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);
136 static void vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot,
137 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags);
139 static void vm_map_zdtor(void *mem, int size, void *arg);
140 static void vmspace_zdtor(void *mem, int size, void *arg);
142 static int vm_map_stack_locked(vm_map_t map, vm_offset_t addrbos,
143 vm_size_t max_ssize, vm_size_t growsize, vm_prot_t prot, vm_prot_t max,
145 static void vm_map_wire_entry_failure(vm_map_t map, vm_map_entry_t entry,
146 vm_offset_t failed_addr);
148 #define ENTRY_CHARGED(e) ((e)->cred != NULL || \
149 ((e)->object.vm_object != NULL && (e)->object.vm_object->cred != NULL && \
150 !((e)->eflags & MAP_ENTRY_NEEDS_COPY)))
153 * PROC_VMSPACE_{UN,}LOCK() can be a noop as long as vmspaces are type
156 #define PROC_VMSPACE_LOCK(p) do { } while (0)
157 #define PROC_VMSPACE_UNLOCK(p) do { } while (0)
160 * VM_MAP_RANGE_CHECK: [ internal use only ]
162 * Asserts that the starting and ending region
163 * addresses fall within the valid range of the map.
165 #define VM_MAP_RANGE_CHECK(map, start, end) \
167 if (start < vm_map_min(map)) \
168 start = vm_map_min(map); \
169 if (end > vm_map_max(map)) \
170 end = vm_map_max(map); \
178 * Initialize the vm_map module. Must be called before
179 * any other vm_map routines.
181 * Map and entry structures are allocated from the general
182 * purpose memory pool with some exceptions:
184 * - The kernel map and kmem submap are allocated statically.
185 * - Kernel map entries are allocated out of a static pool.
187 * These restrictions are necessary since malloc() uses the
188 * maps and requires map entries.
194 mtx_init(&map_sleep_mtx, "vm map sleep mutex", NULL, MTX_DEF);
195 mapzone = uma_zcreate("MAP", sizeof(struct vm_map), NULL,
201 vm_map_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
202 uma_prealloc(mapzone, MAX_KMAP);
203 kmapentzone = uma_zcreate("KMAP ENTRY", sizeof(struct vm_map_entry),
204 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR,
205 UMA_ZONE_MTXCLASS | UMA_ZONE_VM);
206 mapentzone = uma_zcreate("MAP ENTRY", sizeof(struct vm_map_entry),
207 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
208 vmspace_zone = uma_zcreate("VMSPACE", sizeof(struct vmspace), NULL,
214 vmspace_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
218 vmspace_zinit(void *mem, int size, int flags)
222 vm = (struct vmspace *)mem;
224 vm->vm_map.pmap = NULL;
225 (void)vm_map_zinit(&vm->vm_map, sizeof(vm->vm_map), flags);
226 PMAP_LOCK_INIT(vmspace_pmap(vm));
231 vm_map_zinit(void *mem, int size, int flags)
236 memset(map, 0, sizeof(*map));
237 mtx_init(&map->system_mtx, "vm map (system)", NULL, MTX_DEF | MTX_DUPOK);
238 sx_init(&map->lock, "vm map (user)");
244 vmspace_zdtor(void *mem, int size, void *arg)
248 vm = (struct vmspace *)mem;
250 vm_map_zdtor(&vm->vm_map, sizeof(vm->vm_map), arg);
253 vm_map_zdtor(void *mem, int size, void *arg)
258 KASSERT(map->nentries == 0,
259 ("map %p nentries == %d on free.",
260 map, map->nentries));
261 KASSERT(map->size == 0,
262 ("map %p size == %lu on free.",
263 map, (unsigned long)map->size));
265 #endif /* INVARIANTS */
268 * Allocate a vmspace structure, including a vm_map and pmap,
269 * and initialize those structures. The refcnt is set to 1.
271 * If 'pinit' is NULL then the embedded pmap is initialized via pmap_pinit().
274 vmspace_alloc(vm_offset_t min, vm_offset_t max, pmap_pinit_t pinit)
278 vm = uma_zalloc(vmspace_zone, M_WAITOK);
280 KASSERT(vm->vm_map.pmap == NULL, ("vm_map.pmap must be NULL"));
285 if (!pinit(vmspace_pmap(vm))) {
286 uma_zfree(vmspace_zone, vm);
289 CTR1(KTR_VM, "vmspace_alloc: %p", vm);
290 _vm_map_init(&vm->vm_map, vmspace_pmap(vm), min, max);
304 vmspace_container_reset(struct proc *p)
309 racct_set(p, RACCT_DATA, 0);
310 racct_set(p, RACCT_STACK, 0);
311 racct_set(p, RACCT_RSS, 0);
312 racct_set(p, RACCT_MEMLOCK, 0);
313 racct_set(p, RACCT_VMEM, 0);
319 vmspace_dofree(struct vmspace *vm)
322 CTR1(KTR_VM, "vmspace_free: %p", vm);
325 * Make sure any SysV shm is freed, it might not have been in
331 * Lock the map, to wait out all other references to it.
332 * Delete all of the mappings and pages they hold, then call
333 * the pmap module to reclaim anything left.
335 (void)vm_map_remove(&vm->vm_map, vm->vm_map.min_offset,
336 vm->vm_map.max_offset);
338 pmap_release(vmspace_pmap(vm));
339 vm->vm_map.pmap = NULL;
340 uma_zfree(vmspace_zone, vm);
344 vmspace_free(struct vmspace *vm)
347 if (vm->vm_refcnt == 0)
348 panic("vmspace_free: attempt to free already freed vmspace");
350 if (atomic_fetchadd_int(&vm->vm_refcnt, -1) == 1)
355 vmspace_exitfree(struct proc *p)
359 PROC_VMSPACE_LOCK(p);
362 PROC_VMSPACE_UNLOCK(p);
363 KASSERT(vm == &vmspace0, ("vmspace_exitfree: wrong vmspace"));
368 vmspace_exit(struct thread *td)
375 * Release user portion of address space.
376 * This releases references to vnodes,
377 * which could cause I/O if the file has been unlinked.
378 * Need to do this early enough that we can still sleep.
380 * The last exiting process to reach this point releases as
381 * much of the environment as it can. vmspace_dofree() is the
382 * slower fallback in case another process had a temporary
383 * reference to the vmspace.
388 atomic_add_int(&vmspace0.vm_refcnt, 1);
390 refcnt = vm->vm_refcnt;
391 if (refcnt > 1 && p->p_vmspace != &vmspace0) {
392 /* Switch now since other proc might free vmspace */
393 PROC_VMSPACE_LOCK(p);
394 p->p_vmspace = &vmspace0;
395 PROC_VMSPACE_UNLOCK(p);
398 } while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt - 1));
400 if (p->p_vmspace != vm) {
401 /* vmspace not yet freed, switch back */
402 PROC_VMSPACE_LOCK(p);
404 PROC_VMSPACE_UNLOCK(p);
407 pmap_remove_pages(vmspace_pmap(vm));
408 /* Switch now since this proc will free vmspace */
409 PROC_VMSPACE_LOCK(p);
410 p->p_vmspace = &vmspace0;
411 PROC_VMSPACE_UNLOCK(p);
415 vmspace_container_reset(p);
418 /* Acquire reference to vmspace owned by another process. */
421 vmspace_acquire_ref(struct proc *p)
426 PROC_VMSPACE_LOCK(p);
429 PROC_VMSPACE_UNLOCK(p);
433 refcnt = vm->vm_refcnt;
434 if (refcnt <= 0) { /* Avoid 0->1 transition */
435 PROC_VMSPACE_UNLOCK(p);
438 } while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt + 1));
439 if (vm != p->p_vmspace) {
440 PROC_VMSPACE_UNLOCK(p);
444 PROC_VMSPACE_UNLOCK(p);
449 _vm_map_lock(vm_map_t map, const char *file, int line)
453 mtx_lock_flags_(&map->system_mtx, 0, file, line);
455 sx_xlock_(&map->lock, file, line);
460 vm_map_process_deferred(void)
463 vm_map_entry_t entry, next;
467 entry = td->td_map_def_user;
468 td->td_map_def_user = NULL;
469 while (entry != NULL) {
471 if ((entry->eflags & MAP_ENTRY_VN_WRITECNT) != 0) {
473 * Decrement the object's writemappings and
474 * possibly the vnode's v_writecount.
476 KASSERT((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0,
477 ("Submap with writecount"));
478 object = entry->object.vm_object;
479 KASSERT(object != NULL, ("No object for writecount"));
480 vnode_pager_release_writecount(object, entry->start,
483 vm_map_entry_deallocate(entry, FALSE);
489 _vm_map_unlock(vm_map_t map, const char *file, int line)
493 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
495 sx_xunlock_(&map->lock, file, line);
496 vm_map_process_deferred();
501 _vm_map_lock_read(vm_map_t map, const char *file, int line)
505 mtx_lock_flags_(&map->system_mtx, 0, file, line);
507 sx_slock_(&map->lock, file, line);
511 _vm_map_unlock_read(vm_map_t map, const char *file, int line)
515 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
517 sx_sunlock_(&map->lock, file, line);
518 vm_map_process_deferred();
523 _vm_map_trylock(vm_map_t map, const char *file, int line)
527 error = map->system_map ?
528 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) :
529 !sx_try_xlock_(&map->lock, file, line);
536 _vm_map_trylock_read(vm_map_t map, const char *file, int line)
540 error = map->system_map ?
541 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) :
542 !sx_try_slock_(&map->lock, file, line);
547 * _vm_map_lock_upgrade: [ internal use only ]
549 * Tries to upgrade a read (shared) lock on the specified map to a write
550 * (exclusive) lock. Returns the value "0" if the upgrade succeeds and a
551 * non-zero value if the upgrade fails. If the upgrade fails, the map is
552 * returned without a read or write lock held.
554 * Requires that the map be read locked.
557 _vm_map_lock_upgrade(vm_map_t map, const char *file, int line)
559 unsigned int last_timestamp;
561 if (map->system_map) {
562 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
564 if (!sx_try_upgrade_(&map->lock, file, line)) {
565 last_timestamp = map->timestamp;
566 sx_sunlock_(&map->lock, file, line);
567 vm_map_process_deferred();
569 * If the map's timestamp does not change while the
570 * map is unlocked, then the upgrade succeeds.
572 sx_xlock_(&map->lock, file, line);
573 if (last_timestamp != map->timestamp) {
574 sx_xunlock_(&map->lock, file, line);
584 _vm_map_lock_downgrade(vm_map_t map, const char *file, int line)
587 if (map->system_map) {
588 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
590 sx_downgrade_(&map->lock, file, line);
596 * Returns a non-zero value if the caller holds a write (exclusive) lock
597 * on the specified map and the value "0" otherwise.
600 vm_map_locked(vm_map_t map)
604 return (mtx_owned(&map->system_mtx));
606 return (sx_xlocked(&map->lock));
611 _vm_map_assert_locked(vm_map_t map, const char *file, int line)
615 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
617 sx_assert_(&map->lock, SA_XLOCKED, file, line);
620 #define VM_MAP_ASSERT_LOCKED(map) \
621 _vm_map_assert_locked(map, LOCK_FILE, LOCK_LINE)
623 #define VM_MAP_ASSERT_LOCKED(map)
627 * _vm_map_unlock_and_wait:
629 * Atomically releases the lock on the specified map and puts the calling
630 * thread to sleep. The calling thread will remain asleep until either
631 * vm_map_wakeup() is performed on the map or the specified timeout is
634 * WARNING! This function does not perform deferred deallocations of
635 * objects and map entries. Therefore, the calling thread is expected to
636 * reacquire the map lock after reawakening and later perform an ordinary
637 * unlock operation, such as vm_map_unlock(), before completing its
638 * operation on the map.
641 _vm_map_unlock_and_wait(vm_map_t map, int timo, const char *file, int line)
644 mtx_lock(&map_sleep_mtx);
646 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
648 sx_xunlock_(&map->lock, file, line);
649 return (msleep(&map->root, &map_sleep_mtx, PDROP | PVM, "vmmaps",
656 * Awaken any threads that have slept on the map using
657 * vm_map_unlock_and_wait().
660 vm_map_wakeup(vm_map_t map)
664 * Acquire and release map_sleep_mtx to prevent a wakeup()
665 * from being performed (and lost) between the map unlock
666 * and the msleep() in _vm_map_unlock_and_wait().
668 mtx_lock(&map_sleep_mtx);
669 mtx_unlock(&map_sleep_mtx);
674 vm_map_busy(vm_map_t map)
677 VM_MAP_ASSERT_LOCKED(map);
682 vm_map_unbusy(vm_map_t map)
685 VM_MAP_ASSERT_LOCKED(map);
686 KASSERT(map->busy, ("vm_map_unbusy: not busy"));
687 if (--map->busy == 0 && (map->flags & MAP_BUSY_WAKEUP)) {
688 vm_map_modflags(map, 0, MAP_BUSY_WAKEUP);
694 vm_map_wait_busy(vm_map_t map)
697 VM_MAP_ASSERT_LOCKED(map);
699 vm_map_modflags(map, MAP_BUSY_WAKEUP, 0);
701 msleep(&map->busy, &map->system_mtx, 0, "mbusy", 0);
703 sx_sleep(&map->busy, &map->lock, 0, "mbusy", 0);
709 vmspace_resident_count(struct vmspace *vmspace)
711 return pmap_resident_count(vmspace_pmap(vmspace));
717 * Creates and returns a new empty VM map with
718 * the given physical map structure, and having
719 * the given lower and upper address bounds.
722 vm_map_create(pmap_t pmap, vm_offset_t min, vm_offset_t max)
726 result = uma_zalloc(mapzone, M_WAITOK);
727 CTR1(KTR_VM, "vm_map_create: %p", result);
728 _vm_map_init(result, pmap, min, max);
733 * Initialize an existing vm_map structure
734 * such as that in the vmspace structure.
737 _vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max)
740 map->header.next = map->header.prev = &map->header;
741 map->needs_wakeup = FALSE;
744 map->min_offset = min;
745 map->max_offset = max;
753 vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max)
756 _vm_map_init(map, pmap, min, max);
757 mtx_init(&map->system_mtx, "system map", NULL, MTX_DEF | MTX_DUPOK);
758 sx_init(&map->lock, "user map");
762 * vm_map_entry_dispose: [ internal use only ]
764 * Inverse of vm_map_entry_create.
767 vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry)
769 uma_zfree(map->system_map ? kmapentzone : mapentzone, entry);
773 * vm_map_entry_create: [ internal use only ]
775 * Allocates a VM map entry for insertion.
776 * No entry fields are filled in.
778 static vm_map_entry_t
779 vm_map_entry_create(vm_map_t map)
781 vm_map_entry_t new_entry;
784 new_entry = uma_zalloc(kmapentzone, M_NOWAIT);
786 new_entry = uma_zalloc(mapentzone, M_WAITOK);
787 if (new_entry == NULL)
788 panic("vm_map_entry_create: kernel resources exhausted");
793 * vm_map_entry_set_behavior:
795 * Set the expected access behavior, either normal, random, or
799 vm_map_entry_set_behavior(vm_map_entry_t entry, u_char behavior)
801 entry->eflags = (entry->eflags & ~MAP_ENTRY_BEHAV_MASK) |
802 (behavior & MAP_ENTRY_BEHAV_MASK);
806 * vm_map_entry_set_max_free:
808 * Set the max_free field in a vm_map_entry.
811 vm_map_entry_set_max_free(vm_map_entry_t entry)
814 entry->max_free = entry->adj_free;
815 if (entry->left != NULL && entry->left->max_free > entry->max_free)
816 entry->max_free = entry->left->max_free;
817 if (entry->right != NULL && entry->right->max_free > entry->max_free)
818 entry->max_free = entry->right->max_free;
822 * vm_map_entry_splay:
824 * The Sleator and Tarjan top-down splay algorithm with the
825 * following variation. Max_free must be computed bottom-up, so
826 * on the downward pass, maintain the left and right spines in
827 * reverse order. Then, make a second pass up each side to fix
828 * the pointers and compute max_free. The time bound is O(log n)
831 * The new root is the vm_map_entry containing "addr", or else an
832 * adjacent entry (lower or higher) if addr is not in the tree.
834 * The map must be locked, and leaves it so.
836 * Returns: the new root.
838 static vm_map_entry_t
839 vm_map_entry_splay(vm_offset_t addr, vm_map_entry_t root)
841 vm_map_entry_t llist, rlist;
842 vm_map_entry_t ltree, rtree;
845 /* Special case of empty tree. */
850 * Pass One: Splay down the tree until we find addr or a NULL
851 * pointer where addr would go. llist and rlist are the two
852 * sides in reverse order (bottom-up), with llist linked by
853 * the right pointer and rlist linked by the left pointer in
854 * the vm_map_entry. Wait until Pass Two to set max_free on
860 /* root is never NULL in here. */
861 if (addr < root->start) {
865 if (addr < y->start && y->left != NULL) {
866 /* Rotate right and put y on rlist. */
867 root->left = y->right;
869 vm_map_entry_set_max_free(root);
874 /* Put root on rlist. */
879 } else if (addr >= root->end) {
883 if (addr >= y->end && y->right != NULL) {
884 /* Rotate left and put y on llist. */
885 root->right = y->left;
887 vm_map_entry_set_max_free(root);
892 /* Put root on llist. */
902 * Pass Two: Walk back up the two spines, flip the pointers
903 * and set max_free. The subtrees of the root go at the
904 * bottom of llist and rlist.
907 while (llist != NULL) {
909 llist->right = ltree;
910 vm_map_entry_set_max_free(llist);
915 while (rlist != NULL) {
918 vm_map_entry_set_max_free(rlist);
924 * Final assembly: add ltree and rtree as subtrees of root.
928 vm_map_entry_set_max_free(root);
934 * vm_map_entry_{un,}link:
936 * Insert/remove entries from maps.
939 vm_map_entry_link(vm_map_t map,
940 vm_map_entry_t after_where,
941 vm_map_entry_t entry)
945 "vm_map_entry_link: map %p, nentries %d, entry %p, after %p", map,
946 map->nentries, entry, after_where);
947 VM_MAP_ASSERT_LOCKED(map);
948 KASSERT(after_where == &map->header ||
949 after_where->end <= entry->start,
950 ("vm_map_entry_link: prev end %jx new start %jx overlap",
951 (uintmax_t)after_where->end, (uintmax_t)entry->start));
952 KASSERT(after_where->next == &map->header ||
953 entry->end <= after_where->next->start,
954 ("vm_map_entry_link: new end %jx next start %jx overlap",
955 (uintmax_t)entry->end, (uintmax_t)after_where->next->start));
958 entry->prev = after_where;
959 entry->next = after_where->next;
960 entry->next->prev = entry;
961 after_where->next = entry;
963 if (after_where != &map->header) {
964 if (after_where != map->root)
965 vm_map_entry_splay(after_where->start, map->root);
966 entry->right = after_where->right;
967 entry->left = after_where;
968 after_where->right = NULL;
969 after_where->adj_free = entry->start - after_where->end;
970 vm_map_entry_set_max_free(after_where);
972 entry->right = map->root;
975 entry->adj_free = (entry->next == &map->header ? map->max_offset :
976 entry->next->start) - entry->end;
977 vm_map_entry_set_max_free(entry);
982 vm_map_entry_unlink(vm_map_t map,
983 vm_map_entry_t entry)
985 vm_map_entry_t next, prev, root;
987 VM_MAP_ASSERT_LOCKED(map);
988 if (entry != map->root)
989 vm_map_entry_splay(entry->start, map->root);
990 if (entry->left == NULL)
993 root = vm_map_entry_splay(entry->start, entry->left);
994 root->right = entry->right;
995 root->adj_free = (entry->next == &map->header ? map->max_offset :
996 entry->next->start) - root->end;
997 vm_map_entry_set_max_free(root);
1006 CTR3(KTR_VM, "vm_map_entry_unlink: map %p, nentries %d, entry %p", map,
1007 map->nentries, entry);
1011 * vm_map_entry_resize_free:
1013 * Recompute the amount of free space following a vm_map_entry
1014 * and propagate that value up the tree. Call this function after
1015 * resizing a map entry in-place, that is, without a call to
1016 * vm_map_entry_link() or _unlink().
1018 * The map must be locked, and leaves it so.
1021 vm_map_entry_resize_free(vm_map_t map, vm_map_entry_t entry)
1025 * Using splay trees without parent pointers, propagating
1026 * max_free up the tree is done by moving the entry to the
1027 * root and making the change there.
1029 if (entry != map->root)
1030 map->root = vm_map_entry_splay(entry->start, map->root);
1032 entry->adj_free = (entry->next == &map->header ? map->max_offset :
1033 entry->next->start) - entry->end;
1034 vm_map_entry_set_max_free(entry);
1038 * vm_map_lookup_entry: [ internal use only ]
1040 * Finds the map entry containing (or
1041 * immediately preceding) the specified address
1042 * in the given map; the entry is returned
1043 * in the "entry" parameter. The boolean
1044 * result indicates whether the address is
1045 * actually contained in the map.
1048 vm_map_lookup_entry(
1050 vm_offset_t address,
1051 vm_map_entry_t *entry) /* OUT */
1057 * If the map is empty, then the map entry immediately preceding
1058 * "address" is the map's header.
1062 *entry = &map->header;
1063 else if (address >= cur->start && cur->end > address) {
1066 } else if ((locked = vm_map_locked(map)) ||
1067 sx_try_upgrade(&map->lock)) {
1069 * Splay requires a write lock on the map. However, it only
1070 * restructures the binary search tree; it does not otherwise
1071 * change the map. Thus, the map's timestamp need not change
1072 * on a temporary upgrade.
1074 map->root = cur = vm_map_entry_splay(address, cur);
1076 sx_downgrade(&map->lock);
1079 * If "address" is contained within a map entry, the new root
1080 * is that map entry. Otherwise, the new root is a map entry
1081 * immediately before or after "address".
1083 if (address >= cur->start) {
1085 if (cur->end > address)
1091 * Since the map is only locked for read access, perform a
1092 * standard binary search tree lookup for "address".
1095 if (address < cur->start) {
1096 if (cur->left == NULL) {
1101 } else if (cur->end > address) {
1105 if (cur->right == NULL) {
1118 * Inserts the given whole VM object into the target
1119 * map at the specified address range. The object's
1120 * size should match that of the address range.
1122 * Requires that the map be locked, and leaves it so.
1124 * If object is non-NULL, ref count must be bumped by caller
1125 * prior to making call to account for the new entry.
1128 vm_map_insert(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1129 vm_offset_t start, vm_offset_t end, vm_prot_t prot, vm_prot_t max, int cow)
1131 vm_map_entry_t new_entry, prev_entry, temp_entry;
1132 vm_eflags_t protoeflags;
1134 vm_inherit_t inheritance;
1136 VM_MAP_ASSERT_LOCKED(map);
1137 KASSERT((object != kmem_object && object != kernel_object) ||
1138 (cow & MAP_COPY_ON_WRITE) == 0,
1139 ("vm_map_insert: kmem or kernel object and COW"));
1140 KASSERT(object == NULL || (cow & MAP_NOFAULT) == 0,
1141 ("vm_map_insert: paradoxical MAP_NOFAULT request"));
1144 * Check that the start and end points are not bogus.
1146 if ((start < map->min_offset) || (end > map->max_offset) ||
1148 return (KERN_INVALID_ADDRESS);
1151 * Find the entry prior to the proposed starting address; if it's part
1152 * of an existing entry, this range is bogus.
1154 if (vm_map_lookup_entry(map, start, &temp_entry))
1155 return (KERN_NO_SPACE);
1157 prev_entry = temp_entry;
1160 * Assert that the next entry doesn't overlap the end point.
1162 if ((prev_entry->next != &map->header) &&
1163 (prev_entry->next->start < end))
1164 return (KERN_NO_SPACE);
1167 if (cow & MAP_COPY_ON_WRITE)
1168 protoeflags |= MAP_ENTRY_COW | MAP_ENTRY_NEEDS_COPY;
1169 if (cow & MAP_NOFAULT)
1170 protoeflags |= MAP_ENTRY_NOFAULT;
1171 if (cow & MAP_DISABLE_SYNCER)
1172 protoeflags |= MAP_ENTRY_NOSYNC;
1173 if (cow & MAP_DISABLE_COREDUMP)
1174 protoeflags |= MAP_ENTRY_NOCOREDUMP;
1175 if (cow & MAP_STACK_GROWS_DOWN)
1176 protoeflags |= MAP_ENTRY_GROWS_DOWN;
1177 if (cow & MAP_STACK_GROWS_UP)
1178 protoeflags |= MAP_ENTRY_GROWS_UP;
1179 if (cow & MAP_VN_WRITECOUNT)
1180 protoeflags |= MAP_ENTRY_VN_WRITECNT;
1181 if (cow & MAP_INHERIT_SHARE)
1182 inheritance = VM_INHERIT_SHARE;
1184 inheritance = VM_INHERIT_DEFAULT;
1187 if (cow & (MAP_ACC_NO_CHARGE | MAP_NOFAULT))
1189 if ((cow & MAP_ACC_CHARGED) || ((prot & VM_PROT_WRITE) &&
1190 ((protoeflags & MAP_ENTRY_NEEDS_COPY) || object == NULL))) {
1191 if (!(cow & MAP_ACC_CHARGED) && !swap_reserve(end - start))
1192 return (KERN_RESOURCE_SHORTAGE);
1193 KASSERT(object == NULL || (protoeflags & MAP_ENTRY_NEEDS_COPY) ||
1194 object->cred == NULL,
1195 ("OVERCOMMIT: vm_map_insert o %p", object));
1196 cred = curthread->td_ucred;
1200 /* Expand the kernel pmap, if necessary. */
1201 if (map == kernel_map && end > kernel_vm_end)
1202 pmap_growkernel(end);
1203 if (object != NULL) {
1205 * OBJ_ONEMAPPING must be cleared unless this mapping
1206 * is trivially proven to be the only mapping for any
1207 * of the object's pages. (Object granularity
1208 * reference counting is insufficient to recognize
1209 * aliases with precision.)
1211 VM_OBJECT_WLOCK(object);
1212 if (object->ref_count > 1 || object->shadow_count != 0)
1213 vm_object_clear_flag(object, OBJ_ONEMAPPING);
1214 VM_OBJECT_WUNLOCK(object);
1216 else if ((prev_entry != &map->header) &&
1217 (prev_entry->eflags == protoeflags) &&
1218 (cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 &&
1219 (prev_entry->end == start) &&
1220 (prev_entry->wired_count == 0) &&
1221 (prev_entry->cred == cred ||
1222 (prev_entry->object.vm_object != NULL &&
1223 (prev_entry->object.vm_object->cred == cred))) &&
1224 vm_object_coalesce(prev_entry->object.vm_object,
1226 (vm_size_t)(prev_entry->end - prev_entry->start),
1227 (vm_size_t)(end - prev_entry->end), cred != NULL &&
1228 (protoeflags & MAP_ENTRY_NEEDS_COPY) == 0)) {
1230 * We were able to extend the object. Determine if we
1231 * can extend the previous map entry to include the
1232 * new range as well.
1234 if ((prev_entry->inheritance == inheritance) &&
1235 (prev_entry->protection == prot) &&
1236 (prev_entry->max_protection == max)) {
1237 map->size += (end - prev_entry->end);
1238 prev_entry->end = end;
1239 vm_map_entry_resize_free(map, prev_entry);
1240 vm_map_simplify_entry(map, prev_entry);
1241 return (KERN_SUCCESS);
1245 * If we can extend the object but cannot extend the
1246 * map entry, we have to create a new map entry. We
1247 * must bump the ref count on the extended object to
1248 * account for it. object may be NULL.
1250 object = prev_entry->object.vm_object;
1251 offset = prev_entry->offset +
1252 (prev_entry->end - prev_entry->start);
1253 vm_object_reference(object);
1254 if (cred != NULL && object != NULL && object->cred != NULL &&
1255 !(prev_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
1256 /* Object already accounts for this uid. */
1264 * Create a new entry
1266 new_entry = vm_map_entry_create(map);
1267 new_entry->start = start;
1268 new_entry->end = end;
1269 new_entry->cred = NULL;
1271 new_entry->eflags = protoeflags;
1272 new_entry->object.vm_object = object;
1273 new_entry->offset = offset;
1274 new_entry->avail_ssize = 0;
1276 new_entry->inheritance = inheritance;
1277 new_entry->protection = prot;
1278 new_entry->max_protection = max;
1279 new_entry->wired_count = 0;
1280 new_entry->wiring_thread = NULL;
1281 new_entry->read_ahead = VM_FAULT_READ_AHEAD_INIT;
1282 new_entry->next_read = OFF_TO_IDX(offset);
1284 KASSERT(cred == NULL || !ENTRY_CHARGED(new_entry),
1285 ("OVERCOMMIT: vm_map_insert leaks vm_map %p", new_entry));
1286 new_entry->cred = cred;
1289 * Insert the new entry into the list
1291 vm_map_entry_link(map, prev_entry, new_entry);
1292 map->size += new_entry->end - new_entry->start;
1295 * Try to coalesce the new entry with both the previous and next
1296 * entries in the list. Previously, we only attempted to coalesce
1297 * with the previous entry when object is NULL. Here, we handle the
1298 * other cases, which are less common.
1300 vm_map_simplify_entry(map, new_entry);
1302 if (cow & (MAP_PREFAULT|MAP_PREFAULT_PARTIAL)) {
1303 vm_map_pmap_enter(map, start, prot,
1304 object, OFF_TO_IDX(offset), end - start,
1305 cow & MAP_PREFAULT_PARTIAL);
1308 return (KERN_SUCCESS);
1314 * Find the first fit (lowest VM address) for "length" free bytes
1315 * beginning at address >= start in the given map.
1317 * In a vm_map_entry, "adj_free" is the amount of free space
1318 * adjacent (higher address) to this entry, and "max_free" is the
1319 * maximum amount of contiguous free space in its subtree. This
1320 * allows finding a free region in one path down the tree, so
1321 * O(log n) amortized with splay trees.
1323 * The map must be locked, and leaves it so.
1325 * Returns: 0 on success, and starting address in *addr,
1326 * 1 if insufficient space.
1329 vm_map_findspace(vm_map_t map, vm_offset_t start, vm_size_t length,
1330 vm_offset_t *addr) /* OUT */
1332 vm_map_entry_t entry;
1336 * Request must fit within min/max VM address and must avoid
1339 if (start < map->min_offset)
1340 start = map->min_offset;
1341 if (start + length > map->max_offset || start + length < start)
1344 /* Empty tree means wide open address space. */
1345 if (map->root == NULL) {
1351 * After splay, if start comes before root node, then there
1352 * must be a gap from start to the root.
1354 map->root = vm_map_entry_splay(start, map->root);
1355 if (start + length <= map->root->start) {
1361 * Root is the last node that might begin its gap before
1362 * start, and this is the last comparison where address
1363 * wrap might be a problem.
1365 st = (start > map->root->end) ? start : map->root->end;
1366 if (length <= map->root->end + map->root->adj_free - st) {
1371 /* With max_free, can immediately tell if no solution. */
1372 entry = map->root->right;
1373 if (entry == NULL || length > entry->max_free)
1377 * Search the right subtree in the order: left subtree, root,
1378 * right subtree (first fit). The previous splay implies that
1379 * all regions in the right subtree have addresses > start.
1381 while (entry != NULL) {
1382 if (entry->left != NULL && entry->left->max_free >= length)
1383 entry = entry->left;
1384 else if (entry->adj_free >= length) {
1388 entry = entry->right;
1391 /* Can't get here, so panic if we do. */
1392 panic("vm_map_findspace: max_free corrupt");
1396 vm_map_fixed(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1397 vm_offset_t start, vm_size_t length, vm_prot_t prot,
1398 vm_prot_t max, int cow)
1403 end = start + length;
1404 KASSERT((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 ||
1406 ("vm_map_fixed: non-NULL backing object for stack"));
1408 VM_MAP_RANGE_CHECK(map, start, end);
1409 if ((cow & MAP_CHECK_EXCL) == 0)
1410 vm_map_delete(map, start, end);
1411 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) != 0) {
1412 result = vm_map_stack_locked(map, start, length, sgrowsiz,
1415 result = vm_map_insert(map, object, offset, start, end,
1423 * vm_map_find finds an unallocated region in the target address
1424 * map with the given length. The search is defined to be
1425 * first-fit from the specified address; the region found is
1426 * returned in the same parameter.
1428 * If object is non-NULL, ref count must be bumped by caller
1429 * prior to making call to account for the new entry.
1432 vm_map_find(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1433 vm_offset_t *addr, /* IN/OUT */
1434 vm_size_t length, vm_offset_t max_addr, int find_space,
1435 vm_prot_t prot, vm_prot_t max, int cow)
1437 vm_offset_t alignment, initial_addr, start;
1440 KASSERT((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 ||
1442 ("vm_map_find: non-NULL backing object for stack"));
1443 if (find_space == VMFS_OPTIMAL_SPACE && (object == NULL ||
1444 (object->flags & OBJ_COLORED) == 0))
1445 find_space = VMFS_ANY_SPACE;
1446 if (find_space >> 8 != 0) {
1447 KASSERT((find_space & 0xff) == 0, ("bad VMFS flags"));
1448 alignment = (vm_offset_t)1 << (find_space >> 8);
1451 initial_addr = *addr;
1453 start = initial_addr;
1456 if (find_space != VMFS_NO_SPACE) {
1457 if (vm_map_findspace(map, start, length, addr) ||
1458 (max_addr != 0 && *addr + length > max_addr)) {
1460 if (find_space == VMFS_OPTIMAL_SPACE) {
1461 find_space = VMFS_ANY_SPACE;
1464 return (KERN_NO_SPACE);
1466 switch (find_space) {
1467 case VMFS_SUPER_SPACE:
1468 case VMFS_OPTIMAL_SPACE:
1469 pmap_align_superpage(object, offset, addr,
1472 case VMFS_ANY_SPACE:
1475 if ((*addr & (alignment - 1)) != 0) {
1476 *addr &= ~(alignment - 1);
1484 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) != 0) {
1485 result = vm_map_stack_locked(map, start, length,
1486 sgrowsiz, prot, max, cow);
1488 result = vm_map_insert(map, object, offset, start,
1489 start + length, prot, max, cow);
1491 } while (result == KERN_NO_SPACE && find_space != VMFS_NO_SPACE &&
1492 find_space != VMFS_ANY_SPACE);
1498 * vm_map_simplify_entry:
1500 * Simplify the given map entry by merging with either neighbor. This
1501 * routine also has the ability to merge with both neighbors.
1503 * The map must be locked.
1505 * This routine guarentees that the passed entry remains valid (though
1506 * possibly extended). When merging, this routine may delete one or
1510 vm_map_simplify_entry(vm_map_t map, vm_map_entry_t entry)
1512 vm_map_entry_t next, prev;
1513 vm_size_t prevsize, esize;
1515 if ((entry->eflags & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP |
1516 MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_IS_SUB_MAP)) != 0)
1520 if (prev != &map->header) {
1521 prevsize = prev->end - prev->start;
1522 if ( (prev->end == entry->start) &&
1523 (prev->object.vm_object == entry->object.vm_object) &&
1524 (!prev->object.vm_object ||
1525 (prev->offset + prevsize == entry->offset)) &&
1526 (prev->eflags == entry->eflags) &&
1527 (prev->protection == entry->protection) &&
1528 (prev->max_protection == entry->max_protection) &&
1529 (prev->inheritance == entry->inheritance) &&
1530 (prev->wired_count == entry->wired_count) &&
1531 (prev->cred == entry->cred)) {
1532 vm_map_entry_unlink(map, prev);
1533 entry->start = prev->start;
1534 entry->offset = prev->offset;
1535 if (entry->prev != &map->header)
1536 vm_map_entry_resize_free(map, entry->prev);
1539 * If the backing object is a vnode object,
1540 * vm_object_deallocate() calls vrele().
1541 * However, vrele() does not lock the vnode
1542 * because the vnode has additional
1543 * references. Thus, the map lock can be kept
1544 * without causing a lock-order reversal with
1547 * Since we count the number of virtual page
1548 * mappings in object->un_pager.vnp.writemappings,
1549 * the writemappings value should not be adjusted
1550 * when the entry is disposed of.
1552 if (prev->object.vm_object)
1553 vm_object_deallocate(prev->object.vm_object);
1554 if (prev->cred != NULL)
1556 vm_map_entry_dispose(map, prev);
1561 if (next != &map->header) {
1562 esize = entry->end - entry->start;
1563 if ((entry->end == next->start) &&
1564 (next->object.vm_object == entry->object.vm_object) &&
1565 (!entry->object.vm_object ||
1566 (entry->offset + esize == next->offset)) &&
1567 (next->eflags == entry->eflags) &&
1568 (next->protection == entry->protection) &&
1569 (next->max_protection == entry->max_protection) &&
1570 (next->inheritance == entry->inheritance) &&
1571 (next->wired_count == entry->wired_count) &&
1572 (next->cred == entry->cred)) {
1573 vm_map_entry_unlink(map, next);
1574 entry->end = next->end;
1575 vm_map_entry_resize_free(map, entry);
1578 * See comment above.
1580 if (next->object.vm_object)
1581 vm_object_deallocate(next->object.vm_object);
1582 if (next->cred != NULL)
1584 vm_map_entry_dispose(map, next);
1589 * vm_map_clip_start: [ internal use only ]
1591 * Asserts that the given entry begins at or after
1592 * the specified address; if necessary,
1593 * it splits the entry into two.
1595 #define vm_map_clip_start(map, entry, startaddr) \
1597 if (startaddr > entry->start) \
1598 _vm_map_clip_start(map, entry, startaddr); \
1602 * This routine is called only when it is known that
1603 * the entry must be split.
1606 _vm_map_clip_start(vm_map_t map, vm_map_entry_t entry, vm_offset_t start)
1608 vm_map_entry_t new_entry;
1610 VM_MAP_ASSERT_LOCKED(map);
1613 * Split off the front portion -- note that we must insert the new
1614 * entry BEFORE this one, so that this entry has the specified
1617 vm_map_simplify_entry(map, entry);
1620 * If there is no object backing this entry, we might as well create
1621 * one now. If we defer it, an object can get created after the map
1622 * is clipped, and individual objects will be created for the split-up
1623 * map. This is a bit of a hack, but is also about the best place to
1624 * put this improvement.
1626 if (entry->object.vm_object == NULL && !map->system_map) {
1628 object = vm_object_allocate(OBJT_DEFAULT,
1629 atop(entry->end - entry->start));
1630 entry->object.vm_object = object;
1632 if (entry->cred != NULL) {
1633 object->cred = entry->cred;
1634 object->charge = entry->end - entry->start;
1637 } else if (entry->object.vm_object != NULL &&
1638 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) &&
1639 entry->cred != NULL) {
1640 VM_OBJECT_WLOCK(entry->object.vm_object);
1641 KASSERT(entry->object.vm_object->cred == NULL,
1642 ("OVERCOMMIT: vm_entry_clip_start: both cred e %p", entry));
1643 entry->object.vm_object->cred = entry->cred;
1644 entry->object.vm_object->charge = entry->end - entry->start;
1645 VM_OBJECT_WUNLOCK(entry->object.vm_object);
1649 new_entry = vm_map_entry_create(map);
1650 *new_entry = *entry;
1652 new_entry->end = start;
1653 entry->offset += (start - entry->start);
1654 entry->start = start;
1655 if (new_entry->cred != NULL)
1656 crhold(entry->cred);
1658 vm_map_entry_link(map, entry->prev, new_entry);
1660 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1661 vm_object_reference(new_entry->object.vm_object);
1663 * The object->un_pager.vnp.writemappings for the
1664 * object of MAP_ENTRY_VN_WRITECNT type entry shall be
1665 * kept as is here. The virtual pages are
1666 * re-distributed among the clipped entries, so the sum is
1673 * vm_map_clip_end: [ internal use only ]
1675 * Asserts that the given entry ends at or before
1676 * the specified address; if necessary,
1677 * it splits the entry into two.
1679 #define vm_map_clip_end(map, entry, endaddr) \
1681 if ((endaddr) < (entry->end)) \
1682 _vm_map_clip_end((map), (entry), (endaddr)); \
1686 * This routine is called only when it is known that
1687 * the entry must be split.
1690 _vm_map_clip_end(vm_map_t map, vm_map_entry_t entry, vm_offset_t end)
1692 vm_map_entry_t new_entry;
1694 VM_MAP_ASSERT_LOCKED(map);
1697 * If there is no object backing this entry, we might as well create
1698 * one now. If we defer it, an object can get created after the map
1699 * is clipped, and individual objects will be created for the split-up
1700 * map. This is a bit of a hack, but is also about the best place to
1701 * put this improvement.
1703 if (entry->object.vm_object == NULL && !map->system_map) {
1705 object = vm_object_allocate(OBJT_DEFAULT,
1706 atop(entry->end - entry->start));
1707 entry->object.vm_object = object;
1709 if (entry->cred != NULL) {
1710 object->cred = entry->cred;
1711 object->charge = entry->end - entry->start;
1714 } else if (entry->object.vm_object != NULL &&
1715 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) &&
1716 entry->cred != NULL) {
1717 VM_OBJECT_WLOCK(entry->object.vm_object);
1718 KASSERT(entry->object.vm_object->cred == NULL,
1719 ("OVERCOMMIT: vm_entry_clip_end: both cred e %p", entry));
1720 entry->object.vm_object->cred = entry->cred;
1721 entry->object.vm_object->charge = entry->end - entry->start;
1722 VM_OBJECT_WUNLOCK(entry->object.vm_object);
1727 * Create a new entry and insert it AFTER the specified entry
1729 new_entry = vm_map_entry_create(map);
1730 *new_entry = *entry;
1732 new_entry->start = entry->end = end;
1733 new_entry->offset += (end - entry->start);
1734 if (new_entry->cred != NULL)
1735 crhold(entry->cred);
1737 vm_map_entry_link(map, entry, new_entry);
1739 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1740 vm_object_reference(new_entry->object.vm_object);
1745 * vm_map_submap: [ kernel use only ]
1747 * Mark the given range as handled by a subordinate map.
1749 * This range must have been created with vm_map_find,
1750 * and no other operations may have been performed on this
1751 * range prior to calling vm_map_submap.
1753 * Only a limited number of operations can be performed
1754 * within this rage after calling vm_map_submap:
1756 * [Don't try vm_map_copy!]
1758 * To remove a submapping, one must first remove the
1759 * range from the superior map, and then destroy the
1760 * submap (if desired). [Better yet, don't try it.]
1769 vm_map_entry_t entry;
1770 int result = KERN_INVALID_ARGUMENT;
1774 VM_MAP_RANGE_CHECK(map, start, end);
1776 if (vm_map_lookup_entry(map, start, &entry)) {
1777 vm_map_clip_start(map, entry, start);
1779 entry = entry->next;
1781 vm_map_clip_end(map, entry, end);
1783 if ((entry->start == start) && (entry->end == end) &&
1784 ((entry->eflags & MAP_ENTRY_COW) == 0) &&
1785 (entry->object.vm_object == NULL)) {
1786 entry->object.sub_map = submap;
1787 entry->eflags |= MAP_ENTRY_IS_SUB_MAP;
1788 result = KERN_SUCCESS;
1796 * The maximum number of pages to map if MAP_PREFAULT_PARTIAL is specified
1798 #define MAX_INIT_PT 96
1801 * vm_map_pmap_enter:
1803 * Preload the specified map's pmap with mappings to the specified
1804 * object's memory-resident pages. No further physical pages are
1805 * allocated, and no further virtual pages are retrieved from secondary
1806 * storage. If the specified flags include MAP_PREFAULT_PARTIAL, then a
1807 * limited number of page mappings are created at the low-end of the
1808 * specified address range. (For this purpose, a superpage mapping
1809 * counts as one page mapping.) Otherwise, all resident pages within
1810 * the specified address range are mapped. Because these mappings are
1811 * being created speculatively, cached pages are not reactivated and
1815 vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot,
1816 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags)
1819 vm_page_t p, p_start;
1820 vm_pindex_t mask, psize, threshold, tmpidx;
1822 if ((prot & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0 || object == NULL)
1824 VM_OBJECT_RLOCK(object);
1825 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) {
1826 VM_OBJECT_RUNLOCK(object);
1827 VM_OBJECT_WLOCK(object);
1828 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) {
1829 pmap_object_init_pt(map->pmap, addr, object, pindex,
1831 VM_OBJECT_WUNLOCK(object);
1834 VM_OBJECT_LOCK_DOWNGRADE(object);
1838 if (psize + pindex > object->size) {
1839 if (object->size < pindex) {
1840 VM_OBJECT_RUNLOCK(object);
1843 psize = object->size - pindex;
1848 threshold = MAX_INIT_PT;
1850 p = vm_page_find_least(object, pindex);
1852 * Assert: the variable p is either (1) the page with the
1853 * least pindex greater than or equal to the parameter pindex
1857 p != NULL && (tmpidx = p->pindex - pindex) < psize;
1858 p = TAILQ_NEXT(p, listq)) {
1860 * don't allow an madvise to blow away our really
1861 * free pages allocating pv entries.
1863 if (((flags & MAP_PREFAULT_MADVISE) != 0 &&
1864 vm_cnt.v_free_count < vm_cnt.v_free_reserved) ||
1865 ((flags & MAP_PREFAULT_PARTIAL) != 0 &&
1866 tmpidx >= threshold)) {
1870 if (p->valid == VM_PAGE_BITS_ALL) {
1871 if (p_start == NULL) {
1872 start = addr + ptoa(tmpidx);
1875 /* Jump ahead if a superpage mapping is possible. */
1876 if (p->psind > 0 && ((addr + ptoa(tmpidx)) &
1877 (pagesizes[p->psind] - 1)) == 0) {
1878 mask = atop(pagesizes[p->psind]) - 1;
1879 if (tmpidx + mask < psize &&
1880 vm_page_ps_is_valid(p)) {
1885 } else if (p_start != NULL) {
1886 pmap_enter_object(map->pmap, start, addr +
1887 ptoa(tmpidx), p_start, prot);
1891 if (p_start != NULL)
1892 pmap_enter_object(map->pmap, start, addr + ptoa(psize),
1894 VM_OBJECT_RUNLOCK(object);
1900 * Sets the protection of the specified address
1901 * region in the target map. If "set_max" is
1902 * specified, the maximum protection is to be set;
1903 * otherwise, only the current protection is affected.
1906 vm_map_protect(vm_map_t map, vm_offset_t start, vm_offset_t end,
1907 vm_prot_t new_prot, boolean_t set_max)
1909 vm_map_entry_t current, entry;
1915 return (KERN_SUCCESS);
1919 VM_MAP_RANGE_CHECK(map, start, end);
1921 if (vm_map_lookup_entry(map, start, &entry)) {
1922 vm_map_clip_start(map, entry, start);
1924 entry = entry->next;
1928 * Make a first pass to check for protection violations.
1931 while ((current != &map->header) && (current->start < end)) {
1932 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
1934 return (KERN_INVALID_ARGUMENT);
1936 if ((new_prot & current->max_protection) != new_prot) {
1938 return (KERN_PROTECTION_FAILURE);
1940 current = current->next;
1945 * Do an accounting pass for private read-only mappings that
1946 * now will do cow due to allowed write (e.g. debugger sets
1947 * breakpoint on text segment)
1949 for (current = entry; (current != &map->header) &&
1950 (current->start < end); current = current->next) {
1952 vm_map_clip_end(map, current, end);
1955 ((new_prot & ~(current->protection)) & VM_PROT_WRITE) == 0 ||
1956 ENTRY_CHARGED(current)) {
1960 cred = curthread->td_ucred;
1961 obj = current->object.vm_object;
1963 if (obj == NULL || (current->eflags & MAP_ENTRY_NEEDS_COPY)) {
1964 if (!swap_reserve(current->end - current->start)) {
1966 return (KERN_RESOURCE_SHORTAGE);
1969 current->cred = cred;
1973 VM_OBJECT_WLOCK(obj);
1974 if (obj->type != OBJT_DEFAULT && obj->type != OBJT_SWAP) {
1975 VM_OBJECT_WUNLOCK(obj);
1980 * Charge for the whole object allocation now, since
1981 * we cannot distinguish between non-charged and
1982 * charged clipped mapping of the same object later.
1984 KASSERT(obj->charge == 0,
1985 ("vm_map_protect: object %p overcharged (entry %p)",
1987 if (!swap_reserve(ptoa(obj->size))) {
1988 VM_OBJECT_WUNLOCK(obj);
1990 return (KERN_RESOURCE_SHORTAGE);
1995 obj->charge = ptoa(obj->size);
1996 VM_OBJECT_WUNLOCK(obj);
2000 * Go back and fix up protections. [Note that clipping is not
2001 * necessary the second time.]
2004 while ((current != &map->header) && (current->start < end)) {
2005 old_prot = current->protection;
2008 current->protection =
2009 (current->max_protection = new_prot) &
2012 current->protection = new_prot;
2015 * For user wired map entries, the normal lazy evaluation of
2016 * write access upgrades through soft page faults is
2017 * undesirable. Instead, immediately copy any pages that are
2018 * copy-on-write and enable write access in the physical map.
2020 if ((current->eflags & MAP_ENTRY_USER_WIRED) != 0 &&
2021 (current->protection & VM_PROT_WRITE) != 0 &&
2022 (old_prot & VM_PROT_WRITE) == 0)
2023 vm_fault_copy_entry(map, map, current, current, NULL);
2026 * When restricting access, update the physical map. Worry
2027 * about copy-on-write here.
2029 if ((old_prot & ~current->protection) != 0) {
2030 #define MASK(entry) (((entry)->eflags & MAP_ENTRY_COW) ? ~VM_PROT_WRITE : \
2032 pmap_protect(map->pmap, current->start,
2034 current->protection & MASK(current));
2037 vm_map_simplify_entry(map, current);
2038 current = current->next;
2041 return (KERN_SUCCESS);
2047 * This routine traverses a processes map handling the madvise
2048 * system call. Advisories are classified as either those effecting
2049 * the vm_map_entry structure, or those effecting the underlying
2059 vm_map_entry_t current, entry;
2063 * Some madvise calls directly modify the vm_map_entry, in which case
2064 * we need to use an exclusive lock on the map and we need to perform
2065 * various clipping operations. Otherwise we only need a read-lock
2070 case MADV_SEQUENTIAL:
2077 return (KERN_SUCCESS);
2085 return (KERN_SUCCESS);
2086 vm_map_lock_read(map);
2089 return (KERN_INVALID_ARGUMENT);
2093 * Locate starting entry and clip if necessary.
2095 VM_MAP_RANGE_CHECK(map, start, end);
2097 if (vm_map_lookup_entry(map, start, &entry)) {
2099 vm_map_clip_start(map, entry, start);
2101 entry = entry->next;
2106 * madvise behaviors that are implemented in the vm_map_entry.
2108 * We clip the vm_map_entry so that behavioral changes are
2109 * limited to the specified address range.
2111 for (current = entry;
2112 (current != &map->header) && (current->start < end);
2113 current = current->next
2115 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
2118 vm_map_clip_end(map, current, end);
2122 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_NORMAL);
2124 case MADV_SEQUENTIAL:
2125 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_SEQUENTIAL);
2128 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_RANDOM);
2131 current->eflags |= MAP_ENTRY_NOSYNC;
2134 current->eflags &= ~MAP_ENTRY_NOSYNC;
2137 current->eflags |= MAP_ENTRY_NOCOREDUMP;
2140 current->eflags &= ~MAP_ENTRY_NOCOREDUMP;
2145 vm_map_simplify_entry(map, current);
2149 vm_pindex_t pstart, pend;
2152 * madvise behaviors that are implemented in the underlying
2155 * Since we don't clip the vm_map_entry, we have to clip
2156 * the vm_object pindex and count.
2158 for (current = entry;
2159 (current != &map->header) && (current->start < end);
2160 current = current->next
2162 vm_offset_t useEnd, useStart;
2164 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
2167 pstart = OFF_TO_IDX(current->offset);
2168 pend = pstart + atop(current->end - current->start);
2169 useStart = current->start;
2170 useEnd = current->end;
2172 if (current->start < start) {
2173 pstart += atop(start - current->start);
2176 if (current->end > end) {
2177 pend -= atop(current->end - end);
2185 * Perform the pmap_advise() before clearing
2186 * PGA_REFERENCED in vm_page_advise(). Otherwise, a
2187 * concurrent pmap operation, such as pmap_remove(),
2188 * could clear a reference in the pmap and set
2189 * PGA_REFERENCED on the page before the pmap_advise()
2190 * had completed. Consequently, the page would appear
2191 * referenced based upon an old reference that
2192 * occurred before this pmap_advise() ran.
2194 if (behav == MADV_DONTNEED || behav == MADV_FREE)
2195 pmap_advise(map->pmap, useStart, useEnd,
2198 vm_object_madvise(current->object.vm_object, pstart,
2202 * Pre-populate paging structures in the
2203 * WILLNEED case. For wired entries, the
2204 * paging structures are already populated.
2206 if (behav == MADV_WILLNEED &&
2207 current->wired_count == 0) {
2208 vm_map_pmap_enter(map,
2210 current->protection,
2211 current->object.vm_object,
2213 ptoa(pend - pstart),
2214 MAP_PREFAULT_MADVISE
2218 vm_map_unlock_read(map);
2227 * Sets the inheritance of the specified address
2228 * range in the target map. Inheritance
2229 * affects how the map will be shared with
2230 * child maps at the time of vmspace_fork.
2233 vm_map_inherit(vm_map_t map, vm_offset_t start, vm_offset_t end,
2234 vm_inherit_t new_inheritance)
2236 vm_map_entry_t entry;
2237 vm_map_entry_t temp_entry;
2239 switch (new_inheritance) {
2240 case VM_INHERIT_NONE:
2241 case VM_INHERIT_COPY:
2242 case VM_INHERIT_SHARE:
2245 return (KERN_INVALID_ARGUMENT);
2248 return (KERN_SUCCESS);
2250 VM_MAP_RANGE_CHECK(map, start, end);
2251 if (vm_map_lookup_entry(map, start, &temp_entry)) {
2253 vm_map_clip_start(map, entry, start);
2255 entry = temp_entry->next;
2256 while ((entry != &map->header) && (entry->start < end)) {
2257 vm_map_clip_end(map, entry, end);
2258 entry->inheritance = new_inheritance;
2259 vm_map_simplify_entry(map, entry);
2260 entry = entry->next;
2263 return (KERN_SUCCESS);
2269 * Implements both kernel and user unwiring.
2272 vm_map_unwire(vm_map_t map, vm_offset_t start, vm_offset_t end,
2275 vm_map_entry_t entry, first_entry, tmp_entry;
2276 vm_offset_t saved_start;
2277 unsigned int last_timestamp;
2279 boolean_t need_wakeup, result, user_unwire;
2282 return (KERN_SUCCESS);
2283 user_unwire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE;
2285 VM_MAP_RANGE_CHECK(map, start, end);
2286 if (!vm_map_lookup_entry(map, start, &first_entry)) {
2287 if (flags & VM_MAP_WIRE_HOLESOK)
2288 first_entry = first_entry->next;
2291 return (KERN_INVALID_ADDRESS);
2294 last_timestamp = map->timestamp;
2295 entry = first_entry;
2296 while (entry != &map->header && entry->start < end) {
2297 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
2299 * We have not yet clipped the entry.
2301 saved_start = (start >= entry->start) ? start :
2303 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2304 if (vm_map_unlock_and_wait(map, 0)) {
2306 * Allow interruption of user unwiring?
2310 if (last_timestamp+1 != map->timestamp) {
2312 * Look again for the entry because the map was
2313 * modified while it was unlocked.
2314 * Specifically, the entry may have been
2315 * clipped, merged, or deleted.
2317 if (!vm_map_lookup_entry(map, saved_start,
2319 if (flags & VM_MAP_WIRE_HOLESOK)
2320 tmp_entry = tmp_entry->next;
2322 if (saved_start == start) {
2324 * First_entry has been deleted.
2327 return (KERN_INVALID_ADDRESS);
2330 rv = KERN_INVALID_ADDRESS;
2334 if (entry == first_entry)
2335 first_entry = tmp_entry;
2340 last_timestamp = map->timestamp;
2343 vm_map_clip_start(map, entry, start);
2344 vm_map_clip_end(map, entry, end);
2346 * Mark the entry in case the map lock is released. (See
2349 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 &&
2350 entry->wiring_thread == NULL,
2351 ("owned map entry %p", entry));
2352 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
2353 entry->wiring_thread = curthread;
2355 * Check the map for holes in the specified region.
2356 * If VM_MAP_WIRE_HOLESOK was specified, skip this check.
2358 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) &&
2359 (entry->end < end && (entry->next == &map->header ||
2360 entry->next->start > entry->end))) {
2362 rv = KERN_INVALID_ADDRESS;
2366 * If system unwiring, require that the entry is system wired.
2369 vm_map_entry_system_wired_count(entry) == 0) {
2371 rv = KERN_INVALID_ARGUMENT;
2374 entry = entry->next;
2378 need_wakeup = FALSE;
2379 if (first_entry == NULL) {
2380 result = vm_map_lookup_entry(map, start, &first_entry);
2381 if (!result && (flags & VM_MAP_WIRE_HOLESOK))
2382 first_entry = first_entry->next;
2384 KASSERT(result, ("vm_map_unwire: lookup failed"));
2386 for (entry = first_entry; entry != &map->header && entry->start < end;
2387 entry = entry->next) {
2389 * If VM_MAP_WIRE_HOLESOK was specified, an empty
2390 * space in the unwired region could have been mapped
2391 * while the map lock was dropped for draining
2392 * MAP_ENTRY_IN_TRANSITION. Moreover, another thread
2393 * could be simultaneously wiring this new mapping
2394 * entry. Detect these cases and skip any entries
2395 * marked as in transition by us.
2397 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 ||
2398 entry->wiring_thread != curthread) {
2399 KASSERT((flags & VM_MAP_WIRE_HOLESOK) != 0,
2400 ("vm_map_unwire: !HOLESOK and new/changed entry"));
2404 if (rv == KERN_SUCCESS && (!user_unwire ||
2405 (entry->eflags & MAP_ENTRY_USER_WIRED))) {
2407 entry->eflags &= ~MAP_ENTRY_USER_WIRED;
2408 if (entry->wired_count == 1)
2409 vm_map_entry_unwire(map, entry);
2411 entry->wired_count--;
2413 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0,
2414 ("vm_map_unwire: in-transition flag missing %p", entry));
2415 KASSERT(entry->wiring_thread == curthread,
2416 ("vm_map_unwire: alien wire %p", entry));
2417 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION;
2418 entry->wiring_thread = NULL;
2419 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
2420 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
2423 vm_map_simplify_entry(map, entry);
2432 * vm_map_wire_entry_failure:
2434 * Handle a wiring failure on the given entry.
2436 * The map should be locked.
2439 vm_map_wire_entry_failure(vm_map_t map, vm_map_entry_t entry,
2440 vm_offset_t failed_addr)
2443 VM_MAP_ASSERT_LOCKED(map);
2444 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 &&
2445 entry->wired_count == 1,
2446 ("vm_map_wire_entry_failure: entry %p isn't being wired", entry));
2447 KASSERT(failed_addr < entry->end,
2448 ("vm_map_wire_entry_failure: entry %p was fully wired", entry));
2451 * If any pages at the start of this entry were successfully wired,
2454 if (failed_addr > entry->start) {
2455 pmap_unwire(map->pmap, entry->start, failed_addr);
2456 vm_object_unwire(entry->object.vm_object, entry->offset,
2457 failed_addr - entry->start, PQ_ACTIVE);
2461 * Assign an out-of-range value to represent the failure to wire this
2464 entry->wired_count = -1;
2470 * Implements both kernel and user wiring.
2473 vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t end,
2476 vm_map_entry_t entry, first_entry, tmp_entry;
2477 vm_offset_t faddr, saved_end, saved_start;
2478 unsigned int last_timestamp;
2480 boolean_t need_wakeup, result, user_wire;
2484 return (KERN_SUCCESS);
2486 if (flags & VM_MAP_WIRE_WRITE)
2487 prot |= VM_PROT_WRITE;
2488 user_wire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE;
2490 VM_MAP_RANGE_CHECK(map, start, end);
2491 if (!vm_map_lookup_entry(map, start, &first_entry)) {
2492 if (flags & VM_MAP_WIRE_HOLESOK)
2493 first_entry = first_entry->next;
2496 return (KERN_INVALID_ADDRESS);
2499 last_timestamp = map->timestamp;
2500 entry = first_entry;
2501 while (entry != &map->header && entry->start < end) {
2502 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
2504 * We have not yet clipped the entry.
2506 saved_start = (start >= entry->start) ? start :
2508 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2509 if (vm_map_unlock_and_wait(map, 0)) {
2511 * Allow interruption of user wiring?
2515 if (last_timestamp + 1 != map->timestamp) {
2517 * Look again for the entry because the map was
2518 * modified while it was unlocked.
2519 * Specifically, the entry may have been
2520 * clipped, merged, or deleted.
2522 if (!vm_map_lookup_entry(map, saved_start,
2524 if (flags & VM_MAP_WIRE_HOLESOK)
2525 tmp_entry = tmp_entry->next;
2527 if (saved_start == start) {
2529 * first_entry has been deleted.
2532 return (KERN_INVALID_ADDRESS);
2535 rv = KERN_INVALID_ADDRESS;
2539 if (entry == first_entry)
2540 first_entry = tmp_entry;
2545 last_timestamp = map->timestamp;
2548 vm_map_clip_start(map, entry, start);
2549 vm_map_clip_end(map, entry, end);
2551 * Mark the entry in case the map lock is released. (See
2554 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 &&
2555 entry->wiring_thread == NULL,
2556 ("owned map entry %p", entry));
2557 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
2558 entry->wiring_thread = curthread;
2559 if ((entry->protection & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0
2560 || (entry->protection & prot) != prot) {
2561 entry->eflags |= MAP_ENTRY_WIRE_SKIPPED;
2562 if ((flags & VM_MAP_WIRE_HOLESOK) == 0) {
2564 rv = KERN_INVALID_ADDRESS;
2569 if (entry->wired_count == 0) {
2570 entry->wired_count++;
2571 saved_start = entry->start;
2572 saved_end = entry->end;
2575 * Release the map lock, relying on the in-transition
2576 * mark. Mark the map busy for fork.
2581 faddr = saved_start;
2584 * Simulate a fault to get the page and enter
2585 * it into the physical map.
2587 if ((rv = vm_fault(map, faddr, VM_PROT_NONE,
2588 VM_FAULT_CHANGE_WIRING)) != KERN_SUCCESS)
2590 } while ((faddr += PAGE_SIZE) < saved_end);
2593 if (last_timestamp + 1 != map->timestamp) {
2595 * Look again for the entry because the map was
2596 * modified while it was unlocked. The entry
2597 * may have been clipped, but NOT merged or
2600 result = vm_map_lookup_entry(map, saved_start,
2602 KASSERT(result, ("vm_map_wire: lookup failed"));
2603 if (entry == first_entry)
2604 first_entry = tmp_entry;
2608 while (entry->end < saved_end) {
2610 * In case of failure, handle entries
2611 * that were not fully wired here;
2612 * fully wired entries are handled
2615 if (rv != KERN_SUCCESS &&
2617 vm_map_wire_entry_failure(map,
2619 entry = entry->next;
2622 last_timestamp = map->timestamp;
2623 if (rv != KERN_SUCCESS) {
2624 vm_map_wire_entry_failure(map, entry, faddr);
2628 } else if (!user_wire ||
2629 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
2630 entry->wired_count++;
2633 * Check the map for holes in the specified region.
2634 * If VM_MAP_WIRE_HOLESOK was specified, skip this check.
2637 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) &&
2638 (entry->end < end && (entry->next == &map->header ||
2639 entry->next->start > entry->end))) {
2641 rv = KERN_INVALID_ADDRESS;
2644 entry = entry->next;
2648 need_wakeup = FALSE;
2649 if (first_entry == NULL) {
2650 result = vm_map_lookup_entry(map, start, &first_entry);
2651 if (!result && (flags & VM_MAP_WIRE_HOLESOK))
2652 first_entry = first_entry->next;
2654 KASSERT(result, ("vm_map_wire: lookup failed"));
2656 for (entry = first_entry; entry != &map->header && entry->start < end;
2657 entry = entry->next) {
2658 if ((entry->eflags & MAP_ENTRY_WIRE_SKIPPED) != 0)
2659 goto next_entry_done;
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 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 (rv == KERN_SUCCESS) {
2679 entry->eflags |= MAP_ENTRY_USER_WIRED;
2680 } else if (entry->wired_count == -1) {
2682 * Wiring failed on this entry. Thus, unwiring is
2685 entry->wired_count = 0;
2686 } else if (!user_wire ||
2687 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
2689 * Undo the wiring. Wiring succeeded on this entry
2690 * but failed on a later entry.
2692 if (entry->wired_count == 1)
2693 vm_map_entry_unwire(map, entry);
2695 entry->wired_count--;
2698 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0,
2699 ("vm_map_wire: in-transition flag missing %p", entry));
2700 KASSERT(entry->wiring_thread == curthread,
2701 ("vm_map_wire: alien wire %p", entry));
2702 entry->eflags &= ~(MAP_ENTRY_IN_TRANSITION |
2703 MAP_ENTRY_WIRE_SKIPPED);
2704 entry->wiring_thread = NULL;
2705 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
2706 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
2709 vm_map_simplify_entry(map, entry);
2720 * Push any dirty cached pages in the address range to their pager.
2721 * If syncio is TRUE, dirty pages are written synchronously.
2722 * If invalidate is TRUE, any cached pages are freed as well.
2724 * If the size of the region from start to end is zero, we are
2725 * supposed to flush all modified pages within the region containing
2726 * start. Unfortunately, a region can be split or coalesced with
2727 * neighboring regions, making it difficult to determine what the
2728 * original region was. Therefore, we approximate this requirement by
2729 * flushing the current region containing start.
2731 * Returns an error if any part of the specified range is not mapped.
2739 boolean_t invalidate)
2741 vm_map_entry_t current;
2742 vm_map_entry_t entry;
2745 vm_ooffset_t offset;
2746 unsigned int last_timestamp;
2749 vm_map_lock_read(map);
2750 VM_MAP_RANGE_CHECK(map, start, end);
2751 if (!vm_map_lookup_entry(map, start, &entry)) {
2752 vm_map_unlock_read(map);
2753 return (KERN_INVALID_ADDRESS);
2754 } else if (start == end) {
2755 start = entry->start;
2759 * Make a first pass to check for user-wired memory and holes.
2761 for (current = entry; current != &map->header && current->start < end;
2762 current = current->next) {
2763 if (invalidate && (current->eflags & MAP_ENTRY_USER_WIRED)) {
2764 vm_map_unlock_read(map);
2765 return (KERN_INVALID_ARGUMENT);
2767 if (end > current->end &&
2768 (current->next == &map->header ||
2769 current->end != current->next->start)) {
2770 vm_map_unlock_read(map);
2771 return (KERN_INVALID_ADDRESS);
2776 pmap_remove(map->pmap, start, end);
2780 * Make a second pass, cleaning/uncaching pages from the indicated
2783 for (current = entry; current != &map->header && current->start < end;) {
2784 offset = current->offset + (start - current->start);
2785 size = (end <= current->end ? end : current->end) - start;
2786 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
2788 vm_map_entry_t tentry;
2791 smap = current->object.sub_map;
2792 vm_map_lock_read(smap);
2793 (void) vm_map_lookup_entry(smap, offset, &tentry);
2794 tsize = tentry->end - offset;
2797 object = tentry->object.vm_object;
2798 offset = tentry->offset + (offset - tentry->start);
2799 vm_map_unlock_read(smap);
2801 object = current->object.vm_object;
2803 vm_object_reference(object);
2804 last_timestamp = map->timestamp;
2805 vm_map_unlock_read(map);
2806 if (!vm_object_sync(object, offset, size, syncio, invalidate))
2809 vm_object_deallocate(object);
2810 vm_map_lock_read(map);
2811 if (last_timestamp == map->timestamp ||
2812 !vm_map_lookup_entry(map, start, ¤t))
2813 current = current->next;
2816 vm_map_unlock_read(map);
2817 return (failed ? KERN_FAILURE : KERN_SUCCESS);
2821 * vm_map_entry_unwire: [ internal use only ]
2823 * Make the region specified by this entry pageable.
2825 * The map in question should be locked.
2826 * [This is the reason for this routine's existence.]
2829 vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry)
2832 VM_MAP_ASSERT_LOCKED(map);
2833 KASSERT(entry->wired_count > 0,
2834 ("vm_map_entry_unwire: entry %p isn't wired", entry));
2835 pmap_unwire(map->pmap, entry->start, entry->end);
2836 vm_object_unwire(entry->object.vm_object, entry->offset, entry->end -
2837 entry->start, PQ_ACTIVE);
2838 entry->wired_count = 0;
2842 vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map)
2845 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0)
2846 vm_object_deallocate(entry->object.vm_object);
2847 uma_zfree(system_map ? kmapentzone : mapentzone, entry);
2851 * vm_map_entry_delete: [ internal use only ]
2853 * Deallocate the given entry from the target map.
2856 vm_map_entry_delete(vm_map_t map, vm_map_entry_t entry)
2859 vm_pindex_t offidxstart, offidxend, count, size1;
2862 vm_map_entry_unlink(map, entry);
2863 object = entry->object.vm_object;
2864 size = entry->end - entry->start;
2867 if (entry->cred != NULL) {
2868 swap_release_by_cred(size, entry->cred);
2869 crfree(entry->cred);
2872 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0 &&
2874 KASSERT(entry->cred == NULL || object->cred == NULL ||
2875 (entry->eflags & MAP_ENTRY_NEEDS_COPY),
2876 ("OVERCOMMIT vm_map_entry_delete: both cred %p", entry));
2877 count = OFF_TO_IDX(size);
2878 offidxstart = OFF_TO_IDX(entry->offset);
2879 offidxend = offidxstart + count;
2880 VM_OBJECT_WLOCK(object);
2881 if (object->ref_count != 1 &&
2882 ((object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING ||
2883 object == kernel_object || object == kmem_object)) {
2884 vm_object_collapse(object);
2887 * The option OBJPR_NOTMAPPED can be passed here
2888 * because vm_map_delete() already performed
2889 * pmap_remove() on the only mapping to this range
2892 vm_object_page_remove(object, offidxstart, offidxend,
2894 if (object->type == OBJT_SWAP)
2895 swap_pager_freespace(object, offidxstart, count);
2896 if (offidxend >= object->size &&
2897 offidxstart < object->size) {
2898 size1 = object->size;
2899 object->size = offidxstart;
2900 if (object->cred != NULL) {
2901 size1 -= object->size;
2902 KASSERT(object->charge >= ptoa(size1),
2903 ("vm_map_entry_delete: object->charge < 0"));
2904 swap_release_by_cred(ptoa(size1), object->cred);
2905 object->charge -= ptoa(size1);
2909 VM_OBJECT_WUNLOCK(object);
2911 entry->object.vm_object = NULL;
2912 if (map->system_map)
2913 vm_map_entry_deallocate(entry, TRUE);
2915 entry->next = curthread->td_map_def_user;
2916 curthread->td_map_def_user = entry;
2921 * vm_map_delete: [ internal use only ]
2923 * Deallocates the given address range from the target
2927 vm_map_delete(vm_map_t map, vm_offset_t start, vm_offset_t end)
2929 vm_map_entry_t entry;
2930 vm_map_entry_t first_entry;
2932 VM_MAP_ASSERT_LOCKED(map);
2934 return (KERN_SUCCESS);
2937 * Find the start of the region, and clip it
2939 if (!vm_map_lookup_entry(map, start, &first_entry))
2940 entry = first_entry->next;
2942 entry = first_entry;
2943 vm_map_clip_start(map, entry, start);
2947 * Step through all entries in this region
2949 while ((entry != &map->header) && (entry->start < end)) {
2950 vm_map_entry_t next;
2953 * Wait for wiring or unwiring of an entry to complete.
2954 * Also wait for any system wirings to disappear on
2957 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 ||
2958 (vm_map_pmap(map) != kernel_pmap &&
2959 vm_map_entry_system_wired_count(entry) != 0)) {
2960 unsigned int last_timestamp;
2961 vm_offset_t saved_start;
2962 vm_map_entry_t tmp_entry;
2964 saved_start = entry->start;
2965 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2966 last_timestamp = map->timestamp;
2967 (void) vm_map_unlock_and_wait(map, 0);
2969 if (last_timestamp + 1 != map->timestamp) {
2971 * Look again for the entry because the map was
2972 * modified while it was unlocked.
2973 * Specifically, the entry may have been
2974 * clipped, merged, or deleted.
2976 if (!vm_map_lookup_entry(map, saved_start,
2978 entry = tmp_entry->next;
2981 vm_map_clip_start(map, entry,
2987 vm_map_clip_end(map, entry, end);
2992 * Unwire before removing addresses from the pmap; otherwise,
2993 * unwiring will put the entries back in the pmap.
2995 if (entry->wired_count != 0) {
2996 vm_map_entry_unwire(map, entry);
2999 pmap_remove(map->pmap, entry->start, entry->end);
3002 * Delete the entry only after removing all pmap
3003 * entries pointing to its pages. (Otherwise, its
3004 * page frames may be reallocated, and any modify bits
3005 * will be set in the wrong object!)
3007 vm_map_entry_delete(map, entry);
3010 return (KERN_SUCCESS);
3016 * Remove the given address range from the target map.
3017 * This is the exported form of vm_map_delete.
3020 vm_map_remove(vm_map_t map, vm_offset_t start, vm_offset_t end)
3025 VM_MAP_RANGE_CHECK(map, start, end);
3026 result = vm_map_delete(map, start, end);
3032 * vm_map_check_protection:
3034 * Assert that the target map allows the specified privilege on the
3035 * entire address region given. The entire region must be allocated.
3037 * WARNING! This code does not and should not check whether the
3038 * contents of the region is accessible. For example a smaller file
3039 * might be mapped into a larger address space.
3041 * NOTE! This code is also called by munmap().
3043 * The map must be locked. A read lock is sufficient.
3046 vm_map_check_protection(vm_map_t map, vm_offset_t start, vm_offset_t end,
3047 vm_prot_t protection)
3049 vm_map_entry_t entry;
3050 vm_map_entry_t tmp_entry;
3052 if (!vm_map_lookup_entry(map, start, &tmp_entry))
3056 while (start < end) {
3057 if (entry == &map->header)
3062 if (start < entry->start)
3065 * Check protection associated with entry.
3067 if ((entry->protection & protection) != protection)
3069 /* go to next entry */
3071 entry = entry->next;
3077 * vm_map_copy_entry:
3079 * Copies the contents of the source entry to the destination
3080 * entry. The entries *must* be aligned properly.
3086 vm_map_entry_t src_entry,
3087 vm_map_entry_t dst_entry,
3088 vm_ooffset_t *fork_charge)
3090 vm_object_t src_object;
3091 vm_map_entry_t fake_entry;
3096 VM_MAP_ASSERT_LOCKED(dst_map);
3098 if ((dst_entry->eflags|src_entry->eflags) & MAP_ENTRY_IS_SUB_MAP)
3101 if (src_entry->wired_count == 0 ||
3102 (src_entry->protection & VM_PROT_WRITE) == 0) {
3104 * If the source entry is marked needs_copy, it is already
3107 if ((src_entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0 &&
3108 (src_entry->protection & VM_PROT_WRITE) != 0) {
3109 pmap_protect(src_map->pmap,
3112 src_entry->protection & ~VM_PROT_WRITE);
3116 * Make a copy of the object.
3118 size = src_entry->end - src_entry->start;
3119 if ((src_object = src_entry->object.vm_object) != NULL) {
3120 VM_OBJECT_WLOCK(src_object);
3121 charged = ENTRY_CHARGED(src_entry);
3122 if ((src_object->handle == NULL) &&
3123 (src_object->type == OBJT_DEFAULT ||
3124 src_object->type == OBJT_SWAP)) {
3125 vm_object_collapse(src_object);
3126 if ((src_object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING) {
3127 vm_object_split(src_entry);
3128 src_object = src_entry->object.vm_object;
3131 vm_object_reference_locked(src_object);
3132 vm_object_clear_flag(src_object, OBJ_ONEMAPPING);
3133 if (src_entry->cred != NULL &&
3134 !(src_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
3135 KASSERT(src_object->cred == NULL,
3136 ("OVERCOMMIT: vm_map_copy_entry: cred %p",
3138 src_object->cred = src_entry->cred;
3139 src_object->charge = size;
3141 VM_OBJECT_WUNLOCK(src_object);
3142 dst_entry->object.vm_object = src_object;
3144 cred = curthread->td_ucred;
3146 dst_entry->cred = cred;
3147 *fork_charge += size;
3148 if (!(src_entry->eflags &
3149 MAP_ENTRY_NEEDS_COPY)) {
3151 src_entry->cred = cred;
3152 *fork_charge += size;
3155 src_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY);
3156 dst_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY);
3157 dst_entry->offset = src_entry->offset;
3158 if (src_entry->eflags & MAP_ENTRY_VN_WRITECNT) {
3160 * MAP_ENTRY_VN_WRITECNT cannot
3161 * indicate write reference from
3162 * src_entry, since the entry is
3163 * marked as needs copy. Allocate a
3164 * fake entry that is used to
3165 * decrement object->un_pager.vnp.writecount
3166 * at the appropriate time. Attach
3167 * fake_entry to the deferred list.
3169 fake_entry = vm_map_entry_create(dst_map);
3170 fake_entry->eflags = MAP_ENTRY_VN_WRITECNT;
3171 src_entry->eflags &= ~MAP_ENTRY_VN_WRITECNT;
3172 vm_object_reference(src_object);
3173 fake_entry->object.vm_object = src_object;
3174 fake_entry->start = src_entry->start;
3175 fake_entry->end = src_entry->end;
3176 fake_entry->next = curthread->td_map_def_user;
3177 curthread->td_map_def_user = fake_entry;
3180 dst_entry->object.vm_object = NULL;
3181 dst_entry->offset = 0;
3182 if (src_entry->cred != NULL) {
3183 dst_entry->cred = curthread->td_ucred;
3184 crhold(dst_entry->cred);
3185 *fork_charge += size;
3189 pmap_copy(dst_map->pmap, src_map->pmap, dst_entry->start,
3190 dst_entry->end - dst_entry->start, src_entry->start);
3193 * We don't want to make writeable wired pages copy-on-write.
3194 * Immediately copy these pages into the new map by simulating
3195 * page faults. The new pages are pageable.
3197 vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry,
3203 * vmspace_map_entry_forked:
3204 * Update the newly-forked vmspace each time a map entry is inherited
3205 * or copied. The values for vm_dsize and vm_tsize are approximate
3206 * (and mostly-obsolete ideas in the face of mmap(2) et al.)
3209 vmspace_map_entry_forked(const struct vmspace *vm1, struct vmspace *vm2,
3210 vm_map_entry_t entry)
3212 vm_size_t entrysize;
3215 entrysize = entry->end - entry->start;
3216 vm2->vm_map.size += entrysize;
3217 if (entry->eflags & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP)) {
3218 vm2->vm_ssize += btoc(entrysize);
3219 } else if (entry->start >= (vm_offset_t)vm1->vm_daddr &&
3220 entry->start < (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)) {
3221 newend = MIN(entry->end,
3222 (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize));
3223 vm2->vm_dsize += btoc(newend - entry->start);
3224 } else if (entry->start >= (vm_offset_t)vm1->vm_taddr &&
3225 entry->start < (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)) {
3226 newend = MIN(entry->end,
3227 (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize));
3228 vm2->vm_tsize += btoc(newend - entry->start);
3234 * Create a new process vmspace structure and vm_map
3235 * based on those of an existing process. The new map
3236 * is based on the old map, according to the inheritance
3237 * values on the regions in that map.
3239 * XXX It might be worth coalescing the entries added to the new vmspace.
3241 * The source map must not be locked.
3244 vmspace_fork(struct vmspace *vm1, vm_ooffset_t *fork_charge)
3246 struct vmspace *vm2;
3247 vm_map_t new_map, old_map;
3248 vm_map_entry_t new_entry, old_entry;
3252 old_map = &vm1->vm_map;
3253 /* Copy immutable fields of vm1 to vm2. */
3254 vm2 = vmspace_alloc(old_map->min_offset, old_map->max_offset, NULL);
3257 vm2->vm_taddr = vm1->vm_taddr;
3258 vm2->vm_daddr = vm1->vm_daddr;
3259 vm2->vm_maxsaddr = vm1->vm_maxsaddr;
3260 vm_map_lock(old_map);
3262 vm_map_wait_busy(old_map);
3263 new_map = &vm2->vm_map;
3264 locked = vm_map_trylock(new_map); /* trylock to silence WITNESS */
3265 KASSERT(locked, ("vmspace_fork: lock failed"));
3267 old_entry = old_map->header.next;
3269 while (old_entry != &old_map->header) {
3270 if (old_entry->eflags & MAP_ENTRY_IS_SUB_MAP)
3271 panic("vm_map_fork: encountered a submap");
3273 switch (old_entry->inheritance) {
3274 case VM_INHERIT_NONE:
3277 case VM_INHERIT_SHARE:
3279 * Clone the entry, creating the shared object if necessary.
3281 object = old_entry->object.vm_object;
3282 if (object == NULL) {
3283 object = vm_object_allocate(OBJT_DEFAULT,
3284 atop(old_entry->end - old_entry->start));
3285 old_entry->object.vm_object = object;
3286 old_entry->offset = 0;
3287 if (old_entry->cred != NULL) {
3288 object->cred = old_entry->cred;
3289 object->charge = old_entry->end -
3291 old_entry->cred = NULL;
3296 * Add the reference before calling vm_object_shadow
3297 * to insure that a shadow object is created.
3299 vm_object_reference(object);
3300 if (old_entry->eflags & MAP_ENTRY_NEEDS_COPY) {
3301 vm_object_shadow(&old_entry->object.vm_object,
3303 old_entry->end - old_entry->start);
3304 old_entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
3305 /* Transfer the second reference too. */
3306 vm_object_reference(
3307 old_entry->object.vm_object);
3310 * As in vm_map_simplify_entry(), the
3311 * vnode lock will not be acquired in
3312 * this call to vm_object_deallocate().
3314 vm_object_deallocate(object);
3315 object = old_entry->object.vm_object;
3317 VM_OBJECT_WLOCK(object);
3318 vm_object_clear_flag(object, OBJ_ONEMAPPING);
3319 if (old_entry->cred != NULL) {
3320 KASSERT(object->cred == NULL, ("vmspace_fork both cred"));
3321 object->cred = old_entry->cred;
3322 object->charge = old_entry->end - old_entry->start;
3323 old_entry->cred = NULL;
3327 * Assert the correct state of the vnode
3328 * v_writecount while the object is locked, to
3329 * not relock it later for the assertion
3332 if (old_entry->eflags & MAP_ENTRY_VN_WRITECNT &&
3333 object->type == OBJT_VNODE) {
3334 KASSERT(((struct vnode *)object->handle)->
3336 ("vmspace_fork: v_writecount %p", object));
3337 KASSERT(object->un_pager.vnp.writemappings > 0,
3338 ("vmspace_fork: vnp.writecount %p",
3341 VM_OBJECT_WUNLOCK(object);
3344 * Clone the entry, referencing the shared object.
3346 new_entry = vm_map_entry_create(new_map);
3347 *new_entry = *old_entry;
3348 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED |
3349 MAP_ENTRY_IN_TRANSITION);
3350 new_entry->wiring_thread = NULL;
3351 new_entry->wired_count = 0;
3352 if (new_entry->eflags & MAP_ENTRY_VN_WRITECNT) {
3353 vnode_pager_update_writecount(object,
3354 new_entry->start, new_entry->end);
3358 * Insert the entry into the new map -- we know we're
3359 * inserting at the end of the new map.
3361 vm_map_entry_link(new_map, new_map->header.prev,
3363 vmspace_map_entry_forked(vm1, vm2, new_entry);
3366 * Update the physical map
3368 pmap_copy(new_map->pmap, old_map->pmap,
3370 (old_entry->end - old_entry->start),
3374 case VM_INHERIT_COPY:
3376 * Clone the entry and link into the map.
3378 new_entry = vm_map_entry_create(new_map);
3379 *new_entry = *old_entry;
3381 * Copied entry is COW over the old object.
3383 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED |
3384 MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_VN_WRITECNT);
3385 new_entry->wiring_thread = NULL;
3386 new_entry->wired_count = 0;
3387 new_entry->object.vm_object = NULL;
3388 new_entry->cred = NULL;
3389 vm_map_entry_link(new_map, new_map->header.prev,
3391 vmspace_map_entry_forked(vm1, vm2, new_entry);
3392 vm_map_copy_entry(old_map, new_map, old_entry,
3393 new_entry, fork_charge);
3396 old_entry = old_entry->next;
3399 * Use inlined vm_map_unlock() to postpone handling the deferred
3400 * map entries, which cannot be done until both old_map and
3401 * new_map locks are released.
3403 sx_xunlock(&old_map->lock);
3404 sx_xunlock(&new_map->lock);
3405 vm_map_process_deferred();
3411 vm_map_stack(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
3412 vm_prot_t prot, vm_prot_t max, int cow)
3414 vm_size_t growsize, init_ssize;
3415 rlim_t lmemlim, vmemlim;
3418 growsize = sgrowsiz;
3419 init_ssize = (max_ssize < growsize) ? max_ssize : growsize;
3422 lmemlim = lim_cur(curproc, RLIMIT_MEMLOCK);
3423 vmemlim = lim_cur(curproc, RLIMIT_VMEM);
3424 PROC_UNLOCK(curproc);
3425 if (!old_mlock && map->flags & MAP_WIREFUTURE) {
3426 if (ptoa(pmap_wired_count(map->pmap)) + init_ssize > lmemlim) {
3431 /* If we would blow our VMEM resource limit, no go */
3432 if (map->size + init_ssize > vmemlim) {
3436 rv = vm_map_stack_locked(map, addrbos, max_ssize, growsize, prot,
3444 vm_map_stack_locked(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
3445 vm_size_t growsize, vm_prot_t prot, vm_prot_t max, int cow)
3447 vm_map_entry_t new_entry, prev_entry;
3448 vm_offset_t bot, top;
3449 vm_size_t init_ssize;
3453 * The stack orientation is piggybacked with the cow argument.
3454 * Extract it into orient and mask the cow argument so that we
3455 * don't pass it around further.
3456 * NOTE: We explicitly allow bi-directional stacks.
3458 orient = cow & (MAP_STACK_GROWS_DOWN|MAP_STACK_GROWS_UP);
3459 KASSERT(orient != 0, ("No stack grow direction"));
3461 if (addrbos < vm_map_min(map) ||
3462 addrbos > vm_map_max(map) ||
3463 addrbos + max_ssize < addrbos)
3464 return (KERN_NO_SPACE);
3466 init_ssize = (max_ssize < growsize) ? max_ssize : growsize;
3468 /* If addr is already mapped, no go */
3469 if (vm_map_lookup_entry(map, addrbos, &prev_entry))
3470 return (KERN_NO_SPACE);
3473 * If we can't accomodate max_ssize in the current mapping, no go.
3474 * However, we need to be aware that subsequent user mappings might
3475 * map into the space we have reserved for stack, and currently this
3476 * space is not protected.
3478 * Hopefully we will at least detect this condition when we try to
3481 if ((prev_entry->next != &map->header) &&
3482 (prev_entry->next->start < addrbos + max_ssize))
3483 return (KERN_NO_SPACE);
3486 * We initially map a stack of only init_ssize. We will grow as
3487 * needed later. Depending on the orientation of the stack (i.e.
3488 * the grow direction) we either map at the top of the range, the
3489 * bottom of the range or in the middle.
3491 * Note: we would normally expect prot and max to be VM_PROT_ALL,
3492 * and cow to be 0. Possibly we should eliminate these as input
3493 * parameters, and just pass these values here in the insert call.
3495 if (orient == MAP_STACK_GROWS_DOWN)
3496 bot = addrbos + max_ssize - init_ssize;
3497 else if (orient == MAP_STACK_GROWS_UP)
3500 bot = round_page(addrbos + max_ssize/2 - init_ssize/2);
3501 top = bot + init_ssize;
3502 rv = vm_map_insert(map, NULL, 0, bot, top, prot, max, cow);
3504 /* Now set the avail_ssize amount. */
3505 if (rv == KERN_SUCCESS) {
3506 new_entry = prev_entry->next;
3507 if (new_entry->end != top || new_entry->start != bot)
3508 panic("Bad entry start/end for new stack entry");
3510 new_entry->avail_ssize = max_ssize - init_ssize;
3511 KASSERT((orient & MAP_STACK_GROWS_DOWN) == 0 ||
3512 (new_entry->eflags & MAP_ENTRY_GROWS_DOWN) != 0,
3513 ("new entry lacks MAP_ENTRY_GROWS_DOWN"));
3514 KASSERT((orient & MAP_STACK_GROWS_UP) == 0 ||
3515 (new_entry->eflags & MAP_ENTRY_GROWS_UP) != 0,
3516 ("new entry lacks MAP_ENTRY_GROWS_UP"));
3522 static int stack_guard_page = 0;
3523 SYSCTL_INT(_security_bsd, OID_AUTO, stack_guard_page, CTLFLAG_RWTUN,
3524 &stack_guard_page, 0,
3525 "Insert stack guard page ahead of the growable segments.");
3527 /* Attempts to grow a vm stack entry. Returns KERN_SUCCESS if the
3528 * desired address is already mapped, or if we successfully grow
3529 * the stack. Also returns KERN_SUCCESS if addr is outside the
3530 * stack range (this is strange, but preserves compatibility with
3531 * the grow function in vm_machdep.c).
3534 vm_map_growstack(struct proc *p, vm_offset_t addr)
3536 vm_map_entry_t next_entry, prev_entry;
3537 vm_map_entry_t new_entry, stack_entry;
3538 struct vmspace *vm = p->p_vmspace;
3539 vm_map_t map = &vm->vm_map;
3542 size_t grow_amount, max_grow;
3543 rlim_t lmemlim, stacklim, vmemlim;
3544 int is_procstack, rv;
3555 lmemlim = lim_cur(p, RLIMIT_MEMLOCK);
3556 stacklim = lim_cur(p, RLIMIT_STACK);
3557 vmemlim = lim_cur(p, RLIMIT_VMEM);
3560 vm_map_lock_read(map);
3562 /* If addr is already in the entry range, no need to grow.*/
3563 if (vm_map_lookup_entry(map, addr, &prev_entry)) {
3564 vm_map_unlock_read(map);
3565 return (KERN_SUCCESS);
3568 next_entry = prev_entry->next;
3569 if (!(prev_entry->eflags & MAP_ENTRY_GROWS_UP)) {
3571 * This entry does not grow upwards. Since the address lies
3572 * beyond this entry, the next entry (if one exists) has to
3573 * be a downward growable entry. The entry list header is
3574 * never a growable entry, so it suffices to check the flags.
3576 if (!(next_entry->eflags & MAP_ENTRY_GROWS_DOWN)) {
3577 vm_map_unlock_read(map);
3578 return (KERN_SUCCESS);
3580 stack_entry = next_entry;
3583 * This entry grows upward. If the next entry does not at
3584 * least grow downwards, this is the entry we need to grow.
3585 * otherwise we have two possible choices and we have to
3588 if (next_entry->eflags & MAP_ENTRY_GROWS_DOWN) {
3590 * We have two choices; grow the entry closest to
3591 * the address to minimize the amount of growth.
3593 if (addr - prev_entry->end <= next_entry->start - addr)
3594 stack_entry = prev_entry;
3596 stack_entry = next_entry;
3598 stack_entry = prev_entry;
3601 if (stack_entry == next_entry) {
3602 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_DOWN, ("foo"));
3603 KASSERT(addr < stack_entry->start, ("foo"));
3604 end = (prev_entry != &map->header) ? prev_entry->end :
3605 stack_entry->start - stack_entry->avail_ssize;
3606 grow_amount = roundup(stack_entry->start - addr, PAGE_SIZE);
3607 max_grow = stack_entry->start - end;
3609 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_UP, ("foo"));
3610 KASSERT(addr >= stack_entry->end, ("foo"));
3611 end = (next_entry != &map->header) ? next_entry->start :
3612 stack_entry->end + stack_entry->avail_ssize;
3613 grow_amount = roundup(addr + 1 - stack_entry->end, PAGE_SIZE);
3614 max_grow = end - stack_entry->end;
3617 if (grow_amount > stack_entry->avail_ssize) {
3618 vm_map_unlock_read(map);
3619 return (KERN_NO_SPACE);
3623 * If there is no longer enough space between the entries nogo, and
3624 * adjust the available space. Note: this should only happen if the
3625 * user has mapped into the stack area after the stack was created,
3626 * and is probably an error.
3628 * This also effectively destroys any guard page the user might have
3629 * intended by limiting the stack size.
3631 if (grow_amount + (stack_guard_page ? PAGE_SIZE : 0) > max_grow) {
3632 if (vm_map_lock_upgrade(map))
3635 stack_entry->avail_ssize = max_grow;
3638 return (KERN_NO_SPACE);
3641 is_procstack = (addr >= (vm_offset_t)vm->vm_maxsaddr) ? 1 : 0;
3644 * If this is the main process stack, see if we're over the stack
3647 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) {
3648 vm_map_unlock_read(map);
3649 return (KERN_NO_SPACE);
3654 racct_set(p, RACCT_STACK, ctob(vm->vm_ssize) + grow_amount)) {
3656 vm_map_unlock_read(map);
3657 return (KERN_NO_SPACE);
3662 /* Round up the grow amount modulo sgrowsiz */
3663 growsize = sgrowsiz;
3664 grow_amount = roundup(grow_amount, growsize);
3665 if (grow_amount > stack_entry->avail_ssize)
3666 grow_amount = stack_entry->avail_ssize;
3667 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) {
3668 grow_amount = trunc_page((vm_size_t)stacklim) -
3673 limit = racct_get_available(p, RACCT_STACK);
3675 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > limit))
3676 grow_amount = limit - ctob(vm->vm_ssize);
3678 if (!old_mlock && map->flags & MAP_WIREFUTURE) {
3679 if (ptoa(pmap_wired_count(map->pmap)) + grow_amount > lmemlim) {
3680 vm_map_unlock_read(map);
3686 if (racct_set(p, RACCT_MEMLOCK,
3687 ptoa(pmap_wired_count(map->pmap)) + grow_amount)) {
3689 vm_map_unlock_read(map);
3696 /* If we would blow our VMEM resource limit, no go */
3697 if (map->size + grow_amount > vmemlim) {
3698 vm_map_unlock_read(map);
3704 if (racct_set(p, RACCT_VMEM, map->size + grow_amount)) {
3706 vm_map_unlock_read(map);
3713 if (vm_map_lock_upgrade(map))
3716 if (stack_entry == next_entry) {
3720 /* Get the preliminary new entry start value */
3721 addr = stack_entry->start - grow_amount;
3724 * If this puts us into the previous entry, cut back our
3725 * growth to the available space. Also, see the note above.
3728 stack_entry->avail_ssize = max_grow;
3730 if (stack_guard_page)
3734 rv = vm_map_insert(map, NULL, 0, addr, stack_entry->start,
3735 next_entry->protection, next_entry->max_protection,
3736 MAP_STACK_GROWS_DOWN);
3738 /* Adjust the available stack space by the amount we grew. */
3739 if (rv == KERN_SUCCESS) {
3740 new_entry = prev_entry->next;
3741 KASSERT(new_entry == stack_entry->prev, ("foo"));
3742 KASSERT(new_entry->end == stack_entry->start, ("foo"));
3743 KASSERT(new_entry->start == addr, ("foo"));
3744 KASSERT((new_entry->eflags & MAP_ENTRY_GROWS_DOWN) !=
3745 0, ("new entry lacks MAP_ENTRY_GROWS_DOWN"));
3746 grow_amount = new_entry->end - new_entry->start;
3747 new_entry->avail_ssize = stack_entry->avail_ssize -
3749 stack_entry->eflags &= ~MAP_ENTRY_GROWS_DOWN;
3755 addr = stack_entry->end + grow_amount;
3758 * If this puts us into the next entry, cut back our growth
3759 * to the available space. Also, see the note above.
3762 stack_entry->avail_ssize = end - stack_entry->end;
3764 if (stack_guard_page)
3768 grow_amount = addr - stack_entry->end;
3769 cred = stack_entry->cred;
3770 if (cred == NULL && stack_entry->object.vm_object != NULL)
3771 cred = stack_entry->object.vm_object->cred;
3772 if (cred != NULL && !swap_reserve_by_cred(grow_amount, cred))
3774 /* Grow the underlying object if applicable. */
3775 else if (stack_entry->object.vm_object == NULL ||
3776 vm_object_coalesce(stack_entry->object.vm_object,
3777 stack_entry->offset,
3778 (vm_size_t)(stack_entry->end - stack_entry->start),
3779 (vm_size_t)grow_amount, cred != NULL)) {
3780 map->size += (addr - stack_entry->end);
3781 /* Update the current entry. */
3782 stack_entry->end = addr;
3783 stack_entry->avail_ssize -= grow_amount;
3784 vm_map_entry_resize_free(map, stack_entry);
3790 if (rv == KERN_SUCCESS && is_procstack)
3791 vm->vm_ssize += btoc(grow_amount);
3796 * Heed the MAP_WIREFUTURE flag if it was set for this process.
3798 if (rv == KERN_SUCCESS && (map->flags & MAP_WIREFUTURE)) {
3800 (stack_entry == next_entry) ? addr : addr - grow_amount,
3801 (stack_entry == next_entry) ? stack_entry->start : addr,
3802 (p->p_flag & P_SYSTEM)
3803 ? VM_MAP_WIRE_SYSTEM|VM_MAP_WIRE_NOHOLES
3804 : VM_MAP_WIRE_USER|VM_MAP_WIRE_NOHOLES);
3809 if (rv != KERN_SUCCESS) {
3811 error = racct_set(p, RACCT_VMEM, map->size);
3812 KASSERT(error == 0, ("decreasing RACCT_VMEM failed"));
3814 error = racct_set(p, RACCT_MEMLOCK,
3815 ptoa(pmap_wired_count(map->pmap)));
3816 KASSERT(error == 0, ("decreasing RACCT_MEMLOCK failed"));
3818 error = racct_set(p, RACCT_STACK, ctob(vm->vm_ssize));
3819 KASSERT(error == 0, ("decreasing RACCT_STACK failed"));
3828 * Unshare the specified VM space for exec. If other processes are
3829 * mapped to it, then create a new one. The new vmspace is null.
3832 vmspace_exec(struct proc *p, vm_offset_t minuser, vm_offset_t maxuser)
3834 struct vmspace *oldvmspace = p->p_vmspace;
3835 struct vmspace *newvmspace;
3837 KASSERT((curthread->td_pflags & TDP_EXECVMSPC) == 0,
3838 ("vmspace_exec recursed"));
3839 newvmspace = vmspace_alloc(minuser, maxuser, NULL);
3840 if (newvmspace == NULL)
3842 newvmspace->vm_swrss = oldvmspace->vm_swrss;
3844 * This code is written like this for prototype purposes. The
3845 * goal is to avoid running down the vmspace here, but let the
3846 * other process's that are still using the vmspace to finally
3847 * run it down. Even though there is little or no chance of blocking
3848 * here, it is a good idea to keep this form for future mods.
3850 PROC_VMSPACE_LOCK(p);
3851 p->p_vmspace = newvmspace;
3852 PROC_VMSPACE_UNLOCK(p);
3853 if (p == curthread->td_proc)
3854 pmap_activate(curthread);
3855 curthread->td_pflags |= TDP_EXECVMSPC;
3860 * Unshare the specified VM space for forcing COW. This
3861 * is called by rfork, for the (RFMEM|RFPROC) == 0 case.
3864 vmspace_unshare(struct proc *p)
3866 struct vmspace *oldvmspace = p->p_vmspace;
3867 struct vmspace *newvmspace;
3868 vm_ooffset_t fork_charge;
3870 if (oldvmspace->vm_refcnt == 1)
3873 newvmspace = vmspace_fork(oldvmspace, &fork_charge);
3874 if (newvmspace == NULL)
3876 if (!swap_reserve_by_cred(fork_charge, p->p_ucred)) {
3877 vmspace_free(newvmspace);
3880 PROC_VMSPACE_LOCK(p);
3881 p->p_vmspace = newvmspace;
3882 PROC_VMSPACE_UNLOCK(p);
3883 if (p == curthread->td_proc)
3884 pmap_activate(curthread);
3885 vmspace_free(oldvmspace);
3892 * Finds the VM object, offset, and
3893 * protection for a given virtual address in the
3894 * specified map, assuming a page fault of the
3897 * Leaves the map in question locked for read; return
3898 * values are guaranteed until a vm_map_lookup_done
3899 * call is performed. Note that the map argument
3900 * is in/out; the returned map must be used in
3901 * the call to vm_map_lookup_done.
3903 * A handle (out_entry) is returned for use in
3904 * vm_map_lookup_done, to make that fast.
3906 * If a lookup is requested with "write protection"
3907 * specified, the map may be changed to perform virtual
3908 * copying operations, although the data referenced will
3912 vm_map_lookup(vm_map_t *var_map, /* IN/OUT */
3914 vm_prot_t fault_typea,
3915 vm_map_entry_t *out_entry, /* OUT */
3916 vm_object_t *object, /* OUT */
3917 vm_pindex_t *pindex, /* OUT */
3918 vm_prot_t *out_prot, /* OUT */
3919 boolean_t *wired) /* OUT */
3921 vm_map_entry_t entry;
3922 vm_map_t map = *var_map;
3924 vm_prot_t fault_type = fault_typea;
3925 vm_object_t eobject;
3931 vm_map_lock_read(map);
3934 * Lookup the faulting address.
3936 if (!vm_map_lookup_entry(map, vaddr, out_entry)) {
3937 vm_map_unlock_read(map);
3938 return (KERN_INVALID_ADDRESS);
3946 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
3947 vm_map_t old_map = map;
3949 *var_map = map = entry->object.sub_map;
3950 vm_map_unlock_read(old_map);
3955 * Check whether this task is allowed to have this page.
3957 prot = entry->protection;
3958 fault_type &= (VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE);
3959 if ((fault_type & prot) != fault_type || prot == VM_PROT_NONE) {
3960 vm_map_unlock_read(map);
3961 return (KERN_PROTECTION_FAILURE);
3963 if ((entry->eflags & MAP_ENTRY_USER_WIRED) &&
3964 (entry->eflags & MAP_ENTRY_COW) &&
3965 (fault_type & VM_PROT_WRITE)) {
3966 vm_map_unlock_read(map);
3967 return (KERN_PROTECTION_FAILURE);
3969 if ((fault_typea & VM_PROT_COPY) != 0 &&
3970 (entry->max_protection & VM_PROT_WRITE) == 0 &&
3971 (entry->eflags & MAP_ENTRY_COW) == 0) {
3972 vm_map_unlock_read(map);
3973 return (KERN_PROTECTION_FAILURE);
3977 * If this page is not pageable, we have to get it for all possible
3980 *wired = (entry->wired_count != 0);
3982 fault_type = entry->protection;
3983 size = entry->end - entry->start;
3985 * If the entry was copy-on-write, we either ...
3987 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
3989 * If we want to write the page, we may as well handle that
3990 * now since we've got the map locked.
3992 * If we don't need to write the page, we just demote the
3993 * permissions allowed.
3995 if ((fault_type & VM_PROT_WRITE) != 0 ||
3996 (fault_typea & VM_PROT_COPY) != 0) {
3998 * Make a new object, and place it in the object
3999 * chain. Note that no new references have appeared
4000 * -- one just moved from the map to the new
4003 if (vm_map_lock_upgrade(map))
4006 if (entry->cred == NULL) {
4008 * The debugger owner is charged for
4011 cred = curthread->td_ucred;
4013 if (!swap_reserve_by_cred(size, cred)) {
4016 return (KERN_RESOURCE_SHORTAGE);
4020 vm_object_shadow(&entry->object.vm_object,
4021 &entry->offset, size);
4022 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
4023 eobject = entry->object.vm_object;
4024 if (eobject->cred != NULL) {
4026 * The object was not shadowed.
4028 swap_release_by_cred(size, entry->cred);
4029 crfree(entry->cred);
4031 } else if (entry->cred != NULL) {
4032 VM_OBJECT_WLOCK(eobject);
4033 eobject->cred = entry->cred;
4034 eobject->charge = size;
4035 VM_OBJECT_WUNLOCK(eobject);
4039 vm_map_lock_downgrade(map);
4042 * We're attempting to read a copy-on-write page --
4043 * don't allow writes.
4045 prot &= ~VM_PROT_WRITE;
4050 * Create an object if necessary.
4052 if (entry->object.vm_object == NULL &&
4054 if (vm_map_lock_upgrade(map))
4056 entry->object.vm_object = vm_object_allocate(OBJT_DEFAULT,
4059 if (entry->cred != NULL) {
4060 VM_OBJECT_WLOCK(entry->object.vm_object);
4061 entry->object.vm_object->cred = entry->cred;
4062 entry->object.vm_object->charge = size;
4063 VM_OBJECT_WUNLOCK(entry->object.vm_object);
4066 vm_map_lock_downgrade(map);
4070 * Return the object/offset from this entry. If the entry was
4071 * copy-on-write or empty, it has been fixed up.
4073 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
4074 *object = entry->object.vm_object;
4077 return (KERN_SUCCESS);
4081 * vm_map_lookup_locked:
4083 * Lookup the faulting address. A version of vm_map_lookup that returns
4084 * KERN_FAILURE instead of blocking on map lock or memory allocation.
4087 vm_map_lookup_locked(vm_map_t *var_map, /* IN/OUT */
4089 vm_prot_t fault_typea,
4090 vm_map_entry_t *out_entry, /* OUT */
4091 vm_object_t *object, /* OUT */
4092 vm_pindex_t *pindex, /* OUT */
4093 vm_prot_t *out_prot, /* OUT */
4094 boolean_t *wired) /* OUT */
4096 vm_map_entry_t entry;
4097 vm_map_t map = *var_map;
4099 vm_prot_t fault_type = fault_typea;
4102 * Lookup the faulting address.
4104 if (!vm_map_lookup_entry(map, vaddr, out_entry))
4105 return (KERN_INVALID_ADDRESS);
4110 * Fail if the entry refers to a submap.
4112 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP)
4113 return (KERN_FAILURE);
4116 * Check whether this task is allowed to have this page.
4118 prot = entry->protection;
4119 fault_type &= VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE;
4120 if ((fault_type & prot) != fault_type)
4121 return (KERN_PROTECTION_FAILURE);
4122 if ((entry->eflags & MAP_ENTRY_USER_WIRED) &&
4123 (entry->eflags & MAP_ENTRY_COW) &&
4124 (fault_type & VM_PROT_WRITE))
4125 return (KERN_PROTECTION_FAILURE);
4128 * If this page is not pageable, we have to get it for all possible
4131 *wired = (entry->wired_count != 0);
4133 fault_type = entry->protection;
4135 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
4137 * Fail if the entry was copy-on-write for a write fault.
4139 if (fault_type & VM_PROT_WRITE)
4140 return (KERN_FAILURE);
4142 * We're attempting to read a copy-on-write page --
4143 * don't allow writes.
4145 prot &= ~VM_PROT_WRITE;
4149 * Fail if an object should be created.
4151 if (entry->object.vm_object == NULL && !map->system_map)
4152 return (KERN_FAILURE);
4155 * Return the object/offset from this entry. If the entry was
4156 * copy-on-write or empty, it has been fixed up.
4158 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
4159 *object = entry->object.vm_object;
4162 return (KERN_SUCCESS);
4166 * vm_map_lookup_done:
4168 * Releases locks acquired by a vm_map_lookup
4169 * (according to the handle returned by that lookup).
4172 vm_map_lookup_done(vm_map_t map, vm_map_entry_t entry)
4175 * Unlock the main-level map
4177 vm_map_unlock_read(map);
4180 #include "opt_ddb.h"
4182 #include <sys/kernel.h>
4184 #include <ddb/ddb.h>
4187 vm_map_print(vm_map_t map)
4189 vm_map_entry_t entry;
4191 db_iprintf("Task map %p: pmap=%p, nentries=%d, version=%u\n",
4193 (void *)map->pmap, map->nentries, map->timestamp);
4196 for (entry = map->header.next; entry != &map->header;
4197 entry = entry->next) {
4198 db_iprintf("map entry %p: start=%p, end=%p\n",
4199 (void *)entry, (void *)entry->start, (void *)entry->end);
4201 static char *inheritance_name[4] =
4202 {"share", "copy", "none", "donate_copy"};
4204 db_iprintf(" prot=%x/%x/%s",
4206 entry->max_protection,
4207 inheritance_name[(int)(unsigned char)entry->inheritance]);
4208 if (entry->wired_count != 0)
4209 db_printf(", wired");
4211 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
4212 db_printf(", share=%p, offset=0x%jx\n",
4213 (void *)entry->object.sub_map,
4214 (uintmax_t)entry->offset);
4215 if ((entry->prev == &map->header) ||
4216 (entry->prev->object.sub_map !=
4217 entry->object.sub_map)) {
4219 vm_map_print((vm_map_t)entry->object.sub_map);
4223 if (entry->cred != NULL)
4224 db_printf(", ruid %d", entry->cred->cr_ruid);
4225 db_printf(", object=%p, offset=0x%jx",
4226 (void *)entry->object.vm_object,
4227 (uintmax_t)entry->offset);
4228 if (entry->object.vm_object && entry->object.vm_object->cred)
4229 db_printf(", obj ruid %d charge %jx",
4230 entry->object.vm_object->cred->cr_ruid,
4231 (uintmax_t)entry->object.vm_object->charge);
4232 if (entry->eflags & MAP_ENTRY_COW)
4233 db_printf(", copy (%s)",
4234 (entry->eflags & MAP_ENTRY_NEEDS_COPY) ? "needed" : "done");
4237 if ((entry->prev == &map->header) ||
4238 (entry->prev->object.vm_object !=
4239 entry->object.vm_object)) {
4241 vm_object_print((db_expr_t)(intptr_t)
4242 entry->object.vm_object,
4251 DB_SHOW_COMMAND(map, map)
4255 db_printf("usage: show map <addr>\n");
4258 vm_map_print((vm_map_t)addr);
4261 DB_SHOW_COMMAND(procvm, procvm)
4266 p = (struct proc *) addr;
4271 db_printf("p = %p, vmspace = %p, map = %p, pmap = %p\n",
4272 (void *)p, (void *)p->p_vmspace, (void *)&p->p_vmspace->vm_map,
4273 (void *)vmspace_pmap(p->p_vmspace));
4275 vm_map_print((vm_map_t)&p->p_vmspace->vm_map);