2 * Copyright (c) 1991, 1993
3 * The Regents of the University of California. All rights reserved.
5 * This code is derived from software contributed to Berkeley by
6 * The Mach Operating System project at Carnegie-Mellon University.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 4. Neither the name of the University nor the names of its contributors
17 * may be used to endorse or promote products derived from this software
18 * without specific prior written permission.
20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * from: @(#)vm_map.c 8.3 (Berkeley) 1/12/94
35 * Copyright (c) 1987, 1990 Carnegie-Mellon University.
36 * All rights reserved.
38 * Authors: Avadis Tevanian, Jr., Michael Wayne Young
40 * Permission to use, copy, modify and distribute this software and
41 * its documentation is hereby granted, provided that both the copyright
42 * notice and this permission notice appear in all copies of the
43 * software, derivative works or modified versions, and any portions
44 * thereof, and that both notices appear in supporting documentation.
46 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
47 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
48 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
50 * Carnegie Mellon requests users of this software to return to
52 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
53 * School of Computer Science
54 * Carnegie Mellon University
55 * Pittsburgh PA 15213-3890
57 * any improvements or extensions that they make and grant Carnegie the
58 * rights to redistribute these changes.
62 * Virtual memory mapping module.
65 #include <sys/cdefs.h>
66 __FBSDID("$FreeBSD$");
68 #include <sys/param.h>
69 #include <sys/systm.h>
70 #include <sys/kernel.h>
73 #include <sys/mutex.h>
75 #include <sys/vmmeter.h>
77 #include <sys/vnode.h>
78 #include <sys/racct.h>
79 #include <sys/resourcevar.h>
80 #include <sys/rwlock.h>
82 #include <sys/sysctl.h>
83 #include <sys/sysent.h>
87 #include <vm/vm_param.h>
89 #include <vm/vm_map.h>
90 #include <vm/vm_page.h>
91 #include <vm/vm_object.h>
92 #include <vm/vm_pager.h>
93 #include <vm/vm_kern.h>
94 #include <vm/vm_extern.h>
95 #include <vm/vnode_pager.h>
96 #include <vm/swap_pager.h>
100 * Virtual memory maps provide for the mapping, protection,
101 * and sharing of virtual memory objects. In addition,
102 * this module provides for an efficient virtual copy of
103 * memory from one map to another.
105 * Synchronization is required prior to most operations.
107 * Maps consist of an ordered doubly-linked list of simple
108 * entries; a self-adjusting binary search tree of these
109 * entries is used to speed up lookups.
111 * Since portions of maps are specified by start/end addresses,
112 * which may not align with existing map entries, all
113 * routines merely "clip" entries to these start/end values.
114 * [That is, an entry is split into two, bordering at a
115 * start or end value.] Note that these clippings may not
116 * always be necessary (as the two resulting entries are then
117 * not changed); however, the clipping is done for convenience.
119 * As mentioned above, virtual copy operations are performed
120 * by copying VM object references from one map to
121 * another, and then marking both regions as copy-on-write.
124 static struct mtx map_sleep_mtx;
125 static uma_zone_t mapentzone;
126 static uma_zone_t kmapentzone;
127 static uma_zone_t mapzone;
128 static uma_zone_t vmspace_zone;
129 static int vmspace_zinit(void *mem, int size, int flags);
130 static int vm_map_zinit(void *mem, int ize, int flags);
131 static void _vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min,
133 static void vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map);
134 static void vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry);
135 static void vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry);
136 static void vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot,
137 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags);
139 static void vm_map_zdtor(void *mem, int size, void *arg);
140 static void vmspace_zdtor(void *mem, int size, void *arg);
142 static int vm_map_stack_locked(vm_map_t map, vm_offset_t addrbos,
143 vm_size_t max_ssize, vm_size_t growsize, vm_prot_t prot, vm_prot_t max,
145 static void vm_map_wire_entry_failure(vm_map_t map, vm_map_entry_t entry,
146 vm_offset_t failed_addr);
148 #define ENTRY_CHARGED(e) ((e)->cred != NULL || \
149 ((e)->object.vm_object != NULL && (e)->object.vm_object->cred != NULL && \
150 !((e)->eflags & MAP_ENTRY_NEEDS_COPY)))
153 * PROC_VMSPACE_{UN,}LOCK() can be a noop as long as vmspaces are type
156 #define PROC_VMSPACE_LOCK(p) do { } while (0)
157 #define PROC_VMSPACE_UNLOCK(p) do { } while (0)
160 * VM_MAP_RANGE_CHECK: [ internal use only ]
162 * Asserts that the starting and ending region
163 * addresses fall within the valid range of the map.
165 #define VM_MAP_RANGE_CHECK(map, start, end) \
167 if (start < vm_map_min(map)) \
168 start = vm_map_min(map); \
169 if (end > vm_map_max(map)) \
170 end = vm_map_max(map); \
178 * Initialize the vm_map module. Must be called before
179 * any other vm_map routines.
181 * Map and entry structures are allocated from the general
182 * purpose memory pool with some exceptions:
184 * - The kernel map and kmem submap are allocated statically.
185 * - Kernel map entries are allocated out of a static pool.
187 * These restrictions are necessary since malloc() uses the
188 * maps and requires map entries.
194 mtx_init(&map_sleep_mtx, "vm map sleep mutex", NULL, MTX_DEF);
195 mapzone = uma_zcreate("MAP", sizeof(struct vm_map), NULL,
201 vm_map_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
202 uma_prealloc(mapzone, MAX_KMAP);
203 kmapentzone = uma_zcreate("KMAP ENTRY", sizeof(struct vm_map_entry),
204 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR,
205 UMA_ZONE_MTXCLASS | UMA_ZONE_VM);
206 mapentzone = uma_zcreate("MAP ENTRY", sizeof(struct vm_map_entry),
207 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
208 vmspace_zone = uma_zcreate("VMSPACE", sizeof(struct vmspace), NULL,
214 vmspace_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
218 vmspace_zinit(void *mem, int size, int flags)
222 vm = (struct vmspace *)mem;
224 vm->vm_map.pmap = NULL;
225 (void)vm_map_zinit(&vm->vm_map, sizeof(vm->vm_map), flags);
226 PMAP_LOCK_INIT(vmspace_pmap(vm));
231 vm_map_zinit(void *mem, int size, int flags)
236 memset(map, 0, sizeof(*map));
237 mtx_init(&map->system_mtx, "vm map (system)", NULL, MTX_DEF | MTX_DUPOK);
238 sx_init(&map->lock, "vm map (user)");
244 vmspace_zdtor(void *mem, int size, void *arg)
248 vm = (struct vmspace *)mem;
250 vm_map_zdtor(&vm->vm_map, sizeof(vm->vm_map), arg);
253 vm_map_zdtor(void *mem, int size, void *arg)
258 KASSERT(map->nentries == 0,
259 ("map %p nentries == %d on free.",
260 map, map->nentries));
261 KASSERT(map->size == 0,
262 ("map %p size == %lu on free.",
263 map, (unsigned long)map->size));
265 #endif /* INVARIANTS */
268 * Allocate a vmspace structure, including a vm_map and pmap,
269 * and initialize those structures. The refcnt is set to 1.
271 * If 'pinit' is NULL then the embedded pmap is initialized via pmap_pinit().
274 vmspace_alloc(vm_offset_t min, vm_offset_t max, pmap_pinit_t pinit)
278 vm = uma_zalloc(vmspace_zone, M_WAITOK);
280 KASSERT(vm->vm_map.pmap == NULL, ("vm_map.pmap must be NULL"));
285 if (!pinit(vmspace_pmap(vm))) {
286 uma_zfree(vmspace_zone, vm);
289 CTR1(KTR_VM, "vmspace_alloc: %p", vm);
290 _vm_map_init(&vm->vm_map, vmspace_pmap(vm), min, max);
304 vmspace_container_reset(struct proc *p)
309 racct_set(p, RACCT_DATA, 0);
310 racct_set(p, RACCT_STACK, 0);
311 racct_set(p, RACCT_RSS, 0);
312 racct_set(p, RACCT_MEMLOCK, 0);
313 racct_set(p, RACCT_VMEM, 0);
319 vmspace_dofree(struct vmspace *vm)
322 CTR1(KTR_VM, "vmspace_free: %p", vm);
325 * Make sure any SysV shm is freed, it might not have been in
331 * Lock the map, to wait out all other references to it.
332 * Delete all of the mappings and pages they hold, then call
333 * the pmap module to reclaim anything left.
335 (void)vm_map_remove(&vm->vm_map, vm->vm_map.min_offset,
336 vm->vm_map.max_offset);
338 pmap_release(vmspace_pmap(vm));
339 vm->vm_map.pmap = NULL;
340 uma_zfree(vmspace_zone, vm);
344 vmspace_free(struct vmspace *vm)
347 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);
418 vmspace_container_reset(p);
421 /* Acquire reference to vmspace owned by another process. */
424 vmspace_acquire_ref(struct proc *p)
429 PROC_VMSPACE_LOCK(p);
432 PROC_VMSPACE_UNLOCK(p);
436 refcnt = vm->vm_refcnt;
437 if (refcnt <= 0) { /* Avoid 0->1 transition */
438 PROC_VMSPACE_UNLOCK(p);
441 } while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt + 1));
442 if (vm != p->p_vmspace) {
443 PROC_VMSPACE_UNLOCK(p);
447 PROC_VMSPACE_UNLOCK(p);
452 _vm_map_lock(vm_map_t map, const char *file, int line)
456 mtx_lock_flags_(&map->system_mtx, 0, file, line);
458 sx_xlock_(&map->lock, file, line);
463 vm_map_process_deferred(void)
466 vm_map_entry_t entry, next;
470 entry = td->td_map_def_user;
471 td->td_map_def_user = NULL;
472 while (entry != NULL) {
474 if ((entry->eflags & MAP_ENTRY_VN_WRITECNT) != 0) {
476 * Decrement the object's writemappings and
477 * possibly the vnode's v_writecount.
479 KASSERT((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0,
480 ("Submap with writecount"));
481 object = entry->object.vm_object;
482 KASSERT(object != NULL, ("No object for writecount"));
483 vnode_pager_release_writecount(object, entry->start,
486 vm_map_entry_deallocate(entry, FALSE);
492 _vm_map_unlock(vm_map_t map, const char *file, int line)
496 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
498 sx_xunlock_(&map->lock, file, line);
499 vm_map_process_deferred();
504 _vm_map_lock_read(vm_map_t map, const char *file, int line)
508 mtx_lock_flags_(&map->system_mtx, 0, file, line);
510 sx_slock_(&map->lock, file, line);
514 _vm_map_unlock_read(vm_map_t map, const char *file, int line)
518 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
520 sx_sunlock_(&map->lock, file, line);
521 vm_map_process_deferred();
526 _vm_map_trylock(vm_map_t map, const char *file, int line)
530 error = map->system_map ?
531 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) :
532 !sx_try_xlock_(&map->lock, file, line);
539 _vm_map_trylock_read(vm_map_t map, const char *file, int line)
543 error = map->system_map ?
544 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) :
545 !sx_try_slock_(&map->lock, file, line);
550 * _vm_map_lock_upgrade: [ internal use only ]
552 * Tries to upgrade a read (shared) lock on the specified map to a write
553 * (exclusive) lock. Returns the value "0" if the upgrade succeeds and a
554 * non-zero value if the upgrade fails. If the upgrade fails, the map is
555 * returned without a read or write lock held.
557 * Requires that the map be read locked.
560 _vm_map_lock_upgrade(vm_map_t map, const char *file, int line)
562 unsigned int last_timestamp;
564 if (map->system_map) {
565 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
567 if (!sx_try_upgrade_(&map->lock, file, line)) {
568 last_timestamp = map->timestamp;
569 sx_sunlock_(&map->lock, file, line);
570 vm_map_process_deferred();
572 * If the map's timestamp does not change while the
573 * map is unlocked, then the upgrade succeeds.
575 sx_xlock_(&map->lock, file, line);
576 if (last_timestamp != map->timestamp) {
577 sx_xunlock_(&map->lock, file, line);
587 _vm_map_lock_downgrade(vm_map_t map, const char *file, int line)
590 if (map->system_map) {
591 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
593 sx_downgrade_(&map->lock, file, line);
599 * Returns a non-zero value if the caller holds a write (exclusive) lock
600 * on the specified map and the value "0" otherwise.
603 vm_map_locked(vm_map_t map)
607 return (mtx_owned(&map->system_mtx));
609 return (sx_xlocked(&map->lock));
614 _vm_map_assert_locked(vm_map_t map, const char *file, int line)
618 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
620 sx_assert_(&map->lock, SA_XLOCKED, file, line);
623 #define VM_MAP_ASSERT_LOCKED(map) \
624 _vm_map_assert_locked(map, LOCK_FILE, LOCK_LINE)
626 #define VM_MAP_ASSERT_LOCKED(map)
630 * _vm_map_unlock_and_wait:
632 * Atomically releases the lock on the specified map and puts the calling
633 * thread to sleep. The calling thread will remain asleep until either
634 * vm_map_wakeup() is performed on the map or the specified timeout is
637 * WARNING! This function does not perform deferred deallocations of
638 * objects and map entries. Therefore, the calling thread is expected to
639 * reacquire the map lock after reawakening and later perform an ordinary
640 * unlock operation, such as vm_map_unlock(), before completing its
641 * operation on the map.
644 _vm_map_unlock_and_wait(vm_map_t map, int timo, const char *file, int line)
647 mtx_lock(&map_sleep_mtx);
649 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
651 sx_xunlock_(&map->lock, file, line);
652 return (msleep(&map->root, &map_sleep_mtx, PDROP | PVM, "vmmaps",
659 * Awaken any threads that have slept on the map using
660 * vm_map_unlock_and_wait().
663 vm_map_wakeup(vm_map_t map)
667 * Acquire and release map_sleep_mtx to prevent a wakeup()
668 * from being performed (and lost) between the map unlock
669 * and the msleep() in _vm_map_unlock_and_wait().
671 mtx_lock(&map_sleep_mtx);
672 mtx_unlock(&map_sleep_mtx);
677 vm_map_busy(vm_map_t map)
680 VM_MAP_ASSERT_LOCKED(map);
685 vm_map_unbusy(vm_map_t map)
688 VM_MAP_ASSERT_LOCKED(map);
689 KASSERT(map->busy, ("vm_map_unbusy: not busy"));
690 if (--map->busy == 0 && (map->flags & MAP_BUSY_WAKEUP)) {
691 vm_map_modflags(map, 0, MAP_BUSY_WAKEUP);
697 vm_map_wait_busy(vm_map_t map)
700 VM_MAP_ASSERT_LOCKED(map);
702 vm_map_modflags(map, MAP_BUSY_WAKEUP, 0);
704 msleep(&map->busy, &map->system_mtx, 0, "mbusy", 0);
706 sx_sleep(&map->busy, &map->lock, 0, "mbusy", 0);
712 vmspace_resident_count(struct vmspace *vmspace)
714 return pmap_resident_count(vmspace_pmap(vmspace));
720 * Creates and returns a new empty VM map with
721 * the given physical map structure, and having
722 * the given lower and upper address bounds.
725 vm_map_create(pmap_t pmap, vm_offset_t min, vm_offset_t max)
729 result = uma_zalloc(mapzone, M_WAITOK);
730 CTR1(KTR_VM, "vm_map_create: %p", result);
731 _vm_map_init(result, pmap, min, max);
736 * Initialize an existing vm_map structure
737 * such as that in the vmspace structure.
740 _vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max)
743 map->header.next = map->header.prev = &map->header;
744 map->needs_wakeup = FALSE;
747 map->min_offset = min;
748 map->max_offset = max;
756 vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max)
759 _vm_map_init(map, pmap, min, max);
760 mtx_init(&map->system_mtx, "system map", NULL, MTX_DEF | MTX_DUPOK);
761 sx_init(&map->lock, "user map");
765 * vm_map_entry_dispose: [ internal use only ]
767 * Inverse of vm_map_entry_create.
770 vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry)
772 uma_zfree(map->system_map ? kmapentzone : mapentzone, entry);
776 * vm_map_entry_create: [ internal use only ]
778 * Allocates a VM map entry for insertion.
779 * No entry fields are filled in.
781 static vm_map_entry_t
782 vm_map_entry_create(vm_map_t map)
784 vm_map_entry_t new_entry;
787 new_entry = uma_zalloc(kmapentzone, M_NOWAIT);
789 new_entry = uma_zalloc(mapentzone, M_WAITOK);
790 if (new_entry == NULL)
791 panic("vm_map_entry_create: kernel resources exhausted");
796 * vm_map_entry_set_behavior:
798 * Set the expected access behavior, either normal, random, or
802 vm_map_entry_set_behavior(vm_map_entry_t entry, u_char behavior)
804 entry->eflags = (entry->eflags & ~MAP_ENTRY_BEHAV_MASK) |
805 (behavior & MAP_ENTRY_BEHAV_MASK);
809 * vm_map_entry_set_max_free:
811 * Set the max_free field in a vm_map_entry.
814 vm_map_entry_set_max_free(vm_map_entry_t entry)
817 entry->max_free = entry->adj_free;
818 if (entry->left != NULL && entry->left->max_free > entry->max_free)
819 entry->max_free = entry->left->max_free;
820 if (entry->right != NULL && entry->right->max_free > entry->max_free)
821 entry->max_free = entry->right->max_free;
825 * vm_map_entry_splay:
827 * The Sleator and Tarjan top-down splay algorithm with the
828 * following variation. Max_free must be computed bottom-up, so
829 * on the downward pass, maintain the left and right spines in
830 * reverse order. Then, make a second pass up each side to fix
831 * the pointers and compute max_free. The time bound is O(log n)
834 * The new root is the vm_map_entry containing "addr", or else an
835 * adjacent entry (lower or higher) if addr is not in the tree.
837 * The map must be locked, and leaves it so.
839 * Returns: the new root.
841 static vm_map_entry_t
842 vm_map_entry_splay(vm_offset_t addr, vm_map_entry_t root)
844 vm_map_entry_t llist, rlist;
845 vm_map_entry_t ltree, rtree;
848 /* Special case of empty tree. */
853 * Pass One: Splay down the tree until we find addr or a NULL
854 * pointer where addr would go. llist and rlist are the two
855 * sides in reverse order (bottom-up), with llist linked by
856 * the right pointer and rlist linked by the left pointer in
857 * the vm_map_entry. Wait until Pass Two to set max_free on
863 /* root is never NULL in here. */
864 if (addr < root->start) {
868 if (addr < y->start && y->left != NULL) {
869 /* Rotate right and put y on rlist. */
870 root->left = y->right;
872 vm_map_entry_set_max_free(root);
877 /* Put root on rlist. */
882 } else if (addr >= root->end) {
886 if (addr >= y->end && y->right != NULL) {
887 /* Rotate left and put y on llist. */
888 root->right = y->left;
890 vm_map_entry_set_max_free(root);
895 /* Put root on llist. */
905 * Pass Two: Walk back up the two spines, flip the pointers
906 * and set max_free. The subtrees of the root go at the
907 * bottom of llist and rlist.
910 while (llist != NULL) {
912 llist->right = ltree;
913 vm_map_entry_set_max_free(llist);
918 while (rlist != NULL) {
921 vm_map_entry_set_max_free(rlist);
927 * Final assembly: add ltree and rtree as subtrees of root.
931 vm_map_entry_set_max_free(root);
937 * vm_map_entry_{un,}link:
939 * Insert/remove entries from maps.
942 vm_map_entry_link(vm_map_t map,
943 vm_map_entry_t after_where,
944 vm_map_entry_t entry)
948 "vm_map_entry_link: map %p, nentries %d, entry %p, after %p", map,
949 map->nentries, entry, after_where);
950 VM_MAP_ASSERT_LOCKED(map);
951 KASSERT(after_where == &map->header ||
952 after_where->end <= entry->start,
953 ("vm_map_entry_link: prev end %jx new start %jx overlap",
954 (uintmax_t)after_where->end, (uintmax_t)entry->start));
955 KASSERT(after_where->next == &map->header ||
956 entry->end <= after_where->next->start,
957 ("vm_map_entry_link: new end %jx next start %jx overlap",
958 (uintmax_t)entry->end, (uintmax_t)after_where->next->start));
961 entry->prev = after_where;
962 entry->next = after_where->next;
963 entry->next->prev = entry;
964 after_where->next = entry;
966 if (after_where != &map->header) {
967 if (after_where != map->root)
968 vm_map_entry_splay(after_where->start, map->root);
969 entry->right = after_where->right;
970 entry->left = after_where;
971 after_where->right = NULL;
972 after_where->adj_free = entry->start - after_where->end;
973 vm_map_entry_set_max_free(after_where);
975 entry->right = map->root;
978 entry->adj_free = (entry->next == &map->header ? map->max_offset :
979 entry->next->start) - entry->end;
980 vm_map_entry_set_max_free(entry);
985 vm_map_entry_unlink(vm_map_t map,
986 vm_map_entry_t entry)
988 vm_map_entry_t next, prev, root;
990 VM_MAP_ASSERT_LOCKED(map);
991 if (entry != map->root)
992 vm_map_entry_splay(entry->start, map->root);
993 if (entry->left == NULL)
996 root = vm_map_entry_splay(entry->start, entry->left);
997 root->right = entry->right;
998 root->adj_free = (entry->next == &map->header ? map->max_offset :
999 entry->next->start) - root->end;
1000 vm_map_entry_set_max_free(root);
1009 CTR3(KTR_VM, "vm_map_entry_unlink: map %p, nentries %d, entry %p", map,
1010 map->nentries, entry);
1014 * vm_map_entry_resize_free:
1016 * Recompute the amount of free space following a vm_map_entry
1017 * and propagate that value up the tree. Call this function after
1018 * resizing a map entry in-place, that is, without a call to
1019 * vm_map_entry_link() or _unlink().
1021 * The map must be locked, and leaves it so.
1024 vm_map_entry_resize_free(vm_map_t map, vm_map_entry_t entry)
1028 * Using splay trees without parent pointers, propagating
1029 * max_free up the tree is done by moving the entry to the
1030 * root and making the change there.
1032 if (entry != map->root)
1033 map->root = vm_map_entry_splay(entry->start, map->root);
1035 entry->adj_free = (entry->next == &map->header ? map->max_offset :
1036 entry->next->start) - entry->end;
1037 vm_map_entry_set_max_free(entry);
1041 * vm_map_lookup_entry: [ internal use only ]
1043 * Finds the map entry containing (or
1044 * immediately preceding) the specified address
1045 * in the given map; the entry is returned
1046 * in the "entry" parameter. The boolean
1047 * result indicates whether the address is
1048 * actually contained in the map.
1051 vm_map_lookup_entry(
1053 vm_offset_t address,
1054 vm_map_entry_t *entry) /* OUT */
1060 * If the map is empty, then the map entry immediately preceding
1061 * "address" is the map's header.
1065 *entry = &map->header;
1066 else if (address >= cur->start && cur->end > address) {
1069 } else if ((locked = vm_map_locked(map)) ||
1070 sx_try_upgrade(&map->lock)) {
1072 * Splay requires a write lock on the map. However, it only
1073 * restructures the binary search tree; it does not otherwise
1074 * change the map. Thus, the map's timestamp need not change
1075 * on a temporary upgrade.
1077 map->root = cur = vm_map_entry_splay(address, cur);
1079 sx_downgrade(&map->lock);
1082 * If "address" is contained within a map entry, the new root
1083 * is that map entry. Otherwise, the new root is a map entry
1084 * immediately before or after "address".
1086 if (address >= cur->start) {
1088 if (cur->end > address)
1094 * Since the map is only locked for read access, perform a
1095 * standard binary search tree lookup for "address".
1098 if (address < cur->start) {
1099 if (cur->left == NULL) {
1104 } else if (cur->end > address) {
1108 if (cur->right == NULL) {
1121 * Inserts the given whole VM object into the target
1122 * map at the specified address range. The object's
1123 * size should match that of the address range.
1125 * Requires that the map be locked, and leaves it so.
1127 * If object is non-NULL, ref count must be bumped by caller
1128 * prior to making call to account for the new entry.
1131 vm_map_insert(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1132 vm_offset_t start, vm_offset_t end, vm_prot_t prot, vm_prot_t max, int cow)
1134 vm_map_entry_t new_entry, prev_entry, temp_entry;
1135 vm_eflags_t protoeflags;
1137 vm_inherit_t inheritance;
1139 VM_MAP_ASSERT_LOCKED(map);
1140 KASSERT((object != kmem_object && object != kernel_object) ||
1141 (cow & MAP_COPY_ON_WRITE) == 0,
1142 ("vm_map_insert: kmem or kernel object and COW"));
1143 KASSERT(object == NULL || (cow & MAP_NOFAULT) == 0,
1144 ("vm_map_insert: paradoxical MAP_NOFAULT request"));
1147 * Check that the start and end points are not bogus.
1149 if ((start < map->min_offset) || (end > map->max_offset) ||
1151 return (KERN_INVALID_ADDRESS);
1154 * Find the entry prior to the proposed starting address; if it's part
1155 * of an existing entry, this range is bogus.
1157 if (vm_map_lookup_entry(map, start, &temp_entry))
1158 return (KERN_NO_SPACE);
1160 prev_entry = temp_entry;
1163 * Assert that the next entry doesn't overlap the end point.
1165 if ((prev_entry->next != &map->header) &&
1166 (prev_entry->next->start < end))
1167 return (KERN_NO_SPACE);
1170 if (cow & MAP_COPY_ON_WRITE)
1171 protoeflags |= MAP_ENTRY_COW | MAP_ENTRY_NEEDS_COPY;
1172 if (cow & MAP_NOFAULT)
1173 protoeflags |= MAP_ENTRY_NOFAULT;
1174 if (cow & MAP_DISABLE_SYNCER)
1175 protoeflags |= MAP_ENTRY_NOSYNC;
1176 if (cow & MAP_DISABLE_COREDUMP)
1177 protoeflags |= MAP_ENTRY_NOCOREDUMP;
1178 if (cow & MAP_STACK_GROWS_DOWN)
1179 protoeflags |= MAP_ENTRY_GROWS_DOWN;
1180 if (cow & MAP_STACK_GROWS_UP)
1181 protoeflags |= MAP_ENTRY_GROWS_UP;
1182 if (cow & MAP_VN_WRITECOUNT)
1183 protoeflags |= MAP_ENTRY_VN_WRITECNT;
1184 if (cow & MAP_INHERIT_SHARE)
1185 inheritance = VM_INHERIT_SHARE;
1187 inheritance = VM_INHERIT_DEFAULT;
1190 if (cow & (MAP_ACC_NO_CHARGE | MAP_NOFAULT))
1192 if ((cow & MAP_ACC_CHARGED) || ((prot & VM_PROT_WRITE) &&
1193 ((protoeflags & MAP_ENTRY_NEEDS_COPY) || object == NULL))) {
1194 if (!(cow & MAP_ACC_CHARGED) && !swap_reserve(end - start))
1195 return (KERN_RESOURCE_SHORTAGE);
1196 KASSERT(object == NULL || (protoeflags & MAP_ENTRY_NEEDS_COPY) ||
1197 object->cred == NULL,
1198 ("OVERCOMMIT: vm_map_insert o %p", object));
1199 cred = curthread->td_ucred;
1203 /* Expand the kernel pmap, if necessary. */
1204 if (map == kernel_map && end > kernel_vm_end)
1205 pmap_growkernel(end);
1206 if (object != NULL) {
1208 * OBJ_ONEMAPPING must be cleared unless this mapping
1209 * is trivially proven to be the only mapping for any
1210 * of the object's pages. (Object granularity
1211 * reference counting is insufficient to recognize
1212 * aliases with precision.)
1214 VM_OBJECT_WLOCK(object);
1215 if (object->ref_count > 1 || object->shadow_count != 0)
1216 vm_object_clear_flag(object, OBJ_ONEMAPPING);
1217 VM_OBJECT_WUNLOCK(object);
1219 else if ((prev_entry != &map->header) &&
1220 (prev_entry->eflags == protoeflags) &&
1221 (cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 &&
1222 (prev_entry->end == start) &&
1223 (prev_entry->wired_count == 0) &&
1224 (prev_entry->cred == cred ||
1225 (prev_entry->object.vm_object != NULL &&
1226 (prev_entry->object.vm_object->cred == cred))) &&
1227 vm_object_coalesce(prev_entry->object.vm_object,
1229 (vm_size_t)(prev_entry->end - prev_entry->start),
1230 (vm_size_t)(end - prev_entry->end), cred != NULL &&
1231 (protoeflags & MAP_ENTRY_NEEDS_COPY) == 0)) {
1233 * We were able to extend the object. Determine if we
1234 * can extend the previous map entry to include the
1235 * new range as well.
1237 if ((prev_entry->inheritance == inheritance) &&
1238 (prev_entry->protection == prot) &&
1239 (prev_entry->max_protection == max)) {
1240 map->size += (end - prev_entry->end);
1241 prev_entry->end = end;
1242 vm_map_entry_resize_free(map, prev_entry);
1243 vm_map_simplify_entry(map, prev_entry);
1244 return (KERN_SUCCESS);
1248 * If we can extend the object but cannot extend the
1249 * map entry, we have to create a new map entry. We
1250 * must bump the ref count on the extended object to
1251 * account for it. object may be NULL.
1253 object = prev_entry->object.vm_object;
1254 offset = prev_entry->offset +
1255 (prev_entry->end - prev_entry->start);
1256 vm_object_reference(object);
1257 if (cred != NULL && object != NULL && object->cred != NULL &&
1258 !(prev_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
1259 /* Object already accounts for this uid. */
1267 * Create a new entry
1269 new_entry = vm_map_entry_create(map);
1270 new_entry->start = start;
1271 new_entry->end = end;
1272 new_entry->cred = NULL;
1274 new_entry->eflags = protoeflags;
1275 new_entry->object.vm_object = object;
1276 new_entry->offset = offset;
1277 new_entry->avail_ssize = 0;
1279 new_entry->inheritance = inheritance;
1280 new_entry->protection = prot;
1281 new_entry->max_protection = max;
1282 new_entry->wired_count = 0;
1283 new_entry->wiring_thread = NULL;
1284 new_entry->read_ahead = VM_FAULT_READ_AHEAD_INIT;
1285 new_entry->next_read = OFF_TO_IDX(offset);
1287 KASSERT(cred == NULL || !ENTRY_CHARGED(new_entry),
1288 ("OVERCOMMIT: vm_map_insert leaks vm_map %p", new_entry));
1289 new_entry->cred = cred;
1292 * Insert the new entry into the list
1294 vm_map_entry_link(map, prev_entry, new_entry);
1295 map->size += new_entry->end - new_entry->start;
1298 * Try to coalesce the new entry with both the previous and next
1299 * entries in the list. Previously, we only attempted to coalesce
1300 * with the previous entry when object is NULL. Here, we handle the
1301 * other cases, which are less common.
1303 vm_map_simplify_entry(map, new_entry);
1305 if (cow & (MAP_PREFAULT|MAP_PREFAULT_PARTIAL)) {
1306 vm_map_pmap_enter(map, start, prot,
1307 object, OFF_TO_IDX(offset), end - start,
1308 cow & MAP_PREFAULT_PARTIAL);
1311 return (KERN_SUCCESS);
1317 * Find the first fit (lowest VM address) for "length" free bytes
1318 * beginning at address >= start in the given map.
1320 * In a vm_map_entry, "adj_free" is the amount of free space
1321 * adjacent (higher address) to this entry, and "max_free" is the
1322 * maximum amount of contiguous free space in its subtree. This
1323 * allows finding a free region in one path down the tree, so
1324 * O(log n) amortized with splay trees.
1326 * The map must be locked, and leaves it so.
1328 * Returns: 0 on success, and starting address in *addr,
1329 * 1 if insufficient space.
1332 vm_map_findspace(vm_map_t map, vm_offset_t start, vm_size_t length,
1333 vm_offset_t *addr) /* OUT */
1335 vm_map_entry_t entry;
1339 * Request must fit within min/max VM address and must avoid
1342 if (start < map->min_offset)
1343 start = map->min_offset;
1344 if (start + length > map->max_offset || start + length < start)
1347 /* Empty tree means wide open address space. */
1348 if (map->root == NULL) {
1354 * After splay, if start comes before root node, then there
1355 * must be a gap from start to the root.
1357 map->root = vm_map_entry_splay(start, map->root);
1358 if (start + length <= map->root->start) {
1364 * Root is the last node that might begin its gap before
1365 * start, and this is the last comparison where address
1366 * wrap might be a problem.
1368 st = (start > map->root->end) ? start : map->root->end;
1369 if (length <= map->root->end + map->root->adj_free - st) {
1374 /* With max_free, can immediately tell if no solution. */
1375 entry = map->root->right;
1376 if (entry == NULL || length > entry->max_free)
1380 * Search the right subtree in the order: left subtree, root,
1381 * right subtree (first fit). The previous splay implies that
1382 * all regions in the right subtree have addresses > start.
1384 while (entry != NULL) {
1385 if (entry->left != NULL && entry->left->max_free >= length)
1386 entry = entry->left;
1387 else if (entry->adj_free >= length) {
1391 entry = entry->right;
1394 /* Can't get here, so panic if we do. */
1395 panic("vm_map_findspace: max_free corrupt");
1399 vm_map_fixed(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1400 vm_offset_t start, vm_size_t length, vm_prot_t prot,
1401 vm_prot_t max, int cow)
1406 end = start + length;
1407 KASSERT((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 ||
1409 ("vm_map_fixed: non-NULL backing object for stack"));
1411 VM_MAP_RANGE_CHECK(map, start, end);
1412 if ((cow & MAP_CHECK_EXCL) == 0)
1413 vm_map_delete(map, start, end);
1414 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) != 0) {
1415 result = vm_map_stack_locked(map, start, length, sgrowsiz,
1418 result = vm_map_insert(map, object, offset, start, end,
1426 * vm_map_find finds an unallocated region in the target address
1427 * map with the given length. The search is defined to be
1428 * first-fit from the specified address; the region found is
1429 * returned in the same parameter.
1431 * If object is non-NULL, ref count must be bumped by caller
1432 * prior to making call to account for the new entry.
1435 vm_map_find(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1436 vm_offset_t *addr, /* IN/OUT */
1437 vm_size_t length, vm_offset_t max_addr, int find_space,
1438 vm_prot_t prot, vm_prot_t max, int cow)
1440 vm_offset_t alignment, initial_addr, start;
1443 KASSERT((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 ||
1445 ("vm_map_find: non-NULL backing object for stack"));
1446 if (find_space == VMFS_OPTIMAL_SPACE && (object == NULL ||
1447 (object->flags & OBJ_COLORED) == 0))
1448 find_space = VMFS_ANY_SPACE;
1449 if (find_space >> 8 != 0) {
1450 KASSERT((find_space & 0xff) == 0, ("bad VMFS flags"));
1451 alignment = (vm_offset_t)1 << (find_space >> 8);
1454 initial_addr = *addr;
1456 start = initial_addr;
1459 if (find_space != VMFS_NO_SPACE) {
1460 if (vm_map_findspace(map, start, length, addr) ||
1461 (max_addr != 0 && *addr + length > max_addr)) {
1463 if (find_space == VMFS_OPTIMAL_SPACE) {
1464 find_space = VMFS_ANY_SPACE;
1467 return (KERN_NO_SPACE);
1469 switch (find_space) {
1470 case VMFS_SUPER_SPACE:
1471 case VMFS_OPTIMAL_SPACE:
1472 pmap_align_superpage(object, offset, addr,
1475 case VMFS_ANY_SPACE:
1478 if ((*addr & (alignment - 1)) != 0) {
1479 *addr &= ~(alignment - 1);
1487 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) != 0) {
1488 result = vm_map_stack_locked(map, start, length,
1489 sgrowsiz, prot, max, cow);
1491 result = vm_map_insert(map, object, offset, start,
1492 start + length, prot, max, cow);
1494 } while (result == KERN_NO_SPACE && find_space != VMFS_NO_SPACE &&
1495 find_space != VMFS_ANY_SPACE);
1501 * vm_map_simplify_entry:
1503 * Simplify the given map entry by merging with either neighbor. This
1504 * routine also has the ability to merge with both neighbors.
1506 * The map must be locked.
1508 * This routine guarentees that the passed entry remains valid (though
1509 * possibly extended). When merging, this routine may delete one or
1513 vm_map_simplify_entry(vm_map_t map, vm_map_entry_t entry)
1515 vm_map_entry_t next, prev;
1516 vm_size_t prevsize, esize;
1518 if ((entry->eflags & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP |
1519 MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_IS_SUB_MAP)) != 0)
1523 if (prev != &map->header) {
1524 prevsize = prev->end - prev->start;
1525 if ( (prev->end == entry->start) &&
1526 (prev->object.vm_object == entry->object.vm_object) &&
1527 (!prev->object.vm_object ||
1528 (prev->offset + prevsize == entry->offset)) &&
1529 (prev->eflags == entry->eflags) &&
1530 (prev->protection == entry->protection) &&
1531 (prev->max_protection == entry->max_protection) &&
1532 (prev->inheritance == entry->inheritance) &&
1533 (prev->wired_count == entry->wired_count) &&
1534 (prev->cred == entry->cred)) {
1535 vm_map_entry_unlink(map, prev);
1536 entry->start = prev->start;
1537 entry->offset = prev->offset;
1538 if (entry->prev != &map->header)
1539 vm_map_entry_resize_free(map, entry->prev);
1542 * If the backing object is a vnode object,
1543 * vm_object_deallocate() calls vrele().
1544 * However, vrele() does not lock the vnode
1545 * because the vnode has additional
1546 * references. Thus, the map lock can be kept
1547 * without causing a lock-order reversal with
1550 * Since we count the number of virtual page
1551 * mappings in object->un_pager.vnp.writemappings,
1552 * the writemappings value should not be adjusted
1553 * when the entry is disposed of.
1555 if (prev->object.vm_object)
1556 vm_object_deallocate(prev->object.vm_object);
1557 if (prev->cred != NULL)
1559 vm_map_entry_dispose(map, prev);
1564 if (next != &map->header) {
1565 esize = entry->end - entry->start;
1566 if ((entry->end == next->start) &&
1567 (next->object.vm_object == entry->object.vm_object) &&
1568 (!entry->object.vm_object ||
1569 (entry->offset + esize == next->offset)) &&
1570 (next->eflags == entry->eflags) &&
1571 (next->protection == entry->protection) &&
1572 (next->max_protection == entry->max_protection) &&
1573 (next->inheritance == entry->inheritance) &&
1574 (next->wired_count == entry->wired_count) &&
1575 (next->cred == entry->cred)) {
1576 vm_map_entry_unlink(map, next);
1577 entry->end = next->end;
1578 vm_map_entry_resize_free(map, entry);
1581 * See comment above.
1583 if (next->object.vm_object)
1584 vm_object_deallocate(next->object.vm_object);
1585 if (next->cred != NULL)
1587 vm_map_entry_dispose(map, next);
1592 * vm_map_clip_start: [ internal use only ]
1594 * Asserts that the given entry begins at or after
1595 * the specified address; if necessary,
1596 * it splits the entry into two.
1598 #define vm_map_clip_start(map, entry, startaddr) \
1600 if (startaddr > entry->start) \
1601 _vm_map_clip_start(map, entry, startaddr); \
1605 * This routine is called only when it is known that
1606 * the entry must be split.
1609 _vm_map_clip_start(vm_map_t map, vm_map_entry_t entry, vm_offset_t start)
1611 vm_map_entry_t new_entry;
1613 VM_MAP_ASSERT_LOCKED(map);
1616 * Split off the front portion -- note that we must insert the new
1617 * entry BEFORE this one, so that this entry has the specified
1620 vm_map_simplify_entry(map, entry);
1623 * If there is no object backing this entry, we might as well create
1624 * one now. If we defer it, an object can get created after the map
1625 * is clipped, and individual objects will be created for the split-up
1626 * map. This is a bit of a hack, but is also about the best place to
1627 * put this improvement.
1629 if (entry->object.vm_object == NULL && !map->system_map) {
1631 object = vm_object_allocate(OBJT_DEFAULT,
1632 atop(entry->end - entry->start));
1633 entry->object.vm_object = object;
1635 if (entry->cred != NULL) {
1636 object->cred = entry->cred;
1637 object->charge = entry->end - entry->start;
1640 } else if (entry->object.vm_object != NULL &&
1641 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) &&
1642 entry->cred != NULL) {
1643 VM_OBJECT_WLOCK(entry->object.vm_object);
1644 KASSERT(entry->object.vm_object->cred == NULL,
1645 ("OVERCOMMIT: vm_entry_clip_start: both cred e %p", entry));
1646 entry->object.vm_object->cred = entry->cred;
1647 entry->object.vm_object->charge = entry->end - entry->start;
1648 VM_OBJECT_WUNLOCK(entry->object.vm_object);
1652 new_entry = vm_map_entry_create(map);
1653 *new_entry = *entry;
1655 new_entry->end = start;
1656 entry->offset += (start - entry->start);
1657 entry->start = start;
1658 if (new_entry->cred != NULL)
1659 crhold(entry->cred);
1661 vm_map_entry_link(map, entry->prev, new_entry);
1663 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1664 vm_object_reference(new_entry->object.vm_object);
1666 * The object->un_pager.vnp.writemappings for the
1667 * object of MAP_ENTRY_VN_WRITECNT type entry shall be
1668 * kept as is here. The virtual pages are
1669 * re-distributed among the clipped entries, so the sum is
1676 * vm_map_clip_end: [ internal use only ]
1678 * Asserts that the given entry ends at or before
1679 * the specified address; if necessary,
1680 * it splits the entry into two.
1682 #define vm_map_clip_end(map, entry, endaddr) \
1684 if ((endaddr) < (entry->end)) \
1685 _vm_map_clip_end((map), (entry), (endaddr)); \
1689 * This routine is called only when it is known that
1690 * the entry must be split.
1693 _vm_map_clip_end(vm_map_t map, vm_map_entry_t entry, vm_offset_t end)
1695 vm_map_entry_t new_entry;
1697 VM_MAP_ASSERT_LOCKED(map);
1700 * If there is no object backing this entry, we might as well create
1701 * one now. If we defer it, an object can get created after the map
1702 * is clipped, and individual objects will be created for the split-up
1703 * map. This is a bit of a hack, but is also about the best place to
1704 * put this improvement.
1706 if (entry->object.vm_object == NULL && !map->system_map) {
1708 object = vm_object_allocate(OBJT_DEFAULT,
1709 atop(entry->end - entry->start));
1710 entry->object.vm_object = object;
1712 if (entry->cred != NULL) {
1713 object->cred = entry->cred;
1714 object->charge = entry->end - entry->start;
1717 } else if (entry->object.vm_object != NULL &&
1718 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) &&
1719 entry->cred != NULL) {
1720 VM_OBJECT_WLOCK(entry->object.vm_object);
1721 KASSERT(entry->object.vm_object->cred == NULL,
1722 ("OVERCOMMIT: vm_entry_clip_end: both cred e %p", entry));
1723 entry->object.vm_object->cred = entry->cred;
1724 entry->object.vm_object->charge = entry->end - entry->start;
1725 VM_OBJECT_WUNLOCK(entry->object.vm_object);
1730 * Create a new entry and insert it AFTER the specified entry
1732 new_entry = vm_map_entry_create(map);
1733 *new_entry = *entry;
1735 new_entry->start = entry->end = end;
1736 new_entry->offset += (end - entry->start);
1737 if (new_entry->cred != NULL)
1738 crhold(entry->cred);
1740 vm_map_entry_link(map, entry, new_entry);
1742 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1743 vm_object_reference(new_entry->object.vm_object);
1748 * vm_map_submap: [ kernel use only ]
1750 * Mark the given range as handled by a subordinate map.
1752 * This range must have been created with vm_map_find,
1753 * and no other operations may have been performed on this
1754 * range prior to calling vm_map_submap.
1756 * Only a limited number of operations can be performed
1757 * within this rage after calling vm_map_submap:
1759 * [Don't try vm_map_copy!]
1761 * To remove a submapping, one must first remove the
1762 * range from the superior map, and then destroy the
1763 * submap (if desired). [Better yet, don't try it.]
1772 vm_map_entry_t entry;
1773 int result = KERN_INVALID_ARGUMENT;
1777 VM_MAP_RANGE_CHECK(map, start, end);
1779 if (vm_map_lookup_entry(map, start, &entry)) {
1780 vm_map_clip_start(map, entry, start);
1782 entry = entry->next;
1784 vm_map_clip_end(map, entry, end);
1786 if ((entry->start == start) && (entry->end == end) &&
1787 ((entry->eflags & MAP_ENTRY_COW) == 0) &&
1788 (entry->object.vm_object == NULL)) {
1789 entry->object.sub_map = submap;
1790 entry->eflags |= MAP_ENTRY_IS_SUB_MAP;
1791 result = KERN_SUCCESS;
1799 * The maximum number of pages to map if MAP_PREFAULT_PARTIAL is specified
1801 #define MAX_INIT_PT 96
1804 * vm_map_pmap_enter:
1806 * Preload the specified map's pmap with mappings to the specified
1807 * object's memory-resident pages. No further physical pages are
1808 * allocated, and no further virtual pages are retrieved from secondary
1809 * storage. If the specified flags include MAP_PREFAULT_PARTIAL, then a
1810 * limited number of page mappings are created at the low-end of the
1811 * specified address range. (For this purpose, a superpage mapping
1812 * counts as one page mapping.) Otherwise, all resident pages within
1813 * the specified address range are mapped. Because these mappings are
1814 * being created speculatively, cached pages are not reactivated and
1818 vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot,
1819 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags)
1822 vm_page_t p, p_start;
1823 vm_pindex_t mask, psize, threshold, tmpidx;
1825 if ((prot & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0 || object == NULL)
1827 VM_OBJECT_RLOCK(object);
1828 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) {
1829 VM_OBJECT_RUNLOCK(object);
1830 VM_OBJECT_WLOCK(object);
1831 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) {
1832 pmap_object_init_pt(map->pmap, addr, object, pindex,
1834 VM_OBJECT_WUNLOCK(object);
1837 VM_OBJECT_LOCK_DOWNGRADE(object);
1841 if (psize + pindex > object->size) {
1842 if (object->size < pindex) {
1843 VM_OBJECT_RUNLOCK(object);
1846 psize = object->size - pindex;
1851 threshold = MAX_INIT_PT;
1853 p = vm_page_find_least(object, pindex);
1855 * Assert: the variable p is either (1) the page with the
1856 * least pindex greater than or equal to the parameter pindex
1860 p != NULL && (tmpidx = p->pindex - pindex) < psize;
1861 p = TAILQ_NEXT(p, listq)) {
1863 * don't allow an madvise to blow away our really
1864 * free pages allocating pv entries.
1866 if (((flags & MAP_PREFAULT_MADVISE) != 0 &&
1867 vm_cnt.v_free_count < vm_cnt.v_free_reserved) ||
1868 ((flags & MAP_PREFAULT_PARTIAL) != 0 &&
1869 tmpidx >= threshold)) {
1873 if (p->valid == VM_PAGE_BITS_ALL) {
1874 if (p_start == NULL) {
1875 start = addr + ptoa(tmpidx);
1878 /* Jump ahead if a superpage mapping is possible. */
1879 if (p->psind > 0 && ((addr + ptoa(tmpidx)) &
1880 (pagesizes[p->psind] - 1)) == 0) {
1881 mask = atop(pagesizes[p->psind]) - 1;
1882 if (tmpidx + mask < psize &&
1883 vm_page_ps_is_valid(p)) {
1888 } else if (p_start != NULL) {
1889 pmap_enter_object(map->pmap, start, addr +
1890 ptoa(tmpidx), p_start, prot);
1894 if (p_start != NULL)
1895 pmap_enter_object(map->pmap, start, addr + ptoa(psize),
1897 VM_OBJECT_RUNLOCK(object);
1903 * Sets the protection of the specified address
1904 * region in the target map. If "set_max" is
1905 * specified, the maximum protection is to be set;
1906 * otherwise, only the current protection is affected.
1909 vm_map_protect(vm_map_t map, vm_offset_t start, vm_offset_t end,
1910 vm_prot_t new_prot, boolean_t set_max)
1912 vm_map_entry_t current, entry;
1918 return (KERN_SUCCESS);
1922 VM_MAP_RANGE_CHECK(map, start, end);
1924 if (vm_map_lookup_entry(map, start, &entry)) {
1925 vm_map_clip_start(map, entry, start);
1927 entry = entry->next;
1931 * Make a first pass to check for protection violations.
1934 while ((current != &map->header) && (current->start < end)) {
1935 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
1937 return (KERN_INVALID_ARGUMENT);
1939 if ((new_prot & current->max_protection) != new_prot) {
1941 return (KERN_PROTECTION_FAILURE);
1943 current = current->next;
1948 * Do an accounting pass for private read-only mappings that
1949 * now will do cow due to allowed write (e.g. debugger sets
1950 * breakpoint on text segment)
1952 for (current = entry; (current != &map->header) &&
1953 (current->start < end); current = current->next) {
1955 vm_map_clip_end(map, current, end);
1958 ((new_prot & ~(current->protection)) & VM_PROT_WRITE) == 0 ||
1959 ENTRY_CHARGED(current)) {
1963 cred = curthread->td_ucred;
1964 obj = current->object.vm_object;
1966 if (obj == NULL || (current->eflags & MAP_ENTRY_NEEDS_COPY)) {
1967 if (!swap_reserve(current->end - current->start)) {
1969 return (KERN_RESOURCE_SHORTAGE);
1972 current->cred = cred;
1976 VM_OBJECT_WLOCK(obj);
1977 if (obj->type != OBJT_DEFAULT && obj->type != OBJT_SWAP) {
1978 VM_OBJECT_WUNLOCK(obj);
1983 * Charge for the whole object allocation now, since
1984 * we cannot distinguish between non-charged and
1985 * charged clipped mapping of the same object later.
1987 KASSERT(obj->charge == 0,
1988 ("vm_map_protect: object %p overcharged (entry %p)",
1990 if (!swap_reserve(ptoa(obj->size))) {
1991 VM_OBJECT_WUNLOCK(obj);
1993 return (KERN_RESOURCE_SHORTAGE);
1998 obj->charge = ptoa(obj->size);
1999 VM_OBJECT_WUNLOCK(obj);
2003 * Go back and fix up protections. [Note that clipping is not
2004 * necessary the second time.]
2007 while ((current != &map->header) && (current->start < end)) {
2008 old_prot = current->protection;
2011 current->protection =
2012 (current->max_protection = new_prot) &
2015 current->protection = new_prot;
2018 * For user wired map entries, the normal lazy evaluation of
2019 * write access upgrades through soft page faults is
2020 * undesirable. Instead, immediately copy any pages that are
2021 * copy-on-write and enable write access in the physical map.
2023 if ((current->eflags & MAP_ENTRY_USER_WIRED) != 0 &&
2024 (current->protection & VM_PROT_WRITE) != 0 &&
2025 (old_prot & VM_PROT_WRITE) == 0)
2026 vm_fault_copy_entry(map, map, current, current, NULL);
2029 * When restricting access, update the physical map. Worry
2030 * about copy-on-write here.
2032 if ((old_prot & ~current->protection) != 0) {
2033 #define MASK(entry) (((entry)->eflags & MAP_ENTRY_COW) ? ~VM_PROT_WRITE : \
2035 pmap_protect(map->pmap, current->start,
2037 current->protection & MASK(current));
2040 vm_map_simplify_entry(map, current);
2041 current = current->next;
2044 return (KERN_SUCCESS);
2050 * This routine traverses a processes map handling the madvise
2051 * system call. Advisories are classified as either those effecting
2052 * the vm_map_entry structure, or those effecting the underlying
2062 vm_map_entry_t current, entry;
2066 * Some madvise calls directly modify the vm_map_entry, in which case
2067 * we need to use an exclusive lock on the map and we need to perform
2068 * various clipping operations. Otherwise we only need a read-lock
2073 case MADV_SEQUENTIAL:
2080 return (KERN_SUCCESS);
2088 return (KERN_SUCCESS);
2089 vm_map_lock_read(map);
2092 return (KERN_INVALID_ARGUMENT);
2096 * Locate starting entry and clip if necessary.
2098 VM_MAP_RANGE_CHECK(map, start, end);
2100 if (vm_map_lookup_entry(map, start, &entry)) {
2102 vm_map_clip_start(map, entry, start);
2104 entry = entry->next;
2109 * madvise behaviors that are implemented in the vm_map_entry.
2111 * We clip the vm_map_entry so that behavioral changes are
2112 * limited to the specified address range.
2114 for (current = entry;
2115 (current != &map->header) && (current->start < end);
2116 current = current->next
2118 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
2121 vm_map_clip_end(map, current, end);
2125 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_NORMAL);
2127 case MADV_SEQUENTIAL:
2128 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_SEQUENTIAL);
2131 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_RANDOM);
2134 current->eflags |= MAP_ENTRY_NOSYNC;
2137 current->eflags &= ~MAP_ENTRY_NOSYNC;
2140 current->eflags |= MAP_ENTRY_NOCOREDUMP;
2143 current->eflags &= ~MAP_ENTRY_NOCOREDUMP;
2148 vm_map_simplify_entry(map, current);
2152 vm_pindex_t pstart, pend;
2155 * madvise behaviors that are implemented in the underlying
2158 * Since we don't clip the vm_map_entry, we have to clip
2159 * the vm_object pindex and count.
2161 for (current = entry;
2162 (current != &map->header) && (current->start < end);
2163 current = current->next
2165 vm_offset_t useEnd, useStart;
2167 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
2170 pstart = OFF_TO_IDX(current->offset);
2171 pend = pstart + atop(current->end - current->start);
2172 useStart = current->start;
2173 useEnd = current->end;
2175 if (current->start < start) {
2176 pstart += atop(start - current->start);
2179 if (current->end > end) {
2180 pend -= atop(current->end - end);
2188 * Perform the pmap_advise() before clearing
2189 * PGA_REFERENCED in vm_page_advise(). Otherwise, a
2190 * concurrent pmap operation, such as pmap_remove(),
2191 * could clear a reference in the pmap and set
2192 * PGA_REFERENCED on the page before the pmap_advise()
2193 * had completed. Consequently, the page would appear
2194 * referenced based upon an old reference that
2195 * occurred before this pmap_advise() ran.
2197 if (behav == MADV_DONTNEED || behav == MADV_FREE)
2198 pmap_advise(map->pmap, useStart, useEnd,
2201 vm_object_madvise(current->object.vm_object, pstart,
2205 * Pre-populate paging structures in the
2206 * WILLNEED case. For wired entries, the
2207 * paging structures are already populated.
2209 if (behav == MADV_WILLNEED &&
2210 current->wired_count == 0) {
2211 vm_map_pmap_enter(map,
2213 current->protection,
2214 current->object.vm_object,
2216 ptoa(pend - pstart),
2217 MAP_PREFAULT_MADVISE
2221 vm_map_unlock_read(map);
2230 * Sets the inheritance of the specified address
2231 * range in the target map. Inheritance
2232 * affects how the map will be shared with
2233 * child maps at the time of vmspace_fork.
2236 vm_map_inherit(vm_map_t map, vm_offset_t start, vm_offset_t end,
2237 vm_inherit_t new_inheritance)
2239 vm_map_entry_t entry;
2240 vm_map_entry_t temp_entry;
2242 switch (new_inheritance) {
2243 case VM_INHERIT_NONE:
2244 case VM_INHERIT_COPY:
2245 case VM_INHERIT_SHARE:
2248 return (KERN_INVALID_ARGUMENT);
2251 return (KERN_SUCCESS);
2253 VM_MAP_RANGE_CHECK(map, start, end);
2254 if (vm_map_lookup_entry(map, start, &temp_entry)) {
2256 vm_map_clip_start(map, entry, start);
2258 entry = temp_entry->next;
2259 while ((entry != &map->header) && (entry->start < end)) {
2260 vm_map_clip_end(map, entry, end);
2261 entry->inheritance = new_inheritance;
2262 vm_map_simplify_entry(map, entry);
2263 entry = entry->next;
2266 return (KERN_SUCCESS);
2272 * Implements both kernel and user unwiring.
2275 vm_map_unwire(vm_map_t map, vm_offset_t start, vm_offset_t end,
2278 vm_map_entry_t entry, first_entry, tmp_entry;
2279 vm_offset_t saved_start;
2280 unsigned int last_timestamp;
2282 boolean_t need_wakeup, result, user_unwire;
2285 return (KERN_SUCCESS);
2286 user_unwire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE;
2288 VM_MAP_RANGE_CHECK(map, start, end);
2289 if (!vm_map_lookup_entry(map, start, &first_entry)) {
2290 if (flags & VM_MAP_WIRE_HOLESOK)
2291 first_entry = first_entry->next;
2294 return (KERN_INVALID_ADDRESS);
2297 last_timestamp = map->timestamp;
2298 entry = first_entry;
2299 while (entry != &map->header && entry->start < end) {
2300 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
2302 * We have not yet clipped the entry.
2304 saved_start = (start >= entry->start) ? start :
2306 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2307 if (vm_map_unlock_and_wait(map, 0)) {
2309 * Allow interruption of user unwiring?
2313 if (last_timestamp+1 != map->timestamp) {
2315 * Look again for the entry because the map was
2316 * modified while it was unlocked.
2317 * Specifically, the entry may have been
2318 * clipped, merged, or deleted.
2320 if (!vm_map_lookup_entry(map, saved_start,
2322 if (flags & VM_MAP_WIRE_HOLESOK)
2323 tmp_entry = tmp_entry->next;
2325 if (saved_start == start) {
2327 * First_entry has been deleted.
2330 return (KERN_INVALID_ADDRESS);
2333 rv = KERN_INVALID_ADDRESS;
2337 if (entry == first_entry)
2338 first_entry = tmp_entry;
2343 last_timestamp = map->timestamp;
2346 vm_map_clip_start(map, entry, start);
2347 vm_map_clip_end(map, entry, end);
2349 * Mark the entry in case the map lock is released. (See
2352 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 &&
2353 entry->wiring_thread == NULL,
2354 ("owned map entry %p", entry));
2355 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
2356 entry->wiring_thread = curthread;
2358 * Check the map for holes in the specified region.
2359 * If VM_MAP_WIRE_HOLESOK was specified, skip this check.
2361 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) &&
2362 (entry->end < end && (entry->next == &map->header ||
2363 entry->next->start > entry->end))) {
2365 rv = KERN_INVALID_ADDRESS;
2369 * If system unwiring, require that the entry is system wired.
2372 vm_map_entry_system_wired_count(entry) == 0) {
2374 rv = KERN_INVALID_ARGUMENT;
2377 entry = entry->next;
2381 need_wakeup = FALSE;
2382 if (first_entry == NULL) {
2383 result = vm_map_lookup_entry(map, start, &first_entry);
2384 if (!result && (flags & VM_MAP_WIRE_HOLESOK))
2385 first_entry = first_entry->next;
2387 KASSERT(result, ("vm_map_unwire: lookup failed"));
2389 for (entry = first_entry; entry != &map->header && entry->start < end;
2390 entry = entry->next) {
2392 * If VM_MAP_WIRE_HOLESOK was specified, an empty
2393 * space in the unwired region could have been mapped
2394 * while the map lock was dropped for draining
2395 * MAP_ENTRY_IN_TRANSITION. Moreover, another thread
2396 * could be simultaneously wiring this new mapping
2397 * entry. Detect these cases and skip any entries
2398 * marked as in transition by us.
2400 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 ||
2401 entry->wiring_thread != curthread) {
2402 KASSERT((flags & VM_MAP_WIRE_HOLESOK) != 0,
2403 ("vm_map_unwire: !HOLESOK and new/changed entry"));
2407 if (rv == KERN_SUCCESS && (!user_unwire ||
2408 (entry->eflags & MAP_ENTRY_USER_WIRED))) {
2410 entry->eflags &= ~MAP_ENTRY_USER_WIRED;
2411 if (entry->wired_count == 1)
2412 vm_map_entry_unwire(map, entry);
2414 entry->wired_count--;
2416 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0,
2417 ("vm_map_unwire: in-transition flag missing %p", entry));
2418 KASSERT(entry->wiring_thread == curthread,
2419 ("vm_map_unwire: alien wire %p", entry));
2420 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION;
2421 entry->wiring_thread = NULL;
2422 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
2423 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
2426 vm_map_simplify_entry(map, entry);
2435 * vm_map_wire_entry_failure:
2437 * Handle a wiring failure on the given entry.
2439 * The map should be locked.
2442 vm_map_wire_entry_failure(vm_map_t map, vm_map_entry_t entry,
2443 vm_offset_t failed_addr)
2446 VM_MAP_ASSERT_LOCKED(map);
2447 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 &&
2448 entry->wired_count == 1,
2449 ("vm_map_wire_entry_failure: entry %p isn't being wired", entry));
2450 KASSERT(failed_addr < entry->end,
2451 ("vm_map_wire_entry_failure: entry %p was fully wired", entry));
2454 * If any pages at the start of this entry were successfully wired,
2457 if (failed_addr > entry->start) {
2458 pmap_unwire(map->pmap, entry->start, failed_addr);
2459 vm_object_unwire(entry->object.vm_object, entry->offset,
2460 failed_addr - entry->start, PQ_ACTIVE);
2464 * Assign an out-of-range value to represent the failure to wire this
2467 entry->wired_count = -1;
2473 * Implements both kernel and user wiring.
2476 vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t end,
2479 vm_map_entry_t entry, first_entry, tmp_entry;
2480 vm_offset_t faddr, saved_end, saved_start;
2481 unsigned int last_timestamp;
2483 boolean_t need_wakeup, result, user_wire;
2487 return (KERN_SUCCESS);
2489 if (flags & VM_MAP_WIRE_WRITE)
2490 prot |= VM_PROT_WRITE;
2491 user_wire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE;
2493 VM_MAP_RANGE_CHECK(map, start, end);
2494 if (!vm_map_lookup_entry(map, start, &first_entry)) {
2495 if (flags & VM_MAP_WIRE_HOLESOK)
2496 first_entry = first_entry->next;
2499 return (KERN_INVALID_ADDRESS);
2502 last_timestamp = map->timestamp;
2503 entry = first_entry;
2504 while (entry != &map->header && entry->start < end) {
2505 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
2507 * We have not yet clipped the entry.
2509 saved_start = (start >= entry->start) ? start :
2511 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2512 if (vm_map_unlock_and_wait(map, 0)) {
2514 * Allow interruption of user wiring?
2518 if (last_timestamp + 1 != map->timestamp) {
2520 * Look again for the entry because the map was
2521 * modified while it was unlocked.
2522 * Specifically, the entry may have been
2523 * clipped, merged, or deleted.
2525 if (!vm_map_lookup_entry(map, saved_start,
2527 if (flags & VM_MAP_WIRE_HOLESOK)
2528 tmp_entry = tmp_entry->next;
2530 if (saved_start == start) {
2532 * first_entry has been deleted.
2535 return (KERN_INVALID_ADDRESS);
2538 rv = KERN_INVALID_ADDRESS;
2542 if (entry == first_entry)
2543 first_entry = tmp_entry;
2548 last_timestamp = map->timestamp;
2551 vm_map_clip_start(map, entry, start);
2552 vm_map_clip_end(map, entry, end);
2554 * Mark the entry in case the map lock is released. (See
2557 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 &&
2558 entry->wiring_thread == NULL,
2559 ("owned map entry %p", entry));
2560 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
2561 entry->wiring_thread = curthread;
2562 if ((entry->protection & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0
2563 || (entry->protection & prot) != prot) {
2564 entry->eflags |= MAP_ENTRY_WIRE_SKIPPED;
2565 if ((flags & VM_MAP_WIRE_HOLESOK) == 0) {
2567 rv = KERN_INVALID_ADDRESS;
2572 if (entry->wired_count == 0) {
2573 entry->wired_count++;
2574 saved_start = entry->start;
2575 saved_end = entry->end;
2578 * Release the map lock, relying on the in-transition
2579 * mark. Mark the map busy for fork.
2584 faddr = saved_start;
2587 * Simulate a fault to get the page and enter
2588 * it into the physical map.
2590 if ((rv = vm_fault(map, faddr, VM_PROT_NONE,
2591 VM_FAULT_CHANGE_WIRING)) != KERN_SUCCESS)
2593 } while ((faddr += PAGE_SIZE) < saved_end);
2596 if (last_timestamp + 1 != map->timestamp) {
2598 * Look again for the entry because the map was
2599 * modified while it was unlocked. The entry
2600 * may have been clipped, but NOT merged or
2603 result = vm_map_lookup_entry(map, saved_start,
2605 KASSERT(result, ("vm_map_wire: lookup failed"));
2606 if (entry == first_entry)
2607 first_entry = tmp_entry;
2611 while (entry->end < saved_end) {
2613 * In case of failure, handle entries
2614 * that were not fully wired here;
2615 * fully wired entries are handled
2618 if (rv != KERN_SUCCESS &&
2620 vm_map_wire_entry_failure(map,
2622 entry = entry->next;
2625 last_timestamp = map->timestamp;
2626 if (rv != KERN_SUCCESS) {
2627 vm_map_wire_entry_failure(map, entry, faddr);
2631 } else if (!user_wire ||
2632 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
2633 entry->wired_count++;
2636 * Check the map for holes in the specified region.
2637 * If VM_MAP_WIRE_HOLESOK was specified, skip this check.
2640 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) &&
2641 (entry->end < end && (entry->next == &map->header ||
2642 entry->next->start > entry->end))) {
2644 rv = KERN_INVALID_ADDRESS;
2647 entry = entry->next;
2651 need_wakeup = FALSE;
2652 if (first_entry == NULL) {
2653 result = vm_map_lookup_entry(map, start, &first_entry);
2654 if (!result && (flags & VM_MAP_WIRE_HOLESOK))
2655 first_entry = first_entry->next;
2657 KASSERT(result, ("vm_map_wire: lookup failed"));
2659 for (entry = first_entry; entry != &map->header && entry->start < end;
2660 entry = entry->next) {
2661 if ((entry->eflags & MAP_ENTRY_WIRE_SKIPPED) != 0)
2662 goto next_entry_done;
2665 * If VM_MAP_WIRE_HOLESOK was specified, an empty
2666 * space in the unwired region could have been mapped
2667 * while the map lock was dropped for faulting in the
2668 * pages or draining MAP_ENTRY_IN_TRANSITION.
2669 * Moreover, another thread could be simultaneously
2670 * wiring this new mapping entry. Detect these cases
2671 * and skip any entries marked as in transition by us.
2673 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 ||
2674 entry->wiring_thread != curthread) {
2675 KASSERT((flags & VM_MAP_WIRE_HOLESOK) != 0,
2676 ("vm_map_wire: !HOLESOK and new/changed entry"));
2680 if (rv == KERN_SUCCESS) {
2682 entry->eflags |= MAP_ENTRY_USER_WIRED;
2683 } else if (entry->wired_count == -1) {
2685 * Wiring failed on this entry. Thus, unwiring is
2688 entry->wired_count = 0;
2689 } else if (!user_wire ||
2690 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
2692 * Undo the wiring. Wiring succeeded on this entry
2693 * but failed on a later entry.
2695 if (entry->wired_count == 1)
2696 vm_map_entry_unwire(map, entry);
2698 entry->wired_count--;
2701 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0,
2702 ("vm_map_wire: in-transition flag missing %p", entry));
2703 KASSERT(entry->wiring_thread == curthread,
2704 ("vm_map_wire: alien wire %p", entry));
2705 entry->eflags &= ~(MAP_ENTRY_IN_TRANSITION |
2706 MAP_ENTRY_WIRE_SKIPPED);
2707 entry->wiring_thread = NULL;
2708 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
2709 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
2712 vm_map_simplify_entry(map, entry);
2723 * Push any dirty cached pages in the address range to their pager.
2724 * If syncio is TRUE, dirty pages are written synchronously.
2725 * If invalidate is TRUE, any cached pages are freed as well.
2727 * If the size of the region from start to end is zero, we are
2728 * supposed to flush all modified pages within the region containing
2729 * start. Unfortunately, a region can be split or coalesced with
2730 * neighboring regions, making it difficult to determine what the
2731 * original region was. Therefore, we approximate this requirement by
2732 * flushing the current region containing start.
2734 * Returns an error if any part of the specified range is not mapped.
2742 boolean_t invalidate)
2744 vm_map_entry_t current;
2745 vm_map_entry_t entry;
2748 vm_ooffset_t offset;
2749 unsigned int last_timestamp;
2752 vm_map_lock_read(map);
2753 VM_MAP_RANGE_CHECK(map, start, end);
2754 if (!vm_map_lookup_entry(map, start, &entry)) {
2755 vm_map_unlock_read(map);
2756 return (KERN_INVALID_ADDRESS);
2757 } else if (start == end) {
2758 start = entry->start;
2762 * Make a first pass to check for user-wired memory and holes.
2764 for (current = entry; current != &map->header && current->start < end;
2765 current = current->next) {
2766 if (invalidate && (current->eflags & MAP_ENTRY_USER_WIRED)) {
2767 vm_map_unlock_read(map);
2768 return (KERN_INVALID_ARGUMENT);
2770 if (end > current->end &&
2771 (current->next == &map->header ||
2772 current->end != current->next->start)) {
2773 vm_map_unlock_read(map);
2774 return (KERN_INVALID_ADDRESS);
2779 pmap_remove(map->pmap, start, end);
2783 * Make a second pass, cleaning/uncaching pages from the indicated
2786 for (current = entry; current != &map->header && current->start < end;) {
2787 offset = current->offset + (start - current->start);
2788 size = (end <= current->end ? end : current->end) - start;
2789 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
2791 vm_map_entry_t tentry;
2794 smap = current->object.sub_map;
2795 vm_map_lock_read(smap);
2796 (void) vm_map_lookup_entry(smap, offset, &tentry);
2797 tsize = tentry->end - offset;
2800 object = tentry->object.vm_object;
2801 offset = tentry->offset + (offset - tentry->start);
2802 vm_map_unlock_read(smap);
2804 object = current->object.vm_object;
2806 vm_object_reference(object);
2807 last_timestamp = map->timestamp;
2808 vm_map_unlock_read(map);
2809 if (!vm_object_sync(object, offset, size, syncio, invalidate))
2812 vm_object_deallocate(object);
2813 vm_map_lock_read(map);
2814 if (last_timestamp == map->timestamp ||
2815 !vm_map_lookup_entry(map, start, ¤t))
2816 current = current->next;
2819 vm_map_unlock_read(map);
2820 return (failed ? KERN_FAILURE : KERN_SUCCESS);
2824 * vm_map_entry_unwire: [ internal use only ]
2826 * Make the region specified by this entry pageable.
2828 * The map in question should be locked.
2829 * [This is the reason for this routine's existence.]
2832 vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry)
2835 VM_MAP_ASSERT_LOCKED(map);
2836 KASSERT(entry->wired_count > 0,
2837 ("vm_map_entry_unwire: entry %p isn't wired", entry));
2838 pmap_unwire(map->pmap, entry->start, entry->end);
2839 vm_object_unwire(entry->object.vm_object, entry->offset, entry->end -
2840 entry->start, PQ_ACTIVE);
2841 entry->wired_count = 0;
2845 vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map)
2848 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0)
2849 vm_object_deallocate(entry->object.vm_object);
2850 uma_zfree(system_map ? kmapentzone : mapentzone, entry);
2854 * vm_map_entry_delete: [ internal use only ]
2856 * Deallocate the given entry from the target map.
2859 vm_map_entry_delete(vm_map_t map, vm_map_entry_t entry)
2862 vm_pindex_t offidxstart, offidxend, count, size1;
2865 vm_map_entry_unlink(map, entry);
2866 object = entry->object.vm_object;
2867 size = entry->end - entry->start;
2870 if (entry->cred != NULL) {
2871 swap_release_by_cred(size, entry->cred);
2872 crfree(entry->cred);
2875 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0 &&
2877 KASSERT(entry->cred == NULL || object->cred == NULL ||
2878 (entry->eflags & MAP_ENTRY_NEEDS_COPY),
2879 ("OVERCOMMIT vm_map_entry_delete: both cred %p", entry));
2880 count = OFF_TO_IDX(size);
2881 offidxstart = OFF_TO_IDX(entry->offset);
2882 offidxend = offidxstart + count;
2883 VM_OBJECT_WLOCK(object);
2884 if (object->ref_count != 1 &&
2885 ((object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING ||
2886 object == kernel_object || object == kmem_object)) {
2887 vm_object_collapse(object);
2890 * The option OBJPR_NOTMAPPED can be passed here
2891 * because vm_map_delete() already performed
2892 * pmap_remove() on the only mapping to this range
2895 vm_object_page_remove(object, offidxstart, offidxend,
2897 if (object->type == OBJT_SWAP)
2898 swap_pager_freespace(object, offidxstart, count);
2899 if (offidxend >= object->size &&
2900 offidxstart < object->size) {
2901 size1 = object->size;
2902 object->size = offidxstart;
2903 if (object->cred != NULL) {
2904 size1 -= object->size;
2905 KASSERT(object->charge >= ptoa(size1),
2906 ("vm_map_entry_delete: object->charge < 0"));
2907 swap_release_by_cred(ptoa(size1), object->cred);
2908 object->charge -= ptoa(size1);
2912 VM_OBJECT_WUNLOCK(object);
2914 entry->object.vm_object = NULL;
2915 if (map->system_map)
2916 vm_map_entry_deallocate(entry, TRUE);
2918 entry->next = curthread->td_map_def_user;
2919 curthread->td_map_def_user = entry;
2924 * vm_map_delete: [ internal use only ]
2926 * Deallocates the given address range from the target
2930 vm_map_delete(vm_map_t map, vm_offset_t start, vm_offset_t end)
2932 vm_map_entry_t entry;
2933 vm_map_entry_t first_entry;
2935 VM_MAP_ASSERT_LOCKED(map);
2937 return (KERN_SUCCESS);
2940 * Find the start of the region, and clip it
2942 if (!vm_map_lookup_entry(map, start, &first_entry))
2943 entry = first_entry->next;
2945 entry = first_entry;
2946 vm_map_clip_start(map, entry, start);
2950 * Step through all entries in this region
2952 while ((entry != &map->header) && (entry->start < end)) {
2953 vm_map_entry_t next;
2956 * Wait for wiring or unwiring of an entry to complete.
2957 * Also wait for any system wirings to disappear on
2960 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 ||
2961 (vm_map_pmap(map) != kernel_pmap &&
2962 vm_map_entry_system_wired_count(entry) != 0)) {
2963 unsigned int last_timestamp;
2964 vm_offset_t saved_start;
2965 vm_map_entry_t tmp_entry;
2967 saved_start = entry->start;
2968 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2969 last_timestamp = map->timestamp;
2970 (void) vm_map_unlock_and_wait(map, 0);
2972 if (last_timestamp + 1 != map->timestamp) {
2974 * Look again for the entry because the map was
2975 * modified while it was unlocked.
2976 * Specifically, the entry may have been
2977 * clipped, merged, or deleted.
2979 if (!vm_map_lookup_entry(map, saved_start,
2981 entry = tmp_entry->next;
2984 vm_map_clip_start(map, entry,
2990 vm_map_clip_end(map, entry, end);
2995 * Unwire before removing addresses from the pmap; otherwise,
2996 * unwiring will put the entries back in the pmap.
2998 if (entry->wired_count != 0) {
2999 vm_map_entry_unwire(map, entry);
3002 pmap_remove(map->pmap, entry->start, entry->end);
3005 * Delete the entry only after removing all pmap
3006 * entries pointing to its pages. (Otherwise, its
3007 * page frames may be reallocated, and any modify bits
3008 * will be set in the wrong object!)
3010 vm_map_entry_delete(map, entry);
3013 return (KERN_SUCCESS);
3019 * Remove the given address range from the target map.
3020 * This is the exported form of vm_map_delete.
3023 vm_map_remove(vm_map_t map, vm_offset_t start, vm_offset_t end)
3028 VM_MAP_RANGE_CHECK(map, start, end);
3029 result = vm_map_delete(map, start, end);
3035 * vm_map_check_protection:
3037 * Assert that the target map allows the specified privilege on the
3038 * entire address region given. The entire region must be allocated.
3040 * WARNING! This code does not and should not check whether the
3041 * contents of the region is accessible. For example a smaller file
3042 * might be mapped into a larger address space.
3044 * NOTE! This code is also called by munmap().
3046 * The map must be locked. A read lock is sufficient.
3049 vm_map_check_protection(vm_map_t map, vm_offset_t start, vm_offset_t end,
3050 vm_prot_t protection)
3052 vm_map_entry_t entry;
3053 vm_map_entry_t tmp_entry;
3055 if (!vm_map_lookup_entry(map, start, &tmp_entry))
3059 while (start < end) {
3060 if (entry == &map->header)
3065 if (start < entry->start)
3068 * Check protection associated with entry.
3070 if ((entry->protection & protection) != protection)
3072 /* go to next entry */
3074 entry = entry->next;
3080 * vm_map_copy_entry:
3082 * Copies the contents of the source entry to the destination
3083 * entry. The entries *must* be aligned properly.
3089 vm_map_entry_t src_entry,
3090 vm_map_entry_t dst_entry,
3091 vm_ooffset_t *fork_charge)
3093 vm_object_t src_object;
3094 vm_map_entry_t fake_entry;
3099 VM_MAP_ASSERT_LOCKED(dst_map);
3101 if ((dst_entry->eflags|src_entry->eflags) & MAP_ENTRY_IS_SUB_MAP)
3104 if (src_entry->wired_count == 0 ||
3105 (src_entry->protection & VM_PROT_WRITE) == 0) {
3107 * If the source entry is marked needs_copy, it is already
3110 if ((src_entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0 &&
3111 (src_entry->protection & VM_PROT_WRITE) != 0) {
3112 pmap_protect(src_map->pmap,
3115 src_entry->protection & ~VM_PROT_WRITE);
3119 * Make a copy of the object.
3121 size = src_entry->end - src_entry->start;
3122 if ((src_object = src_entry->object.vm_object) != NULL) {
3123 VM_OBJECT_WLOCK(src_object);
3124 charged = ENTRY_CHARGED(src_entry);
3125 if ((src_object->handle == NULL) &&
3126 (src_object->type == OBJT_DEFAULT ||
3127 src_object->type == OBJT_SWAP)) {
3128 vm_object_collapse(src_object);
3129 if ((src_object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING) {
3130 vm_object_split(src_entry);
3131 src_object = src_entry->object.vm_object;
3134 vm_object_reference_locked(src_object);
3135 vm_object_clear_flag(src_object, OBJ_ONEMAPPING);
3136 if (src_entry->cred != NULL &&
3137 !(src_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
3138 KASSERT(src_object->cred == NULL,
3139 ("OVERCOMMIT: vm_map_copy_entry: cred %p",
3141 src_object->cred = src_entry->cred;
3142 src_object->charge = size;
3144 VM_OBJECT_WUNLOCK(src_object);
3145 dst_entry->object.vm_object = src_object;
3147 cred = curthread->td_ucred;
3149 dst_entry->cred = cred;
3150 *fork_charge += size;
3151 if (!(src_entry->eflags &
3152 MAP_ENTRY_NEEDS_COPY)) {
3154 src_entry->cred = cred;
3155 *fork_charge += size;
3158 src_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY);
3159 dst_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY);
3160 dst_entry->offset = src_entry->offset;
3161 if (src_entry->eflags & MAP_ENTRY_VN_WRITECNT) {
3163 * MAP_ENTRY_VN_WRITECNT cannot
3164 * indicate write reference from
3165 * src_entry, since the entry is
3166 * marked as needs copy. Allocate a
3167 * fake entry that is used to
3168 * decrement object->un_pager.vnp.writecount
3169 * at the appropriate time. Attach
3170 * fake_entry to the deferred list.
3172 fake_entry = vm_map_entry_create(dst_map);
3173 fake_entry->eflags = MAP_ENTRY_VN_WRITECNT;
3174 src_entry->eflags &= ~MAP_ENTRY_VN_WRITECNT;
3175 vm_object_reference(src_object);
3176 fake_entry->object.vm_object = src_object;
3177 fake_entry->start = src_entry->start;
3178 fake_entry->end = src_entry->end;
3179 fake_entry->next = curthread->td_map_def_user;
3180 curthread->td_map_def_user = fake_entry;
3183 dst_entry->object.vm_object = NULL;
3184 dst_entry->offset = 0;
3185 if (src_entry->cred != NULL) {
3186 dst_entry->cred = curthread->td_ucred;
3187 crhold(dst_entry->cred);
3188 *fork_charge += size;
3192 pmap_copy(dst_map->pmap, src_map->pmap, dst_entry->start,
3193 dst_entry->end - dst_entry->start, src_entry->start);
3196 * We don't want to make writeable wired pages copy-on-write.
3197 * Immediately copy these pages into the new map by simulating
3198 * page faults. The new pages are pageable.
3200 vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry,
3206 * vmspace_map_entry_forked:
3207 * Update the newly-forked vmspace each time a map entry is inherited
3208 * or copied. The values for vm_dsize and vm_tsize are approximate
3209 * (and mostly-obsolete ideas in the face of mmap(2) et al.)
3212 vmspace_map_entry_forked(const struct vmspace *vm1, struct vmspace *vm2,
3213 vm_map_entry_t entry)
3215 vm_size_t entrysize;
3218 entrysize = entry->end - entry->start;
3219 vm2->vm_map.size += entrysize;
3220 if (entry->eflags & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP)) {
3221 vm2->vm_ssize += btoc(entrysize);
3222 } else if (entry->start >= (vm_offset_t)vm1->vm_daddr &&
3223 entry->start < (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)) {
3224 newend = MIN(entry->end,
3225 (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize));
3226 vm2->vm_dsize += btoc(newend - entry->start);
3227 } else if (entry->start >= (vm_offset_t)vm1->vm_taddr &&
3228 entry->start < (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)) {
3229 newend = MIN(entry->end,
3230 (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize));
3231 vm2->vm_tsize += btoc(newend - entry->start);
3237 * Create a new process vmspace structure and vm_map
3238 * based on those of an existing process. The new map
3239 * is based on the old map, according to the inheritance
3240 * values on the regions in that map.
3242 * XXX It might be worth coalescing the entries added to the new vmspace.
3244 * The source map must not be locked.
3247 vmspace_fork(struct vmspace *vm1, vm_ooffset_t *fork_charge)
3249 struct vmspace *vm2;
3250 vm_map_t new_map, old_map;
3251 vm_map_entry_t new_entry, old_entry;
3255 old_map = &vm1->vm_map;
3256 /* Copy immutable fields of vm1 to vm2. */
3257 vm2 = vmspace_alloc(old_map->min_offset, old_map->max_offset, NULL);
3260 vm2->vm_taddr = vm1->vm_taddr;
3261 vm2->vm_daddr = vm1->vm_daddr;
3262 vm2->vm_maxsaddr = vm1->vm_maxsaddr;
3263 vm_map_lock(old_map);
3265 vm_map_wait_busy(old_map);
3266 new_map = &vm2->vm_map;
3267 locked = vm_map_trylock(new_map); /* trylock to silence WITNESS */
3268 KASSERT(locked, ("vmspace_fork: lock failed"));
3270 old_entry = old_map->header.next;
3272 while (old_entry != &old_map->header) {
3273 if (old_entry->eflags & MAP_ENTRY_IS_SUB_MAP)
3274 panic("vm_map_fork: encountered a submap");
3276 switch (old_entry->inheritance) {
3277 case VM_INHERIT_NONE:
3280 case VM_INHERIT_SHARE:
3282 * Clone the entry, creating the shared object if necessary.
3284 object = old_entry->object.vm_object;
3285 if (object == NULL) {
3286 object = vm_object_allocate(OBJT_DEFAULT,
3287 atop(old_entry->end - old_entry->start));
3288 old_entry->object.vm_object = object;
3289 old_entry->offset = 0;
3290 if (old_entry->cred != NULL) {
3291 object->cred = old_entry->cred;
3292 object->charge = old_entry->end -
3294 old_entry->cred = NULL;
3299 * Add the reference before calling vm_object_shadow
3300 * to insure that a shadow object is created.
3302 vm_object_reference(object);
3303 if (old_entry->eflags & MAP_ENTRY_NEEDS_COPY) {
3304 vm_object_shadow(&old_entry->object.vm_object,
3306 old_entry->end - old_entry->start);
3307 old_entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
3308 /* Transfer the second reference too. */
3309 vm_object_reference(
3310 old_entry->object.vm_object);
3313 * As in vm_map_simplify_entry(), the
3314 * vnode lock will not be acquired in
3315 * this call to vm_object_deallocate().
3317 vm_object_deallocate(object);
3318 object = old_entry->object.vm_object;
3320 VM_OBJECT_WLOCK(object);
3321 vm_object_clear_flag(object, OBJ_ONEMAPPING);
3322 if (old_entry->cred != NULL) {
3323 KASSERT(object->cred == NULL, ("vmspace_fork both cred"));
3324 object->cred = old_entry->cred;
3325 object->charge = old_entry->end - old_entry->start;
3326 old_entry->cred = NULL;
3330 * Assert the correct state of the vnode
3331 * v_writecount while the object is locked, to
3332 * not relock it later for the assertion
3335 if (old_entry->eflags & MAP_ENTRY_VN_WRITECNT &&
3336 object->type == OBJT_VNODE) {
3337 KASSERT(((struct vnode *)object->handle)->
3339 ("vmspace_fork: v_writecount %p", object));
3340 KASSERT(object->un_pager.vnp.writemappings > 0,
3341 ("vmspace_fork: vnp.writecount %p",
3344 VM_OBJECT_WUNLOCK(object);
3347 * Clone the entry, referencing the shared object.
3349 new_entry = vm_map_entry_create(new_map);
3350 *new_entry = *old_entry;
3351 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED |
3352 MAP_ENTRY_IN_TRANSITION);
3353 new_entry->wiring_thread = NULL;
3354 new_entry->wired_count = 0;
3355 if (new_entry->eflags & MAP_ENTRY_VN_WRITECNT) {
3356 vnode_pager_update_writecount(object,
3357 new_entry->start, new_entry->end);
3361 * Insert the entry into the new map -- we know we're
3362 * inserting at the end of the new map.
3364 vm_map_entry_link(new_map, new_map->header.prev,
3366 vmspace_map_entry_forked(vm1, vm2, new_entry);
3369 * Update the physical map
3371 pmap_copy(new_map->pmap, old_map->pmap,
3373 (old_entry->end - old_entry->start),
3377 case VM_INHERIT_COPY:
3379 * Clone the entry and link into the map.
3381 new_entry = vm_map_entry_create(new_map);
3382 *new_entry = *old_entry;
3384 * Copied entry is COW over the old object.
3386 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED |
3387 MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_VN_WRITECNT);
3388 new_entry->wiring_thread = NULL;
3389 new_entry->wired_count = 0;
3390 new_entry->object.vm_object = NULL;
3391 new_entry->cred = NULL;
3392 vm_map_entry_link(new_map, new_map->header.prev,
3394 vmspace_map_entry_forked(vm1, vm2, new_entry);
3395 vm_map_copy_entry(old_map, new_map, old_entry,
3396 new_entry, fork_charge);
3399 old_entry = old_entry->next;
3402 * Use inlined vm_map_unlock() to postpone handling the deferred
3403 * map entries, which cannot be done until both old_map and
3404 * new_map locks are released.
3406 sx_xunlock(&old_map->lock);
3407 sx_xunlock(&new_map->lock);
3408 vm_map_process_deferred();
3414 vm_map_stack(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
3415 vm_prot_t prot, vm_prot_t max, int cow)
3417 vm_size_t growsize, init_ssize;
3418 rlim_t lmemlim, vmemlim;
3421 growsize = sgrowsiz;
3422 init_ssize = (max_ssize < growsize) ? max_ssize : growsize;
3425 lmemlim = lim_cur(curproc, RLIMIT_MEMLOCK);
3426 vmemlim = lim_cur(curproc, RLIMIT_VMEM);
3427 PROC_UNLOCK(curproc);
3428 if (!old_mlock && map->flags & MAP_WIREFUTURE) {
3429 if (ptoa(pmap_wired_count(map->pmap)) + init_ssize > lmemlim) {
3434 /* If we would blow our VMEM resource limit, no go */
3435 if (map->size + init_ssize > vmemlim) {
3439 rv = vm_map_stack_locked(map, addrbos, max_ssize, growsize, prot,
3447 vm_map_stack_locked(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
3448 vm_size_t growsize, vm_prot_t prot, vm_prot_t max, int cow)
3450 vm_map_entry_t new_entry, prev_entry;
3451 vm_offset_t bot, top;
3452 vm_size_t init_ssize;
3456 * The stack orientation is piggybacked with the cow argument.
3457 * Extract it into orient and mask the cow argument so that we
3458 * don't pass it around further.
3459 * NOTE: We explicitly allow bi-directional stacks.
3461 orient = cow & (MAP_STACK_GROWS_DOWN|MAP_STACK_GROWS_UP);
3462 KASSERT(orient != 0, ("No stack grow direction"));
3464 if (addrbos < vm_map_min(map) ||
3465 addrbos > vm_map_max(map) ||
3466 addrbos + max_ssize < addrbos)
3467 return (KERN_NO_SPACE);
3469 init_ssize = (max_ssize < growsize) ? max_ssize : growsize;
3471 /* If addr is already mapped, no go */
3472 if (vm_map_lookup_entry(map, addrbos, &prev_entry))
3473 return (KERN_NO_SPACE);
3476 * If we can't accomodate max_ssize in the current mapping, no go.
3477 * However, we need to be aware that subsequent user mappings might
3478 * map into the space we have reserved for stack, and currently this
3479 * space is not protected.
3481 * Hopefully we will at least detect this condition when we try to
3484 if ((prev_entry->next != &map->header) &&
3485 (prev_entry->next->start < addrbos + max_ssize))
3486 return (KERN_NO_SPACE);
3489 * We initially map a stack of only init_ssize. We will grow as
3490 * needed later. Depending on the orientation of the stack (i.e.
3491 * the grow direction) we either map at the top of the range, the
3492 * bottom of the range or in the middle.
3494 * Note: we would normally expect prot and max to be VM_PROT_ALL,
3495 * and cow to be 0. Possibly we should eliminate these as input
3496 * parameters, and just pass these values here in the insert call.
3498 if (orient == MAP_STACK_GROWS_DOWN)
3499 bot = addrbos + max_ssize - init_ssize;
3500 else if (orient == MAP_STACK_GROWS_UP)
3503 bot = round_page(addrbos + max_ssize/2 - init_ssize/2);
3504 top = bot + init_ssize;
3505 rv = vm_map_insert(map, NULL, 0, bot, top, prot, max, cow);
3507 /* Now set the avail_ssize amount. */
3508 if (rv == KERN_SUCCESS) {
3509 new_entry = prev_entry->next;
3510 if (new_entry->end != top || new_entry->start != bot)
3511 panic("Bad entry start/end for new stack entry");
3513 new_entry->avail_ssize = max_ssize - init_ssize;
3514 KASSERT((orient & MAP_STACK_GROWS_DOWN) == 0 ||
3515 (new_entry->eflags & MAP_ENTRY_GROWS_DOWN) != 0,
3516 ("new entry lacks MAP_ENTRY_GROWS_DOWN"));
3517 KASSERT((orient & MAP_STACK_GROWS_UP) == 0 ||
3518 (new_entry->eflags & MAP_ENTRY_GROWS_UP) != 0,
3519 ("new entry lacks MAP_ENTRY_GROWS_UP"));
3525 static int stack_guard_page = 0;
3526 SYSCTL_INT(_security_bsd, OID_AUTO, stack_guard_page, CTLFLAG_RWTUN,
3527 &stack_guard_page, 0,
3528 "Insert stack guard page ahead of the growable segments.");
3530 /* Attempts to grow a vm stack entry. Returns KERN_SUCCESS if the
3531 * desired address is already mapped, or if we successfully grow
3532 * the stack. Also returns KERN_SUCCESS if addr is outside the
3533 * stack range (this is strange, but preserves compatibility with
3534 * the grow function in vm_machdep.c).
3537 vm_map_growstack(struct proc *p, vm_offset_t addr)
3539 vm_map_entry_t next_entry, prev_entry;
3540 vm_map_entry_t new_entry, stack_entry;
3541 struct vmspace *vm = p->p_vmspace;
3542 vm_map_t map = &vm->vm_map;
3545 size_t grow_amount, max_grow;
3546 rlim_t lmemlim, stacklim, vmemlim;
3547 int is_procstack, rv;
3558 lmemlim = lim_cur(p, RLIMIT_MEMLOCK);
3559 stacklim = lim_cur(p, RLIMIT_STACK);
3560 vmemlim = lim_cur(p, RLIMIT_VMEM);
3563 vm_map_lock_read(map);
3565 /* If addr is already in the entry range, no need to grow.*/
3566 if (vm_map_lookup_entry(map, addr, &prev_entry)) {
3567 vm_map_unlock_read(map);
3568 return (KERN_SUCCESS);
3571 next_entry = prev_entry->next;
3572 if (!(prev_entry->eflags & MAP_ENTRY_GROWS_UP)) {
3574 * This entry does not grow upwards. Since the address lies
3575 * beyond this entry, the next entry (if one exists) has to
3576 * be a downward growable entry. The entry list header is
3577 * never a growable entry, so it suffices to check the flags.
3579 if (!(next_entry->eflags & MAP_ENTRY_GROWS_DOWN)) {
3580 vm_map_unlock_read(map);
3581 return (KERN_SUCCESS);
3583 stack_entry = next_entry;
3586 * This entry grows upward. If the next entry does not at
3587 * least grow downwards, this is the entry we need to grow.
3588 * otherwise we have two possible choices and we have to
3591 if (next_entry->eflags & MAP_ENTRY_GROWS_DOWN) {
3593 * We have two choices; grow the entry closest to
3594 * the address to minimize the amount of growth.
3596 if (addr - prev_entry->end <= next_entry->start - addr)
3597 stack_entry = prev_entry;
3599 stack_entry = next_entry;
3601 stack_entry = prev_entry;
3604 if (stack_entry == next_entry) {
3605 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_DOWN, ("foo"));
3606 KASSERT(addr < stack_entry->start, ("foo"));
3607 end = (prev_entry != &map->header) ? prev_entry->end :
3608 stack_entry->start - stack_entry->avail_ssize;
3609 grow_amount = roundup(stack_entry->start - addr, PAGE_SIZE);
3610 max_grow = stack_entry->start - end;
3612 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_UP, ("foo"));
3613 KASSERT(addr >= stack_entry->end, ("foo"));
3614 end = (next_entry != &map->header) ? next_entry->start :
3615 stack_entry->end + stack_entry->avail_ssize;
3616 grow_amount = roundup(addr + 1 - stack_entry->end, PAGE_SIZE);
3617 max_grow = end - stack_entry->end;
3620 if (grow_amount > stack_entry->avail_ssize) {
3621 vm_map_unlock_read(map);
3622 return (KERN_NO_SPACE);
3626 * If there is no longer enough space between the entries nogo, and
3627 * adjust the available space. Note: this should only happen if the
3628 * user has mapped into the stack area after the stack was created,
3629 * and is probably an error.
3631 * This also effectively destroys any guard page the user might have
3632 * intended by limiting the stack size.
3634 if (grow_amount + (stack_guard_page ? PAGE_SIZE : 0) > max_grow) {
3635 if (vm_map_lock_upgrade(map))
3638 stack_entry->avail_ssize = max_grow;
3641 return (KERN_NO_SPACE);
3644 is_procstack = (addr >= (vm_offset_t)vm->vm_maxsaddr) ? 1 : 0;
3647 * If this is the main process stack, see if we're over the stack
3650 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) {
3651 vm_map_unlock_read(map);
3652 return (KERN_NO_SPACE);
3657 racct_set(p, RACCT_STACK, ctob(vm->vm_ssize) + grow_amount)) {
3659 vm_map_unlock_read(map);
3660 return (KERN_NO_SPACE);
3665 /* Round up the grow amount modulo sgrowsiz */
3666 growsize = sgrowsiz;
3667 grow_amount = roundup(grow_amount, growsize);
3668 if (grow_amount > stack_entry->avail_ssize)
3669 grow_amount = stack_entry->avail_ssize;
3670 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) {
3671 grow_amount = trunc_page((vm_size_t)stacklim) -
3676 limit = racct_get_available(p, RACCT_STACK);
3678 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > limit))
3679 grow_amount = limit - ctob(vm->vm_ssize);
3681 if (!old_mlock && map->flags & MAP_WIREFUTURE) {
3682 if (ptoa(pmap_wired_count(map->pmap)) + grow_amount > lmemlim) {
3683 vm_map_unlock_read(map);
3689 if (racct_set(p, RACCT_MEMLOCK,
3690 ptoa(pmap_wired_count(map->pmap)) + grow_amount)) {
3692 vm_map_unlock_read(map);
3699 /* If we would blow our VMEM resource limit, no go */
3700 if (map->size + grow_amount > vmemlim) {
3701 vm_map_unlock_read(map);
3707 if (racct_set(p, RACCT_VMEM, map->size + grow_amount)) {
3709 vm_map_unlock_read(map);
3716 if (vm_map_lock_upgrade(map))
3719 if (stack_entry == next_entry) {
3723 /* Get the preliminary new entry start value */
3724 addr = stack_entry->start - grow_amount;
3727 * If this puts us into the previous entry, cut back our
3728 * growth to the available space. Also, see the note above.
3731 stack_entry->avail_ssize = max_grow;
3733 if (stack_guard_page)
3737 rv = vm_map_insert(map, NULL, 0, addr, stack_entry->start,
3738 next_entry->protection, next_entry->max_protection,
3739 MAP_STACK_GROWS_DOWN);
3741 /* Adjust the available stack space by the amount we grew. */
3742 if (rv == KERN_SUCCESS) {
3743 new_entry = prev_entry->next;
3744 KASSERT(new_entry == stack_entry->prev, ("foo"));
3745 KASSERT(new_entry->end == stack_entry->start, ("foo"));
3746 KASSERT(new_entry->start == addr, ("foo"));
3747 KASSERT((new_entry->eflags & MAP_ENTRY_GROWS_DOWN) !=
3748 0, ("new entry lacks MAP_ENTRY_GROWS_DOWN"));
3749 grow_amount = new_entry->end - new_entry->start;
3750 new_entry->avail_ssize = stack_entry->avail_ssize -
3752 stack_entry->eflags &= ~MAP_ENTRY_GROWS_DOWN;
3758 addr = stack_entry->end + grow_amount;
3761 * If this puts us into the next entry, cut back our growth
3762 * to the available space. Also, see the note above.
3765 stack_entry->avail_ssize = end - stack_entry->end;
3767 if (stack_guard_page)
3771 grow_amount = addr - stack_entry->end;
3772 cred = stack_entry->cred;
3773 if (cred == NULL && stack_entry->object.vm_object != NULL)
3774 cred = stack_entry->object.vm_object->cred;
3775 if (cred != NULL && !swap_reserve_by_cred(grow_amount, cred))
3777 /* Grow the underlying object if applicable. */
3778 else if (stack_entry->object.vm_object == NULL ||
3779 vm_object_coalesce(stack_entry->object.vm_object,
3780 stack_entry->offset,
3781 (vm_size_t)(stack_entry->end - stack_entry->start),
3782 (vm_size_t)grow_amount, cred != NULL)) {
3783 map->size += (addr - stack_entry->end);
3784 /* Update the current entry. */
3785 stack_entry->end = addr;
3786 stack_entry->avail_ssize -= grow_amount;
3787 vm_map_entry_resize_free(map, stack_entry);
3793 if (rv == KERN_SUCCESS && is_procstack)
3794 vm->vm_ssize += btoc(grow_amount);
3799 * Heed the MAP_WIREFUTURE flag if it was set for this process.
3801 if (rv == KERN_SUCCESS && (map->flags & MAP_WIREFUTURE)) {
3803 (stack_entry == next_entry) ? addr : addr - grow_amount,
3804 (stack_entry == next_entry) ? stack_entry->start : addr,
3805 (p->p_flag & P_SYSTEM)
3806 ? VM_MAP_WIRE_SYSTEM|VM_MAP_WIRE_NOHOLES
3807 : VM_MAP_WIRE_USER|VM_MAP_WIRE_NOHOLES);
3812 if (rv != KERN_SUCCESS) {
3814 error = racct_set(p, RACCT_VMEM, map->size);
3815 KASSERT(error == 0, ("decreasing RACCT_VMEM failed"));
3817 error = racct_set(p, RACCT_MEMLOCK,
3818 ptoa(pmap_wired_count(map->pmap)));
3819 KASSERT(error == 0, ("decreasing RACCT_MEMLOCK failed"));
3821 error = racct_set(p, RACCT_STACK, ctob(vm->vm_ssize));
3822 KASSERT(error == 0, ("decreasing RACCT_STACK failed"));
3831 * Unshare the specified VM space for exec. If other processes are
3832 * mapped to it, then create a new one. The new vmspace is null.
3835 vmspace_exec(struct proc *p, vm_offset_t minuser, vm_offset_t maxuser)
3837 struct vmspace *oldvmspace = p->p_vmspace;
3838 struct vmspace *newvmspace;
3840 KASSERT((curthread->td_pflags & TDP_EXECVMSPC) == 0,
3841 ("vmspace_exec recursed"));
3842 newvmspace = vmspace_alloc(minuser, maxuser, NULL);
3843 if (newvmspace == NULL)
3845 newvmspace->vm_swrss = oldvmspace->vm_swrss;
3847 * This code is written like this for prototype purposes. The
3848 * goal is to avoid running down the vmspace here, but let the
3849 * other process's that are still using the vmspace to finally
3850 * run it down. Even though there is little or no chance of blocking
3851 * here, it is a good idea to keep this form for future mods.
3853 PROC_VMSPACE_LOCK(p);
3854 p->p_vmspace = newvmspace;
3855 PROC_VMSPACE_UNLOCK(p);
3856 if (p == curthread->td_proc)
3857 pmap_activate(curthread);
3858 curthread->td_pflags |= TDP_EXECVMSPC;
3863 * Unshare the specified VM space for forcing COW. This
3864 * is called by rfork, for the (RFMEM|RFPROC) == 0 case.
3867 vmspace_unshare(struct proc *p)
3869 struct vmspace *oldvmspace = p->p_vmspace;
3870 struct vmspace *newvmspace;
3871 vm_ooffset_t fork_charge;
3873 if (oldvmspace->vm_refcnt == 1)
3876 newvmspace = vmspace_fork(oldvmspace, &fork_charge);
3877 if (newvmspace == NULL)
3879 if (!swap_reserve_by_cred(fork_charge, p->p_ucred)) {
3880 vmspace_free(newvmspace);
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 vmspace_free(oldvmspace);
3895 * Finds the VM object, offset, and
3896 * protection for a given virtual address in the
3897 * specified map, assuming a page fault of the
3900 * Leaves the map in question locked for read; return
3901 * values are guaranteed until a vm_map_lookup_done
3902 * call is performed. Note that the map argument
3903 * is in/out; the returned map must be used in
3904 * the call to vm_map_lookup_done.
3906 * A handle (out_entry) is returned for use in
3907 * vm_map_lookup_done, to make that fast.
3909 * If a lookup is requested with "write protection"
3910 * specified, the map may be changed to perform virtual
3911 * copying operations, although the data referenced will
3915 vm_map_lookup(vm_map_t *var_map, /* IN/OUT */
3917 vm_prot_t fault_typea,
3918 vm_map_entry_t *out_entry, /* OUT */
3919 vm_object_t *object, /* OUT */
3920 vm_pindex_t *pindex, /* OUT */
3921 vm_prot_t *out_prot, /* OUT */
3922 boolean_t *wired) /* OUT */
3924 vm_map_entry_t entry;
3925 vm_map_t map = *var_map;
3927 vm_prot_t fault_type = fault_typea;
3928 vm_object_t eobject;
3934 vm_map_lock_read(map);
3937 * Lookup the faulting address.
3939 if (!vm_map_lookup_entry(map, vaddr, out_entry)) {
3940 vm_map_unlock_read(map);
3941 return (KERN_INVALID_ADDRESS);
3949 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
3950 vm_map_t old_map = map;
3952 *var_map = map = entry->object.sub_map;
3953 vm_map_unlock_read(old_map);
3958 * Check whether this task is allowed to have this page.
3960 prot = entry->protection;
3961 fault_type &= (VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE);
3962 if ((fault_type & prot) != fault_type || prot == VM_PROT_NONE) {
3963 vm_map_unlock_read(map);
3964 return (KERN_PROTECTION_FAILURE);
3966 if ((entry->eflags & MAP_ENTRY_USER_WIRED) &&
3967 (entry->eflags & MAP_ENTRY_COW) &&
3968 (fault_type & VM_PROT_WRITE)) {
3969 vm_map_unlock_read(map);
3970 return (KERN_PROTECTION_FAILURE);
3972 if ((fault_typea & VM_PROT_COPY) != 0 &&
3973 (entry->max_protection & VM_PROT_WRITE) == 0 &&
3974 (entry->eflags & MAP_ENTRY_COW) == 0) {
3975 vm_map_unlock_read(map);
3976 return (KERN_PROTECTION_FAILURE);
3980 * If this page is not pageable, we have to get it for all possible
3983 *wired = (entry->wired_count != 0);
3985 fault_type = entry->protection;
3986 size = entry->end - entry->start;
3988 * If the entry was copy-on-write, we either ...
3990 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
3992 * If we want to write the page, we may as well handle that
3993 * now since we've got the map locked.
3995 * If we don't need to write the page, we just demote the
3996 * permissions allowed.
3998 if ((fault_type & VM_PROT_WRITE) != 0 ||
3999 (fault_typea & VM_PROT_COPY) != 0) {
4001 * Make a new object, and place it in the object
4002 * chain. Note that no new references have appeared
4003 * -- one just moved from the map to the new
4006 if (vm_map_lock_upgrade(map))
4009 if (entry->cred == NULL) {
4011 * The debugger owner is charged for
4014 cred = curthread->td_ucred;
4016 if (!swap_reserve_by_cred(size, cred)) {
4019 return (KERN_RESOURCE_SHORTAGE);
4023 vm_object_shadow(&entry->object.vm_object,
4024 &entry->offset, size);
4025 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
4026 eobject = entry->object.vm_object;
4027 if (eobject->cred != NULL) {
4029 * The object was not shadowed.
4031 swap_release_by_cred(size, entry->cred);
4032 crfree(entry->cred);
4034 } else if (entry->cred != NULL) {
4035 VM_OBJECT_WLOCK(eobject);
4036 eobject->cred = entry->cred;
4037 eobject->charge = size;
4038 VM_OBJECT_WUNLOCK(eobject);
4042 vm_map_lock_downgrade(map);
4045 * We're attempting to read a copy-on-write page --
4046 * don't allow writes.
4048 prot &= ~VM_PROT_WRITE;
4053 * Create an object if necessary.
4055 if (entry->object.vm_object == NULL &&
4057 if (vm_map_lock_upgrade(map))
4059 entry->object.vm_object = vm_object_allocate(OBJT_DEFAULT,
4062 if (entry->cred != NULL) {
4063 VM_OBJECT_WLOCK(entry->object.vm_object);
4064 entry->object.vm_object->cred = entry->cred;
4065 entry->object.vm_object->charge = size;
4066 VM_OBJECT_WUNLOCK(entry->object.vm_object);
4069 vm_map_lock_downgrade(map);
4073 * Return the object/offset from this entry. If the entry was
4074 * copy-on-write or empty, it has been fixed up.
4076 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
4077 *object = entry->object.vm_object;
4080 return (KERN_SUCCESS);
4084 * vm_map_lookup_locked:
4086 * Lookup the faulting address. A version of vm_map_lookup that returns
4087 * KERN_FAILURE instead of blocking on map lock or memory allocation.
4090 vm_map_lookup_locked(vm_map_t *var_map, /* IN/OUT */
4092 vm_prot_t fault_typea,
4093 vm_map_entry_t *out_entry, /* OUT */
4094 vm_object_t *object, /* OUT */
4095 vm_pindex_t *pindex, /* OUT */
4096 vm_prot_t *out_prot, /* OUT */
4097 boolean_t *wired) /* OUT */
4099 vm_map_entry_t entry;
4100 vm_map_t map = *var_map;
4102 vm_prot_t fault_type = fault_typea;
4105 * Lookup the faulting address.
4107 if (!vm_map_lookup_entry(map, vaddr, out_entry))
4108 return (KERN_INVALID_ADDRESS);
4113 * Fail if the entry refers to a submap.
4115 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP)
4116 return (KERN_FAILURE);
4119 * Check whether this task is allowed to have this page.
4121 prot = entry->protection;
4122 fault_type &= VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE;
4123 if ((fault_type & prot) != fault_type)
4124 return (KERN_PROTECTION_FAILURE);
4125 if ((entry->eflags & MAP_ENTRY_USER_WIRED) &&
4126 (entry->eflags & MAP_ENTRY_COW) &&
4127 (fault_type & VM_PROT_WRITE))
4128 return (KERN_PROTECTION_FAILURE);
4131 * If this page is not pageable, we have to get it for all possible
4134 *wired = (entry->wired_count != 0);
4136 fault_type = entry->protection;
4138 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
4140 * Fail if the entry was copy-on-write for a write fault.
4142 if (fault_type & VM_PROT_WRITE)
4143 return (KERN_FAILURE);
4145 * We're attempting to read a copy-on-write page --
4146 * don't allow writes.
4148 prot &= ~VM_PROT_WRITE;
4152 * Fail if an object should be created.
4154 if (entry->object.vm_object == NULL && !map->system_map)
4155 return (KERN_FAILURE);
4158 * Return the object/offset from this entry. If the entry was
4159 * copy-on-write or empty, it has been fixed up.
4161 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
4162 *object = entry->object.vm_object;
4165 return (KERN_SUCCESS);
4169 * vm_map_lookup_done:
4171 * Releases locks acquired by a vm_map_lookup
4172 * (according to the handle returned by that lookup).
4175 vm_map_lookup_done(vm_map_t map, vm_map_entry_t entry)
4178 * Unlock the main-level map
4180 vm_map_unlock_read(map);
4183 #include "opt_ddb.h"
4185 #include <sys/kernel.h>
4187 #include <ddb/ddb.h>
4190 vm_map_print(vm_map_t map)
4192 vm_map_entry_t entry;
4194 db_iprintf("Task map %p: pmap=%p, nentries=%d, version=%u\n",
4196 (void *)map->pmap, map->nentries, map->timestamp);
4199 for (entry = map->header.next; entry != &map->header;
4200 entry = entry->next) {
4201 db_iprintf("map entry %p: start=%p, end=%p\n",
4202 (void *)entry, (void *)entry->start, (void *)entry->end);
4204 static char *inheritance_name[4] =
4205 {"share", "copy", "none", "donate_copy"};
4207 db_iprintf(" prot=%x/%x/%s",
4209 entry->max_protection,
4210 inheritance_name[(int)(unsigned char)entry->inheritance]);
4211 if (entry->wired_count != 0)
4212 db_printf(", wired");
4214 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
4215 db_printf(", share=%p, offset=0x%jx\n",
4216 (void *)entry->object.sub_map,
4217 (uintmax_t)entry->offset);
4218 if ((entry->prev == &map->header) ||
4219 (entry->prev->object.sub_map !=
4220 entry->object.sub_map)) {
4222 vm_map_print((vm_map_t)entry->object.sub_map);
4226 if (entry->cred != NULL)
4227 db_printf(", ruid %d", entry->cred->cr_ruid);
4228 db_printf(", object=%p, offset=0x%jx",
4229 (void *)entry->object.vm_object,
4230 (uintmax_t)entry->offset);
4231 if (entry->object.vm_object && entry->object.vm_object->cred)
4232 db_printf(", obj ruid %d charge %jx",
4233 entry->object.vm_object->cred->cr_ruid,
4234 (uintmax_t)entry->object.vm_object->charge);
4235 if (entry->eflags & MAP_ENTRY_COW)
4236 db_printf(", copy (%s)",
4237 (entry->eflags & MAP_ENTRY_NEEDS_COPY) ? "needed" : "done");
4240 if ((entry->prev == &map->header) ||
4241 (entry->prev->object.vm_object !=
4242 entry->object.vm_object)) {
4244 vm_object_print((db_expr_t)(intptr_t)
4245 entry->object.vm_object,
4254 DB_SHOW_COMMAND(map, map)
4258 db_printf("usage: show map <addr>\n");
4261 vm_map_print((vm_map_t)addr);
4264 DB_SHOW_COMMAND(procvm, procvm)
4269 p = (struct proc *) addr;
4274 db_printf("p = %p, vmspace = %p, map = %p, pmap = %p\n",
4275 (void *)p, (void *)p->p_vmspace, (void *)&p->p_vmspace->vm_map,
4276 (void *)vmspace_pmap(p->p_vmspace));
4278 vm_map_print((vm_map_t)&p->p_vmspace->vm_map);