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 void vmspace_zfini(void *mem, int size);
131 static int vm_map_zinit(void *mem, int ize, int flags);
132 static void vm_map_zfini(void *mem, int size);
133 static void _vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min,
135 static void vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map);
136 static void vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry);
138 static void vm_map_zdtor(void *mem, int size, void *arg);
139 static void vmspace_zdtor(void *mem, int size, void *arg);
142 #define ENTRY_CHARGED(e) ((e)->cred != NULL || \
143 ((e)->object.vm_object != NULL && (e)->object.vm_object->cred != NULL && \
144 !((e)->eflags & MAP_ENTRY_NEEDS_COPY)))
147 * PROC_VMSPACE_{UN,}LOCK() can be a noop as long as vmspaces are type
150 #define PROC_VMSPACE_LOCK(p) do { } while (0)
151 #define PROC_VMSPACE_UNLOCK(p) do { } while (0)
154 * VM_MAP_RANGE_CHECK: [ internal use only ]
156 * Asserts that the starting and ending region
157 * addresses fall within the valid range of the map.
159 #define VM_MAP_RANGE_CHECK(map, start, end) \
161 if (start < vm_map_min(map)) \
162 start = vm_map_min(map); \
163 if (end > vm_map_max(map)) \
164 end = vm_map_max(map); \
172 * Initialize the vm_map module. Must be called before
173 * any other vm_map routines.
175 * Map and entry structures are allocated from the general
176 * purpose memory pool with some exceptions:
178 * - The kernel map and kmem submap are allocated statically.
179 * - Kernel map entries are allocated out of a static pool.
181 * These restrictions are necessary since malloc() uses the
182 * maps and requires map entries.
188 mtx_init(&map_sleep_mtx, "vm map sleep mutex", NULL, MTX_DEF);
189 mapzone = uma_zcreate("MAP", sizeof(struct vm_map), NULL,
195 vm_map_zinit, vm_map_zfini, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
196 uma_prealloc(mapzone, MAX_KMAP);
197 kmapentzone = uma_zcreate("KMAP ENTRY", sizeof(struct vm_map_entry),
198 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR,
199 UMA_ZONE_MTXCLASS | UMA_ZONE_VM);
200 mapentzone = uma_zcreate("MAP ENTRY", sizeof(struct vm_map_entry),
201 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
202 vmspace_zone = uma_zcreate("VMSPACE", sizeof(struct vmspace), NULL,
208 vmspace_zinit, vmspace_zfini, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
212 vmspace_zfini(void *mem, int size)
216 vm = (struct vmspace *)mem;
217 vm_map_zfini(&vm->vm_map, sizeof(vm->vm_map));
221 vmspace_zinit(void *mem, int size, int flags)
225 vm = (struct vmspace *)mem;
227 vm->vm_map.pmap = NULL;
228 (void)vm_map_zinit(&vm->vm_map, sizeof(vm->vm_map), flags);
229 PMAP_LOCK_INIT(vmspace_pmap(vm));
234 vm_map_zfini(void *mem, int size)
239 mtx_destroy(&map->system_mtx);
240 sx_destroy(&map->lock);
244 vm_map_zinit(void *mem, int size, int flags)
249 memset(map, 0, sizeof(*map));
250 mtx_init(&map->system_mtx, "vm map (system)", NULL, MTX_DEF | MTX_DUPOK);
251 sx_init(&map->lock, "vm map (user)");
257 vmspace_zdtor(void *mem, int size, void *arg)
261 vm = (struct vmspace *)mem;
263 vm_map_zdtor(&vm->vm_map, sizeof(vm->vm_map), arg);
266 vm_map_zdtor(void *mem, int size, void *arg)
271 KASSERT(map->nentries == 0,
272 ("map %p nentries == %d on free.",
273 map, map->nentries));
274 KASSERT(map->size == 0,
275 ("map %p size == %lu on free.",
276 map, (unsigned long)map->size));
278 #endif /* INVARIANTS */
281 * Allocate a vmspace structure, including a vm_map and pmap,
282 * and initialize those structures. The refcnt is set to 1.
285 vmspace_alloc(min, max)
286 vm_offset_t min, max;
290 vm = uma_zalloc(vmspace_zone, M_WAITOK);
291 if (vm->vm_map.pmap == NULL && !pmap_pinit(vmspace_pmap(vm))) {
292 uma_zfree(vmspace_zone, vm);
295 CTR1(KTR_VM, "vmspace_alloc: %p", vm);
296 _vm_map_init(&vm->vm_map, vmspace_pmap(vm), min, max);
310 vmspace_container_reset(struct proc *p)
315 racct_set(p, RACCT_DATA, 0);
316 racct_set(p, RACCT_STACK, 0);
317 racct_set(p, RACCT_RSS, 0);
318 racct_set(p, RACCT_MEMLOCK, 0);
319 racct_set(p, RACCT_VMEM, 0);
325 vmspace_dofree(struct vmspace *vm)
328 CTR1(KTR_VM, "vmspace_free: %p", vm);
331 * Make sure any SysV shm is freed, it might not have been in
337 * Lock the map, to wait out all other references to it.
338 * Delete all of the mappings and pages they hold, then call
339 * the pmap module to reclaim anything left.
341 (void)vm_map_remove(&vm->vm_map, vm->vm_map.min_offset,
342 vm->vm_map.max_offset);
344 pmap_release(vmspace_pmap(vm));
345 vm->vm_map.pmap = NULL;
346 uma_zfree(vmspace_zone, vm);
350 vmspace_free(struct vmspace *vm)
353 if (vm->vm_refcnt == 0)
354 panic("vmspace_free: attempt to free already freed vmspace");
356 if (atomic_fetchadd_int(&vm->vm_refcnt, -1) == 1)
361 vmspace_exitfree(struct proc *p)
365 PROC_VMSPACE_LOCK(p);
368 PROC_VMSPACE_UNLOCK(p);
369 KASSERT(vm == &vmspace0, ("vmspace_exitfree: wrong vmspace"));
374 vmspace_exit(struct thread *td)
381 * Release user portion of address space.
382 * This releases references to vnodes,
383 * which could cause I/O if the file has been unlinked.
384 * Need to do this early enough that we can still sleep.
386 * The last exiting process to reach this point releases as
387 * much of the environment as it can. vmspace_dofree() is the
388 * slower fallback in case another process had a temporary
389 * reference to the vmspace.
394 atomic_add_int(&vmspace0.vm_refcnt, 1);
396 refcnt = vm->vm_refcnt;
397 if (refcnt > 1 && p->p_vmspace != &vmspace0) {
398 /* Switch now since other proc might free vmspace */
399 PROC_VMSPACE_LOCK(p);
400 p->p_vmspace = &vmspace0;
401 PROC_VMSPACE_UNLOCK(p);
404 } while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt - 1));
406 if (p->p_vmspace != vm) {
407 /* vmspace not yet freed, switch back */
408 PROC_VMSPACE_LOCK(p);
410 PROC_VMSPACE_UNLOCK(p);
413 pmap_remove_pages(vmspace_pmap(vm));
414 /* Switch now since this proc will free vmspace */
415 PROC_VMSPACE_LOCK(p);
416 p->p_vmspace = &vmspace0;
417 PROC_VMSPACE_UNLOCK(p);
421 vmspace_container_reset(p);
424 /* Acquire reference to vmspace owned by another process. */
427 vmspace_acquire_ref(struct proc *p)
432 PROC_VMSPACE_LOCK(p);
435 PROC_VMSPACE_UNLOCK(p);
439 refcnt = vm->vm_refcnt;
440 if (refcnt <= 0) { /* Avoid 0->1 transition */
441 PROC_VMSPACE_UNLOCK(p);
444 } while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt + 1));
445 if (vm != p->p_vmspace) {
446 PROC_VMSPACE_UNLOCK(p);
450 PROC_VMSPACE_UNLOCK(p);
455 _vm_map_lock(vm_map_t map, const char *file, int line)
459 mtx_lock_flags_(&map->system_mtx, 0, file, line);
461 sx_xlock_(&map->lock, file, line);
466 vm_map_process_deferred(void)
469 vm_map_entry_t entry, next;
473 entry = td->td_map_def_user;
474 td->td_map_def_user = NULL;
475 while (entry != NULL) {
477 if ((entry->eflags & MAP_ENTRY_VN_WRITECNT) != 0) {
479 * Decrement the object's writemappings and
480 * possibly the vnode's v_writecount.
482 KASSERT((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0,
483 ("Submap with writecount"));
484 object = entry->object.vm_object;
485 KASSERT(object != NULL, ("No object for writecount"));
486 vnode_pager_release_writecount(object, entry->start,
489 vm_map_entry_deallocate(entry, FALSE);
495 _vm_map_unlock(vm_map_t map, const char *file, int line)
499 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
501 sx_xunlock_(&map->lock, file, line);
502 vm_map_process_deferred();
507 _vm_map_lock_read(vm_map_t map, const char *file, int line)
511 mtx_lock_flags_(&map->system_mtx, 0, file, line);
513 sx_slock_(&map->lock, file, line);
517 _vm_map_unlock_read(vm_map_t map, const char *file, int line)
521 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
523 sx_sunlock_(&map->lock, file, line);
524 vm_map_process_deferred();
529 _vm_map_trylock(vm_map_t map, const char *file, int line)
533 error = map->system_map ?
534 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) :
535 !sx_try_xlock_(&map->lock, file, line);
542 _vm_map_trylock_read(vm_map_t map, const char *file, int line)
546 error = map->system_map ?
547 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) :
548 !sx_try_slock_(&map->lock, file, line);
553 * _vm_map_lock_upgrade: [ internal use only ]
555 * Tries to upgrade a read (shared) lock on the specified map to a write
556 * (exclusive) lock. Returns the value "0" if the upgrade succeeds and a
557 * non-zero value if the upgrade fails. If the upgrade fails, the map is
558 * returned without a read or write lock held.
560 * Requires that the map be read locked.
563 _vm_map_lock_upgrade(vm_map_t map, const char *file, int line)
565 unsigned int last_timestamp;
567 if (map->system_map) {
568 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
570 if (!sx_try_upgrade_(&map->lock, file, line)) {
571 last_timestamp = map->timestamp;
572 sx_sunlock_(&map->lock, file, line);
573 vm_map_process_deferred();
575 * If the map's timestamp does not change while the
576 * map is unlocked, then the upgrade succeeds.
578 sx_xlock_(&map->lock, file, line);
579 if (last_timestamp != map->timestamp) {
580 sx_xunlock_(&map->lock, file, line);
590 _vm_map_lock_downgrade(vm_map_t map, const char *file, int line)
593 if (map->system_map) {
594 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
596 sx_downgrade_(&map->lock, file, line);
602 * Returns a non-zero value if the caller holds a write (exclusive) lock
603 * on the specified map and the value "0" otherwise.
606 vm_map_locked(vm_map_t map)
610 return (mtx_owned(&map->system_mtx));
612 return (sx_xlocked(&map->lock));
617 _vm_map_assert_locked(vm_map_t map, const char *file, int line)
621 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
623 sx_assert_(&map->lock, SA_XLOCKED, file, line);
626 #define VM_MAP_ASSERT_LOCKED(map) \
627 _vm_map_assert_locked(map, LOCK_FILE, LOCK_LINE)
629 #define VM_MAP_ASSERT_LOCKED(map)
633 * _vm_map_unlock_and_wait:
635 * Atomically releases the lock on the specified map and puts the calling
636 * thread to sleep. The calling thread will remain asleep until either
637 * vm_map_wakeup() is performed on the map or the specified timeout is
640 * WARNING! This function does not perform deferred deallocations of
641 * objects and map entries. Therefore, the calling thread is expected to
642 * reacquire the map lock after reawakening and later perform an ordinary
643 * unlock operation, such as vm_map_unlock(), before completing its
644 * operation on the map.
647 _vm_map_unlock_and_wait(vm_map_t map, int timo, const char *file, int line)
650 mtx_lock(&map_sleep_mtx);
652 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
654 sx_xunlock_(&map->lock, file, line);
655 return (msleep(&map->root, &map_sleep_mtx, PDROP | PVM, "vmmaps",
662 * Awaken any threads that have slept on the map using
663 * vm_map_unlock_and_wait().
666 vm_map_wakeup(vm_map_t map)
670 * Acquire and release map_sleep_mtx to prevent a wakeup()
671 * from being performed (and lost) between the map unlock
672 * and the msleep() in _vm_map_unlock_and_wait().
674 mtx_lock(&map_sleep_mtx);
675 mtx_unlock(&map_sleep_mtx);
680 vm_map_busy(vm_map_t map)
683 VM_MAP_ASSERT_LOCKED(map);
688 vm_map_unbusy(vm_map_t map)
691 VM_MAP_ASSERT_LOCKED(map);
692 KASSERT(map->busy, ("vm_map_unbusy: not busy"));
693 if (--map->busy == 0 && (map->flags & MAP_BUSY_WAKEUP)) {
694 vm_map_modflags(map, 0, MAP_BUSY_WAKEUP);
700 vm_map_wait_busy(vm_map_t map)
703 VM_MAP_ASSERT_LOCKED(map);
705 vm_map_modflags(map, MAP_BUSY_WAKEUP, 0);
707 msleep(&map->busy, &map->system_mtx, 0, "mbusy", 0);
709 sx_sleep(&map->busy, &map->lock, 0, "mbusy", 0);
715 vmspace_resident_count(struct vmspace *vmspace)
717 return pmap_resident_count(vmspace_pmap(vmspace));
723 * Creates and returns a new empty VM map with
724 * the given physical map structure, and having
725 * the given lower and upper address bounds.
728 vm_map_create(pmap_t pmap, vm_offset_t min, vm_offset_t max)
732 result = uma_zalloc(mapzone, M_WAITOK);
733 CTR1(KTR_VM, "vm_map_create: %p", result);
734 _vm_map_init(result, pmap, min, max);
739 * Initialize an existing vm_map structure
740 * such as that in the vmspace structure.
743 _vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max)
746 map->header.next = map->header.prev = &map->header;
747 map->needs_wakeup = FALSE;
750 map->min_offset = min;
751 map->max_offset = max;
759 vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max)
762 _vm_map_init(map, pmap, min, max);
763 mtx_init(&map->system_mtx, "system map", NULL, MTX_DEF | MTX_DUPOK);
764 sx_init(&map->lock, "user map");
768 * vm_map_entry_dispose: [ internal use only ]
770 * Inverse of vm_map_entry_create.
773 vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry)
775 uma_zfree(map->system_map ? kmapentzone : mapentzone, entry);
779 * vm_map_entry_create: [ internal use only ]
781 * Allocates a VM map entry for insertion.
782 * No entry fields are filled in.
784 static vm_map_entry_t
785 vm_map_entry_create(vm_map_t map)
787 vm_map_entry_t new_entry;
790 new_entry = uma_zalloc(kmapentzone, M_NOWAIT);
792 new_entry = uma_zalloc(mapentzone, M_WAITOK);
793 if (new_entry == NULL)
794 panic("vm_map_entry_create: kernel resources exhausted");
799 * vm_map_entry_set_behavior:
801 * Set the expected access behavior, either normal, random, or
805 vm_map_entry_set_behavior(vm_map_entry_t entry, u_char behavior)
807 entry->eflags = (entry->eflags & ~MAP_ENTRY_BEHAV_MASK) |
808 (behavior & MAP_ENTRY_BEHAV_MASK);
812 * vm_map_entry_set_max_free:
814 * Set the max_free field in a vm_map_entry.
817 vm_map_entry_set_max_free(vm_map_entry_t entry)
820 entry->max_free = entry->adj_free;
821 if (entry->left != NULL && entry->left->max_free > entry->max_free)
822 entry->max_free = entry->left->max_free;
823 if (entry->right != NULL && entry->right->max_free > entry->max_free)
824 entry->max_free = entry->right->max_free;
828 * vm_map_entry_splay:
830 * The Sleator and Tarjan top-down splay algorithm with the
831 * following variation. Max_free must be computed bottom-up, so
832 * on the downward pass, maintain the left and right spines in
833 * reverse order. Then, make a second pass up each side to fix
834 * the pointers and compute max_free. The time bound is O(log n)
837 * The new root is the vm_map_entry containing "addr", or else an
838 * adjacent entry (lower or higher) if addr is not in the tree.
840 * The map must be locked, and leaves it so.
842 * Returns: the new root.
844 static vm_map_entry_t
845 vm_map_entry_splay(vm_offset_t addr, vm_map_entry_t root)
847 vm_map_entry_t llist, rlist;
848 vm_map_entry_t ltree, rtree;
851 /* Special case of empty tree. */
856 * Pass One: Splay down the tree until we find addr or a NULL
857 * pointer where addr would go. llist and rlist are the two
858 * sides in reverse order (bottom-up), with llist linked by
859 * the right pointer and rlist linked by the left pointer in
860 * the vm_map_entry. Wait until Pass Two to set max_free on
866 /* root is never NULL in here. */
867 if (addr < root->start) {
871 if (addr < y->start && y->left != NULL) {
872 /* Rotate right and put y on rlist. */
873 root->left = y->right;
875 vm_map_entry_set_max_free(root);
880 /* Put root on rlist. */
885 } else if (addr >= root->end) {
889 if (addr >= y->end && y->right != NULL) {
890 /* Rotate left and put y on llist. */
891 root->right = y->left;
893 vm_map_entry_set_max_free(root);
898 /* Put root on llist. */
908 * Pass Two: Walk back up the two spines, flip the pointers
909 * and set max_free. The subtrees of the root go at the
910 * bottom of llist and rlist.
913 while (llist != NULL) {
915 llist->right = ltree;
916 vm_map_entry_set_max_free(llist);
921 while (rlist != NULL) {
924 vm_map_entry_set_max_free(rlist);
930 * Final assembly: add ltree and rtree as subtrees of root.
934 vm_map_entry_set_max_free(root);
940 * vm_map_entry_{un,}link:
942 * Insert/remove entries from maps.
945 vm_map_entry_link(vm_map_t map,
946 vm_map_entry_t after_where,
947 vm_map_entry_t entry)
951 "vm_map_entry_link: map %p, nentries %d, entry %p, after %p", map,
952 map->nentries, entry, after_where);
953 VM_MAP_ASSERT_LOCKED(map);
955 entry->prev = after_where;
956 entry->next = after_where->next;
957 entry->next->prev = entry;
958 after_where->next = entry;
960 if (after_where != &map->header) {
961 if (after_where != map->root)
962 vm_map_entry_splay(after_where->start, map->root);
963 entry->right = after_where->right;
964 entry->left = after_where;
965 after_where->right = NULL;
966 after_where->adj_free = entry->start - after_where->end;
967 vm_map_entry_set_max_free(after_where);
969 entry->right = map->root;
972 entry->adj_free = (entry->next == &map->header ? map->max_offset :
973 entry->next->start) - entry->end;
974 vm_map_entry_set_max_free(entry);
979 vm_map_entry_unlink(vm_map_t map,
980 vm_map_entry_t entry)
982 vm_map_entry_t next, prev, root;
984 VM_MAP_ASSERT_LOCKED(map);
985 if (entry != map->root)
986 vm_map_entry_splay(entry->start, map->root);
987 if (entry->left == NULL)
990 root = vm_map_entry_splay(entry->start, entry->left);
991 root->right = entry->right;
992 root->adj_free = (entry->next == &map->header ? map->max_offset :
993 entry->next->start) - root->end;
994 vm_map_entry_set_max_free(root);
1003 CTR3(KTR_VM, "vm_map_entry_unlink: map %p, nentries %d, entry %p", map,
1004 map->nentries, entry);
1008 * vm_map_entry_resize_free:
1010 * Recompute the amount of free space following a vm_map_entry
1011 * and propagate that value up the tree. Call this function after
1012 * resizing a map entry in-place, that is, without a call to
1013 * vm_map_entry_link() or _unlink().
1015 * The map must be locked, and leaves it so.
1018 vm_map_entry_resize_free(vm_map_t map, vm_map_entry_t entry)
1022 * Using splay trees without parent pointers, propagating
1023 * max_free up the tree is done by moving the entry to the
1024 * root and making the change there.
1026 if (entry != map->root)
1027 map->root = vm_map_entry_splay(entry->start, map->root);
1029 entry->adj_free = (entry->next == &map->header ? map->max_offset :
1030 entry->next->start) - entry->end;
1031 vm_map_entry_set_max_free(entry);
1035 * vm_map_lookup_entry: [ internal use only ]
1037 * Finds the map entry containing (or
1038 * immediately preceding) the specified address
1039 * in the given map; the entry is returned
1040 * in the "entry" parameter. The boolean
1041 * result indicates whether the address is
1042 * actually contained in the map.
1045 vm_map_lookup_entry(
1047 vm_offset_t address,
1048 vm_map_entry_t *entry) /* OUT */
1054 * If the map is empty, then the map entry immediately preceding
1055 * "address" is the map's header.
1059 *entry = &map->header;
1060 else if (address >= cur->start && cur->end > address) {
1063 } else if ((locked = vm_map_locked(map)) ||
1064 sx_try_upgrade(&map->lock)) {
1066 * Splay requires a write lock on the map. However, it only
1067 * restructures the binary search tree; it does not otherwise
1068 * change the map. Thus, the map's timestamp need not change
1069 * on a temporary upgrade.
1071 map->root = cur = vm_map_entry_splay(address, cur);
1073 sx_downgrade(&map->lock);
1076 * If "address" is contained within a map entry, the new root
1077 * is that map entry. Otherwise, the new root is a map entry
1078 * immediately before or after "address".
1080 if (address >= cur->start) {
1082 if (cur->end > address)
1088 * Since the map is only locked for read access, perform a
1089 * standard binary search tree lookup for "address".
1092 if (address < cur->start) {
1093 if (cur->left == NULL) {
1098 } else if (cur->end > address) {
1102 if (cur->right == NULL) {
1115 * Inserts the given whole VM object into the target
1116 * map at the specified address range. The object's
1117 * size should match that of the address range.
1119 * Requires that the map be locked, and leaves it so.
1121 * If object is non-NULL, ref count must be bumped by caller
1122 * prior to making call to account for the new entry.
1125 vm_map_insert(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1126 vm_offset_t start, vm_offset_t end, vm_prot_t prot, vm_prot_t max,
1129 vm_map_entry_t new_entry;
1130 vm_map_entry_t prev_entry;
1131 vm_map_entry_t temp_entry;
1132 vm_eflags_t protoeflags;
1134 vm_inherit_t inheritance;
1135 boolean_t charge_prev_obj;
1137 VM_MAP_ASSERT_LOCKED(map);
1140 * Check that the start and end points are not bogus.
1142 if ((start < map->min_offset) || (end > map->max_offset) ||
1144 return (KERN_INVALID_ADDRESS);
1147 * Find the entry prior to the proposed starting address; if it's part
1148 * of an existing entry, this range is bogus.
1150 if (vm_map_lookup_entry(map, start, &temp_entry))
1151 return (KERN_NO_SPACE);
1153 prev_entry = temp_entry;
1156 * Assert that the next entry doesn't overlap the end point.
1158 if ((prev_entry->next != &map->header) &&
1159 (prev_entry->next->start < end))
1160 return (KERN_NO_SPACE);
1163 charge_prev_obj = FALSE;
1165 if (cow & MAP_COPY_ON_WRITE)
1166 protoeflags |= MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY;
1168 if (cow & MAP_NOFAULT) {
1169 protoeflags |= MAP_ENTRY_NOFAULT;
1171 KASSERT(object == NULL,
1172 ("vm_map_insert: paradoxical MAP_NOFAULT request"));
1174 if (cow & MAP_DISABLE_SYNCER)
1175 protoeflags |= MAP_ENTRY_NOSYNC;
1176 if (cow & MAP_DISABLE_COREDUMP)
1177 protoeflags |= MAP_ENTRY_NOCOREDUMP;
1178 if (cow & MAP_VN_WRITECOUNT)
1179 protoeflags |= MAP_ENTRY_VN_WRITECNT;
1180 if (cow & MAP_INHERIT_SHARE)
1181 inheritance = VM_INHERIT_SHARE;
1183 inheritance = VM_INHERIT_DEFAULT;
1186 KASSERT((object != kmem_object && object != kernel_object) ||
1187 ((object == kmem_object || object == kernel_object) &&
1188 !(protoeflags & MAP_ENTRY_NEEDS_COPY)),
1189 ("kmem or kernel object and cow"));
1190 if (cow & (MAP_ACC_NO_CHARGE | MAP_NOFAULT))
1192 if ((cow & MAP_ACC_CHARGED) || ((prot & VM_PROT_WRITE) &&
1193 ((protoeflags & MAP_ENTRY_NEEDS_COPY) || object == NULL))) {
1194 if (!(cow & MAP_ACC_CHARGED) && !swap_reserve(end - start))
1195 return (KERN_RESOURCE_SHORTAGE);
1196 KASSERT(object == NULL || (protoeflags & MAP_ENTRY_NEEDS_COPY) ||
1197 object->cred == NULL,
1198 ("OVERCOMMIT: vm_map_insert o %p", object));
1199 cred = curthread->td_ucred;
1201 if (object == NULL && !(protoeflags & MAP_ENTRY_NEEDS_COPY))
1202 charge_prev_obj = TRUE;
1206 /* Expand the kernel pmap, if necessary. */
1207 if (map == kernel_map && end > kernel_vm_end)
1208 pmap_growkernel(end);
1209 if (object != NULL) {
1211 * OBJ_ONEMAPPING must be cleared unless this mapping
1212 * is trivially proven to be the only mapping for any
1213 * of the object's pages. (Object granularity
1214 * reference counting is insufficient to recognize
1215 * aliases with precision.)
1217 VM_OBJECT_WLOCK(object);
1218 if (object->ref_count > 1 || object->shadow_count != 0)
1219 vm_object_clear_flag(object, OBJ_ONEMAPPING);
1220 VM_OBJECT_WUNLOCK(object);
1222 else if ((prev_entry != &map->header) &&
1223 (prev_entry->eflags == protoeflags) &&
1224 (prev_entry->end == start) &&
1225 (prev_entry->wired_count == 0) &&
1226 (prev_entry->cred == cred ||
1227 (prev_entry->object.vm_object != NULL &&
1228 (prev_entry->object.vm_object->cred == cred))) &&
1229 vm_object_coalesce(prev_entry->object.vm_object,
1231 (vm_size_t)(prev_entry->end - prev_entry->start),
1232 (vm_size_t)(end - prev_entry->end), charge_prev_obj)) {
1234 * We were able to extend the object. Determine if we
1235 * can extend the previous map entry to include the
1236 * new range as well.
1238 if ((prev_entry->inheritance == inheritance) &&
1239 (prev_entry->protection == prot) &&
1240 (prev_entry->max_protection == max)) {
1241 map->size += (end - prev_entry->end);
1242 prev_entry->end = end;
1243 vm_map_entry_resize_free(map, prev_entry);
1244 vm_map_simplify_entry(map, prev_entry);
1247 return (KERN_SUCCESS);
1251 * If we can extend the object but cannot extend the
1252 * map entry, we have to create a new map entry. We
1253 * must bump the ref count on the extended object to
1254 * account for it. object may be NULL.
1256 object = prev_entry->object.vm_object;
1257 offset = prev_entry->offset +
1258 (prev_entry->end - prev_entry->start);
1259 vm_object_reference(object);
1260 if (cred != NULL && object != NULL && object->cred != NULL &&
1261 !(prev_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
1262 /* Object already accounts for this uid. */
1269 * NOTE: if conditionals fail, object can be NULL here. This occurs
1270 * in things like the buffer map where we manage kva but do not manage
1275 * Create a new entry
1277 new_entry = vm_map_entry_create(map);
1278 new_entry->start = start;
1279 new_entry->end = end;
1280 new_entry->cred = NULL;
1282 new_entry->eflags = protoeflags;
1283 new_entry->object.vm_object = object;
1284 new_entry->offset = offset;
1285 new_entry->avail_ssize = 0;
1287 new_entry->inheritance = inheritance;
1288 new_entry->protection = prot;
1289 new_entry->max_protection = max;
1290 new_entry->wired_count = 0;
1291 new_entry->read_ahead = VM_FAULT_READ_AHEAD_INIT;
1292 new_entry->next_read = OFF_TO_IDX(offset);
1294 KASSERT(cred == NULL || !ENTRY_CHARGED(new_entry),
1295 ("OVERCOMMIT: vm_map_insert leaks vm_map %p", new_entry));
1296 new_entry->cred = cred;
1299 * Insert the new entry into the list
1301 vm_map_entry_link(map, prev_entry, new_entry);
1302 map->size += new_entry->end - new_entry->start;
1305 * It may be possible to merge the new entry with the next and/or
1306 * previous entries. However, due to MAP_STACK_* being a hack, a
1307 * panic can result from merging such entries.
1309 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0)
1310 vm_map_simplify_entry(map, new_entry);
1312 if (cow & (MAP_PREFAULT|MAP_PREFAULT_PARTIAL)) {
1313 vm_map_pmap_enter(map, start, prot,
1314 object, OFF_TO_IDX(offset), end - start,
1315 cow & MAP_PREFAULT_PARTIAL);
1318 return (KERN_SUCCESS);
1324 * Find the first fit (lowest VM address) for "length" free bytes
1325 * beginning at address >= start in the given map.
1327 * In a vm_map_entry, "adj_free" is the amount of free space
1328 * adjacent (higher address) to this entry, and "max_free" is the
1329 * maximum amount of contiguous free space in its subtree. This
1330 * allows finding a free region in one path down the tree, so
1331 * O(log n) amortized with splay trees.
1333 * The map must be locked, and leaves it so.
1335 * Returns: 0 on success, and starting address in *addr,
1336 * 1 if insufficient space.
1339 vm_map_findspace(vm_map_t map, vm_offset_t start, vm_size_t length,
1340 vm_offset_t *addr) /* OUT */
1342 vm_map_entry_t entry;
1346 * Request must fit within min/max VM address and must avoid
1349 if (start < map->min_offset)
1350 start = map->min_offset;
1351 if (start + length > map->max_offset || start + length < start)
1354 /* Empty tree means wide open address space. */
1355 if (map->root == NULL) {
1361 * After splay, if start comes before root node, then there
1362 * must be a gap from start to the root.
1364 map->root = vm_map_entry_splay(start, map->root);
1365 if (start + length <= map->root->start) {
1371 * Root is the last node that might begin its gap before
1372 * start, and this is the last comparison where address
1373 * wrap might be a problem.
1375 st = (start > map->root->end) ? start : map->root->end;
1376 if (length <= map->root->end + map->root->adj_free - st) {
1381 /* With max_free, can immediately tell if no solution. */
1382 entry = map->root->right;
1383 if (entry == NULL || length > entry->max_free)
1387 * Search the right subtree in the order: left subtree, root,
1388 * right subtree (first fit). The previous splay implies that
1389 * all regions in the right subtree have addresses > start.
1391 while (entry != NULL) {
1392 if (entry->left != NULL && entry->left->max_free >= length)
1393 entry = entry->left;
1394 else if (entry->adj_free >= length) {
1398 entry = entry->right;
1401 /* Can't get here, so panic if we do. */
1402 panic("vm_map_findspace: max_free corrupt");
1406 vm_map_fixed(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1407 vm_offset_t start, vm_size_t length, vm_prot_t prot,
1408 vm_prot_t max, int cow)
1413 end = start + length;
1415 VM_MAP_RANGE_CHECK(map, start, end);
1416 (void) vm_map_delete(map, start, end);
1417 result = vm_map_insert(map, object, offset, start, end, prot,
1424 * vm_map_find finds an unallocated region in the target address
1425 * map with the given length. The search is defined to be
1426 * first-fit from the specified address; the region found is
1427 * returned in the same parameter.
1429 * If object is non-NULL, ref count must be bumped by caller
1430 * prior to making call to account for the new entry.
1433 vm_map_find(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1434 vm_offset_t *addr, /* IN/OUT */
1435 vm_size_t length, int find_space, vm_prot_t prot,
1436 vm_prot_t max, int cow)
1438 vm_offset_t alignment, initial_addr, start;
1441 if (find_space == VMFS_OPTIMAL_SPACE && (object == NULL ||
1442 (object->flags & OBJ_COLORED) == 0))
1443 find_space = VMFS_ANY_SPACE;
1444 if (find_space >> 8 != 0) {
1445 KASSERT((find_space & 0xff) == 0, ("bad VMFS flags"));
1446 alignment = (vm_offset_t)1 << (find_space >> 8);
1449 initial_addr = *addr;
1451 start = initial_addr;
1454 if (find_space != VMFS_NO_SPACE) {
1455 if (vm_map_findspace(map, start, length, addr)) {
1457 if (find_space == VMFS_OPTIMAL_SPACE) {
1458 find_space = VMFS_ANY_SPACE;
1461 return (KERN_NO_SPACE);
1463 switch (find_space) {
1464 case VMFS_SUPER_SPACE:
1465 case VMFS_OPTIMAL_SPACE:
1466 pmap_align_superpage(object, offset, addr,
1469 case VMFS_ANY_SPACE:
1472 if ((*addr & (alignment - 1)) != 0) {
1473 *addr &= ~(alignment - 1);
1481 result = vm_map_insert(map, object, offset, start, start +
1482 length, prot, max, cow);
1483 } while (result == KERN_NO_SPACE && find_space != VMFS_NO_SPACE &&
1484 find_space != VMFS_ANY_SPACE);
1490 * vm_map_simplify_entry:
1492 * Simplify the given map entry by merging with either neighbor. This
1493 * routine also has the ability to merge with both neighbors.
1495 * The map must be locked.
1497 * This routine guarentees that the passed entry remains valid (though
1498 * possibly extended). When merging, this routine may delete one or
1502 vm_map_simplify_entry(vm_map_t map, vm_map_entry_t entry)
1504 vm_map_entry_t next, prev;
1505 vm_size_t prevsize, esize;
1507 if (entry->eflags & (MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_IS_SUB_MAP))
1511 if (prev != &map->header) {
1512 prevsize = prev->end - prev->start;
1513 if ( (prev->end == entry->start) &&
1514 (prev->object.vm_object == entry->object.vm_object) &&
1515 (!prev->object.vm_object ||
1516 (prev->offset + prevsize == entry->offset)) &&
1517 (prev->eflags == entry->eflags) &&
1518 (prev->protection == entry->protection) &&
1519 (prev->max_protection == entry->max_protection) &&
1520 (prev->inheritance == entry->inheritance) &&
1521 (prev->wired_count == entry->wired_count) &&
1522 (prev->cred == entry->cred)) {
1523 vm_map_entry_unlink(map, prev);
1524 entry->start = prev->start;
1525 entry->offset = prev->offset;
1526 if (entry->prev != &map->header)
1527 vm_map_entry_resize_free(map, entry->prev);
1530 * If the backing object is a vnode object,
1531 * vm_object_deallocate() calls vrele().
1532 * However, vrele() does not lock the vnode
1533 * because the vnode has additional
1534 * references. Thus, the map lock can be kept
1535 * without causing a lock-order reversal with
1538 * Since we count the number of virtual page
1539 * mappings in object->un_pager.vnp.writemappings,
1540 * the writemappings value should not be adjusted
1541 * when the entry is disposed of.
1543 if (prev->object.vm_object)
1544 vm_object_deallocate(prev->object.vm_object);
1545 if (prev->cred != NULL)
1547 vm_map_entry_dispose(map, prev);
1552 if (next != &map->header) {
1553 esize = entry->end - entry->start;
1554 if ((entry->end == next->start) &&
1555 (next->object.vm_object == entry->object.vm_object) &&
1556 (!entry->object.vm_object ||
1557 (entry->offset + esize == next->offset)) &&
1558 (next->eflags == entry->eflags) &&
1559 (next->protection == entry->protection) &&
1560 (next->max_protection == entry->max_protection) &&
1561 (next->inheritance == entry->inheritance) &&
1562 (next->wired_count == entry->wired_count) &&
1563 (next->cred == entry->cred)) {
1564 vm_map_entry_unlink(map, next);
1565 entry->end = next->end;
1566 vm_map_entry_resize_free(map, entry);
1569 * See comment above.
1571 if (next->object.vm_object)
1572 vm_object_deallocate(next->object.vm_object);
1573 if (next->cred != NULL)
1575 vm_map_entry_dispose(map, next);
1580 * vm_map_clip_start: [ internal use only ]
1582 * Asserts that the given entry begins at or after
1583 * the specified address; if necessary,
1584 * it splits the entry into two.
1586 #define vm_map_clip_start(map, entry, startaddr) \
1588 if (startaddr > entry->start) \
1589 _vm_map_clip_start(map, entry, startaddr); \
1593 * This routine is called only when it is known that
1594 * the entry must be split.
1597 _vm_map_clip_start(vm_map_t map, vm_map_entry_t entry, vm_offset_t start)
1599 vm_map_entry_t new_entry;
1601 VM_MAP_ASSERT_LOCKED(map);
1604 * Split off the front portion -- note that we must insert the new
1605 * entry BEFORE this one, so that this entry has the specified
1608 vm_map_simplify_entry(map, entry);
1611 * If there is no object backing this entry, we might as well create
1612 * one now. If we defer it, an object can get created after the map
1613 * is clipped, and individual objects will be created for the split-up
1614 * map. This is a bit of a hack, but is also about the best place to
1615 * put this improvement.
1617 if (entry->object.vm_object == NULL && !map->system_map) {
1619 object = vm_object_allocate(OBJT_DEFAULT,
1620 atop(entry->end - entry->start));
1621 entry->object.vm_object = object;
1623 if (entry->cred != NULL) {
1624 object->cred = entry->cred;
1625 object->charge = entry->end - entry->start;
1628 } else if (entry->object.vm_object != NULL &&
1629 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) &&
1630 entry->cred != NULL) {
1631 VM_OBJECT_WLOCK(entry->object.vm_object);
1632 KASSERT(entry->object.vm_object->cred == NULL,
1633 ("OVERCOMMIT: vm_entry_clip_start: both cred e %p", entry));
1634 entry->object.vm_object->cred = entry->cred;
1635 entry->object.vm_object->charge = entry->end - entry->start;
1636 VM_OBJECT_WUNLOCK(entry->object.vm_object);
1640 new_entry = vm_map_entry_create(map);
1641 *new_entry = *entry;
1643 new_entry->end = start;
1644 entry->offset += (start - entry->start);
1645 entry->start = start;
1646 if (new_entry->cred != NULL)
1647 crhold(entry->cred);
1649 vm_map_entry_link(map, entry->prev, new_entry);
1651 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1652 vm_object_reference(new_entry->object.vm_object);
1654 * The object->un_pager.vnp.writemappings for the
1655 * object of MAP_ENTRY_VN_WRITECNT type entry shall be
1656 * kept as is here. The virtual pages are
1657 * re-distributed among the clipped entries, so the sum is
1664 * vm_map_clip_end: [ internal use only ]
1666 * Asserts that the given entry ends at or before
1667 * the specified address; if necessary,
1668 * it splits the entry into two.
1670 #define vm_map_clip_end(map, entry, endaddr) \
1672 if ((endaddr) < (entry->end)) \
1673 _vm_map_clip_end((map), (entry), (endaddr)); \
1677 * This routine is called only when it is known that
1678 * the entry must be split.
1681 _vm_map_clip_end(vm_map_t map, vm_map_entry_t entry, vm_offset_t end)
1683 vm_map_entry_t new_entry;
1685 VM_MAP_ASSERT_LOCKED(map);
1688 * If there is no object backing this entry, we might as well create
1689 * one now. If we defer it, an object can get created after the map
1690 * is clipped, and individual objects will be created for the split-up
1691 * map. This is a bit of a hack, but is also about the best place to
1692 * put this improvement.
1694 if (entry->object.vm_object == NULL && !map->system_map) {
1696 object = vm_object_allocate(OBJT_DEFAULT,
1697 atop(entry->end - entry->start));
1698 entry->object.vm_object = object;
1700 if (entry->cred != NULL) {
1701 object->cred = entry->cred;
1702 object->charge = entry->end - entry->start;
1705 } else if (entry->object.vm_object != NULL &&
1706 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) &&
1707 entry->cred != NULL) {
1708 VM_OBJECT_WLOCK(entry->object.vm_object);
1709 KASSERT(entry->object.vm_object->cred == NULL,
1710 ("OVERCOMMIT: vm_entry_clip_end: both cred e %p", entry));
1711 entry->object.vm_object->cred = entry->cred;
1712 entry->object.vm_object->charge = entry->end - entry->start;
1713 VM_OBJECT_WUNLOCK(entry->object.vm_object);
1718 * Create a new entry and insert it AFTER the specified entry
1720 new_entry = vm_map_entry_create(map);
1721 *new_entry = *entry;
1723 new_entry->start = entry->end = end;
1724 new_entry->offset += (end - entry->start);
1725 if (new_entry->cred != NULL)
1726 crhold(entry->cred);
1728 vm_map_entry_link(map, entry, new_entry);
1730 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1731 vm_object_reference(new_entry->object.vm_object);
1736 * vm_map_submap: [ kernel use only ]
1738 * Mark the given range as handled by a subordinate map.
1740 * This range must have been created with vm_map_find,
1741 * and no other operations may have been performed on this
1742 * range prior to calling vm_map_submap.
1744 * Only a limited number of operations can be performed
1745 * within this rage after calling vm_map_submap:
1747 * [Don't try vm_map_copy!]
1749 * To remove a submapping, one must first remove the
1750 * range from the superior map, and then destroy the
1751 * submap (if desired). [Better yet, don't try it.]
1760 vm_map_entry_t entry;
1761 int result = KERN_INVALID_ARGUMENT;
1765 VM_MAP_RANGE_CHECK(map, start, end);
1767 if (vm_map_lookup_entry(map, start, &entry)) {
1768 vm_map_clip_start(map, entry, start);
1770 entry = entry->next;
1772 vm_map_clip_end(map, entry, end);
1774 if ((entry->start == start) && (entry->end == end) &&
1775 ((entry->eflags & MAP_ENTRY_COW) == 0) &&
1776 (entry->object.vm_object == NULL)) {
1777 entry->object.sub_map = submap;
1778 entry->eflags |= MAP_ENTRY_IS_SUB_MAP;
1779 result = KERN_SUCCESS;
1787 * The maximum number of pages to map
1789 #define MAX_INIT_PT 96
1792 * vm_map_pmap_enter:
1794 * Preload read-only mappings for the specified object's resident pages
1795 * into the target map. If "flags" is MAP_PREFAULT_PARTIAL, then only
1796 * the resident pages within the address range [addr, addr + ulmin(size,
1797 * ptoa(MAX_INIT_PT))) are mapped. Otherwise, all resident pages within
1798 * the specified address range are mapped. This eliminates many soft
1799 * faults on process startup and immediately after an mmap(2). Because
1800 * these are speculative mappings, cached pages are not reactivated and
1804 vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot,
1805 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags)
1808 vm_page_t p, p_start;
1809 vm_pindex_t psize, tmpidx;
1811 if ((prot & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0 || object == NULL)
1813 VM_OBJECT_RLOCK(object);
1814 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) {
1815 VM_OBJECT_RUNLOCK(object);
1816 VM_OBJECT_WLOCK(object);
1817 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) {
1818 pmap_object_init_pt(map->pmap, addr, object, pindex,
1820 VM_OBJECT_WUNLOCK(object);
1823 VM_OBJECT_LOCK_DOWNGRADE(object);
1827 if (psize > MAX_INIT_PT && (flags & MAP_PREFAULT_PARTIAL) != 0)
1828 psize = MAX_INIT_PT;
1829 if (psize + pindex > object->size) {
1830 if (object->size < pindex) {
1831 VM_OBJECT_RUNLOCK(object);
1834 psize = object->size - pindex;
1840 p = vm_page_find_least(object, pindex);
1842 * Assert: the variable p is either (1) the page with the
1843 * least pindex greater than or equal to the parameter pindex
1847 p != NULL && (tmpidx = p->pindex - pindex) < psize;
1848 p = TAILQ_NEXT(p, listq)) {
1850 * don't allow an madvise to blow away our really
1851 * free pages allocating pv entries.
1853 if ((flags & MAP_PREFAULT_MADVISE) &&
1854 cnt.v_free_count < cnt.v_free_reserved) {
1858 if (p->valid == VM_PAGE_BITS_ALL) {
1859 if (p_start == NULL) {
1860 start = addr + ptoa(tmpidx);
1863 } else if (p_start != NULL) {
1864 pmap_enter_object(map->pmap, start, addr +
1865 ptoa(tmpidx), p_start, prot);
1869 if (p_start != NULL)
1870 pmap_enter_object(map->pmap, start, addr + ptoa(psize),
1872 VM_OBJECT_RUNLOCK(object);
1878 * Sets the protection of the specified address
1879 * region in the target map. If "set_max" is
1880 * specified, the maximum protection is to be set;
1881 * otherwise, only the current protection is affected.
1884 vm_map_protect(vm_map_t map, vm_offset_t start, vm_offset_t end,
1885 vm_prot_t new_prot, boolean_t set_max)
1887 vm_map_entry_t current, entry;
1894 VM_MAP_RANGE_CHECK(map, start, end);
1896 if (vm_map_lookup_entry(map, start, &entry)) {
1897 vm_map_clip_start(map, entry, start);
1899 entry = entry->next;
1903 * Make a first pass to check for protection violations.
1906 while ((current != &map->header) && (current->start < end)) {
1907 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
1909 return (KERN_INVALID_ARGUMENT);
1911 if ((new_prot & current->max_protection) != new_prot) {
1913 return (KERN_PROTECTION_FAILURE);
1915 current = current->next;
1920 * Do an accounting pass for private read-only mappings that
1921 * now will do cow due to allowed write (e.g. debugger sets
1922 * breakpoint on text segment)
1924 for (current = entry; (current != &map->header) &&
1925 (current->start < end); current = current->next) {
1927 vm_map_clip_end(map, current, end);
1930 ((new_prot & ~(current->protection)) & VM_PROT_WRITE) == 0 ||
1931 ENTRY_CHARGED(current)) {
1935 cred = curthread->td_ucred;
1936 obj = current->object.vm_object;
1938 if (obj == NULL || (current->eflags & MAP_ENTRY_NEEDS_COPY)) {
1939 if (!swap_reserve(current->end - current->start)) {
1941 return (KERN_RESOURCE_SHORTAGE);
1944 current->cred = cred;
1948 VM_OBJECT_WLOCK(obj);
1949 if (obj->type != OBJT_DEFAULT && obj->type != OBJT_SWAP) {
1950 VM_OBJECT_WUNLOCK(obj);
1955 * Charge for the whole object allocation now, since
1956 * we cannot distinguish between non-charged and
1957 * charged clipped mapping of the same object later.
1959 KASSERT(obj->charge == 0,
1960 ("vm_map_protect: object %p overcharged\n", obj));
1961 if (!swap_reserve(ptoa(obj->size))) {
1962 VM_OBJECT_WUNLOCK(obj);
1964 return (KERN_RESOURCE_SHORTAGE);
1969 obj->charge = ptoa(obj->size);
1970 VM_OBJECT_WUNLOCK(obj);
1974 * Go back and fix up protections. [Note that clipping is not
1975 * necessary the second time.]
1978 while ((current != &map->header) && (current->start < end)) {
1979 old_prot = current->protection;
1982 current->protection =
1983 (current->max_protection = new_prot) &
1986 current->protection = new_prot;
1988 if ((current->eflags & (MAP_ENTRY_COW | MAP_ENTRY_USER_WIRED))
1989 == (MAP_ENTRY_COW | MAP_ENTRY_USER_WIRED) &&
1990 (current->protection & VM_PROT_WRITE) != 0 &&
1991 (old_prot & VM_PROT_WRITE) == 0) {
1992 vm_fault_copy_entry(map, map, current, current, NULL);
1996 * When restricting access, update the physical map. Worry
1997 * about copy-on-write here.
1999 if ((old_prot & ~current->protection) != 0) {
2000 #define MASK(entry) (((entry)->eflags & MAP_ENTRY_COW) ? ~VM_PROT_WRITE : \
2002 pmap_protect(map->pmap, current->start,
2004 current->protection & MASK(current));
2007 vm_map_simplify_entry(map, current);
2008 current = current->next;
2011 return (KERN_SUCCESS);
2017 * This routine traverses a processes map handling the madvise
2018 * system call. Advisories are classified as either those effecting
2019 * the vm_map_entry structure, or those effecting the underlying
2029 vm_map_entry_t current, entry;
2033 * Some madvise calls directly modify the vm_map_entry, in which case
2034 * we need to use an exclusive lock on the map and we need to perform
2035 * various clipping operations. Otherwise we only need a read-lock
2040 case MADV_SEQUENTIAL:
2052 vm_map_lock_read(map);
2055 return (KERN_INVALID_ARGUMENT);
2059 * Locate starting entry and clip if necessary.
2061 VM_MAP_RANGE_CHECK(map, start, end);
2063 if (vm_map_lookup_entry(map, start, &entry)) {
2065 vm_map_clip_start(map, entry, start);
2067 entry = entry->next;
2072 * madvise behaviors that are implemented in the vm_map_entry.
2074 * We clip the vm_map_entry so that behavioral changes are
2075 * limited to the specified address range.
2077 for (current = entry;
2078 (current != &map->header) && (current->start < end);
2079 current = current->next
2081 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
2084 vm_map_clip_end(map, current, end);
2088 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_NORMAL);
2090 case MADV_SEQUENTIAL:
2091 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_SEQUENTIAL);
2094 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_RANDOM);
2097 current->eflags |= MAP_ENTRY_NOSYNC;
2100 current->eflags &= ~MAP_ENTRY_NOSYNC;
2103 current->eflags |= MAP_ENTRY_NOCOREDUMP;
2106 current->eflags &= ~MAP_ENTRY_NOCOREDUMP;
2111 vm_map_simplify_entry(map, current);
2115 vm_pindex_t pstart, pend;
2118 * madvise behaviors that are implemented in the underlying
2121 * Since we don't clip the vm_map_entry, we have to clip
2122 * the vm_object pindex and count.
2124 for (current = entry;
2125 (current != &map->header) && (current->start < end);
2126 current = current->next
2128 vm_offset_t useEnd, useStart;
2130 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
2133 pstart = OFF_TO_IDX(current->offset);
2134 pend = pstart + atop(current->end - current->start);
2135 useStart = current->start;
2136 useEnd = current->end;
2138 if (current->start < start) {
2139 pstart += atop(start - current->start);
2142 if (current->end > end) {
2143 pend -= atop(current->end - end);
2151 * Perform the pmap_advise() before clearing
2152 * PGA_REFERENCED in vm_page_advise(). Otherwise, a
2153 * concurrent pmap operation, such as pmap_remove(),
2154 * could clear a reference in the pmap and set
2155 * PGA_REFERENCED on the page before the pmap_advise()
2156 * had completed. Consequently, the page would appear
2157 * referenced based upon an old reference that
2158 * occurred before this pmap_advise() ran.
2160 if (behav == MADV_DONTNEED || behav == MADV_FREE)
2161 pmap_advise(map->pmap, useStart, useEnd,
2164 vm_object_madvise(current->object.vm_object, pstart,
2166 if (behav == MADV_WILLNEED) {
2167 vm_map_pmap_enter(map,
2169 current->protection,
2170 current->object.vm_object,
2172 ptoa(pend - pstart),
2173 MAP_PREFAULT_MADVISE
2177 vm_map_unlock_read(map);
2186 * Sets the inheritance of the specified address
2187 * range in the target map. Inheritance
2188 * affects how the map will be shared with
2189 * child maps at the time of vmspace_fork.
2192 vm_map_inherit(vm_map_t map, vm_offset_t start, vm_offset_t end,
2193 vm_inherit_t new_inheritance)
2195 vm_map_entry_t entry;
2196 vm_map_entry_t temp_entry;
2198 switch (new_inheritance) {
2199 case VM_INHERIT_NONE:
2200 case VM_INHERIT_COPY:
2201 case VM_INHERIT_SHARE:
2204 return (KERN_INVALID_ARGUMENT);
2207 VM_MAP_RANGE_CHECK(map, start, end);
2208 if (vm_map_lookup_entry(map, start, &temp_entry)) {
2210 vm_map_clip_start(map, entry, start);
2212 entry = temp_entry->next;
2213 while ((entry != &map->header) && (entry->start < end)) {
2214 vm_map_clip_end(map, entry, end);
2215 entry->inheritance = new_inheritance;
2216 vm_map_simplify_entry(map, entry);
2217 entry = entry->next;
2220 return (KERN_SUCCESS);
2226 * Implements both kernel and user unwiring.
2229 vm_map_unwire(vm_map_t map, vm_offset_t start, vm_offset_t end,
2232 vm_map_entry_t entry, first_entry, tmp_entry;
2233 vm_offset_t saved_start;
2234 unsigned int last_timestamp;
2236 boolean_t need_wakeup, result, user_unwire;
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 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
2305 entry->wiring_thread = curthread;
2307 * Check the map for holes in the specified region.
2308 * If VM_MAP_WIRE_HOLESOK was specified, skip this check.
2310 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) &&
2311 (entry->end < end && (entry->next == &map->header ||
2312 entry->next->start > entry->end))) {
2314 rv = KERN_INVALID_ADDRESS;
2318 * If system unwiring, require that the entry is system wired.
2321 vm_map_entry_system_wired_count(entry) == 0) {
2323 rv = KERN_INVALID_ARGUMENT;
2326 entry = entry->next;
2330 need_wakeup = FALSE;
2331 if (first_entry == NULL) {
2332 result = vm_map_lookup_entry(map, start, &first_entry);
2333 if (!result && (flags & VM_MAP_WIRE_HOLESOK))
2334 first_entry = first_entry->next;
2336 KASSERT(result, ("vm_map_unwire: lookup failed"));
2338 for (entry = first_entry; entry != &map->header && entry->start < end;
2339 entry = entry->next) {
2341 * If VM_MAP_WIRE_HOLESOK was specified, an empty
2342 * space in the unwired region could have been mapped
2343 * while the map lock was dropped for draining
2344 * MAP_ENTRY_IN_TRANSITION. Moreover, another thread
2345 * could be simultaneously wiring this new mapping
2346 * entry. Detect these cases and skip any entries
2347 * marked as in transition by us.
2349 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 ||
2350 entry->wiring_thread != curthread) {
2351 KASSERT((flags & VM_MAP_WIRE_HOLESOK) != 0,
2352 ("vm_map_unwire: !HOLESOK and new/changed entry"));
2356 if (rv == KERN_SUCCESS && (!user_unwire ||
2357 (entry->eflags & MAP_ENTRY_USER_WIRED))) {
2359 entry->eflags &= ~MAP_ENTRY_USER_WIRED;
2360 entry->wired_count--;
2361 if (entry->wired_count == 0) {
2363 * Retain the map lock.
2365 vm_fault_unwire(map, entry->start, entry->end,
2366 entry->object.vm_object != NULL &&
2367 (entry->object.vm_object->flags &
2368 OBJ_FICTITIOUS) != 0);
2371 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0,
2372 ("vm_map_unwire: in-transition flag missing"));
2373 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION;
2374 entry->wiring_thread = NULL;
2375 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
2376 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
2379 vm_map_simplify_entry(map, entry);
2390 * Implements both kernel and user wiring.
2393 vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t end,
2396 vm_map_entry_t entry, first_entry, tmp_entry;
2397 vm_offset_t saved_end, saved_start;
2398 unsigned int last_timestamp;
2400 boolean_t fictitious, need_wakeup, result, user_wire;
2404 if (flags & VM_MAP_WIRE_WRITE)
2405 prot |= VM_PROT_WRITE;
2406 user_wire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE;
2408 VM_MAP_RANGE_CHECK(map, start, end);
2409 if (!vm_map_lookup_entry(map, start, &first_entry)) {
2410 if (flags & VM_MAP_WIRE_HOLESOK)
2411 first_entry = first_entry->next;
2414 return (KERN_INVALID_ADDRESS);
2417 last_timestamp = map->timestamp;
2418 entry = first_entry;
2419 while (entry != &map->header && entry->start < end) {
2420 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
2422 * We have not yet clipped the entry.
2424 saved_start = (start >= entry->start) ? start :
2426 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2427 if (vm_map_unlock_and_wait(map, 0)) {
2429 * Allow interruption of user wiring?
2433 if (last_timestamp + 1 != map->timestamp) {
2435 * Look again for the entry because the map was
2436 * modified while it was unlocked.
2437 * Specifically, the entry may have been
2438 * clipped, merged, or deleted.
2440 if (!vm_map_lookup_entry(map, saved_start,
2442 if (flags & VM_MAP_WIRE_HOLESOK)
2443 tmp_entry = tmp_entry->next;
2445 if (saved_start == start) {
2447 * first_entry has been deleted.
2450 return (KERN_INVALID_ADDRESS);
2453 rv = KERN_INVALID_ADDRESS;
2457 if (entry == first_entry)
2458 first_entry = tmp_entry;
2463 last_timestamp = map->timestamp;
2466 vm_map_clip_start(map, entry, start);
2467 vm_map_clip_end(map, entry, end);
2469 * Mark the entry in case the map lock is released. (See
2472 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
2473 entry->wiring_thread = curthread;
2474 if ((entry->protection & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0
2475 || (entry->protection & prot) != prot) {
2476 entry->eflags |= MAP_ENTRY_WIRE_SKIPPED;
2477 if ((flags & VM_MAP_WIRE_HOLESOK) == 0) {
2479 rv = KERN_INVALID_ADDRESS;
2484 if (entry->wired_count == 0) {
2485 entry->wired_count++;
2486 saved_start = entry->start;
2487 saved_end = entry->end;
2488 fictitious = entry->object.vm_object != NULL &&
2489 (entry->object.vm_object->flags &
2490 OBJ_FICTITIOUS) != 0;
2492 * Release the map lock, relying on the in-transition
2493 * mark. Mark the map busy for fork.
2497 rv = vm_fault_wire(map, saved_start, saved_end,
2501 if (last_timestamp + 1 != map->timestamp) {
2503 * Look again for the entry because the map was
2504 * modified while it was unlocked. The entry
2505 * may have been clipped, but NOT merged or
2508 result = vm_map_lookup_entry(map, saved_start,
2510 KASSERT(result, ("vm_map_wire: lookup failed"));
2511 if (entry == first_entry)
2512 first_entry = tmp_entry;
2516 while (entry->end < saved_end) {
2517 if (rv != KERN_SUCCESS) {
2518 KASSERT(entry->wired_count == 1,
2519 ("vm_map_wire: bad count"));
2520 entry->wired_count = -1;
2522 entry = entry->next;
2525 last_timestamp = map->timestamp;
2526 if (rv != KERN_SUCCESS) {
2527 KASSERT(entry->wired_count == 1,
2528 ("vm_map_wire: bad count"));
2530 * Assign an out-of-range value to represent
2531 * the failure to wire this entry.
2533 entry->wired_count = -1;
2537 } else if (!user_wire ||
2538 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
2539 entry->wired_count++;
2542 * Check the map for holes in the specified region.
2543 * If VM_MAP_WIRE_HOLESOK was specified, skip this check.
2546 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) &&
2547 (entry->end < end && (entry->next == &map->header ||
2548 entry->next->start > entry->end))) {
2550 rv = KERN_INVALID_ADDRESS;
2553 entry = entry->next;
2557 need_wakeup = FALSE;
2558 if (first_entry == NULL) {
2559 result = vm_map_lookup_entry(map, start, &first_entry);
2560 if (!result && (flags & VM_MAP_WIRE_HOLESOK))
2561 first_entry = first_entry->next;
2563 KASSERT(result, ("vm_map_wire: lookup failed"));
2565 for (entry = first_entry; entry != &map->header && entry->start < end;
2566 entry = entry->next) {
2567 if ((entry->eflags & MAP_ENTRY_WIRE_SKIPPED) != 0)
2568 goto next_entry_done;
2571 * If VM_MAP_WIRE_HOLESOK was specified, an empty
2572 * space in the unwired region could have been mapped
2573 * while the map lock was dropped for faulting in the
2574 * pages or draining MAP_ENTRY_IN_TRANSITION.
2575 * Moreover, another thread could be simultaneously
2576 * wiring this new mapping entry. Detect these cases
2577 * and skip any entries marked as in transition by us.
2579 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 ||
2580 entry->wiring_thread != curthread) {
2581 KASSERT((flags & VM_MAP_WIRE_HOLESOK) != 0,
2582 ("vm_map_wire: !HOLESOK and new/changed entry"));
2586 if (rv == KERN_SUCCESS) {
2588 entry->eflags |= MAP_ENTRY_USER_WIRED;
2589 } else if (entry->wired_count == -1) {
2591 * Wiring failed on this entry. Thus, unwiring is
2594 entry->wired_count = 0;
2597 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0)
2598 entry->wired_count--;
2599 if (entry->wired_count == 0) {
2601 * Retain the map lock.
2603 vm_fault_unwire(map, entry->start, entry->end,
2604 entry->object.vm_object != NULL &&
2605 (entry->object.vm_object->flags &
2606 OBJ_FICTITIOUS) != 0);
2610 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0,
2611 ("vm_map_wire: in-transition flag missing %p", entry));
2612 KASSERT(entry->wiring_thread == curthread,
2613 ("vm_map_wire: alien wire %p", entry));
2614 entry->eflags &= ~(MAP_ENTRY_IN_TRANSITION |
2615 MAP_ENTRY_WIRE_SKIPPED);
2616 entry->wiring_thread = NULL;
2617 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
2618 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
2621 vm_map_simplify_entry(map, entry);
2632 * Push any dirty cached pages in the address range to their pager.
2633 * If syncio is TRUE, dirty pages are written synchronously.
2634 * If invalidate is TRUE, any cached pages are freed as well.
2636 * If the size of the region from start to end is zero, we are
2637 * supposed to flush all modified pages within the region containing
2638 * start. Unfortunately, a region can be split or coalesced with
2639 * neighboring regions, making it difficult to determine what the
2640 * original region was. Therefore, we approximate this requirement by
2641 * flushing the current region containing start.
2643 * Returns an error if any part of the specified range is not mapped.
2651 boolean_t invalidate)
2653 vm_map_entry_t current;
2654 vm_map_entry_t entry;
2657 vm_ooffset_t offset;
2658 unsigned int last_timestamp;
2661 vm_map_lock_read(map);
2662 VM_MAP_RANGE_CHECK(map, start, end);
2663 if (!vm_map_lookup_entry(map, start, &entry)) {
2664 vm_map_unlock_read(map);
2665 return (KERN_INVALID_ADDRESS);
2666 } else if (start == end) {
2667 start = entry->start;
2671 * Make a first pass to check for user-wired memory and holes.
2673 for (current = entry; current != &map->header && current->start < end;
2674 current = current->next) {
2675 if (invalidate && (current->eflags & MAP_ENTRY_USER_WIRED)) {
2676 vm_map_unlock_read(map);
2677 return (KERN_INVALID_ARGUMENT);
2679 if (end > current->end &&
2680 (current->next == &map->header ||
2681 current->end != current->next->start)) {
2682 vm_map_unlock_read(map);
2683 return (KERN_INVALID_ADDRESS);
2688 pmap_remove(map->pmap, start, end);
2692 * Make a second pass, cleaning/uncaching pages from the indicated
2695 for (current = entry; current != &map->header && current->start < end;) {
2696 offset = current->offset + (start - current->start);
2697 size = (end <= current->end ? end : current->end) - start;
2698 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
2700 vm_map_entry_t tentry;
2703 smap = current->object.sub_map;
2704 vm_map_lock_read(smap);
2705 (void) vm_map_lookup_entry(smap, offset, &tentry);
2706 tsize = tentry->end - offset;
2709 object = tentry->object.vm_object;
2710 offset = tentry->offset + (offset - tentry->start);
2711 vm_map_unlock_read(smap);
2713 object = current->object.vm_object;
2715 vm_object_reference(object);
2716 last_timestamp = map->timestamp;
2717 vm_map_unlock_read(map);
2718 if (!vm_object_sync(object, offset, size, syncio, invalidate))
2721 vm_object_deallocate(object);
2722 vm_map_lock_read(map);
2723 if (last_timestamp == map->timestamp ||
2724 !vm_map_lookup_entry(map, start, ¤t))
2725 current = current->next;
2728 vm_map_unlock_read(map);
2729 return (failed ? KERN_FAILURE : KERN_SUCCESS);
2733 * vm_map_entry_unwire: [ internal use only ]
2735 * Make the region specified by this entry pageable.
2737 * The map in question should be locked.
2738 * [This is the reason for this routine's existence.]
2741 vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry)
2743 vm_fault_unwire(map, entry->start, entry->end,
2744 entry->object.vm_object != NULL &&
2745 (entry->object.vm_object->flags & OBJ_FICTITIOUS) != 0);
2746 entry->wired_count = 0;
2750 vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map)
2753 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0)
2754 vm_object_deallocate(entry->object.vm_object);
2755 uma_zfree(system_map ? kmapentzone : mapentzone, entry);
2759 * vm_map_entry_delete: [ internal use only ]
2761 * Deallocate the given entry from the target map.
2764 vm_map_entry_delete(vm_map_t map, vm_map_entry_t entry)
2767 vm_pindex_t offidxstart, offidxend, count, size1;
2770 vm_map_entry_unlink(map, entry);
2771 object = entry->object.vm_object;
2772 size = entry->end - entry->start;
2775 if (entry->cred != NULL) {
2776 swap_release_by_cred(size, entry->cred);
2777 crfree(entry->cred);
2780 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0 &&
2782 KASSERT(entry->cred == NULL || object->cred == NULL ||
2783 (entry->eflags & MAP_ENTRY_NEEDS_COPY),
2784 ("OVERCOMMIT vm_map_entry_delete: both cred %p", entry));
2785 count = OFF_TO_IDX(size);
2786 offidxstart = OFF_TO_IDX(entry->offset);
2787 offidxend = offidxstart + count;
2788 VM_OBJECT_WLOCK(object);
2789 if (object->ref_count != 1 &&
2790 ((object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING ||
2791 object == kernel_object || object == kmem_object)) {
2792 vm_object_collapse(object);
2795 * The option OBJPR_NOTMAPPED can be passed here
2796 * because vm_map_delete() already performed
2797 * pmap_remove() on the only mapping to this range
2800 vm_object_page_remove(object, offidxstart, offidxend,
2802 if (object->type == OBJT_SWAP)
2803 swap_pager_freespace(object, offidxstart, count);
2804 if (offidxend >= object->size &&
2805 offidxstart < object->size) {
2806 size1 = object->size;
2807 object->size = offidxstart;
2808 if (object->cred != NULL) {
2809 size1 -= object->size;
2810 KASSERT(object->charge >= ptoa(size1),
2811 ("vm_map_entry_delete: object->charge < 0"));
2812 swap_release_by_cred(ptoa(size1), object->cred);
2813 object->charge -= ptoa(size1);
2817 VM_OBJECT_WUNLOCK(object);
2819 entry->object.vm_object = NULL;
2820 if (map->system_map)
2821 vm_map_entry_deallocate(entry, TRUE);
2823 entry->next = curthread->td_map_def_user;
2824 curthread->td_map_def_user = entry;
2829 * vm_map_delete: [ internal use only ]
2831 * Deallocates the given address range from the target
2835 vm_map_delete(vm_map_t map, vm_offset_t start, vm_offset_t end)
2837 vm_map_entry_t entry;
2838 vm_map_entry_t first_entry;
2840 VM_MAP_ASSERT_LOCKED(map);
2843 * Find the start of the region, and clip it
2845 if (!vm_map_lookup_entry(map, start, &first_entry))
2846 entry = first_entry->next;
2848 entry = first_entry;
2849 vm_map_clip_start(map, entry, start);
2853 * Step through all entries in this region
2855 while ((entry != &map->header) && (entry->start < end)) {
2856 vm_map_entry_t next;
2859 * Wait for wiring or unwiring of an entry to complete.
2860 * Also wait for any system wirings to disappear on
2863 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 ||
2864 (vm_map_pmap(map) != kernel_pmap &&
2865 vm_map_entry_system_wired_count(entry) != 0)) {
2866 unsigned int last_timestamp;
2867 vm_offset_t saved_start;
2868 vm_map_entry_t tmp_entry;
2870 saved_start = entry->start;
2871 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2872 last_timestamp = map->timestamp;
2873 (void) vm_map_unlock_and_wait(map, 0);
2875 if (last_timestamp + 1 != map->timestamp) {
2877 * Look again for the entry because the map was
2878 * modified while it was unlocked.
2879 * Specifically, the entry may have been
2880 * clipped, merged, or deleted.
2882 if (!vm_map_lookup_entry(map, saved_start,
2884 entry = tmp_entry->next;
2887 vm_map_clip_start(map, entry,
2893 vm_map_clip_end(map, entry, end);
2898 * Unwire before removing addresses from the pmap; otherwise,
2899 * unwiring will put the entries back in the pmap.
2901 if (entry->wired_count != 0) {
2902 vm_map_entry_unwire(map, entry);
2905 pmap_remove(map->pmap, entry->start, entry->end);
2908 * Delete the entry only after removing all pmap
2909 * entries pointing to its pages. (Otherwise, its
2910 * page frames may be reallocated, and any modify bits
2911 * will be set in the wrong object!)
2913 vm_map_entry_delete(map, entry);
2916 return (KERN_SUCCESS);
2922 * Remove the given address range from the target map.
2923 * This is the exported form of vm_map_delete.
2926 vm_map_remove(vm_map_t map, vm_offset_t start, vm_offset_t end)
2931 VM_MAP_RANGE_CHECK(map, start, end);
2932 result = vm_map_delete(map, start, end);
2938 * vm_map_check_protection:
2940 * Assert that the target map allows the specified privilege on the
2941 * entire address region given. The entire region must be allocated.
2943 * WARNING! This code does not and should not check whether the
2944 * contents of the region is accessible. For example a smaller file
2945 * might be mapped into a larger address space.
2947 * NOTE! This code is also called by munmap().
2949 * The map must be locked. A read lock is sufficient.
2952 vm_map_check_protection(vm_map_t map, vm_offset_t start, vm_offset_t end,
2953 vm_prot_t protection)
2955 vm_map_entry_t entry;
2956 vm_map_entry_t tmp_entry;
2958 if (!vm_map_lookup_entry(map, start, &tmp_entry))
2962 while (start < end) {
2963 if (entry == &map->header)
2968 if (start < entry->start)
2971 * Check protection associated with entry.
2973 if ((entry->protection & protection) != protection)
2975 /* go to next entry */
2977 entry = entry->next;
2983 * vm_map_copy_entry:
2985 * Copies the contents of the source entry to the destination
2986 * entry. The entries *must* be aligned properly.
2992 vm_map_entry_t src_entry,
2993 vm_map_entry_t dst_entry,
2994 vm_ooffset_t *fork_charge)
2996 vm_object_t src_object;
2997 vm_map_entry_t fake_entry;
3002 VM_MAP_ASSERT_LOCKED(dst_map);
3004 if ((dst_entry->eflags|src_entry->eflags) & MAP_ENTRY_IS_SUB_MAP)
3007 if (src_entry->wired_count == 0) {
3010 * If the source entry is marked needs_copy, it is already
3013 if ((src_entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) {
3014 pmap_protect(src_map->pmap,
3017 src_entry->protection & ~VM_PROT_WRITE);
3021 * Make a copy of the object.
3023 size = src_entry->end - src_entry->start;
3024 if ((src_object = src_entry->object.vm_object) != NULL) {
3025 VM_OBJECT_WLOCK(src_object);
3026 charged = ENTRY_CHARGED(src_entry);
3027 if ((src_object->handle == NULL) &&
3028 (src_object->type == OBJT_DEFAULT ||
3029 src_object->type == OBJT_SWAP)) {
3030 vm_object_collapse(src_object);
3031 if ((src_object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING) {
3032 vm_object_split(src_entry);
3033 src_object = src_entry->object.vm_object;
3036 vm_object_reference_locked(src_object);
3037 vm_object_clear_flag(src_object, OBJ_ONEMAPPING);
3038 if (src_entry->cred != NULL &&
3039 !(src_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
3040 KASSERT(src_object->cred == NULL,
3041 ("OVERCOMMIT: vm_map_copy_entry: cred %p",
3043 src_object->cred = src_entry->cred;
3044 src_object->charge = size;
3046 VM_OBJECT_WUNLOCK(src_object);
3047 dst_entry->object.vm_object = src_object;
3049 cred = curthread->td_ucred;
3051 dst_entry->cred = cred;
3052 *fork_charge += size;
3053 if (!(src_entry->eflags &
3054 MAP_ENTRY_NEEDS_COPY)) {
3056 src_entry->cred = cred;
3057 *fork_charge += size;
3060 src_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY);
3061 dst_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY);
3062 dst_entry->offset = src_entry->offset;
3063 if (src_entry->eflags & MAP_ENTRY_VN_WRITECNT) {
3065 * MAP_ENTRY_VN_WRITECNT cannot
3066 * indicate write reference from
3067 * src_entry, since the entry is
3068 * marked as needs copy. Allocate a
3069 * fake entry that is used to
3070 * decrement object->un_pager.vnp.writecount
3071 * at the appropriate time. Attach
3072 * fake_entry to the deferred list.
3074 fake_entry = vm_map_entry_create(dst_map);
3075 fake_entry->eflags = MAP_ENTRY_VN_WRITECNT;
3076 src_entry->eflags &= ~MAP_ENTRY_VN_WRITECNT;
3077 vm_object_reference(src_object);
3078 fake_entry->object.vm_object = src_object;
3079 fake_entry->start = src_entry->start;
3080 fake_entry->end = src_entry->end;
3081 fake_entry->next = curthread->td_map_def_user;
3082 curthread->td_map_def_user = fake_entry;
3085 dst_entry->object.vm_object = NULL;
3086 dst_entry->offset = 0;
3087 if (src_entry->cred != NULL) {
3088 dst_entry->cred = curthread->td_ucred;
3089 crhold(dst_entry->cred);
3090 *fork_charge += size;
3094 pmap_copy(dst_map->pmap, src_map->pmap, dst_entry->start,
3095 dst_entry->end - dst_entry->start, src_entry->start);
3098 * Of course, wired down pages can't be set copy-on-write.
3099 * Cause wired pages to be copied into the new map by
3100 * simulating faults (the new pages are pageable)
3102 vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry,
3108 * vmspace_map_entry_forked:
3109 * Update the newly-forked vmspace each time a map entry is inherited
3110 * or copied. The values for vm_dsize and vm_tsize are approximate
3111 * (and mostly-obsolete ideas in the face of mmap(2) et al.)
3114 vmspace_map_entry_forked(const struct vmspace *vm1, struct vmspace *vm2,
3115 vm_map_entry_t entry)
3117 vm_size_t entrysize;
3120 entrysize = entry->end - entry->start;
3121 vm2->vm_map.size += entrysize;
3122 if (entry->eflags & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP)) {
3123 vm2->vm_ssize += btoc(entrysize);
3124 } else if (entry->start >= (vm_offset_t)vm1->vm_daddr &&
3125 entry->start < (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)) {
3126 newend = MIN(entry->end,
3127 (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize));
3128 vm2->vm_dsize += btoc(newend - entry->start);
3129 } else if (entry->start >= (vm_offset_t)vm1->vm_taddr &&
3130 entry->start < (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)) {
3131 newend = MIN(entry->end,
3132 (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize));
3133 vm2->vm_tsize += btoc(newend - entry->start);
3139 * Create a new process vmspace structure and vm_map
3140 * based on those of an existing process. The new map
3141 * is based on the old map, according to the inheritance
3142 * values on the regions in that map.
3144 * XXX It might be worth coalescing the entries added to the new vmspace.
3146 * The source map must not be locked.
3149 vmspace_fork(struct vmspace *vm1, vm_ooffset_t *fork_charge)
3151 struct vmspace *vm2;
3152 vm_map_t new_map, old_map;
3153 vm_map_entry_t new_entry, old_entry;
3157 old_map = &vm1->vm_map;
3158 /* Copy immutable fields of vm1 to vm2. */
3159 vm2 = vmspace_alloc(old_map->min_offset, old_map->max_offset);
3162 vm2->vm_taddr = vm1->vm_taddr;
3163 vm2->vm_daddr = vm1->vm_daddr;
3164 vm2->vm_maxsaddr = vm1->vm_maxsaddr;
3165 vm_map_lock(old_map);
3167 vm_map_wait_busy(old_map);
3168 new_map = &vm2->vm_map;
3169 locked = vm_map_trylock(new_map); /* trylock to silence WITNESS */
3170 KASSERT(locked, ("vmspace_fork: lock failed"));
3172 old_entry = old_map->header.next;
3174 while (old_entry != &old_map->header) {
3175 if (old_entry->eflags & MAP_ENTRY_IS_SUB_MAP)
3176 panic("vm_map_fork: encountered a submap");
3178 switch (old_entry->inheritance) {
3179 case VM_INHERIT_NONE:
3182 case VM_INHERIT_SHARE:
3184 * Clone the entry, creating the shared object if necessary.
3186 object = old_entry->object.vm_object;
3187 if (object == NULL) {
3188 object = vm_object_allocate(OBJT_DEFAULT,
3189 atop(old_entry->end - old_entry->start));
3190 old_entry->object.vm_object = object;
3191 old_entry->offset = 0;
3192 if (old_entry->cred != NULL) {
3193 object->cred = old_entry->cred;
3194 object->charge = old_entry->end -
3196 old_entry->cred = NULL;
3201 * Add the reference before calling vm_object_shadow
3202 * to insure that a shadow object is created.
3204 vm_object_reference(object);
3205 if (old_entry->eflags & MAP_ENTRY_NEEDS_COPY) {
3206 vm_object_shadow(&old_entry->object.vm_object,
3208 old_entry->end - old_entry->start);
3209 old_entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
3210 /* Transfer the second reference too. */
3211 vm_object_reference(
3212 old_entry->object.vm_object);
3215 * As in vm_map_simplify_entry(), the
3216 * vnode lock will not be acquired in
3217 * this call to vm_object_deallocate().
3219 vm_object_deallocate(object);
3220 object = old_entry->object.vm_object;
3222 VM_OBJECT_WLOCK(object);
3223 vm_object_clear_flag(object, OBJ_ONEMAPPING);
3224 if (old_entry->cred != NULL) {
3225 KASSERT(object->cred == NULL, ("vmspace_fork both cred"));
3226 object->cred = old_entry->cred;
3227 object->charge = old_entry->end - old_entry->start;
3228 old_entry->cred = NULL;
3232 * Assert the correct state of the vnode
3233 * v_writecount while the object is locked, to
3234 * not relock it later for the assertion
3237 if (old_entry->eflags & MAP_ENTRY_VN_WRITECNT &&
3238 object->type == OBJT_VNODE) {
3239 KASSERT(((struct vnode *)object->handle)->
3241 ("vmspace_fork: v_writecount %p", object));
3242 KASSERT(object->un_pager.vnp.writemappings > 0,
3243 ("vmspace_fork: vnp.writecount %p",
3246 VM_OBJECT_WUNLOCK(object);
3249 * Clone the entry, referencing the shared object.
3251 new_entry = vm_map_entry_create(new_map);
3252 *new_entry = *old_entry;
3253 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED |
3254 MAP_ENTRY_IN_TRANSITION);
3255 new_entry->wiring_thread = NULL;
3256 new_entry->wired_count = 0;
3257 if (new_entry->eflags & MAP_ENTRY_VN_WRITECNT) {
3258 vnode_pager_update_writecount(object,
3259 new_entry->start, new_entry->end);
3263 * Insert the entry into the new map -- we know we're
3264 * inserting at the end of the new map.
3266 vm_map_entry_link(new_map, new_map->header.prev,
3268 vmspace_map_entry_forked(vm1, vm2, new_entry);
3271 * Update the physical map
3273 pmap_copy(new_map->pmap, old_map->pmap,
3275 (old_entry->end - old_entry->start),
3279 case VM_INHERIT_COPY:
3281 * Clone the entry and link into the map.
3283 new_entry = vm_map_entry_create(new_map);
3284 *new_entry = *old_entry;
3286 * Copied entry is COW over the old object.
3288 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED |
3289 MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_VN_WRITECNT);
3290 new_entry->wiring_thread = NULL;
3291 new_entry->wired_count = 0;
3292 new_entry->object.vm_object = NULL;
3293 new_entry->cred = NULL;
3294 vm_map_entry_link(new_map, new_map->header.prev,
3296 vmspace_map_entry_forked(vm1, vm2, new_entry);
3297 vm_map_copy_entry(old_map, new_map, old_entry,
3298 new_entry, fork_charge);
3301 old_entry = old_entry->next;
3304 * Use inlined vm_map_unlock() to postpone handling the deferred
3305 * map entries, which cannot be done until both old_map and
3306 * new_map locks are released.
3308 sx_xunlock(&old_map->lock);
3309 sx_xunlock(&new_map->lock);
3310 vm_map_process_deferred();
3316 vm_map_stack(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
3317 vm_prot_t prot, vm_prot_t max, int cow)
3319 vm_map_entry_t new_entry, prev_entry;
3320 vm_offset_t bot, top;
3321 vm_size_t growsize, init_ssize;
3323 rlim_t lmemlim, vmemlim;
3326 * The stack orientation is piggybacked with the cow argument.
3327 * Extract it into orient and mask the cow argument so that we
3328 * don't pass it around further.
3329 * NOTE: We explicitly allow bi-directional stacks.
3331 orient = cow & (MAP_STACK_GROWS_DOWN|MAP_STACK_GROWS_UP);
3333 KASSERT(orient != 0, ("No stack grow direction"));
3335 if (addrbos < vm_map_min(map) ||
3336 addrbos > vm_map_max(map) ||
3337 addrbos + max_ssize < addrbos)
3338 return (KERN_NO_SPACE);
3340 growsize = sgrowsiz;
3341 init_ssize = (max_ssize < growsize) ? max_ssize : growsize;
3344 lmemlim = lim_cur(curproc, RLIMIT_MEMLOCK);
3345 vmemlim = lim_cur(curproc, RLIMIT_VMEM);
3346 PROC_UNLOCK(curproc);
3350 /* If addr is already mapped, no go */
3351 if (vm_map_lookup_entry(map, addrbos, &prev_entry)) {
3353 return (KERN_NO_SPACE);
3356 if (!old_mlock && map->flags & MAP_WIREFUTURE) {
3357 if (ptoa(pmap_wired_count(map->pmap)) + init_ssize > lmemlim) {
3359 return (KERN_NO_SPACE);
3363 /* If we would blow our VMEM resource limit, no go */
3364 if (map->size + init_ssize > vmemlim) {
3366 return (KERN_NO_SPACE);
3370 * If we can't accomodate max_ssize in the current mapping, no go.
3371 * However, we need to be aware that subsequent user mappings might
3372 * map into the space we have reserved for stack, and currently this
3373 * space is not protected.
3375 * Hopefully we will at least detect this condition when we try to
3378 if ((prev_entry->next != &map->header) &&
3379 (prev_entry->next->start < addrbos + max_ssize)) {
3381 return (KERN_NO_SPACE);
3385 * We initially map a stack of only init_ssize. We will grow as
3386 * needed later. Depending on the orientation of the stack (i.e.
3387 * the grow direction) we either map at the top of the range, the
3388 * bottom of the range or in the middle.
3390 * Note: we would normally expect prot and max to be VM_PROT_ALL,
3391 * and cow to be 0. Possibly we should eliminate these as input
3392 * parameters, and just pass these values here in the insert call.
3394 if (orient == MAP_STACK_GROWS_DOWN)
3395 bot = addrbos + max_ssize - init_ssize;
3396 else if (orient == MAP_STACK_GROWS_UP)
3399 bot = round_page(addrbos + max_ssize/2 - init_ssize/2);
3400 top = bot + init_ssize;
3401 rv = vm_map_insert(map, NULL, 0, bot, top, prot, max, cow);
3403 /* Now set the avail_ssize amount. */
3404 if (rv == KERN_SUCCESS) {
3405 if (prev_entry != &map->header)
3406 vm_map_clip_end(map, prev_entry, bot);
3407 new_entry = prev_entry->next;
3408 if (new_entry->end != top || new_entry->start != bot)
3409 panic("Bad entry start/end for new stack entry");
3411 new_entry->avail_ssize = max_ssize - init_ssize;
3412 if (orient & MAP_STACK_GROWS_DOWN)
3413 new_entry->eflags |= MAP_ENTRY_GROWS_DOWN;
3414 if (orient & MAP_STACK_GROWS_UP)
3415 new_entry->eflags |= MAP_ENTRY_GROWS_UP;
3422 static int stack_guard_page = 0;
3423 TUNABLE_INT("security.bsd.stack_guard_page", &stack_guard_page);
3424 SYSCTL_INT(_security_bsd, OID_AUTO, stack_guard_page, CTLFLAG_RW,
3425 &stack_guard_page, 0,
3426 "Insert stack guard page ahead of the growable segments.");
3428 /* Attempts to grow a vm stack entry. Returns KERN_SUCCESS if the
3429 * desired address is already mapped, or if we successfully grow
3430 * the stack. Also returns KERN_SUCCESS if addr is outside the
3431 * stack range (this is strange, but preserves compatibility with
3432 * the grow function in vm_machdep.c).
3435 vm_map_growstack(struct proc *p, vm_offset_t addr)
3437 vm_map_entry_t next_entry, prev_entry;
3438 vm_map_entry_t new_entry, stack_entry;
3439 struct vmspace *vm = p->p_vmspace;
3440 vm_map_t map = &vm->vm_map;
3443 size_t grow_amount, max_grow;
3444 rlim_t lmemlim, stacklim, vmemlim;
3445 int is_procstack, rv;
3456 lmemlim = lim_cur(p, RLIMIT_MEMLOCK);
3457 stacklim = lim_cur(p, RLIMIT_STACK);
3458 vmemlim = lim_cur(p, RLIMIT_VMEM);
3461 vm_map_lock_read(map);
3463 /* If addr is already in the entry range, no need to grow.*/
3464 if (vm_map_lookup_entry(map, addr, &prev_entry)) {
3465 vm_map_unlock_read(map);
3466 return (KERN_SUCCESS);
3469 next_entry = prev_entry->next;
3470 if (!(prev_entry->eflags & MAP_ENTRY_GROWS_UP)) {
3472 * This entry does not grow upwards. Since the address lies
3473 * beyond this entry, the next entry (if one exists) has to
3474 * be a downward growable entry. The entry list header is
3475 * never a growable entry, so it suffices to check the flags.
3477 if (!(next_entry->eflags & MAP_ENTRY_GROWS_DOWN)) {
3478 vm_map_unlock_read(map);
3479 return (KERN_SUCCESS);
3481 stack_entry = next_entry;
3484 * This entry grows upward. If the next entry does not at
3485 * least grow downwards, this is the entry we need to grow.
3486 * otherwise we have two possible choices and we have to
3489 if (next_entry->eflags & MAP_ENTRY_GROWS_DOWN) {
3491 * We have two choices; grow the entry closest to
3492 * the address to minimize the amount of growth.
3494 if (addr - prev_entry->end <= next_entry->start - addr)
3495 stack_entry = prev_entry;
3497 stack_entry = next_entry;
3499 stack_entry = prev_entry;
3502 if (stack_entry == next_entry) {
3503 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_DOWN, ("foo"));
3504 KASSERT(addr < stack_entry->start, ("foo"));
3505 end = (prev_entry != &map->header) ? prev_entry->end :
3506 stack_entry->start - stack_entry->avail_ssize;
3507 grow_amount = roundup(stack_entry->start - addr, PAGE_SIZE);
3508 max_grow = stack_entry->start - end;
3510 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_UP, ("foo"));
3511 KASSERT(addr >= stack_entry->end, ("foo"));
3512 end = (next_entry != &map->header) ? next_entry->start :
3513 stack_entry->end + stack_entry->avail_ssize;
3514 grow_amount = roundup(addr + 1 - stack_entry->end, PAGE_SIZE);
3515 max_grow = end - stack_entry->end;
3518 if (grow_amount > stack_entry->avail_ssize) {
3519 vm_map_unlock_read(map);
3520 return (KERN_NO_SPACE);
3524 * If there is no longer enough space between the entries nogo, and
3525 * adjust the available space. Note: this should only happen if the
3526 * user has mapped into the stack area after the stack was created,
3527 * and is probably an error.
3529 * This also effectively destroys any guard page the user might have
3530 * intended by limiting the stack size.
3532 if (grow_amount + (stack_guard_page ? PAGE_SIZE : 0) > max_grow) {
3533 if (vm_map_lock_upgrade(map))
3536 stack_entry->avail_ssize = max_grow;
3539 return (KERN_NO_SPACE);
3542 is_procstack = (addr >= (vm_offset_t)vm->vm_maxsaddr) ? 1 : 0;
3545 * If this is the main process stack, see if we're over the stack
3548 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) {
3549 vm_map_unlock_read(map);
3550 return (KERN_NO_SPACE);
3555 racct_set(p, RACCT_STACK, ctob(vm->vm_ssize) + grow_amount)) {
3557 vm_map_unlock_read(map);
3558 return (KERN_NO_SPACE);
3563 /* Round up the grow amount modulo sgrowsiz */
3564 growsize = sgrowsiz;
3565 grow_amount = roundup(grow_amount, growsize);
3566 if (grow_amount > stack_entry->avail_ssize)
3567 grow_amount = stack_entry->avail_ssize;
3568 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) {
3569 grow_amount = trunc_page((vm_size_t)stacklim) -
3574 limit = racct_get_available(p, RACCT_STACK);
3576 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > limit))
3577 grow_amount = limit - ctob(vm->vm_ssize);
3579 if (!old_mlock && map->flags & MAP_WIREFUTURE) {
3580 if (ptoa(pmap_wired_count(map->pmap)) + grow_amount > lmemlim) {
3581 vm_map_unlock_read(map);
3587 if (racct_set(p, RACCT_MEMLOCK,
3588 ptoa(pmap_wired_count(map->pmap)) + grow_amount)) {
3590 vm_map_unlock_read(map);
3597 /* If we would blow our VMEM resource limit, no go */
3598 if (map->size + grow_amount > vmemlim) {
3599 vm_map_unlock_read(map);
3605 if (racct_set(p, RACCT_VMEM, map->size + grow_amount)) {
3607 vm_map_unlock_read(map);
3614 if (vm_map_lock_upgrade(map))
3617 if (stack_entry == next_entry) {
3621 /* Get the preliminary new entry start value */
3622 addr = stack_entry->start - grow_amount;
3625 * If this puts us into the previous entry, cut back our
3626 * growth to the available space. Also, see the note above.
3629 stack_entry->avail_ssize = max_grow;
3631 if (stack_guard_page)
3635 rv = vm_map_insert(map, NULL, 0, addr, stack_entry->start,
3636 next_entry->protection, next_entry->max_protection, 0);
3638 /* Adjust the available stack space by the amount we grew. */
3639 if (rv == KERN_SUCCESS) {
3640 if (prev_entry != &map->header)
3641 vm_map_clip_end(map, prev_entry, addr);
3642 new_entry = prev_entry->next;
3643 KASSERT(new_entry == stack_entry->prev, ("foo"));
3644 KASSERT(new_entry->end == stack_entry->start, ("foo"));
3645 KASSERT(new_entry->start == addr, ("foo"));
3646 grow_amount = new_entry->end - new_entry->start;
3647 new_entry->avail_ssize = stack_entry->avail_ssize -
3649 stack_entry->eflags &= ~MAP_ENTRY_GROWS_DOWN;
3650 new_entry->eflags |= MAP_ENTRY_GROWS_DOWN;
3656 addr = stack_entry->end + grow_amount;
3659 * If this puts us into the next entry, cut back our growth
3660 * to the available space. Also, see the note above.
3663 stack_entry->avail_ssize = end - stack_entry->end;
3665 if (stack_guard_page)
3669 grow_amount = addr - stack_entry->end;
3670 cred = stack_entry->cred;
3671 if (cred == NULL && stack_entry->object.vm_object != NULL)
3672 cred = stack_entry->object.vm_object->cred;
3673 if (cred != NULL && !swap_reserve_by_cred(grow_amount, cred))
3675 /* Grow the underlying object if applicable. */
3676 else if (stack_entry->object.vm_object == NULL ||
3677 vm_object_coalesce(stack_entry->object.vm_object,
3678 stack_entry->offset,
3679 (vm_size_t)(stack_entry->end - stack_entry->start),
3680 (vm_size_t)grow_amount, cred != NULL)) {
3681 map->size += (addr - stack_entry->end);
3682 /* Update the current entry. */
3683 stack_entry->end = addr;
3684 stack_entry->avail_ssize -= grow_amount;
3685 vm_map_entry_resize_free(map, stack_entry);
3688 if (next_entry != &map->header)
3689 vm_map_clip_start(map, next_entry, addr);
3694 if (rv == KERN_SUCCESS && is_procstack)
3695 vm->vm_ssize += btoc(grow_amount);
3700 * Heed the MAP_WIREFUTURE flag if it was set for this process.
3702 if (rv == KERN_SUCCESS && (map->flags & MAP_WIREFUTURE)) {
3704 (stack_entry == next_entry) ? addr : addr - grow_amount,
3705 (stack_entry == next_entry) ? stack_entry->start : addr,
3706 (p->p_flag & P_SYSTEM)
3707 ? VM_MAP_WIRE_SYSTEM|VM_MAP_WIRE_NOHOLES
3708 : VM_MAP_WIRE_USER|VM_MAP_WIRE_NOHOLES);
3713 if (rv != KERN_SUCCESS) {
3715 error = racct_set(p, RACCT_VMEM, map->size);
3716 KASSERT(error == 0, ("decreasing RACCT_VMEM failed"));
3718 error = racct_set(p, RACCT_MEMLOCK,
3719 ptoa(pmap_wired_count(map->pmap)));
3720 KASSERT(error == 0, ("decreasing RACCT_MEMLOCK failed"));
3722 error = racct_set(p, RACCT_STACK, ctob(vm->vm_ssize));
3723 KASSERT(error == 0, ("decreasing RACCT_STACK failed"));
3732 * Unshare the specified VM space for exec. If other processes are
3733 * mapped to it, then create a new one. The new vmspace is null.
3736 vmspace_exec(struct proc *p, vm_offset_t minuser, vm_offset_t maxuser)
3738 struct vmspace *oldvmspace = p->p_vmspace;
3739 struct vmspace *newvmspace;
3741 newvmspace = vmspace_alloc(minuser, maxuser);
3742 if (newvmspace == NULL)
3744 newvmspace->vm_swrss = oldvmspace->vm_swrss;
3746 * This code is written like this for prototype purposes. The
3747 * goal is to avoid running down the vmspace here, but let the
3748 * other process's that are still using the vmspace to finally
3749 * run it down. Even though there is little or no chance of blocking
3750 * here, it is a good idea to keep this form for future mods.
3752 PROC_VMSPACE_LOCK(p);
3753 p->p_vmspace = newvmspace;
3754 PROC_VMSPACE_UNLOCK(p);
3755 if (p == curthread->td_proc)
3756 pmap_activate(curthread);
3757 vmspace_free(oldvmspace);
3762 * Unshare the specified VM space for forcing COW. This
3763 * is called by rfork, for the (RFMEM|RFPROC) == 0 case.
3766 vmspace_unshare(struct proc *p)
3768 struct vmspace *oldvmspace = p->p_vmspace;
3769 struct vmspace *newvmspace;
3770 vm_ooffset_t fork_charge;
3772 if (oldvmspace->vm_refcnt == 1)
3775 newvmspace = vmspace_fork(oldvmspace, &fork_charge);
3776 if (newvmspace == NULL)
3778 if (!swap_reserve_by_cred(fork_charge, p->p_ucred)) {
3779 vmspace_free(newvmspace);
3782 PROC_VMSPACE_LOCK(p);
3783 p->p_vmspace = newvmspace;
3784 PROC_VMSPACE_UNLOCK(p);
3785 if (p == curthread->td_proc)
3786 pmap_activate(curthread);
3787 vmspace_free(oldvmspace);
3794 * Finds the VM object, offset, and
3795 * protection for a given virtual address in the
3796 * specified map, assuming a page fault of the
3799 * Leaves the map in question locked for read; return
3800 * values are guaranteed until a vm_map_lookup_done
3801 * call is performed. Note that the map argument
3802 * is in/out; the returned map must be used in
3803 * the call to vm_map_lookup_done.
3805 * A handle (out_entry) is returned for use in
3806 * vm_map_lookup_done, to make that fast.
3808 * If a lookup is requested with "write protection"
3809 * specified, the map may be changed to perform virtual
3810 * copying operations, although the data referenced will
3814 vm_map_lookup(vm_map_t *var_map, /* IN/OUT */
3816 vm_prot_t fault_typea,
3817 vm_map_entry_t *out_entry, /* OUT */
3818 vm_object_t *object, /* OUT */
3819 vm_pindex_t *pindex, /* OUT */
3820 vm_prot_t *out_prot, /* OUT */
3821 boolean_t *wired) /* OUT */
3823 vm_map_entry_t entry;
3824 vm_map_t map = *var_map;
3826 vm_prot_t fault_type = fault_typea;
3827 vm_object_t eobject;
3833 vm_map_lock_read(map);
3836 * Lookup the faulting address.
3838 if (!vm_map_lookup_entry(map, vaddr, out_entry)) {
3839 vm_map_unlock_read(map);
3840 return (KERN_INVALID_ADDRESS);
3848 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
3849 vm_map_t old_map = map;
3851 *var_map = map = entry->object.sub_map;
3852 vm_map_unlock_read(old_map);
3857 * Check whether this task is allowed to have this page.
3859 prot = entry->protection;
3860 fault_type &= (VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE);
3861 if ((fault_type & prot) != fault_type || prot == VM_PROT_NONE) {
3862 vm_map_unlock_read(map);
3863 return (KERN_PROTECTION_FAILURE);
3865 if ((entry->eflags & MAP_ENTRY_USER_WIRED) &&
3866 (entry->eflags & MAP_ENTRY_COW) &&
3867 (fault_type & VM_PROT_WRITE)) {
3868 vm_map_unlock_read(map);
3869 return (KERN_PROTECTION_FAILURE);
3871 if ((fault_typea & VM_PROT_COPY) != 0 &&
3872 (entry->max_protection & VM_PROT_WRITE) == 0 &&
3873 (entry->eflags & MAP_ENTRY_COW) == 0) {
3874 vm_map_unlock_read(map);
3875 return (KERN_PROTECTION_FAILURE);
3879 * If this page is not pageable, we have to get it for all possible
3882 *wired = (entry->wired_count != 0);
3884 fault_type = entry->protection;
3885 size = entry->end - entry->start;
3887 * If the entry was copy-on-write, we either ...
3889 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
3891 * If we want to write the page, we may as well handle that
3892 * now since we've got the map locked.
3894 * If we don't need to write the page, we just demote the
3895 * permissions allowed.
3897 if ((fault_type & VM_PROT_WRITE) != 0 ||
3898 (fault_typea & VM_PROT_COPY) != 0) {
3900 * Make a new object, and place it in the object
3901 * chain. Note that no new references have appeared
3902 * -- one just moved from the map to the new
3905 if (vm_map_lock_upgrade(map))
3908 if (entry->cred == NULL) {
3910 * The debugger owner is charged for
3913 cred = curthread->td_ucred;
3915 if (!swap_reserve_by_cred(size, cred)) {
3918 return (KERN_RESOURCE_SHORTAGE);
3922 vm_object_shadow(&entry->object.vm_object,
3923 &entry->offset, size);
3924 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
3925 eobject = entry->object.vm_object;
3926 if (eobject->cred != NULL) {
3928 * The object was not shadowed.
3930 swap_release_by_cred(size, entry->cred);
3931 crfree(entry->cred);
3933 } else if (entry->cred != NULL) {
3934 VM_OBJECT_WLOCK(eobject);
3935 eobject->cred = entry->cred;
3936 eobject->charge = size;
3937 VM_OBJECT_WUNLOCK(eobject);
3941 vm_map_lock_downgrade(map);
3944 * We're attempting to read a copy-on-write page --
3945 * don't allow writes.
3947 prot &= ~VM_PROT_WRITE;
3952 * Create an object if necessary.
3954 if (entry->object.vm_object == NULL &&
3956 if (vm_map_lock_upgrade(map))
3958 entry->object.vm_object = vm_object_allocate(OBJT_DEFAULT,
3961 if (entry->cred != NULL) {
3962 VM_OBJECT_WLOCK(entry->object.vm_object);
3963 entry->object.vm_object->cred = entry->cred;
3964 entry->object.vm_object->charge = size;
3965 VM_OBJECT_WUNLOCK(entry->object.vm_object);
3968 vm_map_lock_downgrade(map);
3972 * Return the object/offset from this entry. If the entry was
3973 * copy-on-write or empty, it has been fixed up.
3975 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
3976 *object = entry->object.vm_object;
3979 return (KERN_SUCCESS);
3983 * vm_map_lookup_locked:
3985 * Lookup the faulting address. A version of vm_map_lookup that returns
3986 * KERN_FAILURE instead of blocking on map lock or memory allocation.
3989 vm_map_lookup_locked(vm_map_t *var_map, /* IN/OUT */
3991 vm_prot_t fault_typea,
3992 vm_map_entry_t *out_entry, /* OUT */
3993 vm_object_t *object, /* OUT */
3994 vm_pindex_t *pindex, /* OUT */
3995 vm_prot_t *out_prot, /* OUT */
3996 boolean_t *wired) /* OUT */
3998 vm_map_entry_t entry;
3999 vm_map_t map = *var_map;
4001 vm_prot_t fault_type = fault_typea;
4004 * Lookup the faulting address.
4006 if (!vm_map_lookup_entry(map, vaddr, out_entry))
4007 return (KERN_INVALID_ADDRESS);
4012 * Fail if the entry refers to a submap.
4014 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP)
4015 return (KERN_FAILURE);
4018 * Check whether this task is allowed to have this page.
4020 prot = entry->protection;
4021 fault_type &= VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE;
4022 if ((fault_type & prot) != fault_type)
4023 return (KERN_PROTECTION_FAILURE);
4024 if ((entry->eflags & MAP_ENTRY_USER_WIRED) &&
4025 (entry->eflags & MAP_ENTRY_COW) &&
4026 (fault_type & VM_PROT_WRITE))
4027 return (KERN_PROTECTION_FAILURE);
4030 * If this page is not pageable, we have to get it for all possible
4033 *wired = (entry->wired_count != 0);
4035 fault_type = entry->protection;
4037 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
4039 * Fail if the entry was copy-on-write for a write fault.
4041 if (fault_type & VM_PROT_WRITE)
4042 return (KERN_FAILURE);
4044 * We're attempting to read a copy-on-write page --
4045 * don't allow writes.
4047 prot &= ~VM_PROT_WRITE;
4051 * Fail if an object should be created.
4053 if (entry->object.vm_object == NULL && !map->system_map)
4054 return (KERN_FAILURE);
4057 * Return the object/offset from this entry. If the entry was
4058 * copy-on-write or empty, it has been fixed up.
4060 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
4061 *object = entry->object.vm_object;
4064 return (KERN_SUCCESS);
4068 * vm_map_lookup_done:
4070 * Releases locks acquired by a vm_map_lookup
4071 * (according to the handle returned by that lookup).
4074 vm_map_lookup_done(vm_map_t map, vm_map_entry_t entry)
4077 * Unlock the main-level map
4079 vm_map_unlock_read(map);
4082 #include "opt_ddb.h"
4084 #include <sys/kernel.h>
4086 #include <ddb/ddb.h>
4089 vm_map_print(vm_map_t map)
4091 vm_map_entry_t entry;
4093 db_iprintf("Task map %p: pmap=%p, nentries=%d, version=%u\n",
4095 (void *)map->pmap, map->nentries, map->timestamp);
4098 for (entry = map->header.next; entry != &map->header;
4099 entry = entry->next) {
4100 db_iprintf("map entry %p: start=%p, end=%p\n",
4101 (void *)entry, (void *)entry->start, (void *)entry->end);
4103 static char *inheritance_name[4] =
4104 {"share", "copy", "none", "donate_copy"};
4106 db_iprintf(" prot=%x/%x/%s",
4108 entry->max_protection,
4109 inheritance_name[(int)(unsigned char)entry->inheritance]);
4110 if (entry->wired_count != 0)
4111 db_printf(", wired");
4113 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
4114 db_printf(", share=%p, offset=0x%jx\n",
4115 (void *)entry->object.sub_map,
4116 (uintmax_t)entry->offset);
4117 if ((entry->prev == &map->header) ||
4118 (entry->prev->object.sub_map !=
4119 entry->object.sub_map)) {
4121 vm_map_print((vm_map_t)entry->object.sub_map);
4125 if (entry->cred != NULL)
4126 db_printf(", ruid %d", entry->cred->cr_ruid);
4127 db_printf(", object=%p, offset=0x%jx",
4128 (void *)entry->object.vm_object,
4129 (uintmax_t)entry->offset);
4130 if (entry->object.vm_object && entry->object.vm_object->cred)
4131 db_printf(", obj ruid %d charge %jx",
4132 entry->object.vm_object->cred->cr_ruid,
4133 (uintmax_t)entry->object.vm_object->charge);
4134 if (entry->eflags & MAP_ENTRY_COW)
4135 db_printf(", copy (%s)",
4136 (entry->eflags & MAP_ENTRY_NEEDS_COPY) ? "needed" : "done");
4139 if ((entry->prev == &map->header) ||
4140 (entry->prev->object.vm_object !=
4141 entry->object.vm_object)) {
4143 vm_object_print((db_expr_t)(intptr_t)
4144 entry->object.vm_object,
4153 DB_SHOW_COMMAND(map, map)
4157 db_printf("usage: show map <addr>\n");
4160 vm_map_print((vm_map_t)addr);
4163 DB_SHOW_COMMAND(procvm, procvm)
4168 p = (struct proc *) addr;
4173 db_printf("p = %p, vmspace = %p, map = %p, pmap = %p\n",
4174 (void *)p, (void *)p->p_vmspace, (void *)&p->p_vmspace->vm_map,
4175 (void *)vmspace_pmap(p->p_vmspace));
4177 vm_map_print((vm_map_t)&p->p_vmspace->vm_map);