2 * SPDX-License-Identifier: (BSD-3-Clause AND MIT-CMU)
4 * Copyright (c) 1991, 1993
5 * The Regents of the University of California. All rights reserved.
7 * This code is derived from software contributed to Berkeley by
8 * The Mach Operating System project at Carnegie-Mellon University.
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 3. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * from: @(#)vm_map.c 8.3 (Berkeley) 1/12/94
37 * Copyright (c) 1987, 1990 Carnegie-Mellon University.
38 * All rights reserved.
40 * Authors: Avadis Tevanian, Jr., Michael Wayne Young
42 * Permission to use, copy, modify and distribute this software and
43 * its documentation is hereby granted, provided that both the copyright
44 * notice and this permission notice appear in all copies of the
45 * software, derivative works or modified versions, and any portions
46 * thereof, and that both notices appear in supporting documentation.
48 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
49 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
50 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
52 * Carnegie Mellon requests users of this software to return to
54 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
55 * School of Computer Science
56 * Carnegie Mellon University
57 * Pittsburgh PA 15213-3890
59 * any improvements or extensions that they make and grant Carnegie the
60 * rights to redistribute these changes.
64 * Virtual memory mapping module.
67 #include <sys/cdefs.h>
68 __FBSDID("$FreeBSD$");
70 #include <sys/param.h>
71 #include <sys/systm.h>
72 #include <sys/kernel.h>
75 #include <sys/mutex.h>
77 #include <sys/vmmeter.h>
79 #include <sys/vnode.h>
80 #include <sys/racct.h>
81 #include <sys/resourcevar.h>
82 #include <sys/rwlock.h>
84 #include <sys/sysctl.h>
85 #include <sys/sysent.h>
89 #include <vm/vm_param.h>
91 #include <vm/vm_map.h>
92 #include <vm/vm_page.h>
93 #include <vm/vm_object.h>
94 #include <vm/vm_pager.h>
95 #include <vm/vm_kern.h>
96 #include <vm/vm_extern.h>
97 #include <vm/vnode_pager.h>
98 #include <vm/swap_pager.h>
102 * Virtual memory maps provide for the mapping, protection,
103 * and sharing of virtual memory objects. In addition,
104 * this module provides for an efficient virtual copy of
105 * memory from one map to another.
107 * Synchronization is required prior to most operations.
109 * Maps consist of an ordered doubly-linked list of simple
110 * entries; a self-adjusting binary search tree of these
111 * entries is used to speed up lookups.
113 * Since portions of maps are specified by start/end addresses,
114 * which may not align with existing map entries, all
115 * routines merely "clip" entries to these start/end values.
116 * [That is, an entry is split into two, bordering at a
117 * start or end value.] Note that these clippings may not
118 * always be necessary (as the two resulting entries are then
119 * not changed); however, the clipping is done for convenience.
121 * As mentioned above, virtual copy operations are performed
122 * by copying VM object references from one map to
123 * another, and then marking both regions as copy-on-write.
126 static struct mtx map_sleep_mtx;
127 static uma_zone_t mapentzone;
128 static uma_zone_t kmapentzone;
129 static uma_zone_t mapzone;
130 static uma_zone_t vmspace_zone;
131 static int vmspace_zinit(void *mem, int size, int flags);
132 static int vm_map_zinit(void *mem, int ize, int flags);
133 static void _vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min,
135 static int vm_map_alignspace(vm_map_t map, vm_object_t object,
136 vm_ooffset_t offset, vm_offset_t *addr, vm_size_t length,
137 vm_offset_t max_addr, vm_offset_t alignment);
138 static void vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map);
139 static void vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry);
140 static void vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry);
141 static int vm_map_growstack(vm_map_t map, vm_offset_t addr,
142 vm_map_entry_t gap_entry);
143 static void vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot,
144 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags);
146 static void vm_map_zdtor(void *mem, int size, void *arg);
147 static void vmspace_zdtor(void *mem, int size, void *arg);
149 static int vm_map_stack_locked(vm_map_t map, vm_offset_t addrbos,
150 vm_size_t max_ssize, vm_size_t growsize, vm_prot_t prot, vm_prot_t max,
152 static void vm_map_wire_entry_failure(vm_map_t map, vm_map_entry_t entry,
153 vm_offset_t failed_addr);
155 #define ENTRY_CHARGED(e) ((e)->cred != NULL || \
156 ((e)->object.vm_object != NULL && (e)->object.vm_object->cred != NULL && \
157 !((e)->eflags & MAP_ENTRY_NEEDS_COPY)))
160 * PROC_VMSPACE_{UN,}LOCK() can be a noop as long as vmspaces are type
163 #define PROC_VMSPACE_LOCK(p) do { } while (0)
164 #define PROC_VMSPACE_UNLOCK(p) do { } while (0)
167 * VM_MAP_RANGE_CHECK: [ internal use only ]
169 * Asserts that the starting and ending region
170 * addresses fall within the valid range of the map.
172 #define VM_MAP_RANGE_CHECK(map, start, end) \
174 if (start < vm_map_min(map)) \
175 start = vm_map_min(map); \
176 if (end > vm_map_max(map)) \
177 end = vm_map_max(map); \
185 * Initialize the vm_map module. Must be called before
186 * any other vm_map routines.
188 * Map and entry structures are allocated from the general
189 * purpose memory pool with some exceptions:
191 * - The kernel map and kmem submap are allocated statically.
192 * - Kernel map entries are allocated out of a static pool.
194 * These restrictions are necessary since malloc() uses the
195 * maps and requires map entries.
201 mtx_init(&map_sleep_mtx, "vm map sleep mutex", NULL, MTX_DEF);
202 mapzone = uma_zcreate("MAP", sizeof(struct vm_map), NULL,
208 vm_map_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
209 uma_prealloc(mapzone, MAX_KMAP);
210 kmapentzone = uma_zcreate("KMAP ENTRY", sizeof(struct vm_map_entry),
211 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR,
212 UMA_ZONE_MTXCLASS | UMA_ZONE_VM);
213 mapentzone = uma_zcreate("MAP ENTRY", sizeof(struct vm_map_entry),
214 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
215 vmspace_zone = uma_zcreate("VMSPACE", sizeof(struct vmspace), NULL,
221 vmspace_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
225 vmspace_zinit(void *mem, int size, int flags)
229 vm = (struct vmspace *)mem;
231 vm->vm_map.pmap = NULL;
232 (void)vm_map_zinit(&vm->vm_map, sizeof(vm->vm_map), flags);
233 PMAP_LOCK_INIT(vmspace_pmap(vm));
238 vm_map_zinit(void *mem, int size, int flags)
243 memset(map, 0, sizeof(*map));
244 mtx_init(&map->system_mtx, "vm map (system)", NULL, MTX_DEF | MTX_DUPOK);
245 sx_init(&map->lock, "vm map (user)");
251 vmspace_zdtor(void *mem, int size, void *arg)
255 vm = (struct vmspace *)mem;
257 vm_map_zdtor(&vm->vm_map, sizeof(vm->vm_map), arg);
260 vm_map_zdtor(void *mem, int size, void *arg)
265 KASSERT(map->nentries == 0,
266 ("map %p nentries == %d on free.",
267 map, map->nentries));
268 KASSERT(map->size == 0,
269 ("map %p size == %lu on free.",
270 map, (unsigned long)map->size));
272 #endif /* INVARIANTS */
275 * Allocate a vmspace structure, including a vm_map and pmap,
276 * and initialize those structures. The refcnt is set to 1.
278 * If 'pinit' is NULL then the embedded pmap is initialized via pmap_pinit().
281 vmspace_alloc(vm_offset_t min, vm_offset_t max, pmap_pinit_t pinit)
285 vm = uma_zalloc(vmspace_zone, M_WAITOK);
286 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);
394 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_fcmpset_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);
440 refcnt = vm->vm_refcnt;
442 if (refcnt <= 0) { /* Avoid 0->1 transition */
443 PROC_VMSPACE_UNLOCK(p);
446 } while (!atomic_fcmpset_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->header.eflags = MAP_ENTRY_HEADER;
795 map->needs_wakeup = FALSE;
798 map->header.end = min;
799 map->header.start = max;
807 vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max)
810 _vm_map_init(map, pmap, min, max);
811 mtx_init(&map->system_mtx, "system map", NULL, MTX_DEF | MTX_DUPOK);
812 sx_init(&map->lock, "user map");
816 * vm_map_entry_dispose: [ internal use only ]
818 * Inverse of vm_map_entry_create.
821 vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry)
823 uma_zfree(map->system_map ? kmapentzone : mapentzone, entry);
827 * vm_map_entry_create: [ internal use only ]
829 * Allocates a VM map entry for insertion.
830 * No entry fields are filled in.
832 static vm_map_entry_t
833 vm_map_entry_create(vm_map_t map)
835 vm_map_entry_t new_entry;
838 new_entry = uma_zalloc(kmapentzone, M_NOWAIT);
840 new_entry = uma_zalloc(mapentzone, M_WAITOK);
841 if (new_entry == NULL)
842 panic("vm_map_entry_create: kernel resources exhausted");
847 * vm_map_entry_set_behavior:
849 * Set the expected access behavior, either normal, random, or
853 vm_map_entry_set_behavior(vm_map_entry_t entry, u_char behavior)
855 entry->eflags = (entry->eflags & ~MAP_ENTRY_BEHAV_MASK) |
856 (behavior & MAP_ENTRY_BEHAV_MASK);
860 * vm_map_entry_set_max_free:
862 * Set the max_free field in a vm_map_entry.
865 vm_map_entry_set_max_free(vm_map_entry_t entry)
868 entry->max_free = entry->adj_free;
869 if (entry->left != NULL && entry->left->max_free > entry->max_free)
870 entry->max_free = entry->left->max_free;
871 if (entry->right != NULL && entry->right->max_free > entry->max_free)
872 entry->max_free = entry->right->max_free;
876 * vm_map_entry_splay:
878 * The Sleator and Tarjan top-down splay algorithm with the
879 * following variation. Max_free must be computed bottom-up, so
880 * on the downward pass, maintain the left and right spines in
881 * reverse order. Then, make a second pass up each side to fix
882 * the pointers and compute max_free. The time bound is O(log n)
885 * The new root is the vm_map_entry containing "addr", or else an
886 * adjacent entry (lower or higher) if addr is not in the tree.
888 * The map must be locked, and leaves it so.
890 * Returns: the new root.
892 static vm_map_entry_t
893 vm_map_entry_splay(vm_offset_t addr, vm_map_entry_t root)
895 vm_map_entry_t llist, rlist;
896 vm_map_entry_t ltree, rtree;
899 /* Special case of empty tree. */
904 * Pass One: Splay down the tree until we find addr or a NULL
905 * pointer where addr would go. llist and rlist are the two
906 * sides in reverse order (bottom-up), with llist linked by
907 * the right pointer and rlist linked by the left pointer in
908 * the vm_map_entry. Wait until Pass Two to set max_free on
914 /* root is never NULL in here. */
915 if (addr < root->start) {
919 if (addr < y->start && y->left != NULL) {
920 /* Rotate right and put y on rlist. */
921 root->left = y->right;
923 vm_map_entry_set_max_free(root);
928 /* Put root on rlist. */
933 } else if (addr >= root->end) {
937 if (addr >= y->end && y->right != NULL) {
938 /* Rotate left and put y on llist. */
939 root->right = y->left;
941 vm_map_entry_set_max_free(root);
946 /* Put root on llist. */
956 * Pass Two: Walk back up the two spines, flip the pointers
957 * and set max_free. The subtrees of the root go at the
958 * bottom of llist and rlist.
961 while (llist != NULL) {
963 llist->right = ltree;
964 vm_map_entry_set_max_free(llist);
969 while (rlist != NULL) {
972 vm_map_entry_set_max_free(rlist);
978 * Final assembly: add ltree and rtree as subtrees of root.
982 vm_map_entry_set_max_free(root);
988 * vm_map_entry_{un,}link:
990 * Insert/remove entries from maps.
993 vm_map_entry_link(vm_map_t map,
994 vm_map_entry_t after_where,
995 vm_map_entry_t entry)
999 "vm_map_entry_link: map %p, nentries %d, entry %p, after %p", map,
1000 map->nentries, entry, after_where);
1001 VM_MAP_ASSERT_LOCKED(map);
1002 KASSERT(after_where->end <= entry->start,
1003 ("vm_map_entry_link: prev end %jx new start %jx overlap",
1004 (uintmax_t)after_where->end, (uintmax_t)entry->start));
1005 KASSERT(entry->end <= after_where->next->start,
1006 ("vm_map_entry_link: new end %jx next start %jx overlap",
1007 (uintmax_t)entry->end, (uintmax_t)after_where->next->start));
1010 entry->prev = after_where;
1011 entry->next = after_where->next;
1012 entry->next->prev = entry;
1013 after_where->next = entry;
1015 if (after_where != &map->header) {
1016 if (after_where != map->root)
1017 vm_map_entry_splay(after_where->start, map->root);
1018 entry->right = after_where->right;
1019 entry->left = after_where;
1020 after_where->right = NULL;
1021 after_where->adj_free = entry->start - after_where->end;
1022 vm_map_entry_set_max_free(after_where);
1024 entry->right = map->root;
1027 entry->adj_free = entry->next->start - entry->end;
1028 vm_map_entry_set_max_free(entry);
1033 vm_map_entry_unlink(vm_map_t map,
1034 vm_map_entry_t entry)
1036 vm_map_entry_t next, prev, root;
1038 VM_MAP_ASSERT_LOCKED(map);
1039 if (entry != map->root)
1040 vm_map_entry_splay(entry->start, map->root);
1041 if (entry->left == NULL)
1042 root = entry->right;
1044 root = vm_map_entry_splay(entry->start, entry->left);
1045 root->right = entry->right;
1046 root->adj_free = entry->next->start - root->end;
1047 vm_map_entry_set_max_free(root);
1056 CTR3(KTR_VM, "vm_map_entry_unlink: map %p, nentries %d, entry %p", map,
1057 map->nentries, entry);
1061 * vm_map_entry_resize_free:
1063 * Recompute the amount of free space following a vm_map_entry
1064 * and propagate that value up the tree. Call this function after
1065 * resizing a map entry in-place, that is, without a call to
1066 * vm_map_entry_link() or _unlink().
1068 * The map must be locked, and leaves it so.
1071 vm_map_entry_resize_free(vm_map_t map, vm_map_entry_t entry)
1075 * Using splay trees without parent pointers, propagating
1076 * max_free up the tree is done by moving the entry to the
1077 * root and making the change there.
1079 if (entry != map->root)
1080 map->root = vm_map_entry_splay(entry->start, map->root);
1082 entry->adj_free = entry->next->start - entry->end;
1083 vm_map_entry_set_max_free(entry);
1087 * vm_map_lookup_entry: [ internal use only ]
1089 * Finds the map entry containing (or
1090 * immediately preceding) the specified address
1091 * in the given map; the entry is returned
1092 * in the "entry" parameter. The boolean
1093 * result indicates whether the address is
1094 * actually contained in the map.
1097 vm_map_lookup_entry(
1099 vm_offset_t address,
1100 vm_map_entry_t *entry) /* OUT */
1106 * If the map is empty, then the map entry immediately preceding
1107 * "address" is the map's header.
1111 *entry = &map->header;
1112 else if (address >= cur->start && cur->end > address) {
1115 } else if ((locked = vm_map_locked(map)) ||
1116 sx_try_upgrade(&map->lock)) {
1118 * Splay requires a write lock on the map. However, it only
1119 * restructures the binary search tree; it does not otherwise
1120 * change the map. Thus, the map's timestamp need not change
1121 * on a temporary upgrade.
1123 map->root = cur = vm_map_entry_splay(address, cur);
1125 sx_downgrade(&map->lock);
1128 * If "address" is contained within a map entry, the new root
1129 * is that map entry. Otherwise, the new root is a map entry
1130 * immediately before or after "address".
1132 if (address >= cur->start) {
1134 if (cur->end > address)
1140 * Since the map is only locked for read access, perform a
1141 * standard binary search tree lookup for "address".
1144 if (address < cur->start) {
1145 if (cur->left == NULL) {
1150 } else if (cur->end > address) {
1154 if (cur->right == NULL) {
1167 * Inserts the given whole VM object into the target
1168 * map at the specified address range. The object's
1169 * size should match that of the address range.
1171 * Requires that the map be locked, and leaves it so.
1173 * If object is non-NULL, ref count must be bumped by caller
1174 * prior to making call to account for the new entry.
1177 vm_map_insert(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1178 vm_offset_t start, vm_offset_t end, vm_prot_t prot, vm_prot_t max, int cow)
1180 vm_map_entry_t new_entry, prev_entry, temp_entry;
1182 vm_eflags_t protoeflags;
1183 vm_inherit_t inheritance;
1185 VM_MAP_ASSERT_LOCKED(map);
1186 KASSERT(object != kernel_object ||
1187 (cow & MAP_COPY_ON_WRITE) == 0,
1188 ("vm_map_insert: kernel object and COW"));
1189 KASSERT(object == NULL || (cow & MAP_NOFAULT) == 0,
1190 ("vm_map_insert: paradoxical MAP_NOFAULT request"));
1191 KASSERT((prot & ~max) == 0,
1192 ("prot %#x is not subset of max_prot %#x", prot, max));
1195 * Check that the start and end points are not bogus.
1197 if (start < vm_map_min(map) || end > vm_map_max(map) ||
1199 return (KERN_INVALID_ADDRESS);
1202 * Find the entry prior to the proposed starting address; if it's part
1203 * of an existing entry, this range is bogus.
1205 if (vm_map_lookup_entry(map, start, &temp_entry))
1206 return (KERN_NO_SPACE);
1208 prev_entry = temp_entry;
1211 * Assert that the next entry doesn't overlap the end point.
1213 if (prev_entry->next->start < end)
1214 return (KERN_NO_SPACE);
1216 if ((cow & MAP_CREATE_GUARD) != 0 && (object != NULL ||
1217 max != VM_PROT_NONE))
1218 return (KERN_INVALID_ARGUMENT);
1221 if (cow & MAP_COPY_ON_WRITE)
1222 protoeflags |= MAP_ENTRY_COW | MAP_ENTRY_NEEDS_COPY;
1223 if (cow & MAP_NOFAULT)
1224 protoeflags |= MAP_ENTRY_NOFAULT;
1225 if (cow & MAP_DISABLE_SYNCER)
1226 protoeflags |= MAP_ENTRY_NOSYNC;
1227 if (cow & MAP_DISABLE_COREDUMP)
1228 protoeflags |= MAP_ENTRY_NOCOREDUMP;
1229 if (cow & MAP_STACK_GROWS_DOWN)
1230 protoeflags |= MAP_ENTRY_GROWS_DOWN;
1231 if (cow & MAP_STACK_GROWS_UP)
1232 protoeflags |= MAP_ENTRY_GROWS_UP;
1233 if (cow & MAP_VN_WRITECOUNT)
1234 protoeflags |= MAP_ENTRY_VN_WRITECNT;
1235 if ((cow & MAP_CREATE_GUARD) != 0)
1236 protoeflags |= MAP_ENTRY_GUARD;
1237 if ((cow & MAP_CREATE_STACK_GAP_DN) != 0)
1238 protoeflags |= MAP_ENTRY_STACK_GAP_DN;
1239 if ((cow & MAP_CREATE_STACK_GAP_UP) != 0)
1240 protoeflags |= MAP_ENTRY_STACK_GAP_UP;
1241 if (cow & MAP_INHERIT_SHARE)
1242 inheritance = VM_INHERIT_SHARE;
1244 inheritance = VM_INHERIT_DEFAULT;
1247 if ((cow & (MAP_ACC_NO_CHARGE | MAP_NOFAULT | MAP_CREATE_GUARD)) != 0)
1249 if ((cow & MAP_ACC_CHARGED) || ((prot & VM_PROT_WRITE) &&
1250 ((protoeflags & MAP_ENTRY_NEEDS_COPY) || object == NULL))) {
1251 if (!(cow & MAP_ACC_CHARGED) && !swap_reserve(end - start))
1252 return (KERN_RESOURCE_SHORTAGE);
1253 KASSERT(object == NULL ||
1254 (protoeflags & MAP_ENTRY_NEEDS_COPY) != 0 ||
1255 object->cred == NULL,
1256 ("overcommit: vm_map_insert o %p", object));
1257 cred = curthread->td_ucred;
1261 /* Expand the kernel pmap, if necessary. */
1262 if (map == kernel_map && end > kernel_vm_end)
1263 pmap_growkernel(end);
1264 if (object != NULL) {
1266 * OBJ_ONEMAPPING must be cleared unless this mapping
1267 * is trivially proven to be the only mapping for any
1268 * of the object's pages. (Object granularity
1269 * reference counting is insufficient to recognize
1270 * aliases with precision.)
1272 VM_OBJECT_WLOCK(object);
1273 if (object->ref_count > 1 || object->shadow_count != 0)
1274 vm_object_clear_flag(object, OBJ_ONEMAPPING);
1275 VM_OBJECT_WUNLOCK(object);
1276 } else if ((prev_entry->eflags & ~MAP_ENTRY_USER_WIRED) ==
1278 (cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 &&
1279 prev_entry->end == start && (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 prev_entry->wired_count == 0) {
1296 KASSERT((prev_entry->eflags & MAP_ENTRY_USER_WIRED) ==
1297 0, ("prev_entry %p has incoherent wiring",
1299 if ((prev_entry->eflags & MAP_ENTRY_GUARD) == 0)
1300 map->size += end - prev_entry->end;
1301 prev_entry->end = end;
1302 vm_map_entry_resize_free(map, prev_entry);
1303 vm_map_simplify_entry(map, prev_entry);
1304 return (KERN_SUCCESS);
1308 * If we can extend the object but cannot extend the
1309 * map entry, we have to create a new map entry. We
1310 * must bump the ref count on the extended object to
1311 * account for it. object may be NULL.
1313 object = prev_entry->object.vm_object;
1314 offset = prev_entry->offset +
1315 (prev_entry->end - prev_entry->start);
1316 vm_object_reference(object);
1317 if (cred != NULL && object != NULL && object->cred != NULL &&
1318 !(prev_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
1319 /* Object already accounts for this uid. */
1327 * Create a new entry
1329 new_entry = vm_map_entry_create(map);
1330 new_entry->start = start;
1331 new_entry->end = end;
1332 new_entry->cred = NULL;
1334 new_entry->eflags = protoeflags;
1335 new_entry->object.vm_object = object;
1336 new_entry->offset = offset;
1338 new_entry->inheritance = inheritance;
1339 new_entry->protection = prot;
1340 new_entry->max_protection = max;
1341 new_entry->wired_count = 0;
1342 new_entry->wiring_thread = NULL;
1343 new_entry->read_ahead = VM_FAULT_READ_AHEAD_INIT;
1344 new_entry->next_read = start;
1346 KASSERT(cred == NULL || !ENTRY_CHARGED(new_entry),
1347 ("overcommit: vm_map_insert leaks vm_map %p", new_entry));
1348 new_entry->cred = cred;
1351 * Insert the new entry into the list
1353 vm_map_entry_link(map, prev_entry, new_entry);
1354 if ((new_entry->eflags & MAP_ENTRY_GUARD) == 0)
1355 map->size += new_entry->end - new_entry->start;
1358 * Try to coalesce the new entry with both the previous and next
1359 * entries in the list. Previously, we only attempted to coalesce
1360 * with the previous entry when object is NULL. Here, we handle the
1361 * other cases, which are less common.
1363 vm_map_simplify_entry(map, new_entry);
1365 if ((cow & (MAP_PREFAULT | MAP_PREFAULT_PARTIAL)) != 0) {
1366 vm_map_pmap_enter(map, start, prot, object, OFF_TO_IDX(offset),
1367 end - start, cow & MAP_PREFAULT_PARTIAL);
1370 return (KERN_SUCCESS);
1376 * Find the first fit (lowest VM address) for "length" free bytes
1377 * beginning at address >= start in the given map.
1379 * In a vm_map_entry, "adj_free" is the amount of free space
1380 * adjacent (higher address) to this entry, and "max_free" is the
1381 * maximum amount of contiguous free space in its subtree. This
1382 * allows finding a free region in one path down the tree, so
1383 * O(log n) amortized with splay trees.
1385 * The map must be locked, and leaves it so.
1387 * Returns: 0 on success, and starting address in *addr,
1388 * 1 if insufficient space.
1391 vm_map_findspace(vm_map_t map, vm_offset_t start, vm_size_t length,
1392 vm_offset_t *addr) /* OUT */
1394 vm_map_entry_t entry;
1398 * Request must fit within min/max VM address and must avoid
1401 start = MAX(start, vm_map_min(map));
1402 if (start + length > vm_map_max(map) || start + length < start)
1405 /* Empty tree means wide open address space. */
1406 if (map->root == NULL) {
1412 * After splay, if start comes before root node, then there
1413 * must be a gap from start to the root.
1415 map->root = vm_map_entry_splay(start, map->root);
1416 if (start + length <= map->root->start) {
1422 * Root is the last node that might begin its gap before
1423 * start, and this is the last comparison where address
1424 * wrap might be a problem.
1426 st = (start > map->root->end) ? start : map->root->end;
1427 if (length <= map->root->end + map->root->adj_free - st) {
1432 /* With max_free, can immediately tell if no solution. */
1433 entry = map->root->right;
1434 if (entry == NULL || length > entry->max_free)
1438 * Search the right subtree in the order: left subtree, root,
1439 * right subtree (first fit). The previous splay implies that
1440 * all regions in the right subtree have addresses > start.
1442 while (entry != NULL) {
1443 if (entry->left != NULL && entry->left->max_free >= length)
1444 entry = entry->left;
1445 else if (entry->adj_free >= length) {
1449 entry = entry->right;
1452 /* Can't get here, so panic if we do. */
1453 panic("vm_map_findspace: max_free corrupt");
1457 vm_map_fixed(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1458 vm_offset_t start, vm_size_t length, vm_prot_t prot,
1459 vm_prot_t max, int cow)
1464 end = start + length;
1465 KASSERT((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 ||
1467 ("vm_map_fixed: non-NULL backing object for stack"));
1469 VM_MAP_RANGE_CHECK(map, start, end);
1470 if ((cow & MAP_CHECK_EXCL) == 0)
1471 vm_map_delete(map, start, end);
1472 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) != 0) {
1473 result = vm_map_stack_locked(map, start, length, sgrowsiz,
1476 result = vm_map_insert(map, object, offset, start, end,
1484 * Searches for the specified amount of free space in the given map with the
1485 * specified alignment. Performs an address-ordered, first-fit search from
1486 * the given address "*addr", with an optional upper bound "max_addr". If the
1487 * parameter "alignment" is zero, then the alignment is computed from the
1488 * given (object, offset) pair so as to enable the greatest possible use of
1489 * superpage mappings. Returns KERN_SUCCESS and the address of the free space
1490 * in "*addr" if successful. Otherwise, returns KERN_NO_SPACE.
1492 * The map must be locked. Initially, there must be at least "length" bytes
1493 * of free space at the given address.
1496 vm_map_alignspace(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1497 vm_offset_t *addr, vm_size_t length, vm_offset_t max_addr,
1498 vm_offset_t alignment)
1500 vm_offset_t aligned_addr, free_addr;
1502 VM_MAP_ASSERT_LOCKED(map);
1504 KASSERT(!vm_map_findspace(map, free_addr, length, addr) &&
1505 free_addr == *addr, ("caller provided insufficient free space"));
1508 * At the start of every iteration, the free space at address
1509 * "*addr" is at least "length" bytes.
1512 pmap_align_superpage(object, offset, addr, length);
1513 else if ((*addr & (alignment - 1)) != 0) {
1514 *addr &= ~(alignment - 1);
1517 aligned_addr = *addr;
1518 if (aligned_addr == free_addr) {
1520 * Alignment did not change "*addr", so "*addr" must
1521 * still provide sufficient free space.
1523 return (KERN_SUCCESS);
1527 * Test for address wrap on "*addr". A wrapped "*addr" could
1528 * be a valid address, in which case vm_map_findspace() cannot
1529 * be relied upon to fail.
1531 if (aligned_addr < free_addr ||
1532 vm_map_findspace(map, aligned_addr, length, addr) ||
1533 (max_addr != 0 && *addr + length > max_addr))
1534 return (KERN_NO_SPACE);
1536 if (free_addr == aligned_addr) {
1538 * If a successful call to vm_map_findspace() did not
1539 * change "*addr", then "*addr" must still be aligned
1540 * and provide sufficient free space.
1542 return (KERN_SUCCESS);
1548 * vm_map_find finds an unallocated region in the target address
1549 * map with the given length. The search is defined to be
1550 * first-fit from the specified address; the region found is
1551 * returned in the same parameter.
1553 * If object is non-NULL, ref count must be bumped by caller
1554 * prior to making call to account for the new entry.
1557 vm_map_find(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1558 vm_offset_t *addr, /* IN/OUT */
1559 vm_size_t length, vm_offset_t max_addr, int find_space,
1560 vm_prot_t prot, vm_prot_t max, int cow)
1562 vm_offset_t alignment, min_addr;
1565 KASSERT((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 ||
1567 ("vm_map_find: non-NULL backing object for stack"));
1568 if (find_space == VMFS_OPTIMAL_SPACE && (object == NULL ||
1569 (object->flags & OBJ_COLORED) == 0))
1570 find_space = VMFS_ANY_SPACE;
1571 if (find_space >> 8 != 0) {
1572 KASSERT((find_space & 0xff) == 0, ("bad VMFS flags"));
1573 alignment = (vm_offset_t)1 << (find_space >> 8);
1577 if (find_space != VMFS_NO_SPACE) {
1578 KASSERT(find_space == VMFS_ANY_SPACE ||
1579 find_space == VMFS_OPTIMAL_SPACE ||
1580 find_space == VMFS_SUPER_SPACE ||
1581 alignment != 0, ("unexpected VMFS flag"));
1584 if (vm_map_findspace(map, min_addr, length, addr) ||
1585 (max_addr != 0 && *addr + length > max_addr)) {
1589 if (find_space != VMFS_ANY_SPACE &&
1590 (rv = vm_map_alignspace(map, object, offset, addr, length,
1591 max_addr, alignment)) != KERN_SUCCESS) {
1592 if (find_space == VMFS_OPTIMAL_SPACE) {
1593 find_space = VMFS_ANY_SPACE;
1599 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) != 0) {
1600 rv = vm_map_stack_locked(map, *addr, length, sgrowsiz, prot,
1603 rv = vm_map_insert(map, object, offset, *addr, *addr + length,
1612 * vm_map_find_min() is a variant of vm_map_find() that takes an
1613 * additional parameter (min_addr) and treats the given address
1614 * (*addr) differently. Specifically, it treats *addr as a hint
1615 * and not as the minimum address where the mapping is created.
1617 * This function works in two phases. First, it tries to
1618 * allocate above the hint. If that fails and the hint is
1619 * greater than min_addr, it performs a second pass, replacing
1620 * the hint with min_addr as the minimum address for the
1624 vm_map_find_min(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1625 vm_offset_t *addr, vm_size_t length, vm_offset_t min_addr,
1626 vm_offset_t max_addr, int find_space, vm_prot_t prot, vm_prot_t max,
1634 rv = vm_map_find(map, object, offset, addr, length, max_addr,
1635 find_space, prot, max, cow);
1636 if (rv == KERN_SUCCESS || min_addr >= hint)
1638 *addr = hint = min_addr;
1643 * A map entry with any of the following flags set must not be merged with
1646 #define MAP_ENTRY_NOMERGE_MASK (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP | \
1647 MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_IS_SUB_MAP)
1650 vm_map_mergeable_neighbors(vm_map_entry_t prev, vm_map_entry_t entry)
1653 KASSERT((prev->eflags & MAP_ENTRY_NOMERGE_MASK) == 0 ||
1654 (entry->eflags & MAP_ENTRY_NOMERGE_MASK) == 0,
1655 ("vm_map_mergeable_neighbors: neither %p nor %p are mergeable",
1657 return (prev->end == entry->start &&
1658 prev->object.vm_object == entry->object.vm_object &&
1659 (prev->object.vm_object == NULL ||
1660 prev->offset + (prev->end - prev->start) == entry->offset) &&
1661 prev->eflags == entry->eflags &&
1662 prev->protection == entry->protection &&
1663 prev->max_protection == entry->max_protection &&
1664 prev->inheritance == entry->inheritance &&
1665 prev->wired_count == entry->wired_count &&
1666 prev->cred == entry->cred);
1670 vm_map_merged_neighbor_dispose(vm_map_t map, vm_map_entry_t entry)
1674 * If the backing object is a vnode object, vm_object_deallocate()
1675 * calls vrele(). However, vrele() does not lock the vnode because
1676 * the vnode has additional references. Thus, the map lock can be
1677 * kept without causing a lock-order reversal with the vnode lock.
1679 * Since we count the number of virtual page mappings in
1680 * object->un_pager.vnp.writemappings, the writemappings value
1681 * should not be adjusted when the entry is disposed of.
1683 if (entry->object.vm_object != NULL)
1684 vm_object_deallocate(entry->object.vm_object);
1685 if (entry->cred != NULL)
1686 crfree(entry->cred);
1687 vm_map_entry_dispose(map, entry);
1691 * vm_map_simplify_entry:
1693 * Simplify the given map entry by merging with either neighbor. This
1694 * routine also has the ability to merge with both neighbors.
1696 * The map must be locked.
1698 * This routine guarantees that the passed entry remains valid (though
1699 * possibly extended). When merging, this routine may delete one or
1703 vm_map_simplify_entry(vm_map_t map, vm_map_entry_t entry)
1705 vm_map_entry_t next, prev;
1707 if ((entry->eflags & MAP_ENTRY_NOMERGE_MASK) != 0)
1710 if (vm_map_mergeable_neighbors(prev, entry)) {
1711 vm_map_entry_unlink(map, prev);
1712 entry->start = prev->start;
1713 entry->offset = prev->offset;
1714 if (entry->prev != &map->header)
1715 vm_map_entry_resize_free(map, entry->prev);
1716 vm_map_merged_neighbor_dispose(map, prev);
1719 if (vm_map_mergeable_neighbors(entry, next)) {
1720 vm_map_entry_unlink(map, next);
1721 entry->end = next->end;
1722 vm_map_entry_resize_free(map, entry);
1723 vm_map_merged_neighbor_dispose(map, next);
1728 * vm_map_clip_start: [ internal use only ]
1730 * Asserts that the given entry begins at or after
1731 * the specified address; if necessary,
1732 * it splits the entry into two.
1734 #define vm_map_clip_start(map, entry, startaddr) \
1736 if (startaddr > entry->start) \
1737 _vm_map_clip_start(map, entry, startaddr); \
1741 * This routine is called only when it is known that
1742 * the entry must be split.
1745 _vm_map_clip_start(vm_map_t map, vm_map_entry_t entry, vm_offset_t start)
1747 vm_map_entry_t new_entry;
1749 VM_MAP_ASSERT_LOCKED(map);
1750 KASSERT(entry->end > start && entry->start < start,
1751 ("_vm_map_clip_start: invalid clip of entry %p", entry));
1754 * Split off the front portion -- note that we must insert the new
1755 * entry BEFORE this one, so that this entry has the specified
1758 vm_map_simplify_entry(map, entry);
1761 * If there is no object backing this entry, we might as well create
1762 * one now. If we defer it, an object can get created after the map
1763 * is clipped, and individual objects will be created for the split-up
1764 * map. This is a bit of a hack, but is also about the best place to
1765 * put this improvement.
1767 if (entry->object.vm_object == NULL && !map->system_map &&
1768 (entry->eflags & MAP_ENTRY_GUARD) == 0) {
1770 object = vm_object_allocate(OBJT_DEFAULT,
1771 atop(entry->end - entry->start));
1772 entry->object.vm_object = object;
1774 if (entry->cred != NULL) {
1775 object->cred = entry->cred;
1776 object->charge = entry->end - entry->start;
1779 } else if (entry->object.vm_object != NULL &&
1780 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) &&
1781 entry->cred != NULL) {
1782 VM_OBJECT_WLOCK(entry->object.vm_object);
1783 KASSERT(entry->object.vm_object->cred == NULL,
1784 ("OVERCOMMIT: vm_entry_clip_start: both cred e %p", entry));
1785 entry->object.vm_object->cred = entry->cred;
1786 entry->object.vm_object->charge = entry->end - entry->start;
1787 VM_OBJECT_WUNLOCK(entry->object.vm_object);
1791 new_entry = vm_map_entry_create(map);
1792 *new_entry = *entry;
1794 new_entry->end = start;
1795 entry->offset += (start - entry->start);
1796 entry->start = start;
1797 if (new_entry->cred != NULL)
1798 crhold(entry->cred);
1800 vm_map_entry_link(map, entry->prev, new_entry);
1802 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1803 vm_object_reference(new_entry->object.vm_object);
1805 * The object->un_pager.vnp.writemappings for the
1806 * object of MAP_ENTRY_VN_WRITECNT type entry shall be
1807 * kept as is here. The virtual pages are
1808 * re-distributed among the clipped entries, so the sum is
1815 * vm_map_clip_end: [ internal use only ]
1817 * Asserts that the given entry ends at or before
1818 * the specified address; if necessary,
1819 * it splits the entry into two.
1821 #define vm_map_clip_end(map, entry, endaddr) \
1823 if ((endaddr) < (entry->end)) \
1824 _vm_map_clip_end((map), (entry), (endaddr)); \
1828 * This routine is called only when it is known that
1829 * the entry must be split.
1832 _vm_map_clip_end(vm_map_t map, vm_map_entry_t entry, vm_offset_t end)
1834 vm_map_entry_t new_entry;
1836 VM_MAP_ASSERT_LOCKED(map);
1837 KASSERT(entry->start < end && entry->end > end,
1838 ("_vm_map_clip_end: invalid clip of entry %p", entry));
1841 * If there is no object backing this entry, we might as well create
1842 * one now. If we defer it, an object can get created after the map
1843 * is clipped, and individual objects will be created for the split-up
1844 * map. This is a bit of a hack, but is also about the best place to
1845 * put this improvement.
1847 if (entry->object.vm_object == NULL && !map->system_map &&
1848 (entry->eflags & MAP_ENTRY_GUARD) == 0) {
1850 object = vm_object_allocate(OBJT_DEFAULT,
1851 atop(entry->end - entry->start));
1852 entry->object.vm_object = object;
1854 if (entry->cred != NULL) {
1855 object->cred = entry->cred;
1856 object->charge = entry->end - entry->start;
1859 } else if (entry->object.vm_object != NULL &&
1860 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) &&
1861 entry->cred != NULL) {
1862 VM_OBJECT_WLOCK(entry->object.vm_object);
1863 KASSERT(entry->object.vm_object->cred == NULL,
1864 ("OVERCOMMIT: vm_entry_clip_end: both cred e %p", entry));
1865 entry->object.vm_object->cred = entry->cred;
1866 entry->object.vm_object->charge = entry->end - entry->start;
1867 VM_OBJECT_WUNLOCK(entry->object.vm_object);
1872 * Create a new entry and insert it AFTER the specified entry
1874 new_entry = vm_map_entry_create(map);
1875 *new_entry = *entry;
1877 new_entry->start = entry->end = end;
1878 new_entry->offset += (end - entry->start);
1879 if (new_entry->cred != NULL)
1880 crhold(entry->cred);
1882 vm_map_entry_link(map, entry, new_entry);
1884 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1885 vm_object_reference(new_entry->object.vm_object);
1890 * vm_map_submap: [ kernel use only ]
1892 * Mark the given range as handled by a subordinate map.
1894 * This range must have been created with vm_map_find,
1895 * and no other operations may have been performed on this
1896 * range prior to calling vm_map_submap.
1898 * Only a limited number of operations can be performed
1899 * within this rage after calling vm_map_submap:
1901 * [Don't try vm_map_copy!]
1903 * To remove a submapping, one must first remove the
1904 * range from the superior map, and then destroy the
1905 * submap (if desired). [Better yet, don't try it.]
1914 vm_map_entry_t entry;
1915 int result = KERN_INVALID_ARGUMENT;
1919 VM_MAP_RANGE_CHECK(map, start, end);
1921 if (vm_map_lookup_entry(map, start, &entry)) {
1922 vm_map_clip_start(map, entry, start);
1924 entry = entry->next;
1926 vm_map_clip_end(map, entry, end);
1928 if ((entry->start == start) && (entry->end == end) &&
1929 ((entry->eflags & MAP_ENTRY_COW) == 0) &&
1930 (entry->object.vm_object == NULL)) {
1931 entry->object.sub_map = submap;
1932 entry->eflags |= MAP_ENTRY_IS_SUB_MAP;
1933 result = KERN_SUCCESS;
1941 * The maximum number of pages to map if MAP_PREFAULT_PARTIAL is specified
1943 #define MAX_INIT_PT 96
1946 * vm_map_pmap_enter:
1948 * Preload the specified map's pmap with mappings to the specified
1949 * object's memory-resident pages. No further physical pages are
1950 * allocated, and no further virtual pages are retrieved from secondary
1951 * storage. If the specified flags include MAP_PREFAULT_PARTIAL, then a
1952 * limited number of page mappings are created at the low-end of the
1953 * specified address range. (For this purpose, a superpage mapping
1954 * counts as one page mapping.) Otherwise, all resident pages within
1955 * the specified address range are mapped.
1958 vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot,
1959 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags)
1962 vm_page_t p, p_start;
1963 vm_pindex_t mask, psize, threshold, tmpidx;
1965 if ((prot & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0 || object == NULL)
1967 VM_OBJECT_RLOCK(object);
1968 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) {
1969 VM_OBJECT_RUNLOCK(object);
1970 VM_OBJECT_WLOCK(object);
1971 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) {
1972 pmap_object_init_pt(map->pmap, addr, object, pindex,
1974 VM_OBJECT_WUNLOCK(object);
1977 VM_OBJECT_LOCK_DOWNGRADE(object);
1981 if (psize + pindex > object->size) {
1982 if (object->size < pindex) {
1983 VM_OBJECT_RUNLOCK(object);
1986 psize = object->size - pindex;
1991 threshold = MAX_INIT_PT;
1993 p = vm_page_find_least(object, pindex);
1995 * Assert: the variable p is either (1) the page with the
1996 * least pindex greater than or equal to the parameter pindex
2000 p != NULL && (tmpidx = p->pindex - pindex) < psize;
2001 p = TAILQ_NEXT(p, listq)) {
2003 * don't allow an madvise to blow away our really
2004 * free pages allocating pv entries.
2006 if (((flags & MAP_PREFAULT_MADVISE) != 0 &&
2007 vm_page_count_severe()) ||
2008 ((flags & MAP_PREFAULT_PARTIAL) != 0 &&
2009 tmpidx >= threshold)) {
2013 if (p->valid == VM_PAGE_BITS_ALL) {
2014 if (p_start == NULL) {
2015 start = addr + ptoa(tmpidx);
2018 /* Jump ahead if a superpage mapping is possible. */
2019 if (p->psind > 0 && ((addr + ptoa(tmpidx)) &
2020 (pagesizes[p->psind] - 1)) == 0) {
2021 mask = atop(pagesizes[p->psind]) - 1;
2022 if (tmpidx + mask < psize &&
2023 vm_page_ps_test(p, PS_ALL_VALID, NULL)) {
2028 } else if (p_start != NULL) {
2029 pmap_enter_object(map->pmap, start, addr +
2030 ptoa(tmpidx), p_start, prot);
2034 if (p_start != NULL)
2035 pmap_enter_object(map->pmap, start, addr + ptoa(psize),
2037 VM_OBJECT_RUNLOCK(object);
2043 * Sets the protection of the specified address
2044 * region in the target map. If "set_max" is
2045 * specified, the maximum protection is to be set;
2046 * otherwise, only the current protection is affected.
2049 vm_map_protect(vm_map_t map, vm_offset_t start, vm_offset_t end,
2050 vm_prot_t new_prot, boolean_t set_max)
2052 vm_map_entry_t current, entry;
2058 return (KERN_SUCCESS);
2063 * Ensure that we are not concurrently wiring pages. vm_map_wire() may
2064 * need to fault pages into the map and will drop the map lock while
2065 * doing so, and the VM object may end up in an inconsistent state if we
2066 * update the protection on the map entry in between faults.
2068 vm_map_wait_busy(map);
2070 VM_MAP_RANGE_CHECK(map, start, end);
2072 if (vm_map_lookup_entry(map, start, &entry)) {
2073 vm_map_clip_start(map, entry, start);
2075 entry = entry->next;
2079 * Make a first pass to check for protection violations.
2081 for (current = entry; current->start < end; current = current->next) {
2082 if ((current->eflags & MAP_ENTRY_GUARD) != 0)
2084 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
2086 return (KERN_INVALID_ARGUMENT);
2088 if ((new_prot & current->max_protection) != new_prot) {
2090 return (KERN_PROTECTION_FAILURE);
2095 * Do an accounting pass for private read-only mappings that
2096 * now will do cow due to allowed write (e.g. debugger sets
2097 * breakpoint on text segment)
2099 for (current = entry; current->start < end; current = current->next) {
2101 vm_map_clip_end(map, current, end);
2104 ((new_prot & ~(current->protection)) & VM_PROT_WRITE) == 0 ||
2105 ENTRY_CHARGED(current) ||
2106 (current->eflags & MAP_ENTRY_GUARD) != 0) {
2110 cred = curthread->td_ucred;
2111 obj = current->object.vm_object;
2113 if (obj == NULL || (current->eflags & MAP_ENTRY_NEEDS_COPY)) {
2114 if (!swap_reserve(current->end - current->start)) {
2116 return (KERN_RESOURCE_SHORTAGE);
2119 current->cred = cred;
2123 VM_OBJECT_WLOCK(obj);
2124 if (obj->type != OBJT_DEFAULT && obj->type != OBJT_SWAP) {
2125 VM_OBJECT_WUNLOCK(obj);
2130 * Charge for the whole object allocation now, since
2131 * we cannot distinguish between non-charged and
2132 * charged clipped mapping of the same object later.
2134 KASSERT(obj->charge == 0,
2135 ("vm_map_protect: object %p overcharged (entry %p)",
2137 if (!swap_reserve(ptoa(obj->size))) {
2138 VM_OBJECT_WUNLOCK(obj);
2140 return (KERN_RESOURCE_SHORTAGE);
2145 obj->charge = ptoa(obj->size);
2146 VM_OBJECT_WUNLOCK(obj);
2150 * Go back and fix up protections. [Note that clipping is not
2151 * necessary the second time.]
2153 for (current = entry; current->start < end; current = current->next) {
2154 if ((current->eflags & MAP_ENTRY_GUARD) != 0)
2157 old_prot = current->protection;
2160 current->protection =
2161 (current->max_protection = new_prot) &
2164 current->protection = new_prot;
2167 * For user wired map entries, the normal lazy evaluation of
2168 * write access upgrades through soft page faults is
2169 * undesirable. Instead, immediately copy any pages that are
2170 * copy-on-write and enable write access in the physical map.
2172 if ((current->eflags & MAP_ENTRY_USER_WIRED) != 0 &&
2173 (current->protection & VM_PROT_WRITE) != 0 &&
2174 (old_prot & VM_PROT_WRITE) == 0)
2175 vm_fault_copy_entry(map, map, current, current, NULL);
2178 * When restricting access, update the physical map. Worry
2179 * about copy-on-write here.
2181 if ((old_prot & ~current->protection) != 0) {
2182 #define MASK(entry) (((entry)->eflags & MAP_ENTRY_COW) ? ~VM_PROT_WRITE : \
2184 pmap_protect(map->pmap, current->start,
2186 current->protection & MASK(current));
2189 vm_map_simplify_entry(map, current);
2192 return (KERN_SUCCESS);
2198 * This routine traverses a processes map handling the madvise
2199 * system call. Advisories are classified as either those effecting
2200 * the vm_map_entry structure, or those effecting the underlying
2210 vm_map_entry_t current, entry;
2214 * Some madvise calls directly modify the vm_map_entry, in which case
2215 * we need to use an exclusive lock on the map and we need to perform
2216 * various clipping operations. Otherwise we only need a read-lock
2221 case MADV_SEQUENTIAL:
2238 vm_map_lock_read(map);
2245 * Locate starting entry and clip if necessary.
2247 VM_MAP_RANGE_CHECK(map, start, end);
2249 if (vm_map_lookup_entry(map, start, &entry)) {
2251 vm_map_clip_start(map, entry, start);
2253 entry = entry->next;
2258 * madvise behaviors that are implemented in the vm_map_entry.
2260 * We clip the vm_map_entry so that behavioral changes are
2261 * limited to the specified address range.
2263 for (current = entry; current->start < end;
2264 current = current->next) {
2265 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
2268 vm_map_clip_end(map, current, end);
2272 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_NORMAL);
2274 case MADV_SEQUENTIAL:
2275 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_SEQUENTIAL);
2278 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_RANDOM);
2281 current->eflags |= MAP_ENTRY_NOSYNC;
2284 current->eflags &= ~MAP_ENTRY_NOSYNC;
2287 current->eflags |= MAP_ENTRY_NOCOREDUMP;
2290 current->eflags &= ~MAP_ENTRY_NOCOREDUMP;
2295 vm_map_simplify_entry(map, current);
2299 vm_pindex_t pstart, pend;
2302 * madvise behaviors that are implemented in the underlying
2305 * Since we don't clip the vm_map_entry, we have to clip
2306 * the vm_object pindex and count.
2308 for (current = entry; current->start < end;
2309 current = current->next) {
2310 vm_offset_t useEnd, useStart;
2312 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
2315 pstart = OFF_TO_IDX(current->offset);
2316 pend = pstart + atop(current->end - current->start);
2317 useStart = current->start;
2318 useEnd = current->end;
2320 if (current->start < start) {
2321 pstart += atop(start - current->start);
2324 if (current->end > end) {
2325 pend -= atop(current->end - end);
2333 * Perform the pmap_advise() before clearing
2334 * PGA_REFERENCED in vm_page_advise(). Otherwise, a
2335 * concurrent pmap operation, such as pmap_remove(),
2336 * could clear a reference in the pmap and set
2337 * PGA_REFERENCED on the page before the pmap_advise()
2338 * had completed. Consequently, the page would appear
2339 * referenced based upon an old reference that
2340 * occurred before this pmap_advise() ran.
2342 if (behav == MADV_DONTNEED || behav == MADV_FREE)
2343 pmap_advise(map->pmap, useStart, useEnd,
2346 vm_object_madvise(current->object.vm_object, pstart,
2350 * Pre-populate paging structures in the
2351 * WILLNEED case. For wired entries, the
2352 * paging structures are already populated.
2354 if (behav == MADV_WILLNEED &&
2355 current->wired_count == 0) {
2356 vm_map_pmap_enter(map,
2358 current->protection,
2359 current->object.vm_object,
2361 ptoa(pend - pstart),
2362 MAP_PREFAULT_MADVISE
2366 vm_map_unlock_read(map);
2375 * Sets the inheritance of the specified address
2376 * range in the target map. Inheritance
2377 * affects how the map will be shared with
2378 * child maps at the time of vmspace_fork.
2381 vm_map_inherit(vm_map_t map, vm_offset_t start, vm_offset_t end,
2382 vm_inherit_t new_inheritance)
2384 vm_map_entry_t entry;
2385 vm_map_entry_t temp_entry;
2387 switch (new_inheritance) {
2388 case VM_INHERIT_NONE:
2389 case VM_INHERIT_COPY:
2390 case VM_INHERIT_SHARE:
2391 case VM_INHERIT_ZERO:
2394 return (KERN_INVALID_ARGUMENT);
2397 return (KERN_SUCCESS);
2399 VM_MAP_RANGE_CHECK(map, start, end);
2400 if (vm_map_lookup_entry(map, start, &temp_entry)) {
2402 vm_map_clip_start(map, entry, start);
2404 entry = temp_entry->next;
2405 while (entry->start < end) {
2406 vm_map_clip_end(map, entry, end);
2407 if ((entry->eflags & MAP_ENTRY_GUARD) == 0 ||
2408 new_inheritance != VM_INHERIT_ZERO)
2409 entry->inheritance = new_inheritance;
2410 vm_map_simplify_entry(map, entry);
2411 entry = entry->next;
2414 return (KERN_SUCCESS);
2420 * Implements both kernel and user unwiring.
2423 vm_map_unwire(vm_map_t map, vm_offset_t start, vm_offset_t end,
2426 vm_map_entry_t entry, first_entry, tmp_entry;
2427 vm_offset_t saved_start;
2428 unsigned int last_timestamp;
2430 boolean_t need_wakeup, result, user_unwire;
2433 return (KERN_SUCCESS);
2434 user_unwire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE;
2436 VM_MAP_RANGE_CHECK(map, start, end);
2437 if (!vm_map_lookup_entry(map, start, &first_entry)) {
2438 if (flags & VM_MAP_WIRE_HOLESOK)
2439 first_entry = first_entry->next;
2442 return (KERN_INVALID_ADDRESS);
2445 last_timestamp = map->timestamp;
2446 entry = first_entry;
2447 while (entry->start < end) {
2448 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
2450 * We have not yet clipped the entry.
2452 saved_start = (start >= entry->start) ? start :
2454 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2455 if (vm_map_unlock_and_wait(map, 0)) {
2457 * Allow interruption of user unwiring?
2461 if (last_timestamp+1 != map->timestamp) {
2463 * Look again for the entry because the map was
2464 * modified while it was unlocked.
2465 * Specifically, the entry may have been
2466 * clipped, merged, or deleted.
2468 if (!vm_map_lookup_entry(map, saved_start,
2470 if (flags & VM_MAP_WIRE_HOLESOK)
2471 tmp_entry = tmp_entry->next;
2473 if (saved_start == start) {
2475 * First_entry has been deleted.
2478 return (KERN_INVALID_ADDRESS);
2481 rv = KERN_INVALID_ADDRESS;
2485 if (entry == first_entry)
2486 first_entry = tmp_entry;
2491 last_timestamp = map->timestamp;
2494 vm_map_clip_start(map, entry, start);
2495 vm_map_clip_end(map, entry, end);
2497 * Mark the entry in case the map lock is released. (See
2500 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 &&
2501 entry->wiring_thread == NULL,
2502 ("owned map entry %p", entry));
2503 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
2504 entry->wiring_thread = curthread;
2506 * Check the map for holes in the specified region.
2507 * If VM_MAP_WIRE_HOLESOK was specified, skip this check.
2509 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) &&
2510 (entry->end < end && entry->next->start > entry->end)) {
2512 rv = KERN_INVALID_ADDRESS;
2516 * If system unwiring, require that the entry is system wired.
2519 vm_map_entry_system_wired_count(entry) == 0) {
2521 rv = KERN_INVALID_ARGUMENT;
2524 entry = entry->next;
2528 need_wakeup = FALSE;
2529 if (first_entry == NULL) {
2530 result = vm_map_lookup_entry(map, start, &first_entry);
2531 if (!result && (flags & VM_MAP_WIRE_HOLESOK))
2532 first_entry = first_entry->next;
2534 KASSERT(result, ("vm_map_unwire: lookup failed"));
2536 for (entry = first_entry; entry->start < end; entry = entry->next) {
2538 * If VM_MAP_WIRE_HOLESOK was specified, an empty
2539 * space in the unwired region could have been mapped
2540 * while the map lock was dropped for draining
2541 * MAP_ENTRY_IN_TRANSITION. Moreover, another thread
2542 * could be simultaneously wiring this new mapping
2543 * entry. Detect these cases and skip any entries
2544 * marked as in transition by us.
2546 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 ||
2547 entry->wiring_thread != curthread) {
2548 KASSERT((flags & VM_MAP_WIRE_HOLESOK) != 0,
2549 ("vm_map_unwire: !HOLESOK and new/changed entry"));
2553 if (rv == KERN_SUCCESS && (!user_unwire ||
2554 (entry->eflags & MAP_ENTRY_USER_WIRED))) {
2556 entry->eflags &= ~MAP_ENTRY_USER_WIRED;
2557 if (entry->wired_count == 1)
2558 vm_map_entry_unwire(map, entry);
2560 entry->wired_count--;
2562 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0,
2563 ("vm_map_unwire: in-transition flag missing %p", entry));
2564 KASSERT(entry->wiring_thread == curthread,
2565 ("vm_map_unwire: alien wire %p", entry));
2566 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION;
2567 entry->wiring_thread = NULL;
2568 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
2569 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
2572 vm_map_simplify_entry(map, entry);
2581 * vm_map_wire_entry_failure:
2583 * Handle a wiring failure on the given entry.
2585 * The map should be locked.
2588 vm_map_wire_entry_failure(vm_map_t map, vm_map_entry_t entry,
2589 vm_offset_t failed_addr)
2592 VM_MAP_ASSERT_LOCKED(map);
2593 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 &&
2594 entry->wired_count == 1,
2595 ("vm_map_wire_entry_failure: entry %p isn't being wired", entry));
2596 KASSERT(failed_addr < entry->end,
2597 ("vm_map_wire_entry_failure: entry %p was fully wired", entry));
2600 * If any pages at the start of this entry were successfully wired,
2603 if (failed_addr > entry->start) {
2604 pmap_unwire(map->pmap, entry->start, failed_addr);
2605 vm_object_unwire(entry->object.vm_object, entry->offset,
2606 failed_addr - entry->start, PQ_ACTIVE);
2610 * Assign an out-of-range value to represent the failure to wire this
2613 entry->wired_count = -1;
2619 * Implements both kernel and user wiring.
2622 vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t end,
2625 vm_map_entry_t entry, first_entry, tmp_entry;
2626 vm_offset_t faddr, saved_end, saved_start;
2627 unsigned int last_timestamp;
2629 boolean_t need_wakeup, result, user_wire;
2633 return (KERN_SUCCESS);
2635 if (flags & VM_MAP_WIRE_WRITE)
2636 prot |= VM_PROT_WRITE;
2637 user_wire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE;
2639 VM_MAP_RANGE_CHECK(map, start, end);
2640 if (!vm_map_lookup_entry(map, start, &first_entry)) {
2641 if (flags & VM_MAP_WIRE_HOLESOK)
2642 first_entry = first_entry->next;
2645 return (KERN_INVALID_ADDRESS);
2648 last_timestamp = map->timestamp;
2649 entry = first_entry;
2650 while (entry->start < end) {
2651 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
2653 * We have not yet clipped the entry.
2655 saved_start = (start >= entry->start) ? start :
2657 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2658 if (vm_map_unlock_and_wait(map, 0)) {
2660 * Allow interruption of user wiring?
2664 if (last_timestamp + 1 != map->timestamp) {
2666 * Look again for the entry because the map was
2667 * modified while it was unlocked.
2668 * Specifically, the entry may have been
2669 * clipped, merged, or deleted.
2671 if (!vm_map_lookup_entry(map, saved_start,
2673 if (flags & VM_MAP_WIRE_HOLESOK)
2674 tmp_entry = tmp_entry->next;
2676 if (saved_start == start) {
2678 * first_entry has been deleted.
2681 return (KERN_INVALID_ADDRESS);
2684 rv = KERN_INVALID_ADDRESS;
2688 if (entry == first_entry)
2689 first_entry = tmp_entry;
2694 last_timestamp = map->timestamp;
2697 vm_map_clip_start(map, entry, start);
2698 vm_map_clip_end(map, entry, end);
2700 * Mark the entry in case the map lock is released. (See
2703 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 &&
2704 entry->wiring_thread == NULL,
2705 ("owned map entry %p", entry));
2706 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
2707 entry->wiring_thread = curthread;
2708 if ((entry->protection & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0
2709 || (entry->protection & prot) != prot) {
2710 entry->eflags |= MAP_ENTRY_WIRE_SKIPPED;
2711 if ((flags & VM_MAP_WIRE_HOLESOK) == 0) {
2713 rv = KERN_INVALID_ADDRESS;
2718 if (entry->wired_count == 0) {
2719 entry->wired_count++;
2720 saved_start = entry->start;
2721 saved_end = entry->end;
2724 * Release the map lock, relying on the in-transition
2725 * mark. Mark the map busy for fork.
2730 faddr = saved_start;
2733 * Simulate a fault to get the page and enter
2734 * it into the physical map.
2736 if ((rv = vm_fault(map, faddr, VM_PROT_NONE,
2737 VM_FAULT_WIRE)) != KERN_SUCCESS)
2739 } while ((faddr += PAGE_SIZE) < saved_end);
2742 if (last_timestamp + 1 != map->timestamp) {
2744 * Look again for the entry because the map was
2745 * modified while it was unlocked. The entry
2746 * may have been clipped, but NOT merged or
2749 result = vm_map_lookup_entry(map, saved_start,
2751 KASSERT(result, ("vm_map_wire: lookup failed"));
2752 if (entry == first_entry)
2753 first_entry = tmp_entry;
2757 while (entry->end < saved_end) {
2759 * In case of failure, handle entries
2760 * that were not fully wired here;
2761 * fully wired entries are handled
2764 if (rv != KERN_SUCCESS &&
2766 vm_map_wire_entry_failure(map,
2768 entry = entry->next;
2771 last_timestamp = map->timestamp;
2772 if (rv != KERN_SUCCESS) {
2773 vm_map_wire_entry_failure(map, entry, faddr);
2777 } else if (!user_wire ||
2778 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
2779 entry->wired_count++;
2782 * Check the map for holes in the specified region.
2783 * If VM_MAP_WIRE_HOLESOK was specified, skip this check.
2786 if ((flags & VM_MAP_WIRE_HOLESOK) == 0 &&
2787 entry->end < end && entry->next->start > entry->end) {
2789 rv = KERN_INVALID_ADDRESS;
2792 entry = entry->next;
2796 need_wakeup = FALSE;
2797 if (first_entry == NULL) {
2798 result = vm_map_lookup_entry(map, start, &first_entry);
2799 if (!result && (flags & VM_MAP_WIRE_HOLESOK))
2800 first_entry = first_entry->next;
2802 KASSERT(result, ("vm_map_wire: lookup failed"));
2804 for (entry = first_entry; entry->start < end; entry = entry->next) {
2806 * If VM_MAP_WIRE_HOLESOK was specified, an empty
2807 * space in the unwired region could have been mapped
2808 * while the map lock was dropped for faulting in the
2809 * pages or draining MAP_ENTRY_IN_TRANSITION.
2810 * Moreover, another thread could be simultaneously
2811 * wiring this new mapping entry. Detect these cases
2812 * and skip any entries marked as in transition not by us.
2814 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 ||
2815 entry->wiring_thread != curthread) {
2816 KASSERT((flags & VM_MAP_WIRE_HOLESOK) != 0,
2817 ("vm_map_wire: !HOLESOK and new/changed entry"));
2821 if ((entry->eflags & MAP_ENTRY_WIRE_SKIPPED) != 0)
2822 goto next_entry_done;
2824 if (rv == KERN_SUCCESS) {
2826 entry->eflags |= MAP_ENTRY_USER_WIRED;
2827 } else if (entry->wired_count == -1) {
2829 * Wiring failed on this entry. Thus, unwiring is
2832 entry->wired_count = 0;
2833 } else if (!user_wire ||
2834 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
2836 * Undo the wiring. Wiring succeeded on this entry
2837 * but failed on a later entry.
2839 if (entry->wired_count == 1)
2840 vm_map_entry_unwire(map, entry);
2842 entry->wired_count--;
2845 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0,
2846 ("vm_map_wire: in-transition flag missing %p", entry));
2847 KASSERT(entry->wiring_thread == curthread,
2848 ("vm_map_wire: alien wire %p", entry));
2849 entry->eflags &= ~(MAP_ENTRY_IN_TRANSITION |
2850 MAP_ENTRY_WIRE_SKIPPED);
2851 entry->wiring_thread = NULL;
2852 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
2853 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
2856 vm_map_simplify_entry(map, entry);
2867 * Push any dirty cached pages in the address range to their pager.
2868 * If syncio is TRUE, dirty pages are written synchronously.
2869 * If invalidate is TRUE, any cached pages are freed as well.
2871 * If the size of the region from start to end is zero, we are
2872 * supposed to flush all modified pages within the region containing
2873 * start. Unfortunately, a region can be split or coalesced with
2874 * neighboring regions, making it difficult to determine what the
2875 * original region was. Therefore, we approximate this requirement by
2876 * flushing the current region containing start.
2878 * Returns an error if any part of the specified range is not mapped.
2886 boolean_t invalidate)
2888 vm_map_entry_t current;
2889 vm_map_entry_t entry;
2892 vm_ooffset_t offset;
2893 unsigned int last_timestamp;
2896 vm_map_lock_read(map);
2897 VM_MAP_RANGE_CHECK(map, start, end);
2898 if (!vm_map_lookup_entry(map, start, &entry)) {
2899 vm_map_unlock_read(map);
2900 return (KERN_INVALID_ADDRESS);
2901 } else if (start == end) {
2902 start = entry->start;
2906 * Make a first pass to check for user-wired memory and holes.
2908 for (current = entry; current->start < end; current = current->next) {
2909 if (invalidate && (current->eflags & MAP_ENTRY_USER_WIRED)) {
2910 vm_map_unlock_read(map);
2911 return (KERN_INVALID_ARGUMENT);
2913 if (end > current->end &&
2914 current->end != current->next->start) {
2915 vm_map_unlock_read(map);
2916 return (KERN_INVALID_ADDRESS);
2921 pmap_remove(map->pmap, start, end);
2925 * Make a second pass, cleaning/uncaching pages from the indicated
2928 for (current = entry; current->start < end;) {
2929 offset = current->offset + (start - current->start);
2930 size = (end <= current->end ? end : current->end) - start;
2931 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
2933 vm_map_entry_t tentry;
2936 smap = current->object.sub_map;
2937 vm_map_lock_read(smap);
2938 (void) vm_map_lookup_entry(smap, offset, &tentry);
2939 tsize = tentry->end - offset;
2942 object = tentry->object.vm_object;
2943 offset = tentry->offset + (offset - tentry->start);
2944 vm_map_unlock_read(smap);
2946 object = current->object.vm_object;
2948 vm_object_reference(object);
2949 last_timestamp = map->timestamp;
2950 vm_map_unlock_read(map);
2951 if (!vm_object_sync(object, offset, size, syncio, invalidate))
2954 vm_object_deallocate(object);
2955 vm_map_lock_read(map);
2956 if (last_timestamp == map->timestamp ||
2957 !vm_map_lookup_entry(map, start, ¤t))
2958 current = current->next;
2961 vm_map_unlock_read(map);
2962 return (failed ? KERN_FAILURE : KERN_SUCCESS);
2966 * vm_map_entry_unwire: [ internal use only ]
2968 * Make the region specified by this entry pageable.
2970 * The map in question should be locked.
2971 * [This is the reason for this routine's existence.]
2974 vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry)
2977 VM_MAP_ASSERT_LOCKED(map);
2978 KASSERT(entry->wired_count > 0,
2979 ("vm_map_entry_unwire: entry %p isn't wired", entry));
2980 pmap_unwire(map->pmap, entry->start, entry->end);
2981 vm_object_unwire(entry->object.vm_object, entry->offset, entry->end -
2982 entry->start, PQ_ACTIVE);
2983 entry->wired_count = 0;
2987 vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map)
2990 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0)
2991 vm_object_deallocate(entry->object.vm_object);
2992 uma_zfree(system_map ? kmapentzone : mapentzone, entry);
2996 * vm_map_entry_delete: [ internal use only ]
2998 * Deallocate the given entry from the target map.
3001 vm_map_entry_delete(vm_map_t map, vm_map_entry_t entry)
3004 vm_pindex_t offidxstart, offidxend, count, size1;
3007 vm_map_entry_unlink(map, entry);
3008 object = entry->object.vm_object;
3010 if ((entry->eflags & MAP_ENTRY_GUARD) != 0) {
3011 MPASS(entry->cred == NULL);
3012 MPASS((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0);
3013 MPASS(object == NULL);
3014 vm_map_entry_deallocate(entry, map->system_map);
3018 size = entry->end - entry->start;
3021 if (entry->cred != NULL) {
3022 swap_release_by_cred(size, entry->cred);
3023 crfree(entry->cred);
3026 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0 &&
3028 KASSERT(entry->cred == NULL || object->cred == NULL ||
3029 (entry->eflags & MAP_ENTRY_NEEDS_COPY),
3030 ("OVERCOMMIT vm_map_entry_delete: both cred %p", entry));
3032 offidxstart = OFF_TO_IDX(entry->offset);
3033 offidxend = offidxstart + count;
3034 VM_OBJECT_WLOCK(object);
3035 if (object->ref_count != 1 && ((object->flags & (OBJ_NOSPLIT |
3036 OBJ_ONEMAPPING)) == OBJ_ONEMAPPING ||
3037 object == kernel_object)) {
3038 vm_object_collapse(object);
3041 * The option OBJPR_NOTMAPPED can be passed here
3042 * because vm_map_delete() already performed
3043 * pmap_remove() on the only mapping to this range
3046 vm_object_page_remove(object, offidxstart, offidxend,
3048 if (object->type == OBJT_SWAP)
3049 swap_pager_freespace(object, offidxstart,
3051 if (offidxend >= object->size &&
3052 offidxstart < object->size) {
3053 size1 = object->size;
3054 object->size = offidxstart;
3055 if (object->cred != NULL) {
3056 size1 -= object->size;
3057 KASSERT(object->charge >= ptoa(size1),
3058 ("object %p charge < 0", object));
3059 swap_release_by_cred(ptoa(size1),
3061 object->charge -= ptoa(size1);
3065 VM_OBJECT_WUNLOCK(object);
3067 entry->object.vm_object = NULL;
3068 if (map->system_map)
3069 vm_map_entry_deallocate(entry, TRUE);
3071 entry->next = curthread->td_map_def_user;
3072 curthread->td_map_def_user = entry;
3077 * vm_map_delete: [ internal use only ]
3079 * Deallocates the given address range from the target
3083 vm_map_delete(vm_map_t map, vm_offset_t start, vm_offset_t end)
3085 vm_map_entry_t entry;
3086 vm_map_entry_t first_entry;
3088 VM_MAP_ASSERT_LOCKED(map);
3090 return (KERN_SUCCESS);
3093 * Find the start of the region, and clip it
3095 if (!vm_map_lookup_entry(map, start, &first_entry))
3096 entry = first_entry->next;
3098 entry = first_entry;
3099 vm_map_clip_start(map, entry, start);
3103 * Step through all entries in this region
3105 while (entry->start < end) {
3106 vm_map_entry_t next;
3109 * Wait for wiring or unwiring of an entry to complete.
3110 * Also wait for any system wirings to disappear on
3113 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 ||
3114 (vm_map_pmap(map) != kernel_pmap &&
3115 vm_map_entry_system_wired_count(entry) != 0)) {
3116 unsigned int last_timestamp;
3117 vm_offset_t saved_start;
3118 vm_map_entry_t tmp_entry;
3120 saved_start = entry->start;
3121 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
3122 last_timestamp = map->timestamp;
3123 (void) vm_map_unlock_and_wait(map, 0);
3125 if (last_timestamp + 1 != map->timestamp) {
3127 * Look again for the entry because the map was
3128 * modified while it was unlocked.
3129 * Specifically, the entry may have been
3130 * clipped, merged, or deleted.
3132 if (!vm_map_lookup_entry(map, saved_start,
3134 entry = tmp_entry->next;
3137 vm_map_clip_start(map, entry,
3143 vm_map_clip_end(map, entry, end);
3148 * Unwire before removing addresses from the pmap; otherwise,
3149 * unwiring will put the entries back in the pmap.
3151 if (entry->wired_count != 0)
3152 vm_map_entry_unwire(map, entry);
3155 * Remove mappings for the pages, but only if the
3156 * mappings could exist. For instance, it does not
3157 * make sense to call pmap_remove() for guard entries.
3159 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0 ||
3160 entry->object.vm_object != NULL)
3161 pmap_remove(map->pmap, entry->start, entry->end);
3164 * Delete the entry only after removing all pmap
3165 * entries pointing to its pages. (Otherwise, its
3166 * page frames may be reallocated, and any modify bits
3167 * will be set in the wrong object!)
3169 vm_map_entry_delete(map, entry);
3172 return (KERN_SUCCESS);
3178 * Remove the given address range from the target map.
3179 * This is the exported form of vm_map_delete.
3182 vm_map_remove(vm_map_t map, vm_offset_t start, vm_offset_t end)
3187 VM_MAP_RANGE_CHECK(map, start, end);
3188 result = vm_map_delete(map, start, end);
3194 * vm_map_check_protection:
3196 * Assert that the target map allows the specified privilege on the
3197 * entire address region given. The entire region must be allocated.
3199 * WARNING! This code does not and should not check whether the
3200 * contents of the region is accessible. For example a smaller file
3201 * might be mapped into a larger address space.
3203 * NOTE! This code is also called by munmap().
3205 * The map must be locked. A read lock is sufficient.
3208 vm_map_check_protection(vm_map_t map, vm_offset_t start, vm_offset_t end,
3209 vm_prot_t protection)
3211 vm_map_entry_t entry;
3212 vm_map_entry_t tmp_entry;
3214 if (!vm_map_lookup_entry(map, start, &tmp_entry))
3218 while (start < end) {
3222 if (start < entry->start)
3225 * Check protection associated with entry.
3227 if ((entry->protection & protection) != protection)
3229 /* go to next entry */
3231 entry = entry->next;
3237 * vm_map_copy_entry:
3239 * Copies the contents of the source entry to the destination
3240 * entry. The entries *must* be aligned properly.
3246 vm_map_entry_t src_entry,
3247 vm_map_entry_t dst_entry,
3248 vm_ooffset_t *fork_charge)
3250 vm_object_t src_object;
3251 vm_map_entry_t fake_entry;
3256 VM_MAP_ASSERT_LOCKED(dst_map);
3258 if ((dst_entry->eflags|src_entry->eflags) & MAP_ENTRY_IS_SUB_MAP)
3261 if (src_entry->wired_count == 0 ||
3262 (src_entry->protection & VM_PROT_WRITE) == 0) {
3264 * If the source entry is marked needs_copy, it is already
3267 if ((src_entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0 &&
3268 (src_entry->protection & VM_PROT_WRITE) != 0) {
3269 pmap_protect(src_map->pmap,
3272 src_entry->protection & ~VM_PROT_WRITE);
3276 * Make a copy of the object.
3278 size = src_entry->end - src_entry->start;
3279 if ((src_object = src_entry->object.vm_object) != NULL) {
3280 VM_OBJECT_WLOCK(src_object);
3281 charged = ENTRY_CHARGED(src_entry);
3282 if (src_object->handle == NULL &&
3283 (src_object->type == OBJT_DEFAULT ||
3284 src_object->type == OBJT_SWAP)) {
3285 vm_object_collapse(src_object);
3286 if ((src_object->flags & (OBJ_NOSPLIT |
3287 OBJ_ONEMAPPING)) == OBJ_ONEMAPPING) {
3288 vm_object_split(src_entry);
3290 src_entry->object.vm_object;
3293 vm_object_reference_locked(src_object);
3294 vm_object_clear_flag(src_object, OBJ_ONEMAPPING);
3295 if (src_entry->cred != NULL &&
3296 !(src_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
3297 KASSERT(src_object->cred == NULL,
3298 ("OVERCOMMIT: vm_map_copy_entry: cred %p",
3300 src_object->cred = src_entry->cred;
3301 src_object->charge = size;
3303 VM_OBJECT_WUNLOCK(src_object);
3304 dst_entry->object.vm_object = src_object;
3306 cred = curthread->td_ucred;
3308 dst_entry->cred = cred;
3309 *fork_charge += size;
3310 if (!(src_entry->eflags &
3311 MAP_ENTRY_NEEDS_COPY)) {
3313 src_entry->cred = cred;
3314 *fork_charge += size;
3317 src_entry->eflags |= MAP_ENTRY_COW |
3318 MAP_ENTRY_NEEDS_COPY;
3319 dst_entry->eflags |= MAP_ENTRY_COW |
3320 MAP_ENTRY_NEEDS_COPY;
3321 dst_entry->offset = src_entry->offset;
3322 if (src_entry->eflags & MAP_ENTRY_VN_WRITECNT) {
3324 * MAP_ENTRY_VN_WRITECNT cannot
3325 * indicate write reference from
3326 * src_entry, since the entry is
3327 * marked as needs copy. Allocate a
3328 * fake entry that is used to
3329 * decrement object->un_pager.vnp.writecount
3330 * at the appropriate time. Attach
3331 * fake_entry to the deferred list.
3333 fake_entry = vm_map_entry_create(dst_map);
3334 fake_entry->eflags = MAP_ENTRY_VN_WRITECNT;
3335 src_entry->eflags &= ~MAP_ENTRY_VN_WRITECNT;
3336 vm_object_reference(src_object);
3337 fake_entry->object.vm_object = src_object;
3338 fake_entry->start = src_entry->start;
3339 fake_entry->end = src_entry->end;
3340 fake_entry->next = curthread->td_map_def_user;
3341 curthread->td_map_def_user = fake_entry;
3344 pmap_copy(dst_map->pmap, src_map->pmap,
3345 dst_entry->start, dst_entry->end - dst_entry->start,
3348 dst_entry->object.vm_object = NULL;
3349 dst_entry->offset = 0;
3350 if (src_entry->cred != NULL) {
3351 dst_entry->cred = curthread->td_ucred;
3352 crhold(dst_entry->cred);
3353 *fork_charge += size;
3358 * We don't want to make writeable wired pages copy-on-write.
3359 * Immediately copy these pages into the new map by simulating
3360 * page faults. The new pages are pageable.
3362 vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry,
3368 * vmspace_map_entry_forked:
3369 * Update the newly-forked vmspace each time a map entry is inherited
3370 * or copied. The values for vm_dsize and vm_tsize are approximate
3371 * (and mostly-obsolete ideas in the face of mmap(2) et al.)
3374 vmspace_map_entry_forked(const struct vmspace *vm1, struct vmspace *vm2,
3375 vm_map_entry_t entry)
3377 vm_size_t entrysize;
3380 if ((entry->eflags & MAP_ENTRY_GUARD) != 0)
3382 entrysize = entry->end - entry->start;
3383 vm2->vm_map.size += entrysize;
3384 if (entry->eflags & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP)) {
3385 vm2->vm_ssize += btoc(entrysize);
3386 } else if (entry->start >= (vm_offset_t)vm1->vm_daddr &&
3387 entry->start < (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)) {
3388 newend = MIN(entry->end,
3389 (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize));
3390 vm2->vm_dsize += btoc(newend - entry->start);
3391 } else if (entry->start >= (vm_offset_t)vm1->vm_taddr &&
3392 entry->start < (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)) {
3393 newend = MIN(entry->end,
3394 (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize));
3395 vm2->vm_tsize += btoc(newend - entry->start);
3401 * Create a new process vmspace structure and vm_map
3402 * based on those of an existing process. The new map
3403 * is based on the old map, according to the inheritance
3404 * values on the regions in that map.
3406 * XXX It might be worth coalescing the entries added to the new vmspace.
3408 * The source map must not be locked.
3411 vmspace_fork(struct vmspace *vm1, vm_ooffset_t *fork_charge)
3413 struct vmspace *vm2;
3414 vm_map_t new_map, old_map;
3415 vm_map_entry_t new_entry, old_entry;
3420 old_map = &vm1->vm_map;
3421 /* Copy immutable fields of vm1 to vm2. */
3422 vm2 = vmspace_alloc(vm_map_min(old_map), vm_map_max(old_map),
3426 vm2->vm_taddr = vm1->vm_taddr;
3427 vm2->vm_daddr = vm1->vm_daddr;
3428 vm2->vm_maxsaddr = vm1->vm_maxsaddr;
3429 vm_map_lock(old_map);
3431 vm_map_wait_busy(old_map);
3432 new_map = &vm2->vm_map;
3433 locked = vm_map_trylock(new_map); /* trylock to silence WITNESS */
3434 KASSERT(locked, ("vmspace_fork: lock failed"));
3436 old_entry = old_map->header.next;
3438 while (old_entry != &old_map->header) {
3439 if (old_entry->eflags & MAP_ENTRY_IS_SUB_MAP)
3440 panic("vm_map_fork: encountered a submap");
3442 inh = old_entry->inheritance;
3443 if ((old_entry->eflags & MAP_ENTRY_GUARD) != 0 &&
3444 inh != VM_INHERIT_NONE)
3445 inh = VM_INHERIT_COPY;
3448 case VM_INHERIT_NONE:
3451 case VM_INHERIT_SHARE:
3453 * Clone the entry, creating the shared object if necessary.
3455 object = old_entry->object.vm_object;
3456 if (object == NULL) {
3457 object = vm_object_allocate(OBJT_DEFAULT,
3458 atop(old_entry->end - old_entry->start));
3459 old_entry->object.vm_object = object;
3460 old_entry->offset = 0;
3461 if (old_entry->cred != NULL) {
3462 object->cred = old_entry->cred;
3463 object->charge = old_entry->end -
3465 old_entry->cred = NULL;
3470 * Add the reference before calling vm_object_shadow
3471 * to insure that a shadow object is created.
3473 vm_object_reference(object);
3474 if (old_entry->eflags & MAP_ENTRY_NEEDS_COPY) {
3475 vm_object_shadow(&old_entry->object.vm_object,
3477 old_entry->end - old_entry->start);
3478 old_entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
3479 /* Transfer the second reference too. */
3480 vm_object_reference(
3481 old_entry->object.vm_object);
3484 * As in vm_map_simplify_entry(), the
3485 * vnode lock will not be acquired in
3486 * this call to vm_object_deallocate().
3488 vm_object_deallocate(object);
3489 object = old_entry->object.vm_object;
3491 VM_OBJECT_WLOCK(object);
3492 vm_object_clear_flag(object, OBJ_ONEMAPPING);
3493 if (old_entry->cred != NULL) {
3494 KASSERT(object->cred == NULL, ("vmspace_fork both cred"));
3495 object->cred = old_entry->cred;
3496 object->charge = old_entry->end - old_entry->start;
3497 old_entry->cred = NULL;
3501 * Assert the correct state of the vnode
3502 * v_writecount while the object is locked, to
3503 * not relock it later for the assertion
3506 if (old_entry->eflags & MAP_ENTRY_VN_WRITECNT &&
3507 object->type == OBJT_VNODE) {
3508 KASSERT(((struct vnode *)object->handle)->
3510 ("vmspace_fork: v_writecount %p", object));
3511 KASSERT(object->un_pager.vnp.writemappings > 0,
3512 ("vmspace_fork: vnp.writecount %p",
3515 VM_OBJECT_WUNLOCK(object);
3518 * Clone the entry, referencing the shared object.
3520 new_entry = vm_map_entry_create(new_map);
3521 *new_entry = *old_entry;
3522 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED |
3523 MAP_ENTRY_IN_TRANSITION);
3524 new_entry->wiring_thread = NULL;
3525 new_entry->wired_count = 0;
3526 if (new_entry->eflags & MAP_ENTRY_VN_WRITECNT) {
3527 vnode_pager_update_writecount(object,
3528 new_entry->start, new_entry->end);
3532 * Insert the entry into the new map -- we know we're
3533 * inserting at the end of the new map.
3535 vm_map_entry_link(new_map, new_map->header.prev,
3537 vmspace_map_entry_forked(vm1, vm2, new_entry);
3540 * Update the physical map
3542 pmap_copy(new_map->pmap, old_map->pmap,
3544 (old_entry->end - old_entry->start),
3548 case VM_INHERIT_COPY:
3550 * Clone the entry and link into the map.
3552 new_entry = vm_map_entry_create(new_map);
3553 *new_entry = *old_entry;
3555 * Copied entry is COW over the old object.
3557 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED |
3558 MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_VN_WRITECNT);
3559 new_entry->wiring_thread = NULL;
3560 new_entry->wired_count = 0;
3561 new_entry->object.vm_object = NULL;
3562 new_entry->cred = NULL;
3563 vm_map_entry_link(new_map, new_map->header.prev,
3565 vmspace_map_entry_forked(vm1, vm2, new_entry);
3566 vm_map_copy_entry(old_map, new_map, old_entry,
3567 new_entry, fork_charge);
3570 case VM_INHERIT_ZERO:
3572 * Create a new anonymous mapping entry modelled from
3575 new_entry = vm_map_entry_create(new_map);
3576 memset(new_entry, 0, sizeof(*new_entry));
3578 new_entry->start = old_entry->start;
3579 new_entry->end = old_entry->end;
3580 new_entry->eflags = old_entry->eflags &
3581 ~(MAP_ENTRY_USER_WIRED | MAP_ENTRY_IN_TRANSITION |
3582 MAP_ENTRY_VN_WRITECNT);
3583 new_entry->protection = old_entry->protection;
3584 new_entry->max_protection = old_entry->max_protection;
3585 new_entry->inheritance = VM_INHERIT_ZERO;
3587 vm_map_entry_link(new_map, new_map->header.prev,
3589 vmspace_map_entry_forked(vm1, vm2, new_entry);
3591 new_entry->cred = curthread->td_ucred;
3592 crhold(new_entry->cred);
3593 *fork_charge += (new_entry->end - new_entry->start);
3597 old_entry = old_entry->next;
3600 * Use inlined vm_map_unlock() to postpone handling the deferred
3601 * map entries, which cannot be done until both old_map and
3602 * new_map locks are released.
3604 sx_xunlock(&old_map->lock);
3605 sx_xunlock(&new_map->lock);
3606 vm_map_process_deferred();
3612 * Create a process's stack for exec_new_vmspace(). This function is never
3613 * asked to wire the newly created stack.
3616 vm_map_stack(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
3617 vm_prot_t prot, vm_prot_t max, int cow)
3619 vm_size_t growsize, init_ssize;
3623 MPASS((map->flags & MAP_WIREFUTURE) == 0);
3624 growsize = sgrowsiz;
3625 init_ssize = (max_ssize < growsize) ? max_ssize : growsize;
3627 vmemlim = lim_cur(curthread, RLIMIT_VMEM);
3628 /* If we would blow our VMEM resource limit, no go */
3629 if (map->size + init_ssize > vmemlim) {
3633 rv = vm_map_stack_locked(map, addrbos, max_ssize, growsize, prot,
3640 static int stack_guard_page = 1;
3641 SYSCTL_INT(_security_bsd, OID_AUTO, stack_guard_page, CTLFLAG_RWTUN,
3642 &stack_guard_page, 0,
3643 "Specifies the number of guard pages for a stack that grows");
3646 vm_map_stack_locked(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
3647 vm_size_t growsize, vm_prot_t prot, vm_prot_t max, int cow)
3649 vm_map_entry_t new_entry, prev_entry;
3650 vm_offset_t bot, gap_bot, gap_top, top;
3651 vm_size_t init_ssize, sgp;
3655 * The stack orientation is piggybacked with the cow argument.
3656 * Extract it into orient and mask the cow argument so that we
3657 * don't pass it around further.
3659 orient = cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP);
3660 KASSERT(orient != 0, ("No stack grow direction"));
3661 KASSERT(orient != (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP),
3664 if (addrbos < vm_map_min(map) ||
3665 addrbos + max_ssize > vm_map_max(map) ||
3666 addrbos + max_ssize <= addrbos)
3667 return (KERN_INVALID_ADDRESS);
3668 sgp = (vm_size_t)stack_guard_page * PAGE_SIZE;
3669 if (sgp >= max_ssize)
3670 return (KERN_INVALID_ARGUMENT);
3672 init_ssize = growsize;
3673 if (max_ssize < init_ssize + sgp)
3674 init_ssize = max_ssize - sgp;
3676 /* If addr is already mapped, no go */
3677 if (vm_map_lookup_entry(map, addrbos, &prev_entry))
3678 return (KERN_NO_SPACE);
3681 * If we can't accommodate max_ssize in the current mapping, no go.
3683 if (prev_entry->next->start < addrbos + max_ssize)
3684 return (KERN_NO_SPACE);
3687 * We initially map a stack of only init_ssize. We will grow as
3688 * needed later. Depending on the orientation of the stack (i.e.
3689 * the grow direction) we either map at the top of the range, the
3690 * bottom of the range or in the middle.
3692 * Note: we would normally expect prot and max to be VM_PROT_ALL,
3693 * and cow to be 0. Possibly we should eliminate these as input
3694 * parameters, and just pass these values here in the insert call.
3696 if (orient == MAP_STACK_GROWS_DOWN) {
3697 bot = addrbos + max_ssize - init_ssize;
3698 top = bot + init_ssize;
3701 } else /* if (orient == MAP_STACK_GROWS_UP) */ {
3703 top = bot + init_ssize;
3705 gap_top = addrbos + max_ssize;
3707 rv = vm_map_insert(map, NULL, 0, bot, top, prot, max, cow);
3708 if (rv != KERN_SUCCESS)
3710 new_entry = prev_entry->next;
3711 KASSERT(new_entry->end == top || new_entry->start == bot,
3712 ("Bad entry start/end for new stack entry"));
3713 KASSERT((orient & MAP_STACK_GROWS_DOWN) == 0 ||
3714 (new_entry->eflags & MAP_ENTRY_GROWS_DOWN) != 0,
3715 ("new entry lacks MAP_ENTRY_GROWS_DOWN"));
3716 KASSERT((orient & MAP_STACK_GROWS_UP) == 0 ||
3717 (new_entry->eflags & MAP_ENTRY_GROWS_UP) != 0,
3718 ("new entry lacks MAP_ENTRY_GROWS_UP"));
3719 rv = vm_map_insert(map, NULL, 0, gap_bot, gap_top, VM_PROT_NONE,
3720 VM_PROT_NONE, MAP_CREATE_GUARD | (orient == MAP_STACK_GROWS_DOWN ?
3721 MAP_CREATE_STACK_GAP_DN : MAP_CREATE_STACK_GAP_UP));
3722 if (rv != KERN_SUCCESS)
3723 (void)vm_map_delete(map, bot, top);
3728 * Attempts to grow a vm stack entry. Returns KERN_SUCCESS if we
3729 * successfully grow the stack.
3732 vm_map_growstack(vm_map_t map, vm_offset_t addr, vm_map_entry_t gap_entry)
3734 vm_map_entry_t stack_entry;
3738 vm_offset_t gap_end, gap_start, grow_start;
3739 size_t grow_amount, guard, max_grow;
3740 rlim_t lmemlim, stacklim, vmemlim;
3742 bool gap_deleted, grow_down, is_procstack;
3754 * Disallow stack growth when the access is performed by a
3755 * debugger or AIO daemon. The reason is that the wrong
3756 * resource limits are applied.
3758 if (map != &p->p_vmspace->vm_map || p->p_textvp == NULL)
3759 return (KERN_FAILURE);
3761 MPASS(!map->system_map);
3763 guard = stack_guard_page * PAGE_SIZE;
3764 lmemlim = lim_cur(curthread, RLIMIT_MEMLOCK);
3765 stacklim = lim_cur(curthread, RLIMIT_STACK);
3766 vmemlim = lim_cur(curthread, RLIMIT_VMEM);
3768 /* If addr is not in a hole for a stack grow area, no need to grow. */
3769 if (gap_entry == NULL && !vm_map_lookup_entry(map, addr, &gap_entry))
3770 return (KERN_FAILURE);
3771 if ((gap_entry->eflags & MAP_ENTRY_GUARD) == 0)
3772 return (KERN_SUCCESS);
3773 if ((gap_entry->eflags & MAP_ENTRY_STACK_GAP_DN) != 0) {
3774 stack_entry = gap_entry->next;
3775 if ((stack_entry->eflags & MAP_ENTRY_GROWS_DOWN) == 0 ||
3776 stack_entry->start != gap_entry->end)
3777 return (KERN_FAILURE);
3778 grow_amount = round_page(stack_entry->start - addr);
3780 } else if ((gap_entry->eflags & MAP_ENTRY_STACK_GAP_UP) != 0) {
3781 stack_entry = gap_entry->prev;
3782 if ((stack_entry->eflags & MAP_ENTRY_GROWS_UP) == 0 ||
3783 stack_entry->end != gap_entry->start)
3784 return (KERN_FAILURE);
3785 grow_amount = round_page(addr + 1 - stack_entry->end);
3788 return (KERN_FAILURE);
3790 max_grow = gap_entry->end - gap_entry->start;
3791 if (guard > max_grow)
3792 return (KERN_NO_SPACE);
3794 if (grow_amount > max_grow)
3795 return (KERN_NO_SPACE);
3798 * If this is the main process stack, see if we're over the stack
3801 is_procstack = addr >= (vm_offset_t)vm->vm_maxsaddr &&
3802 addr < (vm_offset_t)p->p_sysent->sv_usrstack;
3803 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim))
3804 return (KERN_NO_SPACE);
3809 if (is_procstack && racct_set(p, RACCT_STACK,
3810 ctob(vm->vm_ssize) + grow_amount)) {
3812 return (KERN_NO_SPACE);
3818 grow_amount = roundup(grow_amount, sgrowsiz);
3819 if (grow_amount > max_grow)
3820 grow_amount = max_grow;
3821 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) {
3822 grow_amount = trunc_page((vm_size_t)stacklim) -
3828 limit = racct_get_available(p, RACCT_STACK);
3830 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > limit))
3831 grow_amount = limit - ctob(vm->vm_ssize);
3834 if (!old_mlock && (map->flags & MAP_WIREFUTURE) != 0) {
3835 if (ptoa(pmap_wired_count(map->pmap)) + grow_amount > lmemlim) {
3842 if (racct_set(p, RACCT_MEMLOCK,
3843 ptoa(pmap_wired_count(map->pmap)) + grow_amount)) {
3853 /* If we would blow our VMEM resource limit, no go */
3854 if (map->size + grow_amount > vmemlim) {
3861 if (racct_set(p, RACCT_VMEM, map->size + grow_amount)) {
3870 if (vm_map_lock_upgrade(map)) {
3872 vm_map_lock_read(map);
3877 grow_start = gap_entry->end - grow_amount;
3878 if (gap_entry->start + grow_amount == gap_entry->end) {
3879 gap_start = gap_entry->start;
3880 gap_end = gap_entry->end;
3881 vm_map_entry_delete(map, gap_entry);
3884 MPASS(gap_entry->start < gap_entry->end - grow_amount);
3885 gap_entry->end -= grow_amount;
3886 vm_map_entry_resize_free(map, gap_entry);
3887 gap_deleted = false;
3889 rv = vm_map_insert(map, NULL, 0, grow_start,
3890 grow_start + grow_amount,
3891 stack_entry->protection, stack_entry->max_protection,
3892 MAP_STACK_GROWS_DOWN);
3893 if (rv != KERN_SUCCESS) {
3895 rv1 = vm_map_insert(map, NULL, 0, gap_start,
3896 gap_end, VM_PROT_NONE, VM_PROT_NONE,
3897 MAP_CREATE_GUARD | MAP_CREATE_STACK_GAP_DN);
3898 MPASS(rv1 == KERN_SUCCESS);
3900 gap_entry->end += grow_amount;
3901 vm_map_entry_resize_free(map, gap_entry);
3905 grow_start = stack_entry->end;
3906 cred = stack_entry->cred;
3907 if (cred == NULL && stack_entry->object.vm_object != NULL)
3908 cred = stack_entry->object.vm_object->cred;
3909 if (cred != NULL && !swap_reserve_by_cred(grow_amount, cred))
3911 /* Grow the underlying object if applicable. */
3912 else if (stack_entry->object.vm_object == NULL ||
3913 vm_object_coalesce(stack_entry->object.vm_object,
3914 stack_entry->offset,
3915 (vm_size_t)(stack_entry->end - stack_entry->start),
3916 (vm_size_t)grow_amount, cred != NULL)) {
3917 if (gap_entry->start + grow_amount == gap_entry->end)
3918 vm_map_entry_delete(map, gap_entry);
3920 gap_entry->start += grow_amount;
3921 stack_entry->end += grow_amount;
3922 map->size += grow_amount;
3923 vm_map_entry_resize_free(map, stack_entry);
3928 if (rv == KERN_SUCCESS && is_procstack)
3929 vm->vm_ssize += btoc(grow_amount);
3932 * Heed the MAP_WIREFUTURE flag if it was set for this process.
3934 if (rv == KERN_SUCCESS && (map->flags & MAP_WIREFUTURE) != 0) {
3936 vm_map_wire(map, grow_start, grow_start + grow_amount,
3937 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
3938 vm_map_lock_read(map);
3940 vm_map_lock_downgrade(map);
3944 if (racct_enable && rv != KERN_SUCCESS) {
3946 error = racct_set(p, RACCT_VMEM, map->size);
3947 KASSERT(error == 0, ("decreasing RACCT_VMEM failed"));
3949 error = racct_set(p, RACCT_MEMLOCK,
3950 ptoa(pmap_wired_count(map->pmap)));
3951 KASSERT(error == 0, ("decreasing RACCT_MEMLOCK failed"));
3953 error = racct_set(p, RACCT_STACK, ctob(vm->vm_ssize));
3954 KASSERT(error == 0, ("decreasing RACCT_STACK failed"));
3963 * Unshare the specified VM space for exec. If other processes are
3964 * mapped to it, then create a new one. The new vmspace is null.
3967 vmspace_exec(struct proc *p, vm_offset_t minuser, vm_offset_t maxuser)
3969 struct vmspace *oldvmspace = p->p_vmspace;
3970 struct vmspace *newvmspace;
3972 KASSERT((curthread->td_pflags & TDP_EXECVMSPC) == 0,
3973 ("vmspace_exec recursed"));
3974 newvmspace = vmspace_alloc(minuser, maxuser, pmap_pinit);
3975 if (newvmspace == NULL)
3977 newvmspace->vm_swrss = oldvmspace->vm_swrss;
3979 * This code is written like this for prototype purposes. The
3980 * goal is to avoid running down the vmspace here, but let the
3981 * other process's that are still using the vmspace to finally
3982 * run it down. Even though there is little or no chance of blocking
3983 * here, it is a good idea to keep this form for future mods.
3985 PROC_VMSPACE_LOCK(p);
3986 p->p_vmspace = newvmspace;
3987 PROC_VMSPACE_UNLOCK(p);
3988 if (p == curthread->td_proc)
3989 pmap_activate(curthread);
3990 curthread->td_pflags |= TDP_EXECVMSPC;
3995 * Unshare the specified VM space for forcing COW. This
3996 * is called by rfork, for the (RFMEM|RFPROC) == 0 case.
3999 vmspace_unshare(struct proc *p)
4001 struct vmspace *oldvmspace = p->p_vmspace;
4002 struct vmspace *newvmspace;
4003 vm_ooffset_t fork_charge;
4005 if (oldvmspace->vm_refcnt == 1)
4008 newvmspace = vmspace_fork(oldvmspace, &fork_charge);
4009 if (newvmspace == NULL)
4011 if (!swap_reserve_by_cred(fork_charge, p->p_ucred)) {
4012 vmspace_free(newvmspace);
4015 PROC_VMSPACE_LOCK(p);
4016 p->p_vmspace = newvmspace;
4017 PROC_VMSPACE_UNLOCK(p);
4018 if (p == curthread->td_proc)
4019 pmap_activate(curthread);
4020 vmspace_free(oldvmspace);
4027 * Finds the VM object, offset, and
4028 * protection for a given virtual address in the
4029 * specified map, assuming a page fault of the
4032 * Leaves the map in question locked for read; return
4033 * values are guaranteed until a vm_map_lookup_done
4034 * call is performed. Note that the map argument
4035 * is in/out; the returned map must be used in
4036 * the call to vm_map_lookup_done.
4038 * A handle (out_entry) is returned for use in
4039 * vm_map_lookup_done, to make that fast.
4041 * If a lookup is requested with "write protection"
4042 * specified, the map may be changed to perform virtual
4043 * copying operations, although the data referenced will
4047 vm_map_lookup(vm_map_t *var_map, /* IN/OUT */
4049 vm_prot_t fault_typea,
4050 vm_map_entry_t *out_entry, /* OUT */
4051 vm_object_t *object, /* OUT */
4052 vm_pindex_t *pindex, /* OUT */
4053 vm_prot_t *out_prot, /* OUT */
4054 boolean_t *wired) /* OUT */
4056 vm_map_entry_t entry;
4057 vm_map_t map = *var_map;
4059 vm_prot_t fault_type = fault_typea;
4060 vm_object_t eobject;
4066 vm_map_lock_read(map);
4070 * Lookup the faulting address.
4072 if (!vm_map_lookup_entry(map, vaddr, out_entry)) {
4073 vm_map_unlock_read(map);
4074 return (KERN_INVALID_ADDRESS);
4082 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
4083 vm_map_t old_map = map;
4085 *var_map = map = entry->object.sub_map;
4086 vm_map_unlock_read(old_map);
4091 * Check whether this task is allowed to have this page.
4093 prot = entry->protection;
4094 if ((fault_typea & VM_PROT_FAULT_LOOKUP) != 0) {
4095 fault_typea &= ~VM_PROT_FAULT_LOOKUP;
4096 if (prot == VM_PROT_NONE && map != kernel_map &&
4097 (entry->eflags & MAP_ENTRY_GUARD) != 0 &&
4098 (entry->eflags & (MAP_ENTRY_STACK_GAP_DN |
4099 MAP_ENTRY_STACK_GAP_UP)) != 0 &&
4100 vm_map_growstack(map, vaddr, entry) == KERN_SUCCESS)
4101 goto RetryLookupLocked;
4103 fault_type &= VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE;
4104 if ((fault_type & prot) != fault_type || prot == VM_PROT_NONE) {
4105 vm_map_unlock_read(map);
4106 return (KERN_PROTECTION_FAILURE);
4108 KASSERT((prot & VM_PROT_WRITE) == 0 || (entry->eflags &
4109 (MAP_ENTRY_USER_WIRED | MAP_ENTRY_NEEDS_COPY)) !=
4110 (MAP_ENTRY_USER_WIRED | MAP_ENTRY_NEEDS_COPY),
4111 ("entry %p flags %x", entry, entry->eflags));
4112 if ((fault_typea & VM_PROT_COPY) != 0 &&
4113 (entry->max_protection & VM_PROT_WRITE) == 0 &&
4114 (entry->eflags & MAP_ENTRY_COW) == 0) {
4115 vm_map_unlock_read(map);
4116 return (KERN_PROTECTION_FAILURE);
4120 * If this page is not pageable, we have to get it for all possible
4123 *wired = (entry->wired_count != 0);
4125 fault_type = entry->protection;
4126 size = entry->end - entry->start;
4128 * If the entry was copy-on-write, we either ...
4130 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
4132 * If we want to write the page, we may as well handle that
4133 * now since we've got the map locked.
4135 * If we don't need to write the page, we just demote the
4136 * permissions allowed.
4138 if ((fault_type & VM_PROT_WRITE) != 0 ||
4139 (fault_typea & VM_PROT_COPY) != 0) {
4141 * Make a new object, and place it in the object
4142 * chain. Note that no new references have appeared
4143 * -- one just moved from the map to the new
4146 if (vm_map_lock_upgrade(map))
4149 if (entry->cred == NULL) {
4151 * The debugger owner is charged for
4154 cred = curthread->td_ucred;
4156 if (!swap_reserve_by_cred(size, cred)) {
4159 return (KERN_RESOURCE_SHORTAGE);
4163 vm_object_shadow(&entry->object.vm_object,
4164 &entry->offset, size);
4165 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
4166 eobject = entry->object.vm_object;
4167 if (eobject->cred != NULL) {
4169 * The object was not shadowed.
4171 swap_release_by_cred(size, entry->cred);
4172 crfree(entry->cred);
4174 } else if (entry->cred != NULL) {
4175 VM_OBJECT_WLOCK(eobject);
4176 eobject->cred = entry->cred;
4177 eobject->charge = size;
4178 VM_OBJECT_WUNLOCK(eobject);
4182 vm_map_lock_downgrade(map);
4185 * We're attempting to read a copy-on-write page --
4186 * don't allow writes.
4188 prot &= ~VM_PROT_WRITE;
4193 * Create an object if necessary.
4195 if (entry->object.vm_object == NULL &&
4197 if (vm_map_lock_upgrade(map))
4199 entry->object.vm_object = vm_object_allocate(OBJT_DEFAULT,
4202 if (entry->cred != NULL) {
4203 VM_OBJECT_WLOCK(entry->object.vm_object);
4204 entry->object.vm_object->cred = entry->cred;
4205 entry->object.vm_object->charge = size;
4206 VM_OBJECT_WUNLOCK(entry->object.vm_object);
4209 vm_map_lock_downgrade(map);
4213 * Return the object/offset from this entry. If the entry was
4214 * copy-on-write or empty, it has been fixed up.
4216 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
4217 *object = entry->object.vm_object;
4220 return (KERN_SUCCESS);
4224 * vm_map_lookup_locked:
4226 * Lookup the faulting address. A version of vm_map_lookup that returns
4227 * KERN_FAILURE instead of blocking on map lock or memory allocation.
4230 vm_map_lookup_locked(vm_map_t *var_map, /* IN/OUT */
4232 vm_prot_t fault_typea,
4233 vm_map_entry_t *out_entry, /* OUT */
4234 vm_object_t *object, /* OUT */
4235 vm_pindex_t *pindex, /* OUT */
4236 vm_prot_t *out_prot, /* OUT */
4237 boolean_t *wired) /* OUT */
4239 vm_map_entry_t entry;
4240 vm_map_t map = *var_map;
4242 vm_prot_t fault_type = fault_typea;
4245 * Lookup the faulting address.
4247 if (!vm_map_lookup_entry(map, vaddr, out_entry))
4248 return (KERN_INVALID_ADDRESS);
4253 * Fail if the entry refers to a submap.
4255 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP)
4256 return (KERN_FAILURE);
4259 * Check whether this task is allowed to have this page.
4261 prot = entry->protection;
4262 fault_type &= VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE;
4263 if ((fault_type & prot) != fault_type)
4264 return (KERN_PROTECTION_FAILURE);
4267 * If this page is not pageable, we have to get it for all possible
4270 *wired = (entry->wired_count != 0);
4272 fault_type = entry->protection;
4274 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
4276 * Fail if the entry was copy-on-write for a write fault.
4278 if (fault_type & VM_PROT_WRITE)
4279 return (KERN_FAILURE);
4281 * We're attempting to read a copy-on-write page --
4282 * don't allow writes.
4284 prot &= ~VM_PROT_WRITE;
4288 * Fail if an object should be created.
4290 if (entry->object.vm_object == NULL && !map->system_map)
4291 return (KERN_FAILURE);
4294 * Return the object/offset from this entry. If the entry was
4295 * copy-on-write or empty, it has been fixed up.
4297 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
4298 *object = entry->object.vm_object;
4301 return (KERN_SUCCESS);
4305 * vm_map_lookup_done:
4307 * Releases locks acquired by a vm_map_lookup
4308 * (according to the handle returned by that lookup).
4311 vm_map_lookup_done(vm_map_t map, vm_map_entry_t entry)
4314 * Unlock the main-level map
4316 vm_map_unlock_read(map);
4320 vm_map_max_KBI(const struct vm_map *map)
4323 return (vm_map_max(map));
4327 vm_map_min_KBI(const struct vm_map *map)
4330 return (vm_map_min(map));
4334 vm_map_pmap_KBI(vm_map_t map)
4340 #include "opt_ddb.h"
4342 #include <sys/kernel.h>
4344 #include <ddb/ddb.h>
4347 vm_map_print(vm_map_t map)
4349 vm_map_entry_t entry;
4351 db_iprintf("Task map %p: pmap=%p, nentries=%d, version=%u\n",
4353 (void *)map->pmap, map->nentries, map->timestamp);
4356 for (entry = map->header.next; entry != &map->header;
4357 entry = entry->next) {
4358 db_iprintf("map entry %p: start=%p, end=%p, eflags=%#x, \n",
4359 (void *)entry, (void *)entry->start, (void *)entry->end,
4362 static char *inheritance_name[4] =
4363 {"share", "copy", "none", "donate_copy"};
4365 db_iprintf(" prot=%x/%x/%s",
4367 entry->max_protection,
4368 inheritance_name[(int)(unsigned char)entry->inheritance]);
4369 if (entry->wired_count != 0)
4370 db_printf(", wired");
4372 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
4373 db_printf(", share=%p, offset=0x%jx\n",
4374 (void *)entry->object.sub_map,
4375 (uintmax_t)entry->offset);
4376 if ((entry->prev == &map->header) ||
4377 (entry->prev->object.sub_map !=
4378 entry->object.sub_map)) {
4380 vm_map_print((vm_map_t)entry->object.sub_map);
4384 if (entry->cred != NULL)
4385 db_printf(", ruid %d", entry->cred->cr_ruid);
4386 db_printf(", object=%p, offset=0x%jx",
4387 (void *)entry->object.vm_object,
4388 (uintmax_t)entry->offset);
4389 if (entry->object.vm_object && entry->object.vm_object->cred)
4390 db_printf(", obj ruid %d charge %jx",
4391 entry->object.vm_object->cred->cr_ruid,
4392 (uintmax_t)entry->object.vm_object->charge);
4393 if (entry->eflags & MAP_ENTRY_COW)
4394 db_printf(", copy (%s)",
4395 (entry->eflags & MAP_ENTRY_NEEDS_COPY) ? "needed" : "done");
4398 if ((entry->prev == &map->header) ||
4399 (entry->prev->object.vm_object !=
4400 entry->object.vm_object)) {
4402 vm_object_print((db_expr_t)(intptr_t)
4403 entry->object.vm_object,
4412 DB_SHOW_COMMAND(map, map)
4416 db_printf("usage: show map <addr>\n");
4419 vm_map_print((vm_map_t)addr);
4422 DB_SHOW_COMMAND(procvm, procvm)
4427 p = db_lookup_proc(addr);
4432 db_printf("p = %p, vmspace = %p, map = %p, pmap = %p\n",
4433 (void *)p, (void *)p->p_vmspace, (void *)&p->p_vmspace->vm_map,
4434 (void *)vmspace_pmap(p->p_vmspace));
4436 vm_map_print((vm_map_t)&p->p_vmspace->vm_map);