X-Git-Url: https://git.saurik.com/apple/xnu.git/blobdiff_plain/4bd07ac2140668789aa3ee8ec4dde4a3e0a3bba5..d9a64523371fa019c4575bb400cbbc3a50ac9903:/osfmk/vm/vm_fault.c diff --git a/osfmk/vm/vm_fault.c b/osfmk/vm/vm_fault.c index 655c302d2..abbe202ff 100644 --- a/osfmk/vm/vm_fault.c +++ b/osfmk/vm/vm_fault.c @@ -2,7 +2,7 @@ * Copyright (c) 2000-2009 Apple Inc. All rights reserved. * * @APPLE_OSREFERENCE_LICENSE_HEADER_START@ - * + * * This file contains Original Code and/or Modifications of Original Code * as defined in and that are subject to the Apple Public Source License * Version 2.0 (the 'License'). You may not use this file except in @@ -11,10 +11,10 @@ * unlawful or unlicensed copies of an Apple operating system, or to * circumvent, violate, or enable the circumvention or violation of, any * terms of an Apple operating system software license agreement. - * + * * Please obtain a copy of the License at * http://www.opensource.apple.com/apsl/ and read it before using this file. - * + * * The Original Code and all software distributed under the License are * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES, @@ -22,34 +22,34 @@ * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT. * Please see the License for the specific language governing rights and * limitations under the License. - * + * * @APPLE_OSREFERENCE_LICENSE_HEADER_END@ */ /* * @OSF_COPYRIGHT@ */ -/* +/* * Mach Operating System * Copyright (c) 1991,1990,1989,1988,1987 Carnegie Mellon University * All Rights Reserved. - * + * * Permission to use, copy, modify and distribute this software and its * documentation is hereby granted, provided that both the copyright * notice and this permission notice appear in all copies of the * software, derivative works or modified versions, and any portions * thereof, and that both notices appear in supporting documentation. - * + * * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND FOR * ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. - * + * * Carnegie Mellon requests users of this software to return to - * + * * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU * School of Computer Science * Carnegie Mellon University * Pittsburgh PA 15213-3890 - * + * * any improvements or extensions that they make and grant Carnegie Mellon * the rights to redistribute these changes. */ @@ -87,6 +87,7 @@ #include #include #include +#include #include #include @@ -104,8 +105,10 @@ #include #include +#include +#include -#include /* for struct timespec */ +#include #define VM_FAULT_CLASSIFY 0 @@ -114,11 +117,11 @@ unsigned int vm_object_pagein_throttle = 16; /* - * We apply a hard throttle to the demand zero rate of tasks that we believe are running out of control which + * We apply a hard throttle to the demand zero rate of tasks that we believe are running out of control which * kicks in when swap space runs out. 64-bit programs have massive address spaces and can leak enormous amounts * of memory if they're buggy and can run the system completely out of swap space. If this happens, we * impose a hard throttle on them to prevent them from taking the last bit of memory left. This helps - * keep the UI active so that the user has a chance to kill the offending task before the system + * keep the UI active so that the user has a chance to kill the offending task before the system * completely hangs. * * The hard throttle is only applied when the system is nearly completely out of swap space and is only applied @@ -129,16 +132,19 @@ unsigned int vm_object_pagein_throttle = 16; extern void throttle_lowpri_io(int); +extern struct vnode *vnode_pager_lookup_vnode(memory_object_t); + uint64_t vm_hard_throttle_threshold; #define NEED_TO_HARD_THROTTLE_THIS_TASK() (vm_wants_task_throttled(current_task()) || \ - (vm_page_free_count < vm_page_throttle_limit && \ - proc_get_effective_thread_policy(current_thread(), TASK_POLICY_IO) > THROTTLE_LEVEL_THROTTLED)) + ((vm_page_free_count < vm_page_throttle_limit || \ + HARD_THROTTLE_LIMIT_REACHED()) && \ + proc_get_effective_thread_policy(current_thread(), TASK_POLICY_IO) >= THROTTLE_LEVEL_THROTTLED)) -#define HARD_THROTTLE_DELAY 5000 /* 5000 us == 5 ms */ +#define HARD_THROTTLE_DELAY 10000 /* 10000 us == 10 ms */ #define SOFT_THROTTLE_DELAY 200 /* 200 us == .2 ms */ #define VM_PAGE_CREATION_THROTTLE_PERIOD_SECS 6 @@ -152,6 +158,7 @@ static kern_return_t vm_fault_wire_fast( vm_map_t map, vm_map_offset_t va, vm_prot_t prot, + vm_tag_t wire_tag, vm_map_entry_t entry, pmap_t pmap, vm_map_offset_t pmap_addr, @@ -162,6 +169,7 @@ static kern_return_t vm_fault_internal( vm_map_offset_t vaddr, vm_prot_t caller_prot, boolean_t change_wiring, + vm_tag_t wire_tag, int interruptible, pmap_t pmap, vm_map_offset_t pmap_addr, @@ -190,14 +198,35 @@ unsigned long vm_cs_revalidates = 0; unsigned long vm_cs_query_modified = 0; unsigned long vm_cs_validated_dirtied = 0; unsigned long vm_cs_bitmap_validated = 0; +#if PMAP_CS +uint64_t vm_cs_defer_to_pmap_cs = 0; +uint64_t vm_cs_defer_to_pmap_cs_not = 0; +#endif /* PMAP_CS */ void vm_pre_fault(vm_map_offset_t); -extern int not_in_kdp; extern char *kdp_compressor_decompressed_page; extern addr64_t kdp_compressor_decompressed_page_paddr; extern ppnum_t kdp_compressor_decompressed_page_ppnum; +struct vmrtfr { + int vmrtfr_maxi; + int vmrtfr_curi; + int64_t vmrtf_total; + vm_rtfault_record_t *vm_rtf_records; +} vmrtfrs; +#define VMRTF_DEFAULT_BUFSIZE (4096) +#define VMRTF_NUM_RECORDS_DEFAULT (VMRTF_DEFAULT_BUFSIZE / sizeof(vm_rtfault_record_t)) +int vmrtf_num_records = VMRTF_NUM_RECORDS_DEFAULT; + +static void vm_rtfrecord_lock(void); +static void vm_rtfrecord_unlock(void); +static void vm_record_rtfault(thread_t, uint64_t, vm_map_offset_t, int); + +lck_spin_t vm_rtfr_slock; +extern lck_grp_t vm_page_lck_grp_bucket; +extern lck_attr_t vm_page_lck_attr; + /* * Routine: vm_fault_init * Purpose: @@ -225,7 +254,7 @@ vm_fault_init(void) if (PE_parse_boot_argn("vm_compressor", &vm_compressor_temp, sizeof (vm_compressor_temp))) { for ( i = 0; i < VM_PAGER_MAX_MODES; i++) { - if (vm_compressor_temp > 0 && + if (vm_compressor_temp > 0 && ((vm_compressor_temp & ( 1 << i)) == vm_compressor_temp)) { need_default_val = FALSE; vm_compressor_mode = vm_compressor_temp; @@ -234,28 +263,25 @@ vm_fault_init(void) } if (need_default_val) printf("Ignoring \"vm_compressor\" boot arg %d\n", vm_compressor_temp); - } + } if (need_default_val) { /* If no boot arg or incorrect boot arg, try device tree. */ PE_get_default("kern.vm_compressor", &vm_compressor_mode, sizeof(vm_compressor_mode)); } - PE_parse_boot_argn("vm_compressor_threads", &vm_compressor_thread_count, sizeof (vm_compressor_thread_count)); - - if (PE_parse_boot_argn("vm_compressor_immediate", &vm_compressor_temp, sizeof (vm_compressor_temp))) - vm_compressor_immediate_preferred_override = TRUE; - else { - if (PE_get_default("kern.vm_compressor_immediate", &vm_compressor_temp, sizeof(vm_compressor_temp))) - vm_compressor_immediate_preferred_override = TRUE; - } - if (vm_compressor_immediate_preferred_override == TRUE) { - if (vm_compressor_temp) - vm_compressor_immediate_preferred = TRUE; - else - vm_compressor_immediate_preferred = FALSE; - } printf("\"vm_compressor_mode\" is %d\n", vm_compressor_mode); } +void vm_rtfault_record_init(void) { + PE_parse_boot_argn("vm_rtfault_records", &vmrtf_num_records, sizeof(vmrtf_num_records)); + + assert(vmrtf_num_records >= 1); + vmrtf_num_records = MAX(vmrtf_num_records, 1); + size_t kallocsz = vmrtf_num_records * sizeof(vm_rtfault_record_t); + vmrtfrs.vm_rtf_records = kalloc(kallocsz); + bzero(vmrtfrs.vm_rtf_records, kallocsz); + vmrtfrs.vmrtfr_maxi = vmrtf_num_records - 1; + lck_spin_init(&vm_rtfr_slock, &vm_page_lck_grp_bucket, &vm_page_lck_attr); +} /* * Routine: vm_fault_cleanup * Purpose: @@ -272,14 +298,14 @@ vm_fault_init(void) */ void vm_fault_cleanup( - register vm_object_t object, - register vm_page_t top_page) + vm_object_t object, + vm_page_t top_page) { vm_object_paging_end(object); vm_object_unlock(object); if (top_page != VM_PAGE_NULL) { - object = top_page->object; + object = VM_PAGE_OBJECT(top_page); vm_object_lock(object); VM_PAGE_FREE(top_page); @@ -288,30 +314,12 @@ vm_fault_cleanup( } } -#if MACH_CLUSTER_STATS -#define MAXCLUSTERPAGES 16 -struct { - unsigned long pages_in_cluster; - unsigned long pages_at_higher_offsets; - unsigned long pages_at_lower_offsets; -} cluster_stats_in[MAXCLUSTERPAGES]; -#define CLUSTER_STAT(clause) clause -#define CLUSTER_STAT_HIGHER(x) \ - ((cluster_stats_in[(x)].pages_at_higher_offsets)++) -#define CLUSTER_STAT_LOWER(x) \ - ((cluster_stats_in[(x)].pages_at_lower_offsets)++) -#define CLUSTER_STAT_CLUSTER(x) \ - ((cluster_stats_in[(x)].pages_in_cluster)++) -#else /* MACH_CLUSTER_STATS */ -#define CLUSTER_STAT(clause) -#endif /* MACH_CLUSTER_STATS */ - #define ALIGNED(x) (((x) & (PAGE_SIZE_64 - 1)) == 0) boolean_t vm_page_deactivate_behind = TRUE; -/* - * default sizes given VM_BEHAVIOR_DEFAULT reference behavior +/* + * default sizes given VM_BEHAVIOR_DEFAULT reference behavior */ #define VM_DEFAULT_DEACTIVATE_BEHIND_WINDOW 128 #define VM_DEFAULT_DEACTIVATE_BEHIND_CLUSTER 16 /* don't make this too big... */ @@ -536,7 +544,7 @@ vm_fault_deactivate_behind( for (n = 0; n < max_pages_in_run; n++) { m = vm_page_lookup(object, offset + run_offset + (n * pg_offset)); - if (m && !m->laundry && !m->busy && !m->no_cache && !m->throttled && !m->fictitious && !m->absent) { + if (m && !m->vmp_laundry && !m->vmp_busy && !m->vmp_no_cache && (m->vmp_q_state != VM_PAGE_ON_THROTTLED_Q) && !m->vmp_fictitious && !m->vmp_absent) { page_run[pages_in_run++] = m; /* @@ -550,7 +558,7 @@ vm_fault_deactivate_behind( * in the past (TLB caches don't hang around for very long), and of course could just as easily * have happened before we did the deactivate_behind. */ - pmap_clear_refmod_options(m->phys_page, VM_MEM_REFERENCED, PMAP_OPTIONS_NOFLUSH, (void *)NULL); + pmap_clear_refmod_options(VM_PAGE_GET_PHYS_PAGE(m), VM_MEM_REFERENCED, PMAP_OPTIONS_NOFLUSH, (void *)NULL); } } if (pages_in_run) { @@ -587,15 +595,15 @@ vm_page_throttled(boolean_t page_kept) clock_sec_t elapsed_sec; clock_sec_t tv_sec; clock_usec_t tv_usec; - + thread_t thread = current_thread(); - + if (thread->options & TH_OPT_VMPRIV) return (0); if (thread->t_page_creation_throttled) { thread->t_page_creation_throttled = 0; - + if (page_kept == FALSE) goto no_throttle; } @@ -607,9 +615,9 @@ vm_page_throttled(boolean_t page_kept) return (HARD_THROTTLE_DELAY); } - if ((vm_page_free_count < vm_page_throttle_limit || ((COMPRESSED_PAGER_IS_ACTIVE || DEFAULT_FREEZER_COMPRESSED_PAGER_IS_ACTIVE) && SWAPPER_NEEDS_TO_UNTHROTTLE())) && + if ((vm_page_free_count < vm_page_throttle_limit || (VM_CONFIG_COMPRESSOR_IS_PRESENT && SWAPPER_NEEDS_TO_UNTHROTTLE())) && thread->t_page_creation_count > (VM_PAGE_CREATION_THROTTLE_PERIOD_SECS * VM_PAGE_CREATION_THROTTLE_RATE_PER_SEC)) { - + if (vm_page_free_wanted == 0 && vm_page_free_wanted_privileged == 0) { #if (DEVELOPMENT || DEBUG) OSAddAtomic64(1, &vm_page_creation_throttle_avoided); @@ -630,17 +638,17 @@ vm_page_throttled(boolean_t page_kept) * over a long period of time a chance to get out of * the throttled state... we reset the counter and timestamp * so that if it stays under the rate limit for the next second - * it will be back in our good graces... if it exceeds it, it + * it will be back in our good graces... if it exceeds it, it * will remain in the throttled state */ thread->t_page_creation_time = tv_sec; thread->t_page_creation_count = VM_PAGE_CREATION_THROTTLE_RATE_PER_SEC * (VM_PAGE_CREATION_THROTTLE_PERIOD_SECS - 1); } - ++vm_page_throttle_count; + VM_PAGEOUT_DEBUG(vm_page_throttle_count, 1); thread->t_page_creation_throttled = 1; - if ((COMPRESSED_PAGER_IS_ACTIVE || DEFAULT_FREEZER_COMPRESSED_PAGER_IS_ACTIVE) && HARD_THROTTLE_LIMIT_REACHED()) { + if (VM_CONFIG_COMPRESSOR_IS_PRESENT && HARD_THROTTLE_LIMIT_REACHED()) { #if (DEVELOPMENT || DEBUG) thread->t_page_creation_throttled_hard++; OSAddAtomic(1, &vm_page_creation_throttled_hard); @@ -670,10 +678,10 @@ no_throttle: * cleanup is based on being called from vm_fault_page * * object must be locked - * object == m->object + * object == m->vmp_object */ static vm_fault_return_t -vm_fault_check(vm_object_t object, vm_page_t m, vm_page_t first_m, boolean_t interruptible_state, boolean_t page_throttle) +vm_fault_check(vm_object_t object, vm_page_t m, vm_page_t first_m, wait_interrupt_t interruptible_state, boolean_t page_throttle) { int throttle_delay; @@ -694,26 +702,6 @@ vm_fault_check(vm_object_t object, vm_page_t m, vm_page_t first_m, boolean_t int return (VM_FAULT_MEMORY_ERROR); } - if (vm_backing_store_low) { - /* - * are we protecting the system from - * backing store exhaustion. If so - * sleep unless we are privileged. - */ - if (!(current_task()->priv_flags & VM_BACKING_STORE_PRIV)) { - - if (m != VM_PAGE_NULL) - VM_PAGE_FREE(m); - vm_fault_cleanup(object, first_m); - - assert_wait((event_t)&vm_backing_store_low, THREAD_UNINT); - - thread_block(THREAD_CONTINUE_NULL); - thread_interrupt_level(interruptible_state); - - return (VM_FAULT_RETRY); - } - } if (page_throttle == TRUE) { if ((throttle_delay = vm_page_throttled(FALSE))) { /* @@ -745,13 +733,16 @@ vm_fault_check(vm_object_t object, vm_page_t m, vm_page_t first_m, boolean_t int * do the work to zero fill a page and * inject it into the correct paging queue * - * m->object must be locked + * m->vmp_object must be locked * page queue lock must NOT be held */ static int vm_fault_zero_page(vm_page_t m, boolean_t no_zero_fill) { int my_fault = DBG_ZERO_FILL_FAULT; + vm_object_t object; + + object = VM_PAGE_OBJECT(m); /* * This is is a zero-fill page fault... @@ -765,19 +756,19 @@ vm_fault_zero_page(vm_page_t m, boolean_t no_zero_fill) * execution. i.e. it is the responsibility * of higher layers to call for an instruction * sync after changing the contents and before - * sending a program into this area. We + * sending a program into this area. We * choose this approach for performance */ - m->pmapped = TRUE; + m->vmp_pmapped = TRUE; - m->cs_validated = FALSE; - m->cs_tainted = FALSE; - m->cs_nx = FALSE; + m->vmp_cs_validated = FALSE; + m->vmp_cs_tainted = FALSE; + m->vmp_cs_nx = FALSE; if (no_zero_fill == TRUE) { my_fault = DBG_NZF_PAGE_FAULT; - if (m->absent && m->busy) + if (m->vmp_absent && m->vmp_busy) return (my_fault); } else { vm_page_zero_fill(m); @@ -785,30 +776,28 @@ vm_fault_zero_page(vm_page_t m, boolean_t no_zero_fill) VM_STAT_INCR(zero_fill_count); DTRACE_VM2(zfod, int, 1, (uint64_t *), NULL); } - assert(!m->laundry); - assert(m->object != kernel_object); - //assert(m->pageq.next == NULL && m->pageq.prev == NULL); + assert(!m->vmp_laundry); + assert(object != kernel_object); + //assert(m->vmp_pageq.next == 0 && m->vmp_pageq.prev == 0); - if (!VM_DYNAMIC_PAGING_ENABLED(memory_manager_default) && - (m->object->purgable == VM_PURGABLE_DENY || - m->object->purgable == VM_PURGABLE_NONVOLATILE || - m->object->purgable == VM_PURGABLE_VOLATILE )) { + if (!VM_DYNAMIC_PAGING_ENABLED() && + (object->purgable == VM_PURGABLE_DENY || + object->purgable == VM_PURGABLE_NONVOLATILE || + object->purgable == VM_PURGABLE_VOLATILE )) { vm_page_lockspin_queues(); - if (!VM_DYNAMIC_PAGING_ENABLED(memory_manager_default)) { + if (!VM_DYNAMIC_PAGING_ENABLED()) { assert(!VM_PAGE_WIRED(m)); /* * can't be on the pageout queue since we don't * have a pager to try and clean to */ - assert(!m->pageout_queue); - - vm_page_queues_remove(m); + vm_page_queues_remove(m, TRUE); vm_page_check_pageable_safe(m); - queue_enter(&vm_page_queue_throttled, m, vm_page_t, pageq); - m->throttled = TRUE; + vm_page_queue_enter(&vm_page_queue_throttled, m, vm_page_t, vmp_pageq); + m->vmp_q_state = VM_PAGE_ON_THROTTLED_Q; vm_page_throttled_count++; } vm_page_unlock_queues(); @@ -827,7 +816,7 @@ vm_fault_zero_page(vm_page_t m, boolean_t no_zero_fill) * The required permissions for the page is given * in "fault_type". Desired permissions are included * in "protection". - * fault_info is passed along to determine pagein cluster + * fault_info is passed along to determine pagein cluster * limits... it contains the expected reference pattern, * cluster size if available, etc... * @@ -859,7 +848,7 @@ vm_fault_zero_page(vm_page_t m, boolean_t no_zero_fill) * The "result_page" is also left busy. It is not removed * from the pageout queues. * Special Case: - * A return value of VM_FAULT_SUCCESS_NO_PAGE means that the + * A return value of VM_FAULT_SUCCESS_NO_PAGE means that the * fault succeeded but there's no VM page (i.e. the VM object * does not actually hold VM pages, but device memory or * large pages). The object is still locked and we still hold a @@ -887,7 +876,7 @@ vm_fault_page( /* More arguments: */ kern_return_t *error_code, /* code if page is in error */ boolean_t no_zero_fill, /* don't zero fill absent pages */ - boolean_t data_supply, /* treat as data_supply if + boolean_t data_supply, /* treat as data_supply if * it is a write fault and a full * page is provided */ vm_object_fault_info_t fault_info) @@ -902,10 +891,8 @@ vm_fault_page( boolean_t force_fault_retry = FALSE; vm_prot_t access_required = fault_type; vm_prot_t wants_copy_flag; - CLUSTER_STAT(int pages_at_higher_offsets;) - CLUSTER_STAT(int pages_at_lower_offsets;) kern_return_t wait_result; - boolean_t interruptible_state; + wait_interrupt_t interruptible_state; boolean_t data_already_requested = FALSE; vm_behavior_t orig_behavior; vm_size_t orig_cluster_size; @@ -916,50 +903,25 @@ vm_fault_page( int external_state = VM_EXTERNAL_STATE_UNKNOWN; memory_object_t pager; vm_fault_return_t retval; + int grab_options; /* - * MACH page map - an optional optimization where a bit map is maintained - * by the VM subsystem for internal objects to indicate which pages of - * the object currently reside on backing store. This existence map - * duplicates information maintained by the vnode pager. It is - * created at the time of the first pageout against the object, i.e. - * at the same time pager for the object is created. The optimization - * is designed to eliminate pager interaction overhead, if it is - * 'known' that the page does not exist on backing store. - * - * MUST_ASK_PAGER() evaluates to TRUE if the page specified by object/offset is - * either marked as paged out in the existence map for the object or no - * existence map exists for the object. MUST_ASK_PAGER() is one of the - * criteria in the decision to invoke the pager. It is also used as one - * of the criteria to terminate the scan for adjacent pages in a clustered - * pagein operation. Note that MUST_ASK_PAGER() always evaluates to TRUE for - * permanent objects. Note also that if the pager for an internal object - * has not been created, the pager is not invoked regardless of the value - * of MUST_ASK_PAGER() and that clustered pagein scans are only done on an object - * for which a pager has been created. + * MUST_ASK_PAGER() evaluates to TRUE if the page specified by object/offset is + * marked as paged out in the compressor pager or the pager doesn't exist. + * Note also that if the pager for an internal object + * has not been created, the pager is not invoked regardless of the value + * of MUST_ASK_PAGER(). * * PAGED_OUT() evaluates to TRUE if the page specified by the object/offset - * is marked as paged out in the existence map for the object. PAGED_OUT() + * is marked as paged out in the compressor pager. * PAGED_OUT() is used to determine if a page has already been pushed * into a copy object in order to avoid a redundant page out operation. */ -#if MACH_PAGEMAP -#define MUST_ASK_PAGER(o, f, s) \ - ((vm_external_state_get((o)->existence_map, (f)) \ - != VM_EXTERNAL_STATE_ABSENT) && \ - (s = (VM_COMPRESSOR_PAGER_STATE_GET((o), (f)))) \ - != VM_EXTERNAL_STATE_ABSENT) -#define PAGED_OUT(o, f) \ - ((vm_external_state_get((o)->existence_map, (f)) \ - == VM_EXTERNAL_STATE_EXISTS) || \ - (VM_COMPRESSOR_PAGER_STATE_GET((o), (f)) \ - == VM_EXTERNAL_STATE_EXISTS)) -#else /* MACH_PAGEMAP */ #define MUST_ASK_PAGER(o, f, s) \ ((s = VM_COMPRESSOR_PAGER_STATE_GET((o), (f))) != VM_EXTERNAL_STATE_ABSENT) + #define PAGED_OUT(o, f) \ (VM_COMPRESSOR_PAGER_STATE_GET((o), (f)) == VM_EXTERNAL_STATE_EXISTS) -#endif /* MACH_PAGEMAP */ /* * Recovery actions @@ -967,16 +929,16 @@ vm_fault_page( #define RELEASE_PAGE(m) \ MACRO_BEGIN \ PAGE_WAKEUP_DONE(m); \ - if (!m->active && !m->inactive && !m->throttled) { \ - vm_page_lockspin_queues(); \ - if (!m->active && !m->inactive && !m->throttled) { \ - if (COMPRESSED_PAGER_IS_ACTIVE) \ - vm_page_deactivate(m); \ - else \ - vm_page_activate(m); \ - } \ - vm_page_unlock_queues(); \ - } \ + if ( !VM_PAGE_PAGEABLE(m)) { \ + vm_page_lockspin_queues(); \ + if ( !VM_PAGE_PAGEABLE(m)) { \ + if (VM_CONFIG_COMPRESSOR_IS_ACTIVE) \ + vm_page_deactivate(m); \ + else \ + vm_page_activate(m); \ + } \ + vm_page_unlock_queues(); \ + } \ MACRO_END #if TRACEFAULTPAGE @@ -985,7 +947,7 @@ vm_fault_page( interruptible = fault_info->interruptible; interruptible_state = thread_interrupt_level(interruptible); - + /* * INVARIANTS (through entire routine): * @@ -1031,6 +993,14 @@ vm_fault_page( #if TRACEFAULTPAGE dbgTrace(0xBEEF0003, (unsigned int) 0, (unsigned int) 0); /* (TEST/DEBUG) */ #endif + + grab_options = 0; +#if CONFIG_SECLUDED_MEMORY + if (object->can_grab_secluded) { + grab_options |= VM_PAGE_GRAB_SECLUDED; + } +#endif /* CONFIG_SECLUDED_MEMORY */ + if (!object->alive) { /* * object is no longer valid @@ -1094,7 +1064,7 @@ vm_fault_page( #endif if (m != VM_PAGE_NULL) { - if (m->busy) { + if (m->vmp_busy) { /* * The page is being brought in, * wait for it and then retry. @@ -1121,13 +1091,13 @@ vm_fault_page( } continue; } - if (m->laundry) { - m->pageout = FALSE; + if (m->vmp_laundry) { + m->vmp_free_when_done = FALSE; - if (!m->cleaning) + if (!m->vmp_cleaning) vm_pageout_steal_laundry(m, FALSE); } - if (m->phys_page == vm_page_guard_addr) { + if (VM_PAGE_GET_PHYS_PAGE(m) == vm_page_guard_addr) { /* * Guard page: off limits ! */ @@ -1138,7 +1108,7 @@ vm_fault_page( * be just to wire or unwire it. * Let's pretend it succeeded... */ - m->busy = TRUE; + m->vmp_busy = TRUE; *result_page = m; assert(first_m == VM_PAGE_NULL); *top_page = first_m; @@ -1157,7 +1127,7 @@ vm_fault_page( } } - if (m->error) { + if (m->vmp_error) { /* * The page is in error, give up now. */ @@ -1173,7 +1143,7 @@ vm_fault_page( return (VM_FAULT_MEMORY_ERROR); } - if (m->restart) { + if (m->vmp_restart) { /* * The pager wants us to restart * at the top of the chain, @@ -1190,7 +1160,7 @@ vm_fault_page( return (VM_FAULT_RETRY); } - if (m->absent) { + if (m->vmp_absent) { /* * The page isn't busy, but is absent, * therefore it's deemed "unavailable". @@ -1216,7 +1186,7 @@ vm_fault_page( /* * check for any conditions that prevent * us from creating a new zero-fill page - * vm_fault_check will do all of the + * vm_fault_check will do all of the * fault cleanup in the case of an error condition * including resetting the thread_interrupt_level */ @@ -1244,7 +1214,7 @@ vm_fault_page( vm_object_unlock(object); /* - * grab the original page we + * grab the original page we * 'soldered' in place and * retake lock on 'first_object' */ @@ -1260,11 +1230,11 @@ vm_fault_page( * we're going to use the absent page we just found * so convert it to a 'busy' page */ - m->absent = FALSE; - m->busy = TRUE; + m->vmp_absent = FALSE; + m->vmp_busy = TRUE; } if (fault_info->mark_zf_absent && no_zero_fill == TRUE) - m->absent = TRUE; + m->vmp_absent = TRUE; /* * zero-fill the page and put it on * the correct paging queue @@ -1280,14 +1250,11 @@ vm_fault_page( VM_PAGE_FREE(m); } else { first_m = m; - m->absent = FALSE; - m->busy = TRUE; + m->vmp_absent = FALSE; + m->vmp_busy = TRUE; vm_page_lockspin_queues(); - - assert(!m->pageout_queue); - vm_page_queues_remove(m); - + vm_page_queues_remove(m, FALSE); vm_page_unlock_queues(); } XPR(XPR_VM_FAULT, @@ -1305,7 +1272,7 @@ vm_fault_page( vm_object_unlock(object); object = next_object; vm_object_paging_begin(object); - + /* * reset to default type of fault */ @@ -1314,7 +1281,7 @@ vm_fault_page( continue; } } - if ((m->cleaning) + if ((m->vmp_cleaning) && ((object != first_object) || (object->copy != VM_OBJECT_NULL)) && (fault_type & VM_PROT_WRITE)) { /* @@ -1340,14 +1307,14 @@ vm_fault_page( vm_object_reference_locked(object); vm_fault_cleanup(object, first_m); - + counter(c_vm_fault_page_block_backoff_kernel++); vm_object_lock(object); assert(object->ref_count > 0); m = vm_page_lookup(object, offset); - if (m != VM_PAGE_NULL && m->cleaning) { + if (m != VM_PAGE_NULL && m->vmp_cleaning) { PAGE_ASSERT_WAIT(m, interruptible); vm_object_unlock(object); @@ -1364,14 +1331,14 @@ vm_fault_page( return (VM_FAULT_RETRY); } } - if (type_of_fault == NULL && m->speculative && + if (type_of_fault == NULL && (m->vmp_q_state == VM_PAGE_ON_SPECULATIVE_Q) && !(fault_info != NULL && fault_info->stealth)) { /* * If we were passed a non-NULL pointer for * "type_of_fault", than we came from * vm_fault... we'll let it deal with * this condition, since it - * needs to see m->speculative to correctly + * needs to see m->vmp_speculative to correctly * account the pageins, otherwise... * take it off the speculative queue, we'll * let the caller of vm_fault_page deal @@ -1382,36 +1349,13 @@ vm_fault_page( * the page in the speculative queue. */ vm_page_lockspin_queues(); - if (m->speculative) - vm_page_queues_remove(m); + if (m->vmp_q_state == VM_PAGE_ON_SPECULATIVE_Q) + vm_page_queues_remove(m, FALSE); vm_page_unlock_queues(); } + assert(object == VM_PAGE_OBJECT(m)); - if (m->encrypted) { - /* - * ENCRYPTED SWAP: - * the user needs access to a page that we - * encrypted before paging it out. - * Decrypt the page now. - * Keep it busy to prevent anyone from - * accessing it during the decryption. - */ - m->busy = TRUE; - vm_page_decrypt(m, 0); - assert(object == m->object); - assert(m->busy); - PAGE_WAKEUP_DONE(m); - - /* - * Retry from the top, in case - * something changed while we were - * decrypting. - */ - continue; - } - ASSERT_PAGE_DECRYPTED(m); - - if (m->object->code_signed) { + if (object->code_signed) { /* * CODE SIGNING: * We just paged in a page from a signed @@ -1436,13 +1380,13 @@ vm_fault_page( XPR(XPR_VM_FAULT, "vm_f_page: found page obj 0x%X, offset 0x%X, page 0x%X\n", object, offset, m, 0, 0); - assert(!m->busy); - assert(!m->absent); + assert(!m->vmp_busy); + assert(!m->vmp_absent); - m->busy = TRUE; + m->vmp_busy = TRUE; break; } - + /* * we get here when there is no page present in the object at @@ -1451,6 +1395,7 @@ vm_fault_page( * this object can provide the data or we're the top object... * object is locked; m == NULL */ + if (must_be_resident) { if (fault_type == VM_PROT_NONE && object == kernel_object) { @@ -1470,12 +1415,13 @@ vm_fault_page( goto dont_look_for_page; } -#if !MACH_PAGEMAP + /* Don't expect to fault pages into the kernel object. */ + assert(object != kernel_object); + data_supply = FALSE; -#endif /* !MACH_PAGEMAP */ look_for_page = (object->pager_created && (MUST_ASK_PAGER(object, offset, external_state) == TRUE) && !data_supply); - + #if TRACEFAULTPAGE dbgTrace(0xBEEF000C, (unsigned int) look_for_page, (unsigned int) object); /* (TEST/DEBUG) */ #endif @@ -1483,7 +1429,7 @@ vm_fault_page( /* * Allocate a new page for this object/offset pair as a placeholder */ - m = vm_page_grab(); + m = vm_page_grab_options(grab_options); #if TRACEFAULTPAGE dbgTrace(0xBEEF000D, (unsigned int) m, (unsigned int) object); /* (TEST/DEBUG) */ #endif @@ -1586,16 +1532,16 @@ vm_fault_page( return (VM_FAULT_RETRY); } } - if (object->internal && - (COMPRESSED_PAGER_IS_ACTIVE - || DEFAULT_FREEZER_COMPRESSED_PAGER_IS_ACTIVE)) { + if (object->internal) { int compressed_count_delta; + assert(VM_CONFIG_COMPRESSOR_IS_PRESENT); + if (m == VM_PAGE_NULL) { /* * Allocate a new page for this object/offset pair as a placeholder */ - m = vm_page_grab(); + m = vm_page_grab_options(grab_options); #if TRACEFAULTPAGE dbgTrace(0xBEEF000D, (unsigned int) m, (unsigned int) object); /* (TEST/DEBUG) */ #endif @@ -1607,16 +1553,16 @@ vm_fault_page( return (VM_FAULT_MEMORY_SHORTAGE); } - m->absent = TRUE; + m->vmp_absent = TRUE; if (fault_info && fault_info->batch_pmap_op == TRUE) { vm_page_insert_internal(m, object, offset, VM_KERN_MEMORY_NONE, FALSE, TRUE, TRUE, FALSE, NULL); } else { vm_page_insert(m, object, offset); } } - assert(m->busy); - - m->absent = TRUE; + assert(m->vmp_busy); + + m->vmp_absent = TRUE; pager = object->pager; assert(object->paging_in_progress > 0); @@ -1625,7 +1571,7 @@ vm_fault_page( rc = vm_compressor_pager_get( pager, offset + object->paging_offset, - m->phys_page, + VM_PAGE_GET_PHYS_PAGE(m), &my_fault_type, 0, &compressed_count_delta); @@ -1656,9 +1602,9 @@ vm_fault_page( switch (rc) { case KERN_SUCCESS: - m->absent = FALSE; - m->dirty = TRUE; - if ((m->object->wimg_bits & + m->vmp_absent = FALSE; + m->vmp_dirty = TRUE; + if ((object->wimg_bits & VM_WIMG_MASK) != VM_WIMG_USE_DEFAULT) { /* @@ -1668,9 +1614,9 @@ vm_fault_page( * after the decompression. */ pmap_sync_page_attributes_phys( - m->phys_page); + VM_PAGE_GET_PHYS_PAGE(m)); } else { - m->written_by_kernel = TRUE; + m->vmp_written_by_kernel = TRUE; } /* @@ -1681,27 +1627,28 @@ vm_fault_page( * "compressed purgeable" ledger, so * update that now. */ - if ((object->purgable != - VM_PURGABLE_DENY) && - (object->vo_purgeable_owner != + if (((object->purgable != + VM_PURGABLE_DENY) || + object->vo_ledger_tag) && + (object->vo_owner != NULL)) { /* * One less compressed - * purgeable page. + * purgeable/tagged page. */ - vm_purgeable_compressed_update( + vm_object_owner_compressed_update( object, -1); } break; case KERN_MEMORY_FAILURE: - m->unusual = TRUE; - m->error = TRUE; - m->absent = FALSE; + m->vmp_unusual = TRUE; + m->vmp_error = TRUE; + m->vmp_absent = FALSE; break; case KERN_MEMORY_ERROR: - assert(m->absent); + assert(m->vmp_absent); break; default: panic("vm_fault_page(): unexpected " @@ -1715,7 +1662,7 @@ vm_fault_page( goto data_requested; } my_fault_type = DBG_PAGEIN_FAULT; - + if (m != VM_PAGE_NULL) { VM_PAGE_FREE(m); m = VM_PAGE_NULL; @@ -1727,7 +1674,7 @@ vm_fault_page( /* * It's possible someone called vm_object_destroy while we weren't - * holding the object lock. If that has happened, then bail out + * holding the object lock. If that has happened, then bail out * here. */ @@ -1744,6 +1691,10 @@ vm_fault_page( * so we can release the object lock. */ + if (object->object_is_shared_cache) { + set_thread_rwlock_boost(); + } + vm_object_unlock(object); /* @@ -1786,8 +1737,8 @@ vm_fault_page( * the fault w/o having to go through memory_object_data_request again */ assert(first_m != VM_PAGE_NULL); - assert(first_m->object == first_object); - + assert(VM_PAGE_OBJECT(first_m) == first_object); + vm_object_lock(first_object); VM_PAGE_FREE(first_m); vm_object_paging_end(first_object); @@ -1828,6 +1779,10 @@ vm_fault_page( #endif vm_object_lock(object); + if (object->object_is_shared_cache) { + clear_thread_rwlock_boost(); + } + data_requested: if (rc != KERN_SUCCESS) { @@ -1864,7 +1819,7 @@ vm_fault_page( if (m == VM_PAGE_NULL && object->phys_contiguous) { /* * No page here means that the object we - * initially looked up was "physically + * initially looked up was "physically * contiguous" (i.e. device memory). However, * with Virtual VRAM, the object might not * be backed by that device memory anymore, @@ -1894,10 +1849,10 @@ vm_fault_page( } dont_look_for_page: /* - * We get here if the object has no pager, or an existence map + * We get here if the object has no pager, or an existence map * exists and indicates the page isn't present on the pager * or we're unwiring a page. If a pager exists, but there - * is no existence map, then the m->absent case above handles + * is no existence map, then the m->vmp_absent case above handles * the ZF case when the pager can't provide the page */ #if TRACEFAULTPAGE @@ -1931,13 +1886,13 @@ dont_look_for_page: vm_object_lock(object); } m = first_m; - assert(m->object == object); + assert(VM_PAGE_OBJECT(m) == object); first_m = VM_PAGE_NULL; /* * check for any conditions that prevent * us from creating a new zero-fill page - * vm_fault_check will do all of the + * vm_fault_check will do all of the * fault cleanup in the case of an error condition * including resetting the thread_interrupt_level */ @@ -1947,7 +1902,7 @@ dont_look_for_page: return (error); if (m == VM_PAGE_NULL) { - m = vm_page_grab(); + m = vm_page_grab_options(grab_options); if (m == VM_PAGE_NULL) { vm_fault_cleanup(object, VM_PAGE_NULL); @@ -1958,7 +1913,7 @@ dont_look_for_page: vm_page_insert(m, object, offset); } if (fault_info->mark_zf_absent && no_zero_fill == TRUE) - m->absent = TRUE; + m->vmp_absent = TRUE; my_fault = vm_fault_zero_page(m, no_zero_fill); @@ -2007,19 +1962,12 @@ dont_look_for_page: dbgTrace(0xBEEF0015, (unsigned int) object, (unsigned int) m); /* (TEST/DEBUG) */ #endif #if EXTRA_ASSERTIONS - assert(m->busy && !m->absent); + assert(m->vmp_busy && !m->vmp_absent); assert((first_m == VM_PAGE_NULL) || - (first_m->busy && !first_m->absent && - !first_m->active && !first_m->inactive)); + (first_m->vmp_busy && !first_m->vmp_absent && + !first_m->vmp_active && !first_m->vmp_inactive && !first_m->vmp_secluded)); #endif /* EXTRA_ASSERTIONS */ - /* - * ENCRYPTED SWAP: - * If we found a page, we must have decrypted it before we - * get here... - */ - ASSERT_PAGE_DECRYPTED(m); - XPR(XPR_VM_FAULT, "vm_f_page: FOUND obj 0x%X, off 0x%X, page 0x%X, 1_obj 0x%X, 1_m 0x%X\n", object, offset, m, @@ -2045,25 +1993,6 @@ dont_look_for_page: */ assert(!must_be_resident); - /* - * are we protecting the system from - * backing store exhaustion. If so - * sleep unless we are privileged. - */ - if (vm_backing_store_low) { - if (!(current_task()->priv_flags & VM_BACKING_STORE_PRIV)) { - - RELEASE_PAGE(m); - vm_fault_cleanup(object, first_m); - - assert_wait((event_t)&vm_backing_store_low, THREAD_UNINT); - - thread_block(THREAD_CONTINUE_NULL); - thread_interrupt_level(interruptible_state); - - return (VM_FAULT_RETRY); - } - } /* * If we try to collapse first_object at this * point, we may deadlock when we try to get @@ -2083,7 +2012,7 @@ dont_look_for_page: /* * Allocate a page for the copy */ - copy_m = vm_page_grab(); + copy_m = vm_page_grab_options(grab_options); if (copy_m == VM_PAGE_NULL) { RELEASE_PAGE(m); @@ -2112,20 +2041,30 @@ dont_look_for_page: * access to this page, then we could * avoid the pmap_disconnect() call. */ - if (m->pmapped) - pmap_disconnect(m->phys_page); + if (m->vmp_pmapped) + pmap_disconnect(VM_PAGE_GET_PHYS_PAGE(m)); - if (m->clustered) { + if (m->vmp_clustered) { VM_PAGE_COUNT_AS_PAGEIN(m); VM_PAGE_CONSUME_CLUSTERED(m); } - assert(!m->cleaning); + assert(!m->vmp_cleaning); /* * We no longer need the old page or object. */ RELEASE_PAGE(m); + /* + * This check helps with marking the object as having a sequential pattern + * Normally we'll miss doing this below because this fault is about COW to + * the first_object i.e. bring page in from disk, push to object above but + * don't update the file object's sequential pattern. + */ + if (object->internal == FALSE) { + vm_fault_is_sequential(object, offset, fault_info->behavior); + } + vm_object_paging_end(object); vm_object_unlock(object); @@ -2144,12 +2083,12 @@ dont_look_for_page: */ VM_PAGE_FREE(first_m); first_m = VM_PAGE_NULL; - + /* * and replace it with the * page we just copied into */ - assert(copy_m->busy); + assert(copy_m->vmp_busy); vm_page_insert(copy_m, object, offset); SET_PAGE_DIRTY(copy_m, TRUE); @@ -2159,8 +2098,8 @@ dont_look_for_page: * way, let's try to collapse the top object. * But we have to play ugly games with * paging_in_progress to do that... - */ - vm_object_paging_end(object); + */ + vm_object_paging_end(object); vm_object_collapse(object, offset, TRUE); vm_object_paging_begin(object); @@ -2235,7 +2174,7 @@ dont_look_for_page: /* * Page currently exists in the copy object */ - if (copy_m->busy) { + if (copy_m->vmp_busy) { /* * If the page is being brought * in, wait for it and then retry. @@ -2257,13 +2196,8 @@ dont_look_for_page: copy_object->ref_count--; assert(copy_object->ref_count > 0); copy_m = vm_page_lookup(copy_object, copy_offset); - /* - * ENCRYPTED SWAP: - * it's OK if the "copy_m" page is encrypted, - * because we're not moving it nor handling its - * contents. - */ - if (copy_m != VM_PAGE_NULL && copy_m->busy) { + + if (copy_m != VM_PAGE_NULL && copy_m->vmp_busy) { PAGE_ASSERT_WAIT(copy_m, interruptible); vm_object_unlock(copy_object); @@ -2289,32 +2223,7 @@ dont_look_for_page: * for example) or it hasn't been paged out. * (VM_EXTERNAL_STATE_UNKNOWN||VM_EXTERNAL_STATE_ABSENT) * We must copy the page to the copy object. - */ - - if (vm_backing_store_low) { - /* - * we are protecting the system from - * backing store exhaustion. If so - * sleep unless we are privileged. - */ - if (!(current_task()->priv_flags & VM_BACKING_STORE_PRIV)) { - assert_wait((event_t)&vm_backing_store_low, THREAD_UNINT); - - RELEASE_PAGE(m); - VM_OBJ_RES_DECR(copy_object); - vm_object_lock_assert_exclusive(copy_object); - copy_object->ref_count--; - assert(copy_object->ref_count > 0); - - vm_object_unlock(copy_object); - vm_fault_cleanup(object, first_m); - thread_block(THREAD_CONTINUE_NULL); - thread_interrupt_level(interruptible_state); - - return (VM_FAULT_RETRY); - } - } - /* + * * Allocate a page for the copy */ copy_m = vm_page_alloc(copy_object, copy_offset); @@ -2337,17 +2246,17 @@ dont_look_for_page: * Must copy page into copy-object. */ vm_page_copy(m, copy_m); - + /* * If the old page was in use by any users * of the copy-object, it must be removed * from all pmaps. (We can't know which * pmaps use it.) */ - if (m->pmapped) - pmap_disconnect(m->phys_page); + if (m->vmp_pmapped) + pmap_disconnect(VM_PAGE_GET_PHYS_PAGE(m)); - if (m->clustered) { + if (m->vmp_clustered) { VM_PAGE_COUNT_AS_PAGEIN(m); VM_PAGE_CONSUME_CLUSTERED(m); } @@ -2357,87 +2266,21 @@ dont_look_for_page: * option. Else, we use the copy. */ if ((!copy_object->pager_ready) -#if MACH_PAGEMAP - || vm_external_state_get(copy_object->existence_map, copy_offset) == VM_EXTERNAL_STATE_ABSENT -#endif || VM_COMPRESSOR_PAGER_STATE_GET(copy_object, copy_offset) == VM_EXTERNAL_STATE_ABSENT - ) { + ) { vm_page_lockspin_queues(); - assert(!m->cleaning); + assert(!m->vmp_cleaning); vm_page_activate(copy_m); vm_page_unlock_queues(); SET_PAGE_DIRTY(copy_m, TRUE); PAGE_WAKEUP_DONE(copy_m); - } else if (copy_object->internal && - (DEFAULT_PAGER_IS_ACTIVE || DEFAULT_FREEZER_IS_ACTIVE)) { - /* - * For internal objects check with the pager to see - * if the page already exists in the backing store. - * If yes, then we can drop the copy page. If not, - * then we'll activate it, mark it dirty and keep it - * around. - */ - - kern_return_t kr = KERN_SUCCESS; - - memory_object_t copy_pager = copy_object->pager; - assert(copy_pager != MEMORY_OBJECT_NULL); - vm_object_paging_begin(copy_object); - - vm_object_unlock(copy_object); - - kr = memory_object_data_request( - copy_pager, - copy_offset + copy_object->paging_offset, - 0, /* Only query the pager. */ - VM_PROT_READ, - NULL); - - vm_object_lock(copy_object); - - vm_object_paging_end(copy_object); - - /* - * Since we dropped the copy_object's lock, - * check whether we'll have to deallocate - * the hard way. - */ - if ((copy_object->shadow != object) || (copy_object->ref_count == 1)) { - vm_object_unlock(copy_object); - vm_object_deallocate(copy_object); - vm_object_lock(object); - - continue; - } - if (kr == KERN_SUCCESS) { - /* - * The pager has the page. We don't want to overwrite - * that page by sending this one out to the backing store. - * So we drop the copy page. - */ - VM_PAGE_FREE(copy_m); - - } else { - /* - * The pager doesn't have the page. We'll keep this one - * around in the copy object. It might get sent out to - * the backing store under memory pressure. - */ - vm_page_lockspin_queues(); - assert(!m->cleaning); - vm_page_activate(copy_m); - vm_page_unlock_queues(); - - SET_PAGE_DIRTY(copy_m, TRUE); - PAGE_WAKEUP_DONE(copy_m); - } } else { - - assert(copy_m->busy == TRUE); - assert(!m->cleaning); + + assert(copy_m->vmp_busy == TRUE); + assert(!m->vmp_cleaning); /* * dirty is protected by the object lock @@ -2490,8 +2333,8 @@ dont_look_for_page: * wait result]. Can't turn off the page's * busy bit because we're not done with it. */ - if (m->wanted) { - m->wanted = FALSE; + if (m->vmp_wanted) { + m->vmp_wanted = FALSE; thread_wakeup_with_result((event_t) m, THREAD_RESTART); } } @@ -2506,7 +2349,7 @@ dont_look_for_page: copy_object->ref_count--; assert(copy_object->ref_count > 0); - VM_OBJ_RES_DECR(copy_object); + VM_OBJ_RES_DECR(copy_object); vm_object_unlock(copy_object); break; @@ -2521,13 +2364,15 @@ done: object, offset, m, first_m, 0); if (m != VM_PAGE_NULL) { + assert(VM_PAGE_OBJECT(m) == object); + retval = VM_FAULT_SUCCESS; if (my_fault == DBG_PAGEIN_FAULT) { VM_PAGE_COUNT_AS_PAGEIN(m); - if (m->object->internal) + if (object->internal) my_fault = DBG_PAGEIND_FAULT; else my_fault = DBG_PAGEINV_FAULT; @@ -2538,8 +2383,20 @@ done: * state being up to date */ vm_fault_is_sequential(object, offset, fault_info->behavior); + vm_fault_deactivate_behind(object, offset, fault_info->behavior); + } else if (type_of_fault == NULL && my_fault == DBG_CACHE_HIT_FAULT) { + /* + * we weren't called from vm_fault, so handle the + * accounting here for hits in the cache + */ + if (m->vmp_clustered) { + VM_PAGE_COUNT_AS_PAGEIN(m); + VM_PAGE_CONSUME_CLUSTERED(m); + } + vm_fault_is_sequential(object, offset, fault_info->behavior); vm_fault_deactivate_behind(object, offset, fault_info->behavior); + } else if (my_fault == DBG_COMPRESSOR_FAULT || my_fault == DBG_COMPRESSOR_SWAPIN_FAULT) { VM_STAT_INCR(decompressions); @@ -2579,22 +2436,23 @@ backoff: * 3. the page belongs to a code-signed object * 4. the page has not been validated yet or has been mapped for write. */ -#define VM_FAULT_NEED_CS_VALIDATION(pmap, page) \ +#define VM_FAULT_NEED_CS_VALIDATION(pmap, page, page_obj) \ ((pmap) != kernel_pmap /*1*/ && \ - !(page)->cs_tainted /*2*/ && \ - (page)->object->code_signed /*3*/ && \ - (!(page)->cs_validated || (page)->wpmapped /*4*/)) + !(page)->vmp_cs_tainted /*2*/ && \ + (page_obj)->code_signed /*3*/ && \ + (!(page)->vmp_cs_validated || (page)->vmp_wpmapped /*4*/)) /* * page queue lock must NOT be held - * m->object must be locked + * m->vmp_object must be locked * - * NOTE: m->object could be locked "shared" only if we are called + * NOTE: m->vmp_object could be locked "shared" only if we are called * from vm_fault() as part of a soft fault. If so, we must be * careful not to modify the VM object in any way that is not * legal under a shared lock... */ +extern int panic_on_cs_killed; extern int proc_selfpid(void); extern char *proc_name_address(void *p); unsigned long cs_enter_tainted_rejected = 0; @@ -2607,37 +2465,51 @@ vm_fault_enter(vm_page_t m, vm_prot_t caller_prot, boolean_t wired, boolean_t change_wiring, - boolean_t no_cache, - boolean_t cs_bypass, - __unused int user_tag, - int pmap_options, + vm_tag_t wire_tag, + vm_object_fault_info_t fault_info, boolean_t *need_retry, int *type_of_fault) { kern_return_t kr, pe_result; - boolean_t previously_pmapped = m->pmapped; + boolean_t previously_pmapped = m->vmp_pmapped; boolean_t must_disconnect = 0; boolean_t map_is_switched, map_is_switch_protected; + boolean_t cs_violation; int cs_enforcement_enabled; vm_prot_t fault_type; - + vm_object_t object; + boolean_t no_cache = fault_info->no_cache; + boolean_t cs_bypass = fault_info->cs_bypass; + int pmap_options = fault_info->pmap_options; + fault_type = change_wiring ? VM_PROT_NONE : caller_prot; + object = VM_PAGE_OBJECT(m); + + vm_object_lock_assert_held(object); + +#if KASAN + if (pmap == kernel_pmap) { + kasan_notify_address(vaddr, PAGE_SIZE); + } +#endif - vm_object_lock_assert_held(m->object); -#if DEBUG - lck_mtx_assert(&vm_page_queue_lock, LCK_MTX_ASSERT_NOTOWNED); -#endif /* DEBUG */ + LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_NOTOWNED); - if (m->phys_page == vm_page_guard_addr) { - assert(m->fictitious); + if (VM_PAGE_GET_PHYS_PAGE(m) == vm_page_guard_addr) { + assert(m->vmp_fictitious); return KERN_SUCCESS; } if (*type_of_fault == DBG_ZERO_FILL_FAULT) { - vm_object_lock_assert_exclusive(m->object); + vm_object_lock_assert_exclusive(object); - } else if ((fault_type & VM_PROT_WRITE) == 0) { + } else if ((fault_type & VM_PROT_WRITE) == 0 && + (!m->vmp_wpmapped +#if VM_OBJECT_ACCESS_TRACKING + || object->access_tracking +#endif /* VM_OBJECT_ACCESS_TRACKING */ + )) { /* * This is not a "write" fault, so we * might not have taken the object lock @@ -2649,23 +2521,29 @@ vm_fault_enter(vm_page_t m, * soft-fault again if we need write * access later... */ - prot &= ~VM_PROT_WRITE; + + /* This had better not be a JIT page. */ + if (!pmap_has_prot_policy(prot)) { + prot &= ~VM_PROT_WRITE; + } else { + assert(cs_bypass); + } } - if (m->pmapped == FALSE) { + if (m->vmp_pmapped == FALSE) { - if (m->clustered) { + if (m->vmp_clustered) { if (*type_of_fault == DBG_CACHE_HIT_FAULT) { /* * found it in the cache, but this - * is the first fault-in of the page (m->pmapped == FALSE) + * is the first fault-in of the page (m->vmp_pmapped == FALSE) * so it must have come in as part of * a cluster... account 1 pagein against it */ - if (m->object->internal) + if (object->internal) *type_of_fault = DBG_PAGEIND_FAULT; else *type_of_fault = DBG_PAGEINV_FAULT; - + VM_PAGE_COUNT_AS_PAGEIN(m); } VM_PAGE_CONSUME_CLUSTERED(m); @@ -2681,39 +2559,62 @@ vm_fault_enter(vm_page_t m, } /* Validate code signature if necessary. */ - if (VM_FAULT_NEED_CS_VALIDATION(pmap, m)) { - vm_object_lock_assert_exclusive(m->object); + if (!cs_bypass && + VM_FAULT_NEED_CS_VALIDATION(pmap, m, object)) { + vm_object_lock_assert_exclusive(object); - if (m->cs_validated) { + if (m->vmp_cs_validated) { vm_cs_revalidates++; } - /* VM map is locked, so 1 ref will remain on VM object - + /* VM map is locked, so 1 ref will remain on VM object - * so no harm if vm_page_validate_cs drops the object lock */ + +#if PMAP_CS + if (fault_info->pmap_cs_associated && + pmap_cs_enforced(pmap) && + !m->vmp_cs_validated && + !m->vmp_cs_tainted && + !m->vmp_cs_nx && + (prot & VM_PROT_EXECUTE) && + (caller_prot & VM_PROT_EXECUTE)) { + /* + * With pmap_cs, the pmap layer will validate the + * code signature for any executable pmap mapping. + * No need for us to validate this page too: + * in pmap_cs we trust... + */ + vm_cs_defer_to_pmap_cs++; + } else { + vm_cs_defer_to_pmap_cs_not++; + vm_page_validate_cs(m); + } +#else /* PMAP_CS */ vm_page_validate_cs(m); +#endif /* PMAP_CS */ } -#define page_immutable(m,prot) ((m)->cs_validated /*&& ((prot) & VM_PROT_EXECUTE)*/) -#define page_nx(m) ((m)->cs_nx) +#define page_immutable(m,prot) ((m)->vmp_cs_validated /*&& ((prot) & VM_PROT_EXECUTE)*/) +#define page_nx(m) ((m)->vmp_cs_nx) map_is_switched = ((pmap != vm_map_pmap(current_task()->map)) && (pmap == vm_map_pmap(current_thread()->map))); map_is_switch_protected = current_thread()->map->switch_protect; - + /* If the map is switched, and is switch-protected, we must protect - * some pages from being write-faulted: immutable pages because by + * some pages from being write-faulted: immutable pages because by * definition they may not be written, and executable pages because that * would provide a way to inject unsigned code. * If the page is immutable, we can simply return. However, we can't * immediately determine whether a page is executable anywhere. But, * we can disconnect it everywhere and remove the executable protection - * from the current map. We do that below right before we do the + * from the current map. We do that below right before we do the * PMAP_ENTER. */ - cs_enforcement_enabled = cs_enforcement(NULL); + cs_enforcement_enabled = cs_process_enforcement(NULL); - if(cs_enforcement_enabled && map_is_switched && - map_is_switch_protected && page_immutable(m, prot) && + if(cs_enforcement_enabled && map_is_switched && + map_is_switch_protected && page_immutable(m, prot) && (prot & VM_PROT_WRITE)) { return KERN_CODESIGN_ERROR; @@ -2728,35 +2629,59 @@ vm_fault_enter(vm_page_t m, /* A page could be tainted, or pose a risk of being tainted later. * Check whether the receiving process wants it, and make it feel * the consequences (that hapens in cs_invalid_page()). - * For CS Enforcement, two other conditions will - * cause that page to be tainted as well: + * For CS Enforcement, two other conditions will + * cause that page to be tainted as well: * - pmapping an unsigned page executable - this means unsigned code; * - writeable mapping of a validated page - the content of that page * can be changed without the kernel noticing, therefore unsigned * code can be created */ - if (!cs_bypass && - (m->cs_tainted || - (cs_enforcement_enabled && - (/* The page is unsigned and wants to be executable */ - (!m->cs_validated && (prot & VM_PROT_EXECUTE)) || - /* The page should be immutable, but is in danger of being modified - * This is the case where we want policy from the code directory - - * is the page immutable or not? For now we have to assume that - * code pages will be immutable, data pages not. - * We'll assume a page is a code page if it has a code directory - * and we fault for execution. - * That is good enough since if we faulted the code page for - * writing in another map before, it is wpmapped; if we fault - * it for writing in this map later it will also be faulted for executing - * at the same time; and if we fault for writing in another map - * later, we will disconnect it from this pmap so we'll notice - * the change. - */ - (page_immutable(m, prot) && ((prot & VM_PROT_WRITE) || m->wpmapped)) - )) - )) - { + if (cs_bypass) { + /* code-signing is bypassed */ + cs_violation = FALSE; + } else if (m->vmp_cs_tainted) { + /* tainted page */ + cs_violation = TRUE; + } else if (!cs_enforcement_enabled) { + /* no further code-signing enforcement */ + cs_violation = FALSE; + } else if (page_immutable(m, prot) && + ((prot & VM_PROT_WRITE) || + m->vmp_wpmapped)) { + /* + * The page should be immutable, but is in danger of being + * modified. + * This is the case where we want policy from the code + * directory - is the page immutable or not? For now we have + * to assume that code pages will be immutable, data pages not. + * We'll assume a page is a code page if it has a code directory + * and we fault for execution. + * That is good enough since if we faulted the code page for + * writing in another map before, it is wpmapped; if we fault + * it for writing in this map later it will also be faulted for + * executing at the same time; and if we fault for writing in + * another map later, we will disconnect it from this pmap so + * we'll notice the change. + */ + cs_violation = TRUE; + } else if (!m->vmp_cs_validated && + (prot & VM_PROT_EXECUTE) +#if PMAP_CS + /* + * Executable pages will be validated by pmap_cs; + * in pmap_cs we trust... + * If pmap_cs is turned off, this is a code-signing + * violation. + */ + && ! (pmap_cs_enforced(pmap)) +#endif /* PMAP_CS */ + ) { + cs_violation = TRUE; + } else { + cs_violation = FALSE; + } + + if (cs_violation) { /* We will have a tainted page. Have to handle the special case * of a switched map now. If the map is not switched, standard * procedure applies - call cs_invalid_page(). @@ -2764,23 +2689,22 @@ vm_fault_enter(vm_page_t m, * There is no point in invalidating the switching process since * it will not be executing from the map. So we don't call * cs_invalid_page() in that case. */ - boolean_t reject_page; - if(map_is_switched) { + boolean_t reject_page, cs_killed; + if(map_is_switched) { assert(pmap==vm_map_pmap(current_thread()->map)); assert(!(prot & VM_PROT_WRITE) || (map_is_switch_protected == FALSE)); reject_page = FALSE; } else { if (cs_debug > 5) - printf("vm_fault: signed: %s validate: %s tainted: %s wpmapped: %s slid: %s prot: 0x%x\n", - m->object->code_signed ? "yes" : "no", - m->cs_validated ? "yes" : "no", - m->cs_tainted ? "yes" : "no", - m->wpmapped ? "yes" : "no", - m->slid ? "yes" : "no", + printf("vm_fault: signed: %s validate: %s tainted: %s wpmapped: %s prot: 0x%x\n", + object->code_signed ? "yes" : "no", + m->vmp_cs_validated ? "yes" : "no", + m->vmp_cs_tainted ? "yes" : "no", + m->vmp_wpmapped ? "yes" : "no", (int)prot); - reject_page = cs_invalid_page((addr64_t) vaddr); + reject_page = cs_invalid_page((addr64_t) vaddr, &cs_killed); } - + if (reject_page) { /* reject the invalid page: abort the page fault */ int pid; @@ -2793,6 +2717,8 @@ vm_fault_enter(vm_page_t m, boolean_t truncated_path; #define __PATH_MAX 1024 struct timespec mtime, cs_mtime; + int shadow_depth; + os_reason_t codesigning_exit_reason = OS_REASON_NULL; kr = KERN_CODESIGN_ERROR; cs_enter_tainted_rejected++; @@ -2805,13 +2731,15 @@ vm_fault_enter(vm_page_t m, procname = proc_name_address(task->bsd_info); /* get file's VM object */ - file_object = m->object; - file_offset = m->offset; - for (shadow = file_object->shadow; + file_object = object; + file_offset = m->vmp_offset; + for (shadow = file_object->shadow, + shadow_depth = 0; shadow != VM_OBJECT_NULL; - shadow = file_object->shadow) { + shadow = file_object->shadow, + shadow_depth++) { vm_object_lock_shared(shadow); - if (file_object != m->object) { + if (file_object != object) { vm_object_unlock(file_object); } file_offset += file_object->vo_shadow_offset; @@ -2857,8 +2785,8 @@ vm_fault_enter(vm_page_t m, "rejecting invalid page at address 0x%llx " "from offset 0x%llx in file \"%s%s%s\" " "(cs_mtime:%lu.%ld %s mtime:%lu.%ld) " - "(signed:%d validated:%d tainted:%d " - "wpmapped:%d slid:%d)\n", + "(signed:%d validated:%d tainted:%d nx:%d " + "wpmapped:%d dirty:%d depth:%d)\n", pid, procname, (addr64_t) vaddr, file_offset, (pathname ? pathname : ""), @@ -2870,12 +2798,106 @@ vm_fault_enter(vm_page_t m, ? "==" : "!="), mtime.tv_sec, mtime.tv_nsec, - m->object->code_signed, - m->cs_validated, - m->cs_tainted, - m->wpmapped, - m->slid); - if (file_object != m->object) { + object->code_signed, + m->vmp_cs_validated, + m->vmp_cs_tainted, + m->vmp_cs_nx, + m->vmp_wpmapped, + m->vmp_dirty, + shadow_depth); + + /* + * We currently only generate an exit reason if cs_invalid_page directly killed a process. If cs_invalid_page + * did not kill the process (more the case on desktop), vm_fault_enter will not satisfy the fault and whether the + * process dies is dependent on whether there is a signal handler registered for SIGSEGV and how that handler + * will deal with the segmentation fault. + */ + if (cs_killed) { + KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_PROC, BSD_PROC_EXITREASON_CREATE) | DBG_FUNC_NONE, + pid, OS_REASON_CODESIGNING, CODESIGNING_EXIT_REASON_INVALID_PAGE, 0, 0); + + codesigning_exit_reason = os_reason_create(OS_REASON_CODESIGNING, CODESIGNING_EXIT_REASON_INVALID_PAGE); + if (codesigning_exit_reason == NULL) { + printf("vm_fault_enter: failed to allocate codesigning exit reason\n"); + } else { + mach_vm_address_t data_addr = 0; + struct codesigning_exit_reason_info *ceri = NULL; + uint32_t reason_buffer_size_estimate = kcdata_estimate_required_buffer_size(1, sizeof(*ceri)); + + if (os_reason_alloc_buffer_noblock(codesigning_exit_reason, reason_buffer_size_estimate)) { + printf("vm_fault_enter: failed to allocate buffer for codesigning exit reason\n"); + } else { + if (KERN_SUCCESS == kcdata_get_memory_addr(&codesigning_exit_reason->osr_kcd_descriptor, + EXIT_REASON_CODESIGNING_INFO, sizeof(*ceri), &data_addr)) { + ceri = (struct codesigning_exit_reason_info *)data_addr; + static_assert(__PATH_MAX == sizeof(ceri->ceri_pathname)); + + ceri->ceri_virt_addr = vaddr; + ceri->ceri_file_offset = file_offset; + if (pathname) + strncpy((char *)&ceri->ceri_pathname, pathname, sizeof(ceri->ceri_pathname)); + else + ceri->ceri_pathname[0] = '\0'; + if (filename) + strncpy((char *)&ceri->ceri_filename, filename, sizeof(ceri->ceri_filename)); + else + ceri->ceri_filename[0] = '\0'; + ceri->ceri_path_truncated = (truncated_path); + ceri->ceri_codesig_modtime_secs = cs_mtime.tv_sec; + ceri->ceri_codesig_modtime_nsecs = cs_mtime.tv_nsec; + ceri->ceri_page_modtime_secs = mtime.tv_sec; + ceri->ceri_page_modtime_nsecs = mtime.tv_nsec; + ceri->ceri_object_codesigned = (object->code_signed); + ceri->ceri_page_codesig_validated = (m->vmp_cs_validated); + ceri->ceri_page_codesig_tainted = (m->vmp_cs_tainted); + ceri->ceri_page_codesig_nx = (m->vmp_cs_nx); + ceri->ceri_page_wpmapped = (m->vmp_wpmapped); + ceri->ceri_page_slid = 0; + ceri->ceri_page_dirty = (m->vmp_dirty); + ceri->ceri_page_shadow_depth = shadow_depth; + } else { +#if DEBUG || DEVELOPMENT + panic("vm_fault_enter: failed to allocate kcdata for codesigning exit reason"); +#else + printf("vm_fault_enter: failed to allocate kcdata for codesigning exit reason\n"); +#endif /* DEBUG || DEVELOPMENT */ + /* Free the buffer */ + os_reason_alloc_buffer_noblock(codesigning_exit_reason, 0); + } + } + } + + set_thread_exit_reason(current_thread(), codesigning_exit_reason, FALSE); + } + if (panic_on_cs_killed && + object->object_is_shared_cache) { + panic("CODE SIGNING: process %d[%s]: " + "rejecting invalid page at address 0x%llx " + "from offset 0x%llx in file \"%s%s%s\" " + "(cs_mtime:%lu.%ld %s mtime:%lu.%ld) " + "(signed:%d validated:%d tainted:%d nx:%d" + "wpmapped:%d dirty:%d depth:%d)\n", + pid, procname, (addr64_t) vaddr, + file_offset, + (pathname ? pathname : ""), + (truncated_path ? "/.../" : ""), + (truncated_path ? filename : ""), + cs_mtime.tv_sec, cs_mtime.tv_nsec, + ((cs_mtime.tv_sec == mtime.tv_sec && + cs_mtime.tv_nsec == mtime.tv_nsec) + ? "==" + : "!="), + mtime.tv_sec, mtime.tv_nsec, + object->code_signed, + m->vmp_cs_validated, + m->vmp_cs_tainted, + m->vmp_cs_nx, + m->vmp_wpmapped, + m->vmp_dirty, + shadow_depth); + } + + if (file_object != object) { vm_object_unlock(file_object); } if (pathname_len != 0) { @@ -2886,11 +2908,12 @@ vm_fault_enter(vm_page_t m, } else { /* proceed with the invalid page */ kr = KERN_SUCCESS; - if (!m->cs_validated) { + if (!m->vmp_cs_validated && + !object->code_signed) { /* - * This page has not been validated, so it - * must not belong to a code-signed object - * and should not be forcefully considered + * This page has not been (fully) validated but + * does not belong to a code-signed object + * so it should not be forcefully considered * as tainted. * We're just concerned about it here because * we've been asked to "execute" it but that @@ -2905,7 +2928,6 @@ vm_fault_enter(vm_page_t m, * even though they're just reading it and not * executing from it. */ - assert(!m->object->code_signed); } else { /* * Page might have been tainted before or not; @@ -2915,8 +2937,8 @@ vm_fault_enter(vm_page_t m, * through that code path for re-consideration * of the validity of that page. */ - must_disconnect = !m->cs_tainted; - m->cs_tainted = TRUE; + must_disconnect = !m->vmp_cs_tainted; + m->vmp_cs_tainted = TRUE; } cs_enter_tainted_accepted++; } @@ -2932,7 +2954,7 @@ vm_fault_enter(vm_page_t m, } #endif } - + } else { /* proceed with the valid page */ kr = KERN_SUCCESS; @@ -2959,19 +2981,23 @@ MACRO_END * the page queues. Change wiring * case is obvious. */ - assert(m->compressor || m->object != compressor_object); - if (m->compressor) { + assert((m->vmp_q_state == VM_PAGE_USED_BY_COMPRESSOR) || object != compressor_object); + +#if CONFIG_BACKGROUND_QUEUE + vm_page_update_background_state(m); +#endif + if (m->vmp_q_state == VM_PAGE_USED_BY_COMPRESSOR) { /* * Compressor pages are neither wired * nor pageable and should never change. */ - assert(m->object == compressor_object); + assert(object == compressor_object); } else if (change_wiring) { __VM_PAGE_LOCKSPIN_QUEUES_IF_NEEDED(); if (wired) { if (kr == KERN_SUCCESS) { - vm_page_wire(m, VM_PROT_MEMORY_TAG(caller_prot), TRUE); + vm_page_wire(m, wire_tag, TRUE); } } else { vm_page_unwire(m, TRUE); @@ -2979,24 +3005,33 @@ MACRO_END /* we keep the page queues lock, if we need it later */ } else { + if (object->internal == TRUE) { + /* + * don't allow anonymous pages on + * the speculative queues + */ + no_cache = FALSE; + } if (kr != KERN_SUCCESS) { __VM_PAGE_LOCKSPIN_QUEUES_IF_NEEDED(); vm_page_deactivate(m); /* we keep the page queues lock, if we need it later */ - } else if (((!m->active && !m->inactive) || - m->clean_queue || - no_cache) && - !VM_PAGE_WIRED(m) && !m->throttled) { + } else if (((m->vmp_q_state == VM_PAGE_NOT_ON_Q) || + (m->vmp_q_state == VM_PAGE_ON_SPECULATIVE_Q) || + (m->vmp_q_state == VM_PAGE_ON_INACTIVE_CLEANED_Q) || + ((m->vmp_q_state != VM_PAGE_ON_THROTTLED_Q) && no_cache)) && + !VM_PAGE_WIRED(m)) { if (vm_page_local_q && - !no_cache && (*type_of_fault == DBG_COW_FAULT || *type_of_fault == DBG_ZERO_FILL_FAULT) ) { struct vpl *lq; uint32_t lid; + assert(m->vmp_q_state == VM_PAGE_NOT_ON_Q); + __VM_PAGE_UNLOCK_QUEUES_IF_NEEDED(); - vm_object_lock_assert_exclusive(m->object); + vm_object_lock_assert_exclusive(object); /* * we got a local queue to stuff this @@ -3009,7 +3044,7 @@ MACRO_END * we'll use the current cpu number to * select the queue note that we don't * need to disable preemption... we're - * going to behind the local queue's + * going to be behind the local queue's * lock to do the real work */ lid = cpu_number(); @@ -3019,13 +3054,13 @@ MACRO_END VPL_LOCK(&lq->vpl_lock); vm_page_check_pageable_safe(m); - queue_enter(&lq->vpl_queue, m, - vm_page_t, pageq); - m->local = TRUE; - m->local_id = lid; + vm_page_queue_enter(&lq->vpl_queue, m, + vm_page_t, vmp_pageq); + m->vmp_q_state = VM_PAGE_ON_ACTIVE_LOCAL_Q; + m->vmp_local_id = lid; lq->vpl_count++; - - if (m->object->internal) + + if (object->internal) lq->vpl_internal_count++; else lq->vpl_external_count++; @@ -3061,16 +3096,15 @@ MACRO_END * page queue lock */ if (!VM_PAGE_WIRED(m)) { - if (m->clean_queue) { - vm_page_queues_remove(m); + if (m->vmp_q_state == VM_PAGE_ON_INACTIVE_CLEANED_Q) { + vm_page_queues_remove(m, FALSE); - vm_pageout_cleaned_reactivated++; - vm_pageout_cleaned_fault_reactivated++; + VM_PAGEOUT_DEBUG(vm_pageout_cleaned_reactivated, 1); + VM_PAGEOUT_DEBUG(vm_pageout_cleaned_fault_reactivated, 1); } - if ((!m->active && - !m->inactive) || - no_cache) { + if ( !VM_PAGE_ACTIVE_OR_INACTIVE(m) || + no_cache) { /* * If this is a no_cache mapping * and the page has never been @@ -3081,20 +3115,18 @@ MACRO_END * that they can be readily * recycled if free memory runs * low. Otherwise the page is - * activated as normal. + * activated as normal. */ if (no_cache && (!previously_pmapped || - m->no_cache)) { - m->no_cache = TRUE; + m->vmp_no_cache)) { + m->vmp_no_cache = TRUE; - if (!m->speculative) + if (m->vmp_q_state != VM_PAGE_ON_SPECULATIVE_Q) vm_page_speculate(m, FALSE); - } else if (!m->active && - !m->inactive) { - + } else if ( !VM_PAGE_ACTIVE_OR_INACTIVE(m)) { vm_page_activate(m); } } @@ -3114,36 +3146,38 @@ MACRO_END * now so those processes can take note. */ if (kr == KERN_SUCCESS) { - /* * NOTE: we may only hold the vm_object lock SHARED - * at this point, so we need the phys_page lock to + * at this point, so we need the phys_page lock to * properly serialize updating the pmapped and * xpmapped bits */ - if ((prot & VM_PROT_EXECUTE) && !m->xpmapped) { + if ((prot & VM_PROT_EXECUTE) && !m->vmp_xpmapped) { + ppnum_t phys_page = VM_PAGE_GET_PHYS_PAGE(m); - pmap_lock_phys_page(m->phys_page); + pmap_lock_phys_page(phys_page); /* * go ahead and take the opportunity * to set 'pmapped' here so that we don't * need to grab this lock a 2nd time * just below */ - m->pmapped = TRUE; - - if (!m->xpmapped) { + m->vmp_pmapped = TRUE; + + if (!m->vmp_xpmapped) { - m->xpmapped = TRUE; + m->vmp_xpmapped = TRUE; - pmap_unlock_phys_page(m->phys_page); + pmap_unlock_phys_page(phys_page); - if (!m->object->internal) + if (!object->internal) OSAddAtomic(1, &vm_page_xpmapped_external_count); - if ((COMPRESSED_PAGER_IS_ACTIVE) && - m->object->internal && - m->object->pager != NULL) { +#if defined(__arm__) || defined(__arm64__) + pmap_sync_page_data_phys(phys_page); +#else + if (object->internal && + object->pager != NULL) { /* * This page could have been * uncompressed by the @@ -3155,68 +3189,81 @@ MACRO_END * make sure the icache is in * sync. */ - pmap_sync_page_data_phys(m->phys_page); + assert(VM_CONFIG_COMPRESSOR_IS_PRESENT); + pmap_sync_page_data_phys(phys_page); } +#endif } else - pmap_unlock_phys_page(m->phys_page); + pmap_unlock_phys_page(phys_page); } else { - if (m->pmapped == FALSE) { - pmap_lock_phys_page(m->phys_page); - m->pmapped = TRUE; - pmap_unlock_phys_page(m->phys_page); - } - } - if (vm_page_is_slideable(m)) { - boolean_t was_busy = m->busy; - - vm_object_lock_assert_exclusive(m->object); + if (m->vmp_pmapped == FALSE) { + ppnum_t phys_page = VM_PAGE_GET_PHYS_PAGE(m); - m->busy = TRUE; - kr = vm_page_slide(m, 0); - assert(m->busy); - if(!was_busy) { - PAGE_WAKEUP_DONE(m); - } - if (kr != KERN_SUCCESS) { - /* - * This page has not been slid correctly, - * do not do the pmap_enter() ! - * Let vm_fault_enter() return the error - * so the caller can fail the fault. - */ - goto after_the_pmap_enter; + pmap_lock_phys_page(phys_page); + m->vmp_pmapped = TRUE; + pmap_unlock_phys_page(phys_page); } } if (fault_type & VM_PROT_WRITE) { - if (m->wpmapped == FALSE) { - vm_object_lock_assert_exclusive(m->object); - if (!m->object->internal) - task_update_logical_writes(current_task(), PAGE_SIZE, TASK_WRITE_DEFERRED); - m->wpmapped = TRUE; + if (m->vmp_wpmapped == FALSE) { + vm_object_lock_assert_exclusive(object); + if (!object->internal && object->pager) { + task_update_logical_writes(current_task(), PAGE_SIZE, TASK_WRITE_DEFERRED, vnode_pager_lookup_vnode(object->pager)); + } + m->vmp_wpmapped = TRUE; } if (must_disconnect) { /* - * We can only get here + * We can only get here * because of the CSE logic */ assert(cs_enforcement_enabled); - pmap_disconnect(m->phys_page); - /* + pmap_disconnect(VM_PAGE_GET_PHYS_PAGE(m)); + /* * If we are faulting for a write, we can clear * the execute bit - that will ensure the page is * checked again before being executable, which * protects against a map switch. * This only happens the first time the page - * gets tainted, so we won't get stuck here + * gets tainted, so we won't get stuck here * to make an already writeable page executable. */ if (!cs_bypass){ + assert(!pmap_has_prot_policy(prot)); prot &= ~VM_PROT_EXECUTE; } } } + assert(VM_PAGE_OBJECT(m) == object); + +#if VM_OBJECT_ACCESS_TRACKING + if (object->access_tracking) { + DTRACE_VM2(access_tracking, vm_map_offset_t, vaddr, int, fault_type); + if (fault_type & VM_PROT_WRITE) { + object->access_tracking_writes++; + vm_object_access_tracking_writes++; + } else { + object->access_tracking_reads++; + vm_object_access_tracking_reads++; + } + } +#endif /* VM_OBJECT_ACCESS_TRACKING */ + +#if PMAP_CS + /* + * If CS enforcement is on, we don't ask for an executable page if the + * fault does not call for execution, because that can fail in + * situations where the caller only actually wanted read access. + * However, it may be better to instead retry without execute on + * failure, or pass a flag into pmap_enter to do the right thing. + */ + // TODO: maybe do something better than masking out VM_PROT_EXECUTE on non-execute faults + if (pmap_cs_enforced(pmap) && !(caller_prot & VM_PROT_EXECUTE)) { + prot &= ~VM_PROT_EXECUTE; + } +#endif /* Prevent a deadlock by not * holding the object lock if we need to wait for a page in @@ -3225,6 +3272,19 @@ MACRO_END wired, pmap_options | PMAP_OPTIONS_NOWAIT, pe_result); +#if __x86_64__ + if (pe_result == KERN_INVALID_ARGUMENT && + pmap == PMAP_NULL && + wired) { + /* + * Wiring a page in a pmap-less VM map: + * VMware's "vmmon" kernel extension does this + * to grab pages. + * Let it proceed even though the PMAP_ENTER() failed. + */ + pe_result = KERN_SUCCESS; + } +#endif /* __x86_64__ */ if(pe_result == KERN_RESOURCE_SHORTAGE) { @@ -3234,10 +3294,10 @@ MACRO_END * on the top-object in this chain... we can't just drop * the lock on the object we're inserting the page into * and recall the PMAP_ENTER since we can still cause - * a deadlock if one of the critical paths tries to + * a deadlock if one of the critical paths tries to * acquire the lock on the top-object and we're blocked * in PMAP_ENTER waiting for memory... our only recourse - * is to deal with it at a higher level where we can + * is to deal with it at a higher level where we can * drop both locks. */ *need_retry = TRUE; @@ -3246,31 +3306,35 @@ MACRO_END } /* The nonblocking version of pmap_enter did not succeed. * and we don't need to drop other locks and retry - * at the level above us, so + * at the level above us, so * use the blocking version instead. Requires marking * the page busy and unlocking the object */ - boolean_t was_busy = m->busy; + boolean_t was_busy = m->vmp_busy; - vm_object_lock_assert_exclusive(m->object); + vm_object_lock_assert_exclusive(object); + + m->vmp_busy = TRUE; + vm_object_unlock(object); - m->busy = TRUE; - vm_object_unlock(m->object); - PMAP_ENTER_OPTIONS(pmap, vaddr, m, prot, fault_type, 0, wired, pmap_options, pe_result); - + + assert(VM_PAGE_OBJECT(m) == object); + /* Take the object lock again. */ - vm_object_lock(m->object); - + vm_object_lock(object); + /* If the page was busy, someone else will wake it up. * Otherwise, we have to do it now. */ - assert(m->busy); + assert(m->vmp_busy); if(!was_busy) { PAGE_WAKEUP_DONE(m); } vm_pmap_enter_blocked++; } + + kr = pe_result; } after_the_pmap_enter: @@ -3282,13 +3346,14 @@ vm_pre_fault(vm_map_offset_t vaddr) { if (pmap_find_phys(current_map()->pmap, vaddr) == 0) { - vm_fault(current_map(), /* map */ - vaddr, /* vaddr */ - VM_PROT_READ, /* fault_type */ - FALSE, /* change_wiring */ - THREAD_UNINT, /* interruptible */ - NULL, /* caller_pmap */ - 0 /* caller_pmap_addr */); + vm_fault(current_map(), /* map */ + vaddr, /* vaddr */ + VM_PROT_READ, /* fault_type */ + FALSE, /* change_wiring */ + VM_KERN_MEMORY_NONE, /* tag - not wiring */ + THREAD_UNINT, /* interruptible */ + NULL, /* caller_pmap */ + 0 /* caller_pmap_addr */); } } @@ -3308,13 +3373,14 @@ vm_pre_fault(vm_map_offset_t vaddr) */ extern int _map_enter_debug; +extern uint64_t get_current_unique_pid(void); unsigned long vm_fault_collapse_total = 0; unsigned long vm_fault_collapse_skipped = 0; kern_return_t -vm_fault( +vm_fault_external( vm_map_t map, vm_map_offset_t vaddr, vm_prot_t fault_type, @@ -3323,18 +3389,34 @@ vm_fault( pmap_t caller_pmap, vm_map_offset_t caller_pmap_addr) { - return vm_fault_internal(map, vaddr, fault_type, change_wiring, + return vm_fault_internal(map, vaddr, fault_type, change_wiring, vm_tag_bt(), interruptible, caller_pmap, caller_pmap_addr, NULL); } - kern_return_t -vm_fault_internal( +vm_fault( vm_map_t map, vm_map_offset_t vaddr, - vm_prot_t caller_prot, + vm_prot_t fault_type, boolean_t change_wiring, + vm_tag_t wire_tag, /* if wiring must pass tag != VM_KERN_MEMORY_NONE */ + int interruptible, + pmap_t caller_pmap, + vm_map_offset_t caller_pmap_addr) +{ + return vm_fault_internal(map, vaddr, fault_type, change_wiring, wire_tag, + interruptible, caller_pmap, caller_pmap_addr, + NULL); +} + +kern_return_t +vm_fault_internal( + vm_map_t map, + vm_map_offset_t vaddr, + vm_prot_t caller_prot, + boolean_t change_wiring, + vm_tag_t wire_tag, /* if wiring must pass tag != VM_KERN_MEMORY_NONE */ int interruptible, pmap_t caller_pmap, vm_map_offset_t caller_pmap_addr, @@ -3353,47 +3435,76 @@ vm_fault_internal( vm_page_t m; /* Fast access to result_page */ kern_return_t error_code; vm_object_t cur_object; + vm_object_t m_object = NULL; vm_object_offset_t cur_offset; vm_page_t cur_m; vm_object_t new_object; int type_of_fault; pmap_t pmap; - boolean_t interruptible_state; + wait_interrupt_t interruptible_state; vm_map_t real_map = map; vm_map_t original_map = map; + boolean_t object_locks_dropped = FALSE; vm_prot_t fault_type; vm_prot_t original_fault_type; - struct vm_object_fault_info fault_info; + struct vm_object_fault_info fault_info = {}; boolean_t need_collapse = FALSE; boolean_t need_retry = FALSE; boolean_t *need_retry_ptr = NULL; int object_lock_type = 0; int cur_object_lock_type; vm_object_t top_object = VM_OBJECT_NULL; + vm_object_t written_on_object = VM_OBJECT_NULL; + memory_object_t written_on_pager = NULL; + vm_object_offset_t written_on_offset = 0; int throttle_delay; int compressed_count_delta; + int grab_options; + vm_map_offset_t trace_vaddr; + vm_map_offset_t trace_real_vaddr; +#if DEVELOPMENT || DEBUG + vm_map_offset_t real_vaddr; + real_vaddr = vaddr; +#endif /* DEVELOPMENT || DEBUG */ + trace_real_vaddr = vaddr; + vaddr = vm_map_trunc_page(vaddr, PAGE_MASK); + + if (map == kernel_map) { + trace_vaddr = VM_KERNEL_ADDRHIDE(vaddr); + trace_real_vaddr = VM_KERNEL_ADDRHIDE(trace_real_vaddr); + } else { + trace_vaddr = vaddr; + } - KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE, + KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE, (MACHDBG_CODE(DBG_MACH_VM, 2)) | DBG_FUNC_START, - ((uint64_t)vaddr >> 32), - vaddr, + ((uint64_t)trace_vaddr >> 32), + trace_vaddr, (map == kernel_map), 0, 0); if (get_preemption_level() != 0) { - KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE, + KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE, (MACHDBG_CODE(DBG_MACH_VM, 2)) | DBG_FUNC_END, - ((uint64_t)vaddr >> 32), - vaddr, + ((uint64_t)trace_vaddr >> 32), + trace_vaddr, KERN_FAILURE, 0, 0); return (KERN_FAILURE); } - + + thread_t cthread = current_thread(); + boolean_t rtfault = (cthread->sched_mode == TH_MODE_REALTIME); + uint64_t fstart = 0; + + if (rtfault) { + fstart = mach_continuous_time(); + } + interruptible_state = thread_interrupt_level(interruptible); fault_type = (change_wiring ? VM_PROT_NONE : caller_prot); @@ -3409,7 +3520,17 @@ vm_fault_internal( cur_object_lock_type = OBJECT_LOCK_SHARED; + if ((map == kernel_map) && (caller_prot & VM_PROT_WRITE)) { + if (compressor_map) { + if ((vaddr >= vm_map_min(compressor_map)) && (vaddr < vm_map_max(compressor_map))) { + panic("Write fault on compressor map, va: %p type: %u bounds: %p->%p", (void *) vaddr, caller_prot, (void *) vm_map_min(compressor_map), (void *) vm_map_max(compressor_map)); + + } + } + } RetryFault: + assert(written_on_object == VM_OBJECT_NULL); + /* * assume we will hit a page in the cache * otherwise, explicitly override with @@ -3431,7 +3552,6 @@ RetryFault: &fault_info, &real_map); - if (kr != KERN_SUCCESS) { vm_map_unlock_read(map); goto done; @@ -3505,6 +3625,24 @@ RetryFault: * */ +#if defined(__arm64__) + /* + * Fail if reading an execute-only page in a + * pmap that enforces execute-only protection. + */ + if (fault_type == VM_PROT_READ && + (prot & VM_PROT_EXECUTE) && + !(prot & VM_PROT_READ) && + pmap_enforces_execute_only(pmap)) { + vm_object_unlock(object); + vm_map_unlock_read(map); + if (real_map != map) { + vm_map_unlock(real_map); + } + kr = KERN_PROTECTION_FAILURE; + goto done; + } +#endif /* * If this page is to be inserted in a copy delay object @@ -3518,6 +3656,13 @@ RetryFault: cur_object = object; cur_offset = offset; + grab_options = 0; +#if CONFIG_SECLUDED_MEMORY + if (object->can_grab_secluded) { + grab_options |= VM_PAGE_GRAB_SECLUDED; + } +#endif /* CONFIG_SECLUDED_MEMORY */ + while (TRUE) { if (!cur_object->pager_created && cur_object->phys_contiguous) /* superpage */ @@ -3532,9 +3677,12 @@ RetryFault: } m = vm_page_lookup(cur_object, cur_offset); + m_object = NULL; if (m != VM_PAGE_NULL) { - if (m->busy) { + m_object = cur_object; + + if (m->vmp_busy) { wait_result_t result; /* @@ -3583,9 +3731,9 @@ RetryFault: continue; } } - if (m->pageout_queue && m->object->internal && COMPRESSED_PAGER_IS_ACTIVE) { + if ((m->vmp_q_state == VM_PAGE_ON_PAGEOUT_Q) && m_object->internal) { /* - * m->busy == TRUE and the object is locked exclusively + * m->vmp_busy == TRUE and the object is locked exclusively * if m->pageout_queue == TRUE after we acquire the * queues lock, we are guaranteed that it is stable on * the pageout queue and therefore reclaimable @@ -3593,9 +3741,11 @@ RetryFault: * NOTE: this is only true for the internal pageout queue * in the compressor world */ + assert(VM_CONFIG_COMPRESSOR_IS_PRESENT); + vm_page_lock_queues(); - if (m->pageout_queue) { + if (m->vmp_q_state == VM_PAGE_ON_PAGEOUT_Q) { vm_pageout_throttle_up(m); vm_page_unlock_queues(); @@ -3627,7 +3777,7 @@ RetryFault: goto done; } reclaimed_from_pageout: - if (m->laundry) { + if (m->vmp_laundry) { if (object != cur_object) { if (cur_object_lock_type == OBJECT_LOCK_SHARED) { cur_object_lock_type = OBJECT_LOCK_EXCLUSIVE; @@ -3661,24 +3811,22 @@ reclaimed_from_pageout: continue; } } - m->pageout = FALSE; - vm_pageout_steal_laundry(m, FALSE); } - if (m->phys_page == vm_page_guard_addr) { + if (VM_PAGE_GET_PHYS_PAGE(m) == vm_page_guard_addr) { /* * Guard page: let the slow path deal with it */ break; } - if (m->unusual && (m->error || m->restart || m->private || m->absent)) { + if (m->vmp_unusual && (m->vmp_error || m->vmp_restart || m->vmp_private || m->vmp_absent)) { /* * Unusual case... let the slow path deal with it */ break; } - if (VM_OBJECT_PURGEABLE_FAULT_ERROR(m->object)) { + if (VM_OBJECT_PURGEABLE_FAULT_ERROR(m_object)) { if (object != cur_object) vm_object_unlock(object); vm_map_unlock_read(map); @@ -3688,109 +3836,9 @@ reclaimed_from_pageout: kr = KERN_MEMORY_ERROR; goto done; } + assert(m_object == VM_PAGE_OBJECT(m)); - if (m->encrypted) { - /* - * ENCRYPTED SWAP: - * We've soft-faulted (because it's not in the page - * table) on an encrypted page. - * Keep the page "busy" so that no one messes with - * it during the decryption. - * Release the extra locks we're holding, keep only - * the page's VM object lock. - * - * in order to set 'busy' on 'm', we must - * have object that 'm' belongs to locked exclusively - */ - if (object != cur_object) { - vm_object_unlock(object); - - if (cur_object_lock_type == OBJECT_LOCK_SHARED) { - - cur_object_lock_type = OBJECT_LOCK_EXCLUSIVE; - - if (vm_object_lock_upgrade(cur_object) == FALSE) { - /* - * couldn't upgrade so go do a full retry - * immediately since we've already dropped - * the top object lock associated with this page - * and the current one got dropped due to the - * failed upgrade... the state is no longer valid - */ - vm_map_unlock_read(map); - if (real_map != map) - vm_map_unlock(real_map); - - goto RetryFault; - } - } - } else if (object_lock_type == OBJECT_LOCK_SHARED) { - - object_lock_type = OBJECT_LOCK_EXCLUSIVE; - - if (vm_object_lock_upgrade(object) == FALSE) { - /* - * couldn't upgrade, so explictly take the lock - * exclusively and go relookup the page since we - * will have dropped the object lock and - * a different thread could have inserted - * a page at this offset - * no need for a full retry since we're - * at the top level of the object chain - */ - vm_object_lock(object); - - continue; - } - } - m->busy = TRUE; - - vm_map_unlock_read(map); - if (real_map != map) - vm_map_unlock(real_map); - - vm_page_decrypt(m, 0); - - assert(m->busy); - PAGE_WAKEUP_DONE(m); - - vm_object_unlock(cur_object); - /* - * Retry from the top, in case anything - * changed while we were decrypting... - */ - goto RetryFault; - } - ASSERT_PAGE_DECRYPTED(m); - - if(vm_page_is_slideable(m)) { - /* - * We might need to slide this page, and so, - * we want to hold the VM object exclusively. - */ - if (object != cur_object) { - if (cur_object_lock_type == OBJECT_LOCK_SHARED) { - vm_object_unlock(object); - vm_object_unlock(cur_object); - - cur_object_lock_type = OBJECT_LOCK_EXCLUSIVE; - - vm_map_unlock_read(map); - if (real_map != map) - vm_map_unlock(real_map); - - goto RetryFault; - } - } else if (object_lock_type == OBJECT_LOCK_SHARED) { - - vm_object_unlock(object); - object_lock_type = OBJECT_LOCK_EXCLUSIVE; - vm_map_unlock_read(map); - goto RetryFault; - } - } - - if (VM_FAULT_NEED_CS_VALIDATION(map->pmap, m) || + if (VM_FAULT_NEED_CS_VALIDATION(map->pmap, m, m_object) || (physpage_p != NULL && (prot & VM_PROT_WRITE))) { upgrade_for_validation: /* @@ -3845,6 +3893,20 @@ upgrade_for_validation: } if ((fault_type & VM_PROT_WRITE) == 0) { + if (!pmap_has_prot_policy(prot)) { + prot &= ~VM_PROT_WRITE; + } else { + /* + * For a protection that the pmap cares + * about, we must hand over the full + * set of protections (so that the pmap + * layer can apply any desired policy). + * This means that cs_bypass must be + * set, as this can force us to pass + * RWX. + */ + assert(fault_info.cs_bypass); + } if (object != cur_object) { /* @@ -3874,11 +3936,13 @@ upgrade_for_validation: object_lock_type = cur_object_lock_type; } FastPmapEnter: + assert(m_object == VM_PAGE_OBJECT(m)); + /* * prepare for the pmap_enter... * object and map are both locked * m contains valid data - * object == m->object + * object == m->vmp_object * cur_object == NULL or it's been unlocked * no paging references on either object or cur_object */ @@ -3895,10 +3959,8 @@ FastPmapEnter: caller_prot, wired, change_wiring, - fault_info.no_cache, - fault_info.cs_bypass, - fault_info.user_tag, - fault_info.pmap_options, + wire_tag, + &fault_info, need_retry_ptr, &type_of_fault); } else { @@ -3909,22 +3971,34 @@ FastPmapEnter: caller_prot, wired, change_wiring, - fault_info.no_cache, - fault_info.cs_bypass, - fault_info.user_tag, - fault_info.pmap_options, + wire_tag, + &fault_info, need_retry_ptr, &type_of_fault); } +#if DEVELOPMENT || DEBUG + { + int event_code = 0; + if (m_object->internal) + event_code = (MACHDBG_CODE(DBG_MACH_WORKINGSET, VM_REAL_FAULT_ADDR_INTERNAL)); + else if (m_object->object_is_shared_cache) + event_code = (MACHDBG_CODE(DBG_MACH_WORKINGSET, VM_REAL_FAULT_ADDR_SHAREDCACHE)); + else + event_code = (MACHDBG_CODE(DBG_MACH_WORKINGSET, VM_REAL_FAULT_ADDR_EXTERNAL)); + + KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE, event_code, trace_real_vaddr, (fault_info.user_tag << 16) | (caller_prot << 8) | type_of_fault, m->vmp_offset, get_current_unique_pid(), 0); + + DTRACE_VM6(real_fault, vm_map_offset_t, real_vaddr, vm_map_offset_t, m->vmp_offset, int, event_code, int, caller_prot, int, type_of_fault, int, fault_info.user_tag); + } +#endif if (kr == KERN_SUCCESS && physpage_p != NULL) { /* for vm_map_wire_and_extract() */ - *physpage_p = m->phys_page; + *physpage_p = VM_PAGE_GET_PHYS_PAGE(m); if (prot & VM_PROT_WRITE) { - vm_object_lock_assert_exclusive( - m->object); - m->dirty = TRUE; + vm_object_lock_assert_exclusive(m_object); + m->vmp_dirty = TRUE; } } @@ -3944,7 +4018,7 @@ FastPmapEnter: if (need_collapse == TRUE) vm_object_collapse(object, offset, TRUE); - + if (need_retry == FALSE && (type_of_fault == DBG_PAGEIND_FAULT || type_of_fault == DBG_PAGEINV_FAULT || type_of_fault == DBG_CACHE_HIT_FAULT)) { /* @@ -3952,16 +4026,25 @@ FastPmapEnter: * vm_fault_deactivate_behind depends on the * state being up to date */ - vm_fault_is_sequential(object, cur_offset, fault_info.behavior); + vm_fault_is_sequential(m_object, cur_offset, fault_info.behavior); - vm_fault_deactivate_behind(object, cur_offset, fault_info.behavior); + vm_fault_deactivate_behind(m_object, cur_offset, fault_info.behavior); } /* * That's it, clean up and return. */ - if (m->busy) + if (m->vmp_busy) PAGE_WAKEUP_DONE(m); + if (need_retry == FALSE && !m_object->internal && (fault_type & VM_PROT_WRITE)) { + + vm_object_paging_begin(m_object); + + assert(written_on_object == VM_OBJECT_NULL); + written_on_object = m_object; + written_on_pager = m_object->pager; + written_on_offset = m_object->paging_offset + m->vmp_offset; + } vm_object_unlock(object); vm_map_unlock_read(map); @@ -3980,7 +4063,7 @@ FastPmapEnter: (void)pmap_enter_options( pmap, vaddr, 0, 0, 0, 0, 0, PMAP_OPTIONS_NOENTER, NULL); - + need_retry = FALSE; goto RetryFault; } @@ -4005,15 +4088,16 @@ FastPmapEnter: */ break; } - + /* * This is now a shadow based copy on write * fault -- it requires a copy up the shadow * chain. */ - + assert(m_object == VM_PAGE_OBJECT(m)); + if ((cur_object_lock_type == OBJECT_LOCK_SHARED) && - VM_FAULT_NEED_CS_VALIDATION(NULL, m)) { + VM_FAULT_NEED_CS_VALIDATION(NULL, m, m_object)) { goto upgrade_for_validation; } @@ -4023,14 +4107,15 @@ FastPmapEnter: * need to remember current page, as it's the * source of the copy. * - * at this point we hold locks on both + * at this point we hold locks on both * object and cur_object... no need to take * paging refs or mark pages BUSY since * we don't drop either object lock until * the page has been copied and inserted */ cur_m = m; - m = vm_page_grab(); + m = vm_page_grab_options(grab_options); + m_object = NULL; if (m == VM_PAGE_NULL) { /* @@ -4047,17 +4132,19 @@ FastPmapEnter: */ vm_page_copy(cur_m, m); vm_page_insert(m, object, offset); + m_object = object; SET_PAGE_DIRTY(m, FALSE); /* * Now cope with the source page and object */ - if (object->ref_count > 1 && cur_m->pmapped) - pmap_disconnect(cur_m->phys_page); - - if (cur_m->clustered) { + if (object->ref_count > 1 && cur_m->vmp_pmapped) + pmap_disconnect(VM_PAGE_GET_PHYS_PAGE(cur_m)); + + if (cur_m->vmp_clustered) { VM_PAGE_COUNT_AS_PAGEIN(cur_m); VM_PAGE_CONSUME_CLUSTERED(cur_m); + vm_fault_is_sequential(cur_object, cur_offset, fault_info.behavior); } need_collapse = TRUE; @@ -4194,11 +4281,12 @@ FastPmapEnter: * at the top level of the object chain */ vm_object_lock(object); - + continue; } } - m = vm_page_grab(); + m = vm_page_grab_options(grab_options); + m_object = NULL; if (m == VM_PAGE_NULL) { /* @@ -4227,7 +4315,7 @@ FastPmapEnter: cur_object->pager, (cur_offset + cur_object->paging_offset), - m->phys_page, + VM_PAGE_GET_PHYS_PAGE(m), &my_fault_type, c_flags, &compressed_count_delta); @@ -4239,10 +4327,11 @@ FastPmapEnter: cur_object); if (kr != KERN_SUCCESS) { - vm_page_release(m); + vm_page_release(m, FALSE); + m = VM_PAGE_NULL; break; } - m->dirty = TRUE; + m->vmp_dirty = TRUE; /* * If the object is purgeable, its @@ -4270,40 +4359,45 @@ FastPmapEnter: * no ledger update in that * case. */ - } else if ((cur_object->purgable == - VM_PURGABLE_DENY) || - (cur_object->vo_purgeable_owner == + } else if (((cur_object->purgable == + VM_PURGABLE_DENY) && + (!cur_object->vo_ledger_tag)) || + (cur_object->vo_owner == NULL)) { /* * "cur_object" is not purgeable - * or is not owned, so no - * purgeable ledgers to update. + * and is not ledger-taged, or + * there's no owner for it, + * so no owner's ledgers to + * update. */ } else { /* * One less compressed - * purgeable page for + * purgeable/tagged page for * cur_object's owner. */ - vm_purgeable_compressed_update( + vm_object_owner_compressed_update( cur_object, -1); } if (insert_cur_object) { vm_page_insert(m, cur_object, cur_offset); + m_object = cur_object; } else { vm_page_insert(m, object, offset); + m_object = object; } - if ((m->object->wimg_bits & VM_WIMG_MASK) != VM_WIMG_USE_DEFAULT) { + if ((m_object->wimg_bits & VM_WIMG_MASK) != VM_WIMG_USE_DEFAULT) { /* * If the page is not cacheable, * we can't let its contents * linger in the data cache * after the decompression. */ - pmap_sync_page_attributes_phys(m->phys_page); + pmap_sync_page_attributes_phys(VM_PAGE_GET_PHYS_PAGE(m)); } type_of_fault = my_fault_type; @@ -4336,8 +4430,10 @@ FastPmapEnter: * inserted into the original object. */ if (cur_object->shadow_severed || - VM_OBJECT_PURGEABLE_FAULT_ERROR(cur_object)) - { + VM_OBJECT_PURGEABLE_FAULT_ERROR(cur_object) || + cur_object == compressor_object || + cur_object == kernel_object || + cur_object == vm_submap_object) { if (object != cur_object) vm_object_unlock(cur_object); vm_object_unlock(object); @@ -4349,16 +4445,6 @@ FastPmapEnter: kr = KERN_MEMORY_ERROR; goto done; } - if (vm_backing_store_low) { - /* - * we are protecting the system from - * backing store exhaustion... - * must take the slow path if we're - * not privileged - */ - if (!(current_task()->priv_flags & VM_BACKING_STORE_PRIV)) - break; - } if (cur_object != object) { vm_object_unlock(cur_object); @@ -4383,6 +4469,7 @@ FastPmapEnter: } } m = vm_page_alloc(object, offset); + m_object = NULL; if (m == VM_PAGE_NULL) { /* @@ -4391,10 +4478,11 @@ FastPmapEnter: */ break; } + m_object = object; /* * Now zero fill page... - * the page is probably going to + * the page is probably going to * be written soon, so don't bother * to clear the modified bit * @@ -4457,6 +4545,28 @@ handle_copy_delay: if (real_map != map) vm_map_unlock(real_map); + if (__improbable(object == compressor_object || + object == kernel_object || + object == vm_submap_object)) { + /* + * These objects are explicitly managed and populated by the + * kernel. The virtual ranges backed by these objects should + * either have wired pages or "holes" that are not supposed to + * be accessed at all until they get explicitly populated. + * We should never have to resolve a fault on a mapping backed + * by one of these VM objects and providing a zero-filled page + * would be wrong here, so let's fail the fault and let the + * caller crash or recover. + */ + vm_object_unlock(object); + kr = KERN_MEMORY_ERROR; + goto done; + } + + assert(object != compressor_object); + assert(object != kernel_object); + assert(object != vm_submap_object); + /* * Make a reference to this object to * prevent its disposal while we are messing with @@ -4490,7 +4600,7 @@ handle_copy_delay: * * the object is returned locked with a paging reference * - * if top_page != NULL, then it's BUSY and the + * if top_page != NULL, then it's BUSY and the * object it belongs to has a paging reference * but is returned unlocked */ @@ -4506,7 +4616,7 @@ handle_copy_delay: */ switch (kr) { case VM_FAULT_MEMORY_SHORTAGE: - if (vm_page_wait((change_wiring) ? + if (vm_page_wait((change_wiring) ? THREAD_UNINT : THREAD_ABORTSAFE)) goto RetryFault; @@ -4530,11 +4640,13 @@ handle_copy_delay: } } m = result_page; + m_object = NULL; if (m != VM_PAGE_NULL) { + m_object = VM_PAGE_OBJECT(m); assert((change_wiring && !wired) ? - (top_page == VM_PAGE_NULL) : - ((top_page == VM_PAGE_NULL) == (m->object == object))); + (top_page == VM_PAGE_NULL) : + ((top_page == VM_PAGE_NULL) == (m_object == object))); } /* @@ -4544,30 +4656,55 @@ handle_copy_delay: #define RELEASE_PAGE(m) \ MACRO_BEGIN \ PAGE_WAKEUP_DONE(m); \ - if (!m->active && !m->inactive && !m->throttled) { \ - vm_page_lockspin_queues(); \ - if (!m->active && !m->inactive && !m->throttled) \ - vm_page_activate(m); \ - vm_page_unlock_queues(); \ - } \ + if ( !VM_PAGE_PAGEABLE(m)) { \ + vm_page_lockspin_queues(); \ + if ( !VM_PAGE_PAGEABLE(m)) \ + vm_page_activate(m); \ + vm_page_unlock_queues(); \ + } \ MACRO_END + + object_locks_dropped = FALSE; /* * We must verify that the maps have not changed - * since our last lookup. + * since our last lookup. vm_map_verify() needs the + * map lock (shared) but we are holding object locks. + * So we do a try_lock() first and, if that fails, we + * drop the object locks and go in for the map lock again. */ - if (m != VM_PAGE_NULL) { - old_copy_object = m->object->copy; - vm_object_unlock(m->object); - } else { - old_copy_object = VM_OBJECT_NULL; - vm_object_unlock(object); + if (!vm_map_try_lock_read(original_map)) { + + if (m != VM_PAGE_NULL) { + old_copy_object = m_object->copy; + vm_object_unlock(m_object); + } else { + old_copy_object = VM_OBJECT_NULL; + vm_object_unlock(object); + } + + object_locks_dropped = TRUE; + + vm_map_lock_read(original_map); } - /* - * no object locks are held at this point - */ if ((map != original_map) || !vm_map_verify(map, &version)) { + + if (object_locks_dropped == FALSE) { + if (m != VM_PAGE_NULL) { + old_copy_object = m_object->copy; + vm_object_unlock(m_object); + } else { + old_copy_object = VM_OBJECT_NULL; + vm_object_unlock(object); + } + + object_locks_dropped = TRUE; + } + + /* + * no object locks are held at this point + */ vm_object_t retry_object; vm_object_offset_t retry_offset; vm_prot_t retry_prot; @@ -4582,7 +4719,6 @@ handle_copy_delay: * take another fault. */ map = original_map; - vm_map_lock_read(map); kr = vm_map_lookup_locked(&map, vaddr, fault_type & ~VM_PROT_WRITE, @@ -4597,17 +4733,19 @@ handle_copy_delay: vm_map_unlock_read(map); if (m != VM_PAGE_NULL) { + assert(VM_PAGE_OBJECT(m) == m_object); + /* * retake the lock so that * we can drop the paging reference * in vm_fault_cleanup and do the * PAGE_WAKEUP_DONE in RELEASE_PAGE */ - vm_object_lock(m->object); + vm_object_lock(m_object); RELEASE_PAGE(m); - vm_fault_cleanup(m->object, top_page); + vm_fault_cleanup(m_object, top_page); } else { /* * retake the lock so that @@ -4631,17 +4769,19 @@ handle_copy_delay: vm_map_unlock(real_map); if (m != VM_PAGE_NULL) { + assert(VM_PAGE_OBJECT(m) == m_object); + /* * retake the lock so that * we can drop the paging reference * in vm_fault_cleanup and do the * PAGE_WAKEUP_DONE in RELEASE_PAGE */ - vm_object_lock(m->object); + vm_object_lock(m_object); RELEASE_PAGE(m); - vm_fault_cleanup(m->object, top_page); + vm_fault_cleanup(m_object, top_page); } else { /* * retake the lock so that @@ -4660,20 +4800,31 @@ handle_copy_delay: * Check whether the protection has changed or the object * has been copied while we left the map unlocked. */ - prot &= retry_prot; + if (pmap_has_prot_policy(retry_prot)) { + /* If the pmap layer cares, pass the full set. */ + prot = retry_prot; + } else { + prot &= retry_prot; + } } - if (m != VM_PAGE_NULL) { - vm_object_lock(m->object); - if (m->object->copy != old_copy_object) { - /* - * The copy object changed while the top-level object - * was unlocked, so take away write permission. - */ - prot &= ~VM_PROT_WRITE; - } - } else - vm_object_lock(object); + if (object_locks_dropped == TRUE) { + if (m != VM_PAGE_NULL) { + vm_object_lock(m_object); + + if (m_object->copy != old_copy_object) { + /* + * The copy object changed while the top-level object + * was unlocked, so take away write permission. + */ + assert(!pmap_has_prot_policy(prot)); + prot &= ~VM_PROT_WRITE; + } + } else + vm_object_lock(object); + + object_locks_dropped = FALSE; + } /* * If we want to wire down this page, but no longer have @@ -4681,14 +4832,16 @@ handle_copy_delay: */ if (wired && (fault_type != (prot | VM_PROT_WRITE))) { - vm_map_verify_done(map, &version); + vm_map_unlock_read(map); if (real_map != map) vm_map_unlock(real_map); if (m != VM_PAGE_NULL) { + assert(VM_PAGE_OBJECT(m) == m_object); + RELEASE_PAGE(m); - vm_fault_cleanup(m->object, top_page); + vm_fault_cleanup(m_object, top_page); } else vm_fault_cleanup(object, top_page); @@ -4712,10 +4865,8 @@ handle_copy_delay: caller_prot, wired, change_wiring, - fault_info.no_cache, - fault_info.cs_bypass, - fault_info.user_tag, - fault_info.pmap_options, + wire_tag, + &fault_info, NULL, &type_of_fault); } else { @@ -4726,29 +4877,45 @@ handle_copy_delay: caller_prot, wired, change_wiring, - fault_info.no_cache, - fault_info.cs_bypass, - fault_info.user_tag, - fault_info.pmap_options, + wire_tag, + &fault_info, NULL, &type_of_fault); } + assert(VM_PAGE_OBJECT(m) == m_object); + +#if DEVELOPMENT || DEBUG + { + int event_code = 0; + + if (m_object->internal) + event_code = (MACHDBG_CODE(DBG_MACH_WORKINGSET, VM_REAL_FAULT_ADDR_INTERNAL)); + else if (m_object->object_is_shared_cache) + event_code = (MACHDBG_CODE(DBG_MACH_WORKINGSET, VM_REAL_FAULT_ADDR_SHAREDCACHE)); + else + event_code = (MACHDBG_CODE(DBG_MACH_WORKINGSET, VM_REAL_FAULT_ADDR_EXTERNAL)); + + KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE, event_code, trace_real_vaddr, (fault_info.user_tag << 16) | (caller_prot << 8) | type_of_fault, m->vmp_offset, get_current_unique_pid(), 0); + + DTRACE_VM6(real_fault, vm_map_offset_t, real_vaddr, vm_map_offset_t, m->vmp_offset, int, event_code, int, caller_prot, int, type_of_fault, int, fault_info.user_tag); + } +#endif if (kr != KERN_SUCCESS) { /* abort this page fault */ - vm_map_verify_done(map, &version); + vm_map_unlock_read(map); if (real_map != map) vm_map_unlock(real_map); PAGE_WAKEUP_DONE(m); - vm_fault_cleanup(m->object, top_page); + vm_fault_cleanup(m_object, top_page); vm_object_deallocate(object); goto done; } if (physpage_p != NULL) { /* for vm_map_wire_and_extract() */ - *physpage_p = m->phys_page; + *physpage_p = VM_PAGE_GET_PHYS_PAGE(m); if (prot & VM_PROT_WRITE) { - vm_object_lock_assert_exclusive(m->object); - m->dirty = TRUE; + vm_object_lock_assert_exclusive(m_object); + m->vmp_dirty = TRUE; } } } else { @@ -4757,33 +4924,11 @@ handle_copy_delay: vm_map_offset_t laddr; vm_map_offset_t ldelta, hdelta; - /* + /* * do a pmap block mapping from the physical address - * in the object + * in the object */ -#ifdef ppc - /* While we do not worry about execution protection in */ - /* general, certian pages may have instruction execution */ - /* disallowed. We will check here, and if not allowed */ - /* to execute, we return with a protection failure. */ - - if ((fault_type & VM_PROT_EXECUTE) && - (!pmap_eligible_for_execute((ppnum_t)(object->vo_shadow_offset >> 12)))) { - - vm_map_verify_done(map, &version); - - if (real_map != map) - vm_map_unlock(real_map); - - vm_fault_cleanup(object, top_page); - vm_object_deallocate(object); - - kr = KERN_PROTECTION_FAILURE; - goto done; - } -#endif /* ppc */ - if (real_map != map) vm_map_unlock(real_map); @@ -4804,8 +4949,8 @@ handle_copy_delay: if (hdelta > (entry->vme_end - laddr)) hdelta = entry->vme_end - laddr; if (entry->is_sub_map) { - - laddr = ((laddr - entry->vme_start) + + laddr = ((laddr - entry->vme_start) + VME_OFFSET(entry)); vm_map_lock_read(VME_SUBMAP(entry)); @@ -4816,13 +4961,13 @@ handle_copy_delay: real_map = VME_SUBMAP(entry); } map = VME_SUBMAP(entry); - + } else { break; } } - if (vm_map_lookup_entry(map, laddr, &entry) && + if (vm_map_lookup_entry(map, laddr, &entry) && (VME_OBJECT(entry) != NULL) && (VME_OBJECT(entry) == object)) { int superpage; @@ -4852,38 +4997,66 @@ handle_copy_delay: * Set up a block mapped area */ assert((uint32_t)((ldelta + hdelta) >> PAGE_SHIFT) == ((ldelta + hdelta) >> PAGE_SHIFT)); - pmap_map_block(caller_pmap, - (addr64_t)(caller_pmap_addr - ldelta), - (ppnum_t)((((vm_map_offset_t) (VME_OBJECT(entry)->vo_shadow_offset)) + - VME_OFFSET(entry) + (laddr - entry->vme_start) - ldelta) >> PAGE_SHIFT), - (uint32_t)((ldelta + hdelta) >> PAGE_SHIFT), prot, - (VM_WIMG_MASK & (int)object->wimg_bits) | superpage, 0); - } else { + kr = pmap_map_block(caller_pmap, + (addr64_t)(caller_pmap_addr - ldelta), + (ppnum_t)((((vm_map_offset_t) (VME_OBJECT(entry)->vo_shadow_offset)) + + VME_OFFSET(entry) + (laddr - entry->vme_start) - ldelta) >> PAGE_SHIFT), + (uint32_t)((ldelta + hdelta) >> PAGE_SHIFT), prot, + (VM_WIMG_MASK & (int)object->wimg_bits) | superpage, 0); + + if (kr != KERN_SUCCESS) { + goto cleanup; + } + } else { /* * Set up a block mapped area */ assert((uint32_t)((ldelta + hdelta) >> PAGE_SHIFT) == ((ldelta + hdelta) >> PAGE_SHIFT)); - pmap_map_block(real_map->pmap, - (addr64_t)(vaddr - ldelta), - (ppnum_t)((((vm_map_offset_t)(VME_OBJECT(entry)->vo_shadow_offset)) + - VME_OFFSET(entry) + (laddr - entry->vme_start) - ldelta) >> PAGE_SHIFT), - (uint32_t)((ldelta + hdelta) >> PAGE_SHIFT), prot, - (VM_WIMG_MASK & (int)object->wimg_bits) | superpage, 0); + kr = pmap_map_block(real_map->pmap, + (addr64_t)(vaddr - ldelta), + (ppnum_t)((((vm_map_offset_t)(VME_OBJECT(entry)->vo_shadow_offset)) + + VME_OFFSET(entry) + (laddr - entry->vme_start) - ldelta) >> PAGE_SHIFT), + (uint32_t)((ldelta + hdelta) >> PAGE_SHIFT), prot, + (VM_WIMG_MASK & (int)object->wimg_bits) | superpage, 0); + + if (kr != KERN_SUCCESS) { + goto cleanup; + } } } } + /* + * Success + */ + kr = KERN_SUCCESS; + + /* + * TODO: could most of the done cases just use cleanup? + */ +cleanup: /* * Unlock everything, and return */ - vm_map_verify_done(map, &version); + vm_map_unlock_read(map); if (real_map != map) vm_map_unlock(real_map); if (m != VM_PAGE_NULL) { + assert(VM_PAGE_OBJECT(m) == m_object); + + if (!m_object->internal && (fault_type & VM_PROT_WRITE)) { + + vm_object_paging_begin(m_object); + + assert(written_on_object == VM_OBJECT_NULL); + written_on_object = m_object; + written_on_pager = m_object->pager; + written_on_offset = m_object->paging_offset + m->vmp_offset; + } PAGE_WAKEUP_DONE(m); - vm_fault_cleanup(m->object, top_page); + vm_fault_cleanup(m_object, top_page); } else vm_fault_cleanup(object, top_page); @@ -4891,7 +5064,6 @@ handle_copy_delay: #undef RELEASE_PAGE - kr = KERN_SUCCESS; done: thread_interrupt_level(interruptible_state); @@ -4917,10 +5089,26 @@ done: } } } - KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE, + + if (written_on_object) { + + vnode_pager_dirtied(written_on_pager, written_on_offset, written_on_offset + PAGE_SIZE_64); + + vm_object_lock(written_on_object); + vm_object_paging_end(written_on_object); + vm_object_unlock(written_on_object); + + written_on_object = VM_OBJECT_NULL; + } + + if (rtfault) { + vm_record_rtfault(cthread, fstart, trace_vaddr, type_of_fault); + } + + KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE, (MACHDBG_CODE(DBG_MACH_VM, 2)) | DBG_FUNC_END, - ((uint64_t)vaddr >> 32), - vaddr, + ((uint64_t)trace_vaddr >> 32), + trace_vaddr, kr, type_of_fault, 0); @@ -4938,19 +5126,19 @@ vm_fault_wire( vm_map_t map, vm_map_entry_t entry, vm_prot_t prot, + vm_tag_t wire_tag, pmap_t pmap, vm_map_offset_t pmap_addr, ppnum_t *physpage_p) { - - register vm_map_offset_t va; - register vm_map_offset_t end_addr = entry->vme_end; - register kern_return_t rc; + vm_map_offset_t va; + vm_map_offset_t end_addr = entry->vme_end; + kern_return_t rc; assert(entry->in_transition); - if ((VME_OBJECT(entry) != NULL) && - !entry->is_sub_map && + if ((VME_OBJECT(entry) != NULL) && + !entry->is_sub_map && VME_OBJECT(entry)->phys_contiguous) { return KERN_SUCCESS; } @@ -4961,7 +5149,7 @@ vm_fault_wire( * page tables and such can be locked down as well. */ - pmap_pageable(pmap, pmap_addr, + pmap_pageable(pmap, pmap_addr, pmap_addr + (end_addr - entry->vme_start), FALSE); /* @@ -4970,14 +5158,14 @@ vm_fault_wire( */ for (va = entry->vme_start; va < end_addr; va += PAGE_SIZE) { - rc = vm_fault_wire_fast(map, va, prot, entry, pmap, + rc = vm_fault_wire_fast(map, va, prot, wire_tag, entry, pmap, pmap_addr + (va - entry->vme_start), physpage_p); if (rc != KERN_SUCCESS) { - rc = vm_fault_internal(map, va, prot, TRUE, + rc = vm_fault_internal(map, va, prot, TRUE, wire_tag, ((pmap == kernel_pmap) ? THREAD_UNINT - : THREAD_ABORTSAFE), + : THREAD_ABORTSAFE), pmap, (pmap_addr + (va - entry->vme_start)), @@ -4990,7 +5178,7 @@ vm_fault_wire( /* unwire wired pages */ tmp_entry.vme_end = va; - vm_fault_unwire(map, + vm_fault_unwire(map, &tmp_entry, FALSE, pmap, pmap_addr); return rc; @@ -5012,10 +5200,11 @@ vm_fault_unwire( pmap_t pmap, vm_map_offset_t pmap_addr) { - register vm_map_offset_t va; - register vm_map_offset_t end_addr = entry->vme_end; + vm_map_offset_t va; + vm_map_offset_t end_addr = entry->vme_end; vm_object_t object; - struct vm_object_fault_info fault_info; + struct vm_object_fault_info fault_info = {}; + unsigned int unwired_pages; object = (entry->is_sub_map) ? VM_OBJECT_NULL : VME_OBJECT(entry); @@ -5031,7 +5220,6 @@ vm_fault_unwire( fault_info.interruptible = THREAD_UNINT; fault_info.behavior = entry->behavior; fault_info.user_tag = VME_ALIAS(entry); - fault_info.pmap_options = 0; if (entry->iokit_acct || (!entry->is_sub_map && !entry->use_pmap)) { fault_info.pmap_options |= PMAP_OPTIONS_ALT_ACCT; @@ -5040,10 +5228,8 @@ vm_fault_unwire( fault_info.hi_offset = (entry->vme_end - entry->vme_start) + VME_OFFSET(entry); fault_info.no_cache = entry->no_cache; 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; + + unwired_pages = 0; /* * Since the pages are wired down, we must be able to @@ -5054,11 +5240,11 @@ vm_fault_unwire( if (object == VM_OBJECT_NULL) { if (pmap) { - pmap_change_wiring(pmap, + pmap_change_wiring(pmap, pmap_addr + (va - entry->vme_start), FALSE); } - (void) vm_fault(map, va, VM_PROT_NONE, - TRUE, THREAD_UNINT, pmap, pmap_addr); + (void) vm_fault(map, va, VM_PROT_NONE, + TRUE, VM_KERN_MEMORY_NONE, THREAD_UNINT, pmap, pmap_addr); } else { vm_prot_t prot; vm_page_t result_page; @@ -5066,13 +5252,12 @@ vm_fault_unwire( vm_object_t result_object; vm_fault_return_t result; - if (end_addr - va > (vm_size_t) -1) { - /* 32-bit overflow */ - fault_info.cluster_size = (vm_size_t) (0 - PAGE_SIZE); - } else { - fault_info.cluster_size = (vm_size_t) (end_addr - va); - assert(fault_info.cluster_size == end_addr - va); + /* cap cluster size at maximum UPL size */ + upl_size_t cluster_size; + if (os_sub_overflow(end_addr, va, &cluster_size)) { + cluster_size = 0 - (upl_size_t)PAGE_SIZE; } + fault_info.cluster_size = cluster_size; do { prot = VM_PROT_NONE; @@ -5091,7 +5276,7 @@ vm_fault_unwire( FALSE, /* page not looked up */ &prot, &result_page, &top_page, (int *)0, - NULL, map->no_zero_fill, + NULL, map->no_zero_fill, FALSE, &fault_info); } while (result == VM_FAULT_RETRY); @@ -5101,7 +5286,7 @@ vm_fault_unwire( * move on to the next one in case the remaining pages are mapped from * different objects. During a forced unmount, the object is terminated * so the alive flag will be false if this happens. A forced unmount will - * will occur when an external disk is unplugged before the user does an + * will occur when an external disk is unplugged before the user does an * eject, so we don't want to panic in that situation. */ @@ -5123,16 +5308,19 @@ vm_fault_unwire( if (result != VM_FAULT_SUCCESS) panic("vm_fault_unwire: failure"); - result_object = result_page->object; + result_object = VM_PAGE_OBJECT(result_page); if (deallocate) { - assert(result_page->phys_page != + assert(VM_PAGE_GET_PHYS_PAGE(result_page) != vm_page_fictitious_addr); - pmap_disconnect(result_page->phys_page); + pmap_disconnect(VM_PAGE_GET_PHYS_PAGE(result_page)); + if (VM_PAGE_WIRED(result_page)) { + unwired_pages++; + } VM_PAGE_FREE(result_page); } else { - if ((pmap) && (result_page->phys_page != vm_page_guard_addr)) - pmap_change_wiring(pmap, + if ((pmap) && (VM_PAGE_GET_PHYS_PAGE(result_page) != vm_page_guard_addr)) + pmap_change_wiring(pmap, pmap_addr + (va - entry->vme_start), FALSE); @@ -5140,9 +5328,10 @@ vm_fault_unwire( vm_page_lockspin_queues(); vm_page_unwire(result_page, TRUE); vm_page_unlock_queues(); + unwired_pages++; } if(entry->zero_wired_pages) { - pmap_zero_page(result_page->phys_page); + pmap_zero_page(VM_PAGE_GET_PHYS_PAGE(result_page)); entry->zero_wired_pages = FALSE; } @@ -5158,9 +5347,12 @@ vm_fault_unwire( * such may be unwired themselves. */ - pmap_pageable(pmap, pmap_addr, + pmap_pageable(pmap, pmap_addr, pmap_addr + (end_addr - entry->vme_start), TRUE); + if (kernel_object == object) { + vm_tag_update_size(fault_info.user_tag, -ptoa_64(unwired_pages)); + } } /* @@ -5187,7 +5379,8 @@ static kern_return_t vm_fault_wire_fast( __unused vm_map_t map, vm_map_offset_t va, - vm_prot_t caller_prot, + __unused vm_prot_t caller_prot, + vm_tag_t wire_tag, vm_map_entry_t entry, pmap_t pmap, vm_map_offset_t pmap_addr, @@ -5195,11 +5388,12 @@ vm_fault_wire_fast( { vm_object_t object; vm_object_offset_t offset; - register vm_page_t m; + vm_page_t m; vm_prot_t prot; thread_t thread = current_thread(); int type_of_fault; kern_return_t kr; + struct vm_object_fault_info fault_info = {}; VM_STAT_INCR(faults); @@ -5282,19 +5476,15 @@ vm_fault_wire_fast( /* * Look for page in top-level object. If it's not there or * there's something going on, give up. - * ENCRYPTED SWAP: use the slow fault path, since we'll need to - * decrypt the page before wiring it down. */ m = vm_page_lookup(object, offset); - if ((m == VM_PAGE_NULL) || (m->busy) || (m->encrypted) || - (m->unusual && ( m->error || m->restart || m->absent))) { + if ((m == VM_PAGE_NULL) || (m->vmp_busy) || + (m->vmp_unusual && ( m->vmp_error || m->vmp_restart || m->vmp_absent))) { GIVE_UP; } - ASSERT_PAGE_DECRYPTED(m); - - if (m->fictitious && - m->phys_page == vm_page_guard_addr) { + if (m->vmp_fictitious && + VM_PAGE_GET_PHYS_PAGE(m) == vm_page_guard_addr) { /* * Guard pages are fictitious pages and are never * entered into a pmap, so let's say it's been wired... @@ -5305,19 +5495,19 @@ vm_fault_wire_fast( /* * Wire the page down now. All bail outs beyond this - * point must unwire the page. + * point must unwire the page. */ vm_page_lockspin_queues(); - vm_page_wire(m, VM_PROT_MEMORY_TAG(caller_prot), TRUE); + vm_page_wire(m, wire_tag, TRUE); vm_page_unlock_queues(); /* * Mark page busy for other threads. */ - assert(!m->busy); - m->busy = TRUE; - assert(!m->absent); + assert(!m->vmp_busy); + m->vmp_busy = TRUE; + assert(!m->vmp_absent); /* * Give up if the page is being written and there's a copy object @@ -5327,6 +5517,13 @@ vm_fault_wire_fast( GIVE_UP; } + fault_info.user_tag = VME_ALIAS(entry); + fault_info.pmap_options = 0; + if (entry->iokit_acct || + (!entry->is_sub_map && !entry->use_pmap)) { + fault_info.pmap_options |= PMAP_OPTIONS_ALT_ACCT; + } + /* * Put this page into the physical map. */ @@ -5336,17 +5533,16 @@ vm_fault_wire_fast( pmap_addr, prot, prot, - TRUE, - FALSE, - FALSE, - FALSE, - VME_ALIAS(entry), - ((entry->iokit_acct || - (!entry->is_sub_map && !entry->use_pmap)) - ? PMAP_OPTIONS_ALT_ACCT - : 0), + TRUE, /* wired */ + FALSE, /* change_wiring */ + wire_tag, + &fault_info, NULL, &type_of_fault); + if (kr != KERN_SUCCESS) { + RELEASE_PAGE(m); + GIVE_UP; + } done: /* @@ -5356,10 +5552,11 @@ done: if (physpage_p) { /* for vm_map_wire_and_extract() */ if (kr == KERN_SUCCESS) { - *physpage_p = m->phys_page; + assert(object == VM_PAGE_OBJECT(m)); + *physpage_p = VM_PAGE_GET_PHYS_PAGE(m); if (prot & VM_PROT_WRITE) { - vm_object_lock_assert_exclusive(m->object); - m->dirty = TRUE; + vm_object_lock_assert_exclusive(object); + m->vmp_dirty = TRUE; } } else { *physpage_p = 0; @@ -5384,14 +5581,15 @@ vm_fault_copy_cleanup( vm_page_t page, vm_page_t top_page) { - vm_object_t object = page->object; + vm_object_t object = VM_PAGE_OBJECT(page); vm_object_lock(object); PAGE_WAKEUP_DONE(page); - if (!page->active && !page->inactive && !page->throttled) { + if ( !VM_PAGE_PAGEABLE(page)) { vm_page_lockspin_queues(); - if (!page->active && !page->inactive && !page->throttled) + if ( !VM_PAGE_PAGEABLE(page)) { vm_page_activate(page); + } vm_page_unlock_queues(); } vm_fault_cleanup(object, top_page); @@ -5404,12 +5602,12 @@ vm_fault_copy_dst_cleanup( vm_object_t object; if (page != VM_PAGE_NULL) { - object = page->object; + object = VM_PAGE_OBJECT(page); vm_object_lock(object); vm_page_lockspin_queues(); vm_page_unwire(page, TRUE); vm_page_unlock_queues(); - vm_object_paging_end(object); + vm_object_paging_end(object); vm_object_unlock(object); } } @@ -5453,7 +5651,7 @@ vm_fault_copy( int interruptible) { vm_page_t result_page; - + vm_page_t src_page; vm_page_t src_top_page; vm_prot_t src_prot; @@ -5464,12 +5662,13 @@ vm_fault_copy( vm_map_size_t amount_left; vm_object_t old_copy_object; + vm_object_t result_page_object = NULL; kern_return_t error = 0; vm_fault_return_t result; vm_map_size_t part_size; - struct vm_object_fault_info fault_info_src; - struct vm_object_fault_info fault_info_dst; + struct vm_object_fault_info fault_info_src = {}; + struct vm_object_fault_info fault_info_dst = {}; /* * In order not to confuse the clustered pageins, align @@ -5486,29 +5685,15 @@ vm_fault_copy( fault_info_src.interruptible = interruptible; fault_info_src.behavior = VM_BEHAVIOR_SEQUENTIAL; - fault_info_src.user_tag = 0; - fault_info_src.pmap_options = 0; fault_info_src.lo_offset = vm_object_trunc_page(src_offset); fault_info_src.hi_offset = fault_info_src.lo_offset + amount_left; - fault_info_src.no_cache = FALSE; fault_info_src.stealth = TRUE; - fault_info_src.io_sync = FALSE; - fault_info_src.cs_bypass = FALSE; - fault_info_src.mark_zf_absent = FALSE; - fault_info_src.batch_pmap_op = FALSE; fault_info_dst.interruptible = interruptible; fault_info_dst.behavior = VM_BEHAVIOR_SEQUENTIAL; - fault_info_dst.user_tag = 0; - fault_info_dst.pmap_options = 0; fault_info_dst.lo_offset = vm_object_trunc_page(dst_offset); fault_info_dst.hi_offset = fault_info_dst.lo_offset + amount_left; - fault_info_dst.no_cache = FALSE; fault_info_dst.stealth = TRUE; - fault_info_dst.io_sync = FALSE; - fault_info_dst.cs_bypass = FALSE; - fault_info_dst.mark_zf_absent = FALSE; - fault_info_dst.batch_pmap_op = FALSE; do { /* while (amount_left > 0) */ /* @@ -5525,13 +5710,12 @@ vm_fault_copy( vm_object_lock(dst_object); vm_object_paging_begin(dst_object); - if (amount_left > (vm_size_t) -1) { - /* 32-bit overflow */ - fault_info_dst.cluster_size = (vm_size_t) (0 - PAGE_SIZE); - } else { - fault_info_dst.cluster_size = (vm_size_t) amount_left; - assert(fault_info_dst.cluster_size == amount_left); + /* cap cluster size at maximum UPL size */ + upl_size_t cluster_size; + if (os_convert_overflow(amount_left, &cluster_size)) { + cluster_size = 0 - (upl_size_t)PAGE_SIZE; } + fault_info_dst.cluster_size = cluster_size; XPR(XPR_VM_FAULT,"vm_fault_copy -> vm_fault_page\n",0,0,0,0,0); dst_page = VM_PAGE_NULL; @@ -5572,7 +5756,8 @@ vm_fault_copy( } assert ((dst_prot & VM_PROT_WRITE) != VM_PROT_NONE); - old_copy_object = dst_page->object->copy; + assert(dst_object == VM_PAGE_OBJECT(dst_page)); + old_copy_object = dst_object->copy; /* * There exists the possiblity that the source and @@ -5581,7 +5766,7 @@ vm_fault_copy( * same, the call to vm_fault_page() for the * destination page will deadlock. To prevent this we * wire the page so we can drop busy without having - * the page daemon steal the page. We clean up the + * the page daemon steal the page. We clean up the * top page but keep the paging reference on the object * holding the dest page so it doesn't go away. */ @@ -5590,7 +5775,7 @@ vm_fault_copy( vm_page_wire(dst_page, VM_KERN_MEMORY_OSFMK, TRUE); vm_page_unlock_queues(); PAGE_WAKEUP_DONE(dst_page); - vm_object_unlock(dst_page->object); + vm_object_unlock(dst_object); if (dst_top_page != VM_PAGE_NULL) { vm_object_lock(dst_object); @@ -5619,24 +5804,22 @@ vm_fault_copy( src_prot = VM_PROT_READ; vm_object_paging_begin(src_object); - if (amount_left > (vm_size_t) -1) { - /* 32-bit overflow */ - fault_info_src.cluster_size = (vm_size_t) (0 - PAGE_SIZE); - } else { - fault_info_src.cluster_size = (vm_size_t) amount_left; - assert(fault_info_src.cluster_size == amount_left); + /* cap cluster size at maximum UPL size */ + if (os_convert_overflow(amount_left, &cluster_size)) { + cluster_size = 0 - (upl_size_t)PAGE_SIZE; } + fault_info_src.cluster_size = cluster_size; XPR(XPR_VM_FAULT, "vm_fault_copy(2) -> vm_fault_page\n", 0,0,0,0,0); result_page = VM_PAGE_NULL; result = vm_fault_page( - src_object, + src_object, vm_object_trunc_page(src_offset), VM_PROT_READ, FALSE, FALSE, /* page not looked up */ - &src_prot, + &src_prot, &result_page, &src_top_page, (int *)0, &error, FALSE, FALSE, &fault_info_src); @@ -5670,32 +5853,36 @@ vm_fault_copy( "vm_fault_page()\n", result); } - + result_page_object = VM_PAGE_OBJECT(result_page); assert((src_top_page == VM_PAGE_NULL) == - (result_page->object == src_object)); + (result_page_object == src_object)); } assert ((src_prot & VM_PROT_READ) != VM_PROT_NONE); - vm_object_unlock(result_page->object); + vm_object_unlock(result_page_object); } + vm_map_lock_read(dst_map); + if (!vm_map_verify(dst_map, dst_version)) { + vm_map_unlock_read(dst_map); if (result_page != VM_PAGE_NULL && src_page != dst_page) vm_fault_copy_cleanup(result_page, src_top_page); vm_fault_copy_dst_cleanup(dst_page); break; } + assert(dst_object == VM_PAGE_OBJECT(dst_page)); - vm_object_lock(dst_page->object); + vm_object_lock(dst_object); - if (dst_page->object->copy != old_copy_object) { - vm_object_unlock(dst_page->object); - vm_map_verify_done(dst_map, dst_version); + if (dst_object->copy != old_copy_object) { + vm_object_unlock(dst_object); + vm_map_unlock_read(dst_map); if (result_page != VM_PAGE_NULL && src_page != dst_page) vm_fault_copy_cleanup(result_page, src_top_page); vm_fault_copy_dst_cleanup(dst_page); break; } - vm_object_unlock(dst_page->object); + vm_object_unlock(dst_object); /* * Copy the page, and note that it is dirty @@ -5736,10 +5923,10 @@ vm_fault_copy( dst_page, (vm_offset_t) dst_po, (vm_size_t)part_size); - if(!dst_page->dirty){ + if(!dst_page->vmp_dirty){ vm_object_lock(dst_object); SET_PAGE_DIRTY(dst_page, TRUE); - vm_object_unlock(dst_page->object); + vm_object_unlock(dst_object); } } @@ -5749,14 +5936,14 @@ vm_fault_copy( if (result_page == VM_PAGE_NULL) vm_page_zero_fill(dst_page); else{ - vm_object_lock(result_page->object); + vm_object_lock(result_page_object); vm_page_copy(result_page, dst_page); - vm_object_unlock(result_page->object); + vm_object_unlock(result_page_object); - if(!dst_page->dirty){ + if(!dst_page->vmp_dirty){ vm_object_lock(dst_object); SET_PAGE_DIRTY(dst_page, TRUE); - vm_object_unlock(dst_page->object); + vm_object_unlock(dst_object); } } @@ -5766,7 +5953,7 @@ vm_fault_copy( * Unlock everything, and return */ - vm_map_verify_done(dst_map, dst_version); + vm_map_unlock_read(dst_map); if (result_page != VM_PAGE_NULL && src_page != dst_page) vm_fault_copy_cleanup(result_page, src_top_page); @@ -5780,7 +5967,7 @@ vm_fault_copy( RETURN(KERN_SUCCESS); #undef RETURN - /*NOTREACHED*/ + /*NOTREACHED*/ } #if VM_FAULT_CLASSIFY @@ -5813,15 +6000,15 @@ vm_fault_classify(vm_object_t object, while (TRUE) { m = vm_page_lookup(object, offset); - if (m != VM_PAGE_NULL) { - if (m->busy || m->error || m->restart || m->absent) { + if (m != VM_PAGE_NULL) { + if (m->vmp_busy || m->vmp_error || m->vmp_restart || m->vmp_absent) { type = VM_FAULT_TYPE_OTHER; break; } if (((fault_type & VM_PROT_WRITE) == 0) || ((level == 0) && object->copy == VM_OBJECT_NULL)) { type = VM_FAULT_TYPE_MAP_IN; - break; + break; } type = VM_FAULT_TYPE_COPY; break; @@ -5869,12 +6056,8 @@ vm_fault_classify_init(void) #endif /* VM_FAULT_CLASSIFY */ vm_offset_t -kdp_lightweight_fault(vm_map_t map, vm_offset_t cur_target_addr, uint32_t *fault_results) +kdp_lightweight_fault(vm_map_t map, vm_offset_t cur_target_addr) { -#pragma unused(map, cur_target_addr, fault_results) - - return 0; -#if 0 vm_map_entry_t entry; vm_object_t object; vm_offset_t object_offset; @@ -5884,7 +6067,6 @@ kdp_lightweight_fault(vm_map_t map, vm_offset_t cur_target_addr, uint32_t *fault int my_fault_type = VM_PROT_READ; kern_return_t kr; - if (not_in_kdp) { panic("kdp_lightweight_fault called from outside of debugger context"); } @@ -5933,40 +6115,27 @@ kdp_lightweight_fault(vm_map_t map, vm_offset_t cur_target_addr, uint32_t *fault return 0; } - if (m->laundry || m->busy || m->pageout || m->absent || m->error || m->cleaning || - m->overwriting || m->restart || m->unusual) { - return 0; - } - - assert(!m->private); - if (m->private) { - return 0; - } - - assert(!m->fictitious); - if (m->fictitious) { + if (m->vmp_laundry || m->vmp_busy || m->vmp_free_when_done || m->vmp_absent || m->vmp_error || m->vmp_cleaning || + m->vmp_overwriting || m->vmp_restart || m->vmp_unusual) { return 0; } - assert(!m->encrypted); - if (m->encrypted) { + assert(!m->vmp_private); + if (m->vmp_private) { return 0; } - assert(!m->encrypted_cleaning); - if (m->encrypted_cleaning) { + assert(!m->vmp_fictitious); + if (m->vmp_fictitious) { return 0; } - assert(!m->compressor); - if (m->compressor) { + assert(m->vmp_q_state != VM_PAGE_USED_BY_COMPRESSOR); + if (m->vmp_q_state == VM_PAGE_USED_BY_COMPRESSOR) { return 0; } - if (fault_results) { - *fault_results |= kThreadFaultedBT; - } - return ptoa(m->phys_page); + return ptoa(VM_PAGE_GET_PHYS_PAGE(m)); } compressor_external_state = VM_EXTERNAL_STATE_UNKNOWN; @@ -5977,9 +6146,6 @@ kdp_lightweight_fault(vm_map_t map, vm_offset_t cur_target_addr, uint32_t *fault kdp_compressor_decompressed_page_ppnum, &my_fault_type, compressor_flags, &compressed_count_delta); if (kr == KERN_SUCCESS) { - if (fault_results) { - *fault_results |= kThreadDecompressedBT; - } return kdp_compressor_decompressed_page_paddr; } else { return 0; @@ -5994,29 +6160,29 @@ kdp_lightweight_fault(vm_map_t map, vm_offset_t cur_target_addr, uint32_t *fault object_offset += object->vo_shadow_offset; object = object->shadow; } -#endif /* 0 */ -} +} -#define CODE_SIGNING_CHUNK_SIZE 4096 -void -vm_page_validate_cs_mapped( - vm_page_t page, - const void *kaddr) +/* + * vm_page_validate_cs_fast(): + * Performs a few quick checks to determine if the page's code signature + * really needs to be fully validated. It could: + * 1. have been modified (i.e. automatically tainted), + * 2. have already been validated, + * 3. have already been found to be tainted, + * 4. no longer have a backing store. + * Returns FALSE if the page needs to be fully validated. + */ +static boolean_t +vm_page_validate_cs_fast( + vm_page_t page) { - vm_object_t object; - vm_object_offset_t offset, offset_in_page; - kern_return_t kr; - memory_object_t pager; - void *blobs; - boolean_t validated; - unsigned tainted; - int num_chunks, num_chunks_validated; + vm_object_t object; - assert(page->busy); - vm_object_lock_assert_exclusive(page->object); + object = VM_PAGE_OBJECT(page); + vm_object_lock_assert_held(object); - if (page->wpmapped && !page->cs_tainted) { + if (page->vmp_wpmapped && !page->vmp_cs_tainted) { /* * This page was mapped for "write" access sometime in the * past and could still be modifiable in the future. @@ -6024,77 +6190,120 @@ vm_page_validate_cs_mapped( * [ If the page was already found to be "tainted", no * need to re-validate. ] */ - page->cs_validated = TRUE; - page->cs_tainted = TRUE; + vm_object_lock_assert_exclusive(object); + page->vmp_cs_validated = TRUE; + page->vmp_cs_tainted = TRUE; if (cs_debug) { - printf("CODESIGNING: vm_page_validate_cs: " + printf("CODESIGNING: %s: " "page %p obj %p off 0x%llx " "was modified\n", - page, page->object, page->offset); + __FUNCTION__, + page, object, page->vmp_offset); } vm_cs_validated_dirtied++; } - if (page->cs_validated || page->cs_tainted) { - return; + if (page->vmp_cs_validated || page->vmp_cs_tainted) { + return TRUE; } + vm_object_lock_assert_exclusive(object); - vm_cs_validates++; +#if CHECK_CS_VALIDATION_BITMAP + kern_return_t kr; - object = page->object; - assert(object->code_signed); - offset = page->offset; + kr = vnode_pager_cs_check_validation_bitmap( + object->pager, + page->vmp_offset + object->paging_offset, + CS_BITMAP_CHECK); + if (kr == KERN_SUCCESS) { + page->vmp_cs_validated = TRUE; + page->vmp_cs_tainted = FALSE; + vm_cs_bitmap_validated++; + return TRUE; + } +#endif /* CHECK_CS_VALIDATION_BITMAP */ if (!object->alive || object->terminating || object->pager == NULL) { /* * The object is terminating and we don't have its pager * so we can't validate the data... */ - return; + return TRUE; } + + /* we need to really validate this page */ + vm_object_lock_assert_exclusive(object); + return FALSE; +} + +void +vm_page_validate_cs_mapped_slow( + vm_page_t page, + const void *kaddr) +{ + vm_object_t object; + memory_object_offset_t mo_offset; + memory_object_t pager; + struct vnode *vnode; + boolean_t validated; + unsigned tainted; + + assert(page->vmp_busy); + object = VM_PAGE_OBJECT(page); + vm_object_lock_assert_exclusive(object); + + vm_cs_validates++; + /* * Since we get here to validate a page that was brought in by * the pager, we know that this pager is all setup and ready * by now. */ + assert(object->code_signed); assert(!object->internal); assert(object->pager != NULL); assert(object->pager_ready); pager = object->pager; assert(object->paging_in_progress); - kr = vnode_pager_get_object_cs_blobs(pager, &blobs); - if (kr != KERN_SUCCESS) { - blobs = NULL; - } + vnode = vnode_pager_lookup_vnode(pager); + mo_offset = page->vmp_offset + object->paging_offset; /* verify the SHA1 hash for this page */ - num_chunks_validated = 0; - for (offset_in_page = 0, num_chunks = 0; - offset_in_page < PAGE_SIZE_64; - offset_in_page += CODE_SIGNING_CHUNK_SIZE, num_chunks++) { - tainted = 0; - validated = cs_validate_page(blobs, - pager, - (object->paging_offset + - offset + - offset_in_page), - (const void *)((const char *)kaddr - + offset_in_page), - &tainted); - if (validated) { - num_chunks_validated++; - } - if (tainted & CS_VALIDATE_TAINTED) { - page->cs_tainted = TRUE; - } - if (tainted & CS_VALIDATE_NX) { - page->cs_nx = TRUE; - } + tainted = 0; + validated = cs_validate_range(vnode, + pager, + mo_offset, + (const void *)((const char *)kaddr), + PAGE_SIZE_64, + &tainted); + + if (tainted & CS_VALIDATE_TAINTED) { + page->vmp_cs_tainted = TRUE; } - /* page is validated only if all its chunks are */ - if (num_chunks_validated == num_chunks) { - page->cs_validated = TRUE; + if (tainted & CS_VALIDATE_NX) { + page->vmp_cs_nx = TRUE; + } + if (validated) { + page->vmp_cs_validated = TRUE; + } + +#if CHECK_CS_VALIDATION_BITMAP + if (page->vmp_cs_validated && !page->vmp_cs_tainted) { + vnode_pager_cs_check_validation_bitmap(object->pager, + mo_offset, + CS_BITMAP_SET); + } +#endif /* CHECK_CS_VALIDATION_BITMAP */ +} + +void +vm_page_validate_cs_mapped( + vm_page_t page, + const void *kaddr) +{ + if (!vm_page_validate_cs_fast(page)) { + vm_page_validate_cs_mapped_slow(page, kaddr); } } @@ -6111,58 +6320,23 @@ vm_page_validate_cs( boolean_t busy_page; boolean_t need_unmap; - vm_object_lock_assert_held(page->object); - - if (page->wpmapped && !page->cs_tainted) { - vm_object_lock_assert_exclusive(page->object); - - /* - * This page was mapped for "write" access sometime in the - * past and could still be modifiable in the future. - * Consider it tainted. - * [ If the page was already found to be "tainted", no - * need to re-validate. ] - */ - page->cs_validated = TRUE; - page->cs_tainted = TRUE; - if (cs_debug) { - printf("CODESIGNING: vm_page_validate_cs: " - "page %p obj %p off 0x%llx " - "was modified\n", - page, page->object, page->offset); - } - vm_cs_validated_dirtied++; - } + object = VM_PAGE_OBJECT(page); + vm_object_lock_assert_held(object); - if (page->cs_validated || page->cs_tainted) { + if (vm_page_validate_cs_fast(page)) { return; } + vm_object_lock_assert_exclusive(object); - if (page->slid) { - panic("vm_page_validate_cs(%p): page is slid\n", page); - } - assert(!page->slid); - -#if CHECK_CS_VALIDATION_BITMAP - if ( vnode_pager_cs_check_validation_bitmap( page->object->pager, trunc_page(page->offset + page->object->paging_offset), CS_BITMAP_CHECK ) == KERN_SUCCESS) { - page->cs_validated = TRUE; - page->cs_tainted = FALSE; - vm_cs_bitmap_validated++; - return; - } -#endif - vm_object_lock_assert_exclusive(page->object); - - object = page->object; assert(object->code_signed); - offset = page->offset; + offset = page->vmp_offset; - busy_page = page->busy; + busy_page = page->vmp_busy; if (!busy_page) { /* keep page busy while we map (and unlock) the VM object */ - page->busy = TRUE; + page->vmp_busy = TRUE; } - + /* * Take a paging reference on the VM object * to protect it from collapse or bypass, @@ -6183,20 +6357,15 @@ vm_page_validate_cs( &koffset, &need_unmap); if (kr != KERN_SUCCESS) { - panic("vm_page_validate_cs: could not map page: 0x%x\n", kr); + panic("%s: could not map page: 0x%x\n", __FUNCTION__, kr); } kaddr = CAST_DOWN(vm_offset_t, koffset); /* validate the mapped page */ - vm_page_validate_cs_mapped(page, (const void *) kaddr); + vm_page_validate_cs_mapped_slow(page, (const void *) kaddr); -#if CHECK_CS_VALIDATION_BITMAP - if ( page->cs_validated == TRUE && page->cs_tainted == FALSE ) { - vnode_pager_cs_check_validation_bitmap( object->pager, trunc_page( offset + object->paging_offset), CS_BITMAP_SET ); - } -#endif - assert(page->busy); - assert(object == page->object); + assert(page->vmp_busy); + assert(object == VM_PAGE_OBJECT(page)); vm_object_lock_assert_exclusive(object); if (!busy_page) { @@ -6217,26 +6386,26 @@ vm_page_validate_cs_mapped_chunk( vm_page_t page, const void *kaddr, vm_offset_t chunk_offset, + vm_size_t chunk_size, boolean_t *validated_p, unsigned *tainted_p) { vm_object_t object; vm_object_offset_t offset, offset_in_page; - kern_return_t kr; memory_object_t pager; - void *blobs; + struct vnode *vnode; boolean_t validated; unsigned tainted; *validated_p = FALSE; *tainted_p = 0; - assert(page->busy); - vm_object_lock_assert_exclusive(page->object); + assert(page->vmp_busy); + object = VM_PAGE_OBJECT(page); + vm_object_lock_assert_exclusive(object); - object = page->object; assert(object->code_signed); - offset = page->offset; + offset = page->vmp_offset; if (!object->alive || object->terminating || object->pager == NULL) { /* @@ -6256,25 +6425,22 @@ vm_page_validate_cs_mapped_chunk( pager = object->pager; assert(object->paging_in_progress); - kr = vnode_pager_get_object_cs_blobs(pager, &blobs); - if (kr != KERN_SUCCESS) { - blobs = NULL; - } + vnode = vnode_pager_lookup_vnode(pager); /* verify the signature for this chunk */ offset_in_page = chunk_offset; assert(offset_in_page < PAGE_SIZE); - assert((offset_in_page & (CODE_SIGNING_CHUNK_SIZE-1)) == 0); tainted = 0; - validated = cs_validate_page(blobs, - pager, - (object->paging_offset + - offset + - offset_in_page), - (const void *)((const char *)kaddr + validated = cs_validate_range(vnode, + pager, + (object->paging_offset + + offset + + offset_in_page), + (const void *)((const char *)kaddr + offset_in_page), - &tainted); + chunk_size, + &tainted); if (validated) { *validated_p = TRUE; } @@ -6282,3 +6448,98 @@ vm_page_validate_cs_mapped_chunk( *tainted_p = tainted; } } + +static void vm_rtfrecord_lock(void) { + lck_spin_lock(&vm_rtfr_slock); +} + +static void vm_rtfrecord_unlock(void) { + lck_spin_unlock(&vm_rtfr_slock); +} + +unsigned int vmrtfaultinfo_bufsz(void) { + return (vmrtf_num_records * sizeof(vm_rtfault_record_t)); +} + +#include + +static void vm_record_rtfault(thread_t cthread, uint64_t fstart, vm_map_offset_t fault_vaddr, int type_of_fault) { + uint64_t fend = mach_continuous_time(); + + uint64_t cfpc = 0; + uint64_t ctid = cthread->thread_id; + uint64_t cupid = get_current_unique_pid(); + + uintptr_t bpc = 0; + uint32_t bfrs = 0; + bool u64 = false; + + /* Capture a single-frame backtrace; this extracts just the program + * counter at the point of the fault into "bpc", and should perform no + * further user stack traversals, thus avoiding copyin()s and further + * faults. + */ + int btr = backtrace_thread_user(cthread, &bpc, 1U, &bfrs, &u64); + + if ((btr == 0) && (bfrs > 0)) { + cfpc = bpc; + } + + assert((fstart != 0) && fend >= fstart); + vm_rtfrecord_lock(); + assert(vmrtfrs.vmrtfr_curi <= vmrtfrs.vmrtfr_maxi); + + vmrtfrs.vmrtf_total++; + vm_rtfault_record_t *cvmr = &vmrtfrs.vm_rtf_records[vmrtfrs.vmrtfr_curi++]; + + cvmr->rtfabstime = fstart; + cvmr->rtfduration = fend - fstart; + cvmr->rtfaddr = fault_vaddr; + cvmr->rtfpc = cfpc; + cvmr->rtftype = type_of_fault; + cvmr->rtfupid = cupid; + cvmr->rtftid = ctid; + + if (vmrtfrs.vmrtfr_curi > vmrtfrs.vmrtfr_maxi) { + vmrtfrs.vmrtfr_curi = 0; + } + + vm_rtfrecord_unlock(); +} + +int vmrtf_extract(uint64_t cupid, __unused boolean_t isroot, int vrecordsz, void *vrecords, int *vmrtfrv) { + vm_rtfault_record_t *cvmrd = vrecords; + size_t residue = vrecordsz; + int numextracted = 0; + boolean_t early_exit = FALSE; + + vm_rtfrecord_lock(); + + for (int vmfi = 0; vmfi <= vmrtfrs.vmrtfr_maxi; vmfi++) { + + if (residue < sizeof(vm_rtfault_record_t)) { + early_exit = TRUE; + break; + } + + if (vmrtfrs.vm_rtf_records[vmfi].rtfupid != cupid) { +#if DEVELOPMENT || DEBUG + if (isroot == FALSE) { + continue; + } +#else + continue; +#endif /* DEVDEBUG */ + } + + *cvmrd = vmrtfrs.vm_rtf_records[vmfi]; + cvmrd++; + residue -= sizeof(vm_rtfault_record_t); + numextracted++; + } + + vm_rtfrecord_unlock(); + + *vmrtfrv = numextracted; + return (early_exit); +}