X-Git-Url: https://git.saurik.com/apple/xnu.git/blobdiff_plain/21362eb3e66fd2c787aee132bce100a44d71a99c..143464d58d2bd6378e74eec636961ceb0d32fb91:/osfmk/vm/vm_map.c?ds=inline diff --git a/osfmk/vm/vm_map.c b/osfmk/vm/vm_map.c index 267ca6b37..a30ff18e1 100644 --- a/osfmk/vm/vm_map.c +++ b/osfmk/vm/vm_map.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2000-2005 Apple Computer, Inc. All rights reserved. + * Copyright (c) 2000-2012 Apple Inc. All rights reserved. * * @APPLE_OSREFERENCE_LICENSE_HEADER_START@ * @@ -74,7 +74,9 @@ #include #include #include +#include #include +#include #include #include @@ -82,188 +84,215 @@ #include #include +#include #include #include #include #include #include +#include #include #include #include #include -#include -#include #include #include #include -#include -#include // for vm_map_commpage64 and vm_map_remove_compage64 - -#ifdef ppc -#include -#endif /* ppc */ +#include +#include #include +#include +#include +extern u_int32_t random(void); /* from */ /* Internal prototypes */ - + static void vm_map_simplify_range( vm_map_t map, vm_map_offset_t start, vm_map_offset_t end); /* forward */ static boolean_t vm_map_range_check( - vm_map_t map, - vm_map_offset_t start, - vm_map_offset_t end, - vm_map_entry_t *entry); + vm_map_t map, + vm_map_offset_t start, + vm_map_offset_t end, + vm_map_entry_t *entry); static vm_map_entry_t _vm_map_entry_create( - struct vm_map_header *map_header); + struct vm_map_header *map_header, boolean_t map_locked); static void _vm_map_entry_dispose( - struct vm_map_header *map_header, - vm_map_entry_t entry); + struct vm_map_header *map_header, + vm_map_entry_t entry); static void vm_map_pmap_enter( - vm_map_t map, - vm_map_offset_t addr, - vm_map_offset_t end_addr, - vm_object_t object, - vm_object_offset_t offset, - vm_prot_t protection); + vm_map_t map, + vm_map_offset_t addr, + vm_map_offset_t end_addr, + vm_object_t object, + vm_object_offset_t offset, + vm_prot_t protection); static void _vm_map_clip_end( - struct vm_map_header *map_header, - vm_map_entry_t entry, - vm_map_offset_t end); + struct vm_map_header *map_header, + vm_map_entry_t entry, + vm_map_offset_t end); static void _vm_map_clip_start( - struct vm_map_header *map_header, - vm_map_entry_t entry, - vm_map_offset_t start); + struct vm_map_header *map_header, + vm_map_entry_t entry, + vm_map_offset_t start); static void vm_map_entry_delete( - vm_map_t map, - vm_map_entry_t entry); + vm_map_t map, + vm_map_entry_t entry); static kern_return_t vm_map_delete( - vm_map_t map, - vm_map_offset_t start, - vm_map_offset_t end, - int flags, - vm_map_t zap_map); + vm_map_t map, + vm_map_offset_t start, + vm_map_offset_t end, + int flags, + vm_map_t zap_map); static kern_return_t vm_map_copy_overwrite_unaligned( - vm_map_t dst_map, - vm_map_entry_t entry, - vm_map_copy_t copy, - vm_map_address_t start); + vm_map_t dst_map, + vm_map_entry_t entry, + vm_map_copy_t copy, + vm_map_address_t start, + boolean_t discard_on_success); static kern_return_t vm_map_copy_overwrite_aligned( - vm_map_t dst_map, - vm_map_entry_t tmp_entry, - vm_map_copy_t copy, - vm_map_offset_t start, - pmap_t pmap); + vm_map_t dst_map, + vm_map_entry_t tmp_entry, + vm_map_copy_t copy, + vm_map_offset_t start, + pmap_t pmap); static kern_return_t vm_map_copyin_kernel_buffer( - vm_map_t src_map, - vm_map_address_t src_addr, - vm_map_size_t len, - boolean_t src_destroy, - vm_map_copy_t *copy_result); /* OUT */ + vm_map_t src_map, + vm_map_address_t src_addr, + vm_map_size_t len, + boolean_t src_destroy, + vm_map_copy_t *copy_result); /* OUT */ static kern_return_t vm_map_copyout_kernel_buffer( - vm_map_t map, - vm_map_address_t *addr, /* IN/OUT */ - vm_map_copy_t copy, - boolean_t overwrite); + vm_map_t map, + vm_map_address_t *addr, /* IN/OUT */ + vm_map_copy_t copy, + boolean_t overwrite, + boolean_t consume_on_success); static void vm_map_fork_share( - vm_map_t old_map, - vm_map_entry_t old_entry, - vm_map_t new_map); + vm_map_t old_map, + vm_map_entry_t old_entry, + vm_map_t new_map); static boolean_t vm_map_fork_copy( - vm_map_t old_map, - vm_map_entry_t *old_entry_p, - vm_map_t new_map); - -static void vm_map_region_top_walk( - vm_map_entry_t entry, - vm_region_top_info_t top); - -static 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); + vm_map_t old_map, + vm_map_entry_t *old_entry_p, + vm_map_t new_map); + +void vm_map_region_top_walk( + vm_map_entry_t entry, + vm_region_top_info_t top); + +void vm_map_region_walk( + vm_map_t map, + vm_map_offset_t va, + vm_map_entry_t entry, + vm_object_offset_t offset, + vm_object_size_t range, + vm_region_extended_info_t extended, + boolean_t look_for_pages, + mach_msg_type_number_t count); static kern_return_t vm_map_wire_nested( - vm_map_t map, - vm_map_offset_t start, - vm_map_offset_t end, - vm_prot_t access_type, - boolean_t user_wire, - pmap_t map_pmap, - vm_map_offset_t pmap_addr); + vm_map_t map, + vm_map_offset_t start, + vm_map_offset_t end, + vm_prot_t access_type, + boolean_t user_wire, + pmap_t map_pmap, + vm_map_offset_t pmap_addr); static kern_return_t vm_map_unwire_nested( - vm_map_t map, - vm_map_offset_t start, - vm_map_offset_t end, - boolean_t user_wire, - pmap_t map_pmap, - vm_map_offset_t pmap_addr); + vm_map_t map, + vm_map_offset_t start, + vm_map_offset_t end, + boolean_t user_wire, + pmap_t map_pmap, + vm_map_offset_t pmap_addr); static kern_return_t vm_map_overwrite_submap_recurse( - vm_map_t dst_map, - vm_map_offset_t dst_addr, - vm_map_size_t dst_size); + vm_map_t dst_map, + vm_map_offset_t dst_addr, + vm_map_size_t dst_size); static kern_return_t vm_map_copy_overwrite_nested( - vm_map_t dst_map, - vm_map_offset_t dst_addr, - vm_map_copy_t copy, - boolean_t interruptible, - pmap_t pmap); + vm_map_t dst_map, + vm_map_offset_t dst_addr, + vm_map_copy_t copy, + boolean_t interruptible, + pmap_t pmap, + boolean_t discard_on_success); static kern_return_t vm_map_remap_extract( - vm_map_t map, - vm_map_offset_t addr, - vm_map_size_t size, - boolean_t copy, - struct vm_map_header *map_header, - vm_prot_t *cur_protection, - vm_prot_t *max_protection, - vm_inherit_t inheritance, - boolean_t pageable); + vm_map_t map, + vm_map_offset_t addr, + vm_map_size_t size, + boolean_t copy, + struct vm_map_header *map_header, + vm_prot_t *cur_protection, + vm_prot_t *max_protection, + vm_inherit_t inheritance, + boolean_t pageable); static kern_return_t vm_map_remap_range_allocate( - vm_map_t map, - vm_map_address_t *address, - vm_map_size_t size, - vm_map_offset_t mask, - boolean_t anywhere, - vm_map_entry_t *map_entry); + vm_map_t map, + vm_map_address_t *address, + vm_map_size_t size, + vm_map_offset_t mask, + int flags, + vm_map_entry_t *map_entry); static void vm_map_region_look_for_page( - vm_map_t map, - vm_map_offset_t va, - vm_object_t object, - vm_object_offset_t offset, - int max_refcnt, - int depth, - vm_region_extended_info_t extended); + vm_map_t map, + vm_map_offset_t va, + vm_object_t object, + vm_object_offset_t offset, + int max_refcnt, + int depth, + vm_region_extended_info_t extended, + mach_msg_type_number_t count); static int vm_map_region_count_obj_refs( - vm_map_entry_t entry, - vm_object_t object); + vm_map_entry_t entry, + vm_object_t object); + + +static kern_return_t vm_map_willneed( + vm_map_t map, + vm_map_offset_t start, + vm_map_offset_t end); + +static kern_return_t vm_map_reuse_pages( + vm_map_t map, + vm_map_offset_t start, + vm_map_offset_t end); + +static kern_return_t vm_map_reusable_pages( + vm_map_t map, + vm_map_offset_t start, + vm_map_offset_t end); + +static kern_return_t vm_map_can_reuse( + vm_map_t map, + vm_map_offset_t start, + vm_map_offset_t end); + /* * Macros to copy a vm_map_entry. We must be careful to correctly @@ -274,17 +303,85 @@ static int vm_map_region_count_obj_refs( * wire count; it's used for map splitting and zone changing in * vm_map_copyout. */ -#define vm_map_entry_copy(NEW,OLD) \ -MACRO_BEGIN \ - *(NEW) = *(OLD); \ - (NEW)->is_shared = FALSE; \ - (NEW)->needs_wakeup = FALSE; \ - (NEW)->in_transition = FALSE; \ - (NEW)->wired_count = 0; \ - (NEW)->user_wired_count = 0; \ + +#define vm_map_entry_copy(NEW,OLD) \ +MACRO_BEGIN \ +boolean_t _vmec_reserved = (NEW)->from_reserved_zone; \ + *(NEW) = *(OLD); \ + (NEW)->is_shared = FALSE; \ + (NEW)->needs_wakeup = FALSE; \ + (NEW)->in_transition = FALSE; \ + (NEW)->wired_count = 0; \ + (NEW)->user_wired_count = 0; \ + (NEW)->permanent = FALSE; \ + (NEW)->used_for_jit = FALSE; \ + (NEW)->from_reserved_zone = _vmec_reserved; \ +MACRO_END + +#define vm_map_entry_copy_full(NEW,OLD) \ +MACRO_BEGIN \ +boolean_t _vmecf_reserved = (NEW)->from_reserved_zone; \ +(*(NEW) = *(OLD)); \ +(NEW)->from_reserved_zone = _vmecf_reserved; \ MACRO_END -#define vm_map_entry_copy_full(NEW,OLD) (*(NEW) = *(OLD)) +/* + * Decide if we want to allow processes to execute from their data or stack areas. + * override_nx() returns true if we do. Data/stack execution can be enabled independently + * for 32 and 64 bit processes. Set the VM_ABI_32 or VM_ABI_64 flags in allow_data_exec + * or allow_stack_exec to enable data execution for that type of data area for that particular + * ABI (or both by or'ing the flags together). These are initialized in the architecture + * specific pmap files since the default behavior varies according to architecture. The + * main reason it varies is because of the need to provide binary compatibility with old + * applications that were written before these restrictions came into being. In the old + * days, an app could execute anything it could read, but this has slowly been tightened + * up over time. The default behavior is: + * + * 32-bit PPC apps may execute from both stack and data areas + * 32-bit Intel apps may exeucte from data areas but not stack + * 64-bit PPC/Intel apps may not execute from either data or stack + * + * An application on any architecture may override these defaults by explicitly + * adding PROT_EXEC permission to the page in question with the mprotect(2) + * system call. This code here just determines what happens when an app tries to + * execute from a page that lacks execute permission. + * + * Note that allow_data_exec or allow_stack_exec may also be modified by sysctl to change the + * default behavior for both 32 and 64 bit apps on a system-wide basis. Furthermore, + * a Mach-O header flag bit (MH_NO_HEAP_EXECUTION) can be used to forcibly disallow + * execution from data areas for a particular binary even if the arch normally permits it. As + * a final wrinkle, a posix_spawn attribute flag can be used to negate this opt-in header bit + * to support some complicated use cases, notably browsers with out-of-process plugins that + * are not all NX-safe. + */ + +extern int allow_data_exec, allow_stack_exec; + +int +override_nx(vm_map_t map, uint32_t user_tag) /* map unused on arm */ +{ + int current_abi; + + /* + * Determine if the app is running in 32 or 64 bit mode. + */ + + if (vm_map_is_64bit(map)) + current_abi = VM_ABI_64; + else + current_abi = VM_ABI_32; + + /* + * Determine if we should allow the execution based on whether it's a + * stack or data area and the current architecture. + */ + + if (user_tag == VM_MEMORY_STACK) + return allow_stack_exec & current_abi; + + return (allow_data_exec & current_abi) && (map->map_disallow_data_exec == FALSE); +} + /* * Virtual memory maps provide for the mapping, protection, @@ -331,7 +428,8 @@ MACRO_END 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_kentry_zone; /* zone for kernel 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 */ @@ -343,6 +441,160 @@ static zone_t vm_map_copy_zone; /* zone for vm_map_copy structures */ vm_object_t vm_submap_object; +static void *map_data; +static vm_size_t map_data_size; +static void *kentry_data; +static vm_size_t kentry_data_size; + +#define NO_COALESCE_LIMIT ((1024 * 128) - 1) + +/* Skip acquiring locks if we're in the midst of a kernel core dump */ +unsigned int not_in_kdp = 1; + +unsigned int vm_map_set_cache_attr_count = 0; + +kern_return_t +vm_map_set_cache_attr( + vm_map_t map, + vm_map_offset_t va) +{ + vm_map_entry_t map_entry; + vm_object_t object; + kern_return_t kr = KERN_SUCCESS; + + vm_map_lock_read(map); + + if (!vm_map_lookup_entry(map, va, &map_entry) || + map_entry->is_sub_map) { + /* + * that memory is not properly mapped + */ + kr = KERN_INVALID_ARGUMENT; + goto done; + } + object = map_entry->object.vm_object; + + if (object == VM_OBJECT_NULL) { + /* + * there should be a VM object here at this point + */ + kr = KERN_INVALID_ARGUMENT; + goto done; + } + vm_object_lock(object); + object->set_cache_attr = TRUE; + vm_object_unlock(object); + + vm_map_set_cache_attr_count++; +done: + vm_map_unlock_read(map); + + return kr; +} + + +#if CONFIG_CODE_DECRYPTION +/* + * vm_map_apple_protected: + * This remaps the requested part of the object with an object backed by + * the decrypting pager. + * crypt_info contains entry points and session data for the crypt module. + * The crypt_info block will be copied by vm_map_apple_protected. The data structures + * referenced in crypt_info must remain valid until crypt_info->crypt_end() is called. + */ +kern_return_t +vm_map_apple_protected( + vm_map_t map, + vm_map_offset_t start, + vm_map_offset_t end, + 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; + vm_object_t protected_object; + vm_map_offset_t map_addr; + + 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_map_unlock_read(map); + map_locked = FALSE; + + /* + * 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. + */ + protected_mem_obj = apple_protect_pager_setup(protected_object, crypt_info); + + /* release extra ref on protected object */ + vm_object_deallocate(protected_object); + + if (protected_mem_obj == NULL) { + kr = KERN_FAILURE; + goto done; + } + + /* 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); + +done: + if (map_locked) { + vm_map_unlock_read(map); + } + return kr; +} +#endif /* CONFIG_CODE_DECRYPTION */ + + +lck_grp_t vm_map_lck_grp; +lck_grp_attr_t vm_map_lck_grp_attr; +lck_attr_t vm_map_lck_attr; + + /* * vm_map_init: * @@ -356,7 +608,7 @@ vm_object_t vm_submap_object; * * vm_map_zone: used to allocate maps. * vm_map_entry_zone: used to allocate map entries. - * vm_map_kentry_zone: used to allocate map entries for the kernel. + * vm_map_entry_reserved_zone: fallback zone for kernel map entries * * The kernel allocates map entries from a special zone that is initially * "crammed" with memory. It would be difficult (perhaps impossible) for @@ -364,77 +616,100 @@ vm_object_t vm_submap_object; * empty since the very act of allocating memory implies the creation * of a new entry. */ - -static void *map_data; -static vm_map_size_t map_data_size; -static void *kentry_data; -static vm_map_size_t kentry_data_size; -static int kentry_count = 2048; /* to init kentry_data_size */ - -#define NO_COALESCE_LIMIT (1024 * 128) - -/* - * Threshold for aggressive (eager) page map entering for vm copyout - * operations. Any copyout larger will NOT be aggressively entered. - */ -static vm_map_size_t vm_map_aggressive_enter_max; /* set by bootstrap */ - -/* Skip acquiring locks if we're in the midst of a kernel core dump */ -extern unsigned int not_in_kdp; - void vm_map_init( void) { - vm_map_zone = zinit((vm_map_size_t) sizeof(struct vm_map), 40*1024, - PAGE_SIZE, "maps"); - + vm_size_t entry_zone_alloc_size; + const char *mez_name = "VM map entries"; + + vm_map_zone = zinit((vm_map_size_t) sizeof(struct _vm_map), 40*1024, + PAGE_SIZE, "maps"); + zone_change(vm_map_zone, Z_NOENCRYPT, TRUE); +#if defined(__LP64__) + entry_zone_alloc_size = PAGE_SIZE * 5; +#else + entry_zone_alloc_size = PAGE_SIZE * 6; +#endif vm_map_entry_zone = zinit((vm_map_size_t) sizeof(struct vm_map_entry), - 1024*1024, PAGE_SIZE*5, - "non-kernel map entries"); + 1024*1024, entry_zone_alloc_size, + mez_name); + zone_change(vm_map_entry_zone, Z_NOENCRYPT, TRUE); + zone_change(vm_map_entry_zone, Z_NOCALLOUT, TRUE); + zone_change(vm_map_entry_zone, Z_GZALLOC_EXEMPT, TRUE); - vm_map_kentry_zone = zinit((vm_map_size_t) sizeof(struct vm_map_entry), - kentry_data_size, kentry_data_size, - "kernel map entries"); + vm_map_entry_reserved_zone = zinit((vm_map_size_t) sizeof(struct vm_map_entry), + kentry_data_size * 64, kentry_data_size, + "Reserved VM map entries"); + zone_change(vm_map_entry_reserved_zone, Z_NOENCRYPT, TRUE); vm_map_copy_zone = zinit((vm_map_size_t) sizeof(struct vm_map_copy), - 16*1024, PAGE_SIZE, "map copies"); + 16*1024, PAGE_SIZE, "VM map copies"); + zone_change(vm_map_copy_zone, Z_NOENCRYPT, TRUE); /* * Cram the map and kentry zones with initial data. - * Set kentry_zone non-collectible to aid zone_gc(). + * Set reserved_zone non-collectible to aid zone_gc(). */ zone_change(vm_map_zone, Z_COLLECT, FALSE); - zone_change(vm_map_kentry_zone, Z_COLLECT, FALSE); - zone_change(vm_map_kentry_zone, Z_EXPAND, FALSE); - zcram(vm_map_zone, map_data, map_data_size); - zcram(vm_map_kentry_zone, kentry_data, kentry_data_size); + + zone_change(vm_map_entry_reserved_zone, Z_COLLECT, FALSE); + zone_change(vm_map_entry_reserved_zone, Z_EXPAND, FALSE); + zone_change(vm_map_entry_reserved_zone, Z_FOREIGN, TRUE); + zone_change(vm_map_entry_reserved_zone, Z_NOCALLOUT, TRUE); + zone_change(vm_map_entry_reserved_zone, Z_CALLERACCT, FALSE); /* don't charge caller */ + zone_change(vm_map_copy_zone, Z_CALLERACCT, FALSE); /* don't charge caller */ + zone_change(vm_map_entry_reserved_zone, Z_GZALLOC_EXEMPT, TRUE); + + 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); + + 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 */ } void vm_map_steal_memory( void) { - map_data_size = vm_map_round_page(10 * sizeof(struct vm_map)); + uint32_t kentry_initial_pages; + + map_data_size = round_page(10 * sizeof(struct _vm_map)); map_data = pmap_steal_memory(map_data_size); -#if 0 /* - * Limiting worst case: vm_map_kentry_zone needs to map each "available" - * physical page (i.e. that beyond the kernel image and page tables) - * individually; we guess at most one entry per eight pages in the - * real world. This works out to roughly .1 of 1% of physical memory, - * or roughly 1900 entries (64K) for a 64M machine with 4K pages. + * kentry_initial_pages corresponds to the number of kernel map entries + * required during bootstrap until the asynchronous replenishment + * scheme is activated and/or entries are available from the general + * map entry pool. */ +#if defined(__LP64__) + kentry_initial_pages = 10; +#else + kentry_initial_pages = 6; #endif - kentry_count = pmap_free_pages() / 8; +#if CONFIG_GZALLOC + /* If using the guard allocator, reserve more memory for the kernel + * reserved map entry pool. + */ + if (gzalloc_enabled()) + kentry_initial_pages *= 1024; +#endif - kentry_data_size = - vm_map_round_page(kentry_count * sizeof(struct vm_map_entry)); + kentry_data_size = kentry_initial_pages * PAGE_SIZE; kentry_data = pmap_steal_memory(kentry_data_size); } +void vm_kernel_reserved_entry_init(void) { + zone_prio_refill_configure(vm_map_entry_reserved_zone, (6*PAGE_SIZE)/sizeof(struct vm_map_entry)); +} + /* * vm_map_create: * @@ -449,6 +724,7 @@ vm_map_create( vm_map_offset_t max, boolean_t pageable) { + static int color_seed = 0; register vm_map_t result; result = (vm_map_t) zalloc(vm_map_zone); @@ -460,7 +736,13 @@ vm_map_create( result->hdr.nentries = 0; result->hdr.entries_pageable = pageable; + vm_map_store_init( &(result->hdr) ); + + result->hdr.page_shift = PAGE_SHIFT; + result->size = 0; + result->user_wire_limit = MACH_VM_MAX_ADDRESS; /* default limit is unlimited */ + result->user_wire_size = 0; result->ref_count = 1; #if TASK_SWAPPER result->res_count = 1; @@ -471,13 +753,22 @@ vm_map_create( result->max_offset = max; result->wiring_required = FALSE; result->no_zero_fill = FALSE; - result->mapped = FALSE; + result->mapped_in_other_pmaps = FALSE; result->wait_for_space = FALSE; + result->switch_protect = FALSE; + result->disable_vmentry_reuse = FALSE; + result->map_disallow_data_exec = FALSE; + result->highest_entry_end = 0; result->first_free = vm_map_to_entry(result); result->hint = vm_map_to_entry(result); + result->color_rr = (color_seed++) & vm_color_mask; + result->jit_entry_exists = FALSE; +#if CONFIG_FREEZE + result->default_freezer_handle = NULL; +#endif vm_map_lock_init(result); - mutex_init(&result->s_lock, 0); - + lck_mtx_init_ext(&result->s_lock, &result->s_lock_ext, &vm_map_lck_grp, &vm_map_lck_attr); + return(result); } @@ -487,28 +778,47 @@ vm_map_create( * Allocates a VM map entry for insertion in the * given map (or map copy). No fields are filled. */ -#define vm_map_entry_create(map) \ - _vm_map_entry_create(&(map)->hdr) +#define vm_map_entry_create(map, map_locked) _vm_map_entry_create(&(map)->hdr, map_locked) -#define vm_map_copy_entry_create(copy) \ - _vm_map_entry_create(&(copy)->cpy_hdr) +#define vm_map_copy_entry_create(copy, map_locked) \ + _vm_map_entry_create(&(copy)->cpy_hdr, map_locked) +unsigned reserved_zalloc_count, nonreserved_zalloc_count; static vm_map_entry_t _vm_map_entry_create( - register struct vm_map_header *map_header) + struct vm_map_header *map_header, boolean_t __unused map_locked) { - register zone_t zone; - register vm_map_entry_t entry; + zone_t zone; + vm_map_entry_t entry; - if (map_header->entries_pageable) - zone = vm_map_entry_zone; - else - zone = vm_map_kentry_zone; + zone = vm_map_entry_zone; + + assert(map_header->entries_pageable ? !map_locked : TRUE); + + if (map_header->entries_pageable) { + entry = (vm_map_entry_t) zalloc(zone); + } + else { + entry = (vm_map_entry_t) zalloc_canblock(zone, FALSE); + + if (entry == VM_MAP_ENTRY_NULL) { + zone = vm_map_entry_reserved_zone; + entry = (vm_map_entry_t) zalloc(zone); + OSAddAtomic(1, &reserved_zalloc_count); + } else + OSAddAtomic(1, &nonreserved_zalloc_count); + } - entry = (vm_map_entry_t) zalloc(zone); if (entry == VM_MAP_ENTRY_NULL) panic("vm_map_entry_create"); + entry->from_reserved_zone = (zone == vm_map_entry_reserved_zone); + vm_map_store_update( (vm_map_t) NULL, entry, VM_MAP_ENTRY_CREATE); +#if MAP_ENTRY_CREATION_DEBUG + entry->vme_creation_maphdr = map_header; + fastbacktrace(&entry->vme_creation_bt[0], + (sizeof(entry->vme_creation_bt)/sizeof(uintptr_t))); +#endif return(entry); } @@ -516,15 +826,13 @@ _vm_map_entry_create( * vm_map_entry_dispose: [ internal use only ] * * Inverse of vm_map_entry_create. + * + * write map lock held so no need to + * do anything special to insure correctness + * of the stores */ #define vm_map_entry_dispose(map, entry) \ -MACRO_BEGIN \ - if((entry) == (map)->first_free) \ - (map)->first_free = vm_map_to_entry(map); \ - if((entry) == (map)->hint) \ - (map)->hint = vm_map_to_entry(map); \ - _vm_map_entry_dispose(&(map)->hdr, (entry)); \ -MACRO_END + _vm_map_entry_dispose(&(map)->hdr, (entry)) #define vm_map_copy_entry_dispose(map, entry) \ _vm_map_entry_dispose(&(copy)->cpy_hdr, (entry)) @@ -536,125 +844,40 @@ _vm_map_entry_dispose( { register zone_t zone; - if (map_header->entries_pageable) - zone = vm_map_entry_zone; + if (map_header->entries_pageable || !(entry->from_reserved_zone)) + zone = vm_map_entry_zone; else - zone = vm_map_kentry_zone; + zone = vm_map_entry_reserved_zone; + + if (!map_header->entries_pageable) { + if (zone == vm_map_entry_zone) + OSAddAtomic(-1, &nonreserved_zalloc_count); + else + OSAddAtomic(-1, &reserved_zalloc_count); + } zfree(zone, entry); } #if MACH_ASSERT -static boolean_t first_free_is_valid(vm_map_t map); /* forward */ static boolean_t first_free_check = FALSE; -static boolean_t +boolean_t first_free_is_valid( vm_map_t map) { - vm_map_entry_t entry, next; - if (!first_free_check) return TRUE; - - entry = vm_map_to_entry(map); - next = entry->vme_next; - while (vm_map_trunc_page(next->vme_start) == vm_map_trunc_page(entry->vme_end) || - (vm_map_trunc_page(next->vme_start) == vm_map_trunc_page(entry->vme_start) && - next != vm_map_to_entry(map))) { - entry = next; - next = entry->vme_next; - if (entry == vm_map_to_entry(map)) - break; - } - if (map->first_free != entry) { - printf("Bad first_free for map 0x%x: 0x%x should be 0x%x\n", - map, map->first_free, entry); - return FALSE; - } - return TRUE; + + return( first_free_is_valid_store( map )); } #endif /* MACH_ASSERT */ -/* - * UPDATE_FIRST_FREE: - * - * Updates the map->first_free pointer to the - * entry immediately before the first hole in the map. - * The map should be locked. - */ -#define UPDATE_FIRST_FREE(map, new_first_free) \ -MACRO_BEGIN \ - vm_map_t UFF_map; \ - vm_map_entry_t UFF_first_free; \ - vm_map_entry_t UFF_next_entry; \ - UFF_map = (map); \ - UFF_first_free = (new_first_free); \ - UFF_next_entry = UFF_first_free->vme_next; \ - while (vm_map_trunc_page(UFF_next_entry->vme_start) == \ - vm_map_trunc_page(UFF_first_free->vme_end) || \ - (vm_map_trunc_page(UFF_next_entry->vme_start) == \ - vm_map_trunc_page(UFF_first_free->vme_start) && \ - UFF_next_entry != vm_map_to_entry(UFF_map))) { \ - UFF_first_free = UFF_next_entry; \ - UFF_next_entry = UFF_first_free->vme_next; \ - if (UFF_first_free == vm_map_to_entry(UFF_map)) \ - break; \ - } \ - UFF_map->first_free = UFF_first_free; \ - assert(first_free_is_valid(UFF_map)); \ -MACRO_END - -/* - * vm_map_entry_{un,}link: - * - * Insert/remove entries from maps (or map copies). - */ -#define vm_map_entry_link(map, after_where, entry) \ -MACRO_BEGIN \ - vm_map_t VMEL_map; \ - vm_map_entry_t VMEL_entry; \ - VMEL_map = (map); \ - VMEL_entry = (entry); \ - _vm_map_entry_link(&VMEL_map->hdr, after_where, VMEL_entry); \ - UPDATE_FIRST_FREE(VMEL_map, VMEL_map->first_free); \ -MACRO_END - #define vm_map_copy_entry_link(copy, after_where, entry) \ - _vm_map_entry_link(&(copy)->cpy_hdr, after_where, (entry)) - -#define _vm_map_entry_link(hdr, after_where, entry) \ - MACRO_BEGIN \ - (hdr)->nentries++; \ - (entry)->vme_prev = (after_where); \ - (entry)->vme_next = (after_where)->vme_next; \ - (entry)->vme_prev->vme_next = (entry)->vme_next->vme_prev = (entry); \ - MACRO_END - -#define vm_map_entry_unlink(map, entry) \ -MACRO_BEGIN \ - vm_map_t VMEU_map; \ - vm_map_entry_t VMEU_entry; \ - vm_map_entry_t VMEU_first_free; \ - VMEU_map = (map); \ - VMEU_entry = (entry); \ - if (VMEU_entry->vme_start <= VMEU_map->first_free->vme_start) \ - VMEU_first_free = VMEU_entry->vme_prev; \ - else \ - VMEU_first_free = VMEU_map->first_free; \ - _vm_map_entry_unlink(&VMEU_map->hdr, VMEU_entry); \ - UPDATE_FIRST_FREE(VMEU_map, VMEU_first_free); \ -MACRO_END + _vm_map_store_entry_link(&(copy)->cpy_hdr, after_where, (entry)) #define vm_map_copy_entry_unlink(copy, entry) \ - _vm_map_entry_unlink(&(copy)->cpy_hdr, (entry)) - -#define _vm_map_entry_unlink(hdr, entry) \ - MACRO_BEGIN \ - (hdr)->nentries--; \ - (entry)->vme_next->vme_prev = (entry)->vme_prev; \ - (entry)->vme_prev->vme_next = (entry)->vme_next; \ - MACRO_END + _vm_map_store_entry_unlink(&(copy)->cpy_hdr, (entry)) #if MACH_ASSERT && TASK_SWAPPER /* @@ -672,10 +895,10 @@ void vm_map_res_reference(register vm_map_t map) assert(map->res_count >= 0); assert(map->ref_count >= map->res_count); if (map->res_count == 0) { - mutex_unlock(&map->s_lock); + lck_mtx_unlock(&map->s_lock); vm_map_lock(map); vm_map_swapin(map); - mutex_lock(&map->s_lock); + lck_mtx_lock(&map->s_lock); ++map->res_count; vm_map_unlock(map); } else @@ -693,12 +916,12 @@ void vm_map_res_reference(register vm_map_t map) void vm_map_reference_swap(register vm_map_t map) { assert(map != VM_MAP_NULL); - mutex_lock(&map->s_lock); + lck_mtx_lock(&map->s_lock); assert(map->res_count >= 0); assert(map->ref_count >= map->res_count); map->ref_count++; vm_map_res_reference(map); - mutex_unlock(&map->s_lock); + lck_mtx_unlock(&map->s_lock); } /* @@ -715,11 +938,11 @@ void vm_map_res_deallocate(register vm_map_t map) { assert(map->res_count > 0); if (--map->res_count == 0) { - mutex_unlock(&map->s_lock); + lck_mtx_unlock(&map->s_lock); vm_map_lock(map); vm_map_swapout(map); vm_map_unlock(map); - mutex_lock(&map->s_lock); + lck_mtx_lock(&map->s_lock); } assert(map->ref_count >= map->res_count); } @@ -732,21 +955,28 @@ void vm_map_res_deallocate(register vm_map_t map) */ void vm_map_destroy( - register vm_map_t map) + vm_map_t map, + int flags) { vm_map_lock(map); - (void) vm_map_delete(map, map->min_offset, - map->max_offset, VM_MAP_NO_FLAGS, - VM_MAP_NULL); + + /* clean up regular map entries */ + (void) vm_map_delete(map, map->min_offset, map->max_offset, + flags, VM_MAP_NULL); + /* clean up leftover special mappings (commpage, etc...) */ + (void) vm_map_delete(map, 0x0, 0xFFFFFFFFFFFFF000ULL, + flags, VM_MAP_NULL); + +#if CONFIG_FREEZE + if (map->default_freezer_handle) { + default_freezer_handle_deallocate(map->default_freezer_handle); + map->default_freezer_handle = NULL; + } +#endif vm_map_unlock(map); - -#ifdef __PPC__ - if (map->hdr.nentries!=0) - vm_map_remove_commpage64(map); -#endif /* __PPC__ */ - - assert(map->hdr.nentries==0); + assert(map->hdr.nentries == 0); + if(map->pmap) pmap_destroy(map->pmap); @@ -806,7 +1036,7 @@ int vm_map_swap_enable = 1; void vm_map_swapin (vm_map_t map) { register vm_map_entry_t entry; - + if (!vm_map_swap_enable) /* debug */ return; @@ -851,9 +1081,9 @@ void vm_map_swapin (vm_map_t map) if (entry->object.vm_object != VM_OBJECT_NULL) { if (entry->is_sub_map) { vm_map_t lmap = entry->object.sub_map; - mutex_lock(&lmap->s_lock); + lck_mtx_lock(&lmap->s_lock); vm_map_res_reference(lmap); - mutex_unlock(&lmap->s_lock); + lck_mtx_unlock(&lmap->s_lock); } else { vm_object_t object = entry->object.vm_object; vm_object_lock(object); @@ -881,12 +1111,12 @@ void vm_map_swapout(vm_map_t map) * If we raced with a swapin and lost, the residence count * will have been incremented to 1, and we simply return. */ - mutex_lock(&map->s_lock); + lck_mtx_lock(&map->s_lock); if (map->res_count != 0) { - mutex_unlock(&map->s_lock); + lck_mtx_unlock(&map->s_lock); return; } - mutex_unlock(&map->s_lock); + lck_mtx_unlock(&map->s_lock); /* * There are no intermediate states of a map going out or @@ -912,9 +1142,9 @@ void vm_map_swapout(vm_map_t map) if (entry->object.vm_object != VM_OBJECT_NULL) { if (entry->is_sub_map) { vm_map_t lmap = entry->object.sub_map; - mutex_lock(&lmap->s_lock); + lck_mtx_lock(&lmap->s_lock); vm_map_res_deallocate(lmap); - mutex_unlock(&lmap->s_lock); + lck_mtx_unlock(&lmap->s_lock); } else { vm_object_t object = entry->object.vm_object; vm_object_lock(object); @@ -936,26 +1166,12 @@ void vm_map_swapout(vm_map_t map) #endif /* TASK_SWAPPER */ - -/* - * SAVE_HINT: - * - * Saves the specified entry as the hint for - * future lookups. Performs necessary interlocks. - */ -#define SAVE_HINT(map,value) \ -MACRO_BEGIN \ - mutex_lock(&(map)->s_lock); \ - (map)->hint = (value); \ - mutex_unlock(&(map)->s_lock); \ -MACRO_END - /* * vm_map_lookup_entry: [ internal use only ] * - * Finds the map entry containing (or - * immediately preceding) the specified address - * in the given map; the entry is returned + * Calls into the vm map store layer to find the map + * entry containing (or immediately preceding) the + * specified address in the given map; the entry is returned * in the "entry" parameter. The boolean * result indicates whether the address is * actually contained in the map. @@ -966,74 +1182,8 @@ vm_map_lookup_entry( register vm_map_offset_t address, vm_map_entry_t *entry) /* OUT */ { - register vm_map_entry_t cur; - register vm_map_entry_t last; - - /* - * Start looking either from the head of the - * list, or from the hint. - */ - if (not_in_kdp) - mutex_lock(&map->s_lock); - cur = map->hint; - if (not_in_kdp) - mutex_unlock(&map->s_lock); - - if (cur == vm_map_to_entry(map)) - cur = cur->vme_next; - - if (address >= cur->vme_start) { - /* - * Go from hint to end of list. - * - * But first, make a quick check to see if - * we are already looking at the entry we - * want (which is usually the case). - * Note also that we don't need to save the hint - * here... it is the same hint (unless we are - * at the header, in which case the hint didn't - * buy us anything anyway). - */ - last = vm_map_to_entry(map); - if ((cur != last) && (cur->vme_end > address)) { - *entry = cur; - return(TRUE); - } - } - else { - /* - * Go from start to hint, *inclusively* - */ - last = cur->vme_next; - cur = vm_map_first_entry(map); - } - - /* - * Search linearly - */ - - while (cur != last) { - if (cur->vme_end > address) { - if (address >= cur->vme_start) { - /* - * Save this lookup for future - * hints, and return - */ - - *entry = cur; - if (not_in_kdp) - SAVE_HINT(map, cur); - return(TRUE); - } - break; - } - cur = cur->vme_next; - } - *entry = cur->vme_prev; - if (not_in_kdp) - SAVE_HINT(map, *entry); - return(FALSE); -} + return ( vm_map_store_lookup_entry( map, address, entry )); +} /* * Routine: vm_map_find_space @@ -1054,6 +1204,7 @@ vm_map_find_space( vm_map_offset_t *address, /* OUT */ vm_map_size_t size, vm_map_offset_t mask, + int flags, vm_map_entry_t *o_entry) /* OUT */ { register vm_map_entry_t entry, new_entry; @@ -1065,7 +1216,12 @@ vm_map_find_space( return KERN_INVALID_ARGUMENT; } - new_entry = vm_map_entry_create(map); + if (flags & VM_FLAGS_GUARD_AFTER) { + /* account for the back guard page in the size */ + size += VM_MAP_PAGE_SIZE(map); + } + + new_entry = vm_map_entry_create(map, FALSE); /* * Look for the first possible address; if there's already @@ -1074,11 +1230,15 @@ vm_map_find_space( vm_map_lock(map); - assert(first_free_is_valid(map)); - if ((entry = map->first_free) == vm_map_to_entry(map)) - start = map->min_offset; - else - start = entry->vme_end; + if( map->disable_vmentry_reuse == TRUE) { + VM_MAP_HIGHEST_ENTRY(map, entry, start); + } else { + assert(first_free_is_valid(map)); + if ((entry = map->first_free) == vm_map_to_entry(map)) + start = map->min_offset; + else + start = entry->vme_end; + } /* * In any case, the "entry" always precedes @@ -1094,7 +1254,12 @@ vm_map_find_space( * wrap around the address. */ + if (flags & VM_FLAGS_GUARD_BEFORE) { + /* reserve space for the front guard page */ + start += VM_MAP_PAGE_SIZE(map); + } end = ((start + mask) & ~mask); + if (end < start) { vm_map_entry_dispose(map, new_entry); vm_map_unlock(map); @@ -1143,12 +1308,21 @@ vm_map_find_space( * the map should be locked. */ + if (flags & VM_FLAGS_GUARD_BEFORE) { + /* go back for the front guard page */ + start -= VM_MAP_PAGE_SIZE(map); + } *address = start; + assert(start < end); new_entry->vme_start = start; new_entry->vme_end = end; assert(page_aligned(new_entry->vme_start)); assert(page_aligned(new_entry->vme_end)); + assert(VM_MAP_PAGE_ALIGNED(new_entry->vme_start, + VM_MAP_PAGE_MASK(map))); + assert(VM_MAP_PAGE_ALIGNED(new_entry->vme_end, + VM_MAP_PAGE_MASK(map))); new_entry->is_shared = FALSE; new_entry->is_sub_map = FALSE; @@ -1167,19 +1341,34 @@ vm_map_find_space( new_entry->in_transition = FALSE; new_entry->needs_wakeup = FALSE; + new_entry->no_cache = FALSE; + new_entry->permanent = FALSE; + new_entry->superpage_size = FALSE; + if (VM_MAP_PAGE_SHIFT(map) != PAGE_SHIFT) { + new_entry->map_aligned = TRUE; + } else { + new_entry->map_aligned = FALSE; + } + + new_entry->used_for_jit = 0; + + new_entry->alias = 0; + new_entry->zero_wired_pages = FALSE; + + VM_GET_FLAGS_ALIAS(flags, new_entry->alias); /* * Insert the new entry into the list */ - vm_map_entry_link(map, entry, new_entry); + vm_map_store_entry_link(map, entry, new_entry); map->size += size; /* * Update the lookup hint */ - SAVE_HINT(map, new_entry); + SAVE_HINT_MAP_WRITE(map, new_entry); *o_entry = new_entry; return(KERN_SUCCESS); @@ -1211,7 +1400,8 @@ vm_map_pmap_enter( vm_object_offset_t offset, vm_prot_t protection) { - unsigned int cache_attr; + int type_of_fault; + kern_return_t kr; if(map->pmap == 0) return; @@ -1220,7 +1410,6 @@ vm_map_pmap_enter( register vm_page_t m; vm_object_lock(object); - vm_object_paging_begin(object); m = vm_page_lookup(object, offset); /* @@ -1229,42 +1418,22 @@ vm_map_pmap_enter( * enter an encrypted page in the page table. */ if (m == VM_PAGE_NULL || m->busy || m->encrypted || - (m->unusual && ( m->error || m->restart || m->absent || - protection & m->page_lock))) { - - vm_object_paging_end(object); + m->fictitious || + (m->unusual && ( m->error || m->restart || m->absent))) { vm_object_unlock(object); return; } - assert(!m->fictitious); /* XXX is this possible ??? */ - if (vm_map_pmap_enter_print) { printf("vm_map_pmap_enter:"); - printf("map: %x, addr: %llx, object: %x, offset: %llx\n", - map, (unsigned long long)addr, object, (unsigned long long)offset); + printf("map: %p, addr: %llx, object: %p, offset: %llx\n", + map, (unsigned long long)addr, object, (unsigned long long)offset); } - m->busy = TRUE; - - if (m->no_isync == TRUE) { - pmap_sync_page_data_phys(m->phys_page); - m->no_isync = FALSE; - } - - cache_attr = ((unsigned int)object->wimg_bits) & VM_WIMG_MASK; - vm_object_unlock(object); + 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, + &type_of_fault); - PMAP_ENTER(map->pmap, addr, m, - protection, cache_attr, FALSE); - - vm_object_lock(object); - - PAGE_WAKEUP_DONE(m); - vm_page_lock_queues(); - if (!m->active && !m->inactive) - vm_page_activate(m); - vm_page_unlock_queues(); - vm_object_paging_end(object); vm_object_unlock(object); offset += PAGE_SIZE_64; @@ -1281,12 +1450,16 @@ boolean_t vm_map_pmap_is_empty( vm_map_offset_t start, vm_map_offset_t end) { +#ifdef MACHINE_PMAP_IS_EMPTY + return pmap_is_empty(map->pmap, start, end); +#else /* MACHINE_PMAP_IS_EMPTY */ vm_map_offset_t offset; ppnum_t phys_page; if (map->pmap == NULL) { return TRUE; } + for (offset = start; offset < end; offset += PAGE_SIZE) { @@ -1294,11 +1467,66 @@ boolean_t vm_map_pmap_is_empty( if (phys_page) { kprintf("vm_map_pmap_is_empty(%p,0x%llx,0x%llx): " "page %d at 0x%llx\n", - map, start, end, phys_page, offset); + map, (long long)start, (long long)end, + phys_page, (long long)offset); return FALSE; } } return TRUE; +#endif /* MACHINE_PMAP_IS_EMPTY */ +} + +#define MAX_TRIES_TO_GET_RANDOM_ADDRESS 1000 +kern_return_t +vm_map_random_address_for_size( + vm_map_t map, + vm_map_offset_t *address, + vm_map_size_t size) +{ + kern_return_t kr = KERN_SUCCESS; + int tries = 0; + vm_map_offset_t random_addr = 0; + vm_map_offset_t hole_end; + + vm_map_entry_t next_entry = VM_MAP_ENTRY_NULL; + vm_map_entry_t prev_entry = VM_MAP_ENTRY_NULL; + vm_map_size_t vm_hole_size = 0; + vm_map_size_t addr_space_size; + + addr_space_size = vm_map_max(map) - vm_map_min(map); + + assert(page_aligned(size)); + + while (tries < MAX_TRIES_TO_GET_RANDOM_ADDRESS) { + random_addr = ((vm_map_offset_t)random()) << PAGE_SHIFT; + random_addr = vm_map_trunc_page( + vm_map_min(map) +(random_addr % addr_space_size), + VM_MAP_PAGE_MASK(map)); + + if (vm_map_lookup_entry(map, random_addr, &prev_entry) == FALSE) { + if (prev_entry == vm_map_to_entry(map)) { + next_entry = vm_map_first_entry(map); + } else { + next_entry = prev_entry->vme_next; + } + if (next_entry == vm_map_to_entry(map)) { + hole_end = vm_map_max(map); + } else { + hole_end = next_entry->vme_start; + } + vm_hole_size = hole_end - random_addr; + if (vm_hole_size >= size) { + *address = random_addr; + break; + } + } + tries++; + } + + if (tries == MAX_TRIES_TO_GET_RANDOM_ADDRESS) { + kr = KERN_NO_SPACE; + } + return kr; } /* @@ -1317,9 +1545,9 @@ static unsigned int vm_map_enter_restore_failures = 0; kern_return_t vm_map_enter( vm_map_t map, - vm_map_offset_t *address, /* IN/OUT */ + vm_map_offset_t *address, /* IN/OUT */ vm_map_size_t size, - vm_map_offset_t mask, + vm_map_offset_t mask, int flags, vm_object_t object, vm_object_offset_t offset, @@ -1329,8 +1557,10 @@ vm_map_enter( vm_inherit_t inheritance) { vm_map_entry_t entry, new_entry; - vm_map_offset_t start, tmp_start; + vm_map_offset_t start, tmp_start, tmp_offset; vm_map_offset_t end, tmp_end; + vm_map_offset_t tmp2_start, tmp2_end; + vm_map_offset_t step; kern_return_t result = KERN_SUCCESS; vm_map_t zap_old_map = VM_MAP_NULL; vm_map_t zap_new_map = VM_MAP_NULL; @@ -1340,9 +1570,83 @@ vm_map_enter( 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); + unsigned int superpage_size = ((flags & VM_FLAGS_SUPERPAGE_MASK) >> VM_FLAGS_SUPERPAGE_SHIFT); char alias; + vm_map_offset_t effective_min_offset, effective_max_offset; + kern_return_t kr; + boolean_t clear_map_aligned = FALSE; - if (size == 0) { + if (superpage_size) { + switch (superpage_size) { + /* + * Note that the current implementation only supports + * a single size for superpages, SUPERPAGE_SIZE, per + * architecture. As soon as more sizes are supposed + * to be supported, SUPERPAGE_SIZE has to be replaced + * with a lookup of the size depending on superpage_size. + */ +#ifdef __x86_64__ + case SUPERPAGE_SIZE_ANY: + /* handle it like 2 MB and round up to page size */ + size = (size + 2*1024*1024 - 1) & ~(2*1024*1024 - 1); + case SUPERPAGE_SIZE_2MB: + break; +#endif + default: + return KERN_INVALID_ARGUMENT; + } + mask = SUPERPAGE_SIZE-1; + if (size & (SUPERPAGE_SIZE-1)) + return KERN_INVALID_ARGUMENT; + inheritance = VM_INHERIT_NONE; /* fork() children won't inherit superpages */ + } + + + + if (is_submap) { + if (purgable) { + /* submaps can not be purgeable */ + return KERN_INVALID_ARGUMENT; + } + if (object == VM_OBJECT_NULL) { + /* submaps can not be created lazily */ + return KERN_INVALID_ARGUMENT; + } + } + if (flags & VM_FLAGS_ALREADY) { + /* + * VM_FLAGS_ALREADY says that it's OK if the same mapping + * is already present. For it to be meaningul, the requested + * mapping has to be at a fixed address (!VM_FLAGS_ANYWHERE) and + * we shouldn't try and remove what was mapped there first + * (!VM_FLAGS_OVERWRITE). + */ + if ((flags & VM_FLAGS_ANYWHERE) || + (flags & VM_FLAGS_OVERWRITE)) { + return KERN_INVALID_ARGUMENT; + } + } + + effective_min_offset = map->min_offset; + + if (flags & VM_FLAGS_BEYOND_MAX) { + /* + * Allow an insertion beyond the map's max offset. + */ + if (vm_map_is_64bit(map)) + effective_max_offset = 0xFFFFFFFFFFFFF000ULL; + else + effective_max_offset = 0x00000000FFFFF000ULL; + } else { + effective_max_offset = map->max_offset; + } + + if (size == 0 || + (offset & PAGE_MASK_64) != 0) { *address = 0; return KERN_INVALID_ARGUMENT; } @@ -1354,6 +1658,18 @@ vm_map_enter( assert(page_aligned(*address)); assert(page_aligned(size)); + if (!VM_MAP_PAGE_ALIGNED(size, VM_MAP_PAGE_MASK(map))) { + /* + * In most cases, the caller rounds the size up to the + * map's page size. + * If we get a size that is explicitly not map-aligned here, + * we'll have to respect the caller's wish and mark the + * mapping as "not map-aligned" to avoid tripping the + * map alignment checks later. + */ + clear_map_aligned = TRUE; + } + /* * Only zero-fill objects are allowed to be purgable. * LP64todo - limit purgable objects to 32-bits for now @@ -1361,9 +1677,9 @@ vm_map_enter( if (purgable && (offset != 0 || (object != VM_OBJECT_NULL && - (object->size != size || - object->purgable == VM_OBJECT_NONPURGABLE)) - || size > VM_MAX_ADDRESS)) /* LP64todo: remove when dp capable */ + (object->vo_size != size || + object->purgable == VM_PURGABLE_DENY)) + || size > ANON_MAX_SIZE)) /* LP64todo: remove when dp capable */ return KERN_INVALID_ARGUMENT; if (!anywhere && overwrite) { @@ -1379,24 +1695,41 @@ vm_map_enter( zap_old_map = vm_map_create(PMAP_NULL, *address, *address + size, - TRUE); + map->hdr.entries_pageable); + vm_map_set_page_shift(zap_old_map, VM_MAP_PAGE_SHIFT(map)); } - StartAgain: ; +StartAgain: ; start = *address; if (anywhere) { vm_map_lock(map); map_locked = TRUE; + + if (entry_for_jit) { + if (map->jit_entry_exists) { + result = KERN_INVALID_ARGUMENT; + goto BailOut; + } + /* + * Get a random start address. + */ + result = vm_map_random_address_for_size(map, address, size); + if (result != KERN_SUCCESS) { + goto BailOut; + } + start = *address; + } + /* * Calculate the first possible address. */ - if (start < map->min_offset) - start = map->min_offset; - if (start > map->max_offset) + if (start < effective_min_offset) + start = effective_min_offset; + if (start > effective_max_offset) RETURN(KERN_NO_SPACE); /* @@ -1405,15 +1738,45 @@ vm_map_enter( * address, we have to start after it. */ - assert(first_free_is_valid(map)); - if (start == map->min_offset) { - if ((entry = map->first_free) != vm_map_to_entry(map)) - start = entry->vme_end; + if( map->disable_vmentry_reuse == TRUE) { + VM_MAP_HIGHEST_ENTRY(map, entry, start); } else { - vm_map_entry_t tmp_entry; - if (vm_map_lookup_entry(map, start, &tmp_entry)) - start = tmp_entry->vme_end; - entry = tmp_entry; + assert(first_free_is_valid(map)); + + entry = map->first_free; + + if (entry == vm_map_to_entry(map)) { + entry = NULL; + } else { + if (entry->vme_next == vm_map_to_entry(map)){ + /* + * Hole at the end of the map. + */ + entry = NULL; + } else { + if (start < (entry->vme_next)->vme_start ) { + start = entry->vme_end; + start = vm_map_round_page(start, + VM_MAP_PAGE_MASK(map)); + } else { + /* + * Need to do a lookup. + */ + entry = NULL; + } + } + } + + if (entry == NULL) { + vm_map_entry_t tmp_entry; + if (vm_map_lookup_entry(map, start, &tmp_entry)) { + assert(!entry_for_jit); + start = tmp_entry->vme_end; + start = vm_map_round_page(start, + VM_MAP_PAGE_MASK(map)); + } + entry = tmp_entry; + } } /* @@ -1425,22 +1788,26 @@ vm_map_enter( while (TRUE) { register vm_map_entry_t next; - /* + /* * Find the end of the proposed new region. * Be sure we didn't go beyond the end, or * wrap around the address. */ end = ((start + mask) & ~mask); + end = vm_map_round_page(end, + VM_MAP_PAGE_MASK(map)); if (end < start) RETURN(KERN_NO_SPACE); start = end; + assert(VM_MAP_PAGE_ALIGNED(start, + VM_MAP_PAGE_MASK(map))); end += size; - if ((end > map->max_offset) || (end < start)) { + if ((end > effective_max_offset) || (end < start)) { if (map->wait_for_space) { - if (size <= (map->max_offset - - map->min_offset)) { + if (size <= (effective_max_offset - + effective_min_offset)) { assert_wait((event_t)map, THREAD_ABORTSAFE); vm_map_unlock(map); @@ -1474,11 +1841,13 @@ vm_map_enter( entry = next; start = entry->vme_end; + start = vm_map_round_page(start, + VM_MAP_PAGE_MASK(map)); } *address = start; + assert(VM_MAP_PAGE_ALIGNED(*address, + VM_MAP_PAGE_MASK(map))); } else { - vm_map_entry_t temp_entry; - /* * Verify that: * the address doesn't itself violate @@ -1496,8 +1865,8 @@ vm_map_enter( end = start + size; - if ((start < map->min_offset) || - (end > map->max_offset) || + if ((start < effective_min_offset) || + (end > effective_max_offset) || (start >= end)) { RETURN(KERN_INVALID_ADDRESS); } @@ -1517,10 +1886,78 @@ vm_map_enter( * ... the starting address isn't allocated */ - if (vm_map_lookup_entry(map, start, &temp_entry)) - RETURN(KERN_NO_SPACE); + if (vm_map_lookup_entry(map, start, &entry)) { + if (! (flags & VM_FLAGS_ALREADY)) { + RETURN(KERN_NO_SPACE); + } + /* + * Check if what's already there is what we want. + */ + tmp_start = start; + tmp_offset = offset; + if (entry->vme_start < start) { + tmp_start -= start - entry->vme_start; + tmp_offset -= start - entry->vme_start; + + } + for (; entry->vme_start < end; + entry = entry->vme_next) { + /* + * Check if the mapping's attributes + * match the existing map entry. + */ + if (entry == vm_map_to_entry(map) || + entry->vme_start != tmp_start || + entry->is_sub_map != is_submap || + entry->offset != tmp_offset || + entry->needs_copy != needs_copy || + entry->protection != cur_protection || + entry->max_protection != max_protection || + entry->inheritance != inheritance || + entry->alias != 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 != + (vm_map_t) object) { + /* not the same submap */ + RETURN(KERN_NO_SPACE); + } + } else { + if (entry->object.vm_object != object) { + /* not the same VM object... */ + vm_object_t obj2; + + obj2 = entry->object.vm_object; + if ((obj2 == VM_OBJECT_NULL || + obj2->internal) && + (object == VM_OBJECT_NULL || + object->internal)) { + /* + * ... but both are + * anonymous memory, + * so equivalent. + */ + } else { + RETURN(KERN_NO_SPACE); + } + } + } - entry = temp_entry; + tmp_offset += entry->vme_end - entry->vme_start; + tmp_start += entry->vme_end - entry->vme_start; + if (entry->vme_end >= end) { + /* reached the end of our mapping */ + break; + } + } + /* it all matches: let's use what's already there ! */ + RETURN(KERN_MEMORY_PRESENT); + } /* * ... the next region doesn't overlap the @@ -1551,32 +1988,44 @@ vm_map_enter( * semantics. */ - if (purgable) { + if (purgable || entry_for_jit) { if (object == VM_OBJECT_NULL) { object = vm_object_allocate(size); object->copy_strategy = MEMORY_OBJECT_COPY_NONE; - object->purgable = VM_OBJECT_PURGABLE_NONVOLATILE; + if (purgable) { + object->purgable = VM_PURGABLE_NONVOLATILE; + } offset = (vm_object_offset_t)0; } - } else if ((object == VM_OBJECT_NULL) && - (entry != vm_map_to_entry(map)) && - (entry->vme_end == start) && - (!entry->is_shared) && - (!entry->is_sub_map) && - (entry->alias == alias) && - (entry->inheritance == inheritance) && - (entry->protection == cur_protection) && - (entry->max_protection == max_protection) && - (entry->behavior == VM_BEHAVIOR_DEFAULT) && - (entry->in_transition == 0) && - ((alias == VM_MEMORY_REALLOC) || ((entry->vme_end - entry->vme_start) + size < NO_COALESCE_LIMIT)) && - (entry->wired_count == 0)) { /* implies user_wired_count == 0 */ + } else if ((is_submap == FALSE) && + (object == VM_OBJECT_NULL) && + (entry != vm_map_to_entry(map)) && + (entry->vme_end == start) && + (!entry->is_shared) && + (!entry->is_sub_map) && + ((alias == VM_MEMORY_REALLOC) || (entry->alias == alias)) && + (entry->inheritance == inheritance) && + (entry->protection == cur_protection) && + (entry->max_protection == max_protection) && + (entry->behavior == VM_BEHAVIOR_DEFAULT) && + (entry->in_transition == 0) && + (entry->no_cache == no_cache) && + /* + * No coalescing if not map-aligned, to avoid propagating + * that condition any further than needed: + */ + (!entry->map_aligned || !clear_map_aligned) && + ((entry->vme_end - entry->vme_start) + size <= + (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, - VM_OBJECT_NULL, - entry->offset, - (vm_object_offset_t) 0, - (vm_map_size_t)(entry->vme_end - entry->vme_start), - (vm_map_size_t)(end - entry->vme_end))) { + VM_OBJECT_NULL, + entry->offset, + (vm_object_offset_t) 0, + (vm_map_size_t)(entry->vme_end - entry->vme_start), + (vm_map_size_t)(end - entry->vme_end))) { /* * Coalesced the two objects - can extend @@ -1584,35 +2033,139 @@ vm_map_enter( * new range. */ map->size += (end - entry->vme_end); + assert(entry->vme_start < end); + assert(VM_MAP_PAGE_ALIGNED(end, + VM_MAP_PAGE_MASK(map))); entry->vme_end = end; - UPDATE_FIRST_FREE(map, map->first_free); + vm_map_store_update_first_free(map, map->first_free); RETURN(KERN_SUCCESS); } } - /* - * 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. - */ - tmp_start = start; - if (object == VM_OBJECT_NULL && size > (vm_map_size_t)VM_MAX_ADDRESS) - tmp_end = tmp_start + (vm_map_size_t)VM_MAX_ADDRESS; - else - tmp_end = 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, inheritance, 0); - new_entry->alias = alias; - entry = new_entry; - } while (object == VM_OBJECT_NULL && - tmp_end != end && - (tmp_start = tmp_end) && - (tmp_end = (end - tmp_end > (vm_map_size_t)VM_MAX_ADDRESS) ? - tmp_end + (vm_map_size_t)VM_MAX_ADDRESS : end)); + step = superpage_size ? SUPERPAGE_SIZE : (end - start); + new_entry = NULL; + + for (tmp2_start = start; tmp2_start (vm_map_size_t)ANON_CHUNK_SIZE && + max_protection != VM_PROT_NONE && + superpage_size == 0) + tmp_end = tmp_start + (vm_map_size_t)ANON_CHUNK_SIZE; + else + tmp_end = tmp2_end; + do { + new_entry = vm_map_entry_insert(map, entry, tmp_start, tmp_end, + object, offset, needs_copy, + FALSE, FALSE, + cur_protection, max_protection, + VM_BEHAVIOR_DEFAULT, + (entry_for_jit)? VM_INHERIT_NONE: inheritance, + 0, no_cache, + permanent, + superpage_size, + clear_map_aligned); + new_entry->alias = alias; + if (entry_for_jit){ + if (!(map->jit_entry_exists)){ + new_entry->used_for_jit = TRUE; + map->jit_entry_exists = TRUE; + } + } + + if (is_submap) { + vm_map_t submap; + boolean_t submap_is_64bit; + boolean_t use_pmap; + + new_entry->is_sub_map = TRUE; + submap = (vm_map_t) object; + submap_is_64bit = vm_map_is_64bit(submap); + use_pmap = (alias == VM_MEMORY_SHARED_PMAP); + #ifndef NO_NESTED_PMAP + if (use_pmap && submap->pmap == NULL) { + ledger_t ledger = map->pmap->ledger; + /* we need a sub pmap to nest... */ + submap->pmap = pmap_create(ledger, 0, + submap_is_64bit); + if (submap->pmap == NULL) { + /* let's proceed without nesting... */ + } + } + if (use_pmap && submap->pmap != NULL) { + kr = pmap_nest(map->pmap, + submap->pmap, + tmp_start, + tmp_start, + tmp_end - tmp_start); + if (kr != KERN_SUCCESS) { + printf("vm_map_enter: " + "pmap_nest(0x%llx,0x%llx) " + "error 0x%x\n", + (long long)tmp_start, + (long long)tmp_end, + kr); + } else { + /* we're now nested ! */ + new_entry->use_pmap = TRUE; + pmap_empty = FALSE; + } + } + #endif /* NO_NESTED_PMAP */ + } + entry = new_entry; + + if (superpage_size) { + vm_page_t pages, m; + vm_object_t sp_object; + + entry->offset = 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 */ + 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; + + /* enter the base pages into the object */ + vm_object_lock(sp_object); + for (offset = 0; offset < SUPERPAGE_SIZE; offset += PAGE_SIZE) { + m = pages; + pmap_zero_page(m->phys_page); + pages = NEXT_PAGE(m); + *(NEXT_PAGE_PTR(m)) = VM_PAGE_NULL; + vm_page_insert(m, sp_object, offset); + } + vm_object_unlock(sp_object); + } + } while (tmp_end != tmp2_end && + (tmp_start = tmp_end) && + (tmp_end = (tmp2_end - tmp_end > (vm_map_size_t)ANON_CHUNK_SIZE) ? + tmp_end + (vm_map_size_t)ANON_CHUNK_SIZE : tmp2_end)); + } vm_map_unlock(map); map_locked = FALSE; @@ -1622,11 +2175,11 @@ vm_map_enter( /* Wire down the new entry if the user * requested all new map entries be wired. */ - if (map->wiring_required) { + if ((map->wiring_required)||(superpage_size)) { pmap_empty = FALSE; /* pmap won't be empty */ - result = vm_map_wire(map, start, end, + kr = vm_map_wire(map, start, end, new_entry->protection, TRUE); - RETURN(result); + RETURN(kr); } if ((object != VM_OBJECT_NULL) && @@ -1635,18 +2188,65 @@ vm_map_enter( (!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; + vm_map_pmap_enter(map, start, end, object, offset, cur_protection); } - BailOut: ; - if (result == KERN_SUCCESS && - pmap_empty && - !(flags & VM_FLAGS_NO_PMAP_CHECK)) { - assert(vm_map_pmap_is_empty(map, *address, *address+size)); - } +BailOut: ; + if (result == KERN_SUCCESS) { + vm_prot_t pager_prot; + memory_object_t pager; - if (result != KERN_SUCCESS) { + if (pmap_empty && + !(flags & VM_FLAGS_NO_PMAP_CHECK)) { + assert(vm_map_pmap_is_empty(map, + *address, + *address+size)); + } + + /* + * 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); + } + } else { if (new_mapping_established) { /* * We have to get rid of the new mappings since we @@ -1657,7 +2257,9 @@ vm_map_enter( zap_new_map = vm_map_create(PMAP_NULL, *address, *address + size, - TRUE); + map->hdr.entries_pageable); + vm_map_set_page_shift(zap_new_map, + VM_MAP_PAGE_SHIFT(map)); if (!map_locked) { vm_map_lock(map); map_locked = TRUE; @@ -1700,9 +2302,15 @@ vm_map_enter( 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_entry_unlink(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); - vm_map_entry_link(map, entry1, entry2); + zap_old_map->size -= entry_size; + vm_map_store_entry_link(map, entry1, entry2); + map->size += entry_size; entry1 = entry2; } if (map->wiring_required) { @@ -1725,11 +2333,11 @@ vm_map_enter( * they may still contain. */ if (zap_old_map != VM_MAP_NULL) { - vm_map_destroy(zap_old_map); + 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_destroy(zap_new_map, VM_MAP_REMOVE_NO_PMAP_CLEANUP); zap_new_map = VM_MAP_NULL; } @@ -1738,70 +2346,765 @@ vm_map_enter( #undef RETURN } - -#if VM_CPM - -#ifdef MACH_ASSERT -extern vm_offset_t avail_start, avail_end; -#endif - -/* - * Allocate memory in the specified map, with the caveat that - * the memory is physically contiguous. This call may fail - * if the system can't find sufficient contiguous memory. - * This call may cause or lead to heart-stopping amounts of - * paging activity. - * - * Memory obtained from this call should be freed in the - * normal way, viz., via vm_deallocate. - */ kern_return_t -vm_map_enter_cpm( - vm_map_t map, - vm_map_offset_t *addr, - vm_map_size_t size, - int flags) +vm_map_enter_mem_object( + vm_map_t target_map, + vm_map_offset_t *address, + vm_map_size_t initial_size, + vm_map_offset_t mask, + int flags, + ipc_port_t port, + vm_object_offset_t offset, + boolean_t copy, + vm_prot_t cur_protection, + vm_prot_t max_protection, + vm_inherit_t inheritance) { - vm_object_t cpm_obj; - pmap_t pmap; - vm_page_t m, pages; - kern_return_t kr; - vm_map_offset_t va, start, end, offset; -#if MACH_ASSERT - vm_map_offset_t prev_addr; -#endif /* MACH_ASSERT */ + 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; - boolean_t anywhere = ((VM_FLAGS_ANYWHERE & flags) != 0); + 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 (!vm_allocate_cpm_enabled) - return KERN_FAILURE; + /* + * Check arguments for validity + */ + if ((target_map == VM_MAP_NULL) || + (cur_protection & ~VM_PROT_ALL) || + (max_protection & ~VM_PROT_ALL) || + (inheritance > VM_INHERIT_LAST_VALID) || + initial_size == 0) + return KERN_INVALID_ARGUMENT; + + 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 (size == 0) { - *addr = 0; - return KERN_SUCCESS; - } + /* + * 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 (anywhere) - *addr = vm_map_min(map); - else - *addr = vm_map_trunc_page(*addr); - size = vm_map_round_page(size); + named_entry = (vm_named_entry_t) port->ip_kobject; - /* - * LP64todo - cpm_allocate should probably allow + 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 (offset >= named_entry->size) + return KERN_INVALID_RIGHT; + size = named_entry->size - offset; + } + if (mask_max_protection) { + max_protection &= named_entry->protection; + } + if (mask_cur_protection) { + cur_protection &= named_entry->protection; + } + if ((named_entry->protection & max_protection) != + max_protection) + return KERN_INVALID_RIGHT; + if ((named_entry->protection & cur_protection) != + cur_protection) + return KERN_INVALID_RIGHT; + if (offset + size < offset) { + /* overflow */ + return KERN_INVALID_ARGUMENT; + } + if (named_entry->size < (offset + size)) + return KERN_INVALID_ARGUMENT; + + if (named_entry->is_copy) { + /* for a vm_map_copy, we can only map it whole */ + if ((size != named_entry->size) && + (vm_map_round_page(size, + VM_MAP_PAGE_MASK(target_map)) == + named_entry->size)) { + /* XXX FBDP use the rounded size... */ + size = vm_map_round_page( + size, + VM_MAP_PAGE_MASK(target_map)); + } + + if (offset != 0 || + size != named_entry->size) { + 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))) { + /* + * Let's not map more than requested; + * vm_map_enter() will handle this "not map-aligned" + * case. + */ + map_size = size; + } + + named_entry_lock(named_entry); + if (named_entry->is_sub_map) { + vm_map_t submap; + + if ((flags & VM_FLAGS_RETURN_DATA_ADDR) != 0) { + panic("VM_FLAGS_RETURN_DATA_ADDR not expected for submap."); + } + + submap = named_entry->backing.map; + vm_map_lock(submap); + vm_map_reference(submap); + vm_map_unlock(submap); + named_entry_unlock(named_entry); + + result = vm_map_enter(target_map, + &map_addr, + map_size, + mask, + flags | VM_FLAGS_SUBMAP, + (vm_object_t) submap, + offset, + copy, + cur_protection, + max_protection, + inheritance); + if (result != KERN_SUCCESS) { + vm_map_deallocate(submap); + } else { + /* + * No need to lock "submap" just to check its + * "mapped" flag: that flag is never reset + * once it's been set and if we race, we'll + * just end up setting it twice, which is OK. + */ + if (submap->mapped_in_other_pmaps == FALSE && + vm_map_pmap(submap) != PMAP_NULL && + vm_map_pmap(submap) != + vm_map_pmap(target_map)) { + /* + * This submap is being mapped in a map + * that uses a different pmap. + * Set its "mapped_in_other_pmaps" flag + * to indicate that we now need to + * remove mappings from all pmaps rather + * than just the submap's pmap. + */ + vm_map_lock(submap); + submap->mapped_in_other_pmaps = TRUE; + vm_map_unlock(submap); + } + *address = map_addr; + } + 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; + vm_map_entry_t copy_entry; + vm_map_offset_t copy_addr; + + if (flags & ~(VM_FLAGS_FIXED | + VM_FLAGS_ANYWHERE | + VM_FLAGS_OVERWRITE | + VM_FLAGS_RETURN_DATA_ADDR)) { + named_entry_unlock(named_entry); + return KERN_INVALID_ARGUMENT; + } + + if ((flags & VM_FLAGS_RETURN_DATA_ADDR) != 0) { + offset_in_mapping = offset - vm_object_trunc_page(offset); + offset = vm_object_trunc_page(offset); + map_size = vm_object_round_page(offset + offset_in_mapping + initial_size) - offset; + } + + copy_map = named_entry->backing.copy; + assert(copy_map->type == VM_MAP_COPY_ENTRY_LIST); + if (copy_map->type != VM_MAP_COPY_ENTRY_LIST) { + /* unsupported type; should not happen */ + printf("vm_map_enter_mem_object: " + "memory_entry->backing.copy " + "unsupported type 0x%x\n", + copy_map->type); + named_entry_unlock(named_entry); + return KERN_INVALID_ARGUMENT; + } + + /* reserve a contiguous range */ + kr = vm_map_enter(target_map, + &map_addr, + map_size, + mask, + flags & (VM_FLAGS_ANYWHERE | + VM_FLAGS_OVERWRITE | + VM_FLAGS_RETURN_DATA_ADDR), + VM_OBJECT_NULL, + 0, + FALSE, /* copy */ + cur_protection, + max_protection, + inheritance); + if (kr != KERN_SUCCESS) { + named_entry_unlock(named_entry); + return kr; + } + + copy_addr = map_addr; + + for (copy_entry = vm_map_copy_first_entry(copy_map); + copy_entry != vm_map_copy_to_entry(copy_map); + copy_entry = copy_entry->vme_next) { + int remap_flags = 0; + vm_map_t copy_submap; + vm_object_t copy_object; + vm_map_size_t copy_size; + vm_object_offset_t copy_offset; + + copy_offset = copy_entry->offset; + copy_size = (copy_entry->vme_end - + copy_entry->vme_start); + + /* sanity check */ + if (copy_addr + copy_size > + map_addr + map_size) { + /* over-mapping too much !? */ + kr = KERN_INVALID_ARGUMENT; + /* abort */ + break; + } + + /* take a reference on the object */ + if (copy_entry->is_sub_map) { + remap_flags |= VM_FLAGS_SUBMAP; + copy_submap = + copy_entry->object.sub_map; + vm_map_lock(copy_submap); + vm_map_reference(copy_submap); + vm_map_unlock(copy_submap); + copy_object = (vm_object_t) copy_submap; + } else { + copy_object = + copy_entry->object.vm_object; + vm_object_reference(copy_object); + } + + /* over-map the object into destination */ + remap_flags |= flags; + remap_flags |= VM_FLAGS_FIXED; + remap_flags |= VM_FLAGS_OVERWRITE; + remap_flags &= ~VM_FLAGS_ANYWHERE; + kr = vm_map_enter(target_map, + ©_addr, + copy_size, + (vm_map_offset_t) 0, + remap_flags, + copy_object, + copy_offset, + copy, + cur_protection, + max_protection, + inheritance); + if (kr != KERN_SUCCESS) { + if (copy_entry->is_sub_map) { + vm_map_deallocate(copy_submap); + } else { + vm_object_deallocate(copy_object); + } + /* abort */ + break; + } + + /* next mapping */ + copy_addr += copy_size; + } + + if (kr == KERN_SUCCESS) { + if ((flags & VM_FLAGS_RETURN_DATA_ADDR) != 0) { + *address = map_addr + offset_in_mapping; + } else { + *address = map_addr; + } + } + named_entry_unlock(named_entry); + + if (kr != KERN_SUCCESS) { + if (! (flags & VM_FLAGS_OVERWRITE)) { + /* deallocate the contiguous range */ + (void) vm_deallocate(target_map, + map_addr, + map_size); + } + } + + return kr; + + } else { + /* This is the case where we are going to map */ + /* an already mapped object. If the object is */ + /* not ready it is internal. An external */ + /* object cannot be mapped until it is ready */ + /* we can therefore avoid the ready check */ + /* in this case. */ + if ((flags & VM_FLAGS_RETURN_DATA_ADDR) != 0) { + offset_in_mapping = offset - vm_object_trunc_page(offset); + offset = vm_object_trunc_page(offset); + map_size = vm_object_round_page(offset + offset_in_mapping + initial_size) - offset; + } + + object = named_entry->backing.object; + assert(object != VM_OBJECT_NULL); + named_entry_unlock(named_entry); + vm_object_reference(object); + } + } else if (ip_kotype(port) == IKOT_MEMORY_OBJECT) { + /* + * JMM - This is temporary until we unify named entries + * and raw memory objects. + * + * Detected fake ip_kotype for a memory object. In + * this case, the port isn't really a port at all, but + * instead is just a raw memory object. + */ + if ((flags & VM_FLAGS_RETURN_DATA_ADDR) != 0) { + panic("VM_FLAGS_RETURN_DATA_ADDR not expected for raw memory object."); + } + + object = vm_object_enter((memory_object_t)port, + size, FALSE, FALSE, FALSE); + if (object == VM_OBJECT_NULL) + return KERN_INVALID_OBJECT; + + /* wait for object (if any) to be ready */ + if (object != VM_OBJECT_NULL) { + if (object == kernel_object) { + printf("Warning: Attempt to map kernel object" + " by a non-private kernel entity\n"); + return KERN_INVALID_OBJECT; + } + if (!object->pager_ready) { + vm_object_lock(object); + + while (!object->pager_ready) { + vm_object_wait(object, + VM_OBJECT_EVENT_PAGER_READY, + THREAD_UNINT); + vm_object_lock(object); + } + vm_object_unlock(object); + } + } + } else { + return KERN_INVALID_OBJECT; + } + + if (object != VM_OBJECT_NULL && + object->named && + object->pager != MEMORY_OBJECT_NULL && + object->copy_strategy != MEMORY_OBJECT_COPY_NONE) { + memory_object_t pager; + vm_prot_t pager_prot; + kern_return_t kr; + + /* + * For "named" VM objects, let the pager know that the + * memory object is being mapped. Some pagers need to keep + * track of this, to know when they can reclaim the memory + * object, for example. + * VM calls memory_object_map() for each mapping (specifying + * the protection of each mapping) and calls + * memory_object_last_unmap() when all the mappings are gone. + */ + pager_prot = max_protection; + if (copy) { + /* + * Copy-On-Write mapping: won't modify the + * memory object. + */ + pager_prot &= ~VM_PROT_WRITE; + } + vm_object_lock(object); + pager = object->pager; + if (object->named && + pager != MEMORY_OBJECT_NULL && + object->copy_strategy != MEMORY_OBJECT_COPY_NONE) { + assert(object->pager_ready); + vm_object_mapping_wait(object, THREAD_UNINT); + vm_object_mapping_begin(object); + vm_object_unlock(object); + + kr = memory_object_map(pager, pager_prot); + assert(kr == KERN_SUCCESS); + + vm_object_lock(object); + vm_object_mapping_end(object); + } + vm_object_unlock(object); + } + + /* + * Perform the copy if requested + */ + + if (copy) { + vm_object_t new_object; + vm_object_offset_t new_offset; + + result = vm_object_copy_strategically(object, offset, size, + &new_object, &new_offset, + ©); + + + if (result == KERN_MEMORY_RESTART_COPY) { + boolean_t success; + boolean_t src_needs_copy; + + /* + * XXX + * We currently ignore src_needs_copy. + * This really is the issue of how to make + * MEMORY_OBJECT_COPY_SYMMETRIC safe for + * non-kernel users to use. Solution forthcoming. + * In the meantime, since we don't allow non-kernel + * memory managers to specify symmetric copy, + * we won't run into problems here. + */ + new_object = object; + new_offset = offset; + success = vm_object_copy_quickly(&new_object, + new_offset, size, + &src_needs_copy, + ©); + assert(success); + result = KERN_SUCCESS; + } + /* + * Throw away the reference to the + * original object, as it won't be mapped. + */ + + vm_object_deallocate(object); + + 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 (result != KERN_SUCCESS) + vm_object_deallocate(object); + + if ((flags & VM_FLAGS_RETURN_DATA_ADDR) != 0) { + *address = map_addr + offset_in_mapping; + } else { + *address = map_addr; + } + return result; +} + + + + +kern_return_t +vm_map_enter_mem_object_control( + vm_map_t target_map, + vm_map_offset_t *address, + vm_map_size_t initial_size, + vm_map_offset_t mask, + int flags, + memory_object_control_t control, + vm_object_offset_t offset, + boolean_t copy, + vm_prot_t cur_protection, + vm_prot_t max_protection, + vm_inherit_t inheritance) +{ + vm_map_address_t map_addr; + vm_map_size_t map_size; + vm_object_t object; + vm_object_size_t size; + kern_return_t result; + memory_object_t pager; + vm_prot_t pager_prot; + kern_return_t kr; + + /* + * Check arguments for validity + */ + if ((target_map == VM_MAP_NULL) || + (cur_protection & ~VM_PROT_ALL) || + (max_protection & ~VM_PROT_ALL) || + (inheritance > VM_INHERIT_LAST_VALID) || + initial_size == 0) + return KERN_INVALID_ARGUMENT; + + 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); + + if (object == VM_OBJECT_NULL) + return KERN_INVALID_OBJECT; + + if (object == kernel_object) { + printf("Warning: Attempt to map kernel object" + " by a non-private kernel entity\n"); + return KERN_INVALID_OBJECT; + } + + vm_object_lock(object); + object->ref_count++; + vm_object_res_reference(object); + + /* + * For "named" VM objects, let the pager know that the + * memory object is being mapped. Some pagers need to keep + * track of this, to know when they can reclaim the memory + * object, for example. + * VM calls memory_object_map() for each mapping (specifying + * the protection of each mapping) and calls + * memory_object_last_unmap() when all the mappings are gone. + */ + pager_prot = max_protection; + if (copy) { + pager_prot &= ~VM_PROT_WRITE; + } + pager = object->pager; + if (object->named && + pager != MEMORY_OBJECT_NULL && + object->copy_strategy != MEMORY_OBJECT_COPY_NONE) { + assert(object->pager_ready); + vm_object_mapping_wait(object, THREAD_UNINT); + vm_object_mapping_begin(object); + vm_object_unlock(object); + + kr = memory_object_map(pager, pager_prot); + assert(kr == KERN_SUCCESS); + + vm_object_lock(object); + vm_object_mapping_end(object); + } + vm_object_unlock(object); + + /* + * Perform the copy if requested + */ + + if (copy) { + vm_object_t new_object; + vm_object_offset_t new_offset; + + result = vm_object_copy_strategically(object, offset, size, + &new_object, &new_offset, + ©); + + + if (result == KERN_MEMORY_RESTART_COPY) { + boolean_t success; + boolean_t src_needs_copy; + + /* + * XXX + * We currently ignore src_needs_copy. + * This really is the issue of how to make + * MEMORY_OBJECT_COPY_SYMMETRIC safe for + * non-kernel users to use. Solution forthcoming. + * In the meantime, since we don't allow non-kernel + * memory managers to specify symmetric copy, + * we won't run into problems here. + */ + new_object = object; + new_offset = offset; + success = vm_object_copy_quickly(&new_object, + new_offset, size, + &src_needs_copy, + ©); + assert(success); + result = KERN_SUCCESS; + } + /* + * Throw away the reference to the + * original object, as it won't be mapped. + */ + + vm_object_deallocate(object); + + 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 (result != KERN_SUCCESS) + vm_object_deallocate(object); + *address = map_addr; + + return result; +} + + +#if VM_CPM + +#ifdef MACH_ASSERT +extern pmap_paddr_t avail_start, avail_end; +#endif + +/* + * Allocate memory in the specified map, with the caveat that + * the memory is physically contiguous. This call may fail + * if the system can't find sufficient contiguous memory. + * This call may cause or lead to heart-stopping amounts of + * paging activity. + * + * Memory obtained from this call should be freed in the + * normal way, viz., via vm_deallocate. + */ +kern_return_t +vm_map_enter_cpm( + vm_map_t map, + vm_map_offset_t *addr, + vm_map_size_t size, + int flags) +{ + vm_object_t cpm_obj; + pmap_t pmap; + vm_page_t m, pages; + kern_return_t kr; + vm_map_offset_t va, start, end, offset; +#if MACH_ASSERT + vm_map_offset_t prev_addr = 0; +#endif /* MACH_ASSERT */ + + boolean_t anywhere = ((VM_FLAGS_ANYWHERE & flags) != 0); + + if (size == 0) { + *addr = 0; + return KERN_SUCCESS; + } + if (anywhere) + *addr = vm_map_min(map); + else + *addr = vm_map_trunc_page(*addr, + VM_MAP_PAGE_MASK(map)); + size = vm_map_round_page(size, + VM_MAP_PAGE_MASK(map)); + + /* + * LP64todo - cpm_allocate should probably allow * allocations of >4GB, but not with the current * algorithm, so just cast down the size for now. */ if (size > VM_MAX_ADDRESS) return KERN_RESOURCE_SHORTAGE; if ((kr = cpm_allocate(CAST_DOWN(vm_size_t, size), - &pages, TRUE)) != KERN_SUCCESS) + &pages, 0, 0, TRUE, flags)) != KERN_SUCCESS) return kr; cpm_obj = vm_object_allocate((vm_object_size_t)size); assert(cpm_obj != VM_OBJECT_NULL); assert(cpm_obj->internal); - assert(cpm_obj->size == (vm_object_size_t)size); + assert(cpm_obj->vo_size == (vm_object_size_t)size); assert(cpm_obj->can_persist == FALSE); assert(cpm_obj->pager_created == FALSE); assert(cpm_obj->pageout == FALSE); @@ -1815,11 +3118,13 @@ vm_map_enter_cpm( for (offset = 0; offset < size; offset += PAGE_SIZE) { m = pages; pages = NEXT_PAGE(m); + *(NEXT_PAGE_PTR(m)) = VM_PAGE_NULL; assert(!m->gobbled); assert(!m->wanted); assert(!m->pageout); assert(!m->tabled); + assert(VM_PAGE_WIRED(m)); /* * ENCRYPTED SWAP: * "m" is not supposed to be pageable, so it @@ -1828,7 +3133,7 @@ vm_map_enter_cpm( */ ASSERT_PAGE_DECRYPTED(m); assert(m->busy); - assert(m->phys_page>=avail_start && m->phys_page<=avail_end); + assert(m->phys_page>=(avail_start>>PAGE_SHIFT) && m->phys_page<=(avail_end>>PAGE_SHIFT)); m->busy = FALSE; vm_page_insert(m, cpm_obj, offset); @@ -1854,17 +3159,17 @@ vm_map_enter_cpm( */ kr = vm_map_enter( - map, - addr, - size, - (vm_map_offset_t)0, - flags, - cpm_obj, - (vm_object_offset_t)0, - FALSE, - VM_PROT_ALL, - VM_PROT_ALL, - VM_INHERIT_DEFAULT); + map, + addr, + size, + (vm_map_offset_t)0, + flags, + cpm_obj, + (vm_object_offset_t)0, + FALSE, + VM_PROT_ALL, + VM_PROT_ALL, + VM_INHERIT_DEFAULT); if (kr != KERN_SUCCESS) { /* @@ -1898,13 +3203,21 @@ vm_map_enter_cpm( */ for (offset = 0, va = start; offset < size; va += PAGE_SIZE, offset += PAGE_SIZE) { + int type_of_fault; + vm_object_lock(cpm_obj); m = vm_page_lookup(cpm_obj, (vm_object_offset_t)offset); - vm_object_unlock(cpm_obj); assert(m != VM_PAGE_NULL); - PMAP_ENTER(pmap, va, m, VM_PROT_ALL, - ((unsigned int)(m->object->wimg_bits)) & VM_WIMG_MASK, - TRUE); + + vm_page_zero_fill(m); + + 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_object_unlock(cpm_obj); } #if MACH_ASSERT @@ -1916,8 +3229,8 @@ vm_map_enter_cpm( m = vm_page_lookup(cpm_obj, (vm_object_offset_t)offset); vm_object_unlock(cpm_obj); if (m == VM_PAGE_NULL) - panic("vm_allocate_cpm: obj 0x%x off 0x%x no page", - cpm_obj, offset); + 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); @@ -1926,15 +3239,15 @@ vm_map_enter_cpm( assert(!m->absent); assert(!m->error); assert(!m->cleaning); + assert(!m->laundry); assert(!m->precious); assert(!m->clustered); if (offset != 0) { if (m->phys_page != prev_addr + 1) { - printf("start 0x%x end 0x%x va 0x%x\n", - start, end, va); - printf("obj 0x%x off 0x%x\n", cpm_obj, offset); - printf("m 0x%x prev_address 0x%x\n", m, - prev_addr); + printf("start 0x%llx end 0x%llx va 0x%llx\n", + (uint64_t)start, (uint64_t)end, (uint64_t)va); + printf("obj %p off 0x%llx\n", cpm_obj, (uint64_t)offset); + printf("m %p prev_address 0x%llx\n", m, (uint64_t)prev_addr); panic("vm_allocate_cpm: pages not contig!"); } } @@ -1967,60 +3280,125 @@ vm_map_enter_cpm( } #endif /* VM_CPM */ +/* Not used without nested pmaps */ +#ifndef NO_NESTED_PMAP /* - * vm_map_clip_start: [ internal use only ] - * - * Asserts that the given entry begins at or after - * the specified address; if necessary, - * it splits the entry into two. + * Clip and unnest a portion of a nested submap mapping. */ -#ifndef i386 -#define vm_map_clip_start(map, entry, startaddr) \ -MACRO_BEGIN \ - vm_map_t VMCS_map; \ - vm_map_entry_t VMCS_entry; \ - vm_map_offset_t VMCS_startaddr; \ - VMCS_map = (map); \ - VMCS_entry = (entry); \ - VMCS_startaddr = (startaddr); \ - if (VMCS_startaddr > VMCS_entry->vme_start) { \ - if(entry->use_pmap) { \ - vm_map_offset_t pmap_base_addr; \ - \ - pmap_base_addr = 0xF0000000 & entry->vme_start; \ - pmap_unnest(map->pmap, (addr64_t)pmap_base_addr); \ - entry->use_pmap = FALSE; \ - } else if(entry->object.vm_object \ - && !entry->is_sub_map \ - && entry->object.vm_object->phys_contiguous) { \ - pmap_remove(map->pmap, \ - (addr64_t)(entry->vme_start), \ - (addr64_t)(entry->vme_end)); \ - } \ - _vm_map_clip_start(&VMCS_map->hdr,VMCS_entry,VMCS_startaddr);\ - } \ - UPDATE_FIRST_FREE(VMCS_map, VMCS_map->first_free); \ -MACRO_END -#else -#define vm_map_clip_start(map, entry, startaddr) \ -MACRO_BEGIN \ - vm_map_t VMCS_map; \ - vm_map_entry_t VMCS_entry; \ - vm_map_offset_t VMCS_startaddr; \ - VMCS_map = (map); \ - VMCS_entry = (entry); \ - VMCS_startaddr = (startaddr); \ - if (VMCS_startaddr > VMCS_entry->vme_start) { \ - _vm_map_clip_start(&VMCS_map->hdr,VMCS_entry,VMCS_startaddr);\ - } \ - UPDATE_FIRST_FREE(VMCS_map, VMCS_map->first_free); \ -MACRO_END -#endif -#define vm_map_copy_clip_start(copy, entry, startaddr) \ - MACRO_BEGIN \ - if ((startaddr) > (entry)->vme_start) \ - _vm_map_clip_start(&(copy)->cpy_hdr,(entry),(startaddr)); \ + +static void +vm_map_clip_unnest( + vm_map_t map, + vm_map_entry_t entry, + vm_map_offset_t start_unnest, + vm_map_offset_t end_unnest) +{ + vm_map_offset_t old_start_unnest = start_unnest; + vm_map_offset_t old_end_unnest = end_unnest; + + assert(entry->is_sub_map); + assert(entry->object.sub_map != NULL); + + /* + * Query the platform for the optimal unnest range. + * DRK: There's some duplication of effort here, since + * callers may have adjusted the range to some extent. This + * routine was introduced to support 1GiB subtree nesting + * for x86 platforms, which can also nest on 2MiB boundaries + * depending on size/alignment. + */ + if (pmap_adjust_unnest_parameters(map->pmap, &start_unnest, &end_unnest)) { + log_unnest_badness(map, old_start_unnest, old_end_unnest); + } + + if (entry->vme_start > start_unnest || + entry->vme_end < end_unnest) { + panic("vm_map_clip_unnest(0x%llx,0x%llx): " + "bad nested entry: start=0x%llx end=0x%llx\n", + (long long)start_unnest, (long long)end_unnest, + (long long)entry->vme_start, (long long)entry->vme_end); + } + + if (start_unnest > entry->vme_start) { + _vm_map_clip_start(&map->hdr, + entry, + start_unnest); + vm_map_store_update_first_free(map, map->first_free); + } + 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); + } + + pmap_unnest(map->pmap, + entry->vme_start, + entry->vme_end - entry->vme_start); + if ((map->mapped_in_other_pmaps) && (map->ref_count)) { + /* clean up parent map/maps */ + vm_map_submap_pmap_clean( + map, entry->vme_start, + entry->vme_end, + entry->object.sub_map, + entry->offset); + } + entry->use_pmap = FALSE; + if (entry->alias == VM_MEMORY_SHARED_PMAP) { + entry->alias = VM_MEMORY_UNSHARED_PMAP; + } +} +#endif /* NO_NESTED_PMAP */ + +/* + * vm_map_clip_start: [ internal use only ] + * + * Asserts that the given entry begins at or after + * the specified address; if necessary, + * it splits the entry into two. + */ +void +vm_map_clip_start( + vm_map_t map, + vm_map_entry_t entry, + vm_map_offset_t startaddr) +{ +#ifndef NO_NESTED_PMAP + if (entry->use_pmap && + startaddr >= entry->vme_start) { + vm_map_offset_t start_unnest, end_unnest; + + /* + * Make sure "startaddr" is no longer in a nested range + * before we clip. Unnest only the minimum range the platform + * can handle. + * vm_map_clip_unnest may perform additional adjustments to + * the unnest range. + */ + start_unnest = startaddr & ~(pmap_nesting_size_min - 1); + end_unnest = start_unnest + pmap_nesting_size_min; + vm_map_clip_unnest(map, entry, start_unnest, end_unnest); + } +#endif /* NO_NESTED_PMAP */ + if (startaddr > entry->vme_start) { + if (entry->object.vm_object && + !entry->is_sub_map && + entry->object.vm_object->phys_contiguous) { + pmap_remove(map->pmap, + (addr64_t)(entry->vme_start), + (addr64_t)(entry->vme_end)); + } + _vm_map_clip_start(&map->hdr, entry, startaddr); + vm_map_store_update_first_free(map, map->first_free); + } +} + + +#define vm_map_copy_clip_start(copy, entry, startaddr) \ + MACRO_BEGIN \ + if ((startaddr) > (entry)->vme_start) \ + _vm_map_clip_start(&(copy)->cpy_hdr,(entry),(startaddr)); \ MACRO_END /* @@ -2043,17 +3421,23 @@ _vm_map_clip_start( * address. */ - new_entry = _vm_map_entry_create(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); + assert(start < entry->vme_end); + assert(VM_MAP_PAGE_ALIGNED(start, + VM_MAP_HDR_PAGE_MASK(map_header))); entry->vme_start = start; - _vm_map_entry_link(map_header, entry->vme_prev, new_entry); + _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(new_entry->object.sub_map); else vm_object_reference(new_entry->object.vm_object); } @@ -2066,48 +3450,50 @@ _vm_map_clip_start( * the specified address; if necessary, * it splits the entry into two. */ -#ifndef i386 -#define vm_map_clip_end(map, entry, endaddr) \ -MACRO_BEGIN \ - vm_map_t VMCE_map; \ - vm_map_entry_t VMCE_entry; \ - vm_map_offset_t VMCE_endaddr; \ - VMCE_map = (map); \ - VMCE_entry = (entry); \ - VMCE_endaddr = (endaddr); \ - if (VMCE_endaddr < VMCE_entry->vme_end) { \ - if(entry->use_pmap) { \ - vm_map_offset_t pmap_base_addr; \ - \ - pmap_base_addr = 0xF0000000 & entry->vme_start; \ - pmap_unnest(map->pmap, (addr64_t)pmap_base_addr); \ - entry->use_pmap = FALSE; \ - } else if(entry->object.vm_object \ - && !entry->is_sub_map \ - && entry->object.vm_object->phys_contiguous) { \ - pmap_remove(map->pmap, \ - (addr64_t)(entry->vme_start), \ - (addr64_t)(entry->vme_end)); \ - } \ - _vm_map_clip_end(&VMCE_map->hdr,VMCE_entry,VMCE_endaddr); \ - } \ - UPDATE_FIRST_FREE(VMCE_map, VMCE_map->first_free); \ -MACRO_END -#else -#define vm_map_clip_end(map, entry, endaddr) \ -MACRO_BEGIN \ - vm_map_t VMCE_map; \ - vm_map_entry_t VMCE_entry; \ - vm_map_offset_t VMCE_endaddr; \ - VMCE_map = (map); \ - VMCE_entry = (entry); \ - VMCE_endaddr = (endaddr); \ - if (VMCE_endaddr < VMCE_entry->vme_end) { \ - _vm_map_clip_end(&VMCE_map->hdr,VMCE_entry,VMCE_endaddr); \ - } \ - UPDATE_FIRST_FREE(VMCE_map, VMCE_map->first_free); \ -MACRO_END -#endif +void +vm_map_clip_end( + vm_map_t map, + vm_map_entry_t entry, + vm_map_offset_t endaddr) +{ + if (endaddr > entry->vme_end) { + /* + * Within the scope of this clipping, limit "endaddr" to + * the end of this map entry... + */ + endaddr = entry->vme_end; + } +#ifndef NO_NESTED_PMAP + if (entry->use_pmap) { + vm_map_offset_t start_unnest, end_unnest; + + /* + * Make sure the range between the start of this entry and + * the new "endaddr" is no longer nested before we clip. + * Unnest only the minimum range the platform can handle. + * vm_map_clip_unnest may perform additional adjustments to + * the unnest range. + */ + start_unnest = entry->vme_start; + end_unnest = + (endaddr + pmap_nesting_size_min - 1) & + ~(pmap_nesting_size_min - 1); + vm_map_clip_unnest(map, entry, start_unnest, end_unnest); + } +#endif /* NO_NESTED_PMAP */ + if (endaddr < entry->vme_end) { + if (entry->object.vm_object && + !entry->is_sub_map && + entry->object.vm_object->phys_contiguous) { + pmap_remove(map->pmap, + (addr64_t)(entry->vme_start), + (addr64_t)(entry->vme_end)); + } + _vm_map_clip_end(&map->hdr, entry, endaddr); + vm_map_store_update_first_free(map, map->first_free); + } +} + #define vm_map_copy_clip_end(copy, entry, endaddr) \ MACRO_BEGIN \ @@ -2123,7 +3509,7 @@ static void _vm_map_clip_end( register struct vm_map_header *map_header, register vm_map_entry_t entry, - register vm_map_offset_t end) + register vm_map_offset_t end) { register vm_map_entry_t new_entry; @@ -2132,16 +3518,20 @@ _vm_map_clip_end( * AFTER the specified entry */ - new_entry = _vm_map_entry_create(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); + assert(new_entry->vme_start < new_entry->vme_end); - _vm_map_entry_link(map_header, entry, new_entry); + _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(new_entry->object.sub_map); else vm_object_reference(new_entry->object.vm_object); } @@ -2153,15 +3543,15 @@ _vm_map_clip_end( * Asserts that the starting and ending region * addresses fall within the valid range of the map. */ -#define VM_MAP_RANGE_CHECK(map, start, end) \ - { \ - if (start < vm_map_min(map)) \ - start = vm_map_min(map); \ - if (end > vm_map_max(map)) \ - end = vm_map_max(map); \ - if (start > end) \ - start = end; \ - } +#define VM_MAP_RANGE_CHECK(map, start, end) \ + MACRO_BEGIN \ + if (start < vm_map_min(map)) \ + start = vm_map_min(map); \ + if (end > vm_map_max(map)) \ + end = vm_map_max(map); \ + if (start > end) \ + start = end; \ + MACRO_END /* * vm_map_range_check: [ internal use only ] @@ -2249,9 +3639,9 @@ vm_map_submap( vm_map_offset_t end, vm_map_t submap, vm_map_offset_t offset, -#ifdef i386 +#ifdef NO_NESTED_PMAP __unused -#endif +#endif /* NO_NESTED_PMAP */ boolean_t use_pmap) { vm_map_entry_t entry; @@ -2260,21 +3650,18 @@ vm_map_submap( vm_map_lock(map); - submap->mapped = TRUE; - - VM_MAP_RANGE_CHECK(map, start, end); - - if (vm_map_lookup_entry(map, start, &entry)) { - vm_map_clip_start(map, entry, start); - } - else + if (! vm_map_lookup_entry(map, start, &entry)) { entry = entry->vme_next; + } - if(entry == vm_map_to_entry(map)) { + if (entry == vm_map_to_entry(map) || + entry->is_sub_map) { vm_map_unlock(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) && @@ -2284,35 +3671,51 @@ vm_map_submap( (object->copy == VM_OBJECT_NULL) && (object->shadow == VM_OBJECT_NULL) && (!object->pager_created)) { - entry->offset = (vm_object_offset_t)offset; - entry->object.vm_object = VM_OBJECT_NULL; - vm_object_deallocate(object); - entry->is_sub_map = TRUE; - entry->object.sub_map = submap; - vm_map_reference(submap); -#ifndef i386 - if ((use_pmap) && (offset == 0)) { - /* nest if platform code will allow */ - if(submap->pmap == NULL) { - submap->pmap = pmap_create((vm_map_size_t) 0); - if(submap->pmap == PMAP_NULL) { - vm_map_unlock(map); - return(KERN_NO_SPACE); - } + entry->offset = (vm_object_offset_t)offset; + entry->object.vm_object = VM_OBJECT_NULL; + vm_object_deallocate(object); + entry->is_sub_map = TRUE; + entry->object.sub_map = submap; + vm_map_reference(submap); + if (submap->mapped_in_other_pmaps == FALSE && + vm_map_pmap(submap) != PMAP_NULL && + vm_map_pmap(submap) != vm_map_pmap(map)) { + /* + * This submap is being mapped in a map + * that uses a different pmap. + * Set its "mapped_in_other_pmaps" flag + * to indicate that we now need to + * remove mappings from all pmaps rather + * than just the submap's pmap. + */ + submap->mapped_in_other_pmaps = TRUE; + } + +#ifndef NO_NESTED_PMAP + if (use_pmap) { + /* nest if platform code will allow */ + if(submap->pmap == NULL) { + ledger_t ledger = map->pmap->ledger; + submap->pmap = pmap_create(ledger, + (vm_map_size_t) 0, FALSE); + if(submap->pmap == PMAP_NULL) { + vm_map_unlock(map); + return(KERN_NO_SPACE); } - result = pmap_nest(map->pmap, (entry->object.sub_map)->pmap, - (addr64_t)start, - (addr64_t)start, - (uint64_t)(end - start)); - if(result) - panic("vm_map_submap: pmap_nest failed, rc = %08X\n", result); - entry->use_pmap = TRUE; } -#endif -#ifdef i386 - pmap_remove(map->pmap, (addr64_t)start, (addr64_t)end); -#endif - result = KERN_SUCCESS; + result = pmap_nest(map->pmap, + (entry->object.sub_map)->pmap, + (addr64_t)start, + (addr64_t)start, + (uint64_t)(end - start)); + if(result) + panic("vm_map_submap: pmap_nest failed, rc = %08X\n", result); + entry->use_pmap = TRUE; + } +#else /* NO_NESTED_PMAP */ + pmap_remove(map->pmap, (addr64_t)start, (addr64_t)end); +#endif /* NO_NESTED_PMAP */ + result = KERN_SUCCESS; } vm_map_unlock(map); @@ -2336,14 +3739,13 @@ vm_map_protect( register boolean_t set_max) { register vm_map_entry_t current; - register vm_map_offset_t prev; + register vm_map_offset_t prev; vm_map_entry_t entry; vm_prot_t new_max; - boolean_t clip; XPR(XPR_VM_MAP, - "vm_map_protect, 0x%X start 0x%X end 0x%X, new 0x%X %d", - (integer_t)map, start, end, new_prot, set_max); + "vm_map_protect, 0x%X start 0x%X end 0x%X, new 0x%X %d", + map, start, end, new_prot, set_max); vm_map_lock(map); @@ -2356,18 +3758,24 @@ vm_map_protect( return(KERN_INVALID_ADDRESS); } - /* - * Lookup the entry. If it doesn't start in a valid - * entry, return an error. Remember if we need to - * clip the entry. We don't do it here because we don't - * want to make any changes until we've scanned the - * entire range below for address and protection - * violations. - */ - if (!(clip = vm_map_lookup_entry(map, start, &entry))) { - vm_map_unlock(map); - return(KERN_INVALID_ADDRESS); - } + while(1) { + /* + * Lookup the entry. If it doesn't start in a valid + * entry, return an error. + */ + if (! vm_map_lookup_entry(map, start, &entry)) { + vm_map_unlock(map); + return(KERN_INVALID_ADDRESS); + } + + if (entry->superpage_size && (start & (SUPERPAGE_SIZE-1))) { /* extend request to whole entry */ + start = SUPERPAGE_ROUND_DOWN(start); + continue; + } + break; + } + if (entry->superpage_size) + end = SUPERPAGE_ROUND_UP(end); /* * Make a first pass to check for protection and address @@ -2401,6 +3809,7 @@ vm_map_protect( } } + prev = current->vme_end; current = current->vme_next; } @@ -2416,9 +3825,11 @@ vm_map_protect( */ current = entry; - if (clip) { - vm_map_clip_start(map, entry, start); + if (current != vm_map_to_entry(map)) { + /* clip and unnest if necessary */ + vm_map_clip_start(map, current, start); } + while ((current != vm_map_to_entry(map)) && (current->vme_start < end)) { @@ -2426,6 +3837,8 @@ vm_map_protect( vm_map_clip_end(map, current, end); + assert(!current->use_pmap); /* clipping did unnest if needed */ + old_prot = current->protection; if(new_prot & VM_PROT_COPY) { @@ -2434,6 +3847,11 @@ vm_map_protect( /* 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; } @@ -2441,8 +3859,8 @@ vm_map_protect( if (set_max) current->protection = (current->max_protection = - new_prot & ~VM_PROT_COPY) & - old_prot; + new_prot & ~VM_PROT_COPY) & + old_prot; else current->protection = new_prot & ~VM_PROT_COPY; @@ -2455,45 +3873,29 @@ vm_map_protect( * write-protect fault occurred, it will be fixed up * properly, COW or not. */ - /* the 256M hack for existing hardware limitations */ if (current->protection != old_prot) { - if(current->is_sub_map && current->use_pmap) { - vm_map_offset_t pmap_base_addr; - vm_map_offset_t pmap_end_addr; -#ifdef i386 - __unused -#endif - vm_map_entry_t local_entry; - - pmap_base_addr = 0xF0000000 & current->vme_start; - pmap_end_addr = (pmap_base_addr + 0x10000000) - 1; -#ifndef i386 - if(!vm_map_lookup_entry(map, - pmap_base_addr, &local_entry)) - panic("vm_map_protect: nested pmap area is missing"); - while ((local_entry != vm_map_to_entry(map)) && - (local_entry->vme_start < pmap_end_addr)) { - local_entry->use_pmap = FALSE; - local_entry = local_entry->vme_next; - } - pmap_unnest(map->pmap, (addr64_t)pmap_base_addr); -#endif - } - if (!(current->protection & VM_PROT_WRITE)) { /* Look one level in we support nested pmaps */ /* from mapped submaps which are direct entries */ /* in our map */ - if(current->is_sub_map && current->use_pmap) { + + vm_prot_t prot; + + prot = current->protection & ~VM_PROT_WRITE; + + if (override_nx(map, current->alias) && prot) + prot |= VM_PROT_EXECUTE; + + if (current->is_sub_map && current->use_pmap) { pmap_protect(current->object.sub_map->pmap, - current->vme_start, - current->vme_end, - current->protection); + current->vme_start, + current->vme_end, + prot); } else { - pmap_protect(map->pmap, current->vme_start, - current->vme_end, - current->protection); + pmap_protect(map->pmap, + current->vme_start, + current->vme_end, + prot); } - } } current = current->vme_next; } @@ -2533,7 +3935,6 @@ vm_map_inherit( if (vm_map_lookup_entry(map, start, &temp_entry)) { entry = temp_entry; - vm_map_clip_start(map, entry, start); } else { temp_entry = temp_entry->vme_next; @@ -2554,9 +3955,14 @@ vm_map_inherit( } entry = temp_entry; + if (entry != vm_map_to_entry(map)) { + /* clip and unnest if necessary */ + vm_map_clip_start(map, entry, start); + } 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 */ entry->inheritance = new_inheritance; @@ -2567,6 +3973,118 @@ vm_map_inherit( return(KERN_SUCCESS); } +/* + * Update the accounting for the amount of wired memory in this map. If the user has + * exceeded the defined limits, then we fail. Wiring on behalf of the kernel never fails. + */ + +static kern_return_t +add_wire_counts( + vm_map_t map, + vm_map_entry_t entry, + boolean_t user_wire) +{ + vm_map_size_t size; + + if (user_wire) { + unsigned int total_wire_count = vm_page_wire_count + vm_lopage_free_count; + + /* + * We're wiring memory at the request of the user. Check if this is the first time the user is wiring + * this map entry. + */ + + if (entry->user_wired_count == 0) { + size = entry->vme_end - entry->vme_start; + + /* + * Since this is the first time the user is wiring this map entry, check to see if we're + * exceeding the user wire limits. There is a per map limit which is the smaller of either + * the process's rlimit or the global vm_user_wire_limit which caps this value. There is also + * a system-wide limit on the amount of memory all users can wire. If the user is over either + * limit, then we fail. + */ + + if(size + map->user_wire_size > MIN(map->user_wire_limit, vm_user_wire_limit) || + size + ptoa_64(total_wire_count) > vm_global_user_wire_limit || + size + ptoa_64(total_wire_count) > max_mem - vm_global_no_user_wire_amount) + return KERN_RESOURCE_SHORTAGE; + + /* + * The first time the user wires an entry, we also increment the wired_count and add this to + * the total that has been wired in the map. + */ + + if (entry->wired_count >= MAX_WIRE_COUNT) + return KERN_FAILURE; + + entry->wired_count++; + map->user_wire_size += size; + } + + if (entry->user_wired_count >= MAX_WIRE_COUNT) + return KERN_FAILURE; + + entry->user_wired_count++; + + } else { + + /* + * The kernel's wiring the memory. Just bump the count and continue. + */ + + if (entry->wired_count >= MAX_WIRE_COUNT) + panic("vm_map_wire: too many wirings"); + + entry->wired_count++; + } + + return KERN_SUCCESS; +} + +/* + * Update the memory wiring accounting now that the given map entry is being unwired. + */ + +static void +subtract_wire_counts( + vm_map_t map, + vm_map_entry_t entry, + boolean_t user_wire) +{ + + if (user_wire) { + + /* + * We're unwiring memory at the request of the user. See if we're removing the last user wire reference. + */ + + if (entry->user_wired_count == 1) { + + /* + * We're removing the last user wire reference. Decrement the wired_count and the total + * user wired memory for this map. + */ + + assert(entry->wired_count >= 1); + entry->wired_count--; + map->user_wire_size -= entry->vme_end - entry->vme_start; + } + + assert(entry->user_wired_count >= 1); + entry->user_wired_count--; + + } else { + + /* + * The kernel is unwiring the memory. Just update the count. + */ + + assert(entry->wired_count >= 1); + entry->wired_count--; + } +} + /* * vm_map_wire: * @@ -2610,25 +4128,46 @@ vm_map_wire_nested( VM_MAP_RANGE_CHECK(map, start, end); assert(page_aligned(start)); assert(page_aligned(end)); + assert(VM_MAP_PAGE_ALIGNED(start, VM_MAP_PAGE_MASK(map))); + assert(VM_MAP_PAGE_ALIGNED(end, VM_MAP_PAGE_MASK(map))); if (start == end) { /* We wired what the caller asked for, zero pages */ vm_map_unlock(map); return KERN_SUCCESS; } - if (vm_map_lookup_entry(map, start, &first_entry)) { + need_wakeup = FALSE; + cur_thread = current_thread(); + + s = start; + rc = KERN_SUCCESS; + + if (vm_map_lookup_entry(map, s, &first_entry)) { entry = first_entry; - /* vm_map_clip_start will be done later. */ + /* + * vm_map_clip_start will be done later. + * We don't want to unnest any nested submaps here ! + */ } else { /* Start address is not in map */ - vm_map_unlock(map); - return(KERN_INVALID_ADDRESS); + rc = KERN_INVALID_ADDRESS; + goto done; } - s=start; - need_wakeup = FALSE; - cur_thread = current_thread(); - while ((entry != vm_map_to_entry(map)) && (entry->vme_start < end)) { + while ((entry != vm_map_to_entry(map)) && (s < end)) { + /* + * At this point, we have wired from "start" to "s". + * We still need to wire from "s" to "end". + * + * "entry" hasn't been clipped, so it could start before "s" + * and/or end after "end". + */ + + /* "e" is how far we want to wire in this entry */ + e = entry->vme_end; + if (e > end) + e = end; + /* * If another thread is wiring/unwiring this entry then * block after informing other thread to wake us up. @@ -2640,8 +4179,10 @@ vm_map_wire_nested( * We have not clipped the entry. Make sure that * the start address is in range so that the lookup * below will succeed. + * "s" is the current starting point: we've already + * wired from "start" to "s" and we still have + * to wire from "s" to "end". */ - s = entry->vme_start < start? start: entry->vme_start; entry->needs_wakeup = TRUE; @@ -2657,8 +4198,8 @@ vm_map_wire_nested( * User wiring is interruptible */ wait_result = vm_map_entry_wait(map, - (user_wire) ? THREAD_ABORTSAFE : - THREAD_UNINT); + (user_wire) ? THREAD_ABORTSAFE : + THREAD_UNINT); if (user_wire && wait_result == THREAD_INTERRUPTED) { /* * undo the wirings we have done so far @@ -2666,9 +4207,8 @@ vm_map_wire_nested( * because we cannot tell if we were the * only one waiting. */ - vm_map_unlock(map); - vm_map_unwire(map, start, s, user_wire); - return(KERN_FAILURE); + rc = KERN_FAILURE; + goto done; } /* @@ -2681,38 +4221,42 @@ vm_map_wire_nested( * Worse that can happen is, it may not exist anymore. */ if (!vm_map_lookup_entry(map, s, &first_entry)) { - if (!user_wire) - panic("vm_map_wire: re-lookup failed"); - /* * User: undo everything upto the previous * entry. let vm_map_unwire worry about * checking the validity of the range. */ - vm_map_unlock(map); - vm_map_unwire(map, start, s, user_wire); - return(KERN_FAILURE); + rc = KERN_FAILURE; + goto done; } entry = first_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_start; vm_map_offset_t local_end; pmap_t pmap; - - vm_map_clip_start(map, entry, start); + + vm_map_clip_start(map, entry, s); vm_map_clip_end(map, entry, end); sub_start = entry->offset; - sub_end = entry->vme_end - entry->vme_start; - sub_end += entry->offset; - + sub_end = entry->vme_end; + sub_end += entry->offset - entry->vme_start; + local_end = entry->vme_end; if(map_pmap == NULL) { + vm_object_t object; + vm_object_offset_t offset; + vm_prot_t prot; + boolean_t wired; + vm_map_entry_t local_entry; + vm_map_version_t version; + vm_map_t lookup_map; + if(entry->use_pmap) { pmap = entry->object.sub_map->pmap; /* ppc implementation requires that */ @@ -2721,127 +4265,114 @@ vm_map_wire_nested( #ifdef notdef pmap_addr = sub_start; #endif - pmap_addr = start; + pmap_addr = s; } else { pmap = map->pmap; - pmap_addr = start; + pmap_addr = s; } + if (entry->wired_count) { - if (entry->wired_count - >= MAX_WIRE_COUNT) - panic("vm_map_wire: too many wirings"); - - if (user_wire && - entry->user_wired_count - >= MAX_WIRE_COUNT) { - vm_map_unlock(map); - vm_map_unwire(map, start, - entry->vme_start, user_wire); - return(KERN_FAILURE); - } - if(user_wire) - entry->user_wired_count++; - if((!user_wire) || - (entry->user_wired_count == 0)) - entry->wired_count++; + if ((rc = add_wire_counts(map, entry, user_wire)) != KERN_SUCCESS) + goto done; + + /* + * The map was not unlocked: + * no need to goto re-lookup. + * Just go directly to next entry. + */ entry = entry->vme_next; + s = entry->vme_start; continue; - } else { - vm_object_t object; - vm_map_offset_t offset_hi; - vm_map_offset_t offset_lo; - vm_object_offset_t offset; - vm_prot_t prot; - boolean_t wired; - vm_behavior_t behavior; - vm_map_entry_t local_entry; - vm_map_version_t version; - vm_map_t lookup_map; - - /* call vm_map_lookup_locked to */ - /* cause any needs copy to be */ - /* evaluated */ - local_start = entry->vme_start; - lookup_map = map; - vm_map_lock_write_to_read(map); - if(vm_map_lookup_locked( - &lookup_map, local_start, - access_type, - &version, &object, - &offset, &prot, &wired, - &behavior, &offset_lo, - &offset_hi, &real_map)) { - - vm_map_unlock_read(lookup_map); - vm_map_unwire(map, start, - entry->vme_start, user_wire); - return(KERN_FAILURE); - } - if(real_map != lookup_map) - vm_map_unlock(real_map); - vm_map_unlock_read(lookup_map); - vm_map_lock(map); - vm_object_unlock(object); + } - if (!vm_map_lookup_entry(map, - local_start, &local_entry)) { - vm_map_unlock(map); - vm_map_unwire(map, start, - entry->vme_start, user_wire); - return(KERN_FAILURE); - } - /* did we have a change of type? */ - if (!local_entry->is_sub_map) { - last_timestamp = map->timestamp; - continue; - } - entry = local_entry; - if (user_wire) - entry->user_wired_count++; - if((!user_wire) || - (entry->user_wired_count == 1)) - entry->wired_count++; + /* call vm_map_lookup_locked to */ + /* cause any needs copy to be */ + /* evaluated */ + local_start = entry->vme_start; + lookup_map = map; + vm_map_lock_write_to_read(map); + if(vm_map_lookup_locked( + &lookup_map, local_start, + access_type, + OBJECT_LOCK_EXCLUSIVE, + &version, &object, + &offset, &prot, &wired, + NULL, + &real_map)) { - entry->in_transition = TRUE; + vm_map_unlock_read(lookup_map); + vm_map_unwire(map, start, + s, user_wire); + return(KERN_FAILURE); + } + vm_object_unlock(object); + if(real_map != lookup_map) + vm_map_unlock(real_map); + vm_map_unlock_read(lookup_map); + vm_map_lock(map); - vm_map_unlock(map); - rc = vm_map_wire_nested( - entry->object.sub_map, - sub_start, sub_end, - access_type, - user_wire, pmap, pmap_addr); - vm_map_lock(map); + /* we unlocked, so must re-lookup */ + if (!vm_map_lookup_entry(map, + local_start, + &local_entry)) { + rc = KERN_FAILURE; + goto done; + } + + /* + * entry could have been "simplified", + * so re-clip + */ + entry = local_entry; + assert(s == local_start); + vm_map_clip_start(map, entry, s); + vm_map_clip_end(map, entry, end); + /* re-compute "e" */ + e = entry->vme_end; + if (e > end) + e = end; + + /* did we have a change of type? */ + if (!entry->is_sub_map) { + last_timestamp = map->timestamp; + continue; } } else { local_start = entry->vme_start; - if (user_wire) - entry->user_wired_count++; - if((!user_wire) || - (entry->user_wired_count == 1)) - entry->wired_count++; - vm_map_unlock(map); - rc = vm_map_wire_nested(entry->object.sub_map, - sub_start, sub_end, - access_type, - user_wire, map_pmap, pmap_addr); - vm_map_lock(map); + pmap = map_pmap; } - s = entry->vme_start; - e = entry->vme_end; - /* - * Find the entry again. It could have been clipped - * after we unlocked the map. - */ - if (!vm_map_lookup_entry(map, local_start, + if ((rc = add_wire_counts(map, entry, user_wire)) != KERN_SUCCESS) + goto done; + + entry->in_transition = TRUE; + + vm_map_unlock(map); + rc = vm_map_wire_nested(entry->object.sub_map, + sub_start, sub_end, + access_type, + user_wire, pmap, pmap_addr); + vm_map_lock(map); + + /* + * Find the entry again. It could have been clipped + * after we unlocked the map. + */ + if (!vm_map_lookup_entry(map, local_start, &first_entry)) panic("vm_map_wire: re-lookup failed"); entry = first_entry; + assert(local_start == s); + /* re-compute "e" */ + e = entry->vme_end; + if (e > end) + e = end; + last_timestamp = map->timestamp; while ((entry != vm_map_to_entry(map)) && - (entry->vme_start < e)) { + (entry->vme_start < e)) { assert(entry->in_transition); entry->in_transition = FALSE; if (entry->needs_wakeup) { @@ -2849,24 +4380,16 @@ vm_map_wire_nested( need_wakeup = TRUE; } if (rc != KERN_SUCCESS) {/* from vm_*_wire */ - if (user_wire) - entry->user_wired_count--; - if ((!user_wire) || - (entry->user_wired_count == 0)) - entry->wired_count--; + subtract_wire_counts(map, entry, user_wire); } entry = entry->vme_next; } if (rc != KERN_SUCCESS) { /* from vm_*_wire */ - vm_map_unlock(map); - if (need_wakeup) - vm_map_entry_wakeup(map); - /* - * undo everything upto the previous entry. - */ - (void)vm_map_unwire(map, start, s, user_wire); - return rc; + goto done; } + + /* no need to relookup again */ + s = entry->vme_start; continue; } @@ -2875,29 +4398,19 @@ vm_map_wire_nested( * the appropriate wire reference count. */ if (entry->wired_count) { - /* sanity check: wired_count is a short */ - if (entry->wired_count >= MAX_WIRE_COUNT) - panic("vm_map_wire: too many wirings"); - - if (user_wire && - entry->user_wired_count >= MAX_WIRE_COUNT) { - vm_map_unlock(map); - vm_map_unwire(map, start, - entry->vme_start, user_wire); - return(KERN_FAILURE); - } /* * entry is already wired down, get our reference * after clipping to our range. */ - vm_map_clip_start(map, entry, start); + vm_map_clip_start(map, entry, s); vm_map_clip_end(map, entry, end); - if (user_wire) - entry->user_wired_count++; - if ((!user_wire) || (entry->user_wired_count == 1)) - entry->wired_count++; + if ((rc = add_wire_counts(map, entry, user_wire)) != KERN_SUCCESS) + goto done; + + /* map was not unlocked: no need to relookup */ entry = entry->vme_next; + s = entry->vme_start; continue; } @@ -2927,11 +4440,13 @@ vm_map_wire_nested( entry->offset = (vm_object_offset_t)0; } - vm_map_clip_start(map, entry, start); + vm_map_clip_start(map, entry, s); vm_map_clip_end(map, entry, end); - s = entry->vme_start; + /* re-compute "e" */ e = entry->vme_end; + if (e > end) + e = end; /* * Check for holes and protection mismatch. @@ -2940,30 +4455,23 @@ vm_map_wire_nested( * Protection: Access requested must be allowed, unless * wiring is by protection class */ - if ((((entry->vme_end < end) && - ((entry->vme_next == vm_map_to_entry(map)) || - (entry->vme_next->vme_start > entry->vme_end))) || - ((entry->protection & access_type) != access_type))) { - /* - * Found a hole or protection problem. - * Unwire the region we wired so far. - */ - if (start != entry->vme_start) { - vm_map_unlock(map); - vm_map_unwire(map, start, s, user_wire); - } else { - vm_map_unlock(map); - } - return((entry->protection&access_type) != access_type? - KERN_PROTECTION_FAILURE: KERN_INVALID_ADDRESS); + if ((entry->vme_end < end) && + ((entry->vme_next == vm_map_to_entry(map)) || + (entry->vme_next->vme_start > entry->vme_end))) { + /* found a hole */ + rc = KERN_INVALID_ADDRESS; + goto done; + } + if ((entry->protection & access_type) != access_type) { + /* found a protection problem */ + rc = KERN_PROTECTION_FAILURE; + goto done; } assert(entry->wired_count == 0 && entry->user_wired_count == 0); - if (user_wire) - entry->user_wired_count++; - if ((!user_wire) || (entry->user_wired_count == 1)) - entry->wired_count++; + if ((rc = add_wire_counts(map, entry, user_wire)) != KERN_SUCCESS) + goto done; entry->in_transition = TRUE; @@ -2992,11 +4500,11 @@ vm_map_wire_nested( if(map_pmap) rc = vm_fault_wire(map, - &tmp_entry, map_pmap, pmap_addr); + &tmp_entry, map_pmap, pmap_addr); else rc = vm_fault_wire(map, - &tmp_entry, map->pmap, - tmp_entry.vme_start); + &tmp_entry, map->pmap, + tmp_entry.vme_start); if (!user_wire && cur_thread != THREAD_NULL) thread_interrupt_level(interruptible_state); @@ -3009,7 +4517,7 @@ vm_map_wire_nested( * after we unlocked the map. */ if (!vm_map_lookup_entry(map, tmp_entry.vme_start, - &first_entry)) + &first_entry)) panic("vm_map_wire: re-lookup failed"); entry = first_entry; @@ -3026,26 +4534,24 @@ vm_map_wire_nested( need_wakeup = TRUE; } if (rc != KERN_SUCCESS) { /* from vm_*_wire */ - if (user_wire) - entry->user_wired_count--; - if ((!user_wire) || - (entry->user_wired_count == 0)) - entry->wired_count--; + subtract_wire_counts(map, entry, user_wire); } entry = entry->vme_next; } if (rc != KERN_SUCCESS) { /* from vm_*_wire */ - vm_map_unlock(map); - if (need_wakeup) - vm_map_entry_wakeup(map); - /* - * undo everything upto the previous entry. - */ - (void)vm_map_unwire(map, start, s, user_wire); - return rc; + goto done; } + + s = entry->vme_start; } /* end while loop through map entries */ + +done: + if (rc == KERN_SUCCESS) { + /* repair any damage we may have made to the VM map */ + vm_map_simplify_range(map, start, end); + } + vm_map_unlock(map); /* @@ -3054,7 +4560,12 @@ vm_map_wire_nested( if (need_wakeup) vm_map_entry_wakeup(map); - return(KERN_SUCCESS); + if (rc != KERN_SUCCESS) { + /* undo what has been wired so far */ + vm_map_unwire(map, start, s, user_wire); + } + + return rc; } @@ -3069,24 +4580,8 @@ vm_map_wire( kern_return_t kret; -#ifdef ppc - /* - * the calls to mapping_prealloc and mapping_relpre - * (along with the VM_MAP_RANGE_CHECK to insure a - * resonable range was passed in) are - * currently necessary because - * we haven't enabled kernel pre-emption - * and/or the pmap_enter cannot purge and re-use - * existing mappings - */ - VM_MAP_RANGE_CHECK(map, start, end); - mapping_prealloc(end - start); -#endif kret = vm_map_wire_nested(map, start, end, access_type, - user_wire, (pmap_t)NULL, 0); -#ifdef ppc - mapping_relpre(); -#endif + user_wire, (pmap_t)NULL, 0); return kret; } @@ -3126,17 +4621,37 @@ vm_map_unwire_nested( VM_MAP_RANGE_CHECK(map, start, end); assert(page_aligned(start)); assert(page_aligned(end)); + assert(VM_MAP_PAGE_ALIGNED(start, VM_MAP_PAGE_MASK(map))); + assert(VM_MAP_PAGE_ALIGNED(end, VM_MAP_PAGE_MASK(map))); + + if (start == end) { + /* We unwired what the caller asked for: zero pages */ + vm_map_unlock(map); + return KERN_SUCCESS; + } if (vm_map_lookup_entry(map, start, &first_entry)) { entry = first_entry; - /* vm_map_clip_start will be done later. */ + /* + * vm_map_clip_start will be done later. + * We don't want to unnest any nested sub maps here ! + */ } else { + if (!user_wire) { + panic("vm_map_unwire: start not found"); + } /* Start address is not in map. */ vm_map_unlock(map); return(KERN_INVALID_ADDRESS); } + if (entry->superpage_size) { + /* superpages are always wired */ + vm_map_unlock(map); + return KERN_INVALID_ADDRESS; + } + need_wakeup = FALSE; while ((entry != vm_map_to_entry(map)) && (entry->vme_start < end)) { if (entry->in_transition) { @@ -3153,20 +4668,41 @@ vm_map_unwire_nested( * have a reference to it, because if we did, this * entry will not be getting unwired now. */ - if (!user_wire) + if (!user_wire) { + /* + * XXX FBDP + * This could happen: there could be some + * overlapping vslock/vsunlock operations + * going on. + * We should probably just wait and retry, + * but then we have to be careful that this + * entry could get "simplified" after + * "in_transition" gets unset and before + * we re-lookup the entry, so we would + * have to re-clip the entry to avoid + * re-unwiring what we have already unwired... + * See vm_map_wire_nested(). + * + * Or we could just ignore "in_transition" + * here and proceed to decement the wired + * count(s) on this entry. That should be fine + * as long as "wired_count" doesn't drop all + * the way to 0 (and we should panic if THAT + * happens). + */ panic("vm_map_unwire: in_transition entry"); + } entry = entry->vme_next; 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; pmap_t pmap; - - + vm_map_clip_start(map, entry, start); vm_map_clip_end(map, entry, end); @@ -3175,127 +4711,126 @@ vm_map_unwire_nested( sub_end += entry->offset; local_end = entry->vme_end; if(map_pmap == NULL) { - if(entry->use_pmap) { + if(entry->use_pmap) { pmap = entry->object.sub_map->pmap; pmap_addr = sub_start; - } else { + } else { pmap = map->pmap; pmap_addr = start; - } - if (entry->wired_count == 0 || - (user_wire && entry->user_wired_count == 0)) { - if (!user_wire) - panic("vm_map_unwire: entry is unwired"); - entry = entry->vme_next; - continue; - } - - /* - * Check for holes - * Holes: Next entry should be contiguous unless - * this is the end of the region. - */ - if (((entry->vme_end < end) && - ((entry->vme_next == vm_map_to_entry(map)) || - (entry->vme_next->vme_start - > entry->vme_end)))) { - if (!user_wire) - panic("vm_map_unwire: non-contiguous region"); + } + if (entry->wired_count == 0 || + (user_wire && entry->user_wired_count == 0)) { + if (!user_wire) + panic("vm_map_unwire: entry is unwired"); + entry = entry->vme_next; + continue; + } + + /* + * Check for holes + * Holes: Next entry should be contiguous unless + * this is the end of the region. + */ + if (((entry->vme_end < end) && + ((entry->vme_next == vm_map_to_entry(map)) || + (entry->vme_next->vme_start + > entry->vme_end)))) { + if (!user_wire) + panic("vm_map_unwire: non-contiguous region"); /* - entry = entry->vme_next; - continue; + entry = entry->vme_next; + continue; */ - } + } - if (!user_wire || (--entry->user_wired_count == 0)) - entry->wired_count--; + subtract_wire_counts(map, entry, user_wire); - if (entry->wired_count != 0) { - entry = entry->vme_next; - continue; - } + if (entry->wired_count != 0) { + entry = entry->vme_next; + continue; + } + + entry->in_transition = TRUE; + tmp_entry = *entry;/* see comment in vm_map_wire() */ - entry->in_transition = TRUE; - tmp_entry = *entry;/* see comment in vm_map_wire() */ + /* + * We can unlock the map now. The in_transition state + * guarantees existance of the entry. + */ + vm_map_unlock(map); + vm_map_unwire_nested(entry->object.sub_map, + sub_start, sub_end, user_wire, pmap, pmap_addr); + vm_map_lock(map); - /* - * We can unlock the map now. The in_transition state - * guarantees existance of the entry. - */ - vm_map_unlock(map); - vm_map_unwire_nested(entry->object.sub_map, - sub_start, sub_end, user_wire, pmap, pmap_addr); - vm_map_lock(map); + if (last_timestamp+1 != map->timestamp) { + /* + * Find the entry again. It could have been + * clipped or deleted after we unlocked the map. + */ + if (!vm_map_lookup_entry(map, + tmp_entry.vme_start, + &first_entry)) { + if (!user_wire) + panic("vm_map_unwire: re-lookup failed"); + entry = first_entry->vme_next; + } else + entry = first_entry; + } + last_timestamp = map->timestamp; - if (last_timestamp+1 != map->timestamp) { /* - * Find the entry again. It could have been - * clipped or deleted after we unlocked the map. - */ - if (!vm_map_lookup_entry(map, - tmp_entry.vme_start, - &first_entry)) { - if (!user_wire) - panic("vm_map_unwire: re-lookup failed"); - entry = first_entry->vme_next; - } else - entry = first_entry; - } - last_timestamp = map->timestamp; - - /* - * clear transition bit for all constituent entries - * that were in the original entry (saved in - * tmp_entry). Also check for waiters. - */ - while ((entry != vm_map_to_entry(map)) && - (entry->vme_start < tmp_entry.vme_end)) { - assert(entry->in_transition); - entry->in_transition = FALSE; - if (entry->needs_wakeup) { - entry->needs_wakeup = FALSE; - need_wakeup = TRUE; + * clear transition bit for all constituent entries + * that were in the original entry (saved in + * tmp_entry). Also check for waiters. + */ + while ((entry != vm_map_to_entry(map)) && + (entry->vme_start < tmp_entry.vme_end)) { + assert(entry->in_transition); + entry->in_transition = FALSE; + if (entry->needs_wakeup) { + entry->needs_wakeup = FALSE; + need_wakeup = TRUE; + } + entry = entry->vme_next; } - entry = entry->vme_next; - } - continue; + continue; } else { - vm_map_unlock(map); - vm_map_unwire_nested(entry->object.sub_map, - sub_start, sub_end, user_wire, map_pmap, - pmap_addr); - vm_map_lock(map); + vm_map_unlock(map); + vm_map_unwire_nested(entry->object.sub_map, + sub_start, sub_end, user_wire, map_pmap, + pmap_addr); + vm_map_lock(map); - if (last_timestamp+1 != map->timestamp) { - /* - * Find the entry again. It could have been - * clipped or deleted after we unlocked the map. - */ - if (!vm_map_lookup_entry(map, - tmp_entry.vme_start, - &first_entry)) { - if (!user_wire) - panic("vm_map_unwire: re-lookup failed"); - entry = first_entry->vme_next; - } else - entry = first_entry; - } - last_timestamp = map->timestamp; + if (last_timestamp+1 != map->timestamp) { + /* + * Find the entry again. It could have been + * clipped or deleted after we unlocked the map. + */ + if (!vm_map_lookup_entry(map, + tmp_entry.vme_start, + &first_entry)) { + if (!user_wire) + panic("vm_map_unwire: re-lookup failed"); + entry = first_entry->vme_next; + } else + entry = first_entry; + } + last_timestamp = map->timestamp; } } if ((entry->wired_count == 0) || - (user_wire && entry->user_wired_count == 0)) { + (user_wire && entry->user_wired_count == 0)) { if (!user_wire) panic("vm_map_unwire: entry is unwired"); entry = entry->vme_next; continue; } - + assert(entry->wired_count > 0 && - (!user_wire || entry->user_wired_count > 0)); + (!user_wire || entry->user_wired_count > 0)); vm_map_clip_start(map, entry, start); vm_map_clip_end(map, entry, end); @@ -3306,8 +4841,8 @@ vm_map_unwire_nested( * this is the end of the region. */ if (((entry->vme_end < end) && - ((entry->vme_next == vm_map_to_entry(map)) || - (entry->vme_next->vme_start > entry->vme_end)))) { + ((entry->vme_next == vm_map_to_entry(map)) || + (entry->vme_next->vme_start > entry->vme_end)))) { if (!user_wire) panic("vm_map_unwire: non-contiguous region"); @@ -3315,14 +4850,17 @@ vm_map_unwire_nested( continue; } - if (!user_wire || (--entry->user_wired_count == 0)) - entry->wired_count--; + subtract_wire_counts(map, entry, user_wire); if (entry->wired_count != 0) { entry = entry->vme_next; continue; } + if(entry->zero_wired_pages) { + entry->zero_wired_pages = FALSE; + } + entry->in_transition = TRUE; tmp_entry = *entry; /* see comment in vm_map_wire() */ @@ -3333,11 +4871,11 @@ vm_map_unwire_nested( vm_map_unlock(map); if(map_pmap) { vm_fault_unwire(map, - &tmp_entry, FALSE, map_pmap, pmap_addr); + &tmp_entry, FALSE, map_pmap, pmap_addr); } else { vm_fault_unwire(map, - &tmp_entry, FALSE, map->pmap, - tmp_entry.vme_start); + &tmp_entry, FALSE, map->pmap, + tmp_entry.vme_start); } vm_map_lock(map); @@ -3347,9 +4885,9 @@ vm_map_unwire_nested( * or deleted after we unlocked the map. */ if (!vm_map_lookup_entry(map, tmp_entry.vme_start, - &first_entry)) { + &first_entry)) { if (!user_wire) - panic("vm_map_unwire: re-lookup failed"); + panic("vm_map_unwire: re-lookup failed"); entry = first_entry->vme_next; } else entry = first_entry; @@ -3401,7 +4939,7 @@ vm_map_unwire( boolean_t user_wire) { return vm_map_unwire_nested(map, start, end, - user_wire, (pmap_t)NULL, 0); + user_wire, (pmap_t)NULL, 0); } @@ -3423,8 +4961,13 @@ vm_map_entry_delete( e = entry->vme_end; assert(page_aligned(s)); assert(page_aligned(e)); + if (entry->map_aligned == TRUE) { + assert(VM_MAP_PAGE_ALIGNED(s, VM_MAP_PAGE_MASK(map))); + assert(VM_MAP_PAGE_ALIGNED(e, VM_MAP_PAGE_MASK(map))); + } assert(entry->wired_count == 0); assert(entry->user_wired_count == 0); + assert(!entry->permanent); if (entry->is_sub_map) { object = NULL; @@ -3434,7 +4977,7 @@ vm_map_entry_delete( object = entry->object.vm_object; } - vm_map_entry_unlink(map, entry); + vm_map_store_entry_unlink(map, entry); map->size -= e - s; vm_map_entry_dispose(map, entry); @@ -3447,7 +4990,7 @@ vm_map_entry_delete( if (submap) vm_map_deallocate(submap); else - vm_object_deallocate(object); + vm_object_deallocate(object); } @@ -3466,12 +5009,14 @@ vm_map_submap_pmap_clean( submap_end = offset + (end - start); submap_start = offset; + + 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 -= @@ -3486,27 +5031,27 @@ vm_map_submap_pmap_clean( entry->offset); } else { - if((map->mapped) && (map->ref_count) - && (entry->object.vm_object != NULL)) { + if((map->mapped_in_other_pmaps) && (map->ref_count) + && (entry->object.vm_object != NULL)) { vm_object_pmap_protect( entry->object.vm_object, - entry->offset, + entry->offset+(offset-entry->vme_start), remove_size, PMAP_NULL, entry->vme_start, VM_PROT_NONE); } else { pmap_remove(map->pmap, - (addr64_t)start, - (addr64_t)(start + remove_size)); + (addr64_t)start, + (addr64_t)(start + remove_size)); } } } entry = entry->vme_next; - + while((entry != vm_map_to_entry(sub_map)) - && (entry->vme_start < submap_end)) { + && (entry->vme_start < submap_end)) { remove_size = (entry->vme_end - entry->vme_start); if(submap_end < entry->vme_end) { remove_size -= entry->vme_end - submap_end; @@ -3519,8 +5064,8 @@ vm_map_submap_pmap_clean( entry->object.sub_map, entry->offset); } else { - if((map->mapped) && (map->ref_count) - && (entry->object.vm_object != NULL)) { + if((map->mapped_in_other_pmaps) && (map->ref_count) + && (entry->object.vm_object != NULL)) { vm_object_pmap_protect( entry->object.vm_object, entry->offset, @@ -3530,14 +5075,15 @@ vm_map_submap_pmap_clean( VM_PROT_NONE); } else { pmap_remove(map->pmap, - (addr64_t)((start + entry->vme_start) - - offset), - (addr64_t)(((start + entry->vme_start) - - offset) + remove_size)); + (addr64_t)((start + entry->vme_start) + - offset), + (addr64_t)(((start + entry->vme_start) + - offset) + remove_size)); } } entry = entry->vme_next; - } + } + vm_map_unlock_read(sub_map); return; } @@ -3562,14 +5108,14 @@ vm_map_delete( { vm_map_entry_t entry, next; struct vm_map_entry *first_entry, tmp_entry; - register vm_map_offset_t s, e; + register vm_map_offset_t s; register vm_object_t object; boolean_t need_wakeup; unsigned int last_timestamp = ~0; /* unlikely value */ int interruptible; interruptible = (flags & VM_MAP_REMOVE_INTERRUPTIBLE) ? - THREAD_ABORTSAFE : THREAD_UNINT; + THREAD_ABORTSAFE : THREAD_UNINT; /* * All our DMA I/O operations in IOKit are currently done by @@ -3584,29 +5130,85 @@ vm_map_delete( */ flags |= VM_MAP_REMOVE_WAIT_FOR_KWIRE; - /* - * Find the start of the region, and clip it - */ - if (vm_map_lookup_entry(map, start, &first_entry)) { - entry = first_entry; - vm_map_clip_start(map, entry, start); - + while(1) { /* - * Fix the lookup hint now, rather than each - * time through the loop. + * Find the start of the region, and clip it */ - SAVE_HINT(map, entry->vme_prev); - } else { - entry = first_entry->vme_next; + 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); + 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 { + vm_map_clip_start(map, entry, start); + } + + /* + * Fix the lookup hint now, rather than each + * time through the loop. + */ + SAVE_HINT_MAP_WRITE(map, entry->vme_prev); + } else { + entry = first_entry->vme_next; + } + break; } + if (entry->superpage_size) + end = SUPERPAGE_ROUND_UP(end); need_wakeup = FALSE; /* * Step through all entries in this region */ - while ((entry != vm_map_to_entry(map)) && (entry->vme_start < end)) { + s = entry->vme_start; + while ((entry != vm_map_to_entry(map)) && (s < end)) { + /* + * At this point, we have deleted all the memory entries + * between "start" and "s". We still need to delete + * all memory entries between "s" and "end". + * While we were blocked and the map was unlocked, some + * new memory entries could have been re-allocated between + * "start" and "s" and we don't want to mess with those. + * Some of those entries could even have been re-assembled + * with an entry after "s" (in vm_map_simplify_entry()), so + * we may have to vm_map_clip_start() again. + */ + + if (entry->vme_start >= s) { + /* + * This entry starts on or after "s" + * so no need to clip its start. + */ + } else { + /* + * This entry has been re-assembled by a + * vm_map_simplify_entry(). We need to + * re-clip its start. + */ + vm_map_clip_start(map, entry, s); + } + if (entry->vme_end <= end) { + /* + * This entry is going away completely, so no need + * to clip and possibly cause an unnecessary unnesting. + */ + } else { + 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); + } + - vm_map_clip_end(map, entry, end); if (entry->in_transition) { wait_result_t wait_result; @@ -3614,7 +5216,7 @@ vm_map_delete( * Another thread is wiring/unwiring this entry. * Let the other thread know we are waiting. */ - s = entry->vme_start; + assert(s == entry->vme_start); entry->needs_wakeup = TRUE; /* @@ -3649,29 +5251,37 @@ vm_map_delete( * User: use the next entry */ entry = first_entry->vme_next; + s = entry->vme_start; } else { entry = first_entry; - SAVE_HINT(map, entry->vme_prev); + SAVE_HINT_MAP_WRITE(map, entry->vme_prev); } last_timestamp = map->timestamp; continue; } /* end in_transition */ if (entry->wired_count) { + boolean_t user_wire; + + user_wire = entry->user_wired_count > 0; + /* - * Remove a kernel wiring if requested or if - * there are user wirings. + * Remove a kernel wiring if requested */ - if ((flags & VM_MAP_REMOVE_KUNWIRE) || - (entry->user_wired_count > 0)) + if (flags & VM_MAP_REMOVE_KUNWIRE) { entry->wired_count--; - - /* remove all user wire references */ - entry->user_wired_count = 0; + } + + /* + * Remove all user wirings for proper accounting + */ + if (entry->user_wired_count > 0) { + while (entry->user_wired_count) + subtract_wire_counts(map, entry, user_wire); + } if (entry->wired_count != 0) { - assert((map != kernel_map) && - (!entry->is_sub_map)); + assert(map != kernel_map); /* * Cannot continue. Typical case is when * a user thread has physical io pending on @@ -3682,39 +5292,39 @@ vm_map_delete( if (flags & VM_MAP_REMOVE_WAIT_FOR_KWIRE) { wait_result_t wait_result; - s = entry->vme_start; + assert(s == entry->vme_start); entry->needs_wakeup = TRUE; wait_result = vm_map_entry_wait(map, - interruptible); + interruptible); if (interruptible && - wait_result == THREAD_INTERRUPTED) { + wait_result == THREAD_INTERRUPTED) { /* - * We do not clear 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; } /* - * The entry could have been clipped or + * The entry could have been clipped or * it may not exist anymore. Look it * up again. - */ + */ if (!vm_map_lookup_entry(map, s, - &first_entry)) { - assert((map != kernel_map) && - (!entry->is_sub_map)); + &first_entry)) { + assert(map != kernel_map); /* - * User: use the next entry - */ + * User: use the next entry + */ entry = first_entry->vme_next; + s = entry->vme_start; } else { entry = first_entry; - SAVE_HINT(map, entry->vme_prev); + SAVE_HINT_MAP_WRITE(map, entry->vme_prev); } last_timestamp = map->timestamp; continue; @@ -3729,17 +5339,52 @@ vm_map_delete( * copy current entry. see comment in vm_map_wire() */ tmp_entry = *entry; - s = entry->vme_start; - e = entry->vme_end; + assert(s == entry->vme_start); /* * We can unlock the map now. The in_transition * state guarentees existance of the entry. */ vm_map_unlock(map); - vm_fault_unwire(map, &tmp_entry, - tmp_entry.object.vm_object == kernel_object, - map->pmap, tmp_entry.vme_start); + + if (tmp_entry.is_sub_map) { + vm_map_t sub_map; + vm_map_offset_t sub_start, sub_end; + pmap_t pmap; + vm_map_offset_t pmap_addr; + + + sub_map = tmp_entry.object.sub_map; + sub_start = tmp_entry.offset; + sub_end = sub_start + (tmp_entry.vme_end - + tmp_entry.vme_start); + if (tmp_entry.use_pmap) { + pmap = sub_map->pmap; + pmap_addr = tmp_entry.vme_start; + } else { + pmap = map->pmap; + pmap_addr = tmp_entry.vme_start; + } + (void) vm_map_unwire_nested(sub_map, + sub_start, sub_end, + user_wire, + pmap, pmap_addr); + } else { + + if (tmp_entry.object.vm_object == kernel_object) { + pmap_protect_options( + map->pmap, + tmp_entry.vme_start, + tmp_entry.vme_end, + VM_PROT_NONE, + PMAP_OPTIONS_REMOVE, + NULL); + } + vm_fault_unwire(map, &tmp_entry, + tmp_entry.object.vm_object == kernel_object, + map->pmap, tmp_entry.vme_start); + } + vm_map_lock(map); if (last_timestamp+1 != map->timestamp) { @@ -3749,13 +5394,14 @@ vm_map_delete( */ if (!vm_map_lookup_entry(map, s, &first_entry)){ assert((map != kernel_map) && - (!entry->is_sub_map)); + (!entry->is_sub_map)); first_entry = first_entry->vme_next; + s = first_entry->vme_start; } else { - SAVE_HINT(map, entry->vme_prev); + SAVE_HINT_MAP_WRITE(map, entry->vme_prev); } } else { - SAVE_HINT(map, entry->vme_prev); + SAVE_HINT_MAP_WRITE(map, entry->vme_prev); first_entry = entry; } @@ -3784,15 +5430,27 @@ vm_map_delete( assert(entry->wired_count == 0); assert(entry->user_wired_count == 0); - if ((!entry->is_sub_map && - entry->object.vm_object != kernel_object) || - entry->is_sub_map) { - if(entry->is_sub_map) { - if(entry->use_pmap) { -#ifndef i386 - pmap_unnest(map->pmap, (addr64_t)entry->vme_start); -#endif - if((map->mapped) && (map->ref_count)) { + assert(s == entry->vme_start); + + if (flags & VM_MAP_REMOVE_NO_PMAP_CLEANUP) { + /* + * XXX with the VM_MAP_REMOVE_SAVE_ENTRIES flag to + * vm_map_delete(), some map entries might have been + * transferred to a "zap_map", which doesn't have a + * pmap. The original pmap has already been flushed + * in the vm_map_delete() call targeting the original + * map, but when we get to destroying the "zap_map", + * we don't have any pmap to flush, so let's just skip + * all this. + */ + } else if (entry->is_sub_map) { + if (entry->use_pmap) { +#ifndef NO_NESTED_PMAP + pmap_unnest(map->pmap, + (addr64_t)entry->vme_start, + entry->vme_end - entry->vme_start); +#endif /* NO_NESTED_PMAP */ + if ((map->mapped_in_other_pmaps) && (map->ref_count)) { /* clean up parent map/maps */ vm_map_submap_pmap_clean( map, entry->vme_start, @@ -3800,26 +5458,41 @@ vm_map_delete( entry->object.sub_map, entry->offset); } - } else { + } else { vm_map_submap_pmap_clean( map, entry->vme_start, entry->vme_end, entry->object.sub_map, entry->offset); - } - } else { - object = entry->object.vm_object; - if((map->mapped) && (map->ref_count)) { - vm_object_pmap_protect( + } + } 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)) { + vm_object_pmap_protect_options( object, entry->offset, entry->vme_end - entry->vme_start, PMAP_NULL, entry->vme_start, - VM_PROT_NONE); - } else { - pmap_remove(map->pmap, - entry->vme_start, - entry->vme_end); - } + VM_PROT_NONE, + PMAP_OPTIONS_REMOVE); + } else if ((entry->object.vm_object != + VM_OBJECT_NULL) || + (map->pmap == kernel_pmap)) { + /* Remove translations associated + * with this range unless the entry + * does not have an object, or + * it's the kernel map or a descendant + * since the platform could potentially + * create "backdoor" mappings invisible + * to the VM. It is expected that + * objectless, non-kernel ranges + * do not have such VM invisible + * translations. + */ + pmap_remove_options(map->pmap, + (addr64_t)entry->vme_start, + (addr64_t)entry->vme_end, + PMAP_OPTIONS_REMOVE); } } @@ -3837,17 +5510,23 @@ vm_map_delete( if ((flags & VM_MAP_REMOVE_SAVE_ENTRIES) && zap_map != VM_MAP_NULL) { + vm_map_size_t entry_size; /* * The caller wants to save the affected VM map entries * into the "zap_map". The caller will take care of * these entries. */ /* unlink the entry from "map" ... */ - vm_map_entry_unlink(map, entry); + vm_map_store_entry_unlink(map, entry); /* ... and add it to the end of the "zap_map" */ - vm_map_entry_link(zap_map, + vm_map_store_entry_link(zap_map, vm_map_last_entry(zap_map), entry); + entry_size = entry->vme_end - entry->vme_start; + map->size -= entry_size; + zap_map->size += entry_size; + /* we didn't unlock the map, so no timestamp increase */ + last_timestamp--; } else { vm_map_entry_delete(map, entry); /* vm_map_entry_delete unlocks the map */ @@ -3870,8 +5549,9 @@ vm_map_delete( */ if (!vm_map_lookup_entry(map, s, &entry)){ entry = entry->vme_next; + s = entry->vme_start; } else { - SAVE_HINT(map, entry->vme_prev); + SAVE_HINT_MAP_WRITE(map, entry->vme_prev); } /* * others can not only allocate behind us, we can @@ -3880,7 +5560,6 @@ vm_map_delete( if(entry == vm_map_to_entry(map)) { break; } - vm_map_clip_start(map, entry, s); } last_timestamp = map->timestamp; } @@ -3913,6 +5592,15 @@ vm_map_remove( vm_map_lock(map); VM_MAP_RANGE_CHECK(map, start, end); + /* + * For the zone_map, the kernel controls the allocation/freeing of memory. + * Any free to the zone_map should be within the bounds of the map and + * should free up memory. If the VM_MAP_RANGE_CHECK() silently converts a + * free to the zone_map into a no-op, there is a problem and we should + * panic. + */ + if ((map == zone_map) && (start == end)) + panic("Nothing being freed to the zone_map. start = end = %p\n", (void *)start); result = vm_map_delete(map, start, end, flags, VM_MAP_NULL); vm_map_unlock(map); @@ -3931,21 +5619,21 @@ void vm_map_copy_discard( vm_map_copy_t copy) { - TR_DECL("vm_map_copy_discard"); - -/* tr3("enter: copy 0x%x type %d", copy, copy->type);*/ - if (copy == VM_MAP_COPY_NULL) return; switch (copy->type) { case VM_MAP_COPY_ENTRY_LIST: while (vm_map_copy_first_entry(copy) != - vm_map_copy_to_entry(copy)) { + vm_map_copy_to_entry(copy)) { vm_map_entry_t entry = vm_map_copy_first_entry(copy); vm_map_copy_entry_unlink(copy, entry); - vm_object_deallocate(entry->object.vm_object); + if (entry->is_sub_map) { + vm_map_deallocate(entry->object.sub_map); + } else { + vm_object_deallocate(entry->object.vm_object); + } vm_map_copy_entry_dispose(copy, entry); } break; @@ -4044,7 +5732,8 @@ vm_map_overwrite_submap_recurse( * splitting entries in strange ways. */ - dst_end = vm_map_round_page(dst_addr + dst_size); + dst_end = vm_map_round_page(dst_addr + dst_size, + VM_MAP_PAGE_MASK(dst_map)); vm_map_lock(dst_map); start_pass_1: @@ -4053,7 +5742,11 @@ start_pass_1: return(KERN_INVALID_ADDRESS); } - vm_map_clip_start(dst_map, tmp_entry, vm_map_trunc_page(dst_addr)); + vm_map_clip_start(dst_map, + tmp_entry, + vm_map_trunc_page(dst_addr, + VM_MAP_PAGE_MASK(dst_map))); + assert(!tmp_entry->use_pmap); /* clipping did unnest if needed */ for (entry = tmp_entry;;) { vm_map_entry_t next; @@ -4065,9 +5758,9 @@ start_pass_1: vm_map_offset_t local_end; if (entry->in_transition) { - /* - * Say that we are waiting, and wait for entry. - */ + /* + * Say that we are waiting, and wait for entry. + */ entry->needs_wakeup = TRUE; vm_map_entry_wait(dst_map, THREAD_UNINT); @@ -4087,9 +5780,9 @@ start_pass_1: vm_map_unlock(dst_map); result = vm_map_overwrite_submap_recurse( - entry->object.sub_map, - sub_start, - sub_end - sub_start); + entry->object.sub_map, + sub_start, + sub_end - sub_start); if(result != KERN_SUCCESS) return result; @@ -4146,8 +5839,8 @@ start_pass_1: * 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))) { + ((!entry->object.vm_object->internal) || + (entry->object.vm_object->true_share))) { if(encountered_sub_map) { vm_map_unlock(dst_map); return(KERN_FAILURE); @@ -4218,7 +5911,8 @@ vm_map_copy_overwrite_nested( vm_map_address_t dst_addr, vm_map_copy_t copy, boolean_t interruptible, - pmap_t pmap) + pmap_t pmap, + boolean_t discard_on_success) { vm_map_offset_t dst_end; vm_map_entry_t tmp_entry; @@ -4246,8 +5940,8 @@ vm_map_copy_overwrite_nested( if (copy->type == VM_MAP_COPY_KERNEL_BUFFER) { return(vm_map_copyout_kernel_buffer( - dst_map, &dst_addr, - copy, TRUE)); + dst_map, &dst_addr, + copy, TRUE, discard_on_success)); } /* @@ -4258,7 +5952,8 @@ vm_map_copy_overwrite_nested( assert(copy->type == VM_MAP_COPY_ENTRY_LIST); if (copy->size == 0) { - vm_map_copy_discard(copy); + if (discard_on_success) + vm_map_copy_discard(copy); return(KERN_SUCCESS); } @@ -4269,12 +5964,17 @@ vm_map_copy_overwrite_nested( * splitting entries in strange ways. */ - if (!page_aligned(copy->size) || - !page_aligned (copy->offset) || - !page_aligned (dst_addr)) + if (!VM_MAP_PAGE_ALIGNED(copy->size, + VM_MAP_PAGE_MASK(dst_map)) || + !VM_MAP_PAGE_ALIGNED(copy->offset, + VM_MAP_PAGE_MASK(dst_map)) || + !VM_MAP_PAGE_ALIGNED(dst_addr, + VM_MAP_PAGE_MASK(dst_map)) || + dst_map->hdr.page_shift != copy->cpy_hdr.page_shift) { aligned = FALSE; - dst_end = vm_map_round_page(dst_addr + copy->size); + dst_end = vm_map_round_page(dst_addr + copy->size, + VM_MAP_PAGE_MASK(dst_map)); } else { dst_end = dst_addr + copy->size; } @@ -4295,7 +5995,10 @@ start_pass_1: vm_map_unlock(dst_map); return(KERN_INVALID_ADDRESS); } - vm_map_clip_start(dst_map, tmp_entry, vm_map_trunc_page(dst_addr)); + vm_map_clip_start(dst_map, + tmp_entry, + vm_map_trunc_page(dst_addr, + VM_MAP_PAGE_MASK(dst_map))); for (entry = tmp_entry;;) { vm_map_entry_t next = entry->vme_next; @@ -4306,9 +6009,9 @@ start_pass_1: if (entry->in_transition) { - /* - * Say that we are waiting, and wait for entry. - */ + /* + * Say that we are waiting, and wait for entry. + */ entry->needs_wakeup = TRUE; vm_map_entry_wait(dst_map, THREAD_UNINT); @@ -4391,8 +6094,8 @@ start_pass_1: * 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))) { + ((!entry->object.vm_object->internal) || + (entry->object.vm_object->true_share))) { contains_permanent_objects = TRUE; } @@ -4442,7 +6145,7 @@ start_overwrite: vm_map_entry_t next_copy = VM_MAP_ENTRY_NULL; int nentries; int remaining_entries = 0; - int new_offset = 0; + vm_map_offset_t new_offset = 0; for (entry = tmp_entry; copy_size == 0;) { vm_map_entry_t next; @@ -4465,7 +6168,7 @@ start_overwrite: vm_map_entry_wait(dst_map, THREAD_UNINT); if(!vm_map_lookup_entry(dst_map, base_addr, - &tmp_entry)) { + &tmp_entry)) { vm_map_unlock(dst_map); return(KERN_INVALID_ADDRESS); } @@ -4494,6 +6197,7 @@ start_overwrite: dst_map, entry, sub_end); vm_map_clip_start( dst_map, entry, sub_start); + assert(!entry->use_pmap); entry->is_sub_map = FALSE; vm_map_deallocate( entry->object.sub_map); @@ -4501,13 +6205,21 @@ start_overwrite: 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->max_protection = VM_PROT_ALL; entry->wired_count = 0; entry->user_wired_count = 0; if(entry->inheritance - == VM_INHERIT_SHARE) - entry->inheritance = VM_INHERIT_COPY; + == VM_INHERIT_SHARE) + entry->inheritance = VM_INHERIT_COPY; continue; } /* first take care of any non-sub_map */ @@ -4535,45 +6247,45 @@ start_overwrite: vm_map_size_t local_size = 0; vm_map_size_t entry_size; - nentries = 1; - new_offset = copy->offset; - copy_entry = vm_map_copy_first_entry(copy); - while(copy_entry != - vm_map_copy_to_entry(copy)){ - entry_size = copy_entry->vme_end - - copy_entry->vme_start; - if((local_size < copy_size) && - ((local_size + entry_size) - >= copy_size)) { - vm_map_copy_clip_end(copy, - copy_entry, - copy_entry->vme_start + - (copy_size - local_size)); - entry_size = copy_entry->vme_end - - copy_entry->vme_start; - local_size += entry_size; - new_offset += entry_size; - } - if(local_size >= copy_size) { - next_copy = copy_entry->vme_next; - copy_entry->vme_next = - vm_map_copy_to_entry(copy); - previous_prev = - copy->cpy_hdr.links.prev; - copy->cpy_hdr.links.prev = copy_entry; - copy->size = copy_size; - remaining_entries = - copy->cpy_hdr.nentries; - remaining_entries -= nentries; - copy->cpy_hdr.nentries = nentries; - break; - } else { - local_size += entry_size; - new_offset += entry_size; - nentries++; - } - copy_entry = copy_entry->vme_next; - } + nentries = 1; + new_offset = copy->offset; + copy_entry = vm_map_copy_first_entry(copy); + while(copy_entry != + vm_map_copy_to_entry(copy)){ + entry_size = copy_entry->vme_end - + copy_entry->vme_start; + if((local_size < copy_size) && + ((local_size + entry_size) + >= copy_size)) { + vm_map_copy_clip_end(copy, + copy_entry, + copy_entry->vme_start + + (copy_size - local_size)); + entry_size = copy_entry->vme_end - + copy_entry->vme_start; + local_size += entry_size; + new_offset += entry_size; + } + if(local_size >= copy_size) { + next_copy = copy_entry->vme_next; + copy_entry->vme_next = + vm_map_copy_to_entry(copy); + previous_prev = + copy->cpy_hdr.links.prev; + copy->cpy_hdr.links.prev = copy_entry; + copy->size = copy_size; + remaining_entries = + copy->cpy_hdr.nentries; + remaining_entries -= nentries; + copy->cpy_hdr.nentries = nentries; + break; + } else { + local_size += entry_size; + new_offset += entry_size; + nentries++; + } + copy_entry = copy_entry->vme_next; + } } if((entry->use_pmap) && (pmap == NULL)) { @@ -4582,30 +6294,33 @@ start_overwrite: sub_start, copy, interruptible, - entry->object.sub_map->pmap); + entry->object.sub_map->pmap, + TRUE); } else if (pmap != NULL) { kr = vm_map_copy_overwrite_nested( entry->object.sub_map, sub_start, copy, - interruptible, pmap); + interruptible, pmap, + TRUE); } else { kr = vm_map_copy_overwrite_nested( entry->object.sub_map, sub_start, copy, interruptible, - dst_map->pmap); + dst_map->pmap, + TRUE); } if(kr != KERN_SUCCESS) { if(next_copy != NULL) { - copy->cpy_hdr.nentries += - remaining_entries; - copy->cpy_hdr.links.prev->vme_next = - next_copy; - copy->cpy_hdr.links.prev - = previous_prev; - copy->size = total_size; + copy->cpy_hdr.nentries += + remaining_entries; + copy->cpy_hdr.links.prev->vme_next = + next_copy; + copy->cpy_hdr.links.prev + = previous_prev; + copy->size = total_size; } return kr; } @@ -4615,29 +6330,35 @@ start_overwrite: /* otherwise copy no longer exists, it was */ /* destroyed after successful copy_overwrite */ copy = (vm_map_copy_t) - zalloc(vm_map_copy_zone); + zalloc(vm_map_copy_zone); vm_map_copy_first_entry(copy) = - vm_map_copy_last_entry(copy) = - vm_map_copy_to_entry(copy); + vm_map_copy_last_entry(copy) = + vm_map_copy_to_entry(copy); copy->type = VM_MAP_COPY_ENTRY_LIST; copy->offset = new_offset; + /* + * XXX FBDP + * this does not seem to deal with + * the VM map store (R&B tree) + */ + total_size -= copy_size; copy_size = 0; /* put back remainder of copy in container */ if(next_copy != NULL) { - copy->cpy_hdr.nentries = remaining_entries; - copy->cpy_hdr.links.next = next_copy; - copy->cpy_hdr.links.prev = previous_prev; - copy->size = total_size; - next_copy->vme_prev = - vm_map_copy_to_entry(copy); - next_copy = NULL; + copy->cpy_hdr.nentries = remaining_entries; + copy->cpy_hdr.links.next = next_copy; + copy->cpy_hdr.links.prev = previous_prev; + copy->size = total_size; + next_copy->vme_prev = + vm_map_copy_to_entry(copy); + next_copy = NULL; } base_addr = local_end; vm_map_lock(dst_map); if(!vm_map_lookup_entry(dst_map, - local_end, &tmp_entry)) { + local_end, &tmp_entry)) { vm_map_unlock(dst_map); return(KERN_INVALID_ADDRESS); } @@ -4650,7 +6371,7 @@ start_overwrite: } if ((next == vm_map_to_entry(dst_map)) || - (next->vme_start != entry->vme_end)) { + (next->vme_start != entry->vme_end)) { vm_map_unlock(dst_map); return(KERN_INVALID_ADDRESS); } @@ -4670,15 +6391,15 @@ start_overwrite: copy_entry = vm_map_copy_first_entry(copy); while(copy_entry != vm_map_copy_to_entry(copy)) { entry_size = copy_entry->vme_end - - copy_entry->vme_start; + copy_entry->vme_start; if((local_size < copy_size) && - ((local_size + entry_size) - >= copy_size)) { + ((local_size + entry_size) + >= copy_size)) { vm_map_copy_clip_end(copy, copy_entry, - copy_entry->vme_start + - (copy_size - local_size)); + copy_entry->vme_start + + (copy_size - local_size)); entry_size = copy_entry->vme_end - - copy_entry->vme_start; + copy_entry->vme_start; local_size += entry_size; new_offset += entry_size; } @@ -4713,39 +6434,44 @@ start_overwrite: local_pmap = dst_map->pmap; if ((kr = vm_map_copy_overwrite_aligned( - dst_map, tmp_entry, copy, - base_addr, local_pmap)) != KERN_SUCCESS) { + dst_map, tmp_entry, copy, + base_addr, local_pmap)) != KERN_SUCCESS) { if(next_copy != NULL) { copy->cpy_hdr.nentries += - remaining_entries; + remaining_entries; copy->cpy_hdr.links.prev->vme_next = - next_copy; + next_copy; copy->cpy_hdr.links.prev = - previous_prev; + previous_prev; copy->size += copy_size; } return kr; } vm_map_unlock(dst_map); } else { - /* - * Performance gain: - * - * if the copy and dst address are misaligned but the same - * offset within the page we can copy_not_aligned the - * misaligned parts and copy aligned the rest. If they are - * aligned but len is unaligned we simply need to copy - * the end bit unaligned. We'll need to split the misaligned - * bits of the region in this case ! - */ - /* ALWAYS UNLOCKS THE dst_map MAP */ - if ((kr = vm_map_copy_overwrite_unaligned( dst_map, - tmp_entry, copy, base_addr)) != KERN_SUCCESS) { + /* + * Performance gain: + * + * if the copy and dst address are misaligned but the same + * offset within the page we can copy_not_aligned the + * misaligned parts and copy aligned the rest. If they are + * aligned but len is unaligned we simply need to copy + * the end bit unaligned. We'll need to split the misaligned + * bits of the region in this case ! + */ + /* ALWAYS UNLOCKS THE dst_map MAP */ + kr = vm_map_copy_overwrite_unaligned( + dst_map, + tmp_entry, + copy, + base_addr, + discard_on_success); + if (kr != KERN_SUCCESS) { if(next_copy != NULL) { copy->cpy_hdr.nentries += - remaining_entries; + remaining_entries; copy->cpy_hdr.links.prev->vme_next = - next_copy; + next_copy; copy->cpy_hdr.links.prev = previous_prev; copy->size += copy_size; @@ -4769,7 +6495,7 @@ start_overwrite: vm_map_lock(dst_map); while(TRUE) { if (!vm_map_lookup_entry(dst_map, - base_addr, &tmp_entry)) { + base_addr, &tmp_entry)) { vm_map_unlock(dst_map); return(KERN_INVALID_ADDRESS); } @@ -4780,7 +6506,10 @@ start_overwrite: break; } } - vm_map_clip_start(dst_map, tmp_entry, vm_map_trunc_page(base_addr)); + vm_map_clip_start(dst_map, + tmp_entry, + vm_map_trunc_page(base_addr, + VM_MAP_PAGE_MASK(dst_map))); entry = tmp_entry; } /* while */ @@ -4788,7 +6517,8 @@ start_overwrite: /* * Throw away the vm_map_copy object */ - vm_map_copy_discard(copy); + if (discard_on_success) + vm_map_copy_discard(copy); return(KERN_SUCCESS); }/* vm_map_copy_overwrite */ @@ -4800,8 +6530,239 @@ vm_map_copy_overwrite( vm_map_copy_t copy, boolean_t interruptible) { - return vm_map_copy_overwrite_nested( - dst_map, dst_addr, copy, interruptible, (pmap_t) NULL); + vm_map_size_t head_size, tail_size; + vm_map_copy_t head_copy, tail_copy; + vm_map_offset_t head_addr, tail_addr; + vm_map_entry_t entry; + kern_return_t kr; + + head_size = 0; + tail_size = 0; + head_copy = NULL; + tail_copy = NULL; + head_addr = 0; + tail_addr = 0; + + if (interruptible || + copy == VM_MAP_COPY_NULL || + copy->type != VM_MAP_COPY_ENTRY_LIST) { + /* + * We can't split the "copy" map if we're interruptible + * or if we don't have a "copy" map... + */ + blunt_copy: + return vm_map_copy_overwrite_nested(dst_map, + dst_addr, + copy, + interruptible, + (pmap_t) NULL, + TRUE); + } + + if (copy->size < 3 * PAGE_SIZE) { + /* + * Too small to bother with optimizing... + */ + goto blunt_copy; + } + + if ((dst_addr & VM_MAP_PAGE_MASK(dst_map)) != + (copy->offset & VM_MAP_PAGE_MASK(dst_map))) { + /* + * Incompatible mis-alignment of source and destination... + */ + goto blunt_copy; + } + + /* + * Proper alignment or identical mis-alignment at the beginning. + * Let's try and do a small unaligned copy first (if needed) + * and then an aligned copy for the rest. + */ + if (!page_aligned(dst_addr)) { + head_addr = dst_addr; + head_size = (VM_MAP_PAGE_SIZE(dst_map) - + (copy->offset & VM_MAP_PAGE_MASK(dst_map))); + } + if (!page_aligned(copy->offset + copy->size)) { + /* + * Mis-alignment at the end. + * Do an aligned copy up to the last page and + * then an unaligned copy for the remaining bytes. + */ + tail_size = ((copy->offset + copy->size) & + VM_MAP_PAGE_MASK(dst_map)); + tail_addr = dst_addr + copy->size - tail_size; + } + + if (head_size + tail_size == copy->size) { + /* + * It's all unaligned, no optimization possible... + */ + goto blunt_copy; + } + + /* + * Can't optimize if there are any submaps in the + * destination due to the way we free the "copy" map + * progressively in vm_map_copy_overwrite_nested() + * in that case. + */ + vm_map_lock_read(dst_map); + if (! vm_map_lookup_entry(dst_map, dst_addr, &entry)) { + vm_map_unlock_read(dst_map); + goto blunt_copy; + } + for (; + (entry != vm_map_copy_to_entry(copy) && + entry->vme_start < dst_addr + copy->size); + entry = entry->vme_next) { + if (entry->is_sub_map) { + vm_map_unlock_read(dst_map); + goto blunt_copy; + } + } + vm_map_unlock_read(dst_map); + + if (head_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->type = VM_MAP_COPY_ENTRY_LIST; + head_copy->cpy_hdr.nentries = 0; + head_copy->cpy_hdr.entries_pageable = + copy->cpy_hdr.entries_pageable; + vm_map_store_init(&head_copy->cpy_hdr); + + head_copy->offset = copy->offset; + head_copy->size = head_size; + + copy->offset += head_size; + copy->size -= head_size; + + entry = vm_map_copy_first_entry(copy); + vm_map_copy_clip_end(copy, entry, copy->offset); + vm_map_copy_entry_unlink(copy, entry); + vm_map_copy_entry_link(head_copy, + vm_map_copy_to_entry(head_copy), + entry); + + /* + * Do the unaligned copy. + */ + kr = vm_map_copy_overwrite_nested(dst_map, + head_addr, + head_copy, + interruptible, + (pmap_t) NULL, + FALSE); + if (kr != KERN_SUCCESS) + goto done; + } + + if (tail_size) { + /* + * 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->type = VM_MAP_COPY_ENTRY_LIST; + tail_copy->cpy_hdr.nentries = 0; + tail_copy->cpy_hdr.entries_pageable = + copy->cpy_hdr.entries_pageable; + vm_map_store_init(&tail_copy->cpy_hdr); + + tail_copy->offset = copy->offset + copy->size - tail_size; + tail_copy->size = tail_size; + + copy->size -= tail_size; + + entry = vm_map_copy_last_entry(copy); + vm_map_copy_clip_start(copy, entry, tail_copy->offset); + entry = vm_map_copy_last_entry(copy); + vm_map_copy_entry_unlink(copy, entry); + vm_map_copy_entry_link(tail_copy, + vm_map_copy_last_entry(tail_copy), + entry); + } + + /* + * Copy most (or possibly all) of the data. + */ + kr = vm_map_copy_overwrite_nested(dst_map, + dst_addr + head_size, + copy, + interruptible, + (pmap_t) NULL, + FALSE); + if (kr != KERN_SUCCESS) { + goto done; + } + + if (tail_size) { + kr = vm_map_copy_overwrite_nested(dst_map, + tail_addr, + tail_copy, + interruptible, + (pmap_t) NULL, + FALSE); + } + +done: + assert(copy->type == VM_MAP_COPY_ENTRY_LIST); + if (kr == KERN_SUCCESS) { + /* + * Discard all the copy maps. + */ + if (head_copy) { + vm_map_copy_discard(head_copy); + head_copy = NULL; + } + vm_map_copy_discard(copy); + if (tail_copy) { + vm_map_copy_discard(tail_copy); + tail_copy = NULL; + } + } else { + /* + * Re-assemble the original copy map. + */ + if (head_copy) { + entry = vm_map_copy_first_entry(head_copy); + vm_map_copy_entry_unlink(head_copy, entry); + vm_map_copy_entry_link(copy, + vm_map_copy_to_entry(copy), + entry); + copy->offset -= head_size; + copy->size += head_size; + vm_map_copy_discard(head_copy); + head_copy = NULL; + } + if (tail_copy) { + entry = vm_map_copy_last_entry(tail_copy); + vm_map_copy_entry_unlink(tail_copy, entry); + vm_map_copy_entry_link(copy, + vm_map_copy_last_entry(copy), + entry); + copy->size += tail_size; + vm_map_copy_discard(tail_copy); + tail_copy = NULL; + } + } + return kr; } @@ -4833,9 +6794,11 @@ vm_map_copy_overwrite_unaligned( vm_map_t dst_map, vm_map_entry_t entry, vm_map_copy_t copy, - vm_map_offset_t start) + vm_map_offset_t start, + boolean_t discard_on_success) { - vm_map_entry_t copy_entry = vm_map_copy_first_entry(copy); + vm_map_entry_t copy_entry; + vm_map_entry_t copy_entry_next; vm_map_version_t version; vm_object_t dst_object; vm_object_offset_t dst_offset; @@ -4848,6 +6811,9 @@ vm_map_copy_overwrite_unaligned( amount_left; kern_return_t kr = KERN_SUCCESS; + + copy_entry = vm_map_copy_first_entry(copy); + vm_map_lock_write_to_read(dst_map); src_offset = copy->offset - vm_object_trunc_page(copy->offset); @@ -4895,16 +6861,16 @@ vm_map_copy_overwrite_unaligned( * Copy on write region. */ if (entry->needs_copy && - ((entry->protection & VM_PROT_WRITE) != 0)) + ((entry->protection & VM_PROT_WRITE) != 0)) { if (vm_map_lock_read_to_write(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)); + &entry->offset, + (vm_map_size_t)(entry->vme_end + - entry->vme_start)); entry->needs_copy = FALSE; vm_map_lock_write_to_read(dst_map); } @@ -4919,7 +6885,7 @@ vm_map_copy_overwrite_unaligned( goto RetryLookup; } dst_object = vm_object_allocate((vm_map_size_t) - entry->vme_end - entry->vme_start); + entry->vme_end - entry->vme_start); entry->object.vm_object = dst_object; entry->offset = 0; vm_map_lock_write_to_read(dst_map); @@ -4960,22 +6926,31 @@ vm_map_copy_overwrite_unaligned( return kr; if ((copy_entry->vme_start + src_offset) == copy_entry->vme_end - || amount_left == 0) + || amount_left == 0) { /* * all done with this copy entry, dispose. */ - vm_map_copy_entry_unlink(copy, copy_entry); - vm_object_deallocate(copy_entry->object.vm_object); - vm_map_copy_entry_dispose(copy, copy_entry); + copy_entry_next = copy_entry->vme_next; + + if (discard_on_success) { + vm_map_copy_entry_unlink(copy, copy_entry); + assert(!copy_entry->is_sub_map); + vm_object_deallocate( + copy_entry->object.vm_object); + vm_map_copy_entry_dispose(copy, copy_entry); + } - if ((copy_entry = vm_map_copy_first_entry(copy)) - == vm_map_copy_to_entry(copy) && amount_left) { + if (copy_entry_next == vm_map_copy_to_entry(copy) && + amount_left) { /* * not finished copying but run out of source */ return KERN_INVALID_ADDRESS; } + + copy_entry = copy_entry_next; + src_offset = 0; } @@ -5006,7 +6981,7 @@ vm_map_copy_overwrite_unaligned( * we must lookup the entry because somebody * might have changed the map behind our backs. */ -RetryLookup: + RetryLookup: if (!vm_map_lookup_entry(dst_map, start, &entry)) { vm_map_unlock_read(dst_map); @@ -5037,16 +7012,17 @@ RetryLookup: * to the above pass and make sure that no wiring is involved. */ +int vm_map_copy_overwrite_aligned_src_not_internal = 0; +int vm_map_copy_overwrite_aligned_src_not_symmetric = 0; +int vm_map_copy_overwrite_aligned_src_large = 0; + static kern_return_t vm_map_copy_overwrite_aligned( vm_map_t dst_map, vm_map_entry_t tmp_entry, vm_map_copy_t copy, vm_map_offset_t start, -#if !BAD_OPTIMIZATION - __unused -#endif /* !BAD_OPTIMIZATION */ - pmap_t pmap) + __unused pmap_t pmap) { vm_object_t object; vm_map_entry_t copy_entry; @@ -5055,11 +7031,12 @@ vm_map_copy_overwrite_aligned( vm_map_entry_t entry; while ((copy_entry = vm_map_copy_first_entry(copy)) - != vm_map_copy_to_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 == vm_map_to_entry(dst_map)) { vm_map_unlock(dst_map); return KERN_INVALID_ADDRESS; @@ -5073,7 +7050,7 @@ vm_map_copy_overwrite_aligned( */ if ((entry->vme_start != start) || ((entry->is_sub_map) - && !entry->needs_copy)) { + && !entry->needs_copy)) { vm_map_unlock(dst_map); return(KERN_INVALID_ADDRESS); } @@ -5103,7 +7080,7 @@ vm_map_copy_overwrite_aligned( if (size < copy_size) { vm_map_copy_clip_end(copy, copy_entry, - copy_entry->vme_start + size); + copy_entry->vme_start + size); copy_size = size; } @@ -5119,8 +7096,8 @@ vm_map_copy_overwrite_aligned( object = entry->object.vm_object; if ((!entry->is_shared && - ((object == VM_OBJECT_NULL) || - (object->internal && !object->true_share))) || + ((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; @@ -5143,21 +7120,101 @@ vm_map_copy_overwrite_aligned( continue; } +#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 && + copy_size <= __TRADEOFF1_COPY_SIZE) { + /* + * Virtual vs. Physical copy tradeoff #1. + * + * Copying only a few pages out of a large + * object: do a physical copy instead of + * a virtual copy, to avoid possibly keeping + * the entire large object alive because of + * those few copy-on-write pages. + */ + vm_map_copy_overwrite_aligned_src_large++; + goto slow_copy; + } + + if (entry->alias >= VM_MEMORY_MALLOC && + entry->alias <= 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; + if (new_object != VM_OBJECT_NULL) { + vm_object_lock_shared(new_object); + } + while (new_object != VM_OBJECT_NULL && + !new_object->true_share && + new_object->copy_strategy == MEMORY_OBJECT_COPY_SYMMETRIC && + new_object->internal) { + new_shadow = new_object->shadow; + if (new_shadow == VM_OBJECT_NULL) { + break; + } + vm_object_lock_shared(new_shadow); + vm_object_unlock(new_object); + new_object = new_shadow; + } + if (new_object != VM_OBJECT_NULL) { + if (!new_object->internal) { + /* + * The new mapping is backed + * by an external object. We + * don't want malloc'ed memory + * to be replaced with such a + * non-anonymous mapping, so + * let's go off the optimized + * path... + */ + vm_map_copy_overwrite_aligned_src_not_internal++; + vm_object_unlock(new_object); + goto slow_copy; + } + if (new_object->true_share || + new_object->copy_strategy != MEMORY_OBJECT_COPY_SYMMETRIC) { + /* + * Same if there's a "true_share" + * object in the shadow chain, or + * an object with a non-default + * (SYMMETRIC) copy strategy. + */ + vm_map_copy_overwrite_aligned_src_not_symmetric++; + vm_object_unlock(new_object); + goto slow_copy; + } + vm_object_unlock(new_object); + } + /* + * The new mapping is still backed by + * anonymous (internal) memory, so it's + * OK to substitute it for the original + * malloc() mapping. + */ + } + if (old_object != VM_OBJECT_NULL) { if(entry->is_sub_map) { if(entry->use_pmap) { -#ifndef i386 +#ifndef NO_NESTED_PMAP pmap_unnest(dst_map->pmap, - entry->vme_start); -#endif - if(dst_map->mapped) { + (addr64_t)entry->vme_start, + entry->vme_end - entry->vme_start); +#endif /* NO_NESTED_PMAP */ + if(dst_map->mapped_in_other_pmaps) { /* clean up parent */ /* map/maps */ - vm_map_submap_pmap_clean( - dst_map, entry->vme_start, - entry->vme_end, - entry->object.sub_map, - entry->offset); + vm_map_submap_pmap_clean( + dst_map, entry->vme_start, + entry->vme_end, + entry->object.sub_map, + entry->offset); } } else { vm_map_submap_pmap_clean( @@ -5169,19 +7226,22 @@ vm_map_copy_overwrite_aligned( vm_map_deallocate( entry->object.sub_map); } else { - if(dst_map->mapped) { - vm_object_pmap_protect( + if(dst_map->mapped_in_other_pmaps) { + vm_object_pmap_protect_options( entry->object.vm_object, entry->offset, entry->vme_end - - entry->vme_start, + - entry->vme_start, PMAP_NULL, entry->vme_start, - VM_PROT_NONE); + VM_PROT_NONE, + PMAP_OPTIONS_REMOVE); } else { - pmap_remove(dst_map->pmap, - (addr64_t)(entry->vme_start), - (addr64_t)(entry->vme_end)); + pmap_remove_options( + dst_map->pmap, + (addr64_t)(entry->vme_start), + (addr64_t)(entry->vme_end), + PMAP_OPTIONS_REMOVE); } vm_object_deallocate(old_object); } @@ -5197,85 +7257,16 @@ vm_map_copy_overwrite_aligned( vm_map_copy_entry_unlink(copy, copy_entry); vm_map_copy_entry_dispose(copy, copy_entry); -#if BAD_OPTIMIZATION + /* - * if we turn this optimization back on - * we need to revisit our use of pmap mappings - * large copies will cause us to run out and panic + * we could try to push pages into the pmap at this point, BUT * this optimization only saved on average 2 us per page if ALL * the pages in the source were currently mapped * and ALL the pages in the dest were touched, if there were fewer * than 2/3 of the pages touched, this optimization actually cost more cycles + * it also puts a lot of pressure on the pmap layer w/r to mapping structures */ - /* - * Try to aggressively enter physical mappings - * (but avoid uninstantiated objects) - */ - if (object != VM_OBJECT_NULL) { - vm_map_offset_t va = entry->vme_start; - - while (va < entry->vme_end) { - register vm_page_t m; - vm_prot_t prot; - - /* - * Look for the page in the top object - */ - prot = entry->protection; - vm_object_lock(object); - vm_object_paging_begin(object); - - /* - * ENCRYPTED SWAP: - * If the page is encrypted, skip it: - * we can't let the user see the encrypted - * contents. The page will get decrypted - * on demand when the user generates a - * soft-fault when trying to access it. - */ - if ((m = vm_page_lookup(object,offset)) != - VM_PAGE_NULL && !m->busy && - !m->fictitious && !m->encrypted && - (!m->unusual || (!m->error && - !m->restart && !m->absent && - (prot & m->page_lock) == 0))) { - - m->busy = TRUE; - vm_object_unlock(object); - - /* - * Honor COW obligations - */ - if (entry->needs_copy) - prot &= ~VM_PROT_WRITE; - /* It is our policy to require */ - /* explicit sync from anyone */ - /* writing code and then */ - /* a pc to execute it. */ - /* No isync here */ - - PMAP_ENTER(pmap, va, m, prot, - ((unsigned int) - (m->object->wimg_bits)) - & VM_WIMG_MASK, - FALSE); - - vm_object_lock(object); - vm_page_lock_queues(); - if (!m->active && !m->inactive) - vm_page_activate(m); - vm_page_unlock_queues(); - PAGE_WAKEUP_DONE(m); - } - vm_object_paging_end(object); - vm_object_unlock(object); - - offset += PAGE_SIZE_64; - va += PAGE_SIZE; - } /* end while (va < entry->vme_end) */ - } /* end if (object) */ -#endif /* * Set up for the next iteration. The map * has not been unlocked, so the next @@ -5288,16 +7279,49 @@ vm_map_copy_overwrite_aligned( tmp_entry = tmp_entry->vme_next; } else { vm_map_version_t version; - vm_object_t dst_object = entry->object.vm_object; - vm_object_offset_t dst_offset = entry->offset; + vm_object_t dst_object; + vm_object_offset_t dst_offset; kern_return_t r; + slow_copy: + if (entry->needs_copy) { + vm_object_shadow(&entry->object.vm_object, + &entry->offset, + (entry->vme_end - + entry->vme_start)); + entry->needs_copy = FALSE; + } + + dst_object = entry->object.vm_object; + dst_offset = entry->offset; + /* * Take an object reference, and record * the map version information so that the * map can be safely unlocked. */ + if (dst_object == VM_OBJECT_NULL) { + /* + * We would usually have just taken the + * optimized path above if the destination + * object has not been allocated yet. But we + * now disable that optimization if the copy + * entry's object is not backed by anonymous + * memory to avoid replacing malloc'ed + * (i.e. re-usable) anonymous memory with a + * not-so-anonymous mapping. + * So we have to handle this case here and + * allocate a new VM object for this map entry. + */ + dst_object = vm_object_allocate( + entry->vme_end - entry->vme_start); + dst_offset = 0; + entry->object.vm_object = dst_object; + entry->offset = dst_offset; + + } + vm_object_reference(dst_object); /* account for unlock bumping up timestamp */ @@ -5311,14 +7335,14 @@ vm_map_copy_overwrite_aligned( copy_size = size; r = vm_fault_copy( - copy_entry->object.vm_object, - copy_entry->offset, - ©_size, - dst_object, - dst_offset, - dst_map, - &version, - THREAD_UNINT ); + copy_entry->object.vm_object, + copy_entry->offset, + ©_size, + dst_object, + dst_offset, + dst_map, + &version, + THREAD_UNINT ); /* * Release the object reference @@ -5339,7 +7363,7 @@ vm_map_copy_overwrite_aligned( */ vm_map_copy_clip_end(copy, copy_entry, - copy_entry->vme_start + copy_size); + copy_entry->vme_start + copy_size); vm_map_copy_entry_unlink(copy, copy_entry); vm_object_deallocate(copy_entry->object.vm_object); vm_map_copy_entry_dispose(copy, copy_entry); @@ -5354,7 +7378,8 @@ vm_map_copy_overwrite_aligned( start += copy_size; vm_map_lock(dst_map); - if (version.main_timestamp == dst_map->timestamp) { + if (version.main_timestamp == dst_map->timestamp && + copy_size != 0) { /* We can safely use saved tmp_entry value */ vm_map_clip_end(dst_map, tmp_entry, start); @@ -5394,7 +7419,14 @@ vm_map_copyin_kernel_buffer( { kern_return_t kr; vm_map_copy_t copy; - vm_map_size_t kalloc_size = sizeof(struct vm_map_copy) + len; + 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); copy = (vm_map_copy_t) kalloc(kalloc_size); if (copy == VM_MAP_COPY_NULL) { @@ -5406,18 +7438,21 @@ vm_map_copyin_kernel_buffer( copy->cpy_kdata = (void *) (copy + 1); copy->cpy_kalloc_size = kalloc_size; - kr = copyinmap(src_map, src_addr, copy->cpy_kdata, len); + kr = copyinmap(src_map, src_addr, copy->cpy_kdata, (vm_size_t) len); if (kr != KERN_SUCCESS) { kfree(copy, kalloc_size); return kr; } if (src_destroy) { - (void) vm_map_remove(src_map, vm_map_trunc_page(src_addr), - vm_map_round_page(src_addr + len), - VM_MAP_REMOVE_INTERRUPTIBLE | - VM_MAP_REMOVE_WAIT_FOR_KWIRE | - (src_map == kernel_map) ? - VM_MAP_REMOVE_KUNWIRE : 0); + (void) vm_map_remove( + src_map, + vm_map_trunc_page(src_addr, + VM_MAP_PAGE_MASK(src_map)), + vm_map_round_page(src_addr + len, + VM_MAP_PAGE_MASK(src_map)), + (VM_MAP_REMOVE_INTERRUPTIBLE | + VM_MAP_REMOVE_WAIT_FOR_KWIRE | + (src_map == kernel_map) ? VM_MAP_REMOVE_KUNWIRE : 0)); } *copy_result = copy; return KERN_SUCCESS; @@ -5440,7 +7475,8 @@ vm_map_copyout_kernel_buffer( vm_map_t map, vm_map_address_t *addr, /* IN/OUT */ vm_map_copy_t copy, - boolean_t overwrite) + boolean_t overwrite, + boolean_t consume_on_success) { kern_return_t kr = KERN_SUCCESS; thread_t thread = current_thread(); @@ -5453,7 +7489,8 @@ vm_map_copyout_kernel_buffer( *addr = 0; kr = vm_map_enter(map, addr, - vm_map_round_page(copy->size), + vm_map_round_page(copy->size, + VM_MAP_PAGE_MASK(map)), (vm_map_offset_t) 0, VM_FLAGS_ANYWHERE, VM_OBJECT_NULL, @@ -5475,7 +7512,8 @@ vm_map_copyout_kernel_buffer( * If the target map is the current map, just do * the copy. */ - if (copyout(copy->cpy_kdata, *addr, 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; } } @@ -5490,7 +7528,8 @@ vm_map_copyout_kernel_buffer( vm_map_reference(map); oldmap = vm_map_switch(map); - if (copyout(copy->cpy_kdata, *addr, 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; } @@ -5505,16 +7544,22 @@ vm_map_copyout_kernel_buffer( /* * Deallocate the space we allocated in the target map. */ - (void) vm_map_remove(map, - vm_map_trunc_page(*addr), - vm_map_round_page(*addr + - vm_map_round_page(copy->size)), - VM_MAP_NO_FLAGS); + (void) vm_map_remove( + map, + vm_map_trunc_page(*addr, + VM_MAP_PAGE_MASK(map)), + vm_map_round_page((*addr + + vm_map_round_page(copy->size, + VM_MAP_PAGE_MASK(map))), + VM_MAP_PAGE_MASK(map)), + VM_MAP_NO_FLAGS); *addr = 0; } } else { /* copy was successful, dicard the copy structure */ - kfree(copy, copy->cpy_kalloc_size); + if (consume_on_success) { + kfree(copy, copy->cpy_kalloc_size); + } } return kr; @@ -5533,21 +7578,53 @@ vm_map_copyout_kernel_buffer( */ #define vm_map_copy_insert(map, where, copy) \ MACRO_BEGIN \ - vm_map_t VMCI_map; \ - vm_map_entry_t VMCI_where; \ - vm_map_copy_t VMCI_copy; \ - VMCI_map = (map); \ - VMCI_where = (where); \ - VMCI_copy = (copy); \ - ((VMCI_where->vme_next)->vme_prev = vm_map_copy_last_entry(VMCI_copy))\ - ->vme_next = (VMCI_where->vme_next); \ - ((VMCI_where)->vme_next = vm_map_copy_first_entry(VMCI_copy)) \ - ->vme_prev = VMCI_where; \ - VMCI_map->hdr.nentries += VMCI_copy->cpy_hdr.nentries; \ - UPDATE_FIRST_FREE(VMCI_map, VMCI_map->first_free); \ - zfree(vm_map_copy_zone, VMCI_copy); \ + vm_map_store_copy_insert(map, where, copy); \ + zfree(vm_map_copy_zone, copy); \ MACRO_END +void +vm_map_copy_remap( + vm_map_t map, + vm_map_entry_t where, + vm_map_copy_t copy, + vm_map_offset_t adjustment, + vm_prot_t cur_prot, + vm_prot_t max_prot, + vm_inherit_t inheritance) +{ + vm_map_entry_t copy_entry, new_entry; + + for (copy_entry = vm_map_copy_first_entry(copy); + copy_entry != vm_map_copy_to_entry(copy); + copy_entry = copy_entry->vme_next) { + /* get a new VM map entry for the map */ + new_entry = vm_map_entry_create(map, + !map->hdr.entries_pageable); + /* copy the "copy entry" to the new entry */ + vm_map_entry_copy(new_entry, copy_entry); + /* adjust "start" and "end" */ + new_entry->vme_start += adjustment; + new_entry->vme_end += adjustment; + /* clear some attributes */ + new_entry->inheritance = inheritance; + new_entry->protection = cur_prot; + new_entry->max_protection = max_prot; + new_entry->behavior = VM_BEHAVIOR_DEFAULT; + /* take an extra reference on the entry's "object" */ + if (new_entry->is_sub_map) { + vm_map_lock(new_entry->object.sub_map); + vm_map_reference(new_entry->object.sub_map); + vm_map_unlock(new_entry->object.sub_map); + } else { + vm_object_reference(new_entry->object.vm_object); + } + /* insert the new entry in the map */ + vm_map_store_entry_link(map, where, new_entry); + /* continue inserting the "copy entries" after the new entry */ + where = new_entry; + } +} + /* * Routine: vm_map_copyout * @@ -5558,18 +7635,35 @@ MACRO_END * If successful, consumes the copy object. * Otherwise, the caller is responsible for it. */ + kern_return_t vm_map_copyout( vm_map_t dst_map, vm_map_address_t *dst_addr, /* OUT */ vm_map_copy_t copy) +{ + return vm_map_copyout_internal(dst_map, dst_addr, copy, + TRUE, /* consume_on_success */ + VM_PROT_DEFAULT, + VM_PROT_ALL, + VM_INHERIT_DEFAULT); +} + +kern_return_t +vm_map_copyout_internal( + vm_map_t dst_map, + vm_map_address_t *dst_addr, /* OUT */ + vm_map_copy_t copy, + boolean_t consume_on_success, + vm_prot_t cur_protection, + vm_prot_t max_protection, + vm_inherit_t inheritance) { vm_map_size_t size; vm_map_size_t adjustment; vm_map_offset_t start; vm_object_offset_t vm_copy_start; vm_map_entry_t last; - register vm_map_entry_t entry; /* @@ -5592,8 +7686,10 @@ vm_map_copyout( vm_object_offset_t offset; offset = vm_object_trunc_page(copy->offset); - size = vm_map_round_page(copy->size + - (vm_map_size_t)(copy->offset - offset)); + 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, @@ -5604,7 +7700,8 @@ vm_map_copyout( return(kr); /* Account for non-pagealigned copy object */ *dst_addr += (vm_map_offset_t)(copy->offset - offset); - zfree(vm_map_copy_zone, copy); + if (consume_on_success) + zfree(vm_map_copy_zone, copy); return(KERN_SUCCESS); } @@ -5614,24 +7711,36 @@ vm_map_copyout( */ 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, FALSE, + consume_on_success); } + /* * Find space for the data */ - vm_copy_start = vm_object_trunc_page(copy->offset); - size = vm_map_round_page((vm_map_size_t)copy->offset + copy->size) - - vm_copy_start; + vm_copy_start = vm_map_trunc_page((vm_map_size_t)copy->offset, + VM_MAP_COPY_PAGE_MASK(copy)); + size = vm_map_round_page((vm_map_size_t)copy->offset + copy->size, + VM_MAP_COPY_PAGE_MASK(copy)) + - vm_copy_start; + - StartAgain: ; +StartAgain: ; vm_map_lock(dst_map); - assert(first_free_is_valid(dst_map)); - start = ((last = dst_map->first_free) == vm_map_to_entry(dst_map)) ? + if( dst_map->disable_vmentry_reuse == TRUE) { + 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; + start = vm_map_round_page(start, + VM_MAP_PAGE_MASK(dst_map)); + } while (TRUE) { vm_map_entry_t next = last->vme_next; @@ -5657,6 +7766,23 @@ vm_map_copyout( last = next; start = last->vme_end; + start = vm_map_round_page(start, + VM_MAP_PAGE_MASK(dst_map)); + } + + adjustment = start - vm_copy_start; + if (! consume_on_success) { + /* + * We're not allowed to consume "copy", so we'll have to + * copy its map entries into the destination map below. + * No need to re-allocate map entries from the correct + * (pageable or not) zone, since we'll get new map entries + * during the transfer. + * We'll also adjust the map entries's "start" and "end" + * during the transfer, to keep "copy"'s entries consistent + * with its "offset". + */ + goto after_adjustments; } /* @@ -5666,45 +7792,41 @@ vm_map_copyout( */ if (copy->cpy_hdr.entries_pageable != dst_map->hdr.entries_pageable) { - /* - * Mismatches occur when dealing with the default - * pager. - */ - zone_t old_zone; - vm_map_entry_t next, new; - - /* - * Find the zone that the copies were allocated from - */ - old_zone = (copy->cpy_hdr.entries_pageable) - ? vm_map_entry_zone - : vm_map_kentry_zone; - entry = vm_map_copy_first_entry(copy); - - /* - * Reinitialize the copy so that vm_map_copy_entry_link - * will work. - */ - copy->cpy_hdr.nentries = 0; - copy->cpy_hdr.entries_pageable = dst_map->hdr.entries_pageable; - vm_map_copy_first_entry(copy) = - vm_map_copy_last_entry(copy) = - vm_map_copy_to_entry(copy); - - /* - * Copy each entry. - */ - while (entry != vm_map_copy_to_entry(copy)) { - new = vm_map_copy_entry_create(copy); - vm_map_entry_copy_full(new, entry); - new->use_pmap = FALSE; /* clr address space specifics */ - vm_map_copy_entry_link(copy, - vm_map_copy_last_entry(copy), - new); - next = entry->vme_next; - zfree(old_zone, entry); - entry = next; - } + /* + * Mismatches occur when dealing with the default + * pager. + */ + zone_t old_zone; + vm_map_entry_t next, new; + + /* + * Find the zone that the copies were allocated from + */ + + entry = vm_map_copy_first_entry(copy); + + /* + * Reinitialize the copy so that vm_map_copy_entry_link + * will work. + */ + vm_map_store_copy_reset(copy, entry); + copy->cpy_hdr.entries_pageable = dst_map->hdr.entries_pageable; + + /* + * Copy each entry. + */ + while (entry != vm_map_copy_to_entry(copy)) { + new = vm_map_copy_entry_create(copy, !copy->cpy_hdr.entries_pageable); + vm_map_entry_copy_full(new, entry); + new->use_pmap = FALSE; /* clr address space specifics */ + vm_map_copy_entry_link(copy, + vm_map_copy_last_entry(copy), + new); + next = entry->vme_next; + old_zone = entry->from_reserved_zone ? vm_map_entry_reserved_zone : vm_map_entry_zone; + zfree(old_zone, entry); + entry = next; + } } /* @@ -5712,13 +7834,29 @@ vm_map_copyout( * reset the region attributes. */ - adjustment = start - vm_copy_start; for (entry = vm_map_copy_first_entry(copy); entry != vm_map_copy_to_entry(copy); entry = entry->vme_next) { + if (VM_MAP_PAGE_SHIFT(dst_map) == PAGE_SHIFT) { + /* + * We're injecting this copy entry into a map that + * has the standard page alignment, so clear + * "map_aligned" (which might have been inherited + * from the original map entry). + */ + entry->map_aligned = FALSE; + } + entry->vme_start += adjustment; entry->vme_end += adjustment; + if (entry->map_aligned) { + assert(VM_MAP_PAGE_ALIGNED(entry->vme_start, + VM_MAP_PAGE_MASK(dst_map))); + assert(VM_MAP_PAGE_ALIGNED(entry->vme_end, + VM_MAP_PAGE_MASK(dst_map))); + } + entry->inheritance = VM_INHERIT_DEFAULT; entry->protection = VM_PROT_DEFAULT; entry->max_protection = VM_PROT_ALL; @@ -5729,146 +7867,79 @@ vm_map_copyout( * map the pages into the destination map. */ if (entry->wired_count != 0) { - register vm_map_offset_t va; - vm_object_offset_t offset; - register vm_object_t object; + register vm_map_offset_t va; + vm_object_offset_t offset; + register vm_object_t object; + vm_prot_t prot; + int type_of_fault; - object = entry->object.vm_object; - offset = entry->offset; - va = entry->vme_start; + object = entry->object.vm_object; + offset = entry->offset; + va = entry->vme_start; - pmap_pageable(dst_map->pmap, - entry->vme_start, - entry->vme_end, - TRUE); + pmap_pageable(dst_map->pmap, + entry->vme_start, + entry->vme_end, + TRUE); - while (va < entry->vme_end) { - register vm_page_t m; + while (va < entry->vme_end) { + register vm_page_t m; - /* - * Look up the page in the object. - * Assert that the page will be found in the - * top object: - * either - * the object was newly created by - * vm_object_copy_slowly, and has - * copies of all of the pages from - * the source object - * or - * the object was moved from the old - * map entry; because the old map - * entry was wired, all of the pages - * were in the top-level object. - * (XXX not true if we wire pages for - * reading) - */ - vm_object_lock(object); - vm_object_paging_begin(object); + /* + * Look up the page in the object. + * Assert that the page will be found in the + * top object: + * either + * the object was newly created by + * vm_object_copy_slowly, and has + * copies of all of the pages from + * the source object + * or + * the object was moved from the old + * map entry; because the old map + * entry was wired, all of the pages + * were in the top-level object. + * (XXX not true if we wire pages for + * reading) + */ + vm_object_lock(object); - m = vm_page_lookup(object, offset); - if (m == VM_PAGE_NULL || m->wire_count == 0 || - m->absent) - panic("vm_map_copyout: wiring 0x%x", m); + m = vm_page_lookup(object, offset); + if (m == VM_PAGE_NULL || !VM_PAGE_WIRED(m) || + m->absent) + panic("vm_map_copyout: wiring %p", m); - /* - * ENCRYPTED SWAP: - * The page is assumed to be wired here, so it - * shouldn't be encrypted. Otherwise, we - * couldn't enter it in the page table, since - * we don't want the user to see the encrypted - * data. - */ - ASSERT_PAGE_DECRYPTED(m); + /* + * 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); - m->busy = TRUE; - vm_object_unlock(object); + prot = entry->protection; - PMAP_ENTER(dst_map->pmap, va, m, entry->protection, - ((unsigned int) - (m->object->wimg_bits)) - & VM_WIMG_MASK, - TRUE); - - vm_object_lock(object); - PAGE_WAKEUP_DONE(m); - /* the page is wired, so we don't have to activate */ - vm_object_paging_end(object); - vm_object_unlock(object); - - offset += PAGE_SIZE_64; - va += PAGE_SIZE; - } - } - else if (size <= vm_map_aggressive_enter_max) { - - register vm_map_offset_t va; - vm_object_offset_t offset; - register vm_object_t object; - vm_prot_t prot; + if (override_nx(dst_map, entry->alias) && prot) + prot |= VM_PROT_EXECUTE; - object = entry->object.vm_object; - if (object != VM_OBJECT_NULL) { - - offset = entry->offset; - va = entry->vme_start; - while (va < entry->vme_end) { - register vm_page_t m; - - /* - * Look up the page in the object. - * Assert that the page will be found - * in the top object if at all... - */ - vm_object_lock(object); - vm_object_paging_begin(object); + type_of_fault = DBG_CACHE_HIT_FAULT; - /* - * ENCRYPTED SWAP: - * If the page is encrypted, skip it: - * we can't let the user see the - * encrypted contents. The page will - * get decrypted on demand when the - * user generates a soft-fault when - * trying to access it. - */ - if (((m = vm_page_lookup(object, - offset)) - != VM_PAGE_NULL) && - !m->busy && !m->fictitious && - !m->encrypted && - !m->absent && !m->error) { - m->busy = TRUE; - vm_object_unlock(object); + vm_fault_enter(m, dst_map->pmap, va, prot, prot, + VM_PAGE_WIRED(m), FALSE, FALSE, FALSE, NULL, + &type_of_fault); - /* honor cow obligations */ - prot = entry->protection; - if (entry->needs_copy) - prot &= ~VM_PROT_WRITE; - - PMAP_ENTER(dst_map->pmap, va, - m, prot, - ((unsigned int) - (m->object->wimg_bits)) - & VM_WIMG_MASK, - FALSE); - - vm_object_lock(object); - vm_page_lock_queues(); - if (!m->active && !m->inactive) - vm_page_activate(m); - vm_page_unlock_queues(); - PAGE_WAKEUP_DONE(m); - } - vm_object_paging_end(object); - vm_object_unlock(object); + vm_object_unlock(object); - offset += PAGE_SIZE_64; - va += PAGE_SIZE; - } + offset += PAGE_SIZE_64; + va += PAGE_SIZE; } } } +after_adjustments: + /* * Correct the page alignment for the result */ @@ -5879,7 +7950,11 @@ vm_map_copyout( * Update the hints and the map size */ - SAVE_HINT(dst_map, vm_map_copy_last_entry(copy)); + if (consume_on_success) { + SAVE_HINT_MAP_WRITE(dst_map, vm_map_copy_last_entry(copy)); + } else { + SAVE_HINT_MAP_WRITE(dst_map, last); + } dst_map->size += size; @@ -5887,7 +7962,13 @@ vm_map_copyout( * Link in the copy */ - vm_map_copy_insert(dst_map, last, copy); + if (consume_on_success) { + vm_map_copy_insert(dst_map, last, copy); + } else { + vm_map_copy_remap(dst_map, last, copy, adjustment, + cur_protection, max_protection, + inheritance); + } vm_map_unlock(dst_map); @@ -5902,6 +7983,28 @@ vm_map_copyout( * Routine: vm_map_copyin * * Description: + * see vm_map_copyin_common. Exported via Unsupported.exports. + * + */ + +#undef vm_map_copyin + +kern_return_t +vm_map_copyin( + vm_map_t src_map, + vm_map_address_t src_addr, + vm_map_size_t len, + boolean_t src_destroy, + vm_map_copy_t *copy_result) /* OUT */ +{ + return(vm_map_copyin_common(src_map, src_addr, len, src_destroy, + FALSE, copy_result, FALSE)); +} + +/* + * Routine: vm_map_copyin_common + * + * Description: * Copy the specified region (src_addr, len) from the * source address space (src_map), possibly removing * the region from the source address space (src_destroy). @@ -5921,6 +8024,7 @@ typedef struct submap_map { vm_map_t parent_map; vm_map_offset_t base_start; vm_map_offset_t base_end; + vm_map_size_t base_len; struct submap_map *next; } submap_map_t; @@ -5947,6 +8051,7 @@ vm_map_copyin_common( */ vm_map_offset_t src_end; /* End of entire region to be * copied */ + vm_map_offset_t src_base; vm_map_t base_map = src_map; boolean_t map_share=FALSE; submap_map_t *parent_maps = NULL; @@ -5978,16 +8083,18 @@ vm_map_copyin_common( * for small regions. */ if ((len < msg_ool_size_small) && !use_maxprot) - return vm_map_copyin_kernel_buffer(src_map, src_addr, len, - src_destroy, copy_result); + return 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); - src_end = vm_map_round_page(src_end); + 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", (natural_t)src_map, src_addr, len, src_destroy, 0); + 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. @@ -5998,15 +8105,29 @@ vm_map_copyin_common( 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); + vm_map_copy_last_entry(copy) = vm_map_copy_to_entry(copy); copy->type = VM_MAP_COPY_ENTRY_LIST; copy->cpy_hdr.nentries = 0; copy->cpy_hdr.entries_pageable = TRUE; +#if 00 + copy->cpy_hdr.page_shift = src_map->hdr.page_shift; +#else + /* + * The copy entries can be broken down for a variety of reasons, + * so we can't guarantee that they will remain map-aligned... + * Will need to adjust the first copy_entry's "vme_start" and + * the last copy_entry's "vme_end" to be rounded to PAGE_MASK + * rather than the original map's alignment. + */ + copy->cpy_hdr.page_shift = PAGE_SHIFT; +#endif + + vm_map_store_init( &(copy->cpy_hdr) ); copy->offset = src_addr; copy->size = len; - new_entry = vm_map_copy_entry_create(copy); + new_entry = vm_map_copy_entry_create(copy, !copy->cpy_hdr.entries_pageable); #define RETURN(x) \ MACRO_BEGIN \ @@ -6086,7 +8207,7 @@ vm_map_copyin_common( submap_len = tmp_entry->vme_end - src_start; if(submap_len > (src_end-src_start)) submap_len = src_end-src_start; - ptr->base_start += submap_len; + ptr->base_len = submap_len; src_start -= tmp_entry->vme_start; src_start += tmp_entry->offset; @@ -6098,13 +8219,15 @@ vm_map_copyin_common( vm_map_reference(src_map); vm_map_unlock(ptr->parent_map); if (!vm_map_lookup_entry( - src_map, src_start, &tmp_entry)) + src_map, src_start, &tmp_entry)) RETURN(KERN_INVALID_ADDRESS); map_share = TRUE; if(!tmp_entry->is_sub_map) - vm_map_clip_start(src_map, tmp_entry, src_start); + vm_map_clip_start(src_map, tmp_entry, src_start); src_entry = tmp_entry; } + /* we are now in the lowest level submap... */ + if ((tmp_entry->object.vm_object != VM_OBJECT_NULL) && (tmp_entry->object.vm_object->phys_contiguous)) { /* This is not, supported for now.In future */ @@ -6123,27 +8246,28 @@ vm_map_copyin_common( * to allocate a map entry. */ if (new_entry == VM_MAP_ENTRY_NULL) { - version.main_timestamp = src_map->timestamp; - vm_map_unlock(src_map); + version.main_timestamp = src_map->timestamp; + vm_map_unlock(src_map); - new_entry = vm_map_copy_entry_create(copy); + new_entry = vm_map_copy_entry_create(copy, !copy->cpy_hdr.entries_pageable); - vm_map_lock(src_map); - if ((version.main_timestamp + 1) != src_map->timestamp) { - if (!vm_map_lookup_entry(src_map, src_start, - &tmp_entry)) { - RETURN(KERN_INVALID_ADDRESS); + vm_map_lock(src_map); + if ((version.main_timestamp + 1) != src_map->timestamp) { + if (!vm_map_lookup_entry(src_map, src_start, + &tmp_entry)) { + RETURN(KERN_INVALID_ADDRESS); + } + if (!tmp_entry->is_sub_map) + vm_map_clip_start(src_map, tmp_entry, src_start); + continue; /* restart w/ new tmp_entry */ } - vm_map_clip_start(src_map, tmp_entry, src_start); - continue; /* restart w/ new tmp_entry */ - } } /* * Verify that the region can be read. */ if (((src_entry->protection & VM_PROT_READ) == VM_PROT_NONE && - !use_maxprot) || + !use_maxprot) || (src_entry->max_protection & VM_PROT_READ) == 0) RETURN(KERN_PROTECTION_FAILURE); @@ -6167,39 +8291,39 @@ vm_map_copyin_common( if (src_destroy && (src_object == VM_OBJECT_NULL || - (src_object->internal && !src_object->true_share - && !map_share))) { - /* - * If we are destroying the source, and the object - * is internal, we can move the object reference - * from the source to the copy. The copy is - * copy-on-write only if the source is. - * We make another reference to the object, because - * destroying the source entry will deallocate it. - */ - vm_object_reference(src_object); + (src_object->internal && !src_object->true_share + && !map_share))) { + /* + * If we are destroying the source, and the object + * is internal, we can move the object reference + * from the source to the copy. The copy is + * copy-on-write only if the source is. + * We make another reference to the object, because + * destroying the source entry will deallocate it. + */ + vm_object_reference(src_object); - /* - * Copy is always unwired. vm_map_copy_entry - * set its wired count to zero. - */ + /* + * Copy is always unwired. vm_map_copy_entry + * set its wired count to zero. + */ - goto CopySuccessful; + goto CopySuccessful; } -RestartCopy: + RestartCopy: XPR(XPR_VM_MAP, "vm_map_copyin_common src_obj 0x%x ent 0x%x obj 0x%x was_wired %d\n", src_object, new_entry, new_entry->object.vm_object, was_wired, 0); if ((src_object == VM_OBJECT_NULL || - (!was_wired && !map_share && !tmp_entry->is_shared)) && - vm_object_copy_quickly( - &new_entry->object.vm_object, - src_offset, - src_size, - &src_needs_copy, - &new_entry_needs_copy)) { + (!was_wired && !map_share && !tmp_entry->is_shared)) && + vm_object_copy_quickly( + &new_entry->object.vm_object, + src_offset, + src_size, + &src_needs_copy, + &new_entry_needs_copy)) { new_entry->needs_copy = new_entry_needs_copy; @@ -6208,16 +8332,23 @@ RestartCopy: */ if (src_needs_copy && !tmp_entry->needs_copy) { + vm_prot_t prot; + + prot = src_entry->protection & ~VM_PROT_WRITE; + + if (override_nx(src_map, src_entry->alias) && prot) + prot |= VM_PROT_EXECUTE; + vm_object_pmap_protect( src_object, src_offset, src_size, (src_entry->is_shared ? - PMAP_NULL - : src_map->pmap), + PMAP_NULL + : src_map->pmap), src_entry->vme_start, - src_entry->protection & - ~VM_PROT_WRITE); + prot); + tmp_entry->needs_copy = TRUE; } @@ -6254,24 +8385,25 @@ RestartCopy: CopySlowly: vm_object_lock(src_object); result = vm_object_copy_slowly( - src_object, - src_offset, - src_size, - THREAD_UNINT, - &new_entry->object.vm_object); + src_object, + src_offset, + src_size, + THREAD_UNINT, + &new_entry->object.vm_object); new_entry->offset = 0; new_entry->needs_copy = FALSE; } else if (src_object->copy_strategy == MEMORY_OBJECT_COPY_SYMMETRIC && - (tmp_entry->is_shared || map_share)) { + (tmp_entry->is_shared || map_share)) { vm_object_t new_object; - vm_object_lock(src_object); + vm_object_lock_shared(src_object); new_object = vm_object_copy_delayed( - src_object, - src_offset, - src_size); + src_object, + src_offset, + src_size, + TRUE); if (new_object == VM_OBJECT_NULL) goto CopySlowly; @@ -6281,11 +8413,11 @@ RestartCopy: } else { result = vm_object_copy_strategically(src_object, - src_offset, - src_size, - &new_entry->object.vm_object, - &new_entry->offset, - &new_entry_needs_copy); + src_offset, + src_size, + &new_entry->object.vm_object, + &new_entry->offset, + &new_entry_needs_copy); new_entry->needs_copy = new_entry_needs_copy; } @@ -6336,8 +8468,12 @@ RestartCopy: ((src_entry->max_protection & VM_PROT_READ) == 0)) goto VerificationFailed; - if (src_entry->vme_end < new_entry->vme_end) - src_size = (new_entry->vme_end = src_entry->vme_end) - src_start; + if (src_entry->vme_end < new_entry->vme_end) { + assert(VM_MAP_PAGE_ALIGNED(src_entry->vme_end, + VM_MAP_COPY_PAGE_MASK(copy))); + new_entry->vme_end = src_entry->vme_end; + src_size = new_entry->vme_end - src_start; + } if ((src_entry->object.vm_object != src_object) || (src_entry->offset != src_offset) ) { @@ -6348,7 +8484,7 @@ RestartCopy: * Start over with this top-level entry. */ - VerificationFailed: ; + VerificationFailed: ; vm_object_deallocate(new_entry->object.vm_object); tmp_entry = src_entry; @@ -6359,7 +8495,7 @@ RestartCopy: * Verification succeeded. */ - VerificationSuccessful: ; + VerificationSuccessful: ; if (result == KERN_MEMORY_RESTART_COPY) goto RestartCopy; @@ -6368,7 +8504,7 @@ RestartCopy: * Copy succeeded. */ - CopySuccessful: ; + CopySuccessful: ; /* * Link in the new copy entry. @@ -6381,6 +8517,7 @@ RestartCopy: * Determine whether the entire region * has been copied. */ + src_base = src_start; src_start = new_entry->vme_end; new_entry = VM_MAP_ENTRY_NULL; while ((src_start >= src_end) && (src_end != 0)) { @@ -6390,15 +8527,22 @@ RestartCopy: 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_start = ptr->base_start; + 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)) + !vm_map_lookup_entry( + src_map, src_start, &tmp_entry)) RETURN(KERN_INVALID_ADDRESS); kfree(ptr, sizeof(submap_map_t)); if(parent_maps == NULL) @@ -6416,8 +8560,44 @@ RestartCopy: tmp_entry = src_entry->vme_next; if ((tmp_entry->vme_start != src_start) || - (tmp_entry == vm_map_to_entry(src_map))) + (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); + } } /* @@ -6425,26 +8605,93 @@ RestartCopy: * copy was successful. */ if (src_destroy) { - (void) vm_map_delete(src_map, - vm_map_trunc_page(src_addr), - src_end, - (src_map == kernel_map) ? - VM_MAP_REMOVE_KUNWIRE : - VM_MAP_NO_FLAGS, - VM_MAP_NULL); + (void) vm_map_delete( + src_map, + vm_map_trunc_page(src_addr, + VM_MAP_PAGE_MASK(src_map)), + src_end, + ((src_map == kernel_map) ? + VM_MAP_REMOVE_KUNWIRE : + VM_MAP_NO_FLAGS), + VM_MAP_NULL); + } else { + /* fix up the damage we did in the base map */ + vm_map_simplify_range( + src_map, + vm_map_trunc_page(src_addr, + VM_MAP_PAGE_MASK(src_map)), + vm_map_round_page(src_end, + VM_MAP_PAGE_MASK(src_map))); } vm_map_unlock(src_map); + if (VM_MAP_PAGE_SHIFT(src_map) != PAGE_SHIFT) { + 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; + adjustment = + (vm_map_trunc_page(copy->offset, + PAGE_MASK) - + vm_map_trunc_page(copy->offset, + VM_MAP_PAGE_MASK(src_map))); + if (adjustment) { + assert(page_aligned(adjustment)); + assert(adjustment < VM_MAP_PAGE_SIZE(src_map)); + tmp_entry->vme_start += adjustment; + tmp_entry->offset += adjustment; + copy_addr += adjustment; + assert(tmp_entry->vme_start < tmp_entry->vme_end); + } + } + + /* 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; + adjustment = + (vm_map_round_page((copy->offset + + copy->size), + VM_MAP_PAGE_MASK(src_map)) - + vm_map_round_page((copy->offset + + copy->size), + PAGE_MASK)); + if (adjustment) { + assert(page_aligned(adjustment)); + assert(adjustment < VM_MAP_PAGE_SIZE(src_map)); + tmp_entry->vme_end -= adjustment; + assert(tmp_entry->vme_start < tmp_entry->vme_end); + } + } + } + /* Fix-up start and end points in copy. This is necessary */ /* when the various entries in the copy object were picked */ /* up from different sub-maps */ tmp_entry = vm_map_copy_first_entry(copy); while (tmp_entry != vm_map_copy_to_entry(copy)) { + assert(VM_MAP_PAGE_ALIGNED( + copy_addr + (tmp_entry->vme_end - + tmp_entry->vme_start), + VM_MAP_COPY_PAGE_MASK(copy))); + assert(VM_MAP_PAGE_ALIGNED( + copy_addr, + VM_MAP_COPY_PAGE_MASK(copy))); + + /* + * The copy_entries will be injected directly into the + * destination map and might not be "map aligned" there... + */ + tmp_entry->map_aligned = FALSE; + tmp_entry->vme_end = copy_addr + (tmp_entry->vme_end - tmp_entry->vme_start); tmp_entry->vme_start = copy_addr; + assert(tmp_entry->vme_start < tmp_entry->vme_end); copy_addr += tmp_entry->vme_end - tmp_entry->vme_start; tmp_entry = (struct vm_map_entry *)tmp_entry->vme_next; } @@ -6455,6 +8702,78 @@ RestartCopy: #undef RETURN } +kern_return_t +vm_map_copy_extract( + vm_map_t src_map, + vm_map_address_t src_addr, + vm_map_size_t len, + vm_map_copy_t *copy_result, /* OUT */ + vm_prot_t *cur_prot, /* OUT */ + vm_prot_t *max_prot) +{ + vm_map_offset_t src_start, src_end; + vm_map_copy_t copy; + kern_return_t kr; + + /* + * Check for copies of zero bytes. + */ + + if (len == 0) { + *copy_result = VM_MAP_COPY_NULL; + return(KERN_SUCCESS); + } + + /* + * Check that the end address doesn't overflow + */ + src_end = src_addr + len; + if (src_end < src_addr) + return KERN_INVALID_ADDRESS; + + /* + * Compute (page aligned) start and end of region + */ + src_start = vm_map_trunc_page(src_addr, PAGE_MASK); + src_end = vm_map_round_page(src_end, PAGE_MASK); + + /* + * Allocate a header element for the list. + * + * Use the start and end in the header to + * remember the endpoints prior to rounding. + */ + + copy = (vm_map_copy_t) zalloc(vm_map_copy_zone); + vm_map_copy_first_entry(copy) = + vm_map_copy_last_entry(copy) = vm_map_copy_to_entry(copy); + copy->type = VM_MAP_COPY_ENTRY_LIST; + copy->cpy_hdr.nentries = 0; + copy->cpy_hdr.entries_pageable = TRUE; + + vm_map_store_init(©->cpy_hdr); + + copy->offset = 0; + copy->size = len; + + kr = vm_map_remap_extract(src_map, + src_addr, + len, + FALSE, /* copy */ + ©->cpy_hdr, + cur_prot, + max_prot, + VM_INHERIT_SHARE, + TRUE); /* pageable */ + if (kr != KERN_SUCCESS) { + vm_map_copy_discard(copy); + return kr; + } + + *copy_result = copy; + return KERN_SUCCESS; +} + /* * vm_map_copyin_object: * @@ -6508,27 +8827,27 @@ vm_map_fork_share( object = old_entry->object.vm_object; if (old_entry->is_sub_map) { assert(old_entry->wired_count == 0); -#ifndef i386 +#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, - (addr64_t)old_entry->vme_start, - (addr64_t)old_entry->vme_start, - (uint64_t)(old_entry->vme_end - old_entry->vme_start)); + (old_entry->object.sub_map)->pmap, + (addr64_t)old_entry->vme_start, + (addr64_t)old_entry->vme_start, + (uint64_t)(old_entry->vme_end - old_entry->vme_start)); if(result) panic("vm_map_fork_share: pmap_nest failed!"); } -#endif +#endif /* NO_NESTED_PMAP */ } else if (object == VM_OBJECT_NULL) { object = vm_object_allocate((vm_map_size_t)(old_entry->vme_end - - old_entry->vme_start)); + old_entry->vme_start)); old_entry->offset = 0; old_entry->object.vm_object = object; assert(!old_entry->needs_copy); } else if (object->copy_strategy != - MEMORY_OBJECT_COPY_SYMMETRIC) { + MEMORY_OBJECT_COPY_SYMMETRIC) { /* * We are already using an asymmetric @@ -6541,10 +8860,10 @@ vm_map_fork_share( else if (old_entry->needs_copy || /* case 1 */ object->shadowed || /* case 2 */ (!object->true_share && /* case 3 */ - !old_entry->is_shared && - (object->size > - (vm_map_size_t)(old_entry->vme_end - - old_entry->vme_start)))) { + !old_entry->is_shared && + (object->vo_size > + (vm_map_size_t)(old_entry->vme_end - + old_entry->vme_start)))) { /* * We need to create a shadow. @@ -6622,12 +8941,10 @@ vm_map_fork_share( * (This is a preemptive version of * case 2.) */ - - assert(!(object->shadowed && old_entry->is_shared)); vm_object_shadow(&old_entry->object.vm_object, &old_entry->offset, (vm_map_size_t) (old_entry->vme_end - - old_entry->vme_start)); + old_entry->vme_start)); /* * If we're making a shadow for other than @@ -6637,26 +8954,34 @@ vm_map_fork_share( if (!old_entry->needs_copy && (old_entry->protection & VM_PROT_WRITE)) { - if(old_map->mapped) { + vm_prot_t prot; + + prot = old_entry->protection & ~VM_PROT_WRITE; + + if (override_nx(old_map, old_entry->alias) && prot) + prot |= VM_PROT_EXECUTE; + + if (old_map->mapped_in_other_pmaps) { vm_object_pmap_protect( old_entry->object.vm_object, old_entry->offset, (old_entry->vme_end - - old_entry->vme_start), + old_entry->vme_start), PMAP_NULL, old_entry->vme_start, - old_entry->protection & ~VM_PROT_WRITE); + prot); } else { pmap_protect(old_map->pmap, - old_entry->vme_start, - old_entry->vme_end, - old_entry->protection & ~VM_PROT_WRITE); + old_entry->vme_start, + old_entry->vme_end, + prot); } } old_entry->needs_copy = FALSE; object = old_entry->object.vm_object; } + /* * If object was using a symmetric copy strategy, @@ -6673,8 +8998,7 @@ vm_map_fork_share( vm_map_unlock(old_entry->object.sub_map); } else { vm_object_lock(object); - object->ref_count++; - vm_object_res_reference(object); + vm_object_reference_locked(object); if (object->copy_strategy == MEMORY_OBJECT_COPY_SYMMETRIC) { object->copy_strategy = MEMORY_OBJECT_COPY_DELAY; } @@ -6686,7 +9010,8 @@ vm_map_fork_share( * Mark both entries as shared. */ - new_entry = vm_map_entry_create(new_map); + 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; @@ -6697,7 +9022,7 @@ vm_map_fork_share( * map. */ - vm_map_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); /* * Update the physical map @@ -6707,8 +9032,8 @@ vm_map_fork_share( /* Bill Angell pmap support goes here */ } else { pmap_copy(new_map->pmap, old_map->pmap, new_entry->vme_start, - old_entry->vme_end - old_entry->vme_start, - old_entry->vme_start); + old_entry->vme_end - old_entry->vme_start, + old_entry->vme_start); } } @@ -6772,7 +9097,15 @@ vm_map_fork_copy( if (! vm_map_lookup_entry(old_map, start, &last)) { last = last->vme_next; } else { - vm_map_clip_start(old_map, last, start); + if (last->vme_start == start) { + /* + * No need to clip here and we don't + * want to cause any unnecessary + * unnesting... + */ + } else { + vm_map_clip_start(old_map, last, start); + } } *old_entry_p = last; @@ -6790,9 +9123,10 @@ vm_map_fork_copy( */ vm_map_t vm_map_fork( + ledger_t ledger, vm_map_t old_map) { - pmap_t new_pmap = pmap_create((vm_map_size_t) 0); + pmap_t new_pmap; vm_map_t new_map; vm_map_entry_t old_entry; vm_map_size_t new_size = 0, entry_size; @@ -6800,18 +9134,27 @@ vm_map_fork( boolean_t src_needs_copy; boolean_t new_entry_needs_copy; + new_pmap = pmap_create(ledger, (vm_map_size_t) 0, +#if defined(__i386__) || defined(__x86_64__) + old_map->pmap->pm_task_map != TASK_MAP_32BIT +#else +#error Unknown architecture. +#endif + ); + 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); - + old_map->min_offset, + old_map->max_offset, + old_map->hdr.entries_pageable); + /* inherit the parent map's page size */ + vm_map_set_page_shift(new_map, VM_MAP_PAGE_SHIFT(old_map)); for ( - old_entry = vm_map_first_entry(old_map); - old_entry != vm_map_to_entry(old_map); - ) { + old_entry = vm_map_first_entry(old_map); + old_entry != vm_map_to_entry(old_map); + ) { entry_size = old_entry->vme_end - old_entry->vme_start; @@ -6835,23 +9178,23 @@ vm_map_fork( 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))) { + ((old_entry->object.vm_object != NULL) && + (old_entry->object.vm_object->true_share))) { goto slow_vm_map_fork_copy; } - new_entry = vm_map_entry_create(new_map); + 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 (! vm_object_copy_quickly( - &new_entry->object.vm_object, - old_entry->offset, - (old_entry->vme_end - - old_entry->vme_start), - &src_needs_copy, - &new_entry_needs_copy)) { + &new_entry->object.vm_object, + old_entry->offset, + (old_entry->vme_end - + old_entry->vme_start), + &src_needs_copy, + &new_entry_needs_copy)) { vm_map_entry_dispose(new_map, new_entry); goto slow_vm_map_fork_copy; } @@ -6861,17 +9204,24 @@ vm_map_fork( */ if (src_needs_copy && !old_entry->needs_copy) { + vm_prot_t prot; + + prot = old_entry->protection & ~VM_PROT_WRITE; + + if (override_nx(old_map, old_entry->alias) && prot) + prot |= VM_PROT_EXECUTE; + vm_object_pmap_protect( old_entry->object.vm_object, old_entry->offset, (old_entry->vme_end - - old_entry->vme_start), + old_entry->vme_start), ((old_entry->is_shared - || old_map->mapped) - ? PMAP_NULL : - old_map->pmap), + || old_map->mapped_in_other_pmaps) + ? PMAP_NULL : + old_map->pmap), old_entry->vme_start, - old_entry->protection & ~VM_PROT_WRITE); + prot); old_entry->needs_copy = TRUE; } @@ -6882,7 +9232,7 @@ vm_map_fork( * of the map. */ - vm_map_entry_link(new_map, vm_map_last_entry(new_map), + vm_map_store_entry_link(new_map, vm_map_last_entry(new_map), new_entry); new_size += entry_size; break; @@ -6903,6 +9253,30 @@ vm_map_fork( return(new_map); } +/* + * vm_map_exec: + * + * Setup the "new_map" with the proper execution environment according + * to the type of executable (platform, 64bit, chroot environment). + * Map the comm page and shared region, etc... + */ +kern_return_t +vm_map_exec( + vm_map_t new_map, + task_t task, + void *fsroot, + cpu_type_t cpu) +{ + SHARED_REGION_TRACE_DEBUG( + ("shared_region: task %p: vm_map_exec(%p,%p,%p,0x%x): ->\n", + 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_TRACE_DEBUG( + ("shared_region: task %p: vm_map_exec(%p,%p,%p,0x%x): <-\n", + current_task(), new_map, task, fsroot, cpu)); + return KERN_SUCCESS; +} /* * vm_map_lookup_locked: @@ -6931,16 +9305,15 @@ vm_map_fork( kern_return_t vm_map_lookup_locked( vm_map_t *var_map, /* IN/OUT */ - vm_map_offset_t vaddr, + vm_map_offset_t vaddr, vm_prot_t fault_type, + int object_lock_type, vm_map_version_t *out_version, /* OUT */ vm_object_t *object, /* OUT */ vm_object_offset_t *offset, /* OUT */ vm_prot_t *out_prot, /* OUT */ boolean_t *wired, /* OUT */ - int *behavior, /* OUT */ - vm_map_offset_t *lo_offset, /* OUT */ - vm_map_offset_t *hi_offset, /* OUT */ + vm_object_fault_info_t fault_info, /* OUT */ vm_map_t *real_map) { vm_map_entry_t entry; @@ -6951,18 +9324,28 @@ vm_map_lookup_locked( vm_map_offset_t old_start = 0; vm_map_offset_t old_end = 0; register vm_prot_t prot; + boolean_t mask_protections; + vm_prot_t original_fault_type; + + /* + * VM_PROT_MASK means that the caller wants us to use "fault_type" + * as a mask against the mapping's actual protections, not as an + * absolute value. + */ + mask_protections = (fault_type & VM_PROT_IS_MASK) ? TRUE : FALSE; + fault_type &= ~VM_PROT_IS_MASK; + original_fault_type = fault_type; *real_map = map; - RetryLookup: ; + +RetryLookup: + fault_type = original_fault_type; /* * If the map has an interesting hint, try it before calling * full blown lookup routine. */ - - mutex_lock(&map->s_lock); entry = map->hint; - mutex_unlock(&map->s_lock); if ((entry == vm_map_to_entry(map)) || (vaddr < entry->vme_start) || (vaddr >= entry->vme_end)) { @@ -6976,7 +9359,7 @@ vm_map_lookup_locked( if((cow_sub_map_parent) && (cow_sub_map_parent != map)) vm_map_unlock(cow_sub_map_parent); if((*real_map != map) - && (*real_map != cow_sub_map_parent)) + && (*real_map != cow_sub_map_parent)) vm_map_unlock(*real_map); return KERN_INVALID_ADDRESS; } @@ -7003,23 +9386,23 @@ submap_recurse: local_vaddr = vaddr; - if ((!entry->needs_copy) && (entry->use_pmap)) { + if ((entry->use_pmap && !(fault_type & VM_PROT_WRITE))) { /* if real_map equals map we unlock below */ if ((*real_map != map) && - (*real_map != cow_sub_map_parent)) + (*real_map != cow_sub_map_parent)) vm_map_unlock(*real_map); *real_map = entry->object.sub_map; } - if(entry->needs_copy) { + if(entry->needs_copy && (fault_type & VM_PROT_WRITE)) { if (!mapped_needs_copy) { if (vm_map_lock_read_to_write(map)) { vm_map_lock_read(map); - if(*real_map == entry->object.sub_map) - *real_map = map; + *real_map = map; goto RetryLookup; } vm_map_lock_read(entry->object.sub_map); + *var_map = entry->object.sub_map; cow_sub_map_parent = map; /* reset base to map before cow object */ /* this is the map which will accept */ @@ -7030,12 +9413,14 @@ submap_recurse: mapped_needs_copy = TRUE; } else { vm_map_lock_read(entry->object.sub_map); + *var_map = entry->object.sub_map; if((cow_sub_map_parent != map) && - (*real_map != map)) + (*real_map != map)) vm_map_unlock(map); } } else { vm_map_lock_read(entry->object.sub_map); + *var_map = entry->object.sub_map; /* leave map locked if it is a target */ /* cow sub_map above otherwise, just */ /* follow the maps down to the object */ @@ -7045,23 +9430,24 @@ submap_recurse: vm_map_unlock_read(map); } - *var_map = map = entry->object.sub_map; + map = *var_map; /* calculate the offset in the submap for vaddr */ local_vaddr = (local_vaddr - entry->vme_start) + entry->offset; -RetrySubMap: + RetrySubMap: if(!vm_map_lookup_entry(map, local_vaddr, &submap_entry)) { if((cow_sub_map_parent) && (cow_sub_map_parent != map)){ vm_map_unlock(cow_sub_map_parent); } if((*real_map != map) - && (*real_map != cow_sub_map_parent)) { + && (*real_map != cow_sub_map_parent)) { vm_map_unlock(*real_map); } *real_map = map; return KERN_INVALID_ADDRESS; } + /* find the attenuated shadow of the underlying object */ /* on our target map */ @@ -7073,14 +9459,14 @@ RetrySubMap: /* 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; + submap_entry->vme_start - entry->offset : 0; end_delta = - (entry->offset + start_delta + (old_end - old_start)) <= + (entry->offset + start_delta + (old_end - old_start)) <= submap_entry->vme_end ? - 0 : (entry->offset + - (old_end - old_start)) - - submap_entry->vme_end; + 0 : (entry->offset + + (old_end - old_start)) + - submap_entry->vme_end; old_start += start_delta; old_end -= end_delta; @@ -7093,7 +9479,8 @@ RetrySubMap: if(((fault_type & VM_PROT_WRITE) && cow_sub_map_parent)) { - vm_object_t copy_object; + vm_object_t sub_object, copy_object; + vm_object_offset_t copy_offset; vm_map_offset_t local_start; vm_map_offset_t local_end; boolean_t copied_slowly = FALSE; @@ -7106,20 +9493,24 @@ RetrySubMap: } - if (submap_entry->object.vm_object == VM_OBJECT_NULL) { - submap_entry->object.vm_object = + sub_object = submap_entry->object.vm_object; + 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->vme_end - + submap_entry->vme_start)); + submap_entry->object.vm_object = sub_object; submap_entry->offset = 0; } local_start = local_vaddr - - (cow_parent_vaddr - old_start); + (cow_parent_vaddr - old_start); local_end = local_vaddr + - (old_end - cow_parent_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); /* This is the COW case, lets connect */ /* an entry in our space to the underlying */ @@ -7127,37 +9518,48 @@ RetrySubMap: /* submap. */ - if(submap_entry->wired_count != 0) { - vm_object_lock( - submap_entry->object.vm_object); - vm_object_copy_slowly( - submap_entry->object.vm_object, - submap_entry->offset, - submap_entry->vme_end - - submap_entry->vme_start, - FALSE, - ©_object); - copied_slowly = TRUE; + if(submap_entry->wired_count != 0 || + (sub_object->copy_strategy == + MEMORY_OBJECT_COPY_NONE)) { + vm_object_lock(sub_object); + vm_object_copy_slowly(sub_object, + submap_entry->offset, + (submap_entry->vme_end - + submap_entry->vme_start), + FALSE, + ©_object); + copied_slowly = TRUE; } else { /* set up shadow object */ - copy_object = submap_entry->object.vm_object; + copy_object = sub_object; vm_object_reference(copy_object); - submap_entry->object.vm_object->shadowed = TRUE; + sub_object->shadowed = TRUE; submap_entry->needs_copy = TRUE; + + prot = submap_entry->protection & ~VM_PROT_WRITE; + + if (override_nx(old_map, submap_entry->alias) && prot) + prot |= VM_PROT_EXECUTE; + vm_object_pmap_protect( - submap_entry->object.vm_object, + sub_object, submap_entry->offset, submap_entry->vme_end - - submap_entry->vme_start, + submap_entry->vme_start, (submap_entry->is_shared - || map->mapped) ? - PMAP_NULL : map->pmap, + || map->mapped_in_other_pmaps) ? + PMAP_NULL : map->pmap, submap_entry->vme_start, - submap_entry->protection & - ~VM_PROT_WRITE); + prot); } + /* + * Adjust the fault offset to the submap entry. + */ + copy_offset = (local_vaddr - + submap_entry->vme_start + + submap_entry->offset); /* This works diffently than the */ /* normal submap case. We go back */ @@ -7175,20 +9577,44 @@ RetrySubMap: cow_sub_map_parent = NULL; if(!vm_map_lookup_entry(map, - vaddr, &entry)) { - vm_object_deallocate( - copy_object); - vm_map_lock_write_to_read(map); - return KERN_INVALID_ADDRESS; + vaddr, &entry)) { + vm_object_deallocate( + copy_object); + vm_map_lock_write_to_read(map); + return KERN_INVALID_ADDRESS; } /* clip out the portion of space */ /* mapped by the sub map which */ /* corresponds to the underlying */ /* object */ + + /* + * Clip (and unnest) the smallest nested chunk + * possible around the faulting address... + */ + local_start = vaddr & ~(pmap_nesting_size_min - 1); + local_end = local_start + pmap_nesting_size_min; + /* + * ... but don't go beyond the "old_start" to "old_end" + * range, to avoid spanning over another VM region + * with a possibly different VM object and/or offset. + */ + if (local_start < old_start) { + local_start = old_start; + } + if (local_end > old_end) { + local_end = old_end; + } + /* + * Adjust copy_offset to the start of the range. + */ + copy_offset -= (vaddr - local_start); + vm_map_clip_start(map, entry, local_start); vm_map_clip_end(map, entry, local_end); - + /* unnesting was done in vm_map_clip_start/end() */ + assert(!entry->use_pmap); /* substitute copy object for */ /* shared map entry */ @@ -7196,14 +9622,16 @@ RetrySubMap: entry->is_sub_map = FALSE; entry->object.vm_object = copy_object; - entry->protection |= VM_PROT_WRITE; - entry->max_protection |= VM_PROT_WRITE; + /* propagate the submap entry's protections */ + entry->protection |= submap_entry->protection; + entry->max_protection |= submap_entry->max_protection; + if(copied_slowly) { - entry->offset = 0; + entry->offset = local_start - old_start; entry->needs_copy = FALSE; entry->is_shared = FALSE; } else { - entry->offset = submap_entry->offset; + entry->offset = copy_offset; entry->needs_copy = TRUE; if(entry->inheritance == VM_INHERIT_SHARE) entry->inheritance = VM_INHERIT_COPY; @@ -7216,8 +9644,8 @@ RetrySubMap: vm_map_lock_write_to_read(map); } else { if((cow_sub_map_parent) - && (cow_sub_map_parent != *real_map) - && (cow_sub_map_parent != map)) { + && (cow_sub_map_parent != *real_map) + && (cow_sub_map_parent != map)) { vm_map_unlock(cow_sub_map_parent); } entry = submap_entry; @@ -7231,12 +9659,32 @@ RetrySubMap: */ prot = entry->protection; + + if (override_nx(old_map, entry->alias) && prot) { + /* + * HACK -- if not a stack, then allow execution + */ + prot |= VM_PROT_EXECUTE; + } + + if (mask_protections) { + fault_type &= prot; + if (fault_type == VM_PROT_NONE) { + goto protection_failure; + } + } if ((fault_type & (prot)) != fault_type) { - if (*real_map != map) { - vm_map_unlock(*real_map); - } - *real_map = map; - return KERN_PROTECTION_FAILURE; + protection_failure: + if (*real_map != map) { + vm_map_unlock(*real_map); + } + *real_map = map; + + if ((fault_type & VM_PROT_EXECUTE) && prot) + log_stack_execution_failure((addr64_t)vaddr, prot); + + DTRACE_VM2(prot_fault, int, 1, (uint64_t *), NULL); + return KERN_PROTECTION_FAILURE; } /* @@ -7246,7 +9694,7 @@ RetrySubMap: *wired = (entry->wired_count != 0); if (*wired) - prot = fault_type = entry->protection; + fault_type = prot; /* * If the entry was copy-on-write, we either ... @@ -7276,7 +9724,7 @@ RetrySubMap: vm_object_shadow(&entry->object.vm_object, &entry->offset, (vm_map_size_t) (entry->vme_end - - entry->vme_start)); + entry->vme_start)); entry->object.vm_object->shadowed = TRUE; entry->needs_copy = FALSE; @@ -7317,16 +9765,31 @@ RetrySubMap: *offset = (vaddr - entry->vme_start) + entry->offset; *object = entry->object.vm_object; *out_prot = prot; - *behavior = entry->behavior; - *lo_offset = entry->offset; - *hi_offset = (entry->vme_end - entry->vme_start) + entry->offset; + + 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->behavior = entry->behavior; + fault_info->lo_offset = entry->offset; + fault_info->hi_offset = (entry->vme_end - entry->vme_start) + entry->offset; + 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; + fault_info->mark_zf_absent = FALSE; + fault_info->batch_pmap_op = FALSE; + } /* * Lock the object to prevent it from disappearing */ - - vm_object_lock(*object); - + if (object_lock_type == OBJECT_LOCK_EXCLUSIVE) + vm_object_lock(*object); + else + vm_object_lock_shared(*object); + /* * Save the version number */ @@ -7388,6 +9851,7 @@ vm_map_region_recurse_64( vm_region_submap_info_64_t submap_info, /* IN/OUT */ mach_msg_type_number_t *count) /* IN/OUT */ { + mach_msg_type_number_t original_count; vm_region_extended_info_data_t extended; vm_map_entry_t tmp_entry; vm_map_offset_t user_address; @@ -7397,20 +9861,26 @@ vm_map_region_recurse_64( * "curr_entry" is the VM map entry preceding or including the * address we're looking for. * "curr_map" is the map or sub-map containing "curr_entry". + * "curr_address" is the equivalent of the top map's "user_address" + * in the current map. * "curr_offset" is the cumulated offset of "curr_map" in the * target task's address space. * "curr_depth" is the depth of "curr_map" in the chain of * sub-maps. - * "curr_max_offset" is the maximum offset we should take into - * account in the current map. It may be smaller than the current - * map's "max_offset" because we might not have mapped it all in - * the upper level map. + * + * "curr_max_below" and "curr_max_above" limit the range (around + * "curr_address") we should take into account in the current (sub)map. + * They limit the range to what's visible through the map entries + * we've traversed from the top map to the current map. + */ vm_map_entry_t curr_entry; + vm_map_address_t curr_address; vm_map_offset_t curr_offset; vm_map_t curr_map; unsigned int curr_depth; - vm_map_offset_t curr_max_offset; + vm_map_offset_t curr_max_below, curr_max_above; + vm_map_offset_t curr_skip; /* * "next_" is the same as "curr_" but for the VM region immediately @@ -7420,36 +9890,66 @@ vm_map_region_recurse_64( */ vm_map_entry_t next_entry; vm_map_offset_t next_offset; + vm_map_offset_t next_address; vm_map_t next_map; unsigned int next_depth; - vm_map_offset_t next_max_offset; + vm_map_offset_t next_max_below, next_max_above; + vm_map_offset_t next_skip; + + boolean_t look_for_pages; + vm_region_submap_short_info_64_t short_info; if (map == VM_MAP_NULL) { /* no address space to work on */ return KERN_INVALID_ARGUMENT; } - if (*count < VM_REGION_SUBMAP_INFO_COUNT_64) { - /* "info" structure is not big enough and would overflow */ + + if (*count < VM_REGION_SUBMAP_SHORT_INFO_COUNT_64) { + /* + * "info" structure is not big enough and + * would overflow + */ return KERN_INVALID_ARGUMENT; } - - *count = VM_REGION_SUBMAP_INFO_COUNT_64; - + + original_count = *count; + + if (original_count < VM_REGION_SUBMAP_INFO_V0_COUNT_64) { + *count = VM_REGION_SUBMAP_SHORT_INFO_COUNT_64; + look_for_pages = FALSE; + short_info = (vm_region_submap_short_info_64_t) submap_info; + submap_info = NULL; + } else { + look_for_pages = TRUE; + *count = VM_REGION_SUBMAP_INFO_V0_COUNT_64; + short_info = NULL; + + if (original_count >= VM_REGION_SUBMAP_INFO_V1_COUNT_64) { + *count = VM_REGION_SUBMAP_INFO_V1_COUNT_64; + } + } + user_address = *address; user_max_depth = *nesting_depth; curr_entry = NULL; curr_map = map; + curr_address = user_address; curr_offset = 0; + curr_skip = 0; curr_depth = 0; - curr_max_offset = curr_map->max_offset; + curr_max_above = ((vm_map_offset_t) -1) - curr_address; + curr_max_below = curr_address; next_entry = NULL; next_map = NULL; + next_address = 0; next_offset = 0; + next_skip = 0; next_depth = 0; - next_max_offset = curr_max_offset; + 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); @@ -7457,19 +9957,22 @@ vm_map_region_recurse_64( for (;;) { if (vm_map_lookup_entry(curr_map, - user_address - curr_offset, + curr_address, &tmp_entry)) { /* tmp_entry contains the address we're looking for */ curr_entry = tmp_entry; } else { + vm_map_offset_t skip; /* * The address is not mapped. "tmp_entry" is the * map entry preceding the address. We want the next * one, if it exists. */ curr_entry = tmp_entry->vme_next; + if (curr_entry == vm_map_to_entry(curr_map) || - curr_entry->vme_start >= curr_max_offset) { + (curr_entry->vme_start >= + curr_address + curr_max_above)) { /* no next entry at this level: stop looking */ if (not_in_kdp) { vm_map_unlock_read(curr_map); @@ -7478,9 +9981,18 @@ vm_map_region_recurse_64( curr_map = NULL; curr_offset = 0; curr_depth = 0; - curr_max_offset = 0; + curr_max_above = 0; + curr_max_below = 0; break; } + + /* adjust current address and offset */ + skip = curr_entry->vme_start - curr_address; + curr_address = curr_entry->vme_start; + curr_skip = skip; + curr_offset += skip; + curr_max_above -= skip; + curr_max_below = 0; } /* @@ -7491,7 +10003,8 @@ vm_map_region_recurse_64( tmp_entry = curr_entry->vme_next; if (tmp_entry == vm_map_to_entry(curr_map)) { /* no next entry at this level */ - } else if (tmp_entry->vme_start >= curr_max_offset) { + } else if (tmp_entry->vme_start >= + curr_address + curr_max_above) { /* * tmp_entry is beyond the scope of what we mapped of * this submap in the upper level: ignore it. @@ -7512,11 +10025,32 @@ vm_map_region_recurse_64( } next_entry = tmp_entry; next_map = curr_map; - next_offset = curr_offset; next_depth = curr_depth; - next_max_offset = curr_max_offset; + next_address = next_entry->vme_start; + next_skip = curr_skip; + next_offset = curr_offset; + next_offset += (next_address - curr_address); + next_max_above = MIN(next_max_above, curr_max_above); + next_max_above = MIN(next_max_above, + next_entry->vme_end - next_address); + next_max_below = MIN(next_max_below, curr_max_below); + next_max_below = MIN(next_max_below, + next_address - next_entry->vme_start); } + /* + * "curr_max_{above,below}" allow us to keep track of the + * portion of the submap that is actually mapped at this level: + * the rest of that submap is irrelevant to us, since it's not + * mapped here. + * The relevant portion of the map starts at + * "curr_entry->offset" up to the size of "curr_entry". + */ + curr_max_above = MIN(curr_max_above, + curr_entry->vme_end - curr_address); + curr_max_below = MIN(curr_max_below, + curr_address - curr_entry->vme_start); + if (!curr_entry->is_sub_map || curr_depth >= user_max_depth) { /* @@ -7543,7 +10077,8 @@ vm_map_region_recurse_64( /* keep "next_map" locked in case we need it */ } else { /* release this map */ - vm_map_unlock_read(curr_map); + if (not_in_kdp) + vm_map_unlock_read(curr_map); } /* @@ -7556,21 +10091,11 @@ vm_map_region_recurse_64( * space (i.e. the top-level VM map). */ curr_offset += - (curr_entry->vme_start - curr_entry->offset); + (curr_entry->offset - curr_entry->vme_start); + curr_address = user_address + curr_offset; /* switch to the submap */ curr_map = curr_entry->object.sub_map; curr_depth++; - /* - * "curr_max_offset" allows us to keep track of the - * portion of the submap that is actually mapped at this level: - * the rest of that submap is irrelevant to us, since it's not - * mapped here. - * The relevant portion of the map starts at - * "curr_entry->offset" up to the size of "curr_entry". - */ - curr_max_offset = - curr_entry->vme_end - curr_entry->vme_start + - curr_entry->offset; curr_entry = NULL; } @@ -7583,9 +10108,15 @@ vm_map_region_recurse_64( /* ... gather info about the next VM region */ curr_entry = next_entry; curr_map = next_map; /* still locked ... */ + curr_address = next_address; + curr_skip = next_skip; curr_offset = next_offset; curr_depth = next_depth; - curr_max_offset = next_max_offset; + 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) { @@ -7598,66 +10129,104 @@ vm_map_region_recurse_64( next_entry = NULL; next_map = NULL; next_offset = 0; + next_skip = 0; next_depth = 0; - next_max_offset = 0; + next_max_below = -1; + next_max_above = -1; *nesting_depth = curr_depth; - *size = curr_entry->vme_end - curr_entry->vme_start; - *address = curr_entry->vme_start + curr_offset; - - submap_info->user_tag = curr_entry->alias; - submap_info->offset = curr_entry->offset; - 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 = (uint32_t) curr_entry->object.vm_object; + *size = curr_max_above + curr_max_below; + *address = user_address + curr_skip - curr_max_below; + +// LP64todo: all the current tools are 32bit, obviously never worked for 64b +// so probably should be a real 32b ID vs. ptr. +// Current users just check for equality +#define INFO_MAKE_OBJECT_ID(p) ((uint32_t)(uintptr_t)VM_KERNEL_ADDRPERM(p)) + + if (look_for_pages) { + submap_info->user_tag = curr_entry->alias; + submap_info->offset = curr_entry->offset; + 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); + } else { + short_info->user_tag = curr_entry->alias; + short_info->offset = curr_entry->offset; + 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); + } extended.pages_resident = 0; extended.pages_swapped_out = 0; extended.pages_shared_now_private = 0; extended.pages_dirtied = 0; + extended.pages_reusable = 0; extended.external_pager = 0; extended.shadow_depth = 0; if (not_in_kdp) { if (!curr_entry->is_sub_map) { + vm_map_offset_t range_start, range_end; + range_start = MAX((curr_address - curr_max_below), + curr_entry->vme_start); + range_end = MIN((curr_address + curr_max_above), + curr_entry->vme_end); vm_map_region_walk(curr_map, - curr_entry->vme_start, + range_start, curr_entry, - curr_entry->offset, - (curr_entry->vme_end - - curr_entry->vme_start), - &extended); - submap_info->share_mode = extended.share_mode; + (curr_entry->offset + + (range_start - + curr_entry->vme_start)), + range_end - range_start, + &extended, + look_for_pages, VM_REGION_EXTENDED_INFO_COUNT); if (extended.external_pager && extended.ref_count == 2 && extended.share_mode == SM_SHARED) { - submap_info->share_mode = SM_PRIVATE; + extended.share_mode = SM_PRIVATE; } - submap_info->ref_count = extended.ref_count; } else { if (curr_entry->use_pmap) { - submap_info->share_mode = SM_TRUESHARED; + extended.share_mode = SM_TRUESHARED; } else { - submap_info->share_mode = SM_PRIVATE; + extended.share_mode = SM_PRIVATE; } - submap_info->ref_count = + extended.ref_count = curr_entry->object.sub_map->ref_count; } } - submap_info->pages_resident = extended.pages_resident; - submap_info->pages_swapped_out = extended.pages_swapped_out; - submap_info->pages_shared_now_private = - extended.pages_shared_now_private; - submap_info->pages_dirtied = extended.pages_dirtied; - submap_info->external_pager = extended.external_pager; - submap_info->shadow_depth = extended.shadow_depth; - - if (not_in_kdp) { + if (look_for_pages) { + submap_info->pages_resident = extended.pages_resident; + submap_info->pages_swapped_out = extended.pages_swapped_out; + submap_info->pages_shared_now_private = + extended.pages_shared_now_private; + submap_info->pages_dirtied = extended.pages_dirtied; + submap_info->external_pager = extended.external_pager; + submap_info->shadow_depth = extended.shadow_depth; + submap_info->share_mode = extended.share_mode; + submap_info->ref_count = extended.ref_count; + + if (original_count >= VM_REGION_SUBMAP_INFO_V1_COUNT_64) { + submap_info->pages_reusable = extended.pages_reusable; + } + } else { + short_info->external_pager = extended.external_pager; + short_info->shadow_depth = extended.shadow_depth; + short_info->share_mode = extended.share_mode; + short_info->ref_count = extended.ref_count; + } + + if (not_in_kdp) { vm_map_unlock_read(curr_map); } @@ -7696,253 +10265,293 @@ vm_map_region( switch (flavor) { case VM_REGION_BASIC_INFO: - /* legacy for old 32-bit objects info */ + /* legacy for old 32-bit objects info */ { - vm_region_basic_info_t basic; + vm_region_basic_info_t basic; - if (*count < VM_REGION_BASIC_INFO_COUNT) - return(KERN_INVALID_ARGUMENT); + if (*count < VM_REGION_BASIC_INFO_COUNT) + return(KERN_INVALID_ARGUMENT); - basic = (vm_region_basic_info_t) info; - *count = VM_REGION_BASIC_INFO_COUNT; + basic = (vm_region_basic_info_t) info; + *count = VM_REGION_BASIC_INFO_COUNT; - vm_map_lock_read(map); + vm_map_lock_read(map); - start = *address; - if (!vm_map_lookup_entry(map, start, &tmp_entry)) { - if ((entry = tmp_entry->vme_next) == vm_map_to_entry(map)) { - vm_map_unlock_read(map); - return(KERN_INVALID_ADDRESS); + start = *address; + if (!vm_map_lookup_entry(map, start, &tmp_entry)) { + if ((entry = tmp_entry->vme_next) == vm_map_to_entry(map)) { + vm_map_unlock_read(map); + return(KERN_INVALID_ADDRESS); + } + } else { + entry = tmp_entry; } - } else { - entry = tmp_entry; - } - start = entry->vme_start; + start = entry->vme_start; - basic->offset = (uint32_t)entry->offset; - basic->protection = entry->protection; - basic->inheritance = entry->inheritance; - basic->max_protection = entry->max_protection; - basic->behavior = entry->behavior; - basic->user_wired_count = entry->user_wired_count; - basic->reserved = entry->is_sub_map; - *address = start; - *size = (entry->vme_end - start); + basic->offset = (uint32_t)entry->offset; + basic->protection = entry->protection; + basic->inheritance = entry->inheritance; + basic->max_protection = entry->max_protection; + basic->behavior = entry->behavior; + basic->user_wired_count = entry->user_wired_count; + basic->reserved = entry->is_sub_map; + *address = start; + *size = (entry->vme_end - start); - if (object_name) *object_name = IP_NULL; - if (entry->is_sub_map) { - basic->shared = FALSE; - } else { - basic->shared = entry->is_shared; - } + if (object_name) *object_name = IP_NULL; + if (entry->is_sub_map) { + basic->shared = FALSE; + } else { + basic->shared = entry->is_shared; + } - vm_map_unlock_read(map); - return(KERN_SUCCESS); + vm_map_unlock_read(map); + return(KERN_SUCCESS); } case VM_REGION_BASIC_INFO_64: { - vm_region_basic_info_64_t basic; + vm_region_basic_info_64_t basic; - if (*count < VM_REGION_BASIC_INFO_COUNT_64) - return(KERN_INVALID_ARGUMENT); + if (*count < VM_REGION_BASIC_INFO_COUNT_64) + return(KERN_INVALID_ARGUMENT); - basic = (vm_region_basic_info_64_t) info; - *count = VM_REGION_BASIC_INFO_COUNT_64; + basic = (vm_region_basic_info_64_t) info; + *count = VM_REGION_BASIC_INFO_COUNT_64; - vm_map_lock_read(map); + vm_map_lock_read(map); - start = *address; - if (!vm_map_lookup_entry(map, start, &tmp_entry)) { - if ((entry = tmp_entry->vme_next) == vm_map_to_entry(map)) { - vm_map_unlock_read(map); - return(KERN_INVALID_ADDRESS); + start = *address; + if (!vm_map_lookup_entry(map, start, &tmp_entry)) { + if ((entry = tmp_entry->vme_next) == vm_map_to_entry(map)) { + vm_map_unlock_read(map); + return(KERN_INVALID_ADDRESS); + } + } else { + entry = tmp_entry; } - } else { - entry = tmp_entry; - } - - start = entry->vme_start; - - basic->offset = entry->offset; - basic->protection = entry->protection; - basic->inheritance = entry->inheritance; - basic->max_protection = entry->max_protection; - basic->behavior = entry->behavior; - basic->user_wired_count = entry->user_wired_count; - basic->reserved = entry->is_sub_map; - *address = start; - *size = (entry->vme_end - start); - - if (object_name) *object_name = IP_NULL; - if (entry->is_sub_map) { - basic->shared = FALSE; - } else { - basic->shared = entry->is_shared; - } - - vm_map_unlock_read(map); - return(KERN_SUCCESS); + + start = entry->vme_start; + + basic->offset = entry->offset; + basic->protection = entry->protection; + basic->inheritance = entry->inheritance; + basic->max_protection = entry->max_protection; + basic->behavior = entry->behavior; + basic->user_wired_count = entry->user_wired_count; + basic->reserved = entry->is_sub_map; + *address = start; + *size = (entry->vme_end - start); + + if (object_name) *object_name = IP_NULL; + if (entry->is_sub_map) { + basic->shared = FALSE; + } else { + basic->shared = entry->is_shared; + } + + vm_map_unlock_read(map); + return(KERN_SUCCESS); } case VM_REGION_EXTENDED_INFO: - { - vm_region_extended_info_t extended; + if (*count < VM_REGION_EXTENDED_INFO_COUNT) + return(KERN_INVALID_ARGUMENT); + /*fallthru*/ + case VM_REGION_EXTENDED_INFO__legacy: + if (*count < VM_REGION_EXTENDED_INFO_COUNT__legacy) + return KERN_INVALID_ARGUMENT; - if (*count < VM_REGION_EXTENDED_INFO_COUNT) - return(KERN_INVALID_ARGUMENT); + { + vm_region_extended_info_t extended; + mach_msg_type_number_t original_count; - extended = (vm_region_extended_info_t) info; - *count = VM_REGION_EXTENDED_INFO_COUNT; + extended = (vm_region_extended_info_t) info; - vm_map_lock_read(map); + vm_map_lock_read(map); - start = *address; - if (!vm_map_lookup_entry(map, start, &tmp_entry)) { - if ((entry = tmp_entry->vme_next) == vm_map_to_entry(map)) { - vm_map_unlock_read(map); - return(KERN_INVALID_ADDRESS); + start = *address; + if (!vm_map_lookup_entry(map, start, &tmp_entry)) { + if ((entry = tmp_entry->vme_next) == vm_map_to_entry(map)) { + vm_map_unlock_read(map); + return(KERN_INVALID_ADDRESS); + } + } else { + entry = tmp_entry; } - } else { - entry = tmp_entry; - } - start = entry->vme_start; + start = entry->vme_start; - extended->protection = entry->protection; - extended->user_tag = entry->alias; - extended->pages_resident = 0; - extended->pages_swapped_out = 0; - extended->pages_shared_now_private = 0; - extended->pages_dirtied = 0; - extended->external_pager = 0; - extended->shadow_depth = 0; + extended->protection = entry->protection; + extended->user_tag = entry->alias; + extended->pages_resident = 0; + extended->pages_swapped_out = 0; + extended->pages_shared_now_private = 0; + extended->pages_dirtied = 0; + extended->external_pager = 0; + extended->shadow_depth = 0; + + original_count = *count; + if (flavor == VM_REGION_EXTENDED_INFO__legacy) { + *count = VM_REGION_EXTENDED_INFO_COUNT__legacy; + } else { + extended->pages_reusable = 0; + *count = VM_REGION_EXTENDED_INFO_COUNT; + } - vm_map_region_walk(map, start, entry, entry->offset, entry->vme_end - start, extended); + vm_map_region_walk(map, start, entry, entry->offset, entry->vme_end - start, extended, TRUE, *count); - if (extended->external_pager && extended->ref_count == 2 && extended->share_mode == SM_SHARED) - extended->share_mode = SM_PRIVATE; + if (extended->external_pager && extended->ref_count == 2 && extended->share_mode == SM_SHARED) + extended->share_mode = SM_PRIVATE; - if (object_name) - *object_name = IP_NULL; - *address = start; - *size = (entry->vme_end - start); + if (object_name) + *object_name = IP_NULL; + *address = start; + *size = (entry->vme_end - start); - vm_map_unlock_read(map); - return(KERN_SUCCESS); + vm_map_unlock_read(map); + return(KERN_SUCCESS); } case VM_REGION_TOP_INFO: { - vm_region_top_info_t top; + vm_region_top_info_t top; - if (*count < VM_REGION_TOP_INFO_COUNT) - return(KERN_INVALID_ARGUMENT); + if (*count < VM_REGION_TOP_INFO_COUNT) + return(KERN_INVALID_ARGUMENT); - top = (vm_region_top_info_t) info; - *count = VM_REGION_TOP_INFO_COUNT; + top = (vm_region_top_info_t) info; + *count = VM_REGION_TOP_INFO_COUNT; - vm_map_lock_read(map); + vm_map_lock_read(map); - start = *address; - if (!vm_map_lookup_entry(map, start, &tmp_entry)) { - if ((entry = tmp_entry->vme_next) == vm_map_to_entry(map)) { - vm_map_unlock_read(map); - return(KERN_INVALID_ADDRESS); - } - } else { - entry = tmp_entry; + start = *address; + if (!vm_map_lookup_entry(map, start, &tmp_entry)) { + if ((entry = tmp_entry->vme_next) == vm_map_to_entry(map)) { + vm_map_unlock_read(map); + return(KERN_INVALID_ADDRESS); + } + } else { + entry = tmp_entry; - } - start = entry->vme_start; + } + start = entry->vme_start; - top->private_pages_resident = 0; - top->shared_pages_resident = 0; + top->private_pages_resident = 0; + top->shared_pages_resident = 0; - vm_map_region_top_walk(entry, top); + vm_map_region_top_walk(entry, top); - if (object_name) - *object_name = IP_NULL; - *address = start; - *size = (entry->vme_end - start); + if (object_name) + *object_name = IP_NULL; + *address = start; + *size = (entry->vme_end - start); - vm_map_unlock_read(map); - return(KERN_SUCCESS); + vm_map_unlock_read(map); + return(KERN_SUCCESS); } default: - return(KERN_INVALID_ARGUMENT); + return(KERN_INVALID_ARGUMENT); } } -static void +#define OBJ_RESIDENT_COUNT(obj, entry_size) \ + MIN((entry_size), \ + ((obj)->all_reusable ? \ + (obj)->wired_page_count : \ + (obj)->resident_page_count - (obj)->reusable_page_count)) + +void vm_map_region_top_walk( vm_map_entry_t entry, vm_region_top_info_t top) { - register struct vm_object *obj, *tmp_obj; - register int ref_count; if (entry->object.vm_object == 0 || entry->is_sub_map) { - top->share_mode = SM_EMPTY; - top->ref_count = 0; - top->obj_id = 0; - return; + top->share_mode = SM_EMPTY; + top->ref_count = 0; + top->obj_id = 0; + return; } + { - obj = entry->object.vm_object; + struct vm_object *obj, *tmp_obj; + int ref_count; + uint32_t entry_size; - vm_object_lock(obj); + entry_size = (uint32_t) ((entry->vme_end - entry->vme_start) / PAGE_SIZE_64); - if ((ref_count = obj->ref_count) > 1 && obj->paging_in_progress) - ref_count--; + obj = entry->object.vm_object; - if (obj->shadow) { - if (ref_count == 1) - top->private_pages_resident = obj->resident_page_count; - else - top->shared_pages_resident = obj->resident_page_count; - top->ref_count = ref_count; - top->share_mode = SM_COW; - - while ((tmp_obj = obj->shadow)) { - vm_object_lock(tmp_obj); - vm_object_unlock(obj); - obj = tmp_obj; + 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--; - top->shared_pages_resident += obj->resident_page_count; - top->ref_count += ref_count - 1; - } - } else { - if (entry->needs_copy) { - top->share_mode = SM_COW; - top->shared_pages_resident = obj->resident_page_count; + assert(obj->reusable_page_count <= obj->resident_page_count); + if (obj->shadow) { + if (ref_count == 1) + top->private_pages_resident = + OBJ_RESIDENT_COUNT(obj, entry_size); + else + top->shared_pages_resident = + OBJ_RESIDENT_COUNT(obj, entry_size); + top->ref_count = ref_count; + top->share_mode = SM_COW; + + while ((tmp_obj = obj->shadow)) { + vm_object_lock(tmp_obj); + vm_object_unlock(obj); + obj = tmp_obj; + + if ((ref_count = obj->ref_count) > 1 && obj->paging_in_progress) + ref_count--; + + assert(obj->reusable_page_count <= obj->resident_page_count); + top->shared_pages_resident += + OBJ_RESIDENT_COUNT(obj, entry_size); + top->ref_count += ref_count - 1; + } } else { - if (ref_count == 1 || - (ref_count == 2 && !(obj->pager_trusted) && !(obj->internal))) { - top->share_mode = SM_PRIVATE; - top->private_pages_resident = obj->resident_page_count; - } else { - top->share_mode = SM_SHARED; - top->shared_pages_resident = obj->resident_page_count; - } + if (entry->superpage_size) { + top->share_mode = SM_LARGE_PAGE; + top->shared_pages_resident = 0; + top->private_pages_resident = entry_size; + } else if (entry->needs_copy) { + top->share_mode = SM_COW; + top->shared_pages_resident = + OBJ_RESIDENT_COUNT(obj, entry_size); + } else { + if (ref_count == 1 || + (ref_count == 2 && !(obj->pager_trusted) && !(obj->internal))) { + top->share_mode = SM_PRIVATE; + top->private_pages_resident = + OBJ_RESIDENT_COUNT(obj, + entry_size); + } else { + top->share_mode = SM_SHARED; + top->shared_pages_resident = + OBJ_RESIDENT_COUNT(obj, + entry_size); + } + } + top->ref_count = ref_count; } - top->ref_count = ref_count; - } - top->obj_id = (int)obj; + /* XXX K64: obj_id will be truncated */ + top->obj_id = (unsigned int) (uintptr_t)VM_KERNEL_ADDRPERM(obj); - vm_object_unlock(obj); + vm_object_unlock(obj); } } -static void +void vm_map_region_walk( vm_map_t map, - vm_map_offset_t va, - vm_map_entry_t entry, + 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) + vm_region_extended_info_t extended, + boolean_t look_for_pages, + mach_msg_type_number_t count) { register struct vm_object *obj, *tmp_obj; register vm_map_offset_t last_offset; @@ -7952,89 +10561,116 @@ vm_map_region_walk( int shadow_depth; if ((entry->object.vm_object == 0) || - (entry->is_sub_map) || - (entry->object.vm_object->phys_contiguous)) { - extended->share_mode = SM_EMPTY; - extended->ref_count = 0; - return; + (entry->is_sub_map) || + (entry->object.vm_object->phys_contiguous && + !entry->superpage_size)) { + extended->share_mode = SM_EMPTY; + extended->ref_count = 0; + return; } + + if (entry->superpage_size) { + extended->shadow_depth = 0; + extended->share_mode = SM_LARGE_PAGE; + extended->ref_count = 1; + extended->external_pager = 0; + extended->pages_resident = (unsigned int)(range >> PAGE_SHIFT); + extended->shadow_depth = 0; + return; + } + { - obj = entry->object.vm_object; - - vm_object_lock(obj); - - if ((ref_count = obj->ref_count) > 1 && obj->paging_in_progress) - ref_count--; - - 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); - - shadow_object = obj->shadow; - shadow_depth = 0; - if (shadow_object != VM_OBJECT_NULL) { - vm_object_lock(shadow_object); - for (; - shadow_object != VM_OBJECT_NULL; - shadow_depth++) { - vm_object_t next_shadow; - - 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; + obj = entry->object.vm_object; + + vm_object_lock(obj); + + if ((ref_count = obj->ref_count) > 1 && obj->paging_in_progress) + ref_count--; + + if (look_for_pages) { + for (last_offset = offset + range; + offset < last_offset; + offset += PAGE_SIZE_64, va += PAGE_SIZE) { + vm_map_region_look_for_page(map, va, obj, + offset, ref_count, + 0, extended, count); + } + } else { + shadow_object = obj->shadow; + shadow_depth = 0; + + if ( !(obj->pager_trusted) && !(obj->internal)) + extended->external_pager = 1; + + if (shadow_object != VM_OBJECT_NULL) { + vm_object_lock(shadow_object); + for (; + shadow_object != VM_OBJECT_NULL; + shadow_depth++) { + vm_object_t next_shadow; + + if ( !(shadow_object->pager_trusted) && + !(shadow_object->internal)) + extended->external_pager = 1; + + next_shadow = shadow_object->shadow; + if (next_shadow) { + vm_object_lock(next_shadow); + } + vm_object_unlock(shadow_object); + shadow_object = next_shadow; + } + } + extended->shadow_depth = shadow_depth; + } + + if (extended->shadow_depth || entry->needs_copy) + extended->share_mode = SM_COW; else { - if (obj->true_share) - extended->share_mode = SM_TRUESHARED; - else - extended->share_mode = SM_SHARED; + 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; + 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); + 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--; + 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); + extended->ref_count += ref_count; + obj = tmp_obj; + } + vm_object_unlock(obj); - if (extended->share_mode == SM_SHARED) { - register vm_map_entry_t cur; - register vm_map_entry_t last; - int my_refs; + if (extended->share_mode == SM_SHARED) { + register vm_map_entry_t cur; + register vm_map_entry_t last; + int my_refs; - obj = entry->object.vm_object; - last = vm_map_to_entry(map); - my_refs = 0; + obj = entry->object.vm_object; + last = vm_map_to_entry(map); + my_refs = 0; - if ((ref_count = obj->ref_count) > 1 && obj->paging_in_progress) - ref_count--; - for (cur = vm_map_first_entry(map); cur != last; cur = cur->vme_next) - my_refs += vm_map_region_count_obj_refs(cur, obj); + if ((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; - } + if (my_refs == ref_count) + extended->share_mode = SM_PRIVATE_ALIASED; + else if (my_refs > 1) + extended->share_mode = SM_SHARED_ALIASED; + } } } @@ -8045,18 +10681,19 @@ vm_map_region_walk( static void vm_map_region_look_for_page( __unused vm_map_t map, - __unused vm_map_offset_t va, - vm_object_t object, - vm_object_offset_t offset, + __unused vm_map_offset_t va, + vm_object_t object, + vm_object_offset_t offset, int max_refcnt, int depth, - vm_region_extended_info_t extended) + 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; - + register vm_page_t p; + register vm_object_t shadow; + register int ref_count; + vm_object_t caller_object; + kern_return_t kr; shadow = object->shadow; caller_object = object; @@ -8064,36 +10701,85 @@ vm_map_region_look_for_page( while (TRUE) { if ( !(object->pager_trusted) && !(object->internal)) - extended->external_pager = 1; + extended->external_pager = 1; if ((p = vm_page_lookup(object, offset)) != VM_PAGE_NULL) { if (shadow && (max_refcnt == 1)) extended->pages_shared_now_private++; - if (!p->fictitious && + if (!p->fictitious && (p->dirty || pmap_is_modified(p->phys_page))) extended->pages_dirtied++; + else if (count >= VM_REGION_EXTENDED_INFO_COUNT) { + if (p->reusable || p->object->all_reusable) { + extended->pages_reusable++; + } + } - extended->pages_resident++; + extended->pages_resident++; if(object != caller_object) - vm_object_unlock(object); + vm_object_unlock(object); return; } +#if MACH_PAGEMAP if (object->existence_map) { if (vm_external_state_get(object->existence_map, offset) == VM_EXTERNAL_STATE_EXISTS) { extended->pages_swapped_out++; if(object != caller_object) - vm_object_unlock(object); + vm_object_unlock(object); return; } + } else +#endif /* MACH_PAGEMAP */ + if (object->internal && + object->alive && + !object->terminating && + object->pager_ready) { + + if (COMPRESSED_PAGER_IS_ACTIVE || DEFAULT_FREEZER_COMPRESSED_PAGER_IS_ACTIVE) { + if (VM_COMPRESSOR_PAGER_STATE_GET(object, + offset) + == VM_EXTERNAL_STATE_EXISTS) { + /* the pager has that page */ + extended->pages_swapped_out++; + if (object != caller_object) + vm_object_unlock(object); + return; + } + } else { + memory_object_t pager; + + vm_object_paging_begin(object); + pager = object->pager; + vm_object_unlock(object); + + kr = memory_object_data_request( + pager, + offset + object->paging_offset, + 0, /* just poke the pager */ + VM_PROT_READ, + NULL); + + vm_object_lock(object); + vm_object_paging_end(object); + + if (kr == KERN_SUCCESS) { + /* the pager has that page */ + extended->pages_swapped_out++; + if (object != caller_object) + vm_object_unlock(object); + return; + } + } } + if (shadow) { - vm_object_lock(shadow); + vm_object_lock(shadow); if ((ref_count = shadow->ref_count) > 1 && shadow->paging_in_progress) ref_count--; @@ -8105,15 +10791,15 @@ vm_map_region_look_for_page( max_refcnt = ref_count; if(object != caller_object) - vm_object_unlock(object); + vm_object_unlock(object); - offset = offset + object->shadow_offset; + offset = offset + object->vo_shadow_offset; object = shadow; shadow = object->shadow; continue; } if(object != caller_object) - vm_object_unlock(object); + vm_object_unlock(object); break; } } @@ -8128,26 +10814,26 @@ vm_map_region_count_obj_refs( register vm_object_t tmp_obj; if (entry->object.vm_object == 0) - return(0); + return(0); if (entry->is_sub_map) - return(0); + return(0); else { - ref_count = 0; + ref_count = 0; - chk_obj = entry->object.vm_object; - vm_object_lock(chk_obj); + chk_obj = entry->object.vm_object; + vm_object_lock(chk_obj); - while (chk_obj) { - if (chk_obj == object) - ref_count++; - tmp_obj = chk_obj->shadow; - if (tmp_obj) - vm_object_lock(tmp_obj); - vm_object_unlock(chk_obj); + while (chk_obj) { + if (chk_obj == object) + ref_count++; + tmp_obj = chk_obj->shadow; + if (tmp_obj) + vm_object_lock(tmp_obj); + vm_object_unlock(chk_obj); - chk_obj = tmp_obj; - } + chk_obj = tmp_obj; + } } return(ref_count); } @@ -8178,27 +10864,30 @@ vm_map_simplify_entry( prev_entry = this_entry->vme_prev; if ((this_entry != vm_map_to_entry(map)) && - (prev_entry != vm_map_to_entry(map)) && + (prev_entry != vm_map_to_entry(map)) && (prev_entry->vme_end == this_entry->vme_start) && - (prev_entry->is_sub_map == FALSE) && - (this_entry->is_sub_map == FALSE) && + (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 - prev_entry->vme_start)) == this_entry->offset) && + (prev_entry->map_aligned == this_entry->map_aligned) && (prev_entry->inheritance == this_entry->inheritance) && (prev_entry->protection == this_entry->protection) && (prev_entry->max_protection == this_entry->max_protection) && (prev_entry->behavior == this_entry->behavior) && (prev_entry->alias == this_entry->alias) && + (prev_entry->zero_wired_pages == this_entry->zero_wired_pages) && + (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->use_pmap == FALSE) && (this_entry->use_pmap == FALSE) && @@ -8208,13 +10897,21 @@ vm_map_simplify_entry( (this_entry->needs_wakeup == FALSE) && (prev_entry->is_shared == FALSE) && (this_entry->is_shared == FALSE) - ) { - _vm_map_entry_unlink(&map->hdr, prev_entry); + ) { + vm_map_store_entry_unlink(map, prev_entry); + assert(prev_entry->vme_start < this_entry->vme_end); + if (prev_entry->map_aligned) + assert(VM_MAP_PAGE_ALIGNED(prev_entry->vme_start, + VM_MAP_PAGE_MASK(map))); this_entry->vme_start = prev_entry->vme_start; this_entry->offset = prev_entry->offset; - vm_object_deallocate(prev_entry->object.vm_object); + if (prev_entry->is_sub_map) { + vm_map_deallocate(prev_entry->object.sub_map); + } else { + vm_object_deallocate(prev_entry->object.vm_object); + } vm_map_entry_dispose(map, prev_entry); - SAVE_HINT(map, this_entry); + SAVE_HINT_MAP_WRITE(map, this_entry); counter(c_vm_map_simplified++); } } @@ -8252,6 +10949,11 @@ vm_map_simplify_range( return; } + start = vm_map_trunc_page(start, + VM_MAP_PAGE_MASK(map)); + end = vm_map_round_page(end, + VM_MAP_PAGE_MASK(map)); + if (!vm_map_lookup_entry(map, start, &entry)) { /* "start" is not mapped and "entry" ends before "start" */ if (entry == vm_map_to_entry(map)) { @@ -8325,14 +11027,14 @@ vm_map_machine_attribute( sync_size = 0; } else { sub_size = entry->vme_end - start; - sync_size -= sub_size; + sync_size -= sub_size; } if(entry->is_sub_map) { vm_map_offset_t sub_start; vm_map_offset_t sub_end; sub_start = (start - entry->vme_start) - + entry->offset; + + entry->offset; sub_end = sub_start + sub_size; vm_map_machine_attribute( entry->object.sub_map, @@ -8350,7 +11052,7 @@ vm_map_machine_attribute( vm_map_size_t range; range = sub_size; offset = (start - entry->vme_start) - + entry->offset; + + entry->offset; base_offset = offset; object = entry->object.vm_object; base_object = object; @@ -8364,13 +11066,13 @@ vm_map_machine_attribute( if (m && !m->fictitious) { ret = - pmap_attribute_cache_sync( - m->phys_page, - PAGE_SIZE, - attribute, value); + pmap_attribute_cache_sync( + m->phys_page, + PAGE_SIZE, + attribute, value); } else if (object->shadow) { - offset = offset + object->shadow_offset; + offset = offset + object->vo_shadow_offset; last_object = object; object = object->shadow; vm_object_lock(last_object->shadow); @@ -8423,346 +11125,590 @@ vm_map_behavior_set( vm_map_entry_t temp_entry; XPR(XPR_VM_MAP, - "vm_map_behavior_set, 0x%X start 0x%X end 0x%X behavior %d", - (integer_t)map, start, end, new_behavior, 0); + "vm_map_behavior_set, 0x%X start 0x%X end 0x%X behavior %d", + map, start, end, new_behavior, 0); + + if (start > end || + start < vm_map_min(map) || + end > vm_map_max(map)) { + return KERN_NO_SPACE; + } switch (new_behavior) { + + /* + * This first block of behaviors all set a persistent state on the specified + * memory range. All we have to do here is to record the desired behavior + * in the vm_map_entry_t's. + */ + case VM_BEHAVIOR_DEFAULT: case VM_BEHAVIOR_RANDOM: case VM_BEHAVIOR_SEQUENTIAL: case VM_BEHAVIOR_RSEQNTL: + case VM_BEHAVIOR_ZERO_WIRED_PAGES: + vm_map_lock(map); + + /* + * The entire address range must be valid for the map. + * Note that vm_map_range_check() does a + * vm_map_lookup_entry() internally and returns the + * entry containing the start of the address range if + * the entire range is valid. + */ + if (vm_map_range_check(map, start, end, &temp_entry)) { + entry = temp_entry; + vm_map_clip_start(map, entry, start); + } + else { + vm_map_unlock(map); + return(KERN_INVALID_ADDRESS); + } + + while ((entry != vm_map_to_entry(map)) && (entry->vme_start < end)) { + vm_map_clip_end(map, entry, end); + assert(!entry->use_pmap); + + if( new_behavior == VM_BEHAVIOR_ZERO_WIRED_PAGES ) { + entry->zero_wired_pages = TRUE; + } else { + entry->behavior = new_behavior; + } + entry = entry->vme_next; + } + + vm_map_unlock(map); break; + + /* + * The rest of these are different from the above in that they cause + * an immediate action to take place as opposed to setting a behavior that + * affects future actions. + */ + case VM_BEHAVIOR_WILLNEED: + return vm_map_willneed(map, start, end); + case VM_BEHAVIOR_DONTNEED: - new_behavior = VM_BEHAVIOR_DEFAULT; - break; - default: - return(KERN_INVALID_ARGUMENT); - } + return vm_map_msync(map, start, end - start, VM_SYNC_DEACTIVATE | VM_SYNC_CONTIGUOUS); - vm_map_lock(map); + case VM_BEHAVIOR_FREE: + return vm_map_msync(map, start, end - start, VM_SYNC_KILLPAGES | VM_SYNC_CONTIGUOUS); - /* - * The entire address range must be valid for the map. - * Note that vm_map_range_check() does a - * vm_map_lookup_entry() internally and returns the - * entry containing the start of the address range if - * the entire range is valid. - */ - if (vm_map_range_check(map, start, end, &temp_entry)) { - entry = temp_entry; - vm_map_clip_start(map, entry, start); - } - else { - vm_map_unlock(map); - return(KERN_INVALID_ADDRESS); - } + case VM_BEHAVIOR_REUSABLE: + return vm_map_reusable_pages(map, start, end); - while ((entry != vm_map_to_entry(map)) && (entry->vme_start < end)) { - vm_map_clip_end(map, entry, end); + case VM_BEHAVIOR_REUSE: + return vm_map_reuse_pages(map, start, end); - entry->behavior = new_behavior; + case VM_BEHAVIOR_CAN_REUSE: + return vm_map_can_reuse(map, start, end); - entry = entry->vme_next; + default: + return(KERN_INVALID_ARGUMENT); } - vm_map_unlock(map); return(KERN_SUCCESS); } -#include -#if MACH_KDB -#include -#include - -#define printf db_printf - /* - * Forward declarations for internal functions. + * Internals for madvise(MADV_WILLNEED) system call. + * + * The present implementation is to do a read-ahead if the mapping corresponds + * to a mapped regular file. If it's an anonymous mapping, then we do nothing + * and basically ignore the "advice" (which we are always free to do). */ -extern void vm_map_links_print( - struct vm_map_links *links); - -extern void vm_map_header_print( - struct vm_map_header *header); -extern void vm_map_entry_print( - vm_map_entry_t entry); -extern void vm_follow_entry( - vm_map_entry_t entry); - -extern void vm_follow_map( - vm_map_t map); - -/* - * vm_map_links_print: [ debug ] - */ -void -vm_map_links_print( - struct vm_map_links *links) +static kern_return_t +vm_map_willneed( + vm_map_t map, + vm_map_offset_t start, + vm_map_offset_t end +) { - iprintf("prev = %08X next = %08X start = %016llX end = %016llX\n", - links->prev, - links->next, - (unsigned long long)links->start, - (unsigned long long)links->end); -} + vm_map_entry_t entry; + vm_object_t object; + memory_object_t pager; + struct vm_object_fault_info fault_info; + kern_return_t kr; + vm_object_size_t len; + vm_object_offset_t offset; -/* - * vm_map_header_print: [ debug ] - */ -void -vm_map_header_print( - struct vm_map_header *header) -{ - vm_map_links_print(&header->links); - iprintf("nentries = %08X, %sentries_pageable\n", - header->nentries, - (header->entries_pageable ? "" : "!")); -} + /* + * 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. + */ -/* - * vm_follow_entry: [ debug ] - */ -void -vm_follow_entry( - vm_map_entry_t entry) -{ - int shadows; + 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; - iprintf("map entry %08X\n", entry); + /* + * The MADV_WILLNEED operation doesn't require any changes to the + * vm_map_entry_t's, so the read lock is sufficient. + */ - db_indent += 2; + vm_map_lock_read(map); - shadows = vm_follow_object(entry->object.vm_object); - iprintf("Total objects : %d\n",shadows); + /* + * The madvise semantics require that the address range be fully + * allocated with no holes. Otherwise, we're required to return + * an error. + */ - db_indent -= 2; -} + if (! vm_map_range_check(map, start, end, &entry)) { + vm_map_unlock_read(map); + return KERN_INVALID_ADDRESS; + } -/* - * vm_map_entry_print: [ debug ] - */ -void -vm_map_entry_print( - register vm_map_entry_t entry) -{ - static const char *inheritance_name[4] = - { "share", "copy", "none", "?"}; - static const char *behavior_name[4] = - { "dflt", "rand", "seqtl", "rseqntl" }; - - iprintf("map entry %08X - prev = %08X next = %08X\n", entry, entry->vme_prev, entry->vme_next); + /* + * 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; - db_indent += 2; + /* + * Set the length so we don't go beyond the end of the + * map_entry or beyond the end of the range we were given. + * This range could span also multiple map entries all of which + * map different files, so make sure we only do the right amount + * of I/O for each object. Note that it's possible for there + * to be multiple map entries all referring to the same object + * but with different page permissions, but it's not worth + * trying to optimize that case. + */ + len = MIN(entry->vme_end - start, end - start); - vm_map_links_print(&entry->links); + if ((vm_size_t) len != len) { + /* 32-bit overflow */ + len = (vm_size_t) (0 - PAGE_SIZE); + } + fault_info.cluster_size = (vm_size_t) len; + fault_info.lo_offset = offset; + fault_info.hi_offset = offset + len; + fault_info.user_tag = entry->alias; - iprintf("start = %016llX end = %016llX - prot=%x/%x/%s\n", - (unsigned long long)entry->vme_start, - (unsigned long long)entry->vme_end, - entry->protection, - entry->max_protection, - inheritance_name[(entry->inheritance & 0x3)]); + /* + * If there's no read permission to this mapping, then just + * skip it. + */ + if ((entry->protection & VM_PROT_READ) == 0) { + entry = entry->vme_next; + start = entry->vme_start; + continue; + } - iprintf("behavior = %s, wired_count = %d, user_wired_count = %d\n", - behavior_name[(entry->behavior & 0x3)], - entry->wired_count, - entry->user_wired_count); - iprintf("%sin_transition, %sneeds_wakeup\n", - (entry->in_transition ? "" : "!"), - (entry->needs_wakeup ? "" : "!")); + /* + * Find the file object backing this map entry. If there is + * none, then we simply ignore the "will need" advice for this + * entry and go on to the next one. + */ + if ((object = find_vnode_object(entry)) == VM_OBJECT_NULL) { + entry = entry->vme_next; + start = entry->vme_start; + continue; + } - if (entry->is_sub_map) { - iprintf("submap = %08X - offset = %016llX\n", - entry->object.sub_map, - (unsigned long long)entry->offset); - } else { - iprintf("object = %08X offset = %016llX - ", - entry->object.vm_object, - (unsigned long long)entry->offset); - printf("%sis_shared, %sneeds_copy\n", - (entry->is_shared ? "" : "!"), - (entry->needs_copy ? "" : "!")); - } + /* + * The data_request() could take a long time, so let's + * release the map lock to avoid blocking other threads. + */ + vm_map_unlock_read(map); - db_indent -= 2; -} + vm_object_paging_begin(object); + pager = object->pager; + vm_object_unlock(object); -/* - * vm_follow_map: [ debug ] - */ -void -vm_follow_map( - vm_map_t map) -{ - register vm_map_entry_t entry; + /* + * Get the data from the object asynchronously. + * + * Note that memory_object_data_request() places limits on the + * amount of I/O it will do. Regardless of the len we + * specified, it won't do more than MAX_UPL_TRANSFER and it + * silently truncates the len to that size. This isn't + * necessarily bad since madvise shouldn't really be used to + * page in unlimited amounts of data. Other Unix variants + * limit the willneed case as well. If this turns out to be an + * issue for developers, then we can always adjust the policy + * here and still be backwards compatible since this is all + * just "advice". + */ + kr = memory_object_data_request( + pager, + offset + object->paging_offset, + 0, /* ignored */ + VM_PROT_READ, + (memory_object_fault_info_t)&fault_info); + + vm_object_lock(object); + vm_object_paging_end(object); + vm_object_unlock(object); - iprintf("task map %08X\n", map); + /* + * If we couldn't do the I/O for some reason, just give up on + * the madvise. We still return success to the user since + * madvise isn't supposed to fail when the advice can't be + * taken. + */ + if (kr != KERN_SUCCESS) { + return KERN_SUCCESS; + } - db_indent += 2; + start += len; + if (start >= end) { + /* done */ + return KERN_SUCCESS; + } - for (entry = vm_map_first_entry(map); - entry && entry != vm_map_to_entry(map); - entry = entry->vme_next) { - vm_follow_entry(entry); + /* look up next entry */ + vm_map_lock_read(map); + if (! vm_map_lookup_entry(map, start, &entry)) { + /* + * There's a new hole in the address range. + */ + vm_map_unlock_read(map); + return KERN_INVALID_ADDRESS; + } } - db_indent -= 2; + vm_map_unlock_read(map); + return KERN_SUCCESS; } -/* - * vm_map_print: [ debug ] - */ -void -vm_map_print( - db_addr_t inmap) +static boolean_t +vm_map_entry_is_reusable( + vm_map_entry_t entry) { - register vm_map_entry_t entry; - vm_map_t map; -#if TASK_SWAPPER - char *swstate; -#endif /* TASK_SWAPPER */ + vm_object_t object; + + switch (entry->alias) { + case VM_MEMORY_MALLOC: + case VM_MEMORY_MALLOC_SMALL: + case VM_MEMORY_MALLOC_LARGE: + case VM_MEMORY_REALLOC: + case VM_MEMORY_MALLOC_TINY: + case VM_MEMORY_MALLOC_LARGE_REUSABLE: + case VM_MEMORY_MALLOC_LARGE_REUSED: + /* + * This is a malloc() memory region: check if it's still + * in its original state and can be re-used for more + * malloc() allocations. + */ + break; + default: + /* + * Not a malloc() memory region: let the caller decide if + * it's re-usable. + */ + return TRUE; + } - map = (vm_map_t)(long) - inmap; /* Make sure we have the right type */ + if (entry->is_shared || + entry->is_sub_map || + entry->in_transition || + entry->protection != VM_PROT_DEFAULT || + entry->max_protection != VM_PROT_ALL || + entry->inheritance != VM_INHERIT_DEFAULT || + entry->no_cache || + entry->permanent || + entry->superpage_size != FALSE || + entry->zero_wired_pages || + entry->wired_count != 0 || + entry->user_wired_count != 0) { + return FALSE; + } - iprintf("task map %08X\n", map); + object = entry->object.vm_object; + if (object == VM_OBJECT_NULL) { + return TRUE; + } + if ( +#if 0 + /* + * Let's proceed even if the VM object is potentially + * shared. + * We check for this later when processing the actual + * VM pages, so the contents will be safe if shared. + * + * But we can still mark this memory region as "reusable" to + * acknowledge that the caller did let us know that the memory + * could be re-used and should not be penalized for holding + * on to it. This allows its "resident size" to not include + * the reusable range. + */ + object->ref_count == 1 && +#endif + object->wired_page_count == 0 && + object->copy == VM_OBJECT_NULL && + object->shadow == VM_OBJECT_NULL && + object->copy_strategy == MEMORY_OBJECT_COPY_SYMMETRIC && + object->internal && + !object->true_share && + object->wimg_bits == VM_WIMG_USE_DEFAULT && + !object->code_signed) { + return TRUE; + } + return FALSE; + + +} - db_indent += 2; +static kern_return_t +vm_map_reuse_pages( + vm_map_t map, + vm_map_offset_t start, + vm_map_offset_t end) +{ + vm_map_entry_t entry; + vm_object_t object; + vm_object_offset_t start_offset, end_offset; - vm_map_header_print(&map->hdr); + /* + * The MADV_REUSE operation doesn't require any changes to the + * vm_map_entry_t's, so the read lock is sufficient. + */ - iprintf("pmap = %08X size = %08X ref = %d hint = %08X first_free = %08X\n", - map->pmap, - map->size, - map->ref_count, - map->hint, - map->first_free); + vm_map_lock_read(map); - iprintf("%swait_for_space, %swiring_required, timestamp = %d\n", - (map->wait_for_space ? "" : "!"), - (map->wiring_required ? "" : "!"), - map->timestamp); + /* + * The madvise semantics require that the address range be fully + * allocated with no holes. Otherwise, we're required to return + * an error. + */ -#if TASK_SWAPPER - switch (map->sw_state) { - case MAP_SW_IN: - swstate = "SW_IN"; - break; - case MAP_SW_OUT: - swstate = "SW_OUT"; - break; - default: - swstate = "????"; - break; + if (!vm_map_range_check(map, start, end, &entry)) { + vm_map_unlock_read(map); + vm_page_stats_reusable.reuse_pages_failure++; + return KERN_INVALID_ADDRESS; } - iprintf("res = %d, sw_state = %s\n", map->res_count, swstate); -#endif /* TASK_SWAPPER */ - for (entry = vm_map_first_entry(map); - entry && entry != vm_map_to_entry(map); + /* + * 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_map_entry_print(entry); - } + /* + * Sanity check on the VM map entry. + */ + if (! vm_map_entry_is_reusable(entry)) { + vm_map_unlock_read(map); + vm_page_stats_reusable.reuse_pages_failure++; + return KERN_INVALID_ADDRESS; + } + + /* + * The first time through, the start address could be anywhere + * within the vm_map_entry we found. So adjust the offset to + * correspond. + */ + if (entry->vme_start < start) { + start_offset = start - entry->vme_start; + } else { + start_offset = 0; + } + end_offset = MIN(end, entry->vme_end) - entry->vme_start; + start_offset += entry->offset; + end_offset += entry->offset; + + object = entry->object.vm_object; + 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); + } - db_indent -= 2; + if (entry->alias == VM_MEMORY_MALLOC_LARGE_REUSABLE) { + /* + * XXX + * We do not hold the VM map exclusively here. + * The "alias" field is not that critical, so it's + * safe to update it here, as long as it is the only + * one that can be modified while holding the VM map + * "shared". + */ + entry->alias = VM_MEMORY_MALLOC_LARGE_REUSED; + } + } + + vm_map_unlock_read(map); + vm_page_stats_reusable.reuse_pages_success++; + return KERN_SUCCESS; } -/* - * Routine: vm_map_copy_print - * Purpose: - * Pretty-print a copy object for ddb. - */ -void -vm_map_copy_print( - db_addr_t incopy) +static kern_return_t +vm_map_reusable_pages( + vm_map_t map, + vm_map_offset_t start, + vm_map_offset_t end) { - vm_map_copy_t copy; - vm_map_entry_t entry; - - copy = (vm_map_copy_t)(long) - incopy; /* Make sure we have the right type */ + vm_map_entry_t entry; + vm_object_t object; + vm_object_offset_t start_offset, end_offset; - printf("copy object 0x%x\n", copy); + /* + * The MADV_REUSABLE operation doesn't require any changes to the + * vm_map_entry_t's, so the read lock is sufficient. + */ - db_indent += 2; + vm_map_lock_read(map); - iprintf("type=%d", copy->type); - switch (copy->type) { - case VM_MAP_COPY_ENTRY_LIST: - printf("[entry_list]"); - break; - - case VM_MAP_COPY_OBJECT: - printf("[object]"); - break; - - case VM_MAP_COPY_KERNEL_BUFFER: - printf("[kernel_buffer]"); - break; + /* + * The madvise semantics require that the address range be fully + * allocated with no holes. Otherwise, we're required to return + * an error. + */ - default: - printf("[bad type]"); - break; + if (!vm_map_range_check(map, start, end, &entry)) { + vm_map_unlock_read(map); + vm_page_stats_reusable.reusable_pages_failure++; + return KERN_INVALID_ADDRESS; } - printf(", offset=0x%llx", (unsigned long long)copy->offset); - printf(", size=0x%x\n", copy->size); - switch (copy->type) { - case VM_MAP_COPY_ENTRY_LIST: - vm_map_header_print(©->cpy_hdr); - for (entry = vm_map_copy_first_entry(copy); - entry && entry != vm_map_copy_to_entry(copy); - entry = entry->vme_next) { - vm_map_entry_print(entry); + /* + * Examine each vm_map_entry_t in the range. + */ + for (; entry != vm_map_to_entry(map) && entry->vme_start < end; + entry = entry->vme_next) { + int kill_pages = 0; + + /* + * Sanity check on the VM map entry. + */ + if (! vm_map_entry_is_reusable(entry)) { + vm_map_unlock_read(map); + vm_page_stats_reusable.reusable_pages_failure++; + return KERN_INVALID_ADDRESS; } - break; - case VM_MAP_COPY_OBJECT: - iprintf("object=0x%x\n", copy->cpy_object); - break; + /* + * The first time through, the start address could be anywhere + * within the vm_map_entry we found. So adjust the offset to + * correspond. + */ + if (entry->vme_start < start) { + start_offset = start - entry->vme_start; + } else { + start_offset = 0; + } + end_offset = MIN(end, entry->vme_end) - entry->vme_start; + start_offset += entry->offset; + end_offset += entry->offset; - case VM_MAP_COPY_KERNEL_BUFFER: - iprintf("kernel buffer=0x%x", copy->cpy_kdata); - printf(", kalloc_size=0x%x\n", copy->cpy_kalloc_size); - break; + object = entry->object.vm_object; + if (object == VM_OBJECT_NULL) + continue; - } - db_indent -=2; + vm_object_lock(object); + if (object->ref_count == 1 && !object->shadow) + kill_pages = 1; + 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*/); + } 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) { + /* + * XXX + * We do not hold the VM map exclusively here. + * The "alias" field is not that critical, so it's + * safe to update it here, as long as it is the only + * one that can be modified while holding the VM map + * "shared". + */ + entry->alias = VM_MEMORY_MALLOC_LARGE_REUSABLE; + } + } + + vm_map_unlock_read(map); + vm_page_stats_reusable.reusable_pages_success++; + return KERN_SUCCESS; } -/* - * db_vm_map_total_size(map) [ debug ] - * - * return the total virtual size (in bytes) of the map - */ -vm_map_size_t -db_vm_map_total_size( - db_addr_t inmap) + +static kern_return_t +vm_map_can_reuse( + vm_map_t map, + vm_map_offset_t start, + vm_map_offset_t end) { - vm_map_entry_t entry; - vm_map_size_t total; - vm_map_t map; + vm_map_entry_t entry; + + /* + * The MADV_REUSABLE operation doesn't require any changes to the + * vm_map_entry_t's, so the read lock is sufficient. + */ - map = (vm_map_t)(long) - inmap; /* Make sure we have the right type */ + vm_map_lock_read(map); - total = 0; - for (entry = vm_map_first_entry(map); - entry != vm_map_to_entry(map); - entry = entry->vme_next) { - total += entry->vme_end - entry->vme_start; + /* + * 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); + vm_page_stats_reusable.can_reuse_failure++; + return KERN_INVALID_ADDRESS; } - return total; + /* + * Examine each vm_map_entry_t in the range. + */ + for (; entry != vm_map_to_entry(map) && entry->vme_start < end; + entry = entry->vme_next) { + /* + * Sanity check on the VM map entry. + */ + if (! vm_map_entry_is_reusable(entry)) { + vm_map_unlock_read(map); + vm_page_stats_reusable.can_reuse_failure++; + return KERN_INVALID_ADDRESS; + } + } + + vm_map_unlock_read(map); + vm_page_stats_reusable.can_reuse_success++; + return KERN_SUCCESS; } -#endif /* MACH_KDB */ /* * Routine: vm_map_entry_insert @@ -8784,18 +11730,39 @@ vm_map_entry_insert( vm_prot_t max_protection, vm_behavior_t behavior, vm_inherit_t inheritance, - unsigned wired_count) + unsigned wired_count, + boolean_t no_cache, + boolean_t permanent, + unsigned int superpage_size, + boolean_t clear_map_aligned) { vm_map_entry_t new_entry; assert(insp_entry != (vm_map_entry_t)0); - new_entry = vm_map_entry_create(map); + new_entry = vm_map_entry_create(map, !map->hdr.entries_pageable); + + if (VM_MAP_PAGE_SHIFT(map) != PAGE_SHIFT) { + new_entry->map_aligned = TRUE; + } else { + new_entry->map_aligned = FALSE; + } + if (clear_map_aligned && + ! VM_MAP_PAGE_ALIGNED(end, VM_MAP_PAGE_MASK(map))) { + new_entry->map_aligned = FALSE; + } new_entry->vme_start = start; new_entry->vme_end = end; assert(page_aligned(new_entry->vme_start)); assert(page_aligned(new_entry->vme_end)); + assert(VM_MAP_PAGE_ALIGNED(new_entry->vme_start, + VM_MAP_PAGE_MASK(map))); + if (new_entry->map_aligned) { + 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; @@ -8811,19 +11778,28 @@ vm_map_entry_insert( new_entry->wired_count = wired_count; new_entry->user_wired_count = 0; new_entry->use_pmap = FALSE; + new_entry->alias = 0; + new_entry->zero_wired_pages = FALSE; + new_entry->no_cache = no_cache; + new_entry->permanent = permanent; + if (superpage_size) + new_entry->superpage_size = TRUE; + else + new_entry->superpage_size = FALSE; + new_entry->used_for_jit = FALSE; /* * Insert the new entry into the list. */ - vm_map_entry_link(map, insp_entry, new_entry); + vm_map_store_entry_link(map, insp_entry, new_entry); map->size += end - start; /* * Update the free space hint and the lookup hint. */ - SAVE_HINT(map, new_entry); + SAVE_HINT_MAP_WRITE(map, new_entry); return new_entry; } @@ -8860,7 +11836,8 @@ vm_map_remap_extract( boolean_t new_entry_needs_copy; assert(map != VM_MAP_NULL); - assert(size != 0 && size == vm_map_round_page(size)); + assert(size != 0); + assert(size == vm_map_round_page(size, PAGE_MASK)); assert(inheritance == VM_INHERIT_NONE || inheritance == VM_INHERIT_COPY || inheritance == VM_INHERIT_SHARE); @@ -8868,8 +11845,9 @@ vm_map_remap_extract( /* * Compute start and end of region. */ - src_start = vm_map_trunc_page(addr); - src_end = vm_map_round_page(src_start + size); + src_start = vm_map_trunc_page(addr, PAGE_MASK); + src_end = vm_map_round_page(src_start + size, PAGE_MASK); + /* * Initialize map_header. @@ -8878,6 +11856,9 @@ vm_map_remap_extract( map_header->links.prev = (struct vm_map_entry *)&map_header->links; map_header->nentries = 0; map_header->entries_pageable = pageable; + map_header->page_shift = PAGE_SHIFT; + + vm_map_store_init( map_header ); *cur_protection = VM_PROT_ALL; *max_protection = VM_PROT_ALL; @@ -8908,17 +11889,12 @@ vm_map_remap_extract( break; } - if(src_entry->is_sub_map) { - result = KERN_INVALID_ADDRESS; - break; - } - tmp_size = size - mapped_size; if (src_end > src_entry->vme_end) tmp_size -= (src_end - src_entry->vme_end); entry_size = (vm_map_size_t)(src_entry->vme_end - - src_entry->vme_start); + src_entry->vme_start); if(src_entry->is_sub_map) { vm_map_reference(src_entry->object.sub_map); @@ -8940,8 +11916,8 @@ vm_map_remap_extract( assert(!src_entry->needs_copy); } else if (src_entry->needs_copy || object->shadowed || (object->internal && !object->true_share && - !src_entry->is_shared && - object->size > entry_size)) { + !src_entry->is_shared && + object->vo_size > entry_size)) { vm_object_shadow(&src_entry->object.vm_object, &src_entry->offset, @@ -8949,21 +11925,26 @@ vm_map_remap_extract( if (!src_entry->needs_copy && (src_entry->protection & VM_PROT_WRITE)) { - if(map->mapped) { - vm_object_pmap_protect( - src_entry->object.vm_object, - src_entry->offset, - entry_size, - PMAP_NULL, - src_entry->vme_start, - src_entry->protection & - ~VM_PROT_WRITE); + vm_prot_t prot; + + prot = src_entry->protection & ~VM_PROT_WRITE; + + if (override_nx(map, src_entry->alias) && prot) + prot |= VM_PROT_EXECUTE; + + if(map->mapped_in_other_pmaps) { + vm_object_pmap_protect( + src_entry->object.vm_object, + src_entry->offset, + entry_size, + PMAP_NULL, + src_entry->vme_start, + prot); } else { - pmap_protect(vm_map_pmap(map), - src_entry->vme_start, - src_entry->vme_end, - src_entry->protection & - ~VM_PROT_WRITE); + pmap_protect(vm_map_pmap(map), + src_entry->vme_start, + src_entry->vme_end, + prot); } } @@ -8973,10 +11954,9 @@ vm_map_remap_extract( vm_object_lock(object); - object->ref_count++; /* object ref. for new entry */ - VM_OBJ_RES_INCR(object); + vm_object_reference_locked(object); /* object ref. for new entry */ if (object->copy_strategy == - MEMORY_OBJECT_COPY_SYMMETRIC) { + MEMORY_OBJECT_COPY_SYMMETRIC) { object->copy_strategy = MEMORY_OBJECT_COPY_DELAY; } @@ -8985,12 +11965,15 @@ vm_map_remap_extract( offset = src_entry->offset + (src_start - src_entry->vme_start); - new_entry = _vm_map_entry_create(map_header); + 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 */ + 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; @@ -8999,6 +11982,14 @@ vm_map_remap_extract( */ RestartCopy: if (!copy) { + /* + * Cannot allow an entry describing a JIT + * region to be shared across address spaces. + */ + if (src_entry->used_for_jit == TRUE) { + result = KERN_INVALID_ARGUMENT; + break; + } src_entry->is_shared = TRUE; new_entry->is_shared = TRUE; if (!(new_entry->is_sub_map)) @@ -9009,12 +12000,12 @@ vm_map_remap_extract( 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, - (new_entry->vme_end - - new_entry->vme_start), - &src_needs_copy, - &new_entry_needs_copy)) { + vm_object_copy_quickly(&new_entry->object.vm_object, + new_entry->offset, + (new_entry->vme_end - + new_entry->vme_start), + &src_needs_copy, + &new_entry_needs_copy)) { new_entry->needs_copy = new_entry_needs_copy; new_entry->is_shared = FALSE; @@ -9023,15 +12014,21 @@ vm_map_remap_extract( * Handle copy_on_write semantics. */ if (src_needs_copy && !src_entry->needs_copy) { + vm_prot_t prot; + + prot = src_entry->protection & ~VM_PROT_WRITE; + + if (override_nx(map, src_entry->alias) && prot) + prot |= VM_PROT_EXECUTE; + vm_object_pmap_protect(object, offset, entry_size, ((src_entry->is_shared - || map->mapped) ? + || map->mapped_in_other_pmaps) ? PMAP_NULL : map->pmap), src_entry->vme_start, - src_entry->protection & - ~VM_PROT_WRITE); + prot); src_entry->needs_copy = TRUE; } @@ -9059,22 +12056,22 @@ vm_map_remap_extract( if (src_entry->wired_count > 0) { vm_object_lock(object); result = vm_object_copy_slowly( - object, - offset, - entry_size, - THREAD_UNINT, - &new_entry->object.vm_object); + object, + offset, + entry_size, + THREAD_UNINT, + &new_entry->object.vm_object); new_entry->offset = 0; new_entry->needs_copy = FALSE; } else { result = vm_object_copy_strategically( - object, - offset, - entry_size, - &new_entry->object.vm_object, - &new_entry->offset, - &new_entry_needs_copy); + object, + offset, + entry_size, + &new_entry->object.vm_object, + &new_entry->offset, + &new_entry_needs_copy); new_entry->needs_copy = new_entry_needs_copy; } @@ -9117,12 +12114,14 @@ vm_map_remap_extract( } } - _vm_map_entry_link(map_header, + _vm_map_store_entry_link(map_header, map_header->links.prev, new_entry); - *cur_protection &= src_entry->protection; - *max_protection &= src_entry->max_protection; - + /*Protections for submap mapping are irrelevant here*/ + if( !src_entry->is_sub_map ) { + *cur_protection &= src_entry->protection; + *max_protection &= src_entry->max_protection; + } map_address += tmp_size; mapped_size += tmp_size; src_start += tmp_size; @@ -9138,8 +12137,12 @@ vm_map_remap_extract( src_entry != (struct vm_map_entry *)&map_header->links; src_entry = new_entry) { new_entry = src_entry->vme_next; - _vm_map_entry_unlink(map_header, src_entry); - vm_object_deallocate(src_entry->object.vm_object); + _vm_map_store_entry_unlink(map_header, src_entry); + if (src_entry->is_sub_map) { + vm_map_deallocate(src_entry->object.sub_map); + } else { + vm_object_deallocate(src_entry->object.vm_object); + } _vm_map_entry_dispose(map_header, src_entry); } } @@ -9163,7 +12166,7 @@ vm_map_remap( vm_map_address_t *address, vm_map_size_t size, vm_map_offset_t mask, - boolean_t anywhere, + int flags, vm_map_t src_map, vm_map_offset_t memory_address, boolean_t copy, @@ -9173,33 +12176,58 @@ vm_map_remap( { kern_return_t result; vm_map_entry_t entry; - vm_map_entry_t insp_entry; + vm_map_entry_t insp_entry = VM_MAP_ENTRY_NULL; vm_map_entry_t new_entry; struct vm_map_header map_header; + vm_map_offset_t offset_in_mapping; if (target_map == VM_MAP_NULL) return KERN_INVALID_ARGUMENT; switch (inheritance) { - case VM_INHERIT_NONE: - case VM_INHERIT_COPY: - case VM_INHERIT_SHARE: + case VM_INHERIT_NONE: + case VM_INHERIT_COPY: + case VM_INHERIT_SHARE: if (size != 0 && src_map != VM_MAP_NULL) break; /*FALL THRU*/ - default: + default: return KERN_INVALID_ARGUMENT; } - size = vm_map_round_page(size); + /* + * If the user is requesting that we return the address of the + * first byte of the data (rather than the base of the page), + * then we use different rounding semantics: specifically, + * we assume that (memory_address, size) describes a region + * all of whose pages we must cover, rather than a base to be truncated + * down and a size to be added to that base. So we figure out + * the highest page that the requested region includes and make + * sure that the size will cover it. + * + * The key example we're worried about it is of the form: + * + * memory_address = 0x1ff0, size = 0x20 + * + * With the old semantics, we round down the memory_address to 0x1000 + * and round up the size to 0x1000, resulting in our covering *only* + * page 0x1000. With the new semantics, we'd realize that the region covers + * 0x1ff0-0x2010, and compute a size of 0x2000. Thus, we cover both page + * 0x1000 and page 0x2000 in the region we remap. + */ + if ((flags & VM_FLAGS_RETURN_DATA_ADDR) != 0) { + offset_in_mapping = memory_address - vm_map_trunc_page(memory_address, PAGE_MASK); + size = vm_map_round_page(memory_address + size - vm_map_trunc_page(memory_address, PAGE_MASK), PAGE_MASK); + } else { + size = vm_map_round_page(size, PAGE_MASK); + } result = vm_map_remap_extract(src_map, memory_address, - size, copy, &map_header, - cur_protection, - max_protection, - inheritance, - target_map->hdr. - entries_pageable); + size, copy, &map_header, + cur_protection, + max_protection, + inheritance, + target_map->hdr.entries_pageable); if (result != KERN_SUCCESS) { return result; @@ -9209,40 +12237,57 @@ vm_map_remap( * Allocate/check a range of free virtual address * space for the target */ - *address = vm_map_trunc_page(*address); + *address = vm_map_trunc_page(*address, + VM_MAP_PAGE_MASK(target_map)); vm_map_lock(target_map); result = vm_map_remap_range_allocate(target_map, address, size, - mask, anywhere, &insp_entry); + mask, flags, &insp_entry); for (entry = map_header.links.next; entry != (struct vm_map_entry *)&map_header.links; entry = new_entry) { new_entry = entry->vme_next; - _vm_map_entry_unlink(&map_header, entry); + _vm_map_store_entry_unlink(&map_header, entry); if (result == KERN_SUCCESS) { entry->vme_start += *address; entry->vme_end += *address; - vm_map_entry_link(target_map, insp_entry, entry); + assert(!entry->map_aligned); + vm_map_store_entry_link(target_map, insp_entry, entry); insp_entry = entry; } else { if (!entry->is_sub_map) { vm_object_deallocate(entry->object.vm_object); } else { vm_map_deallocate(entry->object.sub_map); - } + } _vm_map_entry_dispose(&map_header, entry); } } + if( target_map->disable_vmentry_reuse == TRUE) { + if( target_map->highest_entry_end < insp_entry->vme_end ){ + target_map->highest_entry_end = insp_entry->vme_end; + } + } + if (result == KERN_SUCCESS) { target_map->size += size; - SAVE_HINT(target_map, insp_entry); + SAVE_HINT_MAP_WRITE(target_map, insp_entry); } 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); + + /* + * If requested, return the address of the data pointed to by the + * request, rather than the base of the resulting page. + */ + if ((flags & VM_FLAGS_RETURN_DATA_ADDR) != 0) { + *address += offset_in_mapping; + } + return result; } @@ -9263,150 +12308,191 @@ vm_map_remap_range_allocate( vm_map_address_t *address, /* IN/OUT */ vm_map_size_t size, vm_map_offset_t mask, - boolean_t anywhere, + int flags, vm_map_entry_t *map_entry) /* OUT */ { - register vm_map_entry_t entry; - register vm_map_offset_t start; - register vm_map_offset_t end; + vm_map_entry_t entry; + vm_map_offset_t start; + vm_map_offset_t end; + kern_return_t kr; - StartAgain: ; +StartAgain: ; - start = *address; + start = *address; - if (anywhere) - { - /* - * Calculate the first possible address. - */ + if (flags & VM_FLAGS_ANYWHERE) + { + /* + * Calculate the first possible address. + */ - if (start < map->min_offset) - start = map->min_offset; - if (start > map->max_offset) - return(KERN_NO_SPACE); + if (start < map->min_offset) + start = map->min_offset; + if (start > map->max_offset) + return(KERN_NO_SPACE); - /* - * Look for the first possible address; - * if there's already something at this - * address, we have to start after it. - */ + /* + * Look for the first possible address; + * if there's already something at this + * address, we have to start after it. + */ - assert(first_free_is_valid(map)); - if (start == map->min_offset) { - if ((entry = map->first_free) != vm_map_to_entry(map)) - start = entry->vme_end; - } else { - vm_map_entry_t tmp_entry; - if (vm_map_lookup_entry(map, start, &tmp_entry)) - start = tmp_entry->vme_end; - entry = tmp_entry; - } + if( map->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; + } 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 - * loop: - */ + /* + * In any case, the "entry" always precedes + * the proposed new region throughout the + * loop: + */ - while (TRUE) { - register vm_map_entry_t next; - - /* - * Find the end of the proposed new region. - * Be sure we didn't go beyond the end, or - * wrap around the address. - */ - - end = ((start + mask) & ~mask); - if (end < start) - return(KERN_NO_SPACE); - start = end; - end += size; - - if ((end > map->max_offset) || (end < start)) { - if (map->wait_for_space) { - if (size <= (map->max_offset - - map->min_offset)) { - assert_wait((event_t) map, THREAD_INTERRUPTIBLE); - vm_map_unlock(map); - thread_block(THREAD_CONTINUE_NULL); - vm_map_lock(map); - goto StartAgain; - } - } + while (TRUE) { + register vm_map_entry_t next; + + /* + * Find the end of the proposed new region. + * Be sure we didn't go beyond the end, or + * wrap around the address. + */ + + end = ((start + mask) & ~mask); + end = vm_map_round_page(end, + VM_MAP_PAGE_MASK(map)); + if (end < start) + return(KERN_NO_SPACE); + start = end; + end += size; + + if ((end > map->max_offset) || (end < start)) { + if (map->wait_for_space) { + if (size <= (map->max_offset - + map->min_offset)) { + assert_wait((event_t) map, THREAD_INTERRUPTIBLE); + vm_map_unlock(map); + thread_block(THREAD_CONTINUE_NULL); + vm_map_lock(map); + goto StartAgain; + } + } - return(KERN_NO_SPACE); - } + return(KERN_NO_SPACE); + } - /* - * If there are no more entries, we must win. - */ + /* + * If there are no more entries, we must win. + */ - next = entry->vme_next; - if (next == vm_map_to_entry(map)) - break; + 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 there is another entry, it must be + * after the end of the potential new region. + */ - if (next->vme_start >= end) - break; + if (next->vme_start >= end) + break; - /* - * Didn't fit -- move to the next entry. - */ + /* + * Didn't fit -- move to the next entry. + */ - entry = next; - start = entry->vme_end; - } - *address = start; - } else { - vm_map_entry_t temp_entry; + entry = next; + start = entry->vme_end; + } + *address = start; + } else { + vm_map_entry_t temp_entry; - /* - * Verify that: - * the address doesn't itself violate - * the mask requirement. - */ + /* + * Verify that: + * the address doesn't itself violate + * the mask requirement. + */ - if ((start & mask) != 0) - return(KERN_NO_SPACE); + if ((start & mask) != 0) + return(KERN_NO_SPACE); - /* - * ... the address is within bounds - */ + /* + * ... the address is within bounds + */ - end = start + size; + end = start + size; - if ((start < map->min_offset) || - (end > map->max_offset) || - (start >= end)) { - return(KERN_INVALID_ADDRESS); - } + if ((start < map->min_offset) || + (end > map->max_offset) || + (start >= end)) { + return(KERN_INVALID_ADDRESS); + } - /* - * ... the starting address isn't allocated - */ + /* + * If we're asked to overwrite whatever was mapped in that + * range, first deallocate that range. + */ + if (flags & VM_FLAGS_OVERWRITE) { + vm_map_t zap_map; - if (vm_map_lookup_entry(map, start, &temp_entry)) - return(KERN_NO_SPACE); + /* + * We use a "zap_map" to avoid having to unlock + * the "map" in vm_map_delete(), which would compromise + * the atomicity of the "deallocate" and then "remap" + * combination. + */ + zap_map = vm_map_create(PMAP_NULL, + start, + end, + map->hdr.entries_pageable); + if (zap_map == VM_MAP_NULL) { + return KERN_RESOURCE_SHORTAGE; + } + vm_map_set_page_shift(zap_map, VM_MAP_PAGE_SHIFT(map)); + + kr = vm_map_delete(map, start, end, + VM_MAP_REMOVE_SAVE_ENTRIES, + zap_map); + if (kr == KERN_SUCCESS) { + vm_map_destroy(zap_map, + VM_MAP_REMOVE_NO_PMAP_CLEANUP); + zap_map = VM_MAP_NULL; + } + } - entry = temp_entry; + /* + * ... the starting address isn't allocated + */ - /* - * ... the next region doesn't overlap the - * end point. - */ + if (vm_map_lookup_entry(map, start, &temp_entry)) + return(KERN_NO_SPACE); + + entry = temp_entry; + + /* + * ... the next region doesn't overlap the + * end point. + */ - if ((entry->vme_next != vm_map_to_entry(map)) && - (entry->vme_next->vme_start < end)) - return(KERN_NO_SPACE); - } - *map_entry = entry; - return(KERN_SUCCESS); + if ((entry->vme_next != vm_map_to_entry(map)) && + (entry->vme_next->vme_start < end)) + return(KERN_NO_SPACE); + } + *map_entry = entry; + return(KERN_SUCCESS); } /* @@ -9528,21 +12614,19 @@ vm_map_read_user( * privilege on the entire address region given. * The entire region must be allocated. */ -boolean_t vm_map_check_protection(map, start, end, protection) - register vm_map_t map; - register vm_map_offset_t start; - register vm_map_offset_t end; - register vm_prot_t protection; +boolean_t +vm_map_check_protection(vm_map_t map, vm_map_offset_t start, + vm_map_offset_t end, vm_prot_t protection) { - register vm_map_entry_t entry; - vm_map_entry_t tmp_entry; + vm_map_entry_t entry; + vm_map_entry_t tmp_entry; vm_map_lock(map); - if (start < vm_map_min(map) || end > vm_map_max(map) || start > end) + if (start < vm_map_min(map) || end > vm_map_max(map) || start > end) { - vm_map_unlock(map); - return (FALSE); + vm_map_unlock(map); + return (FALSE); } if (!vm_map_lookup_entry(map, start, &tmp_entry)) { @@ -9604,22 +12688,28 @@ vm_map_purgable_control( return(KERN_INVALID_ARGUMENT); if (control != VM_PURGABLE_SET_STATE && - control != VM_PURGABLE_GET_STATE) + control != VM_PURGABLE_GET_STATE && + control != VM_PURGABLE_PURGE_ALL) return(KERN_INVALID_ARGUMENT); + if (control == VM_PURGABLE_PURGE_ALL) { + vm_purgeable_object_purge_all(); + return KERN_SUCCESS; + } + if (control == VM_PURGABLE_SET_STATE && - (*state < VM_PURGABLE_STATE_MIN || - *state > VM_PURGABLE_STATE_MAX)) + (((*state & ~(VM_PURGABLE_ALL_MASKS)) != 0) || + ((*state & VM_PURGABLE_STATE_MASK) > VM_PURGABLE_STATE_MASK))) return(KERN_INVALID_ARGUMENT); - vm_map_lock(map); + vm_map_lock_read(map); if (!vm_map_lookup_entry(map, address, &entry) || entry->is_sub_map) { /* * Must pass a valid non-submap address. */ - vm_map_unlock(map); + vm_map_unlock_read(map); return(KERN_INVALID_ADDRESS); } @@ -9627,7 +12717,7 @@ vm_map_purgable_control( /* * Can't apply purgable controls to something you can't write. */ - vm_map_unlock(map); + vm_map_unlock_read(map); return(KERN_PROTECTION_FAILURE); } @@ -9636,24 +12726,26 @@ vm_map_purgable_control( /* * Object must already be present or it can't be purgable. */ - vm_map_unlock(map); + vm_map_unlock_read(map); return KERN_INVALID_ARGUMENT; } vm_object_lock(object); +#if 00 if (entry->offset != 0 || - entry->vme_end - entry->vme_start != object->size) { + entry->vme_end - entry->vme_start != object->vo_size) { /* * Can only apply purgable controls to the whole (existing) * object at once. */ - vm_map_unlock(map); + vm_map_unlock_read(map); vm_object_unlock(object); return KERN_INVALID_ARGUMENT; } +#endif - vm_map_unlock(map); + vm_map_unlock_read(map); kr = vm_object_purgable_control(object, control, state); @@ -9663,58 +12755,209 @@ vm_map_purgable_control( } kern_return_t -vm_map_page_info( - vm_map_t target_map, +vm_map_page_query_internal( + vm_map_t target_map, vm_map_offset_t offset, - int *disposition, - int *ref_count) + int *disposition, + int *ref_count) { - vm_map_entry_t map_entry; - vm_object_t object; - vm_page_t m; + kern_return_t kr; + vm_page_info_basic_data_t info; + mach_msg_type_number_t count; + + count = VM_PAGE_INFO_BASIC_COUNT; + kr = vm_map_page_info(target_map, + offset, + VM_PAGE_INFO_BASIC, + (vm_page_info_t) &info, + &count); + if (kr == KERN_SUCCESS) { + *disposition = info.disposition; + *ref_count = info.ref_count; + } else { + *disposition = 0; + *ref_count = 0; + } -restart_page_query: - *disposition = 0; - *ref_count = 0; - vm_map_lock(target_map); - if(!vm_map_lookup_entry(target_map, offset, &map_entry)) { - vm_map_unlock(target_map); - return KERN_FAILURE; + return kr; +} + +kern_return_t +vm_map_page_info( + vm_map_t map, + vm_map_offset_t offset, + vm_page_info_flavor_t flavor, + vm_page_info_t info, + mach_msg_type_number_t *count) +{ + vm_map_entry_t map_entry; + vm_object_t object; + vm_page_t m; + kern_return_t kr; + kern_return_t retval = KERN_SUCCESS; + boolean_t top_object; + int disposition; + int ref_count; + vm_page_info_basic_t basic_info; + int depth; + vm_map_offset_t offset_in_page; + + switch (flavor) { + case VM_PAGE_INFO_BASIC: + if (*count != VM_PAGE_INFO_BASIC_COUNT) { + /* + * The "vm_page_info_basic_data" structure was not + * properly padded, so allow the size to be off by + * one to maintain backwards binary compatibility... + */ + if (*count != VM_PAGE_INFO_BASIC_COUNT - 1) + return KERN_INVALID_ARGUMENT; + } + break; + default: + return KERN_INVALID_ARGUMENT; } - offset -= map_entry->vme_start; /* adjust to offset within entry */ - offset += map_entry->offset; /* adjust to target object offset */ - if(map_entry->object.vm_object != VM_OBJECT_NULL) { - if(!map_entry->is_sub_map) { - object = map_entry->object.vm_object; - } else { - vm_map_unlock(target_map); - target_map = map_entry->object.sub_map; - goto restart_page_query; + + disposition = 0; + ref_count = 0; + top_object = TRUE; + depth = 0; + + retval = KERN_SUCCESS; + offset_in_page = offset & PAGE_MASK; + offset = vm_map_trunc_page(offset, PAGE_MASK); + + 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; } - } else { - vm_map_unlock(target_map); - return KERN_FAILURE; + /* compute offset from this map entry's start */ + offset -= map_entry->vme_start; + /* compute offset into this map entry's object (or submap) */ + offset += map_entry->offset; + + if (map_entry->is_sub_map) { + vm_map_t sub_map; + + sub_map = map_entry->object.sub_map; + vm_map_lock_read(sub_map); + vm_map_unlock_read(map); + + map = sub_map; + + 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; + } + vm_object_lock(object); - vm_map_unlock(target_map); - while(TRUE) { + vm_map_unlock_read(map); + + /* + * 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); + m = vm_page_lookup(object, offset); + if (m != VM_PAGE_NULL) { - *disposition |= VM_PAGE_QUERY_PAGE_PRESENT; + disposition |= VM_PAGE_QUERY_PAGE_PRESENT; break; } else { - if(object->shadow) { - offset += object->shadow_offset; +#if MACH_PAGEMAP + if (object->existence_map) { + if (vm_external_state_get(object->existence_map, + offset) == + VM_EXTERNAL_STATE_EXISTS) { + /* + * this page has been paged out + */ + disposition |= VM_PAGE_QUERY_PAGE_PAGED_OUT; + break; + } + } else +#endif + if (object->internal && + object->alive && + !object->terminating && + object->pager_ready) { + + if (COMPRESSED_PAGER_IS_ACTIVE || DEFAULT_FREEZER_COMPRESSED_PAGER_IS_ACTIVE) { + if (VM_COMPRESSOR_PAGER_STATE_GET( + object, + offset) + == VM_EXTERNAL_STATE_EXISTS) { + /* the pager has that page */ + disposition |= VM_PAGE_QUERY_PAGE_PAGED_OUT; + break; + } + } else { + memory_object_t pager; + + vm_object_paging_begin(object); + pager = object->pager; + vm_object_unlock(object); + + /* + * Ask the default pager if + * it has this page. + */ + kr = memory_object_data_request( + pager, + offset + object->paging_offset, + 0, /* just poke the pager */ + VM_PROT_READ, + NULL); + + vm_object_lock(object); + vm_object_paging_end(object); + + if (kr == KERN_SUCCESS) { + /* the default pager has it */ + disposition |= VM_PAGE_QUERY_PAGE_PAGED_OUT; + break; + } + } + } + + if (object->shadow != VM_OBJECT_NULL) { + vm_object_t shadow; + + offset += object->vo_shadow_offset; + shadow = object->shadow; + + vm_object_lock(shadow); vm_object_unlock(object); - object = object->shadow; - vm_object_lock(object); - continue; + + object = shadow; + top_object = FALSE; + depth++; + } else { +// if (!object->internal) +// break; +// retval = KERN_FAILURE; +// goto done_with_object; + break; } - vm_object_unlock(object); - return KERN_FAILURE; } } - /* 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 */ @@ -9725,126 +12968,51 @@ restart_page_query: /* but this would under count as only faulted-in mappings would */ /* show up. */ - *ref_count = object->ref_count; + if (top_object == TRUE && object->shadow) + disposition |= VM_PAGE_QUERY_PAGE_COPIED; + + if (! object->internal) + disposition |= VM_PAGE_QUERY_PAGE_EXTERNAL; + + if (m == VM_PAGE_NULL) + goto done_with_object; if (m->fictitious) { - *disposition |= VM_PAGE_QUERY_PAGE_FICTITIOUS; - vm_object_unlock(object); - return KERN_SUCCESS; + 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; + + if (m->reference || pmap_is_referenced(m->phys_page)) + disposition |= VM_PAGE_QUERY_PAGE_REF; - if (m->dirty) - *disposition |= VM_PAGE_QUERY_PAGE_DIRTY; - else if(pmap_is_modified(m->phys_page)) - *disposition |= VM_PAGE_QUERY_PAGE_DIRTY; + if (m->speculative) + disposition |= VM_PAGE_QUERY_PAGE_SPECULATIVE; - if (m->reference) - *disposition |= VM_PAGE_QUERY_PAGE_REF; - else if(pmap_is_referenced(m->phys_page)) - *disposition |= VM_PAGE_QUERY_PAGE_REF; + if (m->cs_validated) + disposition |= VM_PAGE_QUERY_PAGE_CS_VALIDATED; + if (m->cs_tainted) + disposition |= VM_PAGE_QUERY_PAGE_CS_TAINTED; +done_with_object: vm_object_unlock(object); - return KERN_SUCCESS; - -} +done: + 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; + } -/* For a given range, check all map entries. If the entry coresponds to */ -/* the old vm_region/map provided on the call, replace it with the */ -/* corresponding range in the new vm_region/map */ -kern_return_t vm_map_region_replace( - vm_map_t target_map, - ipc_port_t old_region, - ipc_port_t new_region, - vm_map_offset_t start, - vm_map_offset_t end) -{ - vm_named_entry_t old_object; - vm_named_entry_t new_object; - vm_map_t old_submap; - vm_map_t new_submap; - vm_map_offset_t addr; - vm_map_entry_t entry; - int nested_pmap = 0; - - - vm_map_lock(target_map); - old_object = (vm_named_entry_t)old_region->ip_kobject; - new_object = (vm_named_entry_t)new_region->ip_kobject; - if((!old_object->is_sub_map) || (!new_object->is_sub_map)) { - vm_map_unlock(target_map); - return KERN_INVALID_ARGUMENT; - } - old_submap = (vm_map_t)old_object->backing.map; - new_submap = (vm_map_t)new_object->backing.map; - vm_map_lock(old_submap); - if((old_submap->min_offset != new_submap->min_offset) || - (old_submap->max_offset != new_submap->max_offset)) { - vm_map_unlock(old_submap); - vm_map_unlock(target_map); - return KERN_INVALID_ARGUMENT; - } - if(!vm_map_lookup_entry(target_map, start, &entry)) { - /* if the src is not contained, the entry preceeds */ - /* our range */ - addr = entry->vme_start; - if(entry == vm_map_to_entry(target_map)) { - vm_map_unlock(old_submap); - vm_map_unlock(target_map); - return KERN_SUCCESS; - } - } - if ((entry->use_pmap) && - (new_submap->pmap == NULL)) { - new_submap->pmap = pmap_create((vm_map_size_t) 0); - if(new_submap->pmap == PMAP_NULL) { - vm_map_unlock(old_submap); - vm_map_unlock(target_map); - return(KERN_NO_SPACE); - } - } - addr = entry->vme_start; - vm_map_reference(old_submap); - while((entry != vm_map_to_entry(target_map)) && - (entry->vme_start < end)) { - if((entry->is_sub_map) && - (entry->object.sub_map == old_submap)) { - if(entry->use_pmap) { - if((start & 0x0fffffff) || - ((end - start) != 0x10000000)) { - vm_map_unlock(old_submap); - vm_map_deallocate(old_submap); - vm_map_unlock(target_map); - return KERN_INVALID_ARGUMENT; - } - nested_pmap = 1; - } - entry->object.sub_map = new_submap; - vm_map_reference(new_submap); - vm_map_deallocate(old_submap); - } - entry = entry->vme_next; - addr = entry->vme_start; - } - if(nested_pmap) { -#ifndef i386 - pmap_unnest(target_map->pmap, (addr64_t)start); - if(target_map->mapped) { - vm_map_submap_pmap_clean(target_map, - start, end, old_submap, 0); - } - pmap_nest(target_map->pmap, new_submap->pmap, - (addr64_t)start, (addr64_t)start, - (uint64_t)(end - start)); -#endif /* i386 */ - } else { - vm_map_submap_pmap_clean(target_map, - start, end, old_submap, 0); - } - vm_map_unlock(old_submap); - vm_map_deallocate(old_submap); - vm_map_unlock(target_map); - return KERN_SUCCESS; + return retval; } /* @@ -9899,8 +13067,8 @@ vm_map_msync( vm_map_size_t amount_left; vm_object_offset_t offset; boolean_t do_sync_req; - boolean_t modifiable; boolean_t had_hole = FALSE; + memory_object_t pager; if ((sync_flags & VM_SYNC_ASYNCHRONOUS) && (sync_flags & VM_SYNC_SYNCHRONOUS)) @@ -9909,8 +13077,12 @@ vm_map_msync( /* * align address and size on page boundaries */ - size = vm_map_round_page(address + size) - vm_map_trunc_page(address); - address = vm_map_trunc_page(address); + size = (vm_map_round_page(address + size, + VM_MAP_PAGE_MASK(map)) - + vm_map_trunc_page(address, + VM_MAP_PAGE_MASK(map))); + address = vm_map_trunc_page(address, + VM_MAP_PAGE_MASK(map)); if (map == VM_MAP_NULL) return(KERN_INVALID_TASK); @@ -9927,9 +13099,12 @@ vm_map_msync( vm_map_lock(map); if (!vm_map_lookup_entry(map, - vm_map_trunc_page(address), &entry)) { + vm_map_trunc_page( + address, + VM_MAP_PAGE_MASK(map)), + &entry)) { - vm_size_t skip; + vm_map_size_t skip; /* * hole in the address map. @@ -9974,7 +13149,7 @@ vm_map_msync( */ if (amount_left + entry->vme_start + offset > entry->vme_end) { flush_size = entry->vme_end - - (entry->vme_start + offset); + (entry->vme_start + offset); } else { flush_size = amount_left; } @@ -9989,10 +13164,10 @@ vm_map_msync( local_offset = entry->offset; vm_map_unlock(map); if (vm_map_msync( - local_map, - local_offset, - flush_size, - sync_flags) == KERN_INVALID_ADDRESS) { + local_map, + local_offset, + flush_size, + sync_flags) == KERN_INVALID_ADDRESS) { had_hole = TRUE; } continue; @@ -10008,23 +13183,22 @@ vm_map_msync( continue; } offset += entry->offset; - modifiable = (entry->protection & VM_PROT_WRITE) - != VM_PROT_NONE; vm_object_lock(object); if (sync_flags & (VM_SYNC_KILLPAGES | VM_SYNC_DEACTIVATE)) { - boolean_t kill_pages = 0; + int kill_pages = 0; + boolean_t reusable_pages = FALSE; if (sync_flags & VM_SYNC_KILLPAGES) { - if (object->ref_count == 1 && !entry->needs_copy && !object->shadow) + if (object->ref_count == 1 && !object->shadow) kill_pages = 1; else kill_pages = -1; } if (kill_pages != -1) vm_object_deactivate_pages(object, offset, - (vm_object_size_t)flush_size, kill_pages); + (vm_object_size_t)flush_size, kill_pages, reusable_pages); vm_object_unlock(object); vm_map_unlock(map); continue; @@ -10043,26 +13217,36 @@ vm_map_msync( /* * keep reference on the object until syncing is done */ - assert(object->ref_count > 0); - object->ref_count++; - vm_object_res_reference(object); + vm_object_reference_locked(object); vm_object_unlock(object); vm_map_unlock(map); do_sync_req = vm_object_sync(object, - offset, - flush_size, - sync_flags & VM_SYNC_INVALIDATE, - (modifiable && - (sync_flags & VM_SYNC_SYNCHRONOUS || - sync_flags & VM_SYNC_ASYNCHRONOUS)), - sync_flags & VM_SYNC_SYNCHRONOUS); + offset, + flush_size, + sync_flags & VM_SYNC_INVALIDATE, + ((sync_flags & VM_SYNC_SYNCHRONOUS) || + (sync_flags & VM_SYNC_ASYNCHRONOUS)), + sync_flags & VM_SYNC_SYNCHRONOUS); /* * only send a m_o_s if we returned pages or if the entry * is writable (ie dirty pages may have already been sent back) */ - if (!do_sync_req && !modifiable) { + 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; } @@ -10075,7 +13259,24 @@ vm_map_msync( new_msr->length = flush_size; new_msr->object = object; new_msr->flag = VM_MSYNC_SYNCHRONIZING; -re_iterate: + 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 @@ -10099,15 +13300,21 @@ re_iterate: }/* 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( - object->pager, - offset, - flush_size, - sync_flags & ~VM_SYNC_CONTIGUOUS); + pager, + offset, + flush_size, + sync_flags & ~VM_SYNC_CONTIGUOUS); + + vm_object_lock(object); + vm_object_paging_end(object); + vm_object_unlock(object); }/* while */ /* @@ -10136,75 +13343,6 @@ re_iterate: return(KERN_SUCCESS); }/* vm_msync */ -/* Takes existing source and destination sub-maps and clones the contents of */ -/* the source map */ -kern_return_t -vm_region_clone( - ipc_port_t src_region, - ipc_port_t dst_region) -{ - vm_named_entry_t src_object; - vm_named_entry_t dst_object; - vm_map_t src_map; - vm_map_t dst_map; - vm_map_offset_t addr; - vm_map_offset_t max_off; - vm_map_entry_t entry; - vm_map_entry_t new_entry; - vm_map_entry_t insert_point; - - src_object = (vm_named_entry_t)src_region->ip_kobject; - dst_object = (vm_named_entry_t)dst_region->ip_kobject; - if((!src_object->is_sub_map) || (!dst_object->is_sub_map)) { - return KERN_INVALID_ARGUMENT; - } - src_map = (vm_map_t)src_object->backing.map; - dst_map = (vm_map_t)dst_object->backing.map; - /* destination map is assumed to be unavailable to any other */ - /* activity. i.e. it is new */ - vm_map_lock(src_map); - if((src_map->min_offset != dst_map->min_offset) - || (src_map->max_offset != dst_map->max_offset)) { - vm_map_unlock(src_map); - return KERN_INVALID_ARGUMENT; - } - addr = src_map->min_offset; - vm_map_lookup_entry(dst_map, addr, &entry); - if(entry == vm_map_to_entry(dst_map)) { - entry = entry->vme_next; - } - if(entry == vm_map_to_entry(dst_map)) { - max_off = src_map->max_offset; - } else { - max_off = entry->vme_start; - } - vm_map_lookup_entry(src_map, addr, &entry); - if(entry == vm_map_to_entry(src_map)) { - entry = entry->vme_next; - } - vm_map_lookup_entry(dst_map, addr, &insert_point); - while((entry != vm_map_to_entry(src_map)) && - (entry->vme_end <= max_off)) { - addr = entry->vme_start; - new_entry = vm_map_entry_create(dst_map); - vm_map_entry_copy(new_entry, entry); - vm_map_entry_link(dst_map, insert_point, new_entry); - insert_point = new_entry; - if (entry->object.vm_object != VM_OBJECT_NULL) { - if (new_entry->is_sub_map) { - vm_map_reference(new_entry->object.sub_map); - } else { - vm_object_reference( - new_entry->object.vm_object); - } - } - dst_map->size += new_entry->vme_end - new_entry->vme_start; - entry = entry->vme_next; - } - vm_map_unlock(src_map); - return KERN_SUCCESS; -} - /* * Routine: convert_port_entry_to_map * Purpose: @@ -10223,25 +13361,28 @@ convert_port_entry_to_map( { vm_map_t map; vm_named_entry_t named_entry; + uint32_t try_failed_count = 0; if(IP_VALID(port) && (ip_kotype(port) == IKOT_NAMED_ENTRY)) { while(TRUE) { ip_lock(port); if(ip_active(port) && (ip_kotype(port) - == IKOT_NAMED_ENTRY)) { + == IKOT_NAMED_ENTRY)) { named_entry = - (vm_named_entry_t)port->ip_kobject; - if (!(mutex_try(&(named_entry)->Lock))) { + (vm_named_entry_t)port->ip_kobject; + if (!(lck_mtx_try_lock(&(named_entry)->Lock))) { ip_unlock(port); - mutex_pause(); + + try_failed_count++; + mutex_pause(try_failed_count); continue; } named_entry->ref_count++; - mutex_unlock(&(named_entry)->Lock); + lck_mtx_unlock(&(named_entry)->Lock); ip_unlock(port); if ((named_entry->is_sub_map) && - (named_entry->protection - & VM_PROT_WRITE)) { + (named_entry->protection + & VM_PROT_WRITE)) { map = named_entry->backing.map; } else { mach_destroy_memory_entry(port); @@ -10276,42 +13417,35 @@ vm_object_t convert_port_entry_to_object( ipc_port_t port) { - vm_object_t object; + vm_object_t object = VM_OBJECT_NULL; vm_named_entry_t named_entry; - - if(IP_VALID(port) && (ip_kotype(port) == IKOT_NAMED_ENTRY)) { - while(TRUE) { - ip_lock(port); - if(ip_active(port) && (ip_kotype(port) - == IKOT_NAMED_ENTRY)) { - named_entry = - (vm_named_entry_t)port->ip_kobject; - if (!(mutex_try(&(named_entry)->Lock))) { - ip_unlock(port); - mutex_pause(); - continue; - } - named_entry->ref_count++; - mutex_unlock(&(named_entry)->Lock); + uint32_t try_failed_count = 0; + + if (IP_VALID(port) && + (ip_kotype(port) == IKOT_NAMED_ENTRY)) { + try_again: + ip_lock(port); + if (ip_active(port) && + (ip_kotype(port) == IKOT_NAMED_ENTRY)) { + named_entry = (vm_named_entry_t)port->ip_kobject; + if (!(lck_mtx_try_lock(&(named_entry)->Lock))) { ip_unlock(port); - if ((!named_entry->is_sub_map) && - (!named_entry->is_pager) && - (named_entry->protection - & VM_PROT_WRITE)) { - object = named_entry->backing.object; - } else { - mach_destroy_memory_entry(port); - return (vm_object_t)NULL; - } - vm_object_reference(named_entry->backing.object); - mach_destroy_memory_entry(port); - break; + try_failed_count++; + mutex_pause(try_failed_count); + goto try_again; } - else - return (vm_object_t)NULL; + named_entry->ref_count++; + lck_mtx_unlock(&(named_entry)->Lock); + ip_unlock(port); + if (!(named_entry->is_sub_map) && + !(named_entry->is_pager) && + !(named_entry->is_copy) && + (named_entry->protection & VM_PROT_WRITE)) { + object = named_entry->backing.object; + vm_object_reference(object); + } + mach_destroy_memory_entry(port); } - } else { - return (vm_object_t)NULL; } return object; @@ -10343,14 +13477,14 @@ vm_map_reference( if (map == VM_MAP_NULL) return; - mutex_lock(&map->s_lock); + lck_mtx_lock(&map->s_lock); #if TASK_SWAPPER assert(map->res_count > 0); assert(map->ref_count >= map->res_count); map->res_count++; #endif map->ref_count++; - mutex_unlock(&map->s_lock); + lck_mtx_unlock(&map->s_lock); } /* @@ -10369,15 +13503,15 @@ vm_map_deallocate( if (map == VM_MAP_NULL) return; - mutex_lock(&map->s_lock); + lck_mtx_lock(&map->s_lock); ref = --map->ref_count; if (ref > 0) { vm_map_res_deallocate(map); - mutex_unlock(&map->s_lock); + lck_mtx_unlock(&map->s_lock); return; } assert(map->ref_count == 0); - mutex_unlock(&map->s_lock); + lck_mtx_unlock(&map->s_lock); #if TASK_SWAPPER /* @@ -10388,94 +13522,688 @@ vm_map_deallocate( */ #endif - vm_map_destroy(map); + vm_map_destroy(map, VM_MAP_NO_FLAGS); +} + + +void +vm_map_disable_NX(vm_map_t map) +{ + if (map == NULL) + return; + if (map->pmap == NULL) + return; + + pmap_disable_NX(map->pmap); } -#ifdef __PPC__ +void +vm_map_disallow_data_exec(vm_map_t map) +{ + if (map == NULL) + return; + + map->map_disallow_data_exec = TRUE; +} -/* LP64todo - this whole mechanism is temporary. It should be redone when - * the pmap layer can handle 64-bit address spaces. Until then, we trump - * up a map entry for the 64-bit commpage above the map's max_offset. +/* XXX Consider making these constants (VM_MAX_ADDRESS and MACH_VM_MAX_ADDRESS) + * more descriptive. */ -extern vm_map_t com_region_map64; /* the submap for 64-bit commpage */ -SInt32 commpage64s_in_use = 0; - void -vm_map_commpage64( - vm_map_t map ) +vm_map_set_32bit(vm_map_t map) { - vm_map_entry_t entry; - vm_object_t object; - - vm_map_lock(map); - - /* The commpage is necessarily the last entry in the map. - * See if one is already there (not sure if this can happen???) + map->max_offset = (vm_map_offset_t)VM_MAX_ADDRESS; +} + + +void +vm_map_set_64bit(vm_map_t map) +{ + map->max_offset = (vm_map_offset_t)MACH_VM_MAX_ADDRESS; +} + +vm_map_offset_t +vm_compute_max_offset(unsigned is64) +{ + return (is64 ? (vm_map_offset_t)MACH_VM_MAX_ADDRESS : (vm_map_offset_t)VM_MAX_ADDRESS); +} + +uint64_t +vm_map_get_max_aslr_slide_pages(vm_map_t map) +{ + return (1 << (vm_map_is_64bit(map) ? 16 : 8)); +} + +boolean_t +vm_map_is_64bit( + vm_map_t map) +{ + return map->max_offset > ((vm_map_offset_t)VM_MAX_ADDRESS); +} + +boolean_t +vm_map_has_hard_pagezero( + vm_map_t map, + vm_map_offset_t pagezero_size) +{ + /* + * XXX FBDP + * We should lock the VM map (for read) here but we can get away + * with it for now because there can't really be any race condition: + * the VM map's min_offset is changed only when the VM map is created + * and when the zero page is established (when the binary gets loaded), + * and this routine gets called only when the task terminates and the + * VM map is being torn down, and when a new map is created via + * load_machfile()/execve(). */ - entry = vm_map_last_entry(map); - if (entry != vm_map_to_entry(map)) { - if (entry->vme_end >= (vm_map_offset_t)_COMM_PAGE_BASE_ADDRESS) { - vm_map_unlock(map); - return; + 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. + */ +kern_return_t +vm_map_raise_max_offset( + vm_map_t map, + vm_map_offset_t new_max_offset) +{ + kern_return_t ret; + + vm_map_lock(map); + ret = KERN_INVALID_ADDRESS; + + if (new_max_offset >= map->max_offset) { + if (!vm_map_is_64bit(map)) { + if (new_max_offset <= (vm_map_offset_t)VM_MAX_ADDRESS) { + map->max_offset = new_max_offset; + ret = KERN_SUCCESS; + } + } else { + if (new_max_offset <= (vm_map_offset_t)MACH_VM_MAX_ADDRESS) { + map->max_offset = new_max_offset; + ret = KERN_SUCCESS; + } } } + + vm_map_unlock(map); + return ret; +} + + +/* + * Raise a VM map's minimum offset. + * To strictly enforce "page zero" reservation. + */ +kern_return_t +vm_map_raise_min_offset( + vm_map_t map, + vm_map_offset_t new_min_offset) +{ + vm_map_entry_t first_entry; + + new_min_offset = vm_map_round_page(new_min_offset, + VM_MAP_PAGE_MASK(map)); + + vm_map_lock(map); + + if (new_min_offset < map->min_offset) { + /* + * Can't move min_offset backwards, as that would expose + * a part of the address space that was previously, and for + * possibly good reasons, inaccessible. + */ + vm_map_unlock(map); + return KERN_INVALID_ADDRESS; + } + + first_entry = vm_map_first_entry(map); + if (first_entry != vm_map_to_entry(map) && + first_entry->vme_start < new_min_offset) { + /* + * Some memory was already allocated below the new + * minimun offset. It's too late to change it now... + */ + vm_map_unlock(map); + return KERN_NO_SPACE; + } + + map->min_offset = new_min_offset; + + vm_map_unlock(map); + + return KERN_SUCCESS; +} + +/* + * Set the limit on the maximum amount of user wired memory allowed for this map. + * This is basically a copy of the MEMLOCK rlimit value maintained by the BSD side of + * the kernel. The limits are checked in the mach VM side, so we keep a copy so we + * don't have to reach over to the BSD data structures. + */ + +void +vm_map_set_user_wire_limit(vm_map_t map, + vm_size_t limit) +{ + map->user_wire_limit = limit; +} + + +void vm_map_switch_protect(vm_map_t map, + boolean_t val) +{ + vm_map_lock(map); + map->switch_protect=val; + vm_map_unlock(map); +} + +/* + * IOKit has mapped a region into this map; adjust the pmap's ledgers appropriately. + * phys_footprint is a composite limit consisting of iokit + physmem, so we need to + * bump both counters. + */ +void +vm_map_iokit_mapped_region(vm_map_t map, vm_size_t bytes) +{ + pmap_t pmap = vm_map_pmap(map); + + ledger_credit(pmap->ledger, task_ledgers.iokit_mem, bytes); + ledger_credit(pmap->ledger, task_ledgers.phys_footprint, bytes); +} + +void +vm_map_iokit_unmapped_region(vm_map_t map, vm_size_t bytes) +{ + 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); +} + +/* Add (generate) code signature for memory range */ +#if CONFIG_DYNAMIC_CODE_SIGNING +kern_return_t vm_map_sign(vm_map_t map, + vm_map_offset_t start, + vm_map_offset_t end) +{ + vm_map_entry_t entry; + vm_page_t m; + vm_object_t object; + + /* + * 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 + * this situation currently. + */ + vm_map_unlock_read(map); + return(KERN_INVALID_ARGUMENT); + } - entry = vm_map_first_entry(com_region_map64); /* the 64-bit commpage */ object = entry->object.vm_object; - vm_object_reference(object); - - /* We bypass vm_map_enter() because we are adding the entry past the - * map's max_offset. - */ - entry = vm_map_entry_insert( - map, - vm_map_last_entry(map), /* insert after last entry */ - _COMM_PAGE_BASE_ADDRESS, - _COMM_PAGE_BASE_ADDRESS+_COMM_PAGE_AREA_USED, - object, - 0, /* offset */ - FALSE, /* needs_copy */ - FALSE, /* is_shared */ - FALSE, /* in_transition */ - VM_PROT_READ, - VM_PROT_READ, - VM_BEHAVIOR_DEFAULT, - VM_INHERIT_NONE, - 1 ); /* wired_count */ + 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_map_unlock(map); + 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 ); + if (m==VM_PAGE_NULL) { + /* shoud we try to fault a page here? we can probably + * demand it exists and is locked for this request */ + vm_object_unlock(object); + return KERN_FAILURE; + } + /* deal with special page status */ + if (m->busy || + (m->unusual && (m->error || m->restart || m->private || m->absent))) { + vm_object_unlock(object); + return KERN_FAILURE; + } + + /* Page is OK... now "validate" it */ + /* This is the place where we'll call out to create a code + * directory, later */ + m->cs_validated = TRUE; + + /* The page is now "clean" for codesigning purposes. That means + * we don't consider it as modified (wpmapped) anymore. But + * we'll disconnect the page so we note any future modification + * attempts. */ + m->wpmapped = FALSE; + refmod = pmap_disconnect(m->phys_page); + + /* Pull the dirty status from the pmap, since we cleared the + * wpmapped bit */ + if ((refmod & VM_MEM_MODIFIED) && !m->dirty) { + SET_PAGE_DIRTY(m, FALSE); + } + + /* On to the next page */ + start += PAGE_SIZE; + } + vm_object_unlock(object); - OSIncrementAtomic(&commpage64s_in_use); + return KERN_SUCCESS; } +#endif +#if CONFIG_FREEZE -/* LP64todo - remove this! */ - -void -vm_map_remove_commpage64( - vm_map_t 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_entry_t entry; - int deleted = 0; + vm_map_entry_t entry; - while( 1 ) { - vm_map_lock(map); + vm_map_lock_read(map); - entry = vm_map_last_entry(map); - if ((entry == vm_map_to_entry(map)) || - (entry->vme_start < (vm_map_offset_t)_COMM_PAGE_BASE_ADDRESS)) + *purgeable_count = *wired_count = *clean_count = *dirty_count = 0; + *has_shared = FALSE; + + for (entry = vm_map_first_entry(map); + entry != vm_map_to_entry(map); + entry = entry->vme_next) { + unsigned int purgeable, clean, dirty, wired; + boolean_t shared; + + if ((entry->object.vm_object == 0) || + (entry->is_sub_map) || + (entry->object.vm_object->phys_contiguous)) { + continue; + } + + 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; + } + } + } + + vm_map_unlock_read(map); + + return KERN_SUCCESS; +} + +kern_return_t vm_map_freeze( + vm_map_t map, + unsigned int *purgeable_count, + unsigned int *wired_count, + unsigned int *clean_count, + unsigned int *dirty_count, + unsigned int dirty_budget, + boolean_t *has_shared) +{ + vm_map_entry_t entry2 = VM_MAP_ENTRY_NULL; + kern_return_t kr = KERN_SUCCESS; + boolean_t default_freezer_active = TRUE; + + *purgeable_count = *wired_count = *clean_count = *dirty_count = 0; + *has_shared = FALSE; + + /* + * We need the exclusive lock here so that we can + * block any page faults or lookups while we are + * in the middle of freezing this vm map. + */ + vm_map_lock(map); + + if (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(); + } + + if ((kr = default_freezer_handle_init(map->default_freezer_handle)) != KERN_SUCCESS) { + /* + * Can happen if default_freezer_handle passed in is NULL + * Or, a table has already been allocated and associated + * with this handle, i.e. the map is already frozen. + */ + goto done; + } + } + + 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) { + /* 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; - /* clearing the wired count isn't strictly correct */ - entry->wired_count = 0; - vm_map_entry_delete(map,entry); - deleted++; + /* Adjust pageout budget and finish up if reached */ + if (dirty_budget) { + dirty_budget -= dirty; + if (dirty_budget == 0) { + break; + } + } + + if (shared) { + *has_shared = TRUE; + } + } else { + /* + * To the compressor. + */ + if (entry2->object.vm_object->internal == TRUE) { + vm_object_pageout(entry2->object.vm_object); + } + } + } + } + + if (default_freezer_active) { + /* Finally, throw out the pages to swap */ + default_freezer_pageout(map->default_freezer_handle); } + +done: + vm_map_unlock(map); + return kr; +} + +kern_return_t +vm_map_thaw( + vm_map_t map) +{ + kern_return_t kr = KERN_SUCCESS; + + 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; + } + + vm_map_lock(map); + + if (map->default_freezer_handle == NULL) { + /* + * This map is not in a frozen state. + */ + kr = KERN_FAILURE; + goto out; + } + + kr = default_freezer_unpack(map->default_freezer_handle); +out: vm_map_unlock(map); - if (deleted != 0) - OSDecrementAtomic(&commpage64s_in_use); + return kr; +} +#endif + +/* + * vm_map_entry_should_cow_for_true_share: + * + * Determines if the map entry should be clipped and setup for copy-on-write + * to avoid applying "true_share" to a large VM object when only a subset is + * targeted. + * + * For now, we target only the map entries created for the Objective C + * Garbage Collector, which initially have the following properties: + * - alias == VM_MEMORY_MALLOC + * - wired_count == 0 + * - !needs_copy + * and a VM object with: + * - internal + * - copy_strategy == MEMORY_OBJECT_COPY_SYMMETRIC + * - !true_share + * - vo_size == ANON_CHUNK_SIZE + */ +boolean_t +vm_map_entry_should_cow_for_true_share( + vm_map_entry_t entry) +{ + vm_object_t object; + + if (entry->is_sub_map) { + /* entry does not point at a VM object */ + return FALSE; + } + + if (entry->needs_copy) { + /* already set for copy_on_write: done! */ + return FALSE; + } + + if (entry->alias != VM_MEMORY_MALLOC) { + /* not tagged as an ObjectiveC's Garbage Collector entry */ + return FALSE; + } + + if (entry->wired_count) { + /* wired: can't change the map entry... */ + return FALSE; + } + + object = entry->object.vm_object; + + if (object == VM_OBJECT_NULL) { + /* no object yet... */ + return FALSE; + } + + if (!object->internal) { + /* not an internal object */ + return FALSE; + } + + if (object->copy_strategy != MEMORY_OBJECT_COPY_SYMMETRIC) { + /* not the default copy strategy */ + return FALSE; + } + + if (object->true_share) { + /* already true_share: too late to avoid it */ + return FALSE; + } + + if (object->vo_size != ANON_CHUNK_SIZE) { + /* not an object created for the ObjC Garbage Collector */ + return FALSE; + } + + /* + * All the criteria match: we have a large object being targeted for "true_share". + * To limit the adverse side-effects linked with "true_share", tell the caller to + * try and avoid setting up the entire object for "true_share" by clipping the + * targeted range and setting it up for copy-on-write. + */ + return TRUE; +} + +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_trunc_page_mask( + vm_map_offset_t offset, + vm_map_offset_t mask) +{ + return VM_MAP_TRUNC_PAGE(offset, mask); +} + +int +vm_map_page_shift( + vm_map_t map) +{ + return VM_MAP_PAGE_SHIFT(map); +} + +int +vm_map_page_size( + vm_map_t map) +{ + return VM_MAP_PAGE_SIZE(map); +} + +int +vm_map_page_mask( + vm_map_t map) +{ + return VM_MAP_PAGE_MASK(map); +} + +kern_return_t +vm_map_set_page_shift( + vm_map_t map, + int pageshift) +{ + if (map->hdr.nentries != 0) { + /* too late to change page size */ + return KERN_FAILURE; + } + + map->hdr.page_shift = pageshift; + + return KERN_SUCCESS; } -#endif /* __PPC__ */ +kern_return_t +vm_map_query_volatile( + vm_map_t map, + mach_vm_size_t *volatile_virtual_size_p, + mach_vm_size_t *volatile_resident_size_p, + mach_vm_size_t *volatile_pmap_size_p) +{ + mach_vm_size_t volatile_virtual_size; + mach_vm_size_t volatile_resident_count; + mach_vm_size_t volatile_pmap_count; + mach_vm_size_t resident_count; + vm_map_entry_t entry; + vm_object_t object; + + /* map should be locked by caller */ + + volatile_virtual_size = 0; + volatile_resident_count = 0; + volatile_pmap_count = 0; + + for (entry = vm_map_first_entry(map); + entry != vm_map_to_entry(map); + entry = entry->vme_next) { + if (entry->is_sub_map) { + continue; + } + if (! (entry->protection & VM_PROT_WRITE)) { + continue; + } + object = entry->object.vm_object; + if (object == VM_OBJECT_NULL) { + continue; + } + if (object->purgable != VM_PURGABLE_VOLATILE) { + continue; + } + if (entry->offset != 0) { + /* + * If the map entry has been split and the object now + * appears several times in the VM map, we don't want + * to count the object's resident_page_count more than + * once. We count it only for the first one, starting + * at offset 0 and ignore the other VM map entries. + */ + continue; + } + resident_count = object->resident_page_count; + if ((entry->offset / PAGE_SIZE) >= resident_count) { + resident_count = 0; + } else { + resident_count -= (entry->offset / 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); + } + + /* map is still locked on return */ + + *volatile_virtual_size_p = volatile_virtual_size; + *volatile_resident_size_p = volatile_resident_count * PAGE_SIZE; + *volatile_pmap_size_p = volatile_pmap_count * PAGE_SIZE; + + return KERN_SUCCESS; +}