2 * Copyright (c) 1991, 1993
3 * The Regents of the University of California. All rights reserved.
5 * This code is derived from software contributed to Berkeley by
6 * The Mach Operating System project at Carnegie-Mellon University.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 4. Neither the name of the University nor the names of its contributors
17 * may be used to endorse or promote products derived from this software
18 * without specific prior written permission.
20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * from: @(#)vm_map.c 8.3 (Berkeley) 1/12/94
35 * Copyright (c) 1987, 1990 Carnegie-Mellon University.
36 * All rights reserved.
38 * Authors: Avadis Tevanian, Jr., Michael Wayne Young
40 * Permission to use, copy, modify and distribute this software and
41 * its documentation is hereby granted, provided that both the copyright
42 * notice and this permission notice appear in all copies of the
43 * software, derivative works or modified versions, and any portions
44 * thereof, and that both notices appear in supporting documentation.
46 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
47 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
48 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
50 * Carnegie Mellon requests users of this software to return to
52 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
53 * School of Computer Science
54 * Carnegie Mellon University
55 * Pittsburgh PA 15213-3890
57 * any improvements or extensions that they make and grant Carnegie the
58 * rights to redistribute these changes.
62 * Virtual memory mapping module.
65 #include <sys/cdefs.h>
66 __FBSDID("$FreeBSD$");
68 #include <sys/param.h>
69 #include <sys/systm.h>
70 #include <sys/kernel.h>
73 #include <sys/mutex.h>
75 #include <sys/vmmeter.h>
77 #include <sys/vnode.h>
78 #include <sys/racct.h>
79 #include <sys/resourcevar.h>
81 #include <sys/sysctl.h>
82 #include <sys/sysent.h>
86 #include <vm/vm_param.h>
88 #include <vm/vm_map.h>
89 #include <vm/vm_page.h>
90 #include <vm/vm_object.h>
91 #include <vm/vm_pager.h>
92 #include <vm/vm_kern.h>
93 #include <vm/vm_extern.h>
94 #include <vm/swap_pager.h>
98 * Virtual memory maps provide for the mapping, protection,
99 * and sharing of virtual memory objects. In addition,
100 * this module provides for an efficient virtual copy of
101 * memory from one map to another.
103 * Synchronization is required prior to most operations.
105 * Maps consist of an ordered doubly-linked list of simple
106 * entries; a self-adjusting binary search tree of these
107 * entries is used to speed up lookups.
109 * Since portions of maps are specified by start/end addresses,
110 * which may not align with existing map entries, all
111 * routines merely "clip" entries to these start/end values.
112 * [That is, an entry is split into two, bordering at a
113 * start or end value.] Note that these clippings may not
114 * always be necessary (as the two resulting entries are then
115 * not changed); however, the clipping is done for convenience.
117 * As mentioned above, virtual copy operations are performed
118 * by copying VM object references from one map to
119 * another, and then marking both regions as copy-on-write.
122 static struct mtx map_sleep_mtx;
123 static uma_zone_t mapentzone;
124 static uma_zone_t kmapentzone;
125 static uma_zone_t mapzone;
126 static uma_zone_t vmspace_zone;
127 static struct vm_object kmapentobj;
128 static int vmspace_zinit(void *mem, int size, int flags);
129 static void vmspace_zfini(void *mem, int size);
130 static int vm_map_zinit(void *mem, int ize, int flags);
131 static void vm_map_zfini(void *mem, int size);
132 static void _vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min,
134 static void vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map);
135 static void vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry);
137 static void vm_map_zdtor(void *mem, int size, void *arg);
138 static void vmspace_zdtor(void *mem, int size, void *arg);
141 #define ENTRY_CHARGED(e) ((e)->cred != NULL || \
142 ((e)->object.vm_object != NULL && (e)->object.vm_object->cred != NULL && \
143 !((e)->eflags & MAP_ENTRY_NEEDS_COPY)))
146 * PROC_VMSPACE_{UN,}LOCK() can be a noop as long as vmspaces are type
149 #define PROC_VMSPACE_LOCK(p) do { } while (0)
150 #define PROC_VMSPACE_UNLOCK(p) do { } while (0)
153 * VM_MAP_RANGE_CHECK: [ internal use only ]
155 * Asserts that the starting and ending region
156 * addresses fall within the valid range of the map.
158 #define VM_MAP_RANGE_CHECK(map, start, end) \
160 if (start < vm_map_min(map)) \
161 start = vm_map_min(map); \
162 if (end > vm_map_max(map)) \
163 end = vm_map_max(map); \
171 * Initialize the vm_map module. Must be called before
172 * any other vm_map routines.
174 * Map and entry structures are allocated from the general
175 * purpose memory pool with some exceptions:
177 * - The kernel map and kmem submap are allocated statically.
178 * - Kernel map entries are allocated out of a static pool.
180 * These restrictions are necessary since malloc() uses the
181 * maps and requires map entries.
187 mtx_init(&map_sleep_mtx, "vm map sleep mutex", NULL, MTX_DEF);
188 mapzone = uma_zcreate("MAP", sizeof(struct vm_map), NULL,
194 vm_map_zinit, vm_map_zfini, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
195 uma_prealloc(mapzone, MAX_KMAP);
196 kmapentzone = uma_zcreate("KMAP ENTRY", sizeof(struct vm_map_entry),
197 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR,
198 UMA_ZONE_MTXCLASS | UMA_ZONE_VM);
199 uma_prealloc(kmapentzone, MAX_KMAPENT);
200 mapentzone = uma_zcreate("MAP ENTRY", sizeof(struct vm_map_entry),
201 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
205 vmspace_zfini(void *mem, int size)
209 vm = (struct vmspace *)mem;
210 vm_map_zfini(&vm->vm_map, sizeof(vm->vm_map));
214 vmspace_zinit(void *mem, int size, int flags)
218 vm = (struct vmspace *)mem;
220 vm->vm_map.pmap = NULL;
221 (void)vm_map_zinit(&vm->vm_map, sizeof(vm->vm_map), flags);
226 vm_map_zfini(void *mem, int size)
231 mtx_destroy(&map->system_mtx);
232 sx_destroy(&map->lock);
236 vm_map_zinit(void *mem, int size, int flags)
243 mtx_init(&map->system_mtx, "system map", NULL, MTX_DEF | MTX_DUPOK);
244 sx_init(&map->lock, "user map");
250 vmspace_zdtor(void *mem, int size, void *arg)
254 vm = (struct vmspace *)mem;
256 vm_map_zdtor(&vm->vm_map, sizeof(vm->vm_map), arg);
259 vm_map_zdtor(void *mem, int size, void *arg)
264 KASSERT(map->nentries == 0,
265 ("map %p nentries == %d on free.",
266 map, map->nentries));
267 KASSERT(map->size == 0,
268 ("map %p size == %lu on free.",
269 map, (unsigned long)map->size));
271 #endif /* INVARIANTS */
274 * Allocate a vmspace structure, including a vm_map and pmap,
275 * and initialize those structures. The refcnt is set to 1.
278 vmspace_alloc(min, max)
279 vm_offset_t min, max;
283 vm = uma_zalloc(vmspace_zone, M_WAITOK);
284 if (vm->vm_map.pmap == NULL && !pmap_pinit(vmspace_pmap(vm))) {
285 uma_zfree(vmspace_zone, vm);
288 CTR1(KTR_VM, "vmspace_alloc: %p", vm);
289 _vm_map_init(&vm->vm_map, vmspace_pmap(vm), min, max);
305 uma_zone_set_obj(kmapentzone, &kmapentobj, lmin(cnt.v_page_count,
306 (VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS) / PAGE_SIZE) / 8 +
307 maxproc * 2 + maxfiles);
308 vmspace_zone = uma_zcreate("VMSPACE", sizeof(struct vmspace), NULL,
314 vmspace_zinit, vmspace_zfini, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
318 vmspace_container_reset(struct proc *p)
322 racct_set(p, RACCT_DATA, 0);
323 racct_set(p, RACCT_STACK, 0);
324 racct_set(p, RACCT_RSS, 0);
325 racct_set(p, RACCT_MEMLOCK, 0);
326 racct_set(p, RACCT_VMEM, 0);
331 vmspace_dofree(struct vmspace *vm)
334 CTR1(KTR_VM, "vmspace_free: %p", vm);
337 * Make sure any SysV shm is freed, it might not have been in
343 * Lock the map, to wait out all other references to it.
344 * Delete all of the mappings and pages they hold, then call
345 * the pmap module to reclaim anything left.
347 (void)vm_map_remove(&vm->vm_map, vm->vm_map.min_offset,
348 vm->vm_map.max_offset);
350 pmap_release(vmspace_pmap(vm));
351 vm->vm_map.pmap = NULL;
352 uma_zfree(vmspace_zone, vm);
356 vmspace_free(struct vmspace *vm)
359 if (vm->vm_refcnt == 0)
360 panic("vmspace_free: attempt to free already freed vmspace");
362 if (atomic_fetchadd_int(&vm->vm_refcnt, -1) == 1)
367 vmspace_exitfree(struct proc *p)
371 PROC_VMSPACE_LOCK(p);
374 PROC_VMSPACE_UNLOCK(p);
375 KASSERT(vm == &vmspace0, ("vmspace_exitfree: wrong vmspace"));
380 vmspace_exit(struct thread *td)
387 * Release user portion of address space.
388 * This releases references to vnodes,
389 * which could cause I/O if the file has been unlinked.
390 * Need to do this early enough that we can still sleep.
392 * The last exiting process to reach this point releases as
393 * much of the environment as it can. vmspace_dofree() is the
394 * slower fallback in case another process had a temporary
395 * reference to the vmspace.
400 atomic_add_int(&vmspace0.vm_refcnt, 1);
402 refcnt = vm->vm_refcnt;
403 if (refcnt > 1 && p->p_vmspace != &vmspace0) {
404 /* Switch now since other proc might free vmspace */
405 PROC_VMSPACE_LOCK(p);
406 p->p_vmspace = &vmspace0;
407 PROC_VMSPACE_UNLOCK(p);
410 } while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt - 1));
412 if (p->p_vmspace != vm) {
413 /* vmspace not yet freed, switch back */
414 PROC_VMSPACE_LOCK(p);
416 PROC_VMSPACE_UNLOCK(p);
419 pmap_remove_pages(vmspace_pmap(vm));
420 /* Switch now since this proc will free vmspace */
421 PROC_VMSPACE_LOCK(p);
422 p->p_vmspace = &vmspace0;
423 PROC_VMSPACE_UNLOCK(p);
427 vmspace_container_reset(p);
430 /* Acquire reference to vmspace owned by another process. */
433 vmspace_acquire_ref(struct proc *p)
438 PROC_VMSPACE_LOCK(p);
441 PROC_VMSPACE_UNLOCK(p);
445 refcnt = vm->vm_refcnt;
446 if (refcnt <= 0) { /* Avoid 0->1 transition */
447 PROC_VMSPACE_UNLOCK(p);
450 } while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt + 1));
451 if (vm != p->p_vmspace) {
452 PROC_VMSPACE_UNLOCK(p);
456 PROC_VMSPACE_UNLOCK(p);
461 _vm_map_lock(vm_map_t map, const char *file, int line)
465 _mtx_lock_flags(&map->system_mtx, 0, file, line);
467 (void)_sx_xlock(&map->lock, 0, file, line);
472 vm_map_process_deferred(void)
475 vm_map_entry_t entry;
479 while ((entry = td->td_map_def_user) != NULL) {
480 td->td_map_def_user = entry->next;
481 vm_map_entry_deallocate(entry, FALSE);
486 _vm_map_unlock(vm_map_t map, const char *file, int line)
490 _mtx_unlock_flags(&map->system_mtx, 0, file, line);
492 _sx_xunlock(&map->lock, file, line);
493 vm_map_process_deferred();
498 _vm_map_lock_read(vm_map_t map, const char *file, int line)
502 _mtx_lock_flags(&map->system_mtx, 0, file, line);
504 (void)_sx_slock(&map->lock, 0, file, line);
508 _vm_map_unlock_read(vm_map_t map, const char *file, int line)
512 _mtx_unlock_flags(&map->system_mtx, 0, file, line);
514 _sx_sunlock(&map->lock, file, line);
515 vm_map_process_deferred();
520 _vm_map_trylock(vm_map_t map, const char *file, int line)
524 error = map->system_map ?
525 !_mtx_trylock(&map->system_mtx, 0, file, line) :
526 !_sx_try_xlock(&map->lock, file, line);
533 _vm_map_trylock_read(vm_map_t map, const char *file, int line)
537 error = map->system_map ?
538 !_mtx_trylock(&map->system_mtx, 0, file, line) :
539 !_sx_try_slock(&map->lock, file, line);
544 * _vm_map_lock_upgrade: [ internal use only ]
546 * Tries to upgrade a read (shared) lock on the specified map to a write
547 * (exclusive) lock. Returns the value "0" if the upgrade succeeds and a
548 * non-zero value if the upgrade fails. If the upgrade fails, the map is
549 * returned without a read or write lock held.
551 * Requires that the map be read locked.
554 _vm_map_lock_upgrade(vm_map_t map, const char *file, int line)
556 unsigned int last_timestamp;
558 if (map->system_map) {
560 _mtx_assert(&map->system_mtx, MA_OWNED, file, line);
563 if (!_sx_try_upgrade(&map->lock, file, line)) {
564 last_timestamp = map->timestamp;
565 _sx_sunlock(&map->lock, file, line);
566 vm_map_process_deferred();
568 * If the map's timestamp does not change while the
569 * map is unlocked, then the upgrade succeeds.
571 (void)_sx_xlock(&map->lock, 0, file, line);
572 if (last_timestamp != map->timestamp) {
573 _sx_xunlock(&map->lock, file, line);
583 _vm_map_lock_downgrade(vm_map_t map, const char *file, int line)
586 if (map->system_map) {
588 _mtx_assert(&map->system_mtx, MA_OWNED, file, line);
591 _sx_downgrade(&map->lock, file, line);
597 * Returns a non-zero value if the caller holds a write (exclusive) lock
598 * on the specified map and the value "0" otherwise.
601 vm_map_locked(vm_map_t map)
605 return (mtx_owned(&map->system_mtx));
607 return (sx_xlocked(&map->lock));
612 _vm_map_assert_locked(vm_map_t map, const char *file, int line)
616 _mtx_assert(&map->system_mtx, MA_OWNED, file, line);
618 _sx_assert(&map->lock, SA_XLOCKED, file, line);
623 _vm_map_assert_locked_read(vm_map_t map, const char *file, int line)
627 _mtx_assert(&map->system_mtx, MA_OWNED, file, line);
629 _sx_assert(&map->lock, SA_SLOCKED, file, line);
633 #define VM_MAP_ASSERT_LOCKED(map) \
634 _vm_map_assert_locked(map, LOCK_FILE, LOCK_LINE)
635 #define VM_MAP_ASSERT_LOCKED_READ(map) \
636 _vm_map_assert_locked_read(map, LOCK_FILE, LOCK_LINE)
638 #define VM_MAP_ASSERT_LOCKED(map)
639 #define VM_MAP_ASSERT_LOCKED_READ(map)
643 * _vm_map_unlock_and_wait:
645 * Atomically releases the lock on the specified map and puts the calling
646 * thread to sleep. The calling thread will remain asleep until either
647 * vm_map_wakeup() is performed on the map or the specified timeout is
650 * WARNING! This function does not perform deferred deallocations of
651 * objects and map entries. Therefore, the calling thread is expected to
652 * reacquire the map lock after reawakening and later perform an ordinary
653 * unlock operation, such as vm_map_unlock(), before completing its
654 * operation on the map.
657 _vm_map_unlock_and_wait(vm_map_t map, int timo, const char *file, int line)
660 mtx_lock(&map_sleep_mtx);
662 _mtx_unlock_flags(&map->system_mtx, 0, file, line);
664 _sx_xunlock(&map->lock, file, line);
665 return (msleep(&map->root, &map_sleep_mtx, PDROP | PVM, "vmmaps",
672 * Awaken any threads that have slept on the map using
673 * vm_map_unlock_and_wait().
676 vm_map_wakeup(vm_map_t map)
680 * Acquire and release map_sleep_mtx to prevent a wakeup()
681 * from being performed (and lost) between the map unlock
682 * and the msleep() in _vm_map_unlock_and_wait().
684 mtx_lock(&map_sleep_mtx);
685 mtx_unlock(&map_sleep_mtx);
690 vm_map_busy(vm_map_t map)
693 VM_MAP_ASSERT_LOCKED(map);
698 vm_map_unbusy(vm_map_t map)
701 VM_MAP_ASSERT_LOCKED(map);
702 KASSERT(map->busy, ("vm_map_unbusy: not busy"));
703 if (--map->busy == 0 && (map->flags & MAP_BUSY_WAKEUP)) {
704 vm_map_modflags(map, 0, MAP_BUSY_WAKEUP);
710 vm_map_wait_busy(vm_map_t map)
713 VM_MAP_ASSERT_LOCKED(map);
715 vm_map_modflags(map, MAP_BUSY_WAKEUP, 0);
717 msleep(&map->busy, &map->system_mtx, 0, "mbusy", 0);
719 sx_sleep(&map->busy, &map->lock, 0, "mbusy", 0);
725 vmspace_resident_count(struct vmspace *vmspace)
727 return pmap_resident_count(vmspace_pmap(vmspace));
731 vmspace_wired_count(struct vmspace *vmspace)
733 return pmap_wired_count(vmspace_pmap(vmspace));
739 * Creates and returns a new empty VM map with
740 * the given physical map structure, and having
741 * the given lower and upper address bounds.
744 vm_map_create(pmap_t pmap, vm_offset_t min, vm_offset_t max)
748 result = uma_zalloc(mapzone, M_WAITOK);
749 CTR1(KTR_VM, "vm_map_create: %p", result);
750 _vm_map_init(result, pmap, min, max);
755 * Initialize an existing vm_map structure
756 * such as that in the vmspace structure.
759 _vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max)
762 map->header.next = map->header.prev = &map->header;
763 map->needs_wakeup = FALSE;
766 map->min_offset = min;
767 map->max_offset = max;
775 vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max)
778 _vm_map_init(map, pmap, min, max);
779 mtx_init(&map->system_mtx, "system map", NULL, MTX_DEF | MTX_DUPOK);
780 sx_init(&map->lock, "user map");
784 * vm_map_entry_dispose: [ internal use only ]
786 * Inverse of vm_map_entry_create.
789 vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry)
791 uma_zfree(map->system_map ? kmapentzone : mapentzone, entry);
795 * vm_map_entry_create: [ internal use only ]
797 * Allocates a VM map entry for insertion.
798 * No entry fields are filled in.
800 static vm_map_entry_t
801 vm_map_entry_create(vm_map_t map)
803 vm_map_entry_t new_entry;
806 new_entry = uma_zalloc(kmapentzone, M_NOWAIT);
808 new_entry = uma_zalloc(mapentzone, M_WAITOK);
809 if (new_entry == NULL)
810 panic("vm_map_entry_create: kernel resources exhausted");
815 * vm_map_entry_set_behavior:
817 * Set the expected access behavior, either normal, random, or
821 vm_map_entry_set_behavior(vm_map_entry_t entry, u_char behavior)
823 entry->eflags = (entry->eflags & ~MAP_ENTRY_BEHAV_MASK) |
824 (behavior & MAP_ENTRY_BEHAV_MASK);
828 * vm_map_entry_set_max_free:
830 * Set the max_free field in a vm_map_entry.
833 vm_map_entry_set_max_free(vm_map_entry_t entry)
836 entry->max_free = entry->adj_free;
837 if (entry->left != NULL && entry->left->max_free > entry->max_free)
838 entry->max_free = entry->left->max_free;
839 if (entry->right != NULL && entry->right->max_free > entry->max_free)
840 entry->max_free = entry->right->max_free;
844 * vm_map_entry_splay:
846 * The Sleator and Tarjan top-down splay algorithm with the
847 * following variation. Max_free must be computed bottom-up, so
848 * on the downward pass, maintain the left and right spines in
849 * reverse order. Then, make a second pass up each side to fix
850 * the pointers and compute max_free. The time bound is O(log n)
853 * The new root is the vm_map_entry containing "addr", or else an
854 * adjacent entry (lower or higher) if addr is not in the tree.
856 * The map must be locked, and leaves it so.
858 * Returns: the new root.
860 static vm_map_entry_t
861 vm_map_entry_splay(vm_offset_t addr, vm_map_entry_t root)
863 vm_map_entry_t llist, rlist;
864 vm_map_entry_t ltree, rtree;
867 /* Special case of empty tree. */
872 * Pass One: Splay down the tree until we find addr or a NULL
873 * pointer where addr would go. llist and rlist are the two
874 * sides in reverse order (bottom-up), with llist linked by
875 * the right pointer and rlist linked by the left pointer in
876 * the vm_map_entry. Wait until Pass Two to set max_free on
882 /* root is never NULL in here. */
883 if (addr < root->start) {
887 if (addr < y->start && y->left != NULL) {
888 /* Rotate right and put y on rlist. */
889 root->left = y->right;
891 vm_map_entry_set_max_free(root);
896 /* Put root on rlist. */
901 } else if (addr >= root->end) {
905 if (addr >= y->end && y->right != NULL) {
906 /* Rotate left and put y on llist. */
907 root->right = y->left;
909 vm_map_entry_set_max_free(root);
914 /* Put root on llist. */
924 * Pass Two: Walk back up the two spines, flip the pointers
925 * and set max_free. The subtrees of the root go at the
926 * bottom of llist and rlist.
929 while (llist != NULL) {
931 llist->right = ltree;
932 vm_map_entry_set_max_free(llist);
937 while (rlist != NULL) {
940 vm_map_entry_set_max_free(rlist);
946 * Final assembly: add ltree and rtree as subtrees of root.
950 vm_map_entry_set_max_free(root);
956 * vm_map_entry_{un,}link:
958 * Insert/remove entries from maps.
961 vm_map_entry_link(vm_map_t map,
962 vm_map_entry_t after_where,
963 vm_map_entry_t entry)
967 "vm_map_entry_link: map %p, nentries %d, entry %p, after %p", map,
968 map->nentries, entry, after_where);
969 VM_MAP_ASSERT_LOCKED(map);
971 entry->prev = after_where;
972 entry->next = after_where->next;
973 entry->next->prev = entry;
974 after_where->next = entry;
976 if (after_where != &map->header) {
977 if (after_where != map->root)
978 vm_map_entry_splay(after_where->start, map->root);
979 entry->right = after_where->right;
980 entry->left = after_where;
981 after_where->right = NULL;
982 after_where->adj_free = entry->start - after_where->end;
983 vm_map_entry_set_max_free(after_where);
985 entry->right = map->root;
988 entry->adj_free = (entry->next == &map->header ? map->max_offset :
989 entry->next->start) - entry->end;
990 vm_map_entry_set_max_free(entry);
995 vm_map_entry_unlink(vm_map_t map,
996 vm_map_entry_t entry)
998 vm_map_entry_t next, prev, root;
1000 VM_MAP_ASSERT_LOCKED(map);
1001 if (entry != map->root)
1002 vm_map_entry_splay(entry->start, map->root);
1003 if (entry->left == NULL)
1004 root = entry->right;
1006 root = vm_map_entry_splay(entry->start, entry->left);
1007 root->right = entry->right;
1008 root->adj_free = (entry->next == &map->header ? map->max_offset :
1009 entry->next->start) - root->end;
1010 vm_map_entry_set_max_free(root);
1019 CTR3(KTR_VM, "vm_map_entry_unlink: map %p, nentries %d, entry %p", map,
1020 map->nentries, entry);
1024 * vm_map_entry_resize_free:
1026 * Recompute the amount of free space following a vm_map_entry
1027 * and propagate that value up the tree. Call this function after
1028 * resizing a map entry in-place, that is, without a call to
1029 * vm_map_entry_link() or _unlink().
1031 * The map must be locked, and leaves it so.
1034 vm_map_entry_resize_free(vm_map_t map, vm_map_entry_t entry)
1038 * Using splay trees without parent pointers, propagating
1039 * max_free up the tree is done by moving the entry to the
1040 * root and making the change there.
1042 if (entry != map->root)
1043 map->root = vm_map_entry_splay(entry->start, map->root);
1045 entry->adj_free = (entry->next == &map->header ? map->max_offset :
1046 entry->next->start) - entry->end;
1047 vm_map_entry_set_max_free(entry);
1051 * vm_map_lookup_entry: [ internal use only ]
1053 * Finds the map entry containing (or
1054 * immediately preceding) the specified address
1055 * in the given map; the entry is returned
1056 * in the "entry" parameter. The boolean
1057 * result indicates whether the address is
1058 * actually contained in the map.
1061 vm_map_lookup_entry(
1063 vm_offset_t address,
1064 vm_map_entry_t *entry) /* OUT */
1070 * If the map is empty, then the map entry immediately preceding
1071 * "address" is the map's header.
1075 *entry = &map->header;
1076 else if (address >= cur->start && cur->end > address) {
1079 } else if ((locked = vm_map_locked(map)) ||
1080 sx_try_upgrade(&map->lock)) {
1082 * Splay requires a write lock on the map. However, it only
1083 * restructures the binary search tree; it does not otherwise
1084 * change the map. Thus, the map's timestamp need not change
1085 * on a temporary upgrade.
1087 map->root = cur = vm_map_entry_splay(address, cur);
1089 sx_downgrade(&map->lock);
1092 * If "address" is contained within a map entry, the new root
1093 * is that map entry. Otherwise, the new root is a map entry
1094 * immediately before or after "address".
1096 if (address >= cur->start) {
1098 if (cur->end > address)
1104 * Since the map is only locked for read access, perform a
1105 * standard binary search tree lookup for "address".
1108 if (address < cur->start) {
1109 if (cur->left == NULL) {
1114 } else if (cur->end > address) {
1118 if (cur->right == NULL) {
1131 * Inserts the given whole VM object into the target
1132 * map at the specified address range. The object's
1133 * size should match that of the address range.
1135 * Requires that the map be locked, and leaves it so.
1137 * If object is non-NULL, ref count must be bumped by caller
1138 * prior to making call to account for the new entry.
1141 vm_map_insert(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1142 vm_offset_t start, vm_offset_t end, vm_prot_t prot, vm_prot_t max,
1145 vm_map_entry_t new_entry;
1146 vm_map_entry_t prev_entry;
1147 vm_map_entry_t temp_entry;
1148 vm_eflags_t protoeflags;
1150 boolean_t charge_prev_obj;
1152 VM_MAP_ASSERT_LOCKED(map);
1155 * Check that the start and end points are not bogus.
1157 if ((start < map->min_offset) || (end > map->max_offset) ||
1159 return (KERN_INVALID_ADDRESS);
1162 * Find the entry prior to the proposed starting address; if it's part
1163 * of an existing entry, this range is bogus.
1165 if (vm_map_lookup_entry(map, start, &temp_entry))
1166 return (KERN_NO_SPACE);
1168 prev_entry = temp_entry;
1171 * Assert that the next entry doesn't overlap the end point.
1173 if ((prev_entry->next != &map->header) &&
1174 (prev_entry->next->start < end))
1175 return (KERN_NO_SPACE);
1178 charge_prev_obj = FALSE;
1180 if (cow & MAP_COPY_ON_WRITE)
1181 protoeflags |= MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY;
1183 if (cow & MAP_NOFAULT) {
1184 protoeflags |= MAP_ENTRY_NOFAULT;
1186 KASSERT(object == NULL,
1187 ("vm_map_insert: paradoxical MAP_NOFAULT request"));
1189 if (cow & MAP_DISABLE_SYNCER)
1190 protoeflags |= MAP_ENTRY_NOSYNC;
1191 if (cow & MAP_DISABLE_COREDUMP)
1192 protoeflags |= MAP_ENTRY_NOCOREDUMP;
1195 KASSERT((object != kmem_object && object != kernel_object) ||
1196 ((object == kmem_object || object == kernel_object) &&
1197 !(protoeflags & MAP_ENTRY_NEEDS_COPY)),
1198 ("kmem or kernel object and cow"));
1199 if (cow & (MAP_ACC_NO_CHARGE | MAP_NOFAULT))
1201 if ((cow & MAP_ACC_CHARGED) || ((prot & VM_PROT_WRITE) &&
1202 ((protoeflags & MAP_ENTRY_NEEDS_COPY) || object == NULL))) {
1203 if (!(cow & MAP_ACC_CHARGED) && !swap_reserve(end - start))
1204 return (KERN_RESOURCE_SHORTAGE);
1205 KASSERT(object == NULL || (protoeflags & MAP_ENTRY_NEEDS_COPY) ||
1206 object->cred == NULL,
1207 ("OVERCOMMIT: vm_map_insert o %p", object));
1208 cred = curthread->td_ucred;
1210 if (object == NULL && !(protoeflags & MAP_ENTRY_NEEDS_COPY))
1211 charge_prev_obj = TRUE;
1215 /* Expand the kernel pmap, if necessary. */
1216 if (map == kernel_map && end > kernel_vm_end)
1217 pmap_growkernel(end);
1218 if (object != NULL) {
1220 * OBJ_ONEMAPPING must be cleared unless this mapping
1221 * is trivially proven to be the only mapping for any
1222 * of the object's pages. (Object granularity
1223 * reference counting is insufficient to recognize
1224 * aliases with precision.)
1226 VM_OBJECT_LOCK(object);
1227 if (object->ref_count > 1 || object->shadow_count != 0)
1228 vm_object_clear_flag(object, OBJ_ONEMAPPING);
1229 VM_OBJECT_UNLOCK(object);
1231 else if ((prev_entry != &map->header) &&
1232 (prev_entry->eflags == protoeflags) &&
1233 (prev_entry->end == start) &&
1234 (prev_entry->wired_count == 0) &&
1235 (prev_entry->cred == cred ||
1236 (prev_entry->object.vm_object != NULL &&
1237 (prev_entry->object.vm_object->cred == cred))) &&
1238 vm_object_coalesce(prev_entry->object.vm_object,
1240 (vm_size_t)(prev_entry->end - prev_entry->start),
1241 (vm_size_t)(end - prev_entry->end), charge_prev_obj)) {
1243 * We were able to extend the object. Determine if we
1244 * can extend the previous map entry to include the
1245 * new range as well.
1247 if ((prev_entry->inheritance == VM_INHERIT_DEFAULT) &&
1248 (prev_entry->protection == prot) &&
1249 (prev_entry->max_protection == max)) {
1250 map->size += (end - prev_entry->end);
1251 prev_entry->end = end;
1252 vm_map_entry_resize_free(map, prev_entry);
1253 vm_map_simplify_entry(map, prev_entry);
1256 return (KERN_SUCCESS);
1260 * If we can extend the object but cannot extend the
1261 * map entry, we have to create a new map entry. We
1262 * must bump the ref count on the extended object to
1263 * account for it. object may be NULL.
1265 object = prev_entry->object.vm_object;
1266 offset = prev_entry->offset +
1267 (prev_entry->end - prev_entry->start);
1268 vm_object_reference(object);
1269 if (cred != NULL && object != NULL && object->cred != NULL &&
1270 !(prev_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
1271 /* Object already accounts for this uid. */
1278 * NOTE: if conditionals fail, object can be NULL here. This occurs
1279 * in things like the buffer map where we manage kva but do not manage
1284 * Create a new entry
1286 new_entry = vm_map_entry_create(map);
1287 new_entry->start = start;
1288 new_entry->end = end;
1289 new_entry->cred = NULL;
1291 new_entry->eflags = protoeflags;
1292 new_entry->object.vm_object = object;
1293 new_entry->offset = offset;
1294 new_entry->avail_ssize = 0;
1296 new_entry->inheritance = VM_INHERIT_DEFAULT;
1297 new_entry->protection = prot;
1298 new_entry->max_protection = max;
1299 new_entry->wired_count = 0;
1301 KASSERT(cred == NULL || !ENTRY_CHARGED(new_entry),
1302 ("OVERCOMMIT: vm_map_insert leaks vm_map %p", new_entry));
1303 new_entry->cred = cred;
1306 * Insert the new entry into the list
1308 vm_map_entry_link(map, prev_entry, new_entry);
1309 map->size += new_entry->end - new_entry->start;
1312 * It may be possible to merge the new entry with the next and/or
1313 * previous entries. However, due to MAP_STACK_* being a hack, a
1314 * panic can result from merging such entries.
1316 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0)
1317 vm_map_simplify_entry(map, new_entry);
1319 if (cow & (MAP_PREFAULT|MAP_PREFAULT_PARTIAL)) {
1320 vm_map_pmap_enter(map, start, prot,
1321 object, OFF_TO_IDX(offset), end - start,
1322 cow & MAP_PREFAULT_PARTIAL);
1325 return (KERN_SUCCESS);
1331 * Find the first fit (lowest VM address) for "length" free bytes
1332 * beginning at address >= start in the given map.
1334 * In a vm_map_entry, "adj_free" is the amount of free space
1335 * adjacent (higher address) to this entry, and "max_free" is the
1336 * maximum amount of contiguous free space in its subtree. This
1337 * allows finding a free region in one path down the tree, so
1338 * O(log n) amortized with splay trees.
1340 * The map must be locked, and leaves it so.
1342 * Returns: 0 on success, and starting address in *addr,
1343 * 1 if insufficient space.
1346 vm_map_findspace(vm_map_t map, vm_offset_t start, vm_size_t length,
1347 vm_offset_t *addr) /* OUT */
1349 vm_map_entry_t entry;
1353 * Request must fit within min/max VM address and must avoid
1356 if (start < map->min_offset)
1357 start = map->min_offset;
1358 if (start + length > map->max_offset || start + length < start)
1361 /* Empty tree means wide open address space. */
1362 if (map->root == NULL) {
1368 * After splay, if start comes before root node, then there
1369 * must be a gap from start to the root.
1371 map->root = vm_map_entry_splay(start, map->root);
1372 if (start + length <= map->root->start) {
1378 * Root is the last node that might begin its gap before
1379 * start, and this is the last comparison where address
1380 * wrap might be a problem.
1382 st = (start > map->root->end) ? start : map->root->end;
1383 if (length <= map->root->end + map->root->adj_free - st) {
1388 /* With max_free, can immediately tell if no solution. */
1389 entry = map->root->right;
1390 if (entry == NULL || length > entry->max_free)
1394 * Search the right subtree in the order: left subtree, root,
1395 * right subtree (first fit). The previous splay implies that
1396 * all regions in the right subtree have addresses > start.
1398 while (entry != NULL) {
1399 if (entry->left != NULL && entry->left->max_free >= length)
1400 entry = entry->left;
1401 else if (entry->adj_free >= length) {
1405 entry = entry->right;
1408 /* Can't get here, so panic if we do. */
1409 panic("vm_map_findspace: max_free corrupt");
1413 vm_map_fixed(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1414 vm_offset_t start, vm_size_t length, vm_prot_t prot,
1415 vm_prot_t max, int cow)
1420 end = start + length;
1422 VM_MAP_RANGE_CHECK(map, start, end);
1423 (void) vm_map_delete(map, start, end);
1424 result = vm_map_insert(map, object, offset, start, end, prot,
1431 * vm_map_find finds an unallocated region in the target address
1432 * map with the given length. The search is defined to be
1433 * first-fit from the specified address; the region found is
1434 * returned in the same parameter.
1436 * If object is non-NULL, ref count must be bumped by caller
1437 * prior to making call to account for the new entry.
1440 vm_map_find(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1441 vm_offset_t *addr, /* IN/OUT */
1442 vm_size_t length, int find_space, vm_prot_t prot,
1443 vm_prot_t max, int cow)
1451 if (find_space != VMFS_NO_SPACE) {
1452 if (vm_map_findspace(map, start, length, addr)) {
1454 return (KERN_NO_SPACE);
1456 switch (find_space) {
1457 case VMFS_ALIGNED_SPACE:
1458 pmap_align_superpage(object, offset, addr,
1461 #ifdef VMFS_TLB_ALIGNED_SPACE
1462 case VMFS_TLB_ALIGNED_SPACE:
1463 pmap_align_tlb(addr);
1472 result = vm_map_insert(map, object, offset, start, start +
1473 length, prot, max, cow);
1474 } while (result == KERN_NO_SPACE && (find_space == VMFS_ALIGNED_SPACE
1475 #ifdef VMFS_TLB_ALIGNED_SPACE
1476 || find_space == VMFS_TLB_ALIGNED_SPACE
1484 * vm_map_simplify_entry:
1486 * Simplify the given map entry by merging with either neighbor. This
1487 * routine also has the ability to merge with both neighbors.
1489 * The map must be locked.
1491 * This routine guarentees that the passed entry remains valid (though
1492 * possibly extended). When merging, this routine may delete one or
1496 vm_map_simplify_entry(vm_map_t map, vm_map_entry_t entry)
1498 vm_map_entry_t next, prev;
1499 vm_size_t prevsize, esize;
1501 if (entry->eflags & (MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_IS_SUB_MAP))
1505 if (prev != &map->header) {
1506 prevsize = prev->end - prev->start;
1507 if ( (prev->end == entry->start) &&
1508 (prev->object.vm_object == entry->object.vm_object) &&
1509 (!prev->object.vm_object ||
1510 (prev->offset + prevsize == entry->offset)) &&
1511 (prev->eflags == entry->eflags) &&
1512 (prev->protection == entry->protection) &&
1513 (prev->max_protection == entry->max_protection) &&
1514 (prev->inheritance == entry->inheritance) &&
1515 (prev->wired_count == entry->wired_count) &&
1516 (prev->cred == entry->cred)) {
1517 vm_map_entry_unlink(map, prev);
1518 entry->start = prev->start;
1519 entry->offset = prev->offset;
1520 if (entry->prev != &map->header)
1521 vm_map_entry_resize_free(map, entry->prev);
1524 * If the backing object is a vnode object,
1525 * vm_object_deallocate() calls vrele().
1526 * However, vrele() does not lock the vnode
1527 * because the vnode has additional
1528 * references. Thus, the map lock can be kept
1529 * without causing a lock-order reversal with
1532 if (prev->object.vm_object)
1533 vm_object_deallocate(prev->object.vm_object);
1534 if (prev->cred != NULL)
1536 vm_map_entry_dispose(map, prev);
1541 if (next != &map->header) {
1542 esize = entry->end - entry->start;
1543 if ((entry->end == next->start) &&
1544 (next->object.vm_object == entry->object.vm_object) &&
1545 (!entry->object.vm_object ||
1546 (entry->offset + esize == next->offset)) &&
1547 (next->eflags == entry->eflags) &&
1548 (next->protection == entry->protection) &&
1549 (next->max_protection == entry->max_protection) &&
1550 (next->inheritance == entry->inheritance) &&
1551 (next->wired_count == entry->wired_count) &&
1552 (next->cred == entry->cred)) {
1553 vm_map_entry_unlink(map, next);
1554 entry->end = next->end;
1555 vm_map_entry_resize_free(map, entry);
1558 * See comment above.
1560 if (next->object.vm_object)
1561 vm_object_deallocate(next->object.vm_object);
1562 if (next->cred != NULL)
1564 vm_map_entry_dispose(map, next);
1569 * vm_map_clip_start: [ internal use only ]
1571 * Asserts that the given entry begins at or after
1572 * the specified address; if necessary,
1573 * it splits the entry into two.
1575 #define vm_map_clip_start(map, entry, startaddr) \
1577 if (startaddr > entry->start) \
1578 _vm_map_clip_start(map, entry, startaddr); \
1582 * This routine is called only when it is known that
1583 * the entry must be split.
1586 _vm_map_clip_start(vm_map_t map, vm_map_entry_t entry, vm_offset_t start)
1588 vm_map_entry_t new_entry;
1590 VM_MAP_ASSERT_LOCKED(map);
1593 * Split off the front portion -- note that we must insert the new
1594 * entry BEFORE this one, so that this entry has the specified
1597 vm_map_simplify_entry(map, entry);
1600 * If there is no object backing this entry, we might as well create
1601 * one now. If we defer it, an object can get created after the map
1602 * is clipped, and individual objects will be created for the split-up
1603 * map. This is a bit of a hack, but is also about the best place to
1604 * put this improvement.
1606 if (entry->object.vm_object == NULL && !map->system_map) {
1608 object = vm_object_allocate(OBJT_DEFAULT,
1609 atop(entry->end - entry->start));
1610 entry->object.vm_object = object;
1612 if (entry->cred != NULL) {
1613 object->cred = entry->cred;
1614 object->charge = entry->end - entry->start;
1617 } else if (entry->object.vm_object != NULL &&
1618 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) &&
1619 entry->cred != NULL) {
1620 VM_OBJECT_LOCK(entry->object.vm_object);
1621 KASSERT(entry->object.vm_object->cred == NULL,
1622 ("OVERCOMMIT: vm_entry_clip_start: both cred e %p", entry));
1623 entry->object.vm_object->cred = entry->cred;
1624 entry->object.vm_object->charge = entry->end - entry->start;
1625 VM_OBJECT_UNLOCK(entry->object.vm_object);
1629 new_entry = vm_map_entry_create(map);
1630 *new_entry = *entry;
1632 new_entry->end = start;
1633 entry->offset += (start - entry->start);
1634 entry->start = start;
1635 if (new_entry->cred != NULL)
1636 crhold(entry->cred);
1638 vm_map_entry_link(map, entry->prev, new_entry);
1640 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1641 vm_object_reference(new_entry->object.vm_object);
1646 * vm_map_clip_end: [ internal use only ]
1648 * Asserts that the given entry ends at or before
1649 * the specified address; if necessary,
1650 * it splits the entry into two.
1652 #define vm_map_clip_end(map, entry, endaddr) \
1654 if ((endaddr) < (entry->end)) \
1655 _vm_map_clip_end((map), (entry), (endaddr)); \
1659 * This routine is called only when it is known that
1660 * the entry must be split.
1663 _vm_map_clip_end(vm_map_t map, vm_map_entry_t entry, vm_offset_t end)
1665 vm_map_entry_t new_entry;
1667 VM_MAP_ASSERT_LOCKED(map);
1670 * If there is no object backing this entry, we might as well create
1671 * one now. If we defer it, an object can get created after the map
1672 * is clipped, and individual objects will be created for the split-up
1673 * map. This is a bit of a hack, but is also about the best place to
1674 * put this improvement.
1676 if (entry->object.vm_object == NULL && !map->system_map) {
1678 object = vm_object_allocate(OBJT_DEFAULT,
1679 atop(entry->end - entry->start));
1680 entry->object.vm_object = object;
1682 if (entry->cred != NULL) {
1683 object->cred = entry->cred;
1684 object->charge = entry->end - entry->start;
1687 } else if (entry->object.vm_object != NULL &&
1688 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) &&
1689 entry->cred != NULL) {
1690 VM_OBJECT_LOCK(entry->object.vm_object);
1691 KASSERT(entry->object.vm_object->cred == NULL,
1692 ("OVERCOMMIT: vm_entry_clip_end: both cred e %p", entry));
1693 entry->object.vm_object->cred = entry->cred;
1694 entry->object.vm_object->charge = entry->end - entry->start;
1695 VM_OBJECT_UNLOCK(entry->object.vm_object);
1700 * Create a new entry and insert it AFTER the specified entry
1702 new_entry = vm_map_entry_create(map);
1703 *new_entry = *entry;
1705 new_entry->start = entry->end = end;
1706 new_entry->offset += (end - entry->start);
1707 if (new_entry->cred != NULL)
1708 crhold(entry->cred);
1710 vm_map_entry_link(map, entry, new_entry);
1712 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1713 vm_object_reference(new_entry->object.vm_object);
1718 * vm_map_submap: [ kernel use only ]
1720 * Mark the given range as handled by a subordinate map.
1722 * This range must have been created with vm_map_find,
1723 * and no other operations may have been performed on this
1724 * range prior to calling vm_map_submap.
1726 * Only a limited number of operations can be performed
1727 * within this rage after calling vm_map_submap:
1729 * [Don't try vm_map_copy!]
1731 * To remove a submapping, one must first remove the
1732 * range from the superior map, and then destroy the
1733 * submap (if desired). [Better yet, don't try it.]
1742 vm_map_entry_t entry;
1743 int result = KERN_INVALID_ARGUMENT;
1747 VM_MAP_RANGE_CHECK(map, start, end);
1749 if (vm_map_lookup_entry(map, start, &entry)) {
1750 vm_map_clip_start(map, entry, start);
1752 entry = entry->next;
1754 vm_map_clip_end(map, entry, end);
1756 if ((entry->start == start) && (entry->end == end) &&
1757 ((entry->eflags & MAP_ENTRY_COW) == 0) &&
1758 (entry->object.vm_object == NULL)) {
1759 entry->object.sub_map = submap;
1760 entry->eflags |= MAP_ENTRY_IS_SUB_MAP;
1761 result = KERN_SUCCESS;
1769 * The maximum number of pages to map
1771 #define MAX_INIT_PT 96
1774 * vm_map_pmap_enter:
1776 * Preload read-only mappings for the given object's resident pages into
1777 * the given map. This eliminates the soft faults on process startup and
1778 * immediately after an mmap(2). Because these are speculative mappings,
1779 * cached pages are not reactivated and mapped.
1782 vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot,
1783 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags)
1786 vm_page_t p, p_start;
1787 vm_pindex_t psize, tmpidx;
1789 if ((prot & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0 || object == NULL)
1791 VM_OBJECT_LOCK(object);
1792 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) {
1793 pmap_object_init_pt(map->pmap, addr, object, pindex, size);
1799 if ((flags & MAP_PREFAULT_PARTIAL) && psize > MAX_INIT_PT &&
1800 object->resident_page_count > MAX_INIT_PT)
1803 if (psize + pindex > object->size) {
1804 if (object->size < pindex)
1806 psize = object->size - pindex;
1812 p = vm_page_find_least(object, pindex);
1814 * Assert: the variable p is either (1) the page with the
1815 * least pindex greater than or equal to the parameter pindex
1819 p != NULL && (tmpidx = p->pindex - pindex) < psize;
1820 p = TAILQ_NEXT(p, listq)) {
1822 * don't allow an madvise to blow away our really
1823 * free pages allocating pv entries.
1825 if ((flags & MAP_PREFAULT_MADVISE) &&
1826 cnt.v_free_count < cnt.v_free_reserved) {
1830 if (p->valid == VM_PAGE_BITS_ALL) {
1831 if (p_start == NULL) {
1832 start = addr + ptoa(tmpidx);
1835 } else if (p_start != NULL) {
1836 pmap_enter_object(map->pmap, start, addr +
1837 ptoa(tmpidx), p_start, prot);
1841 if (p_start != NULL)
1842 pmap_enter_object(map->pmap, start, addr + ptoa(psize),
1845 VM_OBJECT_UNLOCK(object);
1851 * Sets the protection of the specified address
1852 * region in the target map. If "set_max" is
1853 * specified, the maximum protection is to be set;
1854 * otherwise, only the current protection is affected.
1857 vm_map_protect(vm_map_t map, vm_offset_t start, vm_offset_t end,
1858 vm_prot_t new_prot, boolean_t set_max)
1860 vm_map_entry_t current, entry;
1867 VM_MAP_RANGE_CHECK(map, start, end);
1869 if (vm_map_lookup_entry(map, start, &entry)) {
1870 vm_map_clip_start(map, entry, start);
1872 entry = entry->next;
1876 * Make a first pass to check for protection violations.
1879 while ((current != &map->header) && (current->start < end)) {
1880 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
1882 return (KERN_INVALID_ARGUMENT);
1884 if ((new_prot & current->max_protection) != new_prot) {
1886 return (KERN_PROTECTION_FAILURE);
1888 current = current->next;
1893 * Do an accounting pass for private read-only mappings that
1894 * now will do cow due to allowed write (e.g. debugger sets
1895 * breakpoint on text segment)
1897 for (current = entry; (current != &map->header) &&
1898 (current->start < end); current = current->next) {
1900 vm_map_clip_end(map, current, end);
1903 ((new_prot & ~(current->protection)) & VM_PROT_WRITE) == 0 ||
1904 ENTRY_CHARGED(current)) {
1908 cred = curthread->td_ucred;
1909 obj = current->object.vm_object;
1911 if (obj == NULL || (current->eflags & MAP_ENTRY_NEEDS_COPY)) {
1912 if (!swap_reserve(current->end - current->start)) {
1914 return (KERN_RESOURCE_SHORTAGE);
1917 current->cred = cred;
1921 VM_OBJECT_LOCK(obj);
1922 if (obj->type != OBJT_DEFAULT && obj->type != OBJT_SWAP) {
1923 VM_OBJECT_UNLOCK(obj);
1928 * Charge for the whole object allocation now, since
1929 * we cannot distinguish between non-charged and
1930 * charged clipped mapping of the same object later.
1932 KASSERT(obj->charge == 0,
1933 ("vm_map_protect: object %p overcharged\n", obj));
1934 if (!swap_reserve(ptoa(obj->size))) {
1935 VM_OBJECT_UNLOCK(obj);
1937 return (KERN_RESOURCE_SHORTAGE);
1942 obj->charge = ptoa(obj->size);
1943 VM_OBJECT_UNLOCK(obj);
1947 * Go back and fix up protections. [Note that clipping is not
1948 * necessary the second time.]
1951 while ((current != &map->header) && (current->start < end)) {
1952 old_prot = current->protection;
1955 current->protection =
1956 (current->max_protection = new_prot) &
1959 current->protection = new_prot;
1961 if ((current->eflags & (MAP_ENTRY_COW | MAP_ENTRY_USER_WIRED))
1962 == (MAP_ENTRY_COW | MAP_ENTRY_USER_WIRED) &&
1963 (current->protection & VM_PROT_WRITE) != 0 &&
1964 (old_prot & VM_PROT_WRITE) == 0) {
1965 vm_fault_copy_entry(map, map, current, current, NULL);
1969 * When restricting access, update the physical map. Worry
1970 * about copy-on-write here.
1972 if ((old_prot & ~current->protection) != 0) {
1973 #define MASK(entry) (((entry)->eflags & MAP_ENTRY_COW) ? ~VM_PROT_WRITE : \
1975 pmap_protect(map->pmap, current->start,
1977 current->protection & MASK(current));
1980 vm_map_simplify_entry(map, current);
1981 current = current->next;
1984 return (KERN_SUCCESS);
1990 * This routine traverses a processes map handling the madvise
1991 * system call. Advisories are classified as either those effecting
1992 * the vm_map_entry structure, or those effecting the underlying
2002 vm_map_entry_t current, entry;
2006 * Some madvise calls directly modify the vm_map_entry, in which case
2007 * we need to use an exclusive lock on the map and we need to perform
2008 * various clipping operations. Otherwise we only need a read-lock
2013 case MADV_SEQUENTIAL:
2025 vm_map_lock_read(map);
2028 return (KERN_INVALID_ARGUMENT);
2032 * Locate starting entry and clip if necessary.
2034 VM_MAP_RANGE_CHECK(map, start, end);
2036 if (vm_map_lookup_entry(map, start, &entry)) {
2038 vm_map_clip_start(map, entry, start);
2040 entry = entry->next;
2045 * madvise behaviors that are implemented in the vm_map_entry.
2047 * We clip the vm_map_entry so that behavioral changes are
2048 * limited to the specified address range.
2050 for (current = entry;
2051 (current != &map->header) && (current->start < end);
2052 current = current->next
2054 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
2057 vm_map_clip_end(map, current, end);
2061 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_NORMAL);
2063 case MADV_SEQUENTIAL:
2064 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_SEQUENTIAL);
2067 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_RANDOM);
2070 current->eflags |= MAP_ENTRY_NOSYNC;
2073 current->eflags &= ~MAP_ENTRY_NOSYNC;
2076 current->eflags |= MAP_ENTRY_NOCOREDUMP;
2079 current->eflags &= ~MAP_ENTRY_NOCOREDUMP;
2084 vm_map_simplify_entry(map, current);
2092 * madvise behaviors that are implemented in the underlying
2095 * Since we don't clip the vm_map_entry, we have to clip
2096 * the vm_object pindex and count.
2098 for (current = entry;
2099 (current != &map->header) && (current->start < end);
2100 current = current->next
2102 vm_offset_t useStart;
2104 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
2107 pindex = OFF_TO_IDX(current->offset);
2108 count = atop(current->end - current->start);
2109 useStart = current->start;
2111 if (current->start < start) {
2112 pindex += atop(start - current->start);
2113 count -= atop(start - current->start);
2116 if (current->end > end)
2117 count -= atop(current->end - end);
2122 vm_object_madvise(current->object.vm_object,
2123 pindex, count, behav);
2124 if (behav == MADV_WILLNEED) {
2125 vm_map_pmap_enter(map,
2127 current->protection,
2128 current->object.vm_object,
2130 (count << PAGE_SHIFT),
2131 MAP_PREFAULT_MADVISE
2135 vm_map_unlock_read(map);
2144 * Sets the inheritance of the specified address
2145 * range in the target map. Inheritance
2146 * affects how the map will be shared with
2147 * child maps at the time of vmspace_fork.
2150 vm_map_inherit(vm_map_t map, vm_offset_t start, vm_offset_t end,
2151 vm_inherit_t new_inheritance)
2153 vm_map_entry_t entry;
2154 vm_map_entry_t temp_entry;
2156 switch (new_inheritance) {
2157 case VM_INHERIT_NONE:
2158 case VM_INHERIT_COPY:
2159 case VM_INHERIT_SHARE:
2162 return (KERN_INVALID_ARGUMENT);
2165 VM_MAP_RANGE_CHECK(map, start, end);
2166 if (vm_map_lookup_entry(map, start, &temp_entry)) {
2168 vm_map_clip_start(map, entry, start);
2170 entry = temp_entry->next;
2171 while ((entry != &map->header) && (entry->start < end)) {
2172 vm_map_clip_end(map, entry, end);
2173 entry->inheritance = new_inheritance;
2174 vm_map_simplify_entry(map, entry);
2175 entry = entry->next;
2178 return (KERN_SUCCESS);
2184 * Implements both kernel and user unwiring.
2187 vm_map_unwire(vm_map_t map, vm_offset_t start, vm_offset_t end,
2190 vm_map_entry_t entry, first_entry, tmp_entry;
2191 vm_offset_t saved_start;
2192 unsigned int last_timestamp;
2194 boolean_t need_wakeup, result, user_unwire;
2196 user_unwire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE;
2198 VM_MAP_RANGE_CHECK(map, start, end);
2199 if (!vm_map_lookup_entry(map, start, &first_entry)) {
2200 if (flags & VM_MAP_WIRE_HOLESOK)
2201 first_entry = first_entry->next;
2204 return (KERN_INVALID_ADDRESS);
2207 last_timestamp = map->timestamp;
2208 entry = first_entry;
2209 while (entry != &map->header && entry->start < end) {
2210 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
2212 * We have not yet clipped the entry.
2214 saved_start = (start >= entry->start) ? start :
2216 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2217 if (vm_map_unlock_and_wait(map, 0)) {
2219 * Allow interruption of user unwiring?
2223 if (last_timestamp+1 != map->timestamp) {
2225 * Look again for the entry because the map was
2226 * modified while it was unlocked.
2227 * Specifically, the entry may have been
2228 * clipped, merged, or deleted.
2230 if (!vm_map_lookup_entry(map, saved_start,
2232 if (flags & VM_MAP_WIRE_HOLESOK)
2233 tmp_entry = tmp_entry->next;
2235 if (saved_start == start) {
2237 * First_entry has been deleted.
2240 return (KERN_INVALID_ADDRESS);
2243 rv = KERN_INVALID_ADDRESS;
2247 if (entry == first_entry)
2248 first_entry = tmp_entry;
2253 last_timestamp = map->timestamp;
2256 vm_map_clip_start(map, entry, start);
2257 vm_map_clip_end(map, entry, end);
2259 * Mark the entry in case the map lock is released. (See
2262 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
2264 * Check the map for holes in the specified region.
2265 * If VM_MAP_WIRE_HOLESOK was specified, skip this check.
2267 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) &&
2268 (entry->end < end && (entry->next == &map->header ||
2269 entry->next->start > entry->end))) {
2271 rv = KERN_INVALID_ADDRESS;
2275 * If system unwiring, require that the entry is system wired.
2278 vm_map_entry_system_wired_count(entry) == 0) {
2280 rv = KERN_INVALID_ARGUMENT;
2283 entry = entry->next;
2287 need_wakeup = FALSE;
2288 if (first_entry == NULL) {
2289 result = vm_map_lookup_entry(map, start, &first_entry);
2290 if (!result && (flags & VM_MAP_WIRE_HOLESOK))
2291 first_entry = first_entry->next;
2293 KASSERT(result, ("vm_map_unwire: lookup failed"));
2295 entry = first_entry;
2296 while (entry != &map->header && entry->start < end) {
2297 if (rv == KERN_SUCCESS && (!user_unwire ||
2298 (entry->eflags & MAP_ENTRY_USER_WIRED))) {
2300 entry->eflags &= ~MAP_ENTRY_USER_WIRED;
2301 entry->wired_count--;
2302 if (entry->wired_count == 0) {
2304 * Retain the map lock.
2306 vm_fault_unwire(map, entry->start, entry->end,
2307 entry->object.vm_object != NULL &&
2308 (entry->object.vm_object->type == OBJT_DEVICE ||
2309 entry->object.vm_object->type == OBJT_SG));
2312 KASSERT(entry->eflags & MAP_ENTRY_IN_TRANSITION,
2313 ("vm_map_unwire: in-transition flag missing"));
2314 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION;
2315 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
2316 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
2319 vm_map_simplify_entry(map, entry);
2320 entry = entry->next;
2331 * Implements both kernel and user wiring.
2334 vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t end,
2337 vm_map_entry_t entry, first_entry, tmp_entry;
2338 vm_offset_t saved_end, saved_start;
2339 unsigned int last_timestamp;
2341 boolean_t fictitious, need_wakeup, result, user_wire;
2345 if (flags & VM_MAP_WIRE_WRITE)
2346 prot |= VM_PROT_WRITE;
2347 user_wire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE;
2349 VM_MAP_RANGE_CHECK(map, start, end);
2350 if (!vm_map_lookup_entry(map, start, &first_entry)) {
2351 if (flags & VM_MAP_WIRE_HOLESOK)
2352 first_entry = first_entry->next;
2355 return (KERN_INVALID_ADDRESS);
2358 last_timestamp = map->timestamp;
2359 entry = first_entry;
2360 while (entry != &map->header && entry->start < end) {
2361 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
2363 * We have not yet clipped the entry.
2365 saved_start = (start >= entry->start) ? start :
2367 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2368 if (vm_map_unlock_and_wait(map, 0)) {
2370 * Allow interruption of user wiring?
2374 if (last_timestamp + 1 != map->timestamp) {
2376 * Look again for the entry because the map was
2377 * modified while it was unlocked.
2378 * Specifically, the entry may have been
2379 * clipped, merged, or deleted.
2381 if (!vm_map_lookup_entry(map, saved_start,
2383 if (flags & VM_MAP_WIRE_HOLESOK)
2384 tmp_entry = tmp_entry->next;
2386 if (saved_start == start) {
2388 * first_entry has been deleted.
2391 return (KERN_INVALID_ADDRESS);
2394 rv = KERN_INVALID_ADDRESS;
2398 if (entry == first_entry)
2399 first_entry = tmp_entry;
2404 last_timestamp = map->timestamp;
2407 vm_map_clip_start(map, entry, start);
2408 vm_map_clip_end(map, entry, end);
2410 * Mark the entry in case the map lock is released. (See
2413 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
2414 if ((entry->protection & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0
2415 || (entry->protection & prot) != prot) {
2416 entry->eflags |= MAP_ENTRY_WIRE_SKIPPED;
2417 if ((flags & VM_MAP_WIRE_HOLESOK) == 0) {
2419 rv = KERN_INVALID_ADDRESS;
2424 if (entry->wired_count == 0) {
2425 entry->wired_count++;
2426 saved_start = entry->start;
2427 saved_end = entry->end;
2428 fictitious = entry->object.vm_object != NULL &&
2429 (entry->object.vm_object->type == OBJT_DEVICE ||
2430 entry->object.vm_object->type == OBJT_SG);
2432 * Release the map lock, relying on the in-transition
2433 * mark. Mark the map busy for fork.
2437 rv = vm_fault_wire(map, saved_start, saved_end,
2441 if (last_timestamp + 1 != map->timestamp) {
2443 * Look again for the entry because the map was
2444 * modified while it was unlocked. The entry
2445 * may have been clipped, but NOT merged or
2448 result = vm_map_lookup_entry(map, saved_start,
2450 KASSERT(result, ("vm_map_wire: lookup failed"));
2451 if (entry == first_entry)
2452 first_entry = tmp_entry;
2456 while (entry->end < saved_end) {
2457 if (rv != KERN_SUCCESS) {
2458 KASSERT(entry->wired_count == 1,
2459 ("vm_map_wire: bad count"));
2460 entry->wired_count = -1;
2462 entry = entry->next;
2465 last_timestamp = map->timestamp;
2466 if (rv != KERN_SUCCESS) {
2467 KASSERT(entry->wired_count == 1,
2468 ("vm_map_wire: bad count"));
2470 * Assign an out-of-range value to represent
2471 * the failure to wire this entry.
2473 entry->wired_count = -1;
2477 } else if (!user_wire ||
2478 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
2479 entry->wired_count++;
2482 * Check the map for holes in the specified region.
2483 * If VM_MAP_WIRE_HOLESOK was specified, skip this check.
2486 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) &&
2487 (entry->end < end && (entry->next == &map->header ||
2488 entry->next->start > entry->end))) {
2490 rv = KERN_INVALID_ADDRESS;
2493 entry = entry->next;
2497 need_wakeup = FALSE;
2498 if (first_entry == NULL) {
2499 result = vm_map_lookup_entry(map, start, &first_entry);
2500 if (!result && (flags & VM_MAP_WIRE_HOLESOK))
2501 first_entry = first_entry->next;
2503 KASSERT(result, ("vm_map_wire: lookup failed"));
2505 entry = first_entry;
2506 while (entry != &map->header && entry->start < end) {
2507 if ((entry->eflags & MAP_ENTRY_WIRE_SKIPPED) != 0)
2508 goto next_entry_done;
2509 if (rv == KERN_SUCCESS) {
2511 entry->eflags |= MAP_ENTRY_USER_WIRED;
2512 } else if (entry->wired_count == -1) {
2514 * Wiring failed on this entry. Thus, unwiring is
2517 entry->wired_count = 0;
2520 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0)
2521 entry->wired_count--;
2522 if (entry->wired_count == 0) {
2524 * Retain the map lock.
2526 vm_fault_unwire(map, entry->start, entry->end,
2527 entry->object.vm_object != NULL &&
2528 (entry->object.vm_object->type == OBJT_DEVICE ||
2529 entry->object.vm_object->type == OBJT_SG));
2533 KASSERT(entry->eflags & MAP_ENTRY_IN_TRANSITION,
2534 ("vm_map_wire: in-transition flag missing"));
2535 entry->eflags &= ~(MAP_ENTRY_IN_TRANSITION|MAP_ENTRY_WIRE_SKIPPED);
2536 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
2537 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
2540 vm_map_simplify_entry(map, entry);
2541 entry = entry->next;
2552 * Push any dirty cached pages in the address range to their pager.
2553 * If syncio is TRUE, dirty pages are written synchronously.
2554 * If invalidate is TRUE, any cached pages are freed as well.
2556 * If the size of the region from start to end is zero, we are
2557 * supposed to flush all modified pages within the region containing
2558 * start. Unfortunately, a region can be split or coalesced with
2559 * neighboring regions, making it difficult to determine what the
2560 * original region was. Therefore, we approximate this requirement by
2561 * flushing the current region containing start.
2563 * Returns an error if any part of the specified range is not mapped.
2571 boolean_t invalidate)
2573 vm_map_entry_t current;
2574 vm_map_entry_t entry;
2577 vm_ooffset_t offset;
2578 unsigned int last_timestamp;
2580 vm_map_lock_read(map);
2581 VM_MAP_RANGE_CHECK(map, start, end);
2582 if (!vm_map_lookup_entry(map, start, &entry)) {
2583 vm_map_unlock_read(map);
2584 return (KERN_INVALID_ADDRESS);
2585 } else if (start == end) {
2586 start = entry->start;
2590 * Make a first pass to check for user-wired memory and holes.
2592 for (current = entry; current != &map->header && current->start < end;
2593 current = current->next) {
2594 if (invalidate && (current->eflags & MAP_ENTRY_USER_WIRED)) {
2595 vm_map_unlock_read(map);
2596 return (KERN_INVALID_ARGUMENT);
2598 if (end > current->end &&
2599 (current->next == &map->header ||
2600 current->end != current->next->start)) {
2601 vm_map_unlock_read(map);
2602 return (KERN_INVALID_ADDRESS);
2607 pmap_remove(map->pmap, start, end);
2610 * Make a second pass, cleaning/uncaching pages from the indicated
2613 for (current = entry; current != &map->header && current->start < end;) {
2614 offset = current->offset + (start - current->start);
2615 size = (end <= current->end ? end : current->end) - start;
2616 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
2618 vm_map_entry_t tentry;
2621 smap = current->object.sub_map;
2622 vm_map_lock_read(smap);
2623 (void) vm_map_lookup_entry(smap, offset, &tentry);
2624 tsize = tentry->end - offset;
2627 object = tentry->object.vm_object;
2628 offset = tentry->offset + (offset - tentry->start);
2629 vm_map_unlock_read(smap);
2631 object = current->object.vm_object;
2633 vm_object_reference(object);
2634 last_timestamp = map->timestamp;
2635 vm_map_unlock_read(map);
2636 vm_object_sync(object, offset, size, syncio, invalidate);
2638 vm_object_deallocate(object);
2639 vm_map_lock_read(map);
2640 if (last_timestamp == map->timestamp ||
2641 !vm_map_lookup_entry(map, start, ¤t))
2642 current = current->next;
2645 vm_map_unlock_read(map);
2646 return (KERN_SUCCESS);
2650 * vm_map_entry_unwire: [ internal use only ]
2652 * Make the region specified by this entry pageable.
2654 * The map in question should be locked.
2655 * [This is the reason for this routine's existence.]
2658 vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry)
2660 vm_fault_unwire(map, entry->start, entry->end,
2661 entry->object.vm_object != NULL &&
2662 (entry->object.vm_object->type == OBJT_DEVICE ||
2663 entry->object.vm_object->type == OBJT_SG));
2664 entry->wired_count = 0;
2668 vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map)
2671 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0)
2672 vm_object_deallocate(entry->object.vm_object);
2673 uma_zfree(system_map ? kmapentzone : mapentzone, entry);
2677 * vm_map_entry_delete: [ internal use only ]
2679 * Deallocate the given entry from the target map.
2682 vm_map_entry_delete(vm_map_t map, vm_map_entry_t entry)
2685 vm_pindex_t offidxstart, offidxend, count, size1;
2688 vm_map_entry_unlink(map, entry);
2689 object = entry->object.vm_object;
2690 size = entry->end - entry->start;
2693 if (entry->cred != NULL) {
2694 swap_release_by_cred(size, entry->cred);
2695 crfree(entry->cred);
2698 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0 &&
2700 KASSERT(entry->cred == NULL || object->cred == NULL ||
2701 (entry->eflags & MAP_ENTRY_NEEDS_COPY),
2702 ("OVERCOMMIT vm_map_entry_delete: both cred %p", entry));
2703 count = OFF_TO_IDX(size);
2704 offidxstart = OFF_TO_IDX(entry->offset);
2705 offidxend = offidxstart + count;
2706 VM_OBJECT_LOCK(object);
2707 if (object->ref_count != 1 &&
2708 ((object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING ||
2709 object == kernel_object || object == kmem_object)) {
2710 vm_object_collapse(object);
2711 vm_object_page_remove(object, offidxstart, offidxend, FALSE);
2712 if (object->type == OBJT_SWAP)
2713 swap_pager_freespace(object, offidxstart, count);
2714 if (offidxend >= object->size &&
2715 offidxstart < object->size) {
2716 size1 = object->size;
2717 object->size = offidxstart;
2718 if (object->cred != NULL) {
2719 size1 -= object->size;
2720 KASSERT(object->charge >= ptoa(size1),
2721 ("vm_map_entry_delete: object->charge < 0"));
2722 swap_release_by_cred(ptoa(size1), object->cred);
2723 object->charge -= ptoa(size1);
2727 VM_OBJECT_UNLOCK(object);
2729 entry->object.vm_object = NULL;
2730 if (map->system_map)
2731 vm_map_entry_deallocate(entry, TRUE);
2733 entry->next = curthread->td_map_def_user;
2734 curthread->td_map_def_user = entry;
2739 * vm_map_delete: [ internal use only ]
2741 * Deallocates the given address range from the target
2745 vm_map_delete(vm_map_t map, vm_offset_t start, vm_offset_t end)
2747 vm_map_entry_t entry;
2748 vm_map_entry_t first_entry;
2750 VM_MAP_ASSERT_LOCKED(map);
2753 * Find the start of the region, and clip it
2755 if (!vm_map_lookup_entry(map, start, &first_entry))
2756 entry = first_entry->next;
2758 entry = first_entry;
2759 vm_map_clip_start(map, entry, start);
2763 * Step through all entries in this region
2765 while ((entry != &map->header) && (entry->start < end)) {
2766 vm_map_entry_t next;
2769 * Wait for wiring or unwiring of an entry to complete.
2770 * Also wait for any system wirings to disappear on
2773 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 ||
2774 (vm_map_pmap(map) != kernel_pmap &&
2775 vm_map_entry_system_wired_count(entry) != 0)) {
2776 unsigned int last_timestamp;
2777 vm_offset_t saved_start;
2778 vm_map_entry_t tmp_entry;
2780 saved_start = entry->start;
2781 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2782 last_timestamp = map->timestamp;
2783 (void) vm_map_unlock_and_wait(map, 0);
2785 if (last_timestamp + 1 != map->timestamp) {
2787 * Look again for the entry because the map was
2788 * modified while it was unlocked.
2789 * Specifically, the entry may have been
2790 * clipped, merged, or deleted.
2792 if (!vm_map_lookup_entry(map, saved_start,
2794 entry = tmp_entry->next;
2797 vm_map_clip_start(map, entry,
2803 vm_map_clip_end(map, entry, end);
2808 * Unwire before removing addresses from the pmap; otherwise,
2809 * unwiring will put the entries back in the pmap.
2811 if (entry->wired_count != 0) {
2812 vm_map_entry_unwire(map, entry);
2815 pmap_remove(map->pmap, entry->start, entry->end);
2818 * Delete the entry only after removing all pmap
2819 * entries pointing to its pages. (Otherwise, its
2820 * page frames may be reallocated, and any modify bits
2821 * will be set in the wrong object!)
2823 vm_map_entry_delete(map, entry);
2826 return (KERN_SUCCESS);
2832 * Remove the given address range from the target map.
2833 * This is the exported form of vm_map_delete.
2836 vm_map_remove(vm_map_t map, vm_offset_t start, vm_offset_t end)
2841 VM_MAP_RANGE_CHECK(map, start, end);
2842 result = vm_map_delete(map, start, end);
2848 * vm_map_check_protection:
2850 * Assert that the target map allows the specified privilege on the
2851 * entire address region given. The entire region must be allocated.
2853 * WARNING! This code does not and should not check whether the
2854 * contents of the region is accessible. For example a smaller file
2855 * might be mapped into a larger address space.
2857 * NOTE! This code is also called by munmap().
2859 * The map must be locked. A read lock is sufficient.
2862 vm_map_check_protection(vm_map_t map, vm_offset_t start, vm_offset_t end,
2863 vm_prot_t protection)
2865 vm_map_entry_t entry;
2866 vm_map_entry_t tmp_entry;
2868 if (!vm_map_lookup_entry(map, start, &tmp_entry))
2872 while (start < end) {
2873 if (entry == &map->header)
2878 if (start < entry->start)
2881 * Check protection associated with entry.
2883 if ((entry->protection & protection) != protection)
2885 /* go to next entry */
2887 entry = entry->next;
2893 * vm_map_copy_entry:
2895 * Copies the contents of the source entry to the destination
2896 * entry. The entries *must* be aligned properly.
2902 vm_map_entry_t src_entry,
2903 vm_map_entry_t dst_entry,
2904 vm_ooffset_t *fork_charge)
2906 vm_object_t src_object;
2911 VM_MAP_ASSERT_LOCKED(dst_map);
2913 if ((dst_entry->eflags|src_entry->eflags) & MAP_ENTRY_IS_SUB_MAP)
2916 if (src_entry->wired_count == 0) {
2919 * If the source entry is marked needs_copy, it is already
2922 if ((src_entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) {
2923 pmap_protect(src_map->pmap,
2926 src_entry->protection & ~VM_PROT_WRITE);
2930 * Make a copy of the object.
2932 size = src_entry->end - src_entry->start;
2933 if ((src_object = src_entry->object.vm_object) != NULL) {
2934 VM_OBJECT_LOCK(src_object);
2935 charged = ENTRY_CHARGED(src_entry);
2936 if ((src_object->handle == NULL) &&
2937 (src_object->type == OBJT_DEFAULT ||
2938 src_object->type == OBJT_SWAP)) {
2939 vm_object_collapse(src_object);
2940 if ((src_object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING) {
2941 vm_object_split(src_entry);
2942 src_object = src_entry->object.vm_object;
2945 vm_object_reference_locked(src_object);
2946 vm_object_clear_flag(src_object, OBJ_ONEMAPPING);
2947 if (src_entry->cred != NULL &&
2948 !(src_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
2949 KASSERT(src_object->cred == NULL,
2950 ("OVERCOMMIT: vm_map_copy_entry: cred %p",
2952 src_object->cred = src_entry->cred;
2953 src_object->charge = size;
2955 VM_OBJECT_UNLOCK(src_object);
2956 dst_entry->object.vm_object = src_object;
2958 cred = curthread->td_ucred;
2960 dst_entry->cred = cred;
2961 *fork_charge += size;
2962 if (!(src_entry->eflags &
2963 MAP_ENTRY_NEEDS_COPY)) {
2965 src_entry->cred = cred;
2966 *fork_charge += size;
2969 src_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY);
2970 dst_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY);
2971 dst_entry->offset = src_entry->offset;
2973 dst_entry->object.vm_object = NULL;
2974 dst_entry->offset = 0;
2975 if (src_entry->cred != NULL) {
2976 dst_entry->cred = curthread->td_ucred;
2977 crhold(dst_entry->cred);
2978 *fork_charge += size;
2982 pmap_copy(dst_map->pmap, src_map->pmap, dst_entry->start,
2983 dst_entry->end - dst_entry->start, src_entry->start);
2986 * Of course, wired down pages can't be set copy-on-write.
2987 * Cause wired pages to be copied into the new map by
2988 * simulating faults (the new pages are pageable)
2990 vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry,
2996 * vmspace_map_entry_forked:
2997 * Update the newly-forked vmspace each time a map entry is inherited
2998 * or copied. The values for vm_dsize and vm_tsize are approximate
2999 * (and mostly-obsolete ideas in the face of mmap(2) et al.)
3002 vmspace_map_entry_forked(const struct vmspace *vm1, struct vmspace *vm2,
3003 vm_map_entry_t entry)
3005 vm_size_t entrysize;
3008 entrysize = entry->end - entry->start;
3009 vm2->vm_map.size += entrysize;
3010 if (entry->eflags & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP)) {
3011 vm2->vm_ssize += btoc(entrysize);
3012 } else if (entry->start >= (vm_offset_t)vm1->vm_daddr &&
3013 entry->start < (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)) {
3014 newend = MIN(entry->end,
3015 (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize));
3016 vm2->vm_dsize += btoc(newend - entry->start);
3017 } else if (entry->start >= (vm_offset_t)vm1->vm_taddr &&
3018 entry->start < (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)) {
3019 newend = MIN(entry->end,
3020 (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize));
3021 vm2->vm_tsize += btoc(newend - entry->start);
3027 * Create a new process vmspace structure and vm_map
3028 * based on those of an existing process. The new map
3029 * is based on the old map, according to the inheritance
3030 * values on the regions in that map.
3032 * XXX It might be worth coalescing the entries added to the new vmspace.
3034 * The source map must not be locked.
3037 vmspace_fork(struct vmspace *vm1, vm_ooffset_t *fork_charge)
3039 struct vmspace *vm2;
3040 vm_map_t old_map = &vm1->vm_map;
3042 vm_map_entry_t old_entry;
3043 vm_map_entry_t new_entry;
3047 vm_map_lock(old_map);
3049 vm_map_wait_busy(old_map);
3050 vm2 = vmspace_alloc(old_map->min_offset, old_map->max_offset);
3052 goto unlock_and_return;
3053 vm2->vm_taddr = vm1->vm_taddr;
3054 vm2->vm_daddr = vm1->vm_daddr;
3055 vm2->vm_maxsaddr = vm1->vm_maxsaddr;
3056 new_map = &vm2->vm_map; /* XXX */
3057 locked = vm_map_trylock(new_map); /* trylock to silence WITNESS */
3058 KASSERT(locked, ("vmspace_fork: lock failed"));
3059 new_map->timestamp = 1;
3061 old_entry = old_map->header.next;
3063 while (old_entry != &old_map->header) {
3064 if (old_entry->eflags & MAP_ENTRY_IS_SUB_MAP)
3065 panic("vm_map_fork: encountered a submap");
3067 switch (old_entry->inheritance) {
3068 case VM_INHERIT_NONE:
3071 case VM_INHERIT_SHARE:
3073 * Clone the entry, creating the shared object if necessary.
3075 object = old_entry->object.vm_object;
3076 if (object == NULL) {
3077 object = vm_object_allocate(OBJT_DEFAULT,
3078 atop(old_entry->end - old_entry->start));
3079 old_entry->object.vm_object = object;
3080 old_entry->offset = 0;
3081 if (old_entry->cred != NULL) {
3082 object->cred = old_entry->cred;
3083 object->charge = old_entry->end -
3085 old_entry->cred = NULL;
3090 * Add the reference before calling vm_object_shadow
3091 * to insure that a shadow object is created.
3093 vm_object_reference(object);
3094 if (old_entry->eflags & MAP_ENTRY_NEEDS_COPY) {
3095 vm_object_shadow(&old_entry->object.vm_object,
3097 old_entry->end - old_entry->start);
3098 old_entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
3099 /* Transfer the second reference too. */
3100 vm_object_reference(
3101 old_entry->object.vm_object);
3104 * As in vm_map_simplify_entry(), the
3105 * vnode lock will not be acquired in
3106 * this call to vm_object_deallocate().
3108 vm_object_deallocate(object);
3109 object = old_entry->object.vm_object;
3111 VM_OBJECT_LOCK(object);
3112 vm_object_clear_flag(object, OBJ_ONEMAPPING);
3113 if (old_entry->cred != NULL) {
3114 KASSERT(object->cred == NULL, ("vmspace_fork both cred"));
3115 object->cred = old_entry->cred;
3116 object->charge = old_entry->end - old_entry->start;
3117 old_entry->cred = NULL;
3119 VM_OBJECT_UNLOCK(object);
3122 * Clone the entry, referencing the shared object.
3124 new_entry = vm_map_entry_create(new_map);
3125 *new_entry = *old_entry;
3126 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED |
3127 MAP_ENTRY_IN_TRANSITION);
3128 new_entry->wired_count = 0;
3131 * Insert the entry into the new map -- we know we're
3132 * inserting at the end of the new map.
3134 vm_map_entry_link(new_map, new_map->header.prev,
3136 vmspace_map_entry_forked(vm1, vm2, new_entry);
3139 * Update the physical map
3141 pmap_copy(new_map->pmap, old_map->pmap,
3143 (old_entry->end - old_entry->start),
3147 case VM_INHERIT_COPY:
3149 * Clone the entry and link into the map.
3151 new_entry = vm_map_entry_create(new_map);
3152 *new_entry = *old_entry;
3153 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED |
3154 MAP_ENTRY_IN_TRANSITION);
3155 new_entry->wired_count = 0;
3156 new_entry->object.vm_object = NULL;
3157 new_entry->cred = NULL;
3158 vm_map_entry_link(new_map, new_map->header.prev,
3160 vmspace_map_entry_forked(vm1, vm2, new_entry);
3161 vm_map_copy_entry(old_map, new_map, old_entry,
3162 new_entry, fork_charge);
3165 old_entry = old_entry->next;
3168 vm_map_unlock(old_map);
3170 vm_map_unlock(new_map);
3176 vm_map_stack(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
3177 vm_prot_t prot, vm_prot_t max, int cow)
3179 vm_map_entry_t new_entry, prev_entry;
3180 vm_offset_t bot, top;
3181 vm_size_t init_ssize;
3186 * The stack orientation is piggybacked with the cow argument.
3187 * Extract it into orient and mask the cow argument so that we
3188 * don't pass it around further.
3189 * NOTE: We explicitly allow bi-directional stacks.
3191 orient = cow & (MAP_STACK_GROWS_DOWN|MAP_STACK_GROWS_UP);
3193 KASSERT(orient != 0, ("No stack grow direction"));
3195 if (addrbos < vm_map_min(map) ||
3196 addrbos > vm_map_max(map) ||
3197 addrbos + max_ssize < addrbos)
3198 return (KERN_NO_SPACE);
3200 init_ssize = (max_ssize < sgrowsiz) ? max_ssize : sgrowsiz;
3202 PROC_LOCK(curthread->td_proc);
3203 vmemlim = lim_cur(curthread->td_proc, RLIMIT_VMEM);
3204 PROC_UNLOCK(curthread->td_proc);
3208 /* If addr is already mapped, no go */
3209 if (vm_map_lookup_entry(map, addrbos, &prev_entry)) {
3211 return (KERN_NO_SPACE);
3214 /* If we would blow our VMEM resource limit, no go */
3215 if (map->size + init_ssize > vmemlim) {
3217 return (KERN_NO_SPACE);
3221 * If we can't accomodate max_ssize in the current mapping, no go.
3222 * However, we need to be aware that subsequent user mappings might
3223 * map into the space we have reserved for stack, and currently this
3224 * space is not protected.
3226 * Hopefully we will at least detect this condition when we try to
3229 if ((prev_entry->next != &map->header) &&
3230 (prev_entry->next->start < addrbos + max_ssize)) {
3232 return (KERN_NO_SPACE);
3236 * We initially map a stack of only init_ssize. We will grow as
3237 * needed later. Depending on the orientation of the stack (i.e.
3238 * the grow direction) we either map at the top of the range, the
3239 * bottom of the range or in the middle.
3241 * Note: we would normally expect prot and max to be VM_PROT_ALL,
3242 * and cow to be 0. Possibly we should eliminate these as input
3243 * parameters, and just pass these values here in the insert call.
3245 if (orient == MAP_STACK_GROWS_DOWN)
3246 bot = addrbos + max_ssize - init_ssize;
3247 else if (orient == MAP_STACK_GROWS_UP)
3250 bot = round_page(addrbos + max_ssize/2 - init_ssize/2);
3251 top = bot + init_ssize;
3252 rv = vm_map_insert(map, NULL, 0, bot, top, prot, max, cow);
3254 /* Now set the avail_ssize amount. */
3255 if (rv == KERN_SUCCESS) {
3256 if (prev_entry != &map->header)
3257 vm_map_clip_end(map, prev_entry, bot);
3258 new_entry = prev_entry->next;
3259 if (new_entry->end != top || new_entry->start != bot)
3260 panic("Bad entry start/end for new stack entry");
3262 new_entry->avail_ssize = max_ssize - init_ssize;
3263 if (orient & MAP_STACK_GROWS_DOWN)
3264 new_entry->eflags |= MAP_ENTRY_GROWS_DOWN;
3265 if (orient & MAP_STACK_GROWS_UP)
3266 new_entry->eflags |= MAP_ENTRY_GROWS_UP;
3273 static int stack_guard_page = 0;
3274 TUNABLE_INT("security.bsd.stack_guard_page", &stack_guard_page);
3275 SYSCTL_INT(_security_bsd, OID_AUTO, stack_guard_page, CTLFLAG_RW,
3276 &stack_guard_page, 0,
3277 "Insert stack guard page ahead of the growable segments.");
3279 /* Attempts to grow a vm stack entry. Returns KERN_SUCCESS if the
3280 * desired address is already mapped, or if we successfully grow
3281 * the stack. Also returns KERN_SUCCESS if addr is outside the
3282 * stack range (this is strange, but preserves compatibility with
3283 * the grow function in vm_machdep.c).
3286 vm_map_growstack(struct proc *p, vm_offset_t addr)
3288 vm_map_entry_t next_entry, prev_entry;
3289 vm_map_entry_t new_entry, stack_entry;
3290 struct vmspace *vm = p->p_vmspace;
3291 vm_map_t map = &vm->vm_map;
3293 size_t grow_amount, max_grow;
3294 rlim_t stacklim, vmemlim;
3295 int is_procstack, rv;
3304 stacklim = lim_cur(p, RLIMIT_STACK);
3305 vmemlim = lim_cur(p, RLIMIT_VMEM);
3308 vm_map_lock_read(map);
3310 /* If addr is already in the entry range, no need to grow.*/
3311 if (vm_map_lookup_entry(map, addr, &prev_entry)) {
3312 vm_map_unlock_read(map);
3313 return (KERN_SUCCESS);
3316 next_entry = prev_entry->next;
3317 if (!(prev_entry->eflags & MAP_ENTRY_GROWS_UP)) {
3319 * This entry does not grow upwards. Since the address lies
3320 * beyond this entry, the next entry (if one exists) has to
3321 * be a downward growable entry. The entry list header is
3322 * never a growable entry, so it suffices to check the flags.
3324 if (!(next_entry->eflags & MAP_ENTRY_GROWS_DOWN)) {
3325 vm_map_unlock_read(map);
3326 return (KERN_SUCCESS);
3328 stack_entry = next_entry;
3331 * This entry grows upward. If the next entry does not at
3332 * least grow downwards, this is the entry we need to grow.
3333 * otherwise we have two possible choices and we have to
3336 if (next_entry->eflags & MAP_ENTRY_GROWS_DOWN) {
3338 * We have two choices; grow the entry closest to
3339 * the address to minimize the amount of growth.
3341 if (addr - prev_entry->end <= next_entry->start - addr)
3342 stack_entry = prev_entry;
3344 stack_entry = next_entry;
3346 stack_entry = prev_entry;
3349 if (stack_entry == next_entry) {
3350 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_DOWN, ("foo"));
3351 KASSERT(addr < stack_entry->start, ("foo"));
3352 end = (prev_entry != &map->header) ? prev_entry->end :
3353 stack_entry->start - stack_entry->avail_ssize;
3354 grow_amount = roundup(stack_entry->start - addr, PAGE_SIZE);
3355 max_grow = stack_entry->start - end;
3357 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_UP, ("foo"));
3358 KASSERT(addr >= stack_entry->end, ("foo"));
3359 end = (next_entry != &map->header) ? next_entry->start :
3360 stack_entry->end + stack_entry->avail_ssize;
3361 grow_amount = roundup(addr + 1 - stack_entry->end, PAGE_SIZE);
3362 max_grow = end - stack_entry->end;
3365 if (grow_amount > stack_entry->avail_ssize) {
3366 vm_map_unlock_read(map);
3367 return (KERN_NO_SPACE);
3371 * If there is no longer enough space between the entries nogo, and
3372 * adjust the available space. Note: this should only happen if the
3373 * user has mapped into the stack area after the stack was created,
3374 * and is probably an error.
3376 * This also effectively destroys any guard page the user might have
3377 * intended by limiting the stack size.
3379 if (grow_amount + (stack_guard_page ? PAGE_SIZE : 0) > max_grow) {
3380 if (vm_map_lock_upgrade(map))
3383 stack_entry->avail_ssize = max_grow;
3386 return (KERN_NO_SPACE);
3389 is_procstack = (addr >= (vm_offset_t)vm->vm_maxsaddr) ? 1 : 0;
3392 * If this is the main process stack, see if we're over the stack
3395 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) {
3396 vm_map_unlock_read(map);
3397 return (KERN_NO_SPACE);
3401 racct_set(p, RACCT_STACK, ctob(vm->vm_ssize) + grow_amount)) {
3403 vm_map_unlock_read(map);
3404 return (KERN_NO_SPACE);
3408 /* Round up the grow amount modulo SGROWSIZ */
3409 grow_amount = roundup (grow_amount, sgrowsiz);
3410 if (grow_amount > stack_entry->avail_ssize)
3411 grow_amount = stack_entry->avail_ssize;
3412 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) {
3413 grow_amount = trunc_page((vm_size_t)stacklim) -
3418 limit = racct_get_available(p, RACCT_STACK);
3420 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > limit))
3421 grow_amount = limit - ctob(vm->vm_ssize);
3424 /* If we would blow our VMEM resource limit, no go */
3425 if (map->size + grow_amount > vmemlim) {
3426 vm_map_unlock_read(map);
3431 if (racct_set(p, RACCT_VMEM, map->size + grow_amount)) {
3433 vm_map_unlock_read(map);
3439 if (vm_map_lock_upgrade(map))
3442 if (stack_entry == next_entry) {
3446 /* Get the preliminary new entry start value */
3447 addr = stack_entry->start - grow_amount;
3450 * If this puts us into the previous entry, cut back our
3451 * growth to the available space. Also, see the note above.
3454 stack_entry->avail_ssize = max_grow;
3456 if (stack_guard_page)
3460 rv = vm_map_insert(map, NULL, 0, addr, stack_entry->start,
3461 p->p_sysent->sv_stackprot, VM_PROT_ALL, 0);
3463 /* Adjust the available stack space by the amount we grew. */
3464 if (rv == KERN_SUCCESS) {
3465 if (prev_entry != &map->header)
3466 vm_map_clip_end(map, prev_entry, addr);
3467 new_entry = prev_entry->next;
3468 KASSERT(new_entry == stack_entry->prev, ("foo"));
3469 KASSERT(new_entry->end == stack_entry->start, ("foo"));
3470 KASSERT(new_entry->start == addr, ("foo"));
3471 grow_amount = new_entry->end - new_entry->start;
3472 new_entry->avail_ssize = stack_entry->avail_ssize -
3474 stack_entry->eflags &= ~MAP_ENTRY_GROWS_DOWN;
3475 new_entry->eflags |= MAP_ENTRY_GROWS_DOWN;
3481 addr = stack_entry->end + grow_amount;
3484 * If this puts us into the next entry, cut back our growth
3485 * to the available space. Also, see the note above.
3488 stack_entry->avail_ssize = end - stack_entry->end;
3490 if (stack_guard_page)
3494 grow_amount = addr - stack_entry->end;
3495 cred = stack_entry->cred;
3496 if (cred == NULL && stack_entry->object.vm_object != NULL)
3497 cred = stack_entry->object.vm_object->cred;
3498 if (cred != NULL && !swap_reserve_by_cred(grow_amount, cred))
3500 /* Grow the underlying object if applicable. */
3501 else if (stack_entry->object.vm_object == NULL ||
3502 vm_object_coalesce(stack_entry->object.vm_object,
3503 stack_entry->offset,
3504 (vm_size_t)(stack_entry->end - stack_entry->start),
3505 (vm_size_t)grow_amount, cred != NULL)) {
3506 map->size += (addr - stack_entry->end);
3507 /* Update the current entry. */
3508 stack_entry->end = addr;
3509 stack_entry->avail_ssize -= grow_amount;
3510 vm_map_entry_resize_free(map, stack_entry);
3513 if (next_entry != &map->header)
3514 vm_map_clip_start(map, next_entry, addr);
3519 if (rv == KERN_SUCCESS && is_procstack)
3520 vm->vm_ssize += btoc(grow_amount);
3525 * Heed the MAP_WIREFUTURE flag if it was set for this process.
3527 if (rv == KERN_SUCCESS && (map->flags & MAP_WIREFUTURE)) {
3529 (stack_entry == next_entry) ? addr : addr - grow_amount,
3530 (stack_entry == next_entry) ? stack_entry->start : addr,
3531 (p->p_flag & P_SYSTEM)
3532 ? VM_MAP_WIRE_SYSTEM|VM_MAP_WIRE_NOHOLES
3533 : VM_MAP_WIRE_USER|VM_MAP_WIRE_NOHOLES);
3537 if (rv != KERN_SUCCESS) {
3539 error = racct_set(p, RACCT_VMEM, map->size);
3540 KASSERT(error == 0, ("decreasing RACCT_VMEM failed"));
3541 error = racct_set(p, RACCT_STACK, ctob(vm->vm_ssize));
3542 KASSERT(error == 0, ("decreasing RACCT_STACK failed"));
3550 * Unshare the specified VM space for exec. If other processes are
3551 * mapped to it, then create a new one. The new vmspace is null.
3554 vmspace_exec(struct proc *p, vm_offset_t minuser, vm_offset_t maxuser)
3556 struct vmspace *oldvmspace = p->p_vmspace;
3557 struct vmspace *newvmspace;
3559 newvmspace = vmspace_alloc(minuser, maxuser);
3560 if (newvmspace == NULL)
3562 newvmspace->vm_swrss = oldvmspace->vm_swrss;
3564 * This code is written like this for prototype purposes. The
3565 * goal is to avoid running down the vmspace here, but let the
3566 * other process's that are still using the vmspace to finally
3567 * run it down. Even though there is little or no chance of blocking
3568 * here, it is a good idea to keep this form for future mods.
3570 PROC_VMSPACE_LOCK(p);
3571 p->p_vmspace = newvmspace;
3572 PROC_VMSPACE_UNLOCK(p);
3573 if (p == curthread->td_proc)
3574 pmap_activate(curthread);
3575 vmspace_free(oldvmspace);
3580 * Unshare the specified VM space for forcing COW. This
3581 * is called by rfork, for the (RFMEM|RFPROC) == 0 case.
3584 vmspace_unshare(struct proc *p)
3586 struct vmspace *oldvmspace = p->p_vmspace;
3587 struct vmspace *newvmspace;
3588 vm_ooffset_t fork_charge;
3590 if (oldvmspace->vm_refcnt == 1)
3593 newvmspace = vmspace_fork(oldvmspace, &fork_charge);
3594 if (newvmspace == NULL)
3596 if (!swap_reserve_by_cred(fork_charge, p->p_ucred)) {
3597 vmspace_free(newvmspace);
3600 PROC_VMSPACE_LOCK(p);
3601 p->p_vmspace = newvmspace;
3602 PROC_VMSPACE_UNLOCK(p);
3603 if (p == curthread->td_proc)
3604 pmap_activate(curthread);
3605 vmspace_free(oldvmspace);
3612 * Finds the VM object, offset, and
3613 * protection for a given virtual address in the
3614 * specified map, assuming a page fault of the
3617 * Leaves the map in question locked for read; return
3618 * values are guaranteed until a vm_map_lookup_done
3619 * call is performed. Note that the map argument
3620 * is in/out; the returned map must be used in
3621 * the call to vm_map_lookup_done.
3623 * A handle (out_entry) is returned for use in
3624 * vm_map_lookup_done, to make that fast.
3626 * If a lookup is requested with "write protection"
3627 * specified, the map may be changed to perform virtual
3628 * copying operations, although the data referenced will
3632 vm_map_lookup(vm_map_t *var_map, /* IN/OUT */
3634 vm_prot_t fault_typea,
3635 vm_map_entry_t *out_entry, /* OUT */
3636 vm_object_t *object, /* OUT */
3637 vm_pindex_t *pindex, /* OUT */
3638 vm_prot_t *out_prot, /* OUT */
3639 boolean_t *wired) /* OUT */
3641 vm_map_entry_t entry;
3642 vm_map_t map = *var_map;
3644 vm_prot_t fault_type = fault_typea;
3645 vm_object_t eobject;
3651 vm_map_lock_read(map);
3654 * Lookup the faulting address.
3656 if (!vm_map_lookup_entry(map, vaddr, out_entry)) {
3657 vm_map_unlock_read(map);
3658 return (KERN_INVALID_ADDRESS);
3666 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
3667 vm_map_t old_map = map;
3669 *var_map = map = entry->object.sub_map;
3670 vm_map_unlock_read(old_map);
3675 * Check whether this task is allowed to have this page.
3677 prot = entry->protection;
3678 fault_type &= (VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE);
3679 if ((fault_type & prot) != fault_type || prot == VM_PROT_NONE) {
3680 vm_map_unlock_read(map);
3681 return (KERN_PROTECTION_FAILURE);
3683 if ((entry->eflags & MAP_ENTRY_USER_WIRED) &&
3684 (entry->eflags & MAP_ENTRY_COW) &&
3685 (fault_type & VM_PROT_WRITE)) {
3686 vm_map_unlock_read(map);
3687 return (KERN_PROTECTION_FAILURE);
3691 * If this page is not pageable, we have to get it for all possible
3694 *wired = (entry->wired_count != 0);
3696 fault_type = entry->protection;
3697 size = entry->end - entry->start;
3699 * If the entry was copy-on-write, we either ...
3701 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
3703 * If we want to write the page, we may as well handle that
3704 * now since we've got the map locked.
3706 * If we don't need to write the page, we just demote the
3707 * permissions allowed.
3709 if ((fault_type & VM_PROT_WRITE) != 0 ||
3710 (fault_typea & VM_PROT_COPY) != 0) {
3712 * Make a new object, and place it in the object
3713 * chain. Note that no new references have appeared
3714 * -- one just moved from the map to the new
3717 if (vm_map_lock_upgrade(map))
3720 if (entry->cred == NULL) {
3722 * The debugger owner is charged for
3725 cred = curthread->td_ucred;
3727 if (!swap_reserve_by_cred(size, cred)) {
3730 return (KERN_RESOURCE_SHORTAGE);
3734 vm_object_shadow(&entry->object.vm_object,
3735 &entry->offset, size);
3736 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
3737 eobject = entry->object.vm_object;
3738 if (eobject->cred != NULL) {
3740 * The object was not shadowed.
3742 swap_release_by_cred(size, entry->cred);
3743 crfree(entry->cred);
3745 } else if (entry->cred != NULL) {
3746 VM_OBJECT_LOCK(eobject);
3747 eobject->cred = entry->cred;
3748 eobject->charge = size;
3749 VM_OBJECT_UNLOCK(eobject);
3753 vm_map_lock_downgrade(map);
3756 * We're attempting to read a copy-on-write page --
3757 * don't allow writes.
3759 prot &= ~VM_PROT_WRITE;
3764 * Create an object if necessary.
3766 if (entry->object.vm_object == NULL &&
3768 if (vm_map_lock_upgrade(map))
3770 entry->object.vm_object = vm_object_allocate(OBJT_DEFAULT,
3773 if (entry->cred != NULL) {
3774 VM_OBJECT_LOCK(entry->object.vm_object);
3775 entry->object.vm_object->cred = entry->cred;
3776 entry->object.vm_object->charge = size;
3777 VM_OBJECT_UNLOCK(entry->object.vm_object);
3780 vm_map_lock_downgrade(map);
3784 * Return the object/offset from this entry. If the entry was
3785 * copy-on-write or empty, it has been fixed up.
3787 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
3788 *object = entry->object.vm_object;
3791 return (KERN_SUCCESS);
3795 * vm_map_lookup_locked:
3797 * Lookup the faulting address. A version of vm_map_lookup that returns
3798 * KERN_FAILURE instead of blocking on map lock or memory allocation.
3801 vm_map_lookup_locked(vm_map_t *var_map, /* IN/OUT */
3803 vm_prot_t fault_typea,
3804 vm_map_entry_t *out_entry, /* OUT */
3805 vm_object_t *object, /* OUT */
3806 vm_pindex_t *pindex, /* OUT */
3807 vm_prot_t *out_prot, /* OUT */
3808 boolean_t *wired) /* OUT */
3810 vm_map_entry_t entry;
3811 vm_map_t map = *var_map;
3813 vm_prot_t fault_type = fault_typea;
3816 * Lookup the faulting address.
3818 if (!vm_map_lookup_entry(map, vaddr, out_entry))
3819 return (KERN_INVALID_ADDRESS);
3824 * Fail if the entry refers to a submap.
3826 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP)
3827 return (KERN_FAILURE);
3830 * Check whether this task is allowed to have this page.
3832 prot = entry->protection;
3833 fault_type &= VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE;
3834 if ((fault_type & prot) != fault_type)
3835 return (KERN_PROTECTION_FAILURE);
3836 if ((entry->eflags & MAP_ENTRY_USER_WIRED) &&
3837 (entry->eflags & MAP_ENTRY_COW) &&
3838 (fault_type & VM_PROT_WRITE))
3839 return (KERN_PROTECTION_FAILURE);
3842 * If this page is not pageable, we have to get it for all possible
3845 *wired = (entry->wired_count != 0);
3847 fault_type = entry->protection;
3849 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
3851 * Fail if the entry was copy-on-write for a write fault.
3853 if (fault_type & VM_PROT_WRITE)
3854 return (KERN_FAILURE);
3856 * We're attempting to read a copy-on-write page --
3857 * don't allow writes.
3859 prot &= ~VM_PROT_WRITE;
3863 * Fail if an object should be created.
3865 if (entry->object.vm_object == NULL && !map->system_map)
3866 return (KERN_FAILURE);
3869 * Return the object/offset from this entry. If the entry was
3870 * copy-on-write or empty, it has been fixed up.
3872 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
3873 *object = entry->object.vm_object;
3876 return (KERN_SUCCESS);
3880 * vm_map_lookup_done:
3882 * Releases locks acquired by a vm_map_lookup
3883 * (according to the handle returned by that lookup).
3886 vm_map_lookup_done(vm_map_t map, vm_map_entry_t entry)
3889 * Unlock the main-level map
3891 vm_map_unlock_read(map);
3894 #include "opt_ddb.h"
3896 #include <sys/kernel.h>
3898 #include <ddb/ddb.h>
3901 * vm_map_print: [ debug ]
3903 DB_SHOW_COMMAND(map, vm_map_print)
3906 /* XXX convert args. */
3907 vm_map_t map = (vm_map_t)addr;
3908 boolean_t full = have_addr;
3910 vm_map_entry_t entry;
3912 db_iprintf("Task map %p: pmap=%p, nentries=%d, version=%u\n",
3914 (void *)map->pmap, map->nentries, map->timestamp);
3917 if (!full && db_indent)
3921 for (entry = map->header.next; entry != &map->header;
3922 entry = entry->next) {
3923 db_iprintf("map entry %p: start=%p, end=%p\n",
3924 (void *)entry, (void *)entry->start, (void *)entry->end);
3927 static char *inheritance_name[4] =
3928 {"share", "copy", "none", "donate_copy"};
3930 db_iprintf(" prot=%x/%x/%s",
3932 entry->max_protection,
3933 inheritance_name[(int)(unsigned char)entry->inheritance]);
3934 if (entry->wired_count != 0)
3935 db_printf(", wired");
3937 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
3938 db_printf(", share=%p, offset=0x%jx\n",
3939 (void *)entry->object.sub_map,
3940 (uintmax_t)entry->offset);
3942 if ((entry->prev == &map->header) ||
3943 (entry->prev->object.sub_map !=
3944 entry->object.sub_map)) {
3946 vm_map_print((db_expr_t)(intptr_t)
3947 entry->object.sub_map,
3948 full, 0, (char *)0);
3952 if (entry->cred != NULL)
3953 db_printf(", ruid %d", entry->cred->cr_ruid);
3954 db_printf(", object=%p, offset=0x%jx",
3955 (void *)entry->object.vm_object,
3956 (uintmax_t)entry->offset);
3957 if (entry->object.vm_object && entry->object.vm_object->cred)
3958 db_printf(", obj ruid %d charge %jx",
3959 entry->object.vm_object->cred->cr_ruid,
3960 (uintmax_t)entry->object.vm_object->charge);
3961 if (entry->eflags & MAP_ENTRY_COW)
3962 db_printf(", copy (%s)",
3963 (entry->eflags & MAP_ENTRY_NEEDS_COPY) ? "needed" : "done");
3967 if ((entry->prev == &map->header) ||
3968 (entry->prev->object.vm_object !=
3969 entry->object.vm_object)) {
3971 vm_object_print((db_expr_t)(intptr_t)
3972 entry->object.vm_object,
3973 full, 0, (char *)0);
3985 DB_SHOW_COMMAND(procvm, procvm)
3990 p = (struct proc *) addr;
3995 db_printf("p = %p, vmspace = %p, map = %p, pmap = %p\n",
3996 (void *)p, (void *)p->p_vmspace, (void *)&p->p_vmspace->vm_map,
3997 (void *)vmspace_pmap(p->p_vmspace));
3999 vm_map_print((db_expr_t)(intptr_t)&p->p_vmspace->vm_map, 1, 0, NULL);