2 * Copyright (c) 1991, 1993
3 * The Regents of the University of California. All rights reserved.
5 * This code is derived from software contributed to Berkeley by
6 * The Mach Operating System project at Carnegie-Mellon University.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 4. Neither the name of the University nor the names of its contributors
17 * may be used to endorse or promote products derived from this software
18 * without specific prior written permission.
20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * from: @(#)vm_map.c 8.3 (Berkeley) 1/12/94
35 * Copyright (c) 1987, 1990 Carnegie-Mellon University.
36 * All rights reserved.
38 * Authors: Avadis Tevanian, Jr., Michael Wayne Young
40 * Permission to use, copy, modify and distribute this software and
41 * its documentation is hereby granted, provided that both the copyright
42 * notice and this permission notice appear in all copies of the
43 * software, derivative works or modified versions, and any portions
44 * thereof, and that both notices appear in supporting documentation.
46 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
47 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
48 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
50 * Carnegie Mellon requests users of this software to return to
52 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
53 * School of Computer Science
54 * Carnegie Mellon University
55 * Pittsburgh PA 15213-3890
57 * any improvements or extensions that they make and grant Carnegie the
58 * rights to redistribute these changes.
62 * Virtual memory mapping module.
65 #include <sys/cdefs.h>
66 __FBSDID("$FreeBSD$");
68 #include <sys/param.h>
69 #include <sys/systm.h>
70 #include <sys/kernel.h>
73 #include <sys/mutex.h>
75 #include <sys/vmmeter.h>
77 #include <sys/vnode.h>
78 #include <sys/racct.h>
79 #include <sys/resourcevar.h>
80 #include <sys/rwlock.h>
82 #include <sys/sysctl.h>
83 #include <sys/sysent.h>
87 #include <vm/vm_param.h>
89 #include <vm/vm_map.h>
90 #include <vm/vm_page.h>
91 #include <vm/vm_object.h>
92 #include <vm/vm_pager.h>
93 #include <vm/vm_kern.h>
94 #include <vm/vm_extern.h>
95 #include <vm/vnode_pager.h>
96 #include <vm/swap_pager.h>
100 * Virtual memory maps provide for the mapping, protection,
101 * and sharing of virtual memory objects. In addition,
102 * this module provides for an efficient virtual copy of
103 * memory from one map to another.
105 * Synchronization is required prior to most operations.
107 * Maps consist of an ordered doubly-linked list of simple
108 * entries; a self-adjusting binary search tree of these
109 * entries is used to speed up lookups.
111 * Since portions of maps are specified by start/end addresses,
112 * which may not align with existing map entries, all
113 * routines merely "clip" entries to these start/end values.
114 * [That is, an entry is split into two, bordering at a
115 * start or end value.] Note that these clippings may not
116 * always be necessary (as the two resulting entries are then
117 * not changed); however, the clipping is done for convenience.
119 * As mentioned above, virtual copy operations are performed
120 * by copying VM object references from one map to
121 * another, and then marking both regions as copy-on-write.
124 static struct mtx map_sleep_mtx;
125 static uma_zone_t mapentzone;
126 static uma_zone_t kmapentzone;
127 static uma_zone_t mapzone;
128 static uma_zone_t vmspace_zone;
129 static int vmspace_zinit(void *mem, int size, int flags);
130 static int vm_map_zinit(void *mem, int ize, int flags);
131 static void _vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min,
133 static void vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map);
134 static void vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry);
135 static void vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry);
137 static void vm_map_zdtor(void *mem, int size, void *arg);
138 static void vmspace_zdtor(void *mem, int size, void *arg);
140 static int vm_map_stack_locked(vm_map_t map, vm_offset_t addrbos,
141 vm_size_t max_ssize, vm_size_t growsize, vm_prot_t prot, vm_prot_t max,
143 static void vm_map_wire_entry_failure(vm_map_t map, vm_map_entry_t entry,
144 vm_offset_t failed_addr);
146 #define ENTRY_CHARGED(e) ((e)->cred != NULL || \
147 ((e)->object.vm_object != NULL && (e)->object.vm_object->cred != NULL && \
148 !((e)->eflags & MAP_ENTRY_NEEDS_COPY)))
151 * PROC_VMSPACE_{UN,}LOCK() can be a noop as long as vmspaces are type
154 #define PROC_VMSPACE_LOCK(p) do { } while (0)
155 #define PROC_VMSPACE_UNLOCK(p) do { } while (0)
158 * VM_MAP_RANGE_CHECK: [ internal use only ]
160 * Asserts that the starting and ending region
161 * addresses fall within the valid range of the map.
163 #define VM_MAP_RANGE_CHECK(map, start, end) \
165 if (start < vm_map_min(map)) \
166 start = vm_map_min(map); \
167 if (end > vm_map_max(map)) \
168 end = vm_map_max(map); \
176 * Initialize the vm_map module. Must be called before
177 * any other vm_map routines.
179 * Map and entry structures are allocated from the general
180 * purpose memory pool with some exceptions:
182 * - The kernel map and kmem submap are allocated statically.
183 * - Kernel map entries are allocated out of a static pool.
185 * These restrictions are necessary since malloc() uses the
186 * maps and requires map entries.
192 mtx_init(&map_sleep_mtx, "vm map sleep mutex", NULL, MTX_DEF);
193 mapzone = uma_zcreate("MAP", sizeof(struct vm_map), NULL,
199 vm_map_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
200 uma_prealloc(mapzone, MAX_KMAP);
201 kmapentzone = uma_zcreate("KMAP ENTRY", sizeof(struct vm_map_entry),
202 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR,
203 UMA_ZONE_MTXCLASS | UMA_ZONE_VM);
204 mapentzone = uma_zcreate("MAP ENTRY", sizeof(struct vm_map_entry),
205 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
206 vmspace_zone = uma_zcreate("VMSPACE", sizeof(struct vmspace), NULL,
212 vmspace_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
216 vmspace_zinit(void *mem, int size, int flags)
220 vm = (struct vmspace *)mem;
222 vm->vm_map.pmap = NULL;
223 (void)vm_map_zinit(&vm->vm_map, sizeof(vm->vm_map), flags);
224 PMAP_LOCK_INIT(vmspace_pmap(vm));
229 vm_map_zinit(void *mem, int size, int flags)
234 memset(map, 0, sizeof(*map));
235 mtx_init(&map->system_mtx, "vm map (system)", NULL, MTX_DEF | MTX_DUPOK);
236 sx_init(&map->lock, "vm map (user)");
242 vmspace_zdtor(void *mem, int size, void *arg)
246 vm = (struct vmspace *)mem;
248 vm_map_zdtor(&vm->vm_map, sizeof(vm->vm_map), arg);
251 vm_map_zdtor(void *mem, int size, void *arg)
256 KASSERT(map->nentries == 0,
257 ("map %p nentries == %d on free.",
258 map, map->nentries));
259 KASSERT(map->size == 0,
260 ("map %p size == %lu on free.",
261 map, (unsigned long)map->size));
263 #endif /* INVARIANTS */
266 * Allocate a vmspace structure, including a vm_map and pmap,
267 * and initialize those structures. The refcnt is set to 1.
269 * If 'pinit' is NULL then the embedded pmap is initialized via pmap_pinit().
272 vmspace_alloc(vm_offset_t min, vm_offset_t max, pmap_pinit_t pinit)
276 vm = uma_zalloc(vmspace_zone, M_WAITOK);
278 KASSERT(vm->vm_map.pmap == NULL, ("vm_map.pmap must be NULL"));
283 if (!pinit(vmspace_pmap(vm))) {
284 uma_zfree(vmspace_zone, vm);
287 CTR1(KTR_VM, "vmspace_alloc: %p", vm);
288 _vm_map_init(&vm->vm_map, vmspace_pmap(vm), min, max);
303 vmspace_container_reset(struct proc *p)
307 racct_set(p, RACCT_DATA, 0);
308 racct_set(p, RACCT_STACK, 0);
309 racct_set(p, RACCT_RSS, 0);
310 racct_set(p, RACCT_MEMLOCK, 0);
311 racct_set(p, RACCT_VMEM, 0);
317 vmspace_dofree(struct vmspace *vm)
320 CTR1(KTR_VM, "vmspace_free: %p", vm);
323 * Make sure any SysV shm is freed, it might not have been in
329 * Lock the map, to wait out all other references to it.
330 * Delete all of the mappings and pages they hold, then call
331 * the pmap module to reclaim anything left.
333 (void)vm_map_remove(&vm->vm_map, vm->vm_map.min_offset,
334 vm->vm_map.max_offset);
336 pmap_release(vmspace_pmap(vm));
337 vm->vm_map.pmap = NULL;
338 uma_zfree(vmspace_zone, vm);
342 vmspace_free(struct vmspace *vm)
345 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
346 "vmspace_free() called with non-sleepable lock held");
348 if (vm->vm_refcnt == 0)
349 panic("vmspace_free: attempt to free already freed vmspace");
351 if (atomic_fetchadd_int(&vm->vm_refcnt, -1) == 1)
356 vmspace_exitfree(struct proc *p)
360 PROC_VMSPACE_LOCK(p);
363 PROC_VMSPACE_UNLOCK(p);
364 KASSERT(vm == &vmspace0, ("vmspace_exitfree: wrong vmspace"));
369 vmspace_exit(struct thread *td)
376 * Release user portion of address space.
377 * This releases references to vnodes,
378 * which could cause I/O if the file has been unlinked.
379 * Need to do this early enough that we can still sleep.
381 * The last exiting process to reach this point releases as
382 * much of the environment as it can. vmspace_dofree() is the
383 * slower fallback in case another process had a temporary
384 * reference to the vmspace.
389 atomic_add_int(&vmspace0.vm_refcnt, 1);
391 refcnt = vm->vm_refcnt;
392 if (refcnt > 1 && p->p_vmspace != &vmspace0) {
393 /* Switch now since other proc might free vmspace */
394 PROC_VMSPACE_LOCK(p);
395 p->p_vmspace = &vmspace0;
396 PROC_VMSPACE_UNLOCK(p);
399 } while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt - 1));
401 if (p->p_vmspace != vm) {
402 /* vmspace not yet freed, switch back */
403 PROC_VMSPACE_LOCK(p);
405 PROC_VMSPACE_UNLOCK(p);
408 pmap_remove_pages(vmspace_pmap(vm));
409 /* Switch now since this proc will free vmspace */
410 PROC_VMSPACE_LOCK(p);
411 p->p_vmspace = &vmspace0;
412 PROC_VMSPACE_UNLOCK(p);
418 vmspace_container_reset(p);
422 /* Acquire reference to vmspace owned by another process. */
425 vmspace_acquire_ref(struct proc *p)
430 PROC_VMSPACE_LOCK(p);
433 PROC_VMSPACE_UNLOCK(p);
437 refcnt = vm->vm_refcnt;
438 if (refcnt <= 0) { /* Avoid 0->1 transition */
439 PROC_VMSPACE_UNLOCK(p);
442 } while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt + 1));
443 if (vm != p->p_vmspace) {
444 PROC_VMSPACE_UNLOCK(p);
448 PROC_VMSPACE_UNLOCK(p);
453 _vm_map_lock(vm_map_t map, const char *file, int line)
457 mtx_lock_flags_(&map->system_mtx, 0, file, line);
459 sx_xlock_(&map->lock, file, line);
464 vm_map_process_deferred(void)
467 vm_map_entry_t entry, next;
471 entry = td->td_map_def_user;
472 td->td_map_def_user = NULL;
473 while (entry != NULL) {
475 if ((entry->eflags & MAP_ENTRY_VN_WRITECNT) != 0) {
477 * Decrement the object's writemappings and
478 * possibly the vnode's v_writecount.
480 KASSERT((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0,
481 ("Submap with writecount"));
482 object = entry->object.vm_object;
483 KASSERT(object != NULL, ("No object for writecount"));
484 vnode_pager_release_writecount(object, entry->start,
487 vm_map_entry_deallocate(entry, FALSE);
493 _vm_map_unlock(vm_map_t map, const char *file, int line)
497 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
499 sx_xunlock_(&map->lock, file, line);
500 vm_map_process_deferred();
505 _vm_map_lock_read(vm_map_t map, const char *file, int line)
509 mtx_lock_flags_(&map->system_mtx, 0, file, line);
511 sx_slock_(&map->lock, file, line);
515 _vm_map_unlock_read(vm_map_t map, const char *file, int line)
519 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
521 sx_sunlock_(&map->lock, file, line);
522 vm_map_process_deferred();
527 _vm_map_trylock(vm_map_t map, const char *file, int line)
531 error = map->system_map ?
532 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) :
533 !sx_try_xlock_(&map->lock, file, line);
540 _vm_map_trylock_read(vm_map_t map, const char *file, int line)
544 error = map->system_map ?
545 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) :
546 !sx_try_slock_(&map->lock, file, line);
551 * _vm_map_lock_upgrade: [ internal use only ]
553 * Tries to upgrade a read (shared) lock on the specified map to a write
554 * (exclusive) lock. Returns the value "0" if the upgrade succeeds and a
555 * non-zero value if the upgrade fails. If the upgrade fails, the map is
556 * returned without a read or write lock held.
558 * Requires that the map be read locked.
561 _vm_map_lock_upgrade(vm_map_t map, const char *file, int line)
563 unsigned int last_timestamp;
565 if (map->system_map) {
566 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
568 if (!sx_try_upgrade_(&map->lock, file, line)) {
569 last_timestamp = map->timestamp;
570 sx_sunlock_(&map->lock, file, line);
571 vm_map_process_deferred();
573 * If the map's timestamp does not change while the
574 * map is unlocked, then the upgrade succeeds.
576 sx_xlock_(&map->lock, file, line);
577 if (last_timestamp != map->timestamp) {
578 sx_xunlock_(&map->lock, file, line);
588 _vm_map_lock_downgrade(vm_map_t map, const char *file, int line)
591 if (map->system_map) {
592 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
594 sx_downgrade_(&map->lock, file, line);
600 * Returns a non-zero value if the caller holds a write (exclusive) lock
601 * on the specified map and the value "0" otherwise.
604 vm_map_locked(vm_map_t map)
608 return (mtx_owned(&map->system_mtx));
610 return (sx_xlocked(&map->lock));
615 _vm_map_assert_locked(vm_map_t map, const char *file, int line)
619 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
621 sx_assert_(&map->lock, SA_XLOCKED, file, line);
624 #define VM_MAP_ASSERT_LOCKED(map) \
625 _vm_map_assert_locked(map, LOCK_FILE, LOCK_LINE)
627 #define VM_MAP_ASSERT_LOCKED(map)
631 * _vm_map_unlock_and_wait:
633 * Atomically releases the lock on the specified map and puts the calling
634 * thread to sleep. The calling thread will remain asleep until either
635 * vm_map_wakeup() is performed on the map or the specified timeout is
638 * WARNING! This function does not perform deferred deallocations of
639 * objects and map entries. Therefore, the calling thread is expected to
640 * reacquire the map lock after reawakening and later perform an ordinary
641 * unlock operation, such as vm_map_unlock(), before completing its
642 * operation on the map.
645 _vm_map_unlock_and_wait(vm_map_t map, int timo, const char *file, int line)
648 mtx_lock(&map_sleep_mtx);
650 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
652 sx_xunlock_(&map->lock, file, line);
653 return (msleep(&map->root, &map_sleep_mtx, PDROP | PVM, "vmmaps",
660 * Awaken any threads that have slept on the map using
661 * vm_map_unlock_and_wait().
664 vm_map_wakeup(vm_map_t map)
668 * Acquire and release map_sleep_mtx to prevent a wakeup()
669 * from being performed (and lost) between the map unlock
670 * and the msleep() in _vm_map_unlock_and_wait().
672 mtx_lock(&map_sleep_mtx);
673 mtx_unlock(&map_sleep_mtx);
678 vm_map_busy(vm_map_t map)
681 VM_MAP_ASSERT_LOCKED(map);
686 vm_map_unbusy(vm_map_t map)
689 VM_MAP_ASSERT_LOCKED(map);
690 KASSERT(map->busy, ("vm_map_unbusy: not busy"));
691 if (--map->busy == 0 && (map->flags & MAP_BUSY_WAKEUP)) {
692 vm_map_modflags(map, 0, MAP_BUSY_WAKEUP);
698 vm_map_wait_busy(vm_map_t map)
701 VM_MAP_ASSERT_LOCKED(map);
703 vm_map_modflags(map, MAP_BUSY_WAKEUP, 0);
705 msleep(&map->busy, &map->system_mtx, 0, "mbusy", 0);
707 sx_sleep(&map->busy, &map->lock, 0, "mbusy", 0);
713 vmspace_resident_count(struct vmspace *vmspace)
715 return pmap_resident_count(vmspace_pmap(vmspace));
721 * Creates and returns a new empty VM map with
722 * the given physical map structure, and having
723 * the given lower and upper address bounds.
726 vm_map_create(pmap_t pmap, vm_offset_t min, vm_offset_t max)
730 result = uma_zalloc(mapzone, M_WAITOK);
731 CTR1(KTR_VM, "vm_map_create: %p", result);
732 _vm_map_init(result, pmap, min, max);
737 * Initialize an existing vm_map structure
738 * such as that in the vmspace structure.
741 _vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max)
744 map->header.next = map->header.prev = &map->header;
745 map->needs_wakeup = FALSE;
748 map->min_offset = min;
749 map->max_offset = max;
757 vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max)
760 _vm_map_init(map, pmap, min, max);
761 mtx_init(&map->system_mtx, "system map", NULL, MTX_DEF | MTX_DUPOK);
762 sx_init(&map->lock, "user map");
766 * vm_map_entry_dispose: [ internal use only ]
768 * Inverse of vm_map_entry_create.
771 vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry)
773 uma_zfree(map->system_map ? kmapentzone : mapentzone, entry);
777 * vm_map_entry_create: [ internal use only ]
779 * Allocates a VM map entry for insertion.
780 * No entry fields are filled in.
782 static vm_map_entry_t
783 vm_map_entry_create(vm_map_t map)
785 vm_map_entry_t new_entry;
788 new_entry = uma_zalloc(kmapentzone, M_NOWAIT);
790 new_entry = uma_zalloc(mapentzone, M_WAITOK);
791 if (new_entry == NULL)
792 panic("vm_map_entry_create: kernel resources exhausted");
797 * vm_map_entry_set_behavior:
799 * Set the expected access behavior, either normal, random, or
803 vm_map_entry_set_behavior(vm_map_entry_t entry, u_char behavior)
805 entry->eflags = (entry->eflags & ~MAP_ENTRY_BEHAV_MASK) |
806 (behavior & MAP_ENTRY_BEHAV_MASK);
810 * vm_map_entry_set_max_free:
812 * Set the max_free field in a vm_map_entry.
815 vm_map_entry_set_max_free(vm_map_entry_t entry)
818 entry->max_free = entry->adj_free;
819 if (entry->left != NULL && entry->left->max_free > entry->max_free)
820 entry->max_free = entry->left->max_free;
821 if (entry->right != NULL && entry->right->max_free > entry->max_free)
822 entry->max_free = entry->right->max_free;
826 * vm_map_entry_splay:
828 * The Sleator and Tarjan top-down splay algorithm with the
829 * following variation. Max_free must be computed bottom-up, so
830 * on the downward pass, maintain the left and right spines in
831 * reverse order. Then, make a second pass up each side to fix
832 * the pointers and compute max_free. The time bound is O(log n)
835 * The new root is the vm_map_entry containing "addr", or else an
836 * adjacent entry (lower or higher) if addr is not in the tree.
838 * The map must be locked, and leaves it so.
840 * Returns: the new root.
842 static vm_map_entry_t
843 vm_map_entry_splay(vm_offset_t addr, vm_map_entry_t root)
845 vm_map_entry_t llist, rlist;
846 vm_map_entry_t ltree, rtree;
849 /* Special case of empty tree. */
854 * Pass One: Splay down the tree until we find addr or a NULL
855 * pointer where addr would go. llist and rlist are the two
856 * sides in reverse order (bottom-up), with llist linked by
857 * the right pointer and rlist linked by the left pointer in
858 * the vm_map_entry. Wait until Pass Two to set max_free on
864 /* root is never NULL in here. */
865 if (addr < root->start) {
869 if (addr < y->start && y->left != NULL) {
870 /* Rotate right and put y on rlist. */
871 root->left = y->right;
873 vm_map_entry_set_max_free(root);
878 /* Put root on rlist. */
883 } else if (addr >= root->end) {
887 if (addr >= y->end && y->right != NULL) {
888 /* Rotate left and put y on llist. */
889 root->right = y->left;
891 vm_map_entry_set_max_free(root);
896 /* Put root on llist. */
906 * Pass Two: Walk back up the two spines, flip the pointers
907 * and set max_free. The subtrees of the root go at the
908 * bottom of llist and rlist.
911 while (llist != NULL) {
913 llist->right = ltree;
914 vm_map_entry_set_max_free(llist);
919 while (rlist != NULL) {
922 vm_map_entry_set_max_free(rlist);
928 * Final assembly: add ltree and rtree as subtrees of root.
932 vm_map_entry_set_max_free(root);
938 * vm_map_entry_{un,}link:
940 * Insert/remove entries from maps.
943 vm_map_entry_link(vm_map_t map,
944 vm_map_entry_t after_where,
945 vm_map_entry_t entry)
949 "vm_map_entry_link: map %p, nentries %d, entry %p, after %p", map,
950 map->nentries, entry, after_where);
951 VM_MAP_ASSERT_LOCKED(map);
952 KASSERT(after_where == &map->header ||
953 after_where->end <= entry->start,
954 ("vm_map_entry_link: prev end %jx new start %jx overlap",
955 (uintmax_t)after_where->end, (uintmax_t)entry->start));
956 KASSERT(after_where->next == &map->header ||
957 entry->end <= after_where->next->start,
958 ("vm_map_entry_link: new end %jx next start %jx overlap",
959 (uintmax_t)entry->end, (uintmax_t)after_where->next->start));
962 entry->prev = after_where;
963 entry->next = after_where->next;
964 entry->next->prev = entry;
965 after_where->next = entry;
967 if (after_where != &map->header) {
968 if (after_where != map->root)
969 vm_map_entry_splay(after_where->start, map->root);
970 entry->right = after_where->right;
971 entry->left = after_where;
972 after_where->right = NULL;
973 after_where->adj_free = entry->start - after_where->end;
974 vm_map_entry_set_max_free(after_where);
976 entry->right = map->root;
979 entry->adj_free = (entry->next == &map->header ? map->max_offset :
980 entry->next->start) - entry->end;
981 vm_map_entry_set_max_free(entry);
986 vm_map_entry_unlink(vm_map_t map,
987 vm_map_entry_t entry)
989 vm_map_entry_t next, prev, root;
991 VM_MAP_ASSERT_LOCKED(map);
992 if (entry != map->root)
993 vm_map_entry_splay(entry->start, map->root);
994 if (entry->left == NULL)
997 root = vm_map_entry_splay(entry->start, entry->left);
998 root->right = entry->right;
999 root->adj_free = (entry->next == &map->header ? map->max_offset :
1000 entry->next->start) - root->end;
1001 vm_map_entry_set_max_free(root);
1010 CTR3(KTR_VM, "vm_map_entry_unlink: map %p, nentries %d, entry %p", map,
1011 map->nentries, entry);
1015 * vm_map_entry_resize_free:
1017 * Recompute the amount of free space following a vm_map_entry
1018 * and propagate that value up the tree. Call this function after
1019 * resizing a map entry in-place, that is, without a call to
1020 * vm_map_entry_link() or _unlink().
1022 * The map must be locked, and leaves it so.
1025 vm_map_entry_resize_free(vm_map_t map, vm_map_entry_t entry)
1029 * Using splay trees without parent pointers, propagating
1030 * max_free up the tree is done by moving the entry to the
1031 * root and making the change there.
1033 if (entry != map->root)
1034 map->root = vm_map_entry_splay(entry->start, map->root);
1036 entry->adj_free = (entry->next == &map->header ? map->max_offset :
1037 entry->next->start) - entry->end;
1038 vm_map_entry_set_max_free(entry);
1042 * vm_map_lookup_entry: [ internal use only ]
1044 * Finds the map entry containing (or
1045 * immediately preceding) the specified address
1046 * in the given map; the entry is returned
1047 * in the "entry" parameter. The boolean
1048 * result indicates whether the address is
1049 * actually contained in the map.
1052 vm_map_lookup_entry(
1054 vm_offset_t address,
1055 vm_map_entry_t *entry) /* OUT */
1061 * If the map is empty, then the map entry immediately preceding
1062 * "address" is the map's header.
1066 *entry = &map->header;
1067 else if (address >= cur->start && cur->end > address) {
1070 } else if ((locked = vm_map_locked(map)) ||
1071 sx_try_upgrade(&map->lock)) {
1073 * Splay requires a write lock on the map. However, it only
1074 * restructures the binary search tree; it does not otherwise
1075 * change the map. Thus, the map's timestamp need not change
1076 * on a temporary upgrade.
1078 map->root = cur = vm_map_entry_splay(address, cur);
1080 sx_downgrade(&map->lock);
1083 * If "address" is contained within a map entry, the new root
1084 * is that map entry. Otherwise, the new root is a map entry
1085 * immediately before or after "address".
1087 if (address >= cur->start) {
1089 if (cur->end > address)
1095 * Since the map is only locked for read access, perform a
1096 * standard binary search tree lookup for "address".
1099 if (address < cur->start) {
1100 if (cur->left == NULL) {
1105 } else if (cur->end > address) {
1109 if (cur->right == NULL) {
1122 * Inserts the given whole VM object into the target
1123 * map at the specified address range. The object's
1124 * size should match that of the address range.
1126 * Requires that the map be locked, and leaves it so.
1128 * If object is non-NULL, ref count must be bumped by caller
1129 * prior to making call to account for the new entry.
1132 vm_map_insert(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1133 vm_offset_t start, vm_offset_t end, vm_prot_t prot, vm_prot_t max, int cow)
1135 vm_map_entry_t new_entry, prev_entry, temp_entry;
1137 vm_eflags_t protoeflags;
1138 vm_inherit_t inheritance;
1140 VM_MAP_ASSERT_LOCKED(map);
1141 KASSERT((object != kmem_object && object != kernel_object) ||
1142 (cow & MAP_COPY_ON_WRITE) == 0,
1143 ("vm_map_insert: kmem or kernel object and COW"));
1144 KASSERT(object == NULL || (cow & MAP_NOFAULT) == 0,
1145 ("vm_map_insert: paradoxical MAP_NOFAULT request"));
1146 KASSERT((prot & ~max) == 0,
1147 ("prot %#x is not subset of max_prot %#x", prot, max));
1150 * Check that the start and end points are not bogus.
1152 if (start < map->min_offset || end > map->max_offset || start >= end)
1153 return (KERN_INVALID_ADDRESS);
1156 * Find the entry prior to the proposed starting address; if it's part
1157 * of an existing entry, this range is bogus.
1159 if (vm_map_lookup_entry(map, start, &temp_entry))
1160 return (KERN_NO_SPACE);
1162 prev_entry = temp_entry;
1165 * Assert that the next entry doesn't overlap the end point.
1167 if (prev_entry->next != &map->header && prev_entry->next->start < end)
1168 return (KERN_NO_SPACE);
1171 if (cow & MAP_COPY_ON_WRITE)
1172 protoeflags |= MAP_ENTRY_COW | MAP_ENTRY_NEEDS_COPY;
1173 if (cow & MAP_NOFAULT)
1174 protoeflags |= MAP_ENTRY_NOFAULT;
1175 if (cow & MAP_DISABLE_SYNCER)
1176 protoeflags |= MAP_ENTRY_NOSYNC;
1177 if (cow & MAP_DISABLE_COREDUMP)
1178 protoeflags |= MAP_ENTRY_NOCOREDUMP;
1179 if (cow & MAP_STACK_GROWS_DOWN)
1180 protoeflags |= MAP_ENTRY_GROWS_DOWN;
1181 if (cow & MAP_STACK_GROWS_UP)
1182 protoeflags |= MAP_ENTRY_GROWS_UP;
1183 if (cow & MAP_VN_WRITECOUNT)
1184 protoeflags |= MAP_ENTRY_VN_WRITECNT;
1185 if (cow & MAP_INHERIT_SHARE)
1186 inheritance = VM_INHERIT_SHARE;
1188 inheritance = VM_INHERIT_DEFAULT;
1191 if (cow & (MAP_ACC_NO_CHARGE | MAP_NOFAULT))
1193 if ((cow & MAP_ACC_CHARGED) || ((prot & VM_PROT_WRITE) &&
1194 ((protoeflags & MAP_ENTRY_NEEDS_COPY) || object == NULL))) {
1195 if (!(cow & MAP_ACC_CHARGED) && !swap_reserve(end - start))
1196 return (KERN_RESOURCE_SHORTAGE);
1197 KASSERT(object == NULL ||
1198 (protoeflags & MAP_ENTRY_NEEDS_COPY) != 0 ||
1199 object->cred == NULL,
1200 ("overcommit: vm_map_insert o %p", object));
1201 cred = curthread->td_ucred;
1205 /* Expand the kernel pmap, if necessary. */
1206 if (map == kernel_map && end > kernel_vm_end)
1207 pmap_growkernel(end);
1208 if (object != NULL) {
1210 * OBJ_ONEMAPPING must be cleared unless this mapping
1211 * is trivially proven to be the only mapping for any
1212 * of the object's pages. (Object granularity
1213 * reference counting is insufficient to recognize
1214 * aliases with precision.)
1216 VM_OBJECT_WLOCK(object);
1217 if (object->ref_count > 1 || object->shadow_count != 0)
1218 vm_object_clear_flag(object, OBJ_ONEMAPPING);
1219 VM_OBJECT_WUNLOCK(object);
1220 } else if (prev_entry != &map->header &&
1221 prev_entry->eflags == protoeflags &&
1222 (cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 &&
1223 prev_entry->end == start && prev_entry->wired_count == 0 &&
1224 (prev_entry->cred == cred ||
1225 (prev_entry->object.vm_object != NULL &&
1226 prev_entry->object.vm_object->cred == cred)) &&
1227 vm_object_coalesce(prev_entry->object.vm_object,
1229 (vm_size_t)(prev_entry->end - prev_entry->start),
1230 (vm_size_t)(end - prev_entry->end), cred != NULL &&
1231 (protoeflags & MAP_ENTRY_NEEDS_COPY) == 0)) {
1233 * We were able to extend the object. Determine if we
1234 * can extend the previous map entry to include the
1235 * new range as well.
1237 if (prev_entry->inheritance == inheritance &&
1238 prev_entry->protection == prot &&
1239 prev_entry->max_protection == max) {
1240 map->size += end - prev_entry->end;
1241 prev_entry->end = end;
1242 vm_map_entry_resize_free(map, prev_entry);
1243 vm_map_simplify_entry(map, prev_entry);
1244 return (KERN_SUCCESS);
1248 * If we can extend the object but cannot extend the
1249 * map entry, we have to create a new map entry. We
1250 * must bump the ref count on the extended object to
1251 * account for it. object may be NULL.
1253 object = prev_entry->object.vm_object;
1254 offset = prev_entry->offset +
1255 (prev_entry->end - prev_entry->start);
1256 vm_object_reference(object);
1257 if (cred != NULL && object != NULL && object->cred != NULL &&
1258 !(prev_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
1259 /* Object already accounts for this uid. */
1267 * Create a new entry
1269 new_entry = vm_map_entry_create(map);
1270 new_entry->start = start;
1271 new_entry->end = end;
1272 new_entry->cred = NULL;
1274 new_entry->eflags = protoeflags;
1275 new_entry->object.vm_object = object;
1276 new_entry->offset = offset;
1277 new_entry->avail_ssize = 0;
1279 new_entry->inheritance = inheritance;
1280 new_entry->protection = prot;
1281 new_entry->max_protection = max;
1282 new_entry->wired_count = 0;
1283 new_entry->wiring_thread = NULL;
1284 new_entry->read_ahead = VM_FAULT_READ_AHEAD_INIT;
1285 new_entry->next_read = OFF_TO_IDX(offset);
1287 KASSERT(cred == NULL || !ENTRY_CHARGED(new_entry),
1288 ("overcommit: vm_map_insert leaks vm_map %p", new_entry));
1289 new_entry->cred = cred;
1292 * Insert the new entry into the list
1294 vm_map_entry_link(map, prev_entry, new_entry);
1295 map->size += new_entry->end - new_entry->start;
1298 * Try to coalesce the new entry with both the previous and next
1299 * entries in the list. Previously, we only attempted to coalesce
1300 * with the previous entry when object is NULL. Here, we handle the
1301 * other cases, which are less common.
1303 vm_map_simplify_entry(map, new_entry);
1305 if ((cow & (MAP_PREFAULT | MAP_PREFAULT_PARTIAL)) != 0) {
1306 vm_map_pmap_enter(map, start, prot, object, OFF_TO_IDX(offset),
1307 end - start, cow & MAP_PREFAULT_PARTIAL);
1310 return (KERN_SUCCESS);
1316 * Find the first fit (lowest VM address) for "length" free bytes
1317 * beginning at address >= start in the given map.
1319 * In a vm_map_entry, "adj_free" is the amount of free space
1320 * adjacent (higher address) to this entry, and "max_free" is the
1321 * maximum amount of contiguous free space in its subtree. This
1322 * allows finding a free region in one path down the tree, so
1323 * O(log n) amortized with splay trees.
1325 * The map must be locked, and leaves it so.
1327 * Returns: 0 on success, and starting address in *addr,
1328 * 1 if insufficient space.
1331 vm_map_findspace(vm_map_t map, vm_offset_t start, vm_size_t length,
1332 vm_offset_t *addr) /* OUT */
1334 vm_map_entry_t entry;
1338 * Request must fit within min/max VM address and must avoid
1341 if (start < map->min_offset)
1342 start = map->min_offset;
1343 if (start + length > map->max_offset || start + length < start)
1346 /* Empty tree means wide open address space. */
1347 if (map->root == NULL) {
1353 * After splay, if start comes before root node, then there
1354 * must be a gap from start to the root.
1356 map->root = vm_map_entry_splay(start, map->root);
1357 if (start + length <= map->root->start) {
1363 * Root is the last node that might begin its gap before
1364 * start, and this is the last comparison where address
1365 * wrap might be a problem.
1367 st = (start > map->root->end) ? start : map->root->end;
1368 if (length <= map->root->end + map->root->adj_free - st) {
1373 /* With max_free, can immediately tell if no solution. */
1374 entry = map->root->right;
1375 if (entry == NULL || length > entry->max_free)
1379 * Search the right subtree in the order: left subtree, root,
1380 * right subtree (first fit). The previous splay implies that
1381 * all regions in the right subtree have addresses > start.
1383 while (entry != NULL) {
1384 if (entry->left != NULL && entry->left->max_free >= length)
1385 entry = entry->left;
1386 else if (entry->adj_free >= length) {
1390 entry = entry->right;
1393 /* Can't get here, so panic if we do. */
1394 panic("vm_map_findspace: max_free corrupt");
1398 vm_map_fixed(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1399 vm_offset_t start, vm_size_t length, vm_prot_t prot,
1400 vm_prot_t max, int cow)
1405 end = start + length;
1406 KASSERT((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 ||
1408 ("vm_map_fixed: non-NULL backing object for stack"));
1410 VM_MAP_RANGE_CHECK(map, start, end);
1411 if ((cow & MAP_CHECK_EXCL) == 0)
1412 vm_map_delete(map, start, end);
1413 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) != 0) {
1414 result = vm_map_stack_locked(map, start, length, sgrowsiz,
1417 result = vm_map_insert(map, object, offset, start, end,
1425 * vm_map_find finds an unallocated region in the target address
1426 * map with the given length. The search is defined to be
1427 * first-fit from the specified address; the region found is
1428 * returned in the same parameter.
1430 * If object is non-NULL, ref count must be bumped by caller
1431 * prior to making call to account for the new entry.
1434 vm_map_find(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1435 vm_offset_t *addr, /* IN/OUT */
1436 vm_size_t length, vm_offset_t max_addr, int find_space,
1437 vm_prot_t prot, vm_prot_t max, int cow)
1439 vm_offset_t alignment, initial_addr, start;
1442 KASSERT((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 ||
1444 ("vm_map_find: non-NULL backing object for stack"));
1445 if (find_space == VMFS_OPTIMAL_SPACE && (object == NULL ||
1446 (object->flags & OBJ_COLORED) == 0))
1447 find_space = VMFS_ANY_SPACE;
1448 if (find_space >> 8 != 0) {
1449 KASSERT((find_space & 0xff) == 0, ("bad VMFS flags"));
1450 alignment = (vm_offset_t)1 << (find_space >> 8);
1453 initial_addr = *addr;
1455 start = initial_addr;
1458 if (find_space != VMFS_NO_SPACE) {
1459 if (vm_map_findspace(map, start, length, addr) ||
1460 (max_addr != 0 && *addr + length > max_addr)) {
1462 if (find_space == VMFS_OPTIMAL_SPACE) {
1463 find_space = VMFS_ANY_SPACE;
1466 return (KERN_NO_SPACE);
1468 switch (find_space) {
1469 case VMFS_SUPER_SPACE:
1470 case VMFS_OPTIMAL_SPACE:
1471 pmap_align_superpage(object, offset, addr,
1474 case VMFS_ANY_SPACE:
1477 if ((*addr & (alignment - 1)) != 0) {
1478 *addr &= ~(alignment - 1);
1486 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) != 0) {
1487 result = vm_map_stack_locked(map, start, length,
1488 sgrowsiz, prot, max, cow);
1490 result = vm_map_insert(map, object, offset, start,
1491 start + length, prot, max, cow);
1493 } while (result == KERN_NO_SPACE && find_space != VMFS_NO_SPACE &&
1494 find_space != VMFS_ANY_SPACE);
1500 * vm_map_simplify_entry:
1502 * Simplify the given map entry by merging with either neighbor. This
1503 * routine also has the ability to merge with both neighbors.
1505 * The map must be locked.
1507 * This routine guarentees that the passed entry remains valid (though
1508 * possibly extended). When merging, this routine may delete one or
1512 vm_map_simplify_entry(vm_map_t map, vm_map_entry_t entry)
1514 vm_map_entry_t next, prev;
1515 vm_size_t prevsize, esize;
1517 if ((entry->eflags & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP |
1518 MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_IS_SUB_MAP)) != 0)
1522 if (prev != &map->header) {
1523 prevsize = prev->end - prev->start;
1524 if ( (prev->end == entry->start) &&
1525 (prev->object.vm_object == entry->object.vm_object) &&
1526 (!prev->object.vm_object ||
1527 (prev->offset + prevsize == entry->offset)) &&
1528 (prev->eflags == entry->eflags) &&
1529 (prev->protection == entry->protection) &&
1530 (prev->max_protection == entry->max_protection) &&
1531 (prev->inheritance == entry->inheritance) &&
1532 (prev->wired_count == entry->wired_count) &&
1533 (prev->cred == entry->cred)) {
1534 vm_map_entry_unlink(map, prev);
1535 entry->start = prev->start;
1536 entry->offset = prev->offset;
1537 if (entry->prev != &map->header)
1538 vm_map_entry_resize_free(map, entry->prev);
1541 * If the backing object is a vnode object,
1542 * vm_object_deallocate() calls vrele().
1543 * However, vrele() does not lock the vnode
1544 * because the vnode has additional
1545 * references. Thus, the map lock can be kept
1546 * without causing a lock-order reversal with
1549 * Since we count the number of virtual page
1550 * mappings in object->un_pager.vnp.writemappings,
1551 * the writemappings value should not be adjusted
1552 * when the entry is disposed of.
1554 if (prev->object.vm_object)
1555 vm_object_deallocate(prev->object.vm_object);
1556 if (prev->cred != NULL)
1558 vm_map_entry_dispose(map, prev);
1563 if (next != &map->header) {
1564 esize = entry->end - entry->start;
1565 if ((entry->end == next->start) &&
1566 (next->object.vm_object == entry->object.vm_object) &&
1567 (!entry->object.vm_object ||
1568 (entry->offset + esize == next->offset)) &&
1569 (next->eflags == entry->eflags) &&
1570 (next->protection == entry->protection) &&
1571 (next->max_protection == entry->max_protection) &&
1572 (next->inheritance == entry->inheritance) &&
1573 (next->wired_count == entry->wired_count) &&
1574 (next->cred == entry->cred)) {
1575 vm_map_entry_unlink(map, next);
1576 entry->end = next->end;
1577 vm_map_entry_resize_free(map, entry);
1580 * See comment above.
1582 if (next->object.vm_object)
1583 vm_object_deallocate(next->object.vm_object);
1584 if (next->cred != NULL)
1586 vm_map_entry_dispose(map, next);
1591 * vm_map_clip_start: [ internal use only ]
1593 * Asserts that the given entry begins at or after
1594 * the specified address; if necessary,
1595 * it splits the entry into two.
1597 #define vm_map_clip_start(map, entry, startaddr) \
1599 if (startaddr > entry->start) \
1600 _vm_map_clip_start(map, entry, startaddr); \
1604 * This routine is called only when it is known that
1605 * the entry must be split.
1608 _vm_map_clip_start(vm_map_t map, vm_map_entry_t entry, vm_offset_t start)
1610 vm_map_entry_t new_entry;
1612 VM_MAP_ASSERT_LOCKED(map);
1615 * Split off the front portion -- note that we must insert the new
1616 * entry BEFORE this one, so that this entry has the specified
1619 vm_map_simplify_entry(map, entry);
1622 * If there is no object backing this entry, we might as well create
1623 * one now. If we defer it, an object can get created after the map
1624 * is clipped, and individual objects will be created for the split-up
1625 * map. This is a bit of a hack, but is also about the best place to
1626 * put this improvement.
1628 if (entry->object.vm_object == NULL && !map->system_map) {
1630 object = vm_object_allocate(OBJT_DEFAULT,
1631 atop(entry->end - entry->start));
1632 entry->object.vm_object = object;
1634 if (entry->cred != NULL) {
1635 object->cred = entry->cred;
1636 object->charge = entry->end - entry->start;
1639 } else if (entry->object.vm_object != NULL &&
1640 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) &&
1641 entry->cred != NULL) {
1642 VM_OBJECT_WLOCK(entry->object.vm_object);
1643 KASSERT(entry->object.vm_object->cred == NULL,
1644 ("OVERCOMMIT: vm_entry_clip_start: both cred e %p", entry));
1645 entry->object.vm_object->cred = entry->cred;
1646 entry->object.vm_object->charge = entry->end - entry->start;
1647 VM_OBJECT_WUNLOCK(entry->object.vm_object);
1651 new_entry = vm_map_entry_create(map);
1652 *new_entry = *entry;
1654 new_entry->end = start;
1655 entry->offset += (start - entry->start);
1656 entry->start = start;
1657 if (new_entry->cred != NULL)
1658 crhold(entry->cred);
1660 vm_map_entry_link(map, entry->prev, new_entry);
1662 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1663 vm_object_reference(new_entry->object.vm_object);
1665 * The object->un_pager.vnp.writemappings for the
1666 * object of MAP_ENTRY_VN_WRITECNT type entry shall be
1667 * kept as is here. The virtual pages are
1668 * re-distributed among the clipped entries, so the sum is
1675 * vm_map_clip_end: [ internal use only ]
1677 * Asserts that the given entry ends at or before
1678 * the specified address; if necessary,
1679 * it splits the entry into two.
1681 #define vm_map_clip_end(map, entry, endaddr) \
1683 if ((endaddr) < (entry->end)) \
1684 _vm_map_clip_end((map), (entry), (endaddr)); \
1688 * This routine is called only when it is known that
1689 * the entry must be split.
1692 _vm_map_clip_end(vm_map_t map, vm_map_entry_t entry, vm_offset_t end)
1694 vm_map_entry_t new_entry;
1696 VM_MAP_ASSERT_LOCKED(map);
1699 * If there is no object backing this entry, we might as well create
1700 * one now. If we defer it, an object can get created after the map
1701 * is clipped, and individual objects will be created for the split-up
1702 * map. This is a bit of a hack, but is also about the best place to
1703 * put this improvement.
1705 if (entry->object.vm_object == NULL && !map->system_map) {
1707 object = vm_object_allocate(OBJT_DEFAULT,
1708 atop(entry->end - entry->start));
1709 entry->object.vm_object = object;
1711 if (entry->cred != NULL) {
1712 object->cred = entry->cred;
1713 object->charge = entry->end - entry->start;
1716 } else if (entry->object.vm_object != NULL &&
1717 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) &&
1718 entry->cred != NULL) {
1719 VM_OBJECT_WLOCK(entry->object.vm_object);
1720 KASSERT(entry->object.vm_object->cred == NULL,
1721 ("OVERCOMMIT: vm_entry_clip_end: both cred e %p", entry));
1722 entry->object.vm_object->cred = entry->cred;
1723 entry->object.vm_object->charge = entry->end - entry->start;
1724 VM_OBJECT_WUNLOCK(entry->object.vm_object);
1729 * Create a new entry and insert it AFTER the specified entry
1731 new_entry = vm_map_entry_create(map);
1732 *new_entry = *entry;
1734 new_entry->start = entry->end = end;
1735 new_entry->offset += (end - entry->start);
1736 if (new_entry->cred != NULL)
1737 crhold(entry->cred);
1739 vm_map_entry_link(map, entry, new_entry);
1741 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1742 vm_object_reference(new_entry->object.vm_object);
1747 * vm_map_submap: [ kernel use only ]
1749 * Mark the given range as handled by a subordinate map.
1751 * This range must have been created with vm_map_find,
1752 * and no other operations may have been performed on this
1753 * range prior to calling vm_map_submap.
1755 * Only a limited number of operations can be performed
1756 * within this rage after calling vm_map_submap:
1758 * [Don't try vm_map_copy!]
1760 * To remove a submapping, one must first remove the
1761 * range from the superior map, and then destroy the
1762 * submap (if desired). [Better yet, don't try it.]
1771 vm_map_entry_t entry;
1772 int result = KERN_INVALID_ARGUMENT;
1776 VM_MAP_RANGE_CHECK(map, start, end);
1778 if (vm_map_lookup_entry(map, start, &entry)) {
1779 vm_map_clip_start(map, entry, start);
1781 entry = entry->next;
1783 vm_map_clip_end(map, entry, end);
1785 if ((entry->start == start) && (entry->end == end) &&
1786 ((entry->eflags & MAP_ENTRY_COW) == 0) &&
1787 (entry->object.vm_object == NULL)) {
1788 entry->object.sub_map = submap;
1789 entry->eflags |= MAP_ENTRY_IS_SUB_MAP;
1790 result = KERN_SUCCESS;
1798 * The maximum number of pages to map if MAP_PREFAULT_PARTIAL is specified
1800 #define MAX_INIT_PT 96
1803 * vm_map_pmap_enter:
1805 * Preload the specified map's pmap with mappings to the specified
1806 * object's memory-resident pages. No further physical pages are
1807 * allocated, and no further virtual pages are retrieved from secondary
1808 * storage. If the specified flags include MAP_PREFAULT_PARTIAL, then a
1809 * limited number of page mappings are created at the low-end of the
1810 * specified address range. (For this purpose, a superpage mapping
1811 * counts as one page mapping.) Otherwise, all resident pages within
1812 * the specified address range are mapped. Because these mappings are
1813 * being created speculatively, cached pages are not reactivated and
1817 vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot,
1818 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags)
1821 vm_page_t p, p_start;
1822 vm_pindex_t mask, psize, threshold, tmpidx;
1824 if ((prot & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0 || object == NULL)
1826 VM_OBJECT_RLOCK(object);
1827 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) {
1828 VM_OBJECT_RUNLOCK(object);
1829 VM_OBJECT_WLOCK(object);
1830 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) {
1831 pmap_object_init_pt(map->pmap, addr, object, pindex,
1833 VM_OBJECT_WUNLOCK(object);
1836 VM_OBJECT_LOCK_DOWNGRADE(object);
1840 if (psize + pindex > object->size) {
1841 if (object->size < pindex) {
1842 VM_OBJECT_RUNLOCK(object);
1845 psize = object->size - pindex;
1850 threshold = MAX_INIT_PT;
1852 p = vm_page_find_least(object, pindex);
1854 * Assert: the variable p is either (1) the page with the
1855 * least pindex greater than or equal to the parameter pindex
1859 p != NULL && (tmpidx = p->pindex - pindex) < psize;
1860 p = TAILQ_NEXT(p, listq)) {
1862 * don't allow an madvise to blow away our really
1863 * free pages allocating pv entries.
1865 if (((flags & MAP_PREFAULT_MADVISE) != 0 &&
1866 cnt.v_free_count < cnt.v_free_reserved) ||
1867 ((flags & MAP_PREFAULT_PARTIAL) != 0 &&
1868 tmpidx >= threshold)) {
1872 if (p->valid == VM_PAGE_BITS_ALL) {
1873 if (p_start == NULL) {
1874 start = addr + ptoa(tmpidx);
1877 /* Jump ahead if a superpage mapping is possible. */
1878 if (p->psind > 0 && ((addr + ptoa(tmpidx)) &
1879 (pagesizes[p->psind] - 1)) == 0) {
1880 mask = atop(pagesizes[p->psind]) - 1;
1881 if (tmpidx + mask < psize &&
1882 vm_page_ps_is_valid(p)) {
1887 } else if (p_start != NULL) {
1888 pmap_enter_object(map->pmap, start, addr +
1889 ptoa(tmpidx), p_start, prot);
1893 if (p_start != NULL)
1894 pmap_enter_object(map->pmap, start, addr + ptoa(psize),
1896 VM_OBJECT_RUNLOCK(object);
1902 * Sets the protection of the specified address
1903 * region in the target map. If "set_max" is
1904 * specified, the maximum protection is to be set;
1905 * otherwise, only the current protection is affected.
1908 vm_map_protect(vm_map_t map, vm_offset_t start, vm_offset_t end,
1909 vm_prot_t new_prot, boolean_t set_max)
1911 vm_map_entry_t current, entry;
1917 return (KERN_SUCCESS);
1921 VM_MAP_RANGE_CHECK(map, start, end);
1923 if (vm_map_lookup_entry(map, start, &entry)) {
1924 vm_map_clip_start(map, entry, start);
1926 entry = entry->next;
1930 * Make a first pass to check for protection violations.
1933 while ((current != &map->header) && (current->start < end)) {
1934 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
1936 return (KERN_INVALID_ARGUMENT);
1938 if ((new_prot & current->max_protection) != new_prot) {
1940 return (KERN_PROTECTION_FAILURE);
1942 current = current->next;
1947 * Do an accounting pass for private read-only mappings that
1948 * now will do cow due to allowed write (e.g. debugger sets
1949 * breakpoint on text segment)
1951 for (current = entry; (current != &map->header) &&
1952 (current->start < end); current = current->next) {
1954 vm_map_clip_end(map, current, end);
1957 ((new_prot & ~(current->protection)) & VM_PROT_WRITE) == 0 ||
1958 ENTRY_CHARGED(current)) {
1962 cred = curthread->td_ucred;
1963 obj = current->object.vm_object;
1965 if (obj == NULL || (current->eflags & MAP_ENTRY_NEEDS_COPY)) {
1966 if (!swap_reserve(current->end - current->start)) {
1968 return (KERN_RESOURCE_SHORTAGE);
1971 current->cred = cred;
1975 VM_OBJECT_WLOCK(obj);
1976 if (obj->type != OBJT_DEFAULT && obj->type != OBJT_SWAP) {
1977 VM_OBJECT_WUNLOCK(obj);
1982 * Charge for the whole object allocation now, since
1983 * we cannot distinguish between non-charged and
1984 * charged clipped mapping of the same object later.
1986 KASSERT(obj->charge == 0,
1987 ("vm_map_protect: object %p overcharged (entry %p)",
1989 if (!swap_reserve(ptoa(obj->size))) {
1990 VM_OBJECT_WUNLOCK(obj);
1992 return (KERN_RESOURCE_SHORTAGE);
1997 obj->charge = ptoa(obj->size);
1998 VM_OBJECT_WUNLOCK(obj);
2002 * Go back and fix up protections. [Note that clipping is not
2003 * necessary the second time.]
2006 while ((current != &map->header) && (current->start < end)) {
2007 old_prot = current->protection;
2010 current->protection =
2011 (current->max_protection = new_prot) &
2014 current->protection = new_prot;
2017 * For user wired map entries, the normal lazy evaluation of
2018 * write access upgrades through soft page faults is
2019 * undesirable. Instead, immediately copy any pages that are
2020 * copy-on-write and enable write access in the physical map.
2022 if ((current->eflags & MAP_ENTRY_USER_WIRED) != 0 &&
2023 (current->protection & VM_PROT_WRITE) != 0 &&
2024 (old_prot & VM_PROT_WRITE) == 0)
2025 vm_fault_copy_entry(map, map, current, current, NULL);
2028 * When restricting access, update the physical map. Worry
2029 * about copy-on-write here.
2031 if ((old_prot & ~current->protection) != 0) {
2032 #define MASK(entry) (((entry)->eflags & MAP_ENTRY_COW) ? ~VM_PROT_WRITE : \
2034 pmap_protect(map->pmap, current->start,
2036 current->protection & MASK(current));
2039 vm_map_simplify_entry(map, current);
2040 current = current->next;
2043 return (KERN_SUCCESS);
2049 * This routine traverses a processes map handling the madvise
2050 * system call. Advisories are classified as either those effecting
2051 * the vm_map_entry structure, or those effecting the underlying
2061 vm_map_entry_t current, entry;
2065 * Some madvise calls directly modify the vm_map_entry, in which case
2066 * we need to use an exclusive lock on the map and we need to perform
2067 * various clipping operations. Otherwise we only need a read-lock
2072 case MADV_SEQUENTIAL:
2079 return (KERN_SUCCESS);
2087 return (KERN_SUCCESS);
2088 vm_map_lock_read(map);
2091 return (KERN_INVALID_ARGUMENT);
2095 * Locate starting entry and clip if necessary.
2097 VM_MAP_RANGE_CHECK(map, start, end);
2099 if (vm_map_lookup_entry(map, start, &entry)) {
2101 vm_map_clip_start(map, entry, start);
2103 entry = entry->next;
2108 * madvise behaviors that are implemented in the vm_map_entry.
2110 * We clip the vm_map_entry so that behavioral changes are
2111 * limited to the specified address range.
2113 for (current = entry;
2114 (current != &map->header) && (current->start < end);
2115 current = current->next
2117 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
2120 vm_map_clip_end(map, current, end);
2124 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_NORMAL);
2126 case MADV_SEQUENTIAL:
2127 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_SEQUENTIAL);
2130 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_RANDOM);
2133 current->eflags |= MAP_ENTRY_NOSYNC;
2136 current->eflags &= ~MAP_ENTRY_NOSYNC;
2139 current->eflags |= MAP_ENTRY_NOCOREDUMP;
2142 current->eflags &= ~MAP_ENTRY_NOCOREDUMP;
2147 vm_map_simplify_entry(map, current);
2151 vm_pindex_t pstart, pend;
2154 * madvise behaviors that are implemented in the underlying
2157 * Since we don't clip the vm_map_entry, we have to clip
2158 * the vm_object pindex and count.
2160 for (current = entry;
2161 (current != &map->header) && (current->start < end);
2162 current = current->next
2164 vm_offset_t useEnd, useStart;
2166 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
2169 pstart = OFF_TO_IDX(current->offset);
2170 pend = pstart + atop(current->end - current->start);
2171 useStart = current->start;
2172 useEnd = current->end;
2174 if (current->start < start) {
2175 pstart += atop(start - current->start);
2178 if (current->end > end) {
2179 pend -= atop(current->end - end);
2187 * Perform the pmap_advise() before clearing
2188 * PGA_REFERENCED in vm_page_advise(). Otherwise, a
2189 * concurrent pmap operation, such as pmap_remove(),
2190 * could clear a reference in the pmap and set
2191 * PGA_REFERENCED on the page before the pmap_advise()
2192 * had completed. Consequently, the page would appear
2193 * referenced based upon an old reference that
2194 * occurred before this pmap_advise() ran.
2196 if (behav == MADV_DONTNEED || behav == MADV_FREE)
2197 pmap_advise(map->pmap, useStart, useEnd,
2200 vm_object_madvise(current->object.vm_object, pstart,
2204 * Pre-populate paging structures in the
2205 * WILLNEED case. For wired entries, the
2206 * paging structures are already populated.
2208 if (behav == MADV_WILLNEED &&
2209 current->wired_count == 0) {
2210 vm_map_pmap_enter(map,
2212 current->protection,
2213 current->object.vm_object,
2215 ptoa(pend - pstart),
2216 MAP_PREFAULT_MADVISE
2220 vm_map_unlock_read(map);
2229 * Sets the inheritance of the specified address
2230 * range in the target map. Inheritance
2231 * affects how the map will be shared with
2232 * child maps at the time of vmspace_fork.
2235 vm_map_inherit(vm_map_t map, vm_offset_t start, vm_offset_t end,
2236 vm_inherit_t new_inheritance)
2238 vm_map_entry_t entry;
2239 vm_map_entry_t temp_entry;
2241 switch (new_inheritance) {
2242 case VM_INHERIT_NONE:
2243 case VM_INHERIT_COPY:
2244 case VM_INHERIT_SHARE:
2245 case VM_INHERIT_ZERO:
2248 return (KERN_INVALID_ARGUMENT);
2251 return (KERN_SUCCESS);
2253 VM_MAP_RANGE_CHECK(map, start, end);
2254 if (vm_map_lookup_entry(map, start, &temp_entry)) {
2256 vm_map_clip_start(map, entry, start);
2258 entry = temp_entry->next;
2259 while ((entry != &map->header) && (entry->start < end)) {
2260 vm_map_clip_end(map, entry, end);
2261 entry->inheritance = new_inheritance;
2262 vm_map_simplify_entry(map, entry);
2263 entry = entry->next;
2266 return (KERN_SUCCESS);
2272 * Implements both kernel and user unwiring.
2275 vm_map_unwire(vm_map_t map, vm_offset_t start, vm_offset_t end,
2278 vm_map_entry_t entry, first_entry, tmp_entry;
2279 vm_offset_t saved_start;
2280 unsigned int last_timestamp;
2282 boolean_t need_wakeup, result, user_unwire;
2285 return (KERN_SUCCESS);
2286 user_unwire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE;
2288 VM_MAP_RANGE_CHECK(map, start, end);
2289 if (!vm_map_lookup_entry(map, start, &first_entry)) {
2290 if (flags & VM_MAP_WIRE_HOLESOK)
2291 first_entry = first_entry->next;
2294 return (KERN_INVALID_ADDRESS);
2297 last_timestamp = map->timestamp;
2298 entry = first_entry;
2299 while (entry != &map->header && entry->start < end) {
2300 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
2302 * We have not yet clipped the entry.
2304 saved_start = (start >= entry->start) ? start :
2306 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2307 if (vm_map_unlock_and_wait(map, 0)) {
2309 * Allow interruption of user unwiring?
2313 if (last_timestamp+1 != map->timestamp) {
2315 * Look again for the entry because the map was
2316 * modified while it was unlocked.
2317 * Specifically, the entry may have been
2318 * clipped, merged, or deleted.
2320 if (!vm_map_lookup_entry(map, saved_start,
2322 if (flags & VM_MAP_WIRE_HOLESOK)
2323 tmp_entry = tmp_entry->next;
2325 if (saved_start == start) {
2327 * First_entry has been deleted.
2330 return (KERN_INVALID_ADDRESS);
2333 rv = KERN_INVALID_ADDRESS;
2337 if (entry == first_entry)
2338 first_entry = tmp_entry;
2343 last_timestamp = map->timestamp;
2346 vm_map_clip_start(map, entry, start);
2347 vm_map_clip_end(map, entry, end);
2349 * Mark the entry in case the map lock is released. (See
2352 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 &&
2353 entry->wiring_thread == NULL,
2354 ("owned map entry %p", entry));
2355 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
2356 entry->wiring_thread = curthread;
2358 * Check the map for holes in the specified region.
2359 * If VM_MAP_WIRE_HOLESOK was specified, skip this check.
2361 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) &&
2362 (entry->end < end && (entry->next == &map->header ||
2363 entry->next->start > entry->end))) {
2365 rv = KERN_INVALID_ADDRESS;
2369 * If system unwiring, require that the entry is system wired.
2372 vm_map_entry_system_wired_count(entry) == 0) {
2374 rv = KERN_INVALID_ARGUMENT;
2377 entry = entry->next;
2381 need_wakeup = FALSE;
2382 if (first_entry == NULL) {
2383 result = vm_map_lookup_entry(map, start, &first_entry);
2384 if (!result && (flags & VM_MAP_WIRE_HOLESOK))
2385 first_entry = first_entry->next;
2387 KASSERT(result, ("vm_map_unwire: lookup failed"));
2389 for (entry = first_entry; entry != &map->header && entry->start < end;
2390 entry = entry->next) {
2392 * If VM_MAP_WIRE_HOLESOK was specified, an empty
2393 * space in the unwired region could have been mapped
2394 * while the map lock was dropped for draining
2395 * MAP_ENTRY_IN_TRANSITION. Moreover, another thread
2396 * could be simultaneously wiring this new mapping
2397 * entry. Detect these cases and skip any entries
2398 * marked as in transition by us.
2400 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 ||
2401 entry->wiring_thread != curthread) {
2402 KASSERT((flags & VM_MAP_WIRE_HOLESOK) != 0,
2403 ("vm_map_unwire: !HOLESOK and new/changed entry"));
2407 if (rv == KERN_SUCCESS && (!user_unwire ||
2408 (entry->eflags & MAP_ENTRY_USER_WIRED))) {
2410 entry->eflags &= ~MAP_ENTRY_USER_WIRED;
2411 if (entry->wired_count == 1)
2412 vm_map_entry_unwire(map, entry);
2414 entry->wired_count--;
2416 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0,
2417 ("vm_map_unwire: in-transition flag missing %p", entry));
2418 KASSERT(entry->wiring_thread == curthread,
2419 ("vm_map_unwire: alien wire %p", entry));
2420 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION;
2421 entry->wiring_thread = NULL;
2422 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
2423 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
2426 vm_map_simplify_entry(map, entry);
2435 * vm_map_wire_entry_failure:
2437 * Handle a wiring failure on the given entry.
2439 * The map should be locked.
2442 vm_map_wire_entry_failure(vm_map_t map, vm_map_entry_t entry,
2443 vm_offset_t failed_addr)
2446 VM_MAP_ASSERT_LOCKED(map);
2447 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 &&
2448 entry->wired_count == 1,
2449 ("vm_map_wire_entry_failure: entry %p isn't being wired", entry));
2450 KASSERT(failed_addr < entry->end,
2451 ("vm_map_wire_entry_failure: entry %p was fully wired", entry));
2454 * If any pages at the start of this entry were successfully wired,
2457 if (failed_addr > entry->start) {
2458 pmap_unwire(map->pmap, entry->start, failed_addr);
2459 vm_object_unwire(entry->object.vm_object, entry->offset,
2460 failed_addr - entry->start, PQ_ACTIVE);
2464 * Assign an out-of-range value to represent the failure to wire this
2467 entry->wired_count = -1;
2473 * Implements both kernel and user wiring.
2476 vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t end,
2479 vm_map_entry_t entry, first_entry, tmp_entry;
2480 vm_offset_t faddr, saved_end, saved_start;
2481 unsigned int last_timestamp;
2483 boolean_t need_wakeup, result, user_wire;
2487 return (KERN_SUCCESS);
2489 if (flags & VM_MAP_WIRE_WRITE)
2490 prot |= VM_PROT_WRITE;
2491 user_wire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE;
2493 VM_MAP_RANGE_CHECK(map, start, end);
2494 if (!vm_map_lookup_entry(map, start, &first_entry)) {
2495 if (flags & VM_MAP_WIRE_HOLESOK)
2496 first_entry = first_entry->next;
2499 return (KERN_INVALID_ADDRESS);
2502 last_timestamp = map->timestamp;
2503 entry = first_entry;
2504 while (entry != &map->header && entry->start < end) {
2505 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
2507 * We have not yet clipped the entry.
2509 saved_start = (start >= entry->start) ? start :
2511 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2512 if (vm_map_unlock_and_wait(map, 0)) {
2514 * Allow interruption of user wiring?
2518 if (last_timestamp + 1 != map->timestamp) {
2520 * Look again for the entry because the map was
2521 * modified while it was unlocked.
2522 * Specifically, the entry may have been
2523 * clipped, merged, or deleted.
2525 if (!vm_map_lookup_entry(map, saved_start,
2527 if (flags & VM_MAP_WIRE_HOLESOK)
2528 tmp_entry = tmp_entry->next;
2530 if (saved_start == start) {
2532 * first_entry has been deleted.
2535 return (KERN_INVALID_ADDRESS);
2538 rv = KERN_INVALID_ADDRESS;
2542 if (entry == first_entry)
2543 first_entry = tmp_entry;
2548 last_timestamp = map->timestamp;
2551 vm_map_clip_start(map, entry, start);
2552 vm_map_clip_end(map, entry, end);
2554 * Mark the entry in case the map lock is released. (See
2557 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 &&
2558 entry->wiring_thread == NULL,
2559 ("owned map entry %p", entry));
2560 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
2561 entry->wiring_thread = curthread;
2562 if ((entry->protection & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0
2563 || (entry->protection & prot) != prot) {
2564 entry->eflags |= MAP_ENTRY_WIRE_SKIPPED;
2565 if ((flags & VM_MAP_WIRE_HOLESOK) == 0) {
2567 rv = KERN_INVALID_ADDRESS;
2572 if (entry->wired_count == 0) {
2573 entry->wired_count++;
2574 saved_start = entry->start;
2575 saved_end = entry->end;
2578 * Release the map lock, relying on the in-transition
2579 * mark. Mark the map busy for fork.
2584 faddr = saved_start;
2587 * Simulate a fault to get the page and enter
2588 * it into the physical map.
2590 if ((rv = vm_fault(map, faddr, VM_PROT_NONE,
2591 VM_FAULT_WIRE)) != KERN_SUCCESS)
2593 } while ((faddr += PAGE_SIZE) < saved_end);
2596 if (last_timestamp + 1 != map->timestamp) {
2598 * Look again for the entry because the map was
2599 * modified while it was unlocked. The entry
2600 * may have been clipped, but NOT merged or
2603 result = vm_map_lookup_entry(map, saved_start,
2605 KASSERT(result, ("vm_map_wire: lookup failed"));
2606 if (entry == first_entry)
2607 first_entry = tmp_entry;
2611 while (entry->end < saved_end) {
2613 * In case of failure, handle entries
2614 * that were not fully wired here;
2615 * fully wired entries are handled
2618 if (rv != KERN_SUCCESS &&
2620 vm_map_wire_entry_failure(map,
2622 entry = entry->next;
2625 last_timestamp = map->timestamp;
2626 if (rv != KERN_SUCCESS) {
2627 vm_map_wire_entry_failure(map, entry, faddr);
2631 } else if (!user_wire ||
2632 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
2633 entry->wired_count++;
2636 * Check the map for holes in the specified region.
2637 * If VM_MAP_WIRE_HOLESOK was specified, skip this check.
2640 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) &&
2641 (entry->end < end && (entry->next == &map->header ||
2642 entry->next->start > entry->end))) {
2644 rv = KERN_INVALID_ADDRESS;
2647 entry = entry->next;
2651 need_wakeup = FALSE;
2652 if (first_entry == NULL) {
2653 result = vm_map_lookup_entry(map, start, &first_entry);
2654 if (!result && (flags & VM_MAP_WIRE_HOLESOK))
2655 first_entry = first_entry->next;
2657 KASSERT(result, ("vm_map_wire: lookup failed"));
2659 for (entry = first_entry; entry != &map->header && entry->start < end;
2660 entry = entry->next) {
2662 * If VM_MAP_WIRE_HOLESOK was specified, an empty
2663 * space in the unwired region could have been mapped
2664 * while the map lock was dropped for faulting in the
2665 * pages or draining MAP_ENTRY_IN_TRANSITION.
2666 * Moreover, another thread could be simultaneously
2667 * wiring this new mapping entry. Detect these cases
2668 * and skip any entries marked as in transition not by us.
2670 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 ||
2671 entry->wiring_thread != curthread) {
2672 KASSERT((flags & VM_MAP_WIRE_HOLESOK) != 0,
2673 ("vm_map_wire: !HOLESOK and new/changed entry"));
2677 if ((entry->eflags & MAP_ENTRY_WIRE_SKIPPED) != 0)
2678 goto next_entry_done;
2680 if (rv == KERN_SUCCESS) {
2682 entry->eflags |= MAP_ENTRY_USER_WIRED;
2683 } else if (entry->wired_count == -1) {
2685 * Wiring failed on this entry. Thus, unwiring is
2688 entry->wired_count = 0;
2689 } else if (!user_wire ||
2690 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
2692 * Undo the wiring. Wiring succeeded on this entry
2693 * but failed on a later entry.
2695 if (entry->wired_count == 1)
2696 vm_map_entry_unwire(map, entry);
2698 entry->wired_count--;
2701 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0,
2702 ("vm_map_wire: in-transition flag missing %p", entry));
2703 KASSERT(entry->wiring_thread == curthread,
2704 ("vm_map_wire: alien wire %p", entry));
2705 entry->eflags &= ~(MAP_ENTRY_IN_TRANSITION |
2706 MAP_ENTRY_WIRE_SKIPPED);
2707 entry->wiring_thread = NULL;
2708 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
2709 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
2712 vm_map_simplify_entry(map, entry);
2723 * Push any dirty cached pages in the address range to their pager.
2724 * If syncio is TRUE, dirty pages are written synchronously.
2725 * If invalidate is TRUE, any cached pages are freed as well.
2727 * If the size of the region from start to end is zero, we are
2728 * supposed to flush all modified pages within the region containing
2729 * start. Unfortunately, a region can be split or coalesced with
2730 * neighboring regions, making it difficult to determine what the
2731 * original region was. Therefore, we approximate this requirement by
2732 * flushing the current region containing start.
2734 * Returns an error if any part of the specified range is not mapped.
2742 boolean_t invalidate)
2744 vm_map_entry_t current;
2745 vm_map_entry_t entry;
2748 vm_ooffset_t offset;
2749 unsigned int last_timestamp;
2752 vm_map_lock_read(map);
2753 VM_MAP_RANGE_CHECK(map, start, end);
2754 if (!vm_map_lookup_entry(map, start, &entry)) {
2755 vm_map_unlock_read(map);
2756 return (KERN_INVALID_ADDRESS);
2757 } else if (start == end) {
2758 start = entry->start;
2762 * Make a first pass to check for user-wired memory and holes.
2764 for (current = entry; current != &map->header && current->start < end;
2765 current = current->next) {
2766 if (invalidate && (current->eflags & MAP_ENTRY_USER_WIRED)) {
2767 vm_map_unlock_read(map);
2768 return (KERN_INVALID_ARGUMENT);
2770 if (end > current->end &&
2771 (current->next == &map->header ||
2772 current->end != current->next->start)) {
2773 vm_map_unlock_read(map);
2774 return (KERN_INVALID_ADDRESS);
2779 pmap_remove(map->pmap, start, end);
2783 * Make a second pass, cleaning/uncaching pages from the indicated
2786 for (current = entry; current != &map->header && current->start < end;) {
2787 offset = current->offset + (start - current->start);
2788 size = (end <= current->end ? end : current->end) - start;
2789 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
2791 vm_map_entry_t tentry;
2794 smap = current->object.sub_map;
2795 vm_map_lock_read(smap);
2796 (void) vm_map_lookup_entry(smap, offset, &tentry);
2797 tsize = tentry->end - offset;
2800 object = tentry->object.vm_object;
2801 offset = tentry->offset + (offset - tentry->start);
2802 vm_map_unlock_read(smap);
2804 object = current->object.vm_object;
2806 vm_object_reference(object);
2807 last_timestamp = map->timestamp;
2808 vm_map_unlock_read(map);
2809 if (!vm_object_sync(object, offset, size, syncio, invalidate))
2812 vm_object_deallocate(object);
2813 vm_map_lock_read(map);
2814 if (last_timestamp == map->timestamp ||
2815 !vm_map_lookup_entry(map, start, ¤t))
2816 current = current->next;
2819 vm_map_unlock_read(map);
2820 return (failed ? KERN_FAILURE : KERN_SUCCESS);
2824 * vm_map_entry_unwire: [ internal use only ]
2826 * Make the region specified by this entry pageable.
2828 * The map in question should be locked.
2829 * [This is the reason for this routine's existence.]
2832 vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry)
2835 VM_MAP_ASSERT_LOCKED(map);
2836 KASSERT(entry->wired_count > 0,
2837 ("vm_map_entry_unwire: entry %p isn't wired", entry));
2838 pmap_unwire(map->pmap, entry->start, entry->end);
2839 vm_object_unwire(entry->object.vm_object, entry->offset, entry->end -
2840 entry->start, PQ_ACTIVE);
2841 entry->wired_count = 0;
2845 vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map)
2848 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0)
2849 vm_object_deallocate(entry->object.vm_object);
2850 uma_zfree(system_map ? kmapentzone : mapentzone, entry);
2854 * vm_map_entry_delete: [ internal use only ]
2856 * Deallocate the given entry from the target map.
2859 vm_map_entry_delete(vm_map_t map, vm_map_entry_t entry)
2862 vm_pindex_t offidxstart, offidxend, count, size1;
2865 vm_map_entry_unlink(map, entry);
2866 object = entry->object.vm_object;
2867 size = entry->end - entry->start;
2870 if (entry->cred != NULL) {
2871 swap_release_by_cred(size, entry->cred);
2872 crfree(entry->cred);
2875 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0 &&
2877 KASSERT(entry->cred == NULL || object->cred == NULL ||
2878 (entry->eflags & MAP_ENTRY_NEEDS_COPY),
2879 ("OVERCOMMIT vm_map_entry_delete: both cred %p", entry));
2880 count = OFF_TO_IDX(size);
2881 offidxstart = OFF_TO_IDX(entry->offset);
2882 offidxend = offidxstart + count;
2883 VM_OBJECT_WLOCK(object);
2884 if (object->ref_count != 1 && ((object->flags & (OBJ_NOSPLIT |
2885 OBJ_ONEMAPPING)) == OBJ_ONEMAPPING ||
2886 object == kernel_object || object == kmem_object)) {
2887 vm_object_collapse(object);
2890 * The option OBJPR_NOTMAPPED can be passed here
2891 * because vm_map_delete() already performed
2892 * pmap_remove() on the only mapping to this range
2895 vm_object_page_remove(object, offidxstart, offidxend,
2897 if (object->type == OBJT_SWAP)
2898 swap_pager_freespace(object, offidxstart,
2900 if (offidxend >= object->size &&
2901 offidxstart < object->size) {
2902 size1 = object->size;
2903 object->size = offidxstart;
2904 if (object->cred != NULL) {
2905 size1 -= object->size;
2906 KASSERT(object->charge >= ptoa(size1),
2907 ("object %p charge < 0", object));
2908 swap_release_by_cred(ptoa(size1),
2910 object->charge -= ptoa(size1);
2914 VM_OBJECT_WUNLOCK(object);
2916 entry->object.vm_object = NULL;
2917 if (map->system_map)
2918 vm_map_entry_deallocate(entry, TRUE);
2920 entry->next = curthread->td_map_def_user;
2921 curthread->td_map_def_user = entry;
2926 * vm_map_delete: [ internal use only ]
2928 * Deallocates the given address range from the target
2932 vm_map_delete(vm_map_t map, vm_offset_t start, vm_offset_t end)
2934 vm_map_entry_t entry;
2935 vm_map_entry_t first_entry;
2937 VM_MAP_ASSERT_LOCKED(map);
2939 return (KERN_SUCCESS);
2942 * Find the start of the region, and clip it
2944 if (!vm_map_lookup_entry(map, start, &first_entry))
2945 entry = first_entry->next;
2947 entry = first_entry;
2948 vm_map_clip_start(map, entry, start);
2952 * Step through all entries in this region
2954 while ((entry != &map->header) && (entry->start < end)) {
2955 vm_map_entry_t next;
2958 * Wait for wiring or unwiring of an entry to complete.
2959 * Also wait for any system wirings to disappear on
2962 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 ||
2963 (vm_map_pmap(map) != kernel_pmap &&
2964 vm_map_entry_system_wired_count(entry) != 0)) {
2965 unsigned int last_timestamp;
2966 vm_offset_t saved_start;
2967 vm_map_entry_t tmp_entry;
2969 saved_start = entry->start;
2970 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2971 last_timestamp = map->timestamp;
2972 (void) vm_map_unlock_and_wait(map, 0);
2974 if (last_timestamp + 1 != map->timestamp) {
2976 * Look again for the entry because the map was
2977 * modified while it was unlocked.
2978 * Specifically, the entry may have been
2979 * clipped, merged, or deleted.
2981 if (!vm_map_lookup_entry(map, saved_start,
2983 entry = tmp_entry->next;
2986 vm_map_clip_start(map, entry,
2992 vm_map_clip_end(map, entry, end);
2997 * Unwire before removing addresses from the pmap; otherwise,
2998 * unwiring will put the entries back in the pmap.
3000 if (entry->wired_count != 0) {
3001 vm_map_entry_unwire(map, entry);
3004 pmap_remove(map->pmap, entry->start, entry->end);
3007 * Delete the entry only after removing all pmap
3008 * entries pointing to its pages. (Otherwise, its
3009 * page frames may be reallocated, and any modify bits
3010 * will be set in the wrong object!)
3012 vm_map_entry_delete(map, entry);
3015 return (KERN_SUCCESS);
3021 * Remove the given address range from the target map.
3022 * This is the exported form of vm_map_delete.
3025 vm_map_remove(vm_map_t map, vm_offset_t start, vm_offset_t end)
3030 VM_MAP_RANGE_CHECK(map, start, end);
3031 result = vm_map_delete(map, start, end);
3037 * vm_map_check_protection:
3039 * Assert that the target map allows the specified privilege on the
3040 * entire address region given. The entire region must be allocated.
3042 * WARNING! This code does not and should not check whether the
3043 * contents of the region is accessible. For example a smaller file
3044 * might be mapped into a larger address space.
3046 * NOTE! This code is also called by munmap().
3048 * The map must be locked. A read lock is sufficient.
3051 vm_map_check_protection(vm_map_t map, vm_offset_t start, vm_offset_t end,
3052 vm_prot_t protection)
3054 vm_map_entry_t entry;
3055 vm_map_entry_t tmp_entry;
3057 if (!vm_map_lookup_entry(map, start, &tmp_entry))
3061 while (start < end) {
3062 if (entry == &map->header)
3067 if (start < entry->start)
3070 * Check protection associated with entry.
3072 if ((entry->protection & protection) != protection)
3074 /* go to next entry */
3076 entry = entry->next;
3082 * vm_map_copy_entry:
3084 * Copies the contents of the source entry to the destination
3085 * entry. The entries *must* be aligned properly.
3091 vm_map_entry_t src_entry,
3092 vm_map_entry_t dst_entry,
3093 vm_ooffset_t *fork_charge)
3095 vm_object_t src_object;
3096 vm_map_entry_t fake_entry;
3101 VM_MAP_ASSERT_LOCKED(dst_map);
3103 if ((dst_entry->eflags|src_entry->eflags) & MAP_ENTRY_IS_SUB_MAP)
3106 if (src_entry->wired_count == 0 ||
3107 (src_entry->protection & VM_PROT_WRITE) == 0) {
3109 * If the source entry is marked needs_copy, it is already
3112 if ((src_entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0 &&
3113 (src_entry->protection & VM_PROT_WRITE) != 0) {
3114 pmap_protect(src_map->pmap,
3117 src_entry->protection & ~VM_PROT_WRITE);
3121 * Make a copy of the object.
3123 size = src_entry->end - src_entry->start;
3124 if ((src_object = src_entry->object.vm_object) != NULL) {
3125 VM_OBJECT_WLOCK(src_object);
3126 charged = ENTRY_CHARGED(src_entry);
3127 if (src_object->handle == NULL &&
3128 (src_object->type == OBJT_DEFAULT ||
3129 src_object->type == OBJT_SWAP)) {
3130 vm_object_collapse(src_object);
3131 if ((src_object->flags & (OBJ_NOSPLIT |
3132 OBJ_ONEMAPPING)) == OBJ_ONEMAPPING) {
3133 vm_object_split(src_entry);
3135 src_entry->object.vm_object;
3138 vm_object_reference_locked(src_object);
3139 vm_object_clear_flag(src_object, OBJ_ONEMAPPING);
3140 if (src_entry->cred != NULL &&
3141 !(src_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
3142 KASSERT(src_object->cred == NULL,
3143 ("OVERCOMMIT: vm_map_copy_entry: cred %p",
3145 src_object->cred = src_entry->cred;
3146 src_object->charge = size;
3148 VM_OBJECT_WUNLOCK(src_object);
3149 dst_entry->object.vm_object = src_object;
3151 cred = curthread->td_ucred;
3153 dst_entry->cred = cred;
3154 *fork_charge += size;
3155 if (!(src_entry->eflags &
3156 MAP_ENTRY_NEEDS_COPY)) {
3158 src_entry->cred = cred;
3159 *fork_charge += size;
3162 src_entry->eflags |= MAP_ENTRY_COW |
3163 MAP_ENTRY_NEEDS_COPY;
3164 dst_entry->eflags |= MAP_ENTRY_COW |
3165 MAP_ENTRY_NEEDS_COPY;
3166 dst_entry->offset = src_entry->offset;
3167 if (src_entry->eflags & MAP_ENTRY_VN_WRITECNT) {
3169 * MAP_ENTRY_VN_WRITECNT cannot
3170 * indicate write reference from
3171 * src_entry, since the entry is
3172 * marked as needs copy. Allocate a
3173 * fake entry that is used to
3174 * decrement object->un_pager.vnp.writecount
3175 * at the appropriate time. Attach
3176 * fake_entry to the deferred list.
3178 fake_entry = vm_map_entry_create(dst_map);
3179 fake_entry->eflags = MAP_ENTRY_VN_WRITECNT;
3180 src_entry->eflags &= ~MAP_ENTRY_VN_WRITECNT;
3181 vm_object_reference(src_object);
3182 fake_entry->object.vm_object = src_object;
3183 fake_entry->start = src_entry->start;
3184 fake_entry->end = src_entry->end;
3185 fake_entry->next = curthread->td_map_def_user;
3186 curthread->td_map_def_user = fake_entry;
3189 pmap_copy(dst_map->pmap, src_map->pmap,
3190 dst_entry->start, dst_entry->end - dst_entry->start,
3193 dst_entry->object.vm_object = NULL;
3194 dst_entry->offset = 0;
3195 if (src_entry->cred != NULL) {
3196 dst_entry->cred = curthread->td_ucred;
3197 crhold(dst_entry->cred);
3198 *fork_charge += size;
3203 * We don't want to make writeable wired pages copy-on-write.
3204 * Immediately copy these pages into the new map by simulating
3205 * page faults. The new pages are pageable.
3207 vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry,
3213 * vmspace_map_entry_forked:
3214 * Update the newly-forked vmspace each time a map entry is inherited
3215 * or copied. The values for vm_dsize and vm_tsize are approximate
3216 * (and mostly-obsolete ideas in the face of mmap(2) et al.)
3219 vmspace_map_entry_forked(const struct vmspace *vm1, struct vmspace *vm2,
3220 vm_map_entry_t entry)
3222 vm_size_t entrysize;
3225 entrysize = entry->end - entry->start;
3226 vm2->vm_map.size += entrysize;
3227 if (entry->eflags & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP)) {
3228 vm2->vm_ssize += btoc(entrysize);
3229 } else if (entry->start >= (vm_offset_t)vm1->vm_daddr &&
3230 entry->start < (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)) {
3231 newend = MIN(entry->end,
3232 (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize));
3233 vm2->vm_dsize += btoc(newend - entry->start);
3234 } else if (entry->start >= (vm_offset_t)vm1->vm_taddr &&
3235 entry->start < (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)) {
3236 newend = MIN(entry->end,
3237 (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize));
3238 vm2->vm_tsize += btoc(newend - entry->start);
3244 * Create a new process vmspace structure and vm_map
3245 * based on those of an existing process. The new map
3246 * is based on the old map, according to the inheritance
3247 * values on the regions in that map.
3249 * XXX It might be worth coalescing the entries added to the new vmspace.
3251 * The source map must not be locked.
3254 vmspace_fork(struct vmspace *vm1, vm_ooffset_t *fork_charge)
3256 struct vmspace *vm2;
3257 vm_map_t new_map, old_map;
3258 vm_map_entry_t new_entry, old_entry;
3262 old_map = &vm1->vm_map;
3263 /* Copy immutable fields of vm1 to vm2. */
3264 vm2 = vmspace_alloc(old_map->min_offset, old_map->max_offset, NULL);
3267 vm2->vm_taddr = vm1->vm_taddr;
3268 vm2->vm_daddr = vm1->vm_daddr;
3269 vm2->vm_maxsaddr = vm1->vm_maxsaddr;
3270 vm_map_lock(old_map);
3272 vm_map_wait_busy(old_map);
3273 new_map = &vm2->vm_map;
3274 locked = vm_map_trylock(new_map); /* trylock to silence WITNESS */
3275 KASSERT(locked, ("vmspace_fork: lock failed"));
3277 old_entry = old_map->header.next;
3279 while (old_entry != &old_map->header) {
3280 if (old_entry->eflags & MAP_ENTRY_IS_SUB_MAP)
3281 panic("vm_map_fork: encountered a submap");
3283 switch (old_entry->inheritance) {
3284 case VM_INHERIT_NONE:
3287 case VM_INHERIT_SHARE:
3289 * Clone the entry, creating the shared object if necessary.
3291 object = old_entry->object.vm_object;
3292 if (object == NULL) {
3293 object = vm_object_allocate(OBJT_DEFAULT,
3294 atop(old_entry->end - old_entry->start));
3295 old_entry->object.vm_object = object;
3296 old_entry->offset = 0;
3297 if (old_entry->cred != NULL) {
3298 object->cred = old_entry->cred;
3299 object->charge = old_entry->end -
3301 old_entry->cred = NULL;
3306 * Add the reference before calling vm_object_shadow
3307 * to insure that a shadow object is created.
3309 vm_object_reference(object);
3310 if (old_entry->eflags & MAP_ENTRY_NEEDS_COPY) {
3311 vm_object_shadow(&old_entry->object.vm_object,
3313 old_entry->end - old_entry->start);
3314 old_entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
3315 /* Transfer the second reference too. */
3316 vm_object_reference(
3317 old_entry->object.vm_object);
3320 * As in vm_map_simplify_entry(), the
3321 * vnode lock will not be acquired in
3322 * this call to vm_object_deallocate().
3324 vm_object_deallocate(object);
3325 object = old_entry->object.vm_object;
3327 VM_OBJECT_WLOCK(object);
3328 vm_object_clear_flag(object, OBJ_ONEMAPPING);
3329 if (old_entry->cred != NULL) {
3330 KASSERT(object->cred == NULL, ("vmspace_fork both cred"));
3331 object->cred = old_entry->cred;
3332 object->charge = old_entry->end - old_entry->start;
3333 old_entry->cred = NULL;
3337 * Assert the correct state of the vnode
3338 * v_writecount while the object is locked, to
3339 * not relock it later for the assertion
3342 if (old_entry->eflags & MAP_ENTRY_VN_WRITECNT &&
3343 object->type == OBJT_VNODE) {
3344 KASSERT(((struct vnode *)object->handle)->
3346 ("vmspace_fork: v_writecount %p", object));
3347 KASSERT(object->un_pager.vnp.writemappings > 0,
3348 ("vmspace_fork: vnp.writecount %p",
3351 VM_OBJECT_WUNLOCK(object);
3354 * Clone the entry, referencing the shared object.
3356 new_entry = vm_map_entry_create(new_map);
3357 *new_entry = *old_entry;
3358 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED |
3359 MAP_ENTRY_IN_TRANSITION);
3360 new_entry->wiring_thread = NULL;
3361 new_entry->wired_count = 0;
3362 if (new_entry->eflags & MAP_ENTRY_VN_WRITECNT) {
3363 vnode_pager_update_writecount(object,
3364 new_entry->start, new_entry->end);
3368 * Insert the entry into the new map -- we know we're
3369 * inserting at the end of the new map.
3371 vm_map_entry_link(new_map, new_map->header.prev,
3373 vmspace_map_entry_forked(vm1, vm2, new_entry);
3376 * Update the physical map
3378 pmap_copy(new_map->pmap, old_map->pmap,
3380 (old_entry->end - old_entry->start),
3384 case VM_INHERIT_COPY:
3386 * Clone the entry and link into the map.
3388 new_entry = vm_map_entry_create(new_map);
3389 *new_entry = *old_entry;
3391 * Copied entry is COW over the old object.
3393 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED |
3394 MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_VN_WRITECNT);
3395 new_entry->wiring_thread = NULL;
3396 new_entry->wired_count = 0;
3397 new_entry->object.vm_object = NULL;
3398 new_entry->cred = NULL;
3399 vm_map_entry_link(new_map, new_map->header.prev,
3401 vmspace_map_entry_forked(vm1, vm2, new_entry);
3402 vm_map_copy_entry(old_map, new_map, old_entry,
3403 new_entry, fork_charge);
3406 case VM_INHERIT_ZERO:
3408 * Create a new anonymous mapping entry modelled from
3411 new_entry = vm_map_entry_create(new_map);
3412 memset(new_entry, 0, sizeof(*new_entry));
3414 new_entry->start = old_entry->start;
3415 new_entry->end = old_entry->end;
3416 new_entry->avail_ssize = old_entry->avail_ssize;
3417 new_entry->eflags = old_entry->eflags &
3418 ~(MAP_ENTRY_USER_WIRED | MAP_ENTRY_IN_TRANSITION |
3419 MAP_ENTRY_VN_WRITECNT);
3420 new_entry->protection = old_entry->protection;
3421 new_entry->max_protection = old_entry->max_protection;
3422 new_entry->inheritance = VM_INHERIT_ZERO;
3424 vm_map_entry_link(new_map, new_map->header.prev,
3426 vmspace_map_entry_forked(vm1, vm2, new_entry);
3428 new_entry->cred = curthread->td_ucred;
3429 crhold(new_entry->cred);
3430 *fork_charge += (new_entry->end - new_entry->start);
3434 old_entry = old_entry->next;
3437 * Use inlined vm_map_unlock() to postpone handling the deferred
3438 * map entries, which cannot be done until both old_map and
3439 * new_map locks are released.
3441 sx_xunlock(&old_map->lock);
3442 sx_xunlock(&new_map->lock);
3443 vm_map_process_deferred();
3449 vm_map_stack(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
3450 vm_prot_t prot, vm_prot_t max, int cow)
3452 vm_size_t growsize, init_ssize;
3453 rlim_t lmemlim, vmemlim;
3456 growsize = sgrowsiz;
3457 init_ssize = (max_ssize < growsize) ? max_ssize : growsize;
3460 lmemlim = lim_cur(curproc, RLIMIT_MEMLOCK);
3461 vmemlim = lim_cur(curproc, RLIMIT_VMEM);
3462 PROC_UNLOCK(curproc);
3463 if (!old_mlock && map->flags & MAP_WIREFUTURE) {
3464 if (ptoa(pmap_wired_count(map->pmap)) + init_ssize > lmemlim) {
3469 /* If we would blow our VMEM resource limit, no go */
3470 if (map->size + init_ssize > vmemlim) {
3474 rv = vm_map_stack_locked(map, addrbos, max_ssize, growsize, prot,
3482 vm_map_stack_locked(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
3483 vm_size_t growsize, vm_prot_t prot, vm_prot_t max, int cow)
3485 vm_map_entry_t new_entry, prev_entry;
3486 vm_offset_t bot, top;
3487 vm_size_t init_ssize;
3491 * The stack orientation is piggybacked with the cow argument.
3492 * Extract it into orient and mask the cow argument so that we
3493 * don't pass it around further.
3494 * NOTE: We explicitly allow bi-directional stacks.
3496 orient = cow & (MAP_STACK_GROWS_DOWN|MAP_STACK_GROWS_UP);
3497 KASSERT(orient != 0, ("No stack grow direction"));
3499 if (addrbos < vm_map_min(map) ||
3500 addrbos > vm_map_max(map) ||
3501 addrbos + max_ssize < addrbos)
3502 return (KERN_NO_SPACE);
3504 init_ssize = (max_ssize < growsize) ? max_ssize : growsize;
3506 /* If addr is already mapped, no go */
3507 if (vm_map_lookup_entry(map, addrbos, &prev_entry))
3508 return (KERN_NO_SPACE);
3511 * If we can't accomodate max_ssize in the current mapping, no go.
3512 * However, we need to be aware that subsequent user mappings might
3513 * map into the space we have reserved for stack, and currently this
3514 * space is not protected.
3516 * Hopefully we will at least detect this condition when we try to
3519 if ((prev_entry->next != &map->header) &&
3520 (prev_entry->next->start < addrbos + max_ssize))
3521 return (KERN_NO_SPACE);
3524 * We initially map a stack of only init_ssize. We will grow as
3525 * needed later. Depending on the orientation of the stack (i.e.
3526 * the grow direction) we either map at the top of the range, the
3527 * bottom of the range or in the middle.
3529 * Note: we would normally expect prot and max to be VM_PROT_ALL,
3530 * and cow to be 0. Possibly we should eliminate these as input
3531 * parameters, and just pass these values here in the insert call.
3533 if (orient == MAP_STACK_GROWS_DOWN)
3534 bot = addrbos + max_ssize - init_ssize;
3535 else if (orient == MAP_STACK_GROWS_UP)
3538 bot = round_page(addrbos + max_ssize/2 - init_ssize/2);
3539 top = bot + init_ssize;
3540 rv = vm_map_insert(map, NULL, 0, bot, top, prot, max, cow);
3542 /* Now set the avail_ssize amount. */
3543 if (rv == KERN_SUCCESS) {
3544 new_entry = prev_entry->next;
3545 if (new_entry->end != top || new_entry->start != bot)
3546 panic("Bad entry start/end for new stack entry");
3548 new_entry->avail_ssize = max_ssize - init_ssize;
3549 KASSERT((orient & MAP_STACK_GROWS_DOWN) == 0 ||
3550 (new_entry->eflags & MAP_ENTRY_GROWS_DOWN) != 0,
3551 ("new entry lacks MAP_ENTRY_GROWS_DOWN"));
3552 KASSERT((orient & MAP_STACK_GROWS_UP) == 0 ||
3553 (new_entry->eflags & MAP_ENTRY_GROWS_UP) != 0,
3554 ("new entry lacks MAP_ENTRY_GROWS_UP"));
3560 static int stack_guard_page = 0;
3561 TUNABLE_INT("security.bsd.stack_guard_page", &stack_guard_page);
3562 SYSCTL_INT(_security_bsd, OID_AUTO, stack_guard_page, CTLFLAG_RW,
3563 &stack_guard_page, 0,
3564 "Insert stack guard page ahead of the growable segments.");
3566 /* Attempts to grow a vm stack entry. Returns KERN_SUCCESS if the
3567 * desired address is already mapped, or if we successfully grow
3568 * the stack. Also returns KERN_SUCCESS if addr is outside the
3569 * stack range (this is strange, but preserves compatibility with
3570 * the grow function in vm_machdep.c).
3573 vm_map_growstack(struct proc *p, vm_offset_t addr)
3575 vm_map_entry_t next_entry, prev_entry;
3576 vm_map_entry_t new_entry, stack_entry;
3577 struct vmspace *vm = p->p_vmspace;
3578 vm_map_t map = &vm->vm_map;
3581 size_t grow_amount, max_grow;
3582 rlim_t lmemlim, stacklim, vmemlim;
3583 int is_procstack, rv;
3594 lmemlim = lim_cur(p, RLIMIT_MEMLOCK);
3595 stacklim = lim_cur(p, RLIMIT_STACK);
3596 vmemlim = lim_cur(p, RLIMIT_VMEM);
3599 vm_map_lock_read(map);
3601 /* If addr is already in the entry range, no need to grow.*/
3602 if (vm_map_lookup_entry(map, addr, &prev_entry)) {
3603 vm_map_unlock_read(map);
3604 return (KERN_SUCCESS);
3607 next_entry = prev_entry->next;
3608 if (!(prev_entry->eflags & MAP_ENTRY_GROWS_UP)) {
3610 * This entry does not grow upwards. Since the address lies
3611 * beyond this entry, the next entry (if one exists) has to
3612 * be a downward growable entry. The entry list header is
3613 * never a growable entry, so it suffices to check the flags.
3615 if (!(next_entry->eflags & MAP_ENTRY_GROWS_DOWN)) {
3616 vm_map_unlock_read(map);
3617 return (KERN_SUCCESS);
3619 stack_entry = next_entry;
3622 * This entry grows upward. If the next entry does not at
3623 * least grow downwards, this is the entry we need to grow.
3624 * otherwise we have two possible choices and we have to
3627 if (next_entry->eflags & MAP_ENTRY_GROWS_DOWN) {
3629 * We have two choices; grow the entry closest to
3630 * the address to minimize the amount of growth.
3632 if (addr - prev_entry->end <= next_entry->start - addr)
3633 stack_entry = prev_entry;
3635 stack_entry = next_entry;
3637 stack_entry = prev_entry;
3640 if (stack_entry == next_entry) {
3641 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_DOWN, ("foo"));
3642 KASSERT(addr < stack_entry->start, ("foo"));
3643 end = (prev_entry != &map->header) ? prev_entry->end :
3644 stack_entry->start - stack_entry->avail_ssize;
3645 grow_amount = roundup(stack_entry->start - addr, PAGE_SIZE);
3646 max_grow = stack_entry->start - end;
3648 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_UP, ("foo"));
3649 KASSERT(addr >= stack_entry->end, ("foo"));
3650 end = (next_entry != &map->header) ? next_entry->start :
3651 stack_entry->end + stack_entry->avail_ssize;
3652 grow_amount = roundup(addr + 1 - stack_entry->end, PAGE_SIZE);
3653 max_grow = end - stack_entry->end;
3656 if (grow_amount > stack_entry->avail_ssize) {
3657 vm_map_unlock_read(map);
3658 return (KERN_NO_SPACE);
3662 * If there is no longer enough space between the entries nogo, and
3663 * adjust the available space. Note: this should only happen if the
3664 * user has mapped into the stack area after the stack was created,
3665 * and is probably an error.
3667 * This also effectively destroys any guard page the user might have
3668 * intended by limiting the stack size.
3670 if (grow_amount + (stack_guard_page ? PAGE_SIZE : 0) > max_grow) {
3671 if (vm_map_lock_upgrade(map))
3674 stack_entry->avail_ssize = max_grow;
3677 return (KERN_NO_SPACE);
3680 is_procstack = (addr >= (vm_offset_t)vm->vm_maxsaddr &&
3681 addr < (vm_offset_t)p->p_sysent->sv_usrstack) ? 1 : 0;
3684 * If this is the main process stack, see if we're over the stack
3687 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) {
3688 vm_map_unlock_read(map);
3689 return (KERN_NO_SPACE);
3694 if (is_procstack && racct_set(p, RACCT_STACK,
3695 ctob(vm->vm_ssize) + grow_amount)) {
3697 vm_map_unlock_read(map);
3698 return (KERN_NO_SPACE);
3704 /* Round up the grow amount modulo sgrowsiz */
3705 growsize = sgrowsiz;
3706 grow_amount = roundup(grow_amount, growsize);
3707 if (grow_amount > stack_entry->avail_ssize)
3708 grow_amount = stack_entry->avail_ssize;
3709 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) {
3710 grow_amount = trunc_page((vm_size_t)stacklim) -
3715 limit = racct_get_available(p, RACCT_STACK);
3717 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > limit))
3718 grow_amount = limit - ctob(vm->vm_ssize);
3720 if (!old_mlock && map->flags & MAP_WIREFUTURE) {
3721 if (ptoa(pmap_wired_count(map->pmap)) + grow_amount > lmemlim) {
3722 vm_map_unlock_read(map);
3729 if (racct_set(p, RACCT_MEMLOCK,
3730 ptoa(pmap_wired_count(map->pmap)) + grow_amount)) {
3732 vm_map_unlock_read(map);
3740 /* If we would blow our VMEM resource limit, no go */
3741 if (map->size + grow_amount > vmemlim) {
3742 vm_map_unlock_read(map);
3749 if (racct_set(p, RACCT_VMEM, map->size + grow_amount)) {
3751 vm_map_unlock_read(map);
3759 if (vm_map_lock_upgrade(map))
3762 if (stack_entry == next_entry) {
3766 /* Get the preliminary new entry start value */
3767 addr = stack_entry->start - grow_amount;
3770 * If this puts us into the previous entry, cut back our
3771 * growth to the available space. Also, see the note above.
3774 stack_entry->avail_ssize = max_grow;
3776 if (stack_guard_page)
3780 rv = vm_map_insert(map, NULL, 0, addr, stack_entry->start,
3781 next_entry->protection, next_entry->max_protection,
3782 MAP_STACK_GROWS_DOWN);
3784 /* Adjust the available stack space by the amount we grew. */
3785 if (rv == KERN_SUCCESS) {
3786 new_entry = prev_entry->next;
3787 KASSERT(new_entry == stack_entry->prev, ("foo"));
3788 KASSERT(new_entry->end == stack_entry->start, ("foo"));
3789 KASSERT(new_entry->start == addr, ("foo"));
3790 KASSERT((new_entry->eflags & MAP_ENTRY_GROWS_DOWN) !=
3791 0, ("new entry lacks MAP_ENTRY_GROWS_DOWN"));
3792 grow_amount = new_entry->end - new_entry->start;
3793 new_entry->avail_ssize = stack_entry->avail_ssize -
3795 stack_entry->eflags &= ~MAP_ENTRY_GROWS_DOWN;
3801 addr = stack_entry->end + grow_amount;
3804 * If this puts us into the next entry, cut back our growth
3805 * to the available space. Also, see the note above.
3808 stack_entry->avail_ssize = end - stack_entry->end;
3810 if (stack_guard_page)
3814 grow_amount = addr - stack_entry->end;
3815 cred = stack_entry->cred;
3816 if (cred == NULL && stack_entry->object.vm_object != NULL)
3817 cred = stack_entry->object.vm_object->cred;
3818 if (cred != NULL && !swap_reserve_by_cred(grow_amount, cred))
3820 /* Grow the underlying object if applicable. */
3821 else if (stack_entry->object.vm_object == NULL ||
3822 vm_object_coalesce(stack_entry->object.vm_object,
3823 stack_entry->offset,
3824 (vm_size_t)(stack_entry->end - stack_entry->start),
3825 (vm_size_t)grow_amount, cred != NULL)) {
3826 map->size += (addr - stack_entry->end);
3827 /* Update the current entry. */
3828 stack_entry->end = addr;
3829 stack_entry->avail_ssize -= grow_amount;
3830 vm_map_entry_resize_free(map, stack_entry);
3833 if (next_entry != &map->header)
3834 vm_map_clip_start(map, next_entry, addr);
3839 if (rv == KERN_SUCCESS && is_procstack)
3840 vm->vm_ssize += btoc(grow_amount);
3845 * Heed the MAP_WIREFUTURE flag if it was set for this process.
3847 if (rv == KERN_SUCCESS && (map->flags & MAP_WIREFUTURE)) {
3849 (stack_entry == next_entry) ? addr : addr - grow_amount,
3850 (stack_entry == next_entry) ? stack_entry->start : addr,
3851 (p->p_flag & P_SYSTEM)
3852 ? VM_MAP_WIRE_SYSTEM|VM_MAP_WIRE_NOHOLES
3853 : VM_MAP_WIRE_USER|VM_MAP_WIRE_NOHOLES);
3858 if (racct_enable && rv != KERN_SUCCESS) {
3860 error = racct_set(p, RACCT_VMEM, map->size);
3861 KASSERT(error == 0, ("decreasing RACCT_VMEM failed"));
3863 error = racct_set(p, RACCT_MEMLOCK,
3864 ptoa(pmap_wired_count(map->pmap)));
3865 KASSERT(error == 0, ("decreasing RACCT_MEMLOCK failed"));
3867 error = racct_set(p, RACCT_STACK, ctob(vm->vm_ssize));
3868 KASSERT(error == 0, ("decreasing RACCT_STACK failed"));
3877 * Unshare the specified VM space for exec. If other processes are
3878 * mapped to it, then create a new one. The new vmspace is null.
3881 vmspace_exec(struct proc *p, vm_offset_t minuser, vm_offset_t maxuser)
3883 struct vmspace *oldvmspace = p->p_vmspace;
3884 struct vmspace *newvmspace;
3886 KASSERT((curthread->td_pflags & TDP_EXECVMSPC) == 0,
3887 ("vmspace_exec recursed"));
3888 newvmspace = vmspace_alloc(minuser, maxuser, NULL);
3889 if (newvmspace == NULL)
3891 newvmspace->vm_swrss = oldvmspace->vm_swrss;
3893 * This code is written like this for prototype purposes. The
3894 * goal is to avoid running down the vmspace here, but let the
3895 * other process's that are still using the vmspace to finally
3896 * run it down. Even though there is little or no chance of blocking
3897 * here, it is a good idea to keep this form for future mods.
3899 PROC_VMSPACE_LOCK(p);
3900 p->p_vmspace = newvmspace;
3901 PROC_VMSPACE_UNLOCK(p);
3902 if (p == curthread->td_proc)
3903 pmap_activate(curthread);
3904 curthread->td_pflags |= TDP_EXECVMSPC;
3909 * Unshare the specified VM space for forcing COW. This
3910 * is called by rfork, for the (RFMEM|RFPROC) == 0 case.
3913 vmspace_unshare(struct proc *p)
3915 struct vmspace *oldvmspace = p->p_vmspace;
3916 struct vmspace *newvmspace;
3917 vm_ooffset_t fork_charge;
3919 if (oldvmspace->vm_refcnt == 1)
3922 newvmspace = vmspace_fork(oldvmspace, &fork_charge);
3923 if (newvmspace == NULL)
3925 if (!swap_reserve_by_cred(fork_charge, p->p_ucred)) {
3926 vmspace_free(newvmspace);
3929 PROC_VMSPACE_LOCK(p);
3930 p->p_vmspace = newvmspace;
3931 PROC_VMSPACE_UNLOCK(p);
3932 if (p == curthread->td_proc)
3933 pmap_activate(curthread);
3934 vmspace_free(oldvmspace);
3941 * Finds the VM object, offset, and
3942 * protection for a given virtual address in the
3943 * specified map, assuming a page fault of the
3946 * Leaves the map in question locked for read; return
3947 * values are guaranteed until a vm_map_lookup_done
3948 * call is performed. Note that the map argument
3949 * is in/out; the returned map must be used in
3950 * the call to vm_map_lookup_done.
3952 * A handle (out_entry) is returned for use in
3953 * vm_map_lookup_done, to make that fast.
3955 * If a lookup is requested with "write protection"
3956 * specified, the map may be changed to perform virtual
3957 * copying operations, although the data referenced will
3961 vm_map_lookup(vm_map_t *var_map, /* IN/OUT */
3963 vm_prot_t fault_typea,
3964 vm_map_entry_t *out_entry, /* OUT */
3965 vm_object_t *object, /* OUT */
3966 vm_pindex_t *pindex, /* OUT */
3967 vm_prot_t *out_prot, /* OUT */
3968 boolean_t *wired) /* OUT */
3970 vm_map_entry_t entry;
3971 vm_map_t map = *var_map;
3973 vm_prot_t fault_type = fault_typea;
3974 vm_object_t eobject;
3980 vm_map_lock_read(map);
3983 * Lookup the faulting address.
3985 if (!vm_map_lookup_entry(map, vaddr, out_entry)) {
3986 vm_map_unlock_read(map);
3987 return (KERN_INVALID_ADDRESS);
3995 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
3996 vm_map_t old_map = map;
3998 *var_map = map = entry->object.sub_map;
3999 vm_map_unlock_read(old_map);
4004 * Check whether this task is allowed to have this page.
4006 prot = entry->protection;
4007 fault_type &= (VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE);
4008 if ((fault_type & prot) != fault_type || prot == VM_PROT_NONE) {
4009 vm_map_unlock_read(map);
4010 return (KERN_PROTECTION_FAILURE);
4012 KASSERT((prot & VM_PROT_WRITE) == 0 || (entry->eflags &
4013 (MAP_ENTRY_USER_WIRED | MAP_ENTRY_NEEDS_COPY)) !=
4014 (MAP_ENTRY_USER_WIRED | MAP_ENTRY_NEEDS_COPY),
4015 ("entry %p flags %x", entry, entry->eflags));
4016 if ((fault_typea & VM_PROT_COPY) != 0 &&
4017 (entry->max_protection & VM_PROT_WRITE) == 0 &&
4018 (entry->eflags & MAP_ENTRY_COW) == 0) {
4019 vm_map_unlock_read(map);
4020 return (KERN_PROTECTION_FAILURE);
4024 * If this page is not pageable, we have to get it for all possible
4027 *wired = (entry->wired_count != 0);
4029 fault_type = entry->protection;
4030 size = entry->end - entry->start;
4032 * If the entry was copy-on-write, we either ...
4034 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
4036 * If we want to write the page, we may as well handle that
4037 * now since we've got the map locked.
4039 * If we don't need to write the page, we just demote the
4040 * permissions allowed.
4042 if ((fault_type & VM_PROT_WRITE) != 0 ||
4043 (fault_typea & VM_PROT_COPY) != 0) {
4045 * Make a new object, and place it in the object
4046 * chain. Note that no new references have appeared
4047 * -- one just moved from the map to the new
4050 if (vm_map_lock_upgrade(map))
4053 if (entry->cred == NULL) {
4055 * The debugger owner is charged for
4058 cred = curthread->td_ucred;
4060 if (!swap_reserve_by_cred(size, cred)) {
4063 return (KERN_RESOURCE_SHORTAGE);
4067 vm_object_shadow(&entry->object.vm_object,
4068 &entry->offset, size);
4069 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
4070 eobject = entry->object.vm_object;
4071 if (eobject->cred != NULL) {
4073 * The object was not shadowed.
4075 swap_release_by_cred(size, entry->cred);
4076 crfree(entry->cred);
4078 } else if (entry->cred != NULL) {
4079 VM_OBJECT_WLOCK(eobject);
4080 eobject->cred = entry->cred;
4081 eobject->charge = size;
4082 VM_OBJECT_WUNLOCK(eobject);
4086 vm_map_lock_downgrade(map);
4089 * We're attempting to read a copy-on-write page --
4090 * don't allow writes.
4092 prot &= ~VM_PROT_WRITE;
4097 * Create an object if necessary.
4099 if (entry->object.vm_object == NULL &&
4101 if (vm_map_lock_upgrade(map))
4103 entry->object.vm_object = vm_object_allocate(OBJT_DEFAULT,
4106 if (entry->cred != NULL) {
4107 VM_OBJECT_WLOCK(entry->object.vm_object);
4108 entry->object.vm_object->cred = entry->cred;
4109 entry->object.vm_object->charge = size;
4110 VM_OBJECT_WUNLOCK(entry->object.vm_object);
4113 vm_map_lock_downgrade(map);
4117 * Return the object/offset from this entry. If the entry was
4118 * copy-on-write or empty, it has been fixed up.
4120 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
4121 *object = entry->object.vm_object;
4124 return (KERN_SUCCESS);
4128 * vm_map_lookup_locked:
4130 * Lookup the faulting address. A version of vm_map_lookup that returns
4131 * KERN_FAILURE instead of blocking on map lock or memory allocation.
4134 vm_map_lookup_locked(vm_map_t *var_map, /* IN/OUT */
4136 vm_prot_t fault_typea,
4137 vm_map_entry_t *out_entry, /* OUT */
4138 vm_object_t *object, /* OUT */
4139 vm_pindex_t *pindex, /* OUT */
4140 vm_prot_t *out_prot, /* OUT */
4141 boolean_t *wired) /* OUT */
4143 vm_map_entry_t entry;
4144 vm_map_t map = *var_map;
4146 vm_prot_t fault_type = fault_typea;
4149 * Lookup the faulting address.
4151 if (!vm_map_lookup_entry(map, vaddr, out_entry))
4152 return (KERN_INVALID_ADDRESS);
4157 * Fail if the entry refers to a submap.
4159 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP)
4160 return (KERN_FAILURE);
4163 * Check whether this task is allowed to have this page.
4165 prot = entry->protection;
4166 fault_type &= VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE;
4167 if ((fault_type & prot) != fault_type)
4168 return (KERN_PROTECTION_FAILURE);
4171 * If this page is not pageable, we have to get it for all possible
4174 *wired = (entry->wired_count != 0);
4176 fault_type = entry->protection;
4178 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
4180 * Fail if the entry was copy-on-write for a write fault.
4182 if (fault_type & VM_PROT_WRITE)
4183 return (KERN_FAILURE);
4185 * We're attempting to read a copy-on-write page --
4186 * don't allow writes.
4188 prot &= ~VM_PROT_WRITE;
4192 * Fail if an object should be created.
4194 if (entry->object.vm_object == NULL && !map->system_map)
4195 return (KERN_FAILURE);
4198 * Return the object/offset from this entry. If the entry was
4199 * copy-on-write or empty, it has been fixed up.
4201 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
4202 *object = entry->object.vm_object;
4205 return (KERN_SUCCESS);
4209 * vm_map_lookup_done:
4211 * Releases locks acquired by a vm_map_lookup
4212 * (according to the handle returned by that lookup).
4215 vm_map_lookup_done(vm_map_t map, vm_map_entry_t entry)
4218 * Unlock the main-level map
4220 vm_map_unlock_read(map);
4223 #include "opt_ddb.h"
4225 #include <sys/kernel.h>
4227 #include <ddb/ddb.h>
4230 vm_map_print(vm_map_t map)
4232 vm_map_entry_t entry;
4234 db_iprintf("Task map %p: pmap=%p, nentries=%d, version=%u\n",
4236 (void *)map->pmap, map->nentries, map->timestamp);
4239 for (entry = map->header.next; entry != &map->header;
4240 entry = entry->next) {
4241 db_iprintf("map entry %p: start=%p, end=%p\n",
4242 (void *)entry, (void *)entry->start, (void *)entry->end);
4244 static char *inheritance_name[4] =
4245 {"share", "copy", "none", "donate_copy"};
4247 db_iprintf(" prot=%x/%x/%s",
4249 entry->max_protection,
4250 inheritance_name[(int)(unsigned char)entry->inheritance]);
4251 if (entry->wired_count != 0)
4252 db_printf(", wired");
4254 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
4255 db_printf(", share=%p, offset=0x%jx\n",
4256 (void *)entry->object.sub_map,
4257 (uintmax_t)entry->offset);
4258 if ((entry->prev == &map->header) ||
4259 (entry->prev->object.sub_map !=
4260 entry->object.sub_map)) {
4262 vm_map_print((vm_map_t)entry->object.sub_map);
4266 if (entry->cred != NULL)
4267 db_printf(", ruid %d", entry->cred->cr_ruid);
4268 db_printf(", object=%p, offset=0x%jx",
4269 (void *)entry->object.vm_object,
4270 (uintmax_t)entry->offset);
4271 if (entry->object.vm_object && entry->object.vm_object->cred)
4272 db_printf(", obj ruid %d charge %jx",
4273 entry->object.vm_object->cred->cr_ruid,
4274 (uintmax_t)entry->object.vm_object->charge);
4275 if (entry->eflags & MAP_ENTRY_COW)
4276 db_printf(", copy (%s)",
4277 (entry->eflags & MAP_ENTRY_NEEDS_COPY) ? "needed" : "done");
4280 if ((entry->prev == &map->header) ||
4281 (entry->prev->object.vm_object !=
4282 entry->object.vm_object)) {
4284 vm_object_print((db_expr_t)(intptr_t)
4285 entry->object.vm_object,
4294 DB_SHOW_COMMAND(map, map)
4298 db_printf("usage: show map <addr>\n");
4301 vm_map_print((vm_map_t)addr);
4304 DB_SHOW_COMMAND(procvm, procvm)
4309 p = db_lookup_proc(addr);
4314 db_printf("p = %p, vmspace = %p, map = %p, pmap = %p\n",
4315 (void *)p, (void *)p->p_vmspace, (void *)&p->p_vmspace->vm_map,
4316 (void *)vmspace_pmap(p->p_vmspace));
4318 vm_map_print((vm_map_t)&p->p_vmspace->vm_map);