2 * SPDX-License-Identifier: (BSD-3-Clause AND MIT-CMU)
4 * Copyright (c) 1991, 1993
5 * The Regents of the University of California. All rights reserved.
7 * This code is derived from software contributed to Berkeley by
8 * The Mach Operating System project at Carnegie-Mellon University.
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 3. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * from: @(#)vm_map.c 8.3 (Berkeley) 1/12/94
37 * Copyright (c) 1987, 1990 Carnegie-Mellon University.
38 * All rights reserved.
40 * Authors: Avadis Tevanian, Jr., Michael Wayne Young
42 * Permission to use, copy, modify and distribute this software and
43 * its documentation is hereby granted, provided that both the copyright
44 * notice and this permission notice appear in all copies of the
45 * software, derivative works or modified versions, and any portions
46 * thereof, and that both notices appear in supporting documentation.
48 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
49 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
50 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
52 * Carnegie Mellon requests users of this software to return to
54 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
55 * School of Computer Science
56 * Carnegie Mellon University
57 * Pittsburgh PA 15213-3890
59 * any improvements or extensions that they make and grant Carnegie the
60 * rights to redistribute these changes.
64 * Virtual memory mapping module.
67 #include <sys/cdefs.h>
68 __FBSDID("$FreeBSD$");
70 #include <sys/param.h>
71 #include <sys/systm.h>
72 #include <sys/kernel.h>
75 #include <sys/mutex.h>
77 #include <sys/vmmeter.h>
79 #include <sys/vnode.h>
80 #include <sys/racct.h>
81 #include <sys/resourcevar.h>
82 #include <sys/rwlock.h>
84 #include <sys/sysctl.h>
85 #include <sys/sysent.h>
89 #include <vm/vm_param.h>
91 #include <vm/vm_map.h>
92 #include <vm/vm_page.h>
93 #include <vm/vm_pageout.h>
94 #include <vm/vm_object.h>
95 #include <vm/vm_pager.h>
96 #include <vm/vm_kern.h>
97 #include <vm/vm_extern.h>
98 #include <vm/vnode_pager.h>
99 #include <vm/swap_pager.h>
103 * Virtual memory maps provide for the mapping, protection,
104 * and sharing of virtual memory objects. In addition,
105 * this module provides for an efficient virtual copy of
106 * memory from one map to another.
108 * Synchronization is required prior to most operations.
110 * Maps consist of an ordered doubly-linked list of simple
111 * entries; a self-adjusting binary search tree of these
112 * entries is used to speed up lookups.
114 * Since portions of maps are specified by start/end addresses,
115 * which may not align with existing map entries, all
116 * routines merely "clip" entries to these start/end values.
117 * [That is, an entry is split into two, bordering at a
118 * start or end value.] Note that these clippings may not
119 * always be necessary (as the two resulting entries are then
120 * not changed); however, the clipping is done for convenience.
122 * As mentioned above, virtual copy operations are performed
123 * by copying VM object references from one map to
124 * another, and then marking both regions as copy-on-write.
127 static struct mtx map_sleep_mtx;
128 static uma_zone_t mapentzone;
129 static uma_zone_t kmapentzone;
130 static uma_zone_t mapzone;
131 static uma_zone_t vmspace_zone;
132 static int vmspace_zinit(void *mem, int size, int flags);
133 static int vm_map_zinit(void *mem, int ize, int flags);
134 static void _vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min,
136 static void vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map);
137 static void vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry);
138 static void vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry);
139 static int vm_map_growstack(vm_map_t map, vm_offset_t addr,
140 vm_map_entry_t gap_entry);
141 static void vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot,
142 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags);
144 static void vm_map_zdtor(void *mem, int size, void *arg);
145 static void vmspace_zdtor(void *mem, int size, void *arg);
147 static int vm_map_stack_locked(vm_map_t map, vm_offset_t addrbos,
148 vm_size_t max_ssize, vm_size_t growsize, vm_prot_t prot, vm_prot_t max,
150 static void vm_map_wire_entry_failure(vm_map_t map, vm_map_entry_t entry,
151 vm_offset_t failed_addr);
153 #define ENTRY_CHARGED(e) ((e)->cred != NULL || \
154 ((e)->object.vm_object != NULL && (e)->object.vm_object->cred != NULL && \
155 !((e)->eflags & MAP_ENTRY_NEEDS_COPY)))
158 * PROC_VMSPACE_{UN,}LOCK() can be a noop as long as vmspaces are type
161 #define PROC_VMSPACE_LOCK(p) do { } while (0)
162 #define PROC_VMSPACE_UNLOCK(p) do { } while (0)
165 * VM_MAP_RANGE_CHECK: [ internal use only ]
167 * Asserts that the starting and ending region
168 * addresses fall within the valid range of the map.
170 #define VM_MAP_RANGE_CHECK(map, start, end) \
172 if (start < vm_map_min(map)) \
173 start = vm_map_min(map); \
174 if (end > vm_map_max(map)) \
175 end = vm_map_max(map); \
183 * Initialize the vm_map module. Must be called before
184 * any other vm_map routines.
186 * Map and entry structures are allocated from the general
187 * purpose memory pool with some exceptions:
189 * - The kernel map and kmem submap are allocated statically.
190 * - Kernel map entries are allocated out of a static pool.
192 * These restrictions are necessary since malloc() uses the
193 * maps and requires map entries.
199 mtx_init(&map_sleep_mtx, "vm map sleep mutex", NULL, MTX_DEF);
200 mapzone = uma_zcreate("MAP", sizeof(struct vm_map), NULL,
206 vm_map_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
207 uma_prealloc(mapzone, MAX_KMAP);
208 kmapentzone = uma_zcreate("KMAP ENTRY", sizeof(struct vm_map_entry),
209 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR,
210 UMA_ZONE_MTXCLASS | UMA_ZONE_VM);
211 mapentzone = uma_zcreate("MAP ENTRY", sizeof(struct vm_map_entry),
212 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
213 vmspace_zone = uma_zcreate("VMSPACE", sizeof(struct vmspace), NULL,
219 vmspace_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
223 vmspace_zinit(void *mem, int size, int flags)
227 vm = (struct vmspace *)mem;
229 vm->vm_map.pmap = NULL;
230 (void)vm_map_zinit(&vm->vm_map, sizeof(vm->vm_map), flags);
231 PMAP_LOCK_INIT(vmspace_pmap(vm));
236 vm_map_zinit(void *mem, int size, int flags)
241 memset(map, 0, sizeof(*map));
242 mtx_init(&map->system_mtx, "vm map (system)", NULL, MTX_DEF | MTX_DUPOK);
243 sx_init(&map->lock, "vm map (user)");
249 vmspace_zdtor(void *mem, int size, void *arg)
253 vm = (struct vmspace *)mem;
255 vm_map_zdtor(&vm->vm_map, sizeof(vm->vm_map), arg);
258 vm_map_zdtor(void *mem, int size, void *arg)
263 KASSERT(map->nentries == 0,
264 ("map %p nentries == %d on free.",
265 map, map->nentries));
266 KASSERT(map->size == 0,
267 ("map %p size == %lu on free.",
268 map, (unsigned long)map->size));
270 #endif /* INVARIANTS */
273 * Allocate a vmspace structure, including a vm_map and pmap,
274 * and initialize those structures. The refcnt is set to 1.
276 * If 'pinit' is NULL then the embedded pmap is initialized via pmap_pinit().
279 vmspace_alloc(vm_offset_t min, vm_offset_t max, pmap_pinit_t pinit)
283 vm = uma_zalloc(vmspace_zone, M_WAITOK);
284 KASSERT(vm->vm_map.pmap == NULL, ("vm_map.pmap must be NULL"));
285 if (!pinit(vmspace_pmap(vm))) {
286 uma_zfree(vmspace_zone, vm);
289 CTR1(KTR_VM, "vmspace_alloc: %p", vm);
290 _vm_map_init(&vm->vm_map, vmspace_pmap(vm), min, max);
305 vmspace_container_reset(struct proc *p)
309 racct_set(p, RACCT_DATA, 0);
310 racct_set(p, RACCT_STACK, 0);
311 racct_set(p, RACCT_RSS, 0);
312 racct_set(p, RACCT_MEMLOCK, 0);
313 racct_set(p, RACCT_VMEM, 0);
319 vmspace_dofree(struct vmspace *vm)
322 CTR1(KTR_VM, "vmspace_free: %p", vm);
325 * Make sure any SysV shm is freed, it might not have been in
331 * Lock the map, to wait out all other references to it.
332 * Delete all of the mappings and pages they hold, then call
333 * the pmap module to reclaim anything left.
335 (void)vm_map_remove(&vm->vm_map, vm_map_min(&vm->vm_map),
336 vm_map_max(&vm->vm_map));
338 pmap_release(vmspace_pmap(vm));
339 vm->vm_map.pmap = NULL;
340 uma_zfree(vmspace_zone, vm);
344 vmspace_free(struct vmspace *vm)
347 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
348 "vmspace_free() called");
350 if (vm->vm_refcnt == 0)
351 panic("vmspace_free: attempt to free already freed vmspace");
353 if (atomic_fetchadd_int(&vm->vm_refcnt, -1) == 1)
358 vmspace_exitfree(struct proc *p)
362 PROC_VMSPACE_LOCK(p);
365 PROC_VMSPACE_UNLOCK(p);
366 KASSERT(vm == &vmspace0, ("vmspace_exitfree: wrong vmspace"));
371 vmspace_exit(struct thread *td)
378 * Release user portion of address space.
379 * This releases references to vnodes,
380 * which could cause I/O if the file has been unlinked.
381 * Need to do this early enough that we can still sleep.
383 * The last exiting process to reach this point releases as
384 * much of the environment as it can. vmspace_dofree() is the
385 * slower fallback in case another process had a temporary
386 * reference to the vmspace.
391 atomic_add_int(&vmspace0.vm_refcnt, 1);
392 refcnt = vm->vm_refcnt;
394 if (refcnt > 1 && p->p_vmspace != &vmspace0) {
395 /* Switch now since other proc might free vmspace */
396 PROC_VMSPACE_LOCK(p);
397 p->p_vmspace = &vmspace0;
398 PROC_VMSPACE_UNLOCK(p);
401 } while (!atomic_fcmpset_int(&vm->vm_refcnt, &refcnt, refcnt - 1));
403 if (p->p_vmspace != vm) {
404 /* vmspace not yet freed, switch back */
405 PROC_VMSPACE_LOCK(p);
407 PROC_VMSPACE_UNLOCK(p);
410 pmap_remove_pages(vmspace_pmap(vm));
411 /* Switch now since this proc will free vmspace */
412 PROC_VMSPACE_LOCK(p);
413 p->p_vmspace = &vmspace0;
414 PROC_VMSPACE_UNLOCK(p);
420 vmspace_container_reset(p);
424 /* Acquire reference to vmspace owned by another process. */
427 vmspace_acquire_ref(struct proc *p)
432 PROC_VMSPACE_LOCK(p);
435 PROC_VMSPACE_UNLOCK(p);
438 refcnt = vm->vm_refcnt;
440 if (refcnt <= 0) { /* Avoid 0->1 transition */
441 PROC_VMSPACE_UNLOCK(p);
444 } while (!atomic_fcmpset_int(&vm->vm_refcnt, &refcnt, refcnt + 1));
445 if (vm != p->p_vmspace) {
446 PROC_VMSPACE_UNLOCK(p);
450 PROC_VMSPACE_UNLOCK(p);
455 * Switch between vmspaces in an AIO kernel process.
457 * The AIO kernel processes switch to and from a user process's
458 * vmspace while performing an I/O operation on behalf of a user
459 * process. The new vmspace is either the vmspace of a user process
460 * obtained from an active AIO request or the initial vmspace of the
461 * AIO kernel process (when it is idling). Because user processes
462 * will block to drain any active AIO requests before proceeding in
463 * exit() or execve(), the vmspace reference count for these vmspaces
464 * can never be 0. This allows for a much simpler implementation than
465 * the loop in vmspace_acquire_ref() above. Similarly, AIO kernel
466 * processes hold an extra reference on their initial vmspace for the
467 * life of the process so that this guarantee is true for any vmspace
471 vmspace_switch_aio(struct vmspace *newvm)
473 struct vmspace *oldvm;
475 /* XXX: Need some way to assert that this is an aio daemon. */
477 KASSERT(newvm->vm_refcnt > 0,
478 ("vmspace_switch_aio: newvm unreferenced"));
480 oldvm = curproc->p_vmspace;
485 * Point to the new address space and refer to it.
487 curproc->p_vmspace = newvm;
488 atomic_add_int(&newvm->vm_refcnt, 1);
490 /* Activate the new mapping. */
491 pmap_activate(curthread);
493 /* Remove the daemon's reference to the old address space. */
494 KASSERT(oldvm->vm_refcnt > 1,
495 ("vmspace_switch_aio: oldvm dropping last reference"));
500 _vm_map_lock(vm_map_t map, const char *file, int line)
504 mtx_lock_flags_(&map->system_mtx, 0, file, line);
506 sx_xlock_(&map->lock, file, line);
511 vm_map_entry_set_vnode_text(vm_map_entry_t entry, bool add)
513 vm_object_t object, object1;
516 if ((entry->eflags & MAP_ENTRY_VN_EXEC) == 0)
518 KASSERT((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0,
519 ("Submap with execs"));
520 object = entry->object.vm_object;
521 KASSERT(object != NULL, ("No object for text, entry %p", entry));
522 VM_OBJECT_RLOCK(object);
523 while ((object1 = object->backing_object) != NULL) {
524 VM_OBJECT_RLOCK(object1);
525 VM_OBJECT_RUNLOCK(object);
530 * For OBJT_DEAD objects, v_writecount was handled in
531 * vnode_pager_dealloc().
533 if (object->type != OBJT_DEAD) {
534 KASSERT(((object->flags & OBJ_TMPFS) == 0 &&
535 object->type == OBJT_VNODE) ||
536 ((object->flags & OBJ_TMPFS) != 0 &&
537 object->type == OBJT_SWAP),
538 ("vm_map_entry_set_vnode_text: wrong object type, "
539 "entry %p, object %p, add %d", entry, object, add));
540 vp = (object->flags & OBJ_TMPFS) == 0 ? object->handle :
541 object->un_pager.swp.swp_tmpfs;
543 VOP_SET_TEXT_CHECKED(vp);
545 VOP_UNSET_TEXT_CHECKED(vp);
547 VM_OBJECT_RUNLOCK(object);
551 vm_map_process_deferred(void)
554 vm_map_entry_t entry, next;
558 entry = td->td_map_def_user;
559 td->td_map_def_user = NULL;
560 while (entry != NULL) {
562 MPASS((entry->eflags & (MAP_ENTRY_VN_WRITECNT |
563 MAP_ENTRY_VN_EXEC)) != (MAP_ENTRY_VN_WRITECNT |
565 if ((entry->eflags & MAP_ENTRY_VN_WRITECNT) != 0) {
567 * Decrement the object's writemappings and
568 * possibly the vnode's v_writecount.
570 KASSERT((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0,
571 ("Submap with writecount"));
572 object = entry->object.vm_object;
573 KASSERT(object != NULL, ("No object for writecount"));
574 vnode_pager_release_writecount(object, entry->start,
577 vm_map_entry_set_vnode_text(entry, false);
578 vm_map_entry_deallocate(entry, FALSE);
584 _vm_map_unlock(vm_map_t map, const char *file, int line)
588 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
590 sx_xunlock_(&map->lock, file, line);
591 vm_map_process_deferred();
596 _vm_map_lock_read(vm_map_t map, const char *file, int line)
600 mtx_lock_flags_(&map->system_mtx, 0, file, line);
602 sx_slock_(&map->lock, file, line);
606 _vm_map_unlock_read(vm_map_t map, const char *file, int line)
610 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
612 sx_sunlock_(&map->lock, file, line);
613 vm_map_process_deferred();
618 _vm_map_trylock(vm_map_t map, const char *file, int line)
622 error = map->system_map ?
623 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) :
624 !sx_try_xlock_(&map->lock, file, line);
631 _vm_map_trylock_read(vm_map_t map, const char *file, int line)
635 error = map->system_map ?
636 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) :
637 !sx_try_slock_(&map->lock, file, line);
642 * _vm_map_lock_upgrade: [ internal use only ]
644 * Tries to upgrade a read (shared) lock on the specified map to a write
645 * (exclusive) lock. Returns the value "0" if the upgrade succeeds and a
646 * non-zero value if the upgrade fails. If the upgrade fails, the map is
647 * returned without a read or write lock held.
649 * Requires that the map be read locked.
652 _vm_map_lock_upgrade(vm_map_t map, const char *file, int line)
654 unsigned int last_timestamp;
656 if (map->system_map) {
657 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
659 if (!sx_try_upgrade_(&map->lock, file, line)) {
660 last_timestamp = map->timestamp;
661 sx_sunlock_(&map->lock, file, line);
662 vm_map_process_deferred();
664 * If the map's timestamp does not change while the
665 * map is unlocked, then the upgrade succeeds.
667 sx_xlock_(&map->lock, file, line);
668 if (last_timestamp != map->timestamp) {
669 sx_xunlock_(&map->lock, file, line);
679 _vm_map_lock_downgrade(vm_map_t map, const char *file, int line)
682 if (map->system_map) {
683 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
685 sx_downgrade_(&map->lock, file, line);
691 * Returns a non-zero value if the caller holds a write (exclusive) lock
692 * on the specified map and the value "0" otherwise.
695 vm_map_locked(vm_map_t map)
699 return (mtx_owned(&map->system_mtx));
701 return (sx_xlocked(&map->lock));
706 _vm_map_assert_locked(vm_map_t map, const char *file, int line)
710 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
712 sx_assert_(&map->lock, SA_XLOCKED, file, line);
715 #define VM_MAP_ASSERT_LOCKED(map) \
716 _vm_map_assert_locked(map, LOCK_FILE, LOCK_LINE)
719 static int enable_vmmap_check = 1;
721 static int enable_vmmap_check = 0;
723 SYSCTL_INT(_debug, OID_AUTO, vmmap_check, CTLFLAG_RWTUN,
724 &enable_vmmap_check, 0, "Enable vm map consistency checking");
727 _vm_map_assert_consistent(vm_map_t map)
729 vm_map_entry_t entry;
730 vm_map_entry_t child;
731 vm_size_t max_left, max_right;
733 if (!enable_vmmap_check)
736 for (entry = map->header.next; entry != &map->header;
737 entry = entry->next) {
738 KASSERT(entry->prev->end <= entry->start,
739 ("map %p prev->end = %jx, start = %jx", map,
740 (uintmax_t)entry->prev->end, (uintmax_t)entry->start));
741 KASSERT(entry->start < entry->end,
742 ("map %p start = %jx, end = %jx", map,
743 (uintmax_t)entry->start, (uintmax_t)entry->end));
744 KASSERT(entry->end <= entry->next->start,
745 ("map %p end = %jx, next->start = %jx", map,
746 (uintmax_t)entry->end, (uintmax_t)entry->next->start));
747 KASSERT(entry->left == NULL ||
748 entry->left->start < entry->start,
749 ("map %p left->start = %jx, start = %jx", map,
750 (uintmax_t)entry->left->start, (uintmax_t)entry->start));
751 KASSERT(entry->right == NULL ||
752 entry->start < entry->right->start,
753 ("map %p start = %jx, right->start = %jx", map,
754 (uintmax_t)entry->start, (uintmax_t)entry->right->start));
756 max_left = (child != NULL) ? child->max_free :
757 entry->start - entry->prev->end;
758 child = entry->right;
759 max_right = (child != NULL) ? child->max_free :
760 entry->next->start - entry->end;
761 KASSERT(entry->max_free == MAX(max_left, max_right),
762 ("map %p max = %jx, max_left = %jx, max_right = %jx", map,
763 (uintmax_t)entry->max_free,
764 (uintmax_t)max_left, (uintmax_t)max_right));
768 #define VM_MAP_ASSERT_CONSISTENT(map) \
769 _vm_map_assert_consistent(map)
771 #define VM_MAP_ASSERT_LOCKED(map)
772 #define VM_MAP_ASSERT_CONSISTENT(map)
773 #endif /* INVARIANTS */
776 * _vm_map_unlock_and_wait:
778 * Atomically releases the lock on the specified map and puts the calling
779 * thread to sleep. The calling thread will remain asleep until either
780 * vm_map_wakeup() is performed on the map or the specified timeout is
783 * WARNING! This function does not perform deferred deallocations of
784 * objects and map entries. Therefore, the calling thread is expected to
785 * reacquire the map lock after reawakening and later perform an ordinary
786 * unlock operation, such as vm_map_unlock(), before completing its
787 * operation on the map.
790 _vm_map_unlock_and_wait(vm_map_t map, int timo, const char *file, int line)
793 mtx_lock(&map_sleep_mtx);
795 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
797 sx_xunlock_(&map->lock, file, line);
798 return (msleep(&map->root, &map_sleep_mtx, PDROP | PVM, "vmmaps",
805 * Awaken any threads that have slept on the map using
806 * vm_map_unlock_and_wait().
809 vm_map_wakeup(vm_map_t map)
813 * Acquire and release map_sleep_mtx to prevent a wakeup()
814 * from being performed (and lost) between the map unlock
815 * and the msleep() in _vm_map_unlock_and_wait().
817 mtx_lock(&map_sleep_mtx);
818 mtx_unlock(&map_sleep_mtx);
823 vm_map_busy(vm_map_t map)
826 VM_MAP_ASSERT_LOCKED(map);
831 vm_map_unbusy(vm_map_t map)
834 VM_MAP_ASSERT_LOCKED(map);
835 KASSERT(map->busy, ("vm_map_unbusy: not busy"));
836 if (--map->busy == 0 && (map->flags & MAP_BUSY_WAKEUP)) {
837 vm_map_modflags(map, 0, MAP_BUSY_WAKEUP);
843 vm_map_wait_busy(vm_map_t map)
846 VM_MAP_ASSERT_LOCKED(map);
848 vm_map_modflags(map, MAP_BUSY_WAKEUP, 0);
850 msleep(&map->busy, &map->system_mtx, 0, "mbusy", 0);
852 sx_sleep(&map->busy, &map->lock, 0, "mbusy", 0);
858 vmspace_resident_count(struct vmspace *vmspace)
860 return pmap_resident_count(vmspace_pmap(vmspace));
866 * Creates and returns a new empty VM map with
867 * the given physical map structure, and having
868 * the given lower and upper address bounds.
871 vm_map_create(pmap_t pmap, vm_offset_t min, vm_offset_t max)
875 result = uma_zalloc(mapzone, M_WAITOK);
876 CTR1(KTR_VM, "vm_map_create: %p", result);
877 _vm_map_init(result, pmap, min, max);
882 * Initialize an existing vm_map structure
883 * such as that in the vmspace structure.
886 _vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max)
889 map->header.next = map->header.prev = &map->header;
890 map->header.eflags = MAP_ENTRY_HEADER;
891 map->needs_wakeup = FALSE;
894 map->header.end = min;
895 map->header.start = max;
904 vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max)
907 _vm_map_init(map, pmap, min, max);
908 mtx_init(&map->system_mtx, "system map", NULL, MTX_DEF | MTX_DUPOK);
909 sx_init(&map->lock, "user map");
913 * vm_map_entry_dispose: [ internal use only ]
915 * Inverse of vm_map_entry_create.
918 vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry)
920 uma_zfree(map->system_map ? kmapentzone : mapentzone, entry);
924 * vm_map_entry_create: [ internal use only ]
926 * Allocates a VM map entry for insertion.
927 * No entry fields are filled in.
929 static vm_map_entry_t
930 vm_map_entry_create(vm_map_t map)
932 vm_map_entry_t new_entry;
935 new_entry = uma_zalloc(kmapentzone, M_NOWAIT);
937 new_entry = uma_zalloc(mapentzone, M_WAITOK);
938 if (new_entry == NULL)
939 panic("vm_map_entry_create: kernel resources exhausted");
944 * vm_map_entry_set_behavior:
946 * Set the expected access behavior, either normal, random, or
950 vm_map_entry_set_behavior(vm_map_entry_t entry, u_char behavior)
952 entry->eflags = (entry->eflags & ~MAP_ENTRY_BEHAV_MASK) |
953 (behavior & MAP_ENTRY_BEHAV_MASK);
957 * vm_map_entry_set_max_free:
959 * Set the max_free field in a vm_map_entry.
962 vm_map_entry_set_max_free(vm_map_entry_t entry)
964 vm_map_entry_t child;
965 vm_size_t max_left, max_right;
968 max_left = (child != NULL) ? child->max_free :
969 entry->start - entry->prev->end;
970 child = entry->right;
971 max_right = (child != NULL) ? child->max_free :
972 entry->next->start - entry->end;
973 entry->max_free = MAX(max_left, max_right);
976 #define SPLAY_LEFT_STEP(root, y, rlist, test) do { \
978 if (y != NULL && (test)) { \
979 /* Rotate right and make y root. */ \
980 root->left = y->right; \
982 vm_map_entry_set_max_free(root); \
986 /* Put root on rlist. */ \
987 root->left = rlist; \
992 #define SPLAY_RIGHT_STEP(root, y, llist, test) do { \
994 if (y != NULL && (test)) { \
995 /* Rotate left and make y root. */ \
996 root->right = y->left; \
998 vm_map_entry_set_max_free(root); \
1002 /* Put root on llist. */ \
1003 root->right = llist; \
1009 * Walk down the tree until we find addr or a NULL pointer where addr would go,
1010 * breaking off left and right subtrees of nodes less than, or greater than
1011 * addr. Treat pointers to nodes with max_free < length as NULL pointers.
1012 * llist and rlist are the two sides in reverse order (bottom-up), with llist
1013 * linked by the right pointer and rlist linked by the left pointer in the
1016 static vm_map_entry_t
1017 vm_map_splay_split(vm_offset_t addr, vm_size_t length,
1018 vm_map_entry_t root, vm_map_entry_t *out_llist, vm_map_entry_t *out_rlist)
1020 vm_map_entry_t llist, rlist;
1025 while (root != NULL && root->max_free >= length) {
1026 if (addr < root->start) {
1027 SPLAY_LEFT_STEP(root, y, rlist,
1028 y->max_free >= length && addr < y->start);
1029 } else if (addr >= root->end) {
1030 SPLAY_RIGHT_STEP(root, y, llist,
1031 y->max_free >= length && addr >= y->end);
1041 vm_map_splay_findnext(vm_map_entry_t root, vm_map_entry_t *iolist)
1043 vm_map_entry_t rlist, y;
1047 while (root != NULL)
1048 SPLAY_LEFT_STEP(root, y, rlist, true);
1053 vm_map_splay_findprev(vm_map_entry_t root, vm_map_entry_t *iolist)
1055 vm_map_entry_t llist, y;
1059 while (root != NULL)
1060 SPLAY_RIGHT_STEP(root, y, llist, true);
1065 * Walk back up the two spines, flip the pointers and set max_free. The
1066 * subtrees of the root go at the bottom of llist and rlist.
1068 static vm_map_entry_t
1069 vm_map_splay_merge(vm_map_entry_t root,
1070 vm_map_entry_t llist, vm_map_entry_t rlist,
1071 vm_map_entry_t ltree, vm_map_entry_t rtree)
1075 while (llist != NULL) {
1077 llist->right = ltree;
1078 vm_map_entry_set_max_free(llist);
1082 while (rlist != NULL) {
1084 rlist->left = rtree;
1085 vm_map_entry_set_max_free(rlist);
1091 * Final assembly: add ltree and rtree as subtrees of root.
1094 root->right = rtree;
1095 vm_map_entry_set_max_free(root);
1101 * vm_map_entry_splay:
1103 * The Sleator and Tarjan top-down splay algorithm with the
1104 * following variation. Max_free must be computed bottom-up, so
1105 * on the downward pass, maintain the left and right spines in
1106 * reverse order. Then, make a second pass up each side to fix
1107 * the pointers and compute max_free. The time bound is O(log n)
1110 * The new root is the vm_map_entry containing "addr", or else an
1111 * adjacent entry (lower if possible) if addr is not in the tree.
1113 * The map must be locked, and leaves it so.
1115 * Returns: the new root.
1117 static vm_map_entry_t
1118 vm_map_entry_splay(vm_offset_t addr, vm_map_entry_t root)
1120 vm_map_entry_t llist, rlist;
1122 root = vm_map_splay_split(addr, 0, root, &llist, &rlist);
1125 } else if (llist != NULL) {
1127 * Recover the greatest node in the left
1128 * subtree and make it the root.
1131 llist = root->right;
1133 } else if (rlist != NULL) {
1135 * Recover the least node in the right
1136 * subtree and make it the root.
1142 /* There is no root. */
1145 return (vm_map_splay_merge(root, llist, rlist,
1146 root->left, root->right));
1150 * vm_map_entry_{un,}link:
1152 * Insert/remove entries from maps.
1155 vm_map_entry_link(vm_map_t map,
1156 vm_map_entry_t entry)
1158 vm_map_entry_t llist, rlist, root;
1161 "vm_map_entry_link: map %p, nentries %d, entry %p", map,
1162 map->nentries, entry);
1163 VM_MAP_ASSERT_LOCKED(map);
1166 root = vm_map_splay_split(entry->start, 0, root, &llist, &rlist);
1167 KASSERT(root == NULL,
1168 ("vm_map_entry_link: link object already mapped"));
1169 entry->prev = (llist == NULL) ? &map->header : llist;
1170 entry->next = (rlist == NULL) ? &map->header : rlist;
1171 entry->prev->next = entry->next->prev = entry;
1172 root = vm_map_splay_merge(entry, llist, rlist, NULL, NULL);
1174 VM_MAP_ASSERT_CONSISTENT(map);
1177 enum unlink_merge_type {
1184 vm_map_entry_unlink(vm_map_t map,
1185 vm_map_entry_t entry,
1186 enum unlink_merge_type op)
1188 vm_map_entry_t llist, rlist, root, y;
1190 VM_MAP_ASSERT_LOCKED(map);
1191 llist = entry->prev;
1192 rlist = entry->next;
1193 llist->next = rlist;
1194 rlist->prev = llist;
1196 root = vm_map_splay_split(entry->start, 0, root, &llist, &rlist);
1197 KASSERT(root != NULL,
1198 ("vm_map_entry_unlink: unlink object not mapped"));
1201 case UNLINK_MERGE_PREV:
1202 vm_map_splay_findprev(root, &llist);
1203 llist->end = root->end;
1206 llist = root->right;
1209 case UNLINK_MERGE_NEXT:
1210 vm_map_splay_findnext(root, &rlist);
1211 rlist->start = root->start;
1212 rlist->offset = root->offset;
1218 case UNLINK_MERGE_NONE:
1219 vm_map_splay_findprev(root, &llist);
1220 vm_map_splay_findnext(root, &rlist);
1221 if (llist != NULL) {
1223 llist = root->right;
1225 } else if (rlist != NULL) {
1234 root = vm_map_splay_merge(root, llist, rlist,
1235 root->left, root->right);
1237 VM_MAP_ASSERT_CONSISTENT(map);
1239 CTR3(KTR_VM, "vm_map_entry_unlink: map %p, nentries %d, entry %p", map,
1240 map->nentries, entry);
1244 * vm_map_entry_resize_free:
1246 * Recompute the amount of free space following a modified vm_map_entry
1247 * and propagate those values up the tree. Call this function after
1248 * resizing a map entry in-place by changing the end value, without a
1249 * call to vm_map_entry_link() or _unlink().
1251 * The map must be locked, and leaves it so.
1254 vm_map_entry_resize_free(vm_map_t map, vm_map_entry_t entry)
1256 vm_map_entry_t llist, rlist, root;
1258 VM_MAP_ASSERT_LOCKED(map);
1260 root = vm_map_splay_split(entry->start, 0, root, &llist, &rlist);
1261 KASSERT(root != NULL,
1262 ("vm_map_entry_resize_free: resize_free object not mapped"));
1263 vm_map_splay_findnext(root, &rlist);
1265 map->root = vm_map_splay_merge(root, llist, rlist,
1266 root->left, root->right);
1267 VM_MAP_ASSERT_CONSISTENT(map);
1268 CTR3(KTR_VM, "vm_map_entry_resize_free: map %p, nentries %d, entry %p", map,
1269 map->nentries, entry);
1273 * vm_map_lookup_entry: [ internal use only ]
1275 * Finds the map entry containing (or
1276 * immediately preceding) the specified address
1277 * in the given map; the entry is returned
1278 * in the "entry" parameter. The boolean
1279 * result indicates whether the address is
1280 * actually contained in the map.
1283 vm_map_lookup_entry(
1285 vm_offset_t address,
1286 vm_map_entry_t *entry) /* OUT */
1288 vm_map_entry_t cur, lbound;
1292 * If the map is empty, then the map entry immediately preceding
1293 * "address" is the map's header.
1297 *entry = &map->header;
1300 if (address >= cur->start && cur->end > address) {
1304 if ((locked = vm_map_locked(map)) ||
1305 sx_try_upgrade(&map->lock)) {
1307 * Splay requires a write lock on the map. However, it only
1308 * restructures the binary search tree; it does not otherwise
1309 * change the map. Thus, the map's timestamp need not change
1310 * on a temporary upgrade.
1312 map->root = cur = vm_map_entry_splay(address, cur);
1313 VM_MAP_ASSERT_CONSISTENT(map);
1315 sx_downgrade(&map->lock);
1318 * If "address" is contained within a map entry, the new root
1319 * is that map entry. Otherwise, the new root is a map entry
1320 * immediately before or after "address".
1322 if (address < cur->start) {
1323 *entry = &map->header;
1327 return (address < cur->end);
1330 * Since the map is only locked for read access, perform a
1331 * standard binary search tree lookup for "address".
1333 lbound = &map->header;
1335 if (address < cur->start) {
1337 } else if (cur->end <= address) {
1344 } while (cur != NULL);
1352 * Inserts the given whole VM object into the target
1353 * map at the specified address range. The object's
1354 * size should match that of the address range.
1356 * Requires that the map be locked, and leaves it so.
1358 * If object is non-NULL, ref count must be bumped by caller
1359 * prior to making call to account for the new entry.
1362 vm_map_insert(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1363 vm_offset_t start, vm_offset_t end, vm_prot_t prot, vm_prot_t max, int cow)
1365 vm_map_entry_t new_entry, prev_entry, temp_entry;
1367 vm_eflags_t protoeflags;
1368 vm_inherit_t inheritance;
1370 VM_MAP_ASSERT_LOCKED(map);
1371 KASSERT(object != kernel_object ||
1372 (cow & MAP_COPY_ON_WRITE) == 0,
1373 ("vm_map_insert: kernel object and COW"));
1374 KASSERT(object == NULL || (cow & MAP_NOFAULT) == 0,
1375 ("vm_map_insert: paradoxical MAP_NOFAULT request"));
1376 KASSERT((prot & ~max) == 0,
1377 ("prot %#x is not subset of max_prot %#x", prot, max));
1380 * Check that the start and end points are not bogus.
1382 if (start < vm_map_min(map) || end > vm_map_max(map) ||
1384 return (KERN_INVALID_ADDRESS);
1387 * Find the entry prior to the proposed starting address; if it's part
1388 * of an existing entry, this range is bogus.
1390 if (vm_map_lookup_entry(map, start, &temp_entry))
1391 return (KERN_NO_SPACE);
1393 prev_entry = temp_entry;
1396 * Assert that the next entry doesn't overlap the end point.
1398 if (prev_entry->next->start < end)
1399 return (KERN_NO_SPACE);
1401 if ((cow & MAP_CREATE_GUARD) != 0 && (object != NULL ||
1402 max != VM_PROT_NONE))
1403 return (KERN_INVALID_ARGUMENT);
1406 if (cow & MAP_COPY_ON_WRITE)
1407 protoeflags |= MAP_ENTRY_COW | MAP_ENTRY_NEEDS_COPY;
1408 if (cow & MAP_NOFAULT)
1409 protoeflags |= MAP_ENTRY_NOFAULT;
1410 if (cow & MAP_DISABLE_SYNCER)
1411 protoeflags |= MAP_ENTRY_NOSYNC;
1412 if (cow & MAP_DISABLE_COREDUMP)
1413 protoeflags |= MAP_ENTRY_NOCOREDUMP;
1414 if (cow & MAP_STACK_GROWS_DOWN)
1415 protoeflags |= MAP_ENTRY_GROWS_DOWN;
1416 if (cow & MAP_STACK_GROWS_UP)
1417 protoeflags |= MAP_ENTRY_GROWS_UP;
1418 if (cow & MAP_VN_WRITECOUNT)
1419 protoeflags |= MAP_ENTRY_VN_WRITECNT;
1420 if (cow & MAP_VN_EXEC)
1421 protoeflags |= MAP_ENTRY_VN_EXEC;
1422 if ((cow & MAP_CREATE_GUARD) != 0)
1423 protoeflags |= MAP_ENTRY_GUARD;
1424 if ((cow & MAP_CREATE_STACK_GAP_DN) != 0)
1425 protoeflags |= MAP_ENTRY_STACK_GAP_DN;
1426 if ((cow & MAP_CREATE_STACK_GAP_UP) != 0)
1427 protoeflags |= MAP_ENTRY_STACK_GAP_UP;
1428 if (cow & MAP_INHERIT_SHARE)
1429 inheritance = VM_INHERIT_SHARE;
1431 inheritance = VM_INHERIT_DEFAULT;
1434 if ((cow & (MAP_ACC_NO_CHARGE | MAP_NOFAULT | MAP_CREATE_GUARD)) != 0)
1436 if ((cow & MAP_ACC_CHARGED) || ((prot & VM_PROT_WRITE) &&
1437 ((protoeflags & MAP_ENTRY_NEEDS_COPY) || object == NULL))) {
1438 if (!(cow & MAP_ACC_CHARGED) && !swap_reserve(end - start))
1439 return (KERN_RESOURCE_SHORTAGE);
1440 KASSERT(object == NULL ||
1441 (protoeflags & MAP_ENTRY_NEEDS_COPY) != 0 ||
1442 object->cred == NULL,
1443 ("overcommit: vm_map_insert o %p", object));
1444 cred = curthread->td_ucred;
1448 /* Expand the kernel pmap, if necessary. */
1449 if (map == kernel_map && end > kernel_vm_end)
1450 pmap_growkernel(end);
1451 if (object != NULL) {
1453 * OBJ_ONEMAPPING must be cleared unless this mapping
1454 * is trivially proven to be the only mapping for any
1455 * of the object's pages. (Object granularity
1456 * reference counting is insufficient to recognize
1457 * aliases with precision.)
1459 VM_OBJECT_WLOCK(object);
1460 if (object->ref_count > 1 || object->shadow_count != 0)
1461 vm_object_clear_flag(object, OBJ_ONEMAPPING);
1462 VM_OBJECT_WUNLOCK(object);
1463 } else if ((prev_entry->eflags & ~MAP_ENTRY_USER_WIRED) ==
1465 (cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP |
1466 MAP_VN_EXEC)) == 0 &&
1467 prev_entry->end == start && (prev_entry->cred == cred ||
1468 (prev_entry->object.vm_object != NULL &&
1469 prev_entry->object.vm_object->cred == cred)) &&
1470 vm_object_coalesce(prev_entry->object.vm_object,
1472 (vm_size_t)(prev_entry->end - prev_entry->start),
1473 (vm_size_t)(end - prev_entry->end), cred != NULL &&
1474 (protoeflags & MAP_ENTRY_NEEDS_COPY) == 0)) {
1476 * We were able to extend the object. Determine if we
1477 * can extend the previous map entry to include the
1478 * new range as well.
1480 if (prev_entry->inheritance == inheritance &&
1481 prev_entry->protection == prot &&
1482 prev_entry->max_protection == max &&
1483 prev_entry->wired_count == 0) {
1484 KASSERT((prev_entry->eflags & MAP_ENTRY_USER_WIRED) ==
1485 0, ("prev_entry %p has incoherent wiring",
1487 if ((prev_entry->eflags & MAP_ENTRY_GUARD) == 0)
1488 map->size += end - prev_entry->end;
1489 prev_entry->end = end;
1490 vm_map_entry_resize_free(map, prev_entry);
1491 vm_map_simplify_entry(map, prev_entry);
1492 return (KERN_SUCCESS);
1496 * If we can extend the object but cannot extend the
1497 * map entry, we have to create a new map entry. We
1498 * must bump the ref count on the extended object to
1499 * account for it. object may be NULL.
1501 object = prev_entry->object.vm_object;
1502 offset = prev_entry->offset +
1503 (prev_entry->end - prev_entry->start);
1504 vm_object_reference(object);
1505 if (cred != NULL && object != NULL && object->cred != NULL &&
1506 !(prev_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
1507 /* Object already accounts for this uid. */
1515 * Create a new entry
1517 new_entry = vm_map_entry_create(map);
1518 new_entry->start = start;
1519 new_entry->end = end;
1520 new_entry->cred = NULL;
1522 new_entry->eflags = protoeflags;
1523 new_entry->object.vm_object = object;
1524 new_entry->offset = offset;
1526 new_entry->inheritance = inheritance;
1527 new_entry->protection = prot;
1528 new_entry->max_protection = max;
1529 new_entry->wired_count = 0;
1530 new_entry->wiring_thread = NULL;
1531 new_entry->read_ahead = VM_FAULT_READ_AHEAD_INIT;
1532 new_entry->next_read = start;
1534 KASSERT(cred == NULL || !ENTRY_CHARGED(new_entry),
1535 ("overcommit: vm_map_insert leaks vm_map %p", new_entry));
1536 new_entry->cred = cred;
1539 * Insert the new entry into the list
1541 vm_map_entry_link(map, new_entry);
1542 if ((new_entry->eflags & MAP_ENTRY_GUARD) == 0)
1543 map->size += new_entry->end - new_entry->start;
1546 * Try to coalesce the new entry with both the previous and next
1547 * entries in the list. Previously, we only attempted to coalesce
1548 * with the previous entry when object is NULL. Here, we handle the
1549 * other cases, which are less common.
1551 vm_map_simplify_entry(map, new_entry);
1553 if ((cow & (MAP_PREFAULT | MAP_PREFAULT_PARTIAL)) != 0) {
1554 vm_map_pmap_enter(map, start, prot, object, OFF_TO_IDX(offset),
1555 end - start, cow & MAP_PREFAULT_PARTIAL);
1558 return (KERN_SUCCESS);
1564 * Find the first fit (lowest VM address) for "length" free bytes
1565 * beginning at address >= start in the given map.
1567 * In a vm_map_entry, "max_free" is the maximum amount of
1568 * contiguous free space between an entry in its subtree and a
1569 * neighbor of that entry. This allows finding a free region in
1570 * one path down the tree, so O(log n) amortized with splay
1573 * The map must be locked, and leaves it so.
1575 * Returns: starting address if sufficient space,
1576 * vm_map_max(map)-length+1 if insufficient space.
1579 vm_map_findspace(vm_map_t map, vm_offset_t start, vm_size_t length)
1581 vm_map_entry_t llist, rlist, root, y;
1582 vm_size_t left_length;
1585 * Request must fit within min/max VM address and must avoid
1588 start = MAX(start, vm_map_min(map));
1589 if (start + length > vm_map_max(map) || start + length < start)
1590 return (vm_map_max(map) - length + 1);
1592 /* Empty tree means wide open address space. */
1593 if (map->root == NULL)
1597 * After splay, if start comes before root node, then there
1598 * must be a gap from start to the root.
1600 root = vm_map_splay_split(start, length, map->root,
1604 else if (rlist != NULL) {
1610 llist = root->right;
1613 map->root = vm_map_splay_merge(root, llist, rlist,
1614 root->left, root->right);
1615 VM_MAP_ASSERT_CONSISTENT(map);
1616 if (start + length <= root->start)
1620 * Root is the last node that might begin its gap before
1621 * start, and this is the last comparison where address
1622 * wrap might be a problem.
1624 if (root->right == NULL &&
1625 start + length <= vm_map_max(map))
1628 /* With max_free, can immediately tell if no solution. */
1629 if (root->right == NULL || length > root->right->max_free)
1630 return (vm_map_max(map) - length + 1);
1633 * Splay for the least large-enough gap in the right subtree.
1637 for (left_length = 0; ;
1638 left_length = root->left != NULL ?
1639 root->left->max_free : root->start - llist->end) {
1640 if (length <= left_length)
1641 SPLAY_LEFT_STEP(root, y, rlist,
1642 length <= (y->left != NULL ?
1643 y->left->max_free : y->start - llist->end));
1645 SPLAY_RIGHT_STEP(root, y, llist,
1646 length > (y->left != NULL ?
1647 y->left->max_free : y->start - root->end));
1652 llist = root->right;
1653 if ((y = rlist) == NULL)
1658 root->right = y->right;
1660 root = vm_map_splay_merge(root, llist, rlist,
1661 root->left, root->right);
1663 y->right = root->right;
1664 vm_map_entry_set_max_free(y);
1666 vm_map_entry_set_max_free(root);
1669 VM_MAP_ASSERT_CONSISTENT(map);
1674 vm_map_fixed(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1675 vm_offset_t start, vm_size_t length, vm_prot_t prot,
1676 vm_prot_t max, int cow)
1681 end = start + length;
1682 KASSERT((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 ||
1684 ("vm_map_fixed: non-NULL backing object for stack"));
1686 VM_MAP_RANGE_CHECK(map, start, end);
1687 if ((cow & MAP_CHECK_EXCL) == 0)
1688 vm_map_delete(map, start, end);
1689 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) != 0) {
1690 result = vm_map_stack_locked(map, start, length, sgrowsiz,
1693 result = vm_map_insert(map, object, offset, start, end,
1700 static const int aslr_pages_rnd_64[2] = {0x1000, 0x10};
1701 static const int aslr_pages_rnd_32[2] = {0x100, 0x4};
1703 static int cluster_anon = 1;
1704 SYSCTL_INT(_vm, OID_AUTO, cluster_anon, CTLFLAG_RW,
1706 "Cluster anonymous mappings: 0 = no, 1 = yes if no hint, 2 = always");
1709 clustering_anon_allowed(vm_offset_t addr)
1712 switch (cluster_anon) {
1723 static long aslr_restarts;
1724 SYSCTL_LONG(_vm, OID_AUTO, aslr_restarts, CTLFLAG_RD,
1726 "Number of aslr failures");
1728 #define MAP_32BIT_MAX_ADDR ((vm_offset_t)1 << 31)
1731 * Searches for the specified amount of free space in the given map with the
1732 * specified alignment. Performs an address-ordered, first-fit search from
1733 * the given address "*addr", with an optional upper bound "max_addr". If the
1734 * parameter "alignment" is zero, then the alignment is computed from the
1735 * given (object, offset) pair so as to enable the greatest possible use of
1736 * superpage mappings. Returns KERN_SUCCESS and the address of the free space
1737 * in "*addr" if successful. Otherwise, returns KERN_NO_SPACE.
1739 * The map must be locked. Initially, there must be at least "length" bytes
1740 * of free space at the given address.
1743 vm_map_alignspace(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1744 vm_offset_t *addr, vm_size_t length, vm_offset_t max_addr,
1745 vm_offset_t alignment)
1747 vm_offset_t aligned_addr, free_addr;
1749 VM_MAP_ASSERT_LOCKED(map);
1751 KASSERT(free_addr == vm_map_findspace(map, free_addr, length),
1752 ("caller failed to provide space %d at address %p",
1753 (int)length, (void*)free_addr));
1756 * At the start of every iteration, the free space at address
1757 * "*addr" is at least "length" bytes.
1760 pmap_align_superpage(object, offset, addr, length);
1761 else if ((*addr & (alignment - 1)) != 0) {
1762 *addr &= ~(alignment - 1);
1765 aligned_addr = *addr;
1766 if (aligned_addr == free_addr) {
1768 * Alignment did not change "*addr", so "*addr" must
1769 * still provide sufficient free space.
1771 return (KERN_SUCCESS);
1775 * Test for address wrap on "*addr". A wrapped "*addr" could
1776 * be a valid address, in which case vm_map_findspace() cannot
1777 * be relied upon to fail.
1779 if (aligned_addr < free_addr)
1780 return (KERN_NO_SPACE);
1781 *addr = vm_map_findspace(map, aligned_addr, length);
1782 if (*addr + length > vm_map_max(map) ||
1783 (max_addr != 0 && *addr + length > max_addr))
1784 return (KERN_NO_SPACE);
1786 if (free_addr == aligned_addr) {
1788 * If a successful call to vm_map_findspace() did not
1789 * change "*addr", then "*addr" must still be aligned
1790 * and provide sufficient free space.
1792 return (KERN_SUCCESS);
1798 * vm_map_find finds an unallocated region in the target address
1799 * map with the given length. The search is defined to be
1800 * first-fit from the specified address; the region found is
1801 * returned in the same parameter.
1803 * If object is non-NULL, ref count must be bumped by caller
1804 * prior to making call to account for the new entry.
1807 vm_map_find(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1808 vm_offset_t *addr, /* IN/OUT */
1809 vm_size_t length, vm_offset_t max_addr, int find_space,
1810 vm_prot_t prot, vm_prot_t max, int cow)
1812 vm_offset_t alignment, curr_min_addr, min_addr;
1813 int gap, pidx, rv, try;
1814 bool cluster, en_aslr, update_anon;
1816 KASSERT((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 ||
1818 ("vm_map_find: non-NULL backing object for stack"));
1819 MPASS((cow & MAP_REMAP) == 0 || (find_space == VMFS_NO_SPACE &&
1820 (cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0));
1821 if (find_space == VMFS_OPTIMAL_SPACE && (object == NULL ||
1822 (object->flags & OBJ_COLORED) == 0))
1823 find_space = VMFS_ANY_SPACE;
1824 if (find_space >> 8 != 0) {
1825 KASSERT((find_space & 0xff) == 0, ("bad VMFS flags"));
1826 alignment = (vm_offset_t)1 << (find_space >> 8);
1829 en_aslr = (map->flags & MAP_ASLR) != 0;
1830 update_anon = cluster = clustering_anon_allowed(*addr) &&
1831 (map->flags & MAP_IS_SUB_MAP) == 0 && max_addr == 0 &&
1832 find_space != VMFS_NO_SPACE && object == NULL &&
1833 (cow & (MAP_INHERIT_SHARE | MAP_STACK_GROWS_UP |
1834 MAP_STACK_GROWS_DOWN)) == 0 && prot != PROT_NONE;
1835 curr_min_addr = min_addr = *addr;
1836 if (en_aslr && min_addr == 0 && !cluster &&
1837 find_space != VMFS_NO_SPACE &&
1838 (map->flags & MAP_ASLR_IGNSTART) != 0)
1839 curr_min_addr = min_addr = vm_map_min(map);
1843 curr_min_addr = map->anon_loc;
1844 if (curr_min_addr == 0)
1847 if (find_space != VMFS_NO_SPACE) {
1848 KASSERT(find_space == VMFS_ANY_SPACE ||
1849 find_space == VMFS_OPTIMAL_SPACE ||
1850 find_space == VMFS_SUPER_SPACE ||
1851 alignment != 0, ("unexpected VMFS flag"));
1854 * When creating an anonymous mapping, try clustering
1855 * with an existing anonymous mapping first.
1857 * We make up to two attempts to find address space
1858 * for a given find_space value. The first attempt may
1859 * apply randomization or may cluster with an existing
1860 * anonymous mapping. If this first attempt fails,
1861 * perform a first-fit search of the available address
1864 * If all tries failed, and find_space is
1865 * VMFS_OPTIMAL_SPACE, fallback to VMFS_ANY_SPACE.
1866 * Again enable clustering and randomization.
1873 * Second try: we failed either to find a
1874 * suitable region for randomizing the
1875 * allocation, or to cluster with an existing
1876 * mapping. Retry with free run.
1878 curr_min_addr = (map->flags & MAP_ASLR_IGNSTART) != 0 ?
1879 vm_map_min(map) : min_addr;
1880 atomic_add_long(&aslr_restarts, 1);
1883 if (try == 1 && en_aslr && !cluster) {
1885 * Find space for allocation, including
1886 * gap needed for later randomization.
1888 pidx = MAXPAGESIZES > 1 && pagesizes[1] != 0 &&
1889 (find_space == VMFS_SUPER_SPACE || find_space ==
1890 VMFS_OPTIMAL_SPACE) ? 1 : 0;
1891 gap = vm_map_max(map) > MAP_32BIT_MAX_ADDR &&
1892 (max_addr == 0 || max_addr > MAP_32BIT_MAX_ADDR) ?
1893 aslr_pages_rnd_64[pidx] : aslr_pages_rnd_32[pidx];
1894 *addr = vm_map_findspace(map, curr_min_addr,
1895 length + gap * pagesizes[pidx]);
1896 if (*addr + length + gap * pagesizes[pidx] >
1899 /* And randomize the start address. */
1900 *addr += (arc4random() % gap) * pagesizes[pidx];
1901 if (max_addr != 0 && *addr + length > max_addr)
1904 *addr = vm_map_findspace(map, curr_min_addr, length);
1905 if (*addr + length > vm_map_max(map) ||
1906 (max_addr != 0 && *addr + length > max_addr)) {
1917 if (find_space != VMFS_ANY_SPACE &&
1918 (rv = vm_map_alignspace(map, object, offset, addr, length,
1919 max_addr, alignment)) != KERN_SUCCESS) {
1920 if (find_space == VMFS_OPTIMAL_SPACE) {
1921 find_space = VMFS_ANY_SPACE;
1922 curr_min_addr = min_addr;
1923 cluster = update_anon;
1929 } else if ((cow & MAP_REMAP) != 0) {
1930 if (*addr < vm_map_min(map) ||
1931 *addr + length > vm_map_max(map) ||
1932 *addr + length <= length) {
1933 rv = KERN_INVALID_ADDRESS;
1936 vm_map_delete(map, *addr, *addr + length);
1938 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) != 0) {
1939 rv = vm_map_stack_locked(map, *addr, length, sgrowsiz, prot,
1942 rv = vm_map_insert(map, object, offset, *addr, *addr + length,
1945 if (rv == KERN_SUCCESS && update_anon)
1946 map->anon_loc = *addr + length;
1953 * vm_map_find_min() is a variant of vm_map_find() that takes an
1954 * additional parameter (min_addr) and treats the given address
1955 * (*addr) differently. Specifically, it treats *addr as a hint
1956 * and not as the minimum address where the mapping is created.
1958 * This function works in two phases. First, it tries to
1959 * allocate above the hint. If that fails and the hint is
1960 * greater than min_addr, it performs a second pass, replacing
1961 * the hint with min_addr as the minimum address for the
1965 vm_map_find_min(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1966 vm_offset_t *addr, vm_size_t length, vm_offset_t min_addr,
1967 vm_offset_t max_addr, int find_space, vm_prot_t prot, vm_prot_t max,
1975 rv = vm_map_find(map, object, offset, addr, length, max_addr,
1976 find_space, prot, max, cow);
1977 if (rv == KERN_SUCCESS || min_addr >= hint)
1979 *addr = hint = min_addr;
1984 * A map entry with any of the following flags set must not be merged with
1987 #define MAP_ENTRY_NOMERGE_MASK (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP | \
1988 MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_IS_SUB_MAP | MAP_ENTRY_VN_EXEC)
1991 vm_map_mergeable_neighbors(vm_map_entry_t prev, vm_map_entry_t entry)
1994 KASSERT((prev->eflags & MAP_ENTRY_NOMERGE_MASK) == 0 ||
1995 (entry->eflags & MAP_ENTRY_NOMERGE_MASK) == 0,
1996 ("vm_map_mergeable_neighbors: neither %p nor %p are mergeable",
1998 return (prev->end == entry->start &&
1999 prev->object.vm_object == entry->object.vm_object &&
2000 (prev->object.vm_object == NULL ||
2001 prev->offset + (prev->end - prev->start) == entry->offset) &&
2002 prev->eflags == entry->eflags &&
2003 prev->protection == entry->protection &&
2004 prev->max_protection == entry->max_protection &&
2005 prev->inheritance == entry->inheritance &&
2006 prev->wired_count == entry->wired_count &&
2007 prev->cred == entry->cred);
2011 vm_map_merged_neighbor_dispose(vm_map_t map, vm_map_entry_t entry)
2015 * If the backing object is a vnode object, vm_object_deallocate()
2016 * calls vrele(). However, vrele() does not lock the vnode because
2017 * the vnode has additional references. Thus, the map lock can be
2018 * kept without causing a lock-order reversal with the vnode lock.
2020 * Since we count the number of virtual page mappings in
2021 * object->un_pager.vnp.writemappings, the writemappings value
2022 * should not be adjusted when the entry is disposed of.
2024 if (entry->object.vm_object != NULL)
2025 vm_object_deallocate(entry->object.vm_object);
2026 if (entry->cred != NULL)
2027 crfree(entry->cred);
2028 vm_map_entry_dispose(map, entry);
2032 * vm_map_simplify_entry:
2034 * Simplify the given map entry by merging with either neighbor. This
2035 * routine also has the ability to merge with both neighbors.
2037 * The map must be locked.
2039 * This routine guarantees that the passed entry remains valid (though
2040 * possibly extended). When merging, this routine may delete one or
2044 vm_map_simplify_entry(vm_map_t map, vm_map_entry_t entry)
2046 vm_map_entry_t next, prev;
2048 if ((entry->eflags & MAP_ENTRY_NOMERGE_MASK) != 0)
2051 if (vm_map_mergeable_neighbors(prev, entry)) {
2052 vm_map_entry_unlink(map, prev, UNLINK_MERGE_NEXT);
2053 vm_map_merged_neighbor_dispose(map, prev);
2056 if (vm_map_mergeable_neighbors(entry, next)) {
2057 vm_map_entry_unlink(map, next, UNLINK_MERGE_PREV);
2058 vm_map_merged_neighbor_dispose(map, next);
2063 * vm_map_clip_start: [ internal use only ]
2065 * Asserts that the given entry begins at or after
2066 * the specified address; if necessary,
2067 * it splits the entry into two.
2069 #define vm_map_clip_start(map, entry, startaddr) \
2071 if (startaddr > entry->start) \
2072 _vm_map_clip_start(map, entry, startaddr); \
2076 * This routine is called only when it is known that
2077 * the entry must be split.
2080 _vm_map_clip_start(vm_map_t map, vm_map_entry_t entry, vm_offset_t start)
2082 vm_map_entry_t new_entry;
2084 VM_MAP_ASSERT_LOCKED(map);
2085 KASSERT(entry->end > start && entry->start < start,
2086 ("_vm_map_clip_start: invalid clip of entry %p", entry));
2089 * Split off the front portion -- note that we must insert the new
2090 * entry BEFORE this one, so that this entry has the specified
2093 vm_map_simplify_entry(map, entry);
2096 * If there is no object backing this entry, we might as well create
2097 * one now. If we defer it, an object can get created after the map
2098 * is clipped, and individual objects will be created for the split-up
2099 * map. This is a bit of a hack, but is also about the best place to
2100 * put this improvement.
2102 if (entry->object.vm_object == NULL && !map->system_map &&
2103 (entry->eflags & MAP_ENTRY_GUARD) == 0) {
2105 object = vm_object_allocate(OBJT_DEFAULT,
2106 atop(entry->end - entry->start));
2107 entry->object.vm_object = object;
2109 if (entry->cred != NULL) {
2110 object->cred = entry->cred;
2111 object->charge = entry->end - entry->start;
2114 } else if (entry->object.vm_object != NULL &&
2115 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) &&
2116 entry->cred != NULL) {
2117 VM_OBJECT_WLOCK(entry->object.vm_object);
2118 KASSERT(entry->object.vm_object->cred == NULL,
2119 ("OVERCOMMIT: vm_entry_clip_start: both cred e %p", entry));
2120 entry->object.vm_object->cred = entry->cred;
2121 entry->object.vm_object->charge = entry->end - entry->start;
2122 VM_OBJECT_WUNLOCK(entry->object.vm_object);
2126 new_entry = vm_map_entry_create(map);
2127 *new_entry = *entry;
2129 new_entry->end = start;
2130 entry->offset += (start - entry->start);
2131 entry->start = start;
2132 if (new_entry->cred != NULL)
2133 crhold(entry->cred);
2135 vm_map_entry_link(map, new_entry);
2137 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
2138 vm_object_reference(new_entry->object.vm_object);
2139 vm_map_entry_set_vnode_text(new_entry, true);
2141 * The object->un_pager.vnp.writemappings for the
2142 * object of MAP_ENTRY_VN_WRITECNT type entry shall be
2143 * kept as is here. The virtual pages are
2144 * re-distributed among the clipped entries, so the sum is
2151 * vm_map_clip_end: [ internal use only ]
2153 * Asserts that the given entry ends at or before
2154 * the specified address; if necessary,
2155 * it splits the entry into two.
2157 #define vm_map_clip_end(map, entry, endaddr) \
2159 if ((endaddr) < (entry->end)) \
2160 _vm_map_clip_end((map), (entry), (endaddr)); \
2164 * This routine is called only when it is known that
2165 * the entry must be split.
2168 _vm_map_clip_end(vm_map_t map, vm_map_entry_t entry, vm_offset_t end)
2170 vm_map_entry_t new_entry;
2172 VM_MAP_ASSERT_LOCKED(map);
2173 KASSERT(entry->start < end && entry->end > end,
2174 ("_vm_map_clip_end: invalid clip of entry %p", entry));
2177 * If there is no object backing this entry, we might as well create
2178 * one now. If we defer it, an object can get created after the map
2179 * is clipped, and individual objects will be created for the split-up
2180 * map. This is a bit of a hack, but is also about the best place to
2181 * put this improvement.
2183 if (entry->object.vm_object == NULL && !map->system_map &&
2184 (entry->eflags & MAP_ENTRY_GUARD) == 0) {
2186 object = vm_object_allocate(OBJT_DEFAULT,
2187 atop(entry->end - entry->start));
2188 entry->object.vm_object = object;
2190 if (entry->cred != NULL) {
2191 object->cred = entry->cred;
2192 object->charge = entry->end - entry->start;
2195 } else if (entry->object.vm_object != NULL &&
2196 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) &&
2197 entry->cred != NULL) {
2198 VM_OBJECT_WLOCK(entry->object.vm_object);
2199 KASSERT(entry->object.vm_object->cred == NULL,
2200 ("OVERCOMMIT: vm_entry_clip_end: both cred e %p", entry));
2201 entry->object.vm_object->cred = entry->cred;
2202 entry->object.vm_object->charge = entry->end - entry->start;
2203 VM_OBJECT_WUNLOCK(entry->object.vm_object);
2208 * Create a new entry and insert it AFTER the specified entry
2210 new_entry = vm_map_entry_create(map);
2211 *new_entry = *entry;
2213 new_entry->start = entry->end = end;
2214 new_entry->offset += (end - entry->start);
2215 if (new_entry->cred != NULL)
2216 crhold(entry->cred);
2218 vm_map_entry_link(map, new_entry);
2220 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
2221 vm_object_reference(new_entry->object.vm_object);
2222 vm_map_entry_set_vnode_text(new_entry, true);
2227 * vm_map_submap: [ kernel use only ]
2229 * Mark the given range as handled by a subordinate map.
2231 * This range must have been created with vm_map_find,
2232 * and no other operations may have been performed on this
2233 * range prior to calling vm_map_submap.
2235 * Only a limited number of operations can be performed
2236 * within this rage after calling vm_map_submap:
2238 * [Don't try vm_map_copy!]
2240 * To remove a submapping, one must first remove the
2241 * range from the superior map, and then destroy the
2242 * submap (if desired). [Better yet, don't try it.]
2251 vm_map_entry_t entry;
2254 result = KERN_INVALID_ARGUMENT;
2256 vm_map_lock(submap);
2257 submap->flags |= MAP_IS_SUB_MAP;
2258 vm_map_unlock(submap);
2262 VM_MAP_RANGE_CHECK(map, start, end);
2264 if (vm_map_lookup_entry(map, start, &entry)) {
2265 vm_map_clip_start(map, entry, start);
2267 entry = entry->next;
2269 vm_map_clip_end(map, entry, end);
2271 if ((entry->start == start) && (entry->end == end) &&
2272 ((entry->eflags & MAP_ENTRY_COW) == 0) &&
2273 (entry->object.vm_object == NULL)) {
2274 entry->object.sub_map = submap;
2275 entry->eflags |= MAP_ENTRY_IS_SUB_MAP;
2276 result = KERN_SUCCESS;
2280 if (result != KERN_SUCCESS) {
2281 vm_map_lock(submap);
2282 submap->flags &= ~MAP_IS_SUB_MAP;
2283 vm_map_unlock(submap);
2289 * The maximum number of pages to map if MAP_PREFAULT_PARTIAL is specified
2291 #define MAX_INIT_PT 96
2294 * vm_map_pmap_enter:
2296 * Preload the specified map's pmap with mappings to the specified
2297 * object's memory-resident pages. No further physical pages are
2298 * allocated, and no further virtual pages are retrieved from secondary
2299 * storage. If the specified flags include MAP_PREFAULT_PARTIAL, then a
2300 * limited number of page mappings are created at the low-end of the
2301 * specified address range. (For this purpose, a superpage mapping
2302 * counts as one page mapping.) Otherwise, all resident pages within
2303 * the specified address range are mapped.
2306 vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot,
2307 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags)
2310 vm_page_t p, p_start;
2311 vm_pindex_t mask, psize, threshold, tmpidx;
2313 if ((prot & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0 || object == NULL)
2315 VM_OBJECT_RLOCK(object);
2316 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) {
2317 VM_OBJECT_RUNLOCK(object);
2318 VM_OBJECT_WLOCK(object);
2319 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) {
2320 pmap_object_init_pt(map->pmap, addr, object, pindex,
2322 VM_OBJECT_WUNLOCK(object);
2325 VM_OBJECT_LOCK_DOWNGRADE(object);
2329 if (psize + pindex > object->size) {
2330 if (object->size < pindex) {
2331 VM_OBJECT_RUNLOCK(object);
2334 psize = object->size - pindex;
2339 threshold = MAX_INIT_PT;
2341 p = vm_page_find_least(object, pindex);
2343 * Assert: the variable p is either (1) the page with the
2344 * least pindex greater than or equal to the parameter pindex
2348 p != NULL && (tmpidx = p->pindex - pindex) < psize;
2349 p = TAILQ_NEXT(p, listq)) {
2351 * don't allow an madvise to blow away our really
2352 * free pages allocating pv entries.
2354 if (((flags & MAP_PREFAULT_MADVISE) != 0 &&
2355 vm_page_count_severe()) ||
2356 ((flags & MAP_PREFAULT_PARTIAL) != 0 &&
2357 tmpidx >= threshold)) {
2361 if (p->valid == VM_PAGE_BITS_ALL) {
2362 if (p_start == NULL) {
2363 start = addr + ptoa(tmpidx);
2366 /* Jump ahead if a superpage mapping is possible. */
2367 if (p->psind > 0 && ((addr + ptoa(tmpidx)) &
2368 (pagesizes[p->psind] - 1)) == 0) {
2369 mask = atop(pagesizes[p->psind]) - 1;
2370 if (tmpidx + mask < psize &&
2371 vm_page_ps_test(p, PS_ALL_VALID, NULL)) {
2376 } else if (p_start != NULL) {
2377 pmap_enter_object(map->pmap, start, addr +
2378 ptoa(tmpidx), p_start, prot);
2382 if (p_start != NULL)
2383 pmap_enter_object(map->pmap, start, addr + ptoa(psize),
2385 VM_OBJECT_RUNLOCK(object);
2391 * Sets the protection of the specified address
2392 * region in the target map. If "set_max" is
2393 * specified, the maximum protection is to be set;
2394 * otherwise, only the current protection is affected.
2397 vm_map_protect(vm_map_t map, vm_offset_t start, vm_offset_t end,
2398 vm_prot_t new_prot, boolean_t set_max)
2400 vm_map_entry_t current, entry, in_tran;
2406 return (KERN_SUCCESS);
2413 * Ensure that we are not concurrently wiring pages. vm_map_wire() may
2414 * need to fault pages into the map and will drop the map lock while
2415 * doing so, and the VM object may end up in an inconsistent state if we
2416 * update the protection on the map entry in between faults.
2418 vm_map_wait_busy(map);
2420 VM_MAP_RANGE_CHECK(map, start, end);
2422 if (vm_map_lookup_entry(map, start, &entry)) {
2423 vm_map_clip_start(map, entry, start);
2425 entry = entry->next;
2429 * Make a first pass to check for protection violations.
2431 for (current = entry; current->start < end; current = current->next) {
2432 if ((current->eflags & MAP_ENTRY_GUARD) != 0)
2434 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
2436 return (KERN_INVALID_ARGUMENT);
2438 if ((new_prot & current->max_protection) != new_prot) {
2440 return (KERN_PROTECTION_FAILURE);
2442 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0)
2447 * Postpone the operation until all in transition map entries
2448 * are stabilized. In-transition entry might already have its
2449 * pages wired and wired_count incremented, but
2450 * MAP_ENTRY_USER_WIRED flag not yet set, and visible to other
2451 * threads because the map lock is dropped. In this case we
2452 * would miss our call to vm_fault_copy_entry().
2454 if (in_tran != NULL) {
2455 in_tran->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2456 vm_map_unlock_and_wait(map, 0);
2461 * Do an accounting pass for private read-only mappings that
2462 * now will do cow due to allowed write (e.g. debugger sets
2463 * breakpoint on text segment)
2465 for (current = entry; current->start < end; current = current->next) {
2467 vm_map_clip_end(map, current, end);
2470 ((new_prot & ~(current->protection)) & VM_PROT_WRITE) == 0 ||
2471 ENTRY_CHARGED(current) ||
2472 (current->eflags & MAP_ENTRY_GUARD) != 0) {
2476 cred = curthread->td_ucred;
2477 obj = current->object.vm_object;
2479 if (obj == NULL || (current->eflags & MAP_ENTRY_NEEDS_COPY)) {
2480 if (!swap_reserve(current->end - current->start)) {
2482 return (KERN_RESOURCE_SHORTAGE);
2485 current->cred = cred;
2489 VM_OBJECT_WLOCK(obj);
2490 if (obj->type != OBJT_DEFAULT && obj->type != OBJT_SWAP) {
2491 VM_OBJECT_WUNLOCK(obj);
2496 * Charge for the whole object allocation now, since
2497 * we cannot distinguish between non-charged and
2498 * charged clipped mapping of the same object later.
2500 KASSERT(obj->charge == 0,
2501 ("vm_map_protect: object %p overcharged (entry %p)",
2503 if (!swap_reserve(ptoa(obj->size))) {
2504 VM_OBJECT_WUNLOCK(obj);
2506 return (KERN_RESOURCE_SHORTAGE);
2511 obj->charge = ptoa(obj->size);
2512 VM_OBJECT_WUNLOCK(obj);
2516 * Go back and fix up protections. [Note that clipping is not
2517 * necessary the second time.]
2519 for (current = entry; current->start < end; current = current->next) {
2520 if ((current->eflags & MAP_ENTRY_GUARD) != 0)
2523 old_prot = current->protection;
2526 current->protection =
2527 (current->max_protection = new_prot) &
2530 current->protection = new_prot;
2533 * For user wired map entries, the normal lazy evaluation of
2534 * write access upgrades through soft page faults is
2535 * undesirable. Instead, immediately copy any pages that are
2536 * copy-on-write and enable write access in the physical map.
2538 if ((current->eflags & MAP_ENTRY_USER_WIRED) != 0 &&
2539 (current->protection & VM_PROT_WRITE) != 0 &&
2540 (old_prot & VM_PROT_WRITE) == 0)
2541 vm_fault_copy_entry(map, map, current, current, NULL);
2544 * When restricting access, update the physical map. Worry
2545 * about copy-on-write here.
2547 if ((old_prot & ~current->protection) != 0) {
2548 #define MASK(entry) (((entry)->eflags & MAP_ENTRY_COW) ? ~VM_PROT_WRITE : \
2550 pmap_protect(map->pmap, current->start,
2552 current->protection & MASK(current));
2555 vm_map_simplify_entry(map, current);
2558 return (KERN_SUCCESS);
2564 * This routine traverses a processes map handling the madvise
2565 * system call. Advisories are classified as either those effecting
2566 * the vm_map_entry structure, or those effecting the underlying
2576 vm_map_entry_t current, entry;
2580 * Some madvise calls directly modify the vm_map_entry, in which case
2581 * we need to use an exclusive lock on the map and we need to perform
2582 * various clipping operations. Otherwise we only need a read-lock
2587 case MADV_SEQUENTIAL:
2604 vm_map_lock_read(map);
2611 * Locate starting entry and clip if necessary.
2613 VM_MAP_RANGE_CHECK(map, start, end);
2615 if (vm_map_lookup_entry(map, start, &entry)) {
2617 vm_map_clip_start(map, entry, start);
2619 entry = entry->next;
2624 * madvise behaviors that are implemented in the vm_map_entry.
2626 * We clip the vm_map_entry so that behavioral changes are
2627 * limited to the specified address range.
2629 for (current = entry; current->start < end;
2630 current = current->next) {
2631 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
2634 vm_map_clip_end(map, current, end);
2638 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_NORMAL);
2640 case MADV_SEQUENTIAL:
2641 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_SEQUENTIAL);
2644 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_RANDOM);
2647 current->eflags |= MAP_ENTRY_NOSYNC;
2650 current->eflags &= ~MAP_ENTRY_NOSYNC;
2653 current->eflags |= MAP_ENTRY_NOCOREDUMP;
2656 current->eflags &= ~MAP_ENTRY_NOCOREDUMP;
2661 vm_map_simplify_entry(map, current);
2665 vm_pindex_t pstart, pend;
2668 * madvise behaviors that are implemented in the underlying
2671 * Since we don't clip the vm_map_entry, we have to clip
2672 * the vm_object pindex and count.
2674 for (current = entry; current->start < end;
2675 current = current->next) {
2676 vm_offset_t useEnd, useStart;
2678 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
2681 pstart = OFF_TO_IDX(current->offset);
2682 pend = pstart + atop(current->end - current->start);
2683 useStart = current->start;
2684 useEnd = current->end;
2686 if (current->start < start) {
2687 pstart += atop(start - current->start);
2690 if (current->end > end) {
2691 pend -= atop(current->end - end);
2699 * Perform the pmap_advise() before clearing
2700 * PGA_REFERENCED in vm_page_advise(). Otherwise, a
2701 * concurrent pmap operation, such as pmap_remove(),
2702 * could clear a reference in the pmap and set
2703 * PGA_REFERENCED on the page before the pmap_advise()
2704 * had completed. Consequently, the page would appear
2705 * referenced based upon an old reference that
2706 * occurred before this pmap_advise() ran.
2708 if (behav == MADV_DONTNEED || behav == MADV_FREE)
2709 pmap_advise(map->pmap, useStart, useEnd,
2712 vm_object_madvise(current->object.vm_object, pstart,
2716 * Pre-populate paging structures in the
2717 * WILLNEED case. For wired entries, the
2718 * paging structures are already populated.
2720 if (behav == MADV_WILLNEED &&
2721 current->wired_count == 0) {
2722 vm_map_pmap_enter(map,
2724 current->protection,
2725 current->object.vm_object,
2727 ptoa(pend - pstart),
2728 MAP_PREFAULT_MADVISE
2732 vm_map_unlock_read(map);
2741 * Sets the inheritance of the specified address
2742 * range in the target map. Inheritance
2743 * affects how the map will be shared with
2744 * child maps at the time of vmspace_fork.
2747 vm_map_inherit(vm_map_t map, vm_offset_t start, vm_offset_t end,
2748 vm_inherit_t new_inheritance)
2750 vm_map_entry_t entry;
2751 vm_map_entry_t temp_entry;
2753 switch (new_inheritance) {
2754 case VM_INHERIT_NONE:
2755 case VM_INHERIT_COPY:
2756 case VM_INHERIT_SHARE:
2757 case VM_INHERIT_ZERO:
2760 return (KERN_INVALID_ARGUMENT);
2763 return (KERN_SUCCESS);
2765 VM_MAP_RANGE_CHECK(map, start, end);
2766 if (vm_map_lookup_entry(map, start, &temp_entry)) {
2768 vm_map_clip_start(map, entry, start);
2770 entry = temp_entry->next;
2771 while (entry->start < end) {
2772 vm_map_clip_end(map, entry, end);
2773 if ((entry->eflags & MAP_ENTRY_GUARD) == 0 ||
2774 new_inheritance != VM_INHERIT_ZERO)
2775 entry->inheritance = new_inheritance;
2776 vm_map_simplify_entry(map, entry);
2777 entry = entry->next;
2780 return (KERN_SUCCESS);
2786 * Implements both kernel and user unwiring.
2789 vm_map_unwire(vm_map_t map, vm_offset_t start, vm_offset_t end,
2792 vm_map_entry_t entry, first_entry, tmp_entry;
2793 vm_offset_t saved_start;
2794 unsigned int last_timestamp;
2796 boolean_t need_wakeup, result, user_unwire;
2799 return (KERN_SUCCESS);
2800 user_unwire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE;
2802 VM_MAP_RANGE_CHECK(map, start, end);
2803 if (!vm_map_lookup_entry(map, start, &first_entry)) {
2804 if (flags & VM_MAP_WIRE_HOLESOK)
2805 first_entry = first_entry->next;
2808 return (KERN_INVALID_ADDRESS);
2811 last_timestamp = map->timestamp;
2812 entry = first_entry;
2813 while (entry->start < end) {
2814 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
2816 * We have not yet clipped the entry.
2818 saved_start = (start >= entry->start) ? start :
2820 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2821 if (vm_map_unlock_and_wait(map, 0)) {
2823 * Allow interruption of user unwiring?
2827 if (last_timestamp+1 != map->timestamp) {
2829 * Look again for the entry because the map was
2830 * modified while it was unlocked.
2831 * Specifically, the entry may have been
2832 * clipped, merged, or deleted.
2834 if (!vm_map_lookup_entry(map, saved_start,
2836 if (flags & VM_MAP_WIRE_HOLESOK)
2837 tmp_entry = tmp_entry->next;
2839 if (saved_start == start) {
2841 * First_entry has been deleted.
2844 return (KERN_INVALID_ADDRESS);
2847 rv = KERN_INVALID_ADDRESS;
2851 if (entry == first_entry)
2852 first_entry = tmp_entry;
2857 last_timestamp = map->timestamp;
2860 vm_map_clip_start(map, entry, start);
2861 vm_map_clip_end(map, entry, end);
2863 * Mark the entry in case the map lock is released. (See
2866 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 &&
2867 entry->wiring_thread == NULL,
2868 ("owned map entry %p", entry));
2869 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
2870 entry->wiring_thread = curthread;
2872 * Check the map for holes in the specified region.
2873 * If VM_MAP_WIRE_HOLESOK was specified, skip this check.
2875 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) &&
2876 (entry->end < end && entry->next->start > entry->end)) {
2878 rv = KERN_INVALID_ADDRESS;
2882 * If system unwiring, require that the entry is system wired.
2885 vm_map_entry_system_wired_count(entry) == 0) {
2887 rv = KERN_INVALID_ARGUMENT;
2890 entry = entry->next;
2894 need_wakeup = FALSE;
2895 if (first_entry == NULL) {
2896 result = vm_map_lookup_entry(map, start, &first_entry);
2897 if (!result && (flags & VM_MAP_WIRE_HOLESOK))
2898 first_entry = first_entry->next;
2900 KASSERT(result, ("vm_map_unwire: lookup failed"));
2902 for (entry = first_entry; entry->start < end; entry = entry->next) {
2904 * If VM_MAP_WIRE_HOLESOK was specified, an empty
2905 * space in the unwired region could have been mapped
2906 * while the map lock was dropped for draining
2907 * MAP_ENTRY_IN_TRANSITION. Moreover, another thread
2908 * could be simultaneously wiring this new mapping
2909 * entry. Detect these cases and skip any entries
2910 * marked as in transition by us.
2912 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 ||
2913 entry->wiring_thread != curthread) {
2914 KASSERT((flags & VM_MAP_WIRE_HOLESOK) != 0,
2915 ("vm_map_unwire: !HOLESOK and new/changed entry"));
2919 if (rv == KERN_SUCCESS && (!user_unwire ||
2920 (entry->eflags & MAP_ENTRY_USER_WIRED))) {
2921 if (entry->wired_count == 1)
2922 vm_map_entry_unwire(map, entry);
2924 entry->wired_count--;
2926 entry->eflags &= ~MAP_ENTRY_USER_WIRED;
2928 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0,
2929 ("vm_map_unwire: in-transition flag missing %p", entry));
2930 KASSERT(entry->wiring_thread == curthread,
2931 ("vm_map_unwire: alien wire %p", entry));
2932 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION;
2933 entry->wiring_thread = NULL;
2934 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
2935 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
2938 vm_map_simplify_entry(map, entry);
2947 vm_map_wire_user_count_sub(u_long npages)
2950 atomic_subtract_long(&vm_user_wire_count, npages);
2954 vm_map_wire_user_count_add(u_long npages)
2958 wired = vm_user_wire_count;
2960 if (npages + wired > vm_page_max_user_wired)
2962 } while (!atomic_fcmpset_long(&vm_user_wire_count, &wired,
2969 * vm_map_wire_entry_failure:
2971 * Handle a wiring failure on the given entry.
2973 * The map should be locked.
2976 vm_map_wire_entry_failure(vm_map_t map, vm_map_entry_t entry,
2977 vm_offset_t failed_addr)
2980 VM_MAP_ASSERT_LOCKED(map);
2981 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 &&
2982 entry->wired_count == 1,
2983 ("vm_map_wire_entry_failure: entry %p isn't being wired", entry));
2984 KASSERT(failed_addr < entry->end,
2985 ("vm_map_wire_entry_failure: entry %p was fully wired", entry));
2988 * If any pages at the start of this entry were successfully wired,
2991 if (failed_addr > entry->start) {
2992 pmap_unwire(map->pmap, entry->start, failed_addr);
2993 vm_object_unwire(entry->object.vm_object, entry->offset,
2994 failed_addr - entry->start, PQ_ACTIVE);
2998 * Assign an out-of-range value to represent the failure to wire this
3001 entry->wired_count = -1;
3005 vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t end, int flags)
3010 rv = vm_map_wire_locked(map, start, end, flags);
3017 * vm_map_wire_locked:
3019 * Implements both kernel and user wiring. Returns with the map locked,
3020 * the map lock may be dropped.
3023 vm_map_wire_locked(vm_map_t map, vm_offset_t start, vm_offset_t end, int flags)
3025 vm_map_entry_t entry, first_entry, tmp_entry;
3026 vm_offset_t faddr, saved_end, saved_start;
3028 u_int last_timestamp;
3030 boolean_t need_wakeup, result, user_wire;
3033 VM_MAP_ASSERT_LOCKED(map);
3036 return (KERN_SUCCESS);
3038 if (flags & VM_MAP_WIRE_WRITE)
3039 prot |= VM_PROT_WRITE;
3040 user_wire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE;
3041 VM_MAP_RANGE_CHECK(map, start, end);
3042 if (!vm_map_lookup_entry(map, start, &first_entry)) {
3043 if (flags & VM_MAP_WIRE_HOLESOK)
3044 first_entry = first_entry->next;
3046 return (KERN_INVALID_ADDRESS);
3048 last_timestamp = map->timestamp;
3049 entry = first_entry;
3050 while (entry->start < end) {
3051 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
3053 * We have not yet clipped the entry.
3055 saved_start = (start >= entry->start) ? start :
3057 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
3058 if (vm_map_unlock_and_wait(map, 0)) {
3060 * Allow interruption of user wiring?
3064 if (last_timestamp + 1 != map->timestamp) {
3066 * Look again for the entry because the map was
3067 * modified while it was unlocked.
3068 * Specifically, the entry may have been
3069 * clipped, merged, or deleted.
3071 if (!vm_map_lookup_entry(map, saved_start,
3073 if (flags & VM_MAP_WIRE_HOLESOK)
3074 tmp_entry = tmp_entry->next;
3076 if (saved_start == start) {
3078 * first_entry has been deleted.
3080 return (KERN_INVALID_ADDRESS);
3083 rv = KERN_INVALID_ADDRESS;
3087 if (entry == first_entry)
3088 first_entry = tmp_entry;
3093 last_timestamp = map->timestamp;
3096 vm_map_clip_start(map, entry, start);
3097 vm_map_clip_end(map, entry, end);
3099 * Mark the entry in case the map lock is released. (See
3102 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 &&
3103 entry->wiring_thread == NULL,
3104 ("owned map entry %p", entry));
3105 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
3106 entry->wiring_thread = curthread;
3107 if ((entry->protection & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0
3108 || (entry->protection & prot) != prot) {
3109 entry->eflags |= MAP_ENTRY_WIRE_SKIPPED;
3110 if ((flags & VM_MAP_WIRE_HOLESOK) == 0) {
3112 rv = KERN_INVALID_ADDRESS;
3117 if (entry->wired_count == 0) {
3118 entry->wired_count++;
3120 npages = atop(entry->end - entry->start);
3121 if (user_wire && !vm_map_wire_user_count_add(npages)) {
3122 vm_map_wire_entry_failure(map, entry,
3125 rv = KERN_RESOURCE_SHORTAGE;
3130 * Release the map lock, relying on the in-transition
3131 * mark. Mark the map busy for fork.
3133 saved_start = entry->start;
3134 saved_end = entry->end;
3138 faddr = saved_start;
3141 * Simulate a fault to get the page and enter
3142 * it into the physical map.
3144 if ((rv = vm_fault(map, faddr, VM_PROT_NONE,
3145 VM_FAULT_WIRE)) != KERN_SUCCESS)
3147 } while ((faddr += PAGE_SIZE) < saved_end);
3150 if (last_timestamp + 1 != map->timestamp) {
3152 * Look again for the entry because the map was
3153 * modified while it was unlocked. The entry
3154 * may have been clipped, but NOT merged or
3157 result = vm_map_lookup_entry(map, saved_start,
3159 KASSERT(result, ("vm_map_wire: lookup failed"));
3160 if (entry == first_entry)
3161 first_entry = tmp_entry;
3165 while (entry->end < saved_end) {
3167 * In case of failure, handle entries
3168 * that were not fully wired here;
3169 * fully wired entries are handled
3172 if (rv != KERN_SUCCESS &&
3174 vm_map_wire_entry_failure(map,
3176 entry = entry->next;
3179 last_timestamp = map->timestamp;
3180 if (rv != KERN_SUCCESS) {
3181 vm_map_wire_entry_failure(map, entry, faddr);
3183 vm_map_wire_user_count_sub(npages);
3187 } else if (!user_wire ||
3188 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
3189 entry->wired_count++;
3192 * Check the map for holes in the specified region.
3193 * If VM_MAP_WIRE_HOLESOK was specified, skip this check.
3196 if ((flags & VM_MAP_WIRE_HOLESOK) == 0 &&
3197 entry->end < end && entry->next->start > entry->end) {
3199 rv = KERN_INVALID_ADDRESS;
3202 entry = entry->next;
3206 need_wakeup = FALSE;
3207 if (first_entry == NULL) {
3208 result = vm_map_lookup_entry(map, start, &first_entry);
3209 if (!result && (flags & VM_MAP_WIRE_HOLESOK))
3210 first_entry = first_entry->next;
3212 KASSERT(result, ("vm_map_wire: lookup failed"));
3214 for (entry = first_entry; entry->start < end; entry = entry->next) {
3216 * If VM_MAP_WIRE_HOLESOK was specified, an empty
3217 * space in the unwired region could have been mapped
3218 * while the map lock was dropped for faulting in the
3219 * pages or draining MAP_ENTRY_IN_TRANSITION.
3220 * Moreover, another thread could be simultaneously
3221 * wiring this new mapping entry. Detect these cases
3222 * and skip any entries marked as in transition not by us.
3224 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 ||
3225 entry->wiring_thread != curthread) {
3226 KASSERT((flags & VM_MAP_WIRE_HOLESOK) != 0,
3227 ("vm_map_wire: !HOLESOK and new/changed entry"));
3231 if ((entry->eflags & MAP_ENTRY_WIRE_SKIPPED) != 0)
3232 goto next_entry_done;
3234 if (rv == KERN_SUCCESS) {
3236 entry->eflags |= MAP_ENTRY_USER_WIRED;
3237 } else if (entry->wired_count == -1) {
3239 * Wiring failed on this entry. Thus, unwiring is
3242 entry->wired_count = 0;
3243 } else if (!user_wire ||
3244 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
3246 * Undo the wiring. Wiring succeeded on this entry
3247 * but failed on a later entry.
3249 if (entry->wired_count == 1) {
3250 vm_map_entry_unwire(map, entry);
3252 vm_map_wire_user_count_sub(
3253 atop(entry->end - entry->start));
3255 entry->wired_count--;
3258 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0,
3259 ("vm_map_wire: in-transition flag missing %p", entry));
3260 KASSERT(entry->wiring_thread == curthread,
3261 ("vm_map_wire: alien wire %p", entry));
3262 entry->eflags &= ~(MAP_ENTRY_IN_TRANSITION |
3263 MAP_ENTRY_WIRE_SKIPPED);
3264 entry->wiring_thread = NULL;
3265 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
3266 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
3269 vm_map_simplify_entry(map, entry);
3279 * Push any dirty cached pages in the address range to their pager.
3280 * If syncio is TRUE, dirty pages are written synchronously.
3281 * If invalidate is TRUE, any cached pages are freed as well.
3283 * If the size of the region from start to end is zero, we are
3284 * supposed to flush all modified pages within the region containing
3285 * start. Unfortunately, a region can be split or coalesced with
3286 * neighboring regions, making it difficult to determine what the
3287 * original region was. Therefore, we approximate this requirement by
3288 * flushing the current region containing start.
3290 * Returns an error if any part of the specified range is not mapped.
3298 boolean_t invalidate)
3300 vm_map_entry_t current;
3301 vm_map_entry_t entry;
3304 vm_ooffset_t offset;
3305 unsigned int last_timestamp;
3308 vm_map_lock_read(map);
3309 VM_MAP_RANGE_CHECK(map, start, end);
3310 if (!vm_map_lookup_entry(map, start, &entry)) {
3311 vm_map_unlock_read(map);
3312 return (KERN_INVALID_ADDRESS);
3313 } else if (start == end) {
3314 start = entry->start;
3318 * Make a first pass to check for user-wired memory and holes.
3320 for (current = entry; current->start < end; current = current->next) {
3321 if (invalidate && (current->eflags & MAP_ENTRY_USER_WIRED)) {
3322 vm_map_unlock_read(map);
3323 return (KERN_INVALID_ARGUMENT);
3325 if (end > current->end &&
3326 current->end != current->next->start) {
3327 vm_map_unlock_read(map);
3328 return (KERN_INVALID_ADDRESS);
3333 pmap_remove(map->pmap, start, end);
3337 * Make a second pass, cleaning/uncaching pages from the indicated
3340 for (current = entry; current->start < end;) {
3341 offset = current->offset + (start - current->start);
3342 size = (end <= current->end ? end : current->end) - start;
3343 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
3345 vm_map_entry_t tentry;
3348 smap = current->object.sub_map;
3349 vm_map_lock_read(smap);
3350 (void) vm_map_lookup_entry(smap, offset, &tentry);
3351 tsize = tentry->end - offset;
3354 object = tentry->object.vm_object;
3355 offset = tentry->offset + (offset - tentry->start);
3356 vm_map_unlock_read(smap);
3358 object = current->object.vm_object;
3360 vm_object_reference(object);
3361 last_timestamp = map->timestamp;
3362 vm_map_unlock_read(map);
3363 if (!vm_object_sync(object, offset, size, syncio, invalidate))
3366 vm_object_deallocate(object);
3367 vm_map_lock_read(map);
3368 if (last_timestamp == map->timestamp ||
3369 !vm_map_lookup_entry(map, start, ¤t))
3370 current = current->next;
3373 vm_map_unlock_read(map);
3374 return (failed ? KERN_FAILURE : KERN_SUCCESS);
3378 * vm_map_entry_unwire: [ internal use only ]
3380 * Make the region specified by this entry pageable.
3382 * The map in question should be locked.
3383 * [This is the reason for this routine's existence.]
3386 vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry)
3390 VM_MAP_ASSERT_LOCKED(map);
3391 KASSERT(entry->wired_count > 0,
3392 ("vm_map_entry_unwire: entry %p isn't wired", entry));
3394 size = entry->end - entry->start;
3395 if ((entry->eflags & MAP_ENTRY_USER_WIRED) != 0)
3396 vm_map_wire_user_count_sub(atop(size));
3397 pmap_unwire(map->pmap, entry->start, entry->end);
3398 vm_object_unwire(entry->object.vm_object, entry->offset, size,
3400 entry->wired_count = 0;
3404 vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map)
3407 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0)
3408 vm_object_deallocate(entry->object.vm_object);
3409 uma_zfree(system_map ? kmapentzone : mapentzone, entry);
3413 * vm_map_entry_delete: [ internal use only ]
3415 * Deallocate the given entry from the target map.
3418 vm_map_entry_delete(vm_map_t map, vm_map_entry_t entry)
3421 vm_pindex_t offidxstart, offidxend, count, size1;
3424 vm_map_entry_unlink(map, entry, UNLINK_MERGE_NONE);
3425 object = entry->object.vm_object;
3427 if ((entry->eflags & MAP_ENTRY_GUARD) != 0) {
3428 MPASS(entry->cred == NULL);
3429 MPASS((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0);
3430 MPASS(object == NULL);
3431 vm_map_entry_deallocate(entry, map->system_map);
3435 size = entry->end - entry->start;
3438 if (entry->cred != NULL) {
3439 swap_release_by_cred(size, entry->cred);
3440 crfree(entry->cred);
3443 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0 &&
3445 KASSERT(entry->cred == NULL || object->cred == NULL ||
3446 (entry->eflags & MAP_ENTRY_NEEDS_COPY),
3447 ("OVERCOMMIT vm_map_entry_delete: both cred %p", entry));
3449 offidxstart = OFF_TO_IDX(entry->offset);
3450 offidxend = offidxstart + count;
3451 VM_OBJECT_WLOCK(object);
3452 if (object->ref_count != 1 && ((object->flags & (OBJ_NOSPLIT |
3453 OBJ_ONEMAPPING)) == OBJ_ONEMAPPING ||
3454 object == kernel_object)) {
3455 vm_object_collapse(object);
3458 * The option OBJPR_NOTMAPPED can be passed here
3459 * because vm_map_delete() already performed
3460 * pmap_remove() on the only mapping to this range
3463 vm_object_page_remove(object, offidxstart, offidxend,
3465 if (object->type == OBJT_SWAP)
3466 swap_pager_freespace(object, offidxstart,
3468 if (offidxend >= object->size &&
3469 offidxstart < object->size) {
3470 size1 = object->size;
3471 object->size = offidxstart;
3472 if (object->cred != NULL) {
3473 size1 -= object->size;
3474 KASSERT(object->charge >= ptoa(size1),
3475 ("object %p charge < 0", object));
3476 swap_release_by_cred(ptoa(size1),
3478 object->charge -= ptoa(size1);
3482 VM_OBJECT_WUNLOCK(object);
3484 entry->object.vm_object = NULL;
3485 if (map->system_map)
3486 vm_map_entry_deallocate(entry, TRUE);
3488 entry->next = curthread->td_map_def_user;
3489 curthread->td_map_def_user = entry;
3494 * vm_map_delete: [ internal use only ]
3496 * Deallocates the given address range from the target
3500 vm_map_delete(vm_map_t map, vm_offset_t start, vm_offset_t end)
3502 vm_map_entry_t entry;
3503 vm_map_entry_t first_entry;
3505 VM_MAP_ASSERT_LOCKED(map);
3507 return (KERN_SUCCESS);
3510 * Find the start of the region, and clip it
3512 if (!vm_map_lookup_entry(map, start, &first_entry))
3513 entry = first_entry->next;
3515 entry = first_entry;
3516 vm_map_clip_start(map, entry, start);
3520 * Step through all entries in this region
3522 while (entry->start < end) {
3523 vm_map_entry_t next;
3526 * Wait for wiring or unwiring of an entry to complete.
3527 * Also wait for any system wirings to disappear on
3530 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 ||
3531 (vm_map_pmap(map) != kernel_pmap &&
3532 vm_map_entry_system_wired_count(entry) != 0)) {
3533 unsigned int last_timestamp;
3534 vm_offset_t saved_start;
3535 vm_map_entry_t tmp_entry;
3537 saved_start = entry->start;
3538 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
3539 last_timestamp = map->timestamp;
3540 (void) vm_map_unlock_and_wait(map, 0);
3542 if (last_timestamp + 1 != map->timestamp) {
3544 * Look again for the entry because the map was
3545 * modified while it was unlocked.
3546 * Specifically, the entry may have been
3547 * clipped, merged, or deleted.
3549 if (!vm_map_lookup_entry(map, saved_start,
3551 entry = tmp_entry->next;
3554 vm_map_clip_start(map, entry,
3560 vm_map_clip_end(map, entry, end);
3565 * Unwire before removing addresses from the pmap; otherwise,
3566 * unwiring will put the entries back in the pmap.
3568 if (entry->wired_count != 0)
3569 vm_map_entry_unwire(map, entry);
3572 * Remove mappings for the pages, but only if the
3573 * mappings could exist. For instance, it does not
3574 * make sense to call pmap_remove() for guard entries.
3576 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0 ||
3577 entry->object.vm_object != NULL)
3578 pmap_remove(map->pmap, entry->start, entry->end);
3580 if (entry->end == map->anon_loc)
3581 map->anon_loc = entry->start;
3584 * Delete the entry only after removing all pmap
3585 * entries pointing to its pages. (Otherwise, its
3586 * page frames may be reallocated, and any modify bits
3587 * will be set in the wrong object!)
3589 vm_map_entry_delete(map, entry);
3592 return (KERN_SUCCESS);
3598 * Remove the given address range from the target map.
3599 * This is the exported form of vm_map_delete.
3602 vm_map_remove(vm_map_t map, vm_offset_t start, vm_offset_t end)
3607 VM_MAP_RANGE_CHECK(map, start, end);
3608 result = vm_map_delete(map, start, end);
3614 * vm_map_check_protection:
3616 * Assert that the target map allows the specified privilege on the
3617 * entire address region given. The entire region must be allocated.
3619 * WARNING! This code does not and should not check whether the
3620 * contents of the region is accessible. For example a smaller file
3621 * might be mapped into a larger address space.
3623 * NOTE! This code is also called by munmap().
3625 * The map must be locked. A read lock is sufficient.
3628 vm_map_check_protection(vm_map_t map, vm_offset_t start, vm_offset_t end,
3629 vm_prot_t protection)
3631 vm_map_entry_t entry;
3632 vm_map_entry_t tmp_entry;
3634 if (!vm_map_lookup_entry(map, start, &tmp_entry))
3638 while (start < end) {
3642 if (start < entry->start)
3645 * Check protection associated with entry.
3647 if ((entry->protection & protection) != protection)
3649 /* go to next entry */
3651 entry = entry->next;
3657 * vm_map_copy_entry:
3659 * Copies the contents of the source entry to the destination
3660 * entry. The entries *must* be aligned properly.
3666 vm_map_entry_t src_entry,
3667 vm_map_entry_t dst_entry,
3668 vm_ooffset_t *fork_charge)
3670 vm_object_t src_object;
3671 vm_map_entry_t fake_entry;
3676 VM_MAP_ASSERT_LOCKED(dst_map);
3678 if ((dst_entry->eflags|src_entry->eflags) & MAP_ENTRY_IS_SUB_MAP)
3681 if (src_entry->wired_count == 0 ||
3682 (src_entry->protection & VM_PROT_WRITE) == 0) {
3684 * If the source entry is marked needs_copy, it is already
3687 if ((src_entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0 &&
3688 (src_entry->protection & VM_PROT_WRITE) != 0) {
3689 pmap_protect(src_map->pmap,
3692 src_entry->protection & ~VM_PROT_WRITE);
3696 * Make a copy of the object.
3698 size = src_entry->end - src_entry->start;
3699 if ((src_object = src_entry->object.vm_object) != NULL) {
3700 VM_OBJECT_WLOCK(src_object);
3701 charged = ENTRY_CHARGED(src_entry);
3702 if (src_object->handle == NULL &&
3703 (src_object->type == OBJT_DEFAULT ||
3704 src_object->type == OBJT_SWAP)) {
3705 vm_object_collapse(src_object);
3706 if ((src_object->flags & (OBJ_NOSPLIT |
3707 OBJ_ONEMAPPING)) == OBJ_ONEMAPPING) {
3708 vm_object_split(src_entry);
3710 src_entry->object.vm_object;
3713 vm_object_reference_locked(src_object);
3714 vm_object_clear_flag(src_object, OBJ_ONEMAPPING);
3715 if (src_entry->cred != NULL &&
3716 !(src_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
3717 KASSERT(src_object->cred == NULL,
3718 ("OVERCOMMIT: vm_map_copy_entry: cred %p",
3720 src_object->cred = src_entry->cred;
3721 src_object->charge = size;
3723 VM_OBJECT_WUNLOCK(src_object);
3724 dst_entry->object.vm_object = src_object;
3726 cred = curthread->td_ucred;
3728 dst_entry->cred = cred;
3729 *fork_charge += size;
3730 if (!(src_entry->eflags &
3731 MAP_ENTRY_NEEDS_COPY)) {
3733 src_entry->cred = cred;
3734 *fork_charge += size;
3737 src_entry->eflags |= MAP_ENTRY_COW |
3738 MAP_ENTRY_NEEDS_COPY;
3739 dst_entry->eflags |= MAP_ENTRY_COW |
3740 MAP_ENTRY_NEEDS_COPY;
3741 dst_entry->offset = src_entry->offset;
3742 if (src_entry->eflags & MAP_ENTRY_VN_WRITECNT) {
3744 * MAP_ENTRY_VN_WRITECNT cannot
3745 * indicate write reference from
3746 * src_entry, since the entry is
3747 * marked as needs copy. Allocate a
3748 * fake entry that is used to
3749 * decrement object->un_pager.vnp.writecount
3750 * at the appropriate time. Attach
3751 * fake_entry to the deferred list.
3753 fake_entry = vm_map_entry_create(dst_map);
3754 fake_entry->eflags = MAP_ENTRY_VN_WRITECNT;
3755 src_entry->eflags &= ~MAP_ENTRY_VN_WRITECNT;
3756 vm_object_reference(src_object);
3757 fake_entry->object.vm_object = src_object;
3758 fake_entry->start = src_entry->start;
3759 fake_entry->end = src_entry->end;
3760 fake_entry->next = curthread->td_map_def_user;
3761 curthread->td_map_def_user = fake_entry;
3764 pmap_copy(dst_map->pmap, src_map->pmap,
3765 dst_entry->start, dst_entry->end - dst_entry->start,
3768 dst_entry->object.vm_object = NULL;
3769 dst_entry->offset = 0;
3770 if (src_entry->cred != NULL) {
3771 dst_entry->cred = curthread->td_ucred;
3772 crhold(dst_entry->cred);
3773 *fork_charge += size;
3778 * We don't want to make writeable wired pages copy-on-write.
3779 * Immediately copy these pages into the new map by simulating
3780 * page faults. The new pages are pageable.
3782 vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry,
3788 * vmspace_map_entry_forked:
3789 * Update the newly-forked vmspace each time a map entry is inherited
3790 * or copied. The values for vm_dsize and vm_tsize are approximate
3791 * (and mostly-obsolete ideas in the face of mmap(2) et al.)
3794 vmspace_map_entry_forked(const struct vmspace *vm1, struct vmspace *vm2,
3795 vm_map_entry_t entry)
3797 vm_size_t entrysize;
3800 if ((entry->eflags & MAP_ENTRY_GUARD) != 0)
3802 entrysize = entry->end - entry->start;
3803 vm2->vm_map.size += entrysize;
3804 if (entry->eflags & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP)) {
3805 vm2->vm_ssize += btoc(entrysize);
3806 } else if (entry->start >= (vm_offset_t)vm1->vm_daddr &&
3807 entry->start < (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)) {
3808 newend = MIN(entry->end,
3809 (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize));
3810 vm2->vm_dsize += btoc(newend - entry->start);
3811 } else if (entry->start >= (vm_offset_t)vm1->vm_taddr &&
3812 entry->start < (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)) {
3813 newend = MIN(entry->end,
3814 (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize));
3815 vm2->vm_tsize += btoc(newend - entry->start);
3821 * Create a new process vmspace structure and vm_map
3822 * based on those of an existing process. The new map
3823 * is based on the old map, according to the inheritance
3824 * values on the regions in that map.
3826 * XXX It might be worth coalescing the entries added to the new vmspace.
3828 * The source map must not be locked.
3831 vmspace_fork(struct vmspace *vm1, vm_ooffset_t *fork_charge)
3833 struct vmspace *vm2;
3834 vm_map_t new_map, old_map;
3835 vm_map_entry_t new_entry, old_entry;
3840 old_map = &vm1->vm_map;
3841 /* Copy immutable fields of vm1 to vm2. */
3842 vm2 = vmspace_alloc(vm_map_min(old_map), vm_map_max(old_map),
3847 vm2->vm_taddr = vm1->vm_taddr;
3848 vm2->vm_daddr = vm1->vm_daddr;
3849 vm2->vm_maxsaddr = vm1->vm_maxsaddr;
3850 vm_map_lock(old_map);
3852 vm_map_wait_busy(old_map);
3853 new_map = &vm2->vm_map;
3854 locked = vm_map_trylock(new_map); /* trylock to silence WITNESS */
3855 KASSERT(locked, ("vmspace_fork: lock failed"));
3857 error = pmap_vmspace_copy(new_map->pmap, old_map->pmap);
3859 sx_xunlock(&old_map->lock);
3860 sx_xunlock(&new_map->lock);
3861 vm_map_process_deferred();
3866 new_map->anon_loc = old_map->anon_loc;
3868 old_entry = old_map->header.next;
3870 while (old_entry != &old_map->header) {
3871 if (old_entry->eflags & MAP_ENTRY_IS_SUB_MAP)
3872 panic("vm_map_fork: encountered a submap");
3874 inh = old_entry->inheritance;
3875 if ((old_entry->eflags & MAP_ENTRY_GUARD) != 0 &&
3876 inh != VM_INHERIT_NONE)
3877 inh = VM_INHERIT_COPY;
3880 case VM_INHERIT_NONE:
3883 case VM_INHERIT_SHARE:
3885 * Clone the entry, creating the shared object if necessary.
3887 object = old_entry->object.vm_object;
3888 if (object == NULL) {
3889 object = vm_object_allocate(OBJT_DEFAULT,
3890 atop(old_entry->end - old_entry->start));
3891 old_entry->object.vm_object = object;
3892 old_entry->offset = 0;
3893 if (old_entry->cred != NULL) {
3894 object->cred = old_entry->cred;
3895 object->charge = old_entry->end -
3897 old_entry->cred = NULL;
3902 * Add the reference before calling vm_object_shadow
3903 * to insure that a shadow object is created.
3905 vm_object_reference(object);
3906 if (old_entry->eflags & MAP_ENTRY_NEEDS_COPY) {
3907 vm_object_shadow(&old_entry->object.vm_object,
3909 old_entry->end - old_entry->start);
3910 old_entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
3911 /* Transfer the second reference too. */
3912 vm_object_reference(
3913 old_entry->object.vm_object);
3916 * As in vm_map_simplify_entry(), the
3917 * vnode lock will not be acquired in
3918 * this call to vm_object_deallocate().
3920 vm_object_deallocate(object);
3921 object = old_entry->object.vm_object;
3923 VM_OBJECT_WLOCK(object);
3924 vm_object_clear_flag(object, OBJ_ONEMAPPING);
3925 if (old_entry->cred != NULL) {
3926 KASSERT(object->cred == NULL, ("vmspace_fork both cred"));
3927 object->cred = old_entry->cred;
3928 object->charge = old_entry->end - old_entry->start;
3929 old_entry->cred = NULL;
3933 * Assert the correct state of the vnode
3934 * v_writecount while the object is locked, to
3935 * not relock it later for the assertion
3938 if (old_entry->eflags & MAP_ENTRY_VN_WRITECNT &&
3939 object->type == OBJT_VNODE) {
3940 KASSERT(((struct vnode *)object->handle)->
3942 ("vmspace_fork: v_writecount %p", object));
3943 KASSERT(object->un_pager.vnp.writemappings > 0,
3944 ("vmspace_fork: vnp.writecount %p",
3947 VM_OBJECT_WUNLOCK(object);
3950 * Clone the entry, referencing the shared object.
3952 new_entry = vm_map_entry_create(new_map);
3953 *new_entry = *old_entry;
3954 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED |
3955 MAP_ENTRY_IN_TRANSITION);
3956 new_entry->wiring_thread = NULL;
3957 new_entry->wired_count = 0;
3958 if (new_entry->eflags & MAP_ENTRY_VN_WRITECNT) {
3959 vnode_pager_update_writecount(object,
3960 new_entry->start, new_entry->end);
3962 vm_map_entry_set_vnode_text(new_entry, true);
3965 * Insert the entry into the new map -- we know we're
3966 * inserting at the end of the new map.
3968 vm_map_entry_link(new_map, new_entry);
3969 vmspace_map_entry_forked(vm1, vm2, new_entry);
3972 * Update the physical map
3974 pmap_copy(new_map->pmap, old_map->pmap,
3976 (old_entry->end - old_entry->start),
3980 case VM_INHERIT_COPY:
3982 * Clone the entry and link into the map.
3984 new_entry = vm_map_entry_create(new_map);
3985 *new_entry = *old_entry;
3987 * Copied entry is COW over the old object.
3989 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED |
3990 MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_VN_WRITECNT);
3991 new_entry->wiring_thread = NULL;
3992 new_entry->wired_count = 0;
3993 new_entry->object.vm_object = NULL;
3994 new_entry->cred = NULL;
3995 vm_map_entry_link(new_map, new_entry);
3996 vmspace_map_entry_forked(vm1, vm2, new_entry);
3997 vm_map_copy_entry(old_map, new_map, old_entry,
3998 new_entry, fork_charge);
3999 vm_map_entry_set_vnode_text(new_entry, true);
4002 case VM_INHERIT_ZERO:
4004 * Create a new anonymous mapping entry modelled from
4007 new_entry = vm_map_entry_create(new_map);
4008 memset(new_entry, 0, sizeof(*new_entry));
4010 new_entry->start = old_entry->start;
4011 new_entry->end = old_entry->end;
4012 new_entry->eflags = old_entry->eflags &
4013 ~(MAP_ENTRY_USER_WIRED | MAP_ENTRY_IN_TRANSITION |
4014 MAP_ENTRY_VN_WRITECNT | MAP_ENTRY_VN_EXEC);
4015 new_entry->protection = old_entry->protection;
4016 new_entry->max_protection = old_entry->max_protection;
4017 new_entry->inheritance = VM_INHERIT_ZERO;
4019 vm_map_entry_link(new_map, new_entry);
4020 vmspace_map_entry_forked(vm1, vm2, new_entry);
4022 new_entry->cred = curthread->td_ucred;
4023 crhold(new_entry->cred);
4024 *fork_charge += (new_entry->end - new_entry->start);
4028 old_entry = old_entry->next;
4031 * Use inlined vm_map_unlock() to postpone handling the deferred
4032 * map entries, which cannot be done until both old_map and
4033 * new_map locks are released.
4035 sx_xunlock(&old_map->lock);
4036 sx_xunlock(&new_map->lock);
4037 vm_map_process_deferred();
4043 * Create a process's stack for exec_new_vmspace(). This function is never
4044 * asked to wire the newly created stack.
4047 vm_map_stack(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
4048 vm_prot_t prot, vm_prot_t max, int cow)
4050 vm_size_t growsize, init_ssize;
4054 MPASS((map->flags & MAP_WIREFUTURE) == 0);
4055 growsize = sgrowsiz;
4056 init_ssize = (max_ssize < growsize) ? max_ssize : growsize;
4058 vmemlim = lim_cur(curthread, RLIMIT_VMEM);
4059 /* If we would blow our VMEM resource limit, no go */
4060 if (map->size + init_ssize > vmemlim) {
4064 rv = vm_map_stack_locked(map, addrbos, max_ssize, growsize, prot,
4071 static int stack_guard_page = 1;
4072 SYSCTL_INT(_security_bsd, OID_AUTO, stack_guard_page, CTLFLAG_RWTUN,
4073 &stack_guard_page, 0,
4074 "Specifies the number of guard pages for a stack that grows");
4077 vm_map_stack_locked(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
4078 vm_size_t growsize, vm_prot_t prot, vm_prot_t max, int cow)
4080 vm_map_entry_t new_entry, prev_entry;
4081 vm_offset_t bot, gap_bot, gap_top, top;
4082 vm_size_t init_ssize, sgp;
4086 * The stack orientation is piggybacked with the cow argument.
4087 * Extract it into orient and mask the cow argument so that we
4088 * don't pass it around further.
4090 orient = cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP);
4091 KASSERT(orient != 0, ("No stack grow direction"));
4092 KASSERT(orient != (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP),
4095 if (addrbos < vm_map_min(map) ||
4096 addrbos + max_ssize > vm_map_max(map) ||
4097 addrbos + max_ssize <= addrbos)
4098 return (KERN_INVALID_ADDRESS);
4099 sgp = (vm_size_t)stack_guard_page * PAGE_SIZE;
4100 if (sgp >= max_ssize)
4101 return (KERN_INVALID_ARGUMENT);
4103 init_ssize = growsize;
4104 if (max_ssize < init_ssize + sgp)
4105 init_ssize = max_ssize - sgp;
4107 /* If addr is already mapped, no go */
4108 if (vm_map_lookup_entry(map, addrbos, &prev_entry))
4109 return (KERN_NO_SPACE);
4112 * If we can't accommodate max_ssize in the current mapping, no go.
4114 if (prev_entry->next->start < addrbos + max_ssize)
4115 return (KERN_NO_SPACE);
4118 * We initially map a stack of only init_ssize. We will grow as
4119 * needed later. Depending on the orientation of the stack (i.e.
4120 * the grow direction) we either map at the top of the range, the
4121 * bottom of the range or in the middle.
4123 * Note: we would normally expect prot and max to be VM_PROT_ALL,
4124 * and cow to be 0. Possibly we should eliminate these as input
4125 * parameters, and just pass these values here in the insert call.
4127 if (orient == MAP_STACK_GROWS_DOWN) {
4128 bot = addrbos + max_ssize - init_ssize;
4129 top = bot + init_ssize;
4132 } else /* if (orient == MAP_STACK_GROWS_UP) */ {
4134 top = bot + init_ssize;
4136 gap_top = addrbos + max_ssize;
4138 rv = vm_map_insert(map, NULL, 0, bot, top, prot, max, cow);
4139 if (rv != KERN_SUCCESS)
4141 new_entry = prev_entry->next;
4142 KASSERT(new_entry->end == top || new_entry->start == bot,
4143 ("Bad entry start/end for new stack entry"));
4144 KASSERT((orient & MAP_STACK_GROWS_DOWN) == 0 ||
4145 (new_entry->eflags & MAP_ENTRY_GROWS_DOWN) != 0,
4146 ("new entry lacks MAP_ENTRY_GROWS_DOWN"));
4147 KASSERT((orient & MAP_STACK_GROWS_UP) == 0 ||
4148 (new_entry->eflags & MAP_ENTRY_GROWS_UP) != 0,
4149 ("new entry lacks MAP_ENTRY_GROWS_UP"));
4150 rv = vm_map_insert(map, NULL, 0, gap_bot, gap_top, VM_PROT_NONE,
4151 VM_PROT_NONE, MAP_CREATE_GUARD | (orient == MAP_STACK_GROWS_DOWN ?
4152 MAP_CREATE_STACK_GAP_DN : MAP_CREATE_STACK_GAP_UP));
4153 if (rv != KERN_SUCCESS)
4154 (void)vm_map_delete(map, bot, top);
4159 * Attempts to grow a vm stack entry. Returns KERN_SUCCESS if we
4160 * successfully grow the stack.
4163 vm_map_growstack(vm_map_t map, vm_offset_t addr, vm_map_entry_t gap_entry)
4165 vm_map_entry_t stack_entry;
4169 vm_offset_t gap_end, gap_start, grow_start;
4170 size_t grow_amount, guard, max_grow;
4171 rlim_t lmemlim, stacklim, vmemlim;
4173 bool gap_deleted, grow_down, is_procstack;
4185 * Disallow stack growth when the access is performed by a
4186 * debugger or AIO daemon. The reason is that the wrong
4187 * resource limits are applied.
4189 if (map != &p->p_vmspace->vm_map || p->p_textvp == NULL)
4190 return (KERN_FAILURE);
4192 MPASS(!map->system_map);
4194 guard = stack_guard_page * PAGE_SIZE;
4195 lmemlim = lim_cur(curthread, RLIMIT_MEMLOCK);
4196 stacklim = lim_cur(curthread, RLIMIT_STACK);
4197 vmemlim = lim_cur(curthread, RLIMIT_VMEM);
4199 /* If addr is not in a hole for a stack grow area, no need to grow. */
4200 if (gap_entry == NULL && !vm_map_lookup_entry(map, addr, &gap_entry))
4201 return (KERN_FAILURE);
4202 if ((gap_entry->eflags & MAP_ENTRY_GUARD) == 0)
4203 return (KERN_SUCCESS);
4204 if ((gap_entry->eflags & MAP_ENTRY_STACK_GAP_DN) != 0) {
4205 stack_entry = gap_entry->next;
4206 if ((stack_entry->eflags & MAP_ENTRY_GROWS_DOWN) == 0 ||
4207 stack_entry->start != gap_entry->end)
4208 return (KERN_FAILURE);
4209 grow_amount = round_page(stack_entry->start - addr);
4211 } else if ((gap_entry->eflags & MAP_ENTRY_STACK_GAP_UP) != 0) {
4212 stack_entry = gap_entry->prev;
4213 if ((stack_entry->eflags & MAP_ENTRY_GROWS_UP) == 0 ||
4214 stack_entry->end != gap_entry->start)
4215 return (KERN_FAILURE);
4216 grow_amount = round_page(addr + 1 - stack_entry->end);
4219 return (KERN_FAILURE);
4221 max_grow = gap_entry->end - gap_entry->start;
4222 if (guard > max_grow)
4223 return (KERN_NO_SPACE);
4225 if (grow_amount > max_grow)
4226 return (KERN_NO_SPACE);
4229 * If this is the main process stack, see if we're over the stack
4232 is_procstack = addr >= (vm_offset_t)vm->vm_maxsaddr &&
4233 addr < (vm_offset_t)p->p_sysent->sv_usrstack;
4234 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim))
4235 return (KERN_NO_SPACE);
4240 if (is_procstack && racct_set(p, RACCT_STACK,
4241 ctob(vm->vm_ssize) + grow_amount)) {
4243 return (KERN_NO_SPACE);
4249 grow_amount = roundup(grow_amount, sgrowsiz);
4250 if (grow_amount > max_grow)
4251 grow_amount = max_grow;
4252 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) {
4253 grow_amount = trunc_page((vm_size_t)stacklim) -
4259 limit = racct_get_available(p, RACCT_STACK);
4261 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > limit))
4262 grow_amount = limit - ctob(vm->vm_ssize);
4265 if (!old_mlock && (map->flags & MAP_WIREFUTURE) != 0) {
4266 if (ptoa(pmap_wired_count(map->pmap)) + grow_amount > lmemlim) {
4273 if (racct_set(p, RACCT_MEMLOCK,
4274 ptoa(pmap_wired_count(map->pmap)) + grow_amount)) {
4284 /* If we would blow our VMEM resource limit, no go */
4285 if (map->size + grow_amount > vmemlim) {
4292 if (racct_set(p, RACCT_VMEM, map->size + grow_amount)) {
4301 if (vm_map_lock_upgrade(map)) {
4303 vm_map_lock_read(map);
4308 grow_start = gap_entry->end - grow_amount;
4309 if (gap_entry->start + grow_amount == gap_entry->end) {
4310 gap_start = gap_entry->start;
4311 gap_end = gap_entry->end;
4312 vm_map_entry_delete(map, gap_entry);
4315 MPASS(gap_entry->start < gap_entry->end - grow_amount);
4316 gap_entry->end -= grow_amount;
4317 vm_map_entry_resize_free(map, gap_entry);
4318 gap_deleted = false;
4320 rv = vm_map_insert(map, NULL, 0, grow_start,
4321 grow_start + grow_amount,
4322 stack_entry->protection, stack_entry->max_protection,
4323 MAP_STACK_GROWS_DOWN);
4324 if (rv != KERN_SUCCESS) {
4326 rv1 = vm_map_insert(map, NULL, 0, gap_start,
4327 gap_end, VM_PROT_NONE, VM_PROT_NONE,
4328 MAP_CREATE_GUARD | MAP_CREATE_STACK_GAP_DN);
4329 MPASS(rv1 == KERN_SUCCESS);
4331 gap_entry->end += grow_amount;
4332 vm_map_entry_resize_free(map, gap_entry);
4336 grow_start = stack_entry->end;
4337 cred = stack_entry->cred;
4338 if (cred == NULL && stack_entry->object.vm_object != NULL)
4339 cred = stack_entry->object.vm_object->cred;
4340 if (cred != NULL && !swap_reserve_by_cred(grow_amount, cred))
4342 /* Grow the underlying object if applicable. */
4343 else if (stack_entry->object.vm_object == NULL ||
4344 vm_object_coalesce(stack_entry->object.vm_object,
4345 stack_entry->offset,
4346 (vm_size_t)(stack_entry->end - stack_entry->start),
4347 (vm_size_t)grow_amount, cred != NULL)) {
4348 if (gap_entry->start + grow_amount == gap_entry->end)
4349 vm_map_entry_delete(map, gap_entry);
4351 gap_entry->start += grow_amount;
4352 stack_entry->end += grow_amount;
4353 map->size += grow_amount;
4354 vm_map_entry_resize_free(map, stack_entry);
4359 if (rv == KERN_SUCCESS && is_procstack)
4360 vm->vm_ssize += btoc(grow_amount);
4363 * Heed the MAP_WIREFUTURE flag if it was set for this process.
4365 if (rv == KERN_SUCCESS && (map->flags & MAP_WIREFUTURE) != 0) {
4366 rv = vm_map_wire_locked(map, grow_start,
4367 grow_start + grow_amount,
4368 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
4370 vm_map_lock_downgrade(map);
4374 if (racct_enable && rv != KERN_SUCCESS) {
4376 error = racct_set(p, RACCT_VMEM, map->size);
4377 KASSERT(error == 0, ("decreasing RACCT_VMEM failed"));
4379 error = racct_set(p, RACCT_MEMLOCK,
4380 ptoa(pmap_wired_count(map->pmap)));
4381 KASSERT(error == 0, ("decreasing RACCT_MEMLOCK failed"));
4383 error = racct_set(p, RACCT_STACK, ctob(vm->vm_ssize));
4384 KASSERT(error == 0, ("decreasing RACCT_STACK failed"));
4393 * Unshare the specified VM space for exec. If other processes are
4394 * mapped to it, then create a new one. The new vmspace is null.
4397 vmspace_exec(struct proc *p, vm_offset_t minuser, vm_offset_t maxuser)
4399 struct vmspace *oldvmspace = p->p_vmspace;
4400 struct vmspace *newvmspace;
4402 KASSERT((curthread->td_pflags & TDP_EXECVMSPC) == 0,
4403 ("vmspace_exec recursed"));
4404 newvmspace = vmspace_alloc(minuser, maxuser, pmap_pinit);
4405 if (newvmspace == NULL)
4407 newvmspace->vm_swrss = oldvmspace->vm_swrss;
4409 * This code is written like this for prototype purposes. The
4410 * goal is to avoid running down the vmspace here, but let the
4411 * other process's that are still using the vmspace to finally
4412 * run it down. Even though there is little or no chance of blocking
4413 * here, it is a good idea to keep this form for future mods.
4415 PROC_VMSPACE_LOCK(p);
4416 p->p_vmspace = newvmspace;
4417 PROC_VMSPACE_UNLOCK(p);
4418 if (p == curthread->td_proc)
4419 pmap_activate(curthread);
4420 curthread->td_pflags |= TDP_EXECVMSPC;
4425 * Unshare the specified VM space for forcing COW. This
4426 * is called by rfork, for the (RFMEM|RFPROC) == 0 case.
4429 vmspace_unshare(struct proc *p)
4431 struct vmspace *oldvmspace = p->p_vmspace;
4432 struct vmspace *newvmspace;
4433 vm_ooffset_t fork_charge;
4435 if (oldvmspace->vm_refcnt == 1)
4438 newvmspace = vmspace_fork(oldvmspace, &fork_charge);
4439 if (newvmspace == NULL)
4441 if (!swap_reserve_by_cred(fork_charge, p->p_ucred)) {
4442 vmspace_free(newvmspace);
4445 PROC_VMSPACE_LOCK(p);
4446 p->p_vmspace = newvmspace;
4447 PROC_VMSPACE_UNLOCK(p);
4448 if (p == curthread->td_proc)
4449 pmap_activate(curthread);
4450 vmspace_free(oldvmspace);
4457 * Finds the VM object, offset, and
4458 * protection for a given virtual address in the
4459 * specified map, assuming a page fault of the
4462 * Leaves the map in question locked for read; return
4463 * values are guaranteed until a vm_map_lookup_done
4464 * call is performed. Note that the map argument
4465 * is in/out; the returned map must be used in
4466 * the call to vm_map_lookup_done.
4468 * A handle (out_entry) is returned for use in
4469 * vm_map_lookup_done, to make that fast.
4471 * If a lookup is requested with "write protection"
4472 * specified, the map may be changed to perform virtual
4473 * copying operations, although the data referenced will
4477 vm_map_lookup(vm_map_t *var_map, /* IN/OUT */
4479 vm_prot_t fault_typea,
4480 vm_map_entry_t *out_entry, /* OUT */
4481 vm_object_t *object, /* OUT */
4482 vm_pindex_t *pindex, /* OUT */
4483 vm_prot_t *out_prot, /* OUT */
4484 boolean_t *wired) /* OUT */
4486 vm_map_entry_t entry;
4487 vm_map_t map = *var_map;
4489 vm_prot_t fault_type = fault_typea;
4490 vm_object_t eobject;
4496 vm_map_lock_read(map);
4500 * Lookup the faulting address.
4502 if (!vm_map_lookup_entry(map, vaddr, out_entry)) {
4503 vm_map_unlock_read(map);
4504 return (KERN_INVALID_ADDRESS);
4512 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
4513 vm_map_t old_map = map;
4515 *var_map = map = entry->object.sub_map;
4516 vm_map_unlock_read(old_map);
4521 * Check whether this task is allowed to have this page.
4523 prot = entry->protection;
4524 if ((fault_typea & VM_PROT_FAULT_LOOKUP) != 0) {
4525 fault_typea &= ~VM_PROT_FAULT_LOOKUP;
4526 if (prot == VM_PROT_NONE && map != kernel_map &&
4527 (entry->eflags & MAP_ENTRY_GUARD) != 0 &&
4528 (entry->eflags & (MAP_ENTRY_STACK_GAP_DN |
4529 MAP_ENTRY_STACK_GAP_UP)) != 0 &&
4530 vm_map_growstack(map, vaddr, entry) == KERN_SUCCESS)
4531 goto RetryLookupLocked;
4533 fault_type &= VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE;
4534 if ((fault_type & prot) != fault_type || prot == VM_PROT_NONE) {
4535 vm_map_unlock_read(map);
4536 return (KERN_PROTECTION_FAILURE);
4538 KASSERT((prot & VM_PROT_WRITE) == 0 || (entry->eflags &
4539 (MAP_ENTRY_USER_WIRED | MAP_ENTRY_NEEDS_COPY)) !=
4540 (MAP_ENTRY_USER_WIRED | MAP_ENTRY_NEEDS_COPY),
4541 ("entry %p flags %x", entry, entry->eflags));
4542 if ((fault_typea & VM_PROT_COPY) != 0 &&
4543 (entry->max_protection & VM_PROT_WRITE) == 0 &&
4544 (entry->eflags & MAP_ENTRY_COW) == 0) {
4545 vm_map_unlock_read(map);
4546 return (KERN_PROTECTION_FAILURE);
4550 * If this page is not pageable, we have to get it for all possible
4553 *wired = (entry->wired_count != 0);
4555 fault_type = entry->protection;
4556 size = entry->end - entry->start;
4558 * If the entry was copy-on-write, we either ...
4560 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
4562 * If we want to write the page, we may as well handle that
4563 * now since we've got the map locked.
4565 * If we don't need to write the page, we just demote the
4566 * permissions allowed.
4568 if ((fault_type & VM_PROT_WRITE) != 0 ||
4569 (fault_typea & VM_PROT_COPY) != 0) {
4571 * Make a new object, and place it in the object
4572 * chain. Note that no new references have appeared
4573 * -- one just moved from the map to the new
4576 if (vm_map_lock_upgrade(map))
4579 if (entry->cred == NULL) {
4581 * The debugger owner is charged for
4584 cred = curthread->td_ucred;
4586 if (!swap_reserve_by_cred(size, cred)) {
4589 return (KERN_RESOURCE_SHORTAGE);
4593 vm_object_shadow(&entry->object.vm_object,
4594 &entry->offset, size);
4595 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
4596 eobject = entry->object.vm_object;
4597 if (eobject->cred != NULL) {
4599 * The object was not shadowed.
4601 swap_release_by_cred(size, entry->cred);
4602 crfree(entry->cred);
4604 } else if (entry->cred != NULL) {
4605 VM_OBJECT_WLOCK(eobject);
4606 eobject->cred = entry->cred;
4607 eobject->charge = size;
4608 VM_OBJECT_WUNLOCK(eobject);
4612 vm_map_lock_downgrade(map);
4615 * We're attempting to read a copy-on-write page --
4616 * don't allow writes.
4618 prot &= ~VM_PROT_WRITE;
4623 * Create an object if necessary.
4625 if (entry->object.vm_object == NULL &&
4627 if (vm_map_lock_upgrade(map))
4629 entry->object.vm_object = vm_object_allocate(OBJT_DEFAULT,
4632 if (entry->cred != NULL) {
4633 VM_OBJECT_WLOCK(entry->object.vm_object);
4634 entry->object.vm_object->cred = entry->cred;
4635 entry->object.vm_object->charge = size;
4636 VM_OBJECT_WUNLOCK(entry->object.vm_object);
4639 vm_map_lock_downgrade(map);
4643 * Return the object/offset from this entry. If the entry was
4644 * copy-on-write or empty, it has been fixed up.
4646 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
4647 *object = entry->object.vm_object;
4650 return (KERN_SUCCESS);
4654 * vm_map_lookup_locked:
4656 * Lookup the faulting address. A version of vm_map_lookup that returns
4657 * KERN_FAILURE instead of blocking on map lock or memory allocation.
4660 vm_map_lookup_locked(vm_map_t *var_map, /* IN/OUT */
4662 vm_prot_t fault_typea,
4663 vm_map_entry_t *out_entry, /* OUT */
4664 vm_object_t *object, /* OUT */
4665 vm_pindex_t *pindex, /* OUT */
4666 vm_prot_t *out_prot, /* OUT */
4667 boolean_t *wired) /* OUT */
4669 vm_map_entry_t entry;
4670 vm_map_t map = *var_map;
4672 vm_prot_t fault_type = fault_typea;
4675 * Lookup the faulting address.
4677 if (!vm_map_lookup_entry(map, vaddr, out_entry))
4678 return (KERN_INVALID_ADDRESS);
4683 * Fail if the entry refers to a submap.
4685 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP)
4686 return (KERN_FAILURE);
4689 * Check whether this task is allowed to have this page.
4691 prot = entry->protection;
4692 fault_type &= VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE;
4693 if ((fault_type & prot) != fault_type)
4694 return (KERN_PROTECTION_FAILURE);
4697 * If this page is not pageable, we have to get it for all possible
4700 *wired = (entry->wired_count != 0);
4702 fault_type = entry->protection;
4704 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
4706 * Fail if the entry was copy-on-write for a write fault.
4708 if (fault_type & VM_PROT_WRITE)
4709 return (KERN_FAILURE);
4711 * We're attempting to read a copy-on-write page --
4712 * don't allow writes.
4714 prot &= ~VM_PROT_WRITE;
4718 * Fail if an object should be created.
4720 if (entry->object.vm_object == NULL && !map->system_map)
4721 return (KERN_FAILURE);
4724 * Return the object/offset from this entry. If the entry was
4725 * copy-on-write or empty, it has been fixed up.
4727 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
4728 *object = entry->object.vm_object;
4731 return (KERN_SUCCESS);
4735 * vm_map_lookup_done:
4737 * Releases locks acquired by a vm_map_lookup
4738 * (according to the handle returned by that lookup).
4741 vm_map_lookup_done(vm_map_t map, vm_map_entry_t entry)
4744 * Unlock the main-level map
4746 vm_map_unlock_read(map);
4750 vm_map_max_KBI(const struct vm_map *map)
4753 return (vm_map_max(map));
4757 vm_map_min_KBI(const struct vm_map *map)
4760 return (vm_map_min(map));
4764 vm_map_pmap_KBI(vm_map_t map)
4770 #include "opt_ddb.h"
4772 #include <sys/kernel.h>
4774 #include <ddb/ddb.h>
4777 vm_map_print(vm_map_t map)
4779 vm_map_entry_t entry;
4781 db_iprintf("Task map %p: pmap=%p, nentries=%d, version=%u\n",
4783 (void *)map->pmap, map->nentries, map->timestamp);
4786 for (entry = map->header.next; entry != &map->header;
4787 entry = entry->next) {
4788 db_iprintf("map entry %p: start=%p, end=%p, eflags=%#x, \n",
4789 (void *)entry, (void *)entry->start, (void *)entry->end,
4792 static char *inheritance_name[4] =
4793 {"share", "copy", "none", "donate_copy"};
4795 db_iprintf(" prot=%x/%x/%s",
4797 entry->max_protection,
4798 inheritance_name[(int)(unsigned char)entry->inheritance]);
4799 if (entry->wired_count != 0)
4800 db_printf(", wired");
4802 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
4803 db_printf(", share=%p, offset=0x%jx\n",
4804 (void *)entry->object.sub_map,
4805 (uintmax_t)entry->offset);
4806 if ((entry->prev == &map->header) ||
4807 (entry->prev->object.sub_map !=
4808 entry->object.sub_map)) {
4810 vm_map_print((vm_map_t)entry->object.sub_map);
4814 if (entry->cred != NULL)
4815 db_printf(", ruid %d", entry->cred->cr_ruid);
4816 db_printf(", object=%p, offset=0x%jx",
4817 (void *)entry->object.vm_object,
4818 (uintmax_t)entry->offset);
4819 if (entry->object.vm_object && entry->object.vm_object->cred)
4820 db_printf(", obj ruid %d charge %jx",
4821 entry->object.vm_object->cred->cr_ruid,
4822 (uintmax_t)entry->object.vm_object->charge);
4823 if (entry->eflags & MAP_ENTRY_COW)
4824 db_printf(", copy (%s)",
4825 (entry->eflags & MAP_ENTRY_NEEDS_COPY) ? "needed" : "done");
4828 if ((entry->prev == &map->header) ||
4829 (entry->prev->object.vm_object !=
4830 entry->object.vm_object)) {
4832 vm_object_print((db_expr_t)(intptr_t)
4833 entry->object.vm_object,
4842 DB_SHOW_COMMAND(map, map)
4846 db_printf("usage: show map <addr>\n");
4849 vm_map_print((vm_map_t)addr);
4852 DB_SHOW_COMMAND(procvm, procvm)
4857 p = db_lookup_proc(addr);
4862 db_printf("p = %p, vmspace = %p, map = %p, pmap = %p\n",
4863 (void *)p, (void *)p->p_vmspace, (void *)&p->p_vmspace->vm_map,
4864 (void *)vmspace_pmap(p->p_vmspace));
4866 vm_map_print((vm_map_t)&p->p_vmspace->vm_map);