2 * Copyright (c) 1991, 1993
3 * The Regents of the University of California. All rights reserved.
5 * This code is derived from software contributed to Berkeley by
6 * The Mach Operating System project at Carnegie-Mellon University.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 4. Neither the name of the University nor the names of its contributors
17 * may be used to endorse or promote products derived from this software
18 * without specific prior written permission.
20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * from: @(#)vm_map.c 8.3 (Berkeley) 1/12/94
35 * Copyright (c) 1987, 1990 Carnegie-Mellon University.
36 * All rights reserved.
38 * Authors: Avadis Tevanian, Jr., Michael Wayne Young
40 * Permission to use, copy, modify and distribute this software and
41 * its documentation is hereby granted, provided that both the copyright
42 * notice and this permission notice appear in all copies of the
43 * software, derivative works or modified versions, and any portions
44 * thereof, and that both notices appear in supporting documentation.
46 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
47 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
48 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
50 * Carnegie Mellon requests users of this software to return to
52 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
53 * School of Computer Science
54 * Carnegie Mellon University
55 * Pittsburgh PA 15213-3890
57 * any improvements or extensions that they make and grant Carnegie the
58 * rights to redistribute these changes.
62 * Virtual memory mapping module.
65 #include <sys/cdefs.h>
66 __FBSDID("$FreeBSD$");
68 #include <sys/param.h>
69 #include <sys/systm.h>
70 #include <sys/kernel.h>
73 #include <sys/mutex.h>
75 #include <sys/vmmeter.h>
77 #include <sys/vnode.h>
78 #include <sys/racct.h>
79 #include <sys/resourcevar.h>
80 #include <sys/rwlock.h>
82 #include <sys/sysctl.h>
83 #include <sys/sysent.h>
87 #include <vm/vm_param.h>
89 #include <vm/vm_map.h>
90 #include <vm/vm_page.h>
91 #include <vm/vm_object.h>
92 #include <vm/vm_pager.h>
93 #include <vm/vm_kern.h>
94 #include <vm/vm_extern.h>
95 #include <vm/vnode_pager.h>
96 #include <vm/swap_pager.h>
100 * Virtual memory maps provide for the mapping, protection,
101 * and sharing of virtual memory objects. In addition,
102 * this module provides for an efficient virtual copy of
103 * memory from one map to another.
105 * Synchronization is required prior to most operations.
107 * Maps consist of an ordered doubly-linked list of simple
108 * entries; a self-adjusting binary search tree of these
109 * entries is used to speed up lookups.
111 * Since portions of maps are specified by start/end addresses,
112 * which may not align with existing map entries, all
113 * routines merely "clip" entries to these start/end values.
114 * [That is, an entry is split into two, bordering at a
115 * start or end value.] Note that these clippings may not
116 * always be necessary (as the two resulting entries are then
117 * not changed); however, the clipping is done for convenience.
119 * As mentioned above, virtual copy operations are performed
120 * by copying VM object references from one map to
121 * another, and then marking both regions as copy-on-write.
124 static struct mtx map_sleep_mtx;
125 static uma_zone_t mapentzone;
126 static uma_zone_t kmapentzone;
127 static uma_zone_t mapzone;
128 static uma_zone_t vmspace_zone;
129 static int vmspace_zinit(void *mem, int size, int flags);
130 static int vm_map_zinit(void *mem, int ize, int flags);
131 static void _vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min,
133 static void vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map);
134 static void vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry);
135 static void vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry);
136 static void vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot,
137 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags);
139 static void vm_map_zdtor(void *mem, int size, void *arg);
140 static void vmspace_zdtor(void *mem, int size, void *arg);
142 static int vm_map_stack_locked(vm_map_t map, vm_offset_t addrbos,
143 vm_size_t max_ssize, vm_size_t growsize, vm_prot_t prot, vm_prot_t max,
145 static void vm_map_wire_entry_failure(vm_map_t map, vm_map_entry_t entry,
146 vm_offset_t failed_addr);
148 #define ENTRY_CHARGED(e) ((e)->cred != NULL || \
149 ((e)->object.vm_object != NULL && (e)->object.vm_object->cred != NULL && \
150 !((e)->eflags & MAP_ENTRY_NEEDS_COPY)))
153 * PROC_VMSPACE_{UN,}LOCK() can be a noop as long as vmspaces are type
156 #define PROC_VMSPACE_LOCK(p) do { } while (0)
157 #define PROC_VMSPACE_UNLOCK(p) do { } while (0)
160 * VM_MAP_RANGE_CHECK: [ internal use only ]
162 * Asserts that the starting and ending region
163 * addresses fall within the valid range of the map.
165 #define VM_MAP_RANGE_CHECK(map, start, end) \
167 if (start < vm_map_min(map)) \
168 start = vm_map_min(map); \
169 if (end > vm_map_max(map)) \
170 end = vm_map_max(map); \
178 * Initialize the vm_map module. Must be called before
179 * any other vm_map routines.
181 * Map and entry structures are allocated from the general
182 * purpose memory pool with some exceptions:
184 * - The kernel map and kmem submap are allocated statically.
185 * - Kernel map entries are allocated out of a static pool.
187 * These restrictions are necessary since malloc() uses the
188 * maps and requires map entries.
194 mtx_init(&map_sleep_mtx, "vm map sleep mutex", NULL, MTX_DEF);
195 mapzone = uma_zcreate("MAP", sizeof(struct vm_map), NULL,
201 vm_map_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
202 uma_prealloc(mapzone, MAX_KMAP);
203 kmapentzone = uma_zcreate("KMAP ENTRY", sizeof(struct vm_map_entry),
204 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR,
205 UMA_ZONE_MTXCLASS | UMA_ZONE_VM);
206 mapentzone = uma_zcreate("MAP ENTRY", sizeof(struct vm_map_entry),
207 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
208 vmspace_zone = uma_zcreate("VMSPACE", sizeof(struct vmspace), NULL,
214 vmspace_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
218 vmspace_zinit(void *mem, int size, int flags)
222 vm = (struct vmspace *)mem;
224 vm->vm_map.pmap = NULL;
225 (void)vm_map_zinit(&vm->vm_map, sizeof(vm->vm_map), flags);
226 PMAP_LOCK_INIT(vmspace_pmap(vm));
231 vm_map_zinit(void *mem, int size, int flags)
236 memset(map, 0, sizeof(*map));
237 mtx_init(&map->system_mtx, "vm map (system)", NULL, MTX_DEF | MTX_DUPOK);
238 sx_init(&map->lock, "vm map (user)");
244 vmspace_zdtor(void *mem, int size, void *arg)
248 vm = (struct vmspace *)mem;
250 vm_map_zdtor(&vm->vm_map, sizeof(vm->vm_map), arg);
253 vm_map_zdtor(void *mem, int size, void *arg)
258 KASSERT(map->nentries == 0,
259 ("map %p nentries == %d on free.",
260 map, map->nentries));
261 KASSERT(map->size == 0,
262 ("map %p size == %lu on free.",
263 map, (unsigned long)map->size));
265 #endif /* INVARIANTS */
268 * Allocate a vmspace structure, including a vm_map and pmap,
269 * and initialize those structures. The refcnt is set to 1.
271 * If 'pinit' is NULL then the embedded pmap is initialized via pmap_pinit().
274 vmspace_alloc(vm_offset_t min, vm_offset_t max, pmap_pinit_t pinit)
278 vm = uma_zalloc(vmspace_zone, M_WAITOK);
280 KASSERT(vm->vm_map.pmap == NULL, ("vm_map.pmap must be NULL"));
285 if (!pinit(vmspace_pmap(vm))) {
286 uma_zfree(vmspace_zone, vm);
289 CTR1(KTR_VM, "vmspace_alloc: %p", vm);
290 _vm_map_init(&vm->vm_map, vmspace_pmap(vm), min, max);
305 vmspace_container_reset(struct proc *p)
309 racct_set(p, RACCT_DATA, 0);
310 racct_set(p, RACCT_STACK, 0);
311 racct_set(p, RACCT_RSS, 0);
312 racct_set(p, RACCT_MEMLOCK, 0);
313 racct_set(p, RACCT_VMEM, 0);
319 vmspace_dofree(struct vmspace *vm)
322 CTR1(KTR_VM, "vmspace_free: %p", vm);
325 * Make sure any SysV shm is freed, it might not have been in
331 * Lock the map, to wait out all other references to it.
332 * Delete all of the mappings and pages they hold, then call
333 * the pmap module to reclaim anything left.
335 (void)vm_map_remove(&vm->vm_map, vm->vm_map.min_offset,
336 vm->vm_map.max_offset);
338 pmap_release(vmspace_pmap(vm));
339 vm->vm_map.pmap = NULL;
340 uma_zfree(vmspace_zone, vm);
344 vmspace_free(struct vmspace *vm)
347 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
348 "vmspace_free() called with non-sleepable lock held");
350 if (vm->vm_refcnt == 0)
351 panic("vmspace_free: attempt to free already freed vmspace");
353 if (atomic_fetchadd_int(&vm->vm_refcnt, -1) == 1)
358 vmspace_exitfree(struct proc *p)
362 PROC_VMSPACE_LOCK(p);
365 PROC_VMSPACE_UNLOCK(p);
366 KASSERT(vm == &vmspace0, ("vmspace_exitfree: wrong vmspace"));
371 vmspace_exit(struct thread *td)
378 * Release user portion of address space.
379 * This releases references to vnodes,
380 * which could cause I/O if the file has been unlinked.
381 * Need to do this early enough that we can still sleep.
383 * The last exiting process to reach this point releases as
384 * much of the environment as it can. vmspace_dofree() is the
385 * slower fallback in case another process had a temporary
386 * reference to the vmspace.
391 atomic_add_int(&vmspace0.vm_refcnt, 1);
393 refcnt = vm->vm_refcnt;
394 if (refcnt > 1 && p->p_vmspace != &vmspace0) {
395 /* Switch now since other proc might free vmspace */
396 PROC_VMSPACE_LOCK(p);
397 p->p_vmspace = &vmspace0;
398 PROC_VMSPACE_UNLOCK(p);
401 } while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt - 1));
403 if (p->p_vmspace != vm) {
404 /* vmspace not yet freed, switch back */
405 PROC_VMSPACE_LOCK(p);
407 PROC_VMSPACE_UNLOCK(p);
410 pmap_remove_pages(vmspace_pmap(vm));
411 /* Switch now since this proc will free vmspace */
412 PROC_VMSPACE_LOCK(p);
413 p->p_vmspace = &vmspace0;
414 PROC_VMSPACE_UNLOCK(p);
420 vmspace_container_reset(p);
424 /* Acquire reference to vmspace owned by another process. */
427 vmspace_acquire_ref(struct proc *p)
432 PROC_VMSPACE_LOCK(p);
435 PROC_VMSPACE_UNLOCK(p);
439 refcnt = vm->vm_refcnt;
440 if (refcnt <= 0) { /* Avoid 0->1 transition */
441 PROC_VMSPACE_UNLOCK(p);
444 } while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt + 1));
445 if (vm != p->p_vmspace) {
446 PROC_VMSPACE_UNLOCK(p);
450 PROC_VMSPACE_UNLOCK(p);
455 _vm_map_lock(vm_map_t map, const char *file, int line)
459 mtx_lock_flags_(&map->system_mtx, 0, file, line);
461 sx_xlock_(&map->lock, file, line);
466 vm_map_process_deferred(void)
469 vm_map_entry_t entry, next;
473 entry = td->td_map_def_user;
474 td->td_map_def_user = NULL;
475 while (entry != NULL) {
477 if ((entry->eflags & MAP_ENTRY_VN_WRITECNT) != 0) {
479 * Decrement the object's writemappings and
480 * possibly the vnode's v_writecount.
482 KASSERT((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0,
483 ("Submap with writecount"));
484 object = entry->object.vm_object;
485 KASSERT(object != NULL, ("No object for writecount"));
486 vnode_pager_release_writecount(object, entry->start,
489 vm_map_entry_deallocate(entry, FALSE);
495 _vm_map_unlock(vm_map_t map, const char *file, int line)
499 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
501 sx_xunlock_(&map->lock, file, line);
502 vm_map_process_deferred();
507 _vm_map_lock_read(vm_map_t map, const char *file, int line)
511 mtx_lock_flags_(&map->system_mtx, 0, file, line);
513 sx_slock_(&map->lock, file, line);
517 _vm_map_unlock_read(vm_map_t map, const char *file, int line)
521 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
523 sx_sunlock_(&map->lock, file, line);
524 vm_map_process_deferred();
529 _vm_map_trylock(vm_map_t map, const char *file, int line)
533 error = map->system_map ?
534 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) :
535 !sx_try_xlock_(&map->lock, file, line);
542 _vm_map_trylock_read(vm_map_t map, const char *file, int line)
546 error = map->system_map ?
547 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) :
548 !sx_try_slock_(&map->lock, file, line);
553 * _vm_map_lock_upgrade: [ internal use only ]
555 * Tries to upgrade a read (shared) lock on the specified map to a write
556 * (exclusive) lock. Returns the value "0" if the upgrade succeeds and a
557 * non-zero value if the upgrade fails. If the upgrade fails, the map is
558 * returned without a read or write lock held.
560 * Requires that the map be read locked.
563 _vm_map_lock_upgrade(vm_map_t map, const char *file, int line)
565 unsigned int last_timestamp;
567 if (map->system_map) {
568 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
570 if (!sx_try_upgrade_(&map->lock, file, line)) {
571 last_timestamp = map->timestamp;
572 sx_sunlock_(&map->lock, file, line);
573 vm_map_process_deferred();
575 * If the map's timestamp does not change while the
576 * map is unlocked, then the upgrade succeeds.
578 sx_xlock_(&map->lock, file, line);
579 if (last_timestamp != map->timestamp) {
580 sx_xunlock_(&map->lock, file, line);
590 _vm_map_lock_downgrade(vm_map_t map, const char *file, int line)
593 if (map->system_map) {
594 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
596 sx_downgrade_(&map->lock, file, line);
602 * Returns a non-zero value if the caller holds a write (exclusive) lock
603 * on the specified map and the value "0" otherwise.
606 vm_map_locked(vm_map_t map)
610 return (mtx_owned(&map->system_mtx));
612 return (sx_xlocked(&map->lock));
617 _vm_map_assert_locked(vm_map_t map, const char *file, int line)
621 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
623 sx_assert_(&map->lock, SA_XLOCKED, file, line);
626 #define VM_MAP_ASSERT_LOCKED(map) \
627 _vm_map_assert_locked(map, LOCK_FILE, LOCK_LINE)
629 #define VM_MAP_ASSERT_LOCKED(map)
633 * _vm_map_unlock_and_wait:
635 * Atomically releases the lock on the specified map and puts the calling
636 * thread to sleep. The calling thread will remain asleep until either
637 * vm_map_wakeup() is performed on the map or the specified timeout is
640 * WARNING! This function does not perform deferred deallocations of
641 * objects and map entries. Therefore, the calling thread is expected to
642 * reacquire the map lock after reawakening and later perform an ordinary
643 * unlock operation, such as vm_map_unlock(), before completing its
644 * operation on the map.
647 _vm_map_unlock_and_wait(vm_map_t map, int timo, const char *file, int line)
650 mtx_lock(&map_sleep_mtx);
652 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
654 sx_xunlock_(&map->lock, file, line);
655 return (msleep(&map->root, &map_sleep_mtx, PDROP | PVM, "vmmaps",
662 * Awaken any threads that have slept on the map using
663 * vm_map_unlock_and_wait().
666 vm_map_wakeup(vm_map_t map)
670 * Acquire and release map_sleep_mtx to prevent a wakeup()
671 * from being performed (and lost) between the map unlock
672 * and the msleep() in _vm_map_unlock_and_wait().
674 mtx_lock(&map_sleep_mtx);
675 mtx_unlock(&map_sleep_mtx);
680 vm_map_busy(vm_map_t map)
683 VM_MAP_ASSERT_LOCKED(map);
688 vm_map_unbusy(vm_map_t map)
691 VM_MAP_ASSERT_LOCKED(map);
692 KASSERT(map->busy, ("vm_map_unbusy: not busy"));
693 if (--map->busy == 0 && (map->flags & MAP_BUSY_WAKEUP)) {
694 vm_map_modflags(map, 0, MAP_BUSY_WAKEUP);
700 vm_map_wait_busy(vm_map_t map)
703 VM_MAP_ASSERT_LOCKED(map);
705 vm_map_modflags(map, MAP_BUSY_WAKEUP, 0);
707 msleep(&map->busy, &map->system_mtx, 0, "mbusy", 0);
709 sx_sleep(&map->busy, &map->lock, 0, "mbusy", 0);
715 vmspace_resident_count(struct vmspace *vmspace)
717 return pmap_resident_count(vmspace_pmap(vmspace));
723 * Creates and returns a new empty VM map with
724 * the given physical map structure, and having
725 * the given lower and upper address bounds.
728 vm_map_create(pmap_t pmap, vm_offset_t min, vm_offset_t max)
732 result = uma_zalloc(mapzone, M_WAITOK);
733 CTR1(KTR_VM, "vm_map_create: %p", result);
734 _vm_map_init(result, pmap, min, max);
739 * Initialize an existing vm_map structure
740 * such as that in the vmspace structure.
743 _vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max)
746 map->header.next = map->header.prev = &map->header;
747 map->needs_wakeup = FALSE;
750 map->min_offset = min;
751 map->max_offset = max;
759 vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max)
762 _vm_map_init(map, pmap, min, max);
763 mtx_init(&map->system_mtx, "system map", NULL, MTX_DEF | MTX_DUPOK);
764 sx_init(&map->lock, "user map");
768 * vm_map_entry_dispose: [ internal use only ]
770 * Inverse of vm_map_entry_create.
773 vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry)
775 uma_zfree(map->system_map ? kmapentzone : mapentzone, entry);
779 * vm_map_entry_create: [ internal use only ]
781 * Allocates a VM map entry for insertion.
782 * No entry fields are filled in.
784 static vm_map_entry_t
785 vm_map_entry_create(vm_map_t map)
787 vm_map_entry_t new_entry;
790 new_entry = uma_zalloc(kmapentzone, M_NOWAIT);
792 new_entry = uma_zalloc(mapentzone, M_WAITOK);
793 if (new_entry == NULL)
794 panic("vm_map_entry_create: kernel resources exhausted");
799 * vm_map_entry_set_behavior:
801 * Set the expected access behavior, either normal, random, or
805 vm_map_entry_set_behavior(vm_map_entry_t entry, u_char behavior)
807 entry->eflags = (entry->eflags & ~MAP_ENTRY_BEHAV_MASK) |
808 (behavior & MAP_ENTRY_BEHAV_MASK);
812 * vm_map_entry_set_max_free:
814 * Set the max_free field in a vm_map_entry.
817 vm_map_entry_set_max_free(vm_map_entry_t entry)
820 entry->max_free = entry->adj_free;
821 if (entry->left != NULL && entry->left->max_free > entry->max_free)
822 entry->max_free = entry->left->max_free;
823 if (entry->right != NULL && entry->right->max_free > entry->max_free)
824 entry->max_free = entry->right->max_free;
828 * vm_map_entry_splay:
830 * The Sleator and Tarjan top-down splay algorithm with the
831 * following variation. Max_free must be computed bottom-up, so
832 * on the downward pass, maintain the left and right spines in
833 * reverse order. Then, make a second pass up each side to fix
834 * the pointers and compute max_free. The time bound is O(log n)
837 * The new root is the vm_map_entry containing "addr", or else an
838 * adjacent entry (lower or higher) if addr is not in the tree.
840 * The map must be locked, and leaves it so.
842 * Returns: the new root.
844 static vm_map_entry_t
845 vm_map_entry_splay(vm_offset_t addr, vm_map_entry_t root)
847 vm_map_entry_t llist, rlist;
848 vm_map_entry_t ltree, rtree;
851 /* Special case of empty tree. */
856 * Pass One: Splay down the tree until we find addr or a NULL
857 * pointer where addr would go. llist and rlist are the two
858 * sides in reverse order (bottom-up), with llist linked by
859 * the right pointer and rlist linked by the left pointer in
860 * the vm_map_entry. Wait until Pass Two to set max_free on
866 /* root is never NULL in here. */
867 if (addr < root->start) {
871 if (addr < y->start && y->left != NULL) {
872 /* Rotate right and put y on rlist. */
873 root->left = y->right;
875 vm_map_entry_set_max_free(root);
880 /* Put root on rlist. */
885 } else if (addr >= root->end) {
889 if (addr >= y->end && y->right != NULL) {
890 /* Rotate left and put y on llist. */
891 root->right = y->left;
893 vm_map_entry_set_max_free(root);
898 /* Put root on llist. */
908 * Pass Two: Walk back up the two spines, flip the pointers
909 * and set max_free. The subtrees of the root go at the
910 * bottom of llist and rlist.
913 while (llist != NULL) {
915 llist->right = ltree;
916 vm_map_entry_set_max_free(llist);
921 while (rlist != NULL) {
924 vm_map_entry_set_max_free(rlist);
930 * Final assembly: add ltree and rtree as subtrees of root.
934 vm_map_entry_set_max_free(root);
940 * vm_map_entry_{un,}link:
942 * Insert/remove entries from maps.
945 vm_map_entry_link(vm_map_t map,
946 vm_map_entry_t after_where,
947 vm_map_entry_t entry)
951 "vm_map_entry_link: map %p, nentries %d, entry %p, after %p", map,
952 map->nentries, entry, after_where);
953 VM_MAP_ASSERT_LOCKED(map);
954 KASSERT(after_where == &map->header ||
955 after_where->end <= entry->start,
956 ("vm_map_entry_link: prev end %jx new start %jx overlap",
957 (uintmax_t)after_where->end, (uintmax_t)entry->start));
958 KASSERT(after_where->next == &map->header ||
959 entry->end <= after_where->next->start,
960 ("vm_map_entry_link: new end %jx next start %jx overlap",
961 (uintmax_t)entry->end, (uintmax_t)after_where->next->start));
964 entry->prev = after_where;
965 entry->next = after_where->next;
966 entry->next->prev = entry;
967 after_where->next = entry;
969 if (after_where != &map->header) {
970 if (after_where != map->root)
971 vm_map_entry_splay(after_where->start, map->root);
972 entry->right = after_where->right;
973 entry->left = after_where;
974 after_where->right = NULL;
975 after_where->adj_free = entry->start - after_where->end;
976 vm_map_entry_set_max_free(after_where);
978 entry->right = map->root;
981 entry->adj_free = (entry->next == &map->header ? map->max_offset :
982 entry->next->start) - entry->end;
983 vm_map_entry_set_max_free(entry);
988 vm_map_entry_unlink(vm_map_t map,
989 vm_map_entry_t entry)
991 vm_map_entry_t next, prev, root;
993 VM_MAP_ASSERT_LOCKED(map);
994 if (entry != map->root)
995 vm_map_entry_splay(entry->start, map->root);
996 if (entry->left == NULL)
999 root = vm_map_entry_splay(entry->start, entry->left);
1000 root->right = entry->right;
1001 root->adj_free = (entry->next == &map->header ? map->max_offset :
1002 entry->next->start) - root->end;
1003 vm_map_entry_set_max_free(root);
1012 CTR3(KTR_VM, "vm_map_entry_unlink: map %p, nentries %d, entry %p", map,
1013 map->nentries, entry);
1017 * vm_map_entry_resize_free:
1019 * Recompute the amount of free space following a vm_map_entry
1020 * and propagate that value up the tree. Call this function after
1021 * resizing a map entry in-place, that is, without a call to
1022 * vm_map_entry_link() or _unlink().
1024 * The map must be locked, and leaves it so.
1027 vm_map_entry_resize_free(vm_map_t map, vm_map_entry_t entry)
1031 * Using splay trees without parent pointers, propagating
1032 * max_free up the tree is done by moving the entry to the
1033 * root and making the change there.
1035 if (entry != map->root)
1036 map->root = vm_map_entry_splay(entry->start, map->root);
1038 entry->adj_free = (entry->next == &map->header ? map->max_offset :
1039 entry->next->start) - entry->end;
1040 vm_map_entry_set_max_free(entry);
1044 * vm_map_lookup_entry: [ internal use only ]
1046 * Finds the map entry containing (or
1047 * immediately preceding) the specified address
1048 * in the given map; the entry is returned
1049 * in the "entry" parameter. The boolean
1050 * result indicates whether the address is
1051 * actually contained in the map.
1054 vm_map_lookup_entry(
1056 vm_offset_t address,
1057 vm_map_entry_t *entry) /* OUT */
1063 * If the map is empty, then the map entry immediately preceding
1064 * "address" is the map's header.
1068 *entry = &map->header;
1069 else if (address >= cur->start && cur->end > address) {
1072 } else if ((locked = vm_map_locked(map)) ||
1073 sx_try_upgrade(&map->lock)) {
1075 * Splay requires a write lock on the map. However, it only
1076 * restructures the binary search tree; it does not otherwise
1077 * change the map. Thus, the map's timestamp need not change
1078 * on a temporary upgrade.
1080 map->root = cur = vm_map_entry_splay(address, cur);
1082 sx_downgrade(&map->lock);
1085 * If "address" is contained within a map entry, the new root
1086 * is that map entry. Otherwise, the new root is a map entry
1087 * immediately before or after "address".
1089 if (address >= cur->start) {
1091 if (cur->end > address)
1097 * Since the map is only locked for read access, perform a
1098 * standard binary search tree lookup for "address".
1101 if (address < cur->start) {
1102 if (cur->left == NULL) {
1107 } else if (cur->end > address) {
1111 if (cur->right == NULL) {
1124 * Inserts the given whole VM object into the target
1125 * map at the specified address range. The object's
1126 * size should match that of the address range.
1128 * Requires that the map be locked, and leaves it so.
1130 * If object is non-NULL, ref count must be bumped by caller
1131 * prior to making call to account for the new entry.
1134 vm_map_insert(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1135 vm_offset_t start, vm_offset_t end, vm_prot_t prot, vm_prot_t max, int cow)
1137 vm_map_entry_t new_entry, prev_entry, temp_entry;
1138 vm_eflags_t protoeflags;
1140 vm_inherit_t inheritance;
1142 VM_MAP_ASSERT_LOCKED(map);
1143 KASSERT((object != kmem_object && object != kernel_object) ||
1144 (cow & MAP_COPY_ON_WRITE) == 0,
1145 ("vm_map_insert: kmem or kernel object and COW"));
1146 KASSERT(object == NULL || (cow & MAP_NOFAULT) == 0,
1147 ("vm_map_insert: paradoxical MAP_NOFAULT request"));
1150 * Check that the start and end points are not bogus.
1152 if ((start < map->min_offset) || (end > map->max_offset) ||
1154 return (KERN_INVALID_ADDRESS);
1157 * Find the entry prior to the proposed starting address; if it's part
1158 * of an existing entry, this range is bogus.
1160 if (vm_map_lookup_entry(map, start, &temp_entry))
1161 return (KERN_NO_SPACE);
1163 prev_entry = temp_entry;
1166 * Assert that the next entry doesn't overlap the end point.
1168 if ((prev_entry->next != &map->header) &&
1169 (prev_entry->next->start < end))
1170 return (KERN_NO_SPACE);
1173 if (cow & MAP_COPY_ON_WRITE)
1174 protoeflags |= MAP_ENTRY_COW | MAP_ENTRY_NEEDS_COPY;
1175 if (cow & MAP_NOFAULT)
1176 protoeflags |= MAP_ENTRY_NOFAULT;
1177 if (cow & MAP_DISABLE_SYNCER)
1178 protoeflags |= MAP_ENTRY_NOSYNC;
1179 if (cow & MAP_DISABLE_COREDUMP)
1180 protoeflags |= MAP_ENTRY_NOCOREDUMP;
1181 if (cow & MAP_STACK_GROWS_DOWN)
1182 protoeflags |= MAP_ENTRY_GROWS_DOWN;
1183 if (cow & MAP_STACK_GROWS_UP)
1184 protoeflags |= MAP_ENTRY_GROWS_UP;
1185 if (cow & MAP_VN_WRITECOUNT)
1186 protoeflags |= MAP_ENTRY_VN_WRITECNT;
1187 if (cow & MAP_INHERIT_SHARE)
1188 inheritance = VM_INHERIT_SHARE;
1190 inheritance = VM_INHERIT_DEFAULT;
1193 if (cow & (MAP_ACC_NO_CHARGE | MAP_NOFAULT))
1195 if ((cow & MAP_ACC_CHARGED) || ((prot & VM_PROT_WRITE) &&
1196 ((protoeflags & MAP_ENTRY_NEEDS_COPY) || object == NULL))) {
1197 if (!(cow & MAP_ACC_CHARGED) && !swap_reserve(end - start))
1198 return (KERN_RESOURCE_SHORTAGE);
1199 KASSERT(object == NULL || (protoeflags & MAP_ENTRY_NEEDS_COPY) ||
1200 object->cred == NULL,
1201 ("OVERCOMMIT: vm_map_insert o %p", object));
1202 cred = curthread->td_ucred;
1206 /* Expand the kernel pmap, if necessary. */
1207 if (map == kernel_map && end > kernel_vm_end)
1208 pmap_growkernel(end);
1209 if (object != NULL) {
1211 * OBJ_ONEMAPPING must be cleared unless this mapping
1212 * is trivially proven to be the only mapping for any
1213 * of the object's pages. (Object granularity
1214 * reference counting is insufficient to recognize
1215 * aliases with precision.)
1217 VM_OBJECT_WLOCK(object);
1218 if (object->ref_count > 1 || object->shadow_count != 0)
1219 vm_object_clear_flag(object, OBJ_ONEMAPPING);
1220 VM_OBJECT_WUNLOCK(object);
1222 else if ((prev_entry != &map->header) &&
1223 (prev_entry->eflags == protoeflags) &&
1224 (cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 &&
1225 (prev_entry->end == start) &&
1226 (prev_entry->wired_count == 0) &&
1227 (prev_entry->cred == cred ||
1228 (prev_entry->object.vm_object != NULL &&
1229 (prev_entry->object.vm_object->cred == cred))) &&
1230 vm_object_coalesce(prev_entry->object.vm_object,
1232 (vm_size_t)(prev_entry->end - prev_entry->start),
1233 (vm_size_t)(end - prev_entry->end), cred != NULL &&
1234 (protoeflags & MAP_ENTRY_NEEDS_COPY) == 0)) {
1236 * We were able to extend the object. Determine if we
1237 * can extend the previous map entry to include the
1238 * new range as well.
1240 if ((prev_entry->inheritance == inheritance) &&
1241 (prev_entry->protection == prot) &&
1242 (prev_entry->max_protection == max)) {
1243 map->size += (end - prev_entry->end);
1244 prev_entry->end = end;
1245 vm_map_entry_resize_free(map, prev_entry);
1246 vm_map_simplify_entry(map, prev_entry);
1247 return (KERN_SUCCESS);
1251 * If we can extend the object but cannot extend the
1252 * map entry, we have to create a new map entry. We
1253 * must bump the ref count on the extended object to
1254 * account for it. object may be NULL.
1256 object = prev_entry->object.vm_object;
1257 offset = prev_entry->offset +
1258 (prev_entry->end - prev_entry->start);
1259 vm_object_reference(object);
1260 if (cred != NULL && object != NULL && object->cred != NULL &&
1261 !(prev_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
1262 /* Object already accounts for this uid. */
1270 * Create a new entry
1272 new_entry = vm_map_entry_create(map);
1273 new_entry->start = start;
1274 new_entry->end = end;
1275 new_entry->cred = NULL;
1277 new_entry->eflags = protoeflags;
1278 new_entry->object.vm_object = object;
1279 new_entry->offset = offset;
1280 new_entry->avail_ssize = 0;
1282 new_entry->inheritance = inheritance;
1283 new_entry->protection = prot;
1284 new_entry->max_protection = max;
1285 new_entry->wired_count = 0;
1286 new_entry->wiring_thread = NULL;
1287 new_entry->read_ahead = VM_FAULT_READ_AHEAD_INIT;
1288 new_entry->next_read = OFF_TO_IDX(offset);
1290 KASSERT(cred == NULL || !ENTRY_CHARGED(new_entry),
1291 ("OVERCOMMIT: vm_map_insert leaks vm_map %p", new_entry));
1292 new_entry->cred = cred;
1295 * Insert the new entry into the list
1297 vm_map_entry_link(map, prev_entry, new_entry);
1298 map->size += new_entry->end - new_entry->start;
1301 * Try to coalesce the new entry with both the previous and next
1302 * entries in the list. Previously, we only attempted to coalesce
1303 * with the previous entry when object is NULL. Here, we handle the
1304 * other cases, which are less common.
1306 vm_map_simplify_entry(map, new_entry);
1308 if (cow & (MAP_PREFAULT|MAP_PREFAULT_PARTIAL)) {
1309 vm_map_pmap_enter(map, start, prot,
1310 object, OFF_TO_IDX(offset), end - start,
1311 cow & MAP_PREFAULT_PARTIAL);
1314 return (KERN_SUCCESS);
1320 * Find the first fit (lowest VM address) for "length" free bytes
1321 * beginning at address >= start in the given map.
1323 * In a vm_map_entry, "adj_free" is the amount of free space
1324 * adjacent (higher address) to this entry, and "max_free" is the
1325 * maximum amount of contiguous free space in its subtree. This
1326 * allows finding a free region in one path down the tree, so
1327 * O(log n) amortized with splay trees.
1329 * The map must be locked, and leaves it so.
1331 * Returns: 0 on success, and starting address in *addr,
1332 * 1 if insufficient space.
1335 vm_map_findspace(vm_map_t map, vm_offset_t start, vm_size_t length,
1336 vm_offset_t *addr) /* OUT */
1338 vm_map_entry_t entry;
1342 * Request must fit within min/max VM address and must avoid
1345 if (start < map->min_offset)
1346 start = map->min_offset;
1347 if (start + length > map->max_offset || start + length < start)
1350 /* Empty tree means wide open address space. */
1351 if (map->root == NULL) {
1357 * After splay, if start comes before root node, then there
1358 * must be a gap from start to the root.
1360 map->root = vm_map_entry_splay(start, map->root);
1361 if (start + length <= map->root->start) {
1367 * Root is the last node that might begin its gap before
1368 * start, and this is the last comparison where address
1369 * wrap might be a problem.
1371 st = (start > map->root->end) ? start : map->root->end;
1372 if (length <= map->root->end + map->root->adj_free - st) {
1377 /* With max_free, can immediately tell if no solution. */
1378 entry = map->root->right;
1379 if (entry == NULL || length > entry->max_free)
1383 * Search the right subtree in the order: left subtree, root,
1384 * right subtree (first fit). The previous splay implies that
1385 * all regions in the right subtree have addresses > start.
1387 while (entry != NULL) {
1388 if (entry->left != NULL && entry->left->max_free >= length)
1389 entry = entry->left;
1390 else if (entry->adj_free >= length) {
1394 entry = entry->right;
1397 /* Can't get here, so panic if we do. */
1398 panic("vm_map_findspace: max_free corrupt");
1402 vm_map_fixed(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1403 vm_offset_t start, vm_size_t length, vm_prot_t prot,
1404 vm_prot_t max, int cow)
1409 end = start + length;
1410 KASSERT((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 ||
1412 ("vm_map_fixed: non-NULL backing object for stack"));
1414 VM_MAP_RANGE_CHECK(map, start, end);
1415 if ((cow & MAP_CHECK_EXCL) == 0)
1416 vm_map_delete(map, start, end);
1417 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) != 0) {
1418 result = vm_map_stack_locked(map, start, length, sgrowsiz,
1421 result = vm_map_insert(map, object, offset, start, end,
1429 * vm_map_find finds an unallocated region in the target address
1430 * map with the given length. The search is defined to be
1431 * first-fit from the specified address; the region found is
1432 * returned in the same parameter.
1434 * If object is non-NULL, ref count must be bumped by caller
1435 * prior to making call to account for the new entry.
1438 vm_map_find(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1439 vm_offset_t *addr, /* IN/OUT */
1440 vm_size_t length, vm_offset_t max_addr, int find_space,
1441 vm_prot_t prot, vm_prot_t max, int cow)
1443 vm_offset_t alignment, initial_addr, start;
1446 KASSERT((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 ||
1448 ("vm_map_find: non-NULL backing object for stack"));
1449 if (find_space == VMFS_OPTIMAL_SPACE && (object == NULL ||
1450 (object->flags & OBJ_COLORED) == 0))
1451 find_space = VMFS_ANY_SPACE;
1452 if (find_space >> 8 != 0) {
1453 KASSERT((find_space & 0xff) == 0, ("bad VMFS flags"));
1454 alignment = (vm_offset_t)1 << (find_space >> 8);
1457 initial_addr = *addr;
1459 start = initial_addr;
1462 if (find_space != VMFS_NO_SPACE) {
1463 if (vm_map_findspace(map, start, length, addr) ||
1464 (max_addr != 0 && *addr + length > max_addr)) {
1466 if (find_space == VMFS_OPTIMAL_SPACE) {
1467 find_space = VMFS_ANY_SPACE;
1470 return (KERN_NO_SPACE);
1472 switch (find_space) {
1473 case VMFS_SUPER_SPACE:
1474 case VMFS_OPTIMAL_SPACE:
1475 pmap_align_superpage(object, offset, addr,
1478 case VMFS_ANY_SPACE:
1481 if ((*addr & (alignment - 1)) != 0) {
1482 *addr &= ~(alignment - 1);
1490 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) != 0) {
1491 result = vm_map_stack_locked(map, start, length,
1492 sgrowsiz, prot, max, cow);
1494 result = vm_map_insert(map, object, offset, start,
1495 start + length, prot, max, cow);
1497 } while (result == KERN_NO_SPACE && find_space != VMFS_NO_SPACE &&
1498 find_space != VMFS_ANY_SPACE);
1504 * vm_map_simplify_entry:
1506 * Simplify the given map entry by merging with either neighbor. This
1507 * routine also has the ability to merge with both neighbors.
1509 * The map must be locked.
1511 * This routine guarentees that the passed entry remains valid (though
1512 * possibly extended). When merging, this routine may delete one or
1516 vm_map_simplify_entry(vm_map_t map, vm_map_entry_t entry)
1518 vm_map_entry_t next, prev;
1519 vm_size_t prevsize, esize;
1521 if ((entry->eflags & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP |
1522 MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_IS_SUB_MAP)) != 0)
1526 if (prev != &map->header) {
1527 prevsize = prev->end - prev->start;
1528 if ( (prev->end == entry->start) &&
1529 (prev->object.vm_object == entry->object.vm_object) &&
1530 (!prev->object.vm_object ||
1531 (prev->offset + prevsize == entry->offset)) &&
1532 (prev->eflags == entry->eflags) &&
1533 (prev->protection == entry->protection) &&
1534 (prev->max_protection == entry->max_protection) &&
1535 (prev->inheritance == entry->inheritance) &&
1536 (prev->wired_count == entry->wired_count) &&
1537 (prev->cred == entry->cred)) {
1538 vm_map_entry_unlink(map, prev);
1539 entry->start = prev->start;
1540 entry->offset = prev->offset;
1541 if (entry->prev != &map->header)
1542 vm_map_entry_resize_free(map, entry->prev);
1545 * If the backing object is a vnode object,
1546 * vm_object_deallocate() calls vrele().
1547 * However, vrele() does not lock the vnode
1548 * because the vnode has additional
1549 * references. Thus, the map lock can be kept
1550 * without causing a lock-order reversal with
1553 * Since we count the number of virtual page
1554 * mappings in object->un_pager.vnp.writemappings,
1555 * the writemappings value should not be adjusted
1556 * when the entry is disposed of.
1558 if (prev->object.vm_object)
1559 vm_object_deallocate(prev->object.vm_object);
1560 if (prev->cred != NULL)
1562 vm_map_entry_dispose(map, prev);
1567 if (next != &map->header) {
1568 esize = entry->end - entry->start;
1569 if ((entry->end == next->start) &&
1570 (next->object.vm_object == entry->object.vm_object) &&
1571 (!entry->object.vm_object ||
1572 (entry->offset + esize == next->offset)) &&
1573 (next->eflags == entry->eflags) &&
1574 (next->protection == entry->protection) &&
1575 (next->max_protection == entry->max_protection) &&
1576 (next->inheritance == entry->inheritance) &&
1577 (next->wired_count == entry->wired_count) &&
1578 (next->cred == entry->cred)) {
1579 vm_map_entry_unlink(map, next);
1580 entry->end = next->end;
1581 vm_map_entry_resize_free(map, entry);
1584 * See comment above.
1586 if (next->object.vm_object)
1587 vm_object_deallocate(next->object.vm_object);
1588 if (next->cred != NULL)
1590 vm_map_entry_dispose(map, next);
1595 * vm_map_clip_start: [ internal use only ]
1597 * Asserts that the given entry begins at or after
1598 * the specified address; if necessary,
1599 * it splits the entry into two.
1601 #define vm_map_clip_start(map, entry, startaddr) \
1603 if (startaddr > entry->start) \
1604 _vm_map_clip_start(map, entry, startaddr); \
1608 * This routine is called only when it is known that
1609 * the entry must be split.
1612 _vm_map_clip_start(vm_map_t map, vm_map_entry_t entry, vm_offset_t start)
1614 vm_map_entry_t new_entry;
1616 VM_MAP_ASSERT_LOCKED(map);
1619 * Split off the front portion -- note that we must insert the new
1620 * entry BEFORE this one, so that this entry has the specified
1623 vm_map_simplify_entry(map, entry);
1626 * If there is no object backing this entry, we might as well create
1627 * one now. If we defer it, an object can get created after the map
1628 * is clipped, and individual objects will be created for the split-up
1629 * map. This is a bit of a hack, but is also about the best place to
1630 * put this improvement.
1632 if (entry->object.vm_object == NULL && !map->system_map) {
1634 object = vm_object_allocate(OBJT_DEFAULT,
1635 atop(entry->end - entry->start));
1636 entry->object.vm_object = object;
1638 if (entry->cred != NULL) {
1639 object->cred = entry->cred;
1640 object->charge = entry->end - entry->start;
1643 } else if (entry->object.vm_object != NULL &&
1644 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) &&
1645 entry->cred != NULL) {
1646 VM_OBJECT_WLOCK(entry->object.vm_object);
1647 KASSERT(entry->object.vm_object->cred == NULL,
1648 ("OVERCOMMIT: vm_entry_clip_start: both cred e %p", entry));
1649 entry->object.vm_object->cred = entry->cred;
1650 entry->object.vm_object->charge = entry->end - entry->start;
1651 VM_OBJECT_WUNLOCK(entry->object.vm_object);
1655 new_entry = vm_map_entry_create(map);
1656 *new_entry = *entry;
1658 new_entry->end = start;
1659 entry->offset += (start - entry->start);
1660 entry->start = start;
1661 if (new_entry->cred != NULL)
1662 crhold(entry->cred);
1664 vm_map_entry_link(map, entry->prev, new_entry);
1666 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1667 vm_object_reference(new_entry->object.vm_object);
1669 * The object->un_pager.vnp.writemappings for the
1670 * object of MAP_ENTRY_VN_WRITECNT type entry shall be
1671 * kept as is here. The virtual pages are
1672 * re-distributed among the clipped entries, so the sum is
1679 * vm_map_clip_end: [ internal use only ]
1681 * Asserts that the given entry ends at or before
1682 * the specified address; if necessary,
1683 * it splits the entry into two.
1685 #define vm_map_clip_end(map, entry, endaddr) \
1687 if ((endaddr) < (entry->end)) \
1688 _vm_map_clip_end((map), (entry), (endaddr)); \
1692 * This routine is called only when it is known that
1693 * the entry must be split.
1696 _vm_map_clip_end(vm_map_t map, vm_map_entry_t entry, vm_offset_t end)
1698 vm_map_entry_t new_entry;
1700 VM_MAP_ASSERT_LOCKED(map);
1703 * If there is no object backing this entry, we might as well create
1704 * one now. If we defer it, an object can get created after the map
1705 * is clipped, and individual objects will be created for the split-up
1706 * map. This is a bit of a hack, but is also about the best place to
1707 * put this improvement.
1709 if (entry->object.vm_object == NULL && !map->system_map) {
1711 object = vm_object_allocate(OBJT_DEFAULT,
1712 atop(entry->end - entry->start));
1713 entry->object.vm_object = object;
1715 if (entry->cred != NULL) {
1716 object->cred = entry->cred;
1717 object->charge = entry->end - entry->start;
1720 } else if (entry->object.vm_object != NULL &&
1721 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) &&
1722 entry->cred != NULL) {
1723 VM_OBJECT_WLOCK(entry->object.vm_object);
1724 KASSERT(entry->object.vm_object->cred == NULL,
1725 ("OVERCOMMIT: vm_entry_clip_end: both cred e %p", entry));
1726 entry->object.vm_object->cred = entry->cred;
1727 entry->object.vm_object->charge = entry->end - entry->start;
1728 VM_OBJECT_WUNLOCK(entry->object.vm_object);
1733 * Create a new entry and insert it AFTER the specified entry
1735 new_entry = vm_map_entry_create(map);
1736 *new_entry = *entry;
1738 new_entry->start = entry->end = end;
1739 new_entry->offset += (end - entry->start);
1740 if (new_entry->cred != NULL)
1741 crhold(entry->cred);
1743 vm_map_entry_link(map, entry, new_entry);
1745 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1746 vm_object_reference(new_entry->object.vm_object);
1751 * vm_map_submap: [ kernel use only ]
1753 * Mark the given range as handled by a subordinate map.
1755 * This range must have been created with vm_map_find,
1756 * and no other operations may have been performed on this
1757 * range prior to calling vm_map_submap.
1759 * Only a limited number of operations can be performed
1760 * within this rage after calling vm_map_submap:
1762 * [Don't try vm_map_copy!]
1764 * To remove a submapping, one must first remove the
1765 * range from the superior map, and then destroy the
1766 * submap (if desired). [Better yet, don't try it.]
1775 vm_map_entry_t entry;
1776 int result = KERN_INVALID_ARGUMENT;
1780 VM_MAP_RANGE_CHECK(map, start, end);
1782 if (vm_map_lookup_entry(map, start, &entry)) {
1783 vm_map_clip_start(map, entry, start);
1785 entry = entry->next;
1787 vm_map_clip_end(map, entry, end);
1789 if ((entry->start == start) && (entry->end == end) &&
1790 ((entry->eflags & MAP_ENTRY_COW) == 0) &&
1791 (entry->object.vm_object == NULL)) {
1792 entry->object.sub_map = submap;
1793 entry->eflags |= MAP_ENTRY_IS_SUB_MAP;
1794 result = KERN_SUCCESS;
1802 * The maximum number of pages to map if MAP_PREFAULT_PARTIAL is specified
1804 #define MAX_INIT_PT 96
1807 * vm_map_pmap_enter:
1809 * Preload the specified map's pmap with mappings to the specified
1810 * object's memory-resident pages. No further physical pages are
1811 * allocated, and no further virtual pages are retrieved from secondary
1812 * storage. If the specified flags include MAP_PREFAULT_PARTIAL, then a
1813 * limited number of page mappings are created at the low-end of the
1814 * specified address range. (For this purpose, a superpage mapping
1815 * counts as one page mapping.) Otherwise, all resident pages within
1816 * the specified address range are mapped. Because these mappings are
1817 * being created speculatively, cached pages are not reactivated and
1821 vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot,
1822 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags)
1825 vm_page_t p, p_start;
1826 vm_pindex_t mask, psize, threshold, tmpidx;
1828 if ((prot & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0 || object == NULL)
1830 VM_OBJECT_RLOCK(object);
1831 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) {
1832 VM_OBJECT_RUNLOCK(object);
1833 VM_OBJECT_WLOCK(object);
1834 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) {
1835 pmap_object_init_pt(map->pmap, addr, object, pindex,
1837 VM_OBJECT_WUNLOCK(object);
1840 VM_OBJECT_LOCK_DOWNGRADE(object);
1844 if (psize + pindex > object->size) {
1845 if (object->size < pindex) {
1846 VM_OBJECT_RUNLOCK(object);
1849 psize = object->size - pindex;
1854 threshold = MAX_INIT_PT;
1856 p = vm_page_find_least(object, pindex);
1858 * Assert: the variable p is either (1) the page with the
1859 * least pindex greater than or equal to the parameter pindex
1863 p != NULL && (tmpidx = p->pindex - pindex) < psize;
1864 p = TAILQ_NEXT(p, listq)) {
1866 * don't allow an madvise to blow away our really
1867 * free pages allocating pv entries.
1869 if (((flags & MAP_PREFAULT_MADVISE) != 0 &&
1870 vm_cnt.v_free_count < vm_cnt.v_free_reserved) ||
1871 ((flags & MAP_PREFAULT_PARTIAL) != 0 &&
1872 tmpidx >= threshold)) {
1876 if (p->valid == VM_PAGE_BITS_ALL) {
1877 if (p_start == NULL) {
1878 start = addr + ptoa(tmpidx);
1881 /* Jump ahead if a superpage mapping is possible. */
1882 if (p->psind > 0 && ((addr + ptoa(tmpidx)) &
1883 (pagesizes[p->psind] - 1)) == 0) {
1884 mask = atop(pagesizes[p->psind]) - 1;
1885 if (tmpidx + mask < psize &&
1886 vm_page_ps_is_valid(p)) {
1891 } else if (p_start != NULL) {
1892 pmap_enter_object(map->pmap, start, addr +
1893 ptoa(tmpidx), p_start, prot);
1897 if (p_start != NULL)
1898 pmap_enter_object(map->pmap, start, addr + ptoa(psize),
1900 VM_OBJECT_RUNLOCK(object);
1906 * Sets the protection of the specified address
1907 * region in the target map. If "set_max" is
1908 * specified, the maximum protection is to be set;
1909 * otherwise, only the current protection is affected.
1912 vm_map_protect(vm_map_t map, vm_offset_t start, vm_offset_t end,
1913 vm_prot_t new_prot, boolean_t set_max)
1915 vm_map_entry_t current, entry;
1921 return (KERN_SUCCESS);
1925 VM_MAP_RANGE_CHECK(map, start, end);
1927 if (vm_map_lookup_entry(map, start, &entry)) {
1928 vm_map_clip_start(map, entry, start);
1930 entry = entry->next;
1934 * Make a first pass to check for protection violations.
1937 while ((current != &map->header) && (current->start < end)) {
1938 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
1940 return (KERN_INVALID_ARGUMENT);
1942 if ((new_prot & current->max_protection) != new_prot) {
1944 return (KERN_PROTECTION_FAILURE);
1946 current = current->next;
1951 * Do an accounting pass for private read-only mappings that
1952 * now will do cow due to allowed write (e.g. debugger sets
1953 * breakpoint on text segment)
1955 for (current = entry; (current != &map->header) &&
1956 (current->start < end); current = current->next) {
1958 vm_map_clip_end(map, current, end);
1961 ((new_prot & ~(current->protection)) & VM_PROT_WRITE) == 0 ||
1962 ENTRY_CHARGED(current)) {
1966 cred = curthread->td_ucred;
1967 obj = current->object.vm_object;
1969 if (obj == NULL || (current->eflags & MAP_ENTRY_NEEDS_COPY)) {
1970 if (!swap_reserve(current->end - current->start)) {
1972 return (KERN_RESOURCE_SHORTAGE);
1975 current->cred = cred;
1979 VM_OBJECT_WLOCK(obj);
1980 if (obj->type != OBJT_DEFAULT && obj->type != OBJT_SWAP) {
1981 VM_OBJECT_WUNLOCK(obj);
1986 * Charge for the whole object allocation now, since
1987 * we cannot distinguish between non-charged and
1988 * charged clipped mapping of the same object later.
1990 KASSERT(obj->charge == 0,
1991 ("vm_map_protect: object %p overcharged (entry %p)",
1993 if (!swap_reserve(ptoa(obj->size))) {
1994 VM_OBJECT_WUNLOCK(obj);
1996 return (KERN_RESOURCE_SHORTAGE);
2001 obj->charge = ptoa(obj->size);
2002 VM_OBJECT_WUNLOCK(obj);
2006 * Go back and fix up protections. [Note that clipping is not
2007 * necessary the second time.]
2010 while ((current != &map->header) && (current->start < end)) {
2011 old_prot = current->protection;
2014 current->protection =
2015 (current->max_protection = new_prot) &
2018 current->protection = new_prot;
2021 * For user wired map entries, the normal lazy evaluation of
2022 * write access upgrades through soft page faults is
2023 * undesirable. Instead, immediately copy any pages that are
2024 * copy-on-write and enable write access in the physical map.
2026 if ((current->eflags & MAP_ENTRY_USER_WIRED) != 0 &&
2027 (current->protection & VM_PROT_WRITE) != 0 &&
2028 (old_prot & VM_PROT_WRITE) == 0)
2029 vm_fault_copy_entry(map, map, current, current, NULL);
2032 * When restricting access, update the physical map. Worry
2033 * about copy-on-write here.
2035 if ((old_prot & ~current->protection) != 0) {
2036 #define MASK(entry) (((entry)->eflags & MAP_ENTRY_COW) ? ~VM_PROT_WRITE : \
2038 pmap_protect(map->pmap, current->start,
2040 current->protection & MASK(current));
2043 vm_map_simplify_entry(map, current);
2044 current = current->next;
2047 return (KERN_SUCCESS);
2053 * This routine traverses a processes map handling the madvise
2054 * system call. Advisories are classified as either those effecting
2055 * the vm_map_entry structure, or those effecting the underlying
2065 vm_map_entry_t current, entry;
2069 * Some madvise calls directly modify the vm_map_entry, in which case
2070 * we need to use an exclusive lock on the map and we need to perform
2071 * various clipping operations. Otherwise we only need a read-lock
2076 case MADV_SEQUENTIAL:
2083 return (KERN_SUCCESS);
2091 return (KERN_SUCCESS);
2092 vm_map_lock_read(map);
2095 return (KERN_INVALID_ARGUMENT);
2099 * Locate starting entry and clip if necessary.
2101 VM_MAP_RANGE_CHECK(map, start, end);
2103 if (vm_map_lookup_entry(map, start, &entry)) {
2105 vm_map_clip_start(map, entry, start);
2107 entry = entry->next;
2112 * madvise behaviors that are implemented in the vm_map_entry.
2114 * We clip the vm_map_entry so that behavioral changes are
2115 * limited to the specified address range.
2117 for (current = entry;
2118 (current != &map->header) && (current->start < end);
2119 current = current->next
2121 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
2124 vm_map_clip_end(map, current, end);
2128 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_NORMAL);
2130 case MADV_SEQUENTIAL:
2131 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_SEQUENTIAL);
2134 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_RANDOM);
2137 current->eflags |= MAP_ENTRY_NOSYNC;
2140 current->eflags &= ~MAP_ENTRY_NOSYNC;
2143 current->eflags |= MAP_ENTRY_NOCOREDUMP;
2146 current->eflags &= ~MAP_ENTRY_NOCOREDUMP;
2151 vm_map_simplify_entry(map, current);
2155 vm_pindex_t pstart, pend;
2158 * madvise behaviors that are implemented in the underlying
2161 * Since we don't clip the vm_map_entry, we have to clip
2162 * the vm_object pindex and count.
2164 for (current = entry;
2165 (current != &map->header) && (current->start < end);
2166 current = current->next
2168 vm_offset_t useEnd, useStart;
2170 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
2173 pstart = OFF_TO_IDX(current->offset);
2174 pend = pstart + atop(current->end - current->start);
2175 useStart = current->start;
2176 useEnd = current->end;
2178 if (current->start < start) {
2179 pstart += atop(start - current->start);
2182 if (current->end > end) {
2183 pend -= atop(current->end - end);
2191 * Perform the pmap_advise() before clearing
2192 * PGA_REFERENCED in vm_page_advise(). Otherwise, a
2193 * concurrent pmap operation, such as pmap_remove(),
2194 * could clear a reference in the pmap and set
2195 * PGA_REFERENCED on the page before the pmap_advise()
2196 * had completed. Consequently, the page would appear
2197 * referenced based upon an old reference that
2198 * occurred before this pmap_advise() ran.
2200 if (behav == MADV_DONTNEED || behav == MADV_FREE)
2201 pmap_advise(map->pmap, useStart, useEnd,
2204 vm_object_madvise(current->object.vm_object, pstart,
2208 * Pre-populate paging structures in the
2209 * WILLNEED case. For wired entries, the
2210 * paging structures are already populated.
2212 if (behav == MADV_WILLNEED &&
2213 current->wired_count == 0) {
2214 vm_map_pmap_enter(map,
2216 current->protection,
2217 current->object.vm_object,
2219 ptoa(pend - pstart),
2220 MAP_PREFAULT_MADVISE
2224 vm_map_unlock_read(map);
2233 * Sets the inheritance of the specified address
2234 * range in the target map. Inheritance
2235 * affects how the map will be shared with
2236 * child maps at the time of vmspace_fork.
2239 vm_map_inherit(vm_map_t map, vm_offset_t start, vm_offset_t end,
2240 vm_inherit_t new_inheritance)
2242 vm_map_entry_t entry;
2243 vm_map_entry_t temp_entry;
2245 switch (new_inheritance) {
2246 case VM_INHERIT_NONE:
2247 case VM_INHERIT_COPY:
2248 case VM_INHERIT_SHARE:
2251 return (KERN_INVALID_ARGUMENT);
2254 return (KERN_SUCCESS);
2256 VM_MAP_RANGE_CHECK(map, start, end);
2257 if (vm_map_lookup_entry(map, start, &temp_entry)) {
2259 vm_map_clip_start(map, entry, start);
2261 entry = temp_entry->next;
2262 while ((entry != &map->header) && (entry->start < end)) {
2263 vm_map_clip_end(map, entry, end);
2264 entry->inheritance = new_inheritance;
2265 vm_map_simplify_entry(map, entry);
2266 entry = entry->next;
2269 return (KERN_SUCCESS);
2275 * Implements both kernel and user unwiring.
2278 vm_map_unwire(vm_map_t map, vm_offset_t start, vm_offset_t end,
2281 vm_map_entry_t entry, first_entry, tmp_entry;
2282 vm_offset_t saved_start;
2283 unsigned int last_timestamp;
2285 boolean_t need_wakeup, result, user_unwire;
2288 return (KERN_SUCCESS);
2289 user_unwire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE;
2291 VM_MAP_RANGE_CHECK(map, start, end);
2292 if (!vm_map_lookup_entry(map, start, &first_entry)) {
2293 if (flags & VM_MAP_WIRE_HOLESOK)
2294 first_entry = first_entry->next;
2297 return (KERN_INVALID_ADDRESS);
2300 last_timestamp = map->timestamp;
2301 entry = first_entry;
2302 while (entry != &map->header && entry->start < end) {
2303 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
2305 * We have not yet clipped the entry.
2307 saved_start = (start >= entry->start) ? start :
2309 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2310 if (vm_map_unlock_and_wait(map, 0)) {
2312 * Allow interruption of user unwiring?
2316 if (last_timestamp+1 != map->timestamp) {
2318 * Look again for the entry because the map was
2319 * modified while it was unlocked.
2320 * Specifically, the entry may have been
2321 * clipped, merged, or deleted.
2323 if (!vm_map_lookup_entry(map, saved_start,
2325 if (flags & VM_MAP_WIRE_HOLESOK)
2326 tmp_entry = tmp_entry->next;
2328 if (saved_start == start) {
2330 * First_entry has been deleted.
2333 return (KERN_INVALID_ADDRESS);
2336 rv = KERN_INVALID_ADDRESS;
2340 if (entry == first_entry)
2341 first_entry = tmp_entry;
2346 last_timestamp = map->timestamp;
2349 vm_map_clip_start(map, entry, start);
2350 vm_map_clip_end(map, entry, end);
2352 * Mark the entry in case the map lock is released. (See
2355 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 &&
2356 entry->wiring_thread == NULL,
2357 ("owned map entry %p", entry));
2358 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
2359 entry->wiring_thread = curthread;
2361 * Check the map for holes in the specified region.
2362 * If VM_MAP_WIRE_HOLESOK was specified, skip this check.
2364 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) &&
2365 (entry->end < end && (entry->next == &map->header ||
2366 entry->next->start > entry->end))) {
2368 rv = KERN_INVALID_ADDRESS;
2372 * If system unwiring, require that the entry is system wired.
2375 vm_map_entry_system_wired_count(entry) == 0) {
2377 rv = KERN_INVALID_ARGUMENT;
2380 entry = entry->next;
2384 need_wakeup = FALSE;
2385 if (first_entry == NULL) {
2386 result = vm_map_lookup_entry(map, start, &first_entry);
2387 if (!result && (flags & VM_MAP_WIRE_HOLESOK))
2388 first_entry = first_entry->next;
2390 KASSERT(result, ("vm_map_unwire: lookup failed"));
2392 for (entry = first_entry; entry != &map->header && entry->start < end;
2393 entry = entry->next) {
2395 * If VM_MAP_WIRE_HOLESOK was specified, an empty
2396 * space in the unwired region could have been mapped
2397 * while the map lock was dropped for draining
2398 * MAP_ENTRY_IN_TRANSITION. Moreover, another thread
2399 * could be simultaneously wiring this new mapping
2400 * entry. Detect these cases and skip any entries
2401 * marked as in transition by us.
2403 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 ||
2404 entry->wiring_thread != curthread) {
2405 KASSERT((flags & VM_MAP_WIRE_HOLESOK) != 0,
2406 ("vm_map_unwire: !HOLESOK and new/changed entry"));
2410 if (rv == KERN_SUCCESS && (!user_unwire ||
2411 (entry->eflags & MAP_ENTRY_USER_WIRED))) {
2413 entry->eflags &= ~MAP_ENTRY_USER_WIRED;
2414 if (entry->wired_count == 1)
2415 vm_map_entry_unwire(map, entry);
2417 entry->wired_count--;
2419 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0,
2420 ("vm_map_unwire: in-transition flag missing %p", entry));
2421 KASSERT(entry->wiring_thread == curthread,
2422 ("vm_map_unwire: alien wire %p", entry));
2423 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION;
2424 entry->wiring_thread = NULL;
2425 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
2426 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
2429 vm_map_simplify_entry(map, entry);
2438 * vm_map_wire_entry_failure:
2440 * Handle a wiring failure on the given entry.
2442 * The map should be locked.
2445 vm_map_wire_entry_failure(vm_map_t map, vm_map_entry_t entry,
2446 vm_offset_t failed_addr)
2449 VM_MAP_ASSERT_LOCKED(map);
2450 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 &&
2451 entry->wired_count == 1,
2452 ("vm_map_wire_entry_failure: entry %p isn't being wired", entry));
2453 KASSERT(failed_addr < entry->end,
2454 ("vm_map_wire_entry_failure: entry %p was fully wired", entry));
2457 * If any pages at the start of this entry were successfully wired,
2460 if (failed_addr > entry->start) {
2461 pmap_unwire(map->pmap, entry->start, failed_addr);
2462 vm_object_unwire(entry->object.vm_object, entry->offset,
2463 failed_addr - entry->start, PQ_ACTIVE);
2467 * Assign an out-of-range value to represent the failure to wire this
2470 entry->wired_count = -1;
2476 * Implements both kernel and user wiring.
2479 vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t end,
2482 vm_map_entry_t entry, first_entry, tmp_entry;
2483 vm_offset_t faddr, saved_end, saved_start;
2484 unsigned int last_timestamp;
2486 boolean_t need_wakeup, result, user_wire;
2490 return (KERN_SUCCESS);
2492 if (flags & VM_MAP_WIRE_WRITE)
2493 prot |= VM_PROT_WRITE;
2494 user_wire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE;
2496 VM_MAP_RANGE_CHECK(map, start, end);
2497 if (!vm_map_lookup_entry(map, start, &first_entry)) {
2498 if (flags & VM_MAP_WIRE_HOLESOK)
2499 first_entry = first_entry->next;
2502 return (KERN_INVALID_ADDRESS);
2505 last_timestamp = map->timestamp;
2506 entry = first_entry;
2507 while (entry != &map->header && entry->start < end) {
2508 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
2510 * We have not yet clipped the entry.
2512 saved_start = (start >= entry->start) ? start :
2514 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2515 if (vm_map_unlock_and_wait(map, 0)) {
2517 * Allow interruption of user wiring?
2521 if (last_timestamp + 1 != map->timestamp) {
2523 * Look again for the entry because the map was
2524 * modified while it was unlocked.
2525 * Specifically, the entry may have been
2526 * clipped, merged, or deleted.
2528 if (!vm_map_lookup_entry(map, saved_start,
2530 if (flags & VM_MAP_WIRE_HOLESOK)
2531 tmp_entry = tmp_entry->next;
2533 if (saved_start == start) {
2535 * first_entry has been deleted.
2538 return (KERN_INVALID_ADDRESS);
2541 rv = KERN_INVALID_ADDRESS;
2545 if (entry == first_entry)
2546 first_entry = tmp_entry;
2551 last_timestamp = map->timestamp;
2554 vm_map_clip_start(map, entry, start);
2555 vm_map_clip_end(map, entry, end);
2557 * Mark the entry in case the map lock is released. (See
2560 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 &&
2561 entry->wiring_thread == NULL,
2562 ("owned map entry %p", entry));
2563 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
2564 entry->wiring_thread = curthread;
2565 if ((entry->protection & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0
2566 || (entry->protection & prot) != prot) {
2567 entry->eflags |= MAP_ENTRY_WIRE_SKIPPED;
2568 if ((flags & VM_MAP_WIRE_HOLESOK) == 0) {
2570 rv = KERN_INVALID_ADDRESS;
2575 if (entry->wired_count == 0) {
2576 entry->wired_count++;
2577 saved_start = entry->start;
2578 saved_end = entry->end;
2581 * Release the map lock, relying on the in-transition
2582 * mark. Mark the map busy for fork.
2587 faddr = saved_start;
2590 * Simulate a fault to get the page and enter
2591 * it into the physical map.
2593 if ((rv = vm_fault(map, faddr, VM_PROT_NONE,
2594 VM_FAULT_WIRE)) != KERN_SUCCESS)
2596 } while ((faddr += PAGE_SIZE) < saved_end);
2599 if (last_timestamp + 1 != map->timestamp) {
2601 * Look again for the entry because the map was
2602 * modified while it was unlocked. The entry
2603 * may have been clipped, but NOT merged or
2606 result = vm_map_lookup_entry(map, saved_start,
2608 KASSERT(result, ("vm_map_wire: lookup failed"));
2609 if (entry == first_entry)
2610 first_entry = tmp_entry;
2614 while (entry->end < saved_end) {
2616 * In case of failure, handle entries
2617 * that were not fully wired here;
2618 * fully wired entries are handled
2621 if (rv != KERN_SUCCESS &&
2623 vm_map_wire_entry_failure(map,
2625 entry = entry->next;
2628 last_timestamp = map->timestamp;
2629 if (rv != KERN_SUCCESS) {
2630 vm_map_wire_entry_failure(map, entry, faddr);
2634 } else if (!user_wire ||
2635 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
2636 entry->wired_count++;
2639 * Check the map for holes in the specified region.
2640 * If VM_MAP_WIRE_HOLESOK was specified, skip this check.
2643 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) &&
2644 (entry->end < end && (entry->next == &map->header ||
2645 entry->next->start > entry->end))) {
2647 rv = KERN_INVALID_ADDRESS;
2650 entry = entry->next;
2654 need_wakeup = FALSE;
2655 if (first_entry == NULL) {
2656 result = vm_map_lookup_entry(map, start, &first_entry);
2657 if (!result && (flags & VM_MAP_WIRE_HOLESOK))
2658 first_entry = first_entry->next;
2660 KASSERT(result, ("vm_map_wire: lookup failed"));
2662 for (entry = first_entry; entry != &map->header && entry->start < end;
2663 entry = entry->next) {
2664 if ((entry->eflags & MAP_ENTRY_WIRE_SKIPPED) != 0)
2665 goto next_entry_done;
2668 * If VM_MAP_WIRE_HOLESOK was specified, an empty
2669 * space in the unwired region could have been mapped
2670 * while the map lock was dropped for faulting in the
2671 * pages or draining MAP_ENTRY_IN_TRANSITION.
2672 * Moreover, another thread could be simultaneously
2673 * wiring this new mapping entry. Detect these cases
2674 * and skip any entries marked as in transition by us.
2676 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 ||
2677 entry->wiring_thread != curthread) {
2678 KASSERT((flags & VM_MAP_WIRE_HOLESOK) != 0,
2679 ("vm_map_wire: !HOLESOK and new/changed entry"));
2683 if (rv == KERN_SUCCESS) {
2685 entry->eflags |= MAP_ENTRY_USER_WIRED;
2686 } else if (entry->wired_count == -1) {
2688 * Wiring failed on this entry. Thus, unwiring is
2691 entry->wired_count = 0;
2692 } else if (!user_wire ||
2693 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
2695 * Undo the wiring. Wiring succeeded on this entry
2696 * but failed on a later entry.
2698 if (entry->wired_count == 1)
2699 vm_map_entry_unwire(map, entry);
2701 entry->wired_count--;
2704 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0,
2705 ("vm_map_wire: in-transition flag missing %p", entry));
2706 KASSERT(entry->wiring_thread == curthread,
2707 ("vm_map_wire: alien wire %p", entry));
2708 entry->eflags &= ~(MAP_ENTRY_IN_TRANSITION |
2709 MAP_ENTRY_WIRE_SKIPPED);
2710 entry->wiring_thread = NULL;
2711 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
2712 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
2715 vm_map_simplify_entry(map, entry);
2726 * Push any dirty cached pages in the address range to their pager.
2727 * If syncio is TRUE, dirty pages are written synchronously.
2728 * If invalidate is TRUE, any cached pages are freed as well.
2730 * If the size of the region from start to end is zero, we are
2731 * supposed to flush all modified pages within the region containing
2732 * start. Unfortunately, a region can be split or coalesced with
2733 * neighboring regions, making it difficult to determine what the
2734 * original region was. Therefore, we approximate this requirement by
2735 * flushing the current region containing start.
2737 * Returns an error if any part of the specified range is not mapped.
2745 boolean_t invalidate)
2747 vm_map_entry_t current;
2748 vm_map_entry_t entry;
2751 vm_ooffset_t offset;
2752 unsigned int last_timestamp;
2755 vm_map_lock_read(map);
2756 VM_MAP_RANGE_CHECK(map, start, end);
2757 if (!vm_map_lookup_entry(map, start, &entry)) {
2758 vm_map_unlock_read(map);
2759 return (KERN_INVALID_ADDRESS);
2760 } else if (start == end) {
2761 start = entry->start;
2765 * Make a first pass to check for user-wired memory and holes.
2767 for (current = entry; current != &map->header && current->start < end;
2768 current = current->next) {
2769 if (invalidate && (current->eflags & MAP_ENTRY_USER_WIRED)) {
2770 vm_map_unlock_read(map);
2771 return (KERN_INVALID_ARGUMENT);
2773 if (end > current->end &&
2774 (current->next == &map->header ||
2775 current->end != current->next->start)) {
2776 vm_map_unlock_read(map);
2777 return (KERN_INVALID_ADDRESS);
2782 pmap_remove(map->pmap, start, end);
2786 * Make a second pass, cleaning/uncaching pages from the indicated
2789 for (current = entry; current != &map->header && current->start < end;) {
2790 offset = current->offset + (start - current->start);
2791 size = (end <= current->end ? end : current->end) - start;
2792 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
2794 vm_map_entry_t tentry;
2797 smap = current->object.sub_map;
2798 vm_map_lock_read(smap);
2799 (void) vm_map_lookup_entry(smap, offset, &tentry);
2800 tsize = tentry->end - offset;
2803 object = tentry->object.vm_object;
2804 offset = tentry->offset + (offset - tentry->start);
2805 vm_map_unlock_read(smap);
2807 object = current->object.vm_object;
2809 vm_object_reference(object);
2810 last_timestamp = map->timestamp;
2811 vm_map_unlock_read(map);
2812 if (!vm_object_sync(object, offset, size, syncio, invalidate))
2815 vm_object_deallocate(object);
2816 vm_map_lock_read(map);
2817 if (last_timestamp == map->timestamp ||
2818 !vm_map_lookup_entry(map, start, ¤t))
2819 current = current->next;
2822 vm_map_unlock_read(map);
2823 return (failed ? KERN_FAILURE : KERN_SUCCESS);
2827 * vm_map_entry_unwire: [ internal use only ]
2829 * Make the region specified by this entry pageable.
2831 * The map in question should be locked.
2832 * [This is the reason for this routine's existence.]
2835 vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry)
2838 VM_MAP_ASSERT_LOCKED(map);
2839 KASSERT(entry->wired_count > 0,
2840 ("vm_map_entry_unwire: entry %p isn't wired", entry));
2841 pmap_unwire(map->pmap, entry->start, entry->end);
2842 vm_object_unwire(entry->object.vm_object, entry->offset, entry->end -
2843 entry->start, PQ_ACTIVE);
2844 entry->wired_count = 0;
2848 vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map)
2851 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0)
2852 vm_object_deallocate(entry->object.vm_object);
2853 uma_zfree(system_map ? kmapentzone : mapentzone, entry);
2857 * vm_map_entry_delete: [ internal use only ]
2859 * Deallocate the given entry from the target map.
2862 vm_map_entry_delete(vm_map_t map, vm_map_entry_t entry)
2865 vm_pindex_t offidxstart, offidxend, count, size1;
2868 vm_map_entry_unlink(map, entry);
2869 object = entry->object.vm_object;
2870 size = entry->end - entry->start;
2873 if (entry->cred != NULL) {
2874 swap_release_by_cred(size, entry->cred);
2875 crfree(entry->cred);
2878 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0 &&
2880 KASSERT(entry->cred == NULL || object->cred == NULL ||
2881 (entry->eflags & MAP_ENTRY_NEEDS_COPY),
2882 ("OVERCOMMIT vm_map_entry_delete: both cred %p", entry));
2883 count = OFF_TO_IDX(size);
2884 offidxstart = OFF_TO_IDX(entry->offset);
2885 offidxend = offidxstart + count;
2886 VM_OBJECT_WLOCK(object);
2887 if (object->ref_count != 1 &&
2888 ((object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING ||
2889 object == kernel_object || object == kmem_object)) {
2890 vm_object_collapse(object);
2893 * The option OBJPR_NOTMAPPED can be passed here
2894 * because vm_map_delete() already performed
2895 * pmap_remove() on the only mapping to this range
2898 vm_object_page_remove(object, offidxstart, offidxend,
2900 if (object->type == OBJT_SWAP)
2901 swap_pager_freespace(object, offidxstart, count);
2902 if (offidxend >= object->size &&
2903 offidxstart < object->size) {
2904 size1 = object->size;
2905 object->size = offidxstart;
2906 if (object->cred != NULL) {
2907 size1 -= object->size;
2908 KASSERT(object->charge >= ptoa(size1),
2909 ("vm_map_entry_delete: object->charge < 0"));
2910 swap_release_by_cred(ptoa(size1), object->cred);
2911 object->charge -= ptoa(size1);
2915 VM_OBJECT_WUNLOCK(object);
2917 entry->object.vm_object = NULL;
2918 if (map->system_map)
2919 vm_map_entry_deallocate(entry, TRUE);
2921 entry->next = curthread->td_map_def_user;
2922 curthread->td_map_def_user = entry;
2927 * vm_map_delete: [ internal use only ]
2929 * Deallocates the given address range from the target
2933 vm_map_delete(vm_map_t map, vm_offset_t start, vm_offset_t end)
2935 vm_map_entry_t entry;
2936 vm_map_entry_t first_entry;
2938 VM_MAP_ASSERT_LOCKED(map);
2940 return (KERN_SUCCESS);
2943 * Find the start of the region, and clip it
2945 if (!vm_map_lookup_entry(map, start, &first_entry))
2946 entry = first_entry->next;
2948 entry = first_entry;
2949 vm_map_clip_start(map, entry, start);
2953 * Step through all entries in this region
2955 while ((entry != &map->header) && (entry->start < end)) {
2956 vm_map_entry_t next;
2959 * Wait for wiring or unwiring of an entry to complete.
2960 * Also wait for any system wirings to disappear on
2963 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 ||
2964 (vm_map_pmap(map) != kernel_pmap &&
2965 vm_map_entry_system_wired_count(entry) != 0)) {
2966 unsigned int last_timestamp;
2967 vm_offset_t saved_start;
2968 vm_map_entry_t tmp_entry;
2970 saved_start = entry->start;
2971 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2972 last_timestamp = map->timestamp;
2973 (void) vm_map_unlock_and_wait(map, 0);
2975 if (last_timestamp + 1 != map->timestamp) {
2977 * Look again for the entry because the map was
2978 * modified while it was unlocked.
2979 * Specifically, the entry may have been
2980 * clipped, merged, or deleted.
2982 if (!vm_map_lookup_entry(map, saved_start,
2984 entry = tmp_entry->next;
2987 vm_map_clip_start(map, entry,
2993 vm_map_clip_end(map, entry, end);
2998 * Unwire before removing addresses from the pmap; otherwise,
2999 * unwiring will put the entries back in the pmap.
3001 if (entry->wired_count != 0) {
3002 vm_map_entry_unwire(map, entry);
3005 pmap_remove(map->pmap, entry->start, entry->end);
3008 * Delete the entry only after removing all pmap
3009 * entries pointing to its pages. (Otherwise, its
3010 * page frames may be reallocated, and any modify bits
3011 * will be set in the wrong object!)
3013 vm_map_entry_delete(map, entry);
3016 return (KERN_SUCCESS);
3022 * Remove the given address range from the target map.
3023 * This is the exported form of vm_map_delete.
3026 vm_map_remove(vm_map_t map, vm_offset_t start, vm_offset_t end)
3031 VM_MAP_RANGE_CHECK(map, start, end);
3032 result = vm_map_delete(map, start, end);
3038 * vm_map_check_protection:
3040 * Assert that the target map allows the specified privilege on the
3041 * entire address region given. The entire region must be allocated.
3043 * WARNING! This code does not and should not check whether the
3044 * contents of the region is accessible. For example a smaller file
3045 * might be mapped into a larger address space.
3047 * NOTE! This code is also called by munmap().
3049 * The map must be locked. A read lock is sufficient.
3052 vm_map_check_protection(vm_map_t map, vm_offset_t start, vm_offset_t end,
3053 vm_prot_t protection)
3055 vm_map_entry_t entry;
3056 vm_map_entry_t tmp_entry;
3058 if (!vm_map_lookup_entry(map, start, &tmp_entry))
3062 while (start < end) {
3063 if (entry == &map->header)
3068 if (start < entry->start)
3071 * Check protection associated with entry.
3073 if ((entry->protection & protection) != protection)
3075 /* go to next entry */
3077 entry = entry->next;
3083 * vm_map_copy_entry:
3085 * Copies the contents of the source entry to the destination
3086 * entry. The entries *must* be aligned properly.
3092 vm_map_entry_t src_entry,
3093 vm_map_entry_t dst_entry,
3094 vm_ooffset_t *fork_charge)
3096 vm_object_t src_object;
3097 vm_map_entry_t fake_entry;
3102 VM_MAP_ASSERT_LOCKED(dst_map);
3104 if ((dst_entry->eflags|src_entry->eflags) & MAP_ENTRY_IS_SUB_MAP)
3107 if (src_entry->wired_count == 0 ||
3108 (src_entry->protection & VM_PROT_WRITE) == 0) {
3110 * If the source entry is marked needs_copy, it is already
3113 if ((src_entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0 &&
3114 (src_entry->protection & VM_PROT_WRITE) != 0) {
3115 pmap_protect(src_map->pmap,
3118 src_entry->protection & ~VM_PROT_WRITE);
3122 * Make a copy of the object.
3124 size = src_entry->end - src_entry->start;
3125 if ((src_object = src_entry->object.vm_object) != NULL) {
3126 VM_OBJECT_WLOCK(src_object);
3127 charged = ENTRY_CHARGED(src_entry);
3128 if ((src_object->handle == NULL) &&
3129 (src_object->type == OBJT_DEFAULT ||
3130 src_object->type == OBJT_SWAP)) {
3131 vm_object_collapse(src_object);
3132 if ((src_object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING) {
3133 vm_object_split(src_entry);
3134 src_object = src_entry->object.vm_object;
3137 vm_object_reference_locked(src_object);
3138 vm_object_clear_flag(src_object, OBJ_ONEMAPPING);
3139 if (src_entry->cred != NULL &&
3140 !(src_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
3141 KASSERT(src_object->cred == NULL,
3142 ("OVERCOMMIT: vm_map_copy_entry: cred %p",
3144 src_object->cred = src_entry->cred;
3145 src_object->charge = size;
3147 VM_OBJECT_WUNLOCK(src_object);
3148 dst_entry->object.vm_object = src_object;
3150 cred = curthread->td_ucred;
3152 dst_entry->cred = cred;
3153 *fork_charge += size;
3154 if (!(src_entry->eflags &
3155 MAP_ENTRY_NEEDS_COPY)) {
3157 src_entry->cred = cred;
3158 *fork_charge += size;
3161 src_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY);
3162 dst_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY);
3163 dst_entry->offset = src_entry->offset;
3164 if (src_entry->eflags & MAP_ENTRY_VN_WRITECNT) {
3166 * MAP_ENTRY_VN_WRITECNT cannot
3167 * indicate write reference from
3168 * src_entry, since the entry is
3169 * marked as needs copy. Allocate a
3170 * fake entry that is used to
3171 * decrement object->un_pager.vnp.writecount
3172 * at the appropriate time. Attach
3173 * fake_entry to the deferred list.
3175 fake_entry = vm_map_entry_create(dst_map);
3176 fake_entry->eflags = MAP_ENTRY_VN_WRITECNT;
3177 src_entry->eflags &= ~MAP_ENTRY_VN_WRITECNT;
3178 vm_object_reference(src_object);
3179 fake_entry->object.vm_object = src_object;
3180 fake_entry->start = src_entry->start;
3181 fake_entry->end = src_entry->end;
3182 fake_entry->next = curthread->td_map_def_user;
3183 curthread->td_map_def_user = fake_entry;
3186 dst_entry->object.vm_object = NULL;
3187 dst_entry->offset = 0;
3188 if (src_entry->cred != NULL) {
3189 dst_entry->cred = curthread->td_ucred;
3190 crhold(dst_entry->cred);
3191 *fork_charge += size;
3195 pmap_copy(dst_map->pmap, src_map->pmap, dst_entry->start,
3196 dst_entry->end - dst_entry->start, src_entry->start);
3199 * We don't want to make writeable wired pages copy-on-write.
3200 * Immediately copy these pages into the new map by simulating
3201 * page faults. The new pages are pageable.
3203 vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry,
3209 * vmspace_map_entry_forked:
3210 * Update the newly-forked vmspace each time a map entry is inherited
3211 * or copied. The values for vm_dsize and vm_tsize are approximate
3212 * (and mostly-obsolete ideas in the face of mmap(2) et al.)
3215 vmspace_map_entry_forked(const struct vmspace *vm1, struct vmspace *vm2,
3216 vm_map_entry_t entry)
3218 vm_size_t entrysize;
3221 entrysize = entry->end - entry->start;
3222 vm2->vm_map.size += entrysize;
3223 if (entry->eflags & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP)) {
3224 vm2->vm_ssize += btoc(entrysize);
3225 } else if (entry->start >= (vm_offset_t)vm1->vm_daddr &&
3226 entry->start < (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)) {
3227 newend = MIN(entry->end,
3228 (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize));
3229 vm2->vm_dsize += btoc(newend - entry->start);
3230 } else if (entry->start >= (vm_offset_t)vm1->vm_taddr &&
3231 entry->start < (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)) {
3232 newend = MIN(entry->end,
3233 (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize));
3234 vm2->vm_tsize += btoc(newend - entry->start);
3240 * Create a new process vmspace structure and vm_map
3241 * based on those of an existing process. The new map
3242 * is based on the old map, according to the inheritance
3243 * values on the regions in that map.
3245 * XXX It might be worth coalescing the entries added to the new vmspace.
3247 * The source map must not be locked.
3250 vmspace_fork(struct vmspace *vm1, vm_ooffset_t *fork_charge)
3252 struct vmspace *vm2;
3253 vm_map_t new_map, old_map;
3254 vm_map_entry_t new_entry, old_entry;
3258 old_map = &vm1->vm_map;
3259 /* Copy immutable fields of vm1 to vm2. */
3260 vm2 = vmspace_alloc(old_map->min_offset, old_map->max_offset, NULL);
3263 vm2->vm_taddr = vm1->vm_taddr;
3264 vm2->vm_daddr = vm1->vm_daddr;
3265 vm2->vm_maxsaddr = vm1->vm_maxsaddr;
3266 vm_map_lock(old_map);
3268 vm_map_wait_busy(old_map);
3269 new_map = &vm2->vm_map;
3270 locked = vm_map_trylock(new_map); /* trylock to silence WITNESS */
3271 KASSERT(locked, ("vmspace_fork: lock failed"));
3273 old_entry = old_map->header.next;
3275 while (old_entry != &old_map->header) {
3276 if (old_entry->eflags & MAP_ENTRY_IS_SUB_MAP)
3277 panic("vm_map_fork: encountered a submap");
3279 switch (old_entry->inheritance) {
3280 case VM_INHERIT_NONE:
3283 case VM_INHERIT_SHARE:
3285 * Clone the entry, creating the shared object if necessary.
3287 object = old_entry->object.vm_object;
3288 if (object == NULL) {
3289 object = vm_object_allocate(OBJT_DEFAULT,
3290 atop(old_entry->end - old_entry->start));
3291 old_entry->object.vm_object = object;
3292 old_entry->offset = 0;
3293 if (old_entry->cred != NULL) {
3294 object->cred = old_entry->cred;
3295 object->charge = old_entry->end -
3297 old_entry->cred = NULL;
3302 * Add the reference before calling vm_object_shadow
3303 * to insure that a shadow object is created.
3305 vm_object_reference(object);
3306 if (old_entry->eflags & MAP_ENTRY_NEEDS_COPY) {
3307 vm_object_shadow(&old_entry->object.vm_object,
3309 old_entry->end - old_entry->start);
3310 old_entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
3311 /* Transfer the second reference too. */
3312 vm_object_reference(
3313 old_entry->object.vm_object);
3316 * As in vm_map_simplify_entry(), the
3317 * vnode lock will not be acquired in
3318 * this call to vm_object_deallocate().
3320 vm_object_deallocate(object);
3321 object = old_entry->object.vm_object;
3323 VM_OBJECT_WLOCK(object);
3324 vm_object_clear_flag(object, OBJ_ONEMAPPING);
3325 if (old_entry->cred != NULL) {
3326 KASSERT(object->cred == NULL, ("vmspace_fork both cred"));
3327 object->cred = old_entry->cred;
3328 object->charge = old_entry->end - old_entry->start;
3329 old_entry->cred = NULL;
3333 * Assert the correct state of the vnode
3334 * v_writecount while the object is locked, to
3335 * not relock it later for the assertion
3338 if (old_entry->eflags & MAP_ENTRY_VN_WRITECNT &&
3339 object->type == OBJT_VNODE) {
3340 KASSERT(((struct vnode *)object->handle)->
3342 ("vmspace_fork: v_writecount %p", object));
3343 KASSERT(object->un_pager.vnp.writemappings > 0,
3344 ("vmspace_fork: vnp.writecount %p",
3347 VM_OBJECT_WUNLOCK(object);
3350 * Clone the entry, referencing the shared object.
3352 new_entry = vm_map_entry_create(new_map);
3353 *new_entry = *old_entry;
3354 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED |
3355 MAP_ENTRY_IN_TRANSITION);
3356 new_entry->wiring_thread = NULL;
3357 new_entry->wired_count = 0;
3358 if (new_entry->eflags & MAP_ENTRY_VN_WRITECNT) {
3359 vnode_pager_update_writecount(object,
3360 new_entry->start, new_entry->end);
3364 * Insert the entry into the new map -- we know we're
3365 * inserting at the end of the new map.
3367 vm_map_entry_link(new_map, new_map->header.prev,
3369 vmspace_map_entry_forked(vm1, vm2, new_entry);
3372 * Update the physical map
3374 pmap_copy(new_map->pmap, old_map->pmap,
3376 (old_entry->end - old_entry->start),
3380 case VM_INHERIT_COPY:
3382 * Clone the entry and link into the map.
3384 new_entry = vm_map_entry_create(new_map);
3385 *new_entry = *old_entry;
3387 * Copied entry is COW over the old object.
3389 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED |
3390 MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_VN_WRITECNT);
3391 new_entry->wiring_thread = NULL;
3392 new_entry->wired_count = 0;
3393 new_entry->object.vm_object = NULL;
3394 new_entry->cred = NULL;
3395 vm_map_entry_link(new_map, new_map->header.prev,
3397 vmspace_map_entry_forked(vm1, vm2, new_entry);
3398 vm_map_copy_entry(old_map, new_map, old_entry,
3399 new_entry, fork_charge);
3402 old_entry = old_entry->next;
3405 * Use inlined vm_map_unlock() to postpone handling the deferred
3406 * map entries, which cannot be done until both old_map and
3407 * new_map locks are released.
3409 sx_xunlock(&old_map->lock);
3410 sx_xunlock(&new_map->lock);
3411 vm_map_process_deferred();
3417 vm_map_stack(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
3418 vm_prot_t prot, vm_prot_t max, int cow)
3420 vm_size_t growsize, init_ssize;
3421 rlim_t lmemlim, vmemlim;
3424 growsize = sgrowsiz;
3425 init_ssize = (max_ssize < growsize) ? max_ssize : growsize;
3427 lmemlim = lim_cur(curthread, RLIMIT_MEMLOCK);
3428 vmemlim = lim_cur(curthread, RLIMIT_VMEM);
3429 if (!old_mlock && map->flags & MAP_WIREFUTURE) {
3430 if (ptoa(pmap_wired_count(map->pmap)) + init_ssize > lmemlim) {
3435 /* If we would blow our VMEM resource limit, no go */
3436 if (map->size + init_ssize > vmemlim) {
3440 rv = vm_map_stack_locked(map, addrbos, max_ssize, growsize, prot,
3448 vm_map_stack_locked(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
3449 vm_size_t growsize, vm_prot_t prot, vm_prot_t max, int cow)
3451 vm_map_entry_t new_entry, prev_entry;
3452 vm_offset_t bot, top;
3453 vm_size_t init_ssize;
3457 * The stack orientation is piggybacked with the cow argument.
3458 * Extract it into orient and mask the cow argument so that we
3459 * don't pass it around further.
3460 * NOTE: We explicitly allow bi-directional stacks.
3462 orient = cow & (MAP_STACK_GROWS_DOWN|MAP_STACK_GROWS_UP);
3463 KASSERT(orient != 0, ("No stack grow direction"));
3465 if (addrbos < vm_map_min(map) ||
3466 addrbos > vm_map_max(map) ||
3467 addrbos + max_ssize < addrbos)
3468 return (KERN_NO_SPACE);
3470 init_ssize = (max_ssize < growsize) ? max_ssize : growsize;
3472 /* If addr is already mapped, no go */
3473 if (vm_map_lookup_entry(map, addrbos, &prev_entry))
3474 return (KERN_NO_SPACE);
3477 * If we can't accomodate max_ssize in the current mapping, no go.
3478 * However, we need to be aware that subsequent user mappings might
3479 * map into the space we have reserved for stack, and currently this
3480 * space is not protected.
3482 * Hopefully we will at least detect this condition when we try to
3485 if ((prev_entry->next != &map->header) &&
3486 (prev_entry->next->start < addrbos + max_ssize))
3487 return (KERN_NO_SPACE);
3490 * We initially map a stack of only init_ssize. We will grow as
3491 * needed later. Depending on the orientation of the stack (i.e.
3492 * the grow direction) we either map at the top of the range, the
3493 * bottom of the range or in the middle.
3495 * Note: we would normally expect prot and max to be VM_PROT_ALL,
3496 * and cow to be 0. Possibly we should eliminate these as input
3497 * parameters, and just pass these values here in the insert call.
3499 if (orient == MAP_STACK_GROWS_DOWN)
3500 bot = addrbos + max_ssize - init_ssize;
3501 else if (orient == MAP_STACK_GROWS_UP)
3504 bot = round_page(addrbos + max_ssize/2 - init_ssize/2);
3505 top = bot + init_ssize;
3506 rv = vm_map_insert(map, NULL, 0, bot, top, prot, max, cow);
3508 /* Now set the avail_ssize amount. */
3509 if (rv == KERN_SUCCESS) {
3510 new_entry = prev_entry->next;
3511 if (new_entry->end != top || new_entry->start != bot)
3512 panic("Bad entry start/end for new stack entry");
3514 new_entry->avail_ssize = max_ssize - init_ssize;
3515 KASSERT((orient & MAP_STACK_GROWS_DOWN) == 0 ||
3516 (new_entry->eflags & MAP_ENTRY_GROWS_DOWN) != 0,
3517 ("new entry lacks MAP_ENTRY_GROWS_DOWN"));
3518 KASSERT((orient & MAP_STACK_GROWS_UP) == 0 ||
3519 (new_entry->eflags & MAP_ENTRY_GROWS_UP) != 0,
3520 ("new entry lacks MAP_ENTRY_GROWS_UP"));
3526 static int stack_guard_page = 0;
3527 SYSCTL_INT(_security_bsd, OID_AUTO, stack_guard_page, CTLFLAG_RWTUN,
3528 &stack_guard_page, 0,
3529 "Insert stack guard page ahead of the growable segments.");
3531 /* Attempts to grow a vm stack entry. Returns KERN_SUCCESS if the
3532 * desired address is already mapped, or if we successfully grow
3533 * the stack. Also returns KERN_SUCCESS if addr is outside the
3534 * stack range (this is strange, but preserves compatibility with
3535 * the grow function in vm_machdep.c).
3538 vm_map_growstack(struct proc *p, vm_offset_t addr)
3540 vm_map_entry_t next_entry, prev_entry;
3541 vm_map_entry_t new_entry, stack_entry;
3542 struct vmspace *vm = p->p_vmspace;
3543 vm_map_t map = &vm->vm_map;
3546 size_t grow_amount, max_grow;
3547 rlim_t lmemlim, stacklim, vmemlim;
3548 int is_procstack, rv;
3557 lmemlim = lim_cur(curthread, RLIMIT_MEMLOCK);
3558 stacklim = lim_cur(curthread, RLIMIT_STACK);
3559 vmemlim = lim_cur(curthread, RLIMIT_VMEM);
3562 vm_map_lock_read(map);
3564 /* If addr is already in the entry range, no need to grow.*/
3565 if (vm_map_lookup_entry(map, addr, &prev_entry)) {
3566 vm_map_unlock_read(map);
3567 return (KERN_SUCCESS);
3570 next_entry = prev_entry->next;
3571 if (!(prev_entry->eflags & MAP_ENTRY_GROWS_UP)) {
3573 * This entry does not grow upwards. Since the address lies
3574 * beyond this entry, the next entry (if one exists) has to
3575 * be a downward growable entry. The entry list header is
3576 * never a growable entry, so it suffices to check the flags.
3578 if (!(next_entry->eflags & MAP_ENTRY_GROWS_DOWN)) {
3579 vm_map_unlock_read(map);
3580 return (KERN_SUCCESS);
3582 stack_entry = next_entry;
3585 * This entry grows upward. If the next entry does not at
3586 * least grow downwards, this is the entry we need to grow.
3587 * otherwise we have two possible choices and we have to
3590 if (next_entry->eflags & MAP_ENTRY_GROWS_DOWN) {
3592 * We have two choices; grow the entry closest to
3593 * the address to minimize the amount of growth.
3595 if (addr - prev_entry->end <= next_entry->start - addr)
3596 stack_entry = prev_entry;
3598 stack_entry = next_entry;
3600 stack_entry = prev_entry;
3603 if (stack_entry == next_entry) {
3604 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_DOWN, ("foo"));
3605 KASSERT(addr < stack_entry->start, ("foo"));
3606 end = (prev_entry != &map->header) ? prev_entry->end :
3607 stack_entry->start - stack_entry->avail_ssize;
3608 grow_amount = roundup(stack_entry->start - addr, PAGE_SIZE);
3609 max_grow = stack_entry->start - end;
3611 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_UP, ("foo"));
3612 KASSERT(addr >= stack_entry->end, ("foo"));
3613 end = (next_entry != &map->header) ? next_entry->start :
3614 stack_entry->end + stack_entry->avail_ssize;
3615 grow_amount = roundup(addr + 1 - stack_entry->end, PAGE_SIZE);
3616 max_grow = end - stack_entry->end;
3619 if (grow_amount > stack_entry->avail_ssize) {
3620 vm_map_unlock_read(map);
3621 return (KERN_NO_SPACE);
3625 * If there is no longer enough space between the entries nogo, and
3626 * adjust the available space. Note: this should only happen if the
3627 * user has mapped into the stack area after the stack was created,
3628 * and is probably an error.
3630 * This also effectively destroys any guard page the user might have
3631 * intended by limiting the stack size.
3633 if (grow_amount + (stack_guard_page ? PAGE_SIZE : 0) > max_grow) {
3634 if (vm_map_lock_upgrade(map))
3637 stack_entry->avail_ssize = max_grow;
3640 return (KERN_NO_SPACE);
3643 is_procstack = (addr >= (vm_offset_t)vm->vm_maxsaddr &&
3644 addr < (vm_offset_t)p->p_sysent->sv_usrstack) ? 1 : 0;
3647 * If this is the main process stack, see if we're over the stack
3650 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) {
3651 vm_map_unlock_read(map);
3652 return (KERN_NO_SPACE);
3657 if (is_procstack && racct_set(p, RACCT_STACK,
3658 ctob(vm->vm_ssize) + grow_amount)) {
3660 vm_map_unlock_read(map);
3661 return (KERN_NO_SPACE);
3667 /* Round up the grow amount modulo sgrowsiz */
3668 growsize = sgrowsiz;
3669 grow_amount = roundup(grow_amount, growsize);
3670 if (grow_amount > stack_entry->avail_ssize)
3671 grow_amount = stack_entry->avail_ssize;
3672 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) {
3673 grow_amount = trunc_page((vm_size_t)stacklim) -
3678 limit = racct_get_available(p, RACCT_STACK);
3680 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > limit))
3681 grow_amount = limit - ctob(vm->vm_ssize);
3683 if (!old_mlock && map->flags & MAP_WIREFUTURE) {
3684 if (ptoa(pmap_wired_count(map->pmap)) + grow_amount > lmemlim) {
3685 vm_map_unlock_read(map);
3692 if (racct_set(p, RACCT_MEMLOCK,
3693 ptoa(pmap_wired_count(map->pmap)) + grow_amount)) {
3695 vm_map_unlock_read(map);
3703 /* If we would blow our VMEM resource limit, no go */
3704 if (map->size + grow_amount > vmemlim) {
3705 vm_map_unlock_read(map);
3712 if (racct_set(p, RACCT_VMEM, map->size + grow_amount)) {
3714 vm_map_unlock_read(map);
3722 if (vm_map_lock_upgrade(map))
3725 if (stack_entry == next_entry) {
3729 /* Get the preliminary new entry start value */
3730 addr = stack_entry->start - grow_amount;
3733 * If this puts us into the previous entry, cut back our
3734 * growth to the available space. Also, see the note above.
3737 stack_entry->avail_ssize = max_grow;
3739 if (stack_guard_page)
3743 rv = vm_map_insert(map, NULL, 0, addr, stack_entry->start,
3744 next_entry->protection, next_entry->max_protection,
3745 MAP_STACK_GROWS_DOWN);
3747 /* Adjust the available stack space by the amount we grew. */
3748 if (rv == KERN_SUCCESS) {
3749 new_entry = prev_entry->next;
3750 KASSERT(new_entry == stack_entry->prev, ("foo"));
3751 KASSERT(new_entry->end == stack_entry->start, ("foo"));
3752 KASSERT(new_entry->start == addr, ("foo"));
3753 KASSERT((new_entry->eflags & MAP_ENTRY_GROWS_DOWN) !=
3754 0, ("new entry lacks MAP_ENTRY_GROWS_DOWN"));
3755 grow_amount = new_entry->end - new_entry->start;
3756 new_entry->avail_ssize = stack_entry->avail_ssize -
3758 stack_entry->eflags &= ~MAP_ENTRY_GROWS_DOWN;
3764 addr = stack_entry->end + grow_amount;
3767 * If this puts us into the next entry, cut back our growth
3768 * to the available space. Also, see the note above.
3771 stack_entry->avail_ssize = end - stack_entry->end;
3773 if (stack_guard_page)
3777 grow_amount = addr - stack_entry->end;
3778 cred = stack_entry->cred;
3779 if (cred == NULL && stack_entry->object.vm_object != NULL)
3780 cred = stack_entry->object.vm_object->cred;
3781 if (cred != NULL && !swap_reserve_by_cred(grow_amount, cred))
3783 /* Grow the underlying object if applicable. */
3784 else if (stack_entry->object.vm_object == NULL ||
3785 vm_object_coalesce(stack_entry->object.vm_object,
3786 stack_entry->offset,
3787 (vm_size_t)(stack_entry->end - stack_entry->start),
3788 (vm_size_t)grow_amount, cred != NULL)) {
3789 map->size += (addr - stack_entry->end);
3790 /* Update the current entry. */
3791 stack_entry->end = addr;
3792 stack_entry->avail_ssize -= grow_amount;
3793 vm_map_entry_resize_free(map, stack_entry);
3799 if (rv == KERN_SUCCESS && is_procstack)
3800 vm->vm_ssize += btoc(grow_amount);
3805 * Heed the MAP_WIREFUTURE flag if it was set for this process.
3807 if (rv == KERN_SUCCESS && (map->flags & MAP_WIREFUTURE)) {
3809 (stack_entry == next_entry) ? addr : addr - grow_amount,
3810 (stack_entry == next_entry) ? stack_entry->start : addr,
3811 (p->p_flag & P_SYSTEM)
3812 ? VM_MAP_WIRE_SYSTEM|VM_MAP_WIRE_NOHOLES
3813 : VM_MAP_WIRE_USER|VM_MAP_WIRE_NOHOLES);
3818 if (racct_enable && rv != KERN_SUCCESS) {
3820 error = racct_set(p, RACCT_VMEM, map->size);
3821 KASSERT(error == 0, ("decreasing RACCT_VMEM failed"));
3823 error = racct_set(p, RACCT_MEMLOCK,
3824 ptoa(pmap_wired_count(map->pmap)));
3825 KASSERT(error == 0, ("decreasing RACCT_MEMLOCK failed"));
3827 error = racct_set(p, RACCT_STACK, ctob(vm->vm_ssize));
3828 KASSERT(error == 0, ("decreasing RACCT_STACK failed"));
3837 * Unshare the specified VM space for exec. If other processes are
3838 * mapped to it, then create a new one. The new vmspace is null.
3841 vmspace_exec(struct proc *p, vm_offset_t minuser, vm_offset_t maxuser)
3843 struct vmspace *oldvmspace = p->p_vmspace;
3844 struct vmspace *newvmspace;
3846 KASSERT((curthread->td_pflags & TDP_EXECVMSPC) == 0,
3847 ("vmspace_exec recursed"));
3848 newvmspace = vmspace_alloc(minuser, maxuser, NULL);
3849 if (newvmspace == NULL)
3851 newvmspace->vm_swrss = oldvmspace->vm_swrss;
3853 * This code is written like this for prototype purposes. The
3854 * goal is to avoid running down the vmspace here, but let the
3855 * other process's that are still using the vmspace to finally
3856 * run it down. Even though there is little or no chance of blocking
3857 * here, it is a good idea to keep this form for future mods.
3859 PROC_VMSPACE_LOCK(p);
3860 p->p_vmspace = newvmspace;
3861 PROC_VMSPACE_UNLOCK(p);
3862 if (p == curthread->td_proc)
3863 pmap_activate(curthread);
3864 curthread->td_pflags |= TDP_EXECVMSPC;
3869 * Unshare the specified VM space for forcing COW. This
3870 * is called by rfork, for the (RFMEM|RFPROC) == 0 case.
3873 vmspace_unshare(struct proc *p)
3875 struct vmspace *oldvmspace = p->p_vmspace;
3876 struct vmspace *newvmspace;
3877 vm_ooffset_t fork_charge;
3879 if (oldvmspace->vm_refcnt == 1)
3882 newvmspace = vmspace_fork(oldvmspace, &fork_charge);
3883 if (newvmspace == NULL)
3885 if (!swap_reserve_by_cred(fork_charge, p->p_ucred)) {
3886 vmspace_free(newvmspace);
3889 PROC_VMSPACE_LOCK(p);
3890 p->p_vmspace = newvmspace;
3891 PROC_VMSPACE_UNLOCK(p);
3892 if (p == curthread->td_proc)
3893 pmap_activate(curthread);
3894 vmspace_free(oldvmspace);
3901 * Finds the VM object, offset, and
3902 * protection for a given virtual address in the
3903 * specified map, assuming a page fault of the
3906 * Leaves the map in question locked for read; return
3907 * values are guaranteed until a vm_map_lookup_done
3908 * call is performed. Note that the map argument
3909 * is in/out; the returned map must be used in
3910 * the call to vm_map_lookup_done.
3912 * A handle (out_entry) is returned for use in
3913 * vm_map_lookup_done, to make that fast.
3915 * If a lookup is requested with "write protection"
3916 * specified, the map may be changed to perform virtual
3917 * copying operations, although the data referenced will
3921 vm_map_lookup(vm_map_t *var_map, /* IN/OUT */
3923 vm_prot_t fault_typea,
3924 vm_map_entry_t *out_entry, /* OUT */
3925 vm_object_t *object, /* OUT */
3926 vm_pindex_t *pindex, /* OUT */
3927 vm_prot_t *out_prot, /* OUT */
3928 boolean_t *wired) /* OUT */
3930 vm_map_entry_t entry;
3931 vm_map_t map = *var_map;
3933 vm_prot_t fault_type = fault_typea;
3934 vm_object_t eobject;
3940 vm_map_lock_read(map);
3943 * Lookup the faulting address.
3945 if (!vm_map_lookup_entry(map, vaddr, out_entry)) {
3946 vm_map_unlock_read(map);
3947 return (KERN_INVALID_ADDRESS);
3955 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
3956 vm_map_t old_map = map;
3958 *var_map = map = entry->object.sub_map;
3959 vm_map_unlock_read(old_map);
3964 * Check whether this task is allowed to have this page.
3966 prot = entry->protection;
3967 fault_type &= (VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE);
3968 if ((fault_type & prot) != fault_type || prot == VM_PROT_NONE) {
3969 vm_map_unlock_read(map);
3970 return (KERN_PROTECTION_FAILURE);
3972 KASSERT((prot & VM_PROT_WRITE) == 0 || (entry->eflags &
3973 (MAP_ENTRY_USER_WIRED | MAP_ENTRY_NEEDS_COPY)) !=
3974 (MAP_ENTRY_USER_WIRED | MAP_ENTRY_NEEDS_COPY),
3975 ("entry %p flags %x", entry, entry->eflags));
3976 if ((fault_typea & VM_PROT_COPY) != 0 &&
3977 (entry->max_protection & VM_PROT_WRITE) == 0 &&
3978 (entry->eflags & MAP_ENTRY_COW) == 0) {
3979 vm_map_unlock_read(map);
3980 return (KERN_PROTECTION_FAILURE);
3984 * If this page is not pageable, we have to get it for all possible
3987 *wired = (entry->wired_count != 0);
3989 fault_type = entry->protection;
3990 size = entry->end - entry->start;
3992 * If the entry was copy-on-write, we either ...
3994 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
3996 * If we want to write the page, we may as well handle that
3997 * now since we've got the map locked.
3999 * If we don't need to write the page, we just demote the
4000 * permissions allowed.
4002 if ((fault_type & VM_PROT_WRITE) != 0 ||
4003 (fault_typea & VM_PROT_COPY) != 0) {
4005 * Make a new object, and place it in the object
4006 * chain. Note that no new references have appeared
4007 * -- one just moved from the map to the new
4010 if (vm_map_lock_upgrade(map))
4013 if (entry->cred == NULL) {
4015 * The debugger owner is charged for
4018 cred = curthread->td_ucred;
4020 if (!swap_reserve_by_cred(size, cred)) {
4023 return (KERN_RESOURCE_SHORTAGE);
4027 vm_object_shadow(&entry->object.vm_object,
4028 &entry->offset, size);
4029 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
4030 eobject = entry->object.vm_object;
4031 if (eobject->cred != NULL) {
4033 * The object was not shadowed.
4035 swap_release_by_cred(size, entry->cred);
4036 crfree(entry->cred);
4038 } else if (entry->cred != NULL) {
4039 VM_OBJECT_WLOCK(eobject);
4040 eobject->cred = entry->cred;
4041 eobject->charge = size;
4042 VM_OBJECT_WUNLOCK(eobject);
4046 vm_map_lock_downgrade(map);
4049 * We're attempting to read a copy-on-write page --
4050 * don't allow writes.
4052 prot &= ~VM_PROT_WRITE;
4057 * Create an object if necessary.
4059 if (entry->object.vm_object == NULL &&
4061 if (vm_map_lock_upgrade(map))
4063 entry->object.vm_object = vm_object_allocate(OBJT_DEFAULT,
4066 if (entry->cred != NULL) {
4067 VM_OBJECT_WLOCK(entry->object.vm_object);
4068 entry->object.vm_object->cred = entry->cred;
4069 entry->object.vm_object->charge = size;
4070 VM_OBJECT_WUNLOCK(entry->object.vm_object);
4073 vm_map_lock_downgrade(map);
4077 * Return the object/offset from this entry. If the entry was
4078 * copy-on-write or empty, it has been fixed up.
4080 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
4081 *object = entry->object.vm_object;
4084 return (KERN_SUCCESS);
4088 * vm_map_lookup_locked:
4090 * Lookup the faulting address. A version of vm_map_lookup that returns
4091 * KERN_FAILURE instead of blocking on map lock or memory allocation.
4094 vm_map_lookup_locked(vm_map_t *var_map, /* IN/OUT */
4096 vm_prot_t fault_typea,
4097 vm_map_entry_t *out_entry, /* OUT */
4098 vm_object_t *object, /* OUT */
4099 vm_pindex_t *pindex, /* OUT */
4100 vm_prot_t *out_prot, /* OUT */
4101 boolean_t *wired) /* OUT */
4103 vm_map_entry_t entry;
4104 vm_map_t map = *var_map;
4106 vm_prot_t fault_type = fault_typea;
4109 * Lookup the faulting address.
4111 if (!vm_map_lookup_entry(map, vaddr, out_entry))
4112 return (KERN_INVALID_ADDRESS);
4117 * Fail if the entry refers to a submap.
4119 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP)
4120 return (KERN_FAILURE);
4123 * Check whether this task is allowed to have this page.
4125 prot = entry->protection;
4126 fault_type &= VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE;
4127 if ((fault_type & prot) != fault_type)
4128 return (KERN_PROTECTION_FAILURE);
4131 * If this page is not pageable, we have to get it for all possible
4134 *wired = (entry->wired_count != 0);
4136 fault_type = entry->protection;
4138 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
4140 * Fail if the entry was copy-on-write for a write fault.
4142 if (fault_type & VM_PROT_WRITE)
4143 return (KERN_FAILURE);
4145 * We're attempting to read a copy-on-write page --
4146 * don't allow writes.
4148 prot &= ~VM_PROT_WRITE;
4152 * Fail if an object should be created.
4154 if (entry->object.vm_object == NULL && !map->system_map)
4155 return (KERN_FAILURE);
4158 * Return the object/offset from this entry. If the entry was
4159 * copy-on-write or empty, it has been fixed up.
4161 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
4162 *object = entry->object.vm_object;
4165 return (KERN_SUCCESS);
4169 * vm_map_lookup_done:
4171 * Releases locks acquired by a vm_map_lookup
4172 * (according to the handle returned by that lookup).
4175 vm_map_lookup_done(vm_map_t map, vm_map_entry_t entry)
4178 * Unlock the main-level map
4180 vm_map_unlock_read(map);
4183 #include "opt_ddb.h"
4185 #include <sys/kernel.h>
4187 #include <ddb/ddb.h>
4190 vm_map_print(vm_map_t map)
4192 vm_map_entry_t entry;
4194 db_iprintf("Task map %p: pmap=%p, nentries=%d, version=%u\n",
4196 (void *)map->pmap, map->nentries, map->timestamp);
4199 for (entry = map->header.next; entry != &map->header;
4200 entry = entry->next) {
4201 db_iprintf("map entry %p: start=%p, end=%p\n",
4202 (void *)entry, (void *)entry->start, (void *)entry->end);
4204 static char *inheritance_name[4] =
4205 {"share", "copy", "none", "donate_copy"};
4207 db_iprintf(" prot=%x/%x/%s",
4209 entry->max_protection,
4210 inheritance_name[(int)(unsigned char)entry->inheritance]);
4211 if (entry->wired_count != 0)
4212 db_printf(", wired");
4214 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
4215 db_printf(", share=%p, offset=0x%jx\n",
4216 (void *)entry->object.sub_map,
4217 (uintmax_t)entry->offset);
4218 if ((entry->prev == &map->header) ||
4219 (entry->prev->object.sub_map !=
4220 entry->object.sub_map)) {
4222 vm_map_print((vm_map_t)entry->object.sub_map);
4226 if (entry->cred != NULL)
4227 db_printf(", ruid %d", entry->cred->cr_ruid);
4228 db_printf(", object=%p, offset=0x%jx",
4229 (void *)entry->object.vm_object,
4230 (uintmax_t)entry->offset);
4231 if (entry->object.vm_object && entry->object.vm_object->cred)
4232 db_printf(", obj ruid %d charge %jx",
4233 entry->object.vm_object->cred->cr_ruid,
4234 (uintmax_t)entry->object.vm_object->charge);
4235 if (entry->eflags & MAP_ENTRY_COW)
4236 db_printf(", copy (%s)",
4237 (entry->eflags & MAP_ENTRY_NEEDS_COPY) ? "needed" : "done");
4240 if ((entry->prev == &map->header) ||
4241 (entry->prev->object.vm_object !=
4242 entry->object.vm_object)) {
4244 vm_object_print((db_expr_t)(intptr_t)
4245 entry->object.vm_object,
4254 DB_SHOW_COMMAND(map, map)
4258 db_printf("usage: show map <addr>\n");
4261 vm_map_print((vm_map_t)addr);
4264 DB_SHOW_COMMAND(procvm, procvm)
4269 p = (struct proc *) addr;
4274 db_printf("p = %p, vmspace = %p, map = %p, pmap = %p\n",
4275 (void *)p, (void *)p->p_vmspace, (void *)&p->p_vmspace->vm_map,
4276 (void *)vmspace_pmap(p->p_vmspace));
4278 vm_map_print((vm_map_t)&p->p_vmspace->vm_map);