2 * SPDX-License-Identifier: (BSD-3-Clause AND MIT-CMU)
4 * Copyright (c) 1991, 1993
5 * The Regents of the University of California. All rights reserved.
7 * This code is derived from software contributed to Berkeley by
8 * The Mach Operating System project at Carnegie-Mellon University.
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 3. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * from: @(#)vm_map.c 8.3 (Berkeley) 1/12/94
37 * Copyright (c) 1987, 1990 Carnegie-Mellon University.
38 * All rights reserved.
40 * Authors: Avadis Tevanian, Jr., Michael Wayne Young
42 * Permission to use, copy, modify and distribute this software and
43 * its documentation is hereby granted, provided that both the copyright
44 * notice and this permission notice appear in all copies of the
45 * software, derivative works or modified versions, and any portions
46 * thereof, and that both notices appear in supporting documentation.
48 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
49 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
50 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
52 * Carnegie Mellon requests users of this software to return to
54 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
55 * School of Computer Science
56 * Carnegie Mellon University
57 * Pittsburgh PA 15213-3890
59 * any improvements or extensions that they make and grant Carnegie the
60 * rights to redistribute these changes.
64 * Virtual memory mapping module.
67 #include <sys/cdefs.h>
68 __FBSDID("$FreeBSD$");
70 #include <sys/param.h>
71 #include <sys/systm.h>
72 #include <sys/kernel.h>
75 #include <sys/mutex.h>
77 #include <sys/vmmeter.h>
79 #include <sys/vnode.h>
80 #include <sys/racct.h>
81 #include <sys/resourcevar.h>
82 #include <sys/rwlock.h>
84 #include <sys/sysctl.h>
85 #include <sys/sysent.h>
89 #include <vm/vm_param.h>
91 #include <vm/vm_map.h>
92 #include <vm/vm_page.h>
93 #include <vm/vm_pageout.h>
94 #include <vm/vm_object.h>
95 #include <vm/vm_pager.h>
96 #include <vm/vm_kern.h>
97 #include <vm/vm_extern.h>
98 #include <vm/vnode_pager.h>
99 #include <vm/swap_pager.h>
103 * Virtual memory maps provide for the mapping, protection,
104 * and sharing of virtual memory objects. In addition,
105 * this module provides for an efficient virtual copy of
106 * memory from one map to another.
108 * Synchronization is required prior to most operations.
110 * Maps consist of an ordered doubly-linked list of simple
111 * entries; a self-adjusting binary search tree of these
112 * entries is used to speed up lookups.
114 * Since portions of maps are specified by start/end addresses,
115 * which may not align with existing map entries, all
116 * routines merely "clip" entries to these start/end values.
117 * [That is, an entry is split into two, bordering at a
118 * start or end value.] Note that these clippings may not
119 * always be necessary (as the two resulting entries are then
120 * not changed); however, the clipping is done for convenience.
122 * As mentioned above, virtual copy operations are performed
123 * by copying VM object references from one map to
124 * another, and then marking both regions as copy-on-write.
127 static struct mtx map_sleep_mtx;
128 static uma_zone_t mapentzone;
129 static uma_zone_t kmapentzone;
130 static uma_zone_t mapzone;
131 static uma_zone_t vmspace_zone;
132 static int vmspace_zinit(void *mem, int size, int flags);
133 static int vm_map_zinit(void *mem, int ize, int flags);
134 static void _vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min,
136 static void vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map);
137 static void vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry);
138 static void vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry);
139 static int vm_map_growstack(vm_map_t map, vm_offset_t addr,
140 vm_map_entry_t gap_entry);
141 static void vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot,
142 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags);
144 static void vm_map_zdtor(void *mem, int size, void *arg);
145 static void vmspace_zdtor(void *mem, int size, void *arg);
147 static int vm_map_stack_locked(vm_map_t map, vm_offset_t addrbos,
148 vm_size_t max_ssize, vm_size_t growsize, vm_prot_t prot, vm_prot_t max,
150 static void vm_map_wire_entry_failure(vm_map_t map, vm_map_entry_t entry,
151 vm_offset_t failed_addr);
153 #define ENTRY_CHARGED(e) ((e)->cred != NULL || \
154 ((e)->object.vm_object != NULL && (e)->object.vm_object->cred != NULL && \
155 !((e)->eflags & MAP_ENTRY_NEEDS_COPY)))
158 * PROC_VMSPACE_{UN,}LOCK() can be a noop as long as vmspaces are type
161 #define PROC_VMSPACE_LOCK(p) do { } while (0)
162 #define PROC_VMSPACE_UNLOCK(p) do { } while (0)
165 * VM_MAP_RANGE_CHECK: [ internal use only ]
167 * Asserts that the starting and ending region
168 * addresses fall within the valid range of the map.
170 #define VM_MAP_RANGE_CHECK(map, start, end) \
172 if (start < vm_map_min(map)) \
173 start = vm_map_min(map); \
174 if (end > vm_map_max(map)) \
175 end = vm_map_max(map); \
183 * Initialize the vm_map module. Must be called before
184 * any other vm_map routines.
186 * Map and entry structures are allocated from the general
187 * purpose memory pool with some exceptions:
189 * - The kernel map and kmem submap are allocated statically.
190 * - Kernel map entries are allocated out of a static pool.
192 * These restrictions are necessary since malloc() uses the
193 * maps and requires map entries.
199 mtx_init(&map_sleep_mtx, "vm map sleep mutex", NULL, MTX_DEF);
200 mapzone = uma_zcreate("MAP", sizeof(struct vm_map), NULL,
206 vm_map_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
207 uma_prealloc(mapzone, MAX_KMAP);
208 kmapentzone = uma_zcreate("KMAP ENTRY", sizeof(struct vm_map_entry),
209 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR,
210 UMA_ZONE_MTXCLASS | UMA_ZONE_VM);
211 mapentzone = uma_zcreate("MAP ENTRY", sizeof(struct vm_map_entry),
212 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
213 vmspace_zone = uma_zcreate("VMSPACE", sizeof(struct vmspace), NULL,
219 vmspace_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
223 vmspace_zinit(void *mem, int size, int flags)
227 vm = (struct vmspace *)mem;
229 vm->vm_map.pmap = NULL;
230 (void)vm_map_zinit(&vm->vm_map, sizeof(vm->vm_map), flags);
231 PMAP_LOCK_INIT(vmspace_pmap(vm));
236 vm_map_zinit(void *mem, int size, int flags)
241 memset(map, 0, sizeof(*map));
242 mtx_init(&map->system_mtx, "vm map (system)", NULL, MTX_DEF | MTX_DUPOK);
243 sx_init(&map->lock, "vm map (user)");
249 vmspace_zdtor(void *mem, int size, void *arg)
253 vm = (struct vmspace *)mem;
255 vm_map_zdtor(&vm->vm_map, sizeof(vm->vm_map), arg);
258 vm_map_zdtor(void *mem, int size, void *arg)
263 KASSERT(map->nentries == 0,
264 ("map %p nentries == %d on free.",
265 map, map->nentries));
266 KASSERT(map->size == 0,
267 ("map %p size == %lu on free.",
268 map, (unsigned long)map->size));
270 #endif /* INVARIANTS */
273 * Allocate a vmspace structure, including a vm_map and pmap,
274 * and initialize those structures. The refcnt is set to 1.
276 * If 'pinit' is NULL then the embedded pmap is initialized via pmap_pinit().
279 vmspace_alloc(vm_offset_t min, vm_offset_t max, pmap_pinit_t pinit)
283 vm = uma_zalloc(vmspace_zone, M_WAITOK);
284 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_map_min(&vm->vm_map),
336 vm_map_max(&vm->vm_map));
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);
392 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_fcmpset_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);
438 refcnt = vm->vm_refcnt;
440 if (refcnt <= 0) { /* Avoid 0->1 transition */
441 PROC_VMSPACE_UNLOCK(p);
444 } while (!atomic_fcmpset_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 new vmspace is either the vmspace of a user process obtained
458 * from an active AIO request or the initial vmspace of the AIO kernel
459 * process (when it is idling). Because user processes will block to
460 * drain any active AIO requests before proceeding in exit() or
461 * execve(), the reference count for vmspaces from AIO requests can
462 * never be 0. Similarly, AIO kernel processes hold an extra
463 * reference on their initial vmspace for the life of the process. As
464 * a result, the 'newvm' vmspace always has a non-zero reference
465 * count. This permits an additional reference on 'newvm' to be
466 * acquired via a simple atomic increment rather than the loop in
467 * vmspace_acquire_ref() above.
470 vmspace_switch_aio(struct vmspace *newvm)
472 struct vmspace *oldvm;
474 /* XXX: Need some way to assert that this is an aio daemon. */
476 KASSERT(newvm->vm_refcnt > 0,
477 ("vmspace_switch_aio: newvm unreferenced"));
479 oldvm = curproc->p_vmspace;
484 * Point to the new address space and refer to it.
486 curproc->p_vmspace = newvm;
487 atomic_add_int(&newvm->vm_refcnt, 1);
489 /* Activate the new mapping. */
490 pmap_activate(curthread);
496 _vm_map_lock(vm_map_t map, const char *file, int line)
500 mtx_lock_flags_(&map->system_mtx, 0, file, line);
502 sx_xlock_(&map->lock, file, line);
507 vm_map_entry_set_vnode_text(vm_map_entry_t entry, bool add)
509 vm_object_t object, object1;
512 if ((entry->eflags & MAP_ENTRY_VN_EXEC) == 0)
514 KASSERT((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0,
515 ("Submap with execs"));
516 object = entry->object.vm_object;
517 KASSERT(object != NULL, ("No object for text, entry %p", entry));
518 VM_OBJECT_RLOCK(object);
519 while ((object1 = object->backing_object) != NULL) {
520 VM_OBJECT_RLOCK(object1);
521 VM_OBJECT_RUNLOCK(object);
526 if (object->type == OBJT_DEAD) {
528 * For OBJT_DEAD objects, v_writecount was handled in
529 * vnode_pager_dealloc().
531 } else if (object->type == OBJT_VNODE) {
533 } else if (object->type == OBJT_SWAP) {
534 KASSERT((object->flags & OBJ_TMPFS_NODE) != 0,
535 ("vm_map_entry_set_vnode_text: swap and !TMPFS "
536 "entry %p, object %p, add %d", entry, object, add));
538 * Tmpfs VREG node, which was reclaimed, has
539 * OBJ_TMPFS_NODE flag set, but not OBJ_TMPFS. In
540 * this case there is no v_writecount to adjust.
542 if ((object->flags & OBJ_TMPFS) != 0)
543 vp = object->un_pager.swp.swp_tmpfs;
546 ("vm_map_entry_set_vnode_text: wrong object type, "
547 "entry %p, object %p, add %d", entry, object, add));
551 VOP_SET_TEXT_CHECKED(vp);
553 VOP_UNSET_TEXT_CHECKED(vp);
555 VM_OBJECT_RUNLOCK(object);
559 vm_map_process_deferred(void)
562 vm_map_entry_t entry, next;
566 entry = td->td_map_def_user;
567 td->td_map_def_user = NULL;
568 while (entry != NULL) {
570 MPASS((entry->eflags & (MAP_ENTRY_VN_WRITECNT |
571 MAP_ENTRY_VN_EXEC)) != (MAP_ENTRY_VN_WRITECNT |
573 if ((entry->eflags & MAP_ENTRY_VN_WRITECNT) != 0) {
575 * Decrement the object's writemappings and
576 * possibly the vnode's v_writecount.
578 KASSERT((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0,
579 ("Submap with writecount"));
580 object = entry->object.vm_object;
581 KASSERT(object != NULL, ("No object for writecount"));
582 vnode_pager_release_writecount(object, entry->start,
585 vm_map_entry_set_vnode_text(entry, false);
586 vm_map_entry_deallocate(entry, FALSE);
592 _vm_map_unlock(vm_map_t map, const char *file, int line)
596 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
598 sx_xunlock_(&map->lock, file, line);
599 vm_map_process_deferred();
604 _vm_map_lock_read(vm_map_t map, const char *file, int line)
608 mtx_lock_flags_(&map->system_mtx, 0, file, line);
610 sx_slock_(&map->lock, file, line);
614 _vm_map_unlock_read(vm_map_t map, const char *file, int line)
618 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
620 sx_sunlock_(&map->lock, file, line);
621 vm_map_process_deferred();
626 _vm_map_trylock(vm_map_t map, const char *file, int line)
630 error = map->system_map ?
631 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) :
632 !sx_try_xlock_(&map->lock, file, line);
639 _vm_map_trylock_read(vm_map_t map, const char *file, int line)
643 error = map->system_map ?
644 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) :
645 !sx_try_slock_(&map->lock, file, line);
650 * _vm_map_lock_upgrade: [ internal use only ]
652 * Tries to upgrade a read (shared) lock on the specified map to a write
653 * (exclusive) lock. Returns the value "0" if the upgrade succeeds and a
654 * non-zero value if the upgrade fails. If the upgrade fails, the map is
655 * returned without a read or write lock held.
657 * Requires that the map be read locked.
660 _vm_map_lock_upgrade(vm_map_t map, const char *file, int line)
662 unsigned int last_timestamp;
664 if (map->system_map) {
665 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
667 if (!sx_try_upgrade_(&map->lock, file, line)) {
668 last_timestamp = map->timestamp;
669 sx_sunlock_(&map->lock, file, line);
670 vm_map_process_deferred();
672 * If the map's timestamp does not change while the
673 * map is unlocked, then the upgrade succeeds.
675 sx_xlock_(&map->lock, file, line);
676 if (last_timestamp != map->timestamp) {
677 sx_xunlock_(&map->lock, file, line);
687 _vm_map_lock_downgrade(vm_map_t map, const char *file, int line)
690 if (map->system_map) {
691 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
693 sx_downgrade_(&map->lock, file, line);
699 * Returns a non-zero value if the caller holds a write (exclusive) lock
700 * on the specified map and the value "0" otherwise.
703 vm_map_locked(vm_map_t map)
707 return (mtx_owned(&map->system_mtx));
709 return (sx_xlocked(&map->lock));
714 _vm_map_assert_locked(vm_map_t map, const char *file, int line)
718 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
720 sx_assert_(&map->lock, SA_XLOCKED, file, line);
723 #define VM_MAP_ASSERT_LOCKED(map) \
724 _vm_map_assert_locked(map, LOCK_FILE, LOCK_LINE)
727 static int enable_vmmap_check = 1;
729 static int enable_vmmap_check = 0;
731 SYSCTL_INT(_debug, OID_AUTO, vmmap_check, CTLFLAG_RWTUN,
732 &enable_vmmap_check, 0, "Enable vm map consistency checking");
735 _vm_map_assert_consistent(vm_map_t map)
737 vm_map_entry_t child, entry, prev;
738 vm_size_t max_left, max_right;
740 if (!enable_vmmap_check)
743 for (prev = &map->header; (entry = prev->next) != &map->header;
745 KASSERT(prev->end <= entry->start,
746 ("map %p prev->end = %jx, start = %jx", map,
747 (uintmax_t)prev->end, (uintmax_t)entry->start));
748 KASSERT(entry->start < entry->end,
749 ("map %p start = %jx, end = %jx", map,
750 (uintmax_t)entry->start, (uintmax_t)entry->end));
751 KASSERT(entry->end <= entry->next->start,
752 ("map %p end = %jx, next->start = %jx", map,
753 (uintmax_t)entry->end, (uintmax_t)entry->next->start));
754 KASSERT(entry->left == NULL ||
755 entry->left->start < entry->start,
756 ("map %p left->start = %jx, start = %jx", map,
757 (uintmax_t)entry->left->start, (uintmax_t)entry->start));
758 KASSERT(entry->right == NULL ||
759 entry->start < entry->right->start,
760 ("map %p start = %jx, right->start = %jx", map,
761 (uintmax_t)entry->start, (uintmax_t)entry->right->start));
763 max_left = (child != NULL) ? child->max_free :
764 entry->start - prev->end;
765 child = entry->right;
766 max_right = (child != NULL) ? child->max_free :
767 entry->next->start - entry->end;
768 KASSERT(entry->max_free == MAX(max_left, max_right),
769 ("map %p max = %jx, max_left = %jx, max_right = %jx", map,
770 (uintmax_t)entry->max_free,
771 (uintmax_t)max_left, (uintmax_t)max_right));
775 #define VM_MAP_ASSERT_CONSISTENT(map) \
776 _vm_map_assert_consistent(map)
778 #define VM_MAP_ASSERT_LOCKED(map)
779 #define VM_MAP_ASSERT_CONSISTENT(map)
780 #endif /* INVARIANTS */
783 * _vm_map_unlock_and_wait:
785 * Atomically releases the lock on the specified map and puts the calling
786 * thread to sleep. The calling thread will remain asleep until either
787 * vm_map_wakeup() is performed on the map or the specified timeout is
790 * WARNING! This function does not perform deferred deallocations of
791 * objects and map entries. Therefore, the calling thread is expected to
792 * reacquire the map lock after reawakening and later perform an ordinary
793 * unlock operation, such as vm_map_unlock(), before completing its
794 * operation on the map.
797 _vm_map_unlock_and_wait(vm_map_t map, int timo, const char *file, int line)
800 mtx_lock(&map_sleep_mtx);
802 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
804 sx_xunlock_(&map->lock, file, line);
805 return (msleep(&map->root, &map_sleep_mtx, PDROP | PVM, "vmmaps",
812 * Awaken any threads that have slept on the map using
813 * vm_map_unlock_and_wait().
816 vm_map_wakeup(vm_map_t map)
820 * Acquire and release map_sleep_mtx to prevent a wakeup()
821 * from being performed (and lost) between the map unlock
822 * and the msleep() in _vm_map_unlock_and_wait().
824 mtx_lock(&map_sleep_mtx);
825 mtx_unlock(&map_sleep_mtx);
830 vm_map_busy(vm_map_t map)
833 VM_MAP_ASSERT_LOCKED(map);
838 vm_map_unbusy(vm_map_t map)
841 VM_MAP_ASSERT_LOCKED(map);
842 KASSERT(map->busy, ("vm_map_unbusy: not busy"));
843 if (--map->busy == 0 && (map->flags & MAP_BUSY_WAKEUP)) {
844 vm_map_modflags(map, 0, MAP_BUSY_WAKEUP);
850 vm_map_wait_busy(vm_map_t map)
853 VM_MAP_ASSERT_LOCKED(map);
855 vm_map_modflags(map, MAP_BUSY_WAKEUP, 0);
857 msleep(&map->busy, &map->system_mtx, 0, "mbusy", 0);
859 sx_sleep(&map->busy, &map->lock, 0, "mbusy", 0);
865 vmspace_resident_count(struct vmspace *vmspace)
867 return pmap_resident_count(vmspace_pmap(vmspace));
873 * Creates and returns a new empty VM map with
874 * the given physical map structure, and having
875 * the given lower and upper address bounds.
878 vm_map_create(pmap_t pmap, vm_offset_t min, vm_offset_t max)
882 result = uma_zalloc(mapzone, M_WAITOK);
883 CTR1(KTR_VM, "vm_map_create: %p", result);
884 _vm_map_init(result, pmap, min, max);
889 * Initialize an existing vm_map structure
890 * such as that in the vmspace structure.
893 _vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max)
896 map->header.next = map->header.prev = &map->header;
897 map->header.eflags = MAP_ENTRY_HEADER;
898 map->needs_wakeup = FALSE;
901 map->header.end = min;
902 map->header.start = max;
911 vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max)
914 _vm_map_init(map, pmap, min, max);
915 mtx_init(&map->system_mtx, "system map", NULL, MTX_DEF | MTX_DUPOK);
916 sx_init(&map->lock, "user map");
920 * vm_map_entry_dispose: [ internal use only ]
922 * Inverse of vm_map_entry_create.
925 vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry)
927 uma_zfree(map->system_map ? kmapentzone : mapentzone, entry);
931 * vm_map_entry_create: [ internal use only ]
933 * Allocates a VM map entry for insertion.
934 * No entry fields are filled in.
936 static vm_map_entry_t
937 vm_map_entry_create(vm_map_t map)
939 vm_map_entry_t new_entry;
942 new_entry = uma_zalloc(kmapentzone, M_NOWAIT);
944 new_entry = uma_zalloc(mapentzone, M_WAITOK);
945 if (new_entry == NULL)
946 panic("vm_map_entry_create: kernel resources exhausted");
951 * vm_map_entry_set_behavior:
953 * Set the expected access behavior, either normal, random, or
957 vm_map_entry_set_behavior(vm_map_entry_t entry, u_char behavior)
959 entry->eflags = (entry->eflags & ~MAP_ENTRY_BEHAV_MASK) |
960 (behavior & MAP_ENTRY_BEHAV_MASK);
964 * vm_map_entry_max_free_{left,right}:
966 * Compute the size of the largest free gap between two entries,
967 * one the root of a tree and the other the ancestor of that root
968 * that is the least or greatest ancestor found on the search path.
970 static inline vm_size_t
971 vm_map_entry_max_free_left(vm_map_entry_t root, vm_map_entry_t left_ancestor)
974 return (root->left != NULL ?
975 root->left->max_free : root->start - left_ancestor->end);
978 static inline vm_size_t
979 vm_map_entry_max_free_right(vm_map_entry_t root, vm_map_entry_t right_ancestor)
982 return (root->right != NULL ?
983 root->right->max_free : right_ancestor->start - root->end);
986 #define SPLAY_LEFT_STEP(root, y, rlist, test) do { \
987 vm_size_t max_free; \
990 * Infer root->right->max_free == root->max_free when \
991 * y->max_free < root->max_free || root->max_free == 0. \
992 * Otherwise, look right to find it. \
995 max_free = root->max_free; \
996 KASSERT(max_free >= vm_map_entry_max_free_right(root, rlist), \
997 ("%s: max_free invariant fails", __func__)); \
998 if (y == NULL ? max_free > 0 : max_free - 1 < y->max_free) \
999 max_free = vm_map_entry_max_free_right(root, rlist); \
1000 if (y != NULL && (test)) { \
1001 /* Rotate right and make y root. */ \
1002 root->left = y->right; \
1004 if (max_free < y->max_free) \
1005 root->max_free = max_free = MAX(max_free, \
1006 vm_map_entry_max_free_left(root, y)); \
1010 /* Copy right->max_free. Put root on rlist. */ \
1011 root->max_free = max_free; \
1012 KASSERT(max_free == vm_map_entry_max_free_right(root, rlist), \
1013 ("%s: max_free not copied from right", __func__)); \
1014 root->left = rlist; \
1019 #define SPLAY_RIGHT_STEP(root, y, llist, test) do { \
1020 vm_size_t max_free; \
1023 * Infer root->left->max_free == root->max_free when \
1024 * y->max_free < root->max_free || root->max_free == 0. \
1025 * Otherwise, look left to find it. \
1028 max_free = root->max_free; \
1029 KASSERT(max_free >= vm_map_entry_max_free_left(root, llist), \
1030 ("%s: max_free invariant fails", __func__)); \
1031 if (y == NULL ? max_free > 0 : max_free - 1 < y->max_free) \
1032 max_free = vm_map_entry_max_free_left(root, llist); \
1033 if (y != NULL && (test)) { \
1034 /* Rotate left and make y root. */ \
1035 root->right = y->left; \
1037 if (max_free < y->max_free) \
1038 root->max_free = max_free = MAX(max_free, \
1039 vm_map_entry_max_free_right(root, y)); \
1043 /* Copy left->max_free. Put root on llist. */ \
1044 root->max_free = max_free; \
1045 KASSERT(max_free == vm_map_entry_max_free_left(root, llist), \
1046 ("%s: max_free not copied from left", __func__)); \
1047 root->right = llist; \
1053 * Walk down the tree until we find addr or a NULL pointer where addr would go,
1054 * breaking off left and right subtrees of nodes less than, or greater than
1055 * addr. Treat pointers to nodes with max_free < length as NULL pointers.
1056 * llist and rlist are the two sides in reverse order (bottom-up), with llist
1057 * linked by the right pointer and rlist linked by the left pointer in the
1058 * vm_map_entry, and both lists terminated by &map->header. This function, and
1059 * the subsequent call to vm_map_splay_merge, rely on the start and end address
1060 * values in &map->header.
1062 static vm_map_entry_t
1063 vm_map_splay_split(vm_map_t map, vm_offset_t addr, vm_size_t length,
1064 vm_map_entry_t *out_llist, vm_map_entry_t *out_rlist)
1066 vm_map_entry_t llist, rlist, root, y;
1068 llist = rlist = &map->header;
1070 while (root != NULL && root->max_free >= length) {
1071 KASSERT(llist->end <= root->start && root->end <= rlist->start,
1072 ("%s: root not within tree bounds", __func__));
1073 if (addr < root->start) {
1074 SPLAY_LEFT_STEP(root, y, rlist,
1075 y->max_free >= length && addr < y->start);
1076 } else if (addr >= root->end) {
1077 SPLAY_RIGHT_STEP(root, y, llist,
1078 y->max_free >= length && addr >= y->end);
1088 vm_map_splay_findnext(vm_map_entry_t root, vm_map_entry_t *iolist)
1090 vm_map_entry_t rlist, y;
1094 while (root != NULL)
1095 SPLAY_LEFT_STEP(root, y, rlist, true);
1100 vm_map_splay_findprev(vm_map_entry_t root, vm_map_entry_t *iolist)
1102 vm_map_entry_t llist, y;
1106 while (root != NULL)
1107 SPLAY_RIGHT_STEP(root, y, llist, true);
1112 vm_map_entry_swap(vm_map_entry_t *a, vm_map_entry_t *b)
1122 * Walk back up the two spines, flip the pointers and set max_free. The
1123 * subtrees of the root go at the bottom of llist and rlist.
1126 vm_map_splay_merge(vm_map_t map, vm_map_entry_t root,
1127 vm_map_entry_t llist, vm_map_entry_t rlist)
1129 vm_map_entry_t prev;
1130 vm_size_t max_free_left, max_free_right;
1132 max_free_left = vm_map_entry_max_free_left(root, llist);
1133 if (llist != &map->header) {
1137 * The max_free values of the children of llist are in
1138 * llist->max_free and max_free_left. Update with the
1141 llist->max_free = max_free_left =
1142 MAX(llist->max_free, max_free_left);
1143 vm_map_entry_swap(&llist->right, &prev);
1144 vm_map_entry_swap(&prev, &llist);
1145 } while (llist != &map->header);
1148 max_free_right = vm_map_entry_max_free_right(root, rlist);
1149 if (rlist != &map->header) {
1153 * The max_free values of the children of rlist are in
1154 * rlist->max_free and max_free_right. Update with the
1157 rlist->max_free = max_free_right =
1158 MAX(rlist->max_free, max_free_right);
1159 vm_map_entry_swap(&rlist->left, &prev);
1160 vm_map_entry_swap(&prev, &rlist);
1161 } while (rlist != &map->header);
1164 root->max_free = MAX(max_free_left, max_free_right);
1171 * The Sleator and Tarjan top-down splay algorithm with the
1172 * following variation. Max_free must be computed bottom-up, so
1173 * on the downward pass, maintain the left and right spines in
1174 * reverse order. Then, make a second pass up each side to fix
1175 * the pointers and compute max_free. The time bound is O(log n)
1178 * The new root is the vm_map_entry containing "addr", or else an
1179 * adjacent entry (lower if possible) if addr is not in the tree.
1181 * The map must be locked, and leaves it so.
1183 * Returns: the new root.
1185 static vm_map_entry_t
1186 vm_map_splay(vm_map_t map, vm_offset_t addr)
1188 vm_map_entry_t llist, rlist, root;
1190 root = vm_map_splay_split(map, addr, 0, &llist, &rlist);
1193 } else if (llist != &map->header) {
1195 * Recover the greatest node in the left
1196 * subtree and make it the root.
1199 llist = root->right;
1201 } else if (rlist != &map->header) {
1203 * Recover the least node in the right
1204 * subtree and make it the root.
1210 /* There is no root. */
1213 vm_map_splay_merge(map, root, llist, rlist);
1214 VM_MAP_ASSERT_CONSISTENT(map);
1219 * vm_map_entry_{un,}link:
1221 * Insert/remove entries from maps.
1224 vm_map_entry_link(vm_map_t map, vm_map_entry_t entry)
1226 vm_map_entry_t llist, rlist, root;
1229 "vm_map_entry_link: map %p, nentries %d, entry %p", map,
1230 map->nentries, entry);
1231 VM_MAP_ASSERT_LOCKED(map);
1233 root = vm_map_splay_split(map, entry->start, 0, &llist, &rlist);
1234 KASSERT(root == NULL,
1235 ("vm_map_entry_link: link object already mapped"));
1236 entry->prev = llist;
1237 entry->next = rlist;
1238 llist->next = rlist->prev = entry;
1239 entry->left = entry->right = NULL;
1240 vm_map_splay_merge(map, entry, llist, rlist);
1241 VM_MAP_ASSERT_CONSISTENT(map);
1244 enum unlink_merge_type {
1251 vm_map_entry_unlink(vm_map_t map, vm_map_entry_t entry,
1252 enum unlink_merge_type op)
1254 vm_map_entry_t llist, rlist, root, y;
1256 VM_MAP_ASSERT_LOCKED(map);
1257 root = vm_map_splay_split(map, entry->start, 0, &llist, &rlist);
1258 KASSERT(root != NULL,
1259 ("vm_map_entry_unlink: unlink object not mapped"));
1262 case UNLINK_MERGE_PREV:
1263 vm_map_splay_findprev(root, &llist);
1264 llist->end = root->end;
1267 llist = root->right;
1270 case UNLINK_MERGE_NEXT:
1271 vm_map_splay_findnext(root, &rlist);
1272 rlist->start = root->start;
1273 rlist->offset = root->offset;
1279 case UNLINK_MERGE_NONE:
1280 vm_map_splay_findprev(root, &llist);
1281 vm_map_splay_findnext(root, &rlist);
1282 if (llist != &map->header) {
1284 llist = root->right;
1286 } else if (rlist != &map->header) {
1295 y->prev = entry->prev;
1298 vm_map_splay_merge(map, root, llist, rlist);
1301 VM_MAP_ASSERT_CONSISTENT(map);
1303 CTR3(KTR_VM, "vm_map_entry_unlink: map %p, nentries %d, entry %p", map,
1304 map->nentries, entry);
1308 * vm_map_entry_resize:
1310 * Resize a vm_map_entry, recompute the amount of free space that
1311 * follows it and propagate that value up the tree.
1313 * The map must be locked, and leaves it so.
1316 vm_map_entry_resize(vm_map_t map, vm_map_entry_t entry, vm_size_t grow_amount)
1318 vm_map_entry_t llist, rlist, root;
1320 VM_MAP_ASSERT_LOCKED(map);
1321 root = vm_map_splay_split(map, entry->start, 0, &llist, &rlist);
1322 KASSERT(root != NULL,
1323 ("%s: resize object not mapped", __func__));
1324 vm_map_splay_findnext(root, &rlist);
1326 entry->end += grow_amount;
1327 vm_map_splay_merge(map, root, llist, rlist);
1328 VM_MAP_ASSERT_CONSISTENT(map);
1329 CTR4(KTR_VM, "%s: map %p, nentries %d, entry %p",
1330 __func__, map, map->nentries, entry);
1334 * vm_map_lookup_entry: [ internal use only ]
1336 * Finds the map entry containing (or
1337 * immediately preceding) the specified address
1338 * in the given map; the entry is returned
1339 * in the "entry" parameter. The boolean
1340 * result indicates whether the address is
1341 * actually contained in the map.
1344 vm_map_lookup_entry(
1346 vm_offset_t address,
1347 vm_map_entry_t *entry) /* OUT */
1349 vm_map_entry_t cur, lbound;
1353 * If the map is empty, then the map entry immediately preceding
1354 * "address" is the map's header.
1358 *entry = &map->header;
1361 if (address >= cur->start && cur->end > address) {
1365 if ((locked = vm_map_locked(map)) ||
1366 sx_try_upgrade(&map->lock)) {
1368 * Splay requires a write lock on the map. However, it only
1369 * restructures the binary search tree; it does not otherwise
1370 * change the map. Thus, the map's timestamp need not change
1371 * on a temporary upgrade.
1373 cur = vm_map_splay(map, address);
1375 sx_downgrade(&map->lock);
1378 * If "address" is contained within a map entry, the new root
1379 * is that map entry. Otherwise, the new root is a map entry
1380 * immediately before or after "address".
1382 if (address < cur->start) {
1383 *entry = &map->header;
1387 return (address < cur->end);
1390 * Since the map is only locked for read access, perform a
1391 * standard binary search tree lookup for "address".
1393 lbound = &map->header;
1395 if (address < cur->start) {
1397 } else if (cur->end <= address) {
1404 } while (cur != NULL);
1412 * Inserts the given whole VM object into the target
1413 * map at the specified address range. The object's
1414 * size should match that of the address range.
1416 * Requires that the map be locked, and leaves it so.
1418 * If object is non-NULL, ref count must be bumped by caller
1419 * prior to making call to account for the new entry.
1422 vm_map_insert(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1423 vm_offset_t start, vm_offset_t end, vm_prot_t prot, vm_prot_t max, int cow)
1425 vm_map_entry_t new_entry, prev_entry;
1427 vm_eflags_t protoeflags;
1428 vm_inherit_t inheritance;
1430 VM_MAP_ASSERT_LOCKED(map);
1431 KASSERT(object != kernel_object ||
1432 (cow & MAP_COPY_ON_WRITE) == 0,
1433 ("vm_map_insert: kernel object and COW"));
1434 KASSERT(object == NULL || (cow & MAP_NOFAULT) == 0,
1435 ("vm_map_insert: paradoxical MAP_NOFAULT request"));
1436 KASSERT((prot & ~max) == 0,
1437 ("prot %#x is not subset of max_prot %#x", prot, max));
1440 * Check that the start and end points are not bogus.
1442 if (start < vm_map_min(map) || end > vm_map_max(map) ||
1444 return (KERN_INVALID_ADDRESS);
1447 * Find the entry prior to the proposed starting address; if it's part
1448 * of an existing entry, this range is bogus.
1450 if (vm_map_lookup_entry(map, start, &prev_entry))
1451 return (KERN_NO_SPACE);
1454 * Assert that the next entry doesn't overlap the end point.
1456 if (prev_entry->next->start < end)
1457 return (KERN_NO_SPACE);
1459 if ((cow & MAP_CREATE_GUARD) != 0 && (object != NULL ||
1460 max != VM_PROT_NONE))
1461 return (KERN_INVALID_ARGUMENT);
1464 if (cow & MAP_COPY_ON_WRITE)
1465 protoeflags |= MAP_ENTRY_COW | MAP_ENTRY_NEEDS_COPY;
1466 if (cow & MAP_NOFAULT)
1467 protoeflags |= MAP_ENTRY_NOFAULT;
1468 if (cow & MAP_DISABLE_SYNCER)
1469 protoeflags |= MAP_ENTRY_NOSYNC;
1470 if (cow & MAP_DISABLE_COREDUMP)
1471 protoeflags |= MAP_ENTRY_NOCOREDUMP;
1472 if (cow & MAP_STACK_GROWS_DOWN)
1473 protoeflags |= MAP_ENTRY_GROWS_DOWN;
1474 if (cow & MAP_STACK_GROWS_UP)
1475 protoeflags |= MAP_ENTRY_GROWS_UP;
1476 if (cow & MAP_VN_WRITECOUNT)
1477 protoeflags |= MAP_ENTRY_VN_WRITECNT;
1478 if (cow & MAP_VN_EXEC)
1479 protoeflags |= MAP_ENTRY_VN_EXEC;
1480 if ((cow & MAP_CREATE_GUARD) != 0)
1481 protoeflags |= MAP_ENTRY_GUARD;
1482 if ((cow & MAP_CREATE_STACK_GAP_DN) != 0)
1483 protoeflags |= MAP_ENTRY_STACK_GAP_DN;
1484 if ((cow & MAP_CREATE_STACK_GAP_UP) != 0)
1485 protoeflags |= MAP_ENTRY_STACK_GAP_UP;
1486 if (cow & MAP_INHERIT_SHARE)
1487 inheritance = VM_INHERIT_SHARE;
1489 inheritance = VM_INHERIT_DEFAULT;
1492 if ((cow & (MAP_ACC_NO_CHARGE | MAP_NOFAULT | MAP_CREATE_GUARD)) != 0)
1494 if ((cow & MAP_ACC_CHARGED) || ((prot & VM_PROT_WRITE) &&
1495 ((protoeflags & MAP_ENTRY_NEEDS_COPY) || object == NULL))) {
1496 if (!(cow & MAP_ACC_CHARGED) && !swap_reserve(end - start))
1497 return (KERN_RESOURCE_SHORTAGE);
1498 KASSERT(object == NULL ||
1499 (protoeflags & MAP_ENTRY_NEEDS_COPY) != 0 ||
1500 object->cred == NULL,
1501 ("overcommit: vm_map_insert o %p", object));
1502 cred = curthread->td_ucred;
1506 /* Expand the kernel pmap, if necessary. */
1507 if (map == kernel_map && end > kernel_vm_end)
1508 pmap_growkernel(end);
1509 if (object != NULL) {
1511 * OBJ_ONEMAPPING must be cleared unless this mapping
1512 * is trivially proven to be the only mapping for any
1513 * of the object's pages. (Object granularity
1514 * reference counting is insufficient to recognize
1515 * aliases with precision.)
1517 VM_OBJECT_WLOCK(object);
1518 if (object->ref_count > 1 || object->shadow_count != 0)
1519 vm_object_clear_flag(object, OBJ_ONEMAPPING);
1520 VM_OBJECT_WUNLOCK(object);
1521 } else if ((prev_entry->eflags & ~MAP_ENTRY_USER_WIRED) ==
1523 (cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP |
1524 MAP_VN_EXEC)) == 0 &&
1525 prev_entry->end == start && (prev_entry->cred == cred ||
1526 (prev_entry->object.vm_object != NULL &&
1527 prev_entry->object.vm_object->cred == cred)) &&
1528 vm_object_coalesce(prev_entry->object.vm_object,
1530 (vm_size_t)(prev_entry->end - prev_entry->start),
1531 (vm_size_t)(end - prev_entry->end), cred != NULL &&
1532 (protoeflags & MAP_ENTRY_NEEDS_COPY) == 0)) {
1534 * We were able to extend the object. Determine if we
1535 * can extend the previous map entry to include the
1536 * new range as well.
1538 if (prev_entry->inheritance == inheritance &&
1539 prev_entry->protection == prot &&
1540 prev_entry->max_protection == max &&
1541 prev_entry->wired_count == 0) {
1542 KASSERT((prev_entry->eflags & MAP_ENTRY_USER_WIRED) ==
1543 0, ("prev_entry %p has incoherent wiring",
1545 if ((prev_entry->eflags & MAP_ENTRY_GUARD) == 0)
1546 map->size += end - prev_entry->end;
1547 vm_map_entry_resize(map, prev_entry,
1548 end - prev_entry->end);
1549 vm_map_simplify_entry(map, prev_entry);
1550 return (KERN_SUCCESS);
1554 * If we can extend the object but cannot extend the
1555 * map entry, we have to create a new map entry. We
1556 * must bump the ref count on the extended object to
1557 * account for it. object may be NULL.
1559 object = prev_entry->object.vm_object;
1560 offset = prev_entry->offset +
1561 (prev_entry->end - prev_entry->start);
1562 vm_object_reference(object);
1563 if (cred != NULL && object != NULL && object->cred != NULL &&
1564 !(prev_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
1565 /* Object already accounts for this uid. */
1573 * Create a new entry
1575 new_entry = vm_map_entry_create(map);
1576 new_entry->start = start;
1577 new_entry->end = end;
1578 new_entry->cred = NULL;
1580 new_entry->eflags = protoeflags;
1581 new_entry->object.vm_object = object;
1582 new_entry->offset = offset;
1584 new_entry->inheritance = inheritance;
1585 new_entry->protection = prot;
1586 new_entry->max_protection = max;
1587 new_entry->wired_count = 0;
1588 new_entry->wiring_thread = NULL;
1589 new_entry->read_ahead = VM_FAULT_READ_AHEAD_INIT;
1590 new_entry->next_read = start;
1592 KASSERT(cred == NULL || !ENTRY_CHARGED(new_entry),
1593 ("overcommit: vm_map_insert leaks vm_map %p", new_entry));
1594 new_entry->cred = cred;
1597 * Insert the new entry into the list
1599 vm_map_entry_link(map, new_entry);
1600 if ((new_entry->eflags & MAP_ENTRY_GUARD) == 0)
1601 map->size += new_entry->end - new_entry->start;
1604 * Try to coalesce the new entry with both the previous and next
1605 * entries in the list. Previously, we only attempted to coalesce
1606 * with the previous entry when object is NULL. Here, we handle the
1607 * other cases, which are less common.
1609 vm_map_simplify_entry(map, new_entry);
1611 if ((cow & (MAP_PREFAULT | MAP_PREFAULT_PARTIAL)) != 0) {
1612 vm_map_pmap_enter(map, start, prot, object, OFF_TO_IDX(offset),
1613 end - start, cow & MAP_PREFAULT_PARTIAL);
1616 return (KERN_SUCCESS);
1622 * Find the first fit (lowest VM address) for "length" free bytes
1623 * beginning at address >= start in the given map.
1625 * In a vm_map_entry, "max_free" is the maximum amount of
1626 * contiguous free space between an entry in its subtree and a
1627 * neighbor of that entry. This allows finding a free region in
1628 * one path down the tree, so O(log n) amortized with splay
1631 * The map must be locked, and leaves it so.
1633 * Returns: starting address if sufficient space,
1634 * vm_map_max(map)-length+1 if insufficient space.
1637 vm_map_findspace(vm_map_t map, vm_offset_t start, vm_size_t length)
1639 vm_map_entry_t llist, rlist, root, y;
1640 vm_size_t left_length;
1641 vm_offset_t gap_end;
1644 * Request must fit within min/max VM address and must avoid
1647 start = MAX(start, vm_map_min(map));
1648 if (start >= vm_map_max(map) || length > vm_map_max(map) - start)
1649 return (vm_map_max(map) - length + 1);
1651 /* Empty tree means wide open address space. */
1652 if (map->root == NULL)
1656 * After splay_split, if start is within an entry, push it to the start
1657 * of the following gap. If rlist is at the end of the gap containing
1658 * start, save the end of that gap in gap_end to see if the gap is big
1659 * enough; otherwise set gap_end to start skip gap-checking and move
1660 * directly to a search of the right subtree.
1662 root = vm_map_splay_split(map, start, length, &llist, &rlist);
1663 gap_end = rlist->start;
1666 if (root->right != NULL)
1668 } else if (rlist != &map->header) {
1674 llist = root->right;
1677 vm_map_splay_merge(map, root, llist, rlist);
1678 VM_MAP_ASSERT_CONSISTENT(map);
1679 if (length <= gap_end - start)
1682 /* With max_free, can immediately tell if no solution. */
1683 if (root->right == NULL || length > root->right->max_free)
1684 return (vm_map_max(map) - length + 1);
1687 * Splay for the least large-enough gap in the right subtree.
1689 llist = rlist = &map->header;
1690 for (left_length = 0;;
1691 left_length = vm_map_entry_max_free_left(root, llist)) {
1692 if (length <= left_length)
1693 SPLAY_LEFT_STEP(root, y, rlist,
1694 length <= vm_map_entry_max_free_left(y, llist));
1696 SPLAY_RIGHT_STEP(root, y, llist,
1697 length > vm_map_entry_max_free_left(y, root));
1702 llist = root->right;
1704 if (rlist != &map->header) {
1708 vm_map_splay_merge(map, y, &map->header, rlist);
1710 vm_map_entry_max_free_left(y, root),
1711 vm_map_entry_max_free_right(y, &map->header));
1714 vm_map_splay_merge(map, root, llist, &map->header);
1715 VM_MAP_ASSERT_CONSISTENT(map);
1720 vm_map_fixed(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1721 vm_offset_t start, vm_size_t length, vm_prot_t prot,
1722 vm_prot_t max, int cow)
1727 end = start + length;
1728 KASSERT((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 ||
1730 ("vm_map_fixed: non-NULL backing object for stack"));
1732 VM_MAP_RANGE_CHECK(map, start, end);
1733 if ((cow & MAP_CHECK_EXCL) == 0)
1734 vm_map_delete(map, start, end);
1735 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) != 0) {
1736 result = vm_map_stack_locked(map, start, length, sgrowsiz,
1739 result = vm_map_insert(map, object, offset, start, end,
1746 static const int aslr_pages_rnd_64[2] = {0x1000, 0x10};
1747 static const int aslr_pages_rnd_32[2] = {0x100, 0x4};
1749 static int cluster_anon = 1;
1750 SYSCTL_INT(_vm, OID_AUTO, cluster_anon, CTLFLAG_RW,
1752 "Cluster anonymous mappings: 0 = no, 1 = yes if no hint, 2 = always");
1755 clustering_anon_allowed(vm_offset_t addr)
1758 switch (cluster_anon) {
1769 static long aslr_restarts;
1770 SYSCTL_LONG(_vm, OID_AUTO, aslr_restarts, CTLFLAG_RD,
1772 "Number of aslr failures");
1774 #define MAP_32BIT_MAX_ADDR ((vm_offset_t)1 << 31)
1777 * Searches for the specified amount of free space in the given map with the
1778 * specified alignment. Performs an address-ordered, first-fit search from
1779 * the given address "*addr", with an optional upper bound "max_addr". If the
1780 * parameter "alignment" is zero, then the alignment is computed from the
1781 * given (object, offset) pair so as to enable the greatest possible use of
1782 * superpage mappings. Returns KERN_SUCCESS and the address of the free space
1783 * in "*addr" if successful. Otherwise, returns KERN_NO_SPACE.
1785 * The map must be locked. Initially, there must be at least "length" bytes
1786 * of free space at the given address.
1789 vm_map_alignspace(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1790 vm_offset_t *addr, vm_size_t length, vm_offset_t max_addr,
1791 vm_offset_t alignment)
1793 vm_offset_t aligned_addr, free_addr;
1795 VM_MAP_ASSERT_LOCKED(map);
1797 KASSERT(free_addr == vm_map_findspace(map, free_addr, length),
1798 ("caller failed to provide space %#jx at address %p",
1799 (uintmax_t)length, (void *)free_addr));
1802 * At the start of every iteration, the free space at address
1803 * "*addr" is at least "length" bytes.
1806 pmap_align_superpage(object, offset, addr, length);
1807 else if ((*addr & (alignment - 1)) != 0) {
1808 *addr &= ~(alignment - 1);
1811 aligned_addr = *addr;
1812 if (aligned_addr == free_addr) {
1814 * Alignment did not change "*addr", so "*addr" must
1815 * still provide sufficient free space.
1817 return (KERN_SUCCESS);
1821 * Test for address wrap on "*addr". A wrapped "*addr" could
1822 * be a valid address, in which case vm_map_findspace() cannot
1823 * be relied upon to fail.
1825 if (aligned_addr < free_addr)
1826 return (KERN_NO_SPACE);
1827 *addr = vm_map_findspace(map, aligned_addr, length);
1828 if (*addr + length > vm_map_max(map) ||
1829 (max_addr != 0 && *addr + length > max_addr))
1830 return (KERN_NO_SPACE);
1832 if (free_addr == aligned_addr) {
1834 * If a successful call to vm_map_findspace() did not
1835 * change "*addr", then "*addr" must still be aligned
1836 * and provide sufficient free space.
1838 return (KERN_SUCCESS);
1844 * vm_map_find finds an unallocated region in the target address
1845 * map with the given length. The search is defined to be
1846 * first-fit from the specified address; the region found is
1847 * returned in the same parameter.
1849 * If object is non-NULL, ref count must be bumped by caller
1850 * prior to making call to account for the new entry.
1853 vm_map_find(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1854 vm_offset_t *addr, /* IN/OUT */
1855 vm_size_t length, vm_offset_t max_addr, int find_space,
1856 vm_prot_t prot, vm_prot_t max, int cow)
1858 vm_offset_t alignment, curr_min_addr, min_addr;
1859 int gap, pidx, rv, try;
1860 bool cluster, en_aslr, update_anon;
1862 KASSERT((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 ||
1864 ("vm_map_find: non-NULL backing object for stack"));
1865 MPASS((cow & MAP_REMAP) == 0 || (find_space == VMFS_NO_SPACE &&
1866 (cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0));
1867 if (find_space == VMFS_OPTIMAL_SPACE && (object == NULL ||
1868 (object->flags & OBJ_COLORED) == 0))
1869 find_space = VMFS_ANY_SPACE;
1870 if (find_space >> 8 != 0) {
1871 KASSERT((find_space & 0xff) == 0, ("bad VMFS flags"));
1872 alignment = (vm_offset_t)1 << (find_space >> 8);
1875 en_aslr = (map->flags & MAP_ASLR) != 0;
1876 update_anon = cluster = clustering_anon_allowed(*addr) &&
1877 (map->flags & MAP_IS_SUB_MAP) == 0 && max_addr == 0 &&
1878 find_space != VMFS_NO_SPACE && object == NULL &&
1879 (cow & (MAP_INHERIT_SHARE | MAP_STACK_GROWS_UP |
1880 MAP_STACK_GROWS_DOWN)) == 0 && prot != PROT_NONE;
1881 curr_min_addr = min_addr = *addr;
1882 if (en_aslr && min_addr == 0 && !cluster &&
1883 find_space != VMFS_NO_SPACE &&
1884 (map->flags & MAP_ASLR_IGNSTART) != 0)
1885 curr_min_addr = min_addr = vm_map_min(map);
1889 curr_min_addr = map->anon_loc;
1890 if (curr_min_addr == 0)
1893 if (find_space != VMFS_NO_SPACE) {
1894 KASSERT(find_space == VMFS_ANY_SPACE ||
1895 find_space == VMFS_OPTIMAL_SPACE ||
1896 find_space == VMFS_SUPER_SPACE ||
1897 alignment != 0, ("unexpected VMFS flag"));
1900 * When creating an anonymous mapping, try clustering
1901 * with an existing anonymous mapping first.
1903 * We make up to two attempts to find address space
1904 * for a given find_space value. The first attempt may
1905 * apply randomization or may cluster with an existing
1906 * anonymous mapping. If this first attempt fails,
1907 * perform a first-fit search of the available address
1910 * If all tries failed, and find_space is
1911 * VMFS_OPTIMAL_SPACE, fallback to VMFS_ANY_SPACE.
1912 * Again enable clustering and randomization.
1919 * Second try: we failed either to find a
1920 * suitable region for randomizing the
1921 * allocation, or to cluster with an existing
1922 * mapping. Retry with free run.
1924 curr_min_addr = (map->flags & MAP_ASLR_IGNSTART) != 0 ?
1925 vm_map_min(map) : min_addr;
1926 atomic_add_long(&aslr_restarts, 1);
1929 if (try == 1 && en_aslr && !cluster) {
1931 * Find space for allocation, including
1932 * gap needed for later randomization.
1934 pidx = MAXPAGESIZES > 1 && pagesizes[1] != 0 &&
1935 (find_space == VMFS_SUPER_SPACE || find_space ==
1936 VMFS_OPTIMAL_SPACE) ? 1 : 0;
1937 gap = vm_map_max(map) > MAP_32BIT_MAX_ADDR &&
1938 (max_addr == 0 || max_addr > MAP_32BIT_MAX_ADDR) ?
1939 aslr_pages_rnd_64[pidx] : aslr_pages_rnd_32[pidx];
1940 *addr = vm_map_findspace(map, curr_min_addr,
1941 length + gap * pagesizes[pidx]);
1942 if (*addr + length + gap * pagesizes[pidx] >
1945 /* And randomize the start address. */
1946 *addr += (arc4random() % gap) * pagesizes[pidx];
1947 if (max_addr != 0 && *addr + length > max_addr)
1950 *addr = vm_map_findspace(map, curr_min_addr, length);
1951 if (*addr + length > vm_map_max(map) ||
1952 (max_addr != 0 && *addr + length > max_addr)) {
1963 if (find_space != VMFS_ANY_SPACE &&
1964 (rv = vm_map_alignspace(map, object, offset, addr, length,
1965 max_addr, alignment)) != KERN_SUCCESS) {
1966 if (find_space == VMFS_OPTIMAL_SPACE) {
1967 find_space = VMFS_ANY_SPACE;
1968 curr_min_addr = min_addr;
1969 cluster = update_anon;
1975 } else if ((cow & MAP_REMAP) != 0) {
1976 if (*addr < vm_map_min(map) ||
1977 *addr + length > vm_map_max(map) ||
1978 *addr + length <= length) {
1979 rv = KERN_INVALID_ADDRESS;
1982 vm_map_delete(map, *addr, *addr + length);
1984 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) != 0) {
1985 rv = vm_map_stack_locked(map, *addr, length, sgrowsiz, prot,
1988 rv = vm_map_insert(map, object, offset, *addr, *addr + length,
1991 if (rv == KERN_SUCCESS && update_anon)
1992 map->anon_loc = *addr + length;
1999 * vm_map_find_min() is a variant of vm_map_find() that takes an
2000 * additional parameter (min_addr) and treats the given address
2001 * (*addr) differently. Specifically, it treats *addr as a hint
2002 * and not as the minimum address where the mapping is created.
2004 * This function works in two phases. First, it tries to
2005 * allocate above the hint. If that fails and the hint is
2006 * greater than min_addr, it performs a second pass, replacing
2007 * the hint with min_addr as the minimum address for the
2011 vm_map_find_min(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
2012 vm_offset_t *addr, vm_size_t length, vm_offset_t min_addr,
2013 vm_offset_t max_addr, int find_space, vm_prot_t prot, vm_prot_t max,
2021 rv = vm_map_find(map, object, offset, addr, length, max_addr,
2022 find_space, prot, max, cow);
2023 if (rv == KERN_SUCCESS || min_addr >= hint)
2025 *addr = hint = min_addr;
2030 * A map entry with any of the following flags set must not be merged with
2033 #define MAP_ENTRY_NOMERGE_MASK (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP | \
2034 MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_IS_SUB_MAP | MAP_ENTRY_VN_EXEC)
2037 vm_map_mergeable_neighbors(vm_map_entry_t prev, vm_map_entry_t entry)
2040 KASSERT((prev->eflags & MAP_ENTRY_NOMERGE_MASK) == 0 ||
2041 (entry->eflags & MAP_ENTRY_NOMERGE_MASK) == 0,
2042 ("vm_map_mergeable_neighbors: neither %p nor %p are mergeable",
2044 return (prev->end == entry->start &&
2045 prev->object.vm_object == entry->object.vm_object &&
2046 (prev->object.vm_object == NULL ||
2047 prev->offset + (prev->end - prev->start) == entry->offset) &&
2048 prev->eflags == entry->eflags &&
2049 prev->protection == entry->protection &&
2050 prev->max_protection == entry->max_protection &&
2051 prev->inheritance == entry->inheritance &&
2052 prev->wired_count == entry->wired_count &&
2053 prev->cred == entry->cred);
2057 vm_map_merged_neighbor_dispose(vm_map_t map, vm_map_entry_t entry)
2061 * If the backing object is a vnode object, vm_object_deallocate()
2062 * calls vrele(). However, vrele() does not lock the vnode because
2063 * the vnode has additional references. Thus, the map lock can be
2064 * kept without causing a lock-order reversal with the vnode lock.
2066 * Since we count the number of virtual page mappings in
2067 * object->un_pager.vnp.writemappings, the writemappings value
2068 * should not be adjusted when the entry is disposed of.
2070 if (entry->object.vm_object != NULL)
2071 vm_object_deallocate(entry->object.vm_object);
2072 if (entry->cred != NULL)
2073 crfree(entry->cred);
2074 vm_map_entry_dispose(map, entry);
2078 * vm_map_simplify_entry:
2080 * Simplify the given map entry by merging with either neighbor. This
2081 * routine also has the ability to merge with both neighbors.
2083 * The map must be locked.
2085 * This routine guarantees that the passed entry remains valid (though
2086 * possibly extended). When merging, this routine may delete one or
2090 vm_map_simplify_entry(vm_map_t map, vm_map_entry_t entry)
2092 vm_map_entry_t next, prev;
2094 if ((entry->eflags & MAP_ENTRY_NOMERGE_MASK) != 0)
2097 if (vm_map_mergeable_neighbors(prev, entry)) {
2098 vm_map_entry_unlink(map, prev, UNLINK_MERGE_NEXT);
2099 vm_map_merged_neighbor_dispose(map, prev);
2102 if (vm_map_mergeable_neighbors(entry, next)) {
2103 vm_map_entry_unlink(map, next, UNLINK_MERGE_PREV);
2104 vm_map_merged_neighbor_dispose(map, next);
2109 * vm_map_entry_back:
2111 * Allocate an object to back a map entry.
2114 vm_map_entry_back(vm_map_entry_t entry)
2118 KASSERT(entry->object.vm_object == NULL,
2119 ("map entry %p has backing object", entry));
2120 KASSERT((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0,
2121 ("map entry %p is a submap", entry));
2122 object = vm_object_allocate(OBJT_DEFAULT,
2123 atop(entry->end - entry->start));
2124 entry->object.vm_object = object;
2126 if (entry->cred != NULL) {
2127 object->cred = entry->cred;
2128 object->charge = entry->end - entry->start;
2134 * vm_map_entry_charge_object
2136 * If there is no object backing this entry, create one. Otherwise, if
2137 * the entry has cred, give it to the backing object.
2140 vm_map_entry_charge_object(vm_map_t map, vm_map_entry_t entry)
2143 VM_MAP_ASSERT_LOCKED(map);
2144 KASSERT((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0,
2145 ("map entry %p is a submap", entry));
2146 if (entry->object.vm_object == NULL && !map->system_map &&
2147 (entry->eflags & MAP_ENTRY_GUARD) == 0)
2148 vm_map_entry_back(entry);
2149 else if (entry->object.vm_object != NULL &&
2150 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) &&
2151 entry->cred != NULL) {
2152 VM_OBJECT_WLOCK(entry->object.vm_object);
2153 KASSERT(entry->object.vm_object->cred == NULL,
2154 ("OVERCOMMIT: %s: both cred e %p", __func__, entry));
2155 entry->object.vm_object->cred = entry->cred;
2156 entry->object.vm_object->charge = entry->end - entry->start;
2157 VM_OBJECT_WUNLOCK(entry->object.vm_object);
2163 * vm_map_clip_start: [ internal use only ]
2165 * Asserts that the given entry begins at or after
2166 * the specified address; if necessary,
2167 * it splits the entry into two.
2169 #define vm_map_clip_start(map, entry, startaddr) \
2171 if (startaddr > entry->start) \
2172 _vm_map_clip_start(map, entry, startaddr); \
2176 * This routine is called only when it is known that
2177 * the entry must be split.
2180 _vm_map_clip_start(vm_map_t map, vm_map_entry_t entry, vm_offset_t start)
2182 vm_map_entry_t new_entry;
2184 VM_MAP_ASSERT_LOCKED(map);
2185 KASSERT(entry->end > start && entry->start < start,
2186 ("_vm_map_clip_start: invalid clip of entry %p", entry));
2189 * Create a backing object now, if none exists, so that more individual
2190 * objects won't be created after the map entry is split.
2192 vm_map_entry_charge_object(map, entry);
2194 /* Clone the entry. */
2195 new_entry = vm_map_entry_create(map);
2196 *new_entry = *entry;
2199 * Split off the front portion. Insert the new entry BEFORE this one,
2200 * so that this entry has the specified starting address.
2202 new_entry->end = start;
2203 entry->offset += (start - entry->start);
2204 entry->start = start;
2205 if (new_entry->cred != NULL)
2206 crhold(entry->cred);
2208 vm_map_entry_link(map, new_entry);
2210 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
2211 vm_object_reference(new_entry->object.vm_object);
2212 vm_map_entry_set_vnode_text(new_entry, true);
2214 * The object->un_pager.vnp.writemappings for the
2215 * object of MAP_ENTRY_VN_WRITECNT type entry shall be
2216 * kept as is here. The virtual pages are
2217 * re-distributed among the clipped entries, so the sum is
2224 * vm_map_clip_end: [ internal use only ]
2226 * Asserts that the given entry ends at or before
2227 * the specified address; if necessary,
2228 * it splits the entry into two.
2230 #define vm_map_clip_end(map, entry, endaddr) \
2232 if ((endaddr) < (entry->end)) \
2233 _vm_map_clip_end((map), (entry), (endaddr)); \
2237 * This routine is called only when it is known that
2238 * the entry must be split.
2241 _vm_map_clip_end(vm_map_t map, vm_map_entry_t entry, vm_offset_t end)
2243 vm_map_entry_t new_entry;
2245 VM_MAP_ASSERT_LOCKED(map);
2246 KASSERT(entry->start < end && entry->end > end,
2247 ("_vm_map_clip_end: invalid clip of entry %p", entry));
2250 * Create a backing object now, if none exists, so that more individual
2251 * objects won't be created after the map entry is split.
2253 vm_map_entry_charge_object(map, entry);
2255 /* Clone the entry. */
2256 new_entry = vm_map_entry_create(map);
2257 *new_entry = *entry;
2260 * Split off the back portion. Insert the new entry AFTER this one,
2261 * so that this entry has the specified ending address.
2263 new_entry->start = entry->end = end;
2264 new_entry->offset += (end - entry->start);
2265 if (new_entry->cred != NULL)
2266 crhold(entry->cred);
2268 vm_map_entry_link(map, new_entry);
2270 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
2271 vm_object_reference(new_entry->object.vm_object);
2272 vm_map_entry_set_vnode_text(new_entry, true);
2277 * vm_map_submap: [ kernel use only ]
2279 * Mark the given range as handled by a subordinate map.
2281 * This range must have been created with vm_map_find,
2282 * and no other operations may have been performed on this
2283 * range prior to calling vm_map_submap.
2285 * Only a limited number of operations can be performed
2286 * within this rage after calling vm_map_submap:
2288 * [Don't try vm_map_copy!]
2290 * To remove a submapping, one must first remove the
2291 * range from the superior map, and then destroy the
2292 * submap (if desired). [Better yet, don't try it.]
2301 vm_map_entry_t entry;
2304 result = KERN_INVALID_ARGUMENT;
2306 vm_map_lock(submap);
2307 submap->flags |= MAP_IS_SUB_MAP;
2308 vm_map_unlock(submap);
2312 VM_MAP_RANGE_CHECK(map, start, end);
2314 if (vm_map_lookup_entry(map, start, &entry)) {
2315 vm_map_clip_start(map, entry, start);
2317 entry = entry->next;
2319 vm_map_clip_end(map, entry, end);
2321 if ((entry->start == start) && (entry->end == end) &&
2322 ((entry->eflags & MAP_ENTRY_COW) == 0) &&
2323 (entry->object.vm_object == NULL)) {
2324 entry->object.sub_map = submap;
2325 entry->eflags |= MAP_ENTRY_IS_SUB_MAP;
2326 result = KERN_SUCCESS;
2330 if (result != KERN_SUCCESS) {
2331 vm_map_lock(submap);
2332 submap->flags &= ~MAP_IS_SUB_MAP;
2333 vm_map_unlock(submap);
2339 * The maximum number of pages to map if MAP_PREFAULT_PARTIAL is specified
2341 #define MAX_INIT_PT 96
2344 * vm_map_pmap_enter:
2346 * Preload the specified map's pmap with mappings to the specified
2347 * object's memory-resident pages. No further physical pages are
2348 * allocated, and no further virtual pages are retrieved from secondary
2349 * storage. If the specified flags include MAP_PREFAULT_PARTIAL, then a
2350 * limited number of page mappings are created at the low-end of the
2351 * specified address range. (For this purpose, a superpage mapping
2352 * counts as one page mapping.) Otherwise, all resident pages within
2353 * the specified address range are mapped.
2356 vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot,
2357 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags)
2360 vm_page_t p, p_start;
2361 vm_pindex_t mask, psize, threshold, tmpidx;
2363 if ((prot & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0 || object == NULL)
2365 VM_OBJECT_RLOCK(object);
2366 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) {
2367 VM_OBJECT_RUNLOCK(object);
2368 VM_OBJECT_WLOCK(object);
2369 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) {
2370 pmap_object_init_pt(map->pmap, addr, object, pindex,
2372 VM_OBJECT_WUNLOCK(object);
2375 VM_OBJECT_LOCK_DOWNGRADE(object);
2379 if (psize + pindex > object->size) {
2380 if (object->size < pindex) {
2381 VM_OBJECT_RUNLOCK(object);
2384 psize = object->size - pindex;
2389 threshold = MAX_INIT_PT;
2391 p = vm_page_find_least(object, pindex);
2393 * Assert: the variable p is either (1) the page with the
2394 * least pindex greater than or equal to the parameter pindex
2398 p != NULL && (tmpidx = p->pindex - pindex) < psize;
2399 p = TAILQ_NEXT(p, listq)) {
2401 * don't allow an madvise to blow away our really
2402 * free pages allocating pv entries.
2404 if (((flags & MAP_PREFAULT_MADVISE) != 0 &&
2405 vm_page_count_severe()) ||
2406 ((flags & MAP_PREFAULT_PARTIAL) != 0 &&
2407 tmpidx >= threshold)) {
2411 if (p->valid == VM_PAGE_BITS_ALL) {
2412 if (p_start == NULL) {
2413 start = addr + ptoa(tmpidx);
2416 /* Jump ahead if a superpage mapping is possible. */
2417 if (p->psind > 0 && ((addr + ptoa(tmpidx)) &
2418 (pagesizes[p->psind] - 1)) == 0) {
2419 mask = atop(pagesizes[p->psind]) - 1;
2420 if (tmpidx + mask < psize &&
2421 vm_page_ps_test(p, PS_ALL_VALID, NULL)) {
2426 } else if (p_start != NULL) {
2427 pmap_enter_object(map->pmap, start, addr +
2428 ptoa(tmpidx), p_start, prot);
2432 if (p_start != NULL)
2433 pmap_enter_object(map->pmap, start, addr + ptoa(psize),
2435 VM_OBJECT_RUNLOCK(object);
2441 * Sets the protection of the specified address
2442 * region in the target map. If "set_max" is
2443 * specified, the maximum protection is to be set;
2444 * otherwise, only the current protection is affected.
2447 vm_map_protect(vm_map_t map, vm_offset_t start, vm_offset_t end,
2448 vm_prot_t new_prot, boolean_t set_max)
2450 vm_map_entry_t current, entry, in_tran;
2457 return (KERN_SUCCESS);
2464 * Ensure that we are not concurrently wiring pages. vm_map_wire() may
2465 * need to fault pages into the map and will drop the map lock while
2466 * doing so, and the VM object may end up in an inconsistent state if we
2467 * update the protection on the map entry in between faults.
2469 vm_map_wait_busy(map);
2471 VM_MAP_RANGE_CHECK(map, start, end);
2473 if (!vm_map_lookup_entry(map, start, &entry))
2474 entry = entry->next;
2477 * Make a first pass to check for protection violations.
2479 for (current = entry; current->start < end; current = current->next) {
2480 if ((current->eflags & MAP_ENTRY_GUARD) != 0)
2482 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
2484 return (KERN_INVALID_ARGUMENT);
2486 if ((new_prot & current->max_protection) != new_prot) {
2488 return (KERN_PROTECTION_FAILURE);
2490 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0)
2495 * Postpone the operation until all in transition map entries
2496 * are stabilized. In-transition entry might already have its
2497 * pages wired and wired_count incremented, but
2498 * MAP_ENTRY_USER_WIRED flag not yet set, and visible to other
2499 * threads because the map lock is dropped. In this case we
2500 * would miss our call to vm_fault_copy_entry().
2502 if (in_tran != NULL) {
2503 in_tran->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2504 vm_map_unlock_and_wait(map, 0);
2509 * Before changing the protections, try to reserve swap space for any
2510 * private (i.e., copy-on-write) mappings that are transitioning from
2511 * read-only to read/write access. If a reservation fails, break out
2512 * of this loop early and let the next loop simplify the entries, since
2513 * some may now be mergeable.
2516 vm_map_clip_start(map, entry, start);
2517 for (current = entry; current->start < end; current = current->next) {
2519 vm_map_clip_end(map, current, end);
2522 ((new_prot & ~(current->protection)) & VM_PROT_WRITE) == 0 ||
2523 ENTRY_CHARGED(current) ||
2524 (current->eflags & MAP_ENTRY_GUARD) != 0) {
2528 cred = curthread->td_ucred;
2529 obj = current->object.vm_object;
2531 if (obj == NULL || (current->eflags & MAP_ENTRY_NEEDS_COPY)) {
2532 if (!swap_reserve(current->end - current->start)) {
2533 rv = KERN_RESOURCE_SHORTAGE;
2538 current->cred = cred;
2542 VM_OBJECT_WLOCK(obj);
2543 if (obj->type != OBJT_DEFAULT && obj->type != OBJT_SWAP) {
2544 VM_OBJECT_WUNLOCK(obj);
2549 * Charge for the whole object allocation now, since
2550 * we cannot distinguish between non-charged and
2551 * charged clipped mapping of the same object later.
2553 KASSERT(obj->charge == 0,
2554 ("vm_map_protect: object %p overcharged (entry %p)",
2556 if (!swap_reserve(ptoa(obj->size))) {
2557 VM_OBJECT_WUNLOCK(obj);
2558 rv = KERN_RESOURCE_SHORTAGE;
2565 obj->charge = ptoa(obj->size);
2566 VM_OBJECT_WUNLOCK(obj);
2570 * If enough swap space was available, go back and fix up protections.
2571 * Otherwise, just simplify entries, since some may have been modified.
2572 * [Note that clipping is not necessary the second time.]
2574 for (current = entry; current->start < end;
2575 vm_map_simplify_entry(map, current), current = current->next) {
2576 if (rv != KERN_SUCCESS ||
2577 (current->eflags & MAP_ENTRY_GUARD) != 0)
2580 old_prot = current->protection;
2583 current->protection =
2584 (current->max_protection = new_prot) &
2587 current->protection = new_prot;
2590 * For user wired map entries, the normal lazy evaluation of
2591 * write access upgrades through soft page faults is
2592 * undesirable. Instead, immediately copy any pages that are
2593 * copy-on-write and enable write access in the physical map.
2595 if ((current->eflags & MAP_ENTRY_USER_WIRED) != 0 &&
2596 (current->protection & VM_PROT_WRITE) != 0 &&
2597 (old_prot & VM_PROT_WRITE) == 0)
2598 vm_fault_copy_entry(map, map, current, current, NULL);
2601 * When restricting access, update the physical map. Worry
2602 * about copy-on-write here.
2604 if ((old_prot & ~current->protection) != 0) {
2605 #define MASK(entry) (((entry)->eflags & MAP_ENTRY_COW) ? ~VM_PROT_WRITE : \
2607 pmap_protect(map->pmap, current->start,
2609 current->protection & MASK(current));
2620 * This routine traverses a processes map handling the madvise
2621 * system call. Advisories are classified as either those effecting
2622 * the vm_map_entry structure, or those effecting the underlying
2632 vm_map_entry_t current, entry;
2636 * Some madvise calls directly modify the vm_map_entry, in which case
2637 * we need to use an exclusive lock on the map and we need to perform
2638 * various clipping operations. Otherwise we only need a read-lock
2643 case MADV_SEQUENTIAL:
2660 vm_map_lock_read(map);
2667 * Locate starting entry and clip if necessary.
2669 VM_MAP_RANGE_CHECK(map, start, end);
2671 if (vm_map_lookup_entry(map, start, &entry)) {
2673 vm_map_clip_start(map, entry, start);
2675 entry = entry->next;
2680 * madvise behaviors that are implemented in the vm_map_entry.
2682 * We clip the vm_map_entry so that behavioral changes are
2683 * limited to the specified address range.
2685 for (current = entry; current->start < end;
2686 current = current->next) {
2687 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
2690 vm_map_clip_end(map, current, end);
2694 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_NORMAL);
2696 case MADV_SEQUENTIAL:
2697 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_SEQUENTIAL);
2700 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_RANDOM);
2703 current->eflags |= MAP_ENTRY_NOSYNC;
2706 current->eflags &= ~MAP_ENTRY_NOSYNC;
2709 current->eflags |= MAP_ENTRY_NOCOREDUMP;
2712 current->eflags &= ~MAP_ENTRY_NOCOREDUMP;
2717 vm_map_simplify_entry(map, current);
2721 vm_pindex_t pstart, pend;
2724 * madvise behaviors that are implemented in the underlying
2727 * Since we don't clip the vm_map_entry, we have to clip
2728 * the vm_object pindex and count.
2730 for (current = entry; current->start < end;
2731 current = current->next) {
2732 vm_offset_t useEnd, useStart;
2734 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
2737 pstart = OFF_TO_IDX(current->offset);
2738 pend = pstart + atop(current->end - current->start);
2739 useStart = current->start;
2740 useEnd = current->end;
2742 if (current->start < start) {
2743 pstart += atop(start - current->start);
2746 if (current->end > end) {
2747 pend -= atop(current->end - end);
2755 * Perform the pmap_advise() before clearing
2756 * PGA_REFERENCED in vm_page_advise(). Otherwise, a
2757 * concurrent pmap operation, such as pmap_remove(),
2758 * could clear a reference in the pmap and set
2759 * PGA_REFERENCED on the page before the pmap_advise()
2760 * had completed. Consequently, the page would appear
2761 * referenced based upon an old reference that
2762 * occurred before this pmap_advise() ran.
2764 if (behav == MADV_DONTNEED || behav == MADV_FREE)
2765 pmap_advise(map->pmap, useStart, useEnd,
2768 vm_object_madvise(current->object.vm_object, pstart,
2772 * Pre-populate paging structures in the
2773 * WILLNEED case. For wired entries, the
2774 * paging structures are already populated.
2776 if (behav == MADV_WILLNEED &&
2777 current->wired_count == 0) {
2778 vm_map_pmap_enter(map,
2780 current->protection,
2781 current->object.vm_object,
2783 ptoa(pend - pstart),
2784 MAP_PREFAULT_MADVISE
2788 vm_map_unlock_read(map);
2797 * Sets the inheritance of the specified address
2798 * range in the target map. Inheritance
2799 * affects how the map will be shared with
2800 * child maps at the time of vmspace_fork.
2803 vm_map_inherit(vm_map_t map, vm_offset_t start, vm_offset_t end,
2804 vm_inherit_t new_inheritance)
2806 vm_map_entry_t entry;
2807 vm_map_entry_t temp_entry;
2809 switch (new_inheritance) {
2810 case VM_INHERIT_NONE:
2811 case VM_INHERIT_COPY:
2812 case VM_INHERIT_SHARE:
2813 case VM_INHERIT_ZERO:
2816 return (KERN_INVALID_ARGUMENT);
2819 return (KERN_SUCCESS);
2821 VM_MAP_RANGE_CHECK(map, start, end);
2822 if (vm_map_lookup_entry(map, start, &temp_entry)) {
2824 vm_map_clip_start(map, entry, start);
2826 entry = temp_entry->next;
2827 while (entry->start < end) {
2828 vm_map_clip_end(map, entry, end);
2829 if ((entry->eflags & MAP_ENTRY_GUARD) == 0 ||
2830 new_inheritance != VM_INHERIT_ZERO)
2831 entry->inheritance = new_inheritance;
2832 vm_map_simplify_entry(map, entry);
2833 entry = entry->next;
2836 return (KERN_SUCCESS);
2842 * Implements both kernel and user unwiring.
2845 vm_map_unwire(vm_map_t map, vm_offset_t start, vm_offset_t end,
2848 vm_map_entry_t entry, first_entry, tmp_entry;
2849 vm_offset_t saved_start;
2850 unsigned int last_timestamp;
2852 bool holes_ok, need_wakeup, user_unwire;
2855 return (KERN_SUCCESS);
2856 holes_ok = (flags & VM_MAP_WIRE_HOLESOK) != 0;
2857 user_unwire = (flags & VM_MAP_WIRE_USER) != 0;
2859 VM_MAP_RANGE_CHECK(map, start, end);
2860 if (!vm_map_lookup_entry(map, start, &first_entry)) {
2862 first_entry = first_entry->next;
2865 return (KERN_INVALID_ADDRESS);
2868 last_timestamp = map->timestamp;
2869 entry = first_entry;
2870 while (entry->start < end) {
2871 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
2873 * We have not yet clipped the entry.
2875 saved_start = (start >= entry->start) ? start :
2877 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2878 if (vm_map_unlock_and_wait(map, 0)) {
2880 * Allow interruption of user unwiring?
2884 if (last_timestamp+1 != map->timestamp) {
2886 * Look again for the entry because the map was
2887 * modified while it was unlocked.
2888 * Specifically, the entry may have been
2889 * clipped, merged, or deleted.
2891 if (!vm_map_lookup_entry(map, saved_start,
2894 tmp_entry = tmp_entry->next;
2896 if (saved_start == start) {
2898 * First_entry has been deleted.
2901 return (KERN_INVALID_ADDRESS);
2904 rv = KERN_INVALID_ADDRESS;
2908 if (entry == first_entry)
2909 first_entry = tmp_entry;
2914 last_timestamp = map->timestamp;
2917 vm_map_clip_start(map, entry, start);
2918 vm_map_clip_end(map, entry, end);
2920 * Mark the entry in case the map lock is released. (See
2923 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 &&
2924 entry->wiring_thread == NULL,
2925 ("owned map entry %p", entry));
2926 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
2927 entry->wiring_thread = curthread;
2929 * Check the map for holes in the specified region.
2930 * If holes_ok, skip this check.
2933 (entry->end < end && entry->next->start > entry->end)) {
2935 rv = KERN_INVALID_ADDRESS;
2939 * If system unwiring, require that the entry is system wired.
2942 vm_map_entry_system_wired_count(entry) == 0) {
2944 rv = KERN_INVALID_ARGUMENT;
2947 entry = entry->next;
2951 need_wakeup = false;
2952 if (first_entry == NULL &&
2953 !vm_map_lookup_entry(map, start, &first_entry)) {
2954 KASSERT(holes_ok, ("vm_map_unwire: lookup failed"));
2955 first_entry = first_entry->next;
2957 for (entry = first_entry; entry->start < end; entry = entry->next) {
2959 * If holes_ok was specified, an empty
2960 * space in the unwired region could have been mapped
2961 * while the map lock was dropped for draining
2962 * MAP_ENTRY_IN_TRANSITION. Moreover, another thread
2963 * could be simultaneously wiring this new mapping
2964 * entry. Detect these cases and skip any entries
2965 * marked as in transition by us.
2967 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 ||
2968 entry->wiring_thread != curthread) {
2970 ("vm_map_unwire: !HOLESOK and new/changed entry"));
2974 if (rv == KERN_SUCCESS && (!user_unwire ||
2975 (entry->eflags & MAP_ENTRY_USER_WIRED))) {
2976 if (entry->wired_count == 1)
2977 vm_map_entry_unwire(map, entry);
2979 entry->wired_count--;
2981 entry->eflags &= ~MAP_ENTRY_USER_WIRED;
2983 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0,
2984 ("vm_map_unwire: in-transition flag missing %p", entry));
2985 KASSERT(entry->wiring_thread == curthread,
2986 ("vm_map_unwire: alien wire %p", entry));
2987 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION;
2988 entry->wiring_thread = NULL;
2989 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
2990 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
2993 vm_map_simplify_entry(map, entry);
3002 vm_map_wire_user_count_sub(u_long npages)
3005 atomic_subtract_long(&vm_user_wire_count, npages);
3009 vm_map_wire_user_count_add(u_long npages)
3013 wired = vm_user_wire_count;
3015 if (npages + wired > vm_page_max_user_wired)
3017 } while (!atomic_fcmpset_long(&vm_user_wire_count, &wired,
3024 * vm_map_wire_entry_failure:
3026 * Handle a wiring failure on the given entry.
3028 * The map should be locked.
3031 vm_map_wire_entry_failure(vm_map_t map, vm_map_entry_t entry,
3032 vm_offset_t failed_addr)
3035 VM_MAP_ASSERT_LOCKED(map);
3036 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 &&
3037 entry->wired_count == 1,
3038 ("vm_map_wire_entry_failure: entry %p isn't being wired", entry));
3039 KASSERT(failed_addr < entry->end,
3040 ("vm_map_wire_entry_failure: entry %p was fully wired", entry));
3043 * If any pages at the start of this entry were successfully wired,
3046 if (failed_addr > entry->start) {
3047 pmap_unwire(map->pmap, entry->start, failed_addr);
3048 vm_object_unwire(entry->object.vm_object, entry->offset,
3049 failed_addr - entry->start, PQ_ACTIVE);
3053 * Assign an out-of-range value to represent the failure to wire this
3056 entry->wired_count = -1;
3060 vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t end, int flags)
3065 rv = vm_map_wire_locked(map, start, end, flags);
3072 * vm_map_wire_locked:
3074 * Implements both kernel and user wiring. Returns with the map locked,
3075 * the map lock may be dropped.
3078 vm_map_wire_locked(vm_map_t map, vm_offset_t start, vm_offset_t end, int flags)
3080 vm_map_entry_t entry, first_entry, tmp_entry;
3081 vm_offset_t faddr, saved_end, saved_start;
3083 u_int last_timestamp;
3085 bool holes_ok, need_wakeup, user_wire;
3088 VM_MAP_ASSERT_LOCKED(map);
3091 return (KERN_SUCCESS);
3093 if (flags & VM_MAP_WIRE_WRITE)
3094 prot |= VM_PROT_WRITE;
3095 holes_ok = (flags & VM_MAP_WIRE_HOLESOK) != 0;
3096 user_wire = (flags & VM_MAP_WIRE_USER) != 0;
3097 VM_MAP_RANGE_CHECK(map, start, end);
3098 if (!vm_map_lookup_entry(map, start, &first_entry)) {
3100 first_entry = first_entry->next;
3102 return (KERN_INVALID_ADDRESS);
3104 last_timestamp = map->timestamp;
3105 entry = first_entry;
3106 while (entry->start < end) {
3107 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
3109 * We have not yet clipped the entry.
3111 saved_start = (start >= entry->start) ? start :
3113 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
3114 if (vm_map_unlock_and_wait(map, 0)) {
3116 * Allow interruption of user wiring?
3120 if (last_timestamp + 1 != map->timestamp) {
3122 * Look again for the entry because the map was
3123 * modified while it was unlocked.
3124 * Specifically, the entry may have been
3125 * clipped, merged, or deleted.
3127 if (!vm_map_lookup_entry(map, saved_start,
3130 tmp_entry = tmp_entry->next;
3132 if (saved_start == start) {
3134 * first_entry has been deleted.
3136 return (KERN_INVALID_ADDRESS);
3139 rv = KERN_INVALID_ADDRESS;
3143 if (entry == first_entry)
3144 first_entry = tmp_entry;
3149 last_timestamp = map->timestamp;
3152 vm_map_clip_start(map, entry, start);
3153 vm_map_clip_end(map, entry, end);
3155 * Mark the entry in case the map lock is released. (See
3158 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 &&
3159 entry->wiring_thread == NULL,
3160 ("owned map entry %p", entry));
3161 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
3162 entry->wiring_thread = curthread;
3163 if ((entry->protection & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0
3164 || (entry->protection & prot) != prot) {
3165 entry->eflags |= MAP_ENTRY_WIRE_SKIPPED;
3168 rv = KERN_INVALID_ADDRESS;
3171 } else if (entry->wired_count == 0) {
3172 entry->wired_count++;
3174 npages = atop(entry->end - entry->start);
3175 if (user_wire && !vm_map_wire_user_count_add(npages)) {
3176 vm_map_wire_entry_failure(map, entry,
3179 rv = KERN_RESOURCE_SHORTAGE;
3184 * Release the map lock, relying on the in-transition
3185 * mark. Mark the map busy for fork.
3187 saved_start = entry->start;
3188 saved_end = entry->end;
3192 faddr = saved_start;
3195 * Simulate a fault to get the page and enter
3196 * it into the physical map.
3198 if ((rv = vm_fault(map, faddr, VM_PROT_NONE,
3199 VM_FAULT_WIRE)) != KERN_SUCCESS)
3201 } while ((faddr += PAGE_SIZE) < saved_end);
3204 if (last_timestamp + 1 != map->timestamp) {
3206 * Look again for the entry because the map was
3207 * modified while it was unlocked. The entry
3208 * may have been clipped, but NOT merged or
3211 if (!vm_map_lookup_entry(map, saved_start,
3214 ("vm_map_wire: lookup failed"));
3215 if (entry == first_entry)
3216 first_entry = tmp_entry;
3220 while (entry->end < saved_end) {
3222 * In case of failure, handle entries
3223 * that were not fully wired here;
3224 * fully wired entries are handled
3227 if (rv != KERN_SUCCESS &&
3229 vm_map_wire_entry_failure(map,
3231 entry = entry->next;
3234 last_timestamp = map->timestamp;
3235 if (rv != KERN_SUCCESS) {
3236 vm_map_wire_entry_failure(map, entry, faddr);
3238 vm_map_wire_user_count_sub(npages);
3242 } else if (!user_wire ||
3243 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
3244 entry->wired_count++;
3247 * Check the map for holes in the specified region.
3248 * If holes_ok was specified, skip this check.
3251 entry->end < end && entry->next->start > entry->end) {
3253 rv = KERN_INVALID_ADDRESS;
3256 entry = entry->next;
3260 need_wakeup = false;
3261 if (first_entry == NULL &&
3262 !vm_map_lookup_entry(map, start, &first_entry)) {
3263 KASSERT(holes_ok, ("vm_map_wire: lookup failed"));
3264 first_entry = first_entry->next;
3266 for (entry = first_entry; entry->start < end; entry = entry->next) {
3268 * If holes_ok was specified, an empty
3269 * space in the unwired region could have been mapped
3270 * while the map lock was dropped for faulting in the
3271 * pages or draining MAP_ENTRY_IN_TRANSITION.
3272 * Moreover, another thread could be simultaneously
3273 * wiring this new mapping entry. Detect these cases
3274 * and skip any entries marked as in transition not by us.
3276 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 ||
3277 entry->wiring_thread != curthread) {
3279 ("vm_map_wire: !HOLESOK and new/changed entry"));
3283 if ((entry->eflags & MAP_ENTRY_WIRE_SKIPPED) != 0)
3284 goto next_entry_done;
3286 if (rv == KERN_SUCCESS) {
3288 entry->eflags |= MAP_ENTRY_USER_WIRED;
3289 } else if (entry->wired_count == -1) {
3291 * Wiring failed on this entry. Thus, unwiring is
3294 entry->wired_count = 0;
3295 } else if (!user_wire ||
3296 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
3298 * Undo the wiring. Wiring succeeded on this entry
3299 * but failed on a later entry.
3301 if (entry->wired_count == 1) {
3302 vm_map_entry_unwire(map, entry);
3304 vm_map_wire_user_count_sub(
3305 atop(entry->end - entry->start));
3307 entry->wired_count--;
3310 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0,
3311 ("vm_map_wire: in-transition flag missing %p", entry));
3312 KASSERT(entry->wiring_thread == curthread,
3313 ("vm_map_wire: alien wire %p", entry));
3314 entry->eflags &= ~(MAP_ENTRY_IN_TRANSITION |
3315 MAP_ENTRY_WIRE_SKIPPED);
3316 entry->wiring_thread = NULL;
3317 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
3318 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
3321 vm_map_simplify_entry(map, entry);
3331 * Push any dirty cached pages in the address range to their pager.
3332 * If syncio is TRUE, dirty pages are written synchronously.
3333 * If invalidate is TRUE, any cached pages are freed as well.
3335 * If the size of the region from start to end is zero, we are
3336 * supposed to flush all modified pages within the region containing
3337 * start. Unfortunately, a region can be split or coalesced with
3338 * neighboring regions, making it difficult to determine what the
3339 * original region was. Therefore, we approximate this requirement by
3340 * flushing the current region containing start.
3342 * Returns an error if any part of the specified range is not mapped.
3350 boolean_t invalidate)
3352 vm_map_entry_t current;
3353 vm_map_entry_t entry;
3356 vm_ooffset_t offset;
3357 unsigned int last_timestamp;
3360 vm_map_lock_read(map);
3361 VM_MAP_RANGE_CHECK(map, start, end);
3362 if (!vm_map_lookup_entry(map, start, &entry)) {
3363 vm_map_unlock_read(map);
3364 return (KERN_INVALID_ADDRESS);
3365 } else if (start == end) {
3366 start = entry->start;
3370 * Make a first pass to check for user-wired memory and holes.
3372 for (current = entry; current->start < end; current = current->next) {
3373 if (invalidate && (current->eflags & MAP_ENTRY_USER_WIRED)) {
3374 vm_map_unlock_read(map);
3375 return (KERN_INVALID_ARGUMENT);
3377 if (end > current->end &&
3378 current->end != current->next->start) {
3379 vm_map_unlock_read(map);
3380 return (KERN_INVALID_ADDRESS);
3385 pmap_remove(map->pmap, start, end);
3389 * Make a second pass, cleaning/uncaching pages from the indicated
3392 for (current = entry; current->start < end;) {
3393 offset = current->offset + (start - current->start);
3394 size = (end <= current->end ? end : current->end) - start;
3395 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
3397 vm_map_entry_t tentry;
3400 smap = current->object.sub_map;
3401 vm_map_lock_read(smap);
3402 (void) vm_map_lookup_entry(smap, offset, &tentry);
3403 tsize = tentry->end - offset;
3406 object = tentry->object.vm_object;
3407 offset = tentry->offset + (offset - tentry->start);
3408 vm_map_unlock_read(smap);
3410 object = current->object.vm_object;
3412 vm_object_reference(object);
3413 last_timestamp = map->timestamp;
3414 vm_map_unlock_read(map);
3415 if (!vm_object_sync(object, offset, size, syncio, invalidate))
3418 vm_object_deallocate(object);
3419 vm_map_lock_read(map);
3420 if (last_timestamp == map->timestamp ||
3421 !vm_map_lookup_entry(map, start, ¤t))
3422 current = current->next;
3425 vm_map_unlock_read(map);
3426 return (failed ? KERN_FAILURE : KERN_SUCCESS);
3430 * vm_map_entry_unwire: [ internal use only ]
3432 * Make the region specified by this entry pageable.
3434 * The map in question should be locked.
3435 * [This is the reason for this routine's existence.]
3438 vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry)
3442 VM_MAP_ASSERT_LOCKED(map);
3443 KASSERT(entry->wired_count > 0,
3444 ("vm_map_entry_unwire: entry %p isn't wired", entry));
3446 size = entry->end - entry->start;
3447 if ((entry->eflags & MAP_ENTRY_USER_WIRED) != 0)
3448 vm_map_wire_user_count_sub(atop(size));
3449 pmap_unwire(map->pmap, entry->start, entry->end);
3450 vm_object_unwire(entry->object.vm_object, entry->offset, size,
3452 entry->wired_count = 0;
3456 vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map)
3459 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0)
3460 vm_object_deallocate(entry->object.vm_object);
3461 uma_zfree(system_map ? kmapentzone : mapentzone, entry);
3465 * vm_map_entry_delete: [ internal use only ]
3467 * Deallocate the given entry from the target map.
3470 vm_map_entry_delete(vm_map_t map, vm_map_entry_t entry)
3473 vm_pindex_t offidxstart, offidxend, count, size1;
3476 vm_map_entry_unlink(map, entry, UNLINK_MERGE_NONE);
3477 object = entry->object.vm_object;
3479 if ((entry->eflags & MAP_ENTRY_GUARD) != 0) {
3480 MPASS(entry->cred == NULL);
3481 MPASS((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0);
3482 MPASS(object == NULL);
3483 vm_map_entry_deallocate(entry, map->system_map);
3487 size = entry->end - entry->start;
3490 if (entry->cred != NULL) {
3491 swap_release_by_cred(size, entry->cred);
3492 crfree(entry->cred);
3495 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0 &&
3497 KASSERT(entry->cred == NULL || object->cred == NULL ||
3498 (entry->eflags & MAP_ENTRY_NEEDS_COPY),
3499 ("OVERCOMMIT vm_map_entry_delete: both cred %p", entry));
3501 offidxstart = OFF_TO_IDX(entry->offset);
3502 offidxend = offidxstart + count;
3503 VM_OBJECT_WLOCK(object);
3504 if (object->ref_count != 1 && ((object->flags & (OBJ_NOSPLIT |
3505 OBJ_ONEMAPPING)) == OBJ_ONEMAPPING ||
3506 object == kernel_object)) {
3507 vm_object_collapse(object);
3510 * The option OBJPR_NOTMAPPED can be passed here
3511 * because vm_map_delete() already performed
3512 * pmap_remove() on the only mapping to this range
3515 vm_object_page_remove(object, offidxstart, offidxend,
3517 if (object->type == OBJT_SWAP)
3518 swap_pager_freespace(object, offidxstart,
3520 if (offidxend >= object->size &&
3521 offidxstart < object->size) {
3522 size1 = object->size;
3523 object->size = offidxstart;
3524 if (object->cred != NULL) {
3525 size1 -= object->size;
3526 KASSERT(object->charge >= ptoa(size1),
3527 ("object %p charge < 0", object));
3528 swap_release_by_cred(ptoa(size1),
3530 object->charge -= ptoa(size1);
3534 VM_OBJECT_WUNLOCK(object);
3536 entry->object.vm_object = NULL;
3537 if (map->system_map)
3538 vm_map_entry_deallocate(entry, TRUE);
3540 entry->next = curthread->td_map_def_user;
3541 curthread->td_map_def_user = entry;
3546 * vm_map_delete: [ internal use only ]
3548 * Deallocates the given address range from the target
3552 vm_map_delete(vm_map_t map, vm_offset_t start, vm_offset_t end)
3554 vm_map_entry_t entry;
3555 vm_map_entry_t first_entry;
3557 VM_MAP_ASSERT_LOCKED(map);
3559 return (KERN_SUCCESS);
3562 * Find the start of the region, and clip it
3564 if (!vm_map_lookup_entry(map, start, &first_entry))
3565 entry = first_entry->next;
3567 entry = first_entry;
3568 vm_map_clip_start(map, entry, start);
3572 * Step through all entries in this region
3574 while (entry->start < end) {
3575 vm_map_entry_t next;
3578 * Wait for wiring or unwiring of an entry to complete.
3579 * Also wait for any system wirings to disappear on
3582 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 ||
3583 (vm_map_pmap(map) != kernel_pmap &&
3584 vm_map_entry_system_wired_count(entry) != 0)) {
3585 unsigned int last_timestamp;
3586 vm_offset_t saved_start;
3587 vm_map_entry_t tmp_entry;
3589 saved_start = entry->start;
3590 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
3591 last_timestamp = map->timestamp;
3592 (void) vm_map_unlock_and_wait(map, 0);
3594 if (last_timestamp + 1 != map->timestamp) {
3596 * Look again for the entry because the map was
3597 * modified while it was unlocked.
3598 * Specifically, the entry may have been
3599 * clipped, merged, or deleted.
3601 if (!vm_map_lookup_entry(map, saved_start,
3603 entry = tmp_entry->next;
3606 vm_map_clip_start(map, entry,
3612 vm_map_clip_end(map, entry, end);
3617 * Unwire before removing addresses from the pmap; otherwise,
3618 * unwiring will put the entries back in the pmap.
3620 if (entry->wired_count != 0)
3621 vm_map_entry_unwire(map, entry);
3624 * Remove mappings for the pages, but only if the
3625 * mappings could exist. For instance, it does not
3626 * make sense to call pmap_remove() for guard entries.
3628 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0 ||
3629 entry->object.vm_object != NULL)
3630 pmap_remove(map->pmap, entry->start, entry->end);
3632 if (entry->end == map->anon_loc)
3633 map->anon_loc = entry->start;
3636 * Delete the entry only after removing all pmap
3637 * entries pointing to its pages. (Otherwise, its
3638 * page frames may be reallocated, and any modify bits
3639 * will be set in the wrong object!)
3641 vm_map_entry_delete(map, entry);
3644 return (KERN_SUCCESS);
3650 * Remove the given address range from the target map.
3651 * This is the exported form of vm_map_delete.
3654 vm_map_remove(vm_map_t map, vm_offset_t start, vm_offset_t end)
3659 VM_MAP_RANGE_CHECK(map, start, end);
3660 result = vm_map_delete(map, start, end);
3666 * vm_map_check_protection:
3668 * Assert that the target map allows the specified privilege on the
3669 * entire address region given. The entire region must be allocated.
3671 * WARNING! This code does not and should not check whether the
3672 * contents of the region is accessible. For example a smaller file
3673 * might be mapped into a larger address space.
3675 * NOTE! This code is also called by munmap().
3677 * The map must be locked. A read lock is sufficient.
3680 vm_map_check_protection(vm_map_t map, vm_offset_t start, vm_offset_t end,
3681 vm_prot_t protection)
3683 vm_map_entry_t entry;
3684 vm_map_entry_t tmp_entry;
3686 if (!vm_map_lookup_entry(map, start, &tmp_entry))
3690 while (start < end) {
3694 if (start < entry->start)
3697 * Check protection associated with entry.
3699 if ((entry->protection & protection) != protection)
3701 /* go to next entry */
3703 entry = entry->next;
3709 * vm_map_copy_entry:
3711 * Copies the contents of the source entry to the destination
3712 * entry. The entries *must* be aligned properly.
3718 vm_map_entry_t src_entry,
3719 vm_map_entry_t dst_entry,
3720 vm_ooffset_t *fork_charge)
3722 vm_object_t src_object;
3723 vm_map_entry_t fake_entry;
3728 VM_MAP_ASSERT_LOCKED(dst_map);
3730 if ((dst_entry->eflags|src_entry->eflags) & MAP_ENTRY_IS_SUB_MAP)
3733 if (src_entry->wired_count == 0 ||
3734 (src_entry->protection & VM_PROT_WRITE) == 0) {
3736 * If the source entry is marked needs_copy, it is already
3739 if ((src_entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0 &&
3740 (src_entry->protection & VM_PROT_WRITE) != 0) {
3741 pmap_protect(src_map->pmap,
3744 src_entry->protection & ~VM_PROT_WRITE);
3748 * Make a copy of the object.
3750 size = src_entry->end - src_entry->start;
3751 if ((src_object = src_entry->object.vm_object) != NULL) {
3752 VM_OBJECT_WLOCK(src_object);
3753 charged = ENTRY_CHARGED(src_entry);
3754 if (src_object->handle == NULL &&
3755 (src_object->type == OBJT_DEFAULT ||
3756 src_object->type == OBJT_SWAP)) {
3757 vm_object_collapse(src_object);
3758 if ((src_object->flags & (OBJ_NOSPLIT |
3759 OBJ_ONEMAPPING)) == OBJ_ONEMAPPING) {
3760 vm_object_split(src_entry);
3762 src_entry->object.vm_object;
3765 vm_object_reference_locked(src_object);
3766 vm_object_clear_flag(src_object, OBJ_ONEMAPPING);
3767 if (src_entry->cred != NULL &&
3768 !(src_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
3769 KASSERT(src_object->cred == NULL,
3770 ("OVERCOMMIT: vm_map_copy_entry: cred %p",
3772 src_object->cred = src_entry->cred;
3773 src_object->charge = size;
3775 VM_OBJECT_WUNLOCK(src_object);
3776 dst_entry->object.vm_object = src_object;
3778 cred = curthread->td_ucred;
3780 dst_entry->cred = cred;
3781 *fork_charge += size;
3782 if (!(src_entry->eflags &
3783 MAP_ENTRY_NEEDS_COPY)) {
3785 src_entry->cred = cred;
3786 *fork_charge += size;
3789 src_entry->eflags |= MAP_ENTRY_COW |
3790 MAP_ENTRY_NEEDS_COPY;
3791 dst_entry->eflags |= MAP_ENTRY_COW |
3792 MAP_ENTRY_NEEDS_COPY;
3793 dst_entry->offset = src_entry->offset;
3794 if (src_entry->eflags & MAP_ENTRY_VN_WRITECNT) {
3796 * MAP_ENTRY_VN_WRITECNT cannot
3797 * indicate write reference from
3798 * src_entry, since the entry is
3799 * marked as needs copy. Allocate a
3800 * fake entry that is used to
3801 * decrement object->un_pager.vnp.writecount
3802 * at the appropriate time. Attach
3803 * fake_entry to the deferred list.
3805 fake_entry = vm_map_entry_create(dst_map);
3806 fake_entry->eflags = MAP_ENTRY_VN_WRITECNT;
3807 src_entry->eflags &= ~MAP_ENTRY_VN_WRITECNT;
3808 vm_object_reference(src_object);
3809 fake_entry->object.vm_object = src_object;
3810 fake_entry->start = src_entry->start;
3811 fake_entry->end = src_entry->end;
3812 fake_entry->next = curthread->td_map_def_user;
3813 curthread->td_map_def_user = fake_entry;
3816 pmap_copy(dst_map->pmap, src_map->pmap,
3817 dst_entry->start, dst_entry->end - dst_entry->start,
3820 dst_entry->object.vm_object = NULL;
3821 dst_entry->offset = 0;
3822 if (src_entry->cred != NULL) {
3823 dst_entry->cred = curthread->td_ucred;
3824 crhold(dst_entry->cred);
3825 *fork_charge += size;
3830 * We don't want to make writeable wired pages copy-on-write.
3831 * Immediately copy these pages into the new map by simulating
3832 * page faults. The new pages are pageable.
3834 vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry,
3840 * vmspace_map_entry_forked:
3841 * Update the newly-forked vmspace each time a map entry is inherited
3842 * or copied. The values for vm_dsize and vm_tsize are approximate
3843 * (and mostly-obsolete ideas in the face of mmap(2) et al.)
3846 vmspace_map_entry_forked(const struct vmspace *vm1, struct vmspace *vm2,
3847 vm_map_entry_t entry)
3849 vm_size_t entrysize;
3852 if ((entry->eflags & MAP_ENTRY_GUARD) != 0)
3854 entrysize = entry->end - entry->start;
3855 vm2->vm_map.size += entrysize;
3856 if (entry->eflags & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP)) {
3857 vm2->vm_ssize += btoc(entrysize);
3858 } else if (entry->start >= (vm_offset_t)vm1->vm_daddr &&
3859 entry->start < (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)) {
3860 newend = MIN(entry->end,
3861 (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize));
3862 vm2->vm_dsize += btoc(newend - entry->start);
3863 } else if (entry->start >= (vm_offset_t)vm1->vm_taddr &&
3864 entry->start < (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)) {
3865 newend = MIN(entry->end,
3866 (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize));
3867 vm2->vm_tsize += btoc(newend - entry->start);
3873 * Create a new process vmspace structure and vm_map
3874 * based on those of an existing process. The new map
3875 * is based on the old map, according to the inheritance
3876 * values on the regions in that map.
3878 * XXX It might be worth coalescing the entries added to the new vmspace.
3880 * The source map must not be locked.
3883 vmspace_fork(struct vmspace *vm1, vm_ooffset_t *fork_charge)
3885 struct vmspace *vm2;
3886 vm_map_t new_map, old_map;
3887 vm_map_entry_t new_entry, old_entry;
3892 old_map = &vm1->vm_map;
3893 /* Copy immutable fields of vm1 to vm2. */
3894 vm2 = vmspace_alloc(vm_map_min(old_map), vm_map_max(old_map),
3899 vm2->vm_taddr = vm1->vm_taddr;
3900 vm2->vm_daddr = vm1->vm_daddr;
3901 vm2->vm_maxsaddr = vm1->vm_maxsaddr;
3902 vm_map_lock(old_map);
3904 vm_map_wait_busy(old_map);
3905 new_map = &vm2->vm_map;
3906 locked = vm_map_trylock(new_map); /* trylock to silence WITNESS */
3907 KASSERT(locked, ("vmspace_fork: lock failed"));
3909 error = pmap_vmspace_copy(new_map->pmap, old_map->pmap);
3911 sx_xunlock(&old_map->lock);
3912 sx_xunlock(&new_map->lock);
3913 vm_map_process_deferred();
3918 new_map->anon_loc = old_map->anon_loc;
3920 old_entry = old_map->header.next;
3922 while (old_entry != &old_map->header) {
3923 if (old_entry->eflags & MAP_ENTRY_IS_SUB_MAP)
3924 panic("vm_map_fork: encountered a submap");
3926 inh = old_entry->inheritance;
3927 if ((old_entry->eflags & MAP_ENTRY_GUARD) != 0 &&
3928 inh != VM_INHERIT_NONE)
3929 inh = VM_INHERIT_COPY;
3932 case VM_INHERIT_NONE:
3935 case VM_INHERIT_SHARE:
3937 * Clone the entry, creating the shared object if necessary.
3939 object = old_entry->object.vm_object;
3940 if (object == NULL) {
3941 vm_map_entry_back(old_entry);
3942 object = old_entry->object.vm_object;
3946 * Add the reference before calling vm_object_shadow
3947 * to insure that a shadow object is created.
3949 vm_object_reference(object);
3950 if (old_entry->eflags & MAP_ENTRY_NEEDS_COPY) {
3951 vm_object_shadow(&old_entry->object.vm_object,
3953 old_entry->end - old_entry->start);
3954 old_entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
3955 /* Transfer the second reference too. */
3956 vm_object_reference(
3957 old_entry->object.vm_object);
3960 * As in vm_map_simplify_entry(), the
3961 * vnode lock will not be acquired in
3962 * this call to vm_object_deallocate().
3964 vm_object_deallocate(object);
3965 object = old_entry->object.vm_object;
3967 VM_OBJECT_WLOCK(object);
3968 vm_object_clear_flag(object, OBJ_ONEMAPPING);
3969 if (old_entry->cred != NULL) {
3970 KASSERT(object->cred == NULL, ("vmspace_fork both cred"));
3971 object->cred = old_entry->cred;
3972 object->charge = old_entry->end - old_entry->start;
3973 old_entry->cred = NULL;
3977 * Assert the correct state of the vnode
3978 * v_writecount while the object is locked, to
3979 * not relock it later for the assertion
3982 if (old_entry->eflags & MAP_ENTRY_VN_WRITECNT &&
3983 object->type == OBJT_VNODE) {
3984 KASSERT(((struct vnode *)object->handle)->
3986 ("vmspace_fork: v_writecount %p", object));
3987 KASSERT(object->un_pager.vnp.writemappings > 0,
3988 ("vmspace_fork: vnp.writecount %p",
3991 VM_OBJECT_WUNLOCK(object);
3994 * Clone the entry, referencing the shared object.
3996 new_entry = vm_map_entry_create(new_map);
3997 *new_entry = *old_entry;
3998 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED |
3999 MAP_ENTRY_IN_TRANSITION);
4000 new_entry->wiring_thread = NULL;
4001 new_entry->wired_count = 0;
4002 if (new_entry->eflags & MAP_ENTRY_VN_WRITECNT) {
4003 vnode_pager_update_writecount(object,
4004 new_entry->start, new_entry->end);
4006 vm_map_entry_set_vnode_text(new_entry, true);
4009 * Insert the entry into the new map -- we know we're
4010 * inserting at the end of the new map.
4012 vm_map_entry_link(new_map, new_entry);
4013 vmspace_map_entry_forked(vm1, vm2, new_entry);
4016 * Update the physical map
4018 pmap_copy(new_map->pmap, old_map->pmap,
4020 (old_entry->end - old_entry->start),
4024 case VM_INHERIT_COPY:
4026 * Clone the entry and link into the map.
4028 new_entry = vm_map_entry_create(new_map);
4029 *new_entry = *old_entry;
4031 * Copied entry is COW over the old object.
4033 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED |
4034 MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_VN_WRITECNT);
4035 new_entry->wiring_thread = NULL;
4036 new_entry->wired_count = 0;
4037 new_entry->object.vm_object = NULL;
4038 new_entry->cred = NULL;
4039 vm_map_entry_link(new_map, new_entry);
4040 vmspace_map_entry_forked(vm1, vm2, new_entry);
4041 vm_map_copy_entry(old_map, new_map, old_entry,
4042 new_entry, fork_charge);
4043 vm_map_entry_set_vnode_text(new_entry, true);
4046 case VM_INHERIT_ZERO:
4048 * Create a new anonymous mapping entry modelled from
4051 new_entry = vm_map_entry_create(new_map);
4052 memset(new_entry, 0, sizeof(*new_entry));
4054 new_entry->start = old_entry->start;
4055 new_entry->end = old_entry->end;
4056 new_entry->eflags = old_entry->eflags &
4057 ~(MAP_ENTRY_USER_WIRED | MAP_ENTRY_IN_TRANSITION |
4058 MAP_ENTRY_VN_WRITECNT | MAP_ENTRY_VN_EXEC);
4059 new_entry->protection = old_entry->protection;
4060 new_entry->max_protection = old_entry->max_protection;
4061 new_entry->inheritance = VM_INHERIT_ZERO;
4063 vm_map_entry_link(new_map, new_entry);
4064 vmspace_map_entry_forked(vm1, vm2, new_entry);
4066 new_entry->cred = curthread->td_ucred;
4067 crhold(new_entry->cred);
4068 *fork_charge += (new_entry->end - new_entry->start);
4072 old_entry = old_entry->next;
4075 * Use inlined vm_map_unlock() to postpone handling the deferred
4076 * map entries, which cannot be done until both old_map and
4077 * new_map locks are released.
4079 sx_xunlock(&old_map->lock);
4080 sx_xunlock(&new_map->lock);
4081 vm_map_process_deferred();
4087 * Create a process's stack for exec_new_vmspace(). This function is never
4088 * asked to wire the newly created stack.
4091 vm_map_stack(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
4092 vm_prot_t prot, vm_prot_t max, int cow)
4094 vm_size_t growsize, init_ssize;
4098 MPASS((map->flags & MAP_WIREFUTURE) == 0);
4099 growsize = sgrowsiz;
4100 init_ssize = (max_ssize < growsize) ? max_ssize : growsize;
4102 vmemlim = lim_cur(curthread, RLIMIT_VMEM);
4103 /* If we would blow our VMEM resource limit, no go */
4104 if (map->size + init_ssize > vmemlim) {
4108 rv = vm_map_stack_locked(map, addrbos, max_ssize, growsize, prot,
4115 static int stack_guard_page = 1;
4116 SYSCTL_INT(_security_bsd, OID_AUTO, stack_guard_page, CTLFLAG_RWTUN,
4117 &stack_guard_page, 0,
4118 "Specifies the number of guard pages for a stack that grows");
4121 vm_map_stack_locked(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
4122 vm_size_t growsize, vm_prot_t prot, vm_prot_t max, int cow)
4124 vm_map_entry_t new_entry, prev_entry;
4125 vm_offset_t bot, gap_bot, gap_top, top;
4126 vm_size_t init_ssize, sgp;
4130 * The stack orientation is piggybacked with the cow argument.
4131 * Extract it into orient and mask the cow argument so that we
4132 * don't pass it around further.
4134 orient = cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP);
4135 KASSERT(orient != 0, ("No stack grow direction"));
4136 KASSERT(orient != (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP),
4139 if (addrbos < vm_map_min(map) ||
4140 addrbos + max_ssize > vm_map_max(map) ||
4141 addrbos + max_ssize <= addrbos)
4142 return (KERN_INVALID_ADDRESS);
4143 sgp = (vm_size_t)stack_guard_page * PAGE_SIZE;
4144 if (sgp >= max_ssize)
4145 return (KERN_INVALID_ARGUMENT);
4147 init_ssize = growsize;
4148 if (max_ssize < init_ssize + sgp)
4149 init_ssize = max_ssize - sgp;
4151 /* If addr is already mapped, no go */
4152 if (vm_map_lookup_entry(map, addrbos, &prev_entry))
4153 return (KERN_NO_SPACE);
4156 * If we can't accommodate max_ssize in the current mapping, no go.
4158 if (prev_entry->next->start < addrbos + max_ssize)
4159 return (KERN_NO_SPACE);
4162 * We initially map a stack of only init_ssize. We will grow as
4163 * needed later. Depending on the orientation of the stack (i.e.
4164 * the grow direction) we either map at the top of the range, the
4165 * bottom of the range or in the middle.
4167 * Note: we would normally expect prot and max to be VM_PROT_ALL,
4168 * and cow to be 0. Possibly we should eliminate these as input
4169 * parameters, and just pass these values here in the insert call.
4171 if (orient == MAP_STACK_GROWS_DOWN) {
4172 bot = addrbos + max_ssize - init_ssize;
4173 top = bot + init_ssize;
4176 } else /* if (orient == MAP_STACK_GROWS_UP) */ {
4178 top = bot + init_ssize;
4180 gap_top = addrbos + max_ssize;
4182 rv = vm_map_insert(map, NULL, 0, bot, top, prot, max, cow);
4183 if (rv != KERN_SUCCESS)
4185 new_entry = prev_entry->next;
4186 KASSERT(new_entry->end == top || new_entry->start == bot,
4187 ("Bad entry start/end for new stack entry"));
4188 KASSERT((orient & MAP_STACK_GROWS_DOWN) == 0 ||
4189 (new_entry->eflags & MAP_ENTRY_GROWS_DOWN) != 0,
4190 ("new entry lacks MAP_ENTRY_GROWS_DOWN"));
4191 KASSERT((orient & MAP_STACK_GROWS_UP) == 0 ||
4192 (new_entry->eflags & MAP_ENTRY_GROWS_UP) != 0,
4193 ("new entry lacks MAP_ENTRY_GROWS_UP"));
4194 rv = vm_map_insert(map, NULL, 0, gap_bot, gap_top, VM_PROT_NONE,
4195 VM_PROT_NONE, MAP_CREATE_GUARD | (orient == MAP_STACK_GROWS_DOWN ?
4196 MAP_CREATE_STACK_GAP_DN : MAP_CREATE_STACK_GAP_UP));
4197 if (rv != KERN_SUCCESS)
4198 (void)vm_map_delete(map, bot, top);
4203 * Attempts to grow a vm stack entry. Returns KERN_SUCCESS if we
4204 * successfully grow the stack.
4207 vm_map_growstack(vm_map_t map, vm_offset_t addr, vm_map_entry_t gap_entry)
4209 vm_map_entry_t stack_entry;
4213 vm_offset_t gap_end, gap_start, grow_start;
4214 vm_size_t grow_amount, guard, max_grow;
4215 rlim_t lmemlim, stacklim, vmemlim;
4217 bool gap_deleted, grow_down, is_procstack;
4229 * Disallow stack growth when the access is performed by a
4230 * debugger or AIO daemon. The reason is that the wrong
4231 * resource limits are applied.
4233 if (map != &p->p_vmspace->vm_map || p->p_textvp == NULL)
4234 return (KERN_FAILURE);
4236 MPASS(!map->system_map);
4238 guard = stack_guard_page * PAGE_SIZE;
4239 lmemlim = lim_cur(curthread, RLIMIT_MEMLOCK);
4240 stacklim = lim_cur(curthread, RLIMIT_STACK);
4241 vmemlim = lim_cur(curthread, RLIMIT_VMEM);
4243 /* If addr is not in a hole for a stack grow area, no need to grow. */
4244 if (gap_entry == NULL && !vm_map_lookup_entry(map, addr, &gap_entry))
4245 return (KERN_FAILURE);
4246 if ((gap_entry->eflags & MAP_ENTRY_GUARD) == 0)
4247 return (KERN_SUCCESS);
4248 if ((gap_entry->eflags & MAP_ENTRY_STACK_GAP_DN) != 0) {
4249 stack_entry = gap_entry->next;
4250 if ((stack_entry->eflags & MAP_ENTRY_GROWS_DOWN) == 0 ||
4251 stack_entry->start != gap_entry->end)
4252 return (KERN_FAILURE);
4253 grow_amount = round_page(stack_entry->start - addr);
4255 } else if ((gap_entry->eflags & MAP_ENTRY_STACK_GAP_UP) != 0) {
4256 stack_entry = gap_entry->prev;
4257 if ((stack_entry->eflags & MAP_ENTRY_GROWS_UP) == 0 ||
4258 stack_entry->end != gap_entry->start)
4259 return (KERN_FAILURE);
4260 grow_amount = round_page(addr + 1 - stack_entry->end);
4263 return (KERN_FAILURE);
4265 max_grow = gap_entry->end - gap_entry->start;
4266 if (guard > max_grow)
4267 return (KERN_NO_SPACE);
4269 if (grow_amount > max_grow)
4270 return (KERN_NO_SPACE);
4273 * If this is the main process stack, see if we're over the stack
4276 is_procstack = addr >= (vm_offset_t)vm->vm_maxsaddr &&
4277 addr < (vm_offset_t)p->p_sysent->sv_usrstack;
4278 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim))
4279 return (KERN_NO_SPACE);
4284 if (is_procstack && racct_set(p, RACCT_STACK,
4285 ctob(vm->vm_ssize) + grow_amount)) {
4287 return (KERN_NO_SPACE);
4293 grow_amount = roundup(grow_amount, sgrowsiz);
4294 if (grow_amount > max_grow)
4295 grow_amount = max_grow;
4296 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) {
4297 grow_amount = trunc_page((vm_size_t)stacklim) -
4303 limit = racct_get_available(p, RACCT_STACK);
4305 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > limit))
4306 grow_amount = limit - ctob(vm->vm_ssize);
4309 if (!old_mlock && (map->flags & MAP_WIREFUTURE) != 0) {
4310 if (ptoa(pmap_wired_count(map->pmap)) + grow_amount > lmemlim) {
4317 if (racct_set(p, RACCT_MEMLOCK,
4318 ptoa(pmap_wired_count(map->pmap)) + grow_amount)) {
4328 /* If we would blow our VMEM resource limit, no go */
4329 if (map->size + grow_amount > vmemlim) {
4336 if (racct_set(p, RACCT_VMEM, map->size + grow_amount)) {
4345 if (vm_map_lock_upgrade(map)) {
4347 vm_map_lock_read(map);
4352 grow_start = gap_entry->end - grow_amount;
4353 if (gap_entry->start + grow_amount == gap_entry->end) {
4354 gap_start = gap_entry->start;
4355 gap_end = gap_entry->end;
4356 vm_map_entry_delete(map, gap_entry);
4359 MPASS(gap_entry->start < gap_entry->end - grow_amount);
4360 vm_map_entry_resize(map, gap_entry, -grow_amount);
4361 gap_deleted = false;
4363 rv = vm_map_insert(map, NULL, 0, grow_start,
4364 grow_start + grow_amount,
4365 stack_entry->protection, stack_entry->max_protection,
4366 MAP_STACK_GROWS_DOWN);
4367 if (rv != KERN_SUCCESS) {
4369 rv1 = vm_map_insert(map, NULL, 0, gap_start,
4370 gap_end, VM_PROT_NONE, VM_PROT_NONE,
4371 MAP_CREATE_GUARD | MAP_CREATE_STACK_GAP_DN);
4372 MPASS(rv1 == KERN_SUCCESS);
4374 vm_map_entry_resize(map, gap_entry,
4378 grow_start = stack_entry->end;
4379 cred = stack_entry->cred;
4380 if (cred == NULL && stack_entry->object.vm_object != NULL)
4381 cred = stack_entry->object.vm_object->cred;
4382 if (cred != NULL && !swap_reserve_by_cred(grow_amount, cred))
4384 /* Grow the underlying object if applicable. */
4385 else if (stack_entry->object.vm_object == NULL ||
4386 vm_object_coalesce(stack_entry->object.vm_object,
4387 stack_entry->offset,
4388 (vm_size_t)(stack_entry->end - stack_entry->start),
4389 grow_amount, cred != NULL)) {
4390 if (gap_entry->start + grow_amount == gap_entry->end) {
4391 vm_map_entry_delete(map, gap_entry);
4392 vm_map_entry_resize(map, stack_entry,
4395 gap_entry->start += grow_amount;
4396 stack_entry->end += grow_amount;
4398 map->size += grow_amount;
4403 if (rv == KERN_SUCCESS && is_procstack)
4404 vm->vm_ssize += btoc(grow_amount);
4407 * Heed the MAP_WIREFUTURE flag if it was set for this process.
4409 if (rv == KERN_SUCCESS && (map->flags & MAP_WIREFUTURE) != 0) {
4410 rv = vm_map_wire_locked(map, grow_start,
4411 grow_start + grow_amount,
4412 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
4414 vm_map_lock_downgrade(map);
4418 if (racct_enable && rv != KERN_SUCCESS) {
4420 error = racct_set(p, RACCT_VMEM, map->size);
4421 KASSERT(error == 0, ("decreasing RACCT_VMEM failed"));
4423 error = racct_set(p, RACCT_MEMLOCK,
4424 ptoa(pmap_wired_count(map->pmap)));
4425 KASSERT(error == 0, ("decreasing RACCT_MEMLOCK failed"));
4427 error = racct_set(p, RACCT_STACK, ctob(vm->vm_ssize));
4428 KASSERT(error == 0, ("decreasing RACCT_STACK failed"));
4437 * Unshare the specified VM space for exec. If other processes are
4438 * mapped to it, then create a new one. The new vmspace is null.
4441 vmspace_exec(struct proc *p, vm_offset_t minuser, vm_offset_t maxuser)
4443 struct vmspace *oldvmspace = p->p_vmspace;
4444 struct vmspace *newvmspace;
4446 KASSERT((curthread->td_pflags & TDP_EXECVMSPC) == 0,
4447 ("vmspace_exec recursed"));
4448 newvmspace = vmspace_alloc(minuser, maxuser, pmap_pinit);
4449 if (newvmspace == NULL)
4451 newvmspace->vm_swrss = oldvmspace->vm_swrss;
4453 * This code is written like this for prototype purposes. The
4454 * goal is to avoid running down the vmspace here, but let the
4455 * other process's that are still using the vmspace to finally
4456 * run it down. Even though there is little or no chance of blocking
4457 * here, it is a good idea to keep this form for future mods.
4459 PROC_VMSPACE_LOCK(p);
4460 p->p_vmspace = newvmspace;
4461 PROC_VMSPACE_UNLOCK(p);
4462 if (p == curthread->td_proc)
4463 pmap_activate(curthread);
4464 curthread->td_pflags |= TDP_EXECVMSPC;
4469 * Unshare the specified VM space for forcing COW. This
4470 * is called by rfork, for the (RFMEM|RFPROC) == 0 case.
4473 vmspace_unshare(struct proc *p)
4475 struct vmspace *oldvmspace = p->p_vmspace;
4476 struct vmspace *newvmspace;
4477 vm_ooffset_t fork_charge;
4479 if (oldvmspace->vm_refcnt == 1)
4482 newvmspace = vmspace_fork(oldvmspace, &fork_charge);
4483 if (newvmspace == NULL)
4485 if (!swap_reserve_by_cred(fork_charge, p->p_ucred)) {
4486 vmspace_free(newvmspace);
4489 PROC_VMSPACE_LOCK(p);
4490 p->p_vmspace = newvmspace;
4491 PROC_VMSPACE_UNLOCK(p);
4492 if (p == curthread->td_proc)
4493 pmap_activate(curthread);
4494 vmspace_free(oldvmspace);
4501 * Finds the VM object, offset, and
4502 * protection for a given virtual address in the
4503 * specified map, assuming a page fault of the
4506 * Leaves the map in question locked for read; return
4507 * values are guaranteed until a vm_map_lookup_done
4508 * call is performed. Note that the map argument
4509 * is in/out; the returned map must be used in
4510 * the call to vm_map_lookup_done.
4512 * A handle (out_entry) is returned for use in
4513 * vm_map_lookup_done, to make that fast.
4515 * If a lookup is requested with "write protection"
4516 * specified, the map may be changed to perform virtual
4517 * copying operations, although the data referenced will
4521 vm_map_lookup(vm_map_t *var_map, /* IN/OUT */
4523 vm_prot_t fault_typea,
4524 vm_map_entry_t *out_entry, /* OUT */
4525 vm_object_t *object, /* OUT */
4526 vm_pindex_t *pindex, /* OUT */
4527 vm_prot_t *out_prot, /* OUT */
4528 boolean_t *wired) /* OUT */
4530 vm_map_entry_t entry;
4531 vm_map_t map = *var_map;
4533 vm_prot_t fault_type = fault_typea;
4534 vm_object_t eobject;
4540 vm_map_lock_read(map);
4544 * Lookup the faulting address.
4546 if (!vm_map_lookup_entry(map, vaddr, out_entry)) {
4547 vm_map_unlock_read(map);
4548 return (KERN_INVALID_ADDRESS);
4556 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
4557 vm_map_t old_map = map;
4559 *var_map = map = entry->object.sub_map;
4560 vm_map_unlock_read(old_map);
4565 * Check whether this task is allowed to have this page.
4567 prot = entry->protection;
4568 if ((fault_typea & VM_PROT_FAULT_LOOKUP) != 0) {
4569 fault_typea &= ~VM_PROT_FAULT_LOOKUP;
4570 if (prot == VM_PROT_NONE && map != kernel_map &&
4571 (entry->eflags & MAP_ENTRY_GUARD) != 0 &&
4572 (entry->eflags & (MAP_ENTRY_STACK_GAP_DN |
4573 MAP_ENTRY_STACK_GAP_UP)) != 0 &&
4574 vm_map_growstack(map, vaddr, entry) == KERN_SUCCESS)
4575 goto RetryLookupLocked;
4577 fault_type &= VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE;
4578 if ((fault_type & prot) != fault_type || prot == VM_PROT_NONE) {
4579 vm_map_unlock_read(map);
4580 return (KERN_PROTECTION_FAILURE);
4582 KASSERT((prot & VM_PROT_WRITE) == 0 || (entry->eflags &
4583 (MAP_ENTRY_USER_WIRED | MAP_ENTRY_NEEDS_COPY)) !=
4584 (MAP_ENTRY_USER_WIRED | MAP_ENTRY_NEEDS_COPY),
4585 ("entry %p flags %x", entry, entry->eflags));
4586 if ((fault_typea & VM_PROT_COPY) != 0 &&
4587 (entry->max_protection & VM_PROT_WRITE) == 0 &&
4588 (entry->eflags & MAP_ENTRY_COW) == 0) {
4589 vm_map_unlock_read(map);
4590 return (KERN_PROTECTION_FAILURE);
4594 * If this page is not pageable, we have to get it for all possible
4597 *wired = (entry->wired_count != 0);
4599 fault_type = entry->protection;
4600 size = entry->end - entry->start;
4602 * If the entry was copy-on-write, we either ...
4604 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
4606 * If we want to write the page, we may as well handle that
4607 * now since we've got the map locked.
4609 * If we don't need to write the page, we just demote the
4610 * permissions allowed.
4612 if ((fault_type & VM_PROT_WRITE) != 0 ||
4613 (fault_typea & VM_PROT_COPY) != 0) {
4615 * Make a new object, and place it in the object
4616 * chain. Note that no new references have appeared
4617 * -- one just moved from the map to the new
4620 if (vm_map_lock_upgrade(map))
4623 if (entry->cred == NULL) {
4625 * The debugger owner is charged for
4628 cred = curthread->td_ucred;
4630 if (!swap_reserve_by_cred(size, cred)) {
4633 return (KERN_RESOURCE_SHORTAGE);
4637 vm_object_shadow(&entry->object.vm_object,
4638 &entry->offset, size);
4639 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
4640 eobject = entry->object.vm_object;
4641 if (eobject->cred != NULL) {
4643 * The object was not shadowed.
4645 swap_release_by_cred(size, entry->cred);
4646 crfree(entry->cred);
4648 } else if (entry->cred != NULL) {
4649 VM_OBJECT_WLOCK(eobject);
4650 eobject->cred = entry->cred;
4651 eobject->charge = size;
4652 VM_OBJECT_WUNLOCK(eobject);
4656 vm_map_lock_downgrade(map);
4659 * We're attempting to read a copy-on-write page --
4660 * don't allow writes.
4662 prot &= ~VM_PROT_WRITE;
4667 * Create an object if necessary.
4669 if (entry->object.vm_object == NULL &&
4671 if (vm_map_lock_upgrade(map))
4673 entry->object.vm_object = vm_object_allocate(OBJT_DEFAULT,
4676 if (entry->cred != NULL) {
4677 VM_OBJECT_WLOCK(entry->object.vm_object);
4678 entry->object.vm_object->cred = entry->cred;
4679 entry->object.vm_object->charge = size;
4680 VM_OBJECT_WUNLOCK(entry->object.vm_object);
4683 vm_map_lock_downgrade(map);
4687 * Return the object/offset from this entry. If the entry was
4688 * copy-on-write or empty, it has been fixed up.
4690 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
4691 *object = entry->object.vm_object;
4694 return (KERN_SUCCESS);
4698 * vm_map_lookup_locked:
4700 * Lookup the faulting address. A version of vm_map_lookup that returns
4701 * KERN_FAILURE instead of blocking on map lock or memory allocation.
4704 vm_map_lookup_locked(vm_map_t *var_map, /* IN/OUT */
4706 vm_prot_t fault_typea,
4707 vm_map_entry_t *out_entry, /* OUT */
4708 vm_object_t *object, /* OUT */
4709 vm_pindex_t *pindex, /* OUT */
4710 vm_prot_t *out_prot, /* OUT */
4711 boolean_t *wired) /* OUT */
4713 vm_map_entry_t entry;
4714 vm_map_t map = *var_map;
4716 vm_prot_t fault_type = fault_typea;
4719 * Lookup the faulting address.
4721 if (!vm_map_lookup_entry(map, vaddr, out_entry))
4722 return (KERN_INVALID_ADDRESS);
4727 * Fail if the entry refers to a submap.
4729 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP)
4730 return (KERN_FAILURE);
4733 * Check whether this task is allowed to have this page.
4735 prot = entry->protection;
4736 fault_type &= VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE;
4737 if ((fault_type & prot) != fault_type)
4738 return (KERN_PROTECTION_FAILURE);
4741 * If this page is not pageable, we have to get it for all possible
4744 *wired = (entry->wired_count != 0);
4746 fault_type = entry->protection;
4748 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
4750 * Fail if the entry was copy-on-write for a write fault.
4752 if (fault_type & VM_PROT_WRITE)
4753 return (KERN_FAILURE);
4755 * We're attempting to read a copy-on-write page --
4756 * don't allow writes.
4758 prot &= ~VM_PROT_WRITE;
4762 * Fail if an object should be created.
4764 if (entry->object.vm_object == NULL && !map->system_map)
4765 return (KERN_FAILURE);
4768 * Return the object/offset from this entry. If the entry was
4769 * copy-on-write or empty, it has been fixed up.
4771 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
4772 *object = entry->object.vm_object;
4775 return (KERN_SUCCESS);
4779 * vm_map_lookup_done:
4781 * Releases locks acquired by a vm_map_lookup
4782 * (according to the handle returned by that lookup).
4785 vm_map_lookup_done(vm_map_t map, vm_map_entry_t entry)
4788 * Unlock the main-level map
4790 vm_map_unlock_read(map);
4794 vm_map_max_KBI(const struct vm_map *map)
4797 return (vm_map_max(map));
4801 vm_map_min_KBI(const struct vm_map *map)
4804 return (vm_map_min(map));
4808 vm_map_pmap_KBI(vm_map_t map)
4814 #include "opt_ddb.h"
4816 #include <sys/kernel.h>
4818 #include <ddb/ddb.h>
4821 vm_map_print(vm_map_t map)
4823 vm_map_entry_t entry, prev;
4825 db_iprintf("Task map %p: pmap=%p, nentries=%d, version=%u\n",
4827 (void *)map->pmap, map->nentries, map->timestamp);
4830 for (prev = &map->header; (entry = prev->next) != &map->header;
4832 db_iprintf("map entry %p: start=%p, end=%p, eflags=%#x, \n",
4833 (void *)entry, (void *)entry->start, (void *)entry->end,
4836 static char *inheritance_name[4] =
4837 {"share", "copy", "none", "donate_copy"};
4839 db_iprintf(" prot=%x/%x/%s",
4841 entry->max_protection,
4842 inheritance_name[(int)(unsigned char)
4843 entry->inheritance]);
4844 if (entry->wired_count != 0)
4845 db_printf(", wired");
4847 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
4848 db_printf(", share=%p, offset=0x%jx\n",
4849 (void *)entry->object.sub_map,
4850 (uintmax_t)entry->offset);
4851 if (prev == &map->header ||
4852 prev->object.sub_map !=
4853 entry->object.sub_map) {
4855 vm_map_print((vm_map_t)entry->object.sub_map);
4859 if (entry->cred != NULL)
4860 db_printf(", ruid %d", entry->cred->cr_ruid);
4861 db_printf(", object=%p, offset=0x%jx",
4862 (void *)entry->object.vm_object,
4863 (uintmax_t)entry->offset);
4864 if (entry->object.vm_object && entry->object.vm_object->cred)
4865 db_printf(", obj ruid %d charge %jx",
4866 entry->object.vm_object->cred->cr_ruid,
4867 (uintmax_t)entry->object.vm_object->charge);
4868 if (entry->eflags & MAP_ENTRY_COW)
4869 db_printf(", copy (%s)",
4870 (entry->eflags & MAP_ENTRY_NEEDS_COPY) ? "needed" : "done");
4873 if (prev == &map->header ||
4874 prev->object.vm_object !=
4875 entry->object.vm_object) {
4877 vm_object_print((db_expr_t)(intptr_t)
4878 entry->object.vm_object,
4887 DB_SHOW_COMMAND(map, map)
4891 db_printf("usage: show map <addr>\n");
4894 vm_map_print((vm_map_t)addr);
4897 DB_SHOW_COMMAND(procvm, procvm)
4902 p = db_lookup_proc(addr);
4907 db_printf("p = %p, vmspace = %p, map = %p, pmap = %p\n",
4908 (void *)p, (void *)p->p_vmspace, (void *)&p->p_vmspace->vm_map,
4909 (void *)vmspace_pmap(p->p_vmspace));
4911 vm_map_print((vm_map_t)&p->p_vmspace->vm_map);