2 * Copyright (c) 1991, 1993
3 * The Regents of the University of California. All rights reserved.
5 * This code is derived from software contributed to Berkeley by
6 * The Mach Operating System project at Carnegie-Mellon University.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 4. Neither the name of the University nor the names of its contributors
17 * may be used to endorse or promote products derived from this software
18 * without specific prior written permission.
20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * from: @(#)vm_map.c 8.3 (Berkeley) 1/12/94
35 * Copyright (c) 1987, 1990 Carnegie-Mellon University.
36 * All rights reserved.
38 * Authors: Avadis Tevanian, Jr., Michael Wayne Young
40 * Permission to use, copy, modify and distribute this software and
41 * its documentation is hereby granted, provided that both the copyright
42 * notice and this permission notice appear in all copies of the
43 * software, derivative works or modified versions, and any portions
44 * thereof, and that both notices appear in supporting documentation.
46 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
47 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
48 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
50 * Carnegie Mellon requests users of this software to return to
52 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
53 * School of Computer Science
54 * Carnegie Mellon University
55 * Pittsburgh PA 15213-3890
57 * any improvements or extensions that they make and grant Carnegie the
58 * rights to redistribute these changes.
62 * Virtual memory mapping module.
65 #include <sys/cdefs.h>
66 __FBSDID("$FreeBSD$");
68 #include <sys/param.h>
69 #include <sys/systm.h>
70 #include <sys/kernel.h>
73 #include <sys/mutex.h>
75 #include <sys/vmmeter.h>
77 #include <sys/vnode.h>
78 #include <sys/racct.h>
79 #include <sys/resourcevar.h>
81 #include <sys/sysctl.h>
82 #include <sys/sysent.h>
86 #include <vm/vm_param.h>
88 #include <vm/vm_map.h>
89 #include <vm/vm_page.h>
90 #include <vm/vm_object.h>
91 #include <vm/vm_pager.h>
92 #include <vm/vm_kern.h>
93 #include <vm/vm_extern.h>
94 #include <vm/vnode_pager.h>
95 #include <vm/swap_pager.h>
99 * Virtual memory maps provide for the mapping, protection,
100 * and sharing of virtual memory objects. In addition,
101 * this module provides for an efficient virtual copy of
102 * memory from one map to another.
104 * Synchronization is required prior to most operations.
106 * Maps consist of an ordered doubly-linked list of simple
107 * entries; a self-adjusting binary search tree of these
108 * entries is used to speed up lookups.
110 * Since portions of maps are specified by start/end addresses,
111 * which may not align with existing map entries, all
112 * routines merely "clip" entries to these start/end values.
113 * [That is, an entry is split into two, bordering at a
114 * start or end value.] Note that these clippings may not
115 * always be necessary (as the two resulting entries are then
116 * not changed); however, the clipping is done for convenience.
118 * As mentioned above, virtual copy operations are performed
119 * by copying VM object references from one map to
120 * another, and then marking both regions as copy-on-write.
123 static struct mtx map_sleep_mtx;
124 static uma_zone_t mapentzone;
125 static uma_zone_t kmapentzone;
126 static uma_zone_t mapzone;
127 static uma_zone_t vmspace_zone;
128 static struct vm_object kmapentobj;
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)
244 mtx_init(&map->system_mtx, "system map", NULL, MTX_DEF | MTX_DUPOK);
245 sx_init(&map->lock, "user map");
251 vmspace_zdtor(void *mem, int size, void *arg)
255 vm = (struct vmspace *)mem;
257 vm_map_zdtor(&vm->vm_map, sizeof(vm->vm_map), arg);
260 vm_map_zdtor(void *mem, int size, void *arg)
265 KASSERT(map->nentries == 0,
266 ("map %p nentries == %d on free.",
267 map, map->nentries));
268 KASSERT(map->size == 0,
269 ("map %p size == %lu on free.",
270 map, (unsigned long)map->size));
272 #endif /* INVARIANTS */
275 * Allocate a vmspace structure, including a vm_map and pmap,
276 * and initialize those structures. The refcnt is set to 1.
279 vmspace_alloc(min, max)
280 vm_offset_t min, max;
284 vm = uma_zalloc(vmspace_zone, M_WAITOK);
285 if (vm->vm_map.pmap == NULL && !pmap_pinit(vmspace_pmap(vm))) {
286 uma_zfree(vmspace_zone, vm);
289 CTR1(KTR_VM, "vmspace_alloc: %p", vm);
290 _vm_map_init(&vm->vm_map, vmspace_pmap(vm), min, max);
306 uma_zone_set_obj(kmapentzone, &kmapentobj, lmin(cnt.v_page_count,
307 (VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS) / PAGE_SIZE) / 8 +
308 maxproc * 2 + maxfiles);
309 vmspace_zone = uma_zcreate("VMSPACE", sizeof(struct vmspace), NULL,
315 vmspace_zinit, vmspace_zfini, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
319 vmspace_container_reset(struct proc *p)
324 racct_set(p, RACCT_DATA, 0);
325 racct_set(p, RACCT_STACK, 0);
326 racct_set(p, RACCT_RSS, 0);
327 racct_set(p, RACCT_MEMLOCK, 0);
328 racct_set(p, RACCT_VMEM, 0);
334 vmspace_dofree(struct vmspace *vm)
337 CTR1(KTR_VM, "vmspace_free: %p", vm);
340 * Make sure any SysV shm is freed, it might not have been in
346 * Lock the map, to wait out all other references to it.
347 * Delete all of the mappings and pages they hold, then call
348 * the pmap module to reclaim anything left.
350 (void)vm_map_remove(&vm->vm_map, vm->vm_map.min_offset,
351 vm->vm_map.max_offset);
353 pmap_release(vmspace_pmap(vm));
354 vm->vm_map.pmap = NULL;
355 uma_zfree(vmspace_zone, vm);
359 vmspace_free(struct vmspace *vm)
362 if (vm->vm_refcnt == 0)
363 panic("vmspace_free: attempt to free already freed vmspace");
365 if (atomic_fetchadd_int(&vm->vm_refcnt, -1) == 1)
370 vmspace_exitfree(struct proc *p)
374 PROC_VMSPACE_LOCK(p);
377 PROC_VMSPACE_UNLOCK(p);
378 KASSERT(vm == &vmspace0, ("vmspace_exitfree: wrong vmspace"));
383 vmspace_exit(struct thread *td)
390 * Release user portion of address space.
391 * This releases references to vnodes,
392 * which could cause I/O if the file has been unlinked.
393 * Need to do this early enough that we can still sleep.
395 * The last exiting process to reach this point releases as
396 * much of the environment as it can. vmspace_dofree() is the
397 * slower fallback in case another process had a temporary
398 * reference to the vmspace.
403 atomic_add_int(&vmspace0.vm_refcnt, 1);
405 refcnt = vm->vm_refcnt;
406 if (refcnt > 1 && p->p_vmspace != &vmspace0) {
407 /* Switch now since other proc might free vmspace */
408 PROC_VMSPACE_LOCK(p);
409 p->p_vmspace = &vmspace0;
410 PROC_VMSPACE_UNLOCK(p);
413 } while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt - 1));
415 if (p->p_vmspace != vm) {
416 /* vmspace not yet freed, switch back */
417 PROC_VMSPACE_LOCK(p);
419 PROC_VMSPACE_UNLOCK(p);
422 pmap_remove_pages(vmspace_pmap(vm));
423 /* Switch now since this proc will free vmspace */
424 PROC_VMSPACE_LOCK(p);
425 p->p_vmspace = &vmspace0;
426 PROC_VMSPACE_UNLOCK(p);
430 vmspace_container_reset(p);
433 /* Acquire reference to vmspace owned by another process. */
436 vmspace_acquire_ref(struct proc *p)
441 PROC_VMSPACE_LOCK(p);
444 PROC_VMSPACE_UNLOCK(p);
448 refcnt = vm->vm_refcnt;
449 if (refcnt <= 0) { /* Avoid 0->1 transition */
450 PROC_VMSPACE_UNLOCK(p);
453 } while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt + 1));
454 if (vm != p->p_vmspace) {
455 PROC_VMSPACE_UNLOCK(p);
459 PROC_VMSPACE_UNLOCK(p);
464 _vm_map_lock(vm_map_t map, const char *file, int line)
468 mtx_lock_flags_(&map->system_mtx, 0, file, line);
470 sx_xlock_(&map->lock, file, line);
475 vm_map_process_deferred(void)
478 vm_map_entry_t entry;
482 while ((entry = td->td_map_def_user) != NULL) {
483 td->td_map_def_user = entry->next;
484 if ((entry->eflags & MAP_ENTRY_VN_WRITECNT) != 0) {
486 * Decrement the object's writemappings and
487 * possibly the vnode's v_writecount.
489 KASSERT((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0,
490 ("Submap with writecount"));
491 object = entry->object.vm_object;
492 KASSERT(object != NULL, ("No object for writecount"));
493 vnode_pager_release_writecount(object, entry->start,
496 vm_map_entry_deallocate(entry, FALSE);
501 _vm_map_unlock(vm_map_t map, const char *file, int line)
505 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
507 sx_xunlock_(&map->lock, file, line);
508 vm_map_process_deferred();
513 _vm_map_lock_read(vm_map_t map, const char *file, int line)
517 mtx_lock_flags_(&map->system_mtx, 0, file, line);
519 sx_slock_(&map->lock, file, line);
523 _vm_map_unlock_read(vm_map_t map, const char *file, int line)
527 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
529 sx_sunlock_(&map->lock, file, line);
530 vm_map_process_deferred();
535 _vm_map_trylock(vm_map_t map, const char *file, int line)
539 error = map->system_map ?
540 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) :
541 !sx_try_xlock_(&map->lock, file, line);
548 _vm_map_trylock_read(vm_map_t map, const char *file, int line)
552 error = map->system_map ?
553 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) :
554 !sx_try_slock_(&map->lock, file, line);
559 * _vm_map_lock_upgrade: [ internal use only ]
561 * Tries to upgrade a read (shared) lock on the specified map to a write
562 * (exclusive) lock. Returns the value "0" if the upgrade succeeds and a
563 * non-zero value if the upgrade fails. If the upgrade fails, the map is
564 * returned without a read or write lock held.
566 * Requires that the map be read locked.
569 _vm_map_lock_upgrade(vm_map_t map, const char *file, int line)
571 unsigned int last_timestamp;
573 if (map->system_map) {
574 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
576 if (!sx_try_upgrade_(&map->lock, file, line)) {
577 last_timestamp = map->timestamp;
578 sx_sunlock_(&map->lock, file, line);
579 vm_map_process_deferred();
581 * If the map's timestamp does not change while the
582 * map is unlocked, then the upgrade succeeds.
584 sx_xlock_(&map->lock, file, line);
585 if (last_timestamp != map->timestamp) {
586 sx_xunlock_(&map->lock, file, line);
596 _vm_map_lock_downgrade(vm_map_t map, const char *file, int line)
599 if (map->system_map) {
600 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
602 sx_downgrade_(&map->lock, file, line);
608 * Returns a non-zero value if the caller holds a write (exclusive) lock
609 * on the specified map and the value "0" otherwise.
612 vm_map_locked(vm_map_t map)
616 return (mtx_owned(&map->system_mtx));
618 return (sx_xlocked(&map->lock));
623 _vm_map_assert_locked(vm_map_t map, const char *file, int line)
627 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
629 sx_assert_(&map->lock, SA_XLOCKED, file, line);
632 #define VM_MAP_ASSERT_LOCKED(map) \
633 _vm_map_assert_locked(map, LOCK_FILE, LOCK_LINE)
635 #define VM_MAP_ASSERT_LOCKED(map)
639 * _vm_map_unlock_and_wait:
641 * Atomically releases the lock on the specified map and puts the calling
642 * thread to sleep. The calling thread will remain asleep until either
643 * vm_map_wakeup() is performed on the map or the specified timeout is
646 * WARNING! This function does not perform deferred deallocations of
647 * objects and map entries. Therefore, the calling thread is expected to
648 * reacquire the map lock after reawakening and later perform an ordinary
649 * unlock operation, such as vm_map_unlock(), before completing its
650 * operation on the map.
653 _vm_map_unlock_and_wait(vm_map_t map, int timo, const char *file, int line)
656 mtx_lock(&map_sleep_mtx);
658 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
660 sx_xunlock_(&map->lock, file, line);
661 return (msleep(&map->root, &map_sleep_mtx, PDROP | PVM, "vmmaps",
668 * Awaken any threads that have slept on the map using
669 * vm_map_unlock_and_wait().
672 vm_map_wakeup(vm_map_t map)
676 * Acquire and release map_sleep_mtx to prevent a wakeup()
677 * from being performed (and lost) between the map unlock
678 * and the msleep() in _vm_map_unlock_and_wait().
680 mtx_lock(&map_sleep_mtx);
681 mtx_unlock(&map_sleep_mtx);
686 vm_map_busy(vm_map_t map)
689 VM_MAP_ASSERT_LOCKED(map);
694 vm_map_unbusy(vm_map_t map)
697 VM_MAP_ASSERT_LOCKED(map);
698 KASSERT(map->busy, ("vm_map_unbusy: not busy"));
699 if (--map->busy == 0 && (map->flags & MAP_BUSY_WAKEUP)) {
700 vm_map_modflags(map, 0, MAP_BUSY_WAKEUP);
706 vm_map_wait_busy(vm_map_t map)
709 VM_MAP_ASSERT_LOCKED(map);
711 vm_map_modflags(map, MAP_BUSY_WAKEUP, 0);
713 msleep(&map->busy, &map->system_mtx, 0, "mbusy", 0);
715 sx_sleep(&map->busy, &map->lock, 0, "mbusy", 0);
721 vmspace_resident_count(struct vmspace *vmspace)
723 return pmap_resident_count(vmspace_pmap(vmspace));
727 vmspace_wired_count(struct vmspace *vmspace)
729 return pmap_wired_count(vmspace_pmap(vmspace));
735 * Creates and returns a new empty VM map with
736 * the given physical map structure, and having
737 * the given lower and upper address bounds.
740 vm_map_create(pmap_t pmap, vm_offset_t min, vm_offset_t max)
744 result = uma_zalloc(mapzone, M_WAITOK);
745 CTR1(KTR_VM, "vm_map_create: %p", result);
746 _vm_map_init(result, pmap, min, max);
751 * Initialize an existing vm_map structure
752 * such as that in the vmspace structure.
755 _vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max)
758 map->header.next = map->header.prev = &map->header;
759 map->needs_wakeup = FALSE;
762 map->min_offset = min;
763 map->max_offset = max;
771 vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max)
774 _vm_map_init(map, pmap, min, max);
775 mtx_init(&map->system_mtx, "system map", NULL, MTX_DEF | MTX_DUPOK);
776 sx_init(&map->lock, "user map");
780 * vm_map_entry_dispose: [ internal use only ]
782 * Inverse of vm_map_entry_create.
785 vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry)
787 uma_zfree(map->system_map ? kmapentzone : mapentzone, entry);
791 * vm_map_entry_create: [ internal use only ]
793 * Allocates a VM map entry for insertion.
794 * No entry fields are filled in.
796 static vm_map_entry_t
797 vm_map_entry_create(vm_map_t map)
799 vm_map_entry_t new_entry;
802 new_entry = uma_zalloc(kmapentzone, M_NOWAIT);
804 new_entry = uma_zalloc(mapentzone, M_WAITOK);
805 if (new_entry == NULL)
806 panic("vm_map_entry_create: kernel resources exhausted");
811 * vm_map_entry_set_behavior:
813 * Set the expected access behavior, either normal, random, or
817 vm_map_entry_set_behavior(vm_map_entry_t entry, u_char behavior)
819 entry->eflags = (entry->eflags & ~MAP_ENTRY_BEHAV_MASK) |
820 (behavior & MAP_ENTRY_BEHAV_MASK);
824 * vm_map_entry_set_max_free:
826 * Set the max_free field in a vm_map_entry.
829 vm_map_entry_set_max_free(vm_map_entry_t entry)
832 entry->max_free = entry->adj_free;
833 if (entry->left != NULL && entry->left->max_free > entry->max_free)
834 entry->max_free = entry->left->max_free;
835 if (entry->right != NULL && entry->right->max_free > entry->max_free)
836 entry->max_free = entry->right->max_free;
840 * vm_map_entry_splay:
842 * The Sleator and Tarjan top-down splay algorithm with the
843 * following variation. Max_free must be computed bottom-up, so
844 * on the downward pass, maintain the left and right spines in
845 * reverse order. Then, make a second pass up each side to fix
846 * the pointers and compute max_free. The time bound is O(log n)
849 * The new root is the vm_map_entry containing "addr", or else an
850 * adjacent entry (lower or higher) if addr is not in the tree.
852 * The map must be locked, and leaves it so.
854 * Returns: the new root.
856 static vm_map_entry_t
857 vm_map_entry_splay(vm_offset_t addr, vm_map_entry_t root)
859 vm_map_entry_t llist, rlist;
860 vm_map_entry_t ltree, rtree;
863 /* Special case of empty tree. */
868 * Pass One: Splay down the tree until we find addr or a NULL
869 * pointer where addr would go. llist and rlist are the two
870 * sides in reverse order (bottom-up), with llist linked by
871 * the right pointer and rlist linked by the left pointer in
872 * the vm_map_entry. Wait until Pass Two to set max_free on
878 /* root is never NULL in here. */
879 if (addr < root->start) {
883 if (addr < y->start && y->left != NULL) {
884 /* Rotate right and put y on rlist. */
885 root->left = y->right;
887 vm_map_entry_set_max_free(root);
892 /* Put root on rlist. */
897 } else if (addr >= root->end) {
901 if (addr >= y->end && y->right != NULL) {
902 /* Rotate left and put y on llist. */
903 root->right = y->left;
905 vm_map_entry_set_max_free(root);
910 /* Put root on llist. */
920 * Pass Two: Walk back up the two spines, flip the pointers
921 * and set max_free. The subtrees of the root go at the
922 * bottom of llist and rlist.
925 while (llist != NULL) {
927 llist->right = ltree;
928 vm_map_entry_set_max_free(llist);
933 while (rlist != NULL) {
936 vm_map_entry_set_max_free(rlist);
942 * Final assembly: add ltree and rtree as subtrees of root.
946 vm_map_entry_set_max_free(root);
952 * vm_map_entry_{un,}link:
954 * Insert/remove entries from maps.
957 vm_map_entry_link(vm_map_t map,
958 vm_map_entry_t after_where,
959 vm_map_entry_t entry)
963 "vm_map_entry_link: map %p, nentries %d, entry %p, after %p", map,
964 map->nentries, entry, after_where);
965 VM_MAP_ASSERT_LOCKED(map);
967 entry->prev = after_where;
968 entry->next = after_where->next;
969 entry->next->prev = entry;
970 after_where->next = entry;
972 if (after_where != &map->header) {
973 if (after_where != map->root)
974 vm_map_entry_splay(after_where->start, map->root);
975 entry->right = after_where->right;
976 entry->left = after_where;
977 after_where->right = NULL;
978 after_where->adj_free = entry->start - after_where->end;
979 vm_map_entry_set_max_free(after_where);
981 entry->right = map->root;
984 entry->adj_free = (entry->next == &map->header ? map->max_offset :
985 entry->next->start) - entry->end;
986 vm_map_entry_set_max_free(entry);
991 vm_map_entry_unlink(vm_map_t map,
992 vm_map_entry_t entry)
994 vm_map_entry_t next, prev, root;
996 VM_MAP_ASSERT_LOCKED(map);
997 if (entry != map->root)
998 vm_map_entry_splay(entry->start, map->root);
999 if (entry->left == NULL)
1000 root = entry->right;
1002 root = vm_map_entry_splay(entry->start, entry->left);
1003 root->right = entry->right;
1004 root->adj_free = (entry->next == &map->header ? map->max_offset :
1005 entry->next->start) - root->end;
1006 vm_map_entry_set_max_free(root);
1015 CTR3(KTR_VM, "vm_map_entry_unlink: map %p, nentries %d, entry %p", map,
1016 map->nentries, entry);
1020 * vm_map_entry_resize_free:
1022 * Recompute the amount of free space following a vm_map_entry
1023 * and propagate that value up the tree. Call this function after
1024 * resizing a map entry in-place, that is, without a call to
1025 * vm_map_entry_link() or _unlink().
1027 * The map must be locked, and leaves it so.
1030 vm_map_entry_resize_free(vm_map_t map, vm_map_entry_t entry)
1034 * Using splay trees without parent pointers, propagating
1035 * max_free up the tree is done by moving the entry to the
1036 * root and making the change there.
1038 if (entry != map->root)
1039 map->root = vm_map_entry_splay(entry->start, map->root);
1041 entry->adj_free = (entry->next == &map->header ? map->max_offset :
1042 entry->next->start) - entry->end;
1043 vm_map_entry_set_max_free(entry);
1047 * vm_map_lookup_entry: [ internal use only ]
1049 * Finds the map entry containing (or
1050 * immediately preceding) the specified address
1051 * in the given map; the entry is returned
1052 * in the "entry" parameter. The boolean
1053 * result indicates whether the address is
1054 * actually contained in the map.
1057 vm_map_lookup_entry(
1059 vm_offset_t address,
1060 vm_map_entry_t *entry) /* OUT */
1066 * If the map is empty, then the map entry immediately preceding
1067 * "address" is the map's header.
1071 *entry = &map->header;
1072 else if (address >= cur->start && cur->end > address) {
1075 } else if ((locked = vm_map_locked(map)) ||
1076 sx_try_upgrade(&map->lock)) {
1078 * Splay requires a write lock on the map. However, it only
1079 * restructures the binary search tree; it does not otherwise
1080 * change the map. Thus, the map's timestamp need not change
1081 * on a temporary upgrade.
1083 map->root = cur = vm_map_entry_splay(address, cur);
1085 sx_downgrade(&map->lock);
1088 * If "address" is contained within a map entry, the new root
1089 * is that map entry. Otherwise, the new root is a map entry
1090 * immediately before or after "address".
1092 if (address >= cur->start) {
1094 if (cur->end > address)
1100 * Since the map is only locked for read access, perform a
1101 * standard binary search tree lookup for "address".
1104 if (address < cur->start) {
1105 if (cur->left == NULL) {
1110 } else if (cur->end > address) {
1114 if (cur->right == NULL) {
1127 * Inserts the given whole VM object into the target
1128 * map at the specified address range. The object's
1129 * size should match that of the address range.
1131 * Requires that the map be locked, and leaves it so.
1133 * If object is non-NULL, ref count must be bumped by caller
1134 * prior to making call to account for the new entry.
1137 vm_map_insert(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1138 vm_offset_t start, vm_offset_t end, vm_prot_t prot, vm_prot_t max,
1141 vm_map_entry_t new_entry;
1142 vm_map_entry_t prev_entry;
1143 vm_map_entry_t temp_entry;
1144 vm_eflags_t protoeflags;
1146 vm_inherit_t inheritance;
1147 boolean_t charge_prev_obj;
1149 VM_MAP_ASSERT_LOCKED(map);
1152 * Check that the start and end points are not bogus.
1154 if ((start < map->min_offset) || (end > map->max_offset) ||
1156 return (KERN_INVALID_ADDRESS);
1159 * Find the entry prior to the proposed starting address; if it's part
1160 * of an existing entry, this range is bogus.
1162 if (vm_map_lookup_entry(map, start, &temp_entry))
1163 return (KERN_NO_SPACE);
1165 prev_entry = temp_entry;
1168 * Assert that the next entry doesn't overlap the end point.
1170 if ((prev_entry->next != &map->header) &&
1171 (prev_entry->next->start < end))
1172 return (KERN_NO_SPACE);
1175 charge_prev_obj = FALSE;
1177 if (cow & MAP_COPY_ON_WRITE)
1178 protoeflags |= MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY;
1180 if (cow & MAP_NOFAULT) {
1181 protoeflags |= MAP_ENTRY_NOFAULT;
1183 KASSERT(object == NULL,
1184 ("vm_map_insert: paradoxical MAP_NOFAULT request"));
1186 if (cow & MAP_DISABLE_SYNCER)
1187 protoeflags |= MAP_ENTRY_NOSYNC;
1188 if (cow & MAP_DISABLE_COREDUMP)
1189 protoeflags |= MAP_ENTRY_NOCOREDUMP;
1190 if (cow & MAP_VN_WRITECOUNT)
1191 protoeflags |= MAP_ENTRY_VN_WRITECNT;
1192 if (cow & MAP_INHERIT_SHARE)
1193 inheritance = VM_INHERIT_SHARE;
1195 inheritance = VM_INHERIT_DEFAULT;
1198 KASSERT((object != kmem_object && object != kernel_object) ||
1199 ((object == kmem_object || object == kernel_object) &&
1200 !(protoeflags & MAP_ENTRY_NEEDS_COPY)),
1201 ("kmem or kernel object and cow"));
1202 if (cow & (MAP_ACC_NO_CHARGE | MAP_NOFAULT))
1204 if ((cow & MAP_ACC_CHARGED) || ((prot & VM_PROT_WRITE) &&
1205 ((protoeflags & MAP_ENTRY_NEEDS_COPY) || object == NULL))) {
1206 if (!(cow & MAP_ACC_CHARGED) && !swap_reserve(end - start))
1207 return (KERN_RESOURCE_SHORTAGE);
1208 KASSERT(object == NULL || (protoeflags & MAP_ENTRY_NEEDS_COPY) ||
1209 object->cred == NULL,
1210 ("OVERCOMMIT: vm_map_insert o %p", object));
1211 cred = curthread->td_ucred;
1213 if (object == NULL && !(protoeflags & MAP_ENTRY_NEEDS_COPY))
1214 charge_prev_obj = TRUE;
1218 /* Expand the kernel pmap, if necessary. */
1219 if (map == kernel_map && end > kernel_vm_end)
1220 pmap_growkernel(end);
1221 if (object != NULL) {
1223 * OBJ_ONEMAPPING must be cleared unless this mapping
1224 * is trivially proven to be the only mapping for any
1225 * of the object's pages. (Object granularity
1226 * reference counting is insufficient to recognize
1227 * aliases with precision.)
1229 VM_OBJECT_LOCK(object);
1230 if (object->ref_count > 1 || object->shadow_count != 0)
1231 vm_object_clear_flag(object, OBJ_ONEMAPPING);
1232 VM_OBJECT_UNLOCK(object);
1234 else if ((prev_entry != &map->header) &&
1235 (prev_entry->eflags == protoeflags) &&
1236 (prev_entry->end == start) &&
1237 (prev_entry->wired_count == 0) &&
1238 (prev_entry->cred == cred ||
1239 (prev_entry->object.vm_object != NULL &&
1240 (prev_entry->object.vm_object->cred == cred))) &&
1241 vm_object_coalesce(prev_entry->object.vm_object,
1243 (vm_size_t)(prev_entry->end - prev_entry->start),
1244 (vm_size_t)(end - prev_entry->end), charge_prev_obj)) {
1246 * We were able to extend the object. Determine if we
1247 * can extend the previous map entry to include the
1248 * new range as well.
1250 if ((prev_entry->inheritance == inheritance) &&
1251 (prev_entry->protection == prot) &&
1252 (prev_entry->max_protection == max)) {
1253 map->size += (end - prev_entry->end);
1254 prev_entry->end = end;
1255 vm_map_entry_resize_free(map, prev_entry);
1256 vm_map_simplify_entry(map, prev_entry);
1259 return (KERN_SUCCESS);
1263 * If we can extend the object but cannot extend the
1264 * map entry, we have to create a new map entry. We
1265 * must bump the ref count on the extended object to
1266 * account for it. object may be NULL.
1268 object = prev_entry->object.vm_object;
1269 offset = prev_entry->offset +
1270 (prev_entry->end - prev_entry->start);
1271 vm_object_reference(object);
1272 if (cred != NULL && object != NULL && object->cred != NULL &&
1273 !(prev_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
1274 /* Object already accounts for this uid. */
1281 * NOTE: if conditionals fail, object can be NULL here. This occurs
1282 * in things like the buffer map where we manage kva but do not manage
1287 * Create a new entry
1289 new_entry = vm_map_entry_create(map);
1290 new_entry->start = start;
1291 new_entry->end = end;
1292 new_entry->cred = NULL;
1294 new_entry->eflags = protoeflags;
1295 new_entry->object.vm_object = object;
1296 new_entry->offset = offset;
1297 new_entry->avail_ssize = 0;
1299 new_entry->inheritance = inheritance;
1300 new_entry->protection = prot;
1301 new_entry->max_protection = max;
1302 new_entry->wired_count = 0;
1304 KASSERT(cred == NULL || !ENTRY_CHARGED(new_entry),
1305 ("OVERCOMMIT: vm_map_insert leaks vm_map %p", new_entry));
1306 new_entry->cred = cred;
1309 * Insert the new entry into the list
1311 vm_map_entry_link(map, prev_entry, new_entry);
1312 map->size += new_entry->end - new_entry->start;
1315 * It may be possible to merge the new entry with the next and/or
1316 * previous entries. However, due to MAP_STACK_* being a hack, a
1317 * panic can result from merging such entries.
1319 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0)
1320 vm_map_simplify_entry(map, new_entry);
1322 if (cow & (MAP_PREFAULT|MAP_PREFAULT_PARTIAL)) {
1323 vm_map_pmap_enter(map, start, prot,
1324 object, OFF_TO_IDX(offset), end - start,
1325 cow & MAP_PREFAULT_PARTIAL);
1328 return (KERN_SUCCESS);
1334 * Find the first fit (lowest VM address) for "length" free bytes
1335 * beginning at address >= start in the given map.
1337 * In a vm_map_entry, "adj_free" is the amount of free space
1338 * adjacent (higher address) to this entry, and "max_free" is the
1339 * maximum amount of contiguous free space in its subtree. This
1340 * allows finding a free region in one path down the tree, so
1341 * O(log n) amortized with splay trees.
1343 * The map must be locked, and leaves it so.
1345 * Returns: 0 on success, and starting address in *addr,
1346 * 1 if insufficient space.
1349 vm_map_findspace(vm_map_t map, vm_offset_t start, vm_size_t length,
1350 vm_offset_t *addr) /* OUT */
1352 vm_map_entry_t entry;
1356 * Request must fit within min/max VM address and must avoid
1359 if (start < map->min_offset)
1360 start = map->min_offset;
1361 if (start + length > map->max_offset || start + length < start)
1364 /* Empty tree means wide open address space. */
1365 if (map->root == NULL) {
1371 * After splay, if start comes before root node, then there
1372 * must be a gap from start to the root.
1374 map->root = vm_map_entry_splay(start, map->root);
1375 if (start + length <= map->root->start) {
1381 * Root is the last node that might begin its gap before
1382 * start, and this is the last comparison where address
1383 * wrap might be a problem.
1385 st = (start > map->root->end) ? start : map->root->end;
1386 if (length <= map->root->end + map->root->adj_free - st) {
1391 /* With max_free, can immediately tell if no solution. */
1392 entry = map->root->right;
1393 if (entry == NULL || length > entry->max_free)
1397 * Search the right subtree in the order: left subtree, root,
1398 * right subtree (first fit). The previous splay implies that
1399 * all regions in the right subtree have addresses > start.
1401 while (entry != NULL) {
1402 if (entry->left != NULL && entry->left->max_free >= length)
1403 entry = entry->left;
1404 else if (entry->adj_free >= length) {
1408 entry = entry->right;
1411 /* Can't get here, so panic if we do. */
1412 panic("vm_map_findspace: max_free corrupt");
1416 vm_map_fixed(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1417 vm_offset_t start, vm_size_t length, vm_prot_t prot,
1418 vm_prot_t max, int cow)
1423 end = start + length;
1425 VM_MAP_RANGE_CHECK(map, start, end);
1426 (void) vm_map_delete(map, start, end);
1427 result = vm_map_insert(map, object, offset, start, end, prot,
1434 * vm_map_find finds an unallocated region in the target address
1435 * map with the given length. The search is defined to be
1436 * first-fit from the specified address; the region found is
1437 * returned in the same parameter.
1439 * If object is non-NULL, ref count must be bumped by caller
1440 * prior to making call to account for the new entry.
1443 vm_map_find(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1444 vm_offset_t *addr, /* IN/OUT */
1445 vm_size_t length, int find_space, vm_prot_t prot,
1446 vm_prot_t max, int cow)
1454 if (find_space != VMFS_NO_SPACE) {
1455 if (vm_map_findspace(map, start, length, addr)) {
1457 return (KERN_NO_SPACE);
1459 switch (find_space) {
1460 case VMFS_ALIGNED_SPACE:
1461 pmap_align_superpage(object, offset, addr,
1464 #ifdef VMFS_TLB_ALIGNED_SPACE
1465 case VMFS_TLB_ALIGNED_SPACE:
1466 pmap_align_tlb(addr);
1475 result = vm_map_insert(map, object, offset, start, start +
1476 length, prot, max, cow);
1477 } while (result == KERN_NO_SPACE && (find_space == VMFS_ALIGNED_SPACE
1478 #ifdef VMFS_TLB_ALIGNED_SPACE
1479 || find_space == VMFS_TLB_ALIGNED_SPACE
1487 * vm_map_simplify_entry:
1489 * Simplify the given map entry by merging with either neighbor. This
1490 * routine also has the ability to merge with both neighbors.
1492 * The map must be locked.
1494 * This routine guarentees that the passed entry remains valid (though
1495 * possibly extended). When merging, this routine may delete one or
1499 vm_map_simplify_entry(vm_map_t map, vm_map_entry_t entry)
1501 vm_map_entry_t next, prev;
1502 vm_size_t prevsize, esize;
1504 if (entry->eflags & (MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_IS_SUB_MAP))
1508 if (prev != &map->header) {
1509 prevsize = prev->end - prev->start;
1510 if ( (prev->end == entry->start) &&
1511 (prev->object.vm_object == entry->object.vm_object) &&
1512 (!prev->object.vm_object ||
1513 (prev->offset + prevsize == entry->offset)) &&
1514 (prev->eflags == entry->eflags) &&
1515 (prev->protection == entry->protection) &&
1516 (prev->max_protection == entry->max_protection) &&
1517 (prev->inheritance == entry->inheritance) &&
1518 (prev->wired_count == entry->wired_count) &&
1519 (prev->cred == entry->cred)) {
1520 vm_map_entry_unlink(map, prev);
1521 entry->start = prev->start;
1522 entry->offset = prev->offset;
1523 if (entry->prev != &map->header)
1524 vm_map_entry_resize_free(map, entry->prev);
1527 * If the backing object is a vnode object,
1528 * vm_object_deallocate() calls vrele().
1529 * However, vrele() does not lock the vnode
1530 * because the vnode has additional
1531 * references. Thus, the map lock can be kept
1532 * without causing a lock-order reversal with
1535 * Since we count the number of virtual page
1536 * mappings in object->un_pager.vnp.writemappings,
1537 * the writemappings value should not be adjusted
1538 * when the entry is disposed of.
1540 if (prev->object.vm_object)
1541 vm_object_deallocate(prev->object.vm_object);
1542 if (prev->cred != NULL)
1544 vm_map_entry_dispose(map, prev);
1549 if (next != &map->header) {
1550 esize = entry->end - entry->start;
1551 if ((entry->end == next->start) &&
1552 (next->object.vm_object == entry->object.vm_object) &&
1553 (!entry->object.vm_object ||
1554 (entry->offset + esize == next->offset)) &&
1555 (next->eflags == entry->eflags) &&
1556 (next->protection == entry->protection) &&
1557 (next->max_protection == entry->max_protection) &&
1558 (next->inheritance == entry->inheritance) &&
1559 (next->wired_count == entry->wired_count) &&
1560 (next->cred == entry->cred)) {
1561 vm_map_entry_unlink(map, next);
1562 entry->end = next->end;
1563 vm_map_entry_resize_free(map, entry);
1566 * See comment above.
1568 if (next->object.vm_object)
1569 vm_object_deallocate(next->object.vm_object);
1570 if (next->cred != NULL)
1572 vm_map_entry_dispose(map, next);
1577 * vm_map_clip_start: [ internal use only ]
1579 * Asserts that the given entry begins at or after
1580 * the specified address; if necessary,
1581 * it splits the entry into two.
1583 #define vm_map_clip_start(map, entry, startaddr) \
1585 if (startaddr > entry->start) \
1586 _vm_map_clip_start(map, entry, startaddr); \
1590 * This routine is called only when it is known that
1591 * the entry must be split.
1594 _vm_map_clip_start(vm_map_t map, vm_map_entry_t entry, vm_offset_t start)
1596 vm_map_entry_t new_entry;
1598 VM_MAP_ASSERT_LOCKED(map);
1601 * Split off the front portion -- note that we must insert the new
1602 * entry BEFORE this one, so that this entry has the specified
1605 vm_map_simplify_entry(map, entry);
1608 * If there is no object backing this entry, we might as well create
1609 * one now. If we defer it, an object can get created after the map
1610 * is clipped, and individual objects will be created for the split-up
1611 * map. This is a bit of a hack, but is also about the best place to
1612 * put this improvement.
1614 if (entry->object.vm_object == NULL && !map->system_map) {
1616 object = vm_object_allocate(OBJT_DEFAULT,
1617 atop(entry->end - entry->start));
1618 entry->object.vm_object = object;
1620 if (entry->cred != NULL) {
1621 object->cred = entry->cred;
1622 object->charge = entry->end - entry->start;
1625 } else if (entry->object.vm_object != NULL &&
1626 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) &&
1627 entry->cred != NULL) {
1628 VM_OBJECT_LOCK(entry->object.vm_object);
1629 KASSERT(entry->object.vm_object->cred == NULL,
1630 ("OVERCOMMIT: vm_entry_clip_start: both cred e %p", entry));
1631 entry->object.vm_object->cred = entry->cred;
1632 entry->object.vm_object->charge = entry->end - entry->start;
1633 VM_OBJECT_UNLOCK(entry->object.vm_object);
1637 new_entry = vm_map_entry_create(map);
1638 *new_entry = *entry;
1640 new_entry->end = start;
1641 entry->offset += (start - entry->start);
1642 entry->start = start;
1643 if (new_entry->cred != NULL)
1644 crhold(entry->cred);
1646 vm_map_entry_link(map, entry->prev, new_entry);
1648 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1649 vm_object_reference(new_entry->object.vm_object);
1651 * The object->un_pager.vnp.writemappings for the
1652 * object of MAP_ENTRY_VN_WRITECNT type entry shall be
1653 * kept as is here. The virtual pages are
1654 * re-distributed among the clipped entries, so the sum is
1661 * vm_map_clip_end: [ internal use only ]
1663 * Asserts that the given entry ends at or before
1664 * the specified address; if necessary,
1665 * it splits the entry into two.
1667 #define vm_map_clip_end(map, entry, endaddr) \
1669 if ((endaddr) < (entry->end)) \
1670 _vm_map_clip_end((map), (entry), (endaddr)); \
1674 * This routine is called only when it is known that
1675 * the entry must be split.
1678 _vm_map_clip_end(vm_map_t map, vm_map_entry_t entry, vm_offset_t end)
1680 vm_map_entry_t new_entry;
1682 VM_MAP_ASSERT_LOCKED(map);
1685 * If there is no object backing this entry, we might as well create
1686 * one now. If we defer it, an object can get created after the map
1687 * is clipped, and individual objects will be created for the split-up
1688 * map. This is a bit of a hack, but is also about the best place to
1689 * put this improvement.
1691 if (entry->object.vm_object == NULL && !map->system_map) {
1693 object = vm_object_allocate(OBJT_DEFAULT,
1694 atop(entry->end - entry->start));
1695 entry->object.vm_object = object;
1697 if (entry->cred != NULL) {
1698 object->cred = entry->cred;
1699 object->charge = entry->end - entry->start;
1702 } else if (entry->object.vm_object != NULL &&
1703 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) &&
1704 entry->cred != NULL) {
1705 VM_OBJECT_LOCK(entry->object.vm_object);
1706 KASSERT(entry->object.vm_object->cred == NULL,
1707 ("OVERCOMMIT: vm_entry_clip_end: both cred e %p", entry));
1708 entry->object.vm_object->cred = entry->cred;
1709 entry->object.vm_object->charge = entry->end - entry->start;
1710 VM_OBJECT_UNLOCK(entry->object.vm_object);
1715 * Create a new entry and insert it AFTER the specified entry
1717 new_entry = vm_map_entry_create(map);
1718 *new_entry = *entry;
1720 new_entry->start = entry->end = end;
1721 new_entry->offset += (end - entry->start);
1722 if (new_entry->cred != NULL)
1723 crhold(entry->cred);
1725 vm_map_entry_link(map, entry, new_entry);
1727 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1728 vm_object_reference(new_entry->object.vm_object);
1733 * vm_map_submap: [ kernel use only ]
1735 * Mark the given range as handled by a subordinate map.
1737 * This range must have been created with vm_map_find,
1738 * and no other operations may have been performed on this
1739 * range prior to calling vm_map_submap.
1741 * Only a limited number of operations can be performed
1742 * within this rage after calling vm_map_submap:
1744 * [Don't try vm_map_copy!]
1746 * To remove a submapping, one must first remove the
1747 * range from the superior map, and then destroy the
1748 * submap (if desired). [Better yet, don't try it.]
1757 vm_map_entry_t entry;
1758 int result = KERN_INVALID_ARGUMENT;
1762 VM_MAP_RANGE_CHECK(map, start, end);
1764 if (vm_map_lookup_entry(map, start, &entry)) {
1765 vm_map_clip_start(map, entry, start);
1767 entry = entry->next;
1769 vm_map_clip_end(map, entry, end);
1771 if ((entry->start == start) && (entry->end == end) &&
1772 ((entry->eflags & MAP_ENTRY_COW) == 0) &&
1773 (entry->object.vm_object == NULL)) {
1774 entry->object.sub_map = submap;
1775 entry->eflags |= MAP_ENTRY_IS_SUB_MAP;
1776 result = KERN_SUCCESS;
1784 * The maximum number of pages to map
1786 #define MAX_INIT_PT 96
1789 * vm_map_pmap_enter:
1791 * Preload read-only mappings for the given object's resident pages into
1792 * the given map. This eliminates the soft faults on process startup and
1793 * immediately after an mmap(2). Because these are speculative mappings,
1794 * cached pages are not reactivated and mapped.
1797 vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot,
1798 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags)
1801 vm_page_t p, p_start;
1802 vm_pindex_t psize, tmpidx;
1804 if ((prot & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0 || object == NULL)
1806 VM_OBJECT_LOCK(object);
1807 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) {
1808 pmap_object_init_pt(map->pmap, addr, object, pindex, size);
1814 if ((flags & MAP_PREFAULT_PARTIAL) && psize > MAX_INIT_PT &&
1815 object->resident_page_count > MAX_INIT_PT)
1818 if (psize + pindex > object->size) {
1819 if (object->size < pindex)
1821 psize = object->size - pindex;
1827 p = vm_page_find_least(object, pindex);
1829 * Assert: the variable p is either (1) the page with the
1830 * least pindex greater than or equal to the parameter pindex
1834 p != NULL && (tmpidx = p->pindex - pindex) < psize;
1835 p = TAILQ_NEXT(p, listq)) {
1837 * don't allow an madvise to blow away our really
1838 * free pages allocating pv entries.
1840 if ((flags & MAP_PREFAULT_MADVISE) &&
1841 cnt.v_free_count < cnt.v_free_reserved) {
1845 if (p->valid == VM_PAGE_BITS_ALL) {
1846 if (p_start == NULL) {
1847 start = addr + ptoa(tmpidx);
1850 } else if (p_start != NULL) {
1851 pmap_enter_object(map->pmap, start, addr +
1852 ptoa(tmpidx), p_start, prot);
1856 if (p_start != NULL)
1857 pmap_enter_object(map->pmap, start, addr + ptoa(psize),
1860 VM_OBJECT_UNLOCK(object);
1866 * Sets the protection of the specified address
1867 * region in the target map. If "set_max" is
1868 * specified, the maximum protection is to be set;
1869 * otherwise, only the current protection is affected.
1872 vm_map_protect(vm_map_t map, vm_offset_t start, vm_offset_t end,
1873 vm_prot_t new_prot, boolean_t set_max)
1875 vm_map_entry_t current, entry;
1882 VM_MAP_RANGE_CHECK(map, start, end);
1884 if (vm_map_lookup_entry(map, start, &entry)) {
1885 vm_map_clip_start(map, entry, start);
1887 entry = entry->next;
1891 * Make a first pass to check for protection violations.
1894 while ((current != &map->header) && (current->start < end)) {
1895 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
1897 return (KERN_INVALID_ARGUMENT);
1899 if ((new_prot & current->max_protection) != new_prot) {
1901 return (KERN_PROTECTION_FAILURE);
1903 current = current->next;
1908 * Do an accounting pass for private read-only mappings that
1909 * now will do cow due to allowed write (e.g. debugger sets
1910 * breakpoint on text segment)
1912 for (current = entry; (current != &map->header) &&
1913 (current->start < end); current = current->next) {
1915 vm_map_clip_end(map, current, end);
1918 ((new_prot & ~(current->protection)) & VM_PROT_WRITE) == 0 ||
1919 ENTRY_CHARGED(current)) {
1923 cred = curthread->td_ucred;
1924 obj = current->object.vm_object;
1926 if (obj == NULL || (current->eflags & MAP_ENTRY_NEEDS_COPY)) {
1927 if (!swap_reserve(current->end - current->start)) {
1929 return (KERN_RESOURCE_SHORTAGE);
1932 current->cred = cred;
1936 VM_OBJECT_LOCK(obj);
1937 if (obj->type != OBJT_DEFAULT && obj->type != OBJT_SWAP) {
1938 VM_OBJECT_UNLOCK(obj);
1943 * Charge for the whole object allocation now, since
1944 * we cannot distinguish between non-charged and
1945 * charged clipped mapping of the same object later.
1947 KASSERT(obj->charge == 0,
1948 ("vm_map_protect: object %p overcharged\n", obj));
1949 if (!swap_reserve(ptoa(obj->size))) {
1950 VM_OBJECT_UNLOCK(obj);
1952 return (KERN_RESOURCE_SHORTAGE);
1957 obj->charge = ptoa(obj->size);
1958 VM_OBJECT_UNLOCK(obj);
1962 * Go back and fix up protections. [Note that clipping is not
1963 * necessary the second time.]
1966 while ((current != &map->header) && (current->start < end)) {
1967 old_prot = current->protection;
1970 current->protection =
1971 (current->max_protection = new_prot) &
1974 current->protection = new_prot;
1976 if ((current->eflags & (MAP_ENTRY_COW | MAP_ENTRY_USER_WIRED))
1977 == (MAP_ENTRY_COW | MAP_ENTRY_USER_WIRED) &&
1978 (current->protection & VM_PROT_WRITE) != 0 &&
1979 (old_prot & VM_PROT_WRITE) == 0) {
1980 vm_fault_copy_entry(map, map, current, current, NULL);
1984 * When restricting access, update the physical map. Worry
1985 * about copy-on-write here.
1987 if ((old_prot & ~current->protection) != 0) {
1988 #define MASK(entry) (((entry)->eflags & MAP_ENTRY_COW) ? ~VM_PROT_WRITE : \
1990 pmap_protect(map->pmap, current->start,
1992 current->protection & MASK(current));
1995 vm_map_simplify_entry(map, current);
1996 current = current->next;
1999 return (KERN_SUCCESS);
2005 * This routine traverses a processes map handling the madvise
2006 * system call. Advisories are classified as either those effecting
2007 * the vm_map_entry structure, or those effecting the underlying
2017 vm_map_entry_t current, entry;
2021 * Some madvise calls directly modify the vm_map_entry, in which case
2022 * we need to use an exclusive lock on the map and we need to perform
2023 * various clipping operations. Otherwise we only need a read-lock
2028 case MADV_SEQUENTIAL:
2040 vm_map_lock_read(map);
2043 return (KERN_INVALID_ARGUMENT);
2047 * Locate starting entry and clip if necessary.
2049 VM_MAP_RANGE_CHECK(map, start, end);
2051 if (vm_map_lookup_entry(map, start, &entry)) {
2053 vm_map_clip_start(map, entry, start);
2055 entry = entry->next;
2060 * madvise behaviors that are implemented in the vm_map_entry.
2062 * We clip the vm_map_entry so that behavioral changes are
2063 * limited to the specified address range.
2065 for (current = entry;
2066 (current != &map->header) && (current->start < end);
2067 current = current->next
2069 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
2072 vm_map_clip_end(map, current, end);
2076 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_NORMAL);
2078 case MADV_SEQUENTIAL:
2079 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_SEQUENTIAL);
2082 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_RANDOM);
2085 current->eflags |= MAP_ENTRY_NOSYNC;
2088 current->eflags &= ~MAP_ENTRY_NOSYNC;
2091 current->eflags |= MAP_ENTRY_NOCOREDUMP;
2094 current->eflags &= ~MAP_ENTRY_NOCOREDUMP;
2099 vm_map_simplify_entry(map, current);
2107 * madvise behaviors that are implemented in the underlying
2110 * Since we don't clip the vm_map_entry, we have to clip
2111 * the vm_object pindex and count.
2113 for (current = entry;
2114 (current != &map->header) && (current->start < end);
2115 current = current->next
2117 vm_offset_t useStart;
2119 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
2122 pindex = OFF_TO_IDX(current->offset);
2123 count = atop(current->end - current->start);
2124 useStart = current->start;
2126 if (current->start < start) {
2127 pindex += atop(start - current->start);
2128 count -= atop(start - current->start);
2131 if (current->end > end)
2132 count -= atop(current->end - end);
2137 vm_object_madvise(current->object.vm_object,
2138 pindex, count, behav);
2139 if (behav == MADV_WILLNEED) {
2140 vm_map_pmap_enter(map,
2142 current->protection,
2143 current->object.vm_object,
2145 (count << PAGE_SHIFT),
2146 MAP_PREFAULT_MADVISE
2150 vm_map_unlock_read(map);
2159 * Sets the inheritance of the specified address
2160 * range in the target map. Inheritance
2161 * affects how the map will be shared with
2162 * child maps at the time of vmspace_fork.
2165 vm_map_inherit(vm_map_t map, vm_offset_t start, vm_offset_t end,
2166 vm_inherit_t new_inheritance)
2168 vm_map_entry_t entry;
2169 vm_map_entry_t temp_entry;
2171 switch (new_inheritance) {
2172 case VM_INHERIT_NONE:
2173 case VM_INHERIT_COPY:
2174 case VM_INHERIT_SHARE:
2177 return (KERN_INVALID_ARGUMENT);
2180 VM_MAP_RANGE_CHECK(map, start, end);
2181 if (vm_map_lookup_entry(map, start, &temp_entry)) {
2183 vm_map_clip_start(map, entry, start);
2185 entry = temp_entry->next;
2186 while ((entry != &map->header) && (entry->start < end)) {
2187 vm_map_clip_end(map, entry, end);
2188 entry->inheritance = new_inheritance;
2189 vm_map_simplify_entry(map, entry);
2190 entry = entry->next;
2193 return (KERN_SUCCESS);
2199 * Implements both kernel and user unwiring.
2202 vm_map_unwire(vm_map_t map, vm_offset_t start, vm_offset_t end,
2205 vm_map_entry_t entry, first_entry, tmp_entry;
2206 vm_offset_t saved_start;
2207 unsigned int last_timestamp;
2209 boolean_t need_wakeup, result, user_unwire;
2211 user_unwire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE;
2213 VM_MAP_RANGE_CHECK(map, start, end);
2214 if (!vm_map_lookup_entry(map, start, &first_entry)) {
2215 if (flags & VM_MAP_WIRE_HOLESOK)
2216 first_entry = first_entry->next;
2219 return (KERN_INVALID_ADDRESS);
2222 last_timestamp = map->timestamp;
2223 entry = first_entry;
2224 while (entry != &map->header && entry->start < end) {
2225 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
2227 * We have not yet clipped the entry.
2229 saved_start = (start >= entry->start) ? start :
2231 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2232 if (vm_map_unlock_and_wait(map, 0)) {
2234 * Allow interruption of user unwiring?
2238 if (last_timestamp+1 != map->timestamp) {
2240 * Look again for the entry because the map was
2241 * modified while it was unlocked.
2242 * Specifically, the entry may have been
2243 * clipped, merged, or deleted.
2245 if (!vm_map_lookup_entry(map, saved_start,
2247 if (flags & VM_MAP_WIRE_HOLESOK)
2248 tmp_entry = tmp_entry->next;
2250 if (saved_start == start) {
2252 * First_entry has been deleted.
2255 return (KERN_INVALID_ADDRESS);
2258 rv = KERN_INVALID_ADDRESS;
2262 if (entry == first_entry)
2263 first_entry = tmp_entry;
2268 last_timestamp = map->timestamp;
2271 vm_map_clip_start(map, entry, start);
2272 vm_map_clip_end(map, entry, end);
2274 * Mark the entry in case the map lock is released. (See
2277 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
2279 * Check the map for holes in the specified region.
2280 * If VM_MAP_WIRE_HOLESOK was specified, skip this check.
2282 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) &&
2283 (entry->end < end && (entry->next == &map->header ||
2284 entry->next->start > entry->end))) {
2286 rv = KERN_INVALID_ADDRESS;
2290 * If system unwiring, require that the entry is system wired.
2293 vm_map_entry_system_wired_count(entry) == 0) {
2295 rv = KERN_INVALID_ARGUMENT;
2298 entry = entry->next;
2302 need_wakeup = FALSE;
2303 if (first_entry == NULL) {
2304 result = vm_map_lookup_entry(map, start, &first_entry);
2305 if (!result && (flags & VM_MAP_WIRE_HOLESOK))
2306 first_entry = first_entry->next;
2308 KASSERT(result, ("vm_map_unwire: lookup failed"));
2310 entry = first_entry;
2311 while (entry != &map->header && entry->start < end) {
2312 if (rv == KERN_SUCCESS && (!user_unwire ||
2313 (entry->eflags & MAP_ENTRY_USER_WIRED))) {
2315 entry->eflags &= ~MAP_ENTRY_USER_WIRED;
2316 entry->wired_count--;
2317 if (entry->wired_count == 0) {
2319 * Retain the map lock.
2321 vm_fault_unwire(map, entry->start, entry->end,
2322 entry->object.vm_object != NULL &&
2323 (entry->object.vm_object->type == OBJT_DEVICE ||
2324 entry->object.vm_object->type == OBJT_SG));
2327 KASSERT(entry->eflags & MAP_ENTRY_IN_TRANSITION,
2328 ("vm_map_unwire: in-transition flag missing"));
2329 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION;
2330 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
2331 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
2334 vm_map_simplify_entry(map, entry);
2335 entry = entry->next;
2346 * Implements both kernel and user wiring.
2349 vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t end,
2352 vm_map_entry_t entry, first_entry, tmp_entry;
2353 vm_offset_t saved_end, saved_start;
2354 unsigned int last_timestamp;
2356 boolean_t fictitious, need_wakeup, result, user_wire;
2360 if (flags & VM_MAP_WIRE_WRITE)
2361 prot |= VM_PROT_WRITE;
2362 user_wire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE;
2364 VM_MAP_RANGE_CHECK(map, start, end);
2365 if (!vm_map_lookup_entry(map, start, &first_entry)) {
2366 if (flags & VM_MAP_WIRE_HOLESOK)
2367 first_entry = first_entry->next;
2370 return (KERN_INVALID_ADDRESS);
2373 last_timestamp = map->timestamp;
2374 entry = first_entry;
2375 while (entry != &map->header && entry->start < end) {
2376 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
2378 * We have not yet clipped the entry.
2380 saved_start = (start >= entry->start) ? start :
2382 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2383 if (vm_map_unlock_and_wait(map, 0)) {
2385 * Allow interruption of user wiring?
2389 if (last_timestamp + 1 != map->timestamp) {
2391 * Look again for the entry because the map was
2392 * modified while it was unlocked.
2393 * Specifically, the entry may have been
2394 * clipped, merged, or deleted.
2396 if (!vm_map_lookup_entry(map, saved_start,
2398 if (flags & VM_MAP_WIRE_HOLESOK)
2399 tmp_entry = tmp_entry->next;
2401 if (saved_start == start) {
2403 * first_entry has been deleted.
2406 return (KERN_INVALID_ADDRESS);
2409 rv = KERN_INVALID_ADDRESS;
2413 if (entry == first_entry)
2414 first_entry = tmp_entry;
2419 last_timestamp = map->timestamp;
2422 vm_map_clip_start(map, entry, start);
2423 vm_map_clip_end(map, entry, end);
2425 * Mark the entry in case the map lock is released. (See
2428 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
2429 if ((entry->protection & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0
2430 || (entry->protection & prot) != prot) {
2431 entry->eflags |= MAP_ENTRY_WIRE_SKIPPED;
2432 if ((flags & VM_MAP_WIRE_HOLESOK) == 0) {
2434 rv = KERN_INVALID_ADDRESS;
2439 if (entry->wired_count == 0) {
2440 entry->wired_count++;
2441 saved_start = entry->start;
2442 saved_end = entry->end;
2443 fictitious = entry->object.vm_object != NULL &&
2444 (entry->object.vm_object->type == OBJT_DEVICE ||
2445 entry->object.vm_object->type == OBJT_SG);
2447 * Release the map lock, relying on the in-transition
2448 * mark. Mark the map busy for fork.
2452 rv = vm_fault_wire(map, saved_start, saved_end,
2456 if (last_timestamp + 1 != map->timestamp) {
2458 * Look again for the entry because the map was
2459 * modified while it was unlocked. The entry
2460 * may have been clipped, but NOT merged or
2463 result = vm_map_lookup_entry(map, saved_start,
2465 KASSERT(result, ("vm_map_wire: lookup failed"));
2466 if (entry == first_entry)
2467 first_entry = tmp_entry;
2471 while (entry->end < saved_end) {
2472 if (rv != KERN_SUCCESS) {
2473 KASSERT(entry->wired_count == 1,
2474 ("vm_map_wire: bad count"));
2475 entry->wired_count = -1;
2477 entry = entry->next;
2480 last_timestamp = map->timestamp;
2481 if (rv != KERN_SUCCESS) {
2482 KASSERT(entry->wired_count == 1,
2483 ("vm_map_wire: bad count"));
2485 * Assign an out-of-range value to represent
2486 * the failure to wire this entry.
2488 entry->wired_count = -1;
2492 } else if (!user_wire ||
2493 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
2494 entry->wired_count++;
2497 * Check the map for holes in the specified region.
2498 * If VM_MAP_WIRE_HOLESOK was specified, skip this check.
2501 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) &&
2502 (entry->end < end && (entry->next == &map->header ||
2503 entry->next->start > entry->end))) {
2505 rv = KERN_INVALID_ADDRESS;
2508 entry = entry->next;
2512 need_wakeup = FALSE;
2513 if (first_entry == NULL) {
2514 result = vm_map_lookup_entry(map, start, &first_entry);
2515 if (!result && (flags & VM_MAP_WIRE_HOLESOK))
2516 first_entry = first_entry->next;
2518 KASSERT(result, ("vm_map_wire: lookup failed"));
2520 entry = first_entry;
2521 while (entry != &map->header && entry->start < end) {
2522 if ((entry->eflags & MAP_ENTRY_WIRE_SKIPPED) != 0)
2523 goto next_entry_done;
2524 if (rv == KERN_SUCCESS) {
2526 entry->eflags |= MAP_ENTRY_USER_WIRED;
2527 } else if (entry->wired_count == -1) {
2529 * Wiring failed on this entry. Thus, unwiring is
2532 entry->wired_count = 0;
2535 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0)
2536 entry->wired_count--;
2537 if (entry->wired_count == 0) {
2539 * Retain the map lock.
2541 vm_fault_unwire(map, entry->start, entry->end,
2542 entry->object.vm_object != NULL &&
2543 (entry->object.vm_object->type == OBJT_DEVICE ||
2544 entry->object.vm_object->type == OBJT_SG));
2548 KASSERT(entry->eflags & MAP_ENTRY_IN_TRANSITION,
2549 ("vm_map_wire: in-transition flag missing"));
2550 entry->eflags &= ~(MAP_ENTRY_IN_TRANSITION|MAP_ENTRY_WIRE_SKIPPED);
2551 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
2552 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
2555 vm_map_simplify_entry(map, entry);
2556 entry = entry->next;
2567 * Push any dirty cached pages in the address range to their pager.
2568 * If syncio is TRUE, dirty pages are written synchronously.
2569 * If invalidate is TRUE, any cached pages are freed as well.
2571 * If the size of the region from start to end is zero, we are
2572 * supposed to flush all modified pages within the region containing
2573 * start. Unfortunately, a region can be split or coalesced with
2574 * neighboring regions, making it difficult to determine what the
2575 * original region was. Therefore, we approximate this requirement by
2576 * flushing the current region containing start.
2578 * Returns an error if any part of the specified range is not mapped.
2586 boolean_t invalidate)
2588 vm_map_entry_t current;
2589 vm_map_entry_t entry;
2592 vm_ooffset_t offset;
2593 unsigned int last_timestamp;
2595 vm_map_lock_read(map);
2596 VM_MAP_RANGE_CHECK(map, start, end);
2597 if (!vm_map_lookup_entry(map, start, &entry)) {
2598 vm_map_unlock_read(map);
2599 return (KERN_INVALID_ADDRESS);
2600 } else if (start == end) {
2601 start = entry->start;
2605 * Make a first pass to check for user-wired memory and holes.
2607 for (current = entry; current != &map->header && current->start < end;
2608 current = current->next) {
2609 if (invalidate && (current->eflags & MAP_ENTRY_USER_WIRED)) {
2610 vm_map_unlock_read(map);
2611 return (KERN_INVALID_ARGUMENT);
2613 if (end > current->end &&
2614 (current->next == &map->header ||
2615 current->end != current->next->start)) {
2616 vm_map_unlock_read(map);
2617 return (KERN_INVALID_ADDRESS);
2622 pmap_remove(map->pmap, start, end);
2625 * Make a second pass, cleaning/uncaching pages from the indicated
2628 for (current = entry; current != &map->header && current->start < end;) {
2629 offset = current->offset + (start - current->start);
2630 size = (end <= current->end ? end : current->end) - start;
2631 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
2633 vm_map_entry_t tentry;
2636 smap = current->object.sub_map;
2637 vm_map_lock_read(smap);
2638 (void) vm_map_lookup_entry(smap, offset, &tentry);
2639 tsize = tentry->end - offset;
2642 object = tentry->object.vm_object;
2643 offset = tentry->offset + (offset - tentry->start);
2644 vm_map_unlock_read(smap);
2646 object = current->object.vm_object;
2648 vm_object_reference(object);
2649 last_timestamp = map->timestamp;
2650 vm_map_unlock_read(map);
2651 vm_object_sync(object, offset, size, syncio, invalidate);
2653 vm_object_deallocate(object);
2654 vm_map_lock_read(map);
2655 if (last_timestamp == map->timestamp ||
2656 !vm_map_lookup_entry(map, start, ¤t))
2657 current = current->next;
2660 vm_map_unlock_read(map);
2661 return (KERN_SUCCESS);
2665 * vm_map_entry_unwire: [ internal use only ]
2667 * Make the region specified by this entry pageable.
2669 * The map in question should be locked.
2670 * [This is the reason for this routine's existence.]
2673 vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry)
2675 vm_fault_unwire(map, entry->start, entry->end,
2676 entry->object.vm_object != NULL &&
2677 (entry->object.vm_object->type == OBJT_DEVICE ||
2678 entry->object.vm_object->type == OBJT_SG));
2679 entry->wired_count = 0;
2683 vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map)
2686 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0)
2687 vm_object_deallocate(entry->object.vm_object);
2688 uma_zfree(system_map ? kmapentzone : mapentzone, entry);
2692 * vm_map_entry_delete: [ internal use only ]
2694 * Deallocate the given entry from the target map.
2697 vm_map_entry_delete(vm_map_t map, vm_map_entry_t entry)
2700 vm_pindex_t offidxstart, offidxend, count, size1;
2703 vm_map_entry_unlink(map, entry);
2704 object = entry->object.vm_object;
2705 size = entry->end - entry->start;
2708 if (entry->cred != NULL) {
2709 swap_release_by_cred(size, entry->cred);
2710 crfree(entry->cred);
2713 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0 &&
2715 KASSERT(entry->cred == NULL || object->cred == NULL ||
2716 (entry->eflags & MAP_ENTRY_NEEDS_COPY),
2717 ("OVERCOMMIT vm_map_entry_delete: both cred %p", entry));
2718 count = OFF_TO_IDX(size);
2719 offidxstart = OFF_TO_IDX(entry->offset);
2720 offidxend = offidxstart + count;
2721 VM_OBJECT_LOCK(object);
2722 if (object->ref_count != 1 &&
2723 ((object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING ||
2724 object == kernel_object || object == kmem_object)) {
2725 vm_object_collapse(object);
2728 * The option OBJPR_NOTMAPPED can be passed here
2729 * because vm_map_delete() already performed
2730 * pmap_remove() on the only mapping to this range
2733 vm_object_page_remove(object, offidxstart, offidxend,
2735 if (object->type == OBJT_SWAP)
2736 swap_pager_freespace(object, offidxstart, count);
2737 if (offidxend >= object->size &&
2738 offidxstart < object->size) {
2739 size1 = object->size;
2740 object->size = offidxstart;
2741 if (object->cred != NULL) {
2742 size1 -= object->size;
2743 KASSERT(object->charge >= ptoa(size1),
2744 ("vm_map_entry_delete: object->charge < 0"));
2745 swap_release_by_cred(ptoa(size1), object->cred);
2746 object->charge -= ptoa(size1);
2750 VM_OBJECT_UNLOCK(object);
2752 entry->object.vm_object = NULL;
2753 if (map->system_map)
2754 vm_map_entry_deallocate(entry, TRUE);
2756 entry->next = curthread->td_map_def_user;
2757 curthread->td_map_def_user = entry;
2762 * vm_map_delete: [ internal use only ]
2764 * Deallocates the given address range from the target
2768 vm_map_delete(vm_map_t map, vm_offset_t start, vm_offset_t end)
2770 vm_map_entry_t entry;
2771 vm_map_entry_t first_entry;
2773 VM_MAP_ASSERT_LOCKED(map);
2776 * Find the start of the region, and clip it
2778 if (!vm_map_lookup_entry(map, start, &first_entry))
2779 entry = first_entry->next;
2781 entry = first_entry;
2782 vm_map_clip_start(map, entry, start);
2786 * Step through all entries in this region
2788 while ((entry != &map->header) && (entry->start < end)) {
2789 vm_map_entry_t next;
2792 * Wait for wiring or unwiring of an entry to complete.
2793 * Also wait for any system wirings to disappear on
2796 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 ||
2797 (vm_map_pmap(map) != kernel_pmap &&
2798 vm_map_entry_system_wired_count(entry) != 0)) {
2799 unsigned int last_timestamp;
2800 vm_offset_t saved_start;
2801 vm_map_entry_t tmp_entry;
2803 saved_start = entry->start;
2804 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2805 last_timestamp = map->timestamp;
2806 (void) vm_map_unlock_and_wait(map, 0);
2808 if (last_timestamp + 1 != map->timestamp) {
2810 * Look again for the entry because the map was
2811 * modified while it was unlocked.
2812 * Specifically, the entry may have been
2813 * clipped, merged, or deleted.
2815 if (!vm_map_lookup_entry(map, saved_start,
2817 entry = tmp_entry->next;
2820 vm_map_clip_start(map, entry,
2826 vm_map_clip_end(map, entry, end);
2831 * Unwire before removing addresses from the pmap; otherwise,
2832 * unwiring will put the entries back in the pmap.
2834 if (entry->wired_count != 0) {
2835 vm_map_entry_unwire(map, entry);
2838 pmap_remove(map->pmap, entry->start, entry->end);
2841 * Delete the entry only after removing all pmap
2842 * entries pointing to its pages. (Otherwise, its
2843 * page frames may be reallocated, and any modify bits
2844 * will be set in the wrong object!)
2846 vm_map_entry_delete(map, entry);
2849 return (KERN_SUCCESS);
2855 * Remove the given address range from the target map.
2856 * This is the exported form of vm_map_delete.
2859 vm_map_remove(vm_map_t map, vm_offset_t start, vm_offset_t end)
2864 VM_MAP_RANGE_CHECK(map, start, end);
2865 result = vm_map_delete(map, start, end);
2871 * vm_map_check_protection:
2873 * Assert that the target map allows the specified privilege on the
2874 * entire address region given. The entire region must be allocated.
2876 * WARNING! This code does not and should not check whether the
2877 * contents of the region is accessible. For example a smaller file
2878 * might be mapped into a larger address space.
2880 * NOTE! This code is also called by munmap().
2882 * The map must be locked. A read lock is sufficient.
2885 vm_map_check_protection(vm_map_t map, vm_offset_t start, vm_offset_t end,
2886 vm_prot_t protection)
2888 vm_map_entry_t entry;
2889 vm_map_entry_t tmp_entry;
2891 if (!vm_map_lookup_entry(map, start, &tmp_entry))
2895 while (start < end) {
2896 if (entry == &map->header)
2901 if (start < entry->start)
2904 * Check protection associated with entry.
2906 if ((entry->protection & protection) != protection)
2908 /* go to next entry */
2910 entry = entry->next;
2916 * vm_map_copy_entry:
2918 * Copies the contents of the source entry to the destination
2919 * entry. The entries *must* be aligned properly.
2925 vm_map_entry_t src_entry,
2926 vm_map_entry_t dst_entry,
2927 vm_ooffset_t *fork_charge)
2929 vm_object_t src_object;
2930 vm_map_entry_t fake_entry;
2935 VM_MAP_ASSERT_LOCKED(dst_map);
2937 if ((dst_entry->eflags|src_entry->eflags) & MAP_ENTRY_IS_SUB_MAP)
2940 if (src_entry->wired_count == 0) {
2943 * If the source entry is marked needs_copy, it is already
2946 if ((src_entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) {
2947 pmap_protect(src_map->pmap,
2950 src_entry->protection & ~VM_PROT_WRITE);
2954 * Make a copy of the object.
2956 size = src_entry->end - src_entry->start;
2957 if ((src_object = src_entry->object.vm_object) != NULL) {
2958 VM_OBJECT_LOCK(src_object);
2959 charged = ENTRY_CHARGED(src_entry);
2960 if ((src_object->handle == NULL) &&
2961 (src_object->type == OBJT_DEFAULT ||
2962 src_object->type == OBJT_SWAP)) {
2963 vm_object_collapse(src_object);
2964 if ((src_object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING) {
2965 vm_object_split(src_entry);
2966 src_object = src_entry->object.vm_object;
2969 vm_object_reference_locked(src_object);
2970 vm_object_clear_flag(src_object, OBJ_ONEMAPPING);
2971 if (src_entry->cred != NULL &&
2972 !(src_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
2973 KASSERT(src_object->cred == NULL,
2974 ("OVERCOMMIT: vm_map_copy_entry: cred %p",
2976 src_object->cred = src_entry->cred;
2977 src_object->charge = size;
2979 VM_OBJECT_UNLOCK(src_object);
2980 dst_entry->object.vm_object = src_object;
2982 cred = curthread->td_ucred;
2984 dst_entry->cred = cred;
2985 *fork_charge += size;
2986 if (!(src_entry->eflags &
2987 MAP_ENTRY_NEEDS_COPY)) {
2989 src_entry->cred = cred;
2990 *fork_charge += size;
2993 src_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY);
2994 dst_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY);
2995 dst_entry->offset = src_entry->offset;
2996 if (src_entry->eflags & MAP_ENTRY_VN_WRITECNT) {
2998 * MAP_ENTRY_VN_WRITECNT cannot
2999 * indicate write reference from
3000 * src_entry, since the entry is
3001 * marked as needs copy. Allocate a
3002 * fake entry that is used to
3003 * decrement object->un_pager.vnp.writecount
3004 * at the appropriate time. Attach
3005 * fake_entry to the deferred list.
3007 fake_entry = vm_map_entry_create(dst_map);
3008 fake_entry->eflags = MAP_ENTRY_VN_WRITECNT;
3009 src_entry->eflags &= ~MAP_ENTRY_VN_WRITECNT;
3010 vm_object_reference(src_object);
3011 fake_entry->object.vm_object = src_object;
3012 fake_entry->start = src_entry->start;
3013 fake_entry->end = src_entry->end;
3014 fake_entry->next = curthread->td_map_def_user;
3015 curthread->td_map_def_user = fake_entry;
3018 dst_entry->object.vm_object = NULL;
3019 dst_entry->offset = 0;
3020 if (src_entry->cred != NULL) {
3021 dst_entry->cred = curthread->td_ucred;
3022 crhold(dst_entry->cred);
3023 *fork_charge += size;
3027 pmap_copy(dst_map->pmap, src_map->pmap, dst_entry->start,
3028 dst_entry->end - dst_entry->start, src_entry->start);
3031 * Of course, wired down pages can't be set copy-on-write.
3032 * Cause wired pages to be copied into the new map by
3033 * simulating faults (the new pages are pageable)
3035 vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry,
3041 * vmspace_map_entry_forked:
3042 * Update the newly-forked vmspace each time a map entry is inherited
3043 * or copied. The values for vm_dsize and vm_tsize are approximate
3044 * (and mostly-obsolete ideas in the face of mmap(2) et al.)
3047 vmspace_map_entry_forked(const struct vmspace *vm1, struct vmspace *vm2,
3048 vm_map_entry_t entry)
3050 vm_size_t entrysize;
3053 entrysize = entry->end - entry->start;
3054 vm2->vm_map.size += entrysize;
3055 if (entry->eflags & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP)) {
3056 vm2->vm_ssize += btoc(entrysize);
3057 } else if (entry->start >= (vm_offset_t)vm1->vm_daddr &&
3058 entry->start < (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)) {
3059 newend = MIN(entry->end,
3060 (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize));
3061 vm2->vm_dsize += btoc(newend - entry->start);
3062 } else if (entry->start >= (vm_offset_t)vm1->vm_taddr &&
3063 entry->start < (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)) {
3064 newend = MIN(entry->end,
3065 (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize));
3066 vm2->vm_tsize += btoc(newend - entry->start);
3072 * Create a new process vmspace structure and vm_map
3073 * based on those of an existing process. The new map
3074 * is based on the old map, according to the inheritance
3075 * values on the regions in that map.
3077 * XXX It might be worth coalescing the entries added to the new vmspace.
3079 * The source map must not be locked.
3082 vmspace_fork(struct vmspace *vm1, vm_ooffset_t *fork_charge)
3084 struct vmspace *vm2;
3085 vm_map_t old_map = &vm1->vm_map;
3087 vm_map_entry_t old_entry;
3088 vm_map_entry_t new_entry;
3092 vm_map_lock(old_map);
3094 vm_map_wait_busy(old_map);
3095 new_map = NULL; /* silence gcc */
3096 vm2 = vmspace_alloc(old_map->min_offset, old_map->max_offset);
3098 goto unlock_and_return;
3099 vm2->vm_taddr = vm1->vm_taddr;
3100 vm2->vm_daddr = vm1->vm_daddr;
3101 vm2->vm_maxsaddr = vm1->vm_maxsaddr;
3102 new_map = &vm2->vm_map; /* XXX */
3103 locked = vm_map_trylock(new_map); /* trylock to silence WITNESS */
3104 KASSERT(locked, ("vmspace_fork: lock failed"));
3105 new_map->timestamp = 1;
3107 old_entry = old_map->header.next;
3109 while (old_entry != &old_map->header) {
3110 if (old_entry->eflags & MAP_ENTRY_IS_SUB_MAP)
3111 panic("vm_map_fork: encountered a submap");
3113 switch (old_entry->inheritance) {
3114 case VM_INHERIT_NONE:
3117 case VM_INHERIT_SHARE:
3119 * Clone the entry, creating the shared object if necessary.
3121 object = old_entry->object.vm_object;
3122 if (object == NULL) {
3123 object = vm_object_allocate(OBJT_DEFAULT,
3124 atop(old_entry->end - old_entry->start));
3125 old_entry->object.vm_object = object;
3126 old_entry->offset = 0;
3127 if (old_entry->cred != NULL) {
3128 object->cred = old_entry->cred;
3129 object->charge = old_entry->end -
3131 old_entry->cred = NULL;
3136 * Add the reference before calling vm_object_shadow
3137 * to insure that a shadow object is created.
3139 vm_object_reference(object);
3140 if (old_entry->eflags & MAP_ENTRY_NEEDS_COPY) {
3141 vm_object_shadow(&old_entry->object.vm_object,
3143 old_entry->end - old_entry->start);
3144 old_entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
3145 /* Transfer the second reference too. */
3146 vm_object_reference(
3147 old_entry->object.vm_object);
3150 * As in vm_map_simplify_entry(), the
3151 * vnode lock will not be acquired in
3152 * this call to vm_object_deallocate().
3154 vm_object_deallocate(object);
3155 object = old_entry->object.vm_object;
3157 VM_OBJECT_LOCK(object);
3158 vm_object_clear_flag(object, OBJ_ONEMAPPING);
3159 if (old_entry->cred != NULL) {
3160 KASSERT(object->cred == NULL, ("vmspace_fork both cred"));
3161 object->cred = old_entry->cred;
3162 object->charge = old_entry->end - old_entry->start;
3163 old_entry->cred = NULL;
3165 VM_OBJECT_UNLOCK(object);
3168 * Clone the entry, referencing the shared object.
3170 new_entry = vm_map_entry_create(new_map);
3171 *new_entry = *old_entry;
3172 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED |
3173 MAP_ENTRY_IN_TRANSITION);
3174 new_entry->wired_count = 0;
3175 if (new_entry->eflags & MAP_ENTRY_VN_WRITECNT) {
3176 object = new_entry->object.vm_object;
3177 KASSERT(((struct vnode *)object->handle)->
3179 ("vmspace_fork: v_writecount"));
3180 KASSERT(object->un_pager.vnp.writemappings > 0,
3181 ("vmspace_fork: vnp.writecount"));
3182 vnode_pager_update_writecount(object,
3183 new_entry->start, new_entry->end);
3187 * Insert the entry into the new map -- we know we're
3188 * inserting at the end of the new map.
3190 vm_map_entry_link(new_map, new_map->header.prev,
3192 vmspace_map_entry_forked(vm1, vm2, new_entry);
3195 * Update the physical map
3197 pmap_copy(new_map->pmap, old_map->pmap,
3199 (old_entry->end - old_entry->start),
3203 case VM_INHERIT_COPY:
3205 * Clone the entry and link into the map.
3207 new_entry = vm_map_entry_create(new_map);
3208 *new_entry = *old_entry;
3210 * Copied entry is COW over the old object.
3212 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED |
3213 MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_VN_WRITECNT);
3214 new_entry->wired_count = 0;
3215 new_entry->object.vm_object = NULL;
3216 new_entry->cred = NULL;
3217 vm_map_entry_link(new_map, new_map->header.prev,
3219 vmspace_map_entry_forked(vm1, vm2, new_entry);
3220 vm_map_copy_entry(old_map, new_map, old_entry,
3221 new_entry, fork_charge);
3224 old_entry = old_entry->next;
3228 * Use inlined vm_map_unlock() to postpone handling the deferred
3229 * map entries, which cannot be done until both old_map and
3230 * new_map locks are released.
3232 sx_xunlock(&old_map->lock);
3234 sx_xunlock(&new_map->lock);
3235 vm_map_process_deferred();
3241 vm_map_stack(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
3242 vm_prot_t prot, vm_prot_t max, int cow)
3244 vm_map_entry_t new_entry, prev_entry;
3245 vm_offset_t bot, top;
3246 vm_size_t init_ssize;
3251 * The stack orientation is piggybacked with the cow argument.
3252 * Extract it into orient and mask the cow argument so that we
3253 * don't pass it around further.
3254 * NOTE: We explicitly allow bi-directional stacks.
3256 orient = cow & (MAP_STACK_GROWS_DOWN|MAP_STACK_GROWS_UP);
3258 KASSERT(orient != 0, ("No stack grow direction"));
3260 if (addrbos < vm_map_min(map) ||
3261 addrbos > vm_map_max(map) ||
3262 addrbos + max_ssize < addrbos)
3263 return (KERN_NO_SPACE);
3265 init_ssize = (max_ssize < sgrowsiz) ? max_ssize : sgrowsiz;
3267 PROC_LOCK(curthread->td_proc);
3268 vmemlim = lim_cur(curthread->td_proc, RLIMIT_VMEM);
3269 PROC_UNLOCK(curthread->td_proc);
3273 /* If addr is already mapped, no go */
3274 if (vm_map_lookup_entry(map, addrbos, &prev_entry)) {
3276 return (KERN_NO_SPACE);
3279 /* If we would blow our VMEM resource limit, no go */
3280 if (map->size + init_ssize > vmemlim) {
3282 return (KERN_NO_SPACE);
3286 * If we can't accomodate max_ssize in the current mapping, no go.
3287 * However, we need to be aware that subsequent user mappings might
3288 * map into the space we have reserved for stack, and currently this
3289 * space is not protected.
3291 * Hopefully we will at least detect this condition when we try to
3294 if ((prev_entry->next != &map->header) &&
3295 (prev_entry->next->start < addrbos + max_ssize)) {
3297 return (KERN_NO_SPACE);
3301 * We initially map a stack of only init_ssize. We will grow as
3302 * needed later. Depending on the orientation of the stack (i.e.
3303 * the grow direction) we either map at the top of the range, the
3304 * bottom of the range or in the middle.
3306 * Note: we would normally expect prot and max to be VM_PROT_ALL,
3307 * and cow to be 0. Possibly we should eliminate these as input
3308 * parameters, and just pass these values here in the insert call.
3310 if (orient == MAP_STACK_GROWS_DOWN)
3311 bot = addrbos + max_ssize - init_ssize;
3312 else if (orient == MAP_STACK_GROWS_UP)
3315 bot = round_page(addrbos + max_ssize/2 - init_ssize/2);
3316 top = bot + init_ssize;
3317 rv = vm_map_insert(map, NULL, 0, bot, top, prot, max, cow);
3319 /* Now set the avail_ssize amount. */
3320 if (rv == KERN_SUCCESS) {
3321 if (prev_entry != &map->header)
3322 vm_map_clip_end(map, prev_entry, bot);
3323 new_entry = prev_entry->next;
3324 if (new_entry->end != top || new_entry->start != bot)
3325 panic("Bad entry start/end for new stack entry");
3327 new_entry->avail_ssize = max_ssize - init_ssize;
3328 if (orient & MAP_STACK_GROWS_DOWN)
3329 new_entry->eflags |= MAP_ENTRY_GROWS_DOWN;
3330 if (orient & MAP_STACK_GROWS_UP)
3331 new_entry->eflags |= MAP_ENTRY_GROWS_UP;
3338 static int stack_guard_page = 0;
3339 TUNABLE_INT("security.bsd.stack_guard_page", &stack_guard_page);
3340 SYSCTL_INT(_security_bsd, OID_AUTO, stack_guard_page, CTLFLAG_RW,
3341 &stack_guard_page, 0,
3342 "Insert stack guard page ahead of the growable segments.");
3344 /* Attempts to grow a vm stack entry. Returns KERN_SUCCESS if the
3345 * desired address is already mapped, or if we successfully grow
3346 * the stack. Also returns KERN_SUCCESS if addr is outside the
3347 * stack range (this is strange, but preserves compatibility with
3348 * the grow function in vm_machdep.c).
3351 vm_map_growstack(struct proc *p, vm_offset_t addr)
3353 vm_map_entry_t next_entry, prev_entry;
3354 vm_map_entry_t new_entry, stack_entry;
3355 struct vmspace *vm = p->p_vmspace;
3356 vm_map_t map = &vm->vm_map;
3358 size_t grow_amount, max_grow;
3359 rlim_t stacklim, vmemlim;
3360 int is_procstack, rv;
3371 stacklim = lim_cur(p, RLIMIT_STACK);
3372 vmemlim = lim_cur(p, RLIMIT_VMEM);
3375 vm_map_lock_read(map);
3377 /* If addr is already in the entry range, no need to grow.*/
3378 if (vm_map_lookup_entry(map, addr, &prev_entry)) {
3379 vm_map_unlock_read(map);
3380 return (KERN_SUCCESS);
3383 next_entry = prev_entry->next;
3384 if (!(prev_entry->eflags & MAP_ENTRY_GROWS_UP)) {
3386 * This entry does not grow upwards. Since the address lies
3387 * beyond this entry, the next entry (if one exists) has to
3388 * be a downward growable entry. The entry list header is
3389 * never a growable entry, so it suffices to check the flags.
3391 if (!(next_entry->eflags & MAP_ENTRY_GROWS_DOWN)) {
3392 vm_map_unlock_read(map);
3393 return (KERN_SUCCESS);
3395 stack_entry = next_entry;
3398 * This entry grows upward. If the next entry does not at
3399 * least grow downwards, this is the entry we need to grow.
3400 * otherwise we have two possible choices and we have to
3403 if (next_entry->eflags & MAP_ENTRY_GROWS_DOWN) {
3405 * We have two choices; grow the entry closest to
3406 * the address to minimize the amount of growth.
3408 if (addr - prev_entry->end <= next_entry->start - addr)
3409 stack_entry = prev_entry;
3411 stack_entry = next_entry;
3413 stack_entry = prev_entry;
3416 if (stack_entry == next_entry) {
3417 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_DOWN, ("foo"));
3418 KASSERT(addr < stack_entry->start, ("foo"));
3419 end = (prev_entry != &map->header) ? prev_entry->end :
3420 stack_entry->start - stack_entry->avail_ssize;
3421 grow_amount = roundup(stack_entry->start - addr, PAGE_SIZE);
3422 max_grow = stack_entry->start - end;
3424 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_UP, ("foo"));
3425 KASSERT(addr >= stack_entry->end, ("foo"));
3426 end = (next_entry != &map->header) ? next_entry->start :
3427 stack_entry->end + stack_entry->avail_ssize;
3428 grow_amount = roundup(addr + 1 - stack_entry->end, PAGE_SIZE);
3429 max_grow = end - stack_entry->end;
3432 if (grow_amount > stack_entry->avail_ssize) {
3433 vm_map_unlock_read(map);
3434 return (KERN_NO_SPACE);
3438 * If there is no longer enough space between the entries nogo, and
3439 * adjust the available space. Note: this should only happen if the
3440 * user has mapped into the stack area after the stack was created,
3441 * and is probably an error.
3443 * This also effectively destroys any guard page the user might have
3444 * intended by limiting the stack size.
3446 if (grow_amount + (stack_guard_page ? PAGE_SIZE : 0) > max_grow) {
3447 if (vm_map_lock_upgrade(map))
3450 stack_entry->avail_ssize = max_grow;
3453 return (KERN_NO_SPACE);
3456 is_procstack = (addr >= (vm_offset_t)vm->vm_maxsaddr) ? 1 : 0;
3459 * If this is the main process stack, see if we're over the stack
3462 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) {
3463 vm_map_unlock_read(map);
3464 return (KERN_NO_SPACE);
3469 racct_set(p, RACCT_STACK, ctob(vm->vm_ssize) + grow_amount)) {
3471 vm_map_unlock_read(map);
3472 return (KERN_NO_SPACE);
3477 /* Round up the grow amount modulo SGROWSIZ */
3478 grow_amount = roundup (grow_amount, sgrowsiz);
3479 if (grow_amount > stack_entry->avail_ssize)
3480 grow_amount = stack_entry->avail_ssize;
3481 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) {
3482 grow_amount = trunc_page((vm_size_t)stacklim) -
3487 limit = racct_get_available(p, RACCT_STACK);
3489 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > limit))
3490 grow_amount = limit - ctob(vm->vm_ssize);
3493 /* If we would blow our VMEM resource limit, no go */
3494 if (map->size + grow_amount > vmemlim) {
3495 vm_map_unlock_read(map);
3501 if (racct_set(p, RACCT_VMEM, map->size + grow_amount)) {
3503 vm_map_unlock_read(map);
3510 if (vm_map_lock_upgrade(map))
3513 if (stack_entry == next_entry) {
3517 /* Get the preliminary new entry start value */
3518 addr = stack_entry->start - grow_amount;
3521 * If this puts us into the previous entry, cut back our
3522 * growth to the available space. Also, see the note above.
3525 stack_entry->avail_ssize = max_grow;
3527 if (stack_guard_page)
3531 rv = vm_map_insert(map, NULL, 0, addr, stack_entry->start,
3532 p->p_sysent->sv_stackprot, VM_PROT_ALL, 0);
3534 /* Adjust the available stack space by the amount we grew. */
3535 if (rv == KERN_SUCCESS) {
3536 if (prev_entry != &map->header)
3537 vm_map_clip_end(map, prev_entry, addr);
3538 new_entry = prev_entry->next;
3539 KASSERT(new_entry == stack_entry->prev, ("foo"));
3540 KASSERT(new_entry->end == stack_entry->start, ("foo"));
3541 KASSERT(new_entry->start == addr, ("foo"));
3542 grow_amount = new_entry->end - new_entry->start;
3543 new_entry->avail_ssize = stack_entry->avail_ssize -
3545 stack_entry->eflags &= ~MAP_ENTRY_GROWS_DOWN;
3546 new_entry->eflags |= MAP_ENTRY_GROWS_DOWN;
3552 addr = stack_entry->end + grow_amount;
3555 * If this puts us into the next entry, cut back our growth
3556 * to the available space. Also, see the note above.
3559 stack_entry->avail_ssize = end - stack_entry->end;
3561 if (stack_guard_page)
3565 grow_amount = addr - stack_entry->end;
3566 cred = stack_entry->cred;
3567 if (cred == NULL && stack_entry->object.vm_object != NULL)
3568 cred = stack_entry->object.vm_object->cred;
3569 if (cred != NULL && !swap_reserve_by_cred(grow_amount, cred))
3571 /* Grow the underlying object if applicable. */
3572 else if (stack_entry->object.vm_object == NULL ||
3573 vm_object_coalesce(stack_entry->object.vm_object,
3574 stack_entry->offset,
3575 (vm_size_t)(stack_entry->end - stack_entry->start),
3576 (vm_size_t)grow_amount, cred != NULL)) {
3577 map->size += (addr - stack_entry->end);
3578 /* Update the current entry. */
3579 stack_entry->end = addr;
3580 stack_entry->avail_ssize -= grow_amount;
3581 vm_map_entry_resize_free(map, stack_entry);
3584 if (next_entry != &map->header)
3585 vm_map_clip_start(map, next_entry, addr);
3590 if (rv == KERN_SUCCESS && is_procstack)
3591 vm->vm_ssize += btoc(grow_amount);
3596 * Heed the MAP_WIREFUTURE flag if it was set for this process.
3598 if (rv == KERN_SUCCESS && (map->flags & MAP_WIREFUTURE)) {
3600 (stack_entry == next_entry) ? addr : addr - grow_amount,
3601 (stack_entry == next_entry) ? stack_entry->start : addr,
3602 (p->p_flag & P_SYSTEM)
3603 ? VM_MAP_WIRE_SYSTEM|VM_MAP_WIRE_NOHOLES
3604 : VM_MAP_WIRE_USER|VM_MAP_WIRE_NOHOLES);
3609 if (rv != KERN_SUCCESS) {
3611 error = racct_set(p, RACCT_VMEM, map->size);
3612 KASSERT(error == 0, ("decreasing RACCT_VMEM failed"));
3613 error = racct_set(p, RACCT_STACK, ctob(vm->vm_ssize));
3614 KASSERT(error == 0, ("decreasing RACCT_STACK failed"));
3623 * Unshare the specified VM space for exec. If other processes are
3624 * mapped to it, then create a new one. The new vmspace is null.
3627 vmspace_exec(struct proc *p, vm_offset_t minuser, vm_offset_t maxuser)
3629 struct vmspace *oldvmspace = p->p_vmspace;
3630 struct vmspace *newvmspace;
3632 newvmspace = vmspace_alloc(minuser, maxuser);
3633 if (newvmspace == NULL)
3635 newvmspace->vm_swrss = oldvmspace->vm_swrss;
3637 * This code is written like this for prototype purposes. The
3638 * goal is to avoid running down the vmspace here, but let the
3639 * other process's that are still using the vmspace to finally
3640 * run it down. Even though there is little or no chance of blocking
3641 * here, it is a good idea to keep this form for future mods.
3643 PROC_VMSPACE_LOCK(p);
3644 p->p_vmspace = newvmspace;
3645 PROC_VMSPACE_UNLOCK(p);
3646 if (p == curthread->td_proc)
3647 pmap_activate(curthread);
3648 vmspace_free(oldvmspace);
3653 * Unshare the specified VM space for forcing COW. This
3654 * is called by rfork, for the (RFMEM|RFPROC) == 0 case.
3657 vmspace_unshare(struct proc *p)
3659 struct vmspace *oldvmspace = p->p_vmspace;
3660 struct vmspace *newvmspace;
3661 vm_ooffset_t fork_charge;
3663 if (oldvmspace->vm_refcnt == 1)
3666 newvmspace = vmspace_fork(oldvmspace, &fork_charge);
3667 if (newvmspace == NULL)
3669 if (!swap_reserve_by_cred(fork_charge, p->p_ucred)) {
3670 vmspace_free(newvmspace);
3673 PROC_VMSPACE_LOCK(p);
3674 p->p_vmspace = newvmspace;
3675 PROC_VMSPACE_UNLOCK(p);
3676 if (p == curthread->td_proc)
3677 pmap_activate(curthread);
3678 vmspace_free(oldvmspace);
3685 * Finds the VM object, offset, and
3686 * protection for a given virtual address in the
3687 * specified map, assuming a page fault of the
3690 * Leaves the map in question locked for read; return
3691 * values are guaranteed until a vm_map_lookup_done
3692 * call is performed. Note that the map argument
3693 * is in/out; the returned map must be used in
3694 * the call to vm_map_lookup_done.
3696 * A handle (out_entry) is returned for use in
3697 * vm_map_lookup_done, to make that fast.
3699 * If a lookup is requested with "write protection"
3700 * specified, the map may be changed to perform virtual
3701 * copying operations, although the data referenced will
3705 vm_map_lookup(vm_map_t *var_map, /* IN/OUT */
3707 vm_prot_t fault_typea,
3708 vm_map_entry_t *out_entry, /* OUT */
3709 vm_object_t *object, /* OUT */
3710 vm_pindex_t *pindex, /* OUT */
3711 vm_prot_t *out_prot, /* OUT */
3712 boolean_t *wired) /* OUT */
3714 vm_map_entry_t entry;
3715 vm_map_t map = *var_map;
3717 vm_prot_t fault_type = fault_typea;
3718 vm_object_t eobject;
3724 vm_map_lock_read(map);
3727 * Lookup the faulting address.
3729 if (!vm_map_lookup_entry(map, vaddr, out_entry)) {
3730 vm_map_unlock_read(map);
3731 return (KERN_INVALID_ADDRESS);
3739 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
3740 vm_map_t old_map = map;
3742 *var_map = map = entry->object.sub_map;
3743 vm_map_unlock_read(old_map);
3748 * Check whether this task is allowed to have this page.
3750 prot = entry->protection;
3751 fault_type &= (VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE);
3752 if ((fault_type & prot) != fault_type || prot == VM_PROT_NONE) {
3753 vm_map_unlock_read(map);
3754 return (KERN_PROTECTION_FAILURE);
3756 if ((entry->eflags & MAP_ENTRY_USER_WIRED) &&
3757 (entry->eflags & MAP_ENTRY_COW) &&
3758 (fault_type & VM_PROT_WRITE)) {
3759 vm_map_unlock_read(map);
3760 return (KERN_PROTECTION_FAILURE);
3764 * If this page is not pageable, we have to get it for all possible
3767 *wired = (entry->wired_count != 0);
3769 fault_type = entry->protection;
3770 size = entry->end - entry->start;
3772 * If the entry was copy-on-write, we either ...
3774 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
3776 * If we want to write the page, we may as well handle that
3777 * now since we've got the map locked.
3779 * If we don't need to write the page, we just demote the
3780 * permissions allowed.
3782 if ((fault_type & VM_PROT_WRITE) != 0 ||
3783 (fault_typea & VM_PROT_COPY) != 0) {
3785 * Make a new object, and place it in the object
3786 * chain. Note that no new references have appeared
3787 * -- one just moved from the map to the new
3790 if (vm_map_lock_upgrade(map))
3793 if (entry->cred == NULL) {
3795 * The debugger owner is charged for
3798 cred = curthread->td_ucred;
3800 if (!swap_reserve_by_cred(size, cred)) {
3803 return (KERN_RESOURCE_SHORTAGE);
3807 vm_object_shadow(&entry->object.vm_object,
3808 &entry->offset, size);
3809 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
3810 eobject = entry->object.vm_object;
3811 if (eobject->cred != NULL) {
3813 * The object was not shadowed.
3815 swap_release_by_cred(size, entry->cred);
3816 crfree(entry->cred);
3818 } else if (entry->cred != NULL) {
3819 VM_OBJECT_LOCK(eobject);
3820 eobject->cred = entry->cred;
3821 eobject->charge = size;
3822 VM_OBJECT_UNLOCK(eobject);
3826 vm_map_lock_downgrade(map);
3829 * We're attempting to read a copy-on-write page --
3830 * don't allow writes.
3832 prot &= ~VM_PROT_WRITE;
3837 * Create an object if necessary.
3839 if (entry->object.vm_object == NULL &&
3841 if (vm_map_lock_upgrade(map))
3843 entry->object.vm_object = vm_object_allocate(OBJT_DEFAULT,
3846 if (entry->cred != NULL) {
3847 VM_OBJECT_LOCK(entry->object.vm_object);
3848 entry->object.vm_object->cred = entry->cred;
3849 entry->object.vm_object->charge = size;
3850 VM_OBJECT_UNLOCK(entry->object.vm_object);
3853 vm_map_lock_downgrade(map);
3857 * Return the object/offset from this entry. If the entry was
3858 * copy-on-write or empty, it has been fixed up.
3860 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
3861 *object = entry->object.vm_object;
3864 return (KERN_SUCCESS);
3868 * vm_map_lookup_locked:
3870 * Lookup the faulting address. A version of vm_map_lookup that returns
3871 * KERN_FAILURE instead of blocking on map lock or memory allocation.
3874 vm_map_lookup_locked(vm_map_t *var_map, /* IN/OUT */
3876 vm_prot_t fault_typea,
3877 vm_map_entry_t *out_entry, /* OUT */
3878 vm_object_t *object, /* OUT */
3879 vm_pindex_t *pindex, /* OUT */
3880 vm_prot_t *out_prot, /* OUT */
3881 boolean_t *wired) /* OUT */
3883 vm_map_entry_t entry;
3884 vm_map_t map = *var_map;
3886 vm_prot_t fault_type = fault_typea;
3889 * Lookup the faulting address.
3891 if (!vm_map_lookup_entry(map, vaddr, out_entry))
3892 return (KERN_INVALID_ADDRESS);
3897 * Fail if the entry refers to a submap.
3899 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP)
3900 return (KERN_FAILURE);
3903 * Check whether this task is allowed to have this page.
3905 prot = entry->protection;
3906 fault_type &= VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE;
3907 if ((fault_type & prot) != fault_type)
3908 return (KERN_PROTECTION_FAILURE);
3909 if ((entry->eflags & MAP_ENTRY_USER_WIRED) &&
3910 (entry->eflags & MAP_ENTRY_COW) &&
3911 (fault_type & VM_PROT_WRITE))
3912 return (KERN_PROTECTION_FAILURE);
3915 * If this page is not pageable, we have to get it for all possible
3918 *wired = (entry->wired_count != 0);
3920 fault_type = entry->protection;
3922 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
3924 * Fail if the entry was copy-on-write for a write fault.
3926 if (fault_type & VM_PROT_WRITE)
3927 return (KERN_FAILURE);
3929 * We're attempting to read a copy-on-write page --
3930 * don't allow writes.
3932 prot &= ~VM_PROT_WRITE;
3936 * Fail if an object should be created.
3938 if (entry->object.vm_object == NULL && !map->system_map)
3939 return (KERN_FAILURE);
3942 * Return the object/offset from this entry. If the entry was
3943 * copy-on-write or empty, it has been fixed up.
3945 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
3946 *object = entry->object.vm_object;
3949 return (KERN_SUCCESS);
3953 * vm_map_lookup_done:
3955 * Releases locks acquired by a vm_map_lookup
3956 * (according to the handle returned by that lookup).
3959 vm_map_lookup_done(vm_map_t map, vm_map_entry_t entry)
3962 * Unlock the main-level map
3964 vm_map_unlock_read(map);
3967 #include "opt_ddb.h"
3969 #include <sys/kernel.h>
3971 #include <ddb/ddb.h>
3974 * vm_map_print: [ debug ]
3976 DB_SHOW_COMMAND(map, vm_map_print)
3979 /* XXX convert args. */
3980 vm_map_t map = (vm_map_t)addr;
3981 boolean_t full = have_addr;
3983 vm_map_entry_t entry;
3985 db_iprintf("Task map %p: pmap=%p, nentries=%d, version=%u\n",
3987 (void *)map->pmap, map->nentries, map->timestamp);
3990 if (!full && db_indent)
3994 for (entry = map->header.next; entry != &map->header;
3995 entry = entry->next) {
3996 db_iprintf("map entry %p: start=%p, end=%p\n",
3997 (void *)entry, (void *)entry->start, (void *)entry->end);
4000 static char *inheritance_name[4] =
4001 {"share", "copy", "none", "donate_copy"};
4003 db_iprintf(" prot=%x/%x/%s",
4005 entry->max_protection,
4006 inheritance_name[(int)(unsigned char)entry->inheritance]);
4007 if (entry->wired_count != 0)
4008 db_printf(", wired");
4010 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
4011 db_printf(", share=%p, offset=0x%jx\n",
4012 (void *)entry->object.sub_map,
4013 (uintmax_t)entry->offset);
4015 if ((entry->prev == &map->header) ||
4016 (entry->prev->object.sub_map !=
4017 entry->object.sub_map)) {
4019 vm_map_print((db_expr_t)(intptr_t)
4020 entry->object.sub_map,
4021 full, 0, (char *)0);
4025 if (entry->cred != NULL)
4026 db_printf(", ruid %d", entry->cred->cr_ruid);
4027 db_printf(", object=%p, offset=0x%jx",
4028 (void *)entry->object.vm_object,
4029 (uintmax_t)entry->offset);
4030 if (entry->object.vm_object && entry->object.vm_object->cred)
4031 db_printf(", obj ruid %d charge %jx",
4032 entry->object.vm_object->cred->cr_ruid,
4033 (uintmax_t)entry->object.vm_object->charge);
4034 if (entry->eflags & MAP_ENTRY_COW)
4035 db_printf(", copy (%s)",
4036 (entry->eflags & MAP_ENTRY_NEEDS_COPY) ? "needed" : "done");
4040 if ((entry->prev == &map->header) ||
4041 (entry->prev->object.vm_object !=
4042 entry->object.vm_object)) {
4044 vm_object_print((db_expr_t)(intptr_t)
4045 entry->object.vm_object,
4046 full, 0, (char *)0);
4058 DB_SHOW_COMMAND(procvm, procvm)
4063 p = (struct proc *) addr;
4068 db_printf("p = %p, vmspace = %p, map = %p, pmap = %p\n",
4069 (void *)p, (void *)p->p_vmspace, (void *)&p->p_vmspace->vm_map,
4070 (void *)vmspace_pmap(p->p_vmspace));
4072 vm_map_print((db_expr_t)(intptr_t)&p->p_vmspace->vm_map, 1, 0, NULL);