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/resourcevar.h>
80 #include <sys/sysctl.h>
81 #include <sys/sysent.h>
85 #include <vm/vm_param.h>
87 #include <vm/vm_map.h>
88 #include <vm/vm_page.h>
89 #include <vm/vm_object.h>
90 #include <vm/vm_pager.h>
91 #include <vm/vm_kern.h>
92 #include <vm/vm_extern.h>
93 #include <vm/swap_pager.h>
97 * Virtual memory maps provide for the mapping, protection,
98 * and sharing of virtual memory objects. In addition,
99 * this module provides for an efficient virtual copy of
100 * memory from one map to another.
102 * Synchronization is required prior to most operations.
104 * Maps consist of an ordered doubly-linked list of simple
105 * entries; a self-adjusting binary search tree of these
106 * entries is used to speed up lookups.
108 * Since portions of maps are specified by start/end addresses,
109 * which may not align with existing map entries, all
110 * routines merely "clip" entries to these start/end values.
111 * [That is, an entry is split into two, bordering at a
112 * start or end value.] Note that these clippings may not
113 * always be necessary (as the two resulting entries are then
114 * not changed); however, the clipping is done for convenience.
116 * As mentioned above, virtual copy operations are performed
117 * by copying VM object references from one map to
118 * another, and then marking both regions as copy-on-write.
121 static struct mtx map_sleep_mtx;
122 static uma_zone_t mapentzone;
123 static uma_zone_t kmapentzone;
124 static uma_zone_t mapzone;
125 static uma_zone_t vmspace_zone;
126 static struct vm_object kmapentobj;
127 static int vmspace_zinit(void *mem, int size, int flags);
128 static void vmspace_zfini(void *mem, int size);
129 static int vm_map_zinit(void *mem, int ize, int flags);
130 static void vm_map_zfini(void *mem, int size);
131 static void _vm_map_init(vm_map_t map, vm_offset_t min, vm_offset_t max);
132 static void vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map);
133 static void vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry);
135 static void vm_map_zdtor(void *mem, int size, void *arg);
136 static void vmspace_zdtor(void *mem, int size, void *arg);
139 #define ENTRY_CHARGED(e) ((e)->uip != NULL || \
140 ((e)->object.vm_object != NULL && (e)->object.vm_object->uip != NULL && \
141 !((e)->eflags & MAP_ENTRY_NEEDS_COPY)))
144 * PROC_VMSPACE_{UN,}LOCK() can be a noop as long as vmspaces are type
147 #define PROC_VMSPACE_LOCK(p) do { } while (0)
148 #define PROC_VMSPACE_UNLOCK(p) do { } while (0)
151 * VM_MAP_RANGE_CHECK: [ internal use only ]
153 * Asserts that the starting and ending region
154 * addresses fall within the valid range of the map.
156 #define VM_MAP_RANGE_CHECK(map, start, end) \
158 if (start < vm_map_min(map)) \
159 start = vm_map_min(map); \
160 if (end > vm_map_max(map)) \
161 end = vm_map_max(map); \
169 * Initialize the vm_map module. Must be called before
170 * any other vm_map routines.
172 * Map and entry structures are allocated from the general
173 * purpose memory pool with some exceptions:
175 * - The kernel map and kmem submap are allocated statically.
176 * - Kernel map entries are allocated out of a static pool.
178 * These restrictions are necessary since malloc() uses the
179 * maps and requires map entries.
185 mtx_init(&map_sleep_mtx, "vm map sleep mutex", NULL, MTX_DEF);
186 mapzone = uma_zcreate("MAP", sizeof(struct vm_map), NULL,
192 vm_map_zinit, vm_map_zfini, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
193 uma_prealloc(mapzone, MAX_KMAP);
194 kmapentzone = uma_zcreate("KMAP ENTRY", sizeof(struct vm_map_entry),
195 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR,
196 UMA_ZONE_MTXCLASS | UMA_ZONE_VM);
197 uma_prealloc(kmapentzone, MAX_KMAPENT);
198 mapentzone = uma_zcreate("MAP ENTRY", sizeof(struct vm_map_entry),
199 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
203 vmspace_zfini(void *mem, int size)
207 vm = (struct vmspace *)mem;
208 vm_map_zfini(&vm->vm_map, sizeof(vm->vm_map));
212 vmspace_zinit(void *mem, int size, int flags)
216 vm = (struct vmspace *)mem;
218 vm->vm_map.pmap = NULL;
219 (void)vm_map_zinit(&vm->vm_map, sizeof(vm->vm_map), flags);
224 vm_map_zfini(void *mem, int size)
229 mtx_destroy(&map->system_mtx);
230 sx_destroy(&map->lock);
234 vm_map_zinit(void *mem, int size, int flags)
241 mtx_init(&map->system_mtx, "system map", NULL, MTX_DEF | MTX_DUPOK);
242 sx_init(&map->lock, "user map");
248 vmspace_zdtor(void *mem, int size, void *arg)
252 vm = (struct vmspace *)mem;
254 vm_map_zdtor(&vm->vm_map, sizeof(vm->vm_map), arg);
257 vm_map_zdtor(void *mem, int size, void *arg)
262 KASSERT(map->nentries == 0,
263 ("map %p nentries == %d on free.",
264 map, map->nentries));
265 KASSERT(map->size == 0,
266 ("map %p size == %lu on free.",
267 map, (unsigned long)map->size));
269 #endif /* INVARIANTS */
272 * Allocate a vmspace structure, including a vm_map and pmap,
273 * and initialize those structures. The refcnt is set to 1.
276 vmspace_alloc(min, max)
277 vm_offset_t min, max;
281 vm = uma_zalloc(vmspace_zone, M_WAITOK);
282 if (vm->vm_map.pmap == NULL && !pmap_pinit(vmspace_pmap(vm))) {
283 uma_zfree(vmspace_zone, vm);
286 CTR1(KTR_VM, "vmspace_alloc: %p", vm);
287 _vm_map_init(&vm->vm_map, min, max);
288 vm->vm_map.pmap = vmspace_pmap(vm); /* XXX */
304 uma_zone_set_obj(kmapentzone, &kmapentobj, lmin(cnt.v_page_count,
305 (VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS) / PAGE_SIZE) / 8 +
306 maxproc * 2 + maxfiles);
307 vmspace_zone = uma_zcreate("VMSPACE", sizeof(struct vmspace), NULL,
313 vmspace_zinit, vmspace_zfini, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
317 vmspace_dofree(struct vmspace *vm)
319 CTR1(KTR_VM, "vmspace_free: %p", vm);
322 * Make sure any SysV shm is freed, it might not have been in
328 * Lock the map, to wait out all other references to it.
329 * Delete all of the mappings and pages they hold, then call
330 * the pmap module to reclaim anything left.
332 (void)vm_map_remove(&vm->vm_map, vm->vm_map.min_offset,
333 vm->vm_map.max_offset);
336 * XXX Comment out the pmap_release call for now. The
337 * vmspace_zone is marked as UMA_ZONE_NOFREE, and bugs cause
338 * pmap.resident_count to be != 0 on exit sometimes.
340 /* pmap_release(vmspace_pmap(vm)); */
341 uma_zfree(vmspace_zone, vm);
345 vmspace_free(struct vmspace *vm)
348 if (vm->vm_refcnt == 0)
349 panic("vmspace_free: attempt to free already freed vmspace");
351 if (atomic_fetchadd_int(&vm->vm_refcnt, -1) == 1)
356 vmspace_exitfree(struct proc *p)
360 PROC_VMSPACE_LOCK(p);
363 PROC_VMSPACE_UNLOCK(p);
364 KASSERT(vm == &vmspace0, ("vmspace_exitfree: wrong vmspace"));
369 vmspace_exit(struct thread *td)
376 * Release user portion of address space.
377 * This releases references to vnodes,
378 * which could cause I/O if the file has been unlinked.
379 * Need to do this early enough that we can still sleep.
381 * The last exiting process to reach this point releases as
382 * much of the environment as it can. vmspace_dofree() is the
383 * slower fallback in case another process had a temporary
384 * reference to the vmspace.
389 atomic_add_int(&vmspace0.vm_refcnt, 1);
391 refcnt = vm->vm_refcnt;
392 if (refcnt > 1 && p->p_vmspace != &vmspace0) {
393 /* Switch now since other proc might free vmspace */
394 PROC_VMSPACE_LOCK(p);
395 p->p_vmspace = &vmspace0;
396 PROC_VMSPACE_UNLOCK(p);
399 } while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt - 1));
401 if (p->p_vmspace != vm) {
402 /* vmspace not yet freed, switch back */
403 PROC_VMSPACE_LOCK(p);
405 PROC_VMSPACE_UNLOCK(p);
408 pmap_remove_pages(vmspace_pmap(vm));
409 /* Switch now since this proc will free vmspace */
410 PROC_VMSPACE_LOCK(p);
411 p->p_vmspace = &vmspace0;
412 PROC_VMSPACE_UNLOCK(p);
418 /* Acquire reference to vmspace owned by another process. */
421 vmspace_acquire_ref(struct proc *p)
426 PROC_VMSPACE_LOCK(p);
429 PROC_VMSPACE_UNLOCK(p);
433 refcnt = vm->vm_refcnt;
434 if (refcnt <= 0) { /* Avoid 0->1 transition */
435 PROC_VMSPACE_UNLOCK(p);
438 } while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt + 1));
439 if (vm != p->p_vmspace) {
440 PROC_VMSPACE_UNLOCK(p);
444 PROC_VMSPACE_UNLOCK(p);
449 _vm_map_lock(vm_map_t map, const char *file, int line)
453 _mtx_lock_flags(&map->system_mtx, 0, file, line);
455 (void)_sx_xlock(&map->lock, 0, file, line);
460 vm_map_process_deferred(void)
463 vm_map_entry_t entry;
467 while ((entry = td->td_map_def_user) != NULL) {
468 td->td_map_def_user = entry->next;
469 vm_map_entry_deallocate(entry, FALSE);
474 _vm_map_unlock(vm_map_t map, const char *file, int line)
478 _mtx_unlock_flags(&map->system_mtx, 0, file, line);
480 _sx_xunlock(&map->lock, file, line);
481 vm_map_process_deferred();
486 _vm_map_lock_read(vm_map_t map, const char *file, int line)
490 _mtx_lock_flags(&map->system_mtx, 0, file, line);
492 (void)_sx_slock(&map->lock, 0, file, line);
496 _vm_map_unlock_read(vm_map_t map, const char *file, int line)
500 _mtx_unlock_flags(&map->system_mtx, 0, file, line);
502 _sx_sunlock(&map->lock, file, line);
503 vm_map_process_deferred();
508 _vm_map_trylock(vm_map_t map, const char *file, int line)
512 error = map->system_map ?
513 !_mtx_trylock(&map->system_mtx, 0, file, line) :
514 !_sx_try_xlock(&map->lock, file, line);
521 _vm_map_trylock_read(vm_map_t map, const char *file, int line)
525 error = map->system_map ?
526 !_mtx_trylock(&map->system_mtx, 0, file, line) :
527 !_sx_try_slock(&map->lock, file, line);
532 * _vm_map_lock_upgrade: [ internal use only ]
534 * Tries to upgrade a read (shared) lock on the specified map to a write
535 * (exclusive) lock. Returns the value "0" if the upgrade succeeds and a
536 * non-zero value if the upgrade fails. If the upgrade fails, the map is
537 * returned without a read or write lock held.
539 * Requires that the map be read locked.
542 _vm_map_lock_upgrade(vm_map_t map, const char *file, int line)
544 unsigned int last_timestamp;
546 if (map->system_map) {
548 _mtx_assert(&map->system_mtx, MA_OWNED, file, line);
551 if (!_sx_try_upgrade(&map->lock, file, line)) {
552 last_timestamp = map->timestamp;
553 _sx_sunlock(&map->lock, file, line);
554 vm_map_process_deferred();
556 * If the map's timestamp does not change while the
557 * map is unlocked, then the upgrade succeeds.
559 (void)_sx_xlock(&map->lock, 0, file, line);
560 if (last_timestamp != map->timestamp) {
561 _sx_xunlock(&map->lock, file, line);
571 _vm_map_lock_downgrade(vm_map_t map, const char *file, int line)
574 if (map->system_map) {
576 _mtx_assert(&map->system_mtx, MA_OWNED, file, line);
579 _sx_downgrade(&map->lock, file, line);
585 * Returns a non-zero value if the caller holds a write (exclusive) lock
586 * on the specified map and the value "0" otherwise.
589 vm_map_locked(vm_map_t map)
593 return (mtx_owned(&map->system_mtx));
595 return (sx_xlocked(&map->lock));
600 _vm_map_assert_locked(vm_map_t map, const char *file, int line)
604 _mtx_assert(&map->system_mtx, MA_OWNED, file, line);
606 _sx_assert(&map->lock, SA_XLOCKED, file, line);
611 _vm_map_assert_locked_read(vm_map_t map, const char *file, int line)
615 _mtx_assert(&map->system_mtx, MA_OWNED, file, line);
617 _sx_assert(&map->lock, SA_SLOCKED, file, line);
621 #define VM_MAP_ASSERT_LOCKED(map) \
622 _vm_map_assert_locked(map, LOCK_FILE, LOCK_LINE)
623 #define VM_MAP_ASSERT_LOCKED_READ(map) \
624 _vm_map_assert_locked_read(map, LOCK_FILE, LOCK_LINE)
626 #define VM_MAP_ASSERT_LOCKED(map)
627 #define VM_MAP_ASSERT_LOCKED_READ(map)
631 * _vm_map_unlock_and_wait:
633 * Atomically releases the lock on the specified map and puts the calling
634 * thread to sleep. The calling thread will remain asleep until either
635 * vm_map_wakeup() is performed on the map or the specified timeout is
638 * WARNING! This function does not perform deferred deallocations of
639 * objects and map entries. Therefore, the calling thread is expected to
640 * reacquire the map lock after reawakening and later perform an ordinary
641 * unlock operation, such as vm_map_unlock(), before completing its
642 * operation on the map.
645 _vm_map_unlock_and_wait(vm_map_t map, int timo, const char *file, int line)
648 mtx_lock(&map_sleep_mtx);
650 _mtx_unlock_flags(&map->system_mtx, 0, file, line);
652 _sx_xunlock(&map->lock, file, line);
653 return (msleep(&map->root, &map_sleep_mtx, PDROP | PVM, "vmmaps",
660 * Awaken any threads that have slept on the map using
661 * vm_map_unlock_and_wait().
664 vm_map_wakeup(vm_map_t map)
668 * Acquire and release map_sleep_mtx to prevent a wakeup()
669 * from being performed (and lost) between the map unlock
670 * and the msleep() in _vm_map_unlock_and_wait().
672 mtx_lock(&map_sleep_mtx);
673 mtx_unlock(&map_sleep_mtx);
678 vm_map_busy(vm_map_t map)
681 VM_MAP_ASSERT_LOCKED(map);
686 vm_map_unbusy(vm_map_t map)
689 VM_MAP_ASSERT_LOCKED(map);
690 KASSERT(map->busy, ("vm_map_unbusy: not busy"));
691 if (--map->busy == 0 && (map->flags & MAP_BUSY_WAKEUP)) {
692 vm_map_modflags(map, 0, MAP_BUSY_WAKEUP);
698 vm_map_wait_busy(vm_map_t map)
701 VM_MAP_ASSERT_LOCKED(map);
703 vm_map_modflags(map, MAP_BUSY_WAKEUP, 0);
705 msleep(&map->busy, &map->system_mtx, 0, "mbusy", 0);
707 sx_sleep(&map->busy, &map->lock, 0, "mbusy", 0);
713 vmspace_resident_count(struct vmspace *vmspace)
715 return pmap_resident_count(vmspace_pmap(vmspace));
719 vmspace_wired_count(struct vmspace *vmspace)
721 return pmap_wired_count(vmspace_pmap(vmspace));
727 * Creates and returns a new empty VM map with
728 * the given physical map structure, and having
729 * the given lower and upper address bounds.
732 vm_map_create(pmap_t pmap, vm_offset_t min, vm_offset_t max)
736 result = uma_zalloc(mapzone, M_WAITOK);
737 CTR1(KTR_VM, "vm_map_create: %p", result);
738 _vm_map_init(result, min, max);
744 * Initialize an existing vm_map structure
745 * such as that in the vmspace structure.
746 * The pmap is set elsewhere.
749 _vm_map_init(vm_map_t map, vm_offset_t min, vm_offset_t max)
752 map->header.next = map->header.prev = &map->header;
753 map->needs_wakeup = FALSE;
755 map->min_offset = min;
756 map->max_offset = max;
764 vm_map_init(vm_map_t map, vm_offset_t min, vm_offset_t max)
766 _vm_map_init(map, min, max);
767 mtx_init(&map->system_mtx, "system map", NULL, MTX_DEF | MTX_DUPOK);
768 sx_init(&map->lock, "user map");
772 * vm_map_entry_dispose: [ internal use only ]
774 * Inverse of vm_map_entry_create.
777 vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry)
779 uma_zfree(map->system_map ? kmapentzone : mapentzone, entry);
783 * vm_map_entry_create: [ internal use only ]
785 * Allocates a VM map entry for insertion.
786 * No entry fields are filled in.
788 static vm_map_entry_t
789 vm_map_entry_create(vm_map_t map)
791 vm_map_entry_t new_entry;
794 new_entry = uma_zalloc(kmapentzone, M_NOWAIT);
796 new_entry = uma_zalloc(mapentzone, M_WAITOK);
797 if (new_entry == NULL)
798 panic("vm_map_entry_create: kernel resources exhausted");
803 * vm_map_entry_set_behavior:
805 * Set the expected access behavior, either normal, random, or
809 vm_map_entry_set_behavior(vm_map_entry_t entry, u_char behavior)
811 entry->eflags = (entry->eflags & ~MAP_ENTRY_BEHAV_MASK) |
812 (behavior & MAP_ENTRY_BEHAV_MASK);
816 * vm_map_entry_set_max_free:
818 * Set the max_free field in a vm_map_entry.
821 vm_map_entry_set_max_free(vm_map_entry_t entry)
824 entry->max_free = entry->adj_free;
825 if (entry->left != NULL && entry->left->max_free > entry->max_free)
826 entry->max_free = entry->left->max_free;
827 if (entry->right != NULL && entry->right->max_free > entry->max_free)
828 entry->max_free = entry->right->max_free;
832 * vm_map_entry_splay:
834 * The Sleator and Tarjan top-down splay algorithm with the
835 * following variation. Max_free must be computed bottom-up, so
836 * on the downward pass, maintain the left and right spines in
837 * reverse order. Then, make a second pass up each side to fix
838 * the pointers and compute max_free. The time bound is O(log n)
841 * The new root is the vm_map_entry containing "addr", or else an
842 * adjacent entry (lower or higher) if addr is not in the tree.
844 * The map must be locked, and leaves it so.
846 * Returns: the new root.
848 static vm_map_entry_t
849 vm_map_entry_splay(vm_offset_t addr, vm_map_entry_t root)
851 vm_map_entry_t llist, rlist;
852 vm_map_entry_t ltree, rtree;
855 /* Special case of empty tree. */
860 * Pass One: Splay down the tree until we find addr or a NULL
861 * pointer where addr would go. llist and rlist are the two
862 * sides in reverse order (bottom-up), with llist linked by
863 * the right pointer and rlist linked by the left pointer in
864 * the vm_map_entry. Wait until Pass Two to set max_free on
870 /* root is never NULL in here. */
871 if (addr < root->start) {
875 if (addr < y->start && y->left != NULL) {
876 /* Rotate right and put y on rlist. */
877 root->left = y->right;
879 vm_map_entry_set_max_free(root);
884 /* Put root on rlist. */
889 } else if (addr >= root->end) {
893 if (addr >= y->end && y->right != NULL) {
894 /* Rotate left and put y on llist. */
895 root->right = y->left;
897 vm_map_entry_set_max_free(root);
902 /* Put root on llist. */
912 * Pass Two: Walk back up the two spines, flip the pointers
913 * and set max_free. The subtrees of the root go at the
914 * bottom of llist and rlist.
917 while (llist != NULL) {
919 llist->right = ltree;
920 vm_map_entry_set_max_free(llist);
925 while (rlist != NULL) {
928 vm_map_entry_set_max_free(rlist);
934 * Final assembly: add ltree and rtree as subtrees of root.
938 vm_map_entry_set_max_free(root);
944 * vm_map_entry_{un,}link:
946 * Insert/remove entries from maps.
949 vm_map_entry_link(vm_map_t map,
950 vm_map_entry_t after_where,
951 vm_map_entry_t entry)
955 "vm_map_entry_link: map %p, nentries %d, entry %p, after %p", map,
956 map->nentries, entry, after_where);
957 VM_MAP_ASSERT_LOCKED(map);
959 entry->prev = after_where;
960 entry->next = after_where->next;
961 entry->next->prev = entry;
962 after_where->next = entry;
964 if (after_where != &map->header) {
965 if (after_where != map->root)
966 vm_map_entry_splay(after_where->start, map->root);
967 entry->right = after_where->right;
968 entry->left = after_where;
969 after_where->right = NULL;
970 after_where->adj_free = entry->start - after_where->end;
971 vm_map_entry_set_max_free(after_where);
973 entry->right = map->root;
976 entry->adj_free = (entry->next == &map->header ? map->max_offset :
977 entry->next->start) - entry->end;
978 vm_map_entry_set_max_free(entry);
983 vm_map_entry_unlink(vm_map_t map,
984 vm_map_entry_t entry)
986 vm_map_entry_t next, prev, root;
988 VM_MAP_ASSERT_LOCKED(map);
989 if (entry != map->root)
990 vm_map_entry_splay(entry->start, map->root);
991 if (entry->left == NULL)
994 root = vm_map_entry_splay(entry->start, entry->left);
995 root->right = entry->right;
996 root->adj_free = (entry->next == &map->header ? map->max_offset :
997 entry->next->start) - root->end;
998 vm_map_entry_set_max_free(root);
1007 CTR3(KTR_VM, "vm_map_entry_unlink: map %p, nentries %d, entry %p", map,
1008 map->nentries, entry);
1012 * vm_map_entry_resize_free:
1014 * Recompute the amount of free space following a vm_map_entry
1015 * and propagate that value up the tree. Call this function after
1016 * resizing a map entry in-place, that is, without a call to
1017 * vm_map_entry_link() or _unlink().
1019 * The map must be locked, and leaves it so.
1022 vm_map_entry_resize_free(vm_map_t map, vm_map_entry_t entry)
1026 * Using splay trees without parent pointers, propagating
1027 * max_free up the tree is done by moving the entry to the
1028 * root and making the change there.
1030 if (entry != map->root)
1031 map->root = vm_map_entry_splay(entry->start, map->root);
1033 entry->adj_free = (entry->next == &map->header ? map->max_offset :
1034 entry->next->start) - entry->end;
1035 vm_map_entry_set_max_free(entry);
1039 * vm_map_lookup_entry: [ internal use only ]
1041 * Finds the map entry containing (or
1042 * immediately preceding) the specified address
1043 * in the given map; the entry is returned
1044 * in the "entry" parameter. The boolean
1045 * result indicates whether the address is
1046 * actually contained in the map.
1049 vm_map_lookup_entry(
1051 vm_offset_t address,
1052 vm_map_entry_t *entry) /* OUT */
1058 * If the map is empty, then the map entry immediately preceding
1059 * "address" is the map's header.
1063 *entry = &map->header;
1064 else if (address >= cur->start && cur->end > address) {
1067 } else if ((locked = vm_map_locked(map)) ||
1068 sx_try_upgrade(&map->lock)) {
1070 * Splay requires a write lock on the map. However, it only
1071 * restructures the binary search tree; it does not otherwise
1072 * change the map. Thus, the map's timestamp need not change
1073 * on a temporary upgrade.
1075 map->root = cur = vm_map_entry_splay(address, cur);
1077 sx_downgrade(&map->lock);
1080 * If "address" is contained within a map entry, the new root
1081 * is that map entry. Otherwise, the new root is a map entry
1082 * immediately before or after "address".
1084 if (address >= cur->start) {
1086 if (cur->end > address)
1092 * Since the map is only locked for read access, perform a
1093 * standard binary search tree lookup for "address".
1096 if (address < cur->start) {
1097 if (cur->left == NULL) {
1102 } else if (cur->end > address) {
1106 if (cur->right == NULL) {
1119 * Inserts the given whole VM object into the target
1120 * map at the specified address range. The object's
1121 * size should match that of the address range.
1123 * Requires that the map be locked, and leaves it so.
1125 * If object is non-NULL, ref count must be bumped by caller
1126 * prior to making call to account for the new entry.
1129 vm_map_insert(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1130 vm_offset_t start, vm_offset_t end, vm_prot_t prot, vm_prot_t max,
1133 vm_map_entry_t new_entry;
1134 vm_map_entry_t prev_entry;
1135 vm_map_entry_t temp_entry;
1136 vm_eflags_t protoeflags;
1137 struct uidinfo *uip;
1138 boolean_t charge_prev_obj;
1140 VM_MAP_ASSERT_LOCKED(map);
1143 * Check that the start and end points are not bogus.
1145 if ((start < map->min_offset) || (end > map->max_offset) ||
1147 return (KERN_INVALID_ADDRESS);
1150 * Find the entry prior to the proposed starting address; if it's part
1151 * of an existing entry, this range is bogus.
1153 if (vm_map_lookup_entry(map, start, &temp_entry))
1154 return (KERN_NO_SPACE);
1156 prev_entry = temp_entry;
1159 * Assert that the next entry doesn't overlap the end point.
1161 if ((prev_entry->next != &map->header) &&
1162 (prev_entry->next->start < end))
1163 return (KERN_NO_SPACE);
1166 charge_prev_obj = FALSE;
1168 if (cow & MAP_COPY_ON_WRITE)
1169 protoeflags |= MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY;
1171 if (cow & MAP_NOFAULT) {
1172 protoeflags |= MAP_ENTRY_NOFAULT;
1174 KASSERT(object == NULL,
1175 ("vm_map_insert: paradoxical MAP_NOFAULT request"));
1177 if (cow & MAP_DISABLE_SYNCER)
1178 protoeflags |= MAP_ENTRY_NOSYNC;
1179 if (cow & MAP_DISABLE_COREDUMP)
1180 protoeflags |= MAP_ENTRY_NOCOREDUMP;
1183 KASSERT((object != kmem_object && object != kernel_object) ||
1184 ((object == kmem_object || object == kernel_object) &&
1185 !(protoeflags & MAP_ENTRY_NEEDS_COPY)),
1186 ("kmem or kernel object and cow"));
1187 if (cow & (MAP_ACC_NO_CHARGE | MAP_NOFAULT))
1189 if ((cow & MAP_ACC_CHARGED) || ((prot & VM_PROT_WRITE) &&
1190 ((protoeflags & MAP_ENTRY_NEEDS_COPY) || object == NULL))) {
1191 if (!(cow & MAP_ACC_CHARGED) && !swap_reserve(end - start))
1192 return (KERN_RESOURCE_SHORTAGE);
1193 KASSERT(object == NULL || (protoeflags & MAP_ENTRY_NEEDS_COPY) ||
1194 object->uip == NULL,
1195 ("OVERCOMMIT: vm_map_insert o %p", object));
1196 uip = curthread->td_ucred->cr_ruidinfo;
1198 if (object == NULL && !(protoeflags & MAP_ENTRY_NEEDS_COPY))
1199 charge_prev_obj = TRUE;
1203 /* Expand the kernel pmap, if necessary. */
1204 if (map == kernel_map && end > kernel_vm_end)
1205 pmap_growkernel(end);
1206 if (object != NULL) {
1208 * OBJ_ONEMAPPING must be cleared unless this mapping
1209 * is trivially proven to be the only mapping for any
1210 * of the object's pages. (Object granularity
1211 * reference counting is insufficient to recognize
1212 * aliases with precision.)
1214 VM_OBJECT_LOCK(object);
1215 if (object->ref_count > 1 || object->shadow_count != 0)
1216 vm_object_clear_flag(object, OBJ_ONEMAPPING);
1217 VM_OBJECT_UNLOCK(object);
1219 else if ((prev_entry != &map->header) &&
1220 (prev_entry->eflags == protoeflags) &&
1221 (prev_entry->end == start) &&
1222 (prev_entry->wired_count == 0) &&
1223 (prev_entry->uip == uip ||
1224 (prev_entry->object.vm_object != NULL &&
1225 (prev_entry->object.vm_object->uip == uip))) &&
1226 vm_object_coalesce(prev_entry->object.vm_object,
1228 (vm_size_t)(prev_entry->end - prev_entry->start),
1229 (vm_size_t)(end - prev_entry->end), charge_prev_obj)) {
1231 * We were able to extend the object. Determine if we
1232 * can extend the previous map entry to include the
1233 * new range as well.
1235 if ((prev_entry->inheritance == VM_INHERIT_DEFAULT) &&
1236 (prev_entry->protection == prot) &&
1237 (prev_entry->max_protection == max)) {
1238 map->size += (end - prev_entry->end);
1239 prev_entry->end = end;
1240 vm_map_entry_resize_free(map, prev_entry);
1241 vm_map_simplify_entry(map, prev_entry);
1244 return (KERN_SUCCESS);
1248 * If we can extend the object but cannot extend the
1249 * map entry, we have to create a new map entry. We
1250 * must bump the ref count on the extended object to
1251 * account for it. object may be NULL.
1253 object = prev_entry->object.vm_object;
1254 offset = prev_entry->offset +
1255 (prev_entry->end - prev_entry->start);
1256 vm_object_reference(object);
1257 if (uip != NULL && object != NULL && object->uip != NULL &&
1258 !(prev_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
1259 /* Object already accounts for this uid. */
1266 * NOTE: if conditionals fail, object can be NULL here. This occurs
1267 * in things like the buffer map where we manage kva but do not manage
1272 * Create a new entry
1274 new_entry = vm_map_entry_create(map);
1275 new_entry->start = start;
1276 new_entry->end = end;
1277 new_entry->uip = NULL;
1279 new_entry->eflags = protoeflags;
1280 new_entry->object.vm_object = object;
1281 new_entry->offset = offset;
1282 new_entry->avail_ssize = 0;
1284 new_entry->inheritance = VM_INHERIT_DEFAULT;
1285 new_entry->protection = prot;
1286 new_entry->max_protection = max;
1287 new_entry->wired_count = 0;
1289 KASSERT(uip == NULL || !ENTRY_CHARGED(new_entry),
1290 ("OVERCOMMIT: vm_map_insert leaks vm_map %p", new_entry));
1291 new_entry->uip = uip;
1294 * Insert the new entry into the list
1296 vm_map_entry_link(map, prev_entry, new_entry);
1297 map->size += new_entry->end - new_entry->start;
1301 * Temporarily removed to avoid MAP_STACK panic, due to
1302 * MAP_STACK being a huge hack. Will be added back in
1303 * when MAP_STACK (and the user stack mapping) is fixed.
1306 * It may be possible to simplify the entry
1308 vm_map_simplify_entry(map, new_entry);
1311 if (cow & (MAP_PREFAULT|MAP_PREFAULT_PARTIAL)) {
1312 vm_map_pmap_enter(map, start, prot,
1313 object, OFF_TO_IDX(offset), end - start,
1314 cow & MAP_PREFAULT_PARTIAL);
1317 return (KERN_SUCCESS);
1323 * Find the first fit (lowest VM address) for "length" free bytes
1324 * beginning at address >= start in the given map.
1326 * In a vm_map_entry, "adj_free" is the amount of free space
1327 * adjacent (higher address) to this entry, and "max_free" is the
1328 * maximum amount of contiguous free space in its subtree. This
1329 * allows finding a free region in one path down the tree, so
1330 * O(log n) amortized with splay trees.
1332 * The map must be locked, and leaves it so.
1334 * Returns: 0 on success, and starting address in *addr,
1335 * 1 if insufficient space.
1338 vm_map_findspace(vm_map_t map, vm_offset_t start, vm_size_t length,
1339 vm_offset_t *addr) /* OUT */
1341 vm_map_entry_t entry;
1345 * Request must fit within min/max VM address and must avoid
1348 if (start < map->min_offset)
1349 start = map->min_offset;
1350 if (start + length > map->max_offset || start + length < start)
1353 /* Empty tree means wide open address space. */
1354 if (map->root == NULL) {
1360 * After splay, if start comes before root node, then there
1361 * must be a gap from start to the root.
1363 map->root = vm_map_entry_splay(start, map->root);
1364 if (start + length <= map->root->start) {
1370 * Root is the last node that might begin its gap before
1371 * start, and this is the last comparison where address
1372 * wrap might be a problem.
1374 st = (start > map->root->end) ? start : map->root->end;
1375 if (length <= map->root->end + map->root->adj_free - st) {
1380 /* With max_free, can immediately tell if no solution. */
1381 entry = map->root->right;
1382 if (entry == NULL || length > entry->max_free)
1386 * Search the right subtree in the order: left subtree, root,
1387 * right subtree (first fit). The previous splay implies that
1388 * all regions in the right subtree have addresses > start.
1390 while (entry != NULL) {
1391 if (entry->left != NULL && entry->left->max_free >= length)
1392 entry = entry->left;
1393 else if (entry->adj_free >= length) {
1397 entry = entry->right;
1400 /* Can't get here, so panic if we do. */
1401 panic("vm_map_findspace: max_free corrupt");
1405 vm_map_fixed(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1406 vm_offset_t start, vm_size_t length, vm_prot_t prot,
1407 vm_prot_t max, int cow)
1412 end = start + length;
1414 VM_MAP_RANGE_CHECK(map, start, end);
1415 (void) vm_map_delete(map, start, end);
1416 result = vm_map_insert(map, object, offset, start, end, prot,
1423 * vm_map_find finds an unallocated region in the target address
1424 * map with the given length. The search is defined to be
1425 * first-fit from the specified address; the region found is
1426 * returned in the same parameter.
1428 * If object is non-NULL, ref count must be bumped by caller
1429 * prior to making call to account for the new entry.
1432 vm_map_find(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1433 vm_offset_t *addr, /* IN/OUT */
1434 vm_size_t length, int find_space, vm_prot_t prot,
1435 vm_prot_t max, int cow)
1443 if (find_space != VMFS_NO_SPACE) {
1444 if (vm_map_findspace(map, start, length, addr)) {
1446 return (KERN_NO_SPACE);
1448 switch (find_space) {
1449 case VMFS_ALIGNED_SPACE:
1450 pmap_align_superpage(object, offset, addr,
1453 #ifdef VMFS_TLB_ALIGNED_SPACE
1454 case VMFS_TLB_ALIGNED_SPACE:
1455 pmap_align_tlb(addr);
1464 result = vm_map_insert(map, object, offset, start, start +
1465 length, prot, max, cow);
1466 } while (result == KERN_NO_SPACE && find_space == VMFS_ALIGNED_SPACE);
1472 * vm_map_simplify_entry:
1474 * Simplify the given map entry by merging with either neighbor. This
1475 * routine also has the ability to merge with both neighbors.
1477 * The map must be locked.
1479 * This routine guarentees that the passed entry remains valid (though
1480 * possibly extended). When merging, this routine may delete one or
1484 vm_map_simplify_entry(vm_map_t map, vm_map_entry_t entry)
1486 vm_map_entry_t next, prev;
1487 vm_size_t prevsize, esize;
1489 if (entry->eflags & (MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_IS_SUB_MAP))
1493 if (prev != &map->header) {
1494 prevsize = prev->end - prev->start;
1495 if ( (prev->end == entry->start) &&
1496 (prev->object.vm_object == entry->object.vm_object) &&
1497 (!prev->object.vm_object ||
1498 (prev->offset + prevsize == entry->offset)) &&
1499 (prev->eflags == entry->eflags) &&
1500 (prev->protection == entry->protection) &&
1501 (prev->max_protection == entry->max_protection) &&
1502 (prev->inheritance == entry->inheritance) &&
1503 (prev->wired_count == entry->wired_count) &&
1504 (prev->uip == entry->uip)) {
1505 vm_map_entry_unlink(map, prev);
1506 entry->start = prev->start;
1507 entry->offset = prev->offset;
1508 if (entry->prev != &map->header)
1509 vm_map_entry_resize_free(map, entry->prev);
1512 * If the backing object is a vnode object,
1513 * vm_object_deallocate() calls vrele().
1514 * However, vrele() does not lock the vnode
1515 * because the vnode has additional
1516 * references. Thus, the map lock can be kept
1517 * without causing a lock-order reversal with
1520 if (prev->object.vm_object)
1521 vm_object_deallocate(prev->object.vm_object);
1522 if (prev->uip != NULL)
1524 vm_map_entry_dispose(map, prev);
1529 if (next != &map->header) {
1530 esize = entry->end - entry->start;
1531 if ((entry->end == next->start) &&
1532 (next->object.vm_object == entry->object.vm_object) &&
1533 (!entry->object.vm_object ||
1534 (entry->offset + esize == next->offset)) &&
1535 (next->eflags == entry->eflags) &&
1536 (next->protection == entry->protection) &&
1537 (next->max_protection == entry->max_protection) &&
1538 (next->inheritance == entry->inheritance) &&
1539 (next->wired_count == entry->wired_count) &&
1540 (next->uip == entry->uip)) {
1541 vm_map_entry_unlink(map, next);
1542 entry->end = next->end;
1543 vm_map_entry_resize_free(map, entry);
1546 * See comment above.
1548 if (next->object.vm_object)
1549 vm_object_deallocate(next->object.vm_object);
1550 if (next->uip != NULL)
1552 vm_map_entry_dispose(map, next);
1557 * vm_map_clip_start: [ internal use only ]
1559 * Asserts that the given entry begins at or after
1560 * the specified address; if necessary,
1561 * it splits the entry into two.
1563 #define vm_map_clip_start(map, entry, startaddr) \
1565 if (startaddr > entry->start) \
1566 _vm_map_clip_start(map, entry, startaddr); \
1570 * This routine is called only when it is known that
1571 * the entry must be split.
1574 _vm_map_clip_start(vm_map_t map, vm_map_entry_t entry, vm_offset_t start)
1576 vm_map_entry_t new_entry;
1578 VM_MAP_ASSERT_LOCKED(map);
1581 * Split off the front portion -- note that we must insert the new
1582 * entry BEFORE this one, so that this entry has the specified
1585 vm_map_simplify_entry(map, entry);
1588 * If there is no object backing this entry, we might as well create
1589 * one now. If we defer it, an object can get created after the map
1590 * is clipped, and individual objects will be created for the split-up
1591 * map. This is a bit of a hack, but is also about the best place to
1592 * put this improvement.
1594 if (entry->object.vm_object == NULL && !map->system_map) {
1596 object = vm_object_allocate(OBJT_DEFAULT,
1597 atop(entry->end - entry->start));
1598 entry->object.vm_object = object;
1600 if (entry->uip != NULL) {
1601 object->uip = entry->uip;
1602 object->charge = entry->end - entry->start;
1605 } else if (entry->object.vm_object != NULL &&
1606 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) &&
1607 entry->uip != NULL) {
1608 VM_OBJECT_LOCK(entry->object.vm_object);
1609 KASSERT(entry->object.vm_object->uip == NULL,
1610 ("OVERCOMMIT: vm_entry_clip_start: both uip e %p", entry));
1611 entry->object.vm_object->uip = entry->uip;
1612 entry->object.vm_object->charge = entry->end - entry->start;
1613 VM_OBJECT_UNLOCK(entry->object.vm_object);
1617 new_entry = vm_map_entry_create(map);
1618 *new_entry = *entry;
1620 new_entry->end = start;
1621 entry->offset += (start - entry->start);
1622 entry->start = start;
1623 if (new_entry->uip != NULL)
1626 vm_map_entry_link(map, entry->prev, new_entry);
1628 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1629 vm_object_reference(new_entry->object.vm_object);
1634 * vm_map_clip_end: [ internal use only ]
1636 * Asserts that the given entry ends at or before
1637 * the specified address; if necessary,
1638 * it splits the entry into two.
1640 #define vm_map_clip_end(map, entry, endaddr) \
1642 if ((endaddr) < (entry->end)) \
1643 _vm_map_clip_end((map), (entry), (endaddr)); \
1647 * This routine is called only when it is known that
1648 * the entry must be split.
1651 _vm_map_clip_end(vm_map_t map, vm_map_entry_t entry, vm_offset_t end)
1653 vm_map_entry_t new_entry;
1655 VM_MAP_ASSERT_LOCKED(map);
1658 * If there is no object backing this entry, we might as well create
1659 * one now. If we defer it, an object can get created after the map
1660 * is clipped, and individual objects will be created for the split-up
1661 * map. This is a bit of a hack, but is also about the best place to
1662 * put this improvement.
1664 if (entry->object.vm_object == NULL && !map->system_map) {
1666 object = vm_object_allocate(OBJT_DEFAULT,
1667 atop(entry->end - entry->start));
1668 entry->object.vm_object = object;
1670 if (entry->uip != NULL) {
1671 object->uip = entry->uip;
1672 object->charge = entry->end - entry->start;
1675 } else if (entry->object.vm_object != NULL &&
1676 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) &&
1677 entry->uip != NULL) {
1678 VM_OBJECT_LOCK(entry->object.vm_object);
1679 KASSERT(entry->object.vm_object->uip == NULL,
1680 ("OVERCOMMIT: vm_entry_clip_end: both uip e %p", entry));
1681 entry->object.vm_object->uip = entry->uip;
1682 entry->object.vm_object->charge = entry->end - entry->start;
1683 VM_OBJECT_UNLOCK(entry->object.vm_object);
1688 * Create a new entry and insert it AFTER the specified entry
1690 new_entry = vm_map_entry_create(map);
1691 *new_entry = *entry;
1693 new_entry->start = entry->end = end;
1694 new_entry->offset += (end - entry->start);
1695 if (new_entry->uip != NULL)
1698 vm_map_entry_link(map, entry, new_entry);
1700 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1701 vm_object_reference(new_entry->object.vm_object);
1706 * vm_map_submap: [ kernel use only ]
1708 * Mark the given range as handled by a subordinate map.
1710 * This range must have been created with vm_map_find,
1711 * and no other operations may have been performed on this
1712 * range prior to calling vm_map_submap.
1714 * Only a limited number of operations can be performed
1715 * within this rage after calling vm_map_submap:
1717 * [Don't try vm_map_copy!]
1719 * To remove a submapping, one must first remove the
1720 * range from the superior map, and then destroy the
1721 * submap (if desired). [Better yet, don't try it.]
1730 vm_map_entry_t entry;
1731 int result = KERN_INVALID_ARGUMENT;
1735 VM_MAP_RANGE_CHECK(map, start, end);
1737 if (vm_map_lookup_entry(map, start, &entry)) {
1738 vm_map_clip_start(map, entry, start);
1740 entry = entry->next;
1742 vm_map_clip_end(map, entry, end);
1744 if ((entry->start == start) && (entry->end == end) &&
1745 ((entry->eflags & MAP_ENTRY_COW) == 0) &&
1746 (entry->object.vm_object == NULL)) {
1747 entry->object.sub_map = submap;
1748 entry->eflags |= MAP_ENTRY_IS_SUB_MAP;
1749 result = KERN_SUCCESS;
1757 * The maximum number of pages to map
1759 #define MAX_INIT_PT 96
1762 * vm_map_pmap_enter:
1764 * Preload read-only mappings for the given object's resident pages into
1765 * the given map. This eliminates the soft faults on process startup and
1766 * immediately after an mmap(2). Because these are speculative mappings,
1767 * cached pages are not reactivated and mapped.
1770 vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot,
1771 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags)
1774 vm_page_t p, p_start;
1775 vm_pindex_t psize, tmpidx;
1776 boolean_t are_queues_locked;
1778 if ((prot & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0 || object == NULL)
1780 VM_OBJECT_LOCK(object);
1781 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) {
1782 pmap_object_init_pt(map->pmap, addr, object, pindex, size);
1788 if ((flags & MAP_PREFAULT_PARTIAL) && psize > MAX_INIT_PT &&
1789 object->resident_page_count > MAX_INIT_PT)
1792 if (psize + pindex > object->size) {
1793 if (object->size < pindex)
1795 psize = object->size - pindex;
1798 are_queues_locked = FALSE;
1802 p = vm_page_find_least(object, pindex);
1804 * Assert: the variable p is either (1) the page with the
1805 * least pindex greater than or equal to the parameter pindex
1809 p != NULL && (tmpidx = p->pindex - pindex) < psize;
1810 p = TAILQ_NEXT(p, listq)) {
1812 * don't allow an madvise to blow away our really
1813 * free pages allocating pv entries.
1815 if ((flags & MAP_PREFAULT_MADVISE) &&
1816 cnt.v_free_count < cnt.v_free_reserved) {
1820 if (p->valid == VM_PAGE_BITS_ALL) {
1821 if (p_start == NULL) {
1822 start = addr + ptoa(tmpidx);
1825 } else if (p_start != NULL) {
1826 if (!are_queues_locked) {
1827 are_queues_locked = TRUE;
1828 vm_page_lock_queues();
1830 pmap_enter_object(map->pmap, start, addr +
1831 ptoa(tmpidx), p_start, prot);
1835 if (p_start != NULL) {
1836 if (!are_queues_locked) {
1837 are_queues_locked = TRUE;
1838 vm_page_lock_queues();
1840 pmap_enter_object(map->pmap, start, addr + ptoa(psize),
1843 if (are_queues_locked)
1844 vm_page_unlock_queues();
1846 VM_OBJECT_UNLOCK(object);
1852 * Sets the protection of the specified address
1853 * region in the target map. If "set_max" is
1854 * specified, the maximum protection is to be set;
1855 * otherwise, only the current protection is affected.
1858 vm_map_protect(vm_map_t map, vm_offset_t start, vm_offset_t end,
1859 vm_prot_t new_prot, boolean_t set_max)
1861 vm_map_entry_t current, entry;
1863 struct uidinfo *uip;
1868 VM_MAP_RANGE_CHECK(map, start, end);
1870 if (vm_map_lookup_entry(map, start, &entry)) {
1871 vm_map_clip_start(map, entry, start);
1873 entry = entry->next;
1877 * Make a first pass to check for protection violations.
1880 while ((current != &map->header) && (current->start < end)) {
1881 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
1883 return (KERN_INVALID_ARGUMENT);
1885 if ((new_prot & current->max_protection) != new_prot) {
1887 return (KERN_PROTECTION_FAILURE);
1889 current = current->next;
1894 * Do an accounting pass for private read-only mappings that
1895 * now will do cow due to allowed write (e.g. debugger sets
1896 * breakpoint on text segment)
1898 for (current = entry; (current != &map->header) &&
1899 (current->start < end); current = current->next) {
1901 vm_map_clip_end(map, current, end);
1904 ((new_prot & ~(current->protection)) & VM_PROT_WRITE) == 0 ||
1905 ENTRY_CHARGED(current)) {
1909 uip = curthread->td_ucred->cr_ruidinfo;
1910 obj = current->object.vm_object;
1912 if (obj == NULL || (current->eflags & MAP_ENTRY_NEEDS_COPY)) {
1913 if (!swap_reserve(current->end - current->start)) {
1915 return (KERN_RESOURCE_SHORTAGE);
1922 VM_OBJECT_LOCK(obj);
1923 if (obj->type != OBJT_DEFAULT && obj->type != OBJT_SWAP) {
1924 VM_OBJECT_UNLOCK(obj);
1929 * Charge for the whole object allocation now, since
1930 * we cannot distinguish between non-charged and
1931 * charged clipped mapping of the same object later.
1933 KASSERT(obj->charge == 0,
1934 ("vm_map_protect: object %p overcharged\n", obj));
1935 if (!swap_reserve(ptoa(obj->size))) {
1936 VM_OBJECT_UNLOCK(obj);
1938 return (KERN_RESOURCE_SHORTAGE);
1943 obj->charge = ptoa(obj->size);
1944 VM_OBJECT_UNLOCK(obj);
1948 * Go back and fix up protections. [Note that clipping is not
1949 * necessary the second time.]
1952 while ((current != &map->header) && (current->start < end)) {
1953 old_prot = current->protection;
1956 current->protection =
1957 (current->max_protection = new_prot) &
1960 current->protection = new_prot;
1962 if ((current->eflags & (MAP_ENTRY_COW | MAP_ENTRY_USER_WIRED))
1963 == (MAP_ENTRY_COW | MAP_ENTRY_USER_WIRED) &&
1964 (current->protection & VM_PROT_WRITE) != 0 &&
1965 (old_prot & VM_PROT_WRITE) == 0) {
1966 vm_fault_copy_entry(map, map, current, current, NULL);
1970 * Update physical map if necessary. Worry about copy-on-write
1973 if (current->protection != old_prot) {
1974 #define MASK(entry) (((entry)->eflags & MAP_ENTRY_COW) ? ~VM_PROT_WRITE : \
1976 pmap_protect(map->pmap, current->start,
1978 current->protection & MASK(current));
1981 vm_map_simplify_entry(map, current);
1982 current = current->next;
1985 return (KERN_SUCCESS);
1991 * This routine traverses a processes map handling the madvise
1992 * system call. Advisories are classified as either those effecting
1993 * the vm_map_entry structure, or those effecting the underlying
2003 vm_map_entry_t current, entry;
2007 * Some madvise calls directly modify the vm_map_entry, in which case
2008 * we need to use an exclusive lock on the map and we need to perform
2009 * various clipping operations. Otherwise we only need a read-lock
2014 case MADV_SEQUENTIAL:
2026 vm_map_lock_read(map);
2029 return (KERN_INVALID_ARGUMENT);
2033 * Locate starting entry and clip if necessary.
2035 VM_MAP_RANGE_CHECK(map, start, end);
2037 if (vm_map_lookup_entry(map, start, &entry)) {
2039 vm_map_clip_start(map, entry, start);
2041 entry = entry->next;
2046 * madvise behaviors that are implemented in the vm_map_entry.
2048 * We clip the vm_map_entry so that behavioral changes are
2049 * limited to the specified address range.
2051 for (current = entry;
2052 (current != &map->header) && (current->start < end);
2053 current = current->next
2055 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
2058 vm_map_clip_end(map, current, end);
2062 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_NORMAL);
2064 case MADV_SEQUENTIAL:
2065 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_SEQUENTIAL);
2068 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_RANDOM);
2071 current->eflags |= MAP_ENTRY_NOSYNC;
2074 current->eflags &= ~MAP_ENTRY_NOSYNC;
2077 current->eflags |= MAP_ENTRY_NOCOREDUMP;
2080 current->eflags &= ~MAP_ENTRY_NOCOREDUMP;
2085 vm_map_simplify_entry(map, current);
2093 * madvise behaviors that are implemented in the underlying
2096 * Since we don't clip the vm_map_entry, we have to clip
2097 * the vm_object pindex and count.
2099 for (current = entry;
2100 (current != &map->header) && (current->start < end);
2101 current = current->next
2103 vm_offset_t useStart;
2105 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
2108 pindex = OFF_TO_IDX(current->offset);
2109 count = atop(current->end - current->start);
2110 useStart = current->start;
2112 if (current->start < start) {
2113 pindex += atop(start - current->start);
2114 count -= atop(start - current->start);
2117 if (current->end > end)
2118 count -= atop(current->end - end);
2123 vm_object_madvise(current->object.vm_object,
2124 pindex, count, behav);
2125 if (behav == MADV_WILLNEED) {
2126 vm_map_pmap_enter(map,
2128 current->protection,
2129 current->object.vm_object,
2131 (count << PAGE_SHIFT),
2132 MAP_PREFAULT_MADVISE
2136 vm_map_unlock_read(map);
2145 * Sets the inheritance of the specified address
2146 * range in the target map. Inheritance
2147 * affects how the map will be shared with
2148 * child maps at the time of vmspace_fork.
2151 vm_map_inherit(vm_map_t map, vm_offset_t start, vm_offset_t end,
2152 vm_inherit_t new_inheritance)
2154 vm_map_entry_t entry;
2155 vm_map_entry_t temp_entry;
2157 switch (new_inheritance) {
2158 case VM_INHERIT_NONE:
2159 case VM_INHERIT_COPY:
2160 case VM_INHERIT_SHARE:
2163 return (KERN_INVALID_ARGUMENT);
2166 VM_MAP_RANGE_CHECK(map, start, end);
2167 if (vm_map_lookup_entry(map, start, &temp_entry)) {
2169 vm_map_clip_start(map, entry, start);
2171 entry = temp_entry->next;
2172 while ((entry != &map->header) && (entry->start < end)) {
2173 vm_map_clip_end(map, entry, end);
2174 entry->inheritance = new_inheritance;
2175 vm_map_simplify_entry(map, entry);
2176 entry = entry->next;
2179 return (KERN_SUCCESS);
2185 * Implements both kernel and user unwiring.
2188 vm_map_unwire(vm_map_t map, vm_offset_t start, vm_offset_t end,
2191 vm_map_entry_t entry, first_entry, tmp_entry;
2192 vm_offset_t saved_start;
2193 unsigned int last_timestamp;
2195 boolean_t need_wakeup, result, user_unwire;
2197 user_unwire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE;
2199 VM_MAP_RANGE_CHECK(map, start, end);
2200 if (!vm_map_lookup_entry(map, start, &first_entry)) {
2201 if (flags & VM_MAP_WIRE_HOLESOK)
2202 first_entry = first_entry->next;
2205 return (KERN_INVALID_ADDRESS);
2208 last_timestamp = map->timestamp;
2209 entry = first_entry;
2210 while (entry != &map->header && entry->start < end) {
2211 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
2213 * We have not yet clipped the entry.
2215 saved_start = (start >= entry->start) ? start :
2217 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2218 if (vm_map_unlock_and_wait(map, 0)) {
2220 * Allow interruption of user unwiring?
2224 if (last_timestamp+1 != map->timestamp) {
2226 * Look again for the entry because the map was
2227 * modified while it was unlocked.
2228 * Specifically, the entry may have been
2229 * clipped, merged, or deleted.
2231 if (!vm_map_lookup_entry(map, saved_start,
2233 if (flags & VM_MAP_WIRE_HOLESOK)
2234 tmp_entry = tmp_entry->next;
2236 if (saved_start == start) {
2238 * First_entry has been deleted.
2241 return (KERN_INVALID_ADDRESS);
2244 rv = KERN_INVALID_ADDRESS;
2248 if (entry == first_entry)
2249 first_entry = tmp_entry;
2254 last_timestamp = map->timestamp;
2257 vm_map_clip_start(map, entry, start);
2258 vm_map_clip_end(map, entry, end);
2260 * Mark the entry in case the map lock is released. (See
2263 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
2265 * Check the map for holes in the specified region.
2266 * If VM_MAP_WIRE_HOLESOK was specified, skip this check.
2268 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) &&
2269 (entry->end < end && (entry->next == &map->header ||
2270 entry->next->start > entry->end))) {
2272 rv = KERN_INVALID_ADDRESS;
2276 * If system unwiring, require that the entry is system wired.
2279 vm_map_entry_system_wired_count(entry) == 0) {
2281 rv = KERN_INVALID_ARGUMENT;
2284 entry = entry->next;
2288 need_wakeup = FALSE;
2289 if (first_entry == NULL) {
2290 result = vm_map_lookup_entry(map, start, &first_entry);
2291 if (!result && (flags & VM_MAP_WIRE_HOLESOK))
2292 first_entry = first_entry->next;
2294 KASSERT(result, ("vm_map_unwire: lookup failed"));
2296 entry = first_entry;
2297 while (entry != &map->header && entry->start < end) {
2298 if (rv == KERN_SUCCESS && (!user_unwire ||
2299 (entry->eflags & MAP_ENTRY_USER_WIRED))) {
2301 entry->eflags &= ~MAP_ENTRY_USER_WIRED;
2302 entry->wired_count--;
2303 if (entry->wired_count == 0) {
2305 * Retain the map lock.
2307 vm_fault_unwire(map, entry->start, entry->end,
2308 entry->object.vm_object != NULL &&
2309 (entry->object.vm_object->type == OBJT_DEVICE ||
2310 entry->object.vm_object->type == OBJT_SG));
2313 KASSERT(entry->eflags & MAP_ENTRY_IN_TRANSITION,
2314 ("vm_map_unwire: in-transition flag missing"));
2315 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION;
2316 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
2317 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
2320 vm_map_simplify_entry(map, entry);
2321 entry = entry->next;
2332 * Implements both kernel and user wiring.
2335 vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t end,
2338 vm_map_entry_t entry, first_entry, tmp_entry;
2339 vm_offset_t saved_end, saved_start;
2340 unsigned int last_timestamp;
2342 boolean_t fictitious, need_wakeup, result, user_wire;
2344 user_wire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE;
2346 VM_MAP_RANGE_CHECK(map, start, end);
2347 if (!vm_map_lookup_entry(map, start, &first_entry)) {
2348 if (flags & VM_MAP_WIRE_HOLESOK)
2349 first_entry = first_entry->next;
2352 return (KERN_INVALID_ADDRESS);
2355 last_timestamp = map->timestamp;
2356 entry = first_entry;
2357 while (entry != &map->header && entry->start < end) {
2358 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
2360 * We have not yet clipped the entry.
2362 saved_start = (start >= entry->start) ? start :
2364 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2365 if (vm_map_unlock_and_wait(map, 0)) {
2367 * Allow interruption of user wiring?
2371 if (last_timestamp + 1 != map->timestamp) {
2373 * Look again for the entry because the map was
2374 * modified while it was unlocked.
2375 * Specifically, the entry may have been
2376 * clipped, merged, or deleted.
2378 if (!vm_map_lookup_entry(map, saved_start,
2380 if (flags & VM_MAP_WIRE_HOLESOK)
2381 tmp_entry = tmp_entry->next;
2383 if (saved_start == start) {
2385 * first_entry has been deleted.
2388 return (KERN_INVALID_ADDRESS);
2391 rv = KERN_INVALID_ADDRESS;
2395 if (entry == first_entry)
2396 first_entry = tmp_entry;
2401 last_timestamp = map->timestamp;
2404 vm_map_clip_start(map, entry, start);
2405 vm_map_clip_end(map, entry, end);
2407 * Mark the entry in case the map lock is released. (See
2410 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
2414 if (entry->wired_count == 0) {
2415 if ((entry->protection & (VM_PROT_READ|VM_PROT_EXECUTE))
2417 entry->eflags |= MAP_ENTRY_WIRE_SKIPPED;
2418 if ((flags & VM_MAP_WIRE_HOLESOK) == 0) {
2420 rv = KERN_INVALID_ADDRESS;
2425 entry->wired_count++;
2426 saved_start = entry->start;
2427 saved_end = entry->end;
2428 fictitious = entry->object.vm_object != NULL &&
2429 (entry->object.vm_object->type == OBJT_DEVICE ||
2430 entry->object.vm_object->type == OBJT_SG);
2432 * Release the map lock, relying on the in-transition
2433 * mark. Mark the map busy for fork.
2437 rv = vm_fault_wire(map, saved_start, saved_end,
2438 user_wire, fictitious);
2441 if (last_timestamp + 1 != map->timestamp) {
2443 * Look again for the entry because the map was
2444 * modified while it was unlocked. The entry
2445 * may have been clipped, but NOT merged or
2448 result = vm_map_lookup_entry(map, saved_start,
2450 KASSERT(result, ("vm_map_wire: lookup failed"));
2451 if (entry == first_entry)
2452 first_entry = tmp_entry;
2456 while (entry->end < saved_end) {
2457 if (rv != KERN_SUCCESS) {
2458 KASSERT(entry->wired_count == 1,
2459 ("vm_map_wire: bad count"));
2460 entry->wired_count = -1;
2462 entry = entry->next;
2465 last_timestamp = map->timestamp;
2466 if (rv != KERN_SUCCESS) {
2467 KASSERT(entry->wired_count == 1,
2468 ("vm_map_wire: bad count"));
2470 * Assign an out-of-range value to represent
2471 * the failure to wire this entry.
2473 entry->wired_count = -1;
2477 } else if (!user_wire ||
2478 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
2479 entry->wired_count++;
2482 * Check the map for holes in the specified region.
2483 * If VM_MAP_WIRE_HOLESOK was specified, skip this check.
2486 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) &&
2487 (entry->end < end && (entry->next == &map->header ||
2488 entry->next->start > entry->end))) {
2490 rv = KERN_INVALID_ADDRESS;
2493 entry = entry->next;
2497 need_wakeup = FALSE;
2498 if (first_entry == NULL) {
2499 result = vm_map_lookup_entry(map, start, &first_entry);
2500 if (!result && (flags & VM_MAP_WIRE_HOLESOK))
2501 first_entry = first_entry->next;
2503 KASSERT(result, ("vm_map_wire: lookup failed"));
2505 entry = first_entry;
2506 while (entry != &map->header && entry->start < end) {
2507 if ((entry->eflags & MAP_ENTRY_WIRE_SKIPPED) != 0)
2508 goto next_entry_done;
2509 if (rv == KERN_SUCCESS) {
2511 entry->eflags |= MAP_ENTRY_USER_WIRED;
2512 } else if (entry->wired_count == -1) {
2514 * Wiring failed on this entry. Thus, unwiring is
2517 entry->wired_count = 0;
2520 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0)
2521 entry->wired_count--;
2522 if (entry->wired_count == 0) {
2524 * Retain the map lock.
2526 vm_fault_unwire(map, entry->start, entry->end,
2527 entry->object.vm_object != NULL &&
2528 (entry->object.vm_object->type == OBJT_DEVICE ||
2529 entry->object.vm_object->type == OBJT_SG));
2533 KASSERT(entry->eflags & MAP_ENTRY_IN_TRANSITION,
2534 ("vm_map_wire: in-transition flag missing"));
2535 entry->eflags &= ~(MAP_ENTRY_IN_TRANSITION|MAP_ENTRY_WIRE_SKIPPED);
2536 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
2537 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
2540 vm_map_simplify_entry(map, entry);
2541 entry = entry->next;
2552 * Push any dirty cached pages in the address range to their pager.
2553 * If syncio is TRUE, dirty pages are written synchronously.
2554 * If invalidate is TRUE, any cached pages are freed as well.
2556 * If the size of the region from start to end is zero, we are
2557 * supposed to flush all modified pages within the region containing
2558 * start. Unfortunately, a region can be split or coalesced with
2559 * neighboring regions, making it difficult to determine what the
2560 * original region was. Therefore, we approximate this requirement by
2561 * flushing the current region containing start.
2563 * Returns an error if any part of the specified range is not mapped.
2571 boolean_t invalidate)
2573 vm_map_entry_t current;
2574 vm_map_entry_t entry;
2577 vm_ooffset_t offset;
2578 unsigned int last_timestamp;
2580 vm_map_lock_read(map);
2581 VM_MAP_RANGE_CHECK(map, start, end);
2582 if (!vm_map_lookup_entry(map, start, &entry)) {
2583 vm_map_unlock_read(map);
2584 return (KERN_INVALID_ADDRESS);
2585 } else if (start == end) {
2586 start = entry->start;
2590 * Make a first pass to check for user-wired memory and holes.
2592 for (current = entry; current != &map->header && current->start < end;
2593 current = current->next) {
2594 if (invalidate && (current->eflags & MAP_ENTRY_USER_WIRED)) {
2595 vm_map_unlock_read(map);
2596 return (KERN_INVALID_ARGUMENT);
2598 if (end > current->end &&
2599 (current->next == &map->header ||
2600 current->end != current->next->start)) {
2601 vm_map_unlock_read(map);
2602 return (KERN_INVALID_ADDRESS);
2607 pmap_remove(map->pmap, start, end);
2610 * Make a second pass, cleaning/uncaching pages from the indicated
2613 for (current = entry; current != &map->header && current->start < end;) {
2614 offset = current->offset + (start - current->start);
2615 size = (end <= current->end ? end : current->end) - start;
2616 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
2618 vm_map_entry_t tentry;
2621 smap = current->object.sub_map;
2622 vm_map_lock_read(smap);
2623 (void) vm_map_lookup_entry(smap, offset, &tentry);
2624 tsize = tentry->end - offset;
2627 object = tentry->object.vm_object;
2628 offset = tentry->offset + (offset - tentry->start);
2629 vm_map_unlock_read(smap);
2631 object = current->object.vm_object;
2633 vm_object_reference(object);
2634 last_timestamp = map->timestamp;
2635 vm_map_unlock_read(map);
2636 vm_object_sync(object, offset, size, syncio, invalidate);
2638 vm_object_deallocate(object);
2639 vm_map_lock_read(map);
2640 if (last_timestamp == map->timestamp ||
2641 !vm_map_lookup_entry(map, start, ¤t))
2642 current = current->next;
2645 vm_map_unlock_read(map);
2646 return (KERN_SUCCESS);
2650 * vm_map_entry_unwire: [ internal use only ]
2652 * Make the region specified by this entry pageable.
2654 * The map in question should be locked.
2655 * [This is the reason for this routine's existence.]
2658 vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry)
2660 vm_fault_unwire(map, entry->start, entry->end,
2661 entry->object.vm_object != NULL &&
2662 (entry->object.vm_object->type == OBJT_DEVICE ||
2663 entry->object.vm_object->type == OBJT_SG));
2664 entry->wired_count = 0;
2668 vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map)
2671 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0)
2672 vm_object_deallocate(entry->object.vm_object);
2673 uma_zfree(system_map ? kmapentzone : mapentzone, entry);
2677 * vm_map_entry_delete: [ internal use only ]
2679 * Deallocate the given entry from the target map.
2682 vm_map_entry_delete(vm_map_t map, vm_map_entry_t entry)
2685 vm_pindex_t offidxstart, offidxend, count, size1;
2688 vm_map_entry_unlink(map, entry);
2689 object = entry->object.vm_object;
2690 size = entry->end - entry->start;
2693 if (entry->uip != NULL) {
2694 swap_release_by_uid(size, entry->uip);
2698 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0 &&
2700 KASSERT(entry->uip == NULL || object->uip == NULL ||
2701 (entry->eflags & MAP_ENTRY_NEEDS_COPY),
2702 ("OVERCOMMIT vm_map_entry_delete: both uip %p", entry));
2703 count = OFF_TO_IDX(size);
2704 offidxstart = OFF_TO_IDX(entry->offset);
2705 offidxend = offidxstart + count;
2706 VM_OBJECT_LOCK(object);
2707 if (object->ref_count != 1 &&
2708 ((object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING ||
2709 object == kernel_object || object == kmem_object)) {
2710 vm_object_collapse(object);
2711 vm_object_page_remove(object, offidxstart, offidxend, FALSE);
2712 if (object->type == OBJT_SWAP)
2713 swap_pager_freespace(object, offidxstart, count);
2714 if (offidxend >= object->size &&
2715 offidxstart < object->size) {
2716 size1 = object->size;
2717 object->size = offidxstart;
2718 if (object->uip != NULL) {
2719 size1 -= object->size;
2720 KASSERT(object->charge >= ptoa(size1),
2721 ("vm_map_entry_delete: object->charge < 0"));
2722 swap_release_by_uid(ptoa(size1), object->uip);
2723 object->charge -= ptoa(size1);
2727 VM_OBJECT_UNLOCK(object);
2729 entry->object.vm_object = NULL;
2730 if (map->system_map)
2731 vm_map_entry_deallocate(entry, TRUE);
2733 entry->next = curthread->td_map_def_user;
2734 curthread->td_map_def_user = entry;
2739 * vm_map_delete: [ internal use only ]
2741 * Deallocates the given address range from the target
2745 vm_map_delete(vm_map_t map, vm_offset_t start, vm_offset_t end)
2747 vm_map_entry_t entry;
2748 vm_map_entry_t first_entry;
2750 VM_MAP_ASSERT_LOCKED(map);
2753 * Find the start of the region, and clip it
2755 if (!vm_map_lookup_entry(map, start, &first_entry))
2756 entry = first_entry->next;
2758 entry = first_entry;
2759 vm_map_clip_start(map, entry, start);
2763 * Step through all entries in this region
2765 while ((entry != &map->header) && (entry->start < end)) {
2766 vm_map_entry_t next;
2769 * Wait for wiring or unwiring of an entry to complete.
2770 * Also wait for any system wirings to disappear on
2773 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 ||
2774 (vm_map_pmap(map) != kernel_pmap &&
2775 vm_map_entry_system_wired_count(entry) != 0)) {
2776 unsigned int last_timestamp;
2777 vm_offset_t saved_start;
2778 vm_map_entry_t tmp_entry;
2780 saved_start = entry->start;
2781 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2782 last_timestamp = map->timestamp;
2783 (void) vm_map_unlock_and_wait(map, 0);
2785 if (last_timestamp + 1 != map->timestamp) {
2787 * Look again for the entry because the map was
2788 * modified while it was unlocked.
2789 * Specifically, the entry may have been
2790 * clipped, merged, or deleted.
2792 if (!vm_map_lookup_entry(map, saved_start,
2794 entry = tmp_entry->next;
2797 vm_map_clip_start(map, entry,
2803 vm_map_clip_end(map, entry, end);
2808 * Unwire before removing addresses from the pmap; otherwise,
2809 * unwiring will put the entries back in the pmap.
2811 if (entry->wired_count != 0) {
2812 vm_map_entry_unwire(map, entry);
2815 pmap_remove(map->pmap, entry->start, entry->end);
2818 * Delete the entry only after removing all pmap
2819 * entries pointing to its pages. (Otherwise, its
2820 * page frames may be reallocated, and any modify bits
2821 * will be set in the wrong object!)
2823 vm_map_entry_delete(map, entry);
2826 return (KERN_SUCCESS);
2832 * Remove the given address range from the target map.
2833 * This is the exported form of vm_map_delete.
2836 vm_map_remove(vm_map_t map, vm_offset_t start, vm_offset_t end)
2841 VM_MAP_RANGE_CHECK(map, start, end);
2842 result = vm_map_delete(map, start, end);
2848 * vm_map_check_protection:
2850 * Assert that the target map allows the specified privilege on the
2851 * entire address region given. The entire region must be allocated.
2853 * WARNING! This code does not and should not check whether the
2854 * contents of the region is accessible. For example a smaller file
2855 * might be mapped into a larger address space.
2857 * NOTE! This code is also called by munmap().
2859 * The map must be locked. A read lock is sufficient.
2862 vm_map_check_protection(vm_map_t map, vm_offset_t start, vm_offset_t end,
2863 vm_prot_t protection)
2865 vm_map_entry_t entry;
2866 vm_map_entry_t tmp_entry;
2868 if (!vm_map_lookup_entry(map, start, &tmp_entry))
2872 while (start < end) {
2873 if (entry == &map->header)
2878 if (start < entry->start)
2881 * Check protection associated with entry.
2883 if ((entry->protection & protection) != protection)
2885 /* go to next entry */
2887 entry = entry->next;
2893 * vm_map_copy_entry:
2895 * Copies the contents of the source entry to the destination
2896 * entry. The entries *must* be aligned properly.
2902 vm_map_entry_t src_entry,
2903 vm_map_entry_t dst_entry,
2904 vm_ooffset_t *fork_charge)
2906 vm_object_t src_object;
2908 struct uidinfo *uip;
2911 VM_MAP_ASSERT_LOCKED(dst_map);
2913 if ((dst_entry->eflags|src_entry->eflags) & MAP_ENTRY_IS_SUB_MAP)
2916 if (src_entry->wired_count == 0) {
2919 * If the source entry is marked needs_copy, it is already
2922 if ((src_entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) {
2923 pmap_protect(src_map->pmap,
2926 src_entry->protection & ~VM_PROT_WRITE);
2930 * Make a copy of the object.
2932 size = src_entry->end - src_entry->start;
2933 if ((src_object = src_entry->object.vm_object) != NULL) {
2934 VM_OBJECT_LOCK(src_object);
2935 charged = ENTRY_CHARGED(src_entry);
2936 if ((src_object->handle == NULL) &&
2937 (src_object->type == OBJT_DEFAULT ||
2938 src_object->type == OBJT_SWAP)) {
2939 vm_object_collapse(src_object);
2940 if ((src_object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING) {
2941 vm_object_split(src_entry);
2942 src_object = src_entry->object.vm_object;
2945 vm_object_reference_locked(src_object);
2946 vm_object_clear_flag(src_object, OBJ_ONEMAPPING);
2947 if (src_entry->uip != NULL &&
2948 !(src_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
2949 KASSERT(src_object->uip == NULL,
2950 ("OVERCOMMIT: vm_map_copy_entry: uip %p",
2952 src_object->uip = src_entry->uip;
2953 src_object->charge = size;
2955 VM_OBJECT_UNLOCK(src_object);
2956 dst_entry->object.vm_object = src_object;
2958 uip = curthread->td_ucred->cr_ruidinfo;
2960 dst_entry->uip = uip;
2961 *fork_charge += size;
2962 if (!(src_entry->eflags &
2963 MAP_ENTRY_NEEDS_COPY)) {
2965 src_entry->uip = uip;
2966 *fork_charge += size;
2969 src_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY);
2970 dst_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY);
2971 dst_entry->offset = src_entry->offset;
2973 dst_entry->object.vm_object = NULL;
2974 dst_entry->offset = 0;
2975 if (src_entry->uip != NULL) {
2976 dst_entry->uip = curthread->td_ucred->cr_ruidinfo;
2977 uihold(dst_entry->uip);
2978 *fork_charge += size;
2982 pmap_copy(dst_map->pmap, src_map->pmap, dst_entry->start,
2983 dst_entry->end - dst_entry->start, src_entry->start);
2986 * Of course, wired down pages can't be set copy-on-write.
2987 * Cause wired pages to be copied into the new map by
2988 * simulating faults (the new pages are pageable)
2990 vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry,
2996 * vmspace_map_entry_forked:
2997 * Update the newly-forked vmspace each time a map entry is inherited
2998 * or copied. The values for vm_dsize and vm_tsize are approximate
2999 * (and mostly-obsolete ideas in the face of mmap(2) et al.)
3002 vmspace_map_entry_forked(const struct vmspace *vm1, struct vmspace *vm2,
3003 vm_map_entry_t entry)
3005 vm_size_t entrysize;
3008 entrysize = entry->end - entry->start;
3009 vm2->vm_map.size += entrysize;
3010 if (entry->eflags & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP)) {
3011 vm2->vm_ssize += btoc(entrysize);
3012 } else if (entry->start >= (vm_offset_t)vm1->vm_daddr &&
3013 entry->start < (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)) {
3014 newend = MIN(entry->end,
3015 (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize));
3016 vm2->vm_dsize += btoc(newend - entry->start);
3017 } else if (entry->start >= (vm_offset_t)vm1->vm_taddr &&
3018 entry->start < (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)) {
3019 newend = MIN(entry->end,
3020 (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize));
3021 vm2->vm_tsize += btoc(newend - entry->start);
3027 * Create a new process vmspace structure and vm_map
3028 * based on those of an existing process. The new map
3029 * is based on the old map, according to the inheritance
3030 * values on the regions in that map.
3032 * XXX It might be worth coalescing the entries added to the new vmspace.
3034 * The source map must not be locked.
3037 vmspace_fork(struct vmspace *vm1, vm_ooffset_t *fork_charge)
3039 struct vmspace *vm2;
3040 vm_map_t old_map = &vm1->vm_map;
3042 vm_map_entry_t old_entry;
3043 vm_map_entry_t new_entry;
3047 vm_map_lock(old_map);
3049 vm_map_wait_busy(old_map);
3050 vm2 = vmspace_alloc(old_map->min_offset, old_map->max_offset);
3052 goto unlock_and_return;
3053 vm2->vm_taddr = vm1->vm_taddr;
3054 vm2->vm_daddr = vm1->vm_daddr;
3055 vm2->vm_maxsaddr = vm1->vm_maxsaddr;
3056 new_map = &vm2->vm_map; /* XXX */
3057 locked = vm_map_trylock(new_map); /* trylock to silence WITNESS */
3058 KASSERT(locked, ("vmspace_fork: lock failed"));
3059 new_map->timestamp = 1;
3061 old_entry = old_map->header.next;
3063 while (old_entry != &old_map->header) {
3064 if (old_entry->eflags & MAP_ENTRY_IS_SUB_MAP)
3065 panic("vm_map_fork: encountered a submap");
3067 switch (old_entry->inheritance) {
3068 case VM_INHERIT_NONE:
3071 case VM_INHERIT_SHARE:
3073 * Clone the entry, creating the shared object if necessary.
3075 object = old_entry->object.vm_object;
3076 if (object == NULL) {
3077 object = vm_object_allocate(OBJT_DEFAULT,
3078 atop(old_entry->end - old_entry->start));
3079 old_entry->object.vm_object = object;
3080 old_entry->offset = 0;
3081 if (old_entry->uip != NULL) {
3082 object->uip = old_entry->uip;
3083 object->charge = old_entry->end -
3085 old_entry->uip = NULL;
3090 * Add the reference before calling vm_object_shadow
3091 * to insure that a shadow object is created.
3093 vm_object_reference(object);
3094 if (old_entry->eflags & MAP_ENTRY_NEEDS_COPY) {
3095 vm_object_shadow(&old_entry->object.vm_object,
3097 atop(old_entry->end - old_entry->start));
3098 old_entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
3099 /* Transfer the second reference too. */
3100 vm_object_reference(
3101 old_entry->object.vm_object);
3104 * As in vm_map_simplify_entry(), the
3105 * vnode lock will not be acquired in
3106 * this call to vm_object_deallocate().
3108 vm_object_deallocate(object);
3109 object = old_entry->object.vm_object;
3111 VM_OBJECT_LOCK(object);
3112 vm_object_clear_flag(object, OBJ_ONEMAPPING);
3113 if (old_entry->uip != NULL) {
3114 KASSERT(object->uip == NULL, ("vmspace_fork both uip"));
3115 object->uip = old_entry->uip;
3116 object->charge = old_entry->end - old_entry->start;
3117 old_entry->uip = NULL;
3119 VM_OBJECT_UNLOCK(object);
3122 * Clone the entry, referencing the shared object.
3124 new_entry = vm_map_entry_create(new_map);
3125 *new_entry = *old_entry;
3126 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED |
3127 MAP_ENTRY_IN_TRANSITION);
3128 new_entry->wired_count = 0;
3131 * Insert the entry into the new map -- we know we're
3132 * inserting at the end of the new map.
3134 vm_map_entry_link(new_map, new_map->header.prev,
3136 vmspace_map_entry_forked(vm1, vm2, new_entry);
3139 * Update the physical map
3141 pmap_copy(new_map->pmap, old_map->pmap,
3143 (old_entry->end - old_entry->start),
3147 case VM_INHERIT_COPY:
3149 * Clone the entry and link into the map.
3151 new_entry = vm_map_entry_create(new_map);
3152 *new_entry = *old_entry;
3153 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED |
3154 MAP_ENTRY_IN_TRANSITION);
3155 new_entry->wired_count = 0;
3156 new_entry->object.vm_object = NULL;
3157 new_entry->uip = NULL;
3158 vm_map_entry_link(new_map, new_map->header.prev,
3160 vmspace_map_entry_forked(vm1, vm2, new_entry);
3161 vm_map_copy_entry(old_map, new_map, old_entry,
3162 new_entry, fork_charge);
3165 old_entry = old_entry->next;
3168 vm_map_unlock(old_map);
3170 vm_map_unlock(new_map);
3176 vm_map_stack(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
3177 vm_prot_t prot, vm_prot_t max, int cow)
3179 vm_map_entry_t new_entry, prev_entry;
3180 vm_offset_t bot, top;
3181 vm_size_t init_ssize;
3186 * The stack orientation is piggybacked with the cow argument.
3187 * Extract it into orient and mask the cow argument so that we
3188 * don't pass it around further.
3189 * NOTE: We explicitly allow bi-directional stacks.
3191 orient = cow & (MAP_STACK_GROWS_DOWN|MAP_STACK_GROWS_UP);
3193 KASSERT(orient != 0, ("No stack grow direction"));
3195 if (addrbos < vm_map_min(map) ||
3196 addrbos > vm_map_max(map) ||
3197 addrbos + max_ssize < addrbos)
3198 return (KERN_NO_SPACE);
3200 init_ssize = (max_ssize < sgrowsiz) ? max_ssize : sgrowsiz;
3202 PROC_LOCK(curthread->td_proc);
3203 vmemlim = lim_cur(curthread->td_proc, RLIMIT_VMEM);
3204 PROC_UNLOCK(curthread->td_proc);
3208 /* If addr is already mapped, no go */
3209 if (vm_map_lookup_entry(map, addrbos, &prev_entry)) {
3211 return (KERN_NO_SPACE);
3214 /* If we would blow our VMEM resource limit, no go */
3215 if (map->size + init_ssize > vmemlim) {
3217 return (KERN_NO_SPACE);
3221 * If we can't accomodate max_ssize in the current mapping, no go.
3222 * However, we need to be aware that subsequent user mappings might
3223 * map into the space we have reserved for stack, and currently this
3224 * space is not protected.
3226 * Hopefully we will at least detect this condition when we try to
3229 if ((prev_entry->next != &map->header) &&
3230 (prev_entry->next->start < addrbos + max_ssize)) {
3232 return (KERN_NO_SPACE);
3236 * We initially map a stack of only init_ssize. We will grow as
3237 * needed later. Depending on the orientation of the stack (i.e.
3238 * the grow direction) we either map at the top of the range, the
3239 * bottom of the range or in the middle.
3241 * Note: we would normally expect prot and max to be VM_PROT_ALL,
3242 * and cow to be 0. Possibly we should eliminate these as input
3243 * parameters, and just pass these values here in the insert call.
3245 if (orient == MAP_STACK_GROWS_DOWN)
3246 bot = addrbos + max_ssize - init_ssize;
3247 else if (orient == MAP_STACK_GROWS_UP)
3250 bot = round_page(addrbos + max_ssize/2 - init_ssize/2);
3251 top = bot + init_ssize;
3252 rv = vm_map_insert(map, NULL, 0, bot, top, prot, max, cow);
3254 /* Now set the avail_ssize amount. */
3255 if (rv == KERN_SUCCESS) {
3256 if (prev_entry != &map->header)
3257 vm_map_clip_end(map, prev_entry, bot);
3258 new_entry = prev_entry->next;
3259 if (new_entry->end != top || new_entry->start != bot)
3260 panic("Bad entry start/end for new stack entry");
3262 new_entry->avail_ssize = max_ssize - init_ssize;
3263 if (orient & MAP_STACK_GROWS_DOWN)
3264 new_entry->eflags |= MAP_ENTRY_GROWS_DOWN;
3265 if (orient & MAP_STACK_GROWS_UP)
3266 new_entry->eflags |= MAP_ENTRY_GROWS_UP;
3273 static int stack_guard_page = 0;
3274 TUNABLE_INT("security.bsd.stack_guard_page", &stack_guard_page);
3275 SYSCTL_INT(_security_bsd, OID_AUTO, stack_guard_page, CTLFLAG_RW,
3276 &stack_guard_page, 0,
3277 "Insert stack guard page ahead of the growable segments.");
3279 /* Attempts to grow a vm stack entry. Returns KERN_SUCCESS if the
3280 * desired address is already mapped, or if we successfully grow
3281 * the stack. Also returns KERN_SUCCESS if addr is outside the
3282 * stack range (this is strange, but preserves compatibility with
3283 * the grow function in vm_machdep.c).
3286 vm_map_growstack(struct proc *p, vm_offset_t addr)
3288 vm_map_entry_t next_entry, prev_entry;
3289 vm_map_entry_t new_entry, stack_entry;
3290 struct vmspace *vm = p->p_vmspace;
3291 vm_map_t map = &vm->vm_map;
3293 size_t grow_amount, max_grow;
3294 rlim_t stacklim, vmemlim;
3295 int is_procstack, rv;
3296 struct uidinfo *uip;
3300 stacklim = lim_cur(p, RLIMIT_STACK);
3301 vmemlim = lim_cur(p, RLIMIT_VMEM);
3304 vm_map_lock_read(map);
3306 /* If addr is already in the entry range, no need to grow.*/
3307 if (vm_map_lookup_entry(map, addr, &prev_entry)) {
3308 vm_map_unlock_read(map);
3309 return (KERN_SUCCESS);
3312 next_entry = prev_entry->next;
3313 if (!(prev_entry->eflags & MAP_ENTRY_GROWS_UP)) {
3315 * This entry does not grow upwards. Since the address lies
3316 * beyond this entry, the next entry (if one exists) has to
3317 * be a downward growable entry. The entry list header is
3318 * never a growable entry, so it suffices to check the flags.
3320 if (!(next_entry->eflags & MAP_ENTRY_GROWS_DOWN)) {
3321 vm_map_unlock_read(map);
3322 return (KERN_SUCCESS);
3324 stack_entry = next_entry;
3327 * This entry grows upward. If the next entry does not at
3328 * least grow downwards, this is the entry we need to grow.
3329 * otherwise we have two possible choices and we have to
3332 if (next_entry->eflags & MAP_ENTRY_GROWS_DOWN) {
3334 * We have two choices; grow the entry closest to
3335 * the address to minimize the amount of growth.
3337 if (addr - prev_entry->end <= next_entry->start - addr)
3338 stack_entry = prev_entry;
3340 stack_entry = next_entry;
3342 stack_entry = prev_entry;
3345 if (stack_entry == next_entry) {
3346 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_DOWN, ("foo"));
3347 KASSERT(addr < stack_entry->start, ("foo"));
3348 end = (prev_entry != &map->header) ? prev_entry->end :
3349 stack_entry->start - stack_entry->avail_ssize;
3350 grow_amount = roundup(stack_entry->start - addr, PAGE_SIZE);
3351 max_grow = stack_entry->start - end;
3353 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_UP, ("foo"));
3354 KASSERT(addr >= stack_entry->end, ("foo"));
3355 end = (next_entry != &map->header) ? next_entry->start :
3356 stack_entry->end + stack_entry->avail_ssize;
3357 grow_amount = roundup(addr + 1 - stack_entry->end, PAGE_SIZE);
3358 max_grow = end - stack_entry->end;
3361 if (grow_amount > stack_entry->avail_ssize) {
3362 vm_map_unlock_read(map);
3363 return (KERN_NO_SPACE);
3367 * If there is no longer enough space between the entries nogo, and
3368 * adjust the available space. Note: this should only happen if the
3369 * user has mapped into the stack area after the stack was created,
3370 * and is probably an error.
3372 * This also effectively destroys any guard page the user might have
3373 * intended by limiting the stack size.
3375 if (grow_amount + (stack_guard_page ? PAGE_SIZE : 0) > max_grow) {
3376 if (vm_map_lock_upgrade(map))
3379 stack_entry->avail_ssize = max_grow;
3382 return (KERN_NO_SPACE);
3385 is_procstack = (addr >= (vm_offset_t)vm->vm_maxsaddr) ? 1 : 0;
3388 * If this is the main process stack, see if we're over the stack
3391 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) {
3392 vm_map_unlock_read(map);
3393 return (KERN_NO_SPACE);
3396 /* Round up the grow amount modulo SGROWSIZ */
3397 grow_amount = roundup (grow_amount, sgrowsiz);
3398 if (grow_amount > stack_entry->avail_ssize)
3399 grow_amount = stack_entry->avail_ssize;
3400 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) {
3401 grow_amount = trunc_page((vm_size_t)stacklim) -
3405 /* If we would blow our VMEM resource limit, no go */
3406 if (map->size + grow_amount > vmemlim) {
3407 vm_map_unlock_read(map);
3408 return (KERN_NO_SPACE);
3411 if (vm_map_lock_upgrade(map))
3414 if (stack_entry == next_entry) {
3418 /* Get the preliminary new entry start value */
3419 addr = stack_entry->start - grow_amount;
3422 * If this puts us into the previous entry, cut back our
3423 * growth to the available space. Also, see the note above.
3426 stack_entry->avail_ssize = max_grow;
3428 if (stack_guard_page)
3432 rv = vm_map_insert(map, NULL, 0, addr, stack_entry->start,
3433 p->p_sysent->sv_stackprot, VM_PROT_ALL, 0);
3435 /* Adjust the available stack space by the amount we grew. */
3436 if (rv == KERN_SUCCESS) {
3437 if (prev_entry != &map->header)
3438 vm_map_clip_end(map, prev_entry, addr);
3439 new_entry = prev_entry->next;
3440 KASSERT(new_entry == stack_entry->prev, ("foo"));
3441 KASSERT(new_entry->end == stack_entry->start, ("foo"));
3442 KASSERT(new_entry->start == addr, ("foo"));
3443 grow_amount = new_entry->end - new_entry->start;
3444 new_entry->avail_ssize = stack_entry->avail_ssize -
3446 stack_entry->eflags &= ~MAP_ENTRY_GROWS_DOWN;
3447 new_entry->eflags |= MAP_ENTRY_GROWS_DOWN;
3453 addr = stack_entry->end + grow_amount;
3456 * If this puts us into the next entry, cut back our growth
3457 * to the available space. Also, see the note above.
3460 stack_entry->avail_ssize = end - stack_entry->end;
3462 if (stack_guard_page)
3466 grow_amount = addr - stack_entry->end;
3467 uip = stack_entry->uip;
3468 if (uip == NULL && stack_entry->object.vm_object != NULL)
3469 uip = stack_entry->object.vm_object->uip;
3470 if (uip != NULL && !swap_reserve_by_uid(grow_amount, uip))
3472 /* Grow the underlying object if applicable. */
3473 else if (stack_entry->object.vm_object == NULL ||
3474 vm_object_coalesce(stack_entry->object.vm_object,
3475 stack_entry->offset,
3476 (vm_size_t)(stack_entry->end - stack_entry->start),
3477 (vm_size_t)grow_amount, uip != NULL)) {
3478 map->size += (addr - stack_entry->end);
3479 /* Update the current entry. */
3480 stack_entry->end = addr;
3481 stack_entry->avail_ssize -= grow_amount;
3482 vm_map_entry_resize_free(map, stack_entry);
3485 if (next_entry != &map->header)
3486 vm_map_clip_start(map, next_entry, addr);
3491 if (rv == KERN_SUCCESS && is_procstack)
3492 vm->vm_ssize += btoc(grow_amount);
3497 * Heed the MAP_WIREFUTURE flag if it was set for this process.
3499 if (rv == KERN_SUCCESS && (map->flags & MAP_WIREFUTURE)) {
3501 (stack_entry == next_entry) ? addr : addr - grow_amount,
3502 (stack_entry == next_entry) ? stack_entry->start : addr,
3503 (p->p_flag & P_SYSTEM)
3504 ? VM_MAP_WIRE_SYSTEM|VM_MAP_WIRE_NOHOLES
3505 : VM_MAP_WIRE_USER|VM_MAP_WIRE_NOHOLES);
3512 * Unshare the specified VM space for exec. If other processes are
3513 * mapped to it, then create a new one. The new vmspace is null.
3516 vmspace_exec(struct proc *p, vm_offset_t minuser, vm_offset_t maxuser)
3518 struct vmspace *oldvmspace = p->p_vmspace;
3519 struct vmspace *newvmspace;
3521 newvmspace = vmspace_alloc(minuser, maxuser);
3522 if (newvmspace == NULL)
3524 newvmspace->vm_swrss = oldvmspace->vm_swrss;
3526 * This code is written like this for prototype purposes. The
3527 * goal is to avoid running down the vmspace here, but let the
3528 * other process's that are still using the vmspace to finally
3529 * run it down. Even though there is little or no chance of blocking
3530 * here, it is a good idea to keep this form for future mods.
3532 PROC_VMSPACE_LOCK(p);
3533 p->p_vmspace = newvmspace;
3534 PROC_VMSPACE_UNLOCK(p);
3535 if (p == curthread->td_proc)
3536 pmap_activate(curthread);
3537 vmspace_free(oldvmspace);
3542 * Unshare the specified VM space for forcing COW. This
3543 * is called by rfork, for the (RFMEM|RFPROC) == 0 case.
3546 vmspace_unshare(struct proc *p)
3548 struct vmspace *oldvmspace = p->p_vmspace;
3549 struct vmspace *newvmspace;
3550 vm_ooffset_t fork_charge;
3552 if (oldvmspace->vm_refcnt == 1)
3555 newvmspace = vmspace_fork(oldvmspace, &fork_charge);
3556 if (newvmspace == NULL)
3558 if (!swap_reserve_by_uid(fork_charge, p->p_ucred->cr_ruidinfo)) {
3559 vmspace_free(newvmspace);
3562 PROC_VMSPACE_LOCK(p);
3563 p->p_vmspace = newvmspace;
3564 PROC_VMSPACE_UNLOCK(p);
3565 if (p == curthread->td_proc)
3566 pmap_activate(curthread);
3567 vmspace_free(oldvmspace);
3574 * Finds the VM object, offset, and
3575 * protection for a given virtual address in the
3576 * specified map, assuming a page fault of the
3579 * Leaves the map in question locked for read; return
3580 * values are guaranteed until a vm_map_lookup_done
3581 * call is performed. Note that the map argument
3582 * is in/out; the returned map must be used in
3583 * the call to vm_map_lookup_done.
3585 * A handle (out_entry) is returned for use in
3586 * vm_map_lookup_done, to make that fast.
3588 * If a lookup is requested with "write protection"
3589 * specified, the map may be changed to perform virtual
3590 * copying operations, although the data referenced will
3594 vm_map_lookup(vm_map_t *var_map, /* IN/OUT */
3596 vm_prot_t fault_typea,
3597 vm_map_entry_t *out_entry, /* OUT */
3598 vm_object_t *object, /* OUT */
3599 vm_pindex_t *pindex, /* OUT */
3600 vm_prot_t *out_prot, /* OUT */
3601 boolean_t *wired) /* OUT */
3603 vm_map_entry_t entry;
3604 vm_map_t map = *var_map;
3606 vm_prot_t fault_type = fault_typea;
3607 vm_object_t eobject;
3608 struct uidinfo *uip;
3613 vm_map_lock_read(map);
3616 * Lookup the faulting address.
3618 if (!vm_map_lookup_entry(map, vaddr, out_entry)) {
3619 vm_map_unlock_read(map);
3620 return (KERN_INVALID_ADDRESS);
3628 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
3629 vm_map_t old_map = map;
3631 *var_map = map = entry->object.sub_map;
3632 vm_map_unlock_read(old_map);
3637 * Check whether this task is allowed to have this page.
3638 * Note the special case for MAP_ENTRY_COW
3639 * pages with an override. This is to implement a forced
3640 * COW for debuggers.
3642 if (fault_type & VM_PROT_OVERRIDE_WRITE)
3643 prot = entry->max_protection;
3645 prot = entry->protection;
3646 fault_type &= (VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE);
3647 if ((fault_type & prot) != fault_type) {
3648 vm_map_unlock_read(map);
3649 return (KERN_PROTECTION_FAILURE);
3651 if ((entry->eflags & MAP_ENTRY_USER_WIRED) &&
3652 (entry->eflags & MAP_ENTRY_COW) &&
3653 (fault_type & VM_PROT_WRITE) &&
3654 (fault_typea & VM_PROT_OVERRIDE_WRITE) == 0) {
3655 vm_map_unlock_read(map);
3656 return (KERN_PROTECTION_FAILURE);
3660 * If this page is not pageable, we have to get it for all possible
3663 *wired = (entry->wired_count != 0);
3665 prot = fault_type = entry->protection;
3666 size = entry->end - entry->start;
3668 * If the entry was copy-on-write, we either ...
3670 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
3672 * If we want to write the page, we may as well handle that
3673 * now since we've got the map locked.
3675 * If we don't need to write the page, we just demote the
3676 * permissions allowed.
3678 if (fault_type & VM_PROT_WRITE) {
3680 * Make a new object, and place it in the object
3681 * chain. Note that no new references have appeared
3682 * -- one just moved from the map to the new
3685 if (vm_map_lock_upgrade(map))
3688 if (entry->uip == NULL) {
3690 * The debugger owner is charged for
3693 uip = curthread->td_ucred->cr_ruidinfo;
3695 if (!swap_reserve_by_uid(size, uip)) {
3698 return (KERN_RESOURCE_SHORTAGE);
3703 &entry->object.vm_object,
3706 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
3707 eobject = entry->object.vm_object;
3708 if (eobject->uip != NULL) {
3710 * The object was not shadowed.
3712 swap_release_by_uid(size, entry->uip);
3715 } else if (entry->uip != NULL) {
3716 VM_OBJECT_LOCK(eobject);
3717 eobject->uip = entry->uip;
3718 eobject->charge = size;
3719 VM_OBJECT_UNLOCK(eobject);
3723 vm_map_lock_downgrade(map);
3726 * We're attempting to read a copy-on-write page --
3727 * don't allow writes.
3729 prot &= ~VM_PROT_WRITE;
3734 * Create an object if necessary.
3736 if (entry->object.vm_object == NULL &&
3738 if (vm_map_lock_upgrade(map))
3740 entry->object.vm_object = vm_object_allocate(OBJT_DEFAULT,
3743 if (entry->uip != NULL) {
3744 VM_OBJECT_LOCK(entry->object.vm_object);
3745 entry->object.vm_object->uip = entry->uip;
3746 entry->object.vm_object->charge = size;
3747 VM_OBJECT_UNLOCK(entry->object.vm_object);
3750 vm_map_lock_downgrade(map);
3754 * Return the object/offset from this entry. If the entry was
3755 * copy-on-write or empty, it has been fixed up.
3757 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
3758 *object = entry->object.vm_object;
3761 return (KERN_SUCCESS);
3765 * vm_map_lookup_locked:
3767 * Lookup the faulting address. A version of vm_map_lookup that returns
3768 * KERN_FAILURE instead of blocking on map lock or memory allocation.
3771 vm_map_lookup_locked(vm_map_t *var_map, /* IN/OUT */
3773 vm_prot_t fault_typea,
3774 vm_map_entry_t *out_entry, /* OUT */
3775 vm_object_t *object, /* OUT */
3776 vm_pindex_t *pindex, /* OUT */
3777 vm_prot_t *out_prot, /* OUT */
3778 boolean_t *wired) /* OUT */
3780 vm_map_entry_t entry;
3781 vm_map_t map = *var_map;
3783 vm_prot_t fault_type = fault_typea;
3786 * Lookup the faulting address.
3788 if (!vm_map_lookup_entry(map, vaddr, out_entry))
3789 return (KERN_INVALID_ADDRESS);
3794 * Fail if the entry refers to a submap.
3796 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP)
3797 return (KERN_FAILURE);
3800 * Check whether this task is allowed to have this page.
3801 * Note the special case for MAP_ENTRY_COW
3802 * pages with an override. This is to implement a forced
3803 * COW for debuggers.
3805 if (fault_type & VM_PROT_OVERRIDE_WRITE)
3806 prot = entry->max_protection;
3808 prot = entry->protection;
3809 fault_type &= VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE;
3810 if ((fault_type & prot) != fault_type)
3811 return (KERN_PROTECTION_FAILURE);
3812 if ((entry->eflags & MAP_ENTRY_USER_WIRED) &&
3813 (entry->eflags & MAP_ENTRY_COW) &&
3814 (fault_type & VM_PROT_WRITE) &&
3815 (fault_typea & VM_PROT_OVERRIDE_WRITE) == 0)
3816 return (KERN_PROTECTION_FAILURE);
3819 * If this page is not pageable, we have to get it for all possible
3822 *wired = (entry->wired_count != 0);
3824 prot = fault_type = entry->protection;
3826 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
3828 * Fail if the entry was copy-on-write for a write fault.
3830 if (fault_type & VM_PROT_WRITE)
3831 return (KERN_FAILURE);
3833 * We're attempting to read a copy-on-write page --
3834 * don't allow writes.
3836 prot &= ~VM_PROT_WRITE;
3840 * Fail if an object should be created.
3842 if (entry->object.vm_object == NULL && !map->system_map)
3843 return (KERN_FAILURE);
3846 * Return the object/offset from this entry. If the entry was
3847 * copy-on-write or empty, it has been fixed up.
3849 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
3850 *object = entry->object.vm_object;
3853 return (KERN_SUCCESS);
3857 * vm_map_lookup_done:
3859 * Releases locks acquired by a vm_map_lookup
3860 * (according to the handle returned by that lookup).
3863 vm_map_lookup_done(vm_map_t map, vm_map_entry_t entry)
3866 * Unlock the main-level map
3868 vm_map_unlock_read(map);
3871 #include "opt_ddb.h"
3873 #include <sys/kernel.h>
3875 #include <ddb/ddb.h>
3878 * vm_map_print: [ debug ]
3880 DB_SHOW_COMMAND(map, vm_map_print)
3883 /* XXX convert args. */
3884 vm_map_t map = (vm_map_t)addr;
3885 boolean_t full = have_addr;
3887 vm_map_entry_t entry;
3889 db_iprintf("Task map %p: pmap=%p, nentries=%d, version=%u\n",
3891 (void *)map->pmap, map->nentries, map->timestamp);
3894 if (!full && db_indent)
3898 for (entry = map->header.next; entry != &map->header;
3899 entry = entry->next) {
3900 db_iprintf("map entry %p: start=%p, end=%p\n",
3901 (void *)entry, (void *)entry->start, (void *)entry->end);
3904 static char *inheritance_name[4] =
3905 {"share", "copy", "none", "donate_copy"};
3907 db_iprintf(" prot=%x/%x/%s",
3909 entry->max_protection,
3910 inheritance_name[(int)(unsigned char)entry->inheritance]);
3911 if (entry->wired_count != 0)
3912 db_printf(", wired");
3914 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
3915 db_printf(", share=%p, offset=0x%jx\n",
3916 (void *)entry->object.sub_map,
3917 (uintmax_t)entry->offset);
3919 if ((entry->prev == &map->header) ||
3920 (entry->prev->object.sub_map !=
3921 entry->object.sub_map)) {
3923 vm_map_print((db_expr_t)(intptr_t)
3924 entry->object.sub_map,
3925 full, 0, (char *)0);
3929 if (entry->uip != NULL)
3930 db_printf(", uip %d", entry->uip->ui_uid);
3931 db_printf(", object=%p, offset=0x%jx",
3932 (void *)entry->object.vm_object,
3933 (uintmax_t)entry->offset);
3934 if (entry->object.vm_object && entry->object.vm_object->uip)
3935 db_printf(", obj uip %d charge %jx",
3936 entry->object.vm_object->uip->ui_uid,
3937 (uintmax_t)entry->object.vm_object->charge);
3938 if (entry->eflags & MAP_ENTRY_COW)
3939 db_printf(", copy (%s)",
3940 (entry->eflags & MAP_ENTRY_NEEDS_COPY) ? "needed" : "done");
3944 if ((entry->prev == &map->header) ||
3945 (entry->prev->object.vm_object !=
3946 entry->object.vm_object)) {
3948 vm_object_print((db_expr_t)(intptr_t)
3949 entry->object.vm_object,
3950 full, 0, (char *)0);
3962 DB_SHOW_COMMAND(procvm, procvm)
3967 p = (struct proc *) addr;
3972 db_printf("p = %p, vmspace = %p, map = %p, pmap = %p\n",
3973 (void *)p, (void *)p->p_vmspace, (void *)&p->p_vmspace->vm_map,
3974 (void *)vmspace_pmap(p->p_vmspace));
3976 vm_map_print((db_expr_t)(intptr_t)&p->p_vmspace->vm_map, 1, 0, NULL);