2 * Copyright (c) 1991, 1993
3 * The Regents of the University of California. All rights reserved.
5 * This code is derived from software contributed to Berkeley by
6 * The Mach Operating System project at Carnegie-Mellon University.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 4. Neither the name of the University nor the names of its contributors
17 * may be used to endorse or promote products derived from this software
18 * without specific prior written permission.
20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * from: @(#)vm_map.c 8.3 (Berkeley) 1/12/94
35 * Copyright (c) 1987, 1990 Carnegie-Mellon University.
36 * All rights reserved.
38 * Authors: Avadis Tevanian, Jr., Michael Wayne Young
40 * Permission to use, copy, modify and distribute this software and
41 * its documentation is hereby granted, provided that both the copyright
42 * notice and this permission notice appear in all copies of the
43 * software, derivative works or modified versions, and any portions
44 * thereof, and that both notices appear in supporting documentation.
46 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
47 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
48 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
50 * Carnegie Mellon requests users of this software to return to
52 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
53 * School of Computer Science
54 * Carnegie Mellon University
55 * Pittsburgh PA 15213-3890
57 * any improvements or extensions that they make and grant Carnegie the
58 * rights to redistribute these changes.
62 * Virtual memory mapping module.
65 #include <sys/cdefs.h>
66 __FBSDID("$FreeBSD$");
68 #include <sys/param.h>
69 #include <sys/systm.h>
70 #include <sys/kernel.h>
73 #include <sys/mutex.h>
75 #include <sys/vmmeter.h>
77 #include <sys/vnode.h>
78 #include <sys/racct.h>
79 #include <sys/resourcevar.h>
80 #include <sys/rwlock.h>
82 #include <sys/sysctl.h>
83 #include <sys/sysent.h>
87 #include <vm/vm_param.h>
89 #include <vm/vm_map.h>
90 #include <vm/vm_page.h>
91 #include <vm/vm_object.h>
92 #include <vm/vm_pager.h>
93 #include <vm/vm_kern.h>
94 #include <vm/vm_extern.h>
95 #include <vm/vnode_pager.h>
96 #include <vm/swap_pager.h>
100 * Virtual memory maps provide for the mapping, protection,
101 * and sharing of virtual memory objects. In addition,
102 * this module provides for an efficient virtual copy of
103 * memory from one map to another.
105 * Synchronization is required prior to most operations.
107 * Maps consist of an ordered doubly-linked list of simple
108 * entries; a self-adjusting binary search tree of these
109 * entries is used to speed up lookups.
111 * Since portions of maps are specified by start/end addresses,
112 * which may not align with existing map entries, all
113 * routines merely "clip" entries to these start/end values.
114 * [That is, an entry is split into two, bordering at a
115 * start or end value.] Note that these clippings may not
116 * always be necessary (as the two resulting entries are then
117 * not changed); however, the clipping is done for convenience.
119 * As mentioned above, virtual copy operations are performed
120 * by copying VM object references from one map to
121 * another, and then marking both regions as copy-on-write.
124 static struct mtx map_sleep_mtx;
125 static uma_zone_t mapentzone;
126 static uma_zone_t kmapentzone;
127 static uma_zone_t mapzone;
128 static uma_zone_t vmspace_zone;
129 static int vmspace_zinit(void *mem, int size, int flags);
130 static int vm_map_zinit(void *mem, int ize, int flags);
131 static void _vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min,
133 static void vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map);
134 static void vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry);
135 static void vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry);
136 static int vm_map_growstack(vm_map_t map, vm_offset_t addr,
137 vm_map_entry_t gap_entry);
138 static void vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot,
139 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags);
141 static void vm_map_zdtor(void *mem, int size, void *arg);
142 static void vmspace_zdtor(void *mem, int size, void *arg);
144 static int vm_map_stack_locked(vm_map_t map, vm_offset_t addrbos,
145 vm_size_t max_ssize, vm_size_t growsize, vm_prot_t prot, vm_prot_t max,
147 static void vm_map_wire_entry_failure(vm_map_t map, vm_map_entry_t entry,
148 vm_offset_t failed_addr);
150 #define ENTRY_CHARGED(e) ((e)->cred != NULL || \
151 ((e)->object.vm_object != NULL && (e)->object.vm_object->cred != NULL && \
152 !((e)->eflags & MAP_ENTRY_NEEDS_COPY)))
155 * PROC_VMSPACE_{UN,}LOCK() can be a noop as long as vmspaces are type
158 #define PROC_VMSPACE_LOCK(p) do { } while (0)
159 #define PROC_VMSPACE_UNLOCK(p) do { } while (0)
162 * VM_MAP_RANGE_CHECK: [ internal use only ]
164 * Asserts that the starting and ending region
165 * addresses fall within the valid range of the map.
167 #define VM_MAP_RANGE_CHECK(map, start, end) \
169 if (start < vm_map_min(map)) \
170 start = vm_map_min(map); \
171 if (end > vm_map_max(map)) \
172 end = vm_map_max(map); \
180 * Initialize the vm_map module. Must be called before
181 * any other vm_map routines.
183 * Map and entry structures are allocated from the general
184 * purpose memory pool with some exceptions:
186 * - The kernel map and kmem submap are allocated statically.
187 * - Kernel map entries are allocated out of a static pool.
189 * These restrictions are necessary since malloc() uses the
190 * maps and requires map entries.
196 mtx_init(&map_sleep_mtx, "vm map sleep mutex", NULL, MTX_DEF);
197 mapzone = uma_zcreate("MAP", sizeof(struct vm_map), NULL,
203 vm_map_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
204 uma_prealloc(mapzone, MAX_KMAP);
205 kmapentzone = uma_zcreate("KMAP ENTRY", sizeof(struct vm_map_entry),
206 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR,
207 UMA_ZONE_MTXCLASS | UMA_ZONE_VM);
208 mapentzone = uma_zcreate("MAP ENTRY", sizeof(struct vm_map_entry),
209 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
210 vmspace_zone = uma_zcreate("VMSPACE", sizeof(struct vmspace), NULL,
216 vmspace_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
220 vmspace_zinit(void *mem, int size, int flags)
224 vm = (struct vmspace *)mem;
226 vm->vm_map.pmap = NULL;
227 (void)vm_map_zinit(&vm->vm_map, sizeof(vm->vm_map), flags);
228 PMAP_LOCK_INIT(vmspace_pmap(vm));
233 vm_map_zinit(void *mem, int size, int flags)
238 memset(map, 0, sizeof(*map));
239 mtx_init(&map->system_mtx, "vm map (system)", NULL, MTX_DEF | MTX_DUPOK);
240 sx_init(&map->lock, "vm map (user)");
246 vmspace_zdtor(void *mem, int size, void *arg)
250 vm = (struct vmspace *)mem;
252 vm_map_zdtor(&vm->vm_map, sizeof(vm->vm_map), arg);
255 vm_map_zdtor(void *mem, int size, void *arg)
260 KASSERT(map->nentries == 0,
261 ("map %p nentries == %d on free.",
262 map, map->nentries));
263 KASSERT(map->size == 0,
264 ("map %p size == %lu on free.",
265 map, (unsigned long)map->size));
267 #endif /* INVARIANTS */
270 * Allocate a vmspace structure, including a vm_map and pmap,
271 * and initialize those structures. The refcnt is set to 1.
273 * If 'pinit' is NULL then the embedded pmap is initialized via pmap_pinit().
276 vmspace_alloc(vm_offset_t min, vm_offset_t max, pmap_pinit_t pinit)
280 vm = uma_zalloc(vmspace_zone, M_WAITOK);
282 KASSERT(vm->vm_map.pmap == NULL, ("vm_map.pmap must be NULL"));
287 if (!pinit(vmspace_pmap(vm))) {
288 uma_zfree(vmspace_zone, vm);
291 CTR1(KTR_VM, "vmspace_alloc: %p", vm);
292 _vm_map_init(&vm->vm_map, vmspace_pmap(vm), min, max);
307 vmspace_container_reset(struct proc *p)
311 racct_set(p, RACCT_DATA, 0);
312 racct_set(p, RACCT_STACK, 0);
313 racct_set(p, RACCT_RSS, 0);
314 racct_set(p, RACCT_MEMLOCK, 0);
315 racct_set(p, RACCT_VMEM, 0);
321 vmspace_dofree(struct vmspace *vm)
324 CTR1(KTR_VM, "vmspace_free: %p", vm);
327 * Make sure any SysV shm is freed, it might not have been in
333 * Lock the map, to wait out all other references to it.
334 * Delete all of the mappings and pages they hold, then call
335 * the pmap module to reclaim anything left.
337 (void)vm_map_remove(&vm->vm_map, vm_map_min(&vm->vm_map),
338 vm_map_max(&vm->vm_map));
340 pmap_release(vmspace_pmap(vm));
341 vm->vm_map.pmap = NULL;
342 uma_zfree(vmspace_zone, vm);
346 vmspace_free(struct vmspace *vm)
349 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
350 "vmspace_free() called");
352 if (vm->vm_refcnt == 0)
353 panic("vmspace_free: attempt to free already freed vmspace");
355 if (atomic_fetchadd_int(&vm->vm_refcnt, -1) == 1)
360 vmspace_exitfree(struct proc *p)
364 PROC_VMSPACE_LOCK(p);
367 PROC_VMSPACE_UNLOCK(p);
368 KASSERT(vm == &vmspace0, ("vmspace_exitfree: wrong vmspace"));
373 vmspace_exit(struct thread *td)
380 * Release user portion of address space.
381 * This releases references to vnodes,
382 * which could cause I/O if the file has been unlinked.
383 * Need to do this early enough that we can still sleep.
385 * The last exiting process to reach this point releases as
386 * much of the environment as it can. vmspace_dofree() is the
387 * slower fallback in case another process had a temporary
388 * reference to the vmspace.
393 atomic_add_int(&vmspace0.vm_refcnt, 1);
395 refcnt = vm->vm_refcnt;
396 if (refcnt > 1 && p->p_vmspace != &vmspace0) {
397 /* Switch now since other proc might free vmspace */
398 PROC_VMSPACE_LOCK(p);
399 p->p_vmspace = &vmspace0;
400 PROC_VMSPACE_UNLOCK(p);
403 } while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt - 1));
405 if (p->p_vmspace != vm) {
406 /* vmspace not yet freed, switch back */
407 PROC_VMSPACE_LOCK(p);
409 PROC_VMSPACE_UNLOCK(p);
412 pmap_remove_pages(vmspace_pmap(vm));
413 /* Switch now since this proc will free vmspace */
414 PROC_VMSPACE_LOCK(p);
415 p->p_vmspace = &vmspace0;
416 PROC_VMSPACE_UNLOCK(p);
422 vmspace_container_reset(p);
426 /* Acquire reference to vmspace owned by another process. */
429 vmspace_acquire_ref(struct proc *p)
434 PROC_VMSPACE_LOCK(p);
437 PROC_VMSPACE_UNLOCK(p);
441 refcnt = vm->vm_refcnt;
442 if (refcnt <= 0) { /* Avoid 0->1 transition */
443 PROC_VMSPACE_UNLOCK(p);
446 } while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt + 1));
447 if (vm != p->p_vmspace) {
448 PROC_VMSPACE_UNLOCK(p);
452 PROC_VMSPACE_UNLOCK(p);
457 * Switch between vmspaces in an AIO kernel process.
459 * The AIO kernel processes switch to and from a user process's
460 * vmspace while performing an I/O operation on behalf of a user
461 * process. The new vmspace is either the vmspace of a user process
462 * obtained from an active AIO request or the initial vmspace of the
463 * AIO kernel process (when it is idling). Because user processes
464 * will block to drain any active AIO requests before proceeding in
465 * exit() or execve(), the vmspace reference count for these vmspaces
466 * can never be 0. This allows for a much simpler implementation than
467 * the loop in vmspace_acquire_ref() above. Similarly, AIO kernel
468 * processes hold an extra reference on their initial vmspace for the
469 * life of the process so that this guarantee is true for any vmspace
473 vmspace_switch_aio(struct vmspace *newvm)
475 struct vmspace *oldvm;
477 /* XXX: Need some way to assert that this is an aio daemon. */
479 KASSERT(newvm->vm_refcnt > 0,
480 ("vmspace_switch_aio: newvm unreferenced"));
482 oldvm = curproc->p_vmspace;
487 * Point to the new address space and refer to it.
489 curproc->p_vmspace = newvm;
490 atomic_add_int(&newvm->vm_refcnt, 1);
492 /* Activate the new mapping. */
493 pmap_activate(curthread);
495 /* Remove the daemon's reference to the old address space. */
496 KASSERT(oldvm->vm_refcnt > 1,
497 ("vmspace_switch_aio: oldvm dropping last reference"));
502 _vm_map_lock(vm_map_t map, const char *file, int line)
506 mtx_lock_flags_(&map->system_mtx, 0, file, line);
508 sx_xlock_(&map->lock, file, line);
513 vm_map_process_deferred(void)
516 vm_map_entry_t entry, next;
520 entry = td->td_map_def_user;
521 td->td_map_def_user = NULL;
522 while (entry != NULL) {
524 if ((entry->eflags & MAP_ENTRY_VN_WRITECNT) != 0) {
526 * Decrement the object's writemappings and
527 * possibly the vnode's v_writecount.
529 KASSERT((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0,
530 ("Submap with writecount"));
531 object = entry->object.vm_object;
532 KASSERT(object != NULL, ("No object for writecount"));
533 vnode_pager_release_writecount(object, entry->start,
536 vm_map_entry_deallocate(entry, FALSE);
542 _vm_map_unlock(vm_map_t map, const char *file, int line)
546 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
548 sx_xunlock_(&map->lock, file, line);
549 vm_map_process_deferred();
554 _vm_map_lock_read(vm_map_t map, const char *file, int line)
558 mtx_lock_flags_(&map->system_mtx, 0, file, line);
560 sx_slock_(&map->lock, file, line);
564 _vm_map_unlock_read(vm_map_t map, const char *file, int line)
568 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
570 sx_sunlock_(&map->lock, file, line);
571 vm_map_process_deferred();
576 _vm_map_trylock(vm_map_t map, const char *file, int line)
580 error = map->system_map ?
581 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) :
582 !sx_try_xlock_(&map->lock, file, line);
589 _vm_map_trylock_read(vm_map_t map, const char *file, int line)
593 error = map->system_map ?
594 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) :
595 !sx_try_slock_(&map->lock, file, line);
600 * _vm_map_lock_upgrade: [ internal use only ]
602 * Tries to upgrade a read (shared) lock on the specified map to a write
603 * (exclusive) lock. Returns the value "0" if the upgrade succeeds and a
604 * non-zero value if the upgrade fails. If the upgrade fails, the map is
605 * returned without a read or write lock held.
607 * Requires that the map be read locked.
610 _vm_map_lock_upgrade(vm_map_t map, const char *file, int line)
612 unsigned int last_timestamp;
614 if (map->system_map) {
615 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
617 if (!sx_try_upgrade_(&map->lock, file, line)) {
618 last_timestamp = map->timestamp;
619 sx_sunlock_(&map->lock, file, line);
620 vm_map_process_deferred();
622 * If the map's timestamp does not change while the
623 * map is unlocked, then the upgrade succeeds.
625 sx_xlock_(&map->lock, file, line);
626 if (last_timestamp != map->timestamp) {
627 sx_xunlock_(&map->lock, file, line);
637 _vm_map_lock_downgrade(vm_map_t map, const char *file, int line)
640 if (map->system_map) {
641 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
643 sx_downgrade_(&map->lock, file, line);
649 * Returns a non-zero value if the caller holds a write (exclusive) lock
650 * on the specified map and the value "0" otherwise.
653 vm_map_locked(vm_map_t map)
657 return (mtx_owned(&map->system_mtx));
659 return (sx_xlocked(&map->lock));
664 _vm_map_assert_locked(vm_map_t map, const char *file, int line)
668 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
670 sx_assert_(&map->lock, SA_XLOCKED, file, line);
673 #define VM_MAP_ASSERT_LOCKED(map) \
674 _vm_map_assert_locked(map, LOCK_FILE, LOCK_LINE)
676 #define VM_MAP_ASSERT_LOCKED(map)
680 * _vm_map_unlock_and_wait:
682 * Atomically releases the lock on the specified map and puts the calling
683 * thread to sleep. The calling thread will remain asleep until either
684 * vm_map_wakeup() is performed on the map or the specified timeout is
687 * WARNING! This function does not perform deferred deallocations of
688 * objects and map entries. Therefore, the calling thread is expected to
689 * reacquire the map lock after reawakening and later perform an ordinary
690 * unlock operation, such as vm_map_unlock(), before completing its
691 * operation on the map.
694 _vm_map_unlock_and_wait(vm_map_t map, int timo, const char *file, int line)
697 mtx_lock(&map_sleep_mtx);
699 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
701 sx_xunlock_(&map->lock, file, line);
702 return (msleep(&map->root, &map_sleep_mtx, PDROP | PVM, "vmmaps",
709 * Awaken any threads that have slept on the map using
710 * vm_map_unlock_and_wait().
713 vm_map_wakeup(vm_map_t map)
717 * Acquire and release map_sleep_mtx to prevent a wakeup()
718 * from being performed (and lost) between the map unlock
719 * and the msleep() in _vm_map_unlock_and_wait().
721 mtx_lock(&map_sleep_mtx);
722 mtx_unlock(&map_sleep_mtx);
727 vm_map_busy(vm_map_t map)
730 VM_MAP_ASSERT_LOCKED(map);
735 vm_map_unbusy(vm_map_t map)
738 VM_MAP_ASSERT_LOCKED(map);
739 KASSERT(map->busy, ("vm_map_unbusy: not busy"));
740 if (--map->busy == 0 && (map->flags & MAP_BUSY_WAKEUP)) {
741 vm_map_modflags(map, 0, MAP_BUSY_WAKEUP);
747 vm_map_wait_busy(vm_map_t map)
750 VM_MAP_ASSERT_LOCKED(map);
752 vm_map_modflags(map, MAP_BUSY_WAKEUP, 0);
754 msleep(&map->busy, &map->system_mtx, 0, "mbusy", 0);
756 sx_sleep(&map->busy, &map->lock, 0, "mbusy", 0);
762 vmspace_resident_count(struct vmspace *vmspace)
764 return pmap_resident_count(vmspace_pmap(vmspace));
770 * Creates and returns a new empty VM map with
771 * the given physical map structure, and having
772 * the given lower and upper address bounds.
775 vm_map_create(pmap_t pmap, vm_offset_t min, vm_offset_t max)
779 result = uma_zalloc(mapzone, M_WAITOK);
780 CTR1(KTR_VM, "vm_map_create: %p", result);
781 _vm_map_init(result, pmap, min, max);
786 * Initialize an existing vm_map structure
787 * such as that in the vmspace structure.
790 _vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max)
793 map->header.next = map->header.prev = &map->header;
794 map->needs_wakeup = FALSE;
797 map->header.end = min;
798 map->header.start = max;
806 vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max)
809 _vm_map_init(map, pmap, min, max);
810 mtx_init(&map->system_mtx, "system map", NULL, MTX_DEF | MTX_DUPOK);
811 sx_init(&map->lock, "user map");
815 * vm_map_entry_dispose: [ internal use only ]
817 * Inverse of vm_map_entry_create.
820 vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry)
822 uma_zfree(map->system_map ? kmapentzone : mapentzone, entry);
826 * vm_map_entry_create: [ internal use only ]
828 * Allocates a VM map entry for insertion.
829 * No entry fields are filled in.
831 static vm_map_entry_t
832 vm_map_entry_create(vm_map_t map)
834 vm_map_entry_t new_entry;
837 new_entry = uma_zalloc(kmapentzone, M_NOWAIT);
839 new_entry = uma_zalloc(mapentzone, M_WAITOK);
840 if (new_entry == NULL)
841 panic("vm_map_entry_create: kernel resources exhausted");
846 * vm_map_entry_set_behavior:
848 * Set the expected access behavior, either normal, random, or
852 vm_map_entry_set_behavior(vm_map_entry_t entry, u_char behavior)
854 entry->eflags = (entry->eflags & ~MAP_ENTRY_BEHAV_MASK) |
855 (behavior & MAP_ENTRY_BEHAV_MASK);
859 * vm_map_entry_set_max_free:
861 * Set the max_free field in a vm_map_entry.
864 vm_map_entry_set_max_free(vm_map_entry_t entry)
867 entry->max_free = entry->adj_free;
868 if (entry->left != NULL && entry->left->max_free > entry->max_free)
869 entry->max_free = entry->left->max_free;
870 if (entry->right != NULL && entry->right->max_free > entry->max_free)
871 entry->max_free = entry->right->max_free;
875 * vm_map_entry_splay:
877 * The Sleator and Tarjan top-down splay algorithm with the
878 * following variation. Max_free must be computed bottom-up, so
879 * on the downward pass, maintain the left and right spines in
880 * reverse order. Then, make a second pass up each side to fix
881 * the pointers and compute max_free. The time bound is O(log n)
884 * The new root is the vm_map_entry containing "addr", or else an
885 * adjacent entry (lower or higher) if addr is not in the tree.
887 * The map must be locked, and leaves it so.
889 * Returns: the new root.
891 static vm_map_entry_t
892 vm_map_entry_splay(vm_offset_t addr, vm_map_entry_t root)
894 vm_map_entry_t llist, rlist;
895 vm_map_entry_t ltree, rtree;
898 /* Special case of empty tree. */
903 * Pass One: Splay down the tree until we find addr or a NULL
904 * pointer where addr would go. llist and rlist are the two
905 * sides in reverse order (bottom-up), with llist linked by
906 * the right pointer and rlist linked by the left pointer in
907 * the vm_map_entry. Wait until Pass Two to set max_free on
913 /* root is never NULL in here. */
914 if (addr < root->start) {
918 if (addr < y->start && y->left != NULL) {
919 /* Rotate right and put y on rlist. */
920 root->left = y->right;
922 vm_map_entry_set_max_free(root);
927 /* Put root on rlist. */
932 } else if (addr >= root->end) {
936 if (addr >= y->end && y->right != NULL) {
937 /* Rotate left and put y on llist. */
938 root->right = y->left;
940 vm_map_entry_set_max_free(root);
945 /* Put root on llist. */
955 * Pass Two: Walk back up the two spines, flip the pointers
956 * and set max_free. The subtrees of the root go at the
957 * bottom of llist and rlist.
960 while (llist != NULL) {
962 llist->right = ltree;
963 vm_map_entry_set_max_free(llist);
968 while (rlist != NULL) {
971 vm_map_entry_set_max_free(rlist);
977 * Final assembly: add ltree and rtree as subtrees of root.
981 vm_map_entry_set_max_free(root);
987 * vm_map_entry_{un,}link:
989 * Insert/remove entries from maps.
992 vm_map_entry_link(vm_map_t map,
993 vm_map_entry_t after_where,
994 vm_map_entry_t entry)
998 "vm_map_entry_link: map %p, nentries %d, entry %p, after %p", map,
999 map->nentries, entry, after_where);
1000 VM_MAP_ASSERT_LOCKED(map);
1001 KASSERT(after_where->end <= entry->start,
1002 ("vm_map_entry_link: prev end %jx new start %jx overlap",
1003 (uintmax_t)after_where->end, (uintmax_t)entry->start));
1004 KASSERT(entry->end <= after_where->next->start,
1005 ("vm_map_entry_link: new end %jx next start %jx overlap",
1006 (uintmax_t)entry->end, (uintmax_t)after_where->next->start));
1009 entry->prev = after_where;
1010 entry->next = after_where->next;
1011 entry->next->prev = entry;
1012 after_where->next = entry;
1014 if (after_where != &map->header) {
1015 if (after_where != map->root)
1016 vm_map_entry_splay(after_where->start, map->root);
1017 entry->right = after_where->right;
1018 entry->left = after_where;
1019 after_where->right = NULL;
1020 after_where->adj_free = entry->start - after_where->end;
1021 vm_map_entry_set_max_free(after_where);
1023 entry->right = map->root;
1026 entry->adj_free = entry->next->start - entry->end;
1027 vm_map_entry_set_max_free(entry);
1032 vm_map_entry_unlink(vm_map_t map,
1033 vm_map_entry_t entry)
1035 vm_map_entry_t next, prev, root;
1037 VM_MAP_ASSERT_LOCKED(map);
1038 if (entry != map->root)
1039 vm_map_entry_splay(entry->start, map->root);
1040 if (entry->left == NULL)
1041 root = entry->right;
1043 root = vm_map_entry_splay(entry->start, entry->left);
1044 root->right = entry->right;
1045 root->adj_free = entry->next->start - root->end;
1046 vm_map_entry_set_max_free(root);
1055 CTR3(KTR_VM, "vm_map_entry_unlink: map %p, nentries %d, entry %p", map,
1056 map->nentries, entry);
1060 * vm_map_entry_resize_free:
1062 * Recompute the amount of free space following a vm_map_entry
1063 * and propagate that value up the tree. Call this function after
1064 * resizing a map entry in-place, that is, without a call to
1065 * vm_map_entry_link() or _unlink().
1067 * The map must be locked, and leaves it so.
1070 vm_map_entry_resize_free(vm_map_t map, vm_map_entry_t entry)
1074 * Using splay trees without parent pointers, propagating
1075 * max_free up the tree is done by moving the entry to the
1076 * root and making the change there.
1078 if (entry != map->root)
1079 map->root = vm_map_entry_splay(entry->start, map->root);
1081 entry->adj_free = entry->next->start - entry->end;
1082 vm_map_entry_set_max_free(entry);
1086 * vm_map_lookup_entry: [ internal use only ]
1088 * Finds the map entry containing (or
1089 * immediately preceding) the specified address
1090 * in the given map; the entry is returned
1091 * in the "entry" parameter. The boolean
1092 * result indicates whether the address is
1093 * actually contained in the map.
1096 vm_map_lookup_entry(
1098 vm_offset_t address,
1099 vm_map_entry_t *entry) /* OUT */
1105 * If the map is empty, then the map entry immediately preceding
1106 * "address" is the map's header.
1110 *entry = &map->header;
1111 else if (address >= cur->start && cur->end > address) {
1114 } else if ((locked = vm_map_locked(map)) ||
1115 sx_try_upgrade(&map->lock)) {
1117 * Splay requires a write lock on the map. However, it only
1118 * restructures the binary search tree; it does not otherwise
1119 * change the map. Thus, the map's timestamp need not change
1120 * on a temporary upgrade.
1122 map->root = cur = vm_map_entry_splay(address, cur);
1124 sx_downgrade(&map->lock);
1127 * If "address" is contained within a map entry, the new root
1128 * is that map entry. Otherwise, the new root is a map entry
1129 * immediately before or after "address".
1131 if (address >= cur->start) {
1133 if (cur->end > address)
1139 * Since the map is only locked for read access, perform a
1140 * standard binary search tree lookup for "address".
1143 if (address < cur->start) {
1144 if (cur->left == NULL) {
1149 } else if (cur->end > address) {
1153 if (cur->right == NULL) {
1166 * Inserts the given whole VM object into the target
1167 * map at the specified address range. The object's
1168 * size should match that of the address range.
1170 * Requires that the map be locked, and leaves it so.
1172 * If object is non-NULL, ref count must be bumped by caller
1173 * prior to making call to account for the new entry.
1176 vm_map_insert(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1177 vm_offset_t start, vm_offset_t end, vm_prot_t prot, vm_prot_t max, int cow)
1179 vm_map_entry_t new_entry, prev_entry, temp_entry;
1181 vm_eflags_t protoeflags;
1182 vm_inherit_t inheritance;
1184 VM_MAP_ASSERT_LOCKED(map);
1185 KASSERT((object != kmem_object && object != kernel_object) ||
1186 (cow & MAP_COPY_ON_WRITE) == 0,
1187 ("vm_map_insert: kmem or kernel object and COW"));
1188 KASSERT(object == NULL || (cow & MAP_NOFAULT) == 0,
1189 ("vm_map_insert: paradoxical MAP_NOFAULT request"));
1190 KASSERT((prot & ~max) == 0,
1191 ("prot %#x is not subset of max_prot %#x", prot, max));
1194 * Check that the start and end points are not bogus.
1196 if (start < vm_map_min(map) || end > vm_map_max(map) ||
1198 return (KERN_INVALID_ADDRESS);
1201 * Find the entry prior to the proposed starting address; if it's part
1202 * of an existing entry, this range is bogus.
1204 if (vm_map_lookup_entry(map, start, &temp_entry))
1205 return (KERN_NO_SPACE);
1207 prev_entry = temp_entry;
1210 * Assert that the next entry doesn't overlap the end point.
1212 if (prev_entry->next->start < end)
1213 return (KERN_NO_SPACE);
1215 if ((cow & MAP_CREATE_GUARD) != 0 && (object != NULL ||
1216 max != VM_PROT_NONE))
1217 return (KERN_INVALID_ARGUMENT);
1220 if (cow & MAP_COPY_ON_WRITE)
1221 protoeflags |= MAP_ENTRY_COW | MAP_ENTRY_NEEDS_COPY;
1222 if (cow & MAP_NOFAULT)
1223 protoeflags |= MAP_ENTRY_NOFAULT;
1224 if (cow & MAP_DISABLE_SYNCER)
1225 protoeflags |= MAP_ENTRY_NOSYNC;
1226 if (cow & MAP_DISABLE_COREDUMP)
1227 protoeflags |= MAP_ENTRY_NOCOREDUMP;
1228 if (cow & MAP_STACK_GROWS_DOWN)
1229 protoeflags |= MAP_ENTRY_GROWS_DOWN;
1230 if (cow & MAP_STACK_GROWS_UP)
1231 protoeflags |= MAP_ENTRY_GROWS_UP;
1232 if (cow & MAP_VN_WRITECOUNT)
1233 protoeflags |= MAP_ENTRY_VN_WRITECNT;
1234 if ((cow & MAP_CREATE_GUARD) != 0)
1235 protoeflags |= MAP_ENTRY_GUARD;
1236 if ((cow & MAP_CREATE_STACK_GAP_DN) != 0)
1237 protoeflags |= MAP_ENTRY_STACK_GAP_DN;
1238 if ((cow & MAP_CREATE_STACK_GAP_UP) != 0)
1239 protoeflags |= MAP_ENTRY_STACK_GAP_UP;
1240 if (cow & MAP_INHERIT_SHARE)
1241 inheritance = VM_INHERIT_SHARE;
1243 inheritance = VM_INHERIT_DEFAULT;
1246 if ((cow & (MAP_ACC_NO_CHARGE | MAP_NOFAULT | MAP_CREATE_GUARD)) != 0)
1248 if ((cow & MAP_ACC_CHARGED) || ((prot & VM_PROT_WRITE) &&
1249 ((protoeflags & MAP_ENTRY_NEEDS_COPY) || object == NULL))) {
1250 if (!(cow & MAP_ACC_CHARGED) && !swap_reserve(end - start))
1251 return (KERN_RESOURCE_SHORTAGE);
1252 KASSERT(object == NULL ||
1253 (protoeflags & MAP_ENTRY_NEEDS_COPY) != 0 ||
1254 object->cred == NULL,
1255 ("overcommit: vm_map_insert o %p", object));
1256 cred = curthread->td_ucred;
1260 /* Expand the kernel pmap, if necessary. */
1261 if (map == kernel_map && end > kernel_vm_end)
1262 pmap_growkernel(end);
1263 if (object != NULL) {
1265 * OBJ_ONEMAPPING must be cleared unless this mapping
1266 * is trivially proven to be the only mapping for any
1267 * of the object's pages. (Object granularity
1268 * reference counting is insufficient to recognize
1269 * aliases with precision.)
1271 VM_OBJECT_WLOCK(object);
1272 if (object->ref_count > 1 || object->shadow_count != 0)
1273 vm_object_clear_flag(object, OBJ_ONEMAPPING);
1274 VM_OBJECT_WUNLOCK(object);
1275 } else if (prev_entry != &map->header &&
1276 prev_entry->eflags == protoeflags &&
1277 (cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 &&
1278 prev_entry->end == start && prev_entry->wired_count == 0 &&
1279 (prev_entry->cred == cred ||
1280 (prev_entry->object.vm_object != NULL &&
1281 prev_entry->object.vm_object->cred == cred)) &&
1282 vm_object_coalesce(prev_entry->object.vm_object,
1284 (vm_size_t)(prev_entry->end - prev_entry->start),
1285 (vm_size_t)(end - prev_entry->end), cred != NULL &&
1286 (protoeflags & MAP_ENTRY_NEEDS_COPY) == 0)) {
1288 * We were able to extend the object. Determine if we
1289 * can extend the previous map entry to include the
1290 * new range as well.
1292 if (prev_entry->inheritance == inheritance &&
1293 prev_entry->protection == prot &&
1294 prev_entry->max_protection == max) {
1295 if ((prev_entry->eflags & MAP_ENTRY_GUARD) == 0)
1296 map->size += end - prev_entry->end;
1297 prev_entry->end = end;
1298 vm_map_entry_resize_free(map, prev_entry);
1299 vm_map_simplify_entry(map, prev_entry);
1300 return (KERN_SUCCESS);
1304 * If we can extend the object but cannot extend the
1305 * map entry, we have to create a new map entry. We
1306 * must bump the ref count on the extended object to
1307 * account for it. object may be NULL.
1309 object = prev_entry->object.vm_object;
1310 offset = prev_entry->offset +
1311 (prev_entry->end - prev_entry->start);
1312 vm_object_reference(object);
1313 if (cred != NULL && object != NULL && object->cred != NULL &&
1314 !(prev_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
1315 /* Object already accounts for this uid. */
1323 * Create a new entry
1325 new_entry = vm_map_entry_create(map);
1326 new_entry->start = start;
1327 new_entry->end = end;
1328 new_entry->cred = NULL;
1330 new_entry->eflags = protoeflags;
1331 new_entry->object.vm_object = object;
1332 new_entry->offset = offset;
1334 new_entry->inheritance = inheritance;
1335 new_entry->protection = prot;
1336 new_entry->max_protection = max;
1337 new_entry->wired_count = 0;
1338 new_entry->wiring_thread = NULL;
1339 new_entry->read_ahead = VM_FAULT_READ_AHEAD_INIT;
1340 new_entry->next_read = start;
1342 KASSERT(cred == NULL || !ENTRY_CHARGED(new_entry),
1343 ("overcommit: vm_map_insert leaks vm_map %p", new_entry));
1344 new_entry->cred = cred;
1347 * Insert the new entry into the list
1349 vm_map_entry_link(map, prev_entry, new_entry);
1350 if ((new_entry->eflags & MAP_ENTRY_GUARD) == 0)
1351 map->size += new_entry->end - new_entry->start;
1354 * Try to coalesce the new entry with both the previous and next
1355 * entries in the list. Previously, we only attempted to coalesce
1356 * with the previous entry when object is NULL. Here, we handle the
1357 * other cases, which are less common.
1359 vm_map_simplify_entry(map, new_entry);
1361 if ((cow & (MAP_PREFAULT | MAP_PREFAULT_PARTIAL)) != 0) {
1362 vm_map_pmap_enter(map, start, prot, object, OFF_TO_IDX(offset),
1363 end - start, cow & MAP_PREFAULT_PARTIAL);
1366 return (KERN_SUCCESS);
1372 * Find the first fit (lowest VM address) for "length" free bytes
1373 * beginning at address >= start in the given map.
1375 * In a vm_map_entry, "adj_free" is the amount of free space
1376 * adjacent (higher address) to this entry, and "max_free" is the
1377 * maximum amount of contiguous free space in its subtree. This
1378 * allows finding a free region in one path down the tree, so
1379 * O(log n) amortized with splay trees.
1381 * The map must be locked, and leaves it so.
1383 * Returns: 0 on success, and starting address in *addr,
1384 * 1 if insufficient space.
1387 vm_map_findspace(vm_map_t map, vm_offset_t start, vm_size_t length,
1388 vm_offset_t *addr) /* OUT */
1390 vm_map_entry_t entry;
1394 * Request must fit within min/max VM address and must avoid
1397 start = MAX(start, vm_map_min(map));
1398 if (start + length > vm_map_max(map) || start + length < start)
1401 /* Empty tree means wide open address space. */
1402 if (map->root == NULL) {
1408 * After splay, if start comes before root node, then there
1409 * must be a gap from start to the root.
1411 map->root = vm_map_entry_splay(start, map->root);
1412 if (start + length <= map->root->start) {
1418 * Root is the last node that might begin its gap before
1419 * start, and this is the last comparison where address
1420 * wrap might be a problem.
1422 st = (start > map->root->end) ? start : map->root->end;
1423 if (length <= map->root->end + map->root->adj_free - st) {
1428 /* With max_free, can immediately tell if no solution. */
1429 entry = map->root->right;
1430 if (entry == NULL || length > entry->max_free)
1434 * Search the right subtree in the order: left subtree, root,
1435 * right subtree (first fit). The previous splay implies that
1436 * all regions in the right subtree have addresses > start.
1438 while (entry != NULL) {
1439 if (entry->left != NULL && entry->left->max_free >= length)
1440 entry = entry->left;
1441 else if (entry->adj_free >= length) {
1445 entry = entry->right;
1448 /* Can't get here, so panic if we do. */
1449 panic("vm_map_findspace: max_free corrupt");
1453 vm_map_fixed(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1454 vm_offset_t start, vm_size_t length, vm_prot_t prot,
1455 vm_prot_t max, int cow)
1460 end = start + length;
1461 KASSERT((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 ||
1463 ("vm_map_fixed: non-NULL backing object for stack"));
1465 VM_MAP_RANGE_CHECK(map, start, end);
1466 if ((cow & MAP_CHECK_EXCL) == 0)
1467 vm_map_delete(map, start, end);
1468 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) != 0) {
1469 result = vm_map_stack_locked(map, start, length, sgrowsiz,
1472 result = vm_map_insert(map, object, offset, start, end,
1480 * vm_map_find finds an unallocated region in the target address
1481 * map with the given length. The search is defined to be
1482 * first-fit from the specified address; the region found is
1483 * returned in the same parameter.
1485 * If object is non-NULL, ref count must be bumped by caller
1486 * prior to making call to account for the new entry.
1489 vm_map_find(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1490 vm_offset_t *addr, /* IN/OUT */
1491 vm_size_t length, vm_offset_t max_addr, int find_space,
1492 vm_prot_t prot, vm_prot_t max, int cow)
1494 vm_offset_t alignment, initial_addr, start;
1497 KASSERT((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 ||
1499 ("vm_map_find: non-NULL backing object for stack"));
1500 if (find_space == VMFS_OPTIMAL_SPACE && (object == NULL ||
1501 (object->flags & OBJ_COLORED) == 0))
1502 find_space = VMFS_ANY_SPACE;
1503 if (find_space >> 8 != 0) {
1504 KASSERT((find_space & 0xff) == 0, ("bad VMFS flags"));
1505 alignment = (vm_offset_t)1 << (find_space >> 8);
1508 initial_addr = *addr;
1510 start = initial_addr;
1513 if (find_space != VMFS_NO_SPACE) {
1514 if (vm_map_findspace(map, start, length, addr) ||
1515 (max_addr != 0 && *addr + length > max_addr)) {
1517 if (find_space == VMFS_OPTIMAL_SPACE) {
1518 find_space = VMFS_ANY_SPACE;
1521 return (KERN_NO_SPACE);
1523 switch (find_space) {
1524 case VMFS_SUPER_SPACE:
1525 case VMFS_OPTIMAL_SPACE:
1526 pmap_align_superpage(object, offset, addr,
1529 case VMFS_ANY_SPACE:
1532 if ((*addr & (alignment - 1)) != 0) {
1533 *addr &= ~(alignment - 1);
1541 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) != 0) {
1542 result = vm_map_stack_locked(map, start, length,
1543 sgrowsiz, prot, max, cow);
1545 result = vm_map_insert(map, object, offset, start,
1546 start + length, prot, max, cow);
1548 } while (result == KERN_NO_SPACE && find_space != VMFS_NO_SPACE &&
1549 find_space != VMFS_ANY_SPACE);
1555 * vm_map_find_min() is a variant of vm_map_find() that takes an
1556 * additional parameter (min_addr) and treats the given address
1557 * (*addr) differently. Specifically, it treats *addr as a hint
1558 * and not as the minimum address where the mapping is created.
1560 * This function works in two phases. First, it tries to
1561 * allocate above the hint. If that fails and the hint is
1562 * greater than min_addr, it performs a second pass, replacing
1563 * the hint with min_addr as the minimum address for the
1567 vm_map_find_min(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1568 vm_offset_t *addr, vm_size_t length, vm_offset_t min_addr,
1569 vm_offset_t max_addr, int find_space, vm_prot_t prot, vm_prot_t max,
1577 rv = vm_map_find(map, object, offset, addr, length, max_addr,
1578 find_space, prot, max, cow);
1579 if (rv == KERN_SUCCESS || min_addr >= hint)
1581 *addr = hint = min_addr;
1586 * vm_map_simplify_entry:
1588 * Simplify the given map entry by merging with either neighbor. This
1589 * routine also has the ability to merge with both neighbors.
1591 * The map must be locked.
1593 * This routine guarantees that the passed entry remains valid (though
1594 * possibly extended). When merging, this routine may delete one or
1598 vm_map_simplify_entry(vm_map_t map, vm_map_entry_t entry)
1600 vm_map_entry_t next, prev;
1601 vm_size_t prevsize, esize;
1603 if ((entry->eflags & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP |
1604 MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_IS_SUB_MAP)) != 0)
1608 if (prev != &map->header) {
1609 prevsize = prev->end - prev->start;
1610 if ( (prev->end == entry->start) &&
1611 (prev->object.vm_object == entry->object.vm_object) &&
1612 (!prev->object.vm_object ||
1613 (prev->offset + prevsize == entry->offset)) &&
1614 (prev->eflags == entry->eflags) &&
1615 (prev->protection == entry->protection) &&
1616 (prev->max_protection == entry->max_protection) &&
1617 (prev->inheritance == entry->inheritance) &&
1618 (prev->wired_count == entry->wired_count) &&
1619 (prev->cred == entry->cred)) {
1620 vm_map_entry_unlink(map, prev);
1621 entry->start = prev->start;
1622 entry->offset = prev->offset;
1623 if (entry->prev != &map->header)
1624 vm_map_entry_resize_free(map, entry->prev);
1627 * If the backing object is a vnode object,
1628 * vm_object_deallocate() calls vrele().
1629 * However, vrele() does not lock the vnode
1630 * because the vnode has additional
1631 * references. Thus, the map lock can be kept
1632 * without causing a lock-order reversal with
1635 * Since we count the number of virtual page
1636 * mappings in object->un_pager.vnp.writemappings,
1637 * the writemappings value should not be adjusted
1638 * when the entry is disposed of.
1640 if (prev->object.vm_object)
1641 vm_object_deallocate(prev->object.vm_object);
1642 if (prev->cred != NULL)
1644 vm_map_entry_dispose(map, prev);
1649 if (next != &map->header) {
1650 esize = entry->end - entry->start;
1651 if ((entry->end == next->start) &&
1652 (next->object.vm_object == entry->object.vm_object) &&
1653 (!entry->object.vm_object ||
1654 (entry->offset + esize == next->offset)) &&
1655 (next->eflags == entry->eflags) &&
1656 (next->protection == entry->protection) &&
1657 (next->max_protection == entry->max_protection) &&
1658 (next->inheritance == entry->inheritance) &&
1659 (next->wired_count == entry->wired_count) &&
1660 (next->cred == entry->cred)) {
1661 vm_map_entry_unlink(map, next);
1662 entry->end = next->end;
1663 vm_map_entry_resize_free(map, entry);
1666 * See comment above.
1668 if (next->object.vm_object)
1669 vm_object_deallocate(next->object.vm_object);
1670 if (next->cred != NULL)
1672 vm_map_entry_dispose(map, next);
1677 * vm_map_clip_start: [ internal use only ]
1679 * Asserts that the given entry begins at or after
1680 * the specified address; if necessary,
1681 * it splits the entry into two.
1683 #define vm_map_clip_start(map, entry, startaddr) \
1685 if (startaddr > entry->start) \
1686 _vm_map_clip_start(map, entry, startaddr); \
1690 * This routine is called only when it is known that
1691 * the entry must be split.
1694 _vm_map_clip_start(vm_map_t map, vm_map_entry_t entry, vm_offset_t start)
1696 vm_map_entry_t new_entry;
1698 VM_MAP_ASSERT_LOCKED(map);
1699 KASSERT(entry->end > start && entry->start < start,
1700 ("_vm_map_clip_start: invalid clip of entry %p", entry));
1703 * Split off the front portion -- note that we must insert the new
1704 * entry BEFORE this one, so that this entry has the specified
1707 vm_map_simplify_entry(map, entry);
1710 * If there is no object backing this entry, we might as well create
1711 * one now. If we defer it, an object can get created after the map
1712 * is clipped, and individual objects will be created for the split-up
1713 * map. This is a bit of a hack, but is also about the best place to
1714 * put this improvement.
1716 if (entry->object.vm_object == NULL && !map->system_map &&
1717 (entry->eflags & MAP_ENTRY_GUARD) == 0) {
1719 object = vm_object_allocate(OBJT_DEFAULT,
1720 atop(entry->end - entry->start));
1721 entry->object.vm_object = object;
1723 if (entry->cred != NULL) {
1724 object->cred = entry->cred;
1725 object->charge = entry->end - entry->start;
1728 } else if (entry->object.vm_object != NULL &&
1729 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) &&
1730 entry->cred != NULL) {
1731 VM_OBJECT_WLOCK(entry->object.vm_object);
1732 KASSERT(entry->object.vm_object->cred == NULL,
1733 ("OVERCOMMIT: vm_entry_clip_start: both cred e %p", entry));
1734 entry->object.vm_object->cred = entry->cred;
1735 entry->object.vm_object->charge = entry->end - entry->start;
1736 VM_OBJECT_WUNLOCK(entry->object.vm_object);
1740 new_entry = vm_map_entry_create(map);
1741 *new_entry = *entry;
1743 new_entry->end = start;
1744 entry->offset += (start - entry->start);
1745 entry->start = start;
1746 if (new_entry->cred != NULL)
1747 crhold(entry->cred);
1749 vm_map_entry_link(map, entry->prev, new_entry);
1751 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1752 vm_object_reference(new_entry->object.vm_object);
1754 * The object->un_pager.vnp.writemappings for the
1755 * object of MAP_ENTRY_VN_WRITECNT type entry shall be
1756 * kept as is here. The virtual pages are
1757 * re-distributed among the clipped entries, so the sum is
1764 * vm_map_clip_end: [ internal use only ]
1766 * Asserts that the given entry ends at or before
1767 * the specified address; if necessary,
1768 * it splits the entry into two.
1770 #define vm_map_clip_end(map, entry, endaddr) \
1772 if ((endaddr) < (entry->end)) \
1773 _vm_map_clip_end((map), (entry), (endaddr)); \
1777 * This routine is called only when it is known that
1778 * the entry must be split.
1781 _vm_map_clip_end(vm_map_t map, vm_map_entry_t entry, vm_offset_t end)
1783 vm_map_entry_t new_entry;
1785 VM_MAP_ASSERT_LOCKED(map);
1786 KASSERT(entry->start < end && entry->end > end,
1787 ("_vm_map_clip_end: invalid clip of entry %p", entry));
1790 * If there is no object backing this entry, we might as well create
1791 * one now. If we defer it, an object can get created after the map
1792 * is clipped, and individual objects will be created for the split-up
1793 * map. This is a bit of a hack, but is also about the best place to
1794 * put this improvement.
1796 if (entry->object.vm_object == NULL && !map->system_map &&
1797 (entry->eflags & MAP_ENTRY_GUARD) == 0) {
1799 object = vm_object_allocate(OBJT_DEFAULT,
1800 atop(entry->end - entry->start));
1801 entry->object.vm_object = object;
1803 if (entry->cred != NULL) {
1804 object->cred = entry->cred;
1805 object->charge = entry->end - entry->start;
1808 } else if (entry->object.vm_object != NULL &&
1809 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) &&
1810 entry->cred != NULL) {
1811 VM_OBJECT_WLOCK(entry->object.vm_object);
1812 KASSERT(entry->object.vm_object->cred == NULL,
1813 ("OVERCOMMIT: vm_entry_clip_end: both cred e %p", entry));
1814 entry->object.vm_object->cred = entry->cred;
1815 entry->object.vm_object->charge = entry->end - entry->start;
1816 VM_OBJECT_WUNLOCK(entry->object.vm_object);
1821 * Create a new entry and insert it AFTER the specified entry
1823 new_entry = vm_map_entry_create(map);
1824 *new_entry = *entry;
1826 new_entry->start = entry->end = end;
1827 new_entry->offset += (end - entry->start);
1828 if (new_entry->cred != NULL)
1829 crhold(entry->cred);
1831 vm_map_entry_link(map, entry, new_entry);
1833 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1834 vm_object_reference(new_entry->object.vm_object);
1839 * vm_map_submap: [ kernel use only ]
1841 * Mark the given range as handled by a subordinate map.
1843 * This range must have been created with vm_map_find,
1844 * and no other operations may have been performed on this
1845 * range prior to calling vm_map_submap.
1847 * Only a limited number of operations can be performed
1848 * within this rage after calling vm_map_submap:
1850 * [Don't try vm_map_copy!]
1852 * To remove a submapping, one must first remove the
1853 * range from the superior map, and then destroy the
1854 * submap (if desired). [Better yet, don't try it.]
1863 vm_map_entry_t entry;
1864 int result = KERN_INVALID_ARGUMENT;
1868 VM_MAP_RANGE_CHECK(map, start, end);
1870 if (vm_map_lookup_entry(map, start, &entry)) {
1871 vm_map_clip_start(map, entry, start);
1873 entry = entry->next;
1875 vm_map_clip_end(map, entry, end);
1877 if ((entry->start == start) && (entry->end == end) &&
1878 ((entry->eflags & MAP_ENTRY_COW) == 0) &&
1879 (entry->object.vm_object == NULL)) {
1880 entry->object.sub_map = submap;
1881 entry->eflags |= MAP_ENTRY_IS_SUB_MAP;
1882 result = KERN_SUCCESS;
1890 * The maximum number of pages to map if MAP_PREFAULT_PARTIAL is specified
1892 #define MAX_INIT_PT 96
1895 * vm_map_pmap_enter:
1897 * Preload the specified map's pmap with mappings to the specified
1898 * object's memory-resident pages. No further physical pages are
1899 * allocated, and no further virtual pages are retrieved from secondary
1900 * storage. If the specified flags include MAP_PREFAULT_PARTIAL, then a
1901 * limited number of page mappings are created at the low-end of the
1902 * specified address range. (For this purpose, a superpage mapping
1903 * counts as one page mapping.) Otherwise, all resident pages within
1904 * the specified address range are mapped.
1907 vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot,
1908 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags)
1911 vm_page_t p, p_start;
1912 vm_pindex_t mask, psize, threshold, tmpidx;
1914 if ((prot & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0 || object == NULL)
1916 VM_OBJECT_RLOCK(object);
1917 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) {
1918 VM_OBJECT_RUNLOCK(object);
1919 VM_OBJECT_WLOCK(object);
1920 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) {
1921 pmap_object_init_pt(map->pmap, addr, object, pindex,
1923 VM_OBJECT_WUNLOCK(object);
1926 VM_OBJECT_LOCK_DOWNGRADE(object);
1930 if (psize + pindex > object->size) {
1931 if (object->size < pindex) {
1932 VM_OBJECT_RUNLOCK(object);
1935 psize = object->size - pindex;
1940 threshold = MAX_INIT_PT;
1942 p = vm_page_find_least(object, pindex);
1944 * Assert: the variable p is either (1) the page with the
1945 * least pindex greater than or equal to the parameter pindex
1949 p != NULL && (tmpidx = p->pindex - pindex) < psize;
1950 p = TAILQ_NEXT(p, listq)) {
1952 * don't allow an madvise to blow away our really
1953 * free pages allocating pv entries.
1955 if (((flags & MAP_PREFAULT_MADVISE) != 0 &&
1956 vm_cnt.v_free_count < vm_cnt.v_free_reserved) ||
1957 ((flags & MAP_PREFAULT_PARTIAL) != 0 &&
1958 tmpidx >= threshold)) {
1962 if (p->valid == VM_PAGE_BITS_ALL) {
1963 if (p_start == NULL) {
1964 start = addr + ptoa(tmpidx);
1967 /* Jump ahead if a superpage mapping is possible. */
1968 if (p->psind > 0 && ((addr + ptoa(tmpidx)) &
1969 (pagesizes[p->psind] - 1)) == 0) {
1970 mask = atop(pagesizes[p->psind]) - 1;
1971 if (tmpidx + mask < psize &&
1972 vm_page_ps_test(p, PS_ALL_VALID, NULL)) {
1977 } else if (p_start != NULL) {
1978 pmap_enter_object(map->pmap, start, addr +
1979 ptoa(tmpidx), p_start, prot);
1983 if (p_start != NULL)
1984 pmap_enter_object(map->pmap, start, addr + ptoa(psize),
1986 VM_OBJECT_RUNLOCK(object);
1992 * Sets the protection of the specified address
1993 * region in the target map. If "set_max" is
1994 * specified, the maximum protection is to be set;
1995 * otherwise, only the current protection is affected.
1998 vm_map_protect(vm_map_t map, vm_offset_t start, vm_offset_t end,
1999 vm_prot_t new_prot, boolean_t set_max)
2001 vm_map_entry_t current, entry;
2007 return (KERN_SUCCESS);
2012 * Ensure that we are not concurrently wiring pages. vm_map_wire() may
2013 * need to fault pages into the map and will drop the map lock while
2014 * doing so, and the VM object may end up in an inconsistent state if we
2015 * update the protection on the map entry in between faults.
2017 vm_map_wait_busy(map);
2019 VM_MAP_RANGE_CHECK(map, start, end);
2021 if (vm_map_lookup_entry(map, start, &entry)) {
2022 vm_map_clip_start(map, entry, start);
2024 entry = entry->next;
2028 * Make a first pass to check for protection violations.
2030 for (current = entry; current->start < end; current = current->next) {
2031 if ((current->eflags & MAP_ENTRY_GUARD) != 0)
2033 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
2035 return (KERN_INVALID_ARGUMENT);
2037 if ((new_prot & current->max_protection) != new_prot) {
2039 return (KERN_PROTECTION_FAILURE);
2044 * Do an accounting pass for private read-only mappings that
2045 * now will do cow due to allowed write (e.g. debugger sets
2046 * breakpoint on text segment)
2048 for (current = entry; current->start < end; current = current->next) {
2050 vm_map_clip_end(map, current, end);
2053 ((new_prot & ~(current->protection)) & VM_PROT_WRITE) == 0 ||
2054 ENTRY_CHARGED(current) ||
2055 (current->eflags & MAP_ENTRY_GUARD) != 0) {
2059 cred = curthread->td_ucred;
2060 obj = current->object.vm_object;
2062 if (obj == NULL || (current->eflags & MAP_ENTRY_NEEDS_COPY)) {
2063 if (!swap_reserve(current->end - current->start)) {
2065 return (KERN_RESOURCE_SHORTAGE);
2068 current->cred = cred;
2072 VM_OBJECT_WLOCK(obj);
2073 if (obj->type != OBJT_DEFAULT && obj->type != OBJT_SWAP) {
2074 VM_OBJECT_WUNLOCK(obj);
2079 * Charge for the whole object allocation now, since
2080 * we cannot distinguish between non-charged and
2081 * charged clipped mapping of the same object later.
2083 KASSERT(obj->charge == 0,
2084 ("vm_map_protect: object %p overcharged (entry %p)",
2086 if (!swap_reserve(ptoa(obj->size))) {
2087 VM_OBJECT_WUNLOCK(obj);
2089 return (KERN_RESOURCE_SHORTAGE);
2094 obj->charge = ptoa(obj->size);
2095 VM_OBJECT_WUNLOCK(obj);
2099 * Go back and fix up protections. [Note that clipping is not
2100 * necessary the second time.]
2102 for (current = entry; current->start < end; current = current->next) {
2103 if ((current->eflags & MAP_ENTRY_GUARD) != 0)
2106 old_prot = current->protection;
2109 current->protection =
2110 (current->max_protection = new_prot) &
2113 current->protection = new_prot;
2116 * For user wired map entries, the normal lazy evaluation of
2117 * write access upgrades through soft page faults is
2118 * undesirable. Instead, immediately copy any pages that are
2119 * copy-on-write and enable write access in the physical map.
2121 if ((current->eflags & MAP_ENTRY_USER_WIRED) != 0 &&
2122 (current->protection & VM_PROT_WRITE) != 0 &&
2123 (old_prot & VM_PROT_WRITE) == 0)
2124 vm_fault_copy_entry(map, map, current, current, NULL);
2127 * When restricting access, update the physical map. Worry
2128 * about copy-on-write here.
2130 if ((old_prot & ~current->protection) != 0) {
2131 #define MASK(entry) (((entry)->eflags & MAP_ENTRY_COW) ? ~VM_PROT_WRITE : \
2133 pmap_protect(map->pmap, current->start,
2135 current->protection & MASK(current));
2138 vm_map_simplify_entry(map, current);
2141 return (KERN_SUCCESS);
2147 * This routine traverses a processes map handling the madvise
2148 * system call. Advisories are classified as either those effecting
2149 * the vm_map_entry structure, or those effecting the underlying
2159 vm_map_entry_t current, entry;
2163 * Some madvise calls directly modify the vm_map_entry, in which case
2164 * we need to use an exclusive lock on the map and we need to perform
2165 * various clipping operations. Otherwise we only need a read-lock
2170 case MADV_SEQUENTIAL:
2177 return (KERN_SUCCESS);
2185 return (KERN_SUCCESS);
2186 vm_map_lock_read(map);
2189 return (KERN_INVALID_ARGUMENT);
2193 * Locate starting entry and clip if necessary.
2195 VM_MAP_RANGE_CHECK(map, start, end);
2197 if (vm_map_lookup_entry(map, start, &entry)) {
2199 vm_map_clip_start(map, entry, start);
2201 entry = entry->next;
2206 * madvise behaviors that are implemented in the vm_map_entry.
2208 * We clip the vm_map_entry so that behavioral changes are
2209 * limited to the specified address range.
2211 for (current = entry; current->start < end;
2212 current = current->next) {
2213 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
2216 vm_map_clip_end(map, current, end);
2220 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_NORMAL);
2222 case MADV_SEQUENTIAL:
2223 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_SEQUENTIAL);
2226 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_RANDOM);
2229 current->eflags |= MAP_ENTRY_NOSYNC;
2232 current->eflags &= ~MAP_ENTRY_NOSYNC;
2235 current->eflags |= MAP_ENTRY_NOCOREDUMP;
2238 current->eflags &= ~MAP_ENTRY_NOCOREDUMP;
2243 vm_map_simplify_entry(map, current);
2247 vm_pindex_t pstart, pend;
2250 * madvise behaviors that are implemented in the underlying
2253 * Since we don't clip the vm_map_entry, we have to clip
2254 * the vm_object pindex and count.
2256 for (current = entry; current->start < end;
2257 current = current->next) {
2258 vm_offset_t useEnd, useStart;
2260 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
2263 pstart = OFF_TO_IDX(current->offset);
2264 pend = pstart + atop(current->end - current->start);
2265 useStart = current->start;
2266 useEnd = current->end;
2268 if (current->start < start) {
2269 pstart += atop(start - current->start);
2272 if (current->end > end) {
2273 pend -= atop(current->end - end);
2281 * Perform the pmap_advise() before clearing
2282 * PGA_REFERENCED in vm_page_advise(). Otherwise, a
2283 * concurrent pmap operation, such as pmap_remove(),
2284 * could clear a reference in the pmap and set
2285 * PGA_REFERENCED on the page before the pmap_advise()
2286 * had completed. Consequently, the page would appear
2287 * referenced based upon an old reference that
2288 * occurred before this pmap_advise() ran.
2290 if (behav == MADV_DONTNEED || behav == MADV_FREE)
2291 pmap_advise(map->pmap, useStart, useEnd,
2294 vm_object_madvise(current->object.vm_object, pstart,
2298 * Pre-populate paging structures in the
2299 * WILLNEED case. For wired entries, the
2300 * paging structures are already populated.
2302 if (behav == MADV_WILLNEED &&
2303 current->wired_count == 0) {
2304 vm_map_pmap_enter(map,
2306 current->protection,
2307 current->object.vm_object,
2309 ptoa(pend - pstart),
2310 MAP_PREFAULT_MADVISE
2314 vm_map_unlock_read(map);
2323 * Sets the inheritance of the specified address
2324 * range in the target map. Inheritance
2325 * affects how the map will be shared with
2326 * child maps at the time of vmspace_fork.
2329 vm_map_inherit(vm_map_t map, vm_offset_t start, vm_offset_t end,
2330 vm_inherit_t new_inheritance)
2332 vm_map_entry_t entry;
2333 vm_map_entry_t temp_entry;
2335 switch (new_inheritance) {
2336 case VM_INHERIT_NONE:
2337 case VM_INHERIT_COPY:
2338 case VM_INHERIT_SHARE:
2339 case VM_INHERIT_ZERO:
2342 return (KERN_INVALID_ARGUMENT);
2345 return (KERN_SUCCESS);
2347 VM_MAP_RANGE_CHECK(map, start, end);
2348 if (vm_map_lookup_entry(map, start, &temp_entry)) {
2350 vm_map_clip_start(map, entry, start);
2352 entry = temp_entry->next;
2353 while (entry->start < end) {
2354 vm_map_clip_end(map, entry, end);
2355 if ((entry->eflags & MAP_ENTRY_GUARD) == 0 ||
2356 new_inheritance != VM_INHERIT_ZERO)
2357 entry->inheritance = new_inheritance;
2358 vm_map_simplify_entry(map, entry);
2359 entry = entry->next;
2362 return (KERN_SUCCESS);
2368 * Implements both kernel and user unwiring.
2371 vm_map_unwire(vm_map_t map, vm_offset_t start, vm_offset_t end,
2374 vm_map_entry_t entry, first_entry, tmp_entry;
2375 vm_offset_t saved_start;
2376 unsigned int last_timestamp;
2378 boolean_t need_wakeup, result, user_unwire;
2381 return (KERN_SUCCESS);
2382 user_unwire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE;
2384 VM_MAP_RANGE_CHECK(map, start, end);
2385 if (!vm_map_lookup_entry(map, start, &first_entry)) {
2386 if (flags & VM_MAP_WIRE_HOLESOK)
2387 first_entry = first_entry->next;
2390 return (KERN_INVALID_ADDRESS);
2393 last_timestamp = map->timestamp;
2394 entry = first_entry;
2395 while (entry->start < end) {
2396 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
2398 * We have not yet clipped the entry.
2400 saved_start = (start >= entry->start) ? start :
2402 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2403 if (vm_map_unlock_and_wait(map, 0)) {
2405 * Allow interruption of user unwiring?
2409 if (last_timestamp+1 != map->timestamp) {
2411 * Look again for the entry because the map was
2412 * modified while it was unlocked.
2413 * Specifically, the entry may have been
2414 * clipped, merged, or deleted.
2416 if (!vm_map_lookup_entry(map, saved_start,
2418 if (flags & VM_MAP_WIRE_HOLESOK)
2419 tmp_entry = tmp_entry->next;
2421 if (saved_start == start) {
2423 * First_entry has been deleted.
2426 return (KERN_INVALID_ADDRESS);
2429 rv = KERN_INVALID_ADDRESS;
2433 if (entry == first_entry)
2434 first_entry = tmp_entry;
2439 last_timestamp = map->timestamp;
2442 vm_map_clip_start(map, entry, start);
2443 vm_map_clip_end(map, entry, end);
2445 * Mark the entry in case the map lock is released. (See
2448 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 &&
2449 entry->wiring_thread == NULL,
2450 ("owned map entry %p", entry));
2451 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
2452 entry->wiring_thread = curthread;
2454 * Check the map for holes in the specified region.
2455 * If VM_MAP_WIRE_HOLESOK was specified, skip this check.
2457 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) &&
2458 (entry->end < end && entry->next->start > entry->end)) {
2460 rv = KERN_INVALID_ADDRESS;
2464 * If system unwiring, require that the entry is system wired.
2467 vm_map_entry_system_wired_count(entry) == 0) {
2469 rv = KERN_INVALID_ARGUMENT;
2472 entry = entry->next;
2476 need_wakeup = FALSE;
2477 if (first_entry == NULL) {
2478 result = vm_map_lookup_entry(map, start, &first_entry);
2479 if (!result && (flags & VM_MAP_WIRE_HOLESOK))
2480 first_entry = first_entry->next;
2482 KASSERT(result, ("vm_map_unwire: lookup failed"));
2484 for (entry = first_entry; entry->start < end; entry = entry->next) {
2486 * If VM_MAP_WIRE_HOLESOK was specified, an empty
2487 * space in the unwired region could have been mapped
2488 * while the map lock was dropped for draining
2489 * MAP_ENTRY_IN_TRANSITION. Moreover, another thread
2490 * could be simultaneously wiring this new mapping
2491 * entry. Detect these cases and skip any entries
2492 * marked as in transition by us.
2494 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 ||
2495 entry->wiring_thread != curthread) {
2496 KASSERT((flags & VM_MAP_WIRE_HOLESOK) != 0,
2497 ("vm_map_unwire: !HOLESOK and new/changed entry"));
2501 if (rv == KERN_SUCCESS && (!user_unwire ||
2502 (entry->eflags & MAP_ENTRY_USER_WIRED))) {
2504 entry->eflags &= ~MAP_ENTRY_USER_WIRED;
2505 if (entry->wired_count == 1)
2506 vm_map_entry_unwire(map, entry);
2508 entry->wired_count--;
2510 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0,
2511 ("vm_map_unwire: in-transition flag missing %p", entry));
2512 KASSERT(entry->wiring_thread == curthread,
2513 ("vm_map_unwire: alien wire %p", entry));
2514 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION;
2515 entry->wiring_thread = NULL;
2516 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
2517 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
2520 vm_map_simplify_entry(map, entry);
2529 * vm_map_wire_entry_failure:
2531 * Handle a wiring failure on the given entry.
2533 * The map should be locked.
2536 vm_map_wire_entry_failure(vm_map_t map, vm_map_entry_t entry,
2537 vm_offset_t failed_addr)
2540 VM_MAP_ASSERT_LOCKED(map);
2541 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 &&
2542 entry->wired_count == 1,
2543 ("vm_map_wire_entry_failure: entry %p isn't being wired", entry));
2544 KASSERT(failed_addr < entry->end,
2545 ("vm_map_wire_entry_failure: entry %p was fully wired", entry));
2548 * If any pages at the start of this entry were successfully wired,
2551 if (failed_addr > entry->start) {
2552 pmap_unwire(map->pmap, entry->start, failed_addr);
2553 vm_object_unwire(entry->object.vm_object, entry->offset,
2554 failed_addr - entry->start, PQ_ACTIVE);
2558 * Assign an out-of-range value to represent the failure to wire this
2561 entry->wired_count = -1;
2567 * Implements both kernel and user wiring.
2570 vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t end,
2573 vm_map_entry_t entry, first_entry, tmp_entry;
2574 vm_offset_t faddr, saved_end, saved_start;
2575 unsigned int last_timestamp;
2577 boolean_t need_wakeup, result, user_wire;
2581 return (KERN_SUCCESS);
2583 if (flags & VM_MAP_WIRE_WRITE)
2584 prot |= VM_PROT_WRITE;
2585 user_wire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE;
2587 VM_MAP_RANGE_CHECK(map, start, end);
2588 if (!vm_map_lookup_entry(map, start, &first_entry)) {
2589 if (flags & VM_MAP_WIRE_HOLESOK)
2590 first_entry = first_entry->next;
2593 return (KERN_INVALID_ADDRESS);
2596 last_timestamp = map->timestamp;
2597 entry = first_entry;
2598 while (entry->start < end) {
2599 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
2601 * We have not yet clipped the entry.
2603 saved_start = (start >= entry->start) ? start :
2605 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2606 if (vm_map_unlock_and_wait(map, 0)) {
2608 * Allow interruption of user wiring?
2612 if (last_timestamp + 1 != map->timestamp) {
2614 * Look again for the entry because the map was
2615 * modified while it was unlocked.
2616 * Specifically, the entry may have been
2617 * clipped, merged, or deleted.
2619 if (!vm_map_lookup_entry(map, saved_start,
2621 if (flags & VM_MAP_WIRE_HOLESOK)
2622 tmp_entry = tmp_entry->next;
2624 if (saved_start == start) {
2626 * first_entry has been deleted.
2629 return (KERN_INVALID_ADDRESS);
2632 rv = KERN_INVALID_ADDRESS;
2636 if (entry == first_entry)
2637 first_entry = tmp_entry;
2642 last_timestamp = map->timestamp;
2645 vm_map_clip_start(map, entry, start);
2646 vm_map_clip_end(map, entry, end);
2648 * Mark the entry in case the map lock is released. (See
2651 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 &&
2652 entry->wiring_thread == NULL,
2653 ("owned map entry %p", entry));
2654 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
2655 entry->wiring_thread = curthread;
2656 if ((entry->protection & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0
2657 || (entry->protection & prot) != prot) {
2658 entry->eflags |= MAP_ENTRY_WIRE_SKIPPED;
2659 if ((flags & VM_MAP_WIRE_HOLESOK) == 0) {
2661 rv = KERN_INVALID_ADDRESS;
2666 if (entry->wired_count == 0) {
2667 entry->wired_count++;
2668 saved_start = entry->start;
2669 saved_end = entry->end;
2672 * Release the map lock, relying on the in-transition
2673 * mark. Mark the map busy for fork.
2678 faddr = saved_start;
2681 * Simulate a fault to get the page and enter
2682 * it into the physical map.
2684 if ((rv = vm_fault(map, faddr, VM_PROT_NONE,
2685 VM_FAULT_WIRE)) != KERN_SUCCESS)
2687 } while ((faddr += PAGE_SIZE) < saved_end);
2690 if (last_timestamp + 1 != map->timestamp) {
2692 * Look again for the entry because the map was
2693 * modified while it was unlocked. The entry
2694 * may have been clipped, but NOT merged or
2697 result = vm_map_lookup_entry(map, saved_start,
2699 KASSERT(result, ("vm_map_wire: lookup failed"));
2700 if (entry == first_entry)
2701 first_entry = tmp_entry;
2705 while (entry->end < saved_end) {
2707 * In case of failure, handle entries
2708 * that were not fully wired here;
2709 * fully wired entries are handled
2712 if (rv != KERN_SUCCESS &&
2714 vm_map_wire_entry_failure(map,
2716 entry = entry->next;
2719 last_timestamp = map->timestamp;
2720 if (rv != KERN_SUCCESS) {
2721 vm_map_wire_entry_failure(map, entry, faddr);
2725 } else if (!user_wire ||
2726 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
2727 entry->wired_count++;
2730 * Check the map for holes in the specified region.
2731 * If VM_MAP_WIRE_HOLESOK was specified, skip this check.
2734 if ((flags & VM_MAP_WIRE_HOLESOK) == 0 &&
2735 entry->end < end && entry->next->start > entry->end) {
2737 rv = KERN_INVALID_ADDRESS;
2740 entry = entry->next;
2744 need_wakeup = FALSE;
2745 if (first_entry == NULL) {
2746 result = vm_map_lookup_entry(map, start, &first_entry);
2747 if (!result && (flags & VM_MAP_WIRE_HOLESOK))
2748 first_entry = first_entry->next;
2750 KASSERT(result, ("vm_map_wire: lookup failed"));
2752 for (entry = first_entry; entry->start < end; entry = entry->next) {
2754 * If VM_MAP_WIRE_HOLESOK was specified, an empty
2755 * space in the unwired region could have been mapped
2756 * while the map lock was dropped for faulting in the
2757 * pages or draining MAP_ENTRY_IN_TRANSITION.
2758 * Moreover, another thread could be simultaneously
2759 * wiring this new mapping entry. Detect these cases
2760 * and skip any entries marked as in transition not by us.
2762 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 ||
2763 entry->wiring_thread != curthread) {
2764 KASSERT((flags & VM_MAP_WIRE_HOLESOK) != 0,
2765 ("vm_map_wire: !HOLESOK and new/changed entry"));
2769 if ((entry->eflags & MAP_ENTRY_WIRE_SKIPPED) != 0)
2770 goto next_entry_done;
2772 if (rv == KERN_SUCCESS) {
2774 entry->eflags |= MAP_ENTRY_USER_WIRED;
2775 } else if (entry->wired_count == -1) {
2777 * Wiring failed on this entry. Thus, unwiring is
2780 entry->wired_count = 0;
2781 } else if (!user_wire ||
2782 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
2784 * Undo the wiring. Wiring succeeded on this entry
2785 * but failed on a later entry.
2787 if (entry->wired_count == 1)
2788 vm_map_entry_unwire(map, entry);
2790 entry->wired_count--;
2793 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0,
2794 ("vm_map_wire: in-transition flag missing %p", entry));
2795 KASSERT(entry->wiring_thread == curthread,
2796 ("vm_map_wire: alien wire %p", entry));
2797 entry->eflags &= ~(MAP_ENTRY_IN_TRANSITION |
2798 MAP_ENTRY_WIRE_SKIPPED);
2799 entry->wiring_thread = NULL;
2800 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
2801 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
2804 vm_map_simplify_entry(map, entry);
2815 * Push any dirty cached pages in the address range to their pager.
2816 * If syncio is TRUE, dirty pages are written synchronously.
2817 * If invalidate is TRUE, any cached pages are freed as well.
2819 * If the size of the region from start to end is zero, we are
2820 * supposed to flush all modified pages within the region containing
2821 * start. Unfortunately, a region can be split or coalesced with
2822 * neighboring regions, making it difficult to determine what the
2823 * original region was. Therefore, we approximate this requirement by
2824 * flushing the current region containing start.
2826 * Returns an error if any part of the specified range is not mapped.
2834 boolean_t invalidate)
2836 vm_map_entry_t current;
2837 vm_map_entry_t entry;
2840 vm_ooffset_t offset;
2841 unsigned int last_timestamp;
2844 vm_map_lock_read(map);
2845 VM_MAP_RANGE_CHECK(map, start, end);
2846 if (!vm_map_lookup_entry(map, start, &entry)) {
2847 vm_map_unlock_read(map);
2848 return (KERN_INVALID_ADDRESS);
2849 } else if (start == end) {
2850 start = entry->start;
2854 * Make a first pass to check for user-wired memory and holes.
2856 for (current = entry; current->start < end; current = current->next) {
2857 if (invalidate && (current->eflags & MAP_ENTRY_USER_WIRED)) {
2858 vm_map_unlock_read(map);
2859 return (KERN_INVALID_ARGUMENT);
2861 if (end > current->end &&
2862 current->end != current->next->start) {
2863 vm_map_unlock_read(map);
2864 return (KERN_INVALID_ADDRESS);
2869 pmap_remove(map->pmap, start, end);
2873 * Make a second pass, cleaning/uncaching pages from the indicated
2876 for (current = entry; current->start < end;) {
2877 offset = current->offset + (start - current->start);
2878 size = (end <= current->end ? end : current->end) - start;
2879 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
2881 vm_map_entry_t tentry;
2884 smap = current->object.sub_map;
2885 vm_map_lock_read(smap);
2886 (void) vm_map_lookup_entry(smap, offset, &tentry);
2887 tsize = tentry->end - offset;
2890 object = tentry->object.vm_object;
2891 offset = tentry->offset + (offset - tentry->start);
2892 vm_map_unlock_read(smap);
2894 object = current->object.vm_object;
2896 vm_object_reference(object);
2897 last_timestamp = map->timestamp;
2898 vm_map_unlock_read(map);
2899 if (!vm_object_sync(object, offset, size, syncio, invalidate))
2902 vm_object_deallocate(object);
2903 vm_map_lock_read(map);
2904 if (last_timestamp == map->timestamp ||
2905 !vm_map_lookup_entry(map, start, ¤t))
2906 current = current->next;
2909 vm_map_unlock_read(map);
2910 return (failed ? KERN_FAILURE : KERN_SUCCESS);
2914 * vm_map_entry_unwire: [ internal use only ]
2916 * Make the region specified by this entry pageable.
2918 * The map in question should be locked.
2919 * [This is the reason for this routine's existence.]
2922 vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry)
2925 VM_MAP_ASSERT_LOCKED(map);
2926 KASSERT(entry->wired_count > 0,
2927 ("vm_map_entry_unwire: entry %p isn't wired", entry));
2928 pmap_unwire(map->pmap, entry->start, entry->end);
2929 vm_object_unwire(entry->object.vm_object, entry->offset, entry->end -
2930 entry->start, PQ_ACTIVE);
2931 entry->wired_count = 0;
2935 vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map)
2938 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0)
2939 vm_object_deallocate(entry->object.vm_object);
2940 uma_zfree(system_map ? kmapentzone : mapentzone, entry);
2944 * vm_map_entry_delete: [ internal use only ]
2946 * Deallocate the given entry from the target map.
2949 vm_map_entry_delete(vm_map_t map, vm_map_entry_t entry)
2952 vm_pindex_t offidxstart, offidxend, count, size1;
2955 vm_map_entry_unlink(map, entry);
2956 object = entry->object.vm_object;
2958 if ((entry->eflags & MAP_ENTRY_GUARD) != 0) {
2959 MPASS(entry->cred == NULL);
2960 MPASS((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0);
2961 MPASS(object == NULL);
2962 vm_map_entry_deallocate(entry, map->system_map);
2966 size = entry->end - entry->start;
2969 if (entry->cred != NULL) {
2970 swap_release_by_cred(size, entry->cred);
2971 crfree(entry->cred);
2974 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0 &&
2976 KASSERT(entry->cred == NULL || object->cred == NULL ||
2977 (entry->eflags & MAP_ENTRY_NEEDS_COPY),
2978 ("OVERCOMMIT vm_map_entry_delete: both cred %p", entry));
2980 offidxstart = OFF_TO_IDX(entry->offset);
2981 offidxend = offidxstart + count;
2982 VM_OBJECT_WLOCK(object);
2983 if (object->ref_count != 1 && ((object->flags & (OBJ_NOSPLIT |
2984 OBJ_ONEMAPPING)) == OBJ_ONEMAPPING ||
2985 object == kernel_object || object == kmem_object)) {
2986 vm_object_collapse(object);
2989 * The option OBJPR_NOTMAPPED can be passed here
2990 * because vm_map_delete() already performed
2991 * pmap_remove() on the only mapping to this range
2994 vm_object_page_remove(object, offidxstart, offidxend,
2996 if (object->type == OBJT_SWAP)
2997 swap_pager_freespace(object, offidxstart,
2999 if (offidxend >= object->size &&
3000 offidxstart < object->size) {
3001 size1 = object->size;
3002 object->size = offidxstart;
3003 if (object->cred != NULL) {
3004 size1 -= object->size;
3005 KASSERT(object->charge >= ptoa(size1),
3006 ("object %p charge < 0", object));
3007 swap_release_by_cred(ptoa(size1),
3009 object->charge -= ptoa(size1);
3013 VM_OBJECT_WUNLOCK(object);
3015 entry->object.vm_object = NULL;
3016 if (map->system_map)
3017 vm_map_entry_deallocate(entry, TRUE);
3019 entry->next = curthread->td_map_def_user;
3020 curthread->td_map_def_user = entry;
3025 * vm_map_delete: [ internal use only ]
3027 * Deallocates the given address range from the target
3031 vm_map_delete(vm_map_t map, vm_offset_t start, vm_offset_t end)
3033 vm_map_entry_t entry;
3034 vm_map_entry_t first_entry;
3036 VM_MAP_ASSERT_LOCKED(map);
3038 return (KERN_SUCCESS);
3041 * Find the start of the region, and clip it
3043 if (!vm_map_lookup_entry(map, start, &first_entry))
3044 entry = first_entry->next;
3046 entry = first_entry;
3047 vm_map_clip_start(map, entry, start);
3051 * Step through all entries in this region
3053 while (entry->start < end) {
3054 vm_map_entry_t next;
3057 * Wait for wiring or unwiring of an entry to complete.
3058 * Also wait for any system wirings to disappear on
3061 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 ||
3062 (vm_map_pmap(map) != kernel_pmap &&
3063 vm_map_entry_system_wired_count(entry) != 0)) {
3064 unsigned int last_timestamp;
3065 vm_offset_t saved_start;
3066 vm_map_entry_t tmp_entry;
3068 saved_start = entry->start;
3069 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
3070 last_timestamp = map->timestamp;
3071 (void) vm_map_unlock_and_wait(map, 0);
3073 if (last_timestamp + 1 != map->timestamp) {
3075 * Look again for the entry because the map was
3076 * modified while it was unlocked.
3077 * Specifically, the entry may have been
3078 * clipped, merged, or deleted.
3080 if (!vm_map_lookup_entry(map, saved_start,
3082 entry = tmp_entry->next;
3085 vm_map_clip_start(map, entry,
3091 vm_map_clip_end(map, entry, end);
3096 * Unwire before removing addresses from the pmap; otherwise,
3097 * unwiring will put the entries back in the pmap.
3099 if (entry->wired_count != 0)
3100 vm_map_entry_unwire(map, entry);
3103 * Remove mappings for the pages, but only if the
3104 * mappings could exist. For instance, it does not
3105 * make sense to call pmap_remove() for guard entries.
3107 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0 ||
3108 entry->object.vm_object != NULL)
3109 pmap_remove(map->pmap, entry->start, entry->end);
3112 * Delete the entry only after removing all pmap
3113 * entries pointing to its pages. (Otherwise, its
3114 * page frames may be reallocated, and any modify bits
3115 * will be set in the wrong object!)
3117 vm_map_entry_delete(map, entry);
3120 return (KERN_SUCCESS);
3126 * Remove the given address range from the target map.
3127 * This is the exported form of vm_map_delete.
3130 vm_map_remove(vm_map_t map, vm_offset_t start, vm_offset_t end)
3135 VM_MAP_RANGE_CHECK(map, start, end);
3136 result = vm_map_delete(map, start, end);
3142 * vm_map_check_protection:
3144 * Assert that the target map allows the specified privilege on the
3145 * entire address region given. The entire region must be allocated.
3147 * WARNING! This code does not and should not check whether the
3148 * contents of the region is accessible. For example a smaller file
3149 * might be mapped into a larger address space.
3151 * NOTE! This code is also called by munmap().
3153 * The map must be locked. A read lock is sufficient.
3156 vm_map_check_protection(vm_map_t map, vm_offset_t start, vm_offset_t end,
3157 vm_prot_t protection)
3159 vm_map_entry_t entry;
3160 vm_map_entry_t tmp_entry;
3162 if (!vm_map_lookup_entry(map, start, &tmp_entry))
3166 while (start < end) {
3170 if (start < entry->start)
3173 * Check protection associated with entry.
3175 if ((entry->protection & protection) != protection)
3177 /* go to next entry */
3179 entry = entry->next;
3185 * vm_map_copy_entry:
3187 * Copies the contents of the source entry to the destination
3188 * entry. The entries *must* be aligned properly.
3194 vm_map_entry_t src_entry,
3195 vm_map_entry_t dst_entry,
3196 vm_ooffset_t *fork_charge)
3198 vm_object_t src_object;
3199 vm_map_entry_t fake_entry;
3204 VM_MAP_ASSERT_LOCKED(dst_map);
3206 if ((dst_entry->eflags|src_entry->eflags) & MAP_ENTRY_IS_SUB_MAP)
3209 if (src_entry->wired_count == 0 ||
3210 (src_entry->protection & VM_PROT_WRITE) == 0) {
3212 * If the source entry is marked needs_copy, it is already
3215 if ((src_entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0 &&
3216 (src_entry->protection & VM_PROT_WRITE) != 0) {
3217 pmap_protect(src_map->pmap,
3220 src_entry->protection & ~VM_PROT_WRITE);
3224 * Make a copy of the object.
3226 size = src_entry->end - src_entry->start;
3227 if ((src_object = src_entry->object.vm_object) != NULL) {
3228 VM_OBJECT_WLOCK(src_object);
3229 charged = ENTRY_CHARGED(src_entry);
3230 if (src_object->handle == NULL &&
3231 (src_object->type == OBJT_DEFAULT ||
3232 src_object->type == OBJT_SWAP)) {
3233 vm_object_collapse(src_object);
3234 if ((src_object->flags & (OBJ_NOSPLIT |
3235 OBJ_ONEMAPPING)) == OBJ_ONEMAPPING) {
3236 vm_object_split(src_entry);
3238 src_entry->object.vm_object;
3241 vm_object_reference_locked(src_object);
3242 vm_object_clear_flag(src_object, OBJ_ONEMAPPING);
3243 if (src_entry->cred != NULL &&
3244 !(src_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
3245 KASSERT(src_object->cred == NULL,
3246 ("OVERCOMMIT: vm_map_copy_entry: cred %p",
3248 src_object->cred = src_entry->cred;
3249 src_object->charge = size;
3251 VM_OBJECT_WUNLOCK(src_object);
3252 dst_entry->object.vm_object = src_object;
3254 cred = curthread->td_ucred;
3256 dst_entry->cred = cred;
3257 *fork_charge += size;
3258 if (!(src_entry->eflags &
3259 MAP_ENTRY_NEEDS_COPY)) {
3261 src_entry->cred = cred;
3262 *fork_charge += size;
3265 src_entry->eflags |= MAP_ENTRY_COW |
3266 MAP_ENTRY_NEEDS_COPY;
3267 dst_entry->eflags |= MAP_ENTRY_COW |
3268 MAP_ENTRY_NEEDS_COPY;
3269 dst_entry->offset = src_entry->offset;
3270 if (src_entry->eflags & MAP_ENTRY_VN_WRITECNT) {
3272 * MAP_ENTRY_VN_WRITECNT cannot
3273 * indicate write reference from
3274 * src_entry, since the entry is
3275 * marked as needs copy. Allocate a
3276 * fake entry that is used to
3277 * decrement object->un_pager.vnp.writecount
3278 * at the appropriate time. Attach
3279 * fake_entry to the deferred list.
3281 fake_entry = vm_map_entry_create(dst_map);
3282 fake_entry->eflags = MAP_ENTRY_VN_WRITECNT;
3283 src_entry->eflags &= ~MAP_ENTRY_VN_WRITECNT;
3284 vm_object_reference(src_object);
3285 fake_entry->object.vm_object = src_object;
3286 fake_entry->start = src_entry->start;
3287 fake_entry->end = src_entry->end;
3288 fake_entry->next = curthread->td_map_def_user;
3289 curthread->td_map_def_user = fake_entry;
3292 pmap_copy(dst_map->pmap, src_map->pmap,
3293 dst_entry->start, dst_entry->end - dst_entry->start,
3296 dst_entry->object.vm_object = NULL;
3297 dst_entry->offset = 0;
3298 if (src_entry->cred != NULL) {
3299 dst_entry->cred = curthread->td_ucred;
3300 crhold(dst_entry->cred);
3301 *fork_charge += size;
3306 * We don't want to make writeable wired pages copy-on-write.
3307 * Immediately copy these pages into the new map by simulating
3308 * page faults. The new pages are pageable.
3310 vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry,
3316 * vmspace_map_entry_forked:
3317 * Update the newly-forked vmspace each time a map entry is inherited
3318 * or copied. The values for vm_dsize and vm_tsize are approximate
3319 * (and mostly-obsolete ideas in the face of mmap(2) et al.)
3322 vmspace_map_entry_forked(const struct vmspace *vm1, struct vmspace *vm2,
3323 vm_map_entry_t entry)
3325 vm_size_t entrysize;
3328 if ((entry->eflags & MAP_ENTRY_GUARD) != 0)
3330 entrysize = entry->end - entry->start;
3331 vm2->vm_map.size += entrysize;
3332 if (entry->eflags & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP)) {
3333 vm2->vm_ssize += btoc(entrysize);
3334 } else if (entry->start >= (vm_offset_t)vm1->vm_daddr &&
3335 entry->start < (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)) {
3336 newend = MIN(entry->end,
3337 (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize));
3338 vm2->vm_dsize += btoc(newend - entry->start);
3339 } else if (entry->start >= (vm_offset_t)vm1->vm_taddr &&
3340 entry->start < (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)) {
3341 newend = MIN(entry->end,
3342 (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize));
3343 vm2->vm_tsize += btoc(newend - entry->start);
3349 * Create a new process vmspace structure and vm_map
3350 * based on those of an existing process. The new map
3351 * is based on the old map, according to the inheritance
3352 * values on the regions in that map.
3354 * XXX It might be worth coalescing the entries added to the new vmspace.
3356 * The source map must not be locked.
3359 vmspace_fork(struct vmspace *vm1, vm_ooffset_t *fork_charge)
3361 struct vmspace *vm2;
3362 vm_map_t new_map, old_map;
3363 vm_map_entry_t new_entry, old_entry;
3368 old_map = &vm1->vm_map;
3369 /* Copy immutable fields of vm1 to vm2. */
3370 vm2 = vmspace_alloc(vm_map_min(old_map), vm_map_max(old_map), NULL);
3373 vm2->vm_taddr = vm1->vm_taddr;
3374 vm2->vm_daddr = vm1->vm_daddr;
3375 vm2->vm_maxsaddr = vm1->vm_maxsaddr;
3376 vm_map_lock(old_map);
3378 vm_map_wait_busy(old_map);
3379 new_map = &vm2->vm_map;
3380 locked = vm_map_trylock(new_map); /* trylock to silence WITNESS */
3381 KASSERT(locked, ("vmspace_fork: lock failed"));
3383 old_entry = old_map->header.next;
3385 while (old_entry != &old_map->header) {
3386 if (old_entry->eflags & MAP_ENTRY_IS_SUB_MAP)
3387 panic("vm_map_fork: encountered a submap");
3389 inh = old_entry->inheritance;
3390 if ((old_entry->eflags & MAP_ENTRY_GUARD) != 0 &&
3391 inh != VM_INHERIT_NONE)
3392 inh = VM_INHERIT_COPY;
3395 case VM_INHERIT_NONE:
3398 case VM_INHERIT_SHARE:
3400 * Clone the entry, creating the shared object if necessary.
3402 object = old_entry->object.vm_object;
3403 if (object == NULL) {
3404 object = vm_object_allocate(OBJT_DEFAULT,
3405 atop(old_entry->end - old_entry->start));
3406 old_entry->object.vm_object = object;
3407 old_entry->offset = 0;
3408 if (old_entry->cred != NULL) {
3409 object->cred = old_entry->cred;
3410 object->charge = old_entry->end -
3412 old_entry->cred = NULL;
3417 * Add the reference before calling vm_object_shadow
3418 * to insure that a shadow object is created.
3420 vm_object_reference(object);
3421 if (old_entry->eflags & MAP_ENTRY_NEEDS_COPY) {
3422 vm_object_shadow(&old_entry->object.vm_object,
3424 old_entry->end - old_entry->start);
3425 old_entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
3426 /* Transfer the second reference too. */
3427 vm_object_reference(
3428 old_entry->object.vm_object);
3431 * As in vm_map_simplify_entry(), the
3432 * vnode lock will not be acquired in
3433 * this call to vm_object_deallocate().
3435 vm_object_deallocate(object);
3436 object = old_entry->object.vm_object;
3438 VM_OBJECT_WLOCK(object);
3439 vm_object_clear_flag(object, OBJ_ONEMAPPING);
3440 if (old_entry->cred != NULL) {
3441 KASSERT(object->cred == NULL, ("vmspace_fork both cred"));
3442 object->cred = old_entry->cred;
3443 object->charge = old_entry->end - old_entry->start;
3444 old_entry->cred = NULL;
3448 * Assert the correct state of the vnode
3449 * v_writecount while the object is locked, to
3450 * not relock it later for the assertion
3453 if (old_entry->eflags & MAP_ENTRY_VN_WRITECNT &&
3454 object->type == OBJT_VNODE) {
3455 KASSERT(((struct vnode *)object->handle)->
3457 ("vmspace_fork: v_writecount %p", object));
3458 KASSERT(object->un_pager.vnp.writemappings > 0,
3459 ("vmspace_fork: vnp.writecount %p",
3462 VM_OBJECT_WUNLOCK(object);
3465 * Clone the entry, referencing the shared object.
3467 new_entry = vm_map_entry_create(new_map);
3468 *new_entry = *old_entry;
3469 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED |
3470 MAP_ENTRY_IN_TRANSITION);
3471 new_entry->wiring_thread = NULL;
3472 new_entry->wired_count = 0;
3473 if (new_entry->eflags & MAP_ENTRY_VN_WRITECNT) {
3474 vnode_pager_update_writecount(object,
3475 new_entry->start, new_entry->end);
3479 * Insert the entry into the new map -- we know we're
3480 * inserting at the end of the new map.
3482 vm_map_entry_link(new_map, new_map->header.prev,
3484 vmspace_map_entry_forked(vm1, vm2, new_entry);
3487 * Update the physical map
3489 pmap_copy(new_map->pmap, old_map->pmap,
3491 (old_entry->end - old_entry->start),
3495 case VM_INHERIT_COPY:
3497 * Clone the entry and link into the map.
3499 new_entry = vm_map_entry_create(new_map);
3500 *new_entry = *old_entry;
3502 * Copied entry is COW over the old object.
3504 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED |
3505 MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_VN_WRITECNT);
3506 new_entry->wiring_thread = NULL;
3507 new_entry->wired_count = 0;
3508 new_entry->object.vm_object = NULL;
3509 new_entry->cred = NULL;
3510 vm_map_entry_link(new_map, new_map->header.prev,
3512 vmspace_map_entry_forked(vm1, vm2, new_entry);
3513 vm_map_copy_entry(old_map, new_map, old_entry,
3514 new_entry, fork_charge);
3517 case VM_INHERIT_ZERO:
3519 * Create a new anonymous mapping entry modelled from
3522 new_entry = vm_map_entry_create(new_map);
3523 memset(new_entry, 0, sizeof(*new_entry));
3525 new_entry->start = old_entry->start;
3526 new_entry->end = old_entry->end;
3527 new_entry->eflags = old_entry->eflags &
3528 ~(MAP_ENTRY_USER_WIRED | MAP_ENTRY_IN_TRANSITION |
3529 MAP_ENTRY_VN_WRITECNT);
3530 new_entry->protection = old_entry->protection;
3531 new_entry->max_protection = old_entry->max_protection;
3532 new_entry->inheritance = VM_INHERIT_ZERO;
3534 vm_map_entry_link(new_map, new_map->header.prev,
3536 vmspace_map_entry_forked(vm1, vm2, new_entry);
3538 new_entry->cred = curthread->td_ucred;
3539 crhold(new_entry->cred);
3540 *fork_charge += (new_entry->end - new_entry->start);
3544 old_entry = old_entry->next;
3547 * Use inlined vm_map_unlock() to postpone handling the deferred
3548 * map entries, which cannot be done until both old_map and
3549 * new_map locks are released.
3551 sx_xunlock(&old_map->lock);
3552 sx_xunlock(&new_map->lock);
3553 vm_map_process_deferred();
3559 * Create a process's stack for exec_new_vmspace(). This function is never
3560 * asked to wire the newly created stack.
3563 vm_map_stack(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
3564 vm_prot_t prot, vm_prot_t max, int cow)
3566 vm_size_t growsize, init_ssize;
3570 MPASS((map->flags & MAP_WIREFUTURE) == 0);
3571 growsize = sgrowsiz;
3572 init_ssize = (max_ssize < growsize) ? max_ssize : growsize;
3574 vmemlim = lim_cur(curthread, RLIMIT_VMEM);
3575 /* If we would blow our VMEM resource limit, no go */
3576 if (map->size + init_ssize > vmemlim) {
3580 rv = vm_map_stack_locked(map, addrbos, max_ssize, growsize, prot,
3587 static int stack_guard_page = 1;
3588 SYSCTL_INT(_security_bsd, OID_AUTO, stack_guard_page, CTLFLAG_RWTUN,
3589 &stack_guard_page, 0,
3590 "Specifies the number of guard pages for a stack that grows");
3593 vm_map_stack_locked(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
3594 vm_size_t growsize, vm_prot_t prot, vm_prot_t max, int cow)
3596 vm_map_entry_t new_entry, prev_entry;
3597 vm_offset_t bot, gap_bot, gap_top, top;
3598 vm_size_t init_ssize, sgp;
3602 * The stack orientation is piggybacked with the cow argument.
3603 * Extract it into orient and mask the cow argument so that we
3604 * don't pass it around further.
3606 orient = cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP);
3607 KASSERT(orient != 0, ("No stack grow direction"));
3608 KASSERT(orient != (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP),
3611 if (addrbos < vm_map_min(map) ||
3612 addrbos + max_ssize > vm_map_max(map) ||
3613 addrbos + max_ssize <= addrbos)
3614 return (KERN_INVALID_ADDRESS);
3615 sgp = (vm_size_t)stack_guard_page * PAGE_SIZE;
3616 if (sgp >= max_ssize)
3617 return (KERN_INVALID_ARGUMENT);
3619 init_ssize = growsize;
3620 if (max_ssize < init_ssize + sgp)
3621 init_ssize = max_ssize - sgp;
3623 /* If addr is already mapped, no go */
3624 if (vm_map_lookup_entry(map, addrbos, &prev_entry))
3625 return (KERN_NO_SPACE);
3628 * If we can't accommodate max_ssize in the current mapping, no go.
3630 if (prev_entry->next->start < addrbos + max_ssize)
3631 return (KERN_NO_SPACE);
3634 * We initially map a stack of only init_ssize. We will grow as
3635 * needed later. Depending on the orientation of the stack (i.e.
3636 * the grow direction) we either map at the top of the range, the
3637 * bottom of the range or in the middle.
3639 * Note: we would normally expect prot and max to be VM_PROT_ALL,
3640 * and cow to be 0. Possibly we should eliminate these as input
3641 * parameters, and just pass these values here in the insert call.
3643 if (orient == MAP_STACK_GROWS_DOWN) {
3644 bot = addrbos + max_ssize - init_ssize;
3645 top = bot + init_ssize;
3648 } else /* if (orient == MAP_STACK_GROWS_UP) */ {
3650 top = bot + init_ssize;
3652 gap_top = addrbos + max_ssize;
3654 rv = vm_map_insert(map, NULL, 0, bot, top, prot, max, cow);
3655 if (rv != KERN_SUCCESS)
3657 new_entry = prev_entry->next;
3658 KASSERT(new_entry->end == top || new_entry->start == bot,
3659 ("Bad entry start/end for new stack entry"));
3660 KASSERT((orient & MAP_STACK_GROWS_DOWN) == 0 ||
3661 (new_entry->eflags & MAP_ENTRY_GROWS_DOWN) != 0,
3662 ("new entry lacks MAP_ENTRY_GROWS_DOWN"));
3663 KASSERT((orient & MAP_STACK_GROWS_UP) == 0 ||
3664 (new_entry->eflags & MAP_ENTRY_GROWS_UP) != 0,
3665 ("new entry lacks MAP_ENTRY_GROWS_UP"));
3666 rv = vm_map_insert(map, NULL, 0, gap_bot, gap_top, VM_PROT_NONE,
3667 VM_PROT_NONE, MAP_CREATE_GUARD | (orient == MAP_STACK_GROWS_DOWN ?
3668 MAP_CREATE_STACK_GAP_DN : MAP_CREATE_STACK_GAP_UP));
3669 if (rv != KERN_SUCCESS)
3670 (void)vm_map_delete(map, bot, top);
3675 * Attempts to grow a vm stack entry. Returns KERN_SUCCESS if we
3676 * successfully grow the stack.
3679 vm_map_growstack(vm_map_t map, vm_offset_t addr, vm_map_entry_t gap_entry)
3681 vm_map_entry_t stack_entry;
3685 vm_offset_t gap_end, gap_start, grow_start;
3686 size_t grow_amount, guard, max_grow;
3687 rlim_t lmemlim, stacklim, vmemlim;
3689 bool gap_deleted, grow_down, is_procstack;
3701 * Disallow stack growth when the access is performed by a
3702 * debugger or AIO daemon. The reason is that the wrong
3703 * resource limits are applied.
3705 if (map != &p->p_vmspace->vm_map || p->p_textvp == NULL)
3706 return (KERN_FAILURE);
3708 MPASS(!map->system_map);
3710 guard = stack_guard_page * PAGE_SIZE;
3711 lmemlim = lim_cur(curthread, RLIMIT_MEMLOCK);
3712 stacklim = lim_cur(curthread, RLIMIT_STACK);
3713 vmemlim = lim_cur(curthread, RLIMIT_VMEM);
3715 /* If addr is not in a hole for a stack grow area, no need to grow. */
3716 if (gap_entry == NULL && !vm_map_lookup_entry(map, addr, &gap_entry))
3717 return (KERN_FAILURE);
3718 if ((gap_entry->eflags & MAP_ENTRY_GUARD) == 0)
3719 return (KERN_SUCCESS);
3720 if ((gap_entry->eflags & MAP_ENTRY_STACK_GAP_DN) != 0) {
3721 stack_entry = gap_entry->next;
3722 if ((stack_entry->eflags & MAP_ENTRY_GROWS_DOWN) == 0 ||
3723 stack_entry->start != gap_entry->end)
3724 return (KERN_FAILURE);
3725 grow_amount = round_page(stack_entry->start - addr);
3727 } else if ((gap_entry->eflags & MAP_ENTRY_STACK_GAP_UP) != 0) {
3728 stack_entry = gap_entry->prev;
3729 if ((stack_entry->eflags & MAP_ENTRY_GROWS_UP) == 0 ||
3730 stack_entry->end != gap_entry->start)
3731 return (KERN_FAILURE);
3732 grow_amount = round_page(addr + 1 - stack_entry->end);
3735 return (KERN_FAILURE);
3737 max_grow = gap_entry->end - gap_entry->start;
3738 if (guard > max_grow)
3739 return (KERN_NO_SPACE);
3741 if (grow_amount > max_grow)
3742 return (KERN_NO_SPACE);
3745 * If this is the main process stack, see if we're over the stack
3748 is_procstack = addr >= (vm_offset_t)vm->vm_maxsaddr &&
3749 addr < (vm_offset_t)p->p_sysent->sv_usrstack;
3750 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim))
3751 return (KERN_NO_SPACE);
3756 if (is_procstack && racct_set(p, RACCT_STACK,
3757 ctob(vm->vm_ssize) + grow_amount)) {
3759 return (KERN_NO_SPACE);
3765 grow_amount = roundup(grow_amount, sgrowsiz);
3766 if (grow_amount > max_grow)
3767 grow_amount = max_grow;
3768 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) {
3769 grow_amount = trunc_page((vm_size_t)stacklim) -
3775 limit = racct_get_available(p, RACCT_STACK);
3777 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > limit))
3778 grow_amount = limit - ctob(vm->vm_ssize);
3781 if (!old_mlock && (map->flags & MAP_WIREFUTURE) != 0) {
3782 if (ptoa(pmap_wired_count(map->pmap)) + grow_amount > lmemlim) {
3789 if (racct_set(p, RACCT_MEMLOCK,
3790 ptoa(pmap_wired_count(map->pmap)) + grow_amount)) {
3800 /* If we would blow our VMEM resource limit, no go */
3801 if (map->size + grow_amount > vmemlim) {
3808 if (racct_set(p, RACCT_VMEM, map->size + grow_amount)) {
3817 if (vm_map_lock_upgrade(map)) {
3819 vm_map_lock_read(map);
3824 grow_start = gap_entry->end - grow_amount;
3825 if (gap_entry->start + grow_amount == gap_entry->end) {
3826 gap_start = gap_entry->start;
3827 gap_end = gap_entry->end;
3828 vm_map_entry_delete(map, gap_entry);
3831 MPASS(gap_entry->start < gap_entry->end - grow_amount);
3832 gap_entry->end -= grow_amount;
3833 vm_map_entry_resize_free(map, gap_entry);
3834 gap_deleted = false;
3836 rv = vm_map_insert(map, NULL, 0, grow_start,
3837 grow_start + grow_amount,
3838 stack_entry->protection, stack_entry->max_protection,
3839 MAP_STACK_GROWS_DOWN);
3840 if (rv != KERN_SUCCESS) {
3842 rv1 = vm_map_insert(map, NULL, 0, gap_start,
3843 gap_end, VM_PROT_NONE, VM_PROT_NONE,
3844 MAP_CREATE_GUARD | MAP_CREATE_STACK_GAP_DN);
3845 MPASS(rv1 == KERN_SUCCESS);
3847 gap_entry->end += grow_amount;
3848 vm_map_entry_resize_free(map, gap_entry);
3852 grow_start = stack_entry->end;
3853 cred = stack_entry->cred;
3854 if (cred == NULL && stack_entry->object.vm_object != NULL)
3855 cred = stack_entry->object.vm_object->cred;
3856 if (cred != NULL && !swap_reserve_by_cred(grow_amount, cred))
3858 /* Grow the underlying object if applicable. */
3859 else if (stack_entry->object.vm_object == NULL ||
3860 vm_object_coalesce(stack_entry->object.vm_object,
3861 stack_entry->offset,
3862 (vm_size_t)(stack_entry->end - stack_entry->start),
3863 (vm_size_t)grow_amount, cred != NULL)) {
3864 if (gap_entry->start + grow_amount == gap_entry->end)
3865 vm_map_entry_delete(map, gap_entry);
3867 gap_entry->start += grow_amount;
3868 stack_entry->end += grow_amount;
3869 map->size += grow_amount;
3870 vm_map_entry_resize_free(map, stack_entry);
3875 if (rv == KERN_SUCCESS && is_procstack)
3876 vm->vm_ssize += btoc(grow_amount);
3879 * Heed the MAP_WIREFUTURE flag if it was set for this process.
3881 if (rv == KERN_SUCCESS && (map->flags & MAP_WIREFUTURE) != 0) {
3883 vm_map_wire(map, grow_start, grow_start + grow_amount,
3884 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
3885 vm_map_lock_read(map);
3887 vm_map_lock_downgrade(map);
3891 if (racct_enable && rv != KERN_SUCCESS) {
3893 error = racct_set(p, RACCT_VMEM, map->size);
3894 KASSERT(error == 0, ("decreasing RACCT_VMEM failed"));
3896 error = racct_set(p, RACCT_MEMLOCK,
3897 ptoa(pmap_wired_count(map->pmap)));
3898 KASSERT(error == 0, ("decreasing RACCT_MEMLOCK failed"));
3900 error = racct_set(p, RACCT_STACK, ctob(vm->vm_ssize));
3901 KASSERT(error == 0, ("decreasing RACCT_STACK failed"));
3910 * Unshare the specified VM space for exec. If other processes are
3911 * mapped to it, then create a new one. The new vmspace is null.
3914 vmspace_exec(struct proc *p, vm_offset_t minuser, vm_offset_t maxuser)
3916 struct vmspace *oldvmspace = p->p_vmspace;
3917 struct vmspace *newvmspace;
3919 KASSERT((curthread->td_pflags & TDP_EXECVMSPC) == 0,
3920 ("vmspace_exec recursed"));
3921 newvmspace = vmspace_alloc(minuser, maxuser, NULL);
3922 if (newvmspace == NULL)
3924 newvmspace->vm_swrss = oldvmspace->vm_swrss;
3926 * This code is written like this for prototype purposes. The
3927 * goal is to avoid running down the vmspace here, but let the
3928 * other process's that are still using the vmspace to finally
3929 * run it down. Even though there is little or no chance of blocking
3930 * here, it is a good idea to keep this form for future mods.
3932 PROC_VMSPACE_LOCK(p);
3933 p->p_vmspace = newvmspace;
3934 PROC_VMSPACE_UNLOCK(p);
3935 if (p == curthread->td_proc)
3936 pmap_activate(curthread);
3937 curthread->td_pflags |= TDP_EXECVMSPC;
3942 * Unshare the specified VM space for forcing COW. This
3943 * is called by rfork, for the (RFMEM|RFPROC) == 0 case.
3946 vmspace_unshare(struct proc *p)
3948 struct vmspace *oldvmspace = p->p_vmspace;
3949 struct vmspace *newvmspace;
3950 vm_ooffset_t fork_charge;
3952 if (oldvmspace->vm_refcnt == 1)
3955 newvmspace = vmspace_fork(oldvmspace, &fork_charge);
3956 if (newvmspace == NULL)
3958 if (!swap_reserve_by_cred(fork_charge, p->p_ucred)) {
3959 vmspace_free(newvmspace);
3962 PROC_VMSPACE_LOCK(p);
3963 p->p_vmspace = newvmspace;
3964 PROC_VMSPACE_UNLOCK(p);
3965 if (p == curthread->td_proc)
3966 pmap_activate(curthread);
3967 vmspace_free(oldvmspace);
3974 * Finds the VM object, offset, and
3975 * protection for a given virtual address in the
3976 * specified map, assuming a page fault of the
3979 * Leaves the map in question locked for read; return
3980 * values are guaranteed until a vm_map_lookup_done
3981 * call is performed. Note that the map argument
3982 * is in/out; the returned map must be used in
3983 * the call to vm_map_lookup_done.
3985 * A handle (out_entry) is returned for use in
3986 * vm_map_lookup_done, to make that fast.
3988 * If a lookup is requested with "write protection"
3989 * specified, the map may be changed to perform virtual
3990 * copying operations, although the data referenced will
3994 vm_map_lookup(vm_map_t *var_map, /* IN/OUT */
3996 vm_prot_t fault_typea,
3997 vm_map_entry_t *out_entry, /* OUT */
3998 vm_object_t *object, /* OUT */
3999 vm_pindex_t *pindex, /* OUT */
4000 vm_prot_t *out_prot, /* OUT */
4001 boolean_t *wired) /* OUT */
4003 vm_map_entry_t entry;
4004 vm_map_t map = *var_map;
4006 vm_prot_t fault_type = fault_typea;
4007 vm_object_t eobject;
4013 vm_map_lock_read(map);
4017 * Lookup the faulting address.
4019 if (!vm_map_lookup_entry(map, vaddr, out_entry)) {
4020 vm_map_unlock_read(map);
4021 return (KERN_INVALID_ADDRESS);
4029 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
4030 vm_map_t old_map = map;
4032 *var_map = map = entry->object.sub_map;
4033 vm_map_unlock_read(old_map);
4038 * Check whether this task is allowed to have this page.
4040 prot = entry->protection;
4041 if ((fault_typea & VM_PROT_FAULT_LOOKUP) != 0) {
4042 fault_typea &= ~VM_PROT_FAULT_LOOKUP;
4043 if (prot == VM_PROT_NONE && map != kernel_map &&
4044 (entry->eflags & MAP_ENTRY_GUARD) != 0 &&
4045 (entry->eflags & (MAP_ENTRY_STACK_GAP_DN |
4046 MAP_ENTRY_STACK_GAP_UP)) != 0 &&
4047 vm_map_growstack(map, vaddr, entry) == KERN_SUCCESS)
4048 goto RetryLookupLocked;
4050 fault_type &= VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE;
4051 if ((fault_type & prot) != fault_type || prot == VM_PROT_NONE) {
4052 vm_map_unlock_read(map);
4053 return (KERN_PROTECTION_FAILURE);
4055 KASSERT((prot & VM_PROT_WRITE) == 0 || (entry->eflags &
4056 (MAP_ENTRY_USER_WIRED | MAP_ENTRY_NEEDS_COPY)) !=
4057 (MAP_ENTRY_USER_WIRED | MAP_ENTRY_NEEDS_COPY),
4058 ("entry %p flags %x", entry, entry->eflags));
4059 if ((fault_typea & VM_PROT_COPY) != 0 &&
4060 (entry->max_protection & VM_PROT_WRITE) == 0 &&
4061 (entry->eflags & MAP_ENTRY_COW) == 0) {
4062 vm_map_unlock_read(map);
4063 return (KERN_PROTECTION_FAILURE);
4067 * If this page is not pageable, we have to get it for all possible
4070 *wired = (entry->wired_count != 0);
4072 fault_type = entry->protection;
4073 size = entry->end - entry->start;
4075 * If the entry was copy-on-write, we either ...
4077 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
4079 * If we want to write the page, we may as well handle that
4080 * now since we've got the map locked.
4082 * If we don't need to write the page, we just demote the
4083 * permissions allowed.
4085 if ((fault_type & VM_PROT_WRITE) != 0 ||
4086 (fault_typea & VM_PROT_COPY) != 0) {
4088 * Make a new object, and place it in the object
4089 * chain. Note that no new references have appeared
4090 * -- one just moved from the map to the new
4093 if (vm_map_lock_upgrade(map))
4096 if (entry->cred == NULL) {
4098 * The debugger owner is charged for
4101 cred = curthread->td_ucred;
4103 if (!swap_reserve_by_cred(size, cred)) {
4106 return (KERN_RESOURCE_SHORTAGE);
4110 vm_object_shadow(&entry->object.vm_object,
4111 &entry->offset, size);
4112 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
4113 eobject = entry->object.vm_object;
4114 if (eobject->cred != NULL) {
4116 * The object was not shadowed.
4118 swap_release_by_cred(size, entry->cred);
4119 crfree(entry->cred);
4121 } else if (entry->cred != NULL) {
4122 VM_OBJECT_WLOCK(eobject);
4123 eobject->cred = entry->cred;
4124 eobject->charge = size;
4125 VM_OBJECT_WUNLOCK(eobject);
4129 vm_map_lock_downgrade(map);
4132 * We're attempting to read a copy-on-write page --
4133 * don't allow writes.
4135 prot &= ~VM_PROT_WRITE;
4140 * Create an object if necessary.
4142 if (entry->object.vm_object == NULL &&
4144 if (vm_map_lock_upgrade(map))
4146 entry->object.vm_object = vm_object_allocate(OBJT_DEFAULT,
4149 if (entry->cred != NULL) {
4150 VM_OBJECT_WLOCK(entry->object.vm_object);
4151 entry->object.vm_object->cred = entry->cred;
4152 entry->object.vm_object->charge = size;
4153 VM_OBJECT_WUNLOCK(entry->object.vm_object);
4156 vm_map_lock_downgrade(map);
4160 * Return the object/offset from this entry. If the entry was
4161 * copy-on-write or empty, it has been fixed up.
4163 *pindex = UOFF_TO_IDX((vaddr - entry->start) + entry->offset);
4164 *object = entry->object.vm_object;
4167 return (KERN_SUCCESS);
4171 * vm_map_lookup_locked:
4173 * Lookup the faulting address. A version of vm_map_lookup that returns
4174 * KERN_FAILURE instead of blocking on map lock or memory allocation.
4177 vm_map_lookup_locked(vm_map_t *var_map, /* IN/OUT */
4179 vm_prot_t fault_typea,
4180 vm_map_entry_t *out_entry, /* OUT */
4181 vm_object_t *object, /* OUT */
4182 vm_pindex_t *pindex, /* OUT */
4183 vm_prot_t *out_prot, /* OUT */
4184 boolean_t *wired) /* OUT */
4186 vm_map_entry_t entry;
4187 vm_map_t map = *var_map;
4189 vm_prot_t fault_type = fault_typea;
4192 * Lookup the faulting address.
4194 if (!vm_map_lookup_entry(map, vaddr, out_entry))
4195 return (KERN_INVALID_ADDRESS);
4200 * Fail if the entry refers to a submap.
4202 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP)
4203 return (KERN_FAILURE);
4206 * Check whether this task is allowed to have this page.
4208 prot = entry->protection;
4209 fault_type &= VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE;
4210 if ((fault_type & prot) != fault_type)
4211 return (KERN_PROTECTION_FAILURE);
4214 * If this page is not pageable, we have to get it for all possible
4217 *wired = (entry->wired_count != 0);
4219 fault_type = entry->protection;
4221 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
4223 * Fail if the entry was copy-on-write for a write fault.
4225 if (fault_type & VM_PROT_WRITE)
4226 return (KERN_FAILURE);
4228 * We're attempting to read a copy-on-write page --
4229 * don't allow writes.
4231 prot &= ~VM_PROT_WRITE;
4235 * Fail if an object should be created.
4237 if (entry->object.vm_object == NULL && !map->system_map)
4238 return (KERN_FAILURE);
4241 * Return the object/offset from this entry. If the entry was
4242 * copy-on-write or empty, it has been fixed up.
4244 *pindex = UOFF_TO_IDX((vaddr - entry->start) + entry->offset);
4245 *object = entry->object.vm_object;
4248 return (KERN_SUCCESS);
4252 * vm_map_lookup_done:
4254 * Releases locks acquired by a vm_map_lookup
4255 * (according to the handle returned by that lookup).
4258 vm_map_lookup_done(vm_map_t map, vm_map_entry_t entry)
4261 * Unlock the main-level map
4263 vm_map_unlock_read(map);
4267 vm_map_max_KBI(const struct vm_map *map)
4270 return (vm_map_max(map));
4274 vm_map_min_KBI(const struct vm_map *map)
4277 return (vm_map_min(map));
4281 vm_map_pmap_KBI(vm_map_t map)
4287 #include "opt_ddb.h"
4289 #include <sys/kernel.h>
4291 #include <ddb/ddb.h>
4294 vm_map_print(vm_map_t map)
4296 vm_map_entry_t entry;
4298 db_iprintf("Task map %p: pmap=%p, nentries=%d, version=%u\n",
4300 (void *)map->pmap, map->nentries, map->timestamp);
4303 for (entry = map->header.next; entry != &map->header;
4304 entry = entry->next) {
4305 db_iprintf("map entry %p: start=%p, end=%p, eflags=%#x, \n",
4306 (void *)entry, (void *)entry->start, (void *)entry->end,
4309 static char *inheritance_name[4] =
4310 {"share", "copy", "none", "donate_copy"};
4312 db_iprintf(" prot=%x/%x/%s",
4314 entry->max_protection,
4315 inheritance_name[(int)(unsigned char)entry->inheritance]);
4316 if (entry->wired_count != 0)
4317 db_printf(", wired");
4319 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
4320 db_printf(", share=%p, offset=0x%jx\n",
4321 (void *)entry->object.sub_map,
4322 (uintmax_t)entry->offset);
4323 if ((entry->prev == &map->header) ||
4324 (entry->prev->object.sub_map !=
4325 entry->object.sub_map)) {
4327 vm_map_print((vm_map_t)entry->object.sub_map);
4331 if (entry->cred != NULL)
4332 db_printf(", ruid %d", entry->cred->cr_ruid);
4333 db_printf(", object=%p, offset=0x%jx",
4334 (void *)entry->object.vm_object,
4335 (uintmax_t)entry->offset);
4336 if (entry->object.vm_object && entry->object.vm_object->cred)
4337 db_printf(", obj ruid %d charge %jx",
4338 entry->object.vm_object->cred->cr_ruid,
4339 (uintmax_t)entry->object.vm_object->charge);
4340 if (entry->eflags & MAP_ENTRY_COW)
4341 db_printf(", copy (%s)",
4342 (entry->eflags & MAP_ENTRY_NEEDS_COPY) ? "needed" : "done");
4345 if ((entry->prev == &map->header) ||
4346 (entry->prev->object.vm_object !=
4347 entry->object.vm_object)) {
4349 vm_object_print((db_expr_t)(intptr_t)
4350 entry->object.vm_object,
4359 DB_SHOW_COMMAND(map, map)
4363 db_printf("usage: show map <addr>\n");
4366 vm_map_print((vm_map_t)addr);
4369 DB_SHOW_COMMAND(procvm, procvm)
4374 p = db_lookup_proc(addr);
4379 db_printf("p = %p, vmspace = %p, map = %p, pmap = %p\n",
4380 (void *)p, (void *)p->p_vmspace, (void *)&p->p_vmspace->vm_map,
4381 (void *)vmspace_pmap(p->p_vmspace));
4383 vm_map_print((vm_map_t)&p->p_vmspace->vm_map);