2 * SPDX-License-Identifier: (BSD-3-Clause AND MIT-CMU)
4 * Copyright (c) 1991, 1993
5 * The Regents of the University of California. All rights reserved.
7 * This code is derived from software contributed to Berkeley by
8 * The Mach Operating System project at Carnegie-Mellon University.
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 3. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * from: @(#)vm_map.c 8.3 (Berkeley) 1/12/94
37 * Copyright (c) 1987, 1990 Carnegie-Mellon University.
38 * All rights reserved.
40 * Authors: Avadis Tevanian, Jr., Michael Wayne Young
42 * Permission to use, copy, modify and distribute this software and
43 * its documentation is hereby granted, provided that both the copyright
44 * notice and this permission notice appear in all copies of the
45 * software, derivative works or modified versions, and any portions
46 * thereof, and that both notices appear in supporting documentation.
48 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
49 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
50 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
52 * Carnegie Mellon requests users of this software to return to
54 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
55 * School of Computer Science
56 * Carnegie Mellon University
57 * Pittsburgh PA 15213-3890
59 * any improvements or extensions that they make and grant Carnegie the
60 * rights to redistribute these changes.
64 * Virtual memory mapping module.
67 #include <sys/cdefs.h>
68 __FBSDID("$FreeBSD$");
70 #include <sys/param.h>
71 #include <sys/systm.h>
73 #include <sys/kernel.h>
76 #include <sys/mutex.h>
78 #include <sys/vmmeter.h>
80 #include <sys/vnode.h>
81 #include <sys/racct.h>
82 #include <sys/resourcevar.h>
83 #include <sys/rwlock.h>
85 #include <sys/sysctl.h>
86 #include <sys/sysent.h>
90 #include <vm/vm_param.h>
92 #include <vm/vm_map.h>
93 #include <vm/vm_page.h>
94 #include <vm/vm_pageout.h>
95 #include <vm/vm_object.h>
96 #include <vm/vm_pager.h>
97 #include <vm/vm_kern.h>
98 #include <vm/vm_extern.h>
99 #include <vm/vnode_pager.h>
100 #include <vm/swap_pager.h>
104 * Virtual memory maps provide for the mapping, protection,
105 * and sharing of virtual memory objects. In addition,
106 * this module provides for an efficient virtual copy of
107 * memory from one map to another.
109 * Synchronization is required prior to most operations.
111 * Maps consist of an ordered doubly-linked list of simple
112 * entries; a self-adjusting binary search tree of these
113 * entries is used to speed up lookups.
115 * Since portions of maps are specified by start/end addresses,
116 * which may not align with existing map entries, all
117 * routines merely "clip" entries to these start/end values.
118 * [That is, an entry is split into two, bordering at a
119 * start or end value.] Note that these clippings may not
120 * always be necessary (as the two resulting entries are then
121 * not changed); however, the clipping is done for convenience.
123 * As mentioned above, virtual copy operations are performed
124 * by copying VM object references from one map to
125 * another, and then marking both regions as copy-on-write.
128 static struct mtx map_sleep_mtx;
129 static uma_zone_t mapentzone;
130 static uma_zone_t kmapentzone;
131 static uma_zone_t vmspace_zone;
132 static int vmspace_zinit(void *mem, int size, int flags);
133 static void _vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min,
135 static void vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map);
136 static void vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry);
137 static void vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry);
138 static int vm_map_growstack(vm_map_t map, vm_offset_t addr,
139 vm_map_entry_t gap_entry);
140 static void vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot,
141 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags);
143 static void vmspace_zdtor(void *mem, int size, void *arg);
145 static int vm_map_stack_locked(vm_map_t map, vm_offset_t addrbos,
146 vm_size_t max_ssize, vm_size_t growsize, vm_prot_t prot, vm_prot_t max,
148 static void vm_map_wire_entry_failure(vm_map_t map, vm_map_entry_t entry,
149 vm_offset_t failed_addr);
151 #define ENTRY_CHARGED(e) ((e)->cred != NULL || \
152 ((e)->object.vm_object != NULL && (e)->object.vm_object->cred != NULL && \
153 !((e)->eflags & MAP_ENTRY_NEEDS_COPY)))
156 * PROC_VMSPACE_{UN,}LOCK() can be a noop as long as vmspaces are type
159 #define PROC_VMSPACE_LOCK(p) do { } while (0)
160 #define PROC_VMSPACE_UNLOCK(p) do { } while (0)
163 * VM_MAP_RANGE_CHECK: [ internal use only ]
165 * Asserts that the starting and ending region
166 * addresses fall within the valid range of the map.
168 #define VM_MAP_RANGE_CHECK(map, start, end) \
170 if (start < vm_map_min(map)) \
171 start = vm_map_min(map); \
172 if (end > vm_map_max(map)) \
173 end = vm_map_max(map); \
181 * Initialize the vm_map module. Must be called before
182 * any other vm_map routines.
184 * Map and entry structures are allocated from the general
185 * purpose memory pool with some exceptions:
187 * - The kernel map and kmem submap are allocated statically.
188 * - Kernel map entries are allocated out of a static pool.
190 * These restrictions are necessary since malloc() uses the
191 * maps and requires map entries.
197 mtx_init(&map_sleep_mtx, "vm map sleep mutex", NULL, MTX_DEF);
198 kmapentzone = uma_zcreate("KMAP ENTRY", sizeof(struct vm_map_entry),
199 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR,
200 UMA_ZONE_MTXCLASS | UMA_ZONE_VM);
201 mapentzone = uma_zcreate("MAP ENTRY", sizeof(struct vm_map_entry),
202 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
203 vmspace_zone = uma_zcreate("VMSPACE", sizeof(struct vmspace), NULL,
209 vmspace_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
213 vmspace_zinit(void *mem, int size, int flags)
218 vm = (struct vmspace *)mem;
221 memset(map, 0, sizeof(*map));
222 mtx_init(&map->system_mtx, "vm map (system)", NULL,
223 MTX_DEF | MTX_DUPOK);
224 sx_init(&map->lock, "vm map (user)");
225 PMAP_LOCK_INIT(vmspace_pmap(vm));
231 vmspace_zdtor(void *mem, int size, void *arg)
235 vm = (struct vmspace *)mem;
236 KASSERT(vm->vm_map.nentries == 0,
237 ("vmspace %p nentries == %d on free", vm, vm->vm_map.nentries));
238 KASSERT(vm->vm_map.size == 0,
239 ("vmspace %p size == %ju on free", vm, (uintmax_t)vm->vm_map.size));
241 #endif /* INVARIANTS */
244 * Allocate a vmspace structure, including a vm_map and pmap,
245 * and initialize those structures. The refcnt is set to 1.
248 vmspace_alloc(vm_offset_t min, vm_offset_t max, pmap_pinit_t pinit)
252 vm = uma_zalloc(vmspace_zone, M_WAITOK);
253 KASSERT(vm->vm_map.pmap == NULL, ("vm_map.pmap must be NULL"));
254 if (!pinit(vmspace_pmap(vm))) {
255 uma_zfree(vmspace_zone, vm);
258 CTR1(KTR_VM, "vmspace_alloc: %p", vm);
259 _vm_map_init(&vm->vm_map, vmspace_pmap(vm), min, max);
274 vmspace_container_reset(struct proc *p)
278 racct_set(p, RACCT_DATA, 0);
279 racct_set(p, RACCT_STACK, 0);
280 racct_set(p, RACCT_RSS, 0);
281 racct_set(p, RACCT_MEMLOCK, 0);
282 racct_set(p, RACCT_VMEM, 0);
288 vmspace_dofree(struct vmspace *vm)
291 CTR1(KTR_VM, "vmspace_free: %p", vm);
294 * Make sure any SysV shm is freed, it might not have been in
300 * Lock the map, to wait out all other references to it.
301 * Delete all of the mappings and pages they hold, then call
302 * the pmap module to reclaim anything left.
304 (void)vm_map_remove(&vm->vm_map, vm_map_min(&vm->vm_map),
305 vm_map_max(&vm->vm_map));
307 pmap_release(vmspace_pmap(vm));
308 vm->vm_map.pmap = NULL;
309 uma_zfree(vmspace_zone, vm);
313 vmspace_free(struct vmspace *vm)
316 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
317 "vmspace_free() called");
319 if (vm->vm_refcnt == 0)
320 panic("vmspace_free: attempt to free already freed vmspace");
322 if (atomic_fetchadd_int(&vm->vm_refcnt, -1) == 1)
327 vmspace_exitfree(struct proc *p)
331 PROC_VMSPACE_LOCK(p);
334 PROC_VMSPACE_UNLOCK(p);
335 KASSERT(vm == &vmspace0, ("vmspace_exitfree: wrong vmspace"));
340 vmspace_exit(struct thread *td)
347 * Release user portion of address space.
348 * This releases references to vnodes,
349 * which could cause I/O if the file has been unlinked.
350 * Need to do this early enough that we can still sleep.
352 * The last exiting process to reach this point releases as
353 * much of the environment as it can. vmspace_dofree() is the
354 * slower fallback in case another process had a temporary
355 * reference to the vmspace.
360 atomic_add_int(&vmspace0.vm_refcnt, 1);
361 refcnt = vm->vm_refcnt;
363 if (refcnt > 1 && p->p_vmspace != &vmspace0) {
364 /* Switch now since other proc might free vmspace */
365 PROC_VMSPACE_LOCK(p);
366 p->p_vmspace = &vmspace0;
367 PROC_VMSPACE_UNLOCK(p);
370 } while (!atomic_fcmpset_int(&vm->vm_refcnt, &refcnt, refcnt - 1));
372 if (p->p_vmspace != vm) {
373 /* vmspace not yet freed, switch back */
374 PROC_VMSPACE_LOCK(p);
376 PROC_VMSPACE_UNLOCK(p);
379 pmap_remove_pages(vmspace_pmap(vm));
380 /* Switch now since this proc will free vmspace */
381 PROC_VMSPACE_LOCK(p);
382 p->p_vmspace = &vmspace0;
383 PROC_VMSPACE_UNLOCK(p);
389 vmspace_container_reset(p);
393 /* Acquire reference to vmspace owned by another process. */
396 vmspace_acquire_ref(struct proc *p)
401 PROC_VMSPACE_LOCK(p);
404 PROC_VMSPACE_UNLOCK(p);
407 refcnt = vm->vm_refcnt;
409 if (refcnt <= 0) { /* Avoid 0->1 transition */
410 PROC_VMSPACE_UNLOCK(p);
413 } while (!atomic_fcmpset_int(&vm->vm_refcnt, &refcnt, refcnt + 1));
414 if (vm != p->p_vmspace) {
415 PROC_VMSPACE_UNLOCK(p);
419 PROC_VMSPACE_UNLOCK(p);
424 * Switch between vmspaces in an AIO kernel process.
426 * The new vmspace is either the vmspace of a user process obtained
427 * from an active AIO request or the initial vmspace of the AIO kernel
428 * process (when it is idling). Because user processes will block to
429 * drain any active AIO requests before proceeding in exit() or
430 * execve(), the reference count for vmspaces from AIO requests can
431 * never be 0. Similarly, AIO kernel processes hold an extra
432 * reference on their initial vmspace for the life of the process. As
433 * a result, the 'newvm' vmspace always has a non-zero reference
434 * count. This permits an additional reference on 'newvm' to be
435 * acquired via a simple atomic increment rather than the loop in
436 * vmspace_acquire_ref() above.
439 vmspace_switch_aio(struct vmspace *newvm)
441 struct vmspace *oldvm;
443 /* XXX: Need some way to assert that this is an aio daemon. */
445 KASSERT(newvm->vm_refcnt > 0,
446 ("vmspace_switch_aio: newvm unreferenced"));
448 oldvm = curproc->p_vmspace;
453 * Point to the new address space and refer to it.
455 curproc->p_vmspace = newvm;
456 atomic_add_int(&newvm->vm_refcnt, 1);
458 /* Activate the new mapping. */
459 pmap_activate(curthread);
465 _vm_map_lock(vm_map_t map, const char *file, int line)
469 mtx_lock_flags_(&map->system_mtx, 0, file, line);
471 sx_xlock_(&map->lock, file, line);
476 vm_map_entry_set_vnode_text(vm_map_entry_t entry, bool add)
482 if ((entry->eflags & MAP_ENTRY_VN_EXEC) == 0)
484 KASSERT((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0,
485 ("Submap with execs"));
486 object = entry->object.vm_object;
487 KASSERT(object != NULL, ("No object for text, entry %p", entry));
488 if ((object->flags & OBJ_ANON) != 0)
489 object = object->handle;
491 KASSERT(object->backing_object == NULL,
492 ("non-anon object %p shadows", object));
493 KASSERT(object != NULL, ("No content object for text, entry %p obj %p",
494 entry, entry->object.vm_object));
497 * Mostly, we do not lock the backing object. It is
498 * referenced by the entry we are processing, so it cannot go
503 if (object->type == OBJT_DEAD) {
505 * For OBJT_DEAD objects, v_writecount was handled in
506 * vnode_pager_dealloc().
508 } else if (object->type == OBJT_VNODE) {
510 } else if (object->type == OBJT_SWAP) {
511 KASSERT((object->flags & OBJ_TMPFS_NODE) != 0,
512 ("vm_map_entry_set_vnode_text: swap and !TMPFS "
513 "entry %p, object %p, add %d", entry, object, add));
515 * Tmpfs VREG node, which was reclaimed, has
516 * OBJ_TMPFS_NODE flag set, but not OBJ_TMPFS. In
517 * this case there is no v_writecount to adjust.
519 VM_OBJECT_RLOCK(object);
520 if ((object->flags & OBJ_TMPFS) != 0) {
521 vp = object->un_pager.swp.swp_tmpfs;
527 VM_OBJECT_RUNLOCK(object);
530 ("vm_map_entry_set_vnode_text: wrong object type, "
531 "entry %p, object %p, add %d", entry, object, add));
535 VOP_SET_TEXT_CHECKED(vp);
537 vn_lock(vp, LK_SHARED | LK_RETRY);
538 VOP_UNSET_TEXT_CHECKED(vp);
547 * Use a different name for this vm_map_entry field when it's use
548 * is not consistent with its use as part of an ordered search tree.
550 #define defer_next right
553 vm_map_process_deferred(void)
556 vm_map_entry_t entry, next;
560 entry = td->td_map_def_user;
561 td->td_map_def_user = NULL;
562 while (entry != NULL) {
563 next = entry->defer_next;
564 MPASS((entry->eflags & (MAP_ENTRY_WRITECNT |
565 MAP_ENTRY_VN_EXEC)) != (MAP_ENTRY_WRITECNT |
567 if ((entry->eflags & MAP_ENTRY_WRITECNT) != 0) {
569 * Decrement the object's writemappings and
570 * possibly the vnode's v_writecount.
572 KASSERT((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0,
573 ("Submap with writecount"));
574 object = entry->object.vm_object;
575 KASSERT(object != NULL, ("No object for writecount"));
576 vm_pager_release_writecount(object, entry->start,
579 vm_map_entry_set_vnode_text(entry, false);
580 vm_map_entry_deallocate(entry, FALSE);
587 _vm_map_assert_locked(vm_map_t map, const char *file, int line)
591 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
593 sx_assert_(&map->lock, SA_XLOCKED, file, line);
596 #define VM_MAP_ASSERT_LOCKED(map) \
597 _vm_map_assert_locked(map, LOCK_FILE, LOCK_LINE)
599 enum { VMMAP_CHECK_NONE, VMMAP_CHECK_UNLOCK, VMMAP_CHECK_ALL };
601 static int enable_vmmap_check = VMMAP_CHECK_UNLOCK;
603 static int enable_vmmap_check = VMMAP_CHECK_NONE;
605 SYSCTL_INT(_debug, OID_AUTO, vmmap_check, CTLFLAG_RWTUN,
606 &enable_vmmap_check, 0, "Enable vm map consistency checking");
608 static void _vm_map_assert_consistent(vm_map_t map, int check);
610 #define VM_MAP_ASSERT_CONSISTENT(map) \
611 _vm_map_assert_consistent(map, VMMAP_CHECK_ALL)
613 #define VM_MAP_UNLOCK_CONSISTENT(map) do { \
614 if (map->nupdates > map->nentries) { \
615 _vm_map_assert_consistent(map, VMMAP_CHECK_UNLOCK); \
620 #define VM_MAP_UNLOCK_CONSISTENT(map)
623 #define VM_MAP_ASSERT_LOCKED(map)
624 #define VM_MAP_ASSERT_CONSISTENT(map)
625 #define VM_MAP_UNLOCK_CONSISTENT(map)
626 #endif /* INVARIANTS */
629 _vm_map_unlock(vm_map_t map, const char *file, int line)
632 VM_MAP_UNLOCK_CONSISTENT(map);
634 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
636 sx_xunlock_(&map->lock, file, line);
637 vm_map_process_deferred();
642 _vm_map_lock_read(vm_map_t map, const char *file, int line)
646 mtx_lock_flags_(&map->system_mtx, 0, file, line);
648 sx_slock_(&map->lock, file, line);
652 _vm_map_unlock_read(vm_map_t map, const char *file, int line)
656 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
658 sx_sunlock_(&map->lock, file, line);
659 vm_map_process_deferred();
664 _vm_map_trylock(vm_map_t map, const char *file, int line)
668 error = map->system_map ?
669 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) :
670 !sx_try_xlock_(&map->lock, file, line);
677 _vm_map_trylock_read(vm_map_t map, const char *file, int line)
681 error = map->system_map ?
682 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) :
683 !sx_try_slock_(&map->lock, file, line);
688 * _vm_map_lock_upgrade: [ internal use only ]
690 * Tries to upgrade a read (shared) lock on the specified map to a write
691 * (exclusive) lock. Returns the value "0" if the upgrade succeeds and a
692 * non-zero value if the upgrade fails. If the upgrade fails, the map is
693 * returned without a read or write lock held.
695 * Requires that the map be read locked.
698 _vm_map_lock_upgrade(vm_map_t map, const char *file, int line)
700 unsigned int last_timestamp;
702 if (map->system_map) {
703 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
705 if (!sx_try_upgrade_(&map->lock, file, line)) {
706 last_timestamp = map->timestamp;
707 sx_sunlock_(&map->lock, file, line);
708 vm_map_process_deferred();
710 * If the map's timestamp does not change while the
711 * map is unlocked, then the upgrade succeeds.
713 sx_xlock_(&map->lock, file, line);
714 if (last_timestamp != map->timestamp) {
715 sx_xunlock_(&map->lock, file, line);
725 _vm_map_lock_downgrade(vm_map_t map, const char *file, int line)
728 if (map->system_map) {
729 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
731 VM_MAP_UNLOCK_CONSISTENT(map);
732 sx_downgrade_(&map->lock, file, line);
739 * Returns a non-zero value if the caller holds a write (exclusive) lock
740 * on the specified map and the value "0" otherwise.
743 vm_map_locked(vm_map_t map)
747 return (mtx_owned(&map->system_mtx));
749 return (sx_xlocked(&map->lock));
753 * _vm_map_unlock_and_wait:
755 * Atomically releases the lock on the specified map and puts the calling
756 * thread to sleep. The calling thread will remain asleep until either
757 * vm_map_wakeup() is performed on the map or the specified timeout is
760 * WARNING! This function does not perform deferred deallocations of
761 * objects and map entries. Therefore, the calling thread is expected to
762 * reacquire the map lock after reawakening and later perform an ordinary
763 * unlock operation, such as vm_map_unlock(), before completing its
764 * operation on the map.
767 _vm_map_unlock_and_wait(vm_map_t map, int timo, const char *file, int line)
770 VM_MAP_UNLOCK_CONSISTENT(map);
771 mtx_lock(&map_sleep_mtx);
773 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
775 sx_xunlock_(&map->lock, file, line);
776 return (msleep(&map->root, &map_sleep_mtx, PDROP | PVM, "vmmaps",
783 * Awaken any threads that have slept on the map using
784 * vm_map_unlock_and_wait().
787 vm_map_wakeup(vm_map_t map)
791 * Acquire and release map_sleep_mtx to prevent a wakeup()
792 * from being performed (and lost) between the map unlock
793 * and the msleep() in _vm_map_unlock_and_wait().
795 mtx_lock(&map_sleep_mtx);
796 mtx_unlock(&map_sleep_mtx);
801 vm_map_busy(vm_map_t map)
804 VM_MAP_ASSERT_LOCKED(map);
809 vm_map_unbusy(vm_map_t map)
812 VM_MAP_ASSERT_LOCKED(map);
813 KASSERT(map->busy, ("vm_map_unbusy: not busy"));
814 if (--map->busy == 0 && (map->flags & MAP_BUSY_WAKEUP)) {
815 vm_map_modflags(map, 0, MAP_BUSY_WAKEUP);
821 vm_map_wait_busy(vm_map_t map)
824 VM_MAP_ASSERT_LOCKED(map);
826 vm_map_modflags(map, MAP_BUSY_WAKEUP, 0);
828 msleep(&map->busy, &map->system_mtx, 0, "mbusy", 0);
830 sx_sleep(&map->busy, &map->lock, 0, "mbusy", 0);
836 vmspace_resident_count(struct vmspace *vmspace)
838 return pmap_resident_count(vmspace_pmap(vmspace));
842 * Initialize an existing vm_map structure
843 * such as that in the vmspace structure.
846 _vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max)
849 map->header.eflags = MAP_ENTRY_HEADER;
850 map->needs_wakeup = FALSE;
853 map->header.end = min;
854 map->header.start = max;
856 map->header.left = map->header.right = &map->header;
867 vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max)
870 _vm_map_init(map, pmap, min, max);
871 mtx_init(&map->system_mtx, "vm map (system)", NULL,
872 MTX_DEF | MTX_DUPOK);
873 sx_init(&map->lock, "vm map (user)");
877 * vm_map_entry_dispose: [ internal use only ]
879 * Inverse of vm_map_entry_create.
882 vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry)
884 uma_zfree(map->system_map ? kmapentzone : mapentzone, entry);
888 * vm_map_entry_create: [ internal use only ]
890 * Allocates a VM map entry for insertion.
891 * No entry fields are filled in.
893 static vm_map_entry_t
894 vm_map_entry_create(vm_map_t map)
896 vm_map_entry_t new_entry;
899 new_entry = uma_zalloc(kmapentzone, M_NOWAIT);
901 new_entry = uma_zalloc(mapentzone, M_WAITOK);
902 if (new_entry == NULL)
903 panic("vm_map_entry_create: kernel resources exhausted");
908 * vm_map_entry_set_behavior:
910 * Set the expected access behavior, either normal, random, or
914 vm_map_entry_set_behavior(vm_map_entry_t entry, u_char behavior)
916 entry->eflags = (entry->eflags & ~MAP_ENTRY_BEHAV_MASK) |
917 (behavior & MAP_ENTRY_BEHAV_MASK);
921 * vm_map_entry_max_free_{left,right}:
923 * Compute the size of the largest free gap between two entries,
924 * one the root of a tree and the other the ancestor of that root
925 * that is the least or greatest ancestor found on the search path.
927 static inline vm_size_t
928 vm_map_entry_max_free_left(vm_map_entry_t root, vm_map_entry_t left_ancestor)
931 return (root->left != left_ancestor ?
932 root->left->max_free : root->start - left_ancestor->end);
935 static inline vm_size_t
936 vm_map_entry_max_free_right(vm_map_entry_t root, vm_map_entry_t right_ancestor)
939 return (root->right != right_ancestor ?
940 root->right->max_free : right_ancestor->start - root->end);
944 * vm_map_entry_{pred,succ}:
946 * Find the {predecessor, successor} of the entry by taking one step
947 * in the appropriate direction and backtracking as much as necessary.
948 * vm_map_entry_succ is defined in vm_map.h.
950 static inline vm_map_entry_t
951 vm_map_entry_pred(vm_map_entry_t entry)
953 vm_map_entry_t prior;
956 if (prior->right->start < entry->start) {
958 prior = prior->right;
959 while (prior->right != entry);
964 static inline vm_size_t
965 vm_size_max(vm_size_t a, vm_size_t b)
968 return (a > b ? a : b);
971 #define SPLAY_LEFT_STEP(root, y, llist, rlist, test) do { \
973 vm_size_t max_free; \
976 * Infer root->right->max_free == root->max_free when \
977 * y->max_free < root->max_free || root->max_free == 0. \
978 * Otherwise, look right to find it. \
981 max_free = root->max_free; \
982 KASSERT(max_free == vm_size_max( \
983 vm_map_entry_max_free_left(root, llist), \
984 vm_map_entry_max_free_right(root, rlist)), \
985 ("%s: max_free invariant fails", __func__)); \
986 if (max_free - 1 < vm_map_entry_max_free_left(root, llist)) \
987 max_free = vm_map_entry_max_free_right(root, rlist); \
988 if (y != llist && (test)) { \
989 /* Rotate right and make y root. */ \
994 if (max_free < y->max_free) \
995 root->max_free = max_free = \
996 vm_size_max(max_free, z->max_free); \
997 } else if (max_free < y->max_free) \
998 root->max_free = max_free = \
999 vm_size_max(max_free, root->start - y->end);\
1003 /* Copy right->max_free. Put root on rlist. */ \
1004 root->max_free = max_free; \
1005 KASSERT(max_free == vm_map_entry_max_free_right(root, rlist), \
1006 ("%s: max_free not copied from right", __func__)); \
1007 root->left = rlist; \
1009 root = y != llist ? y : NULL; \
1012 #define SPLAY_RIGHT_STEP(root, y, llist, rlist, test) do { \
1014 vm_size_t max_free; \
1017 * Infer root->left->max_free == root->max_free when \
1018 * y->max_free < root->max_free || root->max_free == 0. \
1019 * Otherwise, look left to find it. \
1022 max_free = root->max_free; \
1023 KASSERT(max_free == vm_size_max( \
1024 vm_map_entry_max_free_left(root, llist), \
1025 vm_map_entry_max_free_right(root, rlist)), \
1026 ("%s: max_free invariant fails", __func__)); \
1027 if (max_free - 1 < vm_map_entry_max_free_right(root, rlist)) \
1028 max_free = vm_map_entry_max_free_left(root, llist); \
1029 if (y != rlist && (test)) { \
1030 /* Rotate left and make y root. */ \
1035 if (max_free < y->max_free) \
1036 root->max_free = max_free = \
1037 vm_size_max(max_free, z->max_free); \
1038 } else if (max_free < y->max_free) \
1039 root->max_free = max_free = \
1040 vm_size_max(max_free, y->start - root->end);\
1044 /* Copy left->max_free. Put root on llist. */ \
1045 root->max_free = max_free; \
1046 KASSERT(max_free == vm_map_entry_max_free_left(root, llist), \
1047 ("%s: max_free not copied from left", __func__)); \
1048 root->right = llist; \
1050 root = y != rlist ? y : NULL; \
1054 * Walk down the tree until we find addr or a gap where addr would go, breaking
1055 * off left and right subtrees of nodes less than, or greater than addr. Treat
1056 * subtrees with root->max_free < length as empty trees. llist and rlist are
1057 * the two sides in reverse order (bottom-up), with llist linked by the right
1058 * pointer and rlist linked by the left pointer in the vm_map_entry, and both
1059 * lists terminated by &map->header. This function, and the subsequent call to
1060 * vm_map_splay_merge_{left,right,pred,succ}, rely on the start and end address
1061 * values in &map->header.
1063 static __always_inline vm_map_entry_t
1064 vm_map_splay_split(vm_map_t map, vm_offset_t addr, vm_size_t length,
1065 vm_map_entry_t *llist, vm_map_entry_t *rlist)
1067 vm_map_entry_t left, right, root, y;
1069 left = right = &map->header;
1071 while (root != NULL && root->max_free >= length) {
1072 KASSERT(left->end <= root->start &&
1073 root->end <= right->start,
1074 ("%s: root not within tree bounds", __func__));
1075 if (addr < root->start) {
1076 SPLAY_LEFT_STEP(root, y, left, right,
1077 y->max_free >= length && addr < y->start);
1078 } else if (addr >= root->end) {
1079 SPLAY_RIGHT_STEP(root, y, left, right,
1080 y->max_free >= length && addr >= y->end);
1089 static __always_inline void
1090 vm_map_splay_findnext(vm_map_entry_t root, vm_map_entry_t *rlist)
1092 vm_map_entry_t hi, right, y;
1095 hi = root->right == right ? NULL : root->right;
1099 SPLAY_LEFT_STEP(hi, y, root, right, true);
1104 static __always_inline void
1105 vm_map_splay_findprev(vm_map_entry_t root, vm_map_entry_t *llist)
1107 vm_map_entry_t left, lo, y;
1110 lo = root->left == left ? NULL : root->left;
1114 SPLAY_RIGHT_STEP(lo, y, left, root, true);
1120 vm_map_entry_swap(vm_map_entry_t *a, vm_map_entry_t *b)
1130 * Walk back up the two spines, flip the pointers and set max_free. The
1131 * subtrees of the root go at the bottom of llist and rlist.
1134 vm_map_splay_merge_left_walk(vm_map_entry_t header, vm_map_entry_t root,
1135 vm_map_entry_t tail, vm_size_t max_free, vm_map_entry_t llist)
1139 * The max_free values of the children of llist are in
1140 * llist->max_free and max_free. Update with the
1143 llist->max_free = max_free =
1144 vm_size_max(llist->max_free, max_free);
1145 vm_map_entry_swap(&llist->right, &tail);
1146 vm_map_entry_swap(&tail, &llist);
1147 } while (llist != header);
1153 * When llist is known to be the predecessor of root.
1155 static inline vm_size_t
1156 vm_map_splay_merge_pred(vm_map_entry_t header, vm_map_entry_t root,
1157 vm_map_entry_t llist)
1161 max_free = root->start - llist->end;
1162 if (llist != header) {
1163 max_free = vm_map_splay_merge_left_walk(header, root,
1164 root, max_free, llist);
1166 root->left = header;
1167 header->right = root;
1173 * When llist may or may not be the predecessor of root.
1175 static inline vm_size_t
1176 vm_map_splay_merge_left(vm_map_entry_t header, vm_map_entry_t root,
1177 vm_map_entry_t llist)
1181 max_free = vm_map_entry_max_free_left(root, llist);
1182 if (llist != header) {
1183 max_free = vm_map_splay_merge_left_walk(header, root,
1184 root->left == llist ? root : root->left,
1191 vm_map_splay_merge_right_walk(vm_map_entry_t header, vm_map_entry_t root,
1192 vm_map_entry_t tail, vm_size_t max_free, vm_map_entry_t rlist)
1196 * The max_free values of the children of rlist are in
1197 * rlist->max_free and max_free. Update with the
1200 rlist->max_free = max_free =
1201 vm_size_max(rlist->max_free, max_free);
1202 vm_map_entry_swap(&rlist->left, &tail);
1203 vm_map_entry_swap(&tail, &rlist);
1204 } while (rlist != header);
1210 * When rlist is known to be the succecessor of root.
1212 static inline vm_size_t
1213 vm_map_splay_merge_succ(vm_map_entry_t header, vm_map_entry_t root,
1214 vm_map_entry_t rlist)
1218 max_free = rlist->start - root->end;
1219 if (rlist != header) {
1220 max_free = vm_map_splay_merge_right_walk(header, root,
1221 root, max_free, rlist);
1223 root->right = header;
1224 header->left = root;
1230 * When rlist may or may not be the succecessor of root.
1232 static inline vm_size_t
1233 vm_map_splay_merge_right(vm_map_entry_t header, vm_map_entry_t root,
1234 vm_map_entry_t rlist)
1238 max_free = vm_map_entry_max_free_right(root, rlist);
1239 if (rlist != header) {
1240 max_free = vm_map_splay_merge_right_walk(header, root,
1241 root->right == rlist ? root : root->right,
1250 * The Sleator and Tarjan top-down splay algorithm with the
1251 * following variation. Max_free must be computed bottom-up, so
1252 * on the downward pass, maintain the left and right spines in
1253 * reverse order. Then, make a second pass up each side to fix
1254 * the pointers and compute max_free. The time bound is O(log n)
1257 * The tree is threaded, which means that there are no null pointers.
1258 * When a node has no left child, its left pointer points to its
1259 * predecessor, which the last ancestor on the search path from the root
1260 * where the search branched right. Likewise, when a node has no right
1261 * child, its right pointer points to its successor. The map header node
1262 * is the predecessor of the first map entry, and the successor of the
1265 * The new root is the vm_map_entry containing "addr", or else an
1266 * adjacent entry (lower if possible) if addr is not in the tree.
1268 * The map must be locked, and leaves it so.
1270 * Returns: the new root.
1272 static vm_map_entry_t
1273 vm_map_splay(vm_map_t map, vm_offset_t addr)
1275 vm_map_entry_t header, llist, rlist, root;
1276 vm_size_t max_free_left, max_free_right;
1278 header = &map->header;
1279 root = vm_map_splay_split(map, addr, 0, &llist, &rlist);
1281 max_free_left = vm_map_splay_merge_left(header, root, llist);
1282 max_free_right = vm_map_splay_merge_right(header, root, rlist);
1283 } else if (llist != header) {
1285 * Recover the greatest node in the left
1286 * subtree and make it the root.
1289 llist = root->right;
1290 max_free_left = vm_map_splay_merge_left(header, root, llist);
1291 max_free_right = vm_map_splay_merge_succ(header, root, rlist);
1292 } else if (rlist != header) {
1294 * Recover the least node in the right
1295 * subtree and make it the root.
1299 max_free_left = vm_map_splay_merge_pred(header, root, llist);
1300 max_free_right = vm_map_splay_merge_right(header, root, rlist);
1302 /* There is no root. */
1305 root->max_free = vm_size_max(max_free_left, max_free_right);
1307 VM_MAP_ASSERT_CONSISTENT(map);
1312 * vm_map_entry_{un,}link:
1314 * Insert/remove entries from maps. On linking, if new entry clips
1315 * existing entry, trim existing entry to avoid overlap, and manage
1316 * offsets. On unlinking, merge disappearing entry with neighbor, if
1317 * called for, and manage offsets. Callers should not modify fields in
1318 * entries already mapped.
1321 vm_map_entry_link(vm_map_t map, vm_map_entry_t entry)
1323 vm_map_entry_t header, llist, rlist, root;
1324 vm_size_t max_free_left, max_free_right;
1327 "vm_map_entry_link: map %p, nentries %d, entry %p", map,
1328 map->nentries, entry);
1329 VM_MAP_ASSERT_LOCKED(map);
1331 header = &map->header;
1332 root = vm_map_splay_split(map, entry->start, 0, &llist, &rlist);
1335 * The new entry does not overlap any existing entry in the
1336 * map, so it becomes the new root of the map tree.
1338 max_free_left = vm_map_splay_merge_pred(header, entry, llist);
1339 max_free_right = vm_map_splay_merge_succ(header, entry, rlist);
1340 } else if (entry->start == root->start) {
1342 * The new entry is a clone of root, with only the end field
1343 * changed. The root entry will be shrunk to abut the new
1344 * entry, and will be the right child of the new root entry in
1347 KASSERT(entry->end < root->end,
1348 ("%s: clip_start not within entry", __func__));
1349 vm_map_splay_findprev(root, &llist);
1350 root->offset += entry->end - root->start;
1351 root->start = entry->end;
1352 max_free_left = vm_map_splay_merge_pred(header, entry, llist);
1353 max_free_right = root->max_free = vm_size_max(
1354 vm_map_splay_merge_pred(entry, root, entry),
1355 vm_map_splay_merge_right(header, root, rlist));
1358 * The new entry is a clone of root, with only the start field
1359 * changed. The root entry will be shrunk to abut the new
1360 * entry, and will be the left child of the new root entry in
1363 KASSERT(entry->end == root->end,
1364 ("%s: clip_start not within entry", __func__));
1365 vm_map_splay_findnext(root, &rlist);
1366 entry->offset += entry->start - root->start;
1367 root->end = entry->start;
1368 max_free_left = root->max_free = vm_size_max(
1369 vm_map_splay_merge_left(header, root, llist),
1370 vm_map_splay_merge_succ(entry, root, entry));
1371 max_free_right = vm_map_splay_merge_succ(header, entry, rlist);
1373 entry->max_free = vm_size_max(max_free_left, max_free_right);
1375 VM_MAP_ASSERT_CONSISTENT(map);
1378 enum unlink_merge_type {
1384 vm_map_entry_unlink(vm_map_t map, vm_map_entry_t entry,
1385 enum unlink_merge_type op)
1387 vm_map_entry_t header, llist, rlist, root;
1388 vm_size_t max_free_left, max_free_right;
1390 VM_MAP_ASSERT_LOCKED(map);
1391 header = &map->header;
1392 root = vm_map_splay_split(map, entry->start, 0, &llist, &rlist);
1393 KASSERT(root != NULL,
1394 ("vm_map_entry_unlink: unlink object not mapped"));
1396 vm_map_splay_findprev(root, &llist);
1397 vm_map_splay_findnext(root, &rlist);
1398 if (op == UNLINK_MERGE_NEXT) {
1399 rlist->start = root->start;
1400 rlist->offset = root->offset;
1402 if (llist != header) {
1404 llist = root->right;
1405 max_free_left = vm_map_splay_merge_left(header, root, llist);
1406 max_free_right = vm_map_splay_merge_succ(header, root, rlist);
1407 } else if (rlist != header) {
1410 max_free_left = vm_map_splay_merge_pred(header, root, llist);
1411 max_free_right = vm_map_splay_merge_right(header, root, rlist);
1413 header->left = header->right = header;
1417 root->max_free = vm_size_max(max_free_left, max_free_right);
1419 VM_MAP_ASSERT_CONSISTENT(map);
1421 CTR3(KTR_VM, "vm_map_entry_unlink: map %p, nentries %d, entry %p", map,
1422 map->nentries, entry);
1426 * vm_map_entry_resize:
1428 * Resize a vm_map_entry, recompute the amount of free space that
1429 * follows it and propagate that value up the tree.
1431 * The map must be locked, and leaves it so.
1434 vm_map_entry_resize(vm_map_t map, vm_map_entry_t entry, vm_size_t grow_amount)
1436 vm_map_entry_t header, llist, rlist, root;
1438 VM_MAP_ASSERT_LOCKED(map);
1439 header = &map->header;
1440 root = vm_map_splay_split(map, entry->start, 0, &llist, &rlist);
1441 KASSERT(root != NULL, ("%s: resize object not mapped", __func__));
1442 vm_map_splay_findnext(root, &rlist);
1443 entry->end += grow_amount;
1444 root->max_free = vm_size_max(
1445 vm_map_splay_merge_left(header, root, llist),
1446 vm_map_splay_merge_succ(header, root, rlist));
1448 VM_MAP_ASSERT_CONSISTENT(map);
1449 CTR4(KTR_VM, "%s: map %p, nentries %d, entry %p",
1450 __func__, map, map->nentries, entry);
1454 * vm_map_lookup_entry: [ internal use only ]
1456 * Finds the map entry containing (or
1457 * immediately preceding) the specified address
1458 * in the given map; the entry is returned
1459 * in the "entry" parameter. The boolean
1460 * result indicates whether the address is
1461 * actually contained in the map.
1464 vm_map_lookup_entry(
1466 vm_offset_t address,
1467 vm_map_entry_t *entry) /* OUT */
1469 vm_map_entry_t cur, header, lbound, ubound;
1473 * If the map is empty, then the map entry immediately preceding
1474 * "address" is the map's header.
1476 header = &map->header;
1482 if (address >= cur->start && cur->end > address) {
1486 if ((locked = vm_map_locked(map)) ||
1487 sx_try_upgrade(&map->lock)) {
1489 * Splay requires a write lock on the map. However, it only
1490 * restructures the binary search tree; it does not otherwise
1491 * change the map. Thus, the map's timestamp need not change
1492 * on a temporary upgrade.
1494 cur = vm_map_splay(map, address);
1496 VM_MAP_UNLOCK_CONSISTENT(map);
1497 sx_downgrade(&map->lock);
1501 * If "address" is contained within a map entry, the new root
1502 * is that map entry. Otherwise, the new root is a map entry
1503 * immediately before or after "address".
1505 if (address < cur->start) {
1510 return (address < cur->end);
1513 * Since the map is only locked for read access, perform a
1514 * standard binary search tree lookup for "address".
1516 lbound = ubound = header;
1518 if (address < cur->start) {
1523 } else if (cur->end <= address) {
1540 * Inserts the given whole VM object into the target
1541 * map at the specified address range. The object's
1542 * size should match that of the address range.
1544 * Requires that the map be locked, and leaves it so.
1546 * If object is non-NULL, ref count must be bumped by caller
1547 * prior to making call to account for the new entry.
1550 vm_map_insert(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1551 vm_offset_t start, vm_offset_t end, vm_prot_t prot, vm_prot_t max, int cow)
1553 vm_map_entry_t new_entry, next_entry, prev_entry;
1555 vm_eflags_t protoeflags;
1556 vm_inherit_t inheritance;
1560 VM_MAP_ASSERT_LOCKED(map);
1561 KASSERT(object != kernel_object ||
1562 (cow & MAP_COPY_ON_WRITE) == 0,
1563 ("vm_map_insert: kernel object and COW"));
1564 KASSERT(object == NULL || (cow & MAP_NOFAULT) == 0 ||
1565 (cow & MAP_SPLIT_BOUNDARY_MASK) != 0,
1566 ("vm_map_insert: paradoxical MAP_NOFAULT request, obj %p cow %#x",
1568 KASSERT((prot & ~max) == 0,
1569 ("prot %#x is not subset of max_prot %#x", prot, max));
1572 * Check that the start and end points are not bogus.
1574 if (start == end || !vm_map_range_valid(map, start, end))
1575 return (KERN_INVALID_ADDRESS);
1578 * Find the entry prior to the proposed starting address; if it's part
1579 * of an existing entry, this range is bogus.
1581 if (vm_map_lookup_entry(map, start, &prev_entry))
1582 return (KERN_NO_SPACE);
1585 * Assert that the next entry doesn't overlap the end point.
1587 next_entry = vm_map_entry_succ(prev_entry);
1588 if (next_entry->start < end)
1589 return (KERN_NO_SPACE);
1591 if ((cow & MAP_CREATE_GUARD) != 0 && (object != NULL ||
1592 max != VM_PROT_NONE))
1593 return (KERN_INVALID_ARGUMENT);
1596 if (cow & MAP_COPY_ON_WRITE)
1597 protoeflags |= MAP_ENTRY_COW | MAP_ENTRY_NEEDS_COPY;
1598 if (cow & MAP_NOFAULT)
1599 protoeflags |= MAP_ENTRY_NOFAULT;
1600 if (cow & MAP_DISABLE_SYNCER)
1601 protoeflags |= MAP_ENTRY_NOSYNC;
1602 if (cow & MAP_DISABLE_COREDUMP)
1603 protoeflags |= MAP_ENTRY_NOCOREDUMP;
1604 if (cow & MAP_STACK_GROWS_DOWN)
1605 protoeflags |= MAP_ENTRY_GROWS_DOWN;
1606 if (cow & MAP_STACK_GROWS_UP)
1607 protoeflags |= MAP_ENTRY_GROWS_UP;
1608 if (cow & MAP_WRITECOUNT)
1609 protoeflags |= MAP_ENTRY_WRITECNT;
1610 if (cow & MAP_VN_EXEC)
1611 protoeflags |= MAP_ENTRY_VN_EXEC;
1612 if ((cow & MAP_CREATE_GUARD) != 0)
1613 protoeflags |= MAP_ENTRY_GUARD;
1614 if ((cow & MAP_CREATE_STACK_GAP_DN) != 0)
1615 protoeflags |= MAP_ENTRY_STACK_GAP_DN;
1616 if ((cow & MAP_CREATE_STACK_GAP_UP) != 0)
1617 protoeflags |= MAP_ENTRY_STACK_GAP_UP;
1618 if (cow & MAP_INHERIT_SHARE)
1619 inheritance = VM_INHERIT_SHARE;
1621 inheritance = VM_INHERIT_DEFAULT;
1622 if ((cow & MAP_SPLIT_BOUNDARY_MASK) != 0) {
1623 /* This magically ignores index 0, for usual page size. */
1624 bidx = (cow & MAP_SPLIT_BOUNDARY_MASK) >>
1625 MAP_SPLIT_BOUNDARY_SHIFT;
1626 if (bidx >= MAXPAGESIZES)
1627 return (KERN_INVALID_ARGUMENT);
1628 bdry = pagesizes[bidx] - 1;
1629 if ((start & bdry) != 0 || (end & bdry) != 0)
1630 return (KERN_INVALID_ARGUMENT);
1631 protoeflags |= bidx << MAP_ENTRY_SPLIT_BOUNDARY_SHIFT;
1635 if ((cow & (MAP_ACC_NO_CHARGE | MAP_NOFAULT | MAP_CREATE_GUARD)) != 0)
1637 if ((cow & MAP_ACC_CHARGED) || ((prot & VM_PROT_WRITE) &&
1638 ((protoeflags & MAP_ENTRY_NEEDS_COPY) || object == NULL))) {
1639 if (!(cow & MAP_ACC_CHARGED) && !swap_reserve(end - start))
1640 return (KERN_RESOURCE_SHORTAGE);
1641 KASSERT(object == NULL ||
1642 (protoeflags & MAP_ENTRY_NEEDS_COPY) != 0 ||
1643 object->cred == NULL,
1644 ("overcommit: vm_map_insert o %p", object));
1645 cred = curthread->td_ucred;
1649 /* Expand the kernel pmap, if necessary. */
1650 if (map == kernel_map && end > kernel_vm_end)
1651 pmap_growkernel(end);
1652 if (object != NULL) {
1654 * OBJ_ONEMAPPING must be cleared unless this mapping
1655 * is trivially proven to be the only mapping for any
1656 * of the object's pages. (Object granularity
1657 * reference counting is insufficient to recognize
1658 * aliases with precision.)
1660 if ((object->flags & OBJ_ANON) != 0) {
1661 VM_OBJECT_WLOCK(object);
1662 if (object->ref_count > 1 || object->shadow_count != 0)
1663 vm_object_clear_flag(object, OBJ_ONEMAPPING);
1664 VM_OBJECT_WUNLOCK(object);
1666 } else if ((prev_entry->eflags & ~MAP_ENTRY_USER_WIRED) ==
1668 (cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP |
1669 MAP_VN_EXEC)) == 0 &&
1670 prev_entry->end == start && (prev_entry->cred == cred ||
1671 (prev_entry->object.vm_object != NULL &&
1672 prev_entry->object.vm_object->cred == cred)) &&
1673 vm_object_coalesce(prev_entry->object.vm_object,
1675 (vm_size_t)(prev_entry->end - prev_entry->start),
1676 (vm_size_t)(end - prev_entry->end), cred != NULL &&
1677 (protoeflags & MAP_ENTRY_NEEDS_COPY) == 0)) {
1679 * We were able to extend the object. Determine if we
1680 * can extend the previous map entry to include the
1681 * new range as well.
1683 if (prev_entry->inheritance == inheritance &&
1684 prev_entry->protection == prot &&
1685 prev_entry->max_protection == max &&
1686 prev_entry->wired_count == 0) {
1687 KASSERT((prev_entry->eflags & MAP_ENTRY_USER_WIRED) ==
1688 0, ("prev_entry %p has incoherent wiring",
1690 if ((prev_entry->eflags & MAP_ENTRY_GUARD) == 0)
1691 map->size += end - prev_entry->end;
1692 vm_map_entry_resize(map, prev_entry,
1693 end - prev_entry->end);
1694 vm_map_try_merge_entries(map, prev_entry, next_entry);
1695 return (KERN_SUCCESS);
1699 * If we can extend the object but cannot extend the
1700 * map entry, we have to create a new map entry. We
1701 * must bump the ref count on the extended object to
1702 * account for it. object may be NULL.
1704 object = prev_entry->object.vm_object;
1705 offset = prev_entry->offset +
1706 (prev_entry->end - prev_entry->start);
1707 vm_object_reference(object);
1708 if (cred != NULL && object != NULL && object->cred != NULL &&
1709 !(prev_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
1710 /* Object already accounts for this uid. */
1718 * Create a new entry
1720 new_entry = vm_map_entry_create(map);
1721 new_entry->start = start;
1722 new_entry->end = end;
1723 new_entry->cred = NULL;
1725 new_entry->eflags = protoeflags;
1726 new_entry->object.vm_object = object;
1727 new_entry->offset = offset;
1729 new_entry->inheritance = inheritance;
1730 new_entry->protection = prot;
1731 new_entry->max_protection = max;
1732 new_entry->wired_count = 0;
1733 new_entry->wiring_thread = NULL;
1734 new_entry->read_ahead = VM_FAULT_READ_AHEAD_INIT;
1735 new_entry->next_read = start;
1737 KASSERT(cred == NULL || !ENTRY_CHARGED(new_entry),
1738 ("overcommit: vm_map_insert leaks vm_map %p", new_entry));
1739 new_entry->cred = cred;
1742 * Insert the new entry into the list
1744 vm_map_entry_link(map, new_entry);
1745 if ((new_entry->eflags & MAP_ENTRY_GUARD) == 0)
1746 map->size += new_entry->end - new_entry->start;
1749 * Try to coalesce the new entry with both the previous and next
1750 * entries in the list. Previously, we only attempted to coalesce
1751 * with the previous entry when object is NULL. Here, we handle the
1752 * other cases, which are less common.
1754 vm_map_try_merge_entries(map, prev_entry, new_entry);
1755 vm_map_try_merge_entries(map, new_entry, next_entry);
1757 if ((cow & (MAP_PREFAULT | MAP_PREFAULT_PARTIAL)) != 0) {
1758 vm_map_pmap_enter(map, start, prot, object, OFF_TO_IDX(offset),
1759 end - start, cow & MAP_PREFAULT_PARTIAL);
1762 return (KERN_SUCCESS);
1768 * Find the first fit (lowest VM address) for "length" free bytes
1769 * beginning at address >= start in the given map.
1771 * In a vm_map_entry, "max_free" is the maximum amount of
1772 * contiguous free space between an entry in its subtree and a
1773 * neighbor of that entry. This allows finding a free region in
1774 * one path down the tree, so O(log n) amortized with splay
1777 * The map must be locked, and leaves it so.
1779 * Returns: starting address if sufficient space,
1780 * vm_map_max(map)-length+1 if insufficient space.
1783 vm_map_findspace(vm_map_t map, vm_offset_t start, vm_size_t length)
1785 vm_map_entry_t header, llist, rlist, root, y;
1786 vm_size_t left_length, max_free_left, max_free_right;
1787 vm_offset_t gap_end;
1790 * Request must fit within min/max VM address and must avoid
1793 start = MAX(start, vm_map_min(map));
1794 if (start >= vm_map_max(map) || length > vm_map_max(map) - start)
1795 return (vm_map_max(map) - length + 1);
1797 /* Empty tree means wide open address space. */
1798 if (map->root == NULL)
1802 * After splay_split, if start is within an entry, push it to the start
1803 * of the following gap. If rlist is at the end of the gap containing
1804 * start, save the end of that gap in gap_end to see if the gap is big
1805 * enough; otherwise set gap_end to start skip gap-checking and move
1806 * directly to a search of the right subtree.
1808 header = &map->header;
1809 root = vm_map_splay_split(map, start, length, &llist, &rlist);
1810 gap_end = rlist->start;
1813 if (root->right != rlist)
1815 max_free_left = vm_map_splay_merge_left(header, root, llist);
1816 max_free_right = vm_map_splay_merge_right(header, root, rlist);
1817 } else if (rlist != header) {
1820 max_free_left = vm_map_splay_merge_pred(header, root, llist);
1821 max_free_right = vm_map_splay_merge_right(header, root, rlist);
1824 llist = root->right;
1825 max_free_left = vm_map_splay_merge_left(header, root, llist);
1826 max_free_right = vm_map_splay_merge_succ(header, root, rlist);
1828 root->max_free = vm_size_max(max_free_left, max_free_right);
1830 VM_MAP_ASSERT_CONSISTENT(map);
1831 if (length <= gap_end - start)
1834 /* With max_free, can immediately tell if no solution. */
1835 if (root->right == header || length > root->right->max_free)
1836 return (vm_map_max(map) - length + 1);
1839 * Splay for the least large-enough gap in the right subtree.
1841 llist = rlist = header;
1842 for (left_length = 0;;
1843 left_length = vm_map_entry_max_free_left(root, llist)) {
1844 if (length <= left_length)
1845 SPLAY_LEFT_STEP(root, y, llist, rlist,
1846 length <= vm_map_entry_max_free_left(y, llist));
1848 SPLAY_RIGHT_STEP(root, y, llist, rlist,
1849 length > vm_map_entry_max_free_left(y, root));
1854 llist = root->right;
1855 max_free_left = vm_map_splay_merge_left(header, root, llist);
1856 if (rlist == header) {
1857 root->max_free = vm_size_max(max_free_left,
1858 vm_map_splay_merge_succ(header, root, rlist));
1862 y->max_free = vm_size_max(
1863 vm_map_splay_merge_pred(root, y, root),
1864 vm_map_splay_merge_right(header, y, rlist));
1865 root->max_free = vm_size_max(max_free_left, y->max_free);
1868 VM_MAP_ASSERT_CONSISTENT(map);
1873 vm_map_fixed(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1874 vm_offset_t start, vm_size_t length, vm_prot_t prot,
1875 vm_prot_t max, int cow)
1880 end = start + length;
1881 KASSERT((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 ||
1883 ("vm_map_fixed: non-NULL backing object for stack"));
1885 VM_MAP_RANGE_CHECK(map, start, end);
1886 if ((cow & MAP_CHECK_EXCL) == 0) {
1887 result = vm_map_delete(map, start, end);
1888 if (result != KERN_SUCCESS)
1891 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) != 0) {
1892 result = vm_map_stack_locked(map, start, length, sgrowsiz,
1895 result = vm_map_insert(map, object, offset, start, end,
1903 static const int aslr_pages_rnd_64[2] = {0x1000, 0x10};
1904 static const int aslr_pages_rnd_32[2] = {0x100, 0x4};
1906 static int cluster_anon = 1;
1907 SYSCTL_INT(_vm, OID_AUTO, cluster_anon, CTLFLAG_RW,
1909 "Cluster anonymous mappings: 0 = no, 1 = yes if no hint, 2 = always");
1912 clustering_anon_allowed(vm_offset_t addr)
1915 switch (cluster_anon) {
1926 static long aslr_restarts;
1927 SYSCTL_LONG(_vm, OID_AUTO, aslr_restarts, CTLFLAG_RD,
1929 "Number of aslr failures");
1932 * Searches for the specified amount of free space in the given map with the
1933 * specified alignment. Performs an address-ordered, first-fit search from
1934 * the given address "*addr", with an optional upper bound "max_addr". If the
1935 * parameter "alignment" is zero, then the alignment is computed from the
1936 * given (object, offset) pair so as to enable the greatest possible use of
1937 * superpage mappings. Returns KERN_SUCCESS and the address of the free space
1938 * in "*addr" if successful. Otherwise, returns KERN_NO_SPACE.
1940 * The map must be locked. Initially, there must be at least "length" bytes
1941 * of free space at the given address.
1944 vm_map_alignspace(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1945 vm_offset_t *addr, vm_size_t length, vm_offset_t max_addr,
1946 vm_offset_t alignment)
1948 vm_offset_t aligned_addr, free_addr;
1950 VM_MAP_ASSERT_LOCKED(map);
1952 KASSERT(free_addr == vm_map_findspace(map, free_addr, length),
1953 ("caller failed to provide space %#jx at address %p",
1954 (uintmax_t)length, (void *)free_addr));
1957 * At the start of every iteration, the free space at address
1958 * "*addr" is at least "length" bytes.
1961 pmap_align_superpage(object, offset, addr, length);
1962 else if ((*addr & (alignment - 1)) != 0) {
1963 *addr &= ~(alignment - 1);
1966 aligned_addr = *addr;
1967 if (aligned_addr == free_addr) {
1969 * Alignment did not change "*addr", so "*addr" must
1970 * still provide sufficient free space.
1972 return (KERN_SUCCESS);
1976 * Test for address wrap on "*addr". A wrapped "*addr" could
1977 * be a valid address, in which case vm_map_findspace() cannot
1978 * be relied upon to fail.
1980 if (aligned_addr < free_addr)
1981 return (KERN_NO_SPACE);
1982 *addr = vm_map_findspace(map, aligned_addr, length);
1983 if (*addr + length > vm_map_max(map) ||
1984 (max_addr != 0 && *addr + length > max_addr))
1985 return (KERN_NO_SPACE);
1987 if (free_addr == aligned_addr) {
1989 * If a successful call to vm_map_findspace() did not
1990 * change "*addr", then "*addr" must still be aligned
1991 * and provide sufficient free space.
1993 return (KERN_SUCCESS);
1999 vm_map_find_aligned(vm_map_t map, vm_offset_t *addr, vm_size_t length,
2000 vm_offset_t max_addr, vm_offset_t alignment)
2002 /* XXXKIB ASLR eh ? */
2003 *addr = vm_map_findspace(map, *addr, length);
2004 if (*addr + length > vm_map_max(map) ||
2005 (max_addr != 0 && *addr + length > max_addr))
2006 return (KERN_NO_SPACE);
2007 return (vm_map_alignspace(map, NULL, 0, addr, length, max_addr,
2012 * vm_map_find finds an unallocated region in the target address
2013 * map with the given length. The search is defined to be
2014 * first-fit from the specified address; the region found is
2015 * returned in the same parameter.
2017 * If object is non-NULL, ref count must be bumped by caller
2018 * prior to making call to account for the new entry.
2021 vm_map_find(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
2022 vm_offset_t *addr, /* IN/OUT */
2023 vm_size_t length, vm_offset_t max_addr, int find_space,
2024 vm_prot_t prot, vm_prot_t max, int cow)
2026 vm_offset_t alignment, curr_min_addr, min_addr;
2027 int gap, pidx, rv, try;
2028 bool cluster, en_aslr, update_anon;
2030 KASSERT((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 ||
2032 ("vm_map_find: non-NULL backing object for stack"));
2033 MPASS((cow & MAP_REMAP) == 0 || (find_space == VMFS_NO_SPACE &&
2034 (cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0));
2035 if (find_space == VMFS_OPTIMAL_SPACE && (object == NULL ||
2036 (object->flags & OBJ_COLORED) == 0))
2037 find_space = VMFS_ANY_SPACE;
2038 if (find_space >> 8 != 0) {
2039 KASSERT((find_space & 0xff) == 0, ("bad VMFS flags"));
2040 alignment = (vm_offset_t)1 << (find_space >> 8);
2043 en_aslr = (map->flags & MAP_ASLR) != 0;
2044 update_anon = cluster = clustering_anon_allowed(*addr) &&
2045 (map->flags & MAP_IS_SUB_MAP) == 0 && max_addr == 0 &&
2046 find_space != VMFS_NO_SPACE && object == NULL &&
2047 (cow & (MAP_INHERIT_SHARE | MAP_STACK_GROWS_UP |
2048 MAP_STACK_GROWS_DOWN)) == 0 && prot != PROT_NONE;
2049 curr_min_addr = min_addr = *addr;
2050 if (en_aslr && min_addr == 0 && !cluster &&
2051 find_space != VMFS_NO_SPACE &&
2052 (map->flags & MAP_ASLR_IGNSTART) != 0)
2053 curr_min_addr = min_addr = vm_map_min(map);
2057 curr_min_addr = map->anon_loc;
2058 if (curr_min_addr == 0)
2061 if (find_space != VMFS_NO_SPACE) {
2062 KASSERT(find_space == VMFS_ANY_SPACE ||
2063 find_space == VMFS_OPTIMAL_SPACE ||
2064 find_space == VMFS_SUPER_SPACE ||
2065 alignment != 0, ("unexpected VMFS flag"));
2068 * When creating an anonymous mapping, try clustering
2069 * with an existing anonymous mapping first.
2071 * We make up to two attempts to find address space
2072 * for a given find_space value. The first attempt may
2073 * apply randomization or may cluster with an existing
2074 * anonymous mapping. If this first attempt fails,
2075 * perform a first-fit search of the available address
2078 * If all tries failed, and find_space is
2079 * VMFS_OPTIMAL_SPACE, fallback to VMFS_ANY_SPACE.
2080 * Again enable clustering and randomization.
2087 * Second try: we failed either to find a
2088 * suitable region for randomizing the
2089 * allocation, or to cluster with an existing
2090 * mapping. Retry with free run.
2092 curr_min_addr = (map->flags & MAP_ASLR_IGNSTART) != 0 ?
2093 vm_map_min(map) : min_addr;
2094 atomic_add_long(&aslr_restarts, 1);
2097 if (try == 1 && en_aslr && !cluster) {
2099 * Find space for allocation, including
2100 * gap needed for later randomization.
2102 pidx = MAXPAGESIZES > 1 && pagesizes[1] != 0 &&
2103 (find_space == VMFS_SUPER_SPACE || find_space ==
2104 VMFS_OPTIMAL_SPACE) ? 1 : 0;
2105 gap = vm_map_max(map) > MAP_32BIT_MAX_ADDR &&
2106 (max_addr == 0 || max_addr > MAP_32BIT_MAX_ADDR) ?
2107 aslr_pages_rnd_64[pidx] : aslr_pages_rnd_32[pidx];
2108 *addr = vm_map_findspace(map, curr_min_addr,
2109 length + gap * pagesizes[pidx]);
2110 if (*addr + length + gap * pagesizes[pidx] >
2113 /* And randomize the start address. */
2114 *addr += (arc4random() % gap) * pagesizes[pidx];
2115 if (max_addr != 0 && *addr + length > max_addr)
2118 *addr = vm_map_findspace(map, curr_min_addr, length);
2119 if (*addr + length > vm_map_max(map) ||
2120 (max_addr != 0 && *addr + length > max_addr)) {
2131 if (find_space != VMFS_ANY_SPACE &&
2132 (rv = vm_map_alignspace(map, object, offset, addr, length,
2133 max_addr, alignment)) != KERN_SUCCESS) {
2134 if (find_space == VMFS_OPTIMAL_SPACE) {
2135 find_space = VMFS_ANY_SPACE;
2136 curr_min_addr = min_addr;
2137 cluster = update_anon;
2143 } else if ((cow & MAP_REMAP) != 0) {
2144 if (!vm_map_range_valid(map, *addr, *addr + length)) {
2145 rv = KERN_INVALID_ADDRESS;
2148 rv = vm_map_delete(map, *addr, *addr + length);
2149 if (rv != KERN_SUCCESS)
2152 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) != 0) {
2153 rv = vm_map_stack_locked(map, *addr, length, sgrowsiz, prot,
2156 rv = vm_map_insert(map, object, offset, *addr, *addr + length,
2159 if (rv == KERN_SUCCESS && update_anon)
2160 map->anon_loc = *addr + length;
2167 * vm_map_find_min() is a variant of vm_map_find() that takes an
2168 * additional parameter (min_addr) and treats the given address
2169 * (*addr) differently. Specifically, it treats *addr as a hint
2170 * and not as the minimum address where the mapping is created.
2172 * This function works in two phases. First, it tries to
2173 * allocate above the hint. If that fails and the hint is
2174 * greater than min_addr, it performs a second pass, replacing
2175 * the hint with min_addr as the minimum address for the
2179 vm_map_find_min(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
2180 vm_offset_t *addr, vm_size_t length, vm_offset_t min_addr,
2181 vm_offset_t max_addr, int find_space, vm_prot_t prot, vm_prot_t max,
2189 rv = vm_map_find(map, object, offset, addr, length, max_addr,
2190 find_space, prot, max, cow);
2191 if (rv == KERN_SUCCESS || min_addr >= hint)
2193 *addr = hint = min_addr;
2198 * A map entry with any of the following flags set must not be merged with
2201 #define MAP_ENTRY_NOMERGE_MASK (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP | \
2202 MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_IS_SUB_MAP | MAP_ENTRY_VN_EXEC)
2205 vm_map_mergeable_neighbors(vm_map_entry_t prev, vm_map_entry_t entry)
2208 KASSERT((prev->eflags & MAP_ENTRY_NOMERGE_MASK) == 0 ||
2209 (entry->eflags & MAP_ENTRY_NOMERGE_MASK) == 0,
2210 ("vm_map_mergeable_neighbors: neither %p nor %p are mergeable",
2212 return (prev->end == entry->start &&
2213 prev->object.vm_object == entry->object.vm_object &&
2214 (prev->object.vm_object == NULL ||
2215 prev->offset + (prev->end - prev->start) == entry->offset) &&
2216 prev->eflags == entry->eflags &&
2217 prev->protection == entry->protection &&
2218 prev->max_protection == entry->max_protection &&
2219 prev->inheritance == entry->inheritance &&
2220 prev->wired_count == entry->wired_count &&
2221 prev->cred == entry->cred);
2225 vm_map_merged_neighbor_dispose(vm_map_t map, vm_map_entry_t entry)
2229 * If the backing object is a vnode object, vm_object_deallocate()
2230 * calls vrele(). However, vrele() does not lock the vnode because
2231 * the vnode has additional references. Thus, the map lock can be
2232 * kept without causing a lock-order reversal with the vnode lock.
2234 * Since we count the number of virtual page mappings in
2235 * object->un_pager.vnp.writemappings, the writemappings value
2236 * should not be adjusted when the entry is disposed of.
2238 if (entry->object.vm_object != NULL)
2239 vm_object_deallocate(entry->object.vm_object);
2240 if (entry->cred != NULL)
2241 crfree(entry->cred);
2242 vm_map_entry_dispose(map, entry);
2246 * vm_map_try_merge_entries:
2248 * Compare the given map entry to its predecessor, and merge its precessor
2249 * into it if possible. The entry remains valid, and may be extended.
2250 * The predecessor may be deleted.
2252 * The map must be locked.
2255 vm_map_try_merge_entries(vm_map_t map, vm_map_entry_t prev_entry,
2256 vm_map_entry_t entry)
2259 VM_MAP_ASSERT_LOCKED(map);
2260 if ((entry->eflags & MAP_ENTRY_NOMERGE_MASK) == 0 &&
2261 vm_map_mergeable_neighbors(prev_entry, entry)) {
2262 vm_map_entry_unlink(map, prev_entry, UNLINK_MERGE_NEXT);
2263 vm_map_merged_neighbor_dispose(map, prev_entry);
2268 * vm_map_entry_back:
2270 * Allocate an object to back a map entry.
2273 vm_map_entry_back(vm_map_entry_t entry)
2277 KASSERT(entry->object.vm_object == NULL,
2278 ("map entry %p has backing object", entry));
2279 KASSERT((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0,
2280 ("map entry %p is a submap", entry));
2281 object = vm_object_allocate_anon(atop(entry->end - entry->start), NULL,
2282 entry->cred, entry->end - entry->start);
2283 entry->object.vm_object = object;
2289 * vm_map_entry_charge_object
2291 * If there is no object backing this entry, create one. Otherwise, if
2292 * the entry has cred, give it to the backing object.
2295 vm_map_entry_charge_object(vm_map_t map, vm_map_entry_t entry)
2298 VM_MAP_ASSERT_LOCKED(map);
2299 KASSERT((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0,
2300 ("map entry %p is a submap", entry));
2301 if (entry->object.vm_object == NULL && !map->system_map &&
2302 (entry->eflags & MAP_ENTRY_GUARD) == 0)
2303 vm_map_entry_back(entry);
2304 else if (entry->object.vm_object != NULL &&
2305 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) &&
2306 entry->cred != NULL) {
2307 VM_OBJECT_WLOCK(entry->object.vm_object);
2308 KASSERT(entry->object.vm_object->cred == NULL,
2309 ("OVERCOMMIT: %s: both cred e %p", __func__, entry));
2310 entry->object.vm_object->cred = entry->cred;
2311 entry->object.vm_object->charge = entry->end - entry->start;
2312 VM_OBJECT_WUNLOCK(entry->object.vm_object);
2318 * vm_map_entry_clone
2320 * Create a duplicate map entry for clipping.
2322 static vm_map_entry_t
2323 vm_map_entry_clone(vm_map_t map, vm_map_entry_t entry)
2325 vm_map_entry_t new_entry;
2327 VM_MAP_ASSERT_LOCKED(map);
2330 * Create a backing object now, if none exists, so that more individual
2331 * objects won't be created after the map entry is split.
2333 vm_map_entry_charge_object(map, entry);
2335 /* Clone the entry. */
2336 new_entry = vm_map_entry_create(map);
2337 *new_entry = *entry;
2338 if (new_entry->cred != NULL)
2339 crhold(entry->cred);
2340 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
2341 vm_object_reference(new_entry->object.vm_object);
2342 vm_map_entry_set_vnode_text(new_entry, true);
2344 * The object->un_pager.vnp.writemappings for the object of
2345 * MAP_ENTRY_WRITECNT type entry shall be kept as is here. The
2346 * virtual pages are re-distributed among the clipped entries,
2347 * so the sum is left the same.
2354 * vm_map_clip_start: [ internal use only ]
2356 * Asserts that the given entry begins at or after
2357 * the specified address; if necessary,
2358 * it splits the entry into two.
2361 vm_map_clip_start(vm_map_t map, vm_map_entry_t entry, vm_offset_t startaddr)
2363 vm_map_entry_t new_entry;
2366 if (!map->system_map)
2367 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
2368 "%s: map %p entry %p start 0x%jx", __func__, map, entry,
2369 (uintmax_t)startaddr);
2371 if (startaddr <= entry->start)
2372 return (KERN_SUCCESS);
2374 VM_MAP_ASSERT_LOCKED(map);
2375 KASSERT(entry->end > startaddr && entry->start < startaddr,
2376 ("%s: invalid clip of entry %p", __func__, entry));
2378 bdry_idx = (entry->eflags & MAP_ENTRY_SPLIT_BOUNDARY_MASK) >>
2379 MAP_ENTRY_SPLIT_BOUNDARY_SHIFT;
2380 if (bdry_idx != 0) {
2381 if ((startaddr & (pagesizes[bdry_idx] - 1)) != 0)
2382 return (KERN_INVALID_ARGUMENT);
2385 new_entry = vm_map_entry_clone(map, entry);
2388 * Split off the front portion. Insert the new entry BEFORE this one,
2389 * so that this entry has the specified starting address.
2391 new_entry->end = startaddr;
2392 vm_map_entry_link(map, new_entry);
2393 return (KERN_SUCCESS);
2397 * vm_map_lookup_clip_start:
2399 * Find the entry at or just after 'start', and clip it if 'start' is in
2400 * the interior of the entry. Return entry after 'start', and in
2401 * prev_entry set the entry before 'start'.
2404 vm_map_lookup_clip_start(vm_map_t map, vm_offset_t start,
2405 vm_map_entry_t *res_entry, vm_map_entry_t *prev_entry)
2407 vm_map_entry_t entry;
2410 if (!map->system_map)
2411 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
2412 "%s: map %p start 0x%jx prev %p", __func__, map,
2413 (uintmax_t)start, prev_entry);
2415 if (vm_map_lookup_entry(map, start, prev_entry)) {
2416 entry = *prev_entry;
2417 rv = vm_map_clip_start(map, entry, start);
2418 if (rv != KERN_SUCCESS)
2420 *prev_entry = vm_map_entry_pred(entry);
2422 entry = vm_map_entry_succ(*prev_entry);
2424 return (KERN_SUCCESS);
2428 * vm_map_clip_end: [ internal use only ]
2430 * Asserts that the given entry ends at or before
2431 * the specified address; if necessary,
2432 * it splits the entry into two.
2435 vm_map_clip_end(vm_map_t map, vm_map_entry_t entry, vm_offset_t endaddr)
2437 vm_map_entry_t new_entry;
2440 if (!map->system_map)
2441 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
2442 "%s: map %p entry %p end 0x%jx", __func__, map, entry,
2443 (uintmax_t)endaddr);
2445 if (endaddr >= entry->end)
2446 return (KERN_SUCCESS);
2448 VM_MAP_ASSERT_LOCKED(map);
2449 KASSERT(entry->start < endaddr && entry->end > endaddr,
2450 ("%s: invalid clip of entry %p", __func__, entry));
2452 bdry_idx = (entry->eflags & MAP_ENTRY_SPLIT_BOUNDARY_MASK) >>
2453 MAP_ENTRY_SPLIT_BOUNDARY_SHIFT;
2454 if (bdry_idx != 0) {
2455 if ((endaddr & (pagesizes[bdry_idx] - 1)) != 0)
2456 return (KERN_INVALID_ARGUMENT);
2459 new_entry = vm_map_entry_clone(map, entry);
2462 * Split off the back portion. Insert the new entry AFTER this one,
2463 * so that this entry has the specified ending address.
2465 new_entry->start = endaddr;
2466 vm_map_entry_link(map, new_entry);
2468 return (KERN_SUCCESS);
2472 * vm_map_submap: [ kernel use only ]
2474 * Mark the given range as handled by a subordinate map.
2476 * This range must have been created with vm_map_find,
2477 * and no other operations may have been performed on this
2478 * range prior to calling vm_map_submap.
2480 * Only a limited number of operations can be performed
2481 * within this rage after calling vm_map_submap:
2483 * [Don't try vm_map_copy!]
2485 * To remove a submapping, one must first remove the
2486 * range from the superior map, and then destroy the
2487 * submap (if desired). [Better yet, don't try it.]
2496 vm_map_entry_t entry;
2499 result = KERN_INVALID_ARGUMENT;
2501 vm_map_lock(submap);
2502 submap->flags |= MAP_IS_SUB_MAP;
2503 vm_map_unlock(submap);
2506 VM_MAP_RANGE_CHECK(map, start, end);
2507 if (vm_map_lookup_entry(map, start, &entry) && entry->end >= end &&
2508 (entry->eflags & MAP_ENTRY_COW) == 0 &&
2509 entry->object.vm_object == NULL) {
2510 result = vm_map_clip_start(map, entry, start);
2511 if (result != KERN_SUCCESS)
2513 result = vm_map_clip_end(map, entry, end);
2514 if (result != KERN_SUCCESS)
2516 entry->object.sub_map = submap;
2517 entry->eflags |= MAP_ENTRY_IS_SUB_MAP;
2518 result = KERN_SUCCESS;
2523 if (result != KERN_SUCCESS) {
2524 vm_map_lock(submap);
2525 submap->flags &= ~MAP_IS_SUB_MAP;
2526 vm_map_unlock(submap);
2532 * The maximum number of pages to map if MAP_PREFAULT_PARTIAL is specified
2534 #define MAX_INIT_PT 96
2537 * vm_map_pmap_enter:
2539 * Preload the specified map's pmap with mappings to the specified
2540 * object's memory-resident pages. No further physical pages are
2541 * allocated, and no further virtual pages are retrieved from secondary
2542 * storage. If the specified flags include MAP_PREFAULT_PARTIAL, then a
2543 * limited number of page mappings are created at the low-end of the
2544 * specified address range. (For this purpose, a superpage mapping
2545 * counts as one page mapping.) Otherwise, all resident pages within
2546 * the specified address range are mapped.
2549 vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot,
2550 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags)
2553 vm_page_t p, p_start;
2554 vm_pindex_t mask, psize, threshold, tmpidx;
2556 if ((prot & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0 || object == NULL)
2558 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) {
2559 VM_OBJECT_WLOCK(object);
2560 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) {
2561 pmap_object_init_pt(map->pmap, addr, object, pindex,
2563 VM_OBJECT_WUNLOCK(object);
2566 VM_OBJECT_LOCK_DOWNGRADE(object);
2568 VM_OBJECT_RLOCK(object);
2571 if (psize + pindex > object->size) {
2572 if (pindex >= object->size) {
2573 VM_OBJECT_RUNLOCK(object);
2576 psize = object->size - pindex;
2581 threshold = MAX_INIT_PT;
2583 p = vm_page_find_least(object, pindex);
2585 * Assert: the variable p is either (1) the page with the
2586 * least pindex greater than or equal to the parameter pindex
2590 p != NULL && (tmpidx = p->pindex - pindex) < psize;
2591 p = TAILQ_NEXT(p, listq)) {
2593 * don't allow an madvise to blow away our really
2594 * free pages allocating pv entries.
2596 if (((flags & MAP_PREFAULT_MADVISE) != 0 &&
2597 vm_page_count_severe()) ||
2598 ((flags & MAP_PREFAULT_PARTIAL) != 0 &&
2599 tmpidx >= threshold)) {
2603 if (vm_page_all_valid(p)) {
2604 if (p_start == NULL) {
2605 start = addr + ptoa(tmpidx);
2608 /* Jump ahead if a superpage mapping is possible. */
2609 if (p->psind > 0 && ((addr + ptoa(tmpidx)) &
2610 (pagesizes[p->psind] - 1)) == 0) {
2611 mask = atop(pagesizes[p->psind]) - 1;
2612 if (tmpidx + mask < psize &&
2613 vm_page_ps_test(p, PS_ALL_VALID, NULL)) {
2618 } else if (p_start != NULL) {
2619 pmap_enter_object(map->pmap, start, addr +
2620 ptoa(tmpidx), p_start, prot);
2624 if (p_start != NULL)
2625 pmap_enter_object(map->pmap, start, addr + ptoa(psize),
2627 VM_OBJECT_RUNLOCK(object);
2633 * Sets the protection of the specified address
2634 * region in the target map. If "set_max" is
2635 * specified, the maximum protection is to be set;
2636 * otherwise, only the current protection is affected.
2639 vm_map_protect(vm_map_t map, vm_offset_t start, vm_offset_t end,
2640 vm_prot_t new_prot, boolean_t set_max)
2642 vm_map_entry_t entry, first_entry, in_tran, prev_entry;
2649 return (KERN_SUCCESS);
2656 * Ensure that we are not concurrently wiring pages. vm_map_wire() may
2657 * need to fault pages into the map and will drop the map lock while
2658 * doing so, and the VM object may end up in an inconsistent state if we
2659 * update the protection on the map entry in between faults.
2661 vm_map_wait_busy(map);
2663 VM_MAP_RANGE_CHECK(map, start, end);
2665 if (!vm_map_lookup_entry(map, start, &first_entry))
2666 first_entry = vm_map_entry_succ(first_entry);
2669 * Make a first pass to check for protection violations.
2671 for (entry = first_entry; entry->start < end;
2672 entry = vm_map_entry_succ(entry)) {
2673 if ((entry->eflags & MAP_ENTRY_GUARD) != 0)
2675 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0) {
2677 return (KERN_INVALID_ARGUMENT);
2679 if ((new_prot & entry->max_protection) != new_prot) {
2681 return (KERN_PROTECTION_FAILURE);
2683 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0)
2688 * Postpone the operation until all in-transition map entries have
2689 * stabilized. An in-transition entry might already have its pages
2690 * wired and wired_count incremented, but not yet have its
2691 * MAP_ENTRY_USER_WIRED flag set. In which case, we would fail to call
2692 * vm_fault_copy_entry() in the final loop below.
2694 if (in_tran != NULL) {
2695 in_tran->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2696 vm_map_unlock_and_wait(map, 0);
2701 * Before changing the protections, try to reserve swap space for any
2702 * private (i.e., copy-on-write) mappings that are transitioning from
2703 * read-only to read/write access. If a reservation fails, break out
2704 * of this loop early and let the next loop simplify the entries, since
2705 * some may now be mergeable.
2707 rv = vm_map_clip_start(map, first_entry, start);
2708 if (rv != KERN_SUCCESS) {
2712 for (entry = first_entry; entry->start < end;
2713 entry = vm_map_entry_succ(entry)) {
2714 rv = vm_map_clip_end(map, entry, end);
2715 if (rv != KERN_SUCCESS) {
2721 ((new_prot & ~entry->protection) & VM_PROT_WRITE) == 0 ||
2722 ENTRY_CHARGED(entry) ||
2723 (entry->eflags & MAP_ENTRY_GUARD) != 0) {
2727 cred = curthread->td_ucred;
2728 obj = entry->object.vm_object;
2731 (entry->eflags & MAP_ENTRY_NEEDS_COPY) != 0) {
2732 if (!swap_reserve(entry->end - entry->start)) {
2733 rv = KERN_RESOURCE_SHORTAGE;
2742 if (obj->type != OBJT_DEFAULT && obj->type != OBJT_SWAP)
2744 VM_OBJECT_WLOCK(obj);
2745 if (obj->type != OBJT_DEFAULT && obj->type != OBJT_SWAP) {
2746 VM_OBJECT_WUNLOCK(obj);
2751 * Charge for the whole object allocation now, since
2752 * we cannot distinguish between non-charged and
2753 * charged clipped mapping of the same object later.
2755 KASSERT(obj->charge == 0,
2756 ("vm_map_protect: object %p overcharged (entry %p)",
2758 if (!swap_reserve(ptoa(obj->size))) {
2759 VM_OBJECT_WUNLOCK(obj);
2760 rv = KERN_RESOURCE_SHORTAGE;
2767 obj->charge = ptoa(obj->size);
2768 VM_OBJECT_WUNLOCK(obj);
2772 * If enough swap space was available, go back and fix up protections.
2773 * Otherwise, just simplify entries, since some may have been modified.
2774 * [Note that clipping is not necessary the second time.]
2776 for (prev_entry = vm_map_entry_pred(first_entry), entry = first_entry;
2778 vm_map_try_merge_entries(map, prev_entry, entry),
2779 prev_entry = entry, entry = vm_map_entry_succ(entry)) {
2780 if (rv != KERN_SUCCESS ||
2781 (entry->eflags & MAP_ENTRY_GUARD) != 0)
2784 old_prot = entry->protection;
2788 (entry->max_protection = new_prot) &
2791 entry->protection = new_prot;
2794 * For user wired map entries, the normal lazy evaluation of
2795 * write access upgrades through soft page faults is
2796 * undesirable. Instead, immediately copy any pages that are
2797 * copy-on-write and enable write access in the physical map.
2799 if ((entry->eflags & MAP_ENTRY_USER_WIRED) != 0 &&
2800 (entry->protection & VM_PROT_WRITE) != 0 &&
2801 (old_prot & VM_PROT_WRITE) == 0)
2802 vm_fault_copy_entry(map, map, entry, entry, NULL);
2805 * When restricting access, update the physical map. Worry
2806 * about copy-on-write here.
2808 if ((old_prot & ~entry->protection) != 0) {
2809 #define MASK(entry) (((entry)->eflags & MAP_ENTRY_COW) ? ~VM_PROT_WRITE : \
2811 pmap_protect(map->pmap, entry->start,
2813 entry->protection & MASK(entry));
2817 vm_map_try_merge_entries(map, prev_entry, entry);
2825 * This routine traverses a processes map handling the madvise
2826 * system call. Advisories are classified as either those effecting
2827 * the vm_map_entry structure, or those effecting the underlying
2837 vm_map_entry_t entry, prev_entry;
2842 * Some madvise calls directly modify the vm_map_entry, in which case
2843 * we need to use an exclusive lock on the map and we need to perform
2844 * various clipping operations. Otherwise we only need a read-lock
2849 case MADV_SEQUENTIAL:
2866 vm_map_lock_read(map);
2873 * Locate starting entry and clip if necessary.
2875 VM_MAP_RANGE_CHECK(map, start, end);
2879 * madvise behaviors that are implemented in the vm_map_entry.
2881 * We clip the vm_map_entry so that behavioral changes are
2882 * limited to the specified address range.
2884 rv = vm_map_lookup_clip_start(map, start, &entry, &prev_entry);
2885 if (rv != KERN_SUCCESS) {
2887 return (vm_mmap_to_errno(rv));
2890 for (; entry->start < end; prev_entry = entry,
2891 entry = vm_map_entry_succ(entry)) {
2892 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0)
2895 rv = vm_map_clip_end(map, entry, end);
2896 if (rv != KERN_SUCCESS) {
2898 return (vm_mmap_to_errno(rv));
2903 vm_map_entry_set_behavior(entry,
2904 MAP_ENTRY_BEHAV_NORMAL);
2906 case MADV_SEQUENTIAL:
2907 vm_map_entry_set_behavior(entry,
2908 MAP_ENTRY_BEHAV_SEQUENTIAL);
2911 vm_map_entry_set_behavior(entry,
2912 MAP_ENTRY_BEHAV_RANDOM);
2915 entry->eflags |= MAP_ENTRY_NOSYNC;
2918 entry->eflags &= ~MAP_ENTRY_NOSYNC;
2921 entry->eflags |= MAP_ENTRY_NOCOREDUMP;
2924 entry->eflags &= ~MAP_ENTRY_NOCOREDUMP;
2929 vm_map_try_merge_entries(map, prev_entry, entry);
2931 vm_map_try_merge_entries(map, prev_entry, entry);
2934 vm_pindex_t pstart, pend;
2937 * madvise behaviors that are implemented in the underlying
2940 * Since we don't clip the vm_map_entry, we have to clip
2941 * the vm_object pindex and count.
2943 if (!vm_map_lookup_entry(map, start, &entry))
2944 entry = vm_map_entry_succ(entry);
2945 for (; entry->start < end;
2946 entry = vm_map_entry_succ(entry)) {
2947 vm_offset_t useEnd, useStart;
2949 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0)
2953 * MADV_FREE would otherwise rewind time to
2954 * the creation of the shadow object. Because
2955 * we hold the VM map read-locked, neither the
2956 * entry's object nor the presence of a
2957 * backing object can change.
2959 if (behav == MADV_FREE &&
2960 entry->object.vm_object != NULL &&
2961 entry->object.vm_object->backing_object != NULL)
2964 pstart = OFF_TO_IDX(entry->offset);
2965 pend = pstart + atop(entry->end - entry->start);
2966 useStart = entry->start;
2967 useEnd = entry->end;
2969 if (entry->start < start) {
2970 pstart += atop(start - entry->start);
2973 if (entry->end > end) {
2974 pend -= atop(entry->end - end);
2982 * Perform the pmap_advise() before clearing
2983 * PGA_REFERENCED in vm_page_advise(). Otherwise, a
2984 * concurrent pmap operation, such as pmap_remove(),
2985 * could clear a reference in the pmap and set
2986 * PGA_REFERENCED on the page before the pmap_advise()
2987 * had completed. Consequently, the page would appear
2988 * referenced based upon an old reference that
2989 * occurred before this pmap_advise() ran.
2991 if (behav == MADV_DONTNEED || behav == MADV_FREE)
2992 pmap_advise(map->pmap, useStart, useEnd,
2995 vm_object_madvise(entry->object.vm_object, pstart,
2999 * Pre-populate paging structures in the
3000 * WILLNEED case. For wired entries, the
3001 * paging structures are already populated.
3003 if (behav == MADV_WILLNEED &&
3004 entry->wired_count == 0) {
3005 vm_map_pmap_enter(map,
3008 entry->object.vm_object,
3010 ptoa(pend - pstart),
3011 MAP_PREFAULT_MADVISE
3015 vm_map_unlock_read(map);
3023 * Sets the inheritance of the specified address
3024 * range in the target map. Inheritance
3025 * affects how the map will be shared with
3026 * child maps at the time of vmspace_fork.
3029 vm_map_inherit(vm_map_t map, vm_offset_t start, vm_offset_t end,
3030 vm_inherit_t new_inheritance)
3032 vm_map_entry_t entry, lentry, prev_entry, start_entry;
3035 switch (new_inheritance) {
3036 case VM_INHERIT_NONE:
3037 case VM_INHERIT_COPY:
3038 case VM_INHERIT_SHARE:
3039 case VM_INHERIT_ZERO:
3042 return (KERN_INVALID_ARGUMENT);
3045 return (KERN_SUCCESS);
3047 VM_MAP_RANGE_CHECK(map, start, end);
3048 rv = vm_map_lookup_clip_start(map, start, &start_entry, &prev_entry);
3049 if (rv != KERN_SUCCESS)
3051 if (vm_map_lookup_entry(map, end - 1, &lentry)) {
3052 rv = vm_map_clip_end(map, lentry, end);
3053 if (rv != KERN_SUCCESS)
3056 if (new_inheritance == VM_INHERIT_COPY) {
3057 for (entry = start_entry; entry->start < end;
3058 prev_entry = entry, entry = vm_map_entry_succ(entry)) {
3059 if ((entry->eflags & MAP_ENTRY_SPLIT_BOUNDARY_MASK)
3061 rv = KERN_INVALID_ARGUMENT;
3066 for (entry = start_entry; entry->start < end; prev_entry = entry,
3067 entry = vm_map_entry_succ(entry)) {
3068 KASSERT(entry->end <= end, ("non-clipped entry %p end %jx %jx",
3069 entry, (uintmax_t)entry->end, (uintmax_t)end));
3070 if ((entry->eflags & MAP_ENTRY_GUARD) == 0 ||
3071 new_inheritance != VM_INHERIT_ZERO)
3072 entry->inheritance = new_inheritance;
3073 vm_map_try_merge_entries(map, prev_entry, entry);
3075 vm_map_try_merge_entries(map, prev_entry, entry);
3082 * vm_map_entry_in_transition:
3084 * Release the map lock, and sleep until the entry is no longer in
3085 * transition. Awake and acquire the map lock. If the map changed while
3086 * another held the lock, lookup a possibly-changed entry at or after the
3087 * 'start' position of the old entry.
3089 static vm_map_entry_t
3090 vm_map_entry_in_transition(vm_map_t map, vm_offset_t in_start,
3091 vm_offset_t *io_end, bool holes_ok, vm_map_entry_t in_entry)
3093 vm_map_entry_t entry;
3095 u_int last_timestamp;
3097 VM_MAP_ASSERT_LOCKED(map);
3098 KASSERT((in_entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0,
3099 ("not in-tranition map entry %p", in_entry));
3101 * We have not yet clipped the entry.
3103 start = MAX(in_start, in_entry->start);
3104 in_entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
3105 last_timestamp = map->timestamp;
3106 if (vm_map_unlock_and_wait(map, 0)) {
3108 * Allow interruption of user wiring/unwiring?
3112 if (last_timestamp + 1 == map->timestamp)
3116 * Look again for the entry because the map was modified while it was
3117 * unlocked. Specifically, the entry may have been clipped, merged, or
3120 if (!vm_map_lookup_entry(map, start, &entry)) {
3125 entry = vm_map_entry_succ(entry);
3133 * Implements both kernel and user unwiring.
3136 vm_map_unwire(vm_map_t map, vm_offset_t start, vm_offset_t end,
3139 vm_map_entry_t entry, first_entry, next_entry, prev_entry;
3141 bool holes_ok, need_wakeup, user_unwire;
3144 return (KERN_SUCCESS);
3145 holes_ok = (flags & VM_MAP_WIRE_HOLESOK) != 0;
3146 user_unwire = (flags & VM_MAP_WIRE_USER) != 0;
3148 VM_MAP_RANGE_CHECK(map, start, end);
3149 if (!vm_map_lookup_entry(map, start, &first_entry)) {
3151 first_entry = vm_map_entry_succ(first_entry);
3154 return (KERN_INVALID_ADDRESS);
3158 for (entry = first_entry; entry->start < end; entry = next_entry) {
3159 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
3161 * We have not yet clipped the entry.
3163 next_entry = vm_map_entry_in_transition(map, start,
3164 &end, holes_ok, entry);
3165 if (next_entry == NULL) {
3166 if (entry == first_entry) {
3168 return (KERN_INVALID_ADDRESS);
3170 rv = KERN_INVALID_ADDRESS;
3173 first_entry = (entry == first_entry) ?
3177 rv = vm_map_clip_start(map, entry, start);
3178 if (rv != KERN_SUCCESS)
3180 rv = vm_map_clip_end(map, entry, end);
3181 if (rv != KERN_SUCCESS)
3185 * Mark the entry in case the map lock is released. (See
3188 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 &&
3189 entry->wiring_thread == NULL,
3190 ("owned map entry %p", entry));
3191 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
3192 entry->wiring_thread = curthread;
3193 next_entry = vm_map_entry_succ(entry);
3195 * Check the map for holes in the specified region.
3196 * If holes_ok, skip this check.
3199 entry->end < end && next_entry->start > entry->end) {
3201 rv = KERN_INVALID_ADDRESS;
3205 * If system unwiring, require that the entry is system wired.
3208 vm_map_entry_system_wired_count(entry) == 0) {
3210 rv = KERN_INVALID_ARGUMENT;
3214 need_wakeup = false;
3215 if (first_entry == NULL &&
3216 !vm_map_lookup_entry(map, start, &first_entry)) {
3217 KASSERT(holes_ok, ("vm_map_unwire: lookup failed"));
3218 prev_entry = first_entry;
3219 entry = vm_map_entry_succ(first_entry);
3221 prev_entry = vm_map_entry_pred(first_entry);
3222 entry = first_entry;
3224 for (; entry->start < end;
3225 prev_entry = entry, entry = vm_map_entry_succ(entry)) {
3227 * If holes_ok was specified, an empty
3228 * space in the unwired region could have been mapped
3229 * while the map lock was dropped for draining
3230 * MAP_ENTRY_IN_TRANSITION. Moreover, another thread
3231 * could be simultaneously wiring this new mapping
3232 * entry. Detect these cases and skip any entries
3233 * marked as in transition by us.
3235 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 ||
3236 entry->wiring_thread != curthread) {
3238 ("vm_map_unwire: !HOLESOK and new/changed entry"));
3242 if (rv == KERN_SUCCESS && (!user_unwire ||
3243 (entry->eflags & MAP_ENTRY_USER_WIRED))) {
3244 if (entry->wired_count == 1)
3245 vm_map_entry_unwire(map, entry);
3247 entry->wired_count--;
3249 entry->eflags &= ~MAP_ENTRY_USER_WIRED;
3251 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0,
3252 ("vm_map_unwire: in-transition flag missing %p", entry));
3253 KASSERT(entry->wiring_thread == curthread,
3254 ("vm_map_unwire: alien wire %p", entry));
3255 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION;
3256 entry->wiring_thread = NULL;
3257 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
3258 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
3261 vm_map_try_merge_entries(map, prev_entry, entry);
3263 vm_map_try_merge_entries(map, prev_entry, entry);
3271 vm_map_wire_user_count_sub(u_long npages)
3274 atomic_subtract_long(&vm_user_wire_count, npages);
3278 vm_map_wire_user_count_add(u_long npages)
3282 wired = vm_user_wire_count;
3284 if (npages + wired > vm_page_max_user_wired)
3286 } while (!atomic_fcmpset_long(&vm_user_wire_count, &wired,
3293 * vm_map_wire_entry_failure:
3295 * Handle a wiring failure on the given entry.
3297 * The map should be locked.
3300 vm_map_wire_entry_failure(vm_map_t map, vm_map_entry_t entry,
3301 vm_offset_t failed_addr)
3304 VM_MAP_ASSERT_LOCKED(map);
3305 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 &&
3306 entry->wired_count == 1,
3307 ("vm_map_wire_entry_failure: entry %p isn't being wired", entry));
3308 KASSERT(failed_addr < entry->end,
3309 ("vm_map_wire_entry_failure: entry %p was fully wired", entry));
3312 * If any pages at the start of this entry were successfully wired,
3315 if (failed_addr > entry->start) {
3316 pmap_unwire(map->pmap, entry->start, failed_addr);
3317 vm_object_unwire(entry->object.vm_object, entry->offset,
3318 failed_addr - entry->start, PQ_ACTIVE);
3322 * Assign an out-of-range value to represent the failure to wire this
3325 entry->wired_count = -1;
3329 vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t end, int flags)
3334 rv = vm_map_wire_locked(map, start, end, flags);
3340 * vm_map_wire_locked:
3342 * Implements both kernel and user wiring. Returns with the map locked,
3343 * the map lock may be dropped.
3346 vm_map_wire_locked(vm_map_t map, vm_offset_t start, vm_offset_t end, int flags)
3348 vm_map_entry_t entry, first_entry, next_entry, prev_entry;
3349 vm_offset_t faddr, saved_end, saved_start;
3350 u_long incr, npages;
3351 u_int bidx, last_timestamp;
3353 bool holes_ok, need_wakeup, user_wire;
3356 VM_MAP_ASSERT_LOCKED(map);
3359 return (KERN_SUCCESS);
3361 if (flags & VM_MAP_WIRE_WRITE)
3362 prot |= VM_PROT_WRITE;
3363 holes_ok = (flags & VM_MAP_WIRE_HOLESOK) != 0;
3364 user_wire = (flags & VM_MAP_WIRE_USER) != 0;
3365 VM_MAP_RANGE_CHECK(map, start, end);
3366 if (!vm_map_lookup_entry(map, start, &first_entry)) {
3368 first_entry = vm_map_entry_succ(first_entry);
3370 return (KERN_INVALID_ADDRESS);
3372 for (entry = first_entry; entry->start < end; entry = next_entry) {
3373 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
3375 * We have not yet clipped the entry.
3377 next_entry = vm_map_entry_in_transition(map, start,
3378 &end, holes_ok, entry);
3379 if (next_entry == NULL) {
3380 if (entry == first_entry)
3381 return (KERN_INVALID_ADDRESS);
3382 rv = KERN_INVALID_ADDRESS;
3385 first_entry = (entry == first_entry) ?
3389 rv = vm_map_clip_start(map, entry, start);
3390 if (rv != KERN_SUCCESS)
3392 rv = vm_map_clip_end(map, entry, end);
3393 if (rv != KERN_SUCCESS)
3397 * Mark the entry in case the map lock is released. (See
3400 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 &&
3401 entry->wiring_thread == NULL,
3402 ("owned map entry %p", entry));
3403 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
3404 entry->wiring_thread = curthread;
3405 if ((entry->protection & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0
3406 || (entry->protection & prot) != prot) {
3407 entry->eflags |= MAP_ENTRY_WIRE_SKIPPED;
3410 rv = KERN_INVALID_ADDRESS;
3413 } else if (entry->wired_count == 0) {
3414 entry->wired_count++;
3416 npages = atop(entry->end - entry->start);
3417 if (user_wire && !vm_map_wire_user_count_add(npages)) {
3418 vm_map_wire_entry_failure(map, entry,
3421 rv = KERN_RESOURCE_SHORTAGE;
3426 * Release the map lock, relying on the in-transition
3427 * mark. Mark the map busy for fork.
3429 saved_start = entry->start;
3430 saved_end = entry->end;
3431 last_timestamp = map->timestamp;
3432 bidx = (entry->eflags & MAP_ENTRY_SPLIT_BOUNDARY_MASK)
3433 >> MAP_ENTRY_SPLIT_BOUNDARY_SHIFT;
3434 incr = pagesizes[bidx];
3438 for (faddr = saved_start; faddr < saved_end;
3441 * Simulate a fault to get the page and enter
3442 * it into the physical map.
3444 rv = vm_fault(map, faddr, VM_PROT_NONE,
3445 VM_FAULT_WIRE, NULL);
3446 if (rv != KERN_SUCCESS)
3451 if (last_timestamp + 1 != map->timestamp) {
3453 * Look again for the entry because the map was
3454 * modified while it was unlocked. The entry
3455 * may have been clipped, but NOT merged or
3458 if (!vm_map_lookup_entry(map, saved_start,
3461 ("vm_map_wire: lookup failed"));
3462 first_entry = (entry == first_entry) ?
3464 for (entry = next_entry; entry->end < saved_end;
3465 entry = vm_map_entry_succ(entry)) {
3467 * In case of failure, handle entries
3468 * that were not fully wired here;
3469 * fully wired entries are handled
3472 if (rv != KERN_SUCCESS &&
3474 vm_map_wire_entry_failure(map,
3478 if (rv != KERN_SUCCESS) {
3479 vm_map_wire_entry_failure(map, entry, faddr);
3481 vm_map_wire_user_count_sub(npages);
3485 } else if (!user_wire ||
3486 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
3487 entry->wired_count++;
3490 * Check the map for holes in the specified region.
3491 * If holes_ok was specified, skip this check.
3493 next_entry = vm_map_entry_succ(entry);
3495 entry->end < end && next_entry->start > entry->end) {
3497 rv = KERN_INVALID_ADDRESS;
3503 need_wakeup = false;
3504 if (first_entry == NULL &&
3505 !vm_map_lookup_entry(map, start, &first_entry)) {
3506 KASSERT(holes_ok, ("vm_map_wire: lookup failed"));
3507 prev_entry = first_entry;
3508 entry = vm_map_entry_succ(first_entry);
3510 prev_entry = vm_map_entry_pred(first_entry);
3511 entry = first_entry;
3513 for (; entry->start < end;
3514 prev_entry = entry, entry = vm_map_entry_succ(entry)) {
3516 * If holes_ok was specified, an empty
3517 * space in the unwired region could have been mapped
3518 * while the map lock was dropped for faulting in the
3519 * pages or draining MAP_ENTRY_IN_TRANSITION.
3520 * Moreover, another thread could be simultaneously
3521 * wiring this new mapping entry. Detect these cases
3522 * and skip any entries marked as in transition not by us.
3524 * Another way to get an entry not marked with
3525 * MAP_ENTRY_IN_TRANSITION is after failed clipping,
3526 * which set rv to KERN_INVALID_ARGUMENT.
3528 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 ||
3529 entry->wiring_thread != curthread) {
3530 KASSERT(holes_ok || rv == KERN_INVALID_ARGUMENT,
3531 ("vm_map_wire: !HOLESOK and new/changed entry"));
3535 if ((entry->eflags & MAP_ENTRY_WIRE_SKIPPED) != 0) {
3537 } else if (rv == KERN_SUCCESS) {
3539 entry->eflags |= MAP_ENTRY_USER_WIRED;
3540 } else if (entry->wired_count == -1) {
3542 * Wiring failed on this entry. Thus, unwiring is
3545 entry->wired_count = 0;
3546 } else if (!user_wire ||
3547 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
3549 * Undo the wiring. Wiring succeeded on this entry
3550 * but failed on a later entry.
3552 if (entry->wired_count == 1) {
3553 vm_map_entry_unwire(map, entry);
3555 vm_map_wire_user_count_sub(
3556 atop(entry->end - entry->start));
3558 entry->wired_count--;
3560 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0,
3561 ("vm_map_wire: in-transition flag missing %p", entry));
3562 KASSERT(entry->wiring_thread == curthread,
3563 ("vm_map_wire: alien wire %p", entry));
3564 entry->eflags &= ~(MAP_ENTRY_IN_TRANSITION |
3565 MAP_ENTRY_WIRE_SKIPPED);
3566 entry->wiring_thread = NULL;
3567 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
3568 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
3571 vm_map_try_merge_entries(map, prev_entry, entry);
3573 vm_map_try_merge_entries(map, prev_entry, entry);
3582 * Push any dirty cached pages in the address range to their pager.
3583 * If syncio is TRUE, dirty pages are written synchronously.
3584 * If invalidate is TRUE, any cached pages are freed as well.
3586 * If the size of the region from start to end is zero, we are
3587 * supposed to flush all modified pages within the region containing
3588 * start. Unfortunately, a region can be split or coalesced with
3589 * neighboring regions, making it difficult to determine what the
3590 * original region was. Therefore, we approximate this requirement by
3591 * flushing the current region containing start.
3593 * Returns an error if any part of the specified range is not mapped.
3601 boolean_t invalidate)
3603 vm_map_entry_t entry, first_entry, next_entry;
3606 vm_ooffset_t offset;
3607 unsigned int last_timestamp;
3611 vm_map_lock_read(map);
3612 VM_MAP_RANGE_CHECK(map, start, end);
3613 if (!vm_map_lookup_entry(map, start, &first_entry)) {
3614 vm_map_unlock_read(map);
3615 return (KERN_INVALID_ADDRESS);
3616 } else if (start == end) {
3617 start = first_entry->start;
3618 end = first_entry->end;
3622 * Make a first pass to check for user-wired memory, holes,
3623 * and partial invalidation of largepage mappings.
3625 for (entry = first_entry; entry->start < end; entry = next_entry) {
3627 if ((entry->eflags & MAP_ENTRY_USER_WIRED) != 0) {
3628 vm_map_unlock_read(map);
3629 return (KERN_INVALID_ARGUMENT);
3631 bdry_idx = (entry->eflags &
3632 MAP_ENTRY_SPLIT_BOUNDARY_MASK) >>
3633 MAP_ENTRY_SPLIT_BOUNDARY_SHIFT;
3634 if (bdry_idx != 0 &&
3635 ((start & (pagesizes[bdry_idx] - 1)) != 0 ||
3636 (end & (pagesizes[bdry_idx] - 1)) != 0)) {
3637 vm_map_unlock_read(map);
3638 return (KERN_INVALID_ARGUMENT);
3641 next_entry = vm_map_entry_succ(entry);
3642 if (end > entry->end &&
3643 entry->end != next_entry->start) {
3644 vm_map_unlock_read(map);
3645 return (KERN_INVALID_ADDRESS);
3650 pmap_remove(map->pmap, start, end);
3654 * Make a second pass, cleaning/uncaching pages from the indicated
3657 for (entry = first_entry; entry->start < end;) {
3658 offset = entry->offset + (start - entry->start);
3659 size = (end <= entry->end ? end : entry->end) - start;
3660 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0) {
3662 vm_map_entry_t tentry;
3665 smap = entry->object.sub_map;
3666 vm_map_lock_read(smap);
3667 (void) vm_map_lookup_entry(smap, offset, &tentry);
3668 tsize = tentry->end - offset;
3671 object = tentry->object.vm_object;
3672 offset = tentry->offset + (offset - tentry->start);
3673 vm_map_unlock_read(smap);
3675 object = entry->object.vm_object;
3677 vm_object_reference(object);
3678 last_timestamp = map->timestamp;
3679 vm_map_unlock_read(map);
3680 if (!vm_object_sync(object, offset, size, syncio, invalidate))
3683 vm_object_deallocate(object);
3684 vm_map_lock_read(map);
3685 if (last_timestamp == map->timestamp ||
3686 !vm_map_lookup_entry(map, start, &entry))
3687 entry = vm_map_entry_succ(entry);
3690 vm_map_unlock_read(map);
3691 return (failed ? KERN_FAILURE : KERN_SUCCESS);
3695 * vm_map_entry_unwire: [ internal use only ]
3697 * Make the region specified by this entry pageable.
3699 * The map in question should be locked.
3700 * [This is the reason for this routine's existence.]
3703 vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry)
3707 VM_MAP_ASSERT_LOCKED(map);
3708 KASSERT(entry->wired_count > 0,
3709 ("vm_map_entry_unwire: entry %p isn't wired", entry));
3711 size = entry->end - entry->start;
3712 if ((entry->eflags & MAP_ENTRY_USER_WIRED) != 0)
3713 vm_map_wire_user_count_sub(atop(size));
3714 pmap_unwire(map->pmap, entry->start, entry->end);
3715 vm_object_unwire(entry->object.vm_object, entry->offset, size,
3717 entry->wired_count = 0;
3721 vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map)
3724 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0)
3725 vm_object_deallocate(entry->object.vm_object);
3726 uma_zfree(system_map ? kmapentzone : mapentzone, entry);
3730 * vm_map_entry_delete: [ internal use only ]
3732 * Deallocate the given entry from the target map.
3735 vm_map_entry_delete(vm_map_t map, vm_map_entry_t entry)
3738 vm_pindex_t offidxstart, offidxend, size1;
3741 vm_map_entry_unlink(map, entry, UNLINK_MERGE_NONE);
3742 object = entry->object.vm_object;
3744 if ((entry->eflags & MAP_ENTRY_GUARD) != 0) {
3745 MPASS(entry->cred == NULL);
3746 MPASS((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0);
3747 MPASS(object == NULL);
3748 vm_map_entry_deallocate(entry, map->system_map);
3752 size = entry->end - entry->start;
3755 if (entry->cred != NULL) {
3756 swap_release_by_cred(size, entry->cred);
3757 crfree(entry->cred);
3760 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0 || object == NULL) {
3761 entry->object.vm_object = NULL;
3762 } else if ((object->flags & OBJ_ANON) != 0 ||
3763 object == kernel_object) {
3764 KASSERT(entry->cred == NULL || object->cred == NULL ||
3765 (entry->eflags & MAP_ENTRY_NEEDS_COPY),
3766 ("OVERCOMMIT vm_map_entry_delete: both cred %p", entry));
3767 offidxstart = OFF_TO_IDX(entry->offset);
3768 offidxend = offidxstart + atop(size);
3769 VM_OBJECT_WLOCK(object);
3770 if (object->ref_count != 1 &&
3771 ((object->flags & OBJ_ONEMAPPING) != 0 ||
3772 object == kernel_object)) {
3773 vm_object_collapse(object);
3776 * The option OBJPR_NOTMAPPED can be passed here
3777 * because vm_map_delete() already performed
3778 * pmap_remove() on the only mapping to this range
3781 vm_object_page_remove(object, offidxstart, offidxend,
3783 if (offidxend >= object->size &&
3784 offidxstart < object->size) {
3785 size1 = object->size;
3786 object->size = offidxstart;
3787 if (object->cred != NULL) {
3788 size1 -= object->size;
3789 KASSERT(object->charge >= ptoa(size1),
3790 ("object %p charge < 0", object));
3791 swap_release_by_cred(ptoa(size1),
3793 object->charge -= ptoa(size1);
3797 VM_OBJECT_WUNLOCK(object);
3799 if (map->system_map)
3800 vm_map_entry_deallocate(entry, TRUE);
3802 entry->defer_next = curthread->td_map_def_user;
3803 curthread->td_map_def_user = entry;
3808 * vm_map_delete: [ internal use only ]
3810 * Deallocates the given address range from the target
3814 vm_map_delete(vm_map_t map, vm_offset_t start, vm_offset_t end)
3816 vm_map_entry_t entry, next_entry, scratch_entry;
3819 VM_MAP_ASSERT_LOCKED(map);
3822 return (KERN_SUCCESS);
3825 * Find the start of the region, and clip it.
3826 * Step through all entries in this region.
3828 rv = vm_map_lookup_clip_start(map, start, &entry, &scratch_entry);
3829 if (rv != KERN_SUCCESS)
3831 for (; entry->start < end; entry = next_entry) {
3833 * Wait for wiring or unwiring of an entry to complete.
3834 * Also wait for any system wirings to disappear on
3837 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 ||
3838 (vm_map_pmap(map) != kernel_pmap &&
3839 vm_map_entry_system_wired_count(entry) != 0)) {
3840 unsigned int last_timestamp;
3841 vm_offset_t saved_start;
3843 saved_start = entry->start;
3844 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
3845 last_timestamp = map->timestamp;
3846 (void) vm_map_unlock_and_wait(map, 0);
3848 if (last_timestamp + 1 != map->timestamp) {
3850 * Look again for the entry because the map was
3851 * modified while it was unlocked.
3852 * Specifically, the entry may have been
3853 * clipped, merged, or deleted.
3855 rv = vm_map_lookup_clip_start(map, saved_start,
3856 &next_entry, &scratch_entry);
3857 if (rv != KERN_SUCCESS)
3864 /* XXXKIB or delete to the upper superpage boundary ? */
3865 rv = vm_map_clip_end(map, entry, end);
3866 if (rv != KERN_SUCCESS)
3868 next_entry = vm_map_entry_succ(entry);
3871 * Unwire before removing addresses from the pmap; otherwise,
3872 * unwiring will put the entries back in the pmap.
3874 if (entry->wired_count != 0)
3875 vm_map_entry_unwire(map, entry);
3878 * Remove mappings for the pages, but only if the
3879 * mappings could exist. For instance, it does not
3880 * make sense to call pmap_remove() for guard entries.
3882 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0 ||
3883 entry->object.vm_object != NULL)
3884 pmap_remove(map->pmap, entry->start, entry->end);
3886 if (entry->end == map->anon_loc)
3887 map->anon_loc = entry->start;
3890 * Delete the entry only after removing all pmap
3891 * entries pointing to its pages. (Otherwise, its
3892 * page frames may be reallocated, and any modify bits
3893 * will be set in the wrong object!)
3895 vm_map_entry_delete(map, entry);
3903 * Remove the given address range from the target map.
3904 * This is the exported form of vm_map_delete.
3907 vm_map_remove(vm_map_t map, vm_offset_t start, vm_offset_t end)
3912 VM_MAP_RANGE_CHECK(map, start, end);
3913 result = vm_map_delete(map, start, end);
3919 * vm_map_check_protection:
3921 * Assert that the target map allows the specified privilege on the
3922 * entire address region given. The entire region must be allocated.
3924 * WARNING! This code does not and should not check whether the
3925 * contents of the region is accessible. For example a smaller file
3926 * might be mapped into a larger address space.
3928 * NOTE! This code is also called by munmap().
3930 * The map must be locked. A read lock is sufficient.
3933 vm_map_check_protection(vm_map_t map, vm_offset_t start, vm_offset_t end,
3934 vm_prot_t protection)
3936 vm_map_entry_t entry;
3937 vm_map_entry_t tmp_entry;
3939 if (!vm_map_lookup_entry(map, start, &tmp_entry))
3943 while (start < end) {
3947 if (start < entry->start)
3950 * Check protection associated with entry.
3952 if ((entry->protection & protection) != protection)
3954 /* go to next entry */
3956 entry = vm_map_entry_succ(entry);
3963 * vm_map_copy_swap_object:
3965 * Copies a swap-backed object from an existing map entry to a
3966 * new one. Carries forward the swap charge. May change the
3967 * src object on return.
3970 vm_map_copy_swap_object(vm_map_entry_t src_entry, vm_map_entry_t dst_entry,
3971 vm_offset_t size, vm_ooffset_t *fork_charge)
3973 vm_object_t src_object;
3977 src_object = src_entry->object.vm_object;
3978 charged = ENTRY_CHARGED(src_entry);
3979 if ((src_object->flags & OBJ_ANON) != 0) {
3980 VM_OBJECT_WLOCK(src_object);
3981 vm_object_collapse(src_object);
3982 if ((src_object->flags & OBJ_ONEMAPPING) != 0) {
3983 vm_object_split(src_entry);
3984 src_object = src_entry->object.vm_object;
3986 vm_object_reference_locked(src_object);
3987 vm_object_clear_flag(src_object, OBJ_ONEMAPPING);
3988 VM_OBJECT_WUNLOCK(src_object);
3990 vm_object_reference(src_object);
3991 if (src_entry->cred != NULL &&
3992 !(src_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
3993 KASSERT(src_object->cred == NULL,
3994 ("OVERCOMMIT: vm_map_copy_anon_entry: cred %p",
3996 src_object->cred = src_entry->cred;
3997 src_object->charge = size;
3999 dst_entry->object.vm_object = src_object;
4001 cred = curthread->td_ucred;
4003 dst_entry->cred = cred;
4004 *fork_charge += size;
4005 if (!(src_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
4007 src_entry->cred = cred;
4008 *fork_charge += size;
4014 * vm_map_copy_entry:
4016 * Copies the contents of the source entry to the destination
4017 * entry. The entries *must* be aligned properly.
4023 vm_map_entry_t src_entry,
4024 vm_map_entry_t dst_entry,
4025 vm_ooffset_t *fork_charge)
4027 vm_object_t src_object;
4028 vm_map_entry_t fake_entry;
4031 VM_MAP_ASSERT_LOCKED(dst_map);
4033 if ((dst_entry->eflags|src_entry->eflags) & MAP_ENTRY_IS_SUB_MAP)
4036 if (src_entry->wired_count == 0 ||
4037 (src_entry->protection & VM_PROT_WRITE) == 0) {
4039 * If the source entry is marked needs_copy, it is already
4042 if ((src_entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0 &&
4043 (src_entry->protection & VM_PROT_WRITE) != 0) {
4044 pmap_protect(src_map->pmap,
4047 src_entry->protection & ~VM_PROT_WRITE);
4051 * Make a copy of the object.
4053 size = src_entry->end - src_entry->start;
4054 if ((src_object = src_entry->object.vm_object) != NULL) {
4055 if (src_object->type == OBJT_DEFAULT ||
4056 src_object->type == OBJT_SWAP) {
4057 vm_map_copy_swap_object(src_entry, dst_entry,
4059 /* May have split/collapsed, reload obj. */
4060 src_object = src_entry->object.vm_object;
4062 vm_object_reference(src_object);
4063 dst_entry->object.vm_object = src_object;
4065 src_entry->eflags |= MAP_ENTRY_COW |
4066 MAP_ENTRY_NEEDS_COPY;
4067 dst_entry->eflags |= MAP_ENTRY_COW |
4068 MAP_ENTRY_NEEDS_COPY;
4069 dst_entry->offset = src_entry->offset;
4070 if (src_entry->eflags & MAP_ENTRY_WRITECNT) {
4072 * MAP_ENTRY_WRITECNT cannot
4073 * indicate write reference from
4074 * src_entry, since the entry is
4075 * marked as needs copy. Allocate a
4076 * fake entry that is used to
4077 * decrement object->un_pager writecount
4078 * at the appropriate time. Attach
4079 * fake_entry to the deferred list.
4081 fake_entry = vm_map_entry_create(dst_map);
4082 fake_entry->eflags = MAP_ENTRY_WRITECNT;
4083 src_entry->eflags &= ~MAP_ENTRY_WRITECNT;
4084 vm_object_reference(src_object);
4085 fake_entry->object.vm_object = src_object;
4086 fake_entry->start = src_entry->start;
4087 fake_entry->end = src_entry->end;
4088 fake_entry->defer_next =
4089 curthread->td_map_def_user;
4090 curthread->td_map_def_user = fake_entry;
4093 pmap_copy(dst_map->pmap, src_map->pmap,
4094 dst_entry->start, dst_entry->end - dst_entry->start,
4097 dst_entry->object.vm_object = NULL;
4098 dst_entry->offset = 0;
4099 if (src_entry->cred != NULL) {
4100 dst_entry->cred = curthread->td_ucred;
4101 crhold(dst_entry->cred);
4102 *fork_charge += size;
4107 * We don't want to make writeable wired pages copy-on-write.
4108 * Immediately copy these pages into the new map by simulating
4109 * page faults. The new pages are pageable.
4111 vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry,
4117 * vmspace_map_entry_forked:
4118 * Update the newly-forked vmspace each time a map entry is inherited
4119 * or copied. The values for vm_dsize and vm_tsize are approximate
4120 * (and mostly-obsolete ideas in the face of mmap(2) et al.)
4123 vmspace_map_entry_forked(const struct vmspace *vm1, struct vmspace *vm2,
4124 vm_map_entry_t entry)
4126 vm_size_t entrysize;
4129 if ((entry->eflags & MAP_ENTRY_GUARD) != 0)
4131 entrysize = entry->end - entry->start;
4132 vm2->vm_map.size += entrysize;
4133 if (entry->eflags & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP)) {
4134 vm2->vm_ssize += btoc(entrysize);
4135 } else if (entry->start >= (vm_offset_t)vm1->vm_daddr &&
4136 entry->start < (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)) {
4137 newend = MIN(entry->end,
4138 (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize));
4139 vm2->vm_dsize += btoc(newend - entry->start);
4140 } else if (entry->start >= (vm_offset_t)vm1->vm_taddr &&
4141 entry->start < (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)) {
4142 newend = MIN(entry->end,
4143 (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize));
4144 vm2->vm_tsize += btoc(newend - entry->start);
4150 * Create a new process vmspace structure and vm_map
4151 * based on those of an existing process. The new map
4152 * is based on the old map, according to the inheritance
4153 * values on the regions in that map.
4155 * XXX It might be worth coalescing the entries added to the new vmspace.
4157 * The source map must not be locked.
4160 vmspace_fork(struct vmspace *vm1, vm_ooffset_t *fork_charge)
4162 struct vmspace *vm2;
4163 vm_map_t new_map, old_map;
4164 vm_map_entry_t new_entry, old_entry;
4169 old_map = &vm1->vm_map;
4170 /* Copy immutable fields of vm1 to vm2. */
4171 vm2 = vmspace_alloc(vm_map_min(old_map), vm_map_max(old_map),
4176 vm2->vm_taddr = vm1->vm_taddr;
4177 vm2->vm_daddr = vm1->vm_daddr;
4178 vm2->vm_maxsaddr = vm1->vm_maxsaddr;
4179 vm_map_lock(old_map);
4181 vm_map_wait_busy(old_map);
4182 new_map = &vm2->vm_map;
4183 locked = vm_map_trylock(new_map); /* trylock to silence WITNESS */
4184 KASSERT(locked, ("vmspace_fork: lock failed"));
4186 error = pmap_vmspace_copy(new_map->pmap, old_map->pmap);
4188 sx_xunlock(&old_map->lock);
4189 sx_xunlock(&new_map->lock);
4190 vm_map_process_deferred();
4195 new_map->anon_loc = old_map->anon_loc;
4196 new_map->flags |= old_map->flags & (MAP_ASLR | MAP_ASLR_IGNSTART);
4198 VM_MAP_ENTRY_FOREACH(old_entry, old_map) {
4199 if ((old_entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0)
4200 panic("vm_map_fork: encountered a submap");
4202 inh = old_entry->inheritance;
4203 if ((old_entry->eflags & MAP_ENTRY_GUARD) != 0 &&
4204 inh != VM_INHERIT_NONE)
4205 inh = VM_INHERIT_COPY;
4208 case VM_INHERIT_NONE:
4211 case VM_INHERIT_SHARE:
4213 * Clone the entry, creating the shared object if
4216 object = old_entry->object.vm_object;
4217 if (object == NULL) {
4218 vm_map_entry_back(old_entry);
4219 object = old_entry->object.vm_object;
4223 * Add the reference before calling vm_object_shadow
4224 * to insure that a shadow object is created.
4226 vm_object_reference(object);
4227 if (old_entry->eflags & MAP_ENTRY_NEEDS_COPY) {
4228 vm_object_shadow(&old_entry->object.vm_object,
4230 old_entry->end - old_entry->start,
4232 /* Transfer the second reference too. */
4234 old_entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
4235 old_entry->cred = NULL;
4238 * As in vm_map_merged_neighbor_dispose(),
4239 * the vnode lock will not be acquired in
4240 * this call to vm_object_deallocate().
4242 vm_object_deallocate(object);
4243 object = old_entry->object.vm_object;
4245 VM_OBJECT_WLOCK(object);
4246 vm_object_clear_flag(object, OBJ_ONEMAPPING);
4247 if (old_entry->cred != NULL) {
4248 KASSERT(object->cred == NULL,
4249 ("vmspace_fork both cred"));
4250 object->cred = old_entry->cred;
4251 object->charge = old_entry->end -
4253 old_entry->cred = NULL;
4257 * Assert the correct state of the vnode
4258 * v_writecount while the object is locked, to
4259 * not relock it later for the assertion
4262 if (old_entry->eflags & MAP_ENTRY_WRITECNT &&
4263 object->type == OBJT_VNODE) {
4264 KASSERT(((struct vnode *)object->
4265 handle)->v_writecount > 0,
4266 ("vmspace_fork: v_writecount %p",
4268 KASSERT(object->un_pager.vnp.
4270 ("vmspace_fork: vnp.writecount %p",
4273 VM_OBJECT_WUNLOCK(object);
4277 * Clone the entry, referencing the shared object.
4279 new_entry = vm_map_entry_create(new_map);
4280 *new_entry = *old_entry;
4281 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED |
4282 MAP_ENTRY_IN_TRANSITION);
4283 new_entry->wiring_thread = NULL;
4284 new_entry->wired_count = 0;
4285 if (new_entry->eflags & MAP_ENTRY_WRITECNT) {
4286 vm_pager_update_writecount(object,
4287 new_entry->start, new_entry->end);
4289 vm_map_entry_set_vnode_text(new_entry, true);
4292 * Insert the entry into the new map -- we know we're
4293 * inserting at the end of the new map.
4295 vm_map_entry_link(new_map, new_entry);
4296 vmspace_map_entry_forked(vm1, vm2, new_entry);
4299 * Update the physical map
4301 pmap_copy(new_map->pmap, old_map->pmap,
4303 (old_entry->end - old_entry->start),
4307 case VM_INHERIT_COPY:
4309 * Clone the entry and link into the map.
4311 new_entry = vm_map_entry_create(new_map);
4312 *new_entry = *old_entry;
4314 * Copied entry is COW over the old object.
4316 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED |
4317 MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_WRITECNT);
4318 new_entry->wiring_thread = NULL;
4319 new_entry->wired_count = 0;
4320 new_entry->object.vm_object = NULL;
4321 new_entry->cred = NULL;
4322 vm_map_entry_link(new_map, new_entry);
4323 vmspace_map_entry_forked(vm1, vm2, new_entry);
4324 vm_map_copy_entry(old_map, new_map, old_entry,
4325 new_entry, fork_charge);
4326 vm_map_entry_set_vnode_text(new_entry, true);
4329 case VM_INHERIT_ZERO:
4331 * Create a new anonymous mapping entry modelled from
4334 new_entry = vm_map_entry_create(new_map);
4335 memset(new_entry, 0, sizeof(*new_entry));
4337 new_entry->start = old_entry->start;
4338 new_entry->end = old_entry->end;
4339 new_entry->eflags = old_entry->eflags &
4340 ~(MAP_ENTRY_USER_WIRED | MAP_ENTRY_IN_TRANSITION |
4341 MAP_ENTRY_WRITECNT | MAP_ENTRY_VN_EXEC |
4342 MAP_ENTRY_SPLIT_BOUNDARY_MASK);
4343 new_entry->protection = old_entry->protection;
4344 new_entry->max_protection = old_entry->max_protection;
4345 new_entry->inheritance = VM_INHERIT_ZERO;
4347 vm_map_entry_link(new_map, new_entry);
4348 vmspace_map_entry_forked(vm1, vm2, new_entry);
4350 new_entry->cred = curthread->td_ucred;
4351 crhold(new_entry->cred);
4352 *fork_charge += (new_entry->end - new_entry->start);
4358 * Use inlined vm_map_unlock() to postpone handling the deferred
4359 * map entries, which cannot be done until both old_map and
4360 * new_map locks are released.
4362 sx_xunlock(&old_map->lock);
4363 sx_xunlock(&new_map->lock);
4364 vm_map_process_deferred();
4370 * Create a process's stack for exec_new_vmspace(). This function is never
4371 * asked to wire the newly created stack.
4374 vm_map_stack(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
4375 vm_prot_t prot, vm_prot_t max, int cow)
4377 vm_size_t growsize, init_ssize;
4381 MPASS((map->flags & MAP_WIREFUTURE) == 0);
4382 growsize = sgrowsiz;
4383 init_ssize = (max_ssize < growsize) ? max_ssize : growsize;
4385 vmemlim = lim_cur(curthread, RLIMIT_VMEM);
4386 /* If we would blow our VMEM resource limit, no go */
4387 if (map->size + init_ssize > vmemlim) {
4391 rv = vm_map_stack_locked(map, addrbos, max_ssize, growsize, prot,
4398 static int stack_guard_page = 1;
4399 SYSCTL_INT(_security_bsd, OID_AUTO, stack_guard_page, CTLFLAG_RWTUN,
4400 &stack_guard_page, 0,
4401 "Specifies the number of guard pages for a stack that grows");
4404 vm_map_stack_locked(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
4405 vm_size_t growsize, vm_prot_t prot, vm_prot_t max, int cow)
4407 vm_map_entry_t new_entry, prev_entry;
4408 vm_offset_t bot, gap_bot, gap_top, top;
4409 vm_size_t init_ssize, sgp;
4413 * The stack orientation is piggybacked with the cow argument.
4414 * Extract it into orient and mask the cow argument so that we
4415 * don't pass it around further.
4417 orient = cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP);
4418 KASSERT(orient != 0, ("No stack grow direction"));
4419 KASSERT(orient != (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP),
4422 if (max_ssize == 0 ||
4423 !vm_map_range_valid(map, addrbos, addrbos + max_ssize))
4424 return (KERN_INVALID_ADDRESS);
4425 sgp = ((curproc->p_flag2 & P2_STKGAP_DISABLE) != 0 ||
4426 (curproc->p_fctl0 & NT_FREEBSD_FCTL_STKGAP_DISABLE) != 0) ? 0 :
4427 (vm_size_t)stack_guard_page * PAGE_SIZE;
4428 if (sgp >= max_ssize)
4429 return (KERN_INVALID_ARGUMENT);
4431 init_ssize = growsize;
4432 if (max_ssize < init_ssize + sgp)
4433 init_ssize = max_ssize - sgp;
4435 /* If addr is already mapped, no go */
4436 if (vm_map_lookup_entry(map, addrbos, &prev_entry))
4437 return (KERN_NO_SPACE);
4440 * If we can't accommodate max_ssize in the current mapping, no go.
4442 if (vm_map_entry_succ(prev_entry)->start < addrbos + max_ssize)
4443 return (KERN_NO_SPACE);
4446 * We initially map a stack of only init_ssize. We will grow as
4447 * needed later. Depending on the orientation of the stack (i.e.
4448 * the grow direction) we either map at the top of the range, the
4449 * bottom of the range or in the middle.
4451 * Note: we would normally expect prot and max to be VM_PROT_ALL,
4452 * and cow to be 0. Possibly we should eliminate these as input
4453 * parameters, and just pass these values here in the insert call.
4455 if (orient == MAP_STACK_GROWS_DOWN) {
4456 bot = addrbos + max_ssize - init_ssize;
4457 top = bot + init_ssize;
4460 } else /* if (orient == MAP_STACK_GROWS_UP) */ {
4462 top = bot + init_ssize;
4464 gap_top = addrbos + max_ssize;
4466 rv = vm_map_insert(map, NULL, 0, bot, top, prot, max, cow);
4467 if (rv != KERN_SUCCESS)
4469 new_entry = vm_map_entry_succ(prev_entry);
4470 KASSERT(new_entry->end == top || new_entry->start == bot,
4471 ("Bad entry start/end for new stack entry"));
4472 KASSERT((orient & MAP_STACK_GROWS_DOWN) == 0 ||
4473 (new_entry->eflags & MAP_ENTRY_GROWS_DOWN) != 0,
4474 ("new entry lacks MAP_ENTRY_GROWS_DOWN"));
4475 KASSERT((orient & MAP_STACK_GROWS_UP) == 0 ||
4476 (new_entry->eflags & MAP_ENTRY_GROWS_UP) != 0,
4477 ("new entry lacks MAP_ENTRY_GROWS_UP"));
4478 if (gap_bot == gap_top)
4479 return (KERN_SUCCESS);
4480 rv = vm_map_insert(map, NULL, 0, gap_bot, gap_top, VM_PROT_NONE,
4481 VM_PROT_NONE, MAP_CREATE_GUARD | (orient == MAP_STACK_GROWS_DOWN ?
4482 MAP_CREATE_STACK_GAP_DN : MAP_CREATE_STACK_GAP_UP));
4483 if (rv == KERN_SUCCESS) {
4485 * Gap can never successfully handle a fault, so
4486 * read-ahead logic is never used for it. Re-use
4487 * next_read of the gap entry to store
4488 * stack_guard_page for vm_map_growstack().
4490 if (orient == MAP_STACK_GROWS_DOWN)
4491 vm_map_entry_pred(new_entry)->next_read = sgp;
4493 vm_map_entry_succ(new_entry)->next_read = sgp;
4495 (void)vm_map_delete(map, bot, top);
4501 * Attempts to grow a vm stack entry. Returns KERN_SUCCESS if we
4502 * successfully grow the stack.
4505 vm_map_growstack(vm_map_t map, vm_offset_t addr, vm_map_entry_t gap_entry)
4507 vm_map_entry_t stack_entry;
4511 vm_offset_t gap_end, gap_start, grow_start;
4512 vm_size_t grow_amount, guard, max_grow;
4513 rlim_t lmemlim, stacklim, vmemlim;
4515 bool gap_deleted, grow_down, is_procstack;
4527 * Disallow stack growth when the access is performed by a
4528 * debugger or AIO daemon. The reason is that the wrong
4529 * resource limits are applied.
4531 if (p != initproc && (map != &p->p_vmspace->vm_map ||
4532 p->p_textvp == NULL))
4533 return (KERN_FAILURE);
4535 MPASS(!map->system_map);
4537 lmemlim = lim_cur(curthread, RLIMIT_MEMLOCK);
4538 stacklim = lim_cur(curthread, RLIMIT_STACK);
4539 vmemlim = lim_cur(curthread, RLIMIT_VMEM);
4541 /* If addr is not in a hole for a stack grow area, no need to grow. */
4542 if (gap_entry == NULL && !vm_map_lookup_entry(map, addr, &gap_entry))
4543 return (KERN_FAILURE);
4544 if ((gap_entry->eflags & MAP_ENTRY_GUARD) == 0)
4545 return (KERN_SUCCESS);
4546 if ((gap_entry->eflags & MAP_ENTRY_STACK_GAP_DN) != 0) {
4547 stack_entry = vm_map_entry_succ(gap_entry);
4548 if ((stack_entry->eflags & MAP_ENTRY_GROWS_DOWN) == 0 ||
4549 stack_entry->start != gap_entry->end)
4550 return (KERN_FAILURE);
4551 grow_amount = round_page(stack_entry->start - addr);
4553 } else if ((gap_entry->eflags & MAP_ENTRY_STACK_GAP_UP) != 0) {
4554 stack_entry = vm_map_entry_pred(gap_entry);
4555 if ((stack_entry->eflags & MAP_ENTRY_GROWS_UP) == 0 ||
4556 stack_entry->end != gap_entry->start)
4557 return (KERN_FAILURE);
4558 grow_amount = round_page(addr + 1 - stack_entry->end);
4561 return (KERN_FAILURE);
4563 guard = ((curproc->p_flag2 & P2_STKGAP_DISABLE) != 0 ||
4564 (curproc->p_fctl0 & NT_FREEBSD_FCTL_STKGAP_DISABLE) != 0) ? 0 :
4565 gap_entry->next_read;
4566 max_grow = gap_entry->end - gap_entry->start;
4567 if (guard > max_grow)
4568 return (KERN_NO_SPACE);
4570 if (grow_amount > max_grow)
4571 return (KERN_NO_SPACE);
4574 * If this is the main process stack, see if we're over the stack
4577 is_procstack = addr >= (vm_offset_t)vm->vm_maxsaddr &&
4578 addr < (vm_offset_t)p->p_sysent->sv_usrstack;
4579 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim))
4580 return (KERN_NO_SPACE);
4585 if (is_procstack && racct_set(p, RACCT_STACK,
4586 ctob(vm->vm_ssize) + grow_amount)) {
4588 return (KERN_NO_SPACE);
4594 grow_amount = roundup(grow_amount, sgrowsiz);
4595 if (grow_amount > max_grow)
4596 grow_amount = max_grow;
4597 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) {
4598 grow_amount = trunc_page((vm_size_t)stacklim) -
4604 limit = racct_get_available(p, RACCT_STACK);
4606 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > limit))
4607 grow_amount = limit - ctob(vm->vm_ssize);
4610 if (!old_mlock && (map->flags & MAP_WIREFUTURE) != 0) {
4611 if (ptoa(pmap_wired_count(map->pmap)) + grow_amount > lmemlim) {
4618 if (racct_set(p, RACCT_MEMLOCK,
4619 ptoa(pmap_wired_count(map->pmap)) + grow_amount)) {
4629 /* If we would blow our VMEM resource limit, no go */
4630 if (map->size + grow_amount > vmemlim) {
4637 if (racct_set(p, RACCT_VMEM, map->size + grow_amount)) {
4646 if (vm_map_lock_upgrade(map)) {
4648 vm_map_lock_read(map);
4653 grow_start = gap_entry->end - grow_amount;
4654 if (gap_entry->start + grow_amount == gap_entry->end) {
4655 gap_start = gap_entry->start;
4656 gap_end = gap_entry->end;
4657 vm_map_entry_delete(map, gap_entry);
4660 MPASS(gap_entry->start < gap_entry->end - grow_amount);
4661 vm_map_entry_resize(map, gap_entry, -grow_amount);
4662 gap_deleted = false;
4664 rv = vm_map_insert(map, NULL, 0, grow_start,
4665 grow_start + grow_amount,
4666 stack_entry->protection, stack_entry->max_protection,
4667 MAP_STACK_GROWS_DOWN);
4668 if (rv != KERN_SUCCESS) {
4670 rv1 = vm_map_insert(map, NULL, 0, gap_start,
4671 gap_end, VM_PROT_NONE, VM_PROT_NONE,
4672 MAP_CREATE_GUARD | MAP_CREATE_STACK_GAP_DN);
4673 MPASS(rv1 == KERN_SUCCESS);
4675 vm_map_entry_resize(map, gap_entry,
4679 grow_start = stack_entry->end;
4680 cred = stack_entry->cred;
4681 if (cred == NULL && stack_entry->object.vm_object != NULL)
4682 cred = stack_entry->object.vm_object->cred;
4683 if (cred != NULL && !swap_reserve_by_cred(grow_amount, cred))
4685 /* Grow the underlying object if applicable. */
4686 else if (stack_entry->object.vm_object == NULL ||
4687 vm_object_coalesce(stack_entry->object.vm_object,
4688 stack_entry->offset,
4689 (vm_size_t)(stack_entry->end - stack_entry->start),
4690 grow_amount, cred != NULL)) {
4691 if (gap_entry->start + grow_amount == gap_entry->end) {
4692 vm_map_entry_delete(map, gap_entry);
4693 vm_map_entry_resize(map, stack_entry,
4696 gap_entry->start += grow_amount;
4697 stack_entry->end += grow_amount;
4699 map->size += grow_amount;
4704 if (rv == KERN_SUCCESS && is_procstack)
4705 vm->vm_ssize += btoc(grow_amount);
4708 * Heed the MAP_WIREFUTURE flag if it was set for this process.
4710 if (rv == KERN_SUCCESS && (map->flags & MAP_WIREFUTURE) != 0) {
4711 rv = vm_map_wire_locked(map, grow_start,
4712 grow_start + grow_amount,
4713 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
4715 vm_map_lock_downgrade(map);
4719 if (racct_enable && rv != KERN_SUCCESS) {
4721 error = racct_set(p, RACCT_VMEM, map->size);
4722 KASSERT(error == 0, ("decreasing RACCT_VMEM failed"));
4724 error = racct_set(p, RACCT_MEMLOCK,
4725 ptoa(pmap_wired_count(map->pmap)));
4726 KASSERT(error == 0, ("decreasing RACCT_MEMLOCK failed"));
4728 error = racct_set(p, RACCT_STACK, ctob(vm->vm_ssize));
4729 KASSERT(error == 0, ("decreasing RACCT_STACK failed"));
4738 * Unshare the specified VM space for exec. If other processes are
4739 * mapped to it, then create a new one. The new vmspace is null.
4742 vmspace_exec(struct proc *p, vm_offset_t minuser, vm_offset_t maxuser)
4744 struct vmspace *oldvmspace = p->p_vmspace;
4745 struct vmspace *newvmspace;
4747 KASSERT((curthread->td_pflags & TDP_EXECVMSPC) == 0,
4748 ("vmspace_exec recursed"));
4749 newvmspace = vmspace_alloc(minuser, maxuser, pmap_pinit);
4750 if (newvmspace == NULL)
4752 newvmspace->vm_swrss = oldvmspace->vm_swrss;
4754 * This code is written like this for prototype purposes. The
4755 * goal is to avoid running down the vmspace here, but let the
4756 * other process's that are still using the vmspace to finally
4757 * run it down. Even though there is little or no chance of blocking
4758 * here, it is a good idea to keep this form for future mods.
4760 PROC_VMSPACE_LOCK(p);
4761 p->p_vmspace = newvmspace;
4762 PROC_VMSPACE_UNLOCK(p);
4763 if (p == curthread->td_proc)
4764 pmap_activate(curthread);
4765 curthread->td_pflags |= TDP_EXECVMSPC;
4770 * Unshare the specified VM space for forcing COW. This
4771 * is called by rfork, for the (RFMEM|RFPROC) == 0 case.
4774 vmspace_unshare(struct proc *p)
4776 struct vmspace *oldvmspace = p->p_vmspace;
4777 struct vmspace *newvmspace;
4778 vm_ooffset_t fork_charge;
4780 if (oldvmspace->vm_refcnt == 1)
4783 newvmspace = vmspace_fork(oldvmspace, &fork_charge);
4784 if (newvmspace == NULL)
4786 if (!swap_reserve_by_cred(fork_charge, p->p_ucred)) {
4787 vmspace_free(newvmspace);
4790 PROC_VMSPACE_LOCK(p);
4791 p->p_vmspace = newvmspace;
4792 PROC_VMSPACE_UNLOCK(p);
4793 if (p == curthread->td_proc)
4794 pmap_activate(curthread);
4795 vmspace_free(oldvmspace);
4802 * Finds the VM object, offset, and
4803 * protection for a given virtual address in the
4804 * specified map, assuming a page fault of the
4807 * Leaves the map in question locked for read; return
4808 * values are guaranteed until a vm_map_lookup_done
4809 * call is performed. Note that the map argument
4810 * is in/out; the returned map must be used in
4811 * the call to vm_map_lookup_done.
4813 * A handle (out_entry) is returned for use in
4814 * vm_map_lookup_done, to make that fast.
4816 * If a lookup is requested with "write protection"
4817 * specified, the map may be changed to perform virtual
4818 * copying operations, although the data referenced will
4822 vm_map_lookup(vm_map_t *var_map, /* IN/OUT */
4824 vm_prot_t fault_typea,
4825 vm_map_entry_t *out_entry, /* OUT */
4826 vm_object_t *object, /* OUT */
4827 vm_pindex_t *pindex, /* OUT */
4828 vm_prot_t *out_prot, /* OUT */
4829 boolean_t *wired) /* OUT */
4831 vm_map_entry_t entry;
4832 vm_map_t map = *var_map;
4834 vm_prot_t fault_type;
4835 vm_object_t eobject;
4841 vm_map_lock_read(map);
4845 * Lookup the faulting address.
4847 if (!vm_map_lookup_entry(map, vaddr, out_entry)) {
4848 vm_map_unlock_read(map);
4849 return (KERN_INVALID_ADDRESS);
4857 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
4858 vm_map_t old_map = map;
4860 *var_map = map = entry->object.sub_map;
4861 vm_map_unlock_read(old_map);
4866 * Check whether this task is allowed to have this page.
4868 prot = entry->protection;
4869 if ((fault_typea & VM_PROT_FAULT_LOOKUP) != 0) {
4870 fault_typea &= ~VM_PROT_FAULT_LOOKUP;
4871 if (prot == VM_PROT_NONE && map != kernel_map &&
4872 (entry->eflags & MAP_ENTRY_GUARD) != 0 &&
4873 (entry->eflags & (MAP_ENTRY_STACK_GAP_DN |
4874 MAP_ENTRY_STACK_GAP_UP)) != 0 &&
4875 vm_map_growstack(map, vaddr, entry) == KERN_SUCCESS)
4876 goto RetryLookupLocked;
4878 fault_type = fault_typea & VM_PROT_ALL;
4879 if ((fault_type & prot) != fault_type || prot == VM_PROT_NONE) {
4880 vm_map_unlock_read(map);
4881 return (KERN_PROTECTION_FAILURE);
4883 KASSERT((prot & VM_PROT_WRITE) == 0 || (entry->eflags &
4884 (MAP_ENTRY_USER_WIRED | MAP_ENTRY_NEEDS_COPY)) !=
4885 (MAP_ENTRY_USER_WIRED | MAP_ENTRY_NEEDS_COPY),
4886 ("entry %p flags %x", entry, entry->eflags));
4887 if ((fault_typea & VM_PROT_COPY) != 0 &&
4888 (entry->max_protection & VM_PROT_WRITE) == 0 &&
4889 (entry->eflags & MAP_ENTRY_COW) == 0) {
4890 vm_map_unlock_read(map);
4891 return (KERN_PROTECTION_FAILURE);
4895 * If this page is not pageable, we have to get it for all possible
4898 *wired = (entry->wired_count != 0);
4900 fault_type = entry->protection;
4901 size = entry->end - entry->start;
4904 * If the entry was copy-on-write, we either ...
4906 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
4908 * If we want to write the page, we may as well handle that
4909 * now since we've got the map locked.
4911 * If we don't need to write the page, we just demote the
4912 * permissions allowed.
4914 if ((fault_type & VM_PROT_WRITE) != 0 ||
4915 (fault_typea & VM_PROT_COPY) != 0) {
4917 * Make a new object, and place it in the object
4918 * chain. Note that no new references have appeared
4919 * -- one just moved from the map to the new
4922 if (vm_map_lock_upgrade(map))
4925 if (entry->cred == NULL) {
4927 * The debugger owner is charged for
4930 cred = curthread->td_ucred;
4932 if (!swap_reserve_by_cred(size, cred)) {
4935 return (KERN_RESOURCE_SHORTAGE);
4939 eobject = entry->object.vm_object;
4940 vm_object_shadow(&entry->object.vm_object,
4941 &entry->offset, size, entry->cred, false);
4942 if (eobject == entry->object.vm_object) {
4944 * The object was not shadowed.
4946 swap_release_by_cred(size, entry->cred);
4947 crfree(entry->cred);
4950 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
4952 vm_map_lock_downgrade(map);
4955 * We're attempting to read a copy-on-write page --
4956 * don't allow writes.
4958 prot &= ~VM_PROT_WRITE;
4963 * Create an object if necessary.
4965 if (entry->object.vm_object == NULL && !map->system_map) {
4966 if (vm_map_lock_upgrade(map))
4968 entry->object.vm_object = vm_object_allocate_anon(atop(size),
4969 NULL, entry->cred, entry->cred != NULL ? size : 0);
4972 vm_map_lock_downgrade(map);
4976 * Return the object/offset from this entry. If the entry was
4977 * copy-on-write or empty, it has been fixed up.
4979 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
4980 *object = entry->object.vm_object;
4983 return (KERN_SUCCESS);
4987 * vm_map_lookup_locked:
4989 * Lookup the faulting address. A version of vm_map_lookup that returns
4990 * KERN_FAILURE instead of blocking on map lock or memory allocation.
4993 vm_map_lookup_locked(vm_map_t *var_map, /* IN/OUT */
4995 vm_prot_t fault_typea,
4996 vm_map_entry_t *out_entry, /* OUT */
4997 vm_object_t *object, /* OUT */
4998 vm_pindex_t *pindex, /* OUT */
4999 vm_prot_t *out_prot, /* OUT */
5000 boolean_t *wired) /* OUT */
5002 vm_map_entry_t entry;
5003 vm_map_t map = *var_map;
5005 vm_prot_t fault_type = fault_typea;
5008 * Lookup the faulting address.
5010 if (!vm_map_lookup_entry(map, vaddr, out_entry))
5011 return (KERN_INVALID_ADDRESS);
5016 * Fail if the entry refers to a submap.
5018 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP)
5019 return (KERN_FAILURE);
5022 * Check whether this task is allowed to have this page.
5024 prot = entry->protection;
5025 fault_type &= VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE;
5026 if ((fault_type & prot) != fault_type)
5027 return (KERN_PROTECTION_FAILURE);
5030 * If this page is not pageable, we have to get it for all possible
5033 *wired = (entry->wired_count != 0);
5035 fault_type = entry->protection;
5037 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
5039 * Fail if the entry was copy-on-write for a write fault.
5041 if (fault_type & VM_PROT_WRITE)
5042 return (KERN_FAILURE);
5044 * We're attempting to read a copy-on-write page --
5045 * don't allow writes.
5047 prot &= ~VM_PROT_WRITE;
5051 * Fail if an object should be created.
5053 if (entry->object.vm_object == NULL && !map->system_map)
5054 return (KERN_FAILURE);
5057 * Return the object/offset from this entry. If the entry was
5058 * copy-on-write or empty, it has been fixed up.
5060 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
5061 *object = entry->object.vm_object;
5064 return (KERN_SUCCESS);
5068 * vm_map_lookup_done:
5070 * Releases locks acquired by a vm_map_lookup
5071 * (according to the handle returned by that lookup).
5074 vm_map_lookup_done(vm_map_t map, vm_map_entry_t entry)
5077 * Unlock the main-level map
5079 vm_map_unlock_read(map);
5083 vm_map_max_KBI(const struct vm_map *map)
5086 return (vm_map_max(map));
5090 vm_map_min_KBI(const struct vm_map *map)
5093 return (vm_map_min(map));
5097 vm_map_pmap_KBI(vm_map_t map)
5104 vm_map_range_valid_KBI(vm_map_t map, vm_offset_t start, vm_offset_t end)
5107 return (vm_map_range_valid(map, start, end));
5112 _vm_map_assert_consistent(vm_map_t map, int check)
5114 vm_map_entry_t entry, prev;
5115 vm_map_entry_t cur, header, lbound, ubound;
5116 vm_size_t max_left, max_right;
5121 if (enable_vmmap_check != check)
5124 header = prev = &map->header;
5125 VM_MAP_ENTRY_FOREACH(entry, map) {
5126 KASSERT(prev->end <= entry->start,
5127 ("map %p prev->end = %jx, start = %jx", map,
5128 (uintmax_t)prev->end, (uintmax_t)entry->start));
5129 KASSERT(entry->start < entry->end,
5130 ("map %p start = %jx, end = %jx", map,
5131 (uintmax_t)entry->start, (uintmax_t)entry->end));
5132 KASSERT(entry->left == header ||
5133 entry->left->start < entry->start,
5134 ("map %p left->start = %jx, start = %jx", map,
5135 (uintmax_t)entry->left->start, (uintmax_t)entry->start));
5136 KASSERT(entry->right == header ||
5137 entry->start < entry->right->start,
5138 ("map %p start = %jx, right->start = %jx", map,
5139 (uintmax_t)entry->start, (uintmax_t)entry->right->start));
5141 lbound = ubound = header;
5143 if (entry->start < cur->start) {
5146 KASSERT(cur != lbound,
5147 ("map %p cannot find %jx",
5148 map, (uintmax_t)entry->start));
5149 } else if (cur->end <= entry->start) {
5152 KASSERT(cur != ubound,
5153 ("map %p cannot find %jx",
5154 map, (uintmax_t)entry->start));
5156 KASSERT(cur == entry,
5157 ("map %p cannot find %jx",
5158 map, (uintmax_t)entry->start));
5162 max_left = vm_map_entry_max_free_left(entry, lbound);
5163 max_right = vm_map_entry_max_free_right(entry, ubound);
5164 KASSERT(entry->max_free == vm_size_max(max_left, max_right),
5165 ("map %p max = %jx, max_left = %jx, max_right = %jx", map,
5166 (uintmax_t)entry->max_free,
5167 (uintmax_t)max_left, (uintmax_t)max_right));
5170 KASSERT(prev->end <= entry->start,
5171 ("map %p prev->end = %jx, start = %jx", map,
5172 (uintmax_t)prev->end, (uintmax_t)entry->start));
5176 #include "opt_ddb.h"
5178 #include <sys/kernel.h>
5180 #include <ddb/ddb.h>
5183 vm_map_print(vm_map_t map)
5185 vm_map_entry_t entry, prev;
5187 db_iprintf("Task map %p: pmap=%p, nentries=%d, version=%u\n",
5189 (void *)map->pmap, map->nentries, map->timestamp);
5192 prev = &map->header;
5193 VM_MAP_ENTRY_FOREACH(entry, map) {
5194 db_iprintf("map entry %p: start=%p, end=%p, eflags=%#x, \n",
5195 (void *)entry, (void *)entry->start, (void *)entry->end,
5198 static const char * const inheritance_name[4] =
5199 {"share", "copy", "none", "donate_copy"};
5201 db_iprintf(" prot=%x/%x/%s",
5203 entry->max_protection,
5204 inheritance_name[(int)(unsigned char)
5205 entry->inheritance]);
5206 if (entry->wired_count != 0)
5207 db_printf(", wired");
5209 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
5210 db_printf(", share=%p, offset=0x%jx\n",
5211 (void *)entry->object.sub_map,
5212 (uintmax_t)entry->offset);
5213 if (prev == &map->header ||
5214 prev->object.sub_map !=
5215 entry->object.sub_map) {
5217 vm_map_print((vm_map_t)entry->object.sub_map);
5221 if (entry->cred != NULL)
5222 db_printf(", ruid %d", entry->cred->cr_ruid);
5223 db_printf(", object=%p, offset=0x%jx",
5224 (void *)entry->object.vm_object,
5225 (uintmax_t)entry->offset);
5226 if (entry->object.vm_object && entry->object.vm_object->cred)
5227 db_printf(", obj ruid %d charge %jx",
5228 entry->object.vm_object->cred->cr_ruid,
5229 (uintmax_t)entry->object.vm_object->charge);
5230 if (entry->eflags & MAP_ENTRY_COW)
5231 db_printf(", copy (%s)",
5232 (entry->eflags & MAP_ENTRY_NEEDS_COPY) ? "needed" : "done");
5235 if (prev == &map->header ||
5236 prev->object.vm_object !=
5237 entry->object.vm_object) {
5239 vm_object_print((db_expr_t)(intptr_t)
5240 entry->object.vm_object,
5250 DB_SHOW_COMMAND(map, map)
5254 db_printf("usage: show map <addr>\n");
5257 vm_map_print((vm_map_t)addr);
5260 DB_SHOW_COMMAND(procvm, procvm)
5265 p = db_lookup_proc(addr);
5270 db_printf("p = %p, vmspace = %p, map = %p, pmap = %p\n",
5271 (void *)p, (void *)p->p_vmspace, (void *)&p->p_vmspace->vm_map,
5272 (void *)vmspace_pmap(p->p_vmspace));
5274 vm_map_print((vm_map_t)&p->p_vmspace->vm_map);