/*
- * Copyright (c) 2000-2002 Apple Computer, Inc. All rights reserved.
+ * Copyright (c) 2000-2012 Apple Inc. All rights reserved.
*
- * @APPLE_LICENSE_HEADER_START@
+ * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
*
- * The contents of this file constitute Original Code as defined in and
- * are subject to the Apple Public Source License Version 1.1 (the
- * "License"). You may not use this file except in compliance with the
- * License. Please obtain a copy of the License at
- * http://www.apple.com/publicsource and read it before using this file.
+ * This file contains Original Code and/or Modifications of Original Code
+ * as defined in and that are subject to the Apple Public Source License
+ * Version 2.0 (the 'License'). You may not use this file except in
+ * compliance with the License. The rights granted to you under the License
+ * may not be used to create, or enable the creation or redistribution of,
+ * unlawful or unlicensed copies of an Apple operating system, or to
+ * circumvent, violate, or enable the circumvention or violation of, any
+ * terms of an Apple operating system software license agreement.
*
- * This Original Code and all software distributed under the License are
- * distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY KIND, EITHER
+ * Please obtain a copy of the License at
+ * http://www.opensource.apple.com/apsl/ and read it before using this file.
+ *
+ * The Original Code and all software distributed under the License are
+ * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
* EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
* INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE OR NON-INFRINGEMENT. Please see the
- * License for the specific language governing rights and limitations
- * under the License.
+ * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
+ * Please see the License for the specific language governing rights and
+ * limitations under the License.
*
- * @APPLE_LICENSE_HEADER_END@
+ * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
*/
/*
* @OSF_COPYRIGHT@
* Virtual memory mapping module.
*/
-#include <cpus.h>
#include <task_swapper.h>
#include <mach_assert.h>
+#include <vm/vm_options.h>
+
+#include <libkern/OSAtomic.h>
+
#include <mach/kern_return.h>
#include <mach/port.h>
#include <mach/vm_attributes.h>
#include <mach/vm_param.h>
#include <mach/vm_behavior.h>
#include <mach/vm_statistics.h>
+#include <mach/memory_object.h>
+#include <mach/mach_vm.h>
+#include <machine/cpu_capabilities.h>
+#include <mach/sdt.h>
+
#include <kern/assert.h>
#include <kern/counters.h>
+#include <kern/kalloc.h>
#include <kern/zalloc.h>
+
+#include <vm/cpm.h>
+#include <vm/vm_compressor_pager.h>
#include <vm/vm_init.h>
#include <vm/vm_fault.h>
#include <vm/vm_map.h>
#include <vm/vm_object.h>
#include <vm/vm_page.h>
+#include <vm/vm_pageout.h>
#include <vm/vm_kern.h>
#include <ipc/ipc_port.h>
#include <kern/sched_prim.h>
#include <kern/misc_protos.h>
+#include <kern/xpr.h>
+
#include <mach/vm_map_server.h>
#include <mach/mach_host_server.h>
-#include <ddb/tr.h>
-#include <machine/db_machdep.h>
-#include <kern/xpr.h>
+#include <vm/vm_protos.h>
+#include <vm/vm_purgeable_internal.h>
+#include <vm/vm_protos.h>
+#include <vm/vm_shared_region.h>
+#include <vm/vm_map_store.h>
+
+
+extern u_int32_t random(void); /* from <libkern/libkern.h> */
/* Internal prototypes
*/
-extern boolean_t vm_map_range_check(
- vm_map_t map,
- vm_offset_t start,
- vm_offset_t end,
- vm_map_entry_t *entry);
-
-extern vm_map_entry_t _vm_map_entry_create(
- struct vm_map_header *map_header);
-
-extern void _vm_map_entry_dispose(
- struct vm_map_header *map_header,
- vm_map_entry_t entry);
-
-extern void vm_map_pmap_enter(
- vm_map_t map,
- vm_offset_t addr,
- vm_offset_t end_addr,
- vm_object_t object,
- vm_object_offset_t offset,
- vm_prot_t protection);
-
-extern void _vm_map_clip_end(
- struct vm_map_header *map_header,
- vm_map_entry_t entry,
- vm_offset_t end);
-
-extern void vm_map_entry_delete(
- vm_map_t map,
- vm_map_entry_t entry);
-
-extern kern_return_t vm_map_delete(
- vm_map_t map,
- vm_offset_t start,
- vm_offset_t end,
- int flags);
-
-extern void vm_map_copy_steal_pages(
- vm_map_copy_t copy);
-
-extern kern_return_t vm_map_copy_overwrite_unaligned(
- vm_map_t dst_map,
- vm_map_entry_t entry,
- vm_map_copy_t copy,
- vm_offset_t start);
-
-extern kern_return_t vm_map_copy_overwrite_aligned(
- vm_map_t dst_map,
- vm_map_entry_t tmp_entry,
- vm_map_copy_t copy,
- vm_offset_t start,
- pmap_t pmap);
-
-extern kern_return_t vm_map_copyin_kernel_buffer(
- vm_map_t src_map,
- vm_offset_t src_addr,
- vm_size_t len,
- boolean_t src_destroy,
- vm_map_copy_t *copy_result); /* OUT */
-
-extern kern_return_t vm_map_copyout_kernel_buffer(
- vm_map_t map,
- vm_offset_t *addr, /* IN/OUT */
- vm_map_copy_t copy,
- boolean_t overwrite);
-
-extern void vm_map_fork_share(
- vm_map_t old_map,
- vm_map_entry_t old_entry,
- vm_map_t new_map);
-
-extern boolean_t vm_map_fork_copy(
- vm_map_t old_map,
- vm_map_entry_t *old_entry_p,
- vm_map_t new_map);
-
-extern kern_return_t vm_remap_range_allocate(
- vm_map_t map,
- vm_offset_t *address, /* IN/OUT */
- vm_size_t size,
- vm_offset_t mask,
- boolean_t anywhere,
- vm_map_entry_t *map_entry); /* OUT */
-
-extern void _vm_map_clip_start(
- struct vm_map_header *map_header,
- vm_map_entry_t entry,
- vm_offset_t start);
-
-void vm_region_top_walk(
- vm_map_entry_t entry,
- vm_region_top_info_t top);
-
-void vm_region_walk(
- vm_map_entry_t entry,
- vm_region_extended_info_t extended,
- vm_object_offset_t offset,
- vm_offset_t range,
- vm_map_t map,
- vm_offset_t va);
+
+static void vm_map_simplify_range(
+ vm_map_t map,
+ vm_map_offset_t start,
+ vm_map_offset_t end); /* forward */
+
+static boolean_t vm_map_range_check(
+ vm_map_t map,
+ vm_map_offset_t start,
+ vm_map_offset_t end,
+ vm_map_entry_t *entry);
+
+static vm_map_entry_t _vm_map_entry_create(
+ struct vm_map_header *map_header, boolean_t map_locked);
+
+static void _vm_map_entry_dispose(
+ struct vm_map_header *map_header,
+ vm_map_entry_t entry);
+
+static void vm_map_pmap_enter(
+ vm_map_t map,
+ vm_map_offset_t addr,
+ vm_map_offset_t end_addr,
+ vm_object_t object,
+ vm_object_offset_t offset,
+ vm_prot_t protection);
+
+static void _vm_map_clip_end(
+ struct vm_map_header *map_header,
+ vm_map_entry_t entry,
+ vm_map_offset_t end);
+
+static void _vm_map_clip_start(
+ struct vm_map_header *map_header,
+ vm_map_entry_t entry,
+ vm_map_offset_t start);
+
+static void vm_map_entry_delete(
+ vm_map_t map,
+ vm_map_entry_t entry);
+
+static kern_return_t vm_map_delete(
+ vm_map_t map,
+ vm_map_offset_t start,
+ vm_map_offset_t end,
+ int flags,
+ vm_map_t zap_map);
+
+static kern_return_t vm_map_copy_overwrite_unaligned(
+ vm_map_t dst_map,
+ vm_map_entry_t entry,
+ vm_map_copy_t copy,
+ vm_map_address_t start,
+ boolean_t discard_on_success);
+
+static kern_return_t vm_map_copy_overwrite_aligned(
+ vm_map_t dst_map,
+ vm_map_entry_t tmp_entry,
+ vm_map_copy_t copy,
+ vm_map_offset_t start,
+ pmap_t pmap);
+
+static kern_return_t vm_map_copyin_kernel_buffer(
+ vm_map_t src_map,
+ vm_map_address_t src_addr,
+ vm_map_size_t len,
+ boolean_t src_destroy,
+ vm_map_copy_t *copy_result); /* OUT */
+
+static kern_return_t vm_map_copyout_kernel_buffer(
+ vm_map_t map,
+ vm_map_address_t *addr, /* IN/OUT */
+ vm_map_copy_t copy,
+ boolean_t overwrite,
+ boolean_t consume_on_success);
+
+static void vm_map_fork_share(
+ vm_map_t old_map,
+ vm_map_entry_t old_entry,
+ vm_map_t new_map);
+
+static boolean_t vm_map_fork_copy(
+ vm_map_t old_map,
+ vm_map_entry_t *old_entry_p,
+ vm_map_t new_map);
+
+void vm_map_region_top_walk(
+ vm_map_entry_t entry,
+ vm_region_top_info_t top);
+
+void vm_map_region_walk(
+ vm_map_t map,
+ vm_map_offset_t va,
+ vm_map_entry_t entry,
+ vm_object_offset_t offset,
+ vm_object_size_t range,
+ vm_region_extended_info_t extended,
+ boolean_t look_for_pages,
+ mach_msg_type_number_t count);
+
+static kern_return_t vm_map_wire_nested(
+ vm_map_t map,
+ vm_map_offset_t start,
+ vm_map_offset_t end,
+ vm_prot_t caller_prot,
+ boolean_t user_wire,
+ pmap_t map_pmap,
+ vm_map_offset_t pmap_addr,
+ ppnum_t *physpage_p);
+
+static kern_return_t vm_map_unwire_nested(
+ vm_map_t map,
+ vm_map_offset_t start,
+ vm_map_offset_t end,
+ boolean_t user_wire,
+ pmap_t map_pmap,
+ vm_map_offset_t pmap_addr);
+
+static kern_return_t vm_map_overwrite_submap_recurse(
+ vm_map_t dst_map,
+ vm_map_offset_t dst_addr,
+ vm_map_size_t dst_size);
+
+static kern_return_t vm_map_copy_overwrite_nested(
+ vm_map_t dst_map,
+ vm_map_offset_t dst_addr,
+ vm_map_copy_t copy,
+ boolean_t interruptible,
+ pmap_t pmap,
+ boolean_t discard_on_success);
+
+static kern_return_t vm_map_remap_extract(
+ vm_map_t map,
+ vm_map_offset_t addr,
+ vm_map_size_t size,
+ boolean_t copy,
+ struct vm_map_header *map_header,
+ vm_prot_t *cur_protection,
+ vm_prot_t *max_protection,
+ vm_inherit_t inheritance,
+ boolean_t pageable);
+
+static kern_return_t vm_map_remap_range_allocate(
+ vm_map_t map,
+ vm_map_address_t *address,
+ vm_map_size_t size,
+ vm_map_offset_t mask,
+ int flags,
+ vm_map_entry_t *map_entry);
+
+static void vm_map_region_look_for_page(
+ vm_map_t map,
+ vm_map_offset_t va,
+ vm_object_t object,
+ vm_object_offset_t offset,
+ int max_refcnt,
+ int depth,
+ vm_region_extended_info_t extended,
+ mach_msg_type_number_t count);
+
+static int vm_map_region_count_obj_refs(
+ vm_map_entry_t entry,
+ vm_object_t object);
+
+
+static kern_return_t vm_map_willneed(
+ vm_map_t map,
+ vm_map_offset_t start,
+ vm_map_offset_t end);
+
+static kern_return_t vm_map_reuse_pages(
+ vm_map_t map,
+ vm_map_offset_t start,
+ vm_map_offset_t end);
+
+static kern_return_t vm_map_reusable_pages(
+ vm_map_t map,
+ vm_map_offset_t start,
+ vm_map_offset_t end);
+
+static kern_return_t vm_map_can_reuse(
+ vm_map_t map,
+ vm_map_offset_t start,
+ vm_map_offset_t end);
+
+#if MACH_ASSERT
+static kern_return_t vm_map_pageout(
+ vm_map_t map,
+ vm_map_offset_t start,
+ vm_map_offset_t end);
+#endif /* MACH_ASSERT */
/*
* Macros to copy a vm_map_entry. We must be careful to correctly
* wire count; it's used for map splitting and zone changing in
* vm_map_copyout.
*/
-#define vm_map_entry_copy(NEW,OLD) \
-MACRO_BEGIN \
- *(NEW) = *(OLD); \
- (NEW)->is_shared = FALSE; \
- (NEW)->needs_wakeup = FALSE; \
- (NEW)->in_transition = FALSE; \
- (NEW)->wired_count = 0; \
- (NEW)->user_wired_count = 0; \
+
+#define vm_map_entry_copy(NEW,OLD) \
+MACRO_BEGIN \
+boolean_t _vmec_reserved = (NEW)->from_reserved_zone; \
+ *(NEW) = *(OLD); \
+ (NEW)->is_shared = FALSE; \
+ (NEW)->needs_wakeup = FALSE; \
+ (NEW)->in_transition = FALSE; \
+ (NEW)->wired_count = 0; \
+ (NEW)->user_wired_count = 0; \
+ (NEW)->permanent = FALSE; \
+ (NEW)->used_for_jit = FALSE; \
+ (NEW)->from_reserved_zone = _vmec_reserved; \
+ (NEW)->iokit_acct = FALSE; \
+ (NEW)->vme_resilient_codesign = FALSE; \
+ (NEW)->vme_resilient_media = FALSE; \
+MACRO_END
+
+#define vm_map_entry_copy_full(NEW,OLD) \
+MACRO_BEGIN \
+boolean_t _vmecf_reserved = (NEW)->from_reserved_zone; \
+(*(NEW) = *(OLD)); \
+(NEW)->from_reserved_zone = _vmecf_reserved; \
MACRO_END
-#define vm_map_entry_copy_full(NEW,OLD) (*(NEW) = *(OLD))
+/*
+ * Decide if we want to allow processes to execute from their data or stack areas.
+ * override_nx() returns true if we do. Data/stack execution can be enabled independently
+ * for 32 and 64 bit processes. Set the VM_ABI_32 or VM_ABI_64 flags in allow_data_exec
+ * or allow_stack_exec to enable data execution for that type of data area for that particular
+ * ABI (or both by or'ing the flags together). These are initialized in the architecture
+ * specific pmap files since the default behavior varies according to architecture. The
+ * main reason it varies is because of the need to provide binary compatibility with old
+ * applications that were written before these restrictions came into being. In the old
+ * days, an app could execute anything it could read, but this has slowly been tightened
+ * up over time. The default behavior is:
+ *
+ * 32-bit PPC apps may execute from both stack and data areas
+ * 32-bit Intel apps may exeucte from data areas but not stack
+ * 64-bit PPC/Intel apps may not execute from either data or stack
+ *
+ * An application on any architecture may override these defaults by explicitly
+ * adding PROT_EXEC permission to the page in question with the mprotect(2)
+ * system call. This code here just determines what happens when an app tries to
+ * execute from a page that lacks execute permission.
+ *
+ * Note that allow_data_exec or allow_stack_exec may also be modified by sysctl to change the
+ * default behavior for both 32 and 64 bit apps on a system-wide basis. Furthermore,
+ * a Mach-O header flag bit (MH_NO_HEAP_EXECUTION) can be used to forcibly disallow
+ * execution from data areas for a particular binary even if the arch normally permits it. As
+ * a final wrinkle, a posix_spawn attribute flag can be used to negate this opt-in header bit
+ * to support some complicated use cases, notably browsers with out-of-process plugins that
+ * are not all NX-safe.
+ */
+
+extern int allow_data_exec, allow_stack_exec;
+
+int
+override_nx(vm_map_t map, uint32_t user_tag) /* map unused on arm */
+{
+ int current_abi;
+
+ if (map->pmap == kernel_pmap) return FALSE;
+
+ /*
+ * Determine if the app is running in 32 or 64 bit mode.
+ */
+
+ if (vm_map_is_64bit(map))
+ current_abi = VM_ABI_64;
+ else
+ current_abi = VM_ABI_32;
+
+ /*
+ * Determine if we should allow the execution based on whether it's a
+ * stack or data area and the current architecture.
+ */
+
+ if (user_tag == VM_MEMORY_STACK)
+ return allow_stack_exec & current_abi;
+
+ return (allow_data_exec & current_abi) && (map->map_disallow_data_exec == FALSE);
+}
+
/*
* Virtual memory maps provide for the mapping, protection,
* vm_object_copy_strategically() in vm_object.c.
*/
-zone_t vm_map_zone; /* zone for vm_map structures */
-zone_t vm_map_entry_zone; /* zone for vm_map_entry structures */
-zone_t vm_map_kentry_zone; /* zone for kernel entry structures */
-zone_t vm_map_copy_zone; /* zone for vm_map_copy structures */
+static zone_t vm_map_zone; /* zone for vm_map structures */
+static zone_t vm_map_entry_zone; /* zone for vm_map_entry structures */
+static zone_t vm_map_entry_reserved_zone; /* zone with reserve for non-blocking
+ * allocations */
+static zone_t vm_map_copy_zone; /* zone for vm_map_copy structures */
+zone_t vm_map_holes_zone; /* zone for vm map holes (vm_map_links) structures */
/*
vm_object_t vm_submap_object;
+static void *map_data;
+static vm_size_t map_data_size;
+static void *kentry_data;
+static vm_size_t kentry_data_size;
+static void *map_holes_data;
+static vm_size_t map_holes_data_size;
+
+#define NO_COALESCE_LIMIT ((1024 * 128) - 1)
+
+/* Skip acquiring locks if we're in the midst of a kernel core dump */
+unsigned int not_in_kdp = 1;
+
+unsigned int vm_map_set_cache_attr_count = 0;
+
+kern_return_t
+vm_map_set_cache_attr(
+ vm_map_t map,
+ vm_map_offset_t va)
+{
+ vm_map_entry_t map_entry;
+ vm_object_t object;
+ kern_return_t kr = KERN_SUCCESS;
+
+ vm_map_lock_read(map);
+
+ if (!vm_map_lookup_entry(map, va, &map_entry) ||
+ map_entry->is_sub_map) {
+ /*
+ * that memory is not properly mapped
+ */
+ kr = KERN_INVALID_ARGUMENT;
+ goto done;
+ }
+ object = VME_OBJECT(map_entry);
+
+ if (object == VM_OBJECT_NULL) {
+ /*
+ * there should be a VM object here at this point
+ */
+ kr = KERN_INVALID_ARGUMENT;
+ goto done;
+ }
+ vm_object_lock(object);
+ object->set_cache_attr = TRUE;
+ vm_object_unlock(object);
+
+ vm_map_set_cache_attr_count++;
+done:
+ vm_map_unlock_read(map);
+
+ return kr;
+}
+
+
+#if CONFIG_CODE_DECRYPTION
+/*
+ * vm_map_apple_protected:
+ * This remaps the requested part of the object with an object backed by
+ * the decrypting pager.
+ * crypt_info contains entry points and session data for the crypt module.
+ * The crypt_info block will be copied by vm_map_apple_protected. The data structures
+ * referenced in crypt_info must remain valid until crypt_info->crypt_end() is called.
+ */
+kern_return_t
+vm_map_apple_protected(
+ vm_map_t map,
+ vm_map_offset_t start,
+ vm_map_offset_t end,
+ vm_object_offset_t crypto_backing_offset,
+ struct pager_crypt_info *crypt_info)
+{
+ boolean_t map_locked;
+ kern_return_t kr;
+ vm_map_entry_t map_entry;
+ struct vm_map_entry tmp_entry;
+ memory_object_t unprotected_mem_obj;
+ vm_object_t protected_object;
+ vm_map_offset_t map_addr;
+ vm_map_offset_t start_aligned, end_aligned;
+ vm_object_offset_t crypto_start, crypto_end;
+ int vm_flags;
+
+ map_locked = FALSE;
+ unprotected_mem_obj = MEMORY_OBJECT_NULL;
+
+ start_aligned = vm_map_trunc_page(start, PAGE_MASK_64);
+ end_aligned = vm_map_round_page(end, PAGE_MASK_64);
+ start_aligned = vm_map_trunc_page(start_aligned, VM_MAP_PAGE_MASK(map));
+ end_aligned = vm_map_round_page(end_aligned, VM_MAP_PAGE_MASK(map));
+
+ assert(start_aligned == start);
+ assert(end_aligned == end);
+
+ map_addr = start_aligned;
+ for (map_addr = start_aligned;
+ map_addr < end;
+ map_addr = tmp_entry.vme_end) {
+ vm_map_lock(map);
+ map_locked = TRUE;
+
+ /* lookup the protected VM object */
+ if (!vm_map_lookup_entry(map,
+ map_addr,
+ &map_entry) ||
+ map_entry->is_sub_map ||
+ VME_OBJECT(map_entry) == VM_OBJECT_NULL ||
+ !(map_entry->protection & VM_PROT_EXECUTE)) {
+ /* that memory is not properly mapped */
+ kr = KERN_INVALID_ARGUMENT;
+ goto done;
+ }
+
+ /* get the protected object to be decrypted */
+ protected_object = VME_OBJECT(map_entry);
+ if (protected_object == VM_OBJECT_NULL) {
+ /* there should be a VM object here at this point */
+ kr = KERN_INVALID_ARGUMENT;
+ goto done;
+ }
+ /* ensure protected object stays alive while map is unlocked */
+ vm_object_reference(protected_object);
+
+ /* limit the map entry to the area we want to cover */
+ vm_map_clip_start(map, map_entry, start_aligned);
+ vm_map_clip_end(map, map_entry, end_aligned);
+
+ tmp_entry = *map_entry;
+ map_entry = VM_MAP_ENTRY_NULL; /* not valid after unlocking map */
+ vm_map_unlock(map);
+ map_locked = FALSE;
+
+ /*
+ * This map entry might be only partially encrypted
+ * (if not fully "page-aligned").
+ */
+ crypto_start = 0;
+ crypto_end = tmp_entry.vme_end - tmp_entry.vme_start;
+ if (tmp_entry.vme_start < start) {
+ if (tmp_entry.vme_start != start_aligned) {
+ kr = KERN_INVALID_ADDRESS;
+ }
+ crypto_start += (start - tmp_entry.vme_start);
+ }
+ if (tmp_entry.vme_end > end) {
+ if (tmp_entry.vme_end != end_aligned) {
+ kr = KERN_INVALID_ADDRESS;
+ }
+ crypto_end -= (tmp_entry.vme_end - end);
+ }
+
+ /*
+ * This "extra backing offset" is needed to get the decryption
+ * routine to use the right key. It adjusts for the possibly
+ * relative offset of an interposed "4K" pager...
+ */
+ if (crypto_backing_offset == (vm_object_offset_t) -1) {
+ crypto_backing_offset = VME_OFFSET(&tmp_entry);
+ }
+
+ /*
+ * Lookup (and create if necessary) the protected memory object
+ * matching that VM object.
+ * If successful, this also grabs a reference on the memory object,
+ * to guarantee that it doesn't go away before we get a chance to map
+ * it.
+ */
+ unprotected_mem_obj = apple_protect_pager_setup(
+ protected_object,
+ VME_OFFSET(&tmp_entry),
+ crypto_backing_offset,
+ crypt_info,
+ crypto_start,
+ crypto_end);
+
+ /* release extra ref on protected object */
+ vm_object_deallocate(protected_object);
+
+ if (unprotected_mem_obj == NULL) {
+ kr = KERN_FAILURE;
+ goto done;
+ }
+
+ vm_flags = VM_FLAGS_FIXED | VM_FLAGS_OVERWRITE;
+
+ /* map this memory object in place of the current one */
+ map_addr = tmp_entry.vme_start;
+ kr = vm_map_enter_mem_object(map,
+ &map_addr,
+ (tmp_entry.vme_end -
+ tmp_entry.vme_start),
+ (mach_vm_offset_t) 0,
+ vm_flags,
+ (ipc_port_t) unprotected_mem_obj,
+ 0,
+ TRUE,
+ tmp_entry.protection,
+ tmp_entry.max_protection,
+ tmp_entry.inheritance);
+ assert(kr == KERN_SUCCESS);
+ assert(map_addr == tmp_entry.vme_start);
+
+#if VM_MAP_DEBUG_APPLE_PROTECT
+ printf("APPLE_PROTECT: map %p [0x%llx:0x%llx] pager %p: "
+ "backing:[object:%p,offset:0x%llx,"
+ "crypto_backing_offset:0x%llx,"
+ "crypto_start:0x%llx,crypto_end:0x%llx]\n",
+ map,
+ (uint64_t) map_addr,
+ (uint64_t) (map_addr + (tmp_entry.vme_end -
+ tmp_entry.vme_start)),
+ unprotected_mem_obj,
+ protected_object,
+ VME_OFFSET(&tmp_entry),
+ crypto_backing_offset,
+ crypto_start,
+ crypto_end);
+#endif /* VM_MAP_DEBUG_APPLE_PROTECT */
+
+ /*
+ * Release the reference obtained by
+ * apple_protect_pager_setup().
+ * The mapping (if it succeeded) is now holding a reference on
+ * the memory object.
+ */
+ memory_object_deallocate(unprotected_mem_obj);
+ unprotected_mem_obj = MEMORY_OBJECT_NULL;
+
+ /* continue with next map entry */
+ crypto_backing_offset += (tmp_entry.vme_end -
+ tmp_entry.vme_start);
+ crypto_backing_offset -= crypto_start;
+ }
+ kr = KERN_SUCCESS;
+
+done:
+ if (map_locked) {
+ vm_map_unlock(map);
+ }
+ return kr;
+}
+#endif /* CONFIG_CODE_DECRYPTION */
+
+
+lck_grp_t vm_map_lck_grp;
+lck_grp_attr_t vm_map_lck_grp_attr;
+lck_attr_t vm_map_lck_attr;
+lck_attr_t vm_map_lck_rw_attr;
+
+
/*
* vm_map_init:
*
*
* vm_map_zone: used to allocate maps.
* vm_map_entry_zone: used to allocate map entries.
- * vm_map_kentry_zone: used to allocate map entries for the kernel.
+ * vm_map_entry_reserved_zone: fallback zone for kernel map entries
*
* The kernel allocates map entries from a special zone that is initially
* "crammed" with memory. It would be difficult (perhaps impossible) for
* empty since the very act of allocating memory implies the creation
* of a new entry.
*/
-
-vm_offset_t map_data;
-vm_size_t map_data_size;
-vm_offset_t kentry_data;
-vm_size_t kentry_data_size;
-int kentry_count = 2048; /* to init kentry_data_size */
-
-#define NO_COALESCE_LIMIT (1024 * 128)
-
-/*
- * Threshold for aggressive (eager) page map entering for vm copyout
- * operations. Any copyout larger will NOT be aggressively entered.
- */
-vm_size_t vm_map_aggressive_enter_max; /* set by bootstrap */
-
-/* Skip acquiring locks if we're in the midst of a kernel core dump */
-extern unsigned int not_in_kdp;
-
void
vm_map_init(
void)
{
- vm_map_zone = zinit((vm_size_t) sizeof(struct vm_map), 40*1024,
- PAGE_SIZE, "maps");
-
- vm_map_entry_zone = zinit((vm_size_t) sizeof(struct vm_map_entry),
- 1024*1024, PAGE_SIZE*5,
- "non-kernel map entries");
-
- vm_map_kentry_zone = zinit((vm_size_t) sizeof(struct vm_map_entry),
- kentry_data_size, kentry_data_size,
- "kernel map entries");
-
- vm_map_copy_zone = zinit((vm_size_t) sizeof(struct vm_map_copy),
- 16*1024, PAGE_SIZE, "map copies");
+ vm_size_t entry_zone_alloc_size;
+ const char *mez_name = "VM map entries";
+
+ vm_map_zone = zinit((vm_map_size_t) sizeof(struct _vm_map), 40*1024,
+ PAGE_SIZE, "maps");
+ zone_change(vm_map_zone, Z_NOENCRYPT, TRUE);
+#if defined(__LP64__)
+ entry_zone_alloc_size = PAGE_SIZE * 5;
+#else
+ entry_zone_alloc_size = PAGE_SIZE * 6;
+#endif
+ vm_map_entry_zone = zinit((vm_map_size_t) sizeof(struct vm_map_entry),
+ 1024*1024, entry_zone_alloc_size,
+ mez_name);
+ zone_change(vm_map_entry_zone, Z_NOENCRYPT, TRUE);
+ zone_change(vm_map_entry_zone, Z_NOCALLOUT, TRUE);
+ zone_change(vm_map_entry_zone, Z_GZALLOC_EXEMPT, TRUE);
+
+ vm_map_entry_reserved_zone = zinit((vm_map_size_t) sizeof(struct vm_map_entry),
+ kentry_data_size * 64, kentry_data_size,
+ "Reserved VM map entries");
+ zone_change(vm_map_entry_reserved_zone, Z_NOENCRYPT, TRUE);
+
+ vm_map_copy_zone = zinit((vm_map_size_t) sizeof(struct vm_map_copy),
+ 16*1024, PAGE_SIZE, "VM map copies");
+ zone_change(vm_map_copy_zone, Z_NOENCRYPT, TRUE);
+
+ vm_map_holes_zone = zinit((vm_map_size_t) sizeof(struct vm_map_links),
+ 16*1024, PAGE_SIZE, "VM map holes");
+ zone_change(vm_map_holes_zone, Z_NOENCRYPT, TRUE);
/*
* Cram the map and kentry zones with initial data.
- * Set kentry_zone non-collectible to aid zone_gc().
+ * Set reserved_zone non-collectible to aid zone_gc().
*/
zone_change(vm_map_zone, Z_COLLECT, FALSE);
- zone_change(vm_map_kentry_zone, Z_COLLECT, FALSE);
- zone_change(vm_map_kentry_zone, Z_EXPAND, FALSE);
- zcram(vm_map_zone, map_data, map_data_size);
- zcram(vm_map_kentry_zone, kentry_data, kentry_data_size);
+
+ zone_change(vm_map_entry_reserved_zone, Z_COLLECT, FALSE);
+ zone_change(vm_map_entry_reserved_zone, Z_EXPAND, FALSE);
+ zone_change(vm_map_entry_reserved_zone, Z_FOREIGN, TRUE);
+ zone_change(vm_map_entry_reserved_zone, Z_NOCALLOUT, TRUE);
+ zone_change(vm_map_entry_reserved_zone, Z_CALLERACCT, FALSE); /* don't charge caller */
+ zone_change(vm_map_copy_zone, Z_CALLERACCT, FALSE); /* don't charge caller */
+ zone_change(vm_map_entry_reserved_zone, Z_GZALLOC_EXEMPT, TRUE);
+
+ zone_change(vm_map_holes_zone, Z_COLLECT, TRUE);
+ zone_change(vm_map_holes_zone, Z_EXPAND, TRUE);
+ zone_change(vm_map_holes_zone, Z_FOREIGN, TRUE);
+ zone_change(vm_map_holes_zone, Z_NOCALLOUT, TRUE);
+ zone_change(vm_map_holes_zone, Z_CALLERACCT, TRUE);
+ zone_change(vm_map_holes_zone, Z_GZALLOC_EXEMPT, TRUE);
+
+ /*
+ * Add the stolen memory to zones, adjust zone size and stolen counts.
+ */
+ zcram(vm_map_zone, (vm_offset_t)map_data, map_data_size);
+ zcram(vm_map_entry_reserved_zone, (vm_offset_t)kentry_data, kentry_data_size);
+ zcram(vm_map_holes_zone, (vm_offset_t)map_holes_data, map_holes_data_size);
+ VM_PAGE_MOVE_STOLEN(atop_64(map_data_size) + atop_64(kentry_data_size) + atop_64(map_holes_data_size));
+
+ lck_grp_attr_setdefault(&vm_map_lck_grp_attr);
+ lck_grp_init(&vm_map_lck_grp, "vm_map", &vm_map_lck_grp_attr);
+ lck_attr_setdefault(&vm_map_lck_attr);
+
+ lck_attr_setdefault(&vm_map_lck_rw_attr);
+ lck_attr_cleardebug(&vm_map_lck_rw_attr);
+
+#if CONFIG_FREEZE
+ default_freezer_init();
+#endif /* CONFIG_FREEZE */
}
void
vm_map_steal_memory(
void)
{
- map_data_size = round_page_32(10 * sizeof(struct vm_map));
+ uint32_t kentry_initial_pages;
+
+ map_data_size = round_page(10 * sizeof(struct _vm_map));
map_data = pmap_steal_memory(map_data_size);
-#if 0
/*
- * Limiting worst case: vm_map_kentry_zone needs to map each "available"
- * physical page (i.e. that beyond the kernel image and page tables)
- * individually; we guess at most one entry per eight pages in the
- * real world. This works out to roughly .1 of 1% of physical memory,
- * or roughly 1900 entries (64K) for a 64M machine with 4K pages.
+ * kentry_initial_pages corresponds to the number of kernel map entries
+ * required during bootstrap until the asynchronous replenishment
+ * scheme is activated and/or entries are available from the general
+ * map entry pool.
*/
+#if defined(__LP64__)
+ kentry_initial_pages = 10;
+#else
+ kentry_initial_pages = 6;
#endif
- kentry_count = pmap_free_pages() / 8;
+#if CONFIG_GZALLOC
+ /* If using the guard allocator, reserve more memory for the kernel
+ * reserved map entry pool.
+ */
+ if (gzalloc_enabled())
+ kentry_initial_pages *= 1024;
+#endif
- kentry_data_size =
- round_page_32(kentry_count * sizeof(struct vm_map_entry));
+ kentry_data_size = kentry_initial_pages * PAGE_SIZE;
kentry_data = pmap_steal_memory(kentry_data_size);
+
+ map_holes_data_size = kentry_data_size;
+ map_holes_data = pmap_steal_memory(map_holes_data_size);
+}
+
+void
+vm_kernel_reserved_entry_init(void) {
+ zone_prio_refill_configure(vm_map_entry_reserved_zone, (6*PAGE_SIZE)/sizeof(struct vm_map_entry));
+ zone_prio_refill_configure(vm_map_holes_zone, (6*PAGE_SIZE)/sizeof(struct vm_map_links));
+}
+
+void
+vm_map_disable_hole_optimization(vm_map_t map)
+{
+ vm_map_entry_t head_entry, hole_entry, next_hole_entry;
+
+ if (map->holelistenabled) {
+
+ head_entry = hole_entry = (vm_map_entry_t) map->holes_list;
+
+ while (hole_entry != NULL) {
+
+ next_hole_entry = hole_entry->vme_next;
+
+ hole_entry->vme_next = NULL;
+ hole_entry->vme_prev = NULL;
+ zfree(vm_map_holes_zone, hole_entry);
+
+ if (next_hole_entry == head_entry) {
+ hole_entry = NULL;
+ } else {
+ hole_entry = next_hole_entry;
+ }
+ }
+
+ map->holes_list = NULL;
+ map->holelistenabled = FALSE;
+
+ map->first_free = vm_map_first_entry(map);
+ SAVE_HINT_HOLE_WRITE(map, NULL);
+ }
+}
+
+boolean_t
+vm_kernel_map_is_kernel(vm_map_t map) {
+ return (map->pmap == kernel_pmap);
}
/*
* the given physical map structure, and having
* the given lower and upper address bounds.
*/
+
+boolean_t vm_map_supports_hole_optimization = TRUE;
+
vm_map_t
vm_map_create(
- pmap_t pmap,
- vm_offset_t min,
- vm_offset_t max,
- boolean_t pageable)
+ pmap_t pmap,
+ vm_map_offset_t min,
+ vm_map_offset_t max,
+ boolean_t pageable)
{
+ static int color_seed = 0;
register vm_map_t result;
+ struct vm_map_links *hole_entry = NULL;
result = (vm_map_t) zalloc(vm_map_zone);
if (result == VM_MAP_NULL)
result->hdr.nentries = 0;
result->hdr.entries_pageable = pageable;
+ vm_map_store_init( &(result->hdr) );
+
+ result->hdr.page_shift = PAGE_SHIFT;
+
result->size = 0;
+ result->user_wire_limit = MACH_VM_MAX_ADDRESS; /* default limit is unlimited */
+ result->user_wire_size = 0;
result->ref_count = 1;
#if TASK_SWAPPER
result->res_count = 1;
result->max_offset = max;
result->wiring_required = FALSE;
result->no_zero_fill = FALSE;
- result->mapped = FALSE;
+ result->mapped_in_other_pmaps = FALSE;
result->wait_for_space = FALSE;
+ result->switch_protect = FALSE;
+ result->disable_vmentry_reuse = FALSE;
+ result->map_disallow_data_exec = FALSE;
+ result->highest_entry_end = 0;
result->first_free = vm_map_to_entry(result);
result->hint = vm_map_to_entry(result);
- vm_map_lock_init(result);
- mutex_init(&result->s_lock, ETAP_VM_RESULT);
+ result->color_rr = (color_seed++) & vm_color_mask;
+ result->jit_entry_exists = FALSE;
+
+ if (vm_map_supports_hole_optimization && pmap != kernel_pmap) {
+ hole_entry = zalloc(vm_map_holes_zone);
+
+ hole_entry->start = min;
+ hole_entry->end = (max > (vm_map_offset_t)MACH_VM_MAX_ADDRESS) ? max : (vm_map_offset_t)MACH_VM_MAX_ADDRESS;
+ result->holes_list = result->hole_hint = hole_entry;
+ hole_entry->prev = hole_entry->next = (vm_map_entry_t) hole_entry;
+ result->holelistenabled = TRUE;
+
+ } else {
+
+ result->holelistenabled = FALSE;
+ }
+#if CONFIG_FREEZE
+ result->default_freezer_handle = NULL;
+#endif
+ vm_map_lock_init(result);
+ lck_mtx_init_ext(&result->s_lock, &result->s_lock_ext, &vm_map_lck_grp, &vm_map_lck_attr);
+
return(result);
}
* Allocates a VM map entry for insertion in the
* given map (or map copy). No fields are filled.
*/
-#define vm_map_entry_create(map) \
- _vm_map_entry_create(&(map)->hdr)
+#define vm_map_entry_create(map, map_locked) _vm_map_entry_create(&(map)->hdr, map_locked)
-#define vm_map_copy_entry_create(copy) \
- _vm_map_entry_create(&(copy)->cpy_hdr)
+#define vm_map_copy_entry_create(copy, map_locked) \
+ _vm_map_entry_create(&(copy)->cpy_hdr, map_locked)
+unsigned reserved_zalloc_count, nonreserved_zalloc_count;
-vm_map_entry_t
+static vm_map_entry_t
_vm_map_entry_create(
- register struct vm_map_header *map_header)
+ struct vm_map_header *map_header, boolean_t __unused map_locked)
{
- register zone_t zone;
- register vm_map_entry_t entry;
+ zone_t zone;
+ vm_map_entry_t entry;
- if (map_header->entries_pageable)
- zone = vm_map_entry_zone;
- else
- zone = vm_map_kentry_zone;
+ zone = vm_map_entry_zone;
+
+ assert(map_header->entries_pageable ? !map_locked : TRUE);
+
+ if (map_header->entries_pageable) {
+ entry = (vm_map_entry_t) zalloc(zone);
+ }
+ else {
+ entry = (vm_map_entry_t) zalloc_canblock(zone, FALSE);
+
+ if (entry == VM_MAP_ENTRY_NULL) {
+ zone = vm_map_entry_reserved_zone;
+ entry = (vm_map_entry_t) zalloc(zone);
+ OSAddAtomic(1, &reserved_zalloc_count);
+ } else
+ OSAddAtomic(1, &nonreserved_zalloc_count);
+ }
- entry = (vm_map_entry_t) zalloc(zone);
if (entry == VM_MAP_ENTRY_NULL)
panic("vm_map_entry_create");
+ entry->from_reserved_zone = (zone == vm_map_entry_reserved_zone);
+ vm_map_store_update( (vm_map_t) NULL, entry, VM_MAP_ENTRY_CREATE);
+#if MAP_ENTRY_CREATION_DEBUG
+ entry->vme_creation_maphdr = map_header;
+ fastbacktrace(&entry->vme_creation_bt[0],
+ (sizeof(entry->vme_creation_bt)/sizeof(uintptr_t)));
+#endif
return(entry);
}
* vm_map_entry_dispose: [ internal use only ]
*
* Inverse of vm_map_entry_create.
+ *
+ * write map lock held so no need to
+ * do anything special to insure correctness
+ * of the stores
*/
#define vm_map_entry_dispose(map, entry) \
-MACRO_BEGIN \
- if((entry) == (map)->first_free) \
- (map)->first_free = vm_map_to_entry(map); \
- if((entry) == (map)->hint) \
- (map)->hint = vm_map_to_entry(map); \
- _vm_map_entry_dispose(&(map)->hdr, (entry)); \
-MACRO_END
+ _vm_map_entry_dispose(&(map)->hdr, (entry))
#define vm_map_copy_entry_dispose(map, entry) \
_vm_map_entry_dispose(&(copy)->cpy_hdr, (entry))
-void
+static void
_vm_map_entry_dispose(
register struct vm_map_header *map_header,
register vm_map_entry_t entry)
{
register zone_t zone;
- if (map_header->entries_pageable)
- zone = vm_map_entry_zone;
+ if (map_header->entries_pageable || !(entry->from_reserved_zone))
+ zone = vm_map_entry_zone;
else
- zone = vm_map_kentry_zone;
+ zone = vm_map_entry_reserved_zone;
+
+ if (!map_header->entries_pageable) {
+ if (zone == vm_map_entry_zone)
+ OSAddAtomic(-1, &nonreserved_zalloc_count);
+ else
+ OSAddAtomic(-1, &reserved_zalloc_count);
+ }
- zfree(zone, (vm_offset_t) entry);
+ zfree(zone, entry);
}
-boolean_t first_free_is_valid(vm_map_t map); /* forward */
-boolean_t first_free_check = FALSE;
+#if MACH_ASSERT
+static boolean_t first_free_check = FALSE;
boolean_t
first_free_is_valid(
vm_map_t map)
{
- vm_map_entry_t entry, next;
-
if (!first_free_check)
return TRUE;
-
- entry = vm_map_to_entry(map);
- next = entry->vme_next;
- while (trunc_page_32(next->vme_start) == trunc_page_32(entry->vme_end) ||
- (trunc_page_32(next->vme_start) == trunc_page_32(entry->vme_start) &&
- next != vm_map_to_entry(map))) {
- entry = next;
- next = entry->vme_next;
- if (entry == vm_map_to_entry(map))
- break;
- }
- if (map->first_free != entry) {
- printf("Bad first_free for map 0x%x: 0x%x should be 0x%x\n",
- map, map->first_free, entry);
- return FALSE;
- }
- return TRUE;
+
+ return( first_free_is_valid_store( map ));
}
-
-/*
- * UPDATE_FIRST_FREE:
- *
- * Updates the map->first_free pointer to the
- * entry immediately before the first hole in the map.
- * The map should be locked.
- */
-#define UPDATE_FIRST_FREE(map, new_first_free) \
-MACRO_BEGIN \
- vm_map_t UFF_map; \
- vm_map_entry_t UFF_first_free; \
- vm_map_entry_t UFF_next_entry; \
- UFF_map = (map); \
- UFF_first_free = (new_first_free); \
- UFF_next_entry = UFF_first_free->vme_next; \
- while (trunc_page_32(UFF_next_entry->vme_start) == \
- trunc_page_32(UFF_first_free->vme_end) || \
- (trunc_page_32(UFF_next_entry->vme_start) == \
- trunc_page_32(UFF_first_free->vme_start) && \
- UFF_next_entry != vm_map_to_entry(UFF_map))) { \
- UFF_first_free = UFF_next_entry; \
- UFF_next_entry = UFF_first_free->vme_next; \
- if (UFF_first_free == vm_map_to_entry(UFF_map)) \
- break; \
- } \
- UFF_map->first_free = UFF_first_free; \
- assert(first_free_is_valid(UFF_map)); \
-MACRO_END
-
-/*
- * vm_map_entry_{un,}link:
- *
- * Insert/remove entries from maps (or map copies).
- */
-#define vm_map_entry_link(map, after_where, entry) \
-MACRO_BEGIN \
- vm_map_t VMEL_map; \
- vm_map_entry_t VMEL_entry; \
- VMEL_map = (map); \
- VMEL_entry = (entry); \
- _vm_map_entry_link(&VMEL_map->hdr, after_where, VMEL_entry); \
- UPDATE_FIRST_FREE(VMEL_map, VMEL_map->first_free); \
-MACRO_END
+#endif /* MACH_ASSERT */
#define vm_map_copy_entry_link(copy, after_where, entry) \
- _vm_map_entry_link(&(copy)->cpy_hdr, after_where, (entry))
-
-#define _vm_map_entry_link(hdr, after_where, entry) \
- MACRO_BEGIN \
- (hdr)->nentries++; \
- (entry)->vme_prev = (after_where); \
- (entry)->vme_next = (after_where)->vme_next; \
- (entry)->vme_prev->vme_next = (entry)->vme_next->vme_prev = (entry); \
- MACRO_END
-
-#define vm_map_entry_unlink(map, entry) \
-MACRO_BEGIN \
- vm_map_t VMEU_map; \
- vm_map_entry_t VMEU_entry; \
- vm_map_entry_t VMEU_first_free; \
- VMEU_map = (map); \
- VMEU_entry = (entry); \
- if (VMEU_entry->vme_start <= VMEU_map->first_free->vme_start) \
- VMEU_first_free = VMEU_entry->vme_prev; \
- else \
- VMEU_first_free = VMEU_map->first_free; \
- _vm_map_entry_unlink(&VMEU_map->hdr, VMEU_entry); \
- UPDATE_FIRST_FREE(VMEU_map, VMEU_first_free); \
-MACRO_END
+ _vm_map_store_entry_link(&(copy)->cpy_hdr, after_where, (entry))
#define vm_map_copy_entry_unlink(copy, entry) \
- _vm_map_entry_unlink(&(copy)->cpy_hdr, (entry))
-
-#define _vm_map_entry_unlink(hdr, entry) \
- MACRO_BEGIN \
- (hdr)->nentries--; \
- (entry)->vme_next->vme_prev = (entry)->vme_prev; \
- (entry)->vme_prev->vme_next = (entry)->vme_next; \
- MACRO_END
+ _vm_map_store_entry_unlink(&(copy)->cpy_hdr, (entry))
#if MACH_ASSERT && TASK_SWAPPER
/*
assert(map->res_count >= 0);
assert(map->ref_count >= map->res_count);
if (map->res_count == 0) {
- mutex_unlock(&map->s_lock);
+ lck_mtx_unlock(&map->s_lock);
vm_map_lock(map);
vm_map_swapin(map);
- mutex_lock(&map->s_lock);
+ lck_mtx_lock(&map->s_lock);
++map->res_count;
vm_map_unlock(map);
} else
void vm_map_reference_swap(register vm_map_t map)
{
assert(map != VM_MAP_NULL);
- mutex_lock(&map->s_lock);
+ lck_mtx_lock(&map->s_lock);
assert(map->res_count >= 0);
assert(map->ref_count >= map->res_count);
map->ref_count++;
vm_map_res_reference(map);
- mutex_unlock(&map->s_lock);
+ lck_mtx_unlock(&map->s_lock);
}
/*
{
assert(map->res_count > 0);
if (--map->res_count == 0) {
- mutex_unlock(&map->s_lock);
+ lck_mtx_unlock(&map->s_lock);
vm_map_lock(map);
vm_map_swapout(map);
vm_map_unlock(map);
- mutex_lock(&map->s_lock);
+ lck_mtx_lock(&map->s_lock);
}
assert(map->ref_count >= map->res_count);
}
*/
void
vm_map_destroy(
- register vm_map_t map)
-{
+ vm_map_t map,
+ int flags)
+{
vm_map_lock(map);
- (void) vm_map_delete(map, map->min_offset,
- map->max_offset, VM_MAP_NO_FLAGS);
+
+ /* final cleanup: no need to unnest shared region */
+ flags |= VM_MAP_REMOVE_NO_UNNESTING;
+
+ /* clean up regular map entries */
+ (void) vm_map_delete(map, map->min_offset, map->max_offset,
+ flags, VM_MAP_NULL);
+ /* clean up leftover special mappings (commpage, etc...) */
+ (void) vm_map_delete(map, 0x0, 0xFFFFFFFFFFFFF000ULL,
+ flags, VM_MAP_NULL);
+
+#if CONFIG_FREEZE
+ if (map->default_freezer_handle) {
+ default_freezer_handle_deallocate(map->default_freezer_handle);
+ map->default_freezer_handle = NULL;
+ }
+#endif
+ vm_map_disable_hole_optimization(map);
vm_map_unlock(map);
+ assert(map->hdr.nentries == 0);
+
if(map->pmap)
pmap_destroy(map->pmap);
- zfree(vm_map_zone, (vm_offset_t) map);
+ zfree(vm_map_zone, map);
}
#if TASK_SWAPPER
void vm_map_swapin (vm_map_t map)
{
register vm_map_entry_t entry;
-
+
if (!vm_map_swap_enable) /* debug */
return;
entry = vm_map_first_entry(map);
while (entry != vm_map_to_entry(map)) {
- if (entry->object.vm_object != VM_OBJECT_NULL) {
+ if (VME_OBJECT(entry) != VM_OBJECT_NULL) {
if (entry->is_sub_map) {
- vm_map_t lmap = entry->object.sub_map;
- mutex_lock(&lmap->s_lock);
+ vm_map_t lmap = VME_SUBMAP(entry);
+ lck_mtx_lock(&lmap->s_lock);
vm_map_res_reference(lmap);
- mutex_unlock(&lmap->s_lock);
+ lck_mtx_unlock(&lmap->s_lock);
} else {
- vm_object_t object = entry->object.vm_object;
+ vm_object_t object = VME_OBEJCT(entry);
vm_object_lock(object);
/*
* This call may iterate through the
* If we raced with a swapin and lost, the residence count
* will have been incremented to 1, and we simply return.
*/
- mutex_lock(&map->s_lock);
+ lck_mtx_lock(&map->s_lock);
if (map->res_count != 0) {
- mutex_unlock(&map->s_lock);
+ lck_mtx_unlock(&map->s_lock);
return;
}
- mutex_unlock(&map->s_lock);
+ lck_mtx_unlock(&map->s_lock);
/*
* There are no intermediate states of a map going out or
entry = vm_map_first_entry(map);
while (entry != vm_map_to_entry(map)) {
- if (entry->object.vm_object != VM_OBJECT_NULL) {
+ if (VME_OBJECT(entry) != VM_OBJECT_NULL) {
if (entry->is_sub_map) {
- vm_map_t lmap = entry->object.sub_map;
- mutex_lock(&lmap->s_lock);
+ vm_map_t lmap = VME_SUBMAP(entry);
+ lck_mtx_lock(&lmap->s_lock);
vm_map_res_deallocate(lmap);
- mutex_unlock(&lmap->s_lock);
+ lck_mtx_unlock(&lmap->s_lock);
} else {
- vm_object_t object = entry->object.vm_object;
+ vm_object_t object = VME_OBJECT(entry);
vm_object_lock(object);
/*
* This call may take a long time,
#endif /* TASK_SWAPPER */
-
-/*
- * SAVE_HINT:
- *
- * Saves the specified entry as the hint for
- * future lookups. Performs necessary interlocks.
- */
-#define SAVE_HINT(map,value) \
-MACRO_BEGIN \
- mutex_lock(&(map)->s_lock); \
- (map)->hint = (value); \
- mutex_unlock(&(map)->s_lock); \
-MACRO_END
-
/*
* vm_map_lookup_entry: [ internal use only ]
*
- * Finds the map entry containing (or
- * immediately preceding) the specified address
- * in the given map; the entry is returned
+ * Calls into the vm map store layer to find the map
+ * entry containing (or immediately preceding) the
+ * specified address in the given map; the entry is returned
* in the "entry" parameter. The boolean
* result indicates whether the address is
* actually contained in the map.
*/
boolean_t
vm_map_lookup_entry(
- register vm_map_t map,
- register vm_offset_t address,
+ register vm_map_t map,
+ register vm_map_offset_t address,
vm_map_entry_t *entry) /* OUT */
{
- register vm_map_entry_t cur;
- register vm_map_entry_t last;
-
- /*
- * Start looking either from the head of the
- * list, or from the hint.
- */
- if (not_in_kdp)
- mutex_lock(&map->s_lock);
- cur = map->hint;
- if (not_in_kdp)
- mutex_unlock(&map->s_lock);
-
- if (cur == vm_map_to_entry(map))
- cur = cur->vme_next;
-
- if (address >= cur->vme_start) {
- /*
- * Go from hint to end of list.
- *
- * But first, make a quick check to see if
- * we are already looking at the entry we
- * want (which is usually the case).
- * Note also that we don't need to save the hint
- * here... it is the same hint (unless we are
- * at the header, in which case the hint didn't
- * buy us anything anyway).
- */
- last = vm_map_to_entry(map);
- if ((cur != last) && (cur->vme_end > address)) {
- *entry = cur;
- return(TRUE);
- }
- }
- else {
- /*
- * Go from start to hint, *inclusively*
- */
- last = cur->vme_next;
- cur = vm_map_first_entry(map);
- }
-
- /*
- * Search linearly
- */
-
- while (cur != last) {
- if (cur->vme_end > address) {
- if (address >= cur->vme_start) {
- /*
- * Save this lookup for future
- * hints, and return
- */
-
- *entry = cur;
- if (not_in_kdp)
- SAVE_HINT(map, cur);
- return(TRUE);
- }
- break;
- }
- cur = cur->vme_next;
- }
- *entry = cur->vme_prev;
- if (not_in_kdp)
- SAVE_HINT(map, *entry);
- return(FALSE);
+ return ( vm_map_store_lookup_entry( map, address, entry ));
}
/*
kern_return_t
vm_map_find_space(
register vm_map_t map,
- vm_offset_t *address, /* OUT */
- vm_size_t size,
- vm_offset_t mask,
+ vm_map_offset_t *address, /* OUT */
+ vm_map_size_t size,
+ vm_map_offset_t mask,
+ int flags,
vm_map_entry_t *o_entry) /* OUT */
{
- register vm_map_entry_t entry, new_entry;
- register vm_offset_t start;
- register vm_offset_t end;
+ vm_map_entry_t entry, new_entry;
+ register vm_map_offset_t start;
+ register vm_map_offset_t end;
+ vm_map_entry_t hole_entry;
+
+ if (size == 0) {
+ *address = 0;
+ return KERN_INVALID_ARGUMENT;
+ }
+
+ if (flags & VM_FLAGS_GUARD_AFTER) {
+ /* account for the back guard page in the size */
+ size += VM_MAP_PAGE_SIZE(map);
+ }
- new_entry = vm_map_entry_create(map);
+ new_entry = vm_map_entry_create(map, FALSE);
/*
* Look for the first possible address; if there's already
vm_map_lock(map);
- assert(first_free_is_valid(map));
- if ((entry = map->first_free) == vm_map_to_entry(map))
- start = map->min_offset;
- else
- start = entry->vme_end;
+ if( map->disable_vmentry_reuse == TRUE) {
+ VM_MAP_HIGHEST_ENTRY(map, entry, start);
+ } else {
+ if (map->holelistenabled) {
+ hole_entry = (vm_map_entry_t)map->holes_list;
+
+ if (hole_entry == NULL) {
+ /*
+ * No more space in the map?
+ */
+ vm_map_entry_dispose(map, new_entry);
+ vm_map_unlock(map);
+ return(KERN_NO_SPACE);
+ }
+
+ entry = hole_entry;
+ start = entry->vme_start;
+ } else {
+ assert(first_free_is_valid(map));
+ if ((entry = map->first_free) == vm_map_to_entry(map))
+ start = map->min_offset;
+ else
+ start = entry->vme_end;
+ }
+ }
/*
* In any case, the "entry" always precedes
* wrap around the address.
*/
+ if (flags & VM_FLAGS_GUARD_BEFORE) {
+ /* reserve space for the front guard page */
+ start += VM_MAP_PAGE_SIZE(map);
+ }
end = ((start + mask) & ~mask);
+
if (end < start) {
vm_map_entry_dispose(map, new_entry);
vm_map_unlock(map);
return(KERN_NO_SPACE);
}
- /*
- * If there are no more entries, we must win.
- */
-
next = entry->vme_next;
- if (next == vm_map_to_entry(map))
- break;
- /*
- * If there is another entry, it must be
- * after the end of the potential new region.
- */
+ if (map->holelistenabled) {
+ if (entry->vme_end >= end)
+ break;
+ } else {
+ /*
+ * If there are no more entries, we must win.
+ *
+ * OR
+ *
+ * If there is another entry, it must be
+ * after the end of the potential new region.
+ */
- if (next->vme_start >= end)
- break;
+ if (next == vm_map_to_entry(map))
+ break;
+
+ if (next->vme_start >= end)
+ break;
+ }
/*
* Didn't fit -- move to the next entry.
*/
entry = next;
- start = entry->vme_end;
+
+ if (map->holelistenabled) {
+ if (entry == (vm_map_entry_t) map->holes_list) {
+ /*
+ * Wrapped around
+ */
+ vm_map_entry_dispose(map, new_entry);
+ vm_map_unlock(map);
+ return(KERN_NO_SPACE);
+ }
+ start = entry->vme_start;
+ } else {
+ start = entry->vme_end;
+ }
+ }
+
+ if (map->holelistenabled) {
+ if (vm_map_lookup_entry(map, entry->vme_start, &entry)) {
+ panic("Found an existing entry (%p) instead of potential hole at address: 0x%llx.\n", entry, (unsigned long long)entry->vme_start);
+ }
}
/*
* the map should be locked.
*/
+ if (flags & VM_FLAGS_GUARD_BEFORE) {
+ /* go back for the front guard page */
+ start -= VM_MAP_PAGE_SIZE(map);
+ }
*address = start;
+ assert(start < end);
new_entry->vme_start = start;
new_entry->vme_end = end;
assert(page_aligned(new_entry->vme_start));
assert(page_aligned(new_entry->vme_end));
+ assert(VM_MAP_PAGE_ALIGNED(new_entry->vme_start,
+ VM_MAP_PAGE_MASK(map)));
+ assert(VM_MAP_PAGE_ALIGNED(new_entry->vme_end,
+ VM_MAP_PAGE_MASK(map)));
new_entry->is_shared = FALSE;
new_entry->is_sub_map = FALSE;
- new_entry->use_pmap = FALSE;
- new_entry->object.vm_object = VM_OBJECT_NULL;
- new_entry->offset = (vm_object_offset_t) 0;
+ new_entry->use_pmap = TRUE;
+ VME_OBJECT_SET(new_entry, VM_OBJECT_NULL);
+ VME_OFFSET_SET(new_entry, (vm_object_offset_t) 0);
new_entry->needs_copy = FALSE;
new_entry->in_transition = FALSE;
new_entry->needs_wakeup = FALSE;
+ new_entry->no_cache = FALSE;
+ new_entry->permanent = FALSE;
+ new_entry->superpage_size = FALSE;
+ if (VM_MAP_PAGE_SHIFT(map) != PAGE_SHIFT) {
+ new_entry->map_aligned = TRUE;
+ } else {
+ new_entry->map_aligned = FALSE;
+ }
+
+ new_entry->used_for_jit = FALSE;
+ new_entry->zero_wired_pages = FALSE;
+ new_entry->iokit_acct = FALSE;
+ new_entry->vme_resilient_codesign = FALSE;
+ new_entry->vme_resilient_media = FALSE;
+
+ int alias;
+ VM_GET_FLAGS_ALIAS(flags, alias);
+ VME_ALIAS_SET(new_entry, alias);
/*
* Insert the new entry into the list
*/
- vm_map_entry_link(map, entry, new_entry);
+ vm_map_store_entry_link(map, entry, new_entry);
map->size += size;
/*
* Update the lookup hint
*/
- SAVE_HINT(map, new_entry);
+ SAVE_HINT_MAP_WRITE(map, new_entry);
*o_entry = new_entry;
return(KERN_SUCCESS);
int vm_map_pmap_enter_enable = FALSE;
/*
- * Routine: vm_map_pmap_enter
+ * Routine: vm_map_pmap_enter [internal only]
*
* Description:
* Force pages from the specified object to be entered into
* In/out conditions:
* The source map should not be locked on entry.
*/
-void
+__unused static void
vm_map_pmap_enter(
vm_map_t map,
- register vm_offset_t addr,
- register vm_offset_t end_addr,
+ register vm_map_offset_t addr,
+ register vm_map_offset_t end_addr,
register vm_object_t object,
vm_object_offset_t offset,
vm_prot_t protection)
{
- unsigned int cache_attr;
+ int type_of_fault;
+ kern_return_t kr;
if(map->pmap == 0)
return;
while (addr < end_addr) {
register vm_page_t m;
+
+ /*
+ * TODO:
+ * From vm_map_enter(), we come into this function without the map
+ * lock held or the object lock held.
+ * We haven't taken a reference on the object either.
+ * We should do a proper lookup on the map to make sure
+ * that things are sane before we go locking objects that
+ * could have been deallocated from under us.
+ */
+
vm_object_lock(object);
- vm_object_paging_begin(object);
m = vm_page_lookup(object, offset);
- if (m == VM_PAGE_NULL || m->busy ||
- (m->unusual && ( m->error || m->restart || m->absent ||
- protection & m->page_lock))) {
-
- vm_object_paging_end(object);
+ /*
+ * ENCRYPTED SWAP:
+ * The user should never see encrypted data, so do not
+ * enter an encrypted page in the page table.
+ */
+ if (m == VM_PAGE_NULL || m->busy || m->encrypted ||
+ m->fictitious ||
+ (m->unusual && ( m->error || m->restart || m->absent))) {
vm_object_unlock(object);
return;
}
- assert(!m->fictitious); /* XXX is this possible ??? */
-
if (vm_map_pmap_enter_print) {
printf("vm_map_pmap_enter:");
- printf("map: %x, addr: %x, object: %x, offset: %x\n",
- map, addr, object, offset);
+ printf("map: %p, addr: %llx, object: %p, offset: %llx\n",
+ map, (unsigned long long)addr, object, (unsigned long long)offset);
}
- m->busy = TRUE;
+ type_of_fault = DBG_CACHE_HIT_FAULT;
+ kr = vm_fault_enter(m, map->pmap, addr, protection, protection,
+ VM_PAGE_WIRED(m), FALSE, FALSE, FALSE,
+ 0, /* XXX need user tag / alias? */
+ 0, /* alternate accounting? */
+ NULL,
+ &type_of_fault);
+
+ vm_object_unlock(object);
+
+ offset += PAGE_SIZE_64;
+ addr += PAGE_SIZE;
+ }
+}
+
+boolean_t vm_map_pmap_is_empty(
+ vm_map_t map,
+ vm_map_offset_t start,
+ vm_map_offset_t end);
+boolean_t vm_map_pmap_is_empty(
+ vm_map_t map,
+ vm_map_offset_t start,
+ vm_map_offset_t end)
+{
+#ifdef MACHINE_PMAP_IS_EMPTY
+ return pmap_is_empty(map->pmap, start, end);
+#else /* MACHINE_PMAP_IS_EMPTY */
+ vm_map_offset_t offset;
+ ppnum_t phys_page;
+
+ if (map->pmap == NULL) {
+ return TRUE;
+ }
- if (m->no_isync == TRUE) {
- pmap_sync_caches_phys(m->phys_page);
- m->no_isync = FALSE;
+ for (offset = start;
+ offset < end;
+ offset += PAGE_SIZE) {
+ phys_page = pmap_find_phys(map->pmap, offset);
+ if (phys_page) {
+ kprintf("vm_map_pmap_is_empty(%p,0x%llx,0x%llx): "
+ "page %d at 0x%llx\n",
+ map, (long long)start, (long long)end,
+ phys_page, (long long)offset);
+ return FALSE;
}
+ }
+ return TRUE;
+#endif /* MACHINE_PMAP_IS_EMPTY */
+}
- cache_attr = ((unsigned int)object->wimg_bits) & VM_WIMG_MASK;
- vm_object_unlock(object);
+#define MAX_TRIES_TO_GET_RANDOM_ADDRESS 1000
+kern_return_t
+vm_map_random_address_for_size(
+ vm_map_t map,
+ vm_map_offset_t *address,
+ vm_map_size_t size)
+{
+ kern_return_t kr = KERN_SUCCESS;
+ int tries = 0;
+ vm_map_offset_t random_addr = 0;
+ vm_map_offset_t hole_end;
- PMAP_ENTER(map->pmap, addr, m,
- protection, cache_attr, FALSE);
+ vm_map_entry_t next_entry = VM_MAP_ENTRY_NULL;
+ vm_map_entry_t prev_entry = VM_MAP_ENTRY_NULL;
+ vm_map_size_t vm_hole_size = 0;
+ vm_map_size_t addr_space_size;
- vm_object_lock(object);
+ addr_space_size = vm_map_max(map) - vm_map_min(map);
- PAGE_WAKEUP_DONE(m);
- vm_page_lock_queues();
- if (!m->active && !m->inactive)
- vm_page_activate(m);
- vm_page_unlock_queues();
- vm_object_paging_end(object);
- vm_object_unlock(object);
+ assert(page_aligned(size));
- offset += PAGE_SIZE_64;
- addr += PAGE_SIZE;
+ while (tries < MAX_TRIES_TO_GET_RANDOM_ADDRESS) {
+ random_addr = ((vm_map_offset_t)random()) << PAGE_SHIFT;
+ random_addr = vm_map_trunc_page(
+ vm_map_min(map) +(random_addr % addr_space_size),
+ VM_MAP_PAGE_MASK(map));
+
+ if (vm_map_lookup_entry(map, random_addr, &prev_entry) == FALSE) {
+ if (prev_entry == vm_map_to_entry(map)) {
+ next_entry = vm_map_first_entry(map);
+ } else {
+ next_entry = prev_entry->vme_next;
+ }
+ if (next_entry == vm_map_to_entry(map)) {
+ hole_end = vm_map_max(map);
+ } else {
+ hole_end = next_entry->vme_start;
+ }
+ vm_hole_size = hole_end - random_addr;
+ if (vm_hole_size >= size) {
+ *address = random_addr;
+ break;
+ }
+ }
+ tries++;
+ }
+
+ if (tries == MAX_TRIES_TO_GET_RANDOM_ADDRESS) {
+ kr = KERN_NO_SPACE;
}
+ return kr;
}
/*
*
* Arguments are as defined in the vm_map call.
*/
+int _map_enter_debug = 0;
+static unsigned int vm_map_enter_restore_successes = 0;
+static unsigned int vm_map_enter_restore_failures = 0;
kern_return_t
vm_map_enter(
- register vm_map_t map,
- vm_offset_t *address, /* IN/OUT */
- vm_size_t size,
- vm_offset_t mask,
+ vm_map_t map,
+ vm_map_offset_t *address, /* IN/OUT */
+ vm_map_size_t size,
+ vm_map_offset_t mask,
int flags,
vm_object_t object,
vm_object_offset_t offset,
vm_prot_t max_protection,
vm_inherit_t inheritance)
{
- vm_map_entry_t entry;
- register vm_offset_t start;
- register vm_offset_t end;
+ vm_map_entry_t entry, new_entry;
+ vm_map_offset_t start, tmp_start, tmp_offset;
+ vm_map_offset_t end, tmp_end;
+ vm_map_offset_t tmp2_start, tmp2_end;
+ vm_map_offset_t step;
kern_return_t result = KERN_SUCCESS;
+ vm_map_t zap_old_map = VM_MAP_NULL;
+ vm_map_t zap_new_map = VM_MAP_NULL;
+ boolean_t map_locked = FALSE;
+ boolean_t pmap_empty = TRUE;
+ boolean_t new_mapping_established = FALSE;
+ boolean_t keep_map_locked = ((flags & VM_FLAGS_KEEP_MAP_LOCKED) != 0);
+ boolean_t anywhere = ((flags & VM_FLAGS_ANYWHERE) != 0);
+ boolean_t purgable = ((flags & VM_FLAGS_PURGABLE) != 0);
+ boolean_t overwrite = ((flags & VM_FLAGS_OVERWRITE) != 0);
+ boolean_t no_cache = ((flags & VM_FLAGS_NO_CACHE) != 0);
+ boolean_t is_submap = ((flags & VM_FLAGS_SUBMAP) != 0);
+ boolean_t permanent = ((flags & VM_FLAGS_PERMANENT) != 0);
+ boolean_t entry_for_jit = ((flags & VM_FLAGS_MAP_JIT) != 0);
+ boolean_t iokit_acct = ((flags & VM_FLAGS_IOKIT_ACCT) != 0);
+ boolean_t resilient_codesign = ((flags & VM_FLAGS_RESILIENT_CODESIGN) != 0);
+ boolean_t resilient_media = ((flags & VM_FLAGS_RESILIENT_MEDIA) != 0);
+ unsigned int superpage_size = ((flags & VM_FLAGS_SUPERPAGE_MASK) >> VM_FLAGS_SUPERPAGE_SHIFT);
+ vm_tag_t alias, user_alias;
+ vm_map_offset_t effective_min_offset, effective_max_offset;
+ kern_return_t kr;
+ boolean_t clear_map_aligned = FALSE;
+ vm_map_entry_t hole_entry;
+
+ if (superpage_size) {
+ switch (superpage_size) {
+ /*
+ * Note that the current implementation only supports
+ * a single size for superpages, SUPERPAGE_SIZE, per
+ * architecture. As soon as more sizes are supposed
+ * to be supported, SUPERPAGE_SIZE has to be replaced
+ * with a lookup of the size depending on superpage_size.
+ */
+#ifdef __x86_64__
+ case SUPERPAGE_SIZE_ANY:
+ /* handle it like 2 MB and round up to page size */
+ size = (size + 2*1024*1024 - 1) & ~(2*1024*1024 - 1);
+ case SUPERPAGE_SIZE_2MB:
+ break;
+#endif
+ default:
+ return KERN_INVALID_ARGUMENT;
+ }
+ mask = SUPERPAGE_SIZE-1;
+ if (size & (SUPERPAGE_SIZE-1))
+ return KERN_INVALID_ARGUMENT;
+ inheritance = VM_INHERIT_NONE; /* fork() children won't inherit superpages */
+ }
+
+
+
+ if (resilient_codesign || resilient_media) {
+ if ((cur_protection & (VM_PROT_WRITE | VM_PROT_EXECUTE)) ||
+ (max_protection & (VM_PROT_WRITE | VM_PROT_EXECUTE))) {
+ return KERN_PROTECTION_FAILURE;
+ }
+ }
+
+ if (is_submap) {
+ if (purgable) {
+ /* submaps can not be purgeable */
+ return KERN_INVALID_ARGUMENT;
+ }
+ if (object == VM_OBJECT_NULL) {
+ /* submaps can not be created lazily */
+ return KERN_INVALID_ARGUMENT;
+ }
+ }
+ if (flags & VM_FLAGS_ALREADY) {
+ /*
+ * VM_FLAGS_ALREADY says that it's OK if the same mapping
+ * is already present. For it to be meaningul, the requested
+ * mapping has to be at a fixed address (!VM_FLAGS_ANYWHERE) and
+ * we shouldn't try and remove what was mapped there first
+ * (!VM_FLAGS_OVERWRITE).
+ */
+ if ((flags & VM_FLAGS_ANYWHERE) ||
+ (flags & VM_FLAGS_OVERWRITE)) {
+ return KERN_INVALID_ARGUMENT;
+ }
+ }
+
+ effective_min_offset = map->min_offset;
- boolean_t anywhere = VM_FLAGS_ANYWHERE & flags;
- char alias;
+ if (flags & VM_FLAGS_BEYOND_MAX) {
+ /*
+ * Allow an insertion beyond the map's max offset.
+ */
+ if (vm_map_is_64bit(map))
+ effective_max_offset = 0xFFFFFFFFFFFFF000ULL;
+ else
+ effective_max_offset = 0x00000000FFFFF000ULL;
+ } else {
+ effective_max_offset = map->max_offset;
+ }
+
+ if (size == 0 ||
+ (offset & PAGE_MASK_64) != 0) {
+ *address = 0;
+ return KERN_INVALID_ARGUMENT;
+ }
VM_GET_FLAGS_ALIAS(flags, alias);
+ if (map->pmap == kernel_pmap) {
+ user_alias = VM_KERN_MEMORY_NONE;
+ } else {
+ user_alias = alias;
+ }
#define RETURN(value) { result = value; goto BailOut; }
assert(page_aligned(*address));
assert(page_aligned(size));
- StartAgain: ;
+
+ if (!VM_MAP_PAGE_ALIGNED(size, VM_MAP_PAGE_MASK(map))) {
+ /*
+ * In most cases, the caller rounds the size up to the
+ * map's page size.
+ * If we get a size that is explicitly not map-aligned here,
+ * we'll have to respect the caller's wish and mark the
+ * mapping as "not map-aligned" to avoid tripping the
+ * map alignment checks later.
+ */
+ clear_map_aligned = TRUE;
+ }
+ if (!anywhere &&
+ !VM_MAP_PAGE_ALIGNED(*address, VM_MAP_PAGE_MASK(map))) {
+ /*
+ * We've been asked to map at a fixed address and that
+ * address is not aligned to the map's specific alignment.
+ * The caller should know what it's doing (i.e. most likely
+ * mapping some fragmented copy map, transferring memory from
+ * a VM map with a different alignment), so clear map_aligned
+ * for this new VM map entry and proceed.
+ */
+ clear_map_aligned = TRUE;
+ }
+
+ /*
+ * Only zero-fill objects are allowed to be purgable.
+ * LP64todo - limit purgable objects to 32-bits for now
+ */
+ if (purgable &&
+ (offset != 0 ||
+ (object != VM_OBJECT_NULL &&
+ (object->vo_size != size ||
+ object->purgable == VM_PURGABLE_DENY))
+ || size > ANON_MAX_SIZE)) /* LP64todo: remove when dp capable */
+ return KERN_INVALID_ARGUMENT;
+
+ if (!anywhere && overwrite) {
+ /*
+ * Create a temporary VM map to hold the old mappings in the
+ * affected area while we create the new one.
+ * This avoids releasing the VM map lock in
+ * vm_map_entry_delete() and allows atomicity
+ * when we want to replace some mappings with a new one.
+ * It also allows us to restore the old VM mappings if the
+ * new mapping fails.
+ */
+ zap_old_map = vm_map_create(PMAP_NULL,
+ *address,
+ *address + size,
+ map->hdr.entries_pageable);
+ vm_map_set_page_shift(zap_old_map, VM_MAP_PAGE_SHIFT(map));
+ vm_map_disable_hole_optimization(zap_old_map);
+ }
+
+StartAgain: ;
start = *address;
if (anywhere) {
vm_map_lock(map);
+ map_locked = TRUE;
+
+ if (entry_for_jit) {
+ if (map->jit_entry_exists) {
+ result = KERN_INVALID_ARGUMENT;
+ goto BailOut;
+ }
+ /*
+ * Get a random start address.
+ */
+ result = vm_map_random_address_for_size(map, address, size);
+ if (result != KERN_SUCCESS) {
+ goto BailOut;
+ }
+ start = *address;
+ }
+
/*
* Calculate the first possible address.
*/
- if (start < map->min_offset)
- start = map->min_offset;
- if (start > map->max_offset)
+ if (start < effective_min_offset)
+ start = effective_min_offset;
+ if (start > effective_max_offset)
RETURN(KERN_NO_SPACE);
/*
* address, we have to start after it.
*/
- assert(first_free_is_valid(map));
- if (start == map->min_offset) {
- if ((entry = map->first_free) != vm_map_to_entry(map))
- start = entry->vme_end;
+ if( map->disable_vmentry_reuse == TRUE) {
+ VM_MAP_HIGHEST_ENTRY(map, entry, start);
} else {
- vm_map_entry_t tmp_entry;
- if (vm_map_lookup_entry(map, start, &tmp_entry))
- start = tmp_entry->vme_end;
- entry = tmp_entry;
- }
- /*
- * In any case, the "entry" always precedes
- * the proposed new region throughout the
- * loop:
- */
+ if (map->holelistenabled) {
+ hole_entry = (vm_map_entry_t)map->holes_list;
- while (TRUE) {
- register vm_map_entry_t next;
+ if (hole_entry == NULL) {
+ /*
+ * No more space in the map?
+ */
+ result = KERN_NO_SPACE;
+ goto BailOut;
+ } else {
- /*
- * Find the end of the proposed new region.
- * Be sure we didn't go beyond the end, or
- * wrap around the address.
- */
+ boolean_t found_hole = FALSE;
- end = ((start + mask) & ~mask);
- if (end < start)
- RETURN(KERN_NO_SPACE);
- start = end;
- end += size;
+ do {
+ if (hole_entry->vme_start >= start) {
+ start = hole_entry->vme_start;
+ found_hole = TRUE;
+ break;
+ }
- if ((end > map->max_offset) || (end < start)) {
- if (map->wait_for_space) {
- if (size <= (map->max_offset -
- map->min_offset)) {
+ if (hole_entry->vme_end > start) {
+ found_hole = TRUE;
+ break;
+ }
+ hole_entry = hole_entry->vme_next;
+
+ } while (hole_entry != (vm_map_entry_t) map->holes_list);
+
+ if (found_hole == FALSE) {
+ result = KERN_NO_SPACE;
+ goto BailOut;
+ }
+
+ entry = hole_entry;
+
+ if (start == 0)
+ start += PAGE_SIZE_64;
+ }
+ } else {
+ assert(first_free_is_valid(map));
+
+ entry = map->first_free;
+
+ if (entry == vm_map_to_entry(map)) {
+ entry = NULL;
+ } else {
+ if (entry->vme_next == vm_map_to_entry(map)){
+ /*
+ * Hole at the end of the map.
+ */
+ entry = NULL;
+ } else {
+ if (start < (entry->vme_next)->vme_start ) {
+ start = entry->vme_end;
+ start = vm_map_round_page(start,
+ VM_MAP_PAGE_MASK(map));
+ } else {
+ /*
+ * Need to do a lookup.
+ */
+ entry = NULL;
+ }
+ }
+ }
+
+ if (entry == NULL) {
+ vm_map_entry_t tmp_entry;
+ if (vm_map_lookup_entry(map, start, &tmp_entry)) {
+ assert(!entry_for_jit);
+ start = tmp_entry->vme_end;
+ start = vm_map_round_page(start,
+ VM_MAP_PAGE_MASK(map));
+ }
+ entry = tmp_entry;
+ }
+ }
+ }
+
+ /*
+ * In any case, the "entry" always precedes
+ * the proposed new region throughout the
+ * loop:
+ */
+
+ while (TRUE) {
+ register vm_map_entry_t next;
+
+ /*
+ * Find the end of the proposed new region.
+ * Be sure we didn't go beyond the end, or
+ * wrap around the address.
+ */
+
+ end = ((start + mask) & ~mask);
+ end = vm_map_round_page(end,
+ VM_MAP_PAGE_MASK(map));
+ if (end < start)
+ RETURN(KERN_NO_SPACE);
+ start = end;
+ assert(VM_MAP_PAGE_ALIGNED(start,
+ VM_MAP_PAGE_MASK(map)));
+ end += size;
+
+ if ((end > effective_max_offset) || (end < start)) {
+ if (map->wait_for_space) {
+ assert(!keep_map_locked);
+ if (size <= (effective_max_offset -
+ effective_min_offset)) {
assert_wait((event_t)map,
THREAD_ABORTSAFE);
vm_map_unlock(map);
- thread_block((void (*)(void))0);
+ map_locked = FALSE;
+ thread_block(THREAD_CONTINUE_NULL);
goto StartAgain;
}
}
RETURN(KERN_NO_SPACE);
}
- /*
- * If there are no more entries, we must win.
- */
-
next = entry->vme_next;
- if (next == vm_map_to_entry(map))
- break;
- /*
- * If there is another entry, it must be
- * after the end of the potential new region.
- */
+ if (map->holelistenabled) {
+ if (entry->vme_end >= end)
+ break;
+ } else {
+ /*
+ * If there are no more entries, we must win.
+ *
+ * OR
+ *
+ * If there is another entry, it must be
+ * after the end of the potential new region.
+ */
- if (next->vme_start >= end)
- break;
+ if (next == vm_map_to_entry(map))
+ break;
+
+ if (next->vme_start >= end)
+ break;
+ }
/*
* Didn't fit -- move to the next entry.
*/
entry = next;
- start = entry->vme_end;
+
+ if (map->holelistenabled) {
+ if (entry == (vm_map_entry_t) map->holes_list) {
+ /*
+ * Wrapped around
+ */
+ result = KERN_NO_SPACE;
+ goto BailOut;
+ }
+ start = entry->vme_start;
+ } else {
+ start = entry->vme_end;
+ }
+
+ start = vm_map_round_page(start,
+ VM_MAP_PAGE_MASK(map));
+ }
+
+ if (map->holelistenabled) {
+ if (vm_map_lookup_entry(map, entry->vme_start, &entry)) {
+ panic("Found an existing entry (%p) instead of potential hole at address: 0x%llx.\n", entry, (unsigned long long)entry->vme_start);
+ }
}
+
*address = start;
+ assert(VM_MAP_PAGE_ALIGNED(*address,
+ VM_MAP_PAGE_MASK(map)));
} else {
- vm_map_entry_t temp_entry;
-
/*
* Verify that:
* the address doesn't itself violate
*/
vm_map_lock(map);
+ map_locked = TRUE;
if ((start & mask) != 0)
RETURN(KERN_NO_SPACE);
end = start + size;
- if ((start < map->min_offset) ||
- (end > map->max_offset) ||
+ if ((start < effective_min_offset) ||
+ (end > effective_max_offset) ||
(start >= end)) {
RETURN(KERN_INVALID_ADDRESS);
}
+ if (overwrite && zap_old_map != VM_MAP_NULL) {
+ /*
+ * Fixed mapping and "overwrite" flag: attempt to
+ * remove all existing mappings in the specified
+ * address range, saving them in our "zap_old_map".
+ */
+ (void) vm_map_delete(map, start, end,
+ (VM_MAP_REMOVE_SAVE_ENTRIES |
+ VM_MAP_REMOVE_NO_MAP_ALIGN),
+ zap_old_map);
+ }
+
/*
* ... the starting address isn't allocated
*/
- if (vm_map_lookup_entry(map, start, &temp_entry))
- RETURN(KERN_NO_SPACE);
+ if (vm_map_lookup_entry(map, start, &entry)) {
+ if (! (flags & VM_FLAGS_ALREADY)) {
+ RETURN(KERN_NO_SPACE);
+ }
+ /*
+ * Check if what's already there is what we want.
+ */
+ tmp_start = start;
+ tmp_offset = offset;
+ if (entry->vme_start < start) {
+ tmp_start -= start - entry->vme_start;
+ tmp_offset -= start - entry->vme_start;
+
+ }
+ for (; entry->vme_start < end;
+ entry = entry->vme_next) {
+ /*
+ * Check if the mapping's attributes
+ * match the existing map entry.
+ */
+ if (entry == vm_map_to_entry(map) ||
+ entry->vme_start != tmp_start ||
+ entry->is_sub_map != is_submap ||
+ VME_OFFSET(entry) != tmp_offset ||
+ entry->needs_copy != needs_copy ||
+ entry->protection != cur_protection ||
+ entry->max_protection != max_protection ||
+ entry->inheritance != inheritance ||
+ entry->iokit_acct != iokit_acct ||
+ VME_ALIAS(entry) != alias) {
+ /* not the same mapping ! */
+ RETURN(KERN_NO_SPACE);
+ }
+ /*
+ * Check if the same object is being mapped.
+ */
+ if (is_submap) {
+ if (VME_SUBMAP(entry) !=
+ (vm_map_t) object) {
+ /* not the same submap */
+ RETURN(KERN_NO_SPACE);
+ }
+ } else {
+ if (VME_OBJECT(entry) != object) {
+ /* not the same VM object... */
+ vm_object_t obj2;
+
+ obj2 = VME_OBJECT(entry);
+ if ((obj2 == VM_OBJECT_NULL ||
+ obj2->internal) &&
+ (object == VM_OBJECT_NULL ||
+ object->internal)) {
+ /*
+ * ... but both are
+ * anonymous memory,
+ * so equivalent.
+ */
+ } else {
+ RETURN(KERN_NO_SPACE);
+ }
+ }
+ }
- entry = temp_entry;
+ tmp_offset += entry->vme_end - entry->vme_start;
+ tmp_start += entry->vme_end - entry->vme_start;
+ if (entry->vme_end >= end) {
+ /* reached the end of our mapping */
+ break;
+ }
+ }
+ /* it all matches: let's use what's already there ! */
+ RETURN(KERN_MEMORY_PRESENT);
+ }
/*
* ... the next region doesn't overlap the
/*
* See whether we can avoid creating a new entry (and object) by
* extending one of our neighbors. [So far, we only attempt to
- * extend from below.]
+ * extend from below.] Note that we can never extend/join
+ * purgable objects because they need to remain distinct
+ * entities in order to implement their "volatile object"
+ * semantics.
*/
- if ((object == VM_OBJECT_NULL) &&
- (entry != vm_map_to_entry(map)) &&
- (entry->vme_end == start) &&
- (!entry->is_shared) &&
- (!entry->is_sub_map) &&
- (entry->alias == alias) &&
- (entry->inheritance == inheritance) &&
- (entry->protection == cur_protection) &&
- (entry->max_protection == max_protection) &&
- (entry->behavior == VM_BEHAVIOR_DEFAULT) &&
- (entry->in_transition == 0) &&
- ((alias == VM_MEMORY_REALLOC) || ((entry->vme_end - entry->vme_start) + size < NO_COALESCE_LIMIT)) &&
- (entry->wired_count == 0)) { /* implies user_wired_count == 0 */
- if (vm_object_coalesce(entry->object.vm_object,
- VM_OBJECT_NULL,
- entry->offset,
- (vm_object_offset_t) 0,
- (vm_size_t)(entry->vme_end - entry->vme_start),
- (vm_size_t)(end - entry->vme_end))) {
+ if (purgable || entry_for_jit) {
+ if (object == VM_OBJECT_NULL) {
+
+ object = vm_object_allocate(size);
+ object->copy_strategy = MEMORY_OBJECT_COPY_NONE;
+ object->true_share = TRUE;
+ if (purgable) {
+ task_t owner;
+ object->purgable = VM_PURGABLE_NONVOLATILE;
+ if (map->pmap == kernel_pmap) {
+ /*
+ * Purgeable mappings made in a kernel
+ * map are "owned" by the kernel itself
+ * rather than the current user task
+ * because they're likely to be used by
+ * more than this user task (see
+ * execargs_purgeable_allocate(), for
+ * example).
+ */
+ owner = kernel_task;
+ } else {
+ owner = current_task();
+ }
+ assert(object->vo_purgeable_owner == NULL);
+ assert(object->resident_page_count == 0);
+ assert(object->wired_page_count == 0);
+ vm_object_lock(object);
+ vm_purgeable_nonvolatile_enqueue(object, owner);
+ vm_object_unlock(object);
+ }
+ offset = (vm_object_offset_t)0;
+ }
+ } else if ((is_submap == FALSE) &&
+ (object == VM_OBJECT_NULL) &&
+ (entry != vm_map_to_entry(map)) &&
+ (entry->vme_end == start) &&
+ (!entry->is_shared) &&
+ (!entry->is_sub_map) &&
+ (!entry->in_transition) &&
+ (!entry->needs_wakeup) &&
+ (entry->behavior == VM_BEHAVIOR_DEFAULT) &&
+ (entry->protection == cur_protection) &&
+ (entry->max_protection == max_protection) &&
+ (entry->inheritance == inheritance) &&
+ ((user_alias == VM_MEMORY_REALLOC) ||
+ (VME_ALIAS(entry) == alias)) &&
+ (entry->no_cache == no_cache) &&
+ (entry->permanent == permanent) &&
+ (!entry->superpage_size && !superpage_size) &&
+ /*
+ * No coalescing if not map-aligned, to avoid propagating
+ * that condition any further than needed:
+ */
+ (!entry->map_aligned || !clear_map_aligned) &&
+ (!entry->zero_wired_pages) &&
+ (!entry->used_for_jit && !entry_for_jit) &&
+ (entry->iokit_acct == iokit_acct) &&
+ (!entry->vme_resilient_codesign) &&
+ (!entry->vme_resilient_media) &&
+
+ ((entry->vme_end - entry->vme_start) + size <=
+ (user_alias == VM_MEMORY_REALLOC ?
+ ANON_CHUNK_SIZE :
+ NO_COALESCE_LIMIT)) &&
+
+ (entry->wired_count == 0)) { /* implies user_wired_count == 0 */
+ if (vm_object_coalesce(VME_OBJECT(entry),
+ VM_OBJECT_NULL,
+ VME_OFFSET(entry),
+ (vm_object_offset_t) 0,
+ (vm_map_size_t)(entry->vme_end - entry->vme_start),
+ (vm_map_size_t)(end - entry->vme_end))) {
/*
* Coalesced the two objects - can extend
* new range.
*/
map->size += (end - entry->vme_end);
+ assert(entry->vme_start < end);
+ assert(VM_MAP_PAGE_ALIGNED(end,
+ VM_MAP_PAGE_MASK(map)));
+ if (__improbable(vm_debug_events))
+ DTRACE_VM5(map_entry_extend, vm_map_t, map, vm_map_entry_t, entry, vm_address_t, entry->vme_start, vm_address_t, entry->vme_end, vm_address_t, end);
entry->vme_end = end;
- UPDATE_FIRST_FREE(map, map->first_free);
+ if (map->holelistenabled) {
+ vm_map_store_update_first_free(map, entry, TRUE);
+ } else {
+ vm_map_store_update_first_free(map, map->first_free, TRUE);
+ }
+ new_mapping_established = TRUE;
RETURN(KERN_SUCCESS);
}
}
- /*
- * Create a new entry
- */
-
- { /**/
- register vm_map_entry_t new_entry;
-
- new_entry = vm_map_entry_insert(map, entry, start, end, object,
- offset, needs_copy, FALSE, FALSE,
- cur_protection, max_protection,
- VM_BEHAVIOR_DEFAULT, inheritance, 0);
- new_entry->alias = alias;
- vm_map_unlock(map);
-
- /* Wire down the new entry if the user
- * requested all new map entries be wired.
- */
- if (map->wiring_required) {
- result = vm_map_wire(map, start, end,
- new_entry->protection, TRUE);
- return(result);
- }
-
- if ((object != VM_OBJECT_NULL) &&
- (vm_map_pmap_enter_enable) &&
- (!anywhere) &&
- (!needs_copy) &&
- (size < (128*1024))) {
- vm_map_pmap_enter(map, start, end,
- object, offset, cur_protection);
- }
+ step = superpage_size ? SUPERPAGE_SIZE : (end - start);
+ new_entry = NULL;
- return(result);
- } /**/
+ for (tmp2_start = start; tmp2_start<end; tmp2_start += step) {
+ tmp2_end = tmp2_start + step;
+ /*
+ * Create a new entry
+ * LP64todo - for now, we can only allocate 4GB internal objects
+ * because the default pager can't page bigger ones. Remove this
+ * when it can.
+ *
+ * XXX FBDP
+ * The reserved "page zero" in each process's address space can
+ * be arbitrarily large. Splitting it into separate 4GB objects and
+ * therefore different VM map entries serves no purpose and just
+ * slows down operations on the VM map, so let's not split the
+ * allocation into 4GB chunks if the max protection is NONE. That
+ * memory should never be accessible, so it will never get to the
+ * default pager.
+ */
+ tmp_start = tmp2_start;
+ if (object == VM_OBJECT_NULL &&
+ size > (vm_map_size_t)ANON_CHUNK_SIZE &&
+ max_protection != VM_PROT_NONE &&
+ superpage_size == 0)
+ tmp_end = tmp_start + (vm_map_size_t)ANON_CHUNK_SIZE;
+ else
+ tmp_end = tmp2_end;
+ do {
+ new_entry = vm_map_entry_insert(map, entry, tmp_start, tmp_end,
+ object, offset, needs_copy,
+ FALSE, FALSE,
+ cur_protection, max_protection,
+ VM_BEHAVIOR_DEFAULT,
+ (entry_for_jit)? VM_INHERIT_NONE: inheritance,
+ 0, no_cache,
+ permanent,
+ superpage_size,
+ clear_map_aligned,
+ is_submap);
+
+ assert((object != kernel_object) || (VM_KERN_MEMORY_NONE != alias));
+ VME_ALIAS_SET(new_entry, alias);
+
+ if (entry_for_jit){
+ if (!(map->jit_entry_exists)){
+ new_entry->used_for_jit = TRUE;
+ map->jit_entry_exists = TRUE;
+ }
+ }
- BailOut: ;
- vm_map_unlock(map);
- return(result);
+ if (resilient_codesign &&
+ ! ((cur_protection | max_protection) &
+ (VM_PROT_WRITE | VM_PROT_EXECUTE))) {
+ new_entry->vme_resilient_codesign = TRUE;
+ }
-#undef RETURN
-}
+ if (resilient_media &&
+ ! ((cur_protection | max_protection) &
+ (VM_PROT_WRITE | VM_PROT_EXECUTE))) {
+ new_entry->vme_resilient_media = TRUE;
+ }
-/*
- * vm_map_clip_start: [ internal use only ]
- *
- * Asserts that the given entry begins at or after
- * the specified address; if necessary,
- * it splits the entry into two.
- */
-#ifndef i386
-#define vm_map_clip_start(map, entry, startaddr) \
-MACRO_BEGIN \
- vm_map_t VMCS_map; \
- vm_map_entry_t VMCS_entry; \
- vm_offset_t VMCS_startaddr; \
- VMCS_map = (map); \
- VMCS_entry = (entry); \
- VMCS_startaddr = (startaddr); \
- if (VMCS_startaddr > VMCS_entry->vme_start) { \
- if(entry->use_pmap) { \
- vm_offset_t pmap_base_addr; \
- \
- pmap_base_addr = 0xF0000000 & entry->vme_start; \
- pmap_unnest(map->pmap, (addr64_t)pmap_base_addr); \
- entry->use_pmap = FALSE; \
- } else if(entry->object.vm_object \
- && !entry->is_sub_map \
- && entry->object.vm_object->phys_contiguous) { \
- pmap_remove(map->pmap, \
- (addr64_t)(entry->vme_start), \
- (addr64_t)(entry->vme_end)); \
- } \
- _vm_map_clip_start(&VMCS_map->hdr,VMCS_entry,VMCS_startaddr);\
- } \
- UPDATE_FIRST_FREE(VMCS_map, VMCS_map->first_free); \
-MACRO_END
-#else
-#define vm_map_clip_start(map, entry, startaddr) \
-MACRO_BEGIN \
- vm_map_t VMCS_map; \
- vm_map_entry_t VMCS_entry; \
- vm_offset_t VMCS_startaddr; \
- VMCS_map = (map); \
- VMCS_entry = (entry); \
- VMCS_startaddr = (startaddr); \
- if (VMCS_startaddr > VMCS_entry->vme_start) { \
- _vm_map_clip_start(&VMCS_map->hdr,VMCS_entry,VMCS_startaddr);\
- } \
- UPDATE_FIRST_FREE(VMCS_map, VMCS_map->first_free); \
-MACRO_END
-#endif
+ assert(!new_entry->iokit_acct);
+ if (!is_submap &&
+ object != VM_OBJECT_NULL &&
+ object->purgable != VM_PURGABLE_DENY) {
+ assert(new_entry->use_pmap);
+ assert(!new_entry->iokit_acct);
+ /*
+ * Turn off pmap accounting since
+ * purgeable objects have their
+ * own ledgers.
+ */
+ new_entry->use_pmap = FALSE;
+ } else if (!is_submap &&
+ iokit_acct &&
+ object != VM_OBJECT_NULL &&
+ object->internal) {
+ /* alternate accounting */
+ assert(!new_entry->iokit_acct);
+ assert(new_entry->use_pmap);
+ new_entry->iokit_acct = TRUE;
+ new_entry->use_pmap = FALSE;
+ DTRACE_VM4(
+ vm_map_iokit_mapped_region,
+ vm_map_t, map,
+ vm_map_offset_t, new_entry->vme_start,
+ vm_map_offset_t, new_entry->vme_end,
+ int, VME_ALIAS(new_entry));
+ vm_map_iokit_mapped_region(
+ map,
+ (new_entry->vme_end -
+ new_entry->vme_start));
+ } else if (!is_submap) {
+ assert(!new_entry->iokit_acct);
+ assert(new_entry->use_pmap);
+ }
-#define vm_map_copy_clip_start(copy, entry, startaddr) \
- MACRO_BEGIN \
- if ((startaddr) > (entry)->vme_start) \
- _vm_map_clip_start(&(copy)->cpy_hdr,(entry),(startaddr)); \
- MACRO_END
+ if (is_submap) {
+ vm_map_t submap;
+ boolean_t submap_is_64bit;
+ boolean_t use_pmap;
+
+ assert(new_entry->is_sub_map);
+ assert(!new_entry->use_pmap);
+ assert(!new_entry->iokit_acct);
+ submap = (vm_map_t) object;
+ submap_is_64bit = vm_map_is_64bit(submap);
+ use_pmap = (user_alias == VM_MEMORY_SHARED_PMAP);
+#ifndef NO_NESTED_PMAP
+ if (use_pmap && submap->pmap == NULL) {
+ ledger_t ledger = map->pmap->ledger;
+ /* we need a sub pmap to nest... */
+ submap->pmap = pmap_create(ledger, 0,
+ submap_is_64bit);
+ if (submap->pmap == NULL) {
+ /* let's proceed without nesting... */
+ }
+ }
+ if (use_pmap && submap->pmap != NULL) {
+ kr = pmap_nest(map->pmap,
+ submap->pmap,
+ tmp_start,
+ tmp_start,
+ tmp_end - tmp_start);
+ if (kr != KERN_SUCCESS) {
+ printf("vm_map_enter: "
+ "pmap_nest(0x%llx,0x%llx) "
+ "error 0x%x\n",
+ (long long)tmp_start,
+ (long long)tmp_end,
+ kr);
+ } else {
+ /* we're now nested ! */
+ new_entry->use_pmap = TRUE;
+ pmap_empty = FALSE;
+ }
+ }
+#endif /* NO_NESTED_PMAP */
+ }
+ entry = new_entry;
+
+ if (superpage_size) {
+ vm_page_t pages, m;
+ vm_object_t sp_object;
+
+ VME_OFFSET_SET(entry, 0);
+
+ /* allocate one superpage */
+ kr = cpm_allocate(SUPERPAGE_SIZE, &pages, 0, SUPERPAGE_NBASEPAGES-1, TRUE, 0);
+ if (kr != KERN_SUCCESS) {
+ /* deallocate whole range... */
+ new_mapping_established = TRUE;
+ /* ... but only up to "tmp_end" */
+ size -= end - tmp_end;
+ RETURN(kr);
+ }
-/*
- * This routine is called only when it is known that
- * the entry must be split.
- */
-void
-_vm_map_clip_start(
- register struct vm_map_header *map_header,
- register vm_map_entry_t entry,
- register vm_offset_t start)
-{
- register vm_map_entry_t new_entry;
+ /* create one vm_object per superpage */
+ sp_object = vm_object_allocate((vm_map_size_t)(entry->vme_end - entry->vme_start));
+ sp_object->phys_contiguous = TRUE;
+ sp_object->vo_shadow_offset = (vm_object_offset_t)pages->phys_page*PAGE_SIZE;
+ VME_OBJECT_SET(entry, sp_object);
+ assert(entry->use_pmap);
+
+ /* enter the base pages into the object */
+ vm_object_lock(sp_object);
+ for (offset = 0; offset < SUPERPAGE_SIZE; offset += PAGE_SIZE) {
+ m = pages;
+ pmap_zero_page(m->phys_page);
+ pages = NEXT_PAGE(m);
+ *(NEXT_PAGE_PTR(m)) = VM_PAGE_NULL;
+ vm_page_insert_wired(m, sp_object, offset, VM_KERN_MEMORY_OSFMK);
+ }
+ vm_object_unlock(sp_object);
+ }
+ } while (tmp_end != tmp2_end &&
+ (tmp_start = tmp_end) &&
+ (tmp_end = (tmp2_end - tmp_end > (vm_map_size_t)ANON_CHUNK_SIZE) ?
+ tmp_end + (vm_map_size_t)ANON_CHUNK_SIZE : tmp2_end));
+ }
- /*
- * Split off the front portion --
- * note that we must insert the new
- * entry BEFORE this one, so that
- * this entry has the specified starting
- * address.
- */
+ new_mapping_established = TRUE;
- new_entry = _vm_map_entry_create(map_header);
- vm_map_entry_copy_full(new_entry, entry);
+BailOut:
+ assert(map_locked == TRUE);
- new_entry->vme_end = start;
- entry->offset += (start - entry->vme_start);
- entry->vme_start = start;
+ if (result == KERN_SUCCESS) {
+ vm_prot_t pager_prot;
+ memory_object_t pager;
- _vm_map_entry_link(map_header, entry->vme_prev, new_entry);
+#if DEBUG
+ if (pmap_empty &&
+ !(flags & VM_FLAGS_NO_PMAP_CHECK)) {
+ assert(vm_map_pmap_is_empty(map,
+ *address,
+ *address+size));
+ }
+#endif /* DEBUG */
- if (entry->is_sub_map)
- vm_map_reference(new_entry->object.sub_map);
- else
- vm_object_reference(new_entry->object.vm_object);
-}
+ /*
+ * For "named" VM objects, let the pager know that the
+ * memory object is being mapped. Some pagers need to keep
+ * track of this, to know when they can reclaim the memory
+ * object, for example.
+ * VM calls memory_object_map() for each mapping (specifying
+ * the protection of each mapping) and calls
+ * memory_object_last_unmap() when all the mappings are gone.
+ */
+ pager_prot = max_protection;
+ if (needs_copy) {
+ /*
+ * Copy-On-Write mapping: won't modify
+ * the memory object.
+ */
+ pager_prot &= ~VM_PROT_WRITE;
+ }
+ if (!is_submap &&
+ object != VM_OBJECT_NULL &&
+ object->named &&
+ object->pager != MEMORY_OBJECT_NULL) {
+ vm_object_lock(object);
+ pager = object->pager;
+ if (object->named &&
+ pager != MEMORY_OBJECT_NULL) {
+ assert(object->pager_ready);
+ vm_object_mapping_wait(object, THREAD_UNINT);
+ vm_object_mapping_begin(object);
+ vm_object_unlock(object);
+ kr = memory_object_map(pager, pager_prot);
+ assert(kr == KERN_SUCCESS);
-/*
- * vm_map_clip_end: [ internal use only ]
- *
- * Asserts that the given entry ends at or before
- * the specified address; if necessary,
- * it splits the entry into two.
- */
-#ifndef i386
-#define vm_map_clip_end(map, entry, endaddr) \
-MACRO_BEGIN \
- vm_map_t VMCE_map; \
- vm_map_entry_t VMCE_entry; \
- vm_offset_t VMCE_endaddr; \
- VMCE_map = (map); \
- VMCE_entry = (entry); \
- VMCE_endaddr = (endaddr); \
- if (VMCE_endaddr < VMCE_entry->vme_end) { \
- if(entry->use_pmap) { \
- vm_offset_t pmap_base_addr; \
- \
- pmap_base_addr = 0xF0000000 & entry->vme_start; \
- pmap_unnest(map->pmap, (addr64_t)pmap_base_addr); \
- entry->use_pmap = FALSE; \
- } else if(entry->object.vm_object \
- && !entry->is_sub_map \
- && entry->object.vm_object->phys_contiguous) { \
- pmap_remove(map->pmap, \
- (addr64_t)(entry->vme_start), \
- (addr64_t)(entry->vme_end)); \
- } \
- _vm_map_clip_end(&VMCE_map->hdr,VMCE_entry,VMCE_endaddr); \
- } \
- UPDATE_FIRST_FREE(VMCE_map, VMCE_map->first_free); \
-MACRO_END
-#else
-#define vm_map_clip_end(map, entry, endaddr) \
-MACRO_BEGIN \
- vm_map_t VMCE_map; \
- vm_map_entry_t VMCE_entry; \
- vm_offset_t VMCE_endaddr; \
- VMCE_map = (map); \
- VMCE_entry = (entry); \
- VMCE_endaddr = (endaddr); \
- if (VMCE_endaddr < VMCE_entry->vme_end) { \
- _vm_map_clip_end(&VMCE_map->hdr,VMCE_entry,VMCE_endaddr); \
- } \
- UPDATE_FIRST_FREE(VMCE_map, VMCE_map->first_free); \
-MACRO_END
-#endif
+ vm_object_lock(object);
+ vm_object_mapping_end(object);
+ }
+ vm_object_unlock(object);
+ }
+ }
-#define vm_map_copy_clip_end(copy, entry, endaddr) \
- MACRO_BEGIN \
- if ((endaddr) < (entry)->vme_end) \
- _vm_map_clip_end(&(copy)->cpy_hdr,(entry),(endaddr)); \
- MACRO_END
+ assert(map_locked == TRUE);
-/*
- * This routine is called only when it is known that
- * the entry must be split.
- */
-void
-_vm_map_clip_end(
- register struct vm_map_header *map_header,
- register vm_map_entry_t entry,
- register vm_offset_t end)
-{
- register vm_map_entry_t new_entry;
+ if (!keep_map_locked) {
+ vm_map_unlock(map);
+ map_locked = FALSE;
+ }
/*
- * Create a new entry and insert it
- * AFTER the specified entry
+ * We can't hold the map lock if we enter this block.
*/
- new_entry = _vm_map_entry_create(map_header);
- vm_map_entry_copy_full(new_entry, entry);
-
- new_entry->vme_start = entry->vme_end = end;
- new_entry->offset += (end - entry->vme_start);
-
- _vm_map_entry_link(map_header, entry, new_entry);
+ if (result == KERN_SUCCESS) {
- if (entry->is_sub_map)
- vm_map_reference(new_entry->object.sub_map);
- else
- vm_object_reference(new_entry->object.vm_object);
-}
+ /* Wire down the new entry if the user
+ * requested all new map entries be wired.
+ */
+ if ((map->wiring_required)||(superpage_size)) {
+ assert(!keep_map_locked);
+ pmap_empty = FALSE; /* pmap won't be empty */
+ kr = vm_map_wire(map, start, end,
+ new_entry->protection | VM_PROT_MEMORY_TAG_MAKE(VM_KERN_MEMORY_MLOCK),
+ TRUE);
+ result = kr;
+ }
+ }
-/*
- * VM_MAP_RANGE_CHECK: [ internal use only ]
- *
- * Asserts that the starting and ending region
- * addresses fall within the valid range of the map.
- */
-#define VM_MAP_RANGE_CHECK(map, start, end) \
- { \
- if (start < vm_map_min(map)) \
- start = vm_map_min(map); \
- if (end > vm_map_max(map)) \
- end = vm_map_max(map); \
- if (start > end) \
- start = end; \
+ if (result != KERN_SUCCESS) {
+ if (new_mapping_established) {
+ /*
+ * We have to get rid of the new mappings since we
+ * won't make them available to the user.
+ * Try and do that atomically, to minimize the risk
+ * that someone else create new mappings that range.
+ */
+ zap_new_map = vm_map_create(PMAP_NULL,
+ *address,
+ *address + size,
+ map->hdr.entries_pageable);
+ vm_map_set_page_shift(zap_new_map,
+ VM_MAP_PAGE_SHIFT(map));
+ vm_map_disable_hole_optimization(zap_new_map);
+
+ if (!map_locked) {
+ vm_map_lock(map);
+ map_locked = TRUE;
+ }
+ (void) vm_map_delete(map, *address, *address+size,
+ (VM_MAP_REMOVE_SAVE_ENTRIES |
+ VM_MAP_REMOVE_NO_MAP_ALIGN),
+ zap_new_map);
}
+ if (zap_old_map != VM_MAP_NULL &&
+ zap_old_map->hdr.nentries != 0) {
+ vm_map_entry_t entry1, entry2;
-/*
- * vm_map_range_check: [ internal use only ]
- *
- * Check that the region defined by the specified start and
- * end addresses are wholly contained within a single map
- * entry or set of adjacent map entries of the spacified map,
- * i.e. the specified region contains no unmapped space.
- * If any or all of the region is unmapped, FALSE is returned.
- * Otherwise, TRUE is returned and if the output argument 'entry'
- * is not NULL it points to the map entry containing the start
- * of the region.
- *
- * The map is locked for reading on entry and is left locked.
- */
-boolean_t
-vm_map_range_check(
- register vm_map_t map,
- register vm_offset_t start,
- register vm_offset_t end,
- vm_map_entry_t *entry)
-{
- vm_map_entry_t cur;
- register vm_offset_t prev;
-
- /*
- * Basic sanity checks first
- */
- if (start < vm_map_min(map) || end > vm_map_max(map) || start > end)
- return (FALSE);
+ /*
+ * The new mapping failed. Attempt to restore
+ * the old mappings, saved in the "zap_old_map".
+ */
+ if (!map_locked) {
+ vm_map_lock(map);
+ map_locked = TRUE;
+ }
- /*
- * Check first if the region starts within a valid
- * mapping for the map.
- */
- if (!vm_map_lookup_entry(map, start, &cur))
- return (FALSE);
+ /* first check if the coast is still clear */
+ start = vm_map_first_entry(zap_old_map)->vme_start;
+ end = vm_map_last_entry(zap_old_map)->vme_end;
+ if (vm_map_lookup_entry(map, start, &entry1) ||
+ vm_map_lookup_entry(map, end, &entry2) ||
+ entry1 != entry2) {
+ /*
+ * Part of that range has already been
+ * re-mapped: we can't restore the old
+ * mappings...
+ */
+ vm_map_enter_restore_failures++;
+ } else {
+ /*
+ * Transfer the saved map entries from
+ * "zap_old_map" to the original "map",
+ * inserting them all after "entry1".
+ */
+ for (entry2 = vm_map_first_entry(zap_old_map);
+ entry2 != vm_map_to_entry(zap_old_map);
+ entry2 = vm_map_first_entry(zap_old_map)) {
+ vm_map_size_t entry_size;
+
+ entry_size = (entry2->vme_end -
+ entry2->vme_start);
+ vm_map_store_entry_unlink(zap_old_map,
+ entry2);
+ zap_old_map->size -= entry_size;
+ vm_map_store_entry_link(map, entry1, entry2);
+ map->size += entry_size;
+ entry1 = entry2;
+ }
+ if (map->wiring_required) {
+ /*
+ * XXX TODO: we should rewire the
+ * old pages here...
+ */
+ }
+ vm_map_enter_restore_successes++;
+ }
+ }
+ }
/*
- * Optimize for the case that the region is contained
- * in a single map entry.
+ * The caller is responsible for releasing the lock if it requested to
+ * keep the map locked.
*/
- if (entry != (vm_map_entry_t *) NULL)
- *entry = cur;
- if (end <= cur->vme_end)
- return (TRUE);
+ if (map_locked && !keep_map_locked) {
+ vm_map_unlock(map);
+ }
/*
- * If the region is not wholly contained within a
- * single entry, walk the entries looking for holes.
+ * Get rid of the "zap_maps" and all the map entries that
+ * they may still contain.
*/
- prev = cur->vme_end;
- cur = cur->vme_next;
- while ((cur != vm_map_to_entry(map)) && (prev == cur->vme_start)) {
- if (end <= cur->vme_end)
- return (TRUE);
- prev = cur->vme_end;
- cur = cur->vme_next;
+ if (zap_old_map != VM_MAP_NULL) {
+ vm_map_destroy(zap_old_map, VM_MAP_REMOVE_NO_PMAP_CLEANUP);
+ zap_old_map = VM_MAP_NULL;
}
- return (FALSE);
+ if (zap_new_map != VM_MAP_NULL) {
+ vm_map_destroy(zap_new_map, VM_MAP_REMOVE_NO_PMAP_CLEANUP);
+ zap_new_map = VM_MAP_NULL;
+ }
+
+ return result;
+
+#undef RETURN
}
+
/*
- * vm_map_submap: [ kernel use only ]
- *
- * Mark the given range as handled by a subordinate map.
- *
- * This range must have been created with vm_map_find using
- * the vm_submap_object, and no other operations may have been
- * performed on this range prior to calling vm_map_submap.
- *
- * Only a limited number of operations can be performed
- * within this rage after calling vm_map_submap:
- * vm_fault
- * [Don't try vm_map_copyin!]
- *
- * To remove a submapping, one must first remove the
- * range from the superior map, and then destroy the
- * submap (if desired). [Better yet, don't try it.]
+ * Counters for the prefault optimization.
*/
-kern_return_t
-vm_map_submap(
- register vm_map_t map,
- register vm_offset_t start,
- register vm_offset_t end,
- vm_map_t submap,
- vm_offset_t offset,
- boolean_t use_pmap)
+int64_t vm_prefault_nb_pages = 0;
+int64_t vm_prefault_nb_bailout = 0;
+
+static kern_return_t
+vm_map_enter_mem_object_helper(
+ vm_map_t target_map,
+ vm_map_offset_t *address,
+ vm_map_size_t initial_size,
+ vm_map_offset_t mask,
+ int flags,
+ ipc_port_t port,
+ vm_object_offset_t offset,
+ boolean_t copy,
+ vm_prot_t cur_protection,
+ vm_prot_t max_protection,
+ vm_inherit_t inheritance,
+ upl_page_list_ptr_t page_list,
+ unsigned int page_list_count)
{
- vm_map_entry_t entry;
- register kern_return_t result = KERN_INVALID_ARGUMENT;
- register vm_object_t object;
+ vm_map_address_t map_addr;
+ vm_map_size_t map_size;
+ vm_object_t object;
+ vm_object_size_t size;
+ kern_return_t result;
+ boolean_t mask_cur_protection, mask_max_protection;
+ boolean_t try_prefault = (page_list_count != 0);
+ vm_map_offset_t offset_in_mapping = 0;
- vm_map_lock(map);
+ mask_cur_protection = cur_protection & VM_PROT_IS_MASK;
+ mask_max_protection = max_protection & VM_PROT_IS_MASK;
+ cur_protection &= ~VM_PROT_IS_MASK;
+ max_protection &= ~VM_PROT_IS_MASK;
- submap->mapped = TRUE;
+ /*
+ * Check arguments for validity
+ */
+ if ((target_map == VM_MAP_NULL) ||
+ (cur_protection & ~VM_PROT_ALL) ||
+ (max_protection & ~VM_PROT_ALL) ||
+ (inheritance > VM_INHERIT_LAST_VALID) ||
+ (try_prefault && (copy || !page_list)) ||
+ initial_size == 0) {
+ return KERN_INVALID_ARGUMENT;
+ }
+
+ {
+ map_addr = vm_map_trunc_page(*address,
+ VM_MAP_PAGE_MASK(target_map));
+ map_size = vm_map_round_page(initial_size,
+ VM_MAP_PAGE_MASK(target_map));
+ }
+ size = vm_object_round_page(initial_size);
- VM_MAP_RANGE_CHECK(map, start, end);
+ /*
+ * Find the vm object (if any) corresponding to this port.
+ */
+ if (!IP_VALID(port)) {
+ object = VM_OBJECT_NULL;
+ offset = 0;
+ copy = FALSE;
+ } else if (ip_kotype(port) == IKOT_NAMED_ENTRY) {
+ vm_named_entry_t named_entry;
- if (vm_map_lookup_entry(map, start, &entry)) {
- vm_map_clip_start(map, entry, start);
- }
- else
- entry = entry->vme_next;
+ named_entry = (vm_named_entry_t) port->ip_kobject;
- if(entry == vm_map_to_entry(map)) {
- vm_map_unlock(map);
- return KERN_INVALID_ARGUMENT;
- }
+ if (flags & (VM_FLAGS_RETURN_DATA_ADDR |
+ VM_FLAGS_RETURN_4K_DATA_ADDR)) {
+ offset += named_entry->data_offset;
+ }
+
+ /* a few checks to make sure user is obeying rules */
+ if (size == 0) {
+ if (offset >= named_entry->size)
+ return KERN_INVALID_RIGHT;
+ size = named_entry->size - offset;
+ }
+ if (mask_max_protection) {
+ max_protection &= named_entry->protection;
+ }
+ if (mask_cur_protection) {
+ cur_protection &= named_entry->protection;
+ }
+ if ((named_entry->protection & max_protection) !=
+ max_protection)
+ return KERN_INVALID_RIGHT;
+ if ((named_entry->protection & cur_protection) !=
+ cur_protection)
+ return KERN_INVALID_RIGHT;
+ if (offset + size < offset) {
+ /* overflow */
+ return KERN_INVALID_ARGUMENT;
+ }
+ if (named_entry->size < (offset + initial_size)) {
+ return KERN_INVALID_ARGUMENT;
+ }
+
+ if (named_entry->is_copy) {
+ /* for a vm_map_copy, we can only map it whole */
+ if ((size != named_entry->size) &&
+ (vm_map_round_page(size,
+ VM_MAP_PAGE_MASK(target_map)) ==
+ named_entry->size)) {
+ /* XXX FBDP use the rounded size... */
+ size = vm_map_round_page(
+ size,
+ VM_MAP_PAGE_MASK(target_map));
+ }
+
+ if (!(flags & VM_FLAGS_ANYWHERE) &&
+ (offset != 0 ||
+ size != named_entry->size)) {
+ /*
+ * XXX for a mapping at a "fixed" address,
+ * we can't trim after mapping the whole
+ * memory entry, so reject a request for a
+ * partial mapping.
+ */
+ return KERN_INVALID_ARGUMENT;
+ }
+ }
- vm_map_clip_end(map, entry, end);
+ /* the callers parameter offset is defined to be the */
+ /* offset from beginning of named entry offset in object */
+ offset = offset + named_entry->offset;
+
+ if (! VM_MAP_PAGE_ALIGNED(size,
+ VM_MAP_PAGE_MASK(target_map))) {
+ /*
+ * Let's not map more than requested;
+ * vm_map_enter() will handle this "not map-aligned"
+ * case.
+ */
+ map_size = size;
+ }
- if ((entry->vme_start == start) && (entry->vme_end == end) &&
- (!entry->is_sub_map) &&
- ((object = entry->object.vm_object) == vm_submap_object) &&
- (object->resident_page_count == 0) &&
- (object->copy == VM_OBJECT_NULL) &&
- (object->shadow == VM_OBJECT_NULL) &&
- (!object->pager_created)) {
- entry->offset = (vm_object_offset_t)offset;
- entry->object.vm_object = VM_OBJECT_NULL;
- vm_object_deallocate(object);
- entry->is_sub_map = TRUE;
- entry->object.sub_map = submap;
+ named_entry_lock(named_entry);
+ if (named_entry->is_sub_map) {
+ vm_map_t submap;
+
+ if (flags & (VM_FLAGS_RETURN_DATA_ADDR |
+ VM_FLAGS_RETURN_4K_DATA_ADDR)) {
+ panic("VM_FLAGS_RETURN_DATA_ADDR not expected for submap.");
+ }
+
+ submap = named_entry->backing.map;
+ vm_map_lock(submap);
vm_map_reference(submap);
-#ifndef i386
- if ((use_pmap) && (offset == 0)) {
- /* nest if platform code will allow */
- if(submap->pmap == NULL) {
- submap->pmap = pmap_create((vm_size_t) 0);
- if(submap->pmap == PMAP_NULL) {
- return(KERN_NO_SPACE);
- }
+ vm_map_unlock(submap);
+ named_entry_unlock(named_entry);
+
+ result = vm_map_enter(target_map,
+ &map_addr,
+ map_size,
+ mask,
+ flags | VM_FLAGS_SUBMAP,
+ (vm_object_t) submap,
+ offset,
+ copy,
+ cur_protection,
+ max_protection,
+ inheritance);
+ if (result != KERN_SUCCESS) {
+ vm_map_deallocate(submap);
+ } else {
+ /*
+ * No need to lock "submap" just to check its
+ * "mapped" flag: that flag is never reset
+ * once it's been set and if we race, we'll
+ * just end up setting it twice, which is OK.
+ */
+ if (submap->mapped_in_other_pmaps == FALSE &&
+ vm_map_pmap(submap) != PMAP_NULL &&
+ vm_map_pmap(submap) !=
+ vm_map_pmap(target_map)) {
+ /*
+ * This submap is being mapped in a map
+ * that uses a different pmap.
+ * Set its "mapped_in_other_pmaps" flag
+ * to indicate that we now need to
+ * remove mappings from all pmaps rather
+ * than just the submap's pmap.
+ */
+ vm_map_lock(submap);
+ submap->mapped_in_other_pmaps = TRUE;
+ vm_map_unlock(submap);
}
- result = pmap_nest(map->pmap, (entry->object.sub_map)->pmap,
- (addr64_t)start, (addr64_t)start, (uint64_t)(end - start));
- if(result)
- panic("vm_map_submap: pmap_nest failed, rc = %08X\n", result);
- entry->use_pmap = TRUE;
+ *address = map_addr;
}
-#endif
-#ifdef i386
- pmap_remove(map->pmap, (addr64_t)start, (addr64_t)end);
-#endif
- result = KERN_SUCCESS;
- }
- vm_map_unlock(map);
+ return result;
- return(result);
-}
+ } else if (named_entry->is_pager) {
+ unsigned int access;
+ vm_prot_t protections;
+ unsigned int wimg_mode;
-/*
- * vm_map_protect:
- *
- * Sets the protection of the specified address
- * region in the target map. If "set_max" is
- * specified, the maximum protection is to be set;
- * otherwise, only the current protection is affected.
- */
-kern_return_t
-vm_map_protect(
- register vm_map_t map,
- register vm_offset_t start,
- register vm_offset_t end,
- register vm_prot_t new_prot,
- register boolean_t set_max)
-{
- register vm_map_entry_t current;
- register vm_offset_t prev;
- vm_map_entry_t entry;
- vm_prot_t new_max;
- boolean_t clip;
+ protections = named_entry->protection & VM_PROT_ALL;
+ access = GET_MAP_MEM(named_entry->protection);
- XPR(XPR_VM_MAP,
- "vm_map_protect, 0x%X start 0x%X end 0x%X, new 0x%X %d",
- (integer_t)map, start, end, new_prot, set_max);
+ if (flags & (VM_FLAGS_RETURN_DATA_ADDR|
+ VM_FLAGS_RETURN_4K_DATA_ADDR)) {
+ panic("VM_FLAGS_RETURN_DATA_ADDR not expected for submap.");
+ }
- vm_map_lock(map);
+ object = vm_object_enter(named_entry->backing.pager,
+ named_entry->size,
+ named_entry->internal,
+ FALSE,
+ FALSE);
+ if (object == VM_OBJECT_NULL) {
+ named_entry_unlock(named_entry);
+ return KERN_INVALID_OBJECT;
+ }
- /*
- * Lookup the entry. If it doesn't start in a valid
- * entry, return an error. Remember if we need to
- * clip the entry. We don't do it here because we don't
- * want to make any changes until we've scanned the
- * entire range below for address and protection
- * violations.
- */
- if (!(clip = vm_map_lookup_entry(map, start, &entry))) {
- vm_map_unlock(map);
- return(KERN_INVALID_ADDRESS);
- }
+ /* JMM - drop reference on pager here */
- /*
- * Make a first pass to check for protection and address
- * violations.
- */
+ /* create an extra ref for the named entry */
+ vm_object_lock(object);
+ vm_object_reference_locked(object);
+ named_entry->backing.object = object;
+ named_entry->is_pager = FALSE;
+ named_entry_unlock(named_entry);
+
+ wimg_mode = object->wimg_bits;
+
+ if (access == MAP_MEM_IO) {
+ wimg_mode = VM_WIMG_IO;
+ } else if (access == MAP_MEM_COPYBACK) {
+ wimg_mode = VM_WIMG_USE_DEFAULT;
+ } else if (access == MAP_MEM_INNERWBACK) {
+ wimg_mode = VM_WIMG_INNERWBACK;
+ } else if (access == MAP_MEM_WTHRU) {
+ wimg_mode = VM_WIMG_WTHRU;
+ } else if (access == MAP_MEM_WCOMB) {
+ wimg_mode = VM_WIMG_WCOMB;
+ }
- current = entry;
- prev = current->vme_start;
- while ((current != vm_map_to_entry(map)) &&
- (current->vme_start < end)) {
+ /* wait for object (if any) to be ready */
+ if (!named_entry->internal) {
+ while (!object->pager_ready) {
+ vm_object_wait(
+ object,
+ VM_OBJECT_EVENT_PAGER_READY,
+ THREAD_UNINT);
+ vm_object_lock(object);
+ }
+ }
- /*
- * If there is a hole, return an error.
- */
- if (current->vme_start != prev) {
- vm_map_unlock(map);
- return(KERN_INVALID_ADDRESS);
- }
+ if (object->wimg_bits != wimg_mode)
+ vm_object_change_wimg_mode(object, wimg_mode);
- new_max = current->max_protection;
- if(new_prot & VM_PROT_COPY) {
- new_max |= VM_PROT_WRITE;
- if ((new_prot & (new_max | VM_PROT_COPY)) != new_prot) {
- vm_map_unlock(map);
- return(KERN_PROTECTION_FAILURE);
- }
- } else {
- if ((new_prot & new_max) != new_prot) {
- vm_map_unlock(map);
- return(KERN_PROTECTION_FAILURE);
+#if VM_OBJECT_TRACKING_OP_TRUESHARE
+ if (!object->true_share &&
+ vm_object_tracking_inited) {
+ void *bt[VM_OBJECT_TRACKING_BTDEPTH];
+ int num = 0;
+
+ num = OSBacktrace(bt,
+ VM_OBJECT_TRACKING_BTDEPTH);
+ btlog_add_entry(vm_object_tracking_btlog,
+ object,
+ VM_OBJECT_TRACKING_OP_TRUESHARE,
+ bt,
+ num);
}
- }
+#endif /* VM_OBJECT_TRACKING_OP_TRUESHARE */
- prev = current->vme_end;
- current = current->vme_next;
- }
- if (end > prev) {
- vm_map_unlock(map);
- return(KERN_INVALID_ADDRESS);
- }
+ object->true_share = TRUE;
- /*
- * Go back and fix up protections.
- * Clip to start here if the range starts within
- * the entry.
- */
+ if (object->copy_strategy == MEMORY_OBJECT_COPY_SYMMETRIC)
+ object->copy_strategy = MEMORY_OBJECT_COPY_DELAY;
+ vm_object_unlock(object);
- current = entry;
- if (clip) {
- vm_map_clip_start(map, entry, start);
- }
- while ((current != vm_map_to_entry(map)) &&
- (current->vme_start < end)) {
+ } else if (named_entry->is_copy) {
+ kern_return_t kr;
+ vm_map_copy_t copy_map;
+ vm_map_entry_t copy_entry;
+ vm_map_offset_t copy_addr;
+
+ if (flags & ~(VM_FLAGS_FIXED |
+ VM_FLAGS_ANYWHERE |
+ VM_FLAGS_OVERWRITE |
+ VM_FLAGS_RETURN_4K_DATA_ADDR |
+ VM_FLAGS_RETURN_DATA_ADDR)) {
+ named_entry_unlock(named_entry);
+ return KERN_INVALID_ARGUMENT;
+ }
- vm_prot_t old_prot;
+ if (flags & (VM_FLAGS_RETURN_DATA_ADDR |
+ VM_FLAGS_RETURN_4K_DATA_ADDR)) {
+ offset_in_mapping = offset - vm_object_trunc_page(offset);
+ if (flags & VM_FLAGS_RETURN_4K_DATA_ADDR)
+ offset_in_mapping &= ~((signed)(0xFFF));
+ offset = vm_object_trunc_page(offset);
+ map_size = vm_object_round_page(offset + offset_in_mapping + initial_size) - offset;
+ }
- vm_map_clip_end(map, current, end);
+ copy_map = named_entry->backing.copy;
+ assert(copy_map->type == VM_MAP_COPY_ENTRY_LIST);
+ if (copy_map->type != VM_MAP_COPY_ENTRY_LIST) {
+ /* unsupported type; should not happen */
+ printf("vm_map_enter_mem_object: "
+ "memory_entry->backing.copy "
+ "unsupported type 0x%x\n",
+ copy_map->type);
+ named_entry_unlock(named_entry);
+ return KERN_INVALID_ARGUMENT;
+ }
- old_prot = current->protection;
+ /* reserve a contiguous range */
+ kr = vm_map_enter(target_map,
+ &map_addr,
+ /* map whole mem entry, trim later: */
+ named_entry->size,
+ mask,
+ flags & (VM_FLAGS_ANYWHERE |
+ VM_FLAGS_OVERWRITE |
+ VM_FLAGS_RETURN_4K_DATA_ADDR |
+ VM_FLAGS_RETURN_DATA_ADDR),
+ VM_OBJECT_NULL,
+ 0,
+ FALSE, /* copy */
+ cur_protection,
+ max_protection,
+ inheritance);
+ if (kr != KERN_SUCCESS) {
+ named_entry_unlock(named_entry);
+ return kr;
+ }
- if(new_prot & VM_PROT_COPY) {
- /* caller is asking specifically to copy the */
- /* mapped data, this implies that max protection */
- /* will include write. Caller must be prepared */
- /* for loss of shared memory communication in the */
- /* target area after taking this step */
- current->needs_copy = TRUE;
- current->max_protection |= VM_PROT_WRITE;
- }
+ copy_addr = map_addr;
+
+ for (copy_entry = vm_map_copy_first_entry(copy_map);
+ copy_entry != vm_map_copy_to_entry(copy_map);
+ copy_entry = copy_entry->vme_next) {
+ int remap_flags = 0;
+ vm_map_t copy_submap;
+ vm_object_t copy_object;
+ vm_map_size_t copy_size;
+ vm_object_offset_t copy_offset;
+
+ copy_offset = VME_OFFSET(copy_entry);
+ copy_size = (copy_entry->vme_end -
+ copy_entry->vme_start);
+
+ /* sanity check */
+ if ((copy_addr + copy_size) >
+ (map_addr +
+ named_entry->size /* XXX full size */ )) {
+ /* over-mapping too much !? */
+ kr = KERN_INVALID_ARGUMENT;
+ /* abort */
+ break;
+ }
- if (set_max)
- current->protection =
- (current->max_protection =
- new_prot & ~VM_PROT_COPY) &
- old_prot;
- else
- current->protection = new_prot & ~VM_PROT_COPY;
+ /* take a reference on the object */
+ if (copy_entry->is_sub_map) {
+ remap_flags |= VM_FLAGS_SUBMAP;
+ copy_submap = VME_SUBMAP(copy_entry);
+ vm_map_lock(copy_submap);
+ vm_map_reference(copy_submap);
+ vm_map_unlock(copy_submap);
+ copy_object = (vm_object_t) copy_submap;
+ } else {
+ copy_object = VME_OBJECT(copy_entry);
+ vm_object_reference(copy_object);
+ }
+
+ /* over-map the object into destination */
+ remap_flags |= flags;
+ remap_flags |= VM_FLAGS_FIXED;
+ remap_flags |= VM_FLAGS_OVERWRITE;
+ remap_flags &= ~VM_FLAGS_ANYWHERE;
+ kr = vm_map_enter(target_map,
+ ©_addr,
+ copy_size,
+ (vm_map_offset_t) 0,
+ remap_flags,
+ copy_object,
+ copy_offset,
+ copy,
+ cur_protection,
+ max_protection,
+ inheritance);
+ if (kr != KERN_SUCCESS) {
+ if (copy_entry->is_sub_map) {
+ vm_map_deallocate(copy_submap);
+ } else {
+ vm_object_deallocate(copy_object);
+ }
+ /* abort */
+ break;
+ }
+
+ /* next mapping */
+ copy_addr += copy_size;
+ }
+
+ if (kr == KERN_SUCCESS) {
+ if (flags & (VM_FLAGS_RETURN_DATA_ADDR |
+ VM_FLAGS_RETURN_4K_DATA_ADDR)) {
+ *address = map_addr + offset_in_mapping;
+ } else {
+ *address = map_addr;
+ }
+
+ if (offset) {
+ /*
+ * Trim in front, from 0 to "offset".
+ */
+ vm_map_remove(target_map,
+ map_addr,
+ map_addr + offset,
+ 0);
+ *address += offset;
+ }
+ if (offset + map_size < named_entry->size) {
+ /*
+ * Trim in back, from
+ * "offset + map_size" to
+ * "named_entry->size".
+ */
+ vm_map_remove(target_map,
+ (map_addr +
+ offset + map_size),
+ (map_addr +
+ named_entry->size),
+ 0);
+ }
+ }
+ named_entry_unlock(named_entry);
+
+ if (kr != KERN_SUCCESS) {
+ if (! (flags & VM_FLAGS_OVERWRITE)) {
+ /* deallocate the contiguous range */
+ (void) vm_deallocate(target_map,
+ map_addr,
+ map_size);
+ }
+ }
+
+ return kr;
+
+ } else {
+ /* This is the case where we are going to map */
+ /* an already mapped object. If the object is */
+ /* not ready it is internal. An external */
+ /* object cannot be mapped until it is ready */
+ /* we can therefore avoid the ready check */
+ /* in this case. */
+ if (flags & (VM_FLAGS_RETURN_DATA_ADDR |
+ VM_FLAGS_RETURN_4K_DATA_ADDR)) {
+ offset_in_mapping = offset - vm_object_trunc_page(offset);
+ if (flags & VM_FLAGS_RETURN_4K_DATA_ADDR)
+ offset_in_mapping &= ~((signed)(0xFFF));
+ offset = vm_object_trunc_page(offset);
+ map_size = vm_object_round_page(offset + offset_in_mapping + initial_size) - offset;
+ }
+ object = named_entry->backing.object;
+ assert(object != VM_OBJECT_NULL);
+ named_entry_unlock(named_entry);
+ vm_object_reference(object);
+ }
+ } else if (ip_kotype(port) == IKOT_MEMORY_OBJECT) {
/*
- * Update physical map if necessary.
- * If the request is to turn off write protection,
- * we won't do it for real (in pmap). This is because
- * it would cause copy-on-write to fail. We've already
- * set, the new protection in the map, so if a
- * write-protect fault occurred, it will be fixed up
- * properly, COW or not.
+ * JMM - This is temporary until we unify named entries
+ * and raw memory objects.
+ *
+ * Detected fake ip_kotype for a memory object. In
+ * this case, the port isn't really a port at all, but
+ * instead is just a raw memory object.
*/
- /* the 256M hack for existing hardware limitations */
- if (current->protection != old_prot) {
- if(current->is_sub_map && current->use_pmap) {
- vm_offset_t pmap_base_addr;
- vm_offset_t pmap_end_addr;
- vm_map_entry_t local_entry;
-
- pmap_base_addr = 0xF0000000 & current->vme_start;
- pmap_end_addr = (pmap_base_addr + 0x10000000) - 1;
-#ifndef i386
- if(!vm_map_lookup_entry(map,
- pmap_base_addr, &local_entry))
- panic("vm_map_protect: nested pmap area is missing");
- while ((local_entry != vm_map_to_entry(map)) &&
- (local_entry->vme_start < pmap_end_addr)) {
- local_entry->use_pmap = FALSE;
- local_entry = local_entry->vme_next;
- }
- pmap_unnest(map->pmap, (addr64_t)pmap_base_addr);
-#endif
- }
- if (!(current->protection & VM_PROT_WRITE)) {
- /* Look one level in we support nested pmaps */
- /* from mapped submaps which are direct entries */
- /* in our map */
- if(current->is_sub_map && current->use_pmap) {
- pmap_protect(current->object.sub_map->pmap,
- current->vme_start,
- current->vme_end,
- current->protection);
- } else {
- pmap_protect(map->pmap, current->vme_start,
- current->vme_end,
- current->protection);
+ if (flags & (VM_FLAGS_RETURN_DATA_ADDR |
+ VM_FLAGS_RETURN_4K_DATA_ADDR)) {
+ panic("VM_FLAGS_RETURN_DATA_ADDR not expected for raw memory object.");
+ }
+
+ object = vm_object_enter((memory_object_t)port,
+ size, FALSE, FALSE, FALSE);
+ if (object == VM_OBJECT_NULL)
+ return KERN_INVALID_OBJECT;
+
+ /* wait for object (if any) to be ready */
+ if (object != VM_OBJECT_NULL) {
+ if (object == kernel_object) {
+ printf("Warning: Attempt to map kernel object"
+ " by a non-private kernel entity\n");
+ return KERN_INVALID_OBJECT;
+ }
+ if (!object->pager_ready) {
+ vm_object_lock(object);
+
+ while (!object->pager_ready) {
+ vm_object_wait(object,
+ VM_OBJECT_EVENT_PAGER_READY,
+ THREAD_UNINT);
+ vm_object_lock(object);
+ }
+ vm_object_unlock(object);
}
- }
}
- current = current->vme_next;
+ } else {
+ return KERN_INVALID_OBJECT;
}
- /* coalesce the map entries, if possible */
- current = entry;
- while (current != vm_map_to_entry(map) &&
- current->vme_start <= end) {
- vm_map_simplify_entry(map, current);
- current = current->vme_next;
+ if (object != VM_OBJECT_NULL &&
+ object->named &&
+ object->pager != MEMORY_OBJECT_NULL &&
+ object->copy_strategy != MEMORY_OBJECT_COPY_NONE) {
+ memory_object_t pager;
+ vm_prot_t pager_prot;
+ kern_return_t kr;
+
+ /*
+ * For "named" VM objects, let the pager know that the
+ * memory object is being mapped. Some pagers need to keep
+ * track of this, to know when they can reclaim the memory
+ * object, for example.
+ * VM calls memory_object_map() for each mapping (specifying
+ * the protection of each mapping) and calls
+ * memory_object_last_unmap() when all the mappings are gone.
+ */
+ pager_prot = max_protection;
+ if (copy) {
+ /*
+ * Copy-On-Write mapping: won't modify the
+ * memory object.
+ */
+ pager_prot &= ~VM_PROT_WRITE;
+ }
+ vm_object_lock(object);
+ pager = object->pager;
+ if (object->named &&
+ pager != MEMORY_OBJECT_NULL &&
+ object->copy_strategy != MEMORY_OBJECT_COPY_NONE) {
+ assert(object->pager_ready);
+ vm_object_mapping_wait(object, THREAD_UNINT);
+ vm_object_mapping_begin(object);
+ vm_object_unlock(object);
+
+ kr = memory_object_map(pager, pager_prot);
+ assert(kr == KERN_SUCCESS);
+
+ vm_object_lock(object);
+ vm_object_mapping_end(object);
+ }
+ vm_object_unlock(object);
}
- vm_map_unlock(map);
- return(KERN_SUCCESS);
-}
+ /*
+ * Perform the copy if requested
+ */
-/*
- * vm_map_inherit:
- *
- * Sets the inheritance of the specified address
- * range in the target map. Inheritance
- * affects how the map will be shared with
- * child maps at the time of vm_map_fork.
- */
-kern_return_t
-vm_map_inherit(
- register vm_map_t map,
- register vm_offset_t start,
- register vm_offset_t end,
- register vm_inherit_t new_inheritance)
-{
- register vm_map_entry_t entry;
- vm_map_entry_t temp_entry;
+ if (copy) {
+ vm_object_t new_object;
+ vm_object_offset_t new_offset;
- vm_map_lock(map);
+ result = vm_object_copy_strategically(object, offset,
+ map_size,
+ &new_object, &new_offset,
+ ©);
- VM_MAP_RANGE_CHECK(map, start, end);
- if (vm_map_lookup_entry(map, start, &temp_entry)) {
- entry = temp_entry;
- vm_map_clip_start(map, entry, start);
- }
- else {
- temp_entry = temp_entry->vme_next;
- entry = temp_entry;
- }
+ if (result == KERN_MEMORY_RESTART_COPY) {
+ boolean_t success;
+ boolean_t src_needs_copy;
- /* first check entire range for submaps which can't support the */
- /* given inheritance. */
- while ((entry != vm_map_to_entry(map)) && (entry->vme_start < end)) {
- if(entry->is_sub_map) {
- if(new_inheritance == VM_INHERIT_COPY)
- return(KERN_INVALID_ARGUMENT);
+ /*
+ * XXX
+ * We currently ignore src_needs_copy.
+ * This really is the issue of how to make
+ * MEMORY_OBJECT_COPY_SYMMETRIC safe for
+ * non-kernel users to use. Solution forthcoming.
+ * In the meantime, since we don't allow non-kernel
+ * memory managers to specify symmetric copy,
+ * we won't run into problems here.
+ */
+ new_object = object;
+ new_offset = offset;
+ success = vm_object_copy_quickly(&new_object,
+ new_offset,
+ map_size,
+ &src_needs_copy,
+ ©);
+ assert(success);
+ result = KERN_SUCCESS;
}
+ /*
+ * Throw away the reference to the
+ * original object, as it won't be mapped.
+ */
- entry = entry->vme_next;
+ vm_object_deallocate(object);
+
+ if (result != KERN_SUCCESS) {
+ return result;
+ }
+
+ object = new_object;
+ offset = new_offset;
}
- entry = temp_entry;
+ /*
+ * If users want to try to prefault pages, the mapping and prefault
+ * needs to be atomic.
+ */
+ if (try_prefault)
+ flags |= VM_FLAGS_KEEP_MAP_LOCKED;
- while ((entry != vm_map_to_entry(map)) && (entry->vme_start < end)) {
- vm_map_clip_end(map, entry, end);
+ {
+ result = vm_map_enter(target_map,
+ &map_addr, map_size,
+ (vm_map_offset_t)mask,
+ flags,
+ object, offset,
+ copy,
+ cur_protection, max_protection,
+ inheritance);
+ }
+ if (result != KERN_SUCCESS)
+ vm_object_deallocate(object);
- entry->inheritance = new_inheritance;
+ /*
+ * Try to prefault, and do not forget to release the vm map lock.
+ */
+ if (result == KERN_SUCCESS && try_prefault) {
+ mach_vm_address_t va = map_addr;
+ kern_return_t kr = KERN_SUCCESS;
+ unsigned int i = 0;
- entry = entry->vme_next;
+ for (i = 0; i < page_list_count; ++i) {
+ if (UPL_VALID_PAGE(page_list, i)) {
+ /*
+ * If this function call failed, we should stop
+ * trying to optimize, other calls are likely
+ * going to fail too.
+ *
+ * We are not gonna report an error for such
+ * failure though. That's an optimization, not
+ * something critical.
+ */
+ kr = pmap_enter_options(target_map->pmap,
+ va, UPL_PHYS_PAGE(page_list, i),
+ cur_protection, VM_PROT_NONE,
+ 0, TRUE, PMAP_OPTIONS_NOWAIT, NULL);
+ if (kr != KERN_SUCCESS) {
+ OSIncrementAtomic64(&vm_prefault_nb_bailout);
+ break;
+ }
+ OSIncrementAtomic64(&vm_prefault_nb_pages);
+ }
+
+ /* Next virtual address */
+ va += PAGE_SIZE;
+ }
+ vm_map_unlock(target_map);
}
- vm_map_unlock(map);
- return(KERN_SUCCESS);
+ if (flags & (VM_FLAGS_RETURN_DATA_ADDR |
+ VM_FLAGS_RETURN_4K_DATA_ADDR)) {
+ *address = map_addr + offset_in_mapping;
+ } else {
+ *address = map_addr;
+ }
+ return result;
}
-/*
- * vm_map_wire:
- *
- * Sets the pageability of the specified address range in the
- * target map as wired. Regions specified as not pageable require
- * locked-down physical memory and physical page maps. The
- * access_type variable indicates types of accesses that must not
- * generate page faults. This is checked against protection of
- * memory being locked-down.
- *
- * The map must not be locked, but a reference must remain to the
- * map throughout the call.
- */
kern_return_t
-vm_map_wire_nested(
- register vm_map_t map,
- register vm_offset_t start,
- register vm_offset_t end,
- register vm_prot_t access_type,
- boolean_t user_wire,
- pmap_t map_pmap,
- vm_offset_t pmap_addr)
+vm_map_enter_mem_object(
+ vm_map_t target_map,
+ vm_map_offset_t *address,
+ vm_map_size_t initial_size,
+ vm_map_offset_t mask,
+ int flags,
+ ipc_port_t port,
+ vm_object_offset_t offset,
+ boolean_t copy,
+ vm_prot_t cur_protection,
+ vm_prot_t max_protection,
+ vm_inherit_t inheritance)
{
- register vm_map_entry_t entry;
- struct vm_map_entry *first_entry, tmp_entry;
- vm_map_t pmap_map;
- register vm_offset_t s,e;
- kern_return_t rc;
- boolean_t need_wakeup;
- boolean_t main_map = FALSE;
- wait_interrupt_t interruptible_state;
- thread_t cur_thread;
- unsigned int last_timestamp;
- vm_size_t size;
+ return vm_map_enter_mem_object_helper(target_map, address, initial_size, mask, flags,
+ port, offset, copy, cur_protection, max_protection,
+ inheritance, NULL, 0);
+}
- vm_map_lock(map);
- if(map_pmap == NULL)
- main_map = TRUE;
- last_timestamp = map->timestamp;
+kern_return_t
+vm_map_enter_mem_object_prefault(
+ vm_map_t target_map,
+ vm_map_offset_t *address,
+ vm_map_size_t initial_size,
+ vm_map_offset_t mask,
+ int flags,
+ ipc_port_t port,
+ vm_object_offset_t offset,
+ vm_prot_t cur_protection,
+ vm_prot_t max_protection,
+ upl_page_list_ptr_t page_list,
+ unsigned int page_list_count)
+{
+ return vm_map_enter_mem_object_helper(target_map, address, initial_size, mask, flags,
+ port, offset, FALSE, cur_protection, max_protection,
+ VM_INHERIT_DEFAULT, page_list, page_list_count);
+}
- VM_MAP_RANGE_CHECK(map, start, end);
- assert(page_aligned(start));
- assert(page_aligned(end));
- if (start == end) {
- /* We wired what the caller asked for, zero pages */
- vm_map_unlock(map);
- return KERN_SUCCESS;
+
+kern_return_t
+vm_map_enter_mem_object_control(
+ vm_map_t target_map,
+ vm_map_offset_t *address,
+ vm_map_size_t initial_size,
+ vm_map_offset_t mask,
+ int flags,
+ memory_object_control_t control,
+ vm_object_offset_t offset,
+ boolean_t copy,
+ vm_prot_t cur_protection,
+ vm_prot_t max_protection,
+ vm_inherit_t inheritance)
+{
+ vm_map_address_t map_addr;
+ vm_map_size_t map_size;
+ vm_object_t object;
+ vm_object_size_t size;
+ kern_return_t result;
+ memory_object_t pager;
+ vm_prot_t pager_prot;
+ kern_return_t kr;
+
+ /*
+ * Check arguments for validity
+ */
+ if ((target_map == VM_MAP_NULL) ||
+ (cur_protection & ~VM_PROT_ALL) ||
+ (max_protection & ~VM_PROT_ALL) ||
+ (inheritance > VM_INHERIT_LAST_VALID) ||
+ initial_size == 0) {
+ return KERN_INVALID_ARGUMENT;
}
- if (vm_map_lookup_entry(map, start, &first_entry)) {
- entry = first_entry;
- /* vm_map_clip_start will be done later. */
- } else {
- /* Start address is not in map */
- vm_map_unlock(map);
- return(KERN_INVALID_ADDRESS);
+ {
+ map_addr = vm_map_trunc_page(*address,
+ VM_MAP_PAGE_MASK(target_map));
+ map_size = vm_map_round_page(initial_size,
+ VM_MAP_PAGE_MASK(target_map));
}
+ size = vm_object_round_page(initial_size);
- s=start;
- need_wakeup = FALSE;
- cur_thread = current_thread();
- while ((entry != vm_map_to_entry(map)) && (entry->vme_start < end)) {
- /*
- * If another thread is wiring/unwiring this entry then
- * block after informing other thread to wake us up.
- */
- if (entry->in_transition) {
- wait_result_t wait_result;
+ object = memory_object_control_to_vm_object(control);
- /*
- * We have not clipped the entry. Make sure that
- * the start address is in range so that the lookup
- * below will succeed.
- */
- s = entry->vme_start < start? start: entry->vme_start;
+ if (object == VM_OBJECT_NULL)
+ return KERN_INVALID_OBJECT;
- entry->needs_wakeup = TRUE;
+ if (object == kernel_object) {
+ printf("Warning: Attempt to map kernel object"
+ " by a non-private kernel entity\n");
+ return KERN_INVALID_OBJECT;
+ }
+
+ vm_object_lock(object);
+ object->ref_count++;
+ vm_object_res_reference(object);
+
+ /*
+ * For "named" VM objects, let the pager know that the
+ * memory object is being mapped. Some pagers need to keep
+ * track of this, to know when they can reclaim the memory
+ * object, for example.
+ * VM calls memory_object_map() for each mapping (specifying
+ * the protection of each mapping) and calls
+ * memory_object_last_unmap() when all the mappings are gone.
+ */
+ pager_prot = max_protection;
+ if (copy) {
+ pager_prot &= ~VM_PROT_WRITE;
+ }
+ pager = object->pager;
+ if (object->named &&
+ pager != MEMORY_OBJECT_NULL &&
+ object->copy_strategy != MEMORY_OBJECT_COPY_NONE) {
+ assert(object->pager_ready);
+ vm_object_mapping_wait(object, THREAD_UNINT);
+ vm_object_mapping_begin(object);
+ vm_object_unlock(object);
+
+ kr = memory_object_map(pager, pager_prot);
+ assert(kr == KERN_SUCCESS);
+
+ vm_object_lock(object);
+ vm_object_mapping_end(object);
+ }
+ vm_object_unlock(object);
+
+ /*
+ * Perform the copy if requested
+ */
+
+ if (copy) {
+ vm_object_t new_object;
+ vm_object_offset_t new_offset;
+
+ result = vm_object_copy_strategically(object, offset, size,
+ &new_object, &new_offset,
+ ©);
+
+
+ if (result == KERN_MEMORY_RESTART_COPY) {
+ boolean_t success;
+ boolean_t src_needs_copy;
/*
- * wake up anybody waiting on entries that we have
- * already wired.
- */
- if (need_wakeup) {
- vm_map_entry_wakeup(map);
- need_wakeup = FALSE;
- }
- /*
- * User wiring is interruptible
- */
- wait_result = vm_map_entry_wait(map,
- (user_wire) ? THREAD_ABORTSAFE :
- THREAD_UNINT);
- if (user_wire && wait_result == THREAD_INTERRUPTED) {
- /*
- * undo the wirings we have done so far
- * We do not clear the needs_wakeup flag,
- * because we cannot tell if we were the
- * only one waiting.
- */
- vm_map_unlock(map);
- vm_map_unwire(map, start, s, user_wire);
- return(KERN_FAILURE);
- }
-
- /*
- * Cannot avoid a lookup here. reset timestamp.
+ * XXX
+ * We currently ignore src_needs_copy.
+ * This really is the issue of how to make
+ * MEMORY_OBJECT_COPY_SYMMETRIC safe for
+ * non-kernel users to use. Solution forthcoming.
+ * In the meantime, since we don't allow non-kernel
+ * memory managers to specify symmetric copy,
+ * we won't run into problems here.
*/
- last_timestamp = map->timestamp;
+ new_object = object;
+ new_offset = offset;
+ success = vm_object_copy_quickly(&new_object,
+ new_offset, size,
+ &src_needs_copy,
+ ©);
+ assert(success);
+ result = KERN_SUCCESS;
+ }
+ /*
+ * Throw away the reference to the
+ * original object, as it won't be mapped.
+ */
- /*
- * The entry could have been clipped, look it up again.
- * Worse that can happen is, it may not exist anymore.
- */
- if (!vm_map_lookup_entry(map, s, &first_entry)) {
- if (!user_wire)
- panic("vm_map_wire: re-lookup failed");
+ vm_object_deallocate(object);
- /*
- * User: undo everything upto the previous
- * entry. let vm_map_unwire worry about
- * checking the validity of the range.
- */
- vm_map_unlock(map);
- vm_map_unwire(map, start, s, user_wire);
- return(KERN_FAILURE);
- }
- entry = first_entry;
- continue;
+ if (result != KERN_SUCCESS) {
+ return result;
}
-
- if(entry->is_sub_map) {
- vm_offset_t sub_start;
- vm_offset_t sub_end;
- vm_offset_t local_start;
- vm_offset_t local_end;
- pmap_t pmap;
-
- vm_map_clip_start(map, entry, start);
- vm_map_clip_end(map, entry, end);
- sub_start = entry->offset;
- sub_end = entry->vme_end - entry->vme_start;
- sub_end += entry->offset;
-
- local_end = entry->vme_end;
- if(map_pmap == NULL) {
- if(entry->use_pmap) {
- pmap = entry->object.sub_map->pmap;
- /* ppc implementation requires that */
- /* submaps pmap address ranges line */
- /* up with parent map */
-#ifdef notdef
- pmap_addr = sub_start;
-#endif
- pmap_addr = start;
- } else {
- pmap = map->pmap;
- pmap_addr = start;
- }
- if (entry->wired_count) {
- if (entry->wired_count
- >= MAX_WIRE_COUNT)
- panic("vm_map_wire: too many wirings");
-
- if (user_wire &&
- entry->user_wired_count
- >= MAX_WIRE_COUNT) {
- vm_map_unlock(map);
- vm_map_unwire(map, start,
- entry->vme_start, user_wire);
- return(KERN_FAILURE);
- }
- if(user_wire)
- entry->user_wired_count++;
- if((!user_wire) ||
- (entry->user_wired_count == 0))
- entry->wired_count++;
- entry = entry->vme_next;
- continue;
+ object = new_object;
+ offset = new_offset;
+ }
- } else {
- vm_object_t object;
- vm_object_offset_t offset_hi;
- vm_object_offset_t offset_lo;
- vm_object_offset_t offset;
- vm_prot_t prot;
- boolean_t wired;
- vm_behavior_t behavior;
- vm_map_entry_t local_entry;
- vm_map_version_t version;
- vm_map_t lookup_map;
-
- /* call vm_map_lookup_locked to */
- /* cause any needs copy to be */
- /* evaluated */
- local_start = entry->vme_start;
- lookup_map = map;
- vm_map_lock_write_to_read(map);
- if(vm_map_lookup_locked(
- &lookup_map, local_start,
- access_type,
- &version, &object,
- &offset, &prot, &wired,
- &behavior, &offset_lo,
- &offset_hi, &pmap_map)) {
-
- vm_map_unlock(lookup_map);
- vm_map_unwire(map, start,
- entry->vme_start, user_wire);
- return(KERN_FAILURE);
- }
- if(pmap_map != lookup_map)
- vm_map_unlock(pmap_map);
- vm_map_unlock_read(lookup_map);
- vm_map_lock(map);
- vm_object_unlock(object);
+ {
+ result = vm_map_enter(target_map,
+ &map_addr, map_size,
+ (vm_map_offset_t)mask,
+ flags,
+ object, offset,
+ copy,
+ cur_protection, max_protection,
+ inheritance);
+ }
+ if (result != KERN_SUCCESS)
+ vm_object_deallocate(object);
+ *address = map_addr;
- if (!vm_map_lookup_entry(map,
- local_start, &local_entry)) {
- vm_map_unlock(map);
- vm_map_unwire(map, start,
- entry->vme_start, user_wire);
- return(KERN_FAILURE);
- }
- /* did we have a change of type? */
- if (!local_entry->is_sub_map) {
- last_timestamp = map->timestamp;
- continue;
- }
- entry = local_entry;
- if (user_wire)
- entry->user_wired_count++;
- if((!user_wire) ||
- (entry->user_wired_count == 1))
- entry->wired_count++;
+ return result;
+}
- entry->in_transition = TRUE;
- vm_map_unlock(map);
- rc = vm_map_wire_nested(
- entry->object.sub_map,
- sub_start, sub_end,
- access_type,
- user_wire, pmap, pmap_addr);
- vm_map_lock(map);
- }
- } else {
- local_start = entry->vme_start;
- if (user_wire)
- entry->user_wired_count++;
- if((!user_wire) ||
- (entry->user_wired_count == 1))
- entry->wired_count++;
- vm_map_unlock(map);
- rc = vm_map_wire_nested(entry->object.sub_map,
- sub_start, sub_end,
- access_type,
- user_wire, map_pmap, pmap_addr);
- vm_map_lock(map);
- }
- s = entry->vme_start;
- e = entry->vme_end;
+#if VM_CPM
- /*
- * Find the entry again. It could have been clipped
- * after we unlocked the map.
- */
- if (!vm_map_lookup_entry(map, local_start,
- &first_entry))
- panic("vm_map_wire: re-lookup failed");
- entry = first_entry;
+#ifdef MACH_ASSERT
+extern pmap_paddr_t avail_start, avail_end;
+#endif
- last_timestamp = map->timestamp;
- while ((entry != vm_map_to_entry(map)) &&
- (entry->vme_start < e)) {
- assert(entry->in_transition);
- entry->in_transition = FALSE;
- if (entry->needs_wakeup) {
- entry->needs_wakeup = FALSE;
- need_wakeup = TRUE;
- }
- if (rc != KERN_SUCCESS) {/* from vm_*_wire */
- if (user_wire)
- entry->user_wired_count--;
- if ((!user_wire) ||
- (entry->user_wired_count == 0))
- entry->wired_count--;
- }
- entry = entry->vme_next;
- }
- if (rc != KERN_SUCCESS) { /* from vm_*_wire */
- vm_map_unlock(map);
- if (need_wakeup)
- vm_map_entry_wakeup(map);
- /*
- * undo everything upto the previous entry.
- */
- (void)vm_map_unwire(map, start, s, user_wire);
- return rc;
- }
- continue;
- }
+/*
+ * Allocate memory in the specified map, with the caveat that
+ * the memory is physically contiguous. This call may fail
+ * if the system can't find sufficient contiguous memory.
+ * This call may cause or lead to heart-stopping amounts of
+ * paging activity.
+ *
+ * Memory obtained from this call should be freed in the
+ * normal way, viz., via vm_deallocate.
+ */
+kern_return_t
+vm_map_enter_cpm(
+ vm_map_t map,
+ vm_map_offset_t *addr,
+ vm_map_size_t size,
+ int flags)
+{
+ vm_object_t cpm_obj;
+ pmap_t pmap;
+ vm_page_t m, pages;
+ kern_return_t kr;
+ vm_map_offset_t va, start, end, offset;
+#if MACH_ASSERT
+ vm_map_offset_t prev_addr = 0;
+#endif /* MACH_ASSERT */
- /*
- * If this entry is already wired then increment
- * the appropriate wire reference count.
- */
- if (entry->wired_count) {
- /* sanity check: wired_count is a short */
- if (entry->wired_count >= MAX_WIRE_COUNT)
- panic("vm_map_wire: too many wirings");
+ boolean_t anywhere = ((VM_FLAGS_ANYWHERE & flags) != 0);
+ vm_tag_t tag;
- if (user_wire &&
- entry->user_wired_count >= MAX_WIRE_COUNT) {
- vm_map_unlock(map);
- vm_map_unwire(map, start,
- entry->vme_start, user_wire);
- return(KERN_FAILURE);
- }
- /*
- * entry is already wired down, get our reference
- * after clipping to our range.
- */
- vm_map_clip_start(map, entry, start);
- vm_map_clip_end(map, entry, end);
- if (user_wire)
- entry->user_wired_count++;
- if ((!user_wire) || (entry->user_wired_count == 1))
- entry->wired_count++;
+ VM_GET_FLAGS_ALIAS(flags, tag);
- entry = entry->vme_next;
- continue;
- }
+ if (size == 0) {
+ *addr = 0;
+ return KERN_SUCCESS;
+ }
+ if (anywhere)
+ *addr = vm_map_min(map);
+ else
+ *addr = vm_map_trunc_page(*addr,
+ VM_MAP_PAGE_MASK(map));
+ size = vm_map_round_page(size,
+ VM_MAP_PAGE_MASK(map));
- /*
- * Unwired entry or wire request transmitted via submap
- */
+ /*
+ * LP64todo - cpm_allocate should probably allow
+ * allocations of >4GB, but not with the current
+ * algorithm, so just cast down the size for now.
+ */
+ if (size > VM_MAX_ADDRESS)
+ return KERN_RESOURCE_SHORTAGE;
+ if ((kr = cpm_allocate(CAST_DOWN(vm_size_t, size),
+ &pages, 0, 0, TRUE, flags)) != KERN_SUCCESS)
+ return kr;
+
+ cpm_obj = vm_object_allocate((vm_object_size_t)size);
+ assert(cpm_obj != VM_OBJECT_NULL);
+ assert(cpm_obj->internal);
+ assert(cpm_obj->vo_size == (vm_object_size_t)size);
+ assert(cpm_obj->can_persist == FALSE);
+ assert(cpm_obj->pager_created == FALSE);
+ assert(cpm_obj->pageout == FALSE);
+ assert(cpm_obj->shadow == VM_OBJECT_NULL);
+ /*
+ * Insert pages into object.
+ */
+ vm_object_lock(cpm_obj);
+ for (offset = 0; offset < size; offset += PAGE_SIZE) {
+ m = pages;
+ pages = NEXT_PAGE(m);
+ *(NEXT_PAGE_PTR(m)) = VM_PAGE_NULL;
+
+ assert(!m->gobbled);
+ assert(!m->wanted);
+ assert(!m->pageout);
+ assert(!m->tabled);
+ assert(VM_PAGE_WIRED(m));
/*
- * Perform actions of vm_map_lookup that need the write
- * lock on the map: create a shadow object for a
- * copy-on-write region, or an object for a zero-fill
- * region.
- */
- size = entry->vme_end - entry->vme_start;
- /*
- * If wiring a copy-on-write page, we need to copy it now
- * even if we're only (currently) requesting read access.
- * This is aggressive, but once it's wired we can't move it.
+ * ENCRYPTED SWAP:
+ * "m" is not supposed to be pageable, so it
+ * should not be encrypted. It wouldn't be safe
+ * to enter it in a new VM object while encrypted.
*/
- if (entry->needs_copy) {
- vm_object_shadow(&entry->object.vm_object,
- &entry->offset, size);
- entry->needs_copy = FALSE;
- } else if (entry->object.vm_object == VM_OBJECT_NULL) {
- entry->object.vm_object = vm_object_allocate(size);
- entry->offset = (vm_object_offset_t)0;
- }
+ ASSERT_PAGE_DECRYPTED(m);
+ assert(m->busy);
+ assert(m->phys_page>=(avail_start>>PAGE_SHIFT) && m->phys_page<=(avail_end>>PAGE_SHIFT));
- vm_map_clip_start(map, entry, start);
- vm_map_clip_end(map, entry, end);
+ m->busy = FALSE;
+ vm_page_insert(m, cpm_obj, offset);
+ }
+ assert(cpm_obj->resident_page_count == size / PAGE_SIZE);
+ vm_object_unlock(cpm_obj);
- s = entry->vme_start;
- e = entry->vme_end;
+ /*
+ * Hang onto a reference on the object in case a
+ * multi-threaded application for some reason decides
+ * to deallocate the portion of the address space into
+ * which we will insert this object.
+ *
+ * Unfortunately, we must insert the object now before
+ * we can talk to the pmap module about which addresses
+ * must be wired down. Hence, the race with a multi-
+ * threaded app.
+ */
+ vm_object_reference(cpm_obj);
+
+ /*
+ * Insert object into map.
+ */
+ kr = vm_map_enter(
+ map,
+ addr,
+ size,
+ (vm_map_offset_t)0,
+ flags,
+ cpm_obj,
+ (vm_object_offset_t)0,
+ FALSE,
+ VM_PROT_ALL,
+ VM_PROT_ALL,
+ VM_INHERIT_DEFAULT);
+
+ if (kr != KERN_SUCCESS) {
/*
- * Check for holes and protection mismatch.
- * Holes: Next entry should be contiguous unless this
- * is the end of the region.
- * Protection: Access requested must be allowed, unless
- * wiring is by protection class
+ * A CPM object doesn't have can_persist set,
+ * so all we have to do is deallocate it to
+ * free up these pages.
*/
- if ((((entry->vme_end < end) &&
- ((entry->vme_next == vm_map_to_entry(map)) ||
- (entry->vme_next->vme_start > entry->vme_end))) ||
- ((entry->protection & access_type) != access_type))) {
- /*
- * Found a hole or protection problem.
- * Unwire the region we wired so far.
- */
- if (start != entry->vme_start) {
- vm_map_unlock(map);
- vm_map_unwire(map, start, s, user_wire);
- } else {
- vm_map_unlock(map);
- }
- return((entry->protection&access_type) != access_type?
- KERN_PROTECTION_FAILURE: KERN_INVALID_ADDRESS);
- }
+ assert(cpm_obj->pager_created == FALSE);
+ assert(cpm_obj->can_persist == FALSE);
+ assert(cpm_obj->pageout == FALSE);
+ assert(cpm_obj->shadow == VM_OBJECT_NULL);
+ vm_object_deallocate(cpm_obj); /* kill acquired ref */
+ vm_object_deallocate(cpm_obj); /* kill creation ref */
+ }
- assert(entry->wired_count == 0 && entry->user_wired_count == 0);
+ /*
+ * Inform the physical mapping system that the
+ * range of addresses may not fault, so that
+ * page tables and such can be locked down as well.
+ */
+ start = *addr;
+ end = start + size;
+ pmap = vm_map_pmap(map);
+ pmap_pageable(pmap, start, end, FALSE);
- if (user_wire)
- entry->user_wired_count++;
- if ((!user_wire) || (entry->user_wired_count == 1))
- entry->wired_count++;
+ /*
+ * Enter each page into the pmap, to avoid faults.
+ * Note that this loop could be coded more efficiently,
+ * if the need arose, rather than looking up each page
+ * again.
+ */
+ for (offset = 0, va = start; offset < size;
+ va += PAGE_SIZE, offset += PAGE_SIZE) {
+ int type_of_fault;
- entry->in_transition = TRUE;
+ vm_object_lock(cpm_obj);
+ m = vm_page_lookup(cpm_obj, (vm_object_offset_t)offset);
+ assert(m != VM_PAGE_NULL);
- /*
- * This entry might get split once we unlock the map.
- * In vm_fault_wire(), we need the current range as
- * defined by this entry. In order for this to work
- * along with a simultaneous clip operation, we make a
- * temporary copy of this entry and use that for the
- * wiring. Note that the underlying objects do not
- * change during a clip.
- */
- tmp_entry = *entry;
+ vm_page_zero_fill(m);
- /*
- * The in_transition state guarentees that the entry
- * (or entries for this range, if split occured) will be
- * there when the map lock is acquired for the second time.
- */
- vm_map_unlock(map);
+ type_of_fault = DBG_ZERO_FILL_FAULT;
- if (!user_wire && cur_thread != THREAD_NULL)
- interruptible_state = thread_interrupt_level(THREAD_UNINT);
+ vm_fault_enter(m, pmap, va, VM_PROT_ALL, VM_PROT_WRITE,
+ VM_PAGE_WIRED(m), FALSE, FALSE, FALSE, 0, NULL,
+ &type_of_fault);
- if(map_pmap)
- rc = vm_fault_wire(map,
- &tmp_entry, map_pmap, pmap_addr);
- else
- rc = vm_fault_wire(map,
- &tmp_entry, map->pmap,
- tmp_entry.vme_start);
+ vm_object_unlock(cpm_obj);
+ }
- if (!user_wire && cur_thread != THREAD_NULL)
- thread_interrupt_level(interruptible_state);
+#if MACH_ASSERT
+ /*
+ * Verify ordering in address space.
+ */
+ for (offset = 0; offset < size; offset += PAGE_SIZE) {
+ vm_object_lock(cpm_obj);
+ m = vm_page_lookup(cpm_obj, (vm_object_offset_t)offset);
+ vm_object_unlock(cpm_obj);
+ if (m == VM_PAGE_NULL)
+ panic("vm_allocate_cpm: obj %p off 0x%llx no page",
+ cpm_obj, (uint64_t)offset);
+ assert(m->tabled);
+ assert(!m->busy);
+ assert(!m->wanted);
+ assert(!m->fictitious);
+ assert(!m->private);
+ assert(!m->absent);
+ assert(!m->error);
+ assert(!m->cleaning);
+ assert(!m->laundry);
+ assert(!m->precious);
+ assert(!m->clustered);
+ if (offset != 0) {
+ if (m->phys_page != prev_addr + 1) {
+ printf("start 0x%llx end 0x%llx va 0x%llx\n",
+ (uint64_t)start, (uint64_t)end, (uint64_t)va);
+ printf("obj %p off 0x%llx\n", cpm_obj, (uint64_t)offset);
+ printf("m %p prev_address 0x%llx\n", m, (uint64_t)prev_addr);
+ panic("vm_allocate_cpm: pages not contig!");
+ }
+ }
+ prev_addr = m->phys_page;
+ }
+#endif /* MACH_ASSERT */
- vm_map_lock(map);
+ vm_object_deallocate(cpm_obj); /* kill extra ref */
- if (last_timestamp+1 != map->timestamp) {
- /*
- * Find the entry again. It could have been clipped
- * after we unlocked the map.
- */
- if (!vm_map_lookup_entry(map, tmp_entry.vme_start,
- &first_entry))
- panic("vm_map_wire: re-lookup failed");
+ return kr;
+}
- entry = first_entry;
- }
- last_timestamp = map->timestamp;
+#else /* VM_CPM */
- while ((entry != vm_map_to_entry(map)) &&
- (entry->vme_start < tmp_entry.vme_end)) {
- assert(entry->in_transition);
- entry->in_transition = FALSE;
- if (entry->needs_wakeup) {
- entry->needs_wakeup = FALSE;
- need_wakeup = TRUE;
- }
- if (rc != KERN_SUCCESS) { /* from vm_*_wire */
- if (user_wire)
- entry->user_wired_count--;
- if ((!user_wire) ||
- (entry->user_wired_count == 0))
- entry->wired_count--;
- }
- entry = entry->vme_next;
- }
-
- if (rc != KERN_SUCCESS) { /* from vm_*_wire */
- vm_map_unlock(map);
- if (need_wakeup)
- vm_map_entry_wakeup(map);
- /*
- * undo everything upto the previous entry.
- */
- (void)vm_map_unwire(map, start, s, user_wire);
- return rc;
- }
- } /* end while loop through map entries */
- vm_map_unlock(map);
-
- /*
- * wake up anybody waiting on entries we wired.
- */
- if (need_wakeup)
- vm_map_entry_wakeup(map);
-
- return(KERN_SUCCESS);
-
-}
+/*
+ * Interface is defined in all cases, but unless the kernel
+ * is built explicitly for this option, the interface does
+ * nothing.
+ */
kern_return_t
-vm_map_wire(
- register vm_map_t map,
- register vm_offset_t start,
- register vm_offset_t end,
- register vm_prot_t access_type,
- boolean_t user_wire)
+vm_map_enter_cpm(
+ __unused vm_map_t map,
+ __unused vm_map_offset_t *addr,
+ __unused vm_map_size_t size,
+ __unused int flags)
{
-
- kern_return_t kret;
-
-#ifdef ppc
- /*
- * the calls to mapping_prealloc and mapping_relpre
- * (along with the VM_MAP_RANGE_CHECK to insure a
- * resonable range was passed in) are
- * currently necessary because
- * we haven't enabled kernel pre-emption
- * and/or the pmap_enter cannot purge and re-use
- * existing mappings
- */
- VM_MAP_RANGE_CHECK(map, start, end);
- mapping_prealloc(end - start);
-#endif
- kret = vm_map_wire_nested(map, start, end, access_type,
- user_wire, (pmap_t)NULL, 0);
-#ifdef ppc
- mapping_relpre();
-#endif
- return kret;
+ return KERN_FAILURE;
}
+#endif /* VM_CPM */
+/* Not used without nested pmaps */
+#ifndef NO_NESTED_PMAP
/*
- * vm_map_unwire:
- *
- * Sets the pageability of the specified address range in the target
- * as pageable. Regions specified must have been wired previously.
- *
- * The map must not be locked, but a reference must remain to the map
- * throughout the call.
- *
- * Kernel will panic on failures. User unwire ignores holes and
- * unwired and intransition entries to avoid losing memory by leaving
- * it unwired.
+ * Clip and unnest a portion of a nested submap mapping.
*/
-kern_return_t
-vm_map_unwire_nested(
- register vm_map_t map,
- register vm_offset_t start,
- register vm_offset_t end,
- boolean_t user_wire,
- pmap_t map_pmap,
- vm_offset_t pmap_addr)
-{
- register vm_map_entry_t entry;
- struct vm_map_entry *first_entry, tmp_entry;
- boolean_t need_wakeup;
- boolean_t main_map = FALSE;
- unsigned int last_timestamp;
- vm_map_lock(map);
- if(map_pmap == NULL)
- main_map = TRUE;
- last_timestamp = map->timestamp;
- VM_MAP_RANGE_CHECK(map, start, end);
- assert(page_aligned(start));
- assert(page_aligned(end));
+static void
+vm_map_clip_unnest(
+ vm_map_t map,
+ vm_map_entry_t entry,
+ vm_map_offset_t start_unnest,
+ vm_map_offset_t end_unnest)
+{
+ vm_map_offset_t old_start_unnest = start_unnest;
+ vm_map_offset_t old_end_unnest = end_unnest;
- if (vm_map_lookup_entry(map, start, &first_entry)) {
- entry = first_entry;
- /* vm_map_clip_start will be done later. */
- }
- else {
- /* Start address is not in map. */
- vm_map_unlock(map);
- return(KERN_INVALID_ADDRESS);
+ assert(entry->is_sub_map);
+ assert(VME_SUBMAP(entry) != NULL);
+ assert(entry->use_pmap);
+
+ /*
+ * Query the platform for the optimal unnest range.
+ * DRK: There's some duplication of effort here, since
+ * callers may have adjusted the range to some extent. This
+ * routine was introduced to support 1GiB subtree nesting
+ * for x86 platforms, which can also nest on 2MiB boundaries
+ * depending on size/alignment.
+ */
+ if (pmap_adjust_unnest_parameters(map->pmap, &start_unnest, &end_unnest)) {
+ log_unnest_badness(map, old_start_unnest, old_end_unnest);
}
- need_wakeup = FALSE;
- while ((entry != vm_map_to_entry(map)) && (entry->vme_start < end)) {
- if (entry->in_transition) {
- /*
- * 1)
- * Another thread is wiring down this entry. Note
- * that if it is not for the other thread we would
- * be unwiring an unwired entry. This is not
- * permitted. If we wait, we will be unwiring memory
- * we did not wire.
- *
- * 2)
- * Another thread is unwiring this entry. We did not
- * have a reference to it, because if we did, this
- * entry will not be getting unwired now.
- */
- if (!user_wire)
- panic("vm_map_unwire: in_transition entry");
+ if (entry->vme_start > start_unnest ||
+ entry->vme_end < end_unnest) {
+ panic("vm_map_clip_unnest(0x%llx,0x%llx): "
+ "bad nested entry: start=0x%llx end=0x%llx\n",
+ (long long)start_unnest, (long long)end_unnest,
+ (long long)entry->vme_start, (long long)entry->vme_end);
+ }
- entry = entry->vme_next;
- continue;
+ if (start_unnest > entry->vme_start) {
+ _vm_map_clip_start(&map->hdr,
+ entry,
+ start_unnest);
+ if (map->holelistenabled) {
+ vm_map_store_update_first_free(map, NULL, FALSE);
+ } else {
+ vm_map_store_update_first_free(map, map->first_free, FALSE);
}
+ }
+ if (entry->vme_end > end_unnest) {
+ _vm_map_clip_end(&map->hdr,
+ entry,
+ end_unnest);
+ if (map->holelistenabled) {
+ vm_map_store_update_first_free(map, NULL, FALSE);
+ } else {
+ vm_map_store_update_first_free(map, map->first_free, FALSE);
+ }
+ }
- if(entry->is_sub_map) {
- vm_offset_t sub_start;
- vm_offset_t sub_end;
- vm_offset_t local_end;
- pmap_t pmap;
-
-
- vm_map_clip_start(map, entry, start);
- vm_map_clip_end(map, entry, end);
+ pmap_unnest(map->pmap,
+ entry->vme_start,
+ entry->vme_end - entry->vme_start);
+ if ((map->mapped_in_other_pmaps) && (map->ref_count)) {
+ /* clean up parent map/maps */
+ vm_map_submap_pmap_clean(
+ map, entry->vme_start,
+ entry->vme_end,
+ VME_SUBMAP(entry),
+ VME_OFFSET(entry));
+ }
+ entry->use_pmap = FALSE;
+ if ((map->pmap != kernel_pmap) &&
+ (VME_ALIAS(entry) == VM_MEMORY_SHARED_PMAP)) {
+ VME_ALIAS_SET(entry, VM_MEMORY_UNSHARED_PMAP);
+ }
+}
+#endif /* NO_NESTED_PMAP */
- sub_start = entry->offset;
- sub_end = entry->vme_end - entry->vme_start;
- sub_end += entry->offset;
- local_end = entry->vme_end;
- if(map_pmap == NULL) {
- if(entry->use_pmap) {
- pmap = entry->object.sub_map->pmap;
- pmap_addr = sub_start;
- } else {
- pmap = map->pmap;
- pmap_addr = start;
- }
- if (entry->wired_count == 0 ||
- (user_wire && entry->user_wired_count == 0)) {
- if (!user_wire)
- panic("vm_map_unwire: entry is unwired");
- entry = entry->vme_next;
- continue;
- }
-
- /*
- * Check for holes
- * Holes: Next entry should be contiguous unless
- * this is the end of the region.
- */
- if (((entry->vme_end < end) &&
- ((entry->vme_next == vm_map_to_entry(map)) ||
- (entry->vme_next->vme_start
- > entry->vme_end)))) {
- if (!user_wire)
- panic("vm_map_unwire: non-contiguous region");
/*
- entry = entry->vme_next;
- continue;
-*/
- }
+ * vm_map_clip_start: [ internal use only ]
+ *
+ * Asserts that the given entry begins at or after
+ * the specified address; if necessary,
+ * it splits the entry into two.
+ */
+void
+vm_map_clip_start(
+ vm_map_t map,
+ vm_map_entry_t entry,
+ vm_map_offset_t startaddr)
+{
+#ifndef NO_NESTED_PMAP
+ if (entry->is_sub_map &&
+ entry->use_pmap &&
+ startaddr >= entry->vme_start) {
+ vm_map_offset_t start_unnest, end_unnest;
- if (!user_wire || (--entry->user_wired_count == 0))
- entry->wired_count--;
+ /*
+ * Make sure "startaddr" is no longer in a nested range
+ * before we clip. Unnest only the minimum range the platform
+ * can handle.
+ * vm_map_clip_unnest may perform additional adjustments to
+ * the unnest range.
+ */
+ start_unnest = startaddr & ~(pmap_nesting_size_min - 1);
+ end_unnest = start_unnest + pmap_nesting_size_min;
+ vm_map_clip_unnest(map, entry, start_unnest, end_unnest);
+ }
+#endif /* NO_NESTED_PMAP */
+ if (startaddr > entry->vme_start) {
+ if (VME_OBJECT(entry) &&
+ !entry->is_sub_map &&
+ VME_OBJECT(entry)->phys_contiguous) {
+ pmap_remove(map->pmap,
+ (addr64_t)(entry->vme_start),
+ (addr64_t)(entry->vme_end));
+ }
+ _vm_map_clip_start(&map->hdr, entry, startaddr);
+ if (map->holelistenabled) {
+ vm_map_store_update_first_free(map, NULL, FALSE);
+ } else {
+ vm_map_store_update_first_free(map, map->first_free, FALSE);
+ }
+ }
+}
- if (entry->wired_count != 0) {
- entry = entry->vme_next;
- continue;
- }
- entry->in_transition = TRUE;
- tmp_entry = *entry;/* see comment in vm_map_wire() */
+#define vm_map_copy_clip_start(copy, entry, startaddr) \
+ MACRO_BEGIN \
+ if ((startaddr) > (entry)->vme_start) \
+ _vm_map_clip_start(&(copy)->cpy_hdr,(entry),(startaddr)); \
+ MACRO_END
- /*
- * We can unlock the map now. The in_transition state
- * guarantees existance of the entry.
- */
- vm_map_unlock(map);
- vm_map_unwire_nested(entry->object.sub_map,
- sub_start, sub_end, user_wire, pmap, pmap_addr);
- vm_map_lock(map);
+/*
+ * This routine is called only when it is known that
+ * the entry must be split.
+ */
+static void
+_vm_map_clip_start(
+ register struct vm_map_header *map_header,
+ register vm_map_entry_t entry,
+ register vm_map_offset_t start)
+{
+ register vm_map_entry_t new_entry;
- if (last_timestamp+1 != map->timestamp) {
- /*
- * Find the entry again. It could have been
- * clipped or deleted after we unlocked the map.
- */
- if (!vm_map_lookup_entry(map,
- tmp_entry.vme_start,
- &first_entry)) {
- if (!user_wire)
- panic("vm_map_unwire: re-lookup failed");
- entry = first_entry->vme_next;
- } else
- entry = first_entry;
- }
- last_timestamp = map->timestamp;
-
- /*
- * clear transition bit for all constituent entries
- * that were in the original entry (saved in
- * tmp_entry). Also check for waiters.
- */
- while ((entry != vm_map_to_entry(map)) &&
- (entry->vme_start < tmp_entry.vme_end)) {
- assert(entry->in_transition);
- entry->in_transition = FALSE;
- if (entry->needs_wakeup) {
- entry->needs_wakeup = FALSE;
- need_wakeup = TRUE;
- }
- entry = entry->vme_next;
- }
- continue;
- } else {
- vm_map_unlock(map);
- vm_map_unwire_nested(entry->object.sub_map,
- sub_start, sub_end, user_wire, map_pmap,
- pmap_addr);
- vm_map_lock(map);
+ /*
+ * Split off the front portion --
+ * note that we must insert the new
+ * entry BEFORE this one, so that
+ * this entry has the specified starting
+ * address.
+ */
- if (last_timestamp+1 != map->timestamp) {
- /*
- * Find the entry again. It could have been
- * clipped or deleted after we unlocked the map.
- */
- if (!vm_map_lookup_entry(map,
- tmp_entry.vme_start,
- &first_entry)) {
- if (!user_wire)
- panic("vm_map_unwire: re-lookup failed");
- entry = first_entry->vme_next;
- } else
- entry = first_entry;
- }
- last_timestamp = map->timestamp;
- }
- }
+ if (entry->map_aligned) {
+ assert(VM_MAP_PAGE_ALIGNED(start,
+ VM_MAP_HDR_PAGE_MASK(map_header)));
+ }
+ new_entry = _vm_map_entry_create(map_header, !map_header->entries_pageable);
+ vm_map_entry_copy_full(new_entry, entry);
- if ((entry->wired_count == 0) ||
- (user_wire && entry->user_wired_count == 0)) {
- if (!user_wire)
- panic("vm_map_unwire: entry is unwired");
+ new_entry->vme_end = start;
+ assert(new_entry->vme_start < new_entry->vme_end);
+ VME_OFFSET_SET(entry, VME_OFFSET(entry) + (start - entry->vme_start));
+ assert(start < entry->vme_end);
+ entry->vme_start = start;
- entry = entry->vme_next;
- continue;
- }
-
- assert(entry->wired_count > 0 &&
- (!user_wire || entry->user_wired_count > 0));
+ _vm_map_store_entry_link(map_header, entry->vme_prev, new_entry);
+
+ if (entry->is_sub_map)
+ vm_map_reference(VME_SUBMAP(new_entry));
+ else
+ vm_object_reference(VME_OBJECT(new_entry));
+}
- vm_map_clip_start(map, entry, start);
- vm_map_clip_end(map, entry, end);
+/*
+ * vm_map_clip_end: [ internal use only ]
+ *
+ * Asserts that the given entry ends at or before
+ * the specified address; if necessary,
+ * it splits the entry into two.
+ */
+void
+vm_map_clip_end(
+ vm_map_t map,
+ vm_map_entry_t entry,
+ vm_map_offset_t endaddr)
+{
+ if (endaddr > entry->vme_end) {
/*
- * Check for holes
- * Holes: Next entry should be contiguous unless
- * this is the end of the region.
+ * Within the scope of this clipping, limit "endaddr" to
+ * the end of this map entry...
*/
- if (((entry->vme_end < end) &&
- ((entry->vme_next == vm_map_to_entry(map)) ||
- (entry->vme_next->vme_start > entry->vme_end)))) {
-
- if (!user_wire)
- panic("vm_map_unwire: non-contiguous region");
- entry = entry->vme_next;
- continue;
- }
-
- if (!user_wire || (--entry->user_wired_count == 0))
- entry->wired_count--;
-
- if (entry->wired_count != 0) {
- entry = entry->vme_next;
- continue;
- }
-
- entry->in_transition = TRUE;
- tmp_entry = *entry; /* see comment in vm_map_wire() */
+ endaddr = entry->vme_end;
+ }
+#ifndef NO_NESTED_PMAP
+ if (entry->is_sub_map && entry->use_pmap) {
+ vm_map_offset_t start_unnest, end_unnest;
/*
- * We can unlock the map now. The in_transition state
- * guarantees existance of the entry.
+ * Make sure the range between the start of this entry and
+ * the new "endaddr" is no longer nested before we clip.
+ * Unnest only the minimum range the platform can handle.
+ * vm_map_clip_unnest may perform additional adjustments to
+ * the unnest range.
*/
- vm_map_unlock(map);
- if(map_pmap) {
- vm_fault_unwire(map,
- &tmp_entry, FALSE, map_pmap, pmap_addr);
+ start_unnest = entry->vme_start;
+ end_unnest =
+ (endaddr + pmap_nesting_size_min - 1) &
+ ~(pmap_nesting_size_min - 1);
+ vm_map_clip_unnest(map, entry, start_unnest, end_unnest);
+ }
+#endif /* NO_NESTED_PMAP */
+ if (endaddr < entry->vme_end) {
+ if (VME_OBJECT(entry) &&
+ !entry->is_sub_map &&
+ VME_OBJECT(entry)->phys_contiguous) {
+ pmap_remove(map->pmap,
+ (addr64_t)(entry->vme_start),
+ (addr64_t)(entry->vme_end));
+ }
+ _vm_map_clip_end(&map->hdr, entry, endaddr);
+ if (map->holelistenabled) {
+ vm_map_store_update_first_free(map, NULL, FALSE);
} else {
- vm_fault_unwire(map,
- &tmp_entry, FALSE, map->pmap,
- tmp_entry.vme_start);
+ vm_map_store_update_first_free(map, map->first_free, FALSE);
}
- vm_map_lock(map);
+ }
+}
- if (last_timestamp+1 != map->timestamp) {
- /*
- * Find the entry again. It could have been clipped
- * or deleted after we unlocked the map.
- */
- if (!vm_map_lookup_entry(map, tmp_entry.vme_start,
- &first_entry)) {
- if (!user_wire)
- panic("vm_map_unwire: re-lookup failed");
- entry = first_entry->vme_next;
- } else
- entry = first_entry;
- }
- last_timestamp = map->timestamp;
- /*
- * clear transition bit for all constituent entries that
- * were in the original entry (saved in tmp_entry). Also
- * check for waiters.
- */
- while ((entry != vm_map_to_entry(map)) &&
- (entry->vme_start < tmp_entry.vme_end)) {
- assert(entry->in_transition);
- entry->in_transition = FALSE;
- if (entry->needs_wakeup) {
- entry->needs_wakeup = FALSE;
- need_wakeup = TRUE;
- }
- entry = entry->vme_next;
- }
- }
- vm_map_unlock(map);
+#define vm_map_copy_clip_end(copy, entry, endaddr) \
+ MACRO_BEGIN \
+ if ((endaddr) < (entry)->vme_end) \
+ _vm_map_clip_end(&(copy)->cpy_hdr,(entry),(endaddr)); \
+ MACRO_END
+
+/*
+ * This routine is called only when it is known that
+ * the entry must be split.
+ */
+static void
+_vm_map_clip_end(
+ register struct vm_map_header *map_header,
+ register vm_map_entry_t entry,
+ register vm_map_offset_t end)
+{
+ register vm_map_entry_t new_entry;
+
/*
- * wake up anybody waiting on entries that we have unwired.
+ * Create a new entry and insert it
+ * AFTER the specified entry
*/
- if (need_wakeup)
- vm_map_entry_wakeup(map);
- return(KERN_SUCCESS);
-}
+ if (entry->map_aligned) {
+ assert(VM_MAP_PAGE_ALIGNED(end,
+ VM_MAP_HDR_PAGE_MASK(map_header)));
+ }
-kern_return_t
-vm_map_unwire(
- register vm_map_t map,
- register vm_offset_t start,
- register vm_offset_t end,
- boolean_t user_wire)
-{
- return vm_map_unwire_nested(map, start, end,
- user_wire, (pmap_t)NULL, 0);
+ new_entry = _vm_map_entry_create(map_header, !map_header->entries_pageable);
+ vm_map_entry_copy_full(new_entry, entry);
+
+ assert(entry->vme_start < end);
+ new_entry->vme_start = entry->vme_end = end;
+ VME_OFFSET_SET(new_entry,
+ VME_OFFSET(new_entry) + (end - entry->vme_start));
+ assert(new_entry->vme_start < new_entry->vme_end);
+
+ _vm_map_store_entry_link(map_header, entry, new_entry);
+
+ if (entry->is_sub_map)
+ vm_map_reference(VME_SUBMAP(new_entry));
+ else
+ vm_object_reference(VME_OBJECT(new_entry));
}
/*
- * vm_map_entry_delete: [ internal use only ]
+ * VM_MAP_RANGE_CHECK: [ internal use only ]
*
- * Deallocate the given entry from the target map.
- */
-void
-vm_map_entry_delete(
+ * Asserts that the starting and ending region
+ * addresses fall within the valid range of the map.
+ */
+#define VM_MAP_RANGE_CHECK(map, start, end) \
+ MACRO_BEGIN \
+ if (start < vm_map_min(map)) \
+ start = vm_map_min(map); \
+ if (end > vm_map_max(map)) \
+ end = vm_map_max(map); \
+ if (start > end) \
+ start = end; \
+ MACRO_END
+
+/*
+ * vm_map_range_check: [ internal use only ]
+ *
+ * Check that the region defined by the specified start and
+ * end addresses are wholly contained within a single map
+ * entry or set of adjacent map entries of the spacified map,
+ * i.e. the specified region contains no unmapped space.
+ * If any or all of the region is unmapped, FALSE is returned.
+ * Otherwise, TRUE is returned and if the output argument 'entry'
+ * is not NULL it points to the map entry containing the start
+ * of the region.
+ *
+ * The map is locked for reading on entry and is left locked.
+ */
+static boolean_t
+vm_map_range_check(
register vm_map_t map,
- register vm_map_entry_t entry)
+ register vm_map_offset_t start,
+ register vm_map_offset_t end,
+ vm_map_entry_t *entry)
{
- register vm_offset_t s, e;
- register vm_object_t object;
- register vm_map_t submap;
- extern vm_object_t kernel_object;
-
- s = entry->vme_start;
- e = entry->vme_end;
- assert(page_aligned(s));
- assert(page_aligned(e));
- assert(entry->wired_count == 0);
- assert(entry->user_wired_count == 0);
-
- if (entry->is_sub_map) {
- object = NULL;
- submap = entry->object.sub_map;
- } else {
- submap = NULL;
- object = entry->object.vm_object;
- }
+ vm_map_entry_t cur;
+ register vm_map_offset_t prev;
- vm_map_entry_unlink(map, entry);
- map->size -= e - s;
+ /*
+ * Basic sanity checks first
+ */
+ if (start < vm_map_min(map) || end > vm_map_max(map) || start > end)
+ return (FALSE);
- vm_map_entry_dispose(map, entry);
+ /*
+ * Check first if the region starts within a valid
+ * mapping for the map.
+ */
+ if (!vm_map_lookup_entry(map, start, &cur))
+ return (FALSE);
- vm_map_unlock(map);
/*
- * Deallocate the object only after removing all
- * pmap entries pointing to its pages.
+ * Optimize for the case that the region is contained
+ * in a single map entry.
*/
- if (submap)
- vm_map_deallocate(submap);
- else
- vm_object_deallocate(object);
+ if (entry != (vm_map_entry_t *) NULL)
+ *entry = cur;
+ if (end <= cur->vme_end)
+ return (TRUE);
+ /*
+ * If the region is not wholly contained within a
+ * single entry, walk the entries looking for holes.
+ */
+ prev = cur->vme_end;
+ cur = cur->vme_next;
+ while ((cur != vm_map_to_entry(map)) && (prev == cur->vme_start)) {
+ if (end <= cur->vme_end)
+ return (TRUE);
+ prev = cur->vme_end;
+ cur = cur->vme_next;
+ }
+ return (FALSE);
}
-void
-vm_map_submap_pmap_clean(
+/*
+ * vm_map_submap: [ kernel use only ]
+ *
+ * Mark the given range as handled by a subordinate map.
+ *
+ * This range must have been created with vm_map_find using
+ * the vm_submap_object, and no other operations may have been
+ * performed on this range prior to calling vm_map_submap.
+ *
+ * Only a limited number of operations can be performed
+ * within this rage after calling vm_map_submap:
+ * vm_fault
+ * [Don't try vm_map_copyin!]
+ *
+ * To remove a submapping, one must first remove the
+ * range from the superior map, and then destroy the
+ * submap (if desired). [Better yet, don't try it.]
+ */
+kern_return_t
+vm_map_submap(
vm_map_t map,
- vm_offset_t start,
- vm_offset_t end,
- vm_map_t sub_map,
- vm_offset_t offset)
+ vm_map_offset_t start,
+ vm_map_offset_t end,
+ vm_map_t submap,
+ vm_map_offset_t offset,
+#ifdef NO_NESTED_PMAP
+ __unused
+#endif /* NO_NESTED_PMAP */
+ boolean_t use_pmap)
{
- vm_offset_t submap_start;
- vm_offset_t submap_end;
- vm_offset_t addr;
- vm_size_t remove_size;
- vm_map_entry_t entry;
+ vm_map_entry_t entry;
+ register kern_return_t result = KERN_INVALID_ARGUMENT;
+ register vm_object_t object;
- submap_end = offset + (end - start);
- submap_start = offset;
- if(vm_map_lookup_entry(sub_map, offset, &entry)) {
-
- remove_size = (entry->vme_end - entry->vme_start);
- if(offset > entry->vme_start)
- remove_size -= offset - entry->vme_start;
-
+ vm_map_lock(map);
- if(submap_end < entry->vme_end) {
- remove_size -=
- entry->vme_end - submap_end;
- }
- if(entry->is_sub_map) {
- vm_map_submap_pmap_clean(
- sub_map,
- start,
- start + remove_size,
- entry->object.sub_map,
- entry->offset);
- } else {
+ if (! vm_map_lookup_entry(map, start, &entry)) {
+ entry = entry->vme_next;
+ }
- if((map->mapped) && (map->ref_count)
- && (entry->object.vm_object != NULL)) {
- vm_object_pmap_protect(
- entry->object.vm_object,
- entry->offset,
- remove_size,
- PMAP_NULL,
- entry->vme_start,
- VM_PROT_NONE);
- } else {
- pmap_remove(map->pmap,
- (addr64_t)start,
- (addr64_t)(start + remove_size));
- }
- }
+ if (entry == vm_map_to_entry(map) ||
+ entry->is_sub_map) {
+ vm_map_unlock(map);
+ return KERN_INVALID_ARGUMENT;
}
- entry = entry->vme_next;
-
- while((entry != vm_map_to_entry(sub_map))
- && (entry->vme_start < submap_end)) {
- remove_size = (entry->vme_end - entry->vme_start);
- if(submap_end < entry->vme_end) {
- remove_size -= entry->vme_end - submap_end;
+ vm_map_clip_start(map, entry, start);
+ vm_map_clip_end(map, entry, end);
+
+ if ((entry->vme_start == start) && (entry->vme_end == end) &&
+ (!entry->is_sub_map) &&
+ ((object = VME_OBJECT(entry)) == vm_submap_object) &&
+ (object->resident_page_count == 0) &&
+ (object->copy == VM_OBJECT_NULL) &&
+ (object->shadow == VM_OBJECT_NULL) &&
+ (!object->pager_created)) {
+ VME_OFFSET_SET(entry, (vm_object_offset_t)offset);
+ VME_OBJECT_SET(entry, VM_OBJECT_NULL);
+ vm_object_deallocate(object);
+ entry->is_sub_map = TRUE;
+ entry->use_pmap = FALSE;
+ VME_SUBMAP_SET(entry, submap);
+ vm_map_reference(submap);
+ if (submap->mapped_in_other_pmaps == FALSE &&
+ vm_map_pmap(submap) != PMAP_NULL &&
+ vm_map_pmap(submap) != vm_map_pmap(map)) {
+ /*
+ * This submap is being mapped in a map
+ * that uses a different pmap.
+ * Set its "mapped_in_other_pmaps" flag
+ * to indicate that we now need to
+ * remove mappings from all pmaps rather
+ * than just the submap's pmap.
+ */
+ submap->mapped_in_other_pmaps = TRUE;
}
- if(entry->is_sub_map) {
- vm_map_submap_pmap_clean(
- sub_map,
- (start + entry->vme_start) - offset,
- ((start + entry->vme_start) - offset) + remove_size,
- entry->object.sub_map,
- entry->offset);
- } else {
- if((map->mapped) && (map->ref_count)
- && (entry->object.vm_object != NULL)) {
- vm_object_pmap_protect(
- entry->object.vm_object,
- entry->offset,
- remove_size,
- PMAP_NULL,
- entry->vme_start,
- VM_PROT_NONE);
- } else {
- pmap_remove(map->pmap,
- (addr64_t)((start + entry->vme_start)
- - offset),
- (addr64_t)(((start + entry->vme_start)
- - offset) + remove_size));
+
+#ifndef NO_NESTED_PMAP
+ if (use_pmap) {
+ /* nest if platform code will allow */
+ if(submap->pmap == NULL) {
+ ledger_t ledger = map->pmap->ledger;
+ submap->pmap = pmap_create(ledger,
+ (vm_map_size_t) 0, FALSE);
+ if(submap->pmap == PMAP_NULL) {
+ vm_map_unlock(map);
+ return(KERN_NO_SPACE);
+ }
}
+ result = pmap_nest(map->pmap,
+ (VME_SUBMAP(entry))->pmap,
+ (addr64_t)start,
+ (addr64_t)start,
+ (uint64_t)(end - start));
+ if(result)
+ panic("vm_map_submap: pmap_nest failed, rc = %08X\n", result);
+ entry->use_pmap = TRUE;
}
- entry = entry->vme_next;
- }
- return;
+#else /* NO_NESTED_PMAP */
+ pmap_remove(map->pmap, (addr64_t)start, (addr64_t)end);
+#endif /* NO_NESTED_PMAP */
+ result = KERN_SUCCESS;
+ }
+ vm_map_unlock(map);
+
+ return(result);
}
+
/*
- * vm_map_delete: [ internal use only ]
- *
- * Deallocates the given address range from the target map.
- * Removes all user wirings. Unwires one kernel wiring if
- * VM_MAP_REMOVE_KUNWIRE is set. Waits for kernel wirings to go
- * away if VM_MAP_REMOVE_WAIT_FOR_KWIRE is set. Sleeps
- * interruptibly if VM_MAP_REMOVE_INTERRUPTIBLE is set.
+ * vm_map_protect:
*
- * This routine is called with map locked and leaves map locked.
+ * Sets the protection of the specified address
+ * region in the target map. If "set_max" is
+ * specified, the maximum protection is to be set;
+ * otherwise, only the current protection is affected.
*/
kern_return_t
-vm_map_delete(
+vm_map_protect(
register vm_map_t map,
- vm_offset_t start,
- register vm_offset_t end,
- int flags)
+ register vm_map_offset_t start,
+ register vm_map_offset_t end,
+ register vm_prot_t new_prot,
+ register boolean_t set_max)
{
- vm_map_entry_t entry, next;
- struct vm_map_entry *first_entry, tmp_entry;
- register vm_offset_t s, e;
- register vm_object_t object;
- boolean_t need_wakeup;
- unsigned int last_timestamp = ~0; /* unlikely value */
- int interruptible;
- extern vm_map_t kernel_map;
+ register vm_map_entry_t current;
+ register vm_map_offset_t prev;
+ vm_map_entry_t entry;
+ vm_prot_t new_max;
- interruptible = (flags & VM_MAP_REMOVE_INTERRUPTIBLE) ?
- THREAD_ABORTSAFE : THREAD_UNINT;
+ XPR(XPR_VM_MAP,
+ "vm_map_protect, 0x%X start 0x%X end 0x%X, new 0x%X %d",
+ map, start, end, new_prot, set_max);
- /*
- * All our DMA I/O operations in IOKit are currently done by
- * wiring through the map entries of the task requesting the I/O.
- * Because of this, we must always wait for kernel wirings
- * to go away on the entries before deleting them.
- *
- * Any caller who wants to actually remove a kernel wiring
- * should explicitly set the VM_MAP_REMOVE_KUNWIRE flag to
- * properly remove one wiring instead of blasting through
- * them all.
- */
- flags |= VM_MAP_REMOVE_WAIT_FOR_KWIRE;
+ vm_map_lock(map);
- /*
- * Find the start of the region, and clip it
+ /* LP64todo - remove this check when vm_map_commpage64()
+ * no longer has to stuff in a map_entry for the commpage
+ * above the map's max_offset.
*/
- if (vm_map_lookup_entry(map, start, &first_entry)) {
- entry = first_entry;
- vm_map_clip_start(map, entry, start);
+ if (start >= map->max_offset) {
+ vm_map_unlock(map);
+ return(KERN_INVALID_ADDRESS);
+ }
+ while(1) {
/*
- * Fix the lookup hint now, rather than each
- * time through the loop.
+ * Lookup the entry. If it doesn't start in a valid
+ * entry, return an error.
*/
- SAVE_HINT(map, entry->vme_prev);
- } else {
- entry = first_entry->vme_next;
- }
+ if (! vm_map_lookup_entry(map, start, &entry)) {
+ vm_map_unlock(map);
+ return(KERN_INVALID_ADDRESS);
+ }
+
+ if (entry->superpage_size && (start & (SUPERPAGE_SIZE-1))) { /* extend request to whole entry */
+ start = SUPERPAGE_ROUND_DOWN(start);
+ continue;
+ }
+ break;
+ }
+ if (entry->superpage_size)
+ end = SUPERPAGE_ROUND_UP(end);
- need_wakeup = FALSE;
/*
- * Step through all entries in this region
+ * Make a first pass to check for protection and address
+ * violations.
*/
- while ((entry != vm_map_to_entry(map)) && (entry->vme_start < end)) {
- vm_map_clip_end(map, entry, end);
- if (entry->in_transition) {
- wait_result_t wait_result;
+ current = entry;
+ prev = current->vme_start;
+ while ((current != vm_map_to_entry(map)) &&
+ (current->vme_start < end)) {
- /*
- * Another thread is wiring/unwiring this entry.
- * Let the other thread know we are waiting.
- */
- s = entry->vme_start;
- entry->needs_wakeup = TRUE;
+ /*
+ * If there is a hole, return an error.
+ */
+ if (current->vme_start != prev) {
+ vm_map_unlock(map);
+ return(KERN_INVALID_ADDRESS);
+ }
- /*
- * wake up anybody waiting on entries that we have
- * already unwired/deleted.
- */
- if (need_wakeup) {
- vm_map_entry_wakeup(map);
- need_wakeup = FALSE;
+ new_max = current->max_protection;
+ if(new_prot & VM_PROT_COPY) {
+ new_max |= VM_PROT_WRITE;
+ if ((new_prot & (new_max | VM_PROT_COPY)) != new_prot) {
+ vm_map_unlock(map);
+ return(KERN_PROTECTION_FAILURE);
}
-
- wait_result = vm_map_entry_wait(map, interruptible);
-
- if (interruptible &&
- wait_result == THREAD_INTERRUPTED) {
- /*
- * We do not clear the needs_wakeup flag,
- * since we cannot tell if we were the only one.
- */
+ } else {
+ if ((new_prot & new_max) != new_prot) {
vm_map_unlock(map);
- return KERN_ABORTED;
+ return(KERN_PROTECTION_FAILURE);
}
+ }
- /*
- * The entry could have been clipped or it
- * may not exist anymore. Look it up again.
- */
- if (!vm_map_lookup_entry(map, s, &first_entry)) {
- assert((map != kernel_map) &&
- (!entry->is_sub_map));
- /*
- * User: use the next entry
- */
- entry = first_entry->vme_next;
- } else {
- entry = first_entry;
- SAVE_HINT(map, entry->vme_prev);
- }
- last_timestamp = map->timestamp;
- continue;
- } /* end in_transition */
- if (entry->wired_count) {
- /*
- * Remove a kernel wiring if requested or if
- * there are user wirings.
- */
- if ((flags & VM_MAP_REMOVE_KUNWIRE) ||
- (entry->user_wired_count > 0))
- entry->wired_count--;
+ prev = current->vme_end;
+ current = current->vme_next;
+ }
+ if (end > prev) {
+ vm_map_unlock(map);
+ return(KERN_INVALID_ADDRESS);
+ }
- /* remove all user wire references */
- entry->user_wired_count = 0;
+ /*
+ * Go back and fix up protections.
+ * Clip to start here if the range starts within
+ * the entry.
+ */
- if (entry->wired_count != 0) {
- assert((map != kernel_map) &&
- (!entry->is_sub_map));
- /*
- * Cannot continue. Typical case is when
- * a user thread has physical io pending on
- * on this page. Either wait for the
- * kernel wiring to go away or return an
- * error.
- */
- if (flags & VM_MAP_REMOVE_WAIT_FOR_KWIRE) {
- wait_result_t wait_result;
+ current = entry;
+ if (current != vm_map_to_entry(map)) {
+ /* clip and unnest if necessary */
+ vm_map_clip_start(map, current, start);
+ }
- s = entry->vme_start;
- entry->needs_wakeup = TRUE;
- wait_result = vm_map_entry_wait(map,
- interruptible);
+ while ((current != vm_map_to_entry(map)) &&
+ (current->vme_start < end)) {
- if (interruptible &&
- wait_result == THREAD_INTERRUPTED) {
- /*
- * We do not clear the
- * needs_wakeup flag, since we
- * cannot tell if we were the
- * only one.
- */
- vm_map_unlock(map);
- return KERN_ABORTED;
- }
+ vm_prot_t old_prot;
- /*
- * The entry could have been clipped or
- * it may not exist anymore. Look it
- * up again.
- */
- if (!vm_map_lookup_entry(map, s,
- &first_entry)) {
- assert((map != kernel_map) &&
- (!entry->is_sub_map));
- /*
- * User: use the next entry
- */
- entry = first_entry->vme_next;
- } else {
- entry = first_entry;
- SAVE_HINT(map, entry->vme_prev);
- }
- last_timestamp = map->timestamp;
- continue;
- }
- else {
- return KERN_FAILURE;
- }
- }
+ vm_map_clip_end(map, current, end);
- entry->in_transition = TRUE;
- /*
- * copy current entry. see comment in vm_map_wire()
- */
- tmp_entry = *entry;
- s = entry->vme_start;
- e = entry->vme_end;
+ if (current->is_sub_map) {
+ /* clipping did unnest if needed */
+ assert(!current->use_pmap);
+ }
- /*
- * We can unlock the map now. The in_transition
- * state guarentees existance of the entry.
- */
- vm_map_unlock(map);
- vm_fault_unwire(map, &tmp_entry,
- tmp_entry.object.vm_object == kernel_object,
- map->pmap, tmp_entry.vme_start);
- vm_map_lock(map);
+ old_prot = current->protection;
- if (last_timestamp+1 != map->timestamp) {
- /*
- * Find the entry again. It could have
- * been clipped after we unlocked the map.
- */
- if (!vm_map_lookup_entry(map, s, &first_entry)){
- assert((map != kernel_map) &&
- (!entry->is_sub_map));
- first_entry = first_entry->vme_next;
- } else {
- SAVE_HINT(map, entry->vme_prev);
- }
- } else {
- SAVE_HINT(map, entry->vme_prev);
- first_entry = entry;
+ if(new_prot & VM_PROT_COPY) {
+ /* caller is asking specifically to copy the */
+ /* mapped data, this implies that max protection */
+ /* will include write. Caller must be prepared */
+ /* for loss of shared memory communication in the */
+ /* target area after taking this step */
+
+ if (current->is_sub_map == FALSE &&
+ VME_OBJECT(current) == VM_OBJECT_NULL) {
+ VME_OBJECT_SET(current,
+ vm_object_allocate(
+ (vm_map_size_t)
+ (current->vme_end -
+ current->vme_start)));
+ VME_OFFSET_SET(current, 0);
+ assert(current->use_pmap);
}
+ assert(current->wired_count == 0);
+ current->needs_copy = TRUE;
+ current->max_protection |= VM_PROT_WRITE;
+ }
- last_timestamp = map->timestamp;
+ if (set_max)
+ current->protection =
+ (current->max_protection =
+ new_prot & ~VM_PROT_COPY) &
+ old_prot;
+ else
+ current->protection = new_prot & ~VM_PROT_COPY;
- entry = first_entry;
- while ((entry != vm_map_to_entry(map)) &&
- (entry->vme_start < tmp_entry.vme_end)) {
- assert(entry->in_transition);
- entry->in_transition = FALSE;
- if (entry->needs_wakeup) {
- entry->needs_wakeup = FALSE;
- need_wakeup = TRUE;
- }
- entry = entry->vme_next;
+ /*
+ * Update physical map if necessary.
+ * If the request is to turn off write protection,
+ * we won't do it for real (in pmap). This is because
+ * it would cause copy-on-write to fail. We've already
+ * set, the new protection in the map, so if a
+ * write-protect fault occurred, it will be fixed up
+ * properly, COW or not.
+ */
+ if (current->protection != old_prot) {
+ /* Look one level in we support nested pmaps */
+ /* from mapped submaps which are direct entries */
+ /* in our map */
+
+ vm_prot_t prot;
+
+ prot = current->protection & ~VM_PROT_WRITE;
+
+ if (override_nx(map, VME_ALIAS(current)) && prot)
+ prot |= VM_PROT_EXECUTE;
+
+
+ if (current->is_sub_map && current->use_pmap) {
+ pmap_protect(VME_SUBMAP(current)->pmap,
+ current->vme_start,
+ current->vme_end,
+ prot);
+ } else {
+ pmap_protect(map->pmap,
+ current->vme_start,
+ current->vme_end,
+ prot);
}
- /*
- * We have unwired the entry(s). Go back and
- * delete them.
- */
- entry = first_entry;
- continue;
}
+ current = current->vme_next;
+ }
- /* entry is unwired */
- assert(entry->wired_count == 0);
- assert(entry->user_wired_count == 0);
+ current = entry;
+ while ((current != vm_map_to_entry(map)) &&
+ (current->vme_start <= end)) {
+ vm_map_simplify_entry(map, current);
+ current = current->vme_next;
+ }
- if ((!entry->is_sub_map &&
- entry->object.vm_object != kernel_object) ||
- entry->is_sub_map) {
- if(entry->is_sub_map) {
- if(entry->use_pmap) {
-#ifndef i386
- pmap_unnest(map->pmap, (addr64_t)entry->vme_start);
-#endif
- if((map->mapped) && (map->ref_count)) {
- /* clean up parent map/maps */
- vm_map_submap_pmap_clean(
- map, entry->vme_start,
- entry->vme_end,
- entry->object.sub_map,
- entry->offset);
- }
- } else {
- vm_map_submap_pmap_clean(
- map, entry->vme_start, entry->vme_end,
- entry->object.sub_map,
- entry->offset);
- }
- } else {
- object = entry->object.vm_object;
- if((map->mapped) && (map->ref_count)) {
- vm_object_pmap_protect(
- object, entry->offset,
- entry->vme_end - entry->vme_start,
- PMAP_NULL,
- entry->vme_start,
- VM_PROT_NONE);
- } else if(object != NULL) {
- if ((object->shadow != NULL) ||
- (object->phys_contiguous) ||
- (object->resident_page_count >
- atop((entry->vme_end - entry->vme_start)/4))) {
- pmap_remove(map->pmap,
- (addr64_t)(entry->vme_start),
- (addr64_t)(entry->vme_end));
- } else {
- vm_page_t p;
- vm_object_offset_t start_off;
- vm_object_offset_t end_off;
- start_off = entry->offset;
- end_off = start_off +
- (entry->vme_end - entry->vme_start);
- vm_object_lock(object);
- queue_iterate(&object->memq,
- p, vm_page_t, listq) {
- if ((!p->fictitious) &&
- (p->offset >= start_off) &&
- (p->offset < end_off)) {
- vm_offset_t start;
- start = entry->vme_start;
- start += p->offset - start_off;
- pmap_remove(
- map->pmap, start,
- start + PAGE_SIZE);
- }
- }
- vm_object_unlock(object);
- }
- }
- }
- }
-
- next = entry->vme_next;
- s = next->vme_start;
- last_timestamp = map->timestamp;
- vm_map_entry_delete(map, entry);
- /* vm_map_entry_delete unlocks the map */
- vm_map_lock(map);
- entry = next;
-
- if(entry == vm_map_to_entry(map)) {
- break;
- }
- if (last_timestamp+1 != map->timestamp) {
- /*
- * we are responsible for deleting everything
- * from the give space, if someone has interfered
- * we pick up where we left off, back fills should
- * be all right for anyone except map_delete and
- * we have to assume that the task has been fully
- * disabled before we get here
- */
- if (!vm_map_lookup_entry(map, s, &entry)){
- entry = entry->vme_next;
- } else {
- SAVE_HINT(map, entry->vme_prev);
- }
- /*
- * others can not only allocate behind us, we can
- * also see coalesce while we don't have the map lock
- */
- if(entry == vm_map_to_entry(map)) {
- break;
- }
- vm_map_clip_start(map, entry, s);
- }
- last_timestamp = map->timestamp;
- }
-
- if (map->wait_for_space)
- thread_wakeup((event_t) map);
- /*
- * wake up anybody waiting on entries that we have already deleted.
- */
- if (need_wakeup)
- vm_map_entry_wakeup(map);
-
- return KERN_SUCCESS;
+ vm_map_unlock(map);
+ return(KERN_SUCCESS);
}
/*
- * vm_map_remove:
+ * vm_map_inherit:
*
- * Remove the given address range from the target map.
- * This is the exported form of vm_map_delete.
+ * Sets the inheritance of the specified address
+ * range in the target map. Inheritance
+ * affects how the map will be shared with
+ * child maps at the time of vm_map_fork.
*/
kern_return_t
-vm_map_remove(
+vm_map_inherit(
register vm_map_t map,
- register vm_offset_t start,
- register vm_offset_t end,
- register boolean_t flags)
+ register vm_map_offset_t start,
+ register vm_map_offset_t end,
+ register vm_inherit_t new_inheritance)
{
- register kern_return_t result;
- boolean_t funnel_set = FALSE;
- funnel_t *curflock;
- thread_t cur_thread;
-
- cur_thread = current_thread();
+ register vm_map_entry_t entry;
+ vm_map_entry_t temp_entry;
- if ((cur_thread->funnel_state & TH_FN_OWNED) == TH_FN_OWNED) {
- funnel_set = TRUE;
- curflock = cur_thread->funnel_lock;
- thread_funnel_set( curflock , FALSE);
- }
vm_map_lock(map);
+
VM_MAP_RANGE_CHECK(map, start, end);
- result = vm_map_delete(map, start, end, flags);
- vm_map_unlock(map);
- if (funnel_set) {
- thread_funnel_set( curflock, TRUE);
- funnel_set = FALSE;
- }
- return(result);
-}
+ if (vm_map_lookup_entry(map, start, &temp_entry)) {
+ entry = temp_entry;
+ }
+ else {
+ temp_entry = temp_entry->vme_next;
+ entry = temp_entry;
+ }
-/*
- * Routine: vm_map_copy_discard
- *
- * Description:
- * Dispose of a map copy object (returned by
- * vm_map_copyin).
- */
-void
-vm_map_copy_discard(
- vm_map_copy_t copy)
-{
- TR_DECL("vm_map_copy_discard");
+ /* first check entire range for submaps which can't support the */
+ /* given inheritance. */
+ while ((entry != vm_map_to_entry(map)) && (entry->vme_start < end)) {
+ if(entry->is_sub_map) {
+ if(new_inheritance == VM_INHERIT_COPY) {
+ vm_map_unlock(map);
+ return(KERN_INVALID_ARGUMENT);
+ }
+ }
-/* tr3("enter: copy 0x%x type %d", copy, copy->type);*/
-free_next_copy:
- if (copy == VM_MAP_COPY_NULL)
- return;
+ entry = entry->vme_next;
+ }
- switch (copy->type) {
- case VM_MAP_COPY_ENTRY_LIST:
- while (vm_map_copy_first_entry(copy) !=
- vm_map_copy_to_entry(copy)) {
- vm_map_entry_t entry = vm_map_copy_first_entry(copy);
+ entry = temp_entry;
+ if (entry != vm_map_to_entry(map)) {
+ /* clip and unnest if necessary */
+ vm_map_clip_start(map, entry, start);
+ }
- vm_map_copy_entry_unlink(copy, entry);
- vm_object_deallocate(entry->object.vm_object);
- vm_map_copy_entry_dispose(copy, entry);
+ while ((entry != vm_map_to_entry(map)) && (entry->vme_start < end)) {
+ vm_map_clip_end(map, entry, end);
+ if (entry->is_sub_map) {
+ /* clip did unnest if needed */
+ assert(!entry->use_pmap);
}
- break;
- case VM_MAP_COPY_OBJECT:
- vm_object_deallocate(copy->cpy_object);
- break;
- case VM_MAP_COPY_KERNEL_BUFFER:
- /*
- * The vm_map_copy_t and possibly the data buffer were
- * allocated by a single call to kalloc(), i.e. the
- * vm_map_copy_t was not allocated out of the zone.
- */
- kfree((vm_offset_t) copy, copy->cpy_kalloc_size);
- return;
+ entry->inheritance = new_inheritance;
+
+ entry = entry->vme_next;
}
- zfree(vm_map_copy_zone, (vm_offset_t) copy);
+
+ vm_map_unlock(map);
+ return(KERN_SUCCESS);
}
/*
- * Routine: vm_map_copy_copy
- *
- * Description:
- * Move the information in a map copy object to
- * a new map copy object, leaving the old one
- * empty.
- *
- * This is used by kernel routines that need
- * to look at out-of-line data (in copyin form)
- * before deciding whether to return SUCCESS.
- * If the routine returns FAILURE, the original
- * copy object will be deallocated; therefore,
- * these routines must make a copy of the copy
- * object and leave the original empty so that
- * deallocation will not fail.
+ * Update the accounting for the amount of wired memory in this map. If the user has
+ * exceeded the defined limits, then we fail. Wiring on behalf of the kernel never fails.
*/
-vm_map_copy_t
-vm_map_copy_copy(
- vm_map_copy_t copy)
-{
- vm_map_copy_t new_copy;
-
- if (copy == VM_MAP_COPY_NULL)
- return VM_MAP_COPY_NULL;
- /*
- * Allocate a new copy object, and copy the information
- * from the old one into it.
- */
+static kern_return_t
+add_wire_counts(
+ vm_map_t map,
+ vm_map_entry_t entry,
+ boolean_t user_wire)
+{
+ vm_map_size_t size;
- new_copy = (vm_map_copy_t) zalloc(vm_map_copy_zone);
- *new_copy = *copy;
+ if (user_wire) {
+ unsigned int total_wire_count = vm_page_wire_count + vm_lopage_free_count;
- if (copy->type == VM_MAP_COPY_ENTRY_LIST) {
/*
- * The links in the entry chain must be
- * changed to point to the new copy object.
+ * We're wiring memory at the request of the user. Check if this is the first time the user is wiring
+ * this map entry.
*/
- vm_map_copy_first_entry(copy)->vme_prev
- = vm_map_copy_to_entry(new_copy);
- vm_map_copy_last_entry(copy)->vme_next
- = vm_map_copy_to_entry(new_copy);
- }
- /*
- * Change the old copy object into one that contains
- * nothing to be deallocated.
- */
- copy->type = VM_MAP_COPY_OBJECT;
- copy->cpy_object = VM_OBJECT_NULL;
+ if (entry->user_wired_count == 0) {
+ size = entry->vme_end - entry->vme_start;
+
+ /*
+ * Since this is the first time the user is wiring this map entry, check to see if we're
+ * exceeding the user wire limits. There is a per map limit which is the smaller of either
+ * the process's rlimit or the global vm_user_wire_limit which caps this value. There is also
+ * a system-wide limit on the amount of memory all users can wire. If the user is over either
+ * limit, then we fail.
+ */
- /*
- * Return the new object.
- */
- return new_copy;
-}
+ if(size + map->user_wire_size > MIN(map->user_wire_limit, vm_user_wire_limit) ||
+ size + ptoa_64(total_wire_count) > vm_global_user_wire_limit ||
+ size + ptoa_64(total_wire_count) > max_mem - vm_global_no_user_wire_amount)
+ return KERN_RESOURCE_SHORTAGE;
-kern_return_t
-vm_map_overwrite_submap_recurse(
- vm_map_t dst_map,
- vm_offset_t dst_addr,
- vm_size_t dst_size)
-{
- vm_offset_t dst_end;
- vm_map_entry_t tmp_entry;
- vm_map_entry_t entry;
- kern_return_t result;
- boolean_t encountered_sub_map = FALSE;
+ /*
+ * The first time the user wires an entry, we also increment the wired_count and add this to
+ * the total that has been wired in the map.
+ */
+ if (entry->wired_count >= MAX_WIRE_COUNT)
+ return KERN_FAILURE;
+ entry->wired_count++;
+ map->user_wire_size += size;
+ }
- /*
- * Verify that the destination is all writeable
- * initially. We have to trunc the destination
- * address and round the copy size or we'll end up
- * splitting entries in strange ways.
- */
+ if (entry->user_wired_count >= MAX_WIRE_COUNT)
+ return KERN_FAILURE;
- dst_end = round_page_32(dst_addr + dst_size);
- vm_map_lock(dst_map);
+ entry->user_wired_count++;
-start_pass_1:
- if (!vm_map_lookup_entry(dst_map, dst_addr, &tmp_entry)) {
- vm_map_unlock(dst_map);
- return(KERN_INVALID_ADDRESS);
- }
+ } else {
- vm_map_clip_start(dst_map, tmp_entry, trunc_page_32(dst_addr));
+ /*
+ * The kernel's wiring the memory. Just bump the count and continue.
+ */
- for (entry = tmp_entry;;) {
- vm_map_entry_t next;
+ if (entry->wired_count >= MAX_WIRE_COUNT)
+ panic("vm_map_wire: too many wirings");
- next = entry->vme_next;
- while(entry->is_sub_map) {
- vm_offset_t sub_start;
- vm_offset_t sub_end;
- vm_offset_t local_end;
+ entry->wired_count++;
+ }
- if (entry->in_transition) {
- /*
- * Say that we are waiting, and wait for entry.
- */
- entry->needs_wakeup = TRUE;
- vm_map_entry_wait(dst_map, THREAD_UNINT);
+ return KERN_SUCCESS;
+}
- goto start_pass_1;
- }
+/*
+ * Update the memory wiring accounting now that the given map entry is being unwired.
+ */
- encountered_sub_map = TRUE;
- sub_start = entry->offset;
+static void
+subtract_wire_counts(
+ vm_map_t map,
+ vm_map_entry_t entry,
+ boolean_t user_wire)
+{
- if(entry->vme_end < dst_end)
- sub_end = entry->vme_end;
- else
- sub_end = dst_end;
- sub_end -= entry->vme_start;
- sub_end += entry->offset;
- local_end = entry->vme_end;
- vm_map_unlock(dst_map);
-
- result = vm_map_overwrite_submap_recurse(
- entry->object.sub_map,
- sub_start,
- sub_end - sub_start);
-
- if(result != KERN_SUCCESS)
- return result;
- if (dst_end <= entry->vme_end)
- return KERN_SUCCESS;
- vm_map_lock(dst_map);
- if(!vm_map_lookup_entry(dst_map, local_end,
- &tmp_entry)) {
- vm_map_unlock(dst_map);
- return(KERN_INVALID_ADDRESS);
- }
- entry = tmp_entry;
- next = entry->vme_next;
- }
-
- if ( ! (entry->protection & VM_PROT_WRITE)) {
- vm_map_unlock(dst_map);
- return(KERN_PROTECTION_FAILURE);
- }
+ if (user_wire) {
/*
- * If the entry is in transition, we must wait
- * for it to exit that state. Anything could happen
- * when we unlock the map, so start over.
+ * We're unwiring memory at the request of the user. See if we're removing the last user wire reference.
*/
- if (entry->in_transition) {
- /*
- * Say that we are waiting, and wait for entry.
- */
- entry->needs_wakeup = TRUE;
- vm_map_entry_wait(dst_map, THREAD_UNINT);
+ if (entry->user_wired_count == 1) {
- goto start_pass_1;
- }
+ /*
+ * We're removing the last user wire reference. Decrement the wired_count and the total
+ * user wired memory for this map.
+ */
-/*
- * our range is contained completely within this map entry
- */
- if (dst_end <= entry->vme_end) {
- vm_map_unlock(dst_map);
- return KERN_SUCCESS;
- }
-/*
- * check that range specified is contiguous region
- */
- if ((next == vm_map_to_entry(dst_map)) ||
- (next->vme_start != entry->vme_end)) {
- vm_map_unlock(dst_map);
- return(KERN_INVALID_ADDRESS);
+ assert(entry->wired_count >= 1);
+ entry->wired_count--;
+ map->user_wire_size -= entry->vme_end - entry->vme_start;
}
+ assert(entry->user_wired_count >= 1);
+ entry->user_wired_count--;
+
+ } else {
+
/*
- * Check for permanent objects in the destination.
+ * The kernel is unwiring the memory. Just update the count.
*/
- if ((entry->object.vm_object != VM_OBJECT_NULL) &&
- ((!entry->object.vm_object->internal) ||
- (entry->object.vm_object->true_share))) {
- if(encountered_sub_map) {
- vm_map_unlock(dst_map);
- return(KERN_FAILURE);
- }
- }
-
- entry = next;
- }/* for */
- vm_map_unlock(dst_map);
- return(KERN_SUCCESS);
+ assert(entry->wired_count >= 1);
+ entry->wired_count--;
+ }
}
/*
- * Routine: vm_map_copy_overwrite
- *
- * Description:
- * Copy the memory described by the map copy
- * object (copy; returned by vm_map_copyin) onto
- * the specified destination region (dst_map, dst_addr).
- * The destination must be writeable.
- *
- * Unlike vm_map_copyout, this routine actually
- * writes over previously-mapped memory. If the
- * previous mapping was to a permanent (user-supplied)
- * memory object, it is preserved.
- *
- * The attributes (protection and inheritance) of the
- * destination region are preserved.
- *
- * If successful, consumes the copy object.
- * Otherwise, the caller is responsible for it.
- *
- * Implementation notes:
- * To overwrite aligned temporary virtual memory, it is
- * sufficient to remove the previous mapping and insert
- * the new copy. This replacement is done either on
- * the whole region (if no permanent virtual memory
- * objects are embedded in the destination region) or
- * in individual map entries.
- *
- * To overwrite permanent virtual memory , it is necessary
- * to copy each page, as the external memory management
- * interface currently does not provide any optimizations.
- *
- * Unaligned memory also has to be copied. It is possible
- * to use 'vm_trickery' to copy the aligned data. This is
- * not done but not hard to implement.
+ * vm_map_wire:
*
- * Once a page of permanent memory has been overwritten,
- * it is impossible to interrupt this function; otherwise,
- * the call would be neither atomic nor location-independent.
- * The kernel-state portion of a user thread must be
- * interruptible.
+ * Sets the pageability of the specified address range in the
+ * target map as wired. Regions specified as not pageable require
+ * locked-down physical memory and physical page maps. The
+ * access_type variable indicates types of accesses that must not
+ * generate page faults. This is checked against protection of
+ * memory being locked-down.
*
- * It may be expensive to forward all requests that might
- * overwrite permanent memory (vm_write, vm_copy) to
- * uninterruptible kernel threads. This routine may be
- * called by interruptible threads; however, success is
- * not guaranteed -- if the request cannot be performed
- * atomically and interruptibly, an error indication is
- * returned.
+ * The map must not be locked, but a reference must remain to the
+ * map throughout the call.
*/
-
-kern_return_t
-vm_map_copy_overwrite_nested(
- vm_map_t dst_map,
- vm_offset_t dst_addr,
- vm_map_copy_t copy,
- boolean_t interruptible,
- pmap_t pmap)
+static kern_return_t
+vm_map_wire_nested(
+ register vm_map_t map,
+ register vm_map_offset_t start,
+ register vm_map_offset_t end,
+ register vm_prot_t caller_prot,
+ boolean_t user_wire,
+ pmap_t map_pmap,
+ vm_map_offset_t pmap_addr,
+ ppnum_t *physpage_p)
{
- vm_offset_t dst_end;
- vm_map_entry_t tmp_entry;
- vm_map_entry_t entry;
- kern_return_t kr;
- boolean_t aligned = TRUE;
- boolean_t contains_permanent_objects = FALSE;
- boolean_t encountered_sub_map = FALSE;
- vm_offset_t base_addr;
- vm_size_t copy_size;
- vm_size_t total_size;
-
-
- /*
- * Check for null copy object.
- */
-
- if (copy == VM_MAP_COPY_NULL)
- return(KERN_SUCCESS);
+ register vm_map_entry_t entry;
+ register vm_prot_t access_type;
+ struct vm_map_entry *first_entry, tmp_entry;
+ vm_map_t real_map;
+ register vm_map_offset_t s,e;
+ kern_return_t rc;
+ boolean_t need_wakeup;
+ boolean_t main_map = FALSE;
+ wait_interrupt_t interruptible_state;
+ thread_t cur_thread;
+ unsigned int last_timestamp;
+ vm_map_size_t size;
+ boolean_t wire_and_extract;
- /*
- * Check for special kernel buffer allocated
- * by new_ipc_kmsg_copyin.
- */
+ access_type = (caller_prot & VM_PROT_ALL);
- if (copy->type == VM_MAP_COPY_KERNEL_BUFFER) {
- return(vm_map_copyout_kernel_buffer(
- dst_map, &dst_addr,
- copy, TRUE));
+ wire_and_extract = FALSE;
+ if (physpage_p != NULL) {
+ /*
+ * The caller wants the physical page number of the
+ * wired page. We return only one physical page number
+ * so this works for only one page at a time.
+ */
+ if ((end - start) != PAGE_SIZE) {
+ return KERN_INVALID_ARGUMENT;
+ }
+ wire_and_extract = TRUE;
+ *physpage_p = 0;
}
- /*
- * Only works for entry lists at the moment. Will
- * support page lists later.
- */
-
- assert(copy->type == VM_MAP_COPY_ENTRY_LIST);
+ vm_map_lock(map);
+ if(map_pmap == NULL)
+ main_map = TRUE;
+ last_timestamp = map->timestamp;
- if (copy->size == 0) {
- vm_map_copy_discard(copy);
- return(KERN_SUCCESS);
+ VM_MAP_RANGE_CHECK(map, start, end);
+ assert(page_aligned(start));
+ assert(page_aligned(end));
+ assert(VM_MAP_PAGE_ALIGNED(start, VM_MAP_PAGE_MASK(map)));
+ assert(VM_MAP_PAGE_ALIGNED(end, VM_MAP_PAGE_MASK(map)));
+ if (start == end) {
+ /* We wired what the caller asked for, zero pages */
+ vm_map_unlock(map);
+ return KERN_SUCCESS;
}
- /*
- * Verify that the destination is all writeable
- * initially. We have to trunc the destination
- * address and round the copy size or we'll end up
- * splitting entries in strange ways.
- */
+ need_wakeup = FALSE;
+ cur_thread = current_thread();
- if (!page_aligned(copy->size) ||
- !page_aligned (copy->offset) ||
- !page_aligned (dst_addr))
- {
- aligned = FALSE;
- dst_end = round_page_32(dst_addr + copy->size);
+ s = start;
+ rc = KERN_SUCCESS;
+
+ if (vm_map_lookup_entry(map, s, &first_entry)) {
+ entry = first_entry;
+ /*
+ * vm_map_clip_start will be done later.
+ * We don't want to unnest any nested submaps here !
+ */
} else {
- dst_end = dst_addr + copy->size;
+ /* Start address is not in map */
+ rc = KERN_INVALID_ADDRESS;
+ goto done;
}
- vm_map_lock(dst_map);
+ while ((entry != vm_map_to_entry(map)) && (s < end)) {
+ /*
+ * At this point, we have wired from "start" to "s".
+ * We still need to wire from "s" to "end".
+ *
+ * "entry" hasn't been clipped, so it could start before "s"
+ * and/or end after "end".
+ */
-start_pass_1:
- if (!vm_map_lookup_entry(dst_map, dst_addr, &tmp_entry)) {
- vm_map_unlock(dst_map);
- return(KERN_INVALID_ADDRESS);
- }
- vm_map_clip_start(dst_map, tmp_entry, trunc_page_32(dst_addr));
- for (entry = tmp_entry;;) {
- vm_map_entry_t next = entry->vme_next;
+ /* "e" is how far we want to wire in this entry */
+ e = entry->vme_end;
+ if (e > end)
+ e = end;
- while(entry->is_sub_map) {
- vm_offset_t sub_start;
- vm_offset_t sub_end;
- vm_offset_t local_end;
+ /*
+ * If another thread is wiring/unwiring this entry then
+ * block after informing other thread to wake us up.
+ */
+ if (entry->in_transition) {
+ wait_result_t wait_result;
- if (entry->in_transition) {
+ /*
+ * We have not clipped the entry. Make sure that
+ * the start address is in range so that the lookup
+ * below will succeed.
+ * "s" is the current starting point: we've already
+ * wired from "start" to "s" and we still have
+ * to wire from "s" to "end".
+ */
- /*
- * Say that we are waiting, and wait for entry.
- */
- entry->needs_wakeup = TRUE;
- vm_map_entry_wait(dst_map, THREAD_UNINT);
+ entry->needs_wakeup = TRUE;
- goto start_pass_1;
+ /*
+ * wake up anybody waiting on entries that we have
+ * already wired.
+ */
+ if (need_wakeup) {
+ vm_map_entry_wakeup(map);
+ need_wakeup = FALSE;
+ }
+ /*
+ * User wiring is interruptible
+ */
+ wait_result = vm_map_entry_wait(map,
+ (user_wire) ? THREAD_ABORTSAFE :
+ THREAD_UNINT);
+ if (user_wire && wait_result == THREAD_INTERRUPTED) {
+ /*
+ * undo the wirings we have done so far
+ * We do not clear the needs_wakeup flag,
+ * because we cannot tell if we were the
+ * only one waiting.
+ */
+ rc = KERN_FAILURE;
+ goto done;
}
- local_end = entry->vme_end;
- if (!(entry->needs_copy)) {
- /* if needs_copy we are a COW submap */
- /* in such a case we just replace so */
- /* there is no need for the follow- */
- /* ing check. */
- encountered_sub_map = TRUE;
- sub_start = entry->offset;
+ /*
+ * Cannot avoid a lookup here. reset timestamp.
+ */
+ last_timestamp = map->timestamp;
- if(entry->vme_end < dst_end)
- sub_end = entry->vme_end;
- else
- sub_end = dst_end;
- sub_end -= entry->vme_start;
- sub_end += entry->offset;
- vm_map_unlock(dst_map);
-
- kr = vm_map_overwrite_submap_recurse(
- entry->object.sub_map,
- sub_start,
- sub_end - sub_start);
- if(kr != KERN_SUCCESS)
- return kr;
- vm_map_lock(dst_map);
+ /*
+ * The entry could have been clipped, look it up again.
+ * Worse that can happen is, it may not exist anymore.
+ */
+ if (!vm_map_lookup_entry(map, s, &first_entry)) {
+ /*
+ * User: undo everything upto the previous
+ * entry. let vm_map_unwire worry about
+ * checking the validity of the range.
+ */
+ rc = KERN_FAILURE;
+ goto done;
}
+ entry = first_entry;
+ continue;
+ }
+
+ if (entry->is_sub_map) {
+ vm_map_offset_t sub_start;
+ vm_map_offset_t sub_end;
+ vm_map_offset_t local_start;
+ vm_map_offset_t local_end;
+ pmap_t pmap;
- if (dst_end <= entry->vme_end)
- goto start_overwrite;
- if(!vm_map_lookup_entry(dst_map, local_end,
- &entry)) {
- vm_map_unlock(dst_map);
- return(KERN_INVALID_ADDRESS);
+ if (wire_and_extract) {
+ /*
+ * Wiring would result in copy-on-write
+ * which would not be compatible with
+ * the sharing we have with the original
+ * provider of this memory.
+ */
+ rc = KERN_INVALID_ARGUMENT;
+ goto done;
}
- next = entry->vme_next;
- }
- if ( ! (entry->protection & VM_PROT_WRITE)) {
- vm_map_unlock(dst_map);
- return(KERN_PROTECTION_FAILURE);
- }
+ vm_map_clip_start(map, entry, s);
+ vm_map_clip_end(map, entry, end);
- /*
- * If the entry is in transition, we must wait
- * for it to exit that state. Anything could happen
- * when we unlock the map, so start over.
- */
- if (entry->in_transition) {
+ sub_start = VME_OFFSET(entry);
+ sub_end = entry->vme_end;
+ sub_end += VME_OFFSET(entry) - entry->vme_start;
+
+ local_end = entry->vme_end;
+ if(map_pmap == NULL) {
+ vm_object_t object;
+ vm_object_offset_t offset;
+ vm_prot_t prot;
+ boolean_t wired;
+ vm_map_entry_t local_entry;
+ vm_map_version_t version;
+ vm_map_t lookup_map;
- /*
- * Say that we are waiting, and wait for entry.
- */
- entry->needs_wakeup = TRUE;
- vm_map_entry_wait(dst_map, THREAD_UNINT);
+ if(entry->use_pmap) {
+ pmap = VME_SUBMAP(entry)->pmap;
+ /* ppc implementation requires that */
+ /* submaps pmap address ranges line */
+ /* up with parent map */
+#ifdef notdef
+ pmap_addr = sub_start;
+#endif
+ pmap_addr = s;
+ } else {
+ pmap = map->pmap;
+ pmap_addr = s;
+ }
- goto start_pass_1;
- }
+ if (entry->wired_count) {
+ if ((rc = add_wire_counts(map, entry, user_wire)) != KERN_SUCCESS)
+ goto done;
-/*
- * our range is contained completely within this map entry
- */
- if (dst_end <= entry->vme_end)
- break;
-/*
- * check that range specified is contiguous region
- */
- if ((next == vm_map_to_entry(dst_map)) ||
- (next->vme_start != entry->vme_end)) {
- vm_map_unlock(dst_map);
- return(KERN_INVALID_ADDRESS);
- }
+ /*
+ * The map was not unlocked:
+ * no need to goto re-lookup.
+ * Just go directly to next entry.
+ */
+ entry = entry->vme_next;
+ s = entry->vme_start;
+ continue;
+ }
- /*
- * Check for permanent objects in the destination.
- */
- if ((entry->object.vm_object != VM_OBJECT_NULL) &&
- ((!entry->object.vm_object->internal) ||
- (entry->object.vm_object->true_share))) {
- contains_permanent_objects = TRUE;
- }
+ /* call vm_map_lookup_locked to */
+ /* cause any needs copy to be */
+ /* evaluated */
+ local_start = entry->vme_start;
+ lookup_map = map;
+ vm_map_lock_write_to_read(map);
+ if(vm_map_lookup_locked(
+ &lookup_map, local_start,
+ access_type,
+ OBJECT_LOCK_EXCLUSIVE,
+ &version, &object,
+ &offset, &prot, &wired,
+ NULL,
+ &real_map)) {
- entry = next;
- }/* for */
+ vm_map_unlock_read(lookup_map);
+ assert(map_pmap == NULL);
+ vm_map_unwire(map, start,
+ s, user_wire);
+ return(KERN_FAILURE);
+ }
+ vm_object_unlock(object);
+ if(real_map != lookup_map)
+ vm_map_unlock(real_map);
+ vm_map_unlock_read(lookup_map);
+ vm_map_lock(map);
-start_overwrite:
- /*
- * If there are permanent objects in the destination, then
- * the copy cannot be interrupted.
- */
+ /* we unlocked, so must re-lookup */
+ if (!vm_map_lookup_entry(map,
+ local_start,
+ &local_entry)) {
+ rc = KERN_FAILURE;
+ goto done;
+ }
- if (interruptible && contains_permanent_objects) {
- vm_map_unlock(dst_map);
- return(KERN_FAILURE); /* XXX */
- }
+ /*
+ * entry could have been "simplified",
+ * so re-clip
+ */
+ entry = local_entry;
+ assert(s == local_start);
+ vm_map_clip_start(map, entry, s);
+ vm_map_clip_end(map, entry, end);
+ /* re-compute "e" */
+ e = entry->vme_end;
+ if (e > end)
+ e = end;
+
+ /* did we have a change of type? */
+ if (!entry->is_sub_map) {
+ last_timestamp = map->timestamp;
+ continue;
+ }
+ } else {
+ local_start = entry->vme_start;
+ pmap = map_pmap;
+ }
- /*
- *
- * Make a second pass, overwriting the data
- * At the beginning of each loop iteration,
- * the next entry to be overwritten is "tmp_entry"
- * (initially, the value returned from the lookup above),
- * and the starting address expected in that entry
- * is "start".
- */
+ if ((rc = add_wire_counts(map, entry, user_wire)) != KERN_SUCCESS)
+ goto done;
- total_size = copy->size;
- if(encountered_sub_map) {
- copy_size = 0;
- /* re-calculate tmp_entry since we've had the map */
- /* unlocked */
- if (!vm_map_lookup_entry( dst_map, dst_addr, &tmp_entry)) {
- vm_map_unlock(dst_map);
- return(KERN_INVALID_ADDRESS);
- }
- } else {
- copy_size = copy->size;
- }
-
- base_addr = dst_addr;
- while(TRUE) {
- /* deconstruct the copy object and do in parts */
- /* only in sub_map, interruptable case */
- vm_map_entry_t copy_entry;
- vm_map_entry_t previous_prev;
- vm_map_entry_t next_copy;
- int nentries;
- int remaining_entries;
- int new_offset;
-
- for (entry = tmp_entry; copy_size == 0;) {
- vm_map_entry_t next;
+ entry->in_transition = TRUE;
- next = entry->vme_next;
+ vm_map_unlock(map);
+ rc = vm_map_wire_nested(VME_SUBMAP(entry),
+ sub_start, sub_end,
+ caller_prot,
+ user_wire, pmap, pmap_addr,
+ NULL);
+ vm_map_lock(map);
- /* tmp_entry and base address are moved along */
- /* each time we encounter a sub-map. Otherwise */
- /* entry can outpase tmp_entry, and the copy_size */
- /* may reflect the distance between them */
- /* if the current entry is found to be in transition */
- /* we will start over at the beginning or the last */
- /* encounter of a submap as dictated by base_addr */
- /* we will zero copy_size accordingly. */
- if (entry->in_transition) {
- /*
- * Say that we are waiting, and wait for entry.
- */
- entry->needs_wakeup = TRUE;
- vm_map_entry_wait(dst_map, THREAD_UNINT);
+ /*
+ * Find the entry again. It could have been clipped
+ * after we unlocked the map.
+ */
+ if (!vm_map_lookup_entry(map, local_start,
+ &first_entry))
+ panic("vm_map_wire: re-lookup failed");
+ entry = first_entry;
- if(!vm_map_lookup_entry(dst_map, base_addr,
- &tmp_entry)) {
- vm_map_unlock(dst_map);
- return(KERN_INVALID_ADDRESS);
- }
- copy_size = 0;
- entry = tmp_entry;
- continue;
- }
- if(entry->is_sub_map) {
- vm_offset_t sub_start;
- vm_offset_t sub_end;
- vm_offset_t local_end;
+ assert(local_start == s);
+ /* re-compute "e" */
+ e = entry->vme_end;
+ if (e > end)
+ e = end;
- if (entry->needs_copy) {
- /* if this is a COW submap */
- /* just back the range with a */
- /* anonymous entry */
- if(entry->vme_end < dst_end)
- sub_end = entry->vme_end;
- else
- sub_end = dst_end;
- if(entry->vme_start < base_addr)
- sub_start = base_addr;
- else
- sub_start = entry->vme_start;
- vm_map_clip_end(
- dst_map, entry, sub_end);
- vm_map_clip_start(
- dst_map, entry, sub_start);
- entry->is_sub_map = FALSE;
- vm_map_deallocate(
- entry->object.sub_map);
- entry->object.sub_map = NULL;
- entry->is_shared = FALSE;
- entry->needs_copy = FALSE;
- entry->offset = 0;
- entry->protection = VM_PROT_ALL;
- entry->max_protection = VM_PROT_ALL;
- entry->wired_count = 0;
- entry->user_wired_count = 0;
- if(entry->inheritance
- == VM_INHERIT_SHARE)
- entry->inheritance = VM_INHERIT_COPY;
- continue;
+ last_timestamp = map->timestamp;
+ while ((entry != vm_map_to_entry(map)) &&
+ (entry->vme_start < e)) {
+ assert(entry->in_transition);
+ entry->in_transition = FALSE;
+ if (entry->needs_wakeup) {
+ entry->needs_wakeup = FALSE;
+ need_wakeup = TRUE;
}
- /* first take care of any non-sub_map */
- /* entries to send */
- if(base_addr < entry->vme_start) {
- /* stuff to send */
- copy_size =
- entry->vme_start - base_addr;
- break;
+ if (rc != KERN_SUCCESS) {/* from vm_*_wire */
+ subtract_wire_counts(map, entry, user_wire);
}
- sub_start = entry->offset;
+ entry = entry->vme_next;
+ }
+ if (rc != KERN_SUCCESS) { /* from vm_*_wire */
+ goto done;
+ }
- if(entry->vme_end < dst_end)
- sub_end = entry->vme_end;
- else
- sub_end = dst_end;
- sub_end -= entry->vme_start;
- sub_end += entry->offset;
- local_end = entry->vme_end;
- vm_map_unlock(dst_map);
- copy_size = sub_end - sub_start;
+ /* no need to relookup again */
+ s = entry->vme_start;
+ continue;
+ }
- /* adjust the copy object */
- if (total_size > copy_size) {
- vm_size_t local_size = 0;
- vm_size_t entry_size;
-
- nentries = 1;
- new_offset = copy->offset;
- copy_entry = vm_map_copy_first_entry(copy);
- while(copy_entry !=
- vm_map_copy_to_entry(copy)){
- entry_size = copy_entry->vme_end -
- copy_entry->vme_start;
- if((local_size < copy_size) &&
- ((local_size + entry_size)
- >= copy_size)) {
- vm_map_copy_clip_end(copy,
- copy_entry,
- copy_entry->vme_start +
- (copy_size - local_size));
- entry_size = copy_entry->vme_end -
- copy_entry->vme_start;
- local_size += entry_size;
- new_offset += entry_size;
- }
- if(local_size >= copy_size) {
- next_copy = copy_entry->vme_next;
- copy_entry->vme_next =
- vm_map_copy_to_entry(copy);
- previous_prev =
- copy->cpy_hdr.links.prev;
- copy->cpy_hdr.links.prev = copy_entry;
- copy->size = copy_size;
- remaining_entries =
- copy->cpy_hdr.nentries;
- remaining_entries -= nentries;
- copy->cpy_hdr.nentries = nentries;
- break;
- } else {
- local_size += entry_size;
- new_offset += entry_size;
- nentries++;
- }
- copy_entry = copy_entry->vme_next;
- }
+ /*
+ * If this entry is already wired then increment
+ * the appropriate wire reference count.
+ */
+ if (entry->wired_count) {
+
+ if ((entry->protection & access_type) != access_type) {
+ /* found a protection problem */
+
+ /*
+ * XXX FBDP
+ * We should always return an error
+ * in this case but since we didn't
+ * enforce it before, let's do
+ * it only for the new "wire_and_extract"
+ * code path for now...
+ */
+ if (wire_and_extract) {
+ rc = KERN_PROTECTION_FAILURE;
+ goto done;
}
-
- if((entry->use_pmap) && (pmap == NULL)) {
- kr = vm_map_copy_overwrite_nested(
- entry->object.sub_map,
- sub_start,
- copy,
- interruptible,
- entry->object.sub_map->pmap);
- } else if (pmap != NULL) {
- kr = vm_map_copy_overwrite_nested(
- entry->object.sub_map,
- sub_start,
- copy,
- interruptible, pmap);
- } else {
- kr = vm_map_copy_overwrite_nested(
- entry->object.sub_map,
- sub_start,
- copy,
- interruptible,
- dst_map->pmap);
- }
- if(kr != KERN_SUCCESS) {
- if(next_copy != NULL) {
- copy->cpy_hdr.nentries +=
- remaining_entries;
- copy->cpy_hdr.links.prev->vme_next =
- next_copy;
- copy->cpy_hdr.links.prev
- = previous_prev;
- copy->size = total_size;
- }
- return kr;
- }
- if (dst_end <= local_end) {
- return(KERN_SUCCESS);
- }
- /* otherwise copy no longer exists, it was */
- /* destroyed after successful copy_overwrite */
- copy = (vm_map_copy_t)
- zalloc(vm_map_copy_zone);
- vm_map_copy_first_entry(copy) =
- vm_map_copy_last_entry(copy) =
- vm_map_copy_to_entry(copy);
- copy->type = VM_MAP_COPY_ENTRY_LIST;
- copy->offset = new_offset;
-
- total_size -= copy_size;
- copy_size = 0;
- /* put back remainder of copy in container */
- if(next_copy != NULL) {
- copy->cpy_hdr.nentries = remaining_entries;
- copy->cpy_hdr.links.next = next_copy;
- copy->cpy_hdr.links.prev = previous_prev;
- copy->size = total_size;
- next_copy->vme_prev =
- vm_map_copy_to_entry(copy);
- next_copy = NULL;
- }
- base_addr = local_end;
- vm_map_lock(dst_map);
- if(!vm_map_lookup_entry(dst_map,
- local_end, &tmp_entry)) {
- vm_map_unlock(dst_map);
- return(KERN_INVALID_ADDRESS);
- }
- entry = tmp_entry;
- continue;
- }
- if (dst_end <= entry->vme_end) {
- copy_size = dst_end - base_addr;
- break;
}
- if ((next == vm_map_to_entry(dst_map)) ||
- (next->vme_start != entry->vme_end)) {
- vm_map_unlock(dst_map);
- return(KERN_INVALID_ADDRESS);
- }
+ /*
+ * entry is already wired down, get our reference
+ * after clipping to our range.
+ */
+ vm_map_clip_start(map, entry, s);
+ vm_map_clip_end(map, entry, end);
- entry = next;
- }/* for */
+ if ((rc = add_wire_counts(map, entry, user_wire)) != KERN_SUCCESS)
+ goto done;
- next_copy = NULL;
- nentries = 1;
+ if (wire_and_extract) {
+ vm_object_t object;
+ vm_object_offset_t offset;
+ vm_page_t m;
- /* adjust the copy object */
- if (total_size > copy_size) {
- vm_size_t local_size = 0;
- vm_size_t entry_size;
+ /*
+ * We don't have to "wire" the page again
+ * bit we still have to "extract" its
+ * physical page number, after some sanity
+ * checks.
+ */
+ assert((entry->vme_end - entry->vme_start)
+ == PAGE_SIZE);
+ assert(!entry->needs_copy);
+ assert(!entry->is_sub_map);
+ assert(VME_OBJECT(entry));
+ if (((entry->vme_end - entry->vme_start)
+ != PAGE_SIZE) ||
+ entry->needs_copy ||
+ entry->is_sub_map ||
+ VME_OBJECT(entry) == VM_OBJECT_NULL) {
+ rc = KERN_INVALID_ARGUMENT;
+ goto done;
+ }
- new_offset = copy->offset;
- copy_entry = vm_map_copy_first_entry(copy);
- while(copy_entry != vm_map_copy_to_entry(copy)) {
- entry_size = copy_entry->vme_end -
- copy_entry->vme_start;
- if((local_size < copy_size) &&
- ((local_size + entry_size)
- >= copy_size)) {
- vm_map_copy_clip_end(copy, copy_entry,
- copy_entry->vme_start +
- (copy_size - local_size));
- entry_size = copy_entry->vme_end -
- copy_entry->vme_start;
- local_size += entry_size;
- new_offset += entry_size;
+ object = VME_OBJECT(entry);
+ offset = VME_OFFSET(entry);
+ /* need exclusive lock to update m->dirty */
+ if (entry->protection & VM_PROT_WRITE) {
+ vm_object_lock(object);
+ } else {
+ vm_object_lock_shared(object);
}
- if(local_size >= copy_size) {
- next_copy = copy_entry->vme_next;
- copy_entry->vme_next =
- vm_map_copy_to_entry(copy);
- previous_prev =
- copy->cpy_hdr.links.prev;
- copy->cpy_hdr.links.prev = copy_entry;
- copy->size = copy_size;
- remaining_entries =
- copy->cpy_hdr.nentries;
- remaining_entries -= nentries;
- copy->cpy_hdr.nentries = nentries;
- break;
+ m = vm_page_lookup(object, offset);
+ assert(m != VM_PAGE_NULL);
+ assert(m->wire_count);
+ if (m != VM_PAGE_NULL && m->wire_count) {
+ *physpage_p = m->phys_page;
+ if (entry->protection & VM_PROT_WRITE) {
+ vm_object_lock_assert_exclusive(
+ m->object);
+ m->dirty = TRUE;
+ }
} else {
- local_size += entry_size;
- new_offset += entry_size;
- nentries++;
+ /* not already wired !? */
+ *physpage_p = 0;
}
- copy_entry = copy_entry->vme_next;
+ vm_object_unlock(object);
}
+
+ /* map was not unlocked: no need to relookup */
+ entry = entry->vme_next;
+ s = entry->vme_start;
+ continue;
}
- if (aligned) {
- pmap_t local_pmap;
+ /*
+ * Unwired entry or wire request transmitted via submap
+ */
- if(pmap)
- local_pmap = pmap;
- else
- local_pmap = dst_map->pmap;
- if ((kr = vm_map_copy_overwrite_aligned(
- dst_map, tmp_entry, copy,
- base_addr, local_pmap)) != KERN_SUCCESS) {
- if(next_copy != NULL) {
- copy->cpy_hdr.nentries +=
- remaining_entries;
- copy->cpy_hdr.links.prev->vme_next =
- next_copy;
- copy->cpy_hdr.links.prev =
- previous_prev;
- copy->size += copy_size;
- }
- return kr;
- }
- vm_map_unlock(dst_map);
- } else {
/*
- * Performance gain:
- *
- * if the copy and dst address are misaligned but the same
- * offset within the page we can copy_not_aligned the
- * misaligned parts and copy aligned the rest. If they are
- * aligned but len is unaligned we simply need to copy
- * the end bit unaligned. We'll need to split the misaligned
- * bits of the region in this case !
- */
- /* ALWAYS UNLOCKS THE dst_map MAP */
- if ((kr = vm_map_copy_overwrite_unaligned( dst_map,
- tmp_entry, copy, base_addr)) != KERN_SUCCESS) {
- if(next_copy != NULL) {
- copy->cpy_hdr.nentries +=
- remaining_entries;
- copy->cpy_hdr.links.prev->vme_next =
- next_copy;
- copy->cpy_hdr.links.prev =
- previous_prev;
- copy->size += copy_size;
- }
- return kr;
- }
- }
- total_size -= copy_size;
- if(total_size == 0)
- break;
- base_addr += copy_size;
- copy_size = 0;
- copy->offset = new_offset;
- if(next_copy != NULL) {
- copy->cpy_hdr.nentries = remaining_entries;
- copy->cpy_hdr.links.next = next_copy;
- copy->cpy_hdr.links.prev = previous_prev;
- next_copy->vme_prev = vm_map_copy_to_entry(copy);
- copy->size = total_size;
- }
- vm_map_lock(dst_map);
- while(TRUE) {
- if (!vm_map_lookup_entry(dst_map,
- base_addr, &tmp_entry)) {
- vm_map_unlock(dst_map);
- return(KERN_INVALID_ADDRESS);
+ * Perform actions of vm_map_lookup that need the write
+ * lock on the map: create a shadow object for a
+ * copy-on-write region, or an object for a zero-fill
+ * region.
+ */
+ size = entry->vme_end - entry->vme_start;
+ /*
+ * If wiring a copy-on-write page, we need to copy it now
+ * even if we're only (currently) requesting read access.
+ * This is aggressive, but once it's wired we can't move it.
+ */
+ if (entry->needs_copy) {
+ if (wire_and_extract) {
+ /*
+ * We're supposed to share with the original
+ * provider so should not be "needs_copy"
+ */
+ rc = KERN_INVALID_ARGUMENT;
+ goto done;
}
- if (tmp_entry->in_transition) {
- entry->needs_wakeup = TRUE;
- vm_map_entry_wait(dst_map, THREAD_UNINT);
- } else {
- break;
+
+ VME_OBJECT_SHADOW(entry, size);
+ entry->needs_copy = FALSE;
+ } else if (VME_OBJECT(entry) == VM_OBJECT_NULL) {
+ if (wire_and_extract) {
+ /*
+ * We're supposed to share with the original
+ * provider so should already have an object.
+ */
+ rc = KERN_INVALID_ARGUMENT;
+ goto done;
}
+ VME_OBJECT_SET(entry, vm_object_allocate(size));
+ VME_OFFSET_SET(entry, (vm_object_offset_t)0);
+ assert(entry->use_pmap);
}
- vm_map_clip_start(dst_map, tmp_entry, trunc_page_32(base_addr));
- entry = tmp_entry;
- } /* while */
+ vm_map_clip_start(map, entry, s);
+ vm_map_clip_end(map, entry, end);
- /*
- * Throw away the vm_map_copy object
- */
- vm_map_copy_discard(copy);
+ /* re-compute "e" */
+ e = entry->vme_end;
+ if (e > end)
+ e = end;
- return(KERN_SUCCESS);
-}/* vm_map_copy_overwrite */
+ /*
+ * Check for holes and protection mismatch.
+ * Holes: Next entry should be contiguous unless this
+ * is the end of the region.
+ * Protection: Access requested must be allowed, unless
+ * wiring is by protection class
+ */
+ if ((entry->vme_end < end) &&
+ ((entry->vme_next == vm_map_to_entry(map)) ||
+ (entry->vme_next->vme_start > entry->vme_end))) {
+ /* found a hole */
+ rc = KERN_INVALID_ADDRESS;
+ goto done;
+ }
+ if ((entry->protection & access_type) != access_type) {
+ /* found a protection problem */
+ rc = KERN_PROTECTION_FAILURE;
+ goto done;
+ }
-kern_return_t
-vm_map_copy_overwrite(
- vm_map_t dst_map,
- vm_offset_t dst_addr,
- vm_map_copy_t copy,
- boolean_t interruptible)
-{
- return vm_map_copy_overwrite_nested(
- dst_map, dst_addr, copy, interruptible, (pmap_t) NULL);
-}
+ assert(entry->wired_count == 0 && entry->user_wired_count == 0);
+ if ((rc = add_wire_counts(map, entry, user_wire)) != KERN_SUCCESS)
+ goto done;
-/*
- * Routine: vm_map_copy_overwrite_unaligned
- *
- * Decription:
- * Physically copy unaligned data
- *
- * Implementation:
- * Unaligned parts of pages have to be physically copied. We use
- * a modified form of vm_fault_copy (which understands none-aligned
- * page offsets and sizes) to do the copy. We attempt to copy as
- * much memory in one go as possibly, however vm_fault_copy copies
- * within 1 memory object so we have to find the smaller of "amount left"
- * "source object data size" and "target object data size". With
- * unaligned data we don't need to split regions, therefore the source
- * (copy) object should be one map entry, the target range may be split
- * over multiple map entries however. In any event we are pessimistic
- * about these assumptions.
- *
- * Assumptions:
- * dst_map is locked on entry and is return locked on success,
- * unlocked on error.
- */
+ entry->in_transition = TRUE;
-kern_return_t
-vm_map_copy_overwrite_unaligned(
- vm_map_t dst_map,
- vm_map_entry_t entry,
- vm_map_copy_t copy,
- vm_offset_t start)
-{
- vm_map_entry_t copy_entry = vm_map_copy_first_entry(copy);
- vm_map_version_t version;
- vm_object_t dst_object;
- vm_object_offset_t dst_offset;
- vm_object_offset_t src_offset;
- vm_object_offset_t entry_offset;
- vm_offset_t entry_end;
- vm_size_t src_size,
- dst_size,
- copy_size,
- amount_left;
- kern_return_t kr = KERN_SUCCESS;
+ /*
+ * This entry might get split once we unlock the map.
+ * In vm_fault_wire(), we need the current range as
+ * defined by this entry. In order for this to work
+ * along with a simultaneous clip operation, we make a
+ * temporary copy of this entry and use that for the
+ * wiring. Note that the underlying objects do not
+ * change during a clip.
+ */
+ tmp_entry = *entry;
- vm_map_lock_write_to_read(dst_map);
+ /*
+ * The in_transition state guarentees that the entry
+ * (or entries for this range, if split occured) will be
+ * there when the map lock is acquired for the second time.
+ */
+ vm_map_unlock(map);
- src_offset = copy->offset - trunc_page_64(copy->offset);
- amount_left = copy->size;
-/*
- * unaligned so we never clipped this entry, we need the offset into
- * the vm_object not just the data.
- */
- while (amount_left > 0) {
-
- if (entry == vm_map_to_entry(dst_map)) {
- vm_map_unlock_read(dst_map);
- return KERN_INVALID_ADDRESS;
- }
+ if (!user_wire && cur_thread != THREAD_NULL)
+ interruptible_state = thread_interrupt_level(THREAD_UNINT);
+ else
+ interruptible_state = THREAD_UNINT;
- /* "start" must be within the current map entry */
- assert ((start>=entry->vme_start) && (start<entry->vme_end));
+ if(map_pmap)
+ rc = vm_fault_wire(map,
+ &tmp_entry, caller_prot, map_pmap, pmap_addr,
+ physpage_p);
+ else
+ rc = vm_fault_wire(map,
+ &tmp_entry, caller_prot, map->pmap,
+ tmp_entry.vme_start,
+ physpage_p);
- dst_offset = start - entry->vme_start;
+ if (!user_wire && cur_thread != THREAD_NULL)
+ thread_interrupt_level(interruptible_state);
- dst_size = entry->vme_end - start;
+ vm_map_lock(map);
- src_size = copy_entry->vme_end -
- (copy_entry->vme_start + src_offset);
+ if (last_timestamp+1 != map->timestamp) {
+ /*
+ * Find the entry again. It could have been clipped
+ * after we unlocked the map.
+ */
+ if (!vm_map_lookup_entry(map, tmp_entry.vme_start,
+ &first_entry))
+ panic("vm_map_wire: re-lookup failed");
- if (dst_size < src_size) {
-/*
- * we can only copy dst_size bytes before
- * we have to get the next destination entry
- */
- copy_size = dst_size;
- } else {
-/*
- * we can only copy src_size bytes before
- * we have to get the next source copy entry
- */
- copy_size = src_size;
+ entry = first_entry;
}
- if (copy_size > amount_left) {
- copy_size = amount_left;
- }
-/*
- * Entry needs copy, create a shadow shadow object for
- * Copy on write region.
- */
- if (entry->needs_copy &&
- ((entry->protection & VM_PROT_WRITE) != 0))
- {
- if (vm_map_lock_read_to_write(dst_map)) {
- vm_map_lock_read(dst_map);
- goto RetryLookup;
+ last_timestamp = map->timestamp;
+
+ while ((entry != vm_map_to_entry(map)) &&
+ (entry->vme_start < tmp_entry.vme_end)) {
+ assert(entry->in_transition);
+ entry->in_transition = FALSE;
+ if (entry->needs_wakeup) {
+ entry->needs_wakeup = FALSE;
+ need_wakeup = TRUE;
}
- vm_object_shadow(&entry->object.vm_object,
- &entry->offset,
- (vm_size_t)(entry->vme_end
- - entry->vme_start));
- entry->needs_copy = FALSE;
- vm_map_lock_write_to_read(dst_map);
- }
- dst_object = entry->object.vm_object;
-/*
- * unlike with the virtual (aligned) copy we're going
- * to fault on it therefore we need a target object.
- */
- if (dst_object == VM_OBJECT_NULL) {
- if (vm_map_lock_read_to_write(dst_map)) {
- vm_map_lock_read(dst_map);
- goto RetryLookup;
+ if (rc != KERN_SUCCESS) { /* from vm_*_wire */
+ subtract_wire_counts(map, entry, user_wire);
}
- dst_object = vm_object_allocate((vm_size_t)
- entry->vme_end - entry->vme_start);
- entry->object.vm_object = dst_object;
- entry->offset = 0;
- vm_map_lock_write_to_read(dst_map);
+ entry = entry->vme_next;
}
-/*
- * Take an object reference and unlock map. The "entry" may
- * disappear or change when the map is unlocked.
- */
- vm_object_reference(dst_object);
- version.main_timestamp = dst_map->timestamp;
- entry_offset = entry->offset;
- entry_end = entry->vme_end;
- vm_map_unlock_read(dst_map);
-/*
- * Copy as much as possible in one pass
- */
- kr = vm_fault_copy(
- copy_entry->object.vm_object,
- copy_entry->offset + src_offset,
- ©_size,
- dst_object,
- entry_offset + dst_offset,
- dst_map,
- &version,
- THREAD_UNINT );
- start += copy_size;
- src_offset += copy_size;
- amount_left -= copy_size;
-/*
- * Release the object reference
- */
- vm_object_deallocate(dst_object);
-/*
- * If a hard error occurred, return it now
- */
- if (kr != KERN_SUCCESS)
- return kr;
+ if (rc != KERN_SUCCESS) { /* from vm_*_wire */
+ goto done;
+ }
- if ((copy_entry->vme_start + src_offset) == copy_entry->vme_end
- || amount_left == 0)
- {
-/*
- * all done with this copy entry, dispose.
- */
- vm_map_copy_entry_unlink(copy, copy_entry);
- vm_object_deallocate(copy_entry->object.vm_object);
- vm_map_copy_entry_dispose(copy, copy_entry);
+ s = entry->vme_start;
+ } /* end while loop through map entries */
- if ((copy_entry = vm_map_copy_first_entry(copy))
- == vm_map_copy_to_entry(copy) && amount_left) {
-/*
- * not finished copying but run out of source
- */
- return KERN_INVALID_ADDRESS;
- }
- src_offset = 0;
- }
+done:
+ if (rc == KERN_SUCCESS) {
+ /* repair any damage we may have made to the VM map */
+ vm_map_simplify_range(map, start, end);
+ }
- if (amount_left == 0)
- return KERN_SUCCESS;
+ vm_map_unlock(map);
- vm_map_lock_read(dst_map);
- if (version.main_timestamp == dst_map->timestamp) {
- if (start == entry_end) {
-/*
- * destination region is split. Use the version
- * information to avoid a lookup in the normal
- * case.
- */
- entry = entry->vme_next;
-/*
- * should be contiguous. Fail if we encounter
- * a hole in the destination.
- */
- if (start != entry->vme_start) {
- vm_map_unlock_read(dst_map);
- return KERN_INVALID_ADDRESS ;
- }
- }
- } else {
-/*
- * Map version check failed.
- * we must lookup the entry because somebody
- * might have changed the map behind our backs.
- */
-RetryLookup:
- if (!vm_map_lookup_entry(dst_map, start, &entry))
- {
- vm_map_unlock_read(dst_map);
- return KERN_INVALID_ADDRESS ;
- }
- }
- }/* while */
+ /*
+ * wake up anybody waiting on entries we wired.
+ */
+ if (need_wakeup)
+ vm_map_entry_wakeup(map);
- /* NOTREACHED ?? */
- vm_map_unlock_read(dst_map);
+ if (rc != KERN_SUCCESS) {
+ /* undo what has been wired so far */
+ vm_map_unwire_nested(map, start, s, user_wire,
+ map_pmap, pmap_addr);
+ if (physpage_p) {
+ *physpage_p = 0;
+ }
+ }
- return KERN_SUCCESS;
-}/* vm_map_copy_overwrite_unaligned */
+ return rc;
-/*
- * Routine: vm_map_copy_overwrite_aligned
- *
- * Description:
- * Does all the vm_trickery possible for whole pages.
- *
- * Implementation:
- *
- * If there are no permanent objects in the destination,
- * and the source and destination map entry zones match,
- * and the destination map entry is not shared,
- * then the map entries can be deleted and replaced
- * with those from the copy. The following code is the
- * basic idea of what to do, but there are lots of annoying
- * little details about getting protection and inheritance
- * right. Should add protection, inheritance, and sharing checks
- * to the above pass and make sure that no wiring is involved.
- */
+}
kern_return_t
-vm_map_copy_overwrite_aligned(
- vm_map_t dst_map,
- vm_map_entry_t tmp_entry,
- vm_map_copy_t copy,
- vm_offset_t start,
- pmap_t pmap)
+vm_map_wire_external(
+ register vm_map_t map,
+ register vm_map_offset_t start,
+ register vm_map_offset_t end,
+ register vm_prot_t caller_prot,
+ boolean_t user_wire)
{
- vm_object_t object;
- vm_map_entry_t copy_entry;
- vm_size_t copy_size;
- vm_size_t size;
- vm_map_entry_t entry;
-
- while ((copy_entry = vm_map_copy_first_entry(copy))
- != vm_map_copy_to_entry(copy))
- {
- copy_size = (copy_entry->vme_end - copy_entry->vme_start);
-
- entry = tmp_entry;
- if (entry == vm_map_to_entry(dst_map)) {
- vm_map_unlock(dst_map);
- return KERN_INVALID_ADDRESS;
- }
- size = (entry->vme_end - entry->vme_start);
- /*
- * Make sure that no holes popped up in the
- * address map, and that the protection is
- * still valid, in case the map was unlocked
- * earlier.
- */
+ kern_return_t kret;
- if ((entry->vme_start != start) || ((entry->is_sub_map)
- && !entry->needs_copy)) {
- vm_map_unlock(dst_map);
- return(KERN_INVALID_ADDRESS);
- }
- assert(entry != vm_map_to_entry(dst_map));
+ caller_prot &= ~VM_PROT_MEMORY_TAG_MASK;
+ caller_prot |= VM_PROT_MEMORY_TAG_MAKE(vm_tag_bt());
+ kret = vm_map_wire_nested(map, start, end, caller_prot,
+ user_wire, (pmap_t)NULL, 0, NULL);
+ return kret;
+}
- /*
- * Check protection again
- */
+kern_return_t
+vm_map_wire(
+ register vm_map_t map,
+ register vm_map_offset_t start,
+ register vm_map_offset_t end,
+ register vm_prot_t caller_prot,
+ boolean_t user_wire)
+{
+ kern_return_t kret;
- if ( ! (entry->protection & VM_PROT_WRITE)) {
- vm_map_unlock(dst_map);
- return(KERN_PROTECTION_FAILURE);
- }
+ kret = vm_map_wire_nested(map, start, end, caller_prot,
+ user_wire, (pmap_t)NULL, 0, NULL);
+ return kret;
+}
- /*
- * Adjust to source size first
- */
+kern_return_t
+vm_map_wire_and_extract_external(
+ vm_map_t map,
+ vm_map_offset_t start,
+ vm_prot_t caller_prot,
+ boolean_t user_wire,
+ ppnum_t *physpage_p)
+{
+ kern_return_t kret;
- if (copy_size < size) {
- vm_map_clip_end(dst_map, entry, entry->vme_start + copy_size);
- size = copy_size;
- }
+ caller_prot &= ~VM_PROT_MEMORY_TAG_MASK;
+ caller_prot |= VM_PROT_MEMORY_TAG_MAKE(vm_tag_bt());
+ kret = vm_map_wire_nested(map,
+ start,
+ start+VM_MAP_PAGE_SIZE(map),
+ caller_prot,
+ user_wire,
+ (pmap_t)NULL,
+ 0,
+ physpage_p);
+ if (kret != KERN_SUCCESS &&
+ physpage_p != NULL) {
+ *physpage_p = 0;
+ }
+ return kret;
+}
- /*
- * Adjust to destination size
- */
+kern_return_t
+vm_map_wire_and_extract(
+ vm_map_t map,
+ vm_map_offset_t start,
+ vm_prot_t caller_prot,
+ boolean_t user_wire,
+ ppnum_t *physpage_p)
+{
+ kern_return_t kret;
- if (size < copy_size) {
- vm_map_copy_clip_end(copy, copy_entry,
- copy_entry->vme_start + size);
- copy_size = size;
- }
+ kret = vm_map_wire_nested(map,
+ start,
+ start+VM_MAP_PAGE_SIZE(map),
+ caller_prot,
+ user_wire,
+ (pmap_t)NULL,
+ 0,
+ physpage_p);
+ if (kret != KERN_SUCCESS &&
+ physpage_p != NULL) {
+ *physpage_p = 0;
+ }
+ return kret;
+}
- assert((entry->vme_end - entry->vme_start) == size);
- assert((tmp_entry->vme_end - tmp_entry->vme_start) == size);
- assert((copy_entry->vme_end - copy_entry->vme_start) == size);
+/*
+ * vm_map_unwire:
+ *
+ * Sets the pageability of the specified address range in the target
+ * as pageable. Regions specified must have been wired previously.
+ *
+ * The map must not be locked, but a reference must remain to the map
+ * throughout the call.
+ *
+ * Kernel will panic on failures. User unwire ignores holes and
+ * unwired and intransition entries to avoid losing memory by leaving
+ * it unwired.
+ */
+static kern_return_t
+vm_map_unwire_nested(
+ register vm_map_t map,
+ register vm_map_offset_t start,
+ register vm_map_offset_t end,
+ boolean_t user_wire,
+ pmap_t map_pmap,
+ vm_map_offset_t pmap_addr)
+{
+ register vm_map_entry_t entry;
+ struct vm_map_entry *first_entry, tmp_entry;
+ boolean_t need_wakeup;
+ boolean_t main_map = FALSE;
+ unsigned int last_timestamp;
+
+ vm_map_lock(map);
+ if(map_pmap == NULL)
+ main_map = TRUE;
+ last_timestamp = map->timestamp;
+
+ VM_MAP_RANGE_CHECK(map, start, end);
+ assert(page_aligned(start));
+ assert(page_aligned(end));
+ assert(VM_MAP_PAGE_ALIGNED(start, VM_MAP_PAGE_MASK(map)));
+ assert(VM_MAP_PAGE_ALIGNED(end, VM_MAP_PAGE_MASK(map)));
+
+ if (start == end) {
+ /* We unwired what the caller asked for: zero pages */
+ vm_map_unlock(map);
+ return KERN_SUCCESS;
+ }
+ if (vm_map_lookup_entry(map, start, &first_entry)) {
+ entry = first_entry;
/*
- * If the destination contains temporary unshared memory,
- * we can perform the copy by throwing it away and
- * installing the source data.
+ * vm_map_clip_start will be done later.
+ * We don't want to unnest any nested sub maps here !
*/
+ }
+ else {
+ if (!user_wire) {
+ panic("vm_map_unwire: start not found");
+ }
+ /* Start address is not in map. */
+ vm_map_unlock(map);
+ return(KERN_INVALID_ADDRESS);
+ }
- object = entry->object.vm_object;
- if ((!entry->is_shared &&
- ((object == VM_OBJECT_NULL) ||
- (object->internal && !object->true_share))) ||
- entry->needs_copy) {
- vm_object_t old_object = entry->object.vm_object;
- vm_object_offset_t old_offset = entry->offset;
- vm_object_offset_t offset;
+ if (entry->superpage_size) {
+ /* superpages are always wired */
+ vm_map_unlock(map);
+ return KERN_INVALID_ADDRESS;
+ }
+ need_wakeup = FALSE;
+ while ((entry != vm_map_to_entry(map)) && (entry->vme_start < end)) {
+ if (entry->in_transition) {
/*
- * Ensure that the source and destination aren't
- * identical
+ * 1)
+ * Another thread is wiring down this entry. Note
+ * that if it is not for the other thread we would
+ * be unwiring an unwired entry. This is not
+ * permitted. If we wait, we will be unwiring memory
+ * we did not wire.
+ *
+ * 2)
+ * Another thread is unwiring this entry. We did not
+ * have a reference to it, because if we did, this
+ * entry will not be getting unwired now.
*/
- if (old_object == copy_entry->object.vm_object &&
- old_offset == copy_entry->offset) {
- vm_map_copy_entry_unlink(copy, copy_entry);
- vm_map_copy_entry_dispose(copy, copy_entry);
+ if (!user_wire) {
+ /*
+ * XXX FBDP
+ * This could happen: there could be some
+ * overlapping vslock/vsunlock operations
+ * going on.
+ * We should probably just wait and retry,
+ * but then we have to be careful that this
+ * entry could get "simplified" after
+ * "in_transition" gets unset and before
+ * we re-lookup the entry, so we would
+ * have to re-clip the entry to avoid
+ * re-unwiring what we have already unwired...
+ * See vm_map_wire_nested().
+ *
+ * Or we could just ignore "in_transition"
+ * here and proceed to decement the wired
+ * count(s) on this entry. That should be fine
+ * as long as "wired_count" doesn't drop all
+ * the way to 0 (and we should panic if THAT
+ * happens).
+ */
+ panic("vm_map_unwire: in_transition entry");
+ }
- if (old_object != VM_OBJECT_NULL)
- vm_object_deallocate(old_object);
+ entry = entry->vme_next;
+ continue;
+ }
- start = tmp_entry->vme_end;
- tmp_entry = tmp_entry->vme_next;
- continue;
- }
+ if (entry->is_sub_map) {
+ vm_map_offset_t sub_start;
+ vm_map_offset_t sub_end;
+ vm_map_offset_t local_end;
+ pmap_t pmap;
+
+ vm_map_clip_start(map, entry, start);
+ vm_map_clip_end(map, entry, end);
- if (old_object != VM_OBJECT_NULL) {
- if(entry->is_sub_map) {
- if(entry->use_pmap) {
-#ifndef i386
- pmap_unnest(dst_map->pmap,
- entry->vme_start,
- entry->vme_end
- - entry->vme_start);
-#endif
- if(dst_map->mapped) {
- /* clean up parent */
- /* map/maps */
- vm_map_submap_pmap_clean(
- dst_map, entry->vme_start,
- entry->vme_end,
- entry->object.sub_map,
- entry->offset);
- }
- } else {
- vm_map_submap_pmap_clean(
- dst_map, entry->vme_start,
- entry->vme_end,
- entry->object.sub_map,
- entry->offset);
- }
- vm_map_deallocate(
- entry->object.sub_map);
- } else {
- if(dst_map->mapped) {
- vm_object_pmap_protect(
- entry->object.vm_object,
- entry->offset,
- entry->vme_end
- - entry->vme_start,
- PMAP_NULL,
- entry->vme_start,
- VM_PROT_NONE);
- } else {
- pmap_remove(dst_map->pmap,
- (addr64_t)(entry->vme_start),
- (addr64_t)(entry->vme_end));
- }
- vm_object_deallocate(old_object);
- }
- }
+ sub_start = VME_OFFSET(entry);
+ sub_end = entry->vme_end - entry->vme_start;
+ sub_end += VME_OFFSET(entry);
+ local_end = entry->vme_end;
+ if(map_pmap == NULL) {
+ if(entry->use_pmap) {
+ pmap = VME_SUBMAP(entry)->pmap;
+ pmap_addr = sub_start;
+ } else {
+ pmap = map->pmap;
+ pmap_addr = start;
+ }
+ if (entry->wired_count == 0 ||
+ (user_wire && entry->user_wired_count == 0)) {
+ if (!user_wire)
+ panic("vm_map_unwire: entry is unwired");
+ entry = entry->vme_next;
+ continue;
+ }
- entry->is_sub_map = FALSE;
- entry->object = copy_entry->object;
- object = entry->object.vm_object;
- entry->needs_copy = copy_entry->needs_copy;
- entry->wired_count = 0;
- entry->user_wired_count = 0;
- offset = entry->offset = copy_entry->offset;
+ /*
+ * Check for holes
+ * Holes: Next entry should be contiguous unless
+ * this is the end of the region.
+ */
+ if (((entry->vme_end < end) &&
+ ((entry->vme_next == vm_map_to_entry(map)) ||
+ (entry->vme_next->vme_start
+ > entry->vme_end)))) {
+ if (!user_wire)
+ panic("vm_map_unwire: non-contiguous region");
+/*
+ entry = entry->vme_next;
+ continue;
+*/
+ }
- vm_map_copy_entry_unlink(copy, copy_entry);
- vm_map_copy_entry_dispose(copy, copy_entry);
-#if BAD_OPTIMIZATION
- /*
- * if we turn this optimization back on
- * we need to revisit our use of pmap mappings
- * large copies will cause us to run out and panic
- * this optimization only saved on average 2 us per page if ALL
- * the pages in the source were currently mapped
- * and ALL the pages in the dest were touched, if there were fewer
- * than 2/3 of the pages touched, this optimization actually cost more cycles
- */
+ subtract_wire_counts(map, entry, user_wire);
- /*
- * Try to aggressively enter physical mappings
- * (but avoid uninstantiated objects)
- */
- if (object != VM_OBJECT_NULL) {
- vm_offset_t va = entry->vme_start;
+ if (entry->wired_count != 0) {
+ entry = entry->vme_next;
+ continue;
+ }
- while (va < entry->vme_end) {
- register vm_page_t m;
- vm_prot_t prot;
+ entry->in_transition = TRUE;
+ tmp_entry = *entry;/* see comment in vm_map_wire() */
/*
- * Look for the page in the top object
+ * We can unlock the map now. The in_transition state
+ * guarantees existance of the entry.
*/
- prot = entry->protection;
- vm_object_lock(object);
- vm_object_paging_begin(object);
+ vm_map_unlock(map);
+ vm_map_unwire_nested(VME_SUBMAP(entry),
+ sub_start, sub_end, user_wire, pmap, pmap_addr);
+ vm_map_lock(map);
- if ((m = vm_page_lookup(object,offset)) !=
- VM_PAGE_NULL && !m->busy &&
- !m->fictitious &&
- (!m->unusual || (!m->error &&
- !m->restart && !m->absent &&
- (prot & m->page_lock) == 0))) {
-
- m->busy = TRUE;
- vm_object_unlock(object);
-
- /*
- * Honor COW obligations
+ if (last_timestamp+1 != map->timestamp) {
+ /*
+ * Find the entry again. It could have been
+ * clipped or deleted after we unlocked the map.
*/
- if (entry->needs_copy)
- prot &= ~VM_PROT_WRITE;
- /* It is our policy to require */
- /* explicit sync from anyone */
- /* writing code and then */
- /* a pc to execute it. */
- /* No isync here */
-
- PMAP_ENTER(pmap, va, m, prot,
- ((unsigned int)
- (m->object->wimg_bits))
- & VM_WIMG_MASK,
- FALSE);
-
- vm_object_lock(object);
- vm_page_lock_queues();
- if (!m->active && !m->inactive)
- vm_page_activate(m);
- vm_page_unlock_queues();
- PAGE_WAKEUP_DONE(m);
+ if (!vm_map_lookup_entry(map,
+ tmp_entry.vme_start,
+ &first_entry)) {
+ if (!user_wire)
+ panic("vm_map_unwire: re-lookup failed");
+ entry = first_entry->vme_next;
+ } else
+ entry = first_entry;
}
- vm_object_paging_end(object);
- vm_object_unlock(object);
-
- offset += PAGE_SIZE_64;
- va += PAGE_SIZE;
- } /* end while (va < entry->vme_end) */
- } /* end if (object) */
-#endif
- /*
- * Set up for the next iteration. The map
- * has not been unlocked, so the next
- * address should be at the end of this
- * entry, and the next map entry should be
- * the one following it.
- */
+ last_timestamp = map->timestamp;
- start = tmp_entry->vme_end;
- tmp_entry = tmp_entry->vme_next;
- } else {
- vm_map_version_t version;
- vm_object_t dst_object = entry->object.vm_object;
- vm_object_offset_t dst_offset = entry->offset;
- kern_return_t r;
+ /*
+ * clear transition bit for all constituent entries
+ * that were in the original entry (saved in
+ * tmp_entry). Also check for waiters.
+ */
+ while ((entry != vm_map_to_entry(map)) &&
+ (entry->vme_start < tmp_entry.vme_end)) {
+ assert(entry->in_transition);
+ entry->in_transition = FALSE;
+ if (entry->needs_wakeup) {
+ entry->needs_wakeup = FALSE;
+ need_wakeup = TRUE;
+ }
+ entry = entry->vme_next;
+ }
+ continue;
+ } else {
+ vm_map_unlock(map);
+ vm_map_unwire_nested(VME_SUBMAP(entry),
+ sub_start, sub_end, user_wire, map_pmap,
+ pmap_addr);
+ vm_map_lock(map);
- /*
- * Take an object reference, and record
- * the map version information so that the
- * map can be safely unlocked.
- */
+ if (last_timestamp+1 != map->timestamp) {
+ /*
+ * Find the entry again. It could have been
+ * clipped or deleted after we unlocked the map.
+ */
+ if (!vm_map_lookup_entry(map,
+ tmp_entry.vme_start,
+ &first_entry)) {
+ if (!user_wire)
+ panic("vm_map_unwire: re-lookup failed");
+ entry = first_entry->vme_next;
+ } else
+ entry = first_entry;
+ }
+ last_timestamp = map->timestamp;
+ }
+ }
- vm_object_reference(dst_object);
- /* account for unlock bumping up timestamp */
- version.main_timestamp = dst_map->timestamp + 1;
+ if ((entry->wired_count == 0) ||
+ (user_wire && entry->user_wired_count == 0)) {
+ if (!user_wire)
+ panic("vm_map_unwire: entry is unwired");
- vm_map_unlock(dst_map);
+ entry = entry->vme_next;
+ continue;
+ }
+
+ assert(entry->wired_count > 0 &&
+ (!user_wire || entry->user_wired_count > 0));
- /*
- * Copy as much as possible in one pass
- */
+ vm_map_clip_start(map, entry, start);
+ vm_map_clip_end(map, entry, end);
- copy_size = size;
- r = vm_fault_copy(
- copy_entry->object.vm_object,
- copy_entry->offset,
- ©_size,
- dst_object,
- dst_offset,
- dst_map,
- &version,
- THREAD_UNINT );
+ /*
+ * Check for holes
+ * Holes: Next entry should be contiguous unless
+ * this is the end of the region.
+ */
+ if (((entry->vme_end < end) &&
+ ((entry->vme_next == vm_map_to_entry(map)) ||
+ (entry->vme_next->vme_start > entry->vme_end)))) {
- /*
- * Release the object reference
- */
+ if (!user_wire)
+ panic("vm_map_unwire: non-contiguous region");
+ entry = entry->vme_next;
+ continue;
+ }
- vm_object_deallocate(dst_object);
+ subtract_wire_counts(map, entry, user_wire);
- /*
- * If a hard error occurred, return it now
- */
+ if (entry->wired_count != 0) {
+ entry = entry->vme_next;
+ continue;
+ }
- if (r != KERN_SUCCESS)
- return(r);
+ if(entry->zero_wired_pages) {
+ entry->zero_wired_pages = FALSE;
+ }
- if (copy_size != 0) {
- /*
- * Dispose of the copied region
- */
+ entry->in_transition = TRUE;
+ tmp_entry = *entry; /* see comment in vm_map_wire() */
- vm_map_copy_clip_end(copy, copy_entry,
- copy_entry->vme_start + copy_size);
- vm_map_copy_entry_unlink(copy, copy_entry);
- vm_object_deallocate(copy_entry->object.vm_object);
- vm_map_copy_entry_dispose(copy, copy_entry);
- }
+ /*
+ * We can unlock the map now. The in_transition state
+ * guarantees existance of the entry.
+ */
+ vm_map_unlock(map);
+ if(map_pmap) {
+ vm_fault_unwire(map,
+ &tmp_entry, FALSE, map_pmap, pmap_addr);
+ } else {
+ vm_fault_unwire(map,
+ &tmp_entry, FALSE, map->pmap,
+ tmp_entry.vme_start);
+ }
+ vm_map_lock(map);
+ if (last_timestamp+1 != map->timestamp) {
/*
- * Pick up in the destination map where we left off.
- *
- * Use the version information to avoid a lookup
- * in the normal case.
+ * Find the entry again. It could have been clipped
+ * or deleted after we unlocked the map.
*/
+ if (!vm_map_lookup_entry(map, tmp_entry.vme_start,
+ &first_entry)) {
+ if (!user_wire)
+ panic("vm_map_unwire: re-lookup failed");
+ entry = first_entry->vme_next;
+ } else
+ entry = first_entry;
+ }
+ last_timestamp = map->timestamp;
- start += copy_size;
- vm_map_lock(dst_map);
- if (version.main_timestamp == dst_map->timestamp) {
- /* We can safely use saved tmp_entry value */
-
- vm_map_clip_end(dst_map, tmp_entry, start);
- tmp_entry = tmp_entry->vme_next;
- } else {
- /* Must do lookup of tmp_entry */
-
- if (!vm_map_lookup_entry(dst_map, start, &tmp_entry)) {
- vm_map_unlock(dst_map);
- return(KERN_INVALID_ADDRESS);
- }
- vm_map_clip_start(dst_map, tmp_entry, start);
+ /*
+ * clear transition bit for all constituent entries that
+ * were in the original entry (saved in tmp_entry). Also
+ * check for waiters.
+ */
+ while ((entry != vm_map_to_entry(map)) &&
+ (entry->vme_start < tmp_entry.vme_end)) {
+ assert(entry->in_transition);
+ entry->in_transition = FALSE;
+ if (entry->needs_wakeup) {
+ entry->needs_wakeup = FALSE;
+ need_wakeup = TRUE;
}
+ entry = entry->vme_next;
}
- }/* while */
+ }
+
+ /*
+ * We might have fragmented the address space when we wired this
+ * range of addresses. Attempt to re-coalesce these VM map entries
+ * with their neighbors now that they're no longer wired.
+ * Under some circumstances, address space fragmentation can
+ * prevent VM object shadow chain collapsing, which can cause
+ * swap space leaks.
+ */
+ vm_map_simplify_range(map, start, end);
+ vm_map_unlock(map);
+ /*
+ * wake up anybody waiting on entries that we have unwired.
+ */
+ if (need_wakeup)
+ vm_map_entry_wakeup(map);
return(KERN_SUCCESS);
-}/* vm_map_copy_overwrite_aligned */
+
+}
+
+kern_return_t
+vm_map_unwire(
+ register vm_map_t map,
+ register vm_map_offset_t start,
+ register vm_map_offset_t end,
+ boolean_t user_wire)
+{
+ return vm_map_unwire_nested(map, start, end,
+ user_wire, (pmap_t)NULL, 0);
+}
+
/*
- * Routine: vm_map_copyin_kernel_buffer
- *
- * Description:
- * Copy in data to a kernel buffer from space in the
- * source map. The original space may be otpionally
- * deallocated.
+ * vm_map_entry_delete: [ internal use only ]
*
- * If successful, returns a new copy object.
- */
-kern_return_t
-vm_map_copyin_kernel_buffer(
- vm_map_t src_map,
- vm_offset_t src_addr,
- vm_size_t len,
- boolean_t src_destroy,
- vm_map_copy_t *copy_result)
+ * Deallocate the given entry from the target map.
+ */
+static void
+vm_map_entry_delete(
+ register vm_map_t map,
+ register vm_map_entry_t entry)
{
- boolean_t flags;
- vm_map_copy_t copy;
- vm_size_t kalloc_size = sizeof(struct vm_map_copy) + len;
+ register vm_map_offset_t s, e;
+ register vm_object_t object;
+ register vm_map_t submap;
- copy = (vm_map_copy_t) kalloc(kalloc_size);
- if (copy == VM_MAP_COPY_NULL) {
- return KERN_RESOURCE_SHORTAGE;
+ s = entry->vme_start;
+ e = entry->vme_end;
+ assert(page_aligned(s));
+ assert(page_aligned(e));
+ if (entry->map_aligned == TRUE) {
+ assert(VM_MAP_PAGE_ALIGNED(s, VM_MAP_PAGE_MASK(map)));
+ assert(VM_MAP_PAGE_ALIGNED(e, VM_MAP_PAGE_MASK(map)));
}
- copy->type = VM_MAP_COPY_KERNEL_BUFFER;
- copy->size = len;
- copy->offset = 0;
- copy->cpy_kdata = (vm_offset_t) (copy + 1);
- copy->cpy_kalloc_size = kalloc_size;
+ assert(entry->wired_count == 0);
+ assert(entry->user_wired_count == 0);
+ assert(!entry->permanent);
- if (src_map == kernel_map) {
- bcopy((char *)src_addr, (char *)copy->cpy_kdata, len);
- flags = VM_MAP_REMOVE_KUNWIRE | VM_MAP_REMOVE_WAIT_FOR_KWIRE |
- VM_MAP_REMOVE_INTERRUPTIBLE;
+ if (entry->is_sub_map) {
+ object = NULL;
+ submap = VME_SUBMAP(entry);
} else {
- kern_return_t kr;
- kr = copyinmap(src_map, src_addr, copy->cpy_kdata, len);
- if (kr != KERN_SUCCESS) {
- kfree((vm_offset_t)copy, kalloc_size);
- return kr;
- }
- flags = VM_MAP_REMOVE_WAIT_FOR_KWIRE |
- VM_MAP_REMOVE_INTERRUPTIBLE;
- }
- if (src_destroy) {
- (void) vm_map_remove(src_map, trunc_page_32(src_addr),
- round_page_32(src_addr + len),
- flags);
+ submap = NULL;
+ object = VME_OBJECT(entry);
}
- *copy_result = copy;
- return KERN_SUCCESS;
+
+ vm_map_store_entry_unlink(map, entry);
+ map->size -= e - s;
+
+ vm_map_entry_dispose(map, entry);
+
+ vm_map_unlock(map);
+ /*
+ * Deallocate the object only after removing all
+ * pmap entries pointing to its pages.
+ */
+ if (submap)
+ vm_map_deallocate(submap);
+ else
+ vm_object_deallocate(object);
+
}
-/*
- * Routine: vm_map_copyout_kernel_buffer
- *
- * Description:
- * Copy out data from a kernel buffer into space in the
- * destination map. The space may be otpionally dynamically
- * allocated.
- *
- * If successful, consumes the copy object.
- * Otherwise, the caller is responsible for it.
- */
-kern_return_t
-vm_map_copyout_kernel_buffer(
+void
+vm_map_submap_pmap_clean(
vm_map_t map,
- vm_offset_t *addr, /* IN/OUT */
- vm_map_copy_t copy,
- boolean_t overwrite)
+ vm_map_offset_t start,
+ vm_map_offset_t end,
+ vm_map_t sub_map,
+ vm_map_offset_t offset)
{
- kern_return_t kr = KERN_SUCCESS;
- thread_act_t thr_act = current_act();
+ vm_map_offset_t submap_start;
+ vm_map_offset_t submap_end;
+ vm_map_size_t remove_size;
+ vm_map_entry_t entry;
- if (!overwrite) {
+ submap_end = offset + (end - start);
+ submap_start = offset;
- /*
- * Allocate space in the target map for the data
- */
- *addr = 0;
- kr = vm_map_enter(map,
- addr,
- round_page_32(copy->size),
- (vm_offset_t) 0,
- TRUE,
- VM_OBJECT_NULL,
- (vm_object_offset_t) 0,
- FALSE,
- VM_PROT_DEFAULT,
- VM_PROT_ALL,
- VM_INHERIT_DEFAULT);
- if (kr != KERN_SUCCESS)
- return(kr);
- }
+ vm_map_lock_read(sub_map);
+ if(vm_map_lookup_entry(sub_map, offset, &entry)) {
+
+ remove_size = (entry->vme_end - entry->vme_start);
+ if(offset > entry->vme_start)
+ remove_size -= offset - entry->vme_start;
+
- /*
- * Copyout the data from the kernel buffer to the target map.
- */
- if (thr_act->map == map) {
-
- /*
- * If the target map is the current map, just do
- * the copy.
- */
- if (copyout((char *)copy->cpy_kdata, (char *)*addr,
- copy->size)) {
- return(KERN_INVALID_ADDRESS);
+ if(submap_end < entry->vme_end) {
+ remove_size -=
+ entry->vme_end - submap_end;
}
- }
- else {
- vm_map_t oldmap;
-
- /*
- * If the target map is another map, assume the
- * target's address space identity for the duration
- * of the copy.
- */
- vm_map_reference(map);
- oldmap = vm_map_switch(map);
+ if(entry->is_sub_map) {
+ vm_map_submap_pmap_clean(
+ sub_map,
+ start,
+ start + remove_size,
+ VME_SUBMAP(entry),
+ VME_OFFSET(entry));
+ } else {
- if (copyout((char *)copy->cpy_kdata, (char *)*addr,
- copy->size)) {
- return(KERN_INVALID_ADDRESS);
+ if((map->mapped_in_other_pmaps) && (map->ref_count)
+ && (VME_OBJECT(entry) != NULL)) {
+ vm_object_pmap_protect_options(
+ VME_OBJECT(entry),
+ (VME_OFFSET(entry) +
+ offset -
+ entry->vme_start),
+ remove_size,
+ PMAP_NULL,
+ entry->vme_start,
+ VM_PROT_NONE,
+ PMAP_OPTIONS_REMOVE);
+ } else {
+ pmap_remove(map->pmap,
+ (addr64_t)start,
+ (addr64_t)(start + remove_size));
+ }
}
-
- (void) vm_map_switch(oldmap);
- vm_map_deallocate(map);
}
- kfree((vm_offset_t)copy, copy->cpy_kalloc_size);
-
- return(kr);
-}
-
-/*
- * Macro: vm_map_copy_insert
- *
- * Description:
- * Link a copy chain ("copy") into a map at the
- * specified location (after "where").
- * Side effects:
- * The copy chain is destroyed.
- * Warning:
- * The arguments are evaluated multiple times.
- */
-#define vm_map_copy_insert(map, where, copy) \
-MACRO_BEGIN \
- vm_map_t VMCI_map; \
- vm_map_entry_t VMCI_where; \
- vm_map_copy_t VMCI_copy; \
- VMCI_map = (map); \
- VMCI_where = (where); \
- VMCI_copy = (copy); \
- ((VMCI_where->vme_next)->vme_prev = vm_map_copy_last_entry(VMCI_copy))\
- ->vme_next = (VMCI_where->vme_next); \
- ((VMCI_where)->vme_next = vm_map_copy_first_entry(VMCI_copy)) \
- ->vme_prev = VMCI_where; \
- VMCI_map->hdr.nentries += VMCI_copy->cpy_hdr.nentries; \
- UPDATE_FIRST_FREE(VMCI_map, VMCI_map->first_free); \
- zfree(vm_map_copy_zone, (vm_offset_t) VMCI_copy); \
-MACRO_END
+ entry = entry->vme_next;
+
+ while((entry != vm_map_to_entry(sub_map))
+ && (entry->vme_start < submap_end)) {
+ remove_size = (entry->vme_end - entry->vme_start);
+ if(submap_end < entry->vme_end) {
+ remove_size -= entry->vme_end - submap_end;
+ }
+ if(entry->is_sub_map) {
+ vm_map_submap_pmap_clean(
+ sub_map,
+ (start + entry->vme_start) - offset,
+ ((start + entry->vme_start) - offset) + remove_size,
+ VME_SUBMAP(entry),
+ VME_OFFSET(entry));
+ } else {
+ if((map->mapped_in_other_pmaps) && (map->ref_count)
+ && (VME_OBJECT(entry) != NULL)) {
+ vm_object_pmap_protect_options(
+ VME_OBJECT(entry),
+ VME_OFFSET(entry),
+ remove_size,
+ PMAP_NULL,
+ entry->vme_start,
+ VM_PROT_NONE,
+ PMAP_OPTIONS_REMOVE);
+ } else {
+ pmap_remove(map->pmap,
+ (addr64_t)((start + entry->vme_start)
+ - offset),
+ (addr64_t)(((start + entry->vme_start)
+ - offset) + remove_size));
+ }
+ }
+ entry = entry->vme_next;
+ }
+ vm_map_unlock_read(sub_map);
+ return;
+}
/*
- * Routine: vm_map_copyout
+ * vm_map_delete: [ internal use only ]
*
- * Description:
- * Copy out a copy chain ("copy") into newly-allocated
- * space in the destination map.
+ * Deallocates the given address range from the target map.
+ * Removes all user wirings. Unwires one kernel wiring if
+ * VM_MAP_REMOVE_KUNWIRE is set. Waits for kernel wirings to go
+ * away if VM_MAP_REMOVE_WAIT_FOR_KWIRE is set. Sleeps
+ * interruptibly if VM_MAP_REMOVE_INTERRUPTIBLE is set.
*
- * If successful, consumes the copy object.
- * Otherwise, the caller is responsible for it.
+ * This routine is called with map locked and leaves map locked.
*/
-kern_return_t
-vm_map_copyout(
- register vm_map_t dst_map,
- vm_offset_t *dst_addr, /* OUT */
- register vm_map_copy_t copy)
+static kern_return_t
+vm_map_delete(
+ vm_map_t map,
+ vm_map_offset_t start,
+ vm_map_offset_t end,
+ int flags,
+ vm_map_t zap_map)
{
- vm_size_t size;
- vm_size_t adjustment;
- vm_offset_t start;
- vm_object_offset_t vm_copy_start;
- vm_map_entry_t last;
- register
- vm_map_entry_t entry;
-
- /*
- * Check for null copy object.
- */
+ vm_map_entry_t entry, next;
+ struct vm_map_entry *first_entry, tmp_entry;
+ register vm_map_offset_t s;
+ register vm_object_t object;
+ boolean_t need_wakeup;
+ unsigned int last_timestamp = ~0; /* unlikely value */
+ int interruptible;
- if (copy == VM_MAP_COPY_NULL) {
- *dst_addr = 0;
- return(KERN_SUCCESS);
- }
+ interruptible = (flags & VM_MAP_REMOVE_INTERRUPTIBLE) ?
+ THREAD_ABORTSAFE : THREAD_UNINT;
/*
- * Check for special copy object, created
- * by vm_map_copyin_object.
+ * All our DMA I/O operations in IOKit are currently done by
+ * wiring through the map entries of the task requesting the I/O.
+ * Because of this, we must always wait for kernel wirings
+ * to go away on the entries before deleting them.
+ *
+ * Any caller who wants to actually remove a kernel wiring
+ * should explicitly set the VM_MAP_REMOVE_KUNWIRE flag to
+ * properly remove one wiring instead of blasting through
+ * them all.
*/
+ flags |= VM_MAP_REMOVE_WAIT_FOR_KWIRE;
- if (copy->type == VM_MAP_COPY_OBJECT) {
- vm_object_t object = copy->cpy_object;
- kern_return_t kr;
- vm_object_offset_t offset;
-
- offset = trunc_page_64(copy->offset);
- size = round_page_32(copy->size +
- (vm_size_t)(copy->offset - offset));
- *dst_addr = 0;
- kr = vm_map_enter(dst_map, dst_addr, size,
- (vm_offset_t) 0, TRUE,
- object, offset, FALSE,
- VM_PROT_DEFAULT, VM_PROT_ALL,
- VM_INHERIT_DEFAULT);
- if (kr != KERN_SUCCESS)
- return(kr);
- /* Account for non-pagealigned copy object */
- *dst_addr += (vm_offset_t)(copy->offset - offset);
- zfree(vm_map_copy_zone, (vm_offset_t) copy);
- return(KERN_SUCCESS);
- }
-
- /*
- * Check for special kernel buffer allocated
- * by new_ipc_kmsg_copyin.
- */
+ while(1) {
+ /*
+ * Find the start of the region, and clip it
+ */
+ if (vm_map_lookup_entry(map, start, &first_entry)) {
+ entry = first_entry;
+ if (map == kalloc_map &&
+ (entry->vme_start != start ||
+ entry->vme_end != end)) {
+ panic("vm_map_delete(%p,0x%llx,0x%llx): "
+ "mismatched entry %p [0x%llx:0x%llx]\n",
+ map,
+ (uint64_t)start,
+ (uint64_t)end,
+ entry,
+ (uint64_t)entry->vme_start,
+ (uint64_t)entry->vme_end);
+ }
+ if (entry->superpage_size && (start & ~SUPERPAGE_MASK)) { /* extend request to whole entry */ start = SUPERPAGE_ROUND_DOWN(start);
+ start = SUPERPAGE_ROUND_DOWN(start);
+ continue;
+ }
+ if (start == entry->vme_start) {
+ /*
+ * No need to clip. We don't want to cause
+ * any unnecessary unnesting in this case...
+ */
+ } else {
+ if ((flags & VM_MAP_REMOVE_NO_MAP_ALIGN) &&
+ entry->map_aligned &&
+ !VM_MAP_PAGE_ALIGNED(
+ start,
+ VM_MAP_PAGE_MASK(map))) {
+ /*
+ * The entry will no longer be
+ * map-aligned after clipping
+ * and the caller said it's OK.
+ */
+ entry->map_aligned = FALSE;
+ }
+ if (map == kalloc_map) {
+ panic("vm_map_delete(%p,0x%llx,0x%llx):"
+ " clipping %p at 0x%llx\n",
+ map,
+ (uint64_t)start,
+ (uint64_t)end,
+ entry,
+ (uint64_t)start);
+ }
+ vm_map_clip_start(map, entry, start);
+ }
- if (copy->type == VM_MAP_COPY_KERNEL_BUFFER) {
- return(vm_map_copyout_kernel_buffer(dst_map, dst_addr,
- copy, FALSE));
+ /*
+ * Fix the lookup hint now, rather than each
+ * time through the loop.
+ */
+ SAVE_HINT_MAP_WRITE(map, entry->vme_prev);
+ } else {
+ if (map->pmap == kernel_pmap &&
+ map->ref_count != 0) {
+ panic("vm_map_delete(%p,0x%llx,0x%llx): "
+ "no map entry at 0x%llx\n",
+ map,
+ (uint64_t)start,
+ (uint64_t)end,
+ (uint64_t)start);
+ }
+ entry = first_entry->vme_next;
+ }
+ break;
}
+ if (entry->superpage_size)
+ end = SUPERPAGE_ROUND_UP(end);
+ need_wakeup = FALSE;
/*
- * Find space for the data
+ * Step through all entries in this region
*/
+ s = entry->vme_start;
+ while ((entry != vm_map_to_entry(map)) && (s < end)) {
+ /*
+ * At this point, we have deleted all the memory entries
+ * between "start" and "s". We still need to delete
+ * all memory entries between "s" and "end".
+ * While we were blocked and the map was unlocked, some
+ * new memory entries could have been re-allocated between
+ * "start" and "s" and we don't want to mess with those.
+ * Some of those entries could even have been re-assembled
+ * with an entry after "s" (in vm_map_simplify_entry()), so
+ * we may have to vm_map_clip_start() again.
+ */
- vm_copy_start = trunc_page_64(copy->offset);
- size = round_page_32((vm_size_t)copy->offset + copy->size)
- - vm_copy_start;
-
- StartAgain: ;
-
- vm_map_lock(dst_map);
- assert(first_free_is_valid(dst_map));
- start = ((last = dst_map->first_free) == vm_map_to_entry(dst_map)) ?
- vm_map_min(dst_map) : last->vme_end;
-
- while (TRUE) {
- vm_map_entry_t next = last->vme_next;
- vm_offset_t end = start + size;
-
- if ((end > dst_map->max_offset) || (end < start)) {
- if (dst_map->wait_for_space) {
- if (size <= (dst_map->max_offset - dst_map->min_offset)) {
- assert_wait((event_t) dst_map,
- THREAD_INTERRUPTIBLE);
- vm_map_unlock(dst_map);
- thread_block((void (*)(void))0);
- goto StartAgain;
- }
+ if (entry->vme_start >= s) {
+ /*
+ * This entry starts on or after "s"
+ * so no need to clip its start.
+ */
+ } else {
+ /*
+ * This entry has been re-assembled by a
+ * vm_map_simplify_entry(). We need to
+ * re-clip its start.
+ */
+ if ((flags & VM_MAP_REMOVE_NO_MAP_ALIGN) &&
+ entry->map_aligned &&
+ !VM_MAP_PAGE_ALIGNED(s,
+ VM_MAP_PAGE_MASK(map))) {
+ /*
+ * The entry will no longer be map-aligned
+ * after clipping and the caller said it's OK.
+ */
+ entry->map_aligned = FALSE;
}
- vm_map_unlock(dst_map);
- return(KERN_NO_SPACE);
+ if (map == kalloc_map) {
+ panic("vm_map_delete(%p,0x%llx,0x%llx): "
+ "clipping %p at 0x%llx\n",
+ map,
+ (uint64_t)start,
+ (uint64_t)end,
+ entry,
+ (uint64_t)s);
+ }
+ vm_map_clip_start(map, entry, s);
+ }
+ if (entry->vme_end <= end) {
+ /*
+ * This entry is going away completely, so no need
+ * to clip and possibly cause an unnecessary unnesting.
+ */
+ } else {
+ if ((flags & VM_MAP_REMOVE_NO_MAP_ALIGN) &&
+ entry->map_aligned &&
+ !VM_MAP_PAGE_ALIGNED(end,
+ VM_MAP_PAGE_MASK(map))) {
+ /*
+ * The entry will no longer be map-aligned
+ * after clipping and the caller said it's OK.
+ */
+ entry->map_aligned = FALSE;
+ }
+ if (map == kalloc_map) {
+ panic("vm_map_delete(%p,0x%llx,0x%llx): "
+ "clipping %p at 0x%llx\n",
+ map,
+ (uint64_t)start,
+ (uint64_t)end,
+ entry,
+ (uint64_t)end);
+ }
+ vm_map_clip_end(map, entry, end);
}
- if ((next == vm_map_to_entry(dst_map)) ||
- (next->vme_start >= end))
- break;
-
- last = next;
- start = last->vme_end;
- }
+ if (entry->permanent) {
+ panic("attempt to remove permanent VM map entry "
+ "%p [0x%llx:0x%llx]\n",
+ entry, (uint64_t) s, (uint64_t) end);
+ }
- /*
- * Since we're going to just drop the map
- * entries from the copy into the destination
- * map, they must come from the same pool.
- */
- if (copy->cpy_hdr.entries_pageable != dst_map->hdr.entries_pageable) {
- /*
- * Mismatches occur when dealing with the default
- * pager.
- */
- zone_t old_zone;
- vm_map_entry_t next, new;
-
- /*
- * Find the zone that the copies were allocated from
- */
- old_zone = (copy->cpy_hdr.entries_pageable)
- ? vm_map_entry_zone
- : vm_map_kentry_zone;
- entry = vm_map_copy_first_entry(copy);
-
- /*
- * Reinitialize the copy so that vm_map_copy_entry_link
- * will work.
- */
- copy->cpy_hdr.nentries = 0;
- copy->cpy_hdr.entries_pageable = dst_map->hdr.entries_pageable;
- vm_map_copy_first_entry(copy) =
- vm_map_copy_last_entry(copy) =
- vm_map_copy_to_entry(copy);
-
- /*
- * Copy each entry.
- */
- while (entry != vm_map_copy_to_entry(copy)) {
- new = vm_map_copy_entry_create(copy);
- vm_map_entry_copy_full(new, entry);
- new->use_pmap = FALSE; /* clr address space specifics */
- vm_map_copy_entry_link(copy,
- vm_map_copy_last_entry(copy),
- new);
- next = entry->vme_next;
- zfree(old_zone, (vm_offset_t) entry);
- entry = next;
- }
- }
+ if (entry->in_transition) {
+ wait_result_t wait_result;
- /*
- * Adjust the addresses in the copy chain, and
- * reset the region attributes.
- */
+ /*
+ * Another thread is wiring/unwiring this entry.
+ * Let the other thread know we are waiting.
+ */
+ assert(s == entry->vme_start);
+ entry->needs_wakeup = TRUE;
- adjustment = start - vm_copy_start;
- for (entry = vm_map_copy_first_entry(copy);
- entry != vm_map_copy_to_entry(copy);
- entry = entry->vme_next) {
- entry->vme_start += adjustment;
- entry->vme_end += adjustment;
+ /*
+ * wake up anybody waiting on entries that we have
+ * already unwired/deleted.
+ */
+ if (need_wakeup) {
+ vm_map_entry_wakeup(map);
+ need_wakeup = FALSE;
+ }
- entry->inheritance = VM_INHERIT_DEFAULT;
- entry->protection = VM_PROT_DEFAULT;
- entry->max_protection = VM_PROT_ALL;
- entry->behavior = VM_BEHAVIOR_DEFAULT;
+ wait_result = vm_map_entry_wait(map, interruptible);
- /*
- * If the entry is now wired,
- * map the pages into the destination map.
- */
- if (entry->wired_count != 0) {
- register vm_offset_t va;
- vm_object_offset_t offset;
- register vm_object_t object;
+ if (interruptible &&
+ wait_result == THREAD_INTERRUPTED) {
+ /*
+ * We do not clear the needs_wakeup flag,
+ * since we cannot tell if we were the only one.
+ */
+ return KERN_ABORTED;
+ }
- object = entry->object.vm_object;
- offset = entry->offset;
- va = entry->vme_start;
+ /*
+ * The entry could have been clipped or it
+ * may not exist anymore. Look it up again.
+ */
+ if (!vm_map_lookup_entry(map, s, &first_entry)) {
+ /*
+ * User: use the next entry
+ */
+ entry = first_entry->vme_next;
+ s = entry->vme_start;
+ } else {
+ entry = first_entry;
+ SAVE_HINT_MAP_WRITE(map, entry->vme_prev);
+ }
+ last_timestamp = map->timestamp;
+ continue;
+ } /* end in_transition */
- pmap_pageable(dst_map->pmap,
- entry->vme_start,
- entry->vme_end,
- TRUE);
+ if (entry->wired_count) {
+ boolean_t user_wire;
- while (va < entry->vme_end) {
- register vm_page_t m;
+ user_wire = entry->user_wired_count > 0;
/*
- * Look up the page in the object.
- * Assert that the page will be found in the
- * top object:
- * either
- * the object was newly created by
- * vm_object_copy_slowly, and has
- * copies of all of the pages from
- * the source object
- * or
- * the object was moved from the old
- * map entry; because the old map
- * entry was wired, all of the pages
- * were in the top-level object.
- * (XXX not true if we wire pages for
- * reading)
+ * Remove a kernel wiring if requested
*/
- vm_object_lock(object);
- vm_object_paging_begin(object);
+ if (flags & VM_MAP_REMOVE_KUNWIRE) {
+ entry->wired_count--;
+ }
+
+ /*
+ * Remove all user wirings for proper accounting
+ */
+ if (entry->user_wired_count > 0) {
+ while (entry->user_wired_count)
+ subtract_wire_counts(map, entry, user_wire);
+ }
- m = vm_page_lookup(object, offset);
- if (m == VM_PAGE_NULL || m->wire_count == 0 ||
- m->absent)
- panic("vm_map_copyout: wiring 0x%x", m);
+ if (entry->wired_count != 0) {
+ assert(map != kernel_map);
+ /*
+ * Cannot continue. Typical case is when
+ * a user thread has physical io pending on
+ * on this page. Either wait for the
+ * kernel wiring to go away or return an
+ * error.
+ */
+ if (flags & VM_MAP_REMOVE_WAIT_FOR_KWIRE) {
+ wait_result_t wait_result;
- m->busy = TRUE;
- vm_object_unlock(object);
+ assert(s == entry->vme_start);
+ entry->needs_wakeup = TRUE;
+ wait_result = vm_map_entry_wait(map,
+ interruptible);
- PMAP_ENTER(dst_map->pmap, va, m, entry->protection,
- ((unsigned int)
- (m->object->wimg_bits))
- & VM_WIMG_MASK,
- TRUE);
+ if (interruptible &&
+ wait_result == THREAD_INTERRUPTED) {
+ /*
+ * We do not clear the
+ * needs_wakeup flag, since we
+ * cannot tell if we were the
+ * only one.
+ */
+ return KERN_ABORTED;
+ }
- vm_object_lock(object);
- PAGE_WAKEUP_DONE(m);
- /* the page is wired, so we don't have to activate */
- vm_object_paging_end(object);
- vm_object_unlock(object);
-
- offset += PAGE_SIZE_64;
- va += PAGE_SIZE;
- }
- }
- else if (size <= vm_map_aggressive_enter_max) {
-
- register vm_offset_t va;
- vm_object_offset_t offset;
- register vm_object_t object;
- vm_prot_t prot;
-
- object = entry->object.vm_object;
- if (object != VM_OBJECT_NULL) {
-
- offset = entry->offset;
- va = entry->vme_start;
- while (va < entry->vme_end) {
- register vm_page_t m;
-
/*
- * Look up the page in the object.
- * Assert that the page will be found
- * in the top object if at all...
+ * The entry could have been clipped or
+ * it may not exist anymore. Look it
+ * up again.
*/
- vm_object_lock(object);
- vm_object_paging_begin(object);
-
- if (((m = vm_page_lookup(object,
- offset))
- != VM_PAGE_NULL) &&
- !m->busy && !m->fictitious &&
- !m->absent && !m->error) {
- m->busy = TRUE;
- vm_object_unlock(object);
-
- /* honor cow obligations */
- prot = entry->protection;
- if (entry->needs_copy)
- prot &= ~VM_PROT_WRITE;
-
- PMAP_ENTER(dst_map->pmap, va,
- m, prot,
- ((unsigned int)
- (m->object->wimg_bits))
- & VM_WIMG_MASK,
- FALSE);
-
- vm_object_lock(object);
- vm_page_lock_queues();
- if (!m->active && !m->inactive)
- vm_page_activate(m);
- vm_page_unlock_queues();
- PAGE_WAKEUP_DONE(m);
+ if (!vm_map_lookup_entry(map, s,
+ &first_entry)) {
+ assert(map != kernel_map);
+ /*
+ * User: use the next entry
+ */
+ entry = first_entry->vme_next;
+ s = entry->vme_start;
+ } else {
+ entry = first_entry;
+ SAVE_HINT_MAP_WRITE(map, entry->vme_prev);
}
- vm_object_paging_end(object);
- vm_object_unlock(object);
-
- offset += PAGE_SIZE_64;
- va += PAGE_SIZE;
+ last_timestamp = map->timestamp;
+ continue;
+ }
+ else {
+ return KERN_FAILURE;
}
}
- }
- }
- /*
- * Correct the page alignment for the result
- */
+ entry->in_transition = TRUE;
+ /*
+ * copy current entry. see comment in vm_map_wire()
+ */
+ tmp_entry = *entry;
+ assert(s == entry->vme_start);
- *dst_addr = start + (copy->offset - vm_copy_start);
+ /*
+ * We can unlock the map now. The in_transition
+ * state guarentees existance of the entry.
+ */
+ vm_map_unlock(map);
- /*
- * Update the hints and the map size
- */
+ if (tmp_entry.is_sub_map) {
+ vm_map_t sub_map;
+ vm_map_offset_t sub_start, sub_end;
+ pmap_t pmap;
+ vm_map_offset_t pmap_addr;
+
- SAVE_HINT(dst_map, vm_map_copy_last_entry(copy));
+ sub_map = VME_SUBMAP(&tmp_entry);
+ sub_start = VME_OFFSET(&tmp_entry);
+ sub_end = sub_start + (tmp_entry.vme_end -
+ tmp_entry.vme_start);
+ if (tmp_entry.use_pmap) {
+ pmap = sub_map->pmap;
+ pmap_addr = tmp_entry.vme_start;
+ } else {
+ pmap = map->pmap;
+ pmap_addr = tmp_entry.vme_start;
+ }
+ (void) vm_map_unwire_nested(sub_map,
+ sub_start, sub_end,
+ user_wire,
+ pmap, pmap_addr);
+ } else {
- dst_map->size += size;
+ if (VME_OBJECT(&tmp_entry) == kernel_object) {
+ pmap_protect_options(
+ map->pmap,
+ tmp_entry.vme_start,
+ tmp_entry.vme_end,
+ VM_PROT_NONE,
+ PMAP_OPTIONS_REMOVE,
+ NULL);
+ }
+ vm_fault_unwire(map, &tmp_entry,
+ VME_OBJECT(&tmp_entry) == kernel_object,
+ map->pmap, tmp_entry.vme_start);
+ }
- /*
- * Link in the copy
- */
+ vm_map_lock(map);
- vm_map_copy_insert(dst_map, last, copy);
+ if (last_timestamp+1 != map->timestamp) {
+ /*
+ * Find the entry again. It could have
+ * been clipped after we unlocked the map.
+ */
+ if (!vm_map_lookup_entry(map, s, &first_entry)){
+ assert((map != kernel_map) &&
+ (!entry->is_sub_map));
+ first_entry = first_entry->vme_next;
+ s = first_entry->vme_start;
+ } else {
+ SAVE_HINT_MAP_WRITE(map, entry->vme_prev);
+ }
+ } else {
+ SAVE_HINT_MAP_WRITE(map, entry->vme_prev);
+ first_entry = entry;
+ }
- vm_map_unlock(dst_map);
+ last_timestamp = map->timestamp;
- /*
- * XXX If wiring_required, call vm_map_pageable
- */
+ entry = first_entry;
+ while ((entry != vm_map_to_entry(map)) &&
+ (entry->vme_start < tmp_entry.vme_end)) {
+ assert(entry->in_transition);
+ entry->in_transition = FALSE;
+ if (entry->needs_wakeup) {
+ entry->needs_wakeup = FALSE;
+ need_wakeup = TRUE;
+ }
+ entry = entry->vme_next;
+ }
+ /*
+ * We have unwired the entry(s). Go back and
+ * delete them.
+ */
+ entry = first_entry;
+ continue;
+ }
- return(KERN_SUCCESS);
-}
+ /* entry is unwired */
+ assert(entry->wired_count == 0);
+ assert(entry->user_wired_count == 0);
-boolean_t vm_map_aggressive_enter; /* not used yet */
+ assert(s == entry->vme_start);
+ if (flags & VM_MAP_REMOVE_NO_PMAP_CLEANUP) {
+ /*
+ * XXX with the VM_MAP_REMOVE_SAVE_ENTRIES flag to
+ * vm_map_delete(), some map entries might have been
+ * transferred to a "zap_map", which doesn't have a
+ * pmap. The original pmap has already been flushed
+ * in the vm_map_delete() call targeting the original
+ * map, but when we get to destroying the "zap_map",
+ * we don't have any pmap to flush, so let's just skip
+ * all this.
+ */
+ } else if (entry->is_sub_map) {
+ if (entry->use_pmap) {
+#ifndef NO_NESTED_PMAP
+ int pmap_flags;
-/*
- * Routine: vm_map_copyin
- *
- * Description:
- * Copy the specified region (src_addr, len) from the
- * source address space (src_map), possibly removing
- * the region from the source address space (src_destroy).
- *
- * Returns:
- * A vm_map_copy_t object (copy_result), suitable for
- * insertion into another address space (using vm_map_copyout),
- * copying over another address space region (using
- * vm_map_copy_overwrite). If the copy is unused, it
- * should be destroyed (using vm_map_copy_discard).
- *
- * In/out conditions:
- * The source map should not be locked on entry.
- */
+ if (flags & VM_MAP_REMOVE_NO_UNNESTING) {
+ /*
+ * This is the final cleanup of the
+ * address space being terminated.
+ * No new mappings are expected and
+ * we don't really need to unnest the
+ * shared region (and lose the "global"
+ * pmap mappings, if applicable).
+ *
+ * Tell the pmap layer that we're
+ * "clean" wrt nesting.
+ */
+ pmap_flags = PMAP_UNNEST_CLEAN;
+ } else {
+ /*
+ * We're unmapping part of the nested
+ * shared region, so we can't keep the
+ * nested pmap.
+ */
+ pmap_flags = 0;
+ }
+ pmap_unnest_options(
+ map->pmap,
+ (addr64_t)entry->vme_start,
+ entry->vme_end - entry->vme_start,
+ pmap_flags);
+#endif /* NO_NESTED_PMAP */
+ if ((map->mapped_in_other_pmaps) && (map->ref_count)) {
+ /* clean up parent map/maps */
+ vm_map_submap_pmap_clean(
+ map, entry->vme_start,
+ entry->vme_end,
+ VME_SUBMAP(entry),
+ VME_OFFSET(entry));
+ }
+ } else {
+ vm_map_submap_pmap_clean(
+ map, entry->vme_start, entry->vme_end,
+ VME_SUBMAP(entry),
+ VME_OFFSET(entry));
+ }
+ } else if (VME_OBJECT(entry) != kernel_object &&
+ VME_OBJECT(entry) != compressor_object) {
+ object = VME_OBJECT(entry);
+ if ((map->mapped_in_other_pmaps) && (map->ref_count)) {
+ vm_object_pmap_protect_options(
+ object, VME_OFFSET(entry),
+ entry->vme_end - entry->vme_start,
+ PMAP_NULL,
+ entry->vme_start,
+ VM_PROT_NONE,
+ PMAP_OPTIONS_REMOVE);
+ } else if ((VME_OBJECT(entry) != VM_OBJECT_NULL) ||
+ (map->pmap == kernel_pmap)) {
+ /* Remove translations associated
+ * with this range unless the entry
+ * does not have an object, or
+ * it's the kernel map or a descendant
+ * since the platform could potentially
+ * create "backdoor" mappings invisible
+ * to the VM. It is expected that
+ * objectless, non-kernel ranges
+ * do not have such VM invisible
+ * translations.
+ */
+ pmap_remove_options(map->pmap,
+ (addr64_t)entry->vme_start,
+ (addr64_t)entry->vme_end,
+ PMAP_OPTIONS_REMOVE);
+ }
+ }
-typedef struct submap_map {
- vm_map_t parent_map;
- vm_offset_t base_start;
- vm_offset_t base_end;
- struct submap_map *next;
-} submap_map_t;
+ if (entry->iokit_acct) {
+ /* alternate accounting */
+ DTRACE_VM4(vm_map_iokit_unmapped_region,
+ vm_map_t, map,
+ vm_map_offset_t, entry->vme_start,
+ vm_map_offset_t, entry->vme_end,
+ int, VME_ALIAS(entry));
+ vm_map_iokit_unmapped_region(map,
+ (entry->vme_end -
+ entry->vme_start));
+ entry->iokit_acct = FALSE;
+ }
-kern_return_t
-vm_map_copyin_common(
- vm_map_t src_map,
- vm_offset_t src_addr,
- vm_size_t len,
- boolean_t src_destroy,
- boolean_t src_volatile,
- vm_map_copy_t *copy_result, /* OUT */
- boolean_t use_maxprot)
-{
- extern int msg_ool_size_small;
+ /*
+ * All pmap mappings for this map entry must have been
+ * cleared by now.
+ */
+#if DEBUG
+ assert(vm_map_pmap_is_empty(map,
+ entry->vme_start,
+ entry->vme_end));
+#endif /* DEBUG */
- vm_map_entry_t tmp_entry; /* Result of last map lookup --
- * in multi-level lookup, this
- * entry contains the actual
- * vm_object/offset.
- */
- register
- vm_map_entry_t new_entry = VM_MAP_ENTRY_NULL; /* Map entry for copy */
+ next = entry->vme_next;
- vm_offset_t src_start; /* Start of current entry --
- * where copy is taking place now
- */
- vm_offset_t src_end; /* End of entire region to be
- * copied */
- vm_offset_t base_start; /* submap fields to save offsets */
- /* in original map */
- vm_offset_t base_end;
- vm_map_t base_map=src_map;
- vm_map_entry_t base_entry;
- boolean_t map_share=FALSE;
- submap_map_t *parent_maps = NULL;
+ if (map->pmap == kernel_pmap &&
+ map->ref_count != 0 &&
+ entry->vme_end < end &&
+ (next == vm_map_to_entry(map) ||
+ next->vme_start != entry->vme_end)) {
+ panic("vm_map_delete(%p,0x%llx,0x%llx): "
+ "hole after %p at 0x%llx\n",
+ map,
+ (uint64_t)start,
+ (uint64_t)end,
+ entry,
+ (uint64_t)entry->vme_end);
+ }
- register
- vm_map_copy_t copy; /* Resulting copy */
- vm_offset_t copy_addr;
+ s = next->vme_start;
+ last_timestamp = map->timestamp;
- /*
- * Check for copies of zero bytes.
- */
+ if ((flags & VM_MAP_REMOVE_SAVE_ENTRIES) &&
+ zap_map != VM_MAP_NULL) {
+ vm_map_size_t entry_size;
+ /*
+ * The caller wants to save the affected VM map entries
+ * into the "zap_map". The caller will take care of
+ * these entries.
+ */
+ /* unlink the entry from "map" ... */
+ vm_map_store_entry_unlink(map, entry);
+ /* ... and add it to the end of the "zap_map" */
+ vm_map_store_entry_link(zap_map,
+ vm_map_last_entry(zap_map),
+ entry);
+ entry_size = entry->vme_end - entry->vme_start;
+ map->size -= entry_size;
+ zap_map->size += entry_size;
+ /* we didn't unlock the map, so no timestamp increase */
+ last_timestamp--;
+ } else {
+ vm_map_entry_delete(map, entry);
+ /* vm_map_entry_delete unlocks the map */
+ vm_map_lock(map);
+ }
- if (len == 0) {
- *copy_result = VM_MAP_COPY_NULL;
- return(KERN_SUCCESS);
+ entry = next;
+
+ if(entry == vm_map_to_entry(map)) {
+ break;
+ }
+ if (last_timestamp+1 != map->timestamp) {
+ /*
+ * we are responsible for deleting everything
+ * from the give space, if someone has interfered
+ * we pick up where we left off, back fills should
+ * be all right for anyone except map_delete and
+ * we have to assume that the task has been fully
+ * disabled before we get here
+ */
+ if (!vm_map_lookup_entry(map, s, &entry)){
+ entry = entry->vme_next;
+ s = entry->vme_start;
+ } else {
+ SAVE_HINT_MAP_WRITE(map, entry->vme_prev);
+ }
+ /*
+ * others can not only allocate behind us, we can
+ * also see coalesce while we don't have the map lock
+ */
+ if(entry == vm_map_to_entry(map)) {
+ break;
+ }
+ }
+ last_timestamp = map->timestamp;
}
+ if (map->wait_for_space)
+ thread_wakeup((event_t) map);
/*
- * Check that the end address doesn't overflow
+ * wake up anybody waiting on entries that we have already deleted.
*/
- src_end = src_addr + len;
- if (src_end < src_addr)
- return KERN_INVALID_ADDRESS;
+ if (need_wakeup)
+ vm_map_entry_wakeup(map);
- /*
- * If the copy is sufficiently small, use a kernel buffer instead
- * of making a virtual copy. The theory being that the cost of
- * setting up VM (and taking C-O-W faults) dominates the copy costs
- * for small regions.
- */
- if ((len < msg_ool_size_small) && !use_maxprot)
- return vm_map_copyin_kernel_buffer(src_map, src_addr, len,
- src_destroy, copy_result);
+ return KERN_SUCCESS;
+}
+
+/*
+ * vm_map_remove:
+ *
+ * Remove the given address range from the target map.
+ * This is the exported form of vm_map_delete.
+ */
+kern_return_t
+vm_map_remove(
+ register vm_map_t map,
+ register vm_map_offset_t start,
+ register vm_map_offset_t end,
+ register boolean_t flags)
+{
+ register kern_return_t result;
+ vm_map_lock(map);
+ VM_MAP_RANGE_CHECK(map, start, end);
/*
- * Compute (page aligned) start and end of region
+ * For the zone_map, the kernel controls the allocation/freeing of memory.
+ * Any free to the zone_map should be within the bounds of the map and
+ * should free up memory. If the VM_MAP_RANGE_CHECK() silently converts a
+ * free to the zone_map into a no-op, there is a problem and we should
+ * panic.
*/
- src_start = trunc_page_32(src_addr);
- src_end = round_page_32(src_end);
+ if ((map == zone_map) && (start == end))
+ panic("Nothing being freed to the zone_map. start = end = %p\n", (void *)start);
+ result = vm_map_delete(map, start, end, flags, VM_MAP_NULL);
+ vm_map_unlock(map);
- XPR(XPR_VM_MAP, "vm_map_copyin_common map 0x%x addr 0x%x len 0x%x dest %d\n", (natural_t)src_map, src_addr, len, src_destroy, 0);
+ return(result);
+}
- /*
- * Allocate a header element for the list.
- *
- * Use the start and end in the header to
- * remember the endpoints prior to rounding.
- */
- copy = (vm_map_copy_t) zalloc(vm_map_copy_zone);
- vm_map_copy_first_entry(copy) =
- vm_map_copy_last_entry(copy) = vm_map_copy_to_entry(copy);
- copy->type = VM_MAP_COPY_ENTRY_LIST;
- copy->cpy_hdr.nentries = 0;
- copy->cpy_hdr.entries_pageable = TRUE;
+/*
+ * Routine: vm_map_copy_discard
+ *
+ * Description:
+ * Dispose of a map copy object (returned by
+ * vm_map_copyin).
+ */
+void
+vm_map_copy_discard(
+ vm_map_copy_t copy)
+{
+ if (copy == VM_MAP_COPY_NULL)
+ return;
- copy->offset = src_addr;
- copy->size = len;
-
- new_entry = vm_map_copy_entry_create(copy);
+ switch (copy->type) {
+ case VM_MAP_COPY_ENTRY_LIST:
+ while (vm_map_copy_first_entry(copy) !=
+ vm_map_copy_to_entry(copy)) {
+ vm_map_entry_t entry = vm_map_copy_first_entry(copy);
-#define RETURN(x) \
- MACRO_BEGIN \
- vm_map_unlock(src_map); \
- if(src_map != base_map) \
- vm_map_deallocate(src_map); \
- if (new_entry != VM_MAP_ENTRY_NULL) \
- vm_map_copy_entry_dispose(copy,new_entry); \
- vm_map_copy_discard(copy); \
- { \
- submap_map_t *ptr; \
- \
- for(ptr = parent_maps; ptr != NULL; ptr = parent_maps) { \
- parent_maps=parent_maps->next; \
- if (ptr->parent_map != base_map) \
- vm_map_deallocate(ptr->parent_map); \
- kfree((vm_offset_t)ptr, sizeof(submap_map_t)); \
- } \
- } \
- MACRO_RETURN(x); \
- MACRO_END
+ vm_map_copy_entry_unlink(copy, entry);
+ if (entry->is_sub_map) {
+ vm_map_deallocate(VME_SUBMAP(entry));
+ } else {
+ vm_object_deallocate(VME_OBJECT(entry));
+ }
+ vm_map_copy_entry_dispose(copy, entry);
+ }
+ break;
+ case VM_MAP_COPY_OBJECT:
+ vm_object_deallocate(copy->cpy_object);
+ break;
+ case VM_MAP_COPY_KERNEL_BUFFER:
- /*
- * Find the beginning of the region.
- */
+ /*
+ * The vm_map_copy_t and possibly the data buffer were
+ * allocated by a single call to kalloc(), i.e. the
+ * vm_map_copy_t was not allocated out of the zone.
+ */
+ if (copy->size > msg_ool_size_small || copy->offset)
+ panic("Invalid vm_map_copy_t sz:%lld, ofst:%lld",
+ (long long)copy->size, (long long)copy->offset);
+ kfree(copy, copy->size + cpy_kdata_hdr_sz);
+ return;
+ }
+ zfree(vm_map_copy_zone, copy);
+}
- vm_map_lock(src_map);
+/*
+ * Routine: vm_map_copy_copy
+ *
+ * Description:
+ * Move the information in a map copy object to
+ * a new map copy object, leaving the old one
+ * empty.
+ *
+ * This is used by kernel routines that need
+ * to look at out-of-line data (in copyin form)
+ * before deciding whether to return SUCCESS.
+ * If the routine returns FAILURE, the original
+ * copy object will be deallocated; therefore,
+ * these routines must make a copy of the copy
+ * object and leave the original empty so that
+ * deallocation will not fail.
+ */
+vm_map_copy_t
+vm_map_copy_copy(
+ vm_map_copy_t copy)
+{
+ vm_map_copy_t new_copy;
- if (!vm_map_lookup_entry(src_map, src_start, &tmp_entry))
- RETURN(KERN_INVALID_ADDRESS);
- if(!tmp_entry->is_sub_map) {
- vm_map_clip_start(src_map, tmp_entry, src_start);
- }
- /* set for later submap fix-up */
- copy_addr = src_start;
+ if (copy == VM_MAP_COPY_NULL)
+ return VM_MAP_COPY_NULL;
/*
- * Go through entries until we get to the end.
+ * Allocate a new copy object, and copy the information
+ * from the old one into it.
*/
- while (TRUE) {
- register
- vm_map_entry_t src_entry = tmp_entry; /* Top-level entry */
- vm_size_t src_size; /* Size of source
- * map entry (in both
- * maps)
- */
-
- register
- vm_object_t src_object; /* Object to copy */
- vm_object_offset_t src_offset;
-
- boolean_t src_needs_copy; /* Should source map
- * be made read-only
- * for copy-on-write?
- */
-
- boolean_t new_entry_needs_copy; /* Will new entry be COW? */
-
- boolean_t was_wired; /* Was source wired? */
- vm_map_version_t version; /* Version before locks
- * dropped to make copy
- */
- kern_return_t result; /* Return value from
- * copy_strategically.
- */
- while(tmp_entry->is_sub_map) {
- vm_size_t submap_len;
- submap_map_t *ptr;
+ new_copy = (vm_map_copy_t) zalloc(vm_map_copy_zone);
+ new_copy->c_u.hdr.rb_head_store.rbh_root = (void*)(int)SKIP_RB_TREE;
+ *new_copy = *copy;
- ptr = (submap_map_t *)kalloc(sizeof(submap_map_t));
- ptr->next = parent_maps;
- parent_maps = ptr;
- ptr->parent_map = src_map;
- ptr->base_start = src_start;
- ptr->base_end = src_end;
- submap_len = tmp_entry->vme_end - src_start;
- if(submap_len > (src_end-src_start))
- submap_len = src_end-src_start;
- ptr->base_start += submap_len;
-
- src_start -= tmp_entry->vme_start;
- src_start += tmp_entry->offset;
- src_end = src_start + submap_len;
- src_map = tmp_entry->object.sub_map;
- vm_map_lock(src_map);
- /* keep an outstanding reference for all maps in */
- /* the parents tree except the base map */
- vm_map_reference(src_map);
- vm_map_unlock(ptr->parent_map);
- if (!vm_map_lookup_entry(
- src_map, src_start, &tmp_entry))
- RETURN(KERN_INVALID_ADDRESS);
- map_share = TRUE;
- if(!tmp_entry->is_sub_map)
- vm_map_clip_start(src_map, tmp_entry, src_start);
- src_entry = tmp_entry;
- }
- if ((tmp_entry->object.vm_object != VM_OBJECT_NULL) &&
- (tmp_entry->object.vm_object->phys_contiguous)) {
- /* This is not, supported for now.In future */
- /* we will need to detect the phys_contig */
- /* condition and then upgrade copy_slowly */
- /* to do physical copy from the device mem */
- /* based object. We can piggy-back off of */
- /* the was wired boolean to set-up the */
- /* proper handling */
- RETURN(KERN_PROTECTION_FAILURE);
- }
+ if (copy->type == VM_MAP_COPY_ENTRY_LIST) {
/*
- * Create a new address map entry to hold the result.
- * Fill in the fields from the appropriate source entries.
- * We must unlock the source map to do this if we need
- * to allocate a map entry.
+ * The links in the entry chain must be
+ * changed to point to the new copy object.
*/
- if (new_entry == VM_MAP_ENTRY_NULL) {
- version.main_timestamp = src_map->timestamp;
- vm_map_unlock(src_map);
+ vm_map_copy_first_entry(copy)->vme_prev
+ = vm_map_copy_to_entry(new_copy);
+ vm_map_copy_last_entry(copy)->vme_next
+ = vm_map_copy_to_entry(new_copy);
+ }
- new_entry = vm_map_copy_entry_create(copy);
+ /*
+ * Change the old copy object into one that contains
+ * nothing to be deallocated.
+ */
+ copy->type = VM_MAP_COPY_OBJECT;
+ copy->cpy_object = VM_OBJECT_NULL;
- vm_map_lock(src_map);
- if ((version.main_timestamp + 1) != src_map->timestamp) {
- if (!vm_map_lookup_entry(src_map, src_start,
- &tmp_entry)) {
- RETURN(KERN_INVALID_ADDRESS);
- }
- vm_map_clip_start(src_map, tmp_entry, src_start);
- continue; /* restart w/ new tmp_entry */
- }
- }
+ /*
+ * Return the new object.
+ */
+ return new_copy;
+}
- /*
- * Verify that the region can be read.
- */
- if (((src_entry->protection & VM_PROT_READ) == VM_PROT_NONE &&
- !use_maxprot) ||
- (src_entry->max_protection & VM_PROT_READ) == 0)
- RETURN(KERN_PROTECTION_FAILURE);
+static kern_return_t
+vm_map_overwrite_submap_recurse(
+ vm_map_t dst_map,
+ vm_map_offset_t dst_addr,
+ vm_map_size_t dst_size)
+{
+ vm_map_offset_t dst_end;
+ vm_map_entry_t tmp_entry;
+ vm_map_entry_t entry;
+ kern_return_t result;
+ boolean_t encountered_sub_map = FALSE;
- /*
- * Clip against the endpoints of the entire region.
- */
- vm_map_clip_end(src_map, src_entry, src_end);
- src_size = src_entry->vme_end - src_start;
- src_object = src_entry->object.vm_object;
- src_offset = src_entry->offset;
- was_wired = (src_entry->wired_count != 0);
+ /*
+ * Verify that the destination is all writeable
+ * initially. We have to trunc the destination
+ * address and round the copy size or we'll end up
+ * splitting entries in strange ways.
+ */
- vm_map_entry_copy(new_entry, src_entry);
- new_entry->use_pmap = FALSE; /* clr address space specifics */
+ dst_end = vm_map_round_page(dst_addr + dst_size,
+ VM_MAP_PAGE_MASK(dst_map));
+ vm_map_lock(dst_map);
- /*
- * Attempt non-blocking copy-on-write optimizations.
- */
+start_pass_1:
+ if (!vm_map_lookup_entry(dst_map, dst_addr, &tmp_entry)) {
+ vm_map_unlock(dst_map);
+ return(KERN_INVALID_ADDRESS);
+ }
- if (src_destroy &&
- (src_object == VM_OBJECT_NULL ||
- (src_object->internal && !src_object->true_share
- && !map_share))) {
- /*
- * If we are destroying the source, and the object
- * is internal, we can move the object reference
- * from the source to the copy. The copy is
- * copy-on-write only if the source is.
- * We make another reference to the object, because
- * destroying the source entry will deallocate it.
- */
- vm_object_reference(src_object);
+ vm_map_clip_start(dst_map,
+ tmp_entry,
+ vm_map_trunc_page(dst_addr,
+ VM_MAP_PAGE_MASK(dst_map)));
+ if (tmp_entry->is_sub_map) {
+ /* clipping did unnest if needed */
+ assert(!tmp_entry->use_pmap);
+ }
- /*
- * Copy is always unwired. vm_map_copy_entry
- * set its wired count to zero.
- */
+ for (entry = tmp_entry;;) {
+ vm_map_entry_t next;
- goto CopySuccessful;
- }
+ next = entry->vme_next;
+ while(entry->is_sub_map) {
+ vm_map_offset_t sub_start;
+ vm_map_offset_t sub_end;
+ vm_map_offset_t local_end;
+ if (entry->in_transition) {
+ /*
+ * Say that we are waiting, and wait for entry.
+ */
+ entry->needs_wakeup = TRUE;
+ vm_map_entry_wait(dst_map, THREAD_UNINT);
-RestartCopy:
- XPR(XPR_VM_MAP, "vm_map_copyin_common src_obj 0x%x ent 0x%x obj 0x%x was_wired %d\n",
- src_object, new_entry, new_entry->object.vm_object,
- was_wired, 0);
- if ((src_object == VM_OBJECT_NULL ||
- (!was_wired && !map_share && !tmp_entry->is_shared)) &&
- vm_object_copy_quickly(
- &new_entry->object.vm_object,
- src_offset,
- src_size,
- &src_needs_copy,
- &new_entry_needs_copy)) {
+ goto start_pass_1;
+ }
- new_entry->needs_copy = new_entry_needs_copy;
+ encountered_sub_map = TRUE;
+ sub_start = VME_OFFSET(entry);
- /*
- * Handle copy-on-write obligations
- */
+ if(entry->vme_end < dst_end)
+ sub_end = entry->vme_end;
+ else
+ sub_end = dst_end;
+ sub_end -= entry->vme_start;
+ sub_end += VME_OFFSET(entry);
+ local_end = entry->vme_end;
+ vm_map_unlock(dst_map);
+
+ result = vm_map_overwrite_submap_recurse(
+ VME_SUBMAP(entry),
+ sub_start,
+ sub_end - sub_start);
- if (src_needs_copy && !tmp_entry->needs_copy) {
- vm_object_pmap_protect(
- src_object,
- src_offset,
- src_size,
- (src_entry->is_shared ?
- PMAP_NULL
- : src_map->pmap),
- src_entry->vme_start,
- src_entry->protection &
- ~VM_PROT_WRITE);
- tmp_entry->needs_copy = TRUE;
+ if(result != KERN_SUCCESS)
+ return result;
+ if (dst_end <= entry->vme_end)
+ return KERN_SUCCESS;
+ vm_map_lock(dst_map);
+ if(!vm_map_lookup_entry(dst_map, local_end,
+ &tmp_entry)) {
+ vm_map_unlock(dst_map);
+ return(KERN_INVALID_ADDRESS);
}
-
- /*
- * The map has never been unlocked, so it's safe
- * to move to the next entry rather than doing
- * another lookup.
- */
-
- goto CopySuccessful;
+ entry = tmp_entry;
+ next = entry->vme_next;
}
- /*
- * Take an object reference, so that we may
- * release the map lock(s).
- */
-
- assert(src_object != VM_OBJECT_NULL);
- vm_object_reference(src_object);
-
- /*
- * Record the timestamp for later verification.
- * Unlock the map.
- */
-
- version.main_timestamp = src_map->timestamp;
- vm_map_unlock(src_map); /* Increments timestamp once! */
+ if ( ! (entry->protection & VM_PROT_WRITE)) {
+ vm_map_unlock(dst_map);
+ return(KERN_PROTECTION_FAILURE);
+ }
/*
- * Perform the copy
+ * If the entry is in transition, we must wait
+ * for it to exit that state. Anything could happen
+ * when we unlock the map, so start over.
*/
+ if (entry->in_transition) {
- if (was_wired) {
- CopySlowly:
- vm_object_lock(src_object);
- result = vm_object_copy_slowly(
- src_object,
- src_offset,
- src_size,
- THREAD_UNINT,
- &new_entry->object.vm_object);
- new_entry->offset = 0;
- new_entry->needs_copy = FALSE;
+ /*
+ * Say that we are waiting, and wait for entry.
+ */
+ entry->needs_wakeup = TRUE;
+ vm_map_entry_wait(dst_map, THREAD_UNINT);
+ goto start_pass_1;
}
- else if (src_object->copy_strategy == MEMORY_OBJECT_COPY_SYMMETRIC &&
- (tmp_entry->is_shared || map_share)) {
- vm_object_t new_object;
-
- vm_object_lock(src_object);
- new_object = vm_object_copy_delayed(
- src_object,
- src_offset,
- src_size);
- if (new_object == VM_OBJECT_NULL)
- goto CopySlowly;
-
- new_entry->object.vm_object = new_object;
- new_entry->needs_copy = TRUE;
- result = KERN_SUCCESS;
-
- } else {
- result = vm_object_copy_strategically(src_object,
- src_offset,
- src_size,
- &new_entry->object.vm_object,
- &new_entry->offset,
- &new_entry_needs_copy);
- new_entry->needs_copy = new_entry_needs_copy;
+/*
+ * our range is contained completely within this map entry
+ */
+ if (dst_end <= entry->vme_end) {
+ vm_map_unlock(dst_map);
+ return KERN_SUCCESS;
}
-
- if (result != KERN_SUCCESS &&
- result != KERN_MEMORY_RESTART_COPY) {
- vm_map_lock(src_map);
- RETURN(result);
+/*
+ * check that range specified is contiguous region
+ */
+ if ((next == vm_map_to_entry(dst_map)) ||
+ (next->vme_start != entry->vme_end)) {
+ vm_map_unlock(dst_map);
+ return(KERN_INVALID_ADDRESS);
}
/*
- * Throw away the extra reference
+ * Check for permanent objects in the destination.
*/
+ if ((VME_OBJECT(entry) != VM_OBJECT_NULL) &&
+ ((!VME_OBJECT(entry)->internal) ||
+ (VME_OBJECT(entry)->true_share))) {
+ if(encountered_sub_map) {
+ vm_map_unlock(dst_map);
+ return(KERN_FAILURE);
+ }
+ }
- vm_object_deallocate(src_object);
- /*
- * Verify that the map has not substantially
- * changed while the copy was being made.
- */
+ entry = next;
+ }/* for */
+ vm_map_unlock(dst_map);
+ return(KERN_SUCCESS);
+}
- vm_map_lock(src_map);
+/*
+ * Routine: vm_map_copy_overwrite
+ *
+ * Description:
+ * Copy the memory described by the map copy
+ * object (copy; returned by vm_map_copyin) onto
+ * the specified destination region (dst_map, dst_addr).
+ * The destination must be writeable.
+ *
+ * Unlike vm_map_copyout, this routine actually
+ * writes over previously-mapped memory. If the
+ * previous mapping was to a permanent (user-supplied)
+ * memory object, it is preserved.
+ *
+ * The attributes (protection and inheritance) of the
+ * destination region are preserved.
+ *
+ * If successful, consumes the copy object.
+ * Otherwise, the caller is responsible for it.
+ *
+ * Implementation notes:
+ * To overwrite aligned temporary virtual memory, it is
+ * sufficient to remove the previous mapping and insert
+ * the new copy. This replacement is done either on
+ * the whole region (if no permanent virtual memory
+ * objects are embedded in the destination region) or
+ * in individual map entries.
+ *
+ * To overwrite permanent virtual memory , it is necessary
+ * to copy each page, as the external memory management
+ * interface currently does not provide any optimizations.
+ *
+ * Unaligned memory also has to be copied. It is possible
+ * to use 'vm_trickery' to copy the aligned data. This is
+ * not done but not hard to implement.
+ *
+ * Once a page of permanent memory has been overwritten,
+ * it is impossible to interrupt this function; otherwise,
+ * the call would be neither atomic nor location-independent.
+ * The kernel-state portion of a user thread must be
+ * interruptible.
+ *
+ * It may be expensive to forward all requests that might
+ * overwrite permanent memory (vm_write, vm_copy) to
+ * uninterruptible kernel threads. This routine may be
+ * called by interruptible threads; however, success is
+ * not guaranteed -- if the request cannot be performed
+ * atomically and interruptibly, an error indication is
+ * returned.
+ */
- if ((version.main_timestamp + 1) == src_map->timestamp)
- goto VerificationSuccessful;
+static kern_return_t
+vm_map_copy_overwrite_nested(
+ vm_map_t dst_map,
+ vm_map_address_t dst_addr,
+ vm_map_copy_t copy,
+ boolean_t interruptible,
+ pmap_t pmap,
+ boolean_t discard_on_success)
+{
+ vm_map_offset_t dst_end;
+ vm_map_entry_t tmp_entry;
+ vm_map_entry_t entry;
+ kern_return_t kr;
+ boolean_t aligned = TRUE;
+ boolean_t contains_permanent_objects = FALSE;
+ boolean_t encountered_sub_map = FALSE;
+ vm_map_offset_t base_addr;
+ vm_map_size_t copy_size;
+ vm_map_size_t total_size;
- /*
- * Simple version comparison failed.
- *
- * Retry the lookup and verify that the
- * same object/offset are still present.
- *
- * [Note: a memory manager that colludes with
- * the calling task can detect that we have
- * cheated. While the map was unlocked, the
- * mapping could have been changed and restored.]
- */
- if (!vm_map_lookup_entry(src_map, src_start, &tmp_entry)) {
- RETURN(KERN_INVALID_ADDRESS);
- }
+ /*
+ * Check for null copy object.
+ */
- src_entry = tmp_entry;
- vm_map_clip_start(src_map, src_entry, src_start);
+ if (copy == VM_MAP_COPY_NULL)
+ return(KERN_SUCCESS);
- if ((src_entry->protection & VM_PROT_READ == VM_PROT_NONE &&
- !use_maxprot) ||
- src_entry->max_protection & VM_PROT_READ == 0)
- goto VerificationFailed;
+ /*
+ * Check for special kernel buffer allocated
+ * by new_ipc_kmsg_copyin.
+ */
- if (src_entry->vme_end < new_entry->vme_end)
- src_size = (new_entry->vme_end = src_entry->vme_end) - src_start;
+ if (copy->type == VM_MAP_COPY_KERNEL_BUFFER) {
+ return(vm_map_copyout_kernel_buffer(
+ dst_map, &dst_addr,
+ copy, TRUE, discard_on_success));
+ }
- if ((src_entry->object.vm_object != src_object) ||
- (src_entry->offset != src_offset) ) {
+ /*
+ * Only works for entry lists at the moment. Will
+ * support page lists later.
+ */
- /*
- * Verification failed.
- *
- * Start over with this top-level entry.
- */
+ assert(copy->type == VM_MAP_COPY_ENTRY_LIST);
- VerificationFailed: ;
+ if (copy->size == 0) {
+ if (discard_on_success)
+ vm_map_copy_discard(copy);
+ return(KERN_SUCCESS);
+ }
- vm_object_deallocate(new_entry->object.vm_object);
- tmp_entry = src_entry;
- continue;
- }
+ /*
+ * Verify that the destination is all writeable
+ * initially. We have to trunc the destination
+ * address and round the copy size or we'll end up
+ * splitting entries in strange ways.
+ */
- /*
- * Verification succeeded.
- */
+ if (!VM_MAP_PAGE_ALIGNED(copy->size,
+ VM_MAP_PAGE_MASK(dst_map)) ||
+ !VM_MAP_PAGE_ALIGNED(copy->offset,
+ VM_MAP_PAGE_MASK(dst_map)) ||
+ !VM_MAP_PAGE_ALIGNED(dst_addr,
+ VM_MAP_PAGE_MASK(dst_map)))
+ {
+ aligned = FALSE;
+ dst_end = vm_map_round_page(dst_addr + copy->size,
+ VM_MAP_PAGE_MASK(dst_map));
+ } else {
+ dst_end = dst_addr + copy->size;
+ }
- VerificationSuccessful: ;
+ vm_map_lock(dst_map);
- if (result == KERN_MEMORY_RESTART_COPY)
- goto RestartCopy;
+ /* LP64todo - remove this check when vm_map_commpage64()
+ * no longer has to stuff in a map_entry for the commpage
+ * above the map's max_offset.
+ */
+ if (dst_addr >= dst_map->max_offset) {
+ vm_map_unlock(dst_map);
+ return(KERN_INVALID_ADDRESS);
+ }
+
+start_pass_1:
+ if (!vm_map_lookup_entry(dst_map, dst_addr, &tmp_entry)) {
+ vm_map_unlock(dst_map);
+ return(KERN_INVALID_ADDRESS);
+ }
+ vm_map_clip_start(dst_map,
+ tmp_entry,
+ vm_map_trunc_page(dst_addr,
+ VM_MAP_PAGE_MASK(dst_map)));
+ for (entry = tmp_entry;;) {
+ vm_map_entry_t next = entry->vme_next;
- /*
- * Copy succeeded.
- */
+ while(entry->is_sub_map) {
+ vm_map_offset_t sub_start;
+ vm_map_offset_t sub_end;
+ vm_map_offset_t local_end;
- CopySuccessful: ;
+ if (entry->in_transition) {
- /*
- * Link in the new copy entry.
- */
+ /*
+ * Say that we are waiting, and wait for entry.
+ */
+ entry->needs_wakeup = TRUE;
+ vm_map_entry_wait(dst_map, THREAD_UNINT);
+
+ goto start_pass_1;
+ }
+
+ local_end = entry->vme_end;
+ if (!(entry->needs_copy)) {
+ /* if needs_copy we are a COW submap */
+ /* in such a case we just replace so */
+ /* there is no need for the follow- */
+ /* ing check. */
+ encountered_sub_map = TRUE;
+ sub_start = VME_OFFSET(entry);
+
+ if(entry->vme_end < dst_end)
+ sub_end = entry->vme_end;
+ else
+ sub_end = dst_end;
+ sub_end -= entry->vme_start;
+ sub_end += VME_OFFSET(entry);
+ vm_map_unlock(dst_map);
+
+ kr = vm_map_overwrite_submap_recurse(
+ VME_SUBMAP(entry),
+ sub_start,
+ sub_end - sub_start);
+ if(kr != KERN_SUCCESS)
+ return kr;
+ vm_map_lock(dst_map);
+ }
+
+ if (dst_end <= entry->vme_end)
+ goto start_overwrite;
+ if(!vm_map_lookup_entry(dst_map, local_end,
+ &entry)) {
+ vm_map_unlock(dst_map);
+ return(KERN_INVALID_ADDRESS);
+ }
+ next = entry->vme_next;
+ }
+
+ if ( ! (entry->protection & VM_PROT_WRITE)) {
+ vm_map_unlock(dst_map);
+ return(KERN_PROTECTION_FAILURE);
+ }
- vm_map_copy_entry_link(copy, vm_map_copy_last_entry(copy),
- new_entry);
-
/*
- * Determine whether the entire region
- * has been copied.
+ * If the entry is in transition, we must wait
+ * for it to exit that state. Anything could happen
+ * when we unlock the map, so start over.
*/
- src_start = new_entry->vme_end;
- new_entry = VM_MAP_ENTRY_NULL;
- while ((src_start >= src_end) && (src_end != 0)) {
- if (src_map != base_map) {
- submap_map_t *ptr;
-
- ptr = parent_maps;
- assert(ptr != NULL);
- parent_maps = parent_maps->next;
- vm_map_unlock(src_map);
- vm_map_deallocate(src_map);
- vm_map_lock(ptr->parent_map);
- src_map = ptr->parent_map;
- src_start = ptr->base_start;
- src_end = ptr->base_end;
- if ((src_end > src_start) &&
- !vm_map_lookup_entry(
- src_map, src_start, &tmp_entry))
- RETURN(KERN_INVALID_ADDRESS);
- kfree((vm_offset_t)ptr, sizeof(submap_map_t));
- if(parent_maps == NULL)
- map_share = FALSE;
- src_entry = tmp_entry->vme_prev;
- } else
- break;
+ if (entry->in_transition) {
+
+ /*
+ * Say that we are waiting, and wait for entry.
+ */
+ entry->needs_wakeup = TRUE;
+ vm_map_entry_wait(dst_map, THREAD_UNINT);
+
+ goto start_pass_1;
}
- if ((src_start >= src_end) && (src_end != 0))
+
+/*
+ * our range is contained completely within this map entry
+ */
+ if (dst_end <= entry->vme_end)
break;
+/*
+ * check that range specified is contiguous region
+ */
+ if ((next == vm_map_to_entry(dst_map)) ||
+ (next->vme_start != entry->vme_end)) {
+ vm_map_unlock(dst_map);
+ return(KERN_INVALID_ADDRESS);
+ }
+
/*
- * Verify that there are no gaps in the region
+ * Check for permanent objects in the destination.
*/
+ if ((VME_OBJECT(entry) != VM_OBJECT_NULL) &&
+ ((!VME_OBJECT(entry)->internal) ||
+ (VME_OBJECT(entry)->true_share))) {
+ contains_permanent_objects = TRUE;
+ }
- tmp_entry = src_entry->vme_next;
- if ((tmp_entry->vme_start != src_start) ||
- (tmp_entry == vm_map_to_entry(src_map)))
- RETURN(KERN_INVALID_ADDRESS);
- }
+ entry = next;
+ }/* for */
+start_overwrite:
/*
- * If the source should be destroyed, do it now, since the
- * copy was successful.
+ * If there are permanent objects in the destination, then
+ * the copy cannot be interrupted.
*/
- if (src_destroy) {
- (void) vm_map_delete(src_map,
- trunc_page_32(src_addr),
- src_end,
- (src_map == kernel_map) ?
- VM_MAP_REMOVE_KUNWIRE :
- VM_MAP_NO_FLAGS);
- }
-
- vm_map_unlock(src_map);
-
- /* Fix-up start and end points in copy. This is necessary */
- /* when the various entries in the copy object were picked */
- /* up from different sub-maps */
- tmp_entry = vm_map_copy_first_entry(copy);
- while (tmp_entry != vm_map_copy_to_entry(copy)) {
- tmp_entry->vme_end = copy_addr +
- (tmp_entry->vme_end - tmp_entry->vme_start);
- tmp_entry->vme_start = copy_addr;
- copy_addr += tmp_entry->vme_end - tmp_entry->vme_start;
- tmp_entry = (struct vm_map_entry *)tmp_entry->vme_next;
+ if (interruptible && contains_permanent_objects) {
+ vm_map_unlock(dst_map);
+ return(KERN_FAILURE); /* XXX */
}
- *copy_result = copy;
- return(KERN_SUCCESS);
-
-#undef RETURN
-}
+ /*
+ *
+ * Make a second pass, overwriting the data
+ * At the beginning of each loop iteration,
+ * the next entry to be overwritten is "tmp_entry"
+ * (initially, the value returned from the lookup above),
+ * and the starting address expected in that entry
+ * is "start".
+ */
-/*
- * vm_map_copyin_object:
- *
- * Create a copy object from an object.
- * Our caller donates an object reference.
- */
+ total_size = copy->size;
+ if(encountered_sub_map) {
+ copy_size = 0;
+ /* re-calculate tmp_entry since we've had the map */
+ /* unlocked */
+ if (!vm_map_lookup_entry( dst_map, dst_addr, &tmp_entry)) {
+ vm_map_unlock(dst_map);
+ return(KERN_INVALID_ADDRESS);
+ }
+ } else {
+ copy_size = copy->size;
+ }
+
+ base_addr = dst_addr;
+ while(TRUE) {
+ /* deconstruct the copy object and do in parts */
+ /* only in sub_map, interruptable case */
+ vm_map_entry_t copy_entry;
+ vm_map_entry_t previous_prev = VM_MAP_ENTRY_NULL;
+ vm_map_entry_t next_copy = VM_MAP_ENTRY_NULL;
+ int nentries;
+ int remaining_entries = 0;
+ vm_map_offset_t new_offset = 0;
+
+ for (entry = tmp_entry; copy_size == 0;) {
+ vm_map_entry_t next;
-kern_return_t
-vm_map_copyin_object(
- vm_object_t object,
- vm_object_offset_t offset, /* offset of region in object */
- vm_object_size_t size, /* size of region in object */
- vm_map_copy_t *copy_result) /* OUT */
-{
- vm_map_copy_t copy; /* Resulting copy */
+ next = entry->vme_next;
- /*
- * We drop the object into a special copy object
- * that contains the object directly.
- */
+ /* tmp_entry and base address are moved along */
+ /* each time we encounter a sub-map. Otherwise */
+ /* entry can outpase tmp_entry, and the copy_size */
+ /* may reflect the distance between them */
+ /* if the current entry is found to be in transition */
+ /* we will start over at the beginning or the last */
+ /* encounter of a submap as dictated by base_addr */
+ /* we will zero copy_size accordingly. */
+ if (entry->in_transition) {
+ /*
+ * Say that we are waiting, and wait for entry.
+ */
+ entry->needs_wakeup = TRUE;
+ vm_map_entry_wait(dst_map, THREAD_UNINT);
- copy = (vm_map_copy_t) zalloc(vm_map_copy_zone);
- copy->type = VM_MAP_COPY_OBJECT;
- copy->cpy_object = object;
- copy->cpy_index = 0;
- copy->offset = offset;
- copy->size = size;
+ if(!vm_map_lookup_entry(dst_map, base_addr,
+ &tmp_entry)) {
+ vm_map_unlock(dst_map);
+ return(KERN_INVALID_ADDRESS);
+ }
+ copy_size = 0;
+ entry = tmp_entry;
+ continue;
+ }
+ if(entry->is_sub_map) {
+ vm_map_offset_t sub_start;
+ vm_map_offset_t sub_end;
+ vm_map_offset_t local_end;
- *copy_result = copy;
- return(KERN_SUCCESS);
-}
+ if (entry->needs_copy) {
+ /* if this is a COW submap */
+ /* just back the range with a */
+ /* anonymous entry */
+ if(entry->vme_end < dst_end)
+ sub_end = entry->vme_end;
+ else
+ sub_end = dst_end;
+ if(entry->vme_start < base_addr)
+ sub_start = base_addr;
+ else
+ sub_start = entry->vme_start;
+ vm_map_clip_end(
+ dst_map, entry, sub_end);
+ vm_map_clip_start(
+ dst_map, entry, sub_start);
+ assert(!entry->use_pmap);
+ entry->is_sub_map = FALSE;
+ vm_map_deallocate(
+ VME_SUBMAP(entry));
+ VME_SUBMAP_SET(entry, NULL);
+ entry->is_shared = FALSE;
+ entry->needs_copy = FALSE;
+ VME_OFFSET_SET(entry, 0);
+ /*
+ * XXX FBDP
+ * We should propagate the protections
+ * of the submap entry here instead
+ * of forcing them to VM_PROT_ALL...
+ * Or better yet, we should inherit
+ * the protection of the copy_entry.
+ */
+ entry->protection = VM_PROT_ALL;
+ entry->max_protection = VM_PROT_ALL;
+ entry->wired_count = 0;
+ entry->user_wired_count = 0;
+ if(entry->inheritance
+ == VM_INHERIT_SHARE)
+ entry->inheritance = VM_INHERIT_COPY;
+ continue;
+ }
+ /* first take care of any non-sub_map */
+ /* entries to send */
+ if(base_addr < entry->vme_start) {
+ /* stuff to send */
+ copy_size =
+ entry->vme_start - base_addr;
+ break;
+ }
+ sub_start = VME_OFFSET(entry);
-void
-vm_map_fork_share(
- vm_map_t old_map,
- vm_map_entry_t old_entry,
- vm_map_t new_map)
-{
- vm_object_t object;
- vm_map_entry_t new_entry;
- kern_return_t result;
+ if(entry->vme_end < dst_end)
+ sub_end = entry->vme_end;
+ else
+ sub_end = dst_end;
+ sub_end -= entry->vme_start;
+ sub_end += VME_OFFSET(entry);
+ local_end = entry->vme_end;
+ vm_map_unlock(dst_map);
+ copy_size = sub_end - sub_start;
- /*
- * New sharing code. New map entry
- * references original object. Internal
- * objects use asynchronous copy algorithm for
- * future copies. First make sure we have
- * the right object. If we need a shadow,
- * or someone else already has one, then
- * make a new shadow and share it.
+ /* adjust the copy object */
+ if (total_size > copy_size) {
+ vm_map_size_t local_size = 0;
+ vm_map_size_t entry_size;
+
+ nentries = 1;
+ new_offset = copy->offset;
+ copy_entry = vm_map_copy_first_entry(copy);
+ while(copy_entry !=
+ vm_map_copy_to_entry(copy)){
+ entry_size = copy_entry->vme_end -
+ copy_entry->vme_start;
+ if((local_size < copy_size) &&
+ ((local_size + entry_size)
+ >= copy_size)) {
+ vm_map_copy_clip_end(copy,
+ copy_entry,
+ copy_entry->vme_start +
+ (copy_size - local_size));
+ entry_size = copy_entry->vme_end -
+ copy_entry->vme_start;
+ local_size += entry_size;
+ new_offset += entry_size;
+ }
+ if(local_size >= copy_size) {
+ next_copy = copy_entry->vme_next;
+ copy_entry->vme_next =
+ vm_map_copy_to_entry(copy);
+ previous_prev =
+ copy->cpy_hdr.links.prev;
+ copy->cpy_hdr.links.prev = copy_entry;
+ copy->size = copy_size;
+ remaining_entries =
+ copy->cpy_hdr.nentries;
+ remaining_entries -= nentries;
+ copy->cpy_hdr.nentries = nentries;
+ break;
+ } else {
+ local_size += entry_size;
+ new_offset += entry_size;
+ nentries++;
+ }
+ copy_entry = copy_entry->vme_next;
+ }
+ }
+
+ if((entry->use_pmap) && (pmap == NULL)) {
+ kr = vm_map_copy_overwrite_nested(
+ VME_SUBMAP(entry),
+ sub_start,
+ copy,
+ interruptible,
+ VME_SUBMAP(entry)->pmap,
+ TRUE);
+ } else if (pmap != NULL) {
+ kr = vm_map_copy_overwrite_nested(
+ VME_SUBMAP(entry),
+ sub_start,
+ copy,
+ interruptible, pmap,
+ TRUE);
+ } else {
+ kr = vm_map_copy_overwrite_nested(
+ VME_SUBMAP(entry),
+ sub_start,
+ copy,
+ interruptible,
+ dst_map->pmap,
+ TRUE);
+ }
+ if(kr != KERN_SUCCESS) {
+ if(next_copy != NULL) {
+ copy->cpy_hdr.nentries +=
+ remaining_entries;
+ copy->cpy_hdr.links.prev->vme_next =
+ next_copy;
+ copy->cpy_hdr.links.prev
+ = previous_prev;
+ copy->size = total_size;
+ }
+ return kr;
+ }
+ if (dst_end <= local_end) {
+ return(KERN_SUCCESS);
+ }
+ /* otherwise copy no longer exists, it was */
+ /* destroyed after successful copy_overwrite */
+ copy = (vm_map_copy_t)
+ zalloc(vm_map_copy_zone);
+ copy->c_u.hdr.rb_head_store.rbh_root = (void*)(int)SKIP_RB_TREE;
+ vm_map_copy_first_entry(copy) =
+ vm_map_copy_last_entry(copy) =
+ vm_map_copy_to_entry(copy);
+ copy->type = VM_MAP_COPY_ENTRY_LIST;
+ copy->offset = new_offset;
+
+ /*
+ * XXX FBDP
+ * this does not seem to deal with
+ * the VM map store (R&B tree)
+ */
+
+ total_size -= copy_size;
+ copy_size = 0;
+ /* put back remainder of copy in container */
+ if(next_copy != NULL) {
+ copy->cpy_hdr.nentries = remaining_entries;
+ copy->cpy_hdr.links.next = next_copy;
+ copy->cpy_hdr.links.prev = previous_prev;
+ copy->size = total_size;
+ next_copy->vme_prev =
+ vm_map_copy_to_entry(copy);
+ next_copy = NULL;
+ }
+ base_addr = local_end;
+ vm_map_lock(dst_map);
+ if(!vm_map_lookup_entry(dst_map,
+ local_end, &tmp_entry)) {
+ vm_map_unlock(dst_map);
+ return(KERN_INVALID_ADDRESS);
+ }
+ entry = tmp_entry;
+ continue;
+ }
+ if (dst_end <= entry->vme_end) {
+ copy_size = dst_end - base_addr;
+ break;
+ }
+
+ if ((next == vm_map_to_entry(dst_map)) ||
+ (next->vme_start != entry->vme_end)) {
+ vm_map_unlock(dst_map);
+ return(KERN_INVALID_ADDRESS);
+ }
+
+ entry = next;
+ }/* for */
+
+ next_copy = NULL;
+ nentries = 1;
+
+ /* adjust the copy object */
+ if (total_size > copy_size) {
+ vm_map_size_t local_size = 0;
+ vm_map_size_t entry_size;
+
+ new_offset = copy->offset;
+ copy_entry = vm_map_copy_first_entry(copy);
+ while(copy_entry != vm_map_copy_to_entry(copy)) {
+ entry_size = copy_entry->vme_end -
+ copy_entry->vme_start;
+ if((local_size < copy_size) &&
+ ((local_size + entry_size)
+ >= copy_size)) {
+ vm_map_copy_clip_end(copy, copy_entry,
+ copy_entry->vme_start +
+ (copy_size - local_size));
+ entry_size = copy_entry->vme_end -
+ copy_entry->vme_start;
+ local_size += entry_size;
+ new_offset += entry_size;
+ }
+ if(local_size >= copy_size) {
+ next_copy = copy_entry->vme_next;
+ copy_entry->vme_next =
+ vm_map_copy_to_entry(copy);
+ previous_prev =
+ copy->cpy_hdr.links.prev;
+ copy->cpy_hdr.links.prev = copy_entry;
+ copy->size = copy_size;
+ remaining_entries =
+ copy->cpy_hdr.nentries;
+ remaining_entries -= nentries;
+ copy->cpy_hdr.nentries = nentries;
+ break;
+ } else {
+ local_size += entry_size;
+ new_offset += entry_size;
+ nentries++;
+ }
+ copy_entry = copy_entry->vme_next;
+ }
+ }
+
+ if (aligned) {
+ pmap_t local_pmap;
+
+ if(pmap)
+ local_pmap = pmap;
+ else
+ local_pmap = dst_map->pmap;
+
+ if ((kr = vm_map_copy_overwrite_aligned(
+ dst_map, tmp_entry, copy,
+ base_addr, local_pmap)) != KERN_SUCCESS) {
+ if(next_copy != NULL) {
+ copy->cpy_hdr.nentries +=
+ remaining_entries;
+ copy->cpy_hdr.links.prev->vme_next =
+ next_copy;
+ copy->cpy_hdr.links.prev =
+ previous_prev;
+ copy->size += copy_size;
+ }
+ return kr;
+ }
+ vm_map_unlock(dst_map);
+ } else {
+ /*
+ * Performance gain:
+ *
+ * if the copy and dst address are misaligned but the same
+ * offset within the page we can copy_not_aligned the
+ * misaligned parts and copy aligned the rest. If they are
+ * aligned but len is unaligned we simply need to copy
+ * the end bit unaligned. We'll need to split the misaligned
+ * bits of the region in this case !
+ */
+ /* ALWAYS UNLOCKS THE dst_map MAP */
+ kr = vm_map_copy_overwrite_unaligned(
+ dst_map,
+ tmp_entry,
+ copy,
+ base_addr,
+ discard_on_success);
+ if (kr != KERN_SUCCESS) {
+ if(next_copy != NULL) {
+ copy->cpy_hdr.nentries +=
+ remaining_entries;
+ copy->cpy_hdr.links.prev->vme_next =
+ next_copy;
+ copy->cpy_hdr.links.prev =
+ previous_prev;
+ copy->size += copy_size;
+ }
+ return kr;
+ }
+ }
+ total_size -= copy_size;
+ if(total_size == 0)
+ break;
+ base_addr += copy_size;
+ copy_size = 0;
+ copy->offset = new_offset;
+ if(next_copy != NULL) {
+ copy->cpy_hdr.nentries = remaining_entries;
+ copy->cpy_hdr.links.next = next_copy;
+ copy->cpy_hdr.links.prev = previous_prev;
+ next_copy->vme_prev = vm_map_copy_to_entry(copy);
+ copy->size = total_size;
+ }
+ vm_map_lock(dst_map);
+ while(TRUE) {
+ if (!vm_map_lookup_entry(dst_map,
+ base_addr, &tmp_entry)) {
+ vm_map_unlock(dst_map);
+ return(KERN_INVALID_ADDRESS);
+ }
+ if (tmp_entry->in_transition) {
+ entry->needs_wakeup = TRUE;
+ vm_map_entry_wait(dst_map, THREAD_UNINT);
+ } else {
+ break;
+ }
+ }
+ vm_map_clip_start(dst_map,
+ tmp_entry,
+ vm_map_trunc_page(base_addr,
+ VM_MAP_PAGE_MASK(dst_map)));
+
+ entry = tmp_entry;
+ } /* while */
+
+ /*
+ * Throw away the vm_map_copy object
*/
-
- object = old_entry->object.vm_object;
- if (old_entry->is_sub_map) {
- assert(old_entry->wired_count == 0);
-#ifndef i386
- if(old_entry->use_pmap) {
- result = pmap_nest(new_map->pmap,
- (old_entry->object.sub_map)->pmap,
- (addr64_t)old_entry->vme_start,
- (addr64_t)old_entry->vme_start,
- (uint64_t)(old_entry->vme_end - old_entry->vme_start));
- if(result)
- panic("vm_map_fork_share: pmap_nest failed!");
+ if (discard_on_success)
+ vm_map_copy_discard(copy);
+
+ return(KERN_SUCCESS);
+}/* vm_map_copy_overwrite */
+
+kern_return_t
+vm_map_copy_overwrite(
+ vm_map_t dst_map,
+ vm_map_offset_t dst_addr,
+ vm_map_copy_t copy,
+ boolean_t interruptible)
+{
+ vm_map_size_t head_size, tail_size;
+ vm_map_copy_t head_copy, tail_copy;
+ vm_map_offset_t head_addr, tail_addr;
+ vm_map_entry_t entry;
+ kern_return_t kr;
+
+ head_size = 0;
+ tail_size = 0;
+ head_copy = NULL;
+ tail_copy = NULL;
+ head_addr = 0;
+ tail_addr = 0;
+
+ if (interruptible ||
+ copy == VM_MAP_COPY_NULL ||
+ copy->type != VM_MAP_COPY_ENTRY_LIST) {
+ /*
+ * We can't split the "copy" map if we're interruptible
+ * or if we don't have a "copy" map...
+ */
+ blunt_copy:
+ return vm_map_copy_overwrite_nested(dst_map,
+ dst_addr,
+ copy,
+ interruptible,
+ (pmap_t) NULL,
+ TRUE);
+ }
+
+ if (copy->size < 3 * PAGE_SIZE) {
+ /*
+ * Too small to bother with optimizing...
+ */
+ goto blunt_copy;
+ }
+
+ if ((dst_addr & VM_MAP_PAGE_MASK(dst_map)) !=
+ (copy->offset & VM_MAP_PAGE_MASK(dst_map))) {
+ /*
+ * Incompatible mis-alignment of source and destination...
+ */
+ goto blunt_copy;
+ }
+
+ /*
+ * Proper alignment or identical mis-alignment at the beginning.
+ * Let's try and do a small unaligned copy first (if needed)
+ * and then an aligned copy for the rest.
+ */
+ if (!page_aligned(dst_addr)) {
+ head_addr = dst_addr;
+ head_size = (VM_MAP_PAGE_SIZE(dst_map) -
+ (copy->offset & VM_MAP_PAGE_MASK(dst_map)));
+ }
+ if (!page_aligned(copy->offset + copy->size)) {
+ /*
+ * Mis-alignment at the end.
+ * Do an aligned copy up to the last page and
+ * then an unaligned copy for the remaining bytes.
+ */
+ tail_size = ((copy->offset + copy->size) &
+ VM_MAP_PAGE_MASK(dst_map));
+ tail_addr = dst_addr + copy->size - tail_size;
+ }
+
+ if (head_size + tail_size == copy->size) {
+ /*
+ * It's all unaligned, no optimization possible...
+ */
+ goto blunt_copy;
+ }
+
+ /*
+ * Can't optimize if there are any submaps in the
+ * destination due to the way we free the "copy" map
+ * progressively in vm_map_copy_overwrite_nested()
+ * in that case.
+ */
+ vm_map_lock_read(dst_map);
+ if (! vm_map_lookup_entry(dst_map, dst_addr, &entry)) {
+ vm_map_unlock_read(dst_map);
+ goto blunt_copy;
+ }
+ for (;
+ (entry != vm_map_copy_to_entry(copy) &&
+ entry->vme_start < dst_addr + copy->size);
+ entry = entry->vme_next) {
+ if (entry->is_sub_map) {
+ vm_map_unlock_read(dst_map);
+ goto blunt_copy;
}
-#endif
- } else if (object == VM_OBJECT_NULL) {
- object = vm_object_allocate((vm_size_t)(old_entry->vme_end -
- old_entry->vme_start));
- old_entry->offset = 0;
- old_entry->object.vm_object = object;
- assert(!old_entry->needs_copy);
- } else if (object->copy_strategy !=
- MEMORY_OBJECT_COPY_SYMMETRIC) {
-
+ }
+ vm_map_unlock_read(dst_map);
+
+ if (head_size) {
/*
- * We are already using an asymmetric
- * copy, and therefore we already have
- * the right object.
+ * Unaligned copy of the first "head_size" bytes, to reach
+ * a page boundary.
*/
- assert(! old_entry->needs_copy);
+ /*
+ * Extract "head_copy" out of "copy".
+ */
+ head_copy = (vm_map_copy_t) zalloc(vm_map_copy_zone);
+ head_copy->c_u.hdr.rb_head_store.rbh_root = (void*)(int)SKIP_RB_TREE;
+ vm_map_copy_first_entry(head_copy) =
+ vm_map_copy_to_entry(head_copy);
+ vm_map_copy_last_entry(head_copy) =
+ vm_map_copy_to_entry(head_copy);
+ head_copy->type = VM_MAP_COPY_ENTRY_LIST;
+ head_copy->cpy_hdr.nentries = 0;
+ head_copy->cpy_hdr.entries_pageable =
+ copy->cpy_hdr.entries_pageable;
+ vm_map_store_init(&head_copy->cpy_hdr);
+
+ head_copy->offset = copy->offset;
+ head_copy->size = head_size;
+
+ copy->offset += head_size;
+ copy->size -= head_size;
+
+ entry = vm_map_copy_first_entry(copy);
+ vm_map_copy_clip_end(copy, entry, copy->offset);
+ vm_map_copy_entry_unlink(copy, entry);
+ vm_map_copy_entry_link(head_copy,
+ vm_map_copy_to_entry(head_copy),
+ entry);
+
+ /*
+ * Do the unaligned copy.
+ */
+ kr = vm_map_copy_overwrite_nested(dst_map,
+ head_addr,
+ head_copy,
+ interruptible,
+ (pmap_t) NULL,
+ FALSE);
+ if (kr != KERN_SUCCESS)
+ goto done;
}
- else if (old_entry->needs_copy || /* case 1 */
- object->shadowed || /* case 2 */
- (!object->true_share && /* case 3 */
- !old_entry->is_shared &&
- (object->size >
- (vm_size_t)(old_entry->vme_end -
- old_entry->vme_start)))) {
-
+
+ if (tail_size) {
/*
- * We need to create a shadow.
- * There are three cases here.
- * In the first case, we need to
- * complete a deferred symmetrical
- * copy that we participated in.
- * In the second and third cases,
- * we need to create the shadow so
- * that changes that we make to the
- * object do not interfere with
- * any symmetrical copies which
- * have occured (case 2) or which
- * might occur (case 3).
- *
- * The first case is when we had
- * deferred shadow object creation
- * via the entry->needs_copy mechanism.
- * This mechanism only works when
- * only one entry points to the source
- * object, and we are about to create
- * a second entry pointing to the
- * same object. The problem is that
- * there is no way of mapping from
- * an object to the entries pointing
- * to it. (Deferred shadow creation
- * works with one entry because occurs
- * at fault time, and we walk from the
- * entry to the object when handling
- * the fault.)
- *
- * The second case is when the object
- * to be shared has already been copied
- * with a symmetric copy, but we point
- * directly to the object without
- * needs_copy set in our entry. (This
- * can happen because different ranges
- * of an object can be pointed to by
- * different entries. In particular,
- * a single entry pointing to an object
- * can be split by a call to vm_inherit,
- * which, combined with task_create, can
- * result in the different entries
- * having different needs_copy values.)
- * The shadowed flag in the object allows
- * us to detect this case. The problem
- * with this case is that if this object
- * has or will have shadows, then we
- * must not perform an asymmetric copy
- * of this object, since such a copy
- * allows the object to be changed, which
- * will break the previous symmetrical
- * copies (which rely upon the object
- * not changing). In a sense, the shadowed
- * flag says "don't change this object".
- * We fix this by creating a shadow
- * object for this object, and sharing
- * that. This works because we are free
- * to change the shadow object (and thus
- * to use an asymmetric copy strategy);
- * this is also semantically correct,
- * since this object is temporary, and
- * therefore a copy of the object is
- * as good as the object itself. (This
- * is not true for permanent objects,
- * since the pager needs to see changes,
- * which won't happen if the changes
- * are made to a copy.)
- *
- * The third case is when the object
- * to be shared has parts sticking
- * outside of the entry we're working
- * with, and thus may in the future
- * be subject to a symmetrical copy.
- * (This is a preemptive version of
- * case 2.)
+ * Extract "tail_copy" out of "copy".
*/
-
- assert(!(object->shadowed && old_entry->is_shared));
- vm_object_shadow(&old_entry->object.vm_object,
- &old_entry->offset,
- (vm_size_t) (old_entry->vme_end -
- old_entry->vme_start));
-
+ tail_copy = (vm_map_copy_t) zalloc(vm_map_copy_zone);
+ tail_copy->c_u.hdr.rb_head_store.rbh_root = (void*)(int)SKIP_RB_TREE;
+ vm_map_copy_first_entry(tail_copy) =
+ vm_map_copy_to_entry(tail_copy);
+ vm_map_copy_last_entry(tail_copy) =
+ vm_map_copy_to_entry(tail_copy);
+ tail_copy->type = VM_MAP_COPY_ENTRY_LIST;
+ tail_copy->cpy_hdr.nentries = 0;
+ tail_copy->cpy_hdr.entries_pageable =
+ copy->cpy_hdr.entries_pageable;
+ vm_map_store_init(&tail_copy->cpy_hdr);
+
+ tail_copy->offset = copy->offset + copy->size - tail_size;
+ tail_copy->size = tail_size;
+
+ copy->size -= tail_size;
+
+ entry = vm_map_copy_last_entry(copy);
+ vm_map_copy_clip_start(copy, entry, tail_copy->offset);
+ entry = vm_map_copy_last_entry(copy);
+ vm_map_copy_entry_unlink(copy, entry);
+ vm_map_copy_entry_link(tail_copy,
+ vm_map_copy_last_entry(tail_copy),
+ entry);
+ }
+
+ /*
+ * Copy most (or possibly all) of the data.
+ */
+ kr = vm_map_copy_overwrite_nested(dst_map,
+ dst_addr + head_size,
+ copy,
+ interruptible,
+ (pmap_t) NULL,
+ FALSE);
+ if (kr != KERN_SUCCESS) {
+ goto done;
+ }
+
+ if (tail_size) {
+ kr = vm_map_copy_overwrite_nested(dst_map,
+ tail_addr,
+ tail_copy,
+ interruptible,
+ (pmap_t) NULL,
+ FALSE);
+ }
+
+done:
+ assert(copy->type == VM_MAP_COPY_ENTRY_LIST);
+ if (kr == KERN_SUCCESS) {
/*
- * If we're making a shadow for other than
- * copy on write reasons, then we have
- * to remove write permission.
+ * Discard all the copy maps.
+ */
+ if (head_copy) {
+ vm_map_copy_discard(head_copy);
+ head_copy = NULL;
+ }
+ vm_map_copy_discard(copy);
+ if (tail_copy) {
+ vm_map_copy_discard(tail_copy);
+ tail_copy = NULL;
+ }
+ } else {
+ /*
+ * Re-assemble the original copy map.
*/
+ if (head_copy) {
+ entry = vm_map_copy_first_entry(head_copy);
+ vm_map_copy_entry_unlink(head_copy, entry);
+ vm_map_copy_entry_link(copy,
+ vm_map_copy_to_entry(copy),
+ entry);
+ copy->offset -= head_size;
+ copy->size += head_size;
+ vm_map_copy_discard(head_copy);
+ head_copy = NULL;
+ }
+ if (tail_copy) {
+ entry = vm_map_copy_last_entry(tail_copy);
+ vm_map_copy_entry_unlink(tail_copy, entry);
+ vm_map_copy_entry_link(copy,
+ vm_map_copy_last_entry(copy),
+ entry);
+ copy->size += tail_size;
+ vm_map_copy_discard(tail_copy);
+ tail_copy = NULL;
+ }
+ }
+ return kr;
+}
+
+
+/*
+ * Routine: vm_map_copy_overwrite_unaligned [internal use only]
+ *
+ * Decription:
+ * Physically copy unaligned data
+ *
+ * Implementation:
+ * Unaligned parts of pages have to be physically copied. We use
+ * a modified form of vm_fault_copy (which understands none-aligned
+ * page offsets and sizes) to do the copy. We attempt to copy as
+ * much memory in one go as possibly, however vm_fault_copy copies
+ * within 1 memory object so we have to find the smaller of "amount left"
+ * "source object data size" and "target object data size". With
+ * unaligned data we don't need to split regions, therefore the source
+ * (copy) object should be one map entry, the target range may be split
+ * over multiple map entries however. In any event we are pessimistic
+ * about these assumptions.
+ *
+ * Assumptions:
+ * dst_map is locked on entry and is return locked on success,
+ * unlocked on error.
+ */
+
+static kern_return_t
+vm_map_copy_overwrite_unaligned(
+ vm_map_t dst_map,
+ vm_map_entry_t entry,
+ vm_map_copy_t copy,
+ vm_map_offset_t start,
+ boolean_t discard_on_success)
+{
+ vm_map_entry_t copy_entry;
+ vm_map_entry_t copy_entry_next;
+ vm_map_version_t version;
+ vm_object_t dst_object;
+ vm_object_offset_t dst_offset;
+ vm_object_offset_t src_offset;
+ vm_object_offset_t entry_offset;
+ vm_map_offset_t entry_end;
+ vm_map_size_t src_size,
+ dst_size,
+ copy_size,
+ amount_left;
+ kern_return_t kr = KERN_SUCCESS;
+
+
+ copy_entry = vm_map_copy_first_entry(copy);
+
+ vm_map_lock_write_to_read(dst_map);
+
+ src_offset = copy->offset - vm_object_trunc_page(copy->offset);
+ amount_left = copy->size;
+/*
+ * unaligned so we never clipped this entry, we need the offset into
+ * the vm_object not just the data.
+ */
+ while (amount_left > 0) {
+
+ if (entry == vm_map_to_entry(dst_map)) {
+ vm_map_unlock_read(dst_map);
+ return KERN_INVALID_ADDRESS;
+ }
+
+ /* "start" must be within the current map entry */
+ assert ((start>=entry->vme_start) && (start<entry->vme_end));
+
+ dst_offset = start - entry->vme_start;
+
+ dst_size = entry->vme_end - start;
+
+ src_size = copy_entry->vme_end -
+ (copy_entry->vme_start + src_offset);
+
+ if (dst_size < src_size) {
+/*
+ * we can only copy dst_size bytes before
+ * we have to get the next destination entry
+ */
+ copy_size = dst_size;
+ } else {
+/*
+ * we can only copy src_size bytes before
+ * we have to get the next source copy entry
+ */
+ copy_size = src_size;
+ }
+
+ if (copy_size > amount_left) {
+ copy_size = amount_left;
+ }
+/*
+ * Entry needs copy, create a shadow shadow object for
+ * Copy on write region.
+ */
+ if (entry->needs_copy &&
+ ((entry->protection & VM_PROT_WRITE) != 0))
+ {
+ if (vm_map_lock_read_to_write(dst_map)) {
+ vm_map_lock_read(dst_map);
+ goto RetryLookup;
+ }
+ VME_OBJECT_SHADOW(entry,
+ (vm_map_size_t)(entry->vme_end
+ - entry->vme_start));
+ entry->needs_copy = FALSE;
+ vm_map_lock_write_to_read(dst_map);
+ }
+ dst_object = VME_OBJECT(entry);
+/*
+ * unlike with the virtual (aligned) copy we're going
+ * to fault on it therefore we need a target object.
+ */
+ if (dst_object == VM_OBJECT_NULL) {
+ if (vm_map_lock_read_to_write(dst_map)) {
+ vm_map_lock_read(dst_map);
+ goto RetryLookup;
+ }
+ dst_object = vm_object_allocate((vm_map_size_t)
+ entry->vme_end - entry->vme_start);
+ VME_OBJECT(entry) = dst_object;
+ VME_OFFSET_SET(entry, 0);
+ assert(entry->use_pmap);
+ vm_map_lock_write_to_read(dst_map);
+ }
+/*
+ * Take an object reference and unlock map. The "entry" may
+ * disappear or change when the map is unlocked.
+ */
+ vm_object_reference(dst_object);
+ version.main_timestamp = dst_map->timestamp;
+ entry_offset = VME_OFFSET(entry);
+ entry_end = entry->vme_end;
+ vm_map_unlock_read(dst_map);
+/*
+ * Copy as much as possible in one pass
+ */
+ kr = vm_fault_copy(
+ VME_OBJECT(copy_entry),
+ VME_OFFSET(copy_entry) + src_offset,
+ ©_size,
+ dst_object,
+ entry_offset + dst_offset,
+ dst_map,
+ &version,
+ THREAD_UNINT );
+
+ start += copy_size;
+ src_offset += copy_size;
+ amount_left -= copy_size;
+/*
+ * Release the object reference
+ */
+ vm_object_deallocate(dst_object);
+/*
+ * If a hard error occurred, return it now
+ */
+ if (kr != KERN_SUCCESS)
+ return kr;
+
+ if ((copy_entry->vme_start + src_offset) == copy_entry->vme_end
+ || amount_left == 0)
+ {
+/*
+ * all done with this copy entry, dispose.
+ */
+ copy_entry_next = copy_entry->vme_next;
+
+ if (discard_on_success) {
+ vm_map_copy_entry_unlink(copy, copy_entry);
+ assert(!copy_entry->is_sub_map);
+ vm_object_deallocate(VME_OBJECT(copy_entry));
+ vm_map_copy_entry_dispose(copy, copy_entry);
+ }
+
+ if (copy_entry_next == vm_map_copy_to_entry(copy) &&
+ amount_left) {
+/*
+ * not finished copying but run out of source
+ */
+ return KERN_INVALID_ADDRESS;
+ }
+
+ copy_entry = copy_entry_next;
+
+ src_offset = 0;
+ }
+
+ if (amount_left == 0)
+ return KERN_SUCCESS;
+
+ vm_map_lock_read(dst_map);
+ if (version.main_timestamp == dst_map->timestamp) {
+ if (start == entry_end) {
+/*
+ * destination region is split. Use the version
+ * information to avoid a lookup in the normal
+ * case.
+ */
+ entry = entry->vme_next;
+/*
+ * should be contiguous. Fail if we encounter
+ * a hole in the destination.
+ */
+ if (start != entry->vme_start) {
+ vm_map_unlock_read(dst_map);
+ return KERN_INVALID_ADDRESS ;
+ }
+ }
+ } else {
+/*
+ * Map version check failed.
+ * we must lookup the entry because somebody
+ * might have changed the map behind our backs.
+ */
+ RetryLookup:
+ if (!vm_map_lookup_entry(dst_map, start, &entry))
+ {
+ vm_map_unlock_read(dst_map);
+ return KERN_INVALID_ADDRESS ;
+ }
+ }
+ }/* while */
+
+ return KERN_SUCCESS;
+}/* vm_map_copy_overwrite_unaligned */
+
+/*
+ * Routine: vm_map_copy_overwrite_aligned [internal use only]
+ *
+ * Description:
+ * Does all the vm_trickery possible for whole pages.
+ *
+ * Implementation:
+ *
+ * If there are no permanent objects in the destination,
+ * and the source and destination map entry zones match,
+ * and the destination map entry is not shared,
+ * then the map entries can be deleted and replaced
+ * with those from the copy. The following code is the
+ * basic idea of what to do, but there are lots of annoying
+ * little details about getting protection and inheritance
+ * right. Should add protection, inheritance, and sharing checks
+ * to the above pass and make sure that no wiring is involved.
+ */
+
+int vm_map_copy_overwrite_aligned_src_not_internal = 0;
+int vm_map_copy_overwrite_aligned_src_not_symmetric = 0;
+int vm_map_copy_overwrite_aligned_src_large = 0;
+
+static kern_return_t
+vm_map_copy_overwrite_aligned(
+ vm_map_t dst_map,
+ vm_map_entry_t tmp_entry,
+ vm_map_copy_t copy,
+ vm_map_offset_t start,
+ __unused pmap_t pmap)
+{
+ vm_object_t object;
+ vm_map_entry_t copy_entry;
+ vm_map_size_t copy_size;
+ vm_map_size_t size;
+ vm_map_entry_t entry;
+
+ while ((copy_entry = vm_map_copy_first_entry(copy))
+ != vm_map_copy_to_entry(copy))
+ {
+ copy_size = (copy_entry->vme_end - copy_entry->vme_start);
+
+ entry = tmp_entry;
+ if (entry->is_sub_map) {
+ /* unnested when clipped earlier */
+ assert(!entry->use_pmap);
+ }
+ if (entry == vm_map_to_entry(dst_map)) {
+ vm_map_unlock(dst_map);
+ return KERN_INVALID_ADDRESS;
+ }
+ size = (entry->vme_end - entry->vme_start);
+ /*
+ * Make sure that no holes popped up in the
+ * address map, and that the protection is
+ * still valid, in case the map was unlocked
+ * earlier.
+ */
+
+ if ((entry->vme_start != start) || ((entry->is_sub_map)
+ && !entry->needs_copy)) {
+ vm_map_unlock(dst_map);
+ return(KERN_INVALID_ADDRESS);
+ }
+ assert(entry != vm_map_to_entry(dst_map));
+
+ /*
+ * Check protection again
+ */
+
+ if ( ! (entry->protection & VM_PROT_WRITE)) {
+ vm_map_unlock(dst_map);
+ return(KERN_PROTECTION_FAILURE);
+ }
+
+ /*
+ * Adjust to source size first
+ */
+
+ if (copy_size < size) {
+ if (entry->map_aligned &&
+ !VM_MAP_PAGE_ALIGNED(entry->vme_start + copy_size,
+ VM_MAP_PAGE_MASK(dst_map))) {
+ /* no longer map-aligned */
+ entry->map_aligned = FALSE;
+ }
+ vm_map_clip_end(dst_map, entry, entry->vme_start + copy_size);
+ size = copy_size;
+ }
+
+ /*
+ * Adjust to destination size
+ */
+
+ if (size < copy_size) {
+ vm_map_copy_clip_end(copy, copy_entry,
+ copy_entry->vme_start + size);
+ copy_size = size;
+ }
+
+ assert((entry->vme_end - entry->vme_start) == size);
+ assert((tmp_entry->vme_end - tmp_entry->vme_start) == size);
+ assert((copy_entry->vme_end - copy_entry->vme_start) == size);
+
+ /*
+ * If the destination contains temporary unshared memory,
+ * we can perform the copy by throwing it away and
+ * installing the source data.
+ */
+
+ object = VME_OBJECT(entry);
+ if ((!entry->is_shared &&
+ ((object == VM_OBJECT_NULL) ||
+ (object->internal && !object->true_share))) ||
+ entry->needs_copy) {
+ vm_object_t old_object = VME_OBJECT(entry);
+ vm_object_offset_t old_offset = VME_OFFSET(entry);
+ vm_object_offset_t offset;
+
+ /*
+ * Ensure that the source and destination aren't
+ * identical
+ */
+ if (old_object == VME_OBJECT(copy_entry) &&
+ old_offset == VME_OFFSET(copy_entry)) {
+ vm_map_copy_entry_unlink(copy, copy_entry);
+ vm_map_copy_entry_dispose(copy, copy_entry);
+
+ if (old_object != VM_OBJECT_NULL)
+ vm_object_deallocate(old_object);
+
+ start = tmp_entry->vme_end;
+ tmp_entry = tmp_entry->vme_next;
+ continue;
+ }
+
+#define __TRADEOFF1_OBJ_SIZE (64 * 1024 * 1024) /* 64 MB */
+#define __TRADEOFF1_COPY_SIZE (128 * 1024) /* 128 KB */
+ if (VME_OBJECT(copy_entry) != VM_OBJECT_NULL &&
+ VME_OBJECT(copy_entry)->vo_size >= __TRADEOFF1_OBJ_SIZE &&
+ copy_size <= __TRADEOFF1_COPY_SIZE) {
+ /*
+ * Virtual vs. Physical copy tradeoff #1.
+ *
+ * Copying only a few pages out of a large
+ * object: do a physical copy instead of
+ * a virtual copy, to avoid possibly keeping
+ * the entire large object alive because of
+ * those few copy-on-write pages.
+ */
+ vm_map_copy_overwrite_aligned_src_large++;
+ goto slow_copy;
+ }
+
+ if ((dst_map->pmap != kernel_pmap) &&
+ (VME_ALIAS(entry) >= VM_MEMORY_MALLOC) &&
+ (VME_ALIAS(entry) <= VM_MEMORY_MALLOC_LARGE_REUSED)) {
+ vm_object_t new_object, new_shadow;
+
+ /*
+ * We're about to map something over a mapping
+ * established by malloc()...
+ */
+ new_object = VME_OBJECT(copy_entry);
+ if (new_object != VM_OBJECT_NULL) {
+ vm_object_lock_shared(new_object);
+ }
+ while (new_object != VM_OBJECT_NULL &&
+ !new_object->true_share &&
+ new_object->copy_strategy == MEMORY_OBJECT_COPY_SYMMETRIC &&
+ new_object->internal) {
+ new_shadow = new_object->shadow;
+ if (new_shadow == VM_OBJECT_NULL) {
+ break;
+ }
+ vm_object_lock_shared(new_shadow);
+ vm_object_unlock(new_object);
+ new_object = new_shadow;
+ }
+ if (new_object != VM_OBJECT_NULL) {
+ if (!new_object->internal) {
+ /*
+ * The new mapping is backed
+ * by an external object. We
+ * don't want malloc'ed memory
+ * to be replaced with such a
+ * non-anonymous mapping, so
+ * let's go off the optimized
+ * path...
+ */
+ vm_map_copy_overwrite_aligned_src_not_internal++;
+ vm_object_unlock(new_object);
+ goto slow_copy;
+ }
+ if (new_object->true_share ||
+ new_object->copy_strategy != MEMORY_OBJECT_COPY_SYMMETRIC) {
+ /*
+ * Same if there's a "true_share"
+ * object in the shadow chain, or
+ * an object with a non-default
+ * (SYMMETRIC) copy strategy.
+ */
+ vm_map_copy_overwrite_aligned_src_not_symmetric++;
+ vm_object_unlock(new_object);
+ goto slow_copy;
+ }
+ vm_object_unlock(new_object);
+ }
+ /*
+ * The new mapping is still backed by
+ * anonymous (internal) memory, so it's
+ * OK to substitute it for the original
+ * malloc() mapping.
+ */
+ }
+
+ if (old_object != VM_OBJECT_NULL) {
+ if(entry->is_sub_map) {
+ if(entry->use_pmap) {
+#ifndef NO_NESTED_PMAP
+ pmap_unnest(dst_map->pmap,
+ (addr64_t)entry->vme_start,
+ entry->vme_end - entry->vme_start);
+#endif /* NO_NESTED_PMAP */
+ if(dst_map->mapped_in_other_pmaps) {
+ /* clean up parent */
+ /* map/maps */
+ vm_map_submap_pmap_clean(
+ dst_map, entry->vme_start,
+ entry->vme_end,
+ VME_SUBMAP(entry),
+ VME_OFFSET(entry));
+ }
+ } else {
+ vm_map_submap_pmap_clean(
+ dst_map, entry->vme_start,
+ entry->vme_end,
+ VME_SUBMAP(entry),
+ VME_OFFSET(entry));
+ }
+ vm_map_deallocate(VME_SUBMAP(entry));
+ } else {
+ if(dst_map->mapped_in_other_pmaps) {
+ vm_object_pmap_protect_options(
+ VME_OBJECT(entry),
+ VME_OFFSET(entry),
+ entry->vme_end
+ - entry->vme_start,
+ PMAP_NULL,
+ entry->vme_start,
+ VM_PROT_NONE,
+ PMAP_OPTIONS_REMOVE);
+ } else {
+ pmap_remove_options(
+ dst_map->pmap,
+ (addr64_t)(entry->vme_start),
+ (addr64_t)(entry->vme_end),
+ PMAP_OPTIONS_REMOVE);
+ }
+ vm_object_deallocate(old_object);
+ }
+ }
+
+ entry->is_sub_map = FALSE;
+ VME_OBJECT_SET(entry, VME_OBJECT(copy_entry));
+ object = VME_OBJECT(entry);
+ entry->needs_copy = copy_entry->needs_copy;
+ entry->wired_count = 0;
+ entry->user_wired_count = 0;
+ offset = VME_OFFSET(copy_entry);
+ VME_OFFSET_SET(entry, offset);
+
+ vm_map_copy_entry_unlink(copy, copy_entry);
+ vm_map_copy_entry_dispose(copy, copy_entry);
+
+ /*
+ * we could try to push pages into the pmap at this point, BUT
+ * this optimization only saved on average 2 us per page if ALL
+ * the pages in the source were currently mapped
+ * and ALL the pages in the dest were touched, if there were fewer
+ * than 2/3 of the pages touched, this optimization actually cost more cycles
+ * it also puts a lot of pressure on the pmap layer w/r to mapping structures
+ */
+
+ /*
+ * Set up for the next iteration. The map
+ * has not been unlocked, so the next
+ * address should be at the end of this
+ * entry, and the next map entry should be
+ * the one following it.
+ */
+
+ start = tmp_entry->vme_end;
+ tmp_entry = tmp_entry->vme_next;
+ } else {
+ vm_map_version_t version;
+ vm_object_t dst_object;
+ vm_object_offset_t dst_offset;
+ kern_return_t r;
+
+ slow_copy:
+ if (entry->needs_copy) {
+ VME_OBJECT_SHADOW(entry,
+ (entry->vme_end -
+ entry->vme_start));
+ entry->needs_copy = FALSE;
+ }
+
+ dst_object = VME_OBJECT(entry);
+ dst_offset = VME_OFFSET(entry);
+
+ /*
+ * Take an object reference, and record
+ * the map version information so that the
+ * map can be safely unlocked.
+ */
+
+ if (dst_object == VM_OBJECT_NULL) {
+ /*
+ * We would usually have just taken the
+ * optimized path above if the destination
+ * object has not been allocated yet. But we
+ * now disable that optimization if the copy
+ * entry's object is not backed by anonymous
+ * memory to avoid replacing malloc'ed
+ * (i.e. re-usable) anonymous memory with a
+ * not-so-anonymous mapping.
+ * So we have to handle this case here and
+ * allocate a new VM object for this map entry.
+ */
+ dst_object = vm_object_allocate(
+ entry->vme_end - entry->vme_start);
+ dst_offset = 0;
+ VME_OBJECT_SET(entry, dst_object);
+ VME_OFFSET_SET(entry, dst_offset);
+ assert(entry->use_pmap);
+
+ }
+
+ vm_object_reference(dst_object);
+
+ /* account for unlock bumping up timestamp */
+ version.main_timestamp = dst_map->timestamp + 1;
+
+ vm_map_unlock(dst_map);
+
+ /*
+ * Copy as much as possible in one pass
+ */
+
+ copy_size = size;
+ r = vm_fault_copy(
+ VME_OBJECT(copy_entry),
+ VME_OFFSET(copy_entry),
+ ©_size,
+ dst_object,
+ dst_offset,
+ dst_map,
+ &version,
+ THREAD_UNINT );
+
+ /*
+ * Release the object reference
+ */
+
+ vm_object_deallocate(dst_object);
+
+ /*
+ * If a hard error occurred, return it now
+ */
+
+ if (r != KERN_SUCCESS)
+ return(r);
+
+ if (copy_size != 0) {
+ /*
+ * Dispose of the copied region
+ */
+
+ vm_map_copy_clip_end(copy, copy_entry,
+ copy_entry->vme_start + copy_size);
+ vm_map_copy_entry_unlink(copy, copy_entry);
+ vm_object_deallocate(VME_OBJECT(copy_entry));
+ vm_map_copy_entry_dispose(copy, copy_entry);
+ }
+
+ /*
+ * Pick up in the destination map where we left off.
+ *
+ * Use the version information to avoid a lookup
+ * in the normal case.
+ */
+
+ start += copy_size;
+ vm_map_lock(dst_map);
+ if (version.main_timestamp == dst_map->timestamp &&
+ copy_size != 0) {
+ /* We can safely use saved tmp_entry value */
+
+ if (tmp_entry->map_aligned &&
+ !VM_MAP_PAGE_ALIGNED(
+ start,
+ VM_MAP_PAGE_MASK(dst_map))) {
+ /* no longer map-aligned */
+ tmp_entry->map_aligned = FALSE;
+ }
+ vm_map_clip_end(dst_map, tmp_entry, start);
+ tmp_entry = tmp_entry->vme_next;
+ } else {
+ /* Must do lookup of tmp_entry */
+
+ if (!vm_map_lookup_entry(dst_map, start, &tmp_entry)) {
+ vm_map_unlock(dst_map);
+ return(KERN_INVALID_ADDRESS);
+ }
+ if (tmp_entry->map_aligned &&
+ !VM_MAP_PAGE_ALIGNED(
+ start,
+ VM_MAP_PAGE_MASK(dst_map))) {
+ /* no longer map-aligned */
+ tmp_entry->map_aligned = FALSE;
+ }
+ vm_map_clip_start(dst_map, tmp_entry, start);
+ }
+ }
+ }/* while */
+
+ return(KERN_SUCCESS);
+}/* vm_map_copy_overwrite_aligned */
+
+/*
+ * Routine: vm_map_copyin_kernel_buffer [internal use only]
+ *
+ * Description:
+ * Copy in data to a kernel buffer from space in the
+ * source map. The original space may be optionally
+ * deallocated.
+ *
+ * If successful, returns a new copy object.
+ */
+static kern_return_t
+vm_map_copyin_kernel_buffer(
+ vm_map_t src_map,
+ vm_map_offset_t src_addr,
+ vm_map_size_t len,
+ boolean_t src_destroy,
+ vm_map_copy_t *copy_result)
+{
+ kern_return_t kr;
+ vm_map_copy_t copy;
+ vm_size_t kalloc_size;
+
+ if (len > msg_ool_size_small)
+ return KERN_INVALID_ARGUMENT;
+
+ kalloc_size = (vm_size_t)(cpy_kdata_hdr_sz + len);
+
+ copy = (vm_map_copy_t)kalloc(kalloc_size);
+ if (copy == VM_MAP_COPY_NULL)
+ return KERN_RESOURCE_SHORTAGE;
+ copy->type = VM_MAP_COPY_KERNEL_BUFFER;
+ copy->size = len;
+ copy->offset = 0;
+
+ kr = copyinmap(src_map, src_addr, copy->cpy_kdata, (vm_size_t)len);
+ if (kr != KERN_SUCCESS) {
+ kfree(copy, kalloc_size);
+ return kr;
+ }
+ if (src_destroy) {
+ (void) vm_map_remove(
+ src_map,
+ vm_map_trunc_page(src_addr,
+ VM_MAP_PAGE_MASK(src_map)),
+ vm_map_round_page(src_addr + len,
+ VM_MAP_PAGE_MASK(src_map)),
+ (VM_MAP_REMOVE_INTERRUPTIBLE |
+ VM_MAP_REMOVE_WAIT_FOR_KWIRE |
+ (src_map == kernel_map) ? VM_MAP_REMOVE_KUNWIRE : 0));
+ }
+ *copy_result = copy;
+ return KERN_SUCCESS;
+}
+
+/*
+ * Routine: vm_map_copyout_kernel_buffer [internal use only]
+ *
+ * Description:
+ * Copy out data from a kernel buffer into space in the
+ * destination map. The space may be otpionally dynamically
+ * allocated.
+ *
+ * If successful, consumes the copy object.
+ * Otherwise, the caller is responsible for it.
+ */
+static int vm_map_copyout_kernel_buffer_failures = 0;
+static kern_return_t
+vm_map_copyout_kernel_buffer(
+ vm_map_t map,
+ vm_map_address_t *addr, /* IN/OUT */
+ vm_map_copy_t copy,
+ boolean_t overwrite,
+ boolean_t consume_on_success)
+{
+ kern_return_t kr = KERN_SUCCESS;
+ thread_t thread = current_thread();
+
+ /*
+ * check for corrupted vm_map_copy structure
+ */
+ if (copy->size > msg_ool_size_small || copy->offset)
+ panic("Invalid vm_map_copy_t sz:%lld, ofst:%lld",
+ (long long)copy->size, (long long)copy->offset);
+
+ if (!overwrite) {
+
+ /*
+ * Allocate space in the target map for the data
+ */
+ *addr = 0;
+ kr = vm_map_enter(map,
+ addr,
+ vm_map_round_page(copy->size,
+ VM_MAP_PAGE_MASK(map)),
+ (vm_map_offset_t) 0,
+ VM_FLAGS_ANYWHERE,
+ VM_OBJECT_NULL,
+ (vm_object_offset_t) 0,
+ FALSE,
+ VM_PROT_DEFAULT,
+ VM_PROT_ALL,
+ VM_INHERIT_DEFAULT);
+ if (kr != KERN_SUCCESS)
+ return kr;
+ }
+
+ /*
+ * Copyout the data from the kernel buffer to the target map.
+ */
+ if (thread->map == map) {
+
+ /*
+ * If the target map is the current map, just do
+ * the copy.
+ */
+ assert((vm_size_t) copy->size == copy->size);
+ if (copyout(copy->cpy_kdata, *addr, (vm_size_t) copy->size)) {
+ kr = KERN_INVALID_ADDRESS;
+ }
+ }
+ else {
+ vm_map_t oldmap;
+
+ /*
+ * If the target map is another map, assume the
+ * target's address space identity for the duration
+ * of the copy.
+ */
+ vm_map_reference(map);
+ oldmap = vm_map_switch(map);
+
+ assert((vm_size_t) copy->size == copy->size);
+ if (copyout(copy->cpy_kdata, *addr, (vm_size_t) copy->size)) {
+ vm_map_copyout_kernel_buffer_failures++;
+ kr = KERN_INVALID_ADDRESS;
+ }
+
+ (void) vm_map_switch(oldmap);
+ vm_map_deallocate(map);
+ }
+
+ if (kr != KERN_SUCCESS) {
+ /* the copy failed, clean up */
+ if (!overwrite) {
+ /*
+ * Deallocate the space we allocated in the target map.
+ */
+ (void) vm_map_remove(
+ map,
+ vm_map_trunc_page(*addr,
+ VM_MAP_PAGE_MASK(map)),
+ vm_map_round_page((*addr +
+ vm_map_round_page(copy->size,
+ VM_MAP_PAGE_MASK(map))),
+ VM_MAP_PAGE_MASK(map)),
+ VM_MAP_NO_FLAGS);
+ *addr = 0;
+ }
+ } else {
+ /* copy was successful, dicard the copy structure */
+ if (consume_on_success) {
+ kfree(copy, copy->size + cpy_kdata_hdr_sz);
+ }
+ }
+
+ return kr;
+}
+
+/*
+ * Macro: vm_map_copy_insert
+ *
+ * Description:
+ * Link a copy chain ("copy") into a map at the
+ * specified location (after "where").
+ * Side effects:
+ * The copy chain is destroyed.
+ * Warning:
+ * The arguments are evaluated multiple times.
+ */
+#define vm_map_copy_insert(map, where, copy) \
+MACRO_BEGIN \
+ vm_map_store_copy_insert(map, where, copy); \
+ zfree(vm_map_copy_zone, copy); \
+MACRO_END
+
+void
+vm_map_copy_remap(
+ vm_map_t map,
+ vm_map_entry_t where,
+ vm_map_copy_t copy,
+ vm_map_offset_t adjustment,
+ vm_prot_t cur_prot,
+ vm_prot_t max_prot,
+ vm_inherit_t inheritance)
+{
+ vm_map_entry_t copy_entry, new_entry;
+
+ for (copy_entry = vm_map_copy_first_entry(copy);
+ copy_entry != vm_map_copy_to_entry(copy);
+ copy_entry = copy_entry->vme_next) {
+ /* get a new VM map entry for the map */
+ new_entry = vm_map_entry_create(map,
+ !map->hdr.entries_pageable);
+ /* copy the "copy entry" to the new entry */
+ vm_map_entry_copy(new_entry, copy_entry);
+ /* adjust "start" and "end" */
+ new_entry->vme_start += adjustment;
+ new_entry->vme_end += adjustment;
+ /* clear some attributes */
+ new_entry->inheritance = inheritance;
+ new_entry->protection = cur_prot;
+ new_entry->max_protection = max_prot;
+ new_entry->behavior = VM_BEHAVIOR_DEFAULT;
+ /* take an extra reference on the entry's "object" */
+ if (new_entry->is_sub_map) {
+ assert(!new_entry->use_pmap); /* not nested */
+ vm_map_lock(VME_SUBMAP(new_entry));
+ vm_map_reference(VME_SUBMAP(new_entry));
+ vm_map_unlock(VME_SUBMAP(new_entry));
+ } else {
+ vm_object_reference(VME_OBJECT(new_entry));
+ }
+ /* insert the new entry in the map */
+ vm_map_store_entry_link(map, where, new_entry);
+ /* continue inserting the "copy entries" after the new entry */
+ where = new_entry;
+ }
+}
+
+
+boolean_t
+vm_map_copy_validate_size(
+ vm_map_t dst_map,
+ vm_map_copy_t copy,
+ vm_map_size_t size)
+{
+ if (copy == VM_MAP_COPY_NULL)
+ return FALSE;
+ switch (copy->type) {
+ case VM_MAP_COPY_OBJECT:
+ case VM_MAP_COPY_KERNEL_BUFFER:
+ if (size == copy->size)
+ return TRUE;
+ break;
+ case VM_MAP_COPY_ENTRY_LIST:
+ /*
+ * potential page-size rounding prevents us from exactly
+ * validating this flavor of vm_map_copy, but we can at least
+ * assert that it's within a range.
+ */
+ if (copy->size >= size &&
+ copy->size <= vm_map_round_page(size,
+ VM_MAP_PAGE_MASK(dst_map)))
+ return TRUE;
+ break;
+ default:
+ break;
+ }
+ return FALSE;
+}
+
+
+/*
+ * Routine: vm_map_copyout
+ *
+ * Description:
+ * Copy out a copy chain ("copy") into newly-allocated
+ * space in the destination map.
+ *
+ * If successful, consumes the copy object.
+ * Otherwise, the caller is responsible for it.
+ */
+
+kern_return_t
+vm_map_copyout(
+ vm_map_t dst_map,
+ vm_map_address_t *dst_addr, /* OUT */
+ vm_map_copy_t copy)
+{
+ return vm_map_copyout_internal(dst_map, dst_addr, copy,
+ TRUE, /* consume_on_success */
+ VM_PROT_DEFAULT,
+ VM_PROT_ALL,
+ VM_INHERIT_DEFAULT);
+}
+
+kern_return_t
+vm_map_copyout_internal(
+ vm_map_t dst_map,
+ vm_map_address_t *dst_addr, /* OUT */
+ vm_map_copy_t copy,
+ boolean_t consume_on_success,
+ vm_prot_t cur_protection,
+ vm_prot_t max_protection,
+ vm_inherit_t inheritance)
+{
+ vm_map_size_t size;
+ vm_map_size_t adjustment;
+ vm_map_offset_t start;
+ vm_object_offset_t vm_copy_start;
+ vm_map_entry_t last;
+ vm_map_entry_t entry;
+ vm_map_entry_t hole_entry;
+
+ /*
+ * Check for null copy object.
+ */
+
+ if (copy == VM_MAP_COPY_NULL) {
+ *dst_addr = 0;
+ return(KERN_SUCCESS);
+ }
+
+ /*
+ * Check for special copy object, created
+ * by vm_map_copyin_object.
+ */
+
+ if (copy->type == VM_MAP_COPY_OBJECT) {
+ vm_object_t object = copy->cpy_object;
+ kern_return_t kr;
+ vm_object_offset_t offset;
+
+ offset = vm_object_trunc_page(copy->offset);
+ size = vm_map_round_page((copy->size +
+ (vm_map_size_t)(copy->offset -
+ offset)),
+ VM_MAP_PAGE_MASK(dst_map));
+ *dst_addr = 0;
+ kr = vm_map_enter(dst_map, dst_addr, size,
+ (vm_map_offset_t) 0, VM_FLAGS_ANYWHERE,
+ object, offset, FALSE,
+ VM_PROT_DEFAULT, VM_PROT_ALL,
+ VM_INHERIT_DEFAULT);
+ if (kr != KERN_SUCCESS)
+ return(kr);
+ /* Account for non-pagealigned copy object */
+ *dst_addr += (vm_map_offset_t)(copy->offset - offset);
+ if (consume_on_success)
+ zfree(vm_map_copy_zone, copy);
+ return(KERN_SUCCESS);
+ }
+
+ /*
+ * Check for special kernel buffer allocated
+ * by new_ipc_kmsg_copyin.
+ */
+
+ if (copy->type == VM_MAP_COPY_KERNEL_BUFFER) {
+ return vm_map_copyout_kernel_buffer(dst_map, dst_addr,
+ copy, FALSE,
+ consume_on_success);
+ }
+
+
+ /*
+ * Find space for the data
+ */
+
+ vm_copy_start = vm_map_trunc_page((vm_map_size_t)copy->offset,
+ VM_MAP_COPY_PAGE_MASK(copy));
+ size = vm_map_round_page((vm_map_size_t)copy->offset + copy->size,
+ VM_MAP_COPY_PAGE_MASK(copy))
+ - vm_copy_start;
+
+
+StartAgain: ;
+
+ vm_map_lock(dst_map);
+ if( dst_map->disable_vmentry_reuse == TRUE) {
+ VM_MAP_HIGHEST_ENTRY(dst_map, entry, start);
+ last = entry;
+ } else {
+ if (dst_map->holelistenabled) {
+ hole_entry = (vm_map_entry_t)dst_map->holes_list;
+
+ if (hole_entry == NULL) {
+ /*
+ * No more space in the map?
+ */
+ vm_map_unlock(dst_map);
+ return(KERN_NO_SPACE);
+ }
+
+ last = hole_entry;
+ start = last->vme_start;
+ } else {
+ assert(first_free_is_valid(dst_map));
+ start = ((last = dst_map->first_free) == vm_map_to_entry(dst_map)) ?
+ vm_map_min(dst_map) : last->vme_end;
+ }
+ start = vm_map_round_page(start,
+ VM_MAP_PAGE_MASK(dst_map));
+ }
+
+ while (TRUE) {
+ vm_map_entry_t next = last->vme_next;
+ vm_map_offset_t end = start + size;
+
+ if ((end > dst_map->max_offset) || (end < start)) {
+ if (dst_map->wait_for_space) {
+ if (size <= (dst_map->max_offset - dst_map->min_offset)) {
+ assert_wait((event_t) dst_map,
+ THREAD_INTERRUPTIBLE);
+ vm_map_unlock(dst_map);
+ thread_block(THREAD_CONTINUE_NULL);
+ goto StartAgain;
+ }
+ }
+ vm_map_unlock(dst_map);
+ return(KERN_NO_SPACE);
+ }
+
+ if (dst_map->holelistenabled) {
+ if (last->vme_end >= end)
+ break;
+ } else {
+ /*
+ * If there are no more entries, we must win.
+ *
+ * OR
+ *
+ * If there is another entry, it must be
+ * after the end of the potential new region.
+ */
+
+ if (next == vm_map_to_entry(dst_map))
+ break;
+
+ if (next->vme_start >= end)
+ break;
+ }
+
+ last = next;
+
+ if (dst_map->holelistenabled) {
+ if (last == (vm_map_entry_t) dst_map->holes_list) {
+ /*
+ * Wrapped around
+ */
+ vm_map_unlock(dst_map);
+ return(KERN_NO_SPACE);
+ }
+ start = last->vme_start;
+ } else {
+ start = last->vme_end;
+ }
+ start = vm_map_round_page(start,
+ VM_MAP_PAGE_MASK(dst_map));
+ }
+
+ if (dst_map->holelistenabled) {
+ if (vm_map_lookup_entry(dst_map, last->vme_start, &last)) {
+ panic("Found an existing entry (%p) instead of potential hole at address: 0x%llx.\n", last, (unsigned long long)last->vme_start);
+ }
+ }
+
+
+ adjustment = start - vm_copy_start;
+ if (! consume_on_success) {
+ /*
+ * We're not allowed to consume "copy", so we'll have to
+ * copy its map entries into the destination map below.
+ * No need to re-allocate map entries from the correct
+ * (pageable or not) zone, since we'll get new map entries
+ * during the transfer.
+ * We'll also adjust the map entries's "start" and "end"
+ * during the transfer, to keep "copy"'s entries consistent
+ * with its "offset".
+ */
+ goto after_adjustments;
+ }
+
+ /*
+ * Since we're going to just drop the map
+ * entries from the copy into the destination
+ * map, they must come from the same pool.
+ */
+
+ if (copy->cpy_hdr.entries_pageable != dst_map->hdr.entries_pageable) {
+ /*
+ * Mismatches occur when dealing with the default
+ * pager.
+ */
+ zone_t old_zone;
+ vm_map_entry_t next, new;
+
+ /*
+ * Find the zone that the copies were allocated from
+ */
+
+ entry = vm_map_copy_first_entry(copy);
+
+ /*
+ * Reinitialize the copy so that vm_map_copy_entry_link
+ * will work.
+ */
+ vm_map_store_copy_reset(copy, entry);
+ copy->cpy_hdr.entries_pageable = dst_map->hdr.entries_pageable;
+
+ /*
+ * Copy each entry.
+ */
+ while (entry != vm_map_copy_to_entry(copy)) {
+ new = vm_map_copy_entry_create(copy, !copy->cpy_hdr.entries_pageable);
+ vm_map_entry_copy_full(new, entry);
+ assert(!new->iokit_acct);
+ if (new->is_sub_map) {
+ /* clr address space specifics */
+ new->use_pmap = FALSE;
+ }
+ vm_map_copy_entry_link(copy,
+ vm_map_copy_last_entry(copy),
+ new);
+ next = entry->vme_next;
+ old_zone = entry->from_reserved_zone ? vm_map_entry_reserved_zone : vm_map_entry_zone;
+ zfree(old_zone, entry);
+ entry = next;
+ }
+ }
+
+ /*
+ * Adjust the addresses in the copy chain, and
+ * reset the region attributes.
+ */
+
+ for (entry = vm_map_copy_first_entry(copy);
+ entry != vm_map_copy_to_entry(copy);
+ entry = entry->vme_next) {
+ if (VM_MAP_PAGE_SHIFT(dst_map) == PAGE_SHIFT) {
+ /*
+ * We're injecting this copy entry into a map that
+ * has the standard page alignment, so clear
+ * "map_aligned" (which might have been inherited
+ * from the original map entry).
+ */
+ entry->map_aligned = FALSE;
+ }
+
+ entry->vme_start += adjustment;
+ entry->vme_end += adjustment;
+
+ if (entry->map_aligned) {
+ assert(VM_MAP_PAGE_ALIGNED(entry->vme_start,
+ VM_MAP_PAGE_MASK(dst_map)));
+ assert(VM_MAP_PAGE_ALIGNED(entry->vme_end,
+ VM_MAP_PAGE_MASK(dst_map)));
+ }
+
+ entry->inheritance = VM_INHERIT_DEFAULT;
+ entry->protection = VM_PROT_DEFAULT;
+ entry->max_protection = VM_PROT_ALL;
+ entry->behavior = VM_BEHAVIOR_DEFAULT;
+
+ /*
+ * If the entry is now wired,
+ * map the pages into the destination map.
+ */
+ if (entry->wired_count != 0) {
+ register vm_map_offset_t va;
+ vm_object_offset_t offset;
+ register vm_object_t object;
+ vm_prot_t prot;
+ int type_of_fault;
+
+ object = VME_OBJECT(entry);
+ offset = VME_OFFSET(entry);
+ va = entry->vme_start;
+
+ pmap_pageable(dst_map->pmap,
+ entry->vme_start,
+ entry->vme_end,
+ TRUE);
+
+ while (va < entry->vme_end) {
+ register vm_page_t m;
+
+ /*
+ * Look up the page in the object.
+ * Assert that the page will be found in the
+ * top object:
+ * either
+ * the object was newly created by
+ * vm_object_copy_slowly, and has
+ * copies of all of the pages from
+ * the source object
+ * or
+ * the object was moved from the old
+ * map entry; because the old map
+ * entry was wired, all of the pages
+ * were in the top-level object.
+ * (XXX not true if we wire pages for
+ * reading)
+ */
+ vm_object_lock(object);
+
+ m = vm_page_lookup(object, offset);
+ if (m == VM_PAGE_NULL || !VM_PAGE_WIRED(m) ||
+ m->absent)
+ panic("vm_map_copyout: wiring %p", m);
+
+ /*
+ * ENCRYPTED SWAP:
+ * The page is assumed to be wired here, so it
+ * shouldn't be encrypted. Otherwise, we
+ * couldn't enter it in the page table, since
+ * we don't want the user to see the encrypted
+ * data.
+ */
+ ASSERT_PAGE_DECRYPTED(m);
+
+ prot = entry->protection;
+
+ if (override_nx(dst_map, VME_ALIAS(entry)) &&
+ prot)
+ prot |= VM_PROT_EXECUTE;
+
+ type_of_fault = DBG_CACHE_HIT_FAULT;
+
+ vm_fault_enter(m, dst_map->pmap, va, prot, prot,
+ VM_PAGE_WIRED(m), FALSE, FALSE,
+ FALSE, VME_ALIAS(entry),
+ ((entry->iokit_acct ||
+ (!entry->is_sub_map &&
+ !entry->use_pmap))
+ ? PMAP_OPTIONS_ALT_ACCT
+ : 0),
+ NULL, &type_of_fault);
+
+ vm_object_unlock(object);
+
+ offset += PAGE_SIZE_64;
+ va += PAGE_SIZE;
+ }
+ }
+ }
+
+after_adjustments:
+
+ /*
+ * Correct the page alignment for the result
+ */
+
+ *dst_addr = start + (copy->offset - vm_copy_start);
+
+ /*
+ * Update the hints and the map size
+ */
+
+ if (consume_on_success) {
+ SAVE_HINT_MAP_WRITE(dst_map, vm_map_copy_last_entry(copy));
+ } else {
+ SAVE_HINT_MAP_WRITE(dst_map, last);
+ }
+
+ dst_map->size += size;
+
+ /*
+ * Link in the copy
+ */
+
+ if (consume_on_success) {
+ vm_map_copy_insert(dst_map, last, copy);
+ } else {
+ vm_map_copy_remap(dst_map, last, copy, adjustment,
+ cur_protection, max_protection,
+ inheritance);
+ }
+
+ vm_map_unlock(dst_map);
+
+ /*
+ * XXX If wiring_required, call vm_map_pageable
+ */
+
+ return(KERN_SUCCESS);
+}
+
+/*
+ * Routine: vm_map_copyin
+ *
+ * Description:
+ * see vm_map_copyin_common. Exported via Unsupported.exports.
+ *
+ */
+
+#undef vm_map_copyin
+
+kern_return_t
+vm_map_copyin(
+ vm_map_t src_map,
+ vm_map_address_t src_addr,
+ vm_map_size_t len,
+ boolean_t src_destroy,
+ vm_map_copy_t *copy_result) /* OUT */
+{
+ return(vm_map_copyin_common(src_map, src_addr, len, src_destroy,
+ FALSE, copy_result, FALSE));
+}
+
+/*
+ * Routine: vm_map_copyin_common
+ *
+ * Description:
+ * Copy the specified region (src_addr, len) from the
+ * source address space (src_map), possibly removing
+ * the region from the source address space (src_destroy).
+ *
+ * Returns:
+ * A vm_map_copy_t object (copy_result), suitable for
+ * insertion into another address space (using vm_map_copyout),
+ * copying over another address space region (using
+ * vm_map_copy_overwrite). If the copy is unused, it
+ * should be destroyed (using vm_map_copy_discard).
+ *
+ * In/out conditions:
+ * The source map should not be locked on entry.
+ */
+
+typedef struct submap_map {
+ vm_map_t parent_map;
+ vm_map_offset_t base_start;
+ vm_map_offset_t base_end;
+ vm_map_size_t base_len;
+ struct submap_map *next;
+} submap_map_t;
+
+kern_return_t
+vm_map_copyin_common(
+ vm_map_t src_map,
+ vm_map_address_t src_addr,
+ vm_map_size_t len,
+ boolean_t src_destroy,
+ __unused boolean_t src_volatile,
+ vm_map_copy_t *copy_result, /* OUT */
+ boolean_t use_maxprot)
+{
+ int flags;
+
+ flags = 0;
+ if (src_destroy) {
+ flags |= VM_MAP_COPYIN_SRC_DESTROY;
+ }
+ if (use_maxprot) {
+ flags |= VM_MAP_COPYIN_USE_MAXPROT;
+ }
+ return vm_map_copyin_internal(src_map,
+ src_addr,
+ len,
+ flags,
+ copy_result);
+}
+kern_return_t
+vm_map_copyin_internal(
+ vm_map_t src_map,
+ vm_map_address_t src_addr,
+ vm_map_size_t len,
+ int flags,
+ vm_map_copy_t *copy_result) /* OUT */
+{
+ vm_map_entry_t tmp_entry; /* Result of last map lookup --
+ * in multi-level lookup, this
+ * entry contains the actual
+ * vm_object/offset.
+ */
+ vm_map_entry_t new_entry = VM_MAP_ENTRY_NULL; /* Map entry for copy */
+
+ vm_map_offset_t src_start; /* Start of current entry --
+ * where copy is taking place now
+ */
+ vm_map_offset_t src_end; /* End of entire region to be
+ * copied */
+ vm_map_offset_t src_base;
+ vm_map_t base_map = src_map;
+ boolean_t map_share=FALSE;
+ submap_map_t *parent_maps = NULL;
+
+ vm_map_copy_t copy; /* Resulting copy */
+ vm_map_address_t copy_addr;
+ vm_map_size_t copy_size;
+ boolean_t src_destroy;
+ boolean_t use_maxprot;
+
+ if (flags & ~VM_MAP_COPYIN_ALL_FLAGS) {
+ return KERN_INVALID_ARGUMENT;
+ }
+
+ src_destroy = (flags & VM_MAP_COPYIN_SRC_DESTROY) ? TRUE : FALSE;
+ use_maxprot = (flags & VM_MAP_COPYIN_USE_MAXPROT) ? TRUE : FALSE;
+
+ /*
+ * Check for copies of zero bytes.
+ */
+
+ if (len == 0) {
+ *copy_result = VM_MAP_COPY_NULL;
+ return(KERN_SUCCESS);
+ }
+
+ /*
+ * Check that the end address doesn't overflow
+ */
+ src_end = src_addr + len;
+ if (src_end < src_addr)
+ return KERN_INVALID_ADDRESS;
+
+ /*
+ * If the copy is sufficiently small, use a kernel buffer instead
+ * of making a virtual copy. The theory being that the cost of
+ * setting up VM (and taking C-O-W faults) dominates the copy costs
+ * for small regions.
+ */
+ if ((len < msg_ool_size_small) &&
+ !use_maxprot &&
+ !(flags & VM_MAP_COPYIN_ENTRY_LIST))
+ return vm_map_copyin_kernel_buffer(src_map, src_addr, len,
+ src_destroy, copy_result);
+
+ /*
+ * Compute (page aligned) start and end of region
+ */
+ src_start = vm_map_trunc_page(src_addr,
+ VM_MAP_PAGE_MASK(src_map));
+ src_end = vm_map_round_page(src_end,
+ VM_MAP_PAGE_MASK(src_map));
+
+ XPR(XPR_VM_MAP, "vm_map_copyin_common map 0x%x addr 0x%x len 0x%x dest %d\n", src_map, src_addr, len, src_destroy, 0);
+
+ /*
+ * Allocate a header element for the list.
+ *
+ * Use the start and end in the header to
+ * remember the endpoints prior to rounding.
+ */
+
+ copy = (vm_map_copy_t) zalloc(vm_map_copy_zone);
+ copy->c_u.hdr.rb_head_store.rbh_root = (void*)(int)SKIP_RB_TREE;
+ vm_map_copy_first_entry(copy) =
+ vm_map_copy_last_entry(copy) = vm_map_copy_to_entry(copy);
+ copy->type = VM_MAP_COPY_ENTRY_LIST;
+ copy->cpy_hdr.nentries = 0;
+ copy->cpy_hdr.entries_pageable = TRUE;
+#if 00
+ copy->cpy_hdr.page_shift = src_map->hdr.page_shift;
+#else
+ /*
+ * The copy entries can be broken down for a variety of reasons,
+ * so we can't guarantee that they will remain map-aligned...
+ * Will need to adjust the first copy_entry's "vme_start" and
+ * the last copy_entry's "vme_end" to be rounded to PAGE_MASK
+ * rather than the original map's alignment.
+ */
+ copy->cpy_hdr.page_shift = PAGE_SHIFT;
+#endif
+
+ vm_map_store_init( &(copy->cpy_hdr) );
+
+ copy->offset = src_addr;
+ copy->size = len;
+
+ new_entry = vm_map_copy_entry_create(copy, !copy->cpy_hdr.entries_pageable);
+
+#define RETURN(x) \
+ MACRO_BEGIN \
+ vm_map_unlock(src_map); \
+ if(src_map != base_map) \
+ vm_map_deallocate(src_map); \
+ if (new_entry != VM_MAP_ENTRY_NULL) \
+ vm_map_copy_entry_dispose(copy,new_entry); \
+ vm_map_copy_discard(copy); \
+ { \
+ submap_map_t *_ptr; \
+ \
+ for(_ptr = parent_maps; _ptr != NULL; _ptr = parent_maps) { \
+ parent_maps=parent_maps->next; \
+ if (_ptr->parent_map != base_map) \
+ vm_map_deallocate(_ptr->parent_map); \
+ kfree(_ptr, sizeof(submap_map_t)); \
+ } \
+ } \
+ MACRO_RETURN(x); \
+ MACRO_END
+
+ /*
+ * Find the beginning of the region.
+ */
+
+ vm_map_lock(src_map);
+
+ /*
+ * Lookup the original "src_addr" rather than the truncated
+ * "src_start", in case "src_start" falls in a non-map-aligned
+ * map entry *before* the map entry that contains "src_addr"...
+ */
+ if (!vm_map_lookup_entry(src_map, src_addr, &tmp_entry))
+ RETURN(KERN_INVALID_ADDRESS);
+ if(!tmp_entry->is_sub_map) {
+ /*
+ * ... but clip to the map-rounded "src_start" rather than
+ * "src_addr" to preserve map-alignment. We'll adjust the
+ * first copy entry at the end, if needed.
+ */
+ vm_map_clip_start(src_map, tmp_entry, src_start);
+ }
+ if (src_start < tmp_entry->vme_start) {
+ /*
+ * Move "src_start" up to the start of the
+ * first map entry to copy.
+ */
+ src_start = tmp_entry->vme_start;
+ }
+ /* set for later submap fix-up */
+ copy_addr = src_start;
+
+ /*
+ * Go through entries until we get to the end.
+ */
+
+ while (TRUE) {
+ register
+ vm_map_entry_t src_entry = tmp_entry; /* Top-level entry */
+ vm_map_size_t src_size; /* Size of source
+ * map entry (in both
+ * maps)
+ */
+
+ register
+ vm_object_t src_object; /* Object to copy */
+ vm_object_offset_t src_offset;
+
+ boolean_t src_needs_copy; /* Should source map
+ * be made read-only
+ * for copy-on-write?
+ */
+
+ boolean_t new_entry_needs_copy; /* Will new entry be COW? */
+
+ boolean_t was_wired; /* Was source wired? */
+ vm_map_version_t version; /* Version before locks
+ * dropped to make copy
+ */
+ kern_return_t result; /* Return value from
+ * copy_strategically.
+ */
+ while(tmp_entry->is_sub_map) {
+ vm_map_size_t submap_len;
+ submap_map_t *ptr;
+
+ ptr = (submap_map_t *)kalloc(sizeof(submap_map_t));
+ ptr->next = parent_maps;
+ parent_maps = ptr;
+ ptr->parent_map = src_map;
+ ptr->base_start = src_start;
+ ptr->base_end = src_end;
+ submap_len = tmp_entry->vme_end - src_start;
+ if(submap_len > (src_end-src_start))
+ submap_len = src_end-src_start;
+ ptr->base_len = submap_len;
+
+ src_start -= tmp_entry->vme_start;
+ src_start += VME_OFFSET(tmp_entry);
+ src_end = src_start + submap_len;
+ src_map = VME_SUBMAP(tmp_entry);
+ vm_map_lock(src_map);
+ /* keep an outstanding reference for all maps in */
+ /* the parents tree except the base map */
+ vm_map_reference(src_map);
+ vm_map_unlock(ptr->parent_map);
+ if (!vm_map_lookup_entry(
+ src_map, src_start, &tmp_entry))
+ RETURN(KERN_INVALID_ADDRESS);
+ map_share = TRUE;
+ if(!tmp_entry->is_sub_map)
+ vm_map_clip_start(src_map, tmp_entry, src_start);
+ src_entry = tmp_entry;
+ }
+ /* we are now in the lowest level submap... */
+
+ if ((VME_OBJECT(tmp_entry) != VM_OBJECT_NULL) &&
+ (VME_OBJECT(tmp_entry)->phys_contiguous)) {
+ /* This is not, supported for now.In future */
+ /* we will need to detect the phys_contig */
+ /* condition and then upgrade copy_slowly */
+ /* to do physical copy from the device mem */
+ /* based object. We can piggy-back off of */
+ /* the was wired boolean to set-up the */
+ /* proper handling */
+ RETURN(KERN_PROTECTION_FAILURE);
+ }
+ /*
+ * Create a new address map entry to hold the result.
+ * Fill in the fields from the appropriate source entries.
+ * We must unlock the source map to do this if we need
+ * to allocate a map entry.
+ */
+ if (new_entry == VM_MAP_ENTRY_NULL) {
+ version.main_timestamp = src_map->timestamp;
+ vm_map_unlock(src_map);
+
+ new_entry = vm_map_copy_entry_create(copy, !copy->cpy_hdr.entries_pageable);
+
+ vm_map_lock(src_map);
+ if ((version.main_timestamp + 1) != src_map->timestamp) {
+ if (!vm_map_lookup_entry(src_map, src_start,
+ &tmp_entry)) {
+ RETURN(KERN_INVALID_ADDRESS);
+ }
+ if (!tmp_entry->is_sub_map)
+ vm_map_clip_start(src_map, tmp_entry, src_start);
+ continue; /* restart w/ new tmp_entry */
+ }
+ }
+
+ /*
+ * Verify that the region can be read.
+ */
+ if (((src_entry->protection & VM_PROT_READ) == VM_PROT_NONE &&
+ !use_maxprot) ||
+ (src_entry->max_protection & VM_PROT_READ) == 0)
+ RETURN(KERN_PROTECTION_FAILURE);
+
+ /*
+ * Clip against the endpoints of the entire region.
+ */
+
+ vm_map_clip_end(src_map, src_entry, src_end);
+
+ src_size = src_entry->vme_end - src_start;
+ src_object = VME_OBJECT(src_entry);
+ src_offset = VME_OFFSET(src_entry);
+ was_wired = (src_entry->wired_count != 0);
+
+ vm_map_entry_copy(new_entry, src_entry);
+ if (new_entry->is_sub_map) {
+ /* clr address space specifics */
+ new_entry->use_pmap = FALSE;
+ }
+
+ /*
+ * Attempt non-blocking copy-on-write optimizations.
+ */
+
+ if (src_destroy &&
+ (src_object == VM_OBJECT_NULL ||
+ (src_object->internal && !src_object->true_share
+ && !map_share))) {
+ /*
+ * If we are destroying the source, and the object
+ * is internal, we can move the object reference
+ * from the source to the copy. The copy is
+ * copy-on-write only if the source is.
+ * We make another reference to the object, because
+ * destroying the source entry will deallocate it.
+ */
+ vm_object_reference(src_object);
+
+ /*
+ * Copy is always unwired. vm_map_copy_entry
+ * set its wired count to zero.
+ */
+
+ goto CopySuccessful;
+ }
+
+
+ RestartCopy:
+ XPR(XPR_VM_MAP, "vm_map_copyin_common src_obj 0x%x ent 0x%x obj 0x%x was_wired %d\n",
+ src_object, new_entry, VME_OBJECT(new_entry),
+ was_wired, 0);
+ if ((src_object == VM_OBJECT_NULL ||
+ (!was_wired && !map_share && !tmp_entry->is_shared)) &&
+ vm_object_copy_quickly(
+ &VME_OBJECT(new_entry),
+ src_offset,
+ src_size,
+ &src_needs_copy,
+ &new_entry_needs_copy)) {
+
+ new_entry->needs_copy = new_entry_needs_copy;
+
+ /*
+ * Handle copy-on-write obligations
+ */
+
+ if (src_needs_copy && !tmp_entry->needs_copy) {
+ vm_prot_t prot;
+
+ prot = src_entry->protection & ~VM_PROT_WRITE;
+
+ if (override_nx(src_map, VME_ALIAS(src_entry))
+ && prot)
+ prot |= VM_PROT_EXECUTE;
+
+ vm_object_pmap_protect(
+ src_object,
+ src_offset,
+ src_size,
+ (src_entry->is_shared ?
+ PMAP_NULL
+ : src_map->pmap),
+ src_entry->vme_start,
+ prot);
+
+ assert(tmp_entry->wired_count == 0);
+ tmp_entry->needs_copy = TRUE;
+ }
+
+ /*
+ * The map has never been unlocked, so it's safe
+ * to move to the next entry rather than doing
+ * another lookup.
+ */
+
+ goto CopySuccessful;
+ }
+
+ /*
+ * Take an object reference, so that we may
+ * release the map lock(s).
+ */
+
+ assert(src_object != VM_OBJECT_NULL);
+ vm_object_reference(src_object);
+
+ /*
+ * Record the timestamp for later verification.
+ * Unlock the map.
+ */
+
+ version.main_timestamp = src_map->timestamp;
+ vm_map_unlock(src_map); /* Increments timestamp once! */
+
+ /*
+ * Perform the copy
+ */
+
+ if (was_wired) {
+ CopySlowly:
+ vm_object_lock(src_object);
+ result = vm_object_copy_slowly(
+ src_object,
+ src_offset,
+ src_size,
+ THREAD_UNINT,
+ &VME_OBJECT(new_entry));
+ VME_OFFSET_SET(new_entry, 0);
+ new_entry->needs_copy = FALSE;
+
+ }
+ else if (src_object->copy_strategy == MEMORY_OBJECT_COPY_SYMMETRIC &&
+ (tmp_entry->is_shared || map_share)) {
+ vm_object_t new_object;
+
+ vm_object_lock_shared(src_object);
+ new_object = vm_object_copy_delayed(
+ src_object,
+ src_offset,
+ src_size,
+ TRUE);
+ if (new_object == VM_OBJECT_NULL)
+ goto CopySlowly;
+
+ VME_OBJECT_SET(new_entry, new_object);
+ assert(new_entry->wired_count == 0);
+ new_entry->needs_copy = TRUE;
+ assert(!new_entry->iokit_acct);
+ assert(new_object->purgable == VM_PURGABLE_DENY);
+ new_entry->use_pmap = TRUE;
+ result = KERN_SUCCESS;
+
+ } else {
+ vm_object_offset_t new_offset;
+ new_offset = VME_OFFSET(new_entry);
+ result = vm_object_copy_strategically(src_object,
+ src_offset,
+ src_size,
+ &VME_OBJECT(new_entry),
+ &new_offset,
+ &new_entry_needs_copy);
+ if (new_offset != VME_OFFSET(new_entry)) {
+ VME_OFFSET_SET(new_entry, new_offset);
+ }
+
+ new_entry->needs_copy = new_entry_needs_copy;
+ }
+
+ if (result != KERN_SUCCESS &&
+ result != KERN_MEMORY_RESTART_COPY) {
+ vm_map_lock(src_map);
+ RETURN(result);
+ }
+
+ /*
+ * Throw away the extra reference
+ */
+
+ vm_object_deallocate(src_object);
+
+ /*
+ * Verify that the map has not substantially
+ * changed while the copy was being made.
+ */
+
+ vm_map_lock(src_map);
+
+ if ((version.main_timestamp + 1) == src_map->timestamp)
+ goto VerificationSuccessful;
+
+ /*
+ * Simple version comparison failed.
+ *
+ * Retry the lookup and verify that the
+ * same object/offset are still present.
+ *
+ * [Note: a memory manager that colludes with
+ * the calling task can detect that we have
+ * cheated. While the map was unlocked, the
+ * mapping could have been changed and restored.]
+ */
+
+ if (!vm_map_lookup_entry(src_map, src_start, &tmp_entry)) {
+ if (result != KERN_MEMORY_RESTART_COPY) {
+ vm_object_deallocate(VME_OBJECT(new_entry));
+ VME_OBJECT_SET(new_entry, VM_OBJECT_NULL);
+ assert(!new_entry->iokit_acct);
+ new_entry->use_pmap = TRUE;
+ }
+ RETURN(KERN_INVALID_ADDRESS);
+ }
+
+ src_entry = tmp_entry;
+ vm_map_clip_start(src_map, src_entry, src_start);
+
+ if ((((src_entry->protection & VM_PROT_READ) == VM_PROT_NONE) &&
+ !use_maxprot) ||
+ ((src_entry->max_protection & VM_PROT_READ) == 0))
+ goto VerificationFailed;
+
+ if (src_entry->vme_end < new_entry->vme_end) {
+ assert(VM_MAP_PAGE_ALIGNED(src_entry->vme_end,
+ VM_MAP_COPY_PAGE_MASK(copy)));
+ new_entry->vme_end = src_entry->vme_end;
+ src_size = new_entry->vme_end - src_start;
+ }
+
+ if ((VME_OBJECT(src_entry) != src_object) ||
+ (VME_OFFSET(src_entry) != src_offset) ) {
+
+ /*
+ * Verification failed.
+ *
+ * Start over with this top-level entry.
+ */
+
+ VerificationFailed: ;
+
+ vm_object_deallocate(VME_OBJECT(new_entry));
+ tmp_entry = src_entry;
+ continue;
+ }
+
+ /*
+ * Verification succeeded.
+ */
+
+ VerificationSuccessful: ;
+
+ if (result == KERN_MEMORY_RESTART_COPY)
+ goto RestartCopy;
+
+ /*
+ * Copy succeeded.
+ */
+
+ CopySuccessful: ;
+
+ /*
+ * Link in the new copy entry.
+ */
+
+ vm_map_copy_entry_link(copy, vm_map_copy_last_entry(copy),
+ new_entry);
+
+ /*
+ * Determine whether the entire region
+ * has been copied.
+ */
+ src_base = src_start;
+ src_start = new_entry->vme_end;
+ new_entry = VM_MAP_ENTRY_NULL;
+ while ((src_start >= src_end) && (src_end != 0)) {
+ submap_map_t *ptr;
+
+ if (src_map == base_map) {
+ /* back to the top */
+ break;
+ }
+
+ ptr = parent_maps;
+ assert(ptr != NULL);
+ parent_maps = parent_maps->next;
+
+ /* fix up the damage we did in that submap */
+ vm_map_simplify_range(src_map,
+ src_base,
+ src_end);
+
+ vm_map_unlock(src_map);
+ vm_map_deallocate(src_map);
+ vm_map_lock(ptr->parent_map);
+ src_map = ptr->parent_map;
+ src_base = ptr->base_start;
+ src_start = ptr->base_start + ptr->base_len;
+ src_end = ptr->base_end;
+ if (!vm_map_lookup_entry(src_map,
+ src_start,
+ &tmp_entry) &&
+ (src_end > src_start)) {
+ RETURN(KERN_INVALID_ADDRESS);
+ }
+ kfree(ptr, sizeof(submap_map_t));
+ if (parent_maps == NULL)
+ map_share = FALSE;
+ src_entry = tmp_entry->vme_prev;
+ }
+
+ if ((VM_MAP_PAGE_SHIFT(src_map) != PAGE_SHIFT) &&
+ (src_start >= src_addr + len) &&
+ (src_addr + len != 0)) {
+ /*
+ * Stop copying now, even though we haven't reached
+ * "src_end". We'll adjust the end of the last copy
+ * entry at the end, if needed.
+ *
+ * If src_map's aligment is different from the
+ * system's page-alignment, there could be
+ * extra non-map-aligned map entries between
+ * the original (non-rounded) "src_addr + len"
+ * and the rounded "src_end".
+ * We do not want to copy those map entries since
+ * they're not part of the copied range.
+ */
+ break;
+ }
+
+ if ((src_start >= src_end) && (src_end != 0))
+ break;
+
+ /*
+ * Verify that there are no gaps in the region
+ */
+
+ tmp_entry = src_entry->vme_next;
+ if ((tmp_entry->vme_start != src_start) ||
+ (tmp_entry == vm_map_to_entry(src_map))) {
+ RETURN(KERN_INVALID_ADDRESS);
+ }
+ }
+
+ /*
+ * If the source should be destroyed, do it now, since the
+ * copy was successful.
+ */
+ if (src_destroy) {
+ (void) vm_map_delete(
+ src_map,
+ vm_map_trunc_page(src_addr,
+ VM_MAP_PAGE_MASK(src_map)),
+ src_end,
+ ((src_map == kernel_map) ?
+ VM_MAP_REMOVE_KUNWIRE :
+ VM_MAP_NO_FLAGS),
+ VM_MAP_NULL);
+ } else {
+ /* fix up the damage we did in the base map */
+ vm_map_simplify_range(
+ src_map,
+ vm_map_trunc_page(src_addr,
+ VM_MAP_PAGE_MASK(src_map)),
+ vm_map_round_page(src_end,
+ VM_MAP_PAGE_MASK(src_map)));
+ }
+
+ vm_map_unlock(src_map);
+
+ if (VM_MAP_PAGE_SHIFT(src_map) != PAGE_SHIFT) {
+ vm_map_offset_t original_start, original_offset, original_end;
+
+ assert(VM_MAP_COPY_PAGE_MASK(copy) == PAGE_MASK);
+
+ /* adjust alignment of first copy_entry's "vme_start" */
+ tmp_entry = vm_map_copy_first_entry(copy);
+ if (tmp_entry != vm_map_copy_to_entry(copy)) {
+ vm_map_offset_t adjustment;
+
+ original_start = tmp_entry->vme_start;
+ original_offset = VME_OFFSET(tmp_entry);
+
+ /* map-align the start of the first copy entry... */
+ adjustment = (tmp_entry->vme_start -
+ vm_map_trunc_page(
+ tmp_entry->vme_start,
+ VM_MAP_PAGE_MASK(src_map)));
+ tmp_entry->vme_start -= adjustment;
+ VME_OFFSET_SET(tmp_entry,
+ VME_OFFSET(tmp_entry) - adjustment);
+ copy_addr -= adjustment;
+ assert(tmp_entry->vme_start < tmp_entry->vme_end);
+ /* ... adjust for mis-aligned start of copy range */
+ adjustment =
+ (vm_map_trunc_page(copy->offset,
+ PAGE_MASK) -
+ vm_map_trunc_page(copy->offset,
+ VM_MAP_PAGE_MASK(src_map)));
+ if (adjustment) {
+ assert(page_aligned(adjustment));
+ assert(adjustment < VM_MAP_PAGE_SIZE(src_map));
+ tmp_entry->vme_start += adjustment;
+ VME_OFFSET_SET(tmp_entry,
+ (VME_OFFSET(tmp_entry) +
+ adjustment));
+ copy_addr += adjustment;
+ assert(tmp_entry->vme_start < tmp_entry->vme_end);
+ }
+
+ /*
+ * Assert that the adjustments haven't exposed
+ * more than was originally copied...
+ */
+ assert(tmp_entry->vme_start >= original_start);
+ assert(VME_OFFSET(tmp_entry) >= original_offset);
+ /*
+ * ... and that it did not adjust outside of a
+ * a single 16K page.
+ */
+ assert(vm_map_trunc_page(tmp_entry->vme_start,
+ VM_MAP_PAGE_MASK(src_map)) ==
+ vm_map_trunc_page(original_start,
+ VM_MAP_PAGE_MASK(src_map)));
+ }
+
+ /* adjust alignment of last copy_entry's "vme_end" */
+ tmp_entry = vm_map_copy_last_entry(copy);
+ if (tmp_entry != vm_map_copy_to_entry(copy)) {
+ vm_map_offset_t adjustment;
+
+ original_end = tmp_entry->vme_end;
+
+ /* map-align the end of the last copy entry... */
+ tmp_entry->vme_end =
+ vm_map_round_page(tmp_entry->vme_end,
+ VM_MAP_PAGE_MASK(src_map));
+ /* ... adjust for mis-aligned end of copy range */
+ adjustment =
+ (vm_map_round_page((copy->offset +
+ copy->size),
+ VM_MAP_PAGE_MASK(src_map)) -
+ vm_map_round_page((copy->offset +
+ copy->size),
+ PAGE_MASK));
+ if (adjustment) {
+ assert(page_aligned(adjustment));
+ assert(adjustment < VM_MAP_PAGE_SIZE(src_map));
+ tmp_entry->vme_end -= adjustment;
+ assert(tmp_entry->vme_start < tmp_entry->vme_end);
+ }
+
+ /*
+ * Assert that the adjustments haven't exposed
+ * more than was originally copied...
+ */
+ assert(tmp_entry->vme_end <= original_end);
+ /*
+ * ... and that it did not adjust outside of a
+ * a single 16K page.
+ */
+ assert(vm_map_round_page(tmp_entry->vme_end,
+ VM_MAP_PAGE_MASK(src_map)) ==
+ vm_map_round_page(original_end,
+ VM_MAP_PAGE_MASK(src_map)));
+ }
+ }
+
+ /* Fix-up start and end points in copy. This is necessary */
+ /* when the various entries in the copy object were picked */
+ /* up from different sub-maps */
+
+ tmp_entry = vm_map_copy_first_entry(copy);
+ copy_size = 0; /* compute actual size */
+ while (tmp_entry != vm_map_copy_to_entry(copy)) {
+ assert(VM_MAP_PAGE_ALIGNED(
+ copy_addr + (tmp_entry->vme_end -
+ tmp_entry->vme_start),
+ VM_MAP_COPY_PAGE_MASK(copy)));
+ assert(VM_MAP_PAGE_ALIGNED(
+ copy_addr,
+ VM_MAP_COPY_PAGE_MASK(copy)));
+
+ /*
+ * The copy_entries will be injected directly into the
+ * destination map and might not be "map aligned" there...
+ */
+ tmp_entry->map_aligned = FALSE;
+
+ tmp_entry->vme_end = copy_addr +
+ (tmp_entry->vme_end - tmp_entry->vme_start);
+ tmp_entry->vme_start = copy_addr;
+ assert(tmp_entry->vme_start < tmp_entry->vme_end);
+ copy_addr += tmp_entry->vme_end - tmp_entry->vme_start;
+ copy_size += tmp_entry->vme_end - tmp_entry->vme_start;
+ tmp_entry = (struct vm_map_entry *)tmp_entry->vme_next;
+ }
+
+ if (VM_MAP_PAGE_SHIFT(src_map) != PAGE_SHIFT &&
+ copy_size < copy->size) {
+ /*
+ * The actual size of the VM map copy is smaller than what
+ * was requested by the caller. This must be because some
+ * PAGE_SIZE-sized pages are missing at the end of the last
+ * VM_MAP_PAGE_SIZE(src_map)-sized chunk of the range.
+ * The caller might not have been aware of those missing
+ * pages and might not want to be aware of it, which is
+ * fine as long as they don't try to access (and crash on)
+ * those missing pages.
+ * Let's adjust the size of the "copy", to avoid failing
+ * in vm_map_copyout() or vm_map_copy_overwrite().
+ */
+ assert(vm_map_round_page(copy_size,
+ VM_MAP_PAGE_MASK(src_map)) ==
+ vm_map_round_page(copy->size,
+ VM_MAP_PAGE_MASK(src_map)));
+ copy->size = copy_size;
+ }
+
+ *copy_result = copy;
+ return(KERN_SUCCESS);
+
+#undef RETURN
+}
+
+kern_return_t
+vm_map_copy_extract(
+ vm_map_t src_map,
+ vm_map_address_t src_addr,
+ vm_map_size_t len,
+ vm_map_copy_t *copy_result, /* OUT */
+ vm_prot_t *cur_prot, /* OUT */
+ vm_prot_t *max_prot)
+{
+ vm_map_offset_t src_start, src_end;
+ vm_map_copy_t copy;
+ kern_return_t kr;
+
+ /*
+ * Check for copies of zero bytes.
+ */
+
+ if (len == 0) {
+ *copy_result = VM_MAP_COPY_NULL;
+ return(KERN_SUCCESS);
+ }
+
+ /*
+ * Check that the end address doesn't overflow
+ */
+ src_end = src_addr + len;
+ if (src_end < src_addr)
+ return KERN_INVALID_ADDRESS;
+
+ /*
+ * Compute (page aligned) start and end of region
+ */
+ src_start = vm_map_trunc_page(src_addr, PAGE_MASK);
+ src_end = vm_map_round_page(src_end, PAGE_MASK);
+
+ /*
+ * Allocate a header element for the list.
+ *
+ * Use the start and end in the header to
+ * remember the endpoints prior to rounding.
+ */
+
+ copy = (vm_map_copy_t) zalloc(vm_map_copy_zone);
+ copy->c_u.hdr.rb_head_store.rbh_root = (void*)(int)SKIP_RB_TREE;
+ vm_map_copy_first_entry(copy) =
+ vm_map_copy_last_entry(copy) = vm_map_copy_to_entry(copy);
+ copy->type = VM_MAP_COPY_ENTRY_LIST;
+ copy->cpy_hdr.nentries = 0;
+ copy->cpy_hdr.entries_pageable = TRUE;
+
+ vm_map_store_init(©->cpy_hdr);
+
+ copy->offset = 0;
+ copy->size = len;
+
+ kr = vm_map_remap_extract(src_map,
+ src_addr,
+ len,
+ FALSE, /* copy */
+ ©->cpy_hdr,
+ cur_prot,
+ max_prot,
+ VM_INHERIT_SHARE,
+ TRUE); /* pageable */
+ if (kr != KERN_SUCCESS) {
+ vm_map_copy_discard(copy);
+ return kr;
+ }
+
+ *copy_result = copy;
+ return KERN_SUCCESS;
+}
+
+/*
+ * vm_map_copyin_object:
+ *
+ * Create a copy object from an object.
+ * Our caller donates an object reference.
+ */
+
+kern_return_t
+vm_map_copyin_object(
+ vm_object_t object,
+ vm_object_offset_t offset, /* offset of region in object */
+ vm_object_size_t size, /* size of region in object */
+ vm_map_copy_t *copy_result) /* OUT */
+{
+ vm_map_copy_t copy; /* Resulting copy */
+
+ /*
+ * We drop the object into a special copy object
+ * that contains the object directly.
+ */
+
+ copy = (vm_map_copy_t) zalloc(vm_map_copy_zone);
+ copy->c_u.hdr.rb_head_store.rbh_root = (void*)(int)SKIP_RB_TREE;
+ copy->type = VM_MAP_COPY_OBJECT;
+ copy->cpy_object = object;
+ copy->offset = offset;
+ copy->size = size;
+
+ *copy_result = copy;
+ return(KERN_SUCCESS);
+}
+
+static void
+vm_map_fork_share(
+ vm_map_t old_map,
+ vm_map_entry_t old_entry,
+ vm_map_t new_map)
+{
+ vm_object_t object;
+ vm_map_entry_t new_entry;
+
+ /*
+ * New sharing code. New map entry
+ * references original object. Internal
+ * objects use asynchronous copy algorithm for
+ * future copies. First make sure we have
+ * the right object. If we need a shadow,
+ * or someone else already has one, then
+ * make a new shadow and share it.
+ */
+
+ object = VME_OBJECT(old_entry);
+ if (old_entry->is_sub_map) {
+ assert(old_entry->wired_count == 0);
+#ifndef NO_NESTED_PMAP
+ if(old_entry->use_pmap) {
+ kern_return_t result;
+
+ result = pmap_nest(new_map->pmap,
+ (VME_SUBMAP(old_entry))->pmap,
+ (addr64_t)old_entry->vme_start,
+ (addr64_t)old_entry->vme_start,
+ (uint64_t)(old_entry->vme_end - old_entry->vme_start));
+ if(result)
+ panic("vm_map_fork_share: pmap_nest failed!");
+ }
+#endif /* NO_NESTED_PMAP */
+ } else if (object == VM_OBJECT_NULL) {
+ object = vm_object_allocate((vm_map_size_t)(old_entry->vme_end -
+ old_entry->vme_start));
+ VME_OFFSET_SET(old_entry, 0);
+ VME_OBJECT_SET(old_entry, object);
+ old_entry->use_pmap = TRUE;
+ assert(!old_entry->needs_copy);
+ } else if (object->copy_strategy !=
+ MEMORY_OBJECT_COPY_SYMMETRIC) {
+
+ /*
+ * We are already using an asymmetric
+ * copy, and therefore we already have
+ * the right object.
+ */
+
+ assert(! old_entry->needs_copy);
+ }
+ else if (old_entry->needs_copy || /* case 1 */
+ object->shadowed || /* case 2 */
+ (!object->true_share && /* case 3 */
+ !old_entry->is_shared &&
+ (object->vo_size >
+ (vm_map_size_t)(old_entry->vme_end -
+ old_entry->vme_start)))) {
+
+ /*
+ * We need to create a shadow.
+ * There are three cases here.
+ * In the first case, we need to
+ * complete a deferred symmetrical
+ * copy that we participated in.
+ * In the second and third cases,
+ * we need to create the shadow so
+ * that changes that we make to the
+ * object do not interfere with
+ * any symmetrical copies which
+ * have occured (case 2) or which
+ * might occur (case 3).
+ *
+ * The first case is when we had
+ * deferred shadow object creation
+ * via the entry->needs_copy mechanism.
+ * This mechanism only works when
+ * only one entry points to the source
+ * object, and we are about to create
+ * a second entry pointing to the
+ * same object. The problem is that
+ * there is no way of mapping from
+ * an object to the entries pointing
+ * to it. (Deferred shadow creation
+ * works with one entry because occurs
+ * at fault time, and we walk from the
+ * entry to the object when handling
+ * the fault.)
+ *
+ * The second case is when the object
+ * to be shared has already been copied
+ * with a symmetric copy, but we point
+ * directly to the object without
+ * needs_copy set in our entry. (This
+ * can happen because different ranges
+ * of an object can be pointed to by
+ * different entries. In particular,
+ * a single entry pointing to an object
+ * can be split by a call to vm_inherit,
+ * which, combined with task_create, can
+ * result in the different entries
+ * having different needs_copy values.)
+ * The shadowed flag in the object allows
+ * us to detect this case. The problem
+ * with this case is that if this object
+ * has or will have shadows, then we
+ * must not perform an asymmetric copy
+ * of this object, since such a copy
+ * allows the object to be changed, which
+ * will break the previous symmetrical
+ * copies (which rely upon the object
+ * not changing). In a sense, the shadowed
+ * flag says "don't change this object".
+ * We fix this by creating a shadow
+ * object for this object, and sharing
+ * that. This works because we are free
+ * to change the shadow object (and thus
+ * to use an asymmetric copy strategy);
+ * this is also semantically correct,
+ * since this object is temporary, and
+ * therefore a copy of the object is
+ * as good as the object itself. (This
+ * is not true for permanent objects,
+ * since the pager needs to see changes,
+ * which won't happen if the changes
+ * are made to a copy.)
+ *
+ * The third case is when the object
+ * to be shared has parts sticking
+ * outside of the entry we're working
+ * with, and thus may in the future
+ * be subject to a symmetrical copy.
+ * (This is a preemptive version of
+ * case 2.)
+ */
+ VME_OBJECT_SHADOW(old_entry,
+ (vm_map_size_t) (old_entry->vme_end -
+ old_entry->vme_start));
+
+ /*
+ * If we're making a shadow for other than
+ * copy on write reasons, then we have
+ * to remove write permission.
+ */
+
+ if (!old_entry->needs_copy &&
+ (old_entry->protection & VM_PROT_WRITE)) {
+ vm_prot_t prot;
+
+ prot = old_entry->protection & ~VM_PROT_WRITE;
+
+ if (override_nx(old_map, VME_ALIAS(old_entry)) && prot)
+ prot |= VM_PROT_EXECUTE;
+
+ if (old_map->mapped_in_other_pmaps) {
+ vm_object_pmap_protect(
+ VME_OBJECT(old_entry),
+ VME_OFFSET(old_entry),
+ (old_entry->vme_end -
+ old_entry->vme_start),
+ PMAP_NULL,
+ old_entry->vme_start,
+ prot);
+ } else {
+ pmap_protect(old_map->pmap,
+ old_entry->vme_start,
+ old_entry->vme_end,
+ prot);
+ }
+ }
+
+ old_entry->needs_copy = FALSE;
+ object = VME_OBJECT(old_entry);
+ }
+
+
+ /*
+ * If object was using a symmetric copy strategy,
+ * change its copy strategy to the default
+ * asymmetric copy strategy, which is copy_delay
+ * in the non-norma case and copy_call in the
+ * norma case. Bump the reference count for the
+ * new entry.
+ */
+
+ if(old_entry->is_sub_map) {
+ vm_map_lock(VME_SUBMAP(old_entry));
+ vm_map_reference(VME_SUBMAP(old_entry));
+ vm_map_unlock(VME_SUBMAP(old_entry));
+ } else {
+ vm_object_lock(object);
+ vm_object_reference_locked(object);
+ if (object->copy_strategy == MEMORY_OBJECT_COPY_SYMMETRIC) {
+ object->copy_strategy = MEMORY_OBJECT_COPY_DELAY;
+ }
+ vm_object_unlock(object);
+ }
+
+ /*
+ * Clone the entry, using object ref from above.
+ * Mark both entries as shared.
+ */
+
+ new_entry = vm_map_entry_create(new_map, FALSE); /* Never the kernel
+ * map or descendants */
+ vm_map_entry_copy(new_entry, old_entry);
+ old_entry->is_shared = TRUE;
+ new_entry->is_shared = TRUE;
+
+ /*
+ * Insert the entry into the new map -- we
+ * know we're inserting at the end of the new
+ * map.
+ */
+
+ vm_map_store_entry_link(new_map, vm_map_last_entry(new_map), new_entry);
+
+ /*
+ * Update the physical map
+ */
+
+ if (old_entry->is_sub_map) {
+ /* Bill Angell pmap support goes here */
+ } else {
+ pmap_copy(new_map->pmap, old_map->pmap, new_entry->vme_start,
+ old_entry->vme_end - old_entry->vme_start,
+ old_entry->vme_start);
+ }
+}
+
+static boolean_t
+vm_map_fork_copy(
+ vm_map_t old_map,
+ vm_map_entry_t *old_entry_p,
+ vm_map_t new_map)
+{
+ vm_map_entry_t old_entry = *old_entry_p;
+ vm_map_size_t entry_size = old_entry->vme_end - old_entry->vme_start;
+ vm_map_offset_t start = old_entry->vme_start;
+ vm_map_copy_t copy;
+ vm_map_entry_t last = vm_map_last_entry(new_map);
+
+ vm_map_unlock(old_map);
+ /*
+ * Use maxprot version of copyin because we
+ * care about whether this memory can ever
+ * be accessed, not just whether it's accessible
+ * right now.
+ */
+ if (vm_map_copyin_maxprot(old_map, start, entry_size, FALSE, ©)
+ != KERN_SUCCESS) {
+ /*
+ * The map might have changed while it
+ * was unlocked, check it again. Skip
+ * any blank space or permanently
+ * unreadable region.
+ */
+ vm_map_lock(old_map);
+ if (!vm_map_lookup_entry(old_map, start, &last) ||
+ (last->max_protection & VM_PROT_READ) == VM_PROT_NONE) {
+ last = last->vme_next;
+ }
+ *old_entry_p = last;
+
+ /*
+ * XXX For some error returns, want to
+ * XXX skip to the next element. Note
+ * that INVALID_ADDRESS and
+ * PROTECTION_FAILURE are handled above.
+ */
+
+ return FALSE;
+ }
+
+ /*
+ * Insert the copy into the new map
+ */
+
+ vm_map_copy_insert(new_map, last, copy);
+
+ /*
+ * Pick up the traversal at the end of
+ * the copied region.
+ */
+
+ vm_map_lock(old_map);
+ start += entry_size;
+ if (! vm_map_lookup_entry(old_map, start, &last)) {
+ last = last->vme_next;
+ } else {
+ if (last->vme_start == start) {
+ /*
+ * No need to clip here and we don't
+ * want to cause any unnecessary
+ * unnesting...
+ */
+ } else {
+ vm_map_clip_start(old_map, last, start);
+ }
+ }
+ *old_entry_p = last;
+
+ return TRUE;
+}
+
+/*
+ * vm_map_fork:
+ *
+ * Create and return a new map based on the old
+ * map, according to the inheritance values on the
+ * regions in that map.
+ *
+ * The source map must not be locked.
+ */
+vm_map_t
+vm_map_fork(
+ ledger_t ledger,
+ vm_map_t old_map)
+{
+ pmap_t new_pmap;
+ vm_map_t new_map;
+ vm_map_entry_t old_entry;
+ vm_map_size_t new_size = 0, entry_size;
+ vm_map_entry_t new_entry;
+ boolean_t src_needs_copy;
+ boolean_t new_entry_needs_copy;
+ boolean_t pmap_is64bit;
+
+ pmap_is64bit =
+#if defined(__i386__) || defined(__x86_64__)
+ old_map->pmap->pm_task_map != TASK_MAP_32BIT;
+#else
+#error Unknown architecture.
+#endif
+
+ new_pmap = pmap_create(ledger, (vm_map_size_t) 0, pmap_is64bit);
+
+ vm_map_reference_swap(old_map);
+ vm_map_lock(old_map);
+
+ new_map = vm_map_create(new_pmap,
+ old_map->min_offset,
+ old_map->max_offset,
+ old_map->hdr.entries_pageable);
+ /* inherit the parent map's page size */
+ vm_map_set_page_shift(new_map, VM_MAP_PAGE_SHIFT(old_map));
+ for (
+ old_entry = vm_map_first_entry(old_map);
+ old_entry != vm_map_to_entry(old_map);
+ ) {
+
+ entry_size = old_entry->vme_end - old_entry->vme_start;
+
+ switch (old_entry->inheritance) {
+ case VM_INHERIT_NONE:
+ break;
+
+ case VM_INHERIT_SHARE:
+ vm_map_fork_share(old_map, old_entry, new_map);
+ new_size += entry_size;
+ break;
+
+ case VM_INHERIT_COPY:
+
+ /*
+ * Inline the copy_quickly case;
+ * upon failure, fall back on call
+ * to vm_map_fork_copy.
+ */
+
+ if(old_entry->is_sub_map)
+ break;
+ if ((old_entry->wired_count != 0) ||
+ ((VME_OBJECT(old_entry) != NULL) &&
+ (VME_OBJECT(old_entry)->true_share))) {
+ goto slow_vm_map_fork_copy;
+ }
+
+ new_entry = vm_map_entry_create(new_map, FALSE); /* never the kernel map or descendants */
+ vm_map_entry_copy(new_entry, old_entry);
+ if (new_entry->is_sub_map) {
+ /* clear address space specifics */
+ new_entry->use_pmap = FALSE;
+ }
+
+ if (! vm_object_copy_quickly(
+ &VME_OBJECT(new_entry),
+ VME_OFFSET(old_entry),
+ (old_entry->vme_end -
+ old_entry->vme_start),
+ &src_needs_copy,
+ &new_entry_needs_copy)) {
+ vm_map_entry_dispose(new_map, new_entry);
+ goto slow_vm_map_fork_copy;
+ }
+
+ /*
+ * Handle copy-on-write obligations
+ */
+
+ if (src_needs_copy && !old_entry->needs_copy) {
+ vm_prot_t prot;
+
+ prot = old_entry->protection & ~VM_PROT_WRITE;
+
+ if (override_nx(old_map, VME_ALIAS(old_entry))
+ && prot)
+ prot |= VM_PROT_EXECUTE;
+
+ vm_object_pmap_protect(
+ VME_OBJECT(old_entry),
+ VME_OFFSET(old_entry),
+ (old_entry->vme_end -
+ old_entry->vme_start),
+ ((old_entry->is_shared
+ || old_map->mapped_in_other_pmaps)
+ ? PMAP_NULL :
+ old_map->pmap),
+ old_entry->vme_start,
+ prot);
+
+ assert(old_entry->wired_count == 0);
+ old_entry->needs_copy = TRUE;
+ }
+ new_entry->needs_copy = new_entry_needs_copy;
+
+ /*
+ * Insert the entry at the end
+ * of the map.
+ */
+
+ vm_map_store_entry_link(new_map, vm_map_last_entry(new_map),
+ new_entry);
+ new_size += entry_size;
+ break;
+
+ slow_vm_map_fork_copy:
+ if (vm_map_fork_copy(old_map, &old_entry, new_map)) {
+ new_size += entry_size;
+ }
+ continue;
+ }
+ old_entry = old_entry->vme_next;
+ }
+
+
+ new_map->size = new_size;
+ vm_map_unlock(old_map);
+ vm_map_deallocate(old_map);
+
+ return(new_map);
+}
+
+/*
+ * vm_map_exec:
+ *
+ * Setup the "new_map" with the proper execution environment according
+ * to the type of executable (platform, 64bit, chroot environment).
+ * Map the comm page and shared region, etc...
+ */
+kern_return_t
+vm_map_exec(
+ vm_map_t new_map,
+ task_t task,
+ void *fsroot,
+ cpu_type_t cpu)
+{
+ SHARED_REGION_TRACE_DEBUG(
+ ("shared_region: task %p: vm_map_exec(%p,%p,%p,0x%x): ->\n",
+ (void *)VM_KERNEL_ADDRPERM(current_task()),
+ (void *)VM_KERNEL_ADDRPERM(new_map),
+ (void *)VM_KERNEL_ADDRPERM(task),
+ (void *)VM_KERNEL_ADDRPERM(fsroot),
+ cpu));
+ (void) vm_commpage_enter(new_map, task);
+ (void) vm_shared_region_enter(new_map, task, fsroot, cpu);
+ SHARED_REGION_TRACE_DEBUG(
+ ("shared_region: task %p: vm_map_exec(%p,%p,%p,0x%x): <-\n",
+ (void *)VM_KERNEL_ADDRPERM(current_task()),
+ (void *)VM_KERNEL_ADDRPERM(new_map),
+ (void *)VM_KERNEL_ADDRPERM(task),
+ (void *)VM_KERNEL_ADDRPERM(fsroot),
+ cpu));
+ return KERN_SUCCESS;
+}
+
+/*
+ * vm_map_lookup_locked:
+ *
+ * Finds the VM object, offset, and
+ * protection for a given virtual address in the
+ * specified map, assuming a page fault of the
+ * type specified.
+ *
+ * Returns the (object, offset, protection) for
+ * this address, whether it is wired down, and whether
+ * this map has the only reference to the data in question.
+ * In order to later verify this lookup, a "version"
+ * is returned.
+ *
+ * The map MUST be locked by the caller and WILL be
+ * locked on exit. In order to guarantee the
+ * existence of the returned object, it is returned
+ * locked.
+ *
+ * If a lookup is requested with "write protection"
+ * specified, the map may be changed to perform virtual
+ * copying operations, although the data referenced will
+ * remain the same.
+ */
+kern_return_t
+vm_map_lookup_locked(
+ vm_map_t *var_map, /* IN/OUT */
+ vm_map_offset_t vaddr,
+ vm_prot_t fault_type,
+ int object_lock_type,
+ vm_map_version_t *out_version, /* OUT */
+ vm_object_t *object, /* OUT */
+ vm_object_offset_t *offset, /* OUT */
+ vm_prot_t *out_prot, /* OUT */
+ boolean_t *wired, /* OUT */
+ vm_object_fault_info_t fault_info, /* OUT */
+ vm_map_t *real_map)
+{
+ vm_map_entry_t entry;
+ register vm_map_t map = *var_map;
+ vm_map_t old_map = *var_map;
+ vm_map_t cow_sub_map_parent = VM_MAP_NULL;
+ vm_map_offset_t cow_parent_vaddr = 0;
+ vm_map_offset_t old_start = 0;
+ vm_map_offset_t old_end = 0;
+ register vm_prot_t prot;
+ boolean_t mask_protections;
+ boolean_t force_copy;
+ vm_prot_t original_fault_type;
+
+ /*
+ * VM_PROT_MASK means that the caller wants us to use "fault_type"
+ * as a mask against the mapping's actual protections, not as an
+ * absolute value.
+ */
+ mask_protections = (fault_type & VM_PROT_IS_MASK) ? TRUE : FALSE;
+ force_copy = (fault_type & VM_PROT_COPY) ? TRUE : FALSE;
+ fault_type &= VM_PROT_ALL;
+ original_fault_type = fault_type;
+
+ *real_map = map;
+
+RetryLookup:
+ fault_type = original_fault_type;
+
+ /*
+ * If the map has an interesting hint, try it before calling
+ * full blown lookup routine.
+ */
+ entry = map->hint;
+
+ if ((entry == vm_map_to_entry(map)) ||
+ (vaddr < entry->vme_start) || (vaddr >= entry->vme_end)) {
+ vm_map_entry_t tmp_entry;
+
+ /*
+ * Entry was either not a valid hint, or the vaddr
+ * was not contained in the entry, so do a full lookup.
+ */
+ if (!vm_map_lookup_entry(map, vaddr, &tmp_entry)) {
+ if((cow_sub_map_parent) && (cow_sub_map_parent != map))
+ vm_map_unlock(cow_sub_map_parent);
+ if((*real_map != map)
+ && (*real_map != cow_sub_map_parent))
+ vm_map_unlock(*real_map);
+ return KERN_INVALID_ADDRESS;
+ }
+
+ entry = tmp_entry;
+ }
+ if(map == old_map) {
+ old_start = entry->vme_start;
+ old_end = entry->vme_end;
+ }
+
+ /*
+ * Handle submaps. Drop lock on upper map, submap is
+ * returned locked.
+ */
+
+submap_recurse:
+ if (entry->is_sub_map) {
+ vm_map_offset_t local_vaddr;
+ vm_map_offset_t end_delta;
+ vm_map_offset_t start_delta;
+ vm_map_entry_t submap_entry;
+ boolean_t mapped_needs_copy=FALSE;
+
+ local_vaddr = vaddr;
+
+ if ((entry->use_pmap && !(fault_type & VM_PROT_WRITE))) {
+ /* if real_map equals map we unlock below */
+ if ((*real_map != map) &&
+ (*real_map != cow_sub_map_parent))
+ vm_map_unlock(*real_map);
+ *real_map = VME_SUBMAP(entry);
+ }
+
+ if(entry->needs_copy && (fault_type & VM_PROT_WRITE)) {
+ if (!mapped_needs_copy) {
+ if (vm_map_lock_read_to_write(map)) {
+ vm_map_lock_read(map);
+ *real_map = map;
+ goto RetryLookup;
+ }
+ vm_map_lock_read(VME_SUBMAP(entry));
+ *var_map = VME_SUBMAP(entry);
+ cow_sub_map_parent = map;
+ /* reset base to map before cow object */
+ /* this is the map which will accept */
+ /* the new cow object */
+ old_start = entry->vme_start;
+ old_end = entry->vme_end;
+ cow_parent_vaddr = vaddr;
+ mapped_needs_copy = TRUE;
+ } else {
+ vm_map_lock_read(VME_SUBMAP(entry));
+ *var_map = VME_SUBMAP(entry);
+ if((cow_sub_map_parent != map) &&
+ (*real_map != map))
+ vm_map_unlock(map);
+ }
+ } else {
+ vm_map_lock_read(VME_SUBMAP(entry));
+ *var_map = VME_SUBMAP(entry);
+ /* leave map locked if it is a target */
+ /* cow sub_map above otherwise, just */
+ /* follow the maps down to the object */
+ /* here we unlock knowing we are not */
+ /* revisiting the map. */
+ if((*real_map != map) && (map != cow_sub_map_parent))
+ vm_map_unlock_read(map);
+ }
+
+ map = *var_map;
+
+ /* calculate the offset in the submap for vaddr */
+ local_vaddr = (local_vaddr - entry->vme_start) + VME_OFFSET(entry);
+
+ RetrySubMap:
+ if(!vm_map_lookup_entry(map, local_vaddr, &submap_entry)) {
+ if((cow_sub_map_parent) && (cow_sub_map_parent != map)){
+ vm_map_unlock(cow_sub_map_parent);
+ }
+ if((*real_map != map)
+ && (*real_map != cow_sub_map_parent)) {
+ vm_map_unlock(*real_map);
+ }
+ *real_map = map;
+ return KERN_INVALID_ADDRESS;
+ }
+
+ /* find the attenuated shadow of the underlying object */
+ /* on our target map */
+
+ /* in english the submap object may extend beyond the */
+ /* region mapped by the entry or, may only fill a portion */
+ /* of it. For our purposes, we only care if the object */
+ /* doesn't fill. In this case the area which will */
+ /* ultimately be clipped in the top map will only need */
+ /* to be as big as the portion of the underlying entry */
+ /* which is mapped */
+ start_delta = submap_entry->vme_start > VME_OFFSET(entry) ?
+ submap_entry->vme_start - VME_OFFSET(entry) : 0;
+
+ end_delta =
+ (VME_OFFSET(entry) + start_delta + (old_end - old_start)) <=
+ submap_entry->vme_end ?
+ 0 : (VME_OFFSET(entry) +
+ (old_end - old_start))
+ - submap_entry->vme_end;
+
+ old_start += start_delta;
+ old_end -= end_delta;
+
+ if(submap_entry->is_sub_map) {
+ entry = submap_entry;
+ vaddr = local_vaddr;
+ goto submap_recurse;
+ }
+
+ if(((fault_type & VM_PROT_WRITE) && cow_sub_map_parent)) {
+
+ vm_object_t sub_object, copy_object;
+ vm_object_offset_t copy_offset;
+ vm_map_offset_t local_start;
+ vm_map_offset_t local_end;
+ boolean_t copied_slowly = FALSE;
+
+ if (vm_map_lock_read_to_write(map)) {
+ vm_map_lock_read(map);
+ old_start -= start_delta;
+ old_end += end_delta;
+ goto RetrySubMap;
+ }
+
+
+ sub_object = VME_OBJECT(submap_entry);
+ if (sub_object == VM_OBJECT_NULL) {
+ sub_object =
+ vm_object_allocate(
+ (vm_map_size_t)
+ (submap_entry->vme_end -
+ submap_entry->vme_start));
+ VME_OBJECT_SET(submap_entry, sub_object);
+ VME_OFFSET_SET(submap_entry, 0);
+ }
+ local_start = local_vaddr -
+ (cow_parent_vaddr - old_start);
+ local_end = local_vaddr +
+ (old_end - cow_parent_vaddr);
+ vm_map_clip_start(map, submap_entry, local_start);
+ vm_map_clip_end(map, submap_entry, local_end);
+ if (submap_entry->is_sub_map) {
+ /* unnesting was done when clipping */
+ assert(!submap_entry->use_pmap);
+ }
+
+ /* This is the COW case, lets connect */
+ /* an entry in our space to the underlying */
+ /* object in the submap, bypassing the */
+ /* submap. */
+
+
+ if(submap_entry->wired_count != 0 ||
+ (sub_object->copy_strategy ==
+ MEMORY_OBJECT_COPY_NONE)) {
+ vm_object_lock(sub_object);
+ vm_object_copy_slowly(sub_object,
+ VME_OFFSET(submap_entry),
+ (submap_entry->vme_end -
+ submap_entry->vme_start),
+ FALSE,
+ ©_object);
+ copied_slowly = TRUE;
+ } else {
+
+ /* set up shadow object */
+ copy_object = sub_object;
+ vm_object_reference(copy_object);
+ sub_object->shadowed = TRUE;
+ assert(submap_entry->wired_count == 0);
+ submap_entry->needs_copy = TRUE;
+
+ prot = submap_entry->protection & ~VM_PROT_WRITE;
+
+ if (override_nx(old_map,
+ VME_ALIAS(submap_entry))
+ && prot)
+ prot |= VM_PROT_EXECUTE;
+
+ vm_object_pmap_protect(
+ sub_object,
+ VME_OFFSET(submap_entry),
+ submap_entry->vme_end -
+ submap_entry->vme_start,
+ (submap_entry->is_shared
+ || map->mapped_in_other_pmaps) ?
+ PMAP_NULL : map->pmap,
+ submap_entry->vme_start,
+ prot);
+ }
+
+ /*
+ * Adjust the fault offset to the submap entry.
+ */
+ copy_offset = (local_vaddr -
+ submap_entry->vme_start +
+ VME_OFFSET(submap_entry));
+
+ /* This works diffently than the */
+ /* normal submap case. We go back */
+ /* to the parent of the cow map and*/
+ /* clip out the target portion of */
+ /* the sub_map, substituting the */
+ /* new copy object, */
+
+ vm_map_unlock(map);
+ local_start = old_start;
+ local_end = old_end;
+ map = cow_sub_map_parent;
+ *var_map = cow_sub_map_parent;
+ vaddr = cow_parent_vaddr;
+ cow_sub_map_parent = NULL;
+
+ if(!vm_map_lookup_entry(map,
+ vaddr, &entry)) {
+ vm_object_deallocate(
+ copy_object);
+ vm_map_lock_write_to_read(map);
+ return KERN_INVALID_ADDRESS;
+ }
+
+ /* clip out the portion of space */
+ /* mapped by the sub map which */
+ /* corresponds to the underlying */
+ /* object */
+
+ /*
+ * Clip (and unnest) the smallest nested chunk
+ * possible around the faulting address...
+ */
+ local_start = vaddr & ~(pmap_nesting_size_min - 1);
+ local_end = local_start + pmap_nesting_size_min;
+ /*
+ * ... but don't go beyond the "old_start" to "old_end"
+ * range, to avoid spanning over another VM region
+ * with a possibly different VM object and/or offset.
+ */
+ if (local_start < old_start) {
+ local_start = old_start;
+ }
+ if (local_end > old_end) {
+ local_end = old_end;
+ }
+ /*
+ * Adjust copy_offset to the start of the range.
+ */
+ copy_offset -= (vaddr - local_start);
+
+ vm_map_clip_start(map, entry, local_start);
+ vm_map_clip_end(map, entry, local_end);
+ if (entry->is_sub_map) {
+ /* unnesting was done when clipping */
+ assert(!entry->use_pmap);
+ }
+
+ /* substitute copy object for */
+ /* shared map entry */
+ vm_map_deallocate(VME_SUBMAP(entry));
+ assert(!entry->iokit_acct);
+ entry->is_sub_map = FALSE;
+ entry->use_pmap = TRUE;
+ VME_OBJECT_SET(entry, copy_object);
+
+ /* propagate the submap entry's protections */
+ entry->protection |= submap_entry->protection;
+ entry->max_protection |= submap_entry->max_protection;
+
+ if(copied_slowly) {
+ VME_OFFSET_SET(entry, local_start - old_start);
+ entry->needs_copy = FALSE;
+ entry->is_shared = FALSE;
+ } else {
+ VME_OFFSET_SET(entry, copy_offset);
+ assert(entry->wired_count == 0);
+ entry->needs_copy = TRUE;
+ if(entry->inheritance == VM_INHERIT_SHARE)
+ entry->inheritance = VM_INHERIT_COPY;
+ if (map != old_map)
+ entry->is_shared = TRUE;
+ }
+ if(entry->inheritance == VM_INHERIT_SHARE)
+ entry->inheritance = VM_INHERIT_COPY;
+
+ vm_map_lock_write_to_read(map);
+ } else {
+ if((cow_sub_map_parent)
+ && (cow_sub_map_parent != *real_map)
+ && (cow_sub_map_parent != map)) {
+ vm_map_unlock(cow_sub_map_parent);
+ }
+ entry = submap_entry;
+ vaddr = local_vaddr;
+ }
+ }
+
+ /*
+ * Check whether this task is allowed to have
+ * this page.
+ */
+
+ prot = entry->protection;
+
+ if (override_nx(old_map, VME_ALIAS(entry)) && prot) {
+ /*
+ * HACK -- if not a stack, then allow execution
+ */
+ prot |= VM_PROT_EXECUTE;
+ }
+
+ if (mask_protections) {
+ fault_type &= prot;
+ if (fault_type == VM_PROT_NONE) {
+ goto protection_failure;
+ }
+ }
+ if ((fault_type & (prot)) != fault_type) {
+ protection_failure:
+ if (*real_map != map) {
+ vm_map_unlock(*real_map);
+ }
+ *real_map = map;
+
+ if ((fault_type & VM_PROT_EXECUTE) && prot)
+ log_stack_execution_failure((addr64_t)vaddr, prot);
+
+ DTRACE_VM2(prot_fault, int, 1, (uint64_t *), NULL);
+ return KERN_PROTECTION_FAILURE;
+ }
+
+ /*
+ * If this page is not pageable, we have to get
+ * it for all possible accesses.
+ */
+
+ *wired = (entry->wired_count != 0);
+ if (*wired)
+ fault_type = prot;
+
+ /*
+ * If the entry was copy-on-write, we either ...
+ */
+
+ if (entry->needs_copy) {
+ /*
+ * If we want to write the page, we may as well
+ * handle that now since we've got the map locked.
+ *
+ * If we don't need to write the page, we just
+ * demote the permissions allowed.
+ */
+
+ if ((fault_type & VM_PROT_WRITE) || *wired || force_copy) {
+ /*
+ * Make a new object, and place it in the
+ * object chain. Note that no new references
+ * have appeared -- one just moved from the
+ * map to the new object.
+ */
+
+ if (vm_map_lock_read_to_write(map)) {
+ vm_map_lock_read(map);
+ goto RetryLookup;
+ }
+ VME_OBJECT_SHADOW(entry,
+ (vm_map_size_t) (entry->vme_end -
+ entry->vme_start));
+
+ VME_OBJECT(entry)->shadowed = TRUE;
+ entry->needs_copy = FALSE;
+ vm_map_lock_write_to_read(map);
+ }
+ else {
+ /*
+ * We're attempting to read a copy-on-write
+ * page -- don't allow writes.
+ */
+
+ prot &= (~VM_PROT_WRITE);
+ }
+ }
+
+ /*
+ * Create an object if necessary.
+ */
+ if (VME_OBJECT(entry) == VM_OBJECT_NULL) {
+
+ if (vm_map_lock_read_to_write(map)) {
+ vm_map_lock_read(map);
+ goto RetryLookup;
+ }
+
+ VME_OBJECT_SET(entry,
+ vm_object_allocate(
+ (vm_map_size_t)(entry->vme_end -
+ entry->vme_start)));
+ VME_OFFSET_SET(entry, 0);
+ vm_map_lock_write_to_read(map);
+ }
+
+ /*
+ * Return the object/offset from this entry. If the entry
+ * was copy-on-write or empty, it has been fixed up. Also
+ * return the protection.
+ */
+
+ *offset = (vaddr - entry->vme_start) + VME_OFFSET(entry);
+ *object = VME_OBJECT(entry);
+ *out_prot = prot;
+
+ if (fault_info) {
+ fault_info->interruptible = THREAD_UNINT; /* for now... */
+ /* ... the caller will change "interruptible" if needed */
+ fault_info->cluster_size = 0;
+ fault_info->user_tag = VME_ALIAS(entry);
+ fault_info->pmap_options = 0;
+ if (entry->iokit_acct ||
+ (!entry->is_sub_map && !entry->use_pmap)) {
+ fault_info->pmap_options |= PMAP_OPTIONS_ALT_ACCT;
+ }
+ fault_info->behavior = entry->behavior;
+ fault_info->lo_offset = VME_OFFSET(entry);
+ fault_info->hi_offset =
+ (entry->vme_end - entry->vme_start) + VME_OFFSET(entry);
+ fault_info->no_cache = entry->no_cache;
+ fault_info->stealth = FALSE;
+ fault_info->io_sync = FALSE;
+ if (entry->used_for_jit ||
+ entry->vme_resilient_codesign) {
+ fault_info->cs_bypass = TRUE;
+ } else {
+ fault_info->cs_bypass = FALSE;
+ }
+ fault_info->mark_zf_absent = FALSE;
+ fault_info->batch_pmap_op = FALSE;
+ }
+
+ /*
+ * Lock the object to prevent it from disappearing
+ */
+ if (object_lock_type == OBJECT_LOCK_EXCLUSIVE)
+ vm_object_lock(*object);
+ else
+ vm_object_lock_shared(*object);
+
+ /*
+ * Save the version number
+ */
+
+ out_version->main_timestamp = map->timestamp;
+
+ return KERN_SUCCESS;
+}
+
+
+/*
+ * vm_map_verify:
+ *
+ * Verifies that the map in question has not changed
+ * since the given version. If successful, the map
+ * will not change until vm_map_verify_done() is called.
+ */
+boolean_t
+vm_map_verify(
+ register vm_map_t map,
+ register vm_map_version_t *version) /* REF */
+{
+ boolean_t result;
+
+ vm_map_lock_read(map);
+ result = (map->timestamp == version->main_timestamp);
+
+ if (!result)
+ vm_map_unlock_read(map);
+
+ return(result);
+}
+
+/*
+ * vm_map_verify_done:
+ *
+ * Releases locks acquired by a vm_map_verify.
+ *
+ * This is now a macro in vm/vm_map.h. It does a
+ * vm_map_unlock_read on the map.
+ */
+
+
+/*
+ * TEMPORARYTEMPORARYTEMPORARYTEMPORARYTEMPORARYTEMPORARY
+ * Goes away after regular vm_region_recurse function migrates to
+ * 64 bits
+ * vm_region_recurse: A form of vm_region which follows the
+ * submaps in a target map
+ *
+ */
+
+kern_return_t
+vm_map_region_recurse_64(
+ vm_map_t map,
+ vm_map_offset_t *address, /* IN/OUT */
+ vm_map_size_t *size, /* OUT */
+ natural_t *nesting_depth, /* IN/OUT */
+ vm_region_submap_info_64_t submap_info, /* IN/OUT */
+ mach_msg_type_number_t *count) /* IN/OUT */
+{
+ mach_msg_type_number_t original_count;
+ vm_region_extended_info_data_t extended;
+ vm_map_entry_t tmp_entry;
+ vm_map_offset_t user_address;
+ unsigned int user_max_depth;
+
+ /*
+ * "curr_entry" is the VM map entry preceding or including the
+ * address we're looking for.
+ * "curr_map" is the map or sub-map containing "curr_entry".
+ * "curr_address" is the equivalent of the top map's "user_address"
+ * in the current map.
+ * "curr_offset" is the cumulated offset of "curr_map" in the
+ * target task's address space.
+ * "curr_depth" is the depth of "curr_map" in the chain of
+ * sub-maps.
+ *
+ * "curr_max_below" and "curr_max_above" limit the range (around
+ * "curr_address") we should take into account in the current (sub)map.
+ * They limit the range to what's visible through the map entries
+ * we've traversed from the top map to the current map.
+
+ */
+ vm_map_entry_t curr_entry;
+ vm_map_address_t curr_address;
+ vm_map_offset_t curr_offset;
+ vm_map_t curr_map;
+ unsigned int curr_depth;
+ vm_map_offset_t curr_max_below, curr_max_above;
+ vm_map_offset_t curr_skip;
+
+ /*
+ * "next_" is the same as "curr_" but for the VM region immediately
+ * after the address we're looking for. We need to keep track of this
+ * too because we want to return info about that region if the
+ * address we're looking for is not mapped.
+ */
+ vm_map_entry_t next_entry;
+ vm_map_offset_t next_offset;
+ vm_map_offset_t next_address;
+ vm_map_t next_map;
+ unsigned int next_depth;
+ vm_map_offset_t next_max_below, next_max_above;
+ vm_map_offset_t next_skip;
+
+ boolean_t look_for_pages;
+ vm_region_submap_short_info_64_t short_info;
+
+ if (map == VM_MAP_NULL) {
+ /* no address space to work on */
+ return KERN_INVALID_ARGUMENT;
+ }
+
+
+ if (*count < VM_REGION_SUBMAP_SHORT_INFO_COUNT_64) {
+ /*
+ * "info" structure is not big enough and
+ * would overflow
+ */
+ return KERN_INVALID_ARGUMENT;
+ }
+
+ original_count = *count;
+
+ if (original_count < VM_REGION_SUBMAP_INFO_V0_COUNT_64) {
+ *count = VM_REGION_SUBMAP_SHORT_INFO_COUNT_64;
+ look_for_pages = FALSE;
+ short_info = (vm_region_submap_short_info_64_t) submap_info;
+ submap_info = NULL;
+ } else {
+ look_for_pages = TRUE;
+ *count = VM_REGION_SUBMAP_INFO_V0_COUNT_64;
+ short_info = NULL;
+
+ if (original_count >= VM_REGION_SUBMAP_INFO_V1_COUNT_64) {
+ *count = VM_REGION_SUBMAP_INFO_V1_COUNT_64;
+ }
+ }
+
+ user_address = *address;
+ user_max_depth = *nesting_depth;
+
+ if (not_in_kdp) {
+ vm_map_lock_read(map);
+ }
+
+recurse_again:
+ curr_entry = NULL;
+ curr_map = map;
+ curr_address = user_address;
+ curr_offset = 0;
+ curr_skip = 0;
+ curr_depth = 0;
+ curr_max_above = ((vm_map_offset_t) -1) - curr_address;
+ curr_max_below = curr_address;
+
+ next_entry = NULL;
+ next_map = NULL;
+ next_address = 0;
+ next_offset = 0;
+ next_skip = 0;
+ next_depth = 0;
+ next_max_above = (vm_map_offset_t) -1;
+ next_max_below = (vm_map_offset_t) -1;
+
+ for (;;) {
+ if (vm_map_lookup_entry(curr_map,
+ curr_address,
+ &tmp_entry)) {
+ /* tmp_entry contains the address we're looking for */
+ curr_entry = tmp_entry;
+ } else {
+ vm_map_offset_t skip;
+ /*
+ * The address is not mapped. "tmp_entry" is the
+ * map entry preceding the address. We want the next
+ * one, if it exists.
+ */
+ curr_entry = tmp_entry->vme_next;
+
+ if (curr_entry == vm_map_to_entry(curr_map) ||
+ (curr_entry->vme_start >=
+ curr_address + curr_max_above)) {
+ /* no next entry at this level: stop looking */
+ if (not_in_kdp) {
+ vm_map_unlock_read(curr_map);
+ }
+ curr_entry = NULL;
+ curr_map = NULL;
+ curr_skip = 0;
+ curr_offset = 0;
+ curr_depth = 0;
+ curr_max_above = 0;
+ curr_max_below = 0;
+ break;
+ }
+
+ /* adjust current address and offset */
+ skip = curr_entry->vme_start - curr_address;
+ curr_address = curr_entry->vme_start;
+ curr_skip += skip;
+ curr_offset += skip;
+ curr_max_above -= skip;
+ curr_max_below = 0;
+ }
+
+ /*
+ * Is the next entry at this level closer to the address (or
+ * deeper in the submap chain) than the one we had
+ * so far ?
+ */
+ tmp_entry = curr_entry->vme_next;
+ if (tmp_entry == vm_map_to_entry(curr_map)) {
+ /* no next entry at this level */
+ } else if (tmp_entry->vme_start >=
+ curr_address + curr_max_above) {
+ /*
+ * tmp_entry is beyond the scope of what we mapped of
+ * this submap in the upper level: ignore it.
+ */
+ } else if ((next_entry == NULL) ||
+ (tmp_entry->vme_start + curr_offset <=
+ next_entry->vme_start + next_offset)) {
+ /*
+ * We didn't have a "next_entry" or this one is
+ * closer to the address we're looking for:
+ * use this "tmp_entry" as the new "next_entry".
+ */
+ if (next_entry != NULL) {
+ /* unlock the last "next_map" */
+ if (next_map != curr_map && not_in_kdp) {
+ vm_map_unlock_read(next_map);
+ }
+ }
+ next_entry = tmp_entry;
+ next_map = curr_map;
+ next_depth = curr_depth;
+ next_address = next_entry->vme_start;
+ next_skip = curr_skip;
+ next_skip += (next_address - curr_address);
+ next_offset = curr_offset;
+ next_offset += (next_address - curr_address);
+ next_max_above = MIN(next_max_above, curr_max_above);
+ next_max_above = MIN(next_max_above,
+ next_entry->vme_end - next_address);
+ next_max_below = MIN(next_max_below, curr_max_below);
+ next_max_below = MIN(next_max_below,
+ next_address - next_entry->vme_start);
+ }
+
+ /*
+ * "curr_max_{above,below}" allow us to keep track of the
+ * portion of the submap that is actually mapped at this level:
+ * the rest of that submap is irrelevant to us, since it's not
+ * mapped here.
+ * The relevant portion of the map starts at
+ * "VME_OFFSET(curr_entry)" up to the size of "curr_entry".
+ */
+ curr_max_above = MIN(curr_max_above,
+ curr_entry->vme_end - curr_address);
+ curr_max_below = MIN(curr_max_below,
+ curr_address - curr_entry->vme_start);
+
+ if (!curr_entry->is_sub_map ||
+ curr_depth >= user_max_depth) {
+ /*
+ * We hit a leaf map or we reached the maximum depth
+ * we could, so stop looking. Keep the current map
+ * locked.
+ */
+ break;
+ }
+
+ /*
+ * Get down to the next submap level.
+ */
+
+ /*
+ * Lock the next level and unlock the current level,
+ * unless we need to keep it locked to access the "next_entry"
+ * later.
+ */
+ if (not_in_kdp) {
+ vm_map_lock_read(VME_SUBMAP(curr_entry));
+ }
+ if (curr_map == next_map) {
+ /* keep "next_map" locked in case we need it */
+ } else {
+ /* release this map */
+ if (not_in_kdp)
+ vm_map_unlock_read(curr_map);
+ }
+
+ /*
+ * Adjust the offset. "curr_entry" maps the submap
+ * at relative address "curr_entry->vme_start" in the
+ * curr_map but skips the first "VME_OFFSET(curr_entry)"
+ * bytes of the submap.
+ * "curr_offset" always represents the offset of a virtual
+ * address in the curr_map relative to the absolute address
+ * space (i.e. the top-level VM map).
+ */
+ curr_offset +=
+ (VME_OFFSET(curr_entry) - curr_entry->vme_start);
+ curr_address = user_address + curr_offset;
+ /* switch to the submap */
+ curr_map = VME_SUBMAP(curr_entry);
+ curr_depth++;
+ curr_entry = NULL;
+ }
+
+ if (curr_entry == NULL) {
+ /* no VM region contains the address... */
+ if (next_entry == NULL) {
+ /* ... and no VM region follows it either */
+ return KERN_INVALID_ADDRESS;
+ }
+ /* ... gather info about the next VM region */
+ curr_entry = next_entry;
+ curr_map = next_map; /* still locked ... */
+ curr_address = next_address;
+ curr_skip = next_skip;
+ curr_offset = next_offset;
+ curr_depth = next_depth;
+ curr_max_above = next_max_above;
+ curr_max_below = next_max_below;
+ } else {
+ /* we won't need "next_entry" after all */
+ if (next_entry != NULL) {
+ /* release "next_map" */
+ if (next_map != curr_map && not_in_kdp) {
+ vm_map_unlock_read(next_map);
+ }
+ }
+ }
+ next_entry = NULL;
+ next_map = NULL;
+ next_offset = 0;
+ next_skip = 0;
+ next_depth = 0;
+ next_max_below = -1;
+ next_max_above = -1;
+
+ if (curr_entry->is_sub_map &&
+ curr_depth < user_max_depth) {
+ /*
+ * We're not as deep as we could be: we must have
+ * gone back up after not finding anything mapped
+ * below the original top-level map entry's.
+ * Let's move "curr_address" forward and recurse again.
+ */
+ user_address = curr_address;
+ goto recurse_again;
+ }
+
+ *nesting_depth = curr_depth;
+ *size = curr_max_above + curr_max_below;
+ *address = user_address + curr_skip - curr_max_below;
+
+// LP64todo: all the current tools are 32bit, obviously never worked for 64b
+// so probably should be a real 32b ID vs. ptr.
+// Current users just check for equality
+#define INFO_MAKE_OBJECT_ID(p) ((uint32_t)(uintptr_t)VM_KERNEL_ADDRPERM(p))
+
+ if (look_for_pages) {
+ submap_info->user_tag = VME_ALIAS(curr_entry);
+ submap_info->offset = VME_OFFSET(curr_entry);
+ submap_info->protection = curr_entry->protection;
+ submap_info->inheritance = curr_entry->inheritance;
+ submap_info->max_protection = curr_entry->max_protection;
+ submap_info->behavior = curr_entry->behavior;
+ submap_info->user_wired_count = curr_entry->user_wired_count;
+ submap_info->is_submap = curr_entry->is_sub_map;
+ submap_info->object_id = INFO_MAKE_OBJECT_ID(VME_OBJECT(curr_entry));
+ } else {
+ short_info->user_tag = VME_ALIAS(curr_entry);
+ short_info->offset = VME_OFFSET(curr_entry);
+ short_info->protection = curr_entry->protection;
+ short_info->inheritance = curr_entry->inheritance;
+ short_info->max_protection = curr_entry->max_protection;
+ short_info->behavior = curr_entry->behavior;
+ short_info->user_wired_count = curr_entry->user_wired_count;
+ short_info->is_submap = curr_entry->is_sub_map;
+ short_info->object_id = INFO_MAKE_OBJECT_ID(VME_OBJECT(curr_entry));
+ }
+
+ extended.pages_resident = 0;
+ extended.pages_swapped_out = 0;
+ extended.pages_shared_now_private = 0;
+ extended.pages_dirtied = 0;
+ extended.pages_reusable = 0;
+ extended.external_pager = 0;
+ extended.shadow_depth = 0;
+ extended.share_mode = SM_EMPTY;
+ extended.ref_count = 0;
+
+ if (not_in_kdp) {
+ if (!curr_entry->is_sub_map) {
+ vm_map_offset_t range_start, range_end;
+ range_start = MAX((curr_address - curr_max_below),
+ curr_entry->vme_start);
+ range_end = MIN((curr_address + curr_max_above),
+ curr_entry->vme_end);
+ vm_map_region_walk(curr_map,
+ range_start,
+ curr_entry,
+ (VME_OFFSET(curr_entry) +
+ (range_start -
+ curr_entry->vme_start)),
+ range_end - range_start,
+ &extended,
+ look_for_pages, VM_REGION_EXTENDED_INFO_COUNT);
+ if (extended.external_pager &&
+ extended.ref_count == 2 &&
+ extended.share_mode == SM_SHARED) {
+ extended.share_mode = SM_PRIVATE;
+ }
+ } else {
+ if (curr_entry->use_pmap) {
+ extended.share_mode = SM_TRUESHARED;
+ } else {
+ extended.share_mode = SM_PRIVATE;
+ }
+ extended.ref_count = VME_SUBMAP(curr_entry)->ref_count;
+ }
+ }
+
+ if (look_for_pages) {
+ submap_info->pages_resident = extended.pages_resident;
+ submap_info->pages_swapped_out = extended.pages_swapped_out;
+ submap_info->pages_shared_now_private =
+ extended.pages_shared_now_private;
+ submap_info->pages_dirtied = extended.pages_dirtied;
+ submap_info->external_pager = extended.external_pager;
+ submap_info->shadow_depth = extended.shadow_depth;
+ submap_info->share_mode = extended.share_mode;
+ submap_info->ref_count = extended.ref_count;
+
+ if (original_count >= VM_REGION_SUBMAP_INFO_V1_COUNT_64) {
+ submap_info->pages_reusable = extended.pages_reusable;
+ }
+ } else {
+ short_info->external_pager = extended.external_pager;
+ short_info->shadow_depth = extended.shadow_depth;
+ short_info->share_mode = extended.share_mode;
+ short_info->ref_count = extended.ref_count;
+ }
+
+ if (not_in_kdp) {
+ vm_map_unlock_read(curr_map);
+ }
+
+ return KERN_SUCCESS;
+}
+
+/*
+ * vm_region:
+ *
+ * User call to obtain information about a region in
+ * a task's address map. Currently, only one flavor is
+ * supported.
+ *
+ * XXX The reserved and behavior fields cannot be filled
+ * in until the vm merge from the IK is completed, and
+ * vm_reserve is implemented.
+ */
+
+kern_return_t
+vm_map_region(
+ vm_map_t map,
+ vm_map_offset_t *address, /* IN/OUT */
+ vm_map_size_t *size, /* OUT */
+ vm_region_flavor_t flavor, /* IN */
+ vm_region_info_t info, /* OUT */
+ mach_msg_type_number_t *count, /* IN/OUT */
+ mach_port_t *object_name) /* OUT */
+{
+ vm_map_entry_t tmp_entry;
+ vm_map_entry_t entry;
+ vm_map_offset_t start;
+
+ if (map == VM_MAP_NULL)
+ return(KERN_INVALID_ARGUMENT);
+
+ switch (flavor) {
+
+ case VM_REGION_BASIC_INFO:
+ /* legacy for old 32-bit objects info */
+ {
+ vm_region_basic_info_t basic;
+
+ if (*count < VM_REGION_BASIC_INFO_COUNT)
+ return(KERN_INVALID_ARGUMENT);
+
+ basic = (vm_region_basic_info_t) info;
+ *count = VM_REGION_BASIC_INFO_COUNT;
+
+ vm_map_lock_read(map);
+
+ start = *address;
+ if (!vm_map_lookup_entry(map, start, &tmp_entry)) {
+ if ((entry = tmp_entry->vme_next) == vm_map_to_entry(map)) {
+ vm_map_unlock_read(map);
+ return(KERN_INVALID_ADDRESS);
+ }
+ } else {
+ entry = tmp_entry;
+ }
+
+ start = entry->vme_start;
+
+ basic->offset = (uint32_t)VME_OFFSET(entry);
+ basic->protection = entry->protection;
+ basic->inheritance = entry->inheritance;
+ basic->max_protection = entry->max_protection;
+ basic->behavior = entry->behavior;
+ basic->user_wired_count = entry->user_wired_count;
+ basic->reserved = entry->is_sub_map;
+ *address = start;
+ *size = (entry->vme_end - start);
+
+ if (object_name) *object_name = IP_NULL;
+ if (entry->is_sub_map) {
+ basic->shared = FALSE;
+ } else {
+ basic->shared = entry->is_shared;
+ }
+
+ vm_map_unlock_read(map);
+ return(KERN_SUCCESS);
+ }
+
+ case VM_REGION_BASIC_INFO_64:
+ {
+ vm_region_basic_info_64_t basic;
+
+ if (*count < VM_REGION_BASIC_INFO_COUNT_64)
+ return(KERN_INVALID_ARGUMENT);
+
+ basic = (vm_region_basic_info_64_t) info;
+ *count = VM_REGION_BASIC_INFO_COUNT_64;
+
+ vm_map_lock_read(map);
+
+ start = *address;
+ if (!vm_map_lookup_entry(map, start, &tmp_entry)) {
+ if ((entry = tmp_entry->vme_next) == vm_map_to_entry(map)) {
+ vm_map_unlock_read(map);
+ return(KERN_INVALID_ADDRESS);
+ }
+ } else {
+ entry = tmp_entry;
+ }
+
+ start = entry->vme_start;
+
+ basic->offset = VME_OFFSET(entry);
+ basic->protection = entry->protection;
+ basic->inheritance = entry->inheritance;
+ basic->max_protection = entry->max_protection;
+ basic->behavior = entry->behavior;
+ basic->user_wired_count = entry->user_wired_count;
+ basic->reserved = entry->is_sub_map;
+ *address = start;
+ *size = (entry->vme_end - start);
+
+ if (object_name) *object_name = IP_NULL;
+ if (entry->is_sub_map) {
+ basic->shared = FALSE;
+ } else {
+ basic->shared = entry->is_shared;
+ }
+
+ vm_map_unlock_read(map);
+ return(KERN_SUCCESS);
+ }
+ case VM_REGION_EXTENDED_INFO:
+ if (*count < VM_REGION_EXTENDED_INFO_COUNT)
+ return(KERN_INVALID_ARGUMENT);
+ /*fallthru*/
+ case VM_REGION_EXTENDED_INFO__legacy:
+ if (*count < VM_REGION_EXTENDED_INFO_COUNT__legacy)
+ return KERN_INVALID_ARGUMENT;
+
+ {
+ vm_region_extended_info_t extended;
+ mach_msg_type_number_t original_count;
+
+ extended = (vm_region_extended_info_t) info;
+
+ vm_map_lock_read(map);
+
+ start = *address;
+ if (!vm_map_lookup_entry(map, start, &tmp_entry)) {
+ if ((entry = tmp_entry->vme_next) == vm_map_to_entry(map)) {
+ vm_map_unlock_read(map);
+ return(KERN_INVALID_ADDRESS);
+ }
+ } else {
+ entry = tmp_entry;
+ }
+ start = entry->vme_start;
+
+ extended->protection = entry->protection;
+ extended->user_tag = VME_ALIAS(entry);
+ extended->pages_resident = 0;
+ extended->pages_swapped_out = 0;
+ extended->pages_shared_now_private = 0;
+ extended->pages_dirtied = 0;
+ extended->external_pager = 0;
+ extended->shadow_depth = 0;
+
+ original_count = *count;
+ if (flavor == VM_REGION_EXTENDED_INFO__legacy) {
+ *count = VM_REGION_EXTENDED_INFO_COUNT__legacy;
+ } else {
+ extended->pages_reusable = 0;
+ *count = VM_REGION_EXTENDED_INFO_COUNT;
+ }
+
+ vm_map_region_walk(map, start, entry, VME_OFFSET(entry), entry->vme_end - start, extended, TRUE, *count);
+
+ if (extended->external_pager && extended->ref_count == 2 && extended->share_mode == SM_SHARED)
+ extended->share_mode = SM_PRIVATE;
+
+ if (object_name)
+ *object_name = IP_NULL;
+ *address = start;
+ *size = (entry->vme_end - start);
+
+ vm_map_unlock_read(map);
+ return(KERN_SUCCESS);
+ }
+ case VM_REGION_TOP_INFO:
+ {
+ vm_region_top_info_t top;
+
+ if (*count < VM_REGION_TOP_INFO_COUNT)
+ return(KERN_INVALID_ARGUMENT);
+
+ top = (vm_region_top_info_t) info;
+ *count = VM_REGION_TOP_INFO_COUNT;
+
+ vm_map_lock_read(map);
+
+ start = *address;
+ if (!vm_map_lookup_entry(map, start, &tmp_entry)) {
+ if ((entry = tmp_entry->vme_next) == vm_map_to_entry(map)) {
+ vm_map_unlock_read(map);
+ return(KERN_INVALID_ADDRESS);
+ }
+ } else {
+ entry = tmp_entry;
+
+ }
+ start = entry->vme_start;
+
+ top->private_pages_resident = 0;
+ top->shared_pages_resident = 0;
+
+ vm_map_region_top_walk(entry, top);
+
+ if (object_name)
+ *object_name = IP_NULL;
+ *address = start;
+ *size = (entry->vme_end - start);
+
+ vm_map_unlock_read(map);
+ return(KERN_SUCCESS);
+ }
+ default:
+ return(KERN_INVALID_ARGUMENT);
+ }
+}
+
+#define OBJ_RESIDENT_COUNT(obj, entry_size) \
+ MIN((entry_size), \
+ ((obj)->all_reusable ? \
+ (obj)->wired_page_count : \
+ (obj)->resident_page_count - (obj)->reusable_page_count))
+
+void
+vm_map_region_top_walk(
+ vm_map_entry_t entry,
+ vm_region_top_info_t top)
+{
+
+ if (VME_OBJECT(entry) == 0 || entry->is_sub_map) {
+ top->share_mode = SM_EMPTY;
+ top->ref_count = 0;
+ top->obj_id = 0;
+ return;
+ }
+
+ {
+ struct vm_object *obj, *tmp_obj;
+ int ref_count;
+ uint32_t entry_size;
+
+ entry_size = (uint32_t) ((entry->vme_end - entry->vme_start) / PAGE_SIZE_64);
+
+ obj = VME_OBJECT(entry);
+
+ vm_object_lock(obj);
+
+ if ((ref_count = obj->ref_count) > 1 && obj->paging_in_progress)
+ ref_count--;
+
+ assert(obj->reusable_page_count <= obj->resident_page_count);
+ if (obj->shadow) {
+ if (ref_count == 1)
+ top->private_pages_resident =
+ OBJ_RESIDENT_COUNT(obj, entry_size);
+ else
+ top->shared_pages_resident =
+ OBJ_RESIDENT_COUNT(obj, entry_size);
+ top->ref_count = ref_count;
+ top->share_mode = SM_COW;
+
+ while ((tmp_obj = obj->shadow)) {
+ vm_object_lock(tmp_obj);
+ vm_object_unlock(obj);
+ obj = tmp_obj;
+
+ if ((ref_count = obj->ref_count) > 1 && obj->paging_in_progress)
+ ref_count--;
+
+ assert(obj->reusable_page_count <= obj->resident_page_count);
+ top->shared_pages_resident +=
+ OBJ_RESIDENT_COUNT(obj, entry_size);
+ top->ref_count += ref_count - 1;
+ }
+ } else {
+ if (entry->superpage_size) {
+ top->share_mode = SM_LARGE_PAGE;
+ top->shared_pages_resident = 0;
+ top->private_pages_resident = entry_size;
+ } else if (entry->needs_copy) {
+ top->share_mode = SM_COW;
+ top->shared_pages_resident =
+ OBJ_RESIDENT_COUNT(obj, entry_size);
+ } else {
+ if (ref_count == 1 ||
+ (ref_count == 2 && !(obj->pager_trusted) && !(obj->internal))) {
+ top->share_mode = SM_PRIVATE;
+ top->private_pages_resident =
+ OBJ_RESIDENT_COUNT(obj,
+ entry_size);
+ } else {
+ top->share_mode = SM_SHARED;
+ top->shared_pages_resident =
+ OBJ_RESIDENT_COUNT(obj,
+ entry_size);
+ }
+ }
+ top->ref_count = ref_count;
+ }
+ /* XXX K64: obj_id will be truncated */
+ top->obj_id = (unsigned int) (uintptr_t)VM_KERNEL_ADDRPERM(obj);
+
+ vm_object_unlock(obj);
+ }
+}
+
+void
+vm_map_region_walk(
+ vm_map_t map,
+ vm_map_offset_t va,
+ vm_map_entry_t entry,
+ vm_object_offset_t offset,
+ vm_object_size_t range,
+ vm_region_extended_info_t extended,
+ boolean_t look_for_pages,
+ mach_msg_type_number_t count)
+{
+ register struct vm_object *obj, *tmp_obj;
+ register vm_map_offset_t last_offset;
+ register int i;
+ register int ref_count;
+ struct vm_object *shadow_object;
+ int shadow_depth;
+
+ if ((VME_OBJECT(entry) == 0) ||
+ (entry->is_sub_map) ||
+ (VME_OBJECT(entry)->phys_contiguous &&
+ !entry->superpage_size)) {
+ extended->share_mode = SM_EMPTY;
+ extended->ref_count = 0;
+ return;
+ }
+
+ if (entry->superpage_size) {
+ extended->shadow_depth = 0;
+ extended->share_mode = SM_LARGE_PAGE;
+ extended->ref_count = 1;
+ extended->external_pager = 0;
+ extended->pages_resident = (unsigned int)(range >> PAGE_SHIFT);
+ extended->shadow_depth = 0;
+ return;
+ }
+
+ {
+ obj = VME_OBJECT(entry);
+
+ vm_object_lock(obj);
+
+ if ((ref_count = obj->ref_count) > 1 && obj->paging_in_progress)
+ ref_count--;
+
+ if (look_for_pages) {
+ for (last_offset = offset + range;
+ offset < last_offset;
+ offset += PAGE_SIZE_64, va += PAGE_SIZE) {
+ vm_map_region_look_for_page(map, va, obj,
+ offset, ref_count,
+ 0, extended, count);
+ }
+ } else {
+ shadow_object = obj->shadow;
+ shadow_depth = 0;
+
+ if ( !(obj->pager_trusted) && !(obj->internal))
+ extended->external_pager = 1;
+
+ if (shadow_object != VM_OBJECT_NULL) {
+ vm_object_lock(shadow_object);
+ for (;
+ shadow_object != VM_OBJECT_NULL;
+ shadow_depth++) {
+ vm_object_t next_shadow;
+
+ if ( !(shadow_object->pager_trusted) &&
+ !(shadow_object->internal))
+ extended->external_pager = 1;
+
+ next_shadow = shadow_object->shadow;
+ if (next_shadow) {
+ vm_object_lock(next_shadow);
+ }
+ vm_object_unlock(shadow_object);
+ shadow_object = next_shadow;
+ }
+ }
+ extended->shadow_depth = shadow_depth;
+ }
+
+ if (extended->shadow_depth || entry->needs_copy)
+ extended->share_mode = SM_COW;
+ else {
+ if (ref_count == 1)
+ extended->share_mode = SM_PRIVATE;
+ else {
+ if (obj->true_share)
+ extended->share_mode = SM_TRUESHARED;
+ else
+ extended->share_mode = SM_SHARED;
+ }
+ }
+ extended->ref_count = ref_count - extended->shadow_depth;
+
+ for (i = 0; i < extended->shadow_depth; i++) {
+ if ((tmp_obj = obj->shadow) == 0)
+ break;
+ vm_object_lock(tmp_obj);
+ vm_object_unlock(obj);
+
+ if ((ref_count = tmp_obj->ref_count) > 1 && tmp_obj->paging_in_progress)
+ ref_count--;
+
+ extended->ref_count += ref_count;
+ obj = tmp_obj;
+ }
+ vm_object_unlock(obj);
+
+ if (extended->share_mode == SM_SHARED) {
+ register vm_map_entry_t cur;
+ register vm_map_entry_t last;
+ int my_refs;
+
+ obj = VME_OBJECT(entry);
+ last = vm_map_to_entry(map);
+ my_refs = 0;
+
+ if ((ref_count = obj->ref_count) > 1 && obj->paging_in_progress)
+ ref_count--;
+ for (cur = vm_map_first_entry(map); cur != last; cur = cur->vme_next)
+ my_refs += vm_map_region_count_obj_refs(cur, obj);
+
+ if (my_refs == ref_count)
+ extended->share_mode = SM_PRIVATE_ALIASED;
+ else if (my_refs > 1)
+ extended->share_mode = SM_SHARED_ALIASED;
+ }
+ }
+}
+
+
+/* object is locked on entry and locked on return */
+
+
+static void
+vm_map_region_look_for_page(
+ __unused vm_map_t map,
+ __unused vm_map_offset_t va,
+ vm_object_t object,
+ vm_object_offset_t offset,
+ int max_refcnt,
+ int depth,
+ vm_region_extended_info_t extended,
+ mach_msg_type_number_t count)
+{
+ register vm_page_t p;
+ register vm_object_t shadow;
+ register int ref_count;
+ vm_object_t caller_object;
+ kern_return_t kr;
+ shadow = object->shadow;
+ caller_object = object;
+
+
+ while (TRUE) {
+
+ if ( !(object->pager_trusted) && !(object->internal))
+ extended->external_pager = 1;
+
+ if ((p = vm_page_lookup(object, offset)) != VM_PAGE_NULL) {
+ if (shadow && (max_refcnt == 1))
+ extended->pages_shared_now_private++;
+
+ if (!p->fictitious &&
+ (p->dirty || pmap_is_modified(p->phys_page)))
+ extended->pages_dirtied++;
+ else if (count >= VM_REGION_EXTENDED_INFO_COUNT) {
+ if (p->reusable || p->object->all_reusable) {
+ extended->pages_reusable++;
+ }
+ }
+
+ extended->pages_resident++;
+
+ if(object != caller_object)
+ vm_object_unlock(object);
+
+ return;
+ }
+#if MACH_PAGEMAP
+ if (object->existence_map) {
+ if (vm_external_state_get(object->existence_map, offset) == VM_EXTERNAL_STATE_EXISTS) {
+
+ extended->pages_swapped_out++;
+
+ if(object != caller_object)
+ vm_object_unlock(object);
+
+ return;
+ }
+ } else
+#endif /* MACH_PAGEMAP */
+ if (object->internal &&
+ object->alive &&
+ !object->terminating &&
+ object->pager_ready) {
+
+ if (COMPRESSED_PAGER_IS_ACTIVE || DEFAULT_FREEZER_COMPRESSED_PAGER_IS_ACTIVE) {
+ if (VM_COMPRESSOR_PAGER_STATE_GET(object,
+ offset)
+ == VM_EXTERNAL_STATE_EXISTS) {
+ /* the pager has that page */
+ extended->pages_swapped_out++;
+ if (object != caller_object)
+ vm_object_unlock(object);
+ return;
+ }
+ } else {
+ memory_object_t pager;
+
+ vm_object_paging_begin(object);
+ pager = object->pager;
+ vm_object_unlock(object);
+
+ kr = memory_object_data_request(
+ pager,
+ offset + object->paging_offset,
+ 0, /* just poke the pager */
+ VM_PROT_READ,
+ NULL);
+
+ vm_object_lock(object);
+ vm_object_paging_end(object);
+
+ if (kr == KERN_SUCCESS) {
+ /* the pager has that page */
+ extended->pages_swapped_out++;
+ if (object != caller_object)
+ vm_object_unlock(object);
+ return;
+ }
+ }
+ }
+
+ if (shadow) {
+ vm_object_lock(shadow);
+
+ if ((ref_count = shadow->ref_count) > 1 && shadow->paging_in_progress)
+ ref_count--;
+
+ if (++depth > extended->shadow_depth)
+ extended->shadow_depth = depth;
+
+ if (ref_count > max_refcnt)
+ max_refcnt = ref_count;
+
+ if(object != caller_object)
+ vm_object_unlock(object);
+
+ offset = offset + object->vo_shadow_offset;
+ object = shadow;
+ shadow = object->shadow;
+ continue;
+ }
+ if(object != caller_object)
+ vm_object_unlock(object);
+ break;
+ }
+}
- if (!old_entry->needs_copy &&
- (old_entry->protection & VM_PROT_WRITE)) {
- if(old_map->mapped) {
- vm_object_pmap_protect(
- old_entry->object.vm_object,
- old_entry->offset,
- (old_entry->vme_end -
- old_entry->vme_start),
- PMAP_NULL,
- old_entry->vme_start,
- old_entry->protection & ~VM_PROT_WRITE);
- } else {
- pmap_protect(old_map->pmap,
- old_entry->vme_start,
- old_entry->vme_end,
- old_entry->protection & ~VM_PROT_WRITE);
- }
+static int
+vm_map_region_count_obj_refs(
+ vm_map_entry_t entry,
+ vm_object_t object)
+{
+ register int ref_count;
+ register vm_object_t chk_obj;
+ register vm_object_t tmp_obj;
+
+ if (VME_OBJECT(entry) == 0)
+ return(0);
+
+ if (entry->is_sub_map)
+ return(0);
+ else {
+ ref_count = 0;
+
+ chk_obj = VME_OBJECT(entry);
+ vm_object_lock(chk_obj);
+
+ while (chk_obj) {
+ if (chk_obj == object)
+ ref_count++;
+ tmp_obj = chk_obj->shadow;
+ if (tmp_obj)
+ vm_object_lock(tmp_obj);
+ vm_object_unlock(chk_obj);
+
+ chk_obj = tmp_obj;
}
-
- old_entry->needs_copy = FALSE;
- object = old_entry->object.vm_object;
}
-
- /*
- * If object was using a symmetric copy strategy,
- * change its copy strategy to the default
- * asymmetric copy strategy, which is copy_delay
- * in the non-norma case and copy_call in the
- * norma case. Bump the reference count for the
- * new entry.
- */
-
- if(old_entry->is_sub_map) {
- vm_map_lock(old_entry->object.sub_map);
- vm_map_reference(old_entry->object.sub_map);
- vm_map_unlock(old_entry->object.sub_map);
- } else {
- vm_object_lock(object);
- object->ref_count++;
- vm_object_res_reference(object);
- if (object->copy_strategy == MEMORY_OBJECT_COPY_SYMMETRIC) {
- object->copy_strategy = MEMORY_OBJECT_COPY_DELAY;
+ return(ref_count);
+}
+
+
+/*
+ * Routine: vm_map_simplify
+ *
+ * Description:
+ * Attempt to simplify the map representation in
+ * the vicinity of the given starting address.
+ * Note:
+ * This routine is intended primarily to keep the
+ * kernel maps more compact -- they generally don't
+ * benefit from the "expand a map entry" technology
+ * at allocation time because the adjacent entry
+ * is often wired down.
+ */
+void
+vm_map_simplify_entry(
+ vm_map_t map,
+ vm_map_entry_t this_entry)
+{
+ vm_map_entry_t prev_entry;
+
+ counter(c_vm_map_simplify_entry_called++);
+
+ prev_entry = this_entry->vme_prev;
+
+ if ((this_entry != vm_map_to_entry(map)) &&
+ (prev_entry != vm_map_to_entry(map)) &&
+
+ (prev_entry->vme_end == this_entry->vme_start) &&
+
+ (prev_entry->is_sub_map == this_entry->is_sub_map) &&
+ (VME_OBJECT(prev_entry) == VME_OBJECT(this_entry)) &&
+ ((VME_OFFSET(prev_entry) + (prev_entry->vme_end -
+ prev_entry->vme_start))
+ == VME_OFFSET(this_entry)) &&
+
+ (prev_entry->behavior == this_entry->behavior) &&
+ (prev_entry->needs_copy == this_entry->needs_copy) &&
+ (prev_entry->protection == this_entry->protection) &&
+ (prev_entry->max_protection == this_entry->max_protection) &&
+ (prev_entry->inheritance == this_entry->inheritance) &&
+ (prev_entry->use_pmap == this_entry->use_pmap) &&
+ (VME_ALIAS(prev_entry) == VME_ALIAS(this_entry)) &&
+ (prev_entry->no_cache == this_entry->no_cache) &&
+ (prev_entry->permanent == this_entry->permanent) &&
+ (prev_entry->map_aligned == this_entry->map_aligned) &&
+ (prev_entry->zero_wired_pages == this_entry->zero_wired_pages) &&
+ (prev_entry->used_for_jit == this_entry->used_for_jit) &&
+ /* from_reserved_zone: OK if that field doesn't match */
+ (prev_entry->iokit_acct == this_entry->iokit_acct) &&
+ (prev_entry->vme_resilient_codesign ==
+ this_entry->vme_resilient_codesign) &&
+ (prev_entry->vme_resilient_media ==
+ this_entry->vme_resilient_media) &&
+
+ (prev_entry->wired_count == this_entry->wired_count) &&
+ (prev_entry->user_wired_count == this_entry->user_wired_count) &&
+
+ (prev_entry->in_transition == FALSE) &&
+ (this_entry->in_transition == FALSE) &&
+ (prev_entry->needs_wakeup == FALSE) &&
+ (this_entry->needs_wakeup == FALSE) &&
+ (prev_entry->is_shared == FALSE) &&
+ (this_entry->is_shared == FALSE) &&
+ (prev_entry->superpage_size == FALSE) &&
+ (this_entry->superpage_size == FALSE)
+ ) {
+ vm_map_store_entry_unlink(map, prev_entry);
+ assert(prev_entry->vme_start < this_entry->vme_end);
+ if (prev_entry->map_aligned)
+ assert(VM_MAP_PAGE_ALIGNED(prev_entry->vme_start,
+ VM_MAP_PAGE_MASK(map)));
+ this_entry->vme_start = prev_entry->vme_start;
+ VME_OFFSET_SET(this_entry, VME_OFFSET(prev_entry));
+
+ if (map->holelistenabled) {
+ vm_map_store_update_first_free(map, this_entry, TRUE);
}
- vm_object_unlock(object);
+
+ if (prev_entry->is_sub_map) {
+ vm_map_deallocate(VME_SUBMAP(prev_entry));
+ } else {
+ vm_object_deallocate(VME_OBJECT(prev_entry));
+ }
+ vm_map_entry_dispose(map, prev_entry);
+ SAVE_HINT_MAP_WRITE(map, this_entry);
+ counter(c_vm_map_simplified++);
}
-
- /*
- * Clone the entry, using object ref from above.
- * Mark both entries as shared.
- */
-
- new_entry = vm_map_entry_create(new_map);
- vm_map_entry_copy(new_entry, old_entry);
- old_entry->is_shared = TRUE;
- new_entry->is_shared = TRUE;
-
- /*
- * Insert the entry into the new map -- we
- * know we're inserting at the end of the new
- * map.
- */
-
- vm_map_entry_link(new_map, vm_map_last_entry(new_map), new_entry);
-
- /*
- * Update the physical map
- */
-
- if (old_entry->is_sub_map) {
- /* Bill Angell pmap support goes here */
- } else {
- pmap_copy(new_map->pmap, old_map->pmap, new_entry->vme_start,
- old_entry->vme_end - old_entry->vme_start,
- old_entry->vme_start);
+}
+
+void
+vm_map_simplify(
+ vm_map_t map,
+ vm_map_offset_t start)
+{
+ vm_map_entry_t this_entry;
+
+ vm_map_lock(map);
+ if (vm_map_lookup_entry(map, start, &this_entry)) {
+ vm_map_simplify_entry(map, this_entry);
+ vm_map_simplify_entry(map, this_entry->vme_next);
}
+ counter(c_vm_map_simplify_called++);
+ vm_map_unlock(map);
}
-boolean_t
-vm_map_fork_copy(
- vm_map_t old_map,
- vm_map_entry_t *old_entry_p,
- vm_map_t new_map)
+static void
+vm_map_simplify_range(
+ vm_map_t map,
+ vm_map_offset_t start,
+ vm_map_offset_t end)
{
- vm_map_entry_t old_entry = *old_entry_p;
- vm_size_t entry_size = old_entry->vme_end - old_entry->vme_start;
- vm_offset_t start = old_entry->vme_start;
- vm_map_copy_t copy;
- vm_map_entry_t last = vm_map_last_entry(new_map);
+ vm_map_entry_t entry;
- vm_map_unlock(old_map);
/*
- * Use maxprot version of copyin because we
- * care about whether this memory can ever
- * be accessed, not just whether it's accessible
- * right now.
+ * The map should be locked (for "write") by the caller.
*/
- if (vm_map_copyin_maxprot(old_map, start, entry_size, FALSE, ©)
- != KERN_SUCCESS) {
- /*
- * The map might have changed while it
- * was unlocked, check it again. Skip
- * any blank space or permanently
- * unreadable region.
- */
- vm_map_lock(old_map);
- if (!vm_map_lookup_entry(old_map, start, &last) ||
- (last->max_protection & VM_PROT_READ) == VM_PROT_NONE) {
- last = last->vme_next;
+
+ if (start >= end) {
+ /* invalid address range */
+ return;
+ }
+
+ start = vm_map_trunc_page(start,
+ VM_MAP_PAGE_MASK(map));
+ end = vm_map_round_page(end,
+ VM_MAP_PAGE_MASK(map));
+
+ if (!vm_map_lookup_entry(map, start, &entry)) {
+ /* "start" is not mapped and "entry" ends before "start" */
+ if (entry == vm_map_to_entry(map)) {
+ /* start with first entry in the map */
+ entry = vm_map_first_entry(map);
+ } else {
+ /* start with next entry */
+ entry = entry->vme_next;
+ }
+ }
+
+ while (entry != vm_map_to_entry(map) &&
+ entry->vme_start <= end) {
+ /* try and coalesce "entry" with its previous entry */
+ vm_map_simplify_entry(map, entry);
+ entry = entry->vme_next;
+ }
+}
+
+
+/*
+ * Routine: vm_map_machine_attribute
+ * Purpose:
+ * Provide machine-specific attributes to mappings,
+ * such as cachability etc. for machines that provide
+ * them. NUMA architectures and machines with big/strange
+ * caches will use this.
+ * Note:
+ * Responsibilities for locking and checking are handled here,
+ * everything else in the pmap module. If any non-volatile
+ * information must be kept, the pmap module should handle
+ * it itself. [This assumes that attributes do not
+ * need to be inherited, which seems ok to me]
+ */
+kern_return_t
+vm_map_machine_attribute(
+ vm_map_t map,
+ vm_map_offset_t start,
+ vm_map_offset_t end,
+ vm_machine_attribute_t attribute,
+ vm_machine_attribute_val_t* value) /* IN/OUT */
+{
+ kern_return_t ret;
+ vm_map_size_t sync_size;
+ vm_map_entry_t entry;
+
+ if (start < vm_map_min(map) || end > vm_map_max(map))
+ return KERN_INVALID_ADDRESS;
+
+ /* Figure how much memory we need to flush (in page increments) */
+ sync_size = end - start;
+
+ vm_map_lock(map);
+
+ if (attribute != MATTR_CACHE) {
+ /* If we don't have to find physical addresses, we */
+ /* don't have to do an explicit traversal here. */
+ ret = pmap_attribute(map->pmap, start, end-start,
+ attribute, value);
+ vm_map_unlock(map);
+ return ret;
+ }
+
+ ret = KERN_SUCCESS; /* Assume it all worked */
+
+ while(sync_size) {
+ if (vm_map_lookup_entry(map, start, &entry)) {
+ vm_map_size_t sub_size;
+ if((entry->vme_end - start) > sync_size) {
+ sub_size = sync_size;
+ sync_size = 0;
+ } else {
+ sub_size = entry->vme_end - start;
+ sync_size -= sub_size;
+ }
+ if(entry->is_sub_map) {
+ vm_map_offset_t sub_start;
+ vm_map_offset_t sub_end;
+
+ sub_start = (start - entry->vme_start)
+ + VME_OFFSET(entry);
+ sub_end = sub_start + sub_size;
+ vm_map_machine_attribute(
+ VME_SUBMAP(entry),
+ sub_start,
+ sub_end,
+ attribute, value);
+ } else {
+ if (VME_OBJECT(entry)) {
+ vm_page_t m;
+ vm_object_t object;
+ vm_object_t base_object;
+ vm_object_t last_object;
+ vm_object_offset_t offset;
+ vm_object_offset_t base_offset;
+ vm_map_size_t range;
+ range = sub_size;
+ offset = (start - entry->vme_start)
+ + VME_OFFSET(entry);
+ base_offset = offset;
+ object = VME_OBJECT(entry);
+ base_object = object;
+ last_object = NULL;
+
+ vm_object_lock(object);
+
+ while (range) {
+ m = vm_page_lookup(
+ object, offset);
+
+ if (m && !m->fictitious) {
+ ret =
+ pmap_attribute_cache_sync(
+ m->phys_page,
+ PAGE_SIZE,
+ attribute, value);
+
+ } else if (object->shadow) {
+ offset = offset + object->vo_shadow_offset;
+ last_object = object;
+ object = object->shadow;
+ vm_object_lock(last_object->shadow);
+ vm_object_unlock(last_object);
+ continue;
+ }
+ range -= PAGE_SIZE;
+
+ if (base_object != object) {
+ vm_object_unlock(object);
+ vm_object_lock(base_object);
+ object = base_object;
+ }
+ /* Bump to the next page */
+ base_offset += PAGE_SIZE;
+ offset = base_offset;
+ }
+ vm_object_unlock(object);
+ }
+ }
+ start += sub_size;
+ } else {
+ vm_map_unlock(map);
+ return KERN_FAILURE;
}
- *old_entry_p = last;
-
- /*
- * XXX For some error returns, want to
- * XXX skip to the next element. Note
- * that INVALID_ADDRESS and
- * PROTECTION_FAILURE are handled above.
- */
- return FALSE;
- }
-
- /*
- * Insert the copy into the new map
- */
-
- vm_map_copy_insert(new_map, last, copy);
-
- /*
- * Pick up the traversal at the end of
- * the copied region.
- */
-
- vm_map_lock(old_map);
- start += entry_size;
- if (! vm_map_lookup_entry(old_map, start, &last)) {
- last = last->vme_next;
- } else {
- vm_map_clip_start(old_map, last, start);
}
- *old_entry_p = last;
- return TRUE;
+ vm_map_unlock(map);
+
+ return ret;
}
/*
- * vm_map_fork:
- *
- * Create and return a new map based on the old
- * map, according to the inheritance values on the
- * regions in that map.
+ * vm_map_behavior_set:
*
- * The source map must not be locked.
+ * Sets the paging reference behavior of the specified address
+ * range in the target map. Paging reference behavior affects
+ * how pagein operations resulting from faults on the map will be
+ * clustered.
*/
-vm_map_t
-vm_map_fork(
- vm_map_t old_map)
+kern_return_t
+vm_map_behavior_set(
+ vm_map_t map,
+ vm_map_offset_t start,
+ vm_map_offset_t end,
+ vm_behavior_t new_behavior)
{
- pmap_t new_pmap = pmap_create((vm_size_t) 0);
- vm_map_t new_map;
- vm_map_entry_t old_entry;
- vm_size_t new_size = 0, entry_size;
- vm_map_entry_t new_entry;
- boolean_t src_needs_copy;
- boolean_t new_entry_needs_copy;
+ register vm_map_entry_t entry;
+ vm_map_entry_t temp_entry;
- vm_map_reference_swap(old_map);
- vm_map_lock(old_map);
+ XPR(XPR_VM_MAP,
+ "vm_map_behavior_set, 0x%X start 0x%X end 0x%X behavior %d",
+ map, start, end, new_behavior, 0);
- new_map = vm_map_create(new_pmap,
- old_map->min_offset,
- old_map->max_offset,
- old_map->hdr.entries_pageable);
+ if (start > end ||
+ start < vm_map_min(map) ||
+ end > vm_map_max(map)) {
+ return KERN_NO_SPACE;
+ }
- for (
- old_entry = vm_map_first_entry(old_map);
- old_entry != vm_map_to_entry(old_map);
- ) {
+ switch (new_behavior) {
- entry_size = old_entry->vme_end - old_entry->vme_start;
+ /*
+ * This first block of behaviors all set a persistent state on the specified
+ * memory range. All we have to do here is to record the desired behavior
+ * in the vm_map_entry_t's.
+ */
- switch (old_entry->inheritance) {
- case VM_INHERIT_NONE:
- break;
+ case VM_BEHAVIOR_DEFAULT:
+ case VM_BEHAVIOR_RANDOM:
+ case VM_BEHAVIOR_SEQUENTIAL:
+ case VM_BEHAVIOR_RSEQNTL:
+ case VM_BEHAVIOR_ZERO_WIRED_PAGES:
+ vm_map_lock(map);
+
+ /*
+ * The entire address range must be valid for the map.
+ * Note that vm_map_range_check() does a
+ * vm_map_lookup_entry() internally and returns the
+ * entry containing the start of the address range if
+ * the entire range is valid.
+ */
+ if (vm_map_range_check(map, start, end, &temp_entry)) {
+ entry = temp_entry;
+ vm_map_clip_start(map, entry, start);
+ }
+ else {
+ vm_map_unlock(map);
+ return(KERN_INVALID_ADDRESS);
+ }
+
+ while ((entry != vm_map_to_entry(map)) && (entry->vme_start < end)) {
+ vm_map_clip_end(map, entry, end);
+ if (entry->is_sub_map) {
+ assert(!entry->use_pmap);
+ }
+
+ if( new_behavior == VM_BEHAVIOR_ZERO_WIRED_PAGES ) {
+ entry->zero_wired_pages = TRUE;
+ } else {
+ entry->behavior = new_behavior;
+ }
+ entry = entry->vme_next;
+ }
+
+ vm_map_unlock(map);
+ break;
- case VM_INHERIT_SHARE:
- vm_map_fork_share(old_map, old_entry, new_map);
- new_size += entry_size;
- break;
+ /*
+ * The rest of these are different from the above in that they cause
+ * an immediate action to take place as opposed to setting a behavior that
+ * affects future actions.
+ */
- case VM_INHERIT_COPY:
+ case VM_BEHAVIOR_WILLNEED:
+ return vm_map_willneed(map, start, end);
- /*
- * Inline the copy_quickly case;
- * upon failure, fall back on call
- * to vm_map_fork_copy.
- */
+ case VM_BEHAVIOR_DONTNEED:
+ return vm_map_msync(map, start, end - start, VM_SYNC_DEACTIVATE | VM_SYNC_CONTIGUOUS);
- if(old_entry->is_sub_map)
- break;
- if ((old_entry->wired_count != 0) ||
- ((old_entry->object.vm_object != NULL) &&
- (old_entry->object.vm_object->true_share))) {
- goto slow_vm_map_fork_copy;
- }
+ case VM_BEHAVIOR_FREE:
+ return vm_map_msync(map, start, end - start, VM_SYNC_KILLPAGES | VM_SYNC_CONTIGUOUS);
- new_entry = vm_map_entry_create(new_map);
- vm_map_entry_copy(new_entry, old_entry);
- /* clear address space specifics */
- new_entry->use_pmap = FALSE;
+ case VM_BEHAVIOR_REUSABLE:
+ return vm_map_reusable_pages(map, start, end);
- if (! vm_object_copy_quickly(
- &new_entry->object.vm_object,
- old_entry->offset,
- (old_entry->vme_end -
- old_entry->vme_start),
- &src_needs_copy,
- &new_entry_needs_copy)) {
- vm_map_entry_dispose(new_map, new_entry);
- goto slow_vm_map_fork_copy;
- }
+ case VM_BEHAVIOR_REUSE:
+ return vm_map_reuse_pages(map, start, end);
- /*
- * Handle copy-on-write obligations
- */
-
- if (src_needs_copy && !old_entry->needs_copy) {
- vm_object_pmap_protect(
- old_entry->object.vm_object,
- old_entry->offset,
- (old_entry->vme_end -
- old_entry->vme_start),
- ((old_entry->is_shared
- || old_map->mapped)
- ? PMAP_NULL :
- old_map->pmap),
- old_entry->vme_start,
- old_entry->protection & ~VM_PROT_WRITE);
+ case VM_BEHAVIOR_CAN_REUSE:
+ return vm_map_can_reuse(map, start, end);
- old_entry->needs_copy = TRUE;
- }
- new_entry->needs_copy = new_entry_needs_copy;
-
- /*
- * Insert the entry at the end
- * of the map.
- */
-
- vm_map_entry_link(new_map, vm_map_last_entry(new_map),
- new_entry);
- new_size += entry_size;
- break;
+#if MACH_ASSERT
+ case VM_BEHAVIOR_PAGEOUT:
+ return vm_map_pageout(map, start, end);
+#endif /* MACH_ASSERT */
- slow_vm_map_fork_copy:
- if (vm_map_fork_copy(old_map, &old_entry, new_map)) {
- new_size += entry_size;
- }
- continue;
- }
- old_entry = old_entry->vme_next;
+ default:
+ return(KERN_INVALID_ARGUMENT);
}
- new_map->size = new_size;
- vm_map_unlock(old_map);
- vm_map_deallocate(old_map);
-
- return(new_map);
+ return(KERN_SUCCESS);
}
/*
- * vm_map_lookup_locked:
- *
- * Finds the VM object, offset, and
- * protection for a given virtual address in the
- * specified map, assuming a page fault of the
- * type specified.
- *
- * Returns the (object, offset, protection) for
- * this address, whether it is wired down, and whether
- * this map has the only reference to the data in question.
- * In order to later verify this lookup, a "version"
- * is returned.
- *
- * The map MUST be locked by the caller and WILL be
- * locked on exit. In order to guarantee the
- * existence of the returned object, it is returned
- * locked.
+ * Internals for madvise(MADV_WILLNEED) system call.
*
- * If a lookup is requested with "write protection"
- * specified, the map may be changed to perform virtual
- * copying operations, although the data referenced will
- * remain the same.
+ * The present implementation is to do a read-ahead if the mapping corresponds
+ * to a mapped regular file. If it's an anonymous mapping, then we do nothing
+ * and basically ignore the "advice" (which we are always free to do).
*/
-kern_return_t
-vm_map_lookup_locked(
- vm_map_t *var_map, /* IN/OUT */
- register vm_offset_t vaddr,
- register vm_prot_t fault_type,
- vm_map_version_t *out_version, /* OUT */
- vm_object_t *object, /* OUT */
- vm_object_offset_t *offset, /* OUT */
- vm_prot_t *out_prot, /* OUT */
- boolean_t *wired, /* OUT */
- int *behavior, /* OUT */
- vm_object_offset_t *lo_offset, /* OUT */
- vm_object_offset_t *hi_offset, /* OUT */
- vm_map_t *pmap_map)
+
+
+static kern_return_t
+vm_map_willneed(
+ vm_map_t map,
+ vm_map_offset_t start,
+ vm_map_offset_t end
+)
{
- vm_map_entry_t entry;
- register vm_map_t map = *var_map;
- vm_map_t old_map = *var_map;
- vm_map_t cow_sub_map_parent = VM_MAP_NULL;
- vm_offset_t cow_parent_vaddr;
- vm_offset_t old_start;
- vm_offset_t old_end;
- register vm_prot_t prot;
+ vm_map_entry_t entry;
+ vm_object_t object;
+ memory_object_t pager;
+ struct vm_object_fault_info fault_info;
+ kern_return_t kr;
+ vm_object_size_t len;
+ vm_object_offset_t offset;
- *pmap_map = map;
- RetryLookup: ;
+ /*
+ * Fill in static values in fault_info. Several fields get ignored by the code
+ * we call, but we'll fill them in anyway since uninitialized fields are bad
+ * when it comes to future backwards compatibility.
+ */
+
+ fault_info.interruptible = THREAD_UNINT; /* ignored value */
+ fault_info.behavior = VM_BEHAVIOR_SEQUENTIAL;
+ fault_info.no_cache = FALSE; /* ignored value */
+ fault_info.stealth = TRUE;
+ fault_info.io_sync = FALSE;
+ fault_info.cs_bypass = FALSE;
+ fault_info.mark_zf_absent = FALSE;
+ fault_info.batch_pmap_op = FALSE;
/*
- * If the map has an interesting hint, try it before calling
- * full blown lookup routine.
+ * The MADV_WILLNEED operation doesn't require any changes to the
+ * vm_map_entry_t's, so the read lock is sufficient.
*/
- mutex_lock(&map->s_lock);
- entry = map->hint;
- mutex_unlock(&map->s_lock);
+ vm_map_lock_read(map);
- if ((entry == vm_map_to_entry(map)) ||
- (vaddr < entry->vme_start) || (vaddr >= entry->vme_end)) {
- vm_map_entry_t tmp_entry;
+ /*
+ * The madvise semantics require that the address range be fully
+ * allocated with no holes. Otherwise, we're required to return
+ * an error.
+ */
+
+ if (! vm_map_range_check(map, start, end, &entry)) {
+ vm_map_unlock_read(map);
+ return KERN_INVALID_ADDRESS;
+ }
+ /*
+ * Examine each vm_map_entry_t in the range.
+ */
+ for (; entry != vm_map_to_entry(map) && start < end; ) {
+
/*
- * Entry was either not a valid hint, or the vaddr
- * was not contained in the entry, so do a full lookup.
+ * The first time through, the start address could be anywhere
+ * within the vm_map_entry we found. So adjust the offset to
+ * correspond. After that, the offset will always be zero to
+ * correspond to the beginning of the current vm_map_entry.
*/
- if (!vm_map_lookup_entry(map, vaddr, &tmp_entry)) {
- if((cow_sub_map_parent) && (cow_sub_map_parent != map))
- vm_map_unlock(cow_sub_map_parent);
- if((*pmap_map != map)
- && (*pmap_map != cow_sub_map_parent))
- vm_map_unlock(*pmap_map);
- return KERN_INVALID_ADDRESS;
+ offset = (start - entry->vme_start) + VME_OFFSET(entry);
+
+ /*
+ * Set the length so we don't go beyond the end of the
+ * map_entry or beyond the end of the range we were given.
+ * This range could span also multiple map entries all of which
+ * map different files, so make sure we only do the right amount
+ * of I/O for each object. Note that it's possible for there
+ * to be multiple map entries all referring to the same object
+ * but with different page permissions, but it's not worth
+ * trying to optimize that case.
+ */
+ len = MIN(entry->vme_end - start, end - start);
+
+ if ((vm_size_t) len != len) {
+ /* 32-bit overflow */
+ len = (vm_size_t) (0 - PAGE_SIZE);
+ }
+ fault_info.cluster_size = (vm_size_t) len;
+ fault_info.lo_offset = offset;
+ fault_info.hi_offset = offset + len;
+ fault_info.user_tag = VME_ALIAS(entry);
+ fault_info.pmap_options = 0;
+ if (entry->iokit_acct ||
+ (!entry->is_sub_map && !entry->use_pmap)) {
+ fault_info.pmap_options |= PMAP_OPTIONS_ALT_ACCT;
}
- entry = tmp_entry;
- }
- if(map == old_map) {
- old_start = entry->vme_start;
- old_end = entry->vme_end;
- }
+ /*
+ * If there's no read permission to this mapping, then just
+ * skip it.
+ */
+ if ((entry->protection & VM_PROT_READ) == 0) {
+ entry = entry->vme_next;
+ start = entry->vme_start;
+ continue;
+ }
- /*
- * Handle submaps. Drop lock on upper map, submap is
- * returned locked.
- */
+ /*
+ * Find the file object backing this map entry. If there is
+ * none, then we simply ignore the "will need" advice for this
+ * entry and go on to the next one.
+ */
+ if ((object = find_vnode_object(entry)) == VM_OBJECT_NULL) {
+ entry = entry->vme_next;
+ start = entry->vme_start;
+ continue;
+ }
+
+ /*
+ * The data_request() could take a long time, so let's
+ * release the map lock to avoid blocking other threads.
+ */
+ vm_map_unlock_read(map);
+
+ vm_object_paging_begin(object);
+ pager = object->pager;
+ vm_object_unlock(object);
-submap_recurse:
- if (entry->is_sub_map) {
- vm_offset_t local_vaddr;
- vm_offset_t end_delta;
- vm_offset_t start_delta;
- vm_offset_t object_start_delta;
- vm_map_entry_t submap_entry;
- boolean_t mapped_needs_copy=FALSE;
+ /*
+ * Get the data from the object asynchronously.
+ *
+ * Note that memory_object_data_request() places limits on the
+ * amount of I/O it will do. Regardless of the len we
+ * specified, it won't do more than MAX_UPL_TRANSFER_BYTES and it
+ * silently truncates the len to that size. This isn't
+ * necessarily bad since madvise shouldn't really be used to
+ * page in unlimited amounts of data. Other Unix variants
+ * limit the willneed case as well. If this turns out to be an
+ * issue for developers, then we can always adjust the policy
+ * here and still be backwards compatible since this is all
+ * just "advice".
+ */
+ kr = memory_object_data_request(
+ pager,
+ offset + object->paging_offset,
+ 0, /* ignored */
+ VM_PROT_READ,
+ (memory_object_fault_info_t)&fault_info);
- local_vaddr = vaddr;
+ vm_object_lock(object);
+ vm_object_paging_end(object);
+ vm_object_unlock(object);
- if ((!entry->needs_copy) && (entry->use_pmap)) {
- /* if pmap_map equals map we unlock below */
- if ((*pmap_map != map) &&
- (*pmap_map != cow_sub_map_parent))
- vm_map_unlock(*pmap_map);
- *pmap_map = entry->object.sub_map;
+ /*
+ * If we couldn't do the I/O for some reason, just give up on
+ * the madvise. We still return success to the user since
+ * madvise isn't supposed to fail when the advice can't be
+ * taken.
+ */
+ if (kr != KERN_SUCCESS) {
+ return KERN_SUCCESS;
}
- if(entry->needs_copy) {
- if (!mapped_needs_copy) {
- if (vm_map_lock_read_to_write(map)) {
- vm_map_lock_read(map);
- if(*pmap_map == entry->object.sub_map)
- *pmap_map = map;
- goto RetryLookup;
- }
- vm_map_lock_read(entry->object.sub_map);
- cow_sub_map_parent = map;
- /* reset base to map before cow object */
- /* this is the map which will accept */
- /* the new cow object */
- old_start = entry->vme_start;
- old_end = entry->vme_end;
- cow_parent_vaddr = vaddr;
- mapped_needs_copy = TRUE;
- } else {
- vm_map_lock_read(entry->object.sub_map);
- if((cow_sub_map_parent != map) &&
- (*pmap_map != map))
- vm_map_unlock(map);
- }
- } else {
- vm_map_lock_read(entry->object.sub_map);
- /* leave map locked if it is a target */
- /* cow sub_map above otherwise, just */
- /* follow the maps down to the object */
- /* here we unlock knowing we are not */
- /* revisiting the map. */
- if((*pmap_map != map) && (map != cow_sub_map_parent))
- vm_map_unlock_read(map);
+ start += len;
+ if (start >= end) {
+ /* done */
+ return KERN_SUCCESS;
}
- *var_map = map = entry->object.sub_map;
-
- /* calculate the offset in the submap for vaddr */
- local_vaddr = (local_vaddr - entry->vme_start) + entry->offset;
-
-RetrySubMap:
- if(!vm_map_lookup_entry(map, local_vaddr, &submap_entry)) {
- if((cow_sub_map_parent) && (cow_sub_map_parent != map)){
- vm_map_unlock(cow_sub_map_parent);
- }
- if((*pmap_map != map)
- && (*pmap_map != cow_sub_map_parent)) {
- vm_map_unlock(*pmap_map);
- }
- *pmap_map = map;
+ /* look up next entry */
+ vm_map_lock_read(map);
+ if (! vm_map_lookup_entry(map, start, &entry)) {
+ /*
+ * There's a new hole in the address range.
+ */
+ vm_map_unlock_read(map);
return KERN_INVALID_ADDRESS;
}
- /* find the attenuated shadow of the underlying object */
- /* on our target map */
+ }
- /* in english the submap object may extend beyond the */
- /* region mapped by the entry or, may only fill a portion */
- /* of it. For our purposes, we only care if the object */
- /* doesn't fill. In this case the area which will */
- /* ultimately be clipped in the top map will only need */
- /* to be as big as the portion of the underlying entry */
- /* which is mapped */
- start_delta = submap_entry->vme_start > entry->offset ?
- submap_entry->vme_start - entry->offset : 0;
+ vm_map_unlock_read(map);
+ return KERN_SUCCESS;
+}
- end_delta =
- (entry->offset + start_delta + (old_end - old_start)) <=
- submap_entry->vme_end ?
- 0 : (entry->offset +
- (old_end - old_start))
- - submap_entry->vme_end;
+static boolean_t
+vm_map_entry_is_reusable(
+ vm_map_entry_t entry)
+{
+ /* Only user map entries */
- old_start += start_delta;
- old_end -= end_delta;
+ vm_object_t object;
- if(submap_entry->is_sub_map) {
- entry = submap_entry;
- vaddr = local_vaddr;
- goto submap_recurse;
- }
+ if (entry->is_sub_map) {
+ return FALSE;
+ }
- if(((fault_type & VM_PROT_WRITE) && cow_sub_map_parent)) {
+ switch (VME_ALIAS(entry)) {
+ case VM_MEMORY_MALLOC:
+ case VM_MEMORY_MALLOC_SMALL:
+ case VM_MEMORY_MALLOC_LARGE:
+ case VM_MEMORY_REALLOC:
+ case VM_MEMORY_MALLOC_TINY:
+ case VM_MEMORY_MALLOC_LARGE_REUSABLE:
+ case VM_MEMORY_MALLOC_LARGE_REUSED:
+ /*
+ * This is a malloc() memory region: check if it's still
+ * in its original state and can be re-used for more
+ * malloc() allocations.
+ */
+ break;
+ default:
+ /*
+ * Not a malloc() memory region: let the caller decide if
+ * it's re-usable.
+ */
+ return TRUE;
+ }
- vm_object_t copy_object;
- vm_offset_t local_start;
- vm_offset_t local_end;
- boolean_t copied_slowly = FALSE;
+ if (entry->is_shared ||
+ entry->is_sub_map ||
+ entry->in_transition ||
+ entry->protection != VM_PROT_DEFAULT ||
+ entry->max_protection != VM_PROT_ALL ||
+ entry->inheritance != VM_INHERIT_DEFAULT ||
+ entry->no_cache ||
+ entry->permanent ||
+ entry->superpage_size != FALSE ||
+ entry->zero_wired_pages ||
+ entry->wired_count != 0 ||
+ entry->user_wired_count != 0) {
+ return FALSE;
+ }
- if (vm_map_lock_read_to_write(map)) {
- vm_map_lock_read(map);
- old_start -= start_delta;
- old_end += end_delta;
- goto RetrySubMap;
- }
+ object = VME_OBJECT(entry);
+ if (object == VM_OBJECT_NULL) {
+ return TRUE;
+ }
+ if (
+#if 0
+ /*
+ * Let's proceed even if the VM object is potentially
+ * shared.
+ * We check for this later when processing the actual
+ * VM pages, so the contents will be safe if shared.
+ *
+ * But we can still mark this memory region as "reusable" to
+ * acknowledge that the caller did let us know that the memory
+ * could be re-used and should not be penalized for holding
+ * on to it. This allows its "resident size" to not include
+ * the reusable range.
+ */
+ object->ref_count == 1 &&
+#endif
+ object->wired_page_count == 0 &&
+ object->copy == VM_OBJECT_NULL &&
+ object->shadow == VM_OBJECT_NULL &&
+ object->copy_strategy == MEMORY_OBJECT_COPY_SYMMETRIC &&
+ object->internal &&
+ !object->true_share &&
+ object->wimg_bits == VM_WIMG_USE_DEFAULT &&
+ !object->code_signed) {
+ return TRUE;
+ }
+ return FALSE;
+
+
+}
+static kern_return_t
+vm_map_reuse_pages(
+ vm_map_t map,
+ vm_map_offset_t start,
+ vm_map_offset_t end)
+{
+ vm_map_entry_t entry;
+ vm_object_t object;
+ vm_object_offset_t start_offset, end_offset;
- if (submap_entry->object.vm_object == VM_OBJECT_NULL) {
- submap_entry->object.vm_object =
- vm_object_allocate(
- (vm_size_t)
- (submap_entry->vme_end
- - submap_entry->vme_start));
- submap_entry->offset = 0;
- }
- local_start = local_vaddr -
- (cow_parent_vaddr - old_start);
- local_end = local_vaddr +
- (old_end - cow_parent_vaddr);
- vm_map_clip_start(map, submap_entry, local_start);
- vm_map_clip_end(map, submap_entry, local_end);
+ /*
+ * The MADV_REUSE operation doesn't require any changes to the
+ * vm_map_entry_t's, so the read lock is sufficient.
+ */
- /* This is the COW case, lets connect */
- /* an entry in our space to the underlying */
- /* object in the submap, bypassing the */
- /* submap. */
+ vm_map_lock_read(map);
+ assert(map->pmap != kernel_pmap); /* protect alias access */
+ /*
+ * The madvise semantics require that the address range be fully
+ * allocated with no holes. Otherwise, we're required to return
+ * an error.
+ */
- if(submap_entry->wired_count != 0) {
- vm_object_lock(
- submap_entry->object.vm_object);
- vm_object_copy_slowly(
- submap_entry->object.vm_object,
- submap_entry->offset,
- submap_entry->vme_end -
- submap_entry->vme_start,
- FALSE,
- ©_object);
- copied_slowly = TRUE;
- } else {
-
- /* set up shadow object */
- copy_object = submap_entry->object.vm_object;
- vm_object_reference(copy_object);
- submap_entry->object.vm_object->shadowed = TRUE;
- submap_entry->needs_copy = TRUE;
- vm_object_pmap_protect(
- submap_entry->object.vm_object,
- submap_entry->offset,
- submap_entry->vme_end -
- submap_entry->vme_start,
- (submap_entry->is_shared
- || map->mapped) ?
- PMAP_NULL : map->pmap,
- submap_entry->vme_start,
- submap_entry->protection &
- ~VM_PROT_WRITE);
- }
-
+ if (!vm_map_range_check(map, start, end, &entry)) {
+ vm_map_unlock_read(map);
+ vm_page_stats_reusable.reuse_pages_failure++;
+ return KERN_INVALID_ADDRESS;
+ }
- /* This works diffently than the */
- /* normal submap case. We go back */
- /* to the parent of the cow map and*/
- /* clip out the target portion of */
- /* the sub_map, substituting the */
- /* new copy object, */
+ /*
+ * Examine each vm_map_entry_t in the range.
+ */
+ for (; entry != vm_map_to_entry(map) && entry->vme_start < end;
+ entry = entry->vme_next) {
+ /*
+ * Sanity check on the VM map entry.
+ */
+ if (! vm_map_entry_is_reusable(entry)) {
+ vm_map_unlock_read(map);
+ vm_page_stats_reusable.reuse_pages_failure++;
+ return KERN_INVALID_ADDRESS;
+ }
- vm_map_unlock(map);
- local_start = old_start;
- local_end = old_end;
- map = cow_sub_map_parent;
- *var_map = cow_sub_map_parent;
- vaddr = cow_parent_vaddr;
- cow_sub_map_parent = NULL;
+ /*
+ * The first time through, the start address could be anywhere
+ * within the vm_map_entry we found. So adjust the offset to
+ * correspond.
+ */
+ if (entry->vme_start < start) {
+ start_offset = start - entry->vme_start;
+ } else {
+ start_offset = 0;
+ }
+ end_offset = MIN(end, entry->vme_end) - entry->vme_start;
+ start_offset += VME_OFFSET(entry);
+ end_offset += VME_OFFSET(entry);
- if(!vm_map_lookup_entry(map,
- vaddr, &entry)) {
- vm_object_deallocate(
- copy_object);
- vm_map_lock_write_to_read(map);
- return KERN_INVALID_ADDRESS;
- }
-
- /* clip out the portion of space */
- /* mapped by the sub map which */
- /* corresponds to the underlying */
- /* object */
- vm_map_clip_start(map, entry, local_start);
- vm_map_clip_end(map, entry, local_end);
+ assert(!entry->is_sub_map);
+ object = VME_OBJECT(entry);
+ if (object != VM_OBJECT_NULL) {
+ vm_object_lock(object);
+ vm_object_reuse_pages(object, start_offset, end_offset,
+ TRUE);
+ vm_object_unlock(object);
+ }
+ if (VME_ALIAS(entry) == VM_MEMORY_MALLOC_LARGE_REUSABLE) {
+ /*
+ * XXX
+ * We do not hold the VM map exclusively here.
+ * The "alias" field is not that critical, so it's
+ * safe to update it here, as long as it is the only
+ * one that can be modified while holding the VM map
+ * "shared".
+ */
+ VME_ALIAS_SET(entry, VM_MEMORY_MALLOC_LARGE_REUSED);
+ }
+ }
+
+ vm_map_unlock_read(map);
+ vm_page_stats_reusable.reuse_pages_success++;
+ return KERN_SUCCESS;
+}
- /* substitute copy object for */
- /* shared map entry */
- vm_map_deallocate(entry->object.sub_map);
- entry->is_sub_map = FALSE;
- entry->object.vm_object = copy_object;
- entry->protection |= VM_PROT_WRITE;
- entry->max_protection |= VM_PROT_WRITE;
- if(copied_slowly) {
- entry->offset = 0;
- entry->needs_copy = FALSE;
- entry->is_shared = FALSE;
- } else {
- entry->offset = submap_entry->offset;
- entry->needs_copy = TRUE;
- if(entry->inheritance == VM_INHERIT_SHARE)
- entry->inheritance = VM_INHERIT_COPY;
- if (map != old_map)
- entry->is_shared = TRUE;
- }
- if(entry->inheritance == VM_INHERIT_SHARE)
- entry->inheritance = VM_INHERIT_COPY;
+static kern_return_t
+vm_map_reusable_pages(
+ vm_map_t map,
+ vm_map_offset_t start,
+ vm_map_offset_t end)
+{
+ vm_map_entry_t entry;
+ vm_object_t object;
+ vm_object_offset_t start_offset, end_offset;
+ vm_map_offset_t pmap_offset;
- vm_map_lock_write_to_read(map);
- } else {
- if((cow_sub_map_parent)
- && (cow_sub_map_parent != *pmap_map)
- && (cow_sub_map_parent != map)) {
- vm_map_unlock(cow_sub_map_parent);
- }
- entry = submap_entry;
- vaddr = local_vaddr;
- }
- }
-
/*
- * Check whether this task is allowed to have
- * this page.
+ * The MADV_REUSABLE operation doesn't require any changes to the
+ * vm_map_entry_t's, so the read lock is sufficient.
*/
- prot = entry->protection;
- if ((fault_type & (prot)) != fault_type) {
- if (*pmap_map != map) {
- vm_map_unlock(*pmap_map);
- }
- *pmap_map = map;
- return KERN_PROTECTION_FAILURE;
- }
+ vm_map_lock_read(map);
+ assert(map->pmap != kernel_pmap); /* protect alias access */
/*
- * If this page is not pageable, we have to get
- * it for all possible accesses.
+ * The madvise semantics require that the address range be fully
+ * allocated with no holes. Otherwise, we're required to return
+ * an error.
*/
- if (*wired = (entry->wired_count != 0))
- prot = fault_type = entry->protection;
+ if (!vm_map_range_check(map, start, end, &entry)) {
+ vm_map_unlock_read(map);
+ vm_page_stats_reusable.reusable_pages_failure++;
+ return KERN_INVALID_ADDRESS;
+ }
/*
- * If the entry was copy-on-write, we either ...
+ * Examine each vm_map_entry_t in the range.
*/
+ for (; entry != vm_map_to_entry(map) && entry->vme_start < end;
+ entry = entry->vme_next) {
+ int kill_pages = 0;
- if (entry->needs_copy) {
- /*
- * If we want to write the page, we may as well
- * handle that now since we've got the map locked.
- *
- * If we don't need to write the page, we just
- * demote the permissions allowed.
+ /*
+ * Sanity check on the VM map entry.
*/
+ if (! vm_map_entry_is_reusable(entry)) {
+ vm_map_unlock_read(map);
+ vm_page_stats_reusable.reusable_pages_failure++;
+ return KERN_INVALID_ADDRESS;
+ }
- if (fault_type & VM_PROT_WRITE || *wired) {
- /*
- * Make a new object, and place it in the
- * object chain. Note that no new references
- * have appeared -- one just moved from the
- * map to the new object.
- */
+ /*
+ * The first time through, the start address could be anywhere
+ * within the vm_map_entry we found. So adjust the offset to
+ * correspond.
+ */
+ if (entry->vme_start < start) {
+ start_offset = start - entry->vme_start;
+ pmap_offset = start;
+ } else {
+ start_offset = 0;
+ pmap_offset = entry->vme_start;
+ }
+ end_offset = MIN(end, entry->vme_end) - entry->vme_start;
+ start_offset += VME_OFFSET(entry);
+ end_offset += VME_OFFSET(entry);
- if (vm_map_lock_read_to_write(map)) {
- vm_map_lock_read(map);
- goto RetryLookup;
- }
- vm_object_shadow(&entry->object.vm_object,
- &entry->offset,
- (vm_size_t) (entry->vme_end -
- entry->vme_start));
+ assert(!entry->is_sub_map);
+ object = VME_OBJECT(entry);
+ if (object == VM_OBJECT_NULL)
+ continue;
- entry->object.vm_object->shadowed = TRUE;
- entry->needs_copy = FALSE;
- vm_map_lock_write_to_read(map);
+
+ vm_object_lock(object);
+ if (object->ref_count == 1 &&
+ !object->shadow &&
+ /*
+ * "iokit_acct" entries are billed for their virtual size
+ * (rather than for their resident pages only), so they
+ * wouldn't benefit from making pages reusable, and it
+ * would be hard to keep track of pages that are both
+ * "iokit_acct" and "reusable" in the pmap stats and ledgers.
+ */
+ !(entry->iokit_acct ||
+ (!entry->is_sub_map && !entry->use_pmap)))
+ kill_pages = 1;
+ else
+ kill_pages = -1;
+ if (kill_pages != -1) {
+ vm_object_deactivate_pages(object,
+ start_offset,
+ end_offset - start_offset,
+ kill_pages,
+ TRUE /*reusable_pages*/,
+ map->pmap,
+ pmap_offset);
+ } else {
+ vm_page_stats_reusable.reusable_pages_shared++;
}
- else {
+ vm_object_unlock(object);
+
+ if (VME_ALIAS(entry) == VM_MEMORY_MALLOC_LARGE ||
+ VME_ALIAS(entry) == VM_MEMORY_MALLOC_LARGE_REUSED) {
/*
- * We're attempting to read a copy-on-write
- * page -- don't allow writes.
+ * XXX
+ * We do not hold the VM map exclusively here.
+ * The "alias" field is not that critical, so it's
+ * safe to update it here, as long as it is the only
+ * one that can be modified while holding the VM map
+ * "shared".
*/
-
- prot &= (~VM_PROT_WRITE);
+ VME_ALIAS_SET(entry, VM_MEMORY_MALLOC_LARGE_REUSABLE);
}
}
+
+ vm_map_unlock_read(map);
+ vm_page_stats_reusable.reusable_pages_success++;
+ return KERN_SUCCESS;
+}
- /*
- * Create an object if necessary.
- */
- if (entry->object.vm_object == VM_OBJECT_NULL) {
-
- if (vm_map_lock_read_to_write(map)) {
- vm_map_lock_read(map);
- goto RetryLookup;
- }
- entry->object.vm_object = vm_object_allocate(
- (vm_size_t)(entry->vme_end - entry->vme_start));
- entry->offset = 0;
- vm_map_lock_write_to_read(map);
- }
+static kern_return_t
+vm_map_can_reuse(
+ vm_map_t map,
+ vm_map_offset_t start,
+ vm_map_offset_t end)
+{
+ vm_map_entry_t entry;
/*
- * Return the object/offset from this entry. If the entry
- * was copy-on-write or empty, it has been fixed up. Also
- * return the protection.
+ * The MADV_REUSABLE operation doesn't require any changes to the
+ * vm_map_entry_t's, so the read lock is sufficient.
*/
- *offset = (vaddr - entry->vme_start) + entry->offset;
- *object = entry->object.vm_object;
- *out_prot = prot;
- *behavior = entry->behavior;
- *lo_offset = entry->offset;
- *hi_offset = (entry->vme_end - entry->vme_start) + entry->offset;
+ vm_map_lock_read(map);
+ assert(map->pmap != kernel_pmap); /* protect alias access */
/*
- * Lock the object to prevent it from disappearing
+ * The madvise semantics require that the address range be fully
+ * allocated with no holes. Otherwise, we're required to return
+ * an error.
*/
- vm_object_lock(*object);
+ if (!vm_map_range_check(map, start, end, &entry)) {
+ vm_map_unlock_read(map);
+ vm_page_stats_reusable.can_reuse_failure++;
+ return KERN_INVALID_ADDRESS;
+ }
/*
- * Save the version number
+ * Examine each vm_map_entry_t in the range.
*/
-
- out_version->main_timestamp = map->timestamp;
-
+ for (; entry != vm_map_to_entry(map) && entry->vme_start < end;
+ entry = entry->vme_next) {
+ /*
+ * Sanity check on the VM map entry.
+ */
+ if (! vm_map_entry_is_reusable(entry)) {
+ vm_map_unlock_read(map);
+ vm_page_stats_reusable.can_reuse_failure++;
+ return KERN_INVALID_ADDRESS;
+ }
+ }
+
+ vm_map_unlock_read(map);
+ vm_page_stats_reusable.can_reuse_success++;
return KERN_SUCCESS;
}
-/*
- * vm_map_verify:
- *
- * Verifies that the map in question has not changed
- * since the given version. If successful, the map
- * will not change until vm_map_verify_done() is called.
- */
-boolean_t
-vm_map_verify(
- register vm_map_t map,
- register vm_map_version_t *version) /* REF */
+#if MACH_ASSERT
+static kern_return_t
+vm_map_pageout(
+ vm_map_t map,
+ vm_map_offset_t start,
+ vm_map_offset_t end)
{
- boolean_t result;
-
- vm_map_lock_read(map);
- result = (map->timestamp == version->main_timestamp);
+ vm_map_entry_t entry;
- if (!result)
- vm_map_unlock_read(map);
-
- return(result);
-}
-
-/*
- * vm_map_verify_done:
- *
- * Releases locks acquired by a vm_map_verify.
- *
- * This is now a macro in vm/vm_map.h. It does a
- * vm_map_unlock_read on the map.
- */
+ /*
+ * The MADV_PAGEOUT operation doesn't require any changes to the
+ * vm_map_entry_t's, so the read lock is sufficient.
+ */
+ vm_map_lock_read(map);
-/*
- * vm_region:
- *
- * User call to obtain information about a region in
- * a task's address map. Currently, only one flavor is
- * supported.
- *
- * XXX The reserved and behavior fields cannot be filled
- * in until the vm merge from the IK is completed, and
- * vm_reserve is implemented.
- *
- * XXX Dependency: syscall_vm_region() also supports only one flavor.
- */
+ /*
+ * The madvise semantics require that the address range be fully
+ * allocated with no holes. Otherwise, we're required to return
+ * an error.
+ */
-kern_return_t
-vm_region(
- vm_map_t map,
- vm_offset_t *address, /* IN/OUT */
- vm_size_t *size, /* OUT */
- vm_region_flavor_t flavor, /* IN */
- vm_region_info_t info, /* OUT */
- mach_msg_type_number_t *count, /* IN/OUT */
- ipc_port_t *object_name) /* OUT */
-{
- vm_map_entry_t tmp_entry;
- register
- vm_map_entry_t entry;
- register
- vm_offset_t start;
- vm_region_basic_info_t basic;
- vm_region_extended_info_t extended;
- vm_region_top_info_t top;
+ if (!vm_map_range_check(map, start, end, &entry)) {
+ vm_map_unlock_read(map);
+ return KERN_INVALID_ADDRESS;
+ }
- if (map == VM_MAP_NULL)
- return(KERN_INVALID_ARGUMENT);
+ /*
+ * Examine each vm_map_entry_t in the range.
+ */
+ for (; entry != vm_map_to_entry(map) && entry->vme_start < end;
+ entry = entry->vme_next) {
+ vm_object_t object;
- switch (flavor) {
-
- case VM_REGION_BASIC_INFO:
- {
- if (*count < VM_REGION_BASIC_INFO_COUNT)
- return(KERN_INVALID_ARGUMENT);
+ /*
+ * Sanity check on the VM map entry.
+ */
+ if (entry->is_sub_map) {
+ vm_map_t submap;
+ vm_map_offset_t submap_start;
+ vm_map_offset_t submap_end;
+ vm_map_entry_t submap_entry;
+
+ submap = VME_SUBMAP(entry);
+ submap_start = VME_OFFSET(entry);
+ submap_end = submap_start + (entry->vme_end -
+ entry->vme_start);
+
+ vm_map_lock_read(submap);
+
+ if (! vm_map_range_check(submap,
+ submap_start,
+ submap_end,
+ &submap_entry)) {
+ vm_map_unlock_read(submap);
+ vm_map_unlock_read(map);
+ return KERN_INVALID_ADDRESS;
+ }
- basic = (vm_region_basic_info_t) info;
- *count = VM_REGION_BASIC_INFO_COUNT;
+ object = VME_OBJECT(submap_entry);
+ if (submap_entry->is_sub_map ||
+ object == VM_OBJECT_NULL ||
+ !object->internal) {
+ vm_map_unlock_read(submap);
+ continue;
+ }
- vm_map_lock_read(map);
+ vm_object_pageout(object);
- start = *address;
- if (!vm_map_lookup_entry(map, start, &tmp_entry)) {
- if ((entry = tmp_entry->vme_next) == vm_map_to_entry(map)) {
- vm_map_unlock_read(map);
- return(KERN_INVALID_ADDRESS);
+ vm_map_unlock_read(submap);
+ submap = VM_MAP_NULL;
+ submap_entry = VM_MAP_ENTRY_NULL;
+ continue;
}
- } else {
- entry = tmp_entry;
- }
-
- start = entry->vme_start;
-
- basic->offset = entry->offset;
- basic->protection = entry->protection;
- basic->inheritance = entry->inheritance;
- basic->max_protection = entry->max_protection;
- basic->behavior = entry->behavior;
- basic->user_wired_count = entry->user_wired_count;
- basic->reserved = entry->is_sub_map;
- *address = start;
- *size = (entry->vme_end - start);
-
- if (object_name) *object_name = IP_NULL;
- if (entry->is_sub_map) {
- basic->shared = FALSE;
- } else {
- basic->shared = entry->is_shared;
- }
-
- vm_map_unlock_read(map);
- return(KERN_SUCCESS);
- }
- case VM_REGION_EXTENDED_INFO:
- {
- if (*count < VM_REGION_EXTENDED_INFO_COUNT)
- return(KERN_INVALID_ARGUMENT);
-
- extended = (vm_region_extended_info_t) info;
- *count = VM_REGION_EXTENDED_INFO_COUNT;
-
- vm_map_lock_read(map);
-
- start = *address;
- if (!vm_map_lookup_entry(map, start, &tmp_entry)) {
- if ((entry = tmp_entry->vme_next) == vm_map_to_entry(map)) {
- vm_map_unlock_read(map);
- return(KERN_INVALID_ADDRESS);
+ object = VME_OBJECT(entry);
+ if (entry->is_sub_map ||
+ object == VM_OBJECT_NULL ||
+ !object->internal) {
+ continue;
}
- } else {
- entry = tmp_entry;
- }
- start = entry->vme_start;
-
- extended->protection = entry->protection;
- extended->user_tag = entry->alias;
- extended->pages_resident = 0;
- extended->pages_swapped_out = 0;
- extended->pages_shared_now_private = 0;
- extended->pages_dirtied = 0;
- extended->external_pager = 0;
- extended->shadow_depth = 0;
- vm_region_walk(entry, extended, entry->offset, entry->vme_end - start, map, start);
-
- if (extended->external_pager && extended->ref_count == 2 && extended->share_mode == SM_SHARED)
- extended->share_mode = SM_PRIVATE;
+ vm_object_pageout(object);
+ }
+
+ vm_map_unlock_read(map);
+ return KERN_SUCCESS;
+}
+#endif /* MACH_ASSERT */
- if (object_name)
- *object_name = IP_NULL;
- *address = start;
- *size = (entry->vme_end - start);
- vm_map_unlock_read(map);
- return(KERN_SUCCESS);
- }
- case VM_REGION_TOP_INFO:
- {
+/*
+ * Routine: vm_map_entry_insert
+ *
+ * Descritpion: This routine inserts a new vm_entry in a locked map.
+ */
+vm_map_entry_t
+vm_map_entry_insert(
+ vm_map_t map,
+ vm_map_entry_t insp_entry,
+ vm_map_offset_t start,
+ vm_map_offset_t end,
+ vm_object_t object,
+ vm_object_offset_t offset,
+ boolean_t needs_copy,
+ boolean_t is_shared,
+ boolean_t in_transition,
+ vm_prot_t cur_protection,
+ vm_prot_t max_protection,
+ vm_behavior_t behavior,
+ vm_inherit_t inheritance,
+ unsigned wired_count,
+ boolean_t no_cache,
+ boolean_t permanent,
+ unsigned int superpage_size,
+ boolean_t clear_map_aligned,
+ boolean_t is_submap)
+{
+ vm_map_entry_t new_entry;
- if (*count < VM_REGION_TOP_INFO_COUNT)
- return(KERN_INVALID_ARGUMENT);
+ assert(insp_entry != (vm_map_entry_t)0);
- top = (vm_region_top_info_t) info;
- *count = VM_REGION_TOP_INFO_COUNT;
+ new_entry = vm_map_entry_create(map, !map->hdr.entries_pageable);
- vm_map_lock_read(map);
+ if (VM_MAP_PAGE_SHIFT(map) != PAGE_SHIFT) {
+ new_entry->map_aligned = TRUE;
+ } else {
+ new_entry->map_aligned = FALSE;
+ }
+ if (clear_map_aligned &&
+ (! VM_MAP_PAGE_ALIGNED(start, VM_MAP_PAGE_MASK(map)) ||
+ ! VM_MAP_PAGE_ALIGNED(end, VM_MAP_PAGE_MASK(map)))) {
+ new_entry->map_aligned = FALSE;
+ }
- start = *address;
- if (!vm_map_lookup_entry(map, start, &tmp_entry)) {
- if ((entry = tmp_entry->vme_next) == vm_map_to_entry(map)) {
- vm_map_unlock_read(map);
- return(KERN_INVALID_ADDRESS);
- }
- } else {
- entry = tmp_entry;
+ new_entry->vme_start = start;
+ new_entry->vme_end = end;
+ assert(page_aligned(new_entry->vme_start));
+ assert(page_aligned(new_entry->vme_end));
+ if (new_entry->map_aligned) {
+ assert(VM_MAP_PAGE_ALIGNED(new_entry->vme_start,
+ VM_MAP_PAGE_MASK(map)));
+ assert(VM_MAP_PAGE_ALIGNED(new_entry->vme_end,
+ VM_MAP_PAGE_MASK(map)));
+ }
+ assert(new_entry->vme_start < new_entry->vme_end);
- }
- start = entry->vme_start;
+ VME_OBJECT_SET(new_entry, object);
+ VME_OFFSET_SET(new_entry, offset);
+ new_entry->is_shared = is_shared;
+ new_entry->is_sub_map = is_submap;
+ new_entry->needs_copy = needs_copy;
+ new_entry->in_transition = in_transition;
+ new_entry->needs_wakeup = FALSE;
+ new_entry->inheritance = inheritance;
+ new_entry->protection = cur_protection;
+ new_entry->max_protection = max_protection;
+ new_entry->behavior = behavior;
+ new_entry->wired_count = wired_count;
+ new_entry->user_wired_count = 0;
+ if (is_submap) {
+ /*
+ * submap: "use_pmap" means "nested".
+ * default: false.
+ */
+ new_entry->use_pmap = FALSE;
+ } else {
+ /*
+ * object: "use_pmap" means "use pmap accounting" for footprint.
+ * default: true.
+ */
+ new_entry->use_pmap = TRUE;
+ }
+ VME_ALIAS_SET(new_entry, 0);
+ new_entry->zero_wired_pages = FALSE;
+ new_entry->no_cache = no_cache;
+ new_entry->permanent = permanent;
+ if (superpage_size)
+ new_entry->superpage_size = TRUE;
+ else
+ new_entry->superpage_size = FALSE;
+ new_entry->used_for_jit = FALSE;
+ new_entry->iokit_acct = FALSE;
+ new_entry->vme_resilient_codesign = FALSE;
+ new_entry->vme_resilient_media = FALSE;
- top->private_pages_resident = 0;
- top->shared_pages_resident = 0;
+ /*
+ * Insert the new entry into the list.
+ */
- vm_region_top_walk(entry, top);
+ vm_map_store_entry_link(map, insp_entry, new_entry);
+ map->size += end - start;
- if (object_name)
- *object_name = IP_NULL;
- *address = start;
- *size = (entry->vme_end - start);
+ /*
+ * Update the free space hint and the lookup hint.
+ */
- vm_map_unlock_read(map);
- return(KERN_SUCCESS);
- }
- default:
- return(KERN_INVALID_ARGUMENT);
- }
+ SAVE_HINT_MAP_WRITE(map, new_entry);
+ return new_entry;
}
/*
- * vm_region_recurse: A form of vm_region which follows the
- * submaps in a target map
+ * Routine: vm_map_remap_extract
*
+ * Descritpion: This routine returns a vm_entry list from a map.
*/
-
-kern_return_t
-vm_region_recurse(
- vm_map_t map,
- vm_offset_t *address, /* IN/OUT */
- vm_size_t *size, /* OUT */
- natural_t *nesting_depth, /* IN/OUT */
- vm_region_recurse_info_t info, /* IN/OUT */
- mach_msg_type_number_t *count) /* IN/OUT */
+static kern_return_t
+vm_map_remap_extract(
+ vm_map_t map,
+ vm_map_offset_t addr,
+ vm_map_size_t size,
+ boolean_t copy,
+ struct vm_map_header *map_header,
+ vm_prot_t *cur_protection,
+ vm_prot_t *max_protection,
+ /* What, no behavior? */
+ vm_inherit_t inheritance,
+ boolean_t pageable)
{
- vm_map_entry_t tmp_entry;
- register
- vm_map_entry_t entry;
- register
- vm_offset_t start;
-
- unsigned int recurse_count;
- vm_map_t submap;
- vm_map_t base_map;
- vm_map_entry_t base_entry;
- vm_offset_t base_next;
- vm_offset_t base_addr;
- vm_offset_t baddr_start_delta;
- vm_region_submap_info_t submap_info;
- vm_region_extended_info_data_t extended;
+ kern_return_t result;
+ vm_map_size_t mapped_size;
+ vm_map_size_t tmp_size;
+ vm_map_entry_t src_entry; /* result of last map lookup */
+ vm_map_entry_t new_entry;
+ vm_object_offset_t offset;
+ vm_map_offset_t map_address;
+ vm_map_offset_t src_start; /* start of entry to map */
+ vm_map_offset_t src_end; /* end of region to be mapped */
+ vm_object_t object;
+ vm_map_version_t version;
+ boolean_t src_needs_copy;
+ boolean_t new_entry_needs_copy;
- if (map == VM_MAP_NULL)
- return(KERN_INVALID_ARGUMENT);
+ assert(map != VM_MAP_NULL);
+ assert(size != 0);
+ assert(size == vm_map_round_page(size, PAGE_MASK));
+ assert(inheritance == VM_INHERIT_NONE ||
+ inheritance == VM_INHERIT_COPY ||
+ inheritance == VM_INHERIT_SHARE);
- submap_info = (vm_region_submap_info_t) info;
- *count = VM_REGION_SUBMAP_INFO_COUNT;
+ /*
+ * Compute start and end of region.
+ */
+ src_start = vm_map_trunc_page(addr, PAGE_MASK);
+ src_end = vm_map_round_page(src_start + size, PAGE_MASK);
- if (*count < VM_REGION_SUBMAP_INFO_COUNT)
- return(KERN_INVALID_ARGUMENT);
- start = *address;
- base_map = map;
- recurse_count = *nesting_depth;
+ /*
+ * Initialize map_header.
+ */
+ map_header->links.next = (struct vm_map_entry *)&map_header->links;
+ map_header->links.prev = (struct vm_map_entry *)&map_header->links;
+ map_header->nentries = 0;
+ map_header->entries_pageable = pageable;
+ map_header->page_shift = PAGE_SHIFT;
+
+ vm_map_store_init( map_header );
+
+ *cur_protection = VM_PROT_ALL;
+ *max_protection = VM_PROT_ALL;
+
+ map_address = 0;
+ mapped_size = 0;
+ result = KERN_SUCCESS;
+
+ /*
+ * The specified source virtual space might correspond to
+ * multiple map entries, need to loop on them.
+ */
+ vm_map_lock(map);
+ while (mapped_size != size) {
+ vm_map_size_t entry_size;
-LOOKUP_NEXT_BASE_ENTRY:
- vm_map_lock_read(map);
- if (!vm_map_lookup_entry(map, start, &tmp_entry)) {
- if ((entry = tmp_entry->vme_next) == vm_map_to_entry(map)) {
- vm_map_unlock_read(map);
- return(KERN_INVALID_ADDRESS);
+ /*
+ * Find the beginning of the region.
+ */
+ if (! vm_map_lookup_entry(map, src_start, &src_entry)) {
+ result = KERN_INVALID_ADDRESS;
+ break;
}
- } else {
- entry = tmp_entry;
- }
- *size = entry->vme_end - entry->vme_start;
- start = entry->vme_start;
- base_addr = start;
- baddr_start_delta = *address - start;
- base_next = entry->vme_end;
- base_entry = entry;
-
- while(entry->is_sub_map && recurse_count) {
- recurse_count--;
- vm_map_lock_read(entry->object.sub_map);
-
- if(entry == base_entry) {
- start = entry->offset;
- start += *address - entry->vme_start;
+ if (src_start < src_entry->vme_start ||
+ (mapped_size && src_start != src_entry->vme_start)) {
+ result = KERN_INVALID_ADDRESS;
+ break;
}
- submap = entry->object.sub_map;
- vm_map_unlock_read(map);
- map = submap;
+ tmp_size = size - mapped_size;
+ if (src_end > src_entry->vme_end)
+ tmp_size -= (src_end - src_entry->vme_end);
- if (!vm_map_lookup_entry(map, start, &tmp_entry)) {
- if ((entry = tmp_entry->vme_next)
- == vm_map_to_entry(map)) {
- vm_map_unlock_read(map);
- map = base_map;
- start = base_next;
- recurse_count = 0;
- *nesting_depth = 0;
- goto LOOKUP_NEXT_BASE_ENTRY;
- }
- } else {
- entry = tmp_entry;
+ entry_size = (vm_map_size_t)(src_entry->vme_end -
+ src_entry->vme_start);
- }
- if(start <= entry->vme_start) {
- vm_offset_t old_start = start;
- if(baddr_start_delta) {
- base_addr += (baddr_start_delta);
- *size -= baddr_start_delta;
- baddr_start_delta = 0;
- }
- if(base_next <=
- (base_addr += (entry->vme_start - start))) {
- vm_map_unlock_read(map);
- map = base_map;
- start = base_next;
- recurse_count = 0;
- *nesting_depth = 0;
- goto LOOKUP_NEXT_BASE_ENTRY;
- }
- *size -= entry->vme_start - start;
- if (*size > (entry->vme_end - entry->vme_start)) {
- *size = entry->vme_end - entry->vme_start;
- }
- start = 0;
+ if(src_entry->is_sub_map) {
+ vm_map_reference(VME_SUBMAP(src_entry));
+ object = VM_OBJECT_NULL;
} else {
- if(baddr_start_delta) {
- if((start - entry->vme_start)
- < baddr_start_delta) {
- base_addr += start - entry->vme_start;
- *size -= start - entry->vme_start;
- } else {
- base_addr += baddr_start_delta;
- *size += baddr_start_delta;
- }
- baddr_start_delta = 0;
- }
- base_addr += entry->vme_start;
- if(base_addr >= base_next) {
- vm_map_unlock_read(map);
- map = base_map;
- start = base_next;
- recurse_count = 0;
- *nesting_depth = 0;
- goto LOOKUP_NEXT_BASE_ENTRY;
+ object = VME_OBJECT(src_entry);
+ if (src_entry->iokit_acct) {
+ /*
+ * This entry uses "IOKit accounting".
+ */
+ } else if (object != VM_OBJECT_NULL &&
+ object->purgable != VM_PURGABLE_DENY) {
+ /*
+ * Purgeable objects have their own accounting:
+ * no pmap accounting for them.
+ */
+ assert(!src_entry->use_pmap);
+ } else {
+ /*
+ * Not IOKit or purgeable:
+ * must be accounted by pmap stats.
+ */
+ assert(src_entry->use_pmap);
}
- if (*size > (entry->vme_end - start))
- *size = entry->vme_end - start;
- start = entry->vme_start - start;
- }
+ if (object == VM_OBJECT_NULL) {
+ object = vm_object_allocate(entry_size);
+ VME_OFFSET_SET(src_entry, 0);
+ VME_OBJECT_SET(src_entry, object);
+ } else if (object->copy_strategy !=
+ MEMORY_OBJECT_COPY_SYMMETRIC) {
+ /*
+ * We are already using an asymmetric
+ * copy, and therefore we already have
+ * the right object.
+ */
+ assert(!src_entry->needs_copy);
+ } else if (src_entry->needs_copy || object->shadowed ||
+ (object->internal && !object->true_share &&
+ !src_entry->is_shared &&
+ object->vo_size > entry_size)) {
- start += entry->offset;
+ VME_OBJECT_SHADOW(src_entry, entry_size);
- }
- *nesting_depth -= recurse_count;
- if(entry != base_entry) {
- start = entry->vme_start + (start - entry->offset);
- }
+ if (!src_entry->needs_copy &&
+ (src_entry->protection & VM_PROT_WRITE)) {
+ vm_prot_t prot;
+ prot = src_entry->protection & ~VM_PROT_WRITE;
- submap_info->user_tag = entry->alias;
- submap_info->offset = entry->offset;
- submap_info->protection = entry->protection;
- submap_info->inheritance = entry->inheritance;
- submap_info->max_protection = entry->max_protection;
- submap_info->behavior = entry->behavior;
- submap_info->user_wired_count = entry->user_wired_count;
- submap_info->is_submap = entry->is_sub_map;
- submap_info->object_id = (vm_offset_t)entry->object.vm_object;
- *address = base_addr;
+ if (override_nx(map,
+ VME_ALIAS(src_entry))
+ && prot)
+ prot |= VM_PROT_EXECUTE;
+ if(map->mapped_in_other_pmaps) {
+ vm_object_pmap_protect(
+ VME_OBJECT(src_entry),
+ VME_OFFSET(src_entry),
+ entry_size,
+ PMAP_NULL,
+ src_entry->vme_start,
+ prot);
+ } else {
+ pmap_protect(vm_map_pmap(map),
+ src_entry->vme_start,
+ src_entry->vme_end,
+ prot);
+ }
+ }
- extended.pages_resident = 0;
- extended.pages_swapped_out = 0;
- extended.pages_shared_now_private = 0;
- extended.pages_dirtied = 0;
- extended.external_pager = 0;
- extended.shadow_depth = 0;
+ object = VME_OBJECT(src_entry);
+ src_entry->needs_copy = FALSE;
+ }
- if(!entry->is_sub_map) {
- vm_region_walk(entry, &extended, entry->offset,
- entry->vme_end - start, map, start);
- submap_info->share_mode = extended.share_mode;
- if (extended.external_pager && extended.ref_count == 2
- && extended.share_mode == SM_SHARED)
- submap_info->share_mode = SM_PRIVATE;
- submap_info->ref_count = extended.ref_count;
- } else {
- if(entry->use_pmap)
- submap_info->share_mode = SM_TRUESHARED;
- else
- submap_info->share_mode = SM_PRIVATE;
- submap_info->ref_count = entry->object.sub_map->ref_count;
- }
- submap_info->pages_resident = extended.pages_resident;
- submap_info->pages_swapped_out = extended.pages_swapped_out;
- submap_info->pages_shared_now_private =
- extended.pages_shared_now_private;
- submap_info->pages_dirtied = extended.pages_dirtied;
- submap_info->external_pager = extended.external_pager;
- submap_info->shadow_depth = extended.shadow_depth;
+ vm_object_lock(object);
+ vm_object_reference_locked(object); /* object ref. for new entry */
+ if (object->copy_strategy ==
+ MEMORY_OBJECT_COPY_SYMMETRIC) {
+ object->copy_strategy =
+ MEMORY_OBJECT_COPY_DELAY;
+ }
+ vm_object_unlock(object);
+ }
- vm_map_unlock_read(map);
- return(KERN_SUCCESS);
-}
+ offset = (VME_OFFSET(src_entry) +
+ (src_start - src_entry->vme_start));
-/*
- * TEMPORARYTEMPORARYTEMPORARYTEMPORARYTEMPORARYTEMPORARY
- * Goes away after regular vm_region_recurse function migrates to
- * 64 bits
- * vm_region_recurse: A form of vm_region which follows the
- * submaps in a target map
- *
- */
+ new_entry = _vm_map_entry_create(map_header, !map_header->entries_pageable);
+ vm_map_entry_copy(new_entry, src_entry);
+ if (new_entry->is_sub_map) {
+ /* clr address space specifics */
+ new_entry->use_pmap = FALSE;
+ }
-kern_return_t
-vm_region_recurse_64(
- vm_map_t map,
- vm_offset_t *address, /* IN/OUT */
- vm_size_t *size, /* OUT */
- natural_t *nesting_depth, /* IN/OUT */
- vm_region_recurse_info_t info, /* IN/OUT */
- mach_msg_type_number_t *count) /* IN/OUT */
-{
- vm_map_entry_t tmp_entry;
- register
- vm_map_entry_t entry;
- register
- vm_offset_t start;
-
- unsigned int recurse_count;
- vm_map_t submap;
- vm_map_t base_map;
- vm_map_entry_t base_entry;
- vm_offset_t base_next;
- vm_offset_t base_addr;
- vm_offset_t baddr_start_delta;
- vm_region_submap_info_64_t submap_info;
- vm_region_extended_info_data_t extended;
+ new_entry->map_aligned = FALSE;
- if (map == VM_MAP_NULL)
- return(KERN_INVALID_ARGUMENT);
+ new_entry->vme_start = map_address;
+ new_entry->vme_end = map_address + tmp_size;
+ assert(new_entry->vme_start < new_entry->vme_end);
+ new_entry->inheritance = inheritance;
+ VME_OFFSET_SET(new_entry, offset);
- submap_info = (vm_region_submap_info_64_t) info;
- *count = VM_REGION_SUBMAP_INFO_COUNT;
+ /*
+ * The new region has to be copied now if required.
+ */
+ RestartCopy:
+ if (!copy) {
+ /*
+ * Cannot allow an entry describing a JIT
+ * region to be shared across address spaces.
+ */
+ if (src_entry->used_for_jit == TRUE) {
+ result = KERN_INVALID_ARGUMENT;
+ break;
+ }
+ src_entry->is_shared = TRUE;
+ new_entry->is_shared = TRUE;
+ if (!(new_entry->is_sub_map))
+ new_entry->needs_copy = FALSE;
- if (*count < VM_REGION_SUBMAP_INFO_COUNT)
- return(KERN_INVALID_ARGUMENT);
+ } else if (src_entry->is_sub_map) {
+ /* make this a COW sub_map if not already */
+ assert(new_entry->wired_count == 0);
+ new_entry->needs_copy = TRUE;
+ object = VM_OBJECT_NULL;
+ } else if (src_entry->wired_count == 0 &&
+ vm_object_copy_quickly(&VME_OBJECT(new_entry),
+ VME_OFFSET(new_entry),
+ (new_entry->vme_end -
+ new_entry->vme_start),
+ &src_needs_copy,
+ &new_entry_needs_copy)) {
- start = *address;
- base_map = map;
- recurse_count = *nesting_depth;
-
-LOOKUP_NEXT_BASE_ENTRY:
- if (not_in_kdp)
- vm_map_lock_read(map);
- if (!vm_map_lookup_entry(map, start, &tmp_entry)) {
- if ((entry = tmp_entry->vme_next) == vm_map_to_entry(map)) {
- if (not_in_kdp)
- vm_map_unlock_read(map);
- return(KERN_INVALID_ADDRESS);
- }
- } else {
- entry = tmp_entry;
- }
- *size = entry->vme_end - entry->vme_start;
- start = entry->vme_start;
- base_addr = start;
- baddr_start_delta = *address - start;
- base_next = entry->vme_end;
- base_entry = entry;
+ new_entry->needs_copy = new_entry_needs_copy;
+ new_entry->is_shared = FALSE;
- while(entry->is_sub_map && recurse_count) {
- recurse_count--;
- if (not_in_kdp)
- vm_map_lock_read(entry->object.sub_map);
+ /*
+ * Handle copy_on_write semantics.
+ */
+ if (src_needs_copy && !src_entry->needs_copy) {
+ vm_prot_t prot;
+ prot = src_entry->protection & ~VM_PROT_WRITE;
- if(entry == base_entry) {
- start = entry->offset;
- start += *address - entry->vme_start;
- }
+ if (override_nx(map,
+ VME_ALIAS(src_entry))
+ && prot)
+ prot |= VM_PROT_EXECUTE;
- submap = entry->object.sub_map;
- if (not_in_kdp)
- vm_map_unlock_read(map);
- map = submap;
+ vm_object_pmap_protect(object,
+ offset,
+ entry_size,
+ ((src_entry->is_shared
+ || map->mapped_in_other_pmaps) ?
+ PMAP_NULL : map->pmap),
+ src_entry->vme_start,
+ prot);
- if (!vm_map_lookup_entry(map, start, &tmp_entry)) {
- if ((entry = tmp_entry->vme_next)
- == vm_map_to_entry(map)) {
- if (not_in_kdp)
- vm_map_unlock_read(map);
- map = base_map;
- start = base_next;
- recurse_count = 0;
- *nesting_depth = 0;
- goto LOOKUP_NEXT_BASE_ENTRY;
+ assert(src_entry->wired_count == 0);
+ src_entry->needs_copy = TRUE;
}
- } else {
- entry = tmp_entry;
+ /*
+ * Throw away the old object reference of the new entry.
+ */
+ vm_object_deallocate(object);
- }
- if(start <= entry->vme_start) {
- vm_offset_t old_start = start;
- if(baddr_start_delta) {
- base_addr += (baddr_start_delta);
- *size -= baddr_start_delta;
- baddr_start_delta = 0;
- }
- if(base_next <=
- (base_addr += (entry->vme_start - start))) {
- if (not_in_kdp)
- vm_map_unlock_read(map);
- map = base_map;
- start = base_next;
- recurse_count = 0;
- *nesting_depth = 0;
- goto LOOKUP_NEXT_BASE_ENTRY;
- }
- *size -= entry->vme_start - start;
- if (*size > (entry->vme_end - entry->vme_start)) {
- *size = entry->vme_end - entry->vme_start;
- }
- start = 0;
} else {
- if(baddr_start_delta) {
- if((start - entry->vme_start)
- < baddr_start_delta) {
- base_addr += start - entry->vme_start;
- *size -= start - entry->vme_start;
- } else {
- base_addr += baddr_start_delta;
- *size += baddr_start_delta;
+ new_entry->is_shared = FALSE;
+
+ /*
+ * The map can be safely unlocked since we
+ * already hold a reference on the object.
+ *
+ * Record the timestamp of the map for later
+ * verification, and unlock the map.
+ */
+ version.main_timestamp = map->timestamp;
+ vm_map_unlock(map); /* Increments timestamp once! */
+
+ /*
+ * Perform the copy.
+ */
+ if (src_entry->wired_count > 0) {
+ vm_object_lock(object);
+ result = vm_object_copy_slowly(
+ object,
+ offset,
+ entry_size,
+ THREAD_UNINT,
+ &VME_OBJECT(new_entry));
+
+ VME_OFFSET_SET(new_entry, 0);
+ new_entry->needs_copy = FALSE;
+ } else {
+ vm_object_offset_t new_offset;
+
+ new_offset = VME_OFFSET(new_entry);
+ result = vm_object_copy_strategically(
+ object,
+ offset,
+ entry_size,
+ &VME_OBJECT(new_entry),
+ &new_offset,
+ &new_entry_needs_copy);
+ if (new_offset != VME_OFFSET(new_entry)) {
+ VME_OFFSET_SET(new_entry, new_offset);
}
- baddr_start_delta = 0;
+
+ new_entry->needs_copy = new_entry_needs_copy;
}
- base_addr += entry->vme_start;
- if(base_addr >= base_next) {
- if (not_in_kdp)
- vm_map_unlock_read(map);
- map = base_map;
- start = base_next;
- recurse_count = 0;
- *nesting_depth = 0;
- goto LOOKUP_NEXT_BASE_ENTRY;
+
+ /*
+ * Throw away the old object reference of the new entry.
+ */
+ vm_object_deallocate(object);
+
+ if (result != KERN_SUCCESS &&
+ result != KERN_MEMORY_RESTART_COPY) {
+ _vm_map_entry_dispose(map_header, new_entry);
+ break;
}
- if (*size > (entry->vme_end - start))
- *size = entry->vme_end - start;
- start = entry->vme_start - start;
- }
+ /*
+ * Verify that the map has not substantially
+ * changed while the copy was being made.
+ */
- start += entry->offset;
+ vm_map_lock(map);
+ if (version.main_timestamp + 1 != map->timestamp) {
+ /*
+ * Simple version comparison failed.
+ *
+ * Retry the lookup and verify that the
+ * same object/offset are still present.
+ */
+ vm_object_deallocate(VME_OBJECT(new_entry));
+ _vm_map_entry_dispose(map_header, new_entry);
+ if (result == KERN_MEMORY_RESTART_COPY)
+ result = KERN_SUCCESS;
+ continue;
+ }
- }
- *nesting_depth -= recurse_count;
- if(entry != base_entry) {
- start = entry->vme_start + (start - entry->offset);
- }
+ if (result == KERN_MEMORY_RESTART_COPY) {
+ vm_object_reference(object);
+ goto RestartCopy;
+ }
+ }
+ _vm_map_store_entry_link(map_header,
+ map_header->links.prev, new_entry);
- submap_info->user_tag = entry->alias;
- submap_info->offset = entry->offset;
- submap_info->protection = entry->protection;
- submap_info->inheritance = entry->inheritance;
- submap_info->max_protection = entry->max_protection;
- submap_info->behavior = entry->behavior;
- submap_info->user_wired_count = entry->user_wired_count;
- submap_info->is_submap = entry->is_sub_map;
- submap_info->object_id = (vm_offset_t)entry->object.vm_object;
- *address = base_addr;
+ /*Protections for submap mapping are irrelevant here*/
+ if( !src_entry->is_sub_map ) {
+ *cur_protection &= src_entry->protection;
+ *max_protection &= src_entry->max_protection;
+ }
+ map_address += tmp_size;
+ mapped_size += tmp_size;
+ src_start += tmp_size;
+ } /* end while */
- extended.pages_resident = 0;
- extended.pages_swapped_out = 0;
- extended.pages_shared_now_private = 0;
- extended.pages_dirtied = 0;
- extended.external_pager = 0;
- extended.shadow_depth = 0;
-
- if (not_in_kdp)
- if(!entry->is_sub_map) {
- vm_region_walk(entry, &extended, entry->offset,
- entry->vme_end - start, map, start);
- submap_info->share_mode = extended.share_mode;
- if (extended.external_pager && extended.ref_count == 2
- && extended.share_mode == SM_SHARED)
- submap_info->share_mode = SM_PRIVATE;
- submap_info->ref_count = extended.ref_count;
- } else {
- if(entry->use_pmap)
- submap_info->share_mode = SM_TRUESHARED;
- else
- submap_info->share_mode = SM_PRIVATE;
- submap_info->ref_count = entry->object.sub_map->ref_count;
- }
-
- submap_info->pages_resident = extended.pages_resident;
- submap_info->pages_swapped_out = extended.pages_swapped_out;
- submap_info->pages_shared_now_private =
- extended.pages_shared_now_private;
- submap_info->pages_dirtied = extended.pages_dirtied;
- submap_info->external_pager = extended.external_pager;
- submap_info->shadow_depth = extended.shadow_depth;
- if (not_in_kdp)
- vm_map_unlock_read(map);
- return(KERN_SUCCESS);
+ vm_map_unlock(map);
+ if (result != KERN_SUCCESS) {
+ /*
+ * Free all allocated elements.
+ */
+ for (src_entry = map_header->links.next;
+ src_entry != (struct vm_map_entry *)&map_header->links;
+ src_entry = new_entry) {
+ new_entry = src_entry->vme_next;
+ _vm_map_store_entry_unlink(map_header, src_entry);
+ if (src_entry->is_sub_map) {
+ vm_map_deallocate(VME_SUBMAP(src_entry));
+ } else {
+ vm_object_deallocate(VME_OBJECT(src_entry));
+ }
+ _vm_map_entry_dispose(map_header, src_entry);
+ }
+ }
+ return result;
}
-
/*
- * TEMPORARYTEMPORARYTEMPORARYTEMPORARYTEMPORARYTEMPORARY
- * Goes away after regular vm_region function migrates to
- * 64 bits
+ * Routine: vm_remap
+ *
+ * Map portion of a task's address space.
+ * Mapped region must not overlap more than
+ * one vm memory object. Protections and
+ * inheritance attributes remain the same
+ * as in the original task and are out parameters.
+ * Source and Target task can be identical
+ * Other attributes are identical as for vm_map()
*/
-
-
kern_return_t
-vm_region_64(
- vm_map_t map,
- vm_offset_t *address, /* IN/OUT */
- vm_size_t *size, /* OUT */
- vm_region_flavor_t flavor, /* IN */
- vm_region_info_t info, /* OUT */
- mach_msg_type_number_t *count, /* IN/OUT */
- ipc_port_t *object_name) /* OUT */
+vm_map_remap(
+ vm_map_t target_map,
+ vm_map_address_t *address,
+ vm_map_size_t size,
+ vm_map_offset_t mask,
+ int flags,
+ vm_map_t src_map,
+ vm_map_offset_t memory_address,
+ boolean_t copy,
+ vm_prot_t *cur_protection,
+ vm_prot_t *max_protection,
+ vm_inherit_t inheritance)
{
- vm_map_entry_t tmp_entry;
- register
+ kern_return_t result;
vm_map_entry_t entry;
- register
- vm_offset_t start;
- vm_region_basic_info_64_t basic;
- vm_region_extended_info_t extended;
- vm_region_top_info_t top;
- vm_region_object_info_64_t object_info_64;
+ vm_map_entry_t insp_entry = VM_MAP_ENTRY_NULL;
+ vm_map_entry_t new_entry;
+ struct vm_map_header map_header;
+ vm_map_offset_t offset_in_mapping;
- if (map == VM_MAP_NULL)
- return(KERN_INVALID_ARGUMENT);
+ if (target_map == VM_MAP_NULL)
+ return KERN_INVALID_ARGUMENT;
- switch (flavor) {
-
- case VM_REGION_BASIC_INFO:
- {
- if (*count < VM_REGION_BASIC_INFO_COUNT)
- return(KERN_INVALID_ARGUMENT);
+ switch (inheritance) {
+ case VM_INHERIT_NONE:
+ case VM_INHERIT_COPY:
+ case VM_INHERIT_SHARE:
+ if (size != 0 && src_map != VM_MAP_NULL)
+ break;
+ /*FALL THRU*/
+ default:
+ return KERN_INVALID_ARGUMENT;
+ }
- basic = (vm_region_basic_info_64_t) info;
- *count = VM_REGION_BASIC_INFO_COUNT;
+ /*
+ * If the user is requesting that we return the address of the
+ * first byte of the data (rather than the base of the page),
+ * then we use different rounding semantics: specifically,
+ * we assume that (memory_address, size) describes a region
+ * all of whose pages we must cover, rather than a base to be truncated
+ * down and a size to be added to that base. So we figure out
+ * the highest page that the requested region includes and make
+ * sure that the size will cover it.
+ *
+ * The key example we're worried about it is of the form:
+ *
+ * memory_address = 0x1ff0, size = 0x20
+ *
+ * With the old semantics, we round down the memory_address to 0x1000
+ * and round up the size to 0x1000, resulting in our covering *only*
+ * page 0x1000. With the new semantics, we'd realize that the region covers
+ * 0x1ff0-0x2010, and compute a size of 0x2000. Thus, we cover both page
+ * 0x1000 and page 0x2000 in the region we remap.
+ */
+ if ((flags & VM_FLAGS_RETURN_DATA_ADDR) != 0) {
+ offset_in_mapping = memory_address - vm_map_trunc_page(memory_address, PAGE_MASK);
+ size = vm_map_round_page(memory_address + size - vm_map_trunc_page(memory_address, PAGE_MASK), PAGE_MASK);
+ } else {
+ size = vm_map_round_page(size, PAGE_MASK);
+ }
- vm_map_lock_read(map);
+ result = vm_map_remap_extract(src_map, memory_address,
+ size, copy, &map_header,
+ cur_protection,
+ max_protection,
+ inheritance,
+ target_map->hdr.entries_pageable);
- start = *address;
- if (!vm_map_lookup_entry(map, start, &tmp_entry)) {
- if ((entry = tmp_entry->vme_next) == vm_map_to_entry(map)) {
- vm_map_unlock_read(map);
- return(KERN_INVALID_ADDRESS);
- }
- } else {
- entry = tmp_entry;
- }
-
- start = entry->vme_start;
-
- basic->offset = entry->offset;
- basic->protection = entry->protection;
- basic->inheritance = entry->inheritance;
- basic->max_protection = entry->max_protection;
- basic->behavior = entry->behavior;
- basic->user_wired_count = entry->user_wired_count;
- basic->reserved = entry->is_sub_map;
- *address = start;
- *size = (entry->vme_end - start);
-
- if (object_name) *object_name = IP_NULL;
- if (entry->is_sub_map) {
- basic->shared = FALSE;
- } else {
- basic->shared = entry->is_shared;
- }
-
- vm_map_unlock_read(map);
- return(KERN_SUCCESS);
+ if (result != KERN_SUCCESS) {
+ return result;
}
- case VM_REGION_EXTENDED_INFO:
- {
-
- if (*count < VM_REGION_EXTENDED_INFO_COUNT)
- return(KERN_INVALID_ARGUMENT);
-
- extended = (vm_region_extended_info_t) info;
- *count = VM_REGION_EXTENDED_INFO_COUNT;
- vm_map_lock_read(map);
+ /*
+ * Allocate/check a range of free virtual address
+ * space for the target
+ */
+ *address = vm_map_trunc_page(*address,
+ VM_MAP_PAGE_MASK(target_map));
+ vm_map_lock(target_map);
+ result = vm_map_remap_range_allocate(target_map, address, size,
+ mask, flags, &insp_entry);
- start = *address;
- if (!vm_map_lookup_entry(map, start, &tmp_entry)) {
- if ((entry = tmp_entry->vme_next) == vm_map_to_entry(map)) {
- vm_map_unlock_read(map);
- return(KERN_INVALID_ADDRESS);
+ for (entry = map_header.links.next;
+ entry != (struct vm_map_entry *)&map_header.links;
+ entry = new_entry) {
+ new_entry = entry->vme_next;
+ _vm_map_store_entry_unlink(&map_header, entry);
+ if (result == KERN_SUCCESS) {
+ if (flags & VM_FLAGS_RESILIENT_CODESIGN) {
+ /* no codesigning -> read-only access */
+ assert(!entry->used_for_jit);
+ entry->max_protection = VM_PROT_READ;
+ entry->protection = VM_PROT_READ;
+ entry->vme_resilient_codesign = TRUE;
+ }
+ entry->vme_start += *address;
+ entry->vme_end += *address;
+ assert(!entry->map_aligned);
+ vm_map_store_entry_link(target_map, insp_entry, entry);
+ insp_entry = entry;
+ } else {
+ if (!entry->is_sub_map) {
+ vm_object_deallocate(VME_OBJECT(entry));
+ } else {
+ vm_map_deallocate(VME_SUBMAP(entry));
+ }
+ _vm_map_entry_dispose(&map_header, entry);
}
- } else {
- entry = tmp_entry;
- }
- start = entry->vme_start;
-
- extended->protection = entry->protection;
- extended->user_tag = entry->alias;
- extended->pages_resident = 0;
- extended->pages_swapped_out = 0;
- extended->pages_shared_now_private = 0;
- extended->pages_dirtied = 0;
- extended->external_pager = 0;
- extended->shadow_depth = 0;
-
- vm_region_walk(entry, extended, entry->offset, entry->vme_end - start, map, start);
-
- if (extended->external_pager && extended->ref_count == 2 && extended->share_mode == SM_SHARED)
- extended->share_mode = SM_PRIVATE;
+ }
- if (object_name)
- *object_name = IP_NULL;
- *address = start;
- *size = (entry->vme_end - start);
+ if (flags & VM_FLAGS_RESILIENT_CODESIGN) {
+ *cur_protection = VM_PROT_READ;
+ *max_protection = VM_PROT_READ;
+ }
- vm_map_unlock_read(map);
- return(KERN_SUCCESS);
+ if( target_map->disable_vmentry_reuse == TRUE) {
+ if( target_map->highest_entry_end < insp_entry->vme_end ){
+ target_map->highest_entry_end = insp_entry->vme_end;
+ }
}
- case VM_REGION_TOP_INFO:
- {
- if (*count < VM_REGION_TOP_INFO_COUNT)
- return(KERN_INVALID_ARGUMENT);
+ if (result == KERN_SUCCESS) {
+ target_map->size += size;
+ SAVE_HINT_MAP_WRITE(target_map, insp_entry);
+ }
+ vm_map_unlock(target_map);
- top = (vm_region_top_info_t) info;
- *count = VM_REGION_TOP_INFO_COUNT;
+ if (result == KERN_SUCCESS && target_map->wiring_required)
+ result = vm_map_wire(target_map, *address,
+ *address + size, *cur_protection | VM_PROT_MEMORY_TAG_MAKE(VM_KERN_MEMORY_MLOCK),
+ TRUE);
- vm_map_lock_read(map);
+ /*
+ * If requested, return the address of the data pointed to by the
+ * request, rather than the base of the resulting page.
+ */
+ if ((flags & VM_FLAGS_RETURN_DATA_ADDR) != 0) {
+ *address += offset_in_mapping;
+ }
- start = *address;
- if (!vm_map_lookup_entry(map, start, &tmp_entry)) {
- if ((entry = tmp_entry->vme_next) == vm_map_to_entry(map)) {
- vm_map_unlock_read(map);
- return(KERN_INVALID_ADDRESS);
- }
- } else {
- entry = tmp_entry;
+ return result;
+}
- }
- start = entry->vme_start;
+/*
+ * Routine: vm_map_remap_range_allocate
+ *
+ * Description:
+ * Allocate a range in the specified virtual address map.
+ * returns the address and the map entry just before the allocated
+ * range
+ *
+ * Map must be locked.
+ */
- top->private_pages_resident = 0;
- top->shared_pages_resident = 0;
+static kern_return_t
+vm_map_remap_range_allocate(
+ vm_map_t map,
+ vm_map_address_t *address, /* IN/OUT */
+ vm_map_size_t size,
+ vm_map_offset_t mask,
+ int flags,
+ vm_map_entry_t *map_entry) /* OUT */
+{
+ vm_map_entry_t entry;
+ vm_map_offset_t start;
+ vm_map_offset_t end;
+ kern_return_t kr;
+ vm_map_entry_t hole_entry;
- vm_region_top_walk(entry, top);
+StartAgain: ;
- if (object_name)
- *object_name = IP_NULL;
- *address = start;
- *size = (entry->vme_end - start);
+ start = *address;
- vm_map_unlock_read(map);
- return(KERN_SUCCESS);
- }
- case VM_REGION_OBJECT_INFO_64:
+ if (flags & VM_FLAGS_ANYWHERE)
{
- if (*count < VM_REGION_OBJECT_INFO_COUNT_64)
- return(KERN_INVALID_ARGUMENT);
+ /*
+ * Calculate the first possible address.
+ */
- object_info_64 = (vm_region_object_info_64_t) info;
- *count = VM_REGION_OBJECT_INFO_COUNT_64;
+ if (start < map->min_offset)
+ start = map->min_offset;
+ if (start > map->max_offset)
+ return(KERN_NO_SPACE);
+
+ /*
+ * Look for the first possible address;
+ * if there's already something at this
+ * address, we have to start after it.
+ */
- vm_map_lock_read(map);
+ if( map->disable_vmentry_reuse == TRUE) {
+ VM_MAP_HIGHEST_ENTRY(map, entry, start);
+ } else {
- start = *address;
- if (!vm_map_lookup_entry(map, start, &tmp_entry)) {
- if ((entry = tmp_entry->vme_next) == vm_map_to_entry(map)) {
- vm_map_unlock_read(map);
- return(KERN_INVALID_ADDRESS);
- }
- } else {
- entry = tmp_entry;
- }
-
- start = entry->vme_start;
-
- object_info_64->offset = entry->offset;
- object_info_64->protection = entry->protection;
- object_info_64->inheritance = entry->inheritance;
- object_info_64->max_protection = entry->max_protection;
- object_info_64->behavior = entry->behavior;
- object_info_64->user_wired_count = entry->user_wired_count;
- object_info_64->is_sub_map = entry->is_sub_map;
- *address = start;
- *size = (entry->vme_end - start);
-
- if (object_name) *object_name = IP_NULL;
- if (entry->is_sub_map) {
- object_info_64->shared = FALSE;
- object_info_64->object_id = 0;
- } else {
- object_info_64->shared = entry->is_shared;
- object_info_64->object_id =
- (vm_offset_t) entry->object.vm_object;
- }
-
- vm_map_unlock_read(map);
- return(KERN_SUCCESS);
- }
- default:
- return(KERN_INVALID_ARGUMENT);
- }
-}
+ if (map->holelistenabled) {
+ hole_entry = (vm_map_entry_t)map->holes_list;
-void
-vm_region_top_walk(
- vm_map_entry_t entry,
- vm_region_top_info_t top)
-{
- register struct vm_object *obj, *tmp_obj;
- register int ref_count;
+ if (hole_entry == NULL) {
+ /*
+ * No more space in the map?
+ */
+ return(KERN_NO_SPACE);
+ } else {
- if (entry->object.vm_object == 0 || entry->is_sub_map) {
- top->share_mode = SM_EMPTY;
- top->ref_count = 0;
- top->obj_id = 0;
- return;
- }
- {
- obj = entry->object.vm_object;
+ boolean_t found_hole = FALSE;
- vm_object_lock(obj);
+ do {
+ if (hole_entry->vme_start >= start) {
+ start = hole_entry->vme_start;
+ found_hole = TRUE;
+ break;
+ }
- if ((ref_count = obj->ref_count) > 1 && obj->paging_in_progress)
- ref_count--;
+ if (hole_entry->vme_end > start) {
+ found_hole = TRUE;
+ break;
+ }
+ hole_entry = hole_entry->vme_next;
- if (obj->shadow) {
- if (ref_count == 1)
- top->private_pages_resident = obj->resident_page_count;
- else
- top->shared_pages_resident = obj->resident_page_count;
- top->ref_count = ref_count;
- top->share_mode = SM_COW;
-
- while (tmp_obj = obj->shadow) {
- vm_object_lock(tmp_obj);
- vm_object_unlock(obj);
- obj = tmp_obj;
+ } while (hole_entry != (vm_map_entry_t) map->holes_list);
- if ((ref_count = obj->ref_count) > 1 && obj->paging_in_progress)
- ref_count--;
+ if (found_hole == FALSE) {
+ return (KERN_NO_SPACE);
+ }
- top->shared_pages_resident += obj->resident_page_count;
- top->ref_count += ref_count - 1;
- }
- } else {
- if (entry->needs_copy) {
- top->share_mode = SM_COW;
- top->shared_pages_resident = obj->resident_page_count;
- } else {
- if (ref_count == 1 ||
- (ref_count == 2 && !(obj->pager_trusted) && !(obj->internal))) {
- top->share_mode = SM_PRIVATE;
- top->private_pages_resident = obj->resident_page_count;
- } else {
- top->share_mode = SM_SHARED;
- top->shared_pages_resident = obj->resident_page_count;
- }
+ entry = hole_entry;
+ }
+ } else {
+ assert(first_free_is_valid(map));
+ if (start == map->min_offset) {
+ if ((entry = map->first_free) != vm_map_to_entry(map))
+ start = entry->vme_end;
+ } else {
+ vm_map_entry_t tmp_entry;
+ if (vm_map_lookup_entry(map, start, &tmp_entry))
+ start = tmp_entry->vme_end;
+ entry = tmp_entry;
+ }
+ }
+ start = vm_map_round_page(start,
+ VM_MAP_PAGE_MASK(map));
}
- top->ref_count = ref_count;
- }
- top->obj_id = (int)obj;
-
- vm_object_unlock(obj);
- }
-}
-
-void
-vm_region_walk(
- vm_map_entry_t entry,
- vm_region_extended_info_t extended,
- vm_object_offset_t offset,
- vm_offset_t range,
- vm_map_t map,
- vm_offset_t va)
-{
- register struct vm_object *obj, *tmp_obj;
- register vm_offset_t last_offset;
- register int i;
- register int ref_count;
- void vm_region_look_for_page();
-
- if ((entry->object.vm_object == 0) ||
- (entry->is_sub_map) ||
- (entry->object.vm_object->phys_contiguous)) {
- extended->share_mode = SM_EMPTY;
- extended->ref_count = 0;
- return;
- }
- {
- obj = entry->object.vm_object;
+
+ /*
+ * In any case, the "entry" always precedes
+ * the proposed new region throughout the
+ * loop:
+ */
- vm_object_lock(obj);
+ while (TRUE) {
+ register vm_map_entry_t next;
- if ((ref_count = obj->ref_count) > 1 && obj->paging_in_progress)
- ref_count--;
+ /*
+ * Find the end of the proposed new region.
+ * Be sure we didn't go beyond the end, or
+ * wrap around the address.
+ */
- for (last_offset = offset + range; offset < last_offset; offset += PAGE_SIZE_64, va += PAGE_SIZE)
- vm_region_look_for_page(obj, extended, offset, ref_count, 0, map, va);
+ end = ((start + mask) & ~mask);
+ end = vm_map_round_page(end,
+ VM_MAP_PAGE_MASK(map));
+ if (end < start)
+ return(KERN_NO_SPACE);
+ start = end;
+ end += size;
- if (extended->shadow_depth || entry->needs_copy)
- extended->share_mode = SM_COW;
- else {
- if (ref_count == 1)
- extended->share_mode = SM_PRIVATE;
- else {
- if (obj->true_share)
- extended->share_mode = SM_TRUESHARED;
- else
- extended->share_mode = SM_SHARED;
- }
- }
- extended->ref_count = ref_count - extended->shadow_depth;
-
- for (i = 0; i < extended->shadow_depth; i++) {
- if ((tmp_obj = obj->shadow) == 0)
- break;
- vm_object_lock(tmp_obj);
- vm_object_unlock(obj);
+ if ((end > map->max_offset) || (end < start)) {
+ if (map->wait_for_space) {
+ if (size <= (map->max_offset -
+ map->min_offset)) {
+ assert_wait((event_t) map, THREAD_INTERRUPTIBLE);
+ vm_map_unlock(map);
+ thread_block(THREAD_CONTINUE_NULL);
+ vm_map_lock(map);
+ goto StartAgain;
+ }
+ }
+
+ return(KERN_NO_SPACE);
+ }
- if ((ref_count = tmp_obj->ref_count) > 1 && tmp_obj->paging_in_progress)
- ref_count--;
+ next = entry->vme_next;
- extended->ref_count += ref_count;
- obj = tmp_obj;
- }
- vm_object_unlock(obj);
+ if (map->holelistenabled) {
+ if (entry->vme_end >= end)
+ break;
+ } else {
+ /*
+ * If there are no more entries, we must win.
+ *
+ * OR
+ *
+ * If there is another entry, it must be
+ * after the end of the potential new region.
+ */
- if (extended->share_mode == SM_SHARED) {
- register vm_map_entry_t cur;
- register vm_map_entry_t last;
- int my_refs;
+ if (next == vm_map_to_entry(map))
+ break;
- obj = entry->object.vm_object;
- last = vm_map_to_entry(map);
- my_refs = 0;
+ if (next->vme_start >= end)
+ break;
+ }
- if ((ref_count = obj->ref_count) > 1 && obj->paging_in_progress)
- ref_count--;
- for (cur = vm_map_first_entry(map); cur != last; cur = cur->vme_next)
- my_refs += vm_region_count_obj_refs(cur, obj);
+ /*
+ * Didn't fit -- move to the next entry.
+ */
- if (my_refs == ref_count)
- extended->share_mode = SM_PRIVATE_ALIASED;
- else if (my_refs > 1)
- extended->share_mode = SM_SHARED_ALIASED;
- }
- }
-}
+ entry = next;
+ if (map->holelistenabled) {
+ if (entry == (vm_map_entry_t) map->holes_list) {
+ /*
+ * Wrapped around
+ */
+ return(KERN_NO_SPACE);
+ }
+ start = entry->vme_start;
+ } else {
+ start = entry->vme_end;
+ }
+ }
-/* object is locked on entry and locked on return */
+ if (map->holelistenabled) {
+ if (vm_map_lookup_entry(map, entry->vme_start, &entry)) {
+ panic("Found an existing entry (%p) instead of potential hole at address: 0x%llx.\n", entry, (unsigned long long)entry->vme_start);
+ }
+ }
-void
-vm_region_look_for_page(
- vm_object_t object,
- vm_region_extended_info_t extended,
- vm_object_offset_t offset,
- int max_refcnt,
- int depth,
- vm_map_t map,
- vm_offset_t va)
-{
- register vm_page_t p;
- register vm_object_t shadow;
- register int ref_count;
- vm_object_t caller_object;
-
- shadow = object->shadow;
- caller_object = object;
+ *address = start;
+ } else {
+ vm_map_entry_t temp_entry;
- while (TRUE) {
+ /*
+ * Verify that:
+ * the address doesn't itself violate
+ * the mask requirement.
+ */
- if ( !(object->pager_trusted) && !(object->internal))
- extended->external_pager = 1;
+ if ((start & mask) != 0)
+ return(KERN_NO_SPACE);
- if ((p = vm_page_lookup(object, offset)) != VM_PAGE_NULL) {
- if (shadow && (max_refcnt == 1))
- extended->pages_shared_now_private++;
- if (!p->fictitious &&
- (p->dirty || pmap_is_modified(p->phys_page)))
- extended->pages_dirtied++;
- extended->pages_resident++;
+ /*
+ * ... the address is within bounds
+ */
- if(object != caller_object)
- vm_object_unlock(object);
+ end = start + size;
- return;
+ if ((start < map->min_offset) ||
+ (end > map->max_offset) ||
+ (start >= end)) {
+ return(KERN_INVALID_ADDRESS);
}
- if (object->existence_map) {
- if (vm_external_state_get(object->existence_map, offset) == VM_EXTERNAL_STATE_EXISTS) {
-
- extended->pages_swapped_out++;
- if(object != caller_object)
- vm_object_unlock(object);
+ /*
+ * If we're asked to overwrite whatever was mapped in that
+ * range, first deallocate that range.
+ */
+ if (flags & VM_FLAGS_OVERWRITE) {
+ vm_map_t zap_map;
- return;
- }
+ /*
+ * We use a "zap_map" to avoid having to unlock
+ * the "map" in vm_map_delete(), which would compromise
+ * the atomicity of the "deallocate" and then "remap"
+ * combination.
+ */
+ zap_map = vm_map_create(PMAP_NULL,
+ start,
+ end,
+ map->hdr.entries_pageable);
+ if (zap_map == VM_MAP_NULL) {
+ return KERN_RESOURCE_SHORTAGE;
+ }
+ vm_map_set_page_shift(zap_map, VM_MAP_PAGE_SHIFT(map));
+ vm_map_disable_hole_optimization(zap_map);
+
+ kr = vm_map_delete(map, start, end,
+ (VM_MAP_REMOVE_SAVE_ENTRIES |
+ VM_MAP_REMOVE_NO_MAP_ALIGN),
+ zap_map);
+ if (kr == KERN_SUCCESS) {
+ vm_map_destroy(zap_map,
+ VM_MAP_REMOVE_NO_PMAP_CLEANUP);
+ zap_map = VM_MAP_NULL;
+ }
}
- if (shadow) {
- vm_object_lock(shadow);
- if ((ref_count = shadow->ref_count) > 1 && shadow->paging_in_progress)
- ref_count--;
+ /*
+ * ... the starting address isn't allocated
+ */
- if (++depth > extended->shadow_depth)
- extended->shadow_depth = depth;
+ if (vm_map_lookup_entry(map, start, &temp_entry))
+ return(KERN_NO_SPACE);
- if (ref_count > max_refcnt)
- max_refcnt = ref_count;
-
- if(object != caller_object)
- vm_object_unlock(object);
+ entry = temp_entry;
- object = shadow;
- shadow = object->shadow;
- offset = offset + object->shadow_offset;
- continue;
- }
- if(object != caller_object)
- vm_object_unlock(object);
- break;
+ /*
+ * ... the next region doesn't overlap the
+ * end point.
+ */
+
+ if ((entry->vme_next != vm_map_to_entry(map)) &&
+ (entry->vme_next->vme_start < end))
+ return(KERN_NO_SPACE);
}
+ *map_entry = entry;
+ return(KERN_SUCCESS);
}
+/*
+ * vm_map_switch:
+ *
+ * Set the address map for the current thread to the specified map
+ */
-vm_region_count_obj_refs(
- vm_map_entry_t entry,
- vm_object_t object)
+vm_map_t
+vm_map_switch(
+ vm_map_t map)
{
- register int ref_count;
- register vm_object_t chk_obj;
- register vm_object_t tmp_obj;
-
- if (entry->object.vm_object == 0)
- return(0);
-
- if (entry->is_sub_map)
- return(0);
- else {
- ref_count = 0;
+ int mycpu;
+ thread_t thread = current_thread();
+ vm_map_t oldmap = thread->map;
- chk_obj = entry->object.vm_object;
- vm_object_lock(chk_obj);
+ mp_disable_preemption();
+ mycpu = cpu_number();
- while (chk_obj) {
- if (chk_obj == object)
- ref_count++;
- if (tmp_obj = chk_obj->shadow)
- vm_object_lock(tmp_obj);
- vm_object_unlock(chk_obj);
+ /*
+ * Deactivate the current map and activate the requested map
+ */
+ PMAP_SWITCH_USER(thread, map, mycpu);
- chk_obj = tmp_obj;
- }
- }
- return(ref_count);
+ mp_enable_preemption();
+ return(oldmap);
}
/*
- * Routine: vm_map_simplify
+ * Routine: vm_map_write_user
*
* Description:
- * Attempt to simplify the map representation in
- * the vicinity of the given starting address.
- * Note:
- * This routine is intended primarily to keep the
- * kernel maps more compact -- they generally don't
- * benefit from the "expand a map entry" technology
- * at allocation time because the adjacent entry
- * is often wired down.
+ * Copy out data from a kernel space into space in the
+ * destination map. The space must already exist in the
+ * destination map.
+ * NOTE: This routine should only be called by threads
+ * which can block on a page fault. i.e. kernel mode user
+ * threads.
+ *
*/
-void
-vm_map_simplify_entry(
- vm_map_t map,
- vm_map_entry_t this_entry)
+kern_return_t
+vm_map_write_user(
+ vm_map_t map,
+ void *src_p,
+ vm_map_address_t dst_addr,
+ vm_size_t size)
{
- vm_map_entry_t prev_entry;
-
- prev_entry = this_entry->vme_prev;
-
- if ((this_entry != vm_map_to_entry(map)) &&
- (prev_entry != vm_map_to_entry(map)) &&
-
- (prev_entry->vme_end == this_entry->vme_start) &&
+ kern_return_t kr = KERN_SUCCESS;
- (prev_entry->is_sub_map == FALSE) &&
- (this_entry->is_sub_map == FALSE) &&
-
- (prev_entry->object.vm_object == this_entry->object.vm_object) &&
- ((prev_entry->offset + (prev_entry->vme_end -
- prev_entry->vme_start))
- == this_entry->offset) &&
-
- (prev_entry->inheritance == this_entry->inheritance) &&
- (prev_entry->protection == this_entry->protection) &&
- (prev_entry->max_protection == this_entry->max_protection) &&
- (prev_entry->behavior == this_entry->behavior) &&
- (prev_entry->alias == this_entry->alias) &&
- (prev_entry->wired_count == this_entry->wired_count) &&
- (prev_entry->user_wired_count == this_entry->user_wired_count) &&
- (prev_entry->needs_copy == this_entry->needs_copy) &&
-
- (prev_entry->use_pmap == FALSE) &&
- (this_entry->use_pmap == FALSE) &&
- (prev_entry->in_transition == FALSE) &&
- (this_entry->in_transition == FALSE) &&
- (prev_entry->needs_wakeup == FALSE) &&
- (this_entry->needs_wakeup == FALSE) &&
- (prev_entry->is_shared == FALSE) &&
- (this_entry->is_shared == FALSE)
- ) {
- _vm_map_entry_unlink(&map->hdr, prev_entry);
- this_entry->vme_start = prev_entry->vme_start;
- this_entry->offset = prev_entry->offset;
- vm_object_deallocate(prev_entry->object.vm_object);
- vm_map_entry_dispose(map, prev_entry);
- SAVE_HINT(map, this_entry);
- counter(c_vm_map_simplified++);
- }
- counter(c_vm_map_simplify_entry_called++);
-}
+ if(current_map() == map) {
+ if (copyout(src_p, dst_addr, size)) {
+ kr = KERN_INVALID_ADDRESS;
+ }
+ } else {
+ vm_map_t oldmap;
-void
-vm_map_simplify(
- vm_map_t map,
- vm_offset_t start)
-{
- vm_map_entry_t this_entry;
+ /* take on the identity of the target map while doing */
+ /* the transfer */
- vm_map_lock(map);
- if (vm_map_lookup_entry(map, start, &this_entry)) {
- vm_map_simplify_entry(map, this_entry);
- vm_map_simplify_entry(map, this_entry->vme_next);
+ vm_map_reference(map);
+ oldmap = vm_map_switch(map);
+ if (copyout(src_p, dst_addr, size)) {
+ kr = KERN_INVALID_ADDRESS;
+ }
+ vm_map_switch(oldmap);
+ vm_map_deallocate(map);
}
- counter(c_vm_map_simplify_called++);
- vm_map_unlock(map);
+ return kr;
}
/*
- * Routine: vm_map_machine_attribute
- * Purpose:
- * Provide machine-specific attributes to mappings,
- * such as cachability etc. for machines that provide
- * them. NUMA architectures and machines with big/strange
- * caches will use this.
- * Note:
- * Responsibilities for locking and checking are handled here,
- * everything else in the pmap module. If any non-volatile
- * information must be kept, the pmap module should handle
- * it itself. [This assumes that attributes do not
- * need to be inherited, which seems ok to me]
+ * Routine: vm_map_read_user
+ *
+ * Description:
+ * Copy in data from a user space source map into the
+ * kernel map. The space must already exist in the
+ * kernel map.
+ * NOTE: This routine should only be called by threads
+ * which can block on a page fault. i.e. kernel mode user
+ * threads.
+ *
*/
kern_return_t
-vm_map_machine_attribute(
- vm_map_t map,
- vm_offset_t address,
- vm_size_t size,
- vm_machine_attribute_t attribute,
- vm_machine_attribute_val_t* value) /* IN/OUT */
+vm_map_read_user(
+ vm_map_t map,
+ vm_map_address_t src_addr,
+ void *dst_p,
+ vm_size_t size)
{
- kern_return_t ret;
- vm_size_t sync_size;
- vm_offset_t start;
- vm_map_entry_t entry;
-
- if (address < vm_map_min(map) ||
- (address + size) > vm_map_max(map))
- return KERN_INVALID_ADDRESS;
-
- vm_map_lock(map);
-
- if (attribute != MATTR_CACHE) {
- /* If we don't have to find physical addresses, we */
- /* don't have to do an explicit traversal here. */
- ret = pmap_attribute(map->pmap,
- address, size, attribute, value);
- vm_map_unlock(map);
- return ret;
- }
-
- /* Get the starting address */
- start = trunc_page_32(address);
- /* Figure how much memory we need to flush (in page increments) */
- sync_size = round_page_32(start + size) - start;
+ kern_return_t kr = KERN_SUCCESS;
+ if(current_map() == map) {
+ if (copyin(src_addr, dst_p, size)) {
+ kr = KERN_INVALID_ADDRESS;
+ }
+ } else {
+ vm_map_t oldmap;
- ret = KERN_SUCCESS; /* Assume it all worked */
+ /* take on the identity of the target map while doing */
+ /* the transfer */
- while(sync_size) {
- if (vm_map_lookup_entry(map, start, &entry)) {
- vm_size_t sub_size;
- if((entry->vme_end - start) > sync_size) {
- sub_size = sync_size;
- sync_size = 0;
- } else {
- sub_size = entry->vme_end - start;
- sync_size -= sub_size;
- }
- if(entry->is_sub_map) {
- vm_map_machine_attribute(
- entry->object.sub_map,
- (start - entry->vme_start)
- + entry->offset,
- sub_size,
- attribute, value);
- } else {
- if(entry->object.vm_object) {
- vm_page_t m;
- vm_object_t object;
- vm_object_t base_object;
- vm_object_offset_t offset;
- vm_object_offset_t base_offset;
- vm_size_t range;
- range = sub_size;
- offset = (start - entry->vme_start)
- + entry->offset;
- base_offset = offset;
- object = entry->object.vm_object;
- base_object = object;
- while(range) {
- m = vm_page_lookup(
- object, offset);
- if(m && !m->fictitious) {
-
- ret =
- pmap_attribute_cache_sync(
- m->phys_page,
- PAGE_SIZE,
- attribute, value);
- } else if (object->shadow) {
- offset = offset +
- object->shadow_offset;
- object = object->shadow;
- continue;
- }
- range -= PAGE_SIZE;
- /* Bump to the next page */
- base_offset += PAGE_SIZE;
- offset = base_offset;
- object = base_object;
-
- }
- }
- }
- start += sub_size;
- } else {
- vm_map_unlock(map);
- return KERN_FAILURE;
+ vm_map_reference(map);
+ oldmap = vm_map_switch(map);
+ if (copyin(src_addr, dst_p, size)) {
+ kr = KERN_INVALID_ADDRESS;
}
-
+ vm_map_switch(oldmap);
+ vm_map_deallocate(map);
}
-
- vm_map_unlock(map);
-
- return ret;
+ return kr;
}
+
/*
- * vm_map_behavior_set:
+ * vm_map_check_protection:
*
- * Sets the paging reference behavior of the specified address
- * range in the target map. Paging reference behavior affects
- * how pagein operations resulting from faults on the map will be
- * clustered.
+ * Assert that the target map allows the specified
+ * privilege on the entire address region given.
+ * The entire region must be allocated.
*/
-kern_return_t
-vm_map_behavior_set(
- vm_map_t map,
- vm_offset_t start,
- vm_offset_t end,
- vm_behavior_t new_behavior)
+boolean_t
+vm_map_check_protection(vm_map_t map, vm_map_offset_t start,
+ vm_map_offset_t end, vm_prot_t protection)
{
- register vm_map_entry_t entry;
- vm_map_entry_t temp_entry;
-
- XPR(XPR_VM_MAP,
- "vm_map_behavior_set, 0x%X start 0x%X end 0x%X behavior %d",
- (integer_t)map, start, end, new_behavior, 0);
-
- switch (new_behavior) {
- case VM_BEHAVIOR_DEFAULT:
- case VM_BEHAVIOR_RANDOM:
- case VM_BEHAVIOR_SEQUENTIAL:
- case VM_BEHAVIOR_RSEQNTL:
- break;
- case VM_BEHAVIOR_WILLNEED:
- case VM_BEHAVIOR_DONTNEED:
- new_behavior = VM_BEHAVIOR_DEFAULT;
- break;
- default:
- return(KERN_INVALID_ARGUMENT);
- }
+ vm_map_entry_t entry;
+ vm_map_entry_t tmp_entry;
vm_map_lock(map);
- /*
- * The entire address range must be valid for the map.
- * Note that vm_map_range_check() does a
- * vm_map_lookup_entry() internally and returns the
- * entry containing the start of the address range if
- * the entire range is valid.
- */
- if (vm_map_range_check(map, start, end, &temp_entry)) {
- entry = temp_entry;
- vm_map_clip_start(map, entry, start);
- }
- else {
+ if (start < vm_map_min(map) || end > vm_map_max(map) || start > end)
+ {
vm_map_unlock(map);
- return(KERN_INVALID_ADDRESS);
+ return (FALSE);
}
- while ((entry != vm_map_to_entry(map)) && (entry->vme_start < end)) {
- vm_map_clip_end(map, entry, end);
-
- entry->behavior = new_behavior;
-
- entry = entry->vme_next;
+ if (!vm_map_lookup_entry(map, start, &tmp_entry)) {
+ vm_map_unlock(map);
+ return(FALSE);
}
- vm_map_unlock(map);
- return(KERN_SUCCESS);
-}
-
-
-#include <mach_kdb.h>
-#if MACH_KDB
-#include <ddb/db_output.h>
-#include <vm/vm_print.h>
-
-#define printf db_printf
+ entry = tmp_entry;
-/*
- * Forward declarations for internal functions.
- */
-extern void vm_map_links_print(
- struct vm_map_links *links);
+ while (start < end) {
+ if (entry == vm_map_to_entry(map)) {
+ vm_map_unlock(map);
+ return(FALSE);
+ }
-extern void vm_map_header_print(
- struct vm_map_header *header);
+ /*
+ * No holes allowed!
+ */
-extern void vm_map_entry_print(
- vm_map_entry_t entry);
+ if (start < entry->vme_start) {
+ vm_map_unlock(map);
+ return(FALSE);
+ }
-extern void vm_follow_entry(
- vm_map_entry_t entry);
+ /*
+ * Check protection associated with entry.
+ */
-extern void vm_follow_map(
- vm_map_t map);
+ if ((entry->protection & protection) != protection) {
+ vm_map_unlock(map);
+ return(FALSE);
+ }
-/*
- * vm_map_links_print: [ debug ]
- */
-void
-vm_map_links_print(
- struct vm_map_links *links)
-{
- iprintf("prev = %08X next = %08X start = %08X end = %08X\n",
- links->prev,
- links->next,
- links->start,
- links->end);
-}
+ /* go to next entry */
-/*
- * vm_map_header_print: [ debug ]
- */
-void
-vm_map_header_print(
- struct vm_map_header *header)
-{
- vm_map_links_print(&header->links);
- iprintf("nentries = %08X, %sentries_pageable\n",
- header->nentries,
- (header->entries_pageable ? "" : "!"));
+ start = entry->vme_end;
+ entry = entry->vme_next;
+ }
+ vm_map_unlock(map);
+ return(TRUE);
}
-/*
- * vm_follow_entry: [ debug ]
- */
-void
-vm_follow_entry(
- vm_map_entry_t entry)
+kern_return_t
+vm_map_purgable_control(
+ vm_map_t map,
+ vm_map_offset_t address,
+ vm_purgable_t control,
+ int *state)
{
- extern int db_indent;
- int shadows;
+ vm_map_entry_t entry;
+ vm_object_t object;
+ kern_return_t kr;
+ boolean_t was_nonvolatile;
- iprintf("map entry %08X\n", entry);
+ /*
+ * Vet all the input parameters and current type and state of the
+ * underlaying object. Return with an error if anything is amiss.
+ */
+ if (map == VM_MAP_NULL)
+ return(KERN_INVALID_ARGUMENT);
- db_indent += 2;
+ if (control != VM_PURGABLE_SET_STATE &&
+ control != VM_PURGABLE_GET_STATE &&
+ control != VM_PURGABLE_PURGE_ALL)
+ return(KERN_INVALID_ARGUMENT);
- shadows = vm_follow_object(entry->object.vm_object);
- iprintf("Total objects : %d\n",shadows);
+ if (control == VM_PURGABLE_PURGE_ALL) {
+ vm_purgeable_object_purge_all();
+ return KERN_SUCCESS;
+ }
- db_indent -= 2;
-}
+ if (control == VM_PURGABLE_SET_STATE &&
+ (((*state & ~(VM_PURGABLE_ALL_MASKS)) != 0) ||
+ ((*state & VM_PURGABLE_STATE_MASK) > VM_PURGABLE_STATE_MASK)))
+ return(KERN_INVALID_ARGUMENT);
-/*
- * vm_map_entry_print: [ debug ]
- */
-void
-vm_map_entry_print(
- register vm_map_entry_t entry)
-{
- extern int db_indent;
- static char *inheritance_name[4] = { "share", "copy", "none", "?"};
- static char *behavior_name[4] = { "dflt", "rand", "seqtl", "rseqntl" };
-
- iprintf("map entry %08X n", entry);
+ vm_map_lock_read(map);
- db_indent += 2;
+ if (!vm_map_lookup_entry(map, address, &entry) || entry->is_sub_map) {
- vm_map_links_print(&entry->links);
+ /*
+ * Must pass a valid non-submap address.
+ */
+ vm_map_unlock_read(map);
+ return(KERN_INVALID_ADDRESS);
+ }
- iprintf("start = %08X end = %08X, prot=%x/%x/%s\n",
- entry->vme_start,
- entry->vme_end,
- entry->protection,
- entry->max_protection,
- inheritance_name[(entry->inheritance & 0x3)]);
+ if ((entry->protection & VM_PROT_WRITE) == 0) {
+ /*
+ * Can't apply purgable controls to something you can't write.
+ */
+ vm_map_unlock_read(map);
+ return(KERN_PROTECTION_FAILURE);
+ }
- iprintf("behavior = %s, wired_count = %d, user_wired_count = %d\n",
- behavior_name[(entry->behavior & 0x3)],
- entry->wired_count,
- entry->user_wired_count);
- iprintf("%sin_transition, %sneeds_wakeup\n",
- (entry->in_transition ? "" : "!"),
- (entry->needs_wakeup ? "" : "!"));
+ object = VME_OBJECT(entry);
+ if (object == VM_OBJECT_NULL ||
+ object->purgable == VM_PURGABLE_DENY) {
+ /*
+ * Object must already be present and be purgeable.
+ */
+ vm_map_unlock_read(map);
+ return KERN_INVALID_ARGUMENT;
+ }
+
+ vm_object_lock(object);
- if (entry->is_sub_map) {
- iprintf("submap = %08X - offset=%08X\n",
- entry->object.sub_map,
- entry->offset);
- } else {
- iprintf("object=%08X, offset=%08X, ",
- entry->object.vm_object,
- entry->offset);
- printf("%sis_shared, %sneeds_copy\n",
- (entry->is_shared ? "" : "!"),
- (entry->needs_copy ? "" : "!"));
+#if 00
+ if (VME_OFFSET(entry) != 0 ||
+ entry->vme_end - entry->vme_start != object->vo_size) {
+ /*
+ * Can only apply purgable controls to the whole (existing)
+ * object at once.
+ */
+ vm_map_unlock_read(map);
+ vm_object_unlock(object);
+ return KERN_INVALID_ARGUMENT;
}
+#endif
- db_indent -= 2;
-}
+ assert(!entry->is_sub_map);
+ assert(!entry->use_pmap); /* purgeable has its own accounting */
-/*
- * vm_follow_map: [ debug ]
- */
-void
-vm_follow_map(
- vm_map_t map)
-{
- register vm_map_entry_t entry;
- extern int db_indent;
+ vm_map_unlock_read(map);
- iprintf("task map %08X\n", map);
+ was_nonvolatile = (object->purgable == VM_PURGABLE_NONVOLATILE);
- db_indent += 2;
+ kr = vm_object_purgable_control(object, control, state);
- for (entry = vm_map_first_entry(map);
- entry && entry != vm_map_to_entry(map);
- entry = entry->vme_next) {
- vm_follow_entry(entry);
+ if (was_nonvolatile &&
+ object->purgable != VM_PURGABLE_NONVOLATILE &&
+ map->pmap == kernel_pmap) {
+#if DEBUG
+ object->vo_purgeable_volatilizer = kernel_task;
+#endif /* DEBUG */
}
- db_indent -= 2;
+ vm_object_unlock(object);
+
+ return kr;
}
-/*
- * vm_map_print: [ debug ]
- */
-void
-vm_map_print(
- db_addr_t inmap)
+kern_return_t
+vm_map_page_query_internal(
+ vm_map_t target_map,
+ vm_map_offset_t offset,
+ int *disposition,
+ int *ref_count)
{
- register vm_map_entry_t entry;
- vm_map_t map;
- extern int db_indent;
- char *swstate;
-
- map = (vm_map_t)inmap; /* Make sure we have the right type */
-
- iprintf("task map %08X\n", map);
-
- db_indent += 2;
-
- vm_map_header_print(&map->hdr);
-
- iprintf("pmap = %08X, size = %08X, ref = %d, hint = %08X, first_free = %08X\n",
- map->pmap,
- map->size,
- map->ref_count,
- map->hint,
- map->first_free);
+ kern_return_t kr;
+ vm_page_info_basic_data_t info;
+ mach_msg_type_number_t count;
+
+ count = VM_PAGE_INFO_BASIC_COUNT;
+ kr = vm_map_page_info(target_map,
+ offset,
+ VM_PAGE_INFO_BASIC,
+ (vm_page_info_t) &info,
+ &count);
+ if (kr == KERN_SUCCESS) {
+ *disposition = info.disposition;
+ *ref_count = info.ref_count;
+ } else {
+ *disposition = 0;
+ *ref_count = 0;
+ }
- iprintf("%swait_for_space, %swiring_required, timestamp = %d\n",
- (map->wait_for_space ? "" : "!"),
- (map->wiring_required ? "" : "!"),
- map->timestamp);
+ return kr;
+}
+
+kern_return_t
+vm_map_page_info(
+ vm_map_t map,
+ vm_map_offset_t offset,
+ vm_page_info_flavor_t flavor,
+ vm_page_info_t info,
+ mach_msg_type_number_t *count)
+{
+ vm_map_entry_t map_entry;
+ vm_object_t object;
+ vm_page_t m;
+ kern_return_t kr;
+ kern_return_t retval = KERN_SUCCESS;
+ boolean_t top_object;
+ int disposition;
+ int ref_count;
+ vm_page_info_basic_t basic_info;
+ int depth;
+ vm_map_offset_t offset_in_page;
-#if TASK_SWAPPER
- switch (map->sw_state) {
- case MAP_SW_IN:
- swstate = "SW_IN";
- break;
- case MAP_SW_OUT:
- swstate = "SW_OUT";
- break;
- default:
- swstate = "????";
+ switch (flavor) {
+ case VM_PAGE_INFO_BASIC:
+ if (*count != VM_PAGE_INFO_BASIC_COUNT) {
+ /*
+ * The "vm_page_info_basic_data" structure was not
+ * properly padded, so allow the size to be off by
+ * one to maintain backwards binary compatibility...
+ */
+ if (*count != VM_PAGE_INFO_BASIC_COUNT - 1)
+ return KERN_INVALID_ARGUMENT;
+ }
break;
+ default:
+ return KERN_INVALID_ARGUMENT;
}
- iprintf("res = %d, sw_state = %s\n", map->res_count, swstate);
-#endif /* TASK_SWAPPER */
- for (entry = vm_map_first_entry(map);
- entry && entry != vm_map_to_entry(map);
- entry = entry->vme_next) {
- vm_map_entry_print(entry);
- }
+ disposition = 0;
+ ref_count = 0;
+ top_object = TRUE;
+ depth = 0;
- db_indent -= 2;
-}
+ retval = KERN_SUCCESS;
+ offset_in_page = offset & PAGE_MASK;
+ offset = vm_map_trunc_page(offset, PAGE_MASK);
-/*
- * Routine: vm_map_copy_print
- * Purpose:
- * Pretty-print a copy object for ddb.
- */
+ vm_map_lock_read(map);
-void
-vm_map_copy_print(
- db_addr_t incopy)
-{
- extern int db_indent;
- vm_map_copy_t copy;
- int i, npages;
- vm_map_entry_t entry;
+ /*
+ * First, find the map entry covering "offset", going down
+ * submaps if necessary.
+ */
+ for (;;) {
+ if (!vm_map_lookup_entry(map, offset, &map_entry)) {
+ vm_map_unlock_read(map);
+ return KERN_INVALID_ADDRESS;
+ }
+ /* compute offset from this map entry's start */
+ offset -= map_entry->vme_start;
+ /* compute offset into this map entry's object (or submap) */
+ offset += VME_OFFSET(map_entry);
- copy = (vm_map_copy_t)incopy; /* Make sure we have the right type */
+ if (map_entry->is_sub_map) {
+ vm_map_t sub_map;
- printf("copy object 0x%x\n", copy);
+ sub_map = VME_SUBMAP(map_entry);
+ vm_map_lock_read(sub_map);
+ vm_map_unlock_read(map);
- db_indent += 2;
+ map = sub_map;
- iprintf("type=%d", copy->type);
- switch (copy->type) {
- case VM_MAP_COPY_ENTRY_LIST:
- printf("[entry_list]");
- break;
-
- case VM_MAP_COPY_OBJECT:
- printf("[object]");
- break;
-
- case VM_MAP_COPY_KERNEL_BUFFER:
- printf("[kernel_buffer]");
+ ref_count = MAX(ref_count, map->ref_count);
+ continue;
+ }
break;
+ }
- default:
- printf("[bad type]");
- break;
+ object = VME_OBJECT(map_entry);
+ if (object == VM_OBJECT_NULL) {
+ /* no object -> no page */
+ vm_map_unlock_read(map);
+ goto done;
}
- printf(", offset=0x%x", copy->offset);
- printf(", size=0x%x\n", copy->size);
- switch (copy->type) {
- case VM_MAP_COPY_ENTRY_LIST:
- vm_map_header_print(©->cpy_hdr);
- for (entry = vm_map_copy_first_entry(copy);
- entry && entry != vm_map_copy_to_entry(copy);
- entry = entry->vme_next) {
- vm_map_entry_print(entry);
- }
- break;
+ vm_object_lock(object);
+ vm_map_unlock_read(map);
- case VM_MAP_COPY_OBJECT:
- iprintf("object=0x%x\n", copy->cpy_object);
- break;
+ /*
+ * Go down the VM object shadow chain until we find the page
+ * we're looking for.
+ */
+ for (;;) {
+ ref_count = MAX(ref_count, object->ref_count);
- case VM_MAP_COPY_KERNEL_BUFFER:
- iprintf("kernel buffer=0x%x", copy->cpy_kdata);
- printf(", kalloc_size=0x%x\n", copy->cpy_kalloc_size);
- break;
+ m = vm_page_lookup(object, offset);
- }
+ if (m != VM_PAGE_NULL) {
+ disposition |= VM_PAGE_QUERY_PAGE_PRESENT;
+ break;
+ } else {
+#if MACH_PAGEMAP
+ if (object->existence_map) {
+ if (vm_external_state_get(object->existence_map,
+ offset) ==
+ VM_EXTERNAL_STATE_EXISTS) {
+ /*
+ * this page has been paged out
+ */
+ disposition |= VM_PAGE_QUERY_PAGE_PAGED_OUT;
+ break;
+ }
+ } else
+#endif
+ if (object->internal &&
+ object->alive &&
+ !object->terminating &&
+ object->pager_ready) {
+
+ if (COMPRESSED_PAGER_IS_ACTIVE || DEFAULT_FREEZER_COMPRESSED_PAGER_IS_ACTIVE) {
+ if (VM_COMPRESSOR_PAGER_STATE_GET(
+ object,
+ offset)
+ == VM_EXTERNAL_STATE_EXISTS) {
+ /* the pager has that page */
+ disposition |= VM_PAGE_QUERY_PAGE_PAGED_OUT;
+ break;
+ }
+ } else {
+ memory_object_t pager;
- db_indent -=2;
-}
+ vm_object_paging_begin(object);
+ pager = object->pager;
+ vm_object_unlock(object);
-/*
- * db_vm_map_total_size(map) [ debug ]
- *
- * return the total virtual size (in bytes) of the map
- */
-vm_size_t
-db_vm_map_total_size(
- db_addr_t inmap)
-{
- vm_map_entry_t entry;
- vm_size_t total;
- vm_map_t map;
+ /*
+ * Ask the default pager if
+ * it has this page.
+ */
+ kr = memory_object_data_request(
+ pager,
+ offset + object->paging_offset,
+ 0, /* just poke the pager */
+ VM_PROT_READ,
+ NULL);
+
+ vm_object_lock(object);
+ vm_object_paging_end(object);
+
+ if (kr == KERN_SUCCESS) {
+ /* the default pager has it */
+ disposition |= VM_PAGE_QUERY_PAGE_PAGED_OUT;
+ break;
+ }
+ }
+ }
- map = (vm_map_t)inmap; /* Make sure we have the right type */
+ if (object->shadow != VM_OBJECT_NULL) {
+ vm_object_t shadow;
+
+ offset += object->vo_shadow_offset;
+ shadow = object->shadow;
+
+ vm_object_lock(shadow);
+ vm_object_unlock(object);
- total = 0;
- for (entry = vm_map_first_entry(map);
- entry != vm_map_to_entry(map);
- entry = entry->vme_next) {
- total += entry->vme_end - entry->vme_start;
+ object = shadow;
+ top_object = FALSE;
+ depth++;
+ } else {
+// if (!object->internal)
+// break;
+// retval = KERN_FAILURE;
+// goto done_with_object;
+ break;
+ }
+ }
}
+ /* The ref_count is not strictly accurate, it measures the number */
+ /* of entities holding a ref on the object, they may not be mapping */
+ /* the object or may not be mapping the section holding the */
+ /* target page but its still a ball park number and though an over- */
+ /* count, it picks up the copy-on-write cases */
- return total;
-}
+ /* We could also get a picture of page sharing from pmap_attributes */
+ /* but this would under count as only faulted-in mappings would */
+ /* show up. */
-#endif /* MACH_KDB */
+ if (top_object == TRUE && object->shadow)
+ disposition |= VM_PAGE_QUERY_PAGE_COPIED;
-/*
- * Routine: vm_map_entry_insert
- *
- * Descritpion: This routine inserts a new vm_entry in a locked map.
- */
-vm_map_entry_t
-vm_map_entry_insert(
- vm_map_t map,
- vm_map_entry_t insp_entry,
- vm_offset_t start,
- vm_offset_t end,
- vm_object_t object,
- vm_object_offset_t offset,
- boolean_t needs_copy,
- boolean_t is_shared,
- boolean_t in_transition,
- vm_prot_t cur_protection,
- vm_prot_t max_protection,
- vm_behavior_t behavior,
- vm_inherit_t inheritance,
- unsigned wired_count)
-{
- vm_map_entry_t new_entry;
+ if (! object->internal)
+ disposition |= VM_PAGE_QUERY_PAGE_EXTERNAL;
- assert(insp_entry != (vm_map_entry_t)0);
+ if (m == VM_PAGE_NULL)
+ goto done_with_object;
- new_entry = vm_map_entry_create(map);
+ if (m->fictitious) {
+ disposition |= VM_PAGE_QUERY_PAGE_FICTITIOUS;
+ goto done_with_object;
+ }
+ if (m->dirty || pmap_is_modified(m->phys_page))
+ disposition |= VM_PAGE_QUERY_PAGE_DIRTY;
- new_entry->vme_start = start;
- new_entry->vme_end = end;
- assert(page_aligned(new_entry->vme_start));
- assert(page_aligned(new_entry->vme_end));
+ if (m->reference || pmap_is_referenced(m->phys_page))
+ disposition |= VM_PAGE_QUERY_PAGE_REF;
- new_entry->object.vm_object = object;
- new_entry->offset = offset;
- new_entry->is_shared = is_shared;
- new_entry->is_sub_map = FALSE;
- new_entry->needs_copy = needs_copy;
- new_entry->in_transition = in_transition;
- new_entry->needs_wakeup = FALSE;
- new_entry->inheritance = inheritance;
- new_entry->protection = cur_protection;
- new_entry->max_protection = max_protection;
- new_entry->behavior = behavior;
- new_entry->wired_count = wired_count;
- new_entry->user_wired_count = 0;
- new_entry->use_pmap = FALSE;
+ if (m->speculative)
+ disposition |= VM_PAGE_QUERY_PAGE_SPECULATIVE;
- /*
- * Insert the new entry into the list.
- */
+ if (m->cs_validated)
+ disposition |= VM_PAGE_QUERY_PAGE_CS_VALIDATED;
+ if (m->cs_tainted)
+ disposition |= VM_PAGE_QUERY_PAGE_CS_TAINTED;
+ if (m->cs_nx)
+ disposition |= VM_PAGE_QUERY_PAGE_CS_NX;
- vm_map_entry_link(map, insp_entry, new_entry);
- map->size += end - start;
+done_with_object:
+ vm_object_unlock(object);
+done:
- /*
- * Update the free space hint and the lookup hint.
- */
+ switch (flavor) {
+ case VM_PAGE_INFO_BASIC:
+ basic_info = (vm_page_info_basic_t) info;
+ basic_info->disposition = disposition;
+ basic_info->ref_count = ref_count;
+ basic_info->object_id = (vm_object_id_t) (uintptr_t)
+ VM_KERNEL_ADDRPERM(object);
+ basic_info->offset =
+ (memory_object_offset_t) offset + offset_in_page;
+ basic_info->depth = depth;
+ break;
+ }
- SAVE_HINT(map, new_entry);
- return new_entry;
+ return retval;
}
/*
- * Routine: vm_remap_extract
+ * vm_map_msync
*
- * Descritpion: This routine returns a vm_entry list from a map.
+ * Synchronises the memory range specified with its backing store
+ * image by either flushing or cleaning the contents to the appropriate
+ * memory manager engaging in a memory object synchronize dialog with
+ * the manager. The client doesn't return until the manager issues
+ * m_o_s_completed message. MIG Magically converts user task parameter
+ * to the task's address map.
+ *
+ * interpretation of sync_flags
+ * VM_SYNC_INVALIDATE - discard pages, only return precious
+ * pages to manager.
+ *
+ * VM_SYNC_INVALIDATE & (VM_SYNC_SYNCHRONOUS | VM_SYNC_ASYNCHRONOUS)
+ * - discard pages, write dirty or precious
+ * pages back to memory manager.
+ *
+ * VM_SYNC_SYNCHRONOUS | VM_SYNC_ASYNCHRONOUS
+ * - write dirty or precious pages back to
+ * the memory manager.
+ *
+ * VM_SYNC_CONTIGUOUS - does everything normally, but if there
+ * is a hole in the region, and we would
+ * have returned KERN_SUCCESS, return
+ * KERN_INVALID_ADDRESS instead.
+ *
+ * NOTE
+ * The memory object attributes have not yet been implemented, this
+ * function will have to deal with the invalidate attribute
+ *
+ * RETURNS
+ * KERN_INVALID_TASK Bad task parameter
+ * KERN_INVALID_ARGUMENT both sync and async were specified.
+ * KERN_SUCCESS The usual.
+ * KERN_INVALID_ADDRESS There was a hole in the region.
*/
+
kern_return_t
-vm_remap_extract(
+vm_map_msync(
vm_map_t map,
- vm_offset_t addr,
- vm_size_t size,
- boolean_t copy,
- struct vm_map_header *map_header,
- vm_prot_t *cur_protection,
- vm_prot_t *max_protection,
- /* What, no behavior? */
- vm_inherit_t inheritance,
- boolean_t pageable)
+ vm_map_address_t address,
+ vm_map_size_t size,
+ vm_sync_t sync_flags)
{
- kern_return_t result;
- vm_size_t mapped_size;
- vm_size_t tmp_size;
- vm_map_entry_t src_entry; /* result of last map lookup */
- vm_map_entry_t new_entry;
+ msync_req_t msr;
+ msync_req_t new_msr;
+ queue_chain_t req_q; /* queue of requests for this msync */
+ vm_map_entry_t entry;
+ vm_map_size_t amount_left;
vm_object_offset_t offset;
- vm_offset_t map_address;
- vm_offset_t src_start; /* start of entry to map */
- vm_offset_t src_end; /* end of region to be mapped */
- vm_object_t object;
- vm_map_version_t version;
- boolean_t src_needs_copy;
- boolean_t new_entry_needs_copy;
-
- assert(map != VM_MAP_NULL);
- assert(size != 0 && size == round_page_32(size));
- assert(inheritance == VM_INHERIT_NONE ||
- inheritance == VM_INHERIT_COPY ||
- inheritance == VM_INHERIT_SHARE);
+ boolean_t do_sync_req;
+ boolean_t had_hole = FALSE;
+ memory_object_t pager;
+ vm_map_offset_t pmap_offset;
+
+ if ((sync_flags & VM_SYNC_ASYNCHRONOUS) &&
+ (sync_flags & VM_SYNC_SYNCHRONOUS))
+ return(KERN_INVALID_ARGUMENT);
/*
- * Compute start and end of region.
+ * align address and size on page boundaries
*/
- src_start = trunc_page_32(addr);
- src_end = round_page_32(src_start + size);
+ size = (vm_map_round_page(address + size,
+ VM_MAP_PAGE_MASK(map)) -
+ vm_map_trunc_page(address,
+ VM_MAP_PAGE_MASK(map)));
+ address = vm_map_trunc_page(address,
+ VM_MAP_PAGE_MASK(map));
- /*
- * Initialize map_header.
- */
- map_header->links.next = (struct vm_map_entry *)&map_header->links;
- map_header->links.prev = (struct vm_map_entry *)&map_header->links;
- map_header->nentries = 0;
- map_header->entries_pageable = pageable;
+ if (map == VM_MAP_NULL)
+ return(KERN_INVALID_TASK);
- *cur_protection = VM_PROT_ALL;
- *max_protection = VM_PROT_ALL;
+ if (size == 0)
+ return(KERN_SUCCESS);
- map_address = 0;
- mapped_size = 0;
- result = KERN_SUCCESS;
+ queue_init(&req_q);
+ amount_left = size;
- /*
- * The specified source virtual space might correspond to
- * multiple map entries, need to loop on them.
- */
- vm_map_lock(map);
- while (mapped_size != size) {
- vm_size_t entry_size;
+ while (amount_left > 0) {
+ vm_object_size_t flush_size;
+ vm_object_t object;
- /*
- * Find the beginning of the region.
- */
- if (! vm_map_lookup_entry(map, src_start, &src_entry)) {
- result = KERN_INVALID_ADDRESS;
- break;
- }
+ vm_map_lock(map);
+ if (!vm_map_lookup_entry(map,
+ address,
+ &entry)) {
- if (src_start < src_entry->vme_start ||
- (mapped_size && src_start != src_entry->vme_start)) {
- result = KERN_INVALID_ADDRESS;
- break;
- }
+ vm_map_size_t skip;
- if(src_entry->is_sub_map) {
- result = KERN_INVALID_ADDRESS;
- break;
- }
+ /*
+ * hole in the address map.
+ */
+ had_hole = TRUE;
- tmp_size = size - mapped_size;
- if (src_end > src_entry->vme_end)
- tmp_size -= (src_end - src_entry->vme_end);
+ /*
+ * Check for empty map.
+ */
+ if (entry == vm_map_to_entry(map) &&
+ entry->vme_next == entry) {
+ vm_map_unlock(map);
+ break;
+ }
+ /*
+ * Check that we don't wrap and that
+ * we have at least one real map entry.
+ */
+ if ((map->hdr.nentries == 0) ||
+ (entry->vme_next->vme_start < address)) {
+ vm_map_unlock(map);
+ break;
+ }
+ /*
+ * Move up to the next entry if needed
+ */
+ skip = (entry->vme_next->vme_start - address);
+ if (skip >= amount_left)
+ amount_left = 0;
+ else
+ amount_left -= skip;
+ address = entry->vme_next->vme_start;
+ vm_map_unlock(map);
+ continue;
+ }
- entry_size = (vm_size_t)(src_entry->vme_end -
- src_entry->vme_start);
+ offset = address - entry->vme_start;
+ pmap_offset = address;
- if(src_entry->is_sub_map) {
- vm_map_reference(src_entry->object.sub_map);
+ /*
+ * do we have more to flush than is contained in this
+ * entry ?
+ */
+ if (amount_left + entry->vme_start + offset > entry->vme_end) {
+ flush_size = entry->vme_end -
+ (entry->vme_start + offset);
} else {
- object = src_entry->object.vm_object;
+ flush_size = amount_left;
+ }
+ amount_left -= flush_size;
+ address += flush_size;
- if (object == VM_OBJECT_NULL) {
- object = vm_object_allocate(entry_size);
- src_entry->offset = 0;
- src_entry->object.vm_object = object;
- } else if (object->copy_strategy !=
- MEMORY_OBJECT_COPY_SYMMETRIC) {
- /*
- * We are already using an asymmetric
- * copy, and therefore we already have
- * the right object.
- */
- assert(!src_entry->needs_copy);
- } else if (src_entry->needs_copy || object->shadowed ||
- (object->internal && !object->true_share &&
- !src_entry->is_shared &&
- object->size > entry_size)) {
+ if (entry->is_sub_map == TRUE) {
+ vm_map_t local_map;
+ vm_map_offset_t local_offset;
- vm_object_shadow(&src_entry->object.vm_object,
- &src_entry->offset,
- entry_size);
+ local_map = VME_SUBMAP(entry);
+ local_offset = VME_OFFSET(entry);
+ vm_map_unlock(map);
+ if (vm_map_msync(
+ local_map,
+ local_offset,
+ flush_size,
+ sync_flags) == KERN_INVALID_ADDRESS) {
+ had_hole = TRUE;
+ }
+ continue;
+ }
+ object = VME_OBJECT(entry);
- if (!src_entry->needs_copy &&
- (src_entry->protection & VM_PROT_WRITE)) {
- if(map->mapped) {
- vm_object_pmap_protect(
- src_entry->object.vm_object,
- src_entry->offset,
- entry_size,
- PMAP_NULL,
- src_entry->vme_start,
- src_entry->protection &
- ~VM_PROT_WRITE);
- } else {
- pmap_protect(vm_map_pmap(map),
- src_entry->vme_start,
- src_entry->vme_end,
- src_entry->protection &
- ~VM_PROT_WRITE);
- }
- }
+ /*
+ * We can't sync this object if the object has not been
+ * created yet
+ */
+ if (object == VM_OBJECT_NULL) {
+ vm_map_unlock(map);
+ continue;
+ }
+ offset += VME_OFFSET(entry);
- object = src_entry->object.vm_object;
- src_entry->needs_copy = FALSE;
- }
+ vm_object_lock(object);
+ if (sync_flags & (VM_SYNC_KILLPAGES | VM_SYNC_DEACTIVATE)) {
+ int kill_pages = 0;
+ boolean_t reusable_pages = FALSE;
- vm_object_lock(object);
- object->ref_count++; /* object ref. for new entry */
- VM_OBJ_RES_INCR(object);
- if (object->copy_strategy ==
- MEMORY_OBJECT_COPY_SYMMETRIC) {
- object->copy_strategy =
- MEMORY_OBJECT_COPY_DELAY;
+ if (sync_flags & VM_SYNC_KILLPAGES) {
+ if (object->ref_count == 1 && !object->shadow)
+ kill_pages = 1;
+ else
+ kill_pages = -1;
}
+ if (kill_pages != -1)
+ vm_object_deactivate_pages(
+ object,
+ offset,
+ (vm_object_size_t) flush_size,
+ kill_pages,
+ reusable_pages,
+ map->pmap,
+ pmap_offset);
+ vm_object_unlock(object);
+ vm_map_unlock(map);
+ continue;
+ }
+ /*
+ * We can't sync this object if there isn't a pager.
+ * Don't bother to sync internal objects, since there can't
+ * be any "permanent" storage for these objects anyway.
+ */
+ if ((object->pager == MEMORY_OBJECT_NULL) ||
+ (object->internal) || (object->private)) {
vm_object_unlock(object);
+ vm_map_unlock(map);
+ continue;
}
+ /*
+ * keep reference on the object until syncing is done
+ */
+ vm_object_reference_locked(object);
+ vm_object_unlock(object);
- offset = src_entry->offset + (src_start - src_entry->vme_start);
+ vm_map_unlock(map);
- new_entry = _vm_map_entry_create(map_header);
- vm_map_entry_copy(new_entry, src_entry);
- new_entry->use_pmap = FALSE; /* clr address space specifics */
+ do_sync_req = vm_object_sync(object,
+ offset,
+ flush_size,
+ sync_flags & VM_SYNC_INVALIDATE,
+ ((sync_flags & VM_SYNC_SYNCHRONOUS) ||
+ (sync_flags & VM_SYNC_ASYNCHRONOUS)),
+ sync_flags & VM_SYNC_SYNCHRONOUS);
+ /*
+ * only send a m_o_s if we returned pages or if the entry
+ * is writable (ie dirty pages may have already been sent back)
+ */
+ if (!do_sync_req) {
+ if ((sync_flags & VM_SYNC_INVALIDATE) && object->resident_page_count == 0) {
+ /*
+ * clear out the clustering and read-ahead hints
+ */
+ vm_object_lock(object);
- new_entry->vme_start = map_address;
- new_entry->vme_end = map_address + tmp_size;
- new_entry->inheritance = inheritance;
- new_entry->offset = offset;
+ object->pages_created = 0;
+ object->pages_used = 0;
+ object->sequential = 0;
+ object->last_alloc = 0;
+
+ vm_object_unlock(object);
+ }
+ vm_object_deallocate(object);
+ continue;
+ }
+ msync_req_alloc(new_msr);
+
+ vm_object_lock(object);
+ offset += object->paging_offset;
+
+ new_msr->offset = offset;
+ new_msr->length = flush_size;
+ new_msr->object = object;
+ new_msr->flag = VM_MSYNC_SYNCHRONIZING;
+ re_iterate:
/*
- * The new region has to be copied now if required.
+ * We can't sync this object if there isn't a pager. The
+ * pager can disappear anytime we're not holding the object
+ * lock. So this has to be checked anytime we goto re_iterate.
*/
- RestartCopy:
- if (!copy) {
- src_entry->is_shared = TRUE;
- new_entry->is_shared = TRUE;
- if (!(new_entry->is_sub_map))
- new_entry->needs_copy = FALSE;
- } else if (src_entry->is_sub_map) {
- /* make this a COW sub_map if not already */
- new_entry->needs_copy = TRUE;
- } else if (src_entry->wired_count == 0 &&
- vm_object_copy_quickly(&new_entry->object.vm_object,
- new_entry->offset,
- (new_entry->vme_end -
- new_entry->vme_start),
- &src_needs_copy,
- &new_entry_needs_copy)) {
+ pager = object->pager;
- new_entry->needs_copy = new_entry_needs_copy;
- new_entry->is_shared = FALSE;
+ if (pager == MEMORY_OBJECT_NULL) {
+ vm_object_unlock(object);
+ vm_object_deallocate(object);
+ msync_req_free(new_msr);
+ new_msr = NULL;
+ continue;
+ }
+ queue_iterate(&object->msr_q, msr, msync_req_t, msr_q) {
/*
- * Handle copy_on_write semantics.
+ * need to check for overlapping entry, if found, wait
+ * on overlapping msr to be done, then reiterate
*/
- if (src_needs_copy && !src_entry->needs_copy) {
- vm_object_pmap_protect(object,
- offset,
- entry_size,
- ((src_entry->is_shared
- || map->mapped) ?
- PMAP_NULL : map->pmap),
- src_entry->vme_start,
- src_entry->protection &
- ~VM_PROT_WRITE);
-
- src_entry->needs_copy = TRUE;
+ msr_lock(msr);
+ if (msr->flag == VM_MSYNC_SYNCHRONIZING &&
+ ((offset >= msr->offset &&
+ offset < (msr->offset + msr->length)) ||
+ (msr->offset >= offset &&
+ msr->offset < (offset + flush_size))))
+ {
+ assert_wait((event_t) msr,THREAD_INTERRUPTIBLE);
+ msr_unlock(msr);
+ vm_object_unlock(object);
+ thread_block(THREAD_CONTINUE_NULL);
+ vm_object_lock(object);
+ goto re_iterate;
}
- /*
- * Throw away the old object reference of the new entry.
- */
- vm_object_deallocate(object);
+ msr_unlock(msr);
+ }/* queue_iterate */
+
+ queue_enter(&object->msr_q, new_msr, msync_req_t, msr_q);
+
+ vm_object_paging_begin(object);
+ vm_object_unlock(object);
+
+ queue_enter(&req_q, new_msr, msync_req_t, req_q);
+
+ (void) memory_object_synchronize(
+ pager,
+ offset,
+ flush_size,
+ sync_flags & ~VM_SYNC_CONTIGUOUS);
+
+ vm_object_lock(object);
+ vm_object_paging_end(object);
+ vm_object_unlock(object);
+ }/* while */
- } else {
- new_entry->is_shared = FALSE;
+ /*
+ * wait for memory_object_sychronize_completed messages from pager(s)
+ */
- /*
- * The map can be safely unlocked since we
- * already hold a reference on the object.
- *
- * Record the timestamp of the map for later
- * verification, and unlock the map.
- */
- version.main_timestamp = map->timestamp;
- vm_map_unlock(map); /* Increments timestamp once! */
+ while (!queue_empty(&req_q)) {
+ msr = (msync_req_t)queue_first(&req_q);
+ msr_lock(msr);
+ while(msr->flag != VM_MSYNC_DONE) {
+ assert_wait((event_t) msr, THREAD_INTERRUPTIBLE);
+ msr_unlock(msr);
+ thread_block(THREAD_CONTINUE_NULL);
+ msr_lock(msr);
+ }/* while */
+ queue_remove(&req_q, msr, msync_req_t, req_q);
+ msr_unlock(msr);
+ vm_object_deallocate(msr->object);
+ msync_req_free(msr);
+ }/* queue_iterate */
+
+ /* for proper msync() behaviour */
+ if (had_hole == TRUE && (sync_flags & VM_SYNC_CONTIGUOUS))
+ return(KERN_INVALID_ADDRESS);
- /*
- * Perform the copy.
- */
- if (src_entry->wired_count > 0) {
- vm_object_lock(object);
- result = vm_object_copy_slowly(
- object,
- offset,
- entry_size,
- THREAD_UNINT,
- &new_entry->object.vm_object);
+ return(KERN_SUCCESS);
+}/* vm_msync */
- new_entry->offset = 0;
- new_entry->needs_copy = FALSE;
- } else {
- result = vm_object_copy_strategically(
- object,
- offset,
- entry_size,
- &new_entry->object.vm_object,
- &new_entry->offset,
- &new_entry_needs_copy);
+/*
+ * Routine: convert_port_entry_to_map
+ * Purpose:
+ * Convert from a port specifying an entry or a task
+ * to a map. Doesn't consume the port ref; produces a map ref,
+ * which may be null. Unlike convert_port_to_map, the
+ * port may be task or a named entry backed.
+ * Conditions:
+ * Nothing locked.
+ */
- new_entry->needs_copy = new_entry_needs_copy;
- }
- /*
- * Throw away the old object reference of the new entry.
- */
- vm_object_deallocate(object);
+vm_map_t
+convert_port_entry_to_map(
+ ipc_port_t port)
+{
+ vm_map_t map;
+ vm_named_entry_t named_entry;
+ uint32_t try_failed_count = 0;
- if (result != KERN_SUCCESS &&
- result != KERN_MEMORY_RESTART_COPY) {
- _vm_map_entry_dispose(map_header, new_entry);
+ if(IP_VALID(port) && (ip_kotype(port) == IKOT_NAMED_ENTRY)) {
+ while(TRUE) {
+ ip_lock(port);
+ if(ip_active(port) && (ip_kotype(port)
+ == IKOT_NAMED_ENTRY)) {
+ named_entry =
+ (vm_named_entry_t)port->ip_kobject;
+ if (!(lck_mtx_try_lock(&(named_entry)->Lock))) {
+ ip_unlock(port);
+
+ try_failed_count++;
+ mutex_pause(try_failed_count);
+ continue;
+ }
+ named_entry->ref_count++;
+ lck_mtx_unlock(&(named_entry)->Lock);
+ ip_unlock(port);
+ if ((named_entry->is_sub_map) &&
+ (named_entry->protection
+ & VM_PROT_WRITE)) {
+ map = named_entry->backing.map;
+ } else {
+ mach_destroy_memory_entry(port);
+ return VM_MAP_NULL;
+ }
+ vm_map_reference_swap(map);
+ mach_destroy_memory_entry(port);
break;
}
+ else
+ return VM_MAP_NULL;
+ }
+ }
+ else
+ map = convert_port_to_map(port);
- /*
- * Verify that the map has not substantially
- * changed while the copy was being made.
- */
+ return map;
+}
- vm_map_lock(map);
- if (version.main_timestamp + 1 != map->timestamp) {
- /*
- * Simple version comparison failed.
- *
- * Retry the lookup and verify that the
- * same object/offset are still present.
- */
- vm_object_deallocate(new_entry->
- object.vm_object);
- _vm_map_entry_dispose(map_header, new_entry);
- if (result == KERN_MEMORY_RESTART_COPY)
- result = KERN_SUCCESS;
- continue;
- }
+/*
+ * Routine: convert_port_entry_to_object
+ * Purpose:
+ * Convert from a port specifying a named entry to an
+ * object. Doesn't consume the port ref; produces a map ref,
+ * which may be null.
+ * Conditions:
+ * Nothing locked.
+ */
- if (result == KERN_MEMORY_RESTART_COPY) {
+
+vm_object_t
+convert_port_entry_to_object(
+ ipc_port_t port)
+{
+ vm_object_t object = VM_OBJECT_NULL;
+ vm_named_entry_t named_entry;
+ uint32_t try_failed_count = 0;
+
+ if (IP_VALID(port) &&
+ (ip_kotype(port) == IKOT_NAMED_ENTRY)) {
+ try_again:
+ ip_lock(port);
+ if (ip_active(port) &&
+ (ip_kotype(port) == IKOT_NAMED_ENTRY)) {
+ named_entry = (vm_named_entry_t)port->ip_kobject;
+ if (!(lck_mtx_try_lock(&(named_entry)->Lock))) {
+ ip_unlock(port);
+ try_failed_count++;
+ mutex_pause(try_failed_count);
+ goto try_again;
+ }
+ named_entry->ref_count++;
+ lck_mtx_unlock(&(named_entry)->Lock);
+ ip_unlock(port);
+ if (!(named_entry->is_sub_map) &&
+ !(named_entry->is_pager) &&
+ !(named_entry->is_copy) &&
+ (named_entry->protection & VM_PROT_WRITE)) {
+ object = named_entry->backing.object;
vm_object_reference(object);
- goto RestartCopy;
}
+ mach_destroy_memory_entry(port);
}
+ }
- _vm_map_entry_link(map_header,
- map_header->links.prev, new_entry);
-
- *cur_protection &= src_entry->protection;
- *max_protection &= src_entry->max_protection;
+ return object;
+}
- map_address += tmp_size;
- mapped_size += tmp_size;
- src_start += tmp_size;
+/*
+ * Export routines to other components for the things we access locally through
+ * macros.
+ */
+#undef current_map
+vm_map_t
+current_map(void)
+{
+ return (current_map_fast());
+}
- } /* end while */
+/*
+ * vm_map_reference:
+ *
+ * Most code internal to the osfmk will go through a
+ * macro defining this. This is always here for the
+ * use of other kernel components.
+ */
+#undef vm_map_reference
+void
+vm_map_reference(
+ register vm_map_t map)
+{
+ if (map == VM_MAP_NULL)
+ return;
- vm_map_unlock(map);
- if (result != KERN_SUCCESS) {
- /*
- * Free all allocated elements.
- */
- for (src_entry = map_header->links.next;
- src_entry != (struct vm_map_entry *)&map_header->links;
- src_entry = new_entry) {
- new_entry = src_entry->vme_next;
- _vm_map_entry_unlink(map_header, src_entry);
- vm_object_deallocate(src_entry->object.vm_object);
- _vm_map_entry_dispose(map_header, src_entry);
- }
- }
- return result;
+ lck_mtx_lock(&map->s_lock);
+#if TASK_SWAPPER
+ assert(map->res_count > 0);
+ assert(map->ref_count >= map->res_count);
+ map->res_count++;
+#endif
+ map->ref_count++;
+ lck_mtx_unlock(&map->s_lock);
}
/*
- * Routine: vm_remap
+ * vm_map_deallocate:
*
- * Map portion of a task's address space.
- * Mapped region must not overlap more than
- * one vm memory object. Protections and
- * inheritance attributes remain the same
- * as in the original task and are out parameters.
- * Source and Target task can be identical
- * Other attributes are identical as for vm_map()
+ * Removes a reference from the specified map,
+ * destroying it if no references remain.
+ * The map should not be locked.
*/
-kern_return_t
-vm_remap(
- vm_map_t target_map,
- vm_offset_t *address,
- vm_size_t size,
- vm_offset_t mask,
- boolean_t anywhere,
- vm_map_t src_map,
- vm_offset_t memory_address,
- boolean_t copy,
- vm_prot_t *cur_protection,
- vm_prot_t *max_protection,
- vm_inherit_t inheritance)
+void
+vm_map_deallocate(
+ register vm_map_t map)
{
- kern_return_t result;
- vm_map_entry_t entry;
- vm_map_entry_t insp_entry;
- vm_map_entry_t new_entry;
- struct vm_map_header map_header;
+ unsigned int ref;
- if (target_map == VM_MAP_NULL)
- return KERN_INVALID_ARGUMENT;
+ if (map == VM_MAP_NULL)
+ return;
- switch (inheritance) {
- case VM_INHERIT_NONE:
- case VM_INHERIT_COPY:
- case VM_INHERIT_SHARE:
- if (size != 0 && src_map != VM_MAP_NULL)
- break;
- /*FALL THRU*/
- default:
- return KERN_INVALID_ARGUMENT;
+ lck_mtx_lock(&map->s_lock);
+ ref = --map->ref_count;
+ if (ref > 0) {
+ vm_map_res_deallocate(map);
+ lck_mtx_unlock(&map->s_lock);
+ return;
}
+ assert(map->ref_count == 0);
+ lck_mtx_unlock(&map->s_lock);
- size = round_page_32(size);
+#if TASK_SWAPPER
+ /*
+ * The map residence count isn't decremented here because
+ * the vm_map_delete below will traverse the entire map,
+ * deleting entries, and the residence counts on objects
+ * and sharing maps will go away then.
+ */
+#endif
- result = vm_remap_extract(src_map, memory_address,
- size, copy, &map_header,
- cur_protection,
- max_protection,
- inheritance,
- target_map->hdr.
- entries_pageable);
+ vm_map_destroy(map, VM_MAP_NO_FLAGS);
+}
- if (result != KERN_SUCCESS) {
- return result;
- }
+void
+vm_map_disable_NX(vm_map_t map)
+{
+ if (map == NULL)
+ return;
+ if (map->pmap == NULL)
+ return;
+
+ pmap_disable_NX(map->pmap);
+}
+
+void
+vm_map_disallow_data_exec(vm_map_t map)
+{
+ if (map == NULL)
+ return;
+
+ map->map_disallow_data_exec = TRUE;
+}
+
+/* XXX Consider making these constants (VM_MAX_ADDRESS and MACH_VM_MAX_ADDRESS)
+ * more descriptive.
+ */
+void
+vm_map_set_32bit(vm_map_t map)
+{
+ map->max_offset = (vm_map_offset_t)VM_MAX_ADDRESS;
+}
+
+
+void
+vm_map_set_64bit(vm_map_t map)
+{
+ map->max_offset = (vm_map_offset_t)MACH_VM_MAX_ADDRESS;
+}
+
+vm_map_offset_t
+vm_compute_max_offset(boolean_t is64)
+{
+ return (is64 ? (vm_map_offset_t)MACH_VM_MAX_ADDRESS : (vm_map_offset_t)VM_MAX_ADDRESS);
+}
+
+uint64_t
+vm_map_get_max_aslr_slide_pages(vm_map_t map)
+{
+ return (1 << (vm_map_is_64bit(map) ? 16 : 8));
+}
+
+boolean_t
+vm_map_is_64bit(
+ vm_map_t map)
+{
+ return map->max_offset > ((vm_map_offset_t)VM_MAX_ADDRESS);
+}
+
+boolean_t
+vm_map_has_hard_pagezero(
+ vm_map_t map,
+ vm_map_offset_t pagezero_size)
+{
/*
- * Allocate/check a range of free virtual address
- * space for the target
+ * XXX FBDP
+ * We should lock the VM map (for read) here but we can get away
+ * with it for now because there can't really be any race condition:
+ * the VM map's min_offset is changed only when the VM map is created
+ * and when the zero page is established (when the binary gets loaded),
+ * and this routine gets called only when the task terminates and the
+ * VM map is being torn down, and when a new map is created via
+ * load_machfile()/execve().
*/
- *address = trunc_page_32(*address);
- vm_map_lock(target_map);
- result = vm_remap_range_allocate(target_map, address, size,
- mask, anywhere, &insp_entry);
+ return (map->min_offset >= pagezero_size);
+}
- for (entry = map_header.links.next;
- entry != (struct vm_map_entry *)&map_header.links;
- entry = new_entry) {
- new_entry = entry->vme_next;
- _vm_map_entry_unlink(&map_header, entry);
- if (result == KERN_SUCCESS) {
- entry->vme_start += *address;
- entry->vme_end += *address;
- vm_map_entry_link(target_map, insp_entry, entry);
- insp_entry = entry;
+/*
+ * Raise a VM map's maximun offset.
+ */
+kern_return_t
+vm_map_raise_max_offset(
+ vm_map_t map,
+ vm_map_offset_t new_max_offset)
+{
+ kern_return_t ret;
+
+ vm_map_lock(map);
+ ret = KERN_INVALID_ADDRESS;
+
+ if (new_max_offset >= map->max_offset) {
+ if (!vm_map_is_64bit(map)) {
+ if (new_max_offset <= (vm_map_offset_t)VM_MAX_ADDRESS) {
+ map->max_offset = new_max_offset;
+ ret = KERN_SUCCESS;
+ }
} else {
- if (!entry->is_sub_map) {
- vm_object_deallocate(entry->object.vm_object);
- } else {
- vm_map_deallocate(entry->object.sub_map);
- }
- _vm_map_entry_dispose(&map_header, entry);
+ if (new_max_offset <= (vm_map_offset_t)MACH_VM_MAX_ADDRESS) {
+ map->max_offset = new_max_offset;
+ ret = KERN_SUCCESS;
+ }
}
}
- if (result == KERN_SUCCESS) {
- target_map->size += size;
- SAVE_HINT(target_map, insp_entry);
+ vm_map_unlock(map);
+ return ret;
+}
+
+
+/*
+ * Raise a VM map's minimum offset.
+ * To strictly enforce "page zero" reservation.
+ */
+kern_return_t
+vm_map_raise_min_offset(
+ vm_map_t map,
+ vm_map_offset_t new_min_offset)
+{
+ vm_map_entry_t first_entry;
+
+ new_min_offset = vm_map_round_page(new_min_offset,
+ VM_MAP_PAGE_MASK(map));
+
+ vm_map_lock(map);
+
+ if (new_min_offset < map->min_offset) {
+ /*
+ * Can't move min_offset backwards, as that would expose
+ * a part of the address space that was previously, and for
+ * possibly good reasons, inaccessible.
+ */
+ vm_map_unlock(map);
+ return KERN_INVALID_ADDRESS;
+ }
+ if (new_min_offset >= map->max_offset) {
+ /* can't go beyond the end of the address space */
+ vm_map_unlock(map);
+ return KERN_INVALID_ADDRESS;
+ }
+
+ first_entry = vm_map_first_entry(map);
+ if (first_entry != vm_map_to_entry(map) &&
+ first_entry->vme_start < new_min_offset) {
+ /*
+ * Some memory was already allocated below the new
+ * minimun offset. It's too late to change it now...
+ */
+ vm_map_unlock(map);
+ return KERN_NO_SPACE;
}
- vm_map_unlock(target_map);
- if (result == KERN_SUCCESS && target_map->wiring_required)
- result = vm_map_wire(target_map, *address,
- *address + size, *cur_protection, TRUE);
- return result;
+ map->min_offset = new_min_offset;
+
+ assert(map->holes_list);
+ map->holes_list->start = new_min_offset;
+ assert(new_min_offset < map->holes_list->end);
+
+ vm_map_unlock(map);
+
+ return KERN_SUCCESS;
}
/*
- * Routine: vm_remap_range_allocate
- *
- * Description:
- * Allocate a range in the specified virtual address map.
- * returns the address and the map entry just before the allocated
- * range
- *
- * Map must be locked.
+ * Set the limit on the maximum amount of user wired memory allowed for this map.
+ * This is basically a copy of the MEMLOCK rlimit value maintained by the BSD side of
+ * the kernel. The limits are checked in the mach VM side, so we keep a copy so we
+ * don't have to reach over to the BSD data structures.
*/
-kern_return_t
-vm_remap_range_allocate(
- vm_map_t map,
- vm_offset_t *address, /* IN/OUT */
- vm_size_t size,
- vm_offset_t mask,
- boolean_t anywhere,
- vm_map_entry_t *map_entry) /* OUT */
+void
+vm_map_set_user_wire_limit(vm_map_t map,
+ vm_size_t limit)
{
- register vm_map_entry_t entry;
- register vm_offset_t start;
- register vm_offset_t end;
- kern_return_t result = KERN_SUCCESS;
+ map->user_wire_limit = limit;
+}
+
+
+void vm_map_switch_protect(vm_map_t map,
+ boolean_t val)
+{
+ vm_map_lock(map);
+ map->switch_protect=val;
+ vm_map_unlock(map);
+}
- StartAgain: ;
+/*
+ * IOKit has mapped a region into this map; adjust the pmap's ledgers appropriately.
+ * phys_footprint is a composite limit consisting of iokit + physmem, so we need to
+ * bump both counters.
+ */
+void
+vm_map_iokit_mapped_region(vm_map_t map, vm_size_t bytes)
+{
+ pmap_t pmap = vm_map_pmap(map);
- start = *address;
+ ledger_credit(pmap->ledger, task_ledgers.iokit_mapped, bytes);
+ ledger_credit(pmap->ledger, task_ledgers.phys_footprint, bytes);
+}
- if (anywhere)
- {
- /*
- * Calculate the first possible address.
- */
+void
+vm_map_iokit_unmapped_region(vm_map_t map, vm_size_t bytes)
+{
+ pmap_t pmap = vm_map_pmap(map);
- if (start < map->min_offset)
- start = map->min_offset;
- if (start > map->max_offset)
- return(KERN_NO_SPACE);
-
+ ledger_debit(pmap->ledger, task_ledgers.iokit_mapped, bytes);
+ ledger_debit(pmap->ledger, task_ledgers.phys_footprint, bytes);
+}
+
+/* Add (generate) code signature for memory range */
+#if CONFIG_DYNAMIC_CODE_SIGNING
+kern_return_t vm_map_sign(vm_map_t map,
+ vm_map_offset_t start,
+ vm_map_offset_t end)
+{
+ vm_map_entry_t entry;
+ vm_page_t m;
+ vm_object_t object;
+
/*
- * Look for the first possible address;
- * if there's already something at this
- * address, we have to start after it.
+ * Vet all the input parameters and current type and state of the
+ * underlaying object. Return with an error if anything is amiss.
*/
-
- assert(first_free_is_valid(map));
- if (start == map->min_offset) {
- if ((entry = map->first_free) != vm_map_to_entry(map))
- start = entry->vme_end;
- } else {
- vm_map_entry_t tmp_entry;
- if (vm_map_lookup_entry(map, start, &tmp_entry))
- start = tmp_entry->vme_end;
- entry = tmp_entry;
+ if (map == VM_MAP_NULL)
+ return(KERN_INVALID_ARGUMENT);
+
+ vm_map_lock_read(map);
+
+ if (!vm_map_lookup_entry(map, start, &entry) || entry->is_sub_map) {
+ /*
+ * Must pass a valid non-submap address.
+ */
+ vm_map_unlock_read(map);
+ return(KERN_INVALID_ADDRESS);
}
+
+ if((entry->vme_start > start) || (entry->vme_end < end)) {
+ /*
+ * Map entry doesn't cover the requested range. Not handling
+ * this situation currently.
+ */
+ vm_map_unlock_read(map);
+ return(KERN_INVALID_ARGUMENT);
+ }
+
+ object = VME_OBJECT(entry);
+ if (object == VM_OBJECT_NULL) {
+ /*
+ * Object must already be present or we can't sign.
+ */
+ vm_map_unlock_read(map);
+ return KERN_INVALID_ARGUMENT;
+ }
+
+ vm_object_lock(object);
+ vm_map_unlock_read(map);
+
+ while(start < end) {
+ uint32_t refmod;
- /*
- * In any case, the "entry" always precedes
- * the proposed new region throughout the
- * loop:
- */
-
- while (TRUE) {
- register vm_map_entry_t next;
-
- /*
- * Find the end of the proposed new region.
- * Be sure we didn't go beyond the end, or
- * wrap around the address.
- */
-
- end = ((start + mask) & ~mask);
- if (end < start)
- return(KERN_NO_SPACE);
- start = end;
- end += size;
-
- if ((end > map->max_offset) || (end < start)) {
- if (map->wait_for_space) {
- if (size <= (map->max_offset -
- map->min_offset)) {
- assert_wait((event_t) map, THREAD_INTERRUPTIBLE);
- vm_map_unlock(map);
- thread_block((void (*)(void))0);
- vm_map_lock(map);
- goto StartAgain;
- }
+ m = vm_page_lookup(object,
+ start - entry->vme_start + VME_OFFSET(entry));
+ if (m==VM_PAGE_NULL) {
+ /* shoud we try to fault a page here? we can probably
+ * demand it exists and is locked for this request */
+ vm_object_unlock(object);
+ return KERN_FAILURE;
}
+ /* deal with special page status */
+ if (m->busy ||
+ (m->unusual && (m->error || m->restart || m->private || m->absent))) {
+ vm_object_unlock(object);
+ return KERN_FAILURE;
+ }
+
+ /* Page is OK... now "validate" it */
+ /* This is the place where we'll call out to create a code
+ * directory, later */
+ m->cs_validated = TRUE;
+
+ /* The page is now "clean" for codesigning purposes. That means
+ * we don't consider it as modified (wpmapped) anymore. But
+ * we'll disconnect the page so we note any future modification
+ * attempts. */
+ m->wpmapped = FALSE;
+ refmod = pmap_disconnect(m->phys_page);
- return(KERN_NO_SPACE);
- }
+ /* Pull the dirty status from the pmap, since we cleared the
+ * wpmapped bit */
+ if ((refmod & VM_MEM_MODIFIED) && !m->dirty) {
+ SET_PAGE_DIRTY(m, FALSE);
+ }
+
+ /* On to the next page */
+ start += PAGE_SIZE;
+ }
+ vm_object_unlock(object);
+
+ return KERN_SUCCESS;
+}
+#endif
+
+kern_return_t vm_map_partial_reap(vm_map_t map, unsigned int *reclaimed_resident, unsigned int *reclaimed_compressed)
+{
+ vm_map_entry_t entry = VM_MAP_ENTRY_NULL;
+ vm_map_entry_t next_entry;
+ kern_return_t kr = KERN_SUCCESS;
+ vm_map_t zap_map;
- /*
- * If there are no more entries, we must win.
- */
+ vm_map_lock(map);
- next = entry->vme_next;
- if (next == vm_map_to_entry(map))
- break;
+ /*
+ * We use a "zap_map" to avoid having to unlock
+ * the "map" in vm_map_delete().
+ */
+ zap_map = vm_map_create(PMAP_NULL,
+ map->min_offset,
+ map->max_offset,
+ map->hdr.entries_pageable);
- /*
- * If there is another entry, it must be
- * after the end of the potential new region.
- */
+ if (zap_map == VM_MAP_NULL) {
+ return KERN_RESOURCE_SHORTAGE;
+ }
- if (next->vme_start >= end)
- break;
+ vm_map_set_page_shift(zap_map,
+ VM_MAP_PAGE_SHIFT(map));
+ vm_map_disable_hole_optimization(zap_map);
+
+ for (entry = vm_map_first_entry(map);
+ entry != vm_map_to_entry(map);
+ entry = next_entry) {
+ next_entry = entry->vme_next;
+
+ if (VME_OBJECT(entry) &&
+ !entry->is_sub_map &&
+ (VME_OBJECT(entry)->internal == TRUE) &&
+ (VME_OBJECT(entry)->ref_count == 1)) {
- /*
- * Didn't fit -- move to the next entry.
- */
+ *reclaimed_resident += VME_OBJECT(entry)->resident_page_count;
+ *reclaimed_compressed += vm_compressor_pager_get_count(VME_OBJECT(entry)->pager);
- entry = next;
- start = entry->vme_end;
+ (void)vm_map_delete(map,
+ entry->vme_start,
+ entry->vme_end,
+ VM_MAP_REMOVE_SAVE_ENTRIES,
+ zap_map);
+ }
}
- *address = start;
- } else {
- vm_map_entry_t temp_entry;
-
- /*
- * Verify that:
- * the address doesn't itself violate
- * the mask requirement.
- */
- if ((start & mask) != 0)
- return(KERN_NO_SPACE);
+ vm_map_unlock(map);
+ /*
+ * Get rid of the "zap_maps" and all the map entries that
+ * they may still contain.
+ */
+ if (zap_map != VM_MAP_NULL) {
+ vm_map_destroy(zap_map, VM_MAP_REMOVE_NO_PMAP_CLEANUP);
+ zap_map = VM_MAP_NULL;
+ }
- /*
- * ... the address is within bounds
- */
+ return kr;
+}
- end = start + size;
+#if CONFIG_FREEZE
+
+kern_return_t vm_map_freeze_walk(
+ vm_map_t map,
+ unsigned int *purgeable_count,
+ unsigned int *wired_count,
+ unsigned int *clean_count,
+ unsigned int *dirty_count,
+ unsigned int dirty_budget,
+ boolean_t *has_shared)
+{
+ vm_map_entry_t entry;
+
+ vm_map_lock_read(map);
+
+ *purgeable_count = *wired_count = *clean_count = *dirty_count = 0;
+ *has_shared = FALSE;
+
+ for (entry = vm_map_first_entry(map);
+ entry != vm_map_to_entry(map);
+ entry = entry->vme_next) {
+ unsigned int purgeable, clean, dirty, wired;
+ boolean_t shared;
+
+ if ((VME_OBJECT(entry) == 0) ||
+ (entry->is_sub_map) ||
+ (VME_OBJECT(entry)->phys_contiguous)) {
+ continue;
+ }
- if ((start < map->min_offset) ||
- (end > map->max_offset) ||
- (start >= end)) {
- return(KERN_INVALID_ADDRESS);
+ default_freezer_pack(&purgeable, &wired, &clean, &dirty, dirty_budget, &shared, VME_OBJECT(entry), NULL);
+
+ *purgeable_count += purgeable;
+ *wired_count += wired;
+ *clean_count += clean;
+ *dirty_count += dirty;
+
+ if (shared) {
+ *has_shared = TRUE;
+ }
+
+ /* Adjust pageout budget and finish up if reached */
+ if (dirty_budget) {
+ dirty_budget -= dirty;
+ if (dirty_budget == 0) {
+ break;
+ }
+ }
}
- /*
- * ... the starting address isn't allocated
- */
+ vm_map_unlock_read(map);
+
+ return KERN_SUCCESS;
+}
- if (vm_map_lookup_entry(map, start, &temp_entry))
- return(KERN_NO_SPACE);
+int c_freezer_swapout_count;
+int c_freezer_compression_count = 0;
+AbsoluteTime c_freezer_last_yield_ts = 0;
+
+kern_return_t vm_map_freeze(
+ vm_map_t map,
+ unsigned int *purgeable_count,
+ unsigned int *wired_count,
+ unsigned int *clean_count,
+ unsigned int *dirty_count,
+ unsigned int dirty_budget,
+ boolean_t *has_shared)
+{
+ vm_map_entry_t entry2 = VM_MAP_ENTRY_NULL;
+ kern_return_t kr = KERN_SUCCESS;
+ boolean_t default_freezer_active = TRUE;
- entry = temp_entry;
+ *purgeable_count = *wired_count = *clean_count = *dirty_count = 0;
+ *has_shared = FALSE;
/*
- * ... the next region doesn't overlap the
- * end point.
+ * We need the exclusive lock here so that we can
+ * block any page faults or lookups while we are
+ * in the middle of freezing this vm map.
*/
+ vm_map_lock(map);
- if ((entry->vme_next != vm_map_to_entry(map)) &&
- (entry->vme_next->vme_start < end))
- return(KERN_NO_SPACE);
- }
- *map_entry = entry;
- return(KERN_SUCCESS);
-}
+ if (COMPRESSED_PAGER_IS_ACTIVE || DEFAULT_FREEZER_COMPRESSED_PAGER_IS_ACTIVE) {
+ default_freezer_active = FALSE;
+
+ if (vm_compressor_low_on_space() || vm_swap_low_on_space()) {
+ kr = KERN_NO_SPACE;
+ goto done;
+ }
+ }
+ assert(default_freezer_active == FALSE);
+
+ if (default_freezer_active) {
+ if (map->default_freezer_handle == NULL) {
+ map->default_freezer_handle = default_freezer_handle_allocate();
+ }
+
+ if ((kr = default_freezer_handle_init(map->default_freezer_handle)) != KERN_SUCCESS) {
+ /*
+ * Can happen if default_freezer_handle passed in is NULL
+ * Or, a table has already been allocated and associated
+ * with this handle, i.e. the map is already frozen.
+ */
+ goto done;
+ }
+ }
+ c_freezer_compression_count = 0;
+ clock_get_uptime(&c_freezer_last_yield_ts);
-/*
- * vm_map_switch:
- *
- * Set the address map for the current thr_act to the specified map
- */
+ for (entry2 = vm_map_first_entry(map);
+ entry2 != vm_map_to_entry(map);
+ entry2 = entry2->vme_next) {
+
+ vm_object_t src_object = VME_OBJECT(entry2);
+
+ if (VME_OBJECT(entry2) &&
+ !entry2->is_sub_map &&
+ !VME_OBJECT(entry2)->phys_contiguous) {
+ /* If eligible, scan the entry, moving eligible pages over to our parent object */
+ if (default_freezer_active) {
+ unsigned int purgeable, clean, dirty, wired;
+ boolean_t shared;
+
+ default_freezer_pack(&purgeable, &wired, &clean, &dirty, dirty_budget, &shared,
+ src_object, map->default_freezer_handle);
+
+ *purgeable_count += purgeable;
+ *wired_count += wired;
+ *clean_count += clean;
+ *dirty_count += dirty;
+
+ /* Adjust pageout budget and finish up if reached */
+ if (dirty_budget) {
+ dirty_budget -= dirty;
+ if (dirty_budget == 0) {
+ break;
+ }
+ }
-vm_map_t
-vm_map_switch(
- vm_map_t map)
-{
- int mycpu;
- thread_act_t thr_act = current_act();
- vm_map_t oldmap = thr_act->map;
+ if (shared) {
+ *has_shared = TRUE;
+ }
+ } else {
+ if (VME_OBJECT(entry2)->internal == TRUE) {
+
+ if (DEFAULT_FREEZER_COMPRESSED_PAGER_IS_SWAPBACKED) {
+ /*
+ * Pages belonging to this object could be swapped to disk.
+ * Make sure it's not a shared object because we could end
+ * up just bringing it back in again.
+ */
+ if (VME_OBJECT(entry2)->ref_count > 1) {
+ continue;
+ }
+ }
+ vm_object_compressed_freezer_pageout(VME_OBJECT(entry2));
+ }
- mp_disable_preemption();
- mycpu = cpu_number();
+ if (vm_compressor_low_on_space() || vm_swap_low_on_space()) {
+ kr = KERN_NO_SPACE;
+ break;
+ }
+ }
+ }
+ }
- /*
- * Deactivate the current map and activate the requested map
- */
- PMAP_SWITCH_USER(thr_act, map, mycpu);
+ if (default_freezer_active) {
+ /* Finally, throw out the pages to swap */
+ default_freezer_pageout(map->default_freezer_handle);
+ }
- mp_enable_preemption();
- return(oldmap);
+done:
+ vm_map_unlock(map);
+
+ if (!default_freezer_active) {
+ vm_object_compressed_freezer_done();
+ }
+ if (DEFAULT_FREEZER_COMPRESSED_PAGER_IS_SWAPBACKED) {
+ /*
+ * reset the counter tracking the # of swapped c_segs
+ * because we are now done with this freeze session and task.
+ */
+ c_freezer_swapout_count = 0;
+ }
+ return kr;
}
-
-/*
- * Routine: vm_map_write_user
- *
- * Description:
- * Copy out data from a kernel space into space in the
- * destination map. The space must already exist in the
- * destination map.
- * NOTE: This routine should only be called by threads
- * which can block on a page fault. i.e. kernel mode user
- * threads.
- *
- */
kern_return_t
-vm_map_write_user(
- vm_map_t map,
- vm_offset_t src_addr,
- vm_offset_t dst_addr,
- vm_size_t size)
+vm_map_thaw(
+ vm_map_t map)
{
- thread_act_t thr_act = current_act();
- kern_return_t kr = KERN_SUCCESS;
+ kern_return_t kr = KERN_SUCCESS;
- if(thr_act->map == map) {
- if (copyout((char *)src_addr, (char *)dst_addr, size)) {
- kr = KERN_INVALID_ADDRESS;
- }
- } else {
- vm_map_t oldmap;
+ if (COMPRESSED_PAGER_IS_ACTIVE || DEFAULT_FREEZER_COMPRESSED_PAGER_IS_ACTIVE) {
+ /*
+ * We will on-demand thaw in the presence of the compressed pager.
+ */
+ return kr;
+ }
- /* take on the identity of the target map while doing */
- /* the transfer */
+ vm_map_lock(map);
- vm_map_reference(map);
- oldmap = vm_map_switch(map);
- if (copyout((char *)src_addr, (char *)dst_addr, size)) {
- kr = KERN_INVALID_ADDRESS;
- }
- vm_map_switch(oldmap);
- vm_map_deallocate(map);
+ if (map->default_freezer_handle == NULL) {
+ /*
+ * This map is not in a frozen state.
+ */
+ kr = KERN_FAILURE;
+ goto out;
}
+
+ kr = default_freezer_unpack(map->default_freezer_handle);
+out:
+ vm_map_unlock(map);
+
return kr;
}
-
-/*
- * Routine: vm_map_read_user
- *
- * Description:
- * Copy in data from a user space source map into the
- * kernel map. The space must already exist in the
- * kernel map.
- * NOTE: This routine should only be called by threads
- * which can block on a page fault. i.e. kernel mode user
- * threads.
+#endif
+
+/*
+ * vm_map_entry_should_cow_for_true_share:
+ *
+ * Determines if the map entry should be clipped and setup for copy-on-write
+ * to avoid applying "true_share" to a large VM object when only a subset is
+ * targeted.
+ *
+ * For now, we target only the map entries created for the Objective C
+ * Garbage Collector, which initially have the following properties:
+ * - alias == VM_MEMORY_MALLOC
+ * - wired_count == 0
+ * - !needs_copy
+ * and a VM object with:
+ * - internal
+ * - copy_strategy == MEMORY_OBJECT_COPY_SYMMETRIC
+ * - !true_share
+ * - vo_size == ANON_CHUNK_SIZE
*
+ * Only non-kernel map entries.
*/
-kern_return_t
-vm_map_read_user(
- vm_map_t map,
- vm_offset_t src_addr,
- vm_offset_t dst_addr,
- vm_size_t size)
+boolean_t
+vm_map_entry_should_cow_for_true_share(
+ vm_map_entry_t entry)
{
- thread_act_t thr_act = current_act();
- kern_return_t kr = KERN_SUCCESS;
+ vm_object_t object;
- if(thr_act->map == map) {
- if (copyin((char *)src_addr, (char *)dst_addr, size)) {
- kr = KERN_INVALID_ADDRESS;
- }
- } else {
- vm_map_t oldmap;
+ if (entry->is_sub_map) {
+ /* entry does not point at a VM object */
+ return FALSE;
+ }
- /* take on the identity of the target map while doing */
- /* the transfer */
+ if (entry->needs_copy) {
+ /* already set for copy_on_write: done! */
+ return FALSE;
+ }
- vm_map_reference(map);
- oldmap = vm_map_switch(map);
- if (copyin((char *)src_addr, (char *)dst_addr, size)) {
- kr = KERN_INVALID_ADDRESS;
- }
- vm_map_switch(oldmap);
- vm_map_deallocate(map);
+ if (VME_ALIAS(entry) != VM_MEMORY_MALLOC &&
+ VME_ALIAS(entry) != VM_MEMORY_MALLOC_SMALL) {
+ /* not a malloc heap or Obj-C Garbage Collector heap */
+ return FALSE;
}
- return kr;
-}
-/* Takes existing source and destination sub-maps and clones the contents of */
-/* the source map */
+ if (entry->wired_count) {
+ /* wired: can't change the map entry... */
+ vm_counters.should_cow_but_wired++;
+ return FALSE;
+ }
-kern_return_t
-vm_region_clone(
- ipc_port_t src_region,
- ipc_port_t dst_region)
-{
- vm_named_entry_t src_object;
- vm_named_entry_t dst_object;
- vm_map_t src_map;
- vm_map_t dst_map;
- vm_offset_t addr;
- vm_offset_t max_off;
- vm_map_entry_t entry;
- vm_map_entry_t new_entry;
- vm_map_entry_t insert_point;
+ object = VME_OBJECT(entry);
- src_object = (vm_named_entry_t)src_region->ip_kobject;
- dst_object = (vm_named_entry_t)dst_region->ip_kobject;
- if((!src_object->is_sub_map) || (!dst_object->is_sub_map)) {
- return KERN_INVALID_ARGUMENT;
+ if (object == VM_OBJECT_NULL) {
+ /* no object yet... */
+ return FALSE;
}
- src_map = (vm_map_t)src_object->backing.map;
- dst_map = (vm_map_t)dst_object->backing.map;
- /* destination map is assumed to be unavailable to any other */
- /* activity. i.e. it is new */
- vm_map_lock(src_map);
- if((src_map->min_offset != dst_map->min_offset)
- || (src_map->max_offset != dst_map->max_offset)) {
- vm_map_unlock(src_map);
- return KERN_INVALID_ARGUMENT;
+
+ if (!object->internal) {
+ /* not an internal object */
+ return FALSE;
}
- addr = src_map->min_offset;
- vm_map_lookup_entry(dst_map, addr, &entry);
- if(entry == vm_map_to_entry(dst_map)) {
- entry = entry->vme_next;
+
+ if (object->copy_strategy != MEMORY_OBJECT_COPY_SYMMETRIC) {
+ /* not the default copy strategy */
+ return FALSE;
}
- if(entry == vm_map_to_entry(dst_map)) {
- max_off = src_map->max_offset;
- } else {
- max_off = entry->vme_start;
+
+ if (object->true_share) {
+ /* already true_share: too late to avoid it */
+ return FALSE;
}
- vm_map_lookup_entry(src_map, addr, &entry);
- if(entry == vm_map_to_entry(src_map)) {
- entry = entry->vme_next;
+
+ if (VME_ALIAS(entry) == VM_MEMORY_MALLOC &&
+ object->vo_size != ANON_CHUNK_SIZE) {
+ /* ... not an object created for the ObjC Garbage Collector */
+ return FALSE;
}
- vm_map_lookup_entry(dst_map, addr, &insert_point);
- while((entry != vm_map_to_entry(src_map)) &&
- (entry->vme_end <= max_off)) {
- addr = entry->vme_start;
- new_entry = vm_map_entry_create(dst_map);
- vm_map_entry_copy(new_entry, entry);
- vm_map_entry_link(dst_map, insert_point, new_entry);
- insert_point = new_entry;
- if (entry->object.vm_object != VM_OBJECT_NULL) {
- if (new_entry->is_sub_map) {
- vm_map_reference(new_entry->object.sub_map);
- } else {
- vm_object_reference(
- new_entry->object.vm_object);
- }
- }
- dst_map->size += new_entry->vme_end - new_entry->vme_start;
- entry = entry->vme_next;
+
+ if (VME_ALIAS(entry) == VM_MEMORY_MALLOC_SMALL &&
+ object->vo_size != 2048 * 4096) {
+ /* ... not a "MALLOC_SMALL" heap */
+ return FALSE;
}
- vm_map_unlock(src_map);
- return KERN_SUCCESS;
+
+ /*
+ * All the criteria match: we have a large object being targeted for "true_share".
+ * To limit the adverse side-effects linked with "true_share", tell the caller to
+ * try and avoid setting up the entire object for "true_share" by clipping the
+ * targeted range and setting it up for copy-on-write.
+ */
+ return TRUE;
}
-/*
- * Export routines to other components for the things we access locally through
- * macros.
- */
-#undef current_map
-vm_map_t
-current_map(void)
+vm_map_offset_t
+vm_map_round_page_mask(
+ vm_map_offset_t offset,
+ vm_map_offset_t mask)
{
- return (current_map_fast());
+ return VM_MAP_ROUND_PAGE(offset, mask);
}
-/*
- * vm_map_check_protection:
- *
- * Assert that the target map allows the specified
- * privilege on the entire address region given.
- * The entire region must be allocated.
- */
-boolean_t vm_map_check_protection(map, start, end, protection)
- register vm_map_t map;
- register vm_offset_t start;
- register vm_offset_t end;
- register vm_prot_t protection;
+vm_map_offset_t
+vm_map_trunc_page_mask(
+ vm_map_offset_t offset,
+ vm_map_offset_t mask)
{
- register vm_map_entry_t entry;
- vm_map_entry_t tmp_entry;
+ return VM_MAP_TRUNC_PAGE(offset, mask);
+}
- vm_map_lock(map);
+boolean_t
+vm_map_page_aligned(
+ vm_map_offset_t offset,
+ vm_map_offset_t mask)
+{
+ return ((offset) & mask) == 0;
+}
- if (start < vm_map_min(map) || end > vm_map_max(map) || start > end)
- {
- vm_map_unlock(map);
- return (FALSE);
- }
+int
+vm_map_page_shift(
+ vm_map_t map)
+{
+ return VM_MAP_PAGE_SHIFT(map);
+}
- if (!vm_map_lookup_entry(map, start, &tmp_entry)) {
- vm_map_unlock(map);
- return(FALSE);
- }
+int
+vm_map_page_size(
+ vm_map_t map)
+{
+ return VM_MAP_PAGE_SIZE(map);
+}
- entry = tmp_entry;
+vm_map_offset_t
+vm_map_page_mask(
+ vm_map_t map)
+{
+ return VM_MAP_PAGE_MASK(map);
+}
- while (start < end) {
- if (entry == vm_map_to_entry(map)) {
- vm_map_unlock(map);
- return(FALSE);
- }
+kern_return_t
+vm_map_set_page_shift(
+ vm_map_t map,
+ int pageshift)
+{
+ if (map->hdr.nentries != 0) {
+ /* too late to change page size */
+ return KERN_FAILURE;
+ }
- /*
- * No holes allowed!
- */
+ map->hdr.page_shift = pageshift;
- if (start < entry->vme_start) {
- vm_map_unlock(map);
- return(FALSE);
- }
+ return KERN_SUCCESS;
+}
- /*
- * Check protection associated with entry.
- */
+int
+vm_map_purge(
+ vm_map_t map)
+{
+ int num_object_purged;
+ vm_map_entry_t entry;
+ vm_map_offset_t next_address;
+ vm_object_t object;
+ int state;
+ kern_return_t kr;
- if ((entry->protection & protection) != protection) {
- vm_map_unlock(map);
- return(FALSE);
+ num_object_purged = 0;
+
+ vm_map_lock_read(map);
+ entry = vm_map_first_entry(map);
+ while (entry != vm_map_to_entry(map)) {
+ if (entry->is_sub_map) {
+ goto next;
+ }
+ if (! (entry->protection & VM_PROT_WRITE)) {
+ goto next;
+ }
+ object = VME_OBJECT(entry);
+ if (object == VM_OBJECT_NULL) {
+ goto next;
+ }
+ if (object->purgable != VM_PURGABLE_VOLATILE) {
+ goto next;
}
- /* go to next entry */
+ vm_object_lock(object);
+#if 00
+ if (VME_OFFSET(entry) != 0 ||
+ (entry->vme_end - entry->vme_start) != object->vo_size) {
+ vm_object_unlock(object);
+ goto next;
+ }
+#endif
+ next_address = entry->vme_end;
+ vm_map_unlock_read(map);
+ state = VM_PURGABLE_EMPTY;
+ kr = vm_object_purgable_control(object,
+ VM_PURGABLE_SET_STATE,
+ &state);
+ if (kr == KERN_SUCCESS) {
+ num_object_purged++;
+ }
+ vm_object_unlock(object);
- start = entry->vme_end;
+ vm_map_lock_read(map);
+ if (vm_map_lookup_entry(map, next_address, &entry)) {
+ continue;
+ }
+ next:
entry = entry->vme_next;
}
- vm_map_unlock(map);
- return(TRUE);
+ vm_map_unlock_read(map);
+
+ return num_object_purged;
}
-/*
- * This routine is obsolete, but included for backward
- * compatibility for older drivers.
- */
-void
-kernel_vm_map_reference(
- vm_map_t map)
+kern_return_t
+vm_map_query_volatile(
+ vm_map_t map,
+ mach_vm_size_t *volatile_virtual_size_p,
+ mach_vm_size_t *volatile_resident_size_p,
+ mach_vm_size_t *volatile_compressed_size_p,
+ mach_vm_size_t *volatile_pmap_size_p,
+ mach_vm_size_t *volatile_compressed_pmap_size_p)
{
- vm_map_reference(map);
+ mach_vm_size_t volatile_virtual_size;
+ mach_vm_size_t volatile_resident_count;
+ mach_vm_size_t volatile_compressed_count;
+ mach_vm_size_t volatile_pmap_count;
+ mach_vm_size_t volatile_compressed_pmap_count;
+ mach_vm_size_t resident_count;
+ vm_map_entry_t entry;
+ vm_object_t object;
+
+ /* map should be locked by caller */
+
+ volatile_virtual_size = 0;
+ volatile_resident_count = 0;
+ volatile_compressed_count = 0;
+ volatile_pmap_count = 0;
+ volatile_compressed_pmap_count = 0;
+
+ for (entry = vm_map_first_entry(map);
+ entry != vm_map_to_entry(map);
+ entry = entry->vme_next) {
+ mach_vm_size_t pmap_resident_bytes, pmap_compressed_bytes;
+
+ if (entry->is_sub_map) {
+ continue;
+ }
+ if (! (entry->protection & VM_PROT_WRITE)) {
+ continue;
+ }
+ object = VME_OBJECT(entry);
+ if (object == VM_OBJECT_NULL) {
+ continue;
+ }
+ if (object->purgable != VM_PURGABLE_VOLATILE &&
+ object->purgable != VM_PURGABLE_EMPTY) {
+ continue;
+ }
+ if (VME_OFFSET(entry)) {
+ /*
+ * If the map entry has been split and the object now
+ * appears several times in the VM map, we don't want
+ * to count the object's resident_page_count more than
+ * once. We count it only for the first one, starting
+ * at offset 0 and ignore the other VM map entries.
+ */
+ continue;
+ }
+ resident_count = object->resident_page_count;
+ if ((VME_OFFSET(entry) / PAGE_SIZE) >= resident_count) {
+ resident_count = 0;
+ } else {
+ resident_count -= (VME_OFFSET(entry) / PAGE_SIZE);
+ }
+
+ volatile_virtual_size += entry->vme_end - entry->vme_start;
+ volatile_resident_count += resident_count;
+ if (object->pager) {
+ volatile_compressed_count +=
+ vm_compressor_pager_get_count(object->pager);
+ }
+ pmap_compressed_bytes = 0;
+ pmap_resident_bytes =
+ pmap_query_resident(map->pmap,
+ entry->vme_start,
+ entry->vme_end,
+ &pmap_compressed_bytes);
+ volatile_pmap_count += (pmap_resident_bytes / PAGE_SIZE);
+ volatile_compressed_pmap_count += (pmap_compressed_bytes
+ / PAGE_SIZE);
+ }
+
+ /* map is still locked on return */
+
+ *volatile_virtual_size_p = volatile_virtual_size;
+ *volatile_resident_size_p = volatile_resident_count * PAGE_SIZE;
+ *volatile_compressed_size_p = volatile_compressed_count * PAGE_SIZE;
+ *volatile_pmap_size_p = volatile_pmap_count * PAGE_SIZE;
+ *volatile_compressed_pmap_size_p = volatile_compressed_pmap_count * PAGE_SIZE;
+
+ return KERN_SUCCESS;
}
-/*
- * vm_map_reference:
- *
- * Most code internal to the osfmk will go through a
- * macro defining this. This is always here for the
- * use of other kernel components.
- */
-#undef vm_map_reference
void
-vm_map_reference(
- register vm_map_t map)
+vm_map_sizes(vm_map_t map,
+ vm_map_size_t * psize,
+ vm_map_size_t * pfree,
+ vm_map_size_t * plargest_free)
{
- if (map == VM_MAP_NULL)
- return;
+ vm_map_entry_t entry;
+ vm_map_offset_t prev;
+ vm_map_size_t free, total_free, largest_free;
+ boolean_t end;
- mutex_lock(&map->s_lock);
-#if TASK_SWAPPER
- assert(map->res_count > 0);
- assert(map->ref_count >= map->res_count);
- map->res_count++;
-#endif
- map->ref_count++;
- mutex_unlock(&map->s_lock);
+ total_free = largest_free = 0;
+
+ vm_map_lock_read(map);
+ if (psize) *psize = map->max_offset - map->min_offset;
+
+ prev = map->min_offset;
+ for (entry = vm_map_first_entry(map);; entry = entry->vme_next)
+ {
+ end = (entry == vm_map_to_entry(map));
+
+ if (end) free = entry->vme_end - prev;
+ else free = entry->vme_start - prev;
+
+ total_free += free;
+ if (free > largest_free) largest_free = free;
+
+ if (end) break;
+ prev = entry->vme_end;
+ }
+ vm_map_unlock_read(map);
+ if (pfree) *pfree = total_free;
+ if (plargest_free) *plargest_free = largest_free;
}
-/*
- * vm_map_deallocate:
- *
- * Removes a reference from the specified map,
- * destroying it if no references remain.
- * The map should not be locked.
- */
-void
-vm_map_deallocate(
- register vm_map_t map)
+#if VM_SCAN_FOR_SHADOW_CHAIN
+int vm_map_shadow_max(vm_map_t map);
+int vm_map_shadow_max(
+ vm_map_t map)
{
- unsigned int ref;
+ int shadows, shadows_max;
+ vm_map_entry_t entry;
+ vm_object_t object, next_object;
- if (map == VM_MAP_NULL)
- return;
+ if (map == NULL)
+ return 0;
- mutex_lock(&map->s_lock);
- ref = --map->ref_count;
- if (ref > 0) {
- vm_map_res_deallocate(map);
- mutex_unlock(&map->s_lock);
- return;
+ shadows_max = 0;
+
+ vm_map_lock_read(map);
+
+ for (entry = vm_map_first_entry(map);
+ entry != vm_map_to_entry(map);
+ entry = entry->vme_next) {
+ if (entry->is_sub_map) {
+ continue;
+ }
+ object = VME_OBJECT(entry);
+ if (object == NULL) {
+ continue;
+ }
+ vm_object_lock_shared(object);
+ for (shadows = 0;
+ object->shadow != NULL;
+ shadows++, object = next_object) {
+ next_object = object->shadow;
+ vm_object_lock_shared(next_object);
+ vm_object_unlock(object);
+ }
+ vm_object_unlock(object);
+ if (shadows > shadows_max) {
+ shadows_max = shadows;
+ }
}
- assert(map->ref_count == 0);
- mutex_unlock(&map->s_lock);
-#if TASK_SWAPPER
- /*
- * The map residence count isn't decremented here because
- * the vm_map_delete below will traverse the entire map,
- * deleting entries, and the residence counts on objects
- * and sharing maps will go away then.
- */
-#endif
+ vm_map_unlock_read(map);
- vm_map_destroy(map);
+ return shadows_max;
}
+#endif /* VM_SCAN_FOR_SHADOW_CHAIN */
projects / apple / xnu.git / blobdiff