* Copyright (c) 2000-2012 Apple Inc. All rights reserved.
*
* @APPLE_OSREFERENCE_LICENSE_HEADER_START@
- *
+ *
* 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
* 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.
- *
+ *
* 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,
* 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_OSREFERENCE_LICENSE_HEADER_END@
*/
/*
* @OSF_COPYRIGHT@
*/
-/*
+/*
* Mach Operating System
* Copyright (c) 1991,1990,1989,1988,1987 Carnegie Mellon University
* All Rights Reserved.
- *
+ *
* Permission to use, copy, modify and distribute this software and its
* documentation is hereby granted, provided that both the copyright
* notice and this permission notice appear in all copies of the
* software, derivative works or modified versions, and any portions
* thereof, and that both notices appear in supporting documentation.
- *
+ *
* CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
* CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND FOR
* ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
- *
+ *
* Carnegie Mellon requests users of this software to return to
- *
+ *
* Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
* School of Computer Science
* Carnegie Mellon University
* Pittsburgh PA 15213-3890
- *
+ *
* any improvements or extensions that they make and grant Carnegie Mellon
* the rights to redistribute these changes.
*/
#include <task_swapper.h>
#include <mach_assert.h>
+
+#include <vm/vm_options.h>
+
#include <libkern/OSAtomic.h>
#include <mach/kern_return.h>
#include <mach/sdt.h>
#include <kern/assert.h>
+#include <kern/backtrace.h>
#include <kern/counters.h>
+#include <kern/exc_guard.h>
#include <kern/kalloc.h>
#include <kern/zalloc.h>
#include <vm/cpm.h>
+#include <vm/vm_compressor.h>
#include <vm/vm_compressor_pager.h>
#include <vm/vm_init.h>
#include <vm/vm_fault.h>
#include <vm/vm_object.h>
#include <vm/vm_page.h>
#include <vm/vm_pageout.h>
+#include <vm/pmap.h>
#include <vm/vm_kern.h>
#include <ipc/ipc_port.h>
#include <kern/sched_prim.h>
#include <vm/vm_shared_region.h>
#include <vm/vm_map_store.h>
+#include <san/kasan.h>
+
+#include <sys/codesign.h>
+#include <libkern/section_keywords.h>
+#if DEVELOPMENT || DEBUG
+extern int proc_selfcsflags(void);
+#if CONFIG_EMBEDDED
+extern int panic_on_unsigned_execute;
+#endif /* CONFIG_EMBEDDED */
+#endif /* DEVELOPMENT || DEBUG */
+
+#if __arm64__
+extern const int fourk_binary_compatibility_unsafe;
+extern const int fourk_binary_compatibility_allow_wx;
+#endif /* __arm64__ */
+extern int proc_selfpid(void);
+extern char *proc_name_address(void *p);
+
+#if VM_MAP_DEBUG_APPLE_PROTECT
+int vm_map_debug_apple_protect = 0;
+#endif /* VM_MAP_DEBUG_APPLE_PROTECT */
+#if VM_MAP_DEBUG_FOURK
+int vm_map_debug_fourk = 0;
+#endif /* VM_MAP_DEBUG_FOURK */
+
+SECURITY_READ_ONLY_LATE(int) vm_map_executable_immutable = 1;
+int vm_map_executable_immutable_verbose = 0;
+
extern u_int32_t random(void); /* from <libkern/libkern.h> */
/* Internal prototypes
*/
int flags,
vm_map_t zap_map);
+static void vm_map_copy_insert(
+ vm_map_t map,
+ vm_map_entry_t after_where,
+ vm_map_copy_t copy);
+
static kern_return_t vm_map_copy_overwrite_unaligned(
vm_map_t dst_map,
vm_map_entry_t entry,
vm_map_t map,
vm_map_address_t *addr, /* IN/OUT */
vm_map_copy_t copy,
+ vm_map_size_t copy_size,
boolean_t overwrite,
boolean_t consume_on_success);
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);
+ vm_map_t new_map,
+ int vm_map_copyin_flags);
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 access_type,
+ vm_prot_t caller_prot,
+ vm_tag_t tag,
boolean_t user_wire,
- pmap_t map_pmap,
- vm_map_offset_t pmap_addr);
+ 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_prot_t *cur_protection,
vm_prot_t *max_protection,
vm_inherit_t inheritance,
- boolean_t pageable);
+ boolean_t pageable,
+ boolean_t same_map,
+ vm_map_kernel_flags_t vmk_flags);
static kern_return_t vm_map_remap_range_allocate(
vm_map_t map,
vm_map_size_t size,
vm_map_offset_t mask,
int flags,
+ vm_map_kernel_flags_t vmk_flags,
+ vm_tag_t tag,
vm_map_entry_t *map_entry);
static void vm_map_region_look_for_page(
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 */
+
+static void vm_map_corpse_footprint_destroy(
+ vm_map_t map);
+
+pid_t find_largest_process_vm_map_entries(void);
/*
* Macros to copy a vm_map_entry. We must be careful to correctly
* vm_map_copyout.
*/
+#if CONFIG_EMBEDDED
+
+/*
+ * The "used_for_jit" flag was copied from OLD to NEW in vm_map_entry_copy().
+ * But for security reasons on embedded platforms, we don't want the
+ * new mapping to be "used for jit", so we always reset the flag here.
+ * Same for "pmap_cs_associated".
+ */
+#define VM_MAP_ENTRY_COPY_CODE_SIGNING(NEW,OLD) \
+MACRO_BEGIN \
+ (NEW)->used_for_jit = FALSE; \
+ (NEW)->pmap_cs_associated = FALSE; \
+MACRO_END
+
+#else /* CONFIG_EMBEDDED */
+
+/*
+ * The "used_for_jit" flag was copied from OLD to NEW in vm_map_entry_copy().
+ * On macOS, the new mapping can be "used for jit".
+ */
+#define VM_MAP_ENTRY_COPY_CODE_SIGNING(NEW,OLD) \
+MACRO_BEGIN \
+ assert((NEW)->used_for_jit == (OLD)->used_for_jit); \
+ assert((NEW)->pmap_cs_associated == FALSE); \
+MACRO_END
+
+#endif /* CONFIG_EMBEDDED */
+
#define vm_map_entry_copy(NEW,OLD) \
MACRO_BEGIN \
boolean_t _vmec_reserved = (NEW)->from_reserved_zone; \
(NEW)->wired_count = 0; \
(NEW)->user_wired_count = 0; \
(NEW)->permanent = FALSE; \
- (NEW)->used_for_jit = FALSE; \
- (NEW)->from_reserved_zone = _vmec_reserved; \
+ VM_MAP_ENTRY_COPY_CODE_SIGNING((NEW),(OLD)); \
+ (NEW)->from_reserved_zone = _vmec_reserved; \
+ if ((NEW)->iokit_acct) { \
+ assertf(!(NEW)->use_pmap, "old %p new %p\n", (OLD), (NEW)); \
+ (NEW)->iokit_acct = FALSE; \
+ (NEW)->use_pmap = TRUE; \
+ } \
+ (NEW)->vme_resilient_codesign = FALSE; \
+ (NEW)->vme_resilient_media = FALSE; \
+ (NEW)->vme_atomic = FALSE; \
MACRO_END
#define vm_map_entry_copy_full(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
+ * 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
+ * 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
* 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)
+ * 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.
*
{
int current_abi;
+ if (map->pmap == kernel_pmap) return FALSE;
+
/*
* Determine if the app is running in 32 or 64 bit mode.
*/
current_abi = VM_ABI_32;
/*
- * Determine if we should allow the execution based on whether it's a
+ * Determine if we should allow the execution based on whether it's a
* stack or data area and the current architecture.
*/
* the asymmetric (delayed) strategy is used for shared temporary
* objects instead of the symmetric (shadow) strategy. All maps
* are now "top level" maps (either task map, kernel map or submap
- * of the kernel map).
+ * of the kernel map).
*
* Since portions of maps are specified by start/end addreses,
* which may not align with existing map entries, all
* vm_object_copy_strategically() in vm_object.c.
*/
-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 */
+static zone_t vm_map_zone; /* zone for vm_map structures */
+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 */
/*
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;
+#if CONFIG_EMBEDDED
+#define NO_COALESCE_LIMIT 0
+#else
#define NO_COALESCE_LIMIT ((1024 * 128) - 1)
+#endif
/* Skip acquiring locks if we're in the midst of a kernel core dump */
unsigned int not_in_kdp = 1;
kr = KERN_INVALID_ARGUMENT;
goto done;
}
- object = map_entry->object.vm_object;
+ object = VME_OBJECT(map_entry);
if (object == VM_OBJECT_NULL) {
/*
#if CONFIG_CODE_DECRYPTION
/*
* vm_map_apple_protected:
- * This remaps the requested part of the object with an object backed by
+ * 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
*/
kern_return_t
vm_map_apple_protected(
- vm_map_t map,
- vm_map_offset_t start,
- vm_map_offset_t end,
+ 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;
- memory_object_t protected_mem_obj;
+ 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;
+ vm_map_kernel_flags_t vmk_flags;
- vm_map_lock_read(map);
- map_locked = TRUE;
-
- /* lookup the protected VM object */
- if (!vm_map_lookup_entry(map,
- start,
- &map_entry) ||
- map_entry->vme_end < end ||
- map_entry->is_sub_map) {
- /* that memory is not properly mapped */
- kr = KERN_INVALID_ARGUMENT;
- goto done;
- }
- protected_object = map_entry->object.vm_object;
- if (protected_object == VM_OBJECT_NULL) {
- /* there should be a VM object here at this point */
- kr = KERN_INVALID_ARGUMENT;
- goto done;
- }
-
- /* make sure protected object stays alive while map is unlocked */
- vm_object_reference(protected_object);
+ vm_flags = 0;
+ vmk_flags = VM_MAP_KERNEL_FLAGS_NONE;
- vm_map_unlock_read(map);
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));
+
+#if __arm64__
/*
- * 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.
+ * "start" and "end" might be 4K-aligned but not 16K-aligned,
+ * so we might have to loop and establish up to 3 mappings:
+ *
+ * + the first 16K-page, which might overlap with the previous
+ * 4K-aligned mapping,
+ * + the center,
+ * + the last 16K-page, which might overlap with the next
+ * 4K-aligned mapping.
+ * Each of these mapping might be backed by a vnode pager (if
+ * properly page-aligned) or a "fourk_pager", itself backed by a
+ * vnode pager (if 4K-aligned but not page-aligned).
*/
- protected_mem_obj = apple_protect_pager_setup(protected_object, crypt_info);
+#else /* __arm64__ */
+ assert(start_aligned == start);
+ assert(end_aligned == end);
+#endif /* __arm64__ */
+
+ 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;
- /* release extra ref on protected object */
- vm_object_deallocate(protected_object);
+ /* 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;
+ }
- if (protected_mem_obj == NULL) {
- kr = KERN_FAILURE;
- 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;
+ /* can overwrite an immutable mapping */
+ vmk_flags.vmkf_overwrite_immutable = TRUE;
+#if __arm64__
+ if (tmp_entry.used_for_jit &&
+ (VM_MAP_PAGE_SHIFT(map) != FOURK_PAGE_SHIFT ||
+ PAGE_SHIFT != FOURK_PAGE_SHIFT) &&
+ fourk_binary_compatibility_unsafe &&
+ fourk_binary_compatibility_allow_wx) {
+ printf("** FOURK_COMPAT [%d]: "
+ "allowing write+execute at 0x%llx\n",
+ proc_selfpid(), tmp_entry.vme_start);
+ vmk_flags.vmkf_map_jit = TRUE;
+ }
+#endif /* __arm64__ */
+
+ /* 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,
+ vmk_flags,
+ VM_KERN_MEMORY_NONE,
+ (ipc_port_t)(uintptr_t) unprotected_mem_obj,
+ 0,
+ TRUE,
+ tmp_entry.protection,
+ tmp_entry.max_protection,
+ tmp_entry.inheritance);
+ assertf(kr == KERN_SUCCESS,
+ "kr = 0x%x\n", kr);
+ assertf(map_addr == tmp_entry.vme_start,
+ "map_addr=0x%llx vme_start=0x%llx tmp_entry=%p\n",
+ (uint64_t)map_addr,
+ (uint64_t) tmp_entry.vme_start,
+ &tmp_entry);
+
+#if VM_MAP_DEBUG_APPLE_PROTECT
+ 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;
- /* map this memory object in place of the current one */
- map_addr = start;
- kr = vm_map_enter_mem_object(map,
- &map_addr,
- end - start,
- (mach_vm_offset_t) 0,
- VM_FLAGS_FIXED | VM_FLAGS_OVERWRITE,
- (ipc_port_t) protected_mem_obj,
- (map_entry->offset +
- (start - map_entry->vme_start)),
- TRUE,
- map_entry->protection,
- map_entry->max_protection,
- map_entry->inheritance);
- assert(map_addr == start);
- /*
- * 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(protected_mem_obj);
+ /* 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_read(map);
+ vm_map_unlock(map);
}
return kr;
}
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;
+#if CONFIG_EMBEDDED
+int malloc_no_cow = 1;
+#define VM_PROTECT_WX_FAIL 0
+#else /* CONFIG_EMBEDDED */
+int malloc_no_cow = 0;
+#define VM_PROTECT_WX_FAIL 1
+#endif /* CONFIG_EMBEDDED */
+uint64_t vm_memory_malloc_no_cow_mask = 0ULL;
/*
* vm_map_init:
kentry_data_size * 64, kentry_data_size,
"Reserved VM map entries");
zone_change(vm_map_entry_reserved_zone, Z_NOENCRYPT, TRUE);
+ /* Don't quarantine because we always need elements available */
+ zone_change(vm_map_entry_reserved_zone, Z_KASAN_QUARANTINE, FALSE);
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 reserved_zone non-collectible to aid zone_gc().
*/
zone_change(vm_map_zone, Z_COLLECT, FALSE);
+ zone_change(vm_map_zone, Z_FOREIGN, TRUE);
+ zone_change(vm_map_zone, Z_GZALLOC_EXEMPT, TRUE);
zone_change(vm_map_entry_reserved_zone, Z_COLLECT, FALSE);
zone_change(vm_map_entry_reserved_zone, Z_EXPAND, FALSE);
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 only up to the maximum number of pages for each zone chunk.
+ */
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);
-
+
+ const vm_size_t stride = ZONE_CHUNK_MAXPAGES * PAGE_SIZE;
+ for (vm_offset_t off = 0; off < kentry_data_size; off += stride) {
+ zcram(vm_map_entry_reserved_zone,
+ (vm_offset_t)kentry_data + off,
+ MIN(kentry_data_size - off, stride));
+ }
+ for (vm_offset_t off = 0; off < map_holes_data_size; off += stride) {
+ zcram(vm_map_holes_zone,
+ (vm_offset_t)map_holes_data + off,
+ MIN(map_holes_data_size - off, stride));
+ }
+
+ 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);
-
-#if CONFIG_FREEZE
- default_freezer_init();
-#endif /* CONFIG_FREEZE */
+ lck_attr_setdefault(&vm_map_lck_attr);
+
+ lck_attr_setdefault(&vm_map_lck_rw_attr);
+ lck_attr_cleardebug(&vm_map_lck_rw_attr);
+
+#if VM_MAP_DEBUG_APPLE_PROTECT
+ PE_parse_boot_argn("vm_map_debug_apple_protect",
+ &vm_map_debug_apple_protect,
+ sizeof(vm_map_debug_apple_protect));
+#endif /* VM_MAP_DEBUG_APPLE_PROTECT */
+#if VM_MAP_DEBUG_APPLE_FOURK
+ PE_parse_boot_argn("vm_map_debug_fourk",
+ &vm_map_debug_fourk,
+ sizeof(vm_map_debug_fourk));
+#endif /* VM_MAP_DEBUG_FOURK */
+ PE_parse_boot_argn("vm_map_executable_immutable",
+ &vm_map_executable_immutable,
+ sizeof(vm_map_executable_immutable));
+ PE_parse_boot_argn("vm_map_executable_immutable_verbose",
+ &vm_map_executable_immutable_verbose,
+ sizeof(vm_map_executable_immutable_verbose));
+
+ PE_parse_boot_argn("malloc_no_cow",
+ &malloc_no_cow,
+ sizeof(malloc_no_cow));
+ if (malloc_no_cow) {
+ vm_memory_malloc_no_cow_mask = 0ULL;
+ vm_memory_malloc_no_cow_mask |= 1ULL << VM_MEMORY_MALLOC;
+ vm_memory_malloc_no_cow_mask |= 1ULL << VM_MEMORY_MALLOC_SMALL;
+ vm_memory_malloc_no_cow_mask |= 1ULL << VM_MEMORY_MALLOC_LARGE;
+// vm_memory_malloc_no_cow_mask |= 1ULL << VM_MEMORY_MALLOC_HUGE;
+// vm_memory_malloc_no_cow_mask |= 1ULL << VM_MEMORY_REALLOC;
+ vm_memory_malloc_no_cow_mask |= 1ULL << VM_MEMORY_MALLOC_TINY;
+ vm_memory_malloc_no_cow_mask |= 1ULL << VM_MEMORY_MALLOC_LARGE_REUSABLE;
+ vm_memory_malloc_no_cow_mask |= 1ULL << VM_MEMORY_MALLOC_LARGE_REUSED;
+ vm_memory_malloc_no_cow_mask |= 1ULL << VM_MEMORY_MALLOC_NANO;
+// vm_memory_malloc_no_cow_mask |= 1ULL << VM_MEMORY_TCMALLOC;
+ PE_parse_boot_argn("vm_memory_malloc_no_cow_mask",
+ &vm_memory_malloc_no_cow_mask,
+ sizeof(vm_memory_malloc_no_cow_mask));
+ }
}
void
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) {
+boolean_t vm_map_supports_hole_optimization = FALSE;
+
+void
+vm_kernel_reserved_entry_init(void) {
zone_prio_refill_configure(vm_map_entry_reserved_zone, (6*PAGE_SIZE)/sizeof(struct vm_map_entry));
+
+ /*
+ * Once we have our replenish thread set up, we can start using the vm_map_holes zone.
+ */
+ zone_prio_refill_configure(vm_map_holes_zone, (6*PAGE_SIZE)/sizeof(struct vm_map_links));
+ vm_map_supports_hole_optimization = TRUE;
+}
+
+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 = CAST_TO_VM_MAP_ENTRY(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.
*/
+
vm_map_t
vm_map_create(
- pmap_t pmap,
+ pmap_t pmap,
vm_map_offset_t min,
vm_map_offset_t max,
- boolean_t pageable)
+ boolean_t pageable)
+{
+ int options;
+
+ options = 0;
+ if (pageable) {
+ options |= VM_MAP_CREATE_PAGEABLE;
+ }
+ return vm_map_create_options(pmap, min, max, options);
+}
+
+vm_map_t
+vm_map_create_options(
+ pmap_t pmap,
+ vm_map_offset_t min,
+ vm_map_offset_t max,
+ int options)
{
- static int color_seed = 0;
- register vm_map_t result;
+ vm_map_t result;
+ struct vm_map_links *hole_entry = NULL;
+
+ if (options & ~(VM_MAP_CREATE_ALL_OPTIONS)) {
+ /* unknown option */
+ return VM_MAP_NULL;
+ }
result = (vm_map_t) zalloc(vm_map_zone);
if (result == VM_MAP_NULL)
vm_map_first_entry(result) = vm_map_to_entry(result);
vm_map_last_entry(result) = vm_map_to_entry(result);
result->hdr.nentries = 0;
- result->hdr.entries_pageable = pageable;
+ if (options & VM_MAP_CREATE_PAGEABLE) {
+ result->hdr.entries_pageable = TRUE;
+ } else {
+ result->hdr.entries_pageable = FALSE;
+ }
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 __x86_64__
+ result->vmmap_high_start = 0;
+#endif /* __x86_64__ */
+ result->map_refcnt = 1;
#if TASK_SWAPPER
result->res_count = 1;
result->sw_state = MAP_SW_IN;
result->switch_protect = FALSE;
result->disable_vmentry_reuse = FALSE;
result->map_disallow_data_exec = FALSE;
+ result->is_nested_map = FALSE;
+ result->map_disallow_new_exec = FALSE;
result->highest_entry_end = 0;
result->first_free = vm_map_to_entry(result);
result->hint = vm_map_to_entry(result);
- result->color_rr = (color_seed++) & vm_color_mask;
result->jit_entry_exists = FALSE;
-#if CONFIG_FREEZE
- result->default_freezer_handle = NULL;
+
+ /* "has_corpse_footprint" and "holelistenabled" are mutually exclusive */
+ if (options & VM_MAP_CREATE_CORPSE_FOOTPRINT) {
+ result->has_corpse_footprint = TRUE;
+ result->holelistenabled = FALSE;
+ result->vmmap_corpse_footprint = NULL;
+ } else {
+ result->has_corpse_footprint = FALSE;
+ if (vm_map_supports_hole_optimization) {
+ hole_entry = zalloc(vm_map_holes_zone);
+
+ hole_entry->start = min;
+#if defined(__arm__) || defined(__arm64__)
+ hole_entry->end = result->max_offset;
+#else
+ hole_entry->end = (max > (vm_map_offset_t)MACH_VM_MAX_ADDRESS) ? max : (vm_map_offset_t)MACH_VM_MAX_ADDRESS;
#endif
+ result->holes_list = result->hole_hint = hole_entry;
+ hole_entry->prev = hole_entry->next = CAST_TO_VM_MAP_ENTRY(hole_entry);
+ result->holelistenabled = TRUE;
+ } else {
+ result->holelistenabled = FALSE;
+ }
+ }
+
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);
}
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)));
+ backtrace(&entry->vme_creation_bt[0],
+ (sizeof(entry->vme_creation_bt)/sizeof(uintptr_t)));
#endif
return(entry);
}
static void
_vm_map_entry_dispose(
- register struct vm_map_header *map_header,
- register vm_map_entry_t entry)
+ struct vm_map_header *map_header,
+ vm_map_entry_t entry)
{
- register zone_t zone;
+ zone_t zone;
if (map_header->entries_pageable || !(entry->from_reserved_zone))
zone = vm_map_entry_zone;
{
if (!first_free_check)
return TRUE;
-
- return( first_free_is_valid_store( map ));
+
+ return( first_free_is_valid_store( map ));
}
#endif /* MACH_ASSERT */
* vm_map_swapin.
*
*/
-void vm_map_res_reference(register vm_map_t map)
+void vm_map_res_reference(vm_map_t map)
{
/* assert map is locked */
assert(map->res_count >= 0);
- assert(map->ref_count >= map->res_count);
+ assert(map->map_refcnt >= map->res_count);
if (map->res_count == 0) {
lck_mtx_unlock(&map->s_lock);
vm_map_lock(map);
* The map may not be in memory (i.e. zero residence count).
*
*/
-void vm_map_reference_swap(register vm_map_t map)
+void vm_map_reference_swap(vm_map_t map)
{
assert(map != VM_MAP_NULL);
lck_mtx_lock(&map->s_lock);
assert(map->res_count >= 0);
- assert(map->ref_count >= map->res_count);
- map->ref_count++;
+ assert(map->map_refcnt >= map->res_count);
+ map->map_refcnt++;
vm_map_res_reference(map);
lck_mtx_unlock(&map->s_lock);
}
* The map is locked, so this function is callable from vm_map_deallocate.
*
*/
-void vm_map_res_deallocate(register vm_map_t map)
+void vm_map_res_deallocate(vm_map_t map)
{
assert(map->res_count > 0);
if (--map->res_count == 0) {
vm_map_unlock(map);
lck_mtx_lock(&map->s_lock);
}
- assert(map->ref_count >= map->res_count);
+ assert(map->map_refcnt >= map->res_count);
}
#endif /* MACH_ASSERT && TASK_SWAPPER */
vm_map_destroy(
vm_map_t map,
int flags)
-{
+{
vm_map_lock(map);
+ /* final cleanup: no need to unnest shared region */
+ flags |= VM_MAP_REMOVE_NO_UNNESTING;
+ /* final cleanup: ok to remove immutable mappings */
+ flags |= VM_MAP_REMOVE_IMMUTABLE;
+ /* final cleanup: allow gaps in range */
+ flags |= VM_MAP_REMOVE_GAPS_OK;
+
/* 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...) */
+#if !defined(__arm__) && !defined(__arm64__)
(void) vm_map_delete(map, 0x0, 0xFFFFFFFFFFFFF000ULL,
flags, VM_MAP_NULL);
+#endif /* !__arm__ && !__arm64__ */
+
+ vm_map_disable_hole_optimization(map);
+ vm_map_corpse_footprint_destroy(map);
-#if CONFIG_FREEZE
- if (map->default_freezer_handle) {
- default_freezer_handle_deallocate(map->default_freezer_handle);
- map->default_freezer_handle = NULL;
- }
-#endif
vm_map_unlock(map);
assert(map->hdr.nentries == 0);
-
+
if(map->pmap)
pmap_destroy(map->pmap);
+ if (vm_map_lck_attr.lck_attr_val & LCK_ATTR_DEBUG) {
+ /*
+ * If lock debugging is enabled the mutexes get tagged as LCK_MTX_TAG_INDIRECT.
+ * And this is regardless of whether the lck_mtx_ext_t is embedded in the
+ * structure or kalloc'ed via lck_mtx_init.
+ * An example is s_lock_ext within struct _vm_map.
+ *
+ * A lck_mtx_destroy on such a mutex will attempt a kfree and panic. We
+ * can add another tag to detect embedded vs alloc'ed indirect external
+ * mutexes but that'll be additional checks in the lock path and require
+ * updating dependencies for the old vs new tag.
+ *
+ * Since the kfree() is for LCK_MTX_TAG_INDIRECT mutexes and that tag is applied
+ * just when lock debugging is ON, we choose to forego explicitly destroying
+ * the vm_map mutex and rw lock and, as a consequence, will overflow the reference
+ * count on vm_map_lck_grp, which has no serious side-effect.
+ */
+ } else {
+ lck_rw_destroy(&(map)->lock, &vm_map_lck_grp);
+ lck_mtx_destroy(&(map)->s_lock, &vm_map_lck_grp);
+ }
+
zfree(vm_map_zone, map);
}
+/*
+ * Returns pid of the task with the largest number of VM map entries.
+ * Used in the zone-map-exhaustion jetsam path.
+ */
+pid_t
+find_largest_process_vm_map_entries(void)
+{
+ pid_t victim_pid = -1;
+ int max_vm_map_entries = 0;
+ task_t task = TASK_NULL;
+ queue_head_t *task_list = &tasks;
+
+ lck_mtx_lock(&tasks_threads_lock);
+ queue_iterate(task_list, task, task_t, tasks) {
+ if (task == kernel_task || !task->active)
+ continue;
+
+ vm_map_t task_map = task->map;
+ if (task_map != VM_MAP_NULL) {
+ int task_vm_map_entries = task_map->hdr.nentries;
+ if (task_vm_map_entries > max_vm_map_entries) {
+ max_vm_map_entries = task_vm_map_entries;
+ victim_pid = pid_from_task(task);
+ }
+ }
+ }
+ lck_mtx_unlock(&tasks_threads_lock);
+
+ printf("zone_map_exhaustion: victim pid %d, vm region count: %d\n", victim_pid, max_vm_map_entries);
+ return victim_pid;
+}
+
#if TASK_SWAPPER
/*
* vm_map_swapin/vm_map_swapout
*
- * Swap a map in and out, either referencing or releasing its resources.
+ * Swap a map in and out, either referencing or releasing its resources.
* These functions are internal use only; however, they must be exported
* because they may be called from macros, which are exported.
*
- * In the case of swapout, there could be races on the residence count,
- * so if the residence count is up, we return, assuming that a
+ * In the case of swapout, there could be races on the residence count,
+ * so if the residence count is up, we return, assuming that a
* vm_map_deallocate() call in the near future will bring us back.
*
* Locking:
* 2) A vm_map_reference() call at this time is illegal, and will
* cause a panic. vm_map_reference() is only allowed on resident
* maps, since it refuses to block.
- * 3) A vm_map_swapin() call during a swapin will block, and
+ * 3) A vm_map_swapin() call during a swapin will block, and
* proceeed when the first swapin is done, turning into a nop.
* This is the reason the res_count is not incremented until
* after the swapin is complete.
* the map lock is taken, during which a swapin may get the lock
* before a swapout about to happen. If this happens, the swapin
* will detect the state and increment the reference count, causing
- * the swapout to be a nop, thereby delaying it until a later
- * vm_map_deallocate. If the swapout gets the lock first, then
- * the swapin will simply block until the swapout is done, and
+ * the swapout to be a nop, thereby delaying it until a later
+ * vm_map_deallocate. If the swapout gets the lock first, then
+ * the swapin will simply block until the swapout is done, and
* then proceed.
*
* Because vm_map_swapin() is potentially an expensive operation, it
void vm_map_swapin (vm_map_t map)
{
- register vm_map_entry_t entry;
+ vm_map_entry_t entry;
if (!vm_map_swap_enable) /* debug */
return;
* First deal with various races.
*/
if (map->sw_state == MAP_SW_IN)
- /*
+ /*
* we raced with swapout and won. Returning will incr.
* the res_count, turning the swapout into a nop.
*/
assert(map->sw_state == MAP_SW_OUT);
/*
- * We now operate upon each map entry. If the entry is a sub-
+ * We now operate upon each map entry. If the entry is a sub-
* or share-map, we call vm_map_res_reference upon it.
* If the entry is an object, we call vm_object_res_reference
* (this may iterate through the shadow chain).
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;
+ vm_map_t lmap = VME_SUBMAP(entry);
lck_mtx_lock(&lmap->s_lock);
vm_map_res_reference(lmap);
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
void vm_map_swapout(vm_map_t map)
{
- register vm_map_entry_t entry;
-
+ vm_map_entry_t entry;
+
/*
* Map is locked
* First deal with various races.
return;
/*
- * We now operate upon each map entry. If the entry is a sub-
+ * We now operate upon each map entry. If the entry is a sub-
* or share-map, we call vm_map_res_deallocate upon it.
* If the entry is an object, we call vm_object_res_deallocate
* (this may iterate through the shadow chain).
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;
+ vm_map_t lmap = VME_SUBMAP(entry);
lck_mtx_lock(&lmap->s_lock);
vm_map_res_deallocate(lmap);
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,
- * since it could actively push
- * out pages (if we implement it
+ * This call may take a long time,
+ * since it could actively push
+ * out pages (if we implement it
* that way).
*/
vm_object_res_deallocate(object);
/*
* vm_map_lookup_entry: [ internal use only ]
*
- * Calls into the vm map store layer to find the map
- * entry containing (or immediately preceding) the
+ * 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
*/
boolean_t
vm_map_lookup_entry(
- register vm_map_t map,
- register vm_map_offset_t address,
+ vm_map_t map,
+ vm_map_offset_t address,
vm_map_entry_t *entry) /* OUT */
{
return ( vm_map_store_lookup_entry( map, address, entry ));
*/
kern_return_t
vm_map_find_space(
- register vm_map_t map,
+ vm_map_t map,
vm_map_offset_t *address, /* OUT */
vm_map_size_t size,
vm_map_offset_t mask,
- int flags,
+ int flags __unused,
+ vm_map_kernel_flags_t vmk_flags,
+ vm_tag_t tag,
vm_map_entry_t *o_entry) /* OUT */
{
- register vm_map_entry_t entry, new_entry;
- register vm_map_offset_t start;
- register vm_map_offset_t end;
+ vm_map_entry_t entry, new_entry;
+ vm_map_offset_t start;
+ 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) {
+ if (vmk_flags.vmkf_guard_after) {
/* account for the back guard page in the size */
size += VM_MAP_PAGE_SIZE(map);
}
if( map->disable_vmentry_reuse == TRUE) {
VM_MAP_HIGHEST_ENTRY(map, entry, 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;
+ if (map->holelistenabled) {
+ hole_entry = CAST_TO_VM_MAP_ENTRY(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;
+ }
}
/*
*/
while (TRUE) {
- register vm_map_entry_t next;
+ vm_map_entry_t next;
/*
* Find the end of the proposed new region.
* wrap around the address.
*/
- if (flags & VM_FLAGS_GUARD_BEFORE) {
+ if (vmk_flags.vmkf_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);
}
start = end;
+ assert(VM_MAP_PAGE_ALIGNED(start, VM_MAP_PAGE_MASK(map)));
end += size;
+ assert(VM_MAP_PAGE_ALIGNED(end, VM_MAP_PAGE_MASK(map)));
if ((end > map->max_offset) || (end < start)) {
vm_map_entry_dispose(map, new_entry);
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 == CAST_TO_VM_MAP_ENTRY(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;
+ }
}
- /*
- * At this point,
+ 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);
+ }
+ }
+
+ /*
+ * At this point,
* "start" and "end" should define the endpoints of the
* available new range, and
* "entry" should refer to the region before the new
* the map should be locked.
*/
- if (flags & VM_FLAGS_GUARD_BEFORE) {
+ if (vmk_flags.vmkf_guard_before) {
/* go back for the front guard page */
start -= VM_MAP_PAGE_SIZE(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->map_aligned = FALSE;
}
- new_entry->used_for_jit = 0;
-
- new_entry->alias = 0;
+ new_entry->used_for_jit = FALSE;
+ new_entry->pmap_cs_associated = FALSE;
new_entry->zero_wired_pages = FALSE;
+ new_entry->iokit_acct = FALSE;
+ new_entry->vme_resilient_codesign = FALSE;
+ new_entry->vme_resilient_media = FALSE;
+ if (vmk_flags.vmkf_atomic_entry)
+ new_entry->vme_atomic = TRUE;
+ else
+ new_entry->vme_atomic = FALSE;
- VM_GET_FLAGS_ALIAS(flags, new_entry->alias);
+ VME_ALIAS_SET(new_entry, tag);
/*
* Insert the new entry into the list
*/
- vm_map_store_entry_link(map, entry, new_entry);
+ vm_map_store_entry_link(map, entry, new_entry, VM_MAP_KERNEL_FLAGS_NONE);
map->size += size;
* As soon as a page not found in the object the scan ends.
*
* Returns:
- * Nothing.
+ * Nothing.
*
* In/out conditions:
* The source map should not be locked on entry.
*/
-static void
+__unused static void
vm_map_pmap_enter(
vm_map_t map,
- register vm_map_offset_t addr,
- register vm_map_offset_t end_addr,
- register vm_object_t object,
+ vm_map_offset_t addr,
+ vm_map_offset_t end_addr,
+ vm_object_t object,
vm_object_offset_t offset,
vm_prot_t protection)
{
int type_of_fault;
kern_return_t kr;
+ struct vm_object_fault_info fault_info = {};
if(map->pmap == 0)
return;
while (addr < end_addr) {
- register vm_page_t m;
+ vm_page_t m;
- vm_object_lock(object);
- m = vm_page_lookup(object, offset);
/*
- * ENCRYPTED SWAP:
- * The user should never see encrypted data, so do not
- * enter an encrypted page in the page table.
+ * 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.
*/
- if (m == VM_PAGE_NULL || m->busy || m->encrypted ||
- m->fictitious ||
- (m->unusual && ( m->error || m->restart || m->absent))) {
+
+ vm_object_lock(object);
+
+ m = vm_page_lookup(object, offset);
+
+ if (m == VM_PAGE_NULL || m->vmp_busy || m->vmp_fictitious ||
+ (m->vmp_unusual && ( m->vmp_error || m->vmp_restart || m->vmp_absent))) {
vm_object_unlock(object);
return;
}
map, (unsigned long long)addr, object, (unsigned long long)offset);
}
type_of_fault = DBG_CACHE_HIT_FAULT;
- kr = vm_fault_enter(m, map->pmap, addr, protection, protection,
- VM_PAGE_WIRED(m), FALSE, FALSE, FALSE, NULL,
+ kr = vm_fault_enter(m, map->pmap,
+ addr, protection, protection,
+ VM_PAGE_WIRED(m),
+ FALSE, /* change_wiring */
+ VM_KERN_MEMORY_NONE, /* tag - not wiring */
+ &fault_info,
+ NULL, /* need_retry */
&type_of_fault);
vm_object_unlock(object);
return kr;
}
+static boolean_t
+vm_memory_malloc_no_cow(
+ int alias)
+{
+ uint64_t alias_mask;
+
+ alias_mask = 1ULL << alias;
+ if (alias_mask & vm_memory_malloc_no_cow_mask) {
+ return TRUE;
+ }
+ return FALSE;
+}
+
/*
* Routine: vm_map_enter
*
vm_map_size_t size,
vm_map_offset_t mask,
int flags,
+ vm_map_kernel_flags_t vmk_flags,
+ vm_tag_t alias,
vm_object_t object,
vm_object_offset_t offset,
boolean_t needs_copy,
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 desired_empty_end;
vm_map_offset_t step;
kern_return_t result = KERN_SUCCESS;
vm_map_t zap_old_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 = vmk_flags.vmkf_keep_map_locked;
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 is_submap = vmk_flags.vmkf_submap;
+ boolean_t permanent = vmk_flags.vmkf_permanent;
+ boolean_t entry_for_jit = vmk_flags.vmkf_map_jit;
+ boolean_t iokit_acct = vmk_flags.vmkf_iokit_acct;
+ boolean_t resilient_codesign = ((flags & VM_FLAGS_RESILIENT_CODESIGN) != 0);
+ boolean_t resilient_media = ((flags & VM_FLAGS_RESILIENT_MEDIA) != 0);
+ boolean_t random_address = ((flags & VM_FLAGS_RANDOM_ADDR) != 0);
unsigned int superpage_size = ((flags & VM_FLAGS_SUPERPAGE_MASK) >> VM_FLAGS_SUPERPAGE_SHIFT);
- char alias;
+ vm_tag_t 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;
+ vm_map_size_t chunk_size = 0;
+
+ assertf(vmk_flags.__vmkf_unused == 0, "vmk_flags unused=0x%x\n", vmk_flags.__vmkf_unused);
+
+ if (flags & VM_FLAGS_4GB_CHUNK) {
+#if defined(__LP64__)
+ chunk_size = (4ULL * 1024 * 1024 * 1024); /* max. 4GB chunks for the new allocation */
+#else /* __LP64__ */
+ chunk_size = ANON_CHUNK_SIZE;
+#endif /* __LP64__ */
+ } else {
+ chunk_size = ANON_CHUNK_SIZE;
+ }
if (superpage_size) {
switch (superpage_size) {
}
+ if ((cur_protection & VM_PROT_WRITE) &&
+ (cur_protection & VM_PROT_EXECUTE) &&
+#if !CONFIG_EMBEDDED
+ map != kernel_map &&
+ (cs_process_global_enforcement() ||
+ (vmk_flags.vmkf_cs_enforcement_override
+ ? vmk_flags.vmkf_cs_enforcement
+ : cs_process_enforcement(NULL))) &&
+#endif /* !CONFIG_EMBEDDED */
+ !entry_for_jit) {
+ DTRACE_VM3(cs_wx,
+ uint64_t, 0,
+ uint64_t, 0,
+ vm_prot_t, cur_protection);
+ printf("CODE SIGNING: %d[%s] %s: curprot cannot be write+execute. "
+#if VM_PROTECT_WX_FAIL
+ "failing\n",
+#else /* VM_PROTECT_WX_FAIL */
+ "turning off execute\n",
+#endif /* VM_PROTECT_WX_FAIL */
+ proc_selfpid(),
+ (current_task()->bsd_info
+ ? proc_name_address(current_task()->bsd_info)
+ : "?"),
+ __FUNCTION__);
+ cur_protection &= ~VM_PROT_EXECUTE;
+#if VM_PROTECT_WX_FAIL
+ return KERN_PROTECTION_FAILURE;
+#endif /* VM_PROTECT_WX_FAIL */
+ }
+
+ /*
+ * If the task has requested executable lockdown,
+ * deny any new executable mapping.
+ */
+ if (map->map_disallow_new_exec == TRUE) {
+ if (cur_protection & VM_PROT_EXECUTE) {
+ return KERN_PROTECTION_FAILURE;
+ }
+ }
+
+ 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) {
return KERN_INVALID_ARGUMENT;
}
}
- if (flags & VM_FLAGS_ALREADY) {
+ if (vmk_flags.vmkf_already) {
/*
* VM_FLAGS_ALREADY says that it's OK if the same mapping
* is already present. For it to be meaningul, the requested
effective_min_offset = map->min_offset;
- if (flags & VM_FLAGS_BEYOND_MAX) {
+ if (vmk_flags.vmkf_beyond_max) {
/*
* Allow an insertion beyond the map's max offset.
*/
+#if !defined(__arm__) && !defined(__arm64__)
if (vm_map_is_64bit(map))
effective_max_offset = 0xFFFFFFFFFFFFF000ULL;
else
+#endif /* __arm__ */
effective_max_offset = 0x00000000FFFFF000ULL;
} else {
effective_max_offset = map->max_offset;
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; }
*/
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.
*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: ;
if (anywhere) {
vm_map_lock(map);
map_locked = TRUE;
-
+
if (entry_for_jit) {
+#if CONFIG_EMBEDDED
if (map->jit_entry_exists) {
result = KERN_INVALID_ARGUMENT;
goto BailOut;
}
+ random_address = TRUE;
+#endif /* CONFIG_EMBEDDED */
+ }
+
+ if (random_address) {
/*
* Get a random start address.
*/
}
start = *address;
}
+#if __x86_64__
+ else if ((start == 0 || start == vm_map_min(map)) &&
+ !map->disable_vmentry_reuse &&
+ map->vmmap_high_start != 0) {
+ start = map->vmmap_high_start;
+ }
+#endif /* __x86_64__ */
/*
if( map->disable_vmentry_reuse == TRUE) {
VM_MAP_HIGHEST_ENTRY(map, entry, start);
} else {
- assert(first_free_is_valid(map));
- entry = map->first_free;
+ if (map->holelistenabled) {
+ hole_entry = CAST_TO_VM_MAP_ENTRY(map->holes_list);
- if (entry == vm_map_to_entry(map)) {
- entry = NULL;
+ if (hole_entry == NULL) {
+ /*
+ * No more space in the map?
+ */
+ result = KERN_NO_SPACE;
+ goto BailOut;
+ } else {
+
+ boolean_t found_hole = FALSE;
+
+ do {
+ if (hole_entry->vme_start >= start) {
+ start = hole_entry->vme_start;
+ found_hole = TRUE;
+ break;
+ }
+
+ if (hole_entry->vme_end > start) {
+ found_hole = TRUE;
+ break;
+ }
+ hole_entry = hole_entry->vme_next;
+
+ } while (hole_entry != CAST_TO_VM_MAP_ENTRY(map->holes_list));
+
+ if (found_hole == FALSE) {
+ result = KERN_NO_SPACE;
+ goto BailOut;
+ }
+
+ entry = hole_entry;
+
+ if (start == 0)
+ start += PAGE_SIZE_64;
+ }
} else {
- if (entry->vme_next == vm_map_to_entry(map)){
- /*
- * Hole at the end of the map.
- */
+ assert(first_free_is_valid(map));
+
+ entry = map->first_free;
+
+ if (entry == vm_map_to_entry(map)) {
entry = NULL;
- } else {
- if (start < (entry->vme_next)->vme_start ) {
- start = entry->vme_end;
+ } 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));
- } 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;
}
- entry = tmp_entry;
}
}
*/
while (TRUE) {
- register vm_map_entry_t next;
+ vm_map_entry_t next;
/*
* Find the end of the proposed new region.
VM_MAP_PAGE_MASK(map)));
end += size;
- if ((end > effective_max_offset) || (end < start)) {
+ /* We want an entire page of empty space, but don't increase the allocation size. */
+ desired_empty_end = vm_map_round_page(end, VM_MAP_PAGE_MASK(map));
+
+ if ((desired_empty_end > effective_max_offset) || (desired_empty_end < start)) {
if (map->wait_for_space) {
+ assert(!keep_map_locked);
if (size <= (effective_max_offset -
effective_min_offset)) {
assert_wait((event_t)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 >= desired_empty_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 >= desired_empty_end)
+ break;
+ }
/*
* Didn't fit -- move to the next entry.
*/
entry = next;
- start = entry->vme_end;
+
+ if (map->holelistenabled) {
+ if (entry == CAST_TO_VM_MAP_ENTRY(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)));
}
if (overwrite && zap_old_map != VM_MAP_NULL) {
+ int remove_flags;
/*
* Fixed mapping and "overwrite" flag: attempt to
* remove all existing mappings in the specified
* address range, saving them in our "zap_old_map".
*/
+ remove_flags = VM_MAP_REMOVE_SAVE_ENTRIES;
+ remove_flags |= VM_MAP_REMOVE_NO_MAP_ALIGN;
+ if (vmk_flags.vmkf_overwrite_immutable) {
+ /* we can overwrite immutable mappings */
+ remove_flags |= VM_MAP_REMOVE_IMMUTABLE;
+ }
(void) vm_map_delete(map, start, end,
- VM_MAP_REMOVE_SAVE_ENTRIES,
+ remove_flags,
zap_old_map);
}
*/
if (vm_map_lookup_entry(map, start, &entry)) {
- if (! (flags & VM_FLAGS_ALREADY)) {
+ if (! (vmk_flags.vmkf_already)) {
RETURN(KERN_NO_SPACE);
}
/*
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) {
if (entry == vm_map_to_entry(map) ||
entry->vme_start != tmp_start ||
entry->is_sub_map != is_submap ||
- entry->offset != tmp_offset ||
+ VME_OFFSET(entry) != tmp_offset ||
entry->needs_copy != needs_copy ||
entry->protection != cur_protection ||
entry->max_protection != max_protection ||
entry->inheritance != inheritance ||
- entry->alias != alias) {
+ 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 (entry->object.sub_map !=
+ if (VME_SUBMAP(entry) !=
(vm_map_t) object) {
/* not the same submap */
RETURN(KERN_NO_SPACE);
}
} else {
- if (entry->object.vm_object != object) {
+ if (VME_OBJECT(entry) != object) {
/* not the same VM object... */
vm_object_t obj2;
- obj2 = entry->object.vm_object;
+ obj2 = VME_OBJECT(entry);
if ((obj2 == VM_OBJECT_NULL ||
obj2->internal) &&
(object == VM_OBJECT_NULL ||
* semantics.
*/
- if (purgable || entry_for_jit) {
+ if (purgable ||
+ entry_for_jit ||
+ vm_memory_malloc_no_cow(user_alias)) {
if (object == VM_OBJECT_NULL) {
+
object = vm_object_allocate(size);
object->copy_strategy = MEMORY_OBJECT_COPY_NONE;
+ object->true_share = FALSE;
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_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;
}
(entry->vme_end == start) &&
(!entry->is_shared) &&
(!entry->is_sub_map) &&
- ((alias == VM_MEMORY_REALLOC) || (entry->alias == alias)) &&
- (entry->inheritance == inheritance) &&
+ (!entry->in_transition) &&
+ (!entry->needs_wakeup) &&
+ (entry->behavior == VM_BEHAVIOR_DEFAULT) &&
(entry->protection == cur_protection) &&
(entry->max_protection == max_protection) &&
- (entry->behavior == VM_BEHAVIOR_DEFAULT) &&
- (entry->in_transition == 0) &&
+ (entry->inheritance == inheritance) &&
+ ((user_alias == VM_MEMORY_REALLOC) ||
+ (VME_ALIAS(entry) == alias)) &&
(entry->no_cache == no_cache) &&
+ (entry->permanent == permanent) &&
+ /* no coalescing for immutable executable mappings */
+ !((entry->protection & VM_PROT_EXECUTE) &&
+ entry->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->pmap_cs_associated) &&
+ (entry->iokit_acct == iokit_acct) &&
+ (!entry->vme_resilient_codesign) &&
+ (!entry->vme_resilient_media) &&
+ (!entry->vme_atomic) &&
+
((entry->vme_end - entry->vme_start) + size <=
- (alias == VM_MEMORY_REALLOC ?
+ (user_alias == VM_MEMORY_REALLOC ?
ANON_CHUNK_SIZE :
NO_COALESCE_LIMIT)) &&
+
(entry->wired_count == 0)) { /* implies user_wired_count == 0 */
- if (vm_object_coalesce(entry->object.vm_object,
+ if (vm_object_coalesce(VME_OBJECT(entry),
VM_OBJECT_NULL,
- entry->offset,
+ 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))) {
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;
- vm_map_store_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);
}
}
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
+ * be arbitrarily large. Splitting it into separate 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
+ * allocation into 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 &&
+ size > chunk_size &&
max_protection != VM_PROT_NONE &&
- superpage_size == 0)
- tmp_end = tmp_start + (vm_map_size_t)ANON_CHUNK_SIZE;
+ superpage_size == 0)
+ tmp_end = tmp_start + 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);
- new_entry->alias = alias;
- if (entry_for_jit){
- if (!(map->jit_entry_exists)){
- new_entry->used_for_jit = TRUE;
- map->jit_entry_exists = TRUE;
- }
+ 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,
+ entry_for_jit,
+ alias);
+
+ assert((object != kernel_object) || (VM_KERN_MEMORY_NONE != alias));
+
+ if (resilient_codesign &&
+ ! ((cur_protection | max_protection) &
+ (VM_PROT_WRITE | VM_PROT_EXECUTE))) {
+ new_entry->vme_resilient_codesign = TRUE;
+ }
+
+ if (resilient_media &&
+ ! ((cur_protection | max_protection) &
+ (VM_PROT_WRITE | VM_PROT_EXECUTE))) {
+ new_entry->vme_resilient_media = TRUE;
+ }
+
+ assert(!new_entry->iokit_acct);
+ if (!is_submap &&
+ object != VM_OBJECT_NULL &&
+ (object->purgable != VM_PURGABLE_DENY ||
+ object->vo_ledger_tag)) {
+ assert(new_entry->use_pmap);
+ assert(!new_entry->iokit_acct);
+ /*
+ * Turn off pmap accounting since
+ * purgeable (or tagged) 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);
}
if (is_submap) {
boolean_t submap_is_64bit;
boolean_t use_pmap;
- new_entry->is_sub_map = TRUE;
+ 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 = (alias == VM_MEMORY_SHARED_PMAP);
- #ifndef NO_NESTED_PMAP
+ 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... */
if (submap->pmap == NULL) {
/* let's proceed without nesting... */
}
+#if defined(__arm__) || defined(__arm64__)
+ else {
+ pmap_set_nested(submap->pmap);
+ }
+#endif
}
if (use_pmap && submap->pmap != NULL) {
kr = pmap_nest(map->pmap,
pmap_empty = FALSE;
}
}
- #endif /* NO_NESTED_PMAP */
+#endif /* NO_NESTED_PMAP */
}
entry = new_entry;
if (superpage_size) {
vm_page_t pages, m;
vm_object_t sp_object;
+ vm_object_offset_t sp_offset;
- entry->offset = 0;
+ VME_OFFSET_SET(entry, 0);
/* allocate one superpage */
kr = cpm_allocate(SUPERPAGE_SIZE, &pages, 0, SUPERPAGE_NBASEPAGES-1, TRUE, 0);
if (kr != KERN_SUCCESS) {
- new_mapping_established = TRUE; /* will cause deallocation of whole range */
+ /* deallocate whole range... */
+ new_mapping_established = TRUE;
+ /* ... but only up to "tmp_end" */
+ size -= end - tmp_end;
RETURN(kr);
}
/* 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;
- entry->object.vm_object = sp_object;
+ sp_object->vo_shadow_offset = (vm_object_offset_t)VM_PAGE_GET_PHYS_PAGE(pages)*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) {
+ for (sp_offset = 0;
+ sp_offset < SUPERPAGE_SIZE;
+ sp_offset += PAGE_SIZE) {
m = pages;
- pmap_zero_page(m->phys_page);
+ pmap_zero_page(VM_PAGE_GET_PHYS_PAGE(m));
pages = NEXT_PAGE(m);
*(NEXT_PAGE_PTR(m)) = VM_PAGE_NULL;
- vm_page_insert(m, sp_object, offset);
+ vm_page_insert_wired(m, sp_object, sp_offset, VM_KERN_MEMORY_OSFMK);
}
vm_object_unlock(sp_object);
}
- } while (tmp_end != tmp2_end &&
+ } 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));
+ (tmp_end = (tmp2_end - tmp_end > chunk_size) ?
+ tmp_end + chunk_size : tmp2_end));
}
- vm_map_unlock(map);
- map_locked = FALSE;
-
new_mapping_established = TRUE;
- /* Wire down the new entry if the user
- * requested all new map entries be wired.
- */
- if ((map->wiring_required)||(superpage_size)) {
- pmap_empty = FALSE; /* pmap won't be empty */
- kr = vm_map_wire(map, start, end,
- new_entry->protection, TRUE);
- RETURN(kr);
- }
-
- if ((object != VM_OBJECT_NULL) &&
- (vm_map_pmap_enter_enable) &&
- (!anywhere) &&
- (!needs_copy) &&
- (size < (128*1024))) {
- pmap_empty = FALSE; /* pmap won't be empty */
-
- if (override_nx(map, alias) && cur_protection)
- cur_protection |= VM_PROT_EXECUTE;
+BailOut:
+ assert(map_locked == TRUE);
- vm_map_pmap_enter(map, start, end,
- object, offset, cur_protection);
- }
-
-BailOut: ;
if (result == KERN_SUCCESS) {
vm_prot_t pager_prot;
memory_object_t pager;
+#if DEBUG
if (pmap_empty &&
- !(flags & VM_FLAGS_NO_PMAP_CHECK)) {
+ !(vmk_flags.vmkf_no_pmap_check)) {
assert(vm_map_pmap_is_empty(map,
*address,
*address+size));
}
+#endif /* DEBUG */
/*
* For "named" VM objects, let the pager know that the
}
vm_object_unlock(object);
}
- } else {
+ }
+
+ assert(map_locked == TRUE);
+
+ if (!keep_map_locked) {
+ vm_map_unlock(map);
+ map_locked = FALSE;
+ }
+
+ /*
+ * We can't hold the map lock if we enter this block.
+ */
+
+ if (result == KERN_SUCCESS) {
+
+ /* 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_kernel(map, start, end,
+ new_entry->protection, VM_KERN_MEMORY_MLOCK,
+ TRUE);
+ result = kr;
+ }
+
+ }
+
+ if (result != KERN_SUCCESS) {
if (new_mapping_established) {
/*
* We have to get rid of the new mappings since we
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_SAVE_ENTRIES |
+ VM_MAP_REMOVE_NO_MAP_ALIGN),
zap_new_map);
}
if (zap_old_map != VM_MAP_NULL &&
vm_map_store_entry_unlink(zap_old_map,
entry2);
zap_old_map->size -= entry_size;
- vm_map_store_entry_link(map, entry1, entry2);
+ vm_map_store_entry_link(map, entry1, entry2,
+ VM_MAP_KERNEL_FLAGS_NONE);
map->size += entry_size;
entry1 = entry2;
}
}
}
- if (map_locked) {
+ /*
+ * The caller is responsible for releasing the lock if it requested to
+ * keep the map locked.
+ */
+ if (map_locked && !keep_map_locked) {
vm_map_unlock(map);
}
#undef RETURN
}
+#if __arm64__
+extern const struct memory_object_pager_ops fourk_pager_ops;
kern_return_t
-vm_map_enter_mem_object(
- vm_map_t target_map,
- vm_map_offset_t *address,
- vm_map_size_t initial_size,
+vm_map_enter_fourk(
+ vm_map_t map,
+ vm_map_offset_t *address, /* IN/OUT */
+ vm_map_size_t size,
vm_map_offset_t mask,
int flags,
- ipc_port_t port,
+ vm_map_kernel_flags_t vmk_flags,
+ vm_tag_t alias,
+ vm_object_t object,
vm_object_offset_t offset,
- boolean_t copy,
+ boolean_t needs_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;
- boolean_t mask_cur_protection, mask_max_protection;
- vm_map_offset_t offset_in_mapping;
+ vm_map_entry_t entry, new_entry;
+ vm_map_offset_t start, fourk_start;
+ vm_map_offset_t end, fourk_end;
+ vm_map_size_t fourk_size;
+ 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 = vmk_flags.vmkf_keep_map_locked;
+ 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 = vmk_flags.vmkf_submap;
+ boolean_t permanent = vmk_flags.vmkf_permanent;
+ boolean_t entry_for_jit = vmk_flags.vmkf_map_jit;
+// boolean_t iokit_acct = vmk_flags.vmkf_iokit_acct;
+ unsigned int superpage_size = ((flags & VM_FLAGS_SUPERPAGE_MASK) >> VM_FLAGS_SUPERPAGE_SHIFT);
+ vm_map_offset_t effective_min_offset, effective_max_offset;
+ kern_return_t kr;
+ boolean_t clear_map_aligned = FALSE;
+ memory_object_t fourk_mem_obj;
+ vm_object_t fourk_object;
+ vm_map_offset_t fourk_pager_offset;
+ int fourk_pager_index_start, fourk_pager_index_num;
+ int cur_idx;
+ boolean_t fourk_copy;
+ vm_object_t copy_object;
+ vm_object_offset_t copy_offset;
+
+ fourk_mem_obj = MEMORY_OBJECT_NULL;
+ fourk_object = VM_OBJECT_NULL;
- 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;
+ if (superpage_size) {
+ return KERN_NOT_SUPPORTED;
+ }
+
+ if ((cur_protection & VM_PROT_WRITE) &&
+ (cur_protection & VM_PROT_EXECUTE) &&
+#if !CONFIG_EMBEDDED
+ map != kernel_map &&
+ cs_process_enforcement(NULL) &&
+#endif /* !CONFIG_EMBEDDED */
+ !entry_for_jit) {
+ DTRACE_VM3(cs_wx,
+ uint64_t, 0,
+ uint64_t, 0,
+ vm_prot_t, cur_protection);
+ printf("CODE SIGNING: %d[%s] %s: curprot cannot be write+execute. "
+ "turning off execute\n",
+ proc_selfpid(),
+ (current_task()->bsd_info
+ ? proc_name_address(current_task()->bsd_info)
+ : "?"),
+ __FUNCTION__);
+ cur_protection &= ~VM_PROT_EXECUTE;
+ }
/*
- * Check arguments for validity
+ * If the task has requested executable lockdown,
+ * deny any new executable mapping.
*/
- 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;
-
- 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);
+ if (map->map_disallow_new_exec == TRUE) {
+ if (cur_protection & VM_PROT_EXECUTE) {
+ return KERN_PROTECTION_FAILURE;
+ }
+ }
- /*
- * 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 (is_submap) {
+ return KERN_NOT_SUPPORTED;
+ }
+ if (vmk_flags.vmkf_already) {
+ return KERN_NOT_SUPPORTED;
+ }
+ if (purgable || entry_for_jit) {
+ return KERN_NOT_SUPPORTED;
+ }
- named_entry = (vm_named_entry_t) port->ip_kobject;
+ effective_min_offset = map->min_offset;
- if ((flags & VM_FLAGS_RETURN_DATA_ADDR) != 0) {
- offset += named_entry->data_offset;
- }
-
- /* a few checks to make sure user is obeying rules */
- if (size == 0) {
+ if (vmk_flags.vmkf_beyond_max) {
+ return KERN_NOT_SUPPORTED;
+ } else {
+ effective_max_offset = map->max_offset;
+ }
+
+ if (size == 0 ||
+ (offset & FOURK_PAGE_MASK) != 0) {
+ *address = 0;
+ return KERN_INVALID_ARGUMENT;
+ }
+
+#define RETURN(value) { result = value; goto BailOut; }
+
+ assert(VM_MAP_PAGE_ALIGNED(*address, FOURK_PAGE_MASK));
+ assert(VM_MAP_PAGE_ALIGNED(size, FOURK_PAGE_MASK));
+
+ if (!anywhere && overwrite) {
+ return KERN_NOT_SUPPORTED;
+ }
+ 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);
+ }
+
+ fourk_start = *address;
+ fourk_size = size;
+ fourk_end = fourk_start + fourk_size;
+
+ start = vm_map_trunc_page(*address, VM_MAP_PAGE_MASK(map));
+ end = vm_map_round_page(fourk_end, VM_MAP_PAGE_MASK(map));
+ size = end - start;
+
+ if (anywhere) {
+ return KERN_NOT_SUPPORTED;
+ } else {
+ /*
+ * Verify that:
+ * the address doesn't itself violate
+ * the mask requirement.
+ */
+
+ vm_map_lock(map);
+ map_locked = TRUE;
+ if ((start & mask) != 0) {
+ RETURN(KERN_NO_SPACE);
+ }
+
+ /*
+ * ... the address is within bounds
+ */
+
+ end = start + size;
+
+ 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, &entry)) {
+ vm_object_t cur_object, shadow_object;
+
+ /*
+ * We might already some 4K mappings
+ * in a 16K page here.
+ */
+
+ if (entry->vme_end - entry->vme_start
+ != SIXTEENK_PAGE_SIZE) {
+ RETURN(KERN_NO_SPACE);
+ }
+ if (entry->is_sub_map) {
+ RETURN(KERN_NO_SPACE);
+ }
+ if (VME_OBJECT(entry) == VM_OBJECT_NULL) {
+ RETURN(KERN_NO_SPACE);
+ }
+
+ /* go all the way down the shadow chain */
+ cur_object = VME_OBJECT(entry);
+ vm_object_lock(cur_object);
+ while (cur_object->shadow != VM_OBJECT_NULL) {
+ shadow_object = cur_object->shadow;
+ vm_object_lock(shadow_object);
+ vm_object_unlock(cur_object);
+ cur_object = shadow_object;
+ shadow_object = VM_OBJECT_NULL;
+ }
+ if (cur_object->internal ||
+ cur_object->pager == NULL) {
+ vm_object_unlock(cur_object);
+ RETURN(KERN_NO_SPACE);
+ }
+ if (cur_object->pager->mo_pager_ops
+ != &fourk_pager_ops) {
+ vm_object_unlock(cur_object);
+ RETURN(KERN_NO_SPACE);
+ }
+ fourk_object = cur_object;
+ fourk_mem_obj = fourk_object->pager;
+
+ /* keep the "4K" object alive */
+ vm_object_reference_locked(fourk_object);
+ vm_object_unlock(fourk_object);
+
+ /* merge permissions */
+ entry->protection |= cur_protection;
+ entry->max_protection |= max_protection;
+ if ((entry->protection & (VM_PROT_WRITE |
+ VM_PROT_EXECUTE)) ==
+ (VM_PROT_WRITE | VM_PROT_EXECUTE) &&
+ fourk_binary_compatibility_unsafe &&
+ fourk_binary_compatibility_allow_wx) {
+ /* write+execute: need to be "jit" */
+ entry->used_for_jit = TRUE;
+ }
+
+ goto map_in_fourk_pager;
+ }
+
+ /*
+ * ... 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);
+ }
+ }
+
+ /*
+ * At this point,
+ * "start" and "end" should define the endpoints of the
+ * available new range, and
+ * "entry" should refer to the region before the new
+ * range, and
+ *
+ * the map should be locked.
+ */
+
+ /* create a new "4K" pager */
+ fourk_mem_obj = fourk_pager_create();
+ fourk_object = fourk_pager_to_vm_object(fourk_mem_obj);
+ assert(fourk_object);
+
+ /* keep the "4" object alive */
+ vm_object_reference(fourk_object);
+
+ /* create a "copy" object, to map the "4K" object copy-on-write */
+ fourk_copy = TRUE;
+ result = vm_object_copy_strategically(fourk_object,
+ 0,
+ end - start,
+ ©_object,
+ ©_offset,
+ &fourk_copy);
+ assert(result == KERN_SUCCESS);
+ assert(copy_object != VM_OBJECT_NULL);
+ assert(copy_offset == 0);
+
+ /* take a reference on the copy object, for this mapping */
+ vm_object_reference(copy_object);
+
+ /* map the "4K" pager's copy object */
+ new_entry =
+ vm_map_entry_insert(map, entry,
+ vm_map_trunc_page(start,
+ VM_MAP_PAGE_MASK(map)),
+ vm_map_round_page(end,
+ VM_MAP_PAGE_MASK(map)),
+ copy_object,
+ 0, /* offset */
+ FALSE, /* 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,
+ FALSE, /* jit */
+ alias);
+ entry = new_entry;
+
+#if VM_MAP_DEBUG_FOURK
+ if (vm_map_debug_fourk) {
+ printf("FOURK_PAGER: map %p [0x%llx:0x%llx] new pager %p\n",
+ map,
+ (uint64_t) entry->vme_start,
+ (uint64_t) entry->vme_end,
+ fourk_mem_obj);
+ }
+#endif /* VM_MAP_DEBUG_FOURK */
+
+ new_mapping_established = TRUE;
+
+map_in_fourk_pager:
+ /* "map" the original "object" where it belongs in the "4K" pager */
+ fourk_pager_offset = (fourk_start & SIXTEENK_PAGE_MASK);
+ fourk_pager_index_start = (int) (fourk_pager_offset / FOURK_PAGE_SIZE);
+ if (fourk_size > SIXTEENK_PAGE_SIZE) {
+ fourk_pager_index_num = 4;
+ } else {
+ fourk_pager_index_num = (int) (fourk_size / FOURK_PAGE_SIZE);
+ }
+ if (fourk_pager_index_start + fourk_pager_index_num > 4) {
+ fourk_pager_index_num = 4 - fourk_pager_index_start;
+ }
+ for (cur_idx = 0;
+ cur_idx < fourk_pager_index_num;
+ cur_idx++) {
+ vm_object_t old_object;
+ vm_object_offset_t old_offset;
+
+ kr = fourk_pager_populate(fourk_mem_obj,
+ TRUE, /* overwrite */
+ fourk_pager_index_start + cur_idx,
+ object,
+ (object
+ ? (offset +
+ (cur_idx * FOURK_PAGE_SIZE))
+ : 0),
+ &old_object,
+ &old_offset);
+#if VM_MAP_DEBUG_FOURK
+ if (vm_map_debug_fourk) {
+ if (old_object == (vm_object_t) -1 &&
+ old_offset == (vm_object_offset_t) -1) {
+ printf("FOURK_PAGER: map %p [0x%llx:0x%llx] "
+ "pager [%p:0x%llx] "
+ "populate[%d] "
+ "[object:%p,offset:0x%llx]\n",
+ map,
+ (uint64_t) entry->vme_start,
+ (uint64_t) entry->vme_end,
+ fourk_mem_obj,
+ VME_OFFSET(entry),
+ fourk_pager_index_start + cur_idx,
+ object,
+ (object
+ ? (offset + (cur_idx * FOURK_PAGE_SIZE))
+ : 0));
+ } else {
+ printf("FOURK_PAGER: map %p [0x%llx:0x%llx] "
+ "pager [%p:0x%llx] "
+ "populate[%d] [object:%p,offset:0x%llx] "
+ "old [%p:0x%llx]\n",
+ map,
+ (uint64_t) entry->vme_start,
+ (uint64_t) entry->vme_end,
+ fourk_mem_obj,
+ VME_OFFSET(entry),
+ fourk_pager_index_start + cur_idx,
+ object,
+ (object
+ ? (offset + (cur_idx * FOURK_PAGE_SIZE))
+ : 0),
+ old_object,
+ old_offset);
+ }
+ }
+#endif /* VM_MAP_DEBUG_FOURK */
+
+ assert(kr == KERN_SUCCESS);
+ if (object != old_object &&
+ object != VM_OBJECT_NULL &&
+ object != (vm_object_t) -1) {
+ vm_object_reference(object);
+ }
+ if (object != old_object &&
+ old_object != VM_OBJECT_NULL &&
+ old_object != (vm_object_t) -1) {
+ vm_object_deallocate(old_object);
+ }
+ }
+
+BailOut:
+ assert(map_locked == TRUE);
+
+ if (fourk_object != VM_OBJECT_NULL) {
+ vm_object_deallocate(fourk_object);
+ fourk_object = VM_OBJECT_NULL;
+ fourk_mem_obj = MEMORY_OBJECT_NULL;
+ }
+
+ if (result == KERN_SUCCESS) {
+ vm_prot_t pager_prot;
+ memory_object_t pager;
+
+#if DEBUG
+ if (pmap_empty &&
+ !(vmk_flags.vmkf_no_pmap_check)) {
+ assert(vm_map_pmap_is_empty(map,
+ *address,
+ *address+size));
+ }
+#endif /* DEBUG */
+
+ /*
+ * 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_object_lock(object);
+ vm_object_mapping_end(object);
+ }
+ vm_object_unlock(object);
+ }
+ if (!is_submap &&
+ fourk_object != VM_OBJECT_NULL &&
+ fourk_object->named &&
+ fourk_object->pager != MEMORY_OBJECT_NULL) {
+ vm_object_lock(fourk_object);
+ pager = fourk_object->pager;
+ if (fourk_object->named &&
+ pager != MEMORY_OBJECT_NULL) {
+ assert(fourk_object->pager_ready);
+ vm_object_mapping_wait(fourk_object,
+ THREAD_UNINT);
+ vm_object_mapping_begin(fourk_object);
+ vm_object_unlock(fourk_object);
+
+ kr = memory_object_map(pager, VM_PROT_READ);
+ assert(kr == KERN_SUCCESS);
+
+ vm_object_lock(fourk_object);
+ vm_object_mapping_end(fourk_object);
+ }
+ vm_object_unlock(fourk_object);
+ }
+ }
+
+ assert(map_locked == TRUE);
+
+ if (!keep_map_locked) {
+ vm_map_unlock(map);
+ map_locked = FALSE;
+ }
+
+ /*
+ * We can't hold the map lock if we enter this block.
+ */
+
+ if (result == KERN_SUCCESS) {
+
+ /* 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_kernel(map, start, end,
+ new_entry->protection, VM_KERN_MEMORY_MLOCK,
+ TRUE);
+ result = kr;
+ }
+
+ }
+
+ 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;
+
+ /*
+ * 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;
+ }
+
+ /* 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,
+ VM_MAP_KERNEL_FLAGS_NONE);
+ map->size += entry_size;
+ entry1 = entry2;
+ }
+ if (map->wiring_required) {
+ /*
+ * XXX TODO: we should rewire the
+ * old pages here...
+ */
+ }
+ vm_map_enter_restore_successes++;
+ }
+ }
+ }
+
+ /*
+ * The caller is responsible for releasing the lock if it requested to
+ * keep the map locked.
+ */
+ if (map_locked && !keep_map_locked) {
+ vm_map_unlock(map);
+ }
+
+ /*
+ * Get rid of the "zap_maps" and all the map entries that
+ * they may still contain.
+ */
+ 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;
+ }
+ 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
+}
+#endif /* __arm64__ */
+
+/*
+ * Counters for the prefault optimization.
+ */
+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,
+ vm_map_kernel_flags_t vmk_flags,
+ vm_tag_t tag,
+ 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_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 kernel_prefault, try_prefault = (page_list_count != 0);
+ vm_map_offset_t offset_in_mapping = 0;
+#if __arm64__
+ boolean_t fourk = vmk_flags.vmkf_fourk;
+#endif /* __arm64__ */
+
+ assertf(vmk_flags.__vmkf_unused == 0, "vmk_flags unused=0x%x\n", vmk_flags.__vmkf_unused);
+
+ 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;
+
+ /*
+ * 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;
+ }
+
+#if __arm64__
+ if (fourk) {
+ map_addr = vm_map_trunc_page(*address, FOURK_PAGE_MASK);
+ map_size = vm_map_round_page(initial_size, FOURK_PAGE_MASK);
+ } else
+#endif /* __arm64__ */
+ {
+ 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);
+
+ /*
+ * 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;
+
+ named_entry = (vm_named_entry_t) port->ip_kobject;
+
+ 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;
/* overflow */
return KERN_INVALID_ARGUMENT;
}
- if (named_entry->size < (offset + size))
+ 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 */
size,
VM_MAP_PAGE_MASK(target_map));
}
-
- if (offset != 0 ||
- size != named_entry->size) {
+
+ 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;
}
}
/* 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))) {
/*
if (named_entry->is_sub_map) {
vm_map_t submap;
- if ((flags & VM_FLAGS_RETURN_DATA_ADDR) != 0) {
+ 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_unlock(submap);
named_entry_unlock(named_entry);
+ vmk_flags.vmkf_submap = TRUE;
+
result = vm_map_enter(target_map,
&map_addr,
map_size,
mask,
- flags | VM_FLAGS_SUBMAP,
- (vm_object_t) submap,
+ flags,
+ vmk_flags,
+ tag,
+ (vm_object_t)(uintptr_t) submap,
offset,
copy,
cur_protection,
* 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
+ * to indicate that we now need to
* remove mappings from all pmaps rather
* than just the submap's pmap.
*/
}
return result;
- } else if (named_entry->is_pager) {
- unsigned int access;
- vm_prot_t protections;
- unsigned int wimg_mode;
-
- protections = named_entry->protection & VM_PROT_ALL;
- access = GET_MAP_MEM(named_entry->protection);
-
- if ((flags & VM_FLAGS_RETURN_DATA_ADDR) != 0) {
- panic("VM_FLAGS_RETURN_DATA_ADDR not expected for submap.");
- }
-
- 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;
- }
-
- /* JMM - drop reference on pager here */
-
- /* 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;
- }
-
- /* 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 (object->wimg_bits != wimg_mode)
- vm_object_change_wimg_mode(object, wimg_mode);
-
- object->true_share = TRUE;
-
- if (object->copy_strategy == MEMORY_OBJECT_COPY_SYMMETRIC)
- object->copy_strategy = MEMORY_OBJECT_COPY_DELAY;
- vm_object_unlock(object);
-
} else if (named_entry->is_copy) {
kern_return_t kr;
vm_map_copy_t copy_map;
if (flags & ~(VM_FLAGS_FIXED |
VM_FLAGS_ANYWHERE |
VM_FLAGS_OVERWRITE |
- VM_FLAGS_RETURN_DATA_ADDR)) {
+ VM_FLAGS_RETURN_4K_DATA_ADDR |
+ VM_FLAGS_RETURN_DATA_ADDR |
+ VM_FLAGS_ALIAS_MASK)) {
named_entry_unlock(named_entry);
return KERN_INVALID_ARGUMENT;
}
- if ((flags & VM_FLAGS_RETURN_DATA_ADDR) != 0) {
+ 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;
}
/* reserve a contiguous range */
kr = vm_map_enter(target_map,
&map_addr,
- map_size,
+ /* 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),
+ vmk_flags,
+ tag,
VM_OBJECT_NULL,
0,
FALSE, /* copy */
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;
+ int remap_flags;
+ vm_map_kernel_flags_t vmk_remap_flags;
vm_map_t copy_submap;
vm_object_t copy_object;
vm_map_size_t copy_size;
vm_object_offset_t copy_offset;
+ int copy_vm_alias;
+
+ remap_flags = 0;
+ vmk_remap_flags = VM_MAP_KERNEL_FLAGS_NONE;
- copy_offset = copy_entry->offset;
+ copy_object = VME_OBJECT(copy_entry);
+ copy_offset = VME_OFFSET(copy_entry);
copy_size = (copy_entry->vme_end -
copy_entry->vme_start);
+ VM_GET_FLAGS_ALIAS(flags, copy_vm_alias);
+ if (copy_vm_alias == 0) {
+ /*
+ * Caller does not want a specific
+ * alias for this new mapping: use
+ * the alias of the original mapping.
+ */
+ copy_vm_alias = VME_ALIAS(copy_entry);
+ }
/* sanity check */
- if (copy_addr + copy_size >
- map_addr + map_size) {
+ if ((copy_addr + copy_size) >
+ (map_addr +
+ named_entry->size /* XXX full size */ )) {
/* over-mapping too much !? */
kr = KERN_INVALID_ARGUMENT;
/* abort */
/* take a reference on the object */
if (copy_entry->is_sub_map) {
- remap_flags |= VM_FLAGS_SUBMAP;
- copy_submap =
- copy_entry->object.sub_map;
+ vmk_remap_flags.vmkf_submap = TRUE;
+ 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;
+ copy_object = (vm_object_t)(uintptr_t) copy_submap;
+ } else if (!copy &&
+ copy_object != VM_OBJECT_NULL &&
+ (copy_entry->needs_copy ||
+ copy_object->shadowed ||
+ (!copy_object->true_share &&
+ !copy_entry->is_shared &&
+ copy_object->vo_size > copy_size))) {
+ /*
+ * We need to resolve our side of this
+ * "symmetric" copy-on-write now; we
+ * need a new object to map and share,
+ * instead of the current one which
+ * might still be shared with the
+ * original mapping.
+ *
+ * Note: A "vm_map_copy_t" does not
+ * have a lock but we're protected by
+ * the named entry's lock here.
+ */
+ // assert(copy_object->copy_strategy == MEMORY_OBJECT_COPY_SYMMETRIC);
+ VME_OBJECT_SHADOW(copy_entry, copy_size);
+ if (!copy_entry->needs_copy &&
+ copy_entry->protection & VM_PROT_WRITE) {
+ vm_prot_t prot;
+
+ prot = copy_entry->protection & ~VM_PROT_WRITE;
+ vm_object_pmap_protect(copy_object,
+ copy_offset,
+ copy_size,
+ PMAP_NULL,
+ 0,
+ prot);
+ }
+
+ copy_entry->needs_copy = FALSE;
+ copy_entry->is_shared = TRUE;
+ copy_object = VME_OBJECT(copy_entry);
+ copy_offset = VME_OFFSET(copy_entry);
+ vm_object_lock(copy_object);
+ vm_object_reference_locked(copy_object);
+ if (copy_object->copy_strategy == MEMORY_OBJECT_COPY_SYMMETRIC) {
+ /* we're about to make a shared mapping of this object */
+ copy_object->copy_strategy = MEMORY_OBJECT_COPY_DELAY;
+ copy_object->true_share = TRUE;
+ }
+ vm_object_unlock(copy_object);
} else {
- copy_object =
- copy_entry->object.vm_object;
+ /*
+ * We already have the right object
+ * to map.
+ */
+ copy_object = VME_OBJECT(copy_entry);
vm_object_reference(copy_object);
}
remap_flags |= VM_FLAGS_FIXED;
remap_flags |= VM_FLAGS_OVERWRITE;
remap_flags &= ~VM_FLAGS_ANYWHERE;
+ if (!copy && !copy_entry->is_sub_map) {
+ /*
+ * copy-on-write should have been
+ * resolved at this point, or we would
+ * end up sharing instead of copying.
+ */
+ assert(!copy_entry->needs_copy);
+ }
+#if !CONFIG_EMBEDDED
+ if (copy_entry->used_for_jit) {
+ vmk_remap_flags.vmkf_map_jit = TRUE;
+ }
+#endif /* !CONFIG_EMBEDDED */
kr = vm_map_enter(target_map,
©_addr,
copy_size,
(vm_map_offset_t) 0,
remap_flags,
+ vmk_remap_flags,
+ copy_vm_alias,
copy_object,
copy_offset,
copy,
/* next mapping */
copy_addr += copy_size;
}
-
+
if (kr == KERN_SUCCESS) {
- if ((flags & VM_FLAGS_RETURN_DATA_ADDR) != 0) {
+ 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,
+ VM_MAP_REMOVE_NO_FLAGS);
+ *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),
+ VM_MAP_REMOVE_NO_FLAGS);
+ }
}
named_entry_unlock(named_entry);
}
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) != 0) {
+ unsigned int access;
+ vm_prot_t protections;
+ unsigned int wimg_mode;
+
+ /* we are mapping a VM object */
+
+ protections = named_entry->protection & VM_PROT_ALL;
+ access = GET_MAP_MEM(named_entry->protection);
+
+ 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);
+ vm_object_lock(object);
named_entry_unlock(named_entry);
- vm_object_reference(object);
+
+ vm_object_reference_locked(object);
+
+ wimg_mode = object->wimg_bits;
+ vm_prot_to_wimg(access, &wimg_mode);
+ if (object->wimg_bits != wimg_mode)
+ vm_object_change_wimg_mode(object, wimg_mode);
+
+ vm_object_unlock(object);
}
} else if (ip_kotype(port) == IKOT_MEMORY_OBJECT) {
/*
* this case, the port isn't really a port at all, but
* instead is just a raw memory object.
*/
- if ((flags & VM_FLAGS_RETURN_DATA_ADDR) != 0) {
+ 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);
+ object = memory_object_to_vm_object((memory_object_t)port);
if (object == VM_OBJECT_NULL)
return KERN_INVALID_OBJECT;
+ vm_object_reference(object);
/* wait for object (if any) to be ready */
if (object != VM_OBJECT_NULL) {
vm_object_t new_object;
vm_object_offset_t new_offset;
- result = vm_object_copy_strategically(object, offset, size,
+ result = vm_object_copy_strategically(object, offset,
+ map_size,
&new_object, &new_offset,
©);
new_object = object;
new_offset = offset;
success = vm_object_copy_quickly(&new_object,
- new_offset, size,
+ new_offset,
+ map_size,
&src_needs_copy,
©);
assert(success);
vm_object_deallocate(object);
- if (result != KERN_SUCCESS)
+ if (result != KERN_SUCCESS) {
return result;
+ }
object = new_object;
offset = new_offset;
}
- 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 non-kernel users want to try to prefault pages, the mapping and prefault
+ * needs to be atomic.
+ */
+ kernel_prefault = (try_prefault && vm_kernel_map_is_kernel(target_map));
+ vmk_flags.vmkf_keep_map_locked = (try_prefault && !kernel_prefault);
+
+#if __arm64__
+ if (fourk) {
+ /* map this object in a "4K" pager */
+ result = vm_map_enter_fourk(target_map,
+ &map_addr,
+ map_size,
+ (vm_map_offset_t) mask,
+ flags,
+ vmk_flags,
+ tag,
+ object,
+ offset,
+ copy,
+ cur_protection,
+ max_protection,
+ inheritance);
+ } else
+#endif /* __arm64__ */
+ {
+ result = vm_map_enter(target_map,
+ &map_addr, map_size,
+ (vm_map_offset_t)mask,
+ flags,
+ vmk_flags,
+ tag,
+ object, offset,
+ copy,
+ cur_protection, max_protection,
+ inheritance);
+ }
if (result != KERN_SUCCESS)
vm_object_deallocate(object);
- if ((flags & VM_FLAGS_RETURN_DATA_ADDR) != 0) {
+ /*
+ * 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;
+ int pmap_options;
+
+ pmap_options = kernel_prefault ? 0 : PMAP_OPTIONS_NOWAIT;
+ if (object->internal) {
+ pmap_options |= PMAP_OPTIONS_INTERNAL;
+ }
+
+ for (i = 0; i < page_list_count; ++i) {
+ if (!UPL_VALID_PAGE(page_list, i)) {
+ if (kernel_prefault) {
+ assertf(FALSE, "kernel_prefault && !UPL_VALID_PAGE");
+ result = KERN_MEMORY_ERROR;
+ break;
+ }
+ } else {
+ /*
+ * 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, NULL);
+ if (kr != KERN_SUCCESS) {
+ OSIncrementAtomic64(&vm_prefault_nb_bailout);
+ if (kernel_prefault) {
+ result = kr;
+ }
+ break;
+ }
+ OSIncrementAtomic64(&vm_prefault_nb_pages);
+ }
+
+ /* Next virtual address */
+ va += PAGE_SIZE;
+ }
+ if (vmk_flags.vmkf_keep_map_locked) {
+ vm_map_unlock(target_map);
+ }
+ }
+
+ 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;
}
+kern_return_t
+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,
+ vm_map_kernel_flags_t vmk_flags,
+ vm_tag_t tag,
+ 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)
+{
+ kern_return_t ret;
+
+ ret = vm_map_enter_mem_object_helper(target_map,
+ address,
+ initial_size,
+ mask,
+ flags,
+ vmk_flags,
+ tag,
+ port,
+ offset,
+ copy,
+ cur_protection,
+ max_protection,
+ inheritance,
+ NULL,
+ 0);
+
+#if KASAN
+ if (ret == KERN_SUCCESS && address && target_map->pmap == kernel_pmap) {
+ kasan_notify_address(*address, initial_size);
+ }
+#endif
+
+ return ret;
+}
+
+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,
+ vm_map_kernel_flags_t vmk_flags,
+ vm_tag_t tag,
+ 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)
+{
+ kern_return_t ret;
+
+ ret = vm_map_enter_mem_object_helper(target_map,
+ address,
+ initial_size,
+ mask,
+ flags,
+ vmk_flags,
+ tag,
+ port,
+ offset,
+ FALSE,
+ cur_protection,
+ max_protection,
+ VM_INHERIT_DEFAULT,
+ page_list,
+ page_list_count);
+
+#if KASAN
+ if (ret == KERN_SUCCESS && address && target_map->pmap == kernel_pmap) {
+ kasan_notify_address(*address, initial_size);
+ }
+#endif
+ return ret;
+}
kern_return_t
vm_map_size_t initial_size,
vm_map_offset_t mask,
int flags,
+ vm_map_kernel_flags_t vmk_flags,
+ vm_tag_t tag,
memory_object_control_t control,
vm_object_offset_t offset,
boolean_t copy,
memory_object_t pager;
vm_prot_t pager_prot;
kern_return_t kr;
+#if __arm64__
+ boolean_t fourk = vmk_flags.vmkf_fourk;
+#endif /* __arm64__ */
/*
* Check arguments for validity
(cur_protection & ~VM_PROT_ALL) ||
(max_protection & ~VM_PROT_ALL) ||
(inheritance > VM_INHERIT_LAST_VALID) ||
- initial_size == 0)
+ 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);
+#if __arm64__
+ if (fourk) {
+ map_addr = vm_map_trunc_page(*address,
+ FOURK_PAGE_MASK);
+ map_size = vm_map_round_page(initial_size,
+ FOURK_PAGE_MASK);
+ } else
+#endif /* __arm64__ */
+ {
+ 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);
object = memory_object_control_to_vm_object(control);
vm_object_deallocate(object);
- if (result != KERN_SUCCESS)
+ if (result != KERN_SUCCESS) {
return result;
+ }
object = new_object;
offset = new_offset;
}
- 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 __arm64__
+ if (fourk) {
+ result = vm_map_enter_fourk(target_map,
+ &map_addr,
+ map_size,
+ (vm_map_offset_t)mask,
+ flags,
+ vmk_flags,
+ tag,
+ object, offset,
+ copy,
+ cur_protection, max_protection,
+ inheritance);
+ } else
+#endif /* __arm64__ */
+ {
+ result = vm_map_enter(target_map,
+ &map_addr, map_size,
+ (vm_map_offset_t)mask,
+ flags,
+ vmk_flags,
+ tag,
+ object, offset,
+ copy,
+ cur_protection, max_protection,
+ inheritance);
+ }
if (result != KERN_SUCCESS)
vm_object_deallocate(object);
*address = map_addr;
#endif /* MACH_ASSERT */
boolean_t anywhere = ((VM_FLAGS_ANYWHERE & flags) != 0);
+ vm_tag_t tag;
+
+ VM_GET_FLAGS_ALIAS(flags, tag);
if (size == 0) {
*addr = 0;
pages = NEXT_PAGE(m);
*(NEXT_PAGE_PTR(m)) = VM_PAGE_NULL;
- assert(!m->gobbled);
- assert(!m->wanted);
- assert(!m->pageout);
- assert(!m->tabled);
+ assert(!m->vmp_gobbled);
+ assert(!m->vmp_wanted);
+ assert(!m->vmp_pageout);
+ assert(!m->vmp_tabled);
assert(VM_PAGE_WIRED(m));
- /*
- * 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.
- */
- ASSERT_PAGE_DECRYPTED(m);
- assert(m->busy);
- assert(m->phys_page>=(avail_start>>PAGE_SHIFT) && m->phys_page<=(avail_end>>PAGE_SHIFT));
+ assert(m->vmp_busy);
+ assert(VM_PAGE_GET_PHYS_PAGE(m)>=(avail_start>>PAGE_SHIFT) && VM_PAGE_GET_PHYS_PAGE(m)<=(avail_end>>PAGE_SHIFT));
- m->busy = FALSE;
+ m->vmp_busy = FALSE;
vm_page_insert(m, cpm_obj, offset);
}
assert(cpm_obj->resident_page_count == size / PAGE_SIZE);
size,
(vm_map_offset_t)0,
flags,
+ VM_MAP_KERNEL_FLAGS_NONE,
cpm_obj,
(vm_object_offset_t)0,
FALSE,
type_of_fault = DBG_ZERO_FILL_FAULT;
vm_fault_enter(m, pmap, va, VM_PROT_ALL, VM_PROT_WRITE,
- VM_PAGE_WIRED(m), FALSE, FALSE, FALSE, NULL,
- &type_of_fault);
+ VM_PAGE_WIRED(m),
+ FALSE, /* change_wiring */
+ VM_KERN_MEMORY_NONE, /* tag - not wiring */
+ FALSE, /* no_cache */
+ FALSE, /* cs_bypass */
+ 0, /* user_tag */
+ 0, /* pmap_options */
+ NULL, /* need_retry */
+ &type_of_fault);
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);
+ assert(m->vmp_tabled);
+ assert(!m->vmp_busy);
+ assert(!m->vmp_wanted);
+ assert(!m->vmp_fictitious);
+ assert(!m->vmp_private);
+ assert(!m->vmp_absent);
+ assert(!m->vmp_error);
+ assert(!m->vmp_cleaning);
+ assert(!m->vmp_laundry);
+ assert(!m->vmp_precious);
+ assert(!m->vmp_clustered);
if (offset != 0) {
- if (m->phys_page != prev_addr + 1) {
+ if (VM_PAGE_GET_PHYS_PAGE(m) != 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);
panic("vm_allocate_cpm: pages not contig!");
}
}
- prev_addr = m->phys_page;
+ prev_addr = VM_PAGE_GET_PHYS_PAGE(m);
}
#endif /* MACH_ASSERT */
vm_map_offset_t old_end_unnest = end_unnest;
assert(entry->is_sub_map);
- assert(entry->object.sub_map != NULL);
+ assert(VME_SUBMAP(entry) != NULL);
+ assert(entry->use_pmap);
/*
* Query the platform for the optimal unnest range.
* 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);
+ assert(VME_SUBMAP(entry)->is_nested_map);
+ assert(!VME_SUBMAP(entry)->disable_vmentry_reuse);
+ log_unnest_badness(map,
+ old_start_unnest,
+ old_end_unnest,
+ VME_SUBMAP(entry)->is_nested_map,
+ (entry->vme_start +
+ VME_SUBMAP(entry)->lowest_unnestable_start -
+ VME_OFFSET(entry)));
}
if (entry->vme_start > start_unnest ||
_vm_map_clip_start(&map->hdr,
entry,
start_unnest);
- vm_map_store_update_first_free(map, map->first_free);
+ 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);
- vm_map_store_update_first_free(map, map->first_free);
+ if (map->holelistenabled) {
+ vm_map_store_update_first_free(map, NULL, FALSE);
+ } else {
+ vm_map_store_update_first_free(map, map->first_free, FALSE);
+ }
}
pmap_unnest(map->pmap,
entry->vme_start,
entry->vme_end - entry->vme_start);
- if ((map->mapped_in_other_pmaps) && (map->ref_count)) {
+ if ((map->mapped_in_other_pmaps) && (map->map_refcnt)) {
/* clean up parent map/maps */
vm_map_submap_pmap_clean(
map, entry->vme_start,
entry->vme_end,
- entry->object.sub_map,
- entry->offset);
+ VME_SUBMAP(entry),
+ VME_OFFSET(entry));
}
entry->use_pmap = FALSE;
- if (entry->alias == VM_MEMORY_SHARED_PMAP) {
- entry->alias = VM_MEMORY_UNSHARED_PMAP;
+ if ((map->pmap != kernel_pmap) &&
+ (VME_ALIAS(entry) == VM_MEMORY_SHARED_PMAP)) {
+ VME_ALIAS_SET(entry, VM_MEMORY_UNSHARED_PMAP);
}
}
#endif /* NO_NESTED_PMAP */
vm_map_offset_t startaddr)
{
#ifndef NO_NESTED_PMAP
- if (entry->use_pmap &&
+ if (entry->is_sub_map &&
+ entry->use_pmap &&
startaddr >= entry->vme_start) {
vm_map_offset_t start_unnest, end_unnest;
}
#endif /* NO_NESTED_PMAP */
if (startaddr > entry->vme_start) {
- if (entry->object.vm_object &&
+ if (VME_OBJECT(entry) &&
!entry->is_sub_map &&
- entry->object.vm_object->phys_contiguous) {
+ VME_OBJECT(entry)->phys_contiguous) {
pmap_remove(map->pmap,
(addr64_t)(entry->vme_start),
(addr64_t)(entry->vme_end));
}
+ if (entry->vme_atomic) {
+ panic("Attempting to clip an atomic VM entry! (map: %p, entry: %p)\n", map, entry);
+ }
+
+ DTRACE_VM5(
+ vm_map_clip_start,
+ vm_map_t, map,
+ vm_map_offset_t, entry->vme_start,
+ vm_map_offset_t, entry->vme_end,
+ vm_map_offset_t, startaddr,
+ int, VME_ALIAS(entry));
+
_vm_map_clip_start(&map->hdr, entry, startaddr);
- vm_map_store_update_first_free(map, map->first_free);
+ if (map->holelistenabled) {
+ vm_map_store_update_first_free(map, NULL, FALSE);
+ } else {
+ vm_map_store_update_first_free(map, map->first_free, FALSE);
+ }
}
}
*/
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)
+ struct vm_map_header *map_header,
+ vm_map_entry_t entry,
+ vm_map_offset_t start)
{
- register vm_map_entry_t new_entry;
+ vm_map_entry_t new_entry;
/*
* Split off the front portion --
* address.
*/
+ 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);
- assert(VM_MAP_PAGE_ALIGNED(start,
- VM_MAP_HDR_PAGE_MASK(map_header)));
new_entry->vme_end = start;
assert(new_entry->vme_start < new_entry->vme_end);
- entry->offset += (start - entry->vme_start);
+ VME_OFFSET_SET(entry, VME_OFFSET(entry) + (start - entry->vme_start));
assert(start < entry->vme_end);
- assert(VM_MAP_PAGE_ALIGNED(start,
- VM_MAP_HDR_PAGE_MASK(map_header)));
entry->vme_start = start;
_vm_map_store_entry_link(map_header, entry->vme_prev, new_entry);
if (entry->is_sub_map)
- vm_map_reference(new_entry->object.sub_map);
+ vm_map_reference(VME_SUBMAP(new_entry));
else
- vm_object_reference(new_entry->object.vm_object);
+ vm_object_reference(VME_OBJECT(new_entry));
}
endaddr = entry->vme_end;
}
#ifndef NO_NESTED_PMAP
- if (entry->use_pmap) {
+ if (entry->is_sub_map && entry->use_pmap) {
vm_map_offset_t start_unnest, end_unnest;
/*
}
#endif /* NO_NESTED_PMAP */
if (endaddr < entry->vme_end) {
- if (entry->object.vm_object &&
+ if (VME_OBJECT(entry) &&
!entry->is_sub_map &&
- entry->object.vm_object->phys_contiguous) {
+ VME_OBJECT(entry)->phys_contiguous) {
pmap_remove(map->pmap,
(addr64_t)(entry->vme_start),
(addr64_t)(entry->vme_end));
}
+ if (entry->vme_atomic) {
+ panic("Attempting to clip an atomic VM entry! (map: %p, entry: %p)\n", map, entry);
+ }
+ DTRACE_VM5(
+ vm_map_clip_end,
+ vm_map_t, map,
+ vm_map_offset_t, entry->vme_start,
+ vm_map_offset_t, entry->vme_end,
+ vm_map_offset_t, endaddr,
+ int, VME_ALIAS(entry));
+
_vm_map_clip_end(&map->hdr, entry, endaddr);
- vm_map_store_update_first_free(map, map->first_free);
+ if (map->holelistenabled) {
+ vm_map_store_update_first_free(map, NULL, FALSE);
+ } else {
+ vm_map_store_update_first_free(map, map->first_free, FALSE);
+ }
}
}
*/
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)
+ struct vm_map_header *map_header,
+ vm_map_entry_t entry,
+ vm_map_offset_t end)
{
- register vm_map_entry_t new_entry;
+ vm_map_entry_t new_entry;
/*
* Create a new entry and insert it
* AFTER the specified entry
*/
+ if (entry->map_aligned) {
+ assert(VM_MAP_PAGE_ALIGNED(end,
+ 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);
assert(entry->vme_start < end);
- assert(VM_MAP_PAGE_ALIGNED(end,
- VM_MAP_HDR_PAGE_MASK(map_header)));
new_entry->vme_start = entry->vme_end = end;
- new_entry->offset += (end - entry->vme_start);
+ 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(new_entry->object.sub_map);
+ vm_map_reference(VME_SUBMAP(new_entry));
else
- vm_object_reference(new_entry->object.vm_object);
+ vm_object_reference(VME_OBJECT(new_entry));
}
/*
* 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,
*/
static boolean_t
vm_map_range_check(
- register vm_map_t map,
- register vm_map_offset_t start,
- register vm_map_offset_t end,
+ vm_map_t map,
+ vm_map_offset_t start,
+ vm_map_offset_t end,
vm_map_entry_t *entry)
{
vm_map_entry_t cur;
- register vm_map_offset_t prev;
+ vm_map_offset_t prev;
/*
* Basic sanity checks first
return (FALSE);
/*
- * Optimize for the case that the region is contained
+ * Optimize for the case that the region is contained
* in a single map entry.
*/
if (entry != (vm_map_entry_t *) NULL)
*/
kern_return_t
vm_map_submap(
- vm_map_t map,
+ vm_map_t map,
vm_map_offset_t start,
vm_map_offset_t end,
- vm_map_t submap,
+ vm_map_t submap,
vm_map_offset_t offset,
#ifdef NO_NESTED_PMAP
__unused
#endif /* NO_NESTED_PMAP */
- boolean_t use_pmap)
+ boolean_t use_pmap)
{
vm_map_entry_t entry;
- register kern_return_t result = KERN_INVALID_ARGUMENT;
- register vm_object_t object;
+ kern_return_t result = KERN_INVALID_ARGUMENT;
+ vm_object_t object;
vm_map_lock(map);
return KERN_INVALID_ARGUMENT;
}
- assert(!entry->use_pmap); /* we don't want to unnest anything here */
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 = entry->object.vm_object) == vm_submap_object) &&
+ ((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)) {
- entry->offset = (vm_object_offset_t)offset;
- entry->object.vm_object = VM_OBJECT_NULL;
+ 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->object.sub_map = submap;
+ 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 &&
* 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
+ * to indicate that we now need to
* remove mappings from all pmaps rather
* than just the submap's pmap.
*/
vm_map_unlock(map);
return(KERN_NO_SPACE);
}
+#if defined(__arm__) || defined(__arm64__)
+ pmap_set_nested(submap->pmap);
+#endif
}
result = pmap_nest(map->pmap,
- (entry->object.sub_map)->pmap,
+ (VME_SUBMAP(entry))->pmap,
(addr64_t)start,
(addr64_t)start,
(uint64_t)(end - start));
*/
kern_return_t
vm_map_protect(
- register vm_map_t map,
- register vm_map_offset_t start,
- register vm_map_offset_t end,
- register vm_prot_t new_prot,
- register boolean_t set_max)
-{
- register vm_map_entry_t current;
- register vm_map_offset_t prev;
+ vm_map_t map,
+ vm_map_offset_t start,
+ vm_map_offset_t end,
+ vm_prot_t new_prot,
+ boolean_t set_max)
+{
+ vm_map_entry_t current;
+ vm_map_offset_t prev;
vm_map_entry_t entry;
vm_prot_t new_max;
+ int pmap_options = 0;
+ kern_return_t kr;
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);
+ if (new_prot & VM_PROT_COPY) {
+ vm_map_offset_t new_start;
+ vm_prot_t cur_prot, max_prot;
+ vm_map_kernel_flags_t kflags;
+
+ /* LP64todo - see below */
+ if (start >= map->max_offset) {
+ return KERN_INVALID_ADDRESS;
+ }
+
+#if VM_PROTECT_WX_FAIL
+ if ((new_prot & VM_PROT_EXECUTE) &&
+ map != kernel_map &&
+ cs_process_enforcement(NULL)) {
+ DTRACE_VM3(cs_wx,
+ uint64_t, (uint64_t) start,
+ uint64_t, (uint64_t) end,
+ vm_prot_t, new_prot);
+ printf("CODE SIGNING: %d[%s] %s can't have both write and exec at the same time\n",
+ proc_selfpid(),
+ (current_task()->bsd_info
+ ? proc_name_address(current_task()->bsd_info)
+ : "?"),
+ __FUNCTION__);
+ return KERN_PROTECTION_FAILURE;
+ }
+#endif /* VM_PROTECT_WX_FAIL */
+
+ /*
+ * Let vm_map_remap_extract() know that it will need to:
+ * + make a copy of the mapping
+ * + add VM_PROT_WRITE to the max protections
+ * + remove any protections that are no longer allowed from the
+ * max protections (to avoid any WRITE/EXECUTE conflict, for
+ * example).
+ * Note that "max_prot" is an IN/OUT parameter only for this
+ * specific (VM_PROT_COPY) case. It's usually an OUT parameter
+ * only.
+ */
+ max_prot = new_prot & VM_PROT_ALL;
+ kflags = VM_MAP_KERNEL_FLAGS_NONE;
+ kflags.vmkf_remap_prot_copy = TRUE;
+ kflags.vmkf_overwrite_immutable = TRUE;
+ new_start = start;
+ kr = vm_map_remap(map,
+ &new_start,
+ end - start,
+ 0, /* mask */
+ VM_FLAGS_FIXED | VM_FLAGS_OVERWRITE,
+ kflags,
+ 0,
+ map,
+ start,
+ TRUE, /* copy-on-write remapping! */
+ &cur_prot,
+ &max_prot,
+ VM_INHERIT_DEFAULT);
+ if (kr != KERN_SUCCESS) {
+ return kr;
+ }
+ new_prot &= ~VM_PROT_COPY;
+ }
+
vm_map_lock(map);
/* LP64todo - remove this check when vm_map_commpage64()
}
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) {
+ if ((new_prot & new_max) != new_prot) {
+ vm_map_unlock(map);
+ return(KERN_PROTECTION_FAILURE);
+ }
+
+ if ((new_prot & VM_PROT_WRITE) &&
+ (new_prot & VM_PROT_EXECUTE) &&
+#if !CONFIG_EMBEDDED
+ map != kernel_map &&
+ cs_process_enforcement(NULL) &&
+#endif /* !CONFIG_EMBEDDED */
+ !(current->used_for_jit)) {
+ DTRACE_VM3(cs_wx,
+ uint64_t, (uint64_t) current->vme_start,
+ uint64_t, (uint64_t) current->vme_end,
+ vm_prot_t, new_prot);
+ printf("CODE SIGNING: %d[%s] %s can't have both write and exec at the same time\n",
+ proc_selfpid(),
+ (current_task()->bsd_info
+ ? proc_name_address(current_task()->bsd_info)
+ : "?"),
+ __FUNCTION__);
+ new_prot &= ~VM_PROT_EXECUTE;
+#if VM_PROTECT_WX_FAIL
+ vm_map_unlock(map);
+ return KERN_PROTECTION_FAILURE;
+#endif /* VM_PROTECT_WX_FAIL */
+ }
+
+ /*
+ * If the task has requested executable lockdown,
+ * deny both:
+ * - adding executable protections OR
+ * - adding write protections to an existing executable mapping.
+ */
+ if (map->map_disallow_new_exec == TRUE) {
+ if ((new_prot & VM_PROT_EXECUTE) ||
+ ((current->protection & VM_PROT_EXECUTE) && (new_prot & VM_PROT_WRITE))) {
vm_map_unlock(map);
return(KERN_PROTECTION_FAILURE);
}
}
-
prev = current->vme_end;
current = current->vme_next;
}
+
+#if __arm64__
+ if (end > prev &&
+ end == vm_map_round_page(prev, VM_MAP_PAGE_MASK(map))) {
+ vm_map_entry_t prev_entry;
+
+ prev_entry = current->vme_prev;
+ if (prev_entry != vm_map_to_entry(map) &&
+ !prev_entry->map_aligned &&
+ (vm_map_round_page(prev_entry->vme_end,
+ VM_MAP_PAGE_MASK(map))
+ == end)) {
+ /*
+ * The last entry in our range is not "map-aligned"
+ * but it would have reached all the way to "end"
+ * if it had been map-aligned, so this is not really
+ * a hole in the range and we can proceed.
+ */
+ prev = end;
+ }
+ }
+#endif /* __arm64__ */
+
if (end > prev) {
vm_map_unlock(map);
return(KERN_INVALID_ADDRESS);
vm_map_clip_end(map, current, end);
- assert(!current->use_pmap); /* clipping did unnest if needed */
+ if (current->is_sub_map) {
+ /* clipping did unnest if needed */
+ assert(!current->use_pmap);
+ }
old_prot = current->protection;
- 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 && current->object.vm_object == VM_OBJECT_NULL){
- current->object.vm_object = vm_object_allocate((vm_map_size_t)(current->vme_end - current->vme_start));
- current->offset = 0;
- }
- current->needs_copy = TRUE;
- current->max_protection |= VM_PROT_WRITE;
+ if (set_max) {
+ current->max_protection = new_prot;
+ current->protection = new_prot & old_prot;
+ } else {
+ current->protection = new_prot;
}
- if (set_max)
- current->protection =
- (current->max_protection =
- new_prot & ~VM_PROT_COPY) &
- old_prot;
- else
- current->protection = new_prot & ~VM_PROT_COPY;
-
/*
* 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
+ * 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) {
vm_prot_t prot;
- prot = current->protection & ~VM_PROT_WRITE;
+ prot = current->protection;
+ if (current->is_sub_map || (VME_OBJECT(current) == NULL) || (VME_OBJECT(current) != compressor_object)) {
+ prot &= ~VM_PROT_WRITE;
+ } else {
+ assert(!VME_OBJECT(current)->code_signed);
+ assert(VME_OBJECT(current)->copy_strategy == MEMORY_OBJECT_COPY_NONE);
+ }
- if (override_nx(map, current->alias) && prot)
+ if (override_nx(map, VME_ALIAS(current)) && prot)
prot |= VM_PROT_EXECUTE;
+#if CONFIG_EMBEDDED && (DEVELOPMENT || DEBUG)
+ if (!(old_prot & VM_PROT_EXECUTE) &&
+ (prot & VM_PROT_EXECUTE) &&
+ panic_on_unsigned_execute &&
+ (proc_selfcsflags() & CS_KILL)) {
+ panic("vm_map_protect(%p,0x%llx,0x%llx) old=0x%x new=0x%x - <rdar://23770418> code-signing bypass?\n", map, (uint64_t)current->vme_start, (uint64_t)current->vme_end, old_prot, prot);
+ }
+#endif /* CONFIG_EMBEDDED && (DEVELOPMENT || DEBUG) */
+
+ if (pmap_has_prot_policy(prot)) {
+ if (current->wired_count) {
+ panic("vm_map_protect(%p,0x%llx,0x%llx) new=0x%x wired=%x\n",
+ map, (uint64_t)current->vme_start, (uint64_t)current->vme_end, prot, current->wired_count);
+ }
+
+ /* If the pmap layer cares about this
+ * protection type, force a fault for
+ * each page so that vm_fault will
+ * repopulate the page with the full
+ * set of protections.
+ */
+ /*
+ * TODO: We don't seem to need this,
+ * but this is due to an internal
+ * implementation detail of
+ * pmap_protect. Do we want to rely
+ * on this?
+ */
+ prot = VM_PROT_NONE;
+ }
+
if (current->is_sub_map && current->use_pmap) {
- pmap_protect(current->object.sub_map->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);
+ if (prot & VM_PROT_WRITE) {
+ if (VME_OBJECT(current) == compressor_object) {
+ /*
+ * For write requests on the
+ * compressor, we wil ask the
+ * pmap layer to prevent us from
+ * taking a write fault when we
+ * attempt to access the mapping
+ * next.
+ */
+ pmap_options |= PMAP_OPTIONS_PROTECT_IMMEDIATE;
+ }
+ }
+
+ pmap_protect_options(map->pmap,
+ current->vme_start,
+ current->vme_end,
+ prot,
+ pmap_options,
+ NULL);
}
}
current = current->vme_next;
*/
kern_return_t
vm_map_inherit(
- register vm_map_t map,
- register vm_map_offset_t start,
- register vm_map_offset_t end,
- register vm_inherit_t new_inheritance)
+ vm_map_t map,
+ vm_map_offset_t start,
+ vm_map_offset_t end,
+ vm_inherit_t new_inheritance)
{
- register vm_map_entry_t entry;
+ vm_map_entry_t entry;
vm_map_entry_t temp_entry;
vm_map_lock(map);
while ((entry != vm_map_to_entry(map)) && (entry->vme_start < end)) {
vm_map_clip_end(map, entry, end);
- assert(!entry->use_pmap); /* clip did unnest if needed */
+ if (entry->is_sub_map) {
+ /* clip did unnest if needed */
+ assert(!entry->use_pmap);
+ }
entry->inheritance = new_inheritance;
static kern_return_t
add_wire_counts(
vm_map_t map,
- vm_map_entry_t entry,
+ vm_map_entry_t entry,
boolean_t user_wire)
-{
+{
vm_map_size_t size;
if (user_wire) {
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
static void
subtract_wire_counts(
vm_map_t map,
- vm_map_entry_t entry,
+ vm_map_entry_t entry,
boolean_t user_wire)
-{
+{
if (user_wire) {
}
}
+int cs_executable_wire = 0;
+
/*
* vm_map_wire:
*
*/
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 access_type,
+ vm_map_t map,
+ vm_map_offset_t start,
+ vm_map_offset_t end,
+ vm_prot_t caller_prot,
+ vm_tag_t tag,
boolean_t user_wire,
- pmap_t map_pmap,
- vm_map_offset_t pmap_addr)
+ pmap_t map_pmap,
+ vm_map_offset_t pmap_addr,
+ ppnum_t *physpage_p)
{
- register vm_map_entry_t entry;
+ vm_map_entry_t entry;
+ 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;
+ vm_map_offset_t s,e;
kern_return_t rc;
boolean_t need_wakeup;
boolean_t main_map = FALSE;
thread_t cur_thread;
unsigned int last_timestamp;
vm_map_size_t size;
+ boolean_t wire_and_extract;
+
+ access_type = (caller_prot & VM_PROT_ALL);
+
+ 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;
+ }
vm_map_lock(map);
if(map_pmap == NULL)
/*
* User wiring is interruptible
*/
- wait_result = vm_map_entry_wait(map,
+ wait_result = vm_map_entry_wait(map,
(user_wire) ? THREAD_ABORTSAFE :
THREAD_UNINT);
if (user_wire && wait_result == THREAD_INTERRUPTED) {
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_end;
pmap_t pmap;
+ 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;
+ }
+
vm_map_clip_start(map, entry, s);
vm_map_clip_end(map, entry, end);
- sub_start = entry->offset;
+ sub_start = VME_OFFSET(entry);
sub_end = entry->vme_end;
- sub_end += entry->offset - entry->vme_start;
-
+ sub_end += VME_OFFSET(entry) - entry->vme_start;
+
local_end = entry->vme_end;
if(map_pmap == NULL) {
vm_object_t object;
vm_map_t lookup_map;
if(entry->use_pmap) {
- pmap = entry->object.sub_map->pmap;
+ pmap = VME_SUBMAP(entry)->pmap;
/* ppc implementation requires that */
/* submaps pmap address ranges line */
/* up with parent map */
lookup_map = map;
vm_map_lock_write_to_read(map);
if(vm_map_lookup_locked(
- &lookup_map, local_start,
- access_type,
+ &lookup_map, local_start,
+ access_type | VM_PROT_COPY,
OBJECT_LOCK_EXCLUSIVE,
&version, &object,
&offset, &prot, &wired,
&real_map)) {
vm_map_unlock_read(lookup_map);
+ assert(map_pmap == NULL);
vm_map_unwire(map, start,
s, user_wire);
return(KERN_FAILURE);
vm_map_lock(map);
/* we unlocked, so must re-lookup */
- if (!vm_map_lookup_entry(map,
+ if (!vm_map_lookup_entry(map,
local_start,
&local_entry)) {
rc = KERN_FAILURE;
entry->in_transition = TRUE;
vm_map_unlock(map);
- rc = vm_map_wire_nested(entry->object.sub_map,
+ rc = vm_map_wire_nested(VME_SUBMAP(entry),
sub_start, sub_end,
- access_type,
- user_wire, pmap, pmap_addr);
+ caller_prot, tag,
+ user_wire, pmap, pmap_addr,
+ NULL);
vm_map_lock(map);
/*
* 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;
+ }
+ }
+
/*
* 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);
- if ((rc = add_wire_counts(map, entry, user_wire)) != KERN_SUCCESS)
- goto done;
+ if ((rc = add_wire_counts(map, entry, user_wire)) != KERN_SUCCESS)
+ goto done;
+
+ if (wire_and_extract) {
+ vm_object_t object;
+ vm_object_offset_t offset;
+ vm_page_t m;
+
+ /*
+ * 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;
+ }
+
+ 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);
+ }
+ m = vm_page_lookup(object, offset);
+ assert(m != VM_PAGE_NULL);
+ assert(VM_PAGE_WIRED(m));
+ if (m != VM_PAGE_NULL && VM_PAGE_WIRED(m)) {
+ *physpage_p = VM_PAGE_GET_PHYS_PAGE(m);
+ if (entry->protection & VM_PROT_WRITE) {
+ vm_object_lock_assert_exclusive(
+ object);
+ m->vmp_dirty = TRUE;
+ }
+ } else {
+ /* not already wired !? */
+ *physpage_p = 0;
+ }
+ vm_object_unlock(object);
+ }
/* map was not unlocked: no need to relookup */
entry = entry->vme_next;
* Unwired entry or wire request transmitted via submap
*/
+ /*
+ * Wiring would copy the pages to the shadow object.
+ * The shadow object would not be code-signed so
+ * attempting to execute code from these copied pages
+ * would trigger a code-signing violation.
+ */
+
+ if ((entry->protection & VM_PROT_EXECUTE)
+#if !CONFIG_EMBEDDED
+ &&
+ map != kernel_map &&
+ cs_process_enforcement(NULL)
+#endif /* !CONFIG_EMBEDDED */
+ ) {
+#if MACH_ASSERT
+ printf("pid %d[%s] wiring executable range from "
+ "0x%llx to 0x%llx: rejected to preserve "
+ "code-signing\n",
+ proc_selfpid(),
+ (current_task()->bsd_info
+ ? proc_name_address(current_task()->bsd_info)
+ : "?"),
+ (uint64_t) entry->vme_start,
+ (uint64_t) entry->vme_end);
+#endif /* MACH_ASSERT */
+ DTRACE_VM2(cs_executable_wire,
+ uint64_t, (uint64_t)entry->vme_start,
+ uint64_t, (uint64_t)entry->vme_end);
+ cs_executable_wire++;
+ rc = KERN_PROTECTION_FAILURE;
+ goto done;
+ }
/*
* Perform actions of vm_map_lookup that need the write
* This is aggressive, but once it's wired we can't move it.
*/
if (entry->needs_copy) {
- vm_object_shadow(&entry->object.vm_object,
- &entry->offset, size);
+ 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;
+ }
+
+ VME_OBJECT_SHADOW(entry, 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;
+ } 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(map, entry, s);
interruptible_state = THREAD_UNINT;
if(map_pmap)
- rc = vm_fault_wire(map,
- &tmp_entry, map_pmap, pmap_addr);
+ rc = vm_fault_wire(map,
+ &tmp_entry, caller_prot, tag, map_pmap, pmap_addr,
+ physpage_p);
else
- rc = vm_fault_wire(map,
- &tmp_entry, map->pmap,
- tmp_entry.vme_start);
+ rc = vm_fault_wire(map,
+ &tmp_entry, caller_prot, tag, map->pmap,
+ tmp_entry.vme_start,
+ physpage_p);
if (!user_wire && cur_thread != THREAD_NULL)
thread_interrupt_level(interruptible_state);
goto done;
}
+ if ((entry != vm_map_to_entry(map)) && /* we still have entries in the map */
+ (tmp_entry.vme_end != end) && /* AND, we are not at the end of the requested range */
+ (entry->vme_start != tmp_entry.vme_end)) { /* AND, the next entry is not contiguous. */
+ /* found a "new" hole */
+ s = tmp_entry.vme_end;
+ rc = KERN_INVALID_ADDRESS;
+ goto done;
+ }
+
s = entry->vme_start;
+
} /* end while loop through map entries */
done:
if (rc != KERN_SUCCESS) {
/* undo what has been wired so far */
- vm_map_unwire(map, start, s, user_wire);
+ vm_map_unwire_nested(map, start, s, user_wire,
+ map_pmap, pmap_addr);
+ if (physpage_p) {
+ *physpage_p = 0;
+ }
}
return rc;
}
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 access_type,
+vm_map_wire_external(
+ vm_map_t map,
+ vm_map_offset_t start,
+ vm_map_offset_t end,
+ vm_prot_t caller_prot,
+ boolean_t user_wire)
+{
+ kern_return_t kret;
+
+ kret = vm_map_wire_nested(map, start, end, caller_prot, vm_tag_bt(),
+ user_wire, (pmap_t)NULL, 0, NULL);
+ return kret;
+}
+
+kern_return_t
+vm_map_wire_kernel(
+ vm_map_t map,
+ vm_map_offset_t start,
+ vm_map_offset_t end,
+ vm_prot_t caller_prot,
+ vm_tag_t tag,
boolean_t user_wire)
{
+ kern_return_t kret;
+
+ kret = vm_map_wire_nested(map, start, end, caller_prot, tag,
+ user_wire, (pmap_t)NULL, 0, NULL);
+ return kret;
+}
+
+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;
+
+ kret = vm_map_wire_nested(map,
+ start,
+ start+VM_MAP_PAGE_SIZE(map),
+ caller_prot,
+ vm_tag_bt(),
+ user_wire,
+ (pmap_t)NULL,
+ 0,
+ physpage_p);
+ if (kret != KERN_SUCCESS &&
+ physpage_p != NULL) {
+ *physpage_p = 0;
+ }
+ return kret;
+}
+kern_return_t
+vm_map_wire_and_extract_kernel(
+ vm_map_t map,
+ vm_map_offset_t start,
+ vm_prot_t caller_prot,
+ vm_tag_t tag,
+ boolean_t user_wire,
+ ppnum_t *physpage_p)
+{
kern_return_t kret;
- kret = vm_map_wire_nested(map, start, end, access_type,
- user_wire, (pmap_t)NULL, 0);
+ kret = vm_map_wire_nested(map,
+ start,
+ start+VM_MAP_PAGE_SIZE(map),
+ caller_prot,
+ tag,
+ user_wire,
+ (pmap_t)NULL,
+ 0,
+ physpage_p);
+ if (kret != KERN_SUCCESS &&
+ physpage_p != NULL) {
+ *physpage_p = 0;
+ }
return kret;
}
*/
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,
+ 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)
{
- register vm_map_entry_t entry;
+ vm_map_entry_t entry;
struct vm_map_entry *first_entry, tmp_entry;
boolean_t need_wakeup;
boolean_t main_map = FALSE;
* going on.
* We should probably just wait and retry,
* but then we have to be careful that this
- * entry could get "simplified" after
+ * 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
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);
- sub_start = entry->offset;
+ sub_start = VME_OFFSET(entry);
sub_end = entry->vme_end - entry->vme_start;
- sub_end += entry->offset;
+ sub_end += VME_OFFSET(entry);
local_end = entry->vme_end;
if(map_pmap == NULL) {
if(entry->use_pmap) {
- pmap = entry->object.sub_map->pmap;
+ pmap = VME_SUBMAP(entry)->pmap;
pmap_addr = sub_start;
} else {
pmap = map->pmap;
* Holes: Next entry should be contiguous unless
* this is the end of the region.
*/
- if (((entry->vme_end < end) &&
+ if (((entry->vme_end < end) &&
((entry->vme_next == vm_map_to_entry(map)) ||
- (entry->vme_next->vme_start
+ (entry->vme_next->vme_start
> entry->vme_end)))) {
if (!user_wire)
panic("vm_map_unwire: non-contiguous region");
* guarantees existance of the entry.
*/
vm_map_unlock(map);
- vm_map_unwire_nested(entry->object.sub_map,
+ vm_map_unwire_nested(VME_SUBMAP(entry),
sub_start, sub_end, user_wire, pmap, pmap_addr);
vm_map_lock(map);
if (last_timestamp+1 != map->timestamp) {
/*
- * Find the entry again. It could have been
+ * Find the entry again. It could have been
* clipped or deleted after we unlocked the map.
*/
- if (!vm_map_lookup_entry(map,
+ if (!vm_map_lookup_entry(map,
tmp_entry.vme_start,
&first_entry)) {
if (!user_wire)
/*
* clear transition bit for all constituent entries
- * that were in the original entry (saved in
+ * that were in the original entry (saved in
* tmp_entry). Also check for waiters.
*/
while ((entry != vm_map_to_entry(map)) &&
continue;
} else {
vm_map_unlock(map);
- vm_map_unwire_nested(entry->object.sub_map,
+ vm_map_unwire_nested(VME_SUBMAP(entry),
sub_start, sub_end, user_wire, map_pmap,
pmap_addr);
vm_map_lock(map);
if (last_timestamp+1 != map->timestamp) {
/*
- * Find the entry again. It could have been
+ * Find the entry again. It could have been
* clipped or deleted after we unlocked the map.
*/
- if (!vm_map_lookup_entry(map,
+ if (!vm_map_lookup_entry(map,
tmp_entry.vme_start,
&first_entry)) {
if (!user_wire)
entry = entry->vme_next;
continue;
}
-
+
assert(entry->wired_count > 0 &&
(!user_wire || entry->user_wired_count > 0));
* Holes: Next entry should be contiguous unless
* this is the end of the region.
*/
- if (((entry->vme_end < end) &&
+ if (((entry->vme_end < end) &&
((entry->vme_next == vm_map_to_entry(map)) ||
(entry->vme_next->vme_start > entry->vme_end)))) {
*/
vm_map_unlock(map);
if(map_pmap) {
- vm_fault_unwire(map,
+ vm_fault_unwire(map,
&tmp_entry, FALSE, map_pmap, pmap_addr);
} else {
- vm_fault_unwire(map,
- &tmp_entry, FALSE, map->pmap,
+ vm_fault_unwire(map,
+ &tmp_entry, FALSE, map->pmap,
tmp_entry.vme_start);
}
vm_map_lock(map);
kern_return_t
vm_map_unwire(
- register vm_map_t map,
- register vm_map_offset_t start,
- register vm_map_offset_t end,
+ vm_map_t map,
+ vm_map_offset_t start,
+ vm_map_offset_t end,
boolean_t user_wire)
{
- return vm_map_unwire_nested(map, start, end,
+ return vm_map_unwire_nested(map, start, end,
user_wire, (pmap_t)NULL, 0);
}
* vm_map_entry_delete: [ internal use only ]
*
* 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)
+ vm_map_t map,
+ vm_map_entry_t entry)
{
- register vm_map_offset_t s, e;
- register vm_object_t object;
- register vm_map_t submap;
+ vm_map_offset_t s, e;
+ vm_object_t object;
+ vm_map_t submap;
s = entry->vme_start;
e = entry->vme_end;
if (entry->is_sub_map) {
object = NULL;
- submap = entry->object.sub_map;
+ submap = VME_SUBMAP(entry);
} else {
submap = NULL;
- object = entry->object.vm_object;
+ object = VME_OBJECT(entry);
}
vm_map_store_entry_unlink(map, entry);
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;
-
+
if(submap_end < entry->vme_end) {
remove_size -=
sub_map,
start,
start + remove_size,
- entry->object.sub_map,
- entry->offset);
+ VME_SUBMAP(entry),
+ VME_OFFSET(entry));
} else {
- if((map->mapped_in_other_pmaps) && (map->ref_count)
- && (entry->object.vm_object != NULL)) {
- vm_object_pmap_protect(
- entry->object.vm_object,
- entry->offset+(offset-entry->vme_start),
+ if((map->mapped_in_other_pmaps) && (map->map_refcnt)
+ && (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);
+ VM_PROT_NONE,
+ PMAP_OPTIONS_REMOVE);
} else {
- pmap_remove(map->pmap,
- (addr64_t)start,
+ pmap_remove(map->pmap,
+ (addr64_t)start,
(addr64_t)(start + remove_size));
}
}
entry = entry->vme_next;
- while((entry != vm_map_to_entry(sub_map))
+ while((entry != vm_map_to_entry(sub_map))
&& (entry->vme_start < submap_end)) {
- remove_size = (entry->vme_end - entry->vme_start);
+ remove_size = (entry->vme_end - entry->vme_start);
if(submap_end < entry->vme_end) {
remove_size -= entry->vme_end - submap_end;
}
sub_map,
(start + entry->vme_start) - offset,
((start + entry->vme_start) - offset) + remove_size,
- entry->object.sub_map,
- entry->offset);
+ VME_SUBMAP(entry),
+ VME_OFFSET(entry));
} else {
- if((map->mapped_in_other_pmaps) && (map->ref_count)
- && (entry->object.vm_object != NULL)) {
- vm_object_pmap_protect(
- entry->object.vm_object,
- entry->offset,
+ if((map->mapped_in_other_pmaps) && (map->map_refcnt)
+ && (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);
+ VM_PROT_NONE,
+ PMAP_OPTIONS_REMOVE);
} else {
- pmap_remove(map->pmap,
- (addr64_t)((start + entry->vme_start)
+ pmap_remove(map->pmap,
+ (addr64_t)((start + entry->vme_start)
- offset),
- (addr64_t)(((start + entry->vme_start)
+ (addr64_t)(((start + entry->vme_start)
- offset) + remove_size));
}
}
return;
}
+/*
+ * virt_memory_guard_ast:
+ *
+ * Handle the AST callout for a virtual memory guard.
+ * raise an EXC_GUARD exception and terminate the task
+ * if configured to do so.
+ */
+void
+virt_memory_guard_ast(
+ thread_t thread,
+ mach_exception_data_type_t code,
+ mach_exception_data_type_t subcode)
+{
+ task_t task = thread->task;
+ assert(task != kernel_task);
+ assert(task == current_task());
+ uint32_t behavior;
+
+ behavior = task->task_exc_guard;
+
+ /* Is delivery enabled */
+ if ((behavior & TASK_EXC_GUARD_VM_DELIVER) == 0) {
+ return;
+ }
+
+ /* If only once, make sure we're that once */
+ while (behavior & TASK_EXC_GUARD_VM_ONCE) {
+ uint32_t new_behavior = behavior & ~TASK_EXC_GUARD_VM_DELIVER;
+
+ if (OSCompareAndSwap(behavior, new_behavior, &task->task_exc_guard)) {
+ break;
+ }
+ behavior = task->task_exc_guard;
+ if ((behavior & TASK_EXC_GUARD_VM_DELIVER) == 0) {
+ return;
+ }
+ }
+
+ /* Raise exception via corpse fork or synchronously */
+ if ((task->task_exc_guard & TASK_EXC_GUARD_VM_CORPSE) &&
+ (task->task_exc_guard & TASK_EXC_GUARD_VM_FATAL) == 0) {
+ task_violated_guard(code, subcode, NULL);
+ } else {
+ task_exception_notify(EXC_GUARD, code, subcode);
+ }
+
+ /* Terminate the task if desired */
+ if (task->task_exc_guard & TASK_EXC_GUARD_VM_FATAL) {
+ task_bsdtask_kill(current_task());
+ }
+}
+
+/*
+ * vm_map_guard_exception:
+ *
+ * Generate a GUARD_TYPE_VIRTUAL_MEMORY EXC_GUARD exception.
+ *
+ * Right now, we do this when we find nothing mapped, or a
+ * gap in the mapping when a user address space deallocate
+ * was requested. We report the address of the first gap found.
+ */
+static void
+vm_map_guard_exception(
+ vm_map_offset_t gap_start,
+ unsigned reason)
+{
+ mach_exception_code_t code = 0;
+ unsigned int guard_type = GUARD_TYPE_VIRT_MEMORY;
+ unsigned int target = 0; /* should we pass in pid associated with map? */
+ mach_exception_data_type_t subcode = (uint64_t)gap_start;
+
+ /* Can't deliver exceptions to kernel task */
+ if (current_task() == kernel_task)
+ return;
+
+ EXC_GUARD_ENCODE_TYPE(code, guard_type);
+ EXC_GUARD_ENCODE_FLAVOR(code, reason);
+ EXC_GUARD_ENCODE_TARGET(code, target);
+ thread_guard_violation(current_thread(), code, subcode);
+}
+
/*
* vm_map_delete: [ internal use only ]
*
{
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;
+ vm_map_offset_t s;
+ vm_object_t object;
boolean_t need_wakeup;
unsigned int last_timestamp = ~0; /* unlikely value */
int interruptible;
+ vm_map_offset_t gap_start;
+ vm_map_offset_t save_start = start;
+ vm_map_offset_t save_end = end;
+ const vm_map_offset_t FIND_GAP = 1; /* a not page aligned value */
+ const vm_map_offset_t GAPS_OK = 2; /* a different not page aligned value */
+
+ if (map != kernel_map && !(flags & VM_MAP_REMOVE_GAPS_OK))
+ gap_start = FIND_GAP;
+ else
+ gap_start = GAPS_OK;
- interruptible = (flags & VM_MAP_REMOVE_INTERRUPTIBLE) ?
+ interruptible = (flags & VM_MAP_REMOVE_INTERRUPTIBLE) ?
THREAD_ABORTSAFE : THREAD_UNINT;
/*
*/
if (vm_map_lookup_entry(map, start, &first_entry)) {
entry = first_entry;
- if (entry->superpage_size && (start & ~SUPERPAGE_MASK)) { /* extend request to whole entry */ start = SUPERPAGE_ROUND_DOWN(start);
+ 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 in a superpage, extend the range to include the start of the mapping.
+ */
+ if (entry->superpage_size && (start & ~SUPERPAGE_MASK)) {
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);
}
* time through the loop.
*/
SAVE_HINT_MAP_WRITE(map, entry->vme_prev);
+
} else {
+
+ if (map->pmap == kernel_pmap &&
+ map->map_refcnt != 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;
+ if (gap_start == FIND_GAP)
+ gap_start = start;
}
break;
}
* 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;
+ }
+ 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) {
* 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 (entry->permanent) {
- panic("attempt to remove permanent VM map entry "
- "%p [0x%llx:0x%llx]\n",
- entry, (uint64_t) s, (uint64_t) end);
+ if (map->pmap == kernel_pmap) {
+ panic("%s(%p,0x%llx,0x%llx): "
+ "attempt to remove permanent "
+ "VM map entry "
+ "%p [0x%llx:0x%llx]\n",
+ __FUNCTION__,
+ map,
+ (uint64_t) start,
+ (uint64_t) end,
+ entry,
+ (uint64_t) entry->vme_start,
+ (uint64_t) entry->vme_end);
+ } else if (flags & VM_MAP_REMOVE_IMMUTABLE) {
+// printf("FBDP %d[%s] removing permanent entry %p [0x%llx:0x%llx] prot 0x%x/0x%x\n", proc_selfpid(), (current_task()->bsd_info ? proc_name_address(current_task()->bsd_info) : "?"), entry, (uint64_t)entry->vme_start, (uint64_t)entry->vme_end, entry->protection, entry->max_protection);
+ entry->permanent = FALSE;
+#if PMAP_CS
+ } else if ((entry->protection & VM_PROT_EXECUTE) && !pmap_cs_enforced(map->pmap)) {
+ entry->permanent = FALSE;
+
+ printf("%d[%s] %s(0x%llx,0x%llx): "
+ "pmap_cs disabled, allowing for permanent executable entry [0x%llx:0x%llx] "
+ "prot 0x%x/0x%x\n",
+ proc_selfpid(),
+ (current_task()->bsd_info
+ ? proc_name_address(current_task()->bsd_info)
+ : "?"),
+ __FUNCTION__,
+ (uint64_t) start,
+ (uint64_t) end,
+ (uint64_t)entry->vme_start,
+ (uint64_t)entry->vme_end,
+ entry->protection,
+ entry->max_protection);
+#endif
+ } else {
+ if (vm_map_executable_immutable_verbose) {
+ printf("%d[%s] %s(0x%llx,0x%llx): "
+ "permanent entry [0x%llx:0x%llx] "
+ "prot 0x%x/0x%x\n",
+ proc_selfpid(),
+ (current_task()->bsd_info
+ ? proc_name_address(current_task()->bsd_info)
+ : "?"),
+ __FUNCTION__,
+ (uint64_t) start,
+ (uint64_t) end,
+ (uint64_t)entry->vme_start,
+ (uint64_t)entry->vme_end,
+ entry->protection,
+ entry->max_protection);
+ }
+ /*
+ * dtrace -n 'vm_map_delete_permanent { print("start=0x%llx end=0x%llx prot=0x%x/0x%x\n", arg0, arg1, arg2, arg3); stack(); ustack(); }'
+ */
+ DTRACE_VM5(vm_map_delete_permanent,
+ vm_map_offset_t, entry->vme_start,
+ vm_map_offset_t, entry->vme_end,
+ vm_prot_t, entry->protection,
+ vm_prot_t, entry->max_protection,
+ int, VME_ALIAS(entry));
+ }
}
* 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;
}
* 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
*/
+ if (gap_start == FIND_GAP)
+ gap_start = s;
entry = first_entry->vme_next;
s = entry->vme_start;
} else {
if (flags & VM_MAP_REMOVE_KUNWIRE) {
entry->wired_count--;
}
-
+
/*
* Remove all user wirings for proper accounting
*/
if (interruptible &&
wait_result == THREAD_INTERRUPTED) {
/*
- * We do not clear the
- * needs_wakeup flag, since we
- * cannot tell if we were the
+ * 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;
}
* it may not exist anymore. Look it
* up again.
*/
- if (!vm_map_lookup_entry(map, s,
+ if (!vm_map_lookup_entry(map, s,
&first_entry)) {
assert(map != kernel_map);
/*
* User: use the next entry
*/
+ if (gap_start == FIND_GAP)
+ gap_start = s;
entry = first_entry->vme_next;
s = entry->vme_start;
} else {
vm_map_offset_t sub_start, sub_end;
pmap_t pmap;
vm_map_offset_t pmap_addr;
-
- sub_map = tmp_entry.object.sub_map;
- sub_start = tmp_entry.offset;
+
+ 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, pmap_addr);
} else {
- if (tmp_entry.object.vm_object == kernel_object) {
+ if (VME_OBJECT(&tmp_entry) == kernel_object) {
pmap_protect_options(
map->pmap,
tmp_entry.vme_start,
NULL);
}
vm_fault_unwire(map, &tmp_entry,
- tmp_entry.object.vm_object == kernel_object,
+ VME_OBJECT(&tmp_entry) == kernel_object,
map->pmap, tmp_entry.vme_start);
}
* been clipped after we unlocked the map.
*/
if (!vm_map_lookup_entry(map, s, &first_entry)){
- assert((map != kernel_map) &&
+ assert((map != kernel_map) &&
(!entry->is_sub_map));
+ if (gap_start == FIND_GAP)
+ gap_start = s;
first_entry = first_entry->vme_next;
s = first_entry->vme_start;
} else {
} else if (entry->is_sub_map) {
if (entry->use_pmap) {
#ifndef NO_NESTED_PMAP
- pmap_unnest(map->pmap,
- (addr64_t)entry->vme_start,
- entry->vme_end - entry->vme_start);
+ int pmap_flags;
+
+ 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)) {
+ if ((map->mapped_in_other_pmaps) && (map->map_refcnt)) {
/* clean up parent map/maps */
vm_map_submap_pmap_clean(
map, entry->vme_start,
entry->vme_end,
- entry->object.sub_map,
- entry->offset);
+ VME_SUBMAP(entry),
+ VME_OFFSET(entry));
}
} else {
vm_map_submap_pmap_clean(
map, entry->vme_start, entry->vme_end,
- entry->object.sub_map,
- entry->offset);
+ VME_SUBMAP(entry),
+ VME_OFFSET(entry));
}
- } else if (entry->object.vm_object != kernel_object &&
- entry->object.vm_object != compressor_object) {
- object = entry->object.vm_object;
- if ((map->mapped_in_other_pmaps) && (map->ref_count)) {
+ } else if (VME_OBJECT(entry) != kernel_object &&
+ VME_OBJECT(entry) != compressor_object) {
+ object = VME_OBJECT(entry);
+ if ((map->mapped_in_other_pmaps) && (map->map_refcnt)) {
vm_object_pmap_protect_options(
- object, entry->offset,
+ object, VME_OFFSET(entry),
entry->vme_end - entry->vme_start,
PMAP_NULL,
entry->vme_start,
VM_PROT_NONE,
PMAP_OPTIONS_REMOVE);
- } else if ((entry->object.vm_object !=
- VM_OBJECT_NULL) ||
+ } else if ((VME_OBJECT(entry) != VM_OBJECT_NULL) ||
(map->pmap == kernel_pmap)) {
/* Remove translations associated
* with this range unless the entry
}
}
+ 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;
+ entry->use_pmap = FALSE;
+ }
+
/*
* 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 */
next = entry->vme_next;
+
+ if (map->pmap == kernel_pmap &&
+ map->map_refcnt != 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);
+ }
+
+ /*
+ * If the desired range didn't end with "entry", then there is a gap if
+ * we wrapped around to the start of the map or if "entry" and "next"
+ * aren't contiguous.
+ *
+ * The vm_map_round_page() is needed since an entry can be less than VM_MAP_PAGE_MASK() sized.
+ * For example, devices which have h/w 4K pages, but entry sizes are all now 16K.
+ */
+ if (gap_start == FIND_GAP &&
+ vm_map_round_page(entry->vme_end, VM_MAP_PAGE_MASK(map)) < end &&
+ (next == vm_map_to_entry(map) || entry->vme_end != next->vme_start)) {
+ gap_start = entry->vme_end;
+ }
s = next->vme_start;
last_timestamp = map->timestamp;
- if ((flags & VM_MAP_REMOVE_SAVE_ENTRIES) &&
- zap_map != VM_MAP_NULL) {
+ if (entry->permanent) {
+ /*
+ * A permanent entry can not be removed, so leave it
+ * in place but remove all access permissions.
+ */
+ entry->protection = VM_PROT_NONE;
+ entry->max_protection = VM_PROT_NONE;
+ } else 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
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);
+ vm_map_last_entry(zap_map),
+ entry,
+ VM_MAP_KERNEL_FLAGS_NONE);
entry_size = entry->vme_end - entry->vme_start;
map->size -= entry_size;
zap_map->size += entry_size;
if(entry == vm_map_to_entry(map)) {
break;
}
- if (last_timestamp+1 != map->timestamp) {
+ 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 are responsible for deleting everything
+ * from the given 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;
+
+ /*
+ * Nothing found for s. If we weren't already done, then there is a gap.
+ */
+ if (gap_start == FIND_GAP && s < end)
+ gap_start = s;
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
+ /*
+ * 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)) {
+ if (entry == vm_map_to_entry(map)) {
break;
}
}
if (need_wakeup)
vm_map_entry_wakeup(map);
+ if (gap_start != FIND_GAP && gap_start != GAPS_OK) {
+ DTRACE_VM3(kern_vm_deallocate_gap,
+ vm_map_offset_t, gap_start,
+ vm_map_offset_t, save_start,
+ vm_map_offset_t, save_end);
+ if (!(flags & VM_MAP_REMOVE_GAPS_OK)) {
+#if defined(DEVELOPMENT) || defined(DEBUG)
+ /* log just once if not checking, otherwise log each one */
+ if (!map->warned_delete_gap ||
+ (task_exc_guard_default & TASK_EXC_GUARD_VM_ALL) != 0) {
+ printf("vm_map_delete: map %p [%p...%p] nothing at %p\n",
+ (void *)map, (void *)save_start, (void *)save_end,
+ (void *)gap_start);
+ if (!map->warned_delete_gap) {
+ map->warned_delete_gap = 1;
+ }
+ }
+#endif
+ vm_map_guard_exception(gap_start, kGUARD_EXC_DEALLOC_GAP);
+ }
+ }
+
return KERN_SUCCESS;
}
*/
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)
+ vm_map_t map,
+ vm_map_offset_t start,
+ vm_map_offset_t end,
+ boolean_t flags)
{
- register kern_return_t result;
+ kern_return_t result;
vm_map_lock(map);
VM_MAP_RANGE_CHECK(map, start, end);
return(result);
}
+/*
+ * vm_map_remove_locked:
+ *
+ * Remove the given address range from the target locked map.
+ * This is the exported form of vm_map_delete.
+ */
+kern_return_t
+vm_map_remove_locked(
+ vm_map_t map,
+ vm_map_offset_t start,
+ vm_map_offset_t end,
+ boolean_t flags)
+{
+ kern_return_t result;
+
+ VM_MAP_RANGE_CHECK(map, start, end);
+ result = vm_map_delete(map, start, end, flags, VM_MAP_NULL);
+ return(result);
+}
+
+
+/*
+ * Routine: vm_map_copy_allocate
+ *
+ * Description:
+ * Allocates and initializes a map copy object.
+ */
+static vm_map_copy_t
+vm_map_copy_allocate(void)
+{
+ vm_map_copy_t new_copy;
+
+ new_copy = zalloc(vm_map_copy_zone);
+ bzero(new_copy, sizeof (*new_copy));
+ new_copy->c_u.hdr.rb_head_store.rbh_root = (void*)(int)SKIP_RB_TREE;
+ vm_map_copy_first_entry(new_copy) = vm_map_copy_to_entry(new_copy);
+ vm_map_copy_last_entry(new_copy) = vm_map_copy_to_entry(new_copy);
+ return new_copy;
+}
/*
* Routine: vm_map_copy_discard
vm_map_copy_entry_unlink(copy, entry);
if (entry->is_sub_map) {
- vm_map_deallocate(entry->object.sub_map);
+ vm_map_deallocate(VME_SUBMAP(entry));
} else {
- vm_object_deallocate(entry->object.vm_object);
+ vm_object_deallocate(VME_OBJECT(entry));
}
vm_map_copy_entry_dispose(copy, entry);
}
* allocated by a single call to kalloc(), i.e. the
* vm_map_copy_t was not allocated out of the zone.
*/
- kfree(copy, copy->cpy_kalloc_size);
+ 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);
tmp_entry,
vm_map_trunc_page(dst_addr,
VM_MAP_PAGE_MASK(dst_map)));
- assert(!tmp_entry->use_pmap); /* clipping did unnest if needed */
+ if (tmp_entry->is_sub_map) {
+ /* clipping did unnest if needed */
+ assert(!tmp_entry->use_pmap);
+ }
for (entry = tmp_entry;;) {
vm_map_entry_t next;
}
encountered_sub_map = TRUE;
- sub_start = entry->offset;
+ sub_start = VME_OFFSET(entry);
if(entry->vme_end < dst_end)
sub_end = entry->vme_end;
- else
+ else
sub_end = dst_end;
sub_end -= entry->vme_start;
- sub_end += entry->offset;
+ sub_end += VME_OFFSET(entry);
local_end = entry->vme_end;
vm_map_unlock(dst_map);
-
+
result = vm_map_overwrite_submap_recurse(
- entry->object.sub_map,
+ VME_SUBMAP(entry),
sub_start,
sub_end - sub_start);
if (dst_end <= entry->vme_end)
return KERN_SUCCESS;
vm_map_lock(dst_map);
- if(!vm_map_lookup_entry(dst_map, local_end,
+ if(!vm_map_lookup_entry(dst_map, local_end,
&tmp_entry)) {
vm_map_unlock(dst_map);
return(KERN_INVALID_ADDRESS);
/*
* 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))) {
+ 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);
if (copy->type == VM_MAP_COPY_KERNEL_BUFFER) {
return(vm_map_copyout_kernel_buffer(
- dst_map, &dst_addr,
- copy, TRUE, discard_on_success));
+ dst_map, &dst_addr,
+ copy, copy->size, TRUE, discard_on_success));
}
/*
!VM_MAP_PAGE_ALIGNED(copy->offset,
VM_MAP_PAGE_MASK(dst_map)) ||
!VM_MAP_PAGE_ALIGNED(dst_addr,
- VM_MAP_PAGE_MASK(dst_map)) ||
- dst_map->hdr.page_shift != copy->cpy_hdr.page_shift)
+ VM_MAP_PAGE_MASK(dst_map)))
{
aligned = FALSE;
dst_end = vm_map_round_page(dst_addr + copy->size,
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);
/* there is no need for the follow- */
/* ing check. */
encountered_sub_map = TRUE;
- sub_start = entry->offset;
+ sub_start = VME_OFFSET(entry);
if(entry->vme_end < dst_end)
sub_end = entry->vme_end;
- else
+ else
sub_end = dst_end;
sub_end -= entry->vme_start;
- sub_end += entry->offset;
+ sub_end += VME_OFFSET(entry);
vm_map_unlock(dst_map);
-
+
kr = vm_map_overwrite_submap_recurse(
- entry->object.sub_map,
+ VME_SUBMAP(entry),
sub_start,
sub_end - sub_start);
if(kr != KERN_SUCCESS)
if (dst_end <= entry->vme_end)
goto start_overwrite;
- if(!vm_map_lookup_entry(dst_map, local_end,
+ if(!vm_map_lookup_entry(dst_map, local_end,
&entry)) {
vm_map_unlock(dst_map);
return(KERN_INVALID_ADDRESS);
/*
* 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))) {
+ if ((VME_OBJECT(entry) != VM_OBJECT_NULL) &&
+ ((!VME_OBJECT(entry)->internal) ||
+ (VME_OBJECT(entry)->true_share))) {
contains_permanent_objects = TRUE;
}
} else {
copy_size = copy->size;
}
-
+
base_addr = dst_addr;
while(TRUE) {
/* deconstruct the copy object and do in parts */
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;
entry->needs_wakeup = TRUE;
vm_map_entry_wait(dst_map, THREAD_UNINT);
- if(!vm_map_lookup_entry(dst_map, base_addr,
+ if(!vm_map_lookup_entry(dst_map, base_addr,
&tmp_entry)) {
vm_map_unlock(dst_map);
return(KERN_INVALID_ADDRESS);
entry = tmp_entry;
continue;
}
- if(entry->is_sub_map) {
+ if (entry->is_sub_map) {
vm_map_offset_t sub_start;
vm_map_offset_t sub_end;
vm_map_offset_t local_end;
/* anonymous entry */
if(entry->vme_end < dst_end)
sub_end = entry->vme_end;
- else
+ else
sub_end = dst_end;
if(entry->vme_start < base_addr)
sub_start = base_addr;
- else
+ 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);
+ assert(!entry->iokit_acct);
+ entry->use_pmap = TRUE;
entry->is_sub_map = FALSE;
vm_map_deallocate(
- entry->object.sub_map);
- entry->object.sub_map = NULL;
+ VME_SUBMAP(entry));
+ VME_OBJECT_SET(entry, NULL);
+ VME_OFFSET_SET(entry, 0);
entry->is_shared = FALSE;
entry->needs_copy = FALSE;
- entry->offset = 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->protection = VM_PROT_DEFAULT;
entry->max_protection = VM_PROT_ALL;
entry->wired_count = 0;
entry->user_wired_count = 0;
- if(entry->inheritance
- == VM_INHERIT_SHARE)
+ if(entry->inheritance
+ == VM_INHERIT_SHARE)
entry->inheritance = VM_INHERIT_COPY;
continue;
}
/* entries to send */
if(base_addr < entry->vme_start) {
/* stuff to send */
- copy_size =
+ copy_size =
entry->vme_start - base_addr;
break;
}
- sub_start = entry->offset;
+ sub_start = VME_OFFSET(entry);
if(entry->vme_end < dst_end)
sub_end = entry->vme_end;
- else
+ else
sub_end = dst_end;
sub_end -= entry->vme_start;
- sub_end += entry->offset;
+ sub_end += VME_OFFSET(entry);
local_end = entry->vme_end;
vm_map_unlock(dst_map);
copy_size = sub_end - sub_start;
nentries = 1;
new_offset = copy->offset;
copy_entry = vm_map_copy_first_entry(copy);
- while(copy_entry !=
+ while(copy_entry !=
vm_map_copy_to_entry(copy)){
- entry_size = copy_entry->vme_end -
+ entry_size = copy_entry->vme_end -
copy_entry->vme_start;
if((local_size < copy_size) &&
- ((local_size + entry_size)
+ ((local_size + entry_size)
>= copy_size)) {
- vm_map_copy_clip_end(copy,
- copy_entry,
+ vm_map_copy_clip_end(copy,
+ copy_entry,
copy_entry->vme_start +
(copy_size - local_size));
- entry_size = copy_entry->vme_end -
+ 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 =
+ copy_entry->vme_next =
vm_map_copy_to_entry(copy);
- previous_prev =
+ previous_prev =
copy->cpy_hdr.links.prev;
copy->cpy_hdr.links.prev = copy_entry;
copy->size = copy_size;
- remaining_entries =
+ remaining_entries =
copy->cpy_hdr.nentries;
remaining_entries -= nentries;
copy->cpy_hdr.nentries = nentries;
copy_entry = copy_entry->vme_next;
}
}
-
+
if((entry->use_pmap) && (pmap == NULL)) {
kr = vm_map_copy_overwrite_nested(
- entry->object.sub_map,
+ VME_SUBMAP(entry),
sub_start,
copy,
- interruptible,
- entry->object.sub_map->pmap,
+ interruptible,
+ VME_SUBMAP(entry)->pmap,
TRUE);
} else if (pmap != NULL) {
kr = vm_map_copy_overwrite_nested(
- entry->object.sub_map,
+ VME_SUBMAP(entry),
sub_start,
copy,
interruptible, pmap,
TRUE);
} else {
kr = vm_map_copy_overwrite_nested(
- entry->object.sub_map,
+ VME_SUBMAP(entry),
sub_start,
copy,
interruptible,
}
if(kr != KERN_SUCCESS) {
if(next_copy != NULL) {
- copy->cpy_hdr.nentries +=
+ copy->cpy_hdr.nentries +=
remaining_entries;
- copy->cpy_hdr.links.prev->vme_next =
+ copy->cpy_hdr.links.prev->vme_next =
next_copy;
- copy->cpy_hdr.links.prev
+ copy->cpy_hdr.links.prev
= previous_prev;
copy->size = total_size;
}
}
/* 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 = vm_map_copy_allocate();
copy->type = VM_MAP_COPY_ENTRY_LIST;
copy->offset = new_offset;
copy->cpy_hdr.links.next = next_copy;
copy->cpy_hdr.links.prev = previous_prev;
copy->size = total_size;
- next_copy->vme_prev =
+ 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,
+ 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;
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 -
+ entry_size = copy_entry->vme_end -
copy_entry->vme_start;
if((local_size < copy_size) &&
- ((local_size + entry_size)
+ ((local_size + entry_size)
>= copy_size)) {
- vm_map_copy_clip_end(copy, copy_entry,
+ vm_map_copy_clip_end(copy, copy_entry,
copy_entry->vme_start +
(copy_size - local_size));
- entry_size = copy_entry->vme_end -
+ 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 =
+ copy_entry->vme_next =
vm_map_copy_to_entry(copy);
- previous_prev =
+ previous_prev =
copy->cpy_hdr.links.prev;
copy->cpy_hdr.links.prev = copy_entry;
copy->size = copy_size;
- remaining_entries =
+ remaining_entries =
copy->cpy_hdr.nentries;
remaining_entries -= nentries;
copy->cpy_hdr.nentries = nentries;
else
local_pmap = dst_map->pmap;
- if ((kr = vm_map_copy_overwrite_aligned(
+ 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 +=
+ copy->cpy_hdr.nentries +=
remaining_entries;
- copy->cpy_hdr.links.prev->vme_next =
+ copy->cpy_hdr.links.prev->vme_next =
next_copy;
- copy->cpy_hdr.links.prev =
+ copy->cpy_hdr.links.prev =
previous_prev;
copy->size += copy_size;
}
if(next_copy != NULL) {
copy->cpy_hdr.nentries +=
remaining_entries;
- copy->cpy_hdr.links.prev->vme_next =
+ copy->cpy_hdr.links.prev->vme_next =
next_copy;
- copy->cpy_hdr.links.prev =
+ copy->cpy_hdr.links.prev =
previous_prev;
copy->size += copy_size;
}
}
vm_map_lock(dst_map);
while(TRUE) {
- if (!vm_map_lookup_entry(dst_map,
+ if (!vm_map_lookup_entry(dst_map,
base_addr, &tmp_entry)) {
vm_map_unlock(dst_map);
return(KERN_INVALID_ADDRESS);
vm_map_offset_t head_addr, tail_addr;
vm_map_entry_t entry;
kern_return_t kr;
+ vm_map_offset_t effective_page_mask, effective_page_size;
head_size = 0;
tail_size = 0;
TRUE);
}
- if (copy->size < 3 * PAGE_SIZE) {
+ effective_page_mask = MAX(VM_MAP_PAGE_MASK(dst_map), PAGE_MASK);
+ effective_page_mask = MAX(VM_MAP_COPY_PAGE_MASK(copy),
+ effective_page_mask);
+ effective_page_size = effective_page_mask + 1;
+
+ if (copy->size < 3 * effective_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))) {
+ if ((dst_addr & effective_page_mask) !=
+ (copy->offset & effective_page_mask)) {
/*
* Incompatible mis-alignment of source and destination...
*/
* 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)) {
+ if (!vm_map_page_aligned(dst_addr, effective_page_mask)) {
head_addr = dst_addr;
- head_size = (VM_MAP_PAGE_SIZE(dst_map) -
- (copy->offset & VM_MAP_PAGE_MASK(dst_map)));
+ head_size = (effective_page_size -
+ (copy->offset & effective_page_mask));
+ head_size = MIN(head_size, copy->size);
}
- if (!page_aligned(copy->offset + copy->size)) {
+ if (!vm_map_page_aligned(copy->offset + copy->size,
+ effective_page_mask)) {
/*
* 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));
+ effective_page_mask);
+ tail_size = MIN(tail_size, copy->size);
tail_addr = dst_addr + copy->size - tail_size;
+ assert(tail_addr >= head_addr + head_size);
}
+ assert(head_size + tail_size <= copy->size);
if (head_size + tail_size == copy->size) {
/*
* Unaligned copy of the first "head_size" bytes, to reach
* a page boundary.
*/
-
+
/*
* Extract "head_copy" out of "copy".
*/
- head_copy = (vm_map_copy_t) zalloc(vm_map_copy_zone);
- 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 = vm_map_copy_allocate();
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);
+ entry = vm_map_copy_first_entry(copy);
+ if (entry->vme_end < copy->offset + head_size) {
+ head_size = entry->vme_end - copy->offset;
+ }
+
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,
/*
* Extract "tail_copy" out of "copy".
*/
- tail_copy = (vm_map_copy_t) zalloc(vm_map_copy_zone);
- 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 = vm_map_copy_allocate();
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);
amount_left;
kern_return_t kr = KERN_SUCCESS;
-
+
copy_entry = vm_map_copy_first_entry(copy);
vm_map_lock_write_to_read(dst_map);
/*
* 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_lock_read(dst_map);
goto RetryLookup;
}
- vm_object_shadow(&entry->object.vm_object,
- &entry->offset,
- (vm_map_size_t)(entry->vme_end
- - entry->vme_start));
+ 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 = entry->object.vm_object;
+ dst_object = VME_OBJECT(entry);
/*
* unlike with the virtual (aligned) copy we're going
* to fault on it therefore we need a target object.
}
dst_object = vm_object_allocate((vm_map_size_t)
entry->vme_end - entry->vme_start);
- entry->object.vm_object = dst_object;
- entry->offset = 0;
+ VME_OBJECT(entry) = dst_object;
+ VME_OFFSET_SET(entry, 0);
+ assert(entry->use_pmap);
vm_map_lock_write_to_read(dst_map);
}
/*
*/
vm_object_reference(dst_object);
version.main_timestamp = dst_map->timestamp;
- entry_offset = entry->offset;
+ 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(
- copy_entry->object.vm_object,
- copy_entry->offset + src_offset,
+ VME_OBJECT(copy_entry),
+ VME_OFFSET(copy_entry) + src_offset,
©_size,
dst_object,
entry_offset + dst_offset,
if (discard_on_success) {
vm_map_copy_entry_unlink(copy, copy_entry);
assert(!copy_entry->is_sub_map);
- vm_object_deallocate(
- copy_entry->object.vm_object);
+ vm_object_deallocate(VME_OBJECT(copy_entry));
vm_map_copy_entry_dispose(copy, 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;
- assert(!entry->use_pmap); /* unnested when clipped earlier */
+ 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;
*/
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;
}
* installing the source data.
*/
- object = entry->object.vm_object;
- if ((!entry->is_shared &&
- ((object == VM_OBJECT_NULL) ||
+ 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 = entry->object.vm_object;
- vm_object_offset_t old_offset = entry->offset;
+ 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 == copy_entry->object.vm_object &&
- old_offset == copy_entry->offset) {
+ 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);
continue;
}
+#if !CONFIG_EMBEDDED
#define __TRADEOFF1_OBJ_SIZE (64 * 1024 * 1024) /* 64 MB */
#define __TRADEOFF1_COPY_SIZE (128 * 1024) /* 128 KB */
- if (copy_entry->object.vm_object != VM_OBJECT_NULL &&
- copy_entry->object.vm_object->vo_size >= __TRADEOFF1_OBJ_SIZE &&
+ 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.
vm_map_copy_overwrite_aligned_src_large++;
goto slow_copy;
}
+#endif /* !CONFIG_EMBEDDED */
- if (entry->alias >= VM_MEMORY_MALLOC &&
- entry->alias <= VM_MEMORY_MALLOC_LARGE_REUSED) {
+ 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 = copy_entry->object.vm_object;
+ new_object = VME_OBJECT(copy_entry);
if (new_object != VM_OBJECT_NULL) {
vm_object_lock_shared(new_object);
}
while (new_object != VM_OBJECT_NULL &&
+#if !CONFIG_EMBEDDED
!new_object->true_share &&
new_object->copy_strategy == MEMORY_OBJECT_COPY_SYMMETRIC &&
+#endif /* !CONFIG_EMBEDDED */
new_object->internal) {
new_shadow = new_object->shadow;
if (new_shadow == VM_OBJECT_NULL) {
vm_object_unlock(new_object);
goto slow_copy;
}
+#if !CONFIG_EMBEDDED
if (new_object->true_share ||
new_object->copy_strategy != MEMORY_OBJECT_COPY_SYMMETRIC) {
/*
vm_object_unlock(new_object);
goto slow_copy;
}
+#endif /* !CONFIG_EMBEDDED */
vm_object_unlock(new_object);
}
/*
if(entry->is_sub_map) {
if(entry->use_pmap) {
#ifndef NO_NESTED_PMAP
- pmap_unnest(dst_map->pmap,
+ pmap_unnest(dst_map->pmap,
(addr64_t)entry->vme_start,
entry->vme_end - entry->vme_start);
#endif /* NO_NESTED_PMAP */
vm_map_submap_pmap_clean(
dst_map, entry->vme_start,
entry->vme_end,
- entry->object.sub_map,
- entry->offset);
+ VME_SUBMAP(entry),
+ VME_OFFSET(entry));
}
} else {
vm_map_submap_pmap_clean(
- dst_map, entry->vme_start,
+ dst_map, entry->vme_start,
entry->vme_end,
- entry->object.sub_map,
- entry->offset);
+ VME_SUBMAP(entry),
+ VME_OFFSET(entry));
}
- vm_map_deallocate(
- entry->object.sub_map);
+ vm_map_deallocate(VME_SUBMAP(entry));
} else {
if(dst_map->mapped_in_other_pmaps) {
vm_object_pmap_protect_options(
- entry->object.vm_object,
- entry->offset,
- entry->vme_end
+ VME_OBJECT(entry),
+ VME_OFFSET(entry),
+ entry->vme_end
- entry->vme_start,
PMAP_NULL,
entry->vme_start,
PMAP_OPTIONS_REMOVE);
} else {
pmap_remove_options(
- dst_map->pmap,
- (addr64_t)(entry->vme_start),
+ dst_map->pmap,
+ (addr64_t)(entry->vme_start),
(addr64_t)(entry->vme_end),
PMAP_OPTIONS_REMOVE);
}
}
}
+ if (entry->iokit_acct) {
+ /* keep using iokit accounting */
+ entry->use_pmap = FALSE;
+ } else {
+ /* use pmap accounting */
+ entry->use_pmap = TRUE;
+ }
entry->is_sub_map = FALSE;
- entry->object = copy_entry->object;
- object = entry->object.vm_object;
+ 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 = entry->offset = copy_entry->offset;
+ 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);
slow_copy:
if (entry->needs_copy) {
- vm_object_shadow(&entry->object.vm_object,
- &entry->offset,
- (entry->vme_end -
- entry->vme_start));
+ VME_OBJECT_SHADOW(entry,
+ (entry->vme_end -
+ entry->vme_start));
entry->needs_copy = FALSE;
}
- dst_object = entry->object.vm_object;
- dst_offset = entry->offset;
+ dst_object = VME_OBJECT(entry);
+ dst_offset = VME_OFFSET(entry);
/*
* Take an object reference, and record
dst_object = vm_object_allocate(
entry->vme_end - entry->vme_start);
dst_offset = 0;
- entry->object.vm_object = dst_object;
- entry->offset = dst_offset;
-
+ VME_OBJECT_SET(entry, dst_object);
+ VME_OFFSET_SET(entry, dst_offset);
+ assert(entry->use_pmap);
+
}
vm_object_reference(dst_object);
copy_size = size;
r = vm_fault_copy(
- copy_entry->object.vm_object,
- copy_entry->offset,
+ VME_OBJECT(copy_entry),
+ VME_OFFSET(copy_entry),
©_size,
dst_object,
dst_offset,
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_object_deallocate(VME_OBJECT(copy_entry));
vm_map_copy_entry_dispose(copy, copy_entry);
}
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 {
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);
}
}
vm_map_copy_t copy;
vm_size_t kalloc_size;
- if ((vm_size_t) len != len) {
- /* "len" is too big and doesn't fit in a "vm_size_t" */
- return KERN_RESOURCE_SHORTAGE;
- }
- kalloc_size = (vm_size_t) (sizeof(struct vm_map_copy) + len);
- assert((vm_map_size_t) kalloc_size == sizeof (struct vm_map_copy) + len);
+ if (len > msg_ool_size_small)
+ return KERN_INVALID_ARGUMENT;
- copy = (vm_map_copy_t) kalloc(kalloc_size);
- if (copy == VM_MAP_COPY_NULL) {
+ 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;
- copy->cpy_kdata = (void *) (copy + 1);
- copy->cpy_kalloc_size = kalloc_size;
- kr = copyinmap(src_map, src_addr, copy->cpy_kdata, (vm_size_t) len);
+ kr = copyinmap(src_map, src_addr, copy->cpy_kdata, (vm_size_t)len);
if (kr != KERN_SUCCESS) {
kfree(copy, kalloc_size);
return kr;
(void) vm_map_remove(
src_map,
vm_map_trunc_page(src_addr,
- VM_MAP_PAGE_MASK(src_map)),
+ 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));
+ ((src_map == kernel_map) ? VM_MAP_REMOVE_KUNWIRE : VM_MAP_REMOVE_NO_FLAGS)));
}
*copy_result = copy;
return KERN_SUCCESS;
vm_map_t map,
vm_map_address_t *addr, /* IN/OUT */
vm_map_copy_t copy,
+ vm_map_size_t copy_size,
boolean_t overwrite,
boolean_t consume_on_success)
{
kern_return_t kr = KERN_SUCCESS;
thread_t thread = current_thread();
+ assert(copy->size == copy_size);
+
+ /*
+ * 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,
+ kr = vm_map_enter(map,
+ addr,
+ vm_map_round_page(copy_size,
VM_MAP_PAGE_MASK(map)),
- (vm_map_offset_t) 0,
+ (vm_map_offset_t) 0,
VM_FLAGS_ANYWHERE,
- VM_OBJECT_NULL,
- (vm_object_offset_t) 0,
+ VM_MAP_KERNEL_FLAGS_NONE,
+ VM_KERN_MEMORY_NONE,
+ VM_OBJECT_NULL,
+ (vm_object_offset_t) 0,
FALSE,
- VM_PROT_DEFAULT,
+ VM_PROT_DEFAULT,
VM_PROT_ALL,
VM_INHERIT_DEFAULT);
if (kr != KERN_SUCCESS)
return kr;
+#if KASAN
+ if (map->pmap == kernel_pmap) {
+ kasan_notify_address(*addr, copy->size);
+ }
+#endif
}
/*
* 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)) {
+ assert((vm_size_t)copy_size == copy_size);
+ if (copyout(copy->cpy_kdata, *addr, (vm_size_t)copy_size)) {
kr = KERN_INVALID_ADDRESS;
}
}
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)) {
+ 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);
}
vm_map_trunc_page(*addr,
VM_MAP_PAGE_MASK(map)),
vm_map_round_page((*addr +
- vm_map_round_page(copy->size,
+ vm_map_round_page(copy_size,
VM_MAP_PAGE_MASK(map))),
VM_MAP_PAGE_MASK(map)),
- VM_MAP_NO_FLAGS);
+ VM_MAP_REMOVE_NO_FLAGS);
*addr = 0;
}
} else {
/* copy was successful, dicard the copy structure */
if (consume_on_success) {
- kfree(copy, copy->cpy_kalloc_size);
+ kfree(copy, copy_size + cpy_kdata_hdr_sz);
}
}
return kr;
}
-
+
/*
- * Macro: vm_map_copy_insert
- *
+ * Routine: vm_map_copy_insert [internal use only]
+ *
* 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
+static void
+vm_map_copy_insert(
+ vm_map_t map,
+ vm_map_entry_t after_where,
+ vm_map_copy_t copy)
+{
+ vm_map_entry_t entry;
+
+ while (vm_map_copy_first_entry(copy) != vm_map_copy_to_entry(copy)) {
+ entry = vm_map_copy_first_entry(copy);
+ vm_map_copy_entry_unlink(copy, entry);
+ vm_map_store_entry_link(map, after_where, entry,
+ VM_MAP_KERNEL_FLAGS_NONE);
+ after_where = entry;
+ }
+ zfree(vm_map_copy_zone, copy);
+}
void
vm_map_copy_remap(
new_entry->behavior = VM_BEHAVIOR_DEFAULT;
/* take an extra reference on the entry's "object" */
if (new_entry->is_sub_map) {
- vm_map_lock(new_entry->object.sub_map);
- vm_map_reference(new_entry->object.sub_map);
- vm_map_unlock(new_entry->object.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(new_entry->object.vm_object);
+ vm_object_reference(VME_OBJECT(new_entry));
}
/* insert the new entry in the map */
- vm_map_store_entry_link(map, where, new_entry);
+ vm_map_store_entry_link(map, where, new_entry,
+ VM_MAP_KERNEL_FLAGS_NONE);
/* continue inserting the "copy entries" after the new entry */
where = new_entry;
}
}
+
+/*
+ * Returns true if *size matches (or is in the range of) copy->size.
+ * Upon returning true, the *size field is updated with the actual size of the
+ * copy object (may be different for VM_MAP_COPY_ENTRY_LIST types)
+ */
+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;
+ vm_map_size_t copy_sz = copy->size;
+ vm_map_size_t sz = *size;
+ switch (copy->type) {
+ case VM_MAP_COPY_OBJECT:
+ case VM_MAP_COPY_KERNEL_BUFFER:
+ if (sz == copy_sz)
+ 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_sz >= sz &&
+ copy_sz <= vm_map_round_page(sz, VM_MAP_PAGE_MASK(dst_map))) {
+ *size = copy_sz;
+ return TRUE;
+ }
+ break;
+ default:
+ break;
+ }
+ return FALSE;
+}
+
+/*
+ * Routine: vm_map_copyout_size
+ *
+ * Description:
+ * Copy out a copy chain ("copy") into newly-allocated
+ * space in the destination map. Uses a prevalidated
+ * size for the copy object (vm_map_copy_validate_size).
+ *
+ * If successful, consumes the copy object.
+ * Otherwise, the caller is responsible for it.
+ */
+kern_return_t
+vm_map_copyout_size(
+ vm_map_t dst_map,
+ vm_map_address_t *dst_addr, /* OUT */
+ vm_map_copy_t copy,
+ vm_map_size_t copy_size)
+{
+ return vm_map_copyout_internal(dst_map, dst_addr, copy, copy_size,
+ TRUE, /* consume_on_success */
+ VM_PROT_DEFAULT,
+ VM_PROT_ALL,
+ VM_INHERIT_DEFAULT);
+}
+
/*
* Routine: vm_map_copyout
*
* 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);
+ return vm_map_copyout_internal(dst_map, dst_addr, copy, copy ? copy->size : 0,
+ TRUE, /* consume_on_success */
+ VM_PROT_DEFAULT,
+ VM_PROT_ALL,
+ VM_INHERIT_DEFAULT);
}
kern_return_t
vm_map_t dst_map,
vm_map_address_t *dst_addr, /* OUT */
vm_map_copy_t copy,
+ vm_map_size_t copy_size,
boolean_t consume_on_success,
vm_prot_t cur_protection,
vm_prot_t max_protection,
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.
return(KERN_SUCCESS);
}
+ if (copy->size != copy_size) {
+ *dst_addr = 0;
+ return KERN_FAILURE;
+ }
+
/*
* Check for special copy object, created
* by vm_map_copyin_object.
vm_object_offset_t offset;
offset = vm_object_trunc_page(copy->offset);
- size = vm_map_round_page((copy->size +
+ 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,
+ VM_MAP_KERNEL_FLAGS_NONE,
+ VM_KERN_MEMORY_NONE,
object, offset, FALSE,
VM_PROT_DEFAULT, VM_PROT_ALL,
VM_INHERIT_DEFAULT);
*/
if (copy->type == VM_MAP_COPY_KERNEL_BUFFER) {
- return vm_map_copyout_kernel_buffer(dst_map, dst_addr,
- copy, FALSE,
+ return vm_map_copyout_kernel_buffer(dst_map, dst_addr,
+ copy, copy_size, FALSE,
consume_on_success);
}
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,
+ size = vm_map_round_page((vm_map_size_t)copy->offset + copy_size,
VM_MAP_COPY_PAGE_MASK(copy))
- vm_copy_start;
VM_MAP_HIGHEST_ENTRY(dst_map, entry, start);
last = entry;
} 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;
+ if (dst_map->holelistenabled) {
+ hole_entry = CAST_TO_VM_MAP_ENTRY(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));
}
return(KERN_NO_SPACE);
}
- if ((next == vm_map_to_entry(dst_map)) ||
- (next->vme_start >= end))
- break;
+ 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;
- start = last->vme_end;
+
+ if (dst_map->holelistenabled) {
+ if (last == CAST_TO_VM_MAP_ENTRY(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) {
/*
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);
- new->use_pmap = FALSE; /* clr address space specifics */
+ 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);
* map the pages into the destination map.
*/
if (entry->wired_count != 0) {
- register vm_map_offset_t va;
+ vm_map_offset_t va;
vm_object_offset_t offset;
- register vm_object_t object;
+ vm_object_t object;
vm_prot_t prot;
int type_of_fault;
- object = entry->object.vm_object;
- offset = entry->offset;
+ object = VME_OBJECT(entry);
+ offset = VME_OFFSET(entry);
va = entry->vme_start;
pmap_pageable(dst_map->pmap,
TRUE);
while (va < entry->vme_end) {
- register vm_page_t m;
+ vm_page_t m;
+ struct vm_object_fault_info fault_info = {};
/*
* Look up the page in the object.
m = vm_page_lookup(object, offset);
if (m == VM_PAGE_NULL || !VM_PAGE_WIRED(m) ||
- m->absent)
+ m->vmp_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, entry->alias) && prot)
+ 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, NULL,
+ 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;
+ }
+
+ vm_fault_enter(m,
+ dst_map->pmap,
+ va,
+ prot,
+ prot,
+ VM_PAGE_WIRED(m),
+ FALSE, /* change_wiring */
+ VM_KERN_MEMORY_NONE, /* tag - not wiring */
+ &fault_info,
+ NULL, /* need_retry */
&type_of_fault);
vm_object_unlock(object);
*dst_addr = start + (copy->offset - vm_copy_start);
+#if KASAN
+ kasan_notify_address(*dst_addr, size);
+#endif
+
/*
* Update the hints and the map size
*/
__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.
*/
- register
vm_map_entry_t new_entry = VM_MAP_ENTRY_NULL; /* Map entry for copy */
vm_map_offset_t src_start; /* Start of current entry --
boolean_t map_share=FALSE;
submap_map_t *parent_maps = NULL;
- register
vm_map_copy_t copy; /* Resulting copy */
- vm_map_address_t copy_addr;
+ vm_map_address_t copy_addr;
+ vm_map_size_t copy_size;
+ boolean_t src_destroy;
+ boolean_t use_maxprot;
+ boolean_t preserve_purgeable;
+ boolean_t entry_was_shared;
+ vm_map_entry_t saved_src_entry;
+
+ 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;
+ preserve_purgeable =
+ (flags & VM_MAP_COPYIN_PRESERVE_PURGEABLE) ? TRUE : FALSE;
/*
* Check for copies of zero bytes.
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,
+ VM_MAP_PAGE_MASK(src_map));
+ src_end = vm_map_round_page(src_end,
+ VM_MAP_PAGE_MASK(src_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)
+ if ((len < msg_ool_size_small) &&
+ !use_maxprot &&
+ !preserve_purgeable &&
+ !(flags & VM_MAP_COPYIN_ENTRY_LIST) &&
+ /*
+ * Since the "msg_ool_size_small" threshold was increased and
+ * vm_map_copyin_kernel_buffer() doesn't handle accesses beyond the
+ * address space limits, we revert to doing a virtual copy if the
+ * copied range goes beyond those limits. Otherwise, mach_vm_read()
+ * of the commpage would now fail when it used to work.
+ */
+ (src_start >= vm_map_min(src_map) &&
+ src_start < vm_map_max(src_map) &&
+ src_end >= vm_map_min(src_map) &&
+ src_end < vm_map_max(src_map)))
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
+ * 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 = vm_map_copy_allocate();
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;
copy->offset = src_addr;
copy->size = len;
-
+
new_entry = vm_map_copy_entry_create(copy, !copy->cpy_hdr.entries_pageable);
#define RETURN(x) \
vm_map_lock(src_map);
- if (!vm_map_lookup_entry(src_map, src_start, &tmp_entry))
+ /*
+ * 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;
*/
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;
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 += tmp_entry->offset;
+ src_start += VME_OFFSET(tmp_entry);
src_end = src_start + submap_len;
- src_map = tmp_entry->object.sub_map;
+ 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 */
}
/* we are now in the lowest level submap... */
- if ((tmp_entry->object.vm_object != VM_OBJECT_NULL) &&
- (tmp_entry->object.vm_object->phys_contiguous)) {
+ 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 */
RETURN(KERN_PROTECTION_FAILURE);
}
/*
- * Create a new address map entry to hold the result.
+ * 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.
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;
+ 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);
- new_entry->use_pmap = FALSE; /* clr address space specifics */
+ if (new_entry->is_sub_map) {
+ /* clr address space specifics */
+ new_entry->use_pmap = FALSE;
+ } else {
+ /*
+ * We're dealing with a copy-on-write operation,
+ * so the resulting mapping should not inherit the
+ * original mapping's accounting settings.
+ * "iokit_acct" should have been cleared in
+ * vm_map_entry_copy().
+ * "use_pmap" should be reset to its default (TRUE)
+ * so that the new mapping gets accounted for in
+ * the task's memory footprint.
+ */
+ assert(!new_entry->iokit_acct);
+ new_entry->use_pmap = TRUE;
+ }
/*
* 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 (src_destroy &&
+ (src_object == VM_OBJECT_NULL ||
+ (src_object->internal &&
+ src_object->copy_strategy == MEMORY_OBJECT_COPY_SYMMETRIC &&
+ src_entry->vme_start <= src_addr &&
+ src_entry->vme_end >= src_end &&
+ !map_share))) {
/*
* If we are destroying the source, and the object
* is internal, we can move the object reference
* copy-on-write only if the source is.
* We make another reference to the object, because
* destroying the source entry will deallocate it.
+ *
+ * This memory transfer has to be atomic (to prevent
+ * the VM object from being shared or copied while
+ * it's being moved here), so we can only do this
+ * if we won't have to unlock the VM map, i.e. the
+ * entire range must be covered by this map entry.
*/
vm_object_reference(src_object);
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,
+ 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(
- &new_entry->object.vm_object,
+ &VME_OBJECT(new_entry),
src_offset,
src_size,
&src_needs_copy,
prot = src_entry->protection & ~VM_PROT_WRITE;
- if (override_nx(src_map, src_entry->alias) && prot)
+ 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 ?
+ (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;
}
goto CopySuccessful;
}
+ entry_was_shared = tmp_entry->is_shared;
+
/*
* Take an object reference, so that we may
* release the map lock(s).
version.main_timestamp = src_map->timestamp;
vm_map_unlock(src_map); /* Increments timestamp once! */
+ saved_src_entry = src_entry;
+ tmp_entry = VM_MAP_ENTRY_NULL;
+ src_entry = VM_MAP_ENTRY_NULL;
/*
* Perform the copy
src_offset,
src_size,
THREAD_UNINT,
- &new_entry->object.vm_object);
- new_entry->offset = 0;
+ &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)) {
+ (entry_was_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_offset,
src_size,
TRUE);
if (new_object == VM_OBJECT_NULL)
goto CopySlowly;
- new_entry->object.vm_object = new_object;
+ 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);
+ assertf(new_entry->use_pmap, "src_map %p new_entry %p\n", src_map, new_entry);
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,
- &new_entry->object.vm_object,
- &new_entry->offset,
+ &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 &&
+ preserve_purgeable &&
+ src_object->purgable != VM_PURGABLE_DENY) {
+ vm_object_t new_object;
+
+ new_object = VME_OBJECT(new_entry);
+ assert(new_object != src_object);
+ vm_object_lock(new_object);
+ assert(new_object->ref_count == 1);
+ assert(new_object->shadow == VM_OBJECT_NULL);
+ assert(new_object->copy == VM_OBJECT_NULL);
+ assert(new_object->vo_owner == NULL);
+
+ new_object->copy_strategy = MEMORY_OBJECT_COPY_NONE;
+ new_object->true_share = TRUE;
+ /* start as non-volatile with no owner... */
+ new_object->purgable = VM_PURGABLE_NONVOLATILE;
+ vm_purgeable_nonvolatile_enqueue(new_object, NULL);
+ /* ... and move to src_object's purgeable state */
+ if (src_object->purgable != VM_PURGABLE_NONVOLATILE) {
+ int state;
+ state = src_object->purgable;
+ vm_object_purgable_control(
+ new_object,
+ VM_PURGABLE_SET_STATE_FROM_KERNEL,
+ &state);
+ }
+ vm_object_unlock(new_object);
+ new_object = VM_OBJECT_NULL;
+ /* no pmap accounting for purgeable objects */
+ new_entry->use_pmap = FALSE;
+ }
+
if (result != KERN_SUCCESS &&
result != KERN_MEMORY_RESTART_COPY) {
vm_map_lock(src_map);
vm_map_lock(src_map);
- if ((version.main_timestamp + 1) == src_map->timestamp)
+ if ((version.main_timestamp + 1) == src_map->timestamp) {
+ /* src_map hasn't changed: src_entry is still valid */
+ src_entry = saved_src_entry;
goto VerificationSuccessful;
+ }
/*
* Simple version comparison failed.
*/
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);
+ /* reset accounting state */
+ new_entry->iokit_acct = FALSE;
+ new_entry->use_pmap = TRUE;
+ }
RETURN(KERN_INVALID_ADDRESS);
}
goto VerificationFailed;
if (src_entry->vme_end < new_entry->vme_end) {
+ /*
+ * This entry might have been shortened
+ * (vm_map_clip_end) or been replaced with
+ * an entry that ends closer to "src_start"
+ * than before.
+ * Adjust "new_entry" accordingly; copying
+ * less memory would be correct but we also
+ * redo the copy (see below) if the new entry
+ * no longer points at the same object/offset.
+ */
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;
+ } else if (src_entry->vme_end > new_entry->vme_end) {
+ /*
+ * This entry might have been extended
+ * (vm_map_entry_simplify() or coalesce)
+ * or been replaced with an entry that ends farther
+ * from "src_start" than before.
+ *
+ * We've called vm_object_copy_*() only on
+ * the previous <start:end> range, so we can't
+ * just extend new_entry. We have to re-do
+ * the copy based on the new entry as if it was
+ * pointing at a different object/offset (see
+ * "Verification failed" below).
+ */
}
- if ((src_entry->object.vm_object != src_object) ||
- (src_entry->offset != src_offset) ) {
+ if ((VME_OBJECT(src_entry) != src_object) ||
+ (VME_OFFSET(src_entry) != src_offset) ||
+ (src_entry->vme_end > new_entry->vme_end)) {
/*
* Verification failed.
VerificationFailed: ;
- vm_object_deallocate(new_entry->object.vm_object);
+ vm_object_deallocate(VME_OBJECT(new_entry));
tmp_entry = src_entry;
continue;
}
vm_map_copy_entry_link(copy, vm_map_copy_last_entry(copy),
new_entry);
-
+
/*
* Determine whether the entire region
* has been copied.
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;
-
- /* 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 ((src_end > src_start) &&
- !vm_map_lookup_entry(
- src_map, src_start, &tmp_entry))
- RETURN(KERN_INVALID_ADDRESS);
- kfree(ptr, sizeof(submap_map_t));
- if(parent_maps == NULL)
- map_share = FALSE;
- src_entry = tmp_entry->vme_prev;
- } else
+ 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;
*/
tmp_entry = src_entry->vme_next;
- if ((tmp_entry->vme_start != src_start) ||
+ if ((tmp_entry->vme_start != src_start) ||
(tmp_entry == vm_map_to_entry(src_map))) {
-
- if (VM_MAP_PAGE_SHIFT(src_map) != PAGE_SHIFT &&
- (vm_map_round_page(src_entry->vme_end,
- VM_MAP_PAGE_MASK(src_map)) ==
- src_end)) {
- vm_map_entry_t last_copy_entry;
- vm_map_offset_t adjustment;
-
- /*
- * This is the last entry in the range we
- * want and it happens to miss a few pages
- * because it is not map-aligned (must have
- * been imported from a differently-aligned
- * map).
- * Let's say we're done, but first we have
- * to compensate for the alignment adjustment
- * we're about to do before returning.
- */
-
- last_copy_entry = vm_map_copy_last_entry(copy);
- assert(last_copy_entry !=
- vm_map_copy_to_entry(copy));
- 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));
- last_copy_entry->vme_end += adjustment;
- last_copy_entry->map_aligned = FALSE;
- /* ... and we're done */
- break;
- }
-
RETURN(KERN_INVALID_ADDRESS);
}
}
/*
* If the source should be destroyed, do it now, since the
- * copy was successful.
+ * copy was successful.
*/
if (src_destroy) {
(void) vm_map_delete(
src_end,
((src_map == kernel_map) ?
VM_MAP_REMOVE_KUNWIRE :
- VM_MAP_NO_FLAGS),
+ VM_MAP_REMOVE_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_PAGE_MASK(src_map)),
vm_map_round_page(src_end,
VM_MAP_PAGE_MASK(src_map)));
}
vm_map_unlock(src_map);
+ tmp_entry = VM_MAP_ENTRY_NULL;
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) -
assert(page_aligned(adjustment));
assert(adjustment < VM_MAP_PAGE_SIZE(src_map));
tmp_entry->vme_start += adjustment;
- tmp_entry->offset += 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),
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)));
}
}
/* 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->map_aligned = FALSE;
- tmp_entry->vme_end = copy_addr +
+ 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);
/*
* Allocate a header element for the list.
*
- * Use the start and end in the header to
+ * 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 = vm_map_copy_allocate();
copy->type = VM_MAP_COPY_ENTRY_LIST;
- copy->cpy_hdr.nentries = 0;
copy->cpy_hdr.entries_pageable = TRUE;
vm_map_store_init(©->cpy_hdr);
cur_prot,
max_prot,
VM_INHERIT_SHARE,
- TRUE); /* pageable */
+ TRUE, /* pageable */
+ FALSE, /* same_map */
+ VM_MAP_KERNEL_FLAGS_NONE);
if (kr != KERN_SUCCESS) {
vm_map_copy_discard(copy);
return kr;
* that contains the object directly.
*/
- copy = (vm_map_copy_t) zalloc(vm_map_copy_zone);
+ copy = vm_map_copy_allocate();
copy->type = VM_MAP_COPY_OBJECT;
copy->cpy_object = object;
copy->offset = offset;
* or someone else already has one, then
* make a new shadow and share it.
*/
-
- object = old_entry->object.vm_object;
+
+ 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,
- (old_entry->object.sub_map)->pmap,
+ 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));
} else if (object == VM_OBJECT_NULL) {
object = vm_object_allocate((vm_map_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);
+ 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->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.
* (This is a preemptive version of
* case 2.)
*/
- vm_object_shadow(&old_entry->object.vm_object,
- &old_entry->offset,
- (vm_map_size_t) (old_entry->vme_end -
- old_entry->vme_start));
-
+ 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
(old_entry->protection & VM_PROT_WRITE)) {
vm_prot_t prot;
+ assert(!pmap_has_prot_policy(old_entry->protection));
+
prot = old_entry->protection & ~VM_PROT_WRITE;
- if (override_nx(old_map, old_entry->alias) && prot)
+ assert(!pmap_has_prot_policy(prot));
+
+ 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(
- old_entry->object.vm_object,
- old_entry->offset,
+ VME_OBJECT(old_entry),
+ VME_OFFSET(old_entry),
(old_entry->vme_end -
old_entry->vme_start),
PMAP_NULL,
prot);
}
}
-
+
old_entry->needs_copy = FALSE;
- object = old_entry->object.vm_object;
+ object = VME_OBJECT(old_entry);
}
-
+
/*
* If object was using a symmetric copy strategy,
* change its copy strategy to the default
* 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);
+ 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);
}
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;
-
+
+ /*
+ * We're dealing with a shared mapping, so the resulting mapping
+ * should inherit some of the original mapping's accounting settings.
+ * "iokit_acct" should have been cleared in vm_map_entry_copy().
+ * "use_pmap" should stay the same as before (if it hasn't been reset
+ * to TRUE when we cleared "iokit_acct").
+ */
+ assert(!new_entry->iokit_acct);
+
+ /*
+ * If old entry's inheritence is VM_INHERIT_NONE,
+ * the new entry is for corpse fork, remove the
+ * write permission from the new entry.
+ */
+ if (old_entry->inheritance == VM_INHERIT_NONE) {
+
+ new_entry->protection &= ~VM_PROT_WRITE;
+ new_entry->max_protection &= ~VM_PROT_WRITE;
+ }
+
/*
* 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);
-
+
+ vm_map_store_entry_link(new_map, vm_map_last_entry(new_map), new_entry,
+ VM_MAP_KERNEL_FLAGS_NONE);
+
/*
* Update the physical map
*/
-
+
if (old_entry->is_sub_map) {
/* Bill Angell pmap support goes here */
} else {
vm_map_fork_copy(
vm_map_t old_map,
vm_map_entry_t *old_entry_p,
- vm_map_t new_map)
+ vm_map_t new_map,
+ int vm_map_copyin_flags)
{
vm_map_entry_t old_entry = *old_entry_p;
vm_map_size_t entry_size = old_entry->vme_end - old_entry->vme_start;
* be accessed, not just whether it's accessible
* right now.
*/
- if (vm_map_copyin_maxprot(old_map, start, entry_size, FALSE, ©)
+ vm_map_copyin_flags |= VM_MAP_COPYIN_USE_MAXPROT;
+ if (vm_map_copyin_internal(old_map, start, entry_size,
+ vm_map_copyin_flags, ©)
!= KERN_SUCCESS) {
/*
* The map might have changed while it
* 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)) {
*
* Create and return a new map based on the old
* map, according to the inheritance values on the
- * regions in that map.
+ * regions in that map and the options.
*
* The source map must not be locked.
*/
vm_map_t
vm_map_fork(
ledger_t ledger,
- vm_map_t old_map)
+ vm_map_t old_map,
+ int options)
{
pmap_t new_pmap;
vm_map_t new_map;
vm_map_entry_t new_entry;
boolean_t src_needs_copy;
boolean_t new_entry_needs_copy;
+ boolean_t pmap_is64bit;
+ int vm_map_copyin_flags;
+ vm_inherit_t old_entry_inheritance;
+ int map_create_options;
+ kern_return_t footprint_collect_kr;
+
+ if (options & ~(VM_MAP_FORK_SHARE_IF_INHERIT_NONE |
+ VM_MAP_FORK_PRESERVE_PURGEABLE |
+ VM_MAP_FORK_CORPSE_FOOTPRINT)) {
+ /* unsupported option */
+ return VM_MAP_NULL;
+ }
- new_pmap = pmap_create(ledger, (vm_map_size_t) 0,
+ pmap_is64bit =
#if defined(__i386__) || defined(__x86_64__)
- old_map->pmap->pm_task_map != TASK_MAP_32BIT
+ old_map->pmap->pm_task_map != TASK_MAP_32BIT;
+#elif defined(__arm64__)
+ old_map->pmap->max == MACH_VM_MAX_ADDRESS;
+#elif defined(__arm__)
+ FALSE;
#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);
+ map_create_options = 0;
+ if (old_map->hdr.entries_pageable) {
+ map_create_options |= VM_MAP_CREATE_PAGEABLE;
+ }
+ if (options & VM_MAP_FORK_CORPSE_FOOTPRINT) {
+ map_create_options |= VM_MAP_CREATE_CORPSE_FOOTPRINT;
+ footprint_collect_kr = KERN_SUCCESS;
+ }
+ new_map = vm_map_create_options(new_pmap,
+ old_map->min_offset,
+ old_map->max_offset,
+ map_create_options);
+ vm_map_lock(new_map);
+ vm_commit_pagezero_status(new_map);
/* inherit the parent map's page size */
vm_map_set_page_shift(new_map, VM_MAP_PAGE_SHIFT(old_map));
for (
entry_size = old_entry->vme_end - old_entry->vme_start;
- switch (old_entry->inheritance) {
+ old_entry_inheritance = old_entry->inheritance;
+ /*
+ * If caller used the VM_MAP_FORK_SHARE_IF_INHERIT_NONE option
+ * share VM_INHERIT_NONE entries that are not backed by a
+ * device pager.
+ */
+ if (old_entry_inheritance == VM_INHERIT_NONE &&
+ (options & VM_MAP_FORK_SHARE_IF_INHERIT_NONE) &&
+ !(!old_entry->is_sub_map &&
+ VME_OBJECT(old_entry) != NULL &&
+ VME_OBJECT(old_entry)->pager != NULL &&
+ is_device_pager_ops(
+ VME_OBJECT(old_entry)->pager->mo_pager_ops))) {
+ old_entry_inheritance = VM_INHERIT_SHARE;
+ }
+
+ if (old_entry_inheritance != VM_INHERIT_NONE &&
+ (options & VM_MAP_FORK_CORPSE_FOOTPRINT) &&
+ footprint_collect_kr == KERN_SUCCESS) {
+ /*
+ * The corpse won't have old_map->pmap to query
+ * footprint information, so collect that data now
+ * and store it in new_map->vmmap_corpse_footprint
+ * for later autopsy.
+ */
+ footprint_collect_kr =
+ vm_map_corpse_footprint_collect(old_map,
+ old_entry,
+ new_map);
+ }
+
+ switch (old_entry_inheritance) {
case VM_INHERIT_NONE:
break;
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))) {
+ ((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);
- /* clear address space specifics */
- new_entry->use_pmap = FALSE;
+ if (new_entry->is_sub_map) {
+ /* clear address space specifics */
+ new_entry->use_pmap = FALSE;
+ } else {
+ /*
+ * We're dealing with a copy-on-write operation,
+ * so the resulting mapping should not inherit
+ * the original mapping's accounting settings.
+ * "iokit_acct" should have been cleared in
+ * vm_map_entry_copy().
+ * "use_pmap" should be reset to its default
+ * (TRUE) so that the new mapping gets
+ * accounted for in the task's memory footprint.
+ */
+ assert(!new_entry->iokit_acct);
+ new_entry->use_pmap = TRUE;
+ }
if (! vm_object_copy_quickly(
- &new_entry->object.vm_object,
- old_entry->offset,
+ &VME_OBJECT(new_entry),
+ VME_OFFSET(old_entry),
(old_entry->vme_end -
old_entry->vme_start),
&src_needs_copy,
/*
* Handle copy-on-write obligations
*/
-
+
if (src_needs_copy && !old_entry->needs_copy) {
vm_prot_t prot;
+ assert(!pmap_has_prot_policy(old_entry->protection));
+
prot = old_entry->protection & ~VM_PROT_WRITE;
- if (override_nx(old_map, old_entry->alias) && prot)
+ if (override_nx(old_map, VME_ALIAS(old_entry))
+ && prot)
prot |= VM_PROT_EXECUTE;
+ assert(!pmap_has_prot_policy(prot));
+
vm_object_pmap_protect(
- old_entry->object.vm_object,
- old_entry->offset,
+ VME_OBJECT(old_entry),
+ VME_OFFSET(old_entry),
(old_entry->vme_end -
old_entry->vme_start),
- ((old_entry->is_shared
+ ((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);
+
+ vm_map_store_entry_link(new_map,
+ vm_map_last_entry(new_map),
+ new_entry,
+ VM_MAP_KERNEL_FLAGS_NONE);
new_size += entry_size;
break;
slow_vm_map_fork_copy:
- if (vm_map_fork_copy(old_map, &old_entry, new_map)) {
+ vm_map_copyin_flags = 0;
+ if (options & VM_MAP_FORK_PRESERVE_PURGEABLE) {
+ vm_map_copyin_flags |=
+ VM_MAP_COPYIN_PRESERVE_PURGEABLE;
+ }
+ if (vm_map_fork_copy(old_map,
+ &old_entry,
+ new_map,
+ vm_map_copyin_flags)) {
new_size += entry_size;
}
continue;
old_entry = old_entry->vme_next;
}
+#if defined(__arm64__)
+ pmap_insert_sharedpage(new_map->pmap);
+#endif
+
new_map->size = new_size;
+
+ if (options & VM_MAP_FORK_CORPSE_FOOTPRINT) {
+ vm_map_corpse_footprint_collect_done(new_map);
+ }
+
+ vm_map_unlock(new_map);
vm_map_unlock(old_map);
vm_map_deallocate(old_map);
vm_map_exec(
vm_map_t new_map,
task_t task,
+ boolean_t is64bit,
void *fsroot,
- cpu_type_t cpu)
+ cpu_type_t cpu,
+ cpu_subtype_t cpu_subtype)
{
SHARED_REGION_TRACE_DEBUG(
- ("shared_region: task %p: vm_map_exec(%p,%p,%p,0x%x): ->\n",
- current_task(), new_map, task, fsroot, cpu));
- (void) vm_commpage_enter(new_map, task);
- (void) vm_shared_region_enter(new_map, task, fsroot, cpu);
+ ("shared_region: task %p: vm_map_exec(%p,%p,%p,0x%x,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,
+ cpu_subtype));
+ (void) vm_commpage_enter(new_map, task, is64bit);
+ (void) vm_shared_region_enter(new_map, task, is64bit, fsroot, cpu, cpu_subtype);
SHARED_REGION_TRACE_DEBUG(
- ("shared_region: task %p: vm_map_exec(%p,%p,%p,0x%x): <-\n",
- current_task(), new_map, task, fsroot, cpu));
+ ("shared_region: task %p: vm_map_exec(%p,%p,%p,0x%x,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,
+ cpu_subtype));
return KERN_SUCCESS;
}
vm_map_t *real_map)
{
vm_map_entry_t entry;
- register vm_map_t map = *var_map;
+ 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;
+ vm_prot_t prot;
boolean_t mask_protections;
+ boolean_t force_copy;
vm_prot_t original_fault_type;
/*
* absolute value.
*/
mask_protections = (fault_type & VM_PROT_IS_MASK) ? TRUE : FALSE;
- fault_type &= ~VM_PROT_IS_MASK;
+ force_copy = (fault_type & VM_PROT_COPY) ? TRUE : FALSE;
+ fault_type &= VM_PROT_ALL;
original_fault_type = fault_type;
*real_map = map;
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)
+ if((*real_map != map)
&& (*real_map != cow_sub_map_parent))
vm_map_unlock(*real_map);
return KERN_INVALID_ADDRESS;
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_offset_t start_delta;
vm_map_entry_t submap_entry;
+ vm_prot_t subentry_protection;
+ vm_prot_t subentry_max_protection;
boolean_t mapped_needs_copy=FALSE;
local_vaddr = vaddr;
- if ((entry->use_pmap && !(fault_type & VM_PROT_WRITE))) {
+ if ((entry->use_pmap &&
+ ! ((fault_type & VM_PROT_WRITE) ||
+ force_copy))) {
/* if real_map equals map we unlock below */
- if ((*real_map != map) &&
+ if ((*real_map != map) &&
(*real_map != cow_sub_map_parent))
vm_map_unlock(*real_map);
- *real_map = entry->object.sub_map;
+ *real_map = VME_SUBMAP(entry);
}
- if(entry->needs_copy && (fault_type & VM_PROT_WRITE)) {
+ if(entry->needs_copy &&
+ ((fault_type & VM_PROT_WRITE) ||
+ force_copy)) {
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(entry->object.sub_map);
- *var_map = entry->object.sub_map;
+ 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 */
cow_parent_vaddr = vaddr;
mapped_needs_copy = TRUE;
} else {
- vm_map_lock_read(entry->object.sub_map);
- *var_map = entry->object.sub_map;
+ 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(entry->object.sub_map);
- *var_map = entry->object.sub_map;
+ 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 */
map = *var_map;
/* calculate the offset in the submap for vaddr */
- local_vaddr = (local_vaddr - entry->vme_start) + entry->offset;
+ 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)
+ if((*real_map != map)
&& (*real_map != cow_sub_map_parent)) {
vm_map_unlock(*real_map);
}
/* 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;
+ start_delta = submap_entry->vme_start > VME_OFFSET(entry) ?
+ submap_entry->vme_start - VME_OFFSET(entry) : 0;
- end_delta =
- (entry->offset + start_delta + (old_end - old_start)) <=
+ end_delta =
+ (VME_OFFSET(entry) + start_delta + (old_end - old_start)) <=
submap_entry->vme_end ?
- 0 : (entry->offset +
+ 0 : (VME_OFFSET(entry) +
(old_end - old_start))
- - submap_entry->vme_end;
+ - submap_entry->vme_end;
old_start += start_delta;
old_end -= end_delta;
goto submap_recurse;
}
- if(((fault_type & VM_PROT_WRITE) && cow_sub_map_parent)) {
+ if (((fault_type & VM_PROT_WRITE) ||
+ force_copy)
+ && cow_sub_map_parent) {
vm_object_t sub_object, copy_object;
vm_object_offset_t copy_offset;
}
- sub_object = submap_entry->object.vm_object;
+ 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));
- submap_entry->object.vm_object = sub_object;
- submap_entry->offset = 0;
+ VME_OBJECT_SET(submap_entry, sub_object);
+ VME_OFFSET_SET(submap_entry, 0);
+ assert(!submap_entry->is_sub_map);
+ assert(submap_entry->use_pmap);
}
- local_start = local_vaddr -
+ local_start = local_vaddr -
(cow_parent_vaddr - old_start);
- local_end = local_vaddr +
+ 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);
- /* unnesting was done in vm_map_clip_start/end() */
- assert(!submap_entry->use_pmap);
+ 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 */
MEMORY_OBJECT_COPY_NONE)) {
vm_object_lock(sub_object);
vm_object_copy_slowly(sub_object,
- submap_entry->offset,
+ 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);
+ vm_object_lock(sub_object);
+ vm_object_reference_locked(sub_object);
sub_object->shadowed = TRUE;
+ vm_object_unlock(sub_object);
+
+ assert(submap_entry->wired_count == 0);
submap_entry->needs_copy = TRUE;
- prot = submap_entry->protection & ~VM_PROT_WRITE;
+ prot = submap_entry->protection;
+ assert(!pmap_has_prot_policy(prot));
+ prot = prot & ~VM_PROT_WRITE;
+ assert(!pmap_has_prot_policy(prot));
- if (override_nx(old_map, submap_entry->alias) && prot)
+ if (override_nx(old_map,
+ VME_ALIAS(submap_entry))
+ && prot)
prot |= VM_PROT_EXECUTE;
vm_object_pmap_protect(
sub_object,
- submap_entry->offset,
- submap_entry->vme_end -
+ VME_OFFSET(submap_entry),
+ submap_entry->vme_end -
submap_entry->vme_start,
- (submap_entry->is_shared
+ (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 +
- submap_entry->offset);
+ VME_OFFSET(submap_entry));
/* This works diffently than the */
/* normal submap case. We go back */
/* the sub_map, substituting the */
/* new copy object, */
+ subentry_protection = submap_entry->protection;
+ subentry_max_protection = submap_entry->max_protection;
vm_map_unlock(map);
+ submap_entry = NULL; /* not valid after map unlock */
+
local_start = old_start;
local_end = old_end;
map = cow_sub_map_parent;
vaddr = cow_parent_vaddr;
cow_sub_map_parent = NULL;
- if(!vm_map_lookup_entry(map,
+ 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 */
vm_map_clip_start(map, entry, local_start);
vm_map_clip_end(map, entry, local_end);
- /* unnesting was done in vm_map_clip_start/end() */
- assert(!entry->use_pmap);
+ if (entry->is_sub_map) {
+ /* unnesting was done when clipping */
+ assert(!entry->use_pmap);
+ }
/* substitute copy object for */
/* shared map entry */
- vm_map_deallocate(entry->object.sub_map);
+ vm_map_deallocate(VME_SUBMAP(entry));
+ assert(!entry->iokit_acct);
entry->is_sub_map = FALSE;
- entry->object.vm_object = copy_object;
+ 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 (entry->protection != VM_PROT_READ) {
+ /*
+ * Someone has already altered the top entry's
+ * protections via vm_protect(VM_PROT_COPY).
+ * Respect these new values and ignore the
+ * submap entry's protections.
+ */
+ } else {
+ /*
+ * Regular copy-on-write: propagate the submap
+ * entry's protections to the top map entry.
+ */
+ entry->protection |= subentry_protection;
+ }
+ entry->max_protection |= subentry_max_protection;
+
+ if ((entry->protection & VM_PROT_WRITE) &&
+ (entry->protection & VM_PROT_EXECUTE) &&
+#if !CONFIG_EMBEDDED
+ map != kernel_map &&
+ cs_process_enforcement(NULL) &&
+#endif /* !CONFIG_EMBEDDED */
+ !(entry->used_for_jit)) {
+ DTRACE_VM3(cs_wx,
+ uint64_t, (uint64_t)entry->vme_start,
+ uint64_t, (uint64_t)entry->vme_end,
+ vm_prot_t, entry->protection);
+ printf("CODE SIGNING: %d[%s] %s can't have both write and exec at the same time\n",
+ proc_selfpid(),
+ (current_task()->bsd_info
+ ? proc_name_address(current_task()->bsd_info)
+ : "?"),
+ __FUNCTION__);
+ entry->protection &= ~VM_PROT_EXECUTE;
+ }
if(copied_slowly) {
- entry->offset = local_start - old_start;
+ VME_OFFSET_SET(entry, local_start - old_start);
entry->needs_copy = FALSE;
entry->is_shared = FALSE;
} else {
- entry->offset = copy_offset;
+ VME_OFFSET_SET(entry, copy_offset);
+ assert(entry->wired_count == 0);
entry->needs_copy = TRUE;
- if(entry->inheritance == VM_INHERIT_SHARE)
+ 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)
+ if(entry->inheritance == VM_INHERIT_SHARE)
entry->inheritance = VM_INHERIT_COPY;
vm_map_lock_write_to_read(map);
vaddr = local_vaddr;
}
}
-
+
/*
* Check whether this task is allowed to have
* this page.
prot = entry->protection;
- if (override_nx(old_map, entry->alias) && prot) {
+ if (override_nx(old_map, VME_ALIAS(entry)) && prot) {
/*
* HACK -- if not a stack, then allow execution
*/
goto protection_failure;
}
}
- if ((fault_type & (prot)) != fault_type) {
+ if (((fault_type & prot) != fault_type)
+#if __arm64__
+ /* prefetch abort in execute-only page */
+ && !(prot == VM_PROT_EXECUTE && fault_type == (VM_PROT_READ | VM_PROT_EXECUTE))
+#endif
+ ) {
protection_failure:
if (*real_map != map) {
vm_map_unlock(*real_map);
* demote the permissions allowed.
*/
- if ((fault_type & VM_PROT_WRITE) || *wired) {
+ 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
vm_map_lock_read(map);
goto RetryLookup;
}
- vm_object_shadow(&entry->object.vm_object,
- &entry->offset,
- (vm_map_size_t) (entry->vme_end -
- entry->vme_start));
- entry->object.vm_object->shadowed = TRUE;
+ if (VME_OBJECT(entry)->shadowed == FALSE) {
+ vm_object_lock(VME_OBJECT(entry));
+ VME_OBJECT(entry)->shadowed = TRUE;
+ vm_object_unlock(VME_OBJECT(entry));
+ }
+ VME_OBJECT_SHADOW(entry,
+ (vm_map_size_t) (entry->vme_end -
+ entry->vme_start));
entry->needs_copy = FALSE;
+
vm_map_lock_write_to_read(map);
}
- else {
+ if ((fault_type & VM_PROT_WRITE) == 0 && *wired == 0) {
/*
* We're attempting to read a copy-on-write
* page -- don't allow writes.
/*
* Create an object if necessary.
*/
- if (entry->object.vm_object == VM_OBJECT_NULL) {
+ if (VME_OBJECT(entry) == 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_map_size_t)(entry->vme_end - entry->vme_start));
- entry->offset = 0;
+ VME_OBJECT_SET(entry,
+ vm_object_allocate(
+ (vm_map_size_t)(entry->vme_end -
+ entry->vme_start)));
+ VME_OFFSET_SET(entry, 0);
+ assert(entry->use_pmap);
vm_map_lock_write_to_read(map);
}
* return the protection.
*/
- *offset = (vaddr - entry->vme_start) + entry->offset;
- *object = entry->object.vm_object;
+ *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 = entry->alias;
+ 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 = entry->offset;
- fault_info->hi_offset = (entry->vme_end - entry->vme_start) + entry->offset;
+ 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;
- fault_info->cs_bypass = (entry->used_for_jit)? TRUE : FALSE;
+ if (entry->used_for_jit ||
+ entry->vme_resilient_codesign) {
+ fault_info->cs_bypass = TRUE;
+ } else {
+ fault_info->cs_bypass = FALSE;
+ }
+ fault_info->pmap_cs_associated = FALSE;
+#if CONFIG_PMAP_CS
+ if (entry->pmap_cs_associated) {
+ /*
+ * The pmap layer will validate this page
+ * before allowing it to be executed from.
+ */
+ fault_info->pmap_cs_associated = TRUE;
+ }
+#endif /* CONFIG_PMAP_CS */
fault_info->mark_zf_absent = FALSE;
fault_info->batch_pmap_op = FALSE;
}
vm_object_lock(*object);
else
vm_object_lock_shared(*object);
-
+
/*
* Save the version number
*/
* 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.
+ * since the given version. The map has to be locked
+ * ("shared" mode is fine) before calling this function
+ * and it will be returned locked too.
*/
boolean_t
vm_map_verify(
- register vm_map_t map,
- register vm_map_version_t *version) /* REF */
+ vm_map_t map,
+ vm_map_version_t *version) /* REF */
{
boolean_t result;
- vm_map_lock_read(map);
+ vm_map_lock_assert_held(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
* "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"
+ * "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
boolean_t look_for_pages;
vm_region_submap_short_info_64_t short_info;
+ boolean_t do_region_footprint;
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
*/
return KERN_INVALID_ARGUMENT;
}
-
+
+ do_region_footprint = task_self_region_footprint();
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;
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;
next_max_above = (vm_map_offset_t) -1;
next_max_below = (vm_map_offset_t) -1;
- if (not_in_kdp) {
- vm_map_lock_read(curr_map);
- }
-
for (;;) {
if (vm_map_lookup_entry(curr_map,
curr_address,
}
curr_entry = NULL;
curr_map = NULL;
+ curr_skip = 0;
curr_offset = 0;
curr_depth = 0;
curr_max_above = 0;
/* adjust current address and offset */
skip = curr_entry->vme_start - curr_address;
curr_address = curr_entry->vme_start;
- curr_skip = skip;
+ curr_skip += skip;
curr_offset += skip;
curr_max_above -= skip;
curr_max_below = 0;
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);
* the rest of that submap is irrelevant to us, since it's not
* mapped here.
* The relevant portion of the map starts at
- * "curr_entry->offset" up to the size of "curr_entry".
+ * "VME_OFFSET(curr_entry)" up to the size of "curr_entry".
*/
curr_max_above = MIN(curr_max_above,
curr_entry->vme_end - curr_address);
* later.
*/
if (not_in_kdp) {
- vm_map_lock_read(curr_entry->object.sub_map);
+ vm_map_lock_read(VME_SUBMAP(curr_entry));
}
if (curr_map == next_map) {
/* keep "next_map" locked in case we need it */
/*
* Adjust the offset. "curr_entry" maps the submap
* at relative address "curr_entry->vme_start" in the
- * curr_map but skips the first "curr_entry->offset"
+ * 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 +=
- (curr_entry->offset - curr_entry->vme_start);
+ (VME_OFFSET(curr_entry) - curr_entry->vme_start);
curr_address = user_address + curr_offset;
/* switch to the submap */
- curr_map = curr_entry->object.sub_map;
+ curr_map = VME_SUBMAP(curr_entry);
curr_depth++;
curr_entry = NULL;
}
+// 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
+
if (curr_entry == NULL) {
/* no VM region contains the address... */
+
+ if (do_region_footprint && /* we want footprint numbers */
+ next_entry == NULL && /* & there are no more regions */
+ /* & we haven't already provided our fake region: */
+ user_address <= vm_map_last_entry(map)->vme_end) {
+ ledger_amount_t nonvol, nonvol_compressed;
+ /*
+ * Add a fake memory region to account for
+ * purgeable memory that counts towards this
+ * task's memory footprint, i.e. the resident
+ * compressed pages of non-volatile objects
+ * owned by that task.
+ */
+ ledger_get_balance(
+ map->pmap->ledger,
+ task_ledgers.purgeable_nonvolatile,
+ &nonvol);
+ ledger_get_balance(
+ map->pmap->ledger,
+ task_ledgers.purgeable_nonvolatile_compressed,
+ &nonvol_compressed);
+ if (nonvol + nonvol_compressed == 0) {
+ /* no purgeable memory usage to report */
+ return KERN_INVALID_ADDRESS;
+ }
+ /* fake region to show nonvolatile footprint */
+ if (look_for_pages) {
+ submap_info->protection = VM_PROT_DEFAULT;
+ submap_info->max_protection = VM_PROT_DEFAULT;
+ submap_info->inheritance = VM_INHERIT_DEFAULT;
+ submap_info->offset = 0;
+ submap_info->user_tag = -1;
+ submap_info->pages_resident = (unsigned int) (nonvol / PAGE_SIZE);
+ submap_info->pages_shared_now_private = 0;
+ submap_info->pages_swapped_out = (unsigned int) (nonvol_compressed / PAGE_SIZE);
+ submap_info->pages_dirtied = submap_info->pages_resident;
+ submap_info->ref_count = 1;
+ submap_info->shadow_depth = 0;
+ submap_info->external_pager = 0;
+ submap_info->share_mode = SM_PRIVATE;
+ submap_info->is_submap = 0;
+ submap_info->behavior = VM_BEHAVIOR_DEFAULT;
+ submap_info->object_id = INFO_MAKE_FAKE_OBJECT_ID(map, task_ledgers.purgeable_nonvolatile);
+ submap_info->user_wired_count = 0;
+ submap_info->pages_reusable = 0;
+ } else {
+ short_info->user_tag = -1;
+ short_info->offset = 0;
+ short_info->protection = VM_PROT_DEFAULT;
+ short_info->inheritance = VM_INHERIT_DEFAULT;
+ short_info->max_protection = VM_PROT_DEFAULT;
+ short_info->behavior = VM_BEHAVIOR_DEFAULT;
+ short_info->user_wired_count = 0;
+ short_info->is_submap = 0;
+ short_info->object_id = INFO_MAKE_FAKE_OBJECT_ID(map, task_ledgers.purgeable_nonvolatile);
+ short_info->external_pager = 0;
+ short_info->shadow_depth = 0;
+ short_info->share_mode = SM_PRIVATE;
+ short_info->ref_count = 1;
+ }
+ *nesting_depth = 0;
+ *size = (vm_map_size_t) (nonvol + nonvol_compressed);
+// *address = user_address;
+ *address = vm_map_last_entry(map)->vme_end;
+ return KERN_SUCCESS;
+ }
+
if (next_entry == NULL) {
/* ... and no VM region follows it either */
return KERN_INVALID_ADDRESS;
curr_depth = next_depth;
curr_max_above = next_max_above;
curr_max_below = next_max_below;
- if (curr_map == map) {
- user_address = curr_address;
- }
} else {
/* we won't need "next_entry" after all */
if (next_entry != NULL) {
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;
#define INFO_MAKE_OBJECT_ID(p) ((uint32_t)(uintptr_t)VM_KERNEL_ADDRPERM(p))
if (look_for_pages) {
- submap_info->user_tag = curr_entry->alias;
- submap_info->offset = curr_entry->offset;
+ 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(curr_entry->object.vm_object);
+ submap_info->object_id = INFO_MAKE_OBJECT_ID(VME_OBJECT(curr_entry));
} else {
- short_info->user_tag = curr_entry->alias;
- short_info->offset = curr_entry->offset;
+ 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(curr_entry->object.vm_object);
+ short_info->object_id = INFO_MAKE_OBJECT_ID(VME_OBJECT(curr_entry));
}
extended.pages_resident = 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_region_walk(curr_map,
range_start,
curr_entry,
- (curr_entry->offset +
+ (VME_OFFSET(curr_entry) +
(range_start -
curr_entry->vme_start)),
range_end - range_start,
} else {
extended.share_mode = SM_PRIVATE;
}
- extended.ref_count =
- curr_entry->object.sub_map->ref_count;
+ extended.ref_count = VME_SUBMAP(curr_entry)->map_refcnt;
}
}
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;
}
vm_map_entry_t entry;
vm_map_offset_t start;
- if (map == VM_MAP_NULL)
+ if (map == VM_MAP_NULL)
return(KERN_INVALID_ARGUMENT);
switch (flavor) {
start = entry->vme_start;
- basic->offset = (uint32_t)entry->offset;
+ basic->offset = (uint32_t)VME_OFFSET(entry);
basic->protection = entry->protection;
basic->inheritance = entry->inheritance;
basic->max_protection = entry->max_protection;
start = entry->vme_start;
- basic->offset = entry->offset;
+ basic->offset = VME_OFFSET(entry);
basic->protection = entry->protection;
basic->inheritance = entry->inheritance;
basic->max_protection = entry->max_protection;
start = entry->vme_start;
extended->protection = entry->protection;
- extended->user_tag = entry->alias;
+ extended->user_tag = VME_ALIAS(entry);
extended->pages_resident = 0;
extended->pages_swapped_out = 0;
extended->pages_shared_now_private = 0;
*count = VM_REGION_EXTENDED_INFO_COUNT;
}
- vm_map_region_walk(map, start, entry, entry->offset, entry->vme_end - start, extended, TRUE, *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;
return(KERN_SUCCESS);
}
case VM_REGION_TOP_INFO:
- {
+ {
vm_region_top_info_t top;
if (*count < VM_REGION_TOP_INFO_COUNT)
vm_region_top_info_t top)
{
- if (entry->object.vm_object == 0 || entry->is_sub_map) {
+ if (VME_OBJECT(entry) == 0 || entry->is_sub_map) {
top->share_mode = SM_EMPTY;
top->ref_count = 0;
top->obj_id = 0;
entry_size = (uint32_t) ((entry->vme_end - entry->vme_start) / PAGE_SIZE_64);
- obj = entry->object.vm_object;
+ obj = VME_OBJECT(entry);
vm_object_lock(obj);
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);
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 *obj, *tmp_obj;
+ vm_map_offset_t last_offset;
+ int i;
+ int ref_count;
struct vm_object *shadow_object;
int shadow_depth;
+ boolean_t do_region_footprint;
- if ((entry->object.vm_object == 0) ||
+ do_region_footprint = task_self_region_footprint();
+
+ if ((VME_OBJECT(entry) == 0) ||
(entry->is_sub_map) ||
- (entry->object.vm_object->phys_contiguous &&
+ (VME_OBJECT(entry)->phys_contiguous &&
!entry->superpage_size)) {
extended->share_mode = SM_EMPTY;
extended->ref_count = 0;
return;
}
- {
- obj = entry->object.vm_object;
+ obj = VME_OBJECT(entry);
- vm_object_lock(obj);
+ vm_object_lock(obj);
- if ((ref_count = obj->ref_count) > 1 && obj->paging_in_progress)
- ref_count--;
+ 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 (look_for_pages) {
+ for (last_offset = offset + range;
+ offset < last_offset;
+ offset += PAGE_SIZE_64, va += PAGE_SIZE) {
- if ( !(obj->pager_trusted) && !(obj->internal))
- extended->external_pager = 1;
+ if (do_region_footprint) {
+ int disp;
- 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);
+ disp = 0;
+ if (map->has_corpse_footprint) {
+ /*
+ * Query the page info data we saved
+ * while forking the corpse.
+ */
+ vm_map_corpse_footprint_query_page_info(
+ map,
+ va,
+ &disp);
+ } else {
+ /*
+ * Query the pmap.
+ */
+ pmap_query_page_info(map->pmap,
+ va,
+ &disp);
+ }
+ if (disp & PMAP_QUERY_PAGE_PRESENT) {
+ if (!(disp & PMAP_QUERY_PAGE_ALTACCT)) {
+ extended->pages_resident++;
+ }
+ if (disp & PMAP_QUERY_PAGE_REUSABLE) {
+ extended->pages_reusable++;
+ } else if (!(disp & PMAP_QUERY_PAGE_INTERNAL) ||
+ (disp & PMAP_QUERY_PAGE_ALTACCT)) {
+ /* alternate accounting */
+ } else {
+ extended->pages_dirtied++;
+ }
+ } else if (disp & PMAP_QUERY_PAGE_COMPRESSED) {
+ if (disp & PMAP_QUERY_PAGE_COMPRESSED_ALTACCT) {
+ /* alternate accounting */
+ } else {
+ extended->pages_swapped_out++;
+ }
+ }
+ /* deal with alternate accounting */
+ if (obj->purgable == VM_PURGABLE_NONVOLATILE &&
+ /* && not tagged as no-footprint? */
+ VM_OBJECT_OWNER(obj) != NULL &&
+ VM_OBJECT_OWNER(obj)->map == map) {
+ if ((((va
+ - entry->vme_start
+ + VME_OFFSET(entry))
+ / PAGE_SIZE) <
+ (obj->resident_page_count +
+ vm_compressor_pager_get_count(obj->pager)))) {
+ /*
+ * Non-volatile purgeable object owned
+ * by this task: report the first
+ * "#resident + #compressed" pages as
+ * "resident" (to show that they
+ * contribute to the footprint) but not
+ * "dirty" (to avoid double-counting
+ * with the fake "non-volatile" region
+ * we'll report at the end of the
+ * address space to account for all
+ * (mapped or not) non-volatile memory
+ * owned by this task.
+ */
+ extended->pages_resident++;
+ }
+ } else if ((obj->purgable == VM_PURGABLE_VOLATILE ||
+ obj->purgable == VM_PURGABLE_EMPTY) &&
+ /* && not tagged as no-footprint? */
+ VM_OBJECT_OWNER(obj) != NULL &&
+ VM_OBJECT_OWNER(obj)->map == map) {
+ if ((((va
+ - entry->vme_start
+ + VME_OFFSET(entry))
+ / PAGE_SIZE) <
+ obj->wired_page_count)) {
+ /*
+ * Volatile|empty purgeable object owned
+ * by this task: report the first
+ * "#wired" pages as "resident" (to
+ * show that they contribute to the
+ * footprint) but not "dirty" (to avoid
+ * double-counting with the fake
+ * "non-volatile" region we'll report
+ * at the end of the address space to
+ * account for all (mapped or not)
+ * non-volatile memory owned by this
+ * task.
+ */
+ extended->pages_resident++;
}
- vm_object_unlock(shadow_object);
- shadow_object = next_shadow;
+ } else if (obj->purgable != VM_PURGABLE_DENY) {
+ /*
+ * Pages from purgeable objects
+ * will be reported as dirty
+ * appropriately in an extra
+ * fake memory region at the end of
+ * the address space.
+ */
+ } else if (entry->iokit_acct) {
+ /*
+ * IOKit mappings are considered
+ * as fully dirty for footprint's
+ * sake.
+ */
+ extended->pages_dirtied++;
}
+ continue;
}
- extended->shadow_depth = shadow_depth;
+
+ vm_map_region_look_for_page(map, va, obj,
+ offset, ref_count,
+ 0, extended, count);
}
- 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;
- }
+ if (do_region_footprint) {
+ goto collect_object_info;
}
- 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--;
+ } else {
+ collect_object_info:
+ 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;
- extended->ref_count += ref_count;
- obj = tmp_obj;
+ 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 (extended->share_mode == SM_SHARED) {
- register vm_map_entry_t cur;
- register vm_map_entry_t last;
- int my_refs;
+ 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);
- obj = entry->object.vm_object;
- last = vm_map_to_entry(map);
- my_refs = 0;
+ if (extended->share_mode == SM_SHARED) {
+ vm_map_entry_t cur;
+ vm_map_entry_t last;
+ int my_refs;
- 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);
+ obj = VME_OBJECT(entry);
+ last = vm_map_to_entry(map);
+ my_refs = 0;
- if (my_refs == ref_count)
- extended->share_mode = SM_PRIVATE_ALIASED;
- else if (my_refs > 1)
- extended->share_mode = SM_SHARED_ALIASED;
- }
+ 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;
}
}
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;
+ vm_page_t p;
+ vm_object_t shadow;
+ int ref_count;
+ vm_object_t caller_object;
+
shadow = object->shadow;
caller_object = object;
-
+
while (TRUE) {
if ( !(object->pager_trusted) && !(object->internal))
if (shadow && (max_refcnt == 1))
extended->pages_shared_now_private++;
- if (!p->fictitious &&
- (p->dirty || pmap_is_modified(p->phys_page)))
+ if (!p->vmp_fictitious &&
+ (p->vmp_dirty || pmap_is_modified(VM_PAGE_GET_PHYS_PAGE(p))))
extended->pages_dirtied++;
else if (count >= VM_REGION_EXTENDED_INFO_COUNT) {
- if (p->reusable || p->object->all_reusable) {
+ if (p->vmp_reusable || object->all_reusable) {
extended->pages_reusable++;
}
}
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 (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;
}
}
if (ref_count > max_refcnt)
max_refcnt = ref_count;
-
+
if(object != caller_object)
vm_object_unlock(object);
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;
+ int ref_count;
+ vm_object_t chk_obj;
+ vm_object_t tmp_obj;
- if (entry->object.vm_object == 0)
+ if (VME_OBJECT(entry) == 0)
return(0);
if (entry->is_sub_map)
else {
ref_count = 0;
- chk_obj = entry->object.vm_object;
+ chk_obj = VME_OBJECT(entry);
vm_object_lock(chk_obj);
while (chk_obj) {
(prev_entry->vme_end == this_entry->vme_start) &&
(prev_entry->is_sub_map == this_entry->is_sub_map) &&
-
- (prev_entry->object.vm_object == this_entry->object.vm_object) &&
- ((prev_entry->offset + (prev_entry->vme_end -
+ (VME_OBJECT(prev_entry) == VME_OBJECT(this_entry)) &&
+ ((VME_OFFSET(prev_entry) + (prev_entry->vme_end -
prev_entry->vme_start))
- == this_entry->offset) &&
+ == VME_OFFSET(this_entry)) &&
- (prev_entry->map_aligned == this_entry->map_aligned) &&
- (prev_entry->inheritance == this_entry->inheritance) &&
+ (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->behavior == this_entry->behavior) &&
- (prev_entry->alias == this_entry->alias) &&
- (prev_entry->zero_wired_pages == this_entry->zero_wired_pages) &&
+ (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->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->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) &&
+ (prev_entry->pmap_cs_associated == this_entry->pmap_cs_associated) &&
+ /* 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->use_pmap == FALSE) &&
- (this_entry->use_pmap == FALSE) &&
+ ((prev_entry->vme_atomic == FALSE) && (this_entry->vme_atomic == 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)
+ (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);
assert(VM_MAP_PAGE_ALIGNED(prev_entry->vme_start,
VM_MAP_PAGE_MASK(map)));
this_entry->vme_start = prev_entry->vme_start;
- this_entry->offset = prev_entry->offset;
+ VME_OFFSET_SET(this_entry, VME_OFFSET(prev_entry));
+
+ if (map->holelistenabled) {
+ vm_map_store_update_first_free(map, this_entry, TRUE);
+ }
+
if (prev_entry->is_sub_map) {
- vm_map_deallocate(prev_entry->object.sub_map);
+ vm_map_deallocate(VME_SUBMAP(prev_entry));
} else {
- vm_object_deallocate(prev_entry->object.vm_object);
+ vm_object_deallocate(VME_OBJECT(prev_entry));
}
vm_map_entry_dispose(map, prev_entry);
SAVE_HINT_MAP_WRITE(map, this_entry);
entry = entry->vme_next;
}
}
-
+
while (entry != vm_map_to_entry(map) &&
entry->vme_start <= end) {
/* try and coalesce "entry" with its previous entry */
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;
sync_size = end - start;
vm_map_lock(map);
-
- if (attribute != MATTR_CACHE) {
+
+ 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,
vm_map_offset_t sub_start;
vm_map_offset_t sub_end;
- sub_start = (start - entry->vme_start)
- + entry->offset;
+ sub_start = (start - entry->vme_start)
+ + VME_OFFSET(entry);
sub_end = sub_start + sub_size;
vm_map_machine_attribute(
- entry->object.sub_map,
+ VME_SUBMAP(entry),
sub_start,
sub_end,
attribute, value);
} else {
- if(entry->object.vm_object) {
+ if (VME_OBJECT(entry)) {
vm_page_t m;
vm_object_t object;
vm_object_t base_object;
vm_map_size_t range;
range = sub_size;
offset = (start - entry->vme_start)
- + entry->offset;
+ + VME_OFFSET(entry);
base_offset = offset;
- object = entry->object.vm_object;
+ object = VME_OBJECT(entry);
base_object = object;
last_object = NULL;
m = vm_page_lookup(
object, offset);
- if (m && !m->fictitious) {
- ret =
+ if (m && !m->vmp_fictitious) {
+ ret =
pmap_attribute_cache_sync(
- m->phys_page,
- PAGE_SIZE,
+ VM_PAGE_GET_PHYS_PAGE(m),
+ PAGE_SIZE,
attribute, value);
-
+
} else if (object->shadow) {
offset = offset + object->vo_shadow_offset;
last_object = object;
vm_map_unlock(map);
return KERN_FAILURE;
}
-
+
}
vm_map_unlock(map);
*
* 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
+ * how pagein operations resulting from faults on the map will be
* clustered.
*/
-kern_return_t
+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)
{
- register vm_map_entry_t entry;
+ vm_map_entry_t entry;
vm_map_entry_t temp_entry;
XPR(XPR_VM_MAP,
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
+ * 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.
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);
- assert(!entry->use_pmap);
-
+ if (entry->is_sub_map) {
+ assert(!entry->use_pmap);
+ }
+
if( new_behavior == VM_BEHAVIOR_ZERO_WIRED_PAGES ) {
entry->zero_wired_pages = TRUE;
} else {
}
entry = entry->vme_next;
}
-
+
vm_map_unlock(map);
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
+ * an immediate action to take place as opposed to setting a behavior that
* affects future actions.
*/
case VM_BEHAVIOR_CAN_REUSE:
return vm_map_can_reuse(map, start, end);
+#if MACH_ASSERT
+ case VM_BEHAVIOR_PAGEOUT:
+ return vm_map_pageout(map, start, end);
+#endif /* MACH_ASSERT */
+
default:
return(KERN_INVALID_ARGUMENT);
}
vm_map_entry_t entry;
vm_object_t object;
memory_object_t pager;
- struct vm_object_fault_info fault_info;
+ struct vm_object_fault_info fault_info = {};
kern_return_t kr;
vm_object_size_t len;
vm_object_offset_t offset;
- /*
- * 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.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;
/*
* The MADV_WILLNEED operation doesn't require any changes to the
* Examine each vm_map_entry_t in the range.
*/
for (; entry != vm_map_to_entry(map) && start < end; ) {
-
+
/*
* 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.
*/
- offset = (start - entry->vme_start) + entry->offset;
+ offset = (start - entry->vme_start) + VME_OFFSET(entry);
/*
* Set the length so we don't go beyond the end of the
len = (vm_size_t) (0 - PAGE_SIZE);
}
fault_info.cluster_size = (vm_size_t) len;
- fault_info.lo_offset = offset;
+ fault_info.lo_offset = offset;
fault_info.hi_offset = offset + len;
- fault_info.user_tag = entry->alias;
+ 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;
+ }
/*
* If there's no read permission to this mapping, then just
*
* 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 and it
+ * 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
vm_map_entry_is_reusable(
vm_map_entry_t entry)
{
+ /* Only user map entries */
+
vm_object_t object;
- switch (entry->alias) {
+ if (entry->is_sub_map) {
+ return FALSE;
+ }
+
+ switch (VME_ALIAS(entry)) {
case VM_MEMORY_MALLOC:
case VM_MEMORY_MALLOC_SMALL:
case VM_MEMORY_MALLOC_LARGE:
return TRUE;
}
- if (entry->is_shared ||
+ if (/*entry->is_shared ||*/
entry->is_sub_map ||
entry->in_transition ||
entry->protection != VM_PROT_DEFAULT ||
return FALSE;
}
- object = entry->object.vm_object;
+ object = VME_OBJECT(entry);
if (object == VM_OBJECT_NULL) {
return TRUE;
}
* 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
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->purgable == VM_PURGABLE_DENY &&
+ object->copy_strategy != MEMORY_OBJECT_COPY_DELAY &&
!object->true_share &&
object->wimg_bits == VM_WIMG_USE_DEFAULT &&
!object->code_signed) {
return TRUE;
}
return FALSE;
-
-
+
+
}
static kern_return_t
*/
vm_map_lock_read(map);
+ assert(map->pmap != kernel_pmap); /* protect alias access */
/*
* The madvise semantics require that the address range be fully
start_offset = 0;
}
end_offset = MIN(end, entry->vme_end) - entry->vme_start;
- start_offset += entry->offset;
- end_offset += entry->offset;
+ start_offset += VME_OFFSET(entry);
+ end_offset += VME_OFFSET(entry);
- object = entry->object.vm_object;
+ assert(!entry->is_sub_map);
+ object = VME_OBJECT(entry);
if (object != VM_OBJECT_NULL) {
- /* tell pmap to not count this range as "reusable" */
- pmap_reusable(map->pmap,
- MAX(start, entry->vme_start),
- MIN(end, entry->vme_end),
- FALSE);
vm_object_lock(object);
vm_object_reuse_pages(object, start_offset, end_offset,
TRUE);
vm_object_unlock(object);
}
- if (entry->alias == VM_MEMORY_MALLOC_LARGE_REUSABLE) {
+ if (VME_ALIAS(entry) == VM_MEMORY_MALLOC_LARGE_REUSABLE) {
/*
* XXX
* We do not hold the VM map exclusively here.
* one that can be modified while holding the VM map
* "shared".
*/
- entry->alias = VM_MEMORY_MALLOC_LARGE_REUSED;
+ VME_ALIAS_SET(entry, VM_MEMORY_MALLOC_LARGE_REUSED);
}
}
-
+
vm_map_unlock_read(map);
vm_page_stats_reusable.reuse_pages_success++;
return KERN_SUCCESS;
vm_map_entry_t entry;
vm_object_t object;
vm_object_offset_t start_offset, end_offset;
+ vm_map_offset_t pmap_offset;
/*
* The MADV_REUSABLE operation doesn't require any changes to the
*/
vm_map_lock_read(map);
+ assert(map->pmap != kernel_pmap); /* protect alias access */
/*
* The madvise semantics require that the address range be fully
return KERN_INVALID_ADDRESS;
}
+ if (! (entry->protection & VM_PROT_WRITE) && !entry->used_for_jit) {
+ /* not writable: can't discard contents */
+ vm_map_unlock_read(map);
+ vm_page_stats_reusable.reusable_nonwritable++;
+ vm_page_stats_reusable.reusable_pages_failure++;
+ return KERN_PROTECTION_FAILURE;
+ }
+
/*
* The first time through, the start address could be anywhere
* within the vm_map_entry we found. So adjust the offset to
*/
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 += entry->offset;
- end_offset += entry->offset;
+ start_offset += VME_OFFSET(entry);
+ end_offset += VME_OFFSET(entry);
- object = entry->object.vm_object;
+ assert(!entry->is_sub_map);
+ object = VME_OBJECT(entry);
if (object == VM_OBJECT_NULL)
continue;
vm_object_lock(object);
- if (object->ref_count == 1 && !object->shadow)
+ if (((object->ref_count == 1) ||
+ (object->copy_strategy != MEMORY_OBJECT_COPY_SYMMETRIC &&
+ object->copy == VM_OBJECT_NULL)) &&
+ object->shadow == VM_OBJECT_NULL &&
+ /*
+ * "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))) {
+ if (object->ref_count != 1) {
+ vm_page_stats_reusable.reusable_shared++;
+ }
kill_pages = 1;
- else
+ } else {
kill_pages = -1;
+ }
if (kill_pages != -1) {
- /* tell pmap to count this range as "reusable" */
- pmap_reusable(map->pmap,
- MAX(start, entry->vme_start),
- MIN(end, entry->vme_end),
- TRUE);
vm_object_deactivate_pages(object,
start_offset,
end_offset - start_offset,
kill_pages,
- TRUE /*reusable_pages*/);
+ TRUE /*reusable_pages*/,
+ map->pmap,
+ pmap_offset);
} else {
vm_page_stats_reusable.reusable_pages_shared++;
}
vm_object_unlock(object);
- if (entry->alias == VM_MEMORY_MALLOC_LARGE ||
- entry->alias == VM_MEMORY_MALLOC_LARGE_REUSED) {
+ if (VME_ALIAS(entry) == VM_MEMORY_MALLOC_LARGE ||
+ VME_ALIAS(entry) == VM_MEMORY_MALLOC_LARGE_REUSED) {
/*
* XXX
* We do not hold the VM map exclusively here.
* one that can be modified while holding the VM map
* "shared".
*/
- entry->alias = VM_MEMORY_MALLOC_LARGE_REUSABLE;
+ VME_ALIAS_SET(entry, VM_MEMORY_MALLOC_LARGE_REUSABLE);
}
}
-
+
vm_map_unlock_read(map);
vm_page_stats_reusable.reusable_pages_success++;
return KERN_SUCCESS;
*/
vm_map_lock_read(map);
+ assert(map->pmap != kernel_pmap); /* protect alias access */
/*
* The madvise semantics require that the address range be fully
return KERN_INVALID_ADDRESS;
}
}
-
+
vm_map_unlock_read(map);
vm_page_stats_reusable.can_reuse_success++;
return KERN_SUCCESS;
}
+#if MACH_ASSERT
+static kern_return_t
+vm_map_pageout(
+ vm_map_t map,
+ vm_map_offset_t start,
+ vm_map_offset_t end)
+{
+ vm_map_entry_t entry;
+
+ /*
+ * 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);
+
+ /*
+ * 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) && entry->vme_start < end;
+ entry = entry->vme_next) {
+ vm_object_t object;
+
+ /*
+ * 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;
+ }
+
+ object = VME_OBJECT(submap_entry);
+ if (submap_entry->is_sub_map ||
+ object == VM_OBJECT_NULL ||
+ !object->internal) {
+ vm_map_unlock_read(submap);
+ continue;
+ }
+
+ vm_object_pageout(object);
+
+ vm_map_unlock_read(submap);
+ submap = VM_MAP_NULL;
+ submap_entry = VM_MAP_ENTRY_NULL;
+ continue;
+ }
+
+ object = VME_OBJECT(entry);
+ if (entry->is_sub_map ||
+ object == VM_OBJECT_NULL ||
+ !object->internal) {
+ continue;
+ }
+
+ vm_object_pageout(object);
+ }
+
+ vm_map_unlock_read(map);
+ return KERN_SUCCESS;
+}
+#endif /* MACH_ASSERT */
+
+
/*
* Routine: vm_map_entry_insert
*
- * Descritpion: This routine inserts a new vm_entry in a locked map.
+ * Description: This routine inserts a new vm_entry in a locked map.
*/
vm_map_entry_t
vm_map_entry_insert(
boolean_t no_cache,
boolean_t permanent,
unsigned int superpage_size,
- boolean_t clear_map_aligned)
+ boolean_t clear_map_aligned,
+ boolean_t is_submap,
+ boolean_t used_for_jit,
+ int alias)
{
vm_map_entry_t new_entry;
assert(insp_entry != (vm_map_entry_t)0);
+ vm_map_lock_assert_exclusive(map);
+
+#if DEVELOPMENT || DEBUG
+ vm_object_offset_t end_offset = 0;
+ assertf(!os_add_overflow(end - start, offset, &end_offset), "size 0x%llx, offset 0x%llx caused overflow", (uint64_t)(end - start), offset);
+#endif /* DEVELOPMENT || DEBUG */
new_entry = vm_map_entry_create(map, !map->hdr.entries_pageable);
new_entry->map_aligned = FALSE;
}
if (clear_map_aligned &&
- ! VM_MAP_PAGE_ALIGNED(end, VM_MAP_PAGE_MASK(map))) {
+ (! 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;
}
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)));
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);
- new_entry->object.vm_object = object;
- new_entry->offset = offset;
+ VME_OBJECT_SET(new_entry, object);
+ VME_OFFSET_SET(new_entry, offset);
new_entry->is_shared = is_shared;
- new_entry->is_sub_map = FALSE;
+ 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->behavior = behavior;
new_entry->wired_count = wired_count;
new_entry->user_wired_count = 0;
- new_entry->use_pmap = FALSE;
- new_entry->alias = 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, alias);
new_entry->zero_wired_pages = FALSE;
new_entry->no_cache = no_cache;
new_entry->permanent = permanent;
new_entry->superpage_size = TRUE;
else
new_entry->superpage_size = FALSE;
- new_entry->used_for_jit = FALSE;
+ if (used_for_jit){
+#if CONFIG_EMBEDDED
+ if (!(map->jit_entry_exists))
+#endif /* CONFIG_EMBEDDED */
+ {
+ new_entry->used_for_jit = TRUE;
+ map->jit_entry_exists = TRUE;
+
+ /* Tell the pmap that it supports JIT. */
+ pmap_set_jit_entitled(map->pmap);
+ }
+ } else {
+ new_entry->used_for_jit = FALSE;
+ }
+ new_entry->pmap_cs_associated = FALSE;
+ new_entry->iokit_acct = FALSE;
+ new_entry->vme_resilient_codesign = FALSE;
+ new_entry->vme_resilient_media = FALSE;
+ new_entry->vme_atomic = FALSE;
/*
* Insert the new entry into the list.
*/
- vm_map_store_entry_link(map, insp_entry, new_entry);
+ vm_map_store_entry_link(map, insp_entry, new_entry,
+ VM_MAP_KERNEL_FLAGS_NONE);
map->size += end - start;
/*
vm_prot_t *max_protection,
/* What, no behavior? */
vm_inherit_t inheritance,
- boolean_t pageable)
+ boolean_t pageable,
+ boolean_t same_map,
+ vm_map_kernel_flags_t vmk_flags)
{
kern_return_t result;
vm_map_size_t mapped_size;
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_object_t object;
vm_map_version_t version;
boolean_t src_needs_copy;
boolean_t new_entry_needs_copy;
+ vm_map_entry_t saved_src_entry;
+ boolean_t src_entry_was_wired;
+ vm_prot_t max_prot_for_prot_copy;
assert(map != VM_MAP_NULL);
assert(size != 0);
/*
* 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->links.next = CAST_TO_VM_MAP_ENTRY(&map_header->links);
+ map_header->links.prev = CAST_TO_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 );
+ if (copy && vmk_flags.vmkf_remap_prot_copy) {
+ max_prot_for_prot_copy = *max_protection & VM_PROT_ALL;
+ } else {
+ max_prot_for_prot_copy = VM_PROT_NONE;
+ }
*cur_protection = VM_PROT_ALL;
*max_protection = VM_PROT_ALL;
mapped_size = 0;
result = KERN_SUCCESS;
- /*
+ /*
* The specified source virtual space might correspond to
* multiple map entries, need to loop on them.
*/
/*
* Find the beginning of the region.
- */
+ */
if (! vm_map_lookup_entry(map, src_start, &src_entry)) {
result = KERN_INVALID_ADDRESS;
break;
src_entry->vme_start);
if(src_entry->is_sub_map) {
- vm_map_reference(src_entry->object.sub_map);
+ vm_map_reference(VME_SUBMAP(src_entry));
object = VM_OBJECT_NULL;
} else {
- object = src_entry->object.vm_object;
+ 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.
+ */
+ assertf(!src_entry->use_pmap,
+ "map=%p src_entry=%p [0x%llx:0x%llx] 0x%x/0x%x %d",
+ map,
+ src_entry,
+ (uint64_t)src_entry->vme_start,
+ (uint64_t)src_entry->vme_end,
+ src_entry->protection,
+ src_entry->max_protection,
+ VME_ALIAS(src_entry));
+ } else {
+ /*
+ * Not IOKit or purgeable:
+ * must be accounted by pmap stats.
+ */
+ assertf(src_entry->use_pmap,
+ "map=%p src_entry=%p [0x%llx:0x%llx] 0x%x/0x%x %d",
+ map,
+ src_entry,
+ (uint64_t)src_entry->vme_start,
+ (uint64_t)src_entry->vme_end,
+ src_entry->protection,
+ src_entry->max_protection,
+ VME_ALIAS(src_entry));
+ }
if (object == VM_OBJECT_NULL) {
object = vm_object_allocate(entry_size);
- src_entry->offset = 0;
- src_entry->object.vm_object = object;
+ VME_OFFSET_SET(src_entry, 0);
+ VME_OBJECT_SET(src_entry, object);
+ assert(src_entry->use_pmap);
} else if (object->copy_strategy !=
MEMORY_OBJECT_COPY_SYMMETRIC) {
/*
!src_entry->is_shared &&
object->vo_size > entry_size)) {
- vm_object_shadow(&src_entry->object.vm_object,
- &src_entry->offset,
- entry_size);
+ VME_OBJECT_SHADOW(src_entry, entry_size);
+ assert(src_entry->use_pmap);
if (!src_entry->needs_copy &&
(src_entry->protection & VM_PROT_WRITE)) {
vm_prot_t prot;
+ assert(!pmap_has_prot_policy(src_entry->protection));
+
prot = src_entry->protection & ~VM_PROT_WRITE;
- if (override_nx(map, src_entry->alias) && prot)
+ if (override_nx(map,
+ VME_ALIAS(src_entry))
+ && prot)
prot |= VM_PROT_EXECUTE;
+ assert(!pmap_has_prot_policy(prot));
+
if(map->mapped_in_other_pmaps) {
vm_object_pmap_protect(
- src_entry->object.vm_object,
- src_entry->offset,
+ VME_OBJECT(src_entry),
+ VME_OFFSET(src_entry),
entry_size,
PMAP_NULL,
src_entry->vme_start,
}
}
- object = src_entry->object.vm_object;
+ object = VME_OBJECT(src_entry);
src_entry->needs_copy = FALSE;
}
vm_object_lock(object);
vm_object_reference_locked(object); /* object ref. for new entry */
- if (object->copy_strategy ==
+ if (object->copy_strategy ==
MEMORY_OBJECT_COPY_SYMMETRIC) {
- object->copy_strategy =
+ object->copy_strategy =
MEMORY_OBJECT_COPY_DELAY;
}
vm_object_unlock(object);
}
- offset = src_entry->offset + (src_start - src_entry->vme_start);
+ offset = (VME_OFFSET(src_entry) +
+ (src_start - src_entry->vme_start));
new_entry = _vm_map_entry_create(map_header, !map_header->entries_pageable);
vm_map_entry_copy(new_entry, src_entry);
- new_entry->use_pmap = FALSE; /* clr address space specifics */
+ if (new_entry->is_sub_map) {
+ /* clr address space specifics */
+ new_entry->use_pmap = FALSE;
+ } else if (copy) {
+ /*
+ * We're dealing with a copy-on-write operation,
+ * so the resulting mapping should not inherit the
+ * original mapping's accounting settings.
+ * "use_pmap" should be reset to its default (TRUE)
+ * so that the new mapping gets accounted for in
+ * the task's memory footprint.
+ */
+ new_entry->use_pmap = TRUE;
+ }
+ /* "iokit_acct" was cleared in vm_map_entry_copy() */
+ assert(!new_entry->iokit_acct);
new_entry->map_aligned = FALSE;
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;
- new_entry->offset = offset;
-
+ if (copy && vmk_flags.vmkf_remap_prot_copy) {
+ /*
+ * Remapping for vm_map_protect(VM_PROT_COPY)
+ * to convert a read-only mapping into a
+ * copy-on-write version of itself but
+ * with write access:
+ * keep the original inheritance and add
+ * VM_PROT_WRITE to the max protection.
+ */
+ new_entry->inheritance = src_entry->inheritance;
+ new_entry->protection &= max_prot_for_prot_copy;
+ new_entry->max_protection |= VM_PROT_WRITE;
+ } else {
+ new_entry->inheritance = inheritance;
+ }
+ VME_OFFSET_SET(new_entry, offset);
+
/*
* The new region has to be copied now if required.
*/
* Cannot allow an entry describing a JIT
* region to be shared across address spaces.
*/
- if (src_entry->used_for_jit == TRUE) {
+ if (src_entry->used_for_jit == TRUE && !same_map) {
+#if CONFIG_EMBEDDED
result = KERN_INVALID_ARGUMENT;
break;
+#endif /* CONFIG_EMBEDDED */
}
src_entry->is_shared = TRUE;
new_entry->is_shared = TRUE;
- if (!(new_entry->is_sub_map))
+ 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 */
+ 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(&new_entry->object.vm_object,
- new_entry->offset,
+ 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 = new_entry_needs_copy;
new_entry->is_shared = FALSE;
+ assertf(new_entry->use_pmap, "map %p new_entry %p\n", map, new_entry);
/*
* Handle copy_on_write semantics.
if (src_needs_copy && !src_entry->needs_copy) {
vm_prot_t prot;
+ assert(!pmap_has_prot_policy(src_entry->protection));
+
prot = src_entry->protection & ~VM_PROT_WRITE;
- if (override_nx(map, src_entry->alias) && prot)
+ if (override_nx(map,
+ VME_ALIAS(src_entry))
+ && prot)
prot |= VM_PROT_EXECUTE;
+ assert(!pmap_has_prot_policy(prot));
+
vm_object_pmap_protect(object,
offset,
entry_size,
- ((src_entry->is_shared
+ ((src_entry->is_shared
|| map->mapped_in_other_pmaps) ?
PMAP_NULL : map->pmap),
src_entry->vme_start,
prot);
+ assert(src_entry->wired_count == 0);
src_entry->needs_copy = TRUE;
}
/*
} else {
new_entry->is_shared = FALSE;
+ assertf(new_entry->use_pmap, "map %p new_entry %p\n", map, new_entry);
+
+ src_entry_was_wired = (src_entry->wired_count > 0);
+ saved_src_entry = src_entry;
+ src_entry = VM_MAP_ENTRY_NULL;
/*
* The map can be safely unlocked since we
/*
* Perform the copy.
*/
- if (src_entry->wired_count > 0) {
+ if (src_entry_was_wired > 0) {
vm_object_lock(object);
result = vm_object_copy_slowly(
object,
offset,
- entry_size,
+ (new_entry->vme_end -
+ new_entry->vme_start),
THREAD_UNINT,
- &new_entry->object.vm_object);
+ &VME_OBJECT(new_entry));
- new_entry->offset = 0;
+ 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,
- &new_entry->object.vm_object,
- &new_entry->offset,
+ (new_entry->vme_end -
+ new_entry->vme_start),
+ &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_entry_dispose(map_header, new_entry);
+ vm_map_lock(map);
break;
}
* Retry the lookup and verify that the
* same object/offset are still present.
*/
- vm_object_deallocate(new_entry->
- object.vm_object);
+ saved_src_entry = VM_MAP_ENTRY_NULL;
+ 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;
}
+ /* map hasn't changed: src_entry is still valid */
+ src_entry = saved_src_entry;
+ saved_src_entry = VM_MAP_ENTRY_NULL;
if (result == KERN_MEMORY_RESTART_COPY) {
vm_object_reference(object);
* Free all allocated elements.
*/
for (src_entry = map_header->links.next;
- src_entry != (struct vm_map_entry *)&map_header->links;
+ src_entry != CAST_TO_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(src_entry->object.sub_map);
+ vm_map_deallocate(VME_SUBMAP(src_entry));
} else {
- vm_object_deallocate(src_entry->object.vm_object);
+ vm_object_deallocate(VME_OBJECT(src_entry));
}
_vm_map_entry_dispose(map_header, src_entry);
}
vm_map_size_t size,
vm_map_offset_t mask,
int flags,
+ vm_map_kernel_flags_t vmk_flags,
+ vm_tag_t tag,
vm_map_t src_map,
vm_map_offset_t memory_address,
boolean_t copy,
return KERN_INVALID_ARGUMENT;
}
- /*
- * 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,
+ /*
+ * 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
+ *
+ * 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
+ * 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) {
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);
- }
+ }
+ if (size == 0) {
+ return KERN_INVALID_ARGUMENT;
+ }
result = vm_map_remap_extract(src_map, memory_address,
size, copy, &map_header,
cur_protection,
max_protection,
inheritance,
- target_map->hdr.entries_pageable);
+ target_map->hdr.entries_pageable,
+ src_map == target_map,
+ vmk_flags);
if (result != KERN_SUCCESS) {
return result;
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);
+ mask, flags, vmk_flags, tag,
+ &insp_entry);
for (entry = map_header.links.next;
- entry != (struct vm_map_entry *)&map_header.links;
+ entry != CAST_TO_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 */
+ 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);
+ vm_map_store_entry_link(target_map, insp_entry, entry,
+ vmk_flags);
insp_entry = entry;
} else {
if (!entry->is_sub_map) {
- vm_object_deallocate(entry->object.vm_object);
+ vm_object_deallocate(VME_OBJECT(entry));
} else {
- vm_map_deallocate(entry->object.sub_map);
+ vm_map_deallocate(VME_SUBMAP(entry));
}
_vm_map_entry_dispose(&map_header, entry);
}
}
+ if (flags & VM_FLAGS_RESILIENT_CODESIGN) {
+ *cur_protection = VM_PROT_READ;
+ *max_protection = VM_PROT_READ;
+ }
+
if( target_map->disable_vmentry_reuse == TRUE) {
+ assert(!target_map->is_nested_map);
if( target_map->highest_entry_end < insp_entry->vme_end ){
target_map->highest_entry_end = insp_entry->vme_end;
}
if (result == KERN_SUCCESS) {
target_map->size += size;
SAVE_HINT_MAP_WRITE(target_map, insp_entry);
+
+#if PMAP_CS
+ if (*max_protection & VM_PROT_EXECUTE) {
+ vm_map_address_t region_start = 0, region_size = 0;
+ struct pmap_cs_code_directory *region_cd = NULL;
+ vm_map_address_t base = 0;
+ struct pmap_cs_lookup_results results = {};
+ vm_map_size_t page_addr = vm_map_trunc_page(memory_address, PAGE_MASK);
+ vm_map_size_t assoc_size = vm_map_round_page(memory_address + size - page_addr, PAGE_MASK);
+
+ pmap_cs_lookup(src_map->pmap, memory_address, &results);
+ region_size = results.region_size;
+ region_start = results.region_start;
+ region_cd = results.region_cd_entry;
+ base = results.base;
+
+ if (region_cd != NULL && (page_addr != region_start || assoc_size != region_size)) {
+ *cur_protection = VM_PROT_READ;
+ *max_protection = VM_PROT_READ;
+ printf("mismatched remap of executable range 0x%llx-0x%llx to 0x%llx, "
+ "region_start 0x%llx, region_size 0x%llx, cd_entry %sNULL, making non-executable.\n",
+ page_addr, page_addr+assoc_size, *address,
+ region_start, region_size,
+ region_cd != NULL ? "not " : "" // Don't leak kernel slide
+ );
+ }
+ }
+#endif
+
}
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);
+ result = vm_map_wire_kernel(target_map, *address,
+ *address + size, *cur_protection, VM_KERN_MEMORY_MLOCK,
+ TRUE);
- /*
- * If requested, return the address of the data pointed to by the
+ /*
+ * 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) {
vm_map_size_t size,
vm_map_offset_t mask,
int flags,
+ vm_map_kernel_flags_t vmk_flags,
+ __unused vm_tag_t tag,
vm_map_entry_t *map_entry) /* OUT */
{
vm_map_entry_t entry;
vm_map_offset_t start;
vm_map_offset_t end;
+ vm_map_offset_t desired_empty_end;
kern_return_t kr;
+ vm_map_entry_t hole_entry;
StartAgain: ;
if (flags & VM_FLAGS_ANYWHERE)
{
+ if (flags & VM_FLAGS_RANDOM_ADDR)
+ {
+ /*
+ * Get a random start address.
+ */
+ kr = vm_map_random_address_for_size(map, address, size);
+ if (kr != KERN_SUCCESS) {
+ return(kr);
+ }
+ start = *address;
+ }
+
/*
* Calculate the first possible address.
*/
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
if( map->disable_vmentry_reuse == TRUE) {
VM_MAP_HIGHEST_ENTRY(map, entry, start);
} 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;
+
+ if (map->holelistenabled) {
+ hole_entry = CAST_TO_VM_MAP_ENTRY(map->holes_list);
+
+ if (hole_entry == NULL) {
+ /*
+ * No more space in the map?
+ */
+ return(KERN_NO_SPACE);
+ } else {
+
+ boolean_t found_hole = FALSE;
+
+ do {
+ if (hole_entry->vme_start >= start) {
+ start = hole_entry->vme_start;
+ found_hole = TRUE;
+ break;
+ }
+
+ if (hole_entry->vme_end > start) {
+ found_hole = TRUE;
+ break;
+ }
+ hole_entry = hole_entry->vme_next;
+
+ } while (hole_entry != CAST_TO_VM_MAP_ENTRY(map->holes_list));
+
+ if (found_hole == FALSE) {
+ return (KERN_NO_SPACE);
+ }
+
+ entry = hole_entry;
+ }
} else {
- vm_map_entry_t tmp_entry;
- if (vm_map_lookup_entry(map, start, &tmp_entry))
- start = tmp_entry->vme_end;
- entry = tmp_entry;
+ 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));
}
-
+
/*
* In any case, the "entry" always precedes
* the proposed new region throughout the
*/
while (TRUE) {
- register vm_map_entry_t next;
+ vm_map_entry_t next;
/*
* Find the end of the proposed new region.
start = end;
end += size;
- if ((end > map->max_offset) || (end < start)) {
+ /* We want an entire page of empty space, but don't increase the allocation size. */
+ desired_empty_end = vm_map_round_page(end, VM_MAP_PAGE_MASK(map));
+
+ if ((desired_empty_end > map->max_offset) || (desired_empty_end < start)) {
if (map->wait_for_space) {
if (size <= (map->max_offset -
map->min_offset)) {
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 >= desired_empty_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 >= desired_empty_end)
+ break;
+ }
/*
* Didn't fit -- move to the next entry.
*/
entry = next;
- start = entry->vme_end;
+
+ if (map->holelistenabled) {
+ if (entry == CAST_TO_VM_MAP_ENTRY(map->holes_list)) {
+ /*
+ * Wrapped around
+ */
+ 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);
+ }
}
+
*address = start;
+
} else {
vm_map_entry_t temp_entry;
-
+
/*
* Verify that:
* the address doesn't itself violate
*/
if (flags & VM_FLAGS_OVERWRITE) {
vm_map_t zap_map;
+ int remove_flags = VM_MAP_REMOVE_SAVE_ENTRIES | VM_MAP_REMOVE_NO_MAP_ALIGN;
/*
* We use a "zap_map" to avoid having to unlock
return KERN_RESOURCE_SHORTAGE;
}
vm_map_set_page_shift(zap_map, VM_MAP_PAGE_SHIFT(map));
+ vm_map_disable_hole_optimization(zap_map);
+ if (vmk_flags.vmkf_overwrite_immutable) {
+ remove_flags |= VM_MAP_REMOVE_IMMUTABLE;
+ }
kr = vm_map_delete(map, start, end,
- VM_MAP_REMOVE_SAVE_ENTRIES,
+ remove_flags,
zap_map);
if (kr == KERN_SUCCESS) {
vm_map_destroy(zap_map,
vm_map_entry_t entry;
vm_object_t object;
kern_return_t kr;
+ boolean_t was_nonvolatile;
/*
* Vet all the input parameters and current type and state of the
if (control != VM_PURGABLE_SET_STATE &&
control != VM_PURGABLE_GET_STATE &&
- control != VM_PURGABLE_PURGE_ALL)
+ control != VM_PURGABLE_PURGE_ALL &&
+ control != VM_PURGABLE_SET_STATE_FROM_KERNEL)
return(KERN_INVALID_ARGUMENT);
if (control == VM_PURGABLE_PURGE_ALL) {
return KERN_SUCCESS;
}
- if (control == VM_PURGABLE_SET_STATE &&
+ if ((control == VM_PURGABLE_SET_STATE ||
+ control == VM_PURGABLE_SET_STATE_FROM_KERNEL) &&
(((*state & ~(VM_PURGABLE_ALL_MASKS)) != 0) ||
((*state & VM_PURGABLE_STATE_MASK) > VM_PURGABLE_STATE_MASK)))
return(KERN_INVALID_ARGUMENT);
return(KERN_PROTECTION_FAILURE);
}
- object = entry->object.vm_object;
- if (object == VM_OBJECT_NULL) {
+ object = VME_OBJECT(entry);
+ if (object == VM_OBJECT_NULL ||
+ object->purgable == VM_PURGABLE_DENY) {
/*
- * Object must already be present or it can't be purgable.
+ * Object must already be present and be purgeable.
*/
vm_map_unlock_read(map);
return KERN_INVALID_ARGUMENT;
}
-
+
vm_object_lock(object);
#if 00
- if (entry->offset != 0 ||
+ if (VME_OFFSET(entry) != 0 ||
entry->vme_end - entry->vme_start != object->vo_size) {
/*
* Can only apply purgable controls to the whole (existing)
return KERN_INVALID_ARGUMENT;
}
#endif
-
+
+ assert(!entry->is_sub_map);
+ assert(!entry->use_pmap); /* purgeable has its own accounting */
+
vm_map_unlock_read(map);
+ was_nonvolatile = (object->purgable == VM_PURGABLE_NONVOLATILE);
+
kr = vm_object_purgable_control(object, control, state);
+ if (was_nonvolatile &&
+ object->purgable != VM_PURGABLE_NONVOLATILE &&
+ map->pmap == kernel_pmap) {
+#if DEBUG
+ object->vo_purgeable_volatilizer = kernel_task;
+#endif /* DEBUG */
+ }
+
vm_object_unlock(object);
return kr;
return kr;
}
-
+
kern_return_t
vm_map_page_info(
vm_map_t map,
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;
+ return (vm_map_page_range_info_internal(map,
+ offset, /* start of range */
+ (offset + 1), /* this will get rounded in the call to the page boundary */
+ flavor,
+ info,
+ count));
+}
+
+kern_return_t
+vm_map_page_range_info_internal(
+ vm_map_t map,
+ vm_map_offset_t start_offset,
+ vm_map_offset_t end_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_MAP_ENTRY_NULL;
+ vm_object_t object = VM_OBJECT_NULL, curr_object = VM_OBJECT_NULL;
+ vm_page_t m = VM_PAGE_NULL;
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;
+ int disposition = 0;
+ int ref_count = 0;
+ int depth = 0, info_idx = 0;
+ vm_page_info_basic_t basic_info = 0;
+ vm_map_offset_t offset_in_page = 0, offset_in_object = 0, curr_offset_in_object = 0;
+ vm_map_offset_t start = 0, end = 0, curr_s_offset = 0, curr_e_offset = 0;
+ boolean_t do_region_footprint;
switch (flavor) {
case VM_PAGE_INFO_BASIC:
return KERN_INVALID_ARGUMENT;
}
+ do_region_footprint = task_self_region_footprint();
disposition = 0;
ref_count = 0;
- top_object = TRUE;
depth = 0;
-
+ info_idx = 0; /* Tracks the next index within the info structure to be filled.*/
retval = KERN_SUCCESS;
- offset_in_page = offset & PAGE_MASK;
- offset = vm_map_trunc_page(offset, PAGE_MASK);
+
+ offset_in_page = start_offset & PAGE_MASK;
+ start = vm_map_trunc_page(start_offset, PAGE_MASK);
+ end = vm_map_round_page(end_offset, PAGE_MASK);
+
+ assert ((end - start) <= MAX_PAGE_RANGE_QUERY);
vm_map_lock_read(map);
- /*
- * 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;
+ for (curr_s_offset = start; curr_s_offset < end;) {
+ /*
+ * New lookup needs reset of these variables.
+ */
+ curr_object = object = VM_OBJECT_NULL;
+ offset_in_object = 0;
+ ref_count = 0;
+ depth = 0;
+
+ if (do_region_footprint &&
+ curr_s_offset >= vm_map_last_entry(map)->vme_end) {
+ ledger_amount_t nonvol_compressed;
+
+ /*
+ * Request for "footprint" info about a page beyond
+ * the end of address space: this must be for
+ * the fake region vm_map_region_recurse_64()
+ * reported to account for non-volatile purgeable
+ * memory owned by this task.
+ */
+ disposition = 0;
+ nonvol_compressed = 0;
+ ledger_get_balance(
+ map->pmap->ledger,
+ task_ledgers.purgeable_nonvolatile_compressed,
+ &nonvol_compressed);
+ if (curr_s_offset - vm_map_last_entry(map)->vme_end <=
+ (unsigned) nonvol_compressed) {
+ /*
+ * We haven't reported all the "non-volatile
+ * compressed" pages yet, so report this fake
+ * page as "compressed".
+ */
+ disposition |= VM_PAGE_QUERY_PAGE_PAGED_OUT;
+ } else {
+ /*
+ * We've reported all the non-volatile
+ * compressed page but not all the non-volatile
+ * pages , so report this fake page as
+ * "resident dirty".
+ */
+ disposition |= VM_PAGE_QUERY_PAGE_PRESENT;
+ disposition |= VM_PAGE_QUERY_PAGE_DIRTY;
+ disposition |= VM_PAGE_QUERY_PAGE_REF;
+ }
+ switch (flavor) {
+ case VM_PAGE_INFO_BASIC:
+ basic_info = (vm_page_info_basic_t) (((uintptr_t) info) + (info_idx * sizeof(struct vm_page_info_basic)));
+ basic_info->disposition = disposition;
+ basic_info->ref_count = 1;
+ basic_info->object_id = INFO_MAKE_FAKE_OBJECT_ID(map, task_ledgers.purgeable_nonvolatile);
+ basic_info->offset = 0;
+ basic_info->depth = 0;
+
+ info_idx++;
+ break;
+ }
+ curr_s_offset += PAGE_SIZE;
+ continue;
+ }
+
+ /*
+ * First, find the map entry covering "curr_s_offset", going down
+ * submaps if necessary.
+ */
+ if (!vm_map_lookup_entry(map, curr_s_offset, &map_entry)) {
+ /* no entry -> no object -> no page */
+
+ if (curr_s_offset < vm_map_min(map)) {
+ /*
+ * Illegal address that falls below map min.
+ */
+ curr_e_offset = MIN(end, vm_map_min(map));
+
+ } else if (curr_s_offset >= vm_map_max(map)) {
+ /*
+ * Illegal address that falls on/after map max.
+ */
+ curr_e_offset = end;
+
+ } else if (map_entry == vm_map_to_entry(map)) {
+ /*
+ * Hit a hole.
+ */
+ if (map_entry->vme_next == vm_map_to_entry(map)) {
+ /*
+ * Empty map.
+ */
+ curr_e_offset = MIN(map->max_offset, end);
+ } else {
+ /*
+ * Hole at start of the map.
+ */
+ curr_e_offset = MIN(map_entry->vme_next->vme_start, end);
+ }
+ } else {
+ if (map_entry->vme_next == vm_map_to_entry(map)) {
+ /*
+ * Hole at the end of the map.
+ */
+ curr_e_offset = MIN(map->max_offset, end);
+ } else {
+ curr_e_offset = MIN(map_entry->vme_next->vme_start, end);
+ }
+ }
+
+ assert(curr_e_offset >= curr_s_offset);
+
+ uint64_t num_pages = (curr_e_offset - curr_s_offset) >> PAGE_SHIFT;
+
+ void *info_ptr = (void*) (((uintptr_t) info) + (info_idx * sizeof(struct vm_page_info_basic)));
+
+ bzero(info_ptr, num_pages * sizeof(struct vm_page_info_basic));
+
+ curr_s_offset = curr_e_offset;
+
+ info_idx += num_pages;
+
+ continue;
}
+
/* compute offset from this map entry's start */
- offset -= map_entry->vme_start;
+ offset_in_object = curr_s_offset - map_entry->vme_start;
+
/* compute offset into this map entry's object (or submap) */
- offset += map_entry->offset;
+ offset_in_object += VME_OFFSET(map_entry);
if (map_entry->is_sub_map) {
- vm_map_t sub_map;
+ vm_map_t sub_map = VM_MAP_NULL;
+ vm_page_info_t submap_info = 0;
+ vm_map_offset_t submap_s_offset = 0, submap_e_offset = 0, range_len = 0;
+
+ range_len = MIN(map_entry->vme_end, end) - curr_s_offset;
- sub_map = map_entry->object.sub_map;
- vm_map_lock_read(sub_map);
+ submap_s_offset = offset_in_object;
+ submap_e_offset = submap_s_offset + range_len;
+
+ sub_map = VME_SUBMAP(map_entry);
+
+ vm_map_reference(sub_map);
vm_map_unlock_read(map);
- map = sub_map;
+ submap_info = (vm_page_info_t) (((uintptr_t) info) + (info_idx * sizeof(struct vm_page_info_basic)));
+
+ retval = vm_map_page_range_info_internal(sub_map,
+ submap_s_offset,
+ submap_e_offset,
+ VM_PAGE_INFO_BASIC,
+ (vm_page_info_t) submap_info,
+ count);
+
+ assert(retval == KERN_SUCCESS);
+
+ vm_map_lock_read(map);
+ vm_map_deallocate(sub_map);
+
+ /* Move the "info" index by the number of pages we inspected.*/
+ info_idx += range_len >> PAGE_SHIFT;
+
+ /* Move our current offset by the size of the range we inspected.*/
+ curr_s_offset += range_len;
- ref_count = MAX(ref_count, map->ref_count);
continue;
}
- break;
- }
- object = map_entry->object.vm_object;
- if (object == VM_OBJECT_NULL) {
- /* no object -> no page */
- vm_map_unlock_read(map);
- goto done;
- }
+ object = VME_OBJECT(map_entry);
+ if (object == VM_OBJECT_NULL) {
- vm_object_lock(object);
- vm_map_unlock_read(map);
+ /*
+ * We don't have an object here and, hence,
+ * no pages to inspect. We'll fill up the
+ * info structure appropriately.
+ */
- /*
- * 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);
+ curr_e_offset = MIN(map_entry->vme_end, end);
- m = vm_page_lookup(object, offset);
+ uint64_t num_pages = (curr_e_offset - curr_s_offset) >> PAGE_SHIFT;
- 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) {
+ void *info_ptr = (void*) (((uintptr_t) info) + (info_idx * sizeof(struct vm_page_info_basic)));
+
+ bzero(info_ptr, num_pages * sizeof(struct vm_page_info_basic));
+
+ curr_s_offset = curr_e_offset;
+
+ info_idx += num_pages;
+
+ continue;
+ }
+
+ if (do_region_footprint) {
+ int pmap_disp;
+
+ disposition = 0;
+ pmap_disp = 0;
+ if (map->has_corpse_footprint) {
+ /*
+ * Query the page info data we saved
+ * while forking the corpse.
+ */
+ vm_map_corpse_footprint_query_page_info(
+ map,
+ curr_s_offset,
+ &pmap_disp);
+ } else {
+ /*
+ * Query the pmap.
+ */
+ pmap_query_page_info(map->pmap,
+ curr_s_offset,
+ &pmap_disp);
+ }
+ if (object->purgable == VM_PURGABLE_NONVOLATILE &&
+ /* && not tagged as no-footprint? */
+ VM_OBJECT_OWNER(object) != NULL &&
+ VM_OBJECT_OWNER(object)->map == map) {
+ if ((((curr_s_offset
+ - map_entry->vme_start
+ + VME_OFFSET(map_entry))
+ / PAGE_SIZE) <
+ (object->resident_page_count +
+ vm_compressor_pager_get_count(object->pager)))) {
/*
- * this page has been paged out
+ * Non-volatile purgeable object owned
+ * by this task: report the first
+ * "#resident + #compressed" pages as
+ * "resident" (to show that they
+ * contribute to the footprint) but not
+ * "dirty" (to avoid double-counting
+ * with the fake "non-volatile" region
+ * we'll report at the end of the
+ * address space to account for all
+ * (mapped or not) non-volatile memory
+ * owned by this task.
*/
- disposition |= VM_PAGE_QUERY_PAGE_PAGED_OUT;
- break;
+ disposition |= VM_PAGE_QUERY_PAGE_PRESENT;
}
- } 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 if ((object->purgable == VM_PURGABLE_VOLATILE ||
+ object->purgable == VM_PURGABLE_EMPTY) &&
+ /* && not tagged as no-footprint? */
+ VM_OBJECT_OWNER(object) != NULL &&
+ VM_OBJECT_OWNER(object)->map == map) {
+ if ((((curr_s_offset
+ - map_entry->vme_start
+ + VME_OFFSET(map_entry))
+ / PAGE_SIZE) <
+ object->wired_page_count)) {
+ /*
+ * Volatile|empty purgeable object owned
+ * by this task: report the first
+ * "#wired" pages as "resident" (to
+ * show that they contribute to the
+ * footprint) but not "dirty" (to avoid
+ * double-counting with the fake
+ * "non-volatile" region we'll report
+ * at the end of the address space to
+ * account for all (mapped or not)
+ * non-volatile memory owned by this
+ * task.
+ */
+ disposition |= VM_PAGE_QUERY_PAGE_PRESENT;
+ }
+ } else if (map_entry->iokit_acct &&
+ object->internal &&
+ object->purgable == VM_PURGABLE_DENY) {
+ /*
+ * Non-purgeable IOKit memory: phys_footprint
+ * includes the entire virtual mapping.
+ */
+ assertf(!map_entry->use_pmap, "offset 0x%llx map_entry %p", (uint64_t) curr_s_offset, map_entry);
+ disposition |= VM_PAGE_QUERY_PAGE_PRESENT;
+ disposition |= VM_PAGE_QUERY_PAGE_DIRTY;
+ } else if (pmap_disp & (PMAP_QUERY_PAGE_ALTACCT |
+ PMAP_QUERY_PAGE_COMPRESSED_ALTACCT)) {
+ /* alternate accounting */
+#if CONFIG_EMBEDDED && (DEVELOPMENT || DEBUG)
+ if (map->pmap->footprint_was_suspended ||
+ /*
+ * XXX corpse does not know if original
+ * pmap had its footprint suspended...
+ */
+ map->has_corpse_footprint) {
+ /*
+ * The assertion below can fail if dyld
+ * suspended footprint accounting
+ * while doing some adjustments to
+ * this page; the mapping would say
+ * "use pmap accounting" but the page
+ * would be marked "alternate
+ * accounting".
+ */
+ } else
+#endif /* CONFIG_EMBEDDED && (DEVELOPMENT || DEBUG) */
+ assertf(!map_entry->use_pmap, "offset 0x%llx map_entry %p", (uint64_t) curr_s_offset, map_entry);
+ pmap_disp = 0;
+ } else {
+ if (pmap_disp & PMAP_QUERY_PAGE_PRESENT) {
+ assertf(map_entry->use_pmap, "offset 0x%llx map_entry %p", (uint64_t) curr_s_offset, map_entry);
+ disposition |= VM_PAGE_QUERY_PAGE_PRESENT;
+ disposition |= VM_PAGE_QUERY_PAGE_REF;
+ if (pmap_disp & PMAP_QUERY_PAGE_INTERNAL) {
+ disposition |= VM_PAGE_QUERY_PAGE_DIRTY;
+ } else {
+ disposition |= VM_PAGE_QUERY_PAGE_EXTERNAL;
}
- } else {
- memory_object_t pager;
+ } else if (pmap_disp & PMAP_QUERY_PAGE_COMPRESSED) {
+ assertf(map_entry->use_pmap, "offset 0x%llx map_entry %p", (uint64_t) curr_s_offset, map_entry);
+ disposition |= VM_PAGE_QUERY_PAGE_PAGED_OUT;
+ }
+ }
+ switch (flavor) {
+ case VM_PAGE_INFO_BASIC:
+ basic_info = (vm_page_info_basic_t) (((uintptr_t) info) + (info_idx * sizeof(struct vm_page_info_basic)));
+ basic_info->disposition = disposition;
+ basic_info->ref_count = 1;
+ basic_info->object_id = INFO_MAKE_FAKE_OBJECT_ID(map, task_ledgers.purgeable_nonvolatile);
+ basic_info->offset = 0;
+ basic_info->depth = 0;
+
+ info_idx++;
+ break;
+ }
+ curr_s_offset += PAGE_SIZE;
+ continue;
+ }
- vm_object_paging_begin(object);
- pager = object->pager;
- vm_object_unlock(object);
+ vm_object_reference(object);
+ /*
+ * Shared mode -- so we can allow other readers
+ * to grab the lock too.
+ */
+ vm_object_lock_shared(object);
+
+ curr_e_offset = MIN(map_entry->vme_end, end);
+
+ vm_map_unlock_read(map);
+
+ map_entry = NULL; /* map is unlocked, the entry is no longer valid. */
+
+ curr_object = object;
+
+ for (; curr_s_offset < curr_e_offset;) {
+
+ if (object == curr_object) {
+ ref_count = curr_object->ref_count - 1; /* account for our object reference above. */
+ } else {
+ ref_count = curr_object->ref_count;
+ }
+
+ curr_offset_in_object = offset_in_object;
+
+ for (;;) {
+ m = vm_page_lookup(curr_object, curr_offset_in_object);
+
+ if (m != VM_PAGE_NULL) {
+
+ disposition |= VM_PAGE_QUERY_PAGE_PRESENT;
+ break;
+ } else {
+ if (curr_object->internal &&
+ curr_object->alive &&
+ !curr_object->terminating &&
+ curr_object->pager_ready) {
+
+ if (VM_COMPRESSOR_PAGER_STATE_GET(curr_object, curr_offset_in_object)
+ == VM_EXTERNAL_STATE_EXISTS) {
+ /* the pager has that page */
+ disposition |= VM_PAGE_QUERY_PAGE_PAGED_OUT;
+ break;
+ }
+ }
+
/*
- * Ask the default pager if
- * it has this page.
+ * Go down the VM object shadow chain until we find the page
+ * we're looking for.
*/
- 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 (curr_object->shadow != VM_OBJECT_NULL) {
+ vm_object_t shadow = VM_OBJECT_NULL;
+
+ curr_offset_in_object += curr_object->vo_shadow_offset;
+ shadow = curr_object->shadow;
+
+ vm_object_lock_shared(shadow);
+ vm_object_unlock(curr_object);
+
+ curr_object = shadow;
+ depth++;
+ continue;
+ } else {
- if (kr == KERN_SUCCESS) {
- /* the default pager has it */
- disposition |= VM_PAGE_QUERY_PAGE_PAGED_OUT;
break;
}
}
}
- if (object->shadow != VM_OBJECT_NULL) {
- vm_object_t shadow;
+ /* 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 */
- offset += object->vo_shadow_offset;
- shadow = object->shadow;
-
- vm_object_lock(shadow);
- vm_object_unlock(object);
+ /* 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. */
- object = shadow;
- top_object = FALSE;
- depth++;
- } else {
-// if (!object->internal)
-// break;
-// retval = KERN_FAILURE;
-// goto done_with_object;
- break;
+ if ((curr_object == object) && curr_object->shadow)
+ disposition |= VM_PAGE_QUERY_PAGE_COPIED;
+
+ if (! curr_object->internal)
+ disposition |= VM_PAGE_QUERY_PAGE_EXTERNAL;
+
+ if (m != VM_PAGE_NULL) {
+
+ if (m->vmp_fictitious) {
+
+ disposition |= VM_PAGE_QUERY_PAGE_FICTITIOUS;
+
+ } else {
+ if (m->vmp_dirty || pmap_is_modified(VM_PAGE_GET_PHYS_PAGE(m)))
+ disposition |= VM_PAGE_QUERY_PAGE_DIRTY;
+
+ if (m->vmp_reference || pmap_is_referenced(VM_PAGE_GET_PHYS_PAGE(m)))
+ disposition |= VM_PAGE_QUERY_PAGE_REF;
+
+ if (m->vmp_q_state == VM_PAGE_ON_SPECULATIVE_Q)
+ disposition |= VM_PAGE_QUERY_PAGE_SPECULATIVE;
+
+ if (m->vmp_cs_validated)
+ disposition |= VM_PAGE_QUERY_PAGE_CS_VALIDATED;
+ if (m->vmp_cs_tainted)
+ disposition |= VM_PAGE_QUERY_PAGE_CS_TAINTED;
+ if (m->vmp_cs_nx)
+ disposition |= VM_PAGE_QUERY_PAGE_CS_NX;
+ }
}
- }
- }
- /* 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 */
- /* 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. */
+ switch (flavor) {
+ case VM_PAGE_INFO_BASIC:
+ basic_info = (vm_page_info_basic_t) (((uintptr_t) info) + (info_idx * sizeof(struct vm_page_info_basic)));
+ basic_info->disposition = disposition;
+ basic_info->ref_count = ref_count;
+ basic_info->object_id = (vm_object_id_t) (uintptr_t)
+ VM_KERNEL_ADDRPERM(curr_object);
+ basic_info->offset =
+ (memory_object_offset_t) curr_offset_in_object + offset_in_page;
+ basic_info->depth = depth;
+
+ info_idx++;
+ break;
+ }
- if (top_object == TRUE && object->shadow)
- disposition |= VM_PAGE_QUERY_PAGE_COPIED;
+ disposition = 0;
+ offset_in_page = 0; // This doesn't really make sense for any offset other than the starting offset.
- if (! object->internal)
- disposition |= VM_PAGE_QUERY_PAGE_EXTERNAL;
+ /*
+ * Move to next offset in the range and in our object.
+ */
+ curr_s_offset += PAGE_SIZE;
+ offset_in_object += PAGE_SIZE;
+ curr_offset_in_object = offset_in_object;
- if (m == VM_PAGE_NULL)
- goto done_with_object;
+ if (curr_object != object) {
- 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;
+ vm_object_unlock(curr_object);
- if (m->reference || pmap_is_referenced(m->phys_page))
- disposition |= VM_PAGE_QUERY_PAGE_REF;
+ curr_object = object;
- if (m->speculative)
- disposition |= VM_PAGE_QUERY_PAGE_SPECULATIVE;
+ vm_object_lock_shared(curr_object);
+ } else {
- if (m->cs_validated)
- disposition |= VM_PAGE_QUERY_PAGE_CS_VALIDATED;
- if (m->cs_tainted)
- disposition |= VM_PAGE_QUERY_PAGE_CS_TAINTED;
+ vm_object_lock_yield_shared(curr_object);
+ }
+ }
-done_with_object:
- vm_object_unlock(object);
-done:
+ vm_object_unlock(curr_object);
+ vm_object_deallocate(curr_object);
- 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;
+ vm_map_lock_read(map);
}
+ vm_map_unlock_read(map);
return retval;
}
vm_map_size_t size,
vm_sync_t sync_flags)
{
- 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;
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);
if (size == 0)
return(KERN_SUCCESS);
- queue_init(&req_q);
amount_left = size;
while (amount_left > 0) {
vm_map_lock(map);
if (!vm_map_lookup_entry(map,
- vm_map_trunc_page(
- address,
- VM_MAP_PAGE_MASK(map)),
+ address,
&entry)) {
vm_map_size_t skip;
*/
had_hole = TRUE;
+ if (sync_flags & VM_SYNC_KILLPAGES) {
+ /*
+ * For VM_SYNC_KILLPAGES, there should be
+ * no holes in the range, since we couldn't
+ * prevent someone else from allocating in
+ * that hole and we wouldn't want to "kill"
+ * their pages.
+ */
+ vm_map_unlock(map);
+ break;
+ }
+
/*
* Check for empty map.
*/
}
offset = address - entry->vme_start;
+ pmap_offset = address;
/*
* do we have more to flush than is contained in this
vm_map_t local_map;
vm_map_offset_t local_offset;
- local_map = entry->object.sub_map;
- local_offset = entry->offset;
+ local_map = VME_SUBMAP(entry);
+ local_offset = VME_OFFSET(entry);
vm_map_unlock(map);
if (vm_map_msync(
local_map,
}
continue;
}
- object = entry->object.vm_object;
+ object = VME_OBJECT(entry);
/*
* We can't sync this object if the object has not been
vm_map_unlock(map);
continue;
}
- offset += entry->offset;
+ offset += VME_OFFSET(entry);
vm_object_lock(object);
boolean_t reusable_pages = FALSE;
if (sync_flags & VM_SYNC_KILLPAGES) {
- if (object->ref_count == 1 && !object->shadow)
+ if (((object->ref_count == 1) ||
+ ((object->copy_strategy !=
+ MEMORY_OBJECT_COPY_SYMMETRIC) &&
+ (object->copy == VM_OBJECT_NULL))) &&
+ (object->shadow == VM_OBJECT_NULL)) {
+ if (object->ref_count != 1) {
+ vm_page_stats_reusable.free_shared++;
+ }
kill_pages = 1;
- else
+ } else {
kill_pages = -1;
+ }
}
if (kill_pages != -1)
- vm_object_deactivate_pages(object, offset,
- (vm_object_size_t)flush_size, kill_pages, reusable_pages);
+ 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;
((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);
-
- 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:
-
- /*
- * 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.
- */
-
- pager = object->pager;
-
- 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) {
- /*
- * need to check for overlapping entry, if found, wait
- * on overlapping msr to be done, then reiterate
- */
- 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;
- }
- 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 */
+ if ((sync_flags & VM_SYNC_INVALIDATE) && object->resident_page_count == 0) {
+ /*
+ * clear out the clustering and read-ahead hints
+ */
+ vm_object_lock(object);
- /*
- * wait for memory_object_sychronize_completed messages from pager(s)
- */
+ object->pages_created = 0;
+ object->pages_used = 0;
+ object->sequential = 0;
+ object->last_alloc = 0;
- 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 */
+ vm_object_unlock(object);
+ }
+ vm_object_deallocate(object);
+ } /* while */
/* for proper msync() behaviour */
if (had_hole == TRUE && (sync_flags & VM_SYNC_CONTIGUOUS))
if(IP_VALID(port) && (ip_kotype(port) == IKOT_NAMED_ENTRY)) {
while(TRUE) {
ip_lock(port);
- if(ip_active(port) && (ip_kotype(port)
+ if(ip_active(port) && (ip_kotype(port)
== IKOT_NAMED_ENTRY)) {
named_entry =
(vm_named_entry_t)port->ip_kobject;
lck_mtx_unlock(&(named_entry)->Lock);
ip_unlock(port);
if ((named_entry->is_sub_map) &&
- (named_entry->protection
+ (named_entry->protection
& VM_PROT_WRITE)) {
map = named_entry->backing.map;
} else {
mach_destroy_memory_entry(port);
break;
}
- else
+ else
return VM_MAP_NULL;
}
}
* 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.
+ * which may be null.
* Conditions:
* Nothing locked.
*/
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;
#undef vm_map_reference
void
vm_map_reference(
- register vm_map_t map)
+ vm_map_t map)
{
if (map == VM_MAP_NULL)
return;
lck_mtx_lock(&map->s_lock);
#if TASK_SWAPPER
assert(map->res_count > 0);
- assert(map->ref_count >= map->res_count);
+ assert(map->map_refcnt >= map->res_count);
map->res_count++;
#endif
- map->ref_count++;
+ map->map_refcnt++;
lck_mtx_unlock(&map->s_lock);
}
*/
void
vm_map_deallocate(
- register vm_map_t map)
+ vm_map_t map)
{
unsigned int ref;
return;
lck_mtx_lock(&map->s_lock);
- ref = --map->ref_count;
+ ref = --map->map_refcnt;
if (ref > 0) {
vm_map_res_deallocate(map);
lck_mtx_unlock(&map->s_lock);
return;
}
- assert(map->ref_count == 0);
+ assert(map->map_refcnt == 0);
lck_mtx_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,
+ * 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_destroy(map, VM_MAP_NO_FLAGS);
+ vm_map_destroy(map, VM_MAP_REMOVE_NO_FLAGS);
}
void
vm_map_set_32bit(vm_map_t map)
{
+#if defined(__arm__) || defined(__arm64__)
+ map->max_offset = pmap_max_offset(FALSE, ARM_PMAP_MAX_OFFSET_DEVICE);
+#else
map->max_offset = (vm_map_offset_t)VM_MAX_ADDRESS;
+#endif
}
void
vm_map_set_64bit(vm_map_t map)
{
+#if defined(__arm__) || defined(__arm64__)
+ map->max_offset = pmap_max_offset(TRUE, ARM_PMAP_MAX_OFFSET_DEVICE);
+#else
map->max_offset = (vm_map_offset_t)MACH_VM_MAX_ADDRESS;
+#endif
+}
+
+/*
+ * Expand the maximum size of an existing map to the maximum supported.
+ */
+void
+vm_map_set_jumbo(vm_map_t map)
+{
+#if defined (__arm64__)
+ vm_map_set_max_addr(map, ~0);
+#else /* arm64 */
+ (void) map;
+#endif
+}
+
+/*
+ * Expand the maximum size of an existing map.
+ */
+void
+vm_map_set_max_addr(vm_map_t map, vm_map_offset_t new_max_offset)
+{
+#if defined(__arm64__)
+ vm_map_offset_t max_supported_offset = 0;
+ vm_map_offset_t old_max_offset = map->max_offset;
+ max_supported_offset = pmap_max_offset(vm_map_is_64bit(map), ARM_PMAP_MAX_OFFSET_JUMBO);
+
+ new_max_offset = trunc_page(new_max_offset);
+
+ /* The address space cannot be shrunk using this routine. */
+ if (old_max_offset >= new_max_offset) {
+ return;
+ }
+
+ if (max_supported_offset < new_max_offset) {
+ new_max_offset = max_supported_offset;
+ }
+
+ map->max_offset = new_max_offset;
+
+ if (map->holes_list->prev->vme_end == old_max_offset) {
+ /*
+ * There is already a hole at the end of the map; simply make it bigger.
+ */
+ map->holes_list->prev->vme_end = map->max_offset;
+ } else {
+ /*
+ * There is no hole at the end, so we need to create a new hole
+ * for the new empty space we're creating.
+ */
+ struct vm_map_links *new_hole = zalloc(vm_map_holes_zone);
+ new_hole->start = old_max_offset;
+ new_hole->end = map->max_offset;
+ new_hole->prev = map->holes_list->prev;
+ new_hole->next = (struct vm_map_entry *)map->holes_list;
+ map->holes_list->prev->links.next = (struct vm_map_entry *)new_hole;
+ map->holes_list->prev = (struct vm_map_entry *)new_hole;
+ }
+#else
+ (void)map;
+ (void)new_max_offset;
+#endif
}
vm_map_offset_t
-vm_compute_max_offset(unsigned is64)
+vm_compute_max_offset(boolean_t is64)
{
+#if defined(__arm__) || defined(__arm64__)
+ return (pmap_max_offset(is64, ARM_PMAP_MAX_OFFSET_DEVICE));
+#else
return (is64 ? (vm_map_offset_t)MACH_VM_MAX_ADDRESS : (vm_map_offset_t)VM_MAX_ADDRESS);
+#endif
+}
+
+void
+vm_map_get_max_aslr_slide_section(
+ vm_map_t map __unused,
+ int64_t *max_sections,
+ int64_t *section_size)
+{
+#if defined(__arm64__)
+ *max_sections = 3;
+ *section_size = ARM_TT_TWIG_SIZE;
+#else
+ *max_sections = 1;
+ *section_size = 0;
+#endif
+}
+
+uint64_t
+vm_map_get_max_aslr_slide_pages(vm_map_t map)
+{
+#if defined(__arm64__)
+ /* Limit arm64 slide to 16MB to conserve contiguous VA space in the more
+ * limited embedded address space; this is also meant to minimize pmap
+ * memory usage on 16KB page systems.
+ */
+ return (1 << (24 - VM_MAP_PAGE_SHIFT(map)));
+#else
+ return (1 << (vm_map_is_64bit(map) ? 16 : 8));
+#endif
}
uint64_t
-vm_map_get_max_aslr_slide_pages(vm_map_t map)
+vm_map_get_max_loader_aslr_slide_pages(vm_map_t map)
{
+#if defined(__arm64__)
+ /* We limit the loader slide to 4MB, in order to ensure at least 8 bits
+ * of independent entropy on 16KB page systems.
+ */
+ return (1 << (22 - VM_MAP_PAGE_SHIFT(map)));
+#else
return (1 << (vm_map_is_64bit(map) ? 16 : 8));
+#endif
}
+#ifndef __arm__
boolean_t
vm_map_is_64bit(
vm_map_t map)
{
return map->max_offset > ((vm_map_offset_t)VM_MAX_ADDRESS);
}
+#endif
boolean_t
vm_map_has_hard_pagezero(
return (map->min_offset >= pagezero_size);
}
-void
-vm_map_set_4GB_pagezero(vm_map_t map)
-{
-#pragma unused(map)
-
-}
-
-void
-vm_map_clear_4GB_pagezero(vm_map_t map)
-{
-#pragma unused(map)
-}
-
/*
* Raise a VM map's maximun offset.
*/
ret = KERN_INVALID_ADDRESS;
if (new_max_offset >= map->max_offset) {
- if (!vm_map_is_64bit(map)) {
+ 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;
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) &&
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;
}
-void vm_map_switch_protect(vm_map_t map,
- boolean_t val)
+void vm_map_switch_protect(vm_map_t map,
+ boolean_t val)
{
vm_map_lock(map);
map->switch_protect=val;
{
pmap_t pmap = vm_map_pmap(map);
- ledger_credit(pmap->ledger, task_ledgers.iokit_mem, bytes);
- ledger_credit(pmap->ledger, task_ledgers.phys_footprint, bytes);
+ ledger_credit(pmap->ledger, task_ledgers.iokit_mapped, bytes);
+ ledger_credit(pmap->ledger, task_ledgers.phys_footprint, bytes);
}
void
{
pmap_t pmap = vm_map_pmap(map);
- ledger_debit(pmap->ledger, task_ledgers.iokit_mem, bytes);
- ledger_debit(pmap->ledger, task_ledgers.phys_footprint, bytes);
+ 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,
+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;
-
+
/*
* 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);
-
+
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
vm_map_unlock_read(map);
return(KERN_INVALID_ARGUMENT);
}
-
- object = entry->object.vm_object;
+
+ 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;
-
- m = vm_page_lookup(object, start - entry->vme_start + entry->offset );
+
+ 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
+ /* 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))) {
+ if (m->vmp_busy ||
+ (m->vmp_unusual && (m->vmp_error || m->vmp_restart || m->vmp_private || m->vmp_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
+ /* This is the place where we'll call out to create a code
* directory, later */
- m->cs_validated = TRUE;
+ m->vmp_cs_validated = TRUE;
/* The page is now "clean" for codesigning purposes. That means
- * we don't consider it as modified (wpmapped) anymore. But
+ * 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);
-
- /* Pull the dirty status from the pmap, since we cleared the
+ m->vmp_wpmapped = FALSE;
+ refmod = pmap_disconnect(VM_PAGE_GET_PHYS_PAGE(m));
+
+ /* Pull the dirty status from the pmap, since we cleared the
* wpmapped bit */
- if ((refmod & VM_MEM_MODIFIED) && !m->dirty) {
+ if ((refmod & VM_MEM_MODIFIED) && !m->vmp_dirty) {
SET_PAGE_DIRTY(m, FALSE);
}
-
+
/* On to the next page */
start += PAGE_SIZE;
}
vm_object_unlock(object);
-
+
return KERN_SUCCESS;
}
#endif
-#if CONFIG_FREEZE
+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;
-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_lock(map);
+
+ /*
+ * 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 (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);
+
+ 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)) {
+
+ *reclaimed_resident += VME_OBJECT(entry)->resident_page_count;
+ *reclaimed_compressed += vm_compressor_pager_get_count(VME_OBJECT(entry)->pager);
+
+ (void)vm_map_delete(map,
+ entry->vme_start,
+ entry->vme_end,
+ VM_MAP_REMOVE_SAVE_ENTRIES,
+ zap_map);
+ }
+ }
+
+ 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;
+ }
+
+ return kr;
+}
+
+
+#if DEVELOPMENT || DEBUG
+
+int
+vm_map_disconnect_page_mappings(
+ vm_map_t map,
+ boolean_t do_unnest)
{
vm_map_entry_t entry;
-
+ int page_count = 0;
+
+ if (do_unnest == TRUE) {
+#ifndef NO_NESTED_PMAP
+ vm_map_lock(map);
+
+ for (entry = vm_map_first_entry(map);
+ entry != vm_map_to_entry(map);
+ entry = entry->vme_next) {
+
+ if (entry->is_sub_map && entry->use_pmap) {
+ /*
+ * Make sure the range between the start of this entry and
+ * the end of this entry is no longer nested, so that
+ * we will only remove mappings from the pmap in use by this
+ * this task
+ */
+ vm_map_clip_unnest(map, entry, entry->vme_start, entry->vme_end);
+ }
+ }
+ vm_map_unlock(map);
+#endif
+ }
vm_map_lock_read(map);
-
- *purgeable_count = *wired_count = *clean_count = *dirty_count = 0;
- *has_shared = FALSE;
-
+
+ page_count = map->pmap->stats.resident_count;
+
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 ((entry->object.vm_object == 0) ||
- (entry->is_sub_map) ||
- (entry->object.vm_object->phys_contiguous)) {
+ if (!entry->is_sub_map && ((VME_OBJECT(entry) == 0) ||
+ (VME_OBJECT(entry)->phys_contiguous))) {
continue;
}
+ if (entry->is_sub_map)
+ assert(!entry->use_pmap);
- default_freezer_pack(&purgeable, &wired, &clean, &dirty, dirty_budget, &shared, entry->object.vm_object, 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;
- }
- }
+ pmap_remove_options(map->pmap, entry->vme_start, entry->vme_end, 0);
}
-
vm_map_unlock_read(map);
- return KERN_SUCCESS;
+ return page_count;
}
-kern_return_t vm_map_freeze(
+#endif
+
+
+#if CONFIG_FREEZE
+
+
+int c_freezer_swapout_page_count;
+int c_freezer_compression_count = 0;
+AbsoluteTime c_freezer_last_yield_ts = 0;
+
+extern unsigned int memorystatus_freeze_private_shared_pages_ratio;
+extern unsigned int memorystatus_freeze_shared_mb_per_process_max;
+
+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)
-{
+ __unused unsigned int dirty_budget,
+ unsigned int *shared_count,
+ int *freezer_error_code,
+ boolean_t eval_only)
+{
vm_map_entry_t entry2 = VM_MAP_ENTRY_NULL;
kern_return_t kr = KERN_SUCCESS;
- boolean_t default_freezer_active = TRUE;
+ boolean_t evaluation_phase = TRUE;
+ vm_object_t cur_shared_object = NULL;
+ int cur_shared_obj_ref_cnt = 0;
+ unsigned int dirty_private_count = 0, dirty_shared_count = 0, obj_pages_snapshot = 0;
- *purgeable_count = *wired_count = *clean_count = *dirty_count = 0;
- *has_shared = FALSE;
+ *purgeable_count = *wired_count = *clean_count = *dirty_count = *shared_count = 0;
/*
* We need the exclusive lock here so that we can
*/
vm_map_lock(map);
- if (COMPRESSED_PAGER_IS_ACTIVE || DEFAULT_FREEZER_COMPRESSED_PAGER_IS_ACTIVE) {
- default_freezer_active = FALSE;
- }
-
- if (default_freezer_active) {
- if (map->default_freezer_handle == NULL) {
- map->default_freezer_handle = default_freezer_handle_allocate();
+ assert(VM_CONFIG_COMPRESSOR_IS_PRESENT);
+
+ if (vm_compressor_low_on_space() || vm_swap_low_on_space()) {
+ if (vm_compressor_low_on_space()) {
+ *freezer_error_code = FREEZER_ERROR_NO_COMPRESSOR_SPACE;
}
-
- if ((kr = default_freezer_handle_init(map->default_freezer_handle)) != KERN_SUCCESS) {
+
+ if (vm_swap_low_on_space()) {
+ *freezer_error_code = FREEZER_ERROR_NO_SWAP_SPACE;
+ }
+
+ kr = KERN_NO_SPACE;
+ goto done;
+ }
+
+ if (VM_CONFIG_FREEZER_SWAP_IS_ACTIVE == FALSE) {
+ /*
+ * In-memory compressor backing the freezer. No disk.
+ * So no need to do the evaluation phase.
+ */
+ evaluation_phase = FALSE;
+
+ if (eval_only == TRUE) {
/*
- * 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.
+ * We don't support 'eval_only' mode
+ * in this non-swap config.
*/
+ *freezer_error_code = FREEZER_ERROR_GENERIC;
+ kr = KERN_INVALID_ARGUMENT;
goto done;
}
+
+ c_freezer_compression_count = 0;
+ clock_get_uptime(&c_freezer_last_yield_ts);
}
-
+again:
+
for (entry2 = vm_map_first_entry(map);
entry2 != vm_map_to_entry(map);
entry2 = entry2->vme_next) {
-
- vm_object_t src_object = entry2->object.vm_object;
- if (entry2->object.vm_object && !entry2->is_sub_map && !entry2->object.vm_object->phys_contiguous) {
+ vm_object_t src_object = VME_OBJECT(entry2);
+
+ if (src_object &&
+ !entry2->is_sub_map &&
+ !src_object->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;
+
+ if (src_object->internal == TRUE) {
+
+ if (VM_CONFIG_FREEZER_SWAP_IS_ACTIVE) {
+ /*
+ * 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.
+ *
+ * We try to optimize somewhat by checking for objects that are mapped
+ * more than once within our own map. But we don't do full searches,
+ * we just look at the entries following our current entry.
+ */
+ if (src_object->ref_count > 1) {
+ if (src_object != cur_shared_object) {
+ obj_pages_snapshot = (src_object->resident_page_count - src_object->wired_page_count) + vm_compressor_pager_get_count(src_object->pager);
+ dirty_shared_count += obj_pages_snapshot;
+
+ cur_shared_object = src_object;
+ cur_shared_obj_ref_cnt = 1;
+ continue;
+ } else {
+ cur_shared_obj_ref_cnt++;
+ if (src_object->ref_count == cur_shared_obj_ref_cnt) {
+ /*
+ * Fall through to below and treat this object as private.
+ * So deduct its pages from our shared total and add it to the
+ * private total.
+ */
+
+ dirty_shared_count -= obj_pages_snapshot;
+ dirty_private_count += obj_pages_snapshot;
+ } else {
+ continue;
+ }
+ }
}
- }
- if (shared) {
- *has_shared = TRUE;
+
+ if (src_object->ref_count == 1) {
+ dirty_private_count += (src_object->resident_page_count - src_object->wired_page_count) + vm_compressor_pager_get_count(src_object->pager);
+ }
+
+ if (evaluation_phase == TRUE) {
+
+ continue;
+ }
}
- } else {
- /*
- * To the compressor.
- */
- if (entry2->object.vm_object->internal == TRUE) {
- vm_object_pageout(entry2->object.vm_object);
+
+ vm_object_compressed_freezer_pageout(src_object);
+
+ *wired_count += src_object->wired_page_count;
+
+ if (vm_compressor_low_on_space() || vm_swap_low_on_space()) {
+ if (vm_compressor_low_on_space()) {
+ *freezer_error_code = FREEZER_ERROR_NO_COMPRESSOR_SPACE;
+ }
+
+ if (vm_swap_low_on_space()) {
+ *freezer_error_code = FREEZER_ERROR_NO_SWAP_SPACE;
+ }
+
+ kr = KERN_NO_SPACE;
+ break;
}
}
}
}
- if (default_freezer_active) {
- /* Finally, throw out the pages to swap */
- default_freezer_pageout(map->default_freezer_handle);
- }
+ if (evaluation_phase) {
-done:
- vm_map_unlock(map);
+ unsigned int shared_pages_threshold = (memorystatus_freeze_shared_mb_per_process_max * 1024 * 1024ULL) / PAGE_SIZE_64;
+
+ if (dirty_shared_count > shared_pages_threshold) {
+ *freezer_error_code = FREEZER_ERROR_EXCESS_SHARED_MEMORY;
+ kr = KERN_FAILURE;
+ goto done;
+ }
+
+ if (dirty_shared_count &&
+ ((dirty_private_count / dirty_shared_count) < memorystatus_freeze_private_shared_pages_ratio)) {
+ *freezer_error_code = FREEZER_ERROR_LOW_PRIVATE_SHARED_RATIO;
+ kr = KERN_FAILURE;
+ goto done;
+ }
+
+ evaluation_phase = FALSE;
+ dirty_shared_count = dirty_private_count = 0;
- return kr;
-}
+ c_freezer_compression_count = 0;
+ clock_get_uptime(&c_freezer_last_yield_ts);
-kern_return_t
-vm_map_thaw(
- vm_map_t map)
-{
- kern_return_t kr = KERN_SUCCESS;
+ if (eval_only) {
+ kr = KERN_SUCCESS;
+ goto done;
+ }
- 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;
- }
+ goto again;
- vm_map_lock(map);
+ } else {
- if (map->default_freezer_handle == NULL) {
- /*
- * This map is not in a frozen state.
- */
- kr = KERN_FAILURE;
- goto out;
+ kr = KERN_SUCCESS;
+ *shared_count = (unsigned int) ((dirty_shared_count * PAGE_SIZE_64) / (1024 * 1024ULL));
}
- kr = default_freezer_unpack(map->default_freezer_handle);
-out:
- vm_map_unlock(map);
-
+done:
+ vm_map_unlock(map);
+
+ if ((eval_only == FALSE) && (kr == KERN_SUCCESS)) {
+ vm_object_compressed_freezer_done();
+
+ if (VM_CONFIG_FREEZER_SWAP_IS_ACTIVE) {
+ /*
+ * reset the counter tracking the # of swapped compressed pages
+ * because we are now done with this freeze session and task.
+ */
+
+ *dirty_count = c_freezer_swapout_page_count; //used to track pageouts
+ c_freezer_swapout_page_count = 0;
+ }
+ }
return kr;
}
+
#endif
/*
* - copy_strategy == MEMORY_OBJECT_COPY_SYMMETRIC
* - !true_share
* - vo_size == ANON_CHUNK_SIZE
+ *
+ * Only non-kernel map entries.
*/
boolean_t
vm_map_entry_should_cow_for_true_share(
return FALSE;
}
- if (entry->alias != VM_MEMORY_MALLOC) {
- /* not tagged as an ObjectiveC's Garbage Collector entry */
+ 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;
}
if (entry->wired_count) {
/* wired: can't change the map entry... */
+ vm_counters.should_cow_but_wired++;
return FALSE;
}
- object = entry->object.vm_object;
+ object = VME_OBJECT(entry);
if (object == VM_OBJECT_NULL) {
/* no object yet... */
return FALSE;
}
- if (object->vo_size != ANON_CHUNK_SIZE) {
- /* not an object created for the ObjC Garbage Collector */
+ if (VME_ALIAS(entry) == VM_MEMORY_MALLOC &&
+ object->vo_size != ANON_CHUNK_SIZE) {
+ /* ... not an object created for the ObjC Garbage Collector */
+ return FALSE;
+ }
+
+ if (VME_ALIAS(entry) == VM_MEMORY_MALLOC_SMALL &&
+ object->vo_size != 2048 * 4096) {
+ /* ... not a "MALLOC_SMALL" heap */
return FALSE;
}
return TRUE;
}
-vm_map_offset_t
+vm_map_offset_t
vm_map_round_page_mask(
vm_map_offset_t offset,
vm_map_offset_t mask)
return VM_MAP_ROUND_PAGE(offset, mask);
}
-vm_map_offset_t
+vm_map_offset_t
vm_map_trunc_page_mask(
vm_map_offset_t offset,
vm_map_offset_t mask)
return VM_MAP_TRUNC_PAGE(offset, mask);
}
+boolean_t
+vm_map_page_aligned(
+ vm_map_offset_t offset,
+ vm_map_offset_t mask)
+{
+ return ((offset) & mask) == 0;
+}
+
int
vm_map_page_shift(
vm_map_t map)
return VM_MAP_PAGE_SIZE(map);
}
-int
+vm_map_offset_t
vm_map_page_mask(
vm_map_t map)
{
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_pmap_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)
{
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;
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 = entry->object.vm_object;
+ object = VME_OBJECT(entry);
if (object == VM_OBJECT_NULL) {
continue;
}
- if (object->purgable != VM_PURGABLE_VOLATILE) {
+ if (object->purgable != VM_PURGABLE_VOLATILE &&
+ object->purgable != VM_PURGABLE_EMPTY) {
continue;
}
- if (entry->offset != 0) {
+ 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
continue;
}
resident_count = object->resident_page_count;
- if ((entry->offset / PAGE_SIZE) >= resident_count) {
+ if ((VME_OFFSET(entry) / PAGE_SIZE) >= resident_count) {
resident_count = 0;
} else {
- resident_count -= (entry->offset / PAGE_SIZE);
+ resident_count -= (VME_OFFSET(entry) / PAGE_SIZE);
}
volatile_virtual_size += entry->vme_end - entry->vme_start;
volatile_resident_count += resident_count;
- volatile_pmap_count += pmap_query_resident(map->pmap,
- entry->vme_start,
- entry->vme_end);
+ 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;
+}
+
+void
+vm_map_sizes(vm_map_t map,
+ vm_map_size_t * psize,
+ vm_map_size_t * pfree,
+ vm_map_size_t * plargest_free)
+{
+ vm_map_entry_t entry;
+ vm_map_offset_t prev;
+ vm_map_size_t free, total_free, largest_free;
+ boolean_t end;
+
+ if (!map)
+ {
+ *psize = *pfree = *plargest_free = 0;
+ return;
+ }
+ 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;
+}
+
+#if VM_SCAN_FOR_SHADOW_CHAIN
+int vm_map_shadow_max(vm_map_t map);
+int vm_map_shadow_max(
+ vm_map_t map)
+{
+ int shadows, shadows_max;
+ vm_map_entry_t entry;
+ vm_object_t object, next_object;
+
+ if (map == NULL)
+ return 0;
+
+ 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;
+ }
+ }
+
+ vm_map_unlock_read(map);
+
+ return shadows_max;
+}
+#endif /* VM_SCAN_FOR_SHADOW_CHAIN */
+
+void vm_commit_pagezero_status(vm_map_t lmap) {
+ pmap_advise_pagezero_range(lmap->pmap, lmap->min_offset);
+}
+
+#if __x86_64__
+void
+vm_map_set_high_start(
+ vm_map_t map,
+ vm_map_offset_t high_start)
+{
+ map->vmmap_high_start = high_start;
+}
+#endif /* __x86_64__ */
+
+#if PMAP_CS
+kern_return_t
+vm_map_entry_cs_associate(
+ vm_map_t map,
+ vm_map_entry_t entry,
+ vm_map_kernel_flags_t vmk_flags)
+{
+ vm_object_t cs_object, cs_shadow;
+ vm_object_offset_t cs_offset;
+ void *cs_blobs;
+ struct vnode *cs_vnode;
+ kern_return_t cs_ret;
+
+ if (map->pmap == NULL ||
+ entry->is_sub_map || /* XXX FBDP: recurse on sub-range? */
+ VME_OBJECT(entry) == VM_OBJECT_NULL ||
+ ! (entry->protection & VM_PROT_EXECUTE)) {
+ return KERN_SUCCESS;
+ }
+
+ vm_map_lock_assert_exclusive(map);
+
+ if (entry->used_for_jit) {
+ cs_ret = pmap_cs_associate(map->pmap,
+ PMAP_CS_ASSOCIATE_JIT,
+ entry->vme_start,
+ entry->vme_end - entry->vme_start);
+ goto done;
+ }
+
+ if (vmk_flags.vmkf_remap_prot_copy) {
+ cs_ret = pmap_cs_associate(map->pmap,
+ PMAP_CS_ASSOCIATE_COW,
+ entry->vme_start,
+ entry->vme_end - entry->vme_start);
+ goto done;
+ }
+
+ vm_object_lock_shared(VME_OBJECT(entry));
+ cs_offset = VME_OFFSET(entry);
+ for (cs_object = VME_OBJECT(entry);
+ (cs_object != VM_OBJECT_NULL &&
+ !cs_object->code_signed);
+ cs_object = cs_shadow) {
+ cs_shadow = cs_object->shadow;
+ if (cs_shadow != VM_OBJECT_NULL) {
+ cs_offset += cs_object->vo_shadow_offset;
+ vm_object_lock_shared(cs_shadow);
+ }
+ vm_object_unlock(cs_object);
+ }
+ if (cs_object == VM_OBJECT_NULL) {
+ return KERN_SUCCESS;
+ }
+
+ cs_offset += cs_object->paging_offset;
+ cs_vnode = vnode_pager_lookup_vnode(cs_object->pager);
+ cs_ret = vnode_pager_get_cs_blobs(cs_vnode,
+ &cs_blobs);
+ assert(cs_ret == KERN_SUCCESS);
+ cs_ret = cs_associate_blob_with_mapping(map->pmap,
+ entry->vme_start,
+ (entry->vme_end -
+ entry->vme_start),
+ cs_offset,
+ cs_blobs);
+ vm_object_unlock(cs_object);
+ cs_object = VM_OBJECT_NULL;
+
+ done:
+ if (cs_ret == KERN_SUCCESS) {
+ DTRACE_VM2(vm_map_entry_cs_associate_success,
+ vm_map_offset_t, entry->vme_start,
+ vm_map_offset_t, entry->vme_end);
+ if (vm_map_executable_immutable) {
+ /*
+ * Prevent this executable
+ * mapping from being unmapped
+ * or modified.
+ */
+ entry->permanent = TRUE;
+ }
+ /*
+ * pmap says it will validate the
+ * code-signing validity of pages
+ * faulted in via this mapping, so
+ * this map entry should be marked so
+ * that vm_fault() bypasses code-signing
+ * validation for faults coming through
+ * this mapping.
+ */
+ entry->pmap_cs_associated = TRUE;
+ } else if (cs_ret == KERN_NOT_SUPPORTED) {
+ /*
+ * pmap won't check the code-signing
+ * validity of pages faulted in via
+ * this mapping, so VM should keep
+ * doing it.
+ */
+ DTRACE_VM3(vm_map_entry_cs_associate_off,
+ vm_map_offset_t, entry->vme_start,
+ vm_map_offset_t, entry->vme_end,
+ int, cs_ret);
+ } else {
+ /*
+ * A real error: do not allow
+ * execution in this mapping.
+ */
+ DTRACE_VM3(vm_map_entry_cs_associate_failure,
+ vm_map_offset_t, entry->vme_start,
+ vm_map_offset_t, entry->vme_end,
+ int, cs_ret);
+ entry->protection &= ~VM_PROT_EXECUTE;
+ entry->max_protection &= ~VM_PROT_EXECUTE;
+ }
+
+ return cs_ret;
+}
+#endif /* PMAP_CS */
+
+/*
+ * FORKED CORPSE FOOTPRINT
+ *
+ * A forked corpse gets a copy of the original VM map but its pmap is mostly
+ * empty since it never ran and never got to fault in any pages.
+ * Collecting footprint info (via "sysctl vm.self_region_footprint") for
+ * a forked corpse would therefore return very little information.
+ *
+ * When forking a corpse, we can pass the VM_MAP_FORK_CORPSE_FOOTPRINT option
+ * to vm_map_fork() to collect footprint information from the original VM map
+ * and its pmap, and store it in the forked corpse's VM map. That information
+ * is stored in place of the VM map's "hole list" since we'll never need to
+ * lookup for holes in the corpse's map.
+ *
+ * The corpse's footprint info looks like this:
+ *
+ * vm_map->vmmap_corpse_footprint points to pageable kernel memory laid out
+ * as follows:
+ * +---------------------------------------+
+ * header-> | cf_size |
+ * +-------------------+-------------------+
+ * | cf_last_region | cf_last_zeroes |
+ * +-------------------+-------------------+
+ * region1-> | cfr_vaddr |
+ * +-------------------+-------------------+
+ * | cfr_num_pages | d0 | d1 | d2 | d3 |
+ * +---------------------------------------+
+ * | d4 | d5 | ... |
+ * +---------------------------------------+
+ * | ... |
+ * +-------------------+-------------------+
+ * | dy | dz | na | na | cfr_vaddr... | <-region2
+ * +-------------------+-------------------+
+ * | cfr_vaddr (ctd) | cfr_num_pages |
+ * +---------------------------------------+
+ * | d0 | d1 ... |
+ * +---------------------------------------+
+ * ...
+ * +---------------------------------------+
+ * last region-> | cfr_vaddr |
+ * +---------------------------------------+
+ * + cfr_num_pages | d0 | d1 | d2 | d3 |
+ * +---------------------------------------+
+ * ...
+ * +---------------------------------------+
+ * | dx | dy | dz | na | na | na | na | na |
+ * +---------------------------------------+
+ *
+ * where:
+ * cf_size: total size of the buffer (rounded to page size)
+ * cf_last_region: offset in the buffer of the last "region" sub-header
+ * cf_last_zeroes: number of trailing "zero" dispositions at the end
+ * of last region
+ * cfr_vaddr: virtual address of the start of the covered "region"
+ * cfr_num_pages: number of pages in the covered "region"
+ * d*: disposition of the page at that virtual address
+ * Regions in the buffer are word-aligned.
+ *
+ * We estimate the size of the buffer based on the number of memory regions
+ * and the virtual size of the address space. While copying each memory region
+ * during vm_map_fork(), we also collect the footprint info for that region
+ * and store it in the buffer, packing it as much as possible (coalescing
+ * contiguous memory regions to avoid having too many region headers and
+ * avoiding long streaks of "zero" page dispositions by splitting footprint
+ * "regions", so the number of regions in the footprint buffer might not match
+ * the number of memory regions in the address space.
+ *
+ * We also have to copy the original task's "nonvolatile" ledgers since that's
+ * part of the footprint and will need to be reported to any tool asking for
+ * the footprint information of the forked corpse.
+ */
+
+uint64_t vm_map_corpse_footprint_count = 0;
+uint64_t vm_map_corpse_footprint_size_avg = 0;
+uint64_t vm_map_corpse_footprint_size_max = 0;
+uint64_t vm_map_corpse_footprint_full = 0;
+uint64_t vm_map_corpse_footprint_no_buf = 0;
+
+/*
+ * vm_map_corpse_footprint_new_region:
+ * closes the current footprint "region" and creates a new one
+ *
+ * Returns NULL if there's not enough space in the buffer for a new region.
+ */
+static struct vm_map_corpse_footprint_region *
+vm_map_corpse_footprint_new_region(
+ struct vm_map_corpse_footprint_header *footprint_header)
+{
+ uintptr_t footprint_edge;
+ uint32_t new_region_offset;
+ struct vm_map_corpse_footprint_region *footprint_region;
+ struct vm_map_corpse_footprint_region *new_footprint_region;
+
+ footprint_edge = ((uintptr_t)footprint_header +
+ footprint_header->cf_size);
+ footprint_region = ((struct vm_map_corpse_footprint_region *)
+ ((char *)footprint_header +
+ footprint_header->cf_last_region));
+ assert((uintptr_t)footprint_region + sizeof (*footprint_region) <=
+ footprint_edge);
+
+ /* get rid of trailing zeroes in the last region */
+ assert(footprint_region->cfr_num_pages >=
+ footprint_header->cf_last_zeroes);
+ footprint_region->cfr_num_pages -=
+ footprint_header->cf_last_zeroes;
+ footprint_header->cf_last_zeroes = 0;
+
+ /* reuse this region if it's now empty */
+ if (footprint_region->cfr_num_pages == 0) {
+ return footprint_region;
+ }
+
+ /* compute offset of new region */
+ new_region_offset = footprint_header->cf_last_region;
+ new_region_offset += sizeof (*footprint_region);
+ new_region_offset += footprint_region->cfr_num_pages;
+ new_region_offset = roundup(new_region_offset, sizeof (int));
+
+ /* check if we're going over the edge */
+ if (((uintptr_t)footprint_header +
+ new_region_offset +
+ sizeof (*footprint_region)) >=
+ footprint_edge) {
+ /* over the edge: no new region */
+ return NULL;
+ }
+
+ /* adjust offset of last region in header */
+ footprint_header->cf_last_region = new_region_offset;
+
+ new_footprint_region = (struct vm_map_corpse_footprint_region *)
+ ((char *)footprint_header +
+ footprint_header->cf_last_region);
+ new_footprint_region->cfr_vaddr = 0;
+ new_footprint_region->cfr_num_pages = 0;
+ /* caller needs to initialize new region */
+
+ return new_footprint_region;
+}
+
+/*
+ * vm_map_corpse_footprint_collect:
+ * collect footprint information for "old_entry" in "old_map" and
+ * stores it in "new_map"'s vmmap_footprint_info.
+ */
+kern_return_t
+vm_map_corpse_footprint_collect(
+ vm_map_t old_map,
+ vm_map_entry_t old_entry,
+ vm_map_t new_map)
+{
+ vm_map_offset_t va;
+ int disp;
+ kern_return_t kr;
+ struct vm_map_corpse_footprint_header *footprint_header;
+ struct vm_map_corpse_footprint_region *footprint_region;
+ struct vm_map_corpse_footprint_region *new_footprint_region;
+ unsigned char *next_disp_p;
+ uintptr_t footprint_edge;
+ uint32_t num_pages_tmp;
+
+ va = old_entry->vme_start;
+
+ vm_map_lock_assert_exclusive(old_map);
+ vm_map_lock_assert_exclusive(new_map);
+
+ assert(new_map->has_corpse_footprint);
+ assert(!old_map->has_corpse_footprint);
+ if (!new_map->has_corpse_footprint ||
+ old_map->has_corpse_footprint) {
+ /*
+ * This can only transfer footprint info from a
+ * map with a live pmap to a map with a corpse footprint.
+ */
+ return KERN_NOT_SUPPORTED;
+ }
+
+ if (new_map->vmmap_corpse_footprint == NULL) {
+ vm_offset_t buf;
+ vm_size_t buf_size;
+
+ buf = 0;
+ buf_size = (sizeof (*footprint_header) +
+ (old_map->hdr.nentries
+ *
+ (sizeof (*footprint_region) +
+ + 3)) /* potential alignment for each region */
+ +
+ ((old_map->size / PAGE_SIZE)
+ *
+ sizeof (char))); /* disposition for each page */
+// printf("FBDP corpse map %p guestimate footprint size 0x%llx\n", new_map, (uint64_t) buf_size);
+ buf_size = round_page(buf_size);
+
+ /* limit buffer to 1 page to validate overflow detection */
+// buf_size = PAGE_SIZE;
+
+ /* limit size to a somewhat sane amount */
+#if CONFIG_EMBEDDED
+#define VM_MAP_CORPSE_FOOTPRINT_INFO_MAX_SIZE (256*1024) /* 256KB */
+#else /* CONFIG_EMBEDDED */
+#define VM_MAP_CORPSE_FOOTPRINT_INFO_MAX_SIZE (8*1024*1024) /* 8MB */
+#endif /* CONFIG_EMBEDDED */
+ if (buf_size > VM_MAP_CORPSE_FOOTPRINT_INFO_MAX_SIZE) {
+ buf_size = VM_MAP_CORPSE_FOOTPRINT_INFO_MAX_SIZE;
+ }
+
+ /*
+ * Allocate the pageable buffer (with a trailing guard page).
+ * It will be zero-filled on demand.
+ */
+ kr = kernel_memory_allocate(kernel_map,
+ &buf,
+ (buf_size
+ + PAGE_SIZE), /* trailing guard page */
+ 0, /* mask */
+ KMA_PAGEABLE | KMA_GUARD_LAST,
+ VM_KERN_MEMORY_DIAG);
+ if (kr != KERN_SUCCESS) {
+ vm_map_corpse_footprint_no_buf++;
+ return kr;
+ }
+
+ /* initialize header and 1st region */
+ footprint_header = (struct vm_map_corpse_footprint_header *)buf;
+ new_map->vmmap_corpse_footprint = footprint_header;
+
+ footprint_header->cf_size = buf_size;
+ footprint_header->cf_last_region =
+ sizeof (*footprint_header);
+ footprint_header->cf_last_zeroes = 0;
+
+ footprint_region = (struct vm_map_corpse_footprint_region *)
+ ((char *)footprint_header +
+ footprint_header->cf_last_region);
+ footprint_region->cfr_vaddr = 0;
+ footprint_region->cfr_num_pages = 0;
+ } else {
+ /* retrieve header and last region */
+ footprint_header = (struct vm_map_corpse_footprint_header *)
+ new_map->vmmap_corpse_footprint;
+ footprint_region = (struct vm_map_corpse_footprint_region *)
+ ((char *)footprint_header +
+ footprint_header->cf_last_region);
+ }
+ footprint_edge = ((uintptr_t)footprint_header +
+ footprint_header->cf_size);
+
+ if ((footprint_region->cfr_vaddr +
+ (((vm_map_offset_t)footprint_region->cfr_num_pages) *
+ PAGE_SIZE))
+ != old_entry->vme_start) {
+ uint64_t num_pages_delta;
+ uint32_t region_offset_delta;
+
+ /*
+ * Not the next contiguous virtual address:
+ * start a new region or store "zero" dispositions for
+ * the missing pages?
+ */
+ /* size of gap in actual page dispositions */
+ num_pages_delta = (((old_entry->vme_start -
+ footprint_region->cfr_vaddr) / PAGE_SIZE)
+ - footprint_region->cfr_num_pages);
+ /* size of gap as a new footprint region header */
+ region_offset_delta =
+ (sizeof (*footprint_region) +
+ roundup((footprint_region->cfr_num_pages -
+ footprint_header->cf_last_zeroes),
+ sizeof (int)) -
+ (footprint_region->cfr_num_pages -
+ footprint_header->cf_last_zeroes));
+// printf("FBDP %s:%d region 0x%x 0x%llx 0x%x vme_start 0x%llx pages_delta 0x%llx region_delta 0x%x\n", __FUNCTION__, __LINE__, footprint_header->cf_last_region, footprint_region->cfr_vaddr, footprint_region->cfr_num_pages, old_entry->vme_start, num_pages_delta, region_offset_delta);
+ if (region_offset_delta < num_pages_delta ||
+ os_add3_overflow(footprint_region->cfr_num_pages,
+ (uint32_t) num_pages_delta,
+ 1,
+ &num_pages_tmp)) {
+ /*
+ * Storing data for this gap would take more space
+ * than inserting a new footprint region header:
+ * let's start a new region and save space. If it's a
+ * tie, let's avoid using a new region, since that
+ * would require more region hops to find the right
+ * range during lookups.
+ *
+ * If the current region's cfr_num_pages would overflow
+ * if we added "zero" page dispositions for the gap,
+ * no choice but to start a new region.
+ */
+// printf("FBDP %s:%d new region\n", __FUNCTION__, __LINE__);
+ new_footprint_region =
+ vm_map_corpse_footprint_new_region(footprint_header);
+ /* check that we're not going over the edge */
+ if (new_footprint_region == NULL) {
+ goto over_the_edge;
+ }
+ footprint_region = new_footprint_region;
+ /* initialize new region as empty */
+ footprint_region->cfr_vaddr = old_entry->vme_start;
+ footprint_region->cfr_num_pages = 0;
+ } else {
+ /*
+ * Store "zero" page dispositions for the missing
+ * pages.
+ */
+// printf("FBDP %s:%d zero gap\n", __FUNCTION__, __LINE__);
+ for (; num_pages_delta > 0; num_pages_delta--) {
+ next_disp_p =
+ ((unsigned char *) footprint_region +
+ sizeof (*footprint_region) +
+ footprint_region->cfr_num_pages);
+ /* check that we're not going over the edge */
+ if ((uintptr_t)next_disp_p >= footprint_edge) {
+ goto over_the_edge;
+ }
+ /* store "zero" disposition for this gap page */
+ footprint_region->cfr_num_pages++;
+ *next_disp_p = (unsigned char) 0;
+ footprint_header->cf_last_zeroes++;
+ }
+ }
+ }
+
+ for (va = old_entry->vme_start;
+ va < old_entry->vme_end;
+ va += PAGE_SIZE) {
+ vm_object_t object;
+
+ object = VME_OBJECT(old_entry);
+ if (!old_entry->is_sub_map &&
+ old_entry->iokit_acct &&
+ object != VM_OBJECT_NULL &&
+ object->internal &&
+ object->purgable == VM_PURGABLE_DENY) {
+ /*
+ * Non-purgeable IOKit memory: phys_footprint
+ * includes the entire virtual mapping.
+ * Since the forked corpse's VM map entry will not
+ * have "iokit_acct", pretend that this page's
+ * disposition is "present & internal", so that it
+ * shows up in the forked corpse's footprint.
+ */
+ disp = (PMAP_QUERY_PAGE_PRESENT |
+ PMAP_QUERY_PAGE_INTERNAL);
+ } else {
+ disp = 0;
+ pmap_query_page_info(old_map->pmap,
+ va,
+ &disp);
+ }
+
+// if (va < SHARED_REGION_BASE_ARM64) printf("FBDP collect map %p va 0x%llx disp 0x%x\n", new_map, va, disp);
+
+ if (disp == 0 && footprint_region->cfr_num_pages == 0) {
+ /*
+ * Ignore "zero" dispositions at start of
+ * region: just move start of region.
+ */
+ footprint_region->cfr_vaddr += PAGE_SIZE;
+ continue;
+ }
+
+ /* would region's cfr_num_pages overflow? */
+ if (os_add_overflow(footprint_region->cfr_num_pages, 1,
+ &num_pages_tmp)) {
+ /* overflow: create a new region */
+ new_footprint_region =
+ vm_map_corpse_footprint_new_region(
+ footprint_header);
+ if (new_footprint_region == NULL) {
+ goto over_the_edge;
+ }
+ footprint_region = new_footprint_region;
+ footprint_region->cfr_vaddr = va;
+ footprint_region->cfr_num_pages = 0;
+ }
+
+ next_disp_p = ((unsigned char *)footprint_region +
+ sizeof (*footprint_region) +
+ footprint_region->cfr_num_pages);
+ /* check that we're not going over the edge */
+ if ((uintptr_t)next_disp_p >= footprint_edge) {
+ goto over_the_edge;
+ }
+ /* store this dispostion */
+ *next_disp_p = (unsigned char) disp;
+ footprint_region->cfr_num_pages++;
+
+ if (disp != 0) {
+ /* non-zero disp: break the current zero streak */
+ footprint_header->cf_last_zeroes = 0;
+ /* done */
+ continue;
+ }
+
+ /* zero disp: add to the current streak of zeroes */
+ footprint_header->cf_last_zeroes++;
+ if ((footprint_header->cf_last_zeroes +
+ roundup((footprint_region->cfr_num_pages -
+ footprint_header->cf_last_zeroes) &
+ (sizeof (int) - 1),
+ sizeof (int))) <
+ (sizeof (*footprint_header))) {
+ /*
+ * There are not enough trailing "zero" dispositions
+ * (+ the extra padding we would need for the previous
+ * region); creating a new region would not save space
+ * at this point, so let's keep this "zero" disposition
+ * in this region and reconsider later.
+ */
+ continue;
+ }
+ /*
+ * Create a new region to avoid having too many consecutive
+ * "zero" dispositions.
+ */
+ new_footprint_region =
+ vm_map_corpse_footprint_new_region(footprint_header);
+ if (new_footprint_region == NULL) {
+ goto over_the_edge;
+ }
+ footprint_region = new_footprint_region;
+ /* initialize the new region as empty ... */
+ footprint_region->cfr_num_pages = 0;
+ /* ... and skip this "zero" disp */
+ footprint_region->cfr_vaddr = va + PAGE_SIZE;
+ }
return KERN_SUCCESS;
+
+over_the_edge:
+// printf("FBDP map %p footprint was full for va 0x%llx\n", new_map, va);
+ vm_map_corpse_footprint_full++;
+ return KERN_RESOURCE_SHORTAGE;
+}
+
+/*
+ * vm_map_corpse_footprint_collect_done:
+ * completes the footprint collection by getting rid of any remaining
+ * trailing "zero" dispositions and trimming the unused part of the
+ * kernel buffer
+ */
+void
+vm_map_corpse_footprint_collect_done(
+ vm_map_t new_map)
+{
+ struct vm_map_corpse_footprint_header *footprint_header;
+ struct vm_map_corpse_footprint_region *footprint_region;
+ vm_size_t buf_size, actual_size;
+ kern_return_t kr;
+
+ assert(new_map->has_corpse_footprint);
+ if (!new_map->has_corpse_footprint ||
+ new_map->vmmap_corpse_footprint == NULL) {
+ return;
+ }
+
+ footprint_header = (struct vm_map_corpse_footprint_header *)
+ new_map->vmmap_corpse_footprint;
+ buf_size = footprint_header->cf_size;
+
+ footprint_region = (struct vm_map_corpse_footprint_region *)
+ ((char *)footprint_header +
+ footprint_header->cf_last_region);
+
+ /* get rid of trailing zeroes in last region */
+ assert(footprint_region->cfr_num_pages >= footprint_header->cf_last_zeroes);
+ footprint_region->cfr_num_pages -= footprint_header->cf_last_zeroes;
+ footprint_header->cf_last_zeroes = 0;
+
+ actual_size = (vm_size_t)(footprint_header->cf_last_region +
+ sizeof (*footprint_region) +
+ footprint_region->cfr_num_pages);
+
+// printf("FBDP map %p buf_size 0x%llx actual_size 0x%llx\n", new_map, (uint64_t) buf_size, (uint64_t) actual_size);
+ vm_map_corpse_footprint_size_avg =
+ (((vm_map_corpse_footprint_size_avg *
+ vm_map_corpse_footprint_count) +
+ actual_size) /
+ (vm_map_corpse_footprint_count + 1));
+ vm_map_corpse_footprint_count++;
+ if (actual_size > vm_map_corpse_footprint_size_max) {
+ vm_map_corpse_footprint_size_max = actual_size;
+ }
+
+ actual_size = round_page(actual_size);
+ if (buf_size > actual_size) {
+ kr = vm_deallocate(kernel_map,
+ ((vm_address_t)footprint_header +
+ actual_size +
+ PAGE_SIZE), /* trailing guard page */
+ (buf_size - actual_size));
+ assertf(kr == KERN_SUCCESS,
+ "trim: footprint_header %p buf_size 0x%llx actual_size 0x%llx kr=0x%x\n",
+ footprint_header,
+ (uint64_t) buf_size,
+ (uint64_t) actual_size,
+ kr);
+ kr = vm_protect(kernel_map,
+ ((vm_address_t)footprint_header +
+ actual_size),
+ PAGE_SIZE,
+ FALSE, /* set_maximum */
+ VM_PROT_NONE);
+ assertf(kr == KERN_SUCCESS,
+ "guard: footprint_header %p buf_size 0x%llx actual_size 0x%llx kr=0x%x\n",
+ footprint_header,
+ (uint64_t) buf_size,
+ (uint64_t) actual_size,
+ kr);
+ }
+
+ footprint_header->cf_size = actual_size;
+}
+
+/*
+ * vm_map_corpse_footprint_query_page_info:
+ * retrieves the disposition of the page at virtual address "vaddr"
+ * in the forked corpse's VM map
+ *
+ * This is the equivalent of pmap_query_page_info() for a forked corpse.
+ */
+kern_return_t
+vm_map_corpse_footprint_query_page_info(
+ vm_map_t map,
+ vm_map_offset_t va,
+ int *disp)
+{
+ struct vm_map_corpse_footprint_header *footprint_header;
+ struct vm_map_corpse_footprint_region *footprint_region;
+ uint32_t footprint_region_offset;
+ vm_map_offset_t region_start, region_end;
+ int disp_idx;
+ kern_return_t kr;
+
+ if (!map->has_corpse_footprint) {
+ *disp = 0;
+ kr = KERN_INVALID_ARGUMENT;
+ goto done;
+ }
+
+ footprint_header = map->vmmap_corpse_footprint;
+ if (footprint_header == NULL) {
+ *disp = 0;
+// if (va < SHARED_REGION_BASE_ARM64) printf("FBDP %d query map %p va 0x%llx disp 0x%x\n", __LINE__, map, va, *disp);
+ kr = KERN_INVALID_ARGUMENT;
+ goto done;
+ }
+
+ /* start looking at the hint ("cf_hint_region") */
+ footprint_region_offset = footprint_header->cf_hint_region;
+
+lookup_again:
+ if (footprint_region_offset < sizeof (*footprint_header)) {
+ /* hint too low: start from 1st region */
+ footprint_region_offset = sizeof (*footprint_header);
+ }
+ if (footprint_region_offset >= footprint_header->cf_last_region) {
+ /* hint too high: re-start from 1st region */
+ footprint_region_offset = sizeof (*footprint_header);
+ }
+ footprint_region = (struct vm_map_corpse_footprint_region *)
+ ((char *)footprint_header + footprint_region_offset);
+ region_start = footprint_region->cfr_vaddr;
+ region_end = (region_start +
+ ((vm_map_offset_t)(footprint_region->cfr_num_pages) *
+ PAGE_SIZE));
+ if (va < region_start &&
+ footprint_region_offset != sizeof (*footprint_header)) {
+ /* our range starts before the hint region */
+
+ /* reset the hint (in a racy way...) */
+ footprint_header->cf_hint_region = sizeof (*footprint_header);
+ /* lookup "va" again from 1st region */
+ footprint_region_offset = sizeof (*footprint_header);
+ goto lookup_again;
+ }
+
+ while (va >= region_end) {
+ if (footprint_region_offset >= footprint_header->cf_last_region) {
+ break;
+ }
+ /* skip the region's header */
+ footprint_region_offset += sizeof (*footprint_region);
+ /* skip the region's page dispositions */
+ footprint_region_offset += footprint_region->cfr_num_pages;
+ /* align to next word boundary */
+ footprint_region_offset =
+ roundup(footprint_region_offset,
+ sizeof (int));
+ footprint_region = (struct vm_map_corpse_footprint_region *)
+ ((char *)footprint_header + footprint_region_offset);
+ region_start = footprint_region->cfr_vaddr;
+ region_end = (region_start +
+ ((vm_map_offset_t)(footprint_region->cfr_num_pages) *
+ PAGE_SIZE));
+ }
+ if (va < region_start || va >= region_end) {
+ /* page not found */
+ *disp = 0;
+// if (va < SHARED_REGION_BASE_ARM64) printf("FBDP %d query map %p va 0x%llx disp 0x%x\n", __LINE__, map, va, *disp);
+ kr = KERN_SUCCESS;
+ goto done;
+ }
+
+ /* "va" found: set the lookup hint for next lookup (in a racy way...) */
+ footprint_header->cf_hint_region = footprint_region_offset;
+
+ /* get page disposition for "va" in this region */
+ disp_idx = (int) ((va - footprint_region->cfr_vaddr) / PAGE_SIZE);
+ *disp = (int) (footprint_region->cfr_disposition[disp_idx]);
+
+ kr = KERN_SUCCESS;
+done:
+// if (va < SHARED_REGION_BASE_ARM64) printf("FBDP %d query map %p va 0x%llx disp 0x%x\n", __LINE__, map, va, *disp);
+ /* dtrace -n 'vminfo:::footprint_query_page_info { printf("map 0x%p va 0x%llx disp 0x%x kr 0x%x", arg0, arg1, arg2, arg3); }' */
+ DTRACE_VM4(footprint_query_page_info,
+ vm_map_t, map,
+ vm_map_offset_t, va,
+ int, *disp,
+ kern_return_t, kr);
+
+ return kr;
+}
+
+
+static void
+vm_map_corpse_footprint_destroy(
+ vm_map_t map)
+{
+ if (map->has_corpse_footprint &&
+ map->vmmap_corpse_footprint != 0) {
+ struct vm_map_corpse_footprint_header *footprint_header;
+ vm_size_t buf_size;
+ kern_return_t kr;
+
+ footprint_header = map->vmmap_corpse_footprint;
+ buf_size = footprint_header->cf_size;
+ kr = vm_deallocate(kernel_map,
+ (vm_offset_t) map->vmmap_corpse_footprint,
+ ((vm_size_t) buf_size
+ + PAGE_SIZE)); /* trailing guard page */
+ assertf(kr == KERN_SUCCESS, "kr=0x%x\n", kr);
+ map->vmmap_corpse_footprint = 0;
+ map->has_corpse_footprint = FALSE;
+ }
+}
+
+/*
+ * vm_map_copy_footprint_ledgers:
+ * copies any ledger that's relevant to the memory footprint of "old_task"
+ * into the forked corpse's task ("new_task")
+ */
+void
+vm_map_copy_footprint_ledgers(
+ task_t old_task,
+ task_t new_task)
+{
+ vm_map_copy_ledger(old_task, new_task, task_ledgers.phys_footprint);
+ vm_map_copy_ledger(old_task, new_task, task_ledgers.purgeable_nonvolatile);
+ vm_map_copy_ledger(old_task, new_task, task_ledgers.purgeable_nonvolatile_compressed);
+ vm_map_copy_ledger(old_task, new_task, task_ledgers.internal);
+ vm_map_copy_ledger(old_task, new_task, task_ledgers.internal_compressed);
+ vm_map_copy_ledger(old_task, new_task, task_ledgers.iokit_mapped);
+ vm_map_copy_ledger(old_task, new_task, task_ledgers.alternate_accounting);
+ vm_map_copy_ledger(old_task, new_task, task_ledgers.alternate_accounting_compressed);
+ vm_map_copy_ledger(old_task, new_task, task_ledgers.page_table);
+ vm_map_copy_ledger(old_task, new_task, task_ledgers.network_nonvolatile);
+ vm_map_copy_ledger(old_task, new_task, task_ledgers.network_nonvolatile_compressed);
+ vm_map_copy_ledger(old_task, new_task, task_ledgers.wired_mem);
+}
+
+/*
+ * vm_map_copy_ledger:
+ * copy a single ledger from "old_task" to "new_task"
+ */
+void
+vm_map_copy_ledger(
+ task_t old_task,
+ task_t new_task,
+ int ledger_entry)
+{
+ ledger_amount_t old_balance, new_balance, delta;
+
+ assert(new_task->map->has_corpse_footprint);
+ if (!new_task->map->has_corpse_footprint)
+ return;
+
+ /* turn off sanity checks for the ledger we're about to mess with */
+ ledger_disable_panic_on_negative(new_task->ledger,
+ ledger_entry);
+
+ /* adjust "new_task" to match "old_task" */
+ ledger_get_balance(old_task->ledger,
+ ledger_entry,
+ &old_balance);
+ ledger_get_balance(new_task->ledger,
+ ledger_entry,
+ &new_balance);
+ if (new_balance == old_balance) {
+ /* new == old: done */
+ } else if (new_balance > old_balance) {
+ /* new > old ==> new -= new - old */
+ delta = new_balance - old_balance;
+ ledger_debit(new_task->ledger,
+ ledger_entry,
+ delta);
+ } else {
+ /* new < old ==> new += old - new */
+ delta = old_balance - new_balance;
+ ledger_credit(new_task->ledger,
+ ledger_entry,
+ delta);
+ }
}
projects / apple / xnu.git / blobdiff