2 * Copyright (c) 1991, 1993
3 * The Regents of the University of California. All rights reserved.
5 * This code is derived from software contributed to Berkeley by
6 * The Mach Operating System project at Carnegie-Mellon University.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 4. Neither the name of the University nor the names of its contributors
17 * may be used to endorse or promote products derived from this software
18 * without specific prior written permission.
20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * from: @(#)vm_map.c 8.3 (Berkeley) 1/12/94
35 * Copyright (c) 1987, 1990 Carnegie-Mellon University.
36 * All rights reserved.
38 * Authors: Avadis Tevanian, Jr., Michael Wayne Young
40 * Permission to use, copy, modify and distribute this software and
41 * its documentation is hereby granted, provided that both the copyright
42 * notice and this permission notice appear in all copies of the
43 * software, derivative works or modified versions, and any portions
44 * thereof, and that both notices appear in supporting documentation.
46 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
47 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
48 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
50 * Carnegie Mellon requests users of this software to return to
52 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
53 * School of Computer Science
54 * Carnegie Mellon University
55 * Pittsburgh PA 15213-3890
57 * any improvements or extensions that they make and grant Carnegie the
58 * rights to redistribute these changes.
62 * Virtual memory mapping module.
65 #include <sys/cdefs.h>
66 __FBSDID("$FreeBSD$");
68 #include <sys/param.h>
69 #include <sys/systm.h>
70 #include <sys/kernel.h>
73 #include <sys/mutex.h>
75 #include <sys/vmmeter.h>
77 #include <sys/vnode.h>
78 #include <sys/racct.h>
79 #include <sys/resourcevar.h>
80 #include <sys/rwlock.h>
82 #include <sys/sysctl.h>
83 #include <sys/sysent.h>
87 #include <vm/vm_param.h>
89 #include <vm/vm_map.h>
90 #include <vm/vm_page.h>
91 #include <vm/vm_object.h>
92 #include <vm/vm_pager.h>
93 #include <vm/vm_kern.h>
94 #include <vm/vm_extern.h>
95 #include <vm/vnode_pager.h>
96 #include <vm/swap_pager.h>
100 * Virtual memory maps provide for the mapping, protection,
101 * and sharing of virtual memory objects. In addition,
102 * this module provides for an efficient virtual copy of
103 * memory from one map to another.
105 * Synchronization is required prior to most operations.
107 * Maps consist of an ordered doubly-linked list of simple
108 * entries; a self-adjusting binary search tree of these
109 * entries is used to speed up lookups.
111 * Since portions of maps are specified by start/end addresses,
112 * which may not align with existing map entries, all
113 * routines merely "clip" entries to these start/end values.
114 * [That is, an entry is split into two, bordering at a
115 * start or end value.] Note that these clippings may not
116 * always be necessary (as the two resulting entries are then
117 * not changed); however, the clipping is done for convenience.
119 * As mentioned above, virtual copy operations are performed
120 * by copying VM object references from one map to
121 * another, and then marking both regions as copy-on-write.
124 static struct mtx map_sleep_mtx;
125 static uma_zone_t mapentzone;
126 static uma_zone_t kmapentzone;
127 static uma_zone_t mapzone;
128 static uma_zone_t vmspace_zone;
129 static int vmspace_zinit(void *mem, int size, int flags);
130 static void vmspace_zfini(void *mem, int size);
131 static int vm_map_zinit(void *mem, int ize, int flags);
132 static void vm_map_zfini(void *mem, int size);
133 static void _vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min,
135 static void vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map);
136 static void vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry);
138 static void vm_map_zdtor(void *mem, int size, void *arg);
139 static void vmspace_zdtor(void *mem, int size, void *arg);
142 #define ENTRY_CHARGED(e) ((e)->cred != NULL || \
143 ((e)->object.vm_object != NULL && (e)->object.vm_object->cred != NULL && \
144 !((e)->eflags & MAP_ENTRY_NEEDS_COPY)))
147 * PROC_VMSPACE_{UN,}LOCK() can be a noop as long as vmspaces are type
150 #define PROC_VMSPACE_LOCK(p) do { } while (0)
151 #define PROC_VMSPACE_UNLOCK(p) do { } while (0)
154 * VM_MAP_RANGE_CHECK: [ internal use only ]
156 * Asserts that the starting and ending region
157 * addresses fall within the valid range of the map.
159 #define VM_MAP_RANGE_CHECK(map, start, end) \
161 if (start < vm_map_min(map)) \
162 start = vm_map_min(map); \
163 if (end > vm_map_max(map)) \
164 end = vm_map_max(map); \
172 * Initialize the vm_map module. Must be called before
173 * any other vm_map routines.
175 * Map and entry structures are allocated from the general
176 * purpose memory pool with some exceptions:
178 * - The kernel map and kmem submap are allocated statically.
179 * - Kernel map entries are allocated out of a static pool.
181 * These restrictions are necessary since malloc() uses the
182 * maps and requires map entries.
188 mtx_init(&map_sleep_mtx, "vm map sleep mutex", NULL, MTX_DEF);
189 mapzone = uma_zcreate("MAP", sizeof(struct vm_map), NULL,
195 vm_map_zinit, vm_map_zfini, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
196 uma_prealloc(mapzone, MAX_KMAP);
197 kmapentzone = uma_zcreate("KMAP ENTRY", sizeof(struct vm_map_entry),
198 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR,
199 UMA_ZONE_MTXCLASS | UMA_ZONE_VM);
200 uma_prealloc(kmapentzone, MAX_KMAPENT);
201 mapentzone = uma_zcreate("MAP ENTRY", sizeof(struct vm_map_entry),
202 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
206 vmspace_zfini(void *mem, int size)
210 vm = (struct vmspace *)mem;
211 vm_map_zfini(&vm->vm_map, sizeof(vm->vm_map));
215 vmspace_zinit(void *mem, int size, int flags)
219 vm = (struct vmspace *)mem;
221 vm->vm_map.pmap = NULL;
222 (void)vm_map_zinit(&vm->vm_map, sizeof(vm->vm_map), flags);
227 vm_map_zfini(void *mem, int size)
232 mtx_destroy(&map->system_mtx);
233 sx_destroy(&map->lock);
237 vm_map_zinit(void *mem, int size, int flags)
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.
279 vmspace_alloc(min, max)
280 vm_offset_t min, max;
284 vm = uma_zalloc(vmspace_zone, M_WAITOK);
285 if (vm->vm_map.pmap == NULL && !pmap_pinit(vmspace_pmap(vm))) {
286 uma_zfree(vmspace_zone, vm);
289 CTR1(KTR_VM, "vmspace_alloc: %p", vm);
290 _vm_map_init(&vm->vm_map, vmspace_pmap(vm), min, max);
306 uma_zone_reserve_kva(kmapentzone, lmin(cnt.v_page_count,
307 (VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS) / PAGE_SIZE) / 8 +
308 maxproc * 2 + maxfiles);
309 vmspace_zone = uma_zcreate("VMSPACE", sizeof(struct vmspace), NULL,
315 vmspace_zinit, vmspace_zfini, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
319 vmspace_container_reset(struct proc *p)
324 racct_set(p, RACCT_DATA, 0);
325 racct_set(p, RACCT_STACK, 0);
326 racct_set(p, RACCT_RSS, 0);
327 racct_set(p, RACCT_MEMLOCK, 0);
328 racct_set(p, RACCT_VMEM, 0);
334 vmspace_dofree(struct vmspace *vm)
337 CTR1(KTR_VM, "vmspace_free: %p", vm);
340 * Make sure any SysV shm is freed, it might not have been in
346 * Lock the map, to wait out all other references to it.
347 * Delete all of the mappings and pages they hold, then call
348 * the pmap module to reclaim anything left.
350 (void)vm_map_remove(&vm->vm_map, vm->vm_map.min_offset,
351 vm->vm_map.max_offset);
353 pmap_release(vmspace_pmap(vm));
354 vm->vm_map.pmap = NULL;
355 uma_zfree(vmspace_zone, vm);
359 vmspace_free(struct vmspace *vm)
362 if (vm->vm_refcnt == 0)
363 panic("vmspace_free: attempt to free already freed vmspace");
365 if (atomic_fetchadd_int(&vm->vm_refcnt, -1) == 1)
370 vmspace_exitfree(struct proc *p)
374 PROC_VMSPACE_LOCK(p);
377 PROC_VMSPACE_UNLOCK(p);
378 KASSERT(vm == &vmspace0, ("vmspace_exitfree: wrong vmspace"));
383 vmspace_exit(struct thread *td)
390 * Release user portion of address space.
391 * This releases references to vnodes,
392 * which could cause I/O if the file has been unlinked.
393 * Need to do this early enough that we can still sleep.
395 * The last exiting process to reach this point releases as
396 * much of the environment as it can. vmspace_dofree() is the
397 * slower fallback in case another process had a temporary
398 * reference to the vmspace.
403 atomic_add_int(&vmspace0.vm_refcnt, 1);
405 refcnt = vm->vm_refcnt;
406 if (refcnt > 1 && p->p_vmspace != &vmspace0) {
407 /* Switch now since other proc might free vmspace */
408 PROC_VMSPACE_LOCK(p);
409 p->p_vmspace = &vmspace0;
410 PROC_VMSPACE_UNLOCK(p);
413 } while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt - 1));
415 if (p->p_vmspace != vm) {
416 /* vmspace not yet freed, switch back */
417 PROC_VMSPACE_LOCK(p);
419 PROC_VMSPACE_UNLOCK(p);
422 pmap_remove_pages(vmspace_pmap(vm));
423 /* Switch now since this proc will free vmspace */
424 PROC_VMSPACE_LOCK(p);
425 p->p_vmspace = &vmspace0;
426 PROC_VMSPACE_UNLOCK(p);
430 vmspace_container_reset(p);
433 /* Acquire reference to vmspace owned by another process. */
436 vmspace_acquire_ref(struct proc *p)
441 PROC_VMSPACE_LOCK(p);
444 PROC_VMSPACE_UNLOCK(p);
448 refcnt = vm->vm_refcnt;
449 if (refcnt <= 0) { /* Avoid 0->1 transition */
450 PROC_VMSPACE_UNLOCK(p);
453 } while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt + 1));
454 if (vm != p->p_vmspace) {
455 PROC_VMSPACE_UNLOCK(p);
459 PROC_VMSPACE_UNLOCK(p);
464 _vm_map_lock(vm_map_t map, const char *file, int line)
468 mtx_lock_flags_(&map->system_mtx, 0, file, line);
470 sx_xlock_(&map->lock, file, line);
475 vm_map_process_deferred(void)
478 vm_map_entry_t entry, next;
482 entry = td->td_map_def_user;
483 td->td_map_def_user = NULL;
484 while (entry != NULL) {
486 if ((entry->eflags & MAP_ENTRY_VN_WRITECNT) != 0) {
488 * Decrement the object's writemappings and
489 * possibly the vnode's v_writecount.
491 KASSERT((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0,
492 ("Submap with writecount"));
493 object = entry->object.vm_object;
494 KASSERT(object != NULL, ("No object for writecount"));
495 vnode_pager_release_writecount(object, entry->start,
498 vm_map_entry_deallocate(entry, FALSE);
504 _vm_map_unlock(vm_map_t map, const char *file, int line)
508 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
510 sx_xunlock_(&map->lock, file, line);
511 vm_map_process_deferred();
516 _vm_map_lock_read(vm_map_t map, const char *file, int line)
520 mtx_lock_flags_(&map->system_mtx, 0, file, line);
522 sx_slock_(&map->lock, file, line);
526 _vm_map_unlock_read(vm_map_t map, const char *file, int line)
530 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
532 sx_sunlock_(&map->lock, file, line);
533 vm_map_process_deferred();
538 _vm_map_trylock(vm_map_t map, const char *file, int line)
542 error = map->system_map ?
543 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) :
544 !sx_try_xlock_(&map->lock, file, line);
551 _vm_map_trylock_read(vm_map_t map, const char *file, int line)
555 error = map->system_map ?
556 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) :
557 !sx_try_slock_(&map->lock, file, line);
562 * _vm_map_lock_upgrade: [ internal use only ]
564 * Tries to upgrade a read (shared) lock on the specified map to a write
565 * (exclusive) lock. Returns the value "0" if the upgrade succeeds and a
566 * non-zero value if the upgrade fails. If the upgrade fails, the map is
567 * returned without a read or write lock held.
569 * Requires that the map be read locked.
572 _vm_map_lock_upgrade(vm_map_t map, const char *file, int line)
574 unsigned int last_timestamp;
576 if (map->system_map) {
577 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
579 if (!sx_try_upgrade_(&map->lock, file, line)) {
580 last_timestamp = map->timestamp;
581 sx_sunlock_(&map->lock, file, line);
582 vm_map_process_deferred();
584 * If the map's timestamp does not change while the
585 * map is unlocked, then the upgrade succeeds.
587 sx_xlock_(&map->lock, file, line);
588 if (last_timestamp != map->timestamp) {
589 sx_xunlock_(&map->lock, file, line);
599 _vm_map_lock_downgrade(vm_map_t map, const char *file, int line)
602 if (map->system_map) {
603 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
605 sx_downgrade_(&map->lock, file, line);
611 * Returns a non-zero value if the caller holds a write (exclusive) lock
612 * on the specified map and the value "0" otherwise.
615 vm_map_locked(vm_map_t map)
619 return (mtx_owned(&map->system_mtx));
621 return (sx_xlocked(&map->lock));
626 _vm_map_assert_locked(vm_map_t map, const char *file, int line)
630 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
632 sx_assert_(&map->lock, SA_XLOCKED, file, line);
635 #define VM_MAP_ASSERT_LOCKED(map) \
636 _vm_map_assert_locked(map, LOCK_FILE, LOCK_LINE)
638 #define VM_MAP_ASSERT_LOCKED(map)
642 * _vm_map_unlock_and_wait:
644 * Atomically releases the lock on the specified map and puts the calling
645 * thread to sleep. The calling thread will remain asleep until either
646 * vm_map_wakeup() is performed on the map or the specified timeout is
649 * WARNING! This function does not perform deferred deallocations of
650 * objects and map entries. Therefore, the calling thread is expected to
651 * reacquire the map lock after reawakening and later perform an ordinary
652 * unlock operation, such as vm_map_unlock(), before completing its
653 * operation on the map.
656 _vm_map_unlock_and_wait(vm_map_t map, int timo, const char *file, int line)
659 mtx_lock(&map_sleep_mtx);
661 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
663 sx_xunlock_(&map->lock, file, line);
664 return (msleep(&map->root, &map_sleep_mtx, PDROP | PVM, "vmmaps",
671 * Awaken any threads that have slept on the map using
672 * vm_map_unlock_and_wait().
675 vm_map_wakeup(vm_map_t map)
679 * Acquire and release map_sleep_mtx to prevent a wakeup()
680 * from being performed (and lost) between the map unlock
681 * and the msleep() in _vm_map_unlock_and_wait().
683 mtx_lock(&map_sleep_mtx);
684 mtx_unlock(&map_sleep_mtx);
689 vm_map_busy(vm_map_t map)
692 VM_MAP_ASSERT_LOCKED(map);
697 vm_map_unbusy(vm_map_t map)
700 VM_MAP_ASSERT_LOCKED(map);
701 KASSERT(map->busy, ("vm_map_unbusy: not busy"));
702 if (--map->busy == 0 && (map->flags & MAP_BUSY_WAKEUP)) {
703 vm_map_modflags(map, 0, MAP_BUSY_WAKEUP);
709 vm_map_wait_busy(vm_map_t map)
712 VM_MAP_ASSERT_LOCKED(map);
714 vm_map_modflags(map, MAP_BUSY_WAKEUP, 0);
716 msleep(&map->busy, &map->system_mtx, 0, "mbusy", 0);
718 sx_sleep(&map->busy, &map->lock, 0, "mbusy", 0);
724 vmspace_resident_count(struct vmspace *vmspace)
726 return pmap_resident_count(vmspace_pmap(vmspace));
732 * Creates and returns a new empty VM map with
733 * the given physical map structure, and having
734 * the given lower and upper address bounds.
737 vm_map_create(pmap_t pmap, vm_offset_t min, vm_offset_t max)
741 result = uma_zalloc(mapzone, M_WAITOK);
742 CTR1(KTR_VM, "vm_map_create: %p", result);
743 _vm_map_init(result, pmap, min, max);
748 * Initialize an existing vm_map structure
749 * such as that in the vmspace structure.
752 _vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max)
755 map->header.next = map->header.prev = &map->header;
756 map->needs_wakeup = FALSE;
759 map->min_offset = min;
760 map->max_offset = max;
768 vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max)
771 _vm_map_init(map, pmap, min, max);
772 mtx_init(&map->system_mtx, "system map", NULL, MTX_DEF | MTX_DUPOK);
773 sx_init(&map->lock, "user map");
777 * vm_map_entry_dispose: [ internal use only ]
779 * Inverse of vm_map_entry_create.
782 vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry)
784 uma_zfree(map->system_map ? kmapentzone : mapentzone, entry);
788 * vm_map_entry_create: [ internal use only ]
790 * Allocates a VM map entry for insertion.
791 * No entry fields are filled in.
793 static vm_map_entry_t
794 vm_map_entry_create(vm_map_t map)
796 vm_map_entry_t new_entry;
799 new_entry = uma_zalloc(kmapentzone, M_NOWAIT);
801 new_entry = uma_zalloc(mapentzone, M_WAITOK);
802 if (new_entry == NULL)
803 panic("vm_map_entry_create: kernel resources exhausted");
808 * vm_map_entry_set_behavior:
810 * Set the expected access behavior, either normal, random, or
814 vm_map_entry_set_behavior(vm_map_entry_t entry, u_char behavior)
816 entry->eflags = (entry->eflags & ~MAP_ENTRY_BEHAV_MASK) |
817 (behavior & MAP_ENTRY_BEHAV_MASK);
821 * vm_map_entry_set_max_free:
823 * Set the max_free field in a vm_map_entry.
826 vm_map_entry_set_max_free(vm_map_entry_t entry)
829 entry->max_free = entry->adj_free;
830 if (entry->left != NULL && entry->left->max_free > entry->max_free)
831 entry->max_free = entry->left->max_free;
832 if (entry->right != NULL && entry->right->max_free > entry->max_free)
833 entry->max_free = entry->right->max_free;
837 * vm_map_entry_splay:
839 * The Sleator and Tarjan top-down splay algorithm with the
840 * following variation. Max_free must be computed bottom-up, so
841 * on the downward pass, maintain the left and right spines in
842 * reverse order. Then, make a second pass up each side to fix
843 * the pointers and compute max_free. The time bound is O(log n)
846 * The new root is the vm_map_entry containing "addr", or else an
847 * adjacent entry (lower or higher) if addr is not in the tree.
849 * The map must be locked, and leaves it so.
851 * Returns: the new root.
853 static vm_map_entry_t
854 vm_map_entry_splay(vm_offset_t addr, vm_map_entry_t root)
856 vm_map_entry_t llist, rlist;
857 vm_map_entry_t ltree, rtree;
860 /* Special case of empty tree. */
865 * Pass One: Splay down the tree until we find addr or a NULL
866 * pointer where addr would go. llist and rlist are the two
867 * sides in reverse order (bottom-up), with llist linked by
868 * the right pointer and rlist linked by the left pointer in
869 * the vm_map_entry. Wait until Pass Two to set max_free on
875 /* root is never NULL in here. */
876 if (addr < root->start) {
880 if (addr < y->start && y->left != NULL) {
881 /* Rotate right and put y on rlist. */
882 root->left = y->right;
884 vm_map_entry_set_max_free(root);
889 /* Put root on rlist. */
894 } else if (addr >= root->end) {
898 if (addr >= y->end && y->right != NULL) {
899 /* Rotate left and put y on llist. */
900 root->right = y->left;
902 vm_map_entry_set_max_free(root);
907 /* Put root on llist. */
917 * Pass Two: Walk back up the two spines, flip the pointers
918 * and set max_free. The subtrees of the root go at the
919 * bottom of llist and rlist.
922 while (llist != NULL) {
924 llist->right = ltree;
925 vm_map_entry_set_max_free(llist);
930 while (rlist != NULL) {
933 vm_map_entry_set_max_free(rlist);
939 * Final assembly: add ltree and rtree as subtrees of root.
943 vm_map_entry_set_max_free(root);
949 * vm_map_entry_{un,}link:
951 * Insert/remove entries from maps.
954 vm_map_entry_link(vm_map_t map,
955 vm_map_entry_t after_where,
956 vm_map_entry_t entry)
960 "vm_map_entry_link: map %p, nentries %d, entry %p, after %p", map,
961 map->nentries, entry, after_where);
962 VM_MAP_ASSERT_LOCKED(map);
964 entry->prev = after_where;
965 entry->next = after_where->next;
966 entry->next->prev = entry;
967 after_where->next = entry;
969 if (after_where != &map->header) {
970 if (after_where != map->root)
971 vm_map_entry_splay(after_where->start, map->root);
972 entry->right = after_where->right;
973 entry->left = after_where;
974 after_where->right = NULL;
975 after_where->adj_free = entry->start - after_where->end;
976 vm_map_entry_set_max_free(after_where);
978 entry->right = map->root;
981 entry->adj_free = (entry->next == &map->header ? map->max_offset :
982 entry->next->start) - entry->end;
983 vm_map_entry_set_max_free(entry);
988 vm_map_entry_unlink(vm_map_t map,
989 vm_map_entry_t entry)
991 vm_map_entry_t next, prev, root;
993 VM_MAP_ASSERT_LOCKED(map);
994 if (entry != map->root)
995 vm_map_entry_splay(entry->start, map->root);
996 if (entry->left == NULL)
999 root = vm_map_entry_splay(entry->start, entry->left);
1000 root->right = entry->right;
1001 root->adj_free = (entry->next == &map->header ? map->max_offset :
1002 entry->next->start) - root->end;
1003 vm_map_entry_set_max_free(root);
1012 CTR3(KTR_VM, "vm_map_entry_unlink: map %p, nentries %d, entry %p", map,
1013 map->nentries, entry);
1017 * vm_map_entry_resize_free:
1019 * Recompute the amount of free space following a vm_map_entry
1020 * and propagate that value up the tree. Call this function after
1021 * resizing a map entry in-place, that is, without a call to
1022 * vm_map_entry_link() or _unlink().
1024 * The map must be locked, and leaves it so.
1027 vm_map_entry_resize_free(vm_map_t map, vm_map_entry_t entry)
1031 * Using splay trees without parent pointers, propagating
1032 * max_free up the tree is done by moving the entry to the
1033 * root and making the change there.
1035 if (entry != map->root)
1036 map->root = vm_map_entry_splay(entry->start, map->root);
1038 entry->adj_free = (entry->next == &map->header ? map->max_offset :
1039 entry->next->start) - entry->end;
1040 vm_map_entry_set_max_free(entry);
1044 * vm_map_lookup_entry: [ internal use only ]
1046 * Finds the map entry containing (or
1047 * immediately preceding) the specified address
1048 * in the given map; the entry is returned
1049 * in the "entry" parameter. The boolean
1050 * result indicates whether the address is
1051 * actually contained in the map.
1054 vm_map_lookup_entry(
1056 vm_offset_t address,
1057 vm_map_entry_t *entry) /* OUT */
1063 * If the map is empty, then the map entry immediately preceding
1064 * "address" is the map's header.
1068 *entry = &map->header;
1069 else if (address >= cur->start && cur->end > address) {
1072 } else if ((locked = vm_map_locked(map)) ||
1073 sx_try_upgrade(&map->lock)) {
1075 * Splay requires a write lock on the map. However, it only
1076 * restructures the binary search tree; it does not otherwise
1077 * change the map. Thus, the map's timestamp need not change
1078 * on a temporary upgrade.
1080 map->root = cur = vm_map_entry_splay(address, cur);
1082 sx_downgrade(&map->lock);
1085 * If "address" is contained within a map entry, the new root
1086 * is that map entry. Otherwise, the new root is a map entry
1087 * immediately before or after "address".
1089 if (address >= cur->start) {
1091 if (cur->end > address)
1097 * Since the map is only locked for read access, perform a
1098 * standard binary search tree lookup for "address".
1101 if (address < cur->start) {
1102 if (cur->left == NULL) {
1107 } else if (cur->end > address) {
1111 if (cur->right == NULL) {
1124 * Inserts the given whole VM object into the target
1125 * map at the specified address range. The object's
1126 * size should match that of the address range.
1128 * Requires that the map be locked, and leaves it so.
1130 * If object is non-NULL, ref count must be bumped by caller
1131 * prior to making call to account for the new entry.
1134 vm_map_insert(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1135 vm_offset_t start, vm_offset_t end, vm_prot_t prot, vm_prot_t max,
1138 vm_map_entry_t new_entry;
1139 vm_map_entry_t prev_entry;
1140 vm_map_entry_t temp_entry;
1141 vm_eflags_t protoeflags;
1143 vm_inherit_t inheritance;
1144 boolean_t charge_prev_obj;
1146 VM_MAP_ASSERT_LOCKED(map);
1149 * Check that the start and end points are not bogus.
1151 if ((start < map->min_offset) || (end > map->max_offset) ||
1153 return (KERN_INVALID_ADDRESS);
1156 * Find the entry prior to the proposed starting address; if it's part
1157 * of an existing entry, this range is bogus.
1159 if (vm_map_lookup_entry(map, start, &temp_entry))
1160 return (KERN_NO_SPACE);
1162 prev_entry = temp_entry;
1165 * Assert that the next entry doesn't overlap the end point.
1167 if ((prev_entry->next != &map->header) &&
1168 (prev_entry->next->start < end))
1169 return (KERN_NO_SPACE);
1172 charge_prev_obj = FALSE;
1174 if (cow & MAP_COPY_ON_WRITE)
1175 protoeflags |= MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY;
1177 if (cow & MAP_NOFAULT) {
1178 protoeflags |= MAP_ENTRY_NOFAULT;
1180 KASSERT(object == NULL,
1181 ("vm_map_insert: paradoxical MAP_NOFAULT request"));
1183 if (cow & MAP_DISABLE_SYNCER)
1184 protoeflags |= MAP_ENTRY_NOSYNC;
1185 if (cow & MAP_DISABLE_COREDUMP)
1186 protoeflags |= MAP_ENTRY_NOCOREDUMP;
1187 if (cow & MAP_VN_WRITECOUNT)
1188 protoeflags |= MAP_ENTRY_VN_WRITECNT;
1189 if (cow & MAP_INHERIT_SHARE)
1190 inheritance = VM_INHERIT_SHARE;
1192 inheritance = VM_INHERIT_DEFAULT;
1195 KASSERT((object != kmem_object && object != kernel_object) ||
1196 ((object == kmem_object || object == kernel_object) &&
1197 !(protoeflags & MAP_ENTRY_NEEDS_COPY)),
1198 ("kmem or kernel object and cow"));
1199 if (cow & (MAP_ACC_NO_CHARGE | MAP_NOFAULT))
1201 if ((cow & MAP_ACC_CHARGED) || ((prot & VM_PROT_WRITE) &&
1202 ((protoeflags & MAP_ENTRY_NEEDS_COPY) || object == NULL))) {
1203 if (!(cow & MAP_ACC_CHARGED) && !swap_reserve(end - start))
1204 return (KERN_RESOURCE_SHORTAGE);
1205 KASSERT(object == NULL || (protoeflags & MAP_ENTRY_NEEDS_COPY) ||
1206 object->cred == NULL,
1207 ("OVERCOMMIT: vm_map_insert o %p", object));
1208 cred = curthread->td_ucred;
1210 if (object == NULL && !(protoeflags & MAP_ENTRY_NEEDS_COPY))
1211 charge_prev_obj = TRUE;
1215 /* Expand the kernel pmap, if necessary. */
1216 if (map == kernel_map && end > kernel_vm_end)
1217 pmap_growkernel(end);
1218 if (object != NULL) {
1220 * OBJ_ONEMAPPING must be cleared unless this mapping
1221 * is trivially proven to be the only mapping for any
1222 * of the object's pages. (Object granularity
1223 * reference counting is insufficient to recognize
1224 * aliases with precision.)
1226 VM_OBJECT_WLOCK(object);
1227 if (object->ref_count > 1 || object->shadow_count != 0)
1228 vm_object_clear_flag(object, OBJ_ONEMAPPING);
1229 VM_OBJECT_WUNLOCK(object);
1231 else if ((prev_entry != &map->header) &&
1232 (prev_entry->eflags == protoeflags) &&
1233 (prev_entry->end == start) &&
1234 (prev_entry->wired_count == 0) &&
1235 (prev_entry->cred == cred ||
1236 (prev_entry->object.vm_object != NULL &&
1237 (prev_entry->object.vm_object->cred == cred))) &&
1238 vm_object_coalesce(prev_entry->object.vm_object,
1240 (vm_size_t)(prev_entry->end - prev_entry->start),
1241 (vm_size_t)(end - prev_entry->end), charge_prev_obj)) {
1243 * We were able to extend the object. Determine if we
1244 * can extend the previous map entry to include the
1245 * new range as well.
1247 if ((prev_entry->inheritance == inheritance) &&
1248 (prev_entry->protection == prot) &&
1249 (prev_entry->max_protection == max)) {
1250 map->size += (end - prev_entry->end);
1251 prev_entry->end = end;
1252 vm_map_entry_resize_free(map, prev_entry);
1253 vm_map_simplify_entry(map, prev_entry);
1256 return (KERN_SUCCESS);
1260 * If we can extend the object but cannot extend the
1261 * map entry, we have to create a new map entry. We
1262 * must bump the ref count on the extended object to
1263 * account for it. object may be NULL.
1265 object = prev_entry->object.vm_object;
1266 offset = prev_entry->offset +
1267 (prev_entry->end - prev_entry->start);
1268 vm_object_reference(object);
1269 if (cred != NULL && object != NULL && object->cred != NULL &&
1270 !(prev_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
1271 /* Object already accounts for this uid. */
1278 * NOTE: if conditionals fail, object can be NULL here. This occurs
1279 * in things like the buffer map where we manage kva but do not manage
1284 * Create a new entry
1286 new_entry = vm_map_entry_create(map);
1287 new_entry->start = start;
1288 new_entry->end = end;
1289 new_entry->cred = NULL;
1291 new_entry->eflags = protoeflags;
1292 new_entry->object.vm_object = object;
1293 new_entry->offset = offset;
1294 new_entry->avail_ssize = 0;
1296 new_entry->inheritance = inheritance;
1297 new_entry->protection = prot;
1298 new_entry->max_protection = max;
1299 new_entry->wired_count = 0;
1300 new_entry->read_ahead = VM_FAULT_READ_AHEAD_INIT;
1301 new_entry->next_read = OFF_TO_IDX(offset);
1303 KASSERT(cred == NULL || !ENTRY_CHARGED(new_entry),
1304 ("OVERCOMMIT: vm_map_insert leaks vm_map %p", new_entry));
1305 new_entry->cred = cred;
1308 * Insert the new entry into the list
1310 vm_map_entry_link(map, prev_entry, new_entry);
1311 map->size += new_entry->end - new_entry->start;
1314 * It may be possible to merge the new entry with the next and/or
1315 * previous entries. However, due to MAP_STACK_* being a hack, a
1316 * panic can result from merging such entries.
1318 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0)
1319 vm_map_simplify_entry(map, new_entry);
1321 if (cow & (MAP_PREFAULT|MAP_PREFAULT_PARTIAL)) {
1322 vm_map_pmap_enter(map, start, prot,
1323 object, OFF_TO_IDX(offset), end - start,
1324 cow & MAP_PREFAULT_PARTIAL);
1327 return (KERN_SUCCESS);
1333 * Find the first fit (lowest VM address) for "length" free bytes
1334 * beginning at address >= start in the given map.
1336 * In a vm_map_entry, "adj_free" is the amount of free space
1337 * adjacent (higher address) to this entry, and "max_free" is the
1338 * maximum amount of contiguous free space in its subtree. This
1339 * allows finding a free region in one path down the tree, so
1340 * O(log n) amortized with splay trees.
1342 * The map must be locked, and leaves it so.
1344 * Returns: 0 on success, and starting address in *addr,
1345 * 1 if insufficient space.
1348 vm_map_findspace(vm_map_t map, vm_offset_t start, vm_size_t length,
1349 vm_offset_t *addr) /* OUT */
1351 vm_map_entry_t entry;
1355 * Request must fit within min/max VM address and must avoid
1358 if (start < map->min_offset)
1359 start = map->min_offset;
1360 if (start + length > map->max_offset || start + length < start)
1363 /* Empty tree means wide open address space. */
1364 if (map->root == NULL) {
1370 * After splay, if start comes before root node, then there
1371 * must be a gap from start to the root.
1373 map->root = vm_map_entry_splay(start, map->root);
1374 if (start + length <= map->root->start) {
1380 * Root is the last node that might begin its gap before
1381 * start, and this is the last comparison where address
1382 * wrap might be a problem.
1384 st = (start > map->root->end) ? start : map->root->end;
1385 if (length <= map->root->end + map->root->adj_free - st) {
1390 /* With max_free, can immediately tell if no solution. */
1391 entry = map->root->right;
1392 if (entry == NULL || length > entry->max_free)
1396 * Search the right subtree in the order: left subtree, root,
1397 * right subtree (first fit). The previous splay implies that
1398 * all regions in the right subtree have addresses > start.
1400 while (entry != NULL) {
1401 if (entry->left != NULL && entry->left->max_free >= length)
1402 entry = entry->left;
1403 else if (entry->adj_free >= length) {
1407 entry = entry->right;
1410 /* Can't get here, so panic if we do. */
1411 panic("vm_map_findspace: max_free corrupt");
1415 vm_map_fixed(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1416 vm_offset_t start, vm_size_t length, vm_prot_t prot,
1417 vm_prot_t max, int cow)
1422 end = start + length;
1424 VM_MAP_RANGE_CHECK(map, start, end);
1425 (void) vm_map_delete(map, start, end);
1426 result = vm_map_insert(map, object, offset, start, end, prot,
1433 * vm_map_find finds an unallocated region in the target address
1434 * map with the given length. The search is defined to be
1435 * first-fit from the specified address; the region found is
1436 * returned in the same parameter.
1438 * If object is non-NULL, ref count must be bumped by caller
1439 * prior to making call to account for the new entry.
1442 vm_map_find(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1443 vm_offset_t *addr, /* IN/OUT */
1444 vm_size_t length, int find_space, vm_prot_t prot,
1445 vm_prot_t max, int cow)
1453 if (find_space != VMFS_NO_SPACE) {
1454 if (vm_map_findspace(map, start, length, addr)) {
1456 return (KERN_NO_SPACE);
1458 switch (find_space) {
1459 case VMFS_ALIGNED_SPACE:
1460 pmap_align_superpage(object, offset, addr,
1463 #ifdef VMFS_TLB_ALIGNED_SPACE
1464 case VMFS_TLB_ALIGNED_SPACE:
1465 pmap_align_tlb(addr);
1474 result = vm_map_insert(map, object, offset, start, start +
1475 length, prot, max, cow);
1476 } while (result == KERN_NO_SPACE && (find_space == VMFS_ALIGNED_SPACE
1477 #ifdef VMFS_TLB_ALIGNED_SPACE
1478 || find_space == VMFS_TLB_ALIGNED_SPACE
1486 * vm_map_simplify_entry:
1488 * Simplify the given map entry by merging with either neighbor. This
1489 * routine also has the ability to merge with both neighbors.
1491 * The map must be locked.
1493 * This routine guarentees that the passed entry remains valid (though
1494 * possibly extended). When merging, this routine may delete one or
1498 vm_map_simplify_entry(vm_map_t map, vm_map_entry_t entry)
1500 vm_map_entry_t next, prev;
1501 vm_size_t prevsize, esize;
1503 if (entry->eflags & (MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_IS_SUB_MAP))
1507 if (prev != &map->header) {
1508 prevsize = prev->end - prev->start;
1509 if ( (prev->end == entry->start) &&
1510 (prev->object.vm_object == entry->object.vm_object) &&
1511 (!prev->object.vm_object ||
1512 (prev->offset + prevsize == entry->offset)) &&
1513 (prev->eflags == entry->eflags) &&
1514 (prev->protection == entry->protection) &&
1515 (prev->max_protection == entry->max_protection) &&
1516 (prev->inheritance == entry->inheritance) &&
1517 (prev->wired_count == entry->wired_count) &&
1518 (prev->cred == entry->cred)) {
1519 vm_map_entry_unlink(map, prev);
1520 entry->start = prev->start;
1521 entry->offset = prev->offset;
1522 if (entry->prev != &map->header)
1523 vm_map_entry_resize_free(map, entry->prev);
1526 * If the backing object is a vnode object,
1527 * vm_object_deallocate() calls vrele().
1528 * However, vrele() does not lock the vnode
1529 * because the vnode has additional
1530 * references. Thus, the map lock can be kept
1531 * without causing a lock-order reversal with
1534 * Since we count the number of virtual page
1535 * mappings in object->un_pager.vnp.writemappings,
1536 * the writemappings value should not be adjusted
1537 * when the entry is disposed of.
1539 if (prev->object.vm_object)
1540 vm_object_deallocate(prev->object.vm_object);
1541 if (prev->cred != NULL)
1543 vm_map_entry_dispose(map, prev);
1548 if (next != &map->header) {
1549 esize = entry->end - entry->start;
1550 if ((entry->end == next->start) &&
1551 (next->object.vm_object == entry->object.vm_object) &&
1552 (!entry->object.vm_object ||
1553 (entry->offset + esize == next->offset)) &&
1554 (next->eflags == entry->eflags) &&
1555 (next->protection == entry->protection) &&
1556 (next->max_protection == entry->max_protection) &&
1557 (next->inheritance == entry->inheritance) &&
1558 (next->wired_count == entry->wired_count) &&
1559 (next->cred == entry->cred)) {
1560 vm_map_entry_unlink(map, next);
1561 entry->end = next->end;
1562 vm_map_entry_resize_free(map, entry);
1565 * See comment above.
1567 if (next->object.vm_object)
1568 vm_object_deallocate(next->object.vm_object);
1569 if (next->cred != NULL)
1571 vm_map_entry_dispose(map, next);
1576 * vm_map_clip_start: [ internal use only ]
1578 * Asserts that the given entry begins at or after
1579 * the specified address; if necessary,
1580 * it splits the entry into two.
1582 #define vm_map_clip_start(map, entry, startaddr) \
1584 if (startaddr > entry->start) \
1585 _vm_map_clip_start(map, entry, startaddr); \
1589 * This routine is called only when it is known that
1590 * the entry must be split.
1593 _vm_map_clip_start(vm_map_t map, vm_map_entry_t entry, vm_offset_t start)
1595 vm_map_entry_t new_entry;
1597 VM_MAP_ASSERT_LOCKED(map);
1600 * Split off the front portion -- note that we must insert the new
1601 * entry BEFORE this one, so that this entry has the specified
1604 vm_map_simplify_entry(map, entry);
1607 * If there is no object backing this entry, we might as well create
1608 * one now. If we defer it, an object can get created after the map
1609 * is clipped, and individual objects will be created for the split-up
1610 * map. This is a bit of a hack, but is also about the best place to
1611 * put this improvement.
1613 if (entry->object.vm_object == NULL && !map->system_map) {
1615 object = vm_object_allocate(OBJT_DEFAULT,
1616 atop(entry->end - entry->start));
1617 entry->object.vm_object = object;
1619 if (entry->cred != NULL) {
1620 object->cred = entry->cred;
1621 object->charge = entry->end - entry->start;
1624 } else if (entry->object.vm_object != NULL &&
1625 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) &&
1626 entry->cred != NULL) {
1627 VM_OBJECT_WLOCK(entry->object.vm_object);
1628 KASSERT(entry->object.vm_object->cred == NULL,
1629 ("OVERCOMMIT: vm_entry_clip_start: both cred e %p", entry));
1630 entry->object.vm_object->cred = entry->cred;
1631 entry->object.vm_object->charge = entry->end - entry->start;
1632 VM_OBJECT_WUNLOCK(entry->object.vm_object);
1636 new_entry = vm_map_entry_create(map);
1637 *new_entry = *entry;
1639 new_entry->end = start;
1640 entry->offset += (start - entry->start);
1641 entry->start = start;
1642 if (new_entry->cred != NULL)
1643 crhold(entry->cred);
1645 vm_map_entry_link(map, entry->prev, new_entry);
1647 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1648 vm_object_reference(new_entry->object.vm_object);
1650 * The object->un_pager.vnp.writemappings for the
1651 * object of MAP_ENTRY_VN_WRITECNT type entry shall be
1652 * kept as is here. The virtual pages are
1653 * re-distributed among the clipped entries, so the sum is
1660 * vm_map_clip_end: [ internal use only ]
1662 * Asserts that the given entry ends at or before
1663 * the specified address; if necessary,
1664 * it splits the entry into two.
1666 #define vm_map_clip_end(map, entry, endaddr) \
1668 if ((endaddr) < (entry->end)) \
1669 _vm_map_clip_end((map), (entry), (endaddr)); \
1673 * This routine is called only when it is known that
1674 * the entry must be split.
1677 _vm_map_clip_end(vm_map_t map, vm_map_entry_t entry, vm_offset_t end)
1679 vm_map_entry_t new_entry;
1681 VM_MAP_ASSERT_LOCKED(map);
1684 * If there is no object backing this entry, we might as well create
1685 * one now. If we defer it, an object can get created after the map
1686 * is clipped, and individual objects will be created for the split-up
1687 * map. This is a bit of a hack, but is also about the best place to
1688 * put this improvement.
1690 if (entry->object.vm_object == NULL && !map->system_map) {
1692 object = vm_object_allocate(OBJT_DEFAULT,
1693 atop(entry->end - entry->start));
1694 entry->object.vm_object = object;
1696 if (entry->cred != NULL) {
1697 object->cred = entry->cred;
1698 object->charge = entry->end - entry->start;
1701 } else if (entry->object.vm_object != NULL &&
1702 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) &&
1703 entry->cred != NULL) {
1704 VM_OBJECT_WLOCK(entry->object.vm_object);
1705 KASSERT(entry->object.vm_object->cred == NULL,
1706 ("OVERCOMMIT: vm_entry_clip_end: both cred e %p", entry));
1707 entry->object.vm_object->cred = entry->cred;
1708 entry->object.vm_object->charge = entry->end - entry->start;
1709 VM_OBJECT_WUNLOCK(entry->object.vm_object);
1714 * Create a new entry and insert it AFTER the specified entry
1716 new_entry = vm_map_entry_create(map);
1717 *new_entry = *entry;
1719 new_entry->start = entry->end = end;
1720 new_entry->offset += (end - entry->start);
1721 if (new_entry->cred != NULL)
1722 crhold(entry->cred);
1724 vm_map_entry_link(map, entry, new_entry);
1726 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1727 vm_object_reference(new_entry->object.vm_object);
1732 * vm_map_submap: [ kernel use only ]
1734 * Mark the given range as handled by a subordinate map.
1736 * This range must have been created with vm_map_find,
1737 * and no other operations may have been performed on this
1738 * range prior to calling vm_map_submap.
1740 * Only a limited number of operations can be performed
1741 * within this rage after calling vm_map_submap:
1743 * [Don't try vm_map_copy!]
1745 * To remove a submapping, one must first remove the
1746 * range from the superior map, and then destroy the
1747 * submap (if desired). [Better yet, don't try it.]
1756 vm_map_entry_t entry;
1757 int result = KERN_INVALID_ARGUMENT;
1761 VM_MAP_RANGE_CHECK(map, start, end);
1763 if (vm_map_lookup_entry(map, start, &entry)) {
1764 vm_map_clip_start(map, entry, start);
1766 entry = entry->next;
1768 vm_map_clip_end(map, entry, end);
1770 if ((entry->start == start) && (entry->end == end) &&
1771 ((entry->eflags & MAP_ENTRY_COW) == 0) &&
1772 (entry->object.vm_object == NULL)) {
1773 entry->object.sub_map = submap;
1774 entry->eflags |= MAP_ENTRY_IS_SUB_MAP;
1775 result = KERN_SUCCESS;
1783 * The maximum number of pages to map
1785 #define MAX_INIT_PT 96
1788 * vm_map_pmap_enter:
1790 * Preload read-only mappings for the specified object's resident pages
1791 * into the target map. If "flags" is MAP_PREFAULT_PARTIAL, then only
1792 * the resident pages within the address range [addr, addr + ulmin(size,
1793 * ptoa(MAX_INIT_PT))) are mapped. Otherwise, all resident pages within
1794 * the specified address range are mapped. This eliminates many soft
1795 * faults on process startup and immediately after an mmap(2). Because
1796 * these are speculative mappings, cached pages are not reactivated and
1800 vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot,
1801 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags)
1804 vm_page_t p, p_start;
1805 vm_pindex_t psize, tmpidx;
1807 if ((prot & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0 || object == NULL)
1809 VM_OBJECT_RLOCK(object);
1810 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) {
1811 VM_OBJECT_RUNLOCK(object);
1812 VM_OBJECT_WLOCK(object);
1813 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) {
1814 pmap_object_init_pt(map->pmap, addr, object, pindex,
1816 VM_OBJECT_WUNLOCK(object);
1819 VM_OBJECT_LOCK_DOWNGRADE(object);
1823 if (psize > MAX_INIT_PT && (flags & MAP_PREFAULT_PARTIAL) != 0)
1824 psize = MAX_INIT_PT;
1825 if (psize + pindex > object->size) {
1826 if (object->size < pindex) {
1827 VM_OBJECT_RUNLOCK(object);
1830 psize = object->size - pindex;
1836 p = vm_page_find_least(object, pindex);
1838 * Assert: the variable p is either (1) the page with the
1839 * least pindex greater than or equal to the parameter pindex
1843 p != NULL && (tmpidx = p->pindex - pindex) < psize;
1844 p = TAILQ_NEXT(p, listq)) {
1846 * don't allow an madvise to blow away our really
1847 * free pages allocating pv entries.
1849 if ((flags & MAP_PREFAULT_MADVISE) &&
1850 cnt.v_free_count < cnt.v_free_reserved) {
1854 if (p->valid == VM_PAGE_BITS_ALL) {
1855 if (p_start == NULL) {
1856 start = addr + ptoa(tmpidx);
1859 } else if (p_start != NULL) {
1860 pmap_enter_object(map->pmap, start, addr +
1861 ptoa(tmpidx), p_start, prot);
1865 if (p_start != NULL)
1866 pmap_enter_object(map->pmap, start, addr + ptoa(psize),
1868 VM_OBJECT_RUNLOCK(object);
1874 * Sets the protection of the specified address
1875 * region in the target map. If "set_max" is
1876 * specified, the maximum protection is to be set;
1877 * otherwise, only the current protection is affected.
1880 vm_map_protect(vm_map_t map, vm_offset_t start, vm_offset_t end,
1881 vm_prot_t new_prot, boolean_t set_max)
1883 vm_map_entry_t current, entry;
1890 VM_MAP_RANGE_CHECK(map, start, end);
1892 if (vm_map_lookup_entry(map, start, &entry)) {
1893 vm_map_clip_start(map, entry, start);
1895 entry = entry->next;
1899 * Make a first pass to check for protection violations.
1902 while ((current != &map->header) && (current->start < end)) {
1903 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
1905 return (KERN_INVALID_ARGUMENT);
1907 if ((new_prot & current->max_protection) != new_prot) {
1909 return (KERN_PROTECTION_FAILURE);
1911 current = current->next;
1916 * Do an accounting pass for private read-only mappings that
1917 * now will do cow due to allowed write (e.g. debugger sets
1918 * breakpoint on text segment)
1920 for (current = entry; (current != &map->header) &&
1921 (current->start < end); current = current->next) {
1923 vm_map_clip_end(map, current, end);
1926 ((new_prot & ~(current->protection)) & VM_PROT_WRITE) == 0 ||
1927 ENTRY_CHARGED(current)) {
1931 cred = curthread->td_ucred;
1932 obj = current->object.vm_object;
1934 if (obj == NULL || (current->eflags & MAP_ENTRY_NEEDS_COPY)) {
1935 if (!swap_reserve(current->end - current->start)) {
1937 return (KERN_RESOURCE_SHORTAGE);
1940 current->cred = cred;
1944 VM_OBJECT_WLOCK(obj);
1945 if (obj->type != OBJT_DEFAULT && obj->type != OBJT_SWAP) {
1946 VM_OBJECT_WUNLOCK(obj);
1951 * Charge for the whole object allocation now, since
1952 * we cannot distinguish between non-charged and
1953 * charged clipped mapping of the same object later.
1955 KASSERT(obj->charge == 0,
1956 ("vm_map_protect: object %p overcharged\n", obj));
1957 if (!swap_reserve(ptoa(obj->size))) {
1958 VM_OBJECT_WUNLOCK(obj);
1960 return (KERN_RESOURCE_SHORTAGE);
1965 obj->charge = ptoa(obj->size);
1966 VM_OBJECT_WUNLOCK(obj);
1970 * Go back and fix up protections. [Note that clipping is not
1971 * necessary the second time.]
1974 while ((current != &map->header) && (current->start < end)) {
1975 old_prot = current->protection;
1978 current->protection =
1979 (current->max_protection = new_prot) &
1982 current->protection = new_prot;
1984 if ((current->eflags & (MAP_ENTRY_COW | MAP_ENTRY_USER_WIRED))
1985 == (MAP_ENTRY_COW | MAP_ENTRY_USER_WIRED) &&
1986 (current->protection & VM_PROT_WRITE) != 0 &&
1987 (old_prot & VM_PROT_WRITE) == 0) {
1988 vm_fault_copy_entry(map, map, current, current, NULL);
1992 * When restricting access, update the physical map. Worry
1993 * about copy-on-write here.
1995 if ((old_prot & ~current->protection) != 0) {
1996 #define MASK(entry) (((entry)->eflags & MAP_ENTRY_COW) ? ~VM_PROT_WRITE : \
1998 pmap_protect(map->pmap, current->start,
2000 current->protection & MASK(current));
2003 vm_map_simplify_entry(map, current);
2004 current = current->next;
2007 return (KERN_SUCCESS);
2013 * This routine traverses a processes map handling the madvise
2014 * system call. Advisories are classified as either those effecting
2015 * the vm_map_entry structure, or those effecting the underlying
2025 vm_map_entry_t current, entry;
2029 * Some madvise calls directly modify the vm_map_entry, in which case
2030 * we need to use an exclusive lock on the map and we need to perform
2031 * various clipping operations. Otherwise we only need a read-lock
2036 case MADV_SEQUENTIAL:
2048 vm_map_lock_read(map);
2051 return (KERN_INVALID_ARGUMENT);
2055 * Locate starting entry and clip if necessary.
2057 VM_MAP_RANGE_CHECK(map, start, end);
2059 if (vm_map_lookup_entry(map, start, &entry)) {
2061 vm_map_clip_start(map, entry, start);
2063 entry = entry->next;
2068 * madvise behaviors that are implemented in the vm_map_entry.
2070 * We clip the vm_map_entry so that behavioral changes are
2071 * limited to the specified address range.
2073 for (current = entry;
2074 (current != &map->header) && (current->start < end);
2075 current = current->next
2077 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
2080 vm_map_clip_end(map, current, end);
2084 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_NORMAL);
2086 case MADV_SEQUENTIAL:
2087 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_SEQUENTIAL);
2090 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_RANDOM);
2093 current->eflags |= MAP_ENTRY_NOSYNC;
2096 current->eflags &= ~MAP_ENTRY_NOSYNC;
2099 current->eflags |= MAP_ENTRY_NOCOREDUMP;
2102 current->eflags &= ~MAP_ENTRY_NOCOREDUMP;
2107 vm_map_simplify_entry(map, current);
2111 vm_pindex_t pstart, pend;
2114 * madvise behaviors that are implemented in the underlying
2117 * Since we don't clip the vm_map_entry, we have to clip
2118 * the vm_object pindex and count.
2120 for (current = entry;
2121 (current != &map->header) && (current->start < end);
2122 current = current->next
2124 vm_offset_t useStart;
2126 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
2129 pstart = OFF_TO_IDX(current->offset);
2130 pend = pstart + atop(current->end - current->start);
2131 useStart = current->start;
2133 if (current->start < start) {
2134 pstart += atop(start - current->start);
2137 if (current->end > end)
2138 pend -= atop(current->end - end);
2143 vm_object_madvise(current->object.vm_object, pstart,
2145 if (behav == MADV_WILLNEED) {
2146 vm_map_pmap_enter(map,
2148 current->protection,
2149 current->object.vm_object,
2151 ptoa(pend - pstart),
2152 MAP_PREFAULT_MADVISE
2156 vm_map_unlock_read(map);
2165 * Sets the inheritance of the specified address
2166 * range in the target map. Inheritance
2167 * affects how the map will be shared with
2168 * child maps at the time of vmspace_fork.
2171 vm_map_inherit(vm_map_t map, vm_offset_t start, vm_offset_t end,
2172 vm_inherit_t new_inheritance)
2174 vm_map_entry_t entry;
2175 vm_map_entry_t temp_entry;
2177 switch (new_inheritance) {
2178 case VM_INHERIT_NONE:
2179 case VM_INHERIT_COPY:
2180 case VM_INHERIT_SHARE:
2183 return (KERN_INVALID_ARGUMENT);
2186 VM_MAP_RANGE_CHECK(map, start, end);
2187 if (vm_map_lookup_entry(map, start, &temp_entry)) {
2189 vm_map_clip_start(map, entry, start);
2191 entry = temp_entry->next;
2192 while ((entry != &map->header) && (entry->start < end)) {
2193 vm_map_clip_end(map, entry, end);
2194 entry->inheritance = new_inheritance;
2195 vm_map_simplify_entry(map, entry);
2196 entry = entry->next;
2199 return (KERN_SUCCESS);
2205 * Implements both kernel and user unwiring.
2208 vm_map_unwire(vm_map_t map, vm_offset_t start, vm_offset_t end,
2211 vm_map_entry_t entry, first_entry, tmp_entry;
2212 vm_offset_t saved_start;
2213 unsigned int last_timestamp;
2215 boolean_t need_wakeup, result, user_unwire;
2217 user_unwire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE;
2219 VM_MAP_RANGE_CHECK(map, start, end);
2220 if (!vm_map_lookup_entry(map, start, &first_entry)) {
2221 if (flags & VM_MAP_WIRE_HOLESOK)
2222 first_entry = first_entry->next;
2225 return (KERN_INVALID_ADDRESS);
2228 last_timestamp = map->timestamp;
2229 entry = first_entry;
2230 while (entry != &map->header && entry->start < end) {
2231 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
2233 * We have not yet clipped the entry.
2235 saved_start = (start >= entry->start) ? start :
2237 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2238 if (vm_map_unlock_and_wait(map, 0)) {
2240 * Allow interruption of user unwiring?
2244 if (last_timestamp+1 != map->timestamp) {
2246 * Look again for the entry because the map was
2247 * modified while it was unlocked.
2248 * Specifically, the entry may have been
2249 * clipped, merged, or deleted.
2251 if (!vm_map_lookup_entry(map, saved_start,
2253 if (flags & VM_MAP_WIRE_HOLESOK)
2254 tmp_entry = tmp_entry->next;
2256 if (saved_start == start) {
2258 * First_entry has been deleted.
2261 return (KERN_INVALID_ADDRESS);
2264 rv = KERN_INVALID_ADDRESS;
2268 if (entry == first_entry)
2269 first_entry = tmp_entry;
2274 last_timestamp = map->timestamp;
2277 vm_map_clip_start(map, entry, start);
2278 vm_map_clip_end(map, entry, end);
2280 * Mark the entry in case the map lock is released. (See
2283 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
2285 * Check the map for holes in the specified region.
2286 * If VM_MAP_WIRE_HOLESOK was specified, skip this check.
2288 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) &&
2289 (entry->end < end && (entry->next == &map->header ||
2290 entry->next->start > entry->end))) {
2292 rv = KERN_INVALID_ADDRESS;
2296 * If system unwiring, require that the entry is system wired.
2299 vm_map_entry_system_wired_count(entry) == 0) {
2301 rv = KERN_INVALID_ARGUMENT;
2304 entry = entry->next;
2308 need_wakeup = FALSE;
2309 if (first_entry == NULL) {
2310 result = vm_map_lookup_entry(map, start, &first_entry);
2311 if (!result && (flags & VM_MAP_WIRE_HOLESOK))
2312 first_entry = first_entry->next;
2314 KASSERT(result, ("vm_map_unwire: lookup failed"));
2316 entry = first_entry;
2317 while (entry != &map->header && entry->start < end) {
2318 if (rv == KERN_SUCCESS && (!user_unwire ||
2319 (entry->eflags & MAP_ENTRY_USER_WIRED))) {
2321 entry->eflags &= ~MAP_ENTRY_USER_WIRED;
2322 entry->wired_count--;
2323 if (entry->wired_count == 0) {
2325 * Retain the map lock.
2327 vm_fault_unwire(map, entry->start, entry->end,
2328 entry->object.vm_object != NULL &&
2329 (entry->object.vm_object->flags &
2330 OBJ_FICTITIOUS) != 0);
2333 KASSERT(entry->eflags & MAP_ENTRY_IN_TRANSITION,
2334 ("vm_map_unwire: in-transition flag missing"));
2335 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION;
2336 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
2337 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
2340 vm_map_simplify_entry(map, entry);
2341 entry = entry->next;
2352 * Implements both kernel and user wiring.
2355 vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t end,
2358 vm_map_entry_t entry, first_entry, tmp_entry;
2359 vm_offset_t saved_end, saved_start;
2360 unsigned int last_timestamp;
2362 boolean_t fictitious, need_wakeup, result, user_wire;
2366 if (flags & VM_MAP_WIRE_WRITE)
2367 prot |= VM_PROT_WRITE;
2368 user_wire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE;
2370 VM_MAP_RANGE_CHECK(map, start, end);
2371 if (!vm_map_lookup_entry(map, start, &first_entry)) {
2372 if (flags & VM_MAP_WIRE_HOLESOK)
2373 first_entry = first_entry->next;
2376 return (KERN_INVALID_ADDRESS);
2379 last_timestamp = map->timestamp;
2380 entry = first_entry;
2381 while (entry != &map->header && entry->start < end) {
2382 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
2384 * We have not yet clipped the entry.
2386 saved_start = (start >= entry->start) ? start :
2388 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2389 if (vm_map_unlock_and_wait(map, 0)) {
2391 * Allow interruption of user wiring?
2395 if (last_timestamp + 1 != map->timestamp) {
2397 * Look again for the entry because the map was
2398 * modified while it was unlocked.
2399 * Specifically, the entry may have been
2400 * clipped, merged, or deleted.
2402 if (!vm_map_lookup_entry(map, saved_start,
2404 if (flags & VM_MAP_WIRE_HOLESOK)
2405 tmp_entry = tmp_entry->next;
2407 if (saved_start == start) {
2409 * first_entry has been deleted.
2412 return (KERN_INVALID_ADDRESS);
2415 rv = KERN_INVALID_ADDRESS;
2419 if (entry == first_entry)
2420 first_entry = tmp_entry;
2425 last_timestamp = map->timestamp;
2428 vm_map_clip_start(map, entry, start);
2429 vm_map_clip_end(map, entry, end);
2431 * Mark the entry in case the map lock is released. (See
2434 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
2435 if ((entry->protection & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0
2436 || (entry->protection & prot) != prot) {
2437 entry->eflags |= MAP_ENTRY_WIRE_SKIPPED;
2438 if ((flags & VM_MAP_WIRE_HOLESOK) == 0) {
2440 rv = KERN_INVALID_ADDRESS;
2445 if (entry->wired_count == 0) {
2446 entry->wired_count++;
2447 saved_start = entry->start;
2448 saved_end = entry->end;
2449 fictitious = entry->object.vm_object != NULL &&
2450 (entry->object.vm_object->flags &
2451 OBJ_FICTITIOUS) != 0;
2453 * Release the map lock, relying on the in-transition
2454 * mark. Mark the map busy for fork.
2458 rv = vm_fault_wire(map, saved_start, saved_end,
2462 if (last_timestamp + 1 != map->timestamp) {
2464 * Look again for the entry because the map was
2465 * modified while it was unlocked. The entry
2466 * may have been clipped, but NOT merged or
2469 result = vm_map_lookup_entry(map, saved_start,
2471 KASSERT(result, ("vm_map_wire: lookup failed"));
2472 if (entry == first_entry)
2473 first_entry = tmp_entry;
2477 while (entry->end < saved_end) {
2478 if (rv != KERN_SUCCESS) {
2479 KASSERT(entry->wired_count == 1,
2480 ("vm_map_wire: bad count"));
2481 entry->wired_count = -1;
2483 entry = entry->next;
2486 last_timestamp = map->timestamp;
2487 if (rv != KERN_SUCCESS) {
2488 KASSERT(entry->wired_count == 1,
2489 ("vm_map_wire: bad count"));
2491 * Assign an out-of-range value to represent
2492 * the failure to wire this entry.
2494 entry->wired_count = -1;
2498 } else if (!user_wire ||
2499 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
2500 entry->wired_count++;
2503 * Check the map for holes in the specified region.
2504 * If VM_MAP_WIRE_HOLESOK was specified, skip this check.
2507 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) &&
2508 (entry->end < end && (entry->next == &map->header ||
2509 entry->next->start > entry->end))) {
2511 rv = KERN_INVALID_ADDRESS;
2514 entry = entry->next;
2518 need_wakeup = FALSE;
2519 if (first_entry == NULL) {
2520 result = vm_map_lookup_entry(map, start, &first_entry);
2521 if (!result && (flags & VM_MAP_WIRE_HOLESOK))
2522 first_entry = first_entry->next;
2524 KASSERT(result, ("vm_map_wire: lookup failed"));
2526 entry = first_entry;
2527 while (entry != &map->header && entry->start < end) {
2528 if ((entry->eflags & MAP_ENTRY_WIRE_SKIPPED) != 0)
2529 goto next_entry_done;
2530 if (rv == KERN_SUCCESS) {
2532 entry->eflags |= MAP_ENTRY_USER_WIRED;
2533 } else if (entry->wired_count == -1) {
2535 * Wiring failed on this entry. Thus, unwiring is
2538 entry->wired_count = 0;
2541 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0)
2542 entry->wired_count--;
2543 if (entry->wired_count == 0) {
2545 * Retain the map lock.
2547 vm_fault_unwire(map, entry->start, entry->end,
2548 entry->object.vm_object != NULL &&
2549 (entry->object.vm_object->flags &
2550 OBJ_FICTITIOUS) != 0);
2554 KASSERT(entry->eflags & MAP_ENTRY_IN_TRANSITION,
2555 ("vm_map_wire: in-transition flag missing"));
2556 entry->eflags &= ~(MAP_ENTRY_IN_TRANSITION|MAP_ENTRY_WIRE_SKIPPED);
2557 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
2558 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
2561 vm_map_simplify_entry(map, entry);
2562 entry = entry->next;
2573 * Push any dirty cached pages in the address range to their pager.
2574 * If syncio is TRUE, dirty pages are written synchronously.
2575 * If invalidate is TRUE, any cached pages are freed as well.
2577 * If the size of the region from start to end is zero, we are
2578 * supposed to flush all modified pages within the region containing
2579 * start. Unfortunately, a region can be split or coalesced with
2580 * neighboring regions, making it difficult to determine what the
2581 * original region was. Therefore, we approximate this requirement by
2582 * flushing the current region containing start.
2584 * Returns an error if any part of the specified range is not mapped.
2592 boolean_t invalidate)
2594 vm_map_entry_t current;
2595 vm_map_entry_t entry;
2598 vm_ooffset_t offset;
2599 unsigned int last_timestamp;
2602 vm_map_lock_read(map);
2603 VM_MAP_RANGE_CHECK(map, start, end);
2604 if (!vm_map_lookup_entry(map, start, &entry)) {
2605 vm_map_unlock_read(map);
2606 return (KERN_INVALID_ADDRESS);
2607 } else if (start == end) {
2608 start = entry->start;
2612 * Make a first pass to check for user-wired memory and holes.
2614 for (current = entry; current != &map->header && current->start < end;
2615 current = current->next) {
2616 if (invalidate && (current->eflags & MAP_ENTRY_USER_WIRED)) {
2617 vm_map_unlock_read(map);
2618 return (KERN_INVALID_ARGUMENT);
2620 if (end > current->end &&
2621 (current->next == &map->header ||
2622 current->end != current->next->start)) {
2623 vm_map_unlock_read(map);
2624 return (KERN_INVALID_ADDRESS);
2629 pmap_remove(map->pmap, start, end);
2633 * Make a second pass, cleaning/uncaching pages from the indicated
2636 for (current = entry; current != &map->header && current->start < end;) {
2637 offset = current->offset + (start - current->start);
2638 size = (end <= current->end ? end : current->end) - start;
2639 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
2641 vm_map_entry_t tentry;
2644 smap = current->object.sub_map;
2645 vm_map_lock_read(smap);
2646 (void) vm_map_lookup_entry(smap, offset, &tentry);
2647 tsize = tentry->end - offset;
2650 object = tentry->object.vm_object;
2651 offset = tentry->offset + (offset - tentry->start);
2652 vm_map_unlock_read(smap);
2654 object = current->object.vm_object;
2656 vm_object_reference(object);
2657 last_timestamp = map->timestamp;
2658 vm_map_unlock_read(map);
2659 if (!vm_object_sync(object, offset, size, syncio, invalidate))
2662 vm_object_deallocate(object);
2663 vm_map_lock_read(map);
2664 if (last_timestamp == map->timestamp ||
2665 !vm_map_lookup_entry(map, start, ¤t))
2666 current = current->next;
2669 vm_map_unlock_read(map);
2670 return (failed ? KERN_FAILURE : KERN_SUCCESS);
2674 * vm_map_entry_unwire: [ internal use only ]
2676 * Make the region specified by this entry pageable.
2678 * The map in question should be locked.
2679 * [This is the reason for this routine's existence.]
2682 vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry)
2684 vm_fault_unwire(map, entry->start, entry->end,
2685 entry->object.vm_object != NULL &&
2686 (entry->object.vm_object->flags & OBJ_FICTITIOUS) != 0);
2687 entry->wired_count = 0;
2691 vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map)
2694 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0)
2695 vm_object_deallocate(entry->object.vm_object);
2696 uma_zfree(system_map ? kmapentzone : mapentzone, entry);
2700 * vm_map_entry_delete: [ internal use only ]
2702 * Deallocate the given entry from the target map.
2705 vm_map_entry_delete(vm_map_t map, vm_map_entry_t entry)
2708 vm_pindex_t offidxstart, offidxend, count, size1;
2711 vm_map_entry_unlink(map, entry);
2712 object = entry->object.vm_object;
2713 size = entry->end - entry->start;
2716 if (entry->cred != NULL) {
2717 swap_release_by_cred(size, entry->cred);
2718 crfree(entry->cred);
2721 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0 &&
2723 KASSERT(entry->cred == NULL || object->cred == NULL ||
2724 (entry->eflags & MAP_ENTRY_NEEDS_COPY),
2725 ("OVERCOMMIT vm_map_entry_delete: both cred %p", entry));
2726 count = OFF_TO_IDX(size);
2727 offidxstart = OFF_TO_IDX(entry->offset);
2728 offidxend = offidxstart + count;
2729 VM_OBJECT_WLOCK(object);
2730 if (object->ref_count != 1 &&
2731 ((object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING ||
2732 object == kernel_object || object == kmem_object)) {
2733 vm_object_collapse(object);
2736 * The option OBJPR_NOTMAPPED can be passed here
2737 * because vm_map_delete() already performed
2738 * pmap_remove() on the only mapping to this range
2741 vm_object_page_remove(object, offidxstart, offidxend,
2743 if (object->type == OBJT_SWAP)
2744 swap_pager_freespace(object, offidxstart, count);
2745 if (offidxend >= object->size &&
2746 offidxstart < object->size) {
2747 size1 = object->size;
2748 object->size = offidxstart;
2749 if (object->cred != NULL) {
2750 size1 -= object->size;
2751 KASSERT(object->charge >= ptoa(size1),
2752 ("vm_map_entry_delete: object->charge < 0"));
2753 swap_release_by_cred(ptoa(size1), object->cred);
2754 object->charge -= ptoa(size1);
2758 VM_OBJECT_WUNLOCK(object);
2760 entry->object.vm_object = NULL;
2761 if (map->system_map)
2762 vm_map_entry_deallocate(entry, TRUE);
2764 entry->next = curthread->td_map_def_user;
2765 curthread->td_map_def_user = entry;
2770 * vm_map_delete: [ internal use only ]
2772 * Deallocates the given address range from the target
2776 vm_map_delete(vm_map_t map, vm_offset_t start, vm_offset_t end)
2778 vm_map_entry_t entry;
2779 vm_map_entry_t first_entry;
2781 VM_MAP_ASSERT_LOCKED(map);
2784 * Find the start of the region, and clip it
2786 if (!vm_map_lookup_entry(map, start, &first_entry))
2787 entry = first_entry->next;
2789 entry = first_entry;
2790 vm_map_clip_start(map, entry, start);
2794 * Step through all entries in this region
2796 while ((entry != &map->header) && (entry->start < end)) {
2797 vm_map_entry_t next;
2800 * Wait for wiring or unwiring of an entry to complete.
2801 * Also wait for any system wirings to disappear on
2804 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 ||
2805 (vm_map_pmap(map) != kernel_pmap &&
2806 vm_map_entry_system_wired_count(entry) != 0)) {
2807 unsigned int last_timestamp;
2808 vm_offset_t saved_start;
2809 vm_map_entry_t tmp_entry;
2811 saved_start = entry->start;
2812 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2813 last_timestamp = map->timestamp;
2814 (void) vm_map_unlock_and_wait(map, 0);
2816 if (last_timestamp + 1 != map->timestamp) {
2818 * Look again for the entry because the map was
2819 * modified while it was unlocked.
2820 * Specifically, the entry may have been
2821 * clipped, merged, or deleted.
2823 if (!vm_map_lookup_entry(map, saved_start,
2825 entry = tmp_entry->next;
2828 vm_map_clip_start(map, entry,
2834 vm_map_clip_end(map, entry, end);
2839 * Unwire before removing addresses from the pmap; otherwise,
2840 * unwiring will put the entries back in the pmap.
2842 if (entry->wired_count != 0) {
2843 vm_map_entry_unwire(map, entry);
2846 pmap_remove(map->pmap, entry->start, entry->end);
2849 * Delete the entry only after removing all pmap
2850 * entries pointing to its pages. (Otherwise, its
2851 * page frames may be reallocated, and any modify bits
2852 * will be set in the wrong object!)
2854 vm_map_entry_delete(map, entry);
2857 return (KERN_SUCCESS);
2863 * Remove the given address range from the target map.
2864 * This is the exported form of vm_map_delete.
2867 vm_map_remove(vm_map_t map, vm_offset_t start, vm_offset_t end)
2872 VM_MAP_RANGE_CHECK(map, start, end);
2873 result = vm_map_delete(map, start, end);
2879 * vm_map_check_protection:
2881 * Assert that the target map allows the specified privilege on the
2882 * entire address region given. The entire region must be allocated.
2884 * WARNING! This code does not and should not check whether the
2885 * contents of the region is accessible. For example a smaller file
2886 * might be mapped into a larger address space.
2888 * NOTE! This code is also called by munmap().
2890 * The map must be locked. A read lock is sufficient.
2893 vm_map_check_protection(vm_map_t map, vm_offset_t start, vm_offset_t end,
2894 vm_prot_t protection)
2896 vm_map_entry_t entry;
2897 vm_map_entry_t tmp_entry;
2899 if (!vm_map_lookup_entry(map, start, &tmp_entry))
2903 while (start < end) {
2904 if (entry == &map->header)
2909 if (start < entry->start)
2912 * Check protection associated with entry.
2914 if ((entry->protection & protection) != protection)
2916 /* go to next entry */
2918 entry = entry->next;
2924 * vm_map_copy_entry:
2926 * Copies the contents of the source entry to the destination
2927 * entry. The entries *must* be aligned properly.
2933 vm_map_entry_t src_entry,
2934 vm_map_entry_t dst_entry,
2935 vm_ooffset_t *fork_charge)
2937 vm_object_t src_object;
2938 vm_map_entry_t fake_entry;
2943 VM_MAP_ASSERT_LOCKED(dst_map);
2945 if ((dst_entry->eflags|src_entry->eflags) & MAP_ENTRY_IS_SUB_MAP)
2948 if (src_entry->wired_count == 0) {
2951 * If the source entry is marked needs_copy, it is already
2954 if ((src_entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) {
2955 pmap_protect(src_map->pmap,
2958 src_entry->protection & ~VM_PROT_WRITE);
2962 * Make a copy of the object.
2964 size = src_entry->end - src_entry->start;
2965 if ((src_object = src_entry->object.vm_object) != NULL) {
2966 VM_OBJECT_WLOCK(src_object);
2967 charged = ENTRY_CHARGED(src_entry);
2968 if ((src_object->handle == NULL) &&
2969 (src_object->type == OBJT_DEFAULT ||
2970 src_object->type == OBJT_SWAP)) {
2971 vm_object_collapse(src_object);
2972 if ((src_object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING) {
2973 vm_object_split(src_entry);
2974 src_object = src_entry->object.vm_object;
2977 vm_object_reference_locked(src_object);
2978 vm_object_clear_flag(src_object, OBJ_ONEMAPPING);
2979 if (src_entry->cred != NULL &&
2980 !(src_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
2981 KASSERT(src_object->cred == NULL,
2982 ("OVERCOMMIT: vm_map_copy_entry: cred %p",
2984 src_object->cred = src_entry->cred;
2985 src_object->charge = size;
2987 VM_OBJECT_WUNLOCK(src_object);
2988 dst_entry->object.vm_object = src_object;
2990 cred = curthread->td_ucred;
2992 dst_entry->cred = cred;
2993 *fork_charge += size;
2994 if (!(src_entry->eflags &
2995 MAP_ENTRY_NEEDS_COPY)) {
2997 src_entry->cred = cred;
2998 *fork_charge += size;
3001 src_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY);
3002 dst_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY);
3003 dst_entry->offset = src_entry->offset;
3004 if (src_entry->eflags & MAP_ENTRY_VN_WRITECNT) {
3006 * MAP_ENTRY_VN_WRITECNT cannot
3007 * indicate write reference from
3008 * src_entry, since the entry is
3009 * marked as needs copy. Allocate a
3010 * fake entry that is used to
3011 * decrement object->un_pager.vnp.writecount
3012 * at the appropriate time. Attach
3013 * fake_entry to the deferred list.
3015 fake_entry = vm_map_entry_create(dst_map);
3016 fake_entry->eflags = MAP_ENTRY_VN_WRITECNT;
3017 src_entry->eflags &= ~MAP_ENTRY_VN_WRITECNT;
3018 vm_object_reference(src_object);
3019 fake_entry->object.vm_object = src_object;
3020 fake_entry->start = src_entry->start;
3021 fake_entry->end = src_entry->end;
3022 fake_entry->next = curthread->td_map_def_user;
3023 curthread->td_map_def_user = fake_entry;
3026 dst_entry->object.vm_object = NULL;
3027 dst_entry->offset = 0;
3028 if (src_entry->cred != NULL) {
3029 dst_entry->cred = curthread->td_ucred;
3030 crhold(dst_entry->cred);
3031 *fork_charge += size;
3035 pmap_copy(dst_map->pmap, src_map->pmap, dst_entry->start,
3036 dst_entry->end - dst_entry->start, src_entry->start);
3039 * Of course, wired down pages can't be set copy-on-write.
3040 * Cause wired pages to be copied into the new map by
3041 * simulating faults (the new pages are pageable)
3043 vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry,
3049 * vmspace_map_entry_forked:
3050 * Update the newly-forked vmspace each time a map entry is inherited
3051 * or copied. The values for vm_dsize and vm_tsize are approximate
3052 * (and mostly-obsolete ideas in the face of mmap(2) et al.)
3055 vmspace_map_entry_forked(const struct vmspace *vm1, struct vmspace *vm2,
3056 vm_map_entry_t entry)
3058 vm_size_t entrysize;
3061 entrysize = entry->end - entry->start;
3062 vm2->vm_map.size += entrysize;
3063 if (entry->eflags & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP)) {
3064 vm2->vm_ssize += btoc(entrysize);
3065 } else if (entry->start >= (vm_offset_t)vm1->vm_daddr &&
3066 entry->start < (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)) {
3067 newend = MIN(entry->end,
3068 (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize));
3069 vm2->vm_dsize += btoc(newend - entry->start);
3070 } else if (entry->start >= (vm_offset_t)vm1->vm_taddr &&
3071 entry->start < (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)) {
3072 newend = MIN(entry->end,
3073 (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize));
3074 vm2->vm_tsize += btoc(newend - entry->start);
3080 * Create a new process vmspace structure and vm_map
3081 * based on those of an existing process. The new map
3082 * is based on the old map, according to the inheritance
3083 * values on the regions in that map.
3085 * XXX It might be worth coalescing the entries added to the new vmspace.
3087 * The source map must not be locked.
3090 vmspace_fork(struct vmspace *vm1, vm_ooffset_t *fork_charge)
3092 struct vmspace *vm2;
3093 vm_map_t new_map, old_map;
3094 vm_map_entry_t new_entry, old_entry;
3098 old_map = &vm1->vm_map;
3099 /* Copy immutable fields of vm1 to vm2. */
3100 vm2 = vmspace_alloc(old_map->min_offset, old_map->max_offset);
3103 vm2->vm_taddr = vm1->vm_taddr;
3104 vm2->vm_daddr = vm1->vm_daddr;
3105 vm2->vm_maxsaddr = vm1->vm_maxsaddr;
3106 vm_map_lock(old_map);
3108 vm_map_wait_busy(old_map);
3109 new_map = &vm2->vm_map;
3110 locked = vm_map_trylock(new_map); /* trylock to silence WITNESS */
3111 KASSERT(locked, ("vmspace_fork: lock failed"));
3113 old_entry = old_map->header.next;
3115 while (old_entry != &old_map->header) {
3116 if (old_entry->eflags & MAP_ENTRY_IS_SUB_MAP)
3117 panic("vm_map_fork: encountered a submap");
3119 switch (old_entry->inheritance) {
3120 case VM_INHERIT_NONE:
3123 case VM_INHERIT_SHARE:
3125 * Clone the entry, creating the shared object if necessary.
3127 object = old_entry->object.vm_object;
3128 if (object == NULL) {
3129 object = vm_object_allocate(OBJT_DEFAULT,
3130 atop(old_entry->end - old_entry->start));
3131 old_entry->object.vm_object = object;
3132 old_entry->offset = 0;
3133 if (old_entry->cred != NULL) {
3134 object->cred = old_entry->cred;
3135 object->charge = old_entry->end -
3137 old_entry->cred = NULL;
3142 * Add the reference before calling vm_object_shadow
3143 * to insure that a shadow object is created.
3145 vm_object_reference(object);
3146 if (old_entry->eflags & MAP_ENTRY_NEEDS_COPY) {
3147 vm_object_shadow(&old_entry->object.vm_object,
3149 old_entry->end - old_entry->start);
3150 old_entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
3151 /* Transfer the second reference too. */
3152 vm_object_reference(
3153 old_entry->object.vm_object);
3156 * As in vm_map_simplify_entry(), the
3157 * vnode lock will not be acquired in
3158 * this call to vm_object_deallocate().
3160 vm_object_deallocate(object);
3161 object = old_entry->object.vm_object;
3163 VM_OBJECT_WLOCK(object);
3164 vm_object_clear_flag(object, OBJ_ONEMAPPING);
3165 if (old_entry->cred != NULL) {
3166 KASSERT(object->cred == NULL, ("vmspace_fork both cred"));
3167 object->cred = old_entry->cred;
3168 object->charge = old_entry->end - old_entry->start;
3169 old_entry->cred = NULL;
3173 * Assert the correct state of the vnode
3174 * v_writecount while the object is locked, to
3175 * not relock it later for the assertion
3178 if (old_entry->eflags & MAP_ENTRY_VN_WRITECNT &&
3179 object->type == OBJT_VNODE) {
3180 KASSERT(((struct vnode *)object->handle)->
3182 ("vmspace_fork: v_writecount %p", object));
3183 KASSERT(object->un_pager.vnp.writemappings > 0,
3184 ("vmspace_fork: vnp.writecount %p",
3187 VM_OBJECT_WUNLOCK(object);
3190 * Clone the entry, referencing the shared object.
3192 new_entry = vm_map_entry_create(new_map);
3193 *new_entry = *old_entry;
3194 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED |
3195 MAP_ENTRY_IN_TRANSITION);
3196 new_entry->wired_count = 0;
3197 if (new_entry->eflags & MAP_ENTRY_VN_WRITECNT) {
3198 vnode_pager_update_writecount(object,
3199 new_entry->start, new_entry->end);
3203 * Insert the entry into the new map -- we know we're
3204 * inserting at the end of the new map.
3206 vm_map_entry_link(new_map, new_map->header.prev,
3208 vmspace_map_entry_forked(vm1, vm2, new_entry);
3211 * Update the physical map
3213 pmap_copy(new_map->pmap, old_map->pmap,
3215 (old_entry->end - old_entry->start),
3219 case VM_INHERIT_COPY:
3221 * Clone the entry and link into the map.
3223 new_entry = vm_map_entry_create(new_map);
3224 *new_entry = *old_entry;
3226 * Copied entry is COW over the old object.
3228 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED |
3229 MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_VN_WRITECNT);
3230 new_entry->wired_count = 0;
3231 new_entry->object.vm_object = NULL;
3232 new_entry->cred = NULL;
3233 vm_map_entry_link(new_map, new_map->header.prev,
3235 vmspace_map_entry_forked(vm1, vm2, new_entry);
3236 vm_map_copy_entry(old_map, new_map, old_entry,
3237 new_entry, fork_charge);
3240 old_entry = old_entry->next;
3243 * Use inlined vm_map_unlock() to postpone handling the deferred
3244 * map entries, which cannot be done until both old_map and
3245 * new_map locks are released.
3247 sx_xunlock(&old_map->lock);
3248 sx_xunlock(&new_map->lock);
3249 vm_map_process_deferred();
3255 vm_map_stack(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
3256 vm_prot_t prot, vm_prot_t max, int cow)
3258 vm_map_entry_t new_entry, prev_entry;
3259 vm_offset_t bot, top;
3260 vm_size_t growsize, init_ssize;
3262 rlim_t lmemlim, vmemlim;
3265 * The stack orientation is piggybacked with the cow argument.
3266 * Extract it into orient and mask the cow argument so that we
3267 * don't pass it around further.
3268 * NOTE: We explicitly allow bi-directional stacks.
3270 orient = cow & (MAP_STACK_GROWS_DOWN|MAP_STACK_GROWS_UP);
3272 KASSERT(orient != 0, ("No stack grow direction"));
3274 if (addrbos < vm_map_min(map) ||
3275 addrbos > vm_map_max(map) ||
3276 addrbos + max_ssize < addrbos)
3277 return (KERN_NO_SPACE);
3279 growsize = sgrowsiz;
3280 init_ssize = (max_ssize < growsize) ? max_ssize : growsize;
3283 lmemlim = lim_cur(curproc, RLIMIT_MEMLOCK);
3284 vmemlim = lim_cur(curproc, RLIMIT_VMEM);
3285 PROC_UNLOCK(curproc);
3289 /* If addr is already mapped, no go */
3290 if (vm_map_lookup_entry(map, addrbos, &prev_entry)) {
3292 return (KERN_NO_SPACE);
3295 if (!old_mlock && map->flags & MAP_WIREFUTURE) {
3296 if (ptoa(pmap_wired_count(map->pmap)) + init_ssize > lmemlim) {
3298 return (KERN_NO_SPACE);
3302 /* If we would blow our VMEM resource limit, no go */
3303 if (map->size + init_ssize > vmemlim) {
3305 return (KERN_NO_SPACE);
3309 * If we can't accomodate max_ssize in the current mapping, no go.
3310 * However, we need to be aware that subsequent user mappings might
3311 * map into the space we have reserved for stack, and currently this
3312 * space is not protected.
3314 * Hopefully we will at least detect this condition when we try to
3317 if ((prev_entry->next != &map->header) &&
3318 (prev_entry->next->start < addrbos + max_ssize)) {
3320 return (KERN_NO_SPACE);
3324 * We initially map a stack of only init_ssize. We will grow as
3325 * needed later. Depending on the orientation of the stack (i.e.
3326 * the grow direction) we either map at the top of the range, the
3327 * bottom of the range or in the middle.
3329 * Note: we would normally expect prot and max to be VM_PROT_ALL,
3330 * and cow to be 0. Possibly we should eliminate these as input
3331 * parameters, and just pass these values here in the insert call.
3333 if (orient == MAP_STACK_GROWS_DOWN)
3334 bot = addrbos + max_ssize - init_ssize;
3335 else if (orient == MAP_STACK_GROWS_UP)
3338 bot = round_page(addrbos + max_ssize/2 - init_ssize/2);
3339 top = bot + init_ssize;
3340 rv = vm_map_insert(map, NULL, 0, bot, top, prot, max, cow);
3342 /* Now set the avail_ssize amount. */
3343 if (rv == KERN_SUCCESS) {
3344 if (prev_entry != &map->header)
3345 vm_map_clip_end(map, prev_entry, bot);
3346 new_entry = prev_entry->next;
3347 if (new_entry->end != top || new_entry->start != bot)
3348 panic("Bad entry start/end for new stack entry");
3350 new_entry->avail_ssize = max_ssize - init_ssize;
3351 if (orient & MAP_STACK_GROWS_DOWN)
3352 new_entry->eflags |= MAP_ENTRY_GROWS_DOWN;
3353 if (orient & MAP_STACK_GROWS_UP)
3354 new_entry->eflags |= MAP_ENTRY_GROWS_UP;
3361 static int stack_guard_page = 0;
3362 TUNABLE_INT("security.bsd.stack_guard_page", &stack_guard_page);
3363 SYSCTL_INT(_security_bsd, OID_AUTO, stack_guard_page, CTLFLAG_RW,
3364 &stack_guard_page, 0,
3365 "Insert stack guard page ahead of the growable segments.");
3367 /* Attempts to grow a vm stack entry. Returns KERN_SUCCESS if the
3368 * desired address is already mapped, or if we successfully grow
3369 * the stack. Also returns KERN_SUCCESS if addr is outside the
3370 * stack range (this is strange, but preserves compatibility with
3371 * the grow function in vm_machdep.c).
3374 vm_map_growstack(struct proc *p, vm_offset_t addr)
3376 vm_map_entry_t next_entry, prev_entry;
3377 vm_map_entry_t new_entry, stack_entry;
3378 struct vmspace *vm = p->p_vmspace;
3379 vm_map_t map = &vm->vm_map;
3382 size_t grow_amount, max_grow;
3383 rlim_t lmemlim, stacklim, vmemlim;
3384 int is_procstack, rv;
3395 lmemlim = lim_cur(p, RLIMIT_MEMLOCK);
3396 stacklim = lim_cur(p, RLIMIT_STACK);
3397 vmemlim = lim_cur(p, RLIMIT_VMEM);
3400 vm_map_lock_read(map);
3402 /* If addr is already in the entry range, no need to grow.*/
3403 if (vm_map_lookup_entry(map, addr, &prev_entry)) {
3404 vm_map_unlock_read(map);
3405 return (KERN_SUCCESS);
3408 next_entry = prev_entry->next;
3409 if (!(prev_entry->eflags & MAP_ENTRY_GROWS_UP)) {
3411 * This entry does not grow upwards. Since the address lies
3412 * beyond this entry, the next entry (if one exists) has to
3413 * be a downward growable entry. The entry list header is
3414 * never a growable entry, so it suffices to check the flags.
3416 if (!(next_entry->eflags & MAP_ENTRY_GROWS_DOWN)) {
3417 vm_map_unlock_read(map);
3418 return (KERN_SUCCESS);
3420 stack_entry = next_entry;
3423 * This entry grows upward. If the next entry does not at
3424 * least grow downwards, this is the entry we need to grow.
3425 * otherwise we have two possible choices and we have to
3428 if (next_entry->eflags & MAP_ENTRY_GROWS_DOWN) {
3430 * We have two choices; grow the entry closest to
3431 * the address to minimize the amount of growth.
3433 if (addr - prev_entry->end <= next_entry->start - addr)
3434 stack_entry = prev_entry;
3436 stack_entry = next_entry;
3438 stack_entry = prev_entry;
3441 if (stack_entry == next_entry) {
3442 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_DOWN, ("foo"));
3443 KASSERT(addr < stack_entry->start, ("foo"));
3444 end = (prev_entry != &map->header) ? prev_entry->end :
3445 stack_entry->start - stack_entry->avail_ssize;
3446 grow_amount = roundup(stack_entry->start - addr, PAGE_SIZE);
3447 max_grow = stack_entry->start - end;
3449 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_UP, ("foo"));
3450 KASSERT(addr >= stack_entry->end, ("foo"));
3451 end = (next_entry != &map->header) ? next_entry->start :
3452 stack_entry->end + stack_entry->avail_ssize;
3453 grow_amount = roundup(addr + 1 - stack_entry->end, PAGE_SIZE);
3454 max_grow = end - stack_entry->end;
3457 if (grow_amount > stack_entry->avail_ssize) {
3458 vm_map_unlock_read(map);
3459 return (KERN_NO_SPACE);
3463 * If there is no longer enough space between the entries nogo, and
3464 * adjust the available space. Note: this should only happen if the
3465 * user has mapped into the stack area after the stack was created,
3466 * and is probably an error.
3468 * This also effectively destroys any guard page the user might have
3469 * intended by limiting the stack size.
3471 if (grow_amount + (stack_guard_page ? PAGE_SIZE : 0) > max_grow) {
3472 if (vm_map_lock_upgrade(map))
3475 stack_entry->avail_ssize = max_grow;
3478 return (KERN_NO_SPACE);
3481 is_procstack = (addr >= (vm_offset_t)vm->vm_maxsaddr) ? 1 : 0;
3484 * If this is the main process stack, see if we're over the stack
3487 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) {
3488 vm_map_unlock_read(map);
3489 return (KERN_NO_SPACE);
3494 racct_set(p, RACCT_STACK, ctob(vm->vm_ssize) + grow_amount)) {
3496 vm_map_unlock_read(map);
3497 return (KERN_NO_SPACE);
3502 /* Round up the grow amount modulo sgrowsiz */
3503 growsize = sgrowsiz;
3504 grow_amount = roundup(grow_amount, growsize);
3505 if (grow_amount > stack_entry->avail_ssize)
3506 grow_amount = stack_entry->avail_ssize;
3507 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) {
3508 grow_amount = trunc_page((vm_size_t)stacklim) -
3513 limit = racct_get_available(p, RACCT_STACK);
3515 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > limit))
3516 grow_amount = limit - ctob(vm->vm_ssize);
3518 if (!old_mlock && map->flags & MAP_WIREFUTURE) {
3519 if (ptoa(pmap_wired_count(map->pmap)) + grow_amount > lmemlim) {
3520 vm_map_unlock_read(map);
3526 if (racct_set(p, RACCT_MEMLOCK,
3527 ptoa(pmap_wired_count(map->pmap)) + grow_amount)) {
3529 vm_map_unlock_read(map);
3536 /* If we would blow our VMEM resource limit, no go */
3537 if (map->size + grow_amount > vmemlim) {
3538 vm_map_unlock_read(map);
3544 if (racct_set(p, RACCT_VMEM, map->size + grow_amount)) {
3546 vm_map_unlock_read(map);
3553 if (vm_map_lock_upgrade(map))
3556 if (stack_entry == next_entry) {
3560 /* Get the preliminary new entry start value */
3561 addr = stack_entry->start - grow_amount;
3564 * If this puts us into the previous entry, cut back our
3565 * growth to the available space. Also, see the note above.
3568 stack_entry->avail_ssize = max_grow;
3570 if (stack_guard_page)
3574 rv = vm_map_insert(map, NULL, 0, addr, stack_entry->start,
3575 next_entry->protection, next_entry->max_protection, 0);
3577 /* Adjust the available stack space by the amount we grew. */
3578 if (rv == KERN_SUCCESS) {
3579 if (prev_entry != &map->header)
3580 vm_map_clip_end(map, prev_entry, addr);
3581 new_entry = prev_entry->next;
3582 KASSERT(new_entry == stack_entry->prev, ("foo"));
3583 KASSERT(new_entry->end == stack_entry->start, ("foo"));
3584 KASSERT(new_entry->start == addr, ("foo"));
3585 grow_amount = new_entry->end - new_entry->start;
3586 new_entry->avail_ssize = stack_entry->avail_ssize -
3588 stack_entry->eflags &= ~MAP_ENTRY_GROWS_DOWN;
3589 new_entry->eflags |= MAP_ENTRY_GROWS_DOWN;
3595 addr = stack_entry->end + grow_amount;
3598 * If this puts us into the next entry, cut back our growth
3599 * to the available space. Also, see the note above.
3602 stack_entry->avail_ssize = end - stack_entry->end;
3604 if (stack_guard_page)
3608 grow_amount = addr - stack_entry->end;
3609 cred = stack_entry->cred;
3610 if (cred == NULL && stack_entry->object.vm_object != NULL)
3611 cred = stack_entry->object.vm_object->cred;
3612 if (cred != NULL && !swap_reserve_by_cred(grow_amount, cred))
3614 /* Grow the underlying object if applicable. */
3615 else if (stack_entry->object.vm_object == NULL ||
3616 vm_object_coalesce(stack_entry->object.vm_object,
3617 stack_entry->offset,
3618 (vm_size_t)(stack_entry->end - stack_entry->start),
3619 (vm_size_t)grow_amount, cred != NULL)) {
3620 map->size += (addr - stack_entry->end);
3621 /* Update the current entry. */
3622 stack_entry->end = addr;
3623 stack_entry->avail_ssize -= grow_amount;
3624 vm_map_entry_resize_free(map, stack_entry);
3627 if (next_entry != &map->header)
3628 vm_map_clip_start(map, next_entry, addr);
3633 if (rv == KERN_SUCCESS && is_procstack)
3634 vm->vm_ssize += btoc(grow_amount);
3639 * Heed the MAP_WIREFUTURE flag if it was set for this process.
3641 if (rv == KERN_SUCCESS && (map->flags & MAP_WIREFUTURE)) {
3643 (stack_entry == next_entry) ? addr : addr - grow_amount,
3644 (stack_entry == next_entry) ? stack_entry->start : addr,
3645 (p->p_flag & P_SYSTEM)
3646 ? VM_MAP_WIRE_SYSTEM|VM_MAP_WIRE_NOHOLES
3647 : VM_MAP_WIRE_USER|VM_MAP_WIRE_NOHOLES);
3652 if (rv != KERN_SUCCESS) {
3654 error = racct_set(p, RACCT_VMEM, map->size);
3655 KASSERT(error == 0, ("decreasing RACCT_VMEM failed"));
3657 error = racct_set(p, RACCT_MEMLOCK,
3658 ptoa(pmap_wired_count(map->pmap)));
3659 KASSERT(error == 0, ("decreasing RACCT_MEMLOCK failed"));
3661 error = racct_set(p, RACCT_STACK, ctob(vm->vm_ssize));
3662 KASSERT(error == 0, ("decreasing RACCT_STACK failed"));
3671 * Unshare the specified VM space for exec. If other processes are
3672 * mapped to it, then create a new one. The new vmspace is null.
3675 vmspace_exec(struct proc *p, vm_offset_t minuser, vm_offset_t maxuser)
3677 struct vmspace *oldvmspace = p->p_vmspace;
3678 struct vmspace *newvmspace;
3680 newvmspace = vmspace_alloc(minuser, maxuser);
3681 if (newvmspace == NULL)
3683 newvmspace->vm_swrss = oldvmspace->vm_swrss;
3685 * This code is written like this for prototype purposes. The
3686 * goal is to avoid running down the vmspace here, but let the
3687 * other process's that are still using the vmspace to finally
3688 * run it down. Even though there is little or no chance of blocking
3689 * here, it is a good idea to keep this form for future mods.
3691 PROC_VMSPACE_LOCK(p);
3692 p->p_vmspace = newvmspace;
3693 PROC_VMSPACE_UNLOCK(p);
3694 if (p == curthread->td_proc)
3695 pmap_activate(curthread);
3696 vmspace_free(oldvmspace);
3701 * Unshare the specified VM space for forcing COW. This
3702 * is called by rfork, for the (RFMEM|RFPROC) == 0 case.
3705 vmspace_unshare(struct proc *p)
3707 struct vmspace *oldvmspace = p->p_vmspace;
3708 struct vmspace *newvmspace;
3709 vm_ooffset_t fork_charge;
3711 if (oldvmspace->vm_refcnt == 1)
3714 newvmspace = vmspace_fork(oldvmspace, &fork_charge);
3715 if (newvmspace == NULL)
3717 if (!swap_reserve_by_cred(fork_charge, p->p_ucred)) {
3718 vmspace_free(newvmspace);
3721 PROC_VMSPACE_LOCK(p);
3722 p->p_vmspace = newvmspace;
3723 PROC_VMSPACE_UNLOCK(p);
3724 if (p == curthread->td_proc)
3725 pmap_activate(curthread);
3726 vmspace_free(oldvmspace);
3733 * Finds the VM object, offset, and
3734 * protection for a given virtual address in the
3735 * specified map, assuming a page fault of the
3738 * Leaves the map in question locked for read; return
3739 * values are guaranteed until a vm_map_lookup_done
3740 * call is performed. Note that the map argument
3741 * is in/out; the returned map must be used in
3742 * the call to vm_map_lookup_done.
3744 * A handle (out_entry) is returned for use in
3745 * vm_map_lookup_done, to make that fast.
3747 * If a lookup is requested with "write protection"
3748 * specified, the map may be changed to perform virtual
3749 * copying operations, although the data referenced will
3753 vm_map_lookup(vm_map_t *var_map, /* IN/OUT */
3755 vm_prot_t fault_typea,
3756 vm_map_entry_t *out_entry, /* OUT */
3757 vm_object_t *object, /* OUT */
3758 vm_pindex_t *pindex, /* OUT */
3759 vm_prot_t *out_prot, /* OUT */
3760 boolean_t *wired) /* OUT */
3762 vm_map_entry_t entry;
3763 vm_map_t map = *var_map;
3765 vm_prot_t fault_type = fault_typea;
3766 vm_object_t eobject;
3772 vm_map_lock_read(map);
3775 * Lookup the faulting address.
3777 if (!vm_map_lookup_entry(map, vaddr, out_entry)) {
3778 vm_map_unlock_read(map);
3779 return (KERN_INVALID_ADDRESS);
3787 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
3788 vm_map_t old_map = map;
3790 *var_map = map = entry->object.sub_map;
3791 vm_map_unlock_read(old_map);
3796 * Check whether this task is allowed to have this page.
3798 prot = entry->protection;
3799 fault_type &= (VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE);
3800 if ((fault_type & prot) != fault_type || prot == VM_PROT_NONE) {
3801 vm_map_unlock_read(map);
3802 return (KERN_PROTECTION_FAILURE);
3804 if ((entry->eflags & MAP_ENTRY_USER_WIRED) &&
3805 (entry->eflags & MAP_ENTRY_COW) &&
3806 (fault_type & VM_PROT_WRITE)) {
3807 vm_map_unlock_read(map);
3808 return (KERN_PROTECTION_FAILURE);
3812 * If this page is not pageable, we have to get it for all possible
3815 *wired = (entry->wired_count != 0);
3817 fault_type = entry->protection;
3818 size = entry->end - entry->start;
3820 * If the entry was copy-on-write, we either ...
3822 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
3824 * If we want to write the page, we may as well handle that
3825 * now since we've got the map locked.
3827 * If we don't need to write the page, we just demote the
3828 * permissions allowed.
3830 if ((fault_type & VM_PROT_WRITE) != 0 ||
3831 (fault_typea & VM_PROT_COPY) != 0) {
3833 * Make a new object, and place it in the object
3834 * chain. Note that no new references have appeared
3835 * -- one just moved from the map to the new
3838 if (vm_map_lock_upgrade(map))
3841 if (entry->cred == NULL) {
3843 * The debugger owner is charged for
3846 cred = curthread->td_ucred;
3848 if (!swap_reserve_by_cred(size, cred)) {
3851 return (KERN_RESOURCE_SHORTAGE);
3855 vm_object_shadow(&entry->object.vm_object,
3856 &entry->offset, size);
3857 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
3858 eobject = entry->object.vm_object;
3859 if (eobject->cred != NULL) {
3861 * The object was not shadowed.
3863 swap_release_by_cred(size, entry->cred);
3864 crfree(entry->cred);
3866 } else if (entry->cred != NULL) {
3867 VM_OBJECT_WLOCK(eobject);
3868 eobject->cred = entry->cred;
3869 eobject->charge = size;
3870 VM_OBJECT_WUNLOCK(eobject);
3874 vm_map_lock_downgrade(map);
3877 * We're attempting to read a copy-on-write page --
3878 * don't allow writes.
3880 prot &= ~VM_PROT_WRITE;
3885 * Create an object if necessary.
3887 if (entry->object.vm_object == NULL &&
3889 if (vm_map_lock_upgrade(map))
3891 entry->object.vm_object = vm_object_allocate(OBJT_DEFAULT,
3894 if (entry->cred != NULL) {
3895 VM_OBJECT_WLOCK(entry->object.vm_object);
3896 entry->object.vm_object->cred = entry->cred;
3897 entry->object.vm_object->charge = size;
3898 VM_OBJECT_WUNLOCK(entry->object.vm_object);
3901 vm_map_lock_downgrade(map);
3905 * Return the object/offset from this entry. If the entry was
3906 * copy-on-write or empty, it has been fixed up.
3908 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
3909 *object = entry->object.vm_object;
3912 return (KERN_SUCCESS);
3916 * vm_map_lookup_locked:
3918 * Lookup the faulting address. A version of vm_map_lookup that returns
3919 * KERN_FAILURE instead of blocking on map lock or memory allocation.
3922 vm_map_lookup_locked(vm_map_t *var_map, /* IN/OUT */
3924 vm_prot_t fault_typea,
3925 vm_map_entry_t *out_entry, /* OUT */
3926 vm_object_t *object, /* OUT */
3927 vm_pindex_t *pindex, /* OUT */
3928 vm_prot_t *out_prot, /* OUT */
3929 boolean_t *wired) /* OUT */
3931 vm_map_entry_t entry;
3932 vm_map_t map = *var_map;
3934 vm_prot_t fault_type = fault_typea;
3937 * Lookup the faulting address.
3939 if (!vm_map_lookup_entry(map, vaddr, out_entry))
3940 return (KERN_INVALID_ADDRESS);
3945 * Fail if the entry refers to a submap.
3947 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP)
3948 return (KERN_FAILURE);
3951 * Check whether this task is allowed to have this page.
3953 prot = entry->protection;
3954 fault_type &= VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE;
3955 if ((fault_type & prot) != fault_type)
3956 return (KERN_PROTECTION_FAILURE);
3957 if ((entry->eflags & MAP_ENTRY_USER_WIRED) &&
3958 (entry->eflags & MAP_ENTRY_COW) &&
3959 (fault_type & VM_PROT_WRITE))
3960 return (KERN_PROTECTION_FAILURE);
3963 * If this page is not pageable, we have to get it for all possible
3966 *wired = (entry->wired_count != 0);
3968 fault_type = entry->protection;
3970 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
3972 * Fail if the entry was copy-on-write for a write fault.
3974 if (fault_type & VM_PROT_WRITE)
3975 return (KERN_FAILURE);
3977 * We're attempting to read a copy-on-write page --
3978 * don't allow writes.
3980 prot &= ~VM_PROT_WRITE;
3984 * Fail if an object should be created.
3986 if (entry->object.vm_object == NULL && !map->system_map)
3987 return (KERN_FAILURE);
3990 * Return the object/offset from this entry. If the entry was
3991 * copy-on-write or empty, it has been fixed up.
3993 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
3994 *object = entry->object.vm_object;
3997 return (KERN_SUCCESS);
4001 * vm_map_lookup_done:
4003 * Releases locks acquired by a vm_map_lookup
4004 * (according to the handle returned by that lookup).
4007 vm_map_lookup_done(vm_map_t map, vm_map_entry_t entry)
4010 * Unlock the main-level map
4012 vm_map_unlock_read(map);
4015 #include "opt_ddb.h"
4017 #include <sys/kernel.h>
4019 #include <ddb/ddb.h>
4022 vm_map_print(vm_map_t map)
4024 vm_map_entry_t entry;
4026 db_iprintf("Task map %p: pmap=%p, nentries=%d, version=%u\n",
4028 (void *)map->pmap, map->nentries, map->timestamp);
4031 for (entry = map->header.next; entry != &map->header;
4032 entry = entry->next) {
4033 db_iprintf("map entry %p: start=%p, end=%p\n",
4034 (void *)entry, (void *)entry->start, (void *)entry->end);
4036 static char *inheritance_name[4] =
4037 {"share", "copy", "none", "donate_copy"};
4039 db_iprintf(" prot=%x/%x/%s",
4041 entry->max_protection,
4042 inheritance_name[(int)(unsigned char)entry->inheritance]);
4043 if (entry->wired_count != 0)
4044 db_printf(", wired");
4046 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
4047 db_printf(", share=%p, offset=0x%jx\n",
4048 (void *)entry->object.sub_map,
4049 (uintmax_t)entry->offset);
4050 if ((entry->prev == &map->header) ||
4051 (entry->prev->object.sub_map !=
4052 entry->object.sub_map)) {
4054 vm_map_print((vm_map_t)entry->object.sub_map);
4058 if (entry->cred != NULL)
4059 db_printf(", ruid %d", entry->cred->cr_ruid);
4060 db_printf(", object=%p, offset=0x%jx",
4061 (void *)entry->object.vm_object,
4062 (uintmax_t)entry->offset);
4063 if (entry->object.vm_object && entry->object.vm_object->cred)
4064 db_printf(", obj ruid %d charge %jx",
4065 entry->object.vm_object->cred->cr_ruid,
4066 (uintmax_t)entry->object.vm_object->charge);
4067 if (entry->eflags & MAP_ENTRY_COW)
4068 db_printf(", copy (%s)",
4069 (entry->eflags & MAP_ENTRY_NEEDS_COPY) ? "needed" : "done");
4072 if ((entry->prev == &map->header) ||
4073 (entry->prev->object.vm_object !=
4074 entry->object.vm_object)) {
4076 vm_object_print((db_expr_t)(intptr_t)
4077 entry->object.vm_object,
4086 DB_SHOW_COMMAND(map, map)
4090 db_printf("usage: show map <addr>\n");
4093 vm_map_print((vm_map_t)addr);
4096 DB_SHOW_COMMAND(procvm, procvm)
4101 p = (struct proc *) addr;
4106 db_printf("p = %p, vmspace = %p, map = %p, pmap = %p\n",
4107 (void *)p, (void *)p->p_vmspace, (void *)&p->p_vmspace->vm_map,
4108 (void *)vmspace_pmap(p->p_vmspace));
4110 vm_map_print((vm_map_t)&p->p_vmspace->vm_map);