#include <kern/macro_help.h>
#include <kern/zalloc.h>
#include <kern/misc_protos.h>
+#include <kern/policy_internal.h>
#include <vm/vm_compressor.h>
#include <vm/vm_compressor_pager.h>
#include <vm/vm_shared_region.h>
#include <sys/codesign.h>
+#include <sys/reason.h>
+#include <sys/signalvar.h>
#include <libsa/sys/timers.h> /* for struct timespec */
#define TRACEFAULTPAGE 0 /* (TEST/DEBUG) */
-int vm_object_pagein_throttle = 16;
+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
extern void throttle_lowpri_io(int);
+extern struct vnode *vnode_pager_lookup_vnode(memory_object_t);
+
uint64_t vm_hard_throttle_threshold;
-#define NEED_TO_HARD_THROTTLE_THIS_TASK() ((current_task() != kernel_task && \
- get_task_resident_size(current_task()) > (((AVAILABLE_NON_COMPRESSED_MEMORY) * PAGE_SIZE) / 5)) && \
- (vm_low_on_space() || (vm_page_free_count < vm_page_throttle_limit && \
- proc_get_effective_thread_policy(current_thread(), TASK_POLICY_IO) >= THROTTLE_LEVEL_THROTTLED )))
+#define NEED_TO_HARD_THROTTLE_THIS_TASK() (vm_wants_task_throttled(current_task()) || \
+ (vm_page_free_count < vm_page_throttle_limit && \
+ proc_get_effective_thread_policy(current_thread(), TASK_POLICY_IO) > THROTTLE_LEVEL_THROTTLED))
+
+#define HARD_THROTTLE_DELAY 5000 /* 5000 us == 5 ms */
+#define SOFT_THROTTLE_DELAY 200 /* 200 us == .2 ms */
+#define VM_PAGE_CREATION_THROTTLE_PERIOD_SECS 6
+#define VM_PAGE_CREATION_THROTTLE_RATE_PER_SEC 20000
-#define HARD_THROTTLE_DELAY 20000 /* 20000 us == 20 ms */
-#define SOFT_THROTTLE_DELAY 2000 /* 2000 us == 2 ms */
boolean_t current_thread_aborted(void);
/* Forward declarations of internal routines. */
-extern kern_return_t vm_fault_wire_fast(
+static kern_return_t vm_fault_wire_fast(
vm_map_t map,
vm_map_offset_t va,
+ vm_prot_t prot,
vm_map_entry_t entry,
pmap_t pmap,
- vm_map_offset_t pmap_addr);
-
-extern void vm_fault_continue(void);
-
-extern void vm_fault_copy_cleanup(
+ vm_map_offset_t pmap_addr,
+ ppnum_t *physpage_p);
+
+static kern_return_t vm_fault_internal(
+ vm_map_t map,
+ vm_map_offset_t vaddr,
+ vm_prot_t caller_prot,
+ boolean_t change_wiring,
+ int interruptible,
+ pmap_t pmap,
+ vm_map_offset_t pmap_addr,
+ ppnum_t *physpage_p);
+
+static void vm_fault_copy_cleanup(
vm_page_t page,
vm_page_t top_page);
-extern void vm_fault_copy_dst_cleanup(
+static void vm_fault_copy_dst_cleanup(
vm_page_t page);
#if VM_FAULT_CLASSIFY
unsigned long vm_cs_validated_dirtied = 0;
unsigned long vm_cs_bitmap_validated = 0;
+void vm_pre_fault(vm_map_offset_t);
+
+extern char *kdp_compressor_decompressed_page;
+extern addr64_t kdp_compressor_decompressed_page_paddr;
+extern ppnum_t kdp_compressor_decompressed_page_ppnum;
+
/*
* Routine: vm_fault_init
* Purpose:
PE_get_default("kern.vm_compressor", &vm_compressor_mode, sizeof(vm_compressor_mode));
}
PE_parse_boot_argn("vm_compressor_threads", &vm_compressor_thread_count, sizeof (vm_compressor_thread_count));
+
+ if (PE_parse_boot_argn("vm_compressor_immediate", &vm_compressor_temp, sizeof (vm_compressor_temp)))
+ vm_compressor_immediate_preferred_override = TRUE;
+ else {
+ if (PE_get_default("kern.vm_compressor_immediate", &vm_compressor_temp, sizeof(vm_compressor_temp)))
+ vm_compressor_immediate_preferred_override = TRUE;
+ }
+ if (vm_compressor_immediate_preferred_override == TRUE) {
+ if (vm_compressor_temp)
+ vm_compressor_immediate_preferred = TRUE;
+ else
+ vm_compressor_immediate_preferred = FALSE;
+ }
printf("\"vm_compressor_mode\" is %d\n", vm_compressor_mode);
}
*/
void
vm_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);
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;
/*
* 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) {
}
+#if (DEVELOPMENT || DEBUG)
+uint32_t vm_page_creation_throttled_hard = 0;
+uint32_t vm_page_creation_throttled_soft = 0;
+uint64_t vm_page_creation_throttle_avoided = 0;
+#endif /* DEVELOPMENT || DEBUG */
+
static int
-vm_page_throttled(void)
+vm_page_throttled(boolean_t page_kept)
{
clock_sec_t elapsed_sec;
clock_sec_t tv_sec;
if (thread->options & TH_OPT_VMPRIV)
return (0);
- thread->t_page_creation_count++;
-
- if (NEED_TO_HARD_THROTTLE_THIS_TASK())
+ if (thread->t_page_creation_throttled) {
+ thread->t_page_creation_throttled = 0;
+
+ if (page_kept == FALSE)
+ goto no_throttle;
+ }
+ if (NEED_TO_HARD_THROTTLE_THIS_TASK()) {
+#if (DEVELOPMENT || DEBUG)
+ thread->t_page_creation_throttled_hard++;
+ OSAddAtomic(1, &vm_page_creation_throttled_hard);
+#endif /* DEVELOPMENT || DEBUG */
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())) &&
- thread->t_page_creation_count > vm_page_creation_throttle) {
+ 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);
+#endif
+ goto no_throttle;
+ }
clock_get_system_microtime(&tv_sec, &tv_usec);
elapsed_sec = tv_sec - thread->t_page_creation_time;
- if (elapsed_sec <= 6 || (thread->t_page_creation_count / elapsed_sec) >= (vm_page_creation_throttle / 6)) {
+ if (elapsed_sec <= VM_PAGE_CREATION_THROTTLE_PERIOD_SECS ||
+ (thread->t_page_creation_count / elapsed_sec) >= VM_PAGE_CREATION_THROTTLE_RATE_PER_SEC) {
- if (elapsed_sec >= 60) {
+ if (elapsed_sec >= (3 * VM_PAGE_CREATION_THROTTLE_PERIOD_SECS)) {
/*
* we'll reset our stats to give a well behaved app
* that was unlucky enough to accumulate a bunch of pages
* will remain in the throttled state
*/
thread->t_page_creation_time = tv_sec;
- thread->t_page_creation_count = (vm_page_creation_throttle / 6) * 5;
+ thread->t_page_creation_count = VM_PAGE_CREATION_THROTTLE_RATE_PER_SEC * (VM_PAGE_CREATION_THROTTLE_PERIOD_SECS - 1);
}
++vm_page_throttle_count;
- if ((COMPRESSED_PAGER_IS_ACTIVE || DEFAULT_FREEZER_COMPRESSED_PAGER_IS_ACTIVE) && HARD_THROTTLE_LIMIT_REACHED())
+ thread->t_page_creation_throttled = 1;
+
+ 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);
+#endif /* DEVELOPMENT || DEBUG */
return (HARD_THROTTLE_DELAY);
- else
+ } else {
+#if (DEVELOPMENT || DEBUG)
+ thread->t_page_creation_throttled_soft++;
+ OSAddAtomic(1, &vm_page_creation_throttled_soft);
+#endif /* DEVELOPMENT || DEBUG */
return (SOFT_THROTTLE_DELAY);
+ }
}
thread->t_page_creation_time = tv_sec;
thread->t_page_creation_count = 0;
}
+no_throttle:
+ thread->t_page_creation_count++;
+
return (0);
}
* object == m->object
*/
static vm_fault_return_t
-vm_fault_check(vm_object_t object, vm_page_t m, vm_page_t first_m, boolean_t interruptible_state)
+vm_fault_check(vm_object_t object, vm_page_t m, vm_page_t first_m, boolean_t interruptible_state, boolean_t page_throttle)
{
int throttle_delay;
return (VM_FAULT_RETRY);
}
}
- if ((throttle_delay = vm_page_throttled())) {
- /*
- * we're throttling zero-fills...
- * treat this as if we couldn't grab a page
- */
- if (m != VM_PAGE_NULL)
- VM_PAGE_FREE(m);
- vm_fault_cleanup(object, first_m);
+ if (page_throttle == TRUE) {
+ if ((throttle_delay = vm_page_throttled(FALSE))) {
+ /*
+ * we're throttling zero-fills...
+ * treat this as if we couldn't grab a page
+ */
+ if (m != VM_PAGE_NULL)
+ VM_PAGE_FREE(m);
+ vm_fault_cleanup(object, first_m);
- VM_DEBUG_EVENT(vmf_check_zfdelay, VMF_CHECK_ZFDELAY, DBG_FUNC_NONE, throttle_delay, 0, 0, 0);
+ VM_DEBUG_EVENT(vmf_check_zfdelay, VMF_CHECK_ZFDELAY, DBG_FUNC_NONE, throttle_delay, 0, 0, 0);
- delay(throttle_delay);
+ delay(throttle_delay);
- if (current_thread_aborted()) {
+ if (current_thread_aborted()) {
+ thread_interrupt_level(interruptible_state);
+ return VM_FAULT_INTERRUPTED;
+ }
thread_interrupt_level(interruptible_state);
- return VM_FAULT_INTERRUPTED;
- }
- thread_interrupt_level(interruptible_state);
- return (VM_FAULT_MEMORY_SHORTAGE);
+ return (VM_FAULT_MEMORY_SHORTAGE);
+ }
}
return (VM_FAULT_SUCCESS);
}
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...
m->cs_validated = FALSE;
m->cs_tainted = FALSE;
+ m->cs_nx = FALSE;
if (no_zero_fill == TRUE) {
my_fault = DBG_NZF_PAGE_FAULT;
+
+ if (m->absent && m->busy)
+ return (my_fault);
} else {
vm_page_zero_fill(m);
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);
-
- queue_enter(&vm_page_queue_throttled, m, vm_page_t, pageq);
- m->throttled = TRUE;
+ vm_page_queues_remove(m, TRUE);
+ vm_page_check_pageable_safe(m);
+ 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();
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
+ * 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() and that clustered pagein scans are only done on an object
- * for which a pager has been created.
+ * 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
#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 || DEFAULT_FREEZER_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
#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
continue;
}
if (m->laundry) {
- m->pageout = FALSE;
+ m->free_when_done = FALSE;
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 !
*/
* fault cleanup in the case of an error condition
* including resetting the thread_interrupt_level
*/
- error = vm_fault_check(object, m, first_m, interruptible_state);
+ error = vm_fault_check(object, m, first_m, interruptible_state, (type_of_fault == NULL) ? TRUE : FALSE);
if (error != VM_FAULT_SUCCESS)
return (error);
m->absent = FALSE;
m->busy = TRUE;
}
+ if (fault_info->mark_zf_absent && no_zero_fill == TRUE)
+ m->absent = TRUE;
/*
* zero-fill the page and put it on
* the correct paging queue
*/
my_fault = vm_fault_zero_page(m, no_zero_fill);
- if (fault_info->mark_zf_absent && no_zero_fill == TRUE)
- m->absent = TRUE;
-
break;
} else {
if (must_be_resident)
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,
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
* 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) {
/*
*/
m->busy = TRUE;
vm_page_decrypt(m, 0);
- assert(object == m->object);
assert(m->busy);
PAGE_WAKEUP_DONE(m);
}
ASSERT_PAGE_DECRYPTED(m);
- if (m->object->code_signed) {
+ if (object->code_signed) {
/*
* CODE SIGNING:
* We just paged in a page from a signed
goto dont_look_for_page;
}
-
-#if !MACH_PAGEMAP
data_supply = FALSE;
-#endif /* !MACH_PAGEMAP */
look_for_page = (object->pager_created && (MUST_ASK_PAGER(object, offset, external_state) == TRUE) && !data_supply);
/*
* 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
}
if (fault_info && fault_info->batch_pmap_op == TRUE) {
- vm_page_insert_internal(m, object, offset, FALSE, TRUE, TRUE);
+ vm_page_insert_internal(m, object, offset, VM_KERN_MEMORY_NONE, FALSE, TRUE, TRUE, FALSE, NULL);
} else {
vm_page_insert(m, object, offset);
}
return (VM_FAULT_RETRY);
}
}
- if ((COMPRESSED_PAGER_IS_ACTIVE || DEFAULT_FREEZER_COMPRESSED_PAGER_IS_ACTIVE) && object->internal) {
+ 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
m->absent = TRUE;
if (fault_info && fault_info->batch_pmap_op == TRUE) {
- vm_page_insert_internal(m, object, offset, FALSE, TRUE, TRUE);
+ vm_page_insert_internal(m, object, offset, VM_KERN_MEMORY_NONE, FALSE, TRUE, TRUE, FALSE, NULL);
} else {
vm_page_insert(m, object, offset);
}
m->absent = TRUE;
pager = object->pager;
+ assert(object->paging_in_progress > 0);
vm_object_unlock(object);
- rc = vm_compressor_pager_get(pager, offset + object->paging_offset, m->phys_page, &my_fault_type, 0);
+ rc = vm_compressor_pager_get(
+ pager,
+ offset + object->paging_offset,
+ VM_PAGE_GET_PHYS_PAGE(m),
+ &my_fault_type,
+ 0,
+ &compressed_count_delta);
+ if (type_of_fault == NULL) {
+ int throttle_delay;
+
+ /*
+ * we weren't called from vm_fault, so we
+ * need to apply page creation throttling
+ * do it before we re-acquire any locks
+ */
+ if (my_fault_type == DBG_COMPRESSOR_FAULT) {
+ if ((throttle_delay = vm_page_throttled(TRUE))) {
+ VM_DEBUG_EVENT(vmf_compressordelay, VMF_COMPRESSORDELAY, DBG_FUNC_NONE, throttle_delay, 0, 1, 0);
+ delay(throttle_delay);
+ }
+ }
+ }
vm_object_lock(object);
+ assert(object->paging_in_progress > 0);
+
+ vm_compressor_pager_count(
+ pager,
+ compressed_count_delta,
+ FALSE, /* shared_lock */
+ object);
switch (rc) {
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) {
/*
* after the decompression.
*/
pmap_sync_page_attributes_phys(
- m->phys_page);
- } else
+ VM_PAGE_GET_PHYS_PAGE(m));
+ } else {
m->written_by_kernel = TRUE;
+ }
+
+ /*
+ * If the object is purgeable, its
+ * owner's purgeable ledgers have been
+ * updated in vm_page_insert() but the
+ * page was also accounted for in a
+ * "compressed purgeable" ledger, so
+ * update that now.
+ */
+ if ((object->purgable !=
+ VM_PURGABLE_DENY) &&
+ (object->vo_purgeable_owner !=
+ NULL)) {
+ /*
+ * One less compressed
+ * purgeable page.
+ */
+ vm_purgeable_compressed_update(
+ object,
+ -1);
+ }
+
break;
case KERN_MEMORY_FAILURE:
m->unusual = TRUE;
assert(m->absent);
break;
default:
- panic("?");
+ panic("vm_fault_page(): unexpected "
+ "error %d from "
+ "vm_compressor_pager_get()\n",
+ rc);
}
PAGE_WAKEUP_DONE(m);
* 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);
} else
data_already_requested = TRUE;
+ DTRACE_VM2(maj_fault, int, 1, (uint64_t *), NULL);
#if TRACEFAULTPAGE
dbgTrace(0xBEEF0013, (unsigned int) object, (unsigned int) rc); /* (TEST/DEBUG) */
#endif
vm_object_lock(object);
}
m = first_m;
- assert(m->object == object);
+ assert(VM_PAGE_OBJECT(m) == object);
first_m = VM_PAGE_NULL;
/*
* fault cleanup in the case of an error condition
* including resetting the thread_interrupt_level
*/
- error = vm_fault_check(object, m, first_m, interruptible_state);
+ error = vm_fault_check(object, m, first_m, interruptible_state, (type_of_fault == NULL) ? TRUE : FALSE);
if (error != VM_FAULT_SUCCESS)
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);
}
vm_page_insert(m, object, offset);
}
- my_fault = vm_fault_zero_page(m, no_zero_fill);
-
if (fault_info->mark_zf_absent && no_zero_fill == TRUE)
m->absent = TRUE;
+
+ my_fault = vm_fault_zero_page(m, no_zero_fill);
+
break;
} else {
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 */
/*
/*
* 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);
* 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);
+ VM_PAGE_CONSUME_CLUSTERED(m);
+ }
assert(!m->cleaning);
/*
*/
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);
* 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);
+ VM_PAGE_CONSUME_CLUSTERED(m);
+ }
/*
* If there's a pager, then immediately
* page out this page, using the "initialize"
* option. Else, we use the copy.
*/
- if ((!copy_object->pager_created)
-#if MACH_PAGEMAP
- || vm_external_state_get(copy_object->existence_map, copy_offset) == VM_EXTERNAL_STATE_ABSENT
-#endif
+ if ((!copy_object->pager_ready)
|| VM_COMPRESSOR_PAGER_STATE_GET(copy_object, copy_offset) == VM_EXTERNAL_STATE_ABSENT
- ) {
+ ) {
vm_page_lockspin_queues();
assert(!m->cleaning);
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);
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) {
- if (!m->object->internal || (DEFAULT_PAGER_IS_ACTIVE || DEFAULT_FREEZER_IS_ACTIVE))
- VM_STAT_INCR(pageins);
- DTRACE_VM2(pgin, int, 1, (uint64_t *), NULL);
- DTRACE_VM2(maj_fault, int, 1, (uint64_t *), NULL);
- current_task()->pageins++;
+ VM_PAGE_COUNT_AS_PAGEIN(m);
- if (m->object->internal) {
- DTRACE_VM2(anonpgin, int, 1, (uint64_t *), NULL);
+ if (object->internal)
my_fault = DBG_PAGEIND_FAULT;
- } else {
- DTRACE_VM2(fspgin, int, 1, (uint64_t *), NULL);
+ else
my_fault = DBG_PAGEINV_FAULT;
- }
/*
* evaluate access pattern and update state
* 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*/))
* 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;
pmap_t pmap,
vm_map_offset_t vaddr,
vm_prot_t prot,
- vm_prot_t fault_type,
+ vm_prot_t caller_prot,
boolean_t wired,
boolean_t change_wiring,
boolean_t no_cache,
boolean_t cs_bypass,
+ __unused int user_tag,
+ int pmap_options,
boolean_t *need_retry,
int *type_of_fault)
{
boolean_t must_disconnect = 0;
boolean_t map_is_switched, map_is_switch_protected;
int cs_enforcement_enabled;
+ vm_prot_t fault_type;
+ vm_object_t object;
- vm_object_lock_assert_held(m->object);
-#if DEBUG
- lck_mtx_assert(&vm_page_queue_lock, LCK_MTX_ASSERT_NOTOWNED);
-#endif /* DEBUG */
+ fault_type = change_wiring ? VM_PROT_NONE : caller_prot;
+ object = VM_PAGE_OBJECT(m);
+
+ vm_object_lock_assert_held(object);
+ LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_NOTOWNED);
- if (m->phys_page == vm_page_guard_addr) {
+ 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
* access later...
*/
prot &= ~VM_PROT_WRITE;
- }
+ }
if (m->pmapped == FALSE) {
- if ((*type_of_fault == DBG_CACHE_HIT_FAULT) && m->clustered) {
- /*
- * found it in the cache, but this
- * is the first fault-in of the page (m->pmapped == FALSE)
- * so it must have come in as part of
- * a cluster... account 1 pagein against it
- */
- VM_STAT_INCR(pageins);
- DTRACE_VM2(pgin, int, 1, (uint64_t *), NULL);
-
- if (m->object->internal) {
- DTRACE_VM2(anonpgin, int, 1, (uint64_t *), NULL);
- *type_of_fault = DBG_PAGEIND_FAULT;
- } else {
- DTRACE_VM2(fspgin, int, 1, (uint64_t *), NULL);
- *type_of_fault = DBG_PAGEINV_FAULT;
+ if (m->clustered) {
+ if (*type_of_fault == DBG_CACHE_HIT_FAULT) {
+ /*
+ * found it in the cache, but this
+ * is the first fault-in of the page (m->pmapped == FALSE)
+ * so it must have come in as part of
+ * a cluster... account 1 pagein against it
+ */
+ if (object->internal)
+ *type_of_fault = DBG_PAGEIND_FAULT;
+ else
+ *type_of_fault = DBG_PAGEINV_FAULT;
+
+ VM_PAGE_COUNT_AS_PAGEIN(m);
}
-
- current_task()->pageins++;
+ VM_PAGE_CONSUME_CLUSTERED(m);
}
- VM_PAGE_CONSUME_CLUSTERED(m);
-
}
if (*type_of_fault != DBG_COW_FAULT) {
}
/* 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++;
}
#define page_immutable(m,prot) ((m)->cs_validated /*&& ((prot) & VM_PROT_EXECUTE)*/)
+#define page_nx(m) ((m)->cs_nx)
map_is_switched = ((pmap != vm_map_pmap(current_task()->map)) &&
(pmap == vm_map_pmap(current_thread()->map)));
return KERN_CODESIGN_ERROR;
}
+ if (cs_enforcement_enabled && page_nx(m) && (prot & VM_PROT_EXECUTE)) {
+ if (cs_debug)
+ printf("page marked to be NX, not letting it be mapped EXEC\n");
+ 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".
+ */
+ prot &= ~VM_PROT_EXECUTE;
+ }
+
/* 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()).
* can be changed without the kernel noticing, therefore unsigned
* code can be created
*/
- if (m->cs_tainted ||
- ((cs_enforcement_enabled && !cs_bypass ) &&
- (/* The page is unsigned and wants to be executable */
- (!m->cs_validated && (prot & VM_PROT_EXECUTE)) ||
- /* The page should be immutable, but is in danger of being modified
+ if (!cs_bypass &&
+ (m->cs_tainted ||
+ (cs_enforcement_enabled &&
+ (/* The page is unsigned and wants to be executable */
+ (!m->cs_validated && (prot & VM_PROT_EXECUTE)) ||
+ /* The page should be immutable, but is in danger of being modified
* This is the case where we want policy from the code directory -
* is the page immutable or not? For now we have to assume that
* code pages will be immutable, data pages not.
*/
(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
* 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;
+ 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));
} 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",
+ 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 tainted page: abort the page fault */
+ /* reject the invalid page: abort the page fault */
int pid;
const char *procname;
task_t task;
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++;
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;
pathname_len = 0;
filename_len = 0;
truncated_path = FALSE;
- if (file_object->pager == NULL) {
- /* no pager -> no file -> no pathname */
- pathname = (char *) "<nil>";
- } else {
+ /* no pager -> no file -> no pathname, use "<nil>" in that case */
+ if (file_object->pager != NULL) {
pathname = (char *)kalloc(__PATH_MAX * 2);
if (pathname) {
+ pathname[0] = '\0';
pathname_len = __PATH_MAX;
filename = pathname + pathname_len;
filename_len = __PATH_MAX;
filename,
filename_len,
&truncated_path);
+ if (pathname) {
+ /* safety first... */
+ pathname[__PATH_MAX-1] = '\0';
+ filename[__PATH_MAX-1] = '\0';
+ }
vnode_pager_get_object_mtime(file_object->pager,
&mtime,
&cs_mtime);
"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 ? pathname : "<nil>"),
(truncated_path ? "/.../" : ""),
(truncated_path ? filename : ""),
cs_mtime.tv_sec, cs_mtime.tv_nsec,
? "=="
: "!="),
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 : "<nil>"),
+ (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) {
filename = NULL;
}
} else {
- /* proceed with the tainted page */
+ /* proceed with the invalid page */
kr = KERN_SUCCESS;
- /* Page might have been tainted before or not; now it
- * definitively is. If the page wasn't tainted, we must
- * disconnect it from all pmaps later. */
- must_disconnect = !m->cs_tainted;
- m->cs_tainted = TRUE;
+ if (!m->cs_validated &&
+ !object->code_signed) {
+ /*
+ * 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
+ * does not mean that it should cause other
+ * accesses to fail.
+ * This happens when a debugger sets a
+ * breakpoint and we then execute code in
+ * that page. Marking the page as "tainted"
+ * would cause any inspection tool ("leaks",
+ * "vmmap", "CrashReporter", ...) to get killed
+ * due to code-signing violation on that page,
+ * even though they're just reading it and not
+ * executing from it.
+ */
+ } else {
+ /*
+ * Page might have been tainted before or not;
+ * now it definitively is. If the page wasn't
+ * tainted, we must disconnect it from all
+ * pmaps later, to force existing mappings
+ * through that code path for re-consideration
+ * of the validity of that page.
+ */
+ must_disconnect = !m->cs_tainted;
+ m->cs_tainted = TRUE;
+ }
cs_enter_tainted_accepted++;
}
if (kr != KERN_SUCCESS) {
if (cs_debug) {
printf("CODESIGNING: vm_fault_enter(0x%llx): "
- "page %p obj %p off 0x%llx *** INVALID PAGE ***\n",
- (long long)vaddr, m, m->object, m->offset);
+ "*** INVALID PAGE ***\n",
+ (long long)vaddr);
}
#if !SECURE_KERNEL
if (cs_enforcement_panic) {
* 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_page_wire(m, VM_PROT_MEMORY_TAG(caller_prot), TRUE);
}
} else {
vm_page_unwire(m, TRUE);
/* 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 &&
+ if (vm_page_local_q &&
(*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
* 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();
VPL_LOCK(&lq->vpl_lock);
- queue_enter(&lq->vpl_queue, m,
- vm_page_t, pageq);
- m->local = TRUE;
+ vm_page_check_pageable_safe(m);
+ 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++;
* 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
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);
}
}
}
}
}
-
- if ((prot & VM_PROT_EXECUTE) &&
- ! m->xpmapped) {
-
- __VM_PAGE_LOCKSPIN_QUEUES_IF_NEEDED();
-
- /*
- * xpmapped is protected by the page queues lock
- * so it matters not that we might only hold the
- * object lock in the shared state
- */
-
- if (! m->xpmapped) {
-
- m->xpmapped = TRUE;
- __VM_PAGE_UNLOCK_QUEUES_IF_NEEDED();
-
- if ((COMPRESSED_PAGER_IS_ACTIVE || DEFAULT_FREEZER_COMPRESSED_PAGER_IS_ACTIVE) &&
- m->object->internal &&
- m->object->pager != NULL) {
- /*
- * This page could have been
- * uncompressed by the
- * compressor pager and its
- * contents might be only in
- * the data cache.
- * Since it's being mapped for
- * "execute" for the fist time,
- * make sure the icache is in
- * sync.
- */
- pmap_sync_page_data_phys(m->phys_page);
- }
-
- }
- }
/* we're done with the page queues lock, if we ever took it */
__VM_PAGE_UNLOCK_QUEUES_IF_NEEDED();
if (kr == KERN_SUCCESS) {
/*
* NOTE: we may only hold the vm_object lock SHARED
- * at this point, but the update of pmapped is ok
- * since this is the ONLY bit updated behind the SHARED
- * lock... however, we need to figure out how to do an atomic
- * update on a bit field to make this less fragile... right
- * now I don't know how to coerce 'C' to give me the offset info
- * that's needed for an AtomicCompareAndSwap
+ * at this point, so we need the phys_page lock to
+ * properly serialize updating the pmapped and
+ * xpmapped bits
*/
- m->pmapped = TRUE;
- if(vm_page_is_slideable(m)) {
+ if ((prot & VM_PROT_EXECUTE) && !m->xpmapped) {
+ ppnum_t phys_page = VM_PAGE_GET_PHYS_PAGE(m);
+
+ pmap_lock_phys_page(phys_page);
+ /*
+ * go ahead and take the opportunity
+ * to set 'pmapped' here so that we don't
+ * need to grab this lock a 2nd time
+ * just below
+ */
+ m->pmapped = TRUE;
+
+ if (!m->xpmapped) {
+
+ m->xpmapped = TRUE;
+
+ pmap_unlock_phys_page(phys_page);
+
+ if (!object->internal)
+ OSAddAtomic(1, &vm_page_xpmapped_external_count);
+
+ if (object->internal &&
+ object->pager != NULL) {
+ /*
+ * This page could have been
+ * uncompressed by the
+ * compressor pager and its
+ * contents might be only in
+ * the data cache.
+ * Since it's being mapped for
+ * "execute" for the fist time,
+ * make sure the icache is in
+ * sync.
+ */
+ assert(VM_CONFIG_COMPRESSOR_IS_PRESENT);
+ pmap_sync_page_data_phys(phys_page);
+ }
+ } else
+ pmap_unlock_phys_page(phys_page);
+ } else {
+ if (m->pmapped == FALSE) {
+ ppnum_t phys_page = VM_PAGE_GET_PHYS_PAGE(m);
+
+ pmap_lock_phys_page(phys_page);
+ m->pmapped = TRUE;
+ 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);
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) {
* 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
}
}
}
+ assert(VM_PAGE_OBJECT(m) == object);
/* Prevent a deadlock by not
* holding the object lock if we need to wait for a page in
* pmap_enter() - <rdar://problem/7138958> */
PMAP_ENTER_OPTIONS(pmap, vaddr, m, prot, fault_type, 0,
- wired, PMAP_OPTIONS_NOWAIT, pe_result);
+ wired,
+ pmap_options | PMAP_OPTIONS_NOWAIT,
+ pe_result);
if(pe_result == KERN_RESOURCE_SHORTAGE) {
* 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(pmap, vaddr, m, prot, fault_type, 0, wired);
+ 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. */
return kr;
}
+void
+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 */);
+ }
+}
+
/*
* Routine: vm_fault
*/
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;
int interruptible,
pmap_t caller_pmap,
vm_map_offset_t caller_pmap_addr)
+{
+ return vm_fault_internal(map, vaddr, fault_type, change_wiring,
+ interruptible, caller_pmap, caller_pmap_addr,
+ NULL);
+}
+
+
+kern_return_t
+vm_fault_internal(
+ vm_map_t map,
+ vm_map_offset_t vaddr,
+ vm_prot_t caller_prot,
+ boolean_t change_wiring,
+ int interruptible,
+ pmap_t caller_pmap,
+ vm_map_offset_t caller_pmap_addr,
+ ppnum_t *physpage_p)
{
vm_map_version_t version; /* Map version for verificiation */
boolean_t wired; /* Should mapping be wired down? */
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;
boolean_t interruptible_state;
vm_map_t real_map = map;
vm_map_t original_map = map;
+ vm_prot_t fault_type;
vm_prot_t original_fault_type;
struct vm_object_fault_info fault_info;
boolean_t need_collapse = FALSE;
int cur_object_lock_type;
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_UNSLIDE_OR_PERM(vaddr);
+ trace_real_vaddr = VM_KERNEL_UNSLIDE_OR_PERM(trace_real_vaddr);
+ } else {
+ trace_vaddr = vaddr;
+ }
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,
(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);
interruptible_state = thread_interrupt_level(interruptible);
+ fault_type = (change_wiring ? VM_PROT_NONE : caller_prot);
+
VM_STAT_INCR(faults);
current_task()->faults++;
original_fault_type = fault_type;
&fault_info,
&real_map);
+
if (kr != KERN_SUCCESS) {
vm_map_unlock_read(map);
goto done;
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 */
}
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;
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
* 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();
continue;
}
}
- m->pageout = FALSE;
-
vm_pageout_steal_laundry(m, FALSE);
}
- if (m->phys_page == vm_page_guard_addr) {
+ if (VM_PAGE_GET_PHYS_PAGE(m) == vm_page_guard_addr) {
/*
* Guard page: let the slow path deal with it
*/
*/
break;
}
- if (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);
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:
/*
* We might need to validate this page
if ((fault_type & VM_PROT_WRITE) == 0) {
+ prot &= ~VM_PROT_WRITE;
+
if (object != cur_object) {
/*
* We still need to hold the top object
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
caller_pmap,
caller_pmap_addr,
prot,
- fault_type,
+ caller_prot,
wired,
change_wiring,
fault_info.no_cache,
fault_info.cs_bypass,
+ fault_info.user_tag,
+ fault_info.pmap_options,
need_retry_ptr,
&type_of_fault);
} else {
pmap,
vaddr,
prot,
- fault_type,
+ caller_prot,
wired,
change_wiring,
fault_info.no_cache,
fault_info.cs_bypass,
+ fault_info.user_tag,
+ fault_info.pmap_options,
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 = VM_PAGE_GET_PHYS_PAGE(m);
+ if (prot & VM_PROT_WRITE) {
+ vm_object_lock_assert_exclusive(m_object);
+ m->dirty = TRUE;
+ }
+ }
if (top_object != VM_OBJECT_NULL) {
/*
* re-drive the fault which should result in vm_fault_enter
* being able to successfully enter the mapping this time around
*/
- (void)pmap_enter_options(pmap, vaddr, 0, 0, 0, 0, 0, PMAP_OPTIONS_NOENTER, NULL);
+ (void)pmap_enter_options(
+ pmap, vaddr, 0, 0, 0, 0, 0,
+ PMAP_OPTIONS_NOENTER, NULL);
need_retry = FALSE;
goto RetryFault;
*/
assert(object_lock_type == OBJECT_LOCK_EXCLUSIVE);
- if ((throttle_delay = vm_page_throttled())) {
- /*
- * drop all of our locks...
- * wait until the free queue is
- * pumped back up and then
- * redrive the fault
- */
- if (object != cur_object)
- vm_object_unlock(cur_object);
- vm_object_unlock(object);
- vm_map_unlock_read(map);
- if (real_map != map)
- vm_map_unlock(real_map);
-
- VM_DEBUG_EVENT(vmf_cowdelay, VMF_COWDELAY, DBG_FUNC_NONE, throttle_delay, 0, 0, 0);
-
- delay(throttle_delay);
-
- if (!current_thread_aborted() && vm_page_wait((change_wiring) ?
- THREAD_UNINT :
- THREAD_ABORTSAFE))
- goto RetryFault;
- kr = KERN_ABORTED;
- goto done;
- }
/*
* If objects match, then
* object->copy must not be NULL (else control
* 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;
}
* 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) {
/*
*/
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;
if (!cur_object->internal &&
continue;
}
}
- m = vm_page_grab();
+ m = vm_page_grab_options(grab_options);
+ m_object = NULL;
if (m == VM_PAGE_NULL) {
/*
*/
break;
}
- if (vm_compressor_pager_get(cur_object->pager, cur_offset + cur_object->paging_offset,
- m->phys_page, &my_fault_type, c_flags) != KERN_SUCCESS) {
- vm_page_release(m);
+
+ /*
+ * The object is and remains locked
+ * so no need to take a
+ * "paging_in_progress" reference.
+ */
+ boolean_t shared_lock;
+ if ((object == cur_object &&
+ object_lock_type == OBJECT_LOCK_EXCLUSIVE) ||
+ (object != cur_object &&
+ cur_object_lock_type == OBJECT_LOCK_EXCLUSIVE)) {
+ shared_lock = FALSE;
+ } else {
+ shared_lock = TRUE;
+ }
+
+ kr = vm_compressor_pager_get(
+ cur_object->pager,
+ (cur_offset +
+ cur_object->paging_offset),
+ VM_PAGE_GET_PHYS_PAGE(m),
+ &my_fault_type,
+ c_flags,
+ &compressed_count_delta);
+
+ vm_compressor_pager_count(
+ cur_object->pager,
+ compressed_count_delta,
+ shared_lock,
+ cur_object);
+
+ if (kr != KERN_SUCCESS) {
+ vm_page_release(m, FALSE);
+ m = VM_PAGE_NULL;
break;
}
m->dirty = TRUE;
- if (insert_cur_object)
+ /*
+ * If the object is purgeable, its
+ * owner's purgeable ledgers will be
+ * updated in vm_page_insert() but the
+ * page was also accounted for in a
+ * "compressed purgeable" ledger, so
+ * update that now.
+ */
+ if (object != cur_object &&
+ !insert_cur_object) {
+ /*
+ * We're not going to insert
+ * the decompressed page into
+ * the object it came from.
+ *
+ * We're dealing with a
+ * copy-on-write fault on
+ * "object".
+ * We're going to decompress
+ * the page directly into the
+ * target "object" while
+ * keepin the compressed
+ * page for "cur_object", so
+ * no ledger update in that
+ * case.
+ */
+ } else if ((cur_object->purgable ==
+ VM_PURGABLE_DENY) ||
+ (cur_object->vo_purgeable_owner ==
+ NULL)) {
+ /*
+ * "cur_object" is not purgeable
+ * or is not owned, so no
+ * purgeable ledgers to update.
+ */
+ } else {
+ /*
+ * One less compressed
+ * purgeable page for
+ * cur_object's owner.
+ */
+ vm_purgeable_compressed_update(
+ cur_object,
+ -1);
+ }
+
+ if (insert_cur_object) {
vm_page_insert(m, cur_object, cur_offset);
- else
+ 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;
VM_STAT_INCR(decompressions);
* 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);
kr = KERN_MEMORY_ERROR;
goto done;
}
- if ((throttle_delay = vm_page_throttled())) {
- /*
- * drop all of our locks...
- * wait until the free queue is
- * pumped back up and then
- * redrive the fault
- */
- if (object != cur_object)
- vm_object_unlock(cur_object);
- vm_object_unlock(object);
- vm_map_unlock_read(map);
- if (real_map != map)
- vm_map_unlock(real_map);
-
- VM_DEBUG_EVENT(vmf_zfdelay, VMF_ZFDELAY, DBG_FUNC_NONE, throttle_delay, 0, 0, 0);
-
- delay(throttle_delay);
-
- if (!current_thread_aborted() && vm_page_wait((change_wiring) ?
- THREAD_UNINT :
- THREAD_ABORTSAFE))
- goto RetryFault;
- kr = KERN_ABORTED;
- goto done;
- }
if (vm_backing_store_low) {
/*
* we are protecting the system from
}
}
m = vm_page_alloc(object, offset);
+ m_object = NULL;
if (m == VM_PAGE_NULL) {
/*
*/
break;
}
+ m_object = object;
/*
* Now zero fill page...
if (real_map != map)
vm_map_unlock(real_map);
+ 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
}
}
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)));
}
/*
#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
/*
* since our last lookup.
*/
if (m != VM_PAGE_NULL) {
- old_copy_object = m->object->copy;
- vm_object_unlock(m->object);
+ old_copy_object = m_object->copy;
+ vm_object_unlock(m_object);
} else {
old_copy_object = VM_OBJECT_NULL;
vm_object_unlock(object);
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
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
prot &= retry_prot;
}
if (m != VM_PAGE_NULL) {
- vm_object_lock(m->object);
+ vm_object_lock(m_object);
- if (m->object->copy != old_copy_object) {
+ if (m_object->copy != old_copy_object) {
/*
* The copy object changed while the top-level object
* was unlocked, so take away write permission.
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);
caller_pmap,
caller_pmap_addr,
prot,
- fault_type,
+ caller_prot,
wired,
change_wiring,
fault_info.no_cache,
fault_info.cs_bypass,
+ fault_info.user_tag,
+ fault_info.pmap_options,
NULL,
&type_of_fault);
} else {
pmap,
vaddr,
prot,
- fault_type,
+ caller_prot,
wired,
change_wiring,
fault_info.no_cache,
fault_info.cs_bypass,
+ fault_info.user_tag,
+ fault_info.pmap_options,
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);
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 = VM_PAGE_GET_PHYS_PAGE(m);
+ if (prot & VM_PROT_WRITE) {
+ vm_object_lock_assert_exclusive(m_object);
+ m->dirty = TRUE;
+ }
+ }
} else {
vm_map_entry_t entry;
hdelta = entry->vme_end - laddr;
if (entry->is_sub_map) {
- laddr = (laddr - entry->vme_start)
- + entry->offset;
- vm_map_lock_read(entry->object.sub_map);
+ laddr = ((laddr - entry->vme_start)
+ + VME_OFFSET(entry));
+ vm_map_lock_read(VME_SUBMAP(entry));
if (map != real_map)
vm_map_unlock_read(map);
if (entry->use_pmap) {
vm_map_unlock_read(real_map);
- real_map = entry->object.sub_map;
+ real_map = VME_SUBMAP(entry);
}
- map = entry->object.sub_map;
+ map = VME_SUBMAP(entry);
} else {
break;
}
if (vm_map_lookup_entry(map, laddr, &entry) &&
- (entry->object.vm_object != NULL) &&
- (entry->object.vm_object == object)) {
+ (VME_OBJECT(entry) != NULL) &&
+ (VME_OBJECT(entry) == object)) {
+ int superpage;
+
+ if (!object->pager_created &&
+ 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;
+ }
+
+ if (superpage && physpage_p) {
+ /* for vm_map_wire_and_extract() */
+ *physpage_p = (ppnum_t)
+ ((((vm_map_offset_t)
+ object->vo_shadow_offset)
+ + VME_OFFSET(entry)
+ + (laddr - entry->vme_start))
+ >> PAGE_SHIFT);
+ }
- int superpage = (!object->pager_created && object->phys_contiguous)? VM_MEM_SUPERPAGE : 0;
if (caller_pmap) {
/*
* Set up a block mapped area
*/
- assert((uint32_t)((ldelta + hdelta) >> 12) == ((ldelta + hdelta) >> 12));
+ 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) (entry->object.vm_object->vo_shadow_offset)) +
- entry->offset + (laddr - entry->vme_start) - ldelta) >> 12),
- (uint32_t)((ldelta + hdelta) >> 12), prot,
+ (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 {
/*
* Set up a block mapped area
*/
- assert((uint32_t)((ldelta + hdelta) >> 12) == ((ldelta + hdelta) >> 12));
+ 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)(entry->object.vm_object->vo_shadow_offset)) +
- entry->offset + (laddr - entry->vme_start) - ldelta) >> 12),
- (uint32_t)((ldelta + hdelta) >> 12), prot,
+ (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);
}
}
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);
thread_interrupt_level(interruptible_state);
/*
- * Only throttle on faults which cause a pagein.
+ * Only I/O throttle on faults which cause a pagein/swapin.
*/
if ((type_of_fault == DBG_PAGEIND_FAULT) || (type_of_fault == DBG_PAGEINV_FAULT) || (type_of_fault == DBG_COMPRESSOR_SWAPIN_FAULT)) {
throttle_lowpri_io(1);
- }
+ } else {
+ if (kr == KERN_SUCCESS && type_of_fault != DBG_CACHE_HIT_FAULT && type_of_fault != DBG_GUARD_FAULT) {
+ if ((throttle_delay = vm_page_throttled(TRUE))) {
+
+ if (vm_debug_events) {
+ if (type_of_fault == DBG_COMPRESSOR_FAULT)
+ VM_DEBUG_EVENT(vmf_compressordelay, VMF_COMPRESSORDELAY, DBG_FUNC_NONE, throttle_delay, 0, 0, 0);
+ else if (type_of_fault == DBG_COW_FAULT)
+ VM_DEBUG_EVENT(vmf_cowdelay, VMF_COWDELAY, DBG_FUNC_NONE, throttle_delay, 0, 0, 0);
+ else
+ VM_DEBUG_EVENT(vmf_zfdelay, VMF_ZFDELAY, DBG_FUNC_NONE, throttle_delay, 0, 0, 0);
+ }
+ delay(throttle_delay);
+ }
+ }
+ }
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);
vm_fault_wire(
vm_map_t map,
vm_map_entry_t entry,
+ vm_prot_t prot,
pmap_t pmap,
- vm_map_offset_t pmap_addr)
+ 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 ((entry->object.vm_object != NULL) &&
- !entry->is_sub_map &&
- entry->object.vm_object->phys_contiguous) {
+ if ((VME_OBJECT(entry) != NULL) &&
+ !entry->is_sub_map &&
+ VME_OBJECT(entry)->phys_contiguous) {
return KERN_SUCCESS;
}
*/
for (va = entry->vme_start; va < end_addr; va += PAGE_SIZE) {
- if ((rc = vm_fault_wire_fast(
- map, va, entry, pmap,
- pmap_addr + (va - entry->vme_start)
- )) != KERN_SUCCESS) {
- rc = vm_fault(map, va, VM_PROT_NONE, TRUE,
- (pmap == kernel_pmap) ?
- THREAD_UNINT : THREAD_ABORTSAFE,
- pmap, pmap_addr + (va - entry->vme_start));
+ rc = vm_fault_wire_fast(map, va, prot, entry, pmap,
+ pmap_addr + (va - entry->vme_start),
+ physpage_p);
+ if (rc != KERN_SUCCESS) {
+ rc = vm_fault_internal(map, va, prot, TRUE,
+ ((pmap == kernel_pmap)
+ ? THREAD_UNINT
+ : THREAD_ABORTSAFE),
+ pmap,
+ (pmap_addr +
+ (va - entry->vme_start)),
+ physpage_p);
DTRACE_VM2(softlock, int, 1, (uint64_t *), NULL);
}
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;
- object = (entry->is_sub_map)
- ? VM_OBJECT_NULL : entry->object.vm_object;
+ object = (entry->is_sub_map) ? VM_OBJECT_NULL : VME_OBJECT(entry);
/*
* If it's marked phys_contiguous, then vm_fault_wire() didn't actually
fault_info.interruptible = THREAD_UNINT;
fault_info.behavior = entry->behavior;
- fault_info.user_tag = entry->alias;
- fault_info.lo_offset = entry->offset;
- fault_info.hi_offset = (entry->vme_end - entry->vme_start) + entry->offset;
+ fault_info.user_tag = VME_ALIAS(entry);
+ fault_info.pmap_options = 0;
+ if (entry->iokit_acct ||
+ (!entry->is_sub_map && !entry->use_pmap)) {
+ fault_info.pmap_options |= PMAP_OPTIONS_ALT_ACCT;
+ }
+ fault_info.lo_offset = VME_OFFSET(entry);
+ fault_info.hi_offset = (entry->vme_end - entry->vme_start) + VME_OFFSET(entry);
fault_info.no_cache = entry->no_cache;
fault_info.stealth = TRUE;
fault_info.io_sync = FALSE;
result_page = VM_PAGE_NULL;
result = vm_fault_page(
object,
- entry->offset + (va - entry->vme_start),
+ (VME_OFFSET(entry) +
+ (va - entry->vme_start)),
VM_PROT_NONE, TRUE,
FALSE, /* page not looked up */
&prot, &result_page, &top_page,
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));
VM_PAGE_FREE(result_page);
} else {
- if ((pmap) && (result_page->phys_page != vm_page_guard_addr))
+ if ((pmap) && (VM_PAGE_GET_PHYS_PAGE(result_page) != vm_page_guard_addr))
pmap_change_wiring(pmap,
pmap_addr + (va - entry->vme_start), FALSE);
vm_page_unlock_queues();
}
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;
}
* other than the common case will return KERN_FAILURE, and the caller
* is expected to call vm_fault().
*/
-kern_return_t
+static kern_return_t
vm_fault_wire_fast(
__unused vm_map_t map,
vm_map_offset_t va,
+ vm_prot_t caller_prot,
vm_map_entry_t entry,
- pmap_t pmap,
- vm_map_offset_t pmap_addr)
+ pmap_t pmap,
+ vm_map_offset_t pmap_addr,
+ ppnum_t *physpage_p)
{
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;
/*
* If this entry is not directly to a vm_object, bail out.
*/
- if (entry->is_sub_map)
+ if (entry->is_sub_map) {
+ assert(physpage_p == NULL);
return(KERN_FAILURE);
+ }
/*
* Find the backing store object and offset into it.
*/
- object = entry->object.vm_object;
- offset = (va - entry->vme_start) + entry->offset;
+ object = VME_OBJECT(entry);
+ offset = (va - entry->vme_start) + VME_OFFSET(entry);
prot = entry->protection;
/*
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...
*/
vm_page_lockspin_queues();
- vm_page_wire(m);
+ vm_page_wire(m, VM_PROT_MEMORY_TAG(caller_prot), TRUE);
vm_page_unlock_queues();
/*
FALSE,
FALSE,
FALSE,
+ VME_ALIAS(entry),
+ ((entry->iokit_acct ||
+ (!entry->is_sub_map && !entry->use_pmap))
+ ? PMAP_OPTIONS_ALT_ACCT
+ : 0),
NULL,
&type_of_fault);
+ if (kr != KERN_SUCCESS) {
+ RELEASE_PAGE(m);
+ GIVE_UP;
+ }
done:
/*
* Unlock everything, and return
*/
+ if (physpage_p) {
+ /* for vm_map_wire_and_extract() */
+ if (kr == KERN_SUCCESS) {
+ assert(object == VM_PAGE_OBJECT(m));
+ *physpage_p = VM_PAGE_GET_PHYS_PAGE(m);
+ if (prot & VM_PROT_WRITE) {
+ vm_object_lock_assert_exclusive(object);
+ m->dirty = TRUE;
+ }
+ } else {
+ *physpage_p = 0;
+ }
+ }
+
PAGE_WAKEUP_DONE(m);
UNLOCK_AND_DEALLOCATE;
* Release a page used by vm_fault_copy.
*/
-void
+static void
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);
}
-void
+static void
vm_fault_copy_dst_cleanup(
vm_page_t page)
{
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_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;
fault_info_src.interruptible = interruptible;
fault_info_src.behavior = VM_BEHAVIOR_SEQUENTIAL;
fault_info_src.user_tag = 0;
+ fault_info_src.pmap_options = 0;
fault_info_src.lo_offset = vm_object_trunc_page(src_offset);
fault_info_src.hi_offset = fault_info_src.lo_offset + amount_left;
fault_info_src.no_cache = FALSE;
fault_info_dst.interruptible = interruptible;
fault_info_dst.behavior = VM_BEHAVIOR_SEQUENTIAL;
fault_info_dst.user_tag = 0;
+ fault_info_dst.pmap_options = 0;
fault_info_dst.lo_offset = vm_object_trunc_page(dst_offset);
fault_info_dst.hi_offset = fault_info_dst.lo_offset + amount_left;
fault_info_dst.no_cache = FALSE;
}
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
*/
vm_page_lockspin_queues();
- vm_page_wire(dst_page);
+ 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);
"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);
}
if (!vm_map_verify(dst_map, dst_version)) {
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);
+ if (dst_object->copy != old_copy_object) {
+ vm_object_unlock(dst_object);
vm_map_verify_done(dst_map, dst_version);
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
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);
}
}
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);
}
}
}
#endif /* VM_FAULT_CLASSIFY */
+vm_offset_t
+kdp_lightweight_fault(vm_map_t map, vm_offset_t cur_target_addr)
+{
+ vm_map_entry_t entry;
+ vm_object_t object;
+ vm_offset_t object_offset;
+ vm_page_t m;
+ int compressor_external_state, compressed_count_delta;
+ int compressor_flags = (C_DONT_BLOCK | C_KEEP | C_KDP);
+ 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");
+ }
+
+ assert(map != VM_MAP_NULL);
+
+ assert((cur_target_addr & PAGE_MASK) == 0);
+ if ((cur_target_addr & PAGE_MASK) != 0) {
+ return 0;
+ }
+
+ if (kdp_lck_rw_lock_is_acquired_exclusive(&map->lock)) {
+ return 0;
+ }
+
+ if (!vm_map_lookup_entry(map, cur_target_addr, &entry)) {
+ return 0;
+ }
+
+ if (entry->is_sub_map) {
+ return 0;
+ }
+
+ object = VME_OBJECT(entry);
+ if (object == VM_OBJECT_NULL) {
+ return 0;
+ }
+
+ object_offset = cur_target_addr - entry->vme_start + VME_OFFSET(entry);
+
+ while (TRUE) {
+ if (kdp_lck_rw_lock_is_acquired_exclusive(&object->Lock)) {
+ return 0;
+ }
+
+ if (object->pager_created && (object->paging_in_progress ||
+ object->activity_in_progress)) {
+ return 0;
+ }
+
+ m = kdp_vm_page_lookup(object, object_offset);
+
+ if (m != VM_PAGE_NULL) {
+
+ if ((object->wimg_bits & VM_WIMG_MASK) != VM_WIMG_DEFAULT) {
+ return 0;
+ }
+
+ if (m->laundry || m->busy || m->free_when_done || m->absent || m->error || m->cleaning ||
+ m->overwriting || m->restart || m->unusual) {
+ return 0;
+ }
+
+ assert(!m->private);
+ if (m->private) {
+ return 0;
+ }
+
+ assert(!m->fictitious);
+ if (m->fictitious) {
+ return 0;
+ }
+
+ assert(!m->encrypted);
+ if (m->encrypted) {
+ return 0;
+ }
+
+ assert(!m->encrypted_cleaning);
+ if (m->encrypted_cleaning) {
+ return 0;
+ }
+
+ assert(m->vm_page_q_state != VM_PAGE_USED_BY_COMPRESSOR);
+ if (m->vm_page_q_state == VM_PAGE_USED_BY_COMPRESSOR) {
+ return 0;
+ }
+
+ return ptoa(VM_PAGE_GET_PHYS_PAGE(m));
+ }
+
+ compressor_external_state = VM_EXTERNAL_STATE_UNKNOWN;
+
+ if (object->pager_created && MUST_ASK_PAGER(object, object_offset, compressor_external_state)) {
+ if (compressor_external_state == VM_EXTERNAL_STATE_EXISTS) {
+ kr = vm_compressor_pager_get(object->pager, (object_offset + object->paging_offset),
+ kdp_compressor_decompressed_page_ppnum, &my_fault_type,
+ compressor_flags, &compressed_count_delta);
+ if (kr == KERN_SUCCESS) {
+ return kdp_compressor_decompressed_page_paddr;
+ } else {
+ return 0;
+ }
+ }
+ }
+
+ if (object->shadow == VM_OBJECT_NULL) {
+ return 0;
+ }
+
+ object_offset += object->vo_shadow_offset;
+ object = object->shadow;
+ }
+
+}
void
vm_page_validate_cs_mapped(
{
vm_object_t object;
vm_object_offset_t offset;
- kern_return_t kr;
memory_object_t pager;
- void *blobs;
- boolean_t validated, tainted;
+ struct vnode *vnode;
+ boolean_t validated;
+ unsigned tainted;
assert(page->busy);
- vm_object_lock_assert_exclusive(page->object);
-
- if (!cs_validation) {
- return;
- }
+ object = VM_PAGE_OBJECT(page);
+ vm_object_lock_assert_exclusive(object);
if (page->wpmapped && !page->cs_tainted) {
/*
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++;
}
- if (page->cs_validated) {
+ if (page->cs_validated || page->cs_tainted) {
return;
}
vm_cs_validates++;
- object = page->object;
assert(object->code_signed);
offset = page->offset;
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 */
- validated = cs_validate_page(blobs,
- pager,
- offset + object->paging_offset,
- (const void *)kaddr,
- &tainted);
+ 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;
+ }
+ if (tainted & CS_VALIDATE_NX) {
+ page->cs_nx = TRUE;
+ }
- page->cs_validated = validated;
if (validated) {
- page->cs_tainted = tainted;
+ page->cs_validated = TRUE;
}
}
-extern int panic_on_cs_killed;
void
vm_page_validate_cs(
vm_page_t page)
boolean_t busy_page;
boolean_t need_unmap;
- vm_object_lock_assert_held(page->object);
-
- if (!cs_validation) {
- return;
- }
+ 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
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++;
}
- if (page->cs_validated) {
+ if (page->cs_validated || page->cs_tainted) {
return;
}
- if (panic_on_cs_killed &&
- page->slid) {
+ if (page->slid) {
panic("vm_page_validate_cs(%p): page is slid\n", page);
}
assert(!page->slid);
#if CHECK_CS_VALIDATION_BITMAP
- if ( vnode_pager_cs_check_validation_bitmap( page->object->pager, trunc_page(page->offset + page->object->paging_offset), CS_BITMAP_CHECK ) == KERN_SUCCESS) {
+ 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;
}
#endif
assert(page->busy);
- assert(object == page->object);
+ assert(object == VM_PAGE_OBJECT(page));
vm_object_lock_assert_exclusive(object);
if (!busy_page) {
}
vm_object_paging_end(object);
}
+
+void
+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;
+ memory_object_t pager;
+ struct vnode *vnode;
+ boolean_t validated;
+ unsigned tainted;
+
+ *validated_p = FALSE;
+ *tainted_p = 0;
+
+ assert(page->busy);
+ object = VM_PAGE_OBJECT(page);
+ vm_object_lock_assert_exclusive(object);
+
+ assert(object->code_signed);
+ offset = page->offset;
+
+ if (!object->alive || object->terminating || object->pager == NULL) {
+ /*
+ * The object is terminating and we don't have its pager
+ * so we can't validate the data...
+ */
+ return;
+ }
+ /*
+ * Since we get here to validate a page that was brought in by
+ * the pager, we know that this pager is all setup and ready
+ * by now.
+ */
+ assert(!object->internal);
+ assert(object->pager != NULL);
+ assert(object->pager_ready);
+
+ pager = object->pager;
+ assert(object->paging_in_progress);
+ vnode = vnode_pager_lookup_vnode(pager);
+
+ /* verify the signature for this chunk */
+ offset_in_page = chunk_offset;
+ assert(offset_in_page < PAGE_SIZE);
+
+ tainted = 0;
+ validated = cs_validate_range(vnode,
+ pager,
+ (object->paging_offset +
+ offset +
+ offset_in_page),
+ (const void *)((const char *)kaddr
+ + offset_in_page),
+ chunk_size,
+ &tainted);
+ if (validated) {
+ *validated_p = TRUE;
+ }
+ if (tainted) {
+ *tainted_p = tainted;
+ }
+}
projects / apple / xnu.git / blobdiff