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 int vm_map_growstack(vm_map_t map, vm_offset_t addr,
137 vm_map_entry_t gap_entry);
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;
1139 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"));
1148 KASSERT((prot & ~max) == 0,
1149 ("prot %#x is not subset of max_prot %#x", prot, max));
1152 * Check that the start and end points are not bogus.
1154 if (start < map->min_offset || end > map->max_offset || start >= end)
1155 return (KERN_INVALID_ADDRESS);
1158 * Find the entry prior to the proposed starting address; if it's part
1159 * of an existing entry, this range is bogus.
1161 if (vm_map_lookup_entry(map, start, &temp_entry))
1162 return (KERN_NO_SPACE);
1164 prev_entry = temp_entry;
1167 * Assert that the next entry doesn't overlap the end point.
1169 if (prev_entry->next != &map->header && prev_entry->next->start < end)
1170 return (KERN_NO_SPACE);
1172 if ((cow & MAP_CREATE_GUARD) != 0 && (object != NULL ||
1173 max != VM_PROT_NONE))
1174 return (KERN_INVALID_ARGUMENT);
1177 if (cow & MAP_COPY_ON_WRITE)
1178 protoeflags |= MAP_ENTRY_COW | MAP_ENTRY_NEEDS_COPY;
1179 if (cow & MAP_NOFAULT)
1180 protoeflags |= MAP_ENTRY_NOFAULT;
1181 if (cow & MAP_DISABLE_SYNCER)
1182 protoeflags |= MAP_ENTRY_NOSYNC;
1183 if (cow & MAP_DISABLE_COREDUMP)
1184 protoeflags |= MAP_ENTRY_NOCOREDUMP;
1185 if (cow & MAP_STACK_GROWS_DOWN)
1186 protoeflags |= MAP_ENTRY_GROWS_DOWN;
1187 if (cow & MAP_STACK_GROWS_UP)
1188 protoeflags |= MAP_ENTRY_GROWS_UP;
1189 if (cow & MAP_VN_WRITECOUNT)
1190 protoeflags |= MAP_ENTRY_VN_WRITECNT;
1191 if ((cow & MAP_CREATE_GUARD) != 0)
1192 protoeflags |= MAP_ENTRY_GUARD;
1193 if ((cow & MAP_CREATE_STACK_GAP_DN) != 0)
1194 protoeflags |= MAP_ENTRY_STACK_GAP_DN;
1195 if ((cow & MAP_CREATE_STACK_GAP_UP) != 0)
1196 protoeflags |= MAP_ENTRY_STACK_GAP_UP;
1197 if (cow & MAP_INHERIT_SHARE)
1198 inheritance = VM_INHERIT_SHARE;
1200 inheritance = VM_INHERIT_DEFAULT;
1203 if ((cow & (MAP_ACC_NO_CHARGE | MAP_NOFAULT | MAP_CREATE_GUARD)) != 0)
1205 if ((cow & MAP_ACC_CHARGED) || ((prot & VM_PROT_WRITE) &&
1206 ((protoeflags & MAP_ENTRY_NEEDS_COPY) || object == NULL))) {
1207 if (!(cow & MAP_ACC_CHARGED) && !swap_reserve(end - start))
1208 return (KERN_RESOURCE_SHORTAGE);
1209 KASSERT(object == NULL ||
1210 (protoeflags & MAP_ENTRY_NEEDS_COPY) != 0 ||
1211 object->cred == NULL,
1212 ("overcommit: vm_map_insert o %p", object));
1213 cred = curthread->td_ucred;
1217 /* Expand the kernel pmap, if necessary. */
1218 if (map == kernel_map && end > kernel_vm_end)
1219 pmap_growkernel(end);
1220 if (object != NULL) {
1222 * OBJ_ONEMAPPING must be cleared unless this mapping
1223 * is trivially proven to be the only mapping for any
1224 * of the object's pages. (Object granularity
1225 * reference counting is insufficient to recognize
1226 * aliases with precision.)
1228 VM_OBJECT_WLOCK(object);
1229 if (object->ref_count > 1 || object->shadow_count != 0)
1230 vm_object_clear_flag(object, OBJ_ONEMAPPING);
1231 VM_OBJECT_WUNLOCK(object);
1232 } else if (prev_entry != &map->header &&
1233 prev_entry->eflags == protoeflags &&
1234 (cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 &&
1235 prev_entry->end == start && prev_entry->wired_count == 0 &&
1236 (prev_entry->cred == cred ||
1237 (prev_entry->object.vm_object != NULL &&
1238 prev_entry->object.vm_object->cred == cred)) &&
1239 vm_object_coalesce(prev_entry->object.vm_object,
1241 (vm_size_t)(prev_entry->end - prev_entry->start),
1242 (vm_size_t)(end - prev_entry->end), cred != NULL &&
1243 (protoeflags & MAP_ENTRY_NEEDS_COPY) == 0)) {
1245 * We were able to extend the object. Determine if we
1246 * can extend the previous map entry to include the
1247 * new range as well.
1249 if (prev_entry->inheritance == inheritance &&
1250 prev_entry->protection == prot &&
1251 prev_entry->max_protection == max) {
1252 if ((prev_entry->eflags & MAP_ENTRY_GUARD) == 0)
1253 map->size += end - prev_entry->end;
1254 prev_entry->end = end;
1255 vm_map_entry_resize_free(map, prev_entry);
1256 vm_map_simplify_entry(map, prev_entry);
1257 return (KERN_SUCCESS);
1261 * If we can extend the object but cannot extend the
1262 * map entry, we have to create a new map entry. We
1263 * must bump the ref count on the extended object to
1264 * account for it. object may be NULL.
1266 object = prev_entry->object.vm_object;
1267 offset = prev_entry->offset +
1268 (prev_entry->end - prev_entry->start);
1269 vm_object_reference(object);
1270 if (cred != NULL && object != NULL && object->cred != NULL &&
1271 !(prev_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
1272 /* Object already accounts for this uid. */
1280 * Create a new entry
1282 new_entry = vm_map_entry_create(map);
1283 new_entry->start = start;
1284 new_entry->end = end;
1285 new_entry->cred = NULL;
1287 new_entry->eflags = protoeflags;
1288 new_entry->object.vm_object = object;
1289 new_entry->offset = offset;
1291 new_entry->inheritance = inheritance;
1292 new_entry->protection = prot;
1293 new_entry->max_protection = max;
1294 new_entry->wired_count = 0;
1295 new_entry->wiring_thread = NULL;
1296 new_entry->read_ahead = VM_FAULT_READ_AHEAD_INIT;
1297 new_entry->next_read = OFF_TO_IDX(offset);
1299 KASSERT(cred == NULL || !ENTRY_CHARGED(new_entry),
1300 ("overcommit: vm_map_insert leaks vm_map %p", new_entry));
1301 new_entry->cred = cred;
1304 * Insert the new entry into the list
1306 vm_map_entry_link(map, prev_entry, new_entry);
1307 if ((new_entry->eflags & MAP_ENTRY_GUARD) == 0)
1308 map->size += new_entry->end - new_entry->start;
1311 * Try to coalesce the new entry with both the previous and next
1312 * entries in the list. Previously, we only attempted to coalesce
1313 * with the previous entry when object is NULL. Here, we handle the
1314 * other cases, which are less common.
1316 vm_map_simplify_entry(map, new_entry);
1318 if ((cow & (MAP_PREFAULT | MAP_PREFAULT_PARTIAL)) != 0) {
1319 vm_map_pmap_enter(map, start, prot, object, OFF_TO_IDX(offset),
1320 end - start, cow & MAP_PREFAULT_PARTIAL);
1323 return (KERN_SUCCESS);
1329 * Find the first fit (lowest VM address) for "length" free bytes
1330 * beginning at address >= start in the given map.
1332 * In a vm_map_entry, "adj_free" is the amount of free space
1333 * adjacent (higher address) to this entry, and "max_free" is the
1334 * maximum amount of contiguous free space in its subtree. This
1335 * allows finding a free region in one path down the tree, so
1336 * O(log n) amortized with splay trees.
1338 * The map must be locked, and leaves it so.
1340 * Returns: 0 on success, and starting address in *addr,
1341 * 1 if insufficient space.
1344 vm_map_findspace(vm_map_t map, vm_offset_t start, vm_size_t length,
1345 vm_offset_t *addr) /* OUT */
1347 vm_map_entry_t entry;
1351 * Request must fit within min/max VM address and must avoid
1354 if (start < map->min_offset)
1355 start = map->min_offset;
1356 if (start + length > map->max_offset || start + length < start)
1359 /* Empty tree means wide open address space. */
1360 if (map->root == NULL) {
1366 * After splay, if start comes before root node, then there
1367 * must be a gap from start to the root.
1369 map->root = vm_map_entry_splay(start, map->root);
1370 if (start + length <= map->root->start) {
1376 * Root is the last node that might begin its gap before
1377 * start, and this is the last comparison where address
1378 * wrap might be a problem.
1380 st = (start > map->root->end) ? start : map->root->end;
1381 if (length <= map->root->end + map->root->adj_free - st) {
1386 /* With max_free, can immediately tell if no solution. */
1387 entry = map->root->right;
1388 if (entry == NULL || length > entry->max_free)
1392 * Search the right subtree in the order: left subtree, root,
1393 * right subtree (first fit). The previous splay implies that
1394 * all regions in the right subtree have addresses > start.
1396 while (entry != NULL) {
1397 if (entry->left != NULL && entry->left->max_free >= length)
1398 entry = entry->left;
1399 else if (entry->adj_free >= length) {
1403 entry = entry->right;
1406 /* Can't get here, so panic if we do. */
1407 panic("vm_map_findspace: max_free corrupt");
1411 vm_map_fixed(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1412 vm_offset_t start, vm_size_t length, vm_prot_t prot,
1413 vm_prot_t max, int cow)
1418 end = start + length;
1419 KASSERT((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 ||
1421 ("vm_map_fixed: non-NULL backing object for stack"));
1423 VM_MAP_RANGE_CHECK(map, start, end);
1424 if ((cow & MAP_CHECK_EXCL) == 0)
1425 vm_map_delete(map, start, end);
1426 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) != 0) {
1427 result = vm_map_stack_locked(map, start, length, sgrowsiz,
1430 result = vm_map_insert(map, object, offset, start, end,
1438 * vm_map_find finds an unallocated region in the target address
1439 * map with the given length. The search is defined to be
1440 * first-fit from the specified address; the region found is
1441 * returned in the same parameter.
1443 * If object is non-NULL, ref count must be bumped by caller
1444 * prior to making call to account for the new entry.
1447 vm_map_find(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1448 vm_offset_t *addr, /* IN/OUT */
1449 vm_size_t length, vm_offset_t max_addr, int find_space,
1450 vm_prot_t prot, vm_prot_t max, int cow)
1452 vm_offset_t alignment, initial_addr, start;
1455 KASSERT((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 ||
1457 ("vm_map_find: non-NULL backing object for stack"));
1458 if (find_space == VMFS_OPTIMAL_SPACE && (object == NULL ||
1459 (object->flags & OBJ_COLORED) == 0))
1460 find_space = VMFS_ANY_SPACE;
1461 if (find_space >> 8 != 0) {
1462 KASSERT((find_space & 0xff) == 0, ("bad VMFS flags"));
1463 alignment = (vm_offset_t)1 << (find_space >> 8);
1466 initial_addr = *addr;
1468 start = initial_addr;
1471 if (find_space != VMFS_NO_SPACE) {
1472 if (vm_map_findspace(map, start, length, addr) ||
1473 (max_addr != 0 && *addr + length > max_addr)) {
1475 if (find_space == VMFS_OPTIMAL_SPACE) {
1476 find_space = VMFS_ANY_SPACE;
1479 return (KERN_NO_SPACE);
1481 switch (find_space) {
1482 case VMFS_SUPER_SPACE:
1483 case VMFS_OPTIMAL_SPACE:
1484 pmap_align_superpage(object, offset, addr,
1487 case VMFS_ANY_SPACE:
1490 if ((*addr & (alignment - 1)) != 0) {
1491 *addr &= ~(alignment - 1);
1499 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) != 0) {
1500 result = vm_map_stack_locked(map, start, length,
1501 sgrowsiz, prot, max, cow);
1503 result = vm_map_insert(map, object, offset, start,
1504 start + length, prot, max, cow);
1506 } while (result == KERN_NO_SPACE && find_space != VMFS_NO_SPACE &&
1507 find_space != VMFS_ANY_SPACE);
1513 vm_map_find_min(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1514 vm_offset_t *addr, vm_size_t length, vm_offset_t min_addr,
1515 vm_offset_t max_addr, int find_space, vm_prot_t prot, vm_prot_t max,
1523 rv = vm_map_find(map, object, offset, addr, length, max_addr,
1524 find_space, prot, max, cow);
1525 if (rv == KERN_SUCCESS || min_addr >= hint)
1527 *addr = hint = min_addr;
1532 * vm_map_simplify_entry:
1534 * Simplify the given map entry by merging with either neighbor. This
1535 * routine also has the ability to merge with both neighbors.
1537 * The map must be locked.
1539 * This routine guarentees that the passed entry remains valid (though
1540 * possibly extended). When merging, this routine may delete one or
1544 vm_map_simplify_entry(vm_map_t map, vm_map_entry_t entry)
1546 vm_map_entry_t next, prev;
1547 vm_size_t prevsize, esize;
1549 if ((entry->eflags & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP |
1550 MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_IS_SUB_MAP)) != 0)
1554 if (prev != &map->header) {
1555 prevsize = prev->end - prev->start;
1556 if ( (prev->end == entry->start) &&
1557 (prev->object.vm_object == entry->object.vm_object) &&
1558 (!prev->object.vm_object ||
1559 (prev->offset + prevsize == entry->offset)) &&
1560 (prev->eflags == entry->eflags) &&
1561 (prev->protection == entry->protection) &&
1562 (prev->max_protection == entry->max_protection) &&
1563 (prev->inheritance == entry->inheritance) &&
1564 (prev->wired_count == entry->wired_count) &&
1565 (prev->cred == entry->cred)) {
1566 vm_map_entry_unlink(map, prev);
1567 entry->start = prev->start;
1568 entry->offset = prev->offset;
1569 if (entry->prev != &map->header)
1570 vm_map_entry_resize_free(map, entry->prev);
1573 * If the backing object is a vnode object,
1574 * vm_object_deallocate() calls vrele().
1575 * However, vrele() does not lock the vnode
1576 * because the vnode has additional
1577 * references. Thus, the map lock can be kept
1578 * without causing a lock-order reversal with
1581 * Since we count the number of virtual page
1582 * mappings in object->un_pager.vnp.writemappings,
1583 * the writemappings value should not be adjusted
1584 * when the entry is disposed of.
1586 if (prev->object.vm_object)
1587 vm_object_deallocate(prev->object.vm_object);
1588 if (prev->cred != NULL)
1590 vm_map_entry_dispose(map, prev);
1595 if (next != &map->header) {
1596 esize = entry->end - entry->start;
1597 if ((entry->end == next->start) &&
1598 (next->object.vm_object == entry->object.vm_object) &&
1599 (!entry->object.vm_object ||
1600 (entry->offset + esize == next->offset)) &&
1601 (next->eflags == entry->eflags) &&
1602 (next->protection == entry->protection) &&
1603 (next->max_protection == entry->max_protection) &&
1604 (next->inheritance == entry->inheritance) &&
1605 (next->wired_count == entry->wired_count) &&
1606 (next->cred == entry->cred)) {
1607 vm_map_entry_unlink(map, next);
1608 entry->end = next->end;
1609 vm_map_entry_resize_free(map, entry);
1612 * See comment above.
1614 if (next->object.vm_object)
1615 vm_object_deallocate(next->object.vm_object);
1616 if (next->cred != NULL)
1618 vm_map_entry_dispose(map, next);
1623 * vm_map_clip_start: [ internal use only ]
1625 * Asserts that the given entry begins at or after
1626 * the specified address; if necessary,
1627 * it splits the entry into two.
1629 #define vm_map_clip_start(map, entry, startaddr) \
1631 if (startaddr > entry->start) \
1632 _vm_map_clip_start(map, entry, startaddr); \
1636 * This routine is called only when it is known that
1637 * the entry must be split.
1640 _vm_map_clip_start(vm_map_t map, vm_map_entry_t entry, vm_offset_t start)
1642 vm_map_entry_t new_entry;
1644 VM_MAP_ASSERT_LOCKED(map);
1647 * Split off the front portion -- note that we must insert the new
1648 * entry BEFORE this one, so that this entry has the specified
1651 vm_map_simplify_entry(map, entry);
1654 * If there is no object backing this entry, we might as well create
1655 * one now. If we defer it, an object can get created after the map
1656 * is clipped, and individual objects will be created for the split-up
1657 * map. This is a bit of a hack, but is also about the best place to
1658 * put this improvement.
1660 if (entry->object.vm_object == NULL && !map->system_map &&
1661 (entry->eflags & MAP_ENTRY_GUARD) == 0) {
1663 object = vm_object_allocate(OBJT_DEFAULT,
1664 atop(entry->end - entry->start));
1665 entry->object.vm_object = object;
1667 if (entry->cred != NULL) {
1668 object->cred = entry->cred;
1669 object->charge = entry->end - entry->start;
1672 } else if (entry->object.vm_object != NULL &&
1673 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) &&
1674 entry->cred != NULL) {
1675 VM_OBJECT_WLOCK(entry->object.vm_object);
1676 KASSERT(entry->object.vm_object->cred == NULL,
1677 ("OVERCOMMIT: vm_entry_clip_start: both cred e %p", entry));
1678 entry->object.vm_object->cred = entry->cred;
1679 entry->object.vm_object->charge = entry->end - entry->start;
1680 VM_OBJECT_WUNLOCK(entry->object.vm_object);
1684 new_entry = vm_map_entry_create(map);
1685 *new_entry = *entry;
1687 new_entry->end = start;
1688 entry->offset += (start - entry->start);
1689 entry->start = start;
1690 if (new_entry->cred != NULL)
1691 crhold(entry->cred);
1693 vm_map_entry_link(map, entry->prev, new_entry);
1695 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1696 vm_object_reference(new_entry->object.vm_object);
1698 * The object->un_pager.vnp.writemappings for the
1699 * object of MAP_ENTRY_VN_WRITECNT type entry shall be
1700 * kept as is here. The virtual pages are
1701 * re-distributed among the clipped entries, so the sum is
1708 * vm_map_clip_end: [ internal use only ]
1710 * Asserts that the given entry ends at or before
1711 * the specified address; if necessary,
1712 * it splits the entry into two.
1714 #define vm_map_clip_end(map, entry, endaddr) \
1716 if ((endaddr) < (entry->end)) \
1717 _vm_map_clip_end((map), (entry), (endaddr)); \
1721 * This routine is called only when it is known that
1722 * the entry must be split.
1725 _vm_map_clip_end(vm_map_t map, vm_map_entry_t entry, vm_offset_t end)
1727 vm_map_entry_t new_entry;
1729 VM_MAP_ASSERT_LOCKED(map);
1732 * If there is no object backing this entry, we might as well create
1733 * one now. If we defer it, an object can get created after the map
1734 * is clipped, and individual objects will be created for the split-up
1735 * map. This is a bit of a hack, but is also about the best place to
1736 * put this improvement.
1738 if (entry->object.vm_object == NULL && !map->system_map &&
1739 (entry->eflags & MAP_ENTRY_GUARD) == 0) {
1741 object = vm_object_allocate(OBJT_DEFAULT,
1742 atop(entry->end - entry->start));
1743 entry->object.vm_object = object;
1745 if (entry->cred != NULL) {
1746 object->cred = entry->cred;
1747 object->charge = entry->end - entry->start;
1750 } else if (entry->object.vm_object != NULL &&
1751 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) &&
1752 entry->cred != NULL) {
1753 VM_OBJECT_WLOCK(entry->object.vm_object);
1754 KASSERT(entry->object.vm_object->cred == NULL,
1755 ("OVERCOMMIT: vm_entry_clip_end: both cred e %p", entry));
1756 entry->object.vm_object->cred = entry->cred;
1757 entry->object.vm_object->charge = entry->end - entry->start;
1758 VM_OBJECT_WUNLOCK(entry->object.vm_object);
1763 * Create a new entry and insert it AFTER the specified entry
1765 new_entry = vm_map_entry_create(map);
1766 *new_entry = *entry;
1768 new_entry->start = entry->end = end;
1769 new_entry->offset += (end - entry->start);
1770 if (new_entry->cred != NULL)
1771 crhold(entry->cred);
1773 vm_map_entry_link(map, entry, new_entry);
1775 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1776 vm_object_reference(new_entry->object.vm_object);
1781 * vm_map_submap: [ kernel use only ]
1783 * Mark the given range as handled by a subordinate map.
1785 * This range must have been created with vm_map_find,
1786 * and no other operations may have been performed on this
1787 * range prior to calling vm_map_submap.
1789 * Only a limited number of operations can be performed
1790 * within this rage after calling vm_map_submap:
1792 * [Don't try vm_map_copy!]
1794 * To remove a submapping, one must first remove the
1795 * range from the superior map, and then destroy the
1796 * submap (if desired). [Better yet, don't try it.]
1805 vm_map_entry_t entry;
1806 int result = KERN_INVALID_ARGUMENT;
1810 VM_MAP_RANGE_CHECK(map, start, end);
1812 if (vm_map_lookup_entry(map, start, &entry)) {
1813 vm_map_clip_start(map, entry, start);
1815 entry = entry->next;
1817 vm_map_clip_end(map, entry, end);
1819 if ((entry->start == start) && (entry->end == end) &&
1820 ((entry->eflags & MAP_ENTRY_COW) == 0) &&
1821 (entry->object.vm_object == NULL)) {
1822 entry->object.sub_map = submap;
1823 entry->eflags |= MAP_ENTRY_IS_SUB_MAP;
1824 result = KERN_SUCCESS;
1832 * The maximum number of pages to map if MAP_PREFAULT_PARTIAL is specified
1834 #define MAX_INIT_PT 96
1837 * vm_map_pmap_enter:
1839 * Preload the specified map's pmap with mappings to the specified
1840 * object's memory-resident pages. No further physical pages are
1841 * allocated, and no further virtual pages are retrieved from secondary
1842 * storage. If the specified flags include MAP_PREFAULT_PARTIAL, then a
1843 * limited number of page mappings are created at the low-end of the
1844 * specified address range. (For this purpose, a superpage mapping
1845 * counts as one page mapping.) Otherwise, all resident pages within
1846 * the specified address range are mapped. Because these mappings are
1847 * being created speculatively, cached pages are not reactivated and
1851 vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot,
1852 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags)
1855 vm_page_t p, p_start;
1856 vm_pindex_t mask, psize, threshold, tmpidx;
1858 if ((prot & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0 || object == NULL)
1860 VM_OBJECT_RLOCK(object);
1861 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) {
1862 VM_OBJECT_RUNLOCK(object);
1863 VM_OBJECT_WLOCK(object);
1864 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) {
1865 pmap_object_init_pt(map->pmap, addr, object, pindex,
1867 VM_OBJECT_WUNLOCK(object);
1870 VM_OBJECT_LOCK_DOWNGRADE(object);
1874 if (psize + pindex > object->size) {
1875 if (object->size < pindex) {
1876 VM_OBJECT_RUNLOCK(object);
1879 psize = object->size - pindex;
1884 threshold = MAX_INIT_PT;
1886 p = vm_page_find_least(object, pindex);
1888 * Assert: the variable p is either (1) the page with the
1889 * least pindex greater than or equal to the parameter pindex
1893 p != NULL && (tmpidx = p->pindex - pindex) < psize;
1894 p = TAILQ_NEXT(p, listq)) {
1896 * don't allow an madvise to blow away our really
1897 * free pages allocating pv entries.
1899 if (((flags & MAP_PREFAULT_MADVISE) != 0 &&
1900 cnt.v_free_count < cnt.v_free_reserved) ||
1901 ((flags & MAP_PREFAULT_PARTIAL) != 0 &&
1902 tmpidx >= threshold)) {
1906 if (p->valid == VM_PAGE_BITS_ALL) {
1907 if (p_start == NULL) {
1908 start = addr + ptoa(tmpidx);
1911 /* Jump ahead if a superpage mapping is possible. */
1912 if (p->psind > 0 && ((addr + ptoa(tmpidx)) &
1913 (pagesizes[p->psind] - 1)) == 0) {
1914 mask = atop(pagesizes[p->psind]) - 1;
1915 if (tmpidx + mask < psize &&
1916 vm_page_ps_is_valid(p)) {
1921 } else if (p_start != NULL) {
1922 pmap_enter_object(map->pmap, start, addr +
1923 ptoa(tmpidx), p_start, prot);
1927 if (p_start != NULL)
1928 pmap_enter_object(map->pmap, start, addr + ptoa(psize),
1930 VM_OBJECT_RUNLOCK(object);
1936 * Sets the protection of the specified address
1937 * region in the target map. If "set_max" is
1938 * specified, the maximum protection is to be set;
1939 * otherwise, only the current protection is affected.
1942 vm_map_protect(vm_map_t map, vm_offset_t start, vm_offset_t end,
1943 vm_prot_t new_prot, boolean_t set_max)
1945 vm_map_entry_t current, entry;
1951 return (KERN_SUCCESS);
1955 VM_MAP_RANGE_CHECK(map, start, end);
1957 if (vm_map_lookup_entry(map, start, &entry)) {
1958 vm_map_clip_start(map, entry, start);
1960 entry = entry->next;
1964 * Make a first pass to check for protection violations.
1966 for (current = entry; current != &map->header && current->start < end;
1967 current = current->next) {
1968 if ((current->eflags & MAP_ENTRY_GUARD) != 0)
1970 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
1972 return (KERN_INVALID_ARGUMENT);
1974 if ((new_prot & current->max_protection) != new_prot) {
1976 return (KERN_PROTECTION_FAILURE);
1981 * Do an accounting pass for private read-only mappings that
1982 * now will do cow due to allowed write (e.g. debugger sets
1983 * breakpoint on text segment)
1985 for (current = entry; current != &map->header && current->start < end;
1986 current = current->next) {
1988 vm_map_clip_end(map, current, end);
1991 ((new_prot & ~(current->protection)) & VM_PROT_WRITE) == 0 ||
1992 ENTRY_CHARGED(current) ||
1993 (current->eflags & MAP_ENTRY_GUARD) != 0) {
1997 cred = curthread->td_ucred;
1998 obj = current->object.vm_object;
2000 if (obj == NULL || (current->eflags & MAP_ENTRY_NEEDS_COPY)) {
2001 if (!swap_reserve(current->end - current->start)) {
2003 return (KERN_RESOURCE_SHORTAGE);
2006 current->cred = cred;
2010 VM_OBJECT_WLOCK(obj);
2011 if (obj->type != OBJT_DEFAULT && obj->type != OBJT_SWAP) {
2012 VM_OBJECT_WUNLOCK(obj);
2017 * Charge for the whole object allocation now, since
2018 * we cannot distinguish between non-charged and
2019 * charged clipped mapping of the same object later.
2021 KASSERT(obj->charge == 0,
2022 ("vm_map_protect: object %p overcharged (entry %p)",
2024 if (!swap_reserve(ptoa(obj->size))) {
2025 VM_OBJECT_WUNLOCK(obj);
2027 return (KERN_RESOURCE_SHORTAGE);
2032 obj->charge = ptoa(obj->size);
2033 VM_OBJECT_WUNLOCK(obj);
2037 * Go back and fix up protections. [Note that clipping is not
2038 * necessary the second time.]
2040 for (current = entry; current != &map->header && current->start < end;
2041 current = current->next) {
2042 if ((current->eflags & MAP_ENTRY_GUARD) != 0)
2045 old_prot = current->protection;
2048 current->protection =
2049 (current->max_protection = new_prot) &
2052 current->protection = new_prot;
2055 * For user wired map entries, the normal lazy evaluation of
2056 * write access upgrades through soft page faults is
2057 * undesirable. Instead, immediately copy any pages that are
2058 * copy-on-write and enable write access in the physical map.
2060 if ((current->eflags & MAP_ENTRY_USER_WIRED) != 0 &&
2061 (current->protection & VM_PROT_WRITE) != 0 &&
2062 (old_prot & VM_PROT_WRITE) == 0)
2063 vm_fault_copy_entry(map, map, current, current, NULL);
2066 * When restricting access, update the physical map. Worry
2067 * about copy-on-write here.
2069 if ((old_prot & ~current->protection) != 0) {
2070 #define MASK(entry) (((entry)->eflags & MAP_ENTRY_COW) ? ~VM_PROT_WRITE : \
2072 pmap_protect(map->pmap, current->start,
2074 current->protection & MASK(current));
2077 vm_map_simplify_entry(map, current);
2080 return (KERN_SUCCESS);
2086 * This routine traverses a processes map handling the madvise
2087 * system call. Advisories are classified as either those effecting
2088 * the vm_map_entry structure, or those effecting the underlying
2098 vm_map_entry_t current, entry;
2102 * Some madvise calls directly modify the vm_map_entry, in which case
2103 * we need to use an exclusive lock on the map and we need to perform
2104 * various clipping operations. Otherwise we only need a read-lock
2109 case MADV_SEQUENTIAL:
2116 return (KERN_SUCCESS);
2124 return (KERN_SUCCESS);
2125 vm_map_lock_read(map);
2128 return (KERN_INVALID_ARGUMENT);
2132 * Locate starting entry and clip if necessary.
2134 VM_MAP_RANGE_CHECK(map, start, end);
2136 if (vm_map_lookup_entry(map, start, &entry)) {
2138 vm_map_clip_start(map, entry, start);
2140 entry = entry->next;
2145 * madvise behaviors that are implemented in the vm_map_entry.
2147 * We clip the vm_map_entry so that behavioral changes are
2148 * limited to the specified address range.
2150 for (current = entry;
2151 (current != &map->header) && (current->start < end);
2152 current = current->next
2154 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
2157 vm_map_clip_end(map, current, end);
2161 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_NORMAL);
2163 case MADV_SEQUENTIAL:
2164 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_SEQUENTIAL);
2167 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_RANDOM);
2170 current->eflags |= MAP_ENTRY_NOSYNC;
2173 current->eflags &= ~MAP_ENTRY_NOSYNC;
2176 current->eflags |= MAP_ENTRY_NOCOREDUMP;
2179 current->eflags &= ~MAP_ENTRY_NOCOREDUMP;
2184 vm_map_simplify_entry(map, current);
2188 vm_pindex_t pstart, pend;
2191 * madvise behaviors that are implemented in the underlying
2194 * Since we don't clip the vm_map_entry, we have to clip
2195 * the vm_object pindex and count.
2197 for (current = entry;
2198 (current != &map->header) && (current->start < end);
2199 current = current->next
2201 vm_offset_t useEnd, useStart;
2203 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
2206 pstart = OFF_TO_IDX(current->offset);
2207 pend = pstart + atop(current->end - current->start);
2208 useStart = current->start;
2209 useEnd = current->end;
2211 if (current->start < start) {
2212 pstart += atop(start - current->start);
2215 if (current->end > end) {
2216 pend -= atop(current->end - end);
2224 * Perform the pmap_advise() before clearing
2225 * PGA_REFERENCED in vm_page_advise(). Otherwise, a
2226 * concurrent pmap operation, such as pmap_remove(),
2227 * could clear a reference in the pmap and set
2228 * PGA_REFERENCED on the page before the pmap_advise()
2229 * had completed. Consequently, the page would appear
2230 * referenced based upon an old reference that
2231 * occurred before this pmap_advise() ran.
2233 if (behav == MADV_DONTNEED || behav == MADV_FREE)
2234 pmap_advise(map->pmap, useStart, useEnd,
2237 vm_object_madvise(current->object.vm_object, pstart,
2241 * Pre-populate paging structures in the
2242 * WILLNEED case. For wired entries, the
2243 * paging structures are already populated.
2245 if (behav == MADV_WILLNEED &&
2246 current->wired_count == 0) {
2247 vm_map_pmap_enter(map,
2249 current->protection,
2250 current->object.vm_object,
2252 ptoa(pend - pstart),
2253 MAP_PREFAULT_MADVISE
2257 vm_map_unlock_read(map);
2266 * Sets the inheritance of the specified address
2267 * range in the target map. Inheritance
2268 * affects how the map will be shared with
2269 * child maps at the time of vmspace_fork.
2272 vm_map_inherit(vm_map_t map, vm_offset_t start, vm_offset_t end,
2273 vm_inherit_t new_inheritance)
2275 vm_map_entry_t entry;
2276 vm_map_entry_t temp_entry;
2278 switch (new_inheritance) {
2279 case VM_INHERIT_NONE:
2280 case VM_INHERIT_COPY:
2281 case VM_INHERIT_SHARE:
2282 case VM_INHERIT_ZERO:
2285 return (KERN_INVALID_ARGUMENT);
2288 return (KERN_SUCCESS);
2290 VM_MAP_RANGE_CHECK(map, start, end);
2291 if (vm_map_lookup_entry(map, start, &temp_entry)) {
2293 vm_map_clip_start(map, entry, start);
2295 entry = temp_entry->next;
2296 while ((entry != &map->header) && (entry->start < end)) {
2297 vm_map_clip_end(map, entry, end);
2298 if ((entry->eflags & MAP_ENTRY_GUARD) == 0 ||
2299 new_inheritance != VM_INHERIT_ZERO)
2300 entry->inheritance = new_inheritance;
2301 vm_map_simplify_entry(map, entry);
2302 entry = entry->next;
2305 return (KERN_SUCCESS);
2311 * Implements both kernel and user unwiring.
2314 vm_map_unwire(vm_map_t map, vm_offset_t start, vm_offset_t end,
2317 vm_map_entry_t entry, first_entry, tmp_entry;
2318 vm_offset_t saved_start;
2319 unsigned int last_timestamp;
2321 boolean_t need_wakeup, result, user_unwire;
2324 return (KERN_SUCCESS);
2325 user_unwire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE;
2327 VM_MAP_RANGE_CHECK(map, start, end);
2328 if (!vm_map_lookup_entry(map, start, &first_entry)) {
2329 if (flags & VM_MAP_WIRE_HOLESOK)
2330 first_entry = first_entry->next;
2333 return (KERN_INVALID_ADDRESS);
2336 last_timestamp = map->timestamp;
2337 entry = first_entry;
2338 while (entry != &map->header && entry->start < end) {
2339 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
2341 * We have not yet clipped the entry.
2343 saved_start = (start >= entry->start) ? start :
2345 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2346 if (vm_map_unlock_and_wait(map, 0)) {
2348 * Allow interruption of user unwiring?
2352 if (last_timestamp+1 != map->timestamp) {
2354 * Look again for the entry because the map was
2355 * modified while it was unlocked.
2356 * Specifically, the entry may have been
2357 * clipped, merged, or deleted.
2359 if (!vm_map_lookup_entry(map, saved_start,
2361 if (flags & VM_MAP_WIRE_HOLESOK)
2362 tmp_entry = tmp_entry->next;
2364 if (saved_start == start) {
2366 * First_entry has been deleted.
2369 return (KERN_INVALID_ADDRESS);
2372 rv = KERN_INVALID_ADDRESS;
2376 if (entry == first_entry)
2377 first_entry = tmp_entry;
2382 last_timestamp = map->timestamp;
2385 vm_map_clip_start(map, entry, start);
2386 vm_map_clip_end(map, entry, end);
2388 * Mark the entry in case the map lock is released. (See
2391 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 &&
2392 entry->wiring_thread == NULL,
2393 ("owned map entry %p", entry));
2394 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
2395 entry->wiring_thread = curthread;
2397 * Check the map for holes in the specified region.
2398 * If VM_MAP_WIRE_HOLESOK was specified, skip this check.
2400 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) &&
2401 (entry->end < end && (entry->next == &map->header ||
2402 entry->next->start > entry->end))) {
2404 rv = KERN_INVALID_ADDRESS;
2408 * If system unwiring, require that the entry is system wired.
2411 vm_map_entry_system_wired_count(entry) == 0) {
2413 rv = KERN_INVALID_ARGUMENT;
2416 entry = entry->next;
2420 need_wakeup = FALSE;
2421 if (first_entry == NULL) {
2422 result = vm_map_lookup_entry(map, start, &first_entry);
2423 if (!result && (flags & VM_MAP_WIRE_HOLESOK))
2424 first_entry = first_entry->next;
2426 KASSERT(result, ("vm_map_unwire: lookup failed"));
2428 for (entry = first_entry; entry != &map->header && entry->start < end;
2429 entry = entry->next) {
2431 * If VM_MAP_WIRE_HOLESOK was specified, an empty
2432 * space in the unwired region could have been mapped
2433 * while the map lock was dropped for draining
2434 * MAP_ENTRY_IN_TRANSITION. Moreover, another thread
2435 * could be simultaneously wiring this new mapping
2436 * entry. Detect these cases and skip any entries
2437 * marked as in transition by us.
2439 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 ||
2440 entry->wiring_thread != curthread) {
2441 KASSERT((flags & VM_MAP_WIRE_HOLESOK) != 0,
2442 ("vm_map_unwire: !HOLESOK and new/changed entry"));
2446 if (rv == KERN_SUCCESS && (!user_unwire ||
2447 (entry->eflags & MAP_ENTRY_USER_WIRED))) {
2449 entry->eflags &= ~MAP_ENTRY_USER_WIRED;
2450 if (entry->wired_count == 1)
2451 vm_map_entry_unwire(map, entry);
2453 entry->wired_count--;
2455 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0,
2456 ("vm_map_unwire: in-transition flag missing %p", entry));
2457 KASSERT(entry->wiring_thread == curthread,
2458 ("vm_map_unwire: alien wire %p", entry));
2459 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION;
2460 entry->wiring_thread = NULL;
2461 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
2462 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
2465 vm_map_simplify_entry(map, entry);
2474 * vm_map_wire_entry_failure:
2476 * Handle a wiring failure on the given entry.
2478 * The map should be locked.
2481 vm_map_wire_entry_failure(vm_map_t map, vm_map_entry_t entry,
2482 vm_offset_t failed_addr)
2485 VM_MAP_ASSERT_LOCKED(map);
2486 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 &&
2487 entry->wired_count == 1,
2488 ("vm_map_wire_entry_failure: entry %p isn't being wired", entry));
2489 KASSERT(failed_addr < entry->end,
2490 ("vm_map_wire_entry_failure: entry %p was fully wired", entry));
2493 * If any pages at the start of this entry were successfully wired,
2496 if (failed_addr > entry->start) {
2497 pmap_unwire(map->pmap, entry->start, failed_addr);
2498 vm_object_unwire(entry->object.vm_object, entry->offset,
2499 failed_addr - entry->start, PQ_ACTIVE);
2503 * Assign an out-of-range value to represent the failure to wire this
2506 entry->wired_count = -1;
2512 * Implements both kernel and user wiring.
2515 vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t end,
2518 vm_map_entry_t entry, first_entry, tmp_entry;
2519 vm_offset_t faddr, saved_end, saved_start;
2520 unsigned int last_timestamp;
2522 boolean_t need_wakeup, result, user_wire;
2526 return (KERN_SUCCESS);
2528 if (flags & VM_MAP_WIRE_WRITE)
2529 prot |= VM_PROT_WRITE;
2530 user_wire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE;
2532 VM_MAP_RANGE_CHECK(map, start, end);
2533 if (!vm_map_lookup_entry(map, start, &first_entry)) {
2534 if (flags & VM_MAP_WIRE_HOLESOK)
2535 first_entry = first_entry->next;
2538 return (KERN_INVALID_ADDRESS);
2541 last_timestamp = map->timestamp;
2542 entry = first_entry;
2543 while (entry != &map->header && entry->start < end) {
2544 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
2546 * We have not yet clipped the entry.
2548 saved_start = (start >= entry->start) ? start :
2550 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2551 if (vm_map_unlock_and_wait(map, 0)) {
2553 * Allow interruption of user wiring?
2557 if (last_timestamp + 1 != map->timestamp) {
2559 * Look again for the entry because the map was
2560 * modified while it was unlocked.
2561 * Specifically, the entry may have been
2562 * clipped, merged, or deleted.
2564 if (!vm_map_lookup_entry(map, saved_start,
2566 if (flags & VM_MAP_WIRE_HOLESOK)
2567 tmp_entry = tmp_entry->next;
2569 if (saved_start == start) {
2571 * first_entry has been deleted.
2574 return (KERN_INVALID_ADDRESS);
2577 rv = KERN_INVALID_ADDRESS;
2581 if (entry == first_entry)
2582 first_entry = tmp_entry;
2587 last_timestamp = map->timestamp;
2590 vm_map_clip_start(map, entry, start);
2591 vm_map_clip_end(map, entry, end);
2593 * Mark the entry in case the map lock is released. (See
2596 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 &&
2597 entry->wiring_thread == NULL,
2598 ("owned map entry %p", entry));
2599 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
2600 entry->wiring_thread = curthread;
2601 if ((entry->protection & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0
2602 || (entry->protection & prot) != prot) {
2603 entry->eflags |= MAP_ENTRY_WIRE_SKIPPED;
2604 if ((flags & VM_MAP_WIRE_HOLESOK) == 0) {
2606 rv = KERN_INVALID_ADDRESS;
2611 if (entry->wired_count == 0) {
2612 entry->wired_count++;
2613 saved_start = entry->start;
2614 saved_end = entry->end;
2617 * Release the map lock, relying on the in-transition
2618 * mark. Mark the map busy for fork.
2623 faddr = saved_start;
2626 * Simulate a fault to get the page and enter
2627 * it into the physical map.
2629 if ((rv = vm_fault(map, faddr, VM_PROT_NONE,
2630 VM_FAULT_WIRE)) != KERN_SUCCESS)
2632 } while ((faddr += PAGE_SIZE) < saved_end);
2635 if (last_timestamp + 1 != map->timestamp) {
2637 * Look again for the entry because the map was
2638 * modified while it was unlocked. The entry
2639 * may have been clipped, but NOT merged or
2642 result = vm_map_lookup_entry(map, saved_start,
2644 KASSERT(result, ("vm_map_wire: lookup failed"));
2645 if (entry == first_entry)
2646 first_entry = tmp_entry;
2650 while (entry->end < saved_end) {
2652 * In case of failure, handle entries
2653 * that were not fully wired here;
2654 * fully wired entries are handled
2657 if (rv != KERN_SUCCESS &&
2659 vm_map_wire_entry_failure(map,
2661 entry = entry->next;
2664 last_timestamp = map->timestamp;
2665 if (rv != KERN_SUCCESS) {
2666 vm_map_wire_entry_failure(map, entry, faddr);
2670 } else if (!user_wire ||
2671 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
2672 entry->wired_count++;
2675 * Check the map for holes in the specified region.
2676 * If VM_MAP_WIRE_HOLESOK was specified, skip this check.
2679 if ((flags & VM_MAP_WIRE_HOLESOK) == 0 &&
2680 entry->end < end && (entry->next == &map->header ||
2681 entry->next->start > entry->end)) {
2683 rv = KERN_INVALID_ADDRESS;
2686 entry = entry->next;
2690 need_wakeup = FALSE;
2691 if (first_entry == NULL) {
2692 result = vm_map_lookup_entry(map, start, &first_entry);
2693 if (!result && (flags & VM_MAP_WIRE_HOLESOK))
2694 first_entry = first_entry->next;
2696 KASSERT(result, ("vm_map_wire: lookup failed"));
2698 for (entry = first_entry; entry != &map->header && entry->start < end;
2699 entry = entry->next) {
2701 * If VM_MAP_WIRE_HOLESOK was specified, an empty
2702 * space in the unwired region could have been mapped
2703 * while the map lock was dropped for faulting in the
2704 * pages or draining MAP_ENTRY_IN_TRANSITION.
2705 * Moreover, another thread could be simultaneously
2706 * wiring this new mapping entry. Detect these cases
2707 * and skip any entries marked as in transition not by us.
2709 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 ||
2710 entry->wiring_thread != curthread) {
2711 KASSERT((flags & VM_MAP_WIRE_HOLESOK) != 0,
2712 ("vm_map_wire: !HOLESOK and new/changed entry"));
2716 if ((entry->eflags & MAP_ENTRY_WIRE_SKIPPED) != 0)
2717 goto next_entry_done;
2719 if (rv == KERN_SUCCESS) {
2721 entry->eflags |= MAP_ENTRY_USER_WIRED;
2722 } else if (entry->wired_count == -1) {
2724 * Wiring failed on this entry. Thus, unwiring is
2727 entry->wired_count = 0;
2728 } else if (!user_wire ||
2729 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
2731 * Undo the wiring. Wiring succeeded on this entry
2732 * but failed on a later entry.
2734 if (entry->wired_count == 1)
2735 vm_map_entry_unwire(map, entry);
2737 entry->wired_count--;
2740 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0,
2741 ("vm_map_wire: in-transition flag missing %p", entry));
2742 KASSERT(entry->wiring_thread == curthread,
2743 ("vm_map_wire: alien wire %p", entry));
2744 entry->eflags &= ~(MAP_ENTRY_IN_TRANSITION |
2745 MAP_ENTRY_WIRE_SKIPPED);
2746 entry->wiring_thread = NULL;
2747 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
2748 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
2751 vm_map_simplify_entry(map, entry);
2762 * Push any dirty cached pages in the address range to their pager.
2763 * If syncio is TRUE, dirty pages are written synchronously.
2764 * If invalidate is TRUE, any cached pages are freed as well.
2766 * If the size of the region from start to end is zero, we are
2767 * supposed to flush all modified pages within the region containing
2768 * start. Unfortunately, a region can be split or coalesced with
2769 * neighboring regions, making it difficult to determine what the
2770 * original region was. Therefore, we approximate this requirement by
2771 * flushing the current region containing start.
2773 * Returns an error if any part of the specified range is not mapped.
2781 boolean_t invalidate)
2783 vm_map_entry_t current;
2784 vm_map_entry_t entry;
2787 vm_ooffset_t offset;
2788 unsigned int last_timestamp;
2791 vm_map_lock_read(map);
2792 VM_MAP_RANGE_CHECK(map, start, end);
2793 if (!vm_map_lookup_entry(map, start, &entry)) {
2794 vm_map_unlock_read(map);
2795 return (KERN_INVALID_ADDRESS);
2796 } else if (start == end) {
2797 start = entry->start;
2801 * Make a first pass to check for user-wired memory and holes.
2803 for (current = entry; current != &map->header && current->start < end;
2804 current = current->next) {
2805 if (invalidate && (current->eflags & MAP_ENTRY_USER_WIRED)) {
2806 vm_map_unlock_read(map);
2807 return (KERN_INVALID_ARGUMENT);
2809 if (end > current->end &&
2810 (current->next == &map->header ||
2811 current->end != current->next->start)) {
2812 vm_map_unlock_read(map);
2813 return (KERN_INVALID_ADDRESS);
2818 pmap_remove(map->pmap, start, end);
2822 * Make a second pass, cleaning/uncaching pages from the indicated
2825 for (current = entry; current != &map->header && current->start < end;) {
2826 offset = current->offset + (start - current->start);
2827 size = (end <= current->end ? end : current->end) - start;
2828 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
2830 vm_map_entry_t tentry;
2833 smap = current->object.sub_map;
2834 vm_map_lock_read(smap);
2835 (void) vm_map_lookup_entry(smap, offset, &tentry);
2836 tsize = tentry->end - offset;
2839 object = tentry->object.vm_object;
2840 offset = tentry->offset + (offset - tentry->start);
2841 vm_map_unlock_read(smap);
2843 object = current->object.vm_object;
2845 vm_object_reference(object);
2846 last_timestamp = map->timestamp;
2847 vm_map_unlock_read(map);
2848 if (!vm_object_sync(object, offset, size, syncio, invalidate))
2851 vm_object_deallocate(object);
2852 vm_map_lock_read(map);
2853 if (last_timestamp == map->timestamp ||
2854 !vm_map_lookup_entry(map, start, ¤t))
2855 current = current->next;
2858 vm_map_unlock_read(map);
2859 return (failed ? KERN_FAILURE : KERN_SUCCESS);
2863 * vm_map_entry_unwire: [ internal use only ]
2865 * Make the region specified by this entry pageable.
2867 * The map in question should be locked.
2868 * [This is the reason for this routine's existence.]
2871 vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry)
2874 VM_MAP_ASSERT_LOCKED(map);
2875 KASSERT(entry->wired_count > 0,
2876 ("vm_map_entry_unwire: entry %p isn't wired", entry));
2877 pmap_unwire(map->pmap, entry->start, entry->end);
2878 vm_object_unwire(entry->object.vm_object, entry->offset, entry->end -
2879 entry->start, PQ_ACTIVE);
2880 entry->wired_count = 0;
2884 vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map)
2887 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0)
2888 vm_object_deallocate(entry->object.vm_object);
2889 uma_zfree(system_map ? kmapentzone : mapentzone, entry);
2893 * vm_map_entry_delete: [ internal use only ]
2895 * Deallocate the given entry from the target map.
2898 vm_map_entry_delete(vm_map_t map, vm_map_entry_t entry)
2901 vm_pindex_t offidxstart, offidxend, count, size1;
2904 vm_map_entry_unlink(map, entry);
2905 object = entry->object.vm_object;
2907 if ((entry->eflags & MAP_ENTRY_GUARD) != 0) {
2908 MPASS(entry->cred == NULL);
2909 MPASS((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0);
2910 MPASS(object == NULL);
2911 vm_map_entry_deallocate(entry, map->system_map);
2915 size = entry->end - entry->start;
2918 if (entry->cred != NULL) {
2919 swap_release_by_cred(size, entry->cred);
2920 crfree(entry->cred);
2923 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0 &&
2925 KASSERT(entry->cred == NULL || object->cred == NULL ||
2926 (entry->eflags & MAP_ENTRY_NEEDS_COPY),
2927 ("OVERCOMMIT vm_map_entry_delete: both cred %p", entry));
2928 count = OFF_TO_IDX(size);
2929 offidxstart = OFF_TO_IDX(entry->offset);
2930 offidxend = offidxstart + count;
2931 VM_OBJECT_WLOCK(object);
2932 if (object->ref_count != 1 && ((object->flags & (OBJ_NOSPLIT |
2933 OBJ_ONEMAPPING)) == OBJ_ONEMAPPING ||
2934 object == kernel_object || object == kmem_object)) {
2935 vm_object_collapse(object);
2938 * The option OBJPR_NOTMAPPED can be passed here
2939 * because vm_map_delete() already performed
2940 * pmap_remove() on the only mapping to this range
2943 vm_object_page_remove(object, offidxstart, offidxend,
2945 if (object->type == OBJT_SWAP)
2946 swap_pager_freespace(object, offidxstart,
2948 if (offidxend >= object->size &&
2949 offidxstart < object->size) {
2950 size1 = object->size;
2951 object->size = offidxstart;
2952 if (object->cred != NULL) {
2953 size1 -= object->size;
2954 KASSERT(object->charge >= ptoa(size1),
2955 ("object %p charge < 0", object));
2956 swap_release_by_cred(ptoa(size1),
2958 object->charge -= ptoa(size1);
2962 VM_OBJECT_WUNLOCK(object);
2964 entry->object.vm_object = NULL;
2965 if (map->system_map)
2966 vm_map_entry_deallocate(entry, TRUE);
2968 entry->next = curthread->td_map_def_user;
2969 curthread->td_map_def_user = entry;
2974 * vm_map_delete: [ internal use only ]
2976 * Deallocates the given address range from the target
2980 vm_map_delete(vm_map_t map, vm_offset_t start, vm_offset_t end)
2982 vm_map_entry_t entry;
2983 vm_map_entry_t first_entry;
2985 VM_MAP_ASSERT_LOCKED(map);
2987 return (KERN_SUCCESS);
2990 * Find the start of the region, and clip it
2992 if (!vm_map_lookup_entry(map, start, &first_entry))
2993 entry = first_entry->next;
2995 entry = first_entry;
2996 vm_map_clip_start(map, entry, start);
3000 * Step through all entries in this region
3002 while ((entry != &map->header) && (entry->start < end)) {
3003 vm_map_entry_t next;
3006 * Wait for wiring or unwiring of an entry to complete.
3007 * Also wait for any system wirings to disappear on
3010 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 ||
3011 (vm_map_pmap(map) != kernel_pmap &&
3012 vm_map_entry_system_wired_count(entry) != 0)) {
3013 unsigned int last_timestamp;
3014 vm_offset_t saved_start;
3015 vm_map_entry_t tmp_entry;
3017 saved_start = entry->start;
3018 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
3019 last_timestamp = map->timestamp;
3020 (void) vm_map_unlock_and_wait(map, 0);
3022 if (last_timestamp + 1 != map->timestamp) {
3024 * Look again for the entry because the map was
3025 * modified while it was unlocked.
3026 * Specifically, the entry may have been
3027 * clipped, merged, or deleted.
3029 if (!vm_map_lookup_entry(map, saved_start,
3031 entry = tmp_entry->next;
3034 vm_map_clip_start(map, entry,
3040 vm_map_clip_end(map, entry, end);
3045 * Unwire before removing addresses from the pmap; otherwise,
3046 * unwiring will put the entries back in the pmap.
3048 if (entry->wired_count != 0) {
3049 vm_map_entry_unwire(map, entry);
3052 pmap_remove(map->pmap, entry->start, entry->end);
3055 * Delete the entry only after removing all pmap
3056 * entries pointing to its pages. (Otherwise, its
3057 * page frames may be reallocated, and any modify bits
3058 * will be set in the wrong object!)
3060 vm_map_entry_delete(map, entry);
3063 return (KERN_SUCCESS);
3069 * Remove the given address range from the target map.
3070 * This is the exported form of vm_map_delete.
3073 vm_map_remove(vm_map_t map, vm_offset_t start, vm_offset_t end)
3078 VM_MAP_RANGE_CHECK(map, start, end);
3079 result = vm_map_delete(map, start, end);
3085 * vm_map_check_protection:
3087 * Assert that the target map allows the specified privilege on the
3088 * entire address region given. The entire region must be allocated.
3090 * WARNING! This code does not and should not check whether the
3091 * contents of the region is accessible. For example a smaller file
3092 * might be mapped into a larger address space.
3094 * NOTE! This code is also called by munmap().
3096 * The map must be locked. A read lock is sufficient.
3099 vm_map_check_protection(vm_map_t map, vm_offset_t start, vm_offset_t end,
3100 vm_prot_t protection)
3102 vm_map_entry_t entry;
3103 vm_map_entry_t tmp_entry;
3105 if (!vm_map_lookup_entry(map, start, &tmp_entry))
3109 while (start < end) {
3110 if (entry == &map->header)
3115 if (start < entry->start)
3118 * Check protection associated with entry.
3120 if ((entry->protection & protection) != protection)
3122 /* go to next entry */
3124 entry = entry->next;
3130 * vm_map_copy_entry:
3132 * Copies the contents of the source entry to the destination
3133 * entry. The entries *must* be aligned properly.
3139 vm_map_entry_t src_entry,
3140 vm_map_entry_t dst_entry,
3141 vm_ooffset_t *fork_charge)
3143 vm_object_t src_object;
3144 vm_map_entry_t fake_entry;
3149 VM_MAP_ASSERT_LOCKED(dst_map);
3151 if ((dst_entry->eflags|src_entry->eflags) & MAP_ENTRY_IS_SUB_MAP)
3154 if (src_entry->wired_count == 0 ||
3155 (src_entry->protection & VM_PROT_WRITE) == 0) {
3157 * If the source entry is marked needs_copy, it is already
3160 if ((src_entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0 &&
3161 (src_entry->protection & VM_PROT_WRITE) != 0) {
3162 pmap_protect(src_map->pmap,
3165 src_entry->protection & ~VM_PROT_WRITE);
3169 * Make a copy of the object.
3171 size = src_entry->end - src_entry->start;
3172 if ((src_object = src_entry->object.vm_object) != NULL) {
3173 VM_OBJECT_WLOCK(src_object);
3174 charged = ENTRY_CHARGED(src_entry);
3175 if (src_object->handle == NULL &&
3176 (src_object->type == OBJT_DEFAULT ||
3177 src_object->type == OBJT_SWAP)) {
3178 vm_object_collapse(src_object);
3179 if ((src_object->flags & (OBJ_NOSPLIT |
3180 OBJ_ONEMAPPING)) == OBJ_ONEMAPPING) {
3181 vm_object_split(src_entry);
3183 src_entry->object.vm_object;
3186 vm_object_reference_locked(src_object);
3187 vm_object_clear_flag(src_object, OBJ_ONEMAPPING);
3188 if (src_entry->cred != NULL &&
3189 !(src_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
3190 KASSERT(src_object->cred == NULL,
3191 ("OVERCOMMIT: vm_map_copy_entry: cred %p",
3193 src_object->cred = src_entry->cred;
3194 src_object->charge = size;
3196 VM_OBJECT_WUNLOCK(src_object);
3197 dst_entry->object.vm_object = src_object;
3199 cred = curthread->td_ucred;
3201 dst_entry->cred = cred;
3202 *fork_charge += size;
3203 if (!(src_entry->eflags &
3204 MAP_ENTRY_NEEDS_COPY)) {
3206 src_entry->cred = cred;
3207 *fork_charge += size;
3210 src_entry->eflags |= MAP_ENTRY_COW |
3211 MAP_ENTRY_NEEDS_COPY;
3212 dst_entry->eflags |= MAP_ENTRY_COW |
3213 MAP_ENTRY_NEEDS_COPY;
3214 dst_entry->offset = src_entry->offset;
3215 if (src_entry->eflags & MAP_ENTRY_VN_WRITECNT) {
3217 * MAP_ENTRY_VN_WRITECNT cannot
3218 * indicate write reference from
3219 * src_entry, since the entry is
3220 * marked as needs copy. Allocate a
3221 * fake entry that is used to
3222 * decrement object->un_pager.vnp.writecount
3223 * at the appropriate time. Attach
3224 * fake_entry to the deferred list.
3226 fake_entry = vm_map_entry_create(dst_map);
3227 fake_entry->eflags = MAP_ENTRY_VN_WRITECNT;
3228 src_entry->eflags &= ~MAP_ENTRY_VN_WRITECNT;
3229 vm_object_reference(src_object);
3230 fake_entry->object.vm_object = src_object;
3231 fake_entry->start = src_entry->start;
3232 fake_entry->end = src_entry->end;
3233 fake_entry->next = curthread->td_map_def_user;
3234 curthread->td_map_def_user = fake_entry;
3237 pmap_copy(dst_map->pmap, src_map->pmap,
3238 dst_entry->start, dst_entry->end - dst_entry->start,
3241 dst_entry->object.vm_object = NULL;
3242 dst_entry->offset = 0;
3243 if (src_entry->cred != NULL) {
3244 dst_entry->cred = curthread->td_ucred;
3245 crhold(dst_entry->cred);
3246 *fork_charge += size;
3251 * We don't want to make writeable wired pages copy-on-write.
3252 * Immediately copy these pages into the new map by simulating
3253 * page faults. The new pages are pageable.
3255 vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry,
3261 * vmspace_map_entry_forked:
3262 * Update the newly-forked vmspace each time a map entry is inherited
3263 * or copied. The values for vm_dsize and vm_tsize are approximate
3264 * (and mostly-obsolete ideas in the face of mmap(2) et al.)
3267 vmspace_map_entry_forked(const struct vmspace *vm1, struct vmspace *vm2,
3268 vm_map_entry_t entry)
3270 vm_size_t entrysize;
3273 if ((entry->eflags & MAP_ENTRY_GUARD) != 0)
3275 entrysize = entry->end - entry->start;
3276 vm2->vm_map.size += entrysize;
3277 if (entry->eflags & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP)) {
3278 vm2->vm_ssize += btoc(entrysize);
3279 } else if (entry->start >= (vm_offset_t)vm1->vm_daddr &&
3280 entry->start < (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)) {
3281 newend = MIN(entry->end,
3282 (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize));
3283 vm2->vm_dsize += btoc(newend - entry->start);
3284 } else if (entry->start >= (vm_offset_t)vm1->vm_taddr &&
3285 entry->start < (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)) {
3286 newend = MIN(entry->end,
3287 (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize));
3288 vm2->vm_tsize += btoc(newend - entry->start);
3294 * Create a new process vmspace structure and vm_map
3295 * based on those of an existing process. The new map
3296 * is based on the old map, according to the inheritance
3297 * values on the regions in that map.
3299 * XXX It might be worth coalescing the entries added to the new vmspace.
3301 * The source map must not be locked.
3304 vmspace_fork(struct vmspace *vm1, vm_ooffset_t *fork_charge)
3306 struct vmspace *vm2;
3307 vm_map_t new_map, old_map;
3308 vm_map_entry_t new_entry, old_entry;
3313 old_map = &vm1->vm_map;
3314 /* Copy immutable fields of vm1 to vm2. */
3315 vm2 = vmspace_alloc(old_map->min_offset, old_map->max_offset, NULL);
3318 vm2->vm_taddr = vm1->vm_taddr;
3319 vm2->vm_daddr = vm1->vm_daddr;
3320 vm2->vm_maxsaddr = vm1->vm_maxsaddr;
3321 vm_map_lock(old_map);
3323 vm_map_wait_busy(old_map);
3324 new_map = &vm2->vm_map;
3325 locked = vm_map_trylock(new_map); /* trylock to silence WITNESS */
3326 KASSERT(locked, ("vmspace_fork: lock failed"));
3328 old_entry = old_map->header.next;
3330 while (old_entry != &old_map->header) {
3331 if (old_entry->eflags & MAP_ENTRY_IS_SUB_MAP)
3332 panic("vm_map_fork: encountered a submap");
3334 inh = old_entry->inheritance;
3335 if ((old_entry->eflags & MAP_ENTRY_GUARD) != 0 &&
3336 inh != VM_INHERIT_NONE)
3337 inh = VM_INHERIT_COPY;
3340 case VM_INHERIT_NONE:
3343 case VM_INHERIT_SHARE:
3345 * Clone the entry, creating the shared object if necessary.
3347 object = old_entry->object.vm_object;
3348 if (object == NULL) {
3349 object = vm_object_allocate(OBJT_DEFAULT,
3350 atop(old_entry->end - old_entry->start));
3351 old_entry->object.vm_object = object;
3352 old_entry->offset = 0;
3353 if (old_entry->cred != NULL) {
3354 object->cred = old_entry->cred;
3355 object->charge = old_entry->end -
3357 old_entry->cred = NULL;
3362 * Add the reference before calling vm_object_shadow
3363 * to insure that a shadow object is created.
3365 vm_object_reference(object);
3366 if (old_entry->eflags & MAP_ENTRY_NEEDS_COPY) {
3367 vm_object_shadow(&old_entry->object.vm_object,
3369 old_entry->end - old_entry->start);
3370 old_entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
3371 /* Transfer the second reference too. */
3372 vm_object_reference(
3373 old_entry->object.vm_object);
3376 * As in vm_map_simplify_entry(), the
3377 * vnode lock will not be acquired in
3378 * this call to vm_object_deallocate().
3380 vm_object_deallocate(object);
3381 object = old_entry->object.vm_object;
3383 VM_OBJECT_WLOCK(object);
3384 vm_object_clear_flag(object, OBJ_ONEMAPPING);
3385 if (old_entry->cred != NULL) {
3386 KASSERT(object->cred == NULL, ("vmspace_fork both cred"));
3387 object->cred = old_entry->cred;
3388 object->charge = old_entry->end - old_entry->start;
3389 old_entry->cred = NULL;
3393 * Assert the correct state of the vnode
3394 * v_writecount while the object is locked, to
3395 * not relock it later for the assertion
3398 if (old_entry->eflags & MAP_ENTRY_VN_WRITECNT &&
3399 object->type == OBJT_VNODE) {
3400 KASSERT(((struct vnode *)object->handle)->
3402 ("vmspace_fork: v_writecount %p", object));
3403 KASSERT(object->un_pager.vnp.writemappings > 0,
3404 ("vmspace_fork: vnp.writecount %p",
3407 VM_OBJECT_WUNLOCK(object);
3410 * Clone the entry, referencing the shared object.
3412 new_entry = vm_map_entry_create(new_map);
3413 *new_entry = *old_entry;
3414 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED |
3415 MAP_ENTRY_IN_TRANSITION);
3416 new_entry->wiring_thread = NULL;
3417 new_entry->wired_count = 0;
3418 if (new_entry->eflags & MAP_ENTRY_VN_WRITECNT) {
3419 vnode_pager_update_writecount(object,
3420 new_entry->start, new_entry->end);
3424 * Insert the entry into the new map -- we know we're
3425 * inserting at the end of the new map.
3427 vm_map_entry_link(new_map, new_map->header.prev,
3429 vmspace_map_entry_forked(vm1, vm2, new_entry);
3432 * Update the physical map
3434 pmap_copy(new_map->pmap, old_map->pmap,
3436 (old_entry->end - old_entry->start),
3440 case VM_INHERIT_COPY:
3442 * Clone the entry and link into the map.
3444 new_entry = vm_map_entry_create(new_map);
3445 *new_entry = *old_entry;
3447 * Copied entry is COW over the old object.
3449 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED |
3450 MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_VN_WRITECNT);
3451 new_entry->wiring_thread = NULL;
3452 new_entry->wired_count = 0;
3453 new_entry->object.vm_object = NULL;
3454 new_entry->cred = NULL;
3455 vm_map_entry_link(new_map, new_map->header.prev,
3457 vmspace_map_entry_forked(vm1, vm2, new_entry);
3458 vm_map_copy_entry(old_map, new_map, old_entry,
3459 new_entry, fork_charge);
3462 case VM_INHERIT_ZERO:
3464 * Create a new anonymous mapping entry modelled from
3467 new_entry = vm_map_entry_create(new_map);
3468 memset(new_entry, 0, sizeof(*new_entry));
3470 new_entry->start = old_entry->start;
3471 new_entry->end = old_entry->end;
3472 new_entry->eflags = old_entry->eflags &
3473 ~(MAP_ENTRY_USER_WIRED | MAP_ENTRY_IN_TRANSITION |
3474 MAP_ENTRY_VN_WRITECNT);
3475 new_entry->protection = old_entry->protection;
3476 new_entry->max_protection = old_entry->max_protection;
3477 new_entry->inheritance = VM_INHERIT_ZERO;
3479 vm_map_entry_link(new_map, new_map->header.prev,
3481 vmspace_map_entry_forked(vm1, vm2, new_entry);
3483 new_entry->cred = curthread->td_ucred;
3484 crhold(new_entry->cred);
3485 *fork_charge += (new_entry->end - new_entry->start);
3489 old_entry = old_entry->next;
3492 * Use inlined vm_map_unlock() to postpone handling the deferred
3493 * map entries, which cannot be done until both old_map and
3494 * new_map locks are released.
3496 sx_xunlock(&old_map->lock);
3497 sx_xunlock(&new_map->lock);
3498 vm_map_process_deferred();
3504 * Create a process's stack for exec_new_vmspace(). This function is never
3505 * asked to wire the newly created stack.
3508 vm_map_stack(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
3509 vm_prot_t prot, vm_prot_t max, int cow)
3511 vm_size_t growsize, init_ssize;
3515 MPASS((map->flags & MAP_WIREFUTURE) == 0);
3516 growsize = sgrowsiz;
3517 init_ssize = (max_ssize < growsize) ? max_ssize : growsize;
3520 vmemlim = lim_cur(curproc, RLIMIT_VMEM);
3521 PROC_UNLOCK(curproc);
3522 /* If we would blow our VMEM resource limit, no go */
3523 if (map->size + init_ssize > vmemlim) {
3527 rv = vm_map_stack_locked(map, addrbos, max_ssize, growsize, prot,
3534 static int stack_guard_page = 1;
3535 SYSCTL_INT(_security_bsd, OID_AUTO, stack_guard_page, CTLFLAG_RWTUN,
3536 &stack_guard_page, 0,
3537 "Specifies the number of guard pages for a stack that grows");
3540 vm_map_stack_locked(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
3541 vm_size_t growsize, vm_prot_t prot, vm_prot_t max, int cow)
3543 vm_map_entry_t new_entry, prev_entry;
3544 vm_offset_t bot, gap_bot, gap_top, top;
3545 vm_size_t init_ssize, sgp;
3549 * The stack orientation is piggybacked with the cow argument.
3550 * Extract it into orient and mask the cow argument so that we
3551 * don't pass it around further.
3553 orient = cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP);
3554 KASSERT(orient != 0, ("No stack grow direction"));
3555 KASSERT(orient != (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP),
3558 sgp = (vm_size_t)stack_guard_page * PAGE_SIZE;
3559 if (addrbos < vm_map_min(map) ||
3560 addrbos > vm_map_max(map) ||
3561 addrbos + max_ssize < addrbos ||
3563 return (KERN_NO_SPACE);
3565 init_ssize = growsize;
3566 if (max_ssize < init_ssize + sgp)
3567 init_ssize = max_ssize - sgp;
3569 /* If addr is already mapped, no go */
3570 if (vm_map_lookup_entry(map, addrbos, &prev_entry))
3571 return (KERN_NO_SPACE);
3574 * If we can't accomodate max_ssize in the current mapping, no go.
3576 if ((prev_entry->next != &map->header) &&
3577 (prev_entry->next->start < addrbos + max_ssize))
3578 return (KERN_NO_SPACE);
3581 * We initially map a stack of only init_ssize. We will grow as
3582 * needed later. Depending on the orientation of the stack (i.e.
3583 * the grow direction) we either map at the top of the range, the
3584 * bottom of the range or in the middle.
3586 * Note: we would normally expect prot and max to be VM_PROT_ALL,
3587 * and cow to be 0. Possibly we should eliminate these as input
3588 * parameters, and just pass these values here in the insert call.
3590 if (orient == MAP_STACK_GROWS_DOWN) {
3591 bot = addrbos + max_ssize - init_ssize;
3592 top = bot + init_ssize;
3595 } else /* if (orient == MAP_STACK_GROWS_UP) */ {
3597 top = bot + init_ssize;
3599 gap_top = addrbos + max_ssize;
3601 rv = vm_map_insert(map, NULL, 0, bot, top, prot, max, cow);
3602 if (rv != KERN_SUCCESS)
3604 new_entry = prev_entry->next;
3605 KASSERT(new_entry->end == top || new_entry->start == bot,
3606 ("Bad entry start/end for new stack entry"));
3607 KASSERT((orient & MAP_STACK_GROWS_DOWN) == 0 ||
3608 (new_entry->eflags & MAP_ENTRY_GROWS_DOWN) != 0,
3609 ("new entry lacks MAP_ENTRY_GROWS_DOWN"));
3610 KASSERT((orient & MAP_STACK_GROWS_UP) == 0 ||
3611 (new_entry->eflags & MAP_ENTRY_GROWS_UP) != 0,
3612 ("new entry lacks MAP_ENTRY_GROWS_UP"));
3613 rv = vm_map_insert(map, NULL, 0, gap_bot, gap_top, VM_PROT_NONE,
3614 VM_PROT_NONE, MAP_CREATE_GUARD | (orient == MAP_STACK_GROWS_DOWN ?
3615 MAP_CREATE_STACK_GAP_DN : MAP_CREATE_STACK_GAP_UP));
3616 if (rv != KERN_SUCCESS)
3617 (void)vm_map_delete(map, bot, top);
3622 * Attempts to grow a vm stack entry. Returns KERN_SUCCESS if we
3623 * successfully grow the stack.
3626 vm_map_growstack(vm_map_t map, vm_offset_t addr, vm_map_entry_t gap_entry)
3628 vm_map_entry_t stack_entry;
3632 vm_offset_t gap_end, gap_start, grow_start;
3633 size_t grow_amount, guard, max_grow;
3634 rlim_t lmemlim, stacklim, vmemlim;
3636 bool gap_deleted, grow_down, is_procstack;
3648 * Disallow stack growth when the access is performed by a
3649 * debugger or AIO daemon. The reason is that the wrong
3650 * resource limits are applied.
3652 if (map != &p->p_vmspace->vm_map || p->p_textvp == NULL)
3653 return (KERN_FAILURE);
3655 MPASS(!map->system_map);
3657 guard = stack_guard_page * PAGE_SIZE;
3659 lmemlim = lim_cur(p, RLIMIT_MEMLOCK);
3660 stacklim = lim_cur(p, RLIMIT_STACK);
3661 vmemlim = lim_cur(p, RLIMIT_VMEM);
3664 /* If addr is not in a hole for a stack grow area, no need to grow. */
3665 if (gap_entry == NULL && !vm_map_lookup_entry(map, addr, &gap_entry))
3666 return (KERN_FAILURE);
3667 if ((gap_entry->eflags & MAP_ENTRY_GUARD) == 0)
3668 return (KERN_SUCCESS);
3669 if ((gap_entry->eflags & MAP_ENTRY_STACK_GAP_DN) != 0) {
3670 stack_entry = gap_entry->next;
3671 if ((stack_entry->eflags & MAP_ENTRY_GROWS_DOWN) == 0 ||
3672 stack_entry->start != gap_entry->end)
3673 return (KERN_FAILURE);
3674 grow_amount = round_page(stack_entry->start - addr);
3676 } else if ((gap_entry->eflags & MAP_ENTRY_STACK_GAP_UP) != 0) {
3677 stack_entry = gap_entry->prev;
3678 if ((stack_entry->eflags & MAP_ENTRY_GROWS_UP) == 0 ||
3679 stack_entry->end != gap_entry->start)
3680 return (KERN_FAILURE);
3681 grow_amount = round_page(addr + 1 - stack_entry->end);
3684 return (KERN_FAILURE);
3686 max_grow = gap_entry->end - gap_entry->start;
3687 if (guard > max_grow)
3688 return (KERN_NO_SPACE);
3690 if (grow_amount > max_grow)
3691 return (KERN_NO_SPACE);
3694 * If this is the main process stack, see if we're over the stack
3697 is_procstack = addr >= (vm_offset_t)vm->vm_maxsaddr &&
3698 addr < (vm_offset_t)p->p_sysent->sv_usrstack;
3699 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim))
3700 return (KERN_NO_SPACE);
3705 if (is_procstack && racct_set(p, RACCT_STACK,
3706 ctob(vm->vm_ssize) + grow_amount)) {
3708 return (KERN_NO_SPACE);
3714 grow_amount = roundup(grow_amount, sgrowsiz);
3715 if (grow_amount > max_grow)
3716 grow_amount = max_grow;
3717 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) {
3718 grow_amount = trunc_page((vm_size_t)stacklim) -
3724 limit = racct_get_available(p, RACCT_STACK);
3726 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > limit))
3727 grow_amount = limit - ctob(vm->vm_ssize);
3730 if (!old_mlock && (map->flags & MAP_WIREFUTURE) != 0) {
3731 if (ptoa(pmap_wired_count(map->pmap)) + grow_amount > lmemlim) {
3738 if (racct_set(p, RACCT_MEMLOCK,
3739 ptoa(pmap_wired_count(map->pmap)) + grow_amount)) {
3749 /* If we would blow our VMEM resource limit, no go */
3750 if (map->size + grow_amount > vmemlim) {
3757 if (racct_set(p, RACCT_VMEM, map->size + grow_amount)) {
3766 if (vm_map_lock_upgrade(map)) {
3768 vm_map_lock_read(map);
3773 grow_start = gap_entry->end - grow_amount;
3774 if (gap_entry->start + grow_amount == gap_entry->end) {
3775 gap_start = gap_entry->start;
3776 gap_end = gap_entry->end;
3777 vm_map_entry_delete(map, gap_entry);
3780 MPASS(gap_entry->start < gap_entry->end - grow_amount);
3781 gap_entry->end -= grow_amount;
3782 vm_map_entry_resize_free(map, gap_entry);
3783 gap_deleted = false;
3785 rv = vm_map_insert(map, NULL, 0, grow_start,
3786 grow_start + grow_amount,
3787 stack_entry->protection, stack_entry->max_protection,
3788 MAP_STACK_GROWS_DOWN);
3789 if (rv != KERN_SUCCESS) {
3791 rv1 = vm_map_insert(map, NULL, 0, gap_start,
3792 gap_end, VM_PROT_NONE, VM_PROT_NONE,
3793 MAP_CREATE_GUARD | MAP_CREATE_STACK_GAP_DN);
3794 MPASS(rv1 == KERN_SUCCESS);
3796 gap_entry->end += grow_amount;
3797 vm_map_entry_resize_free(map, gap_entry);
3801 grow_start = stack_entry->end;
3802 cred = stack_entry->cred;
3803 if (cred == NULL && stack_entry->object.vm_object != NULL)
3804 cred = stack_entry->object.vm_object->cred;
3805 if (cred != NULL && !swap_reserve_by_cred(grow_amount, cred))
3807 /* Grow the underlying object if applicable. */
3808 else if (stack_entry->object.vm_object == NULL ||
3809 vm_object_coalesce(stack_entry->object.vm_object,
3810 stack_entry->offset,
3811 (vm_size_t)(stack_entry->end - stack_entry->start),
3812 (vm_size_t)grow_amount, cred != NULL)) {
3813 if (gap_entry->start + grow_amount == gap_entry->end)
3814 vm_map_entry_delete(map, gap_entry);
3816 gap_entry->start += grow_amount;
3817 stack_entry->end += grow_amount;
3818 map->size += grow_amount;
3819 vm_map_entry_resize_free(map, stack_entry);
3824 if (rv == KERN_SUCCESS && is_procstack)
3825 vm->vm_ssize += btoc(grow_amount);
3828 * Heed the MAP_WIREFUTURE flag if it was set for this process.
3830 if (rv == KERN_SUCCESS && (map->flags & MAP_WIREFUTURE) != 0) {
3832 vm_map_wire(map, grow_start, grow_start + grow_amount,
3833 (p->p_flag & P_SYSTEM)
3834 ? VM_MAP_WIRE_SYSTEM|VM_MAP_WIRE_NOHOLES
3835 : VM_MAP_WIRE_USER|VM_MAP_WIRE_NOHOLES);
3836 vm_map_lock_read(map);
3838 vm_map_lock_downgrade(map);
3842 if (racct_enable && rv != KERN_SUCCESS) {
3844 error = racct_set(p, RACCT_VMEM, map->size);
3845 KASSERT(error == 0, ("decreasing RACCT_VMEM failed"));
3847 error = racct_set(p, RACCT_MEMLOCK,
3848 ptoa(pmap_wired_count(map->pmap)));
3849 KASSERT(error == 0, ("decreasing RACCT_MEMLOCK failed"));
3851 error = racct_set(p, RACCT_STACK, ctob(vm->vm_ssize));
3852 KASSERT(error == 0, ("decreasing RACCT_STACK failed"));
3861 * Unshare the specified VM space for exec. If other processes are
3862 * mapped to it, then create a new one. The new vmspace is null.
3865 vmspace_exec(struct proc *p, vm_offset_t minuser, vm_offset_t maxuser)
3867 struct vmspace *oldvmspace = p->p_vmspace;
3868 struct vmspace *newvmspace;
3870 KASSERT((curthread->td_pflags & TDP_EXECVMSPC) == 0,
3871 ("vmspace_exec recursed"));
3872 newvmspace = vmspace_alloc(minuser, maxuser, NULL);
3873 if (newvmspace == NULL)
3875 newvmspace->vm_swrss = oldvmspace->vm_swrss;
3877 * This code is written like this for prototype purposes. The
3878 * goal is to avoid running down the vmspace here, but let the
3879 * other process's that are still using the vmspace to finally
3880 * run it down. Even though there is little or no chance of blocking
3881 * here, it is a good idea to keep this form for future mods.
3883 PROC_VMSPACE_LOCK(p);
3884 p->p_vmspace = newvmspace;
3885 PROC_VMSPACE_UNLOCK(p);
3886 if (p == curthread->td_proc)
3887 pmap_activate(curthread);
3888 curthread->td_pflags |= TDP_EXECVMSPC;
3893 * Unshare the specified VM space for forcing COW. This
3894 * is called by rfork, for the (RFMEM|RFPROC) == 0 case.
3897 vmspace_unshare(struct proc *p)
3899 struct vmspace *oldvmspace = p->p_vmspace;
3900 struct vmspace *newvmspace;
3901 vm_ooffset_t fork_charge;
3903 if (oldvmspace->vm_refcnt == 1)
3906 newvmspace = vmspace_fork(oldvmspace, &fork_charge);
3907 if (newvmspace == NULL)
3909 if (!swap_reserve_by_cred(fork_charge, p->p_ucred)) {
3910 vmspace_free(newvmspace);
3913 PROC_VMSPACE_LOCK(p);
3914 p->p_vmspace = newvmspace;
3915 PROC_VMSPACE_UNLOCK(p);
3916 if (p == curthread->td_proc)
3917 pmap_activate(curthread);
3918 vmspace_free(oldvmspace);
3925 * Finds the VM object, offset, and
3926 * protection for a given virtual address in the
3927 * specified map, assuming a page fault of the
3930 * Leaves the map in question locked for read; return
3931 * values are guaranteed until a vm_map_lookup_done
3932 * call is performed. Note that the map argument
3933 * is in/out; the returned map must be used in
3934 * the call to vm_map_lookup_done.
3936 * A handle (out_entry) is returned for use in
3937 * vm_map_lookup_done, to make that fast.
3939 * If a lookup is requested with "write protection"
3940 * specified, the map may be changed to perform virtual
3941 * copying operations, although the data referenced will
3945 vm_map_lookup(vm_map_t *var_map, /* IN/OUT */
3947 vm_prot_t fault_typea,
3948 vm_map_entry_t *out_entry, /* OUT */
3949 vm_object_t *object, /* OUT */
3950 vm_pindex_t *pindex, /* OUT */
3951 vm_prot_t *out_prot, /* OUT */
3952 boolean_t *wired) /* OUT */
3954 vm_map_entry_t entry;
3955 vm_map_t map = *var_map;
3957 vm_prot_t fault_type = fault_typea;
3958 vm_object_t eobject;
3964 vm_map_lock_read(map);
3968 * Lookup the faulting address.
3970 if (!vm_map_lookup_entry(map, vaddr, out_entry)) {
3971 vm_map_unlock_read(map);
3972 return (KERN_INVALID_ADDRESS);
3980 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
3981 vm_map_t old_map = map;
3983 *var_map = map = entry->object.sub_map;
3984 vm_map_unlock_read(old_map);
3989 * Check whether this task is allowed to have this page.
3991 prot = entry->protection;
3992 if ((fault_typea & VM_PROT_FAULT_LOOKUP) != 0) {
3993 fault_typea &= ~VM_PROT_FAULT_LOOKUP;
3994 if (prot == VM_PROT_NONE && map != kernel_map &&
3995 (entry->eflags & MAP_ENTRY_GUARD) != 0 &&
3996 (entry->eflags & (MAP_ENTRY_STACK_GAP_DN |
3997 MAP_ENTRY_STACK_GAP_UP)) != 0 &&
3998 vm_map_growstack(map, vaddr, entry) == KERN_SUCCESS)
3999 goto RetryLookupLocked;
4001 fault_type &= VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE;
4002 if ((fault_type & prot) != fault_type || prot == VM_PROT_NONE) {
4003 vm_map_unlock_read(map);
4004 return (KERN_PROTECTION_FAILURE);
4006 KASSERT((prot & VM_PROT_WRITE) == 0 || (entry->eflags &
4007 (MAP_ENTRY_USER_WIRED | MAP_ENTRY_NEEDS_COPY)) !=
4008 (MAP_ENTRY_USER_WIRED | MAP_ENTRY_NEEDS_COPY),
4009 ("entry %p flags %x", entry, entry->eflags));
4010 if ((fault_typea & VM_PROT_COPY) != 0 &&
4011 (entry->max_protection & VM_PROT_WRITE) == 0 &&
4012 (entry->eflags & MAP_ENTRY_COW) == 0) {
4013 vm_map_unlock_read(map);
4014 return (KERN_PROTECTION_FAILURE);
4018 * If this page is not pageable, we have to get it for all possible
4021 *wired = (entry->wired_count != 0);
4023 fault_type = entry->protection;
4024 size = entry->end - entry->start;
4026 * If the entry was copy-on-write, we either ...
4028 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
4030 * If we want to write the page, we may as well handle that
4031 * now since we've got the map locked.
4033 * If we don't need to write the page, we just demote the
4034 * permissions allowed.
4036 if ((fault_type & VM_PROT_WRITE) != 0 ||
4037 (fault_typea & VM_PROT_COPY) != 0) {
4039 * Make a new object, and place it in the object
4040 * chain. Note that no new references have appeared
4041 * -- one just moved from the map to the new
4044 if (vm_map_lock_upgrade(map))
4047 if (entry->cred == NULL) {
4049 * The debugger owner is charged for
4052 cred = curthread->td_ucred;
4054 if (!swap_reserve_by_cred(size, cred)) {
4057 return (KERN_RESOURCE_SHORTAGE);
4061 vm_object_shadow(&entry->object.vm_object,
4062 &entry->offset, size);
4063 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
4064 eobject = entry->object.vm_object;
4065 if (eobject->cred != NULL) {
4067 * The object was not shadowed.
4069 swap_release_by_cred(size, entry->cred);
4070 crfree(entry->cred);
4072 } else if (entry->cred != NULL) {
4073 VM_OBJECT_WLOCK(eobject);
4074 eobject->cred = entry->cred;
4075 eobject->charge = size;
4076 VM_OBJECT_WUNLOCK(eobject);
4080 vm_map_lock_downgrade(map);
4083 * We're attempting to read a copy-on-write page --
4084 * don't allow writes.
4086 prot &= ~VM_PROT_WRITE;
4091 * Create an object if necessary.
4093 if (entry->object.vm_object == NULL &&
4095 if (vm_map_lock_upgrade(map))
4097 entry->object.vm_object = vm_object_allocate(OBJT_DEFAULT,
4100 if (entry->cred != NULL) {
4101 VM_OBJECT_WLOCK(entry->object.vm_object);
4102 entry->object.vm_object->cred = entry->cred;
4103 entry->object.vm_object->charge = size;
4104 VM_OBJECT_WUNLOCK(entry->object.vm_object);
4107 vm_map_lock_downgrade(map);
4111 * Return the object/offset from this entry. If the entry was
4112 * copy-on-write or empty, it has been fixed up.
4114 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
4115 *object = entry->object.vm_object;
4118 return (KERN_SUCCESS);
4122 * vm_map_lookup_locked:
4124 * Lookup the faulting address. A version of vm_map_lookup that returns
4125 * KERN_FAILURE instead of blocking on map lock or memory allocation.
4128 vm_map_lookup_locked(vm_map_t *var_map, /* IN/OUT */
4130 vm_prot_t fault_typea,
4131 vm_map_entry_t *out_entry, /* OUT */
4132 vm_object_t *object, /* OUT */
4133 vm_pindex_t *pindex, /* OUT */
4134 vm_prot_t *out_prot, /* OUT */
4135 boolean_t *wired) /* OUT */
4137 vm_map_entry_t entry;
4138 vm_map_t map = *var_map;
4140 vm_prot_t fault_type = fault_typea;
4143 * Lookup the faulting address.
4145 if (!vm_map_lookup_entry(map, vaddr, out_entry))
4146 return (KERN_INVALID_ADDRESS);
4151 * Fail if the entry refers to a submap.
4153 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP)
4154 return (KERN_FAILURE);
4157 * Check whether this task is allowed to have this page.
4159 prot = entry->protection;
4160 fault_type &= VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE;
4161 if ((fault_type & prot) != fault_type)
4162 return (KERN_PROTECTION_FAILURE);
4165 * If this page is not pageable, we have to get it for all possible
4168 *wired = (entry->wired_count != 0);
4170 fault_type = entry->protection;
4172 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
4174 * Fail if the entry was copy-on-write for a write fault.
4176 if (fault_type & VM_PROT_WRITE)
4177 return (KERN_FAILURE);
4179 * We're attempting to read a copy-on-write page --
4180 * don't allow writes.
4182 prot &= ~VM_PROT_WRITE;
4186 * Fail if an object should be created.
4188 if (entry->object.vm_object == NULL && !map->system_map)
4189 return (KERN_FAILURE);
4192 * Return the object/offset from this entry. If the entry was
4193 * copy-on-write or empty, it has been fixed up.
4195 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
4196 *object = entry->object.vm_object;
4199 return (KERN_SUCCESS);
4203 * vm_map_lookup_done:
4205 * Releases locks acquired by a vm_map_lookup
4206 * (according to the handle returned by that lookup).
4209 vm_map_lookup_done(vm_map_t map, vm_map_entry_t entry)
4212 * Unlock the main-level map
4214 vm_map_unlock_read(map);
4217 #include "opt_ddb.h"
4219 #include <sys/kernel.h>
4221 #include <ddb/ddb.h>
4224 vm_map_print(vm_map_t map)
4226 vm_map_entry_t entry;
4228 db_iprintf("Task map %p: pmap=%p, nentries=%d, version=%u\n",
4230 (void *)map->pmap, map->nentries, map->timestamp);
4233 for (entry = map->header.next; entry != &map->header;
4234 entry = entry->next) {
4235 db_iprintf("map entry %p: start=%p, end=%p, eflags=%#x, \n",
4236 (void *)entry, (void *)entry->start, (void *)entry->end,
4239 static char *inheritance_name[4] =
4240 {"share", "copy", "none", "donate_copy"};
4242 db_iprintf(" prot=%x/%x/%s",
4244 entry->max_protection,
4245 inheritance_name[(int)(unsigned char)entry->inheritance]);
4246 if (entry->wired_count != 0)
4247 db_printf(", wired");
4249 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
4250 db_printf(", share=%p, offset=0x%jx\n",
4251 (void *)entry->object.sub_map,
4252 (uintmax_t)entry->offset);
4253 if ((entry->prev == &map->header) ||
4254 (entry->prev->object.sub_map !=
4255 entry->object.sub_map)) {
4257 vm_map_print((vm_map_t)entry->object.sub_map);
4261 if (entry->cred != NULL)
4262 db_printf(", ruid %d", entry->cred->cr_ruid);
4263 db_printf(", object=%p, offset=0x%jx",
4264 (void *)entry->object.vm_object,
4265 (uintmax_t)entry->offset);
4266 if (entry->object.vm_object && entry->object.vm_object->cred)
4267 db_printf(", obj ruid %d charge %jx",
4268 entry->object.vm_object->cred->cr_ruid,
4269 (uintmax_t)entry->object.vm_object->charge);
4270 if (entry->eflags & MAP_ENTRY_COW)
4271 db_printf(", copy (%s)",
4272 (entry->eflags & MAP_ENTRY_NEEDS_COPY) ? "needed" : "done");
4275 if ((entry->prev == &map->header) ||
4276 (entry->prev->object.vm_object !=
4277 entry->object.vm_object)) {
4279 vm_object_print((db_expr_t)(intptr_t)
4280 entry->object.vm_object,
4289 DB_SHOW_COMMAND(map, map)
4293 db_printf("usage: show map <addr>\n");
4296 vm_map_print((vm_map_t)addr);
4299 DB_SHOW_COMMAND(procvm, procvm)
4304 p = db_lookup_proc(addr);
4309 db_printf("p = %p, vmspace = %p, map = %p, pmap = %p\n",
4310 (void *)p, (void *)p->p_vmspace, (void *)&p->p_vmspace->vm_map,
4311 (void *)vmspace_pmap(p->p_vmspace));
4313 vm_map_print((vm_map_t)&p->p_vmspace->vm_map);