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, "vm map (system)", NULL, MTX_DEF | MTX_DUPOK);
245 sx_init(&map->lock, "vm map (user)");
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, next;
482 entry = td->td_map_def_user;
483 td->td_map_def_user = NULL;
484 while (entry != NULL) {
486 if ((entry->eflags & MAP_ENTRY_VN_WRITECNT) != 0) {
488 * Decrement the object's writemappings and
489 * possibly the vnode's v_writecount.
491 KASSERT((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0,
492 ("Submap with writecount"));
493 object = entry->object.vm_object;
494 KASSERT(object != NULL, ("No object for writecount"));
495 vnode_pager_release_writecount(object, entry->start,
498 vm_map_entry_deallocate(entry, FALSE);
504 _vm_map_unlock(vm_map_t map, const char *file, int line)
508 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
510 sx_xunlock_(&map->lock, file, line);
511 vm_map_process_deferred();
516 _vm_map_lock_read(vm_map_t map, const char *file, int line)
520 mtx_lock_flags_(&map->system_mtx, 0, file, line);
522 sx_slock_(&map->lock, file, line);
526 _vm_map_unlock_read(vm_map_t map, const char *file, int line)
530 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
532 sx_sunlock_(&map->lock, file, line);
533 vm_map_process_deferred();
538 _vm_map_trylock(vm_map_t map, const char *file, int line)
542 error = map->system_map ?
543 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) :
544 !sx_try_xlock_(&map->lock, file, line);
551 _vm_map_trylock_read(vm_map_t map, const char *file, int line)
555 error = map->system_map ?
556 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) :
557 !sx_try_slock_(&map->lock, file, line);
562 * _vm_map_lock_upgrade: [ internal use only ]
564 * Tries to upgrade a read (shared) lock on the specified map to a write
565 * (exclusive) lock. Returns the value "0" if the upgrade succeeds and a
566 * non-zero value if the upgrade fails. If the upgrade fails, the map is
567 * returned without a read or write lock held.
569 * Requires that the map be read locked.
572 _vm_map_lock_upgrade(vm_map_t map, const char *file, int line)
574 unsigned int last_timestamp;
576 if (map->system_map) {
577 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
579 if (!sx_try_upgrade_(&map->lock, file, line)) {
580 last_timestamp = map->timestamp;
581 sx_sunlock_(&map->lock, file, line);
582 vm_map_process_deferred();
584 * If the map's timestamp does not change while the
585 * map is unlocked, then the upgrade succeeds.
587 sx_xlock_(&map->lock, file, line);
588 if (last_timestamp != map->timestamp) {
589 sx_xunlock_(&map->lock, file, line);
599 _vm_map_lock_downgrade(vm_map_t map, const char *file, int line)
602 if (map->system_map) {
603 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
605 sx_downgrade_(&map->lock, file, line);
611 * Returns a non-zero value if the caller holds a write (exclusive) lock
612 * on the specified map and the value "0" otherwise.
615 vm_map_locked(vm_map_t map)
619 return (mtx_owned(&map->system_mtx));
621 return (sx_xlocked(&map->lock));
626 _vm_map_assert_locked(vm_map_t map, const char *file, int line)
630 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
632 sx_assert_(&map->lock, SA_XLOCKED, file, line);
635 #define VM_MAP_ASSERT_LOCKED(map) \
636 _vm_map_assert_locked(map, LOCK_FILE, LOCK_LINE)
638 #define VM_MAP_ASSERT_LOCKED(map)
642 * _vm_map_unlock_and_wait:
644 * Atomically releases the lock on the specified map and puts the calling
645 * thread to sleep. The calling thread will remain asleep until either
646 * vm_map_wakeup() is performed on the map or the specified timeout is
649 * WARNING! This function does not perform deferred deallocations of
650 * objects and map entries. Therefore, the calling thread is expected to
651 * reacquire the map lock after reawakening and later perform an ordinary
652 * unlock operation, such as vm_map_unlock(), before completing its
653 * operation on the map.
656 _vm_map_unlock_and_wait(vm_map_t map, int timo, const char *file, int line)
659 mtx_lock(&map_sleep_mtx);
661 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
663 sx_xunlock_(&map->lock, file, line);
664 return (msleep(&map->root, &map_sleep_mtx, PDROP | PVM, "vmmaps",
671 * Awaken any threads that have slept on the map using
672 * vm_map_unlock_and_wait().
675 vm_map_wakeup(vm_map_t map)
679 * Acquire and release map_sleep_mtx to prevent a wakeup()
680 * from being performed (and lost) between the map unlock
681 * and the msleep() in _vm_map_unlock_and_wait().
683 mtx_lock(&map_sleep_mtx);
684 mtx_unlock(&map_sleep_mtx);
689 vm_map_busy(vm_map_t map)
692 VM_MAP_ASSERT_LOCKED(map);
697 vm_map_unbusy(vm_map_t map)
700 VM_MAP_ASSERT_LOCKED(map);
701 KASSERT(map->busy, ("vm_map_unbusy: not busy"));
702 if (--map->busy == 0 && (map->flags & MAP_BUSY_WAKEUP)) {
703 vm_map_modflags(map, 0, MAP_BUSY_WAKEUP);
709 vm_map_wait_busy(vm_map_t map)
712 VM_MAP_ASSERT_LOCKED(map);
714 vm_map_modflags(map, MAP_BUSY_WAKEUP, 0);
716 msleep(&map->busy, &map->system_mtx, 0, "mbusy", 0);
718 sx_sleep(&map->busy, &map->lock, 0, "mbusy", 0);
724 vmspace_resident_count(struct vmspace *vmspace)
726 return pmap_resident_count(vmspace_pmap(vmspace));
730 vmspace_wired_count(struct vmspace *vmspace)
732 return pmap_wired_count(vmspace_pmap(vmspace));
738 * Creates and returns a new empty VM map with
739 * the given physical map structure, and having
740 * the given lower and upper address bounds.
743 vm_map_create(pmap_t pmap, vm_offset_t min, vm_offset_t max)
747 result = uma_zalloc(mapzone, M_WAITOK);
748 CTR1(KTR_VM, "vm_map_create: %p", result);
749 _vm_map_init(result, pmap, min, max);
754 * Initialize an existing vm_map structure
755 * such as that in the vmspace structure.
758 _vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max)
761 map->header.next = map->header.prev = &map->header;
762 map->needs_wakeup = FALSE;
765 map->min_offset = min;
766 map->max_offset = max;
774 vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max)
777 _vm_map_init(map, pmap, min, max);
778 mtx_init(&map->system_mtx, "system map", NULL, MTX_DEF | MTX_DUPOK);
779 sx_init(&map->lock, "user map");
783 * vm_map_entry_dispose: [ internal use only ]
785 * Inverse of vm_map_entry_create.
788 vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry)
790 uma_zfree(map->system_map ? kmapentzone : mapentzone, entry);
794 * vm_map_entry_create: [ internal use only ]
796 * Allocates a VM map entry for insertion.
797 * No entry fields are filled in.
799 static vm_map_entry_t
800 vm_map_entry_create(vm_map_t map)
802 vm_map_entry_t new_entry;
805 new_entry = uma_zalloc(kmapentzone, M_NOWAIT);
807 new_entry = uma_zalloc(mapentzone, M_WAITOK);
808 if (new_entry == NULL)
809 panic("vm_map_entry_create: kernel resources exhausted");
814 * vm_map_entry_set_behavior:
816 * Set the expected access behavior, either normal, random, or
820 vm_map_entry_set_behavior(vm_map_entry_t entry, u_char behavior)
822 entry->eflags = (entry->eflags & ~MAP_ENTRY_BEHAV_MASK) |
823 (behavior & MAP_ENTRY_BEHAV_MASK);
827 * vm_map_entry_set_max_free:
829 * Set the max_free field in a vm_map_entry.
832 vm_map_entry_set_max_free(vm_map_entry_t entry)
835 entry->max_free = entry->adj_free;
836 if (entry->left != NULL && entry->left->max_free > entry->max_free)
837 entry->max_free = entry->left->max_free;
838 if (entry->right != NULL && entry->right->max_free > entry->max_free)
839 entry->max_free = entry->right->max_free;
843 * vm_map_entry_splay:
845 * The Sleator and Tarjan top-down splay algorithm with the
846 * following variation. Max_free must be computed bottom-up, so
847 * on the downward pass, maintain the left and right spines in
848 * reverse order. Then, make a second pass up each side to fix
849 * the pointers and compute max_free. The time bound is O(log n)
852 * The new root is the vm_map_entry containing "addr", or else an
853 * adjacent entry (lower or higher) if addr is not in the tree.
855 * The map must be locked, and leaves it so.
857 * Returns: the new root.
859 static vm_map_entry_t
860 vm_map_entry_splay(vm_offset_t addr, vm_map_entry_t root)
862 vm_map_entry_t llist, rlist;
863 vm_map_entry_t ltree, rtree;
866 /* Special case of empty tree. */
871 * Pass One: Splay down the tree until we find addr or a NULL
872 * pointer where addr would go. llist and rlist are the two
873 * sides in reverse order (bottom-up), with llist linked by
874 * the right pointer and rlist linked by the left pointer in
875 * the vm_map_entry. Wait until Pass Two to set max_free on
881 /* root is never NULL in here. */
882 if (addr < root->start) {
886 if (addr < y->start && y->left != NULL) {
887 /* Rotate right and put y on rlist. */
888 root->left = y->right;
890 vm_map_entry_set_max_free(root);
895 /* Put root on rlist. */
900 } else if (addr >= root->end) {
904 if (addr >= y->end && y->right != NULL) {
905 /* Rotate left and put y on llist. */
906 root->right = y->left;
908 vm_map_entry_set_max_free(root);
913 /* Put root on llist. */
923 * Pass Two: Walk back up the two spines, flip the pointers
924 * and set max_free. The subtrees of the root go at the
925 * bottom of llist and rlist.
928 while (llist != NULL) {
930 llist->right = ltree;
931 vm_map_entry_set_max_free(llist);
936 while (rlist != NULL) {
939 vm_map_entry_set_max_free(rlist);
945 * Final assembly: add ltree and rtree as subtrees of root.
949 vm_map_entry_set_max_free(root);
955 * vm_map_entry_{un,}link:
957 * Insert/remove entries from maps.
960 vm_map_entry_link(vm_map_t map,
961 vm_map_entry_t after_where,
962 vm_map_entry_t entry)
966 "vm_map_entry_link: map %p, nentries %d, entry %p, after %p", map,
967 map->nentries, entry, after_where);
968 VM_MAP_ASSERT_LOCKED(map);
970 entry->prev = after_where;
971 entry->next = after_where->next;
972 entry->next->prev = entry;
973 after_where->next = entry;
975 if (after_where != &map->header) {
976 if (after_where != map->root)
977 vm_map_entry_splay(after_where->start, map->root);
978 entry->right = after_where->right;
979 entry->left = after_where;
980 after_where->right = NULL;
981 after_where->adj_free = entry->start - after_where->end;
982 vm_map_entry_set_max_free(after_where);
984 entry->right = map->root;
987 entry->adj_free = (entry->next == &map->header ? map->max_offset :
988 entry->next->start) - entry->end;
989 vm_map_entry_set_max_free(entry);
994 vm_map_entry_unlink(vm_map_t map,
995 vm_map_entry_t entry)
997 vm_map_entry_t next, prev, root;
999 VM_MAP_ASSERT_LOCKED(map);
1000 if (entry != map->root)
1001 vm_map_entry_splay(entry->start, map->root);
1002 if (entry->left == NULL)
1003 root = entry->right;
1005 root = vm_map_entry_splay(entry->start, entry->left);
1006 root->right = entry->right;
1007 root->adj_free = (entry->next == &map->header ? map->max_offset :
1008 entry->next->start) - root->end;
1009 vm_map_entry_set_max_free(root);
1018 CTR3(KTR_VM, "vm_map_entry_unlink: map %p, nentries %d, entry %p", map,
1019 map->nentries, entry);
1023 * vm_map_entry_resize_free:
1025 * Recompute the amount of free space following a vm_map_entry
1026 * and propagate that value up the tree. Call this function after
1027 * resizing a map entry in-place, that is, without a call to
1028 * vm_map_entry_link() or _unlink().
1030 * The map must be locked, and leaves it so.
1033 vm_map_entry_resize_free(vm_map_t map, vm_map_entry_t entry)
1037 * Using splay trees without parent pointers, propagating
1038 * max_free up the tree is done by moving the entry to the
1039 * root and making the change there.
1041 if (entry != map->root)
1042 map->root = vm_map_entry_splay(entry->start, map->root);
1044 entry->adj_free = (entry->next == &map->header ? map->max_offset :
1045 entry->next->start) - entry->end;
1046 vm_map_entry_set_max_free(entry);
1050 * vm_map_lookup_entry: [ internal use only ]
1052 * Finds the map entry containing (or
1053 * immediately preceding) the specified address
1054 * in the given map; the entry is returned
1055 * in the "entry" parameter. The boolean
1056 * result indicates whether the address is
1057 * actually contained in the map.
1060 vm_map_lookup_entry(
1062 vm_offset_t address,
1063 vm_map_entry_t *entry) /* OUT */
1069 * If the map is empty, then the map entry immediately preceding
1070 * "address" is the map's header.
1074 *entry = &map->header;
1075 else if (address >= cur->start && cur->end > address) {
1078 } else if ((locked = vm_map_locked(map)) ||
1079 sx_try_upgrade(&map->lock)) {
1081 * Splay requires a write lock on the map. However, it only
1082 * restructures the binary search tree; it does not otherwise
1083 * change the map. Thus, the map's timestamp need not change
1084 * on a temporary upgrade.
1086 map->root = cur = vm_map_entry_splay(address, cur);
1088 sx_downgrade(&map->lock);
1091 * If "address" is contained within a map entry, the new root
1092 * is that map entry. Otherwise, the new root is a map entry
1093 * immediately before or after "address".
1095 if (address >= cur->start) {
1097 if (cur->end > address)
1103 * Since the map is only locked for read access, perform a
1104 * standard binary search tree lookup for "address".
1107 if (address < cur->start) {
1108 if (cur->left == NULL) {
1113 } else if (cur->end > address) {
1117 if (cur->right == NULL) {
1130 * Inserts the given whole VM object into the target
1131 * map at the specified address range. The object's
1132 * size should match that of the address range.
1134 * Requires that the map be locked, and leaves it so.
1136 * If object is non-NULL, ref count must be bumped by caller
1137 * prior to making call to account for the new entry.
1140 vm_map_insert(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1141 vm_offset_t start, vm_offset_t end, vm_prot_t prot, vm_prot_t max,
1144 vm_map_entry_t new_entry;
1145 vm_map_entry_t prev_entry;
1146 vm_map_entry_t temp_entry;
1147 vm_eflags_t protoeflags;
1149 vm_inherit_t inheritance;
1150 boolean_t charge_prev_obj;
1152 VM_MAP_ASSERT_LOCKED(map);
1155 * Check that the start and end points are not bogus.
1157 if ((start < map->min_offset) || (end > map->max_offset) ||
1159 return (KERN_INVALID_ADDRESS);
1162 * Find the entry prior to the proposed starting address; if it's part
1163 * of an existing entry, this range is bogus.
1165 if (vm_map_lookup_entry(map, start, &temp_entry))
1166 return (KERN_NO_SPACE);
1168 prev_entry = temp_entry;
1171 * Assert that the next entry doesn't overlap the end point.
1173 if ((prev_entry->next != &map->header) &&
1174 (prev_entry->next->start < end))
1175 return (KERN_NO_SPACE);
1178 charge_prev_obj = FALSE;
1180 if (cow & MAP_COPY_ON_WRITE)
1181 protoeflags |= MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY;
1183 if (cow & MAP_NOFAULT) {
1184 protoeflags |= MAP_ENTRY_NOFAULT;
1186 KASSERT(object == NULL,
1187 ("vm_map_insert: paradoxical MAP_NOFAULT request"));
1189 if (cow & MAP_DISABLE_SYNCER)
1190 protoeflags |= MAP_ENTRY_NOSYNC;
1191 if (cow & MAP_DISABLE_COREDUMP)
1192 protoeflags |= MAP_ENTRY_NOCOREDUMP;
1193 if (cow & MAP_VN_WRITECOUNT)
1194 protoeflags |= MAP_ENTRY_VN_WRITECNT;
1195 if (cow & MAP_INHERIT_SHARE)
1196 inheritance = VM_INHERIT_SHARE;
1198 inheritance = VM_INHERIT_DEFAULT;
1201 KASSERT((object != kmem_object && object != kernel_object) ||
1202 ((object == kmem_object || object == kernel_object) &&
1203 !(protoeflags & MAP_ENTRY_NEEDS_COPY)),
1204 ("kmem or kernel object and cow"));
1205 if (cow & (MAP_ACC_NO_CHARGE | MAP_NOFAULT))
1207 if ((cow & MAP_ACC_CHARGED) || ((prot & VM_PROT_WRITE) &&
1208 ((protoeflags & MAP_ENTRY_NEEDS_COPY) || object == NULL))) {
1209 if (!(cow & MAP_ACC_CHARGED) && !swap_reserve(end - start))
1210 return (KERN_RESOURCE_SHORTAGE);
1211 KASSERT(object == NULL || (protoeflags & MAP_ENTRY_NEEDS_COPY) ||
1212 object->cred == NULL,
1213 ("OVERCOMMIT: vm_map_insert o %p", object));
1214 cred = curthread->td_ucred;
1216 if (object == NULL && !(protoeflags & MAP_ENTRY_NEEDS_COPY))
1217 charge_prev_obj = TRUE;
1221 /* Expand the kernel pmap, if necessary. */
1222 if (map == kernel_map && end > kernel_vm_end)
1223 pmap_growkernel(end);
1224 if (object != NULL) {
1226 * OBJ_ONEMAPPING must be cleared unless this mapping
1227 * is trivially proven to be the only mapping for any
1228 * of the object's pages. (Object granularity
1229 * reference counting is insufficient to recognize
1230 * aliases with precision.)
1232 VM_OBJECT_LOCK(object);
1233 if (object->ref_count > 1 || object->shadow_count != 0)
1234 vm_object_clear_flag(object, OBJ_ONEMAPPING);
1235 VM_OBJECT_UNLOCK(object);
1237 else if ((prev_entry != &map->header) &&
1238 (prev_entry->eflags == protoeflags) &&
1239 (prev_entry->end == start) &&
1240 (prev_entry->wired_count == 0) &&
1241 (prev_entry->cred == cred ||
1242 (prev_entry->object.vm_object != NULL &&
1243 (prev_entry->object.vm_object->cred == cred))) &&
1244 vm_object_coalesce(prev_entry->object.vm_object,
1246 (vm_size_t)(prev_entry->end - prev_entry->start),
1247 (vm_size_t)(end - prev_entry->end), charge_prev_obj)) {
1249 * We were able to extend the object. Determine if we
1250 * can extend the previous map entry to include the
1251 * new range as well.
1253 if ((prev_entry->inheritance == inheritance) &&
1254 (prev_entry->protection == prot) &&
1255 (prev_entry->max_protection == max)) {
1256 map->size += (end - prev_entry->end);
1257 prev_entry->end = end;
1258 vm_map_entry_resize_free(map, prev_entry);
1259 vm_map_simplify_entry(map, prev_entry);
1262 return (KERN_SUCCESS);
1266 * If we can extend the object but cannot extend the
1267 * map entry, we have to create a new map entry. We
1268 * must bump the ref count on the extended object to
1269 * account for it. object may be NULL.
1271 object = prev_entry->object.vm_object;
1272 offset = prev_entry->offset +
1273 (prev_entry->end - prev_entry->start);
1274 vm_object_reference(object);
1275 if (cred != NULL && object != NULL && object->cred != NULL &&
1276 !(prev_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
1277 /* Object already accounts for this uid. */
1284 * NOTE: if conditionals fail, object can be NULL here. This occurs
1285 * in things like the buffer map where we manage kva but do not manage
1290 * Create a new entry
1292 new_entry = vm_map_entry_create(map);
1293 new_entry->start = start;
1294 new_entry->end = end;
1295 new_entry->cred = NULL;
1297 new_entry->eflags = protoeflags;
1298 new_entry->object.vm_object = object;
1299 new_entry->offset = offset;
1300 new_entry->avail_ssize = 0;
1302 new_entry->inheritance = inheritance;
1303 new_entry->protection = prot;
1304 new_entry->max_protection = max;
1305 new_entry->wired_count = 0;
1306 new_entry->read_ahead = VM_FAULT_READ_AHEAD_INIT;
1307 new_entry->next_read = OFF_TO_IDX(offset);
1309 KASSERT(cred == NULL || !ENTRY_CHARGED(new_entry),
1310 ("OVERCOMMIT: vm_map_insert leaks vm_map %p", new_entry));
1311 new_entry->cred = cred;
1314 * Insert the new entry into the list
1316 vm_map_entry_link(map, prev_entry, new_entry);
1317 map->size += new_entry->end - new_entry->start;
1320 * It may be possible to merge the new entry with the next and/or
1321 * previous entries. However, due to MAP_STACK_* being a hack, a
1322 * panic can result from merging such entries.
1324 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0)
1325 vm_map_simplify_entry(map, new_entry);
1327 if (cow & (MAP_PREFAULT|MAP_PREFAULT_PARTIAL)) {
1328 vm_map_pmap_enter(map, start, prot,
1329 object, OFF_TO_IDX(offset), end - start,
1330 cow & MAP_PREFAULT_PARTIAL);
1333 return (KERN_SUCCESS);
1339 * Find the first fit (lowest VM address) for "length" free bytes
1340 * beginning at address >= start in the given map.
1342 * In a vm_map_entry, "adj_free" is the amount of free space
1343 * adjacent (higher address) to this entry, and "max_free" is the
1344 * maximum amount of contiguous free space in its subtree. This
1345 * allows finding a free region in one path down the tree, so
1346 * O(log n) amortized with splay trees.
1348 * The map must be locked, and leaves it so.
1350 * Returns: 0 on success, and starting address in *addr,
1351 * 1 if insufficient space.
1354 vm_map_findspace(vm_map_t map, vm_offset_t start, vm_size_t length,
1355 vm_offset_t *addr) /* OUT */
1357 vm_map_entry_t entry;
1361 * Request must fit within min/max VM address and must avoid
1364 if (start < map->min_offset)
1365 start = map->min_offset;
1366 if (start + length > map->max_offset || start + length < start)
1369 /* Empty tree means wide open address space. */
1370 if (map->root == NULL) {
1376 * After splay, if start comes before root node, then there
1377 * must be a gap from start to the root.
1379 map->root = vm_map_entry_splay(start, map->root);
1380 if (start + length <= map->root->start) {
1386 * Root is the last node that might begin its gap before
1387 * start, and this is the last comparison where address
1388 * wrap might be a problem.
1390 st = (start > map->root->end) ? start : map->root->end;
1391 if (length <= map->root->end + map->root->adj_free - st) {
1396 /* With max_free, can immediately tell if no solution. */
1397 entry = map->root->right;
1398 if (entry == NULL || length > entry->max_free)
1402 * Search the right subtree in the order: left subtree, root,
1403 * right subtree (first fit). The previous splay implies that
1404 * all regions in the right subtree have addresses > start.
1406 while (entry != NULL) {
1407 if (entry->left != NULL && entry->left->max_free >= length)
1408 entry = entry->left;
1409 else if (entry->adj_free >= length) {
1413 entry = entry->right;
1416 /* Can't get here, so panic if we do. */
1417 panic("vm_map_findspace: max_free corrupt");
1421 vm_map_fixed(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1422 vm_offset_t start, vm_size_t length, vm_prot_t prot,
1423 vm_prot_t max, int cow)
1428 end = start + length;
1430 VM_MAP_RANGE_CHECK(map, start, end);
1431 (void) vm_map_delete(map, start, end);
1432 result = vm_map_insert(map, object, offset, start, end, prot,
1439 * vm_map_find finds an unallocated region in the target address
1440 * map with the given length. The search is defined to be
1441 * first-fit from the specified address; the region found is
1442 * returned in the same parameter.
1444 * If object is non-NULL, ref count must be bumped by caller
1445 * prior to making call to account for the new entry.
1448 vm_map_find(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1449 vm_offset_t *addr, /* IN/OUT */
1450 vm_size_t length, int find_space, vm_prot_t prot,
1451 vm_prot_t max, int cow)
1459 if (find_space != VMFS_NO_SPACE) {
1460 if (vm_map_findspace(map, start, length, addr)) {
1462 return (KERN_NO_SPACE);
1464 switch (find_space) {
1465 case VMFS_ALIGNED_SPACE:
1466 pmap_align_superpage(object, offset, addr,
1469 #ifdef VMFS_TLB_ALIGNED_SPACE
1470 case VMFS_TLB_ALIGNED_SPACE:
1471 pmap_align_tlb(addr);
1480 result = vm_map_insert(map, object, offset, start, start +
1481 length, prot, max, cow);
1482 } while (result == KERN_NO_SPACE && (find_space == VMFS_ALIGNED_SPACE
1483 #ifdef VMFS_TLB_ALIGNED_SPACE
1484 || find_space == VMFS_TLB_ALIGNED_SPACE
1492 * vm_map_simplify_entry:
1494 * Simplify the given map entry by merging with either neighbor. This
1495 * routine also has the ability to merge with both neighbors.
1497 * The map must be locked.
1499 * This routine guarentees that the passed entry remains valid (though
1500 * possibly extended). When merging, this routine may delete one or
1504 vm_map_simplify_entry(vm_map_t map, vm_map_entry_t entry)
1506 vm_map_entry_t next, prev;
1507 vm_size_t prevsize, esize;
1509 if (entry->eflags & (MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_IS_SUB_MAP))
1513 if (prev != &map->header) {
1514 prevsize = prev->end - prev->start;
1515 if ( (prev->end == entry->start) &&
1516 (prev->object.vm_object == entry->object.vm_object) &&
1517 (!prev->object.vm_object ||
1518 (prev->offset + prevsize == entry->offset)) &&
1519 (prev->eflags == entry->eflags) &&
1520 (prev->protection == entry->protection) &&
1521 (prev->max_protection == entry->max_protection) &&
1522 (prev->inheritance == entry->inheritance) &&
1523 (prev->wired_count == entry->wired_count) &&
1524 (prev->cred == entry->cred)) {
1525 vm_map_entry_unlink(map, prev);
1526 entry->start = prev->start;
1527 entry->offset = prev->offset;
1528 if (entry->prev != &map->header)
1529 vm_map_entry_resize_free(map, entry->prev);
1532 * If the backing object is a vnode object,
1533 * vm_object_deallocate() calls vrele().
1534 * However, vrele() does not lock the vnode
1535 * because the vnode has additional
1536 * references. Thus, the map lock can be kept
1537 * without causing a lock-order reversal with
1540 * Since we count the number of virtual page
1541 * mappings in object->un_pager.vnp.writemappings,
1542 * the writemappings value should not be adjusted
1543 * when the entry is disposed of.
1545 if (prev->object.vm_object)
1546 vm_object_deallocate(prev->object.vm_object);
1547 if (prev->cred != NULL)
1549 vm_map_entry_dispose(map, prev);
1554 if (next != &map->header) {
1555 esize = entry->end - entry->start;
1556 if ((entry->end == next->start) &&
1557 (next->object.vm_object == entry->object.vm_object) &&
1558 (!entry->object.vm_object ||
1559 (entry->offset + esize == next->offset)) &&
1560 (next->eflags == entry->eflags) &&
1561 (next->protection == entry->protection) &&
1562 (next->max_protection == entry->max_protection) &&
1563 (next->inheritance == entry->inheritance) &&
1564 (next->wired_count == entry->wired_count) &&
1565 (next->cred == entry->cred)) {
1566 vm_map_entry_unlink(map, next);
1567 entry->end = next->end;
1568 vm_map_entry_resize_free(map, entry);
1571 * See comment above.
1573 if (next->object.vm_object)
1574 vm_object_deallocate(next->object.vm_object);
1575 if (next->cred != NULL)
1577 vm_map_entry_dispose(map, next);
1582 * vm_map_clip_start: [ internal use only ]
1584 * Asserts that the given entry begins at or after
1585 * the specified address; if necessary,
1586 * it splits the entry into two.
1588 #define vm_map_clip_start(map, entry, startaddr) \
1590 if (startaddr > entry->start) \
1591 _vm_map_clip_start(map, entry, startaddr); \
1595 * This routine is called only when it is known that
1596 * the entry must be split.
1599 _vm_map_clip_start(vm_map_t map, vm_map_entry_t entry, vm_offset_t start)
1601 vm_map_entry_t new_entry;
1603 VM_MAP_ASSERT_LOCKED(map);
1606 * Split off the front portion -- note that we must insert the new
1607 * entry BEFORE this one, so that this entry has the specified
1610 vm_map_simplify_entry(map, entry);
1613 * If there is no object backing this entry, we might as well create
1614 * one now. If we defer it, an object can get created after the map
1615 * is clipped, and individual objects will be created for the split-up
1616 * map. This is a bit of a hack, but is also about the best place to
1617 * put this improvement.
1619 if (entry->object.vm_object == NULL && !map->system_map) {
1621 object = vm_object_allocate(OBJT_DEFAULT,
1622 atop(entry->end - entry->start));
1623 entry->object.vm_object = object;
1625 if (entry->cred != NULL) {
1626 object->cred = entry->cred;
1627 object->charge = entry->end - entry->start;
1630 } else if (entry->object.vm_object != NULL &&
1631 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) &&
1632 entry->cred != NULL) {
1633 VM_OBJECT_LOCK(entry->object.vm_object);
1634 KASSERT(entry->object.vm_object->cred == NULL,
1635 ("OVERCOMMIT: vm_entry_clip_start: both cred e %p", entry));
1636 entry->object.vm_object->cred = entry->cred;
1637 entry->object.vm_object->charge = entry->end - entry->start;
1638 VM_OBJECT_UNLOCK(entry->object.vm_object);
1642 new_entry = vm_map_entry_create(map);
1643 *new_entry = *entry;
1645 new_entry->end = start;
1646 entry->offset += (start - entry->start);
1647 entry->start = start;
1648 if (new_entry->cred != NULL)
1649 crhold(entry->cred);
1651 vm_map_entry_link(map, entry->prev, new_entry);
1653 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1654 vm_object_reference(new_entry->object.vm_object);
1656 * The object->un_pager.vnp.writemappings for the
1657 * object of MAP_ENTRY_VN_WRITECNT type entry shall be
1658 * kept as is here. The virtual pages are
1659 * re-distributed among the clipped entries, so the sum is
1666 * vm_map_clip_end: [ internal use only ]
1668 * Asserts that the given entry ends at or before
1669 * the specified address; if necessary,
1670 * it splits the entry into two.
1672 #define vm_map_clip_end(map, entry, endaddr) \
1674 if ((endaddr) < (entry->end)) \
1675 _vm_map_clip_end((map), (entry), (endaddr)); \
1679 * This routine is called only when it is known that
1680 * the entry must be split.
1683 _vm_map_clip_end(vm_map_t map, vm_map_entry_t entry, vm_offset_t end)
1685 vm_map_entry_t new_entry;
1687 VM_MAP_ASSERT_LOCKED(map);
1690 * If there is no object backing this entry, we might as well create
1691 * one now. If we defer it, an object can get created after the map
1692 * is clipped, and individual objects will be created for the split-up
1693 * map. This is a bit of a hack, but is also about the best place to
1694 * put this improvement.
1696 if (entry->object.vm_object == NULL && !map->system_map) {
1698 object = vm_object_allocate(OBJT_DEFAULT,
1699 atop(entry->end - entry->start));
1700 entry->object.vm_object = object;
1702 if (entry->cred != NULL) {
1703 object->cred = entry->cred;
1704 object->charge = entry->end - entry->start;
1707 } else if (entry->object.vm_object != NULL &&
1708 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) &&
1709 entry->cred != NULL) {
1710 VM_OBJECT_LOCK(entry->object.vm_object);
1711 KASSERT(entry->object.vm_object->cred == NULL,
1712 ("OVERCOMMIT: vm_entry_clip_end: both cred e %p", entry));
1713 entry->object.vm_object->cred = entry->cred;
1714 entry->object.vm_object->charge = entry->end - entry->start;
1715 VM_OBJECT_UNLOCK(entry->object.vm_object);
1720 * Create a new entry and insert it AFTER the specified entry
1722 new_entry = vm_map_entry_create(map);
1723 *new_entry = *entry;
1725 new_entry->start = entry->end = end;
1726 new_entry->offset += (end - entry->start);
1727 if (new_entry->cred != NULL)
1728 crhold(entry->cred);
1730 vm_map_entry_link(map, entry, new_entry);
1732 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1733 vm_object_reference(new_entry->object.vm_object);
1738 * vm_map_submap: [ kernel use only ]
1740 * Mark the given range as handled by a subordinate map.
1742 * This range must have been created with vm_map_find,
1743 * and no other operations may have been performed on this
1744 * range prior to calling vm_map_submap.
1746 * Only a limited number of operations can be performed
1747 * within this rage after calling vm_map_submap:
1749 * [Don't try vm_map_copy!]
1751 * To remove a submapping, one must first remove the
1752 * range from the superior map, and then destroy the
1753 * submap (if desired). [Better yet, don't try it.]
1762 vm_map_entry_t entry;
1763 int result = KERN_INVALID_ARGUMENT;
1767 VM_MAP_RANGE_CHECK(map, start, end);
1769 if (vm_map_lookup_entry(map, start, &entry)) {
1770 vm_map_clip_start(map, entry, start);
1772 entry = entry->next;
1774 vm_map_clip_end(map, entry, end);
1776 if ((entry->start == start) && (entry->end == end) &&
1777 ((entry->eflags & MAP_ENTRY_COW) == 0) &&
1778 (entry->object.vm_object == NULL)) {
1779 entry->object.sub_map = submap;
1780 entry->eflags |= MAP_ENTRY_IS_SUB_MAP;
1781 result = KERN_SUCCESS;
1789 * The maximum number of pages to map
1791 #define MAX_INIT_PT 96
1794 * vm_map_pmap_enter:
1796 * Preload read-only mappings for the specified object's resident pages
1797 * into the target map. If "flags" is MAP_PREFAULT_PARTIAL, then only
1798 * the resident pages within the address range [addr, addr + ulmin(size,
1799 * ptoa(MAX_INIT_PT))) are mapped. Otherwise, all resident pages within
1800 * the specified address range are mapped. This eliminates many soft
1801 * faults on process startup and immediately after an mmap(2). Because
1802 * these are speculative mappings, cached pages are not reactivated and
1806 vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot,
1807 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags)
1810 vm_page_t p, p_start;
1811 vm_pindex_t psize, tmpidx;
1813 if ((prot & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0 || object == NULL)
1815 VM_OBJECT_LOCK(object);
1816 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) {
1817 pmap_object_init_pt(map->pmap, addr, object, pindex, size);
1822 if (psize > MAX_INIT_PT && (flags & MAP_PREFAULT_PARTIAL) != 0)
1823 psize = MAX_INIT_PT;
1824 if (psize + pindex > object->size) {
1825 if (object->size < pindex)
1827 psize = object->size - pindex;
1833 p = vm_page_find_least(object, pindex);
1835 * Assert: the variable p is either (1) the page with the
1836 * least pindex greater than or equal to the parameter pindex
1840 p != NULL && (tmpidx = p->pindex - pindex) < psize;
1841 p = TAILQ_NEXT(p, listq)) {
1843 * don't allow an madvise to blow away our really
1844 * free pages allocating pv entries.
1846 if ((flags & MAP_PREFAULT_MADVISE) &&
1847 cnt.v_free_count < cnt.v_free_reserved) {
1851 if (p->valid == VM_PAGE_BITS_ALL) {
1852 if (p_start == NULL) {
1853 start = addr + ptoa(tmpidx);
1856 } else if (p_start != NULL) {
1857 pmap_enter_object(map->pmap, start, addr +
1858 ptoa(tmpidx), p_start, prot);
1862 if (p_start != NULL)
1863 pmap_enter_object(map->pmap, start, addr + ptoa(psize),
1866 VM_OBJECT_UNLOCK(object);
1872 * Sets the protection of the specified address
1873 * region in the target map. If "set_max" is
1874 * specified, the maximum protection is to be set;
1875 * otherwise, only the current protection is affected.
1878 vm_map_protect(vm_map_t map, vm_offset_t start, vm_offset_t end,
1879 vm_prot_t new_prot, boolean_t set_max)
1881 vm_map_entry_t current, entry;
1888 VM_MAP_RANGE_CHECK(map, start, end);
1890 if (vm_map_lookup_entry(map, start, &entry)) {
1891 vm_map_clip_start(map, entry, start);
1893 entry = entry->next;
1897 * Make a first pass to check for protection violations.
1900 while ((current != &map->header) && (current->start < end)) {
1901 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
1903 return (KERN_INVALID_ARGUMENT);
1905 if ((new_prot & current->max_protection) != new_prot) {
1907 return (KERN_PROTECTION_FAILURE);
1909 current = current->next;
1914 * Do an accounting pass for private read-only mappings that
1915 * now will do cow due to allowed write (e.g. debugger sets
1916 * breakpoint on text segment)
1918 for (current = entry; (current != &map->header) &&
1919 (current->start < end); current = current->next) {
1921 vm_map_clip_end(map, current, end);
1924 ((new_prot & ~(current->protection)) & VM_PROT_WRITE) == 0 ||
1925 ENTRY_CHARGED(current)) {
1929 cred = curthread->td_ucred;
1930 obj = current->object.vm_object;
1932 if (obj == NULL || (current->eflags & MAP_ENTRY_NEEDS_COPY)) {
1933 if (!swap_reserve(current->end - current->start)) {
1935 return (KERN_RESOURCE_SHORTAGE);
1938 current->cred = cred;
1942 VM_OBJECT_LOCK(obj);
1943 if (obj->type != OBJT_DEFAULT && obj->type != OBJT_SWAP) {
1944 VM_OBJECT_UNLOCK(obj);
1949 * Charge for the whole object allocation now, since
1950 * we cannot distinguish between non-charged and
1951 * charged clipped mapping of the same object later.
1953 KASSERT(obj->charge == 0,
1954 ("vm_map_protect: object %p overcharged\n", obj));
1955 if (!swap_reserve(ptoa(obj->size))) {
1956 VM_OBJECT_UNLOCK(obj);
1958 return (KERN_RESOURCE_SHORTAGE);
1963 obj->charge = ptoa(obj->size);
1964 VM_OBJECT_UNLOCK(obj);
1968 * Go back and fix up protections. [Note that clipping is not
1969 * necessary the second time.]
1972 while ((current != &map->header) && (current->start < end)) {
1973 old_prot = current->protection;
1976 current->protection =
1977 (current->max_protection = new_prot) &
1980 current->protection = new_prot;
1982 if ((current->eflags & (MAP_ENTRY_COW | MAP_ENTRY_USER_WIRED))
1983 == (MAP_ENTRY_COW | MAP_ENTRY_USER_WIRED) &&
1984 (current->protection & VM_PROT_WRITE) != 0 &&
1985 (old_prot & VM_PROT_WRITE) == 0) {
1986 vm_fault_copy_entry(map, map, current, current, NULL);
1990 * When restricting access, update the physical map. Worry
1991 * about copy-on-write here.
1993 if ((old_prot & ~current->protection) != 0) {
1994 #define MASK(entry) (((entry)->eflags & MAP_ENTRY_COW) ? ~VM_PROT_WRITE : \
1996 pmap_protect(map->pmap, current->start,
1998 current->protection & MASK(current));
2001 vm_map_simplify_entry(map, current);
2002 current = current->next;
2005 return (KERN_SUCCESS);
2011 * This routine traverses a processes map handling the madvise
2012 * system call. Advisories are classified as either those effecting
2013 * the vm_map_entry structure, or those effecting the underlying
2023 vm_map_entry_t current, entry;
2027 * Some madvise calls directly modify the vm_map_entry, in which case
2028 * we need to use an exclusive lock on the map and we need to perform
2029 * various clipping operations. Otherwise we only need a read-lock
2034 case MADV_SEQUENTIAL:
2046 vm_map_lock_read(map);
2049 return (KERN_INVALID_ARGUMENT);
2053 * Locate starting entry and clip if necessary.
2055 VM_MAP_RANGE_CHECK(map, start, end);
2057 if (vm_map_lookup_entry(map, start, &entry)) {
2059 vm_map_clip_start(map, entry, start);
2061 entry = entry->next;
2066 * madvise behaviors that are implemented in the vm_map_entry.
2068 * We clip the vm_map_entry so that behavioral changes are
2069 * limited to the specified address range.
2071 for (current = entry;
2072 (current != &map->header) && (current->start < end);
2073 current = current->next
2075 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
2078 vm_map_clip_end(map, current, end);
2082 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_NORMAL);
2084 case MADV_SEQUENTIAL:
2085 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_SEQUENTIAL);
2088 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_RANDOM);
2091 current->eflags |= MAP_ENTRY_NOSYNC;
2094 current->eflags &= ~MAP_ENTRY_NOSYNC;
2097 current->eflags |= MAP_ENTRY_NOCOREDUMP;
2100 current->eflags &= ~MAP_ENTRY_NOCOREDUMP;
2105 vm_map_simplify_entry(map, current);
2109 vm_pindex_t pstart, pend;
2112 * madvise behaviors that are implemented in the underlying
2115 * Since we don't clip the vm_map_entry, we have to clip
2116 * the vm_object pindex and count.
2118 for (current = entry;
2119 (current != &map->header) && (current->start < end);
2120 current = current->next
2122 vm_offset_t useStart;
2124 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
2127 pstart = OFF_TO_IDX(current->offset);
2128 pend = pstart + atop(current->end - current->start);
2129 useStart = current->start;
2131 if (current->start < start) {
2132 pstart += atop(start - current->start);
2135 if (current->end > end)
2136 pend -= atop(current->end - end);
2141 vm_object_madvise(current->object.vm_object, pstart,
2143 if (behav == MADV_WILLNEED) {
2144 vm_map_pmap_enter(map,
2146 current->protection,
2147 current->object.vm_object,
2149 ptoa(pend - pstart),
2150 MAP_PREFAULT_MADVISE
2154 vm_map_unlock_read(map);
2163 * Sets the inheritance of the specified address
2164 * range in the target map. Inheritance
2165 * affects how the map will be shared with
2166 * child maps at the time of vmspace_fork.
2169 vm_map_inherit(vm_map_t map, vm_offset_t start, vm_offset_t end,
2170 vm_inherit_t new_inheritance)
2172 vm_map_entry_t entry;
2173 vm_map_entry_t temp_entry;
2175 switch (new_inheritance) {
2176 case VM_INHERIT_NONE:
2177 case VM_INHERIT_COPY:
2178 case VM_INHERIT_SHARE:
2181 return (KERN_INVALID_ARGUMENT);
2184 VM_MAP_RANGE_CHECK(map, start, end);
2185 if (vm_map_lookup_entry(map, start, &temp_entry)) {
2187 vm_map_clip_start(map, entry, start);
2189 entry = temp_entry->next;
2190 while ((entry != &map->header) && (entry->start < end)) {
2191 vm_map_clip_end(map, entry, end);
2192 entry->inheritance = new_inheritance;
2193 vm_map_simplify_entry(map, entry);
2194 entry = entry->next;
2197 return (KERN_SUCCESS);
2203 * Implements both kernel and user unwiring.
2206 vm_map_unwire(vm_map_t map, vm_offset_t start, vm_offset_t end,
2209 vm_map_entry_t entry, first_entry, tmp_entry;
2210 vm_offset_t saved_start;
2211 unsigned int last_timestamp;
2213 boolean_t need_wakeup, result, user_unwire;
2215 user_unwire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE;
2217 VM_MAP_RANGE_CHECK(map, start, end);
2218 if (!vm_map_lookup_entry(map, start, &first_entry)) {
2219 if (flags & VM_MAP_WIRE_HOLESOK)
2220 first_entry = first_entry->next;
2223 return (KERN_INVALID_ADDRESS);
2226 last_timestamp = map->timestamp;
2227 entry = first_entry;
2228 while (entry != &map->header && entry->start < end) {
2229 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
2231 * We have not yet clipped the entry.
2233 saved_start = (start >= entry->start) ? start :
2235 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2236 if (vm_map_unlock_and_wait(map, 0)) {
2238 * Allow interruption of user unwiring?
2242 if (last_timestamp+1 != map->timestamp) {
2244 * Look again for the entry because the map was
2245 * modified while it was unlocked.
2246 * Specifically, the entry may have been
2247 * clipped, merged, or deleted.
2249 if (!vm_map_lookup_entry(map, saved_start,
2251 if (flags & VM_MAP_WIRE_HOLESOK)
2252 tmp_entry = tmp_entry->next;
2254 if (saved_start == start) {
2256 * First_entry has been deleted.
2259 return (KERN_INVALID_ADDRESS);
2262 rv = KERN_INVALID_ADDRESS;
2266 if (entry == first_entry)
2267 first_entry = tmp_entry;
2272 last_timestamp = map->timestamp;
2275 vm_map_clip_start(map, entry, start);
2276 vm_map_clip_end(map, entry, end);
2278 * Mark the entry in case the map lock is released. (See
2281 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
2283 * Check the map for holes in the specified region.
2284 * If VM_MAP_WIRE_HOLESOK was specified, skip this check.
2286 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) &&
2287 (entry->end < end && (entry->next == &map->header ||
2288 entry->next->start > entry->end))) {
2290 rv = KERN_INVALID_ADDRESS;
2294 * If system unwiring, require that the entry is system wired.
2297 vm_map_entry_system_wired_count(entry) == 0) {
2299 rv = KERN_INVALID_ARGUMENT;
2302 entry = entry->next;
2306 need_wakeup = FALSE;
2307 if (first_entry == NULL) {
2308 result = vm_map_lookup_entry(map, start, &first_entry);
2309 if (!result && (flags & VM_MAP_WIRE_HOLESOK))
2310 first_entry = first_entry->next;
2312 KASSERT(result, ("vm_map_unwire: lookup failed"));
2314 entry = first_entry;
2315 while (entry != &map->header && entry->start < end) {
2316 if (rv == KERN_SUCCESS && (!user_unwire ||
2317 (entry->eflags & MAP_ENTRY_USER_WIRED))) {
2319 entry->eflags &= ~MAP_ENTRY_USER_WIRED;
2320 entry->wired_count--;
2321 if (entry->wired_count == 0) {
2323 * Retain the map lock.
2325 vm_fault_unwire(map, entry->start, entry->end,
2326 entry->object.vm_object != NULL &&
2327 (entry->object.vm_object->flags &
2328 OBJ_FICTITIOUS) != 0);
2331 KASSERT(entry->eflags & MAP_ENTRY_IN_TRANSITION,
2332 ("vm_map_unwire: in-transition flag missing"));
2333 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION;
2334 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
2335 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
2338 vm_map_simplify_entry(map, entry);
2339 entry = entry->next;
2350 * Implements both kernel and user wiring.
2353 vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t end,
2356 vm_map_entry_t entry, first_entry, tmp_entry;
2357 vm_offset_t saved_end, saved_start;
2358 unsigned int last_timestamp;
2360 boolean_t fictitious, need_wakeup, result, user_wire;
2364 if (flags & VM_MAP_WIRE_WRITE)
2365 prot |= VM_PROT_WRITE;
2366 user_wire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE;
2368 VM_MAP_RANGE_CHECK(map, start, end);
2369 if (!vm_map_lookup_entry(map, start, &first_entry)) {
2370 if (flags & VM_MAP_WIRE_HOLESOK)
2371 first_entry = first_entry->next;
2374 return (KERN_INVALID_ADDRESS);
2377 last_timestamp = map->timestamp;
2378 entry = first_entry;
2379 while (entry != &map->header && entry->start < end) {
2380 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
2382 * We have not yet clipped the entry.
2384 saved_start = (start >= entry->start) ? start :
2386 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2387 if (vm_map_unlock_and_wait(map, 0)) {
2389 * Allow interruption of user wiring?
2393 if (last_timestamp + 1 != map->timestamp) {
2395 * Look again for the entry because the map was
2396 * modified while it was unlocked.
2397 * Specifically, the entry may have been
2398 * clipped, merged, or deleted.
2400 if (!vm_map_lookup_entry(map, saved_start,
2402 if (flags & VM_MAP_WIRE_HOLESOK)
2403 tmp_entry = tmp_entry->next;
2405 if (saved_start == start) {
2407 * first_entry has been deleted.
2410 return (KERN_INVALID_ADDRESS);
2413 rv = KERN_INVALID_ADDRESS;
2417 if (entry == first_entry)
2418 first_entry = tmp_entry;
2423 last_timestamp = map->timestamp;
2426 vm_map_clip_start(map, entry, start);
2427 vm_map_clip_end(map, entry, end);
2429 * Mark the entry in case the map lock is released. (See
2432 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
2433 if ((entry->protection & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0
2434 || (entry->protection & prot) != prot) {
2435 entry->eflags |= MAP_ENTRY_WIRE_SKIPPED;
2436 if ((flags & VM_MAP_WIRE_HOLESOK) == 0) {
2438 rv = KERN_INVALID_ADDRESS;
2443 if (entry->wired_count == 0) {
2444 entry->wired_count++;
2445 saved_start = entry->start;
2446 saved_end = entry->end;
2447 fictitious = entry->object.vm_object != NULL &&
2448 (entry->object.vm_object->flags &
2449 OBJ_FICTITIOUS) != 0;
2451 * Release the map lock, relying on the in-transition
2452 * mark. Mark the map busy for fork.
2456 rv = vm_fault_wire(map, saved_start, saved_end,
2460 if (last_timestamp + 1 != map->timestamp) {
2462 * Look again for the entry because the map was
2463 * modified while it was unlocked. The entry
2464 * may have been clipped, but NOT merged or
2467 result = vm_map_lookup_entry(map, saved_start,
2469 KASSERT(result, ("vm_map_wire: lookup failed"));
2470 if (entry == first_entry)
2471 first_entry = tmp_entry;
2475 while (entry->end < saved_end) {
2476 if (rv != KERN_SUCCESS) {
2477 KASSERT(entry->wired_count == 1,
2478 ("vm_map_wire: bad count"));
2479 entry->wired_count = -1;
2481 entry = entry->next;
2484 last_timestamp = map->timestamp;
2485 if (rv != KERN_SUCCESS) {
2486 KASSERT(entry->wired_count == 1,
2487 ("vm_map_wire: bad count"));
2489 * Assign an out-of-range value to represent
2490 * the failure to wire this entry.
2492 entry->wired_count = -1;
2496 } else if (!user_wire ||
2497 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
2498 entry->wired_count++;
2501 * Check the map for holes in the specified region.
2502 * If VM_MAP_WIRE_HOLESOK was specified, skip this check.
2505 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) &&
2506 (entry->end < end && (entry->next == &map->header ||
2507 entry->next->start > entry->end))) {
2509 rv = KERN_INVALID_ADDRESS;
2512 entry = entry->next;
2516 need_wakeup = FALSE;
2517 if (first_entry == NULL) {
2518 result = vm_map_lookup_entry(map, start, &first_entry);
2519 if (!result && (flags & VM_MAP_WIRE_HOLESOK))
2520 first_entry = first_entry->next;
2522 KASSERT(result, ("vm_map_wire: lookup failed"));
2524 entry = first_entry;
2525 while (entry != &map->header && entry->start < end) {
2526 if ((entry->eflags & MAP_ENTRY_WIRE_SKIPPED) != 0)
2527 goto next_entry_done;
2528 if (rv == KERN_SUCCESS) {
2530 entry->eflags |= MAP_ENTRY_USER_WIRED;
2531 } else if (entry->wired_count == -1) {
2533 * Wiring failed on this entry. Thus, unwiring is
2536 entry->wired_count = 0;
2539 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0)
2540 entry->wired_count--;
2541 if (entry->wired_count == 0) {
2543 * Retain the map lock.
2545 vm_fault_unwire(map, entry->start, entry->end,
2546 entry->object.vm_object != NULL &&
2547 (entry->object.vm_object->flags &
2548 OBJ_FICTITIOUS) != 0);
2552 KASSERT(entry->eflags & MAP_ENTRY_IN_TRANSITION,
2553 ("vm_map_wire: in-transition flag missing"));
2554 entry->eflags &= ~(MAP_ENTRY_IN_TRANSITION|MAP_ENTRY_WIRE_SKIPPED);
2555 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
2556 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
2559 vm_map_simplify_entry(map, entry);
2560 entry = entry->next;
2571 * Push any dirty cached pages in the address range to their pager.
2572 * If syncio is TRUE, dirty pages are written synchronously.
2573 * If invalidate is TRUE, any cached pages are freed as well.
2575 * If the size of the region from start to end is zero, we are
2576 * supposed to flush all modified pages within the region containing
2577 * start. Unfortunately, a region can be split or coalesced with
2578 * neighboring regions, making it difficult to determine what the
2579 * original region was. Therefore, we approximate this requirement by
2580 * flushing the current region containing start.
2582 * Returns an error if any part of the specified range is not mapped.
2590 boolean_t invalidate)
2592 vm_map_entry_t current;
2593 vm_map_entry_t entry;
2596 vm_ooffset_t offset;
2597 unsigned int last_timestamp;
2600 vm_map_lock_read(map);
2601 VM_MAP_RANGE_CHECK(map, start, end);
2602 if (!vm_map_lookup_entry(map, start, &entry)) {
2603 vm_map_unlock_read(map);
2604 return (KERN_INVALID_ADDRESS);
2605 } else if (start == end) {
2606 start = entry->start;
2610 * Make a first pass to check for user-wired memory and holes.
2612 for (current = entry; current != &map->header && current->start < end;
2613 current = current->next) {
2614 if (invalidate && (current->eflags & MAP_ENTRY_USER_WIRED)) {
2615 vm_map_unlock_read(map);
2616 return (KERN_INVALID_ARGUMENT);
2618 if (end > current->end &&
2619 (current->next == &map->header ||
2620 current->end != current->next->start)) {
2621 vm_map_unlock_read(map);
2622 return (KERN_INVALID_ADDRESS);
2627 pmap_remove(map->pmap, start, end);
2631 * Make a second pass, cleaning/uncaching pages from the indicated
2634 for (current = entry; current != &map->header && current->start < end;) {
2635 offset = current->offset + (start - current->start);
2636 size = (end <= current->end ? end : current->end) - start;
2637 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
2639 vm_map_entry_t tentry;
2642 smap = current->object.sub_map;
2643 vm_map_lock_read(smap);
2644 (void) vm_map_lookup_entry(smap, offset, &tentry);
2645 tsize = tentry->end - offset;
2648 object = tentry->object.vm_object;
2649 offset = tentry->offset + (offset - tentry->start);
2650 vm_map_unlock_read(smap);
2652 object = current->object.vm_object;
2654 vm_object_reference(object);
2655 last_timestamp = map->timestamp;
2656 vm_map_unlock_read(map);
2657 if (!vm_object_sync(object, offset, size, syncio, invalidate))
2660 vm_object_deallocate(object);
2661 vm_map_lock_read(map);
2662 if (last_timestamp == map->timestamp ||
2663 !vm_map_lookup_entry(map, start, ¤t))
2664 current = current->next;
2667 vm_map_unlock_read(map);
2668 return (failed ? KERN_FAILURE : KERN_SUCCESS);
2672 * vm_map_entry_unwire: [ internal use only ]
2674 * Make the region specified by this entry pageable.
2676 * The map in question should be locked.
2677 * [This is the reason for this routine's existence.]
2680 vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry)
2682 vm_fault_unwire(map, entry->start, entry->end,
2683 entry->object.vm_object != NULL &&
2684 (entry->object.vm_object->flags & OBJ_FICTITIOUS) != 0);
2685 entry->wired_count = 0;
2689 vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map)
2692 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0)
2693 vm_object_deallocate(entry->object.vm_object);
2694 uma_zfree(system_map ? kmapentzone : mapentzone, entry);
2698 * vm_map_entry_delete: [ internal use only ]
2700 * Deallocate the given entry from the target map.
2703 vm_map_entry_delete(vm_map_t map, vm_map_entry_t entry)
2706 vm_pindex_t offidxstart, offidxend, count, size1;
2709 vm_map_entry_unlink(map, entry);
2710 object = entry->object.vm_object;
2711 size = entry->end - entry->start;
2714 if (entry->cred != NULL) {
2715 swap_release_by_cred(size, entry->cred);
2716 crfree(entry->cred);
2719 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0 &&
2721 KASSERT(entry->cred == NULL || object->cred == NULL ||
2722 (entry->eflags & MAP_ENTRY_NEEDS_COPY),
2723 ("OVERCOMMIT vm_map_entry_delete: both cred %p", entry));
2724 count = OFF_TO_IDX(size);
2725 offidxstart = OFF_TO_IDX(entry->offset);
2726 offidxend = offidxstart + count;
2727 VM_OBJECT_LOCK(object);
2728 if (object->ref_count != 1 &&
2729 ((object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING ||
2730 object == kernel_object || object == kmem_object)) {
2731 vm_object_collapse(object);
2734 * The option OBJPR_NOTMAPPED can be passed here
2735 * because vm_map_delete() already performed
2736 * pmap_remove() on the only mapping to this range
2739 vm_object_page_remove(object, offidxstart, offidxend,
2741 if (object->type == OBJT_SWAP)
2742 swap_pager_freespace(object, offidxstart, count);
2743 if (offidxend >= object->size &&
2744 offidxstart < object->size) {
2745 size1 = object->size;
2746 object->size = offidxstart;
2747 if (object->cred != NULL) {
2748 size1 -= object->size;
2749 KASSERT(object->charge >= ptoa(size1),
2750 ("vm_map_entry_delete: object->charge < 0"));
2751 swap_release_by_cred(ptoa(size1), object->cred);
2752 object->charge -= ptoa(size1);
2756 VM_OBJECT_UNLOCK(object);
2758 entry->object.vm_object = NULL;
2759 if (map->system_map)
2760 vm_map_entry_deallocate(entry, TRUE);
2762 entry->next = curthread->td_map_def_user;
2763 curthread->td_map_def_user = entry;
2768 * vm_map_delete: [ internal use only ]
2770 * Deallocates the given address range from the target
2774 vm_map_delete(vm_map_t map, vm_offset_t start, vm_offset_t end)
2776 vm_map_entry_t entry;
2777 vm_map_entry_t first_entry;
2779 VM_MAP_ASSERT_LOCKED(map);
2782 * Find the start of the region, and clip it
2784 if (!vm_map_lookup_entry(map, start, &first_entry))
2785 entry = first_entry->next;
2787 entry = first_entry;
2788 vm_map_clip_start(map, entry, start);
2792 * Step through all entries in this region
2794 while ((entry != &map->header) && (entry->start < end)) {
2795 vm_map_entry_t next;
2798 * Wait for wiring or unwiring of an entry to complete.
2799 * Also wait for any system wirings to disappear on
2802 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 ||
2803 (vm_map_pmap(map) != kernel_pmap &&
2804 vm_map_entry_system_wired_count(entry) != 0)) {
2805 unsigned int last_timestamp;
2806 vm_offset_t saved_start;
2807 vm_map_entry_t tmp_entry;
2809 saved_start = entry->start;
2810 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2811 last_timestamp = map->timestamp;
2812 (void) vm_map_unlock_and_wait(map, 0);
2814 if (last_timestamp + 1 != map->timestamp) {
2816 * Look again for the entry because the map was
2817 * modified while it was unlocked.
2818 * Specifically, the entry may have been
2819 * clipped, merged, or deleted.
2821 if (!vm_map_lookup_entry(map, saved_start,
2823 entry = tmp_entry->next;
2826 vm_map_clip_start(map, entry,
2832 vm_map_clip_end(map, entry, end);
2837 * Unwire before removing addresses from the pmap; otherwise,
2838 * unwiring will put the entries back in the pmap.
2840 if (entry->wired_count != 0) {
2841 vm_map_entry_unwire(map, entry);
2844 pmap_remove(map->pmap, entry->start, entry->end);
2847 * Delete the entry only after removing all pmap
2848 * entries pointing to its pages. (Otherwise, its
2849 * page frames may be reallocated, and any modify bits
2850 * will be set in the wrong object!)
2852 vm_map_entry_delete(map, entry);
2855 return (KERN_SUCCESS);
2861 * Remove the given address range from the target map.
2862 * This is the exported form of vm_map_delete.
2865 vm_map_remove(vm_map_t map, vm_offset_t start, vm_offset_t end)
2870 VM_MAP_RANGE_CHECK(map, start, end);
2871 result = vm_map_delete(map, start, end);
2877 * vm_map_check_protection:
2879 * Assert that the target map allows the specified privilege on the
2880 * entire address region given. The entire region must be allocated.
2882 * WARNING! This code does not and should not check whether the
2883 * contents of the region is accessible. For example a smaller file
2884 * might be mapped into a larger address space.
2886 * NOTE! This code is also called by munmap().
2888 * The map must be locked. A read lock is sufficient.
2891 vm_map_check_protection(vm_map_t map, vm_offset_t start, vm_offset_t end,
2892 vm_prot_t protection)
2894 vm_map_entry_t entry;
2895 vm_map_entry_t tmp_entry;
2897 if (!vm_map_lookup_entry(map, start, &tmp_entry))
2901 while (start < end) {
2902 if (entry == &map->header)
2907 if (start < entry->start)
2910 * Check protection associated with entry.
2912 if ((entry->protection & protection) != protection)
2914 /* go to next entry */
2916 entry = entry->next;
2922 * vm_map_copy_entry:
2924 * Copies the contents of the source entry to the destination
2925 * entry. The entries *must* be aligned properly.
2931 vm_map_entry_t src_entry,
2932 vm_map_entry_t dst_entry,
2933 vm_ooffset_t *fork_charge)
2935 vm_object_t src_object;
2936 vm_map_entry_t fake_entry;
2941 VM_MAP_ASSERT_LOCKED(dst_map);
2943 if ((dst_entry->eflags|src_entry->eflags) & MAP_ENTRY_IS_SUB_MAP)
2946 if (src_entry->wired_count == 0) {
2949 * If the source entry is marked needs_copy, it is already
2952 if ((src_entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) {
2953 pmap_protect(src_map->pmap,
2956 src_entry->protection & ~VM_PROT_WRITE);
2960 * Make a copy of the object.
2962 size = src_entry->end - src_entry->start;
2963 if ((src_object = src_entry->object.vm_object) != NULL) {
2964 VM_OBJECT_LOCK(src_object);
2965 charged = ENTRY_CHARGED(src_entry);
2966 if ((src_object->handle == NULL) &&
2967 (src_object->type == OBJT_DEFAULT ||
2968 src_object->type == OBJT_SWAP)) {
2969 vm_object_collapse(src_object);
2970 if ((src_object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING) {
2971 vm_object_split(src_entry);
2972 src_object = src_entry->object.vm_object;
2975 vm_object_reference_locked(src_object);
2976 vm_object_clear_flag(src_object, OBJ_ONEMAPPING);
2977 if (src_entry->cred != NULL &&
2978 !(src_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
2979 KASSERT(src_object->cred == NULL,
2980 ("OVERCOMMIT: vm_map_copy_entry: cred %p",
2982 src_object->cred = src_entry->cred;
2983 src_object->charge = size;
2985 VM_OBJECT_UNLOCK(src_object);
2986 dst_entry->object.vm_object = src_object;
2988 cred = curthread->td_ucred;
2990 dst_entry->cred = cred;
2991 *fork_charge += size;
2992 if (!(src_entry->eflags &
2993 MAP_ENTRY_NEEDS_COPY)) {
2995 src_entry->cred = cred;
2996 *fork_charge += size;
2999 src_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY);
3000 dst_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY);
3001 dst_entry->offset = src_entry->offset;
3002 if (src_entry->eflags & MAP_ENTRY_VN_WRITECNT) {
3004 * MAP_ENTRY_VN_WRITECNT cannot
3005 * indicate write reference from
3006 * src_entry, since the entry is
3007 * marked as needs copy. Allocate a
3008 * fake entry that is used to
3009 * decrement object->un_pager.vnp.writecount
3010 * at the appropriate time. Attach
3011 * fake_entry to the deferred list.
3013 fake_entry = vm_map_entry_create(dst_map);
3014 fake_entry->eflags = MAP_ENTRY_VN_WRITECNT;
3015 src_entry->eflags &= ~MAP_ENTRY_VN_WRITECNT;
3016 vm_object_reference(src_object);
3017 fake_entry->object.vm_object = src_object;
3018 fake_entry->start = src_entry->start;
3019 fake_entry->end = src_entry->end;
3020 fake_entry->next = curthread->td_map_def_user;
3021 curthread->td_map_def_user = fake_entry;
3024 dst_entry->object.vm_object = NULL;
3025 dst_entry->offset = 0;
3026 if (src_entry->cred != NULL) {
3027 dst_entry->cred = curthread->td_ucred;
3028 crhold(dst_entry->cred);
3029 *fork_charge += size;
3033 pmap_copy(dst_map->pmap, src_map->pmap, dst_entry->start,
3034 dst_entry->end - dst_entry->start, src_entry->start);
3037 * Of course, wired down pages can't be set copy-on-write.
3038 * Cause wired pages to be copied into the new map by
3039 * simulating faults (the new pages are pageable)
3041 vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry,
3047 * vmspace_map_entry_forked:
3048 * Update the newly-forked vmspace each time a map entry is inherited
3049 * or copied. The values for vm_dsize and vm_tsize are approximate
3050 * (and mostly-obsolete ideas in the face of mmap(2) et al.)
3053 vmspace_map_entry_forked(const struct vmspace *vm1, struct vmspace *vm2,
3054 vm_map_entry_t entry)
3056 vm_size_t entrysize;
3059 entrysize = entry->end - entry->start;
3060 vm2->vm_map.size += entrysize;
3061 if (entry->eflags & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP)) {
3062 vm2->vm_ssize += btoc(entrysize);
3063 } else if (entry->start >= (vm_offset_t)vm1->vm_daddr &&
3064 entry->start < (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)) {
3065 newend = MIN(entry->end,
3066 (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize));
3067 vm2->vm_dsize += btoc(newend - entry->start);
3068 } else if (entry->start >= (vm_offset_t)vm1->vm_taddr &&
3069 entry->start < (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)) {
3070 newend = MIN(entry->end,
3071 (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize));
3072 vm2->vm_tsize += btoc(newend - entry->start);
3078 * Create a new process vmspace structure and vm_map
3079 * based on those of an existing process. The new map
3080 * is based on the old map, according to the inheritance
3081 * values on the regions in that map.
3083 * XXX It might be worth coalescing the entries added to the new vmspace.
3085 * The source map must not be locked.
3088 vmspace_fork(struct vmspace *vm1, vm_ooffset_t *fork_charge)
3090 struct vmspace *vm2;
3091 vm_map_t new_map, old_map;
3092 vm_map_entry_t new_entry, old_entry;
3096 old_map = &vm1->vm_map;
3097 /* Copy immutable fields of vm1 to vm2. */
3098 vm2 = vmspace_alloc(old_map->min_offset, old_map->max_offset);
3101 vm2->vm_taddr = vm1->vm_taddr;
3102 vm2->vm_daddr = vm1->vm_daddr;
3103 vm2->vm_maxsaddr = vm1->vm_maxsaddr;
3104 vm_map_lock(old_map);
3106 vm_map_wait_busy(old_map);
3107 new_map = &vm2->vm_map;
3108 locked = vm_map_trylock(new_map); /* trylock to silence WITNESS */
3109 KASSERT(locked, ("vmspace_fork: lock failed"));
3111 old_entry = old_map->header.next;
3113 while (old_entry != &old_map->header) {
3114 if (old_entry->eflags & MAP_ENTRY_IS_SUB_MAP)
3115 panic("vm_map_fork: encountered a submap");
3117 switch (old_entry->inheritance) {
3118 case VM_INHERIT_NONE:
3121 case VM_INHERIT_SHARE:
3123 * Clone the entry, creating the shared object if necessary.
3125 object = old_entry->object.vm_object;
3126 if (object == NULL) {
3127 object = vm_object_allocate(OBJT_DEFAULT,
3128 atop(old_entry->end - old_entry->start));
3129 old_entry->object.vm_object = object;
3130 old_entry->offset = 0;
3131 if (old_entry->cred != NULL) {
3132 object->cred = old_entry->cred;
3133 object->charge = old_entry->end -
3135 old_entry->cred = NULL;
3140 * Add the reference before calling vm_object_shadow
3141 * to insure that a shadow object is created.
3143 vm_object_reference(object);
3144 if (old_entry->eflags & MAP_ENTRY_NEEDS_COPY) {
3145 vm_object_shadow(&old_entry->object.vm_object,
3147 old_entry->end - old_entry->start);
3148 old_entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
3149 /* Transfer the second reference too. */
3150 vm_object_reference(
3151 old_entry->object.vm_object);
3154 * As in vm_map_simplify_entry(), the
3155 * vnode lock will not be acquired in
3156 * this call to vm_object_deallocate().
3158 vm_object_deallocate(object);
3159 object = old_entry->object.vm_object;
3161 VM_OBJECT_LOCK(object);
3162 vm_object_clear_flag(object, OBJ_ONEMAPPING);
3163 if (old_entry->cred != NULL) {
3164 KASSERT(object->cred == NULL, ("vmspace_fork both cred"));
3165 object->cred = old_entry->cred;
3166 object->charge = old_entry->end - old_entry->start;
3167 old_entry->cred = NULL;
3169 VM_OBJECT_UNLOCK(object);
3172 * Clone the entry, referencing the shared object.
3174 new_entry = vm_map_entry_create(new_map);
3175 *new_entry = *old_entry;
3176 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED |
3177 MAP_ENTRY_IN_TRANSITION);
3178 new_entry->wired_count = 0;
3179 if (new_entry->eflags & MAP_ENTRY_VN_WRITECNT) {
3180 object = new_entry->object.vm_object;
3181 KASSERT(((struct vnode *)object->handle)->
3183 ("vmspace_fork: v_writecount"));
3184 KASSERT(object->un_pager.vnp.writemappings > 0,
3185 ("vmspace_fork: vnp.writecount"));
3186 vnode_pager_update_writecount(object,
3187 new_entry->start, new_entry->end);
3191 * Insert the entry into the new map -- we know we're
3192 * inserting at the end of the new map.
3194 vm_map_entry_link(new_map, new_map->header.prev,
3196 vmspace_map_entry_forked(vm1, vm2, new_entry);
3199 * Update the physical map
3201 pmap_copy(new_map->pmap, old_map->pmap,
3203 (old_entry->end - old_entry->start),
3207 case VM_INHERIT_COPY:
3209 * Clone the entry and link into the map.
3211 new_entry = vm_map_entry_create(new_map);
3212 *new_entry = *old_entry;
3214 * Copied entry is COW over the old object.
3216 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED |
3217 MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_VN_WRITECNT);
3218 new_entry->wired_count = 0;
3219 new_entry->object.vm_object = NULL;
3220 new_entry->cred = NULL;
3221 vm_map_entry_link(new_map, new_map->header.prev,
3223 vmspace_map_entry_forked(vm1, vm2, new_entry);
3224 vm_map_copy_entry(old_map, new_map, old_entry,
3225 new_entry, fork_charge);
3228 old_entry = old_entry->next;
3231 * Use inlined vm_map_unlock() to postpone handling the deferred
3232 * map entries, which cannot be done until both old_map and
3233 * new_map locks are released.
3235 sx_xunlock(&old_map->lock);
3236 sx_xunlock(&new_map->lock);
3237 vm_map_process_deferred();
3243 vm_map_stack(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
3244 vm_prot_t prot, vm_prot_t max, int cow)
3246 vm_map_entry_t new_entry, prev_entry;
3247 vm_offset_t bot, top;
3248 vm_size_t growsize, init_ssize;
3253 * The stack orientation is piggybacked with the cow argument.
3254 * Extract it into orient and mask the cow argument so that we
3255 * don't pass it around further.
3256 * NOTE: We explicitly allow bi-directional stacks.
3258 orient = cow & (MAP_STACK_GROWS_DOWN|MAP_STACK_GROWS_UP);
3260 KASSERT(orient != 0, ("No stack grow direction"));
3262 if (addrbos < vm_map_min(map) ||
3263 addrbos > vm_map_max(map) ||
3264 addrbos + max_ssize < addrbos)
3265 return (KERN_NO_SPACE);
3267 growsize = sgrowsiz;
3268 init_ssize = (max_ssize < growsize) ? max_ssize : growsize;
3270 PROC_LOCK(curthread->td_proc);
3271 vmemlim = lim_cur(curthread->td_proc, RLIMIT_VMEM);
3272 PROC_UNLOCK(curthread->td_proc);
3276 /* If addr is already mapped, no go */
3277 if (vm_map_lookup_entry(map, addrbos, &prev_entry)) {
3279 return (KERN_NO_SPACE);
3282 /* If we would blow our VMEM resource limit, no go */
3283 if (map->size + init_ssize > vmemlim) {
3285 return (KERN_NO_SPACE);
3289 * If we can't accomodate max_ssize in the current mapping, no go.
3290 * However, we need to be aware that subsequent user mappings might
3291 * map into the space we have reserved for stack, and currently this
3292 * space is not protected.
3294 * Hopefully we will at least detect this condition when we try to
3297 if ((prev_entry->next != &map->header) &&
3298 (prev_entry->next->start < addrbos + max_ssize)) {
3300 return (KERN_NO_SPACE);
3304 * We initially map a stack of only init_ssize. We will grow as
3305 * needed later. Depending on the orientation of the stack (i.e.
3306 * the grow direction) we either map at the top of the range, the
3307 * bottom of the range or in the middle.
3309 * Note: we would normally expect prot and max to be VM_PROT_ALL,
3310 * and cow to be 0. Possibly we should eliminate these as input
3311 * parameters, and just pass these values here in the insert call.
3313 if (orient == MAP_STACK_GROWS_DOWN)
3314 bot = addrbos + max_ssize - init_ssize;
3315 else if (orient == MAP_STACK_GROWS_UP)
3318 bot = round_page(addrbos + max_ssize/2 - init_ssize/2);
3319 top = bot + init_ssize;
3320 rv = vm_map_insert(map, NULL, 0, bot, top, prot, max, cow);
3322 /* Now set the avail_ssize amount. */
3323 if (rv == KERN_SUCCESS) {
3324 if (prev_entry != &map->header)
3325 vm_map_clip_end(map, prev_entry, bot);
3326 new_entry = prev_entry->next;
3327 if (new_entry->end != top || new_entry->start != bot)
3328 panic("Bad entry start/end for new stack entry");
3330 new_entry->avail_ssize = max_ssize - init_ssize;
3331 if (orient & MAP_STACK_GROWS_DOWN)
3332 new_entry->eflags |= MAP_ENTRY_GROWS_DOWN;
3333 if (orient & MAP_STACK_GROWS_UP)
3334 new_entry->eflags |= MAP_ENTRY_GROWS_UP;
3341 static int stack_guard_page = 0;
3342 TUNABLE_INT("security.bsd.stack_guard_page", &stack_guard_page);
3343 SYSCTL_INT(_security_bsd, OID_AUTO, stack_guard_page, CTLFLAG_RW,
3344 &stack_guard_page, 0,
3345 "Insert stack guard page ahead of the growable segments.");
3347 /* Attempts to grow a vm stack entry. Returns KERN_SUCCESS if the
3348 * desired address is already mapped, or if we successfully grow
3349 * the stack. Also returns KERN_SUCCESS if addr is outside the
3350 * stack range (this is strange, but preserves compatibility with
3351 * the grow function in vm_machdep.c).
3354 vm_map_growstack(struct proc *p, vm_offset_t addr)
3356 vm_map_entry_t next_entry, prev_entry;
3357 vm_map_entry_t new_entry, stack_entry;
3358 struct vmspace *vm = p->p_vmspace;
3359 vm_map_t map = &vm->vm_map;
3362 size_t grow_amount, max_grow;
3363 rlim_t stacklim, vmemlim;
3364 int is_procstack, rv;
3375 stacklim = lim_cur(p, RLIMIT_STACK);
3376 vmemlim = lim_cur(p, RLIMIT_VMEM);
3379 vm_map_lock_read(map);
3381 /* If addr is already in the entry range, no need to grow.*/
3382 if (vm_map_lookup_entry(map, addr, &prev_entry)) {
3383 vm_map_unlock_read(map);
3384 return (KERN_SUCCESS);
3387 next_entry = prev_entry->next;
3388 if (!(prev_entry->eflags & MAP_ENTRY_GROWS_UP)) {
3390 * This entry does not grow upwards. Since the address lies
3391 * beyond this entry, the next entry (if one exists) has to
3392 * be a downward growable entry. The entry list header is
3393 * never a growable entry, so it suffices to check the flags.
3395 if (!(next_entry->eflags & MAP_ENTRY_GROWS_DOWN)) {
3396 vm_map_unlock_read(map);
3397 return (KERN_SUCCESS);
3399 stack_entry = next_entry;
3402 * This entry grows upward. If the next entry does not at
3403 * least grow downwards, this is the entry we need to grow.
3404 * otherwise we have two possible choices and we have to
3407 if (next_entry->eflags & MAP_ENTRY_GROWS_DOWN) {
3409 * We have two choices; grow the entry closest to
3410 * the address to minimize the amount of growth.
3412 if (addr - prev_entry->end <= next_entry->start - addr)
3413 stack_entry = prev_entry;
3415 stack_entry = next_entry;
3417 stack_entry = prev_entry;
3420 if (stack_entry == next_entry) {
3421 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_DOWN, ("foo"));
3422 KASSERT(addr < stack_entry->start, ("foo"));
3423 end = (prev_entry != &map->header) ? prev_entry->end :
3424 stack_entry->start - stack_entry->avail_ssize;
3425 grow_amount = roundup(stack_entry->start - addr, PAGE_SIZE);
3426 max_grow = stack_entry->start - end;
3428 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_UP, ("foo"));
3429 KASSERT(addr >= stack_entry->end, ("foo"));
3430 end = (next_entry != &map->header) ? next_entry->start :
3431 stack_entry->end + stack_entry->avail_ssize;
3432 grow_amount = roundup(addr + 1 - stack_entry->end, PAGE_SIZE);
3433 max_grow = end - stack_entry->end;
3436 if (grow_amount > stack_entry->avail_ssize) {
3437 vm_map_unlock_read(map);
3438 return (KERN_NO_SPACE);
3442 * If there is no longer enough space between the entries nogo, and
3443 * adjust the available space. Note: this should only happen if the
3444 * user has mapped into the stack area after the stack was created,
3445 * and is probably an error.
3447 * This also effectively destroys any guard page the user might have
3448 * intended by limiting the stack size.
3450 if (grow_amount + (stack_guard_page ? PAGE_SIZE : 0) > max_grow) {
3451 if (vm_map_lock_upgrade(map))
3454 stack_entry->avail_ssize = max_grow;
3457 return (KERN_NO_SPACE);
3460 is_procstack = (addr >= (vm_offset_t)vm->vm_maxsaddr) ? 1 : 0;
3463 * If this is the main process stack, see if we're over the stack
3466 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) {
3467 vm_map_unlock_read(map);
3468 return (KERN_NO_SPACE);
3473 racct_set(p, RACCT_STACK, ctob(vm->vm_ssize) + grow_amount)) {
3475 vm_map_unlock_read(map);
3476 return (KERN_NO_SPACE);
3481 /* Round up the grow amount modulo sgrowsiz */
3482 growsize = sgrowsiz;
3483 grow_amount = roundup(grow_amount, growsize);
3484 if (grow_amount > stack_entry->avail_ssize)
3485 grow_amount = stack_entry->avail_ssize;
3486 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) {
3487 grow_amount = trunc_page((vm_size_t)stacklim) -
3492 limit = racct_get_available(p, RACCT_STACK);
3494 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > limit))
3495 grow_amount = limit - ctob(vm->vm_ssize);
3498 /* If we would blow our VMEM resource limit, no go */
3499 if (map->size + grow_amount > vmemlim) {
3500 vm_map_unlock_read(map);
3506 if (racct_set(p, RACCT_VMEM, map->size + grow_amount)) {
3508 vm_map_unlock_read(map);
3515 if (vm_map_lock_upgrade(map))
3518 if (stack_entry == next_entry) {
3522 /* Get the preliminary new entry start value */
3523 addr = stack_entry->start - grow_amount;
3526 * If this puts us into the previous entry, cut back our
3527 * growth to the available space. Also, see the note above.
3530 stack_entry->avail_ssize = max_grow;
3532 if (stack_guard_page)
3536 rv = vm_map_insert(map, NULL, 0, addr, stack_entry->start,
3537 next_entry->protection, next_entry->max_protection, 0);
3539 /* Adjust the available stack space by the amount we grew. */
3540 if (rv == KERN_SUCCESS) {
3541 if (prev_entry != &map->header)
3542 vm_map_clip_end(map, prev_entry, addr);
3543 new_entry = prev_entry->next;
3544 KASSERT(new_entry == stack_entry->prev, ("foo"));
3545 KASSERT(new_entry->end == stack_entry->start, ("foo"));
3546 KASSERT(new_entry->start == addr, ("foo"));
3547 grow_amount = new_entry->end - new_entry->start;
3548 new_entry->avail_ssize = stack_entry->avail_ssize -
3550 stack_entry->eflags &= ~MAP_ENTRY_GROWS_DOWN;
3551 new_entry->eflags |= MAP_ENTRY_GROWS_DOWN;
3557 addr = stack_entry->end + grow_amount;
3560 * If this puts us into the next entry, cut back our growth
3561 * to the available space. Also, see the note above.
3564 stack_entry->avail_ssize = end - stack_entry->end;
3566 if (stack_guard_page)
3570 grow_amount = addr - stack_entry->end;
3571 cred = stack_entry->cred;
3572 if (cred == NULL && stack_entry->object.vm_object != NULL)
3573 cred = stack_entry->object.vm_object->cred;
3574 if (cred != NULL && !swap_reserve_by_cred(grow_amount, cred))
3576 /* Grow the underlying object if applicable. */
3577 else if (stack_entry->object.vm_object == NULL ||
3578 vm_object_coalesce(stack_entry->object.vm_object,
3579 stack_entry->offset,
3580 (vm_size_t)(stack_entry->end - stack_entry->start),
3581 (vm_size_t)grow_amount, cred != NULL)) {
3582 map->size += (addr - stack_entry->end);
3583 /* Update the current entry. */
3584 stack_entry->end = addr;
3585 stack_entry->avail_ssize -= grow_amount;
3586 vm_map_entry_resize_free(map, stack_entry);
3589 if (next_entry != &map->header)
3590 vm_map_clip_start(map, next_entry, addr);
3595 if (rv == KERN_SUCCESS && is_procstack)
3596 vm->vm_ssize += btoc(grow_amount);
3601 * Heed the MAP_WIREFUTURE flag if it was set for this process.
3603 if (rv == KERN_SUCCESS && (map->flags & MAP_WIREFUTURE)) {
3605 (stack_entry == next_entry) ? addr : addr - grow_amount,
3606 (stack_entry == next_entry) ? stack_entry->start : addr,
3607 (p->p_flag & P_SYSTEM)
3608 ? VM_MAP_WIRE_SYSTEM|VM_MAP_WIRE_NOHOLES
3609 : VM_MAP_WIRE_USER|VM_MAP_WIRE_NOHOLES);
3614 if (rv != KERN_SUCCESS) {
3616 error = racct_set(p, RACCT_VMEM, map->size);
3617 KASSERT(error == 0, ("decreasing RACCT_VMEM failed"));
3618 error = racct_set(p, RACCT_STACK, ctob(vm->vm_ssize));
3619 KASSERT(error == 0, ("decreasing RACCT_STACK failed"));
3628 * Unshare the specified VM space for exec. If other processes are
3629 * mapped to it, then create a new one. The new vmspace is null.
3632 vmspace_exec(struct proc *p, vm_offset_t minuser, vm_offset_t maxuser)
3634 struct vmspace *oldvmspace = p->p_vmspace;
3635 struct vmspace *newvmspace;
3637 newvmspace = vmspace_alloc(minuser, maxuser);
3638 if (newvmspace == NULL)
3640 newvmspace->vm_swrss = oldvmspace->vm_swrss;
3642 * This code is written like this for prototype purposes. The
3643 * goal is to avoid running down the vmspace here, but let the
3644 * other process's that are still using the vmspace to finally
3645 * run it down. Even though there is little or no chance of blocking
3646 * here, it is a good idea to keep this form for future mods.
3648 PROC_VMSPACE_LOCK(p);
3649 p->p_vmspace = newvmspace;
3650 PROC_VMSPACE_UNLOCK(p);
3651 if (p == curthread->td_proc)
3652 pmap_activate(curthread);
3653 vmspace_free(oldvmspace);
3658 * Unshare the specified VM space for forcing COW. This
3659 * is called by rfork, for the (RFMEM|RFPROC) == 0 case.
3662 vmspace_unshare(struct proc *p)
3664 struct vmspace *oldvmspace = p->p_vmspace;
3665 struct vmspace *newvmspace;
3666 vm_ooffset_t fork_charge;
3668 if (oldvmspace->vm_refcnt == 1)
3671 newvmspace = vmspace_fork(oldvmspace, &fork_charge);
3672 if (newvmspace == NULL)
3674 if (!swap_reserve_by_cred(fork_charge, p->p_ucred)) {
3675 vmspace_free(newvmspace);
3678 PROC_VMSPACE_LOCK(p);
3679 p->p_vmspace = newvmspace;
3680 PROC_VMSPACE_UNLOCK(p);
3681 if (p == curthread->td_proc)
3682 pmap_activate(curthread);
3683 vmspace_free(oldvmspace);
3690 * Finds the VM object, offset, and
3691 * protection for a given virtual address in the
3692 * specified map, assuming a page fault of the
3695 * Leaves the map in question locked for read; return
3696 * values are guaranteed until a vm_map_lookup_done
3697 * call is performed. Note that the map argument
3698 * is in/out; the returned map must be used in
3699 * the call to vm_map_lookup_done.
3701 * A handle (out_entry) is returned for use in
3702 * vm_map_lookup_done, to make that fast.
3704 * If a lookup is requested with "write protection"
3705 * specified, the map may be changed to perform virtual
3706 * copying operations, although the data referenced will
3710 vm_map_lookup(vm_map_t *var_map, /* IN/OUT */
3712 vm_prot_t fault_typea,
3713 vm_map_entry_t *out_entry, /* OUT */
3714 vm_object_t *object, /* OUT */
3715 vm_pindex_t *pindex, /* OUT */
3716 vm_prot_t *out_prot, /* OUT */
3717 boolean_t *wired) /* OUT */
3719 vm_map_entry_t entry;
3720 vm_map_t map = *var_map;
3722 vm_prot_t fault_type = fault_typea;
3723 vm_object_t eobject;
3729 vm_map_lock_read(map);
3732 * Lookup the faulting address.
3734 if (!vm_map_lookup_entry(map, vaddr, out_entry)) {
3735 vm_map_unlock_read(map);
3736 return (KERN_INVALID_ADDRESS);
3744 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
3745 vm_map_t old_map = map;
3747 *var_map = map = entry->object.sub_map;
3748 vm_map_unlock_read(old_map);
3753 * Check whether this task is allowed to have this page.
3755 prot = entry->protection;
3756 fault_type &= (VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE);
3757 if ((fault_type & prot) != fault_type || prot == VM_PROT_NONE) {
3758 vm_map_unlock_read(map);
3759 return (KERN_PROTECTION_FAILURE);
3761 if ((entry->eflags & MAP_ENTRY_USER_WIRED) &&
3762 (entry->eflags & MAP_ENTRY_COW) &&
3763 (fault_type & VM_PROT_WRITE)) {
3764 vm_map_unlock_read(map);
3765 return (KERN_PROTECTION_FAILURE);
3769 * If this page is not pageable, we have to get it for all possible
3772 *wired = (entry->wired_count != 0);
3774 fault_type = entry->protection;
3775 size = entry->end - entry->start;
3777 * If the entry was copy-on-write, we either ...
3779 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
3781 * If we want to write the page, we may as well handle that
3782 * now since we've got the map locked.
3784 * If we don't need to write the page, we just demote the
3785 * permissions allowed.
3787 if ((fault_type & VM_PROT_WRITE) != 0 ||
3788 (fault_typea & VM_PROT_COPY) != 0) {
3790 * Make a new object, and place it in the object
3791 * chain. Note that no new references have appeared
3792 * -- one just moved from the map to the new
3795 if (vm_map_lock_upgrade(map))
3798 if (entry->cred == NULL) {
3800 * The debugger owner is charged for
3803 cred = curthread->td_ucred;
3805 if (!swap_reserve_by_cred(size, cred)) {
3808 return (KERN_RESOURCE_SHORTAGE);
3812 vm_object_shadow(&entry->object.vm_object,
3813 &entry->offset, size);
3814 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
3815 eobject = entry->object.vm_object;
3816 if (eobject->cred != NULL) {
3818 * The object was not shadowed.
3820 swap_release_by_cred(size, entry->cred);
3821 crfree(entry->cred);
3823 } else if (entry->cred != NULL) {
3824 VM_OBJECT_LOCK(eobject);
3825 eobject->cred = entry->cred;
3826 eobject->charge = size;
3827 VM_OBJECT_UNLOCK(eobject);
3831 vm_map_lock_downgrade(map);
3834 * We're attempting to read a copy-on-write page --
3835 * don't allow writes.
3837 prot &= ~VM_PROT_WRITE;
3842 * Create an object if necessary.
3844 if (entry->object.vm_object == NULL &&
3846 if (vm_map_lock_upgrade(map))
3848 entry->object.vm_object = vm_object_allocate(OBJT_DEFAULT,
3851 if (entry->cred != NULL) {
3852 VM_OBJECT_LOCK(entry->object.vm_object);
3853 entry->object.vm_object->cred = entry->cred;
3854 entry->object.vm_object->charge = size;
3855 VM_OBJECT_UNLOCK(entry->object.vm_object);
3858 vm_map_lock_downgrade(map);
3862 * Return the object/offset from this entry. If the entry was
3863 * copy-on-write or empty, it has been fixed up.
3865 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
3866 *object = entry->object.vm_object;
3869 return (KERN_SUCCESS);
3873 * vm_map_lookup_locked:
3875 * Lookup the faulting address. A version of vm_map_lookup that returns
3876 * KERN_FAILURE instead of blocking on map lock or memory allocation.
3879 vm_map_lookup_locked(vm_map_t *var_map, /* IN/OUT */
3881 vm_prot_t fault_typea,
3882 vm_map_entry_t *out_entry, /* OUT */
3883 vm_object_t *object, /* OUT */
3884 vm_pindex_t *pindex, /* OUT */
3885 vm_prot_t *out_prot, /* OUT */
3886 boolean_t *wired) /* OUT */
3888 vm_map_entry_t entry;
3889 vm_map_t map = *var_map;
3891 vm_prot_t fault_type = fault_typea;
3894 * Lookup the faulting address.
3896 if (!vm_map_lookup_entry(map, vaddr, out_entry))
3897 return (KERN_INVALID_ADDRESS);
3902 * Fail if the entry refers to a submap.
3904 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP)
3905 return (KERN_FAILURE);
3908 * Check whether this task is allowed to have this page.
3910 prot = entry->protection;
3911 fault_type &= VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE;
3912 if ((fault_type & prot) != fault_type)
3913 return (KERN_PROTECTION_FAILURE);
3914 if ((entry->eflags & MAP_ENTRY_USER_WIRED) &&
3915 (entry->eflags & MAP_ENTRY_COW) &&
3916 (fault_type & VM_PROT_WRITE))
3917 return (KERN_PROTECTION_FAILURE);
3920 * If this page is not pageable, we have to get it for all possible
3923 *wired = (entry->wired_count != 0);
3925 fault_type = entry->protection;
3927 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
3929 * Fail if the entry was copy-on-write for a write fault.
3931 if (fault_type & VM_PROT_WRITE)
3932 return (KERN_FAILURE);
3934 * We're attempting to read a copy-on-write page --
3935 * don't allow writes.
3937 prot &= ~VM_PROT_WRITE;
3941 * Fail if an object should be created.
3943 if (entry->object.vm_object == NULL && !map->system_map)
3944 return (KERN_FAILURE);
3947 * Return the object/offset from this entry. If the entry was
3948 * copy-on-write or empty, it has been fixed up.
3950 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
3951 *object = entry->object.vm_object;
3954 return (KERN_SUCCESS);
3958 * vm_map_lookup_done:
3960 * Releases locks acquired by a vm_map_lookup
3961 * (according to the handle returned by that lookup).
3964 vm_map_lookup_done(vm_map_t map, vm_map_entry_t entry)
3967 * Unlock the main-level map
3969 vm_map_unlock_read(map);
3972 #include "opt_ddb.h"
3974 #include <sys/kernel.h>
3976 #include <ddb/ddb.h>
3979 vm_map_print(vm_map_t map)
3981 vm_map_entry_t entry;
3983 db_iprintf("Task map %p: pmap=%p, nentries=%d, version=%u\n",
3985 (void *)map->pmap, map->nentries, map->timestamp);
3988 for (entry = map->header.next; entry != &map->header;
3989 entry = entry->next) {
3990 db_iprintf("map entry %p: start=%p, end=%p\n",
3991 (void *)entry, (void *)entry->start, (void *)entry->end);
3993 static char *inheritance_name[4] =
3994 {"share", "copy", "none", "donate_copy"};
3996 db_iprintf(" prot=%x/%x/%s",
3998 entry->max_protection,
3999 inheritance_name[(int)(unsigned char)entry->inheritance]);
4000 if (entry->wired_count != 0)
4001 db_printf(", wired");
4003 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
4004 db_printf(", share=%p, offset=0x%jx\n",
4005 (void *)entry->object.sub_map,
4006 (uintmax_t)entry->offset);
4007 if ((entry->prev == &map->header) ||
4008 (entry->prev->object.sub_map !=
4009 entry->object.sub_map)) {
4011 vm_map_print((vm_map_t)entry->object.sub_map);
4015 if (entry->cred != NULL)
4016 db_printf(", ruid %d", entry->cred->cr_ruid);
4017 db_printf(", object=%p, offset=0x%jx",
4018 (void *)entry->object.vm_object,
4019 (uintmax_t)entry->offset);
4020 if (entry->object.vm_object && entry->object.vm_object->cred)
4021 db_printf(", obj ruid %d charge %jx",
4022 entry->object.vm_object->cred->cr_ruid,
4023 (uintmax_t)entry->object.vm_object->charge);
4024 if (entry->eflags & MAP_ENTRY_COW)
4025 db_printf(", copy (%s)",
4026 (entry->eflags & MAP_ENTRY_NEEDS_COPY) ? "needed" : "done");
4029 if ((entry->prev == &map->header) ||
4030 (entry->prev->object.vm_object !=
4031 entry->object.vm_object)) {
4033 vm_object_print((db_expr_t)(intptr_t)
4034 entry->object.vm_object,
4043 DB_SHOW_COMMAND(map, map)
4047 db_printf("usage: show map <addr>\n");
4050 vm_map_print((vm_map_t)addr);
4053 DB_SHOW_COMMAND(procvm, procvm)
4058 p = (struct proc *) addr;
4063 db_printf("p = %p, vmspace = %p, map = %p, pmap = %p\n",
4064 (void *)p, (void *)p->p_vmspace, (void *)&p->p_vmspace->vm_map,
4065 (void *)vmspace_pmap(p->p_vmspace));
4067 vm_map_print((vm_map_t)&p->p_vmspace->vm_map);