2 * Copyright (c) 1991, 1993
3 * The Regents of the University of California. All rights reserved.
5 * This code is derived from software contributed to Berkeley by
6 * The Mach Operating System project at Carnegie-Mellon University.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 4. Neither the name of the University nor the names of its contributors
17 * may be used to endorse or promote products derived from this software
18 * without specific prior written permission.
20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * from: @(#)vm_map.c 8.3 (Berkeley) 1/12/94
35 * Copyright (c) 1987, 1990 Carnegie-Mellon University.
36 * All rights reserved.
38 * Authors: Avadis Tevanian, Jr., Michael Wayne Young
40 * Permission to use, copy, modify and distribute this software and
41 * its documentation is hereby granted, provided that both the copyright
42 * notice and this permission notice appear in all copies of the
43 * software, derivative works or modified versions, and any portions
44 * thereof, and that both notices appear in supporting documentation.
46 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
47 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
48 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
50 * Carnegie Mellon requests users of this software to return to
52 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
53 * School of Computer Science
54 * Carnegie Mellon University
55 * Pittsburgh PA 15213-3890
57 * any improvements or extensions that they make and grant Carnegie the
58 * rights to redistribute these changes.
62 * Virtual memory mapping module.
65 #include <sys/cdefs.h>
66 __FBSDID("$FreeBSD$");
68 #include <sys/param.h>
69 #include <sys/systm.h>
70 #include <sys/kernel.h>
73 #include <sys/mutex.h>
75 #include <sys/vmmeter.h>
77 #include <sys/vnode.h>
78 #include <sys/racct.h>
79 #include <sys/resourcevar.h>
80 #include <sys/rwlock.h>
82 #include <sys/sysctl.h>
83 #include <sys/sysent.h>
87 #include <vm/vm_param.h>
89 #include <vm/vm_map.h>
90 #include <vm/vm_page.h>
91 #include <vm/vm_object.h>
92 #include <vm/vm_pager.h>
93 #include <vm/vm_kern.h>
94 #include <vm/vm_extern.h>
95 #include <vm/vnode_pager.h>
96 #include <vm/swap_pager.h>
100 * Virtual memory maps provide for the mapping, protection,
101 * and sharing of virtual memory objects. In addition,
102 * this module provides for an efficient virtual copy of
103 * memory from one map to another.
105 * Synchronization is required prior to most operations.
107 * Maps consist of an ordered doubly-linked list of simple
108 * entries; a self-adjusting binary search tree of these
109 * entries is used to speed up lookups.
111 * Since portions of maps are specified by start/end addresses,
112 * which may not align with existing map entries, all
113 * routines merely "clip" entries to these start/end values.
114 * [That is, an entry is split into two, bordering at a
115 * start or end value.] Note that these clippings may not
116 * always be necessary (as the two resulting entries are then
117 * not changed); however, the clipping is done for convenience.
119 * As mentioned above, virtual copy operations are performed
120 * by copying VM object references from one map to
121 * another, and then marking both regions as copy-on-write.
124 static struct mtx map_sleep_mtx;
125 static uma_zone_t mapentzone;
126 static uma_zone_t kmapentzone;
127 static uma_zone_t mapzone;
128 static uma_zone_t vmspace_zone;
129 static int vmspace_zinit(void *mem, int size, int flags);
130 static int vm_map_zinit(void *mem, int ize, int flags);
131 static void _vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min,
133 static void vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map);
134 static void vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry);
135 static void vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry);
137 static void vm_map_zdtor(void *mem, int size, void *arg);
138 static void vmspace_zdtor(void *mem, int size, void *arg);
140 static int vm_map_stack_locked(vm_map_t map, vm_offset_t addrbos,
141 vm_size_t max_ssize, vm_size_t growsize, vm_prot_t prot, vm_prot_t max,
143 static void vm_map_wire_entry_failure(vm_map_t map, vm_map_entry_t entry,
144 vm_offset_t failed_addr);
146 #define ENTRY_CHARGED(e) ((e)->cred != NULL || \
147 ((e)->object.vm_object != NULL && (e)->object.vm_object->cred != NULL && \
148 !((e)->eflags & MAP_ENTRY_NEEDS_COPY)))
151 * PROC_VMSPACE_{UN,}LOCK() can be a noop as long as vmspaces are type
154 #define PROC_VMSPACE_LOCK(p) do { } while (0)
155 #define PROC_VMSPACE_UNLOCK(p) do { } while (0)
158 * VM_MAP_RANGE_CHECK: [ internal use only ]
160 * Asserts that the starting and ending region
161 * addresses fall within the valid range of the map.
163 #define VM_MAP_RANGE_CHECK(map, start, end) \
165 if (start < vm_map_min(map)) \
166 start = vm_map_min(map); \
167 if (end > vm_map_max(map)) \
168 end = vm_map_max(map); \
176 * Initialize the vm_map module. Must be called before
177 * any other vm_map routines.
179 * Map and entry structures are allocated from the general
180 * purpose memory pool with some exceptions:
182 * - The kernel map and kmem submap are allocated statically.
183 * - Kernel map entries are allocated out of a static pool.
185 * These restrictions are necessary since malloc() uses the
186 * maps and requires map entries.
192 mtx_init(&map_sleep_mtx, "vm map sleep mutex", NULL, MTX_DEF);
193 mapzone = uma_zcreate("MAP", sizeof(struct vm_map), NULL,
199 vm_map_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
200 uma_prealloc(mapzone, MAX_KMAP);
201 kmapentzone = uma_zcreate("KMAP ENTRY", sizeof(struct vm_map_entry),
202 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR,
203 UMA_ZONE_MTXCLASS | UMA_ZONE_VM);
204 mapentzone = uma_zcreate("MAP ENTRY", sizeof(struct vm_map_entry),
205 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
206 vmspace_zone = uma_zcreate("VMSPACE", sizeof(struct vmspace), NULL,
212 vmspace_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
216 vmspace_zinit(void *mem, int size, int flags)
220 vm = (struct vmspace *)mem;
222 vm->vm_map.pmap = NULL;
223 (void)vm_map_zinit(&vm->vm_map, sizeof(vm->vm_map), flags);
224 PMAP_LOCK_INIT(vmspace_pmap(vm));
229 vm_map_zinit(void *mem, int size, int flags)
234 memset(map, 0, sizeof(*map));
235 mtx_init(&map->system_mtx, "vm map (system)", NULL, MTX_DEF | MTX_DUPOK);
236 sx_init(&map->lock, "vm map (user)");
242 vmspace_zdtor(void *mem, int size, void *arg)
246 vm = (struct vmspace *)mem;
248 vm_map_zdtor(&vm->vm_map, sizeof(vm->vm_map), arg);
251 vm_map_zdtor(void *mem, int size, void *arg)
256 KASSERT(map->nentries == 0,
257 ("map %p nentries == %d on free.",
258 map, map->nentries));
259 KASSERT(map->size == 0,
260 ("map %p size == %lu on free.",
261 map, (unsigned long)map->size));
263 #endif /* INVARIANTS */
266 * Allocate a vmspace structure, including a vm_map and pmap,
267 * and initialize those structures. The refcnt is set to 1.
269 * If 'pinit' is NULL then the embedded pmap is initialized via pmap_pinit().
272 vmspace_alloc(vm_offset_t min, vm_offset_t max, pmap_pinit_t pinit)
276 vm = uma_zalloc(vmspace_zone, M_WAITOK);
278 KASSERT(vm->vm_map.pmap == NULL, ("vm_map.pmap must be NULL"));
283 if (!pinit(vmspace_pmap(vm))) {
284 uma_zfree(vmspace_zone, vm);
287 CTR1(KTR_VM, "vmspace_alloc: %p", vm);
288 _vm_map_init(&vm->vm_map, vmspace_pmap(vm), min, max);
302 vmspace_container_reset(struct proc *p)
307 racct_set(p, RACCT_DATA, 0);
308 racct_set(p, RACCT_STACK, 0);
309 racct_set(p, RACCT_RSS, 0);
310 racct_set(p, RACCT_MEMLOCK, 0);
311 racct_set(p, RACCT_VMEM, 0);
317 vmspace_dofree(struct vmspace *vm)
320 CTR1(KTR_VM, "vmspace_free: %p", vm);
323 * Make sure any SysV shm is freed, it might not have been in
329 * Lock the map, to wait out all other references to it.
330 * Delete all of the mappings and pages they hold, then call
331 * the pmap module to reclaim anything left.
333 (void)vm_map_remove(&vm->vm_map, vm->vm_map.min_offset,
334 vm->vm_map.max_offset);
336 pmap_release(vmspace_pmap(vm));
337 vm->vm_map.pmap = NULL;
338 uma_zfree(vmspace_zone, vm);
342 vmspace_free(struct vmspace *vm)
345 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
346 "vmspace_free() called with non-sleepable lock held");
348 if (vm->vm_refcnt == 0)
349 panic("vmspace_free: attempt to free already freed vmspace");
351 if (atomic_fetchadd_int(&vm->vm_refcnt, -1) == 1)
356 vmspace_exitfree(struct proc *p)
360 PROC_VMSPACE_LOCK(p);
363 PROC_VMSPACE_UNLOCK(p);
364 KASSERT(vm == &vmspace0, ("vmspace_exitfree: wrong vmspace"));
369 vmspace_exit(struct thread *td)
376 * Release user portion of address space.
377 * This releases references to vnodes,
378 * which could cause I/O if the file has been unlinked.
379 * Need to do this early enough that we can still sleep.
381 * The last exiting process to reach this point releases as
382 * much of the environment as it can. vmspace_dofree() is the
383 * slower fallback in case another process had a temporary
384 * reference to the vmspace.
389 atomic_add_int(&vmspace0.vm_refcnt, 1);
391 refcnt = vm->vm_refcnt;
392 if (refcnt > 1 && p->p_vmspace != &vmspace0) {
393 /* Switch now since other proc might free vmspace */
394 PROC_VMSPACE_LOCK(p);
395 p->p_vmspace = &vmspace0;
396 PROC_VMSPACE_UNLOCK(p);
399 } while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt - 1));
401 if (p->p_vmspace != vm) {
402 /* vmspace not yet freed, switch back */
403 PROC_VMSPACE_LOCK(p);
405 PROC_VMSPACE_UNLOCK(p);
408 pmap_remove_pages(vmspace_pmap(vm));
409 /* Switch now since this proc will free vmspace */
410 PROC_VMSPACE_LOCK(p);
411 p->p_vmspace = &vmspace0;
412 PROC_VMSPACE_UNLOCK(p);
416 vmspace_container_reset(p);
419 /* Acquire reference to vmspace owned by another process. */
422 vmspace_acquire_ref(struct proc *p)
427 PROC_VMSPACE_LOCK(p);
430 PROC_VMSPACE_UNLOCK(p);
434 refcnt = vm->vm_refcnt;
435 if (refcnt <= 0) { /* Avoid 0->1 transition */
436 PROC_VMSPACE_UNLOCK(p);
439 } while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt + 1));
440 if (vm != p->p_vmspace) {
441 PROC_VMSPACE_UNLOCK(p);
445 PROC_VMSPACE_UNLOCK(p);
450 _vm_map_lock(vm_map_t map, const char *file, int line)
454 mtx_lock_flags_(&map->system_mtx, 0, file, line);
456 sx_xlock_(&map->lock, file, line);
461 vm_map_process_deferred(void)
464 vm_map_entry_t entry, next;
468 entry = td->td_map_def_user;
469 td->td_map_def_user = NULL;
470 while (entry != NULL) {
472 if ((entry->eflags & MAP_ENTRY_VN_WRITECNT) != 0) {
474 * Decrement the object's writemappings and
475 * possibly the vnode's v_writecount.
477 KASSERT((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0,
478 ("Submap with writecount"));
479 object = entry->object.vm_object;
480 KASSERT(object != NULL, ("No object for writecount"));
481 vnode_pager_release_writecount(object, entry->start,
484 vm_map_entry_deallocate(entry, FALSE);
490 _vm_map_unlock(vm_map_t map, const char *file, int line)
494 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
496 sx_xunlock_(&map->lock, file, line);
497 vm_map_process_deferred();
502 _vm_map_lock_read(vm_map_t map, const char *file, int line)
506 mtx_lock_flags_(&map->system_mtx, 0, file, line);
508 sx_slock_(&map->lock, file, line);
512 _vm_map_unlock_read(vm_map_t map, const char *file, int line)
516 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
518 sx_sunlock_(&map->lock, file, line);
519 vm_map_process_deferred();
524 _vm_map_trylock(vm_map_t map, const char *file, int line)
528 error = map->system_map ?
529 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) :
530 !sx_try_xlock_(&map->lock, file, line);
537 _vm_map_trylock_read(vm_map_t map, const char *file, int line)
541 error = map->system_map ?
542 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) :
543 !sx_try_slock_(&map->lock, file, line);
548 * _vm_map_lock_upgrade: [ internal use only ]
550 * Tries to upgrade a read (shared) lock on the specified map to a write
551 * (exclusive) lock. Returns the value "0" if the upgrade succeeds and a
552 * non-zero value if the upgrade fails. If the upgrade fails, the map is
553 * returned without a read or write lock held.
555 * Requires that the map be read locked.
558 _vm_map_lock_upgrade(vm_map_t map, const char *file, int line)
560 unsigned int last_timestamp;
562 if (map->system_map) {
563 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
565 if (!sx_try_upgrade_(&map->lock, file, line)) {
566 last_timestamp = map->timestamp;
567 sx_sunlock_(&map->lock, file, line);
568 vm_map_process_deferred();
570 * If the map's timestamp does not change while the
571 * map is unlocked, then the upgrade succeeds.
573 sx_xlock_(&map->lock, file, line);
574 if (last_timestamp != map->timestamp) {
575 sx_xunlock_(&map->lock, file, line);
585 _vm_map_lock_downgrade(vm_map_t map, const char *file, int line)
588 if (map->system_map) {
589 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
591 sx_downgrade_(&map->lock, file, line);
597 * Returns a non-zero value if the caller holds a write (exclusive) lock
598 * on the specified map and the value "0" otherwise.
601 vm_map_locked(vm_map_t map)
605 return (mtx_owned(&map->system_mtx));
607 return (sx_xlocked(&map->lock));
612 _vm_map_assert_locked(vm_map_t map, const char *file, int line)
616 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
618 sx_assert_(&map->lock, SA_XLOCKED, file, line);
621 #define VM_MAP_ASSERT_LOCKED(map) \
622 _vm_map_assert_locked(map, LOCK_FILE, LOCK_LINE)
624 #define VM_MAP_ASSERT_LOCKED(map)
628 * _vm_map_unlock_and_wait:
630 * Atomically releases the lock on the specified map and puts the calling
631 * thread to sleep. The calling thread will remain asleep until either
632 * vm_map_wakeup() is performed on the map or the specified timeout is
635 * WARNING! This function does not perform deferred deallocations of
636 * objects and map entries. Therefore, the calling thread is expected to
637 * reacquire the map lock after reawakening and later perform an ordinary
638 * unlock operation, such as vm_map_unlock(), before completing its
639 * operation on the map.
642 _vm_map_unlock_and_wait(vm_map_t map, int timo, const char *file, int line)
645 mtx_lock(&map_sleep_mtx);
647 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
649 sx_xunlock_(&map->lock, file, line);
650 return (msleep(&map->root, &map_sleep_mtx, PDROP | PVM, "vmmaps",
657 * Awaken any threads that have slept on the map using
658 * vm_map_unlock_and_wait().
661 vm_map_wakeup(vm_map_t map)
665 * Acquire and release map_sleep_mtx to prevent a wakeup()
666 * from being performed (and lost) between the map unlock
667 * and the msleep() in _vm_map_unlock_and_wait().
669 mtx_lock(&map_sleep_mtx);
670 mtx_unlock(&map_sleep_mtx);
675 vm_map_busy(vm_map_t map)
678 VM_MAP_ASSERT_LOCKED(map);
683 vm_map_unbusy(vm_map_t map)
686 VM_MAP_ASSERT_LOCKED(map);
687 KASSERT(map->busy, ("vm_map_unbusy: not busy"));
688 if (--map->busy == 0 && (map->flags & MAP_BUSY_WAKEUP)) {
689 vm_map_modflags(map, 0, MAP_BUSY_WAKEUP);
695 vm_map_wait_busy(vm_map_t map)
698 VM_MAP_ASSERT_LOCKED(map);
700 vm_map_modflags(map, MAP_BUSY_WAKEUP, 0);
702 msleep(&map->busy, &map->system_mtx, 0, "mbusy", 0);
704 sx_sleep(&map->busy, &map->lock, 0, "mbusy", 0);
710 vmspace_resident_count(struct vmspace *vmspace)
712 return pmap_resident_count(vmspace_pmap(vmspace));
718 * Creates and returns a new empty VM map with
719 * the given physical map structure, and having
720 * the given lower and upper address bounds.
723 vm_map_create(pmap_t pmap, vm_offset_t min, vm_offset_t max)
727 result = uma_zalloc(mapzone, M_WAITOK);
728 CTR1(KTR_VM, "vm_map_create: %p", result);
729 _vm_map_init(result, pmap, min, max);
734 * Initialize an existing vm_map structure
735 * such as that in the vmspace structure.
738 _vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max)
741 map->header.next = map->header.prev = &map->header;
742 map->needs_wakeup = FALSE;
745 map->min_offset = min;
746 map->max_offset = max;
754 vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max)
757 _vm_map_init(map, pmap, min, max);
758 mtx_init(&map->system_mtx, "system map", NULL, MTX_DEF | MTX_DUPOK);
759 sx_init(&map->lock, "user map");
763 * vm_map_entry_dispose: [ internal use only ]
765 * Inverse of vm_map_entry_create.
768 vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry)
770 uma_zfree(map->system_map ? kmapentzone : mapentzone, entry);
774 * vm_map_entry_create: [ internal use only ]
776 * Allocates a VM map entry for insertion.
777 * No entry fields are filled in.
779 static vm_map_entry_t
780 vm_map_entry_create(vm_map_t map)
782 vm_map_entry_t new_entry;
785 new_entry = uma_zalloc(kmapentzone, M_NOWAIT);
787 new_entry = uma_zalloc(mapentzone, M_WAITOK);
788 if (new_entry == NULL)
789 panic("vm_map_entry_create: kernel resources exhausted");
794 * vm_map_entry_set_behavior:
796 * Set the expected access behavior, either normal, random, or
800 vm_map_entry_set_behavior(vm_map_entry_t entry, u_char behavior)
802 entry->eflags = (entry->eflags & ~MAP_ENTRY_BEHAV_MASK) |
803 (behavior & MAP_ENTRY_BEHAV_MASK);
807 * vm_map_entry_set_max_free:
809 * Set the max_free field in a vm_map_entry.
812 vm_map_entry_set_max_free(vm_map_entry_t entry)
815 entry->max_free = entry->adj_free;
816 if (entry->left != NULL && entry->left->max_free > entry->max_free)
817 entry->max_free = entry->left->max_free;
818 if (entry->right != NULL && entry->right->max_free > entry->max_free)
819 entry->max_free = entry->right->max_free;
823 * vm_map_entry_splay:
825 * The Sleator and Tarjan top-down splay algorithm with the
826 * following variation. Max_free must be computed bottom-up, so
827 * on the downward pass, maintain the left and right spines in
828 * reverse order. Then, make a second pass up each side to fix
829 * the pointers and compute max_free. The time bound is O(log n)
832 * The new root is the vm_map_entry containing "addr", or else an
833 * adjacent entry (lower or higher) if addr is not in the tree.
835 * The map must be locked, and leaves it so.
837 * Returns: the new root.
839 static vm_map_entry_t
840 vm_map_entry_splay(vm_offset_t addr, vm_map_entry_t root)
842 vm_map_entry_t llist, rlist;
843 vm_map_entry_t ltree, rtree;
846 /* Special case of empty tree. */
851 * Pass One: Splay down the tree until we find addr or a NULL
852 * pointer where addr would go. llist and rlist are the two
853 * sides in reverse order (bottom-up), with llist linked by
854 * the right pointer and rlist linked by the left pointer in
855 * the vm_map_entry. Wait until Pass Two to set max_free on
861 /* root is never NULL in here. */
862 if (addr < root->start) {
866 if (addr < y->start && y->left != NULL) {
867 /* Rotate right and put y on rlist. */
868 root->left = y->right;
870 vm_map_entry_set_max_free(root);
875 /* Put root on rlist. */
880 } else if (addr >= root->end) {
884 if (addr >= y->end && y->right != NULL) {
885 /* Rotate left and put y on llist. */
886 root->right = y->left;
888 vm_map_entry_set_max_free(root);
893 /* Put root on llist. */
903 * Pass Two: Walk back up the two spines, flip the pointers
904 * and set max_free. The subtrees of the root go at the
905 * bottom of llist and rlist.
908 while (llist != NULL) {
910 llist->right = ltree;
911 vm_map_entry_set_max_free(llist);
916 while (rlist != NULL) {
919 vm_map_entry_set_max_free(rlist);
925 * Final assembly: add ltree and rtree as subtrees of root.
929 vm_map_entry_set_max_free(root);
935 * vm_map_entry_{un,}link:
937 * Insert/remove entries from maps.
940 vm_map_entry_link(vm_map_t map,
941 vm_map_entry_t after_where,
942 vm_map_entry_t entry)
946 "vm_map_entry_link: map %p, nentries %d, entry %p, after %p", map,
947 map->nentries, entry, after_where);
948 VM_MAP_ASSERT_LOCKED(map);
949 KASSERT(after_where == &map->header ||
950 after_where->end <= entry->start,
951 ("vm_map_entry_link: prev end %jx new start %jx overlap",
952 (uintmax_t)after_where->end, (uintmax_t)entry->start));
953 KASSERT(after_where->next == &map->header ||
954 entry->end <= after_where->next->start,
955 ("vm_map_entry_link: new end %jx next start %jx overlap",
956 (uintmax_t)entry->end, (uintmax_t)after_where->next->start));
959 entry->prev = after_where;
960 entry->next = after_where->next;
961 entry->next->prev = entry;
962 after_where->next = entry;
964 if (after_where != &map->header) {
965 if (after_where != map->root)
966 vm_map_entry_splay(after_where->start, map->root);
967 entry->right = after_where->right;
968 entry->left = after_where;
969 after_where->right = NULL;
970 after_where->adj_free = entry->start - after_where->end;
971 vm_map_entry_set_max_free(after_where);
973 entry->right = map->root;
976 entry->adj_free = (entry->next == &map->header ? map->max_offset :
977 entry->next->start) - entry->end;
978 vm_map_entry_set_max_free(entry);
983 vm_map_entry_unlink(vm_map_t map,
984 vm_map_entry_t entry)
986 vm_map_entry_t next, prev, root;
988 VM_MAP_ASSERT_LOCKED(map);
989 if (entry != map->root)
990 vm_map_entry_splay(entry->start, map->root);
991 if (entry->left == NULL)
994 root = vm_map_entry_splay(entry->start, entry->left);
995 root->right = entry->right;
996 root->adj_free = (entry->next == &map->header ? map->max_offset :
997 entry->next->start) - root->end;
998 vm_map_entry_set_max_free(root);
1007 CTR3(KTR_VM, "vm_map_entry_unlink: map %p, nentries %d, entry %p", map,
1008 map->nentries, entry);
1012 * vm_map_entry_resize_free:
1014 * Recompute the amount of free space following a vm_map_entry
1015 * and propagate that value up the tree. Call this function after
1016 * resizing a map entry in-place, that is, without a call to
1017 * vm_map_entry_link() or _unlink().
1019 * The map must be locked, and leaves it so.
1022 vm_map_entry_resize_free(vm_map_t map, vm_map_entry_t entry)
1026 * Using splay trees without parent pointers, propagating
1027 * max_free up the tree is done by moving the entry to the
1028 * root and making the change there.
1030 if (entry != map->root)
1031 map->root = vm_map_entry_splay(entry->start, map->root);
1033 entry->adj_free = (entry->next == &map->header ? map->max_offset :
1034 entry->next->start) - entry->end;
1035 vm_map_entry_set_max_free(entry);
1039 * vm_map_lookup_entry: [ internal use only ]
1041 * Finds the map entry containing (or
1042 * immediately preceding) the specified address
1043 * in the given map; the entry is returned
1044 * in the "entry" parameter. The boolean
1045 * result indicates whether the address is
1046 * actually contained in the map.
1049 vm_map_lookup_entry(
1051 vm_offset_t address,
1052 vm_map_entry_t *entry) /* OUT */
1058 * If the map is empty, then the map entry immediately preceding
1059 * "address" is the map's header.
1063 *entry = &map->header;
1064 else if (address >= cur->start && cur->end > address) {
1067 } else if ((locked = vm_map_locked(map)) ||
1068 sx_try_upgrade(&map->lock)) {
1070 * Splay requires a write lock on the map. However, it only
1071 * restructures the binary search tree; it does not otherwise
1072 * change the map. Thus, the map's timestamp need not change
1073 * on a temporary upgrade.
1075 map->root = cur = vm_map_entry_splay(address, cur);
1077 sx_downgrade(&map->lock);
1080 * If "address" is contained within a map entry, the new root
1081 * is that map entry. Otherwise, the new root is a map entry
1082 * immediately before or after "address".
1084 if (address >= cur->start) {
1086 if (cur->end > address)
1092 * Since the map is only locked for read access, perform a
1093 * standard binary search tree lookup for "address".
1096 if (address < cur->start) {
1097 if (cur->left == NULL) {
1102 } else if (cur->end > address) {
1106 if (cur->right == NULL) {
1119 * Inserts the given whole VM object into the target
1120 * map at the specified address range. The object's
1121 * size should match that of the address range.
1123 * Requires that the map be locked, and leaves it so.
1125 * If object is non-NULL, ref count must be bumped by caller
1126 * prior to making call to account for the new entry.
1129 vm_map_insert(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1130 vm_offset_t start, vm_offset_t end, vm_prot_t prot, vm_prot_t max,
1133 vm_map_entry_t new_entry;
1134 vm_map_entry_t prev_entry;
1135 vm_map_entry_t temp_entry;
1136 vm_eflags_t protoeflags;
1138 vm_inherit_t inheritance;
1139 boolean_t charge_prev_obj;
1141 VM_MAP_ASSERT_LOCKED(map);
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 charge_prev_obj = FALSE;
1169 if (cow & MAP_COPY_ON_WRITE)
1170 protoeflags |= MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY;
1172 if (cow & MAP_NOFAULT) {
1173 protoeflags |= MAP_ENTRY_NOFAULT;
1175 KASSERT(object == NULL,
1176 ("vm_map_insert: paradoxical MAP_NOFAULT request"));
1178 if (cow & MAP_DISABLE_SYNCER)
1179 protoeflags |= MAP_ENTRY_NOSYNC;
1180 if (cow & MAP_DISABLE_COREDUMP)
1181 protoeflags |= MAP_ENTRY_NOCOREDUMP;
1182 if (cow & MAP_VN_WRITECOUNT)
1183 protoeflags |= MAP_ENTRY_VN_WRITECNT;
1184 if (cow & MAP_INHERIT_SHARE)
1185 inheritance = VM_INHERIT_SHARE;
1187 inheritance = VM_INHERIT_DEFAULT;
1190 KASSERT((object != kmem_object && object != kernel_object) ||
1191 ((object == kmem_object || object == kernel_object) &&
1192 !(protoeflags & MAP_ENTRY_NEEDS_COPY)),
1193 ("kmem or kernel object and cow"));
1194 if (cow & (MAP_ACC_NO_CHARGE | MAP_NOFAULT))
1196 if ((cow & MAP_ACC_CHARGED) || ((prot & VM_PROT_WRITE) &&
1197 ((protoeflags & MAP_ENTRY_NEEDS_COPY) || object == NULL))) {
1198 if (!(cow & MAP_ACC_CHARGED) && !swap_reserve(end - start))
1199 return (KERN_RESOURCE_SHORTAGE);
1200 KASSERT(object == NULL || (protoeflags & MAP_ENTRY_NEEDS_COPY) ||
1201 object->cred == NULL,
1202 ("OVERCOMMIT: vm_map_insert o %p", object));
1203 cred = curthread->td_ucred;
1205 if (object == NULL && !(protoeflags & MAP_ENTRY_NEEDS_COPY))
1206 charge_prev_obj = TRUE;
1210 /* Expand the kernel pmap, if necessary. */
1211 if (map == kernel_map && end > kernel_vm_end)
1212 pmap_growkernel(end);
1213 if (object != NULL) {
1215 * OBJ_ONEMAPPING must be cleared unless this mapping
1216 * is trivially proven to be the only mapping for any
1217 * of the object's pages. (Object granularity
1218 * reference counting is insufficient to recognize
1219 * aliases with precision.)
1221 VM_OBJECT_WLOCK(object);
1222 if (object->ref_count > 1 || object->shadow_count != 0)
1223 vm_object_clear_flag(object, OBJ_ONEMAPPING);
1224 VM_OBJECT_WUNLOCK(object);
1226 else if ((prev_entry != &map->header) &&
1227 (prev_entry->eflags == protoeflags) &&
1228 (cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 &&
1229 (prev_entry->end == start) &&
1230 (prev_entry->wired_count == 0) &&
1231 (prev_entry->cred == cred ||
1232 (prev_entry->object.vm_object != NULL &&
1233 (prev_entry->object.vm_object->cred == cred))) &&
1234 vm_object_coalesce(prev_entry->object.vm_object,
1236 (vm_size_t)(prev_entry->end - prev_entry->start),
1237 (vm_size_t)(end - prev_entry->end), charge_prev_obj)) {
1239 * We were able to extend the object. Determine if we
1240 * can extend the previous map entry to include the
1241 * new range as well.
1243 if ((prev_entry->inheritance == inheritance) &&
1244 (prev_entry->protection == prot) &&
1245 (prev_entry->max_protection == max)) {
1246 map->size += (end - prev_entry->end);
1247 prev_entry->end = end;
1248 vm_map_entry_resize_free(map, prev_entry);
1249 vm_map_simplify_entry(map, prev_entry);
1252 return (KERN_SUCCESS);
1256 * If we can extend the object but cannot extend the
1257 * map entry, we have to create a new map entry. We
1258 * must bump the ref count on the extended object to
1259 * account for it. object may be NULL.
1261 object = prev_entry->object.vm_object;
1262 offset = prev_entry->offset +
1263 (prev_entry->end - prev_entry->start);
1264 vm_object_reference(object);
1265 if (cred != NULL && object != NULL && object->cred != NULL &&
1266 !(prev_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
1267 /* Object already accounts for this uid. */
1274 * NOTE: if conditionals fail, object can be NULL here. This occurs
1275 * in things like the buffer map where we manage kva but do not manage
1280 * Create a new entry
1282 new_entry = vm_map_entry_create(map);
1283 new_entry->start = start;
1284 new_entry->end = end;
1285 new_entry->cred = NULL;
1287 new_entry->eflags = protoeflags;
1288 new_entry->object.vm_object = object;
1289 new_entry->offset = offset;
1290 new_entry->avail_ssize = 0;
1292 new_entry->inheritance = inheritance;
1293 new_entry->protection = prot;
1294 new_entry->max_protection = max;
1295 new_entry->wired_count = 0;
1296 new_entry->wiring_thread = NULL;
1297 new_entry->read_ahead = VM_FAULT_READ_AHEAD_INIT;
1298 new_entry->next_read = OFF_TO_IDX(offset);
1300 KASSERT(cred == NULL || !ENTRY_CHARGED(new_entry),
1301 ("OVERCOMMIT: vm_map_insert leaks vm_map %p", new_entry));
1302 new_entry->cred = cred;
1305 * Insert the new entry into the list
1307 vm_map_entry_link(map, prev_entry, new_entry);
1308 map->size += new_entry->end - new_entry->start;
1311 * It may be possible to merge the new entry with the next and/or
1312 * previous entries. However, due to MAP_STACK_* being a hack, a
1313 * panic can result from merging such entries.
1315 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0)
1316 vm_map_simplify_entry(map, new_entry);
1318 if (cow & (MAP_PREFAULT|MAP_PREFAULT_PARTIAL)) {
1319 vm_map_pmap_enter(map, start, prot,
1320 object, OFF_TO_IDX(offset), end - start,
1321 cow & MAP_PREFAULT_PARTIAL);
1324 return (KERN_SUCCESS);
1330 * Find the first fit (lowest VM address) for "length" free bytes
1331 * beginning at address >= start in the given map.
1333 * In a vm_map_entry, "adj_free" is the amount of free space
1334 * adjacent (higher address) to this entry, and "max_free" is the
1335 * maximum amount of contiguous free space in its subtree. This
1336 * allows finding a free region in one path down the tree, so
1337 * O(log n) amortized with splay trees.
1339 * The map must be locked, and leaves it so.
1341 * Returns: 0 on success, and starting address in *addr,
1342 * 1 if insufficient space.
1345 vm_map_findspace(vm_map_t map, vm_offset_t start, vm_size_t length,
1346 vm_offset_t *addr) /* OUT */
1348 vm_map_entry_t entry;
1352 * Request must fit within min/max VM address and must avoid
1355 if (start < map->min_offset)
1356 start = map->min_offset;
1357 if (start + length > map->max_offset || start + length < start)
1360 /* Empty tree means wide open address space. */
1361 if (map->root == NULL) {
1367 * After splay, if start comes before root node, then there
1368 * must be a gap from start to the root.
1370 map->root = vm_map_entry_splay(start, map->root);
1371 if (start + length <= map->root->start) {
1377 * Root is the last node that might begin its gap before
1378 * start, and this is the last comparison where address
1379 * wrap might be a problem.
1381 st = (start > map->root->end) ? start : map->root->end;
1382 if (length <= map->root->end + map->root->adj_free - st) {
1387 /* With max_free, can immediately tell if no solution. */
1388 entry = map->root->right;
1389 if (entry == NULL || length > entry->max_free)
1393 * Search the right subtree in the order: left subtree, root,
1394 * right subtree (first fit). The previous splay implies that
1395 * all regions in the right subtree have addresses > start.
1397 while (entry != NULL) {
1398 if (entry->left != NULL && entry->left->max_free >= length)
1399 entry = entry->left;
1400 else if (entry->adj_free >= length) {
1404 entry = entry->right;
1407 /* Can't get here, so panic if we do. */
1408 panic("vm_map_findspace: max_free corrupt");
1412 vm_map_fixed(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1413 vm_offset_t start, vm_size_t length, vm_prot_t prot,
1414 vm_prot_t max, int cow)
1419 end = start + length;
1420 KASSERT((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 ||
1422 ("vm_map_fixed: non-NULL backing object for stack"));
1424 VM_MAP_RANGE_CHECK(map, start, end);
1425 if ((cow & MAP_CHECK_EXCL) == 0)
1426 vm_map_delete(map, start, end);
1427 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) != 0) {
1428 result = vm_map_stack_locked(map, start, length, sgrowsiz,
1431 result = vm_map_insert(map, object, offset, start, end,
1439 * vm_map_find finds an unallocated region in the target address
1440 * map with the given length. The search is defined to be
1441 * first-fit from the specified address; the region found is
1442 * returned in the same parameter.
1444 * If object is non-NULL, ref count must be bumped by caller
1445 * prior to making call to account for the new entry.
1448 vm_map_find(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1449 vm_offset_t *addr, /* IN/OUT */
1450 vm_size_t length, vm_offset_t max_addr, int find_space,
1451 vm_prot_t prot, vm_prot_t max, int cow)
1453 vm_offset_t alignment, initial_addr, start;
1456 KASSERT((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 ||
1458 ("vm_map_find: non-NULL backing object for stack"));
1459 if (find_space == VMFS_OPTIMAL_SPACE && (object == NULL ||
1460 (object->flags & OBJ_COLORED) == 0))
1461 find_space = VMFS_ANY_SPACE;
1462 if (find_space >> 8 != 0) {
1463 KASSERT((find_space & 0xff) == 0, ("bad VMFS flags"));
1464 alignment = (vm_offset_t)1 << (find_space >> 8);
1467 initial_addr = *addr;
1469 start = initial_addr;
1472 if (find_space != VMFS_NO_SPACE) {
1473 if (vm_map_findspace(map, start, length, addr) ||
1474 (max_addr != 0 && *addr + length > max_addr)) {
1476 if (find_space == VMFS_OPTIMAL_SPACE) {
1477 find_space = VMFS_ANY_SPACE;
1480 return (KERN_NO_SPACE);
1482 switch (find_space) {
1483 case VMFS_SUPER_SPACE:
1484 case VMFS_OPTIMAL_SPACE:
1485 pmap_align_superpage(object, offset, addr,
1488 case VMFS_ANY_SPACE:
1491 if ((*addr & (alignment - 1)) != 0) {
1492 *addr &= ~(alignment - 1);
1500 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) != 0) {
1501 result = vm_map_stack_locked(map, start, length,
1502 sgrowsiz, prot, max, cow);
1504 result = vm_map_insert(map, object, offset, start,
1505 start + length, prot, max, cow);
1507 } while (result == KERN_NO_SPACE && find_space != VMFS_NO_SPACE &&
1508 find_space != VMFS_ANY_SPACE);
1514 * vm_map_simplify_entry:
1516 * Simplify the given map entry by merging with either neighbor. This
1517 * routine also has the ability to merge with both neighbors.
1519 * The map must be locked.
1521 * This routine guarentees that the passed entry remains valid (though
1522 * possibly extended). When merging, this routine may delete one or
1526 vm_map_simplify_entry(vm_map_t map, vm_map_entry_t entry)
1528 vm_map_entry_t next, prev;
1529 vm_size_t prevsize, esize;
1531 if (entry->eflags & (MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_IS_SUB_MAP))
1535 if (prev != &map->header) {
1536 prevsize = prev->end - prev->start;
1537 if ( (prev->end == entry->start) &&
1538 (prev->object.vm_object == entry->object.vm_object) &&
1539 (!prev->object.vm_object ||
1540 (prev->offset + prevsize == entry->offset)) &&
1541 (prev->eflags == entry->eflags) &&
1542 (prev->protection == entry->protection) &&
1543 (prev->max_protection == entry->max_protection) &&
1544 (prev->inheritance == entry->inheritance) &&
1545 (prev->wired_count == entry->wired_count) &&
1546 (prev->cred == entry->cred)) {
1547 vm_map_entry_unlink(map, prev);
1548 entry->start = prev->start;
1549 entry->offset = prev->offset;
1550 if (entry->prev != &map->header)
1551 vm_map_entry_resize_free(map, entry->prev);
1554 * If the backing object is a vnode object,
1555 * vm_object_deallocate() calls vrele().
1556 * However, vrele() does not lock the vnode
1557 * because the vnode has additional
1558 * references. Thus, the map lock can be kept
1559 * without causing a lock-order reversal with
1562 * Since we count the number of virtual page
1563 * mappings in object->un_pager.vnp.writemappings,
1564 * the writemappings value should not be adjusted
1565 * when the entry is disposed of.
1567 if (prev->object.vm_object)
1568 vm_object_deallocate(prev->object.vm_object);
1569 if (prev->cred != NULL)
1571 vm_map_entry_dispose(map, prev);
1576 if (next != &map->header) {
1577 esize = entry->end - entry->start;
1578 if ((entry->end == next->start) &&
1579 (next->object.vm_object == entry->object.vm_object) &&
1580 (!entry->object.vm_object ||
1581 (entry->offset + esize == next->offset)) &&
1582 (next->eflags == entry->eflags) &&
1583 (next->protection == entry->protection) &&
1584 (next->max_protection == entry->max_protection) &&
1585 (next->inheritance == entry->inheritance) &&
1586 (next->wired_count == entry->wired_count) &&
1587 (next->cred == entry->cred)) {
1588 vm_map_entry_unlink(map, next);
1589 entry->end = next->end;
1590 vm_map_entry_resize_free(map, entry);
1593 * See comment above.
1595 if (next->object.vm_object)
1596 vm_object_deallocate(next->object.vm_object);
1597 if (next->cred != NULL)
1599 vm_map_entry_dispose(map, next);
1604 * vm_map_clip_start: [ internal use only ]
1606 * Asserts that the given entry begins at or after
1607 * the specified address; if necessary,
1608 * it splits the entry into two.
1610 #define vm_map_clip_start(map, entry, startaddr) \
1612 if (startaddr > entry->start) \
1613 _vm_map_clip_start(map, entry, startaddr); \
1617 * This routine is called only when it is known that
1618 * the entry must be split.
1621 _vm_map_clip_start(vm_map_t map, vm_map_entry_t entry, vm_offset_t start)
1623 vm_map_entry_t new_entry;
1625 VM_MAP_ASSERT_LOCKED(map);
1628 * Split off the front portion -- note that we must insert the new
1629 * entry BEFORE this one, so that this entry has the specified
1632 vm_map_simplify_entry(map, entry);
1635 * If there is no object backing this entry, we might as well create
1636 * one now. If we defer it, an object can get created after the map
1637 * is clipped, and individual objects will be created for the split-up
1638 * map. This is a bit of a hack, but is also about the best place to
1639 * put this improvement.
1641 if (entry->object.vm_object == NULL && !map->system_map) {
1643 object = vm_object_allocate(OBJT_DEFAULT,
1644 atop(entry->end - entry->start));
1645 entry->object.vm_object = object;
1647 if (entry->cred != NULL) {
1648 object->cred = entry->cred;
1649 object->charge = entry->end - entry->start;
1652 } else if (entry->object.vm_object != NULL &&
1653 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) &&
1654 entry->cred != NULL) {
1655 VM_OBJECT_WLOCK(entry->object.vm_object);
1656 KASSERT(entry->object.vm_object->cred == NULL,
1657 ("OVERCOMMIT: vm_entry_clip_start: both cred e %p", entry));
1658 entry->object.vm_object->cred = entry->cred;
1659 entry->object.vm_object->charge = entry->end - entry->start;
1660 VM_OBJECT_WUNLOCK(entry->object.vm_object);
1664 new_entry = vm_map_entry_create(map);
1665 *new_entry = *entry;
1667 new_entry->end = start;
1668 entry->offset += (start - entry->start);
1669 entry->start = start;
1670 if (new_entry->cred != NULL)
1671 crhold(entry->cred);
1673 vm_map_entry_link(map, entry->prev, new_entry);
1675 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1676 vm_object_reference(new_entry->object.vm_object);
1678 * The object->un_pager.vnp.writemappings for the
1679 * object of MAP_ENTRY_VN_WRITECNT type entry shall be
1680 * kept as is here. The virtual pages are
1681 * re-distributed among the clipped entries, so the sum is
1688 * vm_map_clip_end: [ internal use only ]
1690 * Asserts that the given entry ends at or before
1691 * the specified address; if necessary,
1692 * it splits the entry into two.
1694 #define vm_map_clip_end(map, entry, endaddr) \
1696 if ((endaddr) < (entry->end)) \
1697 _vm_map_clip_end((map), (entry), (endaddr)); \
1701 * This routine is called only when it is known that
1702 * the entry must be split.
1705 _vm_map_clip_end(vm_map_t map, vm_map_entry_t entry, vm_offset_t end)
1707 vm_map_entry_t new_entry;
1709 VM_MAP_ASSERT_LOCKED(map);
1712 * If there is no object backing this entry, we might as well create
1713 * one now. If we defer it, an object can get created after the map
1714 * is clipped, and individual objects will be created for the split-up
1715 * map. This is a bit of a hack, but is also about the best place to
1716 * put this improvement.
1718 if (entry->object.vm_object == NULL && !map->system_map) {
1720 object = vm_object_allocate(OBJT_DEFAULT,
1721 atop(entry->end - entry->start));
1722 entry->object.vm_object = object;
1724 if (entry->cred != NULL) {
1725 object->cred = entry->cred;
1726 object->charge = entry->end - entry->start;
1729 } else if (entry->object.vm_object != NULL &&
1730 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) &&
1731 entry->cred != NULL) {
1732 VM_OBJECT_WLOCK(entry->object.vm_object);
1733 KASSERT(entry->object.vm_object->cred == NULL,
1734 ("OVERCOMMIT: vm_entry_clip_end: both cred e %p", entry));
1735 entry->object.vm_object->cred = entry->cred;
1736 entry->object.vm_object->charge = entry->end - entry->start;
1737 VM_OBJECT_WUNLOCK(entry->object.vm_object);
1742 * Create a new entry and insert it AFTER the specified entry
1744 new_entry = vm_map_entry_create(map);
1745 *new_entry = *entry;
1747 new_entry->start = entry->end = end;
1748 new_entry->offset += (end - entry->start);
1749 if (new_entry->cred != NULL)
1750 crhold(entry->cred);
1752 vm_map_entry_link(map, entry, new_entry);
1754 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1755 vm_object_reference(new_entry->object.vm_object);
1760 * vm_map_submap: [ kernel use only ]
1762 * Mark the given range as handled by a subordinate map.
1764 * This range must have been created with vm_map_find,
1765 * and no other operations may have been performed on this
1766 * range prior to calling vm_map_submap.
1768 * Only a limited number of operations can be performed
1769 * within this rage after calling vm_map_submap:
1771 * [Don't try vm_map_copy!]
1773 * To remove a submapping, one must first remove the
1774 * range from the superior map, and then destroy the
1775 * submap (if desired). [Better yet, don't try it.]
1784 vm_map_entry_t entry;
1785 int result = KERN_INVALID_ARGUMENT;
1789 VM_MAP_RANGE_CHECK(map, start, end);
1791 if (vm_map_lookup_entry(map, start, &entry)) {
1792 vm_map_clip_start(map, entry, start);
1794 entry = entry->next;
1796 vm_map_clip_end(map, entry, end);
1798 if ((entry->start == start) && (entry->end == end) &&
1799 ((entry->eflags & MAP_ENTRY_COW) == 0) &&
1800 (entry->object.vm_object == NULL)) {
1801 entry->object.sub_map = submap;
1802 entry->eflags |= MAP_ENTRY_IS_SUB_MAP;
1803 result = KERN_SUCCESS;
1811 * The maximum number of pages to map if MAP_PREFAULT_PARTIAL is specified
1813 #define MAX_INIT_PT 96
1816 * vm_map_pmap_enter:
1818 * Preload the specified map's pmap with mappings to the specified
1819 * object's memory-resident pages. No further physical pages are
1820 * allocated, and no further virtual pages are retrieved from secondary
1821 * storage. If the specified flags include MAP_PREFAULT_PARTIAL, then a
1822 * limited number of page mappings are created at the low-end of the
1823 * specified address range. (For this purpose, a superpage mapping
1824 * counts as one page mapping.) Otherwise, all resident pages within
1825 * the specified address range are mapped. Because these mappings are
1826 * being created speculatively, cached pages are not reactivated and
1830 vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot,
1831 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags)
1834 vm_page_t p, p_start;
1835 vm_pindex_t mask, psize, threshold, tmpidx;
1837 if ((prot & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0 || object == NULL)
1839 VM_OBJECT_RLOCK(object);
1840 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) {
1841 VM_OBJECT_RUNLOCK(object);
1842 VM_OBJECT_WLOCK(object);
1843 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) {
1844 pmap_object_init_pt(map->pmap, addr, object, pindex,
1846 VM_OBJECT_WUNLOCK(object);
1849 VM_OBJECT_LOCK_DOWNGRADE(object);
1853 if (psize + pindex > object->size) {
1854 if (object->size < pindex) {
1855 VM_OBJECT_RUNLOCK(object);
1858 psize = object->size - pindex;
1863 threshold = MAX_INIT_PT;
1865 p = vm_page_find_least(object, pindex);
1867 * Assert: the variable p is either (1) the page with the
1868 * least pindex greater than or equal to the parameter pindex
1872 p != NULL && (tmpidx = p->pindex - pindex) < psize;
1873 p = TAILQ_NEXT(p, listq)) {
1875 * don't allow an madvise to blow away our really
1876 * free pages allocating pv entries.
1878 if (((flags & MAP_PREFAULT_MADVISE) != 0 &&
1879 cnt.v_free_count < cnt.v_free_reserved) ||
1880 ((flags & MAP_PREFAULT_PARTIAL) != 0 &&
1881 tmpidx >= threshold)) {
1885 if (p->valid == VM_PAGE_BITS_ALL) {
1886 if (p_start == NULL) {
1887 start = addr + ptoa(tmpidx);
1890 /* Jump ahead if a superpage mapping is possible. */
1891 if (p->psind > 0 && ((addr + ptoa(tmpidx)) &
1892 (pagesizes[p->psind] - 1)) == 0) {
1893 mask = atop(pagesizes[p->psind]) - 1;
1894 if (tmpidx + mask < psize &&
1895 vm_page_ps_is_valid(p)) {
1900 } else if (p_start != NULL) {
1901 pmap_enter_object(map->pmap, start, addr +
1902 ptoa(tmpidx), p_start, prot);
1906 if (p_start != NULL)
1907 pmap_enter_object(map->pmap, start, addr + ptoa(psize),
1909 VM_OBJECT_RUNLOCK(object);
1915 * Sets the protection of the specified address
1916 * region in the target map. If "set_max" is
1917 * specified, the maximum protection is to be set;
1918 * otherwise, only the current protection is affected.
1921 vm_map_protect(vm_map_t map, vm_offset_t start, vm_offset_t end,
1922 vm_prot_t new_prot, boolean_t set_max)
1924 vm_map_entry_t current, entry;
1930 return (KERN_SUCCESS);
1934 VM_MAP_RANGE_CHECK(map, start, end);
1936 if (vm_map_lookup_entry(map, start, &entry)) {
1937 vm_map_clip_start(map, entry, start);
1939 entry = entry->next;
1943 * Make a first pass to check for protection violations.
1946 while ((current != &map->header) && (current->start < end)) {
1947 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
1949 return (KERN_INVALID_ARGUMENT);
1951 if ((new_prot & current->max_protection) != new_prot) {
1953 return (KERN_PROTECTION_FAILURE);
1955 current = current->next;
1960 * Do an accounting pass for private read-only mappings that
1961 * now will do cow due to allowed write (e.g. debugger sets
1962 * breakpoint on text segment)
1964 for (current = entry; (current != &map->header) &&
1965 (current->start < end); current = current->next) {
1967 vm_map_clip_end(map, current, end);
1970 ((new_prot & ~(current->protection)) & VM_PROT_WRITE) == 0 ||
1971 ENTRY_CHARGED(current)) {
1975 cred = curthread->td_ucred;
1976 obj = current->object.vm_object;
1978 if (obj == NULL || (current->eflags & MAP_ENTRY_NEEDS_COPY)) {
1979 if (!swap_reserve(current->end - current->start)) {
1981 return (KERN_RESOURCE_SHORTAGE);
1984 current->cred = cred;
1988 VM_OBJECT_WLOCK(obj);
1989 if (obj->type != OBJT_DEFAULT && obj->type != OBJT_SWAP) {
1990 VM_OBJECT_WUNLOCK(obj);
1995 * Charge for the whole object allocation now, since
1996 * we cannot distinguish between non-charged and
1997 * charged clipped mapping of the same object later.
1999 KASSERT(obj->charge == 0,
2000 ("vm_map_protect: object %p overcharged (entry %p)",
2002 if (!swap_reserve(ptoa(obj->size))) {
2003 VM_OBJECT_WUNLOCK(obj);
2005 return (KERN_RESOURCE_SHORTAGE);
2010 obj->charge = ptoa(obj->size);
2011 VM_OBJECT_WUNLOCK(obj);
2015 * Go back and fix up protections. [Note that clipping is not
2016 * necessary the second time.]
2019 while ((current != &map->header) && (current->start < end)) {
2020 old_prot = current->protection;
2023 current->protection =
2024 (current->max_protection = new_prot) &
2027 current->protection = new_prot;
2030 * For user wired map entries, the normal lazy evaluation of
2031 * write access upgrades through soft page faults is
2032 * undesirable. Instead, immediately copy any pages that are
2033 * copy-on-write and enable write access in the physical map.
2035 if ((current->eflags & MAP_ENTRY_USER_WIRED) != 0 &&
2036 (current->protection & VM_PROT_WRITE) != 0 &&
2037 (old_prot & VM_PROT_WRITE) == 0)
2038 vm_fault_copy_entry(map, map, current, current, NULL);
2041 * When restricting access, update the physical map. Worry
2042 * about copy-on-write here.
2044 if ((old_prot & ~current->protection) != 0) {
2045 #define MASK(entry) (((entry)->eflags & MAP_ENTRY_COW) ? ~VM_PROT_WRITE : \
2047 pmap_protect(map->pmap, current->start,
2049 current->protection & MASK(current));
2052 vm_map_simplify_entry(map, current);
2053 current = current->next;
2056 return (KERN_SUCCESS);
2062 * This routine traverses a processes map handling the madvise
2063 * system call. Advisories are classified as either those effecting
2064 * the vm_map_entry structure, or those effecting the underlying
2074 vm_map_entry_t current, entry;
2078 * Some madvise calls directly modify the vm_map_entry, in which case
2079 * we need to use an exclusive lock on the map and we need to perform
2080 * various clipping operations. Otherwise we only need a read-lock
2085 case MADV_SEQUENTIAL:
2092 return (KERN_SUCCESS);
2100 return (KERN_SUCCESS);
2101 vm_map_lock_read(map);
2104 return (KERN_INVALID_ARGUMENT);
2108 * Locate starting entry and clip if necessary.
2110 VM_MAP_RANGE_CHECK(map, start, end);
2112 if (vm_map_lookup_entry(map, start, &entry)) {
2114 vm_map_clip_start(map, entry, start);
2116 entry = entry->next;
2121 * madvise behaviors that are implemented in the vm_map_entry.
2123 * We clip the vm_map_entry so that behavioral changes are
2124 * limited to the specified address range.
2126 for (current = entry;
2127 (current != &map->header) && (current->start < end);
2128 current = current->next
2130 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
2133 vm_map_clip_end(map, current, end);
2137 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_NORMAL);
2139 case MADV_SEQUENTIAL:
2140 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_SEQUENTIAL);
2143 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_RANDOM);
2146 current->eflags |= MAP_ENTRY_NOSYNC;
2149 current->eflags &= ~MAP_ENTRY_NOSYNC;
2152 current->eflags |= MAP_ENTRY_NOCOREDUMP;
2155 current->eflags &= ~MAP_ENTRY_NOCOREDUMP;
2160 vm_map_simplify_entry(map, current);
2164 vm_pindex_t pstart, pend;
2167 * madvise behaviors that are implemented in the underlying
2170 * Since we don't clip the vm_map_entry, we have to clip
2171 * the vm_object pindex and count.
2173 for (current = entry;
2174 (current != &map->header) && (current->start < end);
2175 current = current->next
2177 vm_offset_t useEnd, useStart;
2179 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
2182 pstart = OFF_TO_IDX(current->offset);
2183 pend = pstart + atop(current->end - current->start);
2184 useStart = current->start;
2185 useEnd = current->end;
2187 if (current->start < start) {
2188 pstart += atop(start - current->start);
2191 if (current->end > end) {
2192 pend -= atop(current->end - end);
2200 * Perform the pmap_advise() before clearing
2201 * PGA_REFERENCED in vm_page_advise(). Otherwise, a
2202 * concurrent pmap operation, such as pmap_remove(),
2203 * could clear a reference in the pmap and set
2204 * PGA_REFERENCED on the page before the pmap_advise()
2205 * had completed. Consequently, the page would appear
2206 * referenced based upon an old reference that
2207 * occurred before this pmap_advise() ran.
2209 if (behav == MADV_DONTNEED || behav == MADV_FREE)
2210 pmap_advise(map->pmap, useStart, useEnd,
2213 vm_object_madvise(current->object.vm_object, pstart,
2217 * Pre-populate paging structures in the
2218 * WILLNEED case. For wired entries, the
2219 * paging structures are already populated.
2221 if (behav == MADV_WILLNEED &&
2222 current->wired_count == 0) {
2223 vm_map_pmap_enter(map,
2225 current->protection,
2226 current->object.vm_object,
2228 ptoa(pend - pstart),
2229 MAP_PREFAULT_MADVISE
2233 vm_map_unlock_read(map);
2242 * Sets the inheritance of the specified address
2243 * range in the target map. Inheritance
2244 * affects how the map will be shared with
2245 * child maps at the time of vmspace_fork.
2248 vm_map_inherit(vm_map_t map, vm_offset_t start, vm_offset_t end,
2249 vm_inherit_t new_inheritance)
2251 vm_map_entry_t entry;
2252 vm_map_entry_t temp_entry;
2254 switch (new_inheritance) {
2255 case VM_INHERIT_NONE:
2256 case VM_INHERIT_COPY:
2257 case VM_INHERIT_SHARE:
2260 return (KERN_INVALID_ARGUMENT);
2263 return (KERN_SUCCESS);
2265 VM_MAP_RANGE_CHECK(map, start, end);
2266 if (vm_map_lookup_entry(map, start, &temp_entry)) {
2268 vm_map_clip_start(map, entry, start);
2270 entry = temp_entry->next;
2271 while ((entry != &map->header) && (entry->start < end)) {
2272 vm_map_clip_end(map, entry, end);
2273 entry->inheritance = new_inheritance;
2274 vm_map_simplify_entry(map, entry);
2275 entry = entry->next;
2278 return (KERN_SUCCESS);
2284 * Implements both kernel and user unwiring.
2287 vm_map_unwire(vm_map_t map, vm_offset_t start, vm_offset_t end,
2290 vm_map_entry_t entry, first_entry, tmp_entry;
2291 vm_offset_t saved_start;
2292 unsigned int last_timestamp;
2294 boolean_t need_wakeup, result, user_unwire;
2297 return (KERN_SUCCESS);
2298 user_unwire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE;
2300 VM_MAP_RANGE_CHECK(map, start, end);
2301 if (!vm_map_lookup_entry(map, start, &first_entry)) {
2302 if (flags & VM_MAP_WIRE_HOLESOK)
2303 first_entry = first_entry->next;
2306 return (KERN_INVALID_ADDRESS);
2309 last_timestamp = map->timestamp;
2310 entry = first_entry;
2311 while (entry != &map->header && entry->start < end) {
2312 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
2314 * We have not yet clipped the entry.
2316 saved_start = (start >= entry->start) ? start :
2318 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2319 if (vm_map_unlock_and_wait(map, 0)) {
2321 * Allow interruption of user unwiring?
2325 if (last_timestamp+1 != map->timestamp) {
2327 * Look again for the entry because the map was
2328 * modified while it was unlocked.
2329 * Specifically, the entry may have been
2330 * clipped, merged, or deleted.
2332 if (!vm_map_lookup_entry(map, saved_start,
2334 if (flags & VM_MAP_WIRE_HOLESOK)
2335 tmp_entry = tmp_entry->next;
2337 if (saved_start == start) {
2339 * First_entry has been deleted.
2342 return (KERN_INVALID_ADDRESS);
2345 rv = KERN_INVALID_ADDRESS;
2349 if (entry == first_entry)
2350 first_entry = tmp_entry;
2355 last_timestamp = map->timestamp;
2358 vm_map_clip_start(map, entry, start);
2359 vm_map_clip_end(map, entry, end);
2361 * Mark the entry in case the map lock is released. (See
2364 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 &&
2365 entry->wiring_thread == NULL,
2366 ("owned map entry %p", entry));
2367 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
2368 entry->wiring_thread = curthread;
2370 * Check the map for holes in the specified region.
2371 * If VM_MAP_WIRE_HOLESOK was specified, skip this check.
2373 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) &&
2374 (entry->end < end && (entry->next == &map->header ||
2375 entry->next->start > entry->end))) {
2377 rv = KERN_INVALID_ADDRESS;
2381 * If system unwiring, require that the entry is system wired.
2384 vm_map_entry_system_wired_count(entry) == 0) {
2386 rv = KERN_INVALID_ARGUMENT;
2389 entry = entry->next;
2393 need_wakeup = FALSE;
2394 if (first_entry == NULL) {
2395 result = vm_map_lookup_entry(map, start, &first_entry);
2396 if (!result && (flags & VM_MAP_WIRE_HOLESOK))
2397 first_entry = first_entry->next;
2399 KASSERT(result, ("vm_map_unwire: lookup failed"));
2401 for (entry = first_entry; entry != &map->header && entry->start < end;
2402 entry = entry->next) {
2404 * If VM_MAP_WIRE_HOLESOK was specified, an empty
2405 * space in the unwired region could have been mapped
2406 * while the map lock was dropped for draining
2407 * MAP_ENTRY_IN_TRANSITION. Moreover, another thread
2408 * could be simultaneously wiring this new mapping
2409 * entry. Detect these cases and skip any entries
2410 * marked as in transition by us.
2412 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 ||
2413 entry->wiring_thread != curthread) {
2414 KASSERT((flags & VM_MAP_WIRE_HOLESOK) != 0,
2415 ("vm_map_unwire: !HOLESOK and new/changed entry"));
2419 if (rv == KERN_SUCCESS && (!user_unwire ||
2420 (entry->eflags & MAP_ENTRY_USER_WIRED))) {
2422 entry->eflags &= ~MAP_ENTRY_USER_WIRED;
2423 if (entry->wired_count == 1)
2424 vm_map_entry_unwire(map, entry);
2426 entry->wired_count--;
2428 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0,
2429 ("vm_map_unwire: in-transition flag missing %p", entry));
2430 KASSERT(entry->wiring_thread == curthread,
2431 ("vm_map_unwire: alien wire %p", entry));
2432 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION;
2433 entry->wiring_thread = NULL;
2434 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
2435 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
2438 vm_map_simplify_entry(map, entry);
2447 * vm_map_wire_entry_failure:
2449 * Handle a wiring failure on the given entry.
2451 * The map should be locked.
2454 vm_map_wire_entry_failure(vm_map_t map, vm_map_entry_t entry,
2455 vm_offset_t failed_addr)
2458 VM_MAP_ASSERT_LOCKED(map);
2459 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 &&
2460 entry->wired_count == 1,
2461 ("vm_map_wire_entry_failure: entry %p isn't being wired", entry));
2462 KASSERT(failed_addr < entry->end,
2463 ("vm_map_wire_entry_failure: entry %p was fully wired", entry));
2466 * If any pages at the start of this entry were successfully wired,
2469 if (failed_addr > entry->start) {
2470 pmap_unwire(map->pmap, entry->start, failed_addr);
2471 vm_object_unwire(entry->object.vm_object, entry->offset,
2472 failed_addr - entry->start, PQ_ACTIVE);
2476 * Assign an out-of-range value to represent the failure to wire this
2479 entry->wired_count = -1;
2485 * Implements both kernel and user wiring.
2488 vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t end,
2491 vm_map_entry_t entry, first_entry, tmp_entry;
2492 vm_offset_t faddr, saved_end, saved_start;
2493 unsigned int last_timestamp;
2495 boolean_t need_wakeup, result, user_wire;
2499 return (KERN_SUCCESS);
2501 if (flags & VM_MAP_WIRE_WRITE)
2502 prot |= VM_PROT_WRITE;
2503 user_wire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE;
2505 VM_MAP_RANGE_CHECK(map, start, end);
2506 if (!vm_map_lookup_entry(map, start, &first_entry)) {
2507 if (flags & VM_MAP_WIRE_HOLESOK)
2508 first_entry = first_entry->next;
2511 return (KERN_INVALID_ADDRESS);
2514 last_timestamp = map->timestamp;
2515 entry = first_entry;
2516 while (entry != &map->header && entry->start < end) {
2517 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
2519 * We have not yet clipped the entry.
2521 saved_start = (start >= entry->start) ? start :
2523 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2524 if (vm_map_unlock_and_wait(map, 0)) {
2526 * Allow interruption of user wiring?
2530 if (last_timestamp + 1 != map->timestamp) {
2532 * Look again for the entry because the map was
2533 * modified while it was unlocked.
2534 * Specifically, the entry may have been
2535 * clipped, merged, or deleted.
2537 if (!vm_map_lookup_entry(map, saved_start,
2539 if (flags & VM_MAP_WIRE_HOLESOK)
2540 tmp_entry = tmp_entry->next;
2542 if (saved_start == start) {
2544 * first_entry has been deleted.
2547 return (KERN_INVALID_ADDRESS);
2550 rv = KERN_INVALID_ADDRESS;
2554 if (entry == first_entry)
2555 first_entry = tmp_entry;
2560 last_timestamp = map->timestamp;
2563 vm_map_clip_start(map, entry, start);
2564 vm_map_clip_end(map, entry, end);
2566 * Mark the entry in case the map lock is released. (See
2569 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 &&
2570 entry->wiring_thread == NULL,
2571 ("owned map entry %p", entry));
2572 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
2573 entry->wiring_thread = curthread;
2574 if ((entry->protection & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0
2575 || (entry->protection & prot) != prot) {
2576 entry->eflags |= MAP_ENTRY_WIRE_SKIPPED;
2577 if ((flags & VM_MAP_WIRE_HOLESOK) == 0) {
2579 rv = KERN_INVALID_ADDRESS;
2584 if (entry->wired_count == 0) {
2585 entry->wired_count++;
2586 saved_start = entry->start;
2587 saved_end = entry->end;
2590 * Release the map lock, relying on the in-transition
2591 * mark. Mark the map busy for fork.
2596 faddr = saved_start;
2599 * Simulate a fault to get the page and enter
2600 * it into the physical map.
2602 if ((rv = vm_fault(map, faddr, VM_PROT_NONE,
2603 VM_FAULT_CHANGE_WIRING)) != KERN_SUCCESS)
2605 } while ((faddr += PAGE_SIZE) < saved_end);
2608 if (last_timestamp + 1 != map->timestamp) {
2610 * Look again for the entry because the map was
2611 * modified while it was unlocked. The entry
2612 * may have been clipped, but NOT merged or
2615 result = vm_map_lookup_entry(map, saved_start,
2617 KASSERT(result, ("vm_map_wire: lookup failed"));
2618 if (entry == first_entry)
2619 first_entry = tmp_entry;
2623 while (entry->end < saved_end) {
2625 * In case of failure, handle entries
2626 * that were not fully wired here;
2627 * fully wired entries are handled
2630 if (rv != KERN_SUCCESS &&
2632 vm_map_wire_entry_failure(map,
2634 entry = entry->next;
2637 last_timestamp = map->timestamp;
2638 if (rv != KERN_SUCCESS) {
2639 vm_map_wire_entry_failure(map, entry, faddr);
2643 } else if (!user_wire ||
2644 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
2645 entry->wired_count++;
2648 * Check the map for holes in the specified region.
2649 * If VM_MAP_WIRE_HOLESOK was specified, skip this check.
2652 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) &&
2653 (entry->end < end && (entry->next == &map->header ||
2654 entry->next->start > entry->end))) {
2656 rv = KERN_INVALID_ADDRESS;
2659 entry = entry->next;
2663 need_wakeup = FALSE;
2664 if (first_entry == NULL) {
2665 result = vm_map_lookup_entry(map, start, &first_entry);
2666 if (!result && (flags & VM_MAP_WIRE_HOLESOK))
2667 first_entry = first_entry->next;
2669 KASSERT(result, ("vm_map_wire: lookup failed"));
2671 for (entry = first_entry; entry != &map->header && entry->start < end;
2672 entry = entry->next) {
2673 if ((entry->eflags & MAP_ENTRY_WIRE_SKIPPED) != 0)
2674 goto next_entry_done;
2677 * If VM_MAP_WIRE_HOLESOK was specified, an empty
2678 * space in the unwired region could have been mapped
2679 * while the map lock was dropped for faulting in the
2680 * pages or draining MAP_ENTRY_IN_TRANSITION.
2681 * Moreover, another thread could be simultaneously
2682 * wiring this new mapping entry. Detect these cases
2683 * and skip any entries marked as in transition by us.
2685 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 ||
2686 entry->wiring_thread != curthread) {
2687 KASSERT((flags & VM_MAP_WIRE_HOLESOK) != 0,
2688 ("vm_map_wire: !HOLESOK and new/changed entry"));
2692 if (rv == KERN_SUCCESS) {
2694 entry->eflags |= MAP_ENTRY_USER_WIRED;
2695 } else if (entry->wired_count == -1) {
2697 * Wiring failed on this entry. Thus, unwiring is
2700 entry->wired_count = 0;
2701 } else if (!user_wire ||
2702 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
2704 * Undo the wiring. Wiring succeeded on this entry
2705 * but failed on a later entry.
2707 if (entry->wired_count == 1)
2708 vm_map_entry_unwire(map, entry);
2710 entry->wired_count--;
2713 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0,
2714 ("vm_map_wire: in-transition flag missing %p", entry));
2715 KASSERT(entry->wiring_thread == curthread,
2716 ("vm_map_wire: alien wire %p", entry));
2717 entry->eflags &= ~(MAP_ENTRY_IN_TRANSITION |
2718 MAP_ENTRY_WIRE_SKIPPED);
2719 entry->wiring_thread = NULL;
2720 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
2721 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
2724 vm_map_simplify_entry(map, entry);
2735 * Push any dirty cached pages in the address range to their pager.
2736 * If syncio is TRUE, dirty pages are written synchronously.
2737 * If invalidate is TRUE, any cached pages are freed as well.
2739 * If the size of the region from start to end is zero, we are
2740 * supposed to flush all modified pages within the region containing
2741 * start. Unfortunately, a region can be split or coalesced with
2742 * neighboring regions, making it difficult to determine what the
2743 * original region was. Therefore, we approximate this requirement by
2744 * flushing the current region containing start.
2746 * Returns an error if any part of the specified range is not mapped.
2754 boolean_t invalidate)
2756 vm_map_entry_t current;
2757 vm_map_entry_t entry;
2760 vm_ooffset_t offset;
2761 unsigned int last_timestamp;
2764 vm_map_lock_read(map);
2765 VM_MAP_RANGE_CHECK(map, start, end);
2766 if (!vm_map_lookup_entry(map, start, &entry)) {
2767 vm_map_unlock_read(map);
2768 return (KERN_INVALID_ADDRESS);
2769 } else if (start == end) {
2770 start = entry->start;
2774 * Make a first pass to check for user-wired memory and holes.
2776 for (current = entry; current != &map->header && current->start < end;
2777 current = current->next) {
2778 if (invalidate && (current->eflags & MAP_ENTRY_USER_WIRED)) {
2779 vm_map_unlock_read(map);
2780 return (KERN_INVALID_ARGUMENT);
2782 if (end > current->end &&
2783 (current->next == &map->header ||
2784 current->end != current->next->start)) {
2785 vm_map_unlock_read(map);
2786 return (KERN_INVALID_ADDRESS);
2791 pmap_remove(map->pmap, start, end);
2795 * Make a second pass, cleaning/uncaching pages from the indicated
2798 for (current = entry; current != &map->header && current->start < end;) {
2799 offset = current->offset + (start - current->start);
2800 size = (end <= current->end ? end : current->end) - start;
2801 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
2803 vm_map_entry_t tentry;
2806 smap = current->object.sub_map;
2807 vm_map_lock_read(smap);
2808 (void) vm_map_lookup_entry(smap, offset, &tentry);
2809 tsize = tentry->end - offset;
2812 object = tentry->object.vm_object;
2813 offset = tentry->offset + (offset - tentry->start);
2814 vm_map_unlock_read(smap);
2816 object = current->object.vm_object;
2818 vm_object_reference(object);
2819 last_timestamp = map->timestamp;
2820 vm_map_unlock_read(map);
2821 if (!vm_object_sync(object, offset, size, syncio, invalidate))
2824 vm_object_deallocate(object);
2825 vm_map_lock_read(map);
2826 if (last_timestamp == map->timestamp ||
2827 !vm_map_lookup_entry(map, start, ¤t))
2828 current = current->next;
2831 vm_map_unlock_read(map);
2832 return (failed ? KERN_FAILURE : KERN_SUCCESS);
2836 * vm_map_entry_unwire: [ internal use only ]
2838 * Make the region specified by this entry pageable.
2840 * The map in question should be locked.
2841 * [This is the reason for this routine's existence.]
2844 vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry)
2847 VM_MAP_ASSERT_LOCKED(map);
2848 KASSERT(entry->wired_count > 0,
2849 ("vm_map_entry_unwire: entry %p isn't wired", entry));
2850 pmap_unwire(map->pmap, entry->start, entry->end);
2851 vm_object_unwire(entry->object.vm_object, entry->offset, entry->end -
2852 entry->start, PQ_ACTIVE);
2853 entry->wired_count = 0;
2857 vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map)
2860 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0)
2861 vm_object_deallocate(entry->object.vm_object);
2862 uma_zfree(system_map ? kmapentzone : mapentzone, entry);
2866 * vm_map_entry_delete: [ internal use only ]
2868 * Deallocate the given entry from the target map.
2871 vm_map_entry_delete(vm_map_t map, vm_map_entry_t entry)
2874 vm_pindex_t offidxstart, offidxend, count, size1;
2877 vm_map_entry_unlink(map, entry);
2878 object = entry->object.vm_object;
2879 size = entry->end - entry->start;
2882 if (entry->cred != NULL) {
2883 swap_release_by_cred(size, entry->cred);
2884 crfree(entry->cred);
2887 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0 &&
2889 KASSERT(entry->cred == NULL || object->cred == NULL ||
2890 (entry->eflags & MAP_ENTRY_NEEDS_COPY),
2891 ("OVERCOMMIT vm_map_entry_delete: both cred %p", entry));
2892 count = OFF_TO_IDX(size);
2893 offidxstart = OFF_TO_IDX(entry->offset);
2894 offidxend = offidxstart + count;
2895 VM_OBJECT_WLOCK(object);
2896 if (object->ref_count != 1 &&
2897 ((object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING ||
2898 object == kernel_object || object == kmem_object)) {
2899 vm_object_collapse(object);
2902 * The option OBJPR_NOTMAPPED can be passed here
2903 * because vm_map_delete() already performed
2904 * pmap_remove() on the only mapping to this range
2907 vm_object_page_remove(object, offidxstart, offidxend,
2909 if (object->type == OBJT_SWAP)
2910 swap_pager_freespace(object, offidxstart, count);
2911 if (offidxend >= object->size &&
2912 offidxstart < object->size) {
2913 size1 = object->size;
2914 object->size = offidxstart;
2915 if (object->cred != NULL) {
2916 size1 -= object->size;
2917 KASSERT(object->charge >= ptoa(size1),
2918 ("vm_map_entry_delete: object->charge < 0"));
2919 swap_release_by_cred(ptoa(size1), object->cred);
2920 object->charge -= ptoa(size1);
2924 VM_OBJECT_WUNLOCK(object);
2926 entry->object.vm_object = NULL;
2927 if (map->system_map)
2928 vm_map_entry_deallocate(entry, TRUE);
2930 entry->next = curthread->td_map_def_user;
2931 curthread->td_map_def_user = entry;
2936 * vm_map_delete: [ internal use only ]
2938 * Deallocates the given address range from the target
2942 vm_map_delete(vm_map_t map, vm_offset_t start, vm_offset_t end)
2944 vm_map_entry_t entry;
2945 vm_map_entry_t first_entry;
2947 VM_MAP_ASSERT_LOCKED(map);
2949 return (KERN_SUCCESS);
2952 * Find the start of the region, and clip it
2954 if (!vm_map_lookup_entry(map, start, &first_entry))
2955 entry = first_entry->next;
2957 entry = first_entry;
2958 vm_map_clip_start(map, entry, start);
2962 * Step through all entries in this region
2964 while ((entry != &map->header) && (entry->start < end)) {
2965 vm_map_entry_t next;
2968 * Wait for wiring or unwiring of an entry to complete.
2969 * Also wait for any system wirings to disappear on
2972 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 ||
2973 (vm_map_pmap(map) != kernel_pmap &&
2974 vm_map_entry_system_wired_count(entry) != 0)) {
2975 unsigned int last_timestamp;
2976 vm_offset_t saved_start;
2977 vm_map_entry_t tmp_entry;
2979 saved_start = entry->start;
2980 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2981 last_timestamp = map->timestamp;
2982 (void) vm_map_unlock_and_wait(map, 0);
2984 if (last_timestamp + 1 != map->timestamp) {
2986 * Look again for the entry because the map was
2987 * modified while it was unlocked.
2988 * Specifically, the entry may have been
2989 * clipped, merged, or deleted.
2991 if (!vm_map_lookup_entry(map, saved_start,
2993 entry = tmp_entry->next;
2996 vm_map_clip_start(map, entry,
3002 vm_map_clip_end(map, entry, end);
3007 * Unwire before removing addresses from the pmap; otherwise,
3008 * unwiring will put the entries back in the pmap.
3010 if (entry->wired_count != 0) {
3011 vm_map_entry_unwire(map, entry);
3014 pmap_remove(map->pmap, entry->start, entry->end);
3017 * Delete the entry only after removing all pmap
3018 * entries pointing to its pages. (Otherwise, its
3019 * page frames may be reallocated, and any modify bits
3020 * will be set in the wrong object!)
3022 vm_map_entry_delete(map, entry);
3025 return (KERN_SUCCESS);
3031 * Remove the given address range from the target map.
3032 * This is the exported form of vm_map_delete.
3035 vm_map_remove(vm_map_t map, vm_offset_t start, vm_offset_t end)
3040 VM_MAP_RANGE_CHECK(map, start, end);
3041 result = vm_map_delete(map, start, end);
3047 * vm_map_check_protection:
3049 * Assert that the target map allows the specified privilege on the
3050 * entire address region given. The entire region must be allocated.
3052 * WARNING! This code does not and should not check whether the
3053 * contents of the region is accessible. For example a smaller file
3054 * might be mapped into a larger address space.
3056 * NOTE! This code is also called by munmap().
3058 * The map must be locked. A read lock is sufficient.
3061 vm_map_check_protection(vm_map_t map, vm_offset_t start, vm_offset_t end,
3062 vm_prot_t protection)
3064 vm_map_entry_t entry;
3065 vm_map_entry_t tmp_entry;
3067 if (!vm_map_lookup_entry(map, start, &tmp_entry))
3071 while (start < end) {
3072 if (entry == &map->header)
3077 if (start < entry->start)
3080 * Check protection associated with entry.
3082 if ((entry->protection & protection) != protection)
3084 /* go to next entry */
3086 entry = entry->next;
3092 * vm_map_copy_entry:
3094 * Copies the contents of the source entry to the destination
3095 * entry. The entries *must* be aligned properly.
3101 vm_map_entry_t src_entry,
3102 vm_map_entry_t dst_entry,
3103 vm_ooffset_t *fork_charge)
3105 vm_object_t src_object;
3106 vm_map_entry_t fake_entry;
3111 VM_MAP_ASSERT_LOCKED(dst_map);
3113 if ((dst_entry->eflags|src_entry->eflags) & MAP_ENTRY_IS_SUB_MAP)
3116 if (src_entry->wired_count == 0 ||
3117 (src_entry->protection & VM_PROT_WRITE) == 0) {
3119 * If the source entry is marked needs_copy, it is already
3122 if ((src_entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0 &&
3123 (src_entry->protection & VM_PROT_WRITE) != 0) {
3124 pmap_protect(src_map->pmap,
3127 src_entry->protection & ~VM_PROT_WRITE);
3131 * Make a copy of the object.
3133 size = src_entry->end - src_entry->start;
3134 if ((src_object = src_entry->object.vm_object) != NULL) {
3135 VM_OBJECT_WLOCK(src_object);
3136 charged = ENTRY_CHARGED(src_entry);
3137 if ((src_object->handle == NULL) &&
3138 (src_object->type == OBJT_DEFAULT ||
3139 src_object->type == OBJT_SWAP)) {
3140 vm_object_collapse(src_object);
3141 if ((src_object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING) {
3142 vm_object_split(src_entry);
3143 src_object = src_entry->object.vm_object;
3146 vm_object_reference_locked(src_object);
3147 vm_object_clear_flag(src_object, OBJ_ONEMAPPING);
3148 if (src_entry->cred != NULL &&
3149 !(src_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
3150 KASSERT(src_object->cred == NULL,
3151 ("OVERCOMMIT: vm_map_copy_entry: cred %p",
3153 src_object->cred = src_entry->cred;
3154 src_object->charge = size;
3156 VM_OBJECT_WUNLOCK(src_object);
3157 dst_entry->object.vm_object = src_object;
3159 cred = curthread->td_ucred;
3161 dst_entry->cred = cred;
3162 *fork_charge += size;
3163 if (!(src_entry->eflags &
3164 MAP_ENTRY_NEEDS_COPY)) {
3166 src_entry->cred = cred;
3167 *fork_charge += size;
3170 src_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY);
3171 dst_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY);
3172 dst_entry->offset = src_entry->offset;
3173 if (src_entry->eflags & MAP_ENTRY_VN_WRITECNT) {
3175 * MAP_ENTRY_VN_WRITECNT cannot
3176 * indicate write reference from
3177 * src_entry, since the entry is
3178 * marked as needs copy. Allocate a
3179 * fake entry that is used to
3180 * decrement object->un_pager.vnp.writecount
3181 * at the appropriate time. Attach
3182 * fake_entry to the deferred list.
3184 fake_entry = vm_map_entry_create(dst_map);
3185 fake_entry->eflags = MAP_ENTRY_VN_WRITECNT;
3186 src_entry->eflags &= ~MAP_ENTRY_VN_WRITECNT;
3187 vm_object_reference(src_object);
3188 fake_entry->object.vm_object = src_object;
3189 fake_entry->start = src_entry->start;
3190 fake_entry->end = src_entry->end;
3191 fake_entry->next = curthread->td_map_def_user;
3192 curthread->td_map_def_user = fake_entry;
3195 dst_entry->object.vm_object = NULL;
3196 dst_entry->offset = 0;
3197 if (src_entry->cred != NULL) {
3198 dst_entry->cred = curthread->td_ucred;
3199 crhold(dst_entry->cred);
3200 *fork_charge += size;
3204 pmap_copy(dst_map->pmap, src_map->pmap, dst_entry->start,
3205 dst_entry->end - dst_entry->start, src_entry->start);
3208 * We don't want to make writeable wired pages copy-on-write.
3209 * Immediately copy these pages into the new map by simulating
3210 * page faults. The new pages are pageable.
3212 vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry,
3218 * vmspace_map_entry_forked:
3219 * Update the newly-forked vmspace each time a map entry is inherited
3220 * or copied. The values for vm_dsize and vm_tsize are approximate
3221 * (and mostly-obsolete ideas in the face of mmap(2) et al.)
3224 vmspace_map_entry_forked(const struct vmspace *vm1, struct vmspace *vm2,
3225 vm_map_entry_t entry)
3227 vm_size_t entrysize;
3230 entrysize = entry->end - entry->start;
3231 vm2->vm_map.size += entrysize;
3232 if (entry->eflags & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP)) {
3233 vm2->vm_ssize += btoc(entrysize);
3234 } else if (entry->start >= (vm_offset_t)vm1->vm_daddr &&
3235 entry->start < (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)) {
3236 newend = MIN(entry->end,
3237 (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize));
3238 vm2->vm_dsize += btoc(newend - entry->start);
3239 } else if (entry->start >= (vm_offset_t)vm1->vm_taddr &&
3240 entry->start < (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)) {
3241 newend = MIN(entry->end,
3242 (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize));
3243 vm2->vm_tsize += btoc(newend - entry->start);
3249 * Create a new process vmspace structure and vm_map
3250 * based on those of an existing process. The new map
3251 * is based on the old map, according to the inheritance
3252 * values on the regions in that map.
3254 * XXX It might be worth coalescing the entries added to the new vmspace.
3256 * The source map must not be locked.
3259 vmspace_fork(struct vmspace *vm1, vm_ooffset_t *fork_charge)
3261 struct vmspace *vm2;
3262 vm_map_t new_map, old_map;
3263 vm_map_entry_t new_entry, old_entry;
3267 old_map = &vm1->vm_map;
3268 /* Copy immutable fields of vm1 to vm2. */
3269 vm2 = vmspace_alloc(old_map->min_offset, old_map->max_offset, NULL);
3272 vm2->vm_taddr = vm1->vm_taddr;
3273 vm2->vm_daddr = vm1->vm_daddr;
3274 vm2->vm_maxsaddr = vm1->vm_maxsaddr;
3275 vm_map_lock(old_map);
3277 vm_map_wait_busy(old_map);
3278 new_map = &vm2->vm_map;
3279 locked = vm_map_trylock(new_map); /* trylock to silence WITNESS */
3280 KASSERT(locked, ("vmspace_fork: lock failed"));
3282 old_entry = old_map->header.next;
3284 while (old_entry != &old_map->header) {
3285 if (old_entry->eflags & MAP_ENTRY_IS_SUB_MAP)
3286 panic("vm_map_fork: encountered a submap");
3288 switch (old_entry->inheritance) {
3289 case VM_INHERIT_NONE:
3292 case VM_INHERIT_SHARE:
3294 * Clone the entry, creating the shared object if necessary.
3296 object = old_entry->object.vm_object;
3297 if (object == NULL) {
3298 object = vm_object_allocate(OBJT_DEFAULT,
3299 atop(old_entry->end - old_entry->start));
3300 old_entry->object.vm_object = object;
3301 old_entry->offset = 0;
3302 if (old_entry->cred != NULL) {
3303 object->cred = old_entry->cred;
3304 object->charge = old_entry->end -
3306 old_entry->cred = NULL;
3311 * Add the reference before calling vm_object_shadow
3312 * to insure that a shadow object is created.
3314 vm_object_reference(object);
3315 if (old_entry->eflags & MAP_ENTRY_NEEDS_COPY) {
3316 vm_object_shadow(&old_entry->object.vm_object,
3318 old_entry->end - old_entry->start);
3319 old_entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
3320 /* Transfer the second reference too. */
3321 vm_object_reference(
3322 old_entry->object.vm_object);
3325 * As in vm_map_simplify_entry(), the
3326 * vnode lock will not be acquired in
3327 * this call to vm_object_deallocate().
3329 vm_object_deallocate(object);
3330 object = old_entry->object.vm_object;
3332 VM_OBJECT_WLOCK(object);
3333 vm_object_clear_flag(object, OBJ_ONEMAPPING);
3334 if (old_entry->cred != NULL) {
3335 KASSERT(object->cred == NULL, ("vmspace_fork both cred"));
3336 object->cred = old_entry->cred;
3337 object->charge = old_entry->end - old_entry->start;
3338 old_entry->cred = NULL;
3342 * Assert the correct state of the vnode
3343 * v_writecount while the object is locked, to
3344 * not relock it later for the assertion
3347 if (old_entry->eflags & MAP_ENTRY_VN_WRITECNT &&
3348 object->type == OBJT_VNODE) {
3349 KASSERT(((struct vnode *)object->handle)->
3351 ("vmspace_fork: v_writecount %p", object));
3352 KASSERT(object->un_pager.vnp.writemappings > 0,
3353 ("vmspace_fork: vnp.writecount %p",
3356 VM_OBJECT_WUNLOCK(object);
3359 * Clone the entry, referencing the shared object.
3361 new_entry = vm_map_entry_create(new_map);
3362 *new_entry = *old_entry;
3363 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED |
3364 MAP_ENTRY_IN_TRANSITION);
3365 new_entry->wiring_thread = NULL;
3366 new_entry->wired_count = 0;
3367 if (new_entry->eflags & MAP_ENTRY_VN_WRITECNT) {
3368 vnode_pager_update_writecount(object,
3369 new_entry->start, new_entry->end);
3373 * Insert the entry into the new map -- we know we're
3374 * inserting at the end of the new map.
3376 vm_map_entry_link(new_map, new_map->header.prev,
3378 vmspace_map_entry_forked(vm1, vm2, new_entry);
3381 * Update the physical map
3383 pmap_copy(new_map->pmap, old_map->pmap,
3385 (old_entry->end - old_entry->start),
3389 case VM_INHERIT_COPY:
3391 * Clone the entry and link into the map.
3393 new_entry = vm_map_entry_create(new_map);
3394 *new_entry = *old_entry;
3396 * Copied entry is COW over the old object.
3398 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED |
3399 MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_VN_WRITECNT);
3400 new_entry->wiring_thread = NULL;
3401 new_entry->wired_count = 0;
3402 new_entry->object.vm_object = NULL;
3403 new_entry->cred = NULL;
3404 vm_map_entry_link(new_map, new_map->header.prev,
3406 vmspace_map_entry_forked(vm1, vm2, new_entry);
3407 vm_map_copy_entry(old_map, new_map, old_entry,
3408 new_entry, fork_charge);
3411 old_entry = old_entry->next;
3414 * Use inlined vm_map_unlock() to postpone handling the deferred
3415 * map entries, which cannot be done until both old_map and
3416 * new_map locks are released.
3418 sx_xunlock(&old_map->lock);
3419 sx_xunlock(&new_map->lock);
3420 vm_map_process_deferred();
3426 vm_map_stack(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
3427 vm_prot_t prot, vm_prot_t max, int cow)
3429 vm_size_t growsize, init_ssize;
3430 rlim_t lmemlim, vmemlim;
3433 growsize = sgrowsiz;
3434 init_ssize = (max_ssize < growsize) ? max_ssize : growsize;
3437 lmemlim = lim_cur(curproc, RLIMIT_MEMLOCK);
3438 vmemlim = lim_cur(curproc, RLIMIT_VMEM);
3439 PROC_UNLOCK(curproc);
3440 if (!old_mlock && map->flags & MAP_WIREFUTURE) {
3441 if (ptoa(pmap_wired_count(map->pmap)) + init_ssize > lmemlim) {
3446 /* If we would blow our VMEM resource limit, no go */
3447 if (map->size + init_ssize > vmemlim) {
3451 rv = vm_map_stack_locked(map, addrbos, max_ssize, growsize, prot,
3459 vm_map_stack_locked(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
3460 vm_size_t growsize, vm_prot_t prot, vm_prot_t max, int cow)
3462 vm_map_entry_t new_entry, prev_entry;
3463 vm_offset_t bot, top;
3464 vm_size_t init_ssize;
3468 * The stack orientation is piggybacked with the cow argument.
3469 * Extract it into orient and mask the cow argument so that we
3470 * don't pass it around further.
3471 * NOTE: We explicitly allow bi-directional stacks.
3473 orient = cow & (MAP_STACK_GROWS_DOWN|MAP_STACK_GROWS_UP);
3474 KASSERT(orient != 0, ("No stack grow direction"));
3476 if (addrbos < vm_map_min(map) ||
3477 addrbos > vm_map_max(map) ||
3478 addrbos + max_ssize < addrbos)
3479 return (KERN_NO_SPACE);
3481 init_ssize = (max_ssize < growsize) ? max_ssize : growsize;
3483 /* If addr is already mapped, no go */
3484 if (vm_map_lookup_entry(map, addrbos, &prev_entry))
3485 return (KERN_NO_SPACE);
3488 * If we can't accomodate max_ssize in the current mapping, no go.
3489 * However, we need to be aware that subsequent user mappings might
3490 * map into the space we have reserved for stack, and currently this
3491 * space is not protected.
3493 * Hopefully we will at least detect this condition when we try to
3496 if ((prev_entry->next != &map->header) &&
3497 (prev_entry->next->start < addrbos + max_ssize))
3498 return (KERN_NO_SPACE);
3501 * We initially map a stack of only init_ssize. We will grow as
3502 * needed later. Depending on the orientation of the stack (i.e.
3503 * the grow direction) we either map at the top of the range, the
3504 * bottom of the range or in the middle.
3506 * Note: we would normally expect prot and max to be VM_PROT_ALL,
3507 * and cow to be 0. Possibly we should eliminate these as input
3508 * parameters, and just pass these values here in the insert call.
3510 if (orient == MAP_STACK_GROWS_DOWN)
3511 bot = addrbos + max_ssize - init_ssize;
3512 else if (orient == MAP_STACK_GROWS_UP)
3515 bot = round_page(addrbos + max_ssize/2 - init_ssize/2);
3516 top = bot + init_ssize;
3517 rv = vm_map_insert(map, NULL, 0, bot, top, prot, max, cow);
3519 /* Now set the avail_ssize amount. */
3520 if (rv == KERN_SUCCESS) {
3521 if (prev_entry != &map->header)
3522 vm_map_clip_end(map, prev_entry, bot);
3523 new_entry = prev_entry->next;
3524 if (new_entry->end != top || new_entry->start != bot)
3525 panic("Bad entry start/end for new stack entry");
3527 new_entry->avail_ssize = max_ssize - init_ssize;
3528 if (orient & MAP_STACK_GROWS_DOWN)
3529 new_entry->eflags |= MAP_ENTRY_GROWS_DOWN;
3530 if (orient & MAP_STACK_GROWS_UP)
3531 new_entry->eflags |= MAP_ENTRY_GROWS_UP;
3537 static int stack_guard_page = 0;
3538 TUNABLE_INT("security.bsd.stack_guard_page", &stack_guard_page);
3539 SYSCTL_INT(_security_bsd, OID_AUTO, stack_guard_page, CTLFLAG_RW,
3540 &stack_guard_page, 0,
3541 "Insert stack guard page ahead of the growable segments.");
3543 /* Attempts to grow a vm stack entry. Returns KERN_SUCCESS if the
3544 * desired address is already mapped, or if we successfully grow
3545 * the stack. Also returns KERN_SUCCESS if addr is outside the
3546 * stack range (this is strange, but preserves compatibility with
3547 * the grow function in vm_machdep.c).
3550 vm_map_growstack(struct proc *p, vm_offset_t addr)
3552 vm_map_entry_t next_entry, prev_entry;
3553 vm_map_entry_t new_entry, stack_entry;
3554 struct vmspace *vm = p->p_vmspace;
3555 vm_map_t map = &vm->vm_map;
3558 size_t grow_amount, max_grow;
3559 rlim_t lmemlim, stacklim, vmemlim;
3560 int is_procstack, rv;
3571 lmemlim = lim_cur(p, RLIMIT_MEMLOCK);
3572 stacklim = lim_cur(p, RLIMIT_STACK);
3573 vmemlim = lim_cur(p, RLIMIT_VMEM);
3576 vm_map_lock_read(map);
3578 /* If addr is already in the entry range, no need to grow.*/
3579 if (vm_map_lookup_entry(map, addr, &prev_entry)) {
3580 vm_map_unlock_read(map);
3581 return (KERN_SUCCESS);
3584 next_entry = prev_entry->next;
3585 if (!(prev_entry->eflags & MAP_ENTRY_GROWS_UP)) {
3587 * This entry does not grow upwards. Since the address lies
3588 * beyond this entry, the next entry (if one exists) has to
3589 * be a downward growable entry. The entry list header is
3590 * never a growable entry, so it suffices to check the flags.
3592 if (!(next_entry->eflags & MAP_ENTRY_GROWS_DOWN)) {
3593 vm_map_unlock_read(map);
3594 return (KERN_SUCCESS);
3596 stack_entry = next_entry;
3599 * This entry grows upward. If the next entry does not at
3600 * least grow downwards, this is the entry we need to grow.
3601 * otherwise we have two possible choices and we have to
3604 if (next_entry->eflags & MAP_ENTRY_GROWS_DOWN) {
3606 * We have two choices; grow the entry closest to
3607 * the address to minimize the amount of growth.
3609 if (addr - prev_entry->end <= next_entry->start - addr)
3610 stack_entry = prev_entry;
3612 stack_entry = next_entry;
3614 stack_entry = prev_entry;
3617 if (stack_entry == next_entry) {
3618 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_DOWN, ("foo"));
3619 KASSERT(addr < stack_entry->start, ("foo"));
3620 end = (prev_entry != &map->header) ? prev_entry->end :
3621 stack_entry->start - stack_entry->avail_ssize;
3622 grow_amount = roundup(stack_entry->start - addr, PAGE_SIZE);
3623 max_grow = stack_entry->start - end;
3625 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_UP, ("foo"));
3626 KASSERT(addr >= stack_entry->end, ("foo"));
3627 end = (next_entry != &map->header) ? next_entry->start :
3628 stack_entry->end + stack_entry->avail_ssize;
3629 grow_amount = roundup(addr + 1 - stack_entry->end, PAGE_SIZE);
3630 max_grow = end - stack_entry->end;
3633 if (grow_amount > stack_entry->avail_ssize) {
3634 vm_map_unlock_read(map);
3635 return (KERN_NO_SPACE);
3639 * If there is no longer enough space between the entries nogo, and
3640 * adjust the available space. Note: this should only happen if the
3641 * user has mapped into the stack area after the stack was created,
3642 * and is probably an error.
3644 * This also effectively destroys any guard page the user might have
3645 * intended by limiting the stack size.
3647 if (grow_amount + (stack_guard_page ? PAGE_SIZE : 0) > max_grow) {
3648 if (vm_map_lock_upgrade(map))
3651 stack_entry->avail_ssize = max_grow;
3654 return (KERN_NO_SPACE);
3657 is_procstack = (addr >= (vm_offset_t)vm->vm_maxsaddr) ? 1 : 0;
3660 * If this is the main process stack, see if we're over the stack
3663 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) {
3664 vm_map_unlock_read(map);
3665 return (KERN_NO_SPACE);
3670 racct_set(p, RACCT_STACK, ctob(vm->vm_ssize) + grow_amount)) {
3672 vm_map_unlock_read(map);
3673 return (KERN_NO_SPACE);
3678 /* Round up the grow amount modulo sgrowsiz */
3679 growsize = sgrowsiz;
3680 grow_amount = roundup(grow_amount, growsize);
3681 if (grow_amount > stack_entry->avail_ssize)
3682 grow_amount = stack_entry->avail_ssize;
3683 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) {
3684 grow_amount = trunc_page((vm_size_t)stacklim) -
3689 limit = racct_get_available(p, RACCT_STACK);
3691 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > limit))
3692 grow_amount = limit - ctob(vm->vm_ssize);
3694 if (!old_mlock && map->flags & MAP_WIREFUTURE) {
3695 if (ptoa(pmap_wired_count(map->pmap)) + grow_amount > lmemlim) {
3696 vm_map_unlock_read(map);
3702 if (racct_set(p, RACCT_MEMLOCK,
3703 ptoa(pmap_wired_count(map->pmap)) + grow_amount)) {
3705 vm_map_unlock_read(map);
3712 /* If we would blow our VMEM resource limit, no go */
3713 if (map->size + grow_amount > vmemlim) {
3714 vm_map_unlock_read(map);
3720 if (racct_set(p, RACCT_VMEM, map->size + grow_amount)) {
3722 vm_map_unlock_read(map);
3729 if (vm_map_lock_upgrade(map))
3732 if (stack_entry == next_entry) {
3736 /* Get the preliminary new entry start value */
3737 addr = stack_entry->start - grow_amount;
3740 * If this puts us into the previous entry, cut back our
3741 * growth to the available space. Also, see the note above.
3744 stack_entry->avail_ssize = max_grow;
3746 if (stack_guard_page)
3750 rv = vm_map_insert(map, NULL, 0, addr, stack_entry->start,
3751 next_entry->protection, next_entry->max_protection, 0);
3753 /* Adjust the available stack space by the amount we grew. */
3754 if (rv == KERN_SUCCESS) {
3755 if (prev_entry != &map->header)
3756 vm_map_clip_end(map, prev_entry, addr);
3757 new_entry = prev_entry->next;
3758 KASSERT(new_entry == stack_entry->prev, ("foo"));
3759 KASSERT(new_entry->end == stack_entry->start, ("foo"));
3760 KASSERT(new_entry->start == addr, ("foo"));
3761 grow_amount = new_entry->end - new_entry->start;
3762 new_entry->avail_ssize = stack_entry->avail_ssize -
3764 stack_entry->eflags &= ~MAP_ENTRY_GROWS_DOWN;
3765 new_entry->eflags |= MAP_ENTRY_GROWS_DOWN;
3771 addr = stack_entry->end + grow_amount;
3774 * If this puts us into the next entry, cut back our growth
3775 * to the available space. Also, see the note above.
3778 stack_entry->avail_ssize = end - stack_entry->end;
3780 if (stack_guard_page)
3784 grow_amount = addr - stack_entry->end;
3785 cred = stack_entry->cred;
3786 if (cred == NULL && stack_entry->object.vm_object != NULL)
3787 cred = stack_entry->object.vm_object->cred;
3788 if (cred != NULL && !swap_reserve_by_cred(grow_amount, cred))
3790 /* Grow the underlying object if applicable. */
3791 else if (stack_entry->object.vm_object == NULL ||
3792 vm_object_coalesce(stack_entry->object.vm_object,
3793 stack_entry->offset,
3794 (vm_size_t)(stack_entry->end - stack_entry->start),
3795 (vm_size_t)grow_amount, cred != NULL)) {
3796 map->size += (addr - stack_entry->end);
3797 /* Update the current entry. */
3798 stack_entry->end = addr;
3799 stack_entry->avail_ssize -= grow_amount;
3800 vm_map_entry_resize_free(map, stack_entry);
3803 if (next_entry != &map->header)
3804 vm_map_clip_start(map, next_entry, addr);
3809 if (rv == KERN_SUCCESS && is_procstack)
3810 vm->vm_ssize += btoc(grow_amount);
3815 * Heed the MAP_WIREFUTURE flag if it was set for this process.
3817 if (rv == KERN_SUCCESS && (map->flags & MAP_WIREFUTURE)) {
3819 (stack_entry == next_entry) ? addr : addr - grow_amount,
3820 (stack_entry == next_entry) ? stack_entry->start : addr,
3821 (p->p_flag & P_SYSTEM)
3822 ? VM_MAP_WIRE_SYSTEM|VM_MAP_WIRE_NOHOLES
3823 : VM_MAP_WIRE_USER|VM_MAP_WIRE_NOHOLES);
3828 if (rv != KERN_SUCCESS) {
3830 error = racct_set(p, RACCT_VMEM, map->size);
3831 KASSERT(error == 0, ("decreasing RACCT_VMEM failed"));
3833 error = racct_set(p, RACCT_MEMLOCK,
3834 ptoa(pmap_wired_count(map->pmap)));
3835 KASSERT(error == 0, ("decreasing RACCT_MEMLOCK failed"));
3837 error = racct_set(p, RACCT_STACK, ctob(vm->vm_ssize));
3838 KASSERT(error == 0, ("decreasing RACCT_STACK failed"));
3847 * Unshare the specified VM space for exec. If other processes are
3848 * mapped to it, then create a new one. The new vmspace is null.
3851 vmspace_exec(struct proc *p, vm_offset_t minuser, vm_offset_t maxuser)
3853 struct vmspace *oldvmspace = p->p_vmspace;
3854 struct vmspace *newvmspace;
3856 KASSERT((curthread->td_pflags & TDP_EXECVMSPC) == 0,
3857 ("vmspace_exec recursed"));
3858 newvmspace = vmspace_alloc(minuser, maxuser, NULL);
3859 if (newvmspace == NULL)
3861 newvmspace->vm_swrss = oldvmspace->vm_swrss;
3863 * This code is written like this for prototype purposes. The
3864 * goal is to avoid running down the vmspace here, but let the
3865 * other process's that are still using the vmspace to finally
3866 * run it down. Even though there is little or no chance of blocking
3867 * here, it is a good idea to keep this form for future mods.
3869 PROC_VMSPACE_LOCK(p);
3870 p->p_vmspace = newvmspace;
3871 PROC_VMSPACE_UNLOCK(p);
3872 if (p == curthread->td_proc)
3873 pmap_activate(curthread);
3874 curthread->td_pflags |= TDP_EXECVMSPC;
3879 * Unshare the specified VM space for forcing COW. This
3880 * is called by rfork, for the (RFMEM|RFPROC) == 0 case.
3883 vmspace_unshare(struct proc *p)
3885 struct vmspace *oldvmspace = p->p_vmspace;
3886 struct vmspace *newvmspace;
3887 vm_ooffset_t fork_charge;
3889 if (oldvmspace->vm_refcnt == 1)
3892 newvmspace = vmspace_fork(oldvmspace, &fork_charge);
3893 if (newvmspace == NULL)
3895 if (!swap_reserve_by_cred(fork_charge, p->p_ucred)) {
3896 vmspace_free(newvmspace);
3899 PROC_VMSPACE_LOCK(p);
3900 p->p_vmspace = newvmspace;
3901 PROC_VMSPACE_UNLOCK(p);
3902 if (p == curthread->td_proc)
3903 pmap_activate(curthread);
3904 vmspace_free(oldvmspace);
3911 * Finds the VM object, offset, and
3912 * protection for a given virtual address in the
3913 * specified map, assuming a page fault of the
3916 * Leaves the map in question locked for read; return
3917 * values are guaranteed until a vm_map_lookup_done
3918 * call is performed. Note that the map argument
3919 * is in/out; the returned map must be used in
3920 * the call to vm_map_lookup_done.
3922 * A handle (out_entry) is returned for use in
3923 * vm_map_lookup_done, to make that fast.
3925 * If a lookup is requested with "write protection"
3926 * specified, the map may be changed to perform virtual
3927 * copying operations, although the data referenced will
3931 vm_map_lookup(vm_map_t *var_map, /* IN/OUT */
3933 vm_prot_t fault_typea,
3934 vm_map_entry_t *out_entry, /* OUT */
3935 vm_object_t *object, /* OUT */
3936 vm_pindex_t *pindex, /* OUT */
3937 vm_prot_t *out_prot, /* OUT */
3938 boolean_t *wired) /* OUT */
3940 vm_map_entry_t entry;
3941 vm_map_t map = *var_map;
3943 vm_prot_t fault_type = fault_typea;
3944 vm_object_t eobject;
3950 vm_map_lock_read(map);
3953 * Lookup the faulting address.
3955 if (!vm_map_lookup_entry(map, vaddr, out_entry)) {
3956 vm_map_unlock_read(map);
3957 return (KERN_INVALID_ADDRESS);
3965 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
3966 vm_map_t old_map = map;
3968 *var_map = map = entry->object.sub_map;
3969 vm_map_unlock_read(old_map);
3974 * Check whether this task is allowed to have this page.
3976 prot = entry->protection;
3977 fault_type &= (VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE);
3978 if ((fault_type & prot) != fault_type || prot == VM_PROT_NONE) {
3979 vm_map_unlock_read(map);
3980 return (KERN_PROTECTION_FAILURE);
3982 if ((entry->eflags & MAP_ENTRY_USER_WIRED) &&
3983 (entry->eflags & MAP_ENTRY_COW) &&
3984 (fault_type & VM_PROT_WRITE)) {
3985 vm_map_unlock_read(map);
3986 return (KERN_PROTECTION_FAILURE);
3988 if ((fault_typea & VM_PROT_COPY) != 0 &&
3989 (entry->max_protection & VM_PROT_WRITE) == 0 &&
3990 (entry->eflags & MAP_ENTRY_COW) == 0) {
3991 vm_map_unlock_read(map);
3992 return (KERN_PROTECTION_FAILURE);
3996 * If this page is not pageable, we have to get it for all possible
3999 *wired = (entry->wired_count != 0);
4001 fault_type = entry->protection;
4002 size = entry->end - entry->start;
4004 * If the entry was copy-on-write, we either ...
4006 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
4008 * If we want to write the page, we may as well handle that
4009 * now since we've got the map locked.
4011 * If we don't need to write the page, we just demote the
4012 * permissions allowed.
4014 if ((fault_type & VM_PROT_WRITE) != 0 ||
4015 (fault_typea & VM_PROT_COPY) != 0) {
4017 * Make a new object, and place it in the object
4018 * chain. Note that no new references have appeared
4019 * -- one just moved from the map to the new
4022 if (vm_map_lock_upgrade(map))
4025 if (entry->cred == NULL) {
4027 * The debugger owner is charged for
4030 cred = curthread->td_ucred;
4032 if (!swap_reserve_by_cred(size, cred)) {
4035 return (KERN_RESOURCE_SHORTAGE);
4039 vm_object_shadow(&entry->object.vm_object,
4040 &entry->offset, size);
4041 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
4042 eobject = entry->object.vm_object;
4043 if (eobject->cred != NULL) {
4045 * The object was not shadowed.
4047 swap_release_by_cred(size, entry->cred);
4048 crfree(entry->cred);
4050 } else if (entry->cred != NULL) {
4051 VM_OBJECT_WLOCK(eobject);
4052 eobject->cred = entry->cred;
4053 eobject->charge = size;
4054 VM_OBJECT_WUNLOCK(eobject);
4058 vm_map_lock_downgrade(map);
4061 * We're attempting to read a copy-on-write page --
4062 * don't allow writes.
4064 prot &= ~VM_PROT_WRITE;
4069 * Create an object if necessary.
4071 if (entry->object.vm_object == NULL &&
4073 if (vm_map_lock_upgrade(map))
4075 entry->object.vm_object = vm_object_allocate(OBJT_DEFAULT,
4078 if (entry->cred != NULL) {
4079 VM_OBJECT_WLOCK(entry->object.vm_object);
4080 entry->object.vm_object->cred = entry->cred;
4081 entry->object.vm_object->charge = size;
4082 VM_OBJECT_WUNLOCK(entry->object.vm_object);
4085 vm_map_lock_downgrade(map);
4089 * Return the object/offset from this entry. If the entry was
4090 * copy-on-write or empty, it has been fixed up.
4092 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
4093 *object = entry->object.vm_object;
4096 return (KERN_SUCCESS);
4100 * vm_map_lookup_locked:
4102 * Lookup the faulting address. A version of vm_map_lookup that returns
4103 * KERN_FAILURE instead of blocking on map lock or memory allocation.
4106 vm_map_lookup_locked(vm_map_t *var_map, /* IN/OUT */
4108 vm_prot_t fault_typea,
4109 vm_map_entry_t *out_entry, /* OUT */
4110 vm_object_t *object, /* OUT */
4111 vm_pindex_t *pindex, /* OUT */
4112 vm_prot_t *out_prot, /* OUT */
4113 boolean_t *wired) /* OUT */
4115 vm_map_entry_t entry;
4116 vm_map_t map = *var_map;
4118 vm_prot_t fault_type = fault_typea;
4121 * Lookup the faulting address.
4123 if (!vm_map_lookup_entry(map, vaddr, out_entry))
4124 return (KERN_INVALID_ADDRESS);
4129 * Fail if the entry refers to a submap.
4131 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP)
4132 return (KERN_FAILURE);
4135 * Check whether this task is allowed to have this page.
4137 prot = entry->protection;
4138 fault_type &= VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE;
4139 if ((fault_type & prot) != fault_type)
4140 return (KERN_PROTECTION_FAILURE);
4141 if ((entry->eflags & MAP_ENTRY_USER_WIRED) &&
4142 (entry->eflags & MAP_ENTRY_COW) &&
4143 (fault_type & VM_PROT_WRITE))
4144 return (KERN_PROTECTION_FAILURE);
4147 * If this page is not pageable, we have to get it for all possible
4150 *wired = (entry->wired_count != 0);
4152 fault_type = entry->protection;
4154 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
4156 * Fail if the entry was copy-on-write for a write fault.
4158 if (fault_type & VM_PROT_WRITE)
4159 return (KERN_FAILURE);
4161 * We're attempting to read a copy-on-write page --
4162 * don't allow writes.
4164 prot &= ~VM_PROT_WRITE;
4168 * Fail if an object should be created.
4170 if (entry->object.vm_object == NULL && !map->system_map)
4171 return (KERN_FAILURE);
4174 * Return the object/offset from this entry. If the entry was
4175 * copy-on-write or empty, it has been fixed up.
4177 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
4178 *object = entry->object.vm_object;
4181 return (KERN_SUCCESS);
4185 * vm_map_lookup_done:
4187 * Releases locks acquired by a vm_map_lookup
4188 * (according to the handle returned by that lookup).
4191 vm_map_lookup_done(vm_map_t map, vm_map_entry_t entry)
4194 * Unlock the main-level map
4196 vm_map_unlock_read(map);
4199 #include "opt_ddb.h"
4201 #include <sys/kernel.h>
4203 #include <ddb/ddb.h>
4206 vm_map_print(vm_map_t map)
4208 vm_map_entry_t entry;
4210 db_iprintf("Task map %p: pmap=%p, nentries=%d, version=%u\n",
4212 (void *)map->pmap, map->nentries, map->timestamp);
4215 for (entry = map->header.next; entry != &map->header;
4216 entry = entry->next) {
4217 db_iprintf("map entry %p: start=%p, end=%p\n",
4218 (void *)entry, (void *)entry->start, (void *)entry->end);
4220 static char *inheritance_name[4] =
4221 {"share", "copy", "none", "donate_copy"};
4223 db_iprintf(" prot=%x/%x/%s",
4225 entry->max_protection,
4226 inheritance_name[(int)(unsigned char)entry->inheritance]);
4227 if (entry->wired_count != 0)
4228 db_printf(", wired");
4230 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
4231 db_printf(", share=%p, offset=0x%jx\n",
4232 (void *)entry->object.sub_map,
4233 (uintmax_t)entry->offset);
4234 if ((entry->prev == &map->header) ||
4235 (entry->prev->object.sub_map !=
4236 entry->object.sub_map)) {
4238 vm_map_print((vm_map_t)entry->object.sub_map);
4242 if (entry->cred != NULL)
4243 db_printf(", ruid %d", entry->cred->cr_ruid);
4244 db_printf(", object=%p, offset=0x%jx",
4245 (void *)entry->object.vm_object,
4246 (uintmax_t)entry->offset);
4247 if (entry->object.vm_object && entry->object.vm_object->cred)
4248 db_printf(", obj ruid %d charge %jx",
4249 entry->object.vm_object->cred->cr_ruid,
4250 (uintmax_t)entry->object.vm_object->charge);
4251 if (entry->eflags & MAP_ENTRY_COW)
4252 db_printf(", copy (%s)",
4253 (entry->eflags & MAP_ENTRY_NEEDS_COPY) ? "needed" : "done");
4256 if ((entry->prev == &map->header) ||
4257 (entry->prev->object.vm_object !=
4258 entry->object.vm_object)) {
4260 vm_object_print((db_expr_t)(intptr_t)
4261 entry->object.vm_object,
4270 DB_SHOW_COMMAND(map, map)
4274 db_printf("usage: show map <addr>\n");
4277 vm_map_print((vm_map_t)addr);
4280 DB_SHOW_COMMAND(procvm, procvm)
4285 p = (struct proc *) addr;
4290 db_printf("p = %p, vmspace = %p, map = %p, pmap = %p\n",
4291 (void *)p, (void *)p->p_vmspace, (void *)&p->p_vmspace->vm_map,
4292 (void *)vmspace_pmap(p->p_vmspace));
4294 vm_map_print((vm_map_t)&p->p_vmspace->vm_map);