2 * Copyright (c) 1991, 1993
3 * The Regents of the University of California. All rights reserved.
5 * This code is derived from software contributed to Berkeley by
6 * The Mach Operating System project at Carnegie-Mellon University.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 4. Neither the name of the University nor the names of its contributors
17 * may be used to endorse or promote products derived from this software
18 * without specific prior written permission.
20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * from: @(#)vm_map.c 8.3 (Berkeley) 1/12/94
35 * Copyright (c) 1987, 1990 Carnegie-Mellon University.
36 * All rights reserved.
38 * Authors: Avadis Tevanian, Jr., Michael Wayne Young
40 * Permission to use, copy, modify and distribute this software and
41 * its documentation is hereby granted, provided that both the copyright
42 * notice and this permission notice appear in all copies of the
43 * software, derivative works or modified versions, and any portions
44 * thereof, and that both notices appear in supporting documentation.
46 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
47 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
48 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
50 * Carnegie Mellon requests users of this software to return to
52 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
53 * School of Computer Science
54 * Carnegie Mellon University
55 * Pittsburgh PA 15213-3890
57 * any improvements or extensions that they make and grant Carnegie the
58 * rights to redistribute these changes.
62 * Virtual memory mapping module.
65 #include <sys/cdefs.h>
66 __FBSDID("$FreeBSD$");
68 #include <sys/param.h>
69 #include <sys/systm.h>
70 #include <sys/kernel.h>
73 #include <sys/mutex.h>
75 #include <sys/vmmeter.h>
77 #include <sys/vnode.h>
78 #include <sys/racct.h>
79 #include <sys/resourcevar.h>
80 #include <sys/rwlock.h>
82 #include <sys/sysctl.h>
83 #include <sys/sysent.h>
87 #include <vm/vm_param.h>
89 #include <vm/vm_map.h>
90 #include <vm/vm_page.h>
91 #include <vm/vm_object.h>
92 #include <vm/vm_pager.h>
93 #include <vm/vm_kern.h>
94 #include <vm/vm_extern.h>
95 #include <vm/vnode_pager.h>
96 #include <vm/swap_pager.h>
100 * Virtual memory maps provide for the mapping, protection,
101 * and sharing of virtual memory objects. In addition,
102 * this module provides for an efficient virtual copy of
103 * memory from one map to another.
105 * Synchronization is required prior to most operations.
107 * Maps consist of an ordered doubly-linked list of simple
108 * entries; a self-adjusting binary search tree of these
109 * entries is used to speed up lookups.
111 * Since portions of maps are specified by start/end addresses,
112 * which may not align with existing map entries, all
113 * routines merely "clip" entries to these start/end values.
114 * [That is, an entry is split into two, bordering at a
115 * start or end value.] Note that these clippings may not
116 * always be necessary (as the two resulting entries are then
117 * not changed); however, the clipping is done for convenience.
119 * As mentioned above, virtual copy operations are performed
120 * by copying VM object references from one map to
121 * another, and then marking both regions as copy-on-write.
124 static struct mtx map_sleep_mtx;
125 static uma_zone_t mapentzone;
126 static uma_zone_t kmapentzone;
127 static uma_zone_t mapzone;
128 static uma_zone_t vmspace_zone;
129 static int vmspace_zinit(void *mem, int size, int flags);
130 static int vm_map_zinit(void *mem, int ize, int flags);
131 static void _vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min,
133 static void vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map);
134 static void vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry);
135 static void vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry);
136 static void vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot,
137 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags);
139 static void vm_map_zdtor(void *mem, int size, void *arg);
140 static void vmspace_zdtor(void *mem, int size, void *arg);
142 static int vm_map_stack_locked(vm_map_t map, vm_offset_t addrbos,
143 vm_size_t max_ssize, vm_size_t growsize, vm_prot_t prot, vm_prot_t max,
145 static void vm_map_wire_entry_failure(vm_map_t map, vm_map_entry_t entry,
146 vm_offset_t failed_addr);
148 #define ENTRY_CHARGED(e) ((e)->cred != NULL || \
149 ((e)->object.vm_object != NULL && (e)->object.vm_object->cred != NULL && \
150 !((e)->eflags & MAP_ENTRY_NEEDS_COPY)))
153 * PROC_VMSPACE_{UN,}LOCK() can be a noop as long as vmspaces are type
156 #define PROC_VMSPACE_LOCK(p) do { } while (0)
157 #define PROC_VMSPACE_UNLOCK(p) do { } while (0)
160 * VM_MAP_RANGE_CHECK: [ internal use only ]
162 * Asserts that the starting and ending region
163 * addresses fall within the valid range of the map.
165 #define VM_MAP_RANGE_CHECK(map, start, end) \
167 if (start < vm_map_min(map)) \
168 start = vm_map_min(map); \
169 if (end > vm_map_max(map)) \
170 end = vm_map_max(map); \
178 * Initialize the vm_map module. Must be called before
179 * any other vm_map routines.
181 * Map and entry structures are allocated from the general
182 * purpose memory pool with some exceptions:
184 * - The kernel map and kmem submap are allocated statically.
185 * - Kernel map entries are allocated out of a static pool.
187 * These restrictions are necessary since malloc() uses the
188 * maps and requires map entries.
194 mtx_init(&map_sleep_mtx, "vm map sleep mutex", NULL, MTX_DEF);
195 mapzone = uma_zcreate("MAP", sizeof(struct vm_map), NULL,
201 vm_map_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
202 uma_prealloc(mapzone, MAX_KMAP);
203 kmapentzone = uma_zcreate("KMAP ENTRY", sizeof(struct vm_map_entry),
204 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR,
205 UMA_ZONE_MTXCLASS | UMA_ZONE_VM);
206 mapentzone = uma_zcreate("MAP ENTRY", sizeof(struct vm_map_entry),
207 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
208 vmspace_zone = uma_zcreate("VMSPACE", sizeof(struct vmspace), NULL,
214 vmspace_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
218 vmspace_zinit(void *mem, int size, int flags)
222 vm = (struct vmspace *)mem;
224 vm->vm_map.pmap = NULL;
225 (void)vm_map_zinit(&vm->vm_map, sizeof(vm->vm_map), flags);
226 PMAP_LOCK_INIT(vmspace_pmap(vm));
231 vm_map_zinit(void *mem, int size, int flags)
236 memset(map, 0, sizeof(*map));
237 mtx_init(&map->system_mtx, "vm map (system)", NULL, MTX_DEF | MTX_DUPOK);
238 sx_init(&map->lock, "vm map (user)");
244 vmspace_zdtor(void *mem, int size, void *arg)
248 vm = (struct vmspace *)mem;
250 vm_map_zdtor(&vm->vm_map, sizeof(vm->vm_map), arg);
253 vm_map_zdtor(void *mem, int size, void *arg)
258 KASSERT(map->nentries == 0,
259 ("map %p nentries == %d on free.",
260 map, map->nentries));
261 KASSERT(map->size == 0,
262 ("map %p size == %lu on free.",
263 map, (unsigned long)map->size));
265 #endif /* INVARIANTS */
268 * Allocate a vmspace structure, including a vm_map and pmap,
269 * and initialize those structures. The refcnt is set to 1.
271 * If 'pinit' is NULL then the embedded pmap is initialized via pmap_pinit().
274 vmspace_alloc(vm_offset_t min, vm_offset_t max, pmap_pinit_t pinit)
278 vm = uma_zalloc(vmspace_zone, M_WAITOK);
280 KASSERT(vm->vm_map.pmap == NULL, ("vm_map.pmap must be NULL"));
285 if (!pinit(vmspace_pmap(vm))) {
286 uma_zfree(vmspace_zone, vm);
289 CTR1(KTR_VM, "vmspace_alloc: %p", vm);
290 _vm_map_init(&vm->vm_map, vmspace_pmap(vm), min, max);
305 vmspace_container_reset(struct proc *p)
309 racct_set(p, RACCT_DATA, 0);
310 racct_set(p, RACCT_STACK, 0);
311 racct_set(p, RACCT_RSS, 0);
312 racct_set(p, RACCT_MEMLOCK, 0);
313 racct_set(p, RACCT_VMEM, 0);
319 vmspace_dofree(struct vmspace *vm)
322 CTR1(KTR_VM, "vmspace_free: %p", vm);
325 * Make sure any SysV shm is freed, it might not have been in
331 * Lock the map, to wait out all other references to it.
332 * Delete all of the mappings and pages they hold, then call
333 * the pmap module to reclaim anything left.
335 (void)vm_map_remove(&vm->vm_map, vm->vm_map.min_offset,
336 vm->vm_map.max_offset);
338 pmap_release(vmspace_pmap(vm));
339 vm->vm_map.pmap = NULL;
340 uma_zfree(vmspace_zone, vm);
344 vmspace_free(struct vmspace *vm)
347 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
348 "vmspace_free() called");
350 if (vm->vm_refcnt == 0)
351 panic("vmspace_free: attempt to free already freed vmspace");
353 if (atomic_fetchadd_int(&vm->vm_refcnt, -1) == 1)
358 vmspace_exitfree(struct proc *p)
362 PROC_VMSPACE_LOCK(p);
365 PROC_VMSPACE_UNLOCK(p);
366 KASSERT(vm == &vmspace0, ("vmspace_exitfree: wrong vmspace"));
371 vmspace_exit(struct thread *td)
378 * Release user portion of address space.
379 * This releases references to vnodes,
380 * which could cause I/O if the file has been unlinked.
381 * Need to do this early enough that we can still sleep.
383 * The last exiting process to reach this point releases as
384 * much of the environment as it can. vmspace_dofree() is the
385 * slower fallback in case another process had a temporary
386 * reference to the vmspace.
391 atomic_add_int(&vmspace0.vm_refcnt, 1);
393 refcnt = vm->vm_refcnt;
394 if (refcnt > 1 && p->p_vmspace != &vmspace0) {
395 /* Switch now since other proc might free vmspace */
396 PROC_VMSPACE_LOCK(p);
397 p->p_vmspace = &vmspace0;
398 PROC_VMSPACE_UNLOCK(p);
401 } while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt - 1));
403 if (p->p_vmspace != vm) {
404 /* vmspace not yet freed, switch back */
405 PROC_VMSPACE_LOCK(p);
407 PROC_VMSPACE_UNLOCK(p);
410 pmap_remove_pages(vmspace_pmap(vm));
411 /* Switch now since this proc will free vmspace */
412 PROC_VMSPACE_LOCK(p);
413 p->p_vmspace = &vmspace0;
414 PROC_VMSPACE_UNLOCK(p);
420 vmspace_container_reset(p);
424 /* Acquire reference to vmspace owned by another process. */
427 vmspace_acquire_ref(struct proc *p)
432 PROC_VMSPACE_LOCK(p);
435 PROC_VMSPACE_UNLOCK(p);
439 refcnt = vm->vm_refcnt;
440 if (refcnt <= 0) { /* Avoid 0->1 transition */
441 PROC_VMSPACE_UNLOCK(p);
444 } while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt + 1));
445 if (vm != p->p_vmspace) {
446 PROC_VMSPACE_UNLOCK(p);
450 PROC_VMSPACE_UNLOCK(p);
455 * Switch between vmspaces in an AIO kernel process.
457 * The AIO kernel processes switch to and from a user process's
458 * vmspace while performing an I/O operation on behalf of a user
459 * process. The new vmspace is either the vmspace of a user process
460 * obtained from an active AIO request or the initial vmspace of the
461 * AIO kernel process (when it is idling). Because user processes
462 * will block to drain any active AIO requests before proceeding in
463 * exit() or execve(), the vmspace reference count for these vmspaces
464 * can never be 0. This allows for a much simpler implementation than
465 * the loop in vmspace_acquire_ref() above. Similarly, AIO kernel
466 * processes hold an extra reference on their initial vmspace for the
467 * life of the process so that this guarantee is true for any vmspace
471 vmspace_switch_aio(struct vmspace *newvm)
473 struct vmspace *oldvm;
475 /* XXX: Need some way to assert that this is an aio daemon. */
477 KASSERT(newvm->vm_refcnt > 0,
478 ("vmspace_switch_aio: newvm unreferenced"));
480 oldvm = curproc->p_vmspace;
485 * Point to the new address space and refer to it.
487 curproc->p_vmspace = newvm;
488 atomic_add_int(&newvm->vm_refcnt, 1);
490 /* Activate the new mapping. */
491 pmap_activate(curthread);
493 /* Remove the daemon's reference to the old address space. */
494 KASSERT(oldvm->vm_refcnt > 1,
495 ("vmspace_switch_aio: oldvm dropping last reference"));
500 _vm_map_lock(vm_map_t map, const char *file, int line)
504 mtx_lock_flags_(&map->system_mtx, 0, file, line);
506 sx_xlock_(&map->lock, file, line);
511 vm_map_process_deferred(void)
514 vm_map_entry_t entry, next;
518 entry = td->td_map_def_user;
519 td->td_map_def_user = NULL;
520 while (entry != NULL) {
522 if ((entry->eflags & MAP_ENTRY_VN_WRITECNT) != 0) {
524 * Decrement the object's writemappings and
525 * possibly the vnode's v_writecount.
527 KASSERT((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0,
528 ("Submap with writecount"));
529 object = entry->object.vm_object;
530 KASSERT(object != NULL, ("No object for writecount"));
531 vnode_pager_release_writecount(object, entry->start,
534 vm_map_entry_deallocate(entry, FALSE);
540 _vm_map_unlock(vm_map_t map, const char *file, int line)
544 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
546 sx_xunlock_(&map->lock, file, line);
547 vm_map_process_deferred();
552 _vm_map_lock_read(vm_map_t map, const char *file, int line)
556 mtx_lock_flags_(&map->system_mtx, 0, file, line);
558 sx_slock_(&map->lock, file, line);
562 _vm_map_unlock_read(vm_map_t map, const char *file, int line)
566 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
568 sx_sunlock_(&map->lock, file, line);
569 vm_map_process_deferred();
574 _vm_map_trylock(vm_map_t map, const char *file, int line)
578 error = map->system_map ?
579 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) :
580 !sx_try_xlock_(&map->lock, file, line);
587 _vm_map_trylock_read(vm_map_t map, const char *file, int line)
591 error = map->system_map ?
592 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) :
593 !sx_try_slock_(&map->lock, file, line);
598 * _vm_map_lock_upgrade: [ internal use only ]
600 * Tries to upgrade a read (shared) lock on the specified map to a write
601 * (exclusive) lock. Returns the value "0" if the upgrade succeeds and a
602 * non-zero value if the upgrade fails. If the upgrade fails, the map is
603 * returned without a read or write lock held.
605 * Requires that the map be read locked.
608 _vm_map_lock_upgrade(vm_map_t map, const char *file, int line)
610 unsigned int last_timestamp;
612 if (map->system_map) {
613 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
615 if (!sx_try_upgrade_(&map->lock, file, line)) {
616 last_timestamp = map->timestamp;
617 sx_sunlock_(&map->lock, file, line);
618 vm_map_process_deferred();
620 * If the map's timestamp does not change while the
621 * map is unlocked, then the upgrade succeeds.
623 sx_xlock_(&map->lock, file, line);
624 if (last_timestamp != map->timestamp) {
625 sx_xunlock_(&map->lock, file, line);
635 _vm_map_lock_downgrade(vm_map_t map, const char *file, int line)
638 if (map->system_map) {
639 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
641 sx_downgrade_(&map->lock, file, line);
647 * Returns a non-zero value if the caller holds a write (exclusive) lock
648 * on the specified map and the value "0" otherwise.
651 vm_map_locked(vm_map_t map)
655 return (mtx_owned(&map->system_mtx));
657 return (sx_xlocked(&map->lock));
662 _vm_map_assert_locked(vm_map_t map, const char *file, int line)
666 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
668 sx_assert_(&map->lock, SA_XLOCKED, file, line);
671 #define VM_MAP_ASSERT_LOCKED(map) \
672 _vm_map_assert_locked(map, LOCK_FILE, LOCK_LINE)
674 #define VM_MAP_ASSERT_LOCKED(map)
678 * _vm_map_unlock_and_wait:
680 * Atomically releases the lock on the specified map and puts the calling
681 * thread to sleep. The calling thread will remain asleep until either
682 * vm_map_wakeup() is performed on the map or the specified timeout is
685 * WARNING! This function does not perform deferred deallocations of
686 * objects and map entries. Therefore, the calling thread is expected to
687 * reacquire the map lock after reawakening and later perform an ordinary
688 * unlock operation, such as vm_map_unlock(), before completing its
689 * operation on the map.
692 _vm_map_unlock_and_wait(vm_map_t map, int timo, const char *file, int line)
695 mtx_lock(&map_sleep_mtx);
697 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
699 sx_xunlock_(&map->lock, file, line);
700 return (msleep(&map->root, &map_sleep_mtx, PDROP | PVM, "vmmaps",
707 * Awaken any threads that have slept on the map using
708 * vm_map_unlock_and_wait().
711 vm_map_wakeup(vm_map_t map)
715 * Acquire and release map_sleep_mtx to prevent a wakeup()
716 * from being performed (and lost) between the map unlock
717 * and the msleep() in _vm_map_unlock_and_wait().
719 mtx_lock(&map_sleep_mtx);
720 mtx_unlock(&map_sleep_mtx);
725 vm_map_busy(vm_map_t map)
728 VM_MAP_ASSERT_LOCKED(map);
733 vm_map_unbusy(vm_map_t map)
736 VM_MAP_ASSERT_LOCKED(map);
737 KASSERT(map->busy, ("vm_map_unbusy: not busy"));
738 if (--map->busy == 0 && (map->flags & MAP_BUSY_WAKEUP)) {
739 vm_map_modflags(map, 0, MAP_BUSY_WAKEUP);
745 vm_map_wait_busy(vm_map_t map)
748 VM_MAP_ASSERT_LOCKED(map);
750 vm_map_modflags(map, MAP_BUSY_WAKEUP, 0);
752 msleep(&map->busy, &map->system_mtx, 0, "mbusy", 0);
754 sx_sleep(&map->busy, &map->lock, 0, "mbusy", 0);
760 vmspace_resident_count(struct vmspace *vmspace)
762 return pmap_resident_count(vmspace_pmap(vmspace));
768 * Creates and returns a new empty VM map with
769 * the given physical map structure, and having
770 * the given lower and upper address bounds.
773 vm_map_create(pmap_t pmap, vm_offset_t min, vm_offset_t max)
777 result = uma_zalloc(mapzone, M_WAITOK);
778 CTR1(KTR_VM, "vm_map_create: %p", result);
779 _vm_map_init(result, pmap, min, max);
784 * Initialize an existing vm_map structure
785 * such as that in the vmspace structure.
788 _vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max)
791 map->header.next = map->header.prev = &map->header;
792 map->needs_wakeup = FALSE;
795 map->min_offset = min;
796 map->max_offset = max;
804 vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max)
807 _vm_map_init(map, pmap, min, max);
808 mtx_init(&map->system_mtx, "system map", NULL, MTX_DEF | MTX_DUPOK);
809 sx_init(&map->lock, "user map");
813 * vm_map_entry_dispose: [ internal use only ]
815 * Inverse of vm_map_entry_create.
818 vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry)
820 uma_zfree(map->system_map ? kmapentzone : mapentzone, entry);
824 * vm_map_entry_create: [ internal use only ]
826 * Allocates a VM map entry for insertion.
827 * No entry fields are filled in.
829 static vm_map_entry_t
830 vm_map_entry_create(vm_map_t map)
832 vm_map_entry_t new_entry;
835 new_entry = uma_zalloc(kmapentzone, M_NOWAIT);
837 new_entry = uma_zalloc(mapentzone, M_WAITOK);
838 if (new_entry == NULL)
839 panic("vm_map_entry_create: kernel resources exhausted");
844 * vm_map_entry_set_behavior:
846 * Set the expected access behavior, either normal, random, or
850 vm_map_entry_set_behavior(vm_map_entry_t entry, u_char behavior)
852 entry->eflags = (entry->eflags & ~MAP_ENTRY_BEHAV_MASK) |
853 (behavior & MAP_ENTRY_BEHAV_MASK);
857 * vm_map_entry_set_max_free:
859 * Set the max_free field in a vm_map_entry.
862 vm_map_entry_set_max_free(vm_map_entry_t entry)
865 entry->max_free = entry->adj_free;
866 if (entry->left != NULL && entry->left->max_free > entry->max_free)
867 entry->max_free = entry->left->max_free;
868 if (entry->right != NULL && entry->right->max_free > entry->max_free)
869 entry->max_free = entry->right->max_free;
873 * vm_map_entry_splay:
875 * The Sleator and Tarjan top-down splay algorithm with the
876 * following variation. Max_free must be computed bottom-up, so
877 * on the downward pass, maintain the left and right spines in
878 * reverse order. Then, make a second pass up each side to fix
879 * the pointers and compute max_free. The time bound is O(log n)
882 * The new root is the vm_map_entry containing "addr", or else an
883 * adjacent entry (lower or higher) if addr is not in the tree.
885 * The map must be locked, and leaves it so.
887 * Returns: the new root.
889 static vm_map_entry_t
890 vm_map_entry_splay(vm_offset_t addr, vm_map_entry_t root)
892 vm_map_entry_t llist, rlist;
893 vm_map_entry_t ltree, rtree;
896 /* Special case of empty tree. */
901 * Pass One: Splay down the tree until we find addr or a NULL
902 * pointer where addr would go. llist and rlist are the two
903 * sides in reverse order (bottom-up), with llist linked by
904 * the right pointer and rlist linked by the left pointer in
905 * the vm_map_entry. Wait until Pass Two to set max_free on
911 /* root is never NULL in here. */
912 if (addr < root->start) {
916 if (addr < y->start && y->left != NULL) {
917 /* Rotate right and put y on rlist. */
918 root->left = y->right;
920 vm_map_entry_set_max_free(root);
925 /* Put root on rlist. */
930 } else if (addr >= root->end) {
934 if (addr >= y->end && y->right != NULL) {
935 /* Rotate left and put y on llist. */
936 root->right = y->left;
938 vm_map_entry_set_max_free(root);
943 /* Put root on llist. */
953 * Pass Two: Walk back up the two spines, flip the pointers
954 * and set max_free. The subtrees of the root go at the
955 * bottom of llist and rlist.
958 while (llist != NULL) {
960 llist->right = ltree;
961 vm_map_entry_set_max_free(llist);
966 while (rlist != NULL) {
969 vm_map_entry_set_max_free(rlist);
975 * Final assembly: add ltree and rtree as subtrees of root.
979 vm_map_entry_set_max_free(root);
985 * vm_map_entry_{un,}link:
987 * Insert/remove entries from maps.
990 vm_map_entry_link(vm_map_t map,
991 vm_map_entry_t after_where,
992 vm_map_entry_t entry)
996 "vm_map_entry_link: map %p, nentries %d, entry %p, after %p", map,
997 map->nentries, entry, after_where);
998 VM_MAP_ASSERT_LOCKED(map);
999 KASSERT(after_where == &map->header ||
1000 after_where->end <= entry->start,
1001 ("vm_map_entry_link: prev end %jx new start %jx overlap",
1002 (uintmax_t)after_where->end, (uintmax_t)entry->start));
1003 KASSERT(after_where->next == &map->header ||
1004 entry->end <= after_where->next->start,
1005 ("vm_map_entry_link: new end %jx next start %jx overlap",
1006 (uintmax_t)entry->end, (uintmax_t)after_where->next->start));
1009 entry->prev = after_where;
1010 entry->next = after_where->next;
1011 entry->next->prev = entry;
1012 after_where->next = entry;
1014 if (after_where != &map->header) {
1015 if (after_where != map->root)
1016 vm_map_entry_splay(after_where->start, map->root);
1017 entry->right = after_where->right;
1018 entry->left = after_where;
1019 after_where->right = NULL;
1020 after_where->adj_free = entry->start - after_where->end;
1021 vm_map_entry_set_max_free(after_where);
1023 entry->right = map->root;
1026 entry->adj_free = (entry->next == &map->header ? map->max_offset :
1027 entry->next->start) - entry->end;
1028 vm_map_entry_set_max_free(entry);
1033 vm_map_entry_unlink(vm_map_t map,
1034 vm_map_entry_t entry)
1036 vm_map_entry_t next, prev, root;
1038 VM_MAP_ASSERT_LOCKED(map);
1039 if (entry != map->root)
1040 vm_map_entry_splay(entry->start, map->root);
1041 if (entry->left == NULL)
1042 root = entry->right;
1044 root = vm_map_entry_splay(entry->start, entry->left);
1045 root->right = entry->right;
1046 root->adj_free = (entry->next == &map->header ? map->max_offset :
1047 entry->next->start) - root->end;
1048 vm_map_entry_set_max_free(root);
1057 CTR3(KTR_VM, "vm_map_entry_unlink: map %p, nentries %d, entry %p", map,
1058 map->nentries, entry);
1062 * vm_map_entry_resize_free:
1064 * Recompute the amount of free space following a vm_map_entry
1065 * and propagate that value up the tree. Call this function after
1066 * resizing a map entry in-place, that is, without a call to
1067 * vm_map_entry_link() or _unlink().
1069 * The map must be locked, and leaves it so.
1072 vm_map_entry_resize_free(vm_map_t map, vm_map_entry_t entry)
1076 * Using splay trees without parent pointers, propagating
1077 * max_free up the tree is done by moving the entry to the
1078 * root and making the change there.
1080 if (entry != map->root)
1081 map->root = vm_map_entry_splay(entry->start, map->root);
1083 entry->adj_free = (entry->next == &map->header ? map->max_offset :
1084 entry->next->start) - entry->end;
1085 vm_map_entry_set_max_free(entry);
1089 * vm_map_lookup_entry: [ internal use only ]
1091 * Finds the map entry containing (or
1092 * immediately preceding) the specified address
1093 * in the given map; the entry is returned
1094 * in the "entry" parameter. The boolean
1095 * result indicates whether the address is
1096 * actually contained in the map.
1099 vm_map_lookup_entry(
1101 vm_offset_t address,
1102 vm_map_entry_t *entry) /* OUT */
1108 * If the map is empty, then the map entry immediately preceding
1109 * "address" is the map's header.
1113 *entry = &map->header;
1114 else if (address >= cur->start && cur->end > address) {
1117 } else if ((locked = vm_map_locked(map)) ||
1118 sx_try_upgrade(&map->lock)) {
1120 * Splay requires a write lock on the map. However, it only
1121 * restructures the binary search tree; it does not otherwise
1122 * change the map. Thus, the map's timestamp need not change
1123 * on a temporary upgrade.
1125 map->root = cur = vm_map_entry_splay(address, cur);
1127 sx_downgrade(&map->lock);
1130 * If "address" is contained within a map entry, the new root
1131 * is that map entry. Otherwise, the new root is a map entry
1132 * immediately before or after "address".
1134 if (address >= cur->start) {
1136 if (cur->end > address)
1142 * Since the map is only locked for read access, perform a
1143 * standard binary search tree lookup for "address".
1146 if (address < cur->start) {
1147 if (cur->left == NULL) {
1152 } else if (cur->end > address) {
1156 if (cur->right == NULL) {
1169 * Inserts the given whole VM object into the target
1170 * map at the specified address range. The object's
1171 * size should match that of the address range.
1173 * Requires that the map be locked, and leaves it so.
1175 * If object is non-NULL, ref count must be bumped by caller
1176 * prior to making call to account for the new entry.
1179 vm_map_insert(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1180 vm_offset_t start, vm_offset_t end, vm_prot_t prot, vm_prot_t max, int cow)
1182 vm_map_entry_t new_entry, prev_entry, temp_entry;
1184 vm_eflags_t protoeflags;
1185 vm_inherit_t inheritance;
1187 VM_MAP_ASSERT_LOCKED(map);
1188 KASSERT((object != kmem_object && object != kernel_object) ||
1189 (cow & MAP_COPY_ON_WRITE) == 0,
1190 ("vm_map_insert: kmem or kernel object and COW"));
1191 KASSERT(object == NULL || (cow & MAP_NOFAULT) == 0,
1192 ("vm_map_insert: paradoxical MAP_NOFAULT request"));
1195 * Check that the start and end points are not bogus.
1197 if (start < map->min_offset || end > map->max_offset || start >= end)
1198 return (KERN_INVALID_ADDRESS);
1201 * Find the entry prior to the proposed starting address; if it's part
1202 * of an existing entry, this range is bogus.
1204 if (vm_map_lookup_entry(map, start, &temp_entry))
1205 return (KERN_NO_SPACE);
1207 prev_entry = temp_entry;
1210 * Assert that the next entry doesn't overlap the end point.
1212 if (prev_entry->next != &map->header && prev_entry->next->start < end)
1213 return (KERN_NO_SPACE);
1216 if (cow & MAP_COPY_ON_WRITE)
1217 protoeflags |= MAP_ENTRY_COW | MAP_ENTRY_NEEDS_COPY;
1218 if (cow & MAP_NOFAULT)
1219 protoeflags |= MAP_ENTRY_NOFAULT;
1220 if (cow & MAP_DISABLE_SYNCER)
1221 protoeflags |= MAP_ENTRY_NOSYNC;
1222 if (cow & MAP_DISABLE_COREDUMP)
1223 protoeflags |= MAP_ENTRY_NOCOREDUMP;
1224 if (cow & MAP_STACK_GROWS_DOWN)
1225 protoeflags |= MAP_ENTRY_GROWS_DOWN;
1226 if (cow & MAP_STACK_GROWS_UP)
1227 protoeflags |= MAP_ENTRY_GROWS_UP;
1228 if (cow & MAP_VN_WRITECOUNT)
1229 protoeflags |= MAP_ENTRY_VN_WRITECNT;
1230 if (cow & MAP_INHERIT_SHARE)
1231 inheritance = VM_INHERIT_SHARE;
1233 inheritance = VM_INHERIT_DEFAULT;
1236 if (cow & (MAP_ACC_NO_CHARGE | MAP_NOFAULT))
1238 if ((cow & MAP_ACC_CHARGED) || ((prot & VM_PROT_WRITE) &&
1239 ((protoeflags & MAP_ENTRY_NEEDS_COPY) || object == NULL))) {
1240 if (!(cow & MAP_ACC_CHARGED) && !swap_reserve(end - start))
1241 return (KERN_RESOURCE_SHORTAGE);
1242 KASSERT(object == NULL ||
1243 (protoeflags & MAP_ENTRY_NEEDS_COPY) != 0 ||
1244 object->cred == NULL,
1245 ("overcommit: vm_map_insert o %p", object));
1246 cred = curthread->td_ucred;
1250 /* Expand the kernel pmap, if necessary. */
1251 if (map == kernel_map && end > kernel_vm_end)
1252 pmap_growkernel(end);
1253 if (object != NULL) {
1255 * OBJ_ONEMAPPING must be cleared unless this mapping
1256 * is trivially proven to be the only mapping for any
1257 * of the object's pages. (Object granularity
1258 * reference counting is insufficient to recognize
1259 * aliases with precision.)
1261 VM_OBJECT_WLOCK(object);
1262 if (object->ref_count > 1 || object->shadow_count != 0)
1263 vm_object_clear_flag(object, OBJ_ONEMAPPING);
1264 VM_OBJECT_WUNLOCK(object);
1265 } else if (prev_entry != &map->header &&
1266 prev_entry->eflags == protoeflags &&
1267 (cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 &&
1268 prev_entry->end == start && prev_entry->wired_count == 0 &&
1269 (prev_entry->cred == cred ||
1270 (prev_entry->object.vm_object != NULL &&
1271 prev_entry->object.vm_object->cred == cred)) &&
1272 vm_object_coalesce(prev_entry->object.vm_object,
1274 (vm_size_t)(prev_entry->end - prev_entry->start),
1275 (vm_size_t)(end - prev_entry->end), cred != NULL &&
1276 (protoeflags & MAP_ENTRY_NEEDS_COPY) == 0)) {
1278 * We were able to extend the object. Determine if we
1279 * can extend the previous map entry to include the
1280 * new range as well.
1282 if (prev_entry->inheritance == inheritance &&
1283 prev_entry->protection == prot &&
1284 prev_entry->max_protection == max) {
1285 map->size += end - prev_entry->end;
1286 prev_entry->end = end;
1287 vm_map_entry_resize_free(map, prev_entry);
1288 vm_map_simplify_entry(map, prev_entry);
1289 return (KERN_SUCCESS);
1293 * If we can extend the object but cannot extend the
1294 * map entry, we have to create a new map entry. We
1295 * must bump the ref count on the extended object to
1296 * account for it. object may be NULL.
1298 object = prev_entry->object.vm_object;
1299 offset = prev_entry->offset +
1300 (prev_entry->end - prev_entry->start);
1301 vm_object_reference(object);
1302 if (cred != NULL && object != NULL && object->cred != NULL &&
1303 !(prev_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
1304 /* Object already accounts for this uid. */
1312 * Create a new entry
1314 new_entry = vm_map_entry_create(map);
1315 new_entry->start = start;
1316 new_entry->end = end;
1317 new_entry->cred = NULL;
1319 new_entry->eflags = protoeflags;
1320 new_entry->object.vm_object = object;
1321 new_entry->offset = offset;
1322 new_entry->avail_ssize = 0;
1324 new_entry->inheritance = inheritance;
1325 new_entry->protection = prot;
1326 new_entry->max_protection = max;
1327 new_entry->wired_count = 0;
1328 new_entry->wiring_thread = NULL;
1329 new_entry->read_ahead = VM_FAULT_READ_AHEAD_INIT;
1330 new_entry->next_read = start;
1332 KASSERT(cred == NULL || !ENTRY_CHARGED(new_entry),
1333 ("overcommit: vm_map_insert leaks vm_map %p", new_entry));
1334 new_entry->cred = cred;
1337 * Insert the new entry into the list
1339 vm_map_entry_link(map, prev_entry, new_entry);
1340 map->size += new_entry->end - new_entry->start;
1343 * Try to coalesce the new entry with both the previous and next
1344 * entries in the list. Previously, we only attempted to coalesce
1345 * with the previous entry when object is NULL. Here, we handle the
1346 * other cases, which are less common.
1348 vm_map_simplify_entry(map, new_entry);
1350 if ((cow & (MAP_PREFAULT | MAP_PREFAULT_PARTIAL)) != 0) {
1351 vm_map_pmap_enter(map, start, prot, object, OFF_TO_IDX(offset),
1352 end - start, cow & MAP_PREFAULT_PARTIAL);
1355 return (KERN_SUCCESS);
1361 * Find the first fit (lowest VM address) for "length" free bytes
1362 * beginning at address >= start in the given map.
1364 * In a vm_map_entry, "adj_free" is the amount of free space
1365 * adjacent (higher address) to this entry, and "max_free" is the
1366 * maximum amount of contiguous free space in its subtree. This
1367 * allows finding a free region in one path down the tree, so
1368 * O(log n) amortized with splay trees.
1370 * The map must be locked, and leaves it so.
1372 * Returns: 0 on success, and starting address in *addr,
1373 * 1 if insufficient space.
1376 vm_map_findspace(vm_map_t map, vm_offset_t start, vm_size_t length,
1377 vm_offset_t *addr) /* OUT */
1379 vm_map_entry_t entry;
1383 * Request must fit within min/max VM address and must avoid
1386 if (start < map->min_offset)
1387 start = map->min_offset;
1388 if (start + length > map->max_offset || start + length < start)
1391 /* Empty tree means wide open address space. */
1392 if (map->root == NULL) {
1398 * After splay, if start comes before root node, then there
1399 * must be a gap from start to the root.
1401 map->root = vm_map_entry_splay(start, map->root);
1402 if (start + length <= map->root->start) {
1408 * Root is the last node that might begin its gap before
1409 * start, and this is the last comparison where address
1410 * wrap might be a problem.
1412 st = (start > map->root->end) ? start : map->root->end;
1413 if (length <= map->root->end + map->root->adj_free - st) {
1418 /* With max_free, can immediately tell if no solution. */
1419 entry = map->root->right;
1420 if (entry == NULL || length > entry->max_free)
1424 * Search the right subtree in the order: left subtree, root,
1425 * right subtree (first fit). The previous splay implies that
1426 * all regions in the right subtree have addresses > start.
1428 while (entry != NULL) {
1429 if (entry->left != NULL && entry->left->max_free >= length)
1430 entry = entry->left;
1431 else if (entry->adj_free >= length) {
1435 entry = entry->right;
1438 /* Can't get here, so panic if we do. */
1439 panic("vm_map_findspace: max_free corrupt");
1443 vm_map_fixed(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1444 vm_offset_t start, vm_size_t length, vm_prot_t prot,
1445 vm_prot_t max, int cow)
1450 end = start + length;
1451 KASSERT((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 ||
1453 ("vm_map_fixed: non-NULL backing object for stack"));
1455 VM_MAP_RANGE_CHECK(map, start, end);
1456 if ((cow & MAP_CHECK_EXCL) == 0)
1457 vm_map_delete(map, start, end);
1458 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) != 0) {
1459 result = vm_map_stack_locked(map, start, length, sgrowsiz,
1462 result = vm_map_insert(map, object, offset, start, end,
1470 * vm_map_find finds an unallocated region in the target address
1471 * map with the given length. The search is defined to be
1472 * first-fit from the specified address; the region found is
1473 * returned in the same parameter.
1475 * If object is non-NULL, ref count must be bumped by caller
1476 * prior to making call to account for the new entry.
1479 vm_map_find(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1480 vm_offset_t *addr, /* IN/OUT */
1481 vm_size_t length, vm_offset_t max_addr, int find_space,
1482 vm_prot_t prot, vm_prot_t max, int cow)
1484 vm_offset_t alignment, initial_addr, start;
1487 KASSERT((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 ||
1489 ("vm_map_find: non-NULL backing object for stack"));
1490 if (find_space == VMFS_OPTIMAL_SPACE && (object == NULL ||
1491 (object->flags & OBJ_COLORED) == 0))
1492 find_space = VMFS_ANY_SPACE;
1493 if (find_space >> 8 != 0) {
1494 KASSERT((find_space & 0xff) == 0, ("bad VMFS flags"));
1495 alignment = (vm_offset_t)1 << (find_space >> 8);
1498 initial_addr = *addr;
1500 start = initial_addr;
1503 if (find_space != VMFS_NO_SPACE) {
1504 if (vm_map_findspace(map, start, length, addr) ||
1505 (max_addr != 0 && *addr + length > max_addr)) {
1507 if (find_space == VMFS_OPTIMAL_SPACE) {
1508 find_space = VMFS_ANY_SPACE;
1511 return (KERN_NO_SPACE);
1513 switch (find_space) {
1514 case VMFS_SUPER_SPACE:
1515 case VMFS_OPTIMAL_SPACE:
1516 pmap_align_superpage(object, offset, addr,
1519 case VMFS_ANY_SPACE:
1522 if ((*addr & (alignment - 1)) != 0) {
1523 *addr &= ~(alignment - 1);
1531 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) != 0) {
1532 result = vm_map_stack_locked(map, start, length,
1533 sgrowsiz, prot, max, cow);
1535 result = vm_map_insert(map, object, offset, start,
1536 start + length, prot, max, cow);
1538 } while (result == KERN_NO_SPACE && find_space != VMFS_NO_SPACE &&
1539 find_space != VMFS_ANY_SPACE);
1545 * vm_map_simplify_entry:
1547 * Simplify the given map entry by merging with either neighbor. This
1548 * routine also has the ability to merge with both neighbors.
1550 * The map must be locked.
1552 * This routine guarantees that the passed entry remains valid (though
1553 * possibly extended). When merging, this routine may delete one or
1557 vm_map_simplify_entry(vm_map_t map, vm_map_entry_t entry)
1559 vm_map_entry_t next, prev;
1560 vm_size_t prevsize, esize;
1562 if ((entry->eflags & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP |
1563 MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_IS_SUB_MAP)) != 0)
1567 if (prev != &map->header) {
1568 prevsize = prev->end - prev->start;
1569 if ( (prev->end == entry->start) &&
1570 (prev->object.vm_object == entry->object.vm_object) &&
1571 (!prev->object.vm_object ||
1572 (prev->offset + prevsize == entry->offset)) &&
1573 (prev->eflags == entry->eflags) &&
1574 (prev->protection == entry->protection) &&
1575 (prev->max_protection == entry->max_protection) &&
1576 (prev->inheritance == entry->inheritance) &&
1577 (prev->wired_count == entry->wired_count) &&
1578 (prev->cred == entry->cred)) {
1579 vm_map_entry_unlink(map, prev);
1580 entry->start = prev->start;
1581 entry->offset = prev->offset;
1582 if (entry->prev != &map->header)
1583 vm_map_entry_resize_free(map, entry->prev);
1586 * If the backing object is a vnode object,
1587 * vm_object_deallocate() calls vrele().
1588 * However, vrele() does not lock the vnode
1589 * because the vnode has additional
1590 * references. Thus, the map lock can be kept
1591 * without causing a lock-order reversal with
1594 * Since we count the number of virtual page
1595 * mappings in object->un_pager.vnp.writemappings,
1596 * the writemappings value should not be adjusted
1597 * when the entry is disposed of.
1599 if (prev->object.vm_object)
1600 vm_object_deallocate(prev->object.vm_object);
1601 if (prev->cred != NULL)
1603 vm_map_entry_dispose(map, prev);
1608 if (next != &map->header) {
1609 esize = entry->end - entry->start;
1610 if ((entry->end == next->start) &&
1611 (next->object.vm_object == entry->object.vm_object) &&
1612 (!entry->object.vm_object ||
1613 (entry->offset + esize == next->offset)) &&
1614 (next->eflags == entry->eflags) &&
1615 (next->protection == entry->protection) &&
1616 (next->max_protection == entry->max_protection) &&
1617 (next->inheritance == entry->inheritance) &&
1618 (next->wired_count == entry->wired_count) &&
1619 (next->cred == entry->cred)) {
1620 vm_map_entry_unlink(map, next);
1621 entry->end = next->end;
1622 vm_map_entry_resize_free(map, entry);
1625 * See comment above.
1627 if (next->object.vm_object)
1628 vm_object_deallocate(next->object.vm_object);
1629 if (next->cred != NULL)
1631 vm_map_entry_dispose(map, next);
1636 * vm_map_clip_start: [ internal use only ]
1638 * Asserts that the given entry begins at or after
1639 * the specified address; if necessary,
1640 * it splits the entry into two.
1642 #define vm_map_clip_start(map, entry, startaddr) \
1644 if (startaddr > entry->start) \
1645 _vm_map_clip_start(map, entry, startaddr); \
1649 * This routine is called only when it is known that
1650 * the entry must be split.
1653 _vm_map_clip_start(vm_map_t map, vm_map_entry_t entry, vm_offset_t start)
1655 vm_map_entry_t new_entry;
1657 VM_MAP_ASSERT_LOCKED(map);
1658 KASSERT(entry->end > start && entry->start < start,
1659 ("_vm_map_clip_start: invalid clip of entry %p", entry));
1662 * Split off the front portion -- note that we must insert the new
1663 * entry BEFORE this one, so that this entry has the specified
1666 vm_map_simplify_entry(map, entry);
1669 * If there is no object backing this entry, we might as well create
1670 * one now. If we defer it, an object can get created after the map
1671 * is clipped, and individual objects will be created for the split-up
1672 * map. This is a bit of a hack, but is also about the best place to
1673 * put this improvement.
1675 if (entry->object.vm_object == NULL && !map->system_map) {
1677 object = vm_object_allocate(OBJT_DEFAULT,
1678 atop(entry->end - entry->start));
1679 entry->object.vm_object = object;
1681 if (entry->cred != NULL) {
1682 object->cred = entry->cred;
1683 object->charge = entry->end - entry->start;
1686 } else if (entry->object.vm_object != NULL &&
1687 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) &&
1688 entry->cred != NULL) {
1689 VM_OBJECT_WLOCK(entry->object.vm_object);
1690 KASSERT(entry->object.vm_object->cred == NULL,
1691 ("OVERCOMMIT: vm_entry_clip_start: both cred e %p", entry));
1692 entry->object.vm_object->cred = entry->cred;
1693 entry->object.vm_object->charge = entry->end - entry->start;
1694 VM_OBJECT_WUNLOCK(entry->object.vm_object);
1698 new_entry = vm_map_entry_create(map);
1699 *new_entry = *entry;
1701 new_entry->end = start;
1702 entry->offset += (start - entry->start);
1703 entry->start = start;
1704 if (new_entry->cred != NULL)
1705 crhold(entry->cred);
1707 vm_map_entry_link(map, entry->prev, new_entry);
1709 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1710 vm_object_reference(new_entry->object.vm_object);
1712 * The object->un_pager.vnp.writemappings for the
1713 * object of MAP_ENTRY_VN_WRITECNT type entry shall be
1714 * kept as is here. The virtual pages are
1715 * re-distributed among the clipped entries, so the sum is
1722 * vm_map_clip_end: [ internal use only ]
1724 * Asserts that the given entry ends at or before
1725 * the specified address; if necessary,
1726 * it splits the entry into two.
1728 #define vm_map_clip_end(map, entry, endaddr) \
1730 if ((endaddr) < (entry->end)) \
1731 _vm_map_clip_end((map), (entry), (endaddr)); \
1735 * This routine is called only when it is known that
1736 * the entry must be split.
1739 _vm_map_clip_end(vm_map_t map, vm_map_entry_t entry, vm_offset_t end)
1741 vm_map_entry_t new_entry;
1743 VM_MAP_ASSERT_LOCKED(map);
1744 KASSERT(entry->start < end && entry->end > end,
1745 ("_vm_map_clip_end: invalid clip of entry %p", entry));
1748 * If there is no object backing this entry, we might as well create
1749 * one now. If we defer it, an object can get created after the map
1750 * is clipped, and individual objects will be created for the split-up
1751 * map. This is a bit of a hack, but is also about the best place to
1752 * put this improvement.
1754 if (entry->object.vm_object == NULL && !map->system_map) {
1756 object = vm_object_allocate(OBJT_DEFAULT,
1757 atop(entry->end - entry->start));
1758 entry->object.vm_object = object;
1760 if (entry->cred != NULL) {
1761 object->cred = entry->cred;
1762 object->charge = entry->end - entry->start;
1765 } else if (entry->object.vm_object != NULL &&
1766 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) &&
1767 entry->cred != NULL) {
1768 VM_OBJECT_WLOCK(entry->object.vm_object);
1769 KASSERT(entry->object.vm_object->cred == NULL,
1770 ("OVERCOMMIT: vm_entry_clip_end: both cred e %p", entry));
1771 entry->object.vm_object->cred = entry->cred;
1772 entry->object.vm_object->charge = entry->end - entry->start;
1773 VM_OBJECT_WUNLOCK(entry->object.vm_object);
1778 * Create a new entry and insert it AFTER the specified entry
1780 new_entry = vm_map_entry_create(map);
1781 *new_entry = *entry;
1783 new_entry->start = entry->end = end;
1784 new_entry->offset += (end - entry->start);
1785 if (new_entry->cred != NULL)
1786 crhold(entry->cred);
1788 vm_map_entry_link(map, entry, new_entry);
1790 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1791 vm_object_reference(new_entry->object.vm_object);
1796 * vm_map_submap: [ kernel use only ]
1798 * Mark the given range as handled by a subordinate map.
1800 * This range must have been created with vm_map_find,
1801 * and no other operations may have been performed on this
1802 * range prior to calling vm_map_submap.
1804 * Only a limited number of operations can be performed
1805 * within this rage after calling vm_map_submap:
1807 * [Don't try vm_map_copy!]
1809 * To remove a submapping, one must first remove the
1810 * range from the superior map, and then destroy the
1811 * submap (if desired). [Better yet, don't try it.]
1820 vm_map_entry_t entry;
1821 int result = KERN_INVALID_ARGUMENT;
1825 VM_MAP_RANGE_CHECK(map, start, end);
1827 if (vm_map_lookup_entry(map, start, &entry)) {
1828 vm_map_clip_start(map, entry, start);
1830 entry = entry->next;
1832 vm_map_clip_end(map, entry, end);
1834 if ((entry->start == start) && (entry->end == end) &&
1835 ((entry->eflags & MAP_ENTRY_COW) == 0) &&
1836 (entry->object.vm_object == NULL)) {
1837 entry->object.sub_map = submap;
1838 entry->eflags |= MAP_ENTRY_IS_SUB_MAP;
1839 result = KERN_SUCCESS;
1847 * The maximum number of pages to map if MAP_PREFAULT_PARTIAL is specified
1849 #define MAX_INIT_PT 96
1852 * vm_map_pmap_enter:
1854 * Preload the specified map's pmap with mappings to the specified
1855 * object's memory-resident pages. No further physical pages are
1856 * allocated, and no further virtual pages are retrieved from secondary
1857 * storage. If the specified flags include MAP_PREFAULT_PARTIAL, then a
1858 * limited number of page mappings are created at the low-end of the
1859 * specified address range. (For this purpose, a superpage mapping
1860 * counts as one page mapping.) Otherwise, all resident pages within
1861 * the specified address range are mapped.
1864 vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot,
1865 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags)
1868 vm_page_t p, p_start;
1869 vm_pindex_t mask, psize, threshold, tmpidx;
1871 if ((prot & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0 || object == NULL)
1873 VM_OBJECT_RLOCK(object);
1874 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) {
1875 VM_OBJECT_RUNLOCK(object);
1876 VM_OBJECT_WLOCK(object);
1877 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) {
1878 pmap_object_init_pt(map->pmap, addr, object, pindex,
1880 VM_OBJECT_WUNLOCK(object);
1883 VM_OBJECT_LOCK_DOWNGRADE(object);
1887 if (psize + pindex > object->size) {
1888 if (object->size < pindex) {
1889 VM_OBJECT_RUNLOCK(object);
1892 psize = object->size - pindex;
1897 threshold = MAX_INIT_PT;
1899 p = vm_page_find_least(object, pindex);
1901 * Assert: the variable p is either (1) the page with the
1902 * least pindex greater than or equal to the parameter pindex
1906 p != NULL && (tmpidx = p->pindex - pindex) < psize;
1907 p = TAILQ_NEXT(p, listq)) {
1909 * don't allow an madvise to blow away our really
1910 * free pages allocating pv entries.
1912 if (((flags & MAP_PREFAULT_MADVISE) != 0 &&
1913 vm_cnt.v_free_count < vm_cnt.v_free_reserved) ||
1914 ((flags & MAP_PREFAULT_PARTIAL) != 0 &&
1915 tmpidx >= threshold)) {
1919 if (p->valid == VM_PAGE_BITS_ALL) {
1920 if (p_start == NULL) {
1921 start = addr + ptoa(tmpidx);
1924 /* Jump ahead if a superpage mapping is possible. */
1925 if (p->psind > 0 && ((addr + ptoa(tmpidx)) &
1926 (pagesizes[p->psind] - 1)) == 0) {
1927 mask = atop(pagesizes[p->psind]) - 1;
1928 if (tmpidx + mask < psize &&
1929 vm_page_ps_is_valid(p)) {
1934 } else if (p_start != NULL) {
1935 pmap_enter_object(map->pmap, start, addr +
1936 ptoa(tmpidx), p_start, prot);
1940 if (p_start != NULL)
1941 pmap_enter_object(map->pmap, start, addr + ptoa(psize),
1943 VM_OBJECT_RUNLOCK(object);
1949 * Sets the protection of the specified address
1950 * region in the target map. If "set_max" is
1951 * specified, the maximum protection is to be set;
1952 * otherwise, only the current protection is affected.
1955 vm_map_protect(vm_map_t map, vm_offset_t start, vm_offset_t end,
1956 vm_prot_t new_prot, boolean_t set_max)
1958 vm_map_entry_t current, entry;
1964 return (KERN_SUCCESS);
1969 * Ensure that we are not concurrently wiring pages. vm_map_wire() may
1970 * need to fault pages into the map and will drop the map lock while
1971 * doing so, and the VM object may end up in an inconsistent state if we
1972 * update the protection on the map entry in between faults.
1974 vm_map_wait_busy(map);
1976 VM_MAP_RANGE_CHECK(map, start, end);
1978 if (vm_map_lookup_entry(map, start, &entry)) {
1979 vm_map_clip_start(map, entry, start);
1981 entry = entry->next;
1985 * Make a first pass to check for protection violations.
1987 for (current = entry; current != &map->header && current->start < end;
1988 current = current->next) {
1989 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
1991 return (KERN_INVALID_ARGUMENT);
1993 if ((new_prot & current->max_protection) != new_prot) {
1995 return (KERN_PROTECTION_FAILURE);
2000 * Do an accounting pass for private read-only mappings that
2001 * now will do cow due to allowed write (e.g. debugger sets
2002 * breakpoint on text segment)
2004 for (current = entry; current != &map->header && current->start < end;
2005 current = current->next) {
2007 vm_map_clip_end(map, current, end);
2010 ((new_prot & ~(current->protection)) & VM_PROT_WRITE) == 0 ||
2011 ENTRY_CHARGED(current)) {
2015 cred = curthread->td_ucred;
2016 obj = current->object.vm_object;
2018 if (obj == NULL || (current->eflags & MAP_ENTRY_NEEDS_COPY)) {
2019 if (!swap_reserve(current->end - current->start)) {
2021 return (KERN_RESOURCE_SHORTAGE);
2024 current->cred = cred;
2028 VM_OBJECT_WLOCK(obj);
2029 if (obj->type != OBJT_DEFAULT && obj->type != OBJT_SWAP) {
2030 VM_OBJECT_WUNLOCK(obj);
2035 * Charge for the whole object allocation now, since
2036 * we cannot distinguish between non-charged and
2037 * charged clipped mapping of the same object later.
2039 KASSERT(obj->charge == 0,
2040 ("vm_map_protect: object %p overcharged (entry %p)",
2042 if (!swap_reserve(ptoa(obj->size))) {
2043 VM_OBJECT_WUNLOCK(obj);
2045 return (KERN_RESOURCE_SHORTAGE);
2050 obj->charge = ptoa(obj->size);
2051 VM_OBJECT_WUNLOCK(obj);
2055 * Go back and fix up protections. [Note that clipping is not
2056 * necessary the second time.]
2058 for (current = entry; current != &map->header && current->start < end;
2059 current = current->next) {
2060 old_prot = current->protection;
2063 current->protection =
2064 (current->max_protection = new_prot) &
2067 current->protection = new_prot;
2070 * For user wired map entries, the normal lazy evaluation of
2071 * write access upgrades through soft page faults is
2072 * undesirable. Instead, immediately copy any pages that are
2073 * copy-on-write and enable write access in the physical map.
2075 if ((current->eflags & MAP_ENTRY_USER_WIRED) != 0 &&
2076 (current->protection & VM_PROT_WRITE) != 0 &&
2077 (old_prot & VM_PROT_WRITE) == 0)
2078 vm_fault_copy_entry(map, map, current, current, NULL);
2081 * When restricting access, update the physical map. Worry
2082 * about copy-on-write here.
2084 if ((old_prot & ~current->protection) != 0) {
2085 #define MASK(entry) (((entry)->eflags & MAP_ENTRY_COW) ? ~VM_PROT_WRITE : \
2087 pmap_protect(map->pmap, current->start,
2089 current->protection & MASK(current));
2092 vm_map_simplify_entry(map, current);
2095 return (KERN_SUCCESS);
2101 * This routine traverses a processes map handling the madvise
2102 * system call. Advisories are classified as either those effecting
2103 * the vm_map_entry structure, or those effecting the underlying
2113 vm_map_entry_t current, entry;
2117 * Some madvise calls directly modify the vm_map_entry, in which case
2118 * we need to use an exclusive lock on the map and we need to perform
2119 * various clipping operations. Otherwise we only need a read-lock
2124 case MADV_SEQUENTIAL:
2131 return (KERN_SUCCESS);
2139 return (KERN_SUCCESS);
2140 vm_map_lock_read(map);
2143 return (KERN_INVALID_ARGUMENT);
2147 * Locate starting entry and clip if necessary.
2149 VM_MAP_RANGE_CHECK(map, start, end);
2151 if (vm_map_lookup_entry(map, start, &entry)) {
2153 vm_map_clip_start(map, entry, start);
2155 entry = entry->next;
2160 * madvise behaviors that are implemented in the vm_map_entry.
2162 * We clip the vm_map_entry so that behavioral changes are
2163 * limited to the specified address range.
2165 for (current = entry;
2166 (current != &map->header) && (current->start < end);
2167 current = current->next
2169 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
2172 vm_map_clip_end(map, current, end);
2176 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_NORMAL);
2178 case MADV_SEQUENTIAL:
2179 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_SEQUENTIAL);
2182 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_RANDOM);
2185 current->eflags |= MAP_ENTRY_NOSYNC;
2188 current->eflags &= ~MAP_ENTRY_NOSYNC;
2191 current->eflags |= MAP_ENTRY_NOCOREDUMP;
2194 current->eflags &= ~MAP_ENTRY_NOCOREDUMP;
2199 vm_map_simplify_entry(map, current);
2203 vm_pindex_t pstart, pend;
2206 * madvise behaviors that are implemented in the underlying
2209 * Since we don't clip the vm_map_entry, we have to clip
2210 * the vm_object pindex and count.
2212 for (current = entry;
2213 (current != &map->header) && (current->start < end);
2214 current = current->next
2216 vm_offset_t useEnd, useStart;
2218 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
2221 pstart = OFF_TO_IDX(current->offset);
2222 pend = pstart + atop(current->end - current->start);
2223 useStart = current->start;
2224 useEnd = current->end;
2226 if (current->start < start) {
2227 pstart += atop(start - current->start);
2230 if (current->end > end) {
2231 pend -= atop(current->end - end);
2239 * Perform the pmap_advise() before clearing
2240 * PGA_REFERENCED in vm_page_advise(). Otherwise, a
2241 * concurrent pmap operation, such as pmap_remove(),
2242 * could clear a reference in the pmap and set
2243 * PGA_REFERENCED on the page before the pmap_advise()
2244 * had completed. Consequently, the page would appear
2245 * referenced based upon an old reference that
2246 * occurred before this pmap_advise() ran.
2248 if (behav == MADV_DONTNEED || behav == MADV_FREE)
2249 pmap_advise(map->pmap, useStart, useEnd,
2252 vm_object_madvise(current->object.vm_object, pstart,
2256 * Pre-populate paging structures in the
2257 * WILLNEED case. For wired entries, the
2258 * paging structures are already populated.
2260 if (behav == MADV_WILLNEED &&
2261 current->wired_count == 0) {
2262 vm_map_pmap_enter(map,
2264 current->protection,
2265 current->object.vm_object,
2267 ptoa(pend - pstart),
2268 MAP_PREFAULT_MADVISE
2272 vm_map_unlock_read(map);
2281 * Sets the inheritance of the specified address
2282 * range in the target map. Inheritance
2283 * affects how the map will be shared with
2284 * child maps at the time of vmspace_fork.
2287 vm_map_inherit(vm_map_t map, vm_offset_t start, vm_offset_t end,
2288 vm_inherit_t new_inheritance)
2290 vm_map_entry_t entry;
2291 vm_map_entry_t temp_entry;
2293 switch (new_inheritance) {
2294 case VM_INHERIT_NONE:
2295 case VM_INHERIT_COPY:
2296 case VM_INHERIT_SHARE:
2297 case VM_INHERIT_ZERO:
2300 return (KERN_INVALID_ARGUMENT);
2303 return (KERN_SUCCESS);
2305 VM_MAP_RANGE_CHECK(map, start, end);
2306 if (vm_map_lookup_entry(map, start, &temp_entry)) {
2308 vm_map_clip_start(map, entry, start);
2310 entry = temp_entry->next;
2311 while ((entry != &map->header) && (entry->start < end)) {
2312 vm_map_clip_end(map, entry, end);
2313 entry->inheritance = new_inheritance;
2314 vm_map_simplify_entry(map, entry);
2315 entry = entry->next;
2318 return (KERN_SUCCESS);
2324 * Implements both kernel and user unwiring.
2327 vm_map_unwire(vm_map_t map, vm_offset_t start, vm_offset_t end,
2330 vm_map_entry_t entry, first_entry, tmp_entry;
2331 vm_offset_t saved_start;
2332 unsigned int last_timestamp;
2334 boolean_t need_wakeup, result, user_unwire;
2337 return (KERN_SUCCESS);
2338 user_unwire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE;
2340 VM_MAP_RANGE_CHECK(map, start, end);
2341 if (!vm_map_lookup_entry(map, start, &first_entry)) {
2342 if (flags & VM_MAP_WIRE_HOLESOK)
2343 first_entry = first_entry->next;
2346 return (KERN_INVALID_ADDRESS);
2349 last_timestamp = map->timestamp;
2350 entry = first_entry;
2351 while (entry != &map->header && entry->start < end) {
2352 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
2354 * We have not yet clipped the entry.
2356 saved_start = (start >= entry->start) ? start :
2358 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2359 if (vm_map_unlock_and_wait(map, 0)) {
2361 * Allow interruption of user unwiring?
2365 if (last_timestamp+1 != map->timestamp) {
2367 * Look again for the entry because the map was
2368 * modified while it was unlocked.
2369 * Specifically, the entry may have been
2370 * clipped, merged, or deleted.
2372 if (!vm_map_lookup_entry(map, saved_start,
2374 if (flags & VM_MAP_WIRE_HOLESOK)
2375 tmp_entry = tmp_entry->next;
2377 if (saved_start == start) {
2379 * First_entry has been deleted.
2382 return (KERN_INVALID_ADDRESS);
2385 rv = KERN_INVALID_ADDRESS;
2389 if (entry == first_entry)
2390 first_entry = tmp_entry;
2395 last_timestamp = map->timestamp;
2398 vm_map_clip_start(map, entry, start);
2399 vm_map_clip_end(map, entry, end);
2401 * Mark the entry in case the map lock is released. (See
2404 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 &&
2405 entry->wiring_thread == NULL,
2406 ("owned map entry %p", entry));
2407 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
2408 entry->wiring_thread = curthread;
2410 * Check the map for holes in the specified region.
2411 * If VM_MAP_WIRE_HOLESOK was specified, skip this check.
2413 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) &&
2414 (entry->end < end && (entry->next == &map->header ||
2415 entry->next->start > entry->end))) {
2417 rv = KERN_INVALID_ADDRESS;
2421 * If system unwiring, require that the entry is system wired.
2424 vm_map_entry_system_wired_count(entry) == 0) {
2426 rv = KERN_INVALID_ARGUMENT;
2429 entry = entry->next;
2433 need_wakeup = FALSE;
2434 if (first_entry == NULL) {
2435 result = vm_map_lookup_entry(map, start, &first_entry);
2436 if (!result && (flags & VM_MAP_WIRE_HOLESOK))
2437 first_entry = first_entry->next;
2439 KASSERT(result, ("vm_map_unwire: lookup failed"));
2441 for (entry = first_entry; entry != &map->header && entry->start < end;
2442 entry = entry->next) {
2444 * If VM_MAP_WIRE_HOLESOK was specified, an empty
2445 * space in the unwired region could have been mapped
2446 * while the map lock was dropped for draining
2447 * MAP_ENTRY_IN_TRANSITION. Moreover, another thread
2448 * could be simultaneously wiring this new mapping
2449 * entry. Detect these cases and skip any entries
2450 * marked as in transition by us.
2452 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 ||
2453 entry->wiring_thread != curthread) {
2454 KASSERT((flags & VM_MAP_WIRE_HOLESOK) != 0,
2455 ("vm_map_unwire: !HOLESOK and new/changed entry"));
2459 if (rv == KERN_SUCCESS && (!user_unwire ||
2460 (entry->eflags & MAP_ENTRY_USER_WIRED))) {
2462 entry->eflags &= ~MAP_ENTRY_USER_WIRED;
2463 if (entry->wired_count == 1)
2464 vm_map_entry_unwire(map, entry);
2466 entry->wired_count--;
2468 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0,
2469 ("vm_map_unwire: in-transition flag missing %p", entry));
2470 KASSERT(entry->wiring_thread == curthread,
2471 ("vm_map_unwire: alien wire %p", entry));
2472 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION;
2473 entry->wiring_thread = NULL;
2474 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
2475 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
2478 vm_map_simplify_entry(map, entry);
2487 * vm_map_wire_entry_failure:
2489 * Handle a wiring failure on the given entry.
2491 * The map should be locked.
2494 vm_map_wire_entry_failure(vm_map_t map, vm_map_entry_t entry,
2495 vm_offset_t failed_addr)
2498 VM_MAP_ASSERT_LOCKED(map);
2499 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 &&
2500 entry->wired_count == 1,
2501 ("vm_map_wire_entry_failure: entry %p isn't being wired", entry));
2502 KASSERT(failed_addr < entry->end,
2503 ("vm_map_wire_entry_failure: entry %p was fully wired", entry));
2506 * If any pages at the start of this entry were successfully wired,
2509 if (failed_addr > entry->start) {
2510 pmap_unwire(map->pmap, entry->start, failed_addr);
2511 vm_object_unwire(entry->object.vm_object, entry->offset,
2512 failed_addr - entry->start, PQ_ACTIVE);
2516 * Assign an out-of-range value to represent the failure to wire this
2519 entry->wired_count = -1;
2525 * Implements both kernel and user wiring.
2528 vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t end,
2531 vm_map_entry_t entry, first_entry, tmp_entry;
2532 vm_offset_t faddr, saved_end, saved_start;
2533 unsigned int last_timestamp;
2535 boolean_t need_wakeup, result, user_wire;
2539 return (KERN_SUCCESS);
2541 if (flags & VM_MAP_WIRE_WRITE)
2542 prot |= VM_PROT_WRITE;
2543 user_wire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE;
2545 VM_MAP_RANGE_CHECK(map, start, end);
2546 if (!vm_map_lookup_entry(map, start, &first_entry)) {
2547 if (flags & VM_MAP_WIRE_HOLESOK)
2548 first_entry = first_entry->next;
2551 return (KERN_INVALID_ADDRESS);
2554 last_timestamp = map->timestamp;
2555 entry = first_entry;
2556 while (entry != &map->header && entry->start < end) {
2557 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
2559 * We have not yet clipped the entry.
2561 saved_start = (start >= entry->start) ? start :
2563 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2564 if (vm_map_unlock_and_wait(map, 0)) {
2566 * Allow interruption of user wiring?
2570 if (last_timestamp + 1 != map->timestamp) {
2572 * Look again for the entry because the map was
2573 * modified while it was unlocked.
2574 * Specifically, the entry may have been
2575 * clipped, merged, or deleted.
2577 if (!vm_map_lookup_entry(map, saved_start,
2579 if (flags & VM_MAP_WIRE_HOLESOK)
2580 tmp_entry = tmp_entry->next;
2582 if (saved_start == start) {
2584 * first_entry has been deleted.
2587 return (KERN_INVALID_ADDRESS);
2590 rv = KERN_INVALID_ADDRESS;
2594 if (entry == first_entry)
2595 first_entry = tmp_entry;
2600 last_timestamp = map->timestamp;
2603 vm_map_clip_start(map, entry, start);
2604 vm_map_clip_end(map, entry, end);
2606 * Mark the entry in case the map lock is released. (See
2609 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 &&
2610 entry->wiring_thread == NULL,
2611 ("owned map entry %p", entry));
2612 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
2613 entry->wiring_thread = curthread;
2614 if ((entry->protection & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0
2615 || (entry->protection & prot) != prot) {
2616 entry->eflags |= MAP_ENTRY_WIRE_SKIPPED;
2617 if ((flags & VM_MAP_WIRE_HOLESOK) == 0) {
2619 rv = KERN_INVALID_ADDRESS;
2624 if (entry->wired_count == 0) {
2625 entry->wired_count++;
2626 saved_start = entry->start;
2627 saved_end = entry->end;
2630 * Release the map lock, relying on the in-transition
2631 * mark. Mark the map busy for fork.
2636 faddr = saved_start;
2639 * Simulate a fault to get the page and enter
2640 * it into the physical map.
2642 if ((rv = vm_fault(map, faddr, VM_PROT_NONE,
2643 VM_FAULT_WIRE)) != KERN_SUCCESS)
2645 } while ((faddr += PAGE_SIZE) < saved_end);
2648 if (last_timestamp + 1 != map->timestamp) {
2650 * Look again for the entry because the map was
2651 * modified while it was unlocked. The entry
2652 * may have been clipped, but NOT merged or
2655 result = vm_map_lookup_entry(map, saved_start,
2657 KASSERT(result, ("vm_map_wire: lookup failed"));
2658 if (entry == first_entry)
2659 first_entry = tmp_entry;
2663 while (entry->end < saved_end) {
2665 * In case of failure, handle entries
2666 * that were not fully wired here;
2667 * fully wired entries are handled
2670 if (rv != KERN_SUCCESS &&
2672 vm_map_wire_entry_failure(map,
2674 entry = entry->next;
2677 last_timestamp = map->timestamp;
2678 if (rv != KERN_SUCCESS) {
2679 vm_map_wire_entry_failure(map, entry, faddr);
2683 } else if (!user_wire ||
2684 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
2685 entry->wired_count++;
2688 * Check the map for holes in the specified region.
2689 * If VM_MAP_WIRE_HOLESOK was specified, skip this check.
2692 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) &&
2693 (entry->end < end && (entry->next == &map->header ||
2694 entry->next->start > entry->end))) {
2696 rv = KERN_INVALID_ADDRESS;
2699 entry = entry->next;
2703 need_wakeup = FALSE;
2704 if (first_entry == NULL) {
2705 result = vm_map_lookup_entry(map, start, &first_entry);
2706 if (!result && (flags & VM_MAP_WIRE_HOLESOK))
2707 first_entry = first_entry->next;
2709 KASSERT(result, ("vm_map_wire: lookup failed"));
2711 for (entry = first_entry; entry != &map->header && entry->start < end;
2712 entry = entry->next) {
2713 if ((entry->eflags & MAP_ENTRY_WIRE_SKIPPED) != 0)
2714 goto next_entry_done;
2717 * If VM_MAP_WIRE_HOLESOK was specified, an empty
2718 * space in the unwired region could have been mapped
2719 * while the map lock was dropped for faulting in the
2720 * pages or draining MAP_ENTRY_IN_TRANSITION.
2721 * Moreover, another thread could be simultaneously
2722 * wiring this new mapping entry. Detect these cases
2723 * and skip any entries marked as in transition by us.
2725 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 ||
2726 entry->wiring_thread != curthread) {
2727 KASSERT((flags & VM_MAP_WIRE_HOLESOK) != 0,
2728 ("vm_map_wire: !HOLESOK and new/changed entry"));
2732 if (rv == KERN_SUCCESS) {
2734 entry->eflags |= MAP_ENTRY_USER_WIRED;
2735 } else if (entry->wired_count == -1) {
2737 * Wiring failed on this entry. Thus, unwiring is
2740 entry->wired_count = 0;
2741 } else if (!user_wire ||
2742 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
2744 * Undo the wiring. Wiring succeeded on this entry
2745 * but failed on a later entry.
2747 if (entry->wired_count == 1)
2748 vm_map_entry_unwire(map, entry);
2750 entry->wired_count--;
2753 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0,
2754 ("vm_map_wire: in-transition flag missing %p", entry));
2755 KASSERT(entry->wiring_thread == curthread,
2756 ("vm_map_wire: alien wire %p", entry));
2757 entry->eflags &= ~(MAP_ENTRY_IN_TRANSITION |
2758 MAP_ENTRY_WIRE_SKIPPED);
2759 entry->wiring_thread = NULL;
2760 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
2761 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
2764 vm_map_simplify_entry(map, entry);
2775 * Push any dirty cached pages in the address range to their pager.
2776 * If syncio is TRUE, dirty pages are written synchronously.
2777 * If invalidate is TRUE, any cached pages are freed as well.
2779 * If the size of the region from start to end is zero, we are
2780 * supposed to flush all modified pages within the region containing
2781 * start. Unfortunately, a region can be split or coalesced with
2782 * neighboring regions, making it difficult to determine what the
2783 * original region was. Therefore, we approximate this requirement by
2784 * flushing the current region containing start.
2786 * Returns an error if any part of the specified range is not mapped.
2794 boolean_t invalidate)
2796 vm_map_entry_t current;
2797 vm_map_entry_t entry;
2800 vm_ooffset_t offset;
2801 unsigned int last_timestamp;
2804 vm_map_lock_read(map);
2805 VM_MAP_RANGE_CHECK(map, start, end);
2806 if (!vm_map_lookup_entry(map, start, &entry)) {
2807 vm_map_unlock_read(map);
2808 return (KERN_INVALID_ADDRESS);
2809 } else if (start == end) {
2810 start = entry->start;
2814 * Make a first pass to check for user-wired memory and holes.
2816 for (current = entry; current != &map->header && current->start < end;
2817 current = current->next) {
2818 if (invalidate && (current->eflags & MAP_ENTRY_USER_WIRED)) {
2819 vm_map_unlock_read(map);
2820 return (KERN_INVALID_ARGUMENT);
2822 if (end > current->end &&
2823 (current->next == &map->header ||
2824 current->end != current->next->start)) {
2825 vm_map_unlock_read(map);
2826 return (KERN_INVALID_ADDRESS);
2831 pmap_remove(map->pmap, start, end);
2835 * Make a second pass, cleaning/uncaching pages from the indicated
2838 for (current = entry; current != &map->header && current->start < end;) {
2839 offset = current->offset + (start - current->start);
2840 size = (end <= current->end ? end : current->end) - start;
2841 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
2843 vm_map_entry_t tentry;
2846 smap = current->object.sub_map;
2847 vm_map_lock_read(smap);
2848 (void) vm_map_lookup_entry(smap, offset, &tentry);
2849 tsize = tentry->end - offset;
2852 object = tentry->object.vm_object;
2853 offset = tentry->offset + (offset - tentry->start);
2854 vm_map_unlock_read(smap);
2856 object = current->object.vm_object;
2858 vm_object_reference(object);
2859 last_timestamp = map->timestamp;
2860 vm_map_unlock_read(map);
2861 if (!vm_object_sync(object, offset, size, syncio, invalidate))
2864 vm_object_deallocate(object);
2865 vm_map_lock_read(map);
2866 if (last_timestamp == map->timestamp ||
2867 !vm_map_lookup_entry(map, start, ¤t))
2868 current = current->next;
2871 vm_map_unlock_read(map);
2872 return (failed ? KERN_FAILURE : KERN_SUCCESS);
2876 * vm_map_entry_unwire: [ internal use only ]
2878 * Make the region specified by this entry pageable.
2880 * The map in question should be locked.
2881 * [This is the reason for this routine's existence.]
2884 vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry)
2887 VM_MAP_ASSERT_LOCKED(map);
2888 KASSERT(entry->wired_count > 0,
2889 ("vm_map_entry_unwire: entry %p isn't wired", entry));
2890 pmap_unwire(map->pmap, entry->start, entry->end);
2891 vm_object_unwire(entry->object.vm_object, entry->offset, entry->end -
2892 entry->start, PQ_ACTIVE);
2893 entry->wired_count = 0;
2897 vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map)
2900 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0)
2901 vm_object_deallocate(entry->object.vm_object);
2902 uma_zfree(system_map ? kmapentzone : mapentzone, entry);
2906 * vm_map_entry_delete: [ internal use only ]
2908 * Deallocate the given entry from the target map.
2911 vm_map_entry_delete(vm_map_t map, vm_map_entry_t entry)
2914 vm_pindex_t offidxstart, offidxend, count, size1;
2917 vm_map_entry_unlink(map, entry);
2918 object = entry->object.vm_object;
2919 size = entry->end - entry->start;
2922 if (entry->cred != NULL) {
2923 swap_release_by_cred(size, entry->cred);
2924 crfree(entry->cred);
2927 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0 &&
2929 KASSERT(entry->cred == NULL || object->cred == NULL ||
2930 (entry->eflags & MAP_ENTRY_NEEDS_COPY),
2931 ("OVERCOMMIT vm_map_entry_delete: both cred %p", entry));
2933 offidxstart = OFF_TO_IDX(entry->offset);
2934 offidxend = offidxstart + count;
2935 VM_OBJECT_WLOCK(object);
2936 if (object->ref_count != 1 && ((object->flags & (OBJ_NOSPLIT |
2937 OBJ_ONEMAPPING)) == OBJ_ONEMAPPING ||
2938 object == kernel_object || object == kmem_object)) {
2939 vm_object_collapse(object);
2942 * The option OBJPR_NOTMAPPED can be passed here
2943 * because vm_map_delete() already performed
2944 * pmap_remove() on the only mapping to this range
2947 vm_object_page_remove(object, offidxstart, offidxend,
2949 if (object->type == OBJT_SWAP)
2950 swap_pager_freespace(object, offidxstart,
2952 if (offidxend >= object->size &&
2953 offidxstart < object->size) {
2954 size1 = object->size;
2955 object->size = offidxstart;
2956 if (object->cred != NULL) {
2957 size1 -= object->size;
2958 KASSERT(object->charge >= ptoa(size1),
2959 ("object %p charge < 0", object));
2960 swap_release_by_cred(ptoa(size1),
2962 object->charge -= ptoa(size1);
2966 VM_OBJECT_WUNLOCK(object);
2968 entry->object.vm_object = NULL;
2969 if (map->system_map)
2970 vm_map_entry_deallocate(entry, TRUE);
2972 entry->next = curthread->td_map_def_user;
2973 curthread->td_map_def_user = entry;
2978 * vm_map_delete: [ internal use only ]
2980 * Deallocates the given address range from the target
2984 vm_map_delete(vm_map_t map, vm_offset_t start, vm_offset_t end)
2986 vm_map_entry_t entry;
2987 vm_map_entry_t first_entry;
2989 VM_MAP_ASSERT_LOCKED(map);
2991 return (KERN_SUCCESS);
2994 * Find the start of the region, and clip it
2996 if (!vm_map_lookup_entry(map, start, &first_entry))
2997 entry = first_entry->next;
2999 entry = first_entry;
3000 vm_map_clip_start(map, entry, start);
3004 * Step through all entries in this region
3006 while ((entry != &map->header) && (entry->start < end)) {
3007 vm_map_entry_t next;
3010 * Wait for wiring or unwiring of an entry to complete.
3011 * Also wait for any system wirings to disappear on
3014 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 ||
3015 (vm_map_pmap(map) != kernel_pmap &&
3016 vm_map_entry_system_wired_count(entry) != 0)) {
3017 unsigned int last_timestamp;
3018 vm_offset_t saved_start;
3019 vm_map_entry_t tmp_entry;
3021 saved_start = entry->start;
3022 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
3023 last_timestamp = map->timestamp;
3024 (void) vm_map_unlock_and_wait(map, 0);
3026 if (last_timestamp + 1 != map->timestamp) {
3028 * Look again for the entry because the map was
3029 * modified while it was unlocked.
3030 * Specifically, the entry may have been
3031 * clipped, merged, or deleted.
3033 if (!vm_map_lookup_entry(map, saved_start,
3035 entry = tmp_entry->next;
3038 vm_map_clip_start(map, entry,
3044 vm_map_clip_end(map, entry, end);
3049 * Unwire before removing addresses from the pmap; otherwise,
3050 * unwiring will put the entries back in the pmap.
3052 if (entry->wired_count != 0) {
3053 vm_map_entry_unwire(map, entry);
3056 pmap_remove(map->pmap, entry->start, entry->end);
3059 * Delete the entry only after removing all pmap
3060 * entries pointing to its pages. (Otherwise, its
3061 * page frames may be reallocated, and any modify bits
3062 * will be set in the wrong object!)
3064 vm_map_entry_delete(map, entry);
3067 return (KERN_SUCCESS);
3073 * Remove the given address range from the target map.
3074 * This is the exported form of vm_map_delete.
3077 vm_map_remove(vm_map_t map, vm_offset_t start, vm_offset_t end)
3082 VM_MAP_RANGE_CHECK(map, start, end);
3083 result = vm_map_delete(map, start, end);
3089 * vm_map_check_protection:
3091 * Assert that the target map allows the specified privilege on the
3092 * entire address region given. The entire region must be allocated.
3094 * WARNING! This code does not and should not check whether the
3095 * contents of the region is accessible. For example a smaller file
3096 * might be mapped into a larger address space.
3098 * NOTE! This code is also called by munmap().
3100 * The map must be locked. A read lock is sufficient.
3103 vm_map_check_protection(vm_map_t map, vm_offset_t start, vm_offset_t end,
3104 vm_prot_t protection)
3106 vm_map_entry_t entry;
3107 vm_map_entry_t tmp_entry;
3109 if (!vm_map_lookup_entry(map, start, &tmp_entry))
3113 while (start < end) {
3114 if (entry == &map->header)
3119 if (start < entry->start)
3122 * Check protection associated with entry.
3124 if ((entry->protection & protection) != protection)
3126 /* go to next entry */
3128 entry = entry->next;
3134 * vm_map_copy_entry:
3136 * Copies the contents of the source entry to the destination
3137 * entry. The entries *must* be aligned properly.
3143 vm_map_entry_t src_entry,
3144 vm_map_entry_t dst_entry,
3145 vm_ooffset_t *fork_charge)
3147 vm_object_t src_object;
3148 vm_map_entry_t fake_entry;
3153 VM_MAP_ASSERT_LOCKED(dst_map);
3155 if ((dst_entry->eflags|src_entry->eflags) & MAP_ENTRY_IS_SUB_MAP)
3158 if (src_entry->wired_count == 0 ||
3159 (src_entry->protection & VM_PROT_WRITE) == 0) {
3161 * If the source entry is marked needs_copy, it is already
3164 if ((src_entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0 &&
3165 (src_entry->protection & VM_PROT_WRITE) != 0) {
3166 pmap_protect(src_map->pmap,
3169 src_entry->protection & ~VM_PROT_WRITE);
3173 * Make a copy of the object.
3175 size = src_entry->end - src_entry->start;
3176 if ((src_object = src_entry->object.vm_object) != NULL) {
3177 VM_OBJECT_WLOCK(src_object);
3178 charged = ENTRY_CHARGED(src_entry);
3179 if (src_object->handle == NULL &&
3180 (src_object->type == OBJT_DEFAULT ||
3181 src_object->type == OBJT_SWAP)) {
3182 vm_object_collapse(src_object);
3183 if ((src_object->flags & (OBJ_NOSPLIT |
3184 OBJ_ONEMAPPING)) == OBJ_ONEMAPPING) {
3185 vm_object_split(src_entry);
3187 src_entry->object.vm_object;
3190 vm_object_reference_locked(src_object);
3191 vm_object_clear_flag(src_object, OBJ_ONEMAPPING);
3192 if (src_entry->cred != NULL &&
3193 !(src_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
3194 KASSERT(src_object->cred == NULL,
3195 ("OVERCOMMIT: vm_map_copy_entry: cred %p",
3197 src_object->cred = src_entry->cred;
3198 src_object->charge = size;
3200 VM_OBJECT_WUNLOCK(src_object);
3201 dst_entry->object.vm_object = src_object;
3203 cred = curthread->td_ucred;
3205 dst_entry->cred = cred;
3206 *fork_charge += size;
3207 if (!(src_entry->eflags &
3208 MAP_ENTRY_NEEDS_COPY)) {
3210 src_entry->cred = cred;
3211 *fork_charge += size;
3214 src_entry->eflags |= MAP_ENTRY_COW |
3215 MAP_ENTRY_NEEDS_COPY;
3216 dst_entry->eflags |= MAP_ENTRY_COW |
3217 MAP_ENTRY_NEEDS_COPY;
3218 dst_entry->offset = src_entry->offset;
3219 if (src_entry->eflags & MAP_ENTRY_VN_WRITECNT) {
3221 * MAP_ENTRY_VN_WRITECNT cannot
3222 * indicate write reference from
3223 * src_entry, since the entry is
3224 * marked as needs copy. Allocate a
3225 * fake entry that is used to
3226 * decrement object->un_pager.vnp.writecount
3227 * at the appropriate time. Attach
3228 * fake_entry to the deferred list.
3230 fake_entry = vm_map_entry_create(dst_map);
3231 fake_entry->eflags = MAP_ENTRY_VN_WRITECNT;
3232 src_entry->eflags &= ~MAP_ENTRY_VN_WRITECNT;
3233 vm_object_reference(src_object);
3234 fake_entry->object.vm_object = src_object;
3235 fake_entry->start = src_entry->start;
3236 fake_entry->end = src_entry->end;
3237 fake_entry->next = curthread->td_map_def_user;
3238 curthread->td_map_def_user = fake_entry;
3241 dst_entry->object.vm_object = NULL;
3242 dst_entry->offset = 0;
3243 if (src_entry->cred != NULL) {
3244 dst_entry->cred = curthread->td_ucred;
3245 crhold(dst_entry->cred);
3246 *fork_charge += size;
3250 pmap_copy(dst_map->pmap, src_map->pmap, dst_entry->start,
3251 dst_entry->end - dst_entry->start, src_entry->start);
3254 * We don't want to make writeable wired pages copy-on-write.
3255 * Immediately copy these pages into the new map by simulating
3256 * page faults. The new pages are pageable.
3258 vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry,
3264 * vmspace_map_entry_forked:
3265 * Update the newly-forked vmspace each time a map entry is inherited
3266 * or copied. The values for vm_dsize and vm_tsize are approximate
3267 * (and mostly-obsolete ideas in the face of mmap(2) et al.)
3270 vmspace_map_entry_forked(const struct vmspace *vm1, struct vmspace *vm2,
3271 vm_map_entry_t entry)
3273 vm_size_t entrysize;
3276 entrysize = entry->end - entry->start;
3277 vm2->vm_map.size += entrysize;
3278 if (entry->eflags & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP)) {
3279 vm2->vm_ssize += btoc(entrysize);
3280 } else if (entry->start >= (vm_offset_t)vm1->vm_daddr &&
3281 entry->start < (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)) {
3282 newend = MIN(entry->end,
3283 (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize));
3284 vm2->vm_dsize += btoc(newend - entry->start);
3285 } else if (entry->start >= (vm_offset_t)vm1->vm_taddr &&
3286 entry->start < (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)) {
3287 newend = MIN(entry->end,
3288 (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize));
3289 vm2->vm_tsize += btoc(newend - entry->start);
3295 * Create a new process vmspace structure and vm_map
3296 * based on those of an existing process. The new map
3297 * is based on the old map, according to the inheritance
3298 * values on the regions in that map.
3300 * XXX It might be worth coalescing the entries added to the new vmspace.
3302 * The source map must not be locked.
3305 vmspace_fork(struct vmspace *vm1, vm_ooffset_t *fork_charge)
3307 struct vmspace *vm2;
3308 vm_map_t new_map, old_map;
3309 vm_map_entry_t new_entry, old_entry;
3313 old_map = &vm1->vm_map;
3314 /* Copy immutable fields of vm1 to vm2. */
3315 vm2 = vmspace_alloc(old_map->min_offset, old_map->max_offset, NULL);
3318 vm2->vm_taddr = vm1->vm_taddr;
3319 vm2->vm_daddr = vm1->vm_daddr;
3320 vm2->vm_maxsaddr = vm1->vm_maxsaddr;
3321 vm_map_lock(old_map);
3323 vm_map_wait_busy(old_map);
3324 new_map = &vm2->vm_map;
3325 locked = vm_map_trylock(new_map); /* trylock to silence WITNESS */
3326 KASSERT(locked, ("vmspace_fork: lock failed"));
3328 old_entry = old_map->header.next;
3330 while (old_entry != &old_map->header) {
3331 if (old_entry->eflags & MAP_ENTRY_IS_SUB_MAP)
3332 panic("vm_map_fork: encountered a submap");
3334 switch (old_entry->inheritance) {
3335 case VM_INHERIT_NONE:
3338 case VM_INHERIT_SHARE:
3340 * Clone the entry, creating the shared object if necessary.
3342 object = old_entry->object.vm_object;
3343 if (object == NULL) {
3344 object = vm_object_allocate(OBJT_DEFAULT,
3345 atop(old_entry->end - old_entry->start));
3346 old_entry->object.vm_object = object;
3347 old_entry->offset = 0;
3348 if (old_entry->cred != NULL) {
3349 object->cred = old_entry->cred;
3350 object->charge = old_entry->end -
3352 old_entry->cred = NULL;
3357 * Add the reference before calling vm_object_shadow
3358 * to insure that a shadow object is created.
3360 vm_object_reference(object);
3361 if (old_entry->eflags & MAP_ENTRY_NEEDS_COPY) {
3362 vm_object_shadow(&old_entry->object.vm_object,
3364 old_entry->end - old_entry->start);
3365 old_entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
3366 /* Transfer the second reference too. */
3367 vm_object_reference(
3368 old_entry->object.vm_object);
3371 * As in vm_map_simplify_entry(), the
3372 * vnode lock will not be acquired in
3373 * this call to vm_object_deallocate().
3375 vm_object_deallocate(object);
3376 object = old_entry->object.vm_object;
3378 VM_OBJECT_WLOCK(object);
3379 vm_object_clear_flag(object, OBJ_ONEMAPPING);
3380 if (old_entry->cred != NULL) {
3381 KASSERT(object->cred == NULL, ("vmspace_fork both cred"));
3382 object->cred = old_entry->cred;
3383 object->charge = old_entry->end - old_entry->start;
3384 old_entry->cred = NULL;
3388 * Assert the correct state of the vnode
3389 * v_writecount while the object is locked, to
3390 * not relock it later for the assertion
3393 if (old_entry->eflags & MAP_ENTRY_VN_WRITECNT &&
3394 object->type == OBJT_VNODE) {
3395 KASSERT(((struct vnode *)object->handle)->
3397 ("vmspace_fork: v_writecount %p", object));
3398 KASSERT(object->un_pager.vnp.writemappings > 0,
3399 ("vmspace_fork: vnp.writecount %p",
3402 VM_OBJECT_WUNLOCK(object);
3405 * Clone the entry, referencing the shared object.
3407 new_entry = vm_map_entry_create(new_map);
3408 *new_entry = *old_entry;
3409 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED |
3410 MAP_ENTRY_IN_TRANSITION);
3411 new_entry->wiring_thread = NULL;
3412 new_entry->wired_count = 0;
3413 if (new_entry->eflags & MAP_ENTRY_VN_WRITECNT) {
3414 vnode_pager_update_writecount(object,
3415 new_entry->start, new_entry->end);
3419 * Insert the entry into the new map -- we know we're
3420 * inserting at the end of the new map.
3422 vm_map_entry_link(new_map, new_map->header.prev,
3424 vmspace_map_entry_forked(vm1, vm2, new_entry);
3427 * Update the physical map
3429 pmap_copy(new_map->pmap, old_map->pmap,
3431 (old_entry->end - old_entry->start),
3435 case VM_INHERIT_COPY:
3437 * Clone the entry and link into the map.
3439 new_entry = vm_map_entry_create(new_map);
3440 *new_entry = *old_entry;
3442 * Copied entry is COW over the old object.
3444 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED |
3445 MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_VN_WRITECNT);
3446 new_entry->wiring_thread = NULL;
3447 new_entry->wired_count = 0;
3448 new_entry->object.vm_object = NULL;
3449 new_entry->cred = NULL;
3450 vm_map_entry_link(new_map, new_map->header.prev,
3452 vmspace_map_entry_forked(vm1, vm2, new_entry);
3453 vm_map_copy_entry(old_map, new_map, old_entry,
3454 new_entry, fork_charge);
3457 case VM_INHERIT_ZERO:
3459 * Create a new anonymous mapping entry modelled from
3462 new_entry = vm_map_entry_create(new_map);
3463 memset(new_entry, 0, sizeof(*new_entry));
3465 new_entry->start = old_entry->start;
3466 new_entry->end = old_entry->end;
3467 new_entry->avail_ssize = old_entry->avail_ssize;
3468 new_entry->eflags = old_entry->eflags &
3469 ~(MAP_ENTRY_USER_WIRED | MAP_ENTRY_IN_TRANSITION |
3470 MAP_ENTRY_VN_WRITECNT);
3471 new_entry->protection = old_entry->protection;
3472 new_entry->max_protection = old_entry->max_protection;
3473 new_entry->inheritance = VM_INHERIT_ZERO;
3475 vm_map_entry_link(new_map, new_map->header.prev,
3477 vmspace_map_entry_forked(vm1, vm2, new_entry);
3479 new_entry->cred = curthread->td_ucred;
3480 crhold(new_entry->cred);
3481 *fork_charge += (new_entry->end - new_entry->start);
3485 old_entry = old_entry->next;
3488 * Use inlined vm_map_unlock() to postpone handling the deferred
3489 * map entries, which cannot be done until both old_map and
3490 * new_map locks are released.
3492 sx_xunlock(&old_map->lock);
3493 sx_xunlock(&new_map->lock);
3494 vm_map_process_deferred();
3500 vm_map_stack(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
3501 vm_prot_t prot, vm_prot_t max, int cow)
3503 vm_size_t growsize, init_ssize;
3504 rlim_t lmemlim, vmemlim;
3507 growsize = sgrowsiz;
3508 init_ssize = (max_ssize < growsize) ? max_ssize : growsize;
3510 lmemlim = lim_cur(curthread, RLIMIT_MEMLOCK);
3511 vmemlim = lim_cur(curthread, RLIMIT_VMEM);
3512 if (!old_mlock && map->flags & MAP_WIREFUTURE) {
3513 if (ptoa(pmap_wired_count(map->pmap)) + init_ssize > lmemlim) {
3518 /* If we would blow our VMEM resource limit, no go */
3519 if (map->size + init_ssize > vmemlim) {
3523 rv = vm_map_stack_locked(map, addrbos, max_ssize, growsize, prot,
3531 vm_map_stack_locked(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
3532 vm_size_t growsize, vm_prot_t prot, vm_prot_t max, int cow)
3534 vm_map_entry_t new_entry, prev_entry;
3535 vm_offset_t bot, top;
3536 vm_size_t init_ssize;
3540 * The stack orientation is piggybacked with the cow argument.
3541 * Extract it into orient and mask the cow argument so that we
3542 * don't pass it around further.
3543 * NOTE: We explicitly allow bi-directional stacks.
3545 orient = cow & (MAP_STACK_GROWS_DOWN|MAP_STACK_GROWS_UP);
3546 KASSERT(orient != 0, ("No stack grow direction"));
3548 if (addrbos < vm_map_min(map) ||
3549 addrbos > vm_map_max(map) ||
3550 addrbos + max_ssize < addrbos)
3551 return (KERN_NO_SPACE);
3553 init_ssize = (max_ssize < growsize) ? max_ssize : growsize;
3555 /* If addr is already mapped, no go */
3556 if (vm_map_lookup_entry(map, addrbos, &prev_entry))
3557 return (KERN_NO_SPACE);
3560 * If we can't accommodate max_ssize in the current mapping, no go.
3561 * However, we need to be aware that subsequent user mappings might
3562 * map into the space we have reserved for stack, and currently this
3563 * space is not protected.
3565 * Hopefully we will at least detect this condition when we try to
3568 if ((prev_entry->next != &map->header) &&
3569 (prev_entry->next->start < addrbos + max_ssize))
3570 return (KERN_NO_SPACE);
3573 * We initially map a stack of only init_ssize. We will grow as
3574 * needed later. Depending on the orientation of the stack (i.e.
3575 * the grow direction) we either map at the top of the range, the
3576 * bottom of the range or in the middle.
3578 * Note: we would normally expect prot and max to be VM_PROT_ALL,
3579 * and cow to be 0. Possibly we should eliminate these as input
3580 * parameters, and just pass these values here in the insert call.
3582 if (orient == MAP_STACK_GROWS_DOWN)
3583 bot = addrbos + max_ssize - init_ssize;
3584 else if (orient == MAP_STACK_GROWS_UP)
3587 bot = round_page(addrbos + max_ssize/2 - init_ssize/2);
3588 top = bot + init_ssize;
3589 rv = vm_map_insert(map, NULL, 0, bot, top, prot, max, cow);
3591 /* Now set the avail_ssize amount. */
3592 if (rv == KERN_SUCCESS) {
3593 new_entry = prev_entry->next;
3594 if (new_entry->end != top || new_entry->start != bot)
3595 panic("Bad entry start/end for new stack entry");
3597 new_entry->avail_ssize = max_ssize - init_ssize;
3598 KASSERT((orient & MAP_STACK_GROWS_DOWN) == 0 ||
3599 (new_entry->eflags & MAP_ENTRY_GROWS_DOWN) != 0,
3600 ("new entry lacks MAP_ENTRY_GROWS_DOWN"));
3601 KASSERT((orient & MAP_STACK_GROWS_UP) == 0 ||
3602 (new_entry->eflags & MAP_ENTRY_GROWS_UP) != 0,
3603 ("new entry lacks MAP_ENTRY_GROWS_UP"));
3609 static int stack_guard_page = 0;
3610 SYSCTL_INT(_security_bsd, OID_AUTO, stack_guard_page, CTLFLAG_RWTUN,
3611 &stack_guard_page, 0,
3612 "Insert stack guard page ahead of the growable segments.");
3614 /* Attempts to grow a vm stack entry. Returns KERN_SUCCESS if the
3615 * desired address is already mapped, or if we successfully grow
3616 * the stack. Also returns KERN_SUCCESS if addr is outside the
3617 * stack range (this is strange, but preserves compatibility with
3618 * the grow function in vm_machdep.c).
3621 vm_map_growstack(struct proc *p, vm_offset_t addr)
3623 vm_map_entry_t next_entry, prev_entry;
3624 vm_map_entry_t new_entry, stack_entry;
3625 struct vmspace *vm = p->p_vmspace;
3626 vm_map_t map = &vm->vm_map;
3629 size_t grow_amount, max_grow;
3630 rlim_t lmemlim, stacklim, vmemlim;
3631 int is_procstack, rv;
3640 lmemlim = lim_cur(curthread, RLIMIT_MEMLOCK);
3641 stacklim = lim_cur(curthread, RLIMIT_STACK);
3642 vmemlim = lim_cur(curthread, RLIMIT_VMEM);
3645 vm_map_lock_read(map);
3647 /* If addr is already in the entry range, no need to grow.*/
3648 if (vm_map_lookup_entry(map, addr, &prev_entry)) {
3649 vm_map_unlock_read(map);
3650 return (KERN_SUCCESS);
3653 next_entry = prev_entry->next;
3654 if (!(prev_entry->eflags & MAP_ENTRY_GROWS_UP)) {
3656 * This entry does not grow upwards. Since the address lies
3657 * beyond this entry, the next entry (if one exists) has to
3658 * be a downward growable entry. The entry list header is
3659 * never a growable entry, so it suffices to check the flags.
3661 if (!(next_entry->eflags & MAP_ENTRY_GROWS_DOWN)) {
3662 vm_map_unlock_read(map);
3663 return (KERN_SUCCESS);
3665 stack_entry = next_entry;
3668 * This entry grows upward. If the next entry does not at
3669 * least grow downwards, this is the entry we need to grow.
3670 * otherwise we have two possible choices and we have to
3673 if (next_entry->eflags & MAP_ENTRY_GROWS_DOWN) {
3675 * We have two choices; grow the entry closest to
3676 * the address to minimize the amount of growth.
3678 if (addr - prev_entry->end <= next_entry->start - addr)
3679 stack_entry = prev_entry;
3681 stack_entry = next_entry;
3683 stack_entry = prev_entry;
3686 if (stack_entry == next_entry) {
3687 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_DOWN, ("foo"));
3688 KASSERT(addr < stack_entry->start, ("foo"));
3689 end = (prev_entry != &map->header) ? prev_entry->end :
3690 stack_entry->start - stack_entry->avail_ssize;
3691 grow_amount = roundup(stack_entry->start - addr, PAGE_SIZE);
3692 max_grow = stack_entry->start - end;
3694 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_UP, ("foo"));
3695 KASSERT(addr >= stack_entry->end, ("foo"));
3696 end = (next_entry != &map->header) ? next_entry->start :
3697 stack_entry->end + stack_entry->avail_ssize;
3698 grow_amount = roundup(addr + 1 - stack_entry->end, PAGE_SIZE);
3699 max_grow = end - stack_entry->end;
3702 if (grow_amount > stack_entry->avail_ssize) {
3703 vm_map_unlock_read(map);
3704 return (KERN_NO_SPACE);
3708 * If there is no longer enough space between the entries nogo, and
3709 * adjust the available space. Note: this should only happen if the
3710 * user has mapped into the stack area after the stack was created,
3711 * and is probably an error.
3713 * This also effectively destroys any guard page the user might have
3714 * intended by limiting the stack size.
3716 if (grow_amount + (stack_guard_page ? PAGE_SIZE : 0) > max_grow) {
3717 if (vm_map_lock_upgrade(map))
3720 stack_entry->avail_ssize = max_grow;
3723 return (KERN_NO_SPACE);
3726 is_procstack = (addr >= (vm_offset_t)vm->vm_maxsaddr &&
3727 addr < (vm_offset_t)p->p_sysent->sv_usrstack) ? 1 : 0;
3730 * If this is the main process stack, see if we're over the stack
3733 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) {
3734 vm_map_unlock_read(map);
3735 return (KERN_NO_SPACE);
3740 if (is_procstack && racct_set(p, RACCT_STACK,
3741 ctob(vm->vm_ssize) + grow_amount)) {
3743 vm_map_unlock_read(map);
3744 return (KERN_NO_SPACE);
3750 /* Round up the grow amount modulo sgrowsiz */
3751 growsize = sgrowsiz;
3752 grow_amount = roundup(grow_amount, growsize);
3753 if (grow_amount > stack_entry->avail_ssize)
3754 grow_amount = stack_entry->avail_ssize;
3755 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) {
3756 grow_amount = trunc_page((vm_size_t)stacklim) -
3761 limit = racct_get_available(p, RACCT_STACK);
3763 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > limit))
3764 grow_amount = limit - ctob(vm->vm_ssize);
3766 if (!old_mlock && map->flags & MAP_WIREFUTURE) {
3767 if (ptoa(pmap_wired_count(map->pmap)) + grow_amount > lmemlim) {
3768 vm_map_unlock_read(map);
3775 if (racct_set(p, RACCT_MEMLOCK,
3776 ptoa(pmap_wired_count(map->pmap)) + grow_amount)) {
3778 vm_map_unlock_read(map);
3786 /* If we would blow our VMEM resource limit, no go */
3787 if (map->size + grow_amount > vmemlim) {
3788 vm_map_unlock_read(map);
3795 if (racct_set(p, RACCT_VMEM, map->size + grow_amount)) {
3797 vm_map_unlock_read(map);
3805 if (vm_map_lock_upgrade(map))
3808 if (stack_entry == next_entry) {
3812 /* Get the preliminary new entry start value */
3813 addr = stack_entry->start - grow_amount;
3816 * If this puts us into the previous entry, cut back our
3817 * growth to the available space. Also, see the note above.
3820 stack_entry->avail_ssize = max_grow;
3822 if (stack_guard_page)
3826 rv = vm_map_insert(map, NULL, 0, addr, stack_entry->start,
3827 next_entry->protection, next_entry->max_protection,
3828 MAP_STACK_GROWS_DOWN);
3830 /* Adjust the available stack space by the amount we grew. */
3831 if (rv == KERN_SUCCESS) {
3832 new_entry = prev_entry->next;
3833 KASSERT(new_entry == stack_entry->prev, ("foo"));
3834 KASSERT(new_entry->end == stack_entry->start, ("foo"));
3835 KASSERT(new_entry->start == addr, ("foo"));
3836 KASSERT((new_entry->eflags & MAP_ENTRY_GROWS_DOWN) !=
3837 0, ("new entry lacks MAP_ENTRY_GROWS_DOWN"));
3838 grow_amount = new_entry->end - new_entry->start;
3839 new_entry->avail_ssize = stack_entry->avail_ssize -
3841 stack_entry->eflags &= ~MAP_ENTRY_GROWS_DOWN;
3847 addr = stack_entry->end + grow_amount;
3850 * If this puts us into the next entry, cut back our growth
3851 * to the available space. Also, see the note above.
3854 stack_entry->avail_ssize = end - stack_entry->end;
3856 if (stack_guard_page)
3860 grow_amount = addr - stack_entry->end;
3861 cred = stack_entry->cred;
3862 if (cred == NULL && stack_entry->object.vm_object != NULL)
3863 cred = stack_entry->object.vm_object->cred;
3864 if (cred != NULL && !swap_reserve_by_cred(grow_amount, cred))
3866 /* Grow the underlying object if applicable. */
3867 else if (stack_entry->object.vm_object == NULL ||
3868 vm_object_coalesce(stack_entry->object.vm_object,
3869 stack_entry->offset,
3870 (vm_size_t)(stack_entry->end - stack_entry->start),
3871 (vm_size_t)grow_amount, cred != NULL)) {
3872 map->size += (addr - stack_entry->end);
3873 /* Update the current entry. */
3874 stack_entry->end = addr;
3875 stack_entry->avail_ssize -= grow_amount;
3876 vm_map_entry_resize_free(map, stack_entry);
3882 if (rv == KERN_SUCCESS && is_procstack)
3883 vm->vm_ssize += btoc(grow_amount);
3888 * Heed the MAP_WIREFUTURE flag if it was set for this process.
3890 if (rv == KERN_SUCCESS && (map->flags & MAP_WIREFUTURE)) {
3892 (stack_entry == next_entry) ? addr : addr - grow_amount,
3893 (stack_entry == next_entry) ? stack_entry->start : addr,
3894 (p->p_flag & P_SYSTEM)
3895 ? VM_MAP_WIRE_SYSTEM|VM_MAP_WIRE_NOHOLES
3896 : VM_MAP_WIRE_USER|VM_MAP_WIRE_NOHOLES);
3901 if (racct_enable && rv != KERN_SUCCESS) {
3903 error = racct_set(p, RACCT_VMEM, map->size);
3904 KASSERT(error == 0, ("decreasing RACCT_VMEM failed"));
3906 error = racct_set(p, RACCT_MEMLOCK,
3907 ptoa(pmap_wired_count(map->pmap)));
3908 KASSERT(error == 0, ("decreasing RACCT_MEMLOCK failed"));
3910 error = racct_set(p, RACCT_STACK, ctob(vm->vm_ssize));
3911 KASSERT(error == 0, ("decreasing RACCT_STACK failed"));
3920 * Unshare the specified VM space for exec. If other processes are
3921 * mapped to it, then create a new one. The new vmspace is null.
3924 vmspace_exec(struct proc *p, vm_offset_t minuser, vm_offset_t maxuser)
3926 struct vmspace *oldvmspace = p->p_vmspace;
3927 struct vmspace *newvmspace;
3929 KASSERT((curthread->td_pflags & TDP_EXECVMSPC) == 0,
3930 ("vmspace_exec recursed"));
3931 newvmspace = vmspace_alloc(minuser, maxuser, NULL);
3932 if (newvmspace == NULL)
3934 newvmspace->vm_swrss = oldvmspace->vm_swrss;
3936 * This code is written like this for prototype purposes. The
3937 * goal is to avoid running down the vmspace here, but let the
3938 * other process's that are still using the vmspace to finally
3939 * run it down. Even though there is little or no chance of blocking
3940 * here, it is a good idea to keep this form for future mods.
3942 PROC_VMSPACE_LOCK(p);
3943 p->p_vmspace = newvmspace;
3944 PROC_VMSPACE_UNLOCK(p);
3945 if (p == curthread->td_proc)
3946 pmap_activate(curthread);
3947 curthread->td_pflags |= TDP_EXECVMSPC;
3952 * Unshare the specified VM space for forcing COW. This
3953 * is called by rfork, for the (RFMEM|RFPROC) == 0 case.
3956 vmspace_unshare(struct proc *p)
3958 struct vmspace *oldvmspace = p->p_vmspace;
3959 struct vmspace *newvmspace;
3960 vm_ooffset_t fork_charge;
3962 if (oldvmspace->vm_refcnt == 1)
3965 newvmspace = vmspace_fork(oldvmspace, &fork_charge);
3966 if (newvmspace == NULL)
3968 if (!swap_reserve_by_cred(fork_charge, p->p_ucred)) {
3969 vmspace_free(newvmspace);
3972 PROC_VMSPACE_LOCK(p);
3973 p->p_vmspace = newvmspace;
3974 PROC_VMSPACE_UNLOCK(p);
3975 if (p == curthread->td_proc)
3976 pmap_activate(curthread);
3977 vmspace_free(oldvmspace);
3984 * Finds the VM object, offset, and
3985 * protection for a given virtual address in the
3986 * specified map, assuming a page fault of the
3989 * Leaves the map in question locked for read; return
3990 * values are guaranteed until a vm_map_lookup_done
3991 * call is performed. Note that the map argument
3992 * is in/out; the returned map must be used in
3993 * the call to vm_map_lookup_done.
3995 * A handle (out_entry) is returned for use in
3996 * vm_map_lookup_done, to make that fast.
3998 * If a lookup is requested with "write protection"
3999 * specified, the map may be changed to perform virtual
4000 * copying operations, although the data referenced will
4004 vm_map_lookup(vm_map_t *var_map, /* IN/OUT */
4006 vm_prot_t fault_typea,
4007 vm_map_entry_t *out_entry, /* OUT */
4008 vm_object_t *object, /* OUT */
4009 vm_pindex_t *pindex, /* OUT */
4010 vm_prot_t *out_prot, /* OUT */
4011 boolean_t *wired) /* OUT */
4013 vm_map_entry_t entry;
4014 vm_map_t map = *var_map;
4016 vm_prot_t fault_type = fault_typea;
4017 vm_object_t eobject;
4023 vm_map_lock_read(map);
4026 * Lookup the faulting address.
4028 if (!vm_map_lookup_entry(map, vaddr, out_entry)) {
4029 vm_map_unlock_read(map);
4030 return (KERN_INVALID_ADDRESS);
4038 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
4039 vm_map_t old_map = map;
4041 *var_map = map = entry->object.sub_map;
4042 vm_map_unlock_read(old_map);
4047 * Check whether this task is allowed to have this page.
4049 prot = entry->protection;
4050 fault_type &= (VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE);
4051 if ((fault_type & prot) != fault_type || prot == VM_PROT_NONE) {
4052 vm_map_unlock_read(map);
4053 return (KERN_PROTECTION_FAILURE);
4055 KASSERT((prot & VM_PROT_WRITE) == 0 || (entry->eflags &
4056 (MAP_ENTRY_USER_WIRED | MAP_ENTRY_NEEDS_COPY)) !=
4057 (MAP_ENTRY_USER_WIRED | MAP_ENTRY_NEEDS_COPY),
4058 ("entry %p flags %x", entry, entry->eflags));
4059 if ((fault_typea & VM_PROT_COPY) != 0 &&
4060 (entry->max_protection & VM_PROT_WRITE) == 0 &&
4061 (entry->eflags & MAP_ENTRY_COW) == 0) {
4062 vm_map_unlock_read(map);
4063 return (KERN_PROTECTION_FAILURE);
4067 * If this page is not pageable, we have to get it for all possible
4070 *wired = (entry->wired_count != 0);
4072 fault_type = entry->protection;
4073 size = entry->end - entry->start;
4075 * If the entry was copy-on-write, we either ...
4077 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
4079 * If we want to write the page, we may as well handle that
4080 * now since we've got the map locked.
4082 * If we don't need to write the page, we just demote the
4083 * permissions allowed.
4085 if ((fault_type & VM_PROT_WRITE) != 0 ||
4086 (fault_typea & VM_PROT_COPY) != 0) {
4088 * Make a new object, and place it in the object
4089 * chain. Note that no new references have appeared
4090 * -- one just moved from the map to the new
4093 if (vm_map_lock_upgrade(map))
4096 if (entry->cred == NULL) {
4098 * The debugger owner is charged for
4101 cred = curthread->td_ucred;
4103 if (!swap_reserve_by_cred(size, cred)) {
4106 return (KERN_RESOURCE_SHORTAGE);
4110 vm_object_shadow(&entry->object.vm_object,
4111 &entry->offset, size);
4112 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
4113 eobject = entry->object.vm_object;
4114 if (eobject->cred != NULL) {
4116 * The object was not shadowed.
4118 swap_release_by_cred(size, entry->cred);
4119 crfree(entry->cred);
4121 } else if (entry->cred != NULL) {
4122 VM_OBJECT_WLOCK(eobject);
4123 eobject->cred = entry->cred;
4124 eobject->charge = size;
4125 VM_OBJECT_WUNLOCK(eobject);
4129 vm_map_lock_downgrade(map);
4132 * We're attempting to read a copy-on-write page --
4133 * don't allow writes.
4135 prot &= ~VM_PROT_WRITE;
4140 * Create an object if necessary.
4142 if (entry->object.vm_object == NULL &&
4144 if (vm_map_lock_upgrade(map))
4146 entry->object.vm_object = vm_object_allocate(OBJT_DEFAULT,
4149 if (entry->cred != NULL) {
4150 VM_OBJECT_WLOCK(entry->object.vm_object);
4151 entry->object.vm_object->cred = entry->cred;
4152 entry->object.vm_object->charge = size;
4153 VM_OBJECT_WUNLOCK(entry->object.vm_object);
4156 vm_map_lock_downgrade(map);
4160 * Return the object/offset from this entry. If the entry was
4161 * copy-on-write or empty, it has been fixed up.
4163 *pindex = UOFF_TO_IDX((vaddr - entry->start) + entry->offset);
4164 *object = entry->object.vm_object;
4167 return (KERN_SUCCESS);
4171 * vm_map_lookup_locked:
4173 * Lookup the faulting address. A version of vm_map_lookup that returns
4174 * KERN_FAILURE instead of blocking on map lock or memory allocation.
4177 vm_map_lookup_locked(vm_map_t *var_map, /* IN/OUT */
4179 vm_prot_t fault_typea,
4180 vm_map_entry_t *out_entry, /* OUT */
4181 vm_object_t *object, /* OUT */
4182 vm_pindex_t *pindex, /* OUT */
4183 vm_prot_t *out_prot, /* OUT */
4184 boolean_t *wired) /* OUT */
4186 vm_map_entry_t entry;
4187 vm_map_t map = *var_map;
4189 vm_prot_t fault_type = fault_typea;
4192 * Lookup the faulting address.
4194 if (!vm_map_lookup_entry(map, vaddr, out_entry))
4195 return (KERN_INVALID_ADDRESS);
4200 * Fail if the entry refers to a submap.
4202 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP)
4203 return (KERN_FAILURE);
4206 * Check whether this task is allowed to have this page.
4208 prot = entry->protection;
4209 fault_type &= VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE;
4210 if ((fault_type & prot) != fault_type)
4211 return (KERN_PROTECTION_FAILURE);
4214 * If this page is not pageable, we have to get it for all possible
4217 *wired = (entry->wired_count != 0);
4219 fault_type = entry->protection;
4221 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
4223 * Fail if the entry was copy-on-write for a write fault.
4225 if (fault_type & VM_PROT_WRITE)
4226 return (KERN_FAILURE);
4228 * We're attempting to read a copy-on-write page --
4229 * don't allow writes.
4231 prot &= ~VM_PROT_WRITE;
4235 * Fail if an object should be created.
4237 if (entry->object.vm_object == NULL && !map->system_map)
4238 return (KERN_FAILURE);
4241 * Return the object/offset from this entry. If the entry was
4242 * copy-on-write or empty, it has been fixed up.
4244 *pindex = UOFF_TO_IDX((vaddr - entry->start) + entry->offset);
4245 *object = entry->object.vm_object;
4248 return (KERN_SUCCESS);
4252 * vm_map_lookup_done:
4254 * Releases locks acquired by a vm_map_lookup
4255 * (according to the handle returned by that lookup).
4258 vm_map_lookup_done(vm_map_t map, vm_map_entry_t entry)
4261 * Unlock the main-level map
4263 vm_map_unlock_read(map);
4266 #include "opt_ddb.h"
4268 #include <sys/kernel.h>
4270 #include <ddb/ddb.h>
4273 vm_map_print(vm_map_t map)
4275 vm_map_entry_t entry;
4277 db_iprintf("Task map %p: pmap=%p, nentries=%d, version=%u\n",
4279 (void *)map->pmap, map->nentries, map->timestamp);
4282 for (entry = map->header.next; entry != &map->header;
4283 entry = entry->next) {
4284 db_iprintf("map entry %p: start=%p, end=%p\n",
4285 (void *)entry, (void *)entry->start, (void *)entry->end);
4287 static char *inheritance_name[4] =
4288 {"share", "copy", "none", "donate_copy"};
4290 db_iprintf(" prot=%x/%x/%s",
4292 entry->max_protection,
4293 inheritance_name[(int)(unsigned char)entry->inheritance]);
4294 if (entry->wired_count != 0)
4295 db_printf(", wired");
4297 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
4298 db_printf(", share=%p, offset=0x%jx\n",
4299 (void *)entry->object.sub_map,
4300 (uintmax_t)entry->offset);
4301 if ((entry->prev == &map->header) ||
4302 (entry->prev->object.sub_map !=
4303 entry->object.sub_map)) {
4305 vm_map_print((vm_map_t)entry->object.sub_map);
4309 if (entry->cred != NULL)
4310 db_printf(", ruid %d", entry->cred->cr_ruid);
4311 db_printf(", object=%p, offset=0x%jx",
4312 (void *)entry->object.vm_object,
4313 (uintmax_t)entry->offset);
4314 if (entry->object.vm_object && entry->object.vm_object->cred)
4315 db_printf(", obj ruid %d charge %jx",
4316 entry->object.vm_object->cred->cr_ruid,
4317 (uintmax_t)entry->object.vm_object->charge);
4318 if (entry->eflags & MAP_ENTRY_COW)
4319 db_printf(", copy (%s)",
4320 (entry->eflags & MAP_ENTRY_NEEDS_COPY) ? "needed" : "done");
4323 if ((entry->prev == &map->header) ||
4324 (entry->prev->object.vm_object !=
4325 entry->object.vm_object)) {
4327 vm_object_print((db_expr_t)(intptr_t)
4328 entry->object.vm_object,
4337 DB_SHOW_COMMAND(map, map)
4341 db_printf("usage: show map <addr>\n");
4344 vm_map_print((vm_map_t)addr);
4347 DB_SHOW_COMMAND(procvm, procvm)
4352 p = db_lookup_proc(addr);
4357 db_printf("p = %p, vmspace = %p, map = %p, pmap = %p\n",
4358 (void *)p, (void *)p->p_vmspace, (void *)&p->p_vmspace->vm_map,
4359 (void *)vmspace_pmap(p->p_vmspace));
4361 vm_map_print((vm_map_t)&p->p_vmspace->vm_map);