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)
320 CTR1(KTR_VM, "vmspace_free: %p", vm);
323 * Make sure any SysV shm is freed, it might not have been in
329 * Lock the map, to wait out all other references to it.
330 * Delete all of the mappings and pages they hold, then call
331 * the pmap module to reclaim anything left.
333 (void)vm_map_remove(&vm->vm_map, vm->vm_map.min_offset,
334 vm->vm_map.max_offset);
336 pmap_release(vmspace_pmap(vm));
337 vm->vm_map.pmap = NULL;
338 uma_zfree(vmspace_zone, vm);
342 vmspace_free(struct vmspace *vm)
345 if (vm->vm_refcnt == 0)
346 panic("vmspace_free: attempt to free already freed vmspace");
348 if (atomic_fetchadd_int(&vm->vm_refcnt, -1) == 1)
353 vmspace_exitfree(struct proc *p)
357 PROC_VMSPACE_LOCK(p);
360 PROC_VMSPACE_UNLOCK(p);
361 KASSERT(vm == &vmspace0, ("vmspace_exitfree: wrong vmspace"));
366 vmspace_exit(struct thread *td)
373 * Release user portion of address space.
374 * This releases references to vnodes,
375 * which could cause I/O if the file has been unlinked.
376 * Need to do this early enough that we can still sleep.
378 * The last exiting process to reach this point releases as
379 * much of the environment as it can. vmspace_dofree() is the
380 * slower fallback in case another process had a temporary
381 * reference to the vmspace.
386 atomic_add_int(&vmspace0.vm_refcnt, 1);
388 refcnt = vm->vm_refcnt;
389 if (refcnt > 1 && p->p_vmspace != &vmspace0) {
390 /* Switch now since other proc might free vmspace */
391 PROC_VMSPACE_LOCK(p);
392 p->p_vmspace = &vmspace0;
393 PROC_VMSPACE_UNLOCK(p);
396 } while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt - 1));
398 if (p->p_vmspace != vm) {
399 /* vmspace not yet freed, switch back */
400 PROC_VMSPACE_LOCK(p);
402 PROC_VMSPACE_UNLOCK(p);
405 pmap_remove_pages(vmspace_pmap(vm));
406 /* Switch now since this proc will free vmspace */
407 PROC_VMSPACE_LOCK(p);
408 p->p_vmspace = &vmspace0;
409 PROC_VMSPACE_UNLOCK(p);
415 /* Acquire reference to vmspace owned by another process. */
418 vmspace_acquire_ref(struct proc *p)
423 PROC_VMSPACE_LOCK(p);
426 PROC_VMSPACE_UNLOCK(p);
430 refcnt = vm->vm_refcnt;
431 if (refcnt <= 0) { /* Avoid 0->1 transition */
432 PROC_VMSPACE_UNLOCK(p);
435 } while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt + 1));
436 if (vm != p->p_vmspace) {
437 PROC_VMSPACE_UNLOCK(p);
441 PROC_VMSPACE_UNLOCK(p);
446 _vm_map_lock(vm_map_t map, const char *file, int line)
450 _mtx_lock_flags(&map->system_mtx, 0, file, line);
452 (void)_sx_xlock(&map->lock, 0, file, line);
457 vm_map_process_deferred(void)
460 vm_map_entry_t entry;
464 while ((entry = td->td_map_def_user) != NULL) {
465 td->td_map_def_user = entry->next;
466 vm_map_entry_deallocate(entry, FALSE);
471 _vm_map_unlock(vm_map_t map, const char *file, int line)
475 _mtx_unlock_flags(&map->system_mtx, 0, file, line);
477 _sx_xunlock(&map->lock, file, line);
478 vm_map_process_deferred();
483 _vm_map_lock_read(vm_map_t map, const char *file, int line)
487 _mtx_lock_flags(&map->system_mtx, 0, file, line);
489 (void)_sx_slock(&map->lock, 0, file, line);
493 _vm_map_unlock_read(vm_map_t map, const char *file, int line)
497 _mtx_unlock_flags(&map->system_mtx, 0, file, line);
499 _sx_sunlock(&map->lock, file, line);
500 vm_map_process_deferred();
505 _vm_map_trylock(vm_map_t map, const char *file, int line)
509 error = map->system_map ?
510 !_mtx_trylock(&map->system_mtx, 0, file, line) :
511 !_sx_try_xlock(&map->lock, file, line);
518 _vm_map_trylock_read(vm_map_t map, const char *file, int line)
522 error = map->system_map ?
523 !_mtx_trylock(&map->system_mtx, 0, file, line) :
524 !_sx_try_slock(&map->lock, file, line);
529 * _vm_map_lock_upgrade: [ internal use only ]
531 * Tries to upgrade a read (shared) lock on the specified map to a write
532 * (exclusive) lock. Returns the value "0" if the upgrade succeeds and a
533 * non-zero value if the upgrade fails. If the upgrade fails, the map is
534 * returned without a read or write lock held.
536 * Requires that the map be read locked.
539 _vm_map_lock_upgrade(vm_map_t map, const char *file, int line)
541 unsigned int last_timestamp;
543 if (map->system_map) {
545 _mtx_assert(&map->system_mtx, MA_OWNED, file, line);
548 if (!_sx_try_upgrade(&map->lock, file, line)) {
549 last_timestamp = map->timestamp;
550 _sx_sunlock(&map->lock, file, line);
551 vm_map_process_deferred();
553 * If the map's timestamp does not change while the
554 * map is unlocked, then the upgrade succeeds.
556 (void)_sx_xlock(&map->lock, 0, file, line);
557 if (last_timestamp != map->timestamp) {
558 _sx_xunlock(&map->lock, file, line);
568 _vm_map_lock_downgrade(vm_map_t map, const char *file, int line)
571 if (map->system_map) {
573 _mtx_assert(&map->system_mtx, MA_OWNED, file, line);
576 _sx_downgrade(&map->lock, file, line);
582 * Returns a non-zero value if the caller holds a write (exclusive) lock
583 * on the specified map and the value "0" otherwise.
586 vm_map_locked(vm_map_t map)
590 return (mtx_owned(&map->system_mtx));
592 return (sx_xlocked(&map->lock));
597 _vm_map_assert_locked(vm_map_t map, const char *file, int line)
601 _mtx_assert(&map->system_mtx, MA_OWNED, file, line);
603 _sx_assert(&map->lock, SA_XLOCKED, file, line);
608 _vm_map_assert_locked_read(vm_map_t map, const char *file, int line)
612 _mtx_assert(&map->system_mtx, MA_OWNED, file, line);
614 _sx_assert(&map->lock, SA_SLOCKED, file, line);
618 #define VM_MAP_ASSERT_LOCKED(map) \
619 _vm_map_assert_locked(map, LOCK_FILE, LOCK_LINE)
620 #define VM_MAP_ASSERT_LOCKED_READ(map) \
621 _vm_map_assert_locked_read(map, LOCK_FILE, LOCK_LINE)
623 #define VM_MAP_ASSERT_LOCKED(map)
624 #define VM_MAP_ASSERT_LOCKED_READ(map)
628 * _vm_map_unlock_and_wait:
630 * Atomically releases the lock on the specified map and puts the calling
631 * thread to sleep. The calling thread will remain asleep until either
632 * vm_map_wakeup() is performed on the map or the specified timeout is
635 * WARNING! This function does not perform deferred deallocations of
636 * objects and map entries. Therefore, the calling thread is expected to
637 * reacquire the map lock after reawakening and later perform an ordinary
638 * unlock operation, such as vm_map_unlock(), before completing its
639 * operation on the map.
642 _vm_map_unlock_and_wait(vm_map_t map, int timo, const char *file, int line)
645 mtx_lock(&map_sleep_mtx);
647 _mtx_unlock_flags(&map->system_mtx, 0, file, line);
649 _sx_xunlock(&map->lock, file, line);
650 return (msleep(&map->root, &map_sleep_mtx, PDROP | PVM, "vmmaps",
657 * Awaken any threads that have slept on the map using
658 * vm_map_unlock_and_wait().
661 vm_map_wakeup(vm_map_t map)
665 * Acquire and release map_sleep_mtx to prevent a wakeup()
666 * from being performed (and lost) between the map unlock
667 * and the msleep() in _vm_map_unlock_and_wait().
669 mtx_lock(&map_sleep_mtx);
670 mtx_unlock(&map_sleep_mtx);
675 vm_map_busy(vm_map_t map)
678 VM_MAP_ASSERT_LOCKED(map);
683 vm_map_unbusy(vm_map_t map)
686 VM_MAP_ASSERT_LOCKED(map);
687 KASSERT(map->busy, ("vm_map_unbusy: not busy"));
688 if (--map->busy == 0 && (map->flags & MAP_BUSY_WAKEUP)) {
689 vm_map_modflags(map, 0, MAP_BUSY_WAKEUP);
695 vm_map_wait_busy(vm_map_t map)
698 VM_MAP_ASSERT_LOCKED(map);
700 vm_map_modflags(map, MAP_BUSY_WAKEUP, 0);
702 msleep(&map->busy, &map->system_mtx, 0, "mbusy", 0);
704 sx_sleep(&map->busy, &map->lock, 0, "mbusy", 0);
710 vmspace_resident_count(struct vmspace *vmspace)
712 return pmap_resident_count(vmspace_pmap(vmspace));
716 vmspace_wired_count(struct vmspace *vmspace)
718 return pmap_wired_count(vmspace_pmap(vmspace));
724 * Creates and returns a new empty VM map with
725 * the given physical map structure, and having
726 * the given lower and upper address bounds.
729 vm_map_create(pmap_t pmap, vm_offset_t min, vm_offset_t max)
733 result = uma_zalloc(mapzone, M_WAITOK);
734 CTR1(KTR_VM, "vm_map_create: %p", result);
735 _vm_map_init(result, min, max);
741 * Initialize an existing vm_map structure
742 * such as that in the vmspace structure.
743 * The pmap is set elsewhere.
746 _vm_map_init(vm_map_t map, vm_offset_t min, vm_offset_t max)
749 map->header.next = map->header.prev = &map->header;
750 map->needs_wakeup = FALSE;
752 map->min_offset = min;
753 map->max_offset = max;
761 vm_map_init(vm_map_t map, vm_offset_t min, vm_offset_t max)
763 _vm_map_init(map, min, max);
764 mtx_init(&map->system_mtx, "system map", NULL, MTX_DEF | MTX_DUPOK);
765 sx_init(&map->lock, "user map");
769 * vm_map_entry_dispose: [ internal use only ]
771 * Inverse of vm_map_entry_create.
774 vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry)
776 uma_zfree(map->system_map ? kmapentzone : mapentzone, entry);
780 * vm_map_entry_create: [ internal use only ]
782 * Allocates a VM map entry for insertion.
783 * No entry fields are filled in.
785 static vm_map_entry_t
786 vm_map_entry_create(vm_map_t map)
788 vm_map_entry_t new_entry;
791 new_entry = uma_zalloc(kmapentzone, M_NOWAIT);
793 new_entry = uma_zalloc(mapentzone, M_WAITOK);
794 if (new_entry == NULL)
795 panic("vm_map_entry_create: kernel resources exhausted");
800 * vm_map_entry_set_behavior:
802 * Set the expected access behavior, either normal, random, or
806 vm_map_entry_set_behavior(vm_map_entry_t entry, u_char behavior)
808 entry->eflags = (entry->eflags & ~MAP_ENTRY_BEHAV_MASK) |
809 (behavior & MAP_ENTRY_BEHAV_MASK);
813 * vm_map_entry_set_max_free:
815 * Set the max_free field in a vm_map_entry.
818 vm_map_entry_set_max_free(vm_map_entry_t entry)
821 entry->max_free = entry->adj_free;
822 if (entry->left != NULL && entry->left->max_free > entry->max_free)
823 entry->max_free = entry->left->max_free;
824 if (entry->right != NULL && entry->right->max_free > entry->max_free)
825 entry->max_free = entry->right->max_free;
829 * vm_map_entry_splay:
831 * The Sleator and Tarjan top-down splay algorithm with the
832 * following variation. Max_free must be computed bottom-up, so
833 * on the downward pass, maintain the left and right spines in
834 * reverse order. Then, make a second pass up each side to fix
835 * the pointers and compute max_free. The time bound is O(log n)
838 * The new root is the vm_map_entry containing "addr", or else an
839 * adjacent entry (lower or higher) if addr is not in the tree.
841 * The map must be locked, and leaves it so.
843 * Returns: the new root.
845 static vm_map_entry_t
846 vm_map_entry_splay(vm_offset_t addr, vm_map_entry_t root)
848 vm_map_entry_t llist, rlist;
849 vm_map_entry_t ltree, rtree;
852 /* Special case of empty tree. */
857 * Pass One: Splay down the tree until we find addr or a NULL
858 * pointer where addr would go. llist and rlist are the two
859 * sides in reverse order (bottom-up), with llist linked by
860 * the right pointer and rlist linked by the left pointer in
861 * the vm_map_entry. Wait until Pass Two to set max_free on
867 /* root is never NULL in here. */
868 if (addr < root->start) {
872 if (addr < y->start && y->left != NULL) {
873 /* Rotate right and put y on rlist. */
874 root->left = y->right;
876 vm_map_entry_set_max_free(root);
881 /* Put root on rlist. */
886 } else if (addr >= root->end) {
890 if (addr >= y->end && y->right != NULL) {
891 /* Rotate left and put y on llist. */
892 root->right = y->left;
894 vm_map_entry_set_max_free(root);
899 /* Put root on llist. */
909 * Pass Two: Walk back up the two spines, flip the pointers
910 * and set max_free. The subtrees of the root go at the
911 * bottom of llist and rlist.
914 while (llist != NULL) {
916 llist->right = ltree;
917 vm_map_entry_set_max_free(llist);
922 while (rlist != NULL) {
925 vm_map_entry_set_max_free(rlist);
931 * Final assembly: add ltree and rtree as subtrees of root.
935 vm_map_entry_set_max_free(root);
941 * vm_map_entry_{un,}link:
943 * Insert/remove entries from maps.
946 vm_map_entry_link(vm_map_t map,
947 vm_map_entry_t after_where,
948 vm_map_entry_t entry)
952 "vm_map_entry_link: map %p, nentries %d, entry %p, after %p", map,
953 map->nentries, entry, after_where);
954 VM_MAP_ASSERT_LOCKED(map);
956 entry->prev = after_where;
957 entry->next = after_where->next;
958 entry->next->prev = entry;
959 after_where->next = entry;
961 if (after_where != &map->header) {
962 if (after_where != map->root)
963 vm_map_entry_splay(after_where->start, map->root);
964 entry->right = after_where->right;
965 entry->left = after_where;
966 after_where->right = NULL;
967 after_where->adj_free = entry->start - after_where->end;
968 vm_map_entry_set_max_free(after_where);
970 entry->right = map->root;
973 entry->adj_free = (entry->next == &map->header ? map->max_offset :
974 entry->next->start) - entry->end;
975 vm_map_entry_set_max_free(entry);
980 vm_map_entry_unlink(vm_map_t map,
981 vm_map_entry_t entry)
983 vm_map_entry_t next, prev, root;
985 VM_MAP_ASSERT_LOCKED(map);
986 if (entry != map->root)
987 vm_map_entry_splay(entry->start, map->root);
988 if (entry->left == NULL)
991 root = vm_map_entry_splay(entry->start, entry->left);
992 root->right = entry->right;
993 root->adj_free = (entry->next == &map->header ? map->max_offset :
994 entry->next->start) - root->end;
995 vm_map_entry_set_max_free(root);
1004 CTR3(KTR_VM, "vm_map_entry_unlink: map %p, nentries %d, entry %p", map,
1005 map->nentries, entry);
1009 * vm_map_entry_resize_free:
1011 * Recompute the amount of free space following a vm_map_entry
1012 * and propagate that value up the tree. Call this function after
1013 * resizing a map entry in-place, that is, without a call to
1014 * vm_map_entry_link() or _unlink().
1016 * The map must be locked, and leaves it so.
1019 vm_map_entry_resize_free(vm_map_t map, vm_map_entry_t entry)
1023 * Using splay trees without parent pointers, propagating
1024 * max_free up the tree is done by moving the entry to the
1025 * root and making the change there.
1027 if (entry != map->root)
1028 map->root = vm_map_entry_splay(entry->start, map->root);
1030 entry->adj_free = (entry->next == &map->header ? map->max_offset :
1031 entry->next->start) - entry->end;
1032 vm_map_entry_set_max_free(entry);
1036 * vm_map_lookup_entry: [ internal use only ]
1038 * Finds the map entry containing (or
1039 * immediately preceding) the specified address
1040 * in the given map; the entry is returned
1041 * in the "entry" parameter. The boolean
1042 * result indicates whether the address is
1043 * actually contained in the map.
1046 vm_map_lookup_entry(
1048 vm_offset_t address,
1049 vm_map_entry_t *entry) /* OUT */
1055 * If the map is empty, then the map entry immediately preceding
1056 * "address" is the map's header.
1060 *entry = &map->header;
1061 else if (address >= cur->start && cur->end > address) {
1064 } else if ((locked = vm_map_locked(map)) ||
1065 sx_try_upgrade(&map->lock)) {
1067 * Splay requires a write lock on the map. However, it only
1068 * restructures the binary search tree; it does not otherwise
1069 * change the map. Thus, the map's timestamp need not change
1070 * on a temporary upgrade.
1072 map->root = cur = vm_map_entry_splay(address, cur);
1074 sx_downgrade(&map->lock);
1077 * If "address" is contained within a map entry, the new root
1078 * is that map entry. Otherwise, the new root is a map entry
1079 * immediately before or after "address".
1081 if (address >= cur->start) {
1083 if (cur->end > address)
1089 * Since the map is only locked for read access, perform a
1090 * standard binary search tree lookup for "address".
1093 if (address < cur->start) {
1094 if (cur->left == NULL) {
1099 } else if (cur->end > address) {
1103 if (cur->right == NULL) {
1116 * Inserts the given whole VM object into the target
1117 * map at the specified address range. The object's
1118 * size should match that of the address range.
1120 * Requires that the map be locked, and leaves it so.
1122 * If object is non-NULL, ref count must be bumped by caller
1123 * prior to making call to account for the new entry.
1126 vm_map_insert(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1127 vm_offset_t start, vm_offset_t end, vm_prot_t prot, vm_prot_t max,
1130 vm_map_entry_t new_entry;
1131 vm_map_entry_t prev_entry;
1132 vm_map_entry_t temp_entry;
1133 vm_eflags_t protoeflags;
1134 struct uidinfo *uip;
1135 boolean_t charge_prev_obj;
1137 VM_MAP_ASSERT_LOCKED(map);
1140 * Check that the start and end points are not bogus.
1142 if ((start < map->min_offset) || (end > map->max_offset) ||
1144 return (KERN_INVALID_ADDRESS);
1147 * Find the entry prior to the proposed starting address; if it's part
1148 * of an existing entry, this range is bogus.
1150 if (vm_map_lookup_entry(map, start, &temp_entry))
1151 return (KERN_NO_SPACE);
1153 prev_entry = temp_entry;
1156 * Assert that the next entry doesn't overlap the end point.
1158 if ((prev_entry->next != &map->header) &&
1159 (prev_entry->next->start < end))
1160 return (KERN_NO_SPACE);
1163 charge_prev_obj = FALSE;
1165 if (cow & MAP_COPY_ON_WRITE)
1166 protoeflags |= MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY;
1168 if (cow & MAP_NOFAULT) {
1169 protoeflags |= MAP_ENTRY_NOFAULT;
1171 KASSERT(object == NULL,
1172 ("vm_map_insert: paradoxical MAP_NOFAULT request"));
1174 if (cow & MAP_DISABLE_SYNCER)
1175 protoeflags |= MAP_ENTRY_NOSYNC;
1176 if (cow & MAP_DISABLE_COREDUMP)
1177 protoeflags |= MAP_ENTRY_NOCOREDUMP;
1180 KASSERT((object != kmem_object && object != kernel_object) ||
1181 ((object == kmem_object || object == kernel_object) &&
1182 !(protoeflags & MAP_ENTRY_NEEDS_COPY)),
1183 ("kmem or kernel object and cow"));
1184 if (cow & (MAP_ACC_NO_CHARGE | MAP_NOFAULT))
1186 if ((cow & MAP_ACC_CHARGED) || ((prot & VM_PROT_WRITE) &&
1187 ((protoeflags & MAP_ENTRY_NEEDS_COPY) || object == NULL))) {
1188 if (!(cow & MAP_ACC_CHARGED) && !swap_reserve(end - start))
1189 return (KERN_RESOURCE_SHORTAGE);
1190 KASSERT(object == NULL || (protoeflags & MAP_ENTRY_NEEDS_COPY) ||
1191 object->uip == NULL,
1192 ("OVERCOMMIT: vm_map_insert o %p", object));
1193 uip = curthread->td_ucred->cr_ruidinfo;
1195 if (object == NULL && !(protoeflags & MAP_ENTRY_NEEDS_COPY))
1196 charge_prev_obj = TRUE;
1200 /* Expand the kernel pmap, if necessary. */
1201 if (map == kernel_map && end > kernel_vm_end)
1202 pmap_growkernel(end);
1203 if (object != NULL) {
1205 * OBJ_ONEMAPPING must be cleared unless this mapping
1206 * is trivially proven to be the only mapping for any
1207 * of the object's pages. (Object granularity
1208 * reference counting is insufficient to recognize
1209 * aliases with precision.)
1211 VM_OBJECT_LOCK(object);
1212 if (object->ref_count > 1 || object->shadow_count != 0)
1213 vm_object_clear_flag(object, OBJ_ONEMAPPING);
1214 VM_OBJECT_UNLOCK(object);
1216 else if ((prev_entry != &map->header) &&
1217 (prev_entry->eflags == protoeflags) &&
1218 (prev_entry->end == start) &&
1219 (prev_entry->wired_count == 0) &&
1220 (prev_entry->uip == uip ||
1221 (prev_entry->object.vm_object != NULL &&
1222 (prev_entry->object.vm_object->uip == uip))) &&
1223 vm_object_coalesce(prev_entry->object.vm_object,
1225 (vm_size_t)(prev_entry->end - prev_entry->start),
1226 (vm_size_t)(end - prev_entry->end), charge_prev_obj)) {
1228 * We were able to extend the object. Determine if we
1229 * can extend the previous map entry to include the
1230 * new range as well.
1232 if ((prev_entry->inheritance == VM_INHERIT_DEFAULT) &&
1233 (prev_entry->protection == prot) &&
1234 (prev_entry->max_protection == max)) {
1235 map->size += (end - prev_entry->end);
1236 prev_entry->end = end;
1237 vm_map_entry_resize_free(map, prev_entry);
1238 vm_map_simplify_entry(map, prev_entry);
1241 return (KERN_SUCCESS);
1245 * If we can extend the object but cannot extend the
1246 * map entry, we have to create a new map entry. We
1247 * must bump the ref count on the extended object to
1248 * account for it. object may be NULL.
1250 object = prev_entry->object.vm_object;
1251 offset = prev_entry->offset +
1252 (prev_entry->end - prev_entry->start);
1253 vm_object_reference(object);
1254 if (uip != NULL && object != NULL && object->uip != NULL &&
1255 !(prev_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
1256 /* Object already accounts for this uid. */
1263 * NOTE: if conditionals fail, object can be NULL here. This occurs
1264 * in things like the buffer map where we manage kva but do not manage
1269 * Create a new entry
1271 new_entry = vm_map_entry_create(map);
1272 new_entry->start = start;
1273 new_entry->end = end;
1274 new_entry->uip = NULL;
1276 new_entry->eflags = protoeflags;
1277 new_entry->object.vm_object = object;
1278 new_entry->offset = offset;
1279 new_entry->avail_ssize = 0;
1281 new_entry->inheritance = VM_INHERIT_DEFAULT;
1282 new_entry->protection = prot;
1283 new_entry->max_protection = max;
1284 new_entry->wired_count = 0;
1286 KASSERT(uip == NULL || !ENTRY_CHARGED(new_entry),
1287 ("OVERCOMMIT: vm_map_insert leaks vm_map %p", new_entry));
1288 new_entry->uip = uip;
1291 * Insert the new entry into the list
1293 vm_map_entry_link(map, prev_entry, new_entry);
1294 map->size += new_entry->end - new_entry->start;
1298 * Temporarily removed to avoid MAP_STACK panic, due to
1299 * MAP_STACK being a huge hack. Will be added back in
1300 * when MAP_STACK (and the user stack mapping) is fixed.
1303 * It may be possible to simplify the entry
1305 vm_map_simplify_entry(map, new_entry);
1308 if (cow & (MAP_PREFAULT|MAP_PREFAULT_PARTIAL)) {
1309 vm_map_pmap_enter(map, start, prot,
1310 object, OFF_TO_IDX(offset), end - start,
1311 cow & MAP_PREFAULT_PARTIAL);
1314 return (KERN_SUCCESS);
1320 * Find the first fit (lowest VM address) for "length" free bytes
1321 * beginning at address >= start in the given map.
1323 * In a vm_map_entry, "adj_free" is the amount of free space
1324 * adjacent (higher address) to this entry, and "max_free" is the
1325 * maximum amount of contiguous free space in its subtree. This
1326 * allows finding a free region in one path down the tree, so
1327 * O(log n) amortized with splay trees.
1329 * The map must be locked, and leaves it so.
1331 * Returns: 0 on success, and starting address in *addr,
1332 * 1 if insufficient space.
1335 vm_map_findspace(vm_map_t map, vm_offset_t start, vm_size_t length,
1336 vm_offset_t *addr) /* OUT */
1338 vm_map_entry_t entry;
1342 * Request must fit within min/max VM address and must avoid
1345 if (start < map->min_offset)
1346 start = map->min_offset;
1347 if (start + length > map->max_offset || start + length < start)
1350 /* Empty tree means wide open address space. */
1351 if (map->root == NULL) {
1357 * After splay, if start comes before root node, then there
1358 * must be a gap from start to the root.
1360 map->root = vm_map_entry_splay(start, map->root);
1361 if (start + length <= map->root->start) {
1367 * Root is the last node that might begin its gap before
1368 * start, and this is the last comparison where address
1369 * wrap might be a problem.
1371 st = (start > map->root->end) ? start : map->root->end;
1372 if (length <= map->root->end + map->root->adj_free - st) {
1377 /* With max_free, can immediately tell if no solution. */
1378 entry = map->root->right;
1379 if (entry == NULL || length > entry->max_free)
1383 * Search the right subtree in the order: left subtree, root,
1384 * right subtree (first fit). The previous splay implies that
1385 * all regions in the right subtree have addresses > start.
1387 while (entry != NULL) {
1388 if (entry->left != NULL && entry->left->max_free >= length)
1389 entry = entry->left;
1390 else if (entry->adj_free >= length) {
1394 entry = entry->right;
1397 /* Can't get here, so panic if we do. */
1398 panic("vm_map_findspace: max_free corrupt");
1402 vm_map_fixed(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1403 vm_offset_t start, vm_size_t length, vm_prot_t prot,
1404 vm_prot_t max, int cow)
1409 end = start + length;
1411 VM_MAP_RANGE_CHECK(map, start, end);
1412 (void) vm_map_delete(map, start, end);
1413 result = vm_map_insert(map, object, offset, start, end, prot,
1420 * vm_map_find finds an unallocated region in the target address
1421 * map with the given length. The search is defined to be
1422 * first-fit from the specified address; the region found is
1423 * returned in the same parameter.
1425 * If object is non-NULL, ref count must be bumped by caller
1426 * prior to making call to account for the new entry.
1429 vm_map_find(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1430 vm_offset_t *addr, /* IN/OUT */
1431 vm_size_t length, int find_space, vm_prot_t prot,
1432 vm_prot_t max, int cow)
1440 if (find_space != VMFS_NO_SPACE) {
1441 if (vm_map_findspace(map, start, length, addr)) {
1443 return (KERN_NO_SPACE);
1445 switch (find_space) {
1446 case VMFS_ALIGNED_SPACE:
1447 pmap_align_superpage(object, offset, addr,
1450 #ifdef VMFS_TLB_ALIGNED_SPACE
1451 case VMFS_TLB_ALIGNED_SPACE:
1452 pmap_align_tlb(addr);
1461 result = vm_map_insert(map, object, offset, start, start +
1462 length, prot, max, cow);
1463 } while (result == KERN_NO_SPACE && find_space == VMFS_ALIGNED_SPACE);
1469 * vm_map_simplify_entry:
1471 * Simplify the given map entry by merging with either neighbor. This
1472 * routine also has the ability to merge with both neighbors.
1474 * The map must be locked.
1476 * This routine guarentees that the passed entry remains valid (though
1477 * possibly extended). When merging, this routine may delete one or
1481 vm_map_simplify_entry(vm_map_t map, vm_map_entry_t entry)
1483 vm_map_entry_t next, prev;
1484 vm_size_t prevsize, esize;
1486 if (entry->eflags & (MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_IS_SUB_MAP))
1490 if (prev != &map->header) {
1491 prevsize = prev->end - prev->start;
1492 if ( (prev->end == entry->start) &&
1493 (prev->object.vm_object == entry->object.vm_object) &&
1494 (!prev->object.vm_object ||
1495 (prev->offset + prevsize == entry->offset)) &&
1496 (prev->eflags == entry->eflags) &&
1497 (prev->protection == entry->protection) &&
1498 (prev->max_protection == entry->max_protection) &&
1499 (prev->inheritance == entry->inheritance) &&
1500 (prev->wired_count == entry->wired_count) &&
1501 (prev->uip == entry->uip)) {
1502 vm_map_entry_unlink(map, prev);
1503 entry->start = prev->start;
1504 entry->offset = prev->offset;
1505 if (entry->prev != &map->header)
1506 vm_map_entry_resize_free(map, entry->prev);
1509 * If the backing object is a vnode object,
1510 * vm_object_deallocate() calls vrele().
1511 * However, vrele() does not lock the vnode
1512 * because the vnode has additional
1513 * references. Thus, the map lock can be kept
1514 * without causing a lock-order reversal with
1517 if (prev->object.vm_object)
1518 vm_object_deallocate(prev->object.vm_object);
1519 if (prev->uip != NULL)
1521 vm_map_entry_dispose(map, prev);
1526 if (next != &map->header) {
1527 esize = entry->end - entry->start;
1528 if ((entry->end == next->start) &&
1529 (next->object.vm_object == entry->object.vm_object) &&
1530 (!entry->object.vm_object ||
1531 (entry->offset + esize == next->offset)) &&
1532 (next->eflags == entry->eflags) &&
1533 (next->protection == entry->protection) &&
1534 (next->max_protection == entry->max_protection) &&
1535 (next->inheritance == entry->inheritance) &&
1536 (next->wired_count == entry->wired_count) &&
1537 (next->uip == entry->uip)) {
1538 vm_map_entry_unlink(map, next);
1539 entry->end = next->end;
1540 vm_map_entry_resize_free(map, entry);
1543 * See comment above.
1545 if (next->object.vm_object)
1546 vm_object_deallocate(next->object.vm_object);
1547 if (next->uip != NULL)
1549 vm_map_entry_dispose(map, next);
1554 * vm_map_clip_start: [ internal use only ]
1556 * Asserts that the given entry begins at or after
1557 * the specified address; if necessary,
1558 * it splits the entry into two.
1560 #define vm_map_clip_start(map, entry, startaddr) \
1562 if (startaddr > entry->start) \
1563 _vm_map_clip_start(map, entry, startaddr); \
1567 * This routine is called only when it is known that
1568 * the entry must be split.
1571 _vm_map_clip_start(vm_map_t map, vm_map_entry_t entry, vm_offset_t start)
1573 vm_map_entry_t new_entry;
1575 VM_MAP_ASSERT_LOCKED(map);
1578 * Split off the front portion -- note that we must insert the new
1579 * entry BEFORE this one, so that this entry has the specified
1582 vm_map_simplify_entry(map, entry);
1585 * If there is no object backing this entry, we might as well create
1586 * one now. If we defer it, an object can get created after the map
1587 * is clipped, and individual objects will be created for the split-up
1588 * map. This is a bit of a hack, but is also about the best place to
1589 * put this improvement.
1591 if (entry->object.vm_object == NULL && !map->system_map) {
1593 object = vm_object_allocate(OBJT_DEFAULT,
1594 atop(entry->end - entry->start));
1595 entry->object.vm_object = object;
1597 if (entry->uip != NULL) {
1598 object->uip = entry->uip;
1599 object->charge = entry->end - entry->start;
1602 } else if (entry->object.vm_object != NULL &&
1603 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) &&
1604 entry->uip != NULL) {
1605 VM_OBJECT_LOCK(entry->object.vm_object);
1606 KASSERT(entry->object.vm_object->uip == NULL,
1607 ("OVERCOMMIT: vm_entry_clip_start: both uip e %p", entry));
1608 entry->object.vm_object->uip = entry->uip;
1609 entry->object.vm_object->charge = entry->end - entry->start;
1610 VM_OBJECT_UNLOCK(entry->object.vm_object);
1614 new_entry = vm_map_entry_create(map);
1615 *new_entry = *entry;
1617 new_entry->end = start;
1618 entry->offset += (start - entry->start);
1619 entry->start = start;
1620 if (new_entry->uip != NULL)
1623 vm_map_entry_link(map, entry->prev, new_entry);
1625 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1626 vm_object_reference(new_entry->object.vm_object);
1631 * vm_map_clip_end: [ internal use only ]
1633 * Asserts that the given entry ends at or before
1634 * the specified address; if necessary,
1635 * it splits the entry into two.
1637 #define vm_map_clip_end(map, entry, endaddr) \
1639 if ((endaddr) < (entry->end)) \
1640 _vm_map_clip_end((map), (entry), (endaddr)); \
1644 * This routine is called only when it is known that
1645 * the entry must be split.
1648 _vm_map_clip_end(vm_map_t map, vm_map_entry_t entry, vm_offset_t end)
1650 vm_map_entry_t new_entry;
1652 VM_MAP_ASSERT_LOCKED(map);
1655 * If there is no object backing this entry, we might as well create
1656 * one now. If we defer it, an object can get created after the map
1657 * is clipped, and individual objects will be created for the split-up
1658 * map. This is a bit of a hack, but is also about the best place to
1659 * put this improvement.
1661 if (entry->object.vm_object == NULL && !map->system_map) {
1663 object = vm_object_allocate(OBJT_DEFAULT,
1664 atop(entry->end - entry->start));
1665 entry->object.vm_object = object;
1667 if (entry->uip != NULL) {
1668 object->uip = entry->uip;
1669 object->charge = entry->end - entry->start;
1672 } else if (entry->object.vm_object != NULL &&
1673 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) &&
1674 entry->uip != NULL) {
1675 VM_OBJECT_LOCK(entry->object.vm_object);
1676 KASSERT(entry->object.vm_object->uip == NULL,
1677 ("OVERCOMMIT: vm_entry_clip_end: both uip e %p", entry));
1678 entry->object.vm_object->uip = entry->uip;
1679 entry->object.vm_object->charge = entry->end - entry->start;
1680 VM_OBJECT_UNLOCK(entry->object.vm_object);
1685 * Create a new entry and insert it AFTER the specified entry
1687 new_entry = vm_map_entry_create(map);
1688 *new_entry = *entry;
1690 new_entry->start = entry->end = end;
1691 new_entry->offset += (end - entry->start);
1692 if (new_entry->uip != NULL)
1695 vm_map_entry_link(map, entry, new_entry);
1697 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1698 vm_object_reference(new_entry->object.vm_object);
1703 * vm_map_submap: [ kernel use only ]
1705 * Mark the given range as handled by a subordinate map.
1707 * This range must have been created with vm_map_find,
1708 * and no other operations may have been performed on this
1709 * range prior to calling vm_map_submap.
1711 * Only a limited number of operations can be performed
1712 * within this rage after calling vm_map_submap:
1714 * [Don't try vm_map_copy!]
1716 * To remove a submapping, one must first remove the
1717 * range from the superior map, and then destroy the
1718 * submap (if desired). [Better yet, don't try it.]
1727 vm_map_entry_t entry;
1728 int result = KERN_INVALID_ARGUMENT;
1732 VM_MAP_RANGE_CHECK(map, start, end);
1734 if (vm_map_lookup_entry(map, start, &entry)) {
1735 vm_map_clip_start(map, entry, start);
1737 entry = entry->next;
1739 vm_map_clip_end(map, entry, end);
1741 if ((entry->start == start) && (entry->end == end) &&
1742 ((entry->eflags & MAP_ENTRY_COW) == 0) &&
1743 (entry->object.vm_object == NULL)) {
1744 entry->object.sub_map = submap;
1745 entry->eflags |= MAP_ENTRY_IS_SUB_MAP;
1746 result = KERN_SUCCESS;
1754 * The maximum number of pages to map
1756 #define MAX_INIT_PT 96
1759 * vm_map_pmap_enter:
1761 * Preload read-only mappings for the given object's resident pages into
1762 * the given map. This eliminates the soft faults on process startup and
1763 * immediately after an mmap(2). Because these are speculative mappings,
1764 * cached pages are not reactivated and mapped.
1767 vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot,
1768 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags)
1771 vm_page_t p, p_start;
1772 vm_pindex_t psize, tmpidx;
1773 boolean_t are_queues_locked;
1775 if ((prot & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0 || object == NULL)
1777 VM_OBJECT_LOCK(object);
1778 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) {
1779 pmap_object_init_pt(map->pmap, addr, object, pindex, size);
1785 if ((flags & MAP_PREFAULT_PARTIAL) && psize > MAX_INIT_PT &&
1786 object->resident_page_count > MAX_INIT_PT)
1789 if (psize + pindex > object->size) {
1790 if (object->size < pindex)
1792 psize = object->size - pindex;
1795 are_queues_locked = FALSE;
1799 p = vm_page_find_least(object, pindex);
1801 * Assert: the variable p is either (1) the page with the
1802 * least pindex greater than or equal to the parameter pindex
1806 p != NULL && (tmpidx = p->pindex - pindex) < psize;
1807 p = TAILQ_NEXT(p, listq)) {
1809 * don't allow an madvise to blow away our really
1810 * free pages allocating pv entries.
1812 if ((flags & MAP_PREFAULT_MADVISE) &&
1813 cnt.v_free_count < cnt.v_free_reserved) {
1817 if (p->valid == VM_PAGE_BITS_ALL) {
1818 if (p_start == NULL) {
1819 start = addr + ptoa(tmpidx);
1822 } else if (p_start != NULL) {
1823 if (!are_queues_locked) {
1824 are_queues_locked = TRUE;
1825 vm_page_lock_queues();
1827 pmap_enter_object(map->pmap, start, addr +
1828 ptoa(tmpidx), p_start, prot);
1832 if (p_start != NULL) {
1833 if (!are_queues_locked) {
1834 are_queues_locked = TRUE;
1835 vm_page_lock_queues();
1837 pmap_enter_object(map->pmap, start, addr + ptoa(psize),
1840 if (are_queues_locked)
1841 vm_page_unlock_queues();
1843 VM_OBJECT_UNLOCK(object);
1849 * Sets the protection of the specified address
1850 * region in the target map. If "set_max" is
1851 * specified, the maximum protection is to be set;
1852 * otherwise, only the current protection is affected.
1855 vm_map_protect(vm_map_t map, vm_offset_t start, vm_offset_t end,
1856 vm_prot_t new_prot, boolean_t set_max)
1858 vm_map_entry_t current, entry;
1860 struct uidinfo *uip;
1865 VM_MAP_RANGE_CHECK(map, start, end);
1867 if (vm_map_lookup_entry(map, start, &entry)) {
1868 vm_map_clip_start(map, entry, start);
1870 entry = entry->next;
1874 * Make a first pass to check for protection violations.
1877 while ((current != &map->header) && (current->start < end)) {
1878 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
1880 return (KERN_INVALID_ARGUMENT);
1882 if ((new_prot & current->max_protection) != new_prot) {
1884 return (KERN_PROTECTION_FAILURE);
1886 current = current->next;
1891 * Do an accounting pass for private read-only mappings that
1892 * now will do cow due to allowed write (e.g. debugger sets
1893 * breakpoint on text segment)
1895 for (current = entry; (current != &map->header) &&
1896 (current->start < end); current = current->next) {
1898 vm_map_clip_end(map, current, end);
1901 ((new_prot & ~(current->protection)) & VM_PROT_WRITE) == 0 ||
1902 ENTRY_CHARGED(current)) {
1906 uip = curthread->td_ucred->cr_ruidinfo;
1907 obj = current->object.vm_object;
1909 if (obj == NULL || (current->eflags & MAP_ENTRY_NEEDS_COPY)) {
1910 if (!swap_reserve(current->end - current->start)) {
1912 return (KERN_RESOURCE_SHORTAGE);
1919 VM_OBJECT_LOCK(obj);
1920 if (obj->type != OBJT_DEFAULT && obj->type != OBJT_SWAP) {
1921 VM_OBJECT_UNLOCK(obj);
1926 * Charge for the whole object allocation now, since
1927 * we cannot distinguish between non-charged and
1928 * charged clipped mapping of the same object later.
1930 KASSERT(obj->charge == 0,
1931 ("vm_map_protect: object %p overcharged\n", obj));
1932 if (!swap_reserve(ptoa(obj->size))) {
1933 VM_OBJECT_UNLOCK(obj);
1935 return (KERN_RESOURCE_SHORTAGE);
1940 obj->charge = ptoa(obj->size);
1941 VM_OBJECT_UNLOCK(obj);
1945 * Go back and fix up protections. [Note that clipping is not
1946 * necessary the second time.]
1949 while ((current != &map->header) && (current->start < end)) {
1950 old_prot = current->protection;
1953 current->protection =
1954 (current->max_protection = new_prot) &
1957 current->protection = new_prot;
1959 if ((current->eflags & (MAP_ENTRY_COW | MAP_ENTRY_USER_WIRED))
1960 == (MAP_ENTRY_COW | MAP_ENTRY_USER_WIRED) &&
1961 (current->protection & VM_PROT_WRITE) != 0 &&
1962 (old_prot & VM_PROT_WRITE) == 0) {
1963 vm_fault_copy_entry(map, map, current, current, NULL);
1967 * Update physical map if necessary. Worry about copy-on-write
1970 if (current->protection != old_prot) {
1971 #define MASK(entry) (((entry)->eflags & MAP_ENTRY_COW) ? ~VM_PROT_WRITE : \
1973 pmap_protect(map->pmap, current->start,
1975 current->protection & MASK(current));
1978 vm_map_simplify_entry(map, current);
1979 current = current->next;
1982 return (KERN_SUCCESS);
1988 * This routine traverses a processes map handling the madvise
1989 * system call. Advisories are classified as either those effecting
1990 * the vm_map_entry structure, or those effecting the underlying
2000 vm_map_entry_t current, entry;
2004 * Some madvise calls directly modify the vm_map_entry, in which case
2005 * we need to use an exclusive lock on the map and we need to perform
2006 * various clipping operations. Otherwise we only need a read-lock
2011 case MADV_SEQUENTIAL:
2023 vm_map_lock_read(map);
2026 return (KERN_INVALID_ARGUMENT);
2030 * Locate starting entry and clip if necessary.
2032 VM_MAP_RANGE_CHECK(map, start, end);
2034 if (vm_map_lookup_entry(map, start, &entry)) {
2036 vm_map_clip_start(map, entry, start);
2038 entry = entry->next;
2043 * madvise behaviors that are implemented in the vm_map_entry.
2045 * We clip the vm_map_entry so that behavioral changes are
2046 * limited to the specified address range.
2048 for (current = entry;
2049 (current != &map->header) && (current->start < end);
2050 current = current->next
2052 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
2055 vm_map_clip_end(map, current, end);
2059 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_NORMAL);
2061 case MADV_SEQUENTIAL:
2062 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_SEQUENTIAL);
2065 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_RANDOM);
2068 current->eflags |= MAP_ENTRY_NOSYNC;
2071 current->eflags &= ~MAP_ENTRY_NOSYNC;
2074 current->eflags |= MAP_ENTRY_NOCOREDUMP;
2077 current->eflags &= ~MAP_ENTRY_NOCOREDUMP;
2082 vm_map_simplify_entry(map, current);
2090 * madvise behaviors that are implemented in the underlying
2093 * Since we don't clip the vm_map_entry, we have to clip
2094 * the vm_object pindex and count.
2096 for (current = entry;
2097 (current != &map->header) && (current->start < end);
2098 current = current->next
2100 vm_offset_t useStart;
2102 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
2105 pindex = OFF_TO_IDX(current->offset);
2106 count = atop(current->end - current->start);
2107 useStart = current->start;
2109 if (current->start < start) {
2110 pindex += atop(start - current->start);
2111 count -= atop(start - current->start);
2114 if (current->end > end)
2115 count -= atop(current->end - end);
2120 vm_object_madvise(current->object.vm_object,
2121 pindex, count, behav);
2122 if (behav == MADV_WILLNEED) {
2123 vm_map_pmap_enter(map,
2125 current->protection,
2126 current->object.vm_object,
2128 (count << PAGE_SHIFT),
2129 MAP_PREFAULT_MADVISE
2133 vm_map_unlock_read(map);
2142 * Sets the inheritance of the specified address
2143 * range in the target map. Inheritance
2144 * affects how the map will be shared with
2145 * child maps at the time of vmspace_fork.
2148 vm_map_inherit(vm_map_t map, vm_offset_t start, vm_offset_t end,
2149 vm_inherit_t new_inheritance)
2151 vm_map_entry_t entry;
2152 vm_map_entry_t temp_entry;
2154 switch (new_inheritance) {
2155 case VM_INHERIT_NONE:
2156 case VM_INHERIT_COPY:
2157 case VM_INHERIT_SHARE:
2160 return (KERN_INVALID_ARGUMENT);
2163 VM_MAP_RANGE_CHECK(map, start, end);
2164 if (vm_map_lookup_entry(map, start, &temp_entry)) {
2166 vm_map_clip_start(map, entry, start);
2168 entry = temp_entry->next;
2169 while ((entry != &map->header) && (entry->start < end)) {
2170 vm_map_clip_end(map, entry, end);
2171 entry->inheritance = new_inheritance;
2172 vm_map_simplify_entry(map, entry);
2173 entry = entry->next;
2176 return (KERN_SUCCESS);
2182 * Implements both kernel and user unwiring.
2185 vm_map_unwire(vm_map_t map, vm_offset_t start, vm_offset_t end,
2188 vm_map_entry_t entry, first_entry, tmp_entry;
2189 vm_offset_t saved_start;
2190 unsigned int last_timestamp;
2192 boolean_t need_wakeup, result, user_unwire;
2194 user_unwire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE;
2196 VM_MAP_RANGE_CHECK(map, start, end);
2197 if (!vm_map_lookup_entry(map, start, &first_entry)) {
2198 if (flags & VM_MAP_WIRE_HOLESOK)
2199 first_entry = first_entry->next;
2202 return (KERN_INVALID_ADDRESS);
2205 last_timestamp = map->timestamp;
2206 entry = first_entry;
2207 while (entry != &map->header && entry->start < end) {
2208 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
2210 * We have not yet clipped the entry.
2212 saved_start = (start >= entry->start) ? start :
2214 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2215 if (vm_map_unlock_and_wait(map, 0)) {
2217 * Allow interruption of user unwiring?
2221 if (last_timestamp+1 != map->timestamp) {
2223 * Look again for the entry because the map was
2224 * modified while it was unlocked.
2225 * Specifically, the entry may have been
2226 * clipped, merged, or deleted.
2228 if (!vm_map_lookup_entry(map, saved_start,
2230 if (flags & VM_MAP_WIRE_HOLESOK)
2231 tmp_entry = tmp_entry->next;
2233 if (saved_start == start) {
2235 * First_entry has been deleted.
2238 return (KERN_INVALID_ADDRESS);
2241 rv = KERN_INVALID_ADDRESS;
2245 if (entry == first_entry)
2246 first_entry = tmp_entry;
2251 last_timestamp = map->timestamp;
2254 vm_map_clip_start(map, entry, start);
2255 vm_map_clip_end(map, entry, end);
2257 * Mark the entry in case the map lock is released. (See
2260 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
2262 * Check the map for holes in the specified region.
2263 * If VM_MAP_WIRE_HOLESOK was specified, skip this check.
2265 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) &&
2266 (entry->end < end && (entry->next == &map->header ||
2267 entry->next->start > entry->end))) {
2269 rv = KERN_INVALID_ADDRESS;
2273 * If system unwiring, require that the entry is system wired.
2276 vm_map_entry_system_wired_count(entry) == 0) {
2278 rv = KERN_INVALID_ARGUMENT;
2281 entry = entry->next;
2285 need_wakeup = FALSE;
2286 if (first_entry == NULL) {
2287 result = vm_map_lookup_entry(map, start, &first_entry);
2288 if (!result && (flags & VM_MAP_WIRE_HOLESOK))
2289 first_entry = first_entry->next;
2291 KASSERT(result, ("vm_map_unwire: lookup failed"));
2293 entry = first_entry;
2294 while (entry != &map->header && entry->start < end) {
2295 if (rv == KERN_SUCCESS && (!user_unwire ||
2296 (entry->eflags & MAP_ENTRY_USER_WIRED))) {
2298 entry->eflags &= ~MAP_ENTRY_USER_WIRED;
2299 entry->wired_count--;
2300 if (entry->wired_count == 0) {
2302 * Retain the map lock.
2304 vm_fault_unwire(map, entry->start, entry->end,
2305 entry->object.vm_object != NULL &&
2306 (entry->object.vm_object->type == OBJT_DEVICE ||
2307 entry->object.vm_object->type == OBJT_SG));
2310 KASSERT(entry->eflags & MAP_ENTRY_IN_TRANSITION,
2311 ("vm_map_unwire: in-transition flag missing"));
2312 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION;
2313 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
2314 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
2317 vm_map_simplify_entry(map, entry);
2318 entry = entry->next;
2329 * Implements both kernel and user wiring.
2332 vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t end,
2335 vm_map_entry_t entry, first_entry, tmp_entry;
2336 vm_offset_t saved_end, saved_start;
2337 unsigned int last_timestamp;
2339 boolean_t fictitious, need_wakeup, result, user_wire;
2341 user_wire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE;
2343 VM_MAP_RANGE_CHECK(map, start, end);
2344 if (!vm_map_lookup_entry(map, start, &first_entry)) {
2345 if (flags & VM_MAP_WIRE_HOLESOK)
2346 first_entry = first_entry->next;
2349 return (KERN_INVALID_ADDRESS);
2352 last_timestamp = map->timestamp;
2353 entry = first_entry;
2354 while (entry != &map->header && entry->start < end) {
2355 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
2357 * We have not yet clipped the entry.
2359 saved_start = (start >= entry->start) ? start :
2361 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2362 if (vm_map_unlock_and_wait(map, 0)) {
2364 * Allow interruption of user wiring?
2368 if (last_timestamp + 1 != map->timestamp) {
2370 * Look again for the entry because the map was
2371 * modified while it was unlocked.
2372 * Specifically, the entry may have been
2373 * clipped, merged, or deleted.
2375 if (!vm_map_lookup_entry(map, saved_start,
2377 if (flags & VM_MAP_WIRE_HOLESOK)
2378 tmp_entry = tmp_entry->next;
2380 if (saved_start == start) {
2382 * first_entry has been deleted.
2385 return (KERN_INVALID_ADDRESS);
2388 rv = KERN_INVALID_ADDRESS;
2392 if (entry == first_entry)
2393 first_entry = tmp_entry;
2398 last_timestamp = map->timestamp;
2401 vm_map_clip_start(map, entry, start);
2402 vm_map_clip_end(map, entry, end);
2404 * Mark the entry in case the map lock is released. (See
2407 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
2411 if (entry->wired_count == 0) {
2412 if ((entry->protection & (VM_PROT_READ|VM_PROT_EXECUTE))
2414 entry->eflags |= MAP_ENTRY_WIRE_SKIPPED;
2415 if ((flags & VM_MAP_WIRE_HOLESOK) == 0) {
2417 rv = KERN_INVALID_ADDRESS;
2422 entry->wired_count++;
2423 saved_start = entry->start;
2424 saved_end = entry->end;
2425 fictitious = entry->object.vm_object != NULL &&
2426 (entry->object.vm_object->type == OBJT_DEVICE ||
2427 entry->object.vm_object->type == OBJT_SG);
2429 * Release the map lock, relying on the in-transition
2430 * mark. Mark the map busy for fork.
2434 rv = vm_fault_wire(map, saved_start, saved_end,
2435 user_wire, fictitious);
2438 if (last_timestamp + 1 != map->timestamp) {
2440 * Look again for the entry because the map was
2441 * modified while it was unlocked. The entry
2442 * may have been clipped, but NOT merged or
2445 result = vm_map_lookup_entry(map, saved_start,
2447 KASSERT(result, ("vm_map_wire: lookup failed"));
2448 if (entry == first_entry)
2449 first_entry = tmp_entry;
2453 while (entry->end < saved_end) {
2454 if (rv != KERN_SUCCESS) {
2455 KASSERT(entry->wired_count == 1,
2456 ("vm_map_wire: bad count"));
2457 entry->wired_count = -1;
2459 entry = entry->next;
2462 last_timestamp = map->timestamp;
2463 if (rv != KERN_SUCCESS) {
2464 KASSERT(entry->wired_count == 1,
2465 ("vm_map_wire: bad count"));
2467 * Assign an out-of-range value to represent
2468 * the failure to wire this entry.
2470 entry->wired_count = -1;
2474 } else if (!user_wire ||
2475 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
2476 entry->wired_count++;
2479 * Check the map for holes in the specified region.
2480 * If VM_MAP_WIRE_HOLESOK was specified, skip this check.
2483 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) &&
2484 (entry->end < end && (entry->next == &map->header ||
2485 entry->next->start > entry->end))) {
2487 rv = KERN_INVALID_ADDRESS;
2490 entry = entry->next;
2494 need_wakeup = FALSE;
2495 if (first_entry == NULL) {
2496 result = vm_map_lookup_entry(map, start, &first_entry);
2497 if (!result && (flags & VM_MAP_WIRE_HOLESOK))
2498 first_entry = first_entry->next;
2500 KASSERT(result, ("vm_map_wire: lookup failed"));
2502 entry = first_entry;
2503 while (entry != &map->header && entry->start < end) {
2504 if ((entry->eflags & MAP_ENTRY_WIRE_SKIPPED) != 0)
2505 goto next_entry_done;
2506 if (rv == KERN_SUCCESS) {
2508 entry->eflags |= MAP_ENTRY_USER_WIRED;
2509 } else if (entry->wired_count == -1) {
2511 * Wiring failed on this entry. Thus, unwiring is
2514 entry->wired_count = 0;
2517 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0)
2518 entry->wired_count--;
2519 if (entry->wired_count == 0) {
2521 * Retain the map lock.
2523 vm_fault_unwire(map, entry->start, entry->end,
2524 entry->object.vm_object != NULL &&
2525 (entry->object.vm_object->type == OBJT_DEVICE ||
2526 entry->object.vm_object->type == OBJT_SG));
2530 KASSERT(entry->eflags & MAP_ENTRY_IN_TRANSITION,
2531 ("vm_map_wire: in-transition flag missing"));
2532 entry->eflags &= ~(MAP_ENTRY_IN_TRANSITION|MAP_ENTRY_WIRE_SKIPPED);
2533 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
2534 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
2537 vm_map_simplify_entry(map, entry);
2538 entry = entry->next;
2549 * Push any dirty cached pages in the address range to their pager.
2550 * If syncio is TRUE, dirty pages are written synchronously.
2551 * If invalidate is TRUE, any cached pages are freed as well.
2553 * If the size of the region from start to end is zero, we are
2554 * supposed to flush all modified pages within the region containing
2555 * start. Unfortunately, a region can be split or coalesced with
2556 * neighboring regions, making it difficult to determine what the
2557 * original region was. Therefore, we approximate this requirement by
2558 * flushing the current region containing start.
2560 * Returns an error if any part of the specified range is not mapped.
2568 boolean_t invalidate)
2570 vm_map_entry_t current;
2571 vm_map_entry_t entry;
2574 vm_ooffset_t offset;
2575 unsigned int last_timestamp;
2577 vm_map_lock_read(map);
2578 VM_MAP_RANGE_CHECK(map, start, end);
2579 if (!vm_map_lookup_entry(map, start, &entry)) {
2580 vm_map_unlock_read(map);
2581 return (KERN_INVALID_ADDRESS);
2582 } else if (start == end) {
2583 start = entry->start;
2587 * Make a first pass to check for user-wired memory and holes.
2589 for (current = entry; current != &map->header && current->start < end;
2590 current = current->next) {
2591 if (invalidate && (current->eflags & MAP_ENTRY_USER_WIRED)) {
2592 vm_map_unlock_read(map);
2593 return (KERN_INVALID_ARGUMENT);
2595 if (end > current->end &&
2596 (current->next == &map->header ||
2597 current->end != current->next->start)) {
2598 vm_map_unlock_read(map);
2599 return (KERN_INVALID_ADDRESS);
2604 pmap_remove(map->pmap, start, end);
2607 * Make a second pass, cleaning/uncaching pages from the indicated
2610 for (current = entry; current != &map->header && current->start < end;) {
2611 offset = current->offset + (start - current->start);
2612 size = (end <= current->end ? end : current->end) - start;
2613 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
2615 vm_map_entry_t tentry;
2618 smap = current->object.sub_map;
2619 vm_map_lock_read(smap);
2620 (void) vm_map_lookup_entry(smap, offset, &tentry);
2621 tsize = tentry->end - offset;
2624 object = tentry->object.vm_object;
2625 offset = tentry->offset + (offset - tentry->start);
2626 vm_map_unlock_read(smap);
2628 object = current->object.vm_object;
2630 vm_object_reference(object);
2631 last_timestamp = map->timestamp;
2632 vm_map_unlock_read(map);
2633 vm_object_sync(object, offset, size, syncio, invalidate);
2635 vm_object_deallocate(object);
2636 vm_map_lock_read(map);
2637 if (last_timestamp == map->timestamp ||
2638 !vm_map_lookup_entry(map, start, ¤t))
2639 current = current->next;
2642 vm_map_unlock_read(map);
2643 return (KERN_SUCCESS);
2647 * vm_map_entry_unwire: [ internal use only ]
2649 * Make the region specified by this entry pageable.
2651 * The map in question should be locked.
2652 * [This is the reason for this routine's existence.]
2655 vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry)
2657 vm_fault_unwire(map, entry->start, entry->end,
2658 entry->object.vm_object != NULL &&
2659 (entry->object.vm_object->type == OBJT_DEVICE ||
2660 entry->object.vm_object->type == OBJT_SG));
2661 entry->wired_count = 0;
2665 vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map)
2668 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0)
2669 vm_object_deallocate(entry->object.vm_object);
2670 uma_zfree(system_map ? kmapentzone : mapentzone, entry);
2674 * vm_map_entry_delete: [ internal use only ]
2676 * Deallocate the given entry from the target map.
2679 vm_map_entry_delete(vm_map_t map, vm_map_entry_t entry)
2682 vm_pindex_t offidxstart, offidxend, count, size1;
2685 vm_map_entry_unlink(map, entry);
2686 object = entry->object.vm_object;
2687 size = entry->end - entry->start;
2690 if (entry->uip != NULL) {
2691 swap_release_by_uid(size, entry->uip);
2695 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0 &&
2697 KASSERT(entry->uip == NULL || object->uip == NULL ||
2698 (entry->eflags & MAP_ENTRY_NEEDS_COPY),
2699 ("OVERCOMMIT vm_map_entry_delete: both uip %p", entry));
2700 count = OFF_TO_IDX(size);
2701 offidxstart = OFF_TO_IDX(entry->offset);
2702 offidxend = offidxstart + count;
2703 VM_OBJECT_LOCK(object);
2704 if (object->ref_count != 1 &&
2705 ((object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING ||
2706 object == kernel_object || object == kmem_object)) {
2707 vm_object_collapse(object);
2708 vm_object_page_remove(object, offidxstart, offidxend, FALSE);
2709 if (object->type == OBJT_SWAP)
2710 swap_pager_freespace(object, offidxstart, count);
2711 if (offidxend >= object->size &&
2712 offidxstart < object->size) {
2713 size1 = object->size;
2714 object->size = offidxstart;
2715 if (object->uip != NULL) {
2716 size1 -= object->size;
2717 KASSERT(object->charge >= ptoa(size1),
2718 ("vm_map_entry_delete: object->charge < 0"));
2719 swap_release_by_uid(ptoa(size1), object->uip);
2720 object->charge -= ptoa(size1);
2724 VM_OBJECT_UNLOCK(object);
2726 entry->object.vm_object = NULL;
2727 if (map->system_map)
2728 vm_map_entry_deallocate(entry, TRUE);
2730 entry->next = curthread->td_map_def_user;
2731 curthread->td_map_def_user = entry;
2736 * vm_map_delete: [ internal use only ]
2738 * Deallocates the given address range from the target
2742 vm_map_delete(vm_map_t map, vm_offset_t start, vm_offset_t end)
2744 vm_map_entry_t entry;
2745 vm_map_entry_t first_entry;
2747 VM_MAP_ASSERT_LOCKED(map);
2750 * Find the start of the region, and clip it
2752 if (!vm_map_lookup_entry(map, start, &first_entry))
2753 entry = first_entry->next;
2755 entry = first_entry;
2756 vm_map_clip_start(map, entry, start);
2760 * Step through all entries in this region
2762 while ((entry != &map->header) && (entry->start < end)) {
2763 vm_map_entry_t next;
2766 * Wait for wiring or unwiring of an entry to complete.
2767 * Also wait for any system wirings to disappear on
2770 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 ||
2771 (vm_map_pmap(map) != kernel_pmap &&
2772 vm_map_entry_system_wired_count(entry) != 0)) {
2773 unsigned int last_timestamp;
2774 vm_offset_t saved_start;
2775 vm_map_entry_t tmp_entry;
2777 saved_start = entry->start;
2778 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2779 last_timestamp = map->timestamp;
2780 (void) vm_map_unlock_and_wait(map, 0);
2782 if (last_timestamp + 1 != map->timestamp) {
2784 * Look again for the entry because the map was
2785 * modified while it was unlocked.
2786 * Specifically, the entry may have been
2787 * clipped, merged, or deleted.
2789 if (!vm_map_lookup_entry(map, saved_start,
2791 entry = tmp_entry->next;
2794 vm_map_clip_start(map, entry,
2800 vm_map_clip_end(map, entry, end);
2805 * Unwire before removing addresses from the pmap; otherwise,
2806 * unwiring will put the entries back in the pmap.
2808 if (entry->wired_count != 0) {
2809 vm_map_entry_unwire(map, entry);
2812 pmap_remove(map->pmap, entry->start, entry->end);
2815 * Delete the entry only after removing all pmap
2816 * entries pointing to its pages. (Otherwise, its
2817 * page frames may be reallocated, and any modify bits
2818 * will be set in the wrong object!)
2820 vm_map_entry_delete(map, entry);
2823 return (KERN_SUCCESS);
2829 * Remove the given address range from the target map.
2830 * This is the exported form of vm_map_delete.
2833 vm_map_remove(vm_map_t map, vm_offset_t start, vm_offset_t end)
2838 VM_MAP_RANGE_CHECK(map, start, end);
2839 result = vm_map_delete(map, start, end);
2845 * vm_map_check_protection:
2847 * Assert that the target map allows the specified privilege on the
2848 * entire address region given. The entire region must be allocated.
2850 * WARNING! This code does not and should not check whether the
2851 * contents of the region is accessible. For example a smaller file
2852 * might be mapped into a larger address space.
2854 * NOTE! This code is also called by munmap().
2856 * The map must be locked. A read lock is sufficient.
2859 vm_map_check_protection(vm_map_t map, vm_offset_t start, vm_offset_t end,
2860 vm_prot_t protection)
2862 vm_map_entry_t entry;
2863 vm_map_entry_t tmp_entry;
2865 if (!vm_map_lookup_entry(map, start, &tmp_entry))
2869 while (start < end) {
2870 if (entry == &map->header)
2875 if (start < entry->start)
2878 * Check protection associated with entry.
2880 if ((entry->protection & protection) != protection)
2882 /* go to next entry */
2884 entry = entry->next;
2890 * vm_map_copy_entry:
2892 * Copies the contents of the source entry to the destination
2893 * entry. The entries *must* be aligned properly.
2899 vm_map_entry_t src_entry,
2900 vm_map_entry_t dst_entry,
2901 vm_ooffset_t *fork_charge)
2903 vm_object_t src_object;
2905 struct uidinfo *uip;
2908 VM_MAP_ASSERT_LOCKED(dst_map);
2910 if ((dst_entry->eflags|src_entry->eflags) & MAP_ENTRY_IS_SUB_MAP)
2913 if (src_entry->wired_count == 0) {
2916 * If the source entry is marked needs_copy, it is already
2919 if ((src_entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) {
2920 pmap_protect(src_map->pmap,
2923 src_entry->protection & ~VM_PROT_WRITE);
2927 * Make a copy of the object.
2929 size = src_entry->end - src_entry->start;
2930 if ((src_object = src_entry->object.vm_object) != NULL) {
2931 VM_OBJECT_LOCK(src_object);
2932 charged = ENTRY_CHARGED(src_entry);
2933 if ((src_object->handle == NULL) &&
2934 (src_object->type == OBJT_DEFAULT ||
2935 src_object->type == OBJT_SWAP)) {
2936 vm_object_collapse(src_object);
2937 if ((src_object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING) {
2938 vm_object_split(src_entry);
2939 src_object = src_entry->object.vm_object;
2942 vm_object_reference_locked(src_object);
2943 vm_object_clear_flag(src_object, OBJ_ONEMAPPING);
2944 if (src_entry->uip != NULL &&
2945 !(src_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
2946 KASSERT(src_object->uip == NULL,
2947 ("OVERCOMMIT: vm_map_copy_entry: uip %p",
2949 src_object->uip = src_entry->uip;
2950 src_object->charge = size;
2952 VM_OBJECT_UNLOCK(src_object);
2953 dst_entry->object.vm_object = src_object;
2955 uip = curthread->td_ucred->cr_ruidinfo;
2957 dst_entry->uip = uip;
2958 *fork_charge += size;
2959 if (!(src_entry->eflags &
2960 MAP_ENTRY_NEEDS_COPY)) {
2962 src_entry->uip = uip;
2963 *fork_charge += size;
2966 src_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY);
2967 dst_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY);
2968 dst_entry->offset = src_entry->offset;
2970 dst_entry->object.vm_object = NULL;
2971 dst_entry->offset = 0;
2972 if (src_entry->uip != NULL) {
2973 dst_entry->uip = curthread->td_ucred->cr_ruidinfo;
2974 uihold(dst_entry->uip);
2975 *fork_charge += size;
2979 pmap_copy(dst_map->pmap, src_map->pmap, dst_entry->start,
2980 dst_entry->end - dst_entry->start, src_entry->start);
2983 * Of course, wired down pages can't be set copy-on-write.
2984 * Cause wired pages to be copied into the new map by
2985 * simulating faults (the new pages are pageable)
2987 vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry,
2993 * vmspace_map_entry_forked:
2994 * Update the newly-forked vmspace each time a map entry is inherited
2995 * or copied. The values for vm_dsize and vm_tsize are approximate
2996 * (and mostly-obsolete ideas in the face of mmap(2) et al.)
2999 vmspace_map_entry_forked(const struct vmspace *vm1, struct vmspace *vm2,
3000 vm_map_entry_t entry)
3002 vm_size_t entrysize;
3005 entrysize = entry->end - entry->start;
3006 vm2->vm_map.size += entrysize;
3007 if (entry->eflags & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP)) {
3008 vm2->vm_ssize += btoc(entrysize);
3009 } else if (entry->start >= (vm_offset_t)vm1->vm_daddr &&
3010 entry->start < (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)) {
3011 newend = MIN(entry->end,
3012 (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize));
3013 vm2->vm_dsize += btoc(newend - entry->start);
3014 } else if (entry->start >= (vm_offset_t)vm1->vm_taddr &&
3015 entry->start < (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)) {
3016 newend = MIN(entry->end,
3017 (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize));
3018 vm2->vm_tsize += btoc(newend - entry->start);
3024 * Create a new process vmspace structure and vm_map
3025 * based on those of an existing process. The new map
3026 * is based on the old map, according to the inheritance
3027 * values on the regions in that map.
3029 * XXX It might be worth coalescing the entries added to the new vmspace.
3031 * The source map must not be locked.
3034 vmspace_fork(struct vmspace *vm1, vm_ooffset_t *fork_charge)
3036 struct vmspace *vm2;
3037 vm_map_t old_map = &vm1->vm_map;
3039 vm_map_entry_t old_entry;
3040 vm_map_entry_t new_entry;
3044 vm_map_lock(old_map);
3046 vm_map_wait_busy(old_map);
3047 vm2 = vmspace_alloc(old_map->min_offset, old_map->max_offset);
3049 goto unlock_and_return;
3050 vm2->vm_taddr = vm1->vm_taddr;
3051 vm2->vm_daddr = vm1->vm_daddr;
3052 vm2->vm_maxsaddr = vm1->vm_maxsaddr;
3053 new_map = &vm2->vm_map; /* XXX */
3054 locked = vm_map_trylock(new_map); /* trylock to silence WITNESS */
3055 KASSERT(locked, ("vmspace_fork: lock failed"));
3056 new_map->timestamp = 1;
3058 old_entry = old_map->header.next;
3060 while (old_entry != &old_map->header) {
3061 if (old_entry->eflags & MAP_ENTRY_IS_SUB_MAP)
3062 panic("vm_map_fork: encountered a submap");
3064 switch (old_entry->inheritance) {
3065 case VM_INHERIT_NONE:
3068 case VM_INHERIT_SHARE:
3070 * Clone the entry, creating the shared object if necessary.
3072 object = old_entry->object.vm_object;
3073 if (object == NULL) {
3074 object = vm_object_allocate(OBJT_DEFAULT,
3075 atop(old_entry->end - old_entry->start));
3076 old_entry->object.vm_object = object;
3077 old_entry->offset = 0;
3078 if (old_entry->uip != NULL) {
3079 object->uip = old_entry->uip;
3080 object->charge = old_entry->end -
3082 old_entry->uip = NULL;
3087 * Add the reference before calling vm_object_shadow
3088 * to insure that a shadow object is created.
3090 vm_object_reference(object);
3091 if (old_entry->eflags & MAP_ENTRY_NEEDS_COPY) {
3092 vm_object_shadow(&old_entry->object.vm_object,
3094 atop(old_entry->end - old_entry->start));
3095 old_entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
3096 /* Transfer the second reference too. */
3097 vm_object_reference(
3098 old_entry->object.vm_object);
3101 * As in vm_map_simplify_entry(), the
3102 * vnode lock will not be acquired in
3103 * this call to vm_object_deallocate().
3105 vm_object_deallocate(object);
3106 object = old_entry->object.vm_object;
3108 VM_OBJECT_LOCK(object);
3109 vm_object_clear_flag(object, OBJ_ONEMAPPING);
3110 if (old_entry->uip != NULL) {
3111 KASSERT(object->uip == NULL, ("vmspace_fork both uip"));
3112 object->uip = old_entry->uip;
3113 object->charge = old_entry->end - old_entry->start;
3114 old_entry->uip = NULL;
3116 VM_OBJECT_UNLOCK(object);
3119 * Clone the entry, referencing the shared object.
3121 new_entry = vm_map_entry_create(new_map);
3122 *new_entry = *old_entry;
3123 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED |
3124 MAP_ENTRY_IN_TRANSITION);
3125 new_entry->wired_count = 0;
3128 * Insert the entry into the new map -- we know we're
3129 * inserting at the end of the new map.
3131 vm_map_entry_link(new_map, new_map->header.prev,
3133 vmspace_map_entry_forked(vm1, vm2, new_entry);
3136 * Update the physical map
3138 pmap_copy(new_map->pmap, old_map->pmap,
3140 (old_entry->end - old_entry->start),
3144 case VM_INHERIT_COPY:
3146 * Clone the entry and link into the map.
3148 new_entry = vm_map_entry_create(new_map);
3149 *new_entry = *old_entry;
3150 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED |
3151 MAP_ENTRY_IN_TRANSITION);
3152 new_entry->wired_count = 0;
3153 new_entry->object.vm_object = NULL;
3154 new_entry->uip = NULL;
3155 vm_map_entry_link(new_map, new_map->header.prev,
3157 vmspace_map_entry_forked(vm1, vm2, new_entry);
3158 vm_map_copy_entry(old_map, new_map, old_entry,
3159 new_entry, fork_charge);
3162 old_entry = old_entry->next;
3165 vm_map_unlock(old_map);
3167 vm_map_unlock(new_map);
3173 vm_map_stack(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
3174 vm_prot_t prot, vm_prot_t max, int cow)
3176 vm_map_entry_t new_entry, prev_entry;
3177 vm_offset_t bot, top;
3178 vm_size_t init_ssize;
3183 * The stack orientation is piggybacked with the cow argument.
3184 * Extract it into orient and mask the cow argument so that we
3185 * don't pass it around further.
3186 * NOTE: We explicitly allow bi-directional stacks.
3188 orient = cow & (MAP_STACK_GROWS_DOWN|MAP_STACK_GROWS_UP);
3190 KASSERT(orient != 0, ("No stack grow direction"));
3192 if (addrbos < vm_map_min(map) ||
3193 addrbos > vm_map_max(map) ||
3194 addrbos + max_ssize < addrbos)
3195 return (KERN_NO_SPACE);
3197 init_ssize = (max_ssize < sgrowsiz) ? max_ssize : sgrowsiz;
3199 PROC_LOCK(curthread->td_proc);
3200 vmemlim = lim_cur(curthread->td_proc, RLIMIT_VMEM);
3201 PROC_UNLOCK(curthread->td_proc);
3205 /* If addr is already mapped, no go */
3206 if (vm_map_lookup_entry(map, addrbos, &prev_entry)) {
3208 return (KERN_NO_SPACE);
3211 /* If we would blow our VMEM resource limit, no go */
3212 if (map->size + init_ssize > vmemlim) {
3214 return (KERN_NO_SPACE);
3218 * If we can't accomodate max_ssize in the current mapping, no go.
3219 * However, we need to be aware that subsequent user mappings might
3220 * map into the space we have reserved for stack, and currently this
3221 * space is not protected.
3223 * Hopefully we will at least detect this condition when we try to
3226 if ((prev_entry->next != &map->header) &&
3227 (prev_entry->next->start < addrbos + max_ssize)) {
3229 return (KERN_NO_SPACE);
3233 * We initially map a stack of only init_ssize. We will grow as
3234 * needed later. Depending on the orientation of the stack (i.e.
3235 * the grow direction) we either map at the top of the range, the
3236 * bottom of the range or in the middle.
3238 * Note: we would normally expect prot and max to be VM_PROT_ALL,
3239 * and cow to be 0. Possibly we should eliminate these as input
3240 * parameters, and just pass these values here in the insert call.
3242 if (orient == MAP_STACK_GROWS_DOWN)
3243 bot = addrbos + max_ssize - init_ssize;
3244 else if (orient == MAP_STACK_GROWS_UP)
3247 bot = round_page(addrbos + max_ssize/2 - init_ssize/2);
3248 top = bot + init_ssize;
3249 rv = vm_map_insert(map, NULL, 0, bot, top, prot, max, cow);
3251 /* Now set the avail_ssize amount. */
3252 if (rv == KERN_SUCCESS) {
3253 if (prev_entry != &map->header)
3254 vm_map_clip_end(map, prev_entry, bot);
3255 new_entry = prev_entry->next;
3256 if (new_entry->end != top || new_entry->start != bot)
3257 panic("Bad entry start/end for new stack entry");
3259 new_entry->avail_ssize = max_ssize - init_ssize;
3260 if (orient & MAP_STACK_GROWS_DOWN)
3261 new_entry->eflags |= MAP_ENTRY_GROWS_DOWN;
3262 if (orient & MAP_STACK_GROWS_UP)
3263 new_entry->eflags |= MAP_ENTRY_GROWS_UP;
3270 static int stack_guard_page = 0;
3271 TUNABLE_INT("security.bsd.stack_guard_page", &stack_guard_page);
3272 SYSCTL_INT(_security_bsd, OID_AUTO, stack_guard_page, CTLFLAG_RW,
3273 &stack_guard_page, 0,
3274 "Insert stack guard page ahead of the growable segments.");
3276 /* Attempts to grow a vm stack entry. Returns KERN_SUCCESS if the
3277 * desired address is already mapped, or if we successfully grow
3278 * the stack. Also returns KERN_SUCCESS if addr is outside the
3279 * stack range (this is strange, but preserves compatibility with
3280 * the grow function in vm_machdep.c).
3283 vm_map_growstack(struct proc *p, vm_offset_t addr)
3285 vm_map_entry_t next_entry, prev_entry;
3286 vm_map_entry_t new_entry, stack_entry;
3287 struct vmspace *vm = p->p_vmspace;
3288 vm_map_t map = &vm->vm_map;
3290 size_t grow_amount, max_grow;
3291 rlim_t stacklim, vmemlim;
3292 int is_procstack, rv;
3293 struct uidinfo *uip;
3297 stacklim = lim_cur(p, RLIMIT_STACK);
3298 vmemlim = lim_cur(p, RLIMIT_VMEM);
3301 vm_map_lock_read(map);
3303 /* If addr is already in the entry range, no need to grow.*/
3304 if (vm_map_lookup_entry(map, addr, &prev_entry)) {
3305 vm_map_unlock_read(map);
3306 return (KERN_SUCCESS);
3309 next_entry = prev_entry->next;
3310 if (!(prev_entry->eflags & MAP_ENTRY_GROWS_UP)) {
3312 * This entry does not grow upwards. Since the address lies
3313 * beyond this entry, the next entry (if one exists) has to
3314 * be a downward growable entry. The entry list header is
3315 * never a growable entry, so it suffices to check the flags.
3317 if (!(next_entry->eflags & MAP_ENTRY_GROWS_DOWN)) {
3318 vm_map_unlock_read(map);
3319 return (KERN_SUCCESS);
3321 stack_entry = next_entry;
3324 * This entry grows upward. If the next entry does not at
3325 * least grow downwards, this is the entry we need to grow.
3326 * otherwise we have two possible choices and we have to
3329 if (next_entry->eflags & MAP_ENTRY_GROWS_DOWN) {
3331 * We have two choices; grow the entry closest to
3332 * the address to minimize the amount of growth.
3334 if (addr - prev_entry->end <= next_entry->start - addr)
3335 stack_entry = prev_entry;
3337 stack_entry = next_entry;
3339 stack_entry = prev_entry;
3342 if (stack_entry == next_entry) {
3343 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_DOWN, ("foo"));
3344 KASSERT(addr < stack_entry->start, ("foo"));
3345 end = (prev_entry != &map->header) ? prev_entry->end :
3346 stack_entry->start - stack_entry->avail_ssize;
3347 grow_amount = roundup(stack_entry->start - addr, PAGE_SIZE);
3348 max_grow = stack_entry->start - end;
3350 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_UP, ("foo"));
3351 KASSERT(addr >= stack_entry->end, ("foo"));
3352 end = (next_entry != &map->header) ? next_entry->start :
3353 stack_entry->end + stack_entry->avail_ssize;
3354 grow_amount = roundup(addr + 1 - stack_entry->end, PAGE_SIZE);
3355 max_grow = end - stack_entry->end;
3358 if (grow_amount > stack_entry->avail_ssize) {
3359 vm_map_unlock_read(map);
3360 return (KERN_NO_SPACE);
3364 * If there is no longer enough space between the entries nogo, and
3365 * adjust the available space. Note: this should only happen if the
3366 * user has mapped into the stack area after the stack was created,
3367 * and is probably an error.
3369 * This also effectively destroys any guard page the user might have
3370 * intended by limiting the stack size.
3372 if (grow_amount + (stack_guard_page ? PAGE_SIZE : 0) > max_grow) {
3373 if (vm_map_lock_upgrade(map))
3376 stack_entry->avail_ssize = max_grow;
3379 return (KERN_NO_SPACE);
3382 is_procstack = (addr >= (vm_offset_t)vm->vm_maxsaddr) ? 1 : 0;
3385 * If this is the main process stack, see if we're over the stack
3388 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) {
3389 vm_map_unlock_read(map);
3390 return (KERN_NO_SPACE);
3393 /* Round up the grow amount modulo SGROWSIZ */
3394 grow_amount = roundup (grow_amount, sgrowsiz);
3395 if (grow_amount > stack_entry->avail_ssize)
3396 grow_amount = stack_entry->avail_ssize;
3397 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) {
3398 grow_amount = trunc_page((vm_size_t)stacklim) -
3402 /* If we would blow our VMEM resource limit, no go */
3403 if (map->size + grow_amount > vmemlim) {
3404 vm_map_unlock_read(map);
3405 return (KERN_NO_SPACE);
3408 if (vm_map_lock_upgrade(map))
3411 if (stack_entry == next_entry) {
3415 /* Get the preliminary new entry start value */
3416 addr = stack_entry->start - grow_amount;
3419 * If this puts us into the previous entry, cut back our
3420 * growth to the available space. Also, see the note above.
3423 stack_entry->avail_ssize = max_grow;
3425 if (stack_guard_page)
3429 rv = vm_map_insert(map, NULL, 0, addr, stack_entry->start,
3430 p->p_sysent->sv_stackprot, VM_PROT_ALL, 0);
3432 /* Adjust the available stack space by the amount we grew. */
3433 if (rv == KERN_SUCCESS) {
3434 if (prev_entry != &map->header)
3435 vm_map_clip_end(map, prev_entry, addr);
3436 new_entry = prev_entry->next;
3437 KASSERT(new_entry == stack_entry->prev, ("foo"));
3438 KASSERT(new_entry->end == stack_entry->start, ("foo"));
3439 KASSERT(new_entry->start == addr, ("foo"));
3440 grow_amount = new_entry->end - new_entry->start;
3441 new_entry->avail_ssize = stack_entry->avail_ssize -
3443 stack_entry->eflags &= ~MAP_ENTRY_GROWS_DOWN;
3444 new_entry->eflags |= MAP_ENTRY_GROWS_DOWN;
3450 addr = stack_entry->end + grow_amount;
3453 * If this puts us into the next entry, cut back our growth
3454 * to the available space. Also, see the note above.
3457 stack_entry->avail_ssize = end - stack_entry->end;
3459 if (stack_guard_page)
3463 grow_amount = addr - stack_entry->end;
3464 uip = stack_entry->uip;
3465 if (uip == NULL && stack_entry->object.vm_object != NULL)
3466 uip = stack_entry->object.vm_object->uip;
3467 if (uip != NULL && !swap_reserve_by_uid(grow_amount, uip))
3469 /* Grow the underlying object if applicable. */
3470 else if (stack_entry->object.vm_object == NULL ||
3471 vm_object_coalesce(stack_entry->object.vm_object,
3472 stack_entry->offset,
3473 (vm_size_t)(stack_entry->end - stack_entry->start),
3474 (vm_size_t)grow_amount, uip != NULL)) {
3475 map->size += (addr - stack_entry->end);
3476 /* Update the current entry. */
3477 stack_entry->end = addr;
3478 stack_entry->avail_ssize -= grow_amount;
3479 vm_map_entry_resize_free(map, stack_entry);
3482 if (next_entry != &map->header)
3483 vm_map_clip_start(map, next_entry, addr);
3488 if (rv == KERN_SUCCESS && is_procstack)
3489 vm->vm_ssize += btoc(grow_amount);
3494 * Heed the MAP_WIREFUTURE flag if it was set for this process.
3496 if (rv == KERN_SUCCESS && (map->flags & MAP_WIREFUTURE)) {
3498 (stack_entry == next_entry) ? addr : addr - grow_amount,
3499 (stack_entry == next_entry) ? stack_entry->start : addr,
3500 (p->p_flag & P_SYSTEM)
3501 ? VM_MAP_WIRE_SYSTEM|VM_MAP_WIRE_NOHOLES
3502 : VM_MAP_WIRE_USER|VM_MAP_WIRE_NOHOLES);
3509 * Unshare the specified VM space for exec. If other processes are
3510 * mapped to it, then create a new one. The new vmspace is null.
3513 vmspace_exec(struct proc *p, vm_offset_t minuser, vm_offset_t maxuser)
3515 struct vmspace *oldvmspace = p->p_vmspace;
3516 struct vmspace *newvmspace;
3518 newvmspace = vmspace_alloc(minuser, maxuser);
3519 if (newvmspace == NULL)
3521 newvmspace->vm_swrss = oldvmspace->vm_swrss;
3523 * This code is written like this for prototype purposes. The
3524 * goal is to avoid running down the vmspace here, but let the
3525 * other process's that are still using the vmspace to finally
3526 * run it down. Even though there is little or no chance of blocking
3527 * here, it is a good idea to keep this form for future mods.
3529 PROC_VMSPACE_LOCK(p);
3530 p->p_vmspace = newvmspace;
3531 PROC_VMSPACE_UNLOCK(p);
3532 if (p == curthread->td_proc)
3533 pmap_activate(curthread);
3534 vmspace_free(oldvmspace);
3539 * Unshare the specified VM space for forcing COW. This
3540 * is called by rfork, for the (RFMEM|RFPROC) == 0 case.
3543 vmspace_unshare(struct proc *p)
3545 struct vmspace *oldvmspace = p->p_vmspace;
3546 struct vmspace *newvmspace;
3547 vm_ooffset_t fork_charge;
3549 if (oldvmspace->vm_refcnt == 1)
3552 newvmspace = vmspace_fork(oldvmspace, &fork_charge);
3553 if (newvmspace == NULL)
3555 if (!swap_reserve_by_uid(fork_charge, p->p_ucred->cr_ruidinfo)) {
3556 vmspace_free(newvmspace);
3559 PROC_VMSPACE_LOCK(p);
3560 p->p_vmspace = newvmspace;
3561 PROC_VMSPACE_UNLOCK(p);
3562 if (p == curthread->td_proc)
3563 pmap_activate(curthread);
3564 vmspace_free(oldvmspace);
3571 * Finds the VM object, offset, and
3572 * protection for a given virtual address in the
3573 * specified map, assuming a page fault of the
3576 * Leaves the map in question locked for read; return
3577 * values are guaranteed until a vm_map_lookup_done
3578 * call is performed. Note that the map argument
3579 * is in/out; the returned map must be used in
3580 * the call to vm_map_lookup_done.
3582 * A handle (out_entry) is returned for use in
3583 * vm_map_lookup_done, to make that fast.
3585 * If a lookup is requested with "write protection"
3586 * specified, the map may be changed to perform virtual
3587 * copying operations, although the data referenced will
3591 vm_map_lookup(vm_map_t *var_map, /* IN/OUT */
3593 vm_prot_t fault_typea,
3594 vm_map_entry_t *out_entry, /* OUT */
3595 vm_object_t *object, /* OUT */
3596 vm_pindex_t *pindex, /* OUT */
3597 vm_prot_t *out_prot, /* OUT */
3598 boolean_t *wired) /* OUT */
3600 vm_map_entry_t entry;
3601 vm_map_t map = *var_map;
3603 vm_prot_t fault_type = fault_typea;
3604 vm_object_t eobject;
3605 struct uidinfo *uip;
3610 vm_map_lock_read(map);
3613 * Lookup the faulting address.
3615 if (!vm_map_lookup_entry(map, vaddr, out_entry)) {
3616 vm_map_unlock_read(map);
3617 return (KERN_INVALID_ADDRESS);
3625 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
3626 vm_map_t old_map = map;
3628 *var_map = map = entry->object.sub_map;
3629 vm_map_unlock_read(old_map);
3634 * Check whether this task is allowed to have this page.
3635 * Note the special case for MAP_ENTRY_COW
3636 * pages with an override. This is to implement a forced
3637 * COW for debuggers.
3639 if (fault_type & VM_PROT_OVERRIDE_WRITE)
3640 prot = entry->max_protection;
3642 prot = entry->protection;
3643 fault_type &= (VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE);
3644 if ((fault_type & prot) != fault_type) {
3645 vm_map_unlock_read(map);
3646 return (KERN_PROTECTION_FAILURE);
3648 if ((entry->eflags & MAP_ENTRY_USER_WIRED) &&
3649 (entry->eflags & MAP_ENTRY_COW) &&
3650 (fault_type & VM_PROT_WRITE) &&
3651 (fault_typea & VM_PROT_OVERRIDE_WRITE) == 0) {
3652 vm_map_unlock_read(map);
3653 return (KERN_PROTECTION_FAILURE);
3657 * If this page is not pageable, we have to get it for all possible
3660 *wired = (entry->wired_count != 0);
3662 prot = fault_type = entry->protection;
3663 size = entry->end - entry->start;
3665 * If the entry was copy-on-write, we either ...
3667 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
3669 * If we want to write the page, we may as well handle that
3670 * now since we've got the map locked.
3672 * If we don't need to write the page, we just demote the
3673 * permissions allowed.
3675 if (fault_type & VM_PROT_WRITE) {
3677 * Make a new object, and place it in the object
3678 * chain. Note that no new references have appeared
3679 * -- one just moved from the map to the new
3682 if (vm_map_lock_upgrade(map))
3685 if (entry->uip == NULL) {
3687 * The debugger owner is charged for
3690 uip = curthread->td_ucred->cr_ruidinfo;
3692 if (!swap_reserve_by_uid(size, uip)) {
3695 return (KERN_RESOURCE_SHORTAGE);
3700 &entry->object.vm_object,
3703 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
3704 eobject = entry->object.vm_object;
3705 if (eobject->uip != NULL) {
3707 * The object was not shadowed.
3709 swap_release_by_uid(size, entry->uip);
3712 } else if (entry->uip != NULL) {
3713 VM_OBJECT_LOCK(eobject);
3714 eobject->uip = entry->uip;
3715 eobject->charge = size;
3716 VM_OBJECT_UNLOCK(eobject);
3720 vm_map_lock_downgrade(map);
3723 * We're attempting to read a copy-on-write page --
3724 * don't allow writes.
3726 prot &= ~VM_PROT_WRITE;
3731 * Create an object if necessary.
3733 if (entry->object.vm_object == NULL &&
3735 if (vm_map_lock_upgrade(map))
3737 entry->object.vm_object = vm_object_allocate(OBJT_DEFAULT,
3740 if (entry->uip != NULL) {
3741 VM_OBJECT_LOCK(entry->object.vm_object);
3742 entry->object.vm_object->uip = entry->uip;
3743 entry->object.vm_object->charge = size;
3744 VM_OBJECT_UNLOCK(entry->object.vm_object);
3747 vm_map_lock_downgrade(map);
3751 * Return the object/offset from this entry. If the entry was
3752 * copy-on-write or empty, it has been fixed up.
3754 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
3755 *object = entry->object.vm_object;
3758 return (KERN_SUCCESS);
3762 * vm_map_lookup_locked:
3764 * Lookup the faulting address. A version of vm_map_lookup that returns
3765 * KERN_FAILURE instead of blocking on map lock or memory allocation.
3768 vm_map_lookup_locked(vm_map_t *var_map, /* IN/OUT */
3770 vm_prot_t fault_typea,
3771 vm_map_entry_t *out_entry, /* OUT */
3772 vm_object_t *object, /* OUT */
3773 vm_pindex_t *pindex, /* OUT */
3774 vm_prot_t *out_prot, /* OUT */
3775 boolean_t *wired) /* OUT */
3777 vm_map_entry_t entry;
3778 vm_map_t map = *var_map;
3780 vm_prot_t fault_type = fault_typea;
3783 * Lookup the faulting address.
3785 if (!vm_map_lookup_entry(map, vaddr, out_entry))
3786 return (KERN_INVALID_ADDRESS);
3791 * Fail if the entry refers to a submap.
3793 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP)
3794 return (KERN_FAILURE);
3797 * Check whether this task is allowed to have this page.
3798 * Note the special case for MAP_ENTRY_COW
3799 * pages with an override. This is to implement a forced
3800 * COW for debuggers.
3802 if (fault_type & VM_PROT_OVERRIDE_WRITE)
3803 prot = entry->max_protection;
3805 prot = entry->protection;
3806 fault_type &= VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE;
3807 if ((fault_type & prot) != fault_type)
3808 return (KERN_PROTECTION_FAILURE);
3809 if ((entry->eflags & MAP_ENTRY_USER_WIRED) &&
3810 (entry->eflags & MAP_ENTRY_COW) &&
3811 (fault_type & VM_PROT_WRITE) &&
3812 (fault_typea & VM_PROT_OVERRIDE_WRITE) == 0)
3813 return (KERN_PROTECTION_FAILURE);
3816 * If this page is not pageable, we have to get it for all possible
3819 *wired = (entry->wired_count != 0);
3821 prot = fault_type = entry->protection;
3823 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
3825 * Fail if the entry was copy-on-write for a write fault.
3827 if (fault_type & VM_PROT_WRITE)
3828 return (KERN_FAILURE);
3830 * We're attempting to read a copy-on-write page --
3831 * don't allow writes.
3833 prot &= ~VM_PROT_WRITE;
3837 * Fail if an object should be created.
3839 if (entry->object.vm_object == NULL && !map->system_map)
3840 return (KERN_FAILURE);
3843 * Return the object/offset from this entry. If the entry was
3844 * copy-on-write or empty, it has been fixed up.
3846 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
3847 *object = entry->object.vm_object;
3850 return (KERN_SUCCESS);
3854 * vm_map_lookup_done:
3856 * Releases locks acquired by a vm_map_lookup
3857 * (according to the handle returned by that lookup).
3860 vm_map_lookup_done(vm_map_t map, vm_map_entry_t entry)
3863 * Unlock the main-level map
3865 vm_map_unlock_read(map);
3868 #include "opt_ddb.h"
3870 #include <sys/kernel.h>
3872 #include <ddb/ddb.h>
3875 * vm_map_print: [ debug ]
3877 DB_SHOW_COMMAND(map, vm_map_print)
3880 /* XXX convert args. */
3881 vm_map_t map = (vm_map_t)addr;
3882 boolean_t full = have_addr;
3884 vm_map_entry_t entry;
3886 db_iprintf("Task map %p: pmap=%p, nentries=%d, version=%u\n",
3888 (void *)map->pmap, map->nentries, map->timestamp);
3891 if (!full && db_indent)
3895 for (entry = map->header.next; entry != &map->header;
3896 entry = entry->next) {
3897 db_iprintf("map entry %p: start=%p, end=%p\n",
3898 (void *)entry, (void *)entry->start, (void *)entry->end);
3901 static char *inheritance_name[4] =
3902 {"share", "copy", "none", "donate_copy"};
3904 db_iprintf(" prot=%x/%x/%s",
3906 entry->max_protection,
3907 inheritance_name[(int)(unsigned char)entry->inheritance]);
3908 if (entry->wired_count != 0)
3909 db_printf(", wired");
3911 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
3912 db_printf(", share=%p, offset=0x%jx\n",
3913 (void *)entry->object.sub_map,
3914 (uintmax_t)entry->offset);
3916 if ((entry->prev == &map->header) ||
3917 (entry->prev->object.sub_map !=
3918 entry->object.sub_map)) {
3920 vm_map_print((db_expr_t)(intptr_t)
3921 entry->object.sub_map,
3922 full, 0, (char *)0);
3926 if (entry->uip != NULL)
3927 db_printf(", uip %d", entry->uip->ui_uid);
3928 db_printf(", object=%p, offset=0x%jx",
3929 (void *)entry->object.vm_object,
3930 (uintmax_t)entry->offset);
3931 if (entry->object.vm_object && entry->object.vm_object->uip)
3932 db_printf(", obj uip %d charge %jx",
3933 entry->object.vm_object->uip->ui_uid,
3934 (uintmax_t)entry->object.vm_object->charge);
3935 if (entry->eflags & MAP_ENTRY_COW)
3936 db_printf(", copy (%s)",
3937 (entry->eflags & MAP_ENTRY_NEEDS_COPY) ? "needed" : "done");
3941 if ((entry->prev == &map->header) ||
3942 (entry->prev->object.vm_object !=
3943 entry->object.vm_object)) {
3945 vm_object_print((db_expr_t)(intptr_t)
3946 entry->object.vm_object,
3947 full, 0, (char *)0);
3959 DB_SHOW_COMMAND(procvm, procvm)
3964 p = (struct proc *) addr;
3969 db_printf("p = %p, vmspace = %p, map = %p, pmap = %p\n",
3970 (void *)p, (void *)p->p_vmspace, (void *)&p->p_vmspace->vm_map,
3971 (void *)vmspace_pmap(p->p_vmspace));
3973 vm_map_print((db_expr_t)(intptr_t)&p->p_vmspace->vm_map, 1, 0, NULL);