2 * SPDX-License-Identifier: (BSD-3-Clause AND MIT-CMU)
4 * Copyright (c) 1991, 1993
5 * The Regents of the University of California. All rights reserved.
7 * This code is derived from software contributed to Berkeley by
8 * The Mach Operating System project at Carnegie-Mellon University.
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 3. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * from: @(#)vm_map.c 8.3 (Berkeley) 1/12/94
37 * Copyright (c) 1987, 1990 Carnegie-Mellon University.
38 * All rights reserved.
40 * Authors: Avadis Tevanian, Jr., Michael Wayne Young
42 * Permission to use, copy, modify and distribute this software and
43 * its documentation is hereby granted, provided that both the copyright
44 * notice and this permission notice appear in all copies of the
45 * software, derivative works or modified versions, and any portions
46 * thereof, and that both notices appear in supporting documentation.
48 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
49 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
50 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
52 * Carnegie Mellon requests users of this software to return to
54 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
55 * School of Computer Science
56 * Carnegie Mellon University
57 * Pittsburgh PA 15213-3890
59 * any improvements or extensions that they make and grant Carnegie the
60 * rights to redistribute these changes.
64 * Virtual memory mapping module.
67 #include <sys/cdefs.h>
68 __FBSDID("$FreeBSD$");
70 #include <sys/param.h>
71 #include <sys/systm.h>
73 #include <sys/kernel.h>
76 #include <sys/mutex.h>
78 #include <sys/vmmeter.h>
80 #include <sys/vnode.h>
81 #include <sys/racct.h>
82 #include <sys/resourcevar.h>
83 #include <sys/rwlock.h>
85 #include <sys/sysctl.h>
86 #include <sys/sysent.h>
90 #include <vm/vm_param.h>
92 #include <vm/vm_map.h>
93 #include <vm/vm_page.h>
94 #include <vm/vm_pageout.h>
95 #include <vm/vm_object.h>
96 #include <vm/vm_pager.h>
97 #include <vm/vm_kern.h>
98 #include <vm/vm_extern.h>
99 #include <vm/vnode_pager.h>
100 #include <vm/swap_pager.h>
104 * Virtual memory maps provide for the mapping, protection,
105 * and sharing of virtual memory objects. In addition,
106 * this module provides for an efficient virtual copy of
107 * memory from one map to another.
109 * Synchronization is required prior to most operations.
111 * Maps consist of an ordered doubly-linked list of simple
112 * entries; a self-adjusting binary search tree of these
113 * entries is used to speed up lookups.
115 * Since portions of maps are specified by start/end addresses,
116 * which may not align with existing map entries, all
117 * routines merely "clip" entries to these start/end values.
118 * [That is, an entry is split into two, bordering at a
119 * start or end value.] Note that these clippings may not
120 * always be necessary (as the two resulting entries are then
121 * not changed); however, the clipping is done for convenience.
123 * As mentioned above, virtual copy operations are performed
124 * by copying VM object references from one map to
125 * another, and then marking both regions as copy-on-write.
128 static struct mtx map_sleep_mtx;
129 static uma_zone_t mapentzone;
130 static uma_zone_t kmapentzone;
131 static uma_zone_t mapzone;
132 static uma_zone_t vmspace_zone;
133 static int vmspace_zinit(void *mem, int size, int flags);
134 static int vm_map_zinit(void *mem, int ize, int flags);
135 static void _vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min,
137 static void vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map);
138 static void vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry);
139 static void vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry);
140 static int vm_map_growstack(vm_map_t map, vm_offset_t addr,
141 vm_map_entry_t gap_entry);
142 static void vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot,
143 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags);
145 static void vm_map_zdtor(void *mem, int size, void *arg);
146 static void vmspace_zdtor(void *mem, int size, void *arg);
148 static int vm_map_stack_locked(vm_map_t map, vm_offset_t addrbos,
149 vm_size_t max_ssize, vm_size_t growsize, vm_prot_t prot, vm_prot_t max,
151 static void vm_map_wire_entry_failure(vm_map_t map, vm_map_entry_t entry,
152 vm_offset_t failed_addr);
154 #define ENTRY_CHARGED(e) ((e)->cred != NULL || \
155 ((e)->object.vm_object != NULL && (e)->object.vm_object->cred != NULL && \
156 !((e)->eflags & MAP_ENTRY_NEEDS_COPY)))
159 * PROC_VMSPACE_{UN,}LOCK() can be a noop as long as vmspaces are type
162 #define PROC_VMSPACE_LOCK(p) do { } while (0)
163 #define PROC_VMSPACE_UNLOCK(p) do { } while (0)
166 * VM_MAP_RANGE_CHECK: [ internal use only ]
168 * Asserts that the starting and ending region
169 * addresses fall within the valid range of the map.
171 #define VM_MAP_RANGE_CHECK(map, start, end) \
173 if (start < vm_map_min(map)) \
174 start = vm_map_min(map); \
175 if (end > vm_map_max(map)) \
176 end = vm_map_max(map); \
184 * Initialize the vm_map module. Must be called before
185 * any other vm_map routines.
187 * Map and entry structures are allocated from the general
188 * purpose memory pool with some exceptions:
190 * - The kernel map and kmem submap are allocated statically.
191 * - Kernel map entries are allocated out of a static pool.
193 * These restrictions are necessary since malloc() uses the
194 * maps and requires map entries.
200 mtx_init(&map_sleep_mtx, "vm map sleep mutex", NULL, MTX_DEF);
201 mapzone = uma_zcreate("MAP", sizeof(struct vm_map), NULL,
207 vm_map_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
208 uma_prealloc(mapzone, MAX_KMAP);
209 kmapentzone = uma_zcreate("KMAP ENTRY", sizeof(struct vm_map_entry),
210 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR,
211 UMA_ZONE_MTXCLASS | UMA_ZONE_VM);
212 mapentzone = uma_zcreate("MAP ENTRY", sizeof(struct vm_map_entry),
213 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
214 vmspace_zone = uma_zcreate("VMSPACE", sizeof(struct vmspace), NULL,
220 vmspace_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
224 vmspace_zinit(void *mem, int size, int flags)
228 vm = (struct vmspace *)mem;
230 vm->vm_map.pmap = NULL;
231 (void)vm_map_zinit(&vm->vm_map, sizeof(vm->vm_map), flags);
232 PMAP_LOCK_INIT(vmspace_pmap(vm));
237 vm_map_zinit(void *mem, int size, int flags)
242 memset(map, 0, sizeof(*map));
243 mtx_init(&map->system_mtx, "vm map (system)", NULL, MTX_DEF | MTX_DUPOK);
244 sx_init(&map->lock, "vm map (user)");
250 vmspace_zdtor(void *mem, int size, void *arg)
254 vm = (struct vmspace *)mem;
256 vm_map_zdtor(&vm->vm_map, sizeof(vm->vm_map), arg);
259 vm_map_zdtor(void *mem, int size, void *arg)
264 KASSERT(map->nentries == 0,
265 ("map %p nentries == %d on free.",
266 map, map->nentries));
267 KASSERT(map->size == 0,
268 ("map %p size == %lu on free.",
269 map, (unsigned long)map->size));
271 #endif /* INVARIANTS */
274 * Allocate a vmspace structure, including a vm_map and pmap,
275 * and initialize those structures. The refcnt is set to 1.
277 * If 'pinit' is NULL then the embedded pmap is initialized via pmap_pinit().
280 vmspace_alloc(vm_offset_t min, vm_offset_t max, pmap_pinit_t pinit)
284 vm = uma_zalloc(vmspace_zone, M_WAITOK);
285 KASSERT(vm->vm_map.pmap == NULL, ("vm_map.pmap must be NULL"));
286 if (!pinit(vmspace_pmap(vm))) {
287 uma_zfree(vmspace_zone, vm);
290 CTR1(KTR_VM, "vmspace_alloc: %p", vm);
291 _vm_map_init(&vm->vm_map, vmspace_pmap(vm), min, max);
306 vmspace_container_reset(struct proc *p)
310 racct_set(p, RACCT_DATA, 0);
311 racct_set(p, RACCT_STACK, 0);
312 racct_set(p, RACCT_RSS, 0);
313 racct_set(p, RACCT_MEMLOCK, 0);
314 racct_set(p, RACCT_VMEM, 0);
320 vmspace_dofree(struct vmspace *vm)
323 CTR1(KTR_VM, "vmspace_free: %p", vm);
326 * Make sure any SysV shm is freed, it might not have been in
332 * Lock the map, to wait out all other references to it.
333 * Delete all of the mappings and pages they hold, then call
334 * the pmap module to reclaim anything left.
336 (void)vm_map_remove(&vm->vm_map, vm_map_min(&vm->vm_map),
337 vm_map_max(&vm->vm_map));
339 pmap_release(vmspace_pmap(vm));
340 vm->vm_map.pmap = NULL;
341 uma_zfree(vmspace_zone, vm);
345 vmspace_free(struct vmspace *vm)
348 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
349 "vmspace_free() called");
351 if (vm->vm_refcnt == 0)
352 panic("vmspace_free: attempt to free already freed vmspace");
354 if (atomic_fetchadd_int(&vm->vm_refcnt, -1) == 1)
359 vmspace_exitfree(struct proc *p)
363 PROC_VMSPACE_LOCK(p);
366 PROC_VMSPACE_UNLOCK(p);
367 KASSERT(vm == &vmspace0, ("vmspace_exitfree: wrong vmspace"));
372 vmspace_exit(struct thread *td)
379 * Release user portion of address space.
380 * This releases references to vnodes,
381 * which could cause I/O if the file has been unlinked.
382 * Need to do this early enough that we can still sleep.
384 * The last exiting process to reach this point releases as
385 * much of the environment as it can. vmspace_dofree() is the
386 * slower fallback in case another process had a temporary
387 * reference to the vmspace.
392 atomic_add_int(&vmspace0.vm_refcnt, 1);
393 refcnt = vm->vm_refcnt;
395 if (refcnt > 1 && p->p_vmspace != &vmspace0) {
396 /* Switch now since other proc might free vmspace */
397 PROC_VMSPACE_LOCK(p);
398 p->p_vmspace = &vmspace0;
399 PROC_VMSPACE_UNLOCK(p);
402 } while (!atomic_fcmpset_int(&vm->vm_refcnt, &refcnt, refcnt - 1));
404 if (p->p_vmspace != vm) {
405 /* vmspace not yet freed, switch back */
406 PROC_VMSPACE_LOCK(p);
408 PROC_VMSPACE_UNLOCK(p);
411 pmap_remove_pages(vmspace_pmap(vm));
412 /* Switch now since this proc will free vmspace */
413 PROC_VMSPACE_LOCK(p);
414 p->p_vmspace = &vmspace0;
415 PROC_VMSPACE_UNLOCK(p);
421 vmspace_container_reset(p);
425 /* Acquire reference to vmspace owned by another process. */
428 vmspace_acquire_ref(struct proc *p)
433 PROC_VMSPACE_LOCK(p);
436 PROC_VMSPACE_UNLOCK(p);
439 refcnt = vm->vm_refcnt;
441 if (refcnt <= 0) { /* Avoid 0->1 transition */
442 PROC_VMSPACE_UNLOCK(p);
445 } while (!atomic_fcmpset_int(&vm->vm_refcnt, &refcnt, refcnt + 1));
446 if (vm != p->p_vmspace) {
447 PROC_VMSPACE_UNLOCK(p);
451 PROC_VMSPACE_UNLOCK(p);
456 * Switch between vmspaces in an AIO kernel process.
458 * The new vmspace is either the vmspace of a user process obtained
459 * from an active AIO request or the initial vmspace of the AIO kernel
460 * process (when it is idling). Because user processes will block to
461 * drain any active AIO requests before proceeding in exit() or
462 * execve(), the reference count for vmspaces from AIO requests can
463 * never be 0. Similarly, AIO kernel processes hold an extra
464 * reference on their initial vmspace for the life of the process. As
465 * a result, the 'newvm' vmspace always has a non-zero reference
466 * count. This permits an additional reference on 'newvm' to be
467 * acquired via a simple atomic increment rather than the loop in
468 * vmspace_acquire_ref() above.
471 vmspace_switch_aio(struct vmspace *newvm)
473 struct vmspace *oldvm;
475 /* XXX: Need some way to assert that this is an aio daemon. */
477 KASSERT(newvm->vm_refcnt > 0,
478 ("vmspace_switch_aio: newvm unreferenced"));
480 oldvm = curproc->p_vmspace;
485 * Point to the new address space and refer to it.
487 curproc->p_vmspace = newvm;
488 atomic_add_int(&newvm->vm_refcnt, 1);
490 /* Activate the new mapping. */
491 pmap_activate(curthread);
497 _vm_map_lock(vm_map_t map, const char *file, int line)
501 mtx_lock_flags_(&map->system_mtx, 0, file, line);
503 sx_xlock_(&map->lock, file, line);
508 vm_map_entry_set_vnode_text(vm_map_entry_t entry, bool add)
510 vm_object_t object, object1;
513 if ((entry->eflags & MAP_ENTRY_VN_EXEC) == 0)
515 KASSERT((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0,
516 ("Submap with execs"));
517 object = entry->object.vm_object;
518 KASSERT(object != NULL, ("No object for text, entry %p", entry));
519 VM_OBJECT_RLOCK(object);
520 while ((object1 = object->backing_object) != NULL) {
521 VM_OBJECT_RLOCK(object1);
522 VM_OBJECT_RUNLOCK(object);
527 if (object->type == OBJT_DEAD) {
529 * For OBJT_DEAD objects, v_writecount was handled in
530 * vnode_pager_dealloc().
532 } else if (object->type == OBJT_VNODE) {
534 } else if (object->type == OBJT_SWAP) {
535 KASSERT((object->flags & OBJ_TMPFS_NODE) != 0,
536 ("vm_map_entry_set_vnode_text: swap and !TMPFS "
537 "entry %p, object %p, add %d", entry, object, add));
539 * Tmpfs VREG node, which was reclaimed, has
540 * OBJ_TMPFS_NODE flag set, but not OBJ_TMPFS. In
541 * this case there is no v_writecount to adjust.
543 if ((object->flags & OBJ_TMPFS) != 0)
544 vp = object->un_pager.swp.swp_tmpfs;
547 ("vm_map_entry_set_vnode_text: wrong object type, "
548 "entry %p, object %p, add %d", entry, object, add));
552 VOP_SET_TEXT_CHECKED(vp);
553 VM_OBJECT_RUNLOCK(object);
556 VM_OBJECT_RUNLOCK(object);
557 vn_lock(vp, LK_SHARED | LK_RETRY);
558 VOP_UNSET_TEXT_CHECKED(vp);
563 VM_OBJECT_RUNLOCK(object);
568 * Use a different name for this vm_map_entry field when it's use
569 * is not consistent with its use as part of an ordered search tree.
571 #define defer_next right
574 vm_map_process_deferred(void)
577 vm_map_entry_t entry, next;
581 entry = td->td_map_def_user;
582 td->td_map_def_user = NULL;
583 while (entry != NULL) {
584 next = entry->defer_next;
585 MPASS((entry->eflags & (MAP_ENTRY_WRITECNT |
586 MAP_ENTRY_VN_EXEC)) != (MAP_ENTRY_WRITECNT |
588 if ((entry->eflags & MAP_ENTRY_WRITECNT) != 0) {
590 * Decrement the object's writemappings and
591 * possibly the vnode's v_writecount.
593 KASSERT((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0,
594 ("Submap with writecount"));
595 object = entry->object.vm_object;
596 KASSERT(object != NULL, ("No object for writecount"));
597 vm_pager_release_writecount(object, entry->start,
600 vm_map_entry_set_vnode_text(entry, false);
601 vm_map_entry_deallocate(entry, FALSE);
608 _vm_map_assert_locked(vm_map_t map, const char *file, int line)
612 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
614 sx_assert_(&map->lock, SA_XLOCKED, file, line);
617 #define VM_MAP_ASSERT_LOCKED(map) \
618 _vm_map_assert_locked(map, LOCK_FILE, LOCK_LINE)
620 enum { VMMAP_CHECK_NONE, VMMAP_CHECK_UNLOCK, VMMAP_CHECK_ALL };
622 static int enable_vmmap_check = VMMAP_CHECK_UNLOCK;
624 static int enable_vmmap_check = VMMAP_CHECK_NONE;
626 SYSCTL_INT(_debug, OID_AUTO, vmmap_check, CTLFLAG_RWTUN,
627 &enable_vmmap_check, 0, "Enable vm map consistency checking");
629 static void _vm_map_assert_consistent(vm_map_t map, int check);
631 #define VM_MAP_ASSERT_CONSISTENT(map) \
632 _vm_map_assert_consistent(map, VMMAP_CHECK_ALL)
634 #define VM_MAP_UNLOCK_CONSISTENT(map) do { \
635 if (map->nupdates > map->nentries) { \
636 _vm_map_assert_consistent(map, VMMAP_CHECK_UNLOCK); \
641 #define VM_MAP_UNLOCK_CONSISTENT(map)
644 #define VM_MAP_ASSERT_LOCKED(map)
645 #define VM_MAP_ASSERT_CONSISTENT(map)
646 #define VM_MAP_UNLOCK_CONSISTENT(map)
647 #endif /* INVARIANTS */
650 _vm_map_unlock(vm_map_t map, const char *file, int line)
653 VM_MAP_UNLOCK_CONSISTENT(map);
655 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
657 sx_xunlock_(&map->lock, file, line);
658 vm_map_process_deferred();
663 _vm_map_lock_read(vm_map_t map, const char *file, int line)
667 mtx_lock_flags_(&map->system_mtx, 0, file, line);
669 sx_slock_(&map->lock, file, line);
673 _vm_map_unlock_read(vm_map_t map, const char *file, int line)
677 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
679 sx_sunlock_(&map->lock, file, line);
680 vm_map_process_deferred();
685 _vm_map_trylock(vm_map_t map, const char *file, int line)
689 error = map->system_map ?
690 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) :
691 !sx_try_xlock_(&map->lock, file, line);
698 _vm_map_trylock_read(vm_map_t map, const char *file, int line)
702 error = map->system_map ?
703 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) :
704 !sx_try_slock_(&map->lock, file, line);
709 * _vm_map_lock_upgrade: [ internal use only ]
711 * Tries to upgrade a read (shared) lock on the specified map to a write
712 * (exclusive) lock. Returns the value "0" if the upgrade succeeds and a
713 * non-zero value if the upgrade fails. If the upgrade fails, the map is
714 * returned without a read or write lock held.
716 * Requires that the map be read locked.
719 _vm_map_lock_upgrade(vm_map_t map, const char *file, int line)
721 unsigned int last_timestamp;
723 if (map->system_map) {
724 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
726 if (!sx_try_upgrade_(&map->lock, file, line)) {
727 last_timestamp = map->timestamp;
728 sx_sunlock_(&map->lock, file, line);
729 vm_map_process_deferred();
731 * If the map's timestamp does not change while the
732 * map is unlocked, then the upgrade succeeds.
734 sx_xlock_(&map->lock, file, line);
735 if (last_timestamp != map->timestamp) {
736 sx_xunlock_(&map->lock, file, line);
746 _vm_map_lock_downgrade(vm_map_t map, const char *file, int line)
749 if (map->system_map) {
750 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
752 VM_MAP_UNLOCK_CONSISTENT(map);
753 sx_downgrade_(&map->lock, file, line);
760 * Returns a non-zero value if the caller holds a write (exclusive) lock
761 * on the specified map and the value "0" otherwise.
764 vm_map_locked(vm_map_t map)
768 return (mtx_owned(&map->system_mtx));
770 return (sx_xlocked(&map->lock));
774 * _vm_map_unlock_and_wait:
776 * Atomically releases the lock on the specified map and puts the calling
777 * thread to sleep. The calling thread will remain asleep until either
778 * vm_map_wakeup() is performed on the map or the specified timeout is
781 * WARNING! This function does not perform deferred deallocations of
782 * objects and map entries. Therefore, the calling thread is expected to
783 * reacquire the map lock after reawakening and later perform an ordinary
784 * unlock operation, such as vm_map_unlock(), before completing its
785 * operation on the map.
788 _vm_map_unlock_and_wait(vm_map_t map, int timo, const char *file, int line)
791 VM_MAP_UNLOCK_CONSISTENT(map);
792 mtx_lock(&map_sleep_mtx);
794 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
796 sx_xunlock_(&map->lock, file, line);
797 return (msleep(&map->root, &map_sleep_mtx, PDROP | PVM, "vmmaps",
804 * Awaken any threads that have slept on the map using
805 * vm_map_unlock_and_wait().
808 vm_map_wakeup(vm_map_t map)
812 * Acquire and release map_sleep_mtx to prevent a wakeup()
813 * from being performed (and lost) between the map unlock
814 * and the msleep() in _vm_map_unlock_and_wait().
816 mtx_lock(&map_sleep_mtx);
817 mtx_unlock(&map_sleep_mtx);
822 vm_map_busy(vm_map_t map)
825 VM_MAP_ASSERT_LOCKED(map);
830 vm_map_unbusy(vm_map_t map)
833 VM_MAP_ASSERT_LOCKED(map);
834 KASSERT(map->busy, ("vm_map_unbusy: not busy"));
835 if (--map->busy == 0 && (map->flags & MAP_BUSY_WAKEUP)) {
836 vm_map_modflags(map, 0, MAP_BUSY_WAKEUP);
842 vm_map_wait_busy(vm_map_t map)
845 VM_MAP_ASSERT_LOCKED(map);
847 vm_map_modflags(map, MAP_BUSY_WAKEUP, 0);
849 msleep(&map->busy, &map->system_mtx, 0, "mbusy", 0);
851 sx_sleep(&map->busy, &map->lock, 0, "mbusy", 0);
857 vmspace_resident_count(struct vmspace *vmspace)
859 return pmap_resident_count(vmspace_pmap(vmspace));
865 * Creates and returns a new empty VM map with
866 * the given physical map structure, and having
867 * the given lower and upper address bounds.
870 vm_map_create(pmap_t pmap, vm_offset_t min, vm_offset_t max)
874 result = uma_zalloc(mapzone, M_WAITOK);
875 CTR1(KTR_VM, "vm_map_create: %p", result);
876 _vm_map_init(result, pmap, min, max);
881 * Initialize an existing vm_map structure
882 * such as that in the vmspace structure.
885 _vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max)
888 map->header.next = map->header.prev = &map->header;
889 map->header.eflags = MAP_ENTRY_HEADER;
890 map->needs_wakeup = FALSE;
893 map->header.end = min;
894 map->header.start = max;
906 vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max)
909 _vm_map_init(map, pmap, min, max);
910 mtx_init(&map->system_mtx, "system map", NULL, MTX_DEF | MTX_DUPOK);
911 sx_init(&map->lock, "user map");
915 * vm_map_entry_dispose: [ internal use only ]
917 * Inverse of vm_map_entry_create.
920 vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry)
922 uma_zfree(map->system_map ? kmapentzone : mapentzone, entry);
926 * vm_map_entry_create: [ internal use only ]
928 * Allocates a VM map entry for insertion.
929 * No entry fields are filled in.
931 static vm_map_entry_t
932 vm_map_entry_create(vm_map_t map)
934 vm_map_entry_t new_entry;
937 new_entry = uma_zalloc(kmapentzone, M_NOWAIT);
939 new_entry = uma_zalloc(mapentzone, M_WAITOK);
940 if (new_entry == NULL)
941 panic("vm_map_entry_create: kernel resources exhausted");
946 * vm_map_entry_set_behavior:
948 * Set the expected access behavior, either normal, random, or
952 vm_map_entry_set_behavior(vm_map_entry_t entry, u_char behavior)
954 entry->eflags = (entry->eflags & ~MAP_ENTRY_BEHAV_MASK) |
955 (behavior & MAP_ENTRY_BEHAV_MASK);
959 * vm_map_entry_max_free_{left,right}:
961 * Compute the size of the largest free gap between two entries,
962 * one the root of a tree and the other the ancestor of that root
963 * that is the least or greatest ancestor found on the search path.
965 static inline vm_size_t
966 vm_map_entry_max_free_left(vm_map_entry_t root, vm_map_entry_t left_ancestor)
969 return (root->left != NULL ?
970 root->left->max_free : root->start - left_ancestor->end);
973 static inline vm_size_t
974 vm_map_entry_max_free_right(vm_map_entry_t root, vm_map_entry_t right_ancestor)
977 return (root->right != NULL ?
978 root->right->max_free : right_ancestor->start - root->end);
981 #define SPLAY_LEFT_STEP(root, y, rlist, test) do { \
982 vm_size_t max_free; \
985 * Infer root->right->max_free == root->max_free when \
986 * y->max_free < root->max_free || root->max_free == 0. \
987 * Otherwise, look right to find it. \
990 max_free = root->max_free; \
991 KASSERT(max_free >= vm_map_entry_max_free_right(root, rlist), \
992 ("%s: max_free invariant fails", __func__)); \
993 if (y == NULL ? max_free > 0 : max_free - 1 < y->max_free) \
994 max_free = vm_map_entry_max_free_right(root, rlist); \
995 if (y != NULL && (test)) { \
996 /* Rotate right and make y root. */ \
997 root->left = y->right; \
999 if (max_free < y->max_free) \
1000 root->max_free = max_free = MAX(max_free, \
1001 vm_map_entry_max_free_left(root, y)); \
1005 /* Copy right->max_free. Put root on rlist. */ \
1006 root->max_free = max_free; \
1007 KASSERT(max_free == vm_map_entry_max_free_right(root, rlist), \
1008 ("%s: max_free not copied from right", __func__)); \
1009 root->left = rlist; \
1014 #define SPLAY_RIGHT_STEP(root, y, llist, test) do { \
1015 vm_size_t max_free; \
1018 * Infer root->left->max_free == root->max_free when \
1019 * y->max_free < root->max_free || root->max_free == 0. \
1020 * Otherwise, look left to find it. \
1023 max_free = root->max_free; \
1024 KASSERT(max_free >= vm_map_entry_max_free_left(root, llist), \
1025 ("%s: max_free invariant fails", __func__)); \
1026 if (y == NULL ? max_free > 0 : max_free - 1 < y->max_free) \
1027 max_free = vm_map_entry_max_free_left(root, llist); \
1028 if (y != NULL && (test)) { \
1029 /* Rotate left and make y root. */ \
1030 root->right = y->left; \
1032 if (max_free < y->max_free) \
1033 root->max_free = max_free = MAX(max_free, \
1034 vm_map_entry_max_free_right(root, y)); \
1038 /* Copy left->max_free. Put root on llist. */ \
1039 root->max_free = max_free; \
1040 KASSERT(max_free == vm_map_entry_max_free_left(root, llist), \
1041 ("%s: max_free not copied from left", __func__)); \
1042 root->right = llist; \
1048 * Walk down the tree until we find addr or a NULL pointer where addr would go,
1049 * breaking off left and right subtrees of nodes less than, or greater than
1050 * addr. Treat pointers to nodes with max_free < length as NULL pointers.
1051 * llist and rlist are the two sides in reverse order (bottom-up), with llist
1052 * linked by the right pointer and rlist linked by the left pointer in the
1053 * vm_map_entry, and both lists terminated by &map->header. This function, and
1054 * the subsequent call to vm_map_splay_merge, rely on the start and end address
1055 * values in &map->header.
1057 static vm_map_entry_t
1058 vm_map_splay_split(vm_map_t map, vm_offset_t addr, vm_size_t length,
1059 vm_map_entry_t *out_llist, vm_map_entry_t *out_rlist)
1061 vm_map_entry_t llist, rlist, root, y;
1063 llist = rlist = &map->header;
1065 while (root != NULL && root->max_free >= length) {
1066 KASSERT(llist->end <= root->start && root->end <= rlist->start,
1067 ("%s: root not within tree bounds", __func__));
1068 if (addr < root->start) {
1069 SPLAY_LEFT_STEP(root, y, rlist,
1070 y->max_free >= length && addr < y->start);
1071 } else if (addr >= root->end) {
1072 SPLAY_RIGHT_STEP(root, y, llist,
1073 y->max_free >= length && addr >= y->end);
1083 vm_map_splay_findnext(vm_map_entry_t root, vm_map_entry_t *iolist)
1085 vm_map_entry_t rlist, y;
1089 while (root != NULL)
1090 SPLAY_LEFT_STEP(root, y, rlist, true);
1095 vm_map_splay_findprev(vm_map_entry_t root, vm_map_entry_t *iolist)
1097 vm_map_entry_t llist, y;
1101 while (root != NULL)
1102 SPLAY_RIGHT_STEP(root, y, llist, true);
1107 vm_map_entry_swap(vm_map_entry_t *a, vm_map_entry_t *b)
1117 * Walk back up the two spines, flip the pointers and set max_free. The
1118 * subtrees of the root go at the bottom of llist and rlist.
1121 vm_map_splay_merge(vm_map_t map, vm_map_entry_t root,
1122 vm_map_entry_t llist, vm_map_entry_t rlist)
1124 vm_map_entry_t prev;
1125 vm_size_t max_free_left, max_free_right;
1127 max_free_left = vm_map_entry_max_free_left(root, llist);
1128 if (llist != &map->header) {
1132 * The max_free values of the children of llist are in
1133 * llist->max_free and max_free_left. Update with the
1136 llist->max_free = max_free_left =
1137 MAX(llist->max_free, max_free_left);
1138 vm_map_entry_swap(&llist->right, &prev);
1139 vm_map_entry_swap(&prev, &llist);
1140 } while (llist != &map->header);
1143 max_free_right = vm_map_entry_max_free_right(root, rlist);
1144 if (rlist != &map->header) {
1148 * The max_free values of the children of rlist are in
1149 * rlist->max_free and max_free_right. Update with the
1152 rlist->max_free = max_free_right =
1153 MAX(rlist->max_free, max_free_right);
1154 vm_map_entry_swap(&rlist->left, &prev);
1155 vm_map_entry_swap(&prev, &rlist);
1156 } while (rlist != &map->header);
1159 root->max_free = MAX(max_free_left, max_free_right);
1169 * The Sleator and Tarjan top-down splay algorithm with the
1170 * following variation. Max_free must be computed bottom-up, so
1171 * on the downward pass, maintain the left and right spines in
1172 * reverse order. Then, make a second pass up each side to fix
1173 * the pointers and compute max_free. The time bound is O(log n)
1176 * The new root is the vm_map_entry containing "addr", or else an
1177 * adjacent entry (lower if possible) if addr is not in the tree.
1179 * The map must be locked, and leaves it so.
1181 * Returns: the new root.
1183 static vm_map_entry_t
1184 vm_map_splay(vm_map_t map, vm_offset_t addr)
1186 vm_map_entry_t llist, rlist, root;
1188 root = vm_map_splay_split(map, addr, 0, &llist, &rlist);
1191 } else if (llist != &map->header) {
1193 * Recover the greatest node in the left
1194 * subtree and make it the root.
1197 llist = root->right;
1199 } else if (rlist != &map->header) {
1201 * Recover the least node in the right
1202 * subtree and make it the root.
1208 /* There is no root. */
1211 vm_map_splay_merge(map, root, llist, rlist);
1212 VM_MAP_ASSERT_CONSISTENT(map);
1217 * vm_map_entry_{un,}link:
1219 * Insert/remove entries from maps.
1222 vm_map_entry_link(vm_map_t map, vm_map_entry_t entry)
1224 vm_map_entry_t llist, rlist, root;
1227 "vm_map_entry_link: map %p, nentries %d, entry %p", map,
1228 map->nentries, entry);
1229 VM_MAP_ASSERT_LOCKED(map);
1231 root = vm_map_splay_split(map, entry->start, 0, &llist, &rlist);
1232 KASSERT(root == NULL,
1233 ("vm_map_entry_link: link object already mapped"));
1234 entry->prev = llist;
1235 entry->next = rlist;
1236 llist->next = rlist->prev = entry;
1237 entry->left = entry->right = NULL;
1238 vm_map_splay_merge(map, entry, llist, rlist);
1239 VM_MAP_ASSERT_CONSISTENT(map);
1242 enum unlink_merge_type {
1248 vm_map_entry_unlink(vm_map_t map, vm_map_entry_t entry,
1249 enum unlink_merge_type op)
1251 vm_map_entry_t llist, rlist, root, y;
1253 VM_MAP_ASSERT_LOCKED(map);
1254 root = vm_map_splay_split(map, entry->start, 0, &llist, &rlist);
1255 KASSERT(root != NULL,
1256 ("vm_map_entry_unlink: unlink object not mapped"));
1258 vm_map_splay_findnext(root, &rlist);
1260 case UNLINK_MERGE_NEXT:
1261 rlist->start = root->start;
1262 rlist->offset = root->offset;
1268 case UNLINK_MERGE_NONE:
1269 vm_map_splay_findprev(root, &llist);
1270 if (llist != &map->header) {
1272 llist = root->right;
1274 } else if (rlist != &map->header) {
1283 y->prev = entry->prev;
1286 vm_map_splay_merge(map, root, llist, rlist);
1289 VM_MAP_ASSERT_CONSISTENT(map);
1291 CTR3(KTR_VM, "vm_map_entry_unlink: map %p, nentries %d, entry %p", map,
1292 map->nentries, entry);
1296 * vm_map_entry_resize:
1298 * Resize a vm_map_entry, recompute the amount of free space that
1299 * follows it and propagate that value up the tree.
1301 * The map must be locked, and leaves it so.
1304 vm_map_entry_resize(vm_map_t map, vm_map_entry_t entry, vm_size_t grow_amount)
1306 vm_map_entry_t llist, rlist, root;
1308 VM_MAP_ASSERT_LOCKED(map);
1309 root = vm_map_splay_split(map, entry->start, 0, &llist, &rlist);
1310 KASSERT(root != NULL,
1311 ("%s: resize object not mapped", __func__));
1312 vm_map_splay_findnext(root, &rlist);
1314 entry->end += grow_amount;
1315 vm_map_splay_merge(map, root, llist, rlist);
1316 VM_MAP_ASSERT_CONSISTENT(map);
1317 CTR4(KTR_VM, "%s: map %p, nentries %d, entry %p",
1318 __func__, map, map->nentries, entry);
1322 * vm_map_lookup_entry: [ internal use only ]
1324 * Finds the map entry containing (or
1325 * immediately preceding) the specified address
1326 * in the given map; the entry is returned
1327 * in the "entry" parameter. The boolean
1328 * result indicates whether the address is
1329 * actually contained in the map.
1332 vm_map_lookup_entry(
1334 vm_offset_t address,
1335 vm_map_entry_t *entry) /* OUT */
1337 vm_map_entry_t cur, lbound;
1341 * If the map is empty, then the map entry immediately preceding
1342 * "address" is the map's header.
1346 *entry = &map->header;
1349 if (address >= cur->start && cur->end > address) {
1353 if ((locked = vm_map_locked(map)) ||
1354 sx_try_upgrade(&map->lock)) {
1356 * Splay requires a write lock on the map. However, it only
1357 * restructures the binary search tree; it does not otherwise
1358 * change the map. Thus, the map's timestamp need not change
1359 * on a temporary upgrade.
1361 cur = vm_map_splay(map, address);
1363 VM_MAP_UNLOCK_CONSISTENT(map);
1364 sx_downgrade(&map->lock);
1368 * If "address" is contained within a map entry, the new root
1369 * is that map entry. Otherwise, the new root is a map entry
1370 * immediately before or after "address".
1372 if (address < cur->start) {
1373 *entry = &map->header;
1377 return (address < cur->end);
1380 * Since the map is only locked for read access, perform a
1381 * standard binary search tree lookup for "address".
1383 lbound = &map->header;
1385 if (address < cur->start) {
1387 } else if (cur->end <= address) {
1394 } while (cur != NULL);
1402 * Inserts the given whole VM object into the target
1403 * map at the specified address range. The object's
1404 * size should match that of the address range.
1406 * Requires that the map be locked, and leaves it so.
1408 * If object is non-NULL, ref count must be bumped by caller
1409 * prior to making call to account for the new entry.
1412 vm_map_insert(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1413 vm_offset_t start, vm_offset_t end, vm_prot_t prot, vm_prot_t max, int cow)
1415 vm_map_entry_t new_entry, prev_entry;
1417 vm_eflags_t protoeflags;
1418 vm_inherit_t inheritance;
1420 VM_MAP_ASSERT_LOCKED(map);
1421 KASSERT(object != kernel_object ||
1422 (cow & MAP_COPY_ON_WRITE) == 0,
1423 ("vm_map_insert: kernel object and COW"));
1424 KASSERT(object == NULL || (cow & MAP_NOFAULT) == 0,
1425 ("vm_map_insert: paradoxical MAP_NOFAULT request"));
1426 KASSERT((prot & ~max) == 0,
1427 ("prot %#x is not subset of max_prot %#x", prot, max));
1430 * Check that the start and end points are not bogus.
1432 if (start < vm_map_min(map) || end > vm_map_max(map) ||
1434 return (KERN_INVALID_ADDRESS);
1437 * Find the entry prior to the proposed starting address; if it's part
1438 * of an existing entry, this range is bogus.
1440 if (vm_map_lookup_entry(map, start, &prev_entry))
1441 return (KERN_NO_SPACE);
1444 * Assert that the next entry doesn't overlap the end point.
1446 if (vm_map_entry_succ(prev_entry)->start < end)
1447 return (KERN_NO_SPACE);
1449 if ((cow & MAP_CREATE_GUARD) != 0 && (object != NULL ||
1450 max != VM_PROT_NONE))
1451 return (KERN_INVALID_ARGUMENT);
1454 if (cow & MAP_COPY_ON_WRITE)
1455 protoeflags |= MAP_ENTRY_COW | MAP_ENTRY_NEEDS_COPY;
1456 if (cow & MAP_NOFAULT)
1457 protoeflags |= MAP_ENTRY_NOFAULT;
1458 if (cow & MAP_DISABLE_SYNCER)
1459 protoeflags |= MAP_ENTRY_NOSYNC;
1460 if (cow & MAP_DISABLE_COREDUMP)
1461 protoeflags |= MAP_ENTRY_NOCOREDUMP;
1462 if (cow & MAP_STACK_GROWS_DOWN)
1463 protoeflags |= MAP_ENTRY_GROWS_DOWN;
1464 if (cow & MAP_STACK_GROWS_UP)
1465 protoeflags |= MAP_ENTRY_GROWS_UP;
1466 if (cow & MAP_WRITECOUNT)
1467 protoeflags |= MAP_ENTRY_WRITECNT;
1468 if (cow & MAP_VN_EXEC)
1469 protoeflags |= MAP_ENTRY_VN_EXEC;
1470 if ((cow & MAP_CREATE_GUARD) != 0)
1471 protoeflags |= MAP_ENTRY_GUARD;
1472 if ((cow & MAP_CREATE_STACK_GAP_DN) != 0)
1473 protoeflags |= MAP_ENTRY_STACK_GAP_DN;
1474 if ((cow & MAP_CREATE_STACK_GAP_UP) != 0)
1475 protoeflags |= MAP_ENTRY_STACK_GAP_UP;
1476 if (cow & MAP_INHERIT_SHARE)
1477 inheritance = VM_INHERIT_SHARE;
1479 inheritance = VM_INHERIT_DEFAULT;
1482 if ((cow & (MAP_ACC_NO_CHARGE | MAP_NOFAULT | MAP_CREATE_GUARD)) != 0)
1484 if ((cow & MAP_ACC_CHARGED) || ((prot & VM_PROT_WRITE) &&
1485 ((protoeflags & MAP_ENTRY_NEEDS_COPY) || object == NULL))) {
1486 if (!(cow & MAP_ACC_CHARGED) && !swap_reserve(end - start))
1487 return (KERN_RESOURCE_SHORTAGE);
1488 KASSERT(object == NULL ||
1489 (protoeflags & MAP_ENTRY_NEEDS_COPY) != 0 ||
1490 object->cred == NULL,
1491 ("overcommit: vm_map_insert o %p", object));
1492 cred = curthread->td_ucred;
1496 /* Expand the kernel pmap, if necessary. */
1497 if (map == kernel_map && end > kernel_vm_end)
1498 pmap_growkernel(end);
1499 if (object != NULL) {
1501 * OBJ_ONEMAPPING must be cleared unless this mapping
1502 * is trivially proven to be the only mapping for any
1503 * of the object's pages. (Object granularity
1504 * reference counting is insufficient to recognize
1505 * aliases with precision.)
1507 VM_OBJECT_WLOCK(object);
1508 if (object->ref_count > 1 || object->shadow_count != 0)
1509 vm_object_clear_flag(object, OBJ_ONEMAPPING);
1510 VM_OBJECT_WUNLOCK(object);
1511 } else if ((prev_entry->eflags & ~MAP_ENTRY_USER_WIRED) ==
1513 (cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP |
1514 MAP_VN_EXEC)) == 0 &&
1515 prev_entry->end == start && (prev_entry->cred == cred ||
1516 (prev_entry->object.vm_object != NULL &&
1517 prev_entry->object.vm_object->cred == cred)) &&
1518 vm_object_coalesce(prev_entry->object.vm_object,
1520 (vm_size_t)(prev_entry->end - prev_entry->start),
1521 (vm_size_t)(end - prev_entry->end), cred != NULL &&
1522 (protoeflags & MAP_ENTRY_NEEDS_COPY) == 0)) {
1524 * We were able to extend the object. Determine if we
1525 * can extend the previous map entry to include the
1526 * new range as well.
1528 if (prev_entry->inheritance == inheritance &&
1529 prev_entry->protection == prot &&
1530 prev_entry->max_protection == max &&
1531 prev_entry->wired_count == 0) {
1532 KASSERT((prev_entry->eflags & MAP_ENTRY_USER_WIRED) ==
1533 0, ("prev_entry %p has incoherent wiring",
1535 if ((prev_entry->eflags & MAP_ENTRY_GUARD) == 0)
1536 map->size += end - prev_entry->end;
1537 vm_map_entry_resize(map, prev_entry,
1538 end - prev_entry->end);
1539 vm_map_try_merge_entries(map, prev_entry,
1540 vm_map_entry_succ(prev_entry));
1541 return (KERN_SUCCESS);
1545 * If we can extend the object but cannot extend the
1546 * map entry, we have to create a new map entry. We
1547 * must bump the ref count on the extended object to
1548 * account for it. object may be NULL.
1550 object = prev_entry->object.vm_object;
1551 offset = prev_entry->offset +
1552 (prev_entry->end - prev_entry->start);
1553 vm_object_reference(object);
1554 if (cred != NULL && object != NULL && object->cred != NULL &&
1555 !(prev_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
1556 /* Object already accounts for this uid. */
1564 * Create a new entry
1566 new_entry = vm_map_entry_create(map);
1567 new_entry->start = start;
1568 new_entry->end = end;
1569 new_entry->cred = NULL;
1571 new_entry->eflags = protoeflags;
1572 new_entry->object.vm_object = object;
1573 new_entry->offset = offset;
1575 new_entry->inheritance = inheritance;
1576 new_entry->protection = prot;
1577 new_entry->max_protection = max;
1578 new_entry->wired_count = 0;
1579 new_entry->wiring_thread = NULL;
1580 new_entry->read_ahead = VM_FAULT_READ_AHEAD_INIT;
1581 new_entry->next_read = start;
1583 KASSERT(cred == NULL || !ENTRY_CHARGED(new_entry),
1584 ("overcommit: vm_map_insert leaks vm_map %p", new_entry));
1585 new_entry->cred = cred;
1588 * Insert the new entry into the list
1590 vm_map_entry_link(map, new_entry);
1591 if ((new_entry->eflags & MAP_ENTRY_GUARD) == 0)
1592 map->size += new_entry->end - new_entry->start;
1595 * Try to coalesce the new entry with both the previous and next
1596 * entries in the list. Previously, we only attempted to coalesce
1597 * with the previous entry when object is NULL. Here, we handle the
1598 * other cases, which are less common.
1600 vm_map_try_merge_entries(map, prev_entry, new_entry);
1601 vm_map_try_merge_entries(map, new_entry, vm_map_entry_succ(new_entry));
1603 if ((cow & (MAP_PREFAULT | MAP_PREFAULT_PARTIAL)) != 0) {
1604 vm_map_pmap_enter(map, start, prot, object, OFF_TO_IDX(offset),
1605 end - start, cow & MAP_PREFAULT_PARTIAL);
1608 return (KERN_SUCCESS);
1614 * Find the first fit (lowest VM address) for "length" free bytes
1615 * beginning at address >= start in the given map.
1617 * In a vm_map_entry, "max_free" is the maximum amount of
1618 * contiguous free space between an entry in its subtree and a
1619 * neighbor of that entry. This allows finding a free region in
1620 * one path down the tree, so O(log n) amortized with splay
1623 * The map must be locked, and leaves it so.
1625 * Returns: starting address if sufficient space,
1626 * vm_map_max(map)-length+1 if insufficient space.
1629 vm_map_findspace(vm_map_t map, vm_offset_t start, vm_size_t length)
1631 vm_map_entry_t llist, rlist, root, y;
1632 vm_size_t left_length;
1633 vm_offset_t gap_end;
1636 * Request must fit within min/max VM address and must avoid
1639 start = MAX(start, vm_map_min(map));
1640 if (start >= vm_map_max(map) || length > vm_map_max(map) - start)
1641 return (vm_map_max(map) - length + 1);
1643 /* Empty tree means wide open address space. */
1644 if (map->root == NULL)
1648 * After splay_split, if start is within an entry, push it to the start
1649 * of the following gap. If rlist is at the end of the gap containing
1650 * start, save the end of that gap in gap_end to see if the gap is big
1651 * enough; otherwise set gap_end to start skip gap-checking and move
1652 * directly to a search of the right subtree.
1654 root = vm_map_splay_split(map, start, length, &llist, &rlist);
1655 gap_end = rlist->start;
1658 if (root->right != NULL)
1660 } else if (rlist != &map->header) {
1666 llist = root->right;
1669 vm_map_splay_merge(map, root, llist, rlist);
1670 VM_MAP_ASSERT_CONSISTENT(map);
1671 if (length <= gap_end - start)
1674 /* With max_free, can immediately tell if no solution. */
1675 if (root->right == NULL || length > root->right->max_free)
1676 return (vm_map_max(map) - length + 1);
1679 * Splay for the least large-enough gap in the right subtree.
1681 llist = rlist = &map->header;
1682 for (left_length = 0;;
1683 left_length = vm_map_entry_max_free_left(root, llist)) {
1684 if (length <= left_length)
1685 SPLAY_LEFT_STEP(root, y, rlist,
1686 length <= vm_map_entry_max_free_left(y, llist));
1688 SPLAY_RIGHT_STEP(root, y, llist,
1689 length > vm_map_entry_max_free_left(y, root));
1694 llist = root->right;
1696 if (rlist != &map->header) {
1700 vm_map_splay_merge(map, y, &map->header, rlist);
1702 vm_map_entry_max_free_left(y, root),
1703 vm_map_entry_max_free_right(y, &map->header));
1706 vm_map_splay_merge(map, root, llist, &map->header);
1707 VM_MAP_ASSERT_CONSISTENT(map);
1712 vm_map_fixed(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1713 vm_offset_t start, vm_size_t length, vm_prot_t prot,
1714 vm_prot_t max, int cow)
1719 end = start + length;
1720 KASSERT((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 ||
1722 ("vm_map_fixed: non-NULL backing object for stack"));
1724 VM_MAP_RANGE_CHECK(map, start, end);
1725 if ((cow & MAP_CHECK_EXCL) == 0)
1726 vm_map_delete(map, start, end);
1727 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) != 0) {
1728 result = vm_map_stack_locked(map, start, length, sgrowsiz,
1731 result = vm_map_insert(map, object, offset, start, end,
1738 static const int aslr_pages_rnd_64[2] = {0x1000, 0x10};
1739 static const int aslr_pages_rnd_32[2] = {0x100, 0x4};
1741 static int cluster_anon = 1;
1742 SYSCTL_INT(_vm, OID_AUTO, cluster_anon, CTLFLAG_RW,
1744 "Cluster anonymous mappings: 0 = no, 1 = yes if no hint, 2 = always");
1747 clustering_anon_allowed(vm_offset_t addr)
1750 switch (cluster_anon) {
1761 static long aslr_restarts;
1762 SYSCTL_LONG(_vm, OID_AUTO, aslr_restarts, CTLFLAG_RD,
1764 "Number of aslr failures");
1766 #define MAP_32BIT_MAX_ADDR ((vm_offset_t)1 << 31)
1769 * Searches for the specified amount of free space in the given map with the
1770 * specified alignment. Performs an address-ordered, first-fit search from
1771 * the given address "*addr", with an optional upper bound "max_addr". If the
1772 * parameter "alignment" is zero, then the alignment is computed from the
1773 * given (object, offset) pair so as to enable the greatest possible use of
1774 * superpage mappings. Returns KERN_SUCCESS and the address of the free space
1775 * in "*addr" if successful. Otherwise, returns KERN_NO_SPACE.
1777 * The map must be locked. Initially, there must be at least "length" bytes
1778 * of free space at the given address.
1781 vm_map_alignspace(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1782 vm_offset_t *addr, vm_size_t length, vm_offset_t max_addr,
1783 vm_offset_t alignment)
1785 vm_offset_t aligned_addr, free_addr;
1787 VM_MAP_ASSERT_LOCKED(map);
1789 KASSERT(free_addr == vm_map_findspace(map, free_addr, length),
1790 ("caller failed to provide space %#jx at address %p",
1791 (uintmax_t)length, (void *)free_addr));
1794 * At the start of every iteration, the free space at address
1795 * "*addr" is at least "length" bytes.
1798 pmap_align_superpage(object, offset, addr, length);
1799 else if ((*addr & (alignment - 1)) != 0) {
1800 *addr &= ~(alignment - 1);
1803 aligned_addr = *addr;
1804 if (aligned_addr == free_addr) {
1806 * Alignment did not change "*addr", so "*addr" must
1807 * still provide sufficient free space.
1809 return (KERN_SUCCESS);
1813 * Test for address wrap on "*addr". A wrapped "*addr" could
1814 * be a valid address, in which case vm_map_findspace() cannot
1815 * be relied upon to fail.
1817 if (aligned_addr < free_addr)
1818 return (KERN_NO_SPACE);
1819 *addr = vm_map_findspace(map, aligned_addr, length);
1820 if (*addr + length > vm_map_max(map) ||
1821 (max_addr != 0 && *addr + length > max_addr))
1822 return (KERN_NO_SPACE);
1824 if (free_addr == aligned_addr) {
1826 * If a successful call to vm_map_findspace() did not
1827 * change "*addr", then "*addr" must still be aligned
1828 * and provide sufficient free space.
1830 return (KERN_SUCCESS);
1836 * vm_map_find finds an unallocated region in the target address
1837 * map with the given length. The search is defined to be
1838 * first-fit from the specified address; the region found is
1839 * returned in the same parameter.
1841 * If object is non-NULL, ref count must be bumped by caller
1842 * prior to making call to account for the new entry.
1845 vm_map_find(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1846 vm_offset_t *addr, /* IN/OUT */
1847 vm_size_t length, vm_offset_t max_addr, int find_space,
1848 vm_prot_t prot, vm_prot_t max, int cow)
1850 vm_offset_t alignment, curr_min_addr, min_addr;
1851 int gap, pidx, rv, try;
1852 bool cluster, en_aslr, update_anon;
1854 KASSERT((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 ||
1856 ("vm_map_find: non-NULL backing object for stack"));
1857 MPASS((cow & MAP_REMAP) == 0 || (find_space == VMFS_NO_SPACE &&
1858 (cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0));
1859 if (find_space == VMFS_OPTIMAL_SPACE && (object == NULL ||
1860 (object->flags & OBJ_COLORED) == 0))
1861 find_space = VMFS_ANY_SPACE;
1862 if (find_space >> 8 != 0) {
1863 KASSERT((find_space & 0xff) == 0, ("bad VMFS flags"));
1864 alignment = (vm_offset_t)1 << (find_space >> 8);
1867 en_aslr = (map->flags & MAP_ASLR) != 0;
1868 update_anon = cluster = clustering_anon_allowed(*addr) &&
1869 (map->flags & MAP_IS_SUB_MAP) == 0 && max_addr == 0 &&
1870 find_space != VMFS_NO_SPACE && object == NULL &&
1871 (cow & (MAP_INHERIT_SHARE | MAP_STACK_GROWS_UP |
1872 MAP_STACK_GROWS_DOWN)) == 0 && prot != PROT_NONE;
1873 curr_min_addr = min_addr = *addr;
1874 if (en_aslr && min_addr == 0 && !cluster &&
1875 find_space != VMFS_NO_SPACE &&
1876 (map->flags & MAP_ASLR_IGNSTART) != 0)
1877 curr_min_addr = min_addr = vm_map_min(map);
1881 curr_min_addr = map->anon_loc;
1882 if (curr_min_addr == 0)
1885 if (find_space != VMFS_NO_SPACE) {
1886 KASSERT(find_space == VMFS_ANY_SPACE ||
1887 find_space == VMFS_OPTIMAL_SPACE ||
1888 find_space == VMFS_SUPER_SPACE ||
1889 alignment != 0, ("unexpected VMFS flag"));
1892 * When creating an anonymous mapping, try clustering
1893 * with an existing anonymous mapping first.
1895 * We make up to two attempts to find address space
1896 * for a given find_space value. The first attempt may
1897 * apply randomization or may cluster with an existing
1898 * anonymous mapping. If this first attempt fails,
1899 * perform a first-fit search of the available address
1902 * If all tries failed, and find_space is
1903 * VMFS_OPTIMAL_SPACE, fallback to VMFS_ANY_SPACE.
1904 * Again enable clustering and randomization.
1911 * Second try: we failed either to find a
1912 * suitable region for randomizing the
1913 * allocation, or to cluster with an existing
1914 * mapping. Retry with free run.
1916 curr_min_addr = (map->flags & MAP_ASLR_IGNSTART) != 0 ?
1917 vm_map_min(map) : min_addr;
1918 atomic_add_long(&aslr_restarts, 1);
1921 if (try == 1 && en_aslr && !cluster) {
1923 * Find space for allocation, including
1924 * gap needed for later randomization.
1926 pidx = MAXPAGESIZES > 1 && pagesizes[1] != 0 &&
1927 (find_space == VMFS_SUPER_SPACE || find_space ==
1928 VMFS_OPTIMAL_SPACE) ? 1 : 0;
1929 gap = vm_map_max(map) > MAP_32BIT_MAX_ADDR &&
1930 (max_addr == 0 || max_addr > MAP_32BIT_MAX_ADDR) ?
1931 aslr_pages_rnd_64[pidx] : aslr_pages_rnd_32[pidx];
1932 *addr = vm_map_findspace(map, curr_min_addr,
1933 length + gap * pagesizes[pidx]);
1934 if (*addr + length + gap * pagesizes[pidx] >
1937 /* And randomize the start address. */
1938 *addr += (arc4random() % gap) * pagesizes[pidx];
1939 if (max_addr != 0 && *addr + length > max_addr)
1942 *addr = vm_map_findspace(map, curr_min_addr, length);
1943 if (*addr + length > vm_map_max(map) ||
1944 (max_addr != 0 && *addr + length > max_addr)) {
1955 if (find_space != VMFS_ANY_SPACE &&
1956 (rv = vm_map_alignspace(map, object, offset, addr, length,
1957 max_addr, alignment)) != KERN_SUCCESS) {
1958 if (find_space == VMFS_OPTIMAL_SPACE) {
1959 find_space = VMFS_ANY_SPACE;
1960 curr_min_addr = min_addr;
1961 cluster = update_anon;
1967 } else if ((cow & MAP_REMAP) != 0) {
1968 if (*addr < vm_map_min(map) ||
1969 *addr + length > vm_map_max(map) ||
1970 *addr + length <= length) {
1971 rv = KERN_INVALID_ADDRESS;
1974 vm_map_delete(map, *addr, *addr + length);
1976 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) != 0) {
1977 rv = vm_map_stack_locked(map, *addr, length, sgrowsiz, prot,
1980 rv = vm_map_insert(map, object, offset, *addr, *addr + length,
1983 if (rv == KERN_SUCCESS && update_anon)
1984 map->anon_loc = *addr + length;
1991 * vm_map_find_min() is a variant of vm_map_find() that takes an
1992 * additional parameter (min_addr) and treats the given address
1993 * (*addr) differently. Specifically, it treats *addr as a hint
1994 * and not as the minimum address where the mapping is created.
1996 * This function works in two phases. First, it tries to
1997 * allocate above the hint. If that fails and the hint is
1998 * greater than min_addr, it performs a second pass, replacing
1999 * the hint with min_addr as the minimum address for the
2003 vm_map_find_min(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
2004 vm_offset_t *addr, vm_size_t length, vm_offset_t min_addr,
2005 vm_offset_t max_addr, int find_space, vm_prot_t prot, vm_prot_t max,
2013 rv = vm_map_find(map, object, offset, addr, length, max_addr,
2014 find_space, prot, max, cow);
2015 if (rv == KERN_SUCCESS || min_addr >= hint)
2017 *addr = hint = min_addr;
2022 * A map entry with any of the following flags set must not be merged with
2025 #define MAP_ENTRY_NOMERGE_MASK (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP | \
2026 MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_IS_SUB_MAP | MAP_ENTRY_VN_EXEC)
2029 vm_map_mergeable_neighbors(vm_map_entry_t prev, vm_map_entry_t entry)
2032 KASSERT((prev->eflags & MAP_ENTRY_NOMERGE_MASK) == 0 ||
2033 (entry->eflags & MAP_ENTRY_NOMERGE_MASK) == 0,
2034 ("vm_map_mergeable_neighbors: neither %p nor %p are mergeable",
2036 return (prev->end == entry->start &&
2037 prev->object.vm_object == entry->object.vm_object &&
2038 (prev->object.vm_object == NULL ||
2039 prev->offset + (prev->end - prev->start) == entry->offset) &&
2040 prev->eflags == entry->eflags &&
2041 prev->protection == entry->protection &&
2042 prev->max_protection == entry->max_protection &&
2043 prev->inheritance == entry->inheritance &&
2044 prev->wired_count == entry->wired_count &&
2045 prev->cred == entry->cred);
2049 vm_map_merged_neighbor_dispose(vm_map_t map, vm_map_entry_t entry)
2053 * If the backing object is a vnode object, vm_object_deallocate()
2054 * calls vrele(). However, vrele() does not lock the vnode because
2055 * the vnode has additional references. Thus, the map lock can be
2056 * kept without causing a lock-order reversal with the vnode lock.
2058 * Since we count the number of virtual page mappings in
2059 * object->un_pager.vnp.writemappings, the writemappings value
2060 * should not be adjusted when the entry is disposed of.
2062 if (entry->object.vm_object != NULL)
2063 vm_object_deallocate(entry->object.vm_object);
2064 if (entry->cred != NULL)
2065 crfree(entry->cred);
2066 vm_map_entry_dispose(map, entry);
2070 * vm_map_try_merge_entries:
2072 * Compare the given map entry to its predecessor, and merge its precessor
2073 * into it if possible. The entry remains valid, and may be extended.
2074 * The predecessor may be deleted.
2076 * The map must be locked.
2079 vm_map_try_merge_entries(vm_map_t map, vm_map_entry_t prev, vm_map_entry_t entry)
2082 VM_MAP_ASSERT_LOCKED(map);
2083 if ((entry->eflags & MAP_ENTRY_NOMERGE_MASK) == 0 &&
2084 vm_map_mergeable_neighbors(prev, entry)) {
2085 vm_map_entry_unlink(map, prev, UNLINK_MERGE_NEXT);
2086 vm_map_merged_neighbor_dispose(map, prev);
2091 * vm_map_entry_back:
2093 * Allocate an object to back a map entry.
2096 vm_map_entry_back(vm_map_entry_t entry)
2100 KASSERT(entry->object.vm_object == NULL,
2101 ("map entry %p has backing object", entry));
2102 KASSERT((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0,
2103 ("map entry %p is a submap", entry));
2104 object = vm_object_allocate(OBJT_DEFAULT,
2105 atop(entry->end - entry->start));
2106 entry->object.vm_object = object;
2108 if (entry->cred != NULL) {
2109 object->cred = entry->cred;
2110 object->charge = entry->end - entry->start;
2116 * vm_map_entry_charge_object
2118 * If there is no object backing this entry, create one. Otherwise, if
2119 * the entry has cred, give it to the backing object.
2122 vm_map_entry_charge_object(vm_map_t map, vm_map_entry_t entry)
2125 VM_MAP_ASSERT_LOCKED(map);
2126 KASSERT((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0,
2127 ("map entry %p is a submap", entry));
2128 if (entry->object.vm_object == NULL && !map->system_map &&
2129 (entry->eflags & MAP_ENTRY_GUARD) == 0)
2130 vm_map_entry_back(entry);
2131 else if (entry->object.vm_object != NULL &&
2132 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) &&
2133 entry->cred != NULL) {
2134 VM_OBJECT_WLOCK(entry->object.vm_object);
2135 KASSERT(entry->object.vm_object->cred == NULL,
2136 ("OVERCOMMIT: %s: both cred e %p", __func__, entry));
2137 entry->object.vm_object->cred = entry->cred;
2138 entry->object.vm_object->charge = entry->end - entry->start;
2139 VM_OBJECT_WUNLOCK(entry->object.vm_object);
2145 * vm_map_clip_start: [ internal use only ]
2147 * Asserts that the given entry begins at or after
2148 * the specified address; if necessary,
2149 * it splits the entry into two.
2151 #define vm_map_clip_start(map, entry, startaddr) \
2153 if (startaddr > entry->start) \
2154 _vm_map_clip_start(map, entry, startaddr); \
2158 * This routine is called only when it is known that
2159 * the entry must be split.
2162 _vm_map_clip_start(vm_map_t map, vm_map_entry_t entry, vm_offset_t start)
2164 vm_map_entry_t new_entry;
2166 VM_MAP_ASSERT_LOCKED(map);
2167 KASSERT(entry->end > start && entry->start < start,
2168 ("_vm_map_clip_start: invalid clip of entry %p", entry));
2171 * Create a backing object now, if none exists, so that more individual
2172 * objects won't be created after the map entry is split.
2174 vm_map_entry_charge_object(map, entry);
2176 /* Clone the entry. */
2177 new_entry = vm_map_entry_create(map);
2178 *new_entry = *entry;
2181 * Split off the front portion. Insert the new entry BEFORE this one,
2182 * so that this entry has the specified starting address.
2184 new_entry->end = start;
2185 entry->offset += (start - entry->start);
2186 entry->start = start;
2187 if (new_entry->cred != NULL)
2188 crhold(entry->cred);
2190 vm_map_entry_link(map, new_entry);
2192 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
2193 vm_object_reference(new_entry->object.vm_object);
2194 vm_map_entry_set_vnode_text(new_entry, true);
2196 * The object->un_pager.vnp.writemappings for the
2197 * object of MAP_ENTRY_WRITECNT type entry shall be
2198 * kept as is here. The virtual pages are
2199 * re-distributed among the clipped entries, so the sum is
2206 * vm_map_clip_end: [ internal use only ]
2208 * Asserts that the given entry ends at or before
2209 * the specified address; if necessary,
2210 * it splits the entry into two.
2212 #define vm_map_clip_end(map, entry, endaddr) \
2214 if ((endaddr) < (entry->end)) \
2215 _vm_map_clip_end((map), (entry), (endaddr)); \
2219 * This routine is called only when it is known that
2220 * the entry must be split.
2223 _vm_map_clip_end(vm_map_t map, vm_map_entry_t entry, vm_offset_t end)
2225 vm_map_entry_t new_entry;
2227 VM_MAP_ASSERT_LOCKED(map);
2228 KASSERT(entry->start < end && entry->end > end,
2229 ("_vm_map_clip_end: invalid clip of entry %p", entry));
2232 * Create a backing object now, if none exists, so that more individual
2233 * objects won't be created after the map entry is split.
2235 vm_map_entry_charge_object(map, entry);
2237 /* Clone the entry. */
2238 new_entry = vm_map_entry_create(map);
2239 *new_entry = *entry;
2242 * Split off the back portion. Insert the new entry AFTER this one,
2243 * so that this entry has the specified ending address.
2245 new_entry->start = entry->end = end;
2246 new_entry->offset += (end - entry->start);
2247 if (new_entry->cred != NULL)
2248 crhold(entry->cred);
2250 vm_map_entry_link(map, new_entry);
2252 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
2253 vm_object_reference(new_entry->object.vm_object);
2254 vm_map_entry_set_vnode_text(new_entry, true);
2259 * vm_map_submap: [ kernel use only ]
2261 * Mark the given range as handled by a subordinate map.
2263 * This range must have been created with vm_map_find,
2264 * and no other operations may have been performed on this
2265 * range prior to calling vm_map_submap.
2267 * Only a limited number of operations can be performed
2268 * within this rage after calling vm_map_submap:
2270 * [Don't try vm_map_copy!]
2272 * To remove a submapping, one must first remove the
2273 * range from the superior map, and then destroy the
2274 * submap (if desired). [Better yet, don't try it.]
2283 vm_map_entry_t entry;
2286 result = KERN_INVALID_ARGUMENT;
2288 vm_map_lock(submap);
2289 submap->flags |= MAP_IS_SUB_MAP;
2290 vm_map_unlock(submap);
2294 VM_MAP_RANGE_CHECK(map, start, end);
2296 if (vm_map_lookup_entry(map, start, &entry)) {
2297 vm_map_clip_start(map, entry, start);
2299 entry = vm_map_entry_succ(entry);
2301 vm_map_clip_end(map, entry, end);
2303 if ((entry->start == start) && (entry->end == end) &&
2304 ((entry->eflags & MAP_ENTRY_COW) == 0) &&
2305 (entry->object.vm_object == NULL)) {
2306 entry->object.sub_map = submap;
2307 entry->eflags |= MAP_ENTRY_IS_SUB_MAP;
2308 result = KERN_SUCCESS;
2312 if (result != KERN_SUCCESS) {
2313 vm_map_lock(submap);
2314 submap->flags &= ~MAP_IS_SUB_MAP;
2315 vm_map_unlock(submap);
2321 * The maximum number of pages to map if MAP_PREFAULT_PARTIAL is specified
2323 #define MAX_INIT_PT 96
2326 * vm_map_pmap_enter:
2328 * Preload the specified map's pmap with mappings to the specified
2329 * object's memory-resident pages. No further physical pages are
2330 * allocated, and no further virtual pages are retrieved from secondary
2331 * storage. If the specified flags include MAP_PREFAULT_PARTIAL, then a
2332 * limited number of page mappings are created at the low-end of the
2333 * specified address range. (For this purpose, a superpage mapping
2334 * counts as one page mapping.) Otherwise, all resident pages within
2335 * the specified address range are mapped.
2338 vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot,
2339 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags)
2342 vm_page_t p, p_start;
2343 vm_pindex_t mask, psize, threshold, tmpidx;
2345 if ((prot & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0 || object == NULL)
2347 VM_OBJECT_RLOCK(object);
2348 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) {
2349 VM_OBJECT_RUNLOCK(object);
2350 VM_OBJECT_WLOCK(object);
2351 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) {
2352 pmap_object_init_pt(map->pmap, addr, object, pindex,
2354 VM_OBJECT_WUNLOCK(object);
2357 VM_OBJECT_LOCK_DOWNGRADE(object);
2361 if (psize + pindex > object->size) {
2362 if (object->size < pindex) {
2363 VM_OBJECT_RUNLOCK(object);
2366 psize = object->size - pindex;
2371 threshold = MAX_INIT_PT;
2373 p = vm_page_find_least(object, pindex);
2375 * Assert: the variable p is either (1) the page with the
2376 * least pindex greater than or equal to the parameter pindex
2380 p != NULL && (tmpidx = p->pindex - pindex) < psize;
2381 p = TAILQ_NEXT(p, listq)) {
2383 * don't allow an madvise to blow away our really
2384 * free pages allocating pv entries.
2386 if (((flags & MAP_PREFAULT_MADVISE) != 0 &&
2387 vm_page_count_severe()) ||
2388 ((flags & MAP_PREFAULT_PARTIAL) != 0 &&
2389 tmpidx >= threshold)) {
2393 if (vm_page_all_valid(p)) {
2394 if (p_start == NULL) {
2395 start = addr + ptoa(tmpidx);
2398 /* Jump ahead if a superpage mapping is possible. */
2399 if (p->psind > 0 && ((addr + ptoa(tmpidx)) &
2400 (pagesizes[p->psind] - 1)) == 0) {
2401 mask = atop(pagesizes[p->psind]) - 1;
2402 if (tmpidx + mask < psize &&
2403 vm_page_ps_test(p, PS_ALL_VALID, NULL)) {
2408 } else if (p_start != NULL) {
2409 pmap_enter_object(map->pmap, start, addr +
2410 ptoa(tmpidx), p_start, prot);
2414 if (p_start != NULL)
2415 pmap_enter_object(map->pmap, start, addr + ptoa(psize),
2417 VM_OBJECT_RUNLOCK(object);
2423 * Sets the protection of the specified address
2424 * region in the target map. If "set_max" is
2425 * specified, the maximum protection is to be set;
2426 * otherwise, only the current protection is affected.
2429 vm_map_protect(vm_map_t map, vm_offset_t start, vm_offset_t end,
2430 vm_prot_t new_prot, boolean_t set_max)
2432 vm_map_entry_t current, entry, in_tran;
2439 return (KERN_SUCCESS);
2446 * Ensure that we are not concurrently wiring pages. vm_map_wire() may
2447 * need to fault pages into the map and will drop the map lock while
2448 * doing so, and the VM object may end up in an inconsistent state if we
2449 * update the protection on the map entry in between faults.
2451 vm_map_wait_busy(map);
2453 VM_MAP_RANGE_CHECK(map, start, end);
2455 if (!vm_map_lookup_entry(map, start, &entry))
2456 entry = vm_map_entry_succ(entry);
2459 * Make a first pass to check for protection violations.
2461 for (current = entry; current->start < end;
2462 current = vm_map_entry_succ(current)) {
2463 if ((current->eflags & MAP_ENTRY_GUARD) != 0)
2465 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
2467 return (KERN_INVALID_ARGUMENT);
2469 if ((new_prot & current->max_protection) != new_prot) {
2471 return (KERN_PROTECTION_FAILURE);
2473 if ((current->eflags & MAP_ENTRY_IN_TRANSITION) != 0)
2478 * Postpone the operation until all in-transition map entries have
2479 * stabilized. An in-transition entry might already have its pages
2480 * wired and wired_count incremented, but not yet have its
2481 * MAP_ENTRY_USER_WIRED flag set. In which case, we would fail to call
2482 * vm_fault_copy_entry() in the final loop below.
2484 if (in_tran != NULL) {
2485 in_tran->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2486 vm_map_unlock_and_wait(map, 0);
2491 * Before changing the protections, try to reserve swap space for any
2492 * private (i.e., copy-on-write) mappings that are transitioning from
2493 * read-only to read/write access. If a reservation fails, break out
2494 * of this loop early and let the next loop simplify the entries, since
2495 * some may now be mergeable.
2498 vm_map_clip_start(map, entry, start);
2499 for (current = entry; current->start < end;
2500 current = vm_map_entry_succ(current)) {
2502 vm_map_clip_end(map, current, end);
2505 ((new_prot & ~(current->protection)) & VM_PROT_WRITE) == 0 ||
2506 ENTRY_CHARGED(current) ||
2507 (current->eflags & MAP_ENTRY_GUARD) != 0) {
2511 cred = curthread->td_ucred;
2512 obj = current->object.vm_object;
2514 if (obj == NULL || (current->eflags & MAP_ENTRY_NEEDS_COPY)) {
2515 if (!swap_reserve(current->end - current->start)) {
2516 rv = KERN_RESOURCE_SHORTAGE;
2521 current->cred = cred;
2525 VM_OBJECT_WLOCK(obj);
2526 if (obj->type != OBJT_DEFAULT && obj->type != OBJT_SWAP) {
2527 VM_OBJECT_WUNLOCK(obj);
2532 * Charge for the whole object allocation now, since
2533 * we cannot distinguish between non-charged and
2534 * charged clipped mapping of the same object later.
2536 KASSERT(obj->charge == 0,
2537 ("vm_map_protect: object %p overcharged (entry %p)",
2539 if (!swap_reserve(ptoa(obj->size))) {
2540 VM_OBJECT_WUNLOCK(obj);
2541 rv = KERN_RESOURCE_SHORTAGE;
2548 obj->charge = ptoa(obj->size);
2549 VM_OBJECT_WUNLOCK(obj);
2553 * If enough swap space was available, go back and fix up protections.
2554 * Otherwise, just simplify entries, since some may have been modified.
2555 * [Note that clipping is not necessary the second time.]
2557 for (current = entry; current->start < end;
2558 vm_map_try_merge_entries(map, vm_map_entry_pred(current), current),
2559 current = vm_map_entry_succ(current)) {
2560 if (rv != KERN_SUCCESS ||
2561 (current->eflags & MAP_ENTRY_GUARD) != 0)
2564 old_prot = current->protection;
2567 current->protection =
2568 (current->max_protection = new_prot) &
2571 current->protection = new_prot;
2574 * For user wired map entries, the normal lazy evaluation of
2575 * write access upgrades through soft page faults is
2576 * undesirable. Instead, immediately copy any pages that are
2577 * copy-on-write and enable write access in the physical map.
2579 if ((current->eflags & MAP_ENTRY_USER_WIRED) != 0 &&
2580 (current->protection & VM_PROT_WRITE) != 0 &&
2581 (old_prot & VM_PROT_WRITE) == 0)
2582 vm_fault_copy_entry(map, map, current, current, NULL);
2585 * When restricting access, update the physical map. Worry
2586 * about copy-on-write here.
2588 if ((old_prot & ~current->protection) != 0) {
2589 #define MASK(entry) (((entry)->eflags & MAP_ENTRY_COW) ? ~VM_PROT_WRITE : \
2591 pmap_protect(map->pmap, current->start,
2593 current->protection & MASK(current));
2597 vm_map_try_merge_entries(map, vm_map_entry_pred(current), current);
2605 * This routine traverses a processes map handling the madvise
2606 * system call. Advisories are classified as either those effecting
2607 * the vm_map_entry structure, or those effecting the underlying
2617 vm_map_entry_t current, entry;
2621 * Some madvise calls directly modify the vm_map_entry, in which case
2622 * we need to use an exclusive lock on the map and we need to perform
2623 * various clipping operations. Otherwise we only need a read-lock
2628 case MADV_SEQUENTIAL:
2645 vm_map_lock_read(map);
2652 * Locate starting entry and clip if necessary.
2654 VM_MAP_RANGE_CHECK(map, start, end);
2656 if (vm_map_lookup_entry(map, start, &entry)) {
2658 vm_map_clip_start(map, entry, start);
2660 entry = vm_map_entry_succ(entry);
2665 * madvise behaviors that are implemented in the vm_map_entry.
2667 * We clip the vm_map_entry so that behavioral changes are
2668 * limited to the specified address range.
2670 for (current = entry; current->start < end;
2671 current = vm_map_entry_succ(current)) {
2672 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
2675 vm_map_clip_end(map, current, end);
2679 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_NORMAL);
2681 case MADV_SEQUENTIAL:
2682 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_SEQUENTIAL);
2685 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_RANDOM);
2688 current->eflags |= MAP_ENTRY_NOSYNC;
2691 current->eflags &= ~MAP_ENTRY_NOSYNC;
2694 current->eflags |= MAP_ENTRY_NOCOREDUMP;
2697 current->eflags &= ~MAP_ENTRY_NOCOREDUMP;
2702 vm_map_try_merge_entries(map,
2703 vm_map_entry_pred(current), current);
2705 vm_map_try_merge_entries(map, vm_map_entry_pred(current),
2709 vm_pindex_t pstart, pend;
2712 * madvise behaviors that are implemented in the underlying
2715 * Since we don't clip the vm_map_entry, we have to clip
2716 * the vm_object pindex and count.
2718 for (current = entry; current->start < end;
2719 current = vm_map_entry_succ(current)) {
2720 vm_offset_t useEnd, useStart;
2722 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
2726 * MADV_FREE would otherwise rewind time to
2727 * the creation of the shadow object. Because
2728 * we hold the VM map read-locked, neither the
2729 * entry's object nor the presence of a
2730 * backing object can change.
2732 if (behav == MADV_FREE &&
2733 current->object.vm_object != NULL &&
2734 current->object.vm_object->backing_object != NULL)
2737 pstart = OFF_TO_IDX(current->offset);
2738 pend = pstart + atop(current->end - current->start);
2739 useStart = current->start;
2740 useEnd = current->end;
2742 if (current->start < start) {
2743 pstart += atop(start - current->start);
2746 if (current->end > end) {
2747 pend -= atop(current->end - end);
2755 * Perform the pmap_advise() before clearing
2756 * PGA_REFERENCED in vm_page_advise(). Otherwise, a
2757 * concurrent pmap operation, such as pmap_remove(),
2758 * could clear a reference in the pmap and set
2759 * PGA_REFERENCED on the page before the pmap_advise()
2760 * had completed. Consequently, the page would appear
2761 * referenced based upon an old reference that
2762 * occurred before this pmap_advise() ran.
2764 if (behav == MADV_DONTNEED || behav == MADV_FREE)
2765 pmap_advise(map->pmap, useStart, useEnd,
2768 vm_object_madvise(current->object.vm_object, pstart,
2772 * Pre-populate paging structures in the
2773 * WILLNEED case. For wired entries, the
2774 * paging structures are already populated.
2776 if (behav == MADV_WILLNEED &&
2777 current->wired_count == 0) {
2778 vm_map_pmap_enter(map,
2780 current->protection,
2781 current->object.vm_object,
2783 ptoa(pend - pstart),
2784 MAP_PREFAULT_MADVISE
2788 vm_map_unlock_read(map);
2797 * Sets the inheritance of the specified address
2798 * range in the target map. Inheritance
2799 * affects how the map will be shared with
2800 * child maps at the time of vmspace_fork.
2803 vm_map_inherit(vm_map_t map, vm_offset_t start, vm_offset_t end,
2804 vm_inherit_t new_inheritance)
2806 vm_map_entry_t entry;
2807 vm_map_entry_t temp_entry;
2809 switch (new_inheritance) {
2810 case VM_INHERIT_NONE:
2811 case VM_INHERIT_COPY:
2812 case VM_INHERIT_SHARE:
2813 case VM_INHERIT_ZERO:
2816 return (KERN_INVALID_ARGUMENT);
2819 return (KERN_SUCCESS);
2821 VM_MAP_RANGE_CHECK(map, start, end);
2822 if (vm_map_lookup_entry(map, start, &temp_entry)) {
2824 vm_map_clip_start(map, entry, start);
2826 entry = vm_map_entry_succ(temp_entry);
2827 while (entry->start < end) {
2828 vm_map_clip_end(map, entry, end);
2829 if ((entry->eflags & MAP_ENTRY_GUARD) == 0 ||
2830 new_inheritance != VM_INHERIT_ZERO)
2831 entry->inheritance = new_inheritance;
2832 vm_map_try_merge_entries(map, vm_map_entry_pred(entry), entry);
2833 entry = vm_map_entry_succ(entry);
2835 vm_map_try_merge_entries(map, vm_map_entry_pred(entry), entry);
2837 return (KERN_SUCCESS);
2841 * vm_map_entry_in_transition:
2843 * Release the map lock, and sleep until the entry is no longer in
2844 * transition. Awake and acquire the map lock. If the map changed while
2845 * another held the lock, lookup a possibly-changed entry at or after the
2846 * 'start' position of the old entry.
2848 static vm_map_entry_t
2849 vm_map_entry_in_transition(vm_map_t map, vm_offset_t in_start,
2850 vm_offset_t *io_end, bool holes_ok, vm_map_entry_t in_entry)
2852 vm_map_entry_t entry;
2854 u_int last_timestamp;
2856 VM_MAP_ASSERT_LOCKED(map);
2857 KASSERT((in_entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0,
2858 ("not in-tranition map entry %p", in_entry));
2860 * We have not yet clipped the entry.
2862 start = MAX(in_start, in_entry->start);
2863 in_entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2864 last_timestamp = map->timestamp;
2865 if (vm_map_unlock_and_wait(map, 0)) {
2867 * Allow interruption of user wiring/unwiring?
2871 if (last_timestamp + 1 == map->timestamp)
2875 * Look again for the entry because the map was modified while it was
2876 * unlocked. Specifically, the entry may have been clipped, merged, or
2879 if (!vm_map_lookup_entry(map, start, &entry)) {
2884 entry = vm_map_entry_succ(entry);
2892 * Implements both kernel and user unwiring.
2895 vm_map_unwire(vm_map_t map, vm_offset_t start, vm_offset_t end,
2898 vm_map_entry_t entry, first_entry;
2900 bool first_iteration, holes_ok, need_wakeup, user_unwire;
2903 return (KERN_SUCCESS);
2904 holes_ok = (flags & VM_MAP_WIRE_HOLESOK) != 0;
2905 user_unwire = (flags & VM_MAP_WIRE_USER) != 0;
2907 VM_MAP_RANGE_CHECK(map, start, end);
2908 if (!vm_map_lookup_entry(map, start, &first_entry)) {
2910 first_entry = vm_map_entry_succ(first_entry);
2913 return (KERN_INVALID_ADDRESS);
2916 first_iteration = true;
2917 entry = first_entry;
2919 while (entry->start < end) {
2920 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
2922 * We have not yet clipped the entry.
2924 entry = vm_map_entry_in_transition(map, start, &end,
2926 if (entry == NULL) {
2927 if (first_iteration) {
2929 return (KERN_INVALID_ADDRESS);
2931 rv = KERN_INVALID_ADDRESS;
2934 first_entry = first_iteration ? entry : NULL;
2937 first_iteration = false;
2938 vm_map_clip_start(map, entry, start);
2939 vm_map_clip_end(map, entry, end);
2941 * Mark the entry in case the map lock is released. (See
2944 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 &&
2945 entry->wiring_thread == NULL,
2946 ("owned map entry %p", entry));
2947 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
2948 entry->wiring_thread = curthread;
2950 * Check the map for holes in the specified region.
2951 * If holes_ok, skip this check.
2954 (entry->end < end &&
2955 vm_map_entry_succ(entry)->start > entry->end)) {
2957 rv = KERN_INVALID_ADDRESS;
2961 * If system unwiring, require that the entry is system wired.
2964 vm_map_entry_system_wired_count(entry) == 0) {
2966 rv = KERN_INVALID_ARGUMENT;
2969 entry = vm_map_entry_succ(entry);
2971 need_wakeup = false;
2972 if (first_entry == NULL &&
2973 !vm_map_lookup_entry(map, start, &first_entry)) {
2974 KASSERT(holes_ok, ("vm_map_unwire: lookup failed"));
2975 first_entry = vm_map_entry_succ(first_entry);
2977 for (entry = first_entry; entry->start < end;
2978 entry = vm_map_entry_succ(entry)) {
2980 * If holes_ok was specified, an empty
2981 * space in the unwired region could have been mapped
2982 * while the map lock was dropped for draining
2983 * MAP_ENTRY_IN_TRANSITION. Moreover, another thread
2984 * could be simultaneously wiring this new mapping
2985 * entry. Detect these cases and skip any entries
2986 * marked as in transition by us.
2988 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 ||
2989 entry->wiring_thread != curthread) {
2991 ("vm_map_unwire: !HOLESOK and new/changed entry"));
2995 if (rv == KERN_SUCCESS && (!user_unwire ||
2996 (entry->eflags & MAP_ENTRY_USER_WIRED))) {
2997 if (entry->wired_count == 1)
2998 vm_map_entry_unwire(map, entry);
3000 entry->wired_count--;
3002 entry->eflags &= ~MAP_ENTRY_USER_WIRED;
3004 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0,
3005 ("vm_map_unwire: in-transition flag missing %p", entry));
3006 KASSERT(entry->wiring_thread == curthread,
3007 ("vm_map_unwire: alien wire %p", entry));
3008 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION;
3009 entry->wiring_thread = NULL;
3010 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
3011 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
3014 vm_map_try_merge_entries(map, vm_map_entry_pred(entry), entry);
3016 vm_map_try_merge_entries(map, vm_map_entry_pred(entry), entry);
3024 vm_map_wire_user_count_sub(u_long npages)
3027 atomic_subtract_long(&vm_user_wire_count, npages);
3031 vm_map_wire_user_count_add(u_long npages)
3035 wired = vm_user_wire_count;
3037 if (npages + wired > vm_page_max_user_wired)
3039 } while (!atomic_fcmpset_long(&vm_user_wire_count, &wired,
3046 * vm_map_wire_entry_failure:
3048 * Handle a wiring failure on the given entry.
3050 * The map should be locked.
3053 vm_map_wire_entry_failure(vm_map_t map, vm_map_entry_t entry,
3054 vm_offset_t failed_addr)
3057 VM_MAP_ASSERT_LOCKED(map);
3058 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 &&
3059 entry->wired_count == 1,
3060 ("vm_map_wire_entry_failure: entry %p isn't being wired", entry));
3061 KASSERT(failed_addr < entry->end,
3062 ("vm_map_wire_entry_failure: entry %p was fully wired", entry));
3065 * If any pages at the start of this entry were successfully wired,
3068 if (failed_addr > entry->start) {
3069 pmap_unwire(map->pmap, entry->start, failed_addr);
3070 vm_object_unwire(entry->object.vm_object, entry->offset,
3071 failed_addr - entry->start, PQ_ACTIVE);
3075 * Assign an out-of-range value to represent the failure to wire this
3078 entry->wired_count = -1;
3082 vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t end, int flags)
3087 rv = vm_map_wire_locked(map, start, end, flags);
3094 * vm_map_wire_locked:
3096 * Implements both kernel and user wiring. Returns with the map locked,
3097 * the map lock may be dropped.
3100 vm_map_wire_locked(vm_map_t map, vm_offset_t start, vm_offset_t end, int flags)
3102 vm_map_entry_t entry, first_entry, tmp_entry;
3103 vm_offset_t faddr, saved_end, saved_start;
3105 u_int last_timestamp;
3107 bool first_iteration, holes_ok, need_wakeup, user_wire;
3110 VM_MAP_ASSERT_LOCKED(map);
3113 return (KERN_SUCCESS);
3115 if (flags & VM_MAP_WIRE_WRITE)
3116 prot |= VM_PROT_WRITE;
3117 holes_ok = (flags & VM_MAP_WIRE_HOLESOK) != 0;
3118 user_wire = (flags & VM_MAP_WIRE_USER) != 0;
3119 VM_MAP_RANGE_CHECK(map, start, end);
3120 if (!vm_map_lookup_entry(map, start, &first_entry)) {
3122 first_entry = vm_map_entry_succ(first_entry);
3124 return (KERN_INVALID_ADDRESS);
3126 first_iteration = true;
3127 entry = first_entry;
3128 while (entry->start < end) {
3129 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
3131 * We have not yet clipped the entry.
3133 entry = vm_map_entry_in_transition(map, start, &end,
3135 if (entry == NULL) {
3136 if (first_iteration)
3137 return (KERN_INVALID_ADDRESS);
3138 rv = KERN_INVALID_ADDRESS;
3141 first_entry = first_iteration ? entry : NULL;
3144 first_iteration = false;
3145 vm_map_clip_start(map, entry, start);
3146 vm_map_clip_end(map, entry, end);
3148 * Mark the entry in case the map lock is released. (See
3151 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 &&
3152 entry->wiring_thread == NULL,
3153 ("owned map entry %p", entry));
3154 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
3155 entry->wiring_thread = curthread;
3156 if ((entry->protection & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0
3157 || (entry->protection & prot) != prot) {
3158 entry->eflags |= MAP_ENTRY_WIRE_SKIPPED;
3161 rv = KERN_INVALID_ADDRESS;
3164 } else if (entry->wired_count == 0) {
3165 entry->wired_count++;
3167 npages = atop(entry->end - entry->start);
3168 if (user_wire && !vm_map_wire_user_count_add(npages)) {
3169 vm_map_wire_entry_failure(map, entry,
3172 rv = KERN_RESOURCE_SHORTAGE;
3177 * Release the map lock, relying on the in-transition
3178 * mark. Mark the map busy for fork.
3180 saved_start = entry->start;
3181 saved_end = entry->end;
3182 last_timestamp = map->timestamp;
3186 faddr = saved_start;
3189 * Simulate a fault to get the page and enter
3190 * it into the physical map.
3192 if ((rv = vm_fault(map, faddr,
3193 VM_PROT_NONE, VM_FAULT_WIRE, NULL)) !=
3196 } while ((faddr += PAGE_SIZE) < saved_end);
3199 if (last_timestamp + 1 != map->timestamp) {
3201 * Look again for the entry because the map was
3202 * modified while it was unlocked. The entry
3203 * may have been clipped, but NOT merged or
3206 if (!vm_map_lookup_entry(map, saved_start,
3209 ("vm_map_wire: lookup failed"));
3210 if (entry == first_entry)
3211 first_entry = tmp_entry;
3215 while (entry->end < saved_end) {
3217 * In case of failure, handle entries
3218 * that were not fully wired here;
3219 * fully wired entries are handled
3222 if (rv != KERN_SUCCESS &&
3224 vm_map_wire_entry_failure(map,
3226 entry = vm_map_entry_succ(entry);
3229 if (rv != KERN_SUCCESS) {
3230 vm_map_wire_entry_failure(map, entry, faddr);
3232 vm_map_wire_user_count_sub(npages);
3236 } else if (!user_wire ||
3237 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
3238 entry->wired_count++;
3241 * Check the map for holes in the specified region.
3242 * If holes_ok was specified, skip this check.
3246 vm_map_entry_succ(entry)->start > entry->end) {
3248 rv = KERN_INVALID_ADDRESS;
3251 entry = vm_map_entry_succ(entry);
3255 need_wakeup = false;
3256 if (first_entry == NULL &&
3257 !vm_map_lookup_entry(map, start, &first_entry)) {
3258 KASSERT(holes_ok, ("vm_map_wire: lookup failed"));
3259 first_entry = vm_map_entry_succ(first_entry);
3261 for (entry = first_entry; entry->start < end;
3262 entry = vm_map_entry_succ(entry)) {
3264 * If holes_ok was specified, an empty
3265 * space in the unwired region could have been mapped
3266 * while the map lock was dropped for faulting in the
3267 * pages or draining MAP_ENTRY_IN_TRANSITION.
3268 * Moreover, another thread could be simultaneously
3269 * wiring this new mapping entry. Detect these cases
3270 * and skip any entries marked as in transition not by us.
3272 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 ||
3273 entry->wiring_thread != curthread) {
3275 ("vm_map_wire: !HOLESOK and new/changed entry"));
3279 if ((entry->eflags & MAP_ENTRY_WIRE_SKIPPED) != 0) {
3281 } else if (rv == KERN_SUCCESS) {
3283 entry->eflags |= MAP_ENTRY_USER_WIRED;
3284 } else if (entry->wired_count == -1) {
3286 * Wiring failed on this entry. Thus, unwiring is
3289 entry->wired_count = 0;
3290 } else if (!user_wire ||
3291 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
3293 * Undo the wiring. Wiring succeeded on this entry
3294 * but failed on a later entry.
3296 if (entry->wired_count == 1) {
3297 vm_map_entry_unwire(map, entry);
3299 vm_map_wire_user_count_sub(
3300 atop(entry->end - entry->start));
3302 entry->wired_count--;
3304 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0,
3305 ("vm_map_wire: in-transition flag missing %p", entry));
3306 KASSERT(entry->wiring_thread == curthread,
3307 ("vm_map_wire: alien wire %p", entry));
3308 entry->eflags &= ~(MAP_ENTRY_IN_TRANSITION |
3309 MAP_ENTRY_WIRE_SKIPPED);
3310 entry->wiring_thread = NULL;
3311 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
3312 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
3315 vm_map_try_merge_entries(map, vm_map_entry_pred(entry), entry);
3317 vm_map_try_merge_entries(map, vm_map_entry_pred(entry), entry);
3326 * Push any dirty cached pages in the address range to their pager.
3327 * If syncio is TRUE, dirty pages are written synchronously.
3328 * If invalidate is TRUE, any cached pages are freed as well.
3330 * If the size of the region from start to end is zero, we are
3331 * supposed to flush all modified pages within the region containing
3332 * start. Unfortunately, a region can be split or coalesced with
3333 * neighboring regions, making it difficult to determine what the
3334 * original region was. Therefore, we approximate this requirement by
3335 * flushing the current region containing start.
3337 * Returns an error if any part of the specified range is not mapped.
3345 boolean_t invalidate)
3347 vm_map_entry_t current;
3348 vm_map_entry_t entry;
3351 vm_ooffset_t offset;
3352 unsigned int last_timestamp;
3355 vm_map_lock_read(map);
3356 VM_MAP_RANGE_CHECK(map, start, end);
3357 if (!vm_map_lookup_entry(map, start, &entry)) {
3358 vm_map_unlock_read(map);
3359 return (KERN_INVALID_ADDRESS);
3360 } else if (start == end) {
3361 start = entry->start;
3365 * Make a first pass to check for user-wired memory and holes.
3367 for (current = entry; current->start < end;
3368 current = vm_map_entry_succ(current)) {
3369 if (invalidate && (current->eflags & MAP_ENTRY_USER_WIRED)) {
3370 vm_map_unlock_read(map);
3371 return (KERN_INVALID_ARGUMENT);
3373 if (end > current->end &&
3374 current->end != vm_map_entry_succ(current)->start) {
3375 vm_map_unlock_read(map);
3376 return (KERN_INVALID_ADDRESS);
3381 pmap_remove(map->pmap, start, end);
3385 * Make a second pass, cleaning/uncaching pages from the indicated
3388 for (current = entry; current->start < end;) {
3389 offset = current->offset + (start - current->start);
3390 size = (end <= current->end ? end : current->end) - start;
3391 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
3393 vm_map_entry_t tentry;
3396 smap = current->object.sub_map;
3397 vm_map_lock_read(smap);
3398 (void) vm_map_lookup_entry(smap, offset, &tentry);
3399 tsize = tentry->end - offset;
3402 object = tentry->object.vm_object;
3403 offset = tentry->offset + (offset - tentry->start);
3404 vm_map_unlock_read(smap);
3406 object = current->object.vm_object;
3408 vm_object_reference(object);
3409 last_timestamp = map->timestamp;
3410 vm_map_unlock_read(map);
3411 if (!vm_object_sync(object, offset, size, syncio, invalidate))
3414 vm_object_deallocate(object);
3415 vm_map_lock_read(map);
3416 if (last_timestamp == map->timestamp ||
3417 !vm_map_lookup_entry(map, start, ¤t))
3418 current = vm_map_entry_succ(current);
3421 vm_map_unlock_read(map);
3422 return (failed ? KERN_FAILURE : KERN_SUCCESS);
3426 * vm_map_entry_unwire: [ internal use only ]
3428 * Make the region specified by this entry pageable.
3430 * The map in question should be locked.
3431 * [This is the reason for this routine's existence.]
3434 vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry)
3438 VM_MAP_ASSERT_LOCKED(map);
3439 KASSERT(entry->wired_count > 0,
3440 ("vm_map_entry_unwire: entry %p isn't wired", entry));
3442 size = entry->end - entry->start;
3443 if ((entry->eflags & MAP_ENTRY_USER_WIRED) != 0)
3444 vm_map_wire_user_count_sub(atop(size));
3445 pmap_unwire(map->pmap, entry->start, entry->end);
3446 vm_object_unwire(entry->object.vm_object, entry->offset, size,
3448 entry->wired_count = 0;
3452 vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map)
3455 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0)
3456 vm_object_deallocate(entry->object.vm_object);
3457 uma_zfree(system_map ? kmapentzone : mapentzone, entry);
3461 * vm_map_entry_delete: [ internal use only ]
3463 * Deallocate the given entry from the target map.
3466 vm_map_entry_delete(vm_map_t map, vm_map_entry_t entry)
3469 vm_pindex_t offidxstart, offidxend, count, size1;
3472 vm_map_entry_unlink(map, entry, UNLINK_MERGE_NONE);
3473 object = entry->object.vm_object;
3475 if ((entry->eflags & MAP_ENTRY_GUARD) != 0) {
3476 MPASS(entry->cred == NULL);
3477 MPASS((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0);
3478 MPASS(object == NULL);
3479 vm_map_entry_deallocate(entry, map->system_map);
3483 size = entry->end - entry->start;
3486 if (entry->cred != NULL) {
3487 swap_release_by_cred(size, entry->cred);
3488 crfree(entry->cred);
3491 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0 &&
3493 KASSERT(entry->cred == NULL || object->cred == NULL ||
3494 (entry->eflags & MAP_ENTRY_NEEDS_COPY),
3495 ("OVERCOMMIT vm_map_entry_delete: both cred %p", entry));
3497 offidxstart = OFF_TO_IDX(entry->offset);
3498 offidxend = offidxstart + count;
3499 VM_OBJECT_WLOCK(object);
3500 if (object->ref_count != 1 && ((object->flags & (OBJ_NOSPLIT |
3501 OBJ_ONEMAPPING)) == OBJ_ONEMAPPING ||
3502 object == kernel_object)) {
3503 vm_object_collapse(object);
3506 * The option OBJPR_NOTMAPPED can be passed here
3507 * because vm_map_delete() already performed
3508 * pmap_remove() on the only mapping to this range
3511 vm_object_page_remove(object, offidxstart, offidxend,
3513 if (object->type == OBJT_SWAP)
3514 swap_pager_freespace(object, offidxstart,
3516 if (offidxend >= object->size &&
3517 offidxstart < object->size) {
3518 size1 = object->size;
3519 object->size = offidxstart;
3520 if (object->cred != NULL) {
3521 size1 -= object->size;
3522 KASSERT(object->charge >= ptoa(size1),
3523 ("object %p charge < 0", object));
3524 swap_release_by_cred(ptoa(size1),
3526 object->charge -= ptoa(size1);
3530 VM_OBJECT_WUNLOCK(object);
3532 entry->object.vm_object = NULL;
3533 if (map->system_map)
3534 vm_map_entry_deallocate(entry, TRUE);
3536 entry->defer_next = curthread->td_map_def_user;
3537 curthread->td_map_def_user = entry;
3542 * vm_map_delete: [ internal use only ]
3544 * Deallocates the given address range from the target
3548 vm_map_delete(vm_map_t map, vm_offset_t start, vm_offset_t end)
3550 vm_map_entry_t entry;
3551 vm_map_entry_t first_entry;
3553 VM_MAP_ASSERT_LOCKED(map);
3555 return (KERN_SUCCESS);
3558 * Find the start of the region, and clip it
3560 if (!vm_map_lookup_entry(map, start, &first_entry))
3561 entry = vm_map_entry_succ(first_entry);
3563 entry = first_entry;
3564 vm_map_clip_start(map, entry, start);
3568 * Step through all entries in this region
3570 while (entry->start < end) {
3571 vm_map_entry_t next;
3574 * Wait for wiring or unwiring of an entry to complete.
3575 * Also wait for any system wirings to disappear on
3578 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 ||
3579 (vm_map_pmap(map) != kernel_pmap &&
3580 vm_map_entry_system_wired_count(entry) != 0)) {
3581 unsigned int last_timestamp;
3582 vm_offset_t saved_start;
3583 vm_map_entry_t tmp_entry;
3585 saved_start = entry->start;
3586 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
3587 last_timestamp = map->timestamp;
3588 (void) vm_map_unlock_and_wait(map, 0);
3590 if (last_timestamp + 1 != map->timestamp) {
3592 * Look again for the entry because the map was
3593 * modified while it was unlocked.
3594 * Specifically, the entry may have been
3595 * clipped, merged, or deleted.
3597 if (!vm_map_lookup_entry(map, saved_start,
3599 entry = vm_map_entry_succ(tmp_entry);
3602 vm_map_clip_start(map, entry,
3608 vm_map_clip_end(map, entry, end);
3610 next = vm_map_entry_succ(entry);
3613 * Unwire before removing addresses from the pmap; otherwise,
3614 * unwiring will put the entries back in the pmap.
3616 if (entry->wired_count != 0)
3617 vm_map_entry_unwire(map, entry);
3620 * Remove mappings for the pages, but only if the
3621 * mappings could exist. For instance, it does not
3622 * make sense to call pmap_remove() for guard entries.
3624 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0 ||
3625 entry->object.vm_object != NULL)
3626 pmap_remove(map->pmap, entry->start, entry->end);
3628 if (entry->end == map->anon_loc)
3629 map->anon_loc = entry->start;
3632 * Delete the entry only after removing all pmap
3633 * entries pointing to its pages. (Otherwise, its
3634 * page frames may be reallocated, and any modify bits
3635 * will be set in the wrong object!)
3637 vm_map_entry_delete(map, entry);
3640 return (KERN_SUCCESS);
3646 * Remove the given address range from the target map.
3647 * This is the exported form of vm_map_delete.
3650 vm_map_remove(vm_map_t map, vm_offset_t start, vm_offset_t end)
3655 VM_MAP_RANGE_CHECK(map, start, end);
3656 result = vm_map_delete(map, start, end);
3662 * vm_map_check_protection:
3664 * Assert that the target map allows the specified privilege on the
3665 * entire address region given. The entire region must be allocated.
3667 * WARNING! This code does not and should not check whether the
3668 * contents of the region is accessible. For example a smaller file
3669 * might be mapped into a larger address space.
3671 * NOTE! This code is also called by munmap().
3673 * The map must be locked. A read lock is sufficient.
3676 vm_map_check_protection(vm_map_t map, vm_offset_t start, vm_offset_t end,
3677 vm_prot_t protection)
3679 vm_map_entry_t entry;
3680 vm_map_entry_t tmp_entry;
3682 if (!vm_map_lookup_entry(map, start, &tmp_entry))
3686 while (start < end) {
3690 if (start < entry->start)
3693 * Check protection associated with entry.
3695 if ((entry->protection & protection) != protection)
3697 /* go to next entry */
3699 entry = vm_map_entry_succ(entry);
3705 * vm_map_copy_entry:
3707 * Copies the contents of the source entry to the destination
3708 * entry. The entries *must* be aligned properly.
3714 vm_map_entry_t src_entry,
3715 vm_map_entry_t dst_entry,
3716 vm_ooffset_t *fork_charge)
3718 vm_object_t src_object;
3719 vm_map_entry_t fake_entry;
3724 VM_MAP_ASSERT_LOCKED(dst_map);
3726 if ((dst_entry->eflags|src_entry->eflags) & MAP_ENTRY_IS_SUB_MAP)
3729 if (src_entry->wired_count == 0 ||
3730 (src_entry->protection & VM_PROT_WRITE) == 0) {
3732 * If the source entry is marked needs_copy, it is already
3735 if ((src_entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0 &&
3736 (src_entry->protection & VM_PROT_WRITE) != 0) {
3737 pmap_protect(src_map->pmap,
3740 src_entry->protection & ~VM_PROT_WRITE);
3744 * Make a copy of the object.
3746 size = src_entry->end - src_entry->start;
3747 if ((src_object = src_entry->object.vm_object) != NULL) {
3748 VM_OBJECT_WLOCK(src_object);
3749 charged = ENTRY_CHARGED(src_entry);
3750 if (src_object->handle == NULL &&
3751 (src_object->type == OBJT_DEFAULT ||
3752 src_object->type == OBJT_SWAP)) {
3753 vm_object_collapse(src_object);
3754 if ((src_object->flags & (OBJ_NOSPLIT |
3755 OBJ_ONEMAPPING)) == OBJ_ONEMAPPING) {
3756 vm_object_split(src_entry);
3758 src_entry->object.vm_object;
3761 vm_object_reference_locked(src_object);
3762 vm_object_clear_flag(src_object, OBJ_ONEMAPPING);
3763 if (src_entry->cred != NULL &&
3764 !(src_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
3765 KASSERT(src_object->cred == NULL,
3766 ("OVERCOMMIT: vm_map_copy_entry: cred %p",
3768 src_object->cred = src_entry->cred;
3769 src_object->charge = size;
3771 VM_OBJECT_WUNLOCK(src_object);
3772 dst_entry->object.vm_object = src_object;
3774 cred = curthread->td_ucred;
3776 dst_entry->cred = cred;
3777 *fork_charge += size;
3778 if (!(src_entry->eflags &
3779 MAP_ENTRY_NEEDS_COPY)) {
3781 src_entry->cred = cred;
3782 *fork_charge += size;
3785 src_entry->eflags |= MAP_ENTRY_COW |
3786 MAP_ENTRY_NEEDS_COPY;
3787 dst_entry->eflags |= MAP_ENTRY_COW |
3788 MAP_ENTRY_NEEDS_COPY;
3789 dst_entry->offset = src_entry->offset;
3790 if (src_entry->eflags & MAP_ENTRY_WRITECNT) {
3792 * MAP_ENTRY_WRITECNT cannot
3793 * indicate write reference from
3794 * src_entry, since the entry is
3795 * marked as needs copy. Allocate a
3796 * fake entry that is used to
3797 * decrement object->un_pager writecount
3798 * at the appropriate time. Attach
3799 * fake_entry to the deferred list.
3801 fake_entry = vm_map_entry_create(dst_map);
3802 fake_entry->eflags = MAP_ENTRY_WRITECNT;
3803 src_entry->eflags &= ~MAP_ENTRY_WRITECNT;
3804 vm_object_reference(src_object);
3805 fake_entry->object.vm_object = src_object;
3806 fake_entry->start = src_entry->start;
3807 fake_entry->end = src_entry->end;
3808 fake_entry->defer_next =
3809 curthread->td_map_def_user;
3810 curthread->td_map_def_user = fake_entry;
3813 pmap_copy(dst_map->pmap, src_map->pmap,
3814 dst_entry->start, dst_entry->end - dst_entry->start,
3817 dst_entry->object.vm_object = NULL;
3818 dst_entry->offset = 0;
3819 if (src_entry->cred != NULL) {
3820 dst_entry->cred = curthread->td_ucred;
3821 crhold(dst_entry->cred);
3822 *fork_charge += size;
3827 * We don't want to make writeable wired pages copy-on-write.
3828 * Immediately copy these pages into the new map by simulating
3829 * page faults. The new pages are pageable.
3831 vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry,
3837 * vmspace_map_entry_forked:
3838 * Update the newly-forked vmspace each time a map entry is inherited
3839 * or copied. The values for vm_dsize and vm_tsize are approximate
3840 * (and mostly-obsolete ideas in the face of mmap(2) et al.)
3843 vmspace_map_entry_forked(const struct vmspace *vm1, struct vmspace *vm2,
3844 vm_map_entry_t entry)
3846 vm_size_t entrysize;
3849 if ((entry->eflags & MAP_ENTRY_GUARD) != 0)
3851 entrysize = entry->end - entry->start;
3852 vm2->vm_map.size += entrysize;
3853 if (entry->eflags & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP)) {
3854 vm2->vm_ssize += btoc(entrysize);
3855 } else if (entry->start >= (vm_offset_t)vm1->vm_daddr &&
3856 entry->start < (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)) {
3857 newend = MIN(entry->end,
3858 (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize));
3859 vm2->vm_dsize += btoc(newend - entry->start);
3860 } else if (entry->start >= (vm_offset_t)vm1->vm_taddr &&
3861 entry->start < (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)) {
3862 newend = MIN(entry->end,
3863 (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize));
3864 vm2->vm_tsize += btoc(newend - entry->start);
3870 * Create a new process vmspace structure and vm_map
3871 * based on those of an existing process. The new map
3872 * is based on the old map, according to the inheritance
3873 * values on the regions in that map.
3875 * XXX It might be worth coalescing the entries added to the new vmspace.
3877 * The source map must not be locked.
3880 vmspace_fork(struct vmspace *vm1, vm_ooffset_t *fork_charge)
3882 struct vmspace *vm2;
3883 vm_map_t new_map, old_map;
3884 vm_map_entry_t new_entry, old_entry;
3889 old_map = &vm1->vm_map;
3890 /* Copy immutable fields of vm1 to vm2. */
3891 vm2 = vmspace_alloc(vm_map_min(old_map), vm_map_max(old_map),
3896 vm2->vm_taddr = vm1->vm_taddr;
3897 vm2->vm_daddr = vm1->vm_daddr;
3898 vm2->vm_maxsaddr = vm1->vm_maxsaddr;
3899 vm_map_lock(old_map);
3901 vm_map_wait_busy(old_map);
3902 new_map = &vm2->vm_map;
3903 locked = vm_map_trylock(new_map); /* trylock to silence WITNESS */
3904 KASSERT(locked, ("vmspace_fork: lock failed"));
3906 error = pmap_vmspace_copy(new_map->pmap, old_map->pmap);
3908 sx_xunlock(&old_map->lock);
3909 sx_xunlock(&new_map->lock);
3910 vm_map_process_deferred();
3915 new_map->anon_loc = old_map->anon_loc;
3917 old_entry = old_map->header.next;
3919 while (old_entry != &old_map->header) {
3920 if (old_entry->eflags & MAP_ENTRY_IS_SUB_MAP)
3921 panic("vm_map_fork: encountered a submap");
3923 inh = old_entry->inheritance;
3924 if ((old_entry->eflags & MAP_ENTRY_GUARD) != 0 &&
3925 inh != VM_INHERIT_NONE)
3926 inh = VM_INHERIT_COPY;
3929 case VM_INHERIT_NONE:
3932 case VM_INHERIT_SHARE:
3934 * Clone the entry, creating the shared object if necessary.
3936 object = old_entry->object.vm_object;
3937 if (object == NULL) {
3938 vm_map_entry_back(old_entry);
3939 object = old_entry->object.vm_object;
3943 * Add the reference before calling vm_object_shadow
3944 * to insure that a shadow object is created.
3946 vm_object_reference(object);
3947 if (old_entry->eflags & MAP_ENTRY_NEEDS_COPY) {
3948 vm_object_shadow(&old_entry->object.vm_object,
3950 old_entry->end - old_entry->start);
3951 old_entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
3952 /* Transfer the second reference too. */
3953 vm_object_reference(
3954 old_entry->object.vm_object);
3957 * As in vm_map_merged_neighbor_dispose(),
3958 * the vnode lock will not be acquired in
3959 * this call to vm_object_deallocate().
3961 vm_object_deallocate(object);
3962 object = old_entry->object.vm_object;
3964 VM_OBJECT_WLOCK(object);
3965 vm_object_clear_flag(object, OBJ_ONEMAPPING);
3966 if (old_entry->cred != NULL) {
3967 KASSERT(object->cred == NULL, ("vmspace_fork both cred"));
3968 object->cred = old_entry->cred;
3969 object->charge = old_entry->end - old_entry->start;
3970 old_entry->cred = NULL;
3974 * Assert the correct state of the vnode
3975 * v_writecount while the object is locked, to
3976 * not relock it later for the assertion
3979 if (old_entry->eflags & MAP_ENTRY_WRITECNT &&
3980 object->type == OBJT_VNODE) {
3981 KASSERT(((struct vnode *)object->handle)->
3983 ("vmspace_fork: v_writecount %p", object));
3984 KASSERT(object->un_pager.vnp.writemappings > 0,
3985 ("vmspace_fork: vnp.writecount %p",
3988 VM_OBJECT_WUNLOCK(object);
3991 * Clone the entry, referencing the shared object.
3993 new_entry = vm_map_entry_create(new_map);
3994 *new_entry = *old_entry;
3995 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED |
3996 MAP_ENTRY_IN_TRANSITION);
3997 new_entry->wiring_thread = NULL;
3998 new_entry->wired_count = 0;
3999 if (new_entry->eflags & MAP_ENTRY_WRITECNT) {
4000 vm_pager_update_writecount(object,
4001 new_entry->start, new_entry->end);
4003 vm_map_entry_set_vnode_text(new_entry, true);
4006 * Insert the entry into the new map -- we know we're
4007 * inserting at the end of the new map.
4009 vm_map_entry_link(new_map, new_entry);
4010 vmspace_map_entry_forked(vm1, vm2, new_entry);
4013 * Update the physical map
4015 pmap_copy(new_map->pmap, old_map->pmap,
4017 (old_entry->end - old_entry->start),
4021 case VM_INHERIT_COPY:
4023 * Clone the entry and link into the map.
4025 new_entry = vm_map_entry_create(new_map);
4026 *new_entry = *old_entry;
4028 * Copied entry is COW over the old object.
4030 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED |
4031 MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_WRITECNT);
4032 new_entry->wiring_thread = NULL;
4033 new_entry->wired_count = 0;
4034 new_entry->object.vm_object = NULL;
4035 new_entry->cred = NULL;
4036 vm_map_entry_link(new_map, new_entry);
4037 vmspace_map_entry_forked(vm1, vm2, new_entry);
4038 vm_map_copy_entry(old_map, new_map, old_entry,
4039 new_entry, fork_charge);
4040 vm_map_entry_set_vnode_text(new_entry, true);
4043 case VM_INHERIT_ZERO:
4045 * Create a new anonymous mapping entry modelled from
4048 new_entry = vm_map_entry_create(new_map);
4049 memset(new_entry, 0, sizeof(*new_entry));
4051 new_entry->start = old_entry->start;
4052 new_entry->end = old_entry->end;
4053 new_entry->eflags = old_entry->eflags &
4054 ~(MAP_ENTRY_USER_WIRED | MAP_ENTRY_IN_TRANSITION |
4055 MAP_ENTRY_WRITECNT | MAP_ENTRY_VN_EXEC);
4056 new_entry->protection = old_entry->protection;
4057 new_entry->max_protection = old_entry->max_protection;
4058 new_entry->inheritance = VM_INHERIT_ZERO;
4060 vm_map_entry_link(new_map, new_entry);
4061 vmspace_map_entry_forked(vm1, vm2, new_entry);
4063 new_entry->cred = curthread->td_ucred;
4064 crhold(new_entry->cred);
4065 *fork_charge += (new_entry->end - new_entry->start);
4069 old_entry = vm_map_entry_succ(old_entry);
4072 * Use inlined vm_map_unlock() to postpone handling the deferred
4073 * map entries, which cannot be done until both old_map and
4074 * new_map locks are released.
4076 sx_xunlock(&old_map->lock);
4077 sx_xunlock(&new_map->lock);
4078 vm_map_process_deferred();
4084 * Create a process's stack for exec_new_vmspace(). This function is never
4085 * asked to wire the newly created stack.
4088 vm_map_stack(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
4089 vm_prot_t prot, vm_prot_t max, int cow)
4091 vm_size_t growsize, init_ssize;
4095 MPASS((map->flags & MAP_WIREFUTURE) == 0);
4096 growsize = sgrowsiz;
4097 init_ssize = (max_ssize < growsize) ? max_ssize : growsize;
4099 vmemlim = lim_cur(curthread, RLIMIT_VMEM);
4100 /* If we would blow our VMEM resource limit, no go */
4101 if (map->size + init_ssize > vmemlim) {
4105 rv = vm_map_stack_locked(map, addrbos, max_ssize, growsize, prot,
4112 static int stack_guard_page = 1;
4113 SYSCTL_INT(_security_bsd, OID_AUTO, stack_guard_page, CTLFLAG_RWTUN,
4114 &stack_guard_page, 0,
4115 "Specifies the number of guard pages for a stack that grows");
4118 vm_map_stack_locked(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
4119 vm_size_t growsize, vm_prot_t prot, vm_prot_t max, int cow)
4121 vm_map_entry_t new_entry, prev_entry;
4122 vm_offset_t bot, gap_bot, gap_top, top;
4123 vm_size_t init_ssize, sgp;
4127 * The stack orientation is piggybacked with the cow argument.
4128 * Extract it into orient and mask the cow argument so that we
4129 * don't pass it around further.
4131 orient = cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP);
4132 KASSERT(orient != 0, ("No stack grow direction"));
4133 KASSERT(orient != (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP),
4136 if (addrbos < vm_map_min(map) ||
4137 addrbos + max_ssize > vm_map_max(map) ||
4138 addrbos + max_ssize <= addrbos)
4139 return (KERN_INVALID_ADDRESS);
4140 sgp = ((curproc->p_flag2 & P2_STKGAP_DISABLE) != 0 ||
4141 (curproc->p_fctl0 & NT_FREEBSD_FCTL_STKGAP_DISABLE) != 0) ? 0 :
4142 (vm_size_t)stack_guard_page * PAGE_SIZE;
4143 if (sgp >= max_ssize)
4144 return (KERN_INVALID_ARGUMENT);
4146 init_ssize = growsize;
4147 if (max_ssize < init_ssize + sgp)
4148 init_ssize = max_ssize - sgp;
4150 /* If addr is already mapped, no go */
4151 if (vm_map_lookup_entry(map, addrbos, &prev_entry))
4152 return (KERN_NO_SPACE);
4155 * If we can't accommodate max_ssize in the current mapping, no go.
4157 if (vm_map_entry_succ(prev_entry)->start < addrbos + max_ssize)
4158 return (KERN_NO_SPACE);
4161 * We initially map a stack of only init_ssize. We will grow as
4162 * needed later. Depending on the orientation of the stack (i.e.
4163 * the grow direction) we either map at the top of the range, the
4164 * bottom of the range or in the middle.
4166 * Note: we would normally expect prot and max to be VM_PROT_ALL,
4167 * and cow to be 0. Possibly we should eliminate these as input
4168 * parameters, and just pass these values here in the insert call.
4170 if (orient == MAP_STACK_GROWS_DOWN) {
4171 bot = addrbos + max_ssize - init_ssize;
4172 top = bot + init_ssize;
4175 } else /* if (orient == MAP_STACK_GROWS_UP) */ {
4177 top = bot + init_ssize;
4179 gap_top = addrbos + max_ssize;
4181 rv = vm_map_insert(map, NULL, 0, bot, top, prot, max, cow);
4182 if (rv != KERN_SUCCESS)
4184 new_entry = vm_map_entry_succ(prev_entry);
4185 KASSERT(new_entry->end == top || new_entry->start == bot,
4186 ("Bad entry start/end for new stack entry"));
4187 KASSERT((orient & MAP_STACK_GROWS_DOWN) == 0 ||
4188 (new_entry->eflags & MAP_ENTRY_GROWS_DOWN) != 0,
4189 ("new entry lacks MAP_ENTRY_GROWS_DOWN"));
4190 KASSERT((orient & MAP_STACK_GROWS_UP) == 0 ||
4191 (new_entry->eflags & MAP_ENTRY_GROWS_UP) != 0,
4192 ("new entry lacks MAP_ENTRY_GROWS_UP"));
4193 if (gap_bot == gap_top)
4194 return (KERN_SUCCESS);
4195 rv = vm_map_insert(map, NULL, 0, gap_bot, gap_top, VM_PROT_NONE,
4196 VM_PROT_NONE, MAP_CREATE_GUARD | (orient == MAP_STACK_GROWS_DOWN ?
4197 MAP_CREATE_STACK_GAP_DN : MAP_CREATE_STACK_GAP_UP));
4198 if (rv == KERN_SUCCESS) {
4200 * Gap can never successfully handle a fault, so
4201 * read-ahead logic is never used for it. Re-use
4202 * next_read of the gap entry to store
4203 * stack_guard_page for vm_map_growstack().
4205 if (orient == MAP_STACK_GROWS_DOWN)
4206 vm_map_entry_pred(new_entry)->next_read = sgp;
4208 vm_map_entry_succ(new_entry)->next_read = sgp;
4210 (void)vm_map_delete(map, bot, top);
4216 * Attempts to grow a vm stack entry. Returns KERN_SUCCESS if we
4217 * successfully grow the stack.
4220 vm_map_growstack(vm_map_t map, vm_offset_t addr, vm_map_entry_t gap_entry)
4222 vm_map_entry_t stack_entry;
4226 vm_offset_t gap_end, gap_start, grow_start;
4227 vm_size_t grow_amount, guard, max_grow;
4228 rlim_t lmemlim, stacklim, vmemlim;
4230 bool gap_deleted, grow_down, is_procstack;
4242 * Disallow stack growth when the access is performed by a
4243 * debugger or AIO daemon. The reason is that the wrong
4244 * resource limits are applied.
4246 if (p != initproc && (map != &p->p_vmspace->vm_map ||
4247 p->p_textvp == NULL))
4248 return (KERN_FAILURE);
4250 MPASS(!map->system_map);
4252 lmemlim = lim_cur(curthread, RLIMIT_MEMLOCK);
4253 stacklim = lim_cur(curthread, RLIMIT_STACK);
4254 vmemlim = lim_cur(curthread, RLIMIT_VMEM);
4256 /* If addr is not in a hole for a stack grow area, no need to grow. */
4257 if (gap_entry == NULL && !vm_map_lookup_entry(map, addr, &gap_entry))
4258 return (KERN_FAILURE);
4259 if ((gap_entry->eflags & MAP_ENTRY_GUARD) == 0)
4260 return (KERN_SUCCESS);
4261 if ((gap_entry->eflags & MAP_ENTRY_STACK_GAP_DN) != 0) {
4262 stack_entry = vm_map_entry_succ(gap_entry);
4263 if ((stack_entry->eflags & MAP_ENTRY_GROWS_DOWN) == 0 ||
4264 stack_entry->start != gap_entry->end)
4265 return (KERN_FAILURE);
4266 grow_amount = round_page(stack_entry->start - addr);
4268 } else if ((gap_entry->eflags & MAP_ENTRY_STACK_GAP_UP) != 0) {
4269 stack_entry = vm_map_entry_pred(gap_entry);
4270 if ((stack_entry->eflags & MAP_ENTRY_GROWS_UP) == 0 ||
4271 stack_entry->end != gap_entry->start)
4272 return (KERN_FAILURE);
4273 grow_amount = round_page(addr + 1 - stack_entry->end);
4276 return (KERN_FAILURE);
4278 guard = ((curproc->p_flag2 & P2_STKGAP_DISABLE) != 0 ||
4279 (curproc->p_fctl0 & NT_FREEBSD_FCTL_STKGAP_DISABLE) != 0) ? 0 :
4280 gap_entry->next_read;
4281 max_grow = gap_entry->end - gap_entry->start;
4282 if (guard > max_grow)
4283 return (KERN_NO_SPACE);
4285 if (grow_amount > max_grow)
4286 return (KERN_NO_SPACE);
4289 * If this is the main process stack, see if we're over the stack
4292 is_procstack = addr >= (vm_offset_t)vm->vm_maxsaddr &&
4293 addr < (vm_offset_t)p->p_sysent->sv_usrstack;
4294 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim))
4295 return (KERN_NO_SPACE);
4300 if (is_procstack && racct_set(p, RACCT_STACK,
4301 ctob(vm->vm_ssize) + grow_amount)) {
4303 return (KERN_NO_SPACE);
4309 grow_amount = roundup(grow_amount, sgrowsiz);
4310 if (grow_amount > max_grow)
4311 grow_amount = max_grow;
4312 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) {
4313 grow_amount = trunc_page((vm_size_t)stacklim) -
4319 limit = racct_get_available(p, RACCT_STACK);
4321 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > limit))
4322 grow_amount = limit - ctob(vm->vm_ssize);
4325 if (!old_mlock && (map->flags & MAP_WIREFUTURE) != 0) {
4326 if (ptoa(pmap_wired_count(map->pmap)) + grow_amount > lmemlim) {
4333 if (racct_set(p, RACCT_MEMLOCK,
4334 ptoa(pmap_wired_count(map->pmap)) + grow_amount)) {
4344 /* If we would blow our VMEM resource limit, no go */
4345 if (map->size + grow_amount > vmemlim) {
4352 if (racct_set(p, RACCT_VMEM, map->size + grow_amount)) {
4361 if (vm_map_lock_upgrade(map)) {
4363 vm_map_lock_read(map);
4368 grow_start = gap_entry->end - grow_amount;
4369 if (gap_entry->start + grow_amount == gap_entry->end) {
4370 gap_start = gap_entry->start;
4371 gap_end = gap_entry->end;
4372 vm_map_entry_delete(map, gap_entry);
4375 MPASS(gap_entry->start < gap_entry->end - grow_amount);
4376 vm_map_entry_resize(map, gap_entry, -grow_amount);
4377 gap_deleted = false;
4379 rv = vm_map_insert(map, NULL, 0, grow_start,
4380 grow_start + grow_amount,
4381 stack_entry->protection, stack_entry->max_protection,
4382 MAP_STACK_GROWS_DOWN);
4383 if (rv != KERN_SUCCESS) {
4385 rv1 = vm_map_insert(map, NULL, 0, gap_start,
4386 gap_end, VM_PROT_NONE, VM_PROT_NONE,
4387 MAP_CREATE_GUARD | MAP_CREATE_STACK_GAP_DN);
4388 MPASS(rv1 == KERN_SUCCESS);
4390 vm_map_entry_resize(map, gap_entry,
4394 grow_start = stack_entry->end;
4395 cred = stack_entry->cred;
4396 if (cred == NULL && stack_entry->object.vm_object != NULL)
4397 cred = stack_entry->object.vm_object->cred;
4398 if (cred != NULL && !swap_reserve_by_cred(grow_amount, cred))
4400 /* Grow the underlying object if applicable. */
4401 else if (stack_entry->object.vm_object == NULL ||
4402 vm_object_coalesce(stack_entry->object.vm_object,
4403 stack_entry->offset,
4404 (vm_size_t)(stack_entry->end - stack_entry->start),
4405 grow_amount, cred != NULL)) {
4406 if (gap_entry->start + grow_amount == gap_entry->end) {
4407 vm_map_entry_delete(map, gap_entry);
4408 vm_map_entry_resize(map, stack_entry,
4411 gap_entry->start += grow_amount;
4412 stack_entry->end += grow_amount;
4414 map->size += grow_amount;
4419 if (rv == KERN_SUCCESS && is_procstack)
4420 vm->vm_ssize += btoc(grow_amount);
4423 * Heed the MAP_WIREFUTURE flag if it was set for this process.
4425 if (rv == KERN_SUCCESS && (map->flags & MAP_WIREFUTURE) != 0) {
4426 rv = vm_map_wire_locked(map, grow_start,
4427 grow_start + grow_amount,
4428 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
4430 vm_map_lock_downgrade(map);
4434 if (racct_enable && rv != KERN_SUCCESS) {
4436 error = racct_set(p, RACCT_VMEM, map->size);
4437 KASSERT(error == 0, ("decreasing RACCT_VMEM failed"));
4439 error = racct_set(p, RACCT_MEMLOCK,
4440 ptoa(pmap_wired_count(map->pmap)));
4441 KASSERT(error == 0, ("decreasing RACCT_MEMLOCK failed"));
4443 error = racct_set(p, RACCT_STACK, ctob(vm->vm_ssize));
4444 KASSERT(error == 0, ("decreasing RACCT_STACK failed"));
4453 * Unshare the specified VM space for exec. If other processes are
4454 * mapped to it, then create a new one. The new vmspace is null.
4457 vmspace_exec(struct proc *p, vm_offset_t minuser, vm_offset_t maxuser)
4459 struct vmspace *oldvmspace = p->p_vmspace;
4460 struct vmspace *newvmspace;
4462 KASSERT((curthread->td_pflags & TDP_EXECVMSPC) == 0,
4463 ("vmspace_exec recursed"));
4464 newvmspace = vmspace_alloc(minuser, maxuser, pmap_pinit);
4465 if (newvmspace == NULL)
4467 newvmspace->vm_swrss = oldvmspace->vm_swrss;
4469 * This code is written like this for prototype purposes. The
4470 * goal is to avoid running down the vmspace here, but let the
4471 * other process's that are still using the vmspace to finally
4472 * run it down. Even though there is little or no chance of blocking
4473 * here, it is a good idea to keep this form for future mods.
4475 PROC_VMSPACE_LOCK(p);
4476 p->p_vmspace = newvmspace;
4477 PROC_VMSPACE_UNLOCK(p);
4478 if (p == curthread->td_proc)
4479 pmap_activate(curthread);
4480 curthread->td_pflags |= TDP_EXECVMSPC;
4485 * Unshare the specified VM space for forcing COW. This
4486 * is called by rfork, for the (RFMEM|RFPROC) == 0 case.
4489 vmspace_unshare(struct proc *p)
4491 struct vmspace *oldvmspace = p->p_vmspace;
4492 struct vmspace *newvmspace;
4493 vm_ooffset_t fork_charge;
4495 if (oldvmspace->vm_refcnt == 1)
4498 newvmspace = vmspace_fork(oldvmspace, &fork_charge);
4499 if (newvmspace == NULL)
4501 if (!swap_reserve_by_cred(fork_charge, p->p_ucred)) {
4502 vmspace_free(newvmspace);
4505 PROC_VMSPACE_LOCK(p);
4506 p->p_vmspace = newvmspace;
4507 PROC_VMSPACE_UNLOCK(p);
4508 if (p == curthread->td_proc)
4509 pmap_activate(curthread);
4510 vmspace_free(oldvmspace);
4517 * Finds the VM object, offset, and
4518 * protection for a given virtual address in the
4519 * specified map, assuming a page fault of the
4522 * Leaves the map in question locked for read; return
4523 * values are guaranteed until a vm_map_lookup_done
4524 * call is performed. Note that the map argument
4525 * is in/out; the returned map must be used in
4526 * the call to vm_map_lookup_done.
4528 * A handle (out_entry) is returned for use in
4529 * vm_map_lookup_done, to make that fast.
4531 * If a lookup is requested with "write protection"
4532 * specified, the map may be changed to perform virtual
4533 * copying operations, although the data referenced will
4537 vm_map_lookup(vm_map_t *var_map, /* IN/OUT */
4539 vm_prot_t fault_typea,
4540 vm_map_entry_t *out_entry, /* OUT */
4541 vm_object_t *object, /* OUT */
4542 vm_pindex_t *pindex, /* OUT */
4543 vm_prot_t *out_prot, /* OUT */
4544 boolean_t *wired) /* OUT */
4546 vm_map_entry_t entry;
4547 vm_map_t map = *var_map;
4549 vm_prot_t fault_type = fault_typea;
4550 vm_object_t eobject;
4556 vm_map_lock_read(map);
4560 * Lookup the faulting address.
4562 if (!vm_map_lookup_entry(map, vaddr, out_entry)) {
4563 vm_map_unlock_read(map);
4564 return (KERN_INVALID_ADDRESS);
4572 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
4573 vm_map_t old_map = map;
4575 *var_map = map = entry->object.sub_map;
4576 vm_map_unlock_read(old_map);
4581 * Check whether this task is allowed to have this page.
4583 prot = entry->protection;
4584 if ((fault_typea & VM_PROT_FAULT_LOOKUP) != 0) {
4585 fault_typea &= ~VM_PROT_FAULT_LOOKUP;
4586 if (prot == VM_PROT_NONE && map != kernel_map &&
4587 (entry->eflags & MAP_ENTRY_GUARD) != 0 &&
4588 (entry->eflags & (MAP_ENTRY_STACK_GAP_DN |
4589 MAP_ENTRY_STACK_GAP_UP)) != 0 &&
4590 vm_map_growstack(map, vaddr, entry) == KERN_SUCCESS)
4591 goto RetryLookupLocked;
4593 fault_type &= VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE;
4594 if ((fault_type & prot) != fault_type || prot == VM_PROT_NONE) {
4595 vm_map_unlock_read(map);
4596 return (KERN_PROTECTION_FAILURE);
4598 KASSERT((prot & VM_PROT_WRITE) == 0 || (entry->eflags &
4599 (MAP_ENTRY_USER_WIRED | MAP_ENTRY_NEEDS_COPY)) !=
4600 (MAP_ENTRY_USER_WIRED | MAP_ENTRY_NEEDS_COPY),
4601 ("entry %p flags %x", entry, entry->eflags));
4602 if ((fault_typea & VM_PROT_COPY) != 0 &&
4603 (entry->max_protection & VM_PROT_WRITE) == 0 &&
4604 (entry->eflags & MAP_ENTRY_COW) == 0) {
4605 vm_map_unlock_read(map);
4606 return (KERN_PROTECTION_FAILURE);
4610 * If this page is not pageable, we have to get it for all possible
4613 *wired = (entry->wired_count != 0);
4615 fault_type = entry->protection;
4616 size = entry->end - entry->start;
4618 * If the entry was copy-on-write, we either ...
4620 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
4622 * If we want to write the page, we may as well handle that
4623 * now since we've got the map locked.
4625 * If we don't need to write the page, we just demote the
4626 * permissions allowed.
4628 if ((fault_type & VM_PROT_WRITE) != 0 ||
4629 (fault_typea & VM_PROT_COPY) != 0) {
4631 * Make a new object, and place it in the object
4632 * chain. Note that no new references have appeared
4633 * -- one just moved from the map to the new
4636 if (vm_map_lock_upgrade(map))
4639 if (entry->cred == NULL) {
4641 * The debugger owner is charged for
4644 cred = curthread->td_ucred;
4646 if (!swap_reserve_by_cred(size, cred)) {
4649 return (KERN_RESOURCE_SHORTAGE);
4653 vm_object_shadow(&entry->object.vm_object,
4654 &entry->offset, size);
4655 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
4656 eobject = entry->object.vm_object;
4657 if (eobject->cred != NULL) {
4659 * The object was not shadowed.
4661 swap_release_by_cred(size, entry->cred);
4662 crfree(entry->cred);
4664 } else if (entry->cred != NULL) {
4665 VM_OBJECT_WLOCK(eobject);
4666 eobject->cred = entry->cred;
4667 eobject->charge = size;
4668 VM_OBJECT_WUNLOCK(eobject);
4672 vm_map_lock_downgrade(map);
4675 * We're attempting to read a copy-on-write page --
4676 * don't allow writes.
4678 prot &= ~VM_PROT_WRITE;
4683 * Create an object if necessary.
4685 if (entry->object.vm_object == NULL &&
4687 if (vm_map_lock_upgrade(map))
4689 entry->object.vm_object = vm_object_allocate(OBJT_DEFAULT,
4692 if (entry->cred != NULL) {
4693 VM_OBJECT_WLOCK(entry->object.vm_object);
4694 entry->object.vm_object->cred = entry->cred;
4695 entry->object.vm_object->charge = size;
4696 VM_OBJECT_WUNLOCK(entry->object.vm_object);
4699 vm_map_lock_downgrade(map);
4703 * Return the object/offset from this entry. If the entry was
4704 * copy-on-write or empty, it has been fixed up.
4706 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
4707 *object = entry->object.vm_object;
4710 return (KERN_SUCCESS);
4714 * vm_map_lookup_locked:
4716 * Lookup the faulting address. A version of vm_map_lookup that returns
4717 * KERN_FAILURE instead of blocking on map lock or memory allocation.
4720 vm_map_lookup_locked(vm_map_t *var_map, /* IN/OUT */
4722 vm_prot_t fault_typea,
4723 vm_map_entry_t *out_entry, /* OUT */
4724 vm_object_t *object, /* OUT */
4725 vm_pindex_t *pindex, /* OUT */
4726 vm_prot_t *out_prot, /* OUT */
4727 boolean_t *wired) /* OUT */
4729 vm_map_entry_t entry;
4730 vm_map_t map = *var_map;
4732 vm_prot_t fault_type = fault_typea;
4735 * Lookup the faulting address.
4737 if (!vm_map_lookup_entry(map, vaddr, out_entry))
4738 return (KERN_INVALID_ADDRESS);
4743 * Fail if the entry refers to a submap.
4745 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP)
4746 return (KERN_FAILURE);
4749 * Check whether this task is allowed to have this page.
4751 prot = entry->protection;
4752 fault_type &= VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE;
4753 if ((fault_type & prot) != fault_type)
4754 return (KERN_PROTECTION_FAILURE);
4757 * If this page is not pageable, we have to get it for all possible
4760 *wired = (entry->wired_count != 0);
4762 fault_type = entry->protection;
4764 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
4766 * Fail if the entry was copy-on-write for a write fault.
4768 if (fault_type & VM_PROT_WRITE)
4769 return (KERN_FAILURE);
4771 * We're attempting to read a copy-on-write page --
4772 * don't allow writes.
4774 prot &= ~VM_PROT_WRITE;
4778 * Fail if an object should be created.
4780 if (entry->object.vm_object == NULL && !map->system_map)
4781 return (KERN_FAILURE);
4784 * Return the object/offset from this entry. If the entry was
4785 * copy-on-write or empty, it has been fixed up.
4787 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
4788 *object = entry->object.vm_object;
4791 return (KERN_SUCCESS);
4795 * vm_map_lookup_done:
4797 * Releases locks acquired by a vm_map_lookup
4798 * (according to the handle returned by that lookup).
4801 vm_map_lookup_done(vm_map_t map, vm_map_entry_t entry)
4804 * Unlock the main-level map
4806 vm_map_unlock_read(map);
4810 vm_map_max_KBI(const struct vm_map *map)
4813 return (vm_map_max(map));
4817 vm_map_min_KBI(const struct vm_map *map)
4820 return (vm_map_min(map));
4824 vm_map_pmap_KBI(vm_map_t map)
4832 _vm_map_assert_consistent(vm_map_t map, int check)
4834 vm_map_entry_t entry, prev;
4835 vm_size_t max_left, max_right;
4837 if (enable_vmmap_check != check)
4840 prev = &map->header;
4841 VM_MAP_ENTRY_FOREACH(entry, map) {
4842 KASSERT(prev->end <= entry->start,
4843 ("map %p prev->end = %jx, start = %jx", map,
4844 (uintmax_t)prev->end, (uintmax_t)entry->start));
4845 KASSERT(entry->start < entry->end,
4846 ("map %p start = %jx, end = %jx", map,
4847 (uintmax_t)entry->start, (uintmax_t)entry->end));
4848 KASSERT(entry->end <= vm_map_entry_succ(entry)->start,
4849 ("map %p end = %jx, next->start = %jx", map,
4850 (uintmax_t)entry->end,
4851 (uintmax_t)vm_map_entry_succ(entry)->start));
4852 KASSERT(entry->left == NULL ||
4853 entry->left->start < entry->start,
4854 ("map %p left->start = %jx, start = %jx", map,
4855 (uintmax_t)entry->left->start, (uintmax_t)entry->start));
4856 KASSERT(entry->right == NULL ||
4857 entry->start < entry->right->start,
4858 ("map %p start = %jx, right->start = %jx", map,
4859 (uintmax_t)entry->start, (uintmax_t)entry->right->start));
4860 max_left = vm_map_entry_max_free_left(entry,
4861 vm_map_entry_pred(entry));
4862 max_right = vm_map_entry_max_free_right(entry,
4863 vm_map_entry_succ(entry));
4864 KASSERT(entry->max_free == MAX(max_left, max_right),
4865 ("map %p max = %jx, max_left = %jx, max_right = %jx", map,
4866 (uintmax_t)entry->max_free,
4867 (uintmax_t)max_left, (uintmax_t)max_right));
4870 KASSERT(prev->end <= entry->start,
4871 ("map %p prev->end = %jx, start = %jx", map,
4872 (uintmax_t)prev->end, (uintmax_t)entry->start));
4876 #include "opt_ddb.h"
4878 #include <sys/kernel.h>
4880 #include <ddb/ddb.h>
4883 vm_map_print(vm_map_t map)
4885 vm_map_entry_t entry, prev;
4887 db_iprintf("Task map %p: pmap=%p, nentries=%d, version=%u\n",
4889 (void *)map->pmap, map->nentries, map->timestamp);
4892 prev = &map->header;
4893 VM_MAP_ENTRY_FOREACH(entry, map) {
4894 db_iprintf("map entry %p: start=%p, end=%p, eflags=%#x, \n",
4895 (void *)entry, (void *)entry->start, (void *)entry->end,
4898 static char *inheritance_name[4] =
4899 {"share", "copy", "none", "donate_copy"};
4901 db_iprintf(" prot=%x/%x/%s",
4903 entry->max_protection,
4904 inheritance_name[(int)(unsigned char)
4905 entry->inheritance]);
4906 if (entry->wired_count != 0)
4907 db_printf(", wired");
4909 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
4910 db_printf(", share=%p, offset=0x%jx\n",
4911 (void *)entry->object.sub_map,
4912 (uintmax_t)entry->offset);
4913 if (prev == &map->header ||
4914 prev->object.sub_map !=
4915 entry->object.sub_map) {
4917 vm_map_print((vm_map_t)entry->object.sub_map);
4921 if (entry->cred != NULL)
4922 db_printf(", ruid %d", entry->cred->cr_ruid);
4923 db_printf(", object=%p, offset=0x%jx",
4924 (void *)entry->object.vm_object,
4925 (uintmax_t)entry->offset);
4926 if (entry->object.vm_object && entry->object.vm_object->cred)
4927 db_printf(", obj ruid %d charge %jx",
4928 entry->object.vm_object->cred->cr_ruid,
4929 (uintmax_t)entry->object.vm_object->charge);
4930 if (entry->eflags & MAP_ENTRY_COW)
4931 db_printf(", copy (%s)",
4932 (entry->eflags & MAP_ENTRY_NEEDS_COPY) ? "needed" : "done");
4935 if (prev == &map->header ||
4936 prev->object.vm_object !=
4937 entry->object.vm_object) {
4939 vm_object_print((db_expr_t)(intptr_t)
4940 entry->object.vm_object,
4950 DB_SHOW_COMMAND(map, map)
4954 db_printf("usage: show map <addr>\n");
4957 vm_map_print((vm_map_t)addr);
4960 DB_SHOW_COMMAND(procvm, procvm)
4965 p = db_lookup_proc(addr);
4970 db_printf("p = %p, vmspace = %p, map = %p, pmap = %p\n",
4971 (void *)p, (void *)p->p_vmspace, (void *)&p->p_vmspace->vm_map,
4972 (void *)vmspace_pmap(p->p_vmspace));
4974 vm_map_print((vm_map_t)&p->p_vmspace->vm_map);