2 * SPDX-License-Identifier: BSD-3-Clause
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 void vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map);
136 static void vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry);
137 static void vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry);
138 static int vm_map_growstack(vm_map_t map, vm_offset_t addr,
139 vm_map_entry_t gap_entry);
140 static void vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot,
141 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags);
143 static void vm_map_zdtor(void *mem, int size, void *arg);
144 static void vmspace_zdtor(void *mem, int size, void *arg);
146 static int vm_map_stack_locked(vm_map_t map, vm_offset_t addrbos,
147 vm_size_t max_ssize, vm_size_t growsize, vm_prot_t prot, vm_prot_t max,
149 static void vm_map_wire_entry_failure(vm_map_t map, vm_map_entry_t entry,
150 vm_offset_t failed_addr);
152 #define ENTRY_CHARGED(e) ((e)->cred != NULL || \
153 ((e)->object.vm_object != NULL && (e)->object.vm_object->cred != NULL && \
154 !((e)->eflags & MAP_ENTRY_NEEDS_COPY)))
157 * PROC_VMSPACE_{UN,}LOCK() can be a noop as long as vmspaces are type
160 #define PROC_VMSPACE_LOCK(p) do { } while (0)
161 #define PROC_VMSPACE_UNLOCK(p) do { } while (0)
164 * VM_MAP_RANGE_CHECK: [ internal use only ]
166 * Asserts that the starting and ending region
167 * addresses fall within the valid range of the map.
169 #define VM_MAP_RANGE_CHECK(map, start, end) \
171 if (start < vm_map_min(map)) \
172 start = vm_map_min(map); \
173 if (end > vm_map_max(map)) \
174 end = vm_map_max(map); \
182 * Initialize the vm_map module. Must be called before
183 * any other vm_map routines.
185 * Map and entry structures are allocated from the general
186 * purpose memory pool with some exceptions:
188 * - The kernel map and kmem submap are allocated statically.
189 * - Kernel map entries are allocated out of a static pool.
191 * These restrictions are necessary since malloc() uses the
192 * maps and requires map entries.
198 mtx_init(&map_sleep_mtx, "vm map sleep mutex", NULL, MTX_DEF);
199 mapzone = uma_zcreate("MAP", sizeof(struct vm_map), NULL,
205 vm_map_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
206 uma_prealloc(mapzone, MAX_KMAP);
207 kmapentzone = uma_zcreate("KMAP ENTRY", sizeof(struct vm_map_entry),
208 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR,
209 UMA_ZONE_MTXCLASS | UMA_ZONE_VM);
210 mapentzone = uma_zcreate("MAP ENTRY", sizeof(struct vm_map_entry),
211 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
212 vmspace_zone = uma_zcreate("VMSPACE", sizeof(struct vmspace), NULL,
218 vmspace_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
222 vmspace_zinit(void *mem, int size, int flags)
226 vm = (struct vmspace *)mem;
228 vm->vm_map.pmap = NULL;
229 (void)vm_map_zinit(&vm->vm_map, sizeof(vm->vm_map), flags);
230 PMAP_LOCK_INIT(vmspace_pmap(vm));
235 vm_map_zinit(void *mem, int size, int flags)
240 memset(map, 0, sizeof(*map));
241 mtx_init(&map->system_mtx, "vm map (system)", NULL, MTX_DEF | MTX_DUPOK);
242 sx_init(&map->lock, "vm map (user)");
248 vmspace_zdtor(void *mem, int size, void *arg)
252 vm = (struct vmspace *)mem;
254 vm_map_zdtor(&vm->vm_map, sizeof(vm->vm_map), arg);
257 vm_map_zdtor(void *mem, int size, void *arg)
262 KASSERT(map->nentries == 0,
263 ("map %p nentries == %d on free.",
264 map, map->nentries));
265 KASSERT(map->size == 0,
266 ("map %p size == %lu on free.",
267 map, (unsigned long)map->size));
269 #endif /* INVARIANTS */
272 * Allocate a vmspace structure, including a vm_map and pmap,
273 * and initialize those structures. The refcnt is set to 1.
275 * If 'pinit' is NULL then the embedded pmap is initialized via pmap_pinit().
278 vmspace_alloc(vm_offset_t min, vm_offset_t max, pmap_pinit_t pinit)
282 vm = uma_zalloc(vmspace_zone, M_WAITOK);
284 KASSERT(vm->vm_map.pmap == NULL, ("vm_map.pmap must be NULL"));
289 if (!pinit(vmspace_pmap(vm))) {
290 uma_zfree(vmspace_zone, vm);
293 CTR1(KTR_VM, "vmspace_alloc: %p", vm);
294 _vm_map_init(&vm->vm_map, vmspace_pmap(vm), min, max);
309 vmspace_container_reset(struct proc *p)
313 racct_set(p, RACCT_DATA, 0);
314 racct_set(p, RACCT_STACK, 0);
315 racct_set(p, RACCT_RSS, 0);
316 racct_set(p, RACCT_MEMLOCK, 0);
317 racct_set(p, RACCT_VMEM, 0);
323 vmspace_dofree(struct vmspace *vm)
326 CTR1(KTR_VM, "vmspace_free: %p", vm);
329 * Make sure any SysV shm is freed, it might not have been in
335 * Lock the map, to wait out all other references to it.
336 * Delete all of the mappings and pages they hold, then call
337 * the pmap module to reclaim anything left.
339 (void)vm_map_remove(&vm->vm_map, vm->vm_map.min_offset,
340 vm->vm_map.max_offset);
342 pmap_release(vmspace_pmap(vm));
343 vm->vm_map.pmap = NULL;
344 uma_zfree(vmspace_zone, vm);
348 vmspace_free(struct vmspace *vm)
351 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
352 "vmspace_free() called");
354 if (vm->vm_refcnt == 0)
355 panic("vmspace_free: attempt to free already freed vmspace");
357 if (atomic_fetchadd_int(&vm->vm_refcnt, -1) == 1)
362 vmspace_exitfree(struct proc *p)
366 PROC_VMSPACE_LOCK(p);
369 PROC_VMSPACE_UNLOCK(p);
370 KASSERT(vm == &vmspace0, ("vmspace_exitfree: wrong vmspace"));
375 vmspace_exit(struct thread *td)
382 * Release user portion of address space.
383 * This releases references to vnodes,
384 * which could cause I/O if the file has been unlinked.
385 * Need to do this early enough that we can still sleep.
387 * The last exiting process to reach this point releases as
388 * much of the environment as it can. vmspace_dofree() is the
389 * slower fallback in case another process had a temporary
390 * reference to the vmspace.
395 atomic_add_int(&vmspace0.vm_refcnt, 1);
397 refcnt = vm->vm_refcnt;
398 if (refcnt > 1 && p->p_vmspace != &vmspace0) {
399 /* Switch now since other proc might free vmspace */
400 PROC_VMSPACE_LOCK(p);
401 p->p_vmspace = &vmspace0;
402 PROC_VMSPACE_UNLOCK(p);
405 } while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt - 1));
407 if (p->p_vmspace != vm) {
408 /* vmspace not yet freed, switch back */
409 PROC_VMSPACE_LOCK(p);
411 PROC_VMSPACE_UNLOCK(p);
414 pmap_remove_pages(vmspace_pmap(vm));
415 /* Switch now since this proc will free vmspace */
416 PROC_VMSPACE_LOCK(p);
417 p->p_vmspace = &vmspace0;
418 PROC_VMSPACE_UNLOCK(p);
424 vmspace_container_reset(p);
428 /* Acquire reference to vmspace owned by another process. */
431 vmspace_acquire_ref(struct proc *p)
436 PROC_VMSPACE_LOCK(p);
439 PROC_VMSPACE_UNLOCK(p);
443 refcnt = vm->vm_refcnt;
444 if (refcnt <= 0) { /* Avoid 0->1 transition */
445 PROC_VMSPACE_UNLOCK(p);
448 } while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt + 1));
449 if (vm != p->p_vmspace) {
450 PROC_VMSPACE_UNLOCK(p);
454 PROC_VMSPACE_UNLOCK(p);
459 * Switch between vmspaces in an AIO kernel process.
461 * The AIO kernel processes switch to and from a user process's
462 * vmspace while performing an I/O operation on behalf of a user
463 * process. The new vmspace is either the vmspace of a user process
464 * obtained from an active AIO request or the initial vmspace of the
465 * AIO kernel process (when it is idling). Because user processes
466 * will block to drain any active AIO requests before proceeding in
467 * exit() or execve(), the vmspace reference count for these vmspaces
468 * can never be 0. This allows for a much simpler implementation than
469 * the loop in vmspace_acquire_ref() above. Similarly, AIO kernel
470 * processes hold an extra reference on their initial vmspace for the
471 * life of the process so that this guarantee is true for any vmspace
475 vmspace_switch_aio(struct vmspace *newvm)
477 struct vmspace *oldvm;
479 /* XXX: Need some way to assert that this is an aio daemon. */
481 KASSERT(newvm->vm_refcnt > 0,
482 ("vmspace_switch_aio: newvm unreferenced"));
484 oldvm = curproc->p_vmspace;
489 * Point to the new address space and refer to it.
491 curproc->p_vmspace = newvm;
492 atomic_add_int(&newvm->vm_refcnt, 1);
494 /* Activate the new mapping. */
495 pmap_activate(curthread);
497 /* Remove the daemon's reference to the old address space. */
498 KASSERT(oldvm->vm_refcnt > 1,
499 ("vmspace_switch_aio: oldvm dropping last reference"));
504 _vm_map_lock(vm_map_t map, const char *file, int line)
508 mtx_lock_flags_(&map->system_mtx, 0, file, line);
510 sx_xlock_(&map->lock, file, line);
515 vm_map_process_deferred(void)
518 vm_map_entry_t entry, next;
522 entry = td->td_map_def_user;
523 td->td_map_def_user = NULL;
524 while (entry != NULL) {
526 if ((entry->eflags & MAP_ENTRY_VN_WRITECNT) != 0) {
528 * Decrement the object's writemappings and
529 * possibly the vnode's v_writecount.
531 KASSERT((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0,
532 ("Submap with writecount"));
533 object = entry->object.vm_object;
534 KASSERT(object != NULL, ("No object for writecount"));
535 vnode_pager_release_writecount(object, entry->start,
538 vm_map_entry_deallocate(entry, FALSE);
544 _vm_map_unlock(vm_map_t map, const char *file, int line)
548 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
550 sx_xunlock_(&map->lock, file, line);
551 vm_map_process_deferred();
556 _vm_map_lock_read(vm_map_t map, const char *file, int line)
560 mtx_lock_flags_(&map->system_mtx, 0, file, line);
562 sx_slock_(&map->lock, file, line);
566 _vm_map_unlock_read(vm_map_t map, const char *file, int line)
570 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
572 sx_sunlock_(&map->lock, file, line);
573 vm_map_process_deferred();
578 _vm_map_trylock(vm_map_t map, const char *file, int line)
582 error = map->system_map ?
583 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) :
584 !sx_try_xlock_(&map->lock, file, line);
591 _vm_map_trylock_read(vm_map_t map, const char *file, int line)
595 error = map->system_map ?
596 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) :
597 !sx_try_slock_(&map->lock, file, line);
602 * _vm_map_lock_upgrade: [ internal use only ]
604 * Tries to upgrade a read (shared) lock on the specified map to a write
605 * (exclusive) lock. Returns the value "0" if the upgrade succeeds and a
606 * non-zero value if the upgrade fails. If the upgrade fails, the map is
607 * returned without a read or write lock held.
609 * Requires that the map be read locked.
612 _vm_map_lock_upgrade(vm_map_t map, const char *file, int line)
614 unsigned int last_timestamp;
616 if (map->system_map) {
617 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
619 if (!sx_try_upgrade_(&map->lock, file, line)) {
620 last_timestamp = map->timestamp;
621 sx_sunlock_(&map->lock, file, line);
622 vm_map_process_deferred();
624 * If the map's timestamp does not change while the
625 * map is unlocked, then the upgrade succeeds.
627 sx_xlock_(&map->lock, file, line);
628 if (last_timestamp != map->timestamp) {
629 sx_xunlock_(&map->lock, file, line);
639 _vm_map_lock_downgrade(vm_map_t map, const char *file, int line)
642 if (map->system_map) {
643 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
645 sx_downgrade_(&map->lock, file, line);
651 * Returns a non-zero value if the caller holds a write (exclusive) lock
652 * on the specified map and the value "0" otherwise.
655 vm_map_locked(vm_map_t map)
659 return (mtx_owned(&map->system_mtx));
661 return (sx_xlocked(&map->lock));
666 _vm_map_assert_locked(vm_map_t map, const char *file, int line)
670 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
672 sx_assert_(&map->lock, SA_XLOCKED, file, line);
675 #define VM_MAP_ASSERT_LOCKED(map) \
676 _vm_map_assert_locked(map, LOCK_FILE, LOCK_LINE)
678 #define VM_MAP_ASSERT_LOCKED(map)
682 * _vm_map_unlock_and_wait:
684 * Atomically releases the lock on the specified map and puts the calling
685 * thread to sleep. The calling thread will remain asleep until either
686 * vm_map_wakeup() is performed on the map or the specified timeout is
689 * WARNING! This function does not perform deferred deallocations of
690 * objects and map entries. Therefore, the calling thread is expected to
691 * reacquire the map lock after reawakening and later perform an ordinary
692 * unlock operation, such as vm_map_unlock(), before completing its
693 * operation on the map.
696 _vm_map_unlock_and_wait(vm_map_t map, int timo, const char *file, int line)
699 mtx_lock(&map_sleep_mtx);
701 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
703 sx_xunlock_(&map->lock, file, line);
704 return (msleep(&map->root, &map_sleep_mtx, PDROP | PVM, "vmmaps",
711 * Awaken any threads that have slept on the map using
712 * vm_map_unlock_and_wait().
715 vm_map_wakeup(vm_map_t map)
719 * Acquire and release map_sleep_mtx to prevent a wakeup()
720 * from being performed (and lost) between the map unlock
721 * and the msleep() in _vm_map_unlock_and_wait().
723 mtx_lock(&map_sleep_mtx);
724 mtx_unlock(&map_sleep_mtx);
729 vm_map_busy(vm_map_t map)
732 VM_MAP_ASSERT_LOCKED(map);
737 vm_map_unbusy(vm_map_t map)
740 VM_MAP_ASSERT_LOCKED(map);
741 KASSERT(map->busy, ("vm_map_unbusy: not busy"));
742 if (--map->busy == 0 && (map->flags & MAP_BUSY_WAKEUP)) {
743 vm_map_modflags(map, 0, MAP_BUSY_WAKEUP);
749 vm_map_wait_busy(vm_map_t map)
752 VM_MAP_ASSERT_LOCKED(map);
754 vm_map_modflags(map, MAP_BUSY_WAKEUP, 0);
756 msleep(&map->busy, &map->system_mtx, 0, "mbusy", 0);
758 sx_sleep(&map->busy, &map->lock, 0, "mbusy", 0);
764 vmspace_resident_count(struct vmspace *vmspace)
766 return pmap_resident_count(vmspace_pmap(vmspace));
772 * Creates and returns a new empty VM map with
773 * the given physical map structure, and having
774 * the given lower and upper address bounds.
777 vm_map_create(pmap_t pmap, vm_offset_t min, vm_offset_t max)
781 result = uma_zalloc(mapzone, M_WAITOK);
782 CTR1(KTR_VM, "vm_map_create: %p", result);
783 _vm_map_init(result, pmap, min, max);
788 * Initialize an existing vm_map structure
789 * such as that in the vmspace structure.
792 _vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max)
795 map->header.next = map->header.prev = &map->header;
796 map->needs_wakeup = FALSE;
799 map->min_offset = min;
800 map->max_offset = max;
808 vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max)
811 _vm_map_init(map, pmap, min, max);
812 mtx_init(&map->system_mtx, "system map", NULL, MTX_DEF | MTX_DUPOK);
813 sx_init(&map->lock, "user map");
817 * vm_map_entry_dispose: [ internal use only ]
819 * Inverse of vm_map_entry_create.
822 vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry)
824 uma_zfree(map->system_map ? kmapentzone : mapentzone, entry);
828 * vm_map_entry_create: [ internal use only ]
830 * Allocates a VM map entry for insertion.
831 * No entry fields are filled in.
833 static vm_map_entry_t
834 vm_map_entry_create(vm_map_t map)
836 vm_map_entry_t new_entry;
839 new_entry = uma_zalloc(kmapentzone, M_NOWAIT);
841 new_entry = uma_zalloc(mapentzone, M_WAITOK);
842 if (new_entry == NULL)
843 panic("vm_map_entry_create: kernel resources exhausted");
848 * vm_map_entry_set_behavior:
850 * Set the expected access behavior, either normal, random, or
854 vm_map_entry_set_behavior(vm_map_entry_t entry, u_char behavior)
856 entry->eflags = (entry->eflags & ~MAP_ENTRY_BEHAV_MASK) |
857 (behavior & MAP_ENTRY_BEHAV_MASK);
861 * vm_map_entry_set_max_free:
863 * Set the max_free field in a vm_map_entry.
866 vm_map_entry_set_max_free(vm_map_entry_t entry)
869 entry->max_free = entry->adj_free;
870 if (entry->left != NULL && entry->left->max_free > entry->max_free)
871 entry->max_free = entry->left->max_free;
872 if (entry->right != NULL && entry->right->max_free > entry->max_free)
873 entry->max_free = entry->right->max_free;
877 * vm_map_entry_splay:
879 * The Sleator and Tarjan top-down splay algorithm with the
880 * following variation. Max_free must be computed bottom-up, so
881 * on the downward pass, maintain the left and right spines in
882 * reverse order. Then, make a second pass up each side to fix
883 * the pointers and compute max_free. The time bound is O(log n)
886 * The new root is the vm_map_entry containing "addr", or else an
887 * adjacent entry (lower or higher) if addr is not in the tree.
889 * The map must be locked, and leaves it so.
891 * Returns: the new root.
893 static vm_map_entry_t
894 vm_map_entry_splay(vm_offset_t addr, vm_map_entry_t root)
896 vm_map_entry_t llist, rlist;
897 vm_map_entry_t ltree, rtree;
900 /* Special case of empty tree. */
905 * Pass One: Splay down the tree until we find addr or a NULL
906 * pointer where addr would go. llist and rlist are the two
907 * sides in reverse order (bottom-up), with llist linked by
908 * the right pointer and rlist linked by the left pointer in
909 * the vm_map_entry. Wait until Pass Two to set max_free on
915 /* root is never NULL in here. */
916 if (addr < root->start) {
920 if (addr < y->start && y->left != NULL) {
921 /* Rotate right and put y on rlist. */
922 root->left = y->right;
924 vm_map_entry_set_max_free(root);
929 /* Put root on rlist. */
934 } else if (addr >= root->end) {
938 if (addr >= y->end && y->right != NULL) {
939 /* Rotate left and put y on llist. */
940 root->right = y->left;
942 vm_map_entry_set_max_free(root);
947 /* Put root on llist. */
957 * Pass Two: Walk back up the two spines, flip the pointers
958 * and set max_free. The subtrees of the root go at the
959 * bottom of llist and rlist.
962 while (llist != NULL) {
964 llist->right = ltree;
965 vm_map_entry_set_max_free(llist);
970 while (rlist != NULL) {
973 vm_map_entry_set_max_free(rlist);
979 * Final assembly: add ltree and rtree as subtrees of root.
983 vm_map_entry_set_max_free(root);
989 * vm_map_entry_{un,}link:
991 * Insert/remove entries from maps.
994 vm_map_entry_link(vm_map_t map,
995 vm_map_entry_t after_where,
996 vm_map_entry_t entry)
1000 "vm_map_entry_link: map %p, nentries %d, entry %p, after %p", map,
1001 map->nentries, entry, after_where);
1002 VM_MAP_ASSERT_LOCKED(map);
1003 KASSERT(after_where == &map->header ||
1004 after_where->end <= entry->start,
1005 ("vm_map_entry_link: prev end %jx new start %jx overlap",
1006 (uintmax_t)after_where->end, (uintmax_t)entry->start));
1007 KASSERT(after_where->next == &map->header ||
1008 entry->end <= after_where->next->start,
1009 ("vm_map_entry_link: new end %jx next start %jx overlap",
1010 (uintmax_t)entry->end, (uintmax_t)after_where->next->start));
1013 entry->prev = after_where;
1014 entry->next = after_where->next;
1015 entry->next->prev = entry;
1016 after_where->next = entry;
1018 if (after_where != &map->header) {
1019 if (after_where != map->root)
1020 vm_map_entry_splay(after_where->start, map->root);
1021 entry->right = after_where->right;
1022 entry->left = after_where;
1023 after_where->right = NULL;
1024 after_where->adj_free = entry->start - after_where->end;
1025 vm_map_entry_set_max_free(after_where);
1027 entry->right = map->root;
1030 entry->adj_free = (entry->next == &map->header ? map->max_offset :
1031 entry->next->start) - entry->end;
1032 vm_map_entry_set_max_free(entry);
1037 vm_map_entry_unlink(vm_map_t map,
1038 vm_map_entry_t entry)
1040 vm_map_entry_t next, prev, root;
1042 VM_MAP_ASSERT_LOCKED(map);
1043 if (entry != map->root)
1044 vm_map_entry_splay(entry->start, map->root);
1045 if (entry->left == NULL)
1046 root = entry->right;
1048 root = vm_map_entry_splay(entry->start, entry->left);
1049 root->right = entry->right;
1050 root->adj_free = (entry->next == &map->header ? map->max_offset :
1051 entry->next->start) - root->end;
1052 vm_map_entry_set_max_free(root);
1061 CTR3(KTR_VM, "vm_map_entry_unlink: map %p, nentries %d, entry %p", map,
1062 map->nentries, entry);
1066 * vm_map_entry_resize_free:
1068 * Recompute the amount of free space following a vm_map_entry
1069 * and propagate that value up the tree. Call this function after
1070 * resizing a map entry in-place, that is, without a call to
1071 * vm_map_entry_link() or _unlink().
1073 * The map must be locked, and leaves it so.
1076 vm_map_entry_resize_free(vm_map_t map, vm_map_entry_t entry)
1080 * Using splay trees without parent pointers, propagating
1081 * max_free up the tree is done by moving the entry to the
1082 * root and making the change there.
1084 if (entry != map->root)
1085 map->root = vm_map_entry_splay(entry->start, map->root);
1087 entry->adj_free = (entry->next == &map->header ? map->max_offset :
1088 entry->next->start) - entry->end;
1089 vm_map_entry_set_max_free(entry);
1093 * vm_map_lookup_entry: [ internal use only ]
1095 * Finds the map entry containing (or
1096 * immediately preceding) the specified address
1097 * in the given map; the entry is returned
1098 * in the "entry" parameter. The boolean
1099 * result indicates whether the address is
1100 * actually contained in the map.
1103 vm_map_lookup_entry(
1105 vm_offset_t address,
1106 vm_map_entry_t *entry) /* OUT */
1112 * If the map is empty, then the map entry immediately preceding
1113 * "address" is the map's header.
1117 *entry = &map->header;
1118 else if (address >= cur->start && cur->end > address) {
1121 } else if ((locked = vm_map_locked(map)) ||
1122 sx_try_upgrade(&map->lock)) {
1124 * Splay requires a write lock on the map. However, it only
1125 * restructures the binary search tree; it does not otherwise
1126 * change the map. Thus, the map's timestamp need not change
1127 * on a temporary upgrade.
1129 map->root = cur = vm_map_entry_splay(address, cur);
1131 sx_downgrade(&map->lock);
1134 * If "address" is contained within a map entry, the new root
1135 * is that map entry. Otherwise, the new root is a map entry
1136 * immediately before or after "address".
1138 if (address >= cur->start) {
1140 if (cur->end > address)
1146 * Since the map is only locked for read access, perform a
1147 * standard binary search tree lookup for "address".
1150 if (address < cur->start) {
1151 if (cur->left == NULL) {
1156 } else if (cur->end > address) {
1160 if (cur->right == NULL) {
1173 * Inserts the given whole VM object into the target
1174 * map at the specified address range. The object's
1175 * size should match that of the address range.
1177 * Requires that the map be locked, and leaves it so.
1179 * If object is non-NULL, ref count must be bumped by caller
1180 * prior to making call to account for the new entry.
1183 vm_map_insert(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1184 vm_offset_t start, vm_offset_t end, vm_prot_t prot, vm_prot_t max, int cow)
1186 vm_map_entry_t new_entry, prev_entry, temp_entry;
1188 vm_eflags_t protoeflags;
1189 vm_inherit_t inheritance;
1191 VM_MAP_ASSERT_LOCKED(map);
1192 KASSERT(object != kernel_object ||
1193 (cow & MAP_COPY_ON_WRITE) == 0,
1194 ("vm_map_insert: kernel object and COW"));
1195 KASSERT(object == NULL || (cow & MAP_NOFAULT) == 0,
1196 ("vm_map_insert: paradoxical MAP_NOFAULT request"));
1197 KASSERT((prot & ~max) == 0,
1198 ("prot %#x is not subset of max_prot %#x", prot, max));
1201 * Check that the start and end points are not bogus.
1203 if (start < map->min_offset || end > map->max_offset || start >= end)
1204 return (KERN_INVALID_ADDRESS);
1207 * Find the entry prior to the proposed starting address; if it's part
1208 * of an existing entry, this range is bogus.
1210 if (vm_map_lookup_entry(map, start, &temp_entry))
1211 return (KERN_NO_SPACE);
1213 prev_entry = temp_entry;
1216 * Assert that the next entry doesn't overlap the end point.
1218 if (prev_entry->next != &map->header && prev_entry->next->start < end)
1219 return (KERN_NO_SPACE);
1221 if ((cow & MAP_CREATE_GUARD) != 0 && (object != NULL ||
1222 max != VM_PROT_NONE))
1223 return (KERN_INVALID_ARGUMENT);
1226 if (cow & MAP_COPY_ON_WRITE)
1227 protoeflags |= MAP_ENTRY_COW | MAP_ENTRY_NEEDS_COPY;
1228 if (cow & MAP_NOFAULT)
1229 protoeflags |= MAP_ENTRY_NOFAULT;
1230 if (cow & MAP_DISABLE_SYNCER)
1231 protoeflags |= MAP_ENTRY_NOSYNC;
1232 if (cow & MAP_DISABLE_COREDUMP)
1233 protoeflags |= MAP_ENTRY_NOCOREDUMP;
1234 if (cow & MAP_STACK_GROWS_DOWN)
1235 protoeflags |= MAP_ENTRY_GROWS_DOWN;
1236 if (cow & MAP_STACK_GROWS_UP)
1237 protoeflags |= MAP_ENTRY_GROWS_UP;
1238 if (cow & MAP_VN_WRITECOUNT)
1239 protoeflags |= MAP_ENTRY_VN_WRITECNT;
1240 if ((cow & MAP_CREATE_GUARD) != 0)
1241 protoeflags |= MAP_ENTRY_GUARD;
1242 if ((cow & MAP_CREATE_STACK_GAP_DN) != 0)
1243 protoeflags |= MAP_ENTRY_STACK_GAP_DN;
1244 if ((cow & MAP_CREATE_STACK_GAP_UP) != 0)
1245 protoeflags |= MAP_ENTRY_STACK_GAP_UP;
1246 if (cow & MAP_INHERIT_SHARE)
1247 inheritance = VM_INHERIT_SHARE;
1249 inheritance = VM_INHERIT_DEFAULT;
1252 if ((cow & (MAP_ACC_NO_CHARGE | MAP_NOFAULT | MAP_CREATE_GUARD)) != 0)
1254 if ((cow & MAP_ACC_CHARGED) || ((prot & VM_PROT_WRITE) &&
1255 ((protoeflags & MAP_ENTRY_NEEDS_COPY) || object == NULL))) {
1256 if (!(cow & MAP_ACC_CHARGED) && !swap_reserve(end - start))
1257 return (KERN_RESOURCE_SHORTAGE);
1258 KASSERT(object == NULL ||
1259 (protoeflags & MAP_ENTRY_NEEDS_COPY) != 0 ||
1260 object->cred == NULL,
1261 ("overcommit: vm_map_insert o %p", object));
1262 cred = curthread->td_ucred;
1266 /* Expand the kernel pmap, if necessary. */
1267 if (map == kernel_map && end > kernel_vm_end)
1268 pmap_growkernel(end);
1269 if (object != NULL) {
1271 * OBJ_ONEMAPPING must be cleared unless this mapping
1272 * is trivially proven to be the only mapping for any
1273 * of the object's pages. (Object granularity
1274 * reference counting is insufficient to recognize
1275 * aliases with precision.)
1277 VM_OBJECT_WLOCK(object);
1278 if (object->ref_count > 1 || object->shadow_count != 0)
1279 vm_object_clear_flag(object, OBJ_ONEMAPPING);
1280 VM_OBJECT_WUNLOCK(object);
1281 } else if (prev_entry != &map->header &&
1282 prev_entry->eflags == protoeflags &&
1283 (cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 &&
1284 prev_entry->end == start && prev_entry->wired_count == 0 &&
1285 (prev_entry->cred == cred ||
1286 (prev_entry->object.vm_object != NULL &&
1287 prev_entry->object.vm_object->cred == cred)) &&
1288 vm_object_coalesce(prev_entry->object.vm_object,
1290 (vm_size_t)(prev_entry->end - prev_entry->start),
1291 (vm_size_t)(end - prev_entry->end), cred != NULL &&
1292 (protoeflags & MAP_ENTRY_NEEDS_COPY) == 0)) {
1294 * We were able to extend the object. Determine if we
1295 * can extend the previous map entry to include the
1296 * new range as well.
1298 if (prev_entry->inheritance == inheritance &&
1299 prev_entry->protection == prot &&
1300 prev_entry->max_protection == max) {
1301 if ((prev_entry->eflags & MAP_ENTRY_GUARD) == 0)
1302 map->size += end - prev_entry->end;
1303 prev_entry->end = end;
1304 vm_map_entry_resize_free(map, prev_entry);
1305 vm_map_simplify_entry(map, prev_entry);
1306 return (KERN_SUCCESS);
1310 * If we can extend the object but cannot extend the
1311 * map entry, we have to create a new map entry. We
1312 * must bump the ref count on the extended object to
1313 * account for it. object may be NULL.
1315 object = prev_entry->object.vm_object;
1316 offset = prev_entry->offset +
1317 (prev_entry->end - prev_entry->start);
1318 vm_object_reference(object);
1319 if (cred != NULL && object != NULL && object->cred != NULL &&
1320 !(prev_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
1321 /* Object already accounts for this uid. */
1329 * Create a new entry
1331 new_entry = vm_map_entry_create(map);
1332 new_entry->start = start;
1333 new_entry->end = end;
1334 new_entry->cred = NULL;
1336 new_entry->eflags = protoeflags;
1337 new_entry->object.vm_object = object;
1338 new_entry->offset = offset;
1340 new_entry->inheritance = inheritance;
1341 new_entry->protection = prot;
1342 new_entry->max_protection = max;
1343 new_entry->wired_count = 0;
1344 new_entry->wiring_thread = NULL;
1345 new_entry->read_ahead = VM_FAULT_READ_AHEAD_INIT;
1346 new_entry->next_read = start;
1348 KASSERT(cred == NULL || !ENTRY_CHARGED(new_entry),
1349 ("overcommit: vm_map_insert leaks vm_map %p", new_entry));
1350 new_entry->cred = cred;
1353 * Insert the new entry into the list
1355 vm_map_entry_link(map, prev_entry, new_entry);
1356 if ((new_entry->eflags & MAP_ENTRY_GUARD) == 0)
1357 map->size += new_entry->end - new_entry->start;
1360 * Try to coalesce the new entry with both the previous and next
1361 * entries in the list. Previously, we only attempted to coalesce
1362 * with the previous entry when object is NULL. Here, we handle the
1363 * other cases, which are less common.
1365 vm_map_simplify_entry(map, new_entry);
1367 if ((cow & (MAP_PREFAULT | MAP_PREFAULT_PARTIAL)) != 0) {
1368 vm_map_pmap_enter(map, start, prot, object, OFF_TO_IDX(offset),
1369 end - start, cow & MAP_PREFAULT_PARTIAL);
1372 return (KERN_SUCCESS);
1378 * Find the first fit (lowest VM address) for "length" free bytes
1379 * beginning at address >= start in the given map.
1381 * In a vm_map_entry, "adj_free" is the amount of free space
1382 * adjacent (higher address) to this entry, and "max_free" is the
1383 * maximum amount of contiguous free space in its subtree. This
1384 * allows finding a free region in one path down the tree, so
1385 * O(log n) amortized with splay trees.
1387 * The map must be locked, and leaves it so.
1389 * Returns: 0 on success, and starting address in *addr,
1390 * 1 if insufficient space.
1393 vm_map_findspace(vm_map_t map, vm_offset_t start, vm_size_t length,
1394 vm_offset_t *addr) /* OUT */
1396 vm_map_entry_t entry;
1400 * Request must fit within min/max VM address and must avoid
1403 if (start < map->min_offset)
1404 start = map->min_offset;
1405 if (start + length > map->max_offset || start + length < start)
1408 /* Empty tree means wide open address space. */
1409 if (map->root == NULL) {
1415 * After splay, if start comes before root node, then there
1416 * must be a gap from start to the root.
1418 map->root = vm_map_entry_splay(start, map->root);
1419 if (start + length <= map->root->start) {
1425 * Root is the last node that might begin its gap before
1426 * start, and this is the last comparison where address
1427 * wrap might be a problem.
1429 st = (start > map->root->end) ? start : map->root->end;
1430 if (length <= map->root->end + map->root->adj_free - st) {
1435 /* With max_free, can immediately tell if no solution. */
1436 entry = map->root->right;
1437 if (entry == NULL || length > entry->max_free)
1441 * Search the right subtree in the order: left subtree, root,
1442 * right subtree (first fit). The previous splay implies that
1443 * all regions in the right subtree have addresses > start.
1445 while (entry != NULL) {
1446 if (entry->left != NULL && entry->left->max_free >= length)
1447 entry = entry->left;
1448 else if (entry->adj_free >= length) {
1452 entry = entry->right;
1455 /* Can't get here, so panic if we do. */
1456 panic("vm_map_findspace: max_free corrupt");
1460 vm_map_fixed(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1461 vm_offset_t start, vm_size_t length, vm_prot_t prot,
1462 vm_prot_t max, int cow)
1467 end = start + length;
1468 KASSERT((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 ||
1470 ("vm_map_fixed: non-NULL backing object for stack"));
1472 VM_MAP_RANGE_CHECK(map, start, end);
1473 if ((cow & MAP_CHECK_EXCL) == 0)
1474 vm_map_delete(map, start, end);
1475 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) != 0) {
1476 result = vm_map_stack_locked(map, start, length, sgrowsiz,
1479 result = vm_map_insert(map, object, offset, start, end,
1487 * vm_map_find finds an unallocated region in the target address
1488 * map with the given length. The search is defined to be
1489 * first-fit from the specified address; the region found is
1490 * returned in the same parameter.
1492 * If object is non-NULL, ref count must be bumped by caller
1493 * prior to making call to account for the new entry.
1496 vm_map_find(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1497 vm_offset_t *addr, /* IN/OUT */
1498 vm_size_t length, vm_offset_t max_addr, int find_space,
1499 vm_prot_t prot, vm_prot_t max, int cow)
1501 vm_offset_t alignment, initial_addr, start;
1504 KASSERT((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 ||
1506 ("vm_map_find: non-NULL backing object for stack"));
1507 if (find_space == VMFS_OPTIMAL_SPACE && (object == NULL ||
1508 (object->flags & OBJ_COLORED) == 0))
1509 find_space = VMFS_ANY_SPACE;
1510 if (find_space >> 8 != 0) {
1511 KASSERT((find_space & 0xff) == 0, ("bad VMFS flags"));
1512 alignment = (vm_offset_t)1 << (find_space >> 8);
1515 initial_addr = *addr;
1518 start = initial_addr;
1520 if (find_space != VMFS_NO_SPACE) {
1521 if (vm_map_findspace(map, start, length, addr) ||
1522 (max_addr != 0 && *addr + length > max_addr)) {
1523 if (find_space == VMFS_OPTIMAL_SPACE) {
1524 find_space = VMFS_ANY_SPACE;
1528 return (KERN_NO_SPACE);
1530 switch (find_space) {
1531 case VMFS_SUPER_SPACE:
1532 case VMFS_OPTIMAL_SPACE:
1533 pmap_align_superpage(object, offset, addr,
1536 case VMFS_ANY_SPACE:
1539 if ((*addr & (alignment - 1)) != 0) {
1540 *addr &= ~(alignment - 1);
1548 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) != 0) {
1549 result = vm_map_stack_locked(map, start, length,
1550 sgrowsiz, prot, max, cow);
1552 result = vm_map_insert(map, object, offset, start,
1553 start + length, prot, max, cow);
1555 } while (result == KERN_NO_SPACE && find_space != VMFS_NO_SPACE &&
1556 find_space != VMFS_ANY_SPACE);
1562 vm_map_find_min(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1563 vm_offset_t *addr, vm_size_t length, vm_offset_t min_addr,
1564 vm_offset_t max_addr, int find_space, vm_prot_t prot, vm_prot_t max,
1572 rv = vm_map_find(map, object, offset, addr, length, max_addr,
1573 find_space, prot, max, cow);
1574 if (rv == KERN_SUCCESS || min_addr >= hint)
1576 *addr = hint = min_addr;
1581 * vm_map_simplify_entry:
1583 * Simplify the given map entry by merging with either neighbor. This
1584 * routine also has the ability to merge with both neighbors.
1586 * The map must be locked.
1588 * This routine guarantees that the passed entry remains valid (though
1589 * possibly extended). When merging, this routine may delete one or
1593 vm_map_simplify_entry(vm_map_t map, vm_map_entry_t entry)
1595 vm_map_entry_t next, prev;
1596 vm_size_t prevsize, esize;
1598 if ((entry->eflags & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP |
1599 MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_IS_SUB_MAP)) != 0)
1603 if (prev != &map->header) {
1604 prevsize = prev->end - prev->start;
1605 if ( (prev->end == entry->start) &&
1606 (prev->object.vm_object == entry->object.vm_object) &&
1607 (!prev->object.vm_object ||
1608 (prev->offset + prevsize == entry->offset)) &&
1609 (prev->eflags == entry->eflags) &&
1610 (prev->protection == entry->protection) &&
1611 (prev->max_protection == entry->max_protection) &&
1612 (prev->inheritance == entry->inheritance) &&
1613 (prev->wired_count == entry->wired_count) &&
1614 (prev->cred == entry->cred)) {
1615 vm_map_entry_unlink(map, prev);
1616 entry->start = prev->start;
1617 entry->offset = prev->offset;
1618 if (entry->prev != &map->header)
1619 vm_map_entry_resize_free(map, entry->prev);
1622 * If the backing object is a vnode object,
1623 * vm_object_deallocate() calls vrele().
1624 * However, vrele() does not lock the vnode
1625 * because the vnode has additional
1626 * references. Thus, the map lock can be kept
1627 * without causing a lock-order reversal with
1630 * Since we count the number of virtual page
1631 * mappings in object->un_pager.vnp.writemappings,
1632 * the writemappings value should not be adjusted
1633 * when the entry is disposed of.
1635 if (prev->object.vm_object)
1636 vm_object_deallocate(prev->object.vm_object);
1637 if (prev->cred != NULL)
1639 vm_map_entry_dispose(map, prev);
1644 if (next != &map->header) {
1645 esize = entry->end - entry->start;
1646 if ((entry->end == next->start) &&
1647 (next->object.vm_object == entry->object.vm_object) &&
1648 (!entry->object.vm_object ||
1649 (entry->offset + esize == next->offset)) &&
1650 (next->eflags == entry->eflags) &&
1651 (next->protection == entry->protection) &&
1652 (next->max_protection == entry->max_protection) &&
1653 (next->inheritance == entry->inheritance) &&
1654 (next->wired_count == entry->wired_count) &&
1655 (next->cred == entry->cred)) {
1656 vm_map_entry_unlink(map, next);
1657 entry->end = next->end;
1658 vm_map_entry_resize_free(map, entry);
1661 * See comment above.
1663 if (next->object.vm_object)
1664 vm_object_deallocate(next->object.vm_object);
1665 if (next->cred != NULL)
1667 vm_map_entry_dispose(map, next);
1672 * vm_map_clip_start: [ internal use only ]
1674 * Asserts that the given entry begins at or after
1675 * the specified address; if necessary,
1676 * it splits the entry into two.
1678 #define vm_map_clip_start(map, entry, startaddr) \
1680 if (startaddr > entry->start) \
1681 _vm_map_clip_start(map, entry, startaddr); \
1685 * This routine is called only when it is known that
1686 * the entry must be split.
1689 _vm_map_clip_start(vm_map_t map, vm_map_entry_t entry, vm_offset_t start)
1691 vm_map_entry_t new_entry;
1693 VM_MAP_ASSERT_LOCKED(map);
1694 KASSERT(entry->end > start && entry->start < start,
1695 ("_vm_map_clip_start: invalid clip of entry %p", entry));
1698 * Split off the front portion -- note that we must insert the new
1699 * entry BEFORE this one, so that this entry has the specified
1702 vm_map_simplify_entry(map, entry);
1705 * If there is no object backing this entry, we might as well create
1706 * one now. If we defer it, an object can get created after the map
1707 * is clipped, and individual objects will be created for the split-up
1708 * map. This is a bit of a hack, but is also about the best place to
1709 * put this improvement.
1711 if (entry->object.vm_object == NULL && !map->system_map &&
1712 (entry->eflags & MAP_ENTRY_GUARD) == 0) {
1714 object = vm_object_allocate(OBJT_DEFAULT,
1715 atop(entry->end - entry->start));
1716 entry->object.vm_object = object;
1718 if (entry->cred != NULL) {
1719 object->cred = entry->cred;
1720 object->charge = entry->end - entry->start;
1723 } else if (entry->object.vm_object != NULL &&
1724 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) &&
1725 entry->cred != NULL) {
1726 VM_OBJECT_WLOCK(entry->object.vm_object);
1727 KASSERT(entry->object.vm_object->cred == NULL,
1728 ("OVERCOMMIT: vm_entry_clip_start: both cred e %p", entry));
1729 entry->object.vm_object->cred = entry->cred;
1730 entry->object.vm_object->charge = entry->end - entry->start;
1731 VM_OBJECT_WUNLOCK(entry->object.vm_object);
1735 new_entry = vm_map_entry_create(map);
1736 *new_entry = *entry;
1738 new_entry->end = start;
1739 entry->offset += (start - entry->start);
1740 entry->start = start;
1741 if (new_entry->cred != NULL)
1742 crhold(entry->cred);
1744 vm_map_entry_link(map, entry->prev, new_entry);
1746 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1747 vm_object_reference(new_entry->object.vm_object);
1749 * The object->un_pager.vnp.writemappings for the
1750 * object of MAP_ENTRY_VN_WRITECNT type entry shall be
1751 * kept as is here. The virtual pages are
1752 * re-distributed among the clipped entries, so the sum is
1759 * vm_map_clip_end: [ internal use only ]
1761 * Asserts that the given entry ends at or before
1762 * the specified address; if necessary,
1763 * it splits the entry into two.
1765 #define vm_map_clip_end(map, entry, endaddr) \
1767 if ((endaddr) < (entry->end)) \
1768 _vm_map_clip_end((map), (entry), (endaddr)); \
1772 * This routine is called only when it is known that
1773 * the entry must be split.
1776 _vm_map_clip_end(vm_map_t map, vm_map_entry_t entry, vm_offset_t end)
1778 vm_map_entry_t new_entry;
1780 VM_MAP_ASSERT_LOCKED(map);
1781 KASSERT(entry->start < end && entry->end > end,
1782 ("_vm_map_clip_end: invalid clip of entry %p", entry));
1785 * If there is no object backing this entry, we might as well create
1786 * one now. If we defer it, an object can get created after the map
1787 * is clipped, and individual objects will be created for the split-up
1788 * map. This is a bit of a hack, but is also about the best place to
1789 * put this improvement.
1791 if (entry->object.vm_object == NULL && !map->system_map &&
1792 (entry->eflags & MAP_ENTRY_GUARD) == 0) {
1794 object = vm_object_allocate(OBJT_DEFAULT,
1795 atop(entry->end - entry->start));
1796 entry->object.vm_object = object;
1798 if (entry->cred != NULL) {
1799 object->cred = entry->cred;
1800 object->charge = entry->end - entry->start;
1803 } else if (entry->object.vm_object != NULL &&
1804 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) &&
1805 entry->cred != NULL) {
1806 VM_OBJECT_WLOCK(entry->object.vm_object);
1807 KASSERT(entry->object.vm_object->cred == NULL,
1808 ("OVERCOMMIT: vm_entry_clip_end: both cred e %p", entry));
1809 entry->object.vm_object->cred = entry->cred;
1810 entry->object.vm_object->charge = entry->end - entry->start;
1811 VM_OBJECT_WUNLOCK(entry->object.vm_object);
1816 * Create a new entry and insert it AFTER the specified entry
1818 new_entry = vm_map_entry_create(map);
1819 *new_entry = *entry;
1821 new_entry->start = entry->end = end;
1822 new_entry->offset += (end - entry->start);
1823 if (new_entry->cred != NULL)
1824 crhold(entry->cred);
1826 vm_map_entry_link(map, entry, new_entry);
1828 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1829 vm_object_reference(new_entry->object.vm_object);
1834 * vm_map_submap: [ kernel use only ]
1836 * Mark the given range as handled by a subordinate map.
1838 * This range must have been created with vm_map_find,
1839 * and no other operations may have been performed on this
1840 * range prior to calling vm_map_submap.
1842 * Only a limited number of operations can be performed
1843 * within this rage after calling vm_map_submap:
1845 * [Don't try vm_map_copy!]
1847 * To remove a submapping, one must first remove the
1848 * range from the superior map, and then destroy the
1849 * submap (if desired). [Better yet, don't try it.]
1858 vm_map_entry_t entry;
1859 int result = KERN_INVALID_ARGUMENT;
1863 VM_MAP_RANGE_CHECK(map, start, end);
1865 if (vm_map_lookup_entry(map, start, &entry)) {
1866 vm_map_clip_start(map, entry, start);
1868 entry = entry->next;
1870 vm_map_clip_end(map, entry, end);
1872 if ((entry->start == start) && (entry->end == end) &&
1873 ((entry->eflags & MAP_ENTRY_COW) == 0) &&
1874 (entry->object.vm_object == NULL)) {
1875 entry->object.sub_map = submap;
1876 entry->eflags |= MAP_ENTRY_IS_SUB_MAP;
1877 result = KERN_SUCCESS;
1885 * The maximum number of pages to map if MAP_PREFAULT_PARTIAL is specified
1887 #define MAX_INIT_PT 96
1890 * vm_map_pmap_enter:
1892 * Preload the specified map's pmap with mappings to the specified
1893 * object's memory-resident pages. No further physical pages are
1894 * allocated, and no further virtual pages are retrieved from secondary
1895 * storage. If the specified flags include MAP_PREFAULT_PARTIAL, then a
1896 * limited number of page mappings are created at the low-end of the
1897 * specified address range. (For this purpose, a superpage mapping
1898 * counts as one page mapping.) Otherwise, all resident pages within
1899 * the specified address range are mapped.
1902 vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot,
1903 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags)
1906 vm_page_t p, p_start;
1907 vm_pindex_t mask, psize, threshold, tmpidx;
1909 if ((prot & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0 || object == NULL)
1911 VM_OBJECT_RLOCK(object);
1912 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) {
1913 VM_OBJECT_RUNLOCK(object);
1914 VM_OBJECT_WLOCK(object);
1915 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) {
1916 pmap_object_init_pt(map->pmap, addr, object, pindex,
1918 VM_OBJECT_WUNLOCK(object);
1921 VM_OBJECT_LOCK_DOWNGRADE(object);
1925 if (psize + pindex > object->size) {
1926 if (object->size < pindex) {
1927 VM_OBJECT_RUNLOCK(object);
1930 psize = object->size - pindex;
1935 threshold = MAX_INIT_PT;
1937 p = vm_page_find_least(object, pindex);
1939 * Assert: the variable p is either (1) the page with the
1940 * least pindex greater than or equal to the parameter pindex
1944 p != NULL && (tmpidx = p->pindex - pindex) < psize;
1945 p = TAILQ_NEXT(p, listq)) {
1947 * don't allow an madvise to blow away our really
1948 * free pages allocating pv entries.
1950 if (((flags & MAP_PREFAULT_MADVISE) != 0 &&
1951 vm_cnt.v_free_count < vm_cnt.v_free_reserved) ||
1952 ((flags & MAP_PREFAULT_PARTIAL) != 0 &&
1953 tmpidx >= threshold)) {
1957 if (p->valid == VM_PAGE_BITS_ALL) {
1958 if (p_start == NULL) {
1959 start = addr + ptoa(tmpidx);
1962 /* Jump ahead if a superpage mapping is possible. */
1963 if (p->psind > 0 && ((addr + ptoa(tmpidx)) &
1964 (pagesizes[p->psind] - 1)) == 0) {
1965 mask = atop(pagesizes[p->psind]) - 1;
1966 if (tmpidx + mask < psize &&
1967 vm_page_ps_test(p, PS_ALL_VALID, NULL)) {
1972 } else if (p_start != NULL) {
1973 pmap_enter_object(map->pmap, start, addr +
1974 ptoa(tmpidx), p_start, prot);
1978 if (p_start != NULL)
1979 pmap_enter_object(map->pmap, start, addr + ptoa(psize),
1981 VM_OBJECT_RUNLOCK(object);
1987 * Sets the protection of the specified address
1988 * region in the target map. If "set_max" is
1989 * specified, the maximum protection is to be set;
1990 * otherwise, only the current protection is affected.
1993 vm_map_protect(vm_map_t map, vm_offset_t start, vm_offset_t end,
1994 vm_prot_t new_prot, boolean_t set_max)
1996 vm_map_entry_t current, entry;
2002 return (KERN_SUCCESS);
2007 * Ensure that we are not concurrently wiring pages. vm_map_wire() may
2008 * need to fault pages into the map and will drop the map lock while
2009 * doing so, and the VM object may end up in an inconsistent state if we
2010 * update the protection on the map entry in between faults.
2012 vm_map_wait_busy(map);
2014 VM_MAP_RANGE_CHECK(map, start, end);
2016 if (vm_map_lookup_entry(map, start, &entry)) {
2017 vm_map_clip_start(map, entry, start);
2019 entry = entry->next;
2023 * Make a first pass to check for protection violations.
2025 for (current = entry; current != &map->header && current->start < end;
2026 current = current->next) {
2027 if ((current->eflags & MAP_ENTRY_GUARD) != 0)
2029 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
2031 return (KERN_INVALID_ARGUMENT);
2033 if ((new_prot & current->max_protection) != new_prot) {
2035 return (KERN_PROTECTION_FAILURE);
2040 * Do an accounting pass for private read-only mappings that
2041 * now will do cow due to allowed write (e.g. debugger sets
2042 * breakpoint on text segment)
2044 for (current = entry; current != &map->header && current->start < end;
2045 current = current->next) {
2047 vm_map_clip_end(map, current, end);
2050 ((new_prot & ~(current->protection)) & VM_PROT_WRITE) == 0 ||
2051 ENTRY_CHARGED(current) ||
2052 (current->eflags & MAP_ENTRY_GUARD) != 0) {
2056 cred = curthread->td_ucred;
2057 obj = current->object.vm_object;
2059 if (obj == NULL || (current->eflags & MAP_ENTRY_NEEDS_COPY)) {
2060 if (!swap_reserve(current->end - current->start)) {
2062 return (KERN_RESOURCE_SHORTAGE);
2065 current->cred = cred;
2069 VM_OBJECT_WLOCK(obj);
2070 if (obj->type != OBJT_DEFAULT && obj->type != OBJT_SWAP) {
2071 VM_OBJECT_WUNLOCK(obj);
2076 * Charge for the whole object allocation now, since
2077 * we cannot distinguish between non-charged and
2078 * charged clipped mapping of the same object later.
2080 KASSERT(obj->charge == 0,
2081 ("vm_map_protect: object %p overcharged (entry %p)",
2083 if (!swap_reserve(ptoa(obj->size))) {
2084 VM_OBJECT_WUNLOCK(obj);
2086 return (KERN_RESOURCE_SHORTAGE);
2091 obj->charge = ptoa(obj->size);
2092 VM_OBJECT_WUNLOCK(obj);
2096 * Go back and fix up protections. [Note that clipping is not
2097 * necessary the second time.]
2099 for (current = entry; current != &map->header && current->start < end;
2100 current = current->next) {
2101 if ((current->eflags & MAP_ENTRY_GUARD) != 0)
2104 old_prot = current->protection;
2107 current->protection =
2108 (current->max_protection = new_prot) &
2111 current->protection = new_prot;
2114 * For user wired map entries, the normal lazy evaluation of
2115 * write access upgrades through soft page faults is
2116 * undesirable. Instead, immediately copy any pages that are
2117 * copy-on-write and enable write access in the physical map.
2119 if ((current->eflags & MAP_ENTRY_USER_WIRED) != 0 &&
2120 (current->protection & VM_PROT_WRITE) != 0 &&
2121 (old_prot & VM_PROT_WRITE) == 0)
2122 vm_fault_copy_entry(map, map, current, current, NULL);
2125 * When restricting access, update the physical map. Worry
2126 * about copy-on-write here.
2128 if ((old_prot & ~current->protection) != 0) {
2129 #define MASK(entry) (((entry)->eflags & MAP_ENTRY_COW) ? ~VM_PROT_WRITE : \
2131 pmap_protect(map->pmap, current->start,
2133 current->protection & MASK(current));
2136 vm_map_simplify_entry(map, current);
2139 return (KERN_SUCCESS);
2145 * This routine traverses a processes map handling the madvise
2146 * system call. Advisories are classified as either those effecting
2147 * the vm_map_entry structure, or those effecting the underlying
2157 vm_map_entry_t current, entry;
2161 * Some madvise calls directly modify the vm_map_entry, in which case
2162 * we need to use an exclusive lock on the map and we need to perform
2163 * various clipping operations. Otherwise we only need a read-lock
2168 case MADV_SEQUENTIAL:
2175 return (KERN_SUCCESS);
2183 return (KERN_SUCCESS);
2184 vm_map_lock_read(map);
2187 return (KERN_INVALID_ARGUMENT);
2191 * Locate starting entry and clip if necessary.
2193 VM_MAP_RANGE_CHECK(map, start, end);
2195 if (vm_map_lookup_entry(map, start, &entry)) {
2197 vm_map_clip_start(map, entry, start);
2199 entry = entry->next;
2204 * madvise behaviors that are implemented in the vm_map_entry.
2206 * We clip the vm_map_entry so that behavioral changes are
2207 * limited to the specified address range.
2209 for (current = entry;
2210 (current != &map->header) && (current->start < end);
2211 current = current->next
2213 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
2216 vm_map_clip_end(map, current, end);
2220 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_NORMAL);
2222 case MADV_SEQUENTIAL:
2223 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_SEQUENTIAL);
2226 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_RANDOM);
2229 current->eflags |= MAP_ENTRY_NOSYNC;
2232 current->eflags &= ~MAP_ENTRY_NOSYNC;
2235 current->eflags |= MAP_ENTRY_NOCOREDUMP;
2238 current->eflags &= ~MAP_ENTRY_NOCOREDUMP;
2243 vm_map_simplify_entry(map, current);
2247 vm_pindex_t pstart, pend;
2250 * madvise behaviors that are implemented in the underlying
2253 * Since we don't clip the vm_map_entry, we have to clip
2254 * the vm_object pindex and count.
2256 for (current = entry;
2257 (current != &map->header) && (current->start < end);
2258 current = current->next
2260 vm_offset_t useEnd, useStart;
2262 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
2265 pstart = OFF_TO_IDX(current->offset);
2266 pend = pstart + atop(current->end - current->start);
2267 useStart = current->start;
2268 useEnd = current->end;
2270 if (current->start < start) {
2271 pstart += atop(start - current->start);
2274 if (current->end > end) {
2275 pend -= atop(current->end - end);
2283 * Perform the pmap_advise() before clearing
2284 * PGA_REFERENCED in vm_page_advise(). Otherwise, a
2285 * concurrent pmap operation, such as pmap_remove(),
2286 * could clear a reference in the pmap and set
2287 * PGA_REFERENCED on the page before the pmap_advise()
2288 * had completed. Consequently, the page would appear
2289 * referenced based upon an old reference that
2290 * occurred before this pmap_advise() ran.
2292 if (behav == MADV_DONTNEED || behav == MADV_FREE)
2293 pmap_advise(map->pmap, useStart, useEnd,
2296 vm_object_madvise(current->object.vm_object, pstart,
2300 * Pre-populate paging structures in the
2301 * WILLNEED case. For wired entries, the
2302 * paging structures are already populated.
2304 if (behav == MADV_WILLNEED &&
2305 current->wired_count == 0) {
2306 vm_map_pmap_enter(map,
2308 current->protection,
2309 current->object.vm_object,
2311 ptoa(pend - pstart),
2312 MAP_PREFAULT_MADVISE
2316 vm_map_unlock_read(map);
2325 * Sets the inheritance of the specified address
2326 * range in the target map. Inheritance
2327 * affects how the map will be shared with
2328 * child maps at the time of vmspace_fork.
2331 vm_map_inherit(vm_map_t map, vm_offset_t start, vm_offset_t end,
2332 vm_inherit_t new_inheritance)
2334 vm_map_entry_t entry;
2335 vm_map_entry_t temp_entry;
2337 switch (new_inheritance) {
2338 case VM_INHERIT_NONE:
2339 case VM_INHERIT_COPY:
2340 case VM_INHERIT_SHARE:
2341 case VM_INHERIT_ZERO:
2344 return (KERN_INVALID_ARGUMENT);
2347 return (KERN_SUCCESS);
2349 VM_MAP_RANGE_CHECK(map, start, end);
2350 if (vm_map_lookup_entry(map, start, &temp_entry)) {
2352 vm_map_clip_start(map, entry, start);
2354 entry = temp_entry->next;
2355 while ((entry != &map->header) && (entry->start < end)) {
2356 vm_map_clip_end(map, entry, end);
2357 if ((entry->eflags & MAP_ENTRY_GUARD) == 0 ||
2358 new_inheritance != VM_INHERIT_ZERO)
2359 entry->inheritance = new_inheritance;
2360 vm_map_simplify_entry(map, entry);
2361 entry = entry->next;
2364 return (KERN_SUCCESS);
2370 * Implements both kernel and user unwiring.
2373 vm_map_unwire(vm_map_t map, vm_offset_t start, vm_offset_t end,
2376 vm_map_entry_t entry, first_entry, tmp_entry;
2377 vm_offset_t saved_start;
2378 unsigned int last_timestamp;
2380 boolean_t need_wakeup, result, user_unwire;
2383 return (KERN_SUCCESS);
2384 user_unwire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE;
2386 VM_MAP_RANGE_CHECK(map, start, end);
2387 if (!vm_map_lookup_entry(map, start, &first_entry)) {
2388 if (flags & VM_MAP_WIRE_HOLESOK)
2389 first_entry = first_entry->next;
2392 return (KERN_INVALID_ADDRESS);
2395 last_timestamp = map->timestamp;
2396 entry = first_entry;
2397 while (entry != &map->header && entry->start < end) {
2398 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
2400 * We have not yet clipped the entry.
2402 saved_start = (start >= entry->start) ? start :
2404 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2405 if (vm_map_unlock_and_wait(map, 0)) {
2407 * Allow interruption of user unwiring?
2411 if (last_timestamp+1 != map->timestamp) {
2413 * Look again for the entry because the map was
2414 * modified while it was unlocked.
2415 * Specifically, the entry may have been
2416 * clipped, merged, or deleted.
2418 if (!vm_map_lookup_entry(map, saved_start,
2420 if (flags & VM_MAP_WIRE_HOLESOK)
2421 tmp_entry = tmp_entry->next;
2423 if (saved_start == start) {
2425 * First_entry has been deleted.
2428 return (KERN_INVALID_ADDRESS);
2431 rv = KERN_INVALID_ADDRESS;
2435 if (entry == first_entry)
2436 first_entry = tmp_entry;
2441 last_timestamp = map->timestamp;
2444 vm_map_clip_start(map, entry, start);
2445 vm_map_clip_end(map, entry, end);
2447 * Mark the entry in case the map lock is released. (See
2450 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 &&
2451 entry->wiring_thread == NULL,
2452 ("owned map entry %p", entry));
2453 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
2454 entry->wiring_thread = curthread;
2456 * Check the map for holes in the specified region.
2457 * If VM_MAP_WIRE_HOLESOK was specified, skip this check.
2459 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) &&
2460 (entry->end < end && (entry->next == &map->header ||
2461 entry->next->start > entry->end))) {
2463 rv = KERN_INVALID_ADDRESS;
2467 * If system unwiring, require that the entry is system wired.
2470 vm_map_entry_system_wired_count(entry) == 0) {
2472 rv = KERN_INVALID_ARGUMENT;
2475 entry = entry->next;
2479 need_wakeup = FALSE;
2480 if (first_entry == NULL) {
2481 result = vm_map_lookup_entry(map, start, &first_entry);
2482 if (!result && (flags & VM_MAP_WIRE_HOLESOK))
2483 first_entry = first_entry->next;
2485 KASSERT(result, ("vm_map_unwire: lookup failed"));
2487 for (entry = first_entry; entry != &map->header && entry->start < end;
2488 entry = entry->next) {
2490 * If VM_MAP_WIRE_HOLESOK was specified, an empty
2491 * space in the unwired region could have been mapped
2492 * while the map lock was dropped for draining
2493 * MAP_ENTRY_IN_TRANSITION. Moreover, another thread
2494 * could be simultaneously wiring this new mapping
2495 * entry. Detect these cases and skip any entries
2496 * marked as in transition by us.
2498 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 ||
2499 entry->wiring_thread != curthread) {
2500 KASSERT((flags & VM_MAP_WIRE_HOLESOK) != 0,
2501 ("vm_map_unwire: !HOLESOK and new/changed entry"));
2505 if (rv == KERN_SUCCESS && (!user_unwire ||
2506 (entry->eflags & MAP_ENTRY_USER_WIRED))) {
2508 entry->eflags &= ~MAP_ENTRY_USER_WIRED;
2509 if (entry->wired_count == 1)
2510 vm_map_entry_unwire(map, entry);
2512 entry->wired_count--;
2514 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0,
2515 ("vm_map_unwire: in-transition flag missing %p", entry));
2516 KASSERT(entry->wiring_thread == curthread,
2517 ("vm_map_unwire: alien wire %p", entry));
2518 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION;
2519 entry->wiring_thread = NULL;
2520 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
2521 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
2524 vm_map_simplify_entry(map, entry);
2533 * vm_map_wire_entry_failure:
2535 * Handle a wiring failure on the given entry.
2537 * The map should be locked.
2540 vm_map_wire_entry_failure(vm_map_t map, vm_map_entry_t entry,
2541 vm_offset_t failed_addr)
2544 VM_MAP_ASSERT_LOCKED(map);
2545 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 &&
2546 entry->wired_count == 1,
2547 ("vm_map_wire_entry_failure: entry %p isn't being wired", entry));
2548 KASSERT(failed_addr < entry->end,
2549 ("vm_map_wire_entry_failure: entry %p was fully wired", entry));
2552 * If any pages at the start of this entry were successfully wired,
2555 if (failed_addr > entry->start) {
2556 pmap_unwire(map->pmap, entry->start, failed_addr);
2557 vm_object_unwire(entry->object.vm_object, entry->offset,
2558 failed_addr - entry->start, PQ_ACTIVE);
2562 * Assign an out-of-range value to represent the failure to wire this
2565 entry->wired_count = -1;
2571 * Implements both kernel and user wiring.
2574 vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t end,
2577 vm_map_entry_t entry, first_entry, tmp_entry;
2578 vm_offset_t faddr, saved_end, saved_start;
2579 unsigned int last_timestamp;
2581 boolean_t need_wakeup, result, user_wire;
2585 return (KERN_SUCCESS);
2587 if (flags & VM_MAP_WIRE_WRITE)
2588 prot |= VM_PROT_WRITE;
2589 user_wire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE;
2591 VM_MAP_RANGE_CHECK(map, start, end);
2592 if (!vm_map_lookup_entry(map, start, &first_entry)) {
2593 if (flags & VM_MAP_WIRE_HOLESOK)
2594 first_entry = first_entry->next;
2597 return (KERN_INVALID_ADDRESS);
2600 last_timestamp = map->timestamp;
2601 entry = first_entry;
2602 while (entry != &map->header && entry->start < end) {
2603 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
2605 * We have not yet clipped the entry.
2607 saved_start = (start >= entry->start) ? start :
2609 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2610 if (vm_map_unlock_and_wait(map, 0)) {
2612 * Allow interruption of user wiring?
2616 if (last_timestamp + 1 != map->timestamp) {
2618 * Look again for the entry because the map was
2619 * modified while it was unlocked.
2620 * Specifically, the entry may have been
2621 * clipped, merged, or deleted.
2623 if (!vm_map_lookup_entry(map, saved_start,
2625 if (flags & VM_MAP_WIRE_HOLESOK)
2626 tmp_entry = tmp_entry->next;
2628 if (saved_start == start) {
2630 * first_entry has been deleted.
2633 return (KERN_INVALID_ADDRESS);
2636 rv = KERN_INVALID_ADDRESS;
2640 if (entry == first_entry)
2641 first_entry = tmp_entry;
2646 last_timestamp = map->timestamp;
2649 vm_map_clip_start(map, entry, start);
2650 vm_map_clip_end(map, entry, end);
2652 * Mark the entry in case the map lock is released. (See
2655 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 &&
2656 entry->wiring_thread == NULL,
2657 ("owned map entry %p", entry));
2658 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
2659 entry->wiring_thread = curthread;
2660 if ((entry->protection & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0
2661 || (entry->protection & prot) != prot) {
2662 entry->eflags |= MAP_ENTRY_WIRE_SKIPPED;
2663 if ((flags & VM_MAP_WIRE_HOLESOK) == 0) {
2665 rv = KERN_INVALID_ADDRESS;
2670 if (entry->wired_count == 0) {
2671 entry->wired_count++;
2672 saved_start = entry->start;
2673 saved_end = entry->end;
2676 * Release the map lock, relying on the in-transition
2677 * mark. Mark the map busy for fork.
2682 faddr = saved_start;
2685 * Simulate a fault to get the page and enter
2686 * it into the physical map.
2688 if ((rv = vm_fault(map, faddr, VM_PROT_NONE,
2689 VM_FAULT_WIRE)) != KERN_SUCCESS)
2691 } while ((faddr += PAGE_SIZE) < saved_end);
2694 if (last_timestamp + 1 != map->timestamp) {
2696 * Look again for the entry because the map was
2697 * modified while it was unlocked. The entry
2698 * may have been clipped, but NOT merged or
2701 result = vm_map_lookup_entry(map, saved_start,
2703 KASSERT(result, ("vm_map_wire: lookup failed"));
2704 if (entry == first_entry)
2705 first_entry = tmp_entry;
2709 while (entry->end < saved_end) {
2711 * In case of failure, handle entries
2712 * that were not fully wired here;
2713 * fully wired entries are handled
2716 if (rv != KERN_SUCCESS &&
2718 vm_map_wire_entry_failure(map,
2720 entry = entry->next;
2723 last_timestamp = map->timestamp;
2724 if (rv != KERN_SUCCESS) {
2725 vm_map_wire_entry_failure(map, entry, faddr);
2729 } else if (!user_wire ||
2730 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
2731 entry->wired_count++;
2734 * Check the map for holes in the specified region.
2735 * If VM_MAP_WIRE_HOLESOK was specified, skip this check.
2738 if ((flags & VM_MAP_WIRE_HOLESOK) == 0 &&
2739 entry->end < end && (entry->next == &map->header ||
2740 entry->next->start > entry->end)) {
2742 rv = KERN_INVALID_ADDRESS;
2745 entry = entry->next;
2749 need_wakeup = FALSE;
2750 if (first_entry == NULL) {
2751 result = vm_map_lookup_entry(map, start, &first_entry);
2752 if (!result && (flags & VM_MAP_WIRE_HOLESOK))
2753 first_entry = first_entry->next;
2755 KASSERT(result, ("vm_map_wire: lookup failed"));
2757 for (entry = first_entry; entry != &map->header && entry->start < end;
2758 entry = entry->next) {
2760 * If VM_MAP_WIRE_HOLESOK was specified, an empty
2761 * space in the unwired region could have been mapped
2762 * while the map lock was dropped for faulting in the
2763 * pages or draining MAP_ENTRY_IN_TRANSITION.
2764 * Moreover, another thread could be simultaneously
2765 * wiring this new mapping entry. Detect these cases
2766 * and skip any entries marked as in transition not by us.
2768 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 ||
2769 entry->wiring_thread != curthread) {
2770 KASSERT((flags & VM_MAP_WIRE_HOLESOK) != 0,
2771 ("vm_map_wire: !HOLESOK and new/changed entry"));
2775 if ((entry->eflags & MAP_ENTRY_WIRE_SKIPPED) != 0)
2776 goto next_entry_done;
2778 if (rv == KERN_SUCCESS) {
2780 entry->eflags |= MAP_ENTRY_USER_WIRED;
2781 } else if (entry->wired_count == -1) {
2783 * Wiring failed on this entry. Thus, unwiring is
2786 entry->wired_count = 0;
2787 } else if (!user_wire ||
2788 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
2790 * Undo the wiring. Wiring succeeded on this entry
2791 * but failed on a later entry.
2793 if (entry->wired_count == 1)
2794 vm_map_entry_unwire(map, entry);
2796 entry->wired_count--;
2799 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0,
2800 ("vm_map_wire: in-transition flag missing %p", entry));
2801 KASSERT(entry->wiring_thread == curthread,
2802 ("vm_map_wire: alien wire %p", entry));
2803 entry->eflags &= ~(MAP_ENTRY_IN_TRANSITION |
2804 MAP_ENTRY_WIRE_SKIPPED);
2805 entry->wiring_thread = NULL;
2806 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
2807 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
2810 vm_map_simplify_entry(map, entry);
2821 * Push any dirty cached pages in the address range to their pager.
2822 * If syncio is TRUE, dirty pages are written synchronously.
2823 * If invalidate is TRUE, any cached pages are freed as well.
2825 * If the size of the region from start to end is zero, we are
2826 * supposed to flush all modified pages within the region containing
2827 * start. Unfortunately, a region can be split or coalesced with
2828 * neighboring regions, making it difficult to determine what the
2829 * original region was. Therefore, we approximate this requirement by
2830 * flushing the current region containing start.
2832 * Returns an error if any part of the specified range is not mapped.
2840 boolean_t invalidate)
2842 vm_map_entry_t current;
2843 vm_map_entry_t entry;
2846 vm_ooffset_t offset;
2847 unsigned int last_timestamp;
2850 vm_map_lock_read(map);
2851 VM_MAP_RANGE_CHECK(map, start, end);
2852 if (!vm_map_lookup_entry(map, start, &entry)) {
2853 vm_map_unlock_read(map);
2854 return (KERN_INVALID_ADDRESS);
2855 } else if (start == end) {
2856 start = entry->start;
2860 * Make a first pass to check for user-wired memory and holes.
2862 for (current = entry; current != &map->header && current->start < end;
2863 current = current->next) {
2864 if (invalidate && (current->eflags & MAP_ENTRY_USER_WIRED)) {
2865 vm_map_unlock_read(map);
2866 return (KERN_INVALID_ARGUMENT);
2868 if (end > current->end &&
2869 (current->next == &map->header ||
2870 current->end != current->next->start)) {
2871 vm_map_unlock_read(map);
2872 return (KERN_INVALID_ADDRESS);
2877 pmap_remove(map->pmap, start, end);
2881 * Make a second pass, cleaning/uncaching pages from the indicated
2884 for (current = entry; current != &map->header && current->start < end;) {
2885 offset = current->offset + (start - current->start);
2886 size = (end <= current->end ? end : current->end) - start;
2887 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
2889 vm_map_entry_t tentry;
2892 smap = current->object.sub_map;
2893 vm_map_lock_read(smap);
2894 (void) vm_map_lookup_entry(smap, offset, &tentry);
2895 tsize = tentry->end - offset;
2898 object = tentry->object.vm_object;
2899 offset = tentry->offset + (offset - tentry->start);
2900 vm_map_unlock_read(smap);
2902 object = current->object.vm_object;
2904 vm_object_reference(object);
2905 last_timestamp = map->timestamp;
2906 vm_map_unlock_read(map);
2907 if (!vm_object_sync(object, offset, size, syncio, invalidate))
2910 vm_object_deallocate(object);
2911 vm_map_lock_read(map);
2912 if (last_timestamp == map->timestamp ||
2913 !vm_map_lookup_entry(map, start, ¤t))
2914 current = current->next;
2917 vm_map_unlock_read(map);
2918 return (failed ? KERN_FAILURE : KERN_SUCCESS);
2922 * vm_map_entry_unwire: [ internal use only ]
2924 * Make the region specified by this entry pageable.
2926 * The map in question should be locked.
2927 * [This is the reason for this routine's existence.]
2930 vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry)
2933 VM_MAP_ASSERT_LOCKED(map);
2934 KASSERT(entry->wired_count > 0,
2935 ("vm_map_entry_unwire: entry %p isn't wired", entry));
2936 pmap_unwire(map->pmap, entry->start, entry->end);
2937 vm_object_unwire(entry->object.vm_object, entry->offset, entry->end -
2938 entry->start, PQ_ACTIVE);
2939 entry->wired_count = 0;
2943 vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map)
2946 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0)
2947 vm_object_deallocate(entry->object.vm_object);
2948 uma_zfree(system_map ? kmapentzone : mapentzone, entry);
2952 * vm_map_entry_delete: [ internal use only ]
2954 * Deallocate the given entry from the target map.
2957 vm_map_entry_delete(vm_map_t map, vm_map_entry_t entry)
2960 vm_pindex_t offidxstart, offidxend, count, size1;
2963 vm_map_entry_unlink(map, entry);
2964 object = entry->object.vm_object;
2966 if ((entry->eflags & MAP_ENTRY_GUARD) != 0) {
2967 MPASS(entry->cred == NULL);
2968 MPASS((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0);
2969 MPASS(object == NULL);
2970 vm_map_entry_deallocate(entry, map->system_map);
2974 size = entry->end - entry->start;
2977 if (entry->cred != NULL) {
2978 swap_release_by_cred(size, entry->cred);
2979 crfree(entry->cred);
2982 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0 &&
2984 KASSERT(entry->cred == NULL || object->cred == NULL ||
2985 (entry->eflags & MAP_ENTRY_NEEDS_COPY),
2986 ("OVERCOMMIT vm_map_entry_delete: both cred %p", entry));
2988 offidxstart = OFF_TO_IDX(entry->offset);
2989 offidxend = offidxstart + count;
2990 VM_OBJECT_WLOCK(object);
2991 if (object->ref_count != 1 && ((object->flags & (OBJ_NOSPLIT |
2992 OBJ_ONEMAPPING)) == OBJ_ONEMAPPING ||
2993 object == kernel_object)) {
2994 vm_object_collapse(object);
2997 * The option OBJPR_NOTMAPPED can be passed here
2998 * because vm_map_delete() already performed
2999 * pmap_remove() on the only mapping to this range
3002 vm_object_page_remove(object, offidxstart, offidxend,
3004 if (object->type == OBJT_SWAP)
3005 swap_pager_freespace(object, offidxstart,
3007 if (offidxend >= object->size &&
3008 offidxstart < object->size) {
3009 size1 = object->size;
3010 object->size = offidxstart;
3011 if (object->cred != NULL) {
3012 size1 -= object->size;
3013 KASSERT(object->charge >= ptoa(size1),
3014 ("object %p charge < 0", object));
3015 swap_release_by_cred(ptoa(size1),
3017 object->charge -= ptoa(size1);
3021 VM_OBJECT_WUNLOCK(object);
3023 entry->object.vm_object = NULL;
3024 if (map->system_map)
3025 vm_map_entry_deallocate(entry, TRUE);
3027 entry->next = curthread->td_map_def_user;
3028 curthread->td_map_def_user = entry;
3033 * vm_map_delete: [ internal use only ]
3035 * Deallocates the given address range from the target
3039 vm_map_delete(vm_map_t map, vm_offset_t start, vm_offset_t end)
3041 vm_map_entry_t entry;
3042 vm_map_entry_t first_entry;
3044 VM_MAP_ASSERT_LOCKED(map);
3046 return (KERN_SUCCESS);
3049 * Find the start of the region, and clip it
3051 if (!vm_map_lookup_entry(map, start, &first_entry))
3052 entry = first_entry->next;
3054 entry = first_entry;
3055 vm_map_clip_start(map, entry, start);
3059 * Step through all entries in this region
3061 while ((entry != &map->header) && (entry->start < end)) {
3062 vm_map_entry_t next;
3065 * Wait for wiring or unwiring of an entry to complete.
3066 * Also wait for any system wirings to disappear on
3069 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 ||
3070 (vm_map_pmap(map) != kernel_pmap &&
3071 vm_map_entry_system_wired_count(entry) != 0)) {
3072 unsigned int last_timestamp;
3073 vm_offset_t saved_start;
3074 vm_map_entry_t tmp_entry;
3076 saved_start = entry->start;
3077 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
3078 last_timestamp = map->timestamp;
3079 (void) vm_map_unlock_and_wait(map, 0);
3081 if (last_timestamp + 1 != map->timestamp) {
3083 * Look again for the entry because the map was
3084 * modified while it was unlocked.
3085 * Specifically, the entry may have been
3086 * clipped, merged, or deleted.
3088 if (!vm_map_lookup_entry(map, saved_start,
3090 entry = tmp_entry->next;
3093 vm_map_clip_start(map, entry,
3099 vm_map_clip_end(map, entry, end);
3104 * Unwire before removing addresses from the pmap; otherwise,
3105 * unwiring will put the entries back in the pmap.
3107 if (entry->wired_count != 0) {
3108 vm_map_entry_unwire(map, entry);
3111 pmap_remove(map->pmap, entry->start, entry->end);
3114 * Delete the entry only after removing all pmap
3115 * entries pointing to its pages. (Otherwise, its
3116 * page frames may be reallocated, and any modify bits
3117 * will be set in the wrong object!)
3119 vm_map_entry_delete(map, entry);
3122 return (KERN_SUCCESS);
3128 * Remove the given address range from the target map.
3129 * This is the exported form of vm_map_delete.
3132 vm_map_remove(vm_map_t map, vm_offset_t start, vm_offset_t end)
3137 VM_MAP_RANGE_CHECK(map, start, end);
3138 result = vm_map_delete(map, start, end);
3144 * vm_map_check_protection:
3146 * Assert that the target map allows the specified privilege on the
3147 * entire address region given. The entire region must be allocated.
3149 * WARNING! This code does not and should not check whether the
3150 * contents of the region is accessible. For example a smaller file
3151 * might be mapped into a larger address space.
3153 * NOTE! This code is also called by munmap().
3155 * The map must be locked. A read lock is sufficient.
3158 vm_map_check_protection(vm_map_t map, vm_offset_t start, vm_offset_t end,
3159 vm_prot_t protection)
3161 vm_map_entry_t entry;
3162 vm_map_entry_t tmp_entry;
3164 if (!vm_map_lookup_entry(map, start, &tmp_entry))
3168 while (start < end) {
3169 if (entry == &map->header)
3174 if (start < entry->start)
3177 * Check protection associated with entry.
3179 if ((entry->protection & protection) != protection)
3181 /* go to next entry */
3183 entry = entry->next;
3189 * vm_map_copy_entry:
3191 * Copies the contents of the source entry to the destination
3192 * entry. The entries *must* be aligned properly.
3198 vm_map_entry_t src_entry,
3199 vm_map_entry_t dst_entry,
3200 vm_ooffset_t *fork_charge)
3202 vm_object_t src_object;
3203 vm_map_entry_t fake_entry;
3208 VM_MAP_ASSERT_LOCKED(dst_map);
3210 if ((dst_entry->eflags|src_entry->eflags) & MAP_ENTRY_IS_SUB_MAP)
3213 if (src_entry->wired_count == 0 ||
3214 (src_entry->protection & VM_PROT_WRITE) == 0) {
3216 * If the source entry is marked needs_copy, it is already
3219 if ((src_entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0 &&
3220 (src_entry->protection & VM_PROT_WRITE) != 0) {
3221 pmap_protect(src_map->pmap,
3224 src_entry->protection & ~VM_PROT_WRITE);
3228 * Make a copy of the object.
3230 size = src_entry->end - src_entry->start;
3231 if ((src_object = src_entry->object.vm_object) != NULL) {
3232 VM_OBJECT_WLOCK(src_object);
3233 charged = ENTRY_CHARGED(src_entry);
3234 if (src_object->handle == NULL &&
3235 (src_object->type == OBJT_DEFAULT ||
3236 src_object->type == OBJT_SWAP)) {
3237 vm_object_collapse(src_object);
3238 if ((src_object->flags & (OBJ_NOSPLIT |
3239 OBJ_ONEMAPPING)) == OBJ_ONEMAPPING) {
3240 vm_object_split(src_entry);
3242 src_entry->object.vm_object;
3245 vm_object_reference_locked(src_object);
3246 vm_object_clear_flag(src_object, OBJ_ONEMAPPING);
3247 if (src_entry->cred != NULL &&
3248 !(src_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
3249 KASSERT(src_object->cred == NULL,
3250 ("OVERCOMMIT: vm_map_copy_entry: cred %p",
3252 src_object->cred = src_entry->cred;
3253 src_object->charge = size;
3255 VM_OBJECT_WUNLOCK(src_object);
3256 dst_entry->object.vm_object = src_object;
3258 cred = curthread->td_ucred;
3260 dst_entry->cred = cred;
3261 *fork_charge += size;
3262 if (!(src_entry->eflags &
3263 MAP_ENTRY_NEEDS_COPY)) {
3265 src_entry->cred = cred;
3266 *fork_charge += size;
3269 src_entry->eflags |= MAP_ENTRY_COW |
3270 MAP_ENTRY_NEEDS_COPY;
3271 dst_entry->eflags |= MAP_ENTRY_COW |
3272 MAP_ENTRY_NEEDS_COPY;
3273 dst_entry->offset = src_entry->offset;
3274 if (src_entry->eflags & MAP_ENTRY_VN_WRITECNT) {
3276 * MAP_ENTRY_VN_WRITECNT cannot
3277 * indicate write reference from
3278 * src_entry, since the entry is
3279 * marked as needs copy. Allocate a
3280 * fake entry that is used to
3281 * decrement object->un_pager.vnp.writecount
3282 * at the appropriate time. Attach
3283 * fake_entry to the deferred list.
3285 fake_entry = vm_map_entry_create(dst_map);
3286 fake_entry->eflags = MAP_ENTRY_VN_WRITECNT;
3287 src_entry->eflags &= ~MAP_ENTRY_VN_WRITECNT;
3288 vm_object_reference(src_object);
3289 fake_entry->object.vm_object = src_object;
3290 fake_entry->start = src_entry->start;
3291 fake_entry->end = src_entry->end;
3292 fake_entry->next = curthread->td_map_def_user;
3293 curthread->td_map_def_user = fake_entry;
3296 pmap_copy(dst_map->pmap, src_map->pmap,
3297 dst_entry->start, dst_entry->end - dst_entry->start,
3300 dst_entry->object.vm_object = NULL;
3301 dst_entry->offset = 0;
3302 if (src_entry->cred != NULL) {
3303 dst_entry->cred = curthread->td_ucred;
3304 crhold(dst_entry->cred);
3305 *fork_charge += size;
3310 * We don't want to make writeable wired pages copy-on-write.
3311 * Immediately copy these pages into the new map by simulating
3312 * page faults. The new pages are pageable.
3314 vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry,
3320 * vmspace_map_entry_forked:
3321 * Update the newly-forked vmspace each time a map entry is inherited
3322 * or copied. The values for vm_dsize and vm_tsize are approximate
3323 * (and mostly-obsolete ideas in the face of mmap(2) et al.)
3326 vmspace_map_entry_forked(const struct vmspace *vm1, struct vmspace *vm2,
3327 vm_map_entry_t entry)
3329 vm_size_t entrysize;
3332 if ((entry->eflags & MAP_ENTRY_GUARD) != 0)
3334 entrysize = entry->end - entry->start;
3335 vm2->vm_map.size += entrysize;
3336 if (entry->eflags & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP)) {
3337 vm2->vm_ssize += btoc(entrysize);
3338 } else if (entry->start >= (vm_offset_t)vm1->vm_daddr &&
3339 entry->start < (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)) {
3340 newend = MIN(entry->end,
3341 (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize));
3342 vm2->vm_dsize += btoc(newend - entry->start);
3343 } else if (entry->start >= (vm_offset_t)vm1->vm_taddr &&
3344 entry->start < (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)) {
3345 newend = MIN(entry->end,
3346 (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize));
3347 vm2->vm_tsize += btoc(newend - entry->start);
3353 * Create a new process vmspace structure and vm_map
3354 * based on those of an existing process. The new map
3355 * is based on the old map, according to the inheritance
3356 * values on the regions in that map.
3358 * XXX It might be worth coalescing the entries added to the new vmspace.
3360 * The source map must not be locked.
3363 vmspace_fork(struct vmspace *vm1, vm_ooffset_t *fork_charge)
3365 struct vmspace *vm2;
3366 vm_map_t new_map, old_map;
3367 vm_map_entry_t new_entry, old_entry;
3372 old_map = &vm1->vm_map;
3373 /* Copy immutable fields of vm1 to vm2. */
3374 vm2 = vmspace_alloc(old_map->min_offset, old_map->max_offset, NULL);
3377 vm2->vm_taddr = vm1->vm_taddr;
3378 vm2->vm_daddr = vm1->vm_daddr;
3379 vm2->vm_maxsaddr = vm1->vm_maxsaddr;
3380 vm_map_lock(old_map);
3382 vm_map_wait_busy(old_map);
3383 new_map = &vm2->vm_map;
3384 locked = vm_map_trylock(new_map); /* trylock to silence WITNESS */
3385 KASSERT(locked, ("vmspace_fork: lock failed"));
3387 old_entry = old_map->header.next;
3389 while (old_entry != &old_map->header) {
3390 if (old_entry->eflags & MAP_ENTRY_IS_SUB_MAP)
3391 panic("vm_map_fork: encountered a submap");
3393 inh = old_entry->inheritance;
3394 if ((old_entry->eflags & MAP_ENTRY_GUARD) != 0 &&
3395 inh != VM_INHERIT_NONE)
3396 inh = VM_INHERIT_COPY;
3399 case VM_INHERIT_NONE:
3402 case VM_INHERIT_SHARE:
3404 * Clone the entry, creating the shared object if necessary.
3406 object = old_entry->object.vm_object;
3407 if (object == NULL) {
3408 object = vm_object_allocate(OBJT_DEFAULT,
3409 atop(old_entry->end - old_entry->start));
3410 old_entry->object.vm_object = object;
3411 old_entry->offset = 0;
3412 if (old_entry->cred != NULL) {
3413 object->cred = old_entry->cred;
3414 object->charge = old_entry->end -
3416 old_entry->cred = NULL;
3421 * Add the reference before calling vm_object_shadow
3422 * to insure that a shadow object is created.
3424 vm_object_reference(object);
3425 if (old_entry->eflags & MAP_ENTRY_NEEDS_COPY) {
3426 vm_object_shadow(&old_entry->object.vm_object,
3428 old_entry->end - old_entry->start);
3429 old_entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
3430 /* Transfer the second reference too. */
3431 vm_object_reference(
3432 old_entry->object.vm_object);
3435 * As in vm_map_simplify_entry(), the
3436 * vnode lock will not be acquired in
3437 * this call to vm_object_deallocate().
3439 vm_object_deallocate(object);
3440 object = old_entry->object.vm_object;
3442 VM_OBJECT_WLOCK(object);
3443 vm_object_clear_flag(object, OBJ_ONEMAPPING);
3444 if (old_entry->cred != NULL) {
3445 KASSERT(object->cred == NULL, ("vmspace_fork both cred"));
3446 object->cred = old_entry->cred;
3447 object->charge = old_entry->end - old_entry->start;
3448 old_entry->cred = NULL;
3452 * Assert the correct state of the vnode
3453 * v_writecount while the object is locked, to
3454 * not relock it later for the assertion
3457 if (old_entry->eflags & MAP_ENTRY_VN_WRITECNT &&
3458 object->type == OBJT_VNODE) {
3459 KASSERT(((struct vnode *)object->handle)->
3461 ("vmspace_fork: v_writecount %p", object));
3462 KASSERT(object->un_pager.vnp.writemappings > 0,
3463 ("vmspace_fork: vnp.writecount %p",
3466 VM_OBJECT_WUNLOCK(object);
3469 * Clone the entry, referencing the shared object.
3471 new_entry = vm_map_entry_create(new_map);
3472 *new_entry = *old_entry;
3473 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED |
3474 MAP_ENTRY_IN_TRANSITION);
3475 new_entry->wiring_thread = NULL;
3476 new_entry->wired_count = 0;
3477 if (new_entry->eflags & MAP_ENTRY_VN_WRITECNT) {
3478 vnode_pager_update_writecount(object,
3479 new_entry->start, new_entry->end);
3483 * Insert the entry into the new map -- we know we're
3484 * inserting at the end of the new map.
3486 vm_map_entry_link(new_map, new_map->header.prev,
3488 vmspace_map_entry_forked(vm1, vm2, new_entry);
3491 * Update the physical map
3493 pmap_copy(new_map->pmap, old_map->pmap,
3495 (old_entry->end - old_entry->start),
3499 case VM_INHERIT_COPY:
3501 * Clone the entry and link into the map.
3503 new_entry = vm_map_entry_create(new_map);
3504 *new_entry = *old_entry;
3506 * Copied entry is COW over the old object.
3508 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED |
3509 MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_VN_WRITECNT);
3510 new_entry->wiring_thread = NULL;
3511 new_entry->wired_count = 0;
3512 new_entry->object.vm_object = NULL;
3513 new_entry->cred = NULL;
3514 vm_map_entry_link(new_map, new_map->header.prev,
3516 vmspace_map_entry_forked(vm1, vm2, new_entry);
3517 vm_map_copy_entry(old_map, new_map, old_entry,
3518 new_entry, fork_charge);
3521 case VM_INHERIT_ZERO:
3523 * Create a new anonymous mapping entry modelled from
3526 new_entry = vm_map_entry_create(new_map);
3527 memset(new_entry, 0, sizeof(*new_entry));
3529 new_entry->start = old_entry->start;
3530 new_entry->end = old_entry->end;
3531 new_entry->eflags = old_entry->eflags &
3532 ~(MAP_ENTRY_USER_WIRED | MAP_ENTRY_IN_TRANSITION |
3533 MAP_ENTRY_VN_WRITECNT);
3534 new_entry->protection = old_entry->protection;
3535 new_entry->max_protection = old_entry->max_protection;
3536 new_entry->inheritance = VM_INHERIT_ZERO;
3538 vm_map_entry_link(new_map, new_map->header.prev,
3540 vmspace_map_entry_forked(vm1, vm2, new_entry);
3542 new_entry->cred = curthread->td_ucred;
3543 crhold(new_entry->cred);
3544 *fork_charge += (new_entry->end - new_entry->start);
3548 old_entry = old_entry->next;
3551 * Use inlined vm_map_unlock() to postpone handling the deferred
3552 * map entries, which cannot be done until both old_map and
3553 * new_map locks are released.
3555 sx_xunlock(&old_map->lock);
3556 sx_xunlock(&new_map->lock);
3557 vm_map_process_deferred();
3563 * Create a process's stack for exec_new_vmspace(). This function is never
3564 * asked to wire the newly created stack.
3567 vm_map_stack(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
3568 vm_prot_t prot, vm_prot_t max, int cow)
3570 vm_size_t growsize, init_ssize;
3574 MPASS((map->flags & MAP_WIREFUTURE) == 0);
3575 growsize = sgrowsiz;
3576 init_ssize = (max_ssize < growsize) ? max_ssize : growsize;
3578 vmemlim = lim_cur(curthread, RLIMIT_VMEM);
3579 /* If we would blow our VMEM resource limit, no go */
3580 if (map->size + init_ssize > vmemlim) {
3584 rv = vm_map_stack_locked(map, addrbos, max_ssize, growsize, prot,
3591 static int stack_guard_page = 1;
3592 SYSCTL_INT(_security_bsd, OID_AUTO, stack_guard_page, CTLFLAG_RWTUN,
3593 &stack_guard_page, 0,
3594 "Specifies the number of guard pages for a stack that grows");
3597 vm_map_stack_locked(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
3598 vm_size_t growsize, vm_prot_t prot, vm_prot_t max, int cow)
3600 vm_map_entry_t new_entry, prev_entry;
3601 vm_offset_t bot, gap_bot, gap_top, top;
3602 vm_size_t init_ssize, sgp;
3606 * The stack orientation is piggybacked with the cow argument.
3607 * Extract it into orient and mask the cow argument so that we
3608 * don't pass it around further.
3610 orient = cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP);
3611 KASSERT(orient != 0, ("No stack grow direction"));
3612 KASSERT(orient != (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP),
3615 if (addrbos < vm_map_min(map) ||
3616 addrbos + max_ssize > vm_map_max(map) ||
3617 addrbos + max_ssize <= addrbos)
3618 return (KERN_INVALID_ADDRESS);
3619 sgp = (vm_size_t)stack_guard_page * PAGE_SIZE;
3620 if (sgp >= max_ssize)
3621 return (KERN_INVALID_ARGUMENT);
3623 init_ssize = growsize;
3624 if (max_ssize < init_ssize + sgp)
3625 init_ssize = max_ssize - sgp;
3627 /* If addr is already mapped, no go */
3628 if (vm_map_lookup_entry(map, addrbos, &prev_entry))
3629 return (KERN_NO_SPACE);
3632 * If we can't accommodate max_ssize in the current mapping, no go.
3634 if ((prev_entry->next != &map->header) &&
3635 (prev_entry->next->start < addrbos + max_ssize))
3636 return (KERN_NO_SPACE);
3639 * We initially map a stack of only init_ssize. We will grow as
3640 * needed later. Depending on the orientation of the stack (i.e.
3641 * the grow direction) we either map at the top of the range, the
3642 * bottom of the range or in the middle.
3644 * Note: we would normally expect prot and max to be VM_PROT_ALL,
3645 * and cow to be 0. Possibly we should eliminate these as input
3646 * parameters, and just pass these values here in the insert call.
3648 if (orient == MAP_STACK_GROWS_DOWN) {
3649 bot = addrbos + max_ssize - init_ssize;
3650 top = bot + init_ssize;
3653 } else /* if (orient == MAP_STACK_GROWS_UP) */ {
3655 top = bot + init_ssize;
3657 gap_top = addrbos + max_ssize;
3659 rv = vm_map_insert(map, NULL, 0, bot, top, prot, max, cow);
3660 if (rv != KERN_SUCCESS)
3662 new_entry = prev_entry->next;
3663 KASSERT(new_entry->end == top || new_entry->start == bot,
3664 ("Bad entry start/end for new stack entry"));
3665 KASSERT((orient & MAP_STACK_GROWS_DOWN) == 0 ||
3666 (new_entry->eflags & MAP_ENTRY_GROWS_DOWN) != 0,
3667 ("new entry lacks MAP_ENTRY_GROWS_DOWN"));
3668 KASSERT((orient & MAP_STACK_GROWS_UP) == 0 ||
3669 (new_entry->eflags & MAP_ENTRY_GROWS_UP) != 0,
3670 ("new entry lacks MAP_ENTRY_GROWS_UP"));
3671 rv = vm_map_insert(map, NULL, 0, gap_bot, gap_top, VM_PROT_NONE,
3672 VM_PROT_NONE, MAP_CREATE_GUARD | (orient == MAP_STACK_GROWS_DOWN ?
3673 MAP_CREATE_STACK_GAP_DN : MAP_CREATE_STACK_GAP_UP));
3674 if (rv != KERN_SUCCESS)
3675 (void)vm_map_delete(map, bot, top);
3680 * Attempts to grow a vm stack entry. Returns KERN_SUCCESS if we
3681 * successfully grow the stack.
3684 vm_map_growstack(vm_map_t map, vm_offset_t addr, vm_map_entry_t gap_entry)
3686 vm_map_entry_t stack_entry;
3690 vm_offset_t gap_end, gap_start, grow_start;
3691 size_t grow_amount, guard, max_grow;
3692 rlim_t lmemlim, stacklim, vmemlim;
3694 bool gap_deleted, grow_down, is_procstack;
3706 * Disallow stack growth when the access is performed by a
3707 * debugger or AIO daemon. The reason is that the wrong
3708 * resource limits are applied.
3710 if (map != &p->p_vmspace->vm_map || p->p_textvp == NULL)
3711 return (KERN_FAILURE);
3713 MPASS(!map->system_map);
3715 guard = stack_guard_page * PAGE_SIZE;
3716 lmemlim = lim_cur(curthread, RLIMIT_MEMLOCK);
3717 stacklim = lim_cur(curthread, RLIMIT_STACK);
3718 vmemlim = lim_cur(curthread, RLIMIT_VMEM);
3720 /* If addr is not in a hole for a stack grow area, no need to grow. */
3721 if (gap_entry == NULL && !vm_map_lookup_entry(map, addr, &gap_entry))
3722 return (KERN_FAILURE);
3723 if ((gap_entry->eflags & MAP_ENTRY_GUARD) == 0)
3724 return (KERN_SUCCESS);
3725 if ((gap_entry->eflags & MAP_ENTRY_STACK_GAP_DN) != 0) {
3726 stack_entry = gap_entry->next;
3727 if ((stack_entry->eflags & MAP_ENTRY_GROWS_DOWN) == 0 ||
3728 stack_entry->start != gap_entry->end)
3729 return (KERN_FAILURE);
3730 grow_amount = round_page(stack_entry->start - addr);
3732 } else if ((gap_entry->eflags & MAP_ENTRY_STACK_GAP_UP) != 0) {
3733 stack_entry = gap_entry->prev;
3734 if ((stack_entry->eflags & MAP_ENTRY_GROWS_UP) == 0 ||
3735 stack_entry->end != gap_entry->start)
3736 return (KERN_FAILURE);
3737 grow_amount = round_page(addr + 1 - stack_entry->end);
3740 return (KERN_FAILURE);
3742 max_grow = gap_entry->end - gap_entry->start;
3743 if (guard > max_grow)
3744 return (KERN_NO_SPACE);
3746 if (grow_amount > max_grow)
3747 return (KERN_NO_SPACE);
3750 * If this is the main process stack, see if we're over the stack
3753 is_procstack = addr >= (vm_offset_t)vm->vm_maxsaddr &&
3754 addr < (vm_offset_t)p->p_sysent->sv_usrstack;
3755 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim))
3756 return (KERN_NO_SPACE);
3761 if (is_procstack && racct_set(p, RACCT_STACK,
3762 ctob(vm->vm_ssize) + grow_amount)) {
3764 return (KERN_NO_SPACE);
3770 grow_amount = roundup(grow_amount, sgrowsiz);
3771 if (grow_amount > max_grow)
3772 grow_amount = max_grow;
3773 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) {
3774 grow_amount = trunc_page((vm_size_t)stacklim) -
3780 limit = racct_get_available(p, RACCT_STACK);
3782 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > limit))
3783 grow_amount = limit - ctob(vm->vm_ssize);
3786 if (!old_mlock && (map->flags & MAP_WIREFUTURE) != 0) {
3787 if (ptoa(pmap_wired_count(map->pmap)) + grow_amount > lmemlim) {
3794 if (racct_set(p, RACCT_MEMLOCK,
3795 ptoa(pmap_wired_count(map->pmap)) + grow_amount)) {
3805 /* If we would blow our VMEM resource limit, no go */
3806 if (map->size + grow_amount > vmemlim) {
3813 if (racct_set(p, RACCT_VMEM, map->size + grow_amount)) {
3822 if (vm_map_lock_upgrade(map)) {
3824 vm_map_lock_read(map);
3829 grow_start = gap_entry->end - grow_amount;
3830 if (gap_entry->start + grow_amount == gap_entry->end) {
3831 gap_start = gap_entry->start;
3832 gap_end = gap_entry->end;
3833 vm_map_entry_delete(map, gap_entry);
3836 MPASS(gap_entry->start < gap_entry->end - grow_amount);
3837 gap_entry->end -= grow_amount;
3838 vm_map_entry_resize_free(map, gap_entry);
3839 gap_deleted = false;
3841 rv = vm_map_insert(map, NULL, 0, grow_start,
3842 grow_start + grow_amount,
3843 stack_entry->protection, stack_entry->max_protection,
3844 MAP_STACK_GROWS_DOWN);
3845 if (rv != KERN_SUCCESS) {
3847 rv1 = vm_map_insert(map, NULL, 0, gap_start,
3848 gap_end, VM_PROT_NONE, VM_PROT_NONE,
3849 MAP_CREATE_GUARD | MAP_CREATE_STACK_GAP_DN);
3850 MPASS(rv1 == KERN_SUCCESS);
3852 gap_entry->end += grow_amount;
3853 vm_map_entry_resize_free(map, gap_entry);
3857 grow_start = stack_entry->end;
3858 cred = stack_entry->cred;
3859 if (cred == NULL && stack_entry->object.vm_object != NULL)
3860 cred = stack_entry->object.vm_object->cred;
3861 if (cred != NULL && !swap_reserve_by_cred(grow_amount, cred))
3863 /* Grow the underlying object if applicable. */
3864 else if (stack_entry->object.vm_object == NULL ||
3865 vm_object_coalesce(stack_entry->object.vm_object,
3866 stack_entry->offset,
3867 (vm_size_t)(stack_entry->end - stack_entry->start),
3868 (vm_size_t)grow_amount, cred != NULL)) {
3869 if (gap_entry->start + grow_amount == gap_entry->end)
3870 vm_map_entry_delete(map, gap_entry);
3872 gap_entry->start += grow_amount;
3873 stack_entry->end += grow_amount;
3874 map->size += grow_amount;
3875 vm_map_entry_resize_free(map, stack_entry);
3880 if (rv == KERN_SUCCESS && is_procstack)
3881 vm->vm_ssize += btoc(grow_amount);
3884 * Heed the MAP_WIREFUTURE flag if it was set for this process.
3886 if (rv == KERN_SUCCESS && (map->flags & MAP_WIREFUTURE) != 0) {
3888 vm_map_wire(map, grow_start, grow_start + grow_amount,
3889 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
3890 vm_map_lock_read(map);
3892 vm_map_lock_downgrade(map);
3896 if (racct_enable && rv != KERN_SUCCESS) {
3898 error = racct_set(p, RACCT_VMEM, map->size);
3899 KASSERT(error == 0, ("decreasing RACCT_VMEM failed"));
3901 error = racct_set(p, RACCT_MEMLOCK,
3902 ptoa(pmap_wired_count(map->pmap)));
3903 KASSERT(error == 0, ("decreasing RACCT_MEMLOCK failed"));
3905 error = racct_set(p, RACCT_STACK, ctob(vm->vm_ssize));
3906 KASSERT(error == 0, ("decreasing RACCT_STACK failed"));
3915 * Unshare the specified VM space for exec. If other processes are
3916 * mapped to it, then create a new one. The new vmspace is null.
3919 vmspace_exec(struct proc *p, vm_offset_t minuser, vm_offset_t maxuser)
3921 struct vmspace *oldvmspace = p->p_vmspace;
3922 struct vmspace *newvmspace;
3924 KASSERT((curthread->td_pflags & TDP_EXECVMSPC) == 0,
3925 ("vmspace_exec recursed"));
3926 newvmspace = vmspace_alloc(minuser, maxuser, NULL);
3927 if (newvmspace == NULL)
3929 newvmspace->vm_swrss = oldvmspace->vm_swrss;
3931 * This code is written like this for prototype purposes. The
3932 * goal is to avoid running down the vmspace here, but let the
3933 * other process's that are still using the vmspace to finally
3934 * run it down. Even though there is little or no chance of blocking
3935 * here, it is a good idea to keep this form for future mods.
3937 PROC_VMSPACE_LOCK(p);
3938 p->p_vmspace = newvmspace;
3939 PROC_VMSPACE_UNLOCK(p);
3940 if (p == curthread->td_proc)
3941 pmap_activate(curthread);
3942 curthread->td_pflags |= TDP_EXECVMSPC;
3947 * Unshare the specified VM space for forcing COW. This
3948 * is called by rfork, for the (RFMEM|RFPROC) == 0 case.
3951 vmspace_unshare(struct proc *p)
3953 struct vmspace *oldvmspace = p->p_vmspace;
3954 struct vmspace *newvmspace;
3955 vm_ooffset_t fork_charge;
3957 if (oldvmspace->vm_refcnt == 1)
3960 newvmspace = vmspace_fork(oldvmspace, &fork_charge);
3961 if (newvmspace == NULL)
3963 if (!swap_reserve_by_cred(fork_charge, p->p_ucred)) {
3964 vmspace_free(newvmspace);
3967 PROC_VMSPACE_LOCK(p);
3968 p->p_vmspace = newvmspace;
3969 PROC_VMSPACE_UNLOCK(p);
3970 if (p == curthread->td_proc)
3971 pmap_activate(curthread);
3972 vmspace_free(oldvmspace);
3979 * Finds the VM object, offset, and
3980 * protection for a given virtual address in the
3981 * specified map, assuming a page fault of the
3984 * Leaves the map in question locked for read; return
3985 * values are guaranteed until a vm_map_lookup_done
3986 * call is performed. Note that the map argument
3987 * is in/out; the returned map must be used in
3988 * the call to vm_map_lookup_done.
3990 * A handle (out_entry) is returned for use in
3991 * vm_map_lookup_done, to make that fast.
3993 * If a lookup is requested with "write protection"
3994 * specified, the map may be changed to perform virtual
3995 * copying operations, although the data referenced will
3999 vm_map_lookup(vm_map_t *var_map, /* IN/OUT */
4001 vm_prot_t fault_typea,
4002 vm_map_entry_t *out_entry, /* OUT */
4003 vm_object_t *object, /* OUT */
4004 vm_pindex_t *pindex, /* OUT */
4005 vm_prot_t *out_prot, /* OUT */
4006 boolean_t *wired) /* OUT */
4008 vm_map_entry_t entry;
4009 vm_map_t map = *var_map;
4011 vm_prot_t fault_type = fault_typea;
4012 vm_object_t eobject;
4018 vm_map_lock_read(map);
4022 * Lookup the faulting address.
4024 if (!vm_map_lookup_entry(map, vaddr, out_entry)) {
4025 vm_map_unlock_read(map);
4026 return (KERN_INVALID_ADDRESS);
4034 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
4035 vm_map_t old_map = map;
4037 *var_map = map = entry->object.sub_map;
4038 vm_map_unlock_read(old_map);
4043 * Check whether this task is allowed to have this page.
4045 prot = entry->protection;
4046 if ((fault_typea & VM_PROT_FAULT_LOOKUP) != 0) {
4047 fault_typea &= ~VM_PROT_FAULT_LOOKUP;
4048 if (prot == VM_PROT_NONE && map != kernel_map &&
4049 (entry->eflags & MAP_ENTRY_GUARD) != 0 &&
4050 (entry->eflags & (MAP_ENTRY_STACK_GAP_DN |
4051 MAP_ENTRY_STACK_GAP_UP)) != 0 &&
4052 vm_map_growstack(map, vaddr, entry) == KERN_SUCCESS)
4053 goto RetryLookupLocked;
4055 fault_type &= VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE;
4056 if ((fault_type & prot) != fault_type || prot == VM_PROT_NONE) {
4057 vm_map_unlock_read(map);
4058 return (KERN_PROTECTION_FAILURE);
4060 KASSERT((prot & VM_PROT_WRITE) == 0 || (entry->eflags &
4061 (MAP_ENTRY_USER_WIRED | MAP_ENTRY_NEEDS_COPY)) !=
4062 (MAP_ENTRY_USER_WIRED | MAP_ENTRY_NEEDS_COPY),
4063 ("entry %p flags %x", entry, entry->eflags));
4064 if ((fault_typea & VM_PROT_COPY) != 0 &&
4065 (entry->max_protection & VM_PROT_WRITE) == 0 &&
4066 (entry->eflags & MAP_ENTRY_COW) == 0) {
4067 vm_map_unlock_read(map);
4068 return (KERN_PROTECTION_FAILURE);
4072 * If this page is not pageable, we have to get it for all possible
4075 *wired = (entry->wired_count != 0);
4077 fault_type = entry->protection;
4078 size = entry->end - entry->start;
4080 * If the entry was copy-on-write, we either ...
4082 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
4084 * If we want to write the page, we may as well handle that
4085 * now since we've got the map locked.
4087 * If we don't need to write the page, we just demote the
4088 * permissions allowed.
4090 if ((fault_type & VM_PROT_WRITE) != 0 ||
4091 (fault_typea & VM_PROT_COPY) != 0) {
4093 * Make a new object, and place it in the object
4094 * chain. Note that no new references have appeared
4095 * -- one just moved from the map to the new
4098 if (vm_map_lock_upgrade(map))
4101 if (entry->cred == NULL) {
4103 * The debugger owner is charged for
4106 cred = curthread->td_ucred;
4108 if (!swap_reserve_by_cred(size, cred)) {
4111 return (KERN_RESOURCE_SHORTAGE);
4115 vm_object_shadow(&entry->object.vm_object,
4116 &entry->offset, size);
4117 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
4118 eobject = entry->object.vm_object;
4119 if (eobject->cred != NULL) {
4121 * The object was not shadowed.
4123 swap_release_by_cred(size, entry->cred);
4124 crfree(entry->cred);
4126 } else if (entry->cred != NULL) {
4127 VM_OBJECT_WLOCK(eobject);
4128 eobject->cred = entry->cred;
4129 eobject->charge = size;
4130 VM_OBJECT_WUNLOCK(eobject);
4134 vm_map_lock_downgrade(map);
4137 * We're attempting to read a copy-on-write page --
4138 * don't allow writes.
4140 prot &= ~VM_PROT_WRITE;
4145 * Create an object if necessary.
4147 if (entry->object.vm_object == NULL &&
4149 if (vm_map_lock_upgrade(map))
4151 entry->object.vm_object = vm_object_allocate(OBJT_DEFAULT,
4154 if (entry->cred != NULL) {
4155 VM_OBJECT_WLOCK(entry->object.vm_object);
4156 entry->object.vm_object->cred = entry->cred;
4157 entry->object.vm_object->charge = size;
4158 VM_OBJECT_WUNLOCK(entry->object.vm_object);
4161 vm_map_lock_downgrade(map);
4165 * Return the object/offset from this entry. If the entry was
4166 * copy-on-write or empty, it has been fixed up.
4168 *pindex = UOFF_TO_IDX((vaddr - entry->start) + entry->offset);
4169 *object = entry->object.vm_object;
4172 return (KERN_SUCCESS);
4176 * vm_map_lookup_locked:
4178 * Lookup the faulting address. A version of vm_map_lookup that returns
4179 * KERN_FAILURE instead of blocking on map lock or memory allocation.
4182 vm_map_lookup_locked(vm_map_t *var_map, /* IN/OUT */
4184 vm_prot_t fault_typea,
4185 vm_map_entry_t *out_entry, /* OUT */
4186 vm_object_t *object, /* OUT */
4187 vm_pindex_t *pindex, /* OUT */
4188 vm_prot_t *out_prot, /* OUT */
4189 boolean_t *wired) /* OUT */
4191 vm_map_entry_t entry;
4192 vm_map_t map = *var_map;
4194 vm_prot_t fault_type = fault_typea;
4197 * Lookup the faulting address.
4199 if (!vm_map_lookup_entry(map, vaddr, out_entry))
4200 return (KERN_INVALID_ADDRESS);
4205 * Fail if the entry refers to a submap.
4207 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP)
4208 return (KERN_FAILURE);
4211 * Check whether this task is allowed to have this page.
4213 prot = entry->protection;
4214 fault_type &= VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE;
4215 if ((fault_type & prot) != fault_type)
4216 return (KERN_PROTECTION_FAILURE);
4219 * If this page is not pageable, we have to get it for all possible
4222 *wired = (entry->wired_count != 0);
4224 fault_type = entry->protection;
4226 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
4228 * Fail if the entry was copy-on-write for a write fault.
4230 if (fault_type & VM_PROT_WRITE)
4231 return (KERN_FAILURE);
4233 * We're attempting to read a copy-on-write page --
4234 * don't allow writes.
4236 prot &= ~VM_PROT_WRITE;
4240 * Fail if an object should be created.
4242 if (entry->object.vm_object == NULL && !map->system_map)
4243 return (KERN_FAILURE);
4246 * Return the object/offset from this entry. If the entry was
4247 * copy-on-write or empty, it has been fixed up.
4249 *pindex = UOFF_TO_IDX((vaddr - entry->start) + entry->offset);
4250 *object = entry->object.vm_object;
4253 return (KERN_SUCCESS);
4257 * vm_map_lookup_done:
4259 * Releases locks acquired by a vm_map_lookup
4260 * (according to the handle returned by that lookup).
4263 vm_map_lookup_done(vm_map_t map, vm_map_entry_t entry)
4266 * Unlock the main-level map
4268 vm_map_unlock_read(map);
4271 #include "opt_ddb.h"
4273 #include <sys/kernel.h>
4275 #include <ddb/ddb.h>
4278 vm_map_print(vm_map_t map)
4280 vm_map_entry_t entry;
4282 db_iprintf("Task map %p: pmap=%p, nentries=%d, version=%u\n",
4284 (void *)map->pmap, map->nentries, map->timestamp);
4287 for (entry = map->header.next; entry != &map->header;
4288 entry = entry->next) {
4289 db_iprintf("map entry %p: start=%p, end=%p, eflags=%#x, \n",
4290 (void *)entry, (void *)entry->start, (void *)entry->end,
4293 static char *inheritance_name[4] =
4294 {"share", "copy", "none", "donate_copy"};
4296 db_iprintf(" prot=%x/%x/%s",
4298 entry->max_protection,
4299 inheritance_name[(int)(unsigned char)entry->inheritance]);
4300 if (entry->wired_count != 0)
4301 db_printf(", wired");
4303 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
4304 db_printf(", share=%p, offset=0x%jx\n",
4305 (void *)entry->object.sub_map,
4306 (uintmax_t)entry->offset);
4307 if ((entry->prev == &map->header) ||
4308 (entry->prev->object.sub_map !=
4309 entry->object.sub_map)) {
4311 vm_map_print((vm_map_t)entry->object.sub_map);
4315 if (entry->cred != NULL)
4316 db_printf(", ruid %d", entry->cred->cr_ruid);
4317 db_printf(", object=%p, offset=0x%jx",
4318 (void *)entry->object.vm_object,
4319 (uintmax_t)entry->offset);
4320 if (entry->object.vm_object && entry->object.vm_object->cred)
4321 db_printf(", obj ruid %d charge %jx",
4322 entry->object.vm_object->cred->cr_ruid,
4323 (uintmax_t)entry->object.vm_object->charge);
4324 if (entry->eflags & MAP_ENTRY_COW)
4325 db_printf(", copy (%s)",
4326 (entry->eflags & MAP_ENTRY_NEEDS_COPY) ? "needed" : "done");
4329 if ((entry->prev == &map->header) ||
4330 (entry->prev->object.vm_object !=
4331 entry->object.vm_object)) {
4333 vm_object_print((db_expr_t)(intptr_t)
4334 entry->object.vm_object,
4343 DB_SHOW_COMMAND(map, map)
4347 db_printf("usage: show map <addr>\n");
4350 vm_map_print((vm_map_t)addr);
4353 DB_SHOW_COMMAND(procvm, procvm)
4358 p = db_lookup_proc(addr);
4363 db_printf("p = %p, vmspace = %p, map = %p, pmap = %p\n",
4364 (void *)p, (void *)p->p_vmspace, (void *)&p->p_vmspace->vm_map,
4365 (void *)vmspace_pmap(p->p_vmspace));
4367 vm_map_print((vm_map_t)&p->p_vmspace->vm_map);