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>
73 #include <sys/kernel.h>
76 #include <sys/mutex.h>
78 #include <sys/vmmeter.h>
80 #include <sys/vnode.h>
81 #include <sys/racct.h>
82 #include <sys/resourcevar.h>
83 #include <sys/rwlock.h>
85 #include <sys/sysctl.h>
86 #include <sys/sysent.h>
90 #include <vm/vm_param.h>
92 #include <vm/vm_map.h>
93 #include <vm/vm_page.h>
94 #include <vm/vm_pageout.h>
95 #include <vm/vm_object.h>
96 #include <vm/vm_pager.h>
97 #include <vm/vm_kern.h>
98 #include <vm/vm_extern.h>
99 #include <vm/vnode_pager.h>
100 #include <vm/swap_pager.h>
104 * Virtual memory maps provide for the mapping, protection,
105 * and sharing of virtual memory objects. In addition,
106 * this module provides for an efficient virtual copy of
107 * memory from one map to another.
109 * Synchronization is required prior to most operations.
111 * Maps consist of an ordered doubly-linked list of simple
112 * entries; a self-adjusting binary search tree of these
113 * entries is used to speed up lookups.
115 * Since portions of maps are specified by start/end addresses,
116 * which may not align with existing map entries, all
117 * routines merely "clip" entries to these start/end values.
118 * [That is, an entry is split into two, bordering at a
119 * start or end value.] Note that these clippings may not
120 * always be necessary (as the two resulting entries are then
121 * not changed); however, the clipping is done for convenience.
123 * As mentioned above, virtual copy operations are performed
124 * by copying VM object references from one map to
125 * another, and then marking both regions as copy-on-write.
128 static struct mtx map_sleep_mtx;
129 static uma_zone_t mapentzone;
130 static uma_zone_t kmapentzone;
131 static uma_zone_t mapzone;
132 static uma_zone_t vmspace_zone;
133 static int vmspace_zinit(void *mem, int size, int flags);
134 static int vm_map_zinit(void *mem, int ize, int flags);
135 static void _vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min,
137 static void vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map);
138 static void vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry);
139 static void vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry);
140 static int vm_map_growstack(vm_map_t map, vm_offset_t addr,
141 vm_map_entry_t gap_entry);
142 static void vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot,
143 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags);
145 static void vm_map_zdtor(void *mem, int size, void *arg);
146 static void vmspace_zdtor(void *mem, int size, void *arg);
148 static int vm_map_stack_locked(vm_map_t map, vm_offset_t addrbos,
149 vm_size_t max_ssize, vm_size_t growsize, vm_prot_t prot, vm_prot_t max,
151 static void vm_map_wire_entry_failure(vm_map_t map, vm_map_entry_t entry,
152 vm_offset_t failed_addr);
154 #define ENTRY_CHARGED(e) ((e)->cred != NULL || \
155 ((e)->object.vm_object != NULL && (e)->object.vm_object->cred != NULL && \
156 !((e)->eflags & MAP_ENTRY_NEEDS_COPY)))
159 * PROC_VMSPACE_{UN,}LOCK() can be a noop as long as vmspaces are type
162 #define PROC_VMSPACE_LOCK(p) do { } while (0)
163 #define PROC_VMSPACE_UNLOCK(p) do { } while (0)
166 * VM_MAP_RANGE_CHECK: [ internal use only ]
168 * Asserts that the starting and ending region
169 * addresses fall within the valid range of the map.
171 #define VM_MAP_RANGE_CHECK(map, start, end) \
173 if (start < vm_map_min(map)) \
174 start = vm_map_min(map); \
175 if (end > vm_map_max(map)) \
176 end = vm_map_max(map); \
184 * Initialize the vm_map module. Must be called before
185 * any other vm_map routines.
187 * Map and entry structures are allocated from the general
188 * purpose memory pool with some exceptions:
190 * - The kernel map and kmem submap are allocated statically.
191 * - Kernel map entries are allocated out of a static pool.
193 * These restrictions are necessary since malloc() uses the
194 * maps and requires map entries.
200 mtx_init(&map_sleep_mtx, "vm map sleep mutex", NULL, MTX_DEF);
201 mapzone = uma_zcreate("MAP", sizeof(struct vm_map), NULL,
207 vm_map_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
208 uma_prealloc(mapzone, MAX_KMAP);
209 kmapentzone = uma_zcreate("KMAP ENTRY", sizeof(struct vm_map_entry),
210 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR,
211 UMA_ZONE_MTXCLASS | UMA_ZONE_VM);
212 mapentzone = uma_zcreate("MAP ENTRY", sizeof(struct vm_map_entry),
213 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
214 vmspace_zone = uma_zcreate("VMSPACE", sizeof(struct vmspace), NULL,
220 vmspace_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
224 vmspace_zinit(void *mem, int size, int flags)
228 vm = (struct vmspace *)mem;
230 vm->vm_map.pmap = NULL;
231 (void)vm_map_zinit(&vm->vm_map, sizeof(vm->vm_map), flags);
232 PMAP_LOCK_INIT(vmspace_pmap(vm));
237 vm_map_zinit(void *mem, int size, int flags)
242 memset(map, 0, sizeof(*map));
243 mtx_init(&map->system_mtx, "vm map (system)", NULL, MTX_DEF | MTX_DUPOK);
244 sx_init(&map->lock, "vm map (user)");
250 vmspace_zdtor(void *mem, int size, void *arg)
254 vm = (struct vmspace *)mem;
256 vm_map_zdtor(&vm->vm_map, sizeof(vm->vm_map), arg);
259 vm_map_zdtor(void *mem, int size, void *arg)
264 KASSERT(map->nentries == 0,
265 ("map %p nentries == %d on free.",
266 map, map->nentries));
267 KASSERT(map->size == 0,
268 ("map %p size == %lu on free.",
269 map, (unsigned long)map->size));
271 #endif /* INVARIANTS */
274 * Allocate a vmspace structure, including a vm_map and pmap,
275 * and initialize those structures. The refcnt is set to 1.
277 * If 'pinit' is NULL then the embedded pmap is initialized via pmap_pinit().
280 vmspace_alloc(vm_offset_t min, vm_offset_t max, pmap_pinit_t pinit)
284 vm = uma_zalloc(vmspace_zone, M_WAITOK);
285 KASSERT(vm->vm_map.pmap == NULL, ("vm_map.pmap must be NULL"));
286 if (!pinit(vmspace_pmap(vm))) {
287 uma_zfree(vmspace_zone, vm);
290 CTR1(KTR_VM, "vmspace_alloc: %p", vm);
291 _vm_map_init(&vm->vm_map, vmspace_pmap(vm), min, max);
306 vmspace_container_reset(struct proc *p)
310 racct_set(p, RACCT_DATA, 0);
311 racct_set(p, RACCT_STACK, 0);
312 racct_set(p, RACCT_RSS, 0);
313 racct_set(p, RACCT_MEMLOCK, 0);
314 racct_set(p, RACCT_VMEM, 0);
320 vmspace_dofree(struct vmspace *vm)
323 CTR1(KTR_VM, "vmspace_free: %p", vm);
326 * Make sure any SysV shm is freed, it might not have been in
332 * Lock the map, to wait out all other references to it.
333 * Delete all of the mappings and pages they hold, then call
334 * the pmap module to reclaim anything left.
336 (void)vm_map_remove(&vm->vm_map, vm_map_min(&vm->vm_map),
337 vm_map_max(&vm->vm_map));
339 pmap_release(vmspace_pmap(vm));
340 vm->vm_map.pmap = NULL;
341 uma_zfree(vmspace_zone, vm);
345 vmspace_free(struct vmspace *vm)
348 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
349 "vmspace_free() called");
351 if (vm->vm_refcnt == 0)
352 panic("vmspace_free: attempt to free already freed vmspace");
354 if (atomic_fetchadd_int(&vm->vm_refcnt, -1) == 1)
359 vmspace_exitfree(struct proc *p)
363 PROC_VMSPACE_LOCK(p);
366 PROC_VMSPACE_UNLOCK(p);
367 KASSERT(vm == &vmspace0, ("vmspace_exitfree: wrong vmspace"));
372 vmspace_exit(struct thread *td)
379 * Release user portion of address space.
380 * This releases references to vnodes,
381 * which could cause I/O if the file has been unlinked.
382 * Need to do this early enough that we can still sleep.
384 * The last exiting process to reach this point releases as
385 * much of the environment as it can. vmspace_dofree() is the
386 * slower fallback in case another process had a temporary
387 * reference to the vmspace.
392 atomic_add_int(&vmspace0.vm_refcnt, 1);
393 refcnt = vm->vm_refcnt;
395 if (refcnt > 1 && p->p_vmspace != &vmspace0) {
396 /* Switch now since other proc might free vmspace */
397 PROC_VMSPACE_LOCK(p);
398 p->p_vmspace = &vmspace0;
399 PROC_VMSPACE_UNLOCK(p);
402 } while (!atomic_fcmpset_int(&vm->vm_refcnt, &refcnt, refcnt - 1));
404 if (p->p_vmspace != vm) {
405 /* vmspace not yet freed, switch back */
406 PROC_VMSPACE_LOCK(p);
408 PROC_VMSPACE_UNLOCK(p);
411 pmap_remove_pages(vmspace_pmap(vm));
412 /* Switch now since this proc will free vmspace */
413 PROC_VMSPACE_LOCK(p);
414 p->p_vmspace = &vmspace0;
415 PROC_VMSPACE_UNLOCK(p);
421 vmspace_container_reset(p);
425 /* Acquire reference to vmspace owned by another process. */
428 vmspace_acquire_ref(struct proc *p)
433 PROC_VMSPACE_LOCK(p);
436 PROC_VMSPACE_UNLOCK(p);
439 refcnt = vm->vm_refcnt;
441 if (refcnt <= 0) { /* Avoid 0->1 transition */
442 PROC_VMSPACE_UNLOCK(p);
445 } while (!atomic_fcmpset_int(&vm->vm_refcnt, &refcnt, refcnt + 1));
446 if (vm != p->p_vmspace) {
447 PROC_VMSPACE_UNLOCK(p);
451 PROC_VMSPACE_UNLOCK(p);
456 * Switch between vmspaces in an AIO kernel process.
458 * The new vmspace is either the vmspace of a user process obtained
459 * from an active AIO request or the initial vmspace of the AIO kernel
460 * process (when it is idling). Because user processes will block to
461 * drain any active AIO requests before proceeding in exit() or
462 * execve(), the reference count for vmspaces from AIO requests can
463 * never be 0. Similarly, AIO kernel processes hold an extra
464 * reference on their initial vmspace for the life of the process. As
465 * a result, the 'newvm' vmspace always has a non-zero reference
466 * count. This permits an additional reference on 'newvm' to be
467 * acquired via a simple atomic increment rather than the loop in
468 * vmspace_acquire_ref() above.
471 vmspace_switch_aio(struct vmspace *newvm)
473 struct vmspace *oldvm;
475 /* XXX: Need some way to assert that this is an aio daemon. */
477 KASSERT(newvm->vm_refcnt > 0,
478 ("vmspace_switch_aio: newvm unreferenced"));
480 oldvm = curproc->p_vmspace;
485 * Point to the new address space and refer to it.
487 curproc->p_vmspace = newvm;
488 atomic_add_int(&newvm->vm_refcnt, 1);
490 /* Activate the new mapping. */
491 pmap_activate(curthread);
497 _vm_map_lock(vm_map_t map, const char *file, int line)
501 mtx_lock_flags_(&map->system_mtx, 0, file, line);
503 sx_xlock_(&map->lock, file, line);
508 vm_map_entry_set_vnode_text(vm_map_entry_t entry, bool add)
514 if ((entry->eflags & MAP_ENTRY_VN_EXEC) == 0)
516 KASSERT((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0,
517 ("Submap with execs"));
518 object = entry->object.vm_object;
519 KASSERT(object != NULL, ("No object for text, entry %p", entry));
520 if ((object->flags & OBJ_ANON) != 0)
521 object = object->handle;
523 KASSERT(object->backing_object == NULL,
524 ("non-anon object %p shadows", object));
525 KASSERT(object != NULL, ("No content object for text, entry %p obj %p",
526 entry, entry->object.vm_object));
529 * Mostly, we do not lock the backing object. It is
530 * referenced by the entry we are processing, so it cannot go
535 if (object->type == OBJT_DEAD) {
537 * For OBJT_DEAD objects, v_writecount was handled in
538 * vnode_pager_dealloc().
540 } else if (object->type == OBJT_VNODE) {
542 } else if (object->type == OBJT_SWAP) {
543 KASSERT((object->flags & OBJ_TMPFS_NODE) != 0,
544 ("vm_map_entry_set_vnode_text: swap and !TMPFS "
545 "entry %p, object %p, add %d", entry, object, add));
547 * Tmpfs VREG node, which was reclaimed, has
548 * OBJ_TMPFS_NODE flag set, but not OBJ_TMPFS. In
549 * this case there is no v_writecount to adjust.
551 VM_OBJECT_RLOCK(object);
552 if ((object->flags & OBJ_TMPFS) != 0) {
553 vp = object->un_pager.swp.swp_tmpfs;
559 VM_OBJECT_RUNLOCK(object);
562 ("vm_map_entry_set_vnode_text: wrong object type, "
563 "entry %p, object %p, add %d", entry, object, add));
567 VOP_SET_TEXT_CHECKED(vp);
569 vn_lock(vp, LK_SHARED | LK_RETRY);
570 VOP_UNSET_TEXT_CHECKED(vp);
579 * Use a different name for this vm_map_entry field when it's use
580 * is not consistent with its use as part of an ordered search tree.
582 #define defer_next right
585 vm_map_process_deferred(void)
588 vm_map_entry_t entry, next;
592 entry = td->td_map_def_user;
593 td->td_map_def_user = NULL;
594 while (entry != NULL) {
595 next = entry->defer_next;
596 MPASS((entry->eflags & (MAP_ENTRY_WRITECNT |
597 MAP_ENTRY_VN_EXEC)) != (MAP_ENTRY_WRITECNT |
599 if ((entry->eflags & MAP_ENTRY_WRITECNT) != 0) {
601 * Decrement the object's writemappings and
602 * possibly the vnode's v_writecount.
604 KASSERT((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0,
605 ("Submap with writecount"));
606 object = entry->object.vm_object;
607 KASSERT(object != NULL, ("No object for writecount"));
608 vm_pager_release_writecount(object, entry->start,
611 vm_map_entry_set_vnode_text(entry, false);
612 vm_map_entry_deallocate(entry, FALSE);
619 _vm_map_assert_locked(vm_map_t map, const char *file, int line)
623 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
625 sx_assert_(&map->lock, SA_XLOCKED, file, line);
628 #define VM_MAP_ASSERT_LOCKED(map) \
629 _vm_map_assert_locked(map, LOCK_FILE, LOCK_LINE)
631 enum { VMMAP_CHECK_NONE, VMMAP_CHECK_UNLOCK, VMMAP_CHECK_ALL };
633 static int enable_vmmap_check = VMMAP_CHECK_UNLOCK;
635 static int enable_vmmap_check = VMMAP_CHECK_NONE;
637 SYSCTL_INT(_debug, OID_AUTO, vmmap_check, CTLFLAG_RWTUN,
638 &enable_vmmap_check, 0, "Enable vm map consistency checking");
640 static void _vm_map_assert_consistent(vm_map_t map, int check);
642 #define VM_MAP_ASSERT_CONSISTENT(map) \
643 _vm_map_assert_consistent(map, VMMAP_CHECK_ALL)
645 #define VM_MAP_UNLOCK_CONSISTENT(map) do { \
646 if (map->nupdates > map->nentries) { \
647 _vm_map_assert_consistent(map, VMMAP_CHECK_UNLOCK); \
652 #define VM_MAP_UNLOCK_CONSISTENT(map)
655 #define VM_MAP_ASSERT_LOCKED(map)
656 #define VM_MAP_ASSERT_CONSISTENT(map)
657 #define VM_MAP_UNLOCK_CONSISTENT(map)
658 #endif /* INVARIANTS */
661 _vm_map_unlock(vm_map_t map, const char *file, int line)
664 VM_MAP_UNLOCK_CONSISTENT(map);
666 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
668 sx_xunlock_(&map->lock, file, line);
669 vm_map_process_deferred();
674 _vm_map_lock_read(vm_map_t map, const char *file, int line)
678 mtx_lock_flags_(&map->system_mtx, 0, file, line);
680 sx_slock_(&map->lock, file, line);
684 _vm_map_unlock_read(vm_map_t map, const char *file, int line)
688 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
690 sx_sunlock_(&map->lock, file, line);
691 vm_map_process_deferred();
696 _vm_map_trylock(vm_map_t map, const char *file, int line)
700 error = map->system_map ?
701 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) :
702 !sx_try_xlock_(&map->lock, file, line);
709 _vm_map_trylock_read(vm_map_t map, const char *file, int line)
713 error = map->system_map ?
714 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) :
715 !sx_try_slock_(&map->lock, file, line);
720 * _vm_map_lock_upgrade: [ internal use only ]
722 * Tries to upgrade a read (shared) lock on the specified map to a write
723 * (exclusive) lock. Returns the value "0" if the upgrade succeeds and a
724 * non-zero value if the upgrade fails. If the upgrade fails, the map is
725 * returned without a read or write lock held.
727 * Requires that the map be read locked.
730 _vm_map_lock_upgrade(vm_map_t map, const char *file, int line)
732 unsigned int last_timestamp;
734 if (map->system_map) {
735 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
737 if (!sx_try_upgrade_(&map->lock, file, line)) {
738 last_timestamp = map->timestamp;
739 sx_sunlock_(&map->lock, file, line);
740 vm_map_process_deferred();
742 * If the map's timestamp does not change while the
743 * map is unlocked, then the upgrade succeeds.
745 sx_xlock_(&map->lock, file, line);
746 if (last_timestamp != map->timestamp) {
747 sx_xunlock_(&map->lock, file, line);
757 _vm_map_lock_downgrade(vm_map_t map, const char *file, int line)
760 if (map->system_map) {
761 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
763 VM_MAP_UNLOCK_CONSISTENT(map);
764 sx_downgrade_(&map->lock, file, line);
771 * Returns a non-zero value if the caller holds a write (exclusive) lock
772 * on the specified map and the value "0" otherwise.
775 vm_map_locked(vm_map_t map)
779 return (mtx_owned(&map->system_mtx));
781 return (sx_xlocked(&map->lock));
785 * _vm_map_unlock_and_wait:
787 * Atomically releases the lock on the specified map and puts the calling
788 * thread to sleep. The calling thread will remain asleep until either
789 * vm_map_wakeup() is performed on the map or the specified timeout is
792 * WARNING! This function does not perform deferred deallocations of
793 * objects and map entries. Therefore, the calling thread is expected to
794 * reacquire the map lock after reawakening and later perform an ordinary
795 * unlock operation, such as vm_map_unlock(), before completing its
796 * operation on the map.
799 _vm_map_unlock_and_wait(vm_map_t map, int timo, const char *file, int line)
802 VM_MAP_UNLOCK_CONSISTENT(map);
803 mtx_lock(&map_sleep_mtx);
805 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
807 sx_xunlock_(&map->lock, file, line);
808 return (msleep(&map->root, &map_sleep_mtx, PDROP | PVM, "vmmaps",
815 * Awaken any threads that have slept on the map using
816 * vm_map_unlock_and_wait().
819 vm_map_wakeup(vm_map_t map)
823 * Acquire and release map_sleep_mtx to prevent a wakeup()
824 * from being performed (and lost) between the map unlock
825 * and the msleep() in _vm_map_unlock_and_wait().
827 mtx_lock(&map_sleep_mtx);
828 mtx_unlock(&map_sleep_mtx);
833 vm_map_busy(vm_map_t map)
836 VM_MAP_ASSERT_LOCKED(map);
841 vm_map_unbusy(vm_map_t map)
844 VM_MAP_ASSERT_LOCKED(map);
845 KASSERT(map->busy, ("vm_map_unbusy: not busy"));
846 if (--map->busy == 0 && (map->flags & MAP_BUSY_WAKEUP)) {
847 vm_map_modflags(map, 0, MAP_BUSY_WAKEUP);
853 vm_map_wait_busy(vm_map_t map)
856 VM_MAP_ASSERT_LOCKED(map);
858 vm_map_modflags(map, MAP_BUSY_WAKEUP, 0);
860 msleep(&map->busy, &map->system_mtx, 0, "mbusy", 0);
862 sx_sleep(&map->busy, &map->lock, 0, "mbusy", 0);
868 vmspace_resident_count(struct vmspace *vmspace)
870 return pmap_resident_count(vmspace_pmap(vmspace));
876 * Creates and returns a new empty VM map with
877 * the given physical map structure, and having
878 * the given lower and upper address bounds.
881 vm_map_create(pmap_t pmap, vm_offset_t min, vm_offset_t max)
885 result = uma_zalloc(mapzone, M_WAITOK);
886 CTR1(KTR_VM, "vm_map_create: %p", result);
887 _vm_map_init(result, pmap, min, max);
892 * Initialize an existing vm_map structure
893 * such as that in the vmspace structure.
896 _vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max)
899 map->header.eflags = MAP_ENTRY_HEADER;
900 map->needs_wakeup = FALSE;
903 map->header.end = min;
904 map->header.start = max;
906 map->header.left = map->header.right = &map->header;
917 vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max)
920 _vm_map_init(map, pmap, min, max);
921 mtx_init(&map->system_mtx, "system map", NULL, MTX_DEF | MTX_DUPOK);
922 sx_init(&map->lock, "user map");
926 * vm_map_entry_dispose: [ internal use only ]
928 * Inverse of vm_map_entry_create.
931 vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry)
933 uma_zfree(map->system_map ? kmapentzone : mapentzone, entry);
937 * vm_map_entry_create: [ internal use only ]
939 * Allocates a VM map entry for insertion.
940 * No entry fields are filled in.
942 static vm_map_entry_t
943 vm_map_entry_create(vm_map_t map)
945 vm_map_entry_t new_entry;
948 new_entry = uma_zalloc(kmapentzone, M_NOWAIT);
950 new_entry = uma_zalloc(mapentzone, M_WAITOK);
951 if (new_entry == NULL)
952 panic("vm_map_entry_create: kernel resources exhausted");
957 * vm_map_entry_set_behavior:
959 * Set the expected access behavior, either normal, random, or
963 vm_map_entry_set_behavior(vm_map_entry_t entry, u_char behavior)
965 entry->eflags = (entry->eflags & ~MAP_ENTRY_BEHAV_MASK) |
966 (behavior & MAP_ENTRY_BEHAV_MASK);
970 * vm_map_entry_max_free_{left,right}:
972 * Compute the size of the largest free gap between two entries,
973 * one the root of a tree and the other the ancestor of that root
974 * that is the least or greatest ancestor found on the search path.
976 static inline vm_size_t
977 vm_map_entry_max_free_left(vm_map_entry_t root, vm_map_entry_t left_ancestor)
980 return (root->left != left_ancestor ?
981 root->left->max_free : root->start - left_ancestor->end);
984 static inline vm_size_t
985 vm_map_entry_max_free_right(vm_map_entry_t root, vm_map_entry_t right_ancestor)
988 return (root->right != right_ancestor ?
989 root->right->max_free : right_ancestor->start - root->end);
993 * vm_map_entry_{pred,succ}:
995 * Find the {predecessor, successor} of the entry by taking one step
996 * in the appropriate direction and backtracking as much as necessary.
997 * vm_map_entry_succ is defined in vm_map.h.
999 static inline vm_map_entry_t
1000 vm_map_entry_pred(vm_map_entry_t entry)
1002 vm_map_entry_t prior;
1004 prior = entry->left;
1005 if (prior->right->start < entry->start) {
1007 prior = prior->right;
1008 while (prior->right != entry);
1013 static inline vm_size_t
1014 vm_size_max(vm_size_t a, vm_size_t b)
1017 return (a > b ? a : b);
1020 #define SPLAY_LEFT_STEP(root, y, llist, rlist, test) do { \
1022 vm_size_t max_free; \
1025 * Infer root->right->max_free == root->max_free when \
1026 * y->max_free < root->max_free || root->max_free == 0. \
1027 * Otherwise, look right to find it. \
1030 max_free = root->max_free; \
1031 KASSERT(max_free >= vm_map_entry_max_free_right(root, rlist), \
1032 ("%s: max_free invariant fails", __func__)); \
1033 if (y == llist ? max_free > 0 : max_free - 1 < y->max_free) \
1034 max_free = vm_map_entry_max_free_right(root, rlist); \
1035 if (y != llist && (test)) { \
1036 /* Rotate right and make y root. */ \
1041 if (max_free < y->max_free) \
1042 root->max_free = max_free = \
1043 vm_size_max(max_free, z->max_free); \
1044 } else if (max_free < y->max_free) \
1045 root->max_free = max_free = \
1046 vm_size_max(max_free, root->start - y->end);\
1050 /* Copy right->max_free. Put root on rlist. */ \
1051 root->max_free = max_free; \
1052 KASSERT(max_free == vm_map_entry_max_free_right(root, rlist), \
1053 ("%s: max_free not copied from right", __func__)); \
1054 root->left = rlist; \
1056 root = y != llist ? y : NULL; \
1059 #define SPLAY_RIGHT_STEP(root, y, llist, rlist, test) do { \
1061 vm_size_t max_free; \
1064 * Infer root->left->max_free == root->max_free when \
1065 * y->max_free < root->max_free || root->max_free == 0. \
1066 * Otherwise, look left to find it. \
1069 max_free = root->max_free; \
1070 KASSERT(max_free >= vm_map_entry_max_free_left(root, llist), \
1071 ("%s: max_free invariant fails", __func__)); \
1072 if (y == rlist ? max_free > 0 : max_free - 1 < y->max_free) \
1073 max_free = vm_map_entry_max_free_left(root, llist); \
1074 if (y != rlist && (test)) { \
1075 /* Rotate left and make y root. */ \
1080 if (max_free < y->max_free) \
1081 root->max_free = max_free = \
1082 vm_size_max(max_free, z->max_free); \
1083 } else if (max_free < y->max_free) \
1084 root->max_free = max_free = \
1085 vm_size_max(max_free, y->start - root->end);\
1089 /* Copy left->max_free. Put root on llist. */ \
1090 root->max_free = max_free; \
1091 KASSERT(max_free == vm_map_entry_max_free_left(root, llist), \
1092 ("%s: max_free not copied from left", __func__)); \
1093 root->right = llist; \
1095 root = y != rlist ? y : NULL; \
1099 * Walk down the tree until we find addr or a gap where addr would go, breaking
1100 * off left and right subtrees of nodes less than, or greater than addr. Treat
1101 * subtrees with root->max_free < length as empty trees. llist and rlist are
1102 * the two sides in reverse order (bottom-up), with llist linked by the right
1103 * pointer and rlist linked by the left pointer in the vm_map_entry, and both
1104 * lists terminated by &map->header. This function, and the subsequent call to
1105 * vm_map_splay_merge_{left,right,pred,succ}, rely on the start and end address
1106 * values in &map->header.
1108 static __always_inline vm_map_entry_t
1109 vm_map_splay_split(vm_map_t map, vm_offset_t addr, vm_size_t length,
1110 vm_map_entry_t *llist, vm_map_entry_t *rlist)
1112 vm_map_entry_t left, right, root, y;
1114 left = right = &map->header;
1116 while (root != NULL && root->max_free >= length) {
1117 KASSERT(left->end <= root->start &&
1118 root->end <= right->start,
1119 ("%s: root not within tree bounds", __func__));
1120 if (addr < root->start) {
1121 SPLAY_LEFT_STEP(root, y, left, right,
1122 y->max_free >= length && addr < y->start);
1123 } else if (addr >= root->end) {
1124 SPLAY_RIGHT_STEP(root, y, left, right,
1125 y->max_free >= length && addr >= y->end);
1134 static __always_inline void
1135 vm_map_splay_findnext(vm_map_entry_t root, vm_map_entry_t *rlist)
1137 vm_map_entry_t hi, right, y;
1140 hi = root->right == right ? NULL : root->right;
1144 SPLAY_LEFT_STEP(hi, y, root, right, true);
1149 static __always_inline void
1150 vm_map_splay_findprev(vm_map_entry_t root, vm_map_entry_t *llist)
1152 vm_map_entry_t left, lo, y;
1155 lo = root->left == left ? NULL : root->left;
1159 SPLAY_RIGHT_STEP(lo, y, left, root, true);
1165 vm_map_entry_swap(vm_map_entry_t *a, vm_map_entry_t *b)
1175 * Walk back up the two spines, flip the pointers and set max_free. The
1176 * subtrees of the root go at the bottom of llist and rlist.
1179 vm_map_splay_merge_left_walk(vm_map_entry_t header, vm_map_entry_t root,
1180 vm_map_entry_t tail, vm_size_t max_free, vm_map_entry_t llist)
1184 * The max_free values of the children of llist are in
1185 * llist->max_free and max_free. Update with the
1188 llist->max_free = max_free =
1189 vm_size_max(llist->max_free, max_free);
1190 vm_map_entry_swap(&llist->right, &tail);
1191 vm_map_entry_swap(&tail, &llist);
1192 } while (llist != header);
1198 * When llist is known to be the predecessor of root.
1200 static inline vm_size_t
1201 vm_map_splay_merge_pred(vm_map_entry_t header, vm_map_entry_t root,
1202 vm_map_entry_t llist)
1206 max_free = root->start - llist->end;
1207 if (llist != header) {
1208 max_free = vm_map_splay_merge_left_walk(header, root,
1209 root, max_free, llist);
1211 root->left = header;
1212 header->right = root;
1218 * When llist may or may not be the predecessor of root.
1220 static inline vm_size_t
1221 vm_map_splay_merge_left(vm_map_entry_t header, vm_map_entry_t root,
1222 vm_map_entry_t llist)
1226 max_free = vm_map_entry_max_free_left(root, llist);
1227 if (llist != header) {
1228 max_free = vm_map_splay_merge_left_walk(header, root,
1229 root->left == llist ? root : root->left,
1236 vm_map_splay_merge_right_walk(vm_map_entry_t header, vm_map_entry_t root,
1237 vm_map_entry_t tail, vm_size_t max_free, vm_map_entry_t rlist)
1241 * The max_free values of the children of rlist are in
1242 * rlist->max_free and max_free. Update with the
1245 rlist->max_free = max_free =
1246 vm_size_max(rlist->max_free, max_free);
1247 vm_map_entry_swap(&rlist->left, &tail);
1248 vm_map_entry_swap(&tail, &rlist);
1249 } while (rlist != header);
1255 * When rlist is known to be the succecessor of root.
1257 static inline vm_size_t
1258 vm_map_splay_merge_succ(vm_map_entry_t header, vm_map_entry_t root,
1259 vm_map_entry_t rlist)
1263 max_free = rlist->start - root->end;
1264 if (rlist != header) {
1265 max_free = vm_map_splay_merge_right_walk(header, root,
1266 root, max_free, rlist);
1268 root->right = header;
1269 header->left = root;
1275 * When rlist may or may not be the succecessor of root.
1277 static inline vm_size_t
1278 vm_map_splay_merge_right(vm_map_entry_t header, vm_map_entry_t root,
1279 vm_map_entry_t rlist)
1283 max_free = vm_map_entry_max_free_right(root, rlist);
1284 if (rlist != header) {
1285 max_free = vm_map_splay_merge_right_walk(header, root,
1286 root->right == rlist ? root : root->right,
1295 * The Sleator and Tarjan top-down splay algorithm with the
1296 * following variation. Max_free must be computed bottom-up, so
1297 * on the downward pass, maintain the left and right spines in
1298 * reverse order. Then, make a second pass up each side to fix
1299 * the pointers and compute max_free. The time bound is O(log n)
1302 * The tree is threaded, which means that there are no null pointers.
1303 * When a node has no left child, its left pointer points to its
1304 * predecessor, which the last ancestor on the search path from the root
1305 * where the search branched right. Likewise, when a node has no right
1306 * child, its right pointer points to its successor. The map header node
1307 * is the predecessor of the first map entry, and the successor of the
1310 * The new root is the vm_map_entry containing "addr", or else an
1311 * adjacent entry (lower if possible) if addr is not in the tree.
1313 * The map must be locked, and leaves it so.
1315 * Returns: the new root.
1317 static vm_map_entry_t
1318 vm_map_splay(vm_map_t map, vm_offset_t addr)
1320 vm_map_entry_t header, llist, rlist, root;
1321 vm_size_t max_free_left, max_free_right;
1323 header = &map->header;
1324 root = vm_map_splay_split(map, addr, 0, &llist, &rlist);
1326 max_free_left = vm_map_splay_merge_left(header, root, llist);
1327 max_free_right = vm_map_splay_merge_right(header, root, rlist);
1328 } else if (llist != header) {
1330 * Recover the greatest node in the left
1331 * subtree and make it the root.
1334 llist = root->right;
1335 max_free_left = vm_map_splay_merge_left(header, root, llist);
1336 max_free_right = vm_map_splay_merge_succ(header, root, rlist);
1337 } else if (rlist != header) {
1339 * Recover the least node in the right
1340 * subtree and make it the root.
1344 max_free_left = vm_map_splay_merge_pred(header, root, llist);
1345 max_free_right = vm_map_splay_merge_right(header, root, rlist);
1347 /* There is no root. */
1350 root->max_free = vm_size_max(max_free_left, max_free_right);
1352 VM_MAP_ASSERT_CONSISTENT(map);
1357 * vm_map_entry_{un,}link:
1359 * Insert/remove entries from maps.
1362 vm_map_entry_link(vm_map_t map, vm_map_entry_t entry)
1364 vm_map_entry_t header, llist, rlist, root;
1367 "vm_map_entry_link: map %p, nentries %d, entry %p", map,
1368 map->nentries, entry);
1369 VM_MAP_ASSERT_LOCKED(map);
1371 header = &map->header;
1372 root = vm_map_splay_split(map, entry->start, 0, &llist, &rlist);
1373 KASSERT(root == NULL,
1374 ("vm_map_entry_link: link object already mapped"));
1376 root->max_free = vm_size_max(
1377 vm_map_splay_merge_pred(header, root, llist),
1378 vm_map_splay_merge_succ(header, root, rlist));
1380 VM_MAP_ASSERT_CONSISTENT(map);
1383 enum unlink_merge_type {
1389 vm_map_entry_unlink(vm_map_t map, vm_map_entry_t entry,
1390 enum unlink_merge_type op)
1392 vm_map_entry_t header, llist, rlist, root;
1393 vm_size_t max_free_left, max_free_right;
1395 VM_MAP_ASSERT_LOCKED(map);
1396 header = &map->header;
1397 root = vm_map_splay_split(map, entry->start, 0, &llist, &rlist);
1398 KASSERT(root != NULL,
1399 ("vm_map_entry_unlink: unlink object not mapped"));
1401 vm_map_splay_findprev(root, &llist);
1402 vm_map_splay_findnext(root, &rlist);
1403 if (op == UNLINK_MERGE_NEXT) {
1404 rlist->start = root->start;
1405 rlist->offset = root->offset;
1407 if (llist != header) {
1409 llist = root->right;
1410 max_free_left = vm_map_splay_merge_left(header, root, llist);
1411 max_free_right = vm_map_splay_merge_succ(header, root, rlist);
1412 } else if (rlist != header) {
1415 max_free_left = vm_map_splay_merge_pred(header, root, llist);
1416 max_free_right = vm_map_splay_merge_right(header, root, rlist);
1418 header->left = header->right = header;
1422 root->max_free = vm_size_max(max_free_left, max_free_right);
1424 VM_MAP_ASSERT_CONSISTENT(map);
1426 CTR3(KTR_VM, "vm_map_entry_unlink: map %p, nentries %d, entry %p", map,
1427 map->nentries, entry);
1431 * vm_map_entry_resize:
1433 * Resize a vm_map_entry, recompute the amount of free space that
1434 * follows it and propagate that value up the tree.
1436 * The map must be locked, and leaves it so.
1439 vm_map_entry_resize(vm_map_t map, vm_map_entry_t entry, vm_size_t grow_amount)
1441 vm_map_entry_t header, llist, rlist, root;
1443 VM_MAP_ASSERT_LOCKED(map);
1444 header = &map->header;
1445 root = vm_map_splay_split(map, entry->start, 0, &llist, &rlist);
1446 KASSERT(root != NULL, ("%s: resize object not mapped", __func__));
1447 vm_map_splay_findnext(root, &rlist);
1448 entry->end += grow_amount;
1449 root->max_free = vm_size_max(
1450 vm_map_splay_merge_left(header, root, llist),
1451 vm_map_splay_merge_succ(header, root, rlist));
1453 VM_MAP_ASSERT_CONSISTENT(map);
1454 CTR4(KTR_VM, "%s: map %p, nentries %d, entry %p",
1455 __func__, map, map->nentries, entry);
1459 * vm_map_lookup_entry: [ internal use only ]
1461 * Finds the map entry containing (or
1462 * immediately preceding) the specified address
1463 * in the given map; the entry is returned
1464 * in the "entry" parameter. The boolean
1465 * result indicates whether the address is
1466 * actually contained in the map.
1469 vm_map_lookup_entry(
1471 vm_offset_t address,
1472 vm_map_entry_t *entry) /* OUT */
1474 vm_map_entry_t cur, header, lbound, ubound;
1478 * If the map is empty, then the map entry immediately preceding
1479 * "address" is the map's header.
1481 header = &map->header;
1487 if (address >= cur->start && cur->end > address) {
1491 if ((locked = vm_map_locked(map)) ||
1492 sx_try_upgrade(&map->lock)) {
1494 * Splay requires a write lock on the map. However, it only
1495 * restructures the binary search tree; it does not otherwise
1496 * change the map. Thus, the map's timestamp need not change
1497 * on a temporary upgrade.
1499 cur = vm_map_splay(map, address);
1501 VM_MAP_UNLOCK_CONSISTENT(map);
1502 sx_downgrade(&map->lock);
1506 * If "address" is contained within a map entry, the new root
1507 * is that map entry. Otherwise, the new root is a map entry
1508 * immediately before or after "address".
1510 if (address < cur->start) {
1515 return (address < cur->end);
1518 * Since the map is only locked for read access, perform a
1519 * standard binary search tree lookup for "address".
1521 lbound = ubound = header;
1523 if (address < cur->start) {
1528 } else if (cur->end <= address) {
1545 * Inserts the given whole VM object into the target
1546 * map at the specified address range. The object's
1547 * size should match that of the address range.
1549 * Requires that the map be locked, and leaves it so.
1551 * If object is non-NULL, ref count must be bumped by caller
1552 * prior to making call to account for the new entry.
1555 vm_map_insert(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1556 vm_offset_t start, vm_offset_t end, vm_prot_t prot, vm_prot_t max, int cow)
1558 vm_map_entry_t new_entry, next_entry, prev_entry;
1560 vm_eflags_t protoeflags;
1561 vm_inherit_t inheritance;
1563 VM_MAP_ASSERT_LOCKED(map);
1564 KASSERT(object != kernel_object ||
1565 (cow & MAP_COPY_ON_WRITE) == 0,
1566 ("vm_map_insert: kernel object and COW"));
1567 KASSERT(object == NULL || (cow & MAP_NOFAULT) == 0,
1568 ("vm_map_insert: paradoxical MAP_NOFAULT request"));
1569 KASSERT((prot & ~max) == 0,
1570 ("prot %#x is not subset of max_prot %#x", prot, max));
1573 * Check that the start and end points are not bogus.
1575 if (start < vm_map_min(map) || end > vm_map_max(map) ||
1577 return (KERN_INVALID_ADDRESS);
1580 * Find the entry prior to the proposed starting address; if it's part
1581 * of an existing entry, this range is bogus.
1583 if (vm_map_lookup_entry(map, start, &prev_entry))
1584 return (KERN_NO_SPACE);
1587 * Assert that the next entry doesn't overlap the end point.
1589 next_entry = vm_map_entry_succ(prev_entry);
1590 if (next_entry->start < end)
1591 return (KERN_NO_SPACE);
1593 if ((cow & MAP_CREATE_GUARD) != 0 && (object != NULL ||
1594 max != VM_PROT_NONE))
1595 return (KERN_INVALID_ARGUMENT);
1598 if (cow & MAP_COPY_ON_WRITE)
1599 protoeflags |= MAP_ENTRY_COW | MAP_ENTRY_NEEDS_COPY;
1600 if (cow & MAP_NOFAULT)
1601 protoeflags |= MAP_ENTRY_NOFAULT;
1602 if (cow & MAP_DISABLE_SYNCER)
1603 protoeflags |= MAP_ENTRY_NOSYNC;
1604 if (cow & MAP_DISABLE_COREDUMP)
1605 protoeflags |= MAP_ENTRY_NOCOREDUMP;
1606 if (cow & MAP_STACK_GROWS_DOWN)
1607 protoeflags |= MAP_ENTRY_GROWS_DOWN;
1608 if (cow & MAP_STACK_GROWS_UP)
1609 protoeflags |= MAP_ENTRY_GROWS_UP;
1610 if (cow & MAP_WRITECOUNT)
1611 protoeflags |= MAP_ENTRY_WRITECNT;
1612 if (cow & MAP_VN_EXEC)
1613 protoeflags |= MAP_ENTRY_VN_EXEC;
1614 if ((cow & MAP_CREATE_GUARD) != 0)
1615 protoeflags |= MAP_ENTRY_GUARD;
1616 if ((cow & MAP_CREATE_STACK_GAP_DN) != 0)
1617 protoeflags |= MAP_ENTRY_STACK_GAP_DN;
1618 if ((cow & MAP_CREATE_STACK_GAP_UP) != 0)
1619 protoeflags |= MAP_ENTRY_STACK_GAP_UP;
1620 if (cow & MAP_INHERIT_SHARE)
1621 inheritance = VM_INHERIT_SHARE;
1623 inheritance = VM_INHERIT_DEFAULT;
1626 if ((cow & (MAP_ACC_NO_CHARGE | MAP_NOFAULT | MAP_CREATE_GUARD)) != 0)
1628 if ((cow & MAP_ACC_CHARGED) || ((prot & VM_PROT_WRITE) &&
1629 ((protoeflags & MAP_ENTRY_NEEDS_COPY) || object == NULL))) {
1630 if (!(cow & MAP_ACC_CHARGED) && !swap_reserve(end - start))
1631 return (KERN_RESOURCE_SHORTAGE);
1632 KASSERT(object == NULL ||
1633 (protoeflags & MAP_ENTRY_NEEDS_COPY) != 0 ||
1634 object->cred == NULL,
1635 ("overcommit: vm_map_insert o %p", object));
1636 cred = curthread->td_ucred;
1640 /* Expand the kernel pmap, if necessary. */
1641 if (map == kernel_map && end > kernel_vm_end)
1642 pmap_growkernel(end);
1643 if (object != NULL) {
1645 * OBJ_ONEMAPPING must be cleared unless this mapping
1646 * is trivially proven to be the only mapping for any
1647 * of the object's pages. (Object granularity
1648 * reference counting is insufficient to recognize
1649 * aliases with precision.)
1651 if ((object->flags & OBJ_ANON) != 0) {
1652 VM_OBJECT_WLOCK(object);
1653 if (object->ref_count > 1 || object->shadow_count != 0)
1654 vm_object_clear_flag(object, OBJ_ONEMAPPING);
1655 VM_OBJECT_WUNLOCK(object);
1657 } else if ((prev_entry->eflags & ~MAP_ENTRY_USER_WIRED) ==
1659 (cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP |
1660 MAP_VN_EXEC)) == 0 &&
1661 prev_entry->end == start && (prev_entry->cred == cred ||
1662 (prev_entry->object.vm_object != NULL &&
1663 prev_entry->object.vm_object->cred == cred)) &&
1664 vm_object_coalesce(prev_entry->object.vm_object,
1666 (vm_size_t)(prev_entry->end - prev_entry->start),
1667 (vm_size_t)(end - prev_entry->end), cred != NULL &&
1668 (protoeflags & MAP_ENTRY_NEEDS_COPY) == 0)) {
1670 * We were able to extend the object. Determine if we
1671 * can extend the previous map entry to include the
1672 * new range as well.
1674 if (prev_entry->inheritance == inheritance &&
1675 prev_entry->protection == prot &&
1676 prev_entry->max_protection == max &&
1677 prev_entry->wired_count == 0) {
1678 KASSERT((prev_entry->eflags & MAP_ENTRY_USER_WIRED) ==
1679 0, ("prev_entry %p has incoherent wiring",
1681 if ((prev_entry->eflags & MAP_ENTRY_GUARD) == 0)
1682 map->size += end - prev_entry->end;
1683 vm_map_entry_resize(map, prev_entry,
1684 end - prev_entry->end);
1685 vm_map_try_merge_entries(map, prev_entry, next_entry);
1686 return (KERN_SUCCESS);
1690 * If we can extend the object but cannot extend the
1691 * map entry, we have to create a new map entry. We
1692 * must bump the ref count on the extended object to
1693 * account for it. object may be NULL.
1695 object = prev_entry->object.vm_object;
1696 offset = prev_entry->offset +
1697 (prev_entry->end - prev_entry->start);
1698 vm_object_reference(object);
1699 if (cred != NULL && object != NULL && object->cred != NULL &&
1700 !(prev_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
1701 /* Object already accounts for this uid. */
1709 * Create a new entry
1711 new_entry = vm_map_entry_create(map);
1712 new_entry->start = start;
1713 new_entry->end = end;
1714 new_entry->cred = NULL;
1716 new_entry->eflags = protoeflags;
1717 new_entry->object.vm_object = object;
1718 new_entry->offset = offset;
1720 new_entry->inheritance = inheritance;
1721 new_entry->protection = prot;
1722 new_entry->max_protection = max;
1723 new_entry->wired_count = 0;
1724 new_entry->wiring_thread = NULL;
1725 new_entry->read_ahead = VM_FAULT_READ_AHEAD_INIT;
1726 new_entry->next_read = start;
1728 KASSERT(cred == NULL || !ENTRY_CHARGED(new_entry),
1729 ("overcommit: vm_map_insert leaks vm_map %p", new_entry));
1730 new_entry->cred = cred;
1733 * Insert the new entry into the list
1735 vm_map_entry_link(map, new_entry);
1736 if ((new_entry->eflags & MAP_ENTRY_GUARD) == 0)
1737 map->size += new_entry->end - new_entry->start;
1740 * Try to coalesce the new entry with both the previous and next
1741 * entries in the list. Previously, we only attempted to coalesce
1742 * with the previous entry when object is NULL. Here, we handle the
1743 * other cases, which are less common.
1745 vm_map_try_merge_entries(map, prev_entry, new_entry);
1746 vm_map_try_merge_entries(map, new_entry, next_entry);
1748 if ((cow & (MAP_PREFAULT | MAP_PREFAULT_PARTIAL)) != 0) {
1749 vm_map_pmap_enter(map, start, prot, object, OFF_TO_IDX(offset),
1750 end - start, cow & MAP_PREFAULT_PARTIAL);
1753 return (KERN_SUCCESS);
1759 * Find the first fit (lowest VM address) for "length" free bytes
1760 * beginning at address >= start in the given map.
1762 * In a vm_map_entry, "max_free" is the maximum amount of
1763 * contiguous free space between an entry in its subtree and a
1764 * neighbor of that entry. This allows finding a free region in
1765 * one path down the tree, so O(log n) amortized with splay
1768 * The map must be locked, and leaves it so.
1770 * Returns: starting address if sufficient space,
1771 * vm_map_max(map)-length+1 if insufficient space.
1774 vm_map_findspace(vm_map_t map, vm_offset_t start, vm_size_t length)
1776 vm_map_entry_t header, llist, rlist, root, y;
1777 vm_size_t left_length, max_free_left, max_free_right;
1778 vm_offset_t gap_end;
1781 * Request must fit within min/max VM address and must avoid
1784 start = MAX(start, vm_map_min(map));
1785 if (start >= vm_map_max(map) || length > vm_map_max(map) - start)
1786 return (vm_map_max(map) - length + 1);
1788 /* Empty tree means wide open address space. */
1789 if (map->root == NULL)
1793 * After splay_split, if start is within an entry, push it to the start
1794 * of the following gap. If rlist is at the end of the gap containing
1795 * start, save the end of that gap in gap_end to see if the gap is big
1796 * enough; otherwise set gap_end to start skip gap-checking and move
1797 * directly to a search of the right subtree.
1799 header = &map->header;
1800 root = vm_map_splay_split(map, start, length, &llist, &rlist);
1801 gap_end = rlist->start;
1804 if (root->right != rlist)
1806 max_free_left = vm_map_splay_merge_left(header, root, llist);
1807 max_free_right = vm_map_splay_merge_right(header, root, rlist);
1808 } else if (rlist != header) {
1811 max_free_left = vm_map_splay_merge_pred(header, root, llist);
1812 max_free_right = vm_map_splay_merge_right(header, root, rlist);
1815 llist = root->right;
1816 max_free_left = vm_map_splay_merge_left(header, root, llist);
1817 max_free_right = vm_map_splay_merge_succ(header, root, rlist);
1819 root->max_free = vm_size_max(max_free_left, max_free_right);
1821 VM_MAP_ASSERT_CONSISTENT(map);
1822 if (length <= gap_end - start)
1825 /* With max_free, can immediately tell if no solution. */
1826 if (root->right == header || length > root->right->max_free)
1827 return (vm_map_max(map) - length + 1);
1830 * Splay for the least large-enough gap in the right subtree.
1832 llist = rlist = header;
1833 for (left_length = 0;;
1834 left_length = vm_map_entry_max_free_left(root, llist)) {
1835 if (length <= left_length)
1836 SPLAY_LEFT_STEP(root, y, llist, rlist,
1837 length <= vm_map_entry_max_free_left(y, llist));
1839 SPLAY_RIGHT_STEP(root, y, llist, rlist,
1840 length > vm_map_entry_max_free_left(y, root));
1845 llist = root->right;
1846 max_free_left = vm_map_splay_merge_left(header, root, llist);
1847 if (rlist == header) {
1848 root->max_free = vm_size_max(max_free_left,
1849 vm_map_splay_merge_succ(header, root, rlist));
1853 y->max_free = vm_size_max(
1854 vm_map_splay_merge_pred(root, y, root),
1855 vm_map_splay_merge_right(header, y, rlist));
1856 root->max_free = vm_size_max(max_free_left, y->max_free);
1859 VM_MAP_ASSERT_CONSISTENT(map);
1864 vm_map_fixed(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1865 vm_offset_t start, vm_size_t length, vm_prot_t prot,
1866 vm_prot_t max, int cow)
1871 end = start + length;
1872 KASSERT((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 ||
1874 ("vm_map_fixed: non-NULL backing object for stack"));
1876 VM_MAP_RANGE_CHECK(map, start, end);
1877 if ((cow & MAP_CHECK_EXCL) == 0)
1878 vm_map_delete(map, start, end);
1879 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) != 0) {
1880 result = vm_map_stack_locked(map, start, length, sgrowsiz,
1883 result = vm_map_insert(map, object, offset, start, end,
1890 static const int aslr_pages_rnd_64[2] = {0x1000, 0x10};
1891 static const int aslr_pages_rnd_32[2] = {0x100, 0x4};
1893 static int cluster_anon = 1;
1894 SYSCTL_INT(_vm, OID_AUTO, cluster_anon, CTLFLAG_RW,
1896 "Cluster anonymous mappings: 0 = no, 1 = yes if no hint, 2 = always");
1899 clustering_anon_allowed(vm_offset_t addr)
1902 switch (cluster_anon) {
1913 static long aslr_restarts;
1914 SYSCTL_LONG(_vm, OID_AUTO, aslr_restarts, CTLFLAG_RD,
1916 "Number of aslr failures");
1918 #define MAP_32BIT_MAX_ADDR ((vm_offset_t)1 << 31)
1921 * Searches for the specified amount of free space in the given map with the
1922 * specified alignment. Performs an address-ordered, first-fit search from
1923 * the given address "*addr", with an optional upper bound "max_addr". If the
1924 * parameter "alignment" is zero, then the alignment is computed from the
1925 * given (object, offset) pair so as to enable the greatest possible use of
1926 * superpage mappings. Returns KERN_SUCCESS and the address of the free space
1927 * in "*addr" if successful. Otherwise, returns KERN_NO_SPACE.
1929 * The map must be locked. Initially, there must be at least "length" bytes
1930 * of free space at the given address.
1933 vm_map_alignspace(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1934 vm_offset_t *addr, vm_size_t length, vm_offset_t max_addr,
1935 vm_offset_t alignment)
1937 vm_offset_t aligned_addr, free_addr;
1939 VM_MAP_ASSERT_LOCKED(map);
1941 KASSERT(free_addr == vm_map_findspace(map, free_addr, length),
1942 ("caller failed to provide space %#jx at address %p",
1943 (uintmax_t)length, (void *)free_addr));
1946 * At the start of every iteration, the free space at address
1947 * "*addr" is at least "length" bytes.
1950 pmap_align_superpage(object, offset, addr, length);
1951 else if ((*addr & (alignment - 1)) != 0) {
1952 *addr &= ~(alignment - 1);
1955 aligned_addr = *addr;
1956 if (aligned_addr == free_addr) {
1958 * Alignment did not change "*addr", so "*addr" must
1959 * still provide sufficient free space.
1961 return (KERN_SUCCESS);
1965 * Test for address wrap on "*addr". A wrapped "*addr" could
1966 * be a valid address, in which case vm_map_findspace() cannot
1967 * be relied upon to fail.
1969 if (aligned_addr < free_addr)
1970 return (KERN_NO_SPACE);
1971 *addr = vm_map_findspace(map, aligned_addr, length);
1972 if (*addr + length > vm_map_max(map) ||
1973 (max_addr != 0 && *addr + length > max_addr))
1974 return (KERN_NO_SPACE);
1976 if (free_addr == aligned_addr) {
1978 * If a successful call to vm_map_findspace() did not
1979 * change "*addr", then "*addr" must still be aligned
1980 * and provide sufficient free space.
1982 return (KERN_SUCCESS);
1988 * vm_map_find finds an unallocated region in the target address
1989 * map with the given length. The search is defined to be
1990 * first-fit from the specified address; the region found is
1991 * returned in the same parameter.
1993 * If object is non-NULL, ref count must be bumped by caller
1994 * prior to making call to account for the new entry.
1997 vm_map_find(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1998 vm_offset_t *addr, /* IN/OUT */
1999 vm_size_t length, vm_offset_t max_addr, int find_space,
2000 vm_prot_t prot, vm_prot_t max, int cow)
2002 vm_offset_t alignment, curr_min_addr, min_addr;
2003 int gap, pidx, rv, try;
2004 bool cluster, en_aslr, update_anon;
2006 KASSERT((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 ||
2008 ("vm_map_find: non-NULL backing object for stack"));
2009 MPASS((cow & MAP_REMAP) == 0 || (find_space == VMFS_NO_SPACE &&
2010 (cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0));
2011 if (find_space == VMFS_OPTIMAL_SPACE && (object == NULL ||
2012 (object->flags & OBJ_COLORED) == 0))
2013 find_space = VMFS_ANY_SPACE;
2014 if (find_space >> 8 != 0) {
2015 KASSERT((find_space & 0xff) == 0, ("bad VMFS flags"));
2016 alignment = (vm_offset_t)1 << (find_space >> 8);
2019 en_aslr = (map->flags & MAP_ASLR) != 0;
2020 update_anon = cluster = clustering_anon_allowed(*addr) &&
2021 (map->flags & MAP_IS_SUB_MAP) == 0 && max_addr == 0 &&
2022 find_space != VMFS_NO_SPACE && object == NULL &&
2023 (cow & (MAP_INHERIT_SHARE | MAP_STACK_GROWS_UP |
2024 MAP_STACK_GROWS_DOWN)) == 0 && prot != PROT_NONE;
2025 curr_min_addr = min_addr = *addr;
2026 if (en_aslr && min_addr == 0 && !cluster &&
2027 find_space != VMFS_NO_SPACE &&
2028 (map->flags & MAP_ASLR_IGNSTART) != 0)
2029 curr_min_addr = min_addr = vm_map_min(map);
2033 curr_min_addr = map->anon_loc;
2034 if (curr_min_addr == 0)
2037 if (find_space != VMFS_NO_SPACE) {
2038 KASSERT(find_space == VMFS_ANY_SPACE ||
2039 find_space == VMFS_OPTIMAL_SPACE ||
2040 find_space == VMFS_SUPER_SPACE ||
2041 alignment != 0, ("unexpected VMFS flag"));
2044 * When creating an anonymous mapping, try clustering
2045 * with an existing anonymous mapping first.
2047 * We make up to two attempts to find address space
2048 * for a given find_space value. The first attempt may
2049 * apply randomization or may cluster with an existing
2050 * anonymous mapping. If this first attempt fails,
2051 * perform a first-fit search of the available address
2054 * If all tries failed, and find_space is
2055 * VMFS_OPTIMAL_SPACE, fallback to VMFS_ANY_SPACE.
2056 * Again enable clustering and randomization.
2063 * Second try: we failed either to find a
2064 * suitable region for randomizing the
2065 * allocation, or to cluster with an existing
2066 * mapping. Retry with free run.
2068 curr_min_addr = (map->flags & MAP_ASLR_IGNSTART) != 0 ?
2069 vm_map_min(map) : min_addr;
2070 atomic_add_long(&aslr_restarts, 1);
2073 if (try == 1 && en_aslr && !cluster) {
2075 * Find space for allocation, including
2076 * gap needed for later randomization.
2078 pidx = MAXPAGESIZES > 1 && pagesizes[1] != 0 &&
2079 (find_space == VMFS_SUPER_SPACE || find_space ==
2080 VMFS_OPTIMAL_SPACE) ? 1 : 0;
2081 gap = vm_map_max(map) > MAP_32BIT_MAX_ADDR &&
2082 (max_addr == 0 || max_addr > MAP_32BIT_MAX_ADDR) ?
2083 aslr_pages_rnd_64[pidx] : aslr_pages_rnd_32[pidx];
2084 *addr = vm_map_findspace(map, curr_min_addr,
2085 length + gap * pagesizes[pidx]);
2086 if (*addr + length + gap * pagesizes[pidx] >
2089 /* And randomize the start address. */
2090 *addr += (arc4random() % gap) * pagesizes[pidx];
2091 if (max_addr != 0 && *addr + length > max_addr)
2094 *addr = vm_map_findspace(map, curr_min_addr, length);
2095 if (*addr + length > vm_map_max(map) ||
2096 (max_addr != 0 && *addr + length > max_addr)) {
2107 if (find_space != VMFS_ANY_SPACE &&
2108 (rv = vm_map_alignspace(map, object, offset, addr, length,
2109 max_addr, alignment)) != KERN_SUCCESS) {
2110 if (find_space == VMFS_OPTIMAL_SPACE) {
2111 find_space = VMFS_ANY_SPACE;
2112 curr_min_addr = min_addr;
2113 cluster = update_anon;
2119 } else if ((cow & MAP_REMAP) != 0) {
2120 if (*addr < vm_map_min(map) ||
2121 *addr + length > vm_map_max(map) ||
2122 *addr + length <= length) {
2123 rv = KERN_INVALID_ADDRESS;
2126 vm_map_delete(map, *addr, *addr + length);
2128 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) != 0) {
2129 rv = vm_map_stack_locked(map, *addr, length, sgrowsiz, prot,
2132 rv = vm_map_insert(map, object, offset, *addr, *addr + length,
2135 if (rv == KERN_SUCCESS && update_anon)
2136 map->anon_loc = *addr + length;
2143 * vm_map_find_min() is a variant of vm_map_find() that takes an
2144 * additional parameter (min_addr) and treats the given address
2145 * (*addr) differently. Specifically, it treats *addr as a hint
2146 * and not as the minimum address where the mapping is created.
2148 * This function works in two phases. First, it tries to
2149 * allocate above the hint. If that fails and the hint is
2150 * greater than min_addr, it performs a second pass, replacing
2151 * the hint with min_addr as the minimum address for the
2155 vm_map_find_min(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
2156 vm_offset_t *addr, vm_size_t length, vm_offset_t min_addr,
2157 vm_offset_t max_addr, int find_space, vm_prot_t prot, vm_prot_t max,
2165 rv = vm_map_find(map, object, offset, addr, length, max_addr,
2166 find_space, prot, max, cow);
2167 if (rv == KERN_SUCCESS || min_addr >= hint)
2169 *addr = hint = min_addr;
2174 * A map entry with any of the following flags set must not be merged with
2177 #define MAP_ENTRY_NOMERGE_MASK (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP | \
2178 MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_IS_SUB_MAP | MAP_ENTRY_VN_EXEC)
2181 vm_map_mergeable_neighbors(vm_map_entry_t prev, vm_map_entry_t entry)
2184 KASSERT((prev->eflags & MAP_ENTRY_NOMERGE_MASK) == 0 ||
2185 (entry->eflags & MAP_ENTRY_NOMERGE_MASK) == 0,
2186 ("vm_map_mergeable_neighbors: neither %p nor %p are mergeable",
2188 return (prev->end == entry->start &&
2189 prev->object.vm_object == entry->object.vm_object &&
2190 (prev->object.vm_object == NULL ||
2191 prev->offset + (prev->end - prev->start) == entry->offset) &&
2192 prev->eflags == entry->eflags &&
2193 prev->protection == entry->protection &&
2194 prev->max_protection == entry->max_protection &&
2195 prev->inheritance == entry->inheritance &&
2196 prev->wired_count == entry->wired_count &&
2197 prev->cred == entry->cred);
2201 vm_map_merged_neighbor_dispose(vm_map_t map, vm_map_entry_t entry)
2205 * If the backing object is a vnode object, vm_object_deallocate()
2206 * calls vrele(). However, vrele() does not lock the vnode because
2207 * the vnode has additional references. Thus, the map lock can be
2208 * kept without causing a lock-order reversal with the vnode lock.
2210 * Since we count the number of virtual page mappings in
2211 * object->un_pager.vnp.writemappings, the writemappings value
2212 * should not be adjusted when the entry is disposed of.
2214 if (entry->object.vm_object != NULL)
2215 vm_object_deallocate(entry->object.vm_object);
2216 if (entry->cred != NULL)
2217 crfree(entry->cred);
2218 vm_map_entry_dispose(map, entry);
2222 * vm_map_try_merge_entries:
2224 * Compare the given map entry to its predecessor, and merge its precessor
2225 * into it if possible. The entry remains valid, and may be extended.
2226 * The predecessor may be deleted.
2228 * The map must be locked.
2231 vm_map_try_merge_entries(vm_map_t map, vm_map_entry_t prev_entry,
2232 vm_map_entry_t entry)
2235 VM_MAP_ASSERT_LOCKED(map);
2236 if ((entry->eflags & MAP_ENTRY_NOMERGE_MASK) == 0 &&
2237 vm_map_mergeable_neighbors(prev_entry, entry)) {
2238 vm_map_entry_unlink(map, prev_entry, UNLINK_MERGE_NEXT);
2239 vm_map_merged_neighbor_dispose(map, prev_entry);
2244 * vm_map_entry_back:
2246 * Allocate an object to back a map entry.
2249 vm_map_entry_back(vm_map_entry_t entry)
2253 KASSERT(entry->object.vm_object == NULL,
2254 ("map entry %p has backing object", entry));
2255 KASSERT((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0,
2256 ("map entry %p is a submap", entry));
2257 object = vm_object_allocate_anon(atop(entry->end - entry->start), NULL,
2258 entry->cred, entry->end - entry->start);
2259 entry->object.vm_object = object;
2265 * vm_map_entry_charge_object
2267 * If there is no object backing this entry, create one. Otherwise, if
2268 * the entry has cred, give it to the backing object.
2271 vm_map_entry_charge_object(vm_map_t map, vm_map_entry_t entry)
2274 VM_MAP_ASSERT_LOCKED(map);
2275 KASSERT((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0,
2276 ("map entry %p is a submap", entry));
2277 if (entry->object.vm_object == NULL && !map->system_map &&
2278 (entry->eflags & MAP_ENTRY_GUARD) == 0)
2279 vm_map_entry_back(entry);
2280 else if (entry->object.vm_object != NULL &&
2281 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) &&
2282 entry->cred != NULL) {
2283 VM_OBJECT_WLOCK(entry->object.vm_object);
2284 KASSERT(entry->object.vm_object->cred == NULL,
2285 ("OVERCOMMIT: %s: both cred e %p", __func__, entry));
2286 entry->object.vm_object->cred = entry->cred;
2287 entry->object.vm_object->charge = entry->end - entry->start;
2288 VM_OBJECT_WUNLOCK(entry->object.vm_object);
2294 * vm_map_entry_clone
2296 * Create a duplicate map entry for clipping.
2298 static vm_map_entry_t
2299 vm_map_entry_clone(vm_map_t map, vm_map_entry_t entry)
2301 vm_map_entry_t new_entry;
2303 VM_MAP_ASSERT_LOCKED(map);
2306 * Create a backing object now, if none exists, so that more individual
2307 * objects won't be created after the map entry is split.
2309 vm_map_entry_charge_object(map, entry);
2311 /* Clone the entry. */
2312 new_entry = vm_map_entry_create(map);
2313 *new_entry = *entry;
2314 if (new_entry->cred != NULL)
2315 crhold(entry->cred);
2316 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
2317 vm_object_reference(new_entry->object.vm_object);
2318 vm_map_entry_set_vnode_text(new_entry, true);
2320 * The object->un_pager.vnp.writemappings for the object of
2321 * MAP_ENTRY_WRITECNT type entry shall be kept as is here. The
2322 * virtual pages are re-distributed among the clipped entries,
2323 * so the sum is left the same.
2330 * vm_map_clip_start: [ internal use only ]
2332 * Asserts that the given entry begins at or after
2333 * the specified address; if necessary,
2334 * it splits the entry into two.
2336 #define vm_map_clip_start(map, entry, startaddr) \
2338 if (startaddr > entry->start) \
2339 _vm_map_clip_start(map, entry, startaddr); \
2343 * This routine is called only when it is known that
2344 * the entry must be split.
2347 _vm_map_clip_start(vm_map_t map, vm_map_entry_t entry, vm_offset_t start)
2349 vm_map_entry_t new_entry;
2351 VM_MAP_ASSERT_LOCKED(map);
2352 KASSERT(entry->end > start && entry->start < start,
2353 ("_vm_map_clip_start: invalid clip of entry %p", entry));
2355 new_entry = vm_map_entry_clone(map, entry);
2358 * Split off the front portion. Insert the new entry BEFORE this one,
2359 * so that this entry has the specified starting address.
2361 new_entry->end = start;
2362 entry->offset += (start - entry->start);
2363 entry->start = start;
2364 vm_map_entry_link(map, new_entry);
2368 * vm_map_clip_end: [ internal use only ]
2370 * Asserts that the given entry ends at or before
2371 * the specified address; if necessary,
2372 * it splits the entry into two.
2374 #define vm_map_clip_end(map, entry, endaddr) \
2376 if ((endaddr) < (entry->end)) \
2377 _vm_map_clip_end((map), (entry), (endaddr)); \
2381 * This routine is called only when it is known that
2382 * the entry must be split.
2385 _vm_map_clip_end(vm_map_t map, vm_map_entry_t entry, vm_offset_t end)
2387 vm_map_entry_t new_entry;
2389 VM_MAP_ASSERT_LOCKED(map);
2390 KASSERT(entry->start < end && entry->end > end,
2391 ("_vm_map_clip_end: invalid clip of entry %p", entry));
2393 new_entry = vm_map_entry_clone(map, entry);
2396 * Split off the back portion. Insert the new entry AFTER this one,
2397 * so that this entry has the specified ending address.
2399 new_entry->start = entry->end = end;
2400 new_entry->offset += (end - entry->start);
2401 vm_map_entry_link(map, new_entry);
2405 * vm_map_submap: [ kernel use only ]
2407 * Mark the given range as handled by a subordinate map.
2409 * This range must have been created with vm_map_find,
2410 * and no other operations may have been performed on this
2411 * range prior to calling vm_map_submap.
2413 * Only a limited number of operations can be performed
2414 * within this rage after calling vm_map_submap:
2416 * [Don't try vm_map_copy!]
2418 * To remove a submapping, one must first remove the
2419 * range from the superior map, and then destroy the
2420 * submap (if desired). [Better yet, don't try it.]
2429 vm_map_entry_t entry;
2432 result = KERN_INVALID_ARGUMENT;
2434 vm_map_lock(submap);
2435 submap->flags |= MAP_IS_SUB_MAP;
2436 vm_map_unlock(submap);
2440 VM_MAP_RANGE_CHECK(map, start, end);
2442 if (vm_map_lookup_entry(map, start, &entry)) {
2443 vm_map_clip_start(map, entry, start);
2445 entry = vm_map_entry_succ(entry);
2447 vm_map_clip_end(map, entry, end);
2449 if ((entry->start == start) && (entry->end == end) &&
2450 ((entry->eflags & MAP_ENTRY_COW) == 0) &&
2451 (entry->object.vm_object == NULL)) {
2452 entry->object.sub_map = submap;
2453 entry->eflags |= MAP_ENTRY_IS_SUB_MAP;
2454 result = KERN_SUCCESS;
2458 if (result != KERN_SUCCESS) {
2459 vm_map_lock(submap);
2460 submap->flags &= ~MAP_IS_SUB_MAP;
2461 vm_map_unlock(submap);
2467 * The maximum number of pages to map if MAP_PREFAULT_PARTIAL is specified
2469 #define MAX_INIT_PT 96
2472 * vm_map_pmap_enter:
2474 * Preload the specified map's pmap with mappings to the specified
2475 * object's memory-resident pages. No further physical pages are
2476 * allocated, and no further virtual pages are retrieved from secondary
2477 * storage. If the specified flags include MAP_PREFAULT_PARTIAL, then a
2478 * limited number of page mappings are created at the low-end of the
2479 * specified address range. (For this purpose, a superpage mapping
2480 * counts as one page mapping.) Otherwise, all resident pages within
2481 * the specified address range are mapped.
2484 vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot,
2485 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags)
2488 vm_page_t p, p_start;
2489 vm_pindex_t mask, psize, threshold, tmpidx;
2491 if ((prot & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0 || object == NULL)
2493 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) {
2494 VM_OBJECT_WLOCK(object);
2495 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) {
2496 pmap_object_init_pt(map->pmap, addr, object, pindex,
2498 VM_OBJECT_WUNLOCK(object);
2501 VM_OBJECT_LOCK_DOWNGRADE(object);
2503 VM_OBJECT_RLOCK(object);
2506 if (psize + pindex > object->size) {
2507 if (pindex >= object->size) {
2508 VM_OBJECT_RUNLOCK(object);
2511 psize = object->size - pindex;
2516 threshold = MAX_INIT_PT;
2518 p = vm_page_find_least(object, pindex);
2520 * Assert: the variable p is either (1) the page with the
2521 * least pindex greater than or equal to the parameter pindex
2525 p != NULL && (tmpidx = p->pindex - pindex) < psize;
2526 p = TAILQ_NEXT(p, listq)) {
2528 * don't allow an madvise to blow away our really
2529 * free pages allocating pv entries.
2531 if (((flags & MAP_PREFAULT_MADVISE) != 0 &&
2532 vm_page_count_severe()) ||
2533 ((flags & MAP_PREFAULT_PARTIAL) != 0 &&
2534 tmpidx >= threshold)) {
2538 if (vm_page_all_valid(p)) {
2539 if (p_start == NULL) {
2540 start = addr + ptoa(tmpidx);
2543 /* Jump ahead if a superpage mapping is possible. */
2544 if (p->psind > 0 && ((addr + ptoa(tmpidx)) &
2545 (pagesizes[p->psind] - 1)) == 0) {
2546 mask = atop(pagesizes[p->psind]) - 1;
2547 if (tmpidx + mask < psize &&
2548 vm_page_ps_test(p, PS_ALL_VALID, NULL)) {
2553 } else if (p_start != NULL) {
2554 pmap_enter_object(map->pmap, start, addr +
2555 ptoa(tmpidx), p_start, prot);
2559 if (p_start != NULL)
2560 pmap_enter_object(map->pmap, start, addr + ptoa(psize),
2562 VM_OBJECT_RUNLOCK(object);
2568 * Sets the protection of the specified address
2569 * region in the target map. If "set_max" is
2570 * specified, the maximum protection is to be set;
2571 * otherwise, only the current protection is affected.
2574 vm_map_protect(vm_map_t map, vm_offset_t start, vm_offset_t end,
2575 vm_prot_t new_prot, boolean_t set_max)
2577 vm_map_entry_t entry, first_entry, in_tran, prev_entry;
2584 return (KERN_SUCCESS);
2591 * Ensure that we are not concurrently wiring pages. vm_map_wire() may
2592 * need to fault pages into the map and will drop the map lock while
2593 * doing so, and the VM object may end up in an inconsistent state if we
2594 * update the protection on the map entry in between faults.
2596 vm_map_wait_busy(map);
2598 VM_MAP_RANGE_CHECK(map, start, end);
2600 if (!vm_map_lookup_entry(map, start, &first_entry))
2601 first_entry = vm_map_entry_succ(first_entry);
2604 * Make a first pass to check for protection violations.
2606 for (entry = first_entry; entry->start < end;
2607 entry = vm_map_entry_succ(entry)) {
2608 if ((entry->eflags & MAP_ENTRY_GUARD) != 0)
2610 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0) {
2612 return (KERN_INVALID_ARGUMENT);
2614 if ((new_prot & entry->max_protection) != new_prot) {
2616 return (KERN_PROTECTION_FAILURE);
2618 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0)
2623 * Postpone the operation until all in-transition map entries have
2624 * stabilized. An in-transition entry might already have its pages
2625 * wired and wired_count incremented, but not yet have its
2626 * MAP_ENTRY_USER_WIRED flag set. In which case, we would fail to call
2627 * vm_fault_copy_entry() in the final loop below.
2629 if (in_tran != NULL) {
2630 in_tran->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2631 vm_map_unlock_and_wait(map, 0);
2636 * Before changing the protections, try to reserve swap space for any
2637 * private (i.e., copy-on-write) mappings that are transitioning from
2638 * read-only to read/write access. If a reservation fails, break out
2639 * of this loop early and let the next loop simplify the entries, since
2640 * some may now be mergeable.
2643 vm_map_clip_start(map, first_entry, start);
2644 for (entry = first_entry; entry->start < end;
2645 entry = vm_map_entry_succ(entry)) {
2646 vm_map_clip_end(map, entry, end);
2649 ((new_prot & ~entry->protection) & VM_PROT_WRITE) == 0 ||
2650 ENTRY_CHARGED(entry) ||
2651 (entry->eflags & MAP_ENTRY_GUARD) != 0) {
2655 cred = curthread->td_ucred;
2656 obj = entry->object.vm_object;
2659 (entry->eflags & MAP_ENTRY_NEEDS_COPY) != 0) {
2660 if (!swap_reserve(entry->end - entry->start)) {
2661 rv = KERN_RESOURCE_SHORTAGE;
2670 if (obj->type != OBJT_DEFAULT && obj->type != OBJT_SWAP)
2672 VM_OBJECT_WLOCK(obj);
2673 if (obj->type != OBJT_DEFAULT && obj->type != OBJT_SWAP) {
2674 VM_OBJECT_WUNLOCK(obj);
2679 * Charge for the whole object allocation now, since
2680 * we cannot distinguish between non-charged and
2681 * charged clipped mapping of the same object later.
2683 KASSERT(obj->charge == 0,
2684 ("vm_map_protect: object %p overcharged (entry %p)",
2686 if (!swap_reserve(ptoa(obj->size))) {
2687 VM_OBJECT_WUNLOCK(obj);
2688 rv = KERN_RESOURCE_SHORTAGE;
2695 obj->charge = ptoa(obj->size);
2696 VM_OBJECT_WUNLOCK(obj);
2700 * If enough swap space was available, go back and fix up protections.
2701 * Otherwise, just simplify entries, since some may have been modified.
2702 * [Note that clipping is not necessary the second time.]
2704 for (prev_entry = vm_map_entry_pred(first_entry), entry = first_entry;
2706 vm_map_try_merge_entries(map, prev_entry, entry),
2707 prev_entry = entry, entry = vm_map_entry_succ(entry)) {
2708 if (rv != KERN_SUCCESS ||
2709 (entry->eflags & MAP_ENTRY_GUARD) != 0)
2712 old_prot = entry->protection;
2716 (entry->max_protection = new_prot) &
2719 entry->protection = new_prot;
2722 * For user wired map entries, the normal lazy evaluation of
2723 * write access upgrades through soft page faults is
2724 * undesirable. Instead, immediately copy any pages that are
2725 * copy-on-write and enable write access in the physical map.
2727 if ((entry->eflags & MAP_ENTRY_USER_WIRED) != 0 &&
2728 (entry->protection & VM_PROT_WRITE) != 0 &&
2729 (old_prot & VM_PROT_WRITE) == 0)
2730 vm_fault_copy_entry(map, map, entry, entry, NULL);
2733 * When restricting access, update the physical map. Worry
2734 * about copy-on-write here.
2736 if ((old_prot & ~entry->protection) != 0) {
2737 #define MASK(entry) (((entry)->eflags & MAP_ENTRY_COW) ? ~VM_PROT_WRITE : \
2739 pmap_protect(map->pmap, entry->start,
2741 entry->protection & MASK(entry));
2745 vm_map_try_merge_entries(map, prev_entry, entry);
2753 * This routine traverses a processes map handling the madvise
2754 * system call. Advisories are classified as either those effecting
2755 * the vm_map_entry structure, or those effecting the underlying
2765 vm_map_entry_t entry, prev_entry;
2769 * Some madvise calls directly modify the vm_map_entry, in which case
2770 * we need to use an exclusive lock on the map and we need to perform
2771 * various clipping operations. Otherwise we only need a read-lock
2776 case MADV_SEQUENTIAL:
2793 vm_map_lock_read(map);
2800 * Locate starting entry and clip if necessary.
2802 VM_MAP_RANGE_CHECK(map, start, end);
2804 if (vm_map_lookup_entry(map, start, &entry)) {
2806 vm_map_clip_start(map, entry, start);
2807 prev_entry = vm_map_entry_pred(entry);
2810 entry = vm_map_entry_succ(entry);
2815 * madvise behaviors that are implemented in the vm_map_entry.
2817 * We clip the vm_map_entry so that behavioral changes are
2818 * limited to the specified address range.
2820 for (; entry->start < end;
2821 prev_entry = entry, entry = vm_map_entry_succ(entry)) {
2822 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0)
2825 vm_map_clip_end(map, entry, end);
2829 vm_map_entry_set_behavior(entry,
2830 MAP_ENTRY_BEHAV_NORMAL);
2832 case MADV_SEQUENTIAL:
2833 vm_map_entry_set_behavior(entry,
2834 MAP_ENTRY_BEHAV_SEQUENTIAL);
2837 vm_map_entry_set_behavior(entry,
2838 MAP_ENTRY_BEHAV_RANDOM);
2841 entry->eflags |= MAP_ENTRY_NOSYNC;
2844 entry->eflags &= ~MAP_ENTRY_NOSYNC;
2847 entry->eflags |= MAP_ENTRY_NOCOREDUMP;
2850 entry->eflags &= ~MAP_ENTRY_NOCOREDUMP;
2855 vm_map_try_merge_entries(map, prev_entry, entry);
2857 vm_map_try_merge_entries(map, prev_entry, entry);
2860 vm_pindex_t pstart, pend;
2863 * madvise behaviors that are implemented in the underlying
2866 * Since we don't clip the vm_map_entry, we have to clip
2867 * the vm_object pindex and count.
2869 for (; entry->start < end;
2870 entry = vm_map_entry_succ(entry)) {
2871 vm_offset_t useEnd, useStart;
2873 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0)
2877 * MADV_FREE would otherwise rewind time to
2878 * the creation of the shadow object. Because
2879 * we hold the VM map read-locked, neither the
2880 * entry's object nor the presence of a
2881 * backing object can change.
2883 if (behav == MADV_FREE &&
2884 entry->object.vm_object != NULL &&
2885 entry->object.vm_object->backing_object != NULL)
2888 pstart = OFF_TO_IDX(entry->offset);
2889 pend = pstart + atop(entry->end - entry->start);
2890 useStart = entry->start;
2891 useEnd = entry->end;
2893 if (entry->start < start) {
2894 pstart += atop(start - entry->start);
2897 if (entry->end > end) {
2898 pend -= atop(entry->end - end);
2906 * Perform the pmap_advise() before clearing
2907 * PGA_REFERENCED in vm_page_advise(). Otherwise, a
2908 * concurrent pmap operation, such as pmap_remove(),
2909 * could clear a reference in the pmap and set
2910 * PGA_REFERENCED on the page before the pmap_advise()
2911 * had completed. Consequently, the page would appear
2912 * referenced based upon an old reference that
2913 * occurred before this pmap_advise() ran.
2915 if (behav == MADV_DONTNEED || behav == MADV_FREE)
2916 pmap_advise(map->pmap, useStart, useEnd,
2919 vm_object_madvise(entry->object.vm_object, pstart,
2923 * Pre-populate paging structures in the
2924 * WILLNEED case. For wired entries, the
2925 * paging structures are already populated.
2927 if (behav == MADV_WILLNEED &&
2928 entry->wired_count == 0) {
2929 vm_map_pmap_enter(map,
2932 entry->object.vm_object,
2934 ptoa(pend - pstart),
2935 MAP_PREFAULT_MADVISE
2939 vm_map_unlock_read(map);
2948 * Sets the inheritance of the specified address
2949 * range in the target map. Inheritance
2950 * affects how the map will be shared with
2951 * child maps at the time of vmspace_fork.
2954 vm_map_inherit(vm_map_t map, vm_offset_t start, vm_offset_t end,
2955 vm_inherit_t new_inheritance)
2957 vm_map_entry_t entry, prev_entry;
2959 switch (new_inheritance) {
2960 case VM_INHERIT_NONE:
2961 case VM_INHERIT_COPY:
2962 case VM_INHERIT_SHARE:
2963 case VM_INHERIT_ZERO:
2966 return (KERN_INVALID_ARGUMENT);
2969 return (KERN_SUCCESS);
2971 VM_MAP_RANGE_CHECK(map, start, end);
2972 if (vm_map_lookup_entry(map, start, &prev_entry)) {
2974 vm_map_clip_start(map, entry, start);
2975 prev_entry = vm_map_entry_pred(entry);
2977 entry = vm_map_entry_succ(prev_entry);
2978 for (; entry->start < end;
2979 prev_entry = entry, entry = vm_map_entry_succ(entry)) {
2980 vm_map_clip_end(map, entry, end);
2981 if ((entry->eflags & MAP_ENTRY_GUARD) == 0 ||
2982 new_inheritance != VM_INHERIT_ZERO)
2983 entry->inheritance = new_inheritance;
2984 vm_map_try_merge_entries(map, prev_entry, entry);
2986 vm_map_try_merge_entries(map, prev_entry, entry);
2988 return (KERN_SUCCESS);
2992 * vm_map_entry_in_transition:
2994 * Release the map lock, and sleep until the entry is no longer in
2995 * transition. Awake and acquire the map lock. If the map changed while
2996 * another held the lock, lookup a possibly-changed entry at or after the
2997 * 'start' position of the old entry.
2999 static vm_map_entry_t
3000 vm_map_entry_in_transition(vm_map_t map, vm_offset_t in_start,
3001 vm_offset_t *io_end, bool holes_ok, vm_map_entry_t in_entry)
3003 vm_map_entry_t entry;
3005 u_int last_timestamp;
3007 VM_MAP_ASSERT_LOCKED(map);
3008 KASSERT((in_entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0,
3009 ("not in-tranition map entry %p", in_entry));
3011 * We have not yet clipped the entry.
3013 start = MAX(in_start, in_entry->start);
3014 in_entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
3015 last_timestamp = map->timestamp;
3016 if (vm_map_unlock_and_wait(map, 0)) {
3018 * Allow interruption of user wiring/unwiring?
3022 if (last_timestamp + 1 == map->timestamp)
3026 * Look again for the entry because the map was modified while it was
3027 * unlocked. Specifically, the entry may have been clipped, merged, or
3030 if (!vm_map_lookup_entry(map, start, &entry)) {
3035 entry = vm_map_entry_succ(entry);
3043 * Implements both kernel and user unwiring.
3046 vm_map_unwire(vm_map_t map, vm_offset_t start, vm_offset_t end,
3049 vm_map_entry_t entry, first_entry, next_entry, prev_entry;
3051 bool holes_ok, need_wakeup, user_unwire;
3054 return (KERN_SUCCESS);
3055 holes_ok = (flags & VM_MAP_WIRE_HOLESOK) != 0;
3056 user_unwire = (flags & VM_MAP_WIRE_USER) != 0;
3058 VM_MAP_RANGE_CHECK(map, start, end);
3059 if (!vm_map_lookup_entry(map, start, &first_entry)) {
3061 first_entry = vm_map_entry_succ(first_entry);
3064 return (KERN_INVALID_ADDRESS);
3068 for (entry = first_entry; entry->start < end; entry = next_entry) {
3069 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
3071 * We have not yet clipped the entry.
3073 next_entry = vm_map_entry_in_transition(map, start,
3074 &end, holes_ok, entry);
3075 if (next_entry == NULL) {
3076 if (entry == first_entry) {
3078 return (KERN_INVALID_ADDRESS);
3080 rv = KERN_INVALID_ADDRESS;
3083 first_entry = (entry == first_entry) ?
3087 vm_map_clip_start(map, entry, start);
3088 vm_map_clip_end(map, entry, end);
3090 * Mark the entry in case the map lock is released. (See
3093 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 &&
3094 entry->wiring_thread == NULL,
3095 ("owned map entry %p", entry));
3096 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
3097 entry->wiring_thread = curthread;
3098 next_entry = vm_map_entry_succ(entry);
3100 * Check the map for holes in the specified region.
3101 * If holes_ok, skip this check.
3104 entry->end < end && next_entry->start > entry->end) {
3106 rv = KERN_INVALID_ADDRESS;
3110 * If system unwiring, require that the entry is system wired.
3113 vm_map_entry_system_wired_count(entry) == 0) {
3115 rv = KERN_INVALID_ARGUMENT;
3119 need_wakeup = false;
3120 if (first_entry == NULL &&
3121 !vm_map_lookup_entry(map, start, &first_entry)) {
3122 KASSERT(holes_ok, ("vm_map_unwire: lookup failed"));
3123 prev_entry = first_entry;
3124 entry = vm_map_entry_succ(first_entry);
3126 prev_entry = vm_map_entry_pred(first_entry);
3127 entry = first_entry;
3129 for (; entry->start < end;
3130 prev_entry = entry, entry = vm_map_entry_succ(entry)) {
3132 * If holes_ok was specified, an empty
3133 * space in the unwired region could have been mapped
3134 * while the map lock was dropped for draining
3135 * MAP_ENTRY_IN_TRANSITION. Moreover, another thread
3136 * could be simultaneously wiring this new mapping
3137 * entry. Detect these cases and skip any entries
3138 * marked as in transition by us.
3140 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 ||
3141 entry->wiring_thread != curthread) {
3143 ("vm_map_unwire: !HOLESOK and new/changed entry"));
3147 if (rv == KERN_SUCCESS && (!user_unwire ||
3148 (entry->eflags & MAP_ENTRY_USER_WIRED))) {
3149 if (entry->wired_count == 1)
3150 vm_map_entry_unwire(map, entry);
3152 entry->wired_count--;
3154 entry->eflags &= ~MAP_ENTRY_USER_WIRED;
3156 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0,
3157 ("vm_map_unwire: in-transition flag missing %p", entry));
3158 KASSERT(entry->wiring_thread == curthread,
3159 ("vm_map_unwire: alien wire %p", entry));
3160 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION;
3161 entry->wiring_thread = NULL;
3162 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
3163 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
3166 vm_map_try_merge_entries(map, prev_entry, entry);
3168 vm_map_try_merge_entries(map, prev_entry, entry);
3176 vm_map_wire_user_count_sub(u_long npages)
3179 atomic_subtract_long(&vm_user_wire_count, npages);
3183 vm_map_wire_user_count_add(u_long npages)
3187 wired = vm_user_wire_count;
3189 if (npages + wired > vm_page_max_user_wired)
3191 } while (!atomic_fcmpset_long(&vm_user_wire_count, &wired,
3198 * vm_map_wire_entry_failure:
3200 * Handle a wiring failure on the given entry.
3202 * The map should be locked.
3205 vm_map_wire_entry_failure(vm_map_t map, vm_map_entry_t entry,
3206 vm_offset_t failed_addr)
3209 VM_MAP_ASSERT_LOCKED(map);
3210 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 &&
3211 entry->wired_count == 1,
3212 ("vm_map_wire_entry_failure: entry %p isn't being wired", entry));
3213 KASSERT(failed_addr < entry->end,
3214 ("vm_map_wire_entry_failure: entry %p was fully wired", entry));
3217 * If any pages at the start of this entry were successfully wired,
3220 if (failed_addr > entry->start) {
3221 pmap_unwire(map->pmap, entry->start, failed_addr);
3222 vm_object_unwire(entry->object.vm_object, entry->offset,
3223 failed_addr - entry->start, PQ_ACTIVE);
3227 * Assign an out-of-range value to represent the failure to wire this
3230 entry->wired_count = -1;
3234 vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t end, int flags)
3239 rv = vm_map_wire_locked(map, start, end, flags);
3246 * vm_map_wire_locked:
3248 * Implements both kernel and user wiring. Returns with the map locked,
3249 * the map lock may be dropped.
3252 vm_map_wire_locked(vm_map_t map, vm_offset_t start, vm_offset_t end, int flags)
3254 vm_map_entry_t entry, first_entry, next_entry, prev_entry;
3255 vm_offset_t faddr, saved_end, saved_start;
3257 u_int last_timestamp;
3259 bool holes_ok, need_wakeup, user_wire;
3262 VM_MAP_ASSERT_LOCKED(map);
3265 return (KERN_SUCCESS);
3267 if (flags & VM_MAP_WIRE_WRITE)
3268 prot |= VM_PROT_WRITE;
3269 holes_ok = (flags & VM_MAP_WIRE_HOLESOK) != 0;
3270 user_wire = (flags & VM_MAP_WIRE_USER) != 0;
3271 VM_MAP_RANGE_CHECK(map, start, end);
3272 if (!vm_map_lookup_entry(map, start, &first_entry)) {
3274 first_entry = vm_map_entry_succ(first_entry);
3276 return (KERN_INVALID_ADDRESS);
3278 for (entry = first_entry; entry->start < end; entry = next_entry) {
3279 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
3281 * We have not yet clipped the entry.
3283 next_entry = vm_map_entry_in_transition(map, start,
3284 &end, holes_ok, entry);
3285 if (next_entry == NULL) {
3286 if (entry == first_entry)
3287 return (KERN_INVALID_ADDRESS);
3288 rv = KERN_INVALID_ADDRESS;
3291 first_entry = (entry == first_entry) ?
3295 vm_map_clip_start(map, entry, start);
3296 vm_map_clip_end(map, entry, end);
3298 * Mark the entry in case the map lock is released. (See
3301 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 &&
3302 entry->wiring_thread == NULL,
3303 ("owned map entry %p", entry));
3304 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
3305 entry->wiring_thread = curthread;
3306 if ((entry->protection & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0
3307 || (entry->protection & prot) != prot) {
3308 entry->eflags |= MAP_ENTRY_WIRE_SKIPPED;
3311 rv = KERN_INVALID_ADDRESS;
3314 } else if (entry->wired_count == 0) {
3315 entry->wired_count++;
3317 npages = atop(entry->end - entry->start);
3318 if (user_wire && !vm_map_wire_user_count_add(npages)) {
3319 vm_map_wire_entry_failure(map, entry,
3322 rv = KERN_RESOURCE_SHORTAGE;
3327 * Release the map lock, relying on the in-transition
3328 * mark. Mark the map busy for fork.
3330 saved_start = entry->start;
3331 saved_end = entry->end;
3332 last_timestamp = map->timestamp;
3336 faddr = saved_start;
3339 * Simulate a fault to get the page and enter
3340 * it into the physical map.
3342 if ((rv = vm_fault(map, faddr,
3343 VM_PROT_NONE, VM_FAULT_WIRE, NULL)) !=
3346 } while ((faddr += PAGE_SIZE) < saved_end);
3349 if (last_timestamp + 1 != map->timestamp) {
3351 * Look again for the entry because the map was
3352 * modified while it was unlocked. The entry
3353 * may have been clipped, but NOT merged or
3356 if (!vm_map_lookup_entry(map, saved_start,
3359 ("vm_map_wire: lookup failed"));
3360 first_entry = (entry == first_entry) ?
3362 for (entry = next_entry; entry->end < saved_end;
3363 entry = vm_map_entry_succ(entry)) {
3365 * In case of failure, handle entries
3366 * that were not fully wired here;
3367 * fully wired entries are handled
3370 if (rv != KERN_SUCCESS &&
3372 vm_map_wire_entry_failure(map,
3376 if (rv != KERN_SUCCESS) {
3377 vm_map_wire_entry_failure(map, entry, faddr);
3379 vm_map_wire_user_count_sub(npages);
3383 } else if (!user_wire ||
3384 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
3385 entry->wired_count++;
3388 * Check the map for holes in the specified region.
3389 * If holes_ok was specified, skip this check.
3391 next_entry = vm_map_entry_succ(entry);
3393 entry->end < end && next_entry->start > entry->end) {
3395 rv = KERN_INVALID_ADDRESS;
3401 need_wakeup = false;
3402 if (first_entry == NULL &&
3403 !vm_map_lookup_entry(map, start, &first_entry)) {
3404 KASSERT(holes_ok, ("vm_map_wire: lookup failed"));
3405 prev_entry = first_entry;
3406 entry = vm_map_entry_succ(first_entry);
3408 prev_entry = vm_map_entry_pred(first_entry);
3409 entry = first_entry;
3411 for (; entry->start < end;
3412 prev_entry = entry, entry = vm_map_entry_succ(entry)) {
3414 * If holes_ok was specified, an empty
3415 * space in the unwired region could have been mapped
3416 * while the map lock was dropped for faulting in the
3417 * pages or draining MAP_ENTRY_IN_TRANSITION.
3418 * Moreover, another thread could be simultaneously
3419 * wiring this new mapping entry. Detect these cases
3420 * and skip any entries marked as in transition not by us.
3422 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 ||
3423 entry->wiring_thread != curthread) {
3425 ("vm_map_wire: !HOLESOK and new/changed entry"));
3429 if ((entry->eflags & MAP_ENTRY_WIRE_SKIPPED) != 0) {
3431 } else if (rv == KERN_SUCCESS) {
3433 entry->eflags |= MAP_ENTRY_USER_WIRED;
3434 } else if (entry->wired_count == -1) {
3436 * Wiring failed on this entry. Thus, unwiring is
3439 entry->wired_count = 0;
3440 } else if (!user_wire ||
3441 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
3443 * Undo the wiring. Wiring succeeded on this entry
3444 * but failed on a later entry.
3446 if (entry->wired_count == 1) {
3447 vm_map_entry_unwire(map, entry);
3449 vm_map_wire_user_count_sub(
3450 atop(entry->end - entry->start));
3452 entry->wired_count--;
3454 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0,
3455 ("vm_map_wire: in-transition flag missing %p", entry));
3456 KASSERT(entry->wiring_thread == curthread,
3457 ("vm_map_wire: alien wire %p", entry));
3458 entry->eflags &= ~(MAP_ENTRY_IN_TRANSITION |
3459 MAP_ENTRY_WIRE_SKIPPED);
3460 entry->wiring_thread = NULL;
3461 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
3462 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
3465 vm_map_try_merge_entries(map, prev_entry, entry);
3467 vm_map_try_merge_entries(map, prev_entry, entry);
3476 * Push any dirty cached pages in the address range to their pager.
3477 * If syncio is TRUE, dirty pages are written synchronously.
3478 * If invalidate is TRUE, any cached pages are freed as well.
3480 * If the size of the region from start to end is zero, we are
3481 * supposed to flush all modified pages within the region containing
3482 * start. Unfortunately, a region can be split or coalesced with
3483 * neighboring regions, making it difficult to determine what the
3484 * original region was. Therefore, we approximate this requirement by
3485 * flushing the current region containing start.
3487 * Returns an error if any part of the specified range is not mapped.
3495 boolean_t invalidate)
3497 vm_map_entry_t entry, first_entry, next_entry;
3500 vm_ooffset_t offset;
3501 unsigned int last_timestamp;
3504 vm_map_lock_read(map);
3505 VM_MAP_RANGE_CHECK(map, start, end);
3506 if (!vm_map_lookup_entry(map, start, &first_entry)) {
3507 vm_map_unlock_read(map);
3508 return (KERN_INVALID_ADDRESS);
3509 } else if (start == end) {
3510 start = first_entry->start;
3511 end = first_entry->end;
3514 * Make a first pass to check for user-wired memory and holes.
3516 for (entry = first_entry; entry->start < end; entry = next_entry) {
3518 (entry->eflags & MAP_ENTRY_USER_WIRED) != 0) {
3519 vm_map_unlock_read(map);
3520 return (KERN_INVALID_ARGUMENT);
3522 next_entry = vm_map_entry_succ(entry);
3523 if (end > entry->end &&
3524 entry->end != next_entry->start) {
3525 vm_map_unlock_read(map);
3526 return (KERN_INVALID_ADDRESS);
3531 pmap_remove(map->pmap, start, end);
3535 * Make a second pass, cleaning/uncaching pages from the indicated
3538 for (entry = first_entry; entry->start < end;) {
3539 offset = entry->offset + (start - entry->start);
3540 size = (end <= entry->end ? end : entry->end) - start;
3541 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0) {
3543 vm_map_entry_t tentry;
3546 smap = entry->object.sub_map;
3547 vm_map_lock_read(smap);
3548 (void) vm_map_lookup_entry(smap, offset, &tentry);
3549 tsize = tentry->end - offset;
3552 object = tentry->object.vm_object;
3553 offset = tentry->offset + (offset - tentry->start);
3554 vm_map_unlock_read(smap);
3556 object = entry->object.vm_object;
3558 vm_object_reference(object);
3559 last_timestamp = map->timestamp;
3560 vm_map_unlock_read(map);
3561 if (!vm_object_sync(object, offset, size, syncio, invalidate))
3564 vm_object_deallocate(object);
3565 vm_map_lock_read(map);
3566 if (last_timestamp == map->timestamp ||
3567 !vm_map_lookup_entry(map, start, &entry))
3568 entry = vm_map_entry_succ(entry);
3571 vm_map_unlock_read(map);
3572 return (failed ? KERN_FAILURE : KERN_SUCCESS);
3576 * vm_map_entry_unwire: [ internal use only ]
3578 * Make the region specified by this entry pageable.
3580 * The map in question should be locked.
3581 * [This is the reason for this routine's existence.]
3584 vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry)
3588 VM_MAP_ASSERT_LOCKED(map);
3589 KASSERT(entry->wired_count > 0,
3590 ("vm_map_entry_unwire: entry %p isn't wired", entry));
3592 size = entry->end - entry->start;
3593 if ((entry->eflags & MAP_ENTRY_USER_WIRED) != 0)
3594 vm_map_wire_user_count_sub(atop(size));
3595 pmap_unwire(map->pmap, entry->start, entry->end);
3596 vm_object_unwire(entry->object.vm_object, entry->offset, size,
3598 entry->wired_count = 0;
3602 vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map)
3605 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0)
3606 vm_object_deallocate(entry->object.vm_object);
3607 uma_zfree(system_map ? kmapentzone : mapentzone, entry);
3611 * vm_map_entry_delete: [ internal use only ]
3613 * Deallocate the given entry from the target map.
3616 vm_map_entry_delete(vm_map_t map, vm_map_entry_t entry)
3619 vm_pindex_t offidxstart, offidxend, count, size1;
3622 vm_map_entry_unlink(map, entry, UNLINK_MERGE_NONE);
3623 object = entry->object.vm_object;
3625 if ((entry->eflags & MAP_ENTRY_GUARD) != 0) {
3626 MPASS(entry->cred == NULL);
3627 MPASS((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0);
3628 MPASS(object == NULL);
3629 vm_map_entry_deallocate(entry, map->system_map);
3633 size = entry->end - entry->start;
3636 if (entry->cred != NULL) {
3637 swap_release_by_cred(size, entry->cred);
3638 crfree(entry->cred);
3641 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0 || object == NULL) {
3642 entry->object.vm_object = NULL;
3643 } else if ((object->flags & OBJ_ANON) != 0 ||
3644 object == kernel_object) {
3645 KASSERT(entry->cred == NULL || object->cred == NULL ||
3646 (entry->eflags & MAP_ENTRY_NEEDS_COPY),
3647 ("OVERCOMMIT vm_map_entry_delete: both cred %p", entry));
3649 offidxstart = OFF_TO_IDX(entry->offset);
3650 offidxend = offidxstart + count;
3651 VM_OBJECT_WLOCK(object);
3652 if (object->ref_count != 1 &&
3653 ((object->flags & OBJ_ONEMAPPING) != 0 ||
3654 object == kernel_object)) {
3655 vm_object_collapse(object);
3658 * The option OBJPR_NOTMAPPED can be passed here
3659 * because vm_map_delete() already performed
3660 * pmap_remove() on the only mapping to this range
3663 vm_object_page_remove(object, offidxstart, offidxend,
3665 if (object->type == OBJT_SWAP)
3666 swap_pager_freespace(object, offidxstart,
3668 if (offidxend >= object->size &&
3669 offidxstart < object->size) {
3670 size1 = object->size;
3671 object->size = offidxstart;
3672 if (object->cred != NULL) {
3673 size1 -= object->size;
3674 KASSERT(object->charge >= ptoa(size1),
3675 ("object %p charge < 0", object));
3676 swap_release_by_cred(ptoa(size1),
3678 object->charge -= ptoa(size1);
3682 VM_OBJECT_WUNLOCK(object);
3684 if (map->system_map)
3685 vm_map_entry_deallocate(entry, TRUE);
3687 entry->defer_next = curthread->td_map_def_user;
3688 curthread->td_map_def_user = entry;
3693 * vm_map_delete: [ internal use only ]
3695 * Deallocates the given address range from the target
3699 vm_map_delete(vm_map_t map, vm_offset_t start, vm_offset_t end)
3701 vm_map_entry_t entry;
3702 vm_map_entry_t first_entry;
3704 VM_MAP_ASSERT_LOCKED(map);
3706 return (KERN_SUCCESS);
3709 * Find the start of the region, and clip it
3711 if (!vm_map_lookup_entry(map, start, &first_entry))
3712 entry = vm_map_entry_succ(first_entry);
3714 entry = first_entry;
3715 vm_map_clip_start(map, entry, start);
3719 * Step through all entries in this region
3721 while (entry->start < end) {
3722 vm_map_entry_t next;
3725 * Wait for wiring or unwiring of an entry to complete.
3726 * Also wait for any system wirings to disappear on
3729 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 ||
3730 (vm_map_pmap(map) != kernel_pmap &&
3731 vm_map_entry_system_wired_count(entry) != 0)) {
3732 unsigned int last_timestamp;
3733 vm_offset_t saved_start;
3734 vm_map_entry_t tmp_entry;
3736 saved_start = entry->start;
3737 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
3738 last_timestamp = map->timestamp;
3739 (void) vm_map_unlock_and_wait(map, 0);
3741 if (last_timestamp + 1 != map->timestamp) {
3743 * Look again for the entry because the map was
3744 * modified while it was unlocked.
3745 * Specifically, the entry may have been
3746 * clipped, merged, or deleted.
3748 if (!vm_map_lookup_entry(map, saved_start,
3750 entry = vm_map_entry_succ(tmp_entry);
3753 vm_map_clip_start(map, entry,
3759 vm_map_clip_end(map, entry, end);
3761 next = vm_map_entry_succ(entry);
3764 * Unwire before removing addresses from the pmap; otherwise,
3765 * unwiring will put the entries back in the pmap.
3767 if (entry->wired_count != 0)
3768 vm_map_entry_unwire(map, entry);
3771 * Remove mappings for the pages, but only if the
3772 * mappings could exist. For instance, it does not
3773 * make sense to call pmap_remove() for guard entries.
3775 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0 ||
3776 entry->object.vm_object != NULL)
3777 pmap_remove(map->pmap, entry->start, entry->end);
3779 if (entry->end == map->anon_loc)
3780 map->anon_loc = entry->start;
3783 * Delete the entry only after removing all pmap
3784 * entries pointing to its pages. (Otherwise, its
3785 * page frames may be reallocated, and any modify bits
3786 * will be set in the wrong object!)
3788 vm_map_entry_delete(map, entry);
3791 return (KERN_SUCCESS);
3797 * Remove the given address range from the target map.
3798 * This is the exported form of vm_map_delete.
3801 vm_map_remove(vm_map_t map, vm_offset_t start, vm_offset_t end)
3806 VM_MAP_RANGE_CHECK(map, start, end);
3807 result = vm_map_delete(map, start, end);
3813 * vm_map_check_protection:
3815 * Assert that the target map allows the specified privilege on the
3816 * entire address region given. The entire region must be allocated.
3818 * WARNING! This code does not and should not check whether the
3819 * contents of the region is accessible. For example a smaller file
3820 * might be mapped into a larger address space.
3822 * NOTE! This code is also called by munmap().
3824 * The map must be locked. A read lock is sufficient.
3827 vm_map_check_protection(vm_map_t map, vm_offset_t start, vm_offset_t end,
3828 vm_prot_t protection)
3830 vm_map_entry_t entry;
3831 vm_map_entry_t tmp_entry;
3833 if (!vm_map_lookup_entry(map, start, &tmp_entry))
3837 while (start < end) {
3841 if (start < entry->start)
3844 * Check protection associated with entry.
3846 if ((entry->protection & protection) != protection)
3848 /* go to next entry */
3850 entry = vm_map_entry_succ(entry);
3858 * vm_map_copy_swap_object:
3860 * Copies a swap-backed object from an existing map entry to a
3861 * new one. Carries forward the swap charge. May change the
3862 * src object on return.
3865 vm_map_copy_swap_object(vm_map_entry_t src_entry, vm_map_entry_t dst_entry,
3866 vm_offset_t size, vm_ooffset_t *fork_charge)
3868 vm_object_t src_object;
3872 src_object = src_entry->object.vm_object;
3873 charged = ENTRY_CHARGED(src_entry);
3874 if ((src_object->flags & OBJ_ANON) != 0) {
3875 VM_OBJECT_WLOCK(src_object);
3876 vm_object_collapse(src_object);
3877 if ((src_object->flags & OBJ_ONEMAPPING) != 0) {
3878 vm_object_split(src_entry);
3879 src_object = src_entry->object.vm_object;
3881 vm_object_reference_locked(src_object);
3882 vm_object_clear_flag(src_object, OBJ_ONEMAPPING);
3883 VM_OBJECT_WUNLOCK(src_object);
3885 vm_object_reference(src_object);
3886 if (src_entry->cred != NULL &&
3887 !(src_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
3888 KASSERT(src_object->cred == NULL,
3889 ("OVERCOMMIT: vm_map_copy_anon_entry: cred %p",
3891 src_object->cred = src_entry->cred;
3892 src_object->charge = size;
3894 dst_entry->object.vm_object = src_object;
3896 cred = curthread->td_ucred;
3898 dst_entry->cred = cred;
3899 *fork_charge += size;
3900 if (!(src_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
3902 src_entry->cred = cred;
3903 *fork_charge += size;
3909 * vm_map_copy_entry:
3911 * Copies the contents of the source entry to the destination
3912 * entry. The entries *must* be aligned properly.
3918 vm_map_entry_t src_entry,
3919 vm_map_entry_t dst_entry,
3920 vm_ooffset_t *fork_charge)
3922 vm_object_t src_object;
3923 vm_map_entry_t fake_entry;
3926 VM_MAP_ASSERT_LOCKED(dst_map);
3928 if ((dst_entry->eflags|src_entry->eflags) & MAP_ENTRY_IS_SUB_MAP)
3931 if (src_entry->wired_count == 0 ||
3932 (src_entry->protection & VM_PROT_WRITE) == 0) {
3934 * If the source entry is marked needs_copy, it is already
3937 if ((src_entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0 &&
3938 (src_entry->protection & VM_PROT_WRITE) != 0) {
3939 pmap_protect(src_map->pmap,
3942 src_entry->protection & ~VM_PROT_WRITE);
3946 * Make a copy of the object.
3948 size = src_entry->end - src_entry->start;
3949 if ((src_object = src_entry->object.vm_object) != NULL) {
3950 if (src_object->type == OBJT_DEFAULT ||
3951 src_object->type == OBJT_SWAP) {
3952 vm_map_copy_swap_object(src_entry, dst_entry,
3954 /* May have split/collapsed, reload obj. */
3955 src_object = src_entry->object.vm_object;
3957 vm_object_reference(src_object);
3958 dst_entry->object.vm_object = src_object;
3960 src_entry->eflags |= MAP_ENTRY_COW |
3961 MAP_ENTRY_NEEDS_COPY;
3962 dst_entry->eflags |= MAP_ENTRY_COW |
3963 MAP_ENTRY_NEEDS_COPY;
3964 dst_entry->offset = src_entry->offset;
3965 if (src_entry->eflags & MAP_ENTRY_WRITECNT) {
3967 * MAP_ENTRY_WRITECNT cannot
3968 * indicate write reference from
3969 * src_entry, since the entry is
3970 * marked as needs copy. Allocate a
3971 * fake entry that is used to
3972 * decrement object->un_pager writecount
3973 * at the appropriate time. Attach
3974 * fake_entry to the deferred list.
3976 fake_entry = vm_map_entry_create(dst_map);
3977 fake_entry->eflags = MAP_ENTRY_WRITECNT;
3978 src_entry->eflags &= ~MAP_ENTRY_WRITECNT;
3979 vm_object_reference(src_object);
3980 fake_entry->object.vm_object = src_object;
3981 fake_entry->start = src_entry->start;
3982 fake_entry->end = src_entry->end;
3983 fake_entry->defer_next =
3984 curthread->td_map_def_user;
3985 curthread->td_map_def_user = fake_entry;
3988 pmap_copy(dst_map->pmap, src_map->pmap,
3989 dst_entry->start, dst_entry->end - dst_entry->start,
3992 dst_entry->object.vm_object = NULL;
3993 dst_entry->offset = 0;
3994 if (src_entry->cred != NULL) {
3995 dst_entry->cred = curthread->td_ucred;
3996 crhold(dst_entry->cred);
3997 *fork_charge += size;
4002 * We don't want to make writeable wired pages copy-on-write.
4003 * Immediately copy these pages into the new map by simulating
4004 * page faults. The new pages are pageable.
4006 vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry,
4012 * vmspace_map_entry_forked:
4013 * Update the newly-forked vmspace each time a map entry is inherited
4014 * or copied. The values for vm_dsize and vm_tsize are approximate
4015 * (and mostly-obsolete ideas in the face of mmap(2) et al.)
4018 vmspace_map_entry_forked(const struct vmspace *vm1, struct vmspace *vm2,
4019 vm_map_entry_t entry)
4021 vm_size_t entrysize;
4024 if ((entry->eflags & MAP_ENTRY_GUARD) != 0)
4026 entrysize = entry->end - entry->start;
4027 vm2->vm_map.size += entrysize;
4028 if (entry->eflags & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP)) {
4029 vm2->vm_ssize += btoc(entrysize);
4030 } else if (entry->start >= (vm_offset_t)vm1->vm_daddr &&
4031 entry->start < (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)) {
4032 newend = MIN(entry->end,
4033 (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize));
4034 vm2->vm_dsize += btoc(newend - entry->start);
4035 } else if (entry->start >= (vm_offset_t)vm1->vm_taddr &&
4036 entry->start < (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)) {
4037 newend = MIN(entry->end,
4038 (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize));
4039 vm2->vm_tsize += btoc(newend - entry->start);
4045 * Create a new process vmspace structure and vm_map
4046 * based on those of an existing process. The new map
4047 * is based on the old map, according to the inheritance
4048 * values on the regions in that map.
4050 * XXX It might be worth coalescing the entries added to the new vmspace.
4052 * The source map must not be locked.
4055 vmspace_fork(struct vmspace *vm1, vm_ooffset_t *fork_charge)
4057 struct vmspace *vm2;
4058 vm_map_t new_map, old_map;
4059 vm_map_entry_t new_entry, old_entry;
4064 old_map = &vm1->vm_map;
4065 /* Copy immutable fields of vm1 to vm2. */
4066 vm2 = vmspace_alloc(vm_map_min(old_map), vm_map_max(old_map),
4071 vm2->vm_taddr = vm1->vm_taddr;
4072 vm2->vm_daddr = vm1->vm_daddr;
4073 vm2->vm_maxsaddr = vm1->vm_maxsaddr;
4074 vm_map_lock(old_map);
4076 vm_map_wait_busy(old_map);
4077 new_map = &vm2->vm_map;
4078 locked = vm_map_trylock(new_map); /* trylock to silence WITNESS */
4079 KASSERT(locked, ("vmspace_fork: lock failed"));
4081 error = pmap_vmspace_copy(new_map->pmap, old_map->pmap);
4083 sx_xunlock(&old_map->lock);
4084 sx_xunlock(&new_map->lock);
4085 vm_map_process_deferred();
4090 new_map->anon_loc = old_map->anon_loc;
4092 VM_MAP_ENTRY_FOREACH(old_entry, old_map) {
4093 if ((old_entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0)
4094 panic("vm_map_fork: encountered a submap");
4096 inh = old_entry->inheritance;
4097 if ((old_entry->eflags & MAP_ENTRY_GUARD) != 0 &&
4098 inh != VM_INHERIT_NONE)
4099 inh = VM_INHERIT_COPY;
4102 case VM_INHERIT_NONE:
4105 case VM_INHERIT_SHARE:
4107 * Clone the entry, creating the shared object if
4110 object = old_entry->object.vm_object;
4111 if (object == NULL) {
4112 vm_map_entry_back(old_entry);
4113 object = old_entry->object.vm_object;
4117 * Add the reference before calling vm_object_shadow
4118 * to insure that a shadow object is created.
4120 vm_object_reference(object);
4121 if (old_entry->eflags & MAP_ENTRY_NEEDS_COPY) {
4122 vm_object_shadow(&old_entry->object.vm_object,
4124 old_entry->end - old_entry->start,
4126 /* Transfer the second reference too. */
4128 old_entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
4129 old_entry->cred = NULL;
4130 vm_object_reference(
4131 old_entry->object.vm_object);
4134 * As in vm_map_merged_neighbor_dispose(),
4135 * the vnode lock will not be acquired in
4136 * this call to vm_object_deallocate().
4138 vm_object_deallocate(object);
4139 object = old_entry->object.vm_object;
4141 VM_OBJECT_WLOCK(object);
4142 vm_object_clear_flag(object, OBJ_ONEMAPPING);
4143 if (old_entry->cred != NULL) {
4144 KASSERT(object->cred == NULL,
4145 ("vmspace_fork both cred"));
4146 object->cred = old_entry->cred;
4147 object->charge = old_entry->end -
4149 old_entry->cred = NULL;
4153 * Assert the correct state of the vnode
4154 * v_writecount while the object is locked, to
4155 * not relock it later for the assertion
4158 if (old_entry->eflags & MAP_ENTRY_WRITECNT &&
4159 object->type == OBJT_VNODE) {
4160 KASSERT(((struct vnode *)object->
4161 handle)->v_writecount > 0,
4162 ("vmspace_fork: v_writecount %p",
4164 KASSERT(object->un_pager.vnp.
4166 ("vmspace_fork: vnp.writecount %p",
4169 VM_OBJECT_WUNLOCK(object);
4173 * Clone the entry, referencing the shared object.
4175 new_entry = vm_map_entry_create(new_map);
4176 *new_entry = *old_entry;
4177 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED |
4178 MAP_ENTRY_IN_TRANSITION);
4179 new_entry->wiring_thread = NULL;
4180 new_entry->wired_count = 0;
4181 if (new_entry->eflags & MAP_ENTRY_WRITECNT) {
4182 vm_pager_update_writecount(object,
4183 new_entry->start, new_entry->end);
4185 vm_map_entry_set_vnode_text(new_entry, true);
4188 * Insert the entry into the new map -- we know we're
4189 * inserting at the end of the new map.
4191 vm_map_entry_link(new_map, new_entry);
4192 vmspace_map_entry_forked(vm1, vm2, new_entry);
4195 * Update the physical map
4197 pmap_copy(new_map->pmap, old_map->pmap,
4199 (old_entry->end - old_entry->start),
4203 case VM_INHERIT_COPY:
4205 * Clone the entry and link into the map.
4207 new_entry = vm_map_entry_create(new_map);
4208 *new_entry = *old_entry;
4210 * Copied entry is COW over the old object.
4212 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED |
4213 MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_WRITECNT);
4214 new_entry->wiring_thread = NULL;
4215 new_entry->wired_count = 0;
4216 new_entry->object.vm_object = NULL;
4217 new_entry->cred = NULL;
4218 vm_map_entry_link(new_map, new_entry);
4219 vmspace_map_entry_forked(vm1, vm2, new_entry);
4220 vm_map_copy_entry(old_map, new_map, old_entry,
4221 new_entry, fork_charge);
4222 vm_map_entry_set_vnode_text(new_entry, true);
4225 case VM_INHERIT_ZERO:
4227 * Create a new anonymous mapping entry modelled from
4230 new_entry = vm_map_entry_create(new_map);
4231 memset(new_entry, 0, sizeof(*new_entry));
4233 new_entry->start = old_entry->start;
4234 new_entry->end = old_entry->end;
4235 new_entry->eflags = old_entry->eflags &
4236 ~(MAP_ENTRY_USER_WIRED | MAP_ENTRY_IN_TRANSITION |
4237 MAP_ENTRY_WRITECNT | MAP_ENTRY_VN_EXEC);
4238 new_entry->protection = old_entry->protection;
4239 new_entry->max_protection = old_entry->max_protection;
4240 new_entry->inheritance = VM_INHERIT_ZERO;
4242 vm_map_entry_link(new_map, new_entry);
4243 vmspace_map_entry_forked(vm1, vm2, new_entry);
4245 new_entry->cred = curthread->td_ucred;
4246 crhold(new_entry->cred);
4247 *fork_charge += (new_entry->end - new_entry->start);
4253 * Use inlined vm_map_unlock() to postpone handling the deferred
4254 * map entries, which cannot be done until both old_map and
4255 * new_map locks are released.
4257 sx_xunlock(&old_map->lock);
4258 sx_xunlock(&new_map->lock);
4259 vm_map_process_deferred();
4265 * Create a process's stack for exec_new_vmspace(). This function is never
4266 * asked to wire the newly created stack.
4269 vm_map_stack(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
4270 vm_prot_t prot, vm_prot_t max, int cow)
4272 vm_size_t growsize, init_ssize;
4276 MPASS((map->flags & MAP_WIREFUTURE) == 0);
4277 growsize = sgrowsiz;
4278 init_ssize = (max_ssize < growsize) ? max_ssize : growsize;
4280 vmemlim = lim_cur(curthread, RLIMIT_VMEM);
4281 /* If we would blow our VMEM resource limit, no go */
4282 if (map->size + init_ssize > vmemlim) {
4286 rv = vm_map_stack_locked(map, addrbos, max_ssize, growsize, prot,
4293 static int stack_guard_page = 1;
4294 SYSCTL_INT(_security_bsd, OID_AUTO, stack_guard_page, CTLFLAG_RWTUN,
4295 &stack_guard_page, 0,
4296 "Specifies the number of guard pages for a stack that grows");
4299 vm_map_stack_locked(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
4300 vm_size_t growsize, vm_prot_t prot, vm_prot_t max, int cow)
4302 vm_map_entry_t new_entry, prev_entry;
4303 vm_offset_t bot, gap_bot, gap_top, top;
4304 vm_size_t init_ssize, sgp;
4308 * The stack orientation is piggybacked with the cow argument.
4309 * Extract it into orient and mask the cow argument so that we
4310 * don't pass it around further.
4312 orient = cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP);
4313 KASSERT(orient != 0, ("No stack grow direction"));
4314 KASSERT(orient != (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP),
4317 if (addrbos < vm_map_min(map) ||
4318 addrbos + max_ssize > vm_map_max(map) ||
4319 addrbos + max_ssize <= addrbos)
4320 return (KERN_INVALID_ADDRESS);
4321 sgp = ((curproc->p_flag2 & P2_STKGAP_DISABLE) != 0 ||
4322 (curproc->p_fctl0 & NT_FREEBSD_FCTL_STKGAP_DISABLE) != 0) ? 0 :
4323 (vm_size_t)stack_guard_page * PAGE_SIZE;
4324 if (sgp >= max_ssize)
4325 return (KERN_INVALID_ARGUMENT);
4327 init_ssize = growsize;
4328 if (max_ssize < init_ssize + sgp)
4329 init_ssize = max_ssize - sgp;
4331 /* If addr is already mapped, no go */
4332 if (vm_map_lookup_entry(map, addrbos, &prev_entry))
4333 return (KERN_NO_SPACE);
4336 * If we can't accommodate max_ssize in the current mapping, no go.
4338 if (vm_map_entry_succ(prev_entry)->start < addrbos + max_ssize)
4339 return (KERN_NO_SPACE);
4342 * We initially map a stack of only init_ssize. We will grow as
4343 * needed later. Depending on the orientation of the stack (i.e.
4344 * the grow direction) we either map at the top of the range, the
4345 * bottom of the range or in the middle.
4347 * Note: we would normally expect prot and max to be VM_PROT_ALL,
4348 * and cow to be 0. Possibly we should eliminate these as input
4349 * parameters, and just pass these values here in the insert call.
4351 if (orient == MAP_STACK_GROWS_DOWN) {
4352 bot = addrbos + max_ssize - init_ssize;
4353 top = bot + init_ssize;
4356 } else /* if (orient == MAP_STACK_GROWS_UP) */ {
4358 top = bot + init_ssize;
4360 gap_top = addrbos + max_ssize;
4362 rv = vm_map_insert(map, NULL, 0, bot, top, prot, max, cow);
4363 if (rv != KERN_SUCCESS)
4365 new_entry = vm_map_entry_succ(prev_entry);
4366 KASSERT(new_entry->end == top || new_entry->start == bot,
4367 ("Bad entry start/end for new stack entry"));
4368 KASSERT((orient & MAP_STACK_GROWS_DOWN) == 0 ||
4369 (new_entry->eflags & MAP_ENTRY_GROWS_DOWN) != 0,
4370 ("new entry lacks MAP_ENTRY_GROWS_DOWN"));
4371 KASSERT((orient & MAP_STACK_GROWS_UP) == 0 ||
4372 (new_entry->eflags & MAP_ENTRY_GROWS_UP) != 0,
4373 ("new entry lacks MAP_ENTRY_GROWS_UP"));
4374 if (gap_bot == gap_top)
4375 return (KERN_SUCCESS);
4376 rv = vm_map_insert(map, NULL, 0, gap_bot, gap_top, VM_PROT_NONE,
4377 VM_PROT_NONE, MAP_CREATE_GUARD | (orient == MAP_STACK_GROWS_DOWN ?
4378 MAP_CREATE_STACK_GAP_DN : MAP_CREATE_STACK_GAP_UP));
4379 if (rv == KERN_SUCCESS) {
4381 * Gap can never successfully handle a fault, so
4382 * read-ahead logic is never used for it. Re-use
4383 * next_read of the gap entry to store
4384 * stack_guard_page for vm_map_growstack().
4386 if (orient == MAP_STACK_GROWS_DOWN)
4387 vm_map_entry_pred(new_entry)->next_read = sgp;
4389 vm_map_entry_succ(new_entry)->next_read = sgp;
4391 (void)vm_map_delete(map, bot, top);
4397 * Attempts to grow a vm stack entry. Returns KERN_SUCCESS if we
4398 * successfully grow the stack.
4401 vm_map_growstack(vm_map_t map, vm_offset_t addr, vm_map_entry_t gap_entry)
4403 vm_map_entry_t stack_entry;
4407 vm_offset_t gap_end, gap_start, grow_start;
4408 vm_size_t grow_amount, guard, max_grow;
4409 rlim_t lmemlim, stacklim, vmemlim;
4411 bool gap_deleted, grow_down, is_procstack;
4423 * Disallow stack growth when the access is performed by a
4424 * debugger or AIO daemon. The reason is that the wrong
4425 * resource limits are applied.
4427 if (p != initproc && (map != &p->p_vmspace->vm_map ||
4428 p->p_textvp == NULL))
4429 return (KERN_FAILURE);
4431 MPASS(!map->system_map);
4433 lmemlim = lim_cur(curthread, RLIMIT_MEMLOCK);
4434 stacklim = lim_cur(curthread, RLIMIT_STACK);
4435 vmemlim = lim_cur(curthread, RLIMIT_VMEM);
4437 /* If addr is not in a hole for a stack grow area, no need to grow. */
4438 if (gap_entry == NULL && !vm_map_lookup_entry(map, addr, &gap_entry))
4439 return (KERN_FAILURE);
4440 if ((gap_entry->eflags & MAP_ENTRY_GUARD) == 0)
4441 return (KERN_SUCCESS);
4442 if ((gap_entry->eflags & MAP_ENTRY_STACK_GAP_DN) != 0) {
4443 stack_entry = vm_map_entry_succ(gap_entry);
4444 if ((stack_entry->eflags & MAP_ENTRY_GROWS_DOWN) == 0 ||
4445 stack_entry->start != gap_entry->end)
4446 return (KERN_FAILURE);
4447 grow_amount = round_page(stack_entry->start - addr);
4449 } else if ((gap_entry->eflags & MAP_ENTRY_STACK_GAP_UP) != 0) {
4450 stack_entry = vm_map_entry_pred(gap_entry);
4451 if ((stack_entry->eflags & MAP_ENTRY_GROWS_UP) == 0 ||
4452 stack_entry->end != gap_entry->start)
4453 return (KERN_FAILURE);
4454 grow_amount = round_page(addr + 1 - stack_entry->end);
4457 return (KERN_FAILURE);
4459 guard = ((curproc->p_flag2 & P2_STKGAP_DISABLE) != 0 ||
4460 (curproc->p_fctl0 & NT_FREEBSD_FCTL_STKGAP_DISABLE) != 0) ? 0 :
4461 gap_entry->next_read;
4462 max_grow = gap_entry->end - gap_entry->start;
4463 if (guard > max_grow)
4464 return (KERN_NO_SPACE);
4466 if (grow_amount > max_grow)
4467 return (KERN_NO_SPACE);
4470 * If this is the main process stack, see if we're over the stack
4473 is_procstack = addr >= (vm_offset_t)vm->vm_maxsaddr &&
4474 addr < (vm_offset_t)p->p_sysent->sv_usrstack;
4475 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim))
4476 return (KERN_NO_SPACE);
4481 if (is_procstack && racct_set(p, RACCT_STACK,
4482 ctob(vm->vm_ssize) + grow_amount)) {
4484 return (KERN_NO_SPACE);
4490 grow_amount = roundup(grow_amount, sgrowsiz);
4491 if (grow_amount > max_grow)
4492 grow_amount = max_grow;
4493 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) {
4494 grow_amount = trunc_page((vm_size_t)stacklim) -
4500 limit = racct_get_available(p, RACCT_STACK);
4502 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > limit))
4503 grow_amount = limit - ctob(vm->vm_ssize);
4506 if (!old_mlock && (map->flags & MAP_WIREFUTURE) != 0) {
4507 if (ptoa(pmap_wired_count(map->pmap)) + grow_amount > lmemlim) {
4514 if (racct_set(p, RACCT_MEMLOCK,
4515 ptoa(pmap_wired_count(map->pmap)) + grow_amount)) {
4525 /* If we would blow our VMEM resource limit, no go */
4526 if (map->size + grow_amount > vmemlim) {
4533 if (racct_set(p, RACCT_VMEM, map->size + grow_amount)) {
4542 if (vm_map_lock_upgrade(map)) {
4544 vm_map_lock_read(map);
4549 grow_start = gap_entry->end - grow_amount;
4550 if (gap_entry->start + grow_amount == gap_entry->end) {
4551 gap_start = gap_entry->start;
4552 gap_end = gap_entry->end;
4553 vm_map_entry_delete(map, gap_entry);
4556 MPASS(gap_entry->start < gap_entry->end - grow_amount);
4557 vm_map_entry_resize(map, gap_entry, -grow_amount);
4558 gap_deleted = false;
4560 rv = vm_map_insert(map, NULL, 0, grow_start,
4561 grow_start + grow_amount,
4562 stack_entry->protection, stack_entry->max_protection,
4563 MAP_STACK_GROWS_DOWN);
4564 if (rv != KERN_SUCCESS) {
4566 rv1 = vm_map_insert(map, NULL, 0, gap_start,
4567 gap_end, VM_PROT_NONE, VM_PROT_NONE,
4568 MAP_CREATE_GUARD | MAP_CREATE_STACK_GAP_DN);
4569 MPASS(rv1 == KERN_SUCCESS);
4571 vm_map_entry_resize(map, gap_entry,
4575 grow_start = stack_entry->end;
4576 cred = stack_entry->cred;
4577 if (cred == NULL && stack_entry->object.vm_object != NULL)
4578 cred = stack_entry->object.vm_object->cred;
4579 if (cred != NULL && !swap_reserve_by_cred(grow_amount, cred))
4581 /* Grow the underlying object if applicable. */
4582 else if (stack_entry->object.vm_object == NULL ||
4583 vm_object_coalesce(stack_entry->object.vm_object,
4584 stack_entry->offset,
4585 (vm_size_t)(stack_entry->end - stack_entry->start),
4586 grow_amount, cred != NULL)) {
4587 if (gap_entry->start + grow_amount == gap_entry->end) {
4588 vm_map_entry_delete(map, gap_entry);
4589 vm_map_entry_resize(map, stack_entry,
4592 gap_entry->start += grow_amount;
4593 stack_entry->end += grow_amount;
4595 map->size += grow_amount;
4600 if (rv == KERN_SUCCESS && is_procstack)
4601 vm->vm_ssize += btoc(grow_amount);
4604 * Heed the MAP_WIREFUTURE flag if it was set for this process.
4606 if (rv == KERN_SUCCESS && (map->flags & MAP_WIREFUTURE) != 0) {
4607 rv = vm_map_wire_locked(map, grow_start,
4608 grow_start + grow_amount,
4609 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
4611 vm_map_lock_downgrade(map);
4615 if (racct_enable && rv != KERN_SUCCESS) {
4617 error = racct_set(p, RACCT_VMEM, map->size);
4618 KASSERT(error == 0, ("decreasing RACCT_VMEM failed"));
4620 error = racct_set(p, RACCT_MEMLOCK,
4621 ptoa(pmap_wired_count(map->pmap)));
4622 KASSERT(error == 0, ("decreasing RACCT_MEMLOCK failed"));
4624 error = racct_set(p, RACCT_STACK, ctob(vm->vm_ssize));
4625 KASSERT(error == 0, ("decreasing RACCT_STACK failed"));
4634 * Unshare the specified VM space for exec. If other processes are
4635 * mapped to it, then create a new one. The new vmspace is null.
4638 vmspace_exec(struct proc *p, vm_offset_t minuser, vm_offset_t maxuser)
4640 struct vmspace *oldvmspace = p->p_vmspace;
4641 struct vmspace *newvmspace;
4643 KASSERT((curthread->td_pflags & TDP_EXECVMSPC) == 0,
4644 ("vmspace_exec recursed"));
4645 newvmspace = vmspace_alloc(minuser, maxuser, pmap_pinit);
4646 if (newvmspace == NULL)
4648 newvmspace->vm_swrss = oldvmspace->vm_swrss;
4650 * This code is written like this for prototype purposes. The
4651 * goal is to avoid running down the vmspace here, but let the
4652 * other process's that are still using the vmspace to finally
4653 * run it down. Even though there is little or no chance of blocking
4654 * here, it is a good idea to keep this form for future mods.
4656 PROC_VMSPACE_LOCK(p);
4657 p->p_vmspace = newvmspace;
4658 PROC_VMSPACE_UNLOCK(p);
4659 if (p == curthread->td_proc)
4660 pmap_activate(curthread);
4661 curthread->td_pflags |= TDP_EXECVMSPC;
4666 * Unshare the specified VM space for forcing COW. This
4667 * is called by rfork, for the (RFMEM|RFPROC) == 0 case.
4670 vmspace_unshare(struct proc *p)
4672 struct vmspace *oldvmspace = p->p_vmspace;
4673 struct vmspace *newvmspace;
4674 vm_ooffset_t fork_charge;
4676 if (oldvmspace->vm_refcnt == 1)
4679 newvmspace = vmspace_fork(oldvmspace, &fork_charge);
4680 if (newvmspace == NULL)
4682 if (!swap_reserve_by_cred(fork_charge, p->p_ucred)) {
4683 vmspace_free(newvmspace);
4686 PROC_VMSPACE_LOCK(p);
4687 p->p_vmspace = newvmspace;
4688 PROC_VMSPACE_UNLOCK(p);
4689 if (p == curthread->td_proc)
4690 pmap_activate(curthread);
4691 vmspace_free(oldvmspace);
4698 * Finds the VM object, offset, and
4699 * protection for a given virtual address in the
4700 * specified map, assuming a page fault of the
4703 * Leaves the map in question locked for read; return
4704 * values are guaranteed until a vm_map_lookup_done
4705 * call is performed. Note that the map argument
4706 * is in/out; the returned map must be used in
4707 * the call to vm_map_lookup_done.
4709 * A handle (out_entry) is returned for use in
4710 * vm_map_lookup_done, to make that fast.
4712 * If a lookup is requested with "write protection"
4713 * specified, the map may be changed to perform virtual
4714 * copying operations, although the data referenced will
4718 vm_map_lookup(vm_map_t *var_map, /* IN/OUT */
4720 vm_prot_t fault_typea,
4721 vm_map_entry_t *out_entry, /* OUT */
4722 vm_object_t *object, /* OUT */
4723 vm_pindex_t *pindex, /* OUT */
4724 vm_prot_t *out_prot, /* OUT */
4725 boolean_t *wired) /* OUT */
4727 vm_map_entry_t entry;
4728 vm_map_t map = *var_map;
4730 vm_prot_t fault_type;
4731 vm_object_t eobject;
4737 vm_map_lock_read(map);
4741 * Lookup the faulting address.
4743 if (!vm_map_lookup_entry(map, vaddr, out_entry)) {
4744 vm_map_unlock_read(map);
4745 return (KERN_INVALID_ADDRESS);
4753 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
4754 vm_map_t old_map = map;
4756 *var_map = map = entry->object.sub_map;
4757 vm_map_unlock_read(old_map);
4762 * Check whether this task is allowed to have this page.
4764 prot = entry->protection;
4765 if ((fault_typea & VM_PROT_FAULT_LOOKUP) != 0) {
4766 fault_typea &= ~VM_PROT_FAULT_LOOKUP;
4767 if (prot == VM_PROT_NONE && map != kernel_map &&
4768 (entry->eflags & MAP_ENTRY_GUARD) != 0 &&
4769 (entry->eflags & (MAP_ENTRY_STACK_GAP_DN |
4770 MAP_ENTRY_STACK_GAP_UP)) != 0 &&
4771 vm_map_growstack(map, vaddr, entry) == KERN_SUCCESS)
4772 goto RetryLookupLocked;
4774 fault_type = fault_typea & VM_PROT_ALL;
4775 if ((fault_type & prot) != fault_type || prot == VM_PROT_NONE) {
4776 vm_map_unlock_read(map);
4777 return (KERN_PROTECTION_FAILURE);
4779 KASSERT((prot & VM_PROT_WRITE) == 0 || (entry->eflags &
4780 (MAP_ENTRY_USER_WIRED | MAP_ENTRY_NEEDS_COPY)) !=
4781 (MAP_ENTRY_USER_WIRED | MAP_ENTRY_NEEDS_COPY),
4782 ("entry %p flags %x", entry, entry->eflags));
4783 if ((fault_typea & VM_PROT_COPY) != 0 &&
4784 (entry->max_protection & VM_PROT_WRITE) == 0 &&
4785 (entry->eflags & MAP_ENTRY_COW) == 0) {
4786 vm_map_unlock_read(map);
4787 return (KERN_PROTECTION_FAILURE);
4791 * If this page is not pageable, we have to get it for all possible
4794 *wired = (entry->wired_count != 0);
4796 fault_type = entry->protection;
4797 size = entry->end - entry->start;
4800 * If the entry was copy-on-write, we either ...
4802 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
4804 * If we want to write the page, we may as well handle that
4805 * now since we've got the map locked.
4807 * If we don't need to write the page, we just demote the
4808 * permissions allowed.
4810 if ((fault_type & VM_PROT_WRITE) != 0 ||
4811 (fault_typea & VM_PROT_COPY) != 0) {
4813 * Make a new object, and place it in the object
4814 * chain. Note that no new references have appeared
4815 * -- one just moved from the map to the new
4818 if (vm_map_lock_upgrade(map))
4821 if (entry->cred == NULL) {
4823 * The debugger owner is charged for
4826 cred = curthread->td_ucred;
4828 if (!swap_reserve_by_cred(size, cred)) {
4831 return (KERN_RESOURCE_SHORTAGE);
4835 eobject = entry->object.vm_object;
4836 vm_object_shadow(&entry->object.vm_object,
4837 &entry->offset, size, entry->cred, false);
4838 if (eobject == entry->object.vm_object) {
4840 * The object was not shadowed.
4842 swap_release_by_cred(size, entry->cred);
4843 crfree(entry->cred);
4846 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
4848 vm_map_lock_downgrade(map);
4851 * We're attempting to read a copy-on-write page --
4852 * don't allow writes.
4854 prot &= ~VM_PROT_WRITE;
4859 * Create an object if necessary.
4861 if (entry->object.vm_object == NULL && !map->system_map) {
4862 if (vm_map_lock_upgrade(map))
4864 entry->object.vm_object = vm_object_allocate_anon(atop(size),
4865 NULL, entry->cred, entry->cred != NULL ? size : 0);
4868 vm_map_lock_downgrade(map);
4872 * Return the object/offset from this entry. If the entry was
4873 * copy-on-write or empty, it has been fixed up.
4875 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
4876 *object = entry->object.vm_object;
4879 return (KERN_SUCCESS);
4883 * vm_map_lookup_locked:
4885 * Lookup the faulting address. A version of vm_map_lookup that returns
4886 * KERN_FAILURE instead of blocking on map lock or memory allocation.
4889 vm_map_lookup_locked(vm_map_t *var_map, /* IN/OUT */
4891 vm_prot_t fault_typea,
4892 vm_map_entry_t *out_entry, /* OUT */
4893 vm_object_t *object, /* OUT */
4894 vm_pindex_t *pindex, /* OUT */
4895 vm_prot_t *out_prot, /* OUT */
4896 boolean_t *wired) /* OUT */
4898 vm_map_entry_t entry;
4899 vm_map_t map = *var_map;
4901 vm_prot_t fault_type = fault_typea;
4904 * Lookup the faulting address.
4906 if (!vm_map_lookup_entry(map, vaddr, out_entry))
4907 return (KERN_INVALID_ADDRESS);
4912 * Fail if the entry refers to a submap.
4914 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP)
4915 return (KERN_FAILURE);
4918 * Check whether this task is allowed to have this page.
4920 prot = entry->protection;
4921 fault_type &= VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE;
4922 if ((fault_type & prot) != fault_type)
4923 return (KERN_PROTECTION_FAILURE);
4926 * If this page is not pageable, we have to get it for all possible
4929 *wired = (entry->wired_count != 0);
4931 fault_type = entry->protection;
4933 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
4935 * Fail if the entry was copy-on-write for a write fault.
4937 if (fault_type & VM_PROT_WRITE)
4938 return (KERN_FAILURE);
4940 * We're attempting to read a copy-on-write page --
4941 * don't allow writes.
4943 prot &= ~VM_PROT_WRITE;
4947 * Fail if an object should be created.
4949 if (entry->object.vm_object == NULL && !map->system_map)
4950 return (KERN_FAILURE);
4953 * Return the object/offset from this entry. If the entry was
4954 * copy-on-write or empty, it has been fixed up.
4956 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
4957 *object = entry->object.vm_object;
4960 return (KERN_SUCCESS);
4964 * vm_map_lookup_done:
4966 * Releases locks acquired by a vm_map_lookup
4967 * (according to the handle returned by that lookup).
4970 vm_map_lookup_done(vm_map_t map, vm_map_entry_t entry)
4973 * Unlock the main-level map
4975 vm_map_unlock_read(map);
4979 vm_map_max_KBI(const struct vm_map *map)
4982 return (vm_map_max(map));
4986 vm_map_min_KBI(const struct vm_map *map)
4989 return (vm_map_min(map));
4993 vm_map_pmap_KBI(vm_map_t map)
5001 _vm_map_assert_consistent(vm_map_t map, int check)
5003 vm_map_entry_t entry, prev;
5004 vm_map_entry_t cur, header, lbound, ubound;
5005 vm_size_t max_left, max_right;
5010 if (enable_vmmap_check != check)
5013 header = prev = &map->header;
5014 VM_MAP_ENTRY_FOREACH(entry, map) {
5015 KASSERT(prev->end <= entry->start,
5016 ("map %p prev->end = %jx, start = %jx", map,
5017 (uintmax_t)prev->end, (uintmax_t)entry->start));
5018 KASSERT(entry->start < entry->end,
5019 ("map %p start = %jx, end = %jx", map,
5020 (uintmax_t)entry->start, (uintmax_t)entry->end));
5021 KASSERT(entry->left == header ||
5022 entry->left->start < entry->start,
5023 ("map %p left->start = %jx, start = %jx", map,
5024 (uintmax_t)entry->left->start, (uintmax_t)entry->start));
5025 KASSERT(entry->right == header ||
5026 entry->start < entry->right->start,
5027 ("map %p start = %jx, right->start = %jx", map,
5028 (uintmax_t)entry->start, (uintmax_t)entry->right->start));
5030 lbound = ubound = header;
5032 if (entry->start < cur->start) {
5035 KASSERT(cur != lbound,
5036 ("map %p cannot find %jx",
5037 map, (uintmax_t)entry->start));
5038 } else if (cur->end <= entry->start) {
5041 KASSERT(cur != ubound,
5042 ("map %p cannot find %jx",
5043 map, (uintmax_t)entry->start));
5045 KASSERT(cur == entry,
5046 ("map %p cannot find %jx",
5047 map, (uintmax_t)entry->start));
5051 max_left = vm_map_entry_max_free_left(entry, lbound);
5052 max_right = vm_map_entry_max_free_right(entry, ubound);
5053 KASSERT(entry->max_free == vm_size_max(max_left, max_right),
5054 ("map %p max = %jx, max_left = %jx, max_right = %jx", map,
5055 (uintmax_t)entry->max_free,
5056 (uintmax_t)max_left, (uintmax_t)max_right));
5059 KASSERT(prev->end <= entry->start,
5060 ("map %p prev->end = %jx, start = %jx", map,
5061 (uintmax_t)prev->end, (uintmax_t)entry->start));
5065 #include "opt_ddb.h"
5067 #include <sys/kernel.h>
5069 #include <ddb/ddb.h>
5072 vm_map_print(vm_map_t map)
5074 vm_map_entry_t entry, prev;
5076 db_iprintf("Task map %p: pmap=%p, nentries=%d, version=%u\n",
5078 (void *)map->pmap, map->nentries, map->timestamp);
5081 prev = &map->header;
5082 VM_MAP_ENTRY_FOREACH(entry, map) {
5083 db_iprintf("map entry %p: start=%p, end=%p, eflags=%#x, \n",
5084 (void *)entry, (void *)entry->start, (void *)entry->end,
5087 static char *inheritance_name[4] =
5088 {"share", "copy", "none", "donate_copy"};
5090 db_iprintf(" prot=%x/%x/%s",
5092 entry->max_protection,
5093 inheritance_name[(int)(unsigned char)
5094 entry->inheritance]);
5095 if (entry->wired_count != 0)
5096 db_printf(", wired");
5098 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
5099 db_printf(", share=%p, offset=0x%jx\n",
5100 (void *)entry->object.sub_map,
5101 (uintmax_t)entry->offset);
5102 if (prev == &map->header ||
5103 prev->object.sub_map !=
5104 entry->object.sub_map) {
5106 vm_map_print((vm_map_t)entry->object.sub_map);
5110 if (entry->cred != NULL)
5111 db_printf(", ruid %d", entry->cred->cr_ruid);
5112 db_printf(", object=%p, offset=0x%jx",
5113 (void *)entry->object.vm_object,
5114 (uintmax_t)entry->offset);
5115 if (entry->object.vm_object && entry->object.vm_object->cred)
5116 db_printf(", obj ruid %d charge %jx",
5117 entry->object.vm_object->cred->cr_ruid,
5118 (uintmax_t)entry->object.vm_object->charge);
5119 if (entry->eflags & MAP_ENTRY_COW)
5120 db_printf(", copy (%s)",
5121 (entry->eflags & MAP_ENTRY_NEEDS_COPY) ? "needed" : "done");
5124 if (prev == &map->header ||
5125 prev->object.vm_object !=
5126 entry->object.vm_object) {
5128 vm_object_print((db_expr_t)(intptr_t)
5129 entry->object.vm_object,
5139 DB_SHOW_COMMAND(map, map)
5143 db_printf("usage: show map <addr>\n");
5146 vm_map_print((vm_map_t)addr);
5149 DB_SHOW_COMMAND(procvm, procvm)
5154 p = db_lookup_proc(addr);
5159 db_printf("p = %p, vmspace = %p, map = %p, pmap = %p\n",
5160 (void *)p, (void *)p->p_vmspace, (void *)&p->p_vmspace->vm_map,
5161 (void *)vmspace_pmap(p->p_vmspace));
5163 vm_map_print((vm_map_t)&p->p_vmspace->vm_map);