2 * SPDX-License-Identifier: (BSD-3-Clause AND MIT-CMU)
4 * Copyright (c) 1991, 1993
5 * The Regents of the University of California. All rights reserved.
7 * This code is derived from software contributed to Berkeley by
8 * The Mach Operating System project at Carnegie-Mellon University.
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 3. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * from: @(#)vm_map.c 8.3 (Berkeley) 1/12/94
37 * Copyright (c) 1987, 1990 Carnegie-Mellon University.
38 * All rights reserved.
40 * Authors: Avadis Tevanian, Jr., Michael Wayne Young
42 * Permission to use, copy, modify and distribute this software and
43 * its documentation is hereby granted, provided that both the copyright
44 * notice and this permission notice appear in all copies of the
45 * software, derivative works or modified versions, and any portions
46 * thereof, and that both notices appear in supporting documentation.
48 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
49 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
50 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
52 * Carnegie Mellon requests users of this software to return to
54 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
55 * School of Computer Science
56 * Carnegie Mellon University
57 * Pittsburgh PA 15213-3890
59 * any improvements or extensions that they make and grant Carnegie the
60 * rights to redistribute these changes.
64 * Virtual memory mapping module.
67 #include <sys/cdefs.h>
68 __FBSDID("$FreeBSD$");
70 #include <sys/param.h>
71 #include <sys/systm.h>
72 #include <sys/kernel.h>
75 #include <sys/mutex.h>
77 #include <sys/vmmeter.h>
79 #include <sys/vnode.h>
80 #include <sys/racct.h>
81 #include <sys/resourcevar.h>
82 #include <sys/rwlock.h>
84 #include <sys/sysctl.h>
85 #include <sys/sysent.h>
89 #include <vm/vm_param.h>
91 #include <vm/vm_map.h>
92 #include <vm/vm_page.h>
93 #include <vm/vm_pageout.h>
94 #include <vm/vm_object.h>
95 #include <vm/vm_pager.h>
96 #include <vm/vm_kern.h>
97 #include <vm/vm_extern.h>
98 #include <vm/vnode_pager.h>
99 #include <vm/swap_pager.h>
103 * Virtual memory maps provide for the mapping, protection,
104 * and sharing of virtual memory objects. In addition,
105 * this module provides for an efficient virtual copy of
106 * memory from one map to another.
108 * Synchronization is required prior to most operations.
110 * Maps consist of an ordered doubly-linked list of simple
111 * entries; a self-adjusting binary search tree of these
112 * entries is used to speed up lookups.
114 * Since portions of maps are specified by start/end addresses,
115 * which may not align with existing map entries, all
116 * routines merely "clip" entries to these start/end values.
117 * [That is, an entry is split into two, bordering at a
118 * start or end value.] Note that these clippings may not
119 * always be necessary (as the two resulting entries are then
120 * not changed); however, the clipping is done for convenience.
122 * As mentioned above, virtual copy operations are performed
123 * by copying VM object references from one map to
124 * another, and then marking both regions as copy-on-write.
127 static struct mtx map_sleep_mtx;
128 static uma_zone_t mapentzone;
129 static uma_zone_t kmapentzone;
130 static uma_zone_t mapzone;
131 static uma_zone_t vmspace_zone;
132 static int vmspace_zinit(void *mem, int size, int flags);
133 static int vm_map_zinit(void *mem, int ize, int flags);
134 static void _vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min,
136 static void vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map);
137 static void vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry);
138 static void vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry);
139 static int vm_map_growstack(vm_map_t map, vm_offset_t addr,
140 vm_map_entry_t gap_entry);
141 static void vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot,
142 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags);
144 static void vm_map_zdtor(void *mem, int size, void *arg);
145 static void vmspace_zdtor(void *mem, int size, void *arg);
147 static int vm_map_stack_locked(vm_map_t map, vm_offset_t addrbos,
148 vm_size_t max_ssize, vm_size_t growsize, vm_prot_t prot, vm_prot_t max,
150 static void vm_map_wire_entry_failure(vm_map_t map, vm_map_entry_t entry,
151 vm_offset_t failed_addr);
153 #define ENTRY_CHARGED(e) ((e)->cred != NULL || \
154 ((e)->object.vm_object != NULL && (e)->object.vm_object->cred != NULL && \
155 !((e)->eflags & MAP_ENTRY_NEEDS_COPY)))
158 * PROC_VMSPACE_{UN,}LOCK() can be a noop as long as vmspaces are type
161 #define PROC_VMSPACE_LOCK(p) do { } while (0)
162 #define PROC_VMSPACE_UNLOCK(p) do { } while (0)
165 * VM_MAP_RANGE_CHECK: [ internal use only ]
167 * Asserts that the starting and ending region
168 * addresses fall within the valid range of the map.
170 #define VM_MAP_RANGE_CHECK(map, start, end) \
172 if (start < vm_map_min(map)) \
173 start = vm_map_min(map); \
174 if (end > vm_map_max(map)) \
175 end = vm_map_max(map); \
183 * Initialize the vm_map module. Must be called before
184 * any other vm_map routines.
186 * Map and entry structures are allocated from the general
187 * purpose memory pool with some exceptions:
189 * - The kernel map and kmem submap are allocated statically.
190 * - Kernel map entries are allocated out of a static pool.
192 * These restrictions are necessary since malloc() uses the
193 * maps and requires map entries.
199 mtx_init(&map_sleep_mtx, "vm map sleep mutex", NULL, MTX_DEF);
200 mapzone = uma_zcreate("MAP", sizeof(struct vm_map), NULL,
206 vm_map_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
207 uma_prealloc(mapzone, MAX_KMAP);
208 kmapentzone = uma_zcreate("KMAP ENTRY", sizeof(struct vm_map_entry),
209 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR,
210 UMA_ZONE_MTXCLASS | UMA_ZONE_VM);
211 mapentzone = uma_zcreate("MAP ENTRY", sizeof(struct vm_map_entry),
212 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
213 vmspace_zone = uma_zcreate("VMSPACE", sizeof(struct vmspace), NULL,
219 vmspace_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
223 vmspace_zinit(void *mem, int size, int flags)
227 vm = (struct vmspace *)mem;
229 vm->vm_map.pmap = NULL;
230 (void)vm_map_zinit(&vm->vm_map, sizeof(vm->vm_map), flags);
231 PMAP_LOCK_INIT(vmspace_pmap(vm));
236 vm_map_zinit(void *mem, int size, int flags)
241 memset(map, 0, sizeof(*map));
242 mtx_init(&map->system_mtx, "vm map (system)", NULL, MTX_DEF | MTX_DUPOK);
243 sx_init(&map->lock, "vm map (user)");
249 vmspace_zdtor(void *mem, int size, void *arg)
253 vm = (struct vmspace *)mem;
255 vm_map_zdtor(&vm->vm_map, sizeof(vm->vm_map), arg);
258 vm_map_zdtor(void *mem, int size, void *arg)
263 KASSERT(map->nentries == 0,
264 ("map %p nentries == %d on free.",
265 map, map->nentries));
266 KASSERT(map->size == 0,
267 ("map %p size == %lu on free.",
268 map, (unsigned long)map->size));
270 #endif /* INVARIANTS */
273 * Allocate a vmspace structure, including a vm_map and pmap,
274 * and initialize those structures. The refcnt is set to 1.
276 * If 'pinit' is NULL then the embedded pmap is initialized via pmap_pinit().
279 vmspace_alloc(vm_offset_t min, vm_offset_t max, pmap_pinit_t pinit)
283 vm = uma_zalloc(vmspace_zone, M_WAITOK);
284 KASSERT(vm->vm_map.pmap == NULL, ("vm_map.pmap must be NULL"));
285 if (!pinit(vmspace_pmap(vm))) {
286 uma_zfree(vmspace_zone, vm);
289 CTR1(KTR_VM, "vmspace_alloc: %p", vm);
290 _vm_map_init(&vm->vm_map, vmspace_pmap(vm), min, max);
305 vmspace_container_reset(struct proc *p)
309 racct_set(p, RACCT_DATA, 0);
310 racct_set(p, RACCT_STACK, 0);
311 racct_set(p, RACCT_RSS, 0);
312 racct_set(p, RACCT_MEMLOCK, 0);
313 racct_set(p, RACCT_VMEM, 0);
319 vmspace_dofree(struct vmspace *vm)
322 CTR1(KTR_VM, "vmspace_free: %p", vm);
325 * Make sure any SysV shm is freed, it might not have been in
331 * Lock the map, to wait out all other references to it.
332 * Delete all of the mappings and pages they hold, then call
333 * the pmap module to reclaim anything left.
335 (void)vm_map_remove(&vm->vm_map, vm_map_min(&vm->vm_map),
336 vm_map_max(&vm->vm_map));
338 pmap_release(vmspace_pmap(vm));
339 vm->vm_map.pmap = NULL;
340 uma_zfree(vmspace_zone, vm);
344 vmspace_free(struct vmspace *vm)
347 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
348 "vmspace_free() called");
350 if (vm->vm_refcnt == 0)
351 panic("vmspace_free: attempt to free already freed vmspace");
353 if (atomic_fetchadd_int(&vm->vm_refcnt, -1) == 1)
358 vmspace_exitfree(struct proc *p)
362 PROC_VMSPACE_LOCK(p);
365 PROC_VMSPACE_UNLOCK(p);
366 KASSERT(vm == &vmspace0, ("vmspace_exitfree: wrong vmspace"));
371 vmspace_exit(struct thread *td)
378 * Release user portion of address space.
379 * This releases references to vnodes,
380 * which could cause I/O if the file has been unlinked.
381 * Need to do this early enough that we can still sleep.
383 * The last exiting process to reach this point releases as
384 * much of the environment as it can. vmspace_dofree() is the
385 * slower fallback in case another process had a temporary
386 * reference to the vmspace.
391 atomic_add_int(&vmspace0.vm_refcnt, 1);
392 refcnt = vm->vm_refcnt;
394 if (refcnt > 1 && p->p_vmspace != &vmspace0) {
395 /* Switch now since other proc might free vmspace */
396 PROC_VMSPACE_LOCK(p);
397 p->p_vmspace = &vmspace0;
398 PROC_VMSPACE_UNLOCK(p);
401 } while (!atomic_fcmpset_int(&vm->vm_refcnt, &refcnt, refcnt - 1));
403 if (p->p_vmspace != vm) {
404 /* vmspace not yet freed, switch back */
405 PROC_VMSPACE_LOCK(p);
407 PROC_VMSPACE_UNLOCK(p);
410 pmap_remove_pages(vmspace_pmap(vm));
411 /* Switch now since this proc will free vmspace */
412 PROC_VMSPACE_LOCK(p);
413 p->p_vmspace = &vmspace0;
414 PROC_VMSPACE_UNLOCK(p);
420 vmspace_container_reset(p);
424 /* Acquire reference to vmspace owned by another process. */
427 vmspace_acquire_ref(struct proc *p)
432 PROC_VMSPACE_LOCK(p);
435 PROC_VMSPACE_UNLOCK(p);
438 refcnt = vm->vm_refcnt;
440 if (refcnt <= 0) { /* Avoid 0->1 transition */
441 PROC_VMSPACE_UNLOCK(p);
444 } while (!atomic_fcmpset_int(&vm->vm_refcnt, &refcnt, refcnt + 1));
445 if (vm != p->p_vmspace) {
446 PROC_VMSPACE_UNLOCK(p);
450 PROC_VMSPACE_UNLOCK(p);
455 * Switch between vmspaces in an AIO kernel process.
457 * The new vmspace is either the vmspace of a user process obtained
458 * from an active AIO request or the initial vmspace of the AIO kernel
459 * process (when it is idling). Because user processes will block to
460 * drain any active AIO requests before proceeding in exit() or
461 * execve(), the reference count for vmspaces from AIO requests can
462 * never be 0. Similarly, AIO kernel processes hold an extra
463 * reference on their initial vmspace for the life of the process. As
464 * a result, the 'newvm' vmspace always has a non-zero reference
465 * count. This permits an additional reference on 'newvm' to be
466 * acquired via a simple atomic increment rather than the loop in
467 * vmspace_acquire_ref() above.
470 vmspace_switch_aio(struct vmspace *newvm)
472 struct vmspace *oldvm;
474 /* XXX: Need some way to assert that this is an aio daemon. */
476 KASSERT(newvm->vm_refcnt > 0,
477 ("vmspace_switch_aio: newvm unreferenced"));
479 oldvm = curproc->p_vmspace;
484 * Point to the new address space and refer to it.
486 curproc->p_vmspace = newvm;
487 atomic_add_int(&newvm->vm_refcnt, 1);
489 /* Activate the new mapping. */
490 pmap_activate(curthread);
496 _vm_map_lock(vm_map_t map, const char *file, int line)
500 mtx_lock_flags_(&map->system_mtx, 0, file, line);
502 sx_xlock_(&map->lock, file, line);
507 vm_map_entry_set_vnode_text(vm_map_entry_t entry, bool add)
509 vm_object_t object, object1;
512 if ((entry->eflags & MAP_ENTRY_VN_EXEC) == 0)
514 KASSERT((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0,
515 ("Submap with execs"));
516 object = entry->object.vm_object;
517 KASSERT(object != NULL, ("No object for text, entry %p", entry));
518 VM_OBJECT_RLOCK(object);
519 while ((object1 = object->backing_object) != NULL) {
520 VM_OBJECT_RLOCK(object1);
521 VM_OBJECT_RUNLOCK(object);
526 if (object->type == OBJT_DEAD) {
528 * For OBJT_DEAD objects, v_writecount was handled in
529 * vnode_pager_dealloc().
531 } else if (object->type == OBJT_VNODE) {
533 } else if (object->type == OBJT_SWAP) {
534 KASSERT((object->flags & OBJ_TMPFS_NODE) != 0,
535 ("vm_map_entry_set_vnode_text: swap and !TMPFS "
536 "entry %p, object %p, add %d", entry, object, add));
538 * Tmpfs VREG node, which was reclaimed, has
539 * OBJ_TMPFS_NODE flag set, but not OBJ_TMPFS. In
540 * this case there is no v_writecount to adjust.
542 if ((object->flags & OBJ_TMPFS) != 0)
543 vp = object->un_pager.swp.swp_tmpfs;
546 ("vm_map_entry_set_vnode_text: wrong object type, "
547 "entry %p, object %p, add %d", entry, object, add));
551 VOP_SET_TEXT_CHECKED(vp);
552 VM_OBJECT_RUNLOCK(object);
555 VM_OBJECT_RUNLOCK(object);
556 vn_lock(vp, LK_SHARED | LK_RETRY);
557 VOP_UNSET_TEXT_CHECKED(vp);
562 VM_OBJECT_RUNLOCK(object);
567 vm_map_process_deferred(void)
570 vm_map_entry_t entry, next;
574 entry = td->td_map_def_user;
575 td->td_map_def_user = NULL;
576 while (entry != NULL) {
578 MPASS((entry->eflags & (MAP_ENTRY_WRITECNT |
579 MAP_ENTRY_VN_EXEC)) != (MAP_ENTRY_WRITECNT |
581 if ((entry->eflags & MAP_ENTRY_WRITECNT) != 0) {
583 * Decrement the object's writemappings and
584 * possibly the vnode's v_writecount.
586 KASSERT((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0,
587 ("Submap with writecount"));
588 object = entry->object.vm_object;
589 KASSERT(object != NULL, ("No object for writecount"));
590 vm_pager_release_writecount(object, entry->start,
593 vm_map_entry_set_vnode_text(entry, false);
594 vm_map_entry_deallocate(entry, FALSE);
600 _vm_map_unlock(vm_map_t map, const char *file, int line)
604 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
606 sx_xunlock_(&map->lock, file, line);
607 vm_map_process_deferred();
612 _vm_map_lock_read(vm_map_t map, const char *file, int line)
616 mtx_lock_flags_(&map->system_mtx, 0, file, line);
618 sx_slock_(&map->lock, file, line);
622 _vm_map_unlock_read(vm_map_t map, const char *file, int line)
626 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
628 sx_sunlock_(&map->lock, file, line);
629 vm_map_process_deferred();
634 _vm_map_trylock(vm_map_t map, const char *file, int line)
638 error = map->system_map ?
639 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) :
640 !sx_try_xlock_(&map->lock, file, line);
647 _vm_map_trylock_read(vm_map_t map, const char *file, int line)
651 error = map->system_map ?
652 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) :
653 !sx_try_slock_(&map->lock, file, line);
658 * _vm_map_lock_upgrade: [ internal use only ]
660 * Tries to upgrade a read (shared) lock on the specified map to a write
661 * (exclusive) lock. Returns the value "0" if the upgrade succeeds and a
662 * non-zero value if the upgrade fails. If the upgrade fails, the map is
663 * returned without a read or write lock held.
665 * Requires that the map be read locked.
668 _vm_map_lock_upgrade(vm_map_t map, const char *file, int line)
670 unsigned int last_timestamp;
672 if (map->system_map) {
673 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
675 if (!sx_try_upgrade_(&map->lock, file, line)) {
676 last_timestamp = map->timestamp;
677 sx_sunlock_(&map->lock, file, line);
678 vm_map_process_deferred();
680 * If the map's timestamp does not change while the
681 * map is unlocked, then the upgrade succeeds.
683 sx_xlock_(&map->lock, file, line);
684 if (last_timestamp != map->timestamp) {
685 sx_xunlock_(&map->lock, file, line);
695 _vm_map_lock_downgrade(vm_map_t map, const char *file, int line)
698 if (map->system_map) {
699 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
701 sx_downgrade_(&map->lock, file, line);
707 * Returns a non-zero value if the caller holds a write (exclusive) lock
708 * on the specified map and the value "0" otherwise.
711 vm_map_locked(vm_map_t map)
715 return (mtx_owned(&map->system_mtx));
717 return (sx_xlocked(&map->lock));
722 _vm_map_assert_locked(vm_map_t map, const char *file, int line)
726 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
728 sx_assert_(&map->lock, SA_XLOCKED, file, line);
731 #define VM_MAP_ASSERT_LOCKED(map) \
732 _vm_map_assert_locked(map, LOCK_FILE, LOCK_LINE)
735 static int enable_vmmap_check = 1;
737 static int enable_vmmap_check = 0;
739 SYSCTL_INT(_debug, OID_AUTO, vmmap_check, CTLFLAG_RWTUN,
740 &enable_vmmap_check, 0, "Enable vm map consistency checking");
742 static void _vm_map_assert_consistent(vm_map_t map);
744 #define VM_MAP_ASSERT_CONSISTENT(map) \
745 _vm_map_assert_consistent(map)
747 #define VM_MAP_ASSERT_LOCKED(map)
748 #define VM_MAP_ASSERT_CONSISTENT(map)
749 #endif /* INVARIANTS */
752 * _vm_map_unlock_and_wait:
754 * Atomically releases the lock on the specified map and puts the calling
755 * thread to sleep. The calling thread will remain asleep until either
756 * vm_map_wakeup() is performed on the map or the specified timeout is
759 * WARNING! This function does not perform deferred deallocations of
760 * objects and map entries. Therefore, the calling thread is expected to
761 * reacquire the map lock after reawakening and later perform an ordinary
762 * unlock operation, such as vm_map_unlock(), before completing its
763 * operation on the map.
766 _vm_map_unlock_and_wait(vm_map_t map, int timo, const char *file, int line)
769 mtx_lock(&map_sleep_mtx);
771 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
773 sx_xunlock_(&map->lock, file, line);
774 return (msleep(&map->root, &map_sleep_mtx, PDROP | PVM, "vmmaps",
781 * Awaken any threads that have slept on the map using
782 * vm_map_unlock_and_wait().
785 vm_map_wakeup(vm_map_t map)
789 * Acquire and release map_sleep_mtx to prevent a wakeup()
790 * from being performed (and lost) between the map unlock
791 * and the msleep() in _vm_map_unlock_and_wait().
793 mtx_lock(&map_sleep_mtx);
794 mtx_unlock(&map_sleep_mtx);
799 vm_map_busy(vm_map_t map)
802 VM_MAP_ASSERT_LOCKED(map);
807 vm_map_unbusy(vm_map_t map)
810 VM_MAP_ASSERT_LOCKED(map);
811 KASSERT(map->busy, ("vm_map_unbusy: not busy"));
812 if (--map->busy == 0 && (map->flags & MAP_BUSY_WAKEUP)) {
813 vm_map_modflags(map, 0, MAP_BUSY_WAKEUP);
819 vm_map_wait_busy(vm_map_t map)
822 VM_MAP_ASSERT_LOCKED(map);
824 vm_map_modflags(map, MAP_BUSY_WAKEUP, 0);
826 msleep(&map->busy, &map->system_mtx, 0, "mbusy", 0);
828 sx_sleep(&map->busy, &map->lock, 0, "mbusy", 0);
834 vmspace_resident_count(struct vmspace *vmspace)
836 return pmap_resident_count(vmspace_pmap(vmspace));
842 * Creates and returns a new empty VM map with
843 * the given physical map structure, and having
844 * the given lower and upper address bounds.
847 vm_map_create(pmap_t pmap, vm_offset_t min, vm_offset_t max)
851 result = uma_zalloc(mapzone, M_WAITOK);
852 CTR1(KTR_VM, "vm_map_create: %p", result);
853 _vm_map_init(result, pmap, min, max);
858 * Initialize an existing vm_map structure
859 * such as that in the vmspace structure.
862 _vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max)
865 map->header.next = map->header.prev = &map->header;
866 map->header.eflags = MAP_ENTRY_HEADER;
867 map->needs_wakeup = FALSE;
870 map->header.end = min;
871 map->header.start = max;
880 vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max)
883 _vm_map_init(map, pmap, min, max);
884 mtx_init(&map->system_mtx, "system map", NULL, MTX_DEF | MTX_DUPOK);
885 sx_init(&map->lock, "user map");
889 * vm_map_entry_dispose: [ internal use only ]
891 * Inverse of vm_map_entry_create.
894 vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry)
896 uma_zfree(map->system_map ? kmapentzone : mapentzone, entry);
900 * vm_map_entry_create: [ internal use only ]
902 * Allocates a VM map entry for insertion.
903 * No entry fields are filled in.
905 static vm_map_entry_t
906 vm_map_entry_create(vm_map_t map)
908 vm_map_entry_t new_entry;
911 new_entry = uma_zalloc(kmapentzone, M_NOWAIT);
913 new_entry = uma_zalloc(mapentzone, M_WAITOK);
914 if (new_entry == NULL)
915 panic("vm_map_entry_create: kernel resources exhausted");
920 * vm_map_entry_set_behavior:
922 * Set the expected access behavior, either normal, random, or
926 vm_map_entry_set_behavior(vm_map_entry_t entry, u_char behavior)
928 entry->eflags = (entry->eflags & ~MAP_ENTRY_BEHAV_MASK) |
929 (behavior & MAP_ENTRY_BEHAV_MASK);
933 * vm_map_entry_max_free_{left,right}:
935 * Compute the size of the largest free gap between two entries,
936 * one the root of a tree and the other the ancestor of that root
937 * that is the least or greatest ancestor found on the search path.
939 static inline vm_size_t
940 vm_map_entry_max_free_left(vm_map_entry_t root, vm_map_entry_t left_ancestor)
943 return (root->left != NULL ?
944 root->left->max_free : root->start - left_ancestor->end);
947 static inline vm_size_t
948 vm_map_entry_max_free_right(vm_map_entry_t root, vm_map_entry_t right_ancestor)
951 return (root->right != NULL ?
952 root->right->max_free : right_ancestor->start - root->end);
955 #define SPLAY_LEFT_STEP(root, y, rlist, test) do { \
956 vm_size_t max_free; \
959 * Infer root->right->max_free == root->max_free when \
960 * y->max_free < root->max_free || root->max_free == 0. \
961 * Otherwise, look right to find it. \
964 max_free = root->max_free; \
965 KASSERT(max_free >= vm_map_entry_max_free_right(root, rlist), \
966 ("%s: max_free invariant fails", __func__)); \
967 if (y == NULL ? max_free > 0 : max_free - 1 < y->max_free) \
968 max_free = vm_map_entry_max_free_right(root, rlist); \
969 if (y != NULL && (test)) { \
970 /* Rotate right and make y root. */ \
971 root->left = y->right; \
973 if (max_free < y->max_free) \
974 root->max_free = max_free = MAX(max_free, \
975 vm_map_entry_max_free_left(root, y)); \
979 /* Copy right->max_free. Put root on rlist. */ \
980 root->max_free = max_free; \
981 KASSERT(max_free == vm_map_entry_max_free_right(root, rlist), \
982 ("%s: max_free not copied from right", __func__)); \
983 root->left = rlist; \
988 #define SPLAY_RIGHT_STEP(root, y, llist, test) do { \
989 vm_size_t max_free; \
992 * Infer root->left->max_free == root->max_free when \
993 * y->max_free < root->max_free || root->max_free == 0. \
994 * Otherwise, look left to find it. \
997 max_free = root->max_free; \
998 KASSERT(max_free >= vm_map_entry_max_free_left(root, llist), \
999 ("%s: max_free invariant fails", __func__)); \
1000 if (y == NULL ? max_free > 0 : max_free - 1 < y->max_free) \
1001 max_free = vm_map_entry_max_free_left(root, llist); \
1002 if (y != NULL && (test)) { \
1003 /* Rotate left and make y root. */ \
1004 root->right = y->left; \
1006 if (max_free < y->max_free) \
1007 root->max_free = max_free = MAX(max_free, \
1008 vm_map_entry_max_free_right(root, y)); \
1012 /* Copy left->max_free. Put root on llist. */ \
1013 root->max_free = max_free; \
1014 KASSERT(max_free == vm_map_entry_max_free_left(root, llist), \
1015 ("%s: max_free not copied from left", __func__)); \
1016 root->right = llist; \
1022 * Walk down the tree until we find addr or a NULL pointer where addr would go,
1023 * breaking off left and right subtrees of nodes less than, or greater than
1024 * addr. Treat pointers to nodes with max_free < length as NULL pointers.
1025 * llist and rlist are the two sides in reverse order (bottom-up), with llist
1026 * linked by the right pointer and rlist linked by the left pointer in the
1027 * vm_map_entry, and both lists terminated by &map->header. This function, and
1028 * the subsequent call to vm_map_splay_merge, rely on the start and end address
1029 * values in &map->header.
1031 static vm_map_entry_t
1032 vm_map_splay_split(vm_map_t map, vm_offset_t addr, vm_size_t length,
1033 vm_map_entry_t *out_llist, vm_map_entry_t *out_rlist)
1035 vm_map_entry_t llist, rlist, root, y;
1037 llist = rlist = &map->header;
1039 while (root != NULL && root->max_free >= length) {
1040 KASSERT(llist->end <= root->start && root->end <= rlist->start,
1041 ("%s: root not within tree bounds", __func__));
1042 if (addr < root->start) {
1043 SPLAY_LEFT_STEP(root, y, rlist,
1044 y->max_free >= length && addr < y->start);
1045 } else if (addr >= root->end) {
1046 SPLAY_RIGHT_STEP(root, y, llist,
1047 y->max_free >= length && addr >= y->end);
1057 vm_map_splay_findnext(vm_map_entry_t root, vm_map_entry_t *iolist)
1059 vm_map_entry_t rlist, y;
1063 while (root != NULL)
1064 SPLAY_LEFT_STEP(root, y, rlist, true);
1069 vm_map_splay_findprev(vm_map_entry_t root, vm_map_entry_t *iolist)
1071 vm_map_entry_t llist, y;
1075 while (root != NULL)
1076 SPLAY_RIGHT_STEP(root, y, llist, true);
1081 vm_map_entry_swap(vm_map_entry_t *a, vm_map_entry_t *b)
1091 * Walk back up the two spines, flip the pointers and set max_free. The
1092 * subtrees of the root go at the bottom of llist and rlist.
1095 vm_map_splay_merge(vm_map_t map, vm_map_entry_t root,
1096 vm_map_entry_t llist, vm_map_entry_t rlist)
1098 vm_map_entry_t prev;
1099 vm_size_t max_free_left, max_free_right;
1101 max_free_left = vm_map_entry_max_free_left(root, llist);
1102 if (llist != &map->header) {
1106 * The max_free values of the children of llist are in
1107 * llist->max_free and max_free_left. Update with the
1110 llist->max_free = max_free_left =
1111 MAX(llist->max_free, max_free_left);
1112 vm_map_entry_swap(&llist->right, &prev);
1113 vm_map_entry_swap(&prev, &llist);
1114 } while (llist != &map->header);
1117 max_free_right = vm_map_entry_max_free_right(root, rlist);
1118 if (rlist != &map->header) {
1122 * The max_free values of the children of rlist are in
1123 * rlist->max_free and max_free_right. Update with the
1126 rlist->max_free = max_free_right =
1127 MAX(rlist->max_free, max_free_right);
1128 vm_map_entry_swap(&rlist->left, &prev);
1129 vm_map_entry_swap(&prev, &rlist);
1130 } while (rlist != &map->header);
1133 root->max_free = MAX(max_free_left, max_free_right);
1140 * The Sleator and Tarjan top-down splay algorithm with the
1141 * following variation. Max_free must be computed bottom-up, so
1142 * on the downward pass, maintain the left and right spines in
1143 * reverse order. Then, make a second pass up each side to fix
1144 * the pointers and compute max_free. The time bound is O(log n)
1147 * The new root is the vm_map_entry containing "addr", or else an
1148 * adjacent entry (lower if possible) if addr is not in the tree.
1150 * The map must be locked, and leaves it so.
1152 * Returns: the new root.
1154 static vm_map_entry_t
1155 vm_map_splay(vm_map_t map, vm_offset_t addr)
1157 vm_map_entry_t llist, rlist, root;
1159 root = vm_map_splay_split(map, addr, 0, &llist, &rlist);
1162 } else if (llist != &map->header) {
1164 * Recover the greatest node in the left
1165 * subtree and make it the root.
1168 llist = root->right;
1170 } else if (rlist != &map->header) {
1172 * Recover the least node in the right
1173 * subtree and make it the root.
1179 /* There is no root. */
1182 vm_map_splay_merge(map, root, llist, rlist);
1183 VM_MAP_ASSERT_CONSISTENT(map);
1188 * vm_map_entry_{un,}link:
1190 * Insert/remove entries from maps.
1193 vm_map_entry_link(vm_map_t map, vm_map_entry_t entry)
1195 vm_map_entry_t llist, rlist, root;
1198 "vm_map_entry_link: map %p, nentries %d, entry %p", map,
1199 map->nentries, entry);
1200 VM_MAP_ASSERT_LOCKED(map);
1202 root = vm_map_splay_split(map, entry->start, 0, &llist, &rlist);
1203 KASSERT(root == NULL,
1204 ("vm_map_entry_link: link object already mapped"));
1205 entry->prev = llist;
1206 entry->next = rlist;
1207 llist->next = rlist->prev = entry;
1208 entry->left = entry->right = NULL;
1209 vm_map_splay_merge(map, entry, llist, rlist);
1210 VM_MAP_ASSERT_CONSISTENT(map);
1213 enum unlink_merge_type {
1219 vm_map_entry_unlink(vm_map_t map, vm_map_entry_t entry,
1220 enum unlink_merge_type op)
1222 vm_map_entry_t llist, rlist, root, y;
1224 VM_MAP_ASSERT_LOCKED(map);
1225 root = vm_map_splay_split(map, entry->start, 0, &llist, &rlist);
1226 KASSERT(root != NULL,
1227 ("vm_map_entry_unlink: unlink object not mapped"));
1229 vm_map_splay_findnext(root, &rlist);
1231 case UNLINK_MERGE_NEXT:
1232 rlist->start = root->start;
1233 rlist->offset = root->offset;
1239 case UNLINK_MERGE_NONE:
1240 vm_map_splay_findprev(root, &llist);
1241 if (llist != &map->header) {
1243 llist = root->right;
1245 } else if (rlist != &map->header) {
1254 y->prev = entry->prev;
1257 vm_map_splay_merge(map, root, llist, rlist);
1260 VM_MAP_ASSERT_CONSISTENT(map);
1262 CTR3(KTR_VM, "vm_map_entry_unlink: map %p, nentries %d, entry %p", map,
1263 map->nentries, entry);
1267 * vm_map_entry_resize:
1269 * Resize a vm_map_entry, recompute the amount of free space that
1270 * follows it and propagate that value up the tree.
1272 * The map must be locked, and leaves it so.
1275 vm_map_entry_resize(vm_map_t map, vm_map_entry_t entry, vm_size_t grow_amount)
1277 vm_map_entry_t llist, rlist, root;
1279 VM_MAP_ASSERT_LOCKED(map);
1280 root = vm_map_splay_split(map, entry->start, 0, &llist, &rlist);
1281 KASSERT(root != NULL,
1282 ("%s: resize object not mapped", __func__));
1283 vm_map_splay_findnext(root, &rlist);
1285 entry->end += grow_amount;
1286 vm_map_splay_merge(map, root, llist, rlist);
1287 VM_MAP_ASSERT_CONSISTENT(map);
1288 CTR4(KTR_VM, "%s: map %p, nentries %d, entry %p",
1289 __func__, map, map->nentries, entry);
1293 * vm_map_lookup_entry: [ internal use only ]
1295 * Finds the map entry containing (or
1296 * immediately preceding) the specified address
1297 * in the given map; the entry is returned
1298 * in the "entry" parameter. The boolean
1299 * result indicates whether the address is
1300 * actually contained in the map.
1303 vm_map_lookup_entry(
1305 vm_offset_t address,
1306 vm_map_entry_t *entry) /* OUT */
1308 vm_map_entry_t cur, lbound;
1312 * If the map is empty, then the map entry immediately preceding
1313 * "address" is the map's header.
1317 *entry = &map->header;
1320 if (address >= cur->start && cur->end > address) {
1324 if ((locked = vm_map_locked(map)) ||
1325 sx_try_upgrade(&map->lock)) {
1327 * Splay requires a write lock on the map. However, it only
1328 * restructures the binary search tree; it does not otherwise
1329 * change the map. Thus, the map's timestamp need not change
1330 * on a temporary upgrade.
1332 cur = vm_map_splay(map, address);
1334 sx_downgrade(&map->lock);
1337 * If "address" is contained within a map entry, the new root
1338 * is that map entry. Otherwise, the new root is a map entry
1339 * immediately before or after "address".
1341 if (address < cur->start) {
1342 *entry = &map->header;
1346 return (address < cur->end);
1349 * Since the map is only locked for read access, perform a
1350 * standard binary search tree lookup for "address".
1352 lbound = &map->header;
1354 if (address < cur->start) {
1356 } else if (cur->end <= address) {
1363 } while (cur != NULL);
1371 * Inserts the given whole VM object into the target
1372 * map at the specified address range. The object's
1373 * size should match that of the address range.
1375 * Requires that the map be locked, and leaves it so.
1377 * If object is non-NULL, ref count must be bumped by caller
1378 * prior to making call to account for the new entry.
1381 vm_map_insert(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1382 vm_offset_t start, vm_offset_t end, vm_prot_t prot, vm_prot_t max, int cow)
1384 vm_map_entry_t new_entry, prev_entry;
1386 vm_eflags_t protoeflags;
1387 vm_inherit_t inheritance;
1389 VM_MAP_ASSERT_LOCKED(map);
1390 KASSERT(object != kernel_object ||
1391 (cow & MAP_COPY_ON_WRITE) == 0,
1392 ("vm_map_insert: kernel object and COW"));
1393 KASSERT(object == NULL || (cow & MAP_NOFAULT) == 0,
1394 ("vm_map_insert: paradoxical MAP_NOFAULT request"));
1395 KASSERT((prot & ~max) == 0,
1396 ("prot %#x is not subset of max_prot %#x", prot, max));
1399 * Check that the start and end points are not bogus.
1401 if (start < vm_map_min(map) || end > vm_map_max(map) ||
1403 return (KERN_INVALID_ADDRESS);
1406 * Find the entry prior to the proposed starting address; if it's part
1407 * of an existing entry, this range is bogus.
1409 if (vm_map_lookup_entry(map, start, &prev_entry))
1410 return (KERN_NO_SPACE);
1413 * Assert that the next entry doesn't overlap the end point.
1415 if (prev_entry->next->start < end)
1416 return (KERN_NO_SPACE);
1418 if ((cow & MAP_CREATE_GUARD) != 0 && (object != NULL ||
1419 max != VM_PROT_NONE))
1420 return (KERN_INVALID_ARGUMENT);
1423 if (cow & MAP_COPY_ON_WRITE)
1424 protoeflags |= MAP_ENTRY_COW | MAP_ENTRY_NEEDS_COPY;
1425 if (cow & MAP_NOFAULT)
1426 protoeflags |= MAP_ENTRY_NOFAULT;
1427 if (cow & MAP_DISABLE_SYNCER)
1428 protoeflags |= MAP_ENTRY_NOSYNC;
1429 if (cow & MAP_DISABLE_COREDUMP)
1430 protoeflags |= MAP_ENTRY_NOCOREDUMP;
1431 if (cow & MAP_STACK_GROWS_DOWN)
1432 protoeflags |= MAP_ENTRY_GROWS_DOWN;
1433 if (cow & MAP_STACK_GROWS_UP)
1434 protoeflags |= MAP_ENTRY_GROWS_UP;
1435 if (cow & MAP_WRITECOUNT)
1436 protoeflags |= MAP_ENTRY_WRITECNT;
1437 if (cow & MAP_VN_EXEC)
1438 protoeflags |= MAP_ENTRY_VN_EXEC;
1439 if ((cow & MAP_CREATE_GUARD) != 0)
1440 protoeflags |= MAP_ENTRY_GUARD;
1441 if ((cow & MAP_CREATE_STACK_GAP_DN) != 0)
1442 protoeflags |= MAP_ENTRY_STACK_GAP_DN;
1443 if ((cow & MAP_CREATE_STACK_GAP_UP) != 0)
1444 protoeflags |= MAP_ENTRY_STACK_GAP_UP;
1445 if (cow & MAP_INHERIT_SHARE)
1446 inheritance = VM_INHERIT_SHARE;
1448 inheritance = VM_INHERIT_DEFAULT;
1451 if ((cow & (MAP_ACC_NO_CHARGE | MAP_NOFAULT | MAP_CREATE_GUARD)) != 0)
1453 if ((cow & MAP_ACC_CHARGED) || ((prot & VM_PROT_WRITE) &&
1454 ((protoeflags & MAP_ENTRY_NEEDS_COPY) || object == NULL))) {
1455 if (!(cow & MAP_ACC_CHARGED) && !swap_reserve(end - start))
1456 return (KERN_RESOURCE_SHORTAGE);
1457 KASSERT(object == NULL ||
1458 (protoeflags & MAP_ENTRY_NEEDS_COPY) != 0 ||
1459 object->cred == NULL,
1460 ("overcommit: vm_map_insert o %p", object));
1461 cred = curthread->td_ucred;
1465 /* Expand the kernel pmap, if necessary. */
1466 if (map == kernel_map && end > kernel_vm_end)
1467 pmap_growkernel(end);
1468 if (object != NULL) {
1470 * OBJ_ONEMAPPING must be cleared unless this mapping
1471 * is trivially proven to be the only mapping for any
1472 * of the object's pages. (Object granularity
1473 * reference counting is insufficient to recognize
1474 * aliases with precision.)
1476 VM_OBJECT_WLOCK(object);
1477 if (object->ref_count > 1 || object->shadow_count != 0)
1478 vm_object_clear_flag(object, OBJ_ONEMAPPING);
1479 VM_OBJECT_WUNLOCK(object);
1480 } else if ((prev_entry->eflags & ~MAP_ENTRY_USER_WIRED) ==
1482 (cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP |
1483 MAP_VN_EXEC)) == 0 &&
1484 prev_entry->end == start && (prev_entry->cred == cred ||
1485 (prev_entry->object.vm_object != NULL &&
1486 prev_entry->object.vm_object->cred == cred)) &&
1487 vm_object_coalesce(prev_entry->object.vm_object,
1489 (vm_size_t)(prev_entry->end - prev_entry->start),
1490 (vm_size_t)(end - prev_entry->end), cred != NULL &&
1491 (protoeflags & MAP_ENTRY_NEEDS_COPY) == 0)) {
1493 * We were able to extend the object. Determine if we
1494 * can extend the previous map entry to include the
1495 * new range as well.
1497 if (prev_entry->inheritance == inheritance &&
1498 prev_entry->protection == prot &&
1499 prev_entry->max_protection == max &&
1500 prev_entry->wired_count == 0) {
1501 KASSERT((prev_entry->eflags & MAP_ENTRY_USER_WIRED) ==
1502 0, ("prev_entry %p has incoherent wiring",
1504 if ((prev_entry->eflags & MAP_ENTRY_GUARD) == 0)
1505 map->size += end - prev_entry->end;
1506 vm_map_entry_resize(map, prev_entry,
1507 end - prev_entry->end);
1508 vm_map_try_merge_entries(map, prev_entry, prev_entry->next);
1509 return (KERN_SUCCESS);
1513 * If we can extend the object but cannot extend the
1514 * map entry, we have to create a new map entry. We
1515 * must bump the ref count on the extended object to
1516 * account for it. object may be NULL.
1518 object = prev_entry->object.vm_object;
1519 offset = prev_entry->offset +
1520 (prev_entry->end - prev_entry->start);
1521 vm_object_reference(object);
1522 if (cred != NULL && object != NULL && object->cred != NULL &&
1523 !(prev_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
1524 /* Object already accounts for this uid. */
1532 * Create a new entry
1534 new_entry = vm_map_entry_create(map);
1535 new_entry->start = start;
1536 new_entry->end = end;
1537 new_entry->cred = NULL;
1539 new_entry->eflags = protoeflags;
1540 new_entry->object.vm_object = object;
1541 new_entry->offset = offset;
1543 new_entry->inheritance = inheritance;
1544 new_entry->protection = prot;
1545 new_entry->max_protection = max;
1546 new_entry->wired_count = 0;
1547 new_entry->wiring_thread = NULL;
1548 new_entry->read_ahead = VM_FAULT_READ_AHEAD_INIT;
1549 new_entry->next_read = start;
1551 KASSERT(cred == NULL || !ENTRY_CHARGED(new_entry),
1552 ("overcommit: vm_map_insert leaks vm_map %p", new_entry));
1553 new_entry->cred = cred;
1556 * Insert the new entry into the list
1558 vm_map_entry_link(map, new_entry);
1559 if ((new_entry->eflags & MAP_ENTRY_GUARD) == 0)
1560 map->size += new_entry->end - new_entry->start;
1563 * Try to coalesce the new entry with both the previous and next
1564 * entries in the list. Previously, we only attempted to coalesce
1565 * with the previous entry when object is NULL. Here, we handle the
1566 * other cases, which are less common.
1568 vm_map_try_merge_entries(map, prev_entry, new_entry);
1569 vm_map_try_merge_entries(map, new_entry, new_entry->next);
1571 if ((cow & (MAP_PREFAULT | MAP_PREFAULT_PARTIAL)) != 0) {
1572 vm_map_pmap_enter(map, start, prot, object, OFF_TO_IDX(offset),
1573 end - start, cow & MAP_PREFAULT_PARTIAL);
1576 return (KERN_SUCCESS);
1582 * Find the first fit (lowest VM address) for "length" free bytes
1583 * beginning at address >= start in the given map.
1585 * In a vm_map_entry, "max_free" is the maximum amount of
1586 * contiguous free space between an entry in its subtree and a
1587 * neighbor of that entry. This allows finding a free region in
1588 * one path down the tree, so O(log n) amortized with splay
1591 * The map must be locked, and leaves it so.
1593 * Returns: starting address if sufficient space,
1594 * vm_map_max(map)-length+1 if insufficient space.
1597 vm_map_findspace(vm_map_t map, vm_offset_t start, vm_size_t length)
1599 vm_map_entry_t llist, rlist, root, y;
1600 vm_size_t left_length;
1601 vm_offset_t gap_end;
1604 * Request must fit within min/max VM address and must avoid
1607 start = MAX(start, vm_map_min(map));
1608 if (start >= vm_map_max(map) || length > vm_map_max(map) - start)
1609 return (vm_map_max(map) - length + 1);
1611 /* Empty tree means wide open address space. */
1612 if (map->root == NULL)
1616 * After splay_split, if start is within an entry, push it to the start
1617 * of the following gap. If rlist is at the end of the gap containing
1618 * start, save the end of that gap in gap_end to see if the gap is big
1619 * enough; otherwise set gap_end to start skip gap-checking and move
1620 * directly to a search of the right subtree.
1622 root = vm_map_splay_split(map, start, length, &llist, &rlist);
1623 gap_end = rlist->start;
1626 if (root->right != NULL)
1628 } else if (rlist != &map->header) {
1634 llist = root->right;
1637 vm_map_splay_merge(map, root, llist, rlist);
1638 VM_MAP_ASSERT_CONSISTENT(map);
1639 if (length <= gap_end - start)
1642 /* With max_free, can immediately tell if no solution. */
1643 if (root->right == NULL || length > root->right->max_free)
1644 return (vm_map_max(map) - length + 1);
1647 * Splay for the least large-enough gap in the right subtree.
1649 llist = rlist = &map->header;
1650 for (left_length = 0;;
1651 left_length = vm_map_entry_max_free_left(root, llist)) {
1652 if (length <= left_length)
1653 SPLAY_LEFT_STEP(root, y, rlist,
1654 length <= vm_map_entry_max_free_left(y, llist));
1656 SPLAY_RIGHT_STEP(root, y, llist,
1657 length > vm_map_entry_max_free_left(y, root));
1662 llist = root->right;
1664 if (rlist != &map->header) {
1668 vm_map_splay_merge(map, y, &map->header, rlist);
1670 vm_map_entry_max_free_left(y, root),
1671 vm_map_entry_max_free_right(y, &map->header));
1674 vm_map_splay_merge(map, root, llist, &map->header);
1675 VM_MAP_ASSERT_CONSISTENT(map);
1680 vm_map_fixed(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1681 vm_offset_t start, vm_size_t length, vm_prot_t prot,
1682 vm_prot_t max, int cow)
1687 end = start + length;
1688 KASSERT((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 ||
1690 ("vm_map_fixed: non-NULL backing object for stack"));
1692 VM_MAP_RANGE_CHECK(map, start, end);
1693 if ((cow & MAP_CHECK_EXCL) == 0)
1694 vm_map_delete(map, start, end);
1695 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) != 0) {
1696 result = vm_map_stack_locked(map, start, length, sgrowsiz,
1699 result = vm_map_insert(map, object, offset, start, end,
1706 static const int aslr_pages_rnd_64[2] = {0x1000, 0x10};
1707 static const int aslr_pages_rnd_32[2] = {0x100, 0x4};
1709 static int cluster_anon = 1;
1710 SYSCTL_INT(_vm, OID_AUTO, cluster_anon, CTLFLAG_RW,
1712 "Cluster anonymous mappings: 0 = no, 1 = yes if no hint, 2 = always");
1715 clustering_anon_allowed(vm_offset_t addr)
1718 switch (cluster_anon) {
1729 static long aslr_restarts;
1730 SYSCTL_LONG(_vm, OID_AUTO, aslr_restarts, CTLFLAG_RD,
1732 "Number of aslr failures");
1734 #define MAP_32BIT_MAX_ADDR ((vm_offset_t)1 << 31)
1737 * Searches for the specified amount of free space in the given map with the
1738 * specified alignment. Performs an address-ordered, first-fit search from
1739 * the given address "*addr", with an optional upper bound "max_addr". If the
1740 * parameter "alignment" is zero, then the alignment is computed from the
1741 * given (object, offset) pair so as to enable the greatest possible use of
1742 * superpage mappings. Returns KERN_SUCCESS and the address of the free space
1743 * in "*addr" if successful. Otherwise, returns KERN_NO_SPACE.
1745 * The map must be locked. Initially, there must be at least "length" bytes
1746 * of free space at the given address.
1749 vm_map_alignspace(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1750 vm_offset_t *addr, vm_size_t length, vm_offset_t max_addr,
1751 vm_offset_t alignment)
1753 vm_offset_t aligned_addr, free_addr;
1755 VM_MAP_ASSERT_LOCKED(map);
1757 KASSERT(free_addr == vm_map_findspace(map, free_addr, length),
1758 ("caller failed to provide space %#jx at address %p",
1759 (uintmax_t)length, (void *)free_addr));
1762 * At the start of every iteration, the free space at address
1763 * "*addr" is at least "length" bytes.
1766 pmap_align_superpage(object, offset, addr, length);
1767 else if ((*addr & (alignment - 1)) != 0) {
1768 *addr &= ~(alignment - 1);
1771 aligned_addr = *addr;
1772 if (aligned_addr == free_addr) {
1774 * Alignment did not change "*addr", so "*addr" must
1775 * still provide sufficient free space.
1777 return (KERN_SUCCESS);
1781 * Test for address wrap on "*addr". A wrapped "*addr" could
1782 * be a valid address, in which case vm_map_findspace() cannot
1783 * be relied upon to fail.
1785 if (aligned_addr < free_addr)
1786 return (KERN_NO_SPACE);
1787 *addr = vm_map_findspace(map, aligned_addr, length);
1788 if (*addr + length > vm_map_max(map) ||
1789 (max_addr != 0 && *addr + length > max_addr))
1790 return (KERN_NO_SPACE);
1792 if (free_addr == aligned_addr) {
1794 * If a successful call to vm_map_findspace() did not
1795 * change "*addr", then "*addr" must still be aligned
1796 * and provide sufficient free space.
1798 return (KERN_SUCCESS);
1804 * vm_map_find finds an unallocated region in the target address
1805 * map with the given length. The search is defined to be
1806 * first-fit from the specified address; the region found is
1807 * returned in the same parameter.
1809 * If object is non-NULL, ref count must be bumped by caller
1810 * prior to making call to account for the new entry.
1813 vm_map_find(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1814 vm_offset_t *addr, /* IN/OUT */
1815 vm_size_t length, vm_offset_t max_addr, int find_space,
1816 vm_prot_t prot, vm_prot_t max, int cow)
1818 vm_offset_t alignment, curr_min_addr, min_addr;
1819 int gap, pidx, rv, try;
1820 bool cluster, en_aslr, update_anon;
1822 KASSERT((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 ||
1824 ("vm_map_find: non-NULL backing object for stack"));
1825 MPASS((cow & MAP_REMAP) == 0 || (find_space == VMFS_NO_SPACE &&
1826 (cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0));
1827 if (find_space == VMFS_OPTIMAL_SPACE && (object == NULL ||
1828 (object->flags & OBJ_COLORED) == 0))
1829 find_space = VMFS_ANY_SPACE;
1830 if (find_space >> 8 != 0) {
1831 KASSERT((find_space & 0xff) == 0, ("bad VMFS flags"));
1832 alignment = (vm_offset_t)1 << (find_space >> 8);
1835 en_aslr = (map->flags & MAP_ASLR) != 0;
1836 update_anon = cluster = clustering_anon_allowed(*addr) &&
1837 (map->flags & MAP_IS_SUB_MAP) == 0 && max_addr == 0 &&
1838 find_space != VMFS_NO_SPACE && object == NULL &&
1839 (cow & (MAP_INHERIT_SHARE | MAP_STACK_GROWS_UP |
1840 MAP_STACK_GROWS_DOWN)) == 0 && prot != PROT_NONE;
1841 curr_min_addr = min_addr = *addr;
1842 if (en_aslr && min_addr == 0 && !cluster &&
1843 find_space != VMFS_NO_SPACE &&
1844 (map->flags & MAP_ASLR_IGNSTART) != 0)
1845 curr_min_addr = min_addr = vm_map_min(map);
1849 curr_min_addr = map->anon_loc;
1850 if (curr_min_addr == 0)
1853 if (find_space != VMFS_NO_SPACE) {
1854 KASSERT(find_space == VMFS_ANY_SPACE ||
1855 find_space == VMFS_OPTIMAL_SPACE ||
1856 find_space == VMFS_SUPER_SPACE ||
1857 alignment != 0, ("unexpected VMFS flag"));
1860 * When creating an anonymous mapping, try clustering
1861 * with an existing anonymous mapping first.
1863 * We make up to two attempts to find address space
1864 * for a given find_space value. The first attempt may
1865 * apply randomization or may cluster with an existing
1866 * anonymous mapping. If this first attempt fails,
1867 * perform a first-fit search of the available address
1870 * If all tries failed, and find_space is
1871 * VMFS_OPTIMAL_SPACE, fallback to VMFS_ANY_SPACE.
1872 * Again enable clustering and randomization.
1879 * Second try: we failed either to find a
1880 * suitable region for randomizing the
1881 * allocation, or to cluster with an existing
1882 * mapping. Retry with free run.
1884 curr_min_addr = (map->flags & MAP_ASLR_IGNSTART) != 0 ?
1885 vm_map_min(map) : min_addr;
1886 atomic_add_long(&aslr_restarts, 1);
1889 if (try == 1 && en_aslr && !cluster) {
1891 * Find space for allocation, including
1892 * gap needed for later randomization.
1894 pidx = MAXPAGESIZES > 1 && pagesizes[1] != 0 &&
1895 (find_space == VMFS_SUPER_SPACE || find_space ==
1896 VMFS_OPTIMAL_SPACE) ? 1 : 0;
1897 gap = vm_map_max(map) > MAP_32BIT_MAX_ADDR &&
1898 (max_addr == 0 || max_addr > MAP_32BIT_MAX_ADDR) ?
1899 aslr_pages_rnd_64[pidx] : aslr_pages_rnd_32[pidx];
1900 *addr = vm_map_findspace(map, curr_min_addr,
1901 length + gap * pagesizes[pidx]);
1902 if (*addr + length + gap * pagesizes[pidx] >
1905 /* And randomize the start address. */
1906 *addr += (arc4random() % gap) * pagesizes[pidx];
1907 if (max_addr != 0 && *addr + length > max_addr)
1910 *addr = vm_map_findspace(map, curr_min_addr, length);
1911 if (*addr + length > vm_map_max(map) ||
1912 (max_addr != 0 && *addr + length > max_addr)) {
1923 if (find_space != VMFS_ANY_SPACE &&
1924 (rv = vm_map_alignspace(map, object, offset, addr, length,
1925 max_addr, alignment)) != KERN_SUCCESS) {
1926 if (find_space == VMFS_OPTIMAL_SPACE) {
1927 find_space = VMFS_ANY_SPACE;
1928 curr_min_addr = min_addr;
1929 cluster = update_anon;
1935 } else if ((cow & MAP_REMAP) != 0) {
1936 if (*addr < vm_map_min(map) ||
1937 *addr + length > vm_map_max(map) ||
1938 *addr + length <= length) {
1939 rv = KERN_INVALID_ADDRESS;
1942 vm_map_delete(map, *addr, *addr + length);
1944 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) != 0) {
1945 rv = vm_map_stack_locked(map, *addr, length, sgrowsiz, prot,
1948 rv = vm_map_insert(map, object, offset, *addr, *addr + length,
1951 if (rv == KERN_SUCCESS && update_anon)
1952 map->anon_loc = *addr + length;
1959 * vm_map_find_min() is a variant of vm_map_find() that takes an
1960 * additional parameter (min_addr) and treats the given address
1961 * (*addr) differently. Specifically, it treats *addr as a hint
1962 * and not as the minimum address where the mapping is created.
1964 * This function works in two phases. First, it tries to
1965 * allocate above the hint. If that fails and the hint is
1966 * greater than min_addr, it performs a second pass, replacing
1967 * the hint with min_addr as the minimum address for the
1971 vm_map_find_min(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1972 vm_offset_t *addr, vm_size_t length, vm_offset_t min_addr,
1973 vm_offset_t max_addr, int find_space, vm_prot_t prot, vm_prot_t max,
1981 rv = vm_map_find(map, object, offset, addr, length, max_addr,
1982 find_space, prot, max, cow);
1983 if (rv == KERN_SUCCESS || min_addr >= hint)
1985 *addr = hint = min_addr;
1990 * A map entry with any of the following flags set must not be merged with
1993 #define MAP_ENTRY_NOMERGE_MASK (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP | \
1994 MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_IS_SUB_MAP | MAP_ENTRY_VN_EXEC)
1997 vm_map_mergeable_neighbors(vm_map_entry_t prev, vm_map_entry_t entry)
2000 KASSERT((prev->eflags & MAP_ENTRY_NOMERGE_MASK) == 0 ||
2001 (entry->eflags & MAP_ENTRY_NOMERGE_MASK) == 0,
2002 ("vm_map_mergeable_neighbors: neither %p nor %p are mergeable",
2004 return (prev->end == entry->start &&
2005 prev->object.vm_object == entry->object.vm_object &&
2006 (prev->object.vm_object == NULL ||
2007 prev->offset + (prev->end - prev->start) == entry->offset) &&
2008 prev->eflags == entry->eflags &&
2009 prev->protection == entry->protection &&
2010 prev->max_protection == entry->max_protection &&
2011 prev->inheritance == entry->inheritance &&
2012 prev->wired_count == entry->wired_count &&
2013 prev->cred == entry->cred);
2017 vm_map_merged_neighbor_dispose(vm_map_t map, vm_map_entry_t entry)
2021 * If the backing object is a vnode object, vm_object_deallocate()
2022 * calls vrele(). However, vrele() does not lock the vnode because
2023 * the vnode has additional references. Thus, the map lock can be
2024 * kept without causing a lock-order reversal with the vnode lock.
2026 * Since we count the number of virtual page mappings in
2027 * object->un_pager.vnp.writemappings, the writemappings value
2028 * should not be adjusted when the entry is disposed of.
2030 if (entry->object.vm_object != NULL)
2031 vm_object_deallocate(entry->object.vm_object);
2032 if (entry->cred != NULL)
2033 crfree(entry->cred);
2034 vm_map_entry_dispose(map, entry);
2038 * vm_map_try_merge_entries:
2040 * Compare the given map entry to its predecessor, and merge its precessor
2041 * into it if possible. The entry remains valid, and may be extended.
2042 * The predecessor may be deleted.
2044 * The map must be locked.
2047 vm_map_try_merge_entries(vm_map_t map, vm_map_entry_t prev, vm_map_entry_t entry)
2050 VM_MAP_ASSERT_LOCKED(map);
2051 if ((entry->eflags & MAP_ENTRY_NOMERGE_MASK) == 0 &&
2052 vm_map_mergeable_neighbors(prev, entry)) {
2053 vm_map_entry_unlink(map, prev, UNLINK_MERGE_NEXT);
2054 vm_map_merged_neighbor_dispose(map, prev);
2059 * vm_map_entry_back:
2061 * Allocate an object to back a map entry.
2064 vm_map_entry_back(vm_map_entry_t entry)
2068 KASSERT(entry->object.vm_object == NULL,
2069 ("map entry %p has backing object", entry));
2070 KASSERT((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0,
2071 ("map entry %p is a submap", entry));
2072 object = vm_object_allocate(OBJT_DEFAULT,
2073 atop(entry->end - entry->start));
2074 entry->object.vm_object = object;
2076 if (entry->cred != NULL) {
2077 object->cred = entry->cred;
2078 object->charge = entry->end - entry->start;
2084 * vm_map_entry_charge_object
2086 * If there is no object backing this entry, create one. Otherwise, if
2087 * the entry has cred, give it to the backing object.
2090 vm_map_entry_charge_object(vm_map_t map, vm_map_entry_t entry)
2093 VM_MAP_ASSERT_LOCKED(map);
2094 KASSERT((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0,
2095 ("map entry %p is a submap", entry));
2096 if (entry->object.vm_object == NULL && !map->system_map &&
2097 (entry->eflags & MAP_ENTRY_GUARD) == 0)
2098 vm_map_entry_back(entry);
2099 else if (entry->object.vm_object != NULL &&
2100 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) &&
2101 entry->cred != NULL) {
2102 VM_OBJECT_WLOCK(entry->object.vm_object);
2103 KASSERT(entry->object.vm_object->cred == NULL,
2104 ("OVERCOMMIT: %s: both cred e %p", __func__, entry));
2105 entry->object.vm_object->cred = entry->cred;
2106 entry->object.vm_object->charge = entry->end - entry->start;
2107 VM_OBJECT_WUNLOCK(entry->object.vm_object);
2113 * vm_map_clip_start: [ internal use only ]
2115 * Asserts that the given entry begins at or after
2116 * the specified address; if necessary,
2117 * it splits the entry into two.
2119 #define vm_map_clip_start(map, entry, startaddr) \
2121 if (startaddr > entry->start) \
2122 _vm_map_clip_start(map, entry, startaddr); \
2126 * This routine is called only when it is known that
2127 * the entry must be split.
2130 _vm_map_clip_start(vm_map_t map, vm_map_entry_t entry, vm_offset_t start)
2132 vm_map_entry_t new_entry;
2134 VM_MAP_ASSERT_LOCKED(map);
2135 KASSERT(entry->end > start && entry->start < start,
2136 ("_vm_map_clip_start: invalid clip of entry %p", entry));
2139 * Create a backing object now, if none exists, so that more individual
2140 * objects won't be created after the map entry is split.
2142 vm_map_entry_charge_object(map, entry);
2144 /* Clone the entry. */
2145 new_entry = vm_map_entry_create(map);
2146 *new_entry = *entry;
2149 * Split off the front portion. Insert the new entry BEFORE this one,
2150 * so that this entry has the specified starting address.
2152 new_entry->end = start;
2153 entry->offset += (start - entry->start);
2154 entry->start = start;
2155 if (new_entry->cred != NULL)
2156 crhold(entry->cred);
2158 vm_map_entry_link(map, new_entry);
2160 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
2161 vm_object_reference(new_entry->object.vm_object);
2162 vm_map_entry_set_vnode_text(new_entry, true);
2164 * The object->un_pager.vnp.writemappings for the
2165 * object of MAP_ENTRY_WRITECNT type entry shall be
2166 * kept as is here. The virtual pages are
2167 * re-distributed among the clipped entries, so the sum is
2174 * vm_map_clip_end: [ internal use only ]
2176 * Asserts that the given entry ends at or before
2177 * the specified address; if necessary,
2178 * it splits the entry into two.
2180 #define vm_map_clip_end(map, entry, endaddr) \
2182 if ((endaddr) < (entry->end)) \
2183 _vm_map_clip_end((map), (entry), (endaddr)); \
2187 * This routine is called only when it is known that
2188 * the entry must be split.
2191 _vm_map_clip_end(vm_map_t map, vm_map_entry_t entry, vm_offset_t end)
2193 vm_map_entry_t new_entry;
2195 VM_MAP_ASSERT_LOCKED(map);
2196 KASSERT(entry->start < end && entry->end > end,
2197 ("_vm_map_clip_end: invalid clip of entry %p", entry));
2200 * Create a backing object now, if none exists, so that more individual
2201 * objects won't be created after the map entry is split.
2203 vm_map_entry_charge_object(map, entry);
2205 /* Clone the entry. */
2206 new_entry = vm_map_entry_create(map);
2207 *new_entry = *entry;
2210 * Split off the back portion. Insert the new entry AFTER this one,
2211 * so that this entry has the specified ending address.
2213 new_entry->start = entry->end = end;
2214 new_entry->offset += (end - entry->start);
2215 if (new_entry->cred != NULL)
2216 crhold(entry->cred);
2218 vm_map_entry_link(map, new_entry);
2220 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
2221 vm_object_reference(new_entry->object.vm_object);
2222 vm_map_entry_set_vnode_text(new_entry, true);
2227 * vm_map_submap: [ kernel use only ]
2229 * Mark the given range as handled by a subordinate map.
2231 * This range must have been created with vm_map_find,
2232 * and no other operations may have been performed on this
2233 * range prior to calling vm_map_submap.
2235 * Only a limited number of operations can be performed
2236 * within this rage after calling vm_map_submap:
2238 * [Don't try vm_map_copy!]
2240 * To remove a submapping, one must first remove the
2241 * range from the superior map, and then destroy the
2242 * submap (if desired). [Better yet, don't try it.]
2251 vm_map_entry_t entry;
2254 result = KERN_INVALID_ARGUMENT;
2256 vm_map_lock(submap);
2257 submap->flags |= MAP_IS_SUB_MAP;
2258 vm_map_unlock(submap);
2262 VM_MAP_RANGE_CHECK(map, start, end);
2264 if (vm_map_lookup_entry(map, start, &entry)) {
2265 vm_map_clip_start(map, entry, start);
2267 entry = entry->next;
2269 vm_map_clip_end(map, entry, end);
2271 if ((entry->start == start) && (entry->end == end) &&
2272 ((entry->eflags & MAP_ENTRY_COW) == 0) &&
2273 (entry->object.vm_object == NULL)) {
2274 entry->object.sub_map = submap;
2275 entry->eflags |= MAP_ENTRY_IS_SUB_MAP;
2276 result = KERN_SUCCESS;
2280 if (result != KERN_SUCCESS) {
2281 vm_map_lock(submap);
2282 submap->flags &= ~MAP_IS_SUB_MAP;
2283 vm_map_unlock(submap);
2289 * The maximum number of pages to map if MAP_PREFAULT_PARTIAL is specified
2291 #define MAX_INIT_PT 96
2294 * vm_map_pmap_enter:
2296 * Preload the specified map's pmap with mappings to the specified
2297 * object's memory-resident pages. No further physical pages are
2298 * allocated, and no further virtual pages are retrieved from secondary
2299 * storage. If the specified flags include MAP_PREFAULT_PARTIAL, then a
2300 * limited number of page mappings are created at the low-end of the
2301 * specified address range. (For this purpose, a superpage mapping
2302 * counts as one page mapping.) Otherwise, all resident pages within
2303 * the specified address range are mapped.
2306 vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot,
2307 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags)
2310 vm_page_t p, p_start;
2311 vm_pindex_t mask, psize, threshold, tmpidx;
2313 if ((prot & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0 || object == NULL)
2315 VM_OBJECT_RLOCK(object);
2316 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) {
2317 VM_OBJECT_RUNLOCK(object);
2318 VM_OBJECT_WLOCK(object);
2319 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) {
2320 pmap_object_init_pt(map->pmap, addr, object, pindex,
2322 VM_OBJECT_WUNLOCK(object);
2325 VM_OBJECT_LOCK_DOWNGRADE(object);
2329 if (psize + pindex > object->size) {
2330 if (object->size < pindex) {
2331 VM_OBJECT_RUNLOCK(object);
2334 psize = object->size - pindex;
2339 threshold = MAX_INIT_PT;
2341 p = vm_page_find_least(object, pindex);
2343 * Assert: the variable p is either (1) the page with the
2344 * least pindex greater than or equal to the parameter pindex
2348 p != NULL && (tmpidx = p->pindex - pindex) < psize;
2349 p = TAILQ_NEXT(p, listq)) {
2351 * don't allow an madvise to blow away our really
2352 * free pages allocating pv entries.
2354 if (((flags & MAP_PREFAULT_MADVISE) != 0 &&
2355 vm_page_count_severe()) ||
2356 ((flags & MAP_PREFAULT_PARTIAL) != 0 &&
2357 tmpidx >= threshold)) {
2361 if (vm_page_all_valid(p)) {
2362 if (p_start == NULL) {
2363 start = addr + ptoa(tmpidx);
2366 /* Jump ahead if a superpage mapping is possible. */
2367 if (p->psind > 0 && ((addr + ptoa(tmpidx)) &
2368 (pagesizes[p->psind] - 1)) == 0) {
2369 mask = atop(pagesizes[p->psind]) - 1;
2370 if (tmpidx + mask < psize &&
2371 vm_page_ps_test(p, PS_ALL_VALID, NULL)) {
2376 } else if (p_start != NULL) {
2377 pmap_enter_object(map->pmap, start, addr +
2378 ptoa(tmpidx), p_start, prot);
2382 if (p_start != NULL)
2383 pmap_enter_object(map->pmap, start, addr + ptoa(psize),
2385 VM_OBJECT_RUNLOCK(object);
2391 * Sets the protection of the specified address
2392 * region in the target map. If "set_max" is
2393 * specified, the maximum protection is to be set;
2394 * otherwise, only the current protection is affected.
2397 vm_map_protect(vm_map_t map, vm_offset_t start, vm_offset_t end,
2398 vm_prot_t new_prot, boolean_t set_max)
2400 vm_map_entry_t current, entry, in_tran;
2407 return (KERN_SUCCESS);
2414 * Ensure that we are not concurrently wiring pages. vm_map_wire() may
2415 * need to fault pages into the map and will drop the map lock while
2416 * doing so, and the VM object may end up in an inconsistent state if we
2417 * update the protection on the map entry in between faults.
2419 vm_map_wait_busy(map);
2421 VM_MAP_RANGE_CHECK(map, start, end);
2423 if (!vm_map_lookup_entry(map, start, &entry))
2424 entry = entry->next;
2427 * Make a first pass to check for protection violations.
2429 for (current = entry; current->start < end; current = current->next) {
2430 if ((current->eflags & MAP_ENTRY_GUARD) != 0)
2432 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
2434 return (KERN_INVALID_ARGUMENT);
2436 if ((new_prot & current->max_protection) != new_prot) {
2438 return (KERN_PROTECTION_FAILURE);
2440 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0)
2445 * Postpone the operation until all in transition map entries
2446 * are stabilized. In-transition entry might already have its
2447 * pages wired and wired_count incremented, but
2448 * MAP_ENTRY_USER_WIRED flag not yet set, and visible to other
2449 * threads because the map lock is dropped. In this case we
2450 * would miss our call to vm_fault_copy_entry().
2452 if (in_tran != NULL) {
2453 in_tran->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2454 vm_map_unlock_and_wait(map, 0);
2459 * Before changing the protections, try to reserve swap space for any
2460 * private (i.e., copy-on-write) mappings that are transitioning from
2461 * read-only to read/write access. If a reservation fails, break out
2462 * of this loop early and let the next loop simplify the entries, since
2463 * some may now be mergeable.
2466 vm_map_clip_start(map, entry, start);
2467 for (current = entry; current->start < end; current = current->next) {
2469 vm_map_clip_end(map, current, end);
2472 ((new_prot & ~(current->protection)) & VM_PROT_WRITE) == 0 ||
2473 ENTRY_CHARGED(current) ||
2474 (current->eflags & MAP_ENTRY_GUARD) != 0) {
2478 cred = curthread->td_ucred;
2479 obj = current->object.vm_object;
2481 if (obj == NULL || (current->eflags & MAP_ENTRY_NEEDS_COPY)) {
2482 if (!swap_reserve(current->end - current->start)) {
2483 rv = KERN_RESOURCE_SHORTAGE;
2488 current->cred = cred;
2492 VM_OBJECT_WLOCK(obj);
2493 if (obj->type != OBJT_DEFAULT && obj->type != OBJT_SWAP) {
2494 VM_OBJECT_WUNLOCK(obj);
2499 * Charge for the whole object allocation now, since
2500 * we cannot distinguish between non-charged and
2501 * charged clipped mapping of the same object later.
2503 KASSERT(obj->charge == 0,
2504 ("vm_map_protect: object %p overcharged (entry %p)",
2506 if (!swap_reserve(ptoa(obj->size))) {
2507 VM_OBJECT_WUNLOCK(obj);
2508 rv = KERN_RESOURCE_SHORTAGE;
2515 obj->charge = ptoa(obj->size);
2516 VM_OBJECT_WUNLOCK(obj);
2520 * If enough swap space was available, go back and fix up protections.
2521 * Otherwise, just simplify entries, since some may have been modified.
2522 * [Note that clipping is not necessary the second time.]
2524 for (current = entry; current->start < end;
2525 vm_map_try_merge_entries(map, current->prev, current),
2526 current = current->next) {
2527 if (rv != KERN_SUCCESS ||
2528 (current->eflags & MAP_ENTRY_GUARD) != 0)
2531 old_prot = current->protection;
2534 current->protection =
2535 (current->max_protection = new_prot) &
2538 current->protection = new_prot;
2541 * For user wired map entries, the normal lazy evaluation of
2542 * write access upgrades through soft page faults is
2543 * undesirable. Instead, immediately copy any pages that are
2544 * copy-on-write and enable write access in the physical map.
2546 if ((current->eflags & MAP_ENTRY_USER_WIRED) != 0 &&
2547 (current->protection & VM_PROT_WRITE) != 0 &&
2548 (old_prot & VM_PROT_WRITE) == 0)
2549 vm_fault_copy_entry(map, map, current, current, NULL);
2552 * When restricting access, update the physical map. Worry
2553 * about copy-on-write here.
2555 if ((old_prot & ~current->protection) != 0) {
2556 #define MASK(entry) (((entry)->eflags & MAP_ENTRY_COW) ? ~VM_PROT_WRITE : \
2558 pmap_protect(map->pmap, current->start,
2560 current->protection & MASK(current));
2564 vm_map_try_merge_entries(map, current->prev, current);
2572 * This routine traverses a processes map handling the madvise
2573 * system call. Advisories are classified as either those effecting
2574 * the vm_map_entry structure, or those effecting the underlying
2584 vm_map_entry_t current, entry;
2588 * Some madvise calls directly modify the vm_map_entry, in which case
2589 * we need to use an exclusive lock on the map and we need to perform
2590 * various clipping operations. Otherwise we only need a read-lock
2595 case MADV_SEQUENTIAL:
2612 vm_map_lock_read(map);
2619 * Locate starting entry and clip if necessary.
2621 VM_MAP_RANGE_CHECK(map, start, end);
2623 if (vm_map_lookup_entry(map, start, &entry)) {
2625 vm_map_clip_start(map, entry, start);
2627 entry = entry->next;
2632 * madvise behaviors that are implemented in the vm_map_entry.
2634 * We clip the vm_map_entry so that behavioral changes are
2635 * limited to the specified address range.
2637 for (current = entry; current->start < end;
2638 current = current->next) {
2639 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
2642 vm_map_clip_end(map, current, end);
2646 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_NORMAL);
2648 case MADV_SEQUENTIAL:
2649 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_SEQUENTIAL);
2652 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_RANDOM);
2655 current->eflags |= MAP_ENTRY_NOSYNC;
2658 current->eflags &= ~MAP_ENTRY_NOSYNC;
2661 current->eflags |= MAP_ENTRY_NOCOREDUMP;
2664 current->eflags &= ~MAP_ENTRY_NOCOREDUMP;
2669 vm_map_try_merge_entries(map, current->prev, current);
2671 vm_map_try_merge_entries(map, current->prev, current);
2674 vm_pindex_t pstart, pend;
2677 * madvise behaviors that are implemented in the underlying
2680 * Since we don't clip the vm_map_entry, we have to clip
2681 * the vm_object pindex and count.
2683 for (current = entry; current->start < end;
2684 current = current->next) {
2685 vm_offset_t useEnd, useStart;
2687 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
2691 * MADV_FREE would otherwise rewind time to
2692 * the creation of the shadow object. Because
2693 * we hold the VM map read-locked, neither the
2694 * entry's object nor the presence of a
2695 * backing object can change.
2697 if (behav == MADV_FREE &&
2698 current->object.vm_object != NULL &&
2699 current->object.vm_object->backing_object != NULL)
2702 pstart = OFF_TO_IDX(current->offset);
2703 pend = pstart + atop(current->end - current->start);
2704 useStart = current->start;
2705 useEnd = current->end;
2707 if (current->start < start) {
2708 pstart += atop(start - current->start);
2711 if (current->end > end) {
2712 pend -= atop(current->end - end);
2720 * Perform the pmap_advise() before clearing
2721 * PGA_REFERENCED in vm_page_advise(). Otherwise, a
2722 * concurrent pmap operation, such as pmap_remove(),
2723 * could clear a reference in the pmap and set
2724 * PGA_REFERENCED on the page before the pmap_advise()
2725 * had completed. Consequently, the page would appear
2726 * referenced based upon an old reference that
2727 * occurred before this pmap_advise() ran.
2729 if (behav == MADV_DONTNEED || behav == MADV_FREE)
2730 pmap_advise(map->pmap, useStart, useEnd,
2733 vm_object_madvise(current->object.vm_object, pstart,
2737 * Pre-populate paging structures in the
2738 * WILLNEED case. For wired entries, the
2739 * paging structures are already populated.
2741 if (behav == MADV_WILLNEED &&
2742 current->wired_count == 0) {
2743 vm_map_pmap_enter(map,
2745 current->protection,
2746 current->object.vm_object,
2748 ptoa(pend - pstart),
2749 MAP_PREFAULT_MADVISE
2753 vm_map_unlock_read(map);
2762 * Sets the inheritance of the specified address
2763 * range in the target map. Inheritance
2764 * affects how the map will be shared with
2765 * child maps at the time of vmspace_fork.
2768 vm_map_inherit(vm_map_t map, vm_offset_t start, vm_offset_t end,
2769 vm_inherit_t new_inheritance)
2771 vm_map_entry_t entry;
2772 vm_map_entry_t temp_entry;
2774 switch (new_inheritance) {
2775 case VM_INHERIT_NONE:
2776 case VM_INHERIT_COPY:
2777 case VM_INHERIT_SHARE:
2778 case VM_INHERIT_ZERO:
2781 return (KERN_INVALID_ARGUMENT);
2784 return (KERN_SUCCESS);
2786 VM_MAP_RANGE_CHECK(map, start, end);
2787 if (vm_map_lookup_entry(map, start, &temp_entry)) {
2789 vm_map_clip_start(map, entry, start);
2791 entry = temp_entry->next;
2792 while (entry->start < end) {
2793 vm_map_clip_end(map, entry, end);
2794 if ((entry->eflags & MAP_ENTRY_GUARD) == 0 ||
2795 new_inheritance != VM_INHERIT_ZERO)
2796 entry->inheritance = new_inheritance;
2797 vm_map_try_merge_entries(map, entry->prev, entry);
2798 entry = entry->next;
2800 vm_map_try_merge_entries(map, entry->prev, entry);
2802 return (KERN_SUCCESS);
2806 * vm_map_entry_in_transition:
2808 * Release the map lock, and sleep until the entry is no longer in
2809 * transition. Awake and acquire the map lock. If the map changed while
2810 * another held the lock, lookup a possibly-changed entry at or after the
2811 * 'start' position of the old entry.
2813 static vm_map_entry_t
2814 vm_map_entry_in_transition(vm_map_t map, vm_offset_t in_start,
2815 vm_offset_t *io_end, bool holes_ok, vm_map_entry_t in_entry)
2817 vm_map_entry_t entry;
2819 u_int last_timestamp;
2821 VM_MAP_ASSERT_LOCKED(map);
2822 KASSERT((in_entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0,
2823 ("not in-tranition map entry %p", in_entry));
2825 * We have not yet clipped the entry.
2827 start = MAX(in_start, in_entry->start);
2828 in_entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2829 last_timestamp = map->timestamp;
2830 if (vm_map_unlock_and_wait(map, 0)) {
2832 * Allow interruption of user wiring/unwiring?
2836 if (last_timestamp + 1 == map->timestamp)
2840 * Look again for the entry because the map was modified while it was
2841 * unlocked. Specifically, the entry may have been clipped, merged, or
2844 if (!vm_map_lookup_entry(map, start, &entry)) {
2849 entry = entry->next;
2857 * Implements both kernel and user unwiring.
2860 vm_map_unwire(vm_map_t map, vm_offset_t start, vm_offset_t end,
2863 vm_map_entry_t entry, first_entry;
2865 bool first_iteration, holes_ok, need_wakeup, user_unwire;
2868 return (KERN_SUCCESS);
2869 holes_ok = (flags & VM_MAP_WIRE_HOLESOK) != 0;
2870 user_unwire = (flags & VM_MAP_WIRE_USER) != 0;
2872 VM_MAP_RANGE_CHECK(map, start, end);
2873 if (!vm_map_lookup_entry(map, start, &first_entry)) {
2875 first_entry = first_entry->next;
2878 return (KERN_INVALID_ADDRESS);
2881 first_iteration = true;
2882 entry = first_entry;
2884 while (entry->start < end) {
2885 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
2887 * We have not yet clipped the entry.
2889 entry = vm_map_entry_in_transition(map, start, &end,
2891 if (entry == NULL) {
2892 if (first_iteration) {
2894 return (KERN_INVALID_ADDRESS);
2896 rv = KERN_INVALID_ADDRESS;
2899 first_entry = first_iteration ? entry : NULL;
2902 first_iteration = false;
2903 vm_map_clip_start(map, entry, start);
2904 vm_map_clip_end(map, entry, end);
2906 * Mark the entry in case the map lock is released. (See
2909 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 &&
2910 entry->wiring_thread == NULL,
2911 ("owned map entry %p", entry));
2912 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
2913 entry->wiring_thread = curthread;
2915 * Check the map for holes in the specified region.
2916 * If holes_ok, skip this check.
2919 (entry->end < end && entry->next->start > entry->end)) {
2921 rv = KERN_INVALID_ADDRESS;
2925 * If system unwiring, require that the entry is system wired.
2928 vm_map_entry_system_wired_count(entry) == 0) {
2930 rv = KERN_INVALID_ARGUMENT;
2933 entry = entry->next;
2935 need_wakeup = false;
2936 if (first_entry == NULL &&
2937 !vm_map_lookup_entry(map, start, &first_entry)) {
2938 KASSERT(holes_ok, ("vm_map_unwire: lookup failed"));
2939 first_entry = first_entry->next;
2941 for (entry = first_entry; entry->start < end; entry = entry->next) {
2943 * If holes_ok was specified, an empty
2944 * space in the unwired region could have been mapped
2945 * while the map lock was dropped for draining
2946 * MAP_ENTRY_IN_TRANSITION. Moreover, another thread
2947 * could be simultaneously wiring this new mapping
2948 * entry. Detect these cases and skip any entries
2949 * marked as in transition by us.
2951 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 ||
2952 entry->wiring_thread != curthread) {
2954 ("vm_map_unwire: !HOLESOK and new/changed entry"));
2958 if (rv == KERN_SUCCESS && (!user_unwire ||
2959 (entry->eflags & MAP_ENTRY_USER_WIRED))) {
2960 if (entry->wired_count == 1)
2961 vm_map_entry_unwire(map, entry);
2963 entry->wired_count--;
2965 entry->eflags &= ~MAP_ENTRY_USER_WIRED;
2967 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0,
2968 ("vm_map_unwire: in-transition flag missing %p", entry));
2969 KASSERT(entry->wiring_thread == curthread,
2970 ("vm_map_unwire: alien wire %p", entry));
2971 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION;
2972 entry->wiring_thread = NULL;
2973 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
2974 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
2977 vm_map_try_merge_entries(map, entry->prev, entry);
2979 vm_map_try_merge_entries(map, entry->prev, entry);
2987 vm_map_wire_user_count_sub(u_long npages)
2990 atomic_subtract_long(&vm_user_wire_count, npages);
2994 vm_map_wire_user_count_add(u_long npages)
2998 wired = vm_user_wire_count;
3000 if (npages + wired > vm_page_max_user_wired)
3002 } while (!atomic_fcmpset_long(&vm_user_wire_count, &wired,
3009 * vm_map_wire_entry_failure:
3011 * Handle a wiring failure on the given entry.
3013 * The map should be locked.
3016 vm_map_wire_entry_failure(vm_map_t map, vm_map_entry_t entry,
3017 vm_offset_t failed_addr)
3020 VM_MAP_ASSERT_LOCKED(map);
3021 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 &&
3022 entry->wired_count == 1,
3023 ("vm_map_wire_entry_failure: entry %p isn't being wired", entry));
3024 KASSERT(failed_addr < entry->end,
3025 ("vm_map_wire_entry_failure: entry %p was fully wired", entry));
3028 * If any pages at the start of this entry were successfully wired,
3031 if (failed_addr > entry->start) {
3032 pmap_unwire(map->pmap, entry->start, failed_addr);
3033 vm_object_unwire(entry->object.vm_object, entry->offset,
3034 failed_addr - entry->start, PQ_ACTIVE);
3038 * Assign an out-of-range value to represent the failure to wire this
3041 entry->wired_count = -1;
3045 vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t end, int flags)
3050 rv = vm_map_wire_locked(map, start, end, flags);
3057 * vm_map_wire_locked:
3059 * Implements both kernel and user wiring. Returns with the map locked,
3060 * the map lock may be dropped.
3063 vm_map_wire_locked(vm_map_t map, vm_offset_t start, vm_offset_t end, int flags)
3065 vm_map_entry_t entry, first_entry, tmp_entry;
3066 vm_offset_t faddr, saved_end, saved_start;
3068 u_int last_timestamp;
3070 bool first_iteration, holes_ok, need_wakeup, user_wire;
3073 VM_MAP_ASSERT_LOCKED(map);
3076 return (KERN_SUCCESS);
3078 if (flags & VM_MAP_WIRE_WRITE)
3079 prot |= VM_PROT_WRITE;
3080 holes_ok = (flags & VM_MAP_WIRE_HOLESOK) != 0;
3081 user_wire = (flags & VM_MAP_WIRE_USER) != 0;
3082 VM_MAP_RANGE_CHECK(map, start, end);
3083 if (!vm_map_lookup_entry(map, start, &first_entry)) {
3085 first_entry = first_entry->next;
3087 return (KERN_INVALID_ADDRESS);
3089 first_iteration = true;
3090 entry = first_entry;
3091 while (entry->start < end) {
3092 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
3094 * We have not yet clipped the entry.
3096 entry = vm_map_entry_in_transition(map, start, &end,
3098 if (entry == NULL) {
3099 if (first_iteration)
3100 return (KERN_INVALID_ADDRESS);
3101 rv = KERN_INVALID_ADDRESS;
3104 first_entry = first_iteration ? entry : NULL;
3107 first_iteration = false;
3108 vm_map_clip_start(map, entry, start);
3109 vm_map_clip_end(map, entry, end);
3111 * Mark the entry in case the map lock is released. (See
3114 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 &&
3115 entry->wiring_thread == NULL,
3116 ("owned map entry %p", entry));
3117 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
3118 entry->wiring_thread = curthread;
3119 if ((entry->protection & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0
3120 || (entry->protection & prot) != prot) {
3121 entry->eflags |= MAP_ENTRY_WIRE_SKIPPED;
3124 rv = KERN_INVALID_ADDRESS;
3127 } else if (entry->wired_count == 0) {
3128 entry->wired_count++;
3130 npages = atop(entry->end - entry->start);
3131 if (user_wire && !vm_map_wire_user_count_add(npages)) {
3132 vm_map_wire_entry_failure(map, entry,
3135 rv = KERN_RESOURCE_SHORTAGE;
3140 * Release the map lock, relying on the in-transition
3141 * mark. Mark the map busy for fork.
3143 saved_start = entry->start;
3144 saved_end = entry->end;
3145 last_timestamp = map->timestamp;
3149 faddr = saved_start;
3152 * Simulate a fault to get the page and enter
3153 * it into the physical map.
3155 if ((rv = vm_fault(map, faddr,
3156 VM_PROT_NONE, VM_FAULT_WIRE, NULL)) !=
3159 } while ((faddr += PAGE_SIZE) < saved_end);
3162 if (last_timestamp + 1 != map->timestamp) {
3164 * Look again for the entry because the map was
3165 * modified while it was unlocked. The entry
3166 * may have been clipped, but NOT merged or
3169 if (!vm_map_lookup_entry(map, saved_start,
3172 ("vm_map_wire: lookup failed"));
3173 if (entry == first_entry)
3174 first_entry = tmp_entry;
3178 while (entry->end < saved_end) {
3180 * In case of failure, handle entries
3181 * that were not fully wired here;
3182 * fully wired entries are handled
3185 if (rv != KERN_SUCCESS &&
3187 vm_map_wire_entry_failure(map,
3189 entry = entry->next;
3192 if (rv != KERN_SUCCESS) {
3193 vm_map_wire_entry_failure(map, entry, faddr);
3195 vm_map_wire_user_count_sub(npages);
3199 } else if (!user_wire ||
3200 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
3201 entry->wired_count++;
3204 * Check the map for holes in the specified region.
3205 * If holes_ok was specified, skip this check.
3208 entry->end < end && entry->next->start > entry->end) {
3210 rv = KERN_INVALID_ADDRESS;
3213 entry = entry->next;
3217 need_wakeup = false;
3218 if (first_entry == NULL &&
3219 !vm_map_lookup_entry(map, start, &first_entry)) {
3220 KASSERT(holes_ok, ("vm_map_wire: lookup failed"));
3221 first_entry = first_entry->next;
3223 for (entry = first_entry; entry->start < end; entry = entry->next) {
3225 * If holes_ok was specified, an empty
3226 * space in the unwired region could have been mapped
3227 * while the map lock was dropped for faulting in the
3228 * pages or draining MAP_ENTRY_IN_TRANSITION.
3229 * Moreover, another thread could be simultaneously
3230 * wiring this new mapping entry. Detect these cases
3231 * and skip any entries marked as in transition not by us.
3233 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 ||
3234 entry->wiring_thread != curthread) {
3236 ("vm_map_wire: !HOLESOK and new/changed entry"));
3240 if ((entry->eflags & MAP_ENTRY_WIRE_SKIPPED) != 0) {
3242 } else if (rv == KERN_SUCCESS) {
3244 entry->eflags |= MAP_ENTRY_USER_WIRED;
3245 } else if (entry->wired_count == -1) {
3247 * Wiring failed on this entry. Thus, unwiring is
3250 entry->wired_count = 0;
3251 } else if (!user_wire ||
3252 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
3254 * Undo the wiring. Wiring succeeded on this entry
3255 * but failed on a later entry.
3257 if (entry->wired_count == 1) {
3258 vm_map_entry_unwire(map, entry);
3260 vm_map_wire_user_count_sub(
3261 atop(entry->end - entry->start));
3263 entry->wired_count--;
3265 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0,
3266 ("vm_map_wire: in-transition flag missing %p", entry));
3267 KASSERT(entry->wiring_thread == curthread,
3268 ("vm_map_wire: alien wire %p", entry));
3269 entry->eflags &= ~(MAP_ENTRY_IN_TRANSITION |
3270 MAP_ENTRY_WIRE_SKIPPED);
3271 entry->wiring_thread = NULL;
3272 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
3273 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
3276 vm_map_try_merge_entries(map, entry->prev, entry);
3278 vm_map_try_merge_entries(map, entry->prev, entry);
3287 * Push any dirty cached pages in the address range to their pager.
3288 * If syncio is TRUE, dirty pages are written synchronously.
3289 * If invalidate is TRUE, any cached pages are freed as well.
3291 * If the size of the region from start to end is zero, we are
3292 * supposed to flush all modified pages within the region containing
3293 * start. Unfortunately, a region can be split or coalesced with
3294 * neighboring regions, making it difficult to determine what the
3295 * original region was. Therefore, we approximate this requirement by
3296 * flushing the current region containing start.
3298 * Returns an error if any part of the specified range is not mapped.
3306 boolean_t invalidate)
3308 vm_map_entry_t current;
3309 vm_map_entry_t entry;
3312 vm_ooffset_t offset;
3313 unsigned int last_timestamp;
3316 vm_map_lock_read(map);
3317 VM_MAP_RANGE_CHECK(map, start, end);
3318 if (!vm_map_lookup_entry(map, start, &entry)) {
3319 vm_map_unlock_read(map);
3320 return (KERN_INVALID_ADDRESS);
3321 } else if (start == end) {
3322 start = entry->start;
3326 * Make a first pass to check for user-wired memory and holes.
3328 for (current = entry; current->start < end; current = current->next) {
3329 if (invalidate && (current->eflags & MAP_ENTRY_USER_WIRED)) {
3330 vm_map_unlock_read(map);
3331 return (KERN_INVALID_ARGUMENT);
3333 if (end > current->end &&
3334 current->end != current->next->start) {
3335 vm_map_unlock_read(map);
3336 return (KERN_INVALID_ADDRESS);
3341 pmap_remove(map->pmap, start, end);
3345 * Make a second pass, cleaning/uncaching pages from the indicated
3348 for (current = entry; current->start < end;) {
3349 offset = current->offset + (start - current->start);
3350 size = (end <= current->end ? end : current->end) - start;
3351 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
3353 vm_map_entry_t tentry;
3356 smap = current->object.sub_map;
3357 vm_map_lock_read(smap);
3358 (void) vm_map_lookup_entry(smap, offset, &tentry);
3359 tsize = tentry->end - offset;
3362 object = tentry->object.vm_object;
3363 offset = tentry->offset + (offset - tentry->start);
3364 vm_map_unlock_read(smap);
3366 object = current->object.vm_object;
3368 vm_object_reference(object);
3369 last_timestamp = map->timestamp;
3370 vm_map_unlock_read(map);
3371 if (!vm_object_sync(object, offset, size, syncio, invalidate))
3374 vm_object_deallocate(object);
3375 vm_map_lock_read(map);
3376 if (last_timestamp == map->timestamp ||
3377 !vm_map_lookup_entry(map, start, ¤t))
3378 current = current->next;
3381 vm_map_unlock_read(map);
3382 return (failed ? KERN_FAILURE : KERN_SUCCESS);
3386 * vm_map_entry_unwire: [ internal use only ]
3388 * Make the region specified by this entry pageable.
3390 * The map in question should be locked.
3391 * [This is the reason for this routine's existence.]
3394 vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry)
3398 VM_MAP_ASSERT_LOCKED(map);
3399 KASSERT(entry->wired_count > 0,
3400 ("vm_map_entry_unwire: entry %p isn't wired", entry));
3402 size = entry->end - entry->start;
3403 if ((entry->eflags & MAP_ENTRY_USER_WIRED) != 0)
3404 vm_map_wire_user_count_sub(atop(size));
3405 pmap_unwire(map->pmap, entry->start, entry->end);
3406 vm_object_unwire(entry->object.vm_object, entry->offset, size,
3408 entry->wired_count = 0;
3412 vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map)
3415 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0)
3416 vm_object_deallocate(entry->object.vm_object);
3417 uma_zfree(system_map ? kmapentzone : mapentzone, entry);
3421 * vm_map_entry_delete: [ internal use only ]
3423 * Deallocate the given entry from the target map.
3426 vm_map_entry_delete(vm_map_t map, vm_map_entry_t entry)
3429 vm_pindex_t offidxstart, offidxend, count, size1;
3432 vm_map_entry_unlink(map, entry, UNLINK_MERGE_NONE);
3433 object = entry->object.vm_object;
3435 if ((entry->eflags & MAP_ENTRY_GUARD) != 0) {
3436 MPASS(entry->cred == NULL);
3437 MPASS((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0);
3438 MPASS(object == NULL);
3439 vm_map_entry_deallocate(entry, map->system_map);
3443 size = entry->end - entry->start;
3446 if (entry->cred != NULL) {
3447 swap_release_by_cred(size, entry->cred);
3448 crfree(entry->cred);
3451 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0 &&
3453 KASSERT(entry->cred == NULL || object->cred == NULL ||
3454 (entry->eflags & MAP_ENTRY_NEEDS_COPY),
3455 ("OVERCOMMIT vm_map_entry_delete: both cred %p", entry));
3457 offidxstart = OFF_TO_IDX(entry->offset);
3458 offidxend = offidxstart + count;
3459 VM_OBJECT_WLOCK(object);
3460 if (object->ref_count != 1 && ((object->flags & (OBJ_NOSPLIT |
3461 OBJ_ONEMAPPING)) == OBJ_ONEMAPPING ||
3462 object == kernel_object)) {
3463 vm_object_collapse(object);
3466 * The option OBJPR_NOTMAPPED can be passed here
3467 * because vm_map_delete() already performed
3468 * pmap_remove() on the only mapping to this range
3471 vm_object_page_remove(object, offidxstart, offidxend,
3473 if (object->type == OBJT_SWAP)
3474 swap_pager_freespace(object, offidxstart,
3476 if (offidxend >= object->size &&
3477 offidxstart < object->size) {
3478 size1 = object->size;
3479 object->size = offidxstart;
3480 if (object->cred != NULL) {
3481 size1 -= object->size;
3482 KASSERT(object->charge >= ptoa(size1),
3483 ("object %p charge < 0", object));
3484 swap_release_by_cred(ptoa(size1),
3486 object->charge -= ptoa(size1);
3490 VM_OBJECT_WUNLOCK(object);
3492 entry->object.vm_object = NULL;
3493 if (map->system_map)
3494 vm_map_entry_deallocate(entry, TRUE);
3496 entry->next = curthread->td_map_def_user;
3497 curthread->td_map_def_user = entry;
3502 * vm_map_delete: [ internal use only ]
3504 * Deallocates the given address range from the target
3508 vm_map_delete(vm_map_t map, vm_offset_t start, vm_offset_t end)
3510 vm_map_entry_t entry;
3511 vm_map_entry_t first_entry;
3513 VM_MAP_ASSERT_LOCKED(map);
3515 return (KERN_SUCCESS);
3518 * Find the start of the region, and clip it
3520 if (!vm_map_lookup_entry(map, start, &first_entry))
3521 entry = first_entry->next;
3523 entry = first_entry;
3524 vm_map_clip_start(map, entry, start);
3528 * Step through all entries in this region
3530 while (entry->start < end) {
3531 vm_map_entry_t next;
3534 * Wait for wiring or unwiring of an entry to complete.
3535 * Also wait for any system wirings to disappear on
3538 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 ||
3539 (vm_map_pmap(map) != kernel_pmap &&
3540 vm_map_entry_system_wired_count(entry) != 0)) {
3541 unsigned int last_timestamp;
3542 vm_offset_t saved_start;
3543 vm_map_entry_t tmp_entry;
3545 saved_start = entry->start;
3546 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
3547 last_timestamp = map->timestamp;
3548 (void) vm_map_unlock_and_wait(map, 0);
3550 if (last_timestamp + 1 != map->timestamp) {
3552 * Look again for the entry because the map was
3553 * modified while it was unlocked.
3554 * Specifically, the entry may have been
3555 * clipped, merged, or deleted.
3557 if (!vm_map_lookup_entry(map, saved_start,
3559 entry = tmp_entry->next;
3562 vm_map_clip_start(map, entry,
3568 vm_map_clip_end(map, entry, end);
3573 * Unwire before removing addresses from the pmap; otherwise,
3574 * unwiring will put the entries back in the pmap.
3576 if (entry->wired_count != 0)
3577 vm_map_entry_unwire(map, entry);
3580 * Remove mappings for the pages, but only if the
3581 * mappings could exist. For instance, it does not
3582 * make sense to call pmap_remove() for guard entries.
3584 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0 ||
3585 entry->object.vm_object != NULL)
3586 pmap_remove(map->pmap, entry->start, entry->end);
3588 if (entry->end == map->anon_loc)
3589 map->anon_loc = entry->start;
3592 * Delete the entry only after removing all pmap
3593 * entries pointing to its pages. (Otherwise, its
3594 * page frames may be reallocated, and any modify bits
3595 * will be set in the wrong object!)
3597 vm_map_entry_delete(map, entry);
3600 return (KERN_SUCCESS);
3606 * Remove the given address range from the target map.
3607 * This is the exported form of vm_map_delete.
3610 vm_map_remove(vm_map_t map, vm_offset_t start, vm_offset_t end)
3615 VM_MAP_RANGE_CHECK(map, start, end);
3616 result = vm_map_delete(map, start, end);
3622 * vm_map_check_protection:
3624 * Assert that the target map allows the specified privilege on the
3625 * entire address region given. The entire region must be allocated.
3627 * WARNING! This code does not and should not check whether the
3628 * contents of the region is accessible. For example a smaller file
3629 * might be mapped into a larger address space.
3631 * NOTE! This code is also called by munmap().
3633 * The map must be locked. A read lock is sufficient.
3636 vm_map_check_protection(vm_map_t map, vm_offset_t start, vm_offset_t end,
3637 vm_prot_t protection)
3639 vm_map_entry_t entry;
3640 vm_map_entry_t tmp_entry;
3642 if (!vm_map_lookup_entry(map, start, &tmp_entry))
3646 while (start < end) {
3650 if (start < entry->start)
3653 * Check protection associated with entry.
3655 if ((entry->protection & protection) != protection)
3657 /* go to next entry */
3659 entry = entry->next;
3665 * vm_map_copy_entry:
3667 * Copies the contents of the source entry to the destination
3668 * entry. The entries *must* be aligned properly.
3674 vm_map_entry_t src_entry,
3675 vm_map_entry_t dst_entry,
3676 vm_ooffset_t *fork_charge)
3678 vm_object_t src_object;
3679 vm_map_entry_t fake_entry;
3684 VM_MAP_ASSERT_LOCKED(dst_map);
3686 if ((dst_entry->eflags|src_entry->eflags) & MAP_ENTRY_IS_SUB_MAP)
3689 if (src_entry->wired_count == 0 ||
3690 (src_entry->protection & VM_PROT_WRITE) == 0) {
3692 * If the source entry is marked needs_copy, it is already
3695 if ((src_entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0 &&
3696 (src_entry->protection & VM_PROT_WRITE) != 0) {
3697 pmap_protect(src_map->pmap,
3700 src_entry->protection & ~VM_PROT_WRITE);
3704 * Make a copy of the object.
3706 size = src_entry->end - src_entry->start;
3707 if ((src_object = src_entry->object.vm_object) != NULL) {
3708 VM_OBJECT_WLOCK(src_object);
3709 charged = ENTRY_CHARGED(src_entry);
3710 if (src_object->handle == NULL &&
3711 (src_object->type == OBJT_DEFAULT ||
3712 src_object->type == OBJT_SWAP)) {
3713 vm_object_collapse(src_object);
3714 if ((src_object->flags & (OBJ_NOSPLIT |
3715 OBJ_ONEMAPPING)) == OBJ_ONEMAPPING) {
3716 vm_object_split(src_entry);
3718 src_entry->object.vm_object;
3721 vm_object_reference_locked(src_object);
3722 vm_object_clear_flag(src_object, OBJ_ONEMAPPING);
3723 if (src_entry->cred != NULL &&
3724 !(src_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
3725 KASSERT(src_object->cred == NULL,
3726 ("OVERCOMMIT: vm_map_copy_entry: cred %p",
3728 src_object->cred = src_entry->cred;
3729 src_object->charge = size;
3731 VM_OBJECT_WUNLOCK(src_object);
3732 dst_entry->object.vm_object = src_object;
3734 cred = curthread->td_ucred;
3736 dst_entry->cred = cred;
3737 *fork_charge += size;
3738 if (!(src_entry->eflags &
3739 MAP_ENTRY_NEEDS_COPY)) {
3741 src_entry->cred = cred;
3742 *fork_charge += size;
3745 src_entry->eflags |= MAP_ENTRY_COW |
3746 MAP_ENTRY_NEEDS_COPY;
3747 dst_entry->eflags |= MAP_ENTRY_COW |
3748 MAP_ENTRY_NEEDS_COPY;
3749 dst_entry->offset = src_entry->offset;
3750 if (src_entry->eflags & MAP_ENTRY_WRITECNT) {
3752 * MAP_ENTRY_WRITECNT cannot
3753 * indicate write reference from
3754 * src_entry, since the entry is
3755 * marked as needs copy. Allocate a
3756 * fake entry that is used to
3757 * decrement object->un_pager writecount
3758 * at the appropriate time. Attach
3759 * fake_entry to the deferred list.
3761 fake_entry = vm_map_entry_create(dst_map);
3762 fake_entry->eflags = MAP_ENTRY_WRITECNT;
3763 src_entry->eflags &= ~MAP_ENTRY_WRITECNT;
3764 vm_object_reference(src_object);
3765 fake_entry->object.vm_object = src_object;
3766 fake_entry->start = src_entry->start;
3767 fake_entry->end = src_entry->end;
3768 fake_entry->next = curthread->td_map_def_user;
3769 curthread->td_map_def_user = fake_entry;
3772 pmap_copy(dst_map->pmap, src_map->pmap,
3773 dst_entry->start, dst_entry->end - dst_entry->start,
3776 dst_entry->object.vm_object = NULL;
3777 dst_entry->offset = 0;
3778 if (src_entry->cred != NULL) {
3779 dst_entry->cred = curthread->td_ucred;
3780 crhold(dst_entry->cred);
3781 *fork_charge += size;
3786 * We don't want to make writeable wired pages copy-on-write.
3787 * Immediately copy these pages into the new map by simulating
3788 * page faults. The new pages are pageable.
3790 vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry,
3796 * vmspace_map_entry_forked:
3797 * Update the newly-forked vmspace each time a map entry is inherited
3798 * or copied. The values for vm_dsize and vm_tsize are approximate
3799 * (and mostly-obsolete ideas in the face of mmap(2) et al.)
3802 vmspace_map_entry_forked(const struct vmspace *vm1, struct vmspace *vm2,
3803 vm_map_entry_t entry)
3805 vm_size_t entrysize;
3808 if ((entry->eflags & MAP_ENTRY_GUARD) != 0)
3810 entrysize = entry->end - entry->start;
3811 vm2->vm_map.size += entrysize;
3812 if (entry->eflags & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP)) {
3813 vm2->vm_ssize += btoc(entrysize);
3814 } else if (entry->start >= (vm_offset_t)vm1->vm_daddr &&
3815 entry->start < (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)) {
3816 newend = MIN(entry->end,
3817 (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize));
3818 vm2->vm_dsize += btoc(newend - entry->start);
3819 } else if (entry->start >= (vm_offset_t)vm1->vm_taddr &&
3820 entry->start < (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)) {
3821 newend = MIN(entry->end,
3822 (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize));
3823 vm2->vm_tsize += btoc(newend - entry->start);
3829 * Create a new process vmspace structure and vm_map
3830 * based on those of an existing process. The new map
3831 * is based on the old map, according to the inheritance
3832 * values on the regions in that map.
3834 * XXX It might be worth coalescing the entries added to the new vmspace.
3836 * The source map must not be locked.
3839 vmspace_fork(struct vmspace *vm1, vm_ooffset_t *fork_charge)
3841 struct vmspace *vm2;
3842 vm_map_t new_map, old_map;
3843 vm_map_entry_t new_entry, old_entry;
3848 old_map = &vm1->vm_map;
3849 /* Copy immutable fields of vm1 to vm2. */
3850 vm2 = vmspace_alloc(vm_map_min(old_map), vm_map_max(old_map),
3855 vm2->vm_taddr = vm1->vm_taddr;
3856 vm2->vm_daddr = vm1->vm_daddr;
3857 vm2->vm_maxsaddr = vm1->vm_maxsaddr;
3858 vm_map_lock(old_map);
3860 vm_map_wait_busy(old_map);
3861 new_map = &vm2->vm_map;
3862 locked = vm_map_trylock(new_map); /* trylock to silence WITNESS */
3863 KASSERT(locked, ("vmspace_fork: lock failed"));
3865 error = pmap_vmspace_copy(new_map->pmap, old_map->pmap);
3867 sx_xunlock(&old_map->lock);
3868 sx_xunlock(&new_map->lock);
3869 vm_map_process_deferred();
3874 new_map->anon_loc = old_map->anon_loc;
3876 old_entry = old_map->header.next;
3878 while (old_entry != &old_map->header) {
3879 if (old_entry->eflags & MAP_ENTRY_IS_SUB_MAP)
3880 panic("vm_map_fork: encountered a submap");
3882 inh = old_entry->inheritance;
3883 if ((old_entry->eflags & MAP_ENTRY_GUARD) != 0 &&
3884 inh != VM_INHERIT_NONE)
3885 inh = VM_INHERIT_COPY;
3888 case VM_INHERIT_NONE:
3891 case VM_INHERIT_SHARE:
3893 * Clone the entry, creating the shared object if necessary.
3895 object = old_entry->object.vm_object;
3896 if (object == NULL) {
3897 vm_map_entry_back(old_entry);
3898 object = old_entry->object.vm_object;
3902 * Add the reference before calling vm_object_shadow
3903 * to insure that a shadow object is created.
3905 vm_object_reference(object);
3906 if (old_entry->eflags & MAP_ENTRY_NEEDS_COPY) {
3907 vm_object_shadow(&old_entry->object.vm_object,
3909 old_entry->end - old_entry->start);
3910 old_entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
3911 /* Transfer the second reference too. */
3912 vm_object_reference(
3913 old_entry->object.vm_object);
3916 * As in vm_map_merged_neighbor_dispose(),
3917 * the vnode lock will not be acquired in
3918 * this call to vm_object_deallocate().
3920 vm_object_deallocate(object);
3921 object = old_entry->object.vm_object;
3923 VM_OBJECT_WLOCK(object);
3924 vm_object_clear_flag(object, OBJ_ONEMAPPING);
3925 if (old_entry->cred != NULL) {
3926 KASSERT(object->cred == NULL, ("vmspace_fork both cred"));
3927 object->cred = old_entry->cred;
3928 object->charge = old_entry->end - old_entry->start;
3929 old_entry->cred = NULL;
3933 * Assert the correct state of the vnode
3934 * v_writecount while the object is locked, to
3935 * not relock it later for the assertion
3938 if (old_entry->eflags & MAP_ENTRY_WRITECNT &&
3939 object->type == OBJT_VNODE) {
3940 KASSERT(((struct vnode *)object->handle)->
3942 ("vmspace_fork: v_writecount %p", object));
3943 KASSERT(object->un_pager.vnp.writemappings > 0,
3944 ("vmspace_fork: vnp.writecount %p",
3947 VM_OBJECT_WUNLOCK(object);
3950 * Clone the entry, referencing the shared object.
3952 new_entry = vm_map_entry_create(new_map);
3953 *new_entry = *old_entry;
3954 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED |
3955 MAP_ENTRY_IN_TRANSITION);
3956 new_entry->wiring_thread = NULL;
3957 new_entry->wired_count = 0;
3958 if (new_entry->eflags & MAP_ENTRY_WRITECNT) {
3959 vm_pager_update_writecount(object,
3960 new_entry->start, new_entry->end);
3962 vm_map_entry_set_vnode_text(new_entry, true);
3965 * Insert the entry into the new map -- we know we're
3966 * inserting at the end of the new map.
3968 vm_map_entry_link(new_map, new_entry);
3969 vmspace_map_entry_forked(vm1, vm2, new_entry);
3972 * Update the physical map
3974 pmap_copy(new_map->pmap, old_map->pmap,
3976 (old_entry->end - old_entry->start),
3980 case VM_INHERIT_COPY:
3982 * Clone the entry and link into the map.
3984 new_entry = vm_map_entry_create(new_map);
3985 *new_entry = *old_entry;
3987 * Copied entry is COW over the old object.
3989 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED |
3990 MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_WRITECNT);
3991 new_entry->wiring_thread = NULL;
3992 new_entry->wired_count = 0;
3993 new_entry->object.vm_object = NULL;
3994 new_entry->cred = NULL;
3995 vm_map_entry_link(new_map, new_entry);
3996 vmspace_map_entry_forked(vm1, vm2, new_entry);
3997 vm_map_copy_entry(old_map, new_map, old_entry,
3998 new_entry, fork_charge);
3999 vm_map_entry_set_vnode_text(new_entry, true);
4002 case VM_INHERIT_ZERO:
4004 * Create a new anonymous mapping entry modelled from
4007 new_entry = vm_map_entry_create(new_map);
4008 memset(new_entry, 0, sizeof(*new_entry));
4010 new_entry->start = old_entry->start;
4011 new_entry->end = old_entry->end;
4012 new_entry->eflags = old_entry->eflags &
4013 ~(MAP_ENTRY_USER_WIRED | MAP_ENTRY_IN_TRANSITION |
4014 MAP_ENTRY_WRITECNT | MAP_ENTRY_VN_EXEC);
4015 new_entry->protection = old_entry->protection;
4016 new_entry->max_protection = old_entry->max_protection;
4017 new_entry->inheritance = VM_INHERIT_ZERO;
4019 vm_map_entry_link(new_map, new_entry);
4020 vmspace_map_entry_forked(vm1, vm2, new_entry);
4022 new_entry->cred = curthread->td_ucred;
4023 crhold(new_entry->cred);
4024 *fork_charge += (new_entry->end - new_entry->start);
4028 old_entry = old_entry->next;
4031 * Use inlined vm_map_unlock() to postpone handling the deferred
4032 * map entries, which cannot be done until both old_map and
4033 * new_map locks are released.
4035 sx_xunlock(&old_map->lock);
4036 sx_xunlock(&new_map->lock);
4037 vm_map_process_deferred();
4043 * Create a process's stack for exec_new_vmspace(). This function is never
4044 * asked to wire the newly created stack.
4047 vm_map_stack(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
4048 vm_prot_t prot, vm_prot_t max, int cow)
4050 vm_size_t growsize, init_ssize;
4054 MPASS((map->flags & MAP_WIREFUTURE) == 0);
4055 growsize = sgrowsiz;
4056 init_ssize = (max_ssize < growsize) ? max_ssize : growsize;
4058 vmemlim = lim_cur(curthread, RLIMIT_VMEM);
4059 /* If we would blow our VMEM resource limit, no go */
4060 if (map->size + init_ssize > vmemlim) {
4064 rv = vm_map_stack_locked(map, addrbos, max_ssize, growsize, prot,
4071 static int stack_guard_page = 1;
4072 SYSCTL_INT(_security_bsd, OID_AUTO, stack_guard_page, CTLFLAG_RWTUN,
4073 &stack_guard_page, 0,
4074 "Specifies the number of guard pages for a stack that grows");
4077 vm_map_stack_locked(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
4078 vm_size_t growsize, vm_prot_t prot, vm_prot_t max, int cow)
4080 vm_map_entry_t new_entry, prev_entry;
4081 vm_offset_t bot, gap_bot, gap_top, top;
4082 vm_size_t init_ssize, sgp;
4086 * The stack orientation is piggybacked with the cow argument.
4087 * Extract it into orient and mask the cow argument so that we
4088 * don't pass it around further.
4090 orient = cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP);
4091 KASSERT(orient != 0, ("No stack grow direction"));
4092 KASSERT(orient != (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP),
4095 if (addrbos < vm_map_min(map) ||
4096 addrbos + max_ssize > vm_map_max(map) ||
4097 addrbos + max_ssize <= addrbos)
4098 return (KERN_INVALID_ADDRESS);
4099 sgp = (curproc->p_flag2 & P2_STKGAP_DISABLE) != 0 ? 0 :
4100 (vm_size_t)stack_guard_page * PAGE_SIZE;
4101 if (sgp >= max_ssize)
4102 return (KERN_INVALID_ARGUMENT);
4104 init_ssize = growsize;
4105 if (max_ssize < init_ssize + sgp)
4106 init_ssize = max_ssize - sgp;
4108 /* If addr is already mapped, no go */
4109 if (vm_map_lookup_entry(map, addrbos, &prev_entry))
4110 return (KERN_NO_SPACE);
4113 * If we can't accommodate max_ssize in the current mapping, no go.
4115 if (prev_entry->next->start < addrbos + max_ssize)
4116 return (KERN_NO_SPACE);
4119 * We initially map a stack of only init_ssize. We will grow as
4120 * needed later. Depending on the orientation of the stack (i.e.
4121 * the grow direction) we either map at the top of the range, the
4122 * bottom of the range or in the middle.
4124 * Note: we would normally expect prot and max to be VM_PROT_ALL,
4125 * and cow to be 0. Possibly we should eliminate these as input
4126 * parameters, and just pass these values here in the insert call.
4128 if (orient == MAP_STACK_GROWS_DOWN) {
4129 bot = addrbos + max_ssize - init_ssize;
4130 top = bot + init_ssize;
4133 } else /* if (orient == MAP_STACK_GROWS_UP) */ {
4135 top = bot + init_ssize;
4137 gap_top = addrbos + max_ssize;
4139 rv = vm_map_insert(map, NULL, 0, bot, top, prot, max, cow);
4140 if (rv != KERN_SUCCESS)
4142 new_entry = prev_entry->next;
4143 KASSERT(new_entry->end == top || new_entry->start == bot,
4144 ("Bad entry start/end for new stack entry"));
4145 KASSERT((orient & MAP_STACK_GROWS_DOWN) == 0 ||
4146 (new_entry->eflags & MAP_ENTRY_GROWS_DOWN) != 0,
4147 ("new entry lacks MAP_ENTRY_GROWS_DOWN"));
4148 KASSERT((orient & MAP_STACK_GROWS_UP) == 0 ||
4149 (new_entry->eflags & MAP_ENTRY_GROWS_UP) != 0,
4150 ("new entry lacks MAP_ENTRY_GROWS_UP"));
4151 if (gap_bot == gap_top)
4152 return (KERN_SUCCESS);
4153 rv = vm_map_insert(map, NULL, 0, gap_bot, gap_top, VM_PROT_NONE,
4154 VM_PROT_NONE, MAP_CREATE_GUARD | (orient == MAP_STACK_GROWS_DOWN ?
4155 MAP_CREATE_STACK_GAP_DN : MAP_CREATE_STACK_GAP_UP));
4156 if (rv == KERN_SUCCESS) {
4158 * Gap can never successfully handle a fault, so
4159 * read-ahead logic is never used for it. Re-use
4160 * next_read of the gap entry to store
4161 * stack_guard_page for vm_map_growstack().
4163 if (orient == MAP_STACK_GROWS_DOWN)
4164 new_entry->prev->next_read = sgp;
4166 new_entry->next->next_read = sgp;
4168 (void)vm_map_delete(map, bot, top);
4174 * Attempts to grow a vm stack entry. Returns KERN_SUCCESS if we
4175 * successfully grow the stack.
4178 vm_map_growstack(vm_map_t map, vm_offset_t addr, vm_map_entry_t gap_entry)
4180 vm_map_entry_t stack_entry;
4184 vm_offset_t gap_end, gap_start, grow_start;
4185 vm_size_t grow_amount, guard, max_grow;
4186 rlim_t lmemlim, stacklim, vmemlim;
4188 bool gap_deleted, grow_down, is_procstack;
4200 * Disallow stack growth when the access is performed by a
4201 * debugger or AIO daemon. The reason is that the wrong
4202 * resource limits are applied.
4204 if (p != initproc && (map != &p->p_vmspace->vm_map ||
4205 p->p_textvp == NULL))
4206 return (KERN_FAILURE);
4208 MPASS(!map->system_map);
4210 lmemlim = lim_cur(curthread, RLIMIT_MEMLOCK);
4211 stacklim = lim_cur(curthread, RLIMIT_STACK);
4212 vmemlim = lim_cur(curthread, RLIMIT_VMEM);
4214 /* If addr is not in a hole for a stack grow area, no need to grow. */
4215 if (gap_entry == NULL && !vm_map_lookup_entry(map, addr, &gap_entry))
4216 return (KERN_FAILURE);
4217 if ((gap_entry->eflags & MAP_ENTRY_GUARD) == 0)
4218 return (KERN_SUCCESS);
4219 if ((gap_entry->eflags & MAP_ENTRY_STACK_GAP_DN) != 0) {
4220 stack_entry = gap_entry->next;
4221 if ((stack_entry->eflags & MAP_ENTRY_GROWS_DOWN) == 0 ||
4222 stack_entry->start != gap_entry->end)
4223 return (KERN_FAILURE);
4224 grow_amount = round_page(stack_entry->start - addr);
4226 } else if ((gap_entry->eflags & MAP_ENTRY_STACK_GAP_UP) != 0) {
4227 stack_entry = gap_entry->prev;
4228 if ((stack_entry->eflags & MAP_ENTRY_GROWS_UP) == 0 ||
4229 stack_entry->end != gap_entry->start)
4230 return (KERN_FAILURE);
4231 grow_amount = round_page(addr + 1 - stack_entry->end);
4234 return (KERN_FAILURE);
4236 guard = (curproc->p_flag2 & P2_STKGAP_DISABLE) != 0 ? 0 :
4237 gap_entry->next_read;
4238 max_grow = gap_entry->end - gap_entry->start;
4239 if (guard > max_grow)
4240 return (KERN_NO_SPACE);
4242 if (grow_amount > max_grow)
4243 return (KERN_NO_SPACE);
4246 * If this is the main process stack, see if we're over the stack
4249 is_procstack = addr >= (vm_offset_t)vm->vm_maxsaddr &&
4250 addr < (vm_offset_t)p->p_sysent->sv_usrstack;
4251 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim))
4252 return (KERN_NO_SPACE);
4257 if (is_procstack && racct_set(p, RACCT_STACK,
4258 ctob(vm->vm_ssize) + grow_amount)) {
4260 return (KERN_NO_SPACE);
4266 grow_amount = roundup(grow_amount, sgrowsiz);
4267 if (grow_amount > max_grow)
4268 grow_amount = max_grow;
4269 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) {
4270 grow_amount = trunc_page((vm_size_t)stacklim) -
4276 limit = racct_get_available(p, RACCT_STACK);
4278 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > limit))
4279 grow_amount = limit - ctob(vm->vm_ssize);
4282 if (!old_mlock && (map->flags & MAP_WIREFUTURE) != 0) {
4283 if (ptoa(pmap_wired_count(map->pmap)) + grow_amount > lmemlim) {
4290 if (racct_set(p, RACCT_MEMLOCK,
4291 ptoa(pmap_wired_count(map->pmap)) + grow_amount)) {
4301 /* If we would blow our VMEM resource limit, no go */
4302 if (map->size + grow_amount > vmemlim) {
4309 if (racct_set(p, RACCT_VMEM, map->size + grow_amount)) {
4318 if (vm_map_lock_upgrade(map)) {
4320 vm_map_lock_read(map);
4325 grow_start = gap_entry->end - grow_amount;
4326 if (gap_entry->start + grow_amount == gap_entry->end) {
4327 gap_start = gap_entry->start;
4328 gap_end = gap_entry->end;
4329 vm_map_entry_delete(map, gap_entry);
4332 MPASS(gap_entry->start < gap_entry->end - grow_amount);
4333 vm_map_entry_resize(map, gap_entry, -grow_amount);
4334 gap_deleted = false;
4336 rv = vm_map_insert(map, NULL, 0, grow_start,
4337 grow_start + grow_amount,
4338 stack_entry->protection, stack_entry->max_protection,
4339 MAP_STACK_GROWS_DOWN);
4340 if (rv != KERN_SUCCESS) {
4342 rv1 = vm_map_insert(map, NULL, 0, gap_start,
4343 gap_end, VM_PROT_NONE, VM_PROT_NONE,
4344 MAP_CREATE_GUARD | MAP_CREATE_STACK_GAP_DN);
4345 MPASS(rv1 == KERN_SUCCESS);
4347 vm_map_entry_resize(map, gap_entry,
4351 grow_start = stack_entry->end;
4352 cred = stack_entry->cred;
4353 if (cred == NULL && stack_entry->object.vm_object != NULL)
4354 cred = stack_entry->object.vm_object->cred;
4355 if (cred != NULL && !swap_reserve_by_cred(grow_amount, cred))
4357 /* Grow the underlying object if applicable. */
4358 else if (stack_entry->object.vm_object == NULL ||
4359 vm_object_coalesce(stack_entry->object.vm_object,
4360 stack_entry->offset,
4361 (vm_size_t)(stack_entry->end - stack_entry->start),
4362 grow_amount, cred != NULL)) {
4363 if (gap_entry->start + grow_amount == gap_entry->end) {
4364 vm_map_entry_delete(map, gap_entry);
4365 vm_map_entry_resize(map, stack_entry,
4368 gap_entry->start += grow_amount;
4369 stack_entry->end += grow_amount;
4371 map->size += grow_amount;
4376 if (rv == KERN_SUCCESS && is_procstack)
4377 vm->vm_ssize += btoc(grow_amount);
4380 * Heed the MAP_WIREFUTURE flag if it was set for this process.
4382 if (rv == KERN_SUCCESS && (map->flags & MAP_WIREFUTURE) != 0) {
4383 rv = vm_map_wire_locked(map, grow_start,
4384 grow_start + grow_amount,
4385 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
4387 vm_map_lock_downgrade(map);
4391 if (racct_enable && rv != KERN_SUCCESS) {
4393 error = racct_set(p, RACCT_VMEM, map->size);
4394 KASSERT(error == 0, ("decreasing RACCT_VMEM failed"));
4396 error = racct_set(p, RACCT_MEMLOCK,
4397 ptoa(pmap_wired_count(map->pmap)));
4398 KASSERT(error == 0, ("decreasing RACCT_MEMLOCK failed"));
4400 error = racct_set(p, RACCT_STACK, ctob(vm->vm_ssize));
4401 KASSERT(error == 0, ("decreasing RACCT_STACK failed"));
4410 * Unshare the specified VM space for exec. If other processes are
4411 * mapped to it, then create a new one. The new vmspace is null.
4414 vmspace_exec(struct proc *p, vm_offset_t minuser, vm_offset_t maxuser)
4416 struct vmspace *oldvmspace = p->p_vmspace;
4417 struct vmspace *newvmspace;
4419 KASSERT((curthread->td_pflags & TDP_EXECVMSPC) == 0,
4420 ("vmspace_exec recursed"));
4421 newvmspace = vmspace_alloc(minuser, maxuser, pmap_pinit);
4422 if (newvmspace == NULL)
4424 newvmspace->vm_swrss = oldvmspace->vm_swrss;
4426 * This code is written like this for prototype purposes. The
4427 * goal is to avoid running down the vmspace here, but let the
4428 * other process's that are still using the vmspace to finally
4429 * run it down. Even though there is little or no chance of blocking
4430 * here, it is a good idea to keep this form for future mods.
4432 PROC_VMSPACE_LOCK(p);
4433 p->p_vmspace = newvmspace;
4434 PROC_VMSPACE_UNLOCK(p);
4435 if (p == curthread->td_proc)
4436 pmap_activate(curthread);
4437 curthread->td_pflags |= TDP_EXECVMSPC;
4442 * Unshare the specified VM space for forcing COW. This
4443 * is called by rfork, for the (RFMEM|RFPROC) == 0 case.
4446 vmspace_unshare(struct proc *p)
4448 struct vmspace *oldvmspace = p->p_vmspace;
4449 struct vmspace *newvmspace;
4450 vm_ooffset_t fork_charge;
4452 if (oldvmspace->vm_refcnt == 1)
4455 newvmspace = vmspace_fork(oldvmspace, &fork_charge);
4456 if (newvmspace == NULL)
4458 if (!swap_reserve_by_cred(fork_charge, p->p_ucred)) {
4459 vmspace_free(newvmspace);
4462 PROC_VMSPACE_LOCK(p);
4463 p->p_vmspace = newvmspace;
4464 PROC_VMSPACE_UNLOCK(p);
4465 if (p == curthread->td_proc)
4466 pmap_activate(curthread);
4467 vmspace_free(oldvmspace);
4474 * Finds the VM object, offset, and
4475 * protection for a given virtual address in the
4476 * specified map, assuming a page fault of the
4479 * Leaves the map in question locked for read; return
4480 * values are guaranteed until a vm_map_lookup_done
4481 * call is performed. Note that the map argument
4482 * is in/out; the returned map must be used in
4483 * the call to vm_map_lookup_done.
4485 * A handle (out_entry) is returned for use in
4486 * vm_map_lookup_done, to make that fast.
4488 * If a lookup is requested with "write protection"
4489 * specified, the map may be changed to perform virtual
4490 * copying operations, although the data referenced will
4494 vm_map_lookup(vm_map_t *var_map, /* IN/OUT */
4496 vm_prot_t fault_typea,
4497 vm_map_entry_t *out_entry, /* OUT */
4498 vm_object_t *object, /* OUT */
4499 vm_pindex_t *pindex, /* OUT */
4500 vm_prot_t *out_prot, /* OUT */
4501 boolean_t *wired) /* OUT */
4503 vm_map_entry_t entry;
4504 vm_map_t map = *var_map;
4506 vm_prot_t fault_type = fault_typea;
4507 vm_object_t eobject;
4513 vm_map_lock_read(map);
4517 * Lookup the faulting address.
4519 if (!vm_map_lookup_entry(map, vaddr, out_entry)) {
4520 vm_map_unlock_read(map);
4521 return (KERN_INVALID_ADDRESS);
4529 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
4530 vm_map_t old_map = map;
4532 *var_map = map = entry->object.sub_map;
4533 vm_map_unlock_read(old_map);
4538 * Check whether this task is allowed to have this page.
4540 prot = entry->protection;
4541 if ((fault_typea & VM_PROT_FAULT_LOOKUP) != 0) {
4542 fault_typea &= ~VM_PROT_FAULT_LOOKUP;
4543 if (prot == VM_PROT_NONE && map != kernel_map &&
4544 (entry->eflags & MAP_ENTRY_GUARD) != 0 &&
4545 (entry->eflags & (MAP_ENTRY_STACK_GAP_DN |
4546 MAP_ENTRY_STACK_GAP_UP)) != 0 &&
4547 vm_map_growstack(map, vaddr, entry) == KERN_SUCCESS)
4548 goto RetryLookupLocked;
4550 fault_type &= VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE;
4551 if ((fault_type & prot) != fault_type || prot == VM_PROT_NONE) {
4552 vm_map_unlock_read(map);
4553 return (KERN_PROTECTION_FAILURE);
4555 KASSERT((prot & VM_PROT_WRITE) == 0 || (entry->eflags &
4556 (MAP_ENTRY_USER_WIRED | MAP_ENTRY_NEEDS_COPY)) !=
4557 (MAP_ENTRY_USER_WIRED | MAP_ENTRY_NEEDS_COPY),
4558 ("entry %p flags %x", entry, entry->eflags));
4559 if ((fault_typea & VM_PROT_COPY) != 0 &&
4560 (entry->max_protection & VM_PROT_WRITE) == 0 &&
4561 (entry->eflags & MAP_ENTRY_COW) == 0) {
4562 vm_map_unlock_read(map);
4563 return (KERN_PROTECTION_FAILURE);
4567 * If this page is not pageable, we have to get it for all possible
4570 *wired = (entry->wired_count != 0);
4572 fault_type = entry->protection;
4573 size = entry->end - entry->start;
4575 * If the entry was copy-on-write, we either ...
4577 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
4579 * If we want to write the page, we may as well handle that
4580 * now since we've got the map locked.
4582 * If we don't need to write the page, we just demote the
4583 * permissions allowed.
4585 if ((fault_type & VM_PROT_WRITE) != 0 ||
4586 (fault_typea & VM_PROT_COPY) != 0) {
4588 * Make a new object, and place it in the object
4589 * chain. Note that no new references have appeared
4590 * -- one just moved from the map to the new
4593 if (vm_map_lock_upgrade(map))
4596 if (entry->cred == NULL) {
4598 * The debugger owner is charged for
4601 cred = curthread->td_ucred;
4603 if (!swap_reserve_by_cred(size, cred)) {
4606 return (KERN_RESOURCE_SHORTAGE);
4610 vm_object_shadow(&entry->object.vm_object,
4611 &entry->offset, size);
4612 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
4613 eobject = entry->object.vm_object;
4614 if (eobject->cred != NULL) {
4616 * The object was not shadowed.
4618 swap_release_by_cred(size, entry->cred);
4619 crfree(entry->cred);
4621 } else if (entry->cred != NULL) {
4622 VM_OBJECT_WLOCK(eobject);
4623 eobject->cred = entry->cred;
4624 eobject->charge = size;
4625 VM_OBJECT_WUNLOCK(eobject);
4629 vm_map_lock_downgrade(map);
4632 * We're attempting to read a copy-on-write page --
4633 * don't allow writes.
4635 prot &= ~VM_PROT_WRITE;
4640 * Create an object if necessary.
4642 if (entry->object.vm_object == NULL &&
4644 if (vm_map_lock_upgrade(map))
4646 entry->object.vm_object = vm_object_allocate(OBJT_DEFAULT,
4649 if (entry->cred != NULL) {
4650 VM_OBJECT_WLOCK(entry->object.vm_object);
4651 entry->object.vm_object->cred = entry->cred;
4652 entry->object.vm_object->charge = size;
4653 VM_OBJECT_WUNLOCK(entry->object.vm_object);
4656 vm_map_lock_downgrade(map);
4660 * Return the object/offset from this entry. If the entry was
4661 * copy-on-write or empty, it has been fixed up.
4663 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
4664 *object = entry->object.vm_object;
4667 return (KERN_SUCCESS);
4671 * vm_map_lookup_locked:
4673 * Lookup the faulting address. A version of vm_map_lookup that returns
4674 * KERN_FAILURE instead of blocking on map lock or memory allocation.
4677 vm_map_lookup_locked(vm_map_t *var_map, /* IN/OUT */
4679 vm_prot_t fault_typea,
4680 vm_map_entry_t *out_entry, /* OUT */
4681 vm_object_t *object, /* OUT */
4682 vm_pindex_t *pindex, /* OUT */
4683 vm_prot_t *out_prot, /* OUT */
4684 boolean_t *wired) /* OUT */
4686 vm_map_entry_t entry;
4687 vm_map_t map = *var_map;
4689 vm_prot_t fault_type = fault_typea;
4692 * Lookup the faulting address.
4694 if (!vm_map_lookup_entry(map, vaddr, out_entry))
4695 return (KERN_INVALID_ADDRESS);
4700 * Fail if the entry refers to a submap.
4702 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP)
4703 return (KERN_FAILURE);
4706 * Check whether this task is allowed to have this page.
4708 prot = entry->protection;
4709 fault_type &= VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE;
4710 if ((fault_type & prot) != fault_type)
4711 return (KERN_PROTECTION_FAILURE);
4714 * If this page is not pageable, we have to get it for all possible
4717 *wired = (entry->wired_count != 0);
4719 fault_type = entry->protection;
4721 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
4723 * Fail if the entry was copy-on-write for a write fault.
4725 if (fault_type & VM_PROT_WRITE)
4726 return (KERN_FAILURE);
4728 * We're attempting to read a copy-on-write page --
4729 * don't allow writes.
4731 prot &= ~VM_PROT_WRITE;
4735 * Fail if an object should be created.
4737 if (entry->object.vm_object == NULL && !map->system_map)
4738 return (KERN_FAILURE);
4741 * Return the object/offset from this entry. If the entry was
4742 * copy-on-write or empty, it has been fixed up.
4744 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
4745 *object = entry->object.vm_object;
4748 return (KERN_SUCCESS);
4752 * vm_map_lookup_done:
4754 * Releases locks acquired by a vm_map_lookup
4755 * (according to the handle returned by that lookup).
4758 vm_map_lookup_done(vm_map_t map, vm_map_entry_t entry)
4761 * Unlock the main-level map
4763 vm_map_unlock_read(map);
4767 vm_map_max_KBI(const struct vm_map *map)
4770 return (vm_map_max(map));
4774 vm_map_min_KBI(const struct vm_map *map)
4777 return (vm_map_min(map));
4781 vm_map_pmap_KBI(vm_map_t map)
4789 _vm_map_assert_consistent(vm_map_t map)
4791 vm_map_entry_t entry, prev;
4792 vm_size_t max_left, max_right;
4794 if (!enable_vmmap_check)
4797 prev = &map->header;
4798 VM_MAP_ENTRY_FOREACH(entry, map) {
4799 KASSERT(prev->end <= entry->start,
4800 ("map %p prev->end = %jx, start = %jx", map,
4801 (uintmax_t)prev->end, (uintmax_t)entry->start));
4802 KASSERT(entry->start < entry->end,
4803 ("map %p start = %jx, end = %jx", map,
4804 (uintmax_t)entry->start, (uintmax_t)entry->end));
4805 KASSERT(entry->end <= entry->next->start,
4806 ("map %p end = %jx, next->start = %jx", map,
4807 (uintmax_t)entry->end, (uintmax_t)entry->next->start));
4808 KASSERT(entry->left == NULL ||
4809 entry->left->start < entry->start,
4810 ("map %p left->start = %jx, start = %jx", map,
4811 (uintmax_t)entry->left->start, (uintmax_t)entry->start));
4812 KASSERT(entry->right == NULL ||
4813 entry->start < entry->right->start,
4814 ("map %p start = %jx, right->start = %jx", map,
4815 (uintmax_t)entry->start, (uintmax_t)entry->right->start));
4816 max_left = vm_map_entry_max_free_left(entry, entry->prev);
4817 max_right = vm_map_entry_max_free_right(entry, entry->next);
4818 KASSERT(entry->max_free == MAX(max_left, max_right),
4819 ("map %p max = %jx, max_left = %jx, max_right = %jx", map,
4820 (uintmax_t)entry->max_free,
4821 (uintmax_t)max_left, (uintmax_t)max_right));
4824 KASSERT(prev->end <= entry->start,
4825 ("map %p prev->end = %jx, start = %jx", map,
4826 (uintmax_t)prev->end, (uintmax_t)entry->start));
4830 #include "opt_ddb.h"
4832 #include <sys/kernel.h>
4834 #include <ddb/ddb.h>
4837 vm_map_print(vm_map_t map)
4839 vm_map_entry_t entry, prev;
4841 db_iprintf("Task map %p: pmap=%p, nentries=%d, version=%u\n",
4843 (void *)map->pmap, map->nentries, map->timestamp);
4846 prev = &map->header;
4847 VM_MAP_ENTRY_FOREACH(entry, map) {
4848 db_iprintf("map entry %p: start=%p, end=%p, eflags=%#x, \n",
4849 (void *)entry, (void *)entry->start, (void *)entry->end,
4852 static char *inheritance_name[4] =
4853 {"share", "copy", "none", "donate_copy"};
4855 db_iprintf(" prot=%x/%x/%s",
4857 entry->max_protection,
4858 inheritance_name[(int)(unsigned char)
4859 entry->inheritance]);
4860 if (entry->wired_count != 0)
4861 db_printf(", wired");
4863 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
4864 db_printf(", share=%p, offset=0x%jx\n",
4865 (void *)entry->object.sub_map,
4866 (uintmax_t)entry->offset);
4867 if (prev == &map->header ||
4868 prev->object.sub_map !=
4869 entry->object.sub_map) {
4871 vm_map_print((vm_map_t)entry->object.sub_map);
4875 if (entry->cred != NULL)
4876 db_printf(", ruid %d", entry->cred->cr_ruid);
4877 db_printf(", object=%p, offset=0x%jx",
4878 (void *)entry->object.vm_object,
4879 (uintmax_t)entry->offset);
4880 if (entry->object.vm_object && entry->object.vm_object->cred)
4881 db_printf(", obj ruid %d charge %jx",
4882 entry->object.vm_object->cred->cr_ruid,
4883 (uintmax_t)entry->object.vm_object->charge);
4884 if (entry->eflags & MAP_ENTRY_COW)
4885 db_printf(", copy (%s)",
4886 (entry->eflags & MAP_ENTRY_NEEDS_COPY) ? "needed" : "done");
4889 if (prev == &map->header ||
4890 prev->object.vm_object !=
4891 entry->object.vm_object) {
4893 vm_object_print((db_expr_t)(intptr_t)
4894 entry->object.vm_object,
4904 DB_SHOW_COMMAND(map, map)
4908 db_printf("usage: show map <addr>\n");
4911 vm_map_print((vm_map_t)addr);
4914 DB_SHOW_COMMAND(procvm, procvm)
4919 p = db_lookup_proc(addr);
4924 db_printf("p = %p, vmspace = %p, map = %p, pmap = %p\n",
4925 (void *)p, (void *)p->p_vmspace, (void *)&p->p_vmspace->vm_map,
4926 (void *)vmspace_pmap(p->p_vmspace));
4928 vm_map_print((vm_map_t)&p->p_vmspace->vm_map);