2 * Copyright (c) 1991, 1993
3 * The Regents of the University of California. All rights reserved.
5 * This code is derived from software contributed to Berkeley by
6 * The Mach Operating System project at Carnegie-Mellon University.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 4. Neither the name of the University nor the names of its contributors
17 * may be used to endorse or promote products derived from this software
18 * without specific prior written permission.
20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * from: @(#)vm_map.c 8.3 (Berkeley) 1/12/94
35 * Copyright (c) 1987, 1990 Carnegie-Mellon University.
36 * All rights reserved.
38 * Authors: Avadis Tevanian, Jr., Michael Wayne Young
40 * Permission to use, copy, modify and distribute this software and
41 * its documentation is hereby granted, provided that both the copyright
42 * notice and this permission notice appear in all copies of the
43 * software, derivative works or modified versions, and any portions
44 * thereof, and that both notices appear in supporting documentation.
46 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
47 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
48 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
50 * Carnegie Mellon requests users of this software to return to
52 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
53 * School of Computer Science
54 * Carnegie Mellon University
55 * Pittsburgh PA 15213-3890
57 * any improvements or extensions that they make and grant Carnegie the
58 * rights to redistribute these changes.
62 * Virtual memory mapping module.
65 #include <sys/cdefs.h>
66 __FBSDID("$FreeBSD$");
68 #include <sys/param.h>
69 #include <sys/systm.h>
70 #include <sys/kernel.h>
73 #include <sys/mutex.h>
75 #include <sys/vmmeter.h>
77 #include <sys/vnode.h>
78 #include <sys/racct.h>
79 #include <sys/resourcevar.h>
81 #include <sys/sysctl.h>
82 #include <sys/sysent.h>
86 #include <vm/vm_param.h>
88 #include <vm/vm_map.h>
89 #include <vm/vm_page.h>
90 #include <vm/vm_object.h>
91 #include <vm/vm_pager.h>
92 #include <vm/vm_kern.h>
93 #include <vm/vm_extern.h>
94 #include <vm/vnode_pager.h>
95 #include <vm/swap_pager.h>
99 * Virtual memory maps provide for the mapping, protection,
100 * and sharing of virtual memory objects. In addition,
101 * this module provides for an efficient virtual copy of
102 * memory from one map to another.
104 * Synchronization is required prior to most operations.
106 * Maps consist of an ordered doubly-linked list of simple
107 * entries; a self-adjusting binary search tree of these
108 * entries is used to speed up lookups.
110 * Since portions of maps are specified by start/end addresses,
111 * which may not align with existing map entries, all
112 * routines merely "clip" entries to these start/end values.
113 * [That is, an entry is split into two, bordering at a
114 * start or end value.] Note that these clippings may not
115 * always be necessary (as the two resulting entries are then
116 * not changed); however, the clipping is done for convenience.
118 * As mentioned above, virtual copy operations are performed
119 * by copying VM object references from one map to
120 * another, and then marking both regions as copy-on-write.
123 static struct mtx map_sleep_mtx;
124 static uma_zone_t mapentzone;
125 static uma_zone_t kmapentzone;
126 static uma_zone_t mapzone;
127 static uma_zone_t vmspace_zone;
128 static struct vm_object kmapentobj;
129 static int vmspace_zinit(void *mem, int size, int flags);
130 static void vmspace_zfini(void *mem, int size);
131 static int vm_map_zinit(void *mem, int ize, int flags);
132 static void vm_map_zfini(void *mem, int size);
133 static void _vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min,
135 static void vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map);
136 static void vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry);
138 static void vm_map_zdtor(void *mem, int size, void *arg);
139 static void vmspace_zdtor(void *mem, int size, void *arg);
142 #define ENTRY_CHARGED(e) ((e)->cred != NULL || \
143 ((e)->object.vm_object != NULL && (e)->object.vm_object->cred != NULL && \
144 !((e)->eflags & MAP_ENTRY_NEEDS_COPY)))
147 * PROC_VMSPACE_{UN,}LOCK() can be a noop as long as vmspaces are type
150 #define PROC_VMSPACE_LOCK(p) do { } while (0)
151 #define PROC_VMSPACE_UNLOCK(p) do { } while (0)
154 * VM_MAP_RANGE_CHECK: [ internal use only ]
156 * Asserts that the starting and ending region
157 * addresses fall within the valid range of the map.
159 #define VM_MAP_RANGE_CHECK(map, start, end) \
161 if (start < vm_map_min(map)) \
162 start = vm_map_min(map); \
163 if (end > vm_map_max(map)) \
164 end = vm_map_max(map); \
172 * Initialize the vm_map module. Must be called before
173 * any other vm_map routines.
175 * Map and entry structures are allocated from the general
176 * purpose memory pool with some exceptions:
178 * - The kernel map and kmem submap are allocated statically.
179 * - Kernel map entries are allocated out of a static pool.
181 * These restrictions are necessary since malloc() uses the
182 * maps and requires map entries.
188 mtx_init(&map_sleep_mtx, "vm map sleep mutex", NULL, MTX_DEF);
189 mapzone = uma_zcreate("MAP", sizeof(struct vm_map), NULL,
195 vm_map_zinit, vm_map_zfini, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
196 uma_prealloc(mapzone, MAX_KMAP);
197 kmapentzone = uma_zcreate("KMAP ENTRY", sizeof(struct vm_map_entry),
198 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR,
199 UMA_ZONE_MTXCLASS | UMA_ZONE_VM);
200 uma_prealloc(kmapentzone, MAX_KMAPENT);
201 mapentzone = uma_zcreate("MAP ENTRY", sizeof(struct vm_map_entry),
202 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
206 vmspace_zfini(void *mem, int size)
210 vm = (struct vmspace *)mem;
211 vm_map_zfini(&vm->vm_map, sizeof(vm->vm_map));
215 vmspace_zinit(void *mem, int size, int flags)
219 vm = (struct vmspace *)mem;
221 vm->vm_map.pmap = NULL;
222 (void)vm_map_zinit(&vm->vm_map, sizeof(vm->vm_map), flags);
227 vm_map_zfini(void *mem, int size)
232 mtx_destroy(&map->system_mtx);
233 sx_destroy(&map->lock);
237 vm_map_zinit(void *mem, int size, int flags)
244 mtx_init(&map->system_mtx, "system map", NULL, MTX_DEF | MTX_DUPOK);
245 sx_init(&map->lock, "user map");
251 vmspace_zdtor(void *mem, int size, void *arg)
255 vm = (struct vmspace *)mem;
257 vm_map_zdtor(&vm->vm_map, sizeof(vm->vm_map), arg);
260 vm_map_zdtor(void *mem, int size, void *arg)
265 KASSERT(map->nentries == 0,
266 ("map %p nentries == %d on free.",
267 map, map->nentries));
268 KASSERT(map->size == 0,
269 ("map %p size == %lu on free.",
270 map, (unsigned long)map->size));
272 #endif /* INVARIANTS */
275 * Allocate a vmspace structure, including a vm_map and pmap,
276 * and initialize those structures. The refcnt is set to 1.
279 vmspace_alloc(min, max)
280 vm_offset_t min, max;
284 vm = uma_zalloc(vmspace_zone, M_WAITOK);
285 if (vm->vm_map.pmap == NULL && !pmap_pinit(vmspace_pmap(vm))) {
286 uma_zfree(vmspace_zone, vm);
289 CTR1(KTR_VM, "vmspace_alloc: %p", vm);
290 _vm_map_init(&vm->vm_map, vmspace_pmap(vm), min, max);
306 uma_zone_set_obj(kmapentzone, &kmapentobj, lmin(cnt.v_page_count,
307 (VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS) / PAGE_SIZE) / 8 +
308 maxproc * 2 + maxfiles);
309 vmspace_zone = uma_zcreate("VMSPACE", sizeof(struct vmspace), NULL,
315 vmspace_zinit, vmspace_zfini, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
319 vmspace_container_reset(struct proc *p)
324 racct_set(p, RACCT_DATA, 0);
325 racct_set(p, RACCT_STACK, 0);
326 racct_set(p, RACCT_RSS, 0);
327 racct_set(p, RACCT_MEMLOCK, 0);
328 racct_set(p, RACCT_VMEM, 0);
334 vmspace_dofree(struct vmspace *vm)
337 CTR1(KTR_VM, "vmspace_free: %p", vm);
340 * Make sure any SysV shm is freed, it might not have been in
346 * Lock the map, to wait out all other references to it.
347 * Delete all of the mappings and pages they hold, then call
348 * the pmap module to reclaim anything left.
350 (void)vm_map_remove(&vm->vm_map, vm->vm_map.min_offset,
351 vm->vm_map.max_offset);
353 pmap_release(vmspace_pmap(vm));
354 vm->vm_map.pmap = NULL;
355 uma_zfree(vmspace_zone, vm);
359 vmspace_free(struct vmspace *vm)
362 if (vm->vm_refcnt == 0)
363 panic("vmspace_free: attempt to free already freed vmspace");
365 if (atomic_fetchadd_int(&vm->vm_refcnt, -1) == 1)
370 vmspace_exitfree(struct proc *p)
374 PROC_VMSPACE_LOCK(p);
377 PROC_VMSPACE_UNLOCK(p);
378 KASSERT(vm == &vmspace0, ("vmspace_exitfree: wrong vmspace"));
383 vmspace_exit(struct thread *td)
390 * Release user portion of address space.
391 * This releases references to vnodes,
392 * which could cause I/O if the file has been unlinked.
393 * Need to do this early enough that we can still sleep.
395 * The last exiting process to reach this point releases as
396 * much of the environment as it can. vmspace_dofree() is the
397 * slower fallback in case another process had a temporary
398 * reference to the vmspace.
403 atomic_add_int(&vmspace0.vm_refcnt, 1);
405 refcnt = vm->vm_refcnt;
406 if (refcnt > 1 && p->p_vmspace != &vmspace0) {
407 /* Switch now since other proc might free vmspace */
408 PROC_VMSPACE_LOCK(p);
409 p->p_vmspace = &vmspace0;
410 PROC_VMSPACE_UNLOCK(p);
413 } while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt - 1));
415 if (p->p_vmspace != vm) {
416 /* vmspace not yet freed, switch back */
417 PROC_VMSPACE_LOCK(p);
419 PROC_VMSPACE_UNLOCK(p);
422 pmap_remove_pages(vmspace_pmap(vm));
423 /* Switch now since this proc will free vmspace */
424 PROC_VMSPACE_LOCK(p);
425 p->p_vmspace = &vmspace0;
426 PROC_VMSPACE_UNLOCK(p);
430 vmspace_container_reset(p);
433 /* Acquire reference to vmspace owned by another process. */
436 vmspace_acquire_ref(struct proc *p)
441 PROC_VMSPACE_LOCK(p);
444 PROC_VMSPACE_UNLOCK(p);
448 refcnt = vm->vm_refcnt;
449 if (refcnt <= 0) { /* Avoid 0->1 transition */
450 PROC_VMSPACE_UNLOCK(p);
453 } while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt + 1));
454 if (vm != p->p_vmspace) {
455 PROC_VMSPACE_UNLOCK(p);
459 PROC_VMSPACE_UNLOCK(p);
464 _vm_map_lock(vm_map_t map, const char *file, int line)
468 mtx_lock_flags_(&map->system_mtx, 0, file, line);
470 sx_xlock_(&map->lock, file, line);
475 vm_map_process_deferred(void)
478 vm_map_entry_t entry;
482 while ((entry = td->td_map_def_user) != NULL) {
483 td->td_map_def_user = entry->next;
484 if ((entry->eflags & MAP_ENTRY_VN_WRITECNT) != 0) {
486 * Decrement the object's writemappings and
487 * possibly the vnode's v_writecount.
489 KASSERT((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0,
490 ("Submap with writecount"));
491 object = entry->object.vm_object;
492 KASSERT(object != NULL, ("No object for writecount"));
493 vnode_pager_release_writecount(object, entry->start,
496 vm_map_entry_deallocate(entry, FALSE);
501 _vm_map_unlock(vm_map_t map, const char *file, int line)
505 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
507 sx_xunlock_(&map->lock, file, line);
508 vm_map_process_deferred();
513 _vm_map_lock_read(vm_map_t map, const char *file, int line)
517 mtx_lock_flags_(&map->system_mtx, 0, file, line);
519 sx_slock_(&map->lock, file, line);
523 _vm_map_unlock_read(vm_map_t map, const char *file, int line)
527 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
529 sx_sunlock_(&map->lock, file, line);
530 vm_map_process_deferred();
535 _vm_map_trylock(vm_map_t map, const char *file, int line)
539 error = map->system_map ?
540 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) :
541 !sx_try_xlock_(&map->lock, file, line);
548 _vm_map_trylock_read(vm_map_t map, const char *file, int line)
552 error = map->system_map ?
553 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) :
554 !sx_try_slock_(&map->lock, file, line);
559 * _vm_map_lock_upgrade: [ internal use only ]
561 * Tries to upgrade a read (shared) lock on the specified map to a write
562 * (exclusive) lock. Returns the value "0" if the upgrade succeeds and a
563 * non-zero value if the upgrade fails. If the upgrade fails, the map is
564 * returned without a read or write lock held.
566 * Requires that the map be read locked.
569 _vm_map_lock_upgrade(vm_map_t map, const char *file, int line)
571 unsigned int last_timestamp;
573 if (map->system_map) {
574 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
576 if (!sx_try_upgrade_(&map->lock, file, line)) {
577 last_timestamp = map->timestamp;
578 sx_sunlock_(&map->lock, file, line);
579 vm_map_process_deferred();
581 * If the map's timestamp does not change while the
582 * map is unlocked, then the upgrade succeeds.
584 sx_xlock_(&map->lock, file, line);
585 if (last_timestamp != map->timestamp) {
586 sx_xunlock_(&map->lock, file, line);
596 _vm_map_lock_downgrade(vm_map_t map, const char *file, int line)
599 if (map->system_map) {
600 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
602 sx_downgrade_(&map->lock, file, line);
608 * Returns a non-zero value if the caller holds a write (exclusive) lock
609 * on the specified map and the value "0" otherwise.
612 vm_map_locked(vm_map_t map)
616 return (mtx_owned(&map->system_mtx));
618 return (sx_xlocked(&map->lock));
623 _vm_map_assert_locked(vm_map_t map, const char *file, int line)
627 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
629 sx_assert_(&map->lock, SA_XLOCKED, file, line);
632 #define VM_MAP_ASSERT_LOCKED(map) \
633 _vm_map_assert_locked(map, LOCK_FILE, LOCK_LINE)
635 #define VM_MAP_ASSERT_LOCKED(map)
639 * _vm_map_unlock_and_wait:
641 * Atomically releases the lock on the specified map and puts the calling
642 * thread to sleep. The calling thread will remain asleep until either
643 * vm_map_wakeup() is performed on the map or the specified timeout is
646 * WARNING! This function does not perform deferred deallocations of
647 * objects and map entries. Therefore, the calling thread is expected to
648 * reacquire the map lock after reawakening and later perform an ordinary
649 * unlock operation, such as vm_map_unlock(), before completing its
650 * operation on the map.
653 _vm_map_unlock_and_wait(vm_map_t map, int timo, const char *file, int line)
656 mtx_lock(&map_sleep_mtx);
658 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
660 sx_xunlock_(&map->lock, file, line);
661 return (msleep(&map->root, &map_sleep_mtx, PDROP | PVM, "vmmaps",
668 * Awaken any threads that have slept on the map using
669 * vm_map_unlock_and_wait().
672 vm_map_wakeup(vm_map_t map)
676 * Acquire and release map_sleep_mtx to prevent a wakeup()
677 * from being performed (and lost) between the map unlock
678 * and the msleep() in _vm_map_unlock_and_wait().
680 mtx_lock(&map_sleep_mtx);
681 mtx_unlock(&map_sleep_mtx);
686 vm_map_busy(vm_map_t map)
689 VM_MAP_ASSERT_LOCKED(map);
694 vm_map_unbusy(vm_map_t map)
697 VM_MAP_ASSERT_LOCKED(map);
698 KASSERT(map->busy, ("vm_map_unbusy: not busy"));
699 if (--map->busy == 0 && (map->flags & MAP_BUSY_WAKEUP)) {
700 vm_map_modflags(map, 0, MAP_BUSY_WAKEUP);
706 vm_map_wait_busy(vm_map_t map)
709 VM_MAP_ASSERT_LOCKED(map);
711 vm_map_modflags(map, MAP_BUSY_WAKEUP, 0);
713 msleep(&map->busy, &map->system_mtx, 0, "mbusy", 0);
715 sx_sleep(&map->busy, &map->lock, 0, "mbusy", 0);
721 vmspace_resident_count(struct vmspace *vmspace)
723 return pmap_resident_count(vmspace_pmap(vmspace));
727 vmspace_wired_count(struct vmspace *vmspace)
729 return pmap_wired_count(vmspace_pmap(vmspace));
735 * Creates and returns a new empty VM map with
736 * the given physical map structure, and having
737 * the given lower and upper address bounds.
740 vm_map_create(pmap_t pmap, vm_offset_t min, vm_offset_t max)
744 result = uma_zalloc(mapzone, M_WAITOK);
745 CTR1(KTR_VM, "vm_map_create: %p", result);
746 _vm_map_init(result, pmap, min, max);
751 * Initialize an existing vm_map structure
752 * such as that in the vmspace structure.
755 _vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max)
758 map->header.next = map->header.prev = &map->header;
759 map->needs_wakeup = FALSE;
762 map->min_offset = min;
763 map->max_offset = max;
771 vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max)
774 _vm_map_init(map, pmap, min, max);
775 mtx_init(&map->system_mtx, "system map", NULL, MTX_DEF | MTX_DUPOK);
776 sx_init(&map->lock, "user map");
780 * vm_map_entry_dispose: [ internal use only ]
782 * Inverse of vm_map_entry_create.
785 vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry)
787 uma_zfree(map->system_map ? kmapentzone : mapentzone, entry);
791 * vm_map_entry_create: [ internal use only ]
793 * Allocates a VM map entry for insertion.
794 * No entry fields are filled in.
796 static vm_map_entry_t
797 vm_map_entry_create(vm_map_t map)
799 vm_map_entry_t new_entry;
802 new_entry = uma_zalloc(kmapentzone, M_NOWAIT);
804 new_entry = uma_zalloc(mapentzone, M_WAITOK);
805 if (new_entry == NULL)
806 panic("vm_map_entry_create: kernel resources exhausted");
811 * vm_map_entry_set_behavior:
813 * Set the expected access behavior, either normal, random, or
817 vm_map_entry_set_behavior(vm_map_entry_t entry, u_char behavior)
819 entry->eflags = (entry->eflags & ~MAP_ENTRY_BEHAV_MASK) |
820 (behavior & MAP_ENTRY_BEHAV_MASK);
824 * vm_map_entry_set_max_free:
826 * Set the max_free field in a vm_map_entry.
829 vm_map_entry_set_max_free(vm_map_entry_t entry)
832 entry->max_free = entry->adj_free;
833 if (entry->left != NULL && entry->left->max_free > entry->max_free)
834 entry->max_free = entry->left->max_free;
835 if (entry->right != NULL && entry->right->max_free > entry->max_free)
836 entry->max_free = entry->right->max_free;
840 * vm_map_entry_splay:
842 * The Sleator and Tarjan top-down splay algorithm with the
843 * following variation. Max_free must be computed bottom-up, so
844 * on the downward pass, maintain the left and right spines in
845 * reverse order. Then, make a second pass up each side to fix
846 * the pointers and compute max_free. The time bound is O(log n)
849 * The new root is the vm_map_entry containing "addr", or else an
850 * adjacent entry (lower or higher) if addr is not in the tree.
852 * The map must be locked, and leaves it so.
854 * Returns: the new root.
856 static vm_map_entry_t
857 vm_map_entry_splay(vm_offset_t addr, vm_map_entry_t root)
859 vm_map_entry_t llist, rlist;
860 vm_map_entry_t ltree, rtree;
863 /* Special case of empty tree. */
868 * Pass One: Splay down the tree until we find addr or a NULL
869 * pointer where addr would go. llist and rlist are the two
870 * sides in reverse order (bottom-up), with llist linked by
871 * the right pointer and rlist linked by the left pointer in
872 * the vm_map_entry. Wait until Pass Two to set max_free on
878 /* root is never NULL in here. */
879 if (addr < root->start) {
883 if (addr < y->start && y->left != NULL) {
884 /* Rotate right and put y on rlist. */
885 root->left = y->right;
887 vm_map_entry_set_max_free(root);
892 /* Put root on rlist. */
897 } else if (addr >= root->end) {
901 if (addr >= y->end && y->right != NULL) {
902 /* Rotate left and put y on llist. */
903 root->right = y->left;
905 vm_map_entry_set_max_free(root);
910 /* Put root on llist. */
920 * Pass Two: Walk back up the two spines, flip the pointers
921 * and set max_free. The subtrees of the root go at the
922 * bottom of llist and rlist.
925 while (llist != NULL) {
927 llist->right = ltree;
928 vm_map_entry_set_max_free(llist);
933 while (rlist != NULL) {
936 vm_map_entry_set_max_free(rlist);
942 * Final assembly: add ltree and rtree as subtrees of root.
946 vm_map_entry_set_max_free(root);
952 * vm_map_entry_{un,}link:
954 * Insert/remove entries from maps.
957 vm_map_entry_link(vm_map_t map,
958 vm_map_entry_t after_where,
959 vm_map_entry_t entry)
963 "vm_map_entry_link: map %p, nentries %d, entry %p, after %p", map,
964 map->nentries, entry, after_where);
965 VM_MAP_ASSERT_LOCKED(map);
967 entry->prev = after_where;
968 entry->next = after_where->next;
969 entry->next->prev = entry;
970 after_where->next = entry;
972 if (after_where != &map->header) {
973 if (after_where != map->root)
974 vm_map_entry_splay(after_where->start, map->root);
975 entry->right = after_where->right;
976 entry->left = after_where;
977 after_where->right = NULL;
978 after_where->adj_free = entry->start - after_where->end;
979 vm_map_entry_set_max_free(after_where);
981 entry->right = map->root;
984 entry->adj_free = (entry->next == &map->header ? map->max_offset :
985 entry->next->start) - entry->end;
986 vm_map_entry_set_max_free(entry);
991 vm_map_entry_unlink(vm_map_t map,
992 vm_map_entry_t entry)
994 vm_map_entry_t next, prev, root;
996 VM_MAP_ASSERT_LOCKED(map);
997 if (entry != map->root)
998 vm_map_entry_splay(entry->start, map->root);
999 if (entry->left == NULL)
1000 root = entry->right;
1002 root = vm_map_entry_splay(entry->start, entry->left);
1003 root->right = entry->right;
1004 root->adj_free = (entry->next == &map->header ? map->max_offset :
1005 entry->next->start) - root->end;
1006 vm_map_entry_set_max_free(root);
1015 CTR3(KTR_VM, "vm_map_entry_unlink: map %p, nentries %d, entry %p", map,
1016 map->nentries, entry);
1020 * vm_map_entry_resize_free:
1022 * Recompute the amount of free space following a vm_map_entry
1023 * and propagate that value up the tree. Call this function after
1024 * resizing a map entry in-place, that is, without a call to
1025 * vm_map_entry_link() or _unlink().
1027 * The map must be locked, and leaves it so.
1030 vm_map_entry_resize_free(vm_map_t map, vm_map_entry_t entry)
1034 * Using splay trees without parent pointers, propagating
1035 * max_free up the tree is done by moving the entry to the
1036 * root and making the change there.
1038 if (entry != map->root)
1039 map->root = vm_map_entry_splay(entry->start, map->root);
1041 entry->adj_free = (entry->next == &map->header ? map->max_offset :
1042 entry->next->start) - entry->end;
1043 vm_map_entry_set_max_free(entry);
1047 * vm_map_lookup_entry: [ internal use only ]
1049 * Finds the map entry containing (or
1050 * immediately preceding) the specified address
1051 * in the given map; the entry is returned
1052 * in the "entry" parameter. The boolean
1053 * result indicates whether the address is
1054 * actually contained in the map.
1057 vm_map_lookup_entry(
1059 vm_offset_t address,
1060 vm_map_entry_t *entry) /* OUT */
1066 * If the map is empty, then the map entry immediately preceding
1067 * "address" is the map's header.
1071 *entry = &map->header;
1072 else if (address >= cur->start && cur->end > address) {
1075 } else if ((locked = vm_map_locked(map)) ||
1076 sx_try_upgrade(&map->lock)) {
1078 * Splay requires a write lock on the map. However, it only
1079 * restructures the binary search tree; it does not otherwise
1080 * change the map. Thus, the map's timestamp need not change
1081 * on a temporary upgrade.
1083 map->root = cur = vm_map_entry_splay(address, cur);
1085 sx_downgrade(&map->lock);
1088 * If "address" is contained within a map entry, the new root
1089 * is that map entry. Otherwise, the new root is a map entry
1090 * immediately before or after "address".
1092 if (address >= cur->start) {
1094 if (cur->end > address)
1100 * Since the map is only locked for read access, perform a
1101 * standard binary search tree lookup for "address".
1104 if (address < cur->start) {
1105 if (cur->left == NULL) {
1110 } else if (cur->end > address) {
1114 if (cur->right == NULL) {
1127 * Inserts the given whole VM object into the target
1128 * map at the specified address range. The object's
1129 * size should match that of the address range.
1131 * Requires that the map be locked, and leaves it so.
1133 * If object is non-NULL, ref count must be bumped by caller
1134 * prior to making call to account for the new entry.
1137 vm_map_insert(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1138 vm_offset_t start, vm_offset_t end, vm_prot_t prot, vm_prot_t max,
1141 vm_map_entry_t new_entry;
1142 vm_map_entry_t prev_entry;
1143 vm_map_entry_t temp_entry;
1144 vm_eflags_t protoeflags;
1146 vm_inherit_t inheritance;
1147 boolean_t charge_prev_obj;
1149 VM_MAP_ASSERT_LOCKED(map);
1152 * Check that the start and end points are not bogus.
1154 if ((start < map->min_offset) || (end > map->max_offset) ||
1156 return (KERN_INVALID_ADDRESS);
1159 * Find the entry prior to the proposed starting address; if it's part
1160 * of an existing entry, this range is bogus.
1162 if (vm_map_lookup_entry(map, start, &temp_entry))
1163 return (KERN_NO_SPACE);
1165 prev_entry = temp_entry;
1168 * Assert that the next entry doesn't overlap the end point.
1170 if ((prev_entry->next != &map->header) &&
1171 (prev_entry->next->start < end))
1172 return (KERN_NO_SPACE);
1175 charge_prev_obj = FALSE;
1177 if (cow & MAP_COPY_ON_WRITE)
1178 protoeflags |= MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY;
1180 if (cow & MAP_NOFAULT) {
1181 protoeflags |= MAP_ENTRY_NOFAULT;
1183 KASSERT(object == NULL,
1184 ("vm_map_insert: paradoxical MAP_NOFAULT request"));
1186 if (cow & MAP_DISABLE_SYNCER)
1187 protoeflags |= MAP_ENTRY_NOSYNC;
1188 if (cow & MAP_DISABLE_COREDUMP)
1189 protoeflags |= MAP_ENTRY_NOCOREDUMP;
1190 if (cow & MAP_VN_WRITECOUNT)
1191 protoeflags |= MAP_ENTRY_VN_WRITECNT;
1192 if (cow & MAP_INHERIT_SHARE)
1193 inheritance = VM_INHERIT_SHARE;
1195 inheritance = VM_INHERIT_DEFAULT;
1198 KASSERT((object != kmem_object && object != kernel_object) ||
1199 ((object == kmem_object || object == kernel_object) &&
1200 !(protoeflags & MAP_ENTRY_NEEDS_COPY)),
1201 ("kmem or kernel object and cow"));
1202 if (cow & (MAP_ACC_NO_CHARGE | MAP_NOFAULT))
1204 if ((cow & MAP_ACC_CHARGED) || ((prot & VM_PROT_WRITE) &&
1205 ((protoeflags & MAP_ENTRY_NEEDS_COPY) || object == NULL))) {
1206 if (!(cow & MAP_ACC_CHARGED) && !swap_reserve(end - start))
1207 return (KERN_RESOURCE_SHORTAGE);
1208 KASSERT(object == NULL || (protoeflags & MAP_ENTRY_NEEDS_COPY) ||
1209 object->cred == NULL,
1210 ("OVERCOMMIT: vm_map_insert o %p", object));
1211 cred = curthread->td_ucred;
1213 if (object == NULL && !(protoeflags & MAP_ENTRY_NEEDS_COPY))
1214 charge_prev_obj = TRUE;
1218 /* Expand the kernel pmap, if necessary. */
1219 if (map == kernel_map && end > kernel_vm_end)
1220 pmap_growkernel(end);
1221 if (object != NULL) {
1223 * OBJ_ONEMAPPING must be cleared unless this mapping
1224 * is trivially proven to be the only mapping for any
1225 * of the object's pages. (Object granularity
1226 * reference counting is insufficient to recognize
1227 * aliases with precision.)
1229 VM_OBJECT_LOCK(object);
1230 if (object->ref_count > 1 || object->shadow_count != 0)
1231 vm_object_clear_flag(object, OBJ_ONEMAPPING);
1232 VM_OBJECT_UNLOCK(object);
1234 else if ((prev_entry != &map->header) &&
1235 (prev_entry->eflags == protoeflags) &&
1236 (prev_entry->end == start) &&
1237 (prev_entry->wired_count == 0) &&
1238 (prev_entry->cred == cred ||
1239 (prev_entry->object.vm_object != NULL &&
1240 (prev_entry->object.vm_object->cred == cred))) &&
1241 vm_object_coalesce(prev_entry->object.vm_object,
1243 (vm_size_t)(prev_entry->end - prev_entry->start),
1244 (vm_size_t)(end - prev_entry->end), charge_prev_obj)) {
1246 * We were able to extend the object. Determine if we
1247 * can extend the previous map entry to include the
1248 * new range as well.
1250 if ((prev_entry->inheritance == inheritance) &&
1251 (prev_entry->protection == prot) &&
1252 (prev_entry->max_protection == max)) {
1253 map->size += (end - prev_entry->end);
1254 prev_entry->end = end;
1255 vm_map_entry_resize_free(map, prev_entry);
1256 vm_map_simplify_entry(map, prev_entry);
1259 return (KERN_SUCCESS);
1263 * If we can extend the object but cannot extend the
1264 * map entry, we have to create a new map entry. We
1265 * must bump the ref count on the extended object to
1266 * account for it. object may be NULL.
1268 object = prev_entry->object.vm_object;
1269 offset = prev_entry->offset +
1270 (prev_entry->end - prev_entry->start);
1271 vm_object_reference(object);
1272 if (cred != NULL && object != NULL && object->cred != NULL &&
1273 !(prev_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
1274 /* Object already accounts for this uid. */
1281 * NOTE: if conditionals fail, object can be NULL here. This occurs
1282 * in things like the buffer map where we manage kva but do not manage
1287 * Create a new entry
1289 new_entry = vm_map_entry_create(map);
1290 new_entry->start = start;
1291 new_entry->end = end;
1292 new_entry->cred = NULL;
1294 new_entry->eflags = protoeflags;
1295 new_entry->object.vm_object = object;
1296 new_entry->offset = offset;
1297 new_entry->avail_ssize = 0;
1299 new_entry->inheritance = inheritance;
1300 new_entry->protection = prot;
1301 new_entry->max_protection = max;
1302 new_entry->wired_count = 0;
1303 new_entry->read_ahead = VM_FAULT_READ_AHEAD_INIT;
1304 new_entry->next_read = OFF_TO_IDX(offset);
1306 KASSERT(cred == NULL || !ENTRY_CHARGED(new_entry),
1307 ("OVERCOMMIT: vm_map_insert leaks vm_map %p", new_entry));
1308 new_entry->cred = cred;
1311 * Insert the new entry into the list
1313 vm_map_entry_link(map, prev_entry, new_entry);
1314 map->size += new_entry->end - new_entry->start;
1317 * It may be possible to merge the new entry with the next and/or
1318 * previous entries. However, due to MAP_STACK_* being a hack, a
1319 * panic can result from merging such entries.
1321 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0)
1322 vm_map_simplify_entry(map, new_entry);
1324 if (cow & (MAP_PREFAULT|MAP_PREFAULT_PARTIAL)) {
1325 vm_map_pmap_enter(map, start, prot,
1326 object, OFF_TO_IDX(offset), end - start,
1327 cow & MAP_PREFAULT_PARTIAL);
1330 return (KERN_SUCCESS);
1336 * Find the first fit (lowest VM address) for "length" free bytes
1337 * beginning at address >= start in the given map.
1339 * In a vm_map_entry, "adj_free" is the amount of free space
1340 * adjacent (higher address) to this entry, and "max_free" is the
1341 * maximum amount of contiguous free space in its subtree. This
1342 * allows finding a free region in one path down the tree, so
1343 * O(log n) amortized with splay trees.
1345 * The map must be locked, and leaves it so.
1347 * Returns: 0 on success, and starting address in *addr,
1348 * 1 if insufficient space.
1351 vm_map_findspace(vm_map_t map, vm_offset_t start, vm_size_t length,
1352 vm_offset_t *addr) /* OUT */
1354 vm_map_entry_t entry;
1358 * Request must fit within min/max VM address and must avoid
1361 if (start < map->min_offset)
1362 start = map->min_offset;
1363 if (start + length > map->max_offset || start + length < start)
1366 /* Empty tree means wide open address space. */
1367 if (map->root == NULL) {
1373 * After splay, if start comes before root node, then there
1374 * must be a gap from start to the root.
1376 map->root = vm_map_entry_splay(start, map->root);
1377 if (start + length <= map->root->start) {
1383 * Root is the last node that might begin its gap before
1384 * start, and this is the last comparison where address
1385 * wrap might be a problem.
1387 st = (start > map->root->end) ? start : map->root->end;
1388 if (length <= map->root->end + map->root->adj_free - st) {
1393 /* With max_free, can immediately tell if no solution. */
1394 entry = map->root->right;
1395 if (entry == NULL || length > entry->max_free)
1399 * Search the right subtree in the order: left subtree, root,
1400 * right subtree (first fit). The previous splay implies that
1401 * all regions in the right subtree have addresses > start.
1403 while (entry != NULL) {
1404 if (entry->left != NULL && entry->left->max_free >= length)
1405 entry = entry->left;
1406 else if (entry->adj_free >= length) {
1410 entry = entry->right;
1413 /* Can't get here, so panic if we do. */
1414 panic("vm_map_findspace: max_free corrupt");
1418 vm_map_fixed(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1419 vm_offset_t start, vm_size_t length, vm_prot_t prot,
1420 vm_prot_t max, int cow)
1425 end = start + length;
1427 VM_MAP_RANGE_CHECK(map, start, end);
1428 (void) vm_map_delete(map, start, end);
1429 result = vm_map_insert(map, object, offset, start, end, prot,
1436 * vm_map_find finds an unallocated region in the target address
1437 * map with the given length. The search is defined to be
1438 * first-fit from the specified address; the region found is
1439 * returned in the same parameter.
1441 * If object is non-NULL, ref count must be bumped by caller
1442 * prior to making call to account for the new entry.
1445 vm_map_find(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1446 vm_offset_t *addr, /* IN/OUT */
1447 vm_size_t length, int find_space, vm_prot_t prot,
1448 vm_prot_t max, int cow)
1456 if (find_space != VMFS_NO_SPACE) {
1457 if (vm_map_findspace(map, start, length, addr)) {
1459 return (KERN_NO_SPACE);
1461 switch (find_space) {
1462 case VMFS_ALIGNED_SPACE:
1463 pmap_align_superpage(object, offset, addr,
1466 #ifdef VMFS_TLB_ALIGNED_SPACE
1467 case VMFS_TLB_ALIGNED_SPACE:
1468 pmap_align_tlb(addr);
1477 result = vm_map_insert(map, object, offset, start, start +
1478 length, prot, max, cow);
1479 } while (result == KERN_NO_SPACE && (find_space == VMFS_ALIGNED_SPACE
1480 #ifdef VMFS_TLB_ALIGNED_SPACE
1481 || find_space == VMFS_TLB_ALIGNED_SPACE
1489 * vm_map_simplify_entry:
1491 * Simplify the given map entry by merging with either neighbor. This
1492 * routine also has the ability to merge with both neighbors.
1494 * The map must be locked.
1496 * This routine guarentees that the passed entry remains valid (though
1497 * possibly extended). When merging, this routine may delete one or
1501 vm_map_simplify_entry(vm_map_t map, vm_map_entry_t entry)
1503 vm_map_entry_t next, prev;
1504 vm_size_t prevsize, esize;
1506 if (entry->eflags & (MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_IS_SUB_MAP))
1510 if (prev != &map->header) {
1511 prevsize = prev->end - prev->start;
1512 if ( (prev->end == entry->start) &&
1513 (prev->object.vm_object == entry->object.vm_object) &&
1514 (!prev->object.vm_object ||
1515 (prev->offset + prevsize == entry->offset)) &&
1516 (prev->eflags == entry->eflags) &&
1517 (prev->protection == entry->protection) &&
1518 (prev->max_protection == entry->max_protection) &&
1519 (prev->inheritance == entry->inheritance) &&
1520 (prev->wired_count == entry->wired_count) &&
1521 (prev->cred == entry->cred)) {
1522 vm_map_entry_unlink(map, prev);
1523 entry->start = prev->start;
1524 entry->offset = prev->offset;
1525 if (entry->prev != &map->header)
1526 vm_map_entry_resize_free(map, entry->prev);
1529 * If the backing object is a vnode object,
1530 * vm_object_deallocate() calls vrele().
1531 * However, vrele() does not lock the vnode
1532 * because the vnode has additional
1533 * references. Thus, the map lock can be kept
1534 * without causing a lock-order reversal with
1537 * Since we count the number of virtual page
1538 * mappings in object->un_pager.vnp.writemappings,
1539 * the writemappings value should not be adjusted
1540 * when the entry is disposed of.
1542 if (prev->object.vm_object)
1543 vm_object_deallocate(prev->object.vm_object);
1544 if (prev->cred != NULL)
1546 vm_map_entry_dispose(map, prev);
1551 if (next != &map->header) {
1552 esize = entry->end - entry->start;
1553 if ((entry->end == next->start) &&
1554 (next->object.vm_object == entry->object.vm_object) &&
1555 (!entry->object.vm_object ||
1556 (entry->offset + esize == next->offset)) &&
1557 (next->eflags == entry->eflags) &&
1558 (next->protection == entry->protection) &&
1559 (next->max_protection == entry->max_protection) &&
1560 (next->inheritance == entry->inheritance) &&
1561 (next->wired_count == entry->wired_count) &&
1562 (next->cred == entry->cred)) {
1563 vm_map_entry_unlink(map, next);
1564 entry->end = next->end;
1565 vm_map_entry_resize_free(map, entry);
1568 * See comment above.
1570 if (next->object.vm_object)
1571 vm_object_deallocate(next->object.vm_object);
1572 if (next->cred != NULL)
1574 vm_map_entry_dispose(map, next);
1579 * vm_map_clip_start: [ internal use only ]
1581 * Asserts that the given entry begins at or after
1582 * the specified address; if necessary,
1583 * it splits the entry into two.
1585 #define vm_map_clip_start(map, entry, startaddr) \
1587 if (startaddr > entry->start) \
1588 _vm_map_clip_start(map, entry, startaddr); \
1592 * This routine is called only when it is known that
1593 * the entry must be split.
1596 _vm_map_clip_start(vm_map_t map, vm_map_entry_t entry, vm_offset_t start)
1598 vm_map_entry_t new_entry;
1600 VM_MAP_ASSERT_LOCKED(map);
1603 * Split off the front portion -- note that we must insert the new
1604 * entry BEFORE this one, so that this entry has the specified
1607 vm_map_simplify_entry(map, entry);
1610 * If there is no object backing this entry, we might as well create
1611 * one now. If we defer it, an object can get created after the map
1612 * is clipped, and individual objects will be created for the split-up
1613 * map. This is a bit of a hack, but is also about the best place to
1614 * put this improvement.
1616 if (entry->object.vm_object == NULL && !map->system_map) {
1618 object = vm_object_allocate(OBJT_DEFAULT,
1619 atop(entry->end - entry->start));
1620 entry->object.vm_object = object;
1622 if (entry->cred != NULL) {
1623 object->cred = entry->cred;
1624 object->charge = entry->end - entry->start;
1627 } else if (entry->object.vm_object != NULL &&
1628 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) &&
1629 entry->cred != NULL) {
1630 VM_OBJECT_LOCK(entry->object.vm_object);
1631 KASSERT(entry->object.vm_object->cred == NULL,
1632 ("OVERCOMMIT: vm_entry_clip_start: both cred e %p", entry));
1633 entry->object.vm_object->cred = entry->cred;
1634 entry->object.vm_object->charge = entry->end - entry->start;
1635 VM_OBJECT_UNLOCK(entry->object.vm_object);
1639 new_entry = vm_map_entry_create(map);
1640 *new_entry = *entry;
1642 new_entry->end = start;
1643 entry->offset += (start - entry->start);
1644 entry->start = start;
1645 if (new_entry->cred != NULL)
1646 crhold(entry->cred);
1648 vm_map_entry_link(map, entry->prev, new_entry);
1650 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1651 vm_object_reference(new_entry->object.vm_object);
1653 * The object->un_pager.vnp.writemappings for the
1654 * object of MAP_ENTRY_VN_WRITECNT type entry shall be
1655 * kept as is here. The virtual pages are
1656 * re-distributed among the clipped entries, so the sum is
1663 * vm_map_clip_end: [ internal use only ]
1665 * Asserts that the given entry ends at or before
1666 * the specified address; if necessary,
1667 * it splits the entry into two.
1669 #define vm_map_clip_end(map, entry, endaddr) \
1671 if ((endaddr) < (entry->end)) \
1672 _vm_map_clip_end((map), (entry), (endaddr)); \
1676 * This routine is called only when it is known that
1677 * the entry must be split.
1680 _vm_map_clip_end(vm_map_t map, vm_map_entry_t entry, vm_offset_t end)
1682 vm_map_entry_t new_entry;
1684 VM_MAP_ASSERT_LOCKED(map);
1687 * If there is no object backing this entry, we might as well create
1688 * one now. If we defer it, an object can get created after the map
1689 * is clipped, and individual objects will be created for the split-up
1690 * map. This is a bit of a hack, but is also about the best place to
1691 * put this improvement.
1693 if (entry->object.vm_object == NULL && !map->system_map) {
1695 object = vm_object_allocate(OBJT_DEFAULT,
1696 atop(entry->end - entry->start));
1697 entry->object.vm_object = object;
1699 if (entry->cred != NULL) {
1700 object->cred = entry->cred;
1701 object->charge = entry->end - entry->start;
1704 } else if (entry->object.vm_object != NULL &&
1705 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) &&
1706 entry->cred != NULL) {
1707 VM_OBJECT_LOCK(entry->object.vm_object);
1708 KASSERT(entry->object.vm_object->cred == NULL,
1709 ("OVERCOMMIT: vm_entry_clip_end: both cred e %p", entry));
1710 entry->object.vm_object->cred = entry->cred;
1711 entry->object.vm_object->charge = entry->end - entry->start;
1712 VM_OBJECT_UNLOCK(entry->object.vm_object);
1717 * Create a new entry and insert it AFTER the specified entry
1719 new_entry = vm_map_entry_create(map);
1720 *new_entry = *entry;
1722 new_entry->start = entry->end = end;
1723 new_entry->offset += (end - entry->start);
1724 if (new_entry->cred != NULL)
1725 crhold(entry->cred);
1727 vm_map_entry_link(map, entry, new_entry);
1729 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1730 vm_object_reference(new_entry->object.vm_object);
1735 * vm_map_submap: [ kernel use only ]
1737 * Mark the given range as handled by a subordinate map.
1739 * This range must have been created with vm_map_find,
1740 * and no other operations may have been performed on this
1741 * range prior to calling vm_map_submap.
1743 * Only a limited number of operations can be performed
1744 * within this rage after calling vm_map_submap:
1746 * [Don't try vm_map_copy!]
1748 * To remove a submapping, one must first remove the
1749 * range from the superior map, and then destroy the
1750 * submap (if desired). [Better yet, don't try it.]
1759 vm_map_entry_t entry;
1760 int result = KERN_INVALID_ARGUMENT;
1764 VM_MAP_RANGE_CHECK(map, start, end);
1766 if (vm_map_lookup_entry(map, start, &entry)) {
1767 vm_map_clip_start(map, entry, start);
1769 entry = entry->next;
1771 vm_map_clip_end(map, entry, end);
1773 if ((entry->start == start) && (entry->end == end) &&
1774 ((entry->eflags & MAP_ENTRY_COW) == 0) &&
1775 (entry->object.vm_object == NULL)) {
1776 entry->object.sub_map = submap;
1777 entry->eflags |= MAP_ENTRY_IS_SUB_MAP;
1778 result = KERN_SUCCESS;
1786 * The maximum number of pages to map
1788 #define MAX_INIT_PT 96
1791 * vm_map_pmap_enter:
1793 * Preload read-only mappings for the given object's resident pages into
1794 * the given map. This eliminates the soft faults on process startup and
1795 * immediately after an mmap(2). Because these are speculative mappings,
1796 * cached pages are not reactivated and mapped.
1799 vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot,
1800 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags)
1803 vm_page_t p, p_start;
1804 vm_pindex_t psize, tmpidx;
1806 if ((prot & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0 || object == NULL)
1808 VM_OBJECT_LOCK(object);
1809 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) {
1810 pmap_object_init_pt(map->pmap, addr, object, pindex, size);
1816 if ((flags & MAP_PREFAULT_PARTIAL) && psize > MAX_INIT_PT &&
1817 object->resident_page_count > MAX_INIT_PT)
1820 if (psize + pindex > object->size) {
1821 if (object->size < pindex)
1823 psize = object->size - pindex;
1829 p = vm_page_find_least(object, pindex);
1831 * Assert: the variable p is either (1) the page with the
1832 * least pindex greater than or equal to the parameter pindex
1836 p != NULL && (tmpidx = p->pindex - pindex) < psize;
1837 p = TAILQ_NEXT(p, listq)) {
1839 * don't allow an madvise to blow away our really
1840 * free pages allocating pv entries.
1842 if ((flags & MAP_PREFAULT_MADVISE) &&
1843 cnt.v_free_count < cnt.v_free_reserved) {
1847 if (p->valid == VM_PAGE_BITS_ALL) {
1848 if (p_start == NULL) {
1849 start = addr + ptoa(tmpidx);
1852 } else if (p_start != NULL) {
1853 pmap_enter_object(map->pmap, start, addr +
1854 ptoa(tmpidx), p_start, prot);
1858 if (p_start != NULL)
1859 pmap_enter_object(map->pmap, start, addr + ptoa(psize),
1862 VM_OBJECT_UNLOCK(object);
1868 * Sets the protection of the specified address
1869 * region in the target map. If "set_max" is
1870 * specified, the maximum protection is to be set;
1871 * otherwise, only the current protection is affected.
1874 vm_map_protect(vm_map_t map, vm_offset_t start, vm_offset_t end,
1875 vm_prot_t new_prot, boolean_t set_max)
1877 vm_map_entry_t current, entry;
1884 VM_MAP_RANGE_CHECK(map, start, end);
1886 if (vm_map_lookup_entry(map, start, &entry)) {
1887 vm_map_clip_start(map, entry, start);
1889 entry = entry->next;
1893 * Make a first pass to check for protection violations.
1896 while ((current != &map->header) && (current->start < end)) {
1897 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
1899 return (KERN_INVALID_ARGUMENT);
1901 if ((new_prot & current->max_protection) != new_prot) {
1903 return (KERN_PROTECTION_FAILURE);
1905 current = current->next;
1910 * Do an accounting pass for private read-only mappings that
1911 * now will do cow due to allowed write (e.g. debugger sets
1912 * breakpoint on text segment)
1914 for (current = entry; (current != &map->header) &&
1915 (current->start < end); current = current->next) {
1917 vm_map_clip_end(map, current, end);
1920 ((new_prot & ~(current->protection)) & VM_PROT_WRITE) == 0 ||
1921 ENTRY_CHARGED(current)) {
1925 cred = curthread->td_ucred;
1926 obj = current->object.vm_object;
1928 if (obj == NULL || (current->eflags & MAP_ENTRY_NEEDS_COPY)) {
1929 if (!swap_reserve(current->end - current->start)) {
1931 return (KERN_RESOURCE_SHORTAGE);
1934 current->cred = cred;
1938 VM_OBJECT_LOCK(obj);
1939 if (obj->type != OBJT_DEFAULT && obj->type != OBJT_SWAP) {
1940 VM_OBJECT_UNLOCK(obj);
1945 * Charge for the whole object allocation now, since
1946 * we cannot distinguish between non-charged and
1947 * charged clipped mapping of the same object later.
1949 KASSERT(obj->charge == 0,
1950 ("vm_map_protect: object %p overcharged\n", obj));
1951 if (!swap_reserve(ptoa(obj->size))) {
1952 VM_OBJECT_UNLOCK(obj);
1954 return (KERN_RESOURCE_SHORTAGE);
1959 obj->charge = ptoa(obj->size);
1960 VM_OBJECT_UNLOCK(obj);
1964 * Go back and fix up protections. [Note that clipping is not
1965 * necessary the second time.]
1968 while ((current != &map->header) && (current->start < end)) {
1969 old_prot = current->protection;
1972 current->protection =
1973 (current->max_protection = new_prot) &
1976 current->protection = new_prot;
1978 if ((current->eflags & (MAP_ENTRY_COW | MAP_ENTRY_USER_WIRED))
1979 == (MAP_ENTRY_COW | MAP_ENTRY_USER_WIRED) &&
1980 (current->protection & VM_PROT_WRITE) != 0 &&
1981 (old_prot & VM_PROT_WRITE) == 0) {
1982 vm_fault_copy_entry(map, map, current, current, NULL);
1986 * When restricting access, update the physical map. Worry
1987 * about copy-on-write here.
1989 if ((old_prot & ~current->protection) != 0) {
1990 #define MASK(entry) (((entry)->eflags & MAP_ENTRY_COW) ? ~VM_PROT_WRITE : \
1992 pmap_protect(map->pmap, current->start,
1994 current->protection & MASK(current));
1997 vm_map_simplify_entry(map, current);
1998 current = current->next;
2001 return (KERN_SUCCESS);
2007 * This routine traverses a processes map handling the madvise
2008 * system call. Advisories are classified as either those effecting
2009 * the vm_map_entry structure, or those effecting the underlying
2019 vm_map_entry_t current, entry;
2023 * Some madvise calls directly modify the vm_map_entry, in which case
2024 * we need to use an exclusive lock on the map and we need to perform
2025 * various clipping operations. Otherwise we only need a read-lock
2030 case MADV_SEQUENTIAL:
2042 vm_map_lock_read(map);
2045 return (KERN_INVALID_ARGUMENT);
2049 * Locate starting entry and clip if necessary.
2051 VM_MAP_RANGE_CHECK(map, start, end);
2053 if (vm_map_lookup_entry(map, start, &entry)) {
2055 vm_map_clip_start(map, entry, start);
2057 entry = entry->next;
2062 * madvise behaviors that are implemented in the vm_map_entry.
2064 * We clip the vm_map_entry so that behavioral changes are
2065 * limited to the specified address range.
2067 for (current = entry;
2068 (current != &map->header) && (current->start < end);
2069 current = current->next
2071 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
2074 vm_map_clip_end(map, current, end);
2078 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_NORMAL);
2080 case MADV_SEQUENTIAL:
2081 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_SEQUENTIAL);
2084 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_RANDOM);
2087 current->eflags |= MAP_ENTRY_NOSYNC;
2090 current->eflags &= ~MAP_ENTRY_NOSYNC;
2093 current->eflags |= MAP_ENTRY_NOCOREDUMP;
2096 current->eflags &= ~MAP_ENTRY_NOCOREDUMP;
2101 vm_map_simplify_entry(map, current);
2109 * madvise behaviors that are implemented in the underlying
2112 * Since we don't clip the vm_map_entry, we have to clip
2113 * the vm_object pindex and count.
2115 for (current = entry;
2116 (current != &map->header) && (current->start < end);
2117 current = current->next
2119 vm_offset_t useStart;
2121 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
2124 pindex = OFF_TO_IDX(current->offset);
2125 count = atop(current->end - current->start);
2126 useStart = current->start;
2128 if (current->start < start) {
2129 pindex += atop(start - current->start);
2130 count -= atop(start - current->start);
2133 if (current->end > end)
2134 count -= atop(current->end - end);
2139 vm_object_madvise(current->object.vm_object,
2140 pindex, count, behav);
2141 if (behav == MADV_WILLNEED) {
2142 vm_map_pmap_enter(map,
2144 current->protection,
2145 current->object.vm_object,
2147 (count << PAGE_SHIFT),
2148 MAP_PREFAULT_MADVISE
2152 vm_map_unlock_read(map);
2161 * Sets the inheritance of the specified address
2162 * range in the target map. Inheritance
2163 * affects how the map will be shared with
2164 * child maps at the time of vmspace_fork.
2167 vm_map_inherit(vm_map_t map, vm_offset_t start, vm_offset_t end,
2168 vm_inherit_t new_inheritance)
2170 vm_map_entry_t entry;
2171 vm_map_entry_t temp_entry;
2173 switch (new_inheritance) {
2174 case VM_INHERIT_NONE:
2175 case VM_INHERIT_COPY:
2176 case VM_INHERIT_SHARE:
2179 return (KERN_INVALID_ARGUMENT);
2182 VM_MAP_RANGE_CHECK(map, start, end);
2183 if (vm_map_lookup_entry(map, start, &temp_entry)) {
2185 vm_map_clip_start(map, entry, start);
2187 entry = temp_entry->next;
2188 while ((entry != &map->header) && (entry->start < end)) {
2189 vm_map_clip_end(map, entry, end);
2190 entry->inheritance = new_inheritance;
2191 vm_map_simplify_entry(map, entry);
2192 entry = entry->next;
2195 return (KERN_SUCCESS);
2201 * Implements both kernel and user unwiring.
2204 vm_map_unwire(vm_map_t map, vm_offset_t start, vm_offset_t end,
2207 vm_map_entry_t entry, first_entry, tmp_entry;
2208 vm_offset_t saved_start;
2209 unsigned int last_timestamp;
2211 boolean_t need_wakeup, result, user_unwire;
2213 user_unwire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE;
2215 VM_MAP_RANGE_CHECK(map, start, end);
2216 if (!vm_map_lookup_entry(map, start, &first_entry)) {
2217 if (flags & VM_MAP_WIRE_HOLESOK)
2218 first_entry = first_entry->next;
2221 return (KERN_INVALID_ADDRESS);
2224 last_timestamp = map->timestamp;
2225 entry = first_entry;
2226 while (entry != &map->header && entry->start < end) {
2227 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
2229 * We have not yet clipped the entry.
2231 saved_start = (start >= entry->start) ? start :
2233 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2234 if (vm_map_unlock_and_wait(map, 0)) {
2236 * Allow interruption of user unwiring?
2240 if (last_timestamp+1 != map->timestamp) {
2242 * Look again for the entry because the map was
2243 * modified while it was unlocked.
2244 * Specifically, the entry may have been
2245 * clipped, merged, or deleted.
2247 if (!vm_map_lookup_entry(map, saved_start,
2249 if (flags & VM_MAP_WIRE_HOLESOK)
2250 tmp_entry = tmp_entry->next;
2252 if (saved_start == start) {
2254 * First_entry has been deleted.
2257 return (KERN_INVALID_ADDRESS);
2260 rv = KERN_INVALID_ADDRESS;
2264 if (entry == first_entry)
2265 first_entry = tmp_entry;
2270 last_timestamp = map->timestamp;
2273 vm_map_clip_start(map, entry, start);
2274 vm_map_clip_end(map, entry, end);
2276 * Mark the entry in case the map lock is released. (See
2279 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
2281 * Check the map for holes in the specified region.
2282 * If VM_MAP_WIRE_HOLESOK was specified, skip this check.
2284 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) &&
2285 (entry->end < end && (entry->next == &map->header ||
2286 entry->next->start > entry->end))) {
2288 rv = KERN_INVALID_ADDRESS;
2292 * If system unwiring, require that the entry is system wired.
2295 vm_map_entry_system_wired_count(entry) == 0) {
2297 rv = KERN_INVALID_ARGUMENT;
2300 entry = entry->next;
2304 need_wakeup = FALSE;
2305 if (first_entry == NULL) {
2306 result = vm_map_lookup_entry(map, start, &first_entry);
2307 if (!result && (flags & VM_MAP_WIRE_HOLESOK))
2308 first_entry = first_entry->next;
2310 KASSERT(result, ("vm_map_unwire: lookup failed"));
2312 entry = first_entry;
2313 while (entry != &map->header && entry->start < end) {
2314 if (rv == KERN_SUCCESS && (!user_unwire ||
2315 (entry->eflags & MAP_ENTRY_USER_WIRED))) {
2317 entry->eflags &= ~MAP_ENTRY_USER_WIRED;
2318 entry->wired_count--;
2319 if (entry->wired_count == 0) {
2321 * Retain the map lock.
2323 vm_fault_unwire(map, entry->start, entry->end,
2324 entry->object.vm_object != NULL &&
2325 (entry->object.vm_object->type == OBJT_DEVICE ||
2326 entry->object.vm_object->type == OBJT_SG));
2329 KASSERT(entry->eflags & MAP_ENTRY_IN_TRANSITION,
2330 ("vm_map_unwire: in-transition flag missing"));
2331 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION;
2332 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
2333 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
2336 vm_map_simplify_entry(map, entry);
2337 entry = entry->next;
2348 * Implements both kernel and user wiring.
2351 vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t end,
2354 vm_map_entry_t entry, first_entry, tmp_entry;
2355 vm_offset_t saved_end, saved_start;
2356 unsigned int last_timestamp;
2358 boolean_t fictitious, need_wakeup, result, user_wire;
2362 if (flags & VM_MAP_WIRE_WRITE)
2363 prot |= VM_PROT_WRITE;
2364 user_wire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE;
2366 VM_MAP_RANGE_CHECK(map, start, end);
2367 if (!vm_map_lookup_entry(map, start, &first_entry)) {
2368 if (flags & VM_MAP_WIRE_HOLESOK)
2369 first_entry = first_entry->next;
2372 return (KERN_INVALID_ADDRESS);
2375 last_timestamp = map->timestamp;
2376 entry = first_entry;
2377 while (entry != &map->header && entry->start < end) {
2378 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
2380 * We have not yet clipped the entry.
2382 saved_start = (start >= entry->start) ? start :
2384 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2385 if (vm_map_unlock_and_wait(map, 0)) {
2387 * Allow interruption of user wiring?
2391 if (last_timestamp + 1 != map->timestamp) {
2393 * Look again for the entry because the map was
2394 * modified while it was unlocked.
2395 * Specifically, the entry may have been
2396 * clipped, merged, or deleted.
2398 if (!vm_map_lookup_entry(map, saved_start,
2400 if (flags & VM_MAP_WIRE_HOLESOK)
2401 tmp_entry = tmp_entry->next;
2403 if (saved_start == start) {
2405 * first_entry has been deleted.
2408 return (KERN_INVALID_ADDRESS);
2411 rv = KERN_INVALID_ADDRESS;
2415 if (entry == first_entry)
2416 first_entry = tmp_entry;
2421 last_timestamp = map->timestamp;
2424 vm_map_clip_start(map, entry, start);
2425 vm_map_clip_end(map, entry, end);
2427 * Mark the entry in case the map lock is released. (See
2430 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
2431 if ((entry->protection & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0
2432 || (entry->protection & prot) != prot) {
2433 entry->eflags |= MAP_ENTRY_WIRE_SKIPPED;
2434 if ((flags & VM_MAP_WIRE_HOLESOK) == 0) {
2436 rv = KERN_INVALID_ADDRESS;
2441 if (entry->wired_count == 0) {
2442 entry->wired_count++;
2443 saved_start = entry->start;
2444 saved_end = entry->end;
2445 fictitious = entry->object.vm_object != NULL &&
2446 (entry->object.vm_object->type == OBJT_DEVICE ||
2447 entry->object.vm_object->type == OBJT_SG);
2449 * Release the map lock, relying on the in-transition
2450 * mark. Mark the map busy for fork.
2454 rv = vm_fault_wire(map, saved_start, saved_end,
2458 if (last_timestamp + 1 != map->timestamp) {
2460 * Look again for the entry because the map was
2461 * modified while it was unlocked. The entry
2462 * may have been clipped, but NOT merged or
2465 result = vm_map_lookup_entry(map, saved_start,
2467 KASSERT(result, ("vm_map_wire: lookup failed"));
2468 if (entry == first_entry)
2469 first_entry = tmp_entry;
2473 while (entry->end < saved_end) {
2474 if (rv != KERN_SUCCESS) {
2475 KASSERT(entry->wired_count == 1,
2476 ("vm_map_wire: bad count"));
2477 entry->wired_count = -1;
2479 entry = entry->next;
2482 last_timestamp = map->timestamp;
2483 if (rv != KERN_SUCCESS) {
2484 KASSERT(entry->wired_count == 1,
2485 ("vm_map_wire: bad count"));
2487 * Assign an out-of-range value to represent
2488 * the failure to wire this entry.
2490 entry->wired_count = -1;
2494 } else if (!user_wire ||
2495 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
2496 entry->wired_count++;
2499 * Check the map for holes in the specified region.
2500 * If VM_MAP_WIRE_HOLESOK was specified, skip this check.
2503 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) &&
2504 (entry->end < end && (entry->next == &map->header ||
2505 entry->next->start > entry->end))) {
2507 rv = KERN_INVALID_ADDRESS;
2510 entry = entry->next;
2514 need_wakeup = FALSE;
2515 if (first_entry == NULL) {
2516 result = vm_map_lookup_entry(map, start, &first_entry);
2517 if (!result && (flags & VM_MAP_WIRE_HOLESOK))
2518 first_entry = first_entry->next;
2520 KASSERT(result, ("vm_map_wire: lookup failed"));
2522 entry = first_entry;
2523 while (entry != &map->header && entry->start < end) {
2524 if ((entry->eflags & MAP_ENTRY_WIRE_SKIPPED) != 0)
2525 goto next_entry_done;
2526 if (rv == KERN_SUCCESS) {
2528 entry->eflags |= MAP_ENTRY_USER_WIRED;
2529 } else if (entry->wired_count == -1) {
2531 * Wiring failed on this entry. Thus, unwiring is
2534 entry->wired_count = 0;
2537 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0)
2538 entry->wired_count--;
2539 if (entry->wired_count == 0) {
2541 * Retain the map lock.
2543 vm_fault_unwire(map, entry->start, entry->end,
2544 entry->object.vm_object != NULL &&
2545 (entry->object.vm_object->type == OBJT_DEVICE ||
2546 entry->object.vm_object->type == OBJT_SG));
2550 KASSERT(entry->eflags & MAP_ENTRY_IN_TRANSITION,
2551 ("vm_map_wire: in-transition flag missing"));
2552 entry->eflags &= ~(MAP_ENTRY_IN_TRANSITION|MAP_ENTRY_WIRE_SKIPPED);
2553 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
2554 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
2557 vm_map_simplify_entry(map, entry);
2558 entry = entry->next;
2569 * Push any dirty cached pages in the address range to their pager.
2570 * If syncio is TRUE, dirty pages are written synchronously.
2571 * If invalidate is TRUE, any cached pages are freed as well.
2573 * If the size of the region from start to end is zero, we are
2574 * supposed to flush all modified pages within the region containing
2575 * start. Unfortunately, a region can be split or coalesced with
2576 * neighboring regions, making it difficult to determine what the
2577 * original region was. Therefore, we approximate this requirement by
2578 * flushing the current region containing start.
2580 * Returns an error if any part of the specified range is not mapped.
2588 boolean_t invalidate)
2590 vm_map_entry_t current;
2591 vm_map_entry_t entry;
2594 vm_ooffset_t offset;
2595 unsigned int last_timestamp;
2598 vm_map_lock_read(map);
2599 VM_MAP_RANGE_CHECK(map, start, end);
2600 if (!vm_map_lookup_entry(map, start, &entry)) {
2601 vm_map_unlock_read(map);
2602 return (KERN_INVALID_ADDRESS);
2603 } else if (start == end) {
2604 start = entry->start;
2608 * Make a first pass to check for user-wired memory and holes.
2610 for (current = entry; current != &map->header && current->start < end;
2611 current = current->next) {
2612 if (invalidate && (current->eflags & MAP_ENTRY_USER_WIRED)) {
2613 vm_map_unlock_read(map);
2614 return (KERN_INVALID_ARGUMENT);
2616 if (end > current->end &&
2617 (current->next == &map->header ||
2618 current->end != current->next->start)) {
2619 vm_map_unlock_read(map);
2620 return (KERN_INVALID_ADDRESS);
2625 pmap_remove(map->pmap, start, end);
2629 * Make a second pass, cleaning/uncaching pages from the indicated
2632 for (current = entry; current != &map->header && current->start < end;) {
2633 offset = current->offset + (start - current->start);
2634 size = (end <= current->end ? end : current->end) - start;
2635 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
2637 vm_map_entry_t tentry;
2640 smap = current->object.sub_map;
2641 vm_map_lock_read(smap);
2642 (void) vm_map_lookup_entry(smap, offset, &tentry);
2643 tsize = tentry->end - offset;
2646 object = tentry->object.vm_object;
2647 offset = tentry->offset + (offset - tentry->start);
2648 vm_map_unlock_read(smap);
2650 object = current->object.vm_object;
2652 vm_object_reference(object);
2653 last_timestamp = map->timestamp;
2654 vm_map_unlock_read(map);
2655 if (!vm_object_sync(object, offset, size, syncio, invalidate))
2658 vm_object_deallocate(object);
2659 vm_map_lock_read(map);
2660 if (last_timestamp == map->timestamp ||
2661 !vm_map_lookup_entry(map, start, ¤t))
2662 current = current->next;
2665 vm_map_unlock_read(map);
2666 return (failed ? KERN_FAILURE : KERN_SUCCESS);
2670 * vm_map_entry_unwire: [ internal use only ]
2672 * Make the region specified by this entry pageable.
2674 * The map in question should be locked.
2675 * [This is the reason for this routine's existence.]
2678 vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry)
2680 vm_fault_unwire(map, entry->start, entry->end,
2681 entry->object.vm_object != NULL &&
2682 (entry->object.vm_object->type == OBJT_DEVICE ||
2683 entry->object.vm_object->type == OBJT_SG));
2684 entry->wired_count = 0;
2688 vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map)
2691 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0)
2692 vm_object_deallocate(entry->object.vm_object);
2693 uma_zfree(system_map ? kmapentzone : mapentzone, entry);
2697 * vm_map_entry_delete: [ internal use only ]
2699 * Deallocate the given entry from the target map.
2702 vm_map_entry_delete(vm_map_t map, vm_map_entry_t entry)
2705 vm_pindex_t offidxstart, offidxend, count, size1;
2708 vm_map_entry_unlink(map, entry);
2709 object = entry->object.vm_object;
2710 size = entry->end - entry->start;
2713 if (entry->cred != NULL) {
2714 swap_release_by_cred(size, entry->cred);
2715 crfree(entry->cred);
2718 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0 &&
2720 KASSERT(entry->cred == NULL || object->cred == NULL ||
2721 (entry->eflags & MAP_ENTRY_NEEDS_COPY),
2722 ("OVERCOMMIT vm_map_entry_delete: both cred %p", entry));
2723 count = OFF_TO_IDX(size);
2724 offidxstart = OFF_TO_IDX(entry->offset);
2725 offidxend = offidxstart + count;
2726 VM_OBJECT_LOCK(object);
2727 if (object->ref_count != 1 &&
2728 ((object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING ||
2729 object == kernel_object || object == kmem_object)) {
2730 vm_object_collapse(object);
2733 * The option OBJPR_NOTMAPPED can be passed here
2734 * because vm_map_delete() already performed
2735 * pmap_remove() on the only mapping to this range
2738 vm_object_page_remove(object, offidxstart, offidxend,
2740 if (object->type == OBJT_SWAP)
2741 swap_pager_freespace(object, offidxstart, count);
2742 if (offidxend >= object->size &&
2743 offidxstart < object->size) {
2744 size1 = object->size;
2745 object->size = offidxstart;
2746 if (object->cred != NULL) {
2747 size1 -= object->size;
2748 KASSERT(object->charge >= ptoa(size1),
2749 ("vm_map_entry_delete: object->charge < 0"));
2750 swap_release_by_cred(ptoa(size1), object->cred);
2751 object->charge -= ptoa(size1);
2755 VM_OBJECT_UNLOCK(object);
2757 entry->object.vm_object = NULL;
2758 if (map->system_map)
2759 vm_map_entry_deallocate(entry, TRUE);
2761 entry->next = curthread->td_map_def_user;
2762 curthread->td_map_def_user = entry;
2767 * vm_map_delete: [ internal use only ]
2769 * Deallocates the given address range from the target
2773 vm_map_delete(vm_map_t map, vm_offset_t start, vm_offset_t end)
2775 vm_map_entry_t entry;
2776 vm_map_entry_t first_entry;
2778 VM_MAP_ASSERT_LOCKED(map);
2781 * Find the start of the region, and clip it
2783 if (!vm_map_lookup_entry(map, start, &first_entry))
2784 entry = first_entry->next;
2786 entry = first_entry;
2787 vm_map_clip_start(map, entry, start);
2791 * Step through all entries in this region
2793 while ((entry != &map->header) && (entry->start < end)) {
2794 vm_map_entry_t next;
2797 * Wait for wiring or unwiring of an entry to complete.
2798 * Also wait for any system wirings to disappear on
2801 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 ||
2802 (vm_map_pmap(map) != kernel_pmap &&
2803 vm_map_entry_system_wired_count(entry) != 0)) {
2804 unsigned int last_timestamp;
2805 vm_offset_t saved_start;
2806 vm_map_entry_t tmp_entry;
2808 saved_start = entry->start;
2809 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2810 last_timestamp = map->timestamp;
2811 (void) vm_map_unlock_and_wait(map, 0);
2813 if (last_timestamp + 1 != map->timestamp) {
2815 * Look again for the entry because the map was
2816 * modified while it was unlocked.
2817 * Specifically, the entry may have been
2818 * clipped, merged, or deleted.
2820 if (!vm_map_lookup_entry(map, saved_start,
2822 entry = tmp_entry->next;
2825 vm_map_clip_start(map, entry,
2831 vm_map_clip_end(map, entry, end);
2836 * Unwire before removing addresses from the pmap; otherwise,
2837 * unwiring will put the entries back in the pmap.
2839 if (entry->wired_count != 0) {
2840 vm_map_entry_unwire(map, entry);
2843 pmap_remove(map->pmap, entry->start, entry->end);
2846 * Delete the entry only after removing all pmap
2847 * entries pointing to its pages. (Otherwise, its
2848 * page frames may be reallocated, and any modify bits
2849 * will be set in the wrong object!)
2851 vm_map_entry_delete(map, entry);
2854 return (KERN_SUCCESS);
2860 * Remove the given address range from the target map.
2861 * This is the exported form of vm_map_delete.
2864 vm_map_remove(vm_map_t map, vm_offset_t start, vm_offset_t end)
2869 VM_MAP_RANGE_CHECK(map, start, end);
2870 result = vm_map_delete(map, start, end);
2876 * vm_map_check_protection:
2878 * Assert that the target map allows the specified privilege on the
2879 * entire address region given. The entire region must be allocated.
2881 * WARNING! This code does not and should not check whether the
2882 * contents of the region is accessible. For example a smaller file
2883 * might be mapped into a larger address space.
2885 * NOTE! This code is also called by munmap().
2887 * The map must be locked. A read lock is sufficient.
2890 vm_map_check_protection(vm_map_t map, vm_offset_t start, vm_offset_t end,
2891 vm_prot_t protection)
2893 vm_map_entry_t entry;
2894 vm_map_entry_t tmp_entry;
2896 if (!vm_map_lookup_entry(map, start, &tmp_entry))
2900 while (start < end) {
2901 if (entry == &map->header)
2906 if (start < entry->start)
2909 * Check protection associated with entry.
2911 if ((entry->protection & protection) != protection)
2913 /* go to next entry */
2915 entry = entry->next;
2921 * vm_map_copy_entry:
2923 * Copies the contents of the source entry to the destination
2924 * entry. The entries *must* be aligned properly.
2930 vm_map_entry_t src_entry,
2931 vm_map_entry_t dst_entry,
2932 vm_ooffset_t *fork_charge)
2934 vm_object_t src_object;
2935 vm_map_entry_t fake_entry;
2940 VM_MAP_ASSERT_LOCKED(dst_map);
2942 if ((dst_entry->eflags|src_entry->eflags) & MAP_ENTRY_IS_SUB_MAP)
2945 if (src_entry->wired_count == 0) {
2948 * If the source entry is marked needs_copy, it is already
2951 if ((src_entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) {
2952 pmap_protect(src_map->pmap,
2955 src_entry->protection & ~VM_PROT_WRITE);
2959 * Make a copy of the object.
2961 size = src_entry->end - src_entry->start;
2962 if ((src_object = src_entry->object.vm_object) != NULL) {
2963 VM_OBJECT_LOCK(src_object);
2964 charged = ENTRY_CHARGED(src_entry);
2965 if ((src_object->handle == NULL) &&
2966 (src_object->type == OBJT_DEFAULT ||
2967 src_object->type == OBJT_SWAP)) {
2968 vm_object_collapse(src_object);
2969 if ((src_object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING) {
2970 vm_object_split(src_entry);
2971 src_object = src_entry->object.vm_object;
2974 vm_object_reference_locked(src_object);
2975 vm_object_clear_flag(src_object, OBJ_ONEMAPPING);
2976 if (src_entry->cred != NULL &&
2977 !(src_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
2978 KASSERT(src_object->cred == NULL,
2979 ("OVERCOMMIT: vm_map_copy_entry: cred %p",
2981 src_object->cred = src_entry->cred;
2982 src_object->charge = size;
2984 VM_OBJECT_UNLOCK(src_object);
2985 dst_entry->object.vm_object = src_object;
2987 cred = curthread->td_ucred;
2989 dst_entry->cred = cred;
2990 *fork_charge += size;
2991 if (!(src_entry->eflags &
2992 MAP_ENTRY_NEEDS_COPY)) {
2994 src_entry->cred = cred;
2995 *fork_charge += size;
2998 src_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY);
2999 dst_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY);
3000 dst_entry->offset = src_entry->offset;
3001 if (src_entry->eflags & MAP_ENTRY_VN_WRITECNT) {
3003 * MAP_ENTRY_VN_WRITECNT cannot
3004 * indicate write reference from
3005 * src_entry, since the entry is
3006 * marked as needs copy. Allocate a
3007 * fake entry that is used to
3008 * decrement object->un_pager.vnp.writecount
3009 * at the appropriate time. Attach
3010 * fake_entry to the deferred list.
3012 fake_entry = vm_map_entry_create(dst_map);
3013 fake_entry->eflags = MAP_ENTRY_VN_WRITECNT;
3014 src_entry->eflags &= ~MAP_ENTRY_VN_WRITECNT;
3015 vm_object_reference(src_object);
3016 fake_entry->object.vm_object = src_object;
3017 fake_entry->start = src_entry->start;
3018 fake_entry->end = src_entry->end;
3019 fake_entry->next = curthread->td_map_def_user;
3020 curthread->td_map_def_user = fake_entry;
3023 dst_entry->object.vm_object = NULL;
3024 dst_entry->offset = 0;
3025 if (src_entry->cred != NULL) {
3026 dst_entry->cred = curthread->td_ucred;
3027 crhold(dst_entry->cred);
3028 *fork_charge += size;
3032 pmap_copy(dst_map->pmap, src_map->pmap, dst_entry->start,
3033 dst_entry->end - dst_entry->start, src_entry->start);
3036 * Of course, wired down pages can't be set copy-on-write.
3037 * Cause wired pages to be copied into the new map by
3038 * simulating faults (the new pages are pageable)
3040 vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry,
3046 * vmspace_map_entry_forked:
3047 * Update the newly-forked vmspace each time a map entry is inherited
3048 * or copied. The values for vm_dsize and vm_tsize are approximate
3049 * (and mostly-obsolete ideas in the face of mmap(2) et al.)
3052 vmspace_map_entry_forked(const struct vmspace *vm1, struct vmspace *vm2,
3053 vm_map_entry_t entry)
3055 vm_size_t entrysize;
3058 entrysize = entry->end - entry->start;
3059 vm2->vm_map.size += entrysize;
3060 if (entry->eflags & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP)) {
3061 vm2->vm_ssize += btoc(entrysize);
3062 } else if (entry->start >= (vm_offset_t)vm1->vm_daddr &&
3063 entry->start < (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)) {
3064 newend = MIN(entry->end,
3065 (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize));
3066 vm2->vm_dsize += btoc(newend - entry->start);
3067 } else if (entry->start >= (vm_offset_t)vm1->vm_taddr &&
3068 entry->start < (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)) {
3069 newend = MIN(entry->end,
3070 (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize));
3071 vm2->vm_tsize += btoc(newend - entry->start);
3077 * Create a new process vmspace structure and vm_map
3078 * based on those of an existing process. The new map
3079 * is based on the old map, according to the inheritance
3080 * values on the regions in that map.
3082 * XXX It might be worth coalescing the entries added to the new vmspace.
3084 * The source map must not be locked.
3087 vmspace_fork(struct vmspace *vm1, vm_ooffset_t *fork_charge)
3089 struct vmspace *vm2;
3090 vm_map_t new_map, old_map;
3091 vm_map_entry_t new_entry, old_entry;
3095 old_map = &vm1->vm_map;
3096 /* Copy immutable fields of vm1 to vm2. */
3097 vm2 = vmspace_alloc(old_map->min_offset, old_map->max_offset);
3100 vm2->vm_taddr = vm1->vm_taddr;
3101 vm2->vm_daddr = vm1->vm_daddr;
3102 vm2->vm_maxsaddr = vm1->vm_maxsaddr;
3103 vm_map_lock(old_map);
3105 vm_map_wait_busy(old_map);
3106 new_map = &vm2->vm_map;
3107 locked = vm_map_trylock(new_map); /* trylock to silence WITNESS */
3108 KASSERT(locked, ("vmspace_fork: lock failed"));
3110 old_entry = old_map->header.next;
3112 while (old_entry != &old_map->header) {
3113 if (old_entry->eflags & MAP_ENTRY_IS_SUB_MAP)
3114 panic("vm_map_fork: encountered a submap");
3116 switch (old_entry->inheritance) {
3117 case VM_INHERIT_NONE:
3120 case VM_INHERIT_SHARE:
3122 * Clone the entry, creating the shared object if necessary.
3124 object = old_entry->object.vm_object;
3125 if (object == NULL) {
3126 object = vm_object_allocate(OBJT_DEFAULT,
3127 atop(old_entry->end - old_entry->start));
3128 old_entry->object.vm_object = object;
3129 old_entry->offset = 0;
3130 if (old_entry->cred != NULL) {
3131 object->cred = old_entry->cred;
3132 object->charge = old_entry->end -
3134 old_entry->cred = NULL;
3139 * Add the reference before calling vm_object_shadow
3140 * to insure that a shadow object is created.
3142 vm_object_reference(object);
3143 if (old_entry->eflags & MAP_ENTRY_NEEDS_COPY) {
3144 vm_object_shadow(&old_entry->object.vm_object,
3146 old_entry->end - old_entry->start);
3147 old_entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
3148 /* Transfer the second reference too. */
3149 vm_object_reference(
3150 old_entry->object.vm_object);
3153 * As in vm_map_simplify_entry(), the
3154 * vnode lock will not be acquired in
3155 * this call to vm_object_deallocate().
3157 vm_object_deallocate(object);
3158 object = old_entry->object.vm_object;
3160 VM_OBJECT_LOCK(object);
3161 vm_object_clear_flag(object, OBJ_ONEMAPPING);
3162 if (old_entry->cred != NULL) {
3163 KASSERT(object->cred == NULL, ("vmspace_fork both cred"));
3164 object->cred = old_entry->cred;
3165 object->charge = old_entry->end - old_entry->start;
3166 old_entry->cred = NULL;
3168 VM_OBJECT_UNLOCK(object);
3171 * Clone the entry, referencing the shared object.
3173 new_entry = vm_map_entry_create(new_map);
3174 *new_entry = *old_entry;
3175 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED |
3176 MAP_ENTRY_IN_TRANSITION);
3177 new_entry->wired_count = 0;
3178 if (new_entry->eflags & MAP_ENTRY_VN_WRITECNT) {
3179 object = new_entry->object.vm_object;
3180 KASSERT(((struct vnode *)object->handle)->
3182 ("vmspace_fork: v_writecount"));
3183 KASSERT(object->un_pager.vnp.writemappings > 0,
3184 ("vmspace_fork: vnp.writecount"));
3185 vnode_pager_update_writecount(object,
3186 new_entry->start, new_entry->end);
3190 * Insert the entry into the new map -- we know we're
3191 * inserting at the end of the new map.
3193 vm_map_entry_link(new_map, new_map->header.prev,
3195 vmspace_map_entry_forked(vm1, vm2, new_entry);
3198 * Update the physical map
3200 pmap_copy(new_map->pmap, old_map->pmap,
3202 (old_entry->end - old_entry->start),
3206 case VM_INHERIT_COPY:
3208 * Clone the entry and link into the map.
3210 new_entry = vm_map_entry_create(new_map);
3211 *new_entry = *old_entry;
3213 * Copied entry is COW over the old object.
3215 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED |
3216 MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_VN_WRITECNT);
3217 new_entry->wired_count = 0;
3218 new_entry->object.vm_object = NULL;
3219 new_entry->cred = NULL;
3220 vm_map_entry_link(new_map, new_map->header.prev,
3222 vmspace_map_entry_forked(vm1, vm2, new_entry);
3223 vm_map_copy_entry(old_map, new_map, old_entry,
3224 new_entry, fork_charge);
3227 old_entry = old_entry->next;
3230 * Use inlined vm_map_unlock() to postpone handling the deferred
3231 * map entries, which cannot be done until both old_map and
3232 * new_map locks are released.
3234 sx_xunlock(&old_map->lock);
3235 sx_xunlock(&new_map->lock);
3236 vm_map_process_deferred();
3242 vm_map_stack(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
3243 vm_prot_t prot, vm_prot_t max, int cow)
3245 vm_map_entry_t new_entry, prev_entry;
3246 vm_offset_t bot, top;
3247 vm_size_t init_ssize;
3252 * The stack orientation is piggybacked with the cow argument.
3253 * Extract it into orient and mask the cow argument so that we
3254 * don't pass it around further.
3255 * NOTE: We explicitly allow bi-directional stacks.
3257 orient = cow & (MAP_STACK_GROWS_DOWN|MAP_STACK_GROWS_UP);
3259 KASSERT(orient != 0, ("No stack grow direction"));
3261 if (addrbos < vm_map_min(map) ||
3262 addrbos > vm_map_max(map) ||
3263 addrbos + max_ssize < addrbos)
3264 return (KERN_NO_SPACE);
3266 init_ssize = (max_ssize < sgrowsiz) ? max_ssize : sgrowsiz;
3268 PROC_LOCK(curthread->td_proc);
3269 vmemlim = lim_cur(curthread->td_proc, RLIMIT_VMEM);
3270 PROC_UNLOCK(curthread->td_proc);
3274 /* If addr is already mapped, no go */
3275 if (vm_map_lookup_entry(map, addrbos, &prev_entry)) {
3277 return (KERN_NO_SPACE);
3280 /* If we would blow our VMEM resource limit, no go */
3281 if (map->size + init_ssize > vmemlim) {
3283 return (KERN_NO_SPACE);
3287 * If we can't accomodate max_ssize in the current mapping, no go.
3288 * However, we need to be aware that subsequent user mappings might
3289 * map into the space we have reserved for stack, and currently this
3290 * space is not protected.
3292 * Hopefully we will at least detect this condition when we try to
3295 if ((prev_entry->next != &map->header) &&
3296 (prev_entry->next->start < addrbos + max_ssize)) {
3298 return (KERN_NO_SPACE);
3302 * We initially map a stack of only init_ssize. We will grow as
3303 * needed later. Depending on the orientation of the stack (i.e.
3304 * the grow direction) we either map at the top of the range, the
3305 * bottom of the range or in the middle.
3307 * Note: we would normally expect prot and max to be VM_PROT_ALL,
3308 * and cow to be 0. Possibly we should eliminate these as input
3309 * parameters, and just pass these values here in the insert call.
3311 if (orient == MAP_STACK_GROWS_DOWN)
3312 bot = addrbos + max_ssize - init_ssize;
3313 else if (orient == MAP_STACK_GROWS_UP)
3316 bot = round_page(addrbos + max_ssize/2 - init_ssize/2);
3317 top = bot + init_ssize;
3318 rv = vm_map_insert(map, NULL, 0, bot, top, prot, max, cow);
3320 /* Now set the avail_ssize amount. */
3321 if (rv == KERN_SUCCESS) {
3322 if (prev_entry != &map->header)
3323 vm_map_clip_end(map, prev_entry, bot);
3324 new_entry = prev_entry->next;
3325 if (new_entry->end != top || new_entry->start != bot)
3326 panic("Bad entry start/end for new stack entry");
3328 new_entry->avail_ssize = max_ssize - init_ssize;
3329 if (orient & MAP_STACK_GROWS_DOWN)
3330 new_entry->eflags |= MAP_ENTRY_GROWS_DOWN;
3331 if (orient & MAP_STACK_GROWS_UP)
3332 new_entry->eflags |= MAP_ENTRY_GROWS_UP;
3339 static int stack_guard_page = 0;
3340 TUNABLE_INT("security.bsd.stack_guard_page", &stack_guard_page);
3341 SYSCTL_INT(_security_bsd, OID_AUTO, stack_guard_page, CTLFLAG_RW,
3342 &stack_guard_page, 0,
3343 "Insert stack guard page ahead of the growable segments.");
3345 /* Attempts to grow a vm stack entry. Returns KERN_SUCCESS if the
3346 * desired address is already mapped, or if we successfully grow
3347 * the stack. Also returns KERN_SUCCESS if addr is outside the
3348 * stack range (this is strange, but preserves compatibility with
3349 * the grow function in vm_machdep.c).
3352 vm_map_growstack(struct proc *p, vm_offset_t addr)
3354 vm_map_entry_t next_entry, prev_entry;
3355 vm_map_entry_t new_entry, stack_entry;
3356 struct vmspace *vm = p->p_vmspace;
3357 vm_map_t map = &vm->vm_map;
3359 size_t grow_amount, max_grow;
3360 rlim_t stacklim, vmemlim;
3361 int is_procstack, rv;
3372 stacklim = lim_cur(p, RLIMIT_STACK);
3373 vmemlim = lim_cur(p, RLIMIT_VMEM);
3376 vm_map_lock_read(map);
3378 /* If addr is already in the entry range, no need to grow.*/
3379 if (vm_map_lookup_entry(map, addr, &prev_entry)) {
3380 vm_map_unlock_read(map);
3381 return (KERN_SUCCESS);
3384 next_entry = prev_entry->next;
3385 if (!(prev_entry->eflags & MAP_ENTRY_GROWS_UP)) {
3387 * This entry does not grow upwards. Since the address lies
3388 * beyond this entry, the next entry (if one exists) has to
3389 * be a downward growable entry. The entry list header is
3390 * never a growable entry, so it suffices to check the flags.
3392 if (!(next_entry->eflags & MAP_ENTRY_GROWS_DOWN)) {
3393 vm_map_unlock_read(map);
3394 return (KERN_SUCCESS);
3396 stack_entry = next_entry;
3399 * This entry grows upward. If the next entry does not at
3400 * least grow downwards, this is the entry we need to grow.
3401 * otherwise we have two possible choices and we have to
3404 if (next_entry->eflags & MAP_ENTRY_GROWS_DOWN) {
3406 * We have two choices; grow the entry closest to
3407 * the address to minimize the amount of growth.
3409 if (addr - prev_entry->end <= next_entry->start - addr)
3410 stack_entry = prev_entry;
3412 stack_entry = next_entry;
3414 stack_entry = prev_entry;
3417 if (stack_entry == next_entry) {
3418 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_DOWN, ("foo"));
3419 KASSERT(addr < stack_entry->start, ("foo"));
3420 end = (prev_entry != &map->header) ? prev_entry->end :
3421 stack_entry->start - stack_entry->avail_ssize;
3422 grow_amount = roundup(stack_entry->start - addr, PAGE_SIZE);
3423 max_grow = stack_entry->start - end;
3425 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_UP, ("foo"));
3426 KASSERT(addr >= stack_entry->end, ("foo"));
3427 end = (next_entry != &map->header) ? next_entry->start :
3428 stack_entry->end + stack_entry->avail_ssize;
3429 grow_amount = roundup(addr + 1 - stack_entry->end, PAGE_SIZE);
3430 max_grow = end - stack_entry->end;
3433 if (grow_amount > stack_entry->avail_ssize) {
3434 vm_map_unlock_read(map);
3435 return (KERN_NO_SPACE);
3439 * If there is no longer enough space between the entries nogo, and
3440 * adjust the available space. Note: this should only happen if the
3441 * user has mapped into the stack area after the stack was created,
3442 * and is probably an error.
3444 * This also effectively destroys any guard page the user might have
3445 * intended by limiting the stack size.
3447 if (grow_amount + (stack_guard_page ? PAGE_SIZE : 0) > max_grow) {
3448 if (vm_map_lock_upgrade(map))
3451 stack_entry->avail_ssize = max_grow;
3454 return (KERN_NO_SPACE);
3457 is_procstack = (addr >= (vm_offset_t)vm->vm_maxsaddr) ? 1 : 0;
3460 * If this is the main process stack, see if we're over the stack
3463 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) {
3464 vm_map_unlock_read(map);
3465 return (KERN_NO_SPACE);
3470 racct_set(p, RACCT_STACK, ctob(vm->vm_ssize) + grow_amount)) {
3472 vm_map_unlock_read(map);
3473 return (KERN_NO_SPACE);
3478 /* Round up the grow amount modulo SGROWSIZ */
3479 grow_amount = roundup (grow_amount, sgrowsiz);
3480 if (grow_amount > stack_entry->avail_ssize)
3481 grow_amount = stack_entry->avail_ssize;
3482 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) {
3483 grow_amount = trunc_page((vm_size_t)stacklim) -
3488 limit = racct_get_available(p, RACCT_STACK);
3490 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > limit))
3491 grow_amount = limit - ctob(vm->vm_ssize);
3494 /* If we would blow our VMEM resource limit, no go */
3495 if (map->size + grow_amount > vmemlim) {
3496 vm_map_unlock_read(map);
3502 if (racct_set(p, RACCT_VMEM, map->size + grow_amount)) {
3504 vm_map_unlock_read(map);
3511 if (vm_map_lock_upgrade(map))
3514 if (stack_entry == next_entry) {
3518 /* Get the preliminary new entry start value */
3519 addr = stack_entry->start - grow_amount;
3522 * If this puts us into the previous entry, cut back our
3523 * growth to the available space. Also, see the note above.
3526 stack_entry->avail_ssize = max_grow;
3528 if (stack_guard_page)
3532 rv = vm_map_insert(map, NULL, 0, addr, stack_entry->start,
3533 next_entry->protection, next_entry->max_protection, 0);
3535 /* Adjust the available stack space by the amount we grew. */
3536 if (rv == KERN_SUCCESS) {
3537 if (prev_entry != &map->header)
3538 vm_map_clip_end(map, prev_entry, addr);
3539 new_entry = prev_entry->next;
3540 KASSERT(new_entry == stack_entry->prev, ("foo"));
3541 KASSERT(new_entry->end == stack_entry->start, ("foo"));
3542 KASSERT(new_entry->start == addr, ("foo"));
3543 grow_amount = new_entry->end - new_entry->start;
3544 new_entry->avail_ssize = stack_entry->avail_ssize -
3546 stack_entry->eflags &= ~MAP_ENTRY_GROWS_DOWN;
3547 new_entry->eflags |= MAP_ENTRY_GROWS_DOWN;
3553 addr = stack_entry->end + grow_amount;
3556 * If this puts us into the next entry, cut back our growth
3557 * to the available space. Also, see the note above.
3560 stack_entry->avail_ssize = end - stack_entry->end;
3562 if (stack_guard_page)
3566 grow_amount = addr - stack_entry->end;
3567 cred = stack_entry->cred;
3568 if (cred == NULL && stack_entry->object.vm_object != NULL)
3569 cred = stack_entry->object.vm_object->cred;
3570 if (cred != NULL && !swap_reserve_by_cred(grow_amount, cred))
3572 /* Grow the underlying object if applicable. */
3573 else if (stack_entry->object.vm_object == NULL ||
3574 vm_object_coalesce(stack_entry->object.vm_object,
3575 stack_entry->offset,
3576 (vm_size_t)(stack_entry->end - stack_entry->start),
3577 (vm_size_t)grow_amount, cred != NULL)) {
3578 map->size += (addr - stack_entry->end);
3579 /* Update the current entry. */
3580 stack_entry->end = addr;
3581 stack_entry->avail_ssize -= grow_amount;
3582 vm_map_entry_resize_free(map, stack_entry);
3585 if (next_entry != &map->header)
3586 vm_map_clip_start(map, next_entry, addr);
3591 if (rv == KERN_SUCCESS && is_procstack)
3592 vm->vm_ssize += btoc(grow_amount);
3597 * Heed the MAP_WIREFUTURE flag if it was set for this process.
3599 if (rv == KERN_SUCCESS && (map->flags & MAP_WIREFUTURE)) {
3601 (stack_entry == next_entry) ? addr : addr - grow_amount,
3602 (stack_entry == next_entry) ? stack_entry->start : addr,
3603 (p->p_flag & P_SYSTEM)
3604 ? VM_MAP_WIRE_SYSTEM|VM_MAP_WIRE_NOHOLES
3605 : VM_MAP_WIRE_USER|VM_MAP_WIRE_NOHOLES);
3610 if (rv != KERN_SUCCESS) {
3612 error = racct_set(p, RACCT_VMEM, map->size);
3613 KASSERT(error == 0, ("decreasing RACCT_VMEM failed"));
3614 error = racct_set(p, RACCT_STACK, ctob(vm->vm_ssize));
3615 KASSERT(error == 0, ("decreasing RACCT_STACK failed"));
3624 * Unshare the specified VM space for exec. If other processes are
3625 * mapped to it, then create a new one. The new vmspace is null.
3628 vmspace_exec(struct proc *p, vm_offset_t minuser, vm_offset_t maxuser)
3630 struct vmspace *oldvmspace = p->p_vmspace;
3631 struct vmspace *newvmspace;
3633 newvmspace = vmspace_alloc(minuser, maxuser);
3634 if (newvmspace == NULL)
3636 newvmspace->vm_swrss = oldvmspace->vm_swrss;
3638 * This code is written like this for prototype purposes. The
3639 * goal is to avoid running down the vmspace here, but let the
3640 * other process's that are still using the vmspace to finally
3641 * run it down. Even though there is little or no chance of blocking
3642 * here, it is a good idea to keep this form for future mods.
3644 PROC_VMSPACE_LOCK(p);
3645 p->p_vmspace = newvmspace;
3646 PROC_VMSPACE_UNLOCK(p);
3647 if (p == curthread->td_proc)
3648 pmap_activate(curthread);
3649 vmspace_free(oldvmspace);
3654 * Unshare the specified VM space for forcing COW. This
3655 * is called by rfork, for the (RFMEM|RFPROC) == 0 case.
3658 vmspace_unshare(struct proc *p)
3660 struct vmspace *oldvmspace = p->p_vmspace;
3661 struct vmspace *newvmspace;
3662 vm_ooffset_t fork_charge;
3664 if (oldvmspace->vm_refcnt == 1)
3667 newvmspace = vmspace_fork(oldvmspace, &fork_charge);
3668 if (newvmspace == NULL)
3670 if (!swap_reserve_by_cred(fork_charge, p->p_ucred)) {
3671 vmspace_free(newvmspace);
3674 PROC_VMSPACE_LOCK(p);
3675 p->p_vmspace = newvmspace;
3676 PROC_VMSPACE_UNLOCK(p);
3677 if (p == curthread->td_proc)
3678 pmap_activate(curthread);
3679 vmspace_free(oldvmspace);
3686 * Finds the VM object, offset, and
3687 * protection for a given virtual address in the
3688 * specified map, assuming a page fault of the
3691 * Leaves the map in question locked for read; return
3692 * values are guaranteed until a vm_map_lookup_done
3693 * call is performed. Note that the map argument
3694 * is in/out; the returned map must be used in
3695 * the call to vm_map_lookup_done.
3697 * A handle (out_entry) is returned for use in
3698 * vm_map_lookup_done, to make that fast.
3700 * If a lookup is requested with "write protection"
3701 * specified, the map may be changed to perform virtual
3702 * copying operations, although the data referenced will
3706 vm_map_lookup(vm_map_t *var_map, /* IN/OUT */
3708 vm_prot_t fault_typea,
3709 vm_map_entry_t *out_entry, /* OUT */
3710 vm_object_t *object, /* OUT */
3711 vm_pindex_t *pindex, /* OUT */
3712 vm_prot_t *out_prot, /* OUT */
3713 boolean_t *wired) /* OUT */
3715 vm_map_entry_t entry;
3716 vm_map_t map = *var_map;
3718 vm_prot_t fault_type = fault_typea;
3719 vm_object_t eobject;
3725 vm_map_lock_read(map);
3728 * Lookup the faulting address.
3730 if (!vm_map_lookup_entry(map, vaddr, out_entry)) {
3731 vm_map_unlock_read(map);
3732 return (KERN_INVALID_ADDRESS);
3740 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
3741 vm_map_t old_map = map;
3743 *var_map = map = entry->object.sub_map;
3744 vm_map_unlock_read(old_map);
3749 * Check whether this task is allowed to have this page.
3751 prot = entry->protection;
3752 fault_type &= (VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE);
3753 if ((fault_type & prot) != fault_type || prot == VM_PROT_NONE) {
3754 vm_map_unlock_read(map);
3755 return (KERN_PROTECTION_FAILURE);
3757 if ((entry->eflags & MAP_ENTRY_USER_WIRED) &&
3758 (entry->eflags & MAP_ENTRY_COW) &&
3759 (fault_type & VM_PROT_WRITE)) {
3760 vm_map_unlock_read(map);
3761 return (KERN_PROTECTION_FAILURE);
3765 * If this page is not pageable, we have to get it for all possible
3768 *wired = (entry->wired_count != 0);
3770 fault_type = entry->protection;
3771 size = entry->end - entry->start;
3773 * If the entry was copy-on-write, we either ...
3775 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
3777 * If we want to write the page, we may as well handle that
3778 * now since we've got the map locked.
3780 * If we don't need to write the page, we just demote the
3781 * permissions allowed.
3783 if ((fault_type & VM_PROT_WRITE) != 0 ||
3784 (fault_typea & VM_PROT_COPY) != 0) {
3786 * Make a new object, and place it in the object
3787 * chain. Note that no new references have appeared
3788 * -- one just moved from the map to the new
3791 if (vm_map_lock_upgrade(map))
3794 if (entry->cred == NULL) {
3796 * The debugger owner is charged for
3799 cred = curthread->td_ucred;
3801 if (!swap_reserve_by_cred(size, cred)) {
3804 return (KERN_RESOURCE_SHORTAGE);
3808 vm_object_shadow(&entry->object.vm_object,
3809 &entry->offset, size);
3810 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
3811 eobject = entry->object.vm_object;
3812 if (eobject->cred != NULL) {
3814 * The object was not shadowed.
3816 swap_release_by_cred(size, entry->cred);
3817 crfree(entry->cred);
3819 } else if (entry->cred != NULL) {
3820 VM_OBJECT_LOCK(eobject);
3821 eobject->cred = entry->cred;
3822 eobject->charge = size;
3823 VM_OBJECT_UNLOCK(eobject);
3827 vm_map_lock_downgrade(map);
3830 * We're attempting to read a copy-on-write page --
3831 * don't allow writes.
3833 prot &= ~VM_PROT_WRITE;
3838 * Create an object if necessary.
3840 if (entry->object.vm_object == NULL &&
3842 if (vm_map_lock_upgrade(map))
3844 entry->object.vm_object = vm_object_allocate(OBJT_DEFAULT,
3847 if (entry->cred != NULL) {
3848 VM_OBJECT_LOCK(entry->object.vm_object);
3849 entry->object.vm_object->cred = entry->cred;
3850 entry->object.vm_object->charge = size;
3851 VM_OBJECT_UNLOCK(entry->object.vm_object);
3854 vm_map_lock_downgrade(map);
3858 * Return the object/offset from this entry. If the entry was
3859 * copy-on-write or empty, it has been fixed up.
3861 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
3862 *object = entry->object.vm_object;
3865 return (KERN_SUCCESS);
3869 * vm_map_lookup_locked:
3871 * Lookup the faulting address. A version of vm_map_lookup that returns
3872 * KERN_FAILURE instead of blocking on map lock or memory allocation.
3875 vm_map_lookup_locked(vm_map_t *var_map, /* IN/OUT */
3877 vm_prot_t fault_typea,
3878 vm_map_entry_t *out_entry, /* OUT */
3879 vm_object_t *object, /* OUT */
3880 vm_pindex_t *pindex, /* OUT */
3881 vm_prot_t *out_prot, /* OUT */
3882 boolean_t *wired) /* OUT */
3884 vm_map_entry_t entry;
3885 vm_map_t map = *var_map;
3887 vm_prot_t fault_type = fault_typea;
3890 * Lookup the faulting address.
3892 if (!vm_map_lookup_entry(map, vaddr, out_entry))
3893 return (KERN_INVALID_ADDRESS);
3898 * Fail if the entry refers to a submap.
3900 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP)
3901 return (KERN_FAILURE);
3904 * Check whether this task is allowed to have this page.
3906 prot = entry->protection;
3907 fault_type &= VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE;
3908 if ((fault_type & prot) != fault_type)
3909 return (KERN_PROTECTION_FAILURE);
3910 if ((entry->eflags & MAP_ENTRY_USER_WIRED) &&
3911 (entry->eflags & MAP_ENTRY_COW) &&
3912 (fault_type & VM_PROT_WRITE))
3913 return (KERN_PROTECTION_FAILURE);
3916 * If this page is not pageable, we have to get it for all possible
3919 *wired = (entry->wired_count != 0);
3921 fault_type = entry->protection;
3923 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
3925 * Fail if the entry was copy-on-write for a write fault.
3927 if (fault_type & VM_PROT_WRITE)
3928 return (KERN_FAILURE);
3930 * We're attempting to read a copy-on-write page --
3931 * don't allow writes.
3933 prot &= ~VM_PROT_WRITE;
3937 * Fail if an object should be created.
3939 if (entry->object.vm_object == NULL && !map->system_map)
3940 return (KERN_FAILURE);
3943 * Return the object/offset from this entry. If the entry was
3944 * copy-on-write or empty, it has been fixed up.
3946 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
3947 *object = entry->object.vm_object;
3950 return (KERN_SUCCESS);
3954 * vm_map_lookup_done:
3956 * Releases locks acquired by a vm_map_lookup
3957 * (according to the handle returned by that lookup).
3960 vm_map_lookup_done(vm_map_t map, vm_map_entry_t entry)
3963 * Unlock the main-level map
3965 vm_map_unlock_read(map);
3968 #include "opt_ddb.h"
3970 #include <sys/kernel.h>
3972 #include <ddb/ddb.h>
3975 * vm_map_print: [ debug ]
3977 DB_SHOW_COMMAND(map, vm_map_print)
3980 /* XXX convert args. */
3981 vm_map_t map = (vm_map_t)addr;
3982 boolean_t full = have_addr;
3984 vm_map_entry_t entry;
3986 db_iprintf("Task map %p: pmap=%p, nentries=%d, version=%u\n",
3988 (void *)map->pmap, map->nentries, map->timestamp);
3991 if (!full && db_indent)
3995 for (entry = map->header.next; entry != &map->header;
3996 entry = entry->next) {
3997 db_iprintf("map entry %p: start=%p, end=%p\n",
3998 (void *)entry, (void *)entry->start, (void *)entry->end);
4001 static char *inheritance_name[4] =
4002 {"share", "copy", "none", "donate_copy"};
4004 db_iprintf(" prot=%x/%x/%s",
4006 entry->max_protection,
4007 inheritance_name[(int)(unsigned char)entry->inheritance]);
4008 if (entry->wired_count != 0)
4009 db_printf(", wired");
4011 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
4012 db_printf(", share=%p, offset=0x%jx\n",
4013 (void *)entry->object.sub_map,
4014 (uintmax_t)entry->offset);
4016 if ((entry->prev == &map->header) ||
4017 (entry->prev->object.sub_map !=
4018 entry->object.sub_map)) {
4020 vm_map_print((db_expr_t)(intptr_t)
4021 entry->object.sub_map,
4022 full, 0, (char *)0);
4026 if (entry->cred != NULL)
4027 db_printf(", ruid %d", entry->cred->cr_ruid);
4028 db_printf(", object=%p, offset=0x%jx",
4029 (void *)entry->object.vm_object,
4030 (uintmax_t)entry->offset);
4031 if (entry->object.vm_object && entry->object.vm_object->cred)
4032 db_printf(", obj ruid %d charge %jx",
4033 entry->object.vm_object->cred->cr_ruid,
4034 (uintmax_t)entry->object.vm_object->charge);
4035 if (entry->eflags & MAP_ENTRY_COW)
4036 db_printf(", copy (%s)",
4037 (entry->eflags & MAP_ENTRY_NEEDS_COPY) ? "needed" : "done");
4041 if ((entry->prev == &map->header) ||
4042 (entry->prev->object.vm_object !=
4043 entry->object.vm_object)) {
4045 vm_object_print((db_expr_t)(intptr_t)
4046 entry->object.vm_object,
4047 full, 0, (char *)0);
4059 DB_SHOW_COMMAND(procvm, procvm)
4064 p = (struct proc *) addr;
4069 db_printf("p = %p, vmspace = %p, map = %p, pmap = %p\n",
4070 (void *)p, (void *)p->p_vmspace, (void *)&p->p_vmspace->vm_map,
4071 (void *)vmspace_pmap(p->p_vmspace));
4073 vm_map_print((db_expr_t)(intptr_t)&p->p_vmspace->vm_map, 1, 0, NULL);