2 * SPDX-License-Identifier: (BSD-3-Clause AND MIT-CMU)
4 * Copyright (c) 1991, 1993
5 * The Regents of the University of California. All rights reserved.
7 * This code is derived from software contributed to Berkeley by
8 * The Mach Operating System project at Carnegie-Mellon University.
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 3. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * from: @(#)vm_map.c 8.3 (Berkeley) 1/12/94
37 * Copyright (c) 1987, 1990 Carnegie-Mellon University.
38 * All rights reserved.
40 * Authors: Avadis Tevanian, Jr., Michael Wayne Young
42 * Permission to use, copy, modify and distribute this software and
43 * its documentation is hereby granted, provided that both the copyright
44 * notice and this permission notice appear in all copies of the
45 * software, derivative works or modified versions, and any portions
46 * thereof, and that both notices appear in supporting documentation.
48 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
49 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
50 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
52 * Carnegie Mellon requests users of this software to return to
54 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
55 * School of Computer Science
56 * Carnegie Mellon University
57 * Pittsburgh PA 15213-3890
59 * any improvements or extensions that they make and grant Carnegie the
60 * rights to redistribute these changes.
64 * Virtual memory mapping module.
67 #include <sys/cdefs.h>
68 __FBSDID("$FreeBSD$");
70 #include <sys/param.h>
71 #include <sys/systm.h>
72 #include <sys/kernel.h>
75 #include <sys/mutex.h>
77 #include <sys/vmmeter.h>
79 #include <sys/vnode.h>
80 #include <sys/racct.h>
81 #include <sys/resourcevar.h>
82 #include <sys/rwlock.h>
84 #include <sys/sysctl.h>
85 #include <sys/sysent.h>
89 #include <vm/vm_param.h>
91 #include <vm/vm_map.h>
92 #include <vm/vm_page.h>
93 #include <vm/vm_object.h>
94 #include <vm/vm_pager.h>
95 #include <vm/vm_kern.h>
96 #include <vm/vm_extern.h>
97 #include <vm/vnode_pager.h>
98 #include <vm/swap_pager.h>
102 * Virtual memory maps provide for the mapping, protection,
103 * and sharing of virtual memory objects. In addition,
104 * this module provides for an efficient virtual copy of
105 * memory from one map to another.
107 * Synchronization is required prior to most operations.
109 * Maps consist of an ordered doubly-linked list of simple
110 * entries; a self-adjusting binary search tree of these
111 * entries is used to speed up lookups.
113 * Since portions of maps are specified by start/end addresses,
114 * which may not align with existing map entries, all
115 * routines merely "clip" entries to these start/end values.
116 * [That is, an entry is split into two, bordering at a
117 * start or end value.] Note that these clippings may not
118 * always be necessary (as the two resulting entries are then
119 * not changed); however, the clipping is done for convenience.
121 * As mentioned above, virtual copy operations are performed
122 * by copying VM object references from one map to
123 * another, and then marking both regions as copy-on-write.
126 static struct mtx map_sleep_mtx;
127 static uma_zone_t mapentzone;
128 static uma_zone_t kmapentzone;
129 static uma_zone_t mapzone;
130 static uma_zone_t vmspace_zone;
131 static int vmspace_zinit(void *mem, int size, int flags);
132 static int vm_map_zinit(void *mem, int ize, int flags);
133 static void _vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min,
135 static int vm_map_alignspace(vm_map_t map, vm_object_t object,
136 vm_ooffset_t offset, vm_offset_t *addr, vm_size_t length,
137 vm_offset_t max_addr, vm_offset_t alignment);
138 static void vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map);
139 static void vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry);
140 static void vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry);
141 static int vm_map_growstack(vm_map_t map, vm_offset_t addr,
142 vm_map_entry_t gap_entry);
143 static void vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot,
144 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags);
146 static void vm_map_zdtor(void *mem, int size, void *arg);
147 static void vmspace_zdtor(void *mem, int size, void *arg);
149 static int vm_map_stack_locked(vm_map_t map, vm_offset_t addrbos,
150 vm_size_t max_ssize, vm_size_t growsize, vm_prot_t prot, vm_prot_t max,
152 static void vm_map_wire_entry_failure(vm_map_t map, vm_map_entry_t entry,
153 vm_offset_t failed_addr);
155 #define ENTRY_CHARGED(e) ((e)->cred != NULL || \
156 ((e)->object.vm_object != NULL && (e)->object.vm_object->cred != NULL && \
157 !((e)->eflags & MAP_ENTRY_NEEDS_COPY)))
160 * PROC_VMSPACE_{UN,}LOCK() can be a noop as long as vmspaces are type
163 #define PROC_VMSPACE_LOCK(p) do { } while (0)
164 #define PROC_VMSPACE_UNLOCK(p) do { } while (0)
167 * VM_MAP_RANGE_CHECK: [ internal use only ]
169 * Asserts that the starting and ending region
170 * addresses fall within the valid range of the map.
172 #define VM_MAP_RANGE_CHECK(map, start, end) \
174 if (start < vm_map_min(map)) \
175 start = vm_map_min(map); \
176 if (end > vm_map_max(map)) \
177 end = vm_map_max(map); \
185 * Initialize the vm_map module. Must be called before
186 * any other vm_map routines.
188 * Map and entry structures are allocated from the general
189 * purpose memory pool with some exceptions:
191 * - The kernel map and kmem submap are allocated statically.
192 * - Kernel map entries are allocated out of a static pool.
194 * These restrictions are necessary since malloc() uses the
195 * maps and requires map entries.
201 mtx_init(&map_sleep_mtx, "vm map sleep mutex", NULL, MTX_DEF);
202 mapzone = uma_zcreate("MAP", sizeof(struct vm_map), NULL,
208 vm_map_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
209 uma_prealloc(mapzone, MAX_KMAP);
210 kmapentzone = uma_zcreate("KMAP ENTRY", sizeof(struct vm_map_entry),
211 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR,
212 UMA_ZONE_MTXCLASS | UMA_ZONE_VM);
213 mapentzone = uma_zcreate("MAP ENTRY", sizeof(struct vm_map_entry),
214 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
215 vmspace_zone = uma_zcreate("VMSPACE", sizeof(struct vmspace), NULL,
221 vmspace_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
225 vmspace_zinit(void *mem, int size, int flags)
229 vm = (struct vmspace *)mem;
231 vm->vm_map.pmap = NULL;
232 (void)vm_map_zinit(&vm->vm_map, sizeof(vm->vm_map), flags);
233 PMAP_LOCK_INIT(vmspace_pmap(vm));
238 vm_map_zinit(void *mem, int size, int flags)
243 memset(map, 0, sizeof(*map));
244 mtx_init(&map->system_mtx, "vm map (system)", NULL, MTX_DEF | MTX_DUPOK);
245 sx_init(&map->lock, "vm map (user)");
251 vmspace_zdtor(void *mem, int size, void *arg)
255 vm = (struct vmspace *)mem;
257 vm_map_zdtor(&vm->vm_map, sizeof(vm->vm_map), arg);
260 vm_map_zdtor(void *mem, int size, void *arg)
265 KASSERT(map->nentries == 0,
266 ("map %p nentries == %d on free.",
267 map, map->nentries));
268 KASSERT(map->size == 0,
269 ("map %p size == %lu on free.",
270 map, (unsigned long)map->size));
272 #endif /* INVARIANTS */
275 * Allocate a vmspace structure, including a vm_map and pmap,
276 * and initialize those structures. The refcnt is set to 1.
278 * If 'pinit' is NULL then the embedded pmap is initialized via pmap_pinit().
281 vmspace_alloc(vm_offset_t min, vm_offset_t max, pmap_pinit_t pinit)
285 vm = uma_zalloc(vmspace_zone, M_WAITOK);
287 KASSERT(vm->vm_map.pmap == NULL, ("vm_map.pmap must be NULL"));
292 if (!pinit(vmspace_pmap(vm))) {
293 uma_zfree(vmspace_zone, vm);
296 CTR1(KTR_VM, "vmspace_alloc: %p", vm);
297 _vm_map_init(&vm->vm_map, vmspace_pmap(vm), min, max);
312 vmspace_container_reset(struct proc *p)
316 racct_set(p, RACCT_DATA, 0);
317 racct_set(p, RACCT_STACK, 0);
318 racct_set(p, RACCT_RSS, 0);
319 racct_set(p, RACCT_MEMLOCK, 0);
320 racct_set(p, RACCT_VMEM, 0);
326 vmspace_dofree(struct vmspace *vm)
329 CTR1(KTR_VM, "vmspace_free: %p", vm);
332 * Make sure any SysV shm is freed, it might not have been in
338 * Lock the map, to wait out all other references to it.
339 * Delete all of the mappings and pages they hold, then call
340 * the pmap module to reclaim anything left.
342 (void)vm_map_remove(&vm->vm_map, vm->vm_map.min_offset,
343 vm->vm_map.max_offset);
345 pmap_release(vmspace_pmap(vm));
346 vm->vm_map.pmap = NULL;
347 uma_zfree(vmspace_zone, vm);
351 vmspace_free(struct vmspace *vm)
354 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
355 "vmspace_free() called");
357 if (vm->vm_refcnt == 0)
358 panic("vmspace_free: attempt to free already freed vmspace");
360 if (atomic_fetchadd_int(&vm->vm_refcnt, -1) == 1)
365 vmspace_exitfree(struct proc *p)
369 PROC_VMSPACE_LOCK(p);
372 PROC_VMSPACE_UNLOCK(p);
373 KASSERT(vm == &vmspace0, ("vmspace_exitfree: wrong vmspace"));
378 vmspace_exit(struct thread *td)
385 * Release user portion of address space.
386 * This releases references to vnodes,
387 * which could cause I/O if the file has been unlinked.
388 * Need to do this early enough that we can still sleep.
390 * The last exiting process to reach this point releases as
391 * much of the environment as it can. vmspace_dofree() is the
392 * slower fallback in case another process had a temporary
393 * reference to the vmspace.
398 atomic_add_int(&vmspace0.vm_refcnt, 1);
400 refcnt = vm->vm_refcnt;
401 if (refcnt > 1 && p->p_vmspace != &vmspace0) {
402 /* Switch now since other proc might free vmspace */
403 PROC_VMSPACE_LOCK(p);
404 p->p_vmspace = &vmspace0;
405 PROC_VMSPACE_UNLOCK(p);
408 } while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt - 1));
410 if (p->p_vmspace != vm) {
411 /* vmspace not yet freed, switch back */
412 PROC_VMSPACE_LOCK(p);
414 PROC_VMSPACE_UNLOCK(p);
417 pmap_remove_pages(vmspace_pmap(vm));
418 /* Switch now since this proc will free vmspace */
419 PROC_VMSPACE_LOCK(p);
420 p->p_vmspace = &vmspace0;
421 PROC_VMSPACE_UNLOCK(p);
427 vmspace_container_reset(p);
431 /* Acquire reference to vmspace owned by another process. */
434 vmspace_acquire_ref(struct proc *p)
439 PROC_VMSPACE_LOCK(p);
442 PROC_VMSPACE_UNLOCK(p);
446 refcnt = vm->vm_refcnt;
447 if (refcnt <= 0) { /* Avoid 0->1 transition */
448 PROC_VMSPACE_UNLOCK(p);
451 } while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt + 1));
452 if (vm != p->p_vmspace) {
453 PROC_VMSPACE_UNLOCK(p);
457 PROC_VMSPACE_UNLOCK(p);
462 * Switch between vmspaces in an AIO kernel process.
464 * The AIO kernel processes switch to and from a user process's
465 * vmspace while performing an I/O operation on behalf of a user
466 * process. The new vmspace is either the vmspace of a user process
467 * obtained from an active AIO request or the initial vmspace of the
468 * AIO kernel process (when it is idling). Because user processes
469 * will block to drain any active AIO requests before proceeding in
470 * exit() or execve(), the vmspace reference count for these vmspaces
471 * can never be 0. This allows for a much simpler implementation than
472 * the loop in vmspace_acquire_ref() above. Similarly, AIO kernel
473 * processes hold an extra reference on their initial vmspace for the
474 * life of the process so that this guarantee is true for any vmspace
478 vmspace_switch_aio(struct vmspace *newvm)
480 struct vmspace *oldvm;
482 /* XXX: Need some way to assert that this is an aio daemon. */
484 KASSERT(newvm->vm_refcnt > 0,
485 ("vmspace_switch_aio: newvm unreferenced"));
487 oldvm = curproc->p_vmspace;
492 * Point to the new address space and refer to it.
494 curproc->p_vmspace = newvm;
495 atomic_add_int(&newvm->vm_refcnt, 1);
497 /* Activate the new mapping. */
498 pmap_activate(curthread);
500 /* Remove the daemon's reference to the old address space. */
501 KASSERT(oldvm->vm_refcnt > 1,
502 ("vmspace_switch_aio: oldvm dropping last reference"));
507 _vm_map_lock(vm_map_t map, const char *file, int line)
511 mtx_lock_flags_(&map->system_mtx, 0, file, line);
513 sx_xlock_(&map->lock, file, line);
518 vm_map_process_deferred(void)
521 vm_map_entry_t entry, next;
525 entry = td->td_map_def_user;
526 td->td_map_def_user = NULL;
527 while (entry != NULL) {
529 if ((entry->eflags & MAP_ENTRY_VN_WRITECNT) != 0) {
531 * Decrement the object's writemappings and
532 * possibly the vnode's v_writecount.
534 KASSERT((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0,
535 ("Submap with writecount"));
536 object = entry->object.vm_object;
537 KASSERT(object != NULL, ("No object for writecount"));
538 vnode_pager_release_writecount(object, entry->start,
541 vm_map_entry_deallocate(entry, FALSE);
547 _vm_map_unlock(vm_map_t map, const char *file, int line)
551 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
553 sx_xunlock_(&map->lock, file, line);
554 vm_map_process_deferred();
559 _vm_map_lock_read(vm_map_t map, const char *file, int line)
563 mtx_lock_flags_(&map->system_mtx, 0, file, line);
565 sx_slock_(&map->lock, file, line);
569 _vm_map_unlock_read(vm_map_t map, const char *file, int line)
573 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
575 sx_sunlock_(&map->lock, file, line);
576 vm_map_process_deferred();
581 _vm_map_trylock(vm_map_t map, const char *file, int line)
585 error = map->system_map ?
586 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) :
587 !sx_try_xlock_(&map->lock, file, line);
594 _vm_map_trylock_read(vm_map_t map, const char *file, int line)
598 error = map->system_map ?
599 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) :
600 !sx_try_slock_(&map->lock, file, line);
605 * _vm_map_lock_upgrade: [ internal use only ]
607 * Tries to upgrade a read (shared) lock on the specified map to a write
608 * (exclusive) lock. Returns the value "0" if the upgrade succeeds and a
609 * non-zero value if the upgrade fails. If the upgrade fails, the map is
610 * returned without a read or write lock held.
612 * Requires that the map be read locked.
615 _vm_map_lock_upgrade(vm_map_t map, const char *file, int line)
617 unsigned int last_timestamp;
619 if (map->system_map) {
620 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
622 if (!sx_try_upgrade_(&map->lock, file, line)) {
623 last_timestamp = map->timestamp;
624 sx_sunlock_(&map->lock, file, line);
625 vm_map_process_deferred();
627 * If the map's timestamp does not change while the
628 * map is unlocked, then the upgrade succeeds.
630 sx_xlock_(&map->lock, file, line);
631 if (last_timestamp != map->timestamp) {
632 sx_xunlock_(&map->lock, file, line);
642 _vm_map_lock_downgrade(vm_map_t map, const char *file, int line)
645 if (map->system_map) {
646 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
648 sx_downgrade_(&map->lock, file, line);
654 * Returns a non-zero value if the caller holds a write (exclusive) lock
655 * on the specified map and the value "0" otherwise.
658 vm_map_locked(vm_map_t map)
662 return (mtx_owned(&map->system_mtx));
664 return (sx_xlocked(&map->lock));
669 _vm_map_assert_locked(vm_map_t map, const char *file, int line)
673 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
675 sx_assert_(&map->lock, SA_XLOCKED, file, line);
678 #define VM_MAP_ASSERT_LOCKED(map) \
679 _vm_map_assert_locked(map, LOCK_FILE, LOCK_LINE)
681 #define VM_MAP_ASSERT_LOCKED(map)
685 * _vm_map_unlock_and_wait:
687 * Atomically releases the lock on the specified map and puts the calling
688 * thread to sleep. The calling thread will remain asleep until either
689 * vm_map_wakeup() is performed on the map or the specified timeout is
692 * WARNING! This function does not perform deferred deallocations of
693 * objects and map entries. Therefore, the calling thread is expected to
694 * reacquire the map lock after reawakening and later perform an ordinary
695 * unlock operation, such as vm_map_unlock(), before completing its
696 * operation on the map.
699 _vm_map_unlock_and_wait(vm_map_t map, int timo, const char *file, int line)
702 mtx_lock(&map_sleep_mtx);
704 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
706 sx_xunlock_(&map->lock, file, line);
707 return (msleep(&map->root, &map_sleep_mtx, PDROP | PVM, "vmmaps",
714 * Awaken any threads that have slept on the map using
715 * vm_map_unlock_and_wait().
718 vm_map_wakeup(vm_map_t map)
722 * Acquire and release map_sleep_mtx to prevent a wakeup()
723 * from being performed (and lost) between the map unlock
724 * and the msleep() in _vm_map_unlock_and_wait().
726 mtx_lock(&map_sleep_mtx);
727 mtx_unlock(&map_sleep_mtx);
732 vm_map_busy(vm_map_t map)
735 VM_MAP_ASSERT_LOCKED(map);
740 vm_map_unbusy(vm_map_t map)
743 VM_MAP_ASSERT_LOCKED(map);
744 KASSERT(map->busy, ("vm_map_unbusy: not busy"));
745 if (--map->busy == 0 && (map->flags & MAP_BUSY_WAKEUP)) {
746 vm_map_modflags(map, 0, MAP_BUSY_WAKEUP);
752 vm_map_wait_busy(vm_map_t map)
755 VM_MAP_ASSERT_LOCKED(map);
757 vm_map_modflags(map, MAP_BUSY_WAKEUP, 0);
759 msleep(&map->busy, &map->system_mtx, 0, "mbusy", 0);
761 sx_sleep(&map->busy, &map->lock, 0, "mbusy", 0);
767 vmspace_resident_count(struct vmspace *vmspace)
769 return pmap_resident_count(vmspace_pmap(vmspace));
775 * Creates and returns a new empty VM map with
776 * the given physical map structure, and having
777 * the given lower and upper address bounds.
780 vm_map_create(pmap_t pmap, vm_offset_t min, vm_offset_t max)
784 result = uma_zalloc(mapzone, M_WAITOK);
785 CTR1(KTR_VM, "vm_map_create: %p", result);
786 _vm_map_init(result, pmap, min, max);
791 * Initialize an existing vm_map structure
792 * such as that in the vmspace structure.
795 _vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max)
798 map->header.next = map->header.prev = &map->header;
799 map->needs_wakeup = FALSE;
802 map->min_offset = min;
803 map->max_offset = max;
811 vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max)
814 _vm_map_init(map, pmap, min, max);
815 mtx_init(&map->system_mtx, "system map", NULL, MTX_DEF | MTX_DUPOK);
816 sx_init(&map->lock, "user map");
820 * vm_map_entry_dispose: [ internal use only ]
822 * Inverse of vm_map_entry_create.
825 vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry)
827 uma_zfree(map->system_map ? kmapentzone : mapentzone, entry);
831 * vm_map_entry_create: [ internal use only ]
833 * Allocates a VM map entry for insertion.
834 * No entry fields are filled in.
836 static vm_map_entry_t
837 vm_map_entry_create(vm_map_t map)
839 vm_map_entry_t new_entry;
842 new_entry = uma_zalloc(kmapentzone, M_NOWAIT);
844 new_entry = uma_zalloc(mapentzone, M_WAITOK);
845 if (new_entry == NULL)
846 panic("vm_map_entry_create: kernel resources exhausted");
851 * vm_map_entry_set_behavior:
853 * Set the expected access behavior, either normal, random, or
857 vm_map_entry_set_behavior(vm_map_entry_t entry, u_char behavior)
859 entry->eflags = (entry->eflags & ~MAP_ENTRY_BEHAV_MASK) |
860 (behavior & MAP_ENTRY_BEHAV_MASK);
864 * vm_map_entry_set_max_free:
866 * Set the max_free field in a vm_map_entry.
869 vm_map_entry_set_max_free(vm_map_entry_t entry)
872 entry->max_free = entry->adj_free;
873 if (entry->left != NULL && entry->left->max_free > entry->max_free)
874 entry->max_free = entry->left->max_free;
875 if (entry->right != NULL && entry->right->max_free > entry->max_free)
876 entry->max_free = entry->right->max_free;
880 * vm_map_entry_splay:
882 * The Sleator and Tarjan top-down splay algorithm with the
883 * following variation. Max_free must be computed bottom-up, so
884 * on the downward pass, maintain the left and right spines in
885 * reverse order. Then, make a second pass up each side to fix
886 * the pointers and compute max_free. The time bound is O(log n)
889 * The new root is the vm_map_entry containing "addr", or else an
890 * adjacent entry (lower or higher) if addr is not in the tree.
892 * The map must be locked, and leaves it so.
894 * Returns: the new root.
896 static vm_map_entry_t
897 vm_map_entry_splay(vm_offset_t addr, vm_map_entry_t root)
899 vm_map_entry_t llist, rlist;
900 vm_map_entry_t ltree, rtree;
903 /* Special case of empty tree. */
908 * Pass One: Splay down the tree until we find addr or a NULL
909 * pointer where addr would go. llist and rlist are the two
910 * sides in reverse order (bottom-up), with llist linked by
911 * the right pointer and rlist linked by the left pointer in
912 * the vm_map_entry. Wait until Pass Two to set max_free on
918 /* root is never NULL in here. */
919 if (addr < root->start) {
923 if (addr < y->start && y->left != NULL) {
924 /* Rotate right and put y on rlist. */
925 root->left = y->right;
927 vm_map_entry_set_max_free(root);
932 /* Put root on rlist. */
937 } else if (addr >= root->end) {
941 if (addr >= y->end && y->right != NULL) {
942 /* Rotate left and put y on llist. */
943 root->right = y->left;
945 vm_map_entry_set_max_free(root);
950 /* Put root on llist. */
960 * Pass Two: Walk back up the two spines, flip the pointers
961 * and set max_free. The subtrees of the root go at the
962 * bottom of llist and rlist.
965 while (llist != NULL) {
967 llist->right = ltree;
968 vm_map_entry_set_max_free(llist);
973 while (rlist != NULL) {
976 vm_map_entry_set_max_free(rlist);
982 * Final assembly: add ltree and rtree as subtrees of root.
986 vm_map_entry_set_max_free(root);
992 * vm_map_entry_{un,}link:
994 * Insert/remove entries from maps.
997 vm_map_entry_link(vm_map_t map,
998 vm_map_entry_t after_where,
999 vm_map_entry_t entry)
1003 "vm_map_entry_link: map %p, nentries %d, entry %p, after %p", map,
1004 map->nentries, entry, after_where);
1005 VM_MAP_ASSERT_LOCKED(map);
1006 KASSERT(after_where->end <= entry->start,
1007 ("vm_map_entry_link: prev end %jx new start %jx overlap",
1008 (uintmax_t)after_where->end, (uintmax_t)entry->start));
1009 KASSERT(entry->end <= after_where->next->start,
1010 ("vm_map_entry_link: new end %jx next start %jx overlap",
1011 (uintmax_t)entry->end, (uintmax_t)after_where->next->start));
1014 entry->prev = after_where;
1015 entry->next = after_where->next;
1016 entry->next->prev = entry;
1017 after_where->next = entry;
1019 if (after_where != &map->header) {
1020 if (after_where != map->root)
1021 vm_map_entry_splay(after_where->start, map->root);
1022 entry->right = after_where->right;
1023 entry->left = after_where;
1024 after_where->right = NULL;
1025 after_where->adj_free = entry->start - after_where->end;
1026 vm_map_entry_set_max_free(after_where);
1028 entry->right = map->root;
1031 entry->adj_free = entry->next->start - entry->end;
1032 vm_map_entry_set_max_free(entry);
1037 vm_map_entry_unlink(vm_map_t map,
1038 vm_map_entry_t entry)
1040 vm_map_entry_t next, prev, root;
1042 VM_MAP_ASSERT_LOCKED(map);
1043 if (entry != map->root)
1044 vm_map_entry_splay(entry->start, map->root);
1045 if (entry->left == NULL)
1046 root = entry->right;
1048 root = vm_map_entry_splay(entry->start, entry->left);
1049 root->right = entry->right;
1050 root->adj_free = entry->next->start - root->end;
1051 vm_map_entry_set_max_free(root);
1060 CTR3(KTR_VM, "vm_map_entry_unlink: map %p, nentries %d, entry %p", map,
1061 map->nentries, entry);
1065 * vm_map_entry_resize_free:
1067 * Recompute the amount of free space following a vm_map_entry
1068 * and propagate that value up the tree. Call this function after
1069 * resizing a map entry in-place, that is, without a call to
1070 * vm_map_entry_link() or _unlink().
1072 * The map must be locked, and leaves it so.
1075 vm_map_entry_resize_free(vm_map_t map, vm_map_entry_t entry)
1079 * Using splay trees without parent pointers, propagating
1080 * max_free up the tree is done by moving the entry to the
1081 * root and making the change there.
1083 if (entry != map->root)
1084 map->root = vm_map_entry_splay(entry->start, map->root);
1086 entry->adj_free = entry->next->start - entry->end;
1087 vm_map_entry_set_max_free(entry);
1091 * vm_map_lookup_entry: [ internal use only ]
1093 * Finds the map entry containing (or
1094 * immediately preceding) the specified address
1095 * in the given map; the entry is returned
1096 * in the "entry" parameter. The boolean
1097 * result indicates whether the address is
1098 * actually contained in the map.
1101 vm_map_lookup_entry(
1103 vm_offset_t address,
1104 vm_map_entry_t *entry) /* OUT */
1110 * If the map is empty, then the map entry immediately preceding
1111 * "address" is the map's header.
1115 *entry = &map->header;
1116 else if (address >= cur->start && cur->end > address) {
1119 } else if ((locked = vm_map_locked(map)) ||
1120 sx_try_upgrade(&map->lock)) {
1122 * Splay requires a write lock on the map. However, it only
1123 * restructures the binary search tree; it does not otherwise
1124 * change the map. Thus, the map's timestamp need not change
1125 * on a temporary upgrade.
1127 map->root = cur = vm_map_entry_splay(address, cur);
1129 sx_downgrade(&map->lock);
1132 * If "address" is contained within a map entry, the new root
1133 * is that map entry. Otherwise, the new root is a map entry
1134 * immediately before or after "address".
1136 if (address >= cur->start) {
1138 if (cur->end > address)
1144 * Since the map is only locked for read access, perform a
1145 * standard binary search tree lookup for "address".
1148 if (address < cur->start) {
1149 if (cur->left == NULL) {
1154 } else if (cur->end > address) {
1158 if (cur->right == NULL) {
1171 * Inserts the given whole VM object into the target
1172 * map at the specified address range. The object's
1173 * size should match that of the address range.
1175 * Requires that the map be locked, and leaves it so.
1177 * If object is non-NULL, ref count must be bumped by caller
1178 * prior to making call to account for the new entry.
1181 vm_map_insert(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1182 vm_offset_t start, vm_offset_t end, vm_prot_t prot, vm_prot_t max, int cow)
1184 vm_map_entry_t new_entry, prev_entry, temp_entry;
1186 vm_eflags_t protoeflags;
1187 vm_inherit_t inheritance;
1189 VM_MAP_ASSERT_LOCKED(map);
1190 KASSERT(object != kernel_object ||
1191 (cow & MAP_COPY_ON_WRITE) == 0,
1192 ("vm_map_insert: kernel object and COW"));
1193 KASSERT(object == NULL || (cow & MAP_NOFAULT) == 0,
1194 ("vm_map_insert: paradoxical MAP_NOFAULT request"));
1195 KASSERT((prot & ~max) == 0,
1196 ("prot %#x is not subset of max_prot %#x", prot, max));
1199 * Check that the start and end points are not bogus.
1201 if (start < map->min_offset || end > map->max_offset || start >= end)
1202 return (KERN_INVALID_ADDRESS);
1205 * Find the entry prior to the proposed starting address; if it's part
1206 * of an existing entry, this range is bogus.
1208 if (vm_map_lookup_entry(map, start, &temp_entry))
1209 return (KERN_NO_SPACE);
1211 prev_entry = temp_entry;
1214 * Assert that the next entry doesn't overlap the end point.
1216 if (prev_entry->next->start < end)
1217 return (KERN_NO_SPACE);
1219 if ((cow & MAP_CREATE_GUARD) != 0 && (object != NULL ||
1220 max != VM_PROT_NONE))
1221 return (KERN_INVALID_ARGUMENT);
1224 if (cow & MAP_COPY_ON_WRITE)
1225 protoeflags |= MAP_ENTRY_COW | MAP_ENTRY_NEEDS_COPY;
1226 if (cow & MAP_NOFAULT)
1227 protoeflags |= MAP_ENTRY_NOFAULT;
1228 if (cow & MAP_DISABLE_SYNCER)
1229 protoeflags |= MAP_ENTRY_NOSYNC;
1230 if (cow & MAP_DISABLE_COREDUMP)
1231 protoeflags |= MAP_ENTRY_NOCOREDUMP;
1232 if (cow & MAP_STACK_GROWS_DOWN)
1233 protoeflags |= MAP_ENTRY_GROWS_DOWN;
1234 if (cow & MAP_STACK_GROWS_UP)
1235 protoeflags |= MAP_ENTRY_GROWS_UP;
1236 if (cow & MAP_VN_WRITECOUNT)
1237 protoeflags |= MAP_ENTRY_VN_WRITECNT;
1238 if ((cow & MAP_CREATE_GUARD) != 0)
1239 protoeflags |= MAP_ENTRY_GUARD;
1240 if ((cow & MAP_CREATE_STACK_GAP_DN) != 0)
1241 protoeflags |= MAP_ENTRY_STACK_GAP_DN;
1242 if ((cow & MAP_CREATE_STACK_GAP_UP) != 0)
1243 protoeflags |= MAP_ENTRY_STACK_GAP_UP;
1244 if (cow & MAP_INHERIT_SHARE)
1245 inheritance = VM_INHERIT_SHARE;
1247 inheritance = VM_INHERIT_DEFAULT;
1250 if ((cow & (MAP_ACC_NO_CHARGE | MAP_NOFAULT | MAP_CREATE_GUARD)) != 0)
1252 if ((cow & MAP_ACC_CHARGED) || ((prot & VM_PROT_WRITE) &&
1253 ((protoeflags & MAP_ENTRY_NEEDS_COPY) || object == NULL))) {
1254 if (!(cow & MAP_ACC_CHARGED) && !swap_reserve(end - start))
1255 return (KERN_RESOURCE_SHORTAGE);
1256 KASSERT(object == NULL ||
1257 (protoeflags & MAP_ENTRY_NEEDS_COPY) != 0 ||
1258 object->cred == NULL,
1259 ("overcommit: vm_map_insert o %p", object));
1260 cred = curthread->td_ucred;
1264 /* Expand the kernel pmap, if necessary. */
1265 if (map == kernel_map && end > kernel_vm_end)
1266 pmap_growkernel(end);
1267 if (object != NULL) {
1269 * OBJ_ONEMAPPING must be cleared unless this mapping
1270 * is trivially proven to be the only mapping for any
1271 * of the object's pages. (Object granularity
1272 * reference counting is insufficient to recognize
1273 * aliases with precision.)
1275 VM_OBJECT_WLOCK(object);
1276 if (object->ref_count > 1 || object->shadow_count != 0)
1277 vm_object_clear_flag(object, OBJ_ONEMAPPING);
1278 VM_OBJECT_WUNLOCK(object);
1279 } else if (prev_entry != &map->header &&
1280 prev_entry->eflags == protoeflags &&
1281 (cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 &&
1282 prev_entry->end == start && prev_entry->wired_count == 0 &&
1283 (prev_entry->cred == cred ||
1284 (prev_entry->object.vm_object != NULL &&
1285 prev_entry->object.vm_object->cred == cred)) &&
1286 vm_object_coalesce(prev_entry->object.vm_object,
1288 (vm_size_t)(prev_entry->end - prev_entry->start),
1289 (vm_size_t)(end - prev_entry->end), cred != NULL &&
1290 (protoeflags & MAP_ENTRY_NEEDS_COPY) == 0)) {
1292 * We were able to extend the object. Determine if we
1293 * can extend the previous map entry to include the
1294 * new range as well.
1296 if (prev_entry->inheritance == inheritance &&
1297 prev_entry->protection == prot &&
1298 prev_entry->max_protection == max) {
1299 if ((prev_entry->eflags & MAP_ENTRY_GUARD) == 0)
1300 map->size += end - prev_entry->end;
1301 prev_entry->end = end;
1302 vm_map_entry_resize_free(map, prev_entry);
1303 vm_map_simplify_entry(map, prev_entry);
1304 return (KERN_SUCCESS);
1308 * If we can extend the object but cannot extend the
1309 * map entry, we have to create a new map entry. We
1310 * must bump the ref count on the extended object to
1311 * account for it. object may be NULL.
1313 object = prev_entry->object.vm_object;
1314 offset = prev_entry->offset +
1315 (prev_entry->end - prev_entry->start);
1316 vm_object_reference(object);
1317 if (cred != NULL && object != NULL && object->cred != NULL &&
1318 !(prev_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
1319 /* Object already accounts for this uid. */
1327 * Create a new entry
1329 new_entry = vm_map_entry_create(map);
1330 new_entry->start = start;
1331 new_entry->end = end;
1332 new_entry->cred = NULL;
1334 new_entry->eflags = protoeflags;
1335 new_entry->object.vm_object = object;
1336 new_entry->offset = offset;
1338 new_entry->inheritance = inheritance;
1339 new_entry->protection = prot;
1340 new_entry->max_protection = max;
1341 new_entry->wired_count = 0;
1342 new_entry->wiring_thread = NULL;
1343 new_entry->read_ahead = VM_FAULT_READ_AHEAD_INIT;
1344 new_entry->next_read = start;
1346 KASSERT(cred == NULL || !ENTRY_CHARGED(new_entry),
1347 ("overcommit: vm_map_insert leaks vm_map %p", new_entry));
1348 new_entry->cred = cred;
1351 * Insert the new entry into the list
1353 vm_map_entry_link(map, prev_entry, new_entry);
1354 if ((new_entry->eflags & MAP_ENTRY_GUARD) == 0)
1355 map->size += new_entry->end - new_entry->start;
1358 * Try to coalesce the new entry with both the previous and next
1359 * entries in the list. Previously, we only attempted to coalesce
1360 * with the previous entry when object is NULL. Here, we handle the
1361 * other cases, which are less common.
1363 vm_map_simplify_entry(map, new_entry);
1365 if ((cow & (MAP_PREFAULT | MAP_PREFAULT_PARTIAL)) != 0) {
1366 vm_map_pmap_enter(map, start, prot, object, OFF_TO_IDX(offset),
1367 end - start, cow & MAP_PREFAULT_PARTIAL);
1370 return (KERN_SUCCESS);
1376 * Find the first fit (lowest VM address) for "length" free bytes
1377 * beginning at address >= start in the given map.
1379 * In a vm_map_entry, "adj_free" is the amount of free space
1380 * adjacent (higher address) to this entry, and "max_free" is the
1381 * maximum amount of contiguous free space in its subtree. This
1382 * allows finding a free region in one path down the tree, so
1383 * O(log n) amortized with splay trees.
1385 * The map must be locked, and leaves it so.
1387 * Returns: 0 on success, and starting address in *addr,
1388 * 1 if insufficient space.
1391 vm_map_findspace(vm_map_t map, vm_offset_t start, vm_size_t length,
1392 vm_offset_t *addr) /* OUT */
1394 vm_map_entry_t entry;
1398 * Request must fit within min/max VM address and must avoid
1401 if (start < map->min_offset)
1402 start = map->min_offset;
1403 if (start + length > map->max_offset || start + length < start)
1406 /* Empty tree means wide open address space. */
1407 if (map->root == NULL) {
1413 * After splay, if start comes before root node, then there
1414 * must be a gap from start to the root.
1416 map->root = vm_map_entry_splay(start, map->root);
1417 if (start + length <= map->root->start) {
1423 * Root is the last node that might begin its gap before
1424 * start, and this is the last comparison where address
1425 * wrap might be a problem.
1427 st = (start > map->root->end) ? start : map->root->end;
1428 if (length <= map->root->end + map->root->adj_free - st) {
1433 /* With max_free, can immediately tell if no solution. */
1434 entry = map->root->right;
1435 if (entry == NULL || length > entry->max_free)
1439 * Search the right subtree in the order: left subtree, root,
1440 * right subtree (first fit). The previous splay implies that
1441 * all regions in the right subtree have addresses > start.
1443 while (entry != NULL) {
1444 if (entry->left != NULL && entry->left->max_free >= length)
1445 entry = entry->left;
1446 else if (entry->adj_free >= length) {
1450 entry = entry->right;
1453 /* Can't get here, so panic if we do. */
1454 panic("vm_map_findspace: max_free corrupt");
1458 vm_map_fixed(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1459 vm_offset_t start, vm_size_t length, vm_prot_t prot,
1460 vm_prot_t max, int cow)
1465 end = start + length;
1466 KASSERT((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 ||
1468 ("vm_map_fixed: non-NULL backing object for stack"));
1470 VM_MAP_RANGE_CHECK(map, start, end);
1471 if ((cow & MAP_CHECK_EXCL) == 0)
1472 vm_map_delete(map, start, end);
1473 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) != 0) {
1474 result = vm_map_stack_locked(map, start, length, sgrowsiz,
1477 result = vm_map_insert(map, object, offset, start, end,
1485 * Searches for the specified amount of free space in the given map with the
1486 * specified alignment. Performs an address-ordered, first-fit search from
1487 * the given address "*addr", with an optional upper bound "max_addr". If the
1488 * parameter "alignment" is zero, then the alignment is computed from the
1489 * given (object, offset) pair so as to enable the greatest possible use of
1490 * superpage mappings. Returns KERN_SUCCESS and the address of the free space
1491 * in "*addr" if successful. Otherwise, returns KERN_NO_SPACE.
1493 * The map must be locked. Initially, there must be at least "length" bytes
1494 * of free space at the given address.
1497 vm_map_alignspace(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1498 vm_offset_t *addr, vm_size_t length, vm_offset_t max_addr,
1499 vm_offset_t alignment)
1501 vm_offset_t aligned_addr, free_addr;
1503 VM_MAP_ASSERT_LOCKED(map);
1505 KASSERT(!vm_map_findspace(map, free_addr, length, addr) &&
1506 free_addr == *addr, ("caller provided insufficient free space"));
1509 * At the start of every iteration, the free space at address
1510 * "*addr" is at least "length" bytes.
1513 pmap_align_superpage(object, offset, addr, length);
1514 else if ((*addr & (alignment - 1)) != 0) {
1515 *addr &= ~(alignment - 1);
1518 aligned_addr = *addr;
1519 if (aligned_addr == free_addr) {
1521 * Alignment did not change "*addr", so "*addr" must
1522 * still provide sufficient free space.
1524 return (KERN_SUCCESS);
1528 * Test for address wrap on "*addr". A wrapped "*addr" could
1529 * be a valid address, in which case vm_map_findspace() cannot
1530 * be relied upon to fail.
1532 if (aligned_addr < free_addr ||
1533 vm_map_findspace(map, aligned_addr, length, addr) ||
1534 (max_addr != 0 && *addr + length > max_addr))
1535 return (KERN_NO_SPACE);
1537 if (free_addr == aligned_addr) {
1539 * If a successful call to vm_map_findspace() did not
1540 * change "*addr", then "*addr" must still be aligned
1541 * and provide sufficient free space.
1543 return (KERN_SUCCESS);
1549 * vm_map_find finds an unallocated region in the target address
1550 * map with the given length. The search is defined to be
1551 * first-fit from the specified address; the region found is
1552 * returned in the same parameter.
1554 * If object is non-NULL, ref count must be bumped by caller
1555 * prior to making call to account for the new entry.
1558 vm_map_find(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1559 vm_offset_t *addr, /* IN/OUT */
1560 vm_size_t length, vm_offset_t max_addr, int find_space,
1561 vm_prot_t prot, vm_prot_t max, int cow)
1563 vm_offset_t alignment, min_addr;
1566 KASSERT((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 ||
1568 ("vm_map_find: non-NULL backing object for stack"));
1569 if (find_space == VMFS_OPTIMAL_SPACE && (object == NULL ||
1570 (object->flags & OBJ_COLORED) == 0))
1571 find_space = VMFS_ANY_SPACE;
1572 if (find_space >> 8 != 0) {
1573 KASSERT((find_space & 0xff) == 0, ("bad VMFS flags"));
1574 alignment = (vm_offset_t)1 << (find_space >> 8);
1578 if (find_space != VMFS_NO_SPACE) {
1579 KASSERT(find_space == VMFS_ANY_SPACE ||
1580 find_space == VMFS_OPTIMAL_SPACE ||
1581 find_space == VMFS_SUPER_SPACE ||
1582 alignment != 0, ("unexpected VMFS flag"));
1585 if (vm_map_findspace(map, min_addr, length, addr) ||
1586 (max_addr != 0 && *addr + length > max_addr)) {
1590 if (find_space != VMFS_ANY_SPACE &&
1591 (rv = vm_map_alignspace(map, object, offset, addr, length,
1592 max_addr, alignment)) != KERN_SUCCESS) {
1593 if (find_space == VMFS_OPTIMAL_SPACE) {
1594 find_space = VMFS_ANY_SPACE;
1600 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) != 0) {
1601 rv = vm_map_stack_locked(map, *addr, length, sgrowsiz, prot,
1604 rv = vm_map_insert(map, object, offset, *addr, *addr + length,
1613 * vm_map_find_min() is a variant of vm_map_find() that takes an
1614 * additional parameter (min_addr) and treats the given address
1615 * (*addr) differently. Specifically, it treats *addr as a hint
1616 * and not as the minimum address where the mapping is created.
1618 * This function works in two phases. First, it tries to
1619 * allocate above the hint. If that fails and the hint is
1620 * greater than min_addr, it performs a second pass, replacing
1621 * the hint with min_addr as the minimum address for the
1625 vm_map_find_min(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1626 vm_offset_t *addr, vm_size_t length, vm_offset_t min_addr,
1627 vm_offset_t max_addr, int find_space, vm_prot_t prot, vm_prot_t max,
1635 rv = vm_map_find(map, object, offset, addr, length, max_addr,
1636 find_space, prot, max, cow);
1637 if (rv == KERN_SUCCESS || min_addr >= hint)
1639 *addr = hint = min_addr;
1644 * vm_map_simplify_entry:
1646 * Simplify the given map entry by merging with either neighbor. This
1647 * routine also has the ability to merge with both neighbors.
1649 * The map must be locked.
1651 * This routine guarantees that the passed entry remains valid (though
1652 * possibly extended). When merging, this routine may delete one or
1656 vm_map_simplify_entry(vm_map_t map, vm_map_entry_t entry)
1658 vm_map_entry_t next, prev;
1659 vm_size_t prevsize, esize;
1661 if ((entry->eflags & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP |
1662 MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_IS_SUB_MAP)) != 0)
1666 if (prev != &map->header) {
1667 prevsize = prev->end - prev->start;
1668 if ( (prev->end == entry->start) &&
1669 (prev->object.vm_object == entry->object.vm_object) &&
1670 (!prev->object.vm_object ||
1671 (prev->offset + prevsize == entry->offset)) &&
1672 (prev->eflags == entry->eflags) &&
1673 (prev->protection == entry->protection) &&
1674 (prev->max_protection == entry->max_protection) &&
1675 (prev->inheritance == entry->inheritance) &&
1676 (prev->wired_count == entry->wired_count) &&
1677 (prev->cred == entry->cred)) {
1678 vm_map_entry_unlink(map, prev);
1679 entry->start = prev->start;
1680 entry->offset = prev->offset;
1681 if (entry->prev != &map->header)
1682 vm_map_entry_resize_free(map, entry->prev);
1685 * If the backing object is a vnode object,
1686 * vm_object_deallocate() calls vrele().
1687 * However, vrele() does not lock the vnode
1688 * because the vnode has additional
1689 * references. Thus, the map lock can be kept
1690 * without causing a lock-order reversal with
1693 * Since we count the number of virtual page
1694 * mappings in object->un_pager.vnp.writemappings,
1695 * the writemappings value should not be adjusted
1696 * when the entry is disposed of.
1698 if (prev->object.vm_object)
1699 vm_object_deallocate(prev->object.vm_object);
1700 if (prev->cred != NULL)
1702 vm_map_entry_dispose(map, prev);
1707 if (next != &map->header) {
1708 esize = entry->end - entry->start;
1709 if ((entry->end == next->start) &&
1710 (next->object.vm_object == entry->object.vm_object) &&
1711 (!entry->object.vm_object ||
1712 (entry->offset + esize == next->offset)) &&
1713 (next->eflags == entry->eflags) &&
1714 (next->protection == entry->protection) &&
1715 (next->max_protection == entry->max_protection) &&
1716 (next->inheritance == entry->inheritance) &&
1717 (next->wired_count == entry->wired_count) &&
1718 (next->cred == entry->cred)) {
1719 vm_map_entry_unlink(map, next);
1720 entry->end = next->end;
1721 vm_map_entry_resize_free(map, entry);
1724 * See comment above.
1726 if (next->object.vm_object)
1727 vm_object_deallocate(next->object.vm_object);
1728 if (next->cred != NULL)
1730 vm_map_entry_dispose(map, next);
1735 * vm_map_clip_start: [ internal use only ]
1737 * Asserts that the given entry begins at or after
1738 * the specified address; if necessary,
1739 * it splits the entry into two.
1741 #define vm_map_clip_start(map, entry, startaddr) \
1743 if (startaddr > entry->start) \
1744 _vm_map_clip_start(map, entry, startaddr); \
1748 * This routine is called only when it is known that
1749 * the entry must be split.
1752 _vm_map_clip_start(vm_map_t map, vm_map_entry_t entry, vm_offset_t start)
1754 vm_map_entry_t new_entry;
1756 VM_MAP_ASSERT_LOCKED(map);
1757 KASSERT(entry->end > start && entry->start < start,
1758 ("_vm_map_clip_start: invalid clip of entry %p", entry));
1761 * Split off the front portion -- note that we must insert the new
1762 * entry BEFORE this one, so that this entry has the specified
1765 vm_map_simplify_entry(map, entry);
1768 * If there is no object backing this entry, we might as well create
1769 * one now. If we defer it, an object can get created after the map
1770 * is clipped, and individual objects will be created for the split-up
1771 * map. This is a bit of a hack, but is also about the best place to
1772 * put this improvement.
1774 if (entry->object.vm_object == NULL && !map->system_map &&
1775 (entry->eflags & MAP_ENTRY_GUARD) == 0) {
1777 object = vm_object_allocate(OBJT_DEFAULT,
1778 atop(entry->end - entry->start));
1779 entry->object.vm_object = object;
1781 if (entry->cred != NULL) {
1782 object->cred = entry->cred;
1783 object->charge = entry->end - entry->start;
1786 } else if (entry->object.vm_object != NULL &&
1787 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) &&
1788 entry->cred != NULL) {
1789 VM_OBJECT_WLOCK(entry->object.vm_object);
1790 KASSERT(entry->object.vm_object->cred == NULL,
1791 ("OVERCOMMIT: vm_entry_clip_start: both cred e %p", entry));
1792 entry->object.vm_object->cred = entry->cred;
1793 entry->object.vm_object->charge = entry->end - entry->start;
1794 VM_OBJECT_WUNLOCK(entry->object.vm_object);
1798 new_entry = vm_map_entry_create(map);
1799 *new_entry = *entry;
1801 new_entry->end = start;
1802 entry->offset += (start - entry->start);
1803 entry->start = start;
1804 if (new_entry->cred != NULL)
1805 crhold(entry->cred);
1807 vm_map_entry_link(map, entry->prev, new_entry);
1809 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1810 vm_object_reference(new_entry->object.vm_object);
1812 * The object->un_pager.vnp.writemappings for the
1813 * object of MAP_ENTRY_VN_WRITECNT type entry shall be
1814 * kept as is here. The virtual pages are
1815 * re-distributed among the clipped entries, so the sum is
1822 * vm_map_clip_end: [ internal use only ]
1824 * Asserts that the given entry ends at or before
1825 * the specified address; if necessary,
1826 * it splits the entry into two.
1828 #define vm_map_clip_end(map, entry, endaddr) \
1830 if ((endaddr) < (entry->end)) \
1831 _vm_map_clip_end((map), (entry), (endaddr)); \
1835 * This routine is called only when it is known that
1836 * the entry must be split.
1839 _vm_map_clip_end(vm_map_t map, vm_map_entry_t entry, vm_offset_t end)
1841 vm_map_entry_t new_entry;
1843 VM_MAP_ASSERT_LOCKED(map);
1844 KASSERT(entry->start < end && entry->end > end,
1845 ("_vm_map_clip_end: invalid clip of entry %p", entry));
1848 * If there is no object backing this entry, we might as well create
1849 * one now. If we defer it, an object can get created after the map
1850 * is clipped, and individual objects will be created for the split-up
1851 * map. This is a bit of a hack, but is also about the best place to
1852 * put this improvement.
1854 if (entry->object.vm_object == NULL && !map->system_map &&
1855 (entry->eflags & MAP_ENTRY_GUARD) == 0) {
1857 object = vm_object_allocate(OBJT_DEFAULT,
1858 atop(entry->end - entry->start));
1859 entry->object.vm_object = object;
1861 if (entry->cred != NULL) {
1862 object->cred = entry->cred;
1863 object->charge = entry->end - entry->start;
1866 } else if (entry->object.vm_object != NULL &&
1867 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) &&
1868 entry->cred != NULL) {
1869 VM_OBJECT_WLOCK(entry->object.vm_object);
1870 KASSERT(entry->object.vm_object->cred == NULL,
1871 ("OVERCOMMIT: vm_entry_clip_end: both cred e %p", entry));
1872 entry->object.vm_object->cred = entry->cred;
1873 entry->object.vm_object->charge = entry->end - entry->start;
1874 VM_OBJECT_WUNLOCK(entry->object.vm_object);
1879 * Create a new entry and insert it AFTER the specified entry
1881 new_entry = vm_map_entry_create(map);
1882 *new_entry = *entry;
1884 new_entry->start = entry->end = end;
1885 new_entry->offset += (end - entry->start);
1886 if (new_entry->cred != NULL)
1887 crhold(entry->cred);
1889 vm_map_entry_link(map, entry, new_entry);
1891 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1892 vm_object_reference(new_entry->object.vm_object);
1897 * vm_map_submap: [ kernel use only ]
1899 * Mark the given range as handled by a subordinate map.
1901 * This range must have been created with vm_map_find,
1902 * and no other operations may have been performed on this
1903 * range prior to calling vm_map_submap.
1905 * Only a limited number of operations can be performed
1906 * within this rage after calling vm_map_submap:
1908 * [Don't try vm_map_copy!]
1910 * To remove a submapping, one must first remove the
1911 * range from the superior map, and then destroy the
1912 * submap (if desired). [Better yet, don't try it.]
1921 vm_map_entry_t entry;
1922 int result = KERN_INVALID_ARGUMENT;
1926 VM_MAP_RANGE_CHECK(map, start, end);
1928 if (vm_map_lookup_entry(map, start, &entry)) {
1929 vm_map_clip_start(map, entry, start);
1931 entry = entry->next;
1933 vm_map_clip_end(map, entry, end);
1935 if ((entry->start == start) && (entry->end == end) &&
1936 ((entry->eflags & MAP_ENTRY_COW) == 0) &&
1937 (entry->object.vm_object == NULL)) {
1938 entry->object.sub_map = submap;
1939 entry->eflags |= MAP_ENTRY_IS_SUB_MAP;
1940 result = KERN_SUCCESS;
1948 * The maximum number of pages to map if MAP_PREFAULT_PARTIAL is specified
1950 #define MAX_INIT_PT 96
1953 * vm_map_pmap_enter:
1955 * Preload the specified map's pmap with mappings to the specified
1956 * object's memory-resident pages. No further physical pages are
1957 * allocated, and no further virtual pages are retrieved from secondary
1958 * storage. If the specified flags include MAP_PREFAULT_PARTIAL, then a
1959 * limited number of page mappings are created at the low-end of the
1960 * specified address range. (For this purpose, a superpage mapping
1961 * counts as one page mapping.) Otherwise, all resident pages within
1962 * the specified address range are mapped.
1965 vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot,
1966 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags)
1969 vm_page_t p, p_start;
1970 vm_pindex_t mask, psize, threshold, tmpidx;
1972 if ((prot & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0 || object == NULL)
1974 VM_OBJECT_RLOCK(object);
1975 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) {
1976 VM_OBJECT_RUNLOCK(object);
1977 VM_OBJECT_WLOCK(object);
1978 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) {
1979 pmap_object_init_pt(map->pmap, addr, object, pindex,
1981 VM_OBJECT_WUNLOCK(object);
1984 VM_OBJECT_LOCK_DOWNGRADE(object);
1988 if (psize + pindex > object->size) {
1989 if (object->size < pindex) {
1990 VM_OBJECT_RUNLOCK(object);
1993 psize = object->size - pindex;
1998 threshold = MAX_INIT_PT;
2000 p = vm_page_find_least(object, pindex);
2002 * Assert: the variable p is either (1) the page with the
2003 * least pindex greater than or equal to the parameter pindex
2007 p != NULL && (tmpidx = p->pindex - pindex) < psize;
2008 p = TAILQ_NEXT(p, listq)) {
2010 * don't allow an madvise to blow away our really
2011 * free pages allocating pv entries.
2013 if (((flags & MAP_PREFAULT_MADVISE) != 0 &&
2014 vm_page_count_severe()) ||
2015 ((flags & MAP_PREFAULT_PARTIAL) != 0 &&
2016 tmpidx >= threshold)) {
2020 if (p->valid == VM_PAGE_BITS_ALL) {
2021 if (p_start == NULL) {
2022 start = addr + ptoa(tmpidx);
2025 /* Jump ahead if a superpage mapping is possible. */
2026 if (p->psind > 0 && ((addr + ptoa(tmpidx)) &
2027 (pagesizes[p->psind] - 1)) == 0) {
2028 mask = atop(pagesizes[p->psind]) - 1;
2029 if (tmpidx + mask < psize &&
2030 vm_page_ps_test(p, PS_ALL_VALID, NULL)) {
2035 } else if (p_start != NULL) {
2036 pmap_enter_object(map->pmap, start, addr +
2037 ptoa(tmpidx), p_start, prot);
2041 if (p_start != NULL)
2042 pmap_enter_object(map->pmap, start, addr + ptoa(psize),
2044 VM_OBJECT_RUNLOCK(object);
2050 * Sets the protection of the specified address
2051 * region in the target map. If "set_max" is
2052 * specified, the maximum protection is to be set;
2053 * otherwise, only the current protection is affected.
2056 vm_map_protect(vm_map_t map, vm_offset_t start, vm_offset_t end,
2057 vm_prot_t new_prot, boolean_t set_max)
2059 vm_map_entry_t current, entry;
2065 return (KERN_SUCCESS);
2070 * Ensure that we are not concurrently wiring pages. vm_map_wire() may
2071 * need to fault pages into the map and will drop the map lock while
2072 * doing so, and the VM object may end up in an inconsistent state if we
2073 * update the protection on the map entry in between faults.
2075 vm_map_wait_busy(map);
2077 VM_MAP_RANGE_CHECK(map, start, end);
2079 if (vm_map_lookup_entry(map, start, &entry)) {
2080 vm_map_clip_start(map, entry, start);
2082 entry = entry->next;
2086 * Make a first pass to check for protection violations.
2088 for (current = entry; current->start < end; current = current->next) {
2089 if ((current->eflags & MAP_ENTRY_GUARD) != 0)
2091 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
2093 return (KERN_INVALID_ARGUMENT);
2095 if ((new_prot & current->max_protection) != new_prot) {
2097 return (KERN_PROTECTION_FAILURE);
2102 * Do an accounting pass for private read-only mappings that
2103 * now will do cow due to allowed write (e.g. debugger sets
2104 * breakpoint on text segment)
2106 for (current = entry; current->start < end; current = current->next) {
2108 vm_map_clip_end(map, current, end);
2111 ((new_prot & ~(current->protection)) & VM_PROT_WRITE) == 0 ||
2112 ENTRY_CHARGED(current) ||
2113 (current->eflags & MAP_ENTRY_GUARD) != 0) {
2117 cred = curthread->td_ucred;
2118 obj = current->object.vm_object;
2120 if (obj == NULL || (current->eflags & MAP_ENTRY_NEEDS_COPY)) {
2121 if (!swap_reserve(current->end - current->start)) {
2123 return (KERN_RESOURCE_SHORTAGE);
2126 current->cred = cred;
2130 VM_OBJECT_WLOCK(obj);
2131 if (obj->type != OBJT_DEFAULT && obj->type != OBJT_SWAP) {
2132 VM_OBJECT_WUNLOCK(obj);
2137 * Charge for the whole object allocation now, since
2138 * we cannot distinguish between non-charged and
2139 * charged clipped mapping of the same object later.
2141 KASSERT(obj->charge == 0,
2142 ("vm_map_protect: object %p overcharged (entry %p)",
2144 if (!swap_reserve(ptoa(obj->size))) {
2145 VM_OBJECT_WUNLOCK(obj);
2147 return (KERN_RESOURCE_SHORTAGE);
2152 obj->charge = ptoa(obj->size);
2153 VM_OBJECT_WUNLOCK(obj);
2157 * Go back and fix up protections. [Note that clipping is not
2158 * necessary the second time.]
2160 for (current = entry; current->start < end; current = current->next) {
2161 if ((current->eflags & MAP_ENTRY_GUARD) != 0)
2164 old_prot = current->protection;
2167 current->protection =
2168 (current->max_protection = new_prot) &
2171 current->protection = new_prot;
2174 * For user wired map entries, the normal lazy evaluation of
2175 * write access upgrades through soft page faults is
2176 * undesirable. Instead, immediately copy any pages that are
2177 * copy-on-write and enable write access in the physical map.
2179 if ((current->eflags & MAP_ENTRY_USER_WIRED) != 0 &&
2180 (current->protection & VM_PROT_WRITE) != 0 &&
2181 (old_prot & VM_PROT_WRITE) == 0)
2182 vm_fault_copy_entry(map, map, current, current, NULL);
2185 * When restricting access, update the physical map. Worry
2186 * about copy-on-write here.
2188 if ((old_prot & ~current->protection) != 0) {
2189 #define MASK(entry) (((entry)->eflags & MAP_ENTRY_COW) ? ~VM_PROT_WRITE : \
2191 pmap_protect(map->pmap, current->start,
2193 current->protection & MASK(current));
2196 vm_map_simplify_entry(map, current);
2199 return (KERN_SUCCESS);
2205 * This routine traverses a processes map handling the madvise
2206 * system call. Advisories are classified as either those effecting
2207 * the vm_map_entry structure, or those effecting the underlying
2217 vm_map_entry_t current, entry;
2221 * Some madvise calls directly modify the vm_map_entry, in which case
2222 * we need to use an exclusive lock on the map and we need to perform
2223 * various clipping operations. Otherwise we only need a read-lock
2228 case MADV_SEQUENTIAL:
2245 vm_map_lock_read(map);
2252 * Locate starting entry and clip if necessary.
2254 VM_MAP_RANGE_CHECK(map, start, end);
2256 if (vm_map_lookup_entry(map, start, &entry)) {
2258 vm_map_clip_start(map, entry, start);
2260 entry = entry->next;
2265 * madvise behaviors that are implemented in the vm_map_entry.
2267 * We clip the vm_map_entry so that behavioral changes are
2268 * limited to the specified address range.
2270 for (current = entry; current->start < end;
2271 current = current->next) {
2272 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
2275 vm_map_clip_end(map, current, end);
2279 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_NORMAL);
2281 case MADV_SEQUENTIAL:
2282 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_SEQUENTIAL);
2285 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_RANDOM);
2288 current->eflags |= MAP_ENTRY_NOSYNC;
2291 current->eflags &= ~MAP_ENTRY_NOSYNC;
2294 current->eflags |= MAP_ENTRY_NOCOREDUMP;
2297 current->eflags &= ~MAP_ENTRY_NOCOREDUMP;
2302 vm_map_simplify_entry(map, current);
2306 vm_pindex_t pstart, pend;
2309 * madvise behaviors that are implemented in the underlying
2312 * Since we don't clip the vm_map_entry, we have to clip
2313 * the vm_object pindex and count.
2315 for (current = entry; current->start < end;
2316 current = current->next) {
2317 vm_offset_t useEnd, useStart;
2319 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
2322 pstart = OFF_TO_IDX(current->offset);
2323 pend = pstart + atop(current->end - current->start);
2324 useStart = current->start;
2325 useEnd = current->end;
2327 if (current->start < start) {
2328 pstart += atop(start - current->start);
2331 if (current->end > end) {
2332 pend -= atop(current->end - end);
2340 * Perform the pmap_advise() before clearing
2341 * PGA_REFERENCED in vm_page_advise(). Otherwise, a
2342 * concurrent pmap operation, such as pmap_remove(),
2343 * could clear a reference in the pmap and set
2344 * PGA_REFERENCED on the page before the pmap_advise()
2345 * had completed. Consequently, the page would appear
2346 * referenced based upon an old reference that
2347 * occurred before this pmap_advise() ran.
2349 if (behav == MADV_DONTNEED || behav == MADV_FREE)
2350 pmap_advise(map->pmap, useStart, useEnd,
2353 vm_object_madvise(current->object.vm_object, pstart,
2357 * Pre-populate paging structures in the
2358 * WILLNEED case. For wired entries, the
2359 * paging structures are already populated.
2361 if (behav == MADV_WILLNEED &&
2362 current->wired_count == 0) {
2363 vm_map_pmap_enter(map,
2365 current->protection,
2366 current->object.vm_object,
2368 ptoa(pend - pstart),
2369 MAP_PREFAULT_MADVISE
2373 vm_map_unlock_read(map);
2382 * Sets the inheritance of the specified address
2383 * range in the target map. Inheritance
2384 * affects how the map will be shared with
2385 * child maps at the time of vmspace_fork.
2388 vm_map_inherit(vm_map_t map, vm_offset_t start, vm_offset_t end,
2389 vm_inherit_t new_inheritance)
2391 vm_map_entry_t entry;
2392 vm_map_entry_t temp_entry;
2394 switch (new_inheritance) {
2395 case VM_INHERIT_NONE:
2396 case VM_INHERIT_COPY:
2397 case VM_INHERIT_SHARE:
2398 case VM_INHERIT_ZERO:
2401 return (KERN_INVALID_ARGUMENT);
2404 return (KERN_SUCCESS);
2406 VM_MAP_RANGE_CHECK(map, start, end);
2407 if (vm_map_lookup_entry(map, start, &temp_entry)) {
2409 vm_map_clip_start(map, entry, start);
2411 entry = temp_entry->next;
2412 while (entry->start < end) {
2413 vm_map_clip_end(map, entry, end);
2414 if ((entry->eflags & MAP_ENTRY_GUARD) == 0 ||
2415 new_inheritance != VM_INHERIT_ZERO)
2416 entry->inheritance = new_inheritance;
2417 vm_map_simplify_entry(map, entry);
2418 entry = entry->next;
2421 return (KERN_SUCCESS);
2427 * Implements both kernel and user unwiring.
2430 vm_map_unwire(vm_map_t map, vm_offset_t start, vm_offset_t end,
2433 vm_map_entry_t entry, first_entry, tmp_entry;
2434 vm_offset_t saved_start;
2435 unsigned int last_timestamp;
2437 boolean_t need_wakeup, result, user_unwire;
2440 return (KERN_SUCCESS);
2441 user_unwire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE;
2443 VM_MAP_RANGE_CHECK(map, start, end);
2444 if (!vm_map_lookup_entry(map, start, &first_entry)) {
2445 if (flags & VM_MAP_WIRE_HOLESOK)
2446 first_entry = first_entry->next;
2449 return (KERN_INVALID_ADDRESS);
2452 last_timestamp = map->timestamp;
2453 entry = first_entry;
2454 while (entry->start < end) {
2455 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
2457 * We have not yet clipped the entry.
2459 saved_start = (start >= entry->start) ? start :
2461 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2462 if (vm_map_unlock_and_wait(map, 0)) {
2464 * Allow interruption of user unwiring?
2468 if (last_timestamp+1 != map->timestamp) {
2470 * Look again for the entry because the map was
2471 * modified while it was unlocked.
2472 * Specifically, the entry may have been
2473 * clipped, merged, or deleted.
2475 if (!vm_map_lookup_entry(map, saved_start,
2477 if (flags & VM_MAP_WIRE_HOLESOK)
2478 tmp_entry = tmp_entry->next;
2480 if (saved_start == start) {
2482 * First_entry has been deleted.
2485 return (KERN_INVALID_ADDRESS);
2488 rv = KERN_INVALID_ADDRESS;
2492 if (entry == first_entry)
2493 first_entry = tmp_entry;
2498 last_timestamp = map->timestamp;
2501 vm_map_clip_start(map, entry, start);
2502 vm_map_clip_end(map, entry, end);
2504 * Mark the entry in case the map lock is released. (See
2507 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 &&
2508 entry->wiring_thread == NULL,
2509 ("owned map entry %p", entry));
2510 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
2511 entry->wiring_thread = curthread;
2513 * Check the map for holes in the specified region.
2514 * If VM_MAP_WIRE_HOLESOK was specified, skip this check.
2516 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) &&
2517 (entry->end < end && entry->next->start > entry->end)) {
2519 rv = KERN_INVALID_ADDRESS;
2523 * If system unwiring, require that the entry is system wired.
2526 vm_map_entry_system_wired_count(entry) == 0) {
2528 rv = KERN_INVALID_ARGUMENT;
2531 entry = entry->next;
2535 need_wakeup = FALSE;
2536 if (first_entry == NULL) {
2537 result = vm_map_lookup_entry(map, start, &first_entry);
2538 if (!result && (flags & VM_MAP_WIRE_HOLESOK))
2539 first_entry = first_entry->next;
2541 KASSERT(result, ("vm_map_unwire: lookup failed"));
2543 for (entry = first_entry; entry->start < end; entry = entry->next) {
2545 * If VM_MAP_WIRE_HOLESOK was specified, an empty
2546 * space in the unwired region could have been mapped
2547 * while the map lock was dropped for draining
2548 * MAP_ENTRY_IN_TRANSITION. Moreover, another thread
2549 * could be simultaneously wiring this new mapping
2550 * entry. Detect these cases and skip any entries
2551 * marked as in transition by us.
2553 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 ||
2554 entry->wiring_thread != curthread) {
2555 KASSERT((flags & VM_MAP_WIRE_HOLESOK) != 0,
2556 ("vm_map_unwire: !HOLESOK and new/changed entry"));
2560 if (rv == KERN_SUCCESS && (!user_unwire ||
2561 (entry->eflags & MAP_ENTRY_USER_WIRED))) {
2563 entry->eflags &= ~MAP_ENTRY_USER_WIRED;
2564 if (entry->wired_count == 1)
2565 vm_map_entry_unwire(map, entry);
2567 entry->wired_count--;
2569 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0,
2570 ("vm_map_unwire: in-transition flag missing %p", entry));
2571 KASSERT(entry->wiring_thread == curthread,
2572 ("vm_map_unwire: alien wire %p", entry));
2573 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION;
2574 entry->wiring_thread = NULL;
2575 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
2576 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
2579 vm_map_simplify_entry(map, entry);
2588 * vm_map_wire_entry_failure:
2590 * Handle a wiring failure on the given entry.
2592 * The map should be locked.
2595 vm_map_wire_entry_failure(vm_map_t map, vm_map_entry_t entry,
2596 vm_offset_t failed_addr)
2599 VM_MAP_ASSERT_LOCKED(map);
2600 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 &&
2601 entry->wired_count == 1,
2602 ("vm_map_wire_entry_failure: entry %p isn't being wired", entry));
2603 KASSERT(failed_addr < entry->end,
2604 ("vm_map_wire_entry_failure: entry %p was fully wired", entry));
2607 * If any pages at the start of this entry were successfully wired,
2610 if (failed_addr > entry->start) {
2611 pmap_unwire(map->pmap, entry->start, failed_addr);
2612 vm_object_unwire(entry->object.vm_object, entry->offset,
2613 failed_addr - entry->start, PQ_ACTIVE);
2617 * Assign an out-of-range value to represent the failure to wire this
2620 entry->wired_count = -1;
2626 * Implements both kernel and user wiring.
2629 vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t end,
2632 vm_map_entry_t entry, first_entry, tmp_entry;
2633 vm_offset_t faddr, saved_end, saved_start;
2634 unsigned int last_timestamp;
2636 boolean_t need_wakeup, result, user_wire;
2640 return (KERN_SUCCESS);
2642 if (flags & VM_MAP_WIRE_WRITE)
2643 prot |= VM_PROT_WRITE;
2644 user_wire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE;
2646 VM_MAP_RANGE_CHECK(map, start, end);
2647 if (!vm_map_lookup_entry(map, start, &first_entry)) {
2648 if (flags & VM_MAP_WIRE_HOLESOK)
2649 first_entry = first_entry->next;
2652 return (KERN_INVALID_ADDRESS);
2655 last_timestamp = map->timestamp;
2656 entry = first_entry;
2657 while (entry->start < end) {
2658 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
2660 * We have not yet clipped the entry.
2662 saved_start = (start >= entry->start) ? start :
2664 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2665 if (vm_map_unlock_and_wait(map, 0)) {
2667 * Allow interruption of user wiring?
2671 if (last_timestamp + 1 != map->timestamp) {
2673 * Look again for the entry because the map was
2674 * modified while it was unlocked.
2675 * Specifically, the entry may have been
2676 * clipped, merged, or deleted.
2678 if (!vm_map_lookup_entry(map, saved_start,
2680 if (flags & VM_MAP_WIRE_HOLESOK)
2681 tmp_entry = tmp_entry->next;
2683 if (saved_start == start) {
2685 * first_entry has been deleted.
2688 return (KERN_INVALID_ADDRESS);
2691 rv = KERN_INVALID_ADDRESS;
2695 if (entry == first_entry)
2696 first_entry = tmp_entry;
2701 last_timestamp = map->timestamp;
2704 vm_map_clip_start(map, entry, start);
2705 vm_map_clip_end(map, entry, end);
2707 * Mark the entry in case the map lock is released. (See
2710 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 &&
2711 entry->wiring_thread == NULL,
2712 ("owned map entry %p", entry));
2713 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
2714 entry->wiring_thread = curthread;
2715 if ((entry->protection & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0
2716 || (entry->protection & prot) != prot) {
2717 entry->eflags |= MAP_ENTRY_WIRE_SKIPPED;
2718 if ((flags & VM_MAP_WIRE_HOLESOK) == 0) {
2720 rv = KERN_INVALID_ADDRESS;
2725 if (entry->wired_count == 0) {
2726 entry->wired_count++;
2727 saved_start = entry->start;
2728 saved_end = entry->end;
2731 * Release the map lock, relying on the in-transition
2732 * mark. Mark the map busy for fork.
2737 faddr = saved_start;
2740 * Simulate a fault to get the page and enter
2741 * it into the physical map.
2743 if ((rv = vm_fault(map, faddr, VM_PROT_NONE,
2744 VM_FAULT_WIRE)) != KERN_SUCCESS)
2746 } while ((faddr += PAGE_SIZE) < saved_end);
2749 if (last_timestamp + 1 != map->timestamp) {
2751 * Look again for the entry because the map was
2752 * modified while it was unlocked. The entry
2753 * may have been clipped, but NOT merged or
2756 result = vm_map_lookup_entry(map, saved_start,
2758 KASSERT(result, ("vm_map_wire: lookup failed"));
2759 if (entry == first_entry)
2760 first_entry = tmp_entry;
2764 while (entry->end < saved_end) {
2766 * In case of failure, handle entries
2767 * that were not fully wired here;
2768 * fully wired entries are handled
2771 if (rv != KERN_SUCCESS &&
2773 vm_map_wire_entry_failure(map,
2775 entry = entry->next;
2778 last_timestamp = map->timestamp;
2779 if (rv != KERN_SUCCESS) {
2780 vm_map_wire_entry_failure(map, entry, faddr);
2784 } else if (!user_wire ||
2785 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
2786 entry->wired_count++;
2789 * Check the map for holes in the specified region.
2790 * If VM_MAP_WIRE_HOLESOK was specified, skip this check.
2793 if ((flags & VM_MAP_WIRE_HOLESOK) == 0 &&
2794 entry->end < end && entry->next->start > entry->end) {
2796 rv = KERN_INVALID_ADDRESS;
2799 entry = entry->next;
2803 need_wakeup = FALSE;
2804 if (first_entry == NULL) {
2805 result = vm_map_lookup_entry(map, start, &first_entry);
2806 if (!result && (flags & VM_MAP_WIRE_HOLESOK))
2807 first_entry = first_entry->next;
2809 KASSERT(result, ("vm_map_wire: lookup failed"));
2811 for (entry = first_entry; entry->start < end; entry = entry->next) {
2813 * If VM_MAP_WIRE_HOLESOK was specified, an empty
2814 * space in the unwired region could have been mapped
2815 * while the map lock was dropped for faulting in the
2816 * pages or draining MAP_ENTRY_IN_TRANSITION.
2817 * Moreover, another thread could be simultaneously
2818 * wiring this new mapping entry. Detect these cases
2819 * and skip any entries marked as in transition not by us.
2821 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 ||
2822 entry->wiring_thread != curthread) {
2823 KASSERT((flags & VM_MAP_WIRE_HOLESOK) != 0,
2824 ("vm_map_wire: !HOLESOK and new/changed entry"));
2828 if ((entry->eflags & MAP_ENTRY_WIRE_SKIPPED) != 0)
2829 goto next_entry_done;
2831 if (rv == KERN_SUCCESS) {
2833 entry->eflags |= MAP_ENTRY_USER_WIRED;
2834 } else if (entry->wired_count == -1) {
2836 * Wiring failed on this entry. Thus, unwiring is
2839 entry->wired_count = 0;
2840 } else if (!user_wire ||
2841 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
2843 * Undo the wiring. Wiring succeeded on this entry
2844 * but failed on a later entry.
2846 if (entry->wired_count == 1)
2847 vm_map_entry_unwire(map, entry);
2849 entry->wired_count--;
2852 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0,
2853 ("vm_map_wire: in-transition flag missing %p", entry));
2854 KASSERT(entry->wiring_thread == curthread,
2855 ("vm_map_wire: alien wire %p", entry));
2856 entry->eflags &= ~(MAP_ENTRY_IN_TRANSITION |
2857 MAP_ENTRY_WIRE_SKIPPED);
2858 entry->wiring_thread = NULL;
2859 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
2860 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
2863 vm_map_simplify_entry(map, entry);
2874 * Push any dirty cached pages in the address range to their pager.
2875 * If syncio is TRUE, dirty pages are written synchronously.
2876 * If invalidate is TRUE, any cached pages are freed as well.
2878 * If the size of the region from start to end is zero, we are
2879 * supposed to flush all modified pages within the region containing
2880 * start. Unfortunately, a region can be split or coalesced with
2881 * neighboring regions, making it difficult to determine what the
2882 * original region was. Therefore, we approximate this requirement by
2883 * flushing the current region containing start.
2885 * Returns an error if any part of the specified range is not mapped.
2893 boolean_t invalidate)
2895 vm_map_entry_t current;
2896 vm_map_entry_t entry;
2899 vm_ooffset_t offset;
2900 unsigned int last_timestamp;
2903 vm_map_lock_read(map);
2904 VM_MAP_RANGE_CHECK(map, start, end);
2905 if (!vm_map_lookup_entry(map, start, &entry)) {
2906 vm_map_unlock_read(map);
2907 return (KERN_INVALID_ADDRESS);
2908 } else if (start == end) {
2909 start = entry->start;
2913 * Make a first pass to check for user-wired memory and holes.
2915 for (current = entry; current->start < end; current = current->next) {
2916 if (invalidate && (current->eflags & MAP_ENTRY_USER_WIRED)) {
2917 vm_map_unlock_read(map);
2918 return (KERN_INVALID_ARGUMENT);
2920 if (end > current->end &&
2921 current->end != current->next->start) {
2922 vm_map_unlock_read(map);
2923 return (KERN_INVALID_ADDRESS);
2928 pmap_remove(map->pmap, start, end);
2932 * Make a second pass, cleaning/uncaching pages from the indicated
2935 for (current = entry; current->start < end;) {
2936 offset = current->offset + (start - current->start);
2937 size = (end <= current->end ? end : current->end) - start;
2938 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
2940 vm_map_entry_t tentry;
2943 smap = current->object.sub_map;
2944 vm_map_lock_read(smap);
2945 (void) vm_map_lookup_entry(smap, offset, &tentry);
2946 tsize = tentry->end - offset;
2949 object = tentry->object.vm_object;
2950 offset = tentry->offset + (offset - tentry->start);
2951 vm_map_unlock_read(smap);
2953 object = current->object.vm_object;
2955 vm_object_reference(object);
2956 last_timestamp = map->timestamp;
2957 vm_map_unlock_read(map);
2958 if (!vm_object_sync(object, offset, size, syncio, invalidate))
2961 vm_object_deallocate(object);
2962 vm_map_lock_read(map);
2963 if (last_timestamp == map->timestamp ||
2964 !vm_map_lookup_entry(map, start, ¤t))
2965 current = current->next;
2968 vm_map_unlock_read(map);
2969 return (failed ? KERN_FAILURE : KERN_SUCCESS);
2973 * vm_map_entry_unwire: [ internal use only ]
2975 * Make the region specified by this entry pageable.
2977 * The map in question should be locked.
2978 * [This is the reason for this routine's existence.]
2981 vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry)
2984 VM_MAP_ASSERT_LOCKED(map);
2985 KASSERT(entry->wired_count > 0,
2986 ("vm_map_entry_unwire: entry %p isn't wired", entry));
2987 pmap_unwire(map->pmap, entry->start, entry->end);
2988 vm_object_unwire(entry->object.vm_object, entry->offset, entry->end -
2989 entry->start, PQ_ACTIVE);
2990 entry->wired_count = 0;
2994 vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map)
2997 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0)
2998 vm_object_deallocate(entry->object.vm_object);
2999 uma_zfree(system_map ? kmapentzone : mapentzone, entry);
3003 * vm_map_entry_delete: [ internal use only ]
3005 * Deallocate the given entry from the target map.
3008 vm_map_entry_delete(vm_map_t map, vm_map_entry_t entry)
3011 vm_pindex_t offidxstart, offidxend, count, size1;
3014 vm_map_entry_unlink(map, entry);
3015 object = entry->object.vm_object;
3017 if ((entry->eflags & MAP_ENTRY_GUARD) != 0) {
3018 MPASS(entry->cred == NULL);
3019 MPASS((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0);
3020 MPASS(object == NULL);
3021 vm_map_entry_deallocate(entry, map->system_map);
3025 size = entry->end - entry->start;
3028 if (entry->cred != NULL) {
3029 swap_release_by_cred(size, entry->cred);
3030 crfree(entry->cred);
3033 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0 &&
3035 KASSERT(entry->cred == NULL || object->cred == NULL ||
3036 (entry->eflags & MAP_ENTRY_NEEDS_COPY),
3037 ("OVERCOMMIT vm_map_entry_delete: both cred %p", entry));
3039 offidxstart = OFF_TO_IDX(entry->offset);
3040 offidxend = offidxstart + count;
3041 VM_OBJECT_WLOCK(object);
3042 if (object->ref_count != 1 && ((object->flags & (OBJ_NOSPLIT |
3043 OBJ_ONEMAPPING)) == OBJ_ONEMAPPING ||
3044 object == kernel_object)) {
3045 vm_object_collapse(object);
3048 * The option OBJPR_NOTMAPPED can be passed here
3049 * because vm_map_delete() already performed
3050 * pmap_remove() on the only mapping to this range
3053 vm_object_page_remove(object, offidxstart, offidxend,
3055 if (object->type == OBJT_SWAP)
3056 swap_pager_freespace(object, offidxstart,
3058 if (offidxend >= object->size &&
3059 offidxstart < object->size) {
3060 size1 = object->size;
3061 object->size = offidxstart;
3062 if (object->cred != NULL) {
3063 size1 -= object->size;
3064 KASSERT(object->charge >= ptoa(size1),
3065 ("object %p charge < 0", object));
3066 swap_release_by_cred(ptoa(size1),
3068 object->charge -= ptoa(size1);
3072 VM_OBJECT_WUNLOCK(object);
3074 entry->object.vm_object = NULL;
3075 if (map->system_map)
3076 vm_map_entry_deallocate(entry, TRUE);
3078 entry->next = curthread->td_map_def_user;
3079 curthread->td_map_def_user = entry;
3084 * vm_map_delete: [ internal use only ]
3086 * Deallocates the given address range from the target
3090 vm_map_delete(vm_map_t map, vm_offset_t start, vm_offset_t end)
3092 vm_map_entry_t entry;
3093 vm_map_entry_t first_entry;
3095 VM_MAP_ASSERT_LOCKED(map);
3097 return (KERN_SUCCESS);
3100 * Find the start of the region, and clip it
3102 if (!vm_map_lookup_entry(map, start, &first_entry))
3103 entry = first_entry->next;
3105 entry = first_entry;
3106 vm_map_clip_start(map, entry, start);
3110 * Step through all entries in this region
3112 while (entry->start < end) {
3113 vm_map_entry_t next;
3116 * Wait for wiring or unwiring of an entry to complete.
3117 * Also wait for any system wirings to disappear on
3120 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 ||
3121 (vm_map_pmap(map) != kernel_pmap &&
3122 vm_map_entry_system_wired_count(entry) != 0)) {
3123 unsigned int last_timestamp;
3124 vm_offset_t saved_start;
3125 vm_map_entry_t tmp_entry;
3127 saved_start = entry->start;
3128 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
3129 last_timestamp = map->timestamp;
3130 (void) vm_map_unlock_and_wait(map, 0);
3132 if (last_timestamp + 1 != map->timestamp) {
3134 * Look again for the entry because the map was
3135 * modified while it was unlocked.
3136 * Specifically, the entry may have been
3137 * clipped, merged, or deleted.
3139 if (!vm_map_lookup_entry(map, saved_start,
3141 entry = tmp_entry->next;
3144 vm_map_clip_start(map, entry,
3150 vm_map_clip_end(map, entry, end);
3155 * Unwire before removing addresses from the pmap; otherwise,
3156 * unwiring will put the entries back in the pmap.
3158 if (entry->wired_count != 0)
3159 vm_map_entry_unwire(map, entry);
3162 * Remove mappings for the pages, but only if the
3163 * mappings could exist. For instance, it does not
3164 * make sense to call pmap_remove() for guard entries.
3166 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0 ||
3167 entry->object.vm_object != NULL)
3168 pmap_remove(map->pmap, entry->start, entry->end);
3171 * Delete the entry only after removing all pmap
3172 * entries pointing to its pages. (Otherwise, its
3173 * page frames may be reallocated, and any modify bits
3174 * will be set in the wrong object!)
3176 vm_map_entry_delete(map, entry);
3179 return (KERN_SUCCESS);
3185 * Remove the given address range from the target map.
3186 * This is the exported form of vm_map_delete.
3189 vm_map_remove(vm_map_t map, vm_offset_t start, vm_offset_t end)
3194 VM_MAP_RANGE_CHECK(map, start, end);
3195 result = vm_map_delete(map, start, end);
3201 * vm_map_check_protection:
3203 * Assert that the target map allows the specified privilege on the
3204 * entire address region given. The entire region must be allocated.
3206 * WARNING! This code does not and should not check whether the
3207 * contents of the region is accessible. For example a smaller file
3208 * might be mapped into a larger address space.
3210 * NOTE! This code is also called by munmap().
3212 * The map must be locked. A read lock is sufficient.
3215 vm_map_check_protection(vm_map_t map, vm_offset_t start, vm_offset_t end,
3216 vm_prot_t protection)
3218 vm_map_entry_t entry;
3219 vm_map_entry_t tmp_entry;
3221 if (!vm_map_lookup_entry(map, start, &tmp_entry))
3225 while (start < end) {
3229 if (start < entry->start)
3232 * Check protection associated with entry.
3234 if ((entry->protection & protection) != protection)
3236 /* go to next entry */
3238 entry = entry->next;
3244 * vm_map_copy_entry:
3246 * Copies the contents of the source entry to the destination
3247 * entry. The entries *must* be aligned properly.
3253 vm_map_entry_t src_entry,
3254 vm_map_entry_t dst_entry,
3255 vm_ooffset_t *fork_charge)
3257 vm_object_t src_object;
3258 vm_map_entry_t fake_entry;
3263 VM_MAP_ASSERT_LOCKED(dst_map);
3265 if ((dst_entry->eflags|src_entry->eflags) & MAP_ENTRY_IS_SUB_MAP)
3268 if (src_entry->wired_count == 0 ||
3269 (src_entry->protection & VM_PROT_WRITE) == 0) {
3271 * If the source entry is marked needs_copy, it is already
3274 if ((src_entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0 &&
3275 (src_entry->protection & VM_PROT_WRITE) != 0) {
3276 pmap_protect(src_map->pmap,
3279 src_entry->protection & ~VM_PROT_WRITE);
3283 * Make a copy of the object.
3285 size = src_entry->end - src_entry->start;
3286 if ((src_object = src_entry->object.vm_object) != NULL) {
3287 VM_OBJECT_WLOCK(src_object);
3288 charged = ENTRY_CHARGED(src_entry);
3289 if (src_object->handle == NULL &&
3290 (src_object->type == OBJT_DEFAULT ||
3291 src_object->type == OBJT_SWAP)) {
3292 vm_object_collapse(src_object);
3293 if ((src_object->flags & (OBJ_NOSPLIT |
3294 OBJ_ONEMAPPING)) == OBJ_ONEMAPPING) {
3295 vm_object_split(src_entry);
3297 src_entry->object.vm_object;
3300 vm_object_reference_locked(src_object);
3301 vm_object_clear_flag(src_object, OBJ_ONEMAPPING);
3302 if (src_entry->cred != NULL &&
3303 !(src_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
3304 KASSERT(src_object->cred == NULL,
3305 ("OVERCOMMIT: vm_map_copy_entry: cred %p",
3307 src_object->cred = src_entry->cred;
3308 src_object->charge = size;
3310 VM_OBJECT_WUNLOCK(src_object);
3311 dst_entry->object.vm_object = src_object;
3313 cred = curthread->td_ucred;
3315 dst_entry->cred = cred;
3316 *fork_charge += size;
3317 if (!(src_entry->eflags &
3318 MAP_ENTRY_NEEDS_COPY)) {
3320 src_entry->cred = cred;
3321 *fork_charge += size;
3324 src_entry->eflags |= MAP_ENTRY_COW |
3325 MAP_ENTRY_NEEDS_COPY;
3326 dst_entry->eflags |= MAP_ENTRY_COW |
3327 MAP_ENTRY_NEEDS_COPY;
3328 dst_entry->offset = src_entry->offset;
3329 if (src_entry->eflags & MAP_ENTRY_VN_WRITECNT) {
3331 * MAP_ENTRY_VN_WRITECNT cannot
3332 * indicate write reference from
3333 * src_entry, since the entry is
3334 * marked as needs copy. Allocate a
3335 * fake entry that is used to
3336 * decrement object->un_pager.vnp.writecount
3337 * at the appropriate time. Attach
3338 * fake_entry to the deferred list.
3340 fake_entry = vm_map_entry_create(dst_map);
3341 fake_entry->eflags = MAP_ENTRY_VN_WRITECNT;
3342 src_entry->eflags &= ~MAP_ENTRY_VN_WRITECNT;
3343 vm_object_reference(src_object);
3344 fake_entry->object.vm_object = src_object;
3345 fake_entry->start = src_entry->start;
3346 fake_entry->end = src_entry->end;
3347 fake_entry->next = curthread->td_map_def_user;
3348 curthread->td_map_def_user = fake_entry;
3351 pmap_copy(dst_map->pmap, src_map->pmap,
3352 dst_entry->start, dst_entry->end - dst_entry->start,
3355 dst_entry->object.vm_object = NULL;
3356 dst_entry->offset = 0;
3357 if (src_entry->cred != NULL) {
3358 dst_entry->cred = curthread->td_ucred;
3359 crhold(dst_entry->cred);
3360 *fork_charge += size;
3365 * We don't want to make writeable wired pages copy-on-write.
3366 * Immediately copy these pages into the new map by simulating
3367 * page faults. The new pages are pageable.
3369 vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry,
3375 * vmspace_map_entry_forked:
3376 * Update the newly-forked vmspace each time a map entry is inherited
3377 * or copied. The values for vm_dsize and vm_tsize are approximate
3378 * (and mostly-obsolete ideas in the face of mmap(2) et al.)
3381 vmspace_map_entry_forked(const struct vmspace *vm1, struct vmspace *vm2,
3382 vm_map_entry_t entry)
3384 vm_size_t entrysize;
3387 if ((entry->eflags & MAP_ENTRY_GUARD) != 0)
3389 entrysize = entry->end - entry->start;
3390 vm2->vm_map.size += entrysize;
3391 if (entry->eflags & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP)) {
3392 vm2->vm_ssize += btoc(entrysize);
3393 } else if (entry->start >= (vm_offset_t)vm1->vm_daddr &&
3394 entry->start < (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)) {
3395 newend = MIN(entry->end,
3396 (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize));
3397 vm2->vm_dsize += btoc(newend - entry->start);
3398 } else if (entry->start >= (vm_offset_t)vm1->vm_taddr &&
3399 entry->start < (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)) {
3400 newend = MIN(entry->end,
3401 (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize));
3402 vm2->vm_tsize += btoc(newend - entry->start);
3408 * Create a new process vmspace structure and vm_map
3409 * based on those of an existing process. The new map
3410 * is based on the old map, according to the inheritance
3411 * values on the regions in that map.
3413 * XXX It might be worth coalescing the entries added to the new vmspace.
3415 * The source map must not be locked.
3418 vmspace_fork(struct vmspace *vm1, vm_ooffset_t *fork_charge)
3420 struct vmspace *vm2;
3421 vm_map_t new_map, old_map;
3422 vm_map_entry_t new_entry, old_entry;
3427 old_map = &vm1->vm_map;
3428 /* Copy immutable fields of vm1 to vm2. */
3429 vm2 = vmspace_alloc(old_map->min_offset, old_map->max_offset, NULL);
3432 vm2->vm_taddr = vm1->vm_taddr;
3433 vm2->vm_daddr = vm1->vm_daddr;
3434 vm2->vm_maxsaddr = vm1->vm_maxsaddr;
3435 vm_map_lock(old_map);
3437 vm_map_wait_busy(old_map);
3438 new_map = &vm2->vm_map;
3439 locked = vm_map_trylock(new_map); /* trylock to silence WITNESS */
3440 KASSERT(locked, ("vmspace_fork: lock failed"));
3442 old_entry = old_map->header.next;
3444 while (old_entry != &old_map->header) {
3445 if (old_entry->eflags & MAP_ENTRY_IS_SUB_MAP)
3446 panic("vm_map_fork: encountered a submap");
3448 inh = old_entry->inheritance;
3449 if ((old_entry->eflags & MAP_ENTRY_GUARD) != 0 &&
3450 inh != VM_INHERIT_NONE)
3451 inh = VM_INHERIT_COPY;
3454 case VM_INHERIT_NONE:
3457 case VM_INHERIT_SHARE:
3459 * Clone the entry, creating the shared object if necessary.
3461 object = old_entry->object.vm_object;
3462 if (object == NULL) {
3463 object = vm_object_allocate(OBJT_DEFAULT,
3464 atop(old_entry->end - old_entry->start));
3465 old_entry->object.vm_object = object;
3466 old_entry->offset = 0;
3467 if (old_entry->cred != NULL) {
3468 object->cred = old_entry->cred;
3469 object->charge = old_entry->end -
3471 old_entry->cred = NULL;
3476 * Add the reference before calling vm_object_shadow
3477 * to insure that a shadow object is created.
3479 vm_object_reference(object);
3480 if (old_entry->eflags & MAP_ENTRY_NEEDS_COPY) {
3481 vm_object_shadow(&old_entry->object.vm_object,
3483 old_entry->end - old_entry->start);
3484 old_entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
3485 /* Transfer the second reference too. */
3486 vm_object_reference(
3487 old_entry->object.vm_object);
3490 * As in vm_map_simplify_entry(), the
3491 * vnode lock will not be acquired in
3492 * this call to vm_object_deallocate().
3494 vm_object_deallocate(object);
3495 object = old_entry->object.vm_object;
3497 VM_OBJECT_WLOCK(object);
3498 vm_object_clear_flag(object, OBJ_ONEMAPPING);
3499 if (old_entry->cred != NULL) {
3500 KASSERT(object->cred == NULL, ("vmspace_fork both cred"));
3501 object->cred = old_entry->cred;
3502 object->charge = old_entry->end - old_entry->start;
3503 old_entry->cred = NULL;
3507 * Assert the correct state of the vnode
3508 * v_writecount while the object is locked, to
3509 * not relock it later for the assertion
3512 if (old_entry->eflags & MAP_ENTRY_VN_WRITECNT &&
3513 object->type == OBJT_VNODE) {
3514 KASSERT(((struct vnode *)object->handle)->
3516 ("vmspace_fork: v_writecount %p", object));
3517 KASSERT(object->un_pager.vnp.writemappings > 0,
3518 ("vmspace_fork: vnp.writecount %p",
3521 VM_OBJECT_WUNLOCK(object);
3524 * Clone the entry, referencing the shared object.
3526 new_entry = vm_map_entry_create(new_map);
3527 *new_entry = *old_entry;
3528 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED |
3529 MAP_ENTRY_IN_TRANSITION);
3530 new_entry->wiring_thread = NULL;
3531 new_entry->wired_count = 0;
3532 if (new_entry->eflags & MAP_ENTRY_VN_WRITECNT) {
3533 vnode_pager_update_writecount(object,
3534 new_entry->start, new_entry->end);
3538 * Insert the entry into the new map -- we know we're
3539 * inserting at the end of the new map.
3541 vm_map_entry_link(new_map, new_map->header.prev,
3543 vmspace_map_entry_forked(vm1, vm2, new_entry);
3546 * Update the physical map
3548 pmap_copy(new_map->pmap, old_map->pmap,
3550 (old_entry->end - old_entry->start),
3554 case VM_INHERIT_COPY:
3556 * Clone the entry and link into the map.
3558 new_entry = vm_map_entry_create(new_map);
3559 *new_entry = *old_entry;
3561 * Copied entry is COW over the old object.
3563 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED |
3564 MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_VN_WRITECNT);
3565 new_entry->wiring_thread = NULL;
3566 new_entry->wired_count = 0;
3567 new_entry->object.vm_object = NULL;
3568 new_entry->cred = NULL;
3569 vm_map_entry_link(new_map, new_map->header.prev,
3571 vmspace_map_entry_forked(vm1, vm2, new_entry);
3572 vm_map_copy_entry(old_map, new_map, old_entry,
3573 new_entry, fork_charge);
3576 case VM_INHERIT_ZERO:
3578 * Create a new anonymous mapping entry modelled from
3581 new_entry = vm_map_entry_create(new_map);
3582 memset(new_entry, 0, sizeof(*new_entry));
3584 new_entry->start = old_entry->start;
3585 new_entry->end = old_entry->end;
3586 new_entry->eflags = old_entry->eflags &
3587 ~(MAP_ENTRY_USER_WIRED | MAP_ENTRY_IN_TRANSITION |
3588 MAP_ENTRY_VN_WRITECNT);
3589 new_entry->protection = old_entry->protection;
3590 new_entry->max_protection = old_entry->max_protection;
3591 new_entry->inheritance = VM_INHERIT_ZERO;
3593 vm_map_entry_link(new_map, new_map->header.prev,
3595 vmspace_map_entry_forked(vm1, vm2, new_entry);
3597 new_entry->cred = curthread->td_ucred;
3598 crhold(new_entry->cred);
3599 *fork_charge += (new_entry->end - new_entry->start);
3603 old_entry = old_entry->next;
3606 * Use inlined vm_map_unlock() to postpone handling the deferred
3607 * map entries, which cannot be done until both old_map and
3608 * new_map locks are released.
3610 sx_xunlock(&old_map->lock);
3611 sx_xunlock(&new_map->lock);
3612 vm_map_process_deferred();
3618 * Create a process's stack for exec_new_vmspace(). This function is never
3619 * asked to wire the newly created stack.
3622 vm_map_stack(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
3623 vm_prot_t prot, vm_prot_t max, int cow)
3625 vm_size_t growsize, init_ssize;
3629 MPASS((map->flags & MAP_WIREFUTURE) == 0);
3630 growsize = sgrowsiz;
3631 init_ssize = (max_ssize < growsize) ? max_ssize : growsize;
3633 vmemlim = lim_cur(curthread, RLIMIT_VMEM);
3634 /* If we would blow our VMEM resource limit, no go */
3635 if (map->size + init_ssize > vmemlim) {
3639 rv = vm_map_stack_locked(map, addrbos, max_ssize, growsize, prot,
3646 static int stack_guard_page = 1;
3647 SYSCTL_INT(_security_bsd, OID_AUTO, stack_guard_page, CTLFLAG_RWTUN,
3648 &stack_guard_page, 0,
3649 "Specifies the number of guard pages for a stack that grows");
3652 vm_map_stack_locked(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
3653 vm_size_t growsize, vm_prot_t prot, vm_prot_t max, int cow)
3655 vm_map_entry_t new_entry, prev_entry;
3656 vm_offset_t bot, gap_bot, gap_top, top;
3657 vm_size_t init_ssize, sgp;
3661 * The stack orientation is piggybacked with the cow argument.
3662 * Extract it into orient and mask the cow argument so that we
3663 * don't pass it around further.
3665 orient = cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP);
3666 KASSERT(orient != 0, ("No stack grow direction"));
3667 KASSERT(orient != (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP),
3670 if (addrbos < vm_map_min(map) ||
3671 addrbos + max_ssize > vm_map_max(map) ||
3672 addrbos + max_ssize <= addrbos)
3673 return (KERN_INVALID_ADDRESS);
3674 sgp = (vm_size_t)stack_guard_page * PAGE_SIZE;
3675 if (sgp >= max_ssize)
3676 return (KERN_INVALID_ARGUMENT);
3678 init_ssize = growsize;
3679 if (max_ssize < init_ssize + sgp)
3680 init_ssize = max_ssize - sgp;
3682 /* If addr is already mapped, no go */
3683 if (vm_map_lookup_entry(map, addrbos, &prev_entry))
3684 return (KERN_NO_SPACE);
3687 * If we can't accommodate max_ssize in the current mapping, no go.
3689 if (prev_entry->next->start < addrbos + max_ssize)
3690 return (KERN_NO_SPACE);
3693 * We initially map a stack of only init_ssize. We will grow as
3694 * needed later. Depending on the orientation of the stack (i.e.
3695 * the grow direction) we either map at the top of the range, the
3696 * bottom of the range or in the middle.
3698 * Note: we would normally expect prot and max to be VM_PROT_ALL,
3699 * and cow to be 0. Possibly we should eliminate these as input
3700 * parameters, and just pass these values here in the insert call.
3702 if (orient == MAP_STACK_GROWS_DOWN) {
3703 bot = addrbos + max_ssize - init_ssize;
3704 top = bot + init_ssize;
3707 } else /* if (orient == MAP_STACK_GROWS_UP) */ {
3709 top = bot + init_ssize;
3711 gap_top = addrbos + max_ssize;
3713 rv = vm_map_insert(map, NULL, 0, bot, top, prot, max, cow);
3714 if (rv != KERN_SUCCESS)
3716 new_entry = prev_entry->next;
3717 KASSERT(new_entry->end == top || new_entry->start == bot,
3718 ("Bad entry start/end for new stack entry"));
3719 KASSERT((orient & MAP_STACK_GROWS_DOWN) == 0 ||
3720 (new_entry->eflags & MAP_ENTRY_GROWS_DOWN) != 0,
3721 ("new entry lacks MAP_ENTRY_GROWS_DOWN"));
3722 KASSERT((orient & MAP_STACK_GROWS_UP) == 0 ||
3723 (new_entry->eflags & MAP_ENTRY_GROWS_UP) != 0,
3724 ("new entry lacks MAP_ENTRY_GROWS_UP"));
3725 rv = vm_map_insert(map, NULL, 0, gap_bot, gap_top, VM_PROT_NONE,
3726 VM_PROT_NONE, MAP_CREATE_GUARD | (orient == MAP_STACK_GROWS_DOWN ?
3727 MAP_CREATE_STACK_GAP_DN : MAP_CREATE_STACK_GAP_UP));
3728 if (rv != KERN_SUCCESS)
3729 (void)vm_map_delete(map, bot, top);
3734 * Attempts to grow a vm stack entry. Returns KERN_SUCCESS if we
3735 * successfully grow the stack.
3738 vm_map_growstack(vm_map_t map, vm_offset_t addr, vm_map_entry_t gap_entry)
3740 vm_map_entry_t stack_entry;
3744 vm_offset_t gap_end, gap_start, grow_start;
3745 size_t grow_amount, guard, max_grow;
3746 rlim_t lmemlim, stacklim, vmemlim;
3748 bool gap_deleted, grow_down, is_procstack;
3760 * Disallow stack growth when the access is performed by a
3761 * debugger or AIO daemon. The reason is that the wrong
3762 * resource limits are applied.
3764 if (map != &p->p_vmspace->vm_map || p->p_textvp == NULL)
3765 return (KERN_FAILURE);
3767 MPASS(!map->system_map);
3769 guard = stack_guard_page * PAGE_SIZE;
3770 lmemlim = lim_cur(curthread, RLIMIT_MEMLOCK);
3771 stacklim = lim_cur(curthread, RLIMIT_STACK);
3772 vmemlim = lim_cur(curthread, RLIMIT_VMEM);
3774 /* If addr is not in a hole for a stack grow area, no need to grow. */
3775 if (gap_entry == NULL && !vm_map_lookup_entry(map, addr, &gap_entry))
3776 return (KERN_FAILURE);
3777 if ((gap_entry->eflags & MAP_ENTRY_GUARD) == 0)
3778 return (KERN_SUCCESS);
3779 if ((gap_entry->eflags & MAP_ENTRY_STACK_GAP_DN) != 0) {
3780 stack_entry = gap_entry->next;
3781 if ((stack_entry->eflags & MAP_ENTRY_GROWS_DOWN) == 0 ||
3782 stack_entry->start != gap_entry->end)
3783 return (KERN_FAILURE);
3784 grow_amount = round_page(stack_entry->start - addr);
3786 } else if ((gap_entry->eflags & MAP_ENTRY_STACK_GAP_UP) != 0) {
3787 stack_entry = gap_entry->prev;
3788 if ((stack_entry->eflags & MAP_ENTRY_GROWS_UP) == 0 ||
3789 stack_entry->end != gap_entry->start)
3790 return (KERN_FAILURE);
3791 grow_amount = round_page(addr + 1 - stack_entry->end);
3794 return (KERN_FAILURE);
3796 max_grow = gap_entry->end - gap_entry->start;
3797 if (guard > max_grow)
3798 return (KERN_NO_SPACE);
3800 if (grow_amount > max_grow)
3801 return (KERN_NO_SPACE);
3804 * If this is the main process stack, see if we're over the stack
3807 is_procstack = addr >= (vm_offset_t)vm->vm_maxsaddr &&
3808 addr < (vm_offset_t)p->p_sysent->sv_usrstack;
3809 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim))
3810 return (KERN_NO_SPACE);
3815 if (is_procstack && racct_set(p, RACCT_STACK,
3816 ctob(vm->vm_ssize) + grow_amount)) {
3818 return (KERN_NO_SPACE);
3824 grow_amount = roundup(grow_amount, sgrowsiz);
3825 if (grow_amount > max_grow)
3826 grow_amount = max_grow;
3827 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) {
3828 grow_amount = trunc_page((vm_size_t)stacklim) -
3834 limit = racct_get_available(p, RACCT_STACK);
3836 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > limit))
3837 grow_amount = limit - ctob(vm->vm_ssize);
3840 if (!old_mlock && (map->flags & MAP_WIREFUTURE) != 0) {
3841 if (ptoa(pmap_wired_count(map->pmap)) + grow_amount > lmemlim) {
3848 if (racct_set(p, RACCT_MEMLOCK,
3849 ptoa(pmap_wired_count(map->pmap)) + grow_amount)) {
3859 /* If we would blow our VMEM resource limit, no go */
3860 if (map->size + grow_amount > vmemlim) {
3867 if (racct_set(p, RACCT_VMEM, map->size + grow_amount)) {
3876 if (vm_map_lock_upgrade(map)) {
3878 vm_map_lock_read(map);
3883 grow_start = gap_entry->end - grow_amount;
3884 if (gap_entry->start + grow_amount == gap_entry->end) {
3885 gap_start = gap_entry->start;
3886 gap_end = gap_entry->end;
3887 vm_map_entry_delete(map, gap_entry);
3890 MPASS(gap_entry->start < gap_entry->end - grow_amount);
3891 gap_entry->end -= grow_amount;
3892 vm_map_entry_resize_free(map, gap_entry);
3893 gap_deleted = false;
3895 rv = vm_map_insert(map, NULL, 0, grow_start,
3896 grow_start + grow_amount,
3897 stack_entry->protection, stack_entry->max_protection,
3898 MAP_STACK_GROWS_DOWN);
3899 if (rv != KERN_SUCCESS) {
3901 rv1 = vm_map_insert(map, NULL, 0, gap_start,
3902 gap_end, VM_PROT_NONE, VM_PROT_NONE,
3903 MAP_CREATE_GUARD | MAP_CREATE_STACK_GAP_DN);
3904 MPASS(rv1 == KERN_SUCCESS);
3906 gap_entry->end += grow_amount;
3907 vm_map_entry_resize_free(map, gap_entry);
3911 grow_start = stack_entry->end;
3912 cred = stack_entry->cred;
3913 if (cred == NULL && stack_entry->object.vm_object != NULL)
3914 cred = stack_entry->object.vm_object->cred;
3915 if (cred != NULL && !swap_reserve_by_cred(grow_amount, cred))
3917 /* Grow the underlying object if applicable. */
3918 else if (stack_entry->object.vm_object == NULL ||
3919 vm_object_coalesce(stack_entry->object.vm_object,
3920 stack_entry->offset,
3921 (vm_size_t)(stack_entry->end - stack_entry->start),
3922 (vm_size_t)grow_amount, cred != NULL)) {
3923 if (gap_entry->start + grow_amount == gap_entry->end)
3924 vm_map_entry_delete(map, gap_entry);
3926 gap_entry->start += grow_amount;
3927 stack_entry->end += grow_amount;
3928 map->size += grow_amount;
3929 vm_map_entry_resize_free(map, stack_entry);
3934 if (rv == KERN_SUCCESS && is_procstack)
3935 vm->vm_ssize += btoc(grow_amount);
3938 * Heed the MAP_WIREFUTURE flag if it was set for this process.
3940 if (rv == KERN_SUCCESS && (map->flags & MAP_WIREFUTURE) != 0) {
3942 vm_map_wire(map, grow_start, grow_start + grow_amount,
3943 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
3944 vm_map_lock_read(map);
3946 vm_map_lock_downgrade(map);
3950 if (racct_enable && rv != KERN_SUCCESS) {
3952 error = racct_set(p, RACCT_VMEM, map->size);
3953 KASSERT(error == 0, ("decreasing RACCT_VMEM failed"));
3955 error = racct_set(p, RACCT_MEMLOCK,
3956 ptoa(pmap_wired_count(map->pmap)));
3957 KASSERT(error == 0, ("decreasing RACCT_MEMLOCK failed"));
3959 error = racct_set(p, RACCT_STACK, ctob(vm->vm_ssize));
3960 KASSERT(error == 0, ("decreasing RACCT_STACK failed"));
3969 * Unshare the specified VM space for exec. If other processes are
3970 * mapped to it, then create a new one. The new vmspace is null.
3973 vmspace_exec(struct proc *p, vm_offset_t minuser, vm_offset_t maxuser)
3975 struct vmspace *oldvmspace = p->p_vmspace;
3976 struct vmspace *newvmspace;
3978 KASSERT((curthread->td_pflags & TDP_EXECVMSPC) == 0,
3979 ("vmspace_exec recursed"));
3980 newvmspace = vmspace_alloc(minuser, maxuser, NULL);
3981 if (newvmspace == NULL)
3983 newvmspace->vm_swrss = oldvmspace->vm_swrss;
3985 * This code is written like this for prototype purposes. The
3986 * goal is to avoid running down the vmspace here, but let the
3987 * other process's that are still using the vmspace to finally
3988 * run it down. Even though there is little or no chance of blocking
3989 * here, it is a good idea to keep this form for future mods.
3991 PROC_VMSPACE_LOCK(p);
3992 p->p_vmspace = newvmspace;
3993 PROC_VMSPACE_UNLOCK(p);
3994 if (p == curthread->td_proc)
3995 pmap_activate(curthread);
3996 curthread->td_pflags |= TDP_EXECVMSPC;
4001 * Unshare the specified VM space for forcing COW. This
4002 * is called by rfork, for the (RFMEM|RFPROC) == 0 case.
4005 vmspace_unshare(struct proc *p)
4007 struct vmspace *oldvmspace = p->p_vmspace;
4008 struct vmspace *newvmspace;
4009 vm_ooffset_t fork_charge;
4011 if (oldvmspace->vm_refcnt == 1)
4014 newvmspace = vmspace_fork(oldvmspace, &fork_charge);
4015 if (newvmspace == NULL)
4017 if (!swap_reserve_by_cred(fork_charge, p->p_ucred)) {
4018 vmspace_free(newvmspace);
4021 PROC_VMSPACE_LOCK(p);
4022 p->p_vmspace = newvmspace;
4023 PROC_VMSPACE_UNLOCK(p);
4024 if (p == curthread->td_proc)
4025 pmap_activate(curthread);
4026 vmspace_free(oldvmspace);
4033 * Finds the VM object, offset, and
4034 * protection for a given virtual address in the
4035 * specified map, assuming a page fault of the
4038 * Leaves the map in question locked for read; return
4039 * values are guaranteed until a vm_map_lookup_done
4040 * call is performed. Note that the map argument
4041 * is in/out; the returned map must be used in
4042 * the call to vm_map_lookup_done.
4044 * A handle (out_entry) is returned for use in
4045 * vm_map_lookup_done, to make that fast.
4047 * If a lookup is requested with "write protection"
4048 * specified, the map may be changed to perform virtual
4049 * copying operations, although the data referenced will
4053 vm_map_lookup(vm_map_t *var_map, /* IN/OUT */
4055 vm_prot_t fault_typea,
4056 vm_map_entry_t *out_entry, /* OUT */
4057 vm_object_t *object, /* OUT */
4058 vm_pindex_t *pindex, /* OUT */
4059 vm_prot_t *out_prot, /* OUT */
4060 boolean_t *wired) /* OUT */
4062 vm_map_entry_t entry;
4063 vm_map_t map = *var_map;
4065 vm_prot_t fault_type = fault_typea;
4066 vm_object_t eobject;
4072 vm_map_lock_read(map);
4076 * Lookup the faulting address.
4078 if (!vm_map_lookup_entry(map, vaddr, out_entry)) {
4079 vm_map_unlock_read(map);
4080 return (KERN_INVALID_ADDRESS);
4088 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
4089 vm_map_t old_map = map;
4091 *var_map = map = entry->object.sub_map;
4092 vm_map_unlock_read(old_map);
4097 * Check whether this task is allowed to have this page.
4099 prot = entry->protection;
4100 if ((fault_typea & VM_PROT_FAULT_LOOKUP) != 0) {
4101 fault_typea &= ~VM_PROT_FAULT_LOOKUP;
4102 if (prot == VM_PROT_NONE && map != kernel_map &&
4103 (entry->eflags & MAP_ENTRY_GUARD) != 0 &&
4104 (entry->eflags & (MAP_ENTRY_STACK_GAP_DN |
4105 MAP_ENTRY_STACK_GAP_UP)) != 0 &&
4106 vm_map_growstack(map, vaddr, entry) == KERN_SUCCESS)
4107 goto RetryLookupLocked;
4109 fault_type &= VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE;
4110 if ((fault_type & prot) != fault_type || prot == VM_PROT_NONE) {
4111 vm_map_unlock_read(map);
4112 return (KERN_PROTECTION_FAILURE);
4114 KASSERT((prot & VM_PROT_WRITE) == 0 || (entry->eflags &
4115 (MAP_ENTRY_USER_WIRED | MAP_ENTRY_NEEDS_COPY)) !=
4116 (MAP_ENTRY_USER_WIRED | MAP_ENTRY_NEEDS_COPY),
4117 ("entry %p flags %x", entry, entry->eflags));
4118 if ((fault_typea & VM_PROT_COPY) != 0 &&
4119 (entry->max_protection & VM_PROT_WRITE) == 0 &&
4120 (entry->eflags & MAP_ENTRY_COW) == 0) {
4121 vm_map_unlock_read(map);
4122 return (KERN_PROTECTION_FAILURE);
4126 * If this page is not pageable, we have to get it for all possible
4129 *wired = (entry->wired_count != 0);
4131 fault_type = entry->protection;
4132 size = entry->end - entry->start;
4134 * If the entry was copy-on-write, we either ...
4136 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
4138 * If we want to write the page, we may as well handle that
4139 * now since we've got the map locked.
4141 * If we don't need to write the page, we just demote the
4142 * permissions allowed.
4144 if ((fault_type & VM_PROT_WRITE) != 0 ||
4145 (fault_typea & VM_PROT_COPY) != 0) {
4147 * Make a new object, and place it in the object
4148 * chain. Note that no new references have appeared
4149 * -- one just moved from the map to the new
4152 if (vm_map_lock_upgrade(map))
4155 if (entry->cred == NULL) {
4157 * The debugger owner is charged for
4160 cred = curthread->td_ucred;
4162 if (!swap_reserve_by_cred(size, cred)) {
4165 return (KERN_RESOURCE_SHORTAGE);
4169 vm_object_shadow(&entry->object.vm_object,
4170 &entry->offset, size);
4171 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
4172 eobject = entry->object.vm_object;
4173 if (eobject->cred != NULL) {
4175 * The object was not shadowed.
4177 swap_release_by_cred(size, entry->cred);
4178 crfree(entry->cred);
4180 } else if (entry->cred != NULL) {
4181 VM_OBJECT_WLOCK(eobject);
4182 eobject->cred = entry->cred;
4183 eobject->charge = size;
4184 VM_OBJECT_WUNLOCK(eobject);
4188 vm_map_lock_downgrade(map);
4191 * We're attempting to read a copy-on-write page --
4192 * don't allow writes.
4194 prot &= ~VM_PROT_WRITE;
4199 * Create an object if necessary.
4201 if (entry->object.vm_object == NULL &&
4203 if (vm_map_lock_upgrade(map))
4205 entry->object.vm_object = vm_object_allocate(OBJT_DEFAULT,
4208 if (entry->cred != NULL) {
4209 VM_OBJECT_WLOCK(entry->object.vm_object);
4210 entry->object.vm_object->cred = entry->cred;
4211 entry->object.vm_object->charge = size;
4212 VM_OBJECT_WUNLOCK(entry->object.vm_object);
4215 vm_map_lock_downgrade(map);
4219 * Return the object/offset from this entry. If the entry was
4220 * copy-on-write or empty, it has been fixed up.
4222 *pindex = UOFF_TO_IDX((vaddr - entry->start) + entry->offset);
4223 *object = entry->object.vm_object;
4226 return (KERN_SUCCESS);
4230 * vm_map_lookup_locked:
4232 * Lookup the faulting address. A version of vm_map_lookup that returns
4233 * KERN_FAILURE instead of blocking on map lock or memory allocation.
4236 vm_map_lookup_locked(vm_map_t *var_map, /* IN/OUT */
4238 vm_prot_t fault_typea,
4239 vm_map_entry_t *out_entry, /* OUT */
4240 vm_object_t *object, /* OUT */
4241 vm_pindex_t *pindex, /* OUT */
4242 vm_prot_t *out_prot, /* OUT */
4243 boolean_t *wired) /* OUT */
4245 vm_map_entry_t entry;
4246 vm_map_t map = *var_map;
4248 vm_prot_t fault_type = fault_typea;
4251 * Lookup the faulting address.
4253 if (!vm_map_lookup_entry(map, vaddr, out_entry))
4254 return (KERN_INVALID_ADDRESS);
4259 * Fail if the entry refers to a submap.
4261 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP)
4262 return (KERN_FAILURE);
4265 * Check whether this task is allowed to have this page.
4267 prot = entry->protection;
4268 fault_type &= VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE;
4269 if ((fault_type & prot) != fault_type)
4270 return (KERN_PROTECTION_FAILURE);
4273 * If this page is not pageable, we have to get it for all possible
4276 *wired = (entry->wired_count != 0);
4278 fault_type = entry->protection;
4280 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
4282 * Fail if the entry was copy-on-write for a write fault.
4284 if (fault_type & VM_PROT_WRITE)
4285 return (KERN_FAILURE);
4287 * We're attempting to read a copy-on-write page --
4288 * don't allow writes.
4290 prot &= ~VM_PROT_WRITE;
4294 * Fail if an object should be created.
4296 if (entry->object.vm_object == NULL && !map->system_map)
4297 return (KERN_FAILURE);
4300 * Return the object/offset from this entry. If the entry was
4301 * copy-on-write or empty, it has been fixed up.
4303 *pindex = UOFF_TO_IDX((vaddr - entry->start) + entry->offset);
4304 *object = entry->object.vm_object;
4307 return (KERN_SUCCESS);
4311 * vm_map_lookup_done:
4313 * Releases locks acquired by a vm_map_lookup
4314 * (according to the handle returned by that lookup).
4317 vm_map_lookup_done(vm_map_t map, vm_map_entry_t entry)
4320 * Unlock the main-level map
4322 vm_map_unlock_read(map);
4326 vm_map_max_KBI(const struct vm_map *map)
4329 return (map->max_offset);
4333 vm_map_min_KBI(const struct vm_map *map)
4336 return (map->min_offset);
4340 vm_map_pmap_KBI(vm_map_t map)
4346 #include "opt_ddb.h"
4348 #include <sys/kernel.h>
4350 #include <ddb/ddb.h>
4353 vm_map_print(vm_map_t map)
4355 vm_map_entry_t entry;
4357 db_iprintf("Task map %p: pmap=%p, nentries=%d, version=%u\n",
4359 (void *)map->pmap, map->nentries, map->timestamp);
4362 for (entry = map->header.next; entry != &map->header;
4363 entry = entry->next) {
4364 db_iprintf("map entry %p: start=%p, end=%p, eflags=%#x, \n",
4365 (void *)entry, (void *)entry->start, (void *)entry->end,
4368 static char *inheritance_name[4] =
4369 {"share", "copy", "none", "donate_copy"};
4371 db_iprintf(" prot=%x/%x/%s",
4373 entry->max_protection,
4374 inheritance_name[(int)(unsigned char)entry->inheritance]);
4375 if (entry->wired_count != 0)
4376 db_printf(", wired");
4378 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
4379 db_printf(", share=%p, offset=0x%jx\n",
4380 (void *)entry->object.sub_map,
4381 (uintmax_t)entry->offset);
4382 if ((entry->prev == &map->header) ||
4383 (entry->prev->object.sub_map !=
4384 entry->object.sub_map)) {
4386 vm_map_print((vm_map_t)entry->object.sub_map);
4390 if (entry->cred != NULL)
4391 db_printf(", ruid %d", entry->cred->cr_ruid);
4392 db_printf(", object=%p, offset=0x%jx",
4393 (void *)entry->object.vm_object,
4394 (uintmax_t)entry->offset);
4395 if (entry->object.vm_object && entry->object.vm_object->cred)
4396 db_printf(", obj ruid %d charge %jx",
4397 entry->object.vm_object->cred->cr_ruid,
4398 (uintmax_t)entry->object.vm_object->charge);
4399 if (entry->eflags & MAP_ENTRY_COW)
4400 db_printf(", copy (%s)",
4401 (entry->eflags & MAP_ENTRY_NEEDS_COPY) ? "needed" : "done");
4404 if ((entry->prev == &map->header) ||
4405 (entry->prev->object.vm_object !=
4406 entry->object.vm_object)) {
4408 vm_object_print((db_expr_t)(intptr_t)
4409 entry->object.vm_object,
4418 DB_SHOW_COMMAND(map, map)
4422 db_printf("usage: show map <addr>\n");
4425 vm_map_print((vm_map_t)addr);
4428 DB_SHOW_COMMAND(procvm, procvm)
4433 p = db_lookup_proc(addr);
4438 db_printf("p = %p, vmspace = %p, map = %p, pmap = %p\n",
4439 (void *)p, (void *)p->p_vmspace, (void *)&p->p_vmspace->vm_map,
4440 (void *)vmspace_pmap(p->p_vmspace));
4442 vm_map_print((vm_map_t)&p->p_vmspace->vm_map);