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 uma_prealloc(kmapentzone, MAX_KMAPENT);
201 mapentzone = uma_zcreate("MAP ENTRY", sizeof(struct vm_map_entry),
202 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
206 vmspace_zfini(void *mem, int size)
210 vm = (struct vmspace *)mem;
211 vm_map_zfini(&vm->vm_map, sizeof(vm->vm_map));
215 vmspace_zinit(void *mem, int size, int flags)
219 vm = (struct vmspace *)mem;
221 vm->vm_map.pmap = NULL;
222 (void)vm_map_zinit(&vm->vm_map, sizeof(vm->vm_map), flags);
227 vm_map_zfini(void *mem, int size)
232 mtx_destroy(&map->system_mtx);
233 sx_destroy(&map->lock);
237 vm_map_zinit(void *mem, int size, int flags)
242 memset(map, 0, sizeof(*map));
243 mtx_init(&map->system_mtx, "vm map (system)", NULL, MTX_DEF | MTX_DUPOK);
244 sx_init(&map->lock, "vm map (user)");
250 vmspace_zdtor(void *mem, int size, void *arg)
254 vm = (struct vmspace *)mem;
256 vm_map_zdtor(&vm->vm_map, sizeof(vm->vm_map), arg);
259 vm_map_zdtor(void *mem, int size, void *arg)
264 KASSERT(map->nentries == 0,
265 ("map %p nentries == %d on free.",
266 map, map->nentries));
267 KASSERT(map->size == 0,
268 ("map %p size == %lu on free.",
269 map, (unsigned long)map->size));
271 #endif /* INVARIANTS */
274 * Allocate a vmspace structure, including a vm_map and pmap,
275 * and initialize those structures. The refcnt is set to 1.
278 vmspace_alloc(min, max)
279 vm_offset_t min, max;
283 vm = uma_zalloc(vmspace_zone, M_WAITOK);
284 if (vm->vm_map.pmap == NULL && !pmap_pinit(vmspace_pmap(vm))) {
285 uma_zfree(vmspace_zone, vm);
288 CTR1(KTR_VM, "vmspace_alloc: %p", vm);
289 _vm_map_init(&vm->vm_map, vmspace_pmap(vm), min, max);
305 uma_zone_reserve_kva(kmapentzone, lmin(cnt.v_page_count,
306 (VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS) / PAGE_SIZE) / 8 +
307 maxproc * 2 + maxfiles);
308 vmspace_zone = uma_zcreate("VMSPACE", sizeof(struct vmspace), NULL,
314 vmspace_zinit, vmspace_zfini, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
318 vmspace_container_reset(struct proc *p)
323 racct_set(p, RACCT_DATA, 0);
324 racct_set(p, RACCT_STACK, 0);
325 racct_set(p, RACCT_RSS, 0);
326 racct_set(p, RACCT_MEMLOCK, 0);
327 racct_set(p, RACCT_VMEM, 0);
333 vmspace_dofree(struct vmspace *vm)
336 CTR1(KTR_VM, "vmspace_free: %p", vm);
339 * Make sure any SysV shm is freed, it might not have been in
345 * Lock the map, to wait out all other references to it.
346 * Delete all of the mappings and pages they hold, then call
347 * the pmap module to reclaim anything left.
349 (void)vm_map_remove(&vm->vm_map, vm->vm_map.min_offset,
350 vm->vm_map.max_offset);
352 pmap_release(vmspace_pmap(vm));
353 vm->vm_map.pmap = NULL;
354 uma_zfree(vmspace_zone, vm);
358 vmspace_free(struct vmspace *vm)
361 if (vm->vm_refcnt == 0)
362 panic("vmspace_free: attempt to free already freed vmspace");
364 if (atomic_fetchadd_int(&vm->vm_refcnt, -1) == 1)
369 vmspace_exitfree(struct proc *p)
373 PROC_VMSPACE_LOCK(p);
376 PROC_VMSPACE_UNLOCK(p);
377 KASSERT(vm == &vmspace0, ("vmspace_exitfree: wrong vmspace"));
382 vmspace_exit(struct thread *td)
389 * Release user portion of address space.
390 * This releases references to vnodes,
391 * which could cause I/O if the file has been unlinked.
392 * Need to do this early enough that we can still sleep.
394 * The last exiting process to reach this point releases as
395 * much of the environment as it can. vmspace_dofree() is the
396 * slower fallback in case another process had a temporary
397 * reference to the vmspace.
402 atomic_add_int(&vmspace0.vm_refcnt, 1);
404 refcnt = vm->vm_refcnt;
405 if (refcnt > 1 && p->p_vmspace != &vmspace0) {
406 /* Switch now since other proc might free vmspace */
407 PROC_VMSPACE_LOCK(p);
408 p->p_vmspace = &vmspace0;
409 PROC_VMSPACE_UNLOCK(p);
412 } while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt - 1));
414 if (p->p_vmspace != vm) {
415 /* vmspace not yet freed, switch back */
416 PROC_VMSPACE_LOCK(p);
418 PROC_VMSPACE_UNLOCK(p);
421 pmap_remove_pages(vmspace_pmap(vm));
422 /* Switch now since this proc will free vmspace */
423 PROC_VMSPACE_LOCK(p);
424 p->p_vmspace = &vmspace0;
425 PROC_VMSPACE_UNLOCK(p);
429 vmspace_container_reset(p);
432 /* Acquire reference to vmspace owned by another process. */
435 vmspace_acquire_ref(struct proc *p)
440 PROC_VMSPACE_LOCK(p);
443 PROC_VMSPACE_UNLOCK(p);
447 refcnt = vm->vm_refcnt;
448 if (refcnt <= 0) { /* Avoid 0->1 transition */
449 PROC_VMSPACE_UNLOCK(p);
452 } while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt + 1));
453 if (vm != p->p_vmspace) {
454 PROC_VMSPACE_UNLOCK(p);
458 PROC_VMSPACE_UNLOCK(p);
463 _vm_map_lock(vm_map_t map, const char *file, int line)
467 mtx_lock_flags_(&map->system_mtx, 0, file, line);
469 sx_xlock_(&map->lock, file, line);
474 vm_map_process_deferred(void)
477 vm_map_entry_t entry, next;
481 entry = td->td_map_def_user;
482 td->td_map_def_user = NULL;
483 while (entry != NULL) {
485 if ((entry->eflags & MAP_ENTRY_VN_WRITECNT) != 0) {
487 * Decrement the object's writemappings and
488 * possibly the vnode's v_writecount.
490 KASSERT((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0,
491 ("Submap with writecount"));
492 object = entry->object.vm_object;
493 KASSERT(object != NULL, ("No object for writecount"));
494 vnode_pager_release_writecount(object, entry->start,
497 vm_map_entry_deallocate(entry, FALSE);
503 _vm_map_unlock(vm_map_t map, const char *file, int line)
507 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
509 sx_xunlock_(&map->lock, file, line);
510 vm_map_process_deferred();
515 _vm_map_lock_read(vm_map_t map, const char *file, int line)
519 mtx_lock_flags_(&map->system_mtx, 0, file, line);
521 sx_slock_(&map->lock, file, line);
525 _vm_map_unlock_read(vm_map_t map, const char *file, int line)
529 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
531 sx_sunlock_(&map->lock, file, line);
532 vm_map_process_deferred();
537 _vm_map_trylock(vm_map_t map, const char *file, int line)
541 error = map->system_map ?
542 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) :
543 !sx_try_xlock_(&map->lock, file, line);
550 _vm_map_trylock_read(vm_map_t map, const char *file, int line)
554 error = map->system_map ?
555 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) :
556 !sx_try_slock_(&map->lock, file, line);
561 * _vm_map_lock_upgrade: [ internal use only ]
563 * Tries to upgrade a read (shared) lock on the specified map to a write
564 * (exclusive) lock. Returns the value "0" if the upgrade succeeds and a
565 * non-zero value if the upgrade fails. If the upgrade fails, the map is
566 * returned without a read or write lock held.
568 * Requires that the map be read locked.
571 _vm_map_lock_upgrade(vm_map_t map, const char *file, int line)
573 unsigned int last_timestamp;
575 if (map->system_map) {
576 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
578 if (!sx_try_upgrade_(&map->lock, file, line)) {
579 last_timestamp = map->timestamp;
580 sx_sunlock_(&map->lock, file, line);
581 vm_map_process_deferred();
583 * If the map's timestamp does not change while the
584 * map is unlocked, then the upgrade succeeds.
586 sx_xlock_(&map->lock, file, line);
587 if (last_timestamp != map->timestamp) {
588 sx_xunlock_(&map->lock, file, line);
598 _vm_map_lock_downgrade(vm_map_t map, const char *file, int line)
601 if (map->system_map) {
602 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
604 sx_downgrade_(&map->lock, file, line);
610 * Returns a non-zero value if the caller holds a write (exclusive) lock
611 * on the specified map and the value "0" otherwise.
614 vm_map_locked(vm_map_t map)
618 return (mtx_owned(&map->system_mtx));
620 return (sx_xlocked(&map->lock));
625 _vm_map_assert_locked(vm_map_t map, const char *file, int line)
629 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
631 sx_assert_(&map->lock, SA_XLOCKED, file, line);
634 #define VM_MAP_ASSERT_LOCKED(map) \
635 _vm_map_assert_locked(map, LOCK_FILE, LOCK_LINE)
637 #define VM_MAP_ASSERT_LOCKED(map)
641 * _vm_map_unlock_and_wait:
643 * Atomically releases the lock on the specified map and puts the calling
644 * thread to sleep. The calling thread will remain asleep until either
645 * vm_map_wakeup() is performed on the map or the specified timeout is
648 * WARNING! This function does not perform deferred deallocations of
649 * objects and map entries. Therefore, the calling thread is expected to
650 * reacquire the map lock after reawakening and later perform an ordinary
651 * unlock operation, such as vm_map_unlock(), before completing its
652 * operation on the map.
655 _vm_map_unlock_and_wait(vm_map_t map, int timo, const char *file, int line)
658 mtx_lock(&map_sleep_mtx);
660 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
662 sx_xunlock_(&map->lock, file, line);
663 return (msleep(&map->root, &map_sleep_mtx, PDROP | PVM, "vmmaps",
670 * Awaken any threads that have slept on the map using
671 * vm_map_unlock_and_wait().
674 vm_map_wakeup(vm_map_t map)
678 * Acquire and release map_sleep_mtx to prevent a wakeup()
679 * from being performed (and lost) between the map unlock
680 * and the msleep() in _vm_map_unlock_and_wait().
682 mtx_lock(&map_sleep_mtx);
683 mtx_unlock(&map_sleep_mtx);
688 vm_map_busy(vm_map_t map)
691 VM_MAP_ASSERT_LOCKED(map);
696 vm_map_unbusy(vm_map_t map)
699 VM_MAP_ASSERT_LOCKED(map);
700 KASSERT(map->busy, ("vm_map_unbusy: not busy"));
701 if (--map->busy == 0 && (map->flags & MAP_BUSY_WAKEUP)) {
702 vm_map_modflags(map, 0, MAP_BUSY_WAKEUP);
708 vm_map_wait_busy(vm_map_t map)
711 VM_MAP_ASSERT_LOCKED(map);
713 vm_map_modflags(map, MAP_BUSY_WAKEUP, 0);
715 msleep(&map->busy, &map->system_mtx, 0, "mbusy", 0);
717 sx_sleep(&map->busy, &map->lock, 0, "mbusy", 0);
723 vmspace_resident_count(struct vmspace *vmspace)
725 return pmap_resident_count(vmspace_pmap(vmspace));
731 * Creates and returns a new empty VM map with
732 * the given physical map structure, and having
733 * the given lower and upper address bounds.
736 vm_map_create(pmap_t pmap, vm_offset_t min, vm_offset_t max)
740 result = uma_zalloc(mapzone, M_WAITOK);
741 CTR1(KTR_VM, "vm_map_create: %p", result);
742 _vm_map_init(result, pmap, min, max);
747 * Initialize an existing vm_map structure
748 * such as that in the vmspace structure.
751 _vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max)
754 map->header.next = map->header.prev = &map->header;
755 map->needs_wakeup = FALSE;
758 map->min_offset = min;
759 map->max_offset = max;
767 vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max)
770 _vm_map_init(map, pmap, min, max);
771 mtx_init(&map->system_mtx, "system map", NULL, MTX_DEF | MTX_DUPOK);
772 sx_init(&map->lock, "user map");
776 * vm_map_entry_dispose: [ internal use only ]
778 * Inverse of vm_map_entry_create.
781 vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry)
783 uma_zfree(map->system_map ? kmapentzone : mapentzone, entry);
787 * vm_map_entry_create: [ internal use only ]
789 * Allocates a VM map entry for insertion.
790 * No entry fields are filled in.
792 static vm_map_entry_t
793 vm_map_entry_create(vm_map_t map)
795 vm_map_entry_t new_entry;
798 new_entry = uma_zalloc(kmapentzone, M_NOWAIT);
800 new_entry = uma_zalloc(mapentzone, M_WAITOK);
801 if (new_entry == NULL)
802 panic("vm_map_entry_create: kernel resources exhausted");
807 * vm_map_entry_set_behavior:
809 * Set the expected access behavior, either normal, random, or
813 vm_map_entry_set_behavior(vm_map_entry_t entry, u_char behavior)
815 entry->eflags = (entry->eflags & ~MAP_ENTRY_BEHAV_MASK) |
816 (behavior & MAP_ENTRY_BEHAV_MASK);
820 * vm_map_entry_set_max_free:
822 * Set the max_free field in a vm_map_entry.
825 vm_map_entry_set_max_free(vm_map_entry_t entry)
828 entry->max_free = entry->adj_free;
829 if (entry->left != NULL && entry->left->max_free > entry->max_free)
830 entry->max_free = entry->left->max_free;
831 if (entry->right != NULL && entry->right->max_free > entry->max_free)
832 entry->max_free = entry->right->max_free;
836 * vm_map_entry_splay:
838 * The Sleator and Tarjan top-down splay algorithm with the
839 * following variation. Max_free must be computed bottom-up, so
840 * on the downward pass, maintain the left and right spines in
841 * reverse order. Then, make a second pass up each side to fix
842 * the pointers and compute max_free. The time bound is O(log n)
845 * The new root is the vm_map_entry containing "addr", or else an
846 * adjacent entry (lower or higher) if addr is not in the tree.
848 * The map must be locked, and leaves it so.
850 * Returns: the new root.
852 static vm_map_entry_t
853 vm_map_entry_splay(vm_offset_t addr, vm_map_entry_t root)
855 vm_map_entry_t llist, rlist;
856 vm_map_entry_t ltree, rtree;
859 /* Special case of empty tree. */
864 * Pass One: Splay down the tree until we find addr or a NULL
865 * pointer where addr would go. llist and rlist are the two
866 * sides in reverse order (bottom-up), with llist linked by
867 * the right pointer and rlist linked by the left pointer in
868 * the vm_map_entry. Wait until Pass Two to set max_free on
874 /* root is never NULL in here. */
875 if (addr < root->start) {
879 if (addr < y->start && y->left != NULL) {
880 /* Rotate right and put y on rlist. */
881 root->left = y->right;
883 vm_map_entry_set_max_free(root);
888 /* Put root on rlist. */
893 } else if (addr >= root->end) {
897 if (addr >= y->end && y->right != NULL) {
898 /* Rotate left and put y on llist. */
899 root->right = y->left;
901 vm_map_entry_set_max_free(root);
906 /* Put root on llist. */
916 * Pass Two: Walk back up the two spines, flip the pointers
917 * and set max_free. The subtrees of the root go at the
918 * bottom of llist and rlist.
921 while (llist != NULL) {
923 llist->right = ltree;
924 vm_map_entry_set_max_free(llist);
929 while (rlist != NULL) {
932 vm_map_entry_set_max_free(rlist);
938 * Final assembly: add ltree and rtree as subtrees of root.
942 vm_map_entry_set_max_free(root);
948 * vm_map_entry_{un,}link:
950 * Insert/remove entries from maps.
953 vm_map_entry_link(vm_map_t map,
954 vm_map_entry_t after_where,
955 vm_map_entry_t entry)
959 "vm_map_entry_link: map %p, nentries %d, entry %p, after %p", map,
960 map->nentries, entry, after_where);
961 VM_MAP_ASSERT_LOCKED(map);
963 entry->prev = after_where;
964 entry->next = after_where->next;
965 entry->next->prev = entry;
966 after_where->next = entry;
968 if (after_where != &map->header) {
969 if (after_where != map->root)
970 vm_map_entry_splay(after_where->start, map->root);
971 entry->right = after_where->right;
972 entry->left = after_where;
973 after_where->right = NULL;
974 after_where->adj_free = entry->start - after_where->end;
975 vm_map_entry_set_max_free(after_where);
977 entry->right = map->root;
980 entry->adj_free = (entry->next == &map->header ? map->max_offset :
981 entry->next->start) - entry->end;
982 vm_map_entry_set_max_free(entry);
987 vm_map_entry_unlink(vm_map_t map,
988 vm_map_entry_t entry)
990 vm_map_entry_t next, prev, root;
992 VM_MAP_ASSERT_LOCKED(map);
993 if (entry != map->root)
994 vm_map_entry_splay(entry->start, map->root);
995 if (entry->left == NULL)
998 root = vm_map_entry_splay(entry->start, entry->left);
999 root->right = entry->right;
1000 root->adj_free = (entry->next == &map->header ? map->max_offset :
1001 entry->next->start) - root->end;
1002 vm_map_entry_set_max_free(root);
1011 CTR3(KTR_VM, "vm_map_entry_unlink: map %p, nentries %d, entry %p", map,
1012 map->nentries, entry);
1016 * vm_map_entry_resize_free:
1018 * Recompute the amount of free space following a vm_map_entry
1019 * and propagate that value up the tree. Call this function after
1020 * resizing a map entry in-place, that is, without a call to
1021 * vm_map_entry_link() or _unlink().
1023 * The map must be locked, and leaves it so.
1026 vm_map_entry_resize_free(vm_map_t map, vm_map_entry_t entry)
1030 * Using splay trees without parent pointers, propagating
1031 * max_free up the tree is done by moving the entry to the
1032 * root and making the change there.
1034 if (entry != map->root)
1035 map->root = vm_map_entry_splay(entry->start, map->root);
1037 entry->adj_free = (entry->next == &map->header ? map->max_offset :
1038 entry->next->start) - entry->end;
1039 vm_map_entry_set_max_free(entry);
1043 * vm_map_lookup_entry: [ internal use only ]
1045 * Finds the map entry containing (or
1046 * immediately preceding) the specified address
1047 * in the given map; the entry is returned
1048 * in the "entry" parameter. The boolean
1049 * result indicates whether the address is
1050 * actually contained in the map.
1053 vm_map_lookup_entry(
1055 vm_offset_t address,
1056 vm_map_entry_t *entry) /* OUT */
1062 * If the map is empty, then the map entry immediately preceding
1063 * "address" is the map's header.
1067 *entry = &map->header;
1068 else if (address >= cur->start && cur->end > address) {
1071 } else if ((locked = vm_map_locked(map)) ||
1072 sx_try_upgrade(&map->lock)) {
1074 * Splay requires a write lock on the map. However, it only
1075 * restructures the binary search tree; it does not otherwise
1076 * change the map. Thus, the map's timestamp need not change
1077 * on a temporary upgrade.
1079 map->root = cur = vm_map_entry_splay(address, cur);
1081 sx_downgrade(&map->lock);
1084 * If "address" is contained within a map entry, the new root
1085 * is that map entry. Otherwise, the new root is a map entry
1086 * immediately before or after "address".
1088 if (address >= cur->start) {
1090 if (cur->end > address)
1096 * Since the map is only locked for read access, perform a
1097 * standard binary search tree lookup for "address".
1100 if (address < cur->start) {
1101 if (cur->left == NULL) {
1106 } else if (cur->end > address) {
1110 if (cur->right == NULL) {
1123 * Inserts the given whole VM object into the target
1124 * map at the specified address range. The object's
1125 * size should match that of the address range.
1127 * Requires that the map be locked, and leaves it so.
1129 * If object is non-NULL, ref count must be bumped by caller
1130 * prior to making call to account for the new entry.
1133 vm_map_insert(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1134 vm_offset_t start, vm_offset_t end, vm_prot_t prot, vm_prot_t max,
1137 vm_map_entry_t new_entry;
1138 vm_map_entry_t prev_entry;
1139 vm_map_entry_t temp_entry;
1140 vm_eflags_t protoeflags;
1142 vm_inherit_t inheritance;
1143 boolean_t charge_prev_obj;
1145 VM_MAP_ASSERT_LOCKED(map);
1148 * Check that the start and end points are not bogus.
1150 if ((start < map->min_offset) || (end > map->max_offset) ||
1152 return (KERN_INVALID_ADDRESS);
1155 * Find the entry prior to the proposed starting address; if it's part
1156 * of an existing entry, this range is bogus.
1158 if (vm_map_lookup_entry(map, start, &temp_entry))
1159 return (KERN_NO_SPACE);
1161 prev_entry = temp_entry;
1164 * Assert that the next entry doesn't overlap the end point.
1166 if ((prev_entry->next != &map->header) &&
1167 (prev_entry->next->start < end))
1168 return (KERN_NO_SPACE);
1171 charge_prev_obj = FALSE;
1173 if (cow & MAP_COPY_ON_WRITE)
1174 protoeflags |= MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY;
1176 if (cow & MAP_NOFAULT) {
1177 protoeflags |= MAP_ENTRY_NOFAULT;
1179 KASSERT(object == NULL,
1180 ("vm_map_insert: paradoxical MAP_NOFAULT request"));
1182 if (cow & MAP_DISABLE_SYNCER)
1183 protoeflags |= MAP_ENTRY_NOSYNC;
1184 if (cow & MAP_DISABLE_COREDUMP)
1185 protoeflags |= MAP_ENTRY_NOCOREDUMP;
1186 if (cow & MAP_VN_WRITECOUNT)
1187 protoeflags |= MAP_ENTRY_VN_WRITECNT;
1188 if (cow & MAP_INHERIT_SHARE)
1189 inheritance = VM_INHERIT_SHARE;
1191 inheritance = VM_INHERIT_DEFAULT;
1194 KASSERT((object != kmem_object && object != kernel_object) ||
1195 ((object == kmem_object || object == kernel_object) &&
1196 !(protoeflags & MAP_ENTRY_NEEDS_COPY)),
1197 ("kmem or kernel object and cow"));
1198 if (cow & (MAP_ACC_NO_CHARGE | MAP_NOFAULT))
1200 if ((cow & MAP_ACC_CHARGED) || ((prot & VM_PROT_WRITE) &&
1201 ((protoeflags & MAP_ENTRY_NEEDS_COPY) || object == NULL))) {
1202 if (!(cow & MAP_ACC_CHARGED) && !swap_reserve(end - start))
1203 return (KERN_RESOURCE_SHORTAGE);
1204 KASSERT(object == NULL || (protoeflags & MAP_ENTRY_NEEDS_COPY) ||
1205 object->cred == NULL,
1206 ("OVERCOMMIT: vm_map_insert o %p", object));
1207 cred = curthread->td_ucred;
1209 if (object == NULL && !(protoeflags & MAP_ENTRY_NEEDS_COPY))
1210 charge_prev_obj = TRUE;
1214 /* Expand the kernel pmap, if necessary. */
1215 if (map == kernel_map && end > kernel_vm_end)
1216 pmap_growkernel(end);
1217 if (object != NULL) {
1219 * OBJ_ONEMAPPING must be cleared unless this mapping
1220 * is trivially proven to be the only mapping for any
1221 * of the object's pages. (Object granularity
1222 * reference counting is insufficient to recognize
1223 * aliases with precision.)
1225 VM_OBJECT_WLOCK(object);
1226 if (object->ref_count > 1 || object->shadow_count != 0)
1227 vm_object_clear_flag(object, OBJ_ONEMAPPING);
1228 VM_OBJECT_WUNLOCK(object);
1230 else if ((prev_entry != &map->header) &&
1231 (prev_entry->eflags == protoeflags) &&
1232 (prev_entry->end == start) &&
1233 (prev_entry->wired_count == 0) &&
1234 (prev_entry->cred == cred ||
1235 (prev_entry->object.vm_object != NULL &&
1236 (prev_entry->object.vm_object->cred == cred))) &&
1237 vm_object_coalesce(prev_entry->object.vm_object,
1239 (vm_size_t)(prev_entry->end - prev_entry->start),
1240 (vm_size_t)(end - prev_entry->end), charge_prev_obj)) {
1242 * We were able to extend the object. Determine if we
1243 * can extend the previous map entry to include the
1244 * new range as well.
1246 if ((prev_entry->inheritance == inheritance) &&
1247 (prev_entry->protection == prot) &&
1248 (prev_entry->max_protection == max)) {
1249 map->size += (end - prev_entry->end);
1250 prev_entry->end = end;
1251 vm_map_entry_resize_free(map, prev_entry);
1252 vm_map_simplify_entry(map, prev_entry);
1255 return (KERN_SUCCESS);
1259 * If we can extend the object but cannot extend the
1260 * map entry, we have to create a new map entry. We
1261 * must bump the ref count on the extended object to
1262 * account for it. object may be NULL.
1264 object = prev_entry->object.vm_object;
1265 offset = prev_entry->offset +
1266 (prev_entry->end - prev_entry->start);
1267 vm_object_reference(object);
1268 if (cred != NULL && object != NULL && object->cred != NULL &&
1269 !(prev_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
1270 /* Object already accounts for this uid. */
1277 * NOTE: if conditionals fail, object can be NULL here. This occurs
1278 * in things like the buffer map where we manage kva but do not manage
1283 * Create a new entry
1285 new_entry = vm_map_entry_create(map);
1286 new_entry->start = start;
1287 new_entry->end = end;
1288 new_entry->cred = NULL;
1290 new_entry->eflags = protoeflags;
1291 new_entry->object.vm_object = object;
1292 new_entry->offset = offset;
1293 new_entry->avail_ssize = 0;
1295 new_entry->inheritance = inheritance;
1296 new_entry->protection = prot;
1297 new_entry->max_protection = max;
1298 new_entry->wired_count = 0;
1299 new_entry->read_ahead = VM_FAULT_READ_AHEAD_INIT;
1300 new_entry->next_read = OFF_TO_IDX(offset);
1302 KASSERT(cred == NULL || !ENTRY_CHARGED(new_entry),
1303 ("OVERCOMMIT: vm_map_insert leaks vm_map %p", new_entry));
1304 new_entry->cred = cred;
1307 * Insert the new entry into the list
1309 vm_map_entry_link(map, prev_entry, new_entry);
1310 map->size += new_entry->end - new_entry->start;
1313 * It may be possible to merge the new entry with the next and/or
1314 * previous entries. However, due to MAP_STACK_* being a hack, a
1315 * panic can result from merging such entries.
1317 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0)
1318 vm_map_simplify_entry(map, new_entry);
1320 if (cow & (MAP_PREFAULT|MAP_PREFAULT_PARTIAL)) {
1321 vm_map_pmap_enter(map, start, prot,
1322 object, OFF_TO_IDX(offset), end - start,
1323 cow & MAP_PREFAULT_PARTIAL);
1326 return (KERN_SUCCESS);
1332 * Find the first fit (lowest VM address) for "length" free bytes
1333 * beginning at address >= start in the given map.
1335 * In a vm_map_entry, "adj_free" is the amount of free space
1336 * adjacent (higher address) to this entry, and "max_free" is the
1337 * maximum amount of contiguous free space in its subtree. This
1338 * allows finding a free region in one path down the tree, so
1339 * O(log n) amortized with splay trees.
1341 * The map must be locked, and leaves it so.
1343 * Returns: 0 on success, and starting address in *addr,
1344 * 1 if insufficient space.
1347 vm_map_findspace(vm_map_t map, vm_offset_t start, vm_size_t length,
1348 vm_offset_t *addr) /* OUT */
1350 vm_map_entry_t entry;
1354 * Request must fit within min/max VM address and must avoid
1357 if (start < map->min_offset)
1358 start = map->min_offset;
1359 if (start + length > map->max_offset || start + length < start)
1362 /* Empty tree means wide open address space. */
1363 if (map->root == NULL) {
1369 * After splay, if start comes before root node, then there
1370 * must be a gap from start to the root.
1372 map->root = vm_map_entry_splay(start, map->root);
1373 if (start + length <= map->root->start) {
1379 * Root is the last node that might begin its gap before
1380 * start, and this is the last comparison where address
1381 * wrap might be a problem.
1383 st = (start > map->root->end) ? start : map->root->end;
1384 if (length <= map->root->end + map->root->adj_free - st) {
1389 /* With max_free, can immediately tell if no solution. */
1390 entry = map->root->right;
1391 if (entry == NULL || length > entry->max_free)
1395 * Search the right subtree in the order: left subtree, root,
1396 * right subtree (first fit). The previous splay implies that
1397 * all regions in the right subtree have addresses > start.
1399 while (entry != NULL) {
1400 if (entry->left != NULL && entry->left->max_free >= length)
1401 entry = entry->left;
1402 else if (entry->adj_free >= length) {
1406 entry = entry->right;
1409 /* Can't get here, so panic if we do. */
1410 panic("vm_map_findspace: max_free corrupt");
1414 vm_map_fixed(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1415 vm_offset_t start, vm_size_t length, vm_prot_t prot,
1416 vm_prot_t max, int cow)
1421 end = start + length;
1423 VM_MAP_RANGE_CHECK(map, start, end);
1424 (void) vm_map_delete(map, start, end);
1425 result = vm_map_insert(map, object, offset, start, end, prot,
1432 * vm_map_find finds an unallocated region in the target address
1433 * map with the given length. The search is defined to be
1434 * first-fit from the specified address; the region found is
1435 * returned in the same parameter.
1437 * If object is non-NULL, ref count must be bumped by caller
1438 * prior to making call to account for the new entry.
1441 vm_map_find(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1442 vm_offset_t *addr, /* IN/OUT */
1443 vm_size_t length, int find_space, vm_prot_t prot,
1444 vm_prot_t max, int cow)
1446 vm_offset_t start, initial_addr;
1449 if (find_space == VMFS_OPTIMAL_SPACE && (object == NULL ||
1450 (object->flags & OBJ_COLORED) == 0))
1451 find_space = VMFS_ANY_SPACE;
1452 initial_addr = *addr;
1454 start = initial_addr;
1457 if (find_space != VMFS_NO_SPACE) {
1458 if (vm_map_findspace(map, start, length, addr)) {
1460 if (find_space == VMFS_OPTIMAL_SPACE) {
1461 find_space = VMFS_ANY_SPACE;
1464 return (KERN_NO_SPACE);
1466 switch (find_space) {
1467 case VMFS_ALIGNED_SPACE:
1468 case VMFS_OPTIMAL_SPACE:
1469 pmap_align_superpage(object, offset, addr,
1472 #ifdef VMFS_TLB_ALIGNED_SPACE
1473 case VMFS_TLB_ALIGNED_SPACE:
1474 pmap_align_tlb(addr);
1483 result = vm_map_insert(map, object, offset, start, start +
1484 length, prot, max, cow);
1485 } while (result == KERN_NO_SPACE && (find_space == VMFS_ALIGNED_SPACE ||
1486 #ifdef VMFS_TLB_ALIGNED_SPACE
1487 find_space == VMFS_TLB_ALIGNED_SPACE ||
1489 find_space == VMFS_OPTIMAL_SPACE));
1495 * vm_map_simplify_entry:
1497 * Simplify the given map entry by merging with either neighbor. This
1498 * routine also has the ability to merge with both neighbors.
1500 * The map must be locked.
1502 * This routine guarentees that the passed entry remains valid (though
1503 * possibly extended). When merging, this routine may delete one or
1507 vm_map_simplify_entry(vm_map_t map, vm_map_entry_t entry)
1509 vm_map_entry_t next, prev;
1510 vm_size_t prevsize, esize;
1512 if (entry->eflags & (MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_IS_SUB_MAP))
1516 if (prev != &map->header) {
1517 prevsize = prev->end - prev->start;
1518 if ( (prev->end == entry->start) &&
1519 (prev->object.vm_object == entry->object.vm_object) &&
1520 (!prev->object.vm_object ||
1521 (prev->offset + prevsize == entry->offset)) &&
1522 (prev->eflags == entry->eflags) &&
1523 (prev->protection == entry->protection) &&
1524 (prev->max_protection == entry->max_protection) &&
1525 (prev->inheritance == entry->inheritance) &&
1526 (prev->wired_count == entry->wired_count) &&
1527 (prev->cred == entry->cred)) {
1528 vm_map_entry_unlink(map, prev);
1529 entry->start = prev->start;
1530 entry->offset = prev->offset;
1531 if (entry->prev != &map->header)
1532 vm_map_entry_resize_free(map, entry->prev);
1535 * If the backing object is a vnode object,
1536 * vm_object_deallocate() calls vrele().
1537 * However, vrele() does not lock the vnode
1538 * because the vnode has additional
1539 * references. Thus, the map lock can be kept
1540 * without causing a lock-order reversal with
1543 * Since we count the number of virtual page
1544 * mappings in object->un_pager.vnp.writemappings,
1545 * the writemappings value should not be adjusted
1546 * when the entry is disposed of.
1548 if (prev->object.vm_object)
1549 vm_object_deallocate(prev->object.vm_object);
1550 if (prev->cred != NULL)
1552 vm_map_entry_dispose(map, prev);
1557 if (next != &map->header) {
1558 esize = entry->end - entry->start;
1559 if ((entry->end == next->start) &&
1560 (next->object.vm_object == entry->object.vm_object) &&
1561 (!entry->object.vm_object ||
1562 (entry->offset + esize == next->offset)) &&
1563 (next->eflags == entry->eflags) &&
1564 (next->protection == entry->protection) &&
1565 (next->max_protection == entry->max_protection) &&
1566 (next->inheritance == entry->inheritance) &&
1567 (next->wired_count == entry->wired_count) &&
1568 (next->cred == entry->cred)) {
1569 vm_map_entry_unlink(map, next);
1570 entry->end = next->end;
1571 vm_map_entry_resize_free(map, entry);
1574 * See comment above.
1576 if (next->object.vm_object)
1577 vm_object_deallocate(next->object.vm_object);
1578 if (next->cred != NULL)
1580 vm_map_entry_dispose(map, next);
1585 * vm_map_clip_start: [ internal use only ]
1587 * Asserts that the given entry begins at or after
1588 * the specified address; if necessary,
1589 * it splits the entry into two.
1591 #define vm_map_clip_start(map, entry, startaddr) \
1593 if (startaddr > entry->start) \
1594 _vm_map_clip_start(map, entry, startaddr); \
1598 * This routine is called only when it is known that
1599 * the entry must be split.
1602 _vm_map_clip_start(vm_map_t map, vm_map_entry_t entry, vm_offset_t start)
1604 vm_map_entry_t new_entry;
1606 VM_MAP_ASSERT_LOCKED(map);
1609 * Split off the front portion -- note that we must insert the new
1610 * entry BEFORE this one, so that this entry has the specified
1613 vm_map_simplify_entry(map, entry);
1616 * If there is no object backing this entry, we might as well create
1617 * one now. If we defer it, an object can get created after the map
1618 * is clipped, and individual objects will be created for the split-up
1619 * map. This is a bit of a hack, but is also about the best place to
1620 * put this improvement.
1622 if (entry->object.vm_object == NULL && !map->system_map) {
1624 object = vm_object_allocate(OBJT_DEFAULT,
1625 atop(entry->end - entry->start));
1626 entry->object.vm_object = object;
1628 if (entry->cred != NULL) {
1629 object->cred = entry->cred;
1630 object->charge = entry->end - entry->start;
1633 } else if (entry->object.vm_object != NULL &&
1634 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) &&
1635 entry->cred != NULL) {
1636 VM_OBJECT_WLOCK(entry->object.vm_object);
1637 KASSERT(entry->object.vm_object->cred == NULL,
1638 ("OVERCOMMIT: vm_entry_clip_start: both cred e %p", entry));
1639 entry->object.vm_object->cred = entry->cred;
1640 entry->object.vm_object->charge = entry->end - entry->start;
1641 VM_OBJECT_WUNLOCK(entry->object.vm_object);
1645 new_entry = vm_map_entry_create(map);
1646 *new_entry = *entry;
1648 new_entry->end = start;
1649 entry->offset += (start - entry->start);
1650 entry->start = start;
1651 if (new_entry->cred != NULL)
1652 crhold(entry->cred);
1654 vm_map_entry_link(map, entry->prev, new_entry);
1656 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1657 vm_object_reference(new_entry->object.vm_object);
1659 * The object->un_pager.vnp.writemappings for the
1660 * object of MAP_ENTRY_VN_WRITECNT type entry shall be
1661 * kept as is here. The virtual pages are
1662 * re-distributed among the clipped entries, so the sum is
1669 * vm_map_clip_end: [ internal use only ]
1671 * Asserts that the given entry ends at or before
1672 * the specified address; if necessary,
1673 * it splits the entry into two.
1675 #define vm_map_clip_end(map, entry, endaddr) \
1677 if ((endaddr) < (entry->end)) \
1678 _vm_map_clip_end((map), (entry), (endaddr)); \
1682 * This routine is called only when it is known that
1683 * the entry must be split.
1686 _vm_map_clip_end(vm_map_t map, vm_map_entry_t entry, vm_offset_t end)
1688 vm_map_entry_t new_entry;
1690 VM_MAP_ASSERT_LOCKED(map);
1693 * If there is no object backing this entry, we might as well create
1694 * one now. If we defer it, an object can get created after the map
1695 * is clipped, and individual objects will be created for the split-up
1696 * map. This is a bit of a hack, but is also about the best place to
1697 * put this improvement.
1699 if (entry->object.vm_object == NULL && !map->system_map) {
1701 object = vm_object_allocate(OBJT_DEFAULT,
1702 atop(entry->end - entry->start));
1703 entry->object.vm_object = object;
1705 if (entry->cred != NULL) {
1706 object->cred = entry->cred;
1707 object->charge = entry->end - entry->start;
1710 } else if (entry->object.vm_object != NULL &&
1711 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) &&
1712 entry->cred != NULL) {
1713 VM_OBJECT_WLOCK(entry->object.vm_object);
1714 KASSERT(entry->object.vm_object->cred == NULL,
1715 ("OVERCOMMIT: vm_entry_clip_end: both cred e %p", entry));
1716 entry->object.vm_object->cred = entry->cred;
1717 entry->object.vm_object->charge = entry->end - entry->start;
1718 VM_OBJECT_WUNLOCK(entry->object.vm_object);
1723 * Create a new entry and insert it AFTER the specified entry
1725 new_entry = vm_map_entry_create(map);
1726 *new_entry = *entry;
1728 new_entry->start = entry->end = end;
1729 new_entry->offset += (end - entry->start);
1730 if (new_entry->cred != NULL)
1731 crhold(entry->cred);
1733 vm_map_entry_link(map, entry, new_entry);
1735 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1736 vm_object_reference(new_entry->object.vm_object);
1741 * vm_map_submap: [ kernel use only ]
1743 * Mark the given range as handled by a subordinate map.
1745 * This range must have been created with vm_map_find,
1746 * and no other operations may have been performed on this
1747 * range prior to calling vm_map_submap.
1749 * Only a limited number of operations can be performed
1750 * within this rage after calling vm_map_submap:
1752 * [Don't try vm_map_copy!]
1754 * To remove a submapping, one must first remove the
1755 * range from the superior map, and then destroy the
1756 * submap (if desired). [Better yet, don't try it.]
1765 vm_map_entry_t entry;
1766 int result = KERN_INVALID_ARGUMENT;
1770 VM_MAP_RANGE_CHECK(map, start, end);
1772 if (vm_map_lookup_entry(map, start, &entry)) {
1773 vm_map_clip_start(map, entry, start);
1775 entry = entry->next;
1777 vm_map_clip_end(map, entry, end);
1779 if ((entry->start == start) && (entry->end == end) &&
1780 ((entry->eflags & MAP_ENTRY_COW) == 0) &&
1781 (entry->object.vm_object == NULL)) {
1782 entry->object.sub_map = submap;
1783 entry->eflags |= MAP_ENTRY_IS_SUB_MAP;
1784 result = KERN_SUCCESS;
1792 * The maximum number of pages to map
1794 #define MAX_INIT_PT 96
1797 * vm_map_pmap_enter:
1799 * Preload read-only mappings for the specified object's resident pages
1800 * into the target map. If "flags" is MAP_PREFAULT_PARTIAL, then only
1801 * the resident pages within the address range [addr, addr + ulmin(size,
1802 * ptoa(MAX_INIT_PT))) are mapped. Otherwise, all resident pages within
1803 * the specified address range are mapped. This eliminates many soft
1804 * faults on process startup and immediately after an mmap(2). Because
1805 * these are speculative mappings, cached pages are not reactivated and
1809 vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot,
1810 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags)
1813 vm_page_t p, p_start;
1814 vm_pindex_t psize, tmpidx;
1816 if ((prot & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0 || object == NULL)
1818 VM_OBJECT_RLOCK(object);
1819 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) {
1820 VM_OBJECT_RUNLOCK(object);
1821 VM_OBJECT_WLOCK(object);
1822 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) {
1823 pmap_object_init_pt(map->pmap, addr, object, pindex,
1825 VM_OBJECT_WUNLOCK(object);
1828 VM_OBJECT_LOCK_DOWNGRADE(object);
1832 if (psize > MAX_INIT_PT && (flags & MAP_PREFAULT_PARTIAL) != 0)
1833 psize = MAX_INIT_PT;
1834 if (psize + pindex > object->size) {
1835 if (object->size < pindex) {
1836 VM_OBJECT_RUNLOCK(object);
1839 psize = object->size - pindex;
1845 p = vm_page_find_least(object, pindex);
1847 * Assert: the variable p is either (1) the page with the
1848 * least pindex greater than or equal to the parameter pindex
1852 p != NULL && (tmpidx = p->pindex - pindex) < psize;
1853 p = TAILQ_NEXT(p, listq)) {
1855 * don't allow an madvise to blow away our really
1856 * free pages allocating pv entries.
1858 if ((flags & MAP_PREFAULT_MADVISE) &&
1859 cnt.v_free_count < cnt.v_free_reserved) {
1863 if (p->valid == VM_PAGE_BITS_ALL) {
1864 if (p_start == NULL) {
1865 start = addr + ptoa(tmpidx);
1868 } else if (p_start != NULL) {
1869 pmap_enter_object(map->pmap, start, addr +
1870 ptoa(tmpidx), p_start, prot);
1874 if (p_start != NULL)
1875 pmap_enter_object(map->pmap, start, addr + ptoa(psize),
1877 VM_OBJECT_RUNLOCK(object);
1883 * Sets the protection of the specified address
1884 * region in the target map. If "set_max" is
1885 * specified, the maximum protection is to be set;
1886 * otherwise, only the current protection is affected.
1889 vm_map_protect(vm_map_t map, vm_offset_t start, vm_offset_t end,
1890 vm_prot_t new_prot, boolean_t set_max)
1892 vm_map_entry_t current, entry;
1899 VM_MAP_RANGE_CHECK(map, start, end);
1901 if (vm_map_lookup_entry(map, start, &entry)) {
1902 vm_map_clip_start(map, entry, start);
1904 entry = entry->next;
1908 * Make a first pass to check for protection violations.
1911 while ((current != &map->header) && (current->start < end)) {
1912 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
1914 return (KERN_INVALID_ARGUMENT);
1916 if ((new_prot & current->max_protection) != new_prot) {
1918 return (KERN_PROTECTION_FAILURE);
1920 current = current->next;
1925 * Do an accounting pass for private read-only mappings that
1926 * now will do cow due to allowed write (e.g. debugger sets
1927 * breakpoint on text segment)
1929 for (current = entry; (current != &map->header) &&
1930 (current->start < end); current = current->next) {
1932 vm_map_clip_end(map, current, end);
1935 ((new_prot & ~(current->protection)) & VM_PROT_WRITE) == 0 ||
1936 ENTRY_CHARGED(current)) {
1940 cred = curthread->td_ucred;
1941 obj = current->object.vm_object;
1943 if (obj == NULL || (current->eflags & MAP_ENTRY_NEEDS_COPY)) {
1944 if (!swap_reserve(current->end - current->start)) {
1946 return (KERN_RESOURCE_SHORTAGE);
1949 current->cred = cred;
1953 VM_OBJECT_WLOCK(obj);
1954 if (obj->type != OBJT_DEFAULT && obj->type != OBJT_SWAP) {
1955 VM_OBJECT_WUNLOCK(obj);
1960 * Charge for the whole object allocation now, since
1961 * we cannot distinguish between non-charged and
1962 * charged clipped mapping of the same object later.
1964 KASSERT(obj->charge == 0,
1965 ("vm_map_protect: object %p overcharged\n", obj));
1966 if (!swap_reserve(ptoa(obj->size))) {
1967 VM_OBJECT_WUNLOCK(obj);
1969 return (KERN_RESOURCE_SHORTAGE);
1974 obj->charge = ptoa(obj->size);
1975 VM_OBJECT_WUNLOCK(obj);
1979 * Go back and fix up protections. [Note that clipping is not
1980 * necessary the second time.]
1983 while ((current != &map->header) && (current->start < end)) {
1984 old_prot = current->protection;
1987 current->protection =
1988 (current->max_protection = new_prot) &
1991 current->protection = new_prot;
1993 if ((current->eflags & (MAP_ENTRY_COW | MAP_ENTRY_USER_WIRED))
1994 == (MAP_ENTRY_COW | MAP_ENTRY_USER_WIRED) &&
1995 (current->protection & VM_PROT_WRITE) != 0 &&
1996 (old_prot & VM_PROT_WRITE) == 0) {
1997 vm_fault_copy_entry(map, map, current, current, NULL);
2001 * When restricting access, update the physical map. Worry
2002 * about copy-on-write here.
2004 if ((old_prot & ~current->protection) != 0) {
2005 #define MASK(entry) (((entry)->eflags & MAP_ENTRY_COW) ? ~VM_PROT_WRITE : \
2007 pmap_protect(map->pmap, current->start,
2009 current->protection & MASK(current));
2012 vm_map_simplify_entry(map, current);
2013 current = current->next;
2016 return (KERN_SUCCESS);
2022 * This routine traverses a processes map handling the madvise
2023 * system call. Advisories are classified as either those effecting
2024 * the vm_map_entry structure, or those effecting the underlying
2034 vm_map_entry_t current, entry;
2038 * Some madvise calls directly modify the vm_map_entry, in which case
2039 * we need to use an exclusive lock on the map and we need to perform
2040 * various clipping operations. Otherwise we only need a read-lock
2045 case MADV_SEQUENTIAL:
2057 vm_map_lock_read(map);
2060 return (KERN_INVALID_ARGUMENT);
2064 * Locate starting entry and clip if necessary.
2066 VM_MAP_RANGE_CHECK(map, start, end);
2068 if (vm_map_lookup_entry(map, start, &entry)) {
2070 vm_map_clip_start(map, entry, start);
2072 entry = entry->next;
2077 * madvise behaviors that are implemented in the vm_map_entry.
2079 * We clip the vm_map_entry so that behavioral changes are
2080 * limited to the specified address range.
2082 for (current = entry;
2083 (current != &map->header) && (current->start < end);
2084 current = current->next
2086 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
2089 vm_map_clip_end(map, current, end);
2093 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_NORMAL);
2095 case MADV_SEQUENTIAL:
2096 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_SEQUENTIAL);
2099 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_RANDOM);
2102 current->eflags |= MAP_ENTRY_NOSYNC;
2105 current->eflags &= ~MAP_ENTRY_NOSYNC;
2108 current->eflags |= MAP_ENTRY_NOCOREDUMP;
2111 current->eflags &= ~MAP_ENTRY_NOCOREDUMP;
2116 vm_map_simplify_entry(map, current);
2120 vm_pindex_t pstart, pend;
2123 * madvise behaviors that are implemented in the underlying
2126 * Since we don't clip the vm_map_entry, we have to clip
2127 * the vm_object pindex and count.
2129 for (current = entry;
2130 (current != &map->header) && (current->start < end);
2131 current = current->next
2133 vm_offset_t useStart;
2135 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
2138 pstart = OFF_TO_IDX(current->offset);
2139 pend = pstart + atop(current->end - current->start);
2140 useStart = current->start;
2142 if (current->start < start) {
2143 pstart += atop(start - current->start);
2146 if (current->end > end)
2147 pend -= atop(current->end - end);
2152 vm_object_madvise(current->object.vm_object, pstart,
2154 if (behav == MADV_WILLNEED) {
2155 vm_map_pmap_enter(map,
2157 current->protection,
2158 current->object.vm_object,
2160 ptoa(pend - pstart),
2161 MAP_PREFAULT_MADVISE
2165 vm_map_unlock_read(map);
2174 * Sets the inheritance of the specified address
2175 * range in the target map. Inheritance
2176 * affects how the map will be shared with
2177 * child maps at the time of vmspace_fork.
2180 vm_map_inherit(vm_map_t map, vm_offset_t start, vm_offset_t end,
2181 vm_inherit_t new_inheritance)
2183 vm_map_entry_t entry;
2184 vm_map_entry_t temp_entry;
2186 switch (new_inheritance) {
2187 case VM_INHERIT_NONE:
2188 case VM_INHERIT_COPY:
2189 case VM_INHERIT_SHARE:
2192 return (KERN_INVALID_ARGUMENT);
2195 VM_MAP_RANGE_CHECK(map, start, end);
2196 if (vm_map_lookup_entry(map, start, &temp_entry)) {
2198 vm_map_clip_start(map, entry, start);
2200 entry = temp_entry->next;
2201 while ((entry != &map->header) && (entry->start < end)) {
2202 vm_map_clip_end(map, entry, end);
2203 entry->inheritance = new_inheritance;
2204 vm_map_simplify_entry(map, entry);
2205 entry = entry->next;
2208 return (KERN_SUCCESS);
2214 * Implements both kernel and user unwiring.
2217 vm_map_unwire(vm_map_t map, vm_offset_t start, vm_offset_t end,
2220 vm_map_entry_t entry, first_entry, tmp_entry;
2221 vm_offset_t saved_start;
2222 unsigned int last_timestamp;
2224 boolean_t need_wakeup, result, user_unwire;
2226 user_unwire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE;
2228 VM_MAP_RANGE_CHECK(map, start, end);
2229 if (!vm_map_lookup_entry(map, start, &first_entry)) {
2230 if (flags & VM_MAP_WIRE_HOLESOK)
2231 first_entry = first_entry->next;
2234 return (KERN_INVALID_ADDRESS);
2237 last_timestamp = map->timestamp;
2238 entry = first_entry;
2239 while (entry != &map->header && entry->start < end) {
2240 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
2242 * We have not yet clipped the entry.
2244 saved_start = (start >= entry->start) ? start :
2246 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2247 if (vm_map_unlock_and_wait(map, 0)) {
2249 * Allow interruption of user unwiring?
2253 if (last_timestamp+1 != map->timestamp) {
2255 * Look again for the entry because the map was
2256 * modified while it was unlocked.
2257 * Specifically, the entry may have been
2258 * clipped, merged, or deleted.
2260 if (!vm_map_lookup_entry(map, saved_start,
2262 if (flags & VM_MAP_WIRE_HOLESOK)
2263 tmp_entry = tmp_entry->next;
2265 if (saved_start == start) {
2267 * First_entry has been deleted.
2270 return (KERN_INVALID_ADDRESS);
2273 rv = KERN_INVALID_ADDRESS;
2277 if (entry == first_entry)
2278 first_entry = tmp_entry;
2283 last_timestamp = map->timestamp;
2286 vm_map_clip_start(map, entry, start);
2287 vm_map_clip_end(map, entry, end);
2289 * Mark the entry in case the map lock is released. (See
2292 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
2293 entry->wiring_thread = curthread;
2295 * Check the map for holes in the specified region.
2296 * If VM_MAP_WIRE_HOLESOK was specified, skip this check.
2298 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) &&
2299 (entry->end < end && (entry->next == &map->header ||
2300 entry->next->start > entry->end))) {
2302 rv = KERN_INVALID_ADDRESS;
2306 * If system unwiring, require that the entry is system wired.
2309 vm_map_entry_system_wired_count(entry) == 0) {
2311 rv = KERN_INVALID_ARGUMENT;
2314 entry = entry->next;
2318 need_wakeup = FALSE;
2319 if (first_entry == NULL) {
2320 result = vm_map_lookup_entry(map, start, &first_entry);
2321 if (!result && (flags & VM_MAP_WIRE_HOLESOK))
2322 first_entry = first_entry->next;
2324 KASSERT(result, ("vm_map_unwire: lookup failed"));
2326 for (entry = first_entry; entry != &map->header && entry->start < end;
2327 entry = entry->next) {
2329 * If VM_MAP_WIRE_HOLESOK was specified, an empty
2330 * space in the unwired region could have been mapped
2331 * while the map lock was dropped for draining
2332 * MAP_ENTRY_IN_TRANSITION. Moreover, another thread
2333 * could be simultaneously wiring this new mapping
2334 * entry. Detect these cases and skip any entries
2335 * marked as in transition by us.
2337 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 ||
2338 entry->wiring_thread != curthread) {
2339 KASSERT((flags & VM_MAP_WIRE_HOLESOK) != 0,
2340 ("vm_map_unwire: !HOLESOK and new/changed entry"));
2344 if (rv == KERN_SUCCESS && (!user_unwire ||
2345 (entry->eflags & MAP_ENTRY_USER_WIRED))) {
2347 entry->eflags &= ~MAP_ENTRY_USER_WIRED;
2348 entry->wired_count--;
2349 if (entry->wired_count == 0) {
2351 * Retain the map lock.
2353 vm_fault_unwire(map, entry->start, entry->end,
2354 entry->object.vm_object != NULL &&
2355 (entry->object.vm_object->flags &
2356 OBJ_FICTITIOUS) != 0);
2359 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0,
2360 ("vm_map_unwire: in-transition flag missing"));
2361 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION;
2362 entry->wiring_thread = NULL;
2363 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
2364 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
2367 vm_map_simplify_entry(map, entry);
2378 * Implements both kernel and user wiring.
2381 vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t end,
2384 vm_map_entry_t entry, first_entry, tmp_entry;
2385 vm_offset_t saved_end, saved_start;
2386 unsigned int last_timestamp;
2388 boolean_t fictitious, need_wakeup, result, user_wire;
2392 if (flags & VM_MAP_WIRE_WRITE)
2393 prot |= VM_PROT_WRITE;
2394 user_wire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE;
2396 VM_MAP_RANGE_CHECK(map, start, end);
2397 if (!vm_map_lookup_entry(map, start, &first_entry)) {
2398 if (flags & VM_MAP_WIRE_HOLESOK)
2399 first_entry = first_entry->next;
2402 return (KERN_INVALID_ADDRESS);
2405 last_timestamp = map->timestamp;
2406 entry = first_entry;
2407 while (entry != &map->header && entry->start < end) {
2408 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
2410 * We have not yet clipped the entry.
2412 saved_start = (start >= entry->start) ? start :
2414 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2415 if (vm_map_unlock_and_wait(map, 0)) {
2417 * Allow interruption of user wiring?
2421 if (last_timestamp + 1 != map->timestamp) {
2423 * Look again for the entry because the map was
2424 * modified while it was unlocked.
2425 * Specifically, the entry may have been
2426 * clipped, merged, or deleted.
2428 if (!vm_map_lookup_entry(map, saved_start,
2430 if (flags & VM_MAP_WIRE_HOLESOK)
2431 tmp_entry = tmp_entry->next;
2433 if (saved_start == start) {
2435 * first_entry has been deleted.
2438 return (KERN_INVALID_ADDRESS);
2441 rv = KERN_INVALID_ADDRESS;
2445 if (entry == first_entry)
2446 first_entry = tmp_entry;
2451 last_timestamp = map->timestamp;
2454 vm_map_clip_start(map, entry, start);
2455 vm_map_clip_end(map, entry, end);
2457 * Mark the entry in case the map lock is released. (See
2460 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
2461 entry->wiring_thread = curthread;
2462 if ((entry->protection & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0
2463 || (entry->protection & prot) != prot) {
2464 entry->eflags |= MAP_ENTRY_WIRE_SKIPPED;
2465 if ((flags & VM_MAP_WIRE_HOLESOK) == 0) {
2467 rv = KERN_INVALID_ADDRESS;
2472 if (entry->wired_count == 0) {
2473 entry->wired_count++;
2474 saved_start = entry->start;
2475 saved_end = entry->end;
2476 fictitious = entry->object.vm_object != NULL &&
2477 (entry->object.vm_object->flags &
2478 OBJ_FICTITIOUS) != 0;
2480 * Release the map lock, relying on the in-transition
2481 * mark. Mark the map busy for fork.
2485 rv = vm_fault_wire(map, saved_start, saved_end,
2489 if (last_timestamp + 1 != map->timestamp) {
2491 * Look again for the entry because the map was
2492 * modified while it was unlocked. The entry
2493 * may have been clipped, but NOT merged or
2496 result = vm_map_lookup_entry(map, saved_start,
2498 KASSERT(result, ("vm_map_wire: lookup failed"));
2499 if (entry == first_entry)
2500 first_entry = tmp_entry;
2504 while (entry->end < saved_end) {
2505 if (rv != KERN_SUCCESS) {
2506 KASSERT(entry->wired_count == 1,
2507 ("vm_map_wire: bad count"));
2508 entry->wired_count = -1;
2510 entry = entry->next;
2513 last_timestamp = map->timestamp;
2514 if (rv != KERN_SUCCESS) {
2515 KASSERT(entry->wired_count == 1,
2516 ("vm_map_wire: bad count"));
2518 * Assign an out-of-range value to represent
2519 * the failure to wire this entry.
2521 entry->wired_count = -1;
2525 } else if (!user_wire ||
2526 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
2527 entry->wired_count++;
2530 * Check the map for holes in the specified region.
2531 * If VM_MAP_WIRE_HOLESOK was specified, skip this check.
2534 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) &&
2535 (entry->end < end && (entry->next == &map->header ||
2536 entry->next->start > entry->end))) {
2538 rv = KERN_INVALID_ADDRESS;
2541 entry = entry->next;
2545 need_wakeup = FALSE;
2546 if (first_entry == NULL) {
2547 result = vm_map_lookup_entry(map, start, &first_entry);
2548 if (!result && (flags & VM_MAP_WIRE_HOLESOK))
2549 first_entry = first_entry->next;
2551 KASSERT(result, ("vm_map_wire: lookup failed"));
2553 for (entry = first_entry; entry != &map->header && entry->start < end;
2554 entry = entry->next) {
2555 if ((entry->eflags & MAP_ENTRY_WIRE_SKIPPED) != 0)
2556 goto next_entry_done;
2559 * If VM_MAP_WIRE_HOLESOK was specified, an empty
2560 * space in the unwired region could have been mapped
2561 * while the map lock was dropped for faulting in the
2562 * pages or draining MAP_ENTRY_IN_TRANSITION.
2563 * Moreover, another thread could be simultaneously
2564 * wiring this new mapping entry. Detect these cases
2565 * and skip any entries marked as in transition by us.
2567 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 ||
2568 entry->wiring_thread != curthread) {
2569 KASSERT((flags & VM_MAP_WIRE_HOLESOK) != 0,
2570 ("vm_map_wire: !HOLESOK and new/changed entry"));
2574 if (rv == KERN_SUCCESS) {
2576 entry->eflags |= MAP_ENTRY_USER_WIRED;
2577 } else if (entry->wired_count == -1) {
2579 * Wiring failed on this entry. Thus, unwiring is
2582 entry->wired_count = 0;
2585 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0)
2586 entry->wired_count--;
2587 if (entry->wired_count == 0) {
2589 * Retain the map lock.
2591 vm_fault_unwire(map, entry->start, entry->end,
2592 entry->object.vm_object != NULL &&
2593 (entry->object.vm_object->flags &
2594 OBJ_FICTITIOUS) != 0);
2598 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0,
2599 ("vm_map_wire: in-transition flag missing %p", entry));
2600 KASSERT(entry->wiring_thread == curthread,
2601 ("vm_map_wire: alien wire %p", entry));
2602 entry->eflags &= ~(MAP_ENTRY_IN_TRANSITION |
2603 MAP_ENTRY_WIRE_SKIPPED);
2604 entry->wiring_thread = NULL;
2605 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
2606 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
2609 vm_map_simplify_entry(map, entry);
2620 * Push any dirty cached pages in the address range to their pager.
2621 * If syncio is TRUE, dirty pages are written synchronously.
2622 * If invalidate is TRUE, any cached pages are freed as well.
2624 * If the size of the region from start to end is zero, we are
2625 * supposed to flush all modified pages within the region containing
2626 * start. Unfortunately, a region can be split or coalesced with
2627 * neighboring regions, making it difficult to determine what the
2628 * original region was. Therefore, we approximate this requirement by
2629 * flushing the current region containing start.
2631 * Returns an error if any part of the specified range is not mapped.
2639 boolean_t invalidate)
2641 vm_map_entry_t current;
2642 vm_map_entry_t entry;
2645 vm_ooffset_t offset;
2646 unsigned int last_timestamp;
2649 vm_map_lock_read(map);
2650 VM_MAP_RANGE_CHECK(map, start, end);
2651 if (!vm_map_lookup_entry(map, start, &entry)) {
2652 vm_map_unlock_read(map);
2653 return (KERN_INVALID_ADDRESS);
2654 } else if (start == end) {
2655 start = entry->start;
2659 * Make a first pass to check for user-wired memory and holes.
2661 for (current = entry; current != &map->header && current->start < end;
2662 current = current->next) {
2663 if (invalidate && (current->eflags & MAP_ENTRY_USER_WIRED)) {
2664 vm_map_unlock_read(map);
2665 return (KERN_INVALID_ARGUMENT);
2667 if (end > current->end &&
2668 (current->next == &map->header ||
2669 current->end != current->next->start)) {
2670 vm_map_unlock_read(map);
2671 return (KERN_INVALID_ADDRESS);
2676 pmap_remove(map->pmap, start, end);
2680 * Make a second pass, cleaning/uncaching pages from the indicated
2683 for (current = entry; current != &map->header && current->start < end;) {
2684 offset = current->offset + (start - current->start);
2685 size = (end <= current->end ? end : current->end) - start;
2686 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
2688 vm_map_entry_t tentry;
2691 smap = current->object.sub_map;
2692 vm_map_lock_read(smap);
2693 (void) vm_map_lookup_entry(smap, offset, &tentry);
2694 tsize = tentry->end - offset;
2697 object = tentry->object.vm_object;
2698 offset = tentry->offset + (offset - tentry->start);
2699 vm_map_unlock_read(smap);
2701 object = current->object.vm_object;
2703 vm_object_reference(object);
2704 last_timestamp = map->timestamp;
2705 vm_map_unlock_read(map);
2706 if (!vm_object_sync(object, offset, size, syncio, invalidate))
2709 vm_object_deallocate(object);
2710 vm_map_lock_read(map);
2711 if (last_timestamp == map->timestamp ||
2712 !vm_map_lookup_entry(map, start, ¤t))
2713 current = current->next;
2716 vm_map_unlock_read(map);
2717 return (failed ? KERN_FAILURE : KERN_SUCCESS);
2721 * vm_map_entry_unwire: [ internal use only ]
2723 * Make the region specified by this entry pageable.
2725 * The map in question should be locked.
2726 * [This is the reason for this routine's existence.]
2729 vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry)
2731 vm_fault_unwire(map, entry->start, entry->end,
2732 entry->object.vm_object != NULL &&
2733 (entry->object.vm_object->flags & OBJ_FICTITIOUS) != 0);
2734 entry->wired_count = 0;
2738 vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map)
2741 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0)
2742 vm_object_deallocate(entry->object.vm_object);
2743 uma_zfree(system_map ? kmapentzone : mapentzone, entry);
2747 * vm_map_entry_delete: [ internal use only ]
2749 * Deallocate the given entry from the target map.
2752 vm_map_entry_delete(vm_map_t map, vm_map_entry_t entry)
2755 vm_pindex_t offidxstart, offidxend, count, size1;
2758 vm_map_entry_unlink(map, entry);
2759 object = entry->object.vm_object;
2760 size = entry->end - entry->start;
2763 if (entry->cred != NULL) {
2764 swap_release_by_cred(size, entry->cred);
2765 crfree(entry->cred);
2768 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0 &&
2770 KASSERT(entry->cred == NULL || object->cred == NULL ||
2771 (entry->eflags & MAP_ENTRY_NEEDS_COPY),
2772 ("OVERCOMMIT vm_map_entry_delete: both cred %p", entry));
2773 count = OFF_TO_IDX(size);
2774 offidxstart = OFF_TO_IDX(entry->offset);
2775 offidxend = offidxstart + count;
2776 VM_OBJECT_WLOCK(object);
2777 if (object->ref_count != 1 &&
2778 ((object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING ||
2779 object == kernel_object || object == kmem_object)) {
2780 vm_object_collapse(object);
2783 * The option OBJPR_NOTMAPPED can be passed here
2784 * because vm_map_delete() already performed
2785 * pmap_remove() on the only mapping to this range
2788 vm_object_page_remove(object, offidxstart, offidxend,
2790 if (object->type == OBJT_SWAP)
2791 swap_pager_freespace(object, offidxstart, count);
2792 if (offidxend >= object->size &&
2793 offidxstart < object->size) {
2794 size1 = object->size;
2795 object->size = offidxstart;
2796 if (object->cred != NULL) {
2797 size1 -= object->size;
2798 KASSERT(object->charge >= ptoa(size1),
2799 ("vm_map_entry_delete: object->charge < 0"));
2800 swap_release_by_cred(ptoa(size1), object->cred);
2801 object->charge -= ptoa(size1);
2805 VM_OBJECT_WUNLOCK(object);
2807 entry->object.vm_object = NULL;
2808 if (map->system_map)
2809 vm_map_entry_deallocate(entry, TRUE);
2811 entry->next = curthread->td_map_def_user;
2812 curthread->td_map_def_user = entry;
2817 * vm_map_delete: [ internal use only ]
2819 * Deallocates the given address range from the target
2823 vm_map_delete(vm_map_t map, vm_offset_t start, vm_offset_t end)
2825 vm_map_entry_t entry;
2826 vm_map_entry_t first_entry;
2828 VM_MAP_ASSERT_LOCKED(map);
2831 * Find the start of the region, and clip it
2833 if (!vm_map_lookup_entry(map, start, &first_entry))
2834 entry = first_entry->next;
2836 entry = first_entry;
2837 vm_map_clip_start(map, entry, start);
2841 * Step through all entries in this region
2843 while ((entry != &map->header) && (entry->start < end)) {
2844 vm_map_entry_t next;
2847 * Wait for wiring or unwiring of an entry to complete.
2848 * Also wait for any system wirings to disappear on
2851 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 ||
2852 (vm_map_pmap(map) != kernel_pmap &&
2853 vm_map_entry_system_wired_count(entry) != 0)) {
2854 unsigned int last_timestamp;
2855 vm_offset_t saved_start;
2856 vm_map_entry_t tmp_entry;
2858 saved_start = entry->start;
2859 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2860 last_timestamp = map->timestamp;
2861 (void) vm_map_unlock_and_wait(map, 0);
2863 if (last_timestamp + 1 != map->timestamp) {
2865 * Look again for the entry because the map was
2866 * modified while it was unlocked.
2867 * Specifically, the entry may have been
2868 * clipped, merged, or deleted.
2870 if (!vm_map_lookup_entry(map, saved_start,
2872 entry = tmp_entry->next;
2875 vm_map_clip_start(map, entry,
2881 vm_map_clip_end(map, entry, end);
2886 * Unwire before removing addresses from the pmap; otherwise,
2887 * unwiring will put the entries back in the pmap.
2889 if (entry->wired_count != 0) {
2890 vm_map_entry_unwire(map, entry);
2893 pmap_remove(map->pmap, entry->start, entry->end);
2896 * Delete the entry only after removing all pmap
2897 * entries pointing to its pages. (Otherwise, its
2898 * page frames may be reallocated, and any modify bits
2899 * will be set in the wrong object!)
2901 vm_map_entry_delete(map, entry);
2904 return (KERN_SUCCESS);
2910 * Remove the given address range from the target map.
2911 * This is the exported form of vm_map_delete.
2914 vm_map_remove(vm_map_t map, vm_offset_t start, vm_offset_t end)
2919 VM_MAP_RANGE_CHECK(map, start, end);
2920 result = vm_map_delete(map, start, end);
2926 * vm_map_check_protection:
2928 * Assert that the target map allows the specified privilege on the
2929 * entire address region given. The entire region must be allocated.
2931 * WARNING! This code does not and should not check whether the
2932 * contents of the region is accessible. For example a smaller file
2933 * might be mapped into a larger address space.
2935 * NOTE! This code is also called by munmap().
2937 * The map must be locked. A read lock is sufficient.
2940 vm_map_check_protection(vm_map_t map, vm_offset_t start, vm_offset_t end,
2941 vm_prot_t protection)
2943 vm_map_entry_t entry;
2944 vm_map_entry_t tmp_entry;
2946 if (!vm_map_lookup_entry(map, start, &tmp_entry))
2950 while (start < end) {
2951 if (entry == &map->header)
2956 if (start < entry->start)
2959 * Check protection associated with entry.
2961 if ((entry->protection & protection) != protection)
2963 /* go to next entry */
2965 entry = entry->next;
2971 * vm_map_copy_entry:
2973 * Copies the contents of the source entry to the destination
2974 * entry. The entries *must* be aligned properly.
2980 vm_map_entry_t src_entry,
2981 vm_map_entry_t dst_entry,
2982 vm_ooffset_t *fork_charge)
2984 vm_object_t src_object;
2985 vm_map_entry_t fake_entry;
2990 VM_MAP_ASSERT_LOCKED(dst_map);
2992 if ((dst_entry->eflags|src_entry->eflags) & MAP_ENTRY_IS_SUB_MAP)
2995 if (src_entry->wired_count == 0) {
2998 * If the source entry is marked needs_copy, it is already
3001 if ((src_entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) {
3002 pmap_protect(src_map->pmap,
3005 src_entry->protection & ~VM_PROT_WRITE);
3009 * Make a copy of the object.
3011 size = src_entry->end - src_entry->start;
3012 if ((src_object = src_entry->object.vm_object) != NULL) {
3013 VM_OBJECT_WLOCK(src_object);
3014 charged = ENTRY_CHARGED(src_entry);
3015 if ((src_object->handle == NULL) &&
3016 (src_object->type == OBJT_DEFAULT ||
3017 src_object->type == OBJT_SWAP)) {
3018 vm_object_collapse(src_object);
3019 if ((src_object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING) {
3020 vm_object_split(src_entry);
3021 src_object = src_entry->object.vm_object;
3024 vm_object_reference_locked(src_object);
3025 vm_object_clear_flag(src_object, OBJ_ONEMAPPING);
3026 if (src_entry->cred != NULL &&
3027 !(src_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
3028 KASSERT(src_object->cred == NULL,
3029 ("OVERCOMMIT: vm_map_copy_entry: cred %p",
3031 src_object->cred = src_entry->cred;
3032 src_object->charge = size;
3034 VM_OBJECT_WUNLOCK(src_object);
3035 dst_entry->object.vm_object = src_object;
3037 cred = curthread->td_ucred;
3039 dst_entry->cred = cred;
3040 *fork_charge += size;
3041 if (!(src_entry->eflags &
3042 MAP_ENTRY_NEEDS_COPY)) {
3044 src_entry->cred = cred;
3045 *fork_charge += size;
3048 src_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY);
3049 dst_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY);
3050 dst_entry->offset = src_entry->offset;
3051 if (src_entry->eflags & MAP_ENTRY_VN_WRITECNT) {
3053 * MAP_ENTRY_VN_WRITECNT cannot
3054 * indicate write reference from
3055 * src_entry, since the entry is
3056 * marked as needs copy. Allocate a
3057 * fake entry that is used to
3058 * decrement object->un_pager.vnp.writecount
3059 * at the appropriate time. Attach
3060 * fake_entry to the deferred list.
3062 fake_entry = vm_map_entry_create(dst_map);
3063 fake_entry->eflags = MAP_ENTRY_VN_WRITECNT;
3064 src_entry->eflags &= ~MAP_ENTRY_VN_WRITECNT;
3065 vm_object_reference(src_object);
3066 fake_entry->object.vm_object = src_object;
3067 fake_entry->start = src_entry->start;
3068 fake_entry->end = src_entry->end;
3069 fake_entry->next = curthread->td_map_def_user;
3070 curthread->td_map_def_user = fake_entry;
3073 dst_entry->object.vm_object = NULL;
3074 dst_entry->offset = 0;
3075 if (src_entry->cred != NULL) {
3076 dst_entry->cred = curthread->td_ucred;
3077 crhold(dst_entry->cred);
3078 *fork_charge += size;
3082 pmap_copy(dst_map->pmap, src_map->pmap, dst_entry->start,
3083 dst_entry->end - dst_entry->start, src_entry->start);
3086 * Of course, wired down pages can't be set copy-on-write.
3087 * Cause wired pages to be copied into the new map by
3088 * simulating faults (the new pages are pageable)
3090 vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry,
3096 * vmspace_map_entry_forked:
3097 * Update the newly-forked vmspace each time a map entry is inherited
3098 * or copied. The values for vm_dsize and vm_tsize are approximate
3099 * (and mostly-obsolete ideas in the face of mmap(2) et al.)
3102 vmspace_map_entry_forked(const struct vmspace *vm1, struct vmspace *vm2,
3103 vm_map_entry_t entry)
3105 vm_size_t entrysize;
3108 entrysize = entry->end - entry->start;
3109 vm2->vm_map.size += entrysize;
3110 if (entry->eflags & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP)) {
3111 vm2->vm_ssize += btoc(entrysize);
3112 } else if (entry->start >= (vm_offset_t)vm1->vm_daddr &&
3113 entry->start < (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)) {
3114 newend = MIN(entry->end,
3115 (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize));
3116 vm2->vm_dsize += btoc(newend - entry->start);
3117 } else if (entry->start >= (vm_offset_t)vm1->vm_taddr &&
3118 entry->start < (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)) {
3119 newend = MIN(entry->end,
3120 (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize));
3121 vm2->vm_tsize += btoc(newend - entry->start);
3127 * Create a new process vmspace structure and vm_map
3128 * based on those of an existing process. The new map
3129 * is based on the old map, according to the inheritance
3130 * values on the regions in that map.
3132 * XXX It might be worth coalescing the entries added to the new vmspace.
3134 * The source map must not be locked.
3137 vmspace_fork(struct vmspace *vm1, vm_ooffset_t *fork_charge)
3139 struct vmspace *vm2;
3140 vm_map_t new_map, old_map;
3141 vm_map_entry_t new_entry, old_entry;
3145 old_map = &vm1->vm_map;
3146 /* Copy immutable fields of vm1 to vm2. */
3147 vm2 = vmspace_alloc(old_map->min_offset, old_map->max_offset);
3150 vm2->vm_taddr = vm1->vm_taddr;
3151 vm2->vm_daddr = vm1->vm_daddr;
3152 vm2->vm_maxsaddr = vm1->vm_maxsaddr;
3153 vm_map_lock(old_map);
3155 vm_map_wait_busy(old_map);
3156 new_map = &vm2->vm_map;
3157 locked = vm_map_trylock(new_map); /* trylock to silence WITNESS */
3158 KASSERT(locked, ("vmspace_fork: lock failed"));
3160 old_entry = old_map->header.next;
3162 while (old_entry != &old_map->header) {
3163 if (old_entry->eflags & MAP_ENTRY_IS_SUB_MAP)
3164 panic("vm_map_fork: encountered a submap");
3166 switch (old_entry->inheritance) {
3167 case VM_INHERIT_NONE:
3170 case VM_INHERIT_SHARE:
3172 * Clone the entry, creating the shared object if necessary.
3174 object = old_entry->object.vm_object;
3175 if (object == NULL) {
3176 object = vm_object_allocate(OBJT_DEFAULT,
3177 atop(old_entry->end - old_entry->start));
3178 old_entry->object.vm_object = object;
3179 old_entry->offset = 0;
3180 if (old_entry->cred != NULL) {
3181 object->cred = old_entry->cred;
3182 object->charge = old_entry->end -
3184 old_entry->cred = NULL;
3189 * Add the reference before calling vm_object_shadow
3190 * to insure that a shadow object is created.
3192 vm_object_reference(object);
3193 if (old_entry->eflags & MAP_ENTRY_NEEDS_COPY) {
3194 vm_object_shadow(&old_entry->object.vm_object,
3196 old_entry->end - old_entry->start);
3197 old_entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
3198 /* Transfer the second reference too. */
3199 vm_object_reference(
3200 old_entry->object.vm_object);
3203 * As in vm_map_simplify_entry(), the
3204 * vnode lock will not be acquired in
3205 * this call to vm_object_deallocate().
3207 vm_object_deallocate(object);
3208 object = old_entry->object.vm_object;
3210 VM_OBJECT_WLOCK(object);
3211 vm_object_clear_flag(object, OBJ_ONEMAPPING);
3212 if (old_entry->cred != NULL) {
3213 KASSERT(object->cred == NULL, ("vmspace_fork both cred"));
3214 object->cred = old_entry->cred;
3215 object->charge = old_entry->end - old_entry->start;
3216 old_entry->cred = NULL;
3220 * Assert the correct state of the vnode
3221 * v_writecount while the object is locked, to
3222 * not relock it later for the assertion
3225 if (old_entry->eflags & MAP_ENTRY_VN_WRITECNT &&
3226 object->type == OBJT_VNODE) {
3227 KASSERT(((struct vnode *)object->handle)->
3229 ("vmspace_fork: v_writecount %p", object));
3230 KASSERT(object->un_pager.vnp.writemappings > 0,
3231 ("vmspace_fork: vnp.writecount %p",
3234 VM_OBJECT_WUNLOCK(object);
3237 * Clone the entry, referencing the shared object.
3239 new_entry = vm_map_entry_create(new_map);
3240 *new_entry = *old_entry;
3241 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED |
3242 MAP_ENTRY_IN_TRANSITION);
3243 new_entry->wiring_thread = NULL;
3244 new_entry->wired_count = 0;
3245 if (new_entry->eflags & MAP_ENTRY_VN_WRITECNT) {
3246 vnode_pager_update_writecount(object,
3247 new_entry->start, new_entry->end);
3251 * Insert the entry into the new map -- we know we're
3252 * inserting at the end of the new map.
3254 vm_map_entry_link(new_map, new_map->header.prev,
3256 vmspace_map_entry_forked(vm1, vm2, new_entry);
3259 * Update the physical map
3261 pmap_copy(new_map->pmap, old_map->pmap,
3263 (old_entry->end - old_entry->start),
3267 case VM_INHERIT_COPY:
3269 * Clone the entry and link into the map.
3271 new_entry = vm_map_entry_create(new_map);
3272 *new_entry = *old_entry;
3274 * Copied entry is COW over the old object.
3276 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED |
3277 MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_VN_WRITECNT);
3278 new_entry->wiring_thread = NULL;
3279 new_entry->wired_count = 0;
3280 new_entry->object.vm_object = NULL;
3281 new_entry->cred = NULL;
3282 vm_map_entry_link(new_map, new_map->header.prev,
3284 vmspace_map_entry_forked(vm1, vm2, new_entry);
3285 vm_map_copy_entry(old_map, new_map, old_entry,
3286 new_entry, fork_charge);
3289 old_entry = old_entry->next;
3292 * Use inlined vm_map_unlock() to postpone handling the deferred
3293 * map entries, which cannot be done until both old_map and
3294 * new_map locks are released.
3296 sx_xunlock(&old_map->lock);
3297 sx_xunlock(&new_map->lock);
3298 vm_map_process_deferred();
3304 vm_map_stack(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
3305 vm_prot_t prot, vm_prot_t max, int cow)
3307 vm_map_entry_t new_entry, prev_entry;
3308 vm_offset_t bot, top;
3309 vm_size_t growsize, init_ssize;
3311 rlim_t lmemlim, vmemlim;
3314 * The stack orientation is piggybacked with the cow argument.
3315 * Extract it into orient and mask the cow argument so that we
3316 * don't pass it around further.
3317 * NOTE: We explicitly allow bi-directional stacks.
3319 orient = cow & (MAP_STACK_GROWS_DOWN|MAP_STACK_GROWS_UP);
3321 KASSERT(orient != 0, ("No stack grow direction"));
3323 if (addrbos < vm_map_min(map) ||
3324 addrbos > vm_map_max(map) ||
3325 addrbos + max_ssize < addrbos)
3326 return (KERN_NO_SPACE);
3328 growsize = sgrowsiz;
3329 init_ssize = (max_ssize < growsize) ? max_ssize : growsize;
3332 lmemlim = lim_cur(curproc, RLIMIT_MEMLOCK);
3333 vmemlim = lim_cur(curproc, RLIMIT_VMEM);
3334 PROC_UNLOCK(curproc);
3338 /* If addr is already mapped, no go */
3339 if (vm_map_lookup_entry(map, addrbos, &prev_entry)) {
3341 return (KERN_NO_SPACE);
3344 if (!old_mlock && map->flags & MAP_WIREFUTURE) {
3345 if (ptoa(pmap_wired_count(map->pmap)) + init_ssize > lmemlim) {
3347 return (KERN_NO_SPACE);
3351 /* If we would blow our VMEM resource limit, no go */
3352 if (map->size + init_ssize > vmemlim) {
3354 return (KERN_NO_SPACE);
3358 * If we can't accomodate max_ssize in the current mapping, no go.
3359 * However, we need to be aware that subsequent user mappings might
3360 * map into the space we have reserved for stack, and currently this
3361 * space is not protected.
3363 * Hopefully we will at least detect this condition when we try to
3366 if ((prev_entry->next != &map->header) &&
3367 (prev_entry->next->start < addrbos + max_ssize)) {
3369 return (KERN_NO_SPACE);
3373 * We initially map a stack of only init_ssize. We will grow as
3374 * needed later. Depending on the orientation of the stack (i.e.
3375 * the grow direction) we either map at the top of the range, the
3376 * bottom of the range or in the middle.
3378 * Note: we would normally expect prot and max to be VM_PROT_ALL,
3379 * and cow to be 0. Possibly we should eliminate these as input
3380 * parameters, and just pass these values here in the insert call.
3382 if (orient == MAP_STACK_GROWS_DOWN)
3383 bot = addrbos + max_ssize - init_ssize;
3384 else if (orient == MAP_STACK_GROWS_UP)
3387 bot = round_page(addrbos + max_ssize/2 - init_ssize/2);
3388 top = bot + init_ssize;
3389 rv = vm_map_insert(map, NULL, 0, bot, top, prot, max, cow);
3391 /* Now set the avail_ssize amount. */
3392 if (rv == KERN_SUCCESS) {
3393 if (prev_entry != &map->header)
3394 vm_map_clip_end(map, prev_entry, bot);
3395 new_entry = prev_entry->next;
3396 if (new_entry->end != top || new_entry->start != bot)
3397 panic("Bad entry start/end for new stack entry");
3399 new_entry->avail_ssize = max_ssize - init_ssize;
3400 if (orient & MAP_STACK_GROWS_DOWN)
3401 new_entry->eflags |= MAP_ENTRY_GROWS_DOWN;
3402 if (orient & MAP_STACK_GROWS_UP)
3403 new_entry->eflags |= MAP_ENTRY_GROWS_UP;
3410 static int stack_guard_page = 0;
3411 TUNABLE_INT("security.bsd.stack_guard_page", &stack_guard_page);
3412 SYSCTL_INT(_security_bsd, OID_AUTO, stack_guard_page, CTLFLAG_RW,
3413 &stack_guard_page, 0,
3414 "Insert stack guard page ahead of the growable segments.");
3416 /* Attempts to grow a vm stack entry. Returns KERN_SUCCESS if the
3417 * desired address is already mapped, or if we successfully grow
3418 * the stack. Also returns KERN_SUCCESS if addr is outside the
3419 * stack range (this is strange, but preserves compatibility with
3420 * the grow function in vm_machdep.c).
3423 vm_map_growstack(struct proc *p, vm_offset_t addr)
3425 vm_map_entry_t next_entry, prev_entry;
3426 vm_map_entry_t new_entry, stack_entry;
3427 struct vmspace *vm = p->p_vmspace;
3428 vm_map_t map = &vm->vm_map;
3431 size_t grow_amount, max_grow;
3432 rlim_t lmemlim, stacklim, vmemlim;
3433 int is_procstack, rv;
3444 lmemlim = lim_cur(p, RLIMIT_MEMLOCK);
3445 stacklim = lim_cur(p, RLIMIT_STACK);
3446 vmemlim = lim_cur(p, RLIMIT_VMEM);
3449 vm_map_lock_read(map);
3451 /* If addr is already in the entry range, no need to grow.*/
3452 if (vm_map_lookup_entry(map, addr, &prev_entry)) {
3453 vm_map_unlock_read(map);
3454 return (KERN_SUCCESS);
3457 next_entry = prev_entry->next;
3458 if (!(prev_entry->eflags & MAP_ENTRY_GROWS_UP)) {
3460 * This entry does not grow upwards. Since the address lies
3461 * beyond this entry, the next entry (if one exists) has to
3462 * be a downward growable entry. The entry list header is
3463 * never a growable entry, so it suffices to check the flags.
3465 if (!(next_entry->eflags & MAP_ENTRY_GROWS_DOWN)) {
3466 vm_map_unlock_read(map);
3467 return (KERN_SUCCESS);
3469 stack_entry = next_entry;
3472 * This entry grows upward. If the next entry does not at
3473 * least grow downwards, this is the entry we need to grow.
3474 * otherwise we have two possible choices and we have to
3477 if (next_entry->eflags & MAP_ENTRY_GROWS_DOWN) {
3479 * We have two choices; grow the entry closest to
3480 * the address to minimize the amount of growth.
3482 if (addr - prev_entry->end <= next_entry->start - addr)
3483 stack_entry = prev_entry;
3485 stack_entry = next_entry;
3487 stack_entry = prev_entry;
3490 if (stack_entry == next_entry) {
3491 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_DOWN, ("foo"));
3492 KASSERT(addr < stack_entry->start, ("foo"));
3493 end = (prev_entry != &map->header) ? prev_entry->end :
3494 stack_entry->start - stack_entry->avail_ssize;
3495 grow_amount = roundup(stack_entry->start - addr, PAGE_SIZE);
3496 max_grow = stack_entry->start - end;
3498 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_UP, ("foo"));
3499 KASSERT(addr >= stack_entry->end, ("foo"));
3500 end = (next_entry != &map->header) ? next_entry->start :
3501 stack_entry->end + stack_entry->avail_ssize;
3502 grow_amount = roundup(addr + 1 - stack_entry->end, PAGE_SIZE);
3503 max_grow = end - stack_entry->end;
3506 if (grow_amount > stack_entry->avail_ssize) {
3507 vm_map_unlock_read(map);
3508 return (KERN_NO_SPACE);
3512 * If there is no longer enough space between the entries nogo, and
3513 * adjust the available space. Note: this should only happen if the
3514 * user has mapped into the stack area after the stack was created,
3515 * and is probably an error.
3517 * This also effectively destroys any guard page the user might have
3518 * intended by limiting the stack size.
3520 if (grow_amount + (stack_guard_page ? PAGE_SIZE : 0) > max_grow) {
3521 if (vm_map_lock_upgrade(map))
3524 stack_entry->avail_ssize = max_grow;
3527 return (KERN_NO_SPACE);
3530 is_procstack = (addr >= (vm_offset_t)vm->vm_maxsaddr) ? 1 : 0;
3533 * If this is the main process stack, see if we're over the stack
3536 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) {
3537 vm_map_unlock_read(map);
3538 return (KERN_NO_SPACE);
3543 racct_set(p, RACCT_STACK, ctob(vm->vm_ssize) + grow_amount)) {
3545 vm_map_unlock_read(map);
3546 return (KERN_NO_SPACE);
3551 /* Round up the grow amount modulo sgrowsiz */
3552 growsize = sgrowsiz;
3553 grow_amount = roundup(grow_amount, growsize);
3554 if (grow_amount > stack_entry->avail_ssize)
3555 grow_amount = stack_entry->avail_ssize;
3556 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) {
3557 grow_amount = trunc_page((vm_size_t)stacklim) -
3562 limit = racct_get_available(p, RACCT_STACK);
3564 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > limit))
3565 grow_amount = limit - ctob(vm->vm_ssize);
3567 if (!old_mlock && map->flags & MAP_WIREFUTURE) {
3568 if (ptoa(pmap_wired_count(map->pmap)) + grow_amount > lmemlim) {
3569 vm_map_unlock_read(map);
3575 if (racct_set(p, RACCT_MEMLOCK,
3576 ptoa(pmap_wired_count(map->pmap)) + grow_amount)) {
3578 vm_map_unlock_read(map);
3585 /* If we would blow our VMEM resource limit, no go */
3586 if (map->size + grow_amount > vmemlim) {
3587 vm_map_unlock_read(map);
3593 if (racct_set(p, RACCT_VMEM, map->size + grow_amount)) {
3595 vm_map_unlock_read(map);
3602 if (vm_map_lock_upgrade(map))
3605 if (stack_entry == next_entry) {
3609 /* Get the preliminary new entry start value */
3610 addr = stack_entry->start - grow_amount;
3613 * If this puts us into the previous entry, cut back our
3614 * growth to the available space. Also, see the note above.
3617 stack_entry->avail_ssize = max_grow;
3619 if (stack_guard_page)
3623 rv = vm_map_insert(map, NULL, 0, addr, stack_entry->start,
3624 next_entry->protection, next_entry->max_protection, 0);
3626 /* Adjust the available stack space by the amount we grew. */
3627 if (rv == KERN_SUCCESS) {
3628 if (prev_entry != &map->header)
3629 vm_map_clip_end(map, prev_entry, addr);
3630 new_entry = prev_entry->next;
3631 KASSERT(new_entry == stack_entry->prev, ("foo"));
3632 KASSERT(new_entry->end == stack_entry->start, ("foo"));
3633 KASSERT(new_entry->start == addr, ("foo"));
3634 grow_amount = new_entry->end - new_entry->start;
3635 new_entry->avail_ssize = stack_entry->avail_ssize -
3637 stack_entry->eflags &= ~MAP_ENTRY_GROWS_DOWN;
3638 new_entry->eflags |= MAP_ENTRY_GROWS_DOWN;
3644 addr = stack_entry->end + grow_amount;
3647 * If this puts us into the next entry, cut back our growth
3648 * to the available space. Also, see the note above.
3651 stack_entry->avail_ssize = end - stack_entry->end;
3653 if (stack_guard_page)
3657 grow_amount = addr - stack_entry->end;
3658 cred = stack_entry->cred;
3659 if (cred == NULL && stack_entry->object.vm_object != NULL)
3660 cred = stack_entry->object.vm_object->cred;
3661 if (cred != NULL && !swap_reserve_by_cred(grow_amount, cred))
3663 /* Grow the underlying object if applicable. */
3664 else if (stack_entry->object.vm_object == NULL ||
3665 vm_object_coalesce(stack_entry->object.vm_object,
3666 stack_entry->offset,
3667 (vm_size_t)(stack_entry->end - stack_entry->start),
3668 (vm_size_t)grow_amount, cred != NULL)) {
3669 map->size += (addr - stack_entry->end);
3670 /* Update the current entry. */
3671 stack_entry->end = addr;
3672 stack_entry->avail_ssize -= grow_amount;
3673 vm_map_entry_resize_free(map, stack_entry);
3676 if (next_entry != &map->header)
3677 vm_map_clip_start(map, next_entry, addr);
3682 if (rv == KERN_SUCCESS && is_procstack)
3683 vm->vm_ssize += btoc(grow_amount);
3688 * Heed the MAP_WIREFUTURE flag if it was set for this process.
3690 if (rv == KERN_SUCCESS && (map->flags & MAP_WIREFUTURE)) {
3692 (stack_entry == next_entry) ? addr : addr - grow_amount,
3693 (stack_entry == next_entry) ? stack_entry->start : addr,
3694 (p->p_flag & P_SYSTEM)
3695 ? VM_MAP_WIRE_SYSTEM|VM_MAP_WIRE_NOHOLES
3696 : VM_MAP_WIRE_USER|VM_MAP_WIRE_NOHOLES);
3701 if (rv != KERN_SUCCESS) {
3703 error = racct_set(p, RACCT_VMEM, map->size);
3704 KASSERT(error == 0, ("decreasing RACCT_VMEM failed"));
3706 error = racct_set(p, RACCT_MEMLOCK,
3707 ptoa(pmap_wired_count(map->pmap)));
3708 KASSERT(error == 0, ("decreasing RACCT_MEMLOCK failed"));
3710 error = racct_set(p, RACCT_STACK, ctob(vm->vm_ssize));
3711 KASSERT(error == 0, ("decreasing RACCT_STACK failed"));
3720 * Unshare the specified VM space for exec. If other processes are
3721 * mapped to it, then create a new one. The new vmspace is null.
3724 vmspace_exec(struct proc *p, vm_offset_t minuser, vm_offset_t maxuser)
3726 struct vmspace *oldvmspace = p->p_vmspace;
3727 struct vmspace *newvmspace;
3729 newvmspace = vmspace_alloc(minuser, maxuser);
3730 if (newvmspace == NULL)
3732 newvmspace->vm_swrss = oldvmspace->vm_swrss;
3734 * This code is written like this for prototype purposes. The
3735 * goal is to avoid running down the vmspace here, but let the
3736 * other process's that are still using the vmspace to finally
3737 * run it down. Even though there is little or no chance of blocking
3738 * here, it is a good idea to keep this form for future mods.
3740 PROC_VMSPACE_LOCK(p);
3741 p->p_vmspace = newvmspace;
3742 PROC_VMSPACE_UNLOCK(p);
3743 if (p == curthread->td_proc)
3744 pmap_activate(curthread);
3745 vmspace_free(oldvmspace);
3750 * Unshare the specified VM space for forcing COW. This
3751 * is called by rfork, for the (RFMEM|RFPROC) == 0 case.
3754 vmspace_unshare(struct proc *p)
3756 struct vmspace *oldvmspace = p->p_vmspace;
3757 struct vmspace *newvmspace;
3758 vm_ooffset_t fork_charge;
3760 if (oldvmspace->vm_refcnt == 1)
3763 newvmspace = vmspace_fork(oldvmspace, &fork_charge);
3764 if (newvmspace == NULL)
3766 if (!swap_reserve_by_cred(fork_charge, p->p_ucred)) {
3767 vmspace_free(newvmspace);
3770 PROC_VMSPACE_LOCK(p);
3771 p->p_vmspace = newvmspace;
3772 PROC_VMSPACE_UNLOCK(p);
3773 if (p == curthread->td_proc)
3774 pmap_activate(curthread);
3775 vmspace_free(oldvmspace);
3782 * Finds the VM object, offset, and
3783 * protection for a given virtual address in the
3784 * specified map, assuming a page fault of the
3787 * Leaves the map in question locked for read; return
3788 * values are guaranteed until a vm_map_lookup_done
3789 * call is performed. Note that the map argument
3790 * is in/out; the returned map must be used in
3791 * the call to vm_map_lookup_done.
3793 * A handle (out_entry) is returned for use in
3794 * vm_map_lookup_done, to make that fast.
3796 * If a lookup is requested with "write protection"
3797 * specified, the map may be changed to perform virtual
3798 * copying operations, although the data referenced will
3802 vm_map_lookup(vm_map_t *var_map, /* IN/OUT */
3804 vm_prot_t fault_typea,
3805 vm_map_entry_t *out_entry, /* OUT */
3806 vm_object_t *object, /* OUT */
3807 vm_pindex_t *pindex, /* OUT */
3808 vm_prot_t *out_prot, /* OUT */
3809 boolean_t *wired) /* OUT */
3811 vm_map_entry_t entry;
3812 vm_map_t map = *var_map;
3814 vm_prot_t fault_type = fault_typea;
3815 vm_object_t eobject;
3821 vm_map_lock_read(map);
3824 * Lookup the faulting address.
3826 if (!vm_map_lookup_entry(map, vaddr, out_entry)) {
3827 vm_map_unlock_read(map);
3828 return (KERN_INVALID_ADDRESS);
3836 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
3837 vm_map_t old_map = map;
3839 *var_map = map = entry->object.sub_map;
3840 vm_map_unlock_read(old_map);
3845 * Check whether this task is allowed to have this page.
3847 prot = entry->protection;
3848 fault_type &= (VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE);
3849 if ((fault_type & prot) != fault_type || prot == VM_PROT_NONE) {
3850 vm_map_unlock_read(map);
3851 return (KERN_PROTECTION_FAILURE);
3853 if ((entry->eflags & MAP_ENTRY_USER_WIRED) &&
3854 (entry->eflags & MAP_ENTRY_COW) &&
3855 (fault_type & VM_PROT_WRITE)) {
3856 vm_map_unlock_read(map);
3857 return (KERN_PROTECTION_FAILURE);
3859 if ((fault_typea & VM_PROT_COPY) != 0 &&
3860 (entry->max_protection & VM_PROT_WRITE) == 0 &&
3861 (entry->eflags & MAP_ENTRY_COW) == 0) {
3862 vm_map_unlock_read(map);
3863 return (KERN_PROTECTION_FAILURE);
3867 * If this page is not pageable, we have to get it for all possible
3870 *wired = (entry->wired_count != 0);
3872 fault_type = entry->protection;
3873 size = entry->end - entry->start;
3875 * If the entry was copy-on-write, we either ...
3877 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
3879 * If we want to write the page, we may as well handle that
3880 * now since we've got the map locked.
3882 * If we don't need to write the page, we just demote the
3883 * permissions allowed.
3885 if ((fault_type & VM_PROT_WRITE) != 0 ||
3886 (fault_typea & VM_PROT_COPY) != 0) {
3888 * Make a new object, and place it in the object
3889 * chain. Note that no new references have appeared
3890 * -- one just moved from the map to the new
3893 if (vm_map_lock_upgrade(map))
3896 if (entry->cred == NULL) {
3898 * The debugger owner is charged for
3901 cred = curthread->td_ucred;
3903 if (!swap_reserve_by_cred(size, cred)) {
3906 return (KERN_RESOURCE_SHORTAGE);
3910 vm_object_shadow(&entry->object.vm_object,
3911 &entry->offset, size);
3912 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
3913 eobject = entry->object.vm_object;
3914 if (eobject->cred != NULL) {
3916 * The object was not shadowed.
3918 swap_release_by_cred(size, entry->cred);
3919 crfree(entry->cred);
3921 } else if (entry->cred != NULL) {
3922 VM_OBJECT_WLOCK(eobject);
3923 eobject->cred = entry->cred;
3924 eobject->charge = size;
3925 VM_OBJECT_WUNLOCK(eobject);
3929 vm_map_lock_downgrade(map);
3932 * We're attempting to read a copy-on-write page --
3933 * don't allow writes.
3935 prot &= ~VM_PROT_WRITE;
3940 * Create an object if necessary.
3942 if (entry->object.vm_object == NULL &&
3944 if (vm_map_lock_upgrade(map))
3946 entry->object.vm_object = vm_object_allocate(OBJT_DEFAULT,
3949 if (entry->cred != NULL) {
3950 VM_OBJECT_WLOCK(entry->object.vm_object);
3951 entry->object.vm_object->cred = entry->cred;
3952 entry->object.vm_object->charge = size;
3953 VM_OBJECT_WUNLOCK(entry->object.vm_object);
3956 vm_map_lock_downgrade(map);
3960 * Return the object/offset from this entry. If the entry was
3961 * copy-on-write or empty, it has been fixed up.
3963 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
3964 *object = entry->object.vm_object;
3967 return (KERN_SUCCESS);
3971 * vm_map_lookup_locked:
3973 * Lookup the faulting address. A version of vm_map_lookup that returns
3974 * KERN_FAILURE instead of blocking on map lock or memory allocation.
3977 vm_map_lookup_locked(vm_map_t *var_map, /* IN/OUT */
3979 vm_prot_t fault_typea,
3980 vm_map_entry_t *out_entry, /* OUT */
3981 vm_object_t *object, /* OUT */
3982 vm_pindex_t *pindex, /* OUT */
3983 vm_prot_t *out_prot, /* OUT */
3984 boolean_t *wired) /* OUT */
3986 vm_map_entry_t entry;
3987 vm_map_t map = *var_map;
3989 vm_prot_t fault_type = fault_typea;
3992 * Lookup the faulting address.
3994 if (!vm_map_lookup_entry(map, vaddr, out_entry))
3995 return (KERN_INVALID_ADDRESS);
4000 * Fail if the entry refers to a submap.
4002 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP)
4003 return (KERN_FAILURE);
4006 * Check whether this task is allowed to have this page.
4008 prot = entry->protection;
4009 fault_type &= VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE;
4010 if ((fault_type & prot) != fault_type)
4011 return (KERN_PROTECTION_FAILURE);
4012 if ((entry->eflags & MAP_ENTRY_USER_WIRED) &&
4013 (entry->eflags & MAP_ENTRY_COW) &&
4014 (fault_type & VM_PROT_WRITE))
4015 return (KERN_PROTECTION_FAILURE);
4018 * If this page is not pageable, we have to get it for all possible
4021 *wired = (entry->wired_count != 0);
4023 fault_type = entry->protection;
4025 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
4027 * Fail if the entry was copy-on-write for a write fault.
4029 if (fault_type & VM_PROT_WRITE)
4030 return (KERN_FAILURE);
4032 * We're attempting to read a copy-on-write page --
4033 * don't allow writes.
4035 prot &= ~VM_PROT_WRITE;
4039 * Fail if an object should be created.
4041 if (entry->object.vm_object == NULL && !map->system_map)
4042 return (KERN_FAILURE);
4045 * Return the object/offset from this entry. If the entry was
4046 * copy-on-write or empty, it has been fixed up.
4048 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
4049 *object = entry->object.vm_object;
4052 return (KERN_SUCCESS);
4056 * vm_map_lookup_done:
4058 * Releases locks acquired by a vm_map_lookup
4059 * (according to the handle returned by that lookup).
4062 vm_map_lookup_done(vm_map_t map, vm_map_entry_t entry)
4065 * Unlock the main-level map
4067 vm_map_unlock_read(map);
4070 #include "opt_ddb.h"
4072 #include <sys/kernel.h>
4074 #include <ddb/ddb.h>
4077 vm_map_print(vm_map_t map)
4079 vm_map_entry_t entry;
4081 db_iprintf("Task map %p: pmap=%p, nentries=%d, version=%u\n",
4083 (void *)map->pmap, map->nentries, map->timestamp);
4086 for (entry = map->header.next; entry != &map->header;
4087 entry = entry->next) {
4088 db_iprintf("map entry %p: start=%p, end=%p\n",
4089 (void *)entry, (void *)entry->start, (void *)entry->end);
4091 static char *inheritance_name[4] =
4092 {"share", "copy", "none", "donate_copy"};
4094 db_iprintf(" prot=%x/%x/%s",
4096 entry->max_protection,
4097 inheritance_name[(int)(unsigned char)entry->inheritance]);
4098 if (entry->wired_count != 0)
4099 db_printf(", wired");
4101 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
4102 db_printf(", share=%p, offset=0x%jx\n",
4103 (void *)entry->object.sub_map,
4104 (uintmax_t)entry->offset);
4105 if ((entry->prev == &map->header) ||
4106 (entry->prev->object.sub_map !=
4107 entry->object.sub_map)) {
4109 vm_map_print((vm_map_t)entry->object.sub_map);
4113 if (entry->cred != NULL)
4114 db_printf(", ruid %d", entry->cred->cr_ruid);
4115 db_printf(", object=%p, offset=0x%jx",
4116 (void *)entry->object.vm_object,
4117 (uintmax_t)entry->offset);
4118 if (entry->object.vm_object && entry->object.vm_object->cred)
4119 db_printf(", obj ruid %d charge %jx",
4120 entry->object.vm_object->cred->cr_ruid,
4121 (uintmax_t)entry->object.vm_object->charge);
4122 if (entry->eflags & MAP_ENTRY_COW)
4123 db_printf(", copy (%s)",
4124 (entry->eflags & MAP_ENTRY_NEEDS_COPY) ? "needed" : "done");
4127 if ((entry->prev == &map->header) ||
4128 (entry->prev->object.vm_object !=
4129 entry->object.vm_object)) {
4131 vm_object_print((db_expr_t)(intptr_t)
4132 entry->object.vm_object,
4141 DB_SHOW_COMMAND(map, map)
4145 db_printf("usage: show map <addr>\n");
4148 vm_map_print((vm_map_t)addr);
4151 DB_SHOW_COMMAND(procvm, procvm)
4156 p = (struct proc *) addr;
4161 db_printf("p = %p, vmspace = %p, map = %p, pmap = %p\n",
4162 (void *)p, (void *)p->p_vmspace, (void *)&p->p_vmspace->vm_map,
4163 (void *)vmspace_pmap(p->p_vmspace));
4165 vm_map_print((vm_map_t)&p->p_vmspace->vm_map);