2 * Copyright (c) 1991, 1993
3 * The Regents of the University of California. All rights reserved.
5 * This code is derived from software contributed to Berkeley by
6 * The Mach Operating System project at Carnegie-Mellon University.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 4. Neither the name of the University nor the names of its contributors
17 * may be used to endorse or promote products derived from this software
18 * without specific prior written permission.
20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * from: @(#)vm_map.c 8.3 (Berkeley) 1/12/94
35 * Copyright (c) 1987, 1990 Carnegie-Mellon University.
36 * All rights reserved.
38 * Authors: Avadis Tevanian, Jr., Michael Wayne Young
40 * Permission to use, copy, modify and distribute this software and
41 * its documentation is hereby granted, provided that both the copyright
42 * notice and this permission notice appear in all copies of the
43 * software, derivative works or modified versions, and any portions
44 * thereof, and that both notices appear in supporting documentation.
46 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
47 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
48 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
50 * Carnegie Mellon requests users of this software to return to
52 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
53 * School of Computer Science
54 * Carnegie Mellon University
55 * Pittsburgh PA 15213-3890
57 * any improvements or extensions that they make and grant Carnegie the
58 * rights to redistribute these changes.
62 * Virtual memory mapping module.
65 #include <sys/cdefs.h>
66 __FBSDID("$FreeBSD$");
68 #include <sys/param.h>
69 #include <sys/systm.h>
70 #include <sys/kernel.h>
73 #include <sys/mutex.h>
75 #include <sys/vmmeter.h>
77 #include <sys/vnode.h>
78 #include <sys/racct.h>
79 #include <sys/resourcevar.h>
80 #include <sys/rwlock.h>
82 #include <sys/sysctl.h>
83 #include <sys/sysent.h>
87 #include <vm/vm_param.h>
89 #include <vm/vm_map.h>
90 #include <vm/vm_page.h>
91 #include <vm/vm_object.h>
92 #include <vm/vm_pager.h>
93 #include <vm/vm_kern.h>
94 #include <vm/vm_extern.h>
95 #include <vm/vnode_pager.h>
96 #include <vm/swap_pager.h>
100 * Virtual memory maps provide for the mapping, protection,
101 * and sharing of virtual memory objects. In addition,
102 * this module provides for an efficient virtual copy of
103 * memory from one map to another.
105 * Synchronization is required prior to most operations.
107 * Maps consist of an ordered doubly-linked list of simple
108 * entries; a self-adjusting binary search tree of these
109 * entries is used to speed up lookups.
111 * Since portions of maps are specified by start/end addresses,
112 * which may not align with existing map entries, all
113 * routines merely "clip" entries to these start/end values.
114 * [That is, an entry is split into two, bordering at a
115 * start or end value.] Note that these clippings may not
116 * always be necessary (as the two resulting entries are then
117 * not changed); however, the clipping is done for convenience.
119 * As mentioned above, virtual copy operations are performed
120 * by copying VM object references from one map to
121 * another, and then marking both regions as copy-on-write.
124 static struct mtx map_sleep_mtx;
125 static uma_zone_t mapentzone;
126 static uma_zone_t kmapentzone;
127 static uma_zone_t mapzone;
128 static uma_zone_t vmspace_zone;
129 static int vmspace_zinit(void *mem, int size, int flags);
130 static int vm_map_zinit(void *mem, int ize, int flags);
131 static void _vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min,
133 static void vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map);
134 static void vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry);
136 static void vm_map_zdtor(void *mem, int size, void *arg);
137 static void vmspace_zdtor(void *mem, int size, void *arg);
140 #define ENTRY_CHARGED(e) ((e)->cred != NULL || \
141 ((e)->object.vm_object != NULL && (e)->object.vm_object->cred != NULL && \
142 !((e)->eflags & MAP_ENTRY_NEEDS_COPY)))
145 * PROC_VMSPACE_{UN,}LOCK() can be a noop as long as vmspaces are type
148 #define PROC_VMSPACE_LOCK(p) do { } while (0)
149 #define PROC_VMSPACE_UNLOCK(p) do { } while (0)
152 * VM_MAP_RANGE_CHECK: [ internal use only ]
154 * Asserts that the starting and ending region
155 * addresses fall within the valid range of the map.
157 #define VM_MAP_RANGE_CHECK(map, start, end) \
159 if (start < vm_map_min(map)) \
160 start = vm_map_min(map); \
161 if (end > vm_map_max(map)) \
162 end = vm_map_max(map); \
170 * Initialize the vm_map module. Must be called before
171 * any other vm_map routines.
173 * Map and entry structures are allocated from the general
174 * purpose memory pool with some exceptions:
176 * - The kernel map and kmem submap are allocated statically.
177 * - Kernel map entries are allocated out of a static pool.
179 * These restrictions are necessary since malloc() uses the
180 * maps and requires map entries.
186 mtx_init(&map_sleep_mtx, "vm map sleep mutex", NULL, MTX_DEF);
187 mapzone = uma_zcreate("MAP", sizeof(struct vm_map), NULL,
193 vm_map_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
194 uma_prealloc(mapzone, MAX_KMAP);
195 kmapentzone = uma_zcreate("KMAP ENTRY", sizeof(struct vm_map_entry),
196 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR,
197 UMA_ZONE_MTXCLASS | UMA_ZONE_VM);
198 mapentzone = uma_zcreate("MAP ENTRY", sizeof(struct vm_map_entry),
199 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
200 vmspace_zone = uma_zcreate("VMSPACE", sizeof(struct vmspace), NULL,
206 vmspace_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
210 vmspace_zinit(void *mem, int size, int flags)
214 vm = (struct vmspace *)mem;
216 vm->vm_map.pmap = NULL;
217 (void)vm_map_zinit(&vm->vm_map, sizeof(vm->vm_map), flags);
218 PMAP_LOCK_INIT(vmspace_pmap(vm));
223 vm_map_zinit(void *mem, int size, int flags)
228 memset(map, 0, sizeof(*map));
229 mtx_init(&map->system_mtx, "vm map (system)", NULL, MTX_DEF | MTX_DUPOK);
230 sx_init(&map->lock, "vm map (user)");
236 vmspace_zdtor(void *mem, int size, void *arg)
240 vm = (struct vmspace *)mem;
242 vm_map_zdtor(&vm->vm_map, sizeof(vm->vm_map), arg);
245 vm_map_zdtor(void *mem, int size, void *arg)
250 KASSERT(map->nentries == 0,
251 ("map %p nentries == %d on free.",
252 map, map->nentries));
253 KASSERT(map->size == 0,
254 ("map %p size == %lu on free.",
255 map, (unsigned long)map->size));
257 #endif /* INVARIANTS */
260 * Allocate a vmspace structure, including a vm_map and pmap,
261 * and initialize those structures. The refcnt is set to 1.
263 * If 'pinit' is NULL then the embedded pmap is initialized via pmap_pinit().
266 vmspace_alloc(vm_offset_t min, vm_offset_t max, pmap_pinit_t pinit)
270 vm = uma_zalloc(vmspace_zone, M_WAITOK);
272 KASSERT(vm->vm_map.pmap == NULL, ("vm_map.pmap must be NULL"));
277 if (!pinit(vmspace_pmap(vm))) {
278 uma_zfree(vmspace_zone, vm);
281 CTR1(KTR_VM, "vmspace_alloc: %p", vm);
282 _vm_map_init(&vm->vm_map, vmspace_pmap(vm), min, max);
296 vmspace_container_reset(struct proc *p)
301 racct_set(p, RACCT_DATA, 0);
302 racct_set(p, RACCT_STACK, 0);
303 racct_set(p, RACCT_RSS, 0);
304 racct_set(p, RACCT_MEMLOCK, 0);
305 racct_set(p, RACCT_VMEM, 0);
311 vmspace_dofree(struct vmspace *vm)
314 CTR1(KTR_VM, "vmspace_free: %p", vm);
317 * Make sure any SysV shm is freed, it might not have been in
323 * Lock the map, to wait out all other references to it.
324 * Delete all of the mappings and pages they hold, then call
325 * the pmap module to reclaim anything left.
327 (void)vm_map_remove(&vm->vm_map, vm->vm_map.min_offset,
328 vm->vm_map.max_offset);
330 pmap_release(vmspace_pmap(vm));
331 vm->vm_map.pmap = NULL;
332 uma_zfree(vmspace_zone, vm);
336 vmspace_free(struct vmspace *vm)
339 if (vm->vm_refcnt == 0)
340 panic("vmspace_free: attempt to free already freed vmspace");
342 if (atomic_fetchadd_int(&vm->vm_refcnt, -1) == 1)
347 vmspace_exitfree(struct proc *p)
351 PROC_VMSPACE_LOCK(p);
354 PROC_VMSPACE_UNLOCK(p);
355 KASSERT(vm == &vmspace0, ("vmspace_exitfree: wrong vmspace"));
360 vmspace_exit(struct thread *td)
367 * Release user portion of address space.
368 * This releases references to vnodes,
369 * which could cause I/O if the file has been unlinked.
370 * Need to do this early enough that we can still sleep.
372 * The last exiting process to reach this point releases as
373 * much of the environment as it can. vmspace_dofree() is the
374 * slower fallback in case another process had a temporary
375 * reference to the vmspace.
380 atomic_add_int(&vmspace0.vm_refcnt, 1);
382 refcnt = vm->vm_refcnt;
383 if (refcnt > 1 && p->p_vmspace != &vmspace0) {
384 /* Switch now since other proc might free vmspace */
385 PROC_VMSPACE_LOCK(p);
386 p->p_vmspace = &vmspace0;
387 PROC_VMSPACE_UNLOCK(p);
390 } while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt - 1));
392 if (p->p_vmspace != vm) {
393 /* vmspace not yet freed, switch back */
394 PROC_VMSPACE_LOCK(p);
396 PROC_VMSPACE_UNLOCK(p);
399 pmap_remove_pages(vmspace_pmap(vm));
400 /* Switch now since this proc will free vmspace */
401 PROC_VMSPACE_LOCK(p);
402 p->p_vmspace = &vmspace0;
403 PROC_VMSPACE_UNLOCK(p);
407 vmspace_container_reset(p);
410 /* Acquire reference to vmspace owned by another process. */
413 vmspace_acquire_ref(struct proc *p)
418 PROC_VMSPACE_LOCK(p);
421 PROC_VMSPACE_UNLOCK(p);
425 refcnt = vm->vm_refcnt;
426 if (refcnt <= 0) { /* Avoid 0->1 transition */
427 PROC_VMSPACE_UNLOCK(p);
430 } while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt + 1));
431 if (vm != p->p_vmspace) {
432 PROC_VMSPACE_UNLOCK(p);
436 PROC_VMSPACE_UNLOCK(p);
441 _vm_map_lock(vm_map_t map, const char *file, int line)
445 mtx_lock_flags_(&map->system_mtx, 0, file, line);
447 sx_xlock_(&map->lock, file, line);
452 vm_map_process_deferred(void)
455 vm_map_entry_t entry, next;
459 entry = td->td_map_def_user;
460 td->td_map_def_user = NULL;
461 while (entry != NULL) {
463 if ((entry->eflags & MAP_ENTRY_VN_WRITECNT) != 0) {
465 * Decrement the object's writemappings and
466 * possibly the vnode's v_writecount.
468 KASSERT((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0,
469 ("Submap with writecount"));
470 object = entry->object.vm_object;
471 KASSERT(object != NULL, ("No object for writecount"));
472 vnode_pager_release_writecount(object, entry->start,
475 vm_map_entry_deallocate(entry, FALSE);
481 _vm_map_unlock(vm_map_t map, const char *file, int line)
485 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
487 sx_xunlock_(&map->lock, file, line);
488 vm_map_process_deferred();
493 _vm_map_lock_read(vm_map_t map, const char *file, int line)
497 mtx_lock_flags_(&map->system_mtx, 0, file, line);
499 sx_slock_(&map->lock, file, line);
503 _vm_map_unlock_read(vm_map_t map, const char *file, int line)
507 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
509 sx_sunlock_(&map->lock, file, line);
510 vm_map_process_deferred();
515 _vm_map_trylock(vm_map_t map, const char *file, int line)
519 error = map->system_map ?
520 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) :
521 !sx_try_xlock_(&map->lock, file, line);
528 _vm_map_trylock_read(vm_map_t map, const char *file, int line)
532 error = map->system_map ?
533 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) :
534 !sx_try_slock_(&map->lock, file, line);
539 * _vm_map_lock_upgrade: [ internal use only ]
541 * Tries to upgrade a read (shared) lock on the specified map to a write
542 * (exclusive) lock. Returns the value "0" if the upgrade succeeds and a
543 * non-zero value if the upgrade fails. If the upgrade fails, the map is
544 * returned without a read or write lock held.
546 * Requires that the map be read locked.
549 _vm_map_lock_upgrade(vm_map_t map, const char *file, int line)
551 unsigned int last_timestamp;
553 if (map->system_map) {
554 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
556 if (!sx_try_upgrade_(&map->lock, file, line)) {
557 last_timestamp = map->timestamp;
558 sx_sunlock_(&map->lock, file, line);
559 vm_map_process_deferred();
561 * If the map's timestamp does not change while the
562 * map is unlocked, then the upgrade succeeds.
564 sx_xlock_(&map->lock, file, line);
565 if (last_timestamp != map->timestamp) {
566 sx_xunlock_(&map->lock, file, line);
576 _vm_map_lock_downgrade(vm_map_t map, const char *file, int line)
579 if (map->system_map) {
580 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
582 sx_downgrade_(&map->lock, file, line);
588 * Returns a non-zero value if the caller holds a write (exclusive) lock
589 * on the specified map and the value "0" otherwise.
592 vm_map_locked(vm_map_t map)
596 return (mtx_owned(&map->system_mtx));
598 return (sx_xlocked(&map->lock));
603 _vm_map_assert_locked(vm_map_t map, const char *file, int line)
607 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
609 sx_assert_(&map->lock, SA_XLOCKED, file, line);
612 #define VM_MAP_ASSERT_LOCKED(map) \
613 _vm_map_assert_locked(map, LOCK_FILE, LOCK_LINE)
615 #define VM_MAP_ASSERT_LOCKED(map)
619 * _vm_map_unlock_and_wait:
621 * Atomically releases the lock on the specified map and puts the calling
622 * thread to sleep. The calling thread will remain asleep until either
623 * vm_map_wakeup() is performed on the map or the specified timeout is
626 * WARNING! This function does not perform deferred deallocations of
627 * objects and map entries. Therefore, the calling thread is expected to
628 * reacquire the map lock after reawakening and later perform an ordinary
629 * unlock operation, such as vm_map_unlock(), before completing its
630 * operation on the map.
633 _vm_map_unlock_and_wait(vm_map_t map, int timo, const char *file, int line)
636 mtx_lock(&map_sleep_mtx);
638 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
640 sx_xunlock_(&map->lock, file, line);
641 return (msleep(&map->root, &map_sleep_mtx, PDROP | PVM, "vmmaps",
648 * Awaken any threads that have slept on the map using
649 * vm_map_unlock_and_wait().
652 vm_map_wakeup(vm_map_t map)
656 * Acquire and release map_sleep_mtx to prevent a wakeup()
657 * from being performed (and lost) between the map unlock
658 * and the msleep() in _vm_map_unlock_and_wait().
660 mtx_lock(&map_sleep_mtx);
661 mtx_unlock(&map_sleep_mtx);
666 vm_map_busy(vm_map_t map)
669 VM_MAP_ASSERT_LOCKED(map);
674 vm_map_unbusy(vm_map_t map)
677 VM_MAP_ASSERT_LOCKED(map);
678 KASSERT(map->busy, ("vm_map_unbusy: not busy"));
679 if (--map->busy == 0 && (map->flags & MAP_BUSY_WAKEUP)) {
680 vm_map_modflags(map, 0, MAP_BUSY_WAKEUP);
686 vm_map_wait_busy(vm_map_t map)
689 VM_MAP_ASSERT_LOCKED(map);
691 vm_map_modflags(map, MAP_BUSY_WAKEUP, 0);
693 msleep(&map->busy, &map->system_mtx, 0, "mbusy", 0);
695 sx_sleep(&map->busy, &map->lock, 0, "mbusy", 0);
701 vmspace_resident_count(struct vmspace *vmspace)
703 return pmap_resident_count(vmspace_pmap(vmspace));
709 * Creates and returns a new empty VM map with
710 * the given physical map structure, and having
711 * the given lower and upper address bounds.
714 vm_map_create(pmap_t pmap, vm_offset_t min, vm_offset_t max)
718 result = uma_zalloc(mapzone, M_WAITOK);
719 CTR1(KTR_VM, "vm_map_create: %p", result);
720 _vm_map_init(result, pmap, min, max);
725 * Initialize an existing vm_map structure
726 * such as that in the vmspace structure.
729 _vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max)
732 map->header.next = map->header.prev = &map->header;
733 map->needs_wakeup = FALSE;
736 map->min_offset = min;
737 map->max_offset = max;
745 vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max)
748 _vm_map_init(map, pmap, min, max);
749 mtx_init(&map->system_mtx, "system map", NULL, MTX_DEF | MTX_DUPOK);
750 sx_init(&map->lock, "user map");
754 * vm_map_entry_dispose: [ internal use only ]
756 * Inverse of vm_map_entry_create.
759 vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry)
761 uma_zfree(map->system_map ? kmapentzone : mapentzone, entry);
765 * vm_map_entry_create: [ internal use only ]
767 * Allocates a VM map entry for insertion.
768 * No entry fields are filled in.
770 static vm_map_entry_t
771 vm_map_entry_create(vm_map_t map)
773 vm_map_entry_t new_entry;
776 new_entry = uma_zalloc(kmapentzone, M_NOWAIT);
778 new_entry = uma_zalloc(mapentzone, M_WAITOK);
779 if (new_entry == NULL)
780 panic("vm_map_entry_create: kernel resources exhausted");
785 * vm_map_entry_set_behavior:
787 * Set the expected access behavior, either normal, random, or
791 vm_map_entry_set_behavior(vm_map_entry_t entry, u_char behavior)
793 entry->eflags = (entry->eflags & ~MAP_ENTRY_BEHAV_MASK) |
794 (behavior & MAP_ENTRY_BEHAV_MASK);
798 * vm_map_entry_set_max_free:
800 * Set the max_free field in a vm_map_entry.
803 vm_map_entry_set_max_free(vm_map_entry_t entry)
806 entry->max_free = entry->adj_free;
807 if (entry->left != NULL && entry->left->max_free > entry->max_free)
808 entry->max_free = entry->left->max_free;
809 if (entry->right != NULL && entry->right->max_free > entry->max_free)
810 entry->max_free = entry->right->max_free;
814 * vm_map_entry_splay:
816 * The Sleator and Tarjan top-down splay algorithm with the
817 * following variation. Max_free must be computed bottom-up, so
818 * on the downward pass, maintain the left and right spines in
819 * reverse order. Then, make a second pass up each side to fix
820 * the pointers and compute max_free. The time bound is O(log n)
823 * The new root is the vm_map_entry containing "addr", or else an
824 * adjacent entry (lower or higher) if addr is not in the tree.
826 * The map must be locked, and leaves it so.
828 * Returns: the new root.
830 static vm_map_entry_t
831 vm_map_entry_splay(vm_offset_t addr, vm_map_entry_t root)
833 vm_map_entry_t llist, rlist;
834 vm_map_entry_t ltree, rtree;
837 /* Special case of empty tree. */
842 * Pass One: Splay down the tree until we find addr or a NULL
843 * pointer where addr would go. llist and rlist are the two
844 * sides in reverse order (bottom-up), with llist linked by
845 * the right pointer and rlist linked by the left pointer in
846 * the vm_map_entry. Wait until Pass Two to set max_free on
852 /* root is never NULL in here. */
853 if (addr < root->start) {
857 if (addr < y->start && y->left != NULL) {
858 /* Rotate right and put y on rlist. */
859 root->left = y->right;
861 vm_map_entry_set_max_free(root);
866 /* Put root on rlist. */
871 } else if (addr >= root->end) {
875 if (addr >= y->end && y->right != NULL) {
876 /* Rotate left and put y on llist. */
877 root->right = y->left;
879 vm_map_entry_set_max_free(root);
884 /* Put root on llist. */
894 * Pass Two: Walk back up the two spines, flip the pointers
895 * and set max_free. The subtrees of the root go at the
896 * bottom of llist and rlist.
899 while (llist != NULL) {
901 llist->right = ltree;
902 vm_map_entry_set_max_free(llist);
907 while (rlist != NULL) {
910 vm_map_entry_set_max_free(rlist);
916 * Final assembly: add ltree and rtree as subtrees of root.
920 vm_map_entry_set_max_free(root);
926 * vm_map_entry_{un,}link:
928 * Insert/remove entries from maps.
931 vm_map_entry_link(vm_map_t map,
932 vm_map_entry_t after_where,
933 vm_map_entry_t entry)
937 "vm_map_entry_link: map %p, nentries %d, entry %p, after %p", map,
938 map->nentries, entry, after_where);
939 VM_MAP_ASSERT_LOCKED(map);
941 entry->prev = after_where;
942 entry->next = after_where->next;
943 entry->next->prev = entry;
944 after_where->next = entry;
946 if (after_where != &map->header) {
947 if (after_where != map->root)
948 vm_map_entry_splay(after_where->start, map->root);
949 entry->right = after_where->right;
950 entry->left = after_where;
951 after_where->right = NULL;
952 after_where->adj_free = entry->start - after_where->end;
953 vm_map_entry_set_max_free(after_where);
955 entry->right = map->root;
958 entry->adj_free = (entry->next == &map->header ? map->max_offset :
959 entry->next->start) - entry->end;
960 vm_map_entry_set_max_free(entry);
965 vm_map_entry_unlink(vm_map_t map,
966 vm_map_entry_t entry)
968 vm_map_entry_t next, prev, root;
970 VM_MAP_ASSERT_LOCKED(map);
971 if (entry != map->root)
972 vm_map_entry_splay(entry->start, map->root);
973 if (entry->left == NULL)
976 root = vm_map_entry_splay(entry->start, entry->left);
977 root->right = entry->right;
978 root->adj_free = (entry->next == &map->header ? map->max_offset :
979 entry->next->start) - root->end;
980 vm_map_entry_set_max_free(root);
989 CTR3(KTR_VM, "vm_map_entry_unlink: map %p, nentries %d, entry %p", map,
990 map->nentries, entry);
994 * vm_map_entry_resize_free:
996 * Recompute the amount of free space following a vm_map_entry
997 * and propagate that value up the tree. Call this function after
998 * resizing a map entry in-place, that is, without a call to
999 * vm_map_entry_link() or _unlink().
1001 * The map must be locked, and leaves it so.
1004 vm_map_entry_resize_free(vm_map_t map, vm_map_entry_t entry)
1008 * Using splay trees without parent pointers, propagating
1009 * max_free up the tree is done by moving the entry to the
1010 * root and making the change there.
1012 if (entry != map->root)
1013 map->root = vm_map_entry_splay(entry->start, map->root);
1015 entry->adj_free = (entry->next == &map->header ? map->max_offset :
1016 entry->next->start) - entry->end;
1017 vm_map_entry_set_max_free(entry);
1021 * vm_map_lookup_entry: [ internal use only ]
1023 * Finds the map entry containing (or
1024 * immediately preceding) the specified address
1025 * in the given map; the entry is returned
1026 * in the "entry" parameter. The boolean
1027 * result indicates whether the address is
1028 * actually contained in the map.
1031 vm_map_lookup_entry(
1033 vm_offset_t address,
1034 vm_map_entry_t *entry) /* OUT */
1040 * If the map is empty, then the map entry immediately preceding
1041 * "address" is the map's header.
1045 *entry = &map->header;
1046 else if (address >= cur->start && cur->end > address) {
1049 } else if ((locked = vm_map_locked(map)) ||
1050 sx_try_upgrade(&map->lock)) {
1052 * Splay requires a write lock on the map. However, it only
1053 * restructures the binary search tree; it does not otherwise
1054 * change the map. Thus, the map's timestamp need not change
1055 * on a temporary upgrade.
1057 map->root = cur = vm_map_entry_splay(address, cur);
1059 sx_downgrade(&map->lock);
1062 * If "address" is contained within a map entry, the new root
1063 * is that map entry. Otherwise, the new root is a map entry
1064 * immediately before or after "address".
1066 if (address >= cur->start) {
1068 if (cur->end > address)
1074 * Since the map is only locked for read access, perform a
1075 * standard binary search tree lookup for "address".
1078 if (address < cur->start) {
1079 if (cur->left == NULL) {
1084 } else if (cur->end > address) {
1088 if (cur->right == NULL) {
1101 * Inserts the given whole VM object into the target
1102 * map at the specified address range. The object's
1103 * size should match that of the address range.
1105 * Requires that the map be locked, and leaves it so.
1107 * If object is non-NULL, ref count must be bumped by caller
1108 * prior to making call to account for the new entry.
1111 vm_map_insert(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1112 vm_offset_t start, vm_offset_t end, vm_prot_t prot, vm_prot_t max,
1115 vm_map_entry_t new_entry;
1116 vm_map_entry_t prev_entry;
1117 vm_map_entry_t temp_entry;
1118 vm_eflags_t protoeflags;
1120 vm_inherit_t inheritance;
1121 boolean_t charge_prev_obj;
1123 VM_MAP_ASSERT_LOCKED(map);
1126 * Check that the start and end points are not bogus.
1128 if ((start < map->min_offset) || (end > map->max_offset) ||
1130 return (KERN_INVALID_ADDRESS);
1133 * Find the entry prior to the proposed starting address; if it's part
1134 * of an existing entry, this range is bogus.
1136 if (vm_map_lookup_entry(map, start, &temp_entry))
1137 return (KERN_NO_SPACE);
1139 prev_entry = temp_entry;
1142 * Assert that the next entry doesn't overlap the end point.
1144 if ((prev_entry->next != &map->header) &&
1145 (prev_entry->next->start < end))
1146 return (KERN_NO_SPACE);
1149 charge_prev_obj = FALSE;
1151 if (cow & MAP_COPY_ON_WRITE)
1152 protoeflags |= MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY;
1154 if (cow & MAP_NOFAULT) {
1155 protoeflags |= MAP_ENTRY_NOFAULT;
1157 KASSERT(object == NULL,
1158 ("vm_map_insert: paradoxical MAP_NOFAULT request"));
1160 if (cow & MAP_DISABLE_SYNCER)
1161 protoeflags |= MAP_ENTRY_NOSYNC;
1162 if (cow & MAP_DISABLE_COREDUMP)
1163 protoeflags |= MAP_ENTRY_NOCOREDUMP;
1164 if (cow & MAP_VN_WRITECOUNT)
1165 protoeflags |= MAP_ENTRY_VN_WRITECNT;
1166 if (cow & MAP_INHERIT_SHARE)
1167 inheritance = VM_INHERIT_SHARE;
1169 inheritance = VM_INHERIT_DEFAULT;
1172 KASSERT((object != kmem_object && object != kernel_object) ||
1173 ((object == kmem_object || object == kernel_object) &&
1174 !(protoeflags & MAP_ENTRY_NEEDS_COPY)),
1175 ("kmem or kernel object and cow"));
1176 if (cow & (MAP_ACC_NO_CHARGE | MAP_NOFAULT))
1178 if ((cow & MAP_ACC_CHARGED) || ((prot & VM_PROT_WRITE) &&
1179 ((protoeflags & MAP_ENTRY_NEEDS_COPY) || object == NULL))) {
1180 if (!(cow & MAP_ACC_CHARGED) && !swap_reserve(end - start))
1181 return (KERN_RESOURCE_SHORTAGE);
1182 KASSERT(object == NULL || (protoeflags & MAP_ENTRY_NEEDS_COPY) ||
1183 object->cred == NULL,
1184 ("OVERCOMMIT: vm_map_insert o %p", object));
1185 cred = curthread->td_ucred;
1187 if (object == NULL && !(protoeflags & MAP_ENTRY_NEEDS_COPY))
1188 charge_prev_obj = TRUE;
1192 /* Expand the kernel pmap, if necessary. */
1193 if (map == kernel_map && end > kernel_vm_end)
1194 pmap_growkernel(end);
1195 if (object != NULL) {
1197 * OBJ_ONEMAPPING must be cleared unless this mapping
1198 * is trivially proven to be the only mapping for any
1199 * of the object's pages. (Object granularity
1200 * reference counting is insufficient to recognize
1201 * aliases with precision.)
1203 VM_OBJECT_WLOCK(object);
1204 if (object->ref_count > 1 || object->shadow_count != 0)
1205 vm_object_clear_flag(object, OBJ_ONEMAPPING);
1206 VM_OBJECT_WUNLOCK(object);
1208 else if ((prev_entry != &map->header) &&
1209 (prev_entry->eflags == protoeflags) &&
1210 (cow & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP)) == 0 &&
1211 (prev_entry->end == start) &&
1212 (prev_entry->wired_count == 0) &&
1213 (prev_entry->cred == cred ||
1214 (prev_entry->object.vm_object != NULL &&
1215 (prev_entry->object.vm_object->cred == cred))) &&
1216 vm_object_coalesce(prev_entry->object.vm_object,
1218 (vm_size_t)(prev_entry->end - prev_entry->start),
1219 (vm_size_t)(end - prev_entry->end), charge_prev_obj)) {
1221 * We were able to extend the object. Determine if we
1222 * can extend the previous map entry to include the
1223 * new range as well.
1225 if ((prev_entry->inheritance == inheritance) &&
1226 (prev_entry->protection == prot) &&
1227 (prev_entry->max_protection == max)) {
1228 map->size += (end - prev_entry->end);
1229 prev_entry->end = end;
1230 vm_map_entry_resize_free(map, prev_entry);
1231 vm_map_simplify_entry(map, prev_entry);
1234 return (KERN_SUCCESS);
1238 * If we can extend the object but cannot extend the
1239 * map entry, we have to create a new map entry. We
1240 * must bump the ref count on the extended object to
1241 * account for it. object may be NULL.
1243 object = prev_entry->object.vm_object;
1244 offset = prev_entry->offset +
1245 (prev_entry->end - prev_entry->start);
1246 vm_object_reference(object);
1247 if (cred != NULL && object != NULL && object->cred != NULL &&
1248 !(prev_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
1249 /* Object already accounts for this uid. */
1256 * NOTE: if conditionals fail, object can be NULL here. This occurs
1257 * in things like the buffer map where we manage kva but do not manage
1262 * Create a new entry
1264 new_entry = vm_map_entry_create(map);
1265 new_entry->start = start;
1266 new_entry->end = end;
1267 new_entry->cred = NULL;
1269 new_entry->eflags = protoeflags;
1270 new_entry->object.vm_object = object;
1271 new_entry->offset = offset;
1272 new_entry->avail_ssize = 0;
1274 new_entry->inheritance = inheritance;
1275 new_entry->protection = prot;
1276 new_entry->max_protection = max;
1277 new_entry->wired_count = 0;
1278 new_entry->wiring_thread = NULL;
1279 new_entry->read_ahead = VM_FAULT_READ_AHEAD_INIT;
1280 new_entry->next_read = OFF_TO_IDX(offset);
1282 KASSERT(cred == NULL || !ENTRY_CHARGED(new_entry),
1283 ("OVERCOMMIT: vm_map_insert leaks vm_map %p", new_entry));
1284 new_entry->cred = cred;
1287 * Insert the new entry into the list
1289 vm_map_entry_link(map, prev_entry, new_entry);
1290 map->size += new_entry->end - new_entry->start;
1293 * It may be possible to merge the new entry with the next and/or
1294 * previous entries. However, due to MAP_STACK_* being a hack, a
1295 * panic can result from merging such entries.
1297 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0)
1298 vm_map_simplify_entry(map, new_entry);
1300 if (cow & (MAP_PREFAULT|MAP_PREFAULT_PARTIAL)) {
1301 vm_map_pmap_enter(map, start, prot,
1302 object, OFF_TO_IDX(offset), end - start,
1303 cow & MAP_PREFAULT_PARTIAL);
1306 return (KERN_SUCCESS);
1312 * Find the first fit (lowest VM address) for "length" free bytes
1313 * beginning at address >= start in the given map.
1315 * In a vm_map_entry, "adj_free" is the amount of free space
1316 * adjacent (higher address) to this entry, and "max_free" is the
1317 * maximum amount of contiguous free space in its subtree. This
1318 * allows finding a free region in one path down the tree, so
1319 * O(log n) amortized with splay trees.
1321 * The map must be locked, and leaves it so.
1323 * Returns: 0 on success, and starting address in *addr,
1324 * 1 if insufficient space.
1327 vm_map_findspace(vm_map_t map, vm_offset_t start, vm_size_t length,
1328 vm_offset_t *addr) /* OUT */
1330 vm_map_entry_t entry;
1334 * Request must fit within min/max VM address and must avoid
1337 if (start < map->min_offset)
1338 start = map->min_offset;
1339 if (start + length > map->max_offset || start + length < start)
1342 /* Empty tree means wide open address space. */
1343 if (map->root == NULL) {
1349 * After splay, if start comes before root node, then there
1350 * must be a gap from start to the root.
1352 map->root = vm_map_entry_splay(start, map->root);
1353 if (start + length <= map->root->start) {
1359 * Root is the last node that might begin its gap before
1360 * start, and this is the last comparison where address
1361 * wrap might be a problem.
1363 st = (start > map->root->end) ? start : map->root->end;
1364 if (length <= map->root->end + map->root->adj_free - st) {
1369 /* With max_free, can immediately tell if no solution. */
1370 entry = map->root->right;
1371 if (entry == NULL || length > entry->max_free)
1375 * Search the right subtree in the order: left subtree, root,
1376 * right subtree (first fit). The previous splay implies that
1377 * all regions in the right subtree have addresses > start.
1379 while (entry != NULL) {
1380 if (entry->left != NULL && entry->left->max_free >= length)
1381 entry = entry->left;
1382 else if (entry->adj_free >= length) {
1386 entry = entry->right;
1389 /* Can't get here, so panic if we do. */
1390 panic("vm_map_findspace: max_free corrupt");
1394 vm_map_fixed(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1395 vm_offset_t start, vm_size_t length, vm_prot_t prot,
1396 vm_prot_t max, int cow)
1401 end = start + length;
1403 VM_MAP_RANGE_CHECK(map, start, end);
1404 (void) vm_map_delete(map, start, end);
1405 result = vm_map_insert(map, object, offset, start, end, prot,
1412 * vm_map_find finds an unallocated region in the target address
1413 * map with the given length. The search is defined to be
1414 * first-fit from the specified address; the region found is
1415 * returned in the same parameter.
1417 * If object is non-NULL, ref count must be bumped by caller
1418 * prior to making call to account for the new entry.
1421 vm_map_find(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1422 vm_offset_t *addr, /* IN/OUT */
1423 vm_size_t length, vm_offset_t max_addr, int find_space,
1424 vm_prot_t prot, vm_prot_t max, int cow)
1426 vm_offset_t alignment, initial_addr, start;
1429 if (find_space == VMFS_OPTIMAL_SPACE && (object == NULL ||
1430 (object->flags & OBJ_COLORED) == 0))
1431 find_space = VMFS_ANY_SPACE;
1432 if (find_space >> 8 != 0) {
1433 KASSERT((find_space & 0xff) == 0, ("bad VMFS flags"));
1434 alignment = (vm_offset_t)1 << (find_space >> 8);
1437 initial_addr = *addr;
1439 start = initial_addr;
1442 if (find_space != VMFS_NO_SPACE) {
1443 if (vm_map_findspace(map, start, length, addr) ||
1444 (max_addr != 0 && *addr + length > max_addr)) {
1446 if (find_space == VMFS_OPTIMAL_SPACE) {
1447 find_space = VMFS_ANY_SPACE;
1450 return (KERN_NO_SPACE);
1452 switch (find_space) {
1453 case VMFS_SUPER_SPACE:
1454 case VMFS_OPTIMAL_SPACE:
1455 pmap_align_superpage(object, offset, addr,
1458 case VMFS_ANY_SPACE:
1461 if ((*addr & (alignment - 1)) != 0) {
1462 *addr &= ~(alignment - 1);
1470 result = vm_map_insert(map, object, offset, start, start +
1471 length, prot, max, cow);
1472 } while (result == KERN_NO_SPACE && find_space != VMFS_NO_SPACE &&
1473 find_space != VMFS_ANY_SPACE);
1479 * vm_map_simplify_entry:
1481 * Simplify the given map entry by merging with either neighbor. This
1482 * routine also has the ability to merge with both neighbors.
1484 * The map must be locked.
1486 * This routine guarentees that the passed entry remains valid (though
1487 * possibly extended). When merging, this routine may delete one or
1491 vm_map_simplify_entry(vm_map_t map, vm_map_entry_t entry)
1493 vm_map_entry_t next, prev;
1494 vm_size_t prevsize, esize;
1496 if (entry->eflags & (MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_IS_SUB_MAP))
1500 if (prev != &map->header) {
1501 prevsize = prev->end - prev->start;
1502 if ( (prev->end == entry->start) &&
1503 (prev->object.vm_object == entry->object.vm_object) &&
1504 (!prev->object.vm_object ||
1505 (prev->offset + prevsize == entry->offset)) &&
1506 (prev->eflags == entry->eflags) &&
1507 (prev->protection == entry->protection) &&
1508 (prev->max_protection == entry->max_protection) &&
1509 (prev->inheritance == entry->inheritance) &&
1510 (prev->wired_count == entry->wired_count) &&
1511 (prev->cred == entry->cred)) {
1512 vm_map_entry_unlink(map, prev);
1513 entry->start = prev->start;
1514 entry->offset = prev->offset;
1515 if (entry->prev != &map->header)
1516 vm_map_entry_resize_free(map, entry->prev);
1519 * If the backing object is a vnode object,
1520 * vm_object_deallocate() calls vrele().
1521 * However, vrele() does not lock the vnode
1522 * because the vnode has additional
1523 * references. Thus, the map lock can be kept
1524 * without causing a lock-order reversal with
1527 * Since we count the number of virtual page
1528 * mappings in object->un_pager.vnp.writemappings,
1529 * the writemappings value should not be adjusted
1530 * when the entry is disposed of.
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_WLOCK(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_WUNLOCK(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);
1643 * The object->un_pager.vnp.writemappings for the
1644 * object of MAP_ENTRY_VN_WRITECNT type entry shall be
1645 * kept as is here. The virtual pages are
1646 * re-distributed among the clipped entries, so the sum is
1653 * vm_map_clip_end: [ internal use only ]
1655 * Asserts that the given entry ends at or before
1656 * the specified address; if necessary,
1657 * it splits the entry into two.
1659 #define vm_map_clip_end(map, entry, endaddr) \
1661 if ((endaddr) < (entry->end)) \
1662 _vm_map_clip_end((map), (entry), (endaddr)); \
1666 * This routine is called only when it is known that
1667 * the entry must be split.
1670 _vm_map_clip_end(vm_map_t map, vm_map_entry_t entry, vm_offset_t end)
1672 vm_map_entry_t new_entry;
1674 VM_MAP_ASSERT_LOCKED(map);
1677 * If there is no object backing this entry, we might as well create
1678 * one now. If we defer it, an object can get created after the map
1679 * is clipped, and individual objects will be created for the split-up
1680 * map. This is a bit of a hack, but is also about the best place to
1681 * put this improvement.
1683 if (entry->object.vm_object == NULL && !map->system_map) {
1685 object = vm_object_allocate(OBJT_DEFAULT,
1686 atop(entry->end - entry->start));
1687 entry->object.vm_object = object;
1689 if (entry->cred != NULL) {
1690 object->cred = entry->cred;
1691 object->charge = entry->end - entry->start;
1694 } else if (entry->object.vm_object != NULL &&
1695 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) &&
1696 entry->cred != NULL) {
1697 VM_OBJECT_WLOCK(entry->object.vm_object);
1698 KASSERT(entry->object.vm_object->cred == NULL,
1699 ("OVERCOMMIT: vm_entry_clip_end: both cred e %p", entry));
1700 entry->object.vm_object->cred = entry->cred;
1701 entry->object.vm_object->charge = entry->end - entry->start;
1702 VM_OBJECT_WUNLOCK(entry->object.vm_object);
1707 * Create a new entry and insert it AFTER the specified entry
1709 new_entry = vm_map_entry_create(map);
1710 *new_entry = *entry;
1712 new_entry->start = entry->end = end;
1713 new_entry->offset += (end - entry->start);
1714 if (new_entry->cred != NULL)
1715 crhold(entry->cred);
1717 vm_map_entry_link(map, entry, new_entry);
1719 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1720 vm_object_reference(new_entry->object.vm_object);
1725 * vm_map_submap: [ kernel use only ]
1727 * Mark the given range as handled by a subordinate map.
1729 * This range must have been created with vm_map_find,
1730 * and no other operations may have been performed on this
1731 * range prior to calling vm_map_submap.
1733 * Only a limited number of operations can be performed
1734 * within this rage after calling vm_map_submap:
1736 * [Don't try vm_map_copy!]
1738 * To remove a submapping, one must first remove the
1739 * range from the superior map, and then destroy the
1740 * submap (if desired). [Better yet, don't try it.]
1749 vm_map_entry_t entry;
1750 int result = KERN_INVALID_ARGUMENT;
1754 VM_MAP_RANGE_CHECK(map, start, end);
1756 if (vm_map_lookup_entry(map, start, &entry)) {
1757 vm_map_clip_start(map, entry, start);
1759 entry = entry->next;
1761 vm_map_clip_end(map, entry, end);
1763 if ((entry->start == start) && (entry->end == end) &&
1764 ((entry->eflags & MAP_ENTRY_COW) == 0) &&
1765 (entry->object.vm_object == NULL)) {
1766 entry->object.sub_map = submap;
1767 entry->eflags |= MAP_ENTRY_IS_SUB_MAP;
1768 result = KERN_SUCCESS;
1776 * The maximum number of pages to map
1778 #define MAX_INIT_PT 96
1781 * vm_map_pmap_enter:
1783 * Preload read-only mappings for the specified object's resident pages
1784 * into the target map. If "flags" is MAP_PREFAULT_PARTIAL, then only
1785 * the resident pages within the address range [addr, addr + ulmin(size,
1786 * ptoa(MAX_INIT_PT))) are mapped. Otherwise, all resident pages within
1787 * the specified address range are mapped. This eliminates many soft
1788 * faults on process startup and immediately after an mmap(2). Because
1789 * these are speculative mappings, cached pages are not reactivated and
1793 vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot,
1794 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags)
1797 vm_page_t p, p_start;
1798 vm_pindex_t psize, tmpidx;
1800 if ((prot & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0 || object == NULL)
1802 VM_OBJECT_RLOCK(object);
1803 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) {
1804 VM_OBJECT_RUNLOCK(object);
1805 VM_OBJECT_WLOCK(object);
1806 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) {
1807 pmap_object_init_pt(map->pmap, addr, object, pindex,
1809 VM_OBJECT_WUNLOCK(object);
1812 VM_OBJECT_LOCK_DOWNGRADE(object);
1816 if (psize > MAX_INIT_PT && (flags & MAP_PREFAULT_PARTIAL) != 0)
1817 psize = MAX_INIT_PT;
1818 if (psize + pindex > object->size) {
1819 if (object->size < pindex) {
1820 VM_OBJECT_RUNLOCK(object);
1823 psize = object->size - pindex;
1829 p = vm_page_find_least(object, pindex);
1831 * Assert: the variable p is either (1) the page with the
1832 * least pindex greater than or equal to the parameter pindex
1836 p != NULL && (tmpidx = p->pindex - pindex) < psize;
1837 p = TAILQ_NEXT(p, listq)) {
1839 * don't allow an madvise to blow away our really
1840 * free pages allocating pv entries.
1842 if ((flags & MAP_PREFAULT_MADVISE) &&
1843 cnt.v_free_count < cnt.v_free_reserved) {
1847 if (p->valid == VM_PAGE_BITS_ALL) {
1848 if (p_start == NULL) {
1849 start = addr + ptoa(tmpidx);
1852 } else if (p_start != NULL) {
1853 pmap_enter_object(map->pmap, start, addr +
1854 ptoa(tmpidx), p_start, prot);
1858 if (p_start != NULL)
1859 pmap_enter_object(map->pmap, start, addr + ptoa(psize),
1861 VM_OBJECT_RUNLOCK(object);
1867 * Sets the protection of the specified address
1868 * region in the target map. If "set_max" is
1869 * specified, the maximum protection is to be set;
1870 * otherwise, only the current protection is affected.
1873 vm_map_protect(vm_map_t map, vm_offset_t start, vm_offset_t end,
1874 vm_prot_t new_prot, boolean_t set_max)
1876 vm_map_entry_t current, entry;
1882 return (KERN_SUCCESS);
1886 VM_MAP_RANGE_CHECK(map, start, end);
1888 if (vm_map_lookup_entry(map, start, &entry)) {
1889 vm_map_clip_start(map, entry, start);
1891 entry = entry->next;
1895 * Make a first pass to check for protection violations.
1898 while ((current != &map->header) && (current->start < end)) {
1899 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
1901 return (KERN_INVALID_ARGUMENT);
1903 if ((new_prot & current->max_protection) != new_prot) {
1905 return (KERN_PROTECTION_FAILURE);
1907 current = current->next;
1912 * Do an accounting pass for private read-only mappings that
1913 * now will do cow due to allowed write (e.g. debugger sets
1914 * breakpoint on text segment)
1916 for (current = entry; (current != &map->header) &&
1917 (current->start < end); current = current->next) {
1919 vm_map_clip_end(map, current, end);
1922 ((new_prot & ~(current->protection)) & VM_PROT_WRITE) == 0 ||
1923 ENTRY_CHARGED(current)) {
1927 cred = curthread->td_ucred;
1928 obj = current->object.vm_object;
1930 if (obj == NULL || (current->eflags & MAP_ENTRY_NEEDS_COPY)) {
1931 if (!swap_reserve(current->end - current->start)) {
1933 return (KERN_RESOURCE_SHORTAGE);
1936 current->cred = cred;
1940 VM_OBJECT_WLOCK(obj);
1941 if (obj->type != OBJT_DEFAULT && obj->type != OBJT_SWAP) {
1942 VM_OBJECT_WUNLOCK(obj);
1947 * Charge for the whole object allocation now, since
1948 * we cannot distinguish between non-charged and
1949 * charged clipped mapping of the same object later.
1951 KASSERT(obj->charge == 0,
1952 ("vm_map_protect: object %p overcharged\n", obj));
1953 if (!swap_reserve(ptoa(obj->size))) {
1954 VM_OBJECT_WUNLOCK(obj);
1956 return (KERN_RESOURCE_SHORTAGE);
1961 obj->charge = ptoa(obj->size);
1962 VM_OBJECT_WUNLOCK(obj);
1966 * Go back and fix up protections. [Note that clipping is not
1967 * necessary the second time.]
1970 while ((current != &map->header) && (current->start < end)) {
1971 old_prot = current->protection;
1974 current->protection =
1975 (current->max_protection = new_prot) &
1978 current->protection = new_prot;
1980 if ((current->eflags & (MAP_ENTRY_COW | MAP_ENTRY_USER_WIRED))
1981 == (MAP_ENTRY_COW | MAP_ENTRY_USER_WIRED) &&
1982 (current->protection & VM_PROT_WRITE) != 0 &&
1983 (old_prot & VM_PROT_WRITE) == 0) {
1984 vm_fault_copy_entry(map, map, current, current, NULL);
1988 * When restricting access, update the physical map. Worry
1989 * about copy-on-write here.
1991 if ((old_prot & ~current->protection) != 0) {
1992 #define MASK(entry) (((entry)->eflags & MAP_ENTRY_COW) ? ~VM_PROT_WRITE : \
1994 pmap_protect(map->pmap, current->start,
1996 current->protection & MASK(current));
1999 vm_map_simplify_entry(map, current);
2000 current = current->next;
2003 return (KERN_SUCCESS);
2009 * This routine traverses a processes map handling the madvise
2010 * system call. Advisories are classified as either those effecting
2011 * the vm_map_entry structure, or those effecting the underlying
2021 vm_map_entry_t current, entry;
2025 * Some madvise calls directly modify the vm_map_entry, in which case
2026 * we need to use an exclusive lock on the map and we need to perform
2027 * various clipping operations. Otherwise we only need a read-lock
2032 case MADV_SEQUENTIAL:
2039 return (KERN_SUCCESS);
2047 return (KERN_SUCCESS);
2048 vm_map_lock_read(map);
2051 return (KERN_INVALID_ARGUMENT);
2055 * Locate starting entry and clip if necessary.
2057 VM_MAP_RANGE_CHECK(map, start, end);
2059 if (vm_map_lookup_entry(map, start, &entry)) {
2061 vm_map_clip_start(map, entry, start);
2063 entry = entry->next;
2068 * madvise behaviors that are implemented in the vm_map_entry.
2070 * We clip the vm_map_entry so that behavioral changes are
2071 * limited to the specified address range.
2073 for (current = entry;
2074 (current != &map->header) && (current->start < end);
2075 current = current->next
2077 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
2080 vm_map_clip_end(map, current, end);
2084 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_NORMAL);
2086 case MADV_SEQUENTIAL:
2087 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_SEQUENTIAL);
2090 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_RANDOM);
2093 current->eflags |= MAP_ENTRY_NOSYNC;
2096 current->eflags &= ~MAP_ENTRY_NOSYNC;
2099 current->eflags |= MAP_ENTRY_NOCOREDUMP;
2102 current->eflags &= ~MAP_ENTRY_NOCOREDUMP;
2107 vm_map_simplify_entry(map, current);
2111 vm_pindex_t pstart, pend;
2114 * madvise behaviors that are implemented in the underlying
2117 * Since we don't clip the vm_map_entry, we have to clip
2118 * the vm_object pindex and count.
2120 for (current = entry;
2121 (current != &map->header) && (current->start < end);
2122 current = current->next
2124 vm_offset_t useEnd, useStart;
2126 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
2129 pstart = OFF_TO_IDX(current->offset);
2130 pend = pstart + atop(current->end - current->start);
2131 useStart = current->start;
2132 useEnd = current->end;
2134 if (current->start < start) {
2135 pstart += atop(start - current->start);
2138 if (current->end > end) {
2139 pend -= atop(current->end - end);
2147 * Perform the pmap_advise() before clearing
2148 * PGA_REFERENCED in vm_page_advise(). Otherwise, a
2149 * concurrent pmap operation, such as pmap_remove(),
2150 * could clear a reference in the pmap and set
2151 * PGA_REFERENCED on the page before the pmap_advise()
2152 * had completed. Consequently, the page would appear
2153 * referenced based upon an old reference that
2154 * occurred before this pmap_advise() ran.
2156 if (behav == MADV_DONTNEED || behav == MADV_FREE)
2157 pmap_advise(map->pmap, useStart, useEnd,
2160 vm_object_madvise(current->object.vm_object, pstart,
2162 if (behav == MADV_WILLNEED) {
2163 vm_map_pmap_enter(map,
2165 current->protection,
2166 current->object.vm_object,
2168 ptoa(pend - pstart),
2169 MAP_PREFAULT_MADVISE
2173 vm_map_unlock_read(map);
2182 * Sets the inheritance of the specified address
2183 * range in the target map. Inheritance
2184 * affects how the map will be shared with
2185 * child maps at the time of vmspace_fork.
2188 vm_map_inherit(vm_map_t map, vm_offset_t start, vm_offset_t end,
2189 vm_inherit_t new_inheritance)
2191 vm_map_entry_t entry;
2192 vm_map_entry_t temp_entry;
2194 switch (new_inheritance) {
2195 case VM_INHERIT_NONE:
2196 case VM_INHERIT_COPY:
2197 case VM_INHERIT_SHARE:
2200 return (KERN_INVALID_ARGUMENT);
2203 return (KERN_SUCCESS);
2205 VM_MAP_RANGE_CHECK(map, start, end);
2206 if (vm_map_lookup_entry(map, start, &temp_entry)) {
2208 vm_map_clip_start(map, entry, start);
2210 entry = temp_entry->next;
2211 while ((entry != &map->header) && (entry->start < end)) {
2212 vm_map_clip_end(map, entry, end);
2213 entry->inheritance = new_inheritance;
2214 vm_map_simplify_entry(map, entry);
2215 entry = entry->next;
2218 return (KERN_SUCCESS);
2224 * Implements both kernel and user unwiring.
2227 vm_map_unwire(vm_map_t map, vm_offset_t start, vm_offset_t end,
2230 vm_map_entry_t entry, first_entry, tmp_entry;
2231 vm_offset_t saved_start;
2232 unsigned int last_timestamp;
2234 boolean_t need_wakeup, result, user_unwire;
2237 return (KERN_SUCCESS);
2238 user_unwire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE;
2240 VM_MAP_RANGE_CHECK(map, start, end);
2241 if (!vm_map_lookup_entry(map, start, &first_entry)) {
2242 if (flags & VM_MAP_WIRE_HOLESOK)
2243 first_entry = first_entry->next;
2246 return (KERN_INVALID_ADDRESS);
2249 last_timestamp = map->timestamp;
2250 entry = first_entry;
2251 while (entry != &map->header && entry->start < end) {
2252 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
2254 * We have not yet clipped the entry.
2256 saved_start = (start >= entry->start) ? start :
2258 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2259 if (vm_map_unlock_and_wait(map, 0)) {
2261 * Allow interruption of user unwiring?
2265 if (last_timestamp+1 != map->timestamp) {
2267 * Look again for the entry because the map was
2268 * modified while it was unlocked.
2269 * Specifically, the entry may have been
2270 * clipped, merged, or deleted.
2272 if (!vm_map_lookup_entry(map, saved_start,
2274 if (flags & VM_MAP_WIRE_HOLESOK)
2275 tmp_entry = tmp_entry->next;
2277 if (saved_start == start) {
2279 * First_entry has been deleted.
2282 return (KERN_INVALID_ADDRESS);
2285 rv = KERN_INVALID_ADDRESS;
2289 if (entry == first_entry)
2290 first_entry = tmp_entry;
2295 last_timestamp = map->timestamp;
2298 vm_map_clip_start(map, entry, start);
2299 vm_map_clip_end(map, entry, end);
2301 * Mark the entry in case the map lock is released. (See
2304 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 &&
2305 entry->wiring_thread == NULL,
2306 ("owned map entry %p", entry));
2307 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
2308 entry->wiring_thread = curthread;
2310 * Check the map for holes in the specified region.
2311 * If VM_MAP_WIRE_HOLESOK was specified, skip this check.
2313 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) &&
2314 (entry->end < end && (entry->next == &map->header ||
2315 entry->next->start > entry->end))) {
2317 rv = KERN_INVALID_ADDRESS;
2321 * If system unwiring, require that the entry is system wired.
2324 vm_map_entry_system_wired_count(entry) == 0) {
2326 rv = KERN_INVALID_ARGUMENT;
2329 entry = entry->next;
2333 need_wakeup = FALSE;
2334 if (first_entry == NULL) {
2335 result = vm_map_lookup_entry(map, start, &first_entry);
2336 if (!result && (flags & VM_MAP_WIRE_HOLESOK))
2337 first_entry = first_entry->next;
2339 KASSERT(result, ("vm_map_unwire: lookup failed"));
2341 for (entry = first_entry; entry != &map->header && entry->start < end;
2342 entry = entry->next) {
2344 * If VM_MAP_WIRE_HOLESOK was specified, an empty
2345 * space in the unwired region could have been mapped
2346 * while the map lock was dropped for draining
2347 * MAP_ENTRY_IN_TRANSITION. Moreover, another thread
2348 * could be simultaneously wiring this new mapping
2349 * entry. Detect these cases and skip any entries
2350 * marked as in transition by us.
2352 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 ||
2353 entry->wiring_thread != curthread) {
2354 KASSERT((flags & VM_MAP_WIRE_HOLESOK) != 0,
2355 ("vm_map_unwire: !HOLESOK and new/changed entry"));
2359 if (rv == KERN_SUCCESS && (!user_unwire ||
2360 (entry->eflags & MAP_ENTRY_USER_WIRED))) {
2362 entry->eflags &= ~MAP_ENTRY_USER_WIRED;
2363 entry->wired_count--;
2364 if (entry->wired_count == 0) {
2366 * Retain the map lock.
2368 vm_fault_unwire(map, entry->start, entry->end,
2369 entry->object.vm_object != NULL &&
2370 (entry->object.vm_object->flags &
2371 OBJ_FICTITIOUS) != 0);
2374 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0,
2375 ("vm_map_unwire: in-transition flag missing %p", entry));
2376 KASSERT(entry->wiring_thread == curthread,
2377 ("vm_map_unwire: alien wire %p", entry));
2378 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION;
2379 entry->wiring_thread = NULL;
2380 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
2381 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
2384 vm_map_simplify_entry(map, entry);
2395 * Implements both kernel and user wiring.
2398 vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t end,
2401 vm_map_entry_t entry, first_entry, tmp_entry;
2402 vm_offset_t saved_end, saved_start;
2403 unsigned int last_timestamp;
2405 boolean_t fictitious, need_wakeup, result, user_wire;
2409 return (KERN_SUCCESS);
2411 if (flags & VM_MAP_WIRE_WRITE)
2412 prot |= VM_PROT_WRITE;
2413 user_wire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE;
2415 VM_MAP_RANGE_CHECK(map, start, end);
2416 if (!vm_map_lookup_entry(map, start, &first_entry)) {
2417 if (flags & VM_MAP_WIRE_HOLESOK)
2418 first_entry = first_entry->next;
2421 return (KERN_INVALID_ADDRESS);
2424 last_timestamp = map->timestamp;
2425 entry = first_entry;
2426 while (entry != &map->header && entry->start < end) {
2427 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
2429 * We have not yet clipped the entry.
2431 saved_start = (start >= entry->start) ? start :
2433 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2434 if (vm_map_unlock_and_wait(map, 0)) {
2436 * Allow interruption of user wiring?
2440 if (last_timestamp + 1 != map->timestamp) {
2442 * Look again for the entry because the map was
2443 * modified while it was unlocked.
2444 * Specifically, the entry may have been
2445 * clipped, merged, or deleted.
2447 if (!vm_map_lookup_entry(map, saved_start,
2449 if (flags & VM_MAP_WIRE_HOLESOK)
2450 tmp_entry = tmp_entry->next;
2452 if (saved_start == start) {
2454 * first_entry has been deleted.
2457 return (KERN_INVALID_ADDRESS);
2460 rv = KERN_INVALID_ADDRESS;
2464 if (entry == first_entry)
2465 first_entry = tmp_entry;
2470 last_timestamp = map->timestamp;
2473 vm_map_clip_start(map, entry, start);
2474 vm_map_clip_end(map, entry, end);
2476 * Mark the entry in case the map lock is released. (See
2479 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 &&
2480 entry->wiring_thread == NULL,
2481 ("owned map entry %p", entry));
2482 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
2483 entry->wiring_thread = curthread;
2484 if ((entry->protection & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0
2485 || (entry->protection & prot) != prot) {
2486 entry->eflags |= MAP_ENTRY_WIRE_SKIPPED;
2487 if ((flags & VM_MAP_WIRE_HOLESOK) == 0) {
2489 rv = KERN_INVALID_ADDRESS;
2494 if (entry->wired_count == 0) {
2495 entry->wired_count++;
2496 saved_start = entry->start;
2497 saved_end = entry->end;
2498 fictitious = entry->object.vm_object != NULL &&
2499 (entry->object.vm_object->flags &
2500 OBJ_FICTITIOUS) != 0;
2502 * Release the map lock, relying on the in-transition
2503 * mark. Mark the map busy for fork.
2507 rv = vm_fault_wire(map, saved_start, saved_end,
2511 if (last_timestamp + 1 != map->timestamp) {
2513 * Look again for the entry because the map was
2514 * modified while it was unlocked. The entry
2515 * may have been clipped, but NOT merged or
2518 result = vm_map_lookup_entry(map, saved_start,
2520 KASSERT(result, ("vm_map_wire: lookup failed"));
2521 if (entry == first_entry)
2522 first_entry = tmp_entry;
2526 while (entry->end < saved_end) {
2527 if (rv != KERN_SUCCESS) {
2528 KASSERT(entry->wired_count == 1,
2529 ("vm_map_wire: bad count"));
2530 entry->wired_count = -1;
2532 entry = entry->next;
2535 last_timestamp = map->timestamp;
2536 if (rv != KERN_SUCCESS) {
2537 KASSERT(entry->wired_count == 1,
2538 ("vm_map_wire: bad count"));
2540 * Assign an out-of-range value to represent
2541 * the failure to wire this entry.
2543 entry->wired_count = -1;
2547 } else if (!user_wire ||
2548 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
2549 entry->wired_count++;
2552 * Check the map for holes in the specified region.
2553 * If VM_MAP_WIRE_HOLESOK was specified, skip this check.
2556 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) &&
2557 (entry->end < end && (entry->next == &map->header ||
2558 entry->next->start > entry->end))) {
2560 rv = KERN_INVALID_ADDRESS;
2563 entry = entry->next;
2567 need_wakeup = FALSE;
2568 if (first_entry == NULL) {
2569 result = vm_map_lookup_entry(map, start, &first_entry);
2570 if (!result && (flags & VM_MAP_WIRE_HOLESOK))
2571 first_entry = first_entry->next;
2573 KASSERT(result, ("vm_map_wire: lookup failed"));
2575 for (entry = first_entry; entry != &map->header && entry->start < end;
2576 entry = entry->next) {
2577 if ((entry->eflags & MAP_ENTRY_WIRE_SKIPPED) != 0)
2578 goto next_entry_done;
2581 * If VM_MAP_WIRE_HOLESOK was specified, an empty
2582 * space in the unwired region could have been mapped
2583 * while the map lock was dropped for faulting in the
2584 * pages or draining MAP_ENTRY_IN_TRANSITION.
2585 * Moreover, another thread could be simultaneously
2586 * wiring this new mapping entry. Detect these cases
2587 * and skip any entries marked as in transition by us.
2589 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 ||
2590 entry->wiring_thread != curthread) {
2591 KASSERT((flags & VM_MAP_WIRE_HOLESOK) != 0,
2592 ("vm_map_wire: !HOLESOK and new/changed entry"));
2596 if (rv == KERN_SUCCESS) {
2598 entry->eflags |= MAP_ENTRY_USER_WIRED;
2599 } else if (entry->wired_count == -1) {
2601 * Wiring failed on this entry. Thus, unwiring is
2604 entry->wired_count = 0;
2607 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0)
2608 entry->wired_count--;
2609 if (entry->wired_count == 0) {
2611 * Retain the map lock.
2613 vm_fault_unwire(map, entry->start, entry->end,
2614 entry->object.vm_object != NULL &&
2615 (entry->object.vm_object->flags &
2616 OBJ_FICTITIOUS) != 0);
2620 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0,
2621 ("vm_map_wire: in-transition flag missing %p", entry));
2622 KASSERT(entry->wiring_thread == curthread,
2623 ("vm_map_wire: alien wire %p", entry));
2624 entry->eflags &= ~(MAP_ENTRY_IN_TRANSITION |
2625 MAP_ENTRY_WIRE_SKIPPED);
2626 entry->wiring_thread = NULL;
2627 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
2628 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
2631 vm_map_simplify_entry(map, entry);
2642 * Push any dirty cached pages in the address range to their pager.
2643 * If syncio is TRUE, dirty pages are written synchronously.
2644 * If invalidate is TRUE, any cached pages are freed as well.
2646 * If the size of the region from start to end is zero, we are
2647 * supposed to flush all modified pages within the region containing
2648 * start. Unfortunately, a region can be split or coalesced with
2649 * neighboring regions, making it difficult to determine what the
2650 * original region was. Therefore, we approximate this requirement by
2651 * flushing the current region containing start.
2653 * Returns an error if any part of the specified range is not mapped.
2661 boolean_t invalidate)
2663 vm_map_entry_t current;
2664 vm_map_entry_t entry;
2667 vm_ooffset_t offset;
2668 unsigned int last_timestamp;
2671 vm_map_lock_read(map);
2672 VM_MAP_RANGE_CHECK(map, start, end);
2673 if (!vm_map_lookup_entry(map, start, &entry)) {
2674 vm_map_unlock_read(map);
2675 return (KERN_INVALID_ADDRESS);
2676 } else if (start == end) {
2677 start = entry->start;
2681 * Make a first pass to check for user-wired memory and holes.
2683 for (current = entry; current != &map->header && current->start < end;
2684 current = current->next) {
2685 if (invalidate && (current->eflags & MAP_ENTRY_USER_WIRED)) {
2686 vm_map_unlock_read(map);
2687 return (KERN_INVALID_ARGUMENT);
2689 if (end > current->end &&
2690 (current->next == &map->header ||
2691 current->end != current->next->start)) {
2692 vm_map_unlock_read(map);
2693 return (KERN_INVALID_ADDRESS);
2698 pmap_remove(map->pmap, start, end);
2702 * Make a second pass, cleaning/uncaching pages from the indicated
2705 for (current = entry; current != &map->header && current->start < end;) {
2706 offset = current->offset + (start - current->start);
2707 size = (end <= current->end ? end : current->end) - start;
2708 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
2710 vm_map_entry_t tentry;
2713 smap = current->object.sub_map;
2714 vm_map_lock_read(smap);
2715 (void) vm_map_lookup_entry(smap, offset, &tentry);
2716 tsize = tentry->end - offset;
2719 object = tentry->object.vm_object;
2720 offset = tentry->offset + (offset - tentry->start);
2721 vm_map_unlock_read(smap);
2723 object = current->object.vm_object;
2725 vm_object_reference(object);
2726 last_timestamp = map->timestamp;
2727 vm_map_unlock_read(map);
2728 if (!vm_object_sync(object, offset, size, syncio, invalidate))
2731 vm_object_deallocate(object);
2732 vm_map_lock_read(map);
2733 if (last_timestamp == map->timestamp ||
2734 !vm_map_lookup_entry(map, start, ¤t))
2735 current = current->next;
2738 vm_map_unlock_read(map);
2739 return (failed ? KERN_FAILURE : KERN_SUCCESS);
2743 * vm_map_entry_unwire: [ internal use only ]
2745 * Make the region specified by this entry pageable.
2747 * The map in question should be locked.
2748 * [This is the reason for this routine's existence.]
2751 vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry)
2753 vm_fault_unwire(map, entry->start, entry->end,
2754 entry->object.vm_object != NULL &&
2755 (entry->object.vm_object->flags & OBJ_FICTITIOUS) != 0);
2756 entry->wired_count = 0;
2760 vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map)
2763 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0)
2764 vm_object_deallocate(entry->object.vm_object);
2765 uma_zfree(system_map ? kmapentzone : mapentzone, entry);
2769 * vm_map_entry_delete: [ internal use only ]
2771 * Deallocate the given entry from the target map.
2774 vm_map_entry_delete(vm_map_t map, vm_map_entry_t entry)
2777 vm_pindex_t offidxstart, offidxend, count, size1;
2780 vm_map_entry_unlink(map, entry);
2781 object = entry->object.vm_object;
2782 size = entry->end - entry->start;
2785 if (entry->cred != NULL) {
2786 swap_release_by_cred(size, entry->cred);
2787 crfree(entry->cred);
2790 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0 &&
2792 KASSERT(entry->cred == NULL || object->cred == NULL ||
2793 (entry->eflags & MAP_ENTRY_NEEDS_COPY),
2794 ("OVERCOMMIT vm_map_entry_delete: both cred %p", entry));
2795 count = OFF_TO_IDX(size);
2796 offidxstart = OFF_TO_IDX(entry->offset);
2797 offidxend = offidxstart + count;
2798 VM_OBJECT_WLOCK(object);
2799 if (object->ref_count != 1 &&
2800 ((object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING ||
2801 object == kernel_object || object == kmem_object)) {
2802 vm_object_collapse(object);
2805 * The option OBJPR_NOTMAPPED can be passed here
2806 * because vm_map_delete() already performed
2807 * pmap_remove() on the only mapping to this range
2810 vm_object_page_remove(object, offidxstart, offidxend,
2812 if (object->type == OBJT_SWAP)
2813 swap_pager_freespace(object, offidxstart, count);
2814 if (offidxend >= object->size &&
2815 offidxstart < object->size) {
2816 size1 = object->size;
2817 object->size = offidxstart;
2818 if (object->cred != NULL) {
2819 size1 -= object->size;
2820 KASSERT(object->charge >= ptoa(size1),
2821 ("vm_map_entry_delete: object->charge < 0"));
2822 swap_release_by_cred(ptoa(size1), object->cred);
2823 object->charge -= ptoa(size1);
2827 VM_OBJECT_WUNLOCK(object);
2829 entry->object.vm_object = NULL;
2830 if (map->system_map)
2831 vm_map_entry_deallocate(entry, TRUE);
2833 entry->next = curthread->td_map_def_user;
2834 curthread->td_map_def_user = entry;
2839 * vm_map_delete: [ internal use only ]
2841 * Deallocates the given address range from the target
2845 vm_map_delete(vm_map_t map, vm_offset_t start, vm_offset_t end)
2847 vm_map_entry_t entry;
2848 vm_map_entry_t first_entry;
2850 VM_MAP_ASSERT_LOCKED(map);
2852 return (KERN_SUCCESS);
2855 * Find the start of the region, and clip it
2857 if (!vm_map_lookup_entry(map, start, &first_entry))
2858 entry = first_entry->next;
2860 entry = first_entry;
2861 vm_map_clip_start(map, entry, start);
2865 * Step through all entries in this region
2867 while ((entry != &map->header) && (entry->start < end)) {
2868 vm_map_entry_t next;
2871 * Wait for wiring or unwiring of an entry to complete.
2872 * Also wait for any system wirings to disappear on
2875 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 ||
2876 (vm_map_pmap(map) != kernel_pmap &&
2877 vm_map_entry_system_wired_count(entry) != 0)) {
2878 unsigned int last_timestamp;
2879 vm_offset_t saved_start;
2880 vm_map_entry_t tmp_entry;
2882 saved_start = entry->start;
2883 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2884 last_timestamp = map->timestamp;
2885 (void) vm_map_unlock_and_wait(map, 0);
2887 if (last_timestamp + 1 != map->timestamp) {
2889 * Look again for the entry because the map was
2890 * modified while it was unlocked.
2891 * Specifically, the entry may have been
2892 * clipped, merged, or deleted.
2894 if (!vm_map_lookup_entry(map, saved_start,
2896 entry = tmp_entry->next;
2899 vm_map_clip_start(map, entry,
2905 vm_map_clip_end(map, entry, end);
2910 * Unwire before removing addresses from the pmap; otherwise,
2911 * unwiring will put the entries back in the pmap.
2913 if (entry->wired_count != 0) {
2914 vm_map_entry_unwire(map, entry);
2917 pmap_remove(map->pmap, entry->start, entry->end);
2920 * Delete the entry only after removing all pmap
2921 * entries pointing to its pages. (Otherwise, its
2922 * page frames may be reallocated, and any modify bits
2923 * will be set in the wrong object!)
2925 vm_map_entry_delete(map, entry);
2928 return (KERN_SUCCESS);
2934 * Remove the given address range from the target map.
2935 * This is the exported form of vm_map_delete.
2938 vm_map_remove(vm_map_t map, vm_offset_t start, vm_offset_t end)
2943 VM_MAP_RANGE_CHECK(map, start, end);
2944 result = vm_map_delete(map, start, end);
2950 * vm_map_check_protection:
2952 * Assert that the target map allows the specified privilege on the
2953 * entire address region given. The entire region must be allocated.
2955 * WARNING! This code does not and should not check whether the
2956 * contents of the region is accessible. For example a smaller file
2957 * might be mapped into a larger address space.
2959 * NOTE! This code is also called by munmap().
2961 * The map must be locked. A read lock is sufficient.
2964 vm_map_check_protection(vm_map_t map, vm_offset_t start, vm_offset_t end,
2965 vm_prot_t protection)
2967 vm_map_entry_t entry;
2968 vm_map_entry_t tmp_entry;
2970 if (!vm_map_lookup_entry(map, start, &tmp_entry))
2974 while (start < end) {
2975 if (entry == &map->header)
2980 if (start < entry->start)
2983 * Check protection associated with entry.
2985 if ((entry->protection & protection) != protection)
2987 /* go to next entry */
2989 entry = entry->next;
2995 * vm_map_copy_entry:
2997 * Copies the contents of the source entry to the destination
2998 * entry. The entries *must* be aligned properly.
3004 vm_map_entry_t src_entry,
3005 vm_map_entry_t dst_entry,
3006 vm_ooffset_t *fork_charge)
3008 vm_object_t src_object;
3009 vm_map_entry_t fake_entry;
3014 VM_MAP_ASSERT_LOCKED(dst_map);
3016 if ((dst_entry->eflags|src_entry->eflags) & MAP_ENTRY_IS_SUB_MAP)
3019 if (src_entry->wired_count == 0) {
3022 * If the source entry is marked needs_copy, it is already
3025 if ((src_entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) {
3026 pmap_protect(src_map->pmap,
3029 src_entry->protection & ~VM_PROT_WRITE);
3033 * Make a copy of the object.
3035 size = src_entry->end - src_entry->start;
3036 if ((src_object = src_entry->object.vm_object) != NULL) {
3037 VM_OBJECT_WLOCK(src_object);
3038 charged = ENTRY_CHARGED(src_entry);
3039 if ((src_object->handle == NULL) &&
3040 (src_object->type == OBJT_DEFAULT ||
3041 src_object->type == OBJT_SWAP)) {
3042 vm_object_collapse(src_object);
3043 if ((src_object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING) {
3044 vm_object_split(src_entry);
3045 src_object = src_entry->object.vm_object;
3048 vm_object_reference_locked(src_object);
3049 vm_object_clear_flag(src_object, OBJ_ONEMAPPING);
3050 if (src_entry->cred != NULL &&
3051 !(src_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
3052 KASSERT(src_object->cred == NULL,
3053 ("OVERCOMMIT: vm_map_copy_entry: cred %p",
3055 src_object->cred = src_entry->cred;
3056 src_object->charge = size;
3058 VM_OBJECT_WUNLOCK(src_object);
3059 dst_entry->object.vm_object = src_object;
3061 cred = curthread->td_ucred;
3063 dst_entry->cred = cred;
3064 *fork_charge += size;
3065 if (!(src_entry->eflags &
3066 MAP_ENTRY_NEEDS_COPY)) {
3068 src_entry->cred = cred;
3069 *fork_charge += size;
3072 src_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY);
3073 dst_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY);
3074 dst_entry->offset = src_entry->offset;
3075 if (src_entry->eflags & MAP_ENTRY_VN_WRITECNT) {
3077 * MAP_ENTRY_VN_WRITECNT cannot
3078 * indicate write reference from
3079 * src_entry, since the entry is
3080 * marked as needs copy. Allocate a
3081 * fake entry that is used to
3082 * decrement object->un_pager.vnp.writecount
3083 * at the appropriate time. Attach
3084 * fake_entry to the deferred list.
3086 fake_entry = vm_map_entry_create(dst_map);
3087 fake_entry->eflags = MAP_ENTRY_VN_WRITECNT;
3088 src_entry->eflags &= ~MAP_ENTRY_VN_WRITECNT;
3089 vm_object_reference(src_object);
3090 fake_entry->object.vm_object = src_object;
3091 fake_entry->start = src_entry->start;
3092 fake_entry->end = src_entry->end;
3093 fake_entry->next = curthread->td_map_def_user;
3094 curthread->td_map_def_user = fake_entry;
3097 dst_entry->object.vm_object = NULL;
3098 dst_entry->offset = 0;
3099 if (src_entry->cred != NULL) {
3100 dst_entry->cred = curthread->td_ucred;
3101 crhold(dst_entry->cred);
3102 *fork_charge += size;
3106 pmap_copy(dst_map->pmap, src_map->pmap, dst_entry->start,
3107 dst_entry->end - dst_entry->start, src_entry->start);
3110 * Of course, wired down pages can't be set copy-on-write.
3111 * Cause wired pages to be copied into the new map by
3112 * simulating faults (the new pages are pageable)
3114 vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry,
3120 * vmspace_map_entry_forked:
3121 * Update the newly-forked vmspace each time a map entry is inherited
3122 * or copied. The values for vm_dsize and vm_tsize are approximate
3123 * (and mostly-obsolete ideas in the face of mmap(2) et al.)
3126 vmspace_map_entry_forked(const struct vmspace *vm1, struct vmspace *vm2,
3127 vm_map_entry_t entry)
3129 vm_size_t entrysize;
3132 entrysize = entry->end - entry->start;
3133 vm2->vm_map.size += entrysize;
3134 if (entry->eflags & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP)) {
3135 vm2->vm_ssize += btoc(entrysize);
3136 } else if (entry->start >= (vm_offset_t)vm1->vm_daddr &&
3137 entry->start < (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)) {
3138 newend = MIN(entry->end,
3139 (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize));
3140 vm2->vm_dsize += btoc(newend - entry->start);
3141 } else if (entry->start >= (vm_offset_t)vm1->vm_taddr &&
3142 entry->start < (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)) {
3143 newend = MIN(entry->end,
3144 (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize));
3145 vm2->vm_tsize += btoc(newend - entry->start);
3151 * Create a new process vmspace structure and vm_map
3152 * based on those of an existing process. The new map
3153 * is based on the old map, according to the inheritance
3154 * values on the regions in that map.
3156 * XXX It might be worth coalescing the entries added to the new vmspace.
3158 * The source map must not be locked.
3161 vmspace_fork(struct vmspace *vm1, vm_ooffset_t *fork_charge)
3163 struct vmspace *vm2;
3164 vm_map_t new_map, old_map;
3165 vm_map_entry_t new_entry, old_entry;
3169 old_map = &vm1->vm_map;
3170 /* Copy immutable fields of vm1 to vm2. */
3171 vm2 = vmspace_alloc(old_map->min_offset, old_map->max_offset, NULL);
3174 vm2->vm_taddr = vm1->vm_taddr;
3175 vm2->vm_daddr = vm1->vm_daddr;
3176 vm2->vm_maxsaddr = vm1->vm_maxsaddr;
3177 vm_map_lock(old_map);
3179 vm_map_wait_busy(old_map);
3180 new_map = &vm2->vm_map;
3181 locked = vm_map_trylock(new_map); /* trylock to silence WITNESS */
3182 KASSERT(locked, ("vmspace_fork: lock failed"));
3184 old_entry = old_map->header.next;
3186 while (old_entry != &old_map->header) {
3187 if (old_entry->eflags & MAP_ENTRY_IS_SUB_MAP)
3188 panic("vm_map_fork: encountered a submap");
3190 switch (old_entry->inheritance) {
3191 case VM_INHERIT_NONE:
3194 case VM_INHERIT_SHARE:
3196 * Clone the entry, creating the shared object if necessary.
3198 object = old_entry->object.vm_object;
3199 if (object == NULL) {
3200 object = vm_object_allocate(OBJT_DEFAULT,
3201 atop(old_entry->end - old_entry->start));
3202 old_entry->object.vm_object = object;
3203 old_entry->offset = 0;
3204 if (old_entry->cred != NULL) {
3205 object->cred = old_entry->cred;
3206 object->charge = old_entry->end -
3208 old_entry->cred = NULL;
3213 * Add the reference before calling vm_object_shadow
3214 * to insure that a shadow object is created.
3216 vm_object_reference(object);
3217 if (old_entry->eflags & MAP_ENTRY_NEEDS_COPY) {
3218 vm_object_shadow(&old_entry->object.vm_object,
3220 old_entry->end - old_entry->start);
3221 old_entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
3222 /* Transfer the second reference too. */
3223 vm_object_reference(
3224 old_entry->object.vm_object);
3227 * As in vm_map_simplify_entry(), the
3228 * vnode lock will not be acquired in
3229 * this call to vm_object_deallocate().
3231 vm_object_deallocate(object);
3232 object = old_entry->object.vm_object;
3234 VM_OBJECT_WLOCK(object);
3235 vm_object_clear_flag(object, OBJ_ONEMAPPING);
3236 if (old_entry->cred != NULL) {
3237 KASSERT(object->cred == NULL, ("vmspace_fork both cred"));
3238 object->cred = old_entry->cred;
3239 object->charge = old_entry->end - old_entry->start;
3240 old_entry->cred = NULL;
3244 * Assert the correct state of the vnode
3245 * v_writecount while the object is locked, to
3246 * not relock it later for the assertion
3249 if (old_entry->eflags & MAP_ENTRY_VN_WRITECNT &&
3250 object->type == OBJT_VNODE) {
3251 KASSERT(((struct vnode *)object->handle)->
3253 ("vmspace_fork: v_writecount %p", object));
3254 KASSERT(object->un_pager.vnp.writemappings > 0,
3255 ("vmspace_fork: vnp.writecount %p",
3258 VM_OBJECT_WUNLOCK(object);
3261 * Clone the entry, referencing the shared object.
3263 new_entry = vm_map_entry_create(new_map);
3264 *new_entry = *old_entry;
3265 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED |
3266 MAP_ENTRY_IN_TRANSITION);
3267 new_entry->wiring_thread = NULL;
3268 new_entry->wired_count = 0;
3269 if (new_entry->eflags & MAP_ENTRY_VN_WRITECNT) {
3270 vnode_pager_update_writecount(object,
3271 new_entry->start, new_entry->end);
3275 * Insert the entry into the new map -- we know we're
3276 * inserting at the end of the new map.
3278 vm_map_entry_link(new_map, new_map->header.prev,
3280 vmspace_map_entry_forked(vm1, vm2, new_entry);
3283 * Update the physical map
3285 pmap_copy(new_map->pmap, old_map->pmap,
3287 (old_entry->end - old_entry->start),
3291 case VM_INHERIT_COPY:
3293 * Clone the entry and link into the map.
3295 new_entry = vm_map_entry_create(new_map);
3296 *new_entry = *old_entry;
3298 * Copied entry is COW over the old object.
3300 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED |
3301 MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_VN_WRITECNT);
3302 new_entry->wiring_thread = NULL;
3303 new_entry->wired_count = 0;
3304 new_entry->object.vm_object = NULL;
3305 new_entry->cred = NULL;
3306 vm_map_entry_link(new_map, new_map->header.prev,
3308 vmspace_map_entry_forked(vm1, vm2, new_entry);
3309 vm_map_copy_entry(old_map, new_map, old_entry,
3310 new_entry, fork_charge);
3313 old_entry = old_entry->next;
3316 * Use inlined vm_map_unlock() to postpone handling the deferred
3317 * map entries, which cannot be done until both old_map and
3318 * new_map locks are released.
3320 sx_xunlock(&old_map->lock);
3321 sx_xunlock(&new_map->lock);
3322 vm_map_process_deferred();
3328 vm_map_stack(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
3329 vm_prot_t prot, vm_prot_t max, int cow)
3331 vm_map_entry_t new_entry, prev_entry;
3332 vm_offset_t bot, top;
3333 vm_size_t growsize, init_ssize;
3335 rlim_t lmemlim, vmemlim;
3338 * The stack orientation is piggybacked with the cow argument.
3339 * Extract it into orient and mask the cow argument so that we
3340 * don't pass it around further.
3341 * NOTE: We explicitly allow bi-directional stacks.
3343 orient = cow & (MAP_STACK_GROWS_DOWN|MAP_STACK_GROWS_UP);
3344 KASSERT(orient != 0, ("No stack grow direction"));
3346 if (addrbos < vm_map_min(map) ||
3347 addrbos > vm_map_max(map) ||
3348 addrbos + max_ssize < addrbos)
3349 return (KERN_NO_SPACE);
3351 growsize = sgrowsiz;
3352 init_ssize = (max_ssize < growsize) ? max_ssize : growsize;
3355 lmemlim = lim_cur(curproc, RLIMIT_MEMLOCK);
3356 vmemlim = lim_cur(curproc, RLIMIT_VMEM);
3357 PROC_UNLOCK(curproc);
3361 /* If addr is already mapped, no go */
3362 if (vm_map_lookup_entry(map, addrbos, &prev_entry)) {
3364 return (KERN_NO_SPACE);
3367 if (!old_mlock && map->flags & MAP_WIREFUTURE) {
3368 if (ptoa(pmap_wired_count(map->pmap)) + init_ssize > lmemlim) {
3370 return (KERN_NO_SPACE);
3374 /* If we would blow our VMEM resource limit, no go */
3375 if (map->size + init_ssize > vmemlim) {
3377 return (KERN_NO_SPACE);
3381 * If we can't accomodate max_ssize in the current mapping, no go.
3382 * However, we need to be aware that subsequent user mappings might
3383 * map into the space we have reserved for stack, and currently this
3384 * space is not protected.
3386 * Hopefully we will at least detect this condition when we try to
3389 if ((prev_entry->next != &map->header) &&
3390 (prev_entry->next->start < addrbos + max_ssize)) {
3392 return (KERN_NO_SPACE);
3396 * We initially map a stack of only init_ssize. We will grow as
3397 * needed later. Depending on the orientation of the stack (i.e.
3398 * the grow direction) we either map at the top of the range, the
3399 * bottom of the range or in the middle.
3401 * Note: we would normally expect prot and max to be VM_PROT_ALL,
3402 * and cow to be 0. Possibly we should eliminate these as input
3403 * parameters, and just pass these values here in the insert call.
3405 if (orient == MAP_STACK_GROWS_DOWN)
3406 bot = addrbos + max_ssize - init_ssize;
3407 else if (orient == MAP_STACK_GROWS_UP)
3410 bot = round_page(addrbos + max_ssize/2 - init_ssize/2);
3411 top = bot + init_ssize;
3412 rv = vm_map_insert(map, NULL, 0, bot, top, prot, max, cow);
3414 /* Now set the avail_ssize amount. */
3415 if (rv == KERN_SUCCESS) {
3416 if (prev_entry != &map->header)
3417 vm_map_clip_end(map, prev_entry, bot);
3418 new_entry = prev_entry->next;
3419 if (new_entry->end != top || new_entry->start != bot)
3420 panic("Bad entry start/end for new stack entry");
3422 new_entry->avail_ssize = max_ssize - init_ssize;
3423 if (orient & MAP_STACK_GROWS_DOWN)
3424 new_entry->eflags |= MAP_ENTRY_GROWS_DOWN;
3425 if (orient & MAP_STACK_GROWS_UP)
3426 new_entry->eflags |= MAP_ENTRY_GROWS_UP;
3433 static int stack_guard_page = 0;
3434 TUNABLE_INT("security.bsd.stack_guard_page", &stack_guard_page);
3435 SYSCTL_INT(_security_bsd, OID_AUTO, stack_guard_page, CTLFLAG_RW,
3436 &stack_guard_page, 0,
3437 "Insert stack guard page ahead of the growable segments.");
3439 /* Attempts to grow a vm stack entry. Returns KERN_SUCCESS if the
3440 * desired address is already mapped, or if we successfully grow
3441 * the stack. Also returns KERN_SUCCESS if addr is outside the
3442 * stack range (this is strange, but preserves compatibility with
3443 * the grow function in vm_machdep.c).
3446 vm_map_growstack(struct proc *p, vm_offset_t addr)
3448 vm_map_entry_t next_entry, prev_entry;
3449 vm_map_entry_t new_entry, stack_entry;
3450 struct vmspace *vm = p->p_vmspace;
3451 vm_map_t map = &vm->vm_map;
3454 size_t grow_amount, max_grow;
3455 rlim_t lmemlim, stacklim, vmemlim;
3456 int is_procstack, rv;
3467 lmemlim = lim_cur(p, RLIMIT_MEMLOCK);
3468 stacklim = lim_cur(p, RLIMIT_STACK);
3469 vmemlim = lim_cur(p, RLIMIT_VMEM);
3472 vm_map_lock_read(map);
3474 /* If addr is already in the entry range, no need to grow.*/
3475 if (vm_map_lookup_entry(map, addr, &prev_entry)) {
3476 vm_map_unlock_read(map);
3477 return (KERN_SUCCESS);
3480 next_entry = prev_entry->next;
3481 if (!(prev_entry->eflags & MAP_ENTRY_GROWS_UP)) {
3483 * This entry does not grow upwards. Since the address lies
3484 * beyond this entry, the next entry (if one exists) has to
3485 * be a downward growable entry. The entry list header is
3486 * never a growable entry, so it suffices to check the flags.
3488 if (!(next_entry->eflags & MAP_ENTRY_GROWS_DOWN)) {
3489 vm_map_unlock_read(map);
3490 return (KERN_SUCCESS);
3492 stack_entry = next_entry;
3495 * This entry grows upward. If the next entry does not at
3496 * least grow downwards, this is the entry we need to grow.
3497 * otherwise we have two possible choices and we have to
3500 if (next_entry->eflags & MAP_ENTRY_GROWS_DOWN) {
3502 * We have two choices; grow the entry closest to
3503 * the address to minimize the amount of growth.
3505 if (addr - prev_entry->end <= next_entry->start - addr)
3506 stack_entry = prev_entry;
3508 stack_entry = next_entry;
3510 stack_entry = prev_entry;
3513 if (stack_entry == next_entry) {
3514 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_DOWN, ("foo"));
3515 KASSERT(addr < stack_entry->start, ("foo"));
3516 end = (prev_entry != &map->header) ? prev_entry->end :
3517 stack_entry->start - stack_entry->avail_ssize;
3518 grow_amount = roundup(stack_entry->start - addr, PAGE_SIZE);
3519 max_grow = stack_entry->start - end;
3521 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_UP, ("foo"));
3522 KASSERT(addr >= stack_entry->end, ("foo"));
3523 end = (next_entry != &map->header) ? next_entry->start :
3524 stack_entry->end + stack_entry->avail_ssize;
3525 grow_amount = roundup(addr + 1 - stack_entry->end, PAGE_SIZE);
3526 max_grow = end - stack_entry->end;
3529 if (grow_amount > stack_entry->avail_ssize) {
3530 vm_map_unlock_read(map);
3531 return (KERN_NO_SPACE);
3535 * If there is no longer enough space between the entries nogo, and
3536 * adjust the available space. Note: this should only happen if the
3537 * user has mapped into the stack area after the stack was created,
3538 * and is probably an error.
3540 * This also effectively destroys any guard page the user might have
3541 * intended by limiting the stack size.
3543 if (grow_amount + (stack_guard_page ? PAGE_SIZE : 0) > max_grow) {
3544 if (vm_map_lock_upgrade(map))
3547 stack_entry->avail_ssize = max_grow;
3550 return (KERN_NO_SPACE);
3553 is_procstack = (addr >= (vm_offset_t)vm->vm_maxsaddr) ? 1 : 0;
3556 * If this is the main process stack, see if we're over the stack
3559 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) {
3560 vm_map_unlock_read(map);
3561 return (KERN_NO_SPACE);
3566 racct_set(p, RACCT_STACK, ctob(vm->vm_ssize) + grow_amount)) {
3568 vm_map_unlock_read(map);
3569 return (KERN_NO_SPACE);
3574 /* Round up the grow amount modulo sgrowsiz */
3575 growsize = sgrowsiz;
3576 grow_amount = roundup(grow_amount, growsize);
3577 if (grow_amount > stack_entry->avail_ssize)
3578 grow_amount = stack_entry->avail_ssize;
3579 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) {
3580 grow_amount = trunc_page((vm_size_t)stacklim) -
3585 limit = racct_get_available(p, RACCT_STACK);
3587 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > limit))
3588 grow_amount = limit - ctob(vm->vm_ssize);
3590 if (!old_mlock && map->flags & MAP_WIREFUTURE) {
3591 if (ptoa(pmap_wired_count(map->pmap)) + grow_amount > lmemlim) {
3592 vm_map_unlock_read(map);
3598 if (racct_set(p, RACCT_MEMLOCK,
3599 ptoa(pmap_wired_count(map->pmap)) + grow_amount)) {
3601 vm_map_unlock_read(map);
3608 /* If we would blow our VMEM resource limit, no go */
3609 if (map->size + grow_amount > vmemlim) {
3610 vm_map_unlock_read(map);
3616 if (racct_set(p, RACCT_VMEM, map->size + grow_amount)) {
3618 vm_map_unlock_read(map);
3625 if (vm_map_lock_upgrade(map))
3628 if (stack_entry == next_entry) {
3632 /* Get the preliminary new entry start value */
3633 addr = stack_entry->start - grow_amount;
3636 * If this puts us into the previous entry, cut back our
3637 * growth to the available space. Also, see the note above.
3640 stack_entry->avail_ssize = max_grow;
3642 if (stack_guard_page)
3646 rv = vm_map_insert(map, NULL, 0, addr, stack_entry->start,
3647 next_entry->protection, next_entry->max_protection, 0);
3649 /* Adjust the available stack space by the amount we grew. */
3650 if (rv == KERN_SUCCESS) {
3651 if (prev_entry != &map->header)
3652 vm_map_clip_end(map, prev_entry, addr);
3653 new_entry = prev_entry->next;
3654 KASSERT(new_entry == stack_entry->prev, ("foo"));
3655 KASSERT(new_entry->end == stack_entry->start, ("foo"));
3656 KASSERT(new_entry->start == addr, ("foo"));
3657 grow_amount = new_entry->end - new_entry->start;
3658 new_entry->avail_ssize = stack_entry->avail_ssize -
3660 stack_entry->eflags &= ~MAP_ENTRY_GROWS_DOWN;
3661 new_entry->eflags |= MAP_ENTRY_GROWS_DOWN;
3667 addr = stack_entry->end + grow_amount;
3670 * If this puts us into the next entry, cut back our growth
3671 * to the available space. Also, see the note above.
3674 stack_entry->avail_ssize = end - stack_entry->end;
3676 if (stack_guard_page)
3680 grow_amount = addr - stack_entry->end;
3681 cred = stack_entry->cred;
3682 if (cred == NULL && stack_entry->object.vm_object != NULL)
3683 cred = stack_entry->object.vm_object->cred;
3684 if (cred != NULL && !swap_reserve_by_cred(grow_amount, cred))
3686 /* Grow the underlying object if applicable. */
3687 else if (stack_entry->object.vm_object == NULL ||
3688 vm_object_coalesce(stack_entry->object.vm_object,
3689 stack_entry->offset,
3690 (vm_size_t)(stack_entry->end - stack_entry->start),
3691 (vm_size_t)grow_amount, cred != NULL)) {
3692 map->size += (addr - stack_entry->end);
3693 /* Update the current entry. */
3694 stack_entry->end = addr;
3695 stack_entry->avail_ssize -= grow_amount;
3696 vm_map_entry_resize_free(map, stack_entry);
3699 if (next_entry != &map->header)
3700 vm_map_clip_start(map, next_entry, addr);
3705 if (rv == KERN_SUCCESS && is_procstack)
3706 vm->vm_ssize += btoc(grow_amount);
3711 * Heed the MAP_WIREFUTURE flag if it was set for this process.
3713 if (rv == KERN_SUCCESS && (map->flags & MAP_WIREFUTURE)) {
3715 (stack_entry == next_entry) ? addr : addr - grow_amount,
3716 (stack_entry == next_entry) ? stack_entry->start : addr,
3717 (p->p_flag & P_SYSTEM)
3718 ? VM_MAP_WIRE_SYSTEM|VM_MAP_WIRE_NOHOLES
3719 : VM_MAP_WIRE_USER|VM_MAP_WIRE_NOHOLES);
3724 if (rv != KERN_SUCCESS) {
3726 error = racct_set(p, RACCT_VMEM, map->size);
3727 KASSERT(error == 0, ("decreasing RACCT_VMEM failed"));
3729 error = racct_set(p, RACCT_MEMLOCK,
3730 ptoa(pmap_wired_count(map->pmap)));
3731 KASSERT(error == 0, ("decreasing RACCT_MEMLOCK failed"));
3733 error = racct_set(p, RACCT_STACK, ctob(vm->vm_ssize));
3734 KASSERT(error == 0, ("decreasing RACCT_STACK failed"));
3743 * Unshare the specified VM space for exec. If other processes are
3744 * mapped to it, then create a new one. The new vmspace is null.
3747 vmspace_exec(struct proc *p, vm_offset_t minuser, vm_offset_t maxuser)
3749 struct vmspace *oldvmspace = p->p_vmspace;
3750 struct vmspace *newvmspace;
3752 newvmspace = vmspace_alloc(minuser, maxuser, NULL);
3753 if (newvmspace == NULL)
3755 newvmspace->vm_swrss = oldvmspace->vm_swrss;
3757 * This code is written like this for prototype purposes. The
3758 * goal is to avoid running down the vmspace here, but let the
3759 * other process's that are still using the vmspace to finally
3760 * run it down. Even though there is little or no chance of blocking
3761 * here, it is a good idea to keep this form for future mods.
3763 PROC_VMSPACE_LOCK(p);
3764 p->p_vmspace = newvmspace;
3765 PROC_VMSPACE_UNLOCK(p);
3766 if (p == curthread->td_proc)
3767 pmap_activate(curthread);
3768 vmspace_free(oldvmspace);
3773 * Unshare the specified VM space for forcing COW. This
3774 * is called by rfork, for the (RFMEM|RFPROC) == 0 case.
3777 vmspace_unshare(struct proc *p)
3779 struct vmspace *oldvmspace = p->p_vmspace;
3780 struct vmspace *newvmspace;
3781 vm_ooffset_t fork_charge;
3783 if (oldvmspace->vm_refcnt == 1)
3786 newvmspace = vmspace_fork(oldvmspace, &fork_charge);
3787 if (newvmspace == NULL)
3789 if (!swap_reserve_by_cred(fork_charge, p->p_ucred)) {
3790 vmspace_free(newvmspace);
3793 PROC_VMSPACE_LOCK(p);
3794 p->p_vmspace = newvmspace;
3795 PROC_VMSPACE_UNLOCK(p);
3796 if (p == curthread->td_proc)
3797 pmap_activate(curthread);
3798 vmspace_free(oldvmspace);
3805 * Finds the VM object, offset, and
3806 * protection for a given virtual address in the
3807 * specified map, assuming a page fault of the
3810 * Leaves the map in question locked for read; return
3811 * values are guaranteed until a vm_map_lookup_done
3812 * call is performed. Note that the map argument
3813 * is in/out; the returned map must be used in
3814 * the call to vm_map_lookup_done.
3816 * A handle (out_entry) is returned for use in
3817 * vm_map_lookup_done, to make that fast.
3819 * If a lookup is requested with "write protection"
3820 * specified, the map may be changed to perform virtual
3821 * copying operations, although the data referenced will
3825 vm_map_lookup(vm_map_t *var_map, /* IN/OUT */
3827 vm_prot_t fault_typea,
3828 vm_map_entry_t *out_entry, /* OUT */
3829 vm_object_t *object, /* OUT */
3830 vm_pindex_t *pindex, /* OUT */
3831 vm_prot_t *out_prot, /* OUT */
3832 boolean_t *wired) /* OUT */
3834 vm_map_entry_t entry;
3835 vm_map_t map = *var_map;
3837 vm_prot_t fault_type = fault_typea;
3838 vm_object_t eobject;
3844 vm_map_lock_read(map);
3847 * Lookup the faulting address.
3849 if (!vm_map_lookup_entry(map, vaddr, out_entry)) {
3850 vm_map_unlock_read(map);
3851 return (KERN_INVALID_ADDRESS);
3859 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
3860 vm_map_t old_map = map;
3862 *var_map = map = entry->object.sub_map;
3863 vm_map_unlock_read(old_map);
3868 * Check whether this task is allowed to have this page.
3870 prot = entry->protection;
3871 fault_type &= (VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE);
3872 if ((fault_type & prot) != fault_type || prot == VM_PROT_NONE) {
3873 vm_map_unlock_read(map);
3874 return (KERN_PROTECTION_FAILURE);
3876 if ((entry->eflags & MAP_ENTRY_USER_WIRED) &&
3877 (entry->eflags & MAP_ENTRY_COW) &&
3878 (fault_type & VM_PROT_WRITE)) {
3879 vm_map_unlock_read(map);
3880 return (KERN_PROTECTION_FAILURE);
3882 if ((fault_typea & VM_PROT_COPY) != 0 &&
3883 (entry->max_protection & VM_PROT_WRITE) == 0 &&
3884 (entry->eflags & MAP_ENTRY_COW) == 0) {
3885 vm_map_unlock_read(map);
3886 return (KERN_PROTECTION_FAILURE);
3890 * If this page is not pageable, we have to get it for all possible
3893 *wired = (entry->wired_count != 0);
3895 fault_type = entry->protection;
3896 size = entry->end - entry->start;
3898 * If the entry was copy-on-write, we either ...
3900 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
3902 * If we want to write the page, we may as well handle that
3903 * now since we've got the map locked.
3905 * If we don't need to write the page, we just demote the
3906 * permissions allowed.
3908 if ((fault_type & VM_PROT_WRITE) != 0 ||
3909 (fault_typea & VM_PROT_COPY) != 0) {
3911 * Make a new object, and place it in the object
3912 * chain. Note that no new references have appeared
3913 * -- one just moved from the map to the new
3916 if (vm_map_lock_upgrade(map))
3919 if (entry->cred == NULL) {
3921 * The debugger owner is charged for
3924 cred = curthread->td_ucred;
3926 if (!swap_reserve_by_cred(size, cred)) {
3929 return (KERN_RESOURCE_SHORTAGE);
3933 vm_object_shadow(&entry->object.vm_object,
3934 &entry->offset, size);
3935 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
3936 eobject = entry->object.vm_object;
3937 if (eobject->cred != NULL) {
3939 * The object was not shadowed.
3941 swap_release_by_cred(size, entry->cred);
3942 crfree(entry->cred);
3944 } else if (entry->cred != NULL) {
3945 VM_OBJECT_WLOCK(eobject);
3946 eobject->cred = entry->cred;
3947 eobject->charge = size;
3948 VM_OBJECT_WUNLOCK(eobject);
3952 vm_map_lock_downgrade(map);
3955 * We're attempting to read a copy-on-write page --
3956 * don't allow writes.
3958 prot &= ~VM_PROT_WRITE;
3963 * Create an object if necessary.
3965 if (entry->object.vm_object == NULL &&
3967 if (vm_map_lock_upgrade(map))
3969 entry->object.vm_object = vm_object_allocate(OBJT_DEFAULT,
3972 if (entry->cred != NULL) {
3973 VM_OBJECT_WLOCK(entry->object.vm_object);
3974 entry->object.vm_object->cred = entry->cred;
3975 entry->object.vm_object->charge = size;
3976 VM_OBJECT_WUNLOCK(entry->object.vm_object);
3979 vm_map_lock_downgrade(map);
3983 * Return the object/offset from this entry. If the entry was
3984 * copy-on-write or empty, it has been fixed up.
3986 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
3987 *object = entry->object.vm_object;
3990 return (KERN_SUCCESS);
3994 * vm_map_lookup_locked:
3996 * Lookup the faulting address. A version of vm_map_lookup that returns
3997 * KERN_FAILURE instead of blocking on map lock or memory allocation.
4000 vm_map_lookup_locked(vm_map_t *var_map, /* IN/OUT */
4002 vm_prot_t fault_typea,
4003 vm_map_entry_t *out_entry, /* OUT */
4004 vm_object_t *object, /* OUT */
4005 vm_pindex_t *pindex, /* OUT */
4006 vm_prot_t *out_prot, /* OUT */
4007 boolean_t *wired) /* OUT */
4009 vm_map_entry_t entry;
4010 vm_map_t map = *var_map;
4012 vm_prot_t fault_type = fault_typea;
4015 * Lookup the faulting address.
4017 if (!vm_map_lookup_entry(map, vaddr, out_entry))
4018 return (KERN_INVALID_ADDRESS);
4023 * Fail if the entry refers to a submap.
4025 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP)
4026 return (KERN_FAILURE);
4029 * Check whether this task is allowed to have this page.
4031 prot = entry->protection;
4032 fault_type &= VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE;
4033 if ((fault_type & prot) != fault_type)
4034 return (KERN_PROTECTION_FAILURE);
4035 if ((entry->eflags & MAP_ENTRY_USER_WIRED) &&
4036 (entry->eflags & MAP_ENTRY_COW) &&
4037 (fault_type & VM_PROT_WRITE))
4038 return (KERN_PROTECTION_FAILURE);
4041 * If this page is not pageable, we have to get it for all possible
4044 *wired = (entry->wired_count != 0);
4046 fault_type = entry->protection;
4048 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
4050 * Fail if the entry was copy-on-write for a write fault.
4052 if (fault_type & VM_PROT_WRITE)
4053 return (KERN_FAILURE);
4055 * We're attempting to read a copy-on-write page --
4056 * don't allow writes.
4058 prot &= ~VM_PROT_WRITE;
4062 * Fail if an object should be created.
4064 if (entry->object.vm_object == NULL && !map->system_map)
4065 return (KERN_FAILURE);
4068 * Return the object/offset from this entry. If the entry was
4069 * copy-on-write or empty, it has been fixed up.
4071 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
4072 *object = entry->object.vm_object;
4075 return (KERN_SUCCESS);
4079 * vm_map_lookup_done:
4081 * Releases locks acquired by a vm_map_lookup
4082 * (according to the handle returned by that lookup).
4085 vm_map_lookup_done(vm_map_t map, vm_map_entry_t entry)
4088 * Unlock the main-level map
4090 vm_map_unlock_read(map);
4093 #include "opt_ddb.h"
4095 #include <sys/kernel.h>
4097 #include <ddb/ddb.h>
4100 vm_map_print(vm_map_t map)
4102 vm_map_entry_t entry;
4104 db_iprintf("Task map %p: pmap=%p, nentries=%d, version=%u\n",
4106 (void *)map->pmap, map->nentries, map->timestamp);
4109 for (entry = map->header.next; entry != &map->header;
4110 entry = entry->next) {
4111 db_iprintf("map entry %p: start=%p, end=%p\n",
4112 (void *)entry, (void *)entry->start, (void *)entry->end);
4114 static char *inheritance_name[4] =
4115 {"share", "copy", "none", "donate_copy"};
4117 db_iprintf(" prot=%x/%x/%s",
4119 entry->max_protection,
4120 inheritance_name[(int)(unsigned char)entry->inheritance]);
4121 if (entry->wired_count != 0)
4122 db_printf(", wired");
4124 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
4125 db_printf(", share=%p, offset=0x%jx\n",
4126 (void *)entry->object.sub_map,
4127 (uintmax_t)entry->offset);
4128 if ((entry->prev == &map->header) ||
4129 (entry->prev->object.sub_map !=
4130 entry->object.sub_map)) {
4132 vm_map_print((vm_map_t)entry->object.sub_map);
4136 if (entry->cred != NULL)
4137 db_printf(", ruid %d", entry->cred->cr_ruid);
4138 db_printf(", object=%p, offset=0x%jx",
4139 (void *)entry->object.vm_object,
4140 (uintmax_t)entry->offset);
4141 if (entry->object.vm_object && entry->object.vm_object->cred)
4142 db_printf(", obj ruid %d charge %jx",
4143 entry->object.vm_object->cred->cr_ruid,
4144 (uintmax_t)entry->object.vm_object->charge);
4145 if (entry->eflags & MAP_ENTRY_COW)
4146 db_printf(", copy (%s)",
4147 (entry->eflags & MAP_ENTRY_NEEDS_COPY) ? "needed" : "done");
4150 if ((entry->prev == &map->header) ||
4151 (entry->prev->object.vm_object !=
4152 entry->object.vm_object)) {
4154 vm_object_print((db_expr_t)(intptr_t)
4155 entry->object.vm_object,
4164 DB_SHOW_COMMAND(map, map)
4168 db_printf("usage: show map <addr>\n");
4171 vm_map_print((vm_map_t)addr);
4174 DB_SHOW_COMMAND(procvm, procvm)
4179 p = (struct proc *) addr;
4184 db_printf("p = %p, vmspace = %p, map = %p, pmap = %p\n",
4185 (void *)p, (void *)p->p_vmspace, (void *)&p->p_vmspace->vm_map,
4186 (void *)vmspace_pmap(p->p_vmspace));
4188 vm_map_print((vm_map_t)&p->p_vmspace->vm_map);