X-Git-Url: https://git.saurik.com/apple/xnu.git/blobdiff_plain/2dced7af2b695f87fe26496a3e73c219b7880cbc..9d749ea394c01276fa19e397e70f46858e849c76:/osfmk/vm/vm_fault.c diff --git a/osfmk/vm/vm_fault.c b/osfmk/vm/vm_fault.c index 663199f2a..256c70dfe 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 @@ -129,6 +132,8 @@ 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; @@ -152,6 +157,7 @@ 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 +168,7 @@ 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, @@ -193,7 +200,6 @@ unsigned long vm_cs_bitmap_validated = 0; 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; @@ -225,7 +231,7 @@ 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; @@ -234,25 +240,13 @@ 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); } @@ -272,14 +266,14 @@ 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); @@ -310,8 +304,8 @@ struct { 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 +530,7 @@ 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->laundry && !m->busy && !m->no_cache && (m->vm_page_q_state != VM_PAGE_ON_THROTTLED_Q) && !m->fictitious && !m->absent) { page_run[pages_in_run++] = m; /* @@ -550,7 +544,7 @@ 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 +581,15 @@ 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 +601,9 @@ 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); @@ -630,7 +624,7 @@ 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; @@ -640,7 +634,7 @@ vm_page_throttled(boolean_t page_kept) 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); @@ -752,6 +746,9 @@ 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... @@ -765,7 +762,7 @@ 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; @@ -786,29 +783,27 @@ vm_fault_zero_page(vm_page_t m, boolean_t no_zero_fill) 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(object != kernel_object); + //assert(m->pageq.next == 0 && m->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, pageq); + m->vm_page_q_state = VM_PAGE_ON_THROTTLED_Q; vm_page_throttled_count++; } vm_page_unlock_queues(); @@ -827,7 +822,7 @@ 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 +854,7 @@ 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 +882,7 @@ 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) @@ -916,50 +911,25 @@ 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 +937,16 @@ 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 +955,7 @@ vm_fault_page( interruptible = fault_info->interruptible; interruptible_state = thread_interrupt_level(interruptible); - + /* * INVARIANTS (through entire routine): * @@ -1031,6 +1001,14 @@ 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 @@ -1122,12 +1100,12 @@ vm_fault_page( continue; } if (m->laundry) { - m->pageout = FALSE; + m->free_when_done = FALSE; - if (!m->cleaning) + if (!m->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 ! */ @@ -1216,7 +1194,7 @@ 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 +1222,7 @@ 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' */ @@ -1284,10 +1262,7 @@ vm_fault_page( m->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 +1280,7 @@ vm_fault_page( vm_object_unlock(object); object = next_object; vm_object_paging_begin(object); - + /* * reset to default type of fault */ @@ -1340,7 +1315,7 @@ 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); @@ -1364,7 +1339,7 @@ vm_fault_page( return (VM_FAULT_RETRY); } } - if (type_of_fault == NULL && m->speculative && + if (type_of_fault == NULL && (m->vm_page_q_state == VM_PAGE_ON_SPECULATIVE_Q) && !(fault_info != NULL && fault_info->stealth)) { /* * If we were passed a non-NULL pointer for @@ -1382,36 +1357,13 @@ vm_fault_page( * the page in the speculative queue. */ vm_page_lockspin_queues(); - if (m->speculative) - vm_page_queues_remove(m); + if (m->vm_page_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 @@ -1442,7 +1394,7 @@ vm_fault_page( m->busy = TRUE; break; } - + /* * we get here when there is no page present in the object at @@ -1451,6 +1403,7 @@ 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 +1423,13 @@ 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 +1437,7 @@ 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 +1540,16 @@ 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 @@ -1615,7 +1569,7 @@ vm_fault_page( } } assert(m->busy); - + m->absent = TRUE; pager = object->pager; @@ -1625,7 +1579,7 @@ 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); @@ -1658,7 +1612,7 @@ vm_fault_page( case KERN_SUCCESS: m->absent = FALSE; m->dirty = TRUE; - if ((m->object->wimg_bits & + if ((object->wimg_bits & VM_WIMG_MASK) != VM_WIMG_USE_DEFAULT) { /* @@ -1668,7 +1622,7 @@ 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; } @@ -1715,7 +1669,7 @@ 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 +1681,7 @@ 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 +1698,10 @@ vm_fault_page( * so we can release the object lock. */ + if (object->object_slid == TRUE) { + set_thread_rwlock_boost(); + } + vm_object_unlock(object); /* @@ -1786,8 +1744,8 @@ 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 +1786,10 @@ vm_fault_page( #endif vm_object_lock(object); + if (object->object_slid == TRUE) { + clear_thread_rwlock_boost(); + } + data_requested: if (rc != KERN_SUCCESS) { @@ -1864,7 +1826,7 @@ 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, @@ -1894,7 +1856,7 @@ 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 @@ -1931,13 +1893,13 @@ 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 +1909,7 @@ 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); @@ -2010,16 +1972,9 @@ dont_look_for_page: assert(m->busy && !m->absent); assert((first_m == VM_PAGE_NULL) || (first_m->busy && !first_m->absent && - !first_m->active && !first_m->inactive)); + !first_m->active && !first_m->inactive && !first_m->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, @@ -2083,7 +2038,7 @@ 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); @@ -2113,7 +2068,7 @@ dont_look_for_page: * avoid the pmap_disconnect() call. */ if (m->pmapped) - pmap_disconnect(m->phys_page); + pmap_disconnect(VM_PAGE_GET_PHYS_PAGE(m)); if (m->clustered) { VM_PAGE_COUNT_AS_PAGEIN(m); @@ -2126,6 +2081,16 @@ dont_look_for_page: */ 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); @@ -2144,7 +2109,7 @@ dont_look_for_page: */ VM_PAGE_FREE(first_m); first_m = VM_PAGE_NULL; - + /* * and replace it with the * page we just copied into @@ -2159,8 +2124,8 @@ 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); @@ -2257,12 +2222,7 @@ 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) { PAGE_ASSERT_WAIT(copy_m, interruptible); @@ -2337,7 +2297,7 @@ 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 @@ -2345,7 +2305,7 @@ dont_look_for_page: * pmaps use it.) */ if (m->pmapped) - pmap_disconnect(m->phys_page); + pmap_disconnect(VM_PAGE_GET_PHYS_PAGE(m)); if (m->clustered) { VM_PAGE_COUNT_AS_PAGEIN(m); @@ -2357,11 +2317,8 @@ 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); @@ -2371,71 +2328,8 @@ dont_look_for_page: 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); @@ -2506,7 +2400,7 @@ 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 +2415,15 @@ 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; @@ -2579,10 +2475,10 @@ 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_obj)->code_signed /*3*/ && \ (!(page)->cs_validated || (page)->wpmapped /*4*/)) @@ -2595,6 +2491,7 @@ backoff: * 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; @@ -2607,6 +2504,7 @@ vm_fault_enter(vm_page_t m, vm_prot_t caller_prot, boolean_t wired, boolean_t change_wiring, + vm_tag_t wire_tag, boolean_t no_cache, boolean_t cs_bypass, __unused int user_tag, @@ -2620,24 +2518,31 @@ vm_fault_enter(vm_page_t m, boolean_t map_is_switched, map_is_switch_protected; int cs_enforcement_enabled; vm_prot_t fault_type; - + vm_object_t object; + fault_type = change_wiring ? VM_PROT_NONE : caller_prot; + object = VM_PAGE_OBJECT(m); - vm_object_lock_assert_held(m->object); -#if DEBUG - lck_mtx_assert(&vm_page_queue_lock, LCK_MTX_ASSERT_NOTOWNED); -#endif /* DEBUG */ + vm_object_lock_assert_held(object); - if (m->phys_page == vm_page_guard_addr) { +#if KASAN + if (pmap == kernel_pmap) { + kasan_notify_address(vaddr, PAGE_SIZE); + } +#endif + + LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_NOTOWNED); + + if (VM_PAGE_GET_PHYS_PAGE(m) == vm_page_guard_addr) { assert(m->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->wpmapped) { /* * This is not a "write" fault, so we * might not have taken the object lock @@ -2649,7 +2554,13 @@ 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) { @@ -2661,11 +2572,11 @@ vm_fault_enter(vm_page_t m, * 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); @@ -2681,14 +2592,14 @@ 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 (VM_FAULT_NEED_CS_VALIDATION(pmap, m, object)) { + vm_object_lock_assert_exclusive(object); if (m->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 */ vm_page_validate_cs(m); } @@ -2699,21 +2610,21 @@ vm_fault_enter(vm_page_t m, 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); - 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; @@ -2725,11 +2636,36 @@ vm_fault_enter(vm_page_t m, return KERN_CODESIGN_ERROR; } + if (cs_enforcement_enabled && + !m->cs_validated && + (prot & VM_PROT_EXECUTE) && + !(caller_prot & VM_PROT_EXECUTE)) { + /* + * FOURK PAGER: + * This page has not been validated and will not be + * allowed to be mapped for "execute". + * But the caller did not request "execute" access for this + * fault, so we should not raise a code-signing violation + * (and possibly kill the process) below. + * Instead, let's just remove the "execute" access request. + * + * This can happen on devices with a 4K page size if a 16K + * page contains a mix of signed&executable and + * unsigned&non-executable 4K pages, making the whole 16K + * mapping "executable". + */ + if (!pmap_has_prot_policy(prot)) { + prot &= ~VM_PROT_EXECUTE; + } else { + assert(cs_bypass); + } + } + /* 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 @@ -2742,20 +2678,20 @@ vm_fault_enter(vm_page_t m, (!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 + * 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 + * 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 + * 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)) )) - )) + )) { /* We will have a tainted page. Have to handle the special case * of a switched map now. If the map is not switched, standard @@ -2764,23 +2700,23 @@ 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", + printf("vm_fault: signed: %s validate: %s tainted: %s wpmapped: %s slid: %s prot: 0x%x\n", + object->code_signed ? "yes" : "no", m->cs_validated ? "yes" : "no", m->cs_tainted ? "yes" : "no", m->wpmapped ? "yes" : "no", m->slid ? "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 +2729,8 @@ 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 +2743,15 @@ vm_fault_enter(vm_page_t m, procname = proc_name_address(task->bsd_info); /* get file's VM object */ - file_object = m->object; + file_object = object; file_offset = m->offset; - for (shadow = file_object->shadow; + 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 +2797,8 @@ 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 slid:%d dirty:%d depth:%d)\n", pid, procname, (addr64_t) vaddr, file_offset, (pathname ? pathname : ""), @@ -2870,12 +2810,108 @@ vm_fault_enter(vm_page_t m, ? "==" : "!="), mtime.tv_sec, mtime.tv_nsec, - m->object->code_signed, + object->code_signed, m->cs_validated, m->cs_tainted, + m->cs_nx, m->wpmapped, - m->slid); - if (file_object != m->object) { + m->slid, + m->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->cs_validated); + ceri->ceri_page_codesig_tainted = (m->cs_tainted); + ceri->ceri_page_codesig_nx = (m->cs_nx); + ceri->ceri_page_wpmapped = (m->wpmapped); + ceri->ceri_page_slid = (m->slid); + ceri->ceri_page_dirty = (m->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_slid) { + 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 slid:%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->cs_validated, + m->cs_tainted, + m->cs_nx, + m->wpmapped, + m->slid, + m->dirty, + shadow_depth); + } + + if (file_object != object) { vm_object_unlock(file_object); } if (pathname_len != 0) { @@ -2886,11 +2922,12 @@ vm_fault_enter(vm_page_t m, } else { /* proceed with the invalid page */ kr = KERN_SUCCESS; - if (!m->cs_validated) { + if (!m->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 +2942,6 @@ 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; @@ -2932,7 +2968,7 @@ vm_fault_enter(vm_page_t m, } #endif } - + } else { /* proceed with the valid page */ kr = KERN_SUCCESS; @@ -2959,19 +2995,23 @@ MACRO_END * the page queues. Change wiring * case is obvious. */ - assert(m->compressor || m->object != compressor_object); - if (m->compressor) { + assert((m->vm_page_q_state == VM_PAGE_USED_BY_COMPRESSOR) || object != compressor_object); + +#if CONFIG_BACKGROUND_QUEUE + vm_page_update_background_state(m); +#endif + if (m->vm_page_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 +3019,33 @@ 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->vm_page_q_state == VM_PAGE_NOT_ON_Q) || + (m->vm_page_q_state == VM_PAGE_ON_SPECULATIVE_Q) || + (m->vm_page_q_state == VM_PAGE_ON_INACTIVE_CLEANED_Q) || + ((m->vm_page_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->vm_page_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 +3058,7 @@ 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 +3068,13 @@ 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; + vm_page_queue_enter(&lq->vpl_queue, m, + vm_page_t, pageq); + m->vm_page_q_state = VM_PAGE_ON_ACTIVE_LOCAL_Q; m->local_id = lid; lq->vpl_count++; - - if (m->object->internal) + + if (object->internal) lq->vpl_internal_count++; else lq->vpl_external_count++; @@ -3061,16 +3110,15 @@ MACRO_END * page queue lock */ if (!VM_PAGE_WIRED(m)) { - if (m->clean_queue) { - vm_page_queues_remove(m); + if (m->vm_page_q_state == VM_PAGE_ON_INACTIVE_CLEANED_Q) { + vm_page_queues_remove(m, FALSE); vm_pageout_cleaned_reactivated++; vm_pageout_cleaned_fault_reactivated++; } - 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 @@ -3081,7 +3129,7 @@ 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 && @@ -3089,12 +3137,10 @@ MACRO_END m->no_cache)) { m->no_cache = TRUE; - if (!m->speculative) + if (m->vm_page_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); } } @@ -3114,16 +3160,16 @@ 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) { + 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 @@ -3131,19 +3177,21 @@ MACRO_END * just below */ m->pmapped = TRUE; - + if (!m->xpmapped) { m->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 @@ -3155,21 +3203,25 @@ 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); + ppnum_t phys_page = VM_PAGE_GET_PHYS_PAGE(m); + + pmap_lock_phys_page(phys_page); m->pmapped = TRUE; - pmap_unlock_phys_page(m->phys_page); + pmap_unlock_phys_page(phys_page); } } if (vm_page_is_slideable(m)) { boolean_t was_busy = m->busy; - vm_object_lock_assert_exclusive(m->object); + vm_object_lock_assert_exclusive(object); m->busy = TRUE; kr = vm_page_slide(m, 0); @@ -3191,31 +3243,35 @@ MACRO_END if (fault_type & VM_PROT_WRITE) { if (m->wpmapped == FALSE) { - vm_object_lock_assert_exclusive(m->object); - + 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->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); /* Prevent a deadlock by not * holding the object lock if we need to wait for a page in @@ -3224,6 +3280,19 @@ 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) { @@ -3233,10 +3302,10 @@ 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; @@ -3245,23 +3314,25 @@ 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; - vm_object_lock_assert_exclusive(m->object); + vm_object_lock_assert_exclusive(object); m->busy = TRUE; - vm_object_unlock(m->object); - + vm_object_unlock(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); @@ -3270,6 +3341,8 @@ MACRO_END } vm_pmap_enter_blocked++; } + + kr = pe_result; } after_the_pmap_enter: @@ -3281,13 +3354,14 @@ 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 */); } } @@ -3307,13 +3381,14 @@ 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, @@ -3322,11 +3397,26 @@ 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( + vm_map_t map, + vm_map_offset_t vaddr, + 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( @@ -3334,6 +3424,7 @@ vm_fault_internal( 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, @@ -3352,6 +3443,7 @@ 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; @@ -3360,6 +3452,7 @@ vm_fault_internal( boolean_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; @@ -3371,28 +3464,44 @@ vm_fault_internal( vm_object_t top_object = VM_OBJECT_NULL; 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); } - + interruptible_state = thread_interrupt_level(interruptible); fault_type = (change_wiring ? VM_PROT_NONE : caller_prot); @@ -3408,6 +3517,14 @@ 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: /* * assume we will hit a page in the cache @@ -3430,7 +3547,6 @@ RetryFault: &fault_info, &real_map); - if (kr != KERN_SUCCESS) { vm_map_unlock_read(map); goto done; @@ -3504,6 +3620,24 @@ 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 @@ -3517,6 +3651,13 @@ 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 */ @@ -3531,8 +3672,11 @@ RetryFault: } m = vm_page_lookup(cur_object, cur_offset); + m_object = NULL; if (m != VM_PAGE_NULL) { + m_object = cur_object; + if (m->busy) { wait_result_t result; @@ -3582,7 +3726,7 @@ RetryFault: continue; } } - if (m->pageout_queue && m->object->internal && COMPRESSED_PAGER_IS_ACTIVE) { + if ((m->vm_page_q_state == VM_PAGE_ON_PAGEOUT_Q) && m_object->internal) { /* * m->busy == TRUE and the object is locked exclusively * if m->pageout_queue == TRUE after we acquire the @@ -3592,9 +3736,11 @@ 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->vm_page_q_state == VM_PAGE_ON_PAGEOUT_Q) { vm_pageout_throttle_up(m); vm_page_unlock_queues(); @@ -3660,12 +3806,10 @@ 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 */ @@ -3677,7 +3821,7 @@ reclaimed_from_pageout: */ 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); @@ -3687,82 +3831,7 @@ reclaimed_from_pageout: kr = KERN_MEMORY_ERROR; goto done; } - - 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)) { + if (vm_page_is_slideable(m)) { /* * We might need to slide this page, and so, * we want to hold the VM object exclusively. @@ -3788,8 +3857,9 @@ reclaimed_from_pageout: goto RetryFault; } } + assert(m_object == VM_PAGE_OBJECT(m)); - 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: /* @@ -3844,6 +3914,20 @@ 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) { /* @@ -3873,6 +3957,8 @@ 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 @@ -3894,6 +3980,7 @@ FastPmapEnter: caller_prot, wired, change_wiring, + wire_tag, fault_info.no_cache, fault_info.cs_bypass, fault_info.user_tag, @@ -3908,6 +3995,7 @@ FastPmapEnter: caller_prot, wired, change_wiring, + wire_tag, fault_info.no_cache, fault_info.cs_bypass, fault_info.user_tag, @@ -3915,14 +4003,28 @@ FastPmapEnter: 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_slid) + 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->offset, get_current_unique_pid(), 0); + + DTRACE_VM6(real_fault, vm_map_offset_t, real_vaddr, vm_map_offset_t, m->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); + vm_object_lock_assert_exclusive(m_object); m->dirty = TRUE; } } @@ -3943,7 +4045,7 @@ 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)) { /* @@ -3979,7 +4081,7 @@ FastPmapEnter: (void)pmap_enter_options( pmap, vaddr, 0, 0, 0, 0, 0, PMAP_OPTIONS_NOENTER, NULL); - + need_retry = FALSE; goto RetryFault; } @@ -4004,15 +4106,16 @@ 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; } @@ -4022,14 +4125,15 @@ 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) { /* @@ -4046,17 +4150,19 @@ 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); - + pmap_disconnect(VM_PAGE_GET_PHYS_PAGE(cur_m)); + if (cur_m->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; @@ -4193,11 +4299,12 @@ 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) { /* @@ -4226,7 +4333,7 @@ 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); @@ -4238,7 +4345,8 @@ FastPmapEnter: cur_object); if (kr != KERN_SUCCESS) { - vm_page_release(m); + vm_page_release(m, FALSE); + m = VM_PAGE_NULL; break; } m->dirty = TRUE; @@ -4291,18 +4399,20 @@ FastPmapEnter: 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; @@ -4335,8 +4445,10 @@ 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); @@ -4351,7 +4463,7 @@ FastPmapEnter: if (vm_backing_store_low) { /* * we are protecting the system from - * backing store exhaustion... + * backing store exhaustion... * must take the slow path if we're * not privileged */ @@ -4382,6 +4494,7 @@ FastPmapEnter: } } m = vm_page_alloc(object, offset); + m_object = NULL; if (m == VM_PAGE_NULL) { /* @@ -4390,10 +4503,11 @@ 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 * @@ -4456,6 +4570,28 @@ 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 @@ -4489,7 +4625,7 @@ 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 */ @@ -4505,7 +4641,7 @@ 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; @@ -4529,11 +4665,13 @@ 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))); } /* @@ -4543,30 +4681,55 @@ 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; @@ -4581,7 +4744,6 @@ 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, @@ -4596,17 +4758,19 @@ 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 @@ -4630,17 +4794,19 @@ 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 @@ -4659,20 +4825,31 @@ 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 @@ -4680,14 +4857,16 @@ 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); @@ -4711,6 +4890,7 @@ handle_copy_delay: caller_prot, wired, change_wiring, + wire_tag, fault_info.no_cache, fault_info.cs_bypass, fault_info.user_tag, @@ -4725,6 +4905,7 @@ handle_copy_delay: caller_prot, wired, change_wiring, + wire_tag, fault_info.no_cache, fault_info.cs_bypass, fault_info.user_tag, @@ -4732,21 +4913,39 @@ handle_copy_delay: 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_slid) + 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->offset, get_current_unique_pid(), 0); + + DTRACE_VM6(real_fault, vm_map_offset_t, real_vaddr, vm_map_offset_t, m->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); + vm_object_lock_assert_exclusive(m_object); m->dirty = TRUE; } } @@ -4756,9 +4955,9 @@ 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 @@ -4770,7 +4969,7 @@ handle_copy_delay: if ((fault_type & VM_PROT_EXECUTE) && (!pmap_eligible_for_execute((ppnum_t)(object->vo_shadow_offset >> 12)))) { - vm_map_verify_done(map, &version); + vm_map_unlock_read(map); if (real_map != map) vm_map_unlock(real_map); @@ -4803,8 +5002,8 @@ 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)); @@ -4815,19 +5014,22 @@ 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; if (!object->pager_created && - object->phys_contiguous) { + object->phys_contiguous && + VME_OFFSET(entry) == 0 && + (entry->vme_end - entry->vme_start == object->vo_size) && + VM_MAP_PAGE_ALIGNED(entry->vme_start, (object->vo_size-1))) { superpage = VM_MEM_SUPERPAGE; } else { superpage = 0; @@ -4848,38 +5050,57 @@ 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); + 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); @@ -4887,7 +5108,6 @@ handle_copy_delay: #undef RELEASE_PAGE - kr = KERN_SUCCESS; done: thread_interrupt_level(interruptible_state); @@ -4913,10 +5133,10 @@ done: } } } - 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, kr, type_of_fault, 0); @@ -4934,19 +5154,19 @@ 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; } @@ -4957,7 +5177,7 @@ 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); /* @@ -4966,14 +5186,14 @@ 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)), @@ -4986,7 +5206,7 @@ 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; @@ -5008,10 +5228,11 @@ 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; + unsigned int unwired_pages; object = (entry->is_sub_map) ? VM_OBJECT_NULL : VME_OBJECT(entry); @@ -5041,6 +5262,8 @@ vm_fault_unwire( 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 * get their mappings from the physical map system. @@ -5050,11 +5273,11 @@ 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; @@ -5087,7 +5310,7 @@ 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); @@ -5097,7 +5320,7 @@ 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. */ @@ -5119,16 +5342,19 @@ 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); @@ -5136,9 +5362,10 @@ 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; } @@ -5154,9 +5381,12 @@ 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)); + } } /* @@ -5183,7 +5413,8 @@ 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, @@ -5191,7 +5422,7 @@ 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; @@ -5278,19 +5509,15 @@ 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) || + if ((m == VM_PAGE_NULL) || (m->busy) || (m->unusual && ( m->error || m->restart || m->absent))) { GIVE_UP; } - ASSERT_PAGE_DECRYPTED(m); - if (m->fictitious && - m->phys_page == vm_page_guard_addr) { + 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... @@ -5301,11 +5528,11 @@ 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(); /* @@ -5332,10 +5559,11 @@ vm_fault_wire_fast( pmap_addr, prot, prot, - TRUE, - FALSE, - FALSE, - FALSE, + TRUE, /* wired */ + FALSE, /* change_wiring */ + wire_tag, + FALSE, /* no_cache */ + FALSE, /* cs_bypass */ VME_ALIAS(entry), ((entry->iokit_acct || (!entry->is_sub_map && !entry->use_pmap)) @@ -5343,6 +5571,10 @@ vm_fault_wire_fast( : 0), NULL, &type_of_fault); + if (kr != KERN_SUCCESS) { + RELEASE_PAGE(m); + GIVE_UP; + } done: /* @@ -5352,9 +5584,10 @@ 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); + vm_object_lock_assert_exclusive(object); m->dirty = TRUE; } } else { @@ -5380,14 +5613,15 @@ 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); @@ -5400,12 +5634,12 @@ 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); } } @@ -5449,7 +5683,7 @@ 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; @@ -5460,6 +5694,7 @@ 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; @@ -5568,7 +5803,8 @@ 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 @@ -5577,7 +5813,7 @@ 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. */ @@ -5586,7 +5822,7 @@ 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); @@ -5628,11 +5864,11 @@ vm_fault_copy( 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); @@ -5666,32 +5902,36 @@ 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 @@ -5735,7 +5975,7 @@ vm_fault_copy( if(!dst_page->dirty){ vm_object_lock(dst_object); SET_PAGE_DIRTY(dst_page, TRUE); - vm_object_unlock(dst_page->object); + vm_object_unlock(dst_object); } } @@ -5745,14 +5985,14 @@ 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){ vm_object_lock(dst_object); SET_PAGE_DIRTY(dst_page, TRUE); - vm_object_unlock(dst_page->object); + vm_object_unlock(dst_object); } } @@ -5762,7 +6002,7 @@ 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); @@ -5776,7 +6016,7 @@ vm_fault_copy( RETURN(KERN_SUCCESS); #undef RETURN - /*NOTREACHED*/ + /*NOTREACHED*/ } #if VM_FAULT_CLASSIFY @@ -5809,7 +6049,7 @@ vm_fault_classify(vm_object_t object, while (TRUE) { m = vm_page_lookup(object, offset); - if (m != VM_PAGE_NULL) { + if (m != VM_PAGE_NULL) { if (m->busy || m->error || m->restart || m->absent) { type = VM_FAULT_TYPE_OTHER; break; @@ -5817,7 +6057,7 @@ vm_fault_classify(vm_object_t object, 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; @@ -5865,12 +6105,8 @@ 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; @@ -5880,7 +6116,6 @@ 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"); } @@ -5929,7 +6164,7 @@ 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 || + if (m->laundry || m->busy || m->free_when_done || m->absent || m->error || m->cleaning || m->overwriting || m->restart || m->unusual) { return 0; } @@ -5944,25 +6179,12 @@ kdp_lightweight_fault(vm_map_t map, vm_offset_t cur_target_addr, uint32_t *fault return 0; } - assert(!m->encrypted); - if (m->encrypted) { - return 0; - } - - assert(!m->encrypted_cleaning); - if (m->encrypted_cleaning) { + assert(m->vm_page_q_state != VM_PAGE_USED_BY_COMPRESSOR); + if (m->vm_page_q_state == VM_PAGE_USED_BY_COMPRESSOR) { return 0; } - assert(!m->compressor); - if (m->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; @@ -5973,9 +6195,6 @@ 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; @@ -5990,27 +6209,24 @@ 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_object_t object; - vm_object_offset_t offset, offset_in_page; - kern_return_t kr; + vm_object_offset_t offset; memory_object_t pager; - void *blobs; + struct vnode *vnode; boolean_t validated; unsigned tainted; - int num_chunks, num_chunks_validated; assert(page->busy); - vm_object_lock_assert_exclusive(page->object); + object = VM_PAGE_OBJECT(page); + vm_object_lock_assert_exclusive(object); if (page->wpmapped && !page->cs_tainted) { /* @@ -6026,7 +6242,7 @@ vm_page_validate_cs_mapped( printf("CODESIGNING: vm_page_validate_cs: " "page %p obj %p off 0x%llx " "was modified\n", - page, page->object, page->offset); + page, object, page->offset); } vm_cs_validated_dirtied++; } @@ -6037,7 +6253,6 @@ vm_page_validate_cs_mapped( vm_cs_validates++; - object = page->object; assert(object->code_signed); offset = page->offset; @@ -6059,37 +6274,26 @@ vm_page_validate_cs_mapped( 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 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, + (object->paging_offset + + offset), + (const void *)((const char *)kaddr), + PAGE_SIZE_64, + &tainted); + + if (tainted & CS_VALIDATE_TAINTED) { + page->cs_tainted = TRUE; } - /* page is validated only if all its chunks are */ - if (num_chunks_validated == num_chunks) { + if (tainted & CS_VALIDATE_NX) { + page->cs_nx = TRUE; + } + + if (validated) { page->cs_validated = TRUE; } } @@ -6107,10 +6311,11 @@ vm_page_validate_cs( boolean_t busy_page; boolean_t need_unmap; - vm_object_lock_assert_held(page->object); + object = VM_PAGE_OBJECT(page); + vm_object_lock_assert_held(object); if (page->wpmapped && !page->cs_tainted) { - vm_object_lock_assert_exclusive(page->object); + vm_object_lock_assert_exclusive(object); /* * This page was mapped for "write" access sometime in the @@ -6125,7 +6330,7 @@ vm_page_validate_cs( printf("CODESIGNING: vm_page_validate_cs: " "page %p obj %p off 0x%llx " "was modified\n", - page, page->object, page->offset); + page, object, page->offset); } vm_cs_validated_dirtied++; } @@ -6139,17 +6344,16 @@ vm_page_validate_cs( } 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) { +#if CHECK_CS_VALIDATION_BITMAP + if ( vnode_pager_cs_check_validation_bitmap( object->pager, trunc_page(page->offset + 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); + vm_object_lock_assert_exclusive(object); - object = page->object; assert(object->code_signed); offset = page->offset; @@ -6158,7 +6362,7 @@ vm_page_validate_cs( /* keep page busy while we map (and unlock) the VM object */ page->busy = TRUE; } - + /* * Take a paging reference on the VM object * to protect it from collapse or bypass, @@ -6186,13 +6390,13 @@ vm_page_validate_cs( /* validate the mapped page */ vm_page_validate_cs_mapped(page, (const void *) kaddr); -#if CHECK_CS_VALIDATION_BITMAP +#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(object == VM_PAGE_OBJECT(page)); vm_object_lock_assert_exclusive(object); if (!busy_page) { @@ -6213,14 +6417,14 @@ 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; @@ -6228,9 +6432,9 @@ vm_page_validate_cs_mapped_chunk( *tainted_p = 0; assert(page->busy); - vm_object_lock_assert_exclusive(page->object); + object = VM_PAGE_OBJECT(page); + vm_object_lock_assert_exclusive(object); - object = page->object; assert(object->code_signed); offset = page->offset; @@ -6252,25 +6456,22 @@ 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; }