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_object.h>
94 #include <vm/vm_pager.h>
95 #include <vm/vm_kern.h>
96 #include <vm/vm_extern.h>
97 #include <vm/vnode_pager.h>
98 #include <vm/swap_pager.h>
102 * Virtual memory maps provide for the mapping, protection,
103 * and sharing of virtual memory objects. In addition,
104 * this module provides for an efficient virtual copy of
105 * memory from one map to another.
107 * Synchronization is required prior to most operations.
109 * Maps consist of an ordered doubly-linked list of simple
110 * entries; a self-adjusting binary search tree of these
111 * entries is used to speed up lookups.
113 * Since portions of maps are specified by start/end addresses,
114 * which may not align with existing map entries, all
115 * routines merely "clip" entries to these start/end values.
116 * [That is, an entry is split into two, bordering at a
117 * start or end value.] Note that these clippings may not
118 * always be necessary (as the two resulting entries are then
119 * not changed); however, the clipping is done for convenience.
121 * As mentioned above, virtual copy operations are performed
122 * by copying VM object references from one map to
123 * another, and then marking both regions as copy-on-write.
126 static struct mtx map_sleep_mtx;
127 static uma_zone_t mapentzone;
128 static uma_zone_t kmapentzone;
129 static uma_zone_t mapzone;
130 static uma_zone_t vmspace_zone;
131 static int vmspace_zinit(void *mem, int size, int flags);
132 static int vm_map_zinit(void *mem, int ize, int flags);
133 static void _vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min,
135 static int vm_map_alignspace(vm_map_t map, vm_object_t object,
136 vm_ooffset_t offset, vm_offset_t *addr, vm_size_t length,
137 vm_offset_t max_addr, vm_offset_t alignment);
138 static void vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map);
139 static void vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry);
140 static void vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry);
141 static int vm_map_growstack(vm_map_t map, vm_offset_t addr,
142 vm_map_entry_t gap_entry);
143 static void vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot,
144 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags);
146 static void vm_map_zdtor(void *mem, int size, void *arg);
147 static void vmspace_zdtor(void *mem, int size, void *arg);
149 static int vm_map_stack_locked(vm_map_t map, vm_offset_t addrbos,
150 vm_size_t max_ssize, vm_size_t growsize, vm_prot_t prot, vm_prot_t max,
152 static void vm_map_wire_entry_failure(vm_map_t map, vm_map_entry_t entry,
153 vm_offset_t failed_addr);
155 #define ENTRY_CHARGED(e) ((e)->cred != NULL || \
156 ((e)->object.vm_object != NULL && (e)->object.vm_object->cred != NULL && \
157 !((e)->eflags & MAP_ENTRY_NEEDS_COPY)))
160 * PROC_VMSPACE_{UN,}LOCK() can be a noop as long as vmspaces are type
163 #define PROC_VMSPACE_LOCK(p) do { } while (0)
164 #define PROC_VMSPACE_UNLOCK(p) do { } while (0)
167 * VM_MAP_RANGE_CHECK: [ internal use only ]
169 * Asserts that the starting and ending region
170 * addresses fall within the valid range of the map.
172 #define VM_MAP_RANGE_CHECK(map, start, end) \
174 if (start < vm_map_min(map)) \
175 start = vm_map_min(map); \
176 if (end > vm_map_max(map)) \
177 end = vm_map_max(map); \
185 * Initialize the vm_map module. Must be called before
186 * any other vm_map routines.
188 * Map and entry structures are allocated from the general
189 * purpose memory pool with some exceptions:
191 * - The kernel map and kmem submap are allocated statically.
192 * - Kernel map entries are allocated out of a static pool.
194 * These restrictions are necessary since malloc() uses the
195 * maps and requires map entries.
201 mtx_init(&map_sleep_mtx, "vm map sleep mutex", NULL, MTX_DEF);
202 mapzone = uma_zcreate("MAP", sizeof(struct vm_map), NULL,
208 vm_map_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
209 uma_prealloc(mapzone, MAX_KMAP);
210 kmapentzone = uma_zcreate("KMAP ENTRY", sizeof(struct vm_map_entry),
211 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR,
212 UMA_ZONE_MTXCLASS | UMA_ZONE_VM);
213 mapentzone = uma_zcreate("MAP ENTRY", sizeof(struct vm_map_entry),
214 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
215 vmspace_zone = uma_zcreate("VMSPACE", sizeof(struct vmspace), NULL,
221 vmspace_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
225 vmspace_zinit(void *mem, int size, int flags)
229 vm = (struct vmspace *)mem;
231 vm->vm_map.pmap = NULL;
232 (void)vm_map_zinit(&vm->vm_map, sizeof(vm->vm_map), flags);
233 PMAP_LOCK_INIT(vmspace_pmap(vm));
238 vm_map_zinit(void *mem, int size, int flags)
243 memset(map, 0, sizeof(*map));
244 mtx_init(&map->system_mtx, "vm map (system)", NULL, MTX_DEF | MTX_DUPOK);
245 sx_init(&map->lock, "vm map (user)");
251 vmspace_zdtor(void *mem, int size, void *arg)
255 vm = (struct vmspace *)mem;
257 vm_map_zdtor(&vm->vm_map, sizeof(vm->vm_map), arg);
260 vm_map_zdtor(void *mem, int size, void *arg)
265 KASSERT(map->nentries == 0,
266 ("map %p nentries == %d on free.",
267 map, map->nentries));
268 KASSERT(map->size == 0,
269 ("map %p size == %lu on free.",
270 map, (unsigned long)map->size));
272 #endif /* INVARIANTS */
275 * Allocate a vmspace structure, including a vm_map and pmap,
276 * and initialize those structures. The refcnt is set to 1.
278 * If 'pinit' is NULL then the embedded pmap is initialized via pmap_pinit().
281 vmspace_alloc(vm_offset_t min, vm_offset_t max, pmap_pinit_t pinit)
285 vm = uma_zalloc(vmspace_zone, M_WAITOK);
286 KASSERT(vm->vm_map.pmap == NULL, ("vm_map.pmap must be NULL"));
287 if (!pinit(vmspace_pmap(vm))) {
288 uma_zfree(vmspace_zone, vm);
291 CTR1(KTR_VM, "vmspace_alloc: %p", vm);
292 _vm_map_init(&vm->vm_map, vmspace_pmap(vm), min, max);
307 vmspace_container_reset(struct proc *p)
311 racct_set(p, RACCT_DATA, 0);
312 racct_set(p, RACCT_STACK, 0);
313 racct_set(p, RACCT_RSS, 0);
314 racct_set(p, RACCT_MEMLOCK, 0);
315 racct_set(p, RACCT_VMEM, 0);
321 vmspace_dofree(struct vmspace *vm)
324 CTR1(KTR_VM, "vmspace_free: %p", vm);
327 * Make sure any SysV shm is freed, it might not have been in
333 * Lock the map, to wait out all other references to it.
334 * Delete all of the mappings and pages they hold, then call
335 * the pmap module to reclaim anything left.
337 (void)vm_map_remove(&vm->vm_map, vm_map_min(&vm->vm_map),
338 vm_map_max(&vm->vm_map));
340 pmap_release(vmspace_pmap(vm));
341 vm->vm_map.pmap = NULL;
342 uma_zfree(vmspace_zone, vm);
346 vmspace_free(struct vmspace *vm)
349 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
350 "vmspace_free() called");
352 if (vm->vm_refcnt == 0)
353 panic("vmspace_free: attempt to free already freed vmspace");
355 if (atomic_fetchadd_int(&vm->vm_refcnt, -1) == 1)
360 vmspace_exitfree(struct proc *p)
364 PROC_VMSPACE_LOCK(p);
367 PROC_VMSPACE_UNLOCK(p);
368 KASSERT(vm == &vmspace0, ("vmspace_exitfree: wrong vmspace"));
373 vmspace_exit(struct thread *td)
380 * Release user portion of address space.
381 * This releases references to vnodes,
382 * which could cause I/O if the file has been unlinked.
383 * Need to do this early enough that we can still sleep.
385 * The last exiting process to reach this point releases as
386 * much of the environment as it can. vmspace_dofree() is the
387 * slower fallback in case another process had a temporary
388 * reference to the vmspace.
393 atomic_add_int(&vmspace0.vm_refcnt, 1);
394 refcnt = vm->vm_refcnt;
396 if (refcnt > 1 && p->p_vmspace != &vmspace0) {
397 /* Switch now since other proc might free vmspace */
398 PROC_VMSPACE_LOCK(p);
399 p->p_vmspace = &vmspace0;
400 PROC_VMSPACE_UNLOCK(p);
403 } while (!atomic_fcmpset_int(&vm->vm_refcnt, &refcnt, refcnt - 1));
405 if (p->p_vmspace != vm) {
406 /* vmspace not yet freed, switch back */
407 PROC_VMSPACE_LOCK(p);
409 PROC_VMSPACE_UNLOCK(p);
412 pmap_remove_pages(vmspace_pmap(vm));
413 /* Switch now since this proc will free vmspace */
414 PROC_VMSPACE_LOCK(p);
415 p->p_vmspace = &vmspace0;
416 PROC_VMSPACE_UNLOCK(p);
422 vmspace_container_reset(p);
426 /* Acquire reference to vmspace owned by another process. */
429 vmspace_acquire_ref(struct proc *p)
434 PROC_VMSPACE_LOCK(p);
437 PROC_VMSPACE_UNLOCK(p);
440 refcnt = vm->vm_refcnt;
442 if (refcnt <= 0) { /* Avoid 0->1 transition */
443 PROC_VMSPACE_UNLOCK(p);
446 } while (!atomic_fcmpset_int(&vm->vm_refcnt, &refcnt, refcnt + 1));
447 if (vm != p->p_vmspace) {
448 PROC_VMSPACE_UNLOCK(p);
452 PROC_VMSPACE_UNLOCK(p);
457 * Switch between vmspaces in an AIO kernel process.
459 * The AIO kernel processes switch to and from a user process's
460 * vmspace while performing an I/O operation on behalf of a user
461 * process. The new vmspace is either the vmspace of a user process
462 * obtained from an active AIO request or the initial vmspace of the
463 * AIO kernel process (when it is idling). Because user processes
464 * will block to drain any active AIO requests before proceeding in
465 * exit() or execve(), the vmspace reference count for these vmspaces
466 * can never be 0. This allows for a much simpler implementation than
467 * the loop in vmspace_acquire_ref() above. Similarly, AIO kernel
468 * processes hold an extra reference on their initial vmspace for the
469 * life of the process so that this guarantee is true for any vmspace
473 vmspace_switch_aio(struct vmspace *newvm)
475 struct vmspace *oldvm;
477 /* XXX: Need some way to assert that this is an aio daemon. */
479 KASSERT(newvm->vm_refcnt > 0,
480 ("vmspace_switch_aio: newvm unreferenced"));
482 oldvm = curproc->p_vmspace;
487 * Point to the new address space and refer to it.
489 curproc->p_vmspace = newvm;
490 atomic_add_int(&newvm->vm_refcnt, 1);
492 /* Activate the new mapping. */
493 pmap_activate(curthread);
495 /* Remove the daemon's reference to the old address space. */
496 KASSERT(oldvm->vm_refcnt > 1,
497 ("vmspace_switch_aio: oldvm dropping last reference"));
502 _vm_map_lock(vm_map_t map, const char *file, int line)
506 mtx_lock_flags_(&map->system_mtx, 0, file, line);
508 sx_xlock_(&map->lock, file, line);
513 vm_map_process_deferred(void)
516 vm_map_entry_t entry, next;
520 entry = td->td_map_def_user;
521 td->td_map_def_user = NULL;
522 while (entry != NULL) {
524 if ((entry->eflags & MAP_ENTRY_VN_WRITECNT) != 0) {
526 * Decrement the object's writemappings and
527 * possibly the vnode's v_writecount.
529 KASSERT((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0,
530 ("Submap with writecount"));
531 object = entry->object.vm_object;
532 KASSERT(object != NULL, ("No object for writecount"));
533 vnode_pager_release_writecount(object, entry->start,
536 vm_map_entry_deallocate(entry, FALSE);
542 _vm_map_unlock(vm_map_t map, const char *file, int line)
546 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
548 sx_xunlock_(&map->lock, file, line);
549 vm_map_process_deferred();
554 _vm_map_lock_read(vm_map_t map, const char *file, int line)
558 mtx_lock_flags_(&map->system_mtx, 0, file, line);
560 sx_slock_(&map->lock, file, line);
564 _vm_map_unlock_read(vm_map_t map, const char *file, int line)
568 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
570 sx_sunlock_(&map->lock, file, line);
571 vm_map_process_deferred();
576 _vm_map_trylock(vm_map_t map, const char *file, int line)
580 error = map->system_map ?
581 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) :
582 !sx_try_xlock_(&map->lock, file, line);
589 _vm_map_trylock_read(vm_map_t map, const char *file, int line)
593 error = map->system_map ?
594 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) :
595 !sx_try_slock_(&map->lock, file, line);
600 * _vm_map_lock_upgrade: [ internal use only ]
602 * Tries to upgrade a read (shared) lock on the specified map to a write
603 * (exclusive) lock. Returns the value "0" if the upgrade succeeds and a
604 * non-zero value if the upgrade fails. If the upgrade fails, the map is
605 * returned without a read or write lock held.
607 * Requires that the map be read locked.
610 _vm_map_lock_upgrade(vm_map_t map, const char *file, int line)
612 unsigned int last_timestamp;
614 if (map->system_map) {
615 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
617 if (!sx_try_upgrade_(&map->lock, file, line)) {
618 last_timestamp = map->timestamp;
619 sx_sunlock_(&map->lock, file, line);
620 vm_map_process_deferred();
622 * If the map's timestamp does not change while the
623 * map is unlocked, then the upgrade succeeds.
625 sx_xlock_(&map->lock, file, line);
626 if (last_timestamp != map->timestamp) {
627 sx_xunlock_(&map->lock, file, line);
637 _vm_map_lock_downgrade(vm_map_t map, const char *file, int line)
640 if (map->system_map) {
641 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
643 sx_downgrade_(&map->lock, file, line);
649 * Returns a non-zero value if the caller holds a write (exclusive) lock
650 * on the specified map and the value "0" otherwise.
653 vm_map_locked(vm_map_t map)
657 return (mtx_owned(&map->system_mtx));
659 return (sx_xlocked(&map->lock));
664 _vm_map_assert_locked(vm_map_t map, const char *file, int line)
668 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
670 sx_assert_(&map->lock, SA_XLOCKED, file, line);
673 #define VM_MAP_ASSERT_LOCKED(map) \
674 _vm_map_assert_locked(map, LOCK_FILE, LOCK_LINE)
676 #define VM_MAP_ASSERT_LOCKED(map)
680 * _vm_map_unlock_and_wait:
682 * Atomically releases the lock on the specified map and puts the calling
683 * thread to sleep. The calling thread will remain asleep until either
684 * vm_map_wakeup() is performed on the map or the specified timeout is
687 * WARNING! This function does not perform deferred deallocations of
688 * objects and map entries. Therefore, the calling thread is expected to
689 * reacquire the map lock after reawakening and later perform an ordinary
690 * unlock operation, such as vm_map_unlock(), before completing its
691 * operation on the map.
694 _vm_map_unlock_and_wait(vm_map_t map, int timo, const char *file, int line)
697 mtx_lock(&map_sleep_mtx);
699 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
701 sx_xunlock_(&map->lock, file, line);
702 return (msleep(&map->root, &map_sleep_mtx, PDROP | PVM, "vmmaps",
709 * Awaken any threads that have slept on the map using
710 * vm_map_unlock_and_wait().
713 vm_map_wakeup(vm_map_t map)
717 * Acquire and release map_sleep_mtx to prevent a wakeup()
718 * from being performed (and lost) between the map unlock
719 * and the msleep() in _vm_map_unlock_and_wait().
721 mtx_lock(&map_sleep_mtx);
722 mtx_unlock(&map_sleep_mtx);
727 vm_map_busy(vm_map_t map)
730 VM_MAP_ASSERT_LOCKED(map);
735 vm_map_unbusy(vm_map_t map)
738 VM_MAP_ASSERT_LOCKED(map);
739 KASSERT(map->busy, ("vm_map_unbusy: not busy"));
740 if (--map->busy == 0 && (map->flags & MAP_BUSY_WAKEUP)) {
741 vm_map_modflags(map, 0, MAP_BUSY_WAKEUP);
747 vm_map_wait_busy(vm_map_t map)
750 VM_MAP_ASSERT_LOCKED(map);
752 vm_map_modflags(map, MAP_BUSY_WAKEUP, 0);
754 msleep(&map->busy, &map->system_mtx, 0, "mbusy", 0);
756 sx_sleep(&map->busy, &map->lock, 0, "mbusy", 0);
762 vmspace_resident_count(struct vmspace *vmspace)
764 return pmap_resident_count(vmspace_pmap(vmspace));
770 * Creates and returns a new empty VM map with
771 * the given physical map structure, and having
772 * the given lower and upper address bounds.
775 vm_map_create(pmap_t pmap, vm_offset_t min, vm_offset_t max)
779 result = uma_zalloc(mapzone, M_WAITOK);
780 CTR1(KTR_VM, "vm_map_create: %p", result);
781 _vm_map_init(result, pmap, min, max);
786 * Initialize an existing vm_map structure
787 * such as that in the vmspace structure.
790 _vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max)
793 map->header.next = map->header.prev = &map->header;
794 map->header.eflags = MAP_ENTRY_HEADER;
795 map->needs_wakeup = FALSE;
798 map->header.end = min;
799 map->header.start = max;
808 vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max)
811 _vm_map_init(map, pmap, min, max);
812 mtx_init(&map->system_mtx, "system map", NULL, MTX_DEF | MTX_DUPOK);
813 sx_init(&map->lock, "user map");
817 * vm_map_entry_dispose: [ internal use only ]
819 * Inverse of vm_map_entry_create.
822 vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry)
824 uma_zfree(map->system_map ? kmapentzone : mapentzone, entry);
828 * vm_map_entry_create: [ internal use only ]
830 * Allocates a VM map entry for insertion.
831 * No entry fields are filled in.
833 static vm_map_entry_t
834 vm_map_entry_create(vm_map_t map)
836 vm_map_entry_t new_entry;
839 new_entry = uma_zalloc(kmapentzone, M_NOWAIT);
841 new_entry = uma_zalloc(mapentzone, M_WAITOK);
842 if (new_entry == NULL)
843 panic("vm_map_entry_create: kernel resources exhausted");
848 * vm_map_entry_set_behavior:
850 * Set the expected access behavior, either normal, random, or
854 vm_map_entry_set_behavior(vm_map_entry_t entry, u_char behavior)
856 entry->eflags = (entry->eflags & ~MAP_ENTRY_BEHAV_MASK) |
857 (behavior & MAP_ENTRY_BEHAV_MASK);
861 * vm_map_entry_set_max_free:
863 * Set the max_free field in a vm_map_entry.
866 vm_map_entry_set_max_free(vm_map_entry_t entry)
869 entry->max_free = entry->adj_free;
870 if (entry->left != NULL && entry->left->max_free > entry->max_free)
871 entry->max_free = entry->left->max_free;
872 if (entry->right != NULL && entry->right->max_free > entry->max_free)
873 entry->max_free = entry->right->max_free;
877 * vm_map_entry_splay:
879 * The Sleator and Tarjan top-down splay algorithm with the
880 * following variation. Max_free must be computed bottom-up, so
881 * on the downward pass, maintain the left and right spines in
882 * reverse order. Then, make a second pass up each side to fix
883 * the pointers and compute max_free. The time bound is O(log n)
886 * The new root is the vm_map_entry containing "addr", or else an
887 * adjacent entry (lower or higher) if addr is not in the tree.
889 * The map must be locked, and leaves it so.
891 * Returns: the new root.
893 static vm_map_entry_t
894 vm_map_entry_splay(vm_offset_t addr, vm_map_entry_t root)
896 vm_map_entry_t llist, rlist;
897 vm_map_entry_t ltree, rtree;
900 /* Special case of empty tree. */
905 * Pass One: Splay down the tree until we find addr or a NULL
906 * pointer where addr would go. llist and rlist are the two
907 * sides in reverse order (bottom-up), with llist linked by
908 * the right pointer and rlist linked by the left pointer in
909 * the vm_map_entry. Wait until Pass Two to set max_free on
915 /* root is never NULL in here. */
916 if (addr < root->start) {
920 if (addr < y->start && y->left != NULL) {
921 /* Rotate right and put y on rlist. */
922 root->left = y->right;
924 vm_map_entry_set_max_free(root);
929 /* Put root on rlist. */
934 } else if (addr >= root->end) {
938 if (addr >= y->end && y->right != NULL) {
939 /* Rotate left and put y on llist. */
940 root->right = y->left;
942 vm_map_entry_set_max_free(root);
947 /* Put root on llist. */
957 * Pass Two: Walk back up the two spines, flip the pointers
958 * and set max_free. The subtrees of the root go at the
959 * bottom of llist and rlist.
962 while (llist != NULL) {
964 llist->right = ltree;
965 vm_map_entry_set_max_free(llist);
970 while (rlist != NULL) {
973 vm_map_entry_set_max_free(rlist);
979 * Final assembly: add ltree and rtree as subtrees of root.
983 vm_map_entry_set_max_free(root);
989 * vm_map_entry_{un,}link:
991 * Insert/remove entries from maps.
994 vm_map_entry_link(vm_map_t map,
995 vm_map_entry_t after_where,
996 vm_map_entry_t entry)
1000 "vm_map_entry_link: map %p, nentries %d, entry %p, after %p", map,
1001 map->nentries, entry, after_where);
1002 VM_MAP_ASSERT_LOCKED(map);
1003 KASSERT(after_where->end <= entry->start,
1004 ("vm_map_entry_link: prev end %jx new start %jx overlap",
1005 (uintmax_t)after_where->end, (uintmax_t)entry->start));
1006 KASSERT(entry->end <= after_where->next->start,
1007 ("vm_map_entry_link: new end %jx next start %jx overlap",
1008 (uintmax_t)entry->end, (uintmax_t)after_where->next->start));
1011 entry->prev = after_where;
1012 entry->next = after_where->next;
1013 entry->next->prev = entry;
1014 after_where->next = entry;
1016 if (after_where != &map->header) {
1017 if (after_where != map->root)
1018 vm_map_entry_splay(after_where->start, map->root);
1019 entry->right = after_where->right;
1020 entry->left = after_where;
1021 after_where->right = NULL;
1022 after_where->adj_free = entry->start - after_where->end;
1023 vm_map_entry_set_max_free(after_where);
1025 entry->right = map->root;
1028 entry->adj_free = entry->next->start - entry->end;
1029 vm_map_entry_set_max_free(entry);
1034 vm_map_entry_unlink(vm_map_t map,
1035 vm_map_entry_t entry)
1037 vm_map_entry_t next, prev, root;
1039 VM_MAP_ASSERT_LOCKED(map);
1040 if (entry != map->root)
1041 vm_map_entry_splay(entry->start, map->root);
1042 if (entry->left == NULL)
1043 root = entry->right;
1045 root = vm_map_entry_splay(entry->start, entry->left);
1046 root->right = entry->right;
1047 root->adj_free = entry->next->start - root->end;
1048 vm_map_entry_set_max_free(root);
1057 CTR3(KTR_VM, "vm_map_entry_unlink: map %p, nentries %d, entry %p", map,
1058 map->nentries, entry);
1062 * vm_map_entry_resize_free:
1064 * Recompute the amount of free space following a vm_map_entry
1065 * and propagate that value up the tree. Call this function after
1066 * resizing a map entry in-place, that is, without a call to
1067 * vm_map_entry_link() or _unlink().
1069 * The map must be locked, and leaves it so.
1072 vm_map_entry_resize_free(vm_map_t map, vm_map_entry_t entry)
1076 * Using splay trees without parent pointers, propagating
1077 * max_free up the tree is done by moving the entry to the
1078 * root and making the change there.
1080 if (entry != map->root)
1081 map->root = vm_map_entry_splay(entry->start, map->root);
1083 entry->adj_free = entry->next->start - entry->end;
1084 vm_map_entry_set_max_free(entry);
1088 * vm_map_lookup_entry: [ internal use only ]
1090 * Finds the map entry containing (or
1091 * immediately preceding) the specified address
1092 * in the given map; the entry is returned
1093 * in the "entry" parameter. The boolean
1094 * result indicates whether the address is
1095 * actually contained in the map.
1098 vm_map_lookup_entry(
1100 vm_offset_t address,
1101 vm_map_entry_t *entry) /* OUT */
1107 * If the map is empty, then the map entry immediately preceding
1108 * "address" is the map's header.
1112 *entry = &map->header;
1113 else if (address >= cur->start && cur->end > address) {
1116 } else if ((locked = vm_map_locked(map)) ||
1117 sx_try_upgrade(&map->lock)) {
1119 * Splay requires a write lock on the map. However, it only
1120 * restructures the binary search tree; it does not otherwise
1121 * change the map. Thus, the map's timestamp need not change
1122 * on a temporary upgrade.
1124 map->root = cur = vm_map_entry_splay(address, cur);
1126 sx_downgrade(&map->lock);
1129 * If "address" is contained within a map entry, the new root
1130 * is that map entry. Otherwise, the new root is a map entry
1131 * immediately before or after "address".
1133 if (address >= cur->start) {
1135 if (cur->end > address)
1141 * Since the map is only locked for read access, perform a
1142 * standard binary search tree lookup for "address".
1145 if (address < cur->start) {
1146 if (cur->left == NULL) {
1151 } else if (cur->end > address) {
1155 if (cur->right == NULL) {
1168 * Inserts the given whole VM object into the target
1169 * map at the specified address range. The object's
1170 * size should match that of the address range.
1172 * Requires that the map be locked, and leaves it so.
1174 * If object is non-NULL, ref count must be bumped by caller
1175 * prior to making call to account for the new entry.
1178 vm_map_insert(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1179 vm_offset_t start, vm_offset_t end, vm_prot_t prot, vm_prot_t max, int cow)
1181 vm_map_entry_t new_entry, prev_entry, temp_entry;
1183 vm_eflags_t protoeflags;
1184 vm_inherit_t inheritance;
1186 VM_MAP_ASSERT_LOCKED(map);
1187 KASSERT(object != kernel_object ||
1188 (cow & MAP_COPY_ON_WRITE) == 0,
1189 ("vm_map_insert: kernel object and COW"));
1190 KASSERT(object == NULL || (cow & MAP_NOFAULT) == 0,
1191 ("vm_map_insert: paradoxical MAP_NOFAULT request"));
1192 KASSERT((prot & ~max) == 0,
1193 ("prot %#x is not subset of max_prot %#x", prot, max));
1196 * Check that the start and end points are not bogus.
1198 if (start < vm_map_min(map) || end > vm_map_max(map) ||
1200 return (KERN_INVALID_ADDRESS);
1203 * Find the entry prior to the proposed starting address; if it's part
1204 * of an existing entry, this range is bogus.
1206 if (vm_map_lookup_entry(map, start, &temp_entry))
1207 return (KERN_NO_SPACE);
1209 prev_entry = temp_entry;
1212 * Assert that the next entry doesn't overlap the end point.
1214 if (prev_entry->next->start < end)
1215 return (KERN_NO_SPACE);
1217 if ((cow & MAP_CREATE_GUARD) != 0 && (object != NULL ||
1218 max != VM_PROT_NONE))
1219 return (KERN_INVALID_ARGUMENT);
1222 if (cow & MAP_COPY_ON_WRITE)
1223 protoeflags |= MAP_ENTRY_COW | MAP_ENTRY_NEEDS_COPY;
1224 if (cow & MAP_NOFAULT)
1225 protoeflags |= MAP_ENTRY_NOFAULT;
1226 if (cow & MAP_DISABLE_SYNCER)
1227 protoeflags |= MAP_ENTRY_NOSYNC;
1228 if (cow & MAP_DISABLE_COREDUMP)
1229 protoeflags |= MAP_ENTRY_NOCOREDUMP;
1230 if (cow & MAP_STACK_GROWS_DOWN)
1231 protoeflags |= MAP_ENTRY_GROWS_DOWN;
1232 if (cow & MAP_STACK_GROWS_UP)
1233 protoeflags |= MAP_ENTRY_GROWS_UP;
1234 if (cow & MAP_VN_WRITECOUNT)
1235 protoeflags |= MAP_ENTRY_VN_WRITECNT;
1236 if ((cow & MAP_CREATE_GUARD) != 0)
1237 protoeflags |= MAP_ENTRY_GUARD;
1238 if ((cow & MAP_CREATE_STACK_GAP_DN) != 0)
1239 protoeflags |= MAP_ENTRY_STACK_GAP_DN;
1240 if ((cow & MAP_CREATE_STACK_GAP_UP) != 0)
1241 protoeflags |= MAP_ENTRY_STACK_GAP_UP;
1242 if (cow & MAP_INHERIT_SHARE)
1243 inheritance = VM_INHERIT_SHARE;
1245 inheritance = VM_INHERIT_DEFAULT;
1248 if ((cow & (MAP_ACC_NO_CHARGE | MAP_NOFAULT | MAP_CREATE_GUARD)) != 0)
1250 if ((cow & MAP_ACC_CHARGED) || ((prot & VM_PROT_WRITE) &&
1251 ((protoeflags & MAP_ENTRY_NEEDS_COPY) || object == NULL))) {
1252 if (!(cow & MAP_ACC_CHARGED) && !swap_reserve(end - start))
1253 return (KERN_RESOURCE_SHORTAGE);
1254 KASSERT(object == NULL ||
1255 (protoeflags & MAP_ENTRY_NEEDS_COPY) != 0 ||
1256 object->cred == NULL,
1257 ("overcommit: vm_map_insert o %p", object));
1258 cred = curthread->td_ucred;
1262 /* Expand the kernel pmap, if necessary. */
1263 if (map == kernel_map && end > kernel_vm_end)
1264 pmap_growkernel(end);
1265 if (object != NULL) {
1267 * OBJ_ONEMAPPING must be cleared unless this mapping
1268 * is trivially proven to be the only mapping for any
1269 * of the object's pages. (Object granularity
1270 * reference counting is insufficient to recognize
1271 * aliases with precision.)
1273 VM_OBJECT_WLOCK(object);
1274 if (object->ref_count > 1 || object->shadow_count != 0)
1275 vm_object_clear_flag(object, OBJ_ONEMAPPING);
1276 VM_OBJECT_WUNLOCK(object);
1277 } else if ((prev_entry->eflags & ~MAP_ENTRY_USER_WIRED) ==
1279 (cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 &&
1280 prev_entry->end == start && (prev_entry->cred == cred ||
1281 (prev_entry->object.vm_object != NULL &&
1282 prev_entry->object.vm_object->cred == cred)) &&
1283 vm_object_coalesce(prev_entry->object.vm_object,
1285 (vm_size_t)(prev_entry->end - prev_entry->start),
1286 (vm_size_t)(end - prev_entry->end), cred != NULL &&
1287 (protoeflags & MAP_ENTRY_NEEDS_COPY) == 0)) {
1289 * We were able to extend the object. Determine if we
1290 * can extend the previous map entry to include the
1291 * new range as well.
1293 if (prev_entry->inheritance == inheritance &&
1294 prev_entry->protection == prot &&
1295 prev_entry->max_protection == max &&
1296 prev_entry->wired_count == 0) {
1297 KASSERT((prev_entry->eflags & MAP_ENTRY_USER_WIRED) ==
1298 0, ("prev_entry %p has incoherent wiring",
1300 if ((prev_entry->eflags & MAP_ENTRY_GUARD) == 0)
1301 map->size += end - prev_entry->end;
1302 prev_entry->end = end;
1303 vm_map_entry_resize_free(map, prev_entry);
1304 vm_map_simplify_entry(map, prev_entry);
1305 return (KERN_SUCCESS);
1309 * If we can extend the object but cannot extend the
1310 * map entry, we have to create a new map entry. We
1311 * must bump the ref count on the extended object to
1312 * account for it. object may be NULL.
1314 object = prev_entry->object.vm_object;
1315 offset = prev_entry->offset +
1316 (prev_entry->end - prev_entry->start);
1317 vm_object_reference(object);
1318 if (cred != NULL && object != NULL && object->cred != NULL &&
1319 !(prev_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
1320 /* Object already accounts for this uid. */
1328 * Create a new entry
1330 new_entry = vm_map_entry_create(map);
1331 new_entry->start = start;
1332 new_entry->end = end;
1333 new_entry->cred = NULL;
1335 new_entry->eflags = protoeflags;
1336 new_entry->object.vm_object = object;
1337 new_entry->offset = offset;
1339 new_entry->inheritance = inheritance;
1340 new_entry->protection = prot;
1341 new_entry->max_protection = max;
1342 new_entry->wired_count = 0;
1343 new_entry->wiring_thread = NULL;
1344 new_entry->read_ahead = VM_FAULT_READ_AHEAD_INIT;
1345 new_entry->next_read = start;
1347 KASSERT(cred == NULL || !ENTRY_CHARGED(new_entry),
1348 ("overcommit: vm_map_insert leaks vm_map %p", new_entry));
1349 new_entry->cred = cred;
1352 * Insert the new entry into the list
1354 vm_map_entry_link(map, prev_entry, new_entry);
1355 if ((new_entry->eflags & MAP_ENTRY_GUARD) == 0)
1356 map->size += new_entry->end - new_entry->start;
1359 * Try to coalesce the new entry with both the previous and next
1360 * entries in the list. Previously, we only attempted to coalesce
1361 * with the previous entry when object is NULL. Here, we handle the
1362 * other cases, which are less common.
1364 vm_map_simplify_entry(map, new_entry);
1366 if ((cow & (MAP_PREFAULT | MAP_PREFAULT_PARTIAL)) != 0) {
1367 vm_map_pmap_enter(map, start, prot, object, OFF_TO_IDX(offset),
1368 end - start, cow & MAP_PREFAULT_PARTIAL);
1371 return (KERN_SUCCESS);
1377 * Find the first fit (lowest VM address) for "length" free bytes
1378 * beginning at address >= start in the given map.
1380 * In a vm_map_entry, "adj_free" is the amount of free space
1381 * adjacent (higher address) to this entry, and "max_free" is the
1382 * maximum amount of contiguous free space in its subtree. This
1383 * allows finding a free region in one path down the tree, so
1384 * O(log n) amortized with splay trees.
1386 * The map must be locked, and leaves it so.
1388 * Returns: 0 on success, and starting address in *addr,
1389 * 1 if insufficient space.
1392 vm_map_findspace(vm_map_t map, vm_offset_t start, vm_size_t length,
1393 vm_offset_t *addr) /* OUT */
1395 vm_map_entry_t entry;
1399 * Request must fit within min/max VM address and must avoid
1402 start = MAX(start, vm_map_min(map));
1403 if (start + length > vm_map_max(map) || start + length < start)
1406 /* Empty tree means wide open address space. */
1407 if (map->root == NULL) {
1413 * After splay, if start comes before root node, then there
1414 * must be a gap from start to the root.
1416 map->root = vm_map_entry_splay(start, map->root);
1417 if (start + length <= map->root->start) {
1423 * Root is the last node that might begin its gap before
1424 * start, and this is the last comparison where address
1425 * wrap might be a problem.
1427 st = (start > map->root->end) ? start : map->root->end;
1428 if (length <= map->root->end + map->root->adj_free - st) {
1433 /* With max_free, can immediately tell if no solution. */
1434 entry = map->root->right;
1435 if (entry == NULL || length > entry->max_free)
1439 * Search the right subtree in the order: left subtree, root,
1440 * right subtree (first fit). The previous splay implies that
1441 * all regions in the right subtree have addresses > start.
1443 while (entry != NULL) {
1444 if (entry->left != NULL && entry->left->max_free >= length)
1445 entry = entry->left;
1446 else if (entry->adj_free >= length) {
1450 entry = entry->right;
1453 /* Can't get here, so panic if we do. */
1454 panic("vm_map_findspace: max_free corrupt");
1458 vm_map_fixed(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1459 vm_offset_t start, vm_size_t length, vm_prot_t prot,
1460 vm_prot_t max, int cow)
1465 end = start + length;
1466 KASSERT((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 ||
1468 ("vm_map_fixed: non-NULL backing object for stack"));
1470 VM_MAP_RANGE_CHECK(map, start, end);
1471 if ((cow & MAP_CHECK_EXCL) == 0)
1472 vm_map_delete(map, start, end);
1473 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) != 0) {
1474 result = vm_map_stack_locked(map, start, length, sgrowsiz,
1477 result = vm_map_insert(map, object, offset, start, end,
1484 static const int aslr_pages_rnd_64[2] = {0x1000, 0x10};
1485 static const int aslr_pages_rnd_32[2] = {0x100, 0x4};
1487 static int cluster_anon = 1;
1488 SYSCTL_INT(_vm, OID_AUTO, cluster_anon, CTLFLAG_RW,
1490 "Cluster anonymous mappings");
1492 static long aslr_restarts;
1493 SYSCTL_LONG(_vm, OID_AUTO, aslr_restarts, CTLFLAG_RD,
1495 "Number of aslr failures");
1497 #define MAP_32BIT_MAX_ADDR ((vm_offset_t)1 << 31)
1500 * Searches for the specified amount of free space in the given map with the
1501 * specified alignment. Performs an address-ordered, first-fit search from
1502 * the given address "*addr", with an optional upper bound "max_addr". If the
1503 * parameter "alignment" is zero, then the alignment is computed from the
1504 * given (object, offset) pair so as to enable the greatest possible use of
1505 * superpage mappings. Returns KERN_SUCCESS and the address of the free space
1506 * in "*addr" if successful. Otherwise, returns KERN_NO_SPACE.
1508 * The map must be locked. Initially, there must be at least "length" bytes
1509 * of free space at the given address.
1512 vm_map_alignspace(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1513 vm_offset_t *addr, vm_size_t length, vm_offset_t max_addr,
1514 vm_offset_t alignment)
1516 vm_offset_t aligned_addr, free_addr;
1518 VM_MAP_ASSERT_LOCKED(map);
1520 KASSERT(!vm_map_findspace(map, free_addr, length, addr) &&
1521 free_addr == *addr, ("caller provided insufficient free space"));
1524 * At the start of every iteration, the free space at address
1525 * "*addr" is at least "length" bytes.
1528 pmap_align_superpage(object, offset, addr, length);
1529 else if ((*addr & (alignment - 1)) != 0) {
1530 *addr &= ~(alignment - 1);
1533 aligned_addr = *addr;
1534 if (aligned_addr == free_addr) {
1536 * Alignment did not change "*addr", so "*addr" must
1537 * still provide sufficient free space.
1539 return (KERN_SUCCESS);
1543 * Test for address wrap on "*addr". A wrapped "*addr" could
1544 * be a valid address, in which case vm_map_findspace() cannot
1545 * be relied upon to fail.
1547 if (aligned_addr < free_addr ||
1548 vm_map_findspace(map, aligned_addr, length, addr) ||
1549 (max_addr != 0 && *addr + length > max_addr))
1550 return (KERN_NO_SPACE);
1552 if (free_addr == aligned_addr) {
1554 * If a successful call to vm_map_findspace() did not
1555 * change "*addr", then "*addr" must still be aligned
1556 * and provide sufficient free space.
1558 return (KERN_SUCCESS);
1564 * vm_map_find finds an unallocated region in the target address
1565 * map with the given length. The search is defined to be
1566 * first-fit from the specified address; the region found is
1567 * returned in the same parameter.
1569 * If object is non-NULL, ref count must be bumped by caller
1570 * prior to making call to account for the new entry.
1573 vm_map_find(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1574 vm_offset_t *addr, /* IN/OUT */
1575 vm_size_t length, vm_offset_t max_addr, int find_space,
1576 vm_prot_t prot, vm_prot_t max, int cow)
1578 vm_offset_t alignment, curr_min_addr, min_addr;
1579 int gap, pidx, rv, try;
1580 bool cluster, en_aslr, update_anon;
1582 KASSERT((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 ||
1584 ("vm_map_find: non-NULL backing object for stack"));
1585 MPASS((cow & MAP_REMAP) == 0 || (find_space == VMFS_NO_SPACE &&
1586 (cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0));
1587 if (find_space == VMFS_OPTIMAL_SPACE && (object == NULL ||
1588 (object->flags & OBJ_COLORED) == 0))
1589 find_space = VMFS_ANY_SPACE;
1590 if (find_space >> 8 != 0) {
1591 KASSERT((find_space & 0xff) == 0, ("bad VMFS flags"));
1592 alignment = (vm_offset_t)1 << (find_space >> 8);
1595 en_aslr = (map->flags & MAP_ASLR) != 0;
1596 update_anon = cluster = cluster_anon != 0 &&
1597 (map->flags & MAP_IS_SUB_MAP) == 0 && max_addr == 0 &&
1598 find_space != VMFS_NO_SPACE && object == NULL &&
1599 (cow & (MAP_INHERIT_SHARE | MAP_STACK_GROWS_UP |
1600 MAP_STACK_GROWS_DOWN)) == 0 && prot != PROT_NONE;
1601 curr_min_addr = min_addr = *addr;
1602 if (en_aslr && min_addr == 0 && !cluster &&
1603 find_space != VMFS_NO_SPACE &&
1604 (map->flags & MAP_ASLR_IGNSTART) != 0)
1605 curr_min_addr = min_addr = vm_map_min(map);
1609 curr_min_addr = map->anon_loc;
1610 if (curr_min_addr == 0)
1613 if (find_space != VMFS_NO_SPACE) {
1614 KASSERT(find_space == VMFS_ANY_SPACE ||
1615 find_space == VMFS_OPTIMAL_SPACE ||
1616 find_space == VMFS_SUPER_SPACE ||
1617 alignment != 0, ("unexpected VMFS flag"));
1620 * When creating an anonymous mapping, try clustering
1621 * with an existing anonymous mapping first.
1623 * We make up to two attempts to find address space
1624 * for a given find_space value. The first attempt may
1625 * apply randomization or may cluster with an existing
1626 * anonymous mapping. If this first attempt fails,
1627 * perform a first-fit search of the available address
1630 * If all tries failed, and find_space is
1631 * VMFS_OPTIMAL_SPACE, fallback to VMFS_ANY_SPACE.
1632 * Again enable clustering and randomization.
1639 * Second try: we failed either to find a
1640 * suitable region for randomizing the
1641 * allocation, or to cluster with an existing
1642 * mapping. Retry with free run.
1644 curr_min_addr = (map->flags & MAP_ASLR_IGNSTART) != 0 ?
1645 vm_map_min(map) : min_addr;
1646 atomic_add_long(&aslr_restarts, 1);
1649 if (try == 1 && en_aslr && !cluster) {
1651 * Find space for allocation, including
1652 * gap needed for later randomization.
1654 pidx = MAXPAGESIZES > 1 && pagesizes[1] != 0 &&
1655 (find_space == VMFS_SUPER_SPACE || find_space ==
1656 VMFS_OPTIMAL_SPACE) ? 1 : 0;
1657 gap = vm_map_max(map) > MAP_32BIT_MAX_ADDR &&
1658 (max_addr == 0 || max_addr > MAP_32BIT_MAX_ADDR) ?
1659 aslr_pages_rnd_64[pidx] : aslr_pages_rnd_32[pidx];
1660 if (vm_map_findspace(map, curr_min_addr, length +
1661 gap * pagesizes[pidx], addr) ||
1662 (max_addr != 0 && *addr + length > max_addr))
1664 /* And randomize the start address. */
1665 *addr += (arc4random() % gap) * pagesizes[pidx];
1666 } else if (vm_map_findspace(map, curr_min_addr, length, addr) ||
1667 (max_addr != 0 && *addr + length > max_addr)) {
1677 if (find_space != VMFS_ANY_SPACE &&
1678 (rv = vm_map_alignspace(map, object, offset, addr, length,
1679 max_addr, alignment)) != KERN_SUCCESS) {
1680 if (find_space == VMFS_OPTIMAL_SPACE) {
1681 find_space = VMFS_ANY_SPACE;
1682 curr_min_addr = min_addr;
1683 cluster = update_anon;
1689 } else if ((cow & MAP_REMAP) != 0) {
1690 if (*addr < vm_map_min(map) ||
1691 *addr + length > vm_map_max(map) ||
1692 *addr + length <= length) {
1693 rv = KERN_INVALID_ADDRESS;
1696 vm_map_delete(map, *addr, *addr + length);
1698 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) != 0) {
1699 rv = vm_map_stack_locked(map, *addr, length, sgrowsiz, prot,
1702 rv = vm_map_insert(map, object, offset, *addr, *addr + length,
1705 if (rv == KERN_SUCCESS && update_anon)
1706 map->anon_loc = *addr + length;
1713 * vm_map_find_min() is a variant of vm_map_find() that takes an
1714 * additional parameter (min_addr) and treats the given address
1715 * (*addr) differently. Specifically, it treats *addr as a hint
1716 * and not as the minimum address where the mapping is created.
1718 * This function works in two phases. First, it tries to
1719 * allocate above the hint. If that fails and the hint is
1720 * greater than min_addr, it performs a second pass, replacing
1721 * the hint with min_addr as the minimum address for the
1725 vm_map_find_min(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1726 vm_offset_t *addr, vm_size_t length, vm_offset_t min_addr,
1727 vm_offset_t max_addr, int find_space, vm_prot_t prot, vm_prot_t max,
1735 rv = vm_map_find(map, object, offset, addr, length, max_addr,
1736 find_space, prot, max, cow);
1737 if (rv == KERN_SUCCESS || min_addr >= hint)
1739 *addr = hint = min_addr;
1744 * A map entry with any of the following flags set must not be merged with
1747 #define MAP_ENTRY_NOMERGE_MASK (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP | \
1748 MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_IS_SUB_MAP)
1751 vm_map_mergeable_neighbors(vm_map_entry_t prev, vm_map_entry_t entry)
1754 KASSERT((prev->eflags & MAP_ENTRY_NOMERGE_MASK) == 0 ||
1755 (entry->eflags & MAP_ENTRY_NOMERGE_MASK) == 0,
1756 ("vm_map_mergeable_neighbors: neither %p nor %p are mergeable",
1758 return (prev->end == entry->start &&
1759 prev->object.vm_object == entry->object.vm_object &&
1760 (prev->object.vm_object == NULL ||
1761 prev->offset + (prev->end - prev->start) == entry->offset) &&
1762 prev->eflags == entry->eflags &&
1763 prev->protection == entry->protection &&
1764 prev->max_protection == entry->max_protection &&
1765 prev->inheritance == entry->inheritance &&
1766 prev->wired_count == entry->wired_count &&
1767 prev->cred == entry->cred);
1771 vm_map_merged_neighbor_dispose(vm_map_t map, vm_map_entry_t entry)
1775 * If the backing object is a vnode object, vm_object_deallocate()
1776 * calls vrele(). However, vrele() does not lock the vnode because
1777 * the vnode has additional references. Thus, the map lock can be
1778 * kept without causing a lock-order reversal with the vnode lock.
1780 * Since we count the number of virtual page mappings in
1781 * object->un_pager.vnp.writemappings, the writemappings value
1782 * should not be adjusted when the entry is disposed of.
1784 if (entry->object.vm_object != NULL)
1785 vm_object_deallocate(entry->object.vm_object);
1786 if (entry->cred != NULL)
1787 crfree(entry->cred);
1788 vm_map_entry_dispose(map, entry);
1792 * vm_map_simplify_entry:
1794 * Simplify the given map entry by merging with either neighbor. This
1795 * routine also has the ability to merge with both neighbors.
1797 * The map must be locked.
1799 * This routine guarantees that the passed entry remains valid (though
1800 * possibly extended). When merging, this routine may delete one or
1804 vm_map_simplify_entry(vm_map_t map, vm_map_entry_t entry)
1806 vm_map_entry_t next, prev;
1808 if ((entry->eflags & MAP_ENTRY_NOMERGE_MASK) != 0)
1811 if (vm_map_mergeable_neighbors(prev, entry)) {
1812 vm_map_entry_unlink(map, prev);
1813 entry->start = prev->start;
1814 entry->offset = prev->offset;
1815 if (entry->prev != &map->header)
1816 vm_map_entry_resize_free(map, entry->prev);
1817 vm_map_merged_neighbor_dispose(map, prev);
1820 if (vm_map_mergeable_neighbors(entry, next)) {
1821 vm_map_entry_unlink(map, next);
1822 entry->end = next->end;
1823 vm_map_entry_resize_free(map, entry);
1824 vm_map_merged_neighbor_dispose(map, next);
1829 * vm_map_clip_start: [ internal use only ]
1831 * Asserts that the given entry begins at or after
1832 * the specified address; if necessary,
1833 * it splits the entry into two.
1835 #define vm_map_clip_start(map, entry, startaddr) \
1837 if (startaddr > entry->start) \
1838 _vm_map_clip_start(map, entry, startaddr); \
1842 * This routine is called only when it is known that
1843 * the entry must be split.
1846 _vm_map_clip_start(vm_map_t map, vm_map_entry_t entry, vm_offset_t start)
1848 vm_map_entry_t new_entry;
1850 VM_MAP_ASSERT_LOCKED(map);
1851 KASSERT(entry->end > start && entry->start < start,
1852 ("_vm_map_clip_start: invalid clip of entry %p", entry));
1855 * Split off the front portion -- note that we must insert the new
1856 * entry BEFORE this one, so that this entry has the specified
1859 vm_map_simplify_entry(map, entry);
1862 * If there is no object backing this entry, we might as well create
1863 * one now. If we defer it, an object can get created after the map
1864 * is clipped, and individual objects will be created for the split-up
1865 * map. This is a bit of a hack, but is also about the best place to
1866 * put this improvement.
1868 if (entry->object.vm_object == NULL && !map->system_map &&
1869 (entry->eflags & MAP_ENTRY_GUARD) == 0) {
1871 object = vm_object_allocate(OBJT_DEFAULT,
1872 atop(entry->end - entry->start));
1873 entry->object.vm_object = object;
1875 if (entry->cred != NULL) {
1876 object->cred = entry->cred;
1877 object->charge = entry->end - entry->start;
1880 } else if (entry->object.vm_object != NULL &&
1881 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) &&
1882 entry->cred != NULL) {
1883 VM_OBJECT_WLOCK(entry->object.vm_object);
1884 KASSERT(entry->object.vm_object->cred == NULL,
1885 ("OVERCOMMIT: vm_entry_clip_start: both cred e %p", entry));
1886 entry->object.vm_object->cred = entry->cred;
1887 entry->object.vm_object->charge = entry->end - entry->start;
1888 VM_OBJECT_WUNLOCK(entry->object.vm_object);
1892 new_entry = vm_map_entry_create(map);
1893 *new_entry = *entry;
1895 new_entry->end = start;
1896 entry->offset += (start - entry->start);
1897 entry->start = start;
1898 if (new_entry->cred != NULL)
1899 crhold(entry->cred);
1901 vm_map_entry_link(map, entry->prev, new_entry);
1903 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1904 vm_object_reference(new_entry->object.vm_object);
1906 * The object->un_pager.vnp.writemappings for the
1907 * object of MAP_ENTRY_VN_WRITECNT type entry shall be
1908 * kept as is here. The virtual pages are
1909 * re-distributed among the clipped entries, so the sum is
1916 * vm_map_clip_end: [ internal use only ]
1918 * Asserts that the given entry ends at or before
1919 * the specified address; if necessary,
1920 * it splits the entry into two.
1922 #define vm_map_clip_end(map, entry, endaddr) \
1924 if ((endaddr) < (entry->end)) \
1925 _vm_map_clip_end((map), (entry), (endaddr)); \
1929 * This routine is called only when it is known that
1930 * the entry must be split.
1933 _vm_map_clip_end(vm_map_t map, vm_map_entry_t entry, vm_offset_t end)
1935 vm_map_entry_t new_entry;
1937 VM_MAP_ASSERT_LOCKED(map);
1938 KASSERT(entry->start < end && entry->end > end,
1939 ("_vm_map_clip_end: invalid clip of entry %p", entry));
1942 * If there is no object backing this entry, we might as well create
1943 * one now. If we defer it, an object can get created after the map
1944 * is clipped, and individual objects will be created for the split-up
1945 * map. This is a bit of a hack, but is also about the best place to
1946 * put this improvement.
1948 if (entry->object.vm_object == NULL && !map->system_map &&
1949 (entry->eflags & MAP_ENTRY_GUARD) == 0) {
1951 object = vm_object_allocate(OBJT_DEFAULT,
1952 atop(entry->end - entry->start));
1953 entry->object.vm_object = object;
1955 if (entry->cred != NULL) {
1956 object->cred = entry->cred;
1957 object->charge = entry->end - entry->start;
1960 } else if (entry->object.vm_object != NULL &&
1961 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) &&
1962 entry->cred != NULL) {
1963 VM_OBJECT_WLOCK(entry->object.vm_object);
1964 KASSERT(entry->object.vm_object->cred == NULL,
1965 ("OVERCOMMIT: vm_entry_clip_end: both cred e %p", entry));
1966 entry->object.vm_object->cred = entry->cred;
1967 entry->object.vm_object->charge = entry->end - entry->start;
1968 VM_OBJECT_WUNLOCK(entry->object.vm_object);
1973 * Create a new entry and insert it AFTER the specified entry
1975 new_entry = vm_map_entry_create(map);
1976 *new_entry = *entry;
1978 new_entry->start = entry->end = end;
1979 new_entry->offset += (end - entry->start);
1980 if (new_entry->cred != NULL)
1981 crhold(entry->cred);
1983 vm_map_entry_link(map, entry, new_entry);
1985 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1986 vm_object_reference(new_entry->object.vm_object);
1991 * vm_map_submap: [ kernel use only ]
1993 * Mark the given range as handled by a subordinate map.
1995 * This range must have been created with vm_map_find,
1996 * and no other operations may have been performed on this
1997 * range prior to calling vm_map_submap.
1999 * Only a limited number of operations can be performed
2000 * within this rage after calling vm_map_submap:
2002 * [Don't try vm_map_copy!]
2004 * To remove a submapping, one must first remove the
2005 * range from the superior map, and then destroy the
2006 * submap (if desired). [Better yet, don't try it.]
2015 vm_map_entry_t entry;
2018 result = KERN_INVALID_ARGUMENT;
2020 vm_map_lock(submap);
2021 submap->flags |= MAP_IS_SUB_MAP;
2022 vm_map_unlock(submap);
2026 VM_MAP_RANGE_CHECK(map, start, end);
2028 if (vm_map_lookup_entry(map, start, &entry)) {
2029 vm_map_clip_start(map, entry, start);
2031 entry = entry->next;
2033 vm_map_clip_end(map, entry, end);
2035 if ((entry->start == start) && (entry->end == end) &&
2036 ((entry->eflags & MAP_ENTRY_COW) == 0) &&
2037 (entry->object.vm_object == NULL)) {
2038 entry->object.sub_map = submap;
2039 entry->eflags |= MAP_ENTRY_IS_SUB_MAP;
2040 result = KERN_SUCCESS;
2044 if (result != KERN_SUCCESS) {
2045 vm_map_lock(submap);
2046 submap->flags &= ~MAP_IS_SUB_MAP;
2047 vm_map_unlock(submap);
2053 * The maximum number of pages to map if MAP_PREFAULT_PARTIAL is specified
2055 #define MAX_INIT_PT 96
2058 * vm_map_pmap_enter:
2060 * Preload the specified map's pmap with mappings to the specified
2061 * object's memory-resident pages. No further physical pages are
2062 * allocated, and no further virtual pages are retrieved from secondary
2063 * storage. If the specified flags include MAP_PREFAULT_PARTIAL, then a
2064 * limited number of page mappings are created at the low-end of the
2065 * specified address range. (For this purpose, a superpage mapping
2066 * counts as one page mapping.) Otherwise, all resident pages within
2067 * the specified address range are mapped.
2070 vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot,
2071 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags)
2074 vm_page_t p, p_start;
2075 vm_pindex_t mask, psize, threshold, tmpidx;
2077 if ((prot & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0 || object == NULL)
2079 VM_OBJECT_RLOCK(object);
2080 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) {
2081 VM_OBJECT_RUNLOCK(object);
2082 VM_OBJECT_WLOCK(object);
2083 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) {
2084 pmap_object_init_pt(map->pmap, addr, object, pindex,
2086 VM_OBJECT_WUNLOCK(object);
2089 VM_OBJECT_LOCK_DOWNGRADE(object);
2093 if (psize + pindex > object->size) {
2094 if (object->size < pindex) {
2095 VM_OBJECT_RUNLOCK(object);
2098 psize = object->size - pindex;
2103 threshold = MAX_INIT_PT;
2105 p = vm_page_find_least(object, pindex);
2107 * Assert: the variable p is either (1) the page with the
2108 * least pindex greater than or equal to the parameter pindex
2112 p != NULL && (tmpidx = p->pindex - pindex) < psize;
2113 p = TAILQ_NEXT(p, listq)) {
2115 * don't allow an madvise to blow away our really
2116 * free pages allocating pv entries.
2118 if (((flags & MAP_PREFAULT_MADVISE) != 0 &&
2119 vm_page_count_severe()) ||
2120 ((flags & MAP_PREFAULT_PARTIAL) != 0 &&
2121 tmpidx >= threshold)) {
2125 if (p->valid == VM_PAGE_BITS_ALL) {
2126 if (p_start == NULL) {
2127 start = addr + ptoa(tmpidx);
2130 /* Jump ahead if a superpage mapping is possible. */
2131 if (p->psind > 0 && ((addr + ptoa(tmpidx)) &
2132 (pagesizes[p->psind] - 1)) == 0) {
2133 mask = atop(pagesizes[p->psind]) - 1;
2134 if (tmpidx + mask < psize &&
2135 vm_page_ps_test(p, PS_ALL_VALID, NULL)) {
2140 } else if (p_start != NULL) {
2141 pmap_enter_object(map->pmap, start, addr +
2142 ptoa(tmpidx), p_start, prot);
2146 if (p_start != NULL)
2147 pmap_enter_object(map->pmap, start, addr + ptoa(psize),
2149 VM_OBJECT_RUNLOCK(object);
2155 * Sets the protection of the specified address
2156 * region in the target map. If "set_max" is
2157 * specified, the maximum protection is to be set;
2158 * otherwise, only the current protection is affected.
2161 vm_map_protect(vm_map_t map, vm_offset_t start, vm_offset_t end,
2162 vm_prot_t new_prot, boolean_t set_max)
2164 vm_map_entry_t current, entry;
2170 return (KERN_SUCCESS);
2175 * Ensure that we are not concurrently wiring pages. vm_map_wire() may
2176 * need to fault pages into the map and will drop the map lock while
2177 * doing so, and the VM object may end up in an inconsistent state if we
2178 * update the protection on the map entry in between faults.
2180 vm_map_wait_busy(map);
2182 VM_MAP_RANGE_CHECK(map, start, end);
2184 if (vm_map_lookup_entry(map, start, &entry)) {
2185 vm_map_clip_start(map, entry, start);
2187 entry = entry->next;
2191 * Make a first pass to check for protection violations.
2193 for (current = entry; current->start < end; current = current->next) {
2194 if ((current->eflags & MAP_ENTRY_GUARD) != 0)
2196 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
2198 return (KERN_INVALID_ARGUMENT);
2200 if ((new_prot & current->max_protection) != new_prot) {
2202 return (KERN_PROTECTION_FAILURE);
2207 * Do an accounting pass for private read-only mappings that
2208 * now will do cow due to allowed write (e.g. debugger sets
2209 * breakpoint on text segment)
2211 for (current = entry; current->start < end; current = current->next) {
2213 vm_map_clip_end(map, current, end);
2216 ((new_prot & ~(current->protection)) & VM_PROT_WRITE) == 0 ||
2217 ENTRY_CHARGED(current) ||
2218 (current->eflags & MAP_ENTRY_GUARD) != 0) {
2222 cred = curthread->td_ucred;
2223 obj = current->object.vm_object;
2225 if (obj == NULL || (current->eflags & MAP_ENTRY_NEEDS_COPY)) {
2226 if (!swap_reserve(current->end - current->start)) {
2228 return (KERN_RESOURCE_SHORTAGE);
2231 current->cred = cred;
2235 VM_OBJECT_WLOCK(obj);
2236 if (obj->type != OBJT_DEFAULT && obj->type != OBJT_SWAP) {
2237 VM_OBJECT_WUNLOCK(obj);
2242 * Charge for the whole object allocation now, since
2243 * we cannot distinguish between non-charged and
2244 * charged clipped mapping of the same object later.
2246 KASSERT(obj->charge == 0,
2247 ("vm_map_protect: object %p overcharged (entry %p)",
2249 if (!swap_reserve(ptoa(obj->size))) {
2250 VM_OBJECT_WUNLOCK(obj);
2252 return (KERN_RESOURCE_SHORTAGE);
2257 obj->charge = ptoa(obj->size);
2258 VM_OBJECT_WUNLOCK(obj);
2262 * Go back and fix up protections. [Note that clipping is not
2263 * necessary the second time.]
2265 for (current = entry; current->start < end; current = current->next) {
2266 if ((current->eflags & MAP_ENTRY_GUARD) != 0)
2269 old_prot = current->protection;
2272 current->protection =
2273 (current->max_protection = new_prot) &
2276 current->protection = new_prot;
2279 * For user wired map entries, the normal lazy evaluation of
2280 * write access upgrades through soft page faults is
2281 * undesirable. Instead, immediately copy any pages that are
2282 * copy-on-write and enable write access in the physical map.
2284 if ((current->eflags & MAP_ENTRY_USER_WIRED) != 0 &&
2285 (current->protection & VM_PROT_WRITE) != 0 &&
2286 (old_prot & VM_PROT_WRITE) == 0)
2287 vm_fault_copy_entry(map, map, current, current, NULL);
2290 * When restricting access, update the physical map. Worry
2291 * about copy-on-write here.
2293 if ((old_prot & ~current->protection) != 0) {
2294 #define MASK(entry) (((entry)->eflags & MAP_ENTRY_COW) ? ~VM_PROT_WRITE : \
2296 pmap_protect(map->pmap, current->start,
2298 current->protection & MASK(current));
2301 vm_map_simplify_entry(map, current);
2304 return (KERN_SUCCESS);
2310 * This routine traverses a processes map handling the madvise
2311 * system call. Advisories are classified as either those effecting
2312 * the vm_map_entry structure, or those effecting the underlying
2322 vm_map_entry_t current, entry;
2326 * Some madvise calls directly modify the vm_map_entry, in which case
2327 * we need to use an exclusive lock on the map and we need to perform
2328 * various clipping operations. Otherwise we only need a read-lock
2333 case MADV_SEQUENTIAL:
2350 vm_map_lock_read(map);
2357 * Locate starting entry and clip if necessary.
2359 VM_MAP_RANGE_CHECK(map, start, end);
2361 if (vm_map_lookup_entry(map, start, &entry)) {
2363 vm_map_clip_start(map, entry, start);
2365 entry = entry->next;
2370 * madvise behaviors that are implemented in the vm_map_entry.
2372 * We clip the vm_map_entry so that behavioral changes are
2373 * limited to the specified address range.
2375 for (current = entry; current->start < end;
2376 current = current->next) {
2377 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
2380 vm_map_clip_end(map, current, end);
2384 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_NORMAL);
2386 case MADV_SEQUENTIAL:
2387 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_SEQUENTIAL);
2390 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_RANDOM);
2393 current->eflags |= MAP_ENTRY_NOSYNC;
2396 current->eflags &= ~MAP_ENTRY_NOSYNC;
2399 current->eflags |= MAP_ENTRY_NOCOREDUMP;
2402 current->eflags &= ~MAP_ENTRY_NOCOREDUMP;
2407 vm_map_simplify_entry(map, current);
2411 vm_pindex_t pstart, pend;
2414 * madvise behaviors that are implemented in the underlying
2417 * Since we don't clip the vm_map_entry, we have to clip
2418 * the vm_object pindex and count.
2420 for (current = entry; current->start < end;
2421 current = current->next) {
2422 vm_offset_t useEnd, useStart;
2424 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
2427 pstart = OFF_TO_IDX(current->offset);
2428 pend = pstart + atop(current->end - current->start);
2429 useStart = current->start;
2430 useEnd = current->end;
2432 if (current->start < start) {
2433 pstart += atop(start - current->start);
2436 if (current->end > end) {
2437 pend -= atop(current->end - end);
2445 * Perform the pmap_advise() before clearing
2446 * PGA_REFERENCED in vm_page_advise(). Otherwise, a
2447 * concurrent pmap operation, such as pmap_remove(),
2448 * could clear a reference in the pmap and set
2449 * PGA_REFERENCED on the page before the pmap_advise()
2450 * had completed. Consequently, the page would appear
2451 * referenced based upon an old reference that
2452 * occurred before this pmap_advise() ran.
2454 if (behav == MADV_DONTNEED || behav == MADV_FREE)
2455 pmap_advise(map->pmap, useStart, useEnd,
2458 vm_object_madvise(current->object.vm_object, pstart,
2462 * Pre-populate paging structures in the
2463 * WILLNEED case. For wired entries, the
2464 * paging structures are already populated.
2466 if (behav == MADV_WILLNEED &&
2467 current->wired_count == 0) {
2468 vm_map_pmap_enter(map,
2470 current->protection,
2471 current->object.vm_object,
2473 ptoa(pend - pstart),
2474 MAP_PREFAULT_MADVISE
2478 vm_map_unlock_read(map);
2487 * Sets the inheritance of the specified address
2488 * range in the target map. Inheritance
2489 * affects how the map will be shared with
2490 * child maps at the time of vmspace_fork.
2493 vm_map_inherit(vm_map_t map, vm_offset_t start, vm_offset_t end,
2494 vm_inherit_t new_inheritance)
2496 vm_map_entry_t entry;
2497 vm_map_entry_t temp_entry;
2499 switch (new_inheritance) {
2500 case VM_INHERIT_NONE:
2501 case VM_INHERIT_COPY:
2502 case VM_INHERIT_SHARE:
2503 case VM_INHERIT_ZERO:
2506 return (KERN_INVALID_ARGUMENT);
2509 return (KERN_SUCCESS);
2511 VM_MAP_RANGE_CHECK(map, start, end);
2512 if (vm_map_lookup_entry(map, start, &temp_entry)) {
2514 vm_map_clip_start(map, entry, start);
2516 entry = temp_entry->next;
2517 while (entry->start < end) {
2518 vm_map_clip_end(map, entry, end);
2519 if ((entry->eflags & MAP_ENTRY_GUARD) == 0 ||
2520 new_inheritance != VM_INHERIT_ZERO)
2521 entry->inheritance = new_inheritance;
2522 vm_map_simplify_entry(map, entry);
2523 entry = entry->next;
2526 return (KERN_SUCCESS);
2532 * Implements both kernel and user unwiring.
2535 vm_map_unwire(vm_map_t map, vm_offset_t start, vm_offset_t end,
2538 vm_map_entry_t entry, first_entry, tmp_entry;
2539 vm_offset_t saved_start;
2540 unsigned int last_timestamp;
2542 boolean_t need_wakeup, result, user_unwire;
2545 return (KERN_SUCCESS);
2546 user_unwire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE;
2548 VM_MAP_RANGE_CHECK(map, start, end);
2549 if (!vm_map_lookup_entry(map, start, &first_entry)) {
2550 if (flags & VM_MAP_WIRE_HOLESOK)
2551 first_entry = first_entry->next;
2554 return (KERN_INVALID_ADDRESS);
2557 last_timestamp = map->timestamp;
2558 entry = first_entry;
2559 while (entry->start < end) {
2560 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
2562 * We have not yet clipped the entry.
2564 saved_start = (start >= entry->start) ? start :
2566 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2567 if (vm_map_unlock_and_wait(map, 0)) {
2569 * Allow interruption of user unwiring?
2573 if (last_timestamp+1 != map->timestamp) {
2575 * Look again for the entry because the map was
2576 * modified while it was unlocked.
2577 * Specifically, the entry may have been
2578 * clipped, merged, or deleted.
2580 if (!vm_map_lookup_entry(map, saved_start,
2582 if (flags & VM_MAP_WIRE_HOLESOK)
2583 tmp_entry = tmp_entry->next;
2585 if (saved_start == start) {
2587 * First_entry has been deleted.
2590 return (KERN_INVALID_ADDRESS);
2593 rv = KERN_INVALID_ADDRESS;
2597 if (entry == first_entry)
2598 first_entry = tmp_entry;
2603 last_timestamp = map->timestamp;
2606 vm_map_clip_start(map, entry, start);
2607 vm_map_clip_end(map, entry, end);
2609 * Mark the entry in case the map lock is released. (See
2612 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 &&
2613 entry->wiring_thread == NULL,
2614 ("owned map entry %p", entry));
2615 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
2616 entry->wiring_thread = curthread;
2618 * Check the map for holes in the specified region.
2619 * If VM_MAP_WIRE_HOLESOK was specified, skip this check.
2621 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) &&
2622 (entry->end < end && entry->next->start > entry->end)) {
2624 rv = KERN_INVALID_ADDRESS;
2628 * If system unwiring, require that the entry is system wired.
2631 vm_map_entry_system_wired_count(entry) == 0) {
2633 rv = KERN_INVALID_ARGUMENT;
2636 entry = entry->next;
2640 need_wakeup = FALSE;
2641 if (first_entry == NULL) {
2642 result = vm_map_lookup_entry(map, start, &first_entry);
2643 if (!result && (flags & VM_MAP_WIRE_HOLESOK))
2644 first_entry = first_entry->next;
2646 KASSERT(result, ("vm_map_unwire: lookup failed"));
2648 for (entry = first_entry; entry->start < end; entry = entry->next) {
2650 * If VM_MAP_WIRE_HOLESOK was specified, an empty
2651 * space in the unwired region could have been mapped
2652 * while the map lock was dropped for draining
2653 * MAP_ENTRY_IN_TRANSITION. Moreover, another thread
2654 * could be simultaneously wiring this new mapping
2655 * entry. Detect these cases and skip any entries
2656 * marked as in transition by us.
2658 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 ||
2659 entry->wiring_thread != curthread) {
2660 KASSERT((flags & VM_MAP_WIRE_HOLESOK) != 0,
2661 ("vm_map_unwire: !HOLESOK and new/changed entry"));
2665 if (rv == KERN_SUCCESS && (!user_unwire ||
2666 (entry->eflags & MAP_ENTRY_USER_WIRED))) {
2668 entry->eflags &= ~MAP_ENTRY_USER_WIRED;
2669 if (entry->wired_count == 1)
2670 vm_map_entry_unwire(map, entry);
2672 entry->wired_count--;
2674 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0,
2675 ("vm_map_unwire: in-transition flag missing %p", entry));
2676 KASSERT(entry->wiring_thread == curthread,
2677 ("vm_map_unwire: alien wire %p", entry));
2678 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION;
2679 entry->wiring_thread = NULL;
2680 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
2681 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
2684 vm_map_simplify_entry(map, entry);
2693 * vm_map_wire_entry_failure:
2695 * Handle a wiring failure on the given entry.
2697 * The map should be locked.
2700 vm_map_wire_entry_failure(vm_map_t map, vm_map_entry_t entry,
2701 vm_offset_t failed_addr)
2704 VM_MAP_ASSERT_LOCKED(map);
2705 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 &&
2706 entry->wired_count == 1,
2707 ("vm_map_wire_entry_failure: entry %p isn't being wired", entry));
2708 KASSERT(failed_addr < entry->end,
2709 ("vm_map_wire_entry_failure: entry %p was fully wired", entry));
2712 * If any pages at the start of this entry were successfully wired,
2715 if (failed_addr > entry->start) {
2716 pmap_unwire(map->pmap, entry->start, failed_addr);
2717 vm_object_unwire(entry->object.vm_object, entry->offset,
2718 failed_addr - entry->start, PQ_ACTIVE);
2722 * Assign an out-of-range value to represent the failure to wire this
2725 entry->wired_count = -1;
2731 * Implements both kernel and user wiring.
2734 vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t end,
2737 vm_map_entry_t entry, first_entry, tmp_entry;
2738 vm_offset_t faddr, saved_end, saved_start;
2739 unsigned int last_timestamp;
2741 boolean_t need_wakeup, result, user_wire;
2745 return (KERN_SUCCESS);
2747 if (flags & VM_MAP_WIRE_WRITE)
2748 prot |= VM_PROT_WRITE;
2749 user_wire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE;
2751 VM_MAP_RANGE_CHECK(map, start, end);
2752 if (!vm_map_lookup_entry(map, start, &first_entry)) {
2753 if (flags & VM_MAP_WIRE_HOLESOK)
2754 first_entry = first_entry->next;
2757 return (KERN_INVALID_ADDRESS);
2760 last_timestamp = map->timestamp;
2761 entry = first_entry;
2762 while (entry->start < end) {
2763 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
2765 * We have not yet clipped the entry.
2767 saved_start = (start >= entry->start) ? start :
2769 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2770 if (vm_map_unlock_and_wait(map, 0)) {
2772 * Allow interruption of user wiring?
2776 if (last_timestamp + 1 != map->timestamp) {
2778 * Look again for the entry because the map was
2779 * modified while it was unlocked.
2780 * Specifically, the entry may have been
2781 * clipped, merged, or deleted.
2783 if (!vm_map_lookup_entry(map, saved_start,
2785 if (flags & VM_MAP_WIRE_HOLESOK)
2786 tmp_entry = tmp_entry->next;
2788 if (saved_start == start) {
2790 * first_entry has been deleted.
2793 return (KERN_INVALID_ADDRESS);
2796 rv = KERN_INVALID_ADDRESS;
2800 if (entry == first_entry)
2801 first_entry = tmp_entry;
2806 last_timestamp = map->timestamp;
2809 vm_map_clip_start(map, entry, start);
2810 vm_map_clip_end(map, entry, end);
2812 * Mark the entry in case the map lock is released. (See
2815 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 &&
2816 entry->wiring_thread == NULL,
2817 ("owned map entry %p", entry));
2818 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
2819 entry->wiring_thread = curthread;
2820 if ((entry->protection & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0
2821 || (entry->protection & prot) != prot) {
2822 entry->eflags |= MAP_ENTRY_WIRE_SKIPPED;
2823 if ((flags & VM_MAP_WIRE_HOLESOK) == 0) {
2825 rv = KERN_INVALID_ADDRESS;
2830 if (entry->wired_count == 0) {
2831 entry->wired_count++;
2832 saved_start = entry->start;
2833 saved_end = entry->end;
2836 * Release the map lock, relying on the in-transition
2837 * mark. Mark the map busy for fork.
2842 faddr = saved_start;
2845 * Simulate a fault to get the page and enter
2846 * it into the physical map.
2848 if ((rv = vm_fault(map, faddr, VM_PROT_NONE,
2849 VM_FAULT_WIRE)) != KERN_SUCCESS)
2851 } while ((faddr += PAGE_SIZE) < saved_end);
2854 if (last_timestamp + 1 != map->timestamp) {
2856 * Look again for the entry because the map was
2857 * modified while it was unlocked. The entry
2858 * may have been clipped, but NOT merged or
2861 result = vm_map_lookup_entry(map, saved_start,
2863 KASSERT(result, ("vm_map_wire: lookup failed"));
2864 if (entry == first_entry)
2865 first_entry = tmp_entry;
2869 while (entry->end < saved_end) {
2871 * In case of failure, handle entries
2872 * that were not fully wired here;
2873 * fully wired entries are handled
2876 if (rv != KERN_SUCCESS &&
2878 vm_map_wire_entry_failure(map,
2880 entry = entry->next;
2883 last_timestamp = map->timestamp;
2884 if (rv != KERN_SUCCESS) {
2885 vm_map_wire_entry_failure(map, entry, faddr);
2889 } else if (!user_wire ||
2890 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
2891 entry->wired_count++;
2894 * Check the map for holes in the specified region.
2895 * If VM_MAP_WIRE_HOLESOK was specified, skip this check.
2898 if ((flags & VM_MAP_WIRE_HOLESOK) == 0 &&
2899 entry->end < end && entry->next->start > entry->end) {
2901 rv = KERN_INVALID_ADDRESS;
2904 entry = entry->next;
2908 need_wakeup = FALSE;
2909 if (first_entry == NULL) {
2910 result = vm_map_lookup_entry(map, start, &first_entry);
2911 if (!result && (flags & VM_MAP_WIRE_HOLESOK))
2912 first_entry = first_entry->next;
2914 KASSERT(result, ("vm_map_wire: lookup failed"));
2916 for (entry = first_entry; entry->start < end; entry = entry->next) {
2918 * If VM_MAP_WIRE_HOLESOK was specified, an empty
2919 * space in the unwired region could have been mapped
2920 * while the map lock was dropped for faulting in the
2921 * pages or draining MAP_ENTRY_IN_TRANSITION.
2922 * Moreover, another thread could be simultaneously
2923 * wiring this new mapping entry. Detect these cases
2924 * and skip any entries marked as in transition not by us.
2926 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 ||
2927 entry->wiring_thread != curthread) {
2928 KASSERT((flags & VM_MAP_WIRE_HOLESOK) != 0,
2929 ("vm_map_wire: !HOLESOK and new/changed entry"));
2933 if ((entry->eflags & MAP_ENTRY_WIRE_SKIPPED) != 0)
2934 goto next_entry_done;
2936 if (rv == KERN_SUCCESS) {
2938 entry->eflags |= MAP_ENTRY_USER_WIRED;
2939 } else if (entry->wired_count == -1) {
2941 * Wiring failed on this entry. Thus, unwiring is
2944 entry->wired_count = 0;
2945 } else if (!user_wire ||
2946 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
2948 * Undo the wiring. Wiring succeeded on this entry
2949 * but failed on a later entry.
2951 if (entry->wired_count == 1)
2952 vm_map_entry_unwire(map, entry);
2954 entry->wired_count--;
2957 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0,
2958 ("vm_map_wire: in-transition flag missing %p", entry));
2959 KASSERT(entry->wiring_thread == curthread,
2960 ("vm_map_wire: alien wire %p", entry));
2961 entry->eflags &= ~(MAP_ENTRY_IN_TRANSITION |
2962 MAP_ENTRY_WIRE_SKIPPED);
2963 entry->wiring_thread = NULL;
2964 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
2965 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
2968 vm_map_simplify_entry(map, entry);
2979 * Push any dirty cached pages in the address range to their pager.
2980 * If syncio is TRUE, dirty pages are written synchronously.
2981 * If invalidate is TRUE, any cached pages are freed as well.
2983 * If the size of the region from start to end is zero, we are
2984 * supposed to flush all modified pages within the region containing
2985 * start. Unfortunately, a region can be split or coalesced with
2986 * neighboring regions, making it difficult to determine what the
2987 * original region was. Therefore, we approximate this requirement by
2988 * flushing the current region containing start.
2990 * Returns an error if any part of the specified range is not mapped.
2998 boolean_t invalidate)
3000 vm_map_entry_t current;
3001 vm_map_entry_t entry;
3004 vm_ooffset_t offset;
3005 unsigned int last_timestamp;
3008 vm_map_lock_read(map);
3009 VM_MAP_RANGE_CHECK(map, start, end);
3010 if (!vm_map_lookup_entry(map, start, &entry)) {
3011 vm_map_unlock_read(map);
3012 return (KERN_INVALID_ADDRESS);
3013 } else if (start == end) {
3014 start = entry->start;
3018 * Make a first pass to check for user-wired memory and holes.
3020 for (current = entry; current->start < end; current = current->next) {
3021 if (invalidate && (current->eflags & MAP_ENTRY_USER_WIRED)) {
3022 vm_map_unlock_read(map);
3023 return (KERN_INVALID_ARGUMENT);
3025 if (end > current->end &&
3026 current->end != current->next->start) {
3027 vm_map_unlock_read(map);
3028 return (KERN_INVALID_ADDRESS);
3033 pmap_remove(map->pmap, start, end);
3037 * Make a second pass, cleaning/uncaching pages from the indicated
3040 for (current = entry; current->start < end;) {
3041 offset = current->offset + (start - current->start);
3042 size = (end <= current->end ? end : current->end) - start;
3043 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
3045 vm_map_entry_t tentry;
3048 smap = current->object.sub_map;
3049 vm_map_lock_read(smap);
3050 (void) vm_map_lookup_entry(smap, offset, &tentry);
3051 tsize = tentry->end - offset;
3054 object = tentry->object.vm_object;
3055 offset = tentry->offset + (offset - tentry->start);
3056 vm_map_unlock_read(smap);
3058 object = current->object.vm_object;
3060 vm_object_reference(object);
3061 last_timestamp = map->timestamp;
3062 vm_map_unlock_read(map);
3063 if (!vm_object_sync(object, offset, size, syncio, invalidate))
3066 vm_object_deallocate(object);
3067 vm_map_lock_read(map);
3068 if (last_timestamp == map->timestamp ||
3069 !vm_map_lookup_entry(map, start, ¤t))
3070 current = current->next;
3073 vm_map_unlock_read(map);
3074 return (failed ? KERN_FAILURE : KERN_SUCCESS);
3078 * vm_map_entry_unwire: [ internal use only ]
3080 * Make the region specified by this entry pageable.
3082 * The map in question should be locked.
3083 * [This is the reason for this routine's existence.]
3086 vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry)
3089 VM_MAP_ASSERT_LOCKED(map);
3090 KASSERT(entry->wired_count > 0,
3091 ("vm_map_entry_unwire: entry %p isn't wired", entry));
3092 pmap_unwire(map->pmap, entry->start, entry->end);
3093 vm_object_unwire(entry->object.vm_object, entry->offset, entry->end -
3094 entry->start, PQ_ACTIVE);
3095 entry->wired_count = 0;
3099 vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map)
3102 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0)
3103 vm_object_deallocate(entry->object.vm_object);
3104 uma_zfree(system_map ? kmapentzone : mapentzone, entry);
3108 * vm_map_entry_delete: [ internal use only ]
3110 * Deallocate the given entry from the target map.
3113 vm_map_entry_delete(vm_map_t map, vm_map_entry_t entry)
3116 vm_pindex_t offidxstart, offidxend, count, size1;
3119 vm_map_entry_unlink(map, entry);
3120 object = entry->object.vm_object;
3122 if ((entry->eflags & MAP_ENTRY_GUARD) != 0) {
3123 MPASS(entry->cred == NULL);
3124 MPASS((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0);
3125 MPASS(object == NULL);
3126 vm_map_entry_deallocate(entry, map->system_map);
3130 size = entry->end - entry->start;
3133 if (entry->cred != NULL) {
3134 swap_release_by_cred(size, entry->cred);
3135 crfree(entry->cred);
3138 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0 &&
3140 KASSERT(entry->cred == NULL || object->cred == NULL ||
3141 (entry->eflags & MAP_ENTRY_NEEDS_COPY),
3142 ("OVERCOMMIT vm_map_entry_delete: both cred %p", entry));
3144 offidxstart = OFF_TO_IDX(entry->offset);
3145 offidxend = offidxstart + count;
3146 VM_OBJECT_WLOCK(object);
3147 if (object->ref_count != 1 && ((object->flags & (OBJ_NOSPLIT |
3148 OBJ_ONEMAPPING)) == OBJ_ONEMAPPING ||
3149 object == kernel_object)) {
3150 vm_object_collapse(object);
3153 * The option OBJPR_NOTMAPPED can be passed here
3154 * because vm_map_delete() already performed
3155 * pmap_remove() on the only mapping to this range
3158 vm_object_page_remove(object, offidxstart, offidxend,
3160 if (object->type == OBJT_SWAP)
3161 swap_pager_freespace(object, offidxstart,
3163 if (offidxend >= object->size &&
3164 offidxstart < object->size) {
3165 size1 = object->size;
3166 object->size = offidxstart;
3167 if (object->cred != NULL) {
3168 size1 -= object->size;
3169 KASSERT(object->charge >= ptoa(size1),
3170 ("object %p charge < 0", object));
3171 swap_release_by_cred(ptoa(size1),
3173 object->charge -= ptoa(size1);
3177 VM_OBJECT_WUNLOCK(object);
3179 entry->object.vm_object = NULL;
3180 if (map->system_map)
3181 vm_map_entry_deallocate(entry, TRUE);
3183 entry->next = curthread->td_map_def_user;
3184 curthread->td_map_def_user = entry;
3189 * vm_map_delete: [ internal use only ]
3191 * Deallocates the given address range from the target
3195 vm_map_delete(vm_map_t map, vm_offset_t start, vm_offset_t end)
3197 vm_map_entry_t entry;
3198 vm_map_entry_t first_entry;
3200 VM_MAP_ASSERT_LOCKED(map);
3202 return (KERN_SUCCESS);
3205 * Find the start of the region, and clip it
3207 if (!vm_map_lookup_entry(map, start, &first_entry))
3208 entry = first_entry->next;
3210 entry = first_entry;
3211 vm_map_clip_start(map, entry, start);
3215 * Step through all entries in this region
3217 while (entry->start < end) {
3218 vm_map_entry_t next;
3221 * Wait for wiring or unwiring of an entry to complete.
3222 * Also wait for any system wirings to disappear on
3225 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 ||
3226 (vm_map_pmap(map) != kernel_pmap &&
3227 vm_map_entry_system_wired_count(entry) != 0)) {
3228 unsigned int last_timestamp;
3229 vm_offset_t saved_start;
3230 vm_map_entry_t tmp_entry;
3232 saved_start = entry->start;
3233 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
3234 last_timestamp = map->timestamp;
3235 (void) vm_map_unlock_and_wait(map, 0);
3237 if (last_timestamp + 1 != map->timestamp) {
3239 * Look again for the entry because the map was
3240 * modified while it was unlocked.
3241 * Specifically, the entry may have been
3242 * clipped, merged, or deleted.
3244 if (!vm_map_lookup_entry(map, saved_start,
3246 entry = tmp_entry->next;
3249 vm_map_clip_start(map, entry,
3255 vm_map_clip_end(map, entry, end);
3260 * Unwire before removing addresses from the pmap; otherwise,
3261 * unwiring will put the entries back in the pmap.
3263 if (entry->wired_count != 0)
3264 vm_map_entry_unwire(map, entry);
3267 * Remove mappings for the pages, but only if the
3268 * mappings could exist. For instance, it does not
3269 * make sense to call pmap_remove() for guard entries.
3271 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0 ||
3272 entry->object.vm_object != NULL)
3273 pmap_remove(map->pmap, entry->start, entry->end);
3275 if (entry->end == map->anon_loc)
3276 map->anon_loc = entry->start;
3279 * Delete the entry only after removing all pmap
3280 * entries pointing to its pages. (Otherwise, its
3281 * page frames may be reallocated, and any modify bits
3282 * will be set in the wrong object!)
3284 vm_map_entry_delete(map, entry);
3287 return (KERN_SUCCESS);
3293 * Remove the given address range from the target map.
3294 * This is the exported form of vm_map_delete.
3297 vm_map_remove(vm_map_t map, vm_offset_t start, vm_offset_t end)
3302 VM_MAP_RANGE_CHECK(map, start, end);
3303 result = vm_map_delete(map, start, end);
3309 * vm_map_check_protection:
3311 * Assert that the target map allows the specified privilege on the
3312 * entire address region given. The entire region must be allocated.
3314 * WARNING! This code does not and should not check whether the
3315 * contents of the region is accessible. For example a smaller file
3316 * might be mapped into a larger address space.
3318 * NOTE! This code is also called by munmap().
3320 * The map must be locked. A read lock is sufficient.
3323 vm_map_check_protection(vm_map_t map, vm_offset_t start, vm_offset_t end,
3324 vm_prot_t protection)
3326 vm_map_entry_t entry;
3327 vm_map_entry_t tmp_entry;
3329 if (!vm_map_lookup_entry(map, start, &tmp_entry))
3333 while (start < end) {
3337 if (start < entry->start)
3340 * Check protection associated with entry.
3342 if ((entry->protection & protection) != protection)
3344 /* go to next entry */
3346 entry = entry->next;
3352 * vm_map_copy_entry:
3354 * Copies the contents of the source entry to the destination
3355 * entry. The entries *must* be aligned properly.
3361 vm_map_entry_t src_entry,
3362 vm_map_entry_t dst_entry,
3363 vm_ooffset_t *fork_charge)
3365 vm_object_t src_object;
3366 vm_map_entry_t fake_entry;
3371 VM_MAP_ASSERT_LOCKED(dst_map);
3373 if ((dst_entry->eflags|src_entry->eflags) & MAP_ENTRY_IS_SUB_MAP)
3376 if (src_entry->wired_count == 0 ||
3377 (src_entry->protection & VM_PROT_WRITE) == 0) {
3379 * If the source entry is marked needs_copy, it is already
3382 if ((src_entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0 &&
3383 (src_entry->protection & VM_PROT_WRITE) != 0) {
3384 pmap_protect(src_map->pmap,
3387 src_entry->protection & ~VM_PROT_WRITE);
3391 * Make a copy of the object.
3393 size = src_entry->end - src_entry->start;
3394 if ((src_object = src_entry->object.vm_object) != NULL) {
3395 VM_OBJECT_WLOCK(src_object);
3396 charged = ENTRY_CHARGED(src_entry);
3397 if (src_object->handle == NULL &&
3398 (src_object->type == OBJT_DEFAULT ||
3399 src_object->type == OBJT_SWAP)) {
3400 vm_object_collapse(src_object);
3401 if ((src_object->flags & (OBJ_NOSPLIT |
3402 OBJ_ONEMAPPING)) == OBJ_ONEMAPPING) {
3403 vm_object_split(src_entry);
3405 src_entry->object.vm_object;
3408 vm_object_reference_locked(src_object);
3409 vm_object_clear_flag(src_object, OBJ_ONEMAPPING);
3410 if (src_entry->cred != NULL &&
3411 !(src_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
3412 KASSERT(src_object->cred == NULL,
3413 ("OVERCOMMIT: vm_map_copy_entry: cred %p",
3415 src_object->cred = src_entry->cred;
3416 src_object->charge = size;
3418 VM_OBJECT_WUNLOCK(src_object);
3419 dst_entry->object.vm_object = src_object;
3421 cred = curthread->td_ucred;
3423 dst_entry->cred = cred;
3424 *fork_charge += size;
3425 if (!(src_entry->eflags &
3426 MAP_ENTRY_NEEDS_COPY)) {
3428 src_entry->cred = cred;
3429 *fork_charge += size;
3432 src_entry->eflags |= MAP_ENTRY_COW |
3433 MAP_ENTRY_NEEDS_COPY;
3434 dst_entry->eflags |= MAP_ENTRY_COW |
3435 MAP_ENTRY_NEEDS_COPY;
3436 dst_entry->offset = src_entry->offset;
3437 if (src_entry->eflags & MAP_ENTRY_VN_WRITECNT) {
3439 * MAP_ENTRY_VN_WRITECNT cannot
3440 * indicate write reference from
3441 * src_entry, since the entry is
3442 * marked as needs copy. Allocate a
3443 * fake entry that is used to
3444 * decrement object->un_pager.vnp.writecount
3445 * at the appropriate time. Attach
3446 * fake_entry to the deferred list.
3448 fake_entry = vm_map_entry_create(dst_map);
3449 fake_entry->eflags = MAP_ENTRY_VN_WRITECNT;
3450 src_entry->eflags &= ~MAP_ENTRY_VN_WRITECNT;
3451 vm_object_reference(src_object);
3452 fake_entry->object.vm_object = src_object;
3453 fake_entry->start = src_entry->start;
3454 fake_entry->end = src_entry->end;
3455 fake_entry->next = curthread->td_map_def_user;
3456 curthread->td_map_def_user = fake_entry;
3459 pmap_copy(dst_map->pmap, src_map->pmap,
3460 dst_entry->start, dst_entry->end - dst_entry->start,
3463 dst_entry->object.vm_object = NULL;
3464 dst_entry->offset = 0;
3465 if (src_entry->cred != NULL) {
3466 dst_entry->cred = curthread->td_ucred;
3467 crhold(dst_entry->cred);
3468 *fork_charge += size;
3473 * We don't want to make writeable wired pages copy-on-write.
3474 * Immediately copy these pages into the new map by simulating
3475 * page faults. The new pages are pageable.
3477 vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry,
3483 * vmspace_map_entry_forked:
3484 * Update the newly-forked vmspace each time a map entry is inherited
3485 * or copied. The values for vm_dsize and vm_tsize are approximate
3486 * (and mostly-obsolete ideas in the face of mmap(2) et al.)
3489 vmspace_map_entry_forked(const struct vmspace *vm1, struct vmspace *vm2,
3490 vm_map_entry_t entry)
3492 vm_size_t entrysize;
3495 if ((entry->eflags & MAP_ENTRY_GUARD) != 0)
3497 entrysize = entry->end - entry->start;
3498 vm2->vm_map.size += entrysize;
3499 if (entry->eflags & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP)) {
3500 vm2->vm_ssize += btoc(entrysize);
3501 } else if (entry->start >= (vm_offset_t)vm1->vm_daddr &&
3502 entry->start < (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)) {
3503 newend = MIN(entry->end,
3504 (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize));
3505 vm2->vm_dsize += btoc(newend - entry->start);
3506 } else if (entry->start >= (vm_offset_t)vm1->vm_taddr &&
3507 entry->start < (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)) {
3508 newend = MIN(entry->end,
3509 (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize));
3510 vm2->vm_tsize += btoc(newend - entry->start);
3516 * Create a new process vmspace structure and vm_map
3517 * based on those of an existing process. The new map
3518 * is based on the old map, according to the inheritance
3519 * values on the regions in that map.
3521 * XXX It might be worth coalescing the entries added to the new vmspace.
3523 * The source map must not be locked.
3526 vmspace_fork(struct vmspace *vm1, vm_ooffset_t *fork_charge)
3528 struct vmspace *vm2;
3529 vm_map_t new_map, old_map;
3530 vm_map_entry_t new_entry, old_entry;
3535 old_map = &vm1->vm_map;
3536 /* Copy immutable fields of vm1 to vm2. */
3537 vm2 = vmspace_alloc(vm_map_min(old_map), vm_map_max(old_map),
3541 vm2->vm_taddr = vm1->vm_taddr;
3542 vm2->vm_daddr = vm1->vm_daddr;
3543 vm2->vm_maxsaddr = vm1->vm_maxsaddr;
3544 vm_map_lock(old_map);
3546 vm_map_wait_busy(old_map);
3547 new_map = &vm2->vm_map;
3548 locked = vm_map_trylock(new_map); /* trylock to silence WITNESS */
3549 KASSERT(locked, ("vmspace_fork: lock failed"));
3551 new_map->anon_loc = old_map->anon_loc;
3552 old_entry = old_map->header.next;
3554 while (old_entry != &old_map->header) {
3555 if (old_entry->eflags & MAP_ENTRY_IS_SUB_MAP)
3556 panic("vm_map_fork: encountered a submap");
3558 inh = old_entry->inheritance;
3559 if ((old_entry->eflags & MAP_ENTRY_GUARD) != 0 &&
3560 inh != VM_INHERIT_NONE)
3561 inh = VM_INHERIT_COPY;
3564 case VM_INHERIT_NONE:
3567 case VM_INHERIT_SHARE:
3569 * Clone the entry, creating the shared object if necessary.
3571 object = old_entry->object.vm_object;
3572 if (object == NULL) {
3573 object = vm_object_allocate(OBJT_DEFAULT,
3574 atop(old_entry->end - old_entry->start));
3575 old_entry->object.vm_object = object;
3576 old_entry->offset = 0;
3577 if (old_entry->cred != NULL) {
3578 object->cred = old_entry->cred;
3579 object->charge = old_entry->end -
3581 old_entry->cred = NULL;
3586 * Add the reference before calling vm_object_shadow
3587 * to insure that a shadow object is created.
3589 vm_object_reference(object);
3590 if (old_entry->eflags & MAP_ENTRY_NEEDS_COPY) {
3591 vm_object_shadow(&old_entry->object.vm_object,
3593 old_entry->end - old_entry->start);
3594 old_entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
3595 /* Transfer the second reference too. */
3596 vm_object_reference(
3597 old_entry->object.vm_object);
3600 * As in vm_map_simplify_entry(), the
3601 * vnode lock will not be acquired in
3602 * this call to vm_object_deallocate().
3604 vm_object_deallocate(object);
3605 object = old_entry->object.vm_object;
3607 VM_OBJECT_WLOCK(object);
3608 vm_object_clear_flag(object, OBJ_ONEMAPPING);
3609 if (old_entry->cred != NULL) {
3610 KASSERT(object->cred == NULL, ("vmspace_fork both cred"));
3611 object->cred = old_entry->cred;
3612 object->charge = old_entry->end - old_entry->start;
3613 old_entry->cred = NULL;
3617 * Assert the correct state of the vnode
3618 * v_writecount while the object is locked, to
3619 * not relock it later for the assertion
3622 if (old_entry->eflags & MAP_ENTRY_VN_WRITECNT &&
3623 object->type == OBJT_VNODE) {
3624 KASSERT(((struct vnode *)object->handle)->
3626 ("vmspace_fork: v_writecount %p", object));
3627 KASSERT(object->un_pager.vnp.writemappings > 0,
3628 ("vmspace_fork: vnp.writecount %p",
3631 VM_OBJECT_WUNLOCK(object);
3634 * Clone the entry, referencing the shared object.
3636 new_entry = vm_map_entry_create(new_map);
3637 *new_entry = *old_entry;
3638 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED |
3639 MAP_ENTRY_IN_TRANSITION);
3640 new_entry->wiring_thread = NULL;
3641 new_entry->wired_count = 0;
3642 if (new_entry->eflags & MAP_ENTRY_VN_WRITECNT) {
3643 vnode_pager_update_writecount(object,
3644 new_entry->start, new_entry->end);
3648 * Insert the entry into the new map -- we know we're
3649 * inserting at the end of the new map.
3651 vm_map_entry_link(new_map, new_map->header.prev,
3653 vmspace_map_entry_forked(vm1, vm2, new_entry);
3656 * Update the physical map
3658 pmap_copy(new_map->pmap, old_map->pmap,
3660 (old_entry->end - old_entry->start),
3664 case VM_INHERIT_COPY:
3666 * Clone the entry and link into the map.
3668 new_entry = vm_map_entry_create(new_map);
3669 *new_entry = *old_entry;
3671 * Copied entry is COW over the old object.
3673 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED |
3674 MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_VN_WRITECNT);
3675 new_entry->wiring_thread = NULL;
3676 new_entry->wired_count = 0;
3677 new_entry->object.vm_object = NULL;
3678 new_entry->cred = NULL;
3679 vm_map_entry_link(new_map, new_map->header.prev,
3681 vmspace_map_entry_forked(vm1, vm2, new_entry);
3682 vm_map_copy_entry(old_map, new_map, old_entry,
3683 new_entry, fork_charge);
3686 case VM_INHERIT_ZERO:
3688 * Create a new anonymous mapping entry modelled from
3691 new_entry = vm_map_entry_create(new_map);
3692 memset(new_entry, 0, sizeof(*new_entry));
3694 new_entry->start = old_entry->start;
3695 new_entry->end = old_entry->end;
3696 new_entry->eflags = old_entry->eflags &
3697 ~(MAP_ENTRY_USER_WIRED | MAP_ENTRY_IN_TRANSITION |
3698 MAP_ENTRY_VN_WRITECNT);
3699 new_entry->protection = old_entry->protection;
3700 new_entry->max_protection = old_entry->max_protection;
3701 new_entry->inheritance = VM_INHERIT_ZERO;
3703 vm_map_entry_link(new_map, new_map->header.prev,
3705 vmspace_map_entry_forked(vm1, vm2, new_entry);
3707 new_entry->cred = curthread->td_ucred;
3708 crhold(new_entry->cred);
3709 *fork_charge += (new_entry->end - new_entry->start);
3713 old_entry = old_entry->next;
3716 * Use inlined vm_map_unlock() to postpone handling the deferred
3717 * map entries, which cannot be done until both old_map and
3718 * new_map locks are released.
3720 sx_xunlock(&old_map->lock);
3721 sx_xunlock(&new_map->lock);
3722 vm_map_process_deferred();
3728 * Create a process's stack for exec_new_vmspace(). This function is never
3729 * asked to wire the newly created stack.
3732 vm_map_stack(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
3733 vm_prot_t prot, vm_prot_t max, int cow)
3735 vm_size_t growsize, init_ssize;
3739 MPASS((map->flags & MAP_WIREFUTURE) == 0);
3740 growsize = sgrowsiz;
3741 init_ssize = (max_ssize < growsize) ? max_ssize : growsize;
3743 vmemlim = lim_cur(curthread, RLIMIT_VMEM);
3744 /* If we would blow our VMEM resource limit, no go */
3745 if (map->size + init_ssize > vmemlim) {
3749 rv = vm_map_stack_locked(map, addrbos, max_ssize, growsize, prot,
3756 static int stack_guard_page = 1;
3757 SYSCTL_INT(_security_bsd, OID_AUTO, stack_guard_page, CTLFLAG_RWTUN,
3758 &stack_guard_page, 0,
3759 "Specifies the number of guard pages for a stack that grows");
3762 vm_map_stack_locked(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
3763 vm_size_t growsize, vm_prot_t prot, vm_prot_t max, int cow)
3765 vm_map_entry_t new_entry, prev_entry;
3766 vm_offset_t bot, gap_bot, gap_top, top;
3767 vm_size_t init_ssize, sgp;
3771 * The stack orientation is piggybacked with the cow argument.
3772 * Extract it into orient and mask the cow argument so that we
3773 * don't pass it around further.
3775 orient = cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP);
3776 KASSERT(orient != 0, ("No stack grow direction"));
3777 KASSERT(orient != (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP),
3780 if (addrbos < vm_map_min(map) ||
3781 addrbos + max_ssize > vm_map_max(map) ||
3782 addrbos + max_ssize <= addrbos)
3783 return (KERN_INVALID_ADDRESS);
3784 sgp = (vm_size_t)stack_guard_page * PAGE_SIZE;
3785 if (sgp >= max_ssize)
3786 return (KERN_INVALID_ARGUMENT);
3788 init_ssize = growsize;
3789 if (max_ssize < init_ssize + sgp)
3790 init_ssize = max_ssize - sgp;
3792 /* If addr is already mapped, no go */
3793 if (vm_map_lookup_entry(map, addrbos, &prev_entry))
3794 return (KERN_NO_SPACE);
3797 * If we can't accommodate max_ssize in the current mapping, no go.
3799 if (prev_entry->next->start < addrbos + max_ssize)
3800 return (KERN_NO_SPACE);
3803 * We initially map a stack of only init_ssize. We will grow as
3804 * needed later. Depending on the orientation of the stack (i.e.
3805 * the grow direction) we either map at the top of the range, the
3806 * bottom of the range or in the middle.
3808 * Note: we would normally expect prot and max to be VM_PROT_ALL,
3809 * and cow to be 0. Possibly we should eliminate these as input
3810 * parameters, and just pass these values here in the insert call.
3812 if (orient == MAP_STACK_GROWS_DOWN) {
3813 bot = addrbos + max_ssize - init_ssize;
3814 top = bot + init_ssize;
3817 } else /* if (orient == MAP_STACK_GROWS_UP) */ {
3819 top = bot + init_ssize;
3821 gap_top = addrbos + max_ssize;
3823 rv = vm_map_insert(map, NULL, 0, bot, top, prot, max, cow);
3824 if (rv != KERN_SUCCESS)
3826 new_entry = prev_entry->next;
3827 KASSERT(new_entry->end == top || new_entry->start == bot,
3828 ("Bad entry start/end for new stack entry"));
3829 KASSERT((orient & MAP_STACK_GROWS_DOWN) == 0 ||
3830 (new_entry->eflags & MAP_ENTRY_GROWS_DOWN) != 0,
3831 ("new entry lacks MAP_ENTRY_GROWS_DOWN"));
3832 KASSERT((orient & MAP_STACK_GROWS_UP) == 0 ||
3833 (new_entry->eflags & MAP_ENTRY_GROWS_UP) != 0,
3834 ("new entry lacks MAP_ENTRY_GROWS_UP"));
3835 rv = vm_map_insert(map, NULL, 0, gap_bot, gap_top, VM_PROT_NONE,
3836 VM_PROT_NONE, MAP_CREATE_GUARD | (orient == MAP_STACK_GROWS_DOWN ?
3837 MAP_CREATE_STACK_GAP_DN : MAP_CREATE_STACK_GAP_UP));
3838 if (rv != KERN_SUCCESS)
3839 (void)vm_map_delete(map, bot, top);
3844 * Attempts to grow a vm stack entry. Returns KERN_SUCCESS if we
3845 * successfully grow the stack.
3848 vm_map_growstack(vm_map_t map, vm_offset_t addr, vm_map_entry_t gap_entry)
3850 vm_map_entry_t stack_entry;
3854 vm_offset_t gap_end, gap_start, grow_start;
3855 size_t grow_amount, guard, max_grow;
3856 rlim_t lmemlim, stacklim, vmemlim;
3858 bool gap_deleted, grow_down, is_procstack;
3870 * Disallow stack growth when the access is performed by a
3871 * debugger or AIO daemon. The reason is that the wrong
3872 * resource limits are applied.
3874 if (map != &p->p_vmspace->vm_map || p->p_textvp == NULL)
3875 return (KERN_FAILURE);
3877 MPASS(!map->system_map);
3879 guard = stack_guard_page * PAGE_SIZE;
3880 lmemlim = lim_cur(curthread, RLIMIT_MEMLOCK);
3881 stacklim = lim_cur(curthread, RLIMIT_STACK);
3882 vmemlim = lim_cur(curthread, RLIMIT_VMEM);
3884 /* If addr is not in a hole for a stack grow area, no need to grow. */
3885 if (gap_entry == NULL && !vm_map_lookup_entry(map, addr, &gap_entry))
3886 return (KERN_FAILURE);
3887 if ((gap_entry->eflags & MAP_ENTRY_GUARD) == 0)
3888 return (KERN_SUCCESS);
3889 if ((gap_entry->eflags & MAP_ENTRY_STACK_GAP_DN) != 0) {
3890 stack_entry = gap_entry->next;
3891 if ((stack_entry->eflags & MAP_ENTRY_GROWS_DOWN) == 0 ||
3892 stack_entry->start != gap_entry->end)
3893 return (KERN_FAILURE);
3894 grow_amount = round_page(stack_entry->start - addr);
3896 } else if ((gap_entry->eflags & MAP_ENTRY_STACK_GAP_UP) != 0) {
3897 stack_entry = gap_entry->prev;
3898 if ((stack_entry->eflags & MAP_ENTRY_GROWS_UP) == 0 ||
3899 stack_entry->end != gap_entry->start)
3900 return (KERN_FAILURE);
3901 grow_amount = round_page(addr + 1 - stack_entry->end);
3904 return (KERN_FAILURE);
3906 max_grow = gap_entry->end - gap_entry->start;
3907 if (guard > max_grow)
3908 return (KERN_NO_SPACE);
3910 if (grow_amount > max_grow)
3911 return (KERN_NO_SPACE);
3914 * If this is the main process stack, see if we're over the stack
3917 is_procstack = addr >= (vm_offset_t)vm->vm_maxsaddr &&
3918 addr < (vm_offset_t)p->p_sysent->sv_usrstack;
3919 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim))
3920 return (KERN_NO_SPACE);
3925 if (is_procstack && racct_set(p, RACCT_STACK,
3926 ctob(vm->vm_ssize) + grow_amount)) {
3928 return (KERN_NO_SPACE);
3934 grow_amount = roundup(grow_amount, sgrowsiz);
3935 if (grow_amount > max_grow)
3936 grow_amount = max_grow;
3937 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) {
3938 grow_amount = trunc_page((vm_size_t)stacklim) -
3944 limit = racct_get_available(p, RACCT_STACK);
3946 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > limit))
3947 grow_amount = limit - ctob(vm->vm_ssize);
3950 if (!old_mlock && (map->flags & MAP_WIREFUTURE) != 0) {
3951 if (ptoa(pmap_wired_count(map->pmap)) + grow_amount > lmemlim) {
3958 if (racct_set(p, RACCT_MEMLOCK,
3959 ptoa(pmap_wired_count(map->pmap)) + grow_amount)) {
3969 /* If we would blow our VMEM resource limit, no go */
3970 if (map->size + grow_amount > vmemlim) {
3977 if (racct_set(p, RACCT_VMEM, map->size + grow_amount)) {
3986 if (vm_map_lock_upgrade(map)) {
3988 vm_map_lock_read(map);
3993 grow_start = gap_entry->end - grow_amount;
3994 if (gap_entry->start + grow_amount == gap_entry->end) {
3995 gap_start = gap_entry->start;
3996 gap_end = gap_entry->end;
3997 vm_map_entry_delete(map, gap_entry);
4000 MPASS(gap_entry->start < gap_entry->end - grow_amount);
4001 gap_entry->end -= grow_amount;
4002 vm_map_entry_resize_free(map, gap_entry);
4003 gap_deleted = false;
4005 rv = vm_map_insert(map, NULL, 0, grow_start,
4006 grow_start + grow_amount,
4007 stack_entry->protection, stack_entry->max_protection,
4008 MAP_STACK_GROWS_DOWN);
4009 if (rv != KERN_SUCCESS) {
4011 rv1 = vm_map_insert(map, NULL, 0, gap_start,
4012 gap_end, VM_PROT_NONE, VM_PROT_NONE,
4013 MAP_CREATE_GUARD | MAP_CREATE_STACK_GAP_DN);
4014 MPASS(rv1 == KERN_SUCCESS);
4016 gap_entry->end += grow_amount;
4017 vm_map_entry_resize_free(map, gap_entry);
4021 grow_start = stack_entry->end;
4022 cred = stack_entry->cred;
4023 if (cred == NULL && stack_entry->object.vm_object != NULL)
4024 cred = stack_entry->object.vm_object->cred;
4025 if (cred != NULL && !swap_reserve_by_cred(grow_amount, cred))
4027 /* Grow the underlying object if applicable. */
4028 else if (stack_entry->object.vm_object == NULL ||
4029 vm_object_coalesce(stack_entry->object.vm_object,
4030 stack_entry->offset,
4031 (vm_size_t)(stack_entry->end - stack_entry->start),
4032 (vm_size_t)grow_amount, cred != NULL)) {
4033 if (gap_entry->start + grow_amount == gap_entry->end)
4034 vm_map_entry_delete(map, gap_entry);
4036 gap_entry->start += grow_amount;
4037 stack_entry->end += grow_amount;
4038 map->size += grow_amount;
4039 vm_map_entry_resize_free(map, stack_entry);
4044 if (rv == KERN_SUCCESS && is_procstack)
4045 vm->vm_ssize += btoc(grow_amount);
4048 * Heed the MAP_WIREFUTURE flag if it was set for this process.
4050 if (rv == KERN_SUCCESS && (map->flags & MAP_WIREFUTURE) != 0) {
4052 vm_map_wire(map, grow_start, grow_start + grow_amount,
4053 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
4054 vm_map_lock_read(map);
4056 vm_map_lock_downgrade(map);
4060 if (racct_enable && rv != KERN_SUCCESS) {
4062 error = racct_set(p, RACCT_VMEM, map->size);
4063 KASSERT(error == 0, ("decreasing RACCT_VMEM failed"));
4065 error = racct_set(p, RACCT_MEMLOCK,
4066 ptoa(pmap_wired_count(map->pmap)));
4067 KASSERT(error == 0, ("decreasing RACCT_MEMLOCK failed"));
4069 error = racct_set(p, RACCT_STACK, ctob(vm->vm_ssize));
4070 KASSERT(error == 0, ("decreasing RACCT_STACK failed"));
4079 * Unshare the specified VM space for exec. If other processes are
4080 * mapped to it, then create a new one. The new vmspace is null.
4083 vmspace_exec(struct proc *p, vm_offset_t minuser, vm_offset_t maxuser)
4085 struct vmspace *oldvmspace = p->p_vmspace;
4086 struct vmspace *newvmspace;
4088 KASSERT((curthread->td_pflags & TDP_EXECVMSPC) == 0,
4089 ("vmspace_exec recursed"));
4090 newvmspace = vmspace_alloc(minuser, maxuser, pmap_pinit);
4091 if (newvmspace == NULL)
4093 newvmspace->vm_swrss = oldvmspace->vm_swrss;
4095 * This code is written like this for prototype purposes. The
4096 * goal is to avoid running down the vmspace here, but let the
4097 * other process's that are still using the vmspace to finally
4098 * run it down. Even though there is little or no chance of blocking
4099 * here, it is a good idea to keep this form for future mods.
4101 PROC_VMSPACE_LOCK(p);
4102 p->p_vmspace = newvmspace;
4103 PROC_VMSPACE_UNLOCK(p);
4104 if (p == curthread->td_proc)
4105 pmap_activate(curthread);
4106 curthread->td_pflags |= TDP_EXECVMSPC;
4111 * Unshare the specified VM space for forcing COW. This
4112 * is called by rfork, for the (RFMEM|RFPROC) == 0 case.
4115 vmspace_unshare(struct proc *p)
4117 struct vmspace *oldvmspace = p->p_vmspace;
4118 struct vmspace *newvmspace;
4119 vm_ooffset_t fork_charge;
4121 if (oldvmspace->vm_refcnt == 1)
4124 newvmspace = vmspace_fork(oldvmspace, &fork_charge);
4125 if (newvmspace == NULL)
4127 if (!swap_reserve_by_cred(fork_charge, p->p_ucred)) {
4128 vmspace_free(newvmspace);
4131 PROC_VMSPACE_LOCK(p);
4132 p->p_vmspace = newvmspace;
4133 PROC_VMSPACE_UNLOCK(p);
4134 if (p == curthread->td_proc)
4135 pmap_activate(curthread);
4136 vmspace_free(oldvmspace);
4143 * Finds the VM object, offset, and
4144 * protection for a given virtual address in the
4145 * specified map, assuming a page fault of the
4148 * Leaves the map in question locked for read; return
4149 * values are guaranteed until a vm_map_lookup_done
4150 * call is performed. Note that the map argument
4151 * is in/out; the returned map must be used in
4152 * the call to vm_map_lookup_done.
4154 * A handle (out_entry) is returned for use in
4155 * vm_map_lookup_done, to make that fast.
4157 * If a lookup is requested with "write protection"
4158 * specified, the map may be changed to perform virtual
4159 * copying operations, although the data referenced will
4163 vm_map_lookup(vm_map_t *var_map, /* IN/OUT */
4165 vm_prot_t fault_typea,
4166 vm_map_entry_t *out_entry, /* OUT */
4167 vm_object_t *object, /* OUT */
4168 vm_pindex_t *pindex, /* OUT */
4169 vm_prot_t *out_prot, /* OUT */
4170 boolean_t *wired) /* OUT */
4172 vm_map_entry_t entry;
4173 vm_map_t map = *var_map;
4175 vm_prot_t fault_type = fault_typea;
4176 vm_object_t eobject;
4182 vm_map_lock_read(map);
4186 * Lookup the faulting address.
4188 if (!vm_map_lookup_entry(map, vaddr, out_entry)) {
4189 vm_map_unlock_read(map);
4190 return (KERN_INVALID_ADDRESS);
4198 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
4199 vm_map_t old_map = map;
4201 *var_map = map = entry->object.sub_map;
4202 vm_map_unlock_read(old_map);
4207 * Check whether this task is allowed to have this page.
4209 prot = entry->protection;
4210 if ((fault_typea & VM_PROT_FAULT_LOOKUP) != 0) {
4211 fault_typea &= ~VM_PROT_FAULT_LOOKUP;
4212 if (prot == VM_PROT_NONE && map != kernel_map &&
4213 (entry->eflags & MAP_ENTRY_GUARD) != 0 &&
4214 (entry->eflags & (MAP_ENTRY_STACK_GAP_DN |
4215 MAP_ENTRY_STACK_GAP_UP)) != 0 &&
4216 vm_map_growstack(map, vaddr, entry) == KERN_SUCCESS)
4217 goto RetryLookupLocked;
4219 fault_type &= VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE;
4220 if ((fault_type & prot) != fault_type || prot == VM_PROT_NONE) {
4221 vm_map_unlock_read(map);
4222 return (KERN_PROTECTION_FAILURE);
4224 KASSERT((prot & VM_PROT_WRITE) == 0 || (entry->eflags &
4225 (MAP_ENTRY_USER_WIRED | MAP_ENTRY_NEEDS_COPY)) !=
4226 (MAP_ENTRY_USER_WIRED | MAP_ENTRY_NEEDS_COPY),
4227 ("entry %p flags %x", entry, entry->eflags));
4228 if ((fault_typea & VM_PROT_COPY) != 0 &&
4229 (entry->max_protection & VM_PROT_WRITE) == 0 &&
4230 (entry->eflags & MAP_ENTRY_COW) == 0) {
4231 vm_map_unlock_read(map);
4232 return (KERN_PROTECTION_FAILURE);
4236 * If this page is not pageable, we have to get it for all possible
4239 *wired = (entry->wired_count != 0);
4241 fault_type = entry->protection;
4242 size = entry->end - entry->start;
4244 * If the entry was copy-on-write, we either ...
4246 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
4248 * If we want to write the page, we may as well handle that
4249 * now since we've got the map locked.
4251 * If we don't need to write the page, we just demote the
4252 * permissions allowed.
4254 if ((fault_type & VM_PROT_WRITE) != 0 ||
4255 (fault_typea & VM_PROT_COPY) != 0) {
4257 * Make a new object, and place it in the object
4258 * chain. Note that no new references have appeared
4259 * -- one just moved from the map to the new
4262 if (vm_map_lock_upgrade(map))
4265 if (entry->cred == NULL) {
4267 * The debugger owner is charged for
4270 cred = curthread->td_ucred;
4272 if (!swap_reserve_by_cred(size, cred)) {
4275 return (KERN_RESOURCE_SHORTAGE);
4279 vm_object_shadow(&entry->object.vm_object,
4280 &entry->offset, size);
4281 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
4282 eobject = entry->object.vm_object;
4283 if (eobject->cred != NULL) {
4285 * The object was not shadowed.
4287 swap_release_by_cred(size, entry->cred);
4288 crfree(entry->cred);
4290 } else if (entry->cred != NULL) {
4291 VM_OBJECT_WLOCK(eobject);
4292 eobject->cred = entry->cred;
4293 eobject->charge = size;
4294 VM_OBJECT_WUNLOCK(eobject);
4298 vm_map_lock_downgrade(map);
4301 * We're attempting to read a copy-on-write page --
4302 * don't allow writes.
4304 prot &= ~VM_PROT_WRITE;
4309 * Create an object if necessary.
4311 if (entry->object.vm_object == NULL &&
4313 if (vm_map_lock_upgrade(map))
4315 entry->object.vm_object = vm_object_allocate(OBJT_DEFAULT,
4318 if (entry->cred != NULL) {
4319 VM_OBJECT_WLOCK(entry->object.vm_object);
4320 entry->object.vm_object->cred = entry->cred;
4321 entry->object.vm_object->charge = size;
4322 VM_OBJECT_WUNLOCK(entry->object.vm_object);
4325 vm_map_lock_downgrade(map);
4329 * Return the object/offset from this entry. If the entry was
4330 * copy-on-write or empty, it has been fixed up.
4332 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
4333 *object = entry->object.vm_object;
4336 return (KERN_SUCCESS);
4340 * vm_map_lookup_locked:
4342 * Lookup the faulting address. A version of vm_map_lookup that returns
4343 * KERN_FAILURE instead of blocking on map lock or memory allocation.
4346 vm_map_lookup_locked(vm_map_t *var_map, /* IN/OUT */
4348 vm_prot_t fault_typea,
4349 vm_map_entry_t *out_entry, /* OUT */
4350 vm_object_t *object, /* OUT */
4351 vm_pindex_t *pindex, /* OUT */
4352 vm_prot_t *out_prot, /* OUT */
4353 boolean_t *wired) /* OUT */
4355 vm_map_entry_t entry;
4356 vm_map_t map = *var_map;
4358 vm_prot_t fault_type = fault_typea;
4361 * Lookup the faulting address.
4363 if (!vm_map_lookup_entry(map, vaddr, out_entry))
4364 return (KERN_INVALID_ADDRESS);
4369 * Fail if the entry refers to a submap.
4371 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP)
4372 return (KERN_FAILURE);
4375 * Check whether this task is allowed to have this page.
4377 prot = entry->protection;
4378 fault_type &= VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE;
4379 if ((fault_type & prot) != fault_type)
4380 return (KERN_PROTECTION_FAILURE);
4383 * If this page is not pageable, we have to get it for all possible
4386 *wired = (entry->wired_count != 0);
4388 fault_type = entry->protection;
4390 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
4392 * Fail if the entry was copy-on-write for a write fault.
4394 if (fault_type & VM_PROT_WRITE)
4395 return (KERN_FAILURE);
4397 * We're attempting to read a copy-on-write page --
4398 * don't allow writes.
4400 prot &= ~VM_PROT_WRITE;
4404 * Fail if an object should be created.
4406 if (entry->object.vm_object == NULL && !map->system_map)
4407 return (KERN_FAILURE);
4410 * Return the object/offset from this entry. If the entry was
4411 * copy-on-write or empty, it has been fixed up.
4413 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
4414 *object = entry->object.vm_object;
4417 return (KERN_SUCCESS);
4421 * vm_map_lookup_done:
4423 * Releases locks acquired by a vm_map_lookup
4424 * (according to the handle returned by that lookup).
4427 vm_map_lookup_done(vm_map_t map, vm_map_entry_t entry)
4430 * Unlock the main-level map
4432 vm_map_unlock_read(map);
4436 vm_map_max_KBI(const struct vm_map *map)
4439 return (vm_map_max(map));
4443 vm_map_min_KBI(const struct vm_map *map)
4446 return (vm_map_min(map));
4450 vm_map_pmap_KBI(vm_map_t map)
4456 #include "opt_ddb.h"
4458 #include <sys/kernel.h>
4460 #include <ddb/ddb.h>
4463 vm_map_print(vm_map_t map)
4465 vm_map_entry_t entry;
4467 db_iprintf("Task map %p: pmap=%p, nentries=%d, version=%u\n",
4469 (void *)map->pmap, map->nentries, map->timestamp);
4472 for (entry = map->header.next; entry != &map->header;
4473 entry = entry->next) {
4474 db_iprintf("map entry %p: start=%p, end=%p, eflags=%#x, \n",
4475 (void *)entry, (void *)entry->start, (void *)entry->end,
4478 static char *inheritance_name[4] =
4479 {"share", "copy", "none", "donate_copy"};
4481 db_iprintf(" prot=%x/%x/%s",
4483 entry->max_protection,
4484 inheritance_name[(int)(unsigned char)entry->inheritance]);
4485 if (entry->wired_count != 0)
4486 db_printf(", wired");
4488 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
4489 db_printf(", share=%p, offset=0x%jx\n",
4490 (void *)entry->object.sub_map,
4491 (uintmax_t)entry->offset);
4492 if ((entry->prev == &map->header) ||
4493 (entry->prev->object.sub_map !=
4494 entry->object.sub_map)) {
4496 vm_map_print((vm_map_t)entry->object.sub_map);
4500 if (entry->cred != NULL)
4501 db_printf(", ruid %d", entry->cred->cr_ruid);
4502 db_printf(", object=%p, offset=0x%jx",
4503 (void *)entry->object.vm_object,
4504 (uintmax_t)entry->offset);
4505 if (entry->object.vm_object && entry->object.vm_object->cred)
4506 db_printf(", obj ruid %d charge %jx",
4507 entry->object.vm_object->cred->cr_ruid,
4508 (uintmax_t)entry->object.vm_object->charge);
4509 if (entry->eflags & MAP_ENTRY_COW)
4510 db_printf(", copy (%s)",
4511 (entry->eflags & MAP_ENTRY_NEEDS_COPY) ? "needed" : "done");
4514 if ((entry->prev == &map->header) ||
4515 (entry->prev->object.vm_object !=
4516 entry->object.vm_object)) {
4518 vm_object_print((db_expr_t)(intptr_t)
4519 entry->object.vm_object,
4528 DB_SHOW_COMMAND(map, map)
4532 db_printf("usage: show map <addr>\n");
4535 vm_map_print((vm_map_t)addr);
4538 DB_SHOW_COMMAND(procvm, procvm)
4543 p = db_lookup_proc(addr);
4548 db_printf("p = %p, vmspace = %p, map = %p, pmap = %p\n",
4549 (void *)p, (void *)p->p_vmspace, (void *)&p->p_vmspace->vm_map,
4550 (void *)vmspace_pmap(p->p_vmspace));
4552 vm_map_print((vm_map_t)&p->p_vmspace->vm_map);