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>
81 #include <sys/sysctl.h>
82 #include <sys/sysent.h>
86 #include <vm/vm_param.h>
88 #include <vm/vm_map.h>
89 #include <vm/vm_page.h>
90 #include <vm/vm_object.h>
91 #include <vm/vm_pager.h>
92 #include <vm/vm_kern.h>
93 #include <vm/vm_extern.h>
94 #include <vm/vnode_pager.h>
95 #include <vm/swap_pager.h>
99 * Virtual memory maps provide for the mapping, protection,
100 * and sharing of virtual memory objects. In addition,
101 * this module provides for an efficient virtual copy of
102 * memory from one map to another.
104 * Synchronization is required prior to most operations.
106 * Maps consist of an ordered doubly-linked list of simple
107 * entries; a self-adjusting binary search tree of these
108 * entries is used to speed up lookups.
110 * Since portions of maps are specified by start/end addresses,
111 * which may not align with existing map entries, all
112 * routines merely "clip" entries to these start/end values.
113 * [That is, an entry is split into two, bordering at a
114 * start or end value.] Note that these clippings may not
115 * always be necessary (as the two resulting entries are then
116 * not changed); however, the clipping is done for convenience.
118 * As mentioned above, virtual copy operations are performed
119 * by copying VM object references from one map to
120 * another, and then marking both regions as copy-on-write.
123 static struct mtx map_sleep_mtx;
124 static uma_zone_t mapentzone;
125 static uma_zone_t kmapentzone;
126 static uma_zone_t mapzone;
127 static uma_zone_t vmspace_zone;
128 static struct vm_object kmapentobj;
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_set_obj(kmapentzone, &kmapentobj, 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_LOCK(object);
1227 if (object->ref_count > 1 || object->shadow_count != 0)
1228 vm_object_clear_flag(object, OBJ_ONEMAPPING);
1229 VM_OBJECT_UNLOCK(object);
1231 else if ((prev_entry != &map->header) &&
1232 (prev_entry->eflags == protoeflags) &&
1233 (cow & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP)) == 0 &&
1234 (prev_entry->end == start) &&
1235 (prev_entry->wired_count == 0) &&
1236 (prev_entry->cred == cred ||
1237 (prev_entry->object.vm_object != NULL &&
1238 (prev_entry->object.vm_object->cred == cred))) &&
1239 vm_object_coalesce(prev_entry->object.vm_object,
1241 (vm_size_t)(prev_entry->end - prev_entry->start),
1242 (vm_size_t)(end - prev_entry->end), charge_prev_obj)) {
1244 * We were able to extend the object. Determine if we
1245 * can extend the previous map entry to include the
1246 * new range as well.
1248 if ((prev_entry->inheritance == inheritance) &&
1249 (prev_entry->protection == prot) &&
1250 (prev_entry->max_protection == max)) {
1251 map->size += (end - prev_entry->end);
1252 prev_entry->end = end;
1253 vm_map_entry_resize_free(map, prev_entry);
1254 vm_map_simplify_entry(map, prev_entry);
1257 return (KERN_SUCCESS);
1261 * If we can extend the object but cannot extend the
1262 * map entry, we have to create a new map entry. We
1263 * must bump the ref count on the extended object to
1264 * account for it. object may be NULL.
1266 object = prev_entry->object.vm_object;
1267 offset = prev_entry->offset +
1268 (prev_entry->end - prev_entry->start);
1269 vm_object_reference(object);
1270 if (cred != NULL && object != NULL && object->cred != NULL &&
1271 !(prev_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
1272 /* Object already accounts for this uid. */
1279 * NOTE: if conditionals fail, object can be NULL here. This occurs
1280 * in things like the buffer map where we manage kva but do not manage
1285 * Create a new entry
1287 new_entry = vm_map_entry_create(map);
1288 new_entry->start = start;
1289 new_entry->end = end;
1290 new_entry->cred = NULL;
1292 new_entry->eflags = protoeflags;
1293 new_entry->object.vm_object = object;
1294 new_entry->offset = offset;
1295 new_entry->avail_ssize = 0;
1297 new_entry->inheritance = inheritance;
1298 new_entry->protection = prot;
1299 new_entry->max_protection = max;
1300 new_entry->wired_count = 0;
1301 new_entry->wiring_thread = NULL;
1302 new_entry->read_ahead = VM_FAULT_READ_AHEAD_INIT;
1303 new_entry->next_read = OFF_TO_IDX(offset);
1305 KASSERT(cred == NULL || !ENTRY_CHARGED(new_entry),
1306 ("OVERCOMMIT: vm_map_insert leaks vm_map %p", new_entry));
1307 new_entry->cred = cred;
1310 * Insert the new entry into the list
1312 vm_map_entry_link(map, prev_entry, new_entry);
1313 map->size += new_entry->end - new_entry->start;
1316 * It may be possible to merge the new entry with the next and/or
1317 * previous entries. However, due to MAP_STACK_* being a hack, a
1318 * panic can result from merging such entries.
1320 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0)
1321 vm_map_simplify_entry(map, new_entry);
1323 if (cow & (MAP_PREFAULT|MAP_PREFAULT_PARTIAL)) {
1324 vm_map_pmap_enter(map, start, prot,
1325 object, OFF_TO_IDX(offset), end - start,
1326 cow & MAP_PREFAULT_PARTIAL);
1329 return (KERN_SUCCESS);
1335 * Find the first fit (lowest VM address) for "length" free bytes
1336 * beginning at address >= start in the given map.
1338 * In a vm_map_entry, "adj_free" is the amount of free space
1339 * adjacent (higher address) to this entry, and "max_free" is the
1340 * maximum amount of contiguous free space in its subtree. This
1341 * allows finding a free region in one path down the tree, so
1342 * O(log n) amortized with splay trees.
1344 * The map must be locked, and leaves it so.
1346 * Returns: 0 on success, and starting address in *addr,
1347 * 1 if insufficient space.
1350 vm_map_findspace(vm_map_t map, vm_offset_t start, vm_size_t length,
1351 vm_offset_t *addr) /* OUT */
1353 vm_map_entry_t entry;
1357 * Request must fit within min/max VM address and must avoid
1360 if (start < map->min_offset)
1361 start = map->min_offset;
1362 if (start + length > map->max_offset || start + length < start)
1365 /* Empty tree means wide open address space. */
1366 if (map->root == NULL) {
1372 * After splay, if start comes before root node, then there
1373 * must be a gap from start to the root.
1375 map->root = vm_map_entry_splay(start, map->root);
1376 if (start + length <= map->root->start) {
1382 * Root is the last node that might begin its gap before
1383 * start, and this is the last comparison where address
1384 * wrap might be a problem.
1386 st = (start > map->root->end) ? start : map->root->end;
1387 if (length <= map->root->end + map->root->adj_free - st) {
1392 /* With max_free, can immediately tell if no solution. */
1393 entry = map->root->right;
1394 if (entry == NULL || length > entry->max_free)
1398 * Search the right subtree in the order: left subtree, root,
1399 * right subtree (first fit). The previous splay implies that
1400 * all regions in the right subtree have addresses > start.
1402 while (entry != NULL) {
1403 if (entry->left != NULL && entry->left->max_free >= length)
1404 entry = entry->left;
1405 else if (entry->adj_free >= length) {
1409 entry = entry->right;
1412 /* Can't get here, so panic if we do. */
1413 panic("vm_map_findspace: max_free corrupt");
1417 vm_map_fixed(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1418 vm_offset_t start, vm_size_t length, vm_prot_t prot,
1419 vm_prot_t max, int cow)
1424 end = start + length;
1426 VM_MAP_RANGE_CHECK(map, start, end);
1427 (void) vm_map_delete(map, start, end);
1428 result = vm_map_insert(map, object, offset, start, end, prot,
1435 * vm_map_find finds an unallocated region in the target address
1436 * map with the given length. The search is defined to be
1437 * first-fit from the specified address; the region found is
1438 * returned in the same parameter.
1440 * If object is non-NULL, ref count must be bumped by caller
1441 * prior to making call to account for the new entry.
1444 vm_map_find(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1445 vm_offset_t *addr, /* IN/OUT */
1446 vm_size_t length, int find_space, vm_prot_t prot,
1447 vm_prot_t max, int cow)
1449 vm_offset_t alignment, initial_addr, start;
1452 if (find_space == VMFS_OPTIMAL_SPACE && (object == NULL ||
1453 (object->flags & OBJ_COLORED) == 0))
1454 find_space = VMFS_ANY_SPACE;
1455 if (find_space >> 8 != 0) {
1456 KASSERT((find_space & 0xff) == 0, ("bad VMFS flags"));
1457 alignment = (vm_offset_t)1 << (find_space >> 8);
1460 initial_addr = *addr;
1462 start = initial_addr;
1465 if (find_space != VMFS_NO_SPACE) {
1466 if (vm_map_findspace(map, start, length, addr)) {
1468 if (find_space == VMFS_OPTIMAL_SPACE) {
1469 find_space = VMFS_ANY_SPACE;
1472 return (KERN_NO_SPACE);
1474 switch (find_space) {
1475 case VMFS_SUPER_SPACE:
1476 case VMFS_OPTIMAL_SPACE:
1477 pmap_align_superpage(object, offset, addr,
1480 #ifdef VMFS_TLB_ALIGNED_SPACE
1481 case VMFS_TLB_ALIGNED_SPACE:
1482 pmap_align_tlb(addr);
1485 case VMFS_ANY_SPACE:
1488 if ((*addr & (alignment - 1)) != 0) {
1489 *addr &= ~(alignment - 1);
1497 result = vm_map_insert(map, object, offset, start, start +
1498 length, prot, max, cow);
1499 } while (result == KERN_NO_SPACE && find_space != VMFS_NO_SPACE &&
1500 find_space != VMFS_ANY_SPACE);
1506 * vm_map_simplify_entry:
1508 * Simplify the given map entry by merging with either neighbor. This
1509 * routine also has the ability to merge with both neighbors.
1511 * The map must be locked.
1513 * This routine guarentees that the passed entry remains valid (though
1514 * possibly extended). When merging, this routine may delete one or
1518 vm_map_simplify_entry(vm_map_t map, vm_map_entry_t entry)
1520 vm_map_entry_t next, prev;
1521 vm_size_t prevsize, esize;
1523 if (entry->eflags & (MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_IS_SUB_MAP))
1527 if (prev != &map->header) {
1528 prevsize = prev->end - prev->start;
1529 if ( (prev->end == entry->start) &&
1530 (prev->object.vm_object == entry->object.vm_object) &&
1531 (!prev->object.vm_object ||
1532 (prev->offset + prevsize == entry->offset)) &&
1533 (prev->eflags == entry->eflags) &&
1534 (prev->protection == entry->protection) &&
1535 (prev->max_protection == entry->max_protection) &&
1536 (prev->inheritance == entry->inheritance) &&
1537 (prev->wired_count == entry->wired_count) &&
1538 (prev->cred == entry->cred)) {
1539 vm_map_entry_unlink(map, prev);
1540 entry->start = prev->start;
1541 entry->offset = prev->offset;
1542 if (entry->prev != &map->header)
1543 vm_map_entry_resize_free(map, entry->prev);
1546 * If the backing object is a vnode object,
1547 * vm_object_deallocate() calls vrele().
1548 * However, vrele() does not lock the vnode
1549 * because the vnode has additional
1550 * references. Thus, the map lock can be kept
1551 * without causing a lock-order reversal with
1554 * Since we count the number of virtual page
1555 * mappings in object->un_pager.vnp.writemappings,
1556 * the writemappings value should not be adjusted
1557 * when the entry is disposed of.
1559 if (prev->object.vm_object)
1560 vm_object_deallocate(prev->object.vm_object);
1561 if (prev->cred != NULL)
1563 vm_map_entry_dispose(map, prev);
1568 if (next != &map->header) {
1569 esize = entry->end - entry->start;
1570 if ((entry->end == next->start) &&
1571 (next->object.vm_object == entry->object.vm_object) &&
1572 (!entry->object.vm_object ||
1573 (entry->offset + esize == next->offset)) &&
1574 (next->eflags == entry->eflags) &&
1575 (next->protection == entry->protection) &&
1576 (next->max_protection == entry->max_protection) &&
1577 (next->inheritance == entry->inheritance) &&
1578 (next->wired_count == entry->wired_count) &&
1579 (next->cred == entry->cred)) {
1580 vm_map_entry_unlink(map, next);
1581 entry->end = next->end;
1582 vm_map_entry_resize_free(map, entry);
1585 * See comment above.
1587 if (next->object.vm_object)
1588 vm_object_deallocate(next->object.vm_object);
1589 if (next->cred != NULL)
1591 vm_map_entry_dispose(map, next);
1596 * vm_map_clip_start: [ internal use only ]
1598 * Asserts that the given entry begins at or after
1599 * the specified address; if necessary,
1600 * it splits the entry into two.
1602 #define vm_map_clip_start(map, entry, startaddr) \
1604 if (startaddr > entry->start) \
1605 _vm_map_clip_start(map, entry, startaddr); \
1609 * This routine is called only when it is known that
1610 * the entry must be split.
1613 _vm_map_clip_start(vm_map_t map, vm_map_entry_t entry, vm_offset_t start)
1615 vm_map_entry_t new_entry;
1617 VM_MAP_ASSERT_LOCKED(map);
1620 * Split off the front portion -- note that we must insert the new
1621 * entry BEFORE this one, so that this entry has the specified
1624 vm_map_simplify_entry(map, entry);
1627 * If there is no object backing this entry, we might as well create
1628 * one now. If we defer it, an object can get created after the map
1629 * is clipped, and individual objects will be created for the split-up
1630 * map. This is a bit of a hack, but is also about the best place to
1631 * put this improvement.
1633 if (entry->object.vm_object == NULL && !map->system_map) {
1635 object = vm_object_allocate(OBJT_DEFAULT,
1636 atop(entry->end - entry->start));
1637 entry->object.vm_object = object;
1639 if (entry->cred != NULL) {
1640 object->cred = entry->cred;
1641 object->charge = entry->end - entry->start;
1644 } else if (entry->object.vm_object != NULL &&
1645 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) &&
1646 entry->cred != NULL) {
1647 VM_OBJECT_LOCK(entry->object.vm_object);
1648 KASSERT(entry->object.vm_object->cred == NULL,
1649 ("OVERCOMMIT: vm_entry_clip_start: both cred e %p", entry));
1650 entry->object.vm_object->cred = entry->cred;
1651 entry->object.vm_object->charge = entry->end - entry->start;
1652 VM_OBJECT_UNLOCK(entry->object.vm_object);
1656 new_entry = vm_map_entry_create(map);
1657 *new_entry = *entry;
1659 new_entry->end = start;
1660 entry->offset += (start - entry->start);
1661 entry->start = start;
1662 if (new_entry->cred != NULL)
1663 crhold(entry->cred);
1665 vm_map_entry_link(map, entry->prev, new_entry);
1667 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1668 vm_object_reference(new_entry->object.vm_object);
1670 * The object->un_pager.vnp.writemappings for the
1671 * object of MAP_ENTRY_VN_WRITECNT type entry shall be
1672 * kept as is here. The virtual pages are
1673 * re-distributed among the clipped entries, so the sum is
1680 * vm_map_clip_end: [ internal use only ]
1682 * Asserts that the given entry ends at or before
1683 * the specified address; if necessary,
1684 * it splits the entry into two.
1686 #define vm_map_clip_end(map, entry, endaddr) \
1688 if ((endaddr) < (entry->end)) \
1689 _vm_map_clip_end((map), (entry), (endaddr)); \
1693 * This routine is called only when it is known that
1694 * the entry must be split.
1697 _vm_map_clip_end(vm_map_t map, vm_map_entry_t entry, vm_offset_t end)
1699 vm_map_entry_t new_entry;
1701 VM_MAP_ASSERT_LOCKED(map);
1704 * If there is no object backing this entry, we might as well create
1705 * one now. If we defer it, an object can get created after the map
1706 * is clipped, and individual objects will be created for the split-up
1707 * map. This is a bit of a hack, but is also about the best place to
1708 * put this improvement.
1710 if (entry->object.vm_object == NULL && !map->system_map) {
1712 object = vm_object_allocate(OBJT_DEFAULT,
1713 atop(entry->end - entry->start));
1714 entry->object.vm_object = object;
1716 if (entry->cred != NULL) {
1717 object->cred = entry->cred;
1718 object->charge = entry->end - entry->start;
1721 } else if (entry->object.vm_object != NULL &&
1722 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) &&
1723 entry->cred != NULL) {
1724 VM_OBJECT_LOCK(entry->object.vm_object);
1725 KASSERT(entry->object.vm_object->cred == NULL,
1726 ("OVERCOMMIT: vm_entry_clip_end: both cred e %p", entry));
1727 entry->object.vm_object->cred = entry->cred;
1728 entry->object.vm_object->charge = entry->end - entry->start;
1729 VM_OBJECT_UNLOCK(entry->object.vm_object);
1734 * Create a new entry and insert it AFTER the specified entry
1736 new_entry = vm_map_entry_create(map);
1737 *new_entry = *entry;
1739 new_entry->start = entry->end = end;
1740 new_entry->offset += (end - entry->start);
1741 if (new_entry->cred != NULL)
1742 crhold(entry->cred);
1744 vm_map_entry_link(map, entry, new_entry);
1746 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1747 vm_object_reference(new_entry->object.vm_object);
1752 * vm_map_submap: [ kernel use only ]
1754 * Mark the given range as handled by a subordinate map.
1756 * This range must have been created with vm_map_find,
1757 * and no other operations may have been performed on this
1758 * range prior to calling vm_map_submap.
1760 * Only a limited number of operations can be performed
1761 * within this rage after calling vm_map_submap:
1763 * [Don't try vm_map_copy!]
1765 * To remove a submapping, one must first remove the
1766 * range from the superior map, and then destroy the
1767 * submap (if desired). [Better yet, don't try it.]
1776 vm_map_entry_t entry;
1777 int result = KERN_INVALID_ARGUMENT;
1781 VM_MAP_RANGE_CHECK(map, start, end);
1783 if (vm_map_lookup_entry(map, start, &entry)) {
1784 vm_map_clip_start(map, entry, start);
1786 entry = entry->next;
1788 vm_map_clip_end(map, entry, end);
1790 if ((entry->start == start) && (entry->end == end) &&
1791 ((entry->eflags & MAP_ENTRY_COW) == 0) &&
1792 (entry->object.vm_object == NULL)) {
1793 entry->object.sub_map = submap;
1794 entry->eflags |= MAP_ENTRY_IS_SUB_MAP;
1795 result = KERN_SUCCESS;
1803 * The maximum number of pages to map
1805 #define MAX_INIT_PT 96
1808 * vm_map_pmap_enter:
1810 * Preload read-only mappings for the given object's resident pages into
1811 * the given map. This eliminates the soft faults on process startup and
1812 * immediately after an mmap(2). Because these are speculative mappings,
1813 * cached pages are not reactivated and mapped.
1816 vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot,
1817 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags)
1820 vm_page_t p, p_start;
1821 vm_pindex_t psize, tmpidx;
1823 if ((prot & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0 || object == NULL)
1825 VM_OBJECT_LOCK(object);
1826 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) {
1827 pmap_object_init_pt(map->pmap, addr, object, pindex, size);
1833 if ((flags & MAP_PREFAULT_PARTIAL) && psize > MAX_INIT_PT &&
1834 object->resident_page_count > MAX_INIT_PT)
1837 if (psize + pindex > object->size) {
1838 if (object->size < pindex)
1840 psize = object->size - pindex;
1846 p = vm_page_find_least(object, pindex);
1848 * Assert: the variable p is either (1) the page with the
1849 * least pindex greater than or equal to the parameter pindex
1853 p != NULL && (tmpidx = p->pindex - pindex) < psize;
1854 p = TAILQ_NEXT(p, listq)) {
1856 * don't allow an madvise to blow away our really
1857 * free pages allocating pv entries.
1859 if ((flags & MAP_PREFAULT_MADVISE) &&
1860 cnt.v_free_count < cnt.v_free_reserved) {
1864 if (p->valid == VM_PAGE_BITS_ALL) {
1865 if (p_start == NULL) {
1866 start = addr + ptoa(tmpidx);
1869 } else if (p_start != NULL) {
1870 pmap_enter_object(map->pmap, start, addr +
1871 ptoa(tmpidx), p_start, prot);
1875 if (p_start != NULL)
1876 pmap_enter_object(map->pmap, start, addr + ptoa(psize),
1879 VM_OBJECT_UNLOCK(object);
1885 * Sets the protection of the specified address
1886 * region in the target map. If "set_max" is
1887 * specified, the maximum protection is to be set;
1888 * otherwise, only the current protection is affected.
1891 vm_map_protect(vm_map_t map, vm_offset_t start, vm_offset_t end,
1892 vm_prot_t new_prot, boolean_t set_max)
1894 vm_map_entry_t current, entry;
1900 return (KERN_SUCCESS);
1904 VM_MAP_RANGE_CHECK(map, start, end);
1906 if (vm_map_lookup_entry(map, start, &entry)) {
1907 vm_map_clip_start(map, entry, start);
1909 entry = entry->next;
1913 * Make a first pass to check for protection violations.
1916 while ((current != &map->header) && (current->start < end)) {
1917 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
1919 return (KERN_INVALID_ARGUMENT);
1921 if ((new_prot & current->max_protection) != new_prot) {
1923 return (KERN_PROTECTION_FAILURE);
1925 current = current->next;
1930 * Do an accounting pass for private read-only mappings that
1931 * now will do cow due to allowed write (e.g. debugger sets
1932 * breakpoint on text segment)
1934 for (current = entry; (current != &map->header) &&
1935 (current->start < end); current = current->next) {
1937 vm_map_clip_end(map, current, end);
1940 ((new_prot & ~(current->protection)) & VM_PROT_WRITE) == 0 ||
1941 ENTRY_CHARGED(current)) {
1945 cred = curthread->td_ucred;
1946 obj = current->object.vm_object;
1948 if (obj == NULL || (current->eflags & MAP_ENTRY_NEEDS_COPY)) {
1949 if (!swap_reserve(current->end - current->start)) {
1951 return (KERN_RESOURCE_SHORTAGE);
1954 current->cred = cred;
1958 VM_OBJECT_LOCK(obj);
1959 if (obj->type != OBJT_DEFAULT && obj->type != OBJT_SWAP) {
1960 VM_OBJECT_UNLOCK(obj);
1965 * Charge for the whole object allocation now, since
1966 * we cannot distinguish between non-charged and
1967 * charged clipped mapping of the same object later.
1969 KASSERT(obj->charge == 0,
1970 ("vm_map_protect: object %p overcharged (entry %p)",
1972 if (!swap_reserve(ptoa(obj->size))) {
1973 VM_OBJECT_UNLOCK(obj);
1975 return (KERN_RESOURCE_SHORTAGE);
1980 obj->charge = ptoa(obj->size);
1981 VM_OBJECT_UNLOCK(obj);
1985 * Go back and fix up protections. [Note that clipping is not
1986 * necessary the second time.]
1989 while ((current != &map->header) && (current->start < end)) {
1990 old_prot = current->protection;
1993 current->protection =
1994 (current->max_protection = new_prot) &
1997 current->protection = new_prot;
1999 if ((current->eflags & (MAP_ENTRY_COW | MAP_ENTRY_USER_WIRED))
2000 == (MAP_ENTRY_COW | MAP_ENTRY_USER_WIRED) &&
2001 (current->protection & VM_PROT_WRITE) != 0 &&
2002 (old_prot & VM_PROT_WRITE) == 0) {
2003 vm_fault_copy_entry(map, map, current, current, NULL);
2007 * When restricting access, update the physical map. Worry
2008 * about copy-on-write here.
2010 if ((old_prot & ~current->protection) != 0) {
2011 #define MASK(entry) (((entry)->eflags & MAP_ENTRY_COW) ? ~VM_PROT_WRITE : \
2013 pmap_protect(map->pmap, current->start,
2015 current->protection & MASK(current));
2018 vm_map_simplify_entry(map, current);
2019 current = current->next;
2022 return (KERN_SUCCESS);
2028 * This routine traverses a processes map handling the madvise
2029 * system call. Advisories are classified as either those effecting
2030 * the vm_map_entry structure, or those effecting the underlying
2040 vm_map_entry_t current, entry;
2044 * Some madvise calls directly modify the vm_map_entry, in which case
2045 * we need to use an exclusive lock on the map and we need to perform
2046 * various clipping operations. Otherwise we only need a read-lock
2051 case MADV_SEQUENTIAL:
2058 return (KERN_SUCCESS);
2066 return (KERN_SUCCESS);
2067 vm_map_lock_read(map);
2070 return (KERN_INVALID_ARGUMENT);
2074 * Locate starting entry and clip if necessary.
2076 VM_MAP_RANGE_CHECK(map, start, end);
2078 if (vm_map_lookup_entry(map, start, &entry)) {
2080 vm_map_clip_start(map, entry, start);
2082 entry = entry->next;
2087 * madvise behaviors that are implemented in the vm_map_entry.
2089 * We clip the vm_map_entry so that behavioral changes are
2090 * limited to the specified address range.
2092 for (current = entry;
2093 (current != &map->header) && (current->start < end);
2094 current = current->next
2096 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
2099 vm_map_clip_end(map, current, end);
2103 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_NORMAL);
2105 case MADV_SEQUENTIAL:
2106 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_SEQUENTIAL);
2109 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_RANDOM);
2112 current->eflags |= MAP_ENTRY_NOSYNC;
2115 current->eflags &= ~MAP_ENTRY_NOSYNC;
2118 current->eflags |= MAP_ENTRY_NOCOREDUMP;
2121 current->eflags &= ~MAP_ENTRY_NOCOREDUMP;
2126 vm_map_simplify_entry(map, current);
2130 vm_pindex_t pstart, pend;
2133 * madvise behaviors that are implemented in the underlying
2136 * Since we don't clip the vm_map_entry, we have to clip
2137 * the vm_object pindex and count.
2139 for (current = entry;
2140 (current != &map->header) && (current->start < end);
2141 current = current->next
2143 vm_offset_t useStart;
2145 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
2148 pstart = OFF_TO_IDX(current->offset);
2149 pend = pstart + atop(current->end - current->start);
2150 useStart = current->start;
2152 if (current->start < start) {
2153 pstart += atop(start - current->start);
2156 if (current->end > end)
2157 pend -= atop(current->end - end);
2162 vm_object_madvise(current->object.vm_object, pstart,
2164 if (behav == MADV_WILLNEED) {
2165 vm_map_pmap_enter(map,
2167 current->protection,
2168 current->object.vm_object,
2170 ptoa(pend - pstart),
2171 MAP_PREFAULT_MADVISE
2175 vm_map_unlock_read(map);
2184 * Sets the inheritance of the specified address
2185 * range in the target map. Inheritance
2186 * affects how the map will be shared with
2187 * child maps at the time of vmspace_fork.
2190 vm_map_inherit(vm_map_t map, vm_offset_t start, vm_offset_t end,
2191 vm_inherit_t new_inheritance)
2193 vm_map_entry_t entry;
2194 vm_map_entry_t temp_entry;
2196 switch (new_inheritance) {
2197 case VM_INHERIT_NONE:
2198 case VM_INHERIT_COPY:
2199 case VM_INHERIT_SHARE:
2202 return (KERN_INVALID_ARGUMENT);
2205 return (KERN_SUCCESS);
2207 VM_MAP_RANGE_CHECK(map, start, end);
2208 if (vm_map_lookup_entry(map, start, &temp_entry)) {
2210 vm_map_clip_start(map, entry, start);
2212 entry = temp_entry->next;
2213 while ((entry != &map->header) && (entry->start < end)) {
2214 vm_map_clip_end(map, entry, end);
2215 entry->inheritance = new_inheritance;
2216 vm_map_simplify_entry(map, entry);
2217 entry = entry->next;
2220 return (KERN_SUCCESS);
2226 * Implements both kernel and user unwiring.
2229 vm_map_unwire(vm_map_t map, vm_offset_t start, vm_offset_t end,
2232 vm_map_entry_t entry, first_entry, tmp_entry;
2233 vm_offset_t saved_start;
2234 unsigned int last_timestamp;
2236 boolean_t need_wakeup, result, user_unwire;
2239 return (KERN_SUCCESS);
2240 user_unwire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE;
2242 VM_MAP_RANGE_CHECK(map, start, end);
2243 if (!vm_map_lookup_entry(map, start, &first_entry)) {
2244 if (flags & VM_MAP_WIRE_HOLESOK)
2245 first_entry = first_entry->next;
2248 return (KERN_INVALID_ADDRESS);
2251 last_timestamp = map->timestamp;
2252 entry = first_entry;
2253 while (entry != &map->header && entry->start < end) {
2254 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
2256 * We have not yet clipped the entry.
2258 saved_start = (start >= entry->start) ? start :
2260 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2261 if (vm_map_unlock_and_wait(map, 0)) {
2263 * Allow interruption of user unwiring?
2267 if (last_timestamp+1 != map->timestamp) {
2269 * Look again for the entry because the map was
2270 * modified while it was unlocked.
2271 * Specifically, the entry may have been
2272 * clipped, merged, or deleted.
2274 if (!vm_map_lookup_entry(map, saved_start,
2276 if (flags & VM_MAP_WIRE_HOLESOK)
2277 tmp_entry = tmp_entry->next;
2279 if (saved_start == start) {
2281 * First_entry has been deleted.
2284 return (KERN_INVALID_ADDRESS);
2287 rv = KERN_INVALID_ADDRESS;
2291 if (entry == first_entry)
2292 first_entry = tmp_entry;
2297 last_timestamp = map->timestamp;
2300 vm_map_clip_start(map, entry, start);
2301 vm_map_clip_end(map, entry, end);
2303 * Mark the entry in case the map lock is released. (See
2306 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 &&
2307 entry->wiring_thread == NULL,
2308 ("owned map entry %p", entry));
2309 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
2310 entry->wiring_thread = curthread;
2312 * Check the map for holes in the specified region.
2313 * If VM_MAP_WIRE_HOLESOK was specified, skip this check.
2315 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) &&
2316 (entry->end < end && (entry->next == &map->header ||
2317 entry->next->start > entry->end))) {
2319 rv = KERN_INVALID_ADDRESS;
2323 * If system unwiring, require that the entry is system wired.
2326 vm_map_entry_system_wired_count(entry) == 0) {
2328 rv = KERN_INVALID_ARGUMENT;
2331 entry = entry->next;
2335 need_wakeup = FALSE;
2336 if (first_entry == NULL) {
2337 result = vm_map_lookup_entry(map, start, &first_entry);
2338 if (!result && (flags & VM_MAP_WIRE_HOLESOK))
2339 first_entry = first_entry->next;
2341 KASSERT(result, ("vm_map_unwire: lookup failed"));
2343 for (entry = first_entry; entry != &map->header && entry->start < end;
2344 entry = entry->next) {
2346 * If VM_MAP_WIRE_HOLESOK was specified, an empty
2347 * space in the unwired region could have been mapped
2348 * while the map lock was dropped for draining
2349 * MAP_ENTRY_IN_TRANSITION. Moreover, another thread
2350 * could be simultaneously wiring this new mapping
2351 * entry. Detect these cases and skip any entries
2352 * marked as in transition by us.
2354 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 ||
2355 entry->wiring_thread != curthread) {
2356 KASSERT((flags & VM_MAP_WIRE_HOLESOK) != 0,
2357 ("vm_map_unwire: !HOLESOK and new/changed entry"));
2361 if (rv == KERN_SUCCESS && (!user_unwire ||
2362 (entry->eflags & MAP_ENTRY_USER_WIRED))) {
2364 entry->eflags &= ~MAP_ENTRY_USER_WIRED;
2365 entry->wired_count--;
2366 if (entry->wired_count == 0) {
2368 * Retain the map lock.
2370 vm_fault_unwire(map, entry->start, entry->end,
2371 entry->object.vm_object != NULL &&
2372 (entry->object.vm_object->type == OBJT_DEVICE ||
2373 entry->object.vm_object->type == OBJT_SG));
2376 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0,
2377 ("vm_map_unwire: in-transition flag missing %p", entry));
2378 KASSERT(entry->wiring_thread == curthread,
2379 ("vm_map_unwire: alien wire %p", entry));
2380 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION;
2381 entry->wiring_thread = NULL;
2382 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
2383 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
2386 vm_map_simplify_entry(map, entry);
2397 * Implements both kernel and user wiring.
2400 vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t end,
2403 vm_map_entry_t entry, first_entry, tmp_entry;
2404 vm_offset_t saved_end, saved_start;
2405 unsigned int last_timestamp;
2407 boolean_t fictitious, need_wakeup, result, user_wire;
2411 return (KERN_SUCCESS);
2413 if (flags & VM_MAP_WIRE_WRITE)
2414 prot |= VM_PROT_WRITE;
2415 user_wire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE;
2417 VM_MAP_RANGE_CHECK(map, start, end);
2418 if (!vm_map_lookup_entry(map, start, &first_entry)) {
2419 if (flags & VM_MAP_WIRE_HOLESOK)
2420 first_entry = first_entry->next;
2423 return (KERN_INVALID_ADDRESS);
2426 last_timestamp = map->timestamp;
2427 entry = first_entry;
2428 while (entry != &map->header && entry->start < end) {
2429 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
2431 * We have not yet clipped the entry.
2433 saved_start = (start >= entry->start) ? start :
2435 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2436 if (vm_map_unlock_and_wait(map, 0)) {
2438 * Allow interruption of user wiring?
2442 if (last_timestamp + 1 != map->timestamp) {
2444 * Look again for the entry because the map was
2445 * modified while it was unlocked.
2446 * Specifically, the entry may have been
2447 * clipped, merged, or deleted.
2449 if (!vm_map_lookup_entry(map, saved_start,
2451 if (flags & VM_MAP_WIRE_HOLESOK)
2452 tmp_entry = tmp_entry->next;
2454 if (saved_start == start) {
2456 * first_entry has been deleted.
2459 return (KERN_INVALID_ADDRESS);
2462 rv = KERN_INVALID_ADDRESS;
2466 if (entry == first_entry)
2467 first_entry = tmp_entry;
2472 last_timestamp = map->timestamp;
2475 vm_map_clip_start(map, entry, start);
2476 vm_map_clip_end(map, entry, end);
2478 * Mark the entry in case the map lock is released. (See
2481 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 &&
2482 entry->wiring_thread == NULL,
2483 ("owned map entry %p", entry));
2484 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
2485 entry->wiring_thread = curthread;
2486 if ((entry->protection & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0
2487 || (entry->protection & prot) != prot) {
2488 entry->eflags |= MAP_ENTRY_WIRE_SKIPPED;
2489 if ((flags & VM_MAP_WIRE_HOLESOK) == 0) {
2491 rv = KERN_INVALID_ADDRESS;
2496 if (entry->wired_count == 0) {
2497 entry->wired_count++;
2498 saved_start = entry->start;
2499 saved_end = entry->end;
2500 fictitious = entry->object.vm_object != NULL &&
2501 (entry->object.vm_object->type == OBJT_DEVICE ||
2502 entry->object.vm_object->type == OBJT_SG);
2504 * Release the map lock, relying on the in-transition
2505 * mark. Mark the map busy for fork.
2509 rv = vm_fault_wire(map, saved_start, saved_end,
2513 if (last_timestamp + 1 != map->timestamp) {
2515 * Look again for the entry because the map was
2516 * modified while it was unlocked. The entry
2517 * may have been clipped, but NOT merged or
2520 result = vm_map_lookup_entry(map, saved_start,
2522 KASSERT(result, ("vm_map_wire: lookup failed"));
2523 if (entry == first_entry)
2524 first_entry = tmp_entry;
2528 while (entry->end < saved_end) {
2529 if (rv != KERN_SUCCESS) {
2530 KASSERT(entry->wired_count == 1,
2531 ("vm_map_wire: bad count"));
2532 entry->wired_count = -1;
2534 entry = entry->next;
2537 last_timestamp = map->timestamp;
2538 if (rv != KERN_SUCCESS) {
2539 KASSERT(entry->wired_count == 1,
2540 ("vm_map_wire: bad count"));
2542 * Assign an out-of-range value to represent
2543 * the failure to wire this entry.
2545 entry->wired_count = -1;
2549 } else if (!user_wire ||
2550 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
2551 entry->wired_count++;
2554 * Check the map for holes in the specified region.
2555 * If VM_MAP_WIRE_HOLESOK was specified, skip this check.
2558 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) &&
2559 (entry->end < end && (entry->next == &map->header ||
2560 entry->next->start > entry->end))) {
2562 rv = KERN_INVALID_ADDRESS;
2565 entry = entry->next;
2569 need_wakeup = FALSE;
2570 if (first_entry == NULL) {
2571 result = vm_map_lookup_entry(map, start, &first_entry);
2572 if (!result && (flags & VM_MAP_WIRE_HOLESOK))
2573 first_entry = first_entry->next;
2575 KASSERT(result, ("vm_map_wire: lookup failed"));
2577 for (entry = first_entry; entry != &map->header && entry->start < end;
2578 entry = entry->next) {
2579 if ((entry->eflags & MAP_ENTRY_WIRE_SKIPPED) != 0)
2580 goto next_entry_done;
2583 * If VM_MAP_WIRE_HOLESOK was specified, an empty
2584 * space in the unwired region could have been mapped
2585 * while the map lock was dropped for faulting in the
2586 * pages or draining MAP_ENTRY_IN_TRANSITION.
2587 * Moreover, another thread could be simultaneously
2588 * wiring this new mapping entry. Detect these cases
2589 * and skip any entries marked as in transition by us.
2591 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 ||
2592 entry->wiring_thread != curthread) {
2593 KASSERT((flags & VM_MAP_WIRE_HOLESOK) != 0,
2594 ("vm_map_wire: !HOLESOK and new/changed entry"));
2598 if (rv == KERN_SUCCESS) {
2600 entry->eflags |= MAP_ENTRY_USER_WIRED;
2601 } else if (entry->wired_count == -1) {
2603 * Wiring failed on this entry. Thus, unwiring is
2606 entry->wired_count = 0;
2609 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0)
2610 entry->wired_count--;
2611 if (entry->wired_count == 0) {
2613 * Retain the map lock.
2615 vm_fault_unwire(map, entry->start, entry->end,
2616 entry->object.vm_object != NULL &&
2617 (entry->object.vm_object->type == OBJT_DEVICE ||
2618 entry->object.vm_object->type == OBJT_SG));
2622 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0,
2623 ("vm_map_wire: in-transition flag missing %p", entry));
2624 KASSERT(entry->wiring_thread == curthread,
2625 ("vm_map_wire: alien wire %p", entry));
2626 entry->eflags &= ~(MAP_ENTRY_IN_TRANSITION |
2627 MAP_ENTRY_WIRE_SKIPPED);
2628 entry->wiring_thread = NULL;
2629 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
2630 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
2633 vm_map_simplify_entry(map, entry);
2644 * Push any dirty cached pages in the address range to their pager.
2645 * If syncio is TRUE, dirty pages are written synchronously.
2646 * If invalidate is TRUE, any cached pages are freed as well.
2648 * If the size of the region from start to end is zero, we are
2649 * supposed to flush all modified pages within the region containing
2650 * start. Unfortunately, a region can be split or coalesced with
2651 * neighboring regions, making it difficult to determine what the
2652 * original region was. Therefore, we approximate this requirement by
2653 * flushing the current region containing start.
2655 * Returns an error if any part of the specified range is not mapped.
2663 boolean_t invalidate)
2665 vm_map_entry_t current;
2666 vm_map_entry_t entry;
2669 vm_ooffset_t offset;
2670 unsigned int last_timestamp;
2673 vm_map_lock_read(map);
2674 VM_MAP_RANGE_CHECK(map, start, end);
2675 if (!vm_map_lookup_entry(map, start, &entry)) {
2676 vm_map_unlock_read(map);
2677 return (KERN_INVALID_ADDRESS);
2678 } else if (start == end) {
2679 start = entry->start;
2683 * Make a first pass to check for user-wired memory and holes.
2685 for (current = entry; current != &map->header && current->start < end;
2686 current = current->next) {
2687 if (invalidate && (current->eflags & MAP_ENTRY_USER_WIRED)) {
2688 vm_map_unlock_read(map);
2689 return (KERN_INVALID_ARGUMENT);
2691 if (end > current->end &&
2692 (current->next == &map->header ||
2693 current->end != current->next->start)) {
2694 vm_map_unlock_read(map);
2695 return (KERN_INVALID_ADDRESS);
2700 pmap_remove(map->pmap, start, end);
2704 * Make a second pass, cleaning/uncaching pages from the indicated
2707 for (current = entry; current != &map->header && current->start < end;) {
2708 offset = current->offset + (start - current->start);
2709 size = (end <= current->end ? end : current->end) - start;
2710 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
2712 vm_map_entry_t tentry;
2715 smap = current->object.sub_map;
2716 vm_map_lock_read(smap);
2717 (void) vm_map_lookup_entry(smap, offset, &tentry);
2718 tsize = tentry->end - offset;
2721 object = tentry->object.vm_object;
2722 offset = tentry->offset + (offset - tentry->start);
2723 vm_map_unlock_read(smap);
2725 object = current->object.vm_object;
2727 vm_object_reference(object);
2728 last_timestamp = map->timestamp;
2729 vm_map_unlock_read(map);
2730 if (!vm_object_sync(object, offset, size, syncio, invalidate))
2733 vm_object_deallocate(object);
2734 vm_map_lock_read(map);
2735 if (last_timestamp == map->timestamp ||
2736 !vm_map_lookup_entry(map, start, ¤t))
2737 current = current->next;
2740 vm_map_unlock_read(map);
2741 return (failed ? KERN_FAILURE : KERN_SUCCESS);
2745 * vm_map_entry_unwire: [ internal use only ]
2747 * Make the region specified by this entry pageable.
2749 * The map in question should be locked.
2750 * [This is the reason for this routine's existence.]
2753 vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry)
2755 vm_fault_unwire(map, entry->start, entry->end,
2756 entry->object.vm_object != NULL &&
2757 (entry->object.vm_object->type == OBJT_DEVICE ||
2758 entry->object.vm_object->type == OBJT_SG));
2759 entry->wired_count = 0;
2763 vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map)
2766 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0)
2767 vm_object_deallocate(entry->object.vm_object);
2768 uma_zfree(system_map ? kmapentzone : mapentzone, entry);
2772 * vm_map_entry_delete: [ internal use only ]
2774 * Deallocate the given entry from the target map.
2777 vm_map_entry_delete(vm_map_t map, vm_map_entry_t entry)
2780 vm_pindex_t offidxstart, offidxend, count, size1;
2783 vm_map_entry_unlink(map, entry);
2784 object = entry->object.vm_object;
2785 size = entry->end - entry->start;
2788 if (entry->cred != NULL) {
2789 swap_release_by_cred(size, entry->cred);
2790 crfree(entry->cred);
2793 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0 &&
2795 KASSERT(entry->cred == NULL || object->cred == NULL ||
2796 (entry->eflags & MAP_ENTRY_NEEDS_COPY),
2797 ("OVERCOMMIT vm_map_entry_delete: both cred %p", entry));
2798 count = OFF_TO_IDX(size);
2799 offidxstart = OFF_TO_IDX(entry->offset);
2800 offidxend = offidxstart + count;
2801 VM_OBJECT_LOCK(object);
2802 if (object->ref_count != 1 &&
2803 ((object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING ||
2804 object == kernel_object || object == kmem_object)) {
2805 vm_object_collapse(object);
2808 * The option OBJPR_NOTMAPPED can be passed here
2809 * because vm_map_delete() already performed
2810 * pmap_remove() on the only mapping to this range
2813 vm_object_page_remove(object, offidxstart, offidxend,
2815 if (object->type == OBJT_SWAP)
2816 swap_pager_freespace(object, offidxstart, count);
2817 if (offidxend >= object->size &&
2818 offidxstart < object->size) {
2819 size1 = object->size;
2820 object->size = offidxstart;
2821 if (object->cred != NULL) {
2822 size1 -= object->size;
2823 KASSERT(object->charge >= ptoa(size1),
2824 ("vm_map_entry_delete: object->charge < 0"));
2825 swap_release_by_cred(ptoa(size1), object->cred);
2826 object->charge -= ptoa(size1);
2830 VM_OBJECT_UNLOCK(object);
2832 entry->object.vm_object = NULL;
2833 if (map->system_map)
2834 vm_map_entry_deallocate(entry, TRUE);
2836 entry->next = curthread->td_map_def_user;
2837 curthread->td_map_def_user = entry;
2842 * vm_map_delete: [ internal use only ]
2844 * Deallocates the given address range from the target
2848 vm_map_delete(vm_map_t map, vm_offset_t start, vm_offset_t end)
2850 vm_map_entry_t entry;
2851 vm_map_entry_t first_entry;
2853 VM_MAP_ASSERT_LOCKED(map);
2855 return (KERN_SUCCESS);
2858 * Find the start of the region, and clip it
2860 if (!vm_map_lookup_entry(map, start, &first_entry))
2861 entry = first_entry->next;
2863 entry = first_entry;
2864 vm_map_clip_start(map, entry, start);
2868 * Step through all entries in this region
2870 while ((entry != &map->header) && (entry->start < end)) {
2871 vm_map_entry_t next;
2874 * Wait for wiring or unwiring of an entry to complete.
2875 * Also wait for any system wirings to disappear on
2878 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 ||
2879 (vm_map_pmap(map) != kernel_pmap &&
2880 vm_map_entry_system_wired_count(entry) != 0)) {
2881 unsigned int last_timestamp;
2882 vm_offset_t saved_start;
2883 vm_map_entry_t tmp_entry;
2885 saved_start = entry->start;
2886 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2887 last_timestamp = map->timestamp;
2888 (void) vm_map_unlock_and_wait(map, 0);
2890 if (last_timestamp + 1 != map->timestamp) {
2892 * Look again for the entry because the map was
2893 * modified while it was unlocked.
2894 * Specifically, the entry may have been
2895 * clipped, merged, or deleted.
2897 if (!vm_map_lookup_entry(map, saved_start,
2899 entry = tmp_entry->next;
2902 vm_map_clip_start(map, entry,
2908 vm_map_clip_end(map, entry, end);
2913 * Unwire before removing addresses from the pmap; otherwise,
2914 * unwiring will put the entries back in the pmap.
2916 if (entry->wired_count != 0) {
2917 vm_map_entry_unwire(map, entry);
2920 pmap_remove(map->pmap, entry->start, entry->end);
2923 * Delete the entry only after removing all pmap
2924 * entries pointing to its pages. (Otherwise, its
2925 * page frames may be reallocated, and any modify bits
2926 * will be set in the wrong object!)
2928 vm_map_entry_delete(map, entry);
2931 return (KERN_SUCCESS);
2937 * Remove the given address range from the target map.
2938 * This is the exported form of vm_map_delete.
2941 vm_map_remove(vm_map_t map, vm_offset_t start, vm_offset_t end)
2946 VM_MAP_RANGE_CHECK(map, start, end);
2947 result = vm_map_delete(map, start, end);
2953 * vm_map_check_protection:
2955 * Assert that the target map allows the specified privilege on the
2956 * entire address region given. The entire region must be allocated.
2958 * WARNING! This code does not and should not check whether the
2959 * contents of the region is accessible. For example a smaller file
2960 * might be mapped into a larger address space.
2962 * NOTE! This code is also called by munmap().
2964 * The map must be locked. A read lock is sufficient.
2967 vm_map_check_protection(vm_map_t map, vm_offset_t start, vm_offset_t end,
2968 vm_prot_t protection)
2970 vm_map_entry_t entry;
2971 vm_map_entry_t tmp_entry;
2973 if (!vm_map_lookup_entry(map, start, &tmp_entry))
2977 while (start < end) {
2978 if (entry == &map->header)
2983 if (start < entry->start)
2986 * Check protection associated with entry.
2988 if ((entry->protection & protection) != protection)
2990 /* go to next entry */
2992 entry = entry->next;
2998 * vm_map_copy_entry:
3000 * Copies the contents of the source entry to the destination
3001 * entry. The entries *must* be aligned properly.
3007 vm_map_entry_t src_entry,
3008 vm_map_entry_t dst_entry,
3009 vm_ooffset_t *fork_charge)
3011 vm_object_t src_object;
3012 vm_map_entry_t fake_entry;
3017 VM_MAP_ASSERT_LOCKED(dst_map);
3019 if ((dst_entry->eflags|src_entry->eflags) & MAP_ENTRY_IS_SUB_MAP)
3022 if (src_entry->wired_count == 0) {
3025 * If the source entry is marked needs_copy, it is already
3028 if ((src_entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) {
3029 pmap_protect(src_map->pmap,
3032 src_entry->protection & ~VM_PROT_WRITE);
3036 * Make a copy of the object.
3038 size = src_entry->end - src_entry->start;
3039 if ((src_object = src_entry->object.vm_object) != NULL) {
3040 VM_OBJECT_LOCK(src_object);
3041 charged = ENTRY_CHARGED(src_entry);
3042 if ((src_object->handle == NULL) &&
3043 (src_object->type == OBJT_DEFAULT ||
3044 src_object->type == OBJT_SWAP)) {
3045 vm_object_collapse(src_object);
3046 if ((src_object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING) {
3047 vm_object_split(src_entry);
3048 src_object = src_entry->object.vm_object;
3051 vm_object_reference_locked(src_object);
3052 vm_object_clear_flag(src_object, OBJ_ONEMAPPING);
3053 if (src_entry->cred != NULL &&
3054 !(src_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
3055 KASSERT(src_object->cred == NULL,
3056 ("OVERCOMMIT: vm_map_copy_entry: cred %p",
3058 src_object->cred = src_entry->cred;
3059 src_object->charge = size;
3061 VM_OBJECT_UNLOCK(src_object);
3062 dst_entry->object.vm_object = src_object;
3064 cred = curthread->td_ucred;
3066 dst_entry->cred = cred;
3067 *fork_charge += size;
3068 if (!(src_entry->eflags &
3069 MAP_ENTRY_NEEDS_COPY)) {
3071 src_entry->cred = cred;
3072 *fork_charge += size;
3075 src_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY);
3076 dst_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY);
3077 dst_entry->offset = src_entry->offset;
3078 if (src_entry->eflags & MAP_ENTRY_VN_WRITECNT) {
3080 * MAP_ENTRY_VN_WRITECNT cannot
3081 * indicate write reference from
3082 * src_entry, since the entry is
3083 * marked as needs copy. Allocate a
3084 * fake entry that is used to
3085 * decrement object->un_pager.vnp.writecount
3086 * at the appropriate time. Attach
3087 * fake_entry to the deferred list.
3089 fake_entry = vm_map_entry_create(dst_map);
3090 fake_entry->eflags = MAP_ENTRY_VN_WRITECNT;
3091 src_entry->eflags &= ~MAP_ENTRY_VN_WRITECNT;
3092 vm_object_reference(src_object);
3093 fake_entry->object.vm_object = src_object;
3094 fake_entry->start = src_entry->start;
3095 fake_entry->end = src_entry->end;
3096 fake_entry->next = curthread->td_map_def_user;
3097 curthread->td_map_def_user = fake_entry;
3100 dst_entry->object.vm_object = NULL;
3101 dst_entry->offset = 0;
3102 if (src_entry->cred != NULL) {
3103 dst_entry->cred = curthread->td_ucred;
3104 crhold(dst_entry->cred);
3105 *fork_charge += size;
3109 pmap_copy(dst_map->pmap, src_map->pmap, dst_entry->start,
3110 dst_entry->end - dst_entry->start, src_entry->start);
3113 * Of course, wired down pages can't be set copy-on-write.
3114 * Cause wired pages to be copied into the new map by
3115 * simulating faults (the new pages are pageable)
3117 vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry,
3123 * vmspace_map_entry_forked:
3124 * Update the newly-forked vmspace each time a map entry is inherited
3125 * or copied. The values for vm_dsize and vm_tsize are approximate
3126 * (and mostly-obsolete ideas in the face of mmap(2) et al.)
3129 vmspace_map_entry_forked(const struct vmspace *vm1, struct vmspace *vm2,
3130 vm_map_entry_t entry)
3132 vm_size_t entrysize;
3135 entrysize = entry->end - entry->start;
3136 vm2->vm_map.size += entrysize;
3137 if (entry->eflags & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP)) {
3138 vm2->vm_ssize += btoc(entrysize);
3139 } else if (entry->start >= (vm_offset_t)vm1->vm_daddr &&
3140 entry->start < (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)) {
3141 newend = MIN(entry->end,
3142 (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize));
3143 vm2->vm_dsize += btoc(newend - entry->start);
3144 } else if (entry->start >= (vm_offset_t)vm1->vm_taddr &&
3145 entry->start < (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)) {
3146 newend = MIN(entry->end,
3147 (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize));
3148 vm2->vm_tsize += btoc(newend - entry->start);
3154 * Create a new process vmspace structure and vm_map
3155 * based on those of an existing process. The new map
3156 * is based on the old map, according to the inheritance
3157 * values on the regions in that map.
3159 * XXX It might be worth coalescing the entries added to the new vmspace.
3161 * The source map must not be locked.
3164 vmspace_fork(struct vmspace *vm1, vm_ooffset_t *fork_charge)
3166 struct vmspace *vm2;
3167 vm_map_t new_map, old_map;
3168 vm_map_entry_t new_entry, old_entry;
3172 old_map = &vm1->vm_map;
3173 /* Copy immutable fields of vm1 to vm2. */
3174 vm2 = vmspace_alloc(old_map->min_offset, old_map->max_offset);
3177 vm2->vm_taddr = vm1->vm_taddr;
3178 vm2->vm_daddr = vm1->vm_daddr;
3179 vm2->vm_maxsaddr = vm1->vm_maxsaddr;
3180 vm_map_lock(old_map);
3182 vm_map_wait_busy(old_map);
3183 new_map = &vm2->vm_map;
3184 locked = vm_map_trylock(new_map); /* trylock to silence WITNESS */
3185 KASSERT(locked, ("vmspace_fork: lock failed"));
3187 old_entry = old_map->header.next;
3189 while (old_entry != &old_map->header) {
3190 if (old_entry->eflags & MAP_ENTRY_IS_SUB_MAP)
3191 panic("vm_map_fork: encountered a submap");
3193 switch (old_entry->inheritance) {
3194 case VM_INHERIT_NONE:
3197 case VM_INHERIT_SHARE:
3199 * Clone the entry, creating the shared object if necessary.
3201 object = old_entry->object.vm_object;
3202 if (object == NULL) {
3203 object = vm_object_allocate(OBJT_DEFAULT,
3204 atop(old_entry->end - old_entry->start));
3205 old_entry->object.vm_object = object;
3206 old_entry->offset = 0;
3207 if (old_entry->cred != NULL) {
3208 object->cred = old_entry->cred;
3209 object->charge = old_entry->end -
3211 old_entry->cred = NULL;
3216 * Add the reference before calling vm_object_shadow
3217 * to insure that a shadow object is created.
3219 vm_object_reference(object);
3220 if (old_entry->eflags & MAP_ENTRY_NEEDS_COPY) {
3221 vm_object_shadow(&old_entry->object.vm_object,
3223 old_entry->end - old_entry->start);
3224 old_entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
3225 /* Transfer the second reference too. */
3226 vm_object_reference(
3227 old_entry->object.vm_object);
3230 * As in vm_map_simplify_entry(), the
3231 * vnode lock will not be acquired in
3232 * this call to vm_object_deallocate().
3234 vm_object_deallocate(object);
3235 object = old_entry->object.vm_object;
3237 VM_OBJECT_LOCK(object);
3238 vm_object_clear_flag(object, OBJ_ONEMAPPING);
3239 if (old_entry->cred != NULL) {
3240 KASSERT(object->cred == NULL, ("vmspace_fork both cred"));
3241 object->cred = old_entry->cred;
3242 object->charge = old_entry->end - old_entry->start;
3243 old_entry->cred = NULL;
3247 * Assert the correct state of the vnode
3248 * v_writecount while the object is locked, to
3249 * not relock it later for the assertion
3252 if (old_entry->eflags & MAP_ENTRY_VN_WRITECNT &&
3253 object->type == OBJT_VNODE) {
3254 KASSERT(((struct vnode *)object->handle)->
3256 ("vmspace_fork: v_writecount %p", object));
3257 KASSERT(object->un_pager.vnp.writemappings > 0,
3258 ("vmspace_fork: vnp.writecount %p",
3261 VM_OBJECT_UNLOCK(object);
3264 * Clone the entry, referencing the shared object.
3266 new_entry = vm_map_entry_create(new_map);
3267 *new_entry = *old_entry;
3268 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED |
3269 MAP_ENTRY_IN_TRANSITION);
3270 new_entry->wiring_thread = NULL;
3271 new_entry->wired_count = 0;
3272 if (new_entry->eflags & MAP_ENTRY_VN_WRITECNT) {
3273 vnode_pager_update_writecount(object,
3274 new_entry->start, new_entry->end);
3278 * Insert the entry into the new map -- we know we're
3279 * inserting at the end of the new map.
3281 vm_map_entry_link(new_map, new_map->header.prev,
3283 vmspace_map_entry_forked(vm1, vm2, new_entry);
3286 * Update the physical map
3288 pmap_copy(new_map->pmap, old_map->pmap,
3290 (old_entry->end - old_entry->start),
3294 case VM_INHERIT_COPY:
3296 * Clone the entry and link into the map.
3298 new_entry = vm_map_entry_create(new_map);
3299 *new_entry = *old_entry;
3301 * Copied entry is COW over the old object.
3303 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED |
3304 MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_VN_WRITECNT);
3305 new_entry->wiring_thread = NULL;
3306 new_entry->wired_count = 0;
3307 new_entry->object.vm_object = NULL;
3308 new_entry->cred = NULL;
3309 vm_map_entry_link(new_map, new_map->header.prev,
3311 vmspace_map_entry_forked(vm1, vm2, new_entry);
3312 vm_map_copy_entry(old_map, new_map, old_entry,
3313 new_entry, fork_charge);
3316 old_entry = old_entry->next;
3319 * Use inlined vm_map_unlock() to postpone handling the deferred
3320 * map entries, which cannot be done until both old_map and
3321 * new_map locks are released.
3323 sx_xunlock(&old_map->lock);
3324 sx_xunlock(&new_map->lock);
3325 vm_map_process_deferred();
3331 vm_map_stack(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
3332 vm_prot_t prot, vm_prot_t max, int cow)
3334 vm_map_entry_t new_entry, prev_entry;
3335 vm_offset_t bot, top;
3336 vm_size_t growsize, init_ssize;
3338 rlim_t lmemlim, vmemlim;
3341 * The stack orientation is piggybacked with the cow argument.
3342 * Extract it into orient and mask the cow argument so that we
3343 * don't pass it around further.
3344 * NOTE: We explicitly allow bi-directional stacks.
3346 orient = cow & (MAP_STACK_GROWS_DOWN|MAP_STACK_GROWS_UP);
3347 KASSERT(orient != 0, ("No stack grow direction"));
3349 if (addrbos < vm_map_min(map) ||
3350 addrbos > vm_map_max(map) ||
3351 addrbos + max_ssize < addrbos)
3352 return (KERN_NO_SPACE);
3354 growsize = sgrowsiz;
3355 init_ssize = (max_ssize < growsize) ? max_ssize : growsize;
3358 lmemlim = lim_cur(curproc, RLIMIT_MEMLOCK);
3359 vmemlim = lim_cur(curproc, RLIMIT_VMEM);
3360 PROC_UNLOCK(curproc);
3364 /* If addr is already mapped, no go */
3365 if (vm_map_lookup_entry(map, addrbos, &prev_entry)) {
3367 return (KERN_NO_SPACE);
3370 if (!old_mlock && map->flags & MAP_WIREFUTURE) {
3371 if (ptoa(pmap_wired_count(map->pmap)) + init_ssize > lmemlim) {
3373 return (KERN_NO_SPACE);
3377 /* If we would blow our VMEM resource limit, no go */
3378 if (map->size + init_ssize > vmemlim) {
3380 return (KERN_NO_SPACE);
3384 * If we can't accomodate max_ssize in the current mapping, no go.
3385 * However, we need to be aware that subsequent user mappings might
3386 * map into the space we have reserved for stack, and currently this
3387 * space is not protected.
3389 * Hopefully we will at least detect this condition when we try to
3392 if ((prev_entry->next != &map->header) &&
3393 (prev_entry->next->start < addrbos + max_ssize)) {
3395 return (KERN_NO_SPACE);
3399 * We initially map a stack of only init_ssize. We will grow as
3400 * needed later. Depending on the orientation of the stack (i.e.
3401 * the grow direction) we either map at the top of the range, the
3402 * bottom of the range or in the middle.
3404 * Note: we would normally expect prot and max to be VM_PROT_ALL,
3405 * and cow to be 0. Possibly we should eliminate these as input
3406 * parameters, and just pass these values here in the insert call.
3408 if (orient == MAP_STACK_GROWS_DOWN)
3409 bot = addrbos + max_ssize - init_ssize;
3410 else if (orient == MAP_STACK_GROWS_UP)
3413 bot = round_page(addrbos + max_ssize/2 - init_ssize/2);
3414 top = bot + init_ssize;
3415 rv = vm_map_insert(map, NULL, 0, bot, top, prot, max, cow);
3417 /* Now set the avail_ssize amount. */
3418 if (rv == KERN_SUCCESS) {
3419 if (prev_entry != &map->header)
3420 vm_map_clip_end(map, prev_entry, bot);
3421 new_entry = prev_entry->next;
3422 if (new_entry->end != top || new_entry->start != bot)
3423 panic("Bad entry start/end for new stack entry");
3425 new_entry->avail_ssize = max_ssize - init_ssize;
3426 if (orient & MAP_STACK_GROWS_DOWN)
3427 new_entry->eflags |= MAP_ENTRY_GROWS_DOWN;
3428 if (orient & MAP_STACK_GROWS_UP)
3429 new_entry->eflags |= MAP_ENTRY_GROWS_UP;
3436 static int stack_guard_page = 0;
3437 TUNABLE_INT("security.bsd.stack_guard_page", &stack_guard_page);
3438 SYSCTL_INT(_security_bsd, OID_AUTO, stack_guard_page, CTLFLAG_RW,
3439 &stack_guard_page, 0,
3440 "Insert stack guard page ahead of the growable segments.");
3442 /* Attempts to grow a vm stack entry. Returns KERN_SUCCESS if the
3443 * desired address is already mapped, or if we successfully grow
3444 * the stack. Also returns KERN_SUCCESS if addr is outside the
3445 * stack range (this is strange, but preserves compatibility with
3446 * the grow function in vm_machdep.c).
3449 vm_map_growstack(struct proc *p, vm_offset_t addr)
3451 vm_map_entry_t next_entry, prev_entry;
3452 vm_map_entry_t new_entry, stack_entry;
3453 struct vmspace *vm = p->p_vmspace;
3454 vm_map_t map = &vm->vm_map;
3457 size_t grow_amount, max_grow;
3458 rlim_t lmemlim, stacklim, vmemlim;
3459 int is_procstack, rv;
3470 lmemlim = lim_cur(p, RLIMIT_MEMLOCK);
3471 stacklim = lim_cur(p, RLIMIT_STACK);
3472 vmemlim = lim_cur(p, RLIMIT_VMEM);
3475 vm_map_lock_read(map);
3477 /* If addr is already in the entry range, no need to grow.*/
3478 if (vm_map_lookup_entry(map, addr, &prev_entry)) {
3479 vm_map_unlock_read(map);
3480 return (KERN_SUCCESS);
3483 next_entry = prev_entry->next;
3484 if (!(prev_entry->eflags & MAP_ENTRY_GROWS_UP)) {
3486 * This entry does not grow upwards. Since the address lies
3487 * beyond this entry, the next entry (if one exists) has to
3488 * be a downward growable entry. The entry list header is
3489 * never a growable entry, so it suffices to check the flags.
3491 if (!(next_entry->eflags & MAP_ENTRY_GROWS_DOWN)) {
3492 vm_map_unlock_read(map);
3493 return (KERN_SUCCESS);
3495 stack_entry = next_entry;
3498 * This entry grows upward. If the next entry does not at
3499 * least grow downwards, this is the entry we need to grow.
3500 * otherwise we have two possible choices and we have to
3503 if (next_entry->eflags & MAP_ENTRY_GROWS_DOWN) {
3505 * We have two choices; grow the entry closest to
3506 * the address to minimize the amount of growth.
3508 if (addr - prev_entry->end <= next_entry->start - addr)
3509 stack_entry = prev_entry;
3511 stack_entry = next_entry;
3513 stack_entry = prev_entry;
3516 if (stack_entry == next_entry) {
3517 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_DOWN, ("foo"));
3518 KASSERT(addr < stack_entry->start, ("foo"));
3519 end = (prev_entry != &map->header) ? prev_entry->end :
3520 stack_entry->start - stack_entry->avail_ssize;
3521 grow_amount = roundup(stack_entry->start - addr, PAGE_SIZE);
3522 max_grow = stack_entry->start - end;
3524 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_UP, ("foo"));
3525 KASSERT(addr >= stack_entry->end, ("foo"));
3526 end = (next_entry != &map->header) ? next_entry->start :
3527 stack_entry->end + stack_entry->avail_ssize;
3528 grow_amount = roundup(addr + 1 - stack_entry->end, PAGE_SIZE);
3529 max_grow = end - stack_entry->end;
3532 if (grow_amount > stack_entry->avail_ssize) {
3533 vm_map_unlock_read(map);
3534 return (KERN_NO_SPACE);
3538 * If there is no longer enough space between the entries nogo, and
3539 * adjust the available space. Note: this should only happen if the
3540 * user has mapped into the stack area after the stack was created,
3541 * and is probably an error.
3543 * This also effectively destroys any guard page the user might have
3544 * intended by limiting the stack size.
3546 if (grow_amount + (stack_guard_page ? PAGE_SIZE : 0) > max_grow) {
3547 if (vm_map_lock_upgrade(map))
3550 stack_entry->avail_ssize = max_grow;
3553 return (KERN_NO_SPACE);
3556 is_procstack = (addr >= (vm_offset_t)vm->vm_maxsaddr) ? 1 : 0;
3559 * If this is the main process stack, see if we're over the stack
3562 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) {
3563 vm_map_unlock_read(map);
3564 return (KERN_NO_SPACE);
3569 racct_set(p, RACCT_STACK, ctob(vm->vm_ssize) + grow_amount)) {
3571 vm_map_unlock_read(map);
3572 return (KERN_NO_SPACE);
3577 /* Round up the grow amount modulo sgrowsiz */
3578 growsize = sgrowsiz;
3579 grow_amount = roundup(grow_amount, growsize);
3580 if (grow_amount > stack_entry->avail_ssize)
3581 grow_amount = stack_entry->avail_ssize;
3582 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) {
3583 grow_amount = trunc_page((vm_size_t)stacklim) -
3588 limit = racct_get_available(p, RACCT_STACK);
3590 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > limit))
3591 grow_amount = limit - ctob(vm->vm_ssize);
3593 if (!old_mlock && map->flags & MAP_WIREFUTURE) {
3594 if (ptoa(pmap_wired_count(map->pmap)) + grow_amount > lmemlim) {
3595 vm_map_unlock_read(map);
3601 if (racct_set(p, RACCT_MEMLOCK,
3602 ptoa(pmap_wired_count(map->pmap)) + grow_amount)) {
3604 vm_map_unlock_read(map);
3611 /* If we would blow our VMEM resource limit, no go */
3612 if (map->size + grow_amount > vmemlim) {
3613 vm_map_unlock_read(map);
3619 if (racct_set(p, RACCT_VMEM, map->size + grow_amount)) {
3621 vm_map_unlock_read(map);
3628 if (vm_map_lock_upgrade(map))
3631 if (stack_entry == next_entry) {
3635 /* Get the preliminary new entry start value */
3636 addr = stack_entry->start - grow_amount;
3639 * If this puts us into the previous entry, cut back our
3640 * growth to the available space. Also, see the note above.
3643 stack_entry->avail_ssize = max_grow;
3645 if (stack_guard_page)
3649 rv = vm_map_insert(map, NULL, 0, addr, stack_entry->start,
3650 next_entry->protection, next_entry->max_protection, 0);
3652 /* Adjust the available stack space by the amount we grew. */
3653 if (rv == KERN_SUCCESS) {
3654 if (prev_entry != &map->header)
3655 vm_map_clip_end(map, prev_entry, addr);
3656 new_entry = prev_entry->next;
3657 KASSERT(new_entry == stack_entry->prev, ("foo"));
3658 KASSERT(new_entry->end == stack_entry->start, ("foo"));
3659 KASSERT(new_entry->start == addr, ("foo"));
3660 grow_amount = new_entry->end - new_entry->start;
3661 new_entry->avail_ssize = stack_entry->avail_ssize -
3663 stack_entry->eflags &= ~MAP_ENTRY_GROWS_DOWN;
3664 new_entry->eflags |= MAP_ENTRY_GROWS_DOWN;
3670 addr = stack_entry->end + grow_amount;
3673 * If this puts us into the next entry, cut back our growth
3674 * to the available space. Also, see the note above.
3677 stack_entry->avail_ssize = end - stack_entry->end;
3679 if (stack_guard_page)
3683 grow_amount = addr - stack_entry->end;
3684 cred = stack_entry->cred;
3685 if (cred == NULL && stack_entry->object.vm_object != NULL)
3686 cred = stack_entry->object.vm_object->cred;
3687 if (cred != NULL && !swap_reserve_by_cred(grow_amount, cred))
3689 /* Grow the underlying object if applicable. */
3690 else if (stack_entry->object.vm_object == NULL ||
3691 vm_object_coalesce(stack_entry->object.vm_object,
3692 stack_entry->offset,
3693 (vm_size_t)(stack_entry->end - stack_entry->start),
3694 (vm_size_t)grow_amount, cred != NULL)) {
3695 map->size += (addr - stack_entry->end);
3696 /* Update the current entry. */
3697 stack_entry->end = addr;
3698 stack_entry->avail_ssize -= grow_amount;
3699 vm_map_entry_resize_free(map, stack_entry);
3702 if (next_entry != &map->header)
3703 vm_map_clip_start(map, next_entry, addr);
3708 if (rv == KERN_SUCCESS && is_procstack)
3709 vm->vm_ssize += btoc(grow_amount);
3714 * Heed the MAP_WIREFUTURE flag if it was set for this process.
3716 if (rv == KERN_SUCCESS && (map->flags & MAP_WIREFUTURE)) {
3718 (stack_entry == next_entry) ? addr : addr - grow_amount,
3719 (stack_entry == next_entry) ? stack_entry->start : addr,
3720 (p->p_flag & P_SYSTEM)
3721 ? VM_MAP_WIRE_SYSTEM|VM_MAP_WIRE_NOHOLES
3722 : VM_MAP_WIRE_USER|VM_MAP_WIRE_NOHOLES);
3727 if (rv != KERN_SUCCESS) {
3729 error = racct_set(p, RACCT_VMEM, map->size);
3730 KASSERT(error == 0, ("decreasing RACCT_VMEM failed"));
3732 error = racct_set(p, RACCT_MEMLOCK,
3733 ptoa(pmap_wired_count(map->pmap)));
3734 KASSERT(error == 0, ("decreasing RACCT_MEMLOCK failed"));
3736 error = racct_set(p, RACCT_STACK, ctob(vm->vm_ssize));
3737 KASSERT(error == 0, ("decreasing RACCT_STACK failed"));
3746 * Unshare the specified VM space for exec. If other processes are
3747 * mapped to it, then create a new one. The new vmspace is null.
3750 vmspace_exec(struct proc *p, vm_offset_t minuser, vm_offset_t maxuser)
3752 struct vmspace *oldvmspace = p->p_vmspace;
3753 struct vmspace *newvmspace;
3755 KASSERT((curthread->td_pflags & TDP_EXECVMSPC) == 0,
3756 ("vmspace_exec recursed"));
3757 newvmspace = vmspace_alloc(minuser, maxuser);
3758 if (newvmspace == NULL)
3760 newvmspace->vm_swrss = oldvmspace->vm_swrss;
3762 * This code is written like this for prototype purposes. The
3763 * goal is to avoid running down the vmspace here, but let the
3764 * other process's that are still using the vmspace to finally
3765 * run it down. Even though there is little or no chance of blocking
3766 * here, it is a good idea to keep this form for future mods.
3768 PROC_VMSPACE_LOCK(p);
3769 p->p_vmspace = newvmspace;
3770 PROC_VMSPACE_UNLOCK(p);
3771 if (p == curthread->td_proc)
3772 pmap_activate(curthread);
3773 curthread->td_pflags |= TDP_EXECVMSPC;
3778 * Unshare the specified VM space for forcing COW. This
3779 * is called by rfork, for the (RFMEM|RFPROC) == 0 case.
3782 vmspace_unshare(struct proc *p)
3784 struct vmspace *oldvmspace = p->p_vmspace;
3785 struct vmspace *newvmspace;
3786 vm_ooffset_t fork_charge;
3788 if (oldvmspace->vm_refcnt == 1)
3791 newvmspace = vmspace_fork(oldvmspace, &fork_charge);
3792 if (newvmspace == NULL)
3794 if (!swap_reserve_by_cred(fork_charge, p->p_ucred)) {
3795 vmspace_free(newvmspace);
3798 PROC_VMSPACE_LOCK(p);
3799 p->p_vmspace = newvmspace;
3800 PROC_VMSPACE_UNLOCK(p);
3801 if (p == curthread->td_proc)
3802 pmap_activate(curthread);
3803 vmspace_free(oldvmspace);
3810 * Finds the VM object, offset, and
3811 * protection for a given virtual address in the
3812 * specified map, assuming a page fault of the
3815 * Leaves the map in question locked for read; return
3816 * values are guaranteed until a vm_map_lookup_done
3817 * call is performed. Note that the map argument
3818 * is in/out; the returned map must be used in
3819 * the call to vm_map_lookup_done.
3821 * A handle (out_entry) is returned for use in
3822 * vm_map_lookup_done, to make that fast.
3824 * If a lookup is requested with "write protection"
3825 * specified, the map may be changed to perform virtual
3826 * copying operations, although the data referenced will
3830 vm_map_lookup(vm_map_t *var_map, /* IN/OUT */
3832 vm_prot_t fault_typea,
3833 vm_map_entry_t *out_entry, /* OUT */
3834 vm_object_t *object, /* OUT */
3835 vm_pindex_t *pindex, /* OUT */
3836 vm_prot_t *out_prot, /* OUT */
3837 boolean_t *wired) /* OUT */
3839 vm_map_entry_t entry;
3840 vm_map_t map = *var_map;
3842 vm_prot_t fault_type = fault_typea;
3843 vm_object_t eobject;
3849 vm_map_lock_read(map);
3852 * Lookup the faulting address.
3854 if (!vm_map_lookup_entry(map, vaddr, out_entry)) {
3855 vm_map_unlock_read(map);
3856 return (KERN_INVALID_ADDRESS);
3864 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
3865 vm_map_t old_map = map;
3867 *var_map = map = entry->object.sub_map;
3868 vm_map_unlock_read(old_map);
3873 * Check whether this task is allowed to have this page.
3875 prot = entry->protection;
3876 fault_type &= (VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE);
3877 if ((fault_type & prot) != fault_type || prot == VM_PROT_NONE) {
3878 vm_map_unlock_read(map);
3879 return (KERN_PROTECTION_FAILURE);
3881 KASSERT((prot & VM_PROT_WRITE) == 0 || (entry->eflags &
3882 (MAP_ENTRY_USER_WIRED | MAP_ENTRY_NEEDS_COPY)) !=
3883 (MAP_ENTRY_USER_WIRED | MAP_ENTRY_NEEDS_COPY),
3884 ("entry %p flags %x", entry, entry->eflags));
3885 if ((fault_typea & VM_PROT_COPY) != 0 &&
3886 (entry->max_protection & VM_PROT_WRITE) == 0 &&
3887 (entry->eflags & MAP_ENTRY_COW) == 0) {
3888 vm_map_unlock_read(map);
3889 return (KERN_PROTECTION_FAILURE);
3893 * If this page is not pageable, we have to get it for all possible
3896 *wired = (entry->wired_count != 0);
3898 fault_type = entry->protection;
3899 size = entry->end - entry->start;
3901 * If the entry was copy-on-write, we either ...
3903 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
3905 * If we want to write the page, we may as well handle that
3906 * now since we've got the map locked.
3908 * If we don't need to write the page, we just demote the
3909 * permissions allowed.
3911 if ((fault_type & VM_PROT_WRITE) != 0 ||
3912 (fault_typea & VM_PROT_COPY) != 0) {
3914 * Make a new object, and place it in the object
3915 * chain. Note that no new references have appeared
3916 * -- one just moved from the map to the new
3919 if (vm_map_lock_upgrade(map))
3922 if (entry->cred == NULL) {
3924 * The debugger owner is charged for
3927 cred = curthread->td_ucred;
3929 if (!swap_reserve_by_cred(size, cred)) {
3932 return (KERN_RESOURCE_SHORTAGE);
3936 vm_object_shadow(&entry->object.vm_object,
3937 &entry->offset, size);
3938 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
3939 eobject = entry->object.vm_object;
3940 if (eobject->cred != NULL) {
3942 * The object was not shadowed.
3944 swap_release_by_cred(size, entry->cred);
3945 crfree(entry->cred);
3947 } else if (entry->cred != NULL) {
3948 VM_OBJECT_LOCK(eobject);
3949 eobject->cred = entry->cred;
3950 eobject->charge = size;
3951 VM_OBJECT_UNLOCK(eobject);
3955 vm_map_lock_downgrade(map);
3958 * We're attempting to read a copy-on-write page --
3959 * don't allow writes.
3961 prot &= ~VM_PROT_WRITE;
3966 * Create an object if necessary.
3968 if (entry->object.vm_object == NULL &&
3970 if (vm_map_lock_upgrade(map))
3972 entry->object.vm_object = vm_object_allocate(OBJT_DEFAULT,
3975 if (entry->cred != NULL) {
3976 VM_OBJECT_LOCK(entry->object.vm_object);
3977 entry->object.vm_object->cred = entry->cred;
3978 entry->object.vm_object->charge = size;
3979 VM_OBJECT_UNLOCK(entry->object.vm_object);
3982 vm_map_lock_downgrade(map);
3986 * Return the object/offset from this entry. If the entry was
3987 * copy-on-write or empty, it has been fixed up.
3989 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
3990 *object = entry->object.vm_object;
3993 return (KERN_SUCCESS);
3997 * vm_map_lookup_locked:
3999 * Lookup the faulting address. A version of vm_map_lookup that returns
4000 * KERN_FAILURE instead of blocking on map lock or memory allocation.
4003 vm_map_lookup_locked(vm_map_t *var_map, /* IN/OUT */
4005 vm_prot_t fault_typea,
4006 vm_map_entry_t *out_entry, /* OUT */
4007 vm_object_t *object, /* OUT */
4008 vm_pindex_t *pindex, /* OUT */
4009 vm_prot_t *out_prot, /* OUT */
4010 boolean_t *wired) /* OUT */
4012 vm_map_entry_t entry;
4013 vm_map_t map = *var_map;
4015 vm_prot_t fault_type = fault_typea;
4018 * Lookup the faulting address.
4020 if (!vm_map_lookup_entry(map, vaddr, out_entry))
4021 return (KERN_INVALID_ADDRESS);
4026 * Fail if the entry refers to a submap.
4028 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP)
4029 return (KERN_FAILURE);
4032 * Check whether this task is allowed to have this page.
4034 prot = entry->protection;
4035 fault_type &= VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE;
4036 if ((fault_type & prot) != fault_type)
4037 return (KERN_PROTECTION_FAILURE);
4040 * If this page is not pageable, we have to get it for all possible
4043 *wired = (entry->wired_count != 0);
4045 fault_type = entry->protection;
4047 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
4049 * Fail if the entry was copy-on-write for a write fault.
4051 if (fault_type & VM_PROT_WRITE)
4052 return (KERN_FAILURE);
4054 * We're attempting to read a copy-on-write page --
4055 * don't allow writes.
4057 prot &= ~VM_PROT_WRITE;
4061 * Fail if an object should be created.
4063 if (entry->object.vm_object == NULL && !map->system_map)
4064 return (KERN_FAILURE);
4067 * Return the object/offset from this entry. If the entry was
4068 * copy-on-write or empty, it has been fixed up.
4070 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
4071 *object = entry->object.vm_object;
4074 return (KERN_SUCCESS);
4078 * vm_map_lookup_done:
4080 * Releases locks acquired by a vm_map_lookup
4081 * (according to the handle returned by that lookup).
4084 vm_map_lookup_done(vm_map_t map, vm_map_entry_t entry)
4087 * Unlock the main-level map
4089 vm_map_unlock_read(map);
4092 #include "opt_ddb.h"
4094 #include <sys/kernel.h>
4096 #include <ddb/ddb.h>
4099 * vm_map_print: [ debug ]
4101 DB_SHOW_COMMAND(map, vm_map_print)
4104 /* XXX convert args. */
4105 vm_map_t map = (vm_map_t)addr;
4106 boolean_t full = have_addr;
4108 vm_map_entry_t entry;
4110 db_iprintf("Task map %p: pmap=%p, nentries=%d, version=%u\n",
4112 (void *)map->pmap, map->nentries, map->timestamp);
4115 if (!full && db_indent)
4119 for (entry = map->header.next; entry != &map->header;
4120 entry = entry->next) {
4121 db_iprintf("map entry %p: start=%p, end=%p\n",
4122 (void *)entry, (void *)entry->start, (void *)entry->end);
4125 static char *inheritance_name[4] =
4126 {"share", "copy", "none", "donate_copy"};
4128 db_iprintf(" prot=%x/%x/%s",
4130 entry->max_protection,
4131 inheritance_name[(int)(unsigned char)entry->inheritance]);
4132 if (entry->wired_count != 0)
4133 db_printf(", wired");
4135 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
4136 db_printf(", share=%p, offset=0x%jx\n",
4137 (void *)entry->object.sub_map,
4138 (uintmax_t)entry->offset);
4140 if ((entry->prev == &map->header) ||
4141 (entry->prev->object.sub_map !=
4142 entry->object.sub_map)) {
4144 vm_map_print((db_expr_t)(intptr_t)
4145 entry->object.sub_map,
4146 full, 0, (char *)0);
4150 if (entry->cred != NULL)
4151 db_printf(", ruid %d", entry->cred->cr_ruid);
4152 db_printf(", object=%p, offset=0x%jx",
4153 (void *)entry->object.vm_object,
4154 (uintmax_t)entry->offset);
4155 if (entry->object.vm_object && entry->object.vm_object->cred)
4156 db_printf(", obj ruid %d charge %jx",
4157 entry->object.vm_object->cred->cr_ruid,
4158 (uintmax_t)entry->object.vm_object->charge);
4159 if (entry->eflags & MAP_ENTRY_COW)
4160 db_printf(", copy (%s)",
4161 (entry->eflags & MAP_ENTRY_NEEDS_COPY) ? "needed" : "done");
4165 if ((entry->prev == &map->header) ||
4166 (entry->prev->object.vm_object !=
4167 entry->object.vm_object)) {
4169 vm_object_print((db_expr_t)(intptr_t)
4170 entry->object.vm_object,
4171 full, 0, (char *)0);
4183 DB_SHOW_COMMAND(procvm, procvm)
4188 p = (struct proc *) addr;
4193 db_printf("p = %p, vmspace = %p, map = %p, pmap = %p\n",
4194 (void *)p, (void *)p->p_vmspace, (void *)&p->p_vmspace->vm_map,
4195 (void *)vmspace_pmap(p->p_vmspace));
4197 vm_map_print((db_expr_t)(intptr_t)&p->p_vmspace->vm_map, 1, 0, NULL);