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_object.c 8.5 (Berkeley) 3/22/94
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 object module.
67 #include <sys/cdefs.h>
68 __FBSDID("$FreeBSD$");
72 #include <sys/param.h>
73 #include <sys/systm.h>
74 #include <sys/blockcount.h>
75 #include <sys/cpuset.h>
78 #include <sys/mount.h>
79 #include <sys/kernel.h>
80 #include <sys/pctrie.h>
81 #include <sys/sysctl.h>
82 #include <sys/mutex.h>
83 #include <sys/proc.h> /* for curproc, pageproc */
84 #include <sys/refcount.h>
85 #include <sys/socket.h>
86 #include <sys/resourcevar.h>
87 #include <sys/refcount.h>
88 #include <sys/rwlock.h>
90 #include <sys/vnode.h>
91 #include <sys/vmmeter.h>
95 #include <vm/vm_param.h>
97 #include <vm/vm_map.h>
98 #include <vm/vm_object.h>
99 #include <vm/vm_page.h>
100 #include <vm/vm_pageout.h>
101 #include <vm/vm_pager.h>
102 #include <vm/vm_phys.h>
103 #include <vm/vm_pagequeue.h>
104 #include <vm/swap_pager.h>
105 #include <vm/vm_kern.h>
106 #include <vm/vm_extern.h>
107 #include <vm/vm_radix.h>
108 #include <vm/vm_reserv.h>
111 static int old_msync;
112 SYSCTL_INT(_vm, OID_AUTO, old_msync, CTLFLAG_RW, &old_msync, 0,
113 "Use old (insecure) msync behavior");
115 static int vm_object_page_collect_flush(vm_object_t object, vm_page_t p,
116 int pagerflags, int flags, boolean_t *allclean,
118 static boolean_t vm_object_page_remove_write(vm_page_t p, int flags,
119 boolean_t *allclean);
120 static void vm_object_backing_remove(vm_object_t object);
123 * Virtual memory objects maintain the actual data
124 * associated with allocated virtual memory. A given
125 * page of memory exists within exactly one object.
127 * An object is only deallocated when all "references"
128 * are given up. Only one "reference" to a given
129 * region of an object should be writeable.
131 * Associated with each object is a list of all resident
132 * memory pages belonging to that object; this list is
133 * maintained by the "vm_page" module, and locked by the object's
136 * Each object also records a "pager" routine which is
137 * used to retrieve (and store) pages to the proper backing
138 * storage. In addition, objects may be backed by other
139 * objects from which they were virtual-copied.
141 * The only items within the object structure which are
142 * modified after time of creation are:
143 * reference count locked by object's lock
144 * pager routine locked by object's lock
148 struct object_q vm_object_list;
149 struct mtx vm_object_list_mtx; /* lock for object list and count */
151 struct vm_object kernel_object_store;
153 static SYSCTL_NODE(_vm_stats, OID_AUTO, object, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
156 static COUNTER_U64_DEFINE_EARLY(object_collapses);
157 SYSCTL_COUNTER_U64(_vm_stats_object, OID_AUTO, collapses, CTLFLAG_RD,
159 "VM object collapses");
161 static COUNTER_U64_DEFINE_EARLY(object_bypasses);
162 SYSCTL_COUNTER_U64(_vm_stats_object, OID_AUTO, bypasses, CTLFLAG_RD,
164 "VM object bypasses");
166 static COUNTER_U64_DEFINE_EARLY(object_collapse_waits);
167 SYSCTL_COUNTER_U64(_vm_stats_object, OID_AUTO, collapse_waits, CTLFLAG_RD,
168 &object_collapse_waits,
169 "Number of sleeps for collapse");
171 static uma_zone_t obj_zone;
173 static int vm_object_zinit(void *mem, int size, int flags);
176 static void vm_object_zdtor(void *mem, int size, void *arg);
179 vm_object_zdtor(void *mem, int size, void *arg)
183 object = (vm_object_t)mem;
184 KASSERT(object->ref_count == 0,
185 ("object %p ref_count = %d", object, object->ref_count));
186 KASSERT(TAILQ_EMPTY(&object->memq),
187 ("object %p has resident pages in its memq", object));
188 KASSERT(vm_radix_is_empty(&object->rtree),
189 ("object %p has resident pages in its trie", object));
190 #if VM_NRESERVLEVEL > 0
191 KASSERT(LIST_EMPTY(&object->rvq),
192 ("object %p has reservations",
195 KASSERT(!vm_object_busied(object),
196 ("object %p busy = %d", object, blockcount_read(&object->busy)));
197 KASSERT(object->resident_page_count == 0,
198 ("object %p resident_page_count = %d",
199 object, object->resident_page_count));
200 KASSERT(object->shadow_count == 0,
201 ("object %p shadow_count = %d",
202 object, object->shadow_count));
203 KASSERT(object->type == OBJT_DEAD,
204 ("object %p has non-dead type %d",
205 object, object->type));
210 vm_object_zinit(void *mem, int size, int flags)
214 object = (vm_object_t)mem;
215 rw_init_flags(&object->lock, "vm object", RW_DUPOK | RW_NEW);
217 /* These are true for any object that has been freed */
218 object->type = OBJT_DEAD;
219 vm_radix_init(&object->rtree);
220 refcount_init(&object->ref_count, 0);
221 blockcount_init(&object->paging_in_progress);
222 blockcount_init(&object->busy);
223 object->resident_page_count = 0;
224 object->shadow_count = 0;
225 object->flags = OBJ_DEAD;
227 mtx_lock(&vm_object_list_mtx);
228 TAILQ_INSERT_TAIL(&vm_object_list, object, object_list);
229 mtx_unlock(&vm_object_list_mtx);
234 _vm_object_allocate(objtype_t type, vm_pindex_t size, u_short flags,
235 vm_object_t object, void *handle)
238 TAILQ_INIT(&object->memq);
239 LIST_INIT(&object->shadow_head);
242 if (type == OBJT_SWAP)
243 pctrie_init(&object->un_pager.swp.swp_blks);
246 * Ensure that swap_pager_swapoff() iteration over object_list
247 * sees up to date type and pctrie head if it observed
250 atomic_thread_fence_rel();
252 object->pg_color = 0;
253 object->flags = flags;
255 object->domain.dr_policy = NULL;
256 object->generation = 1;
257 object->cleangeneration = 1;
258 refcount_init(&object->ref_count, 1);
259 object->memattr = VM_MEMATTR_DEFAULT;
262 object->handle = handle;
263 object->backing_object = NULL;
264 object->backing_object_offset = (vm_ooffset_t) 0;
265 #if VM_NRESERVLEVEL > 0
266 LIST_INIT(&object->rvq);
268 umtx_shm_object_init(object);
274 * Initialize the VM objects module.
279 TAILQ_INIT(&vm_object_list);
280 mtx_init(&vm_object_list_mtx, "vm object_list", NULL, MTX_DEF);
282 rw_init(&kernel_object->lock, "kernel vm object");
283 _vm_object_allocate(OBJT_PHYS, atop(VM_MAX_KERNEL_ADDRESS -
284 VM_MIN_KERNEL_ADDRESS), OBJ_UNMANAGED, kernel_object, NULL);
285 #if VM_NRESERVLEVEL > 0
286 kernel_object->flags |= OBJ_COLORED;
287 kernel_object->pg_color = (u_short)atop(VM_MIN_KERNEL_ADDRESS);
291 * The lock portion of struct vm_object must be type stable due
292 * to vm_pageout_fallback_object_lock locking a vm object
293 * without holding any references to it.
295 * paging_in_progress is valid always. Lockless references to
296 * the objects may acquire pip and then check OBJ_DEAD.
298 obj_zone = uma_zcreate("VM OBJECT", sizeof (struct vm_object), NULL,
304 vm_object_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
310 vm_object_clear_flag(vm_object_t object, u_short bits)
313 VM_OBJECT_ASSERT_WLOCKED(object);
314 object->flags &= ~bits;
318 * Sets the default memory attribute for the specified object. Pages
319 * that are allocated to this object are by default assigned this memory
322 * Presently, this function must be called before any pages are allocated
323 * to the object. In the future, this requirement may be relaxed for
324 * "default" and "swap" objects.
327 vm_object_set_memattr(vm_object_t object, vm_memattr_t memattr)
330 VM_OBJECT_ASSERT_WLOCKED(object);
331 switch (object->type) {
339 if (!TAILQ_EMPTY(&object->memq))
340 return (KERN_FAILURE);
343 return (KERN_INVALID_ARGUMENT);
345 panic("vm_object_set_memattr: object %p is of undefined type",
348 object->memattr = memattr;
349 return (KERN_SUCCESS);
353 vm_object_pip_add(vm_object_t object, short i)
357 blockcount_acquire(&object->paging_in_progress, i);
361 vm_object_pip_wakeup(vm_object_t object)
364 vm_object_pip_wakeupn(object, 1);
368 vm_object_pip_wakeupn(vm_object_t object, short i)
372 blockcount_release(&object->paging_in_progress, i);
376 * Atomically drop the object lock and wait for pip to drain. This protects
377 * from sleep/wakeup races due to identity changes. The lock is not re-acquired
381 vm_object_pip_sleep(vm_object_t object, const char *waitid)
384 (void)blockcount_sleep(&object->paging_in_progress, &object->lock,
385 waitid, PVM | PDROP);
389 vm_object_pip_wait(vm_object_t object, const char *waitid)
392 VM_OBJECT_ASSERT_WLOCKED(object);
394 blockcount_wait(&object->paging_in_progress, &object->lock, waitid,
399 vm_object_pip_wait_unlocked(vm_object_t object, const char *waitid)
402 VM_OBJECT_ASSERT_UNLOCKED(object);
404 blockcount_wait(&object->paging_in_progress, NULL, waitid, PVM);
408 * vm_object_allocate:
410 * Returns a new object with the given size.
413 vm_object_allocate(objtype_t type, vm_pindex_t size)
420 panic("vm_object_allocate: can't create OBJT_DEAD");
427 flags = OBJ_FICTITIOUS | OBJ_UNMANAGED;
430 flags = OBJ_FICTITIOUS;
433 flags = OBJ_UNMANAGED;
439 panic("vm_object_allocate: type %d is undefined", type);
441 object = (vm_object_t)uma_zalloc(obj_zone, M_WAITOK);
442 _vm_object_allocate(type, size, flags, object, NULL);
448 * vm_object_allocate_anon:
450 * Returns a new default object of the given size and marked as
451 * anonymous memory for special split/collapse handling. Color
452 * to be initialized by the caller.
455 vm_object_allocate_anon(vm_pindex_t size, vm_object_t backing_object,
456 struct ucred *cred, vm_size_t charge)
458 vm_object_t handle, object;
460 if (backing_object == NULL)
462 else if ((backing_object->flags & OBJ_ANON) != 0)
463 handle = backing_object->handle;
465 handle = backing_object;
466 object = uma_zalloc(obj_zone, M_WAITOK);
467 _vm_object_allocate(OBJT_DEFAULT, size, OBJ_ANON | OBJ_ONEMAPPING,
470 object->charge = cred != NULL ? charge : 0;
475 vm_object_reference_vnode(vm_object_t object)
480 * vnode objects need the lock for the first reference
481 * to serialize with vnode_object_deallocate().
483 if (!refcount_acquire_if_gt(&object->ref_count, 0)) {
484 VM_OBJECT_RLOCK(object);
485 old = refcount_acquire(&object->ref_count);
486 if (object->type == OBJT_VNODE && old == 0)
487 vref(object->handle);
488 VM_OBJECT_RUNLOCK(object);
493 * vm_object_reference:
495 * Acquires a reference to the given object.
498 vm_object_reference(vm_object_t object)
504 if (object->type == OBJT_VNODE)
505 vm_object_reference_vnode(object);
507 refcount_acquire(&object->ref_count);
508 KASSERT((object->flags & OBJ_DEAD) == 0,
509 ("vm_object_reference: Referenced dead object."));
513 * vm_object_reference_locked:
515 * Gets another reference to the given object.
517 * The object must be locked.
520 vm_object_reference_locked(vm_object_t object)
524 VM_OBJECT_ASSERT_LOCKED(object);
525 old = refcount_acquire(&object->ref_count);
526 if (object->type == OBJT_VNODE && old == 0)
527 vref(object->handle);
528 KASSERT((object->flags & OBJ_DEAD) == 0,
529 ("vm_object_reference: Referenced dead object."));
533 * Handle deallocating an object of type OBJT_VNODE.
536 vm_object_deallocate_vnode(vm_object_t object)
538 struct vnode *vp = (struct vnode *) object->handle;
541 KASSERT(object->type == OBJT_VNODE,
542 ("vm_object_deallocate_vnode: not a vnode object"));
543 KASSERT(vp != NULL, ("vm_object_deallocate_vnode: missing vp"));
545 /* Object lock to protect handle lookup. */
546 last = refcount_release(&object->ref_count);
547 VM_OBJECT_RUNLOCK(object);
552 if (!umtx_shm_vnobj_persistent)
553 umtx_shm_object_terminated(object);
555 /* vrele may need the vnode lock. */
561 * We dropped a reference on an object and discovered that it had a
562 * single remaining shadow. This is a sibling of the reference we
563 * dropped. Attempt to collapse the sibling and backing object.
566 vm_object_deallocate_anon(vm_object_t backing_object)
570 /* Fetch the final shadow. */
571 object = LIST_FIRST(&backing_object->shadow_head);
572 KASSERT(object != NULL && backing_object->shadow_count == 1,
573 ("vm_object_anon_deallocate: ref_count: %d, shadow_count: %d",
574 backing_object->ref_count, backing_object->shadow_count));
575 KASSERT((object->flags & (OBJ_TMPFS_NODE | OBJ_ANON)) == OBJ_ANON,
576 ("invalid shadow object %p", object));
578 if (!VM_OBJECT_TRYWLOCK(object)) {
580 * Prevent object from disappearing since we do not have a
583 vm_object_pip_add(object, 1);
584 VM_OBJECT_WUNLOCK(backing_object);
585 VM_OBJECT_WLOCK(object);
586 vm_object_pip_wakeup(object);
588 VM_OBJECT_WUNLOCK(backing_object);
591 * Check for a collapse/terminate race with the last reference holder.
593 if ((object->flags & (OBJ_DEAD | OBJ_COLLAPSING)) != 0 ||
594 !refcount_acquire_if_not_zero(&object->ref_count)) {
595 VM_OBJECT_WUNLOCK(object);
598 backing_object = object->backing_object;
599 if (backing_object != NULL && (backing_object->flags & OBJ_ANON) != 0)
600 vm_object_collapse(object);
601 VM_OBJECT_WUNLOCK(object);
607 * vm_object_deallocate:
609 * Release a reference to the specified object,
610 * gained either through a vm_object_allocate
611 * or a vm_object_reference call. When all references
612 * are gone, storage associated with this object
613 * may be relinquished.
615 * No object may be locked.
618 vm_object_deallocate(vm_object_t object)
623 while (object != NULL) {
625 * If the reference count goes to 0 we start calling
626 * vm_object_terminate() on the object chain. A ref count
627 * of 1 may be a special case depending on the shadow count
628 * being 0 or 1. These cases require a write lock on the
631 if ((object->flags & OBJ_ANON) == 0)
632 released = refcount_release_if_gt(&object->ref_count, 1);
634 released = refcount_release_if_gt(&object->ref_count, 2);
638 if (object->type == OBJT_VNODE) {
639 VM_OBJECT_RLOCK(object);
640 if (object->type == OBJT_VNODE) {
641 vm_object_deallocate_vnode(object);
644 VM_OBJECT_RUNLOCK(object);
647 VM_OBJECT_WLOCK(object);
648 KASSERT(object->ref_count > 0,
649 ("vm_object_deallocate: object deallocated too many times: %d",
653 * If this is not the final reference to an anonymous
654 * object we may need to collapse the shadow chain.
656 if (!refcount_release(&object->ref_count)) {
657 if (object->ref_count > 1 ||
658 object->shadow_count == 0) {
659 if ((object->flags & OBJ_ANON) != 0 &&
660 object->ref_count == 1)
661 vm_object_set_flag(object,
663 VM_OBJECT_WUNLOCK(object);
667 /* Handle collapsing last ref on anonymous objects. */
668 object = vm_object_deallocate_anon(object);
673 * Handle the final reference to an object. We restart
674 * the loop with the backing object to avoid recursion.
676 umtx_shm_object_terminated(object);
677 temp = object->backing_object;
679 KASSERT((object->flags & OBJ_TMPFS_NODE) == 0,
680 ("shadowed tmpfs v_object 2 %p", object));
681 vm_object_backing_remove(object);
684 KASSERT((object->flags & OBJ_DEAD) == 0,
685 ("vm_object_deallocate: Terminating dead object."));
686 vm_object_set_flag(object, OBJ_DEAD);
687 vm_object_terminate(object);
693 * vm_object_destroy removes the object from the global object list
694 * and frees the space for the object.
697 vm_object_destroy(vm_object_t object)
701 * Release the allocation charge.
703 if (object->cred != NULL) {
704 swap_release_by_cred(object->charge, object->cred);
706 crfree(object->cred);
711 * Free the space for the object.
713 uma_zfree(obj_zone, object);
717 vm_object_backing_remove_locked(vm_object_t object)
719 vm_object_t backing_object;
721 backing_object = object->backing_object;
722 VM_OBJECT_ASSERT_WLOCKED(object);
723 VM_OBJECT_ASSERT_WLOCKED(backing_object);
725 KASSERT((object->flags & OBJ_COLLAPSING) == 0,
726 ("vm_object_backing_remove: Removing collapsing object."));
728 if ((object->flags & OBJ_SHADOWLIST) != 0) {
729 LIST_REMOVE(object, shadow_list);
730 backing_object->shadow_count--;
731 object->flags &= ~OBJ_SHADOWLIST;
733 object->backing_object = NULL;
737 vm_object_backing_remove(vm_object_t object)
739 vm_object_t backing_object;
741 VM_OBJECT_ASSERT_WLOCKED(object);
743 if ((object->flags & OBJ_SHADOWLIST) != 0) {
744 backing_object = object->backing_object;
745 VM_OBJECT_WLOCK(backing_object);
746 vm_object_backing_remove_locked(object);
747 VM_OBJECT_WUNLOCK(backing_object);
749 object->backing_object = NULL;
753 vm_object_backing_insert_locked(vm_object_t object, vm_object_t backing_object)
756 VM_OBJECT_ASSERT_WLOCKED(object);
758 if ((backing_object->flags & OBJ_ANON) != 0) {
759 VM_OBJECT_ASSERT_WLOCKED(backing_object);
760 LIST_INSERT_HEAD(&backing_object->shadow_head, object,
762 backing_object->shadow_count++;
763 object->flags |= OBJ_SHADOWLIST;
765 object->backing_object = backing_object;
769 vm_object_backing_insert(vm_object_t object, vm_object_t backing_object)
772 VM_OBJECT_ASSERT_WLOCKED(object);
774 if ((backing_object->flags & OBJ_ANON) != 0) {
775 VM_OBJECT_WLOCK(backing_object);
776 vm_object_backing_insert_locked(object, backing_object);
777 VM_OBJECT_WUNLOCK(backing_object);
779 object->backing_object = backing_object;
783 * Insert an object into a backing_object's shadow list with an additional
784 * reference to the backing_object added.
787 vm_object_backing_insert_ref(vm_object_t object, vm_object_t backing_object)
790 VM_OBJECT_ASSERT_WLOCKED(object);
792 if ((backing_object->flags & OBJ_ANON) != 0) {
793 VM_OBJECT_WLOCK(backing_object);
794 KASSERT((backing_object->flags & OBJ_DEAD) == 0,
795 ("shadowing dead anonymous object"));
796 vm_object_reference_locked(backing_object);
797 vm_object_backing_insert_locked(object, backing_object);
798 vm_object_clear_flag(backing_object, OBJ_ONEMAPPING);
799 VM_OBJECT_WUNLOCK(backing_object);
801 vm_object_reference(backing_object);
802 object->backing_object = backing_object;
807 * Transfer a backing reference from backing_object to object.
810 vm_object_backing_transfer(vm_object_t object, vm_object_t backing_object)
812 vm_object_t new_backing_object;
815 * Note that the reference to backing_object->backing_object
816 * moves from within backing_object to within object.
818 vm_object_backing_remove_locked(object);
819 new_backing_object = backing_object->backing_object;
820 if (new_backing_object == NULL)
822 if ((new_backing_object->flags & OBJ_ANON) != 0) {
823 VM_OBJECT_WLOCK(new_backing_object);
824 vm_object_backing_remove_locked(backing_object);
825 vm_object_backing_insert_locked(object, new_backing_object);
826 VM_OBJECT_WUNLOCK(new_backing_object);
828 object->backing_object = new_backing_object;
829 backing_object->backing_object = NULL;
834 * Wait for a concurrent collapse to settle.
837 vm_object_collapse_wait(vm_object_t object)
840 VM_OBJECT_ASSERT_WLOCKED(object);
842 while ((object->flags & OBJ_COLLAPSING) != 0) {
843 vm_object_pip_wait(object, "vmcolwait");
844 counter_u64_add(object_collapse_waits, 1);
849 * Waits for a backing object to clear a pending collapse and returns
850 * it locked if it is an ANON object.
853 vm_object_backing_collapse_wait(vm_object_t object)
855 vm_object_t backing_object;
857 VM_OBJECT_ASSERT_WLOCKED(object);
860 backing_object = object->backing_object;
861 if (backing_object == NULL ||
862 (backing_object->flags & OBJ_ANON) == 0)
864 VM_OBJECT_WLOCK(backing_object);
865 if ((backing_object->flags & (OBJ_DEAD | OBJ_COLLAPSING)) == 0)
867 VM_OBJECT_WUNLOCK(object);
868 vm_object_pip_sleep(backing_object, "vmbckwait");
869 counter_u64_add(object_collapse_waits, 1);
870 VM_OBJECT_WLOCK(object);
872 return (backing_object);
876 * vm_object_terminate_pages removes any remaining pageable pages
877 * from the object and resets the object to an empty state.
880 vm_object_terminate_pages(vm_object_t object)
884 VM_OBJECT_ASSERT_WLOCKED(object);
887 * Free any remaining pageable pages. This also removes them from the
888 * paging queues. However, don't free wired pages, just remove them
889 * from the object. Rather than incrementally removing each page from
890 * the object, the page and object are reset to any empty state.
892 TAILQ_FOREACH_SAFE(p, &object->memq, listq, p_next) {
893 vm_page_assert_unbusied(p);
894 KASSERT(p->object == object &&
895 (p->ref_count & VPRC_OBJREF) != 0,
896 ("vm_object_terminate_pages: page %p is inconsistent", p));
899 if (vm_page_drop(p, VPRC_OBJREF) == VPRC_OBJREF) {
906 * If the object contained any pages, then reset it to an empty state.
907 * None of the object's fields, including "resident_page_count", were
908 * modified by the preceding loop.
910 if (object->resident_page_count != 0) {
911 vm_radix_reclaim_allnodes(&object->rtree);
912 TAILQ_INIT(&object->memq);
913 object->resident_page_count = 0;
914 if (object->type == OBJT_VNODE)
915 vdrop(object->handle);
920 * vm_object_terminate actually destroys the specified object, freeing
921 * up all previously used resources.
923 * The object must be locked.
924 * This routine may block.
927 vm_object_terminate(vm_object_t object)
930 VM_OBJECT_ASSERT_WLOCKED(object);
931 KASSERT((object->flags & OBJ_DEAD) != 0,
932 ("terminating non-dead obj %p", object));
933 KASSERT((object->flags & OBJ_COLLAPSING) == 0,
934 ("terminating collapsing obj %p", object));
935 KASSERT(object->backing_object == NULL,
936 ("terminating shadow obj %p", object));
939 * Wait for the pageout daemon and other current users to be
940 * done with the object. Note that new paging_in_progress
941 * users can come after this wait, but they must check
942 * OBJ_DEAD flag set (without unlocking the object), and avoid
943 * the object being terminated.
945 vm_object_pip_wait(object, "objtrm");
947 KASSERT(object->ref_count == 0,
948 ("vm_object_terminate: object with references, ref_count=%d",
951 if ((object->flags & OBJ_PG_DTOR) == 0)
952 vm_object_terminate_pages(object);
954 #if VM_NRESERVLEVEL > 0
955 if (__predict_false(!LIST_EMPTY(&object->rvq)))
956 vm_reserv_break_all(object);
959 KASSERT(object->cred == NULL || object->type == OBJT_DEFAULT ||
960 object->type == OBJT_SWAP,
961 ("%s: non-swap obj %p has cred", __func__, object));
964 * Let the pager know object is dead.
966 vm_pager_deallocate(object);
967 VM_OBJECT_WUNLOCK(object);
969 vm_object_destroy(object);
973 * Make the page read-only so that we can clear the object flags. However, if
974 * this is a nosync mmap then the object is likely to stay dirty so do not
975 * mess with the page and do not clear the object flags. Returns TRUE if the
976 * page should be flushed, and FALSE otherwise.
979 vm_object_page_remove_write(vm_page_t p, int flags, boolean_t *allclean)
982 vm_page_assert_busied(p);
985 * If we have been asked to skip nosync pages and this is a
986 * nosync page, skip it. Note that the object flags were not
987 * cleared in this case so we do not have to set them.
989 if ((flags & OBJPC_NOSYNC) != 0 && (p->a.flags & PGA_NOSYNC) != 0) {
993 pmap_remove_write(p);
994 return (p->dirty != 0);
999 * vm_object_page_clean
1001 * Clean all dirty pages in the specified range of object. Leaves page
1002 * on whatever queue it is currently on. If NOSYNC is set then do not
1003 * write out pages with PGA_NOSYNC set (originally comes from MAP_NOSYNC),
1004 * leaving the object dirty.
1006 * For swap objects backing tmpfs regular files, do not flush anything,
1007 * but remove write protection on the mapped pages to update mtime through
1010 * When stuffing pages asynchronously, allow clustering. XXX we need a
1011 * synchronous clustering mode implementation.
1013 * Odd semantics: if start == end, we clean everything.
1015 * The object must be locked.
1017 * Returns FALSE if some page from the range was not written, as
1018 * reported by the pager, and TRUE otherwise.
1021 vm_object_page_clean(vm_object_t object, vm_ooffset_t start, vm_ooffset_t end,
1025 vm_pindex_t pi, tend, tstart;
1026 int curgeneration, n, pagerflags;
1027 boolean_t eio, res, allclean;
1029 VM_OBJECT_ASSERT_WLOCKED(object);
1031 if (!vm_object_mightbedirty(object) || object->resident_page_count == 0)
1034 pagerflags = (flags & (OBJPC_SYNC | OBJPC_INVAL)) != 0 ?
1035 VM_PAGER_PUT_SYNC : VM_PAGER_CLUSTER_OK;
1036 pagerflags |= (flags & OBJPC_INVAL) != 0 ? VM_PAGER_PUT_INVAL : 0;
1038 tstart = OFF_TO_IDX(start);
1039 tend = (end == 0) ? object->size : OFF_TO_IDX(end + PAGE_MASK);
1040 allclean = tstart == 0 && tend >= object->size;
1044 curgeneration = object->generation;
1046 for (p = vm_page_find_least(object, tstart); p != NULL; p = np) {
1050 np = TAILQ_NEXT(p, listq);
1051 if (vm_page_none_valid(p))
1053 if (vm_page_busy_acquire(p, VM_ALLOC_WAITFAIL) == 0) {
1054 if (object->generation != curgeneration &&
1055 (flags & OBJPC_SYNC) != 0)
1057 np = vm_page_find_least(object, pi);
1060 if (!vm_object_page_remove_write(p, flags, &allclean)) {
1064 if (object->type == OBJT_VNODE) {
1065 n = vm_object_page_collect_flush(object, p, pagerflags,
1066 flags, &allclean, &eio);
1071 if (object->generation != curgeneration &&
1072 (flags & OBJPC_SYNC) != 0)
1076 * If the VOP_PUTPAGES() did a truncated write, so
1077 * that even the first page of the run is not fully
1078 * written, vm_pageout_flush() returns 0 as the run
1079 * length. Since the condition that caused truncated
1080 * write may be permanent, e.g. exhausted free space,
1081 * accepting n == 0 would cause an infinite loop.
1083 * Forwarding the iterator leaves the unwritten page
1084 * behind, but there is not much we can do there if
1085 * filesystem refuses to write it.
1095 np = vm_page_find_least(object, pi + n);
1098 VOP_FSYNC(vp, (pagerflags & VM_PAGER_PUT_SYNC) ? MNT_WAIT : 0);
1102 * Leave updating cleangeneration for tmpfs objects to tmpfs
1103 * scan. It needs to update mtime, which happens for other
1104 * filesystems during page writeouts.
1106 if (allclean && object->type == OBJT_VNODE)
1107 object->cleangeneration = curgeneration;
1112 vm_object_page_collect_flush(vm_object_t object, vm_page_t p, int pagerflags,
1113 int flags, boolean_t *allclean, boolean_t *eio)
1115 vm_page_t ma[vm_pageout_page_count], p_first, tp;
1116 int count, i, mreq, runlen;
1118 vm_page_lock_assert(p, MA_NOTOWNED);
1119 vm_page_assert_xbusied(p);
1120 VM_OBJECT_ASSERT_WLOCKED(object);
1125 for (tp = p; count < vm_pageout_page_count; count++) {
1126 tp = vm_page_next(tp);
1127 if (tp == NULL || vm_page_tryxbusy(tp) == 0)
1129 if (!vm_object_page_remove_write(tp, flags, allclean)) {
1130 vm_page_xunbusy(tp);
1135 for (p_first = p; count < vm_pageout_page_count; count++) {
1136 tp = vm_page_prev(p_first);
1137 if (tp == NULL || vm_page_tryxbusy(tp) == 0)
1139 if (!vm_object_page_remove_write(tp, flags, allclean)) {
1140 vm_page_xunbusy(tp);
1147 for (tp = p_first, i = 0; i < count; tp = TAILQ_NEXT(tp, listq), i++)
1150 vm_pageout_flush(ma, count, pagerflags, mreq, &runlen, eio);
1155 * Note that there is absolutely no sense in writing out
1156 * anonymous objects, so we track down the vnode object
1158 * We invalidate (remove) all pages from the address space
1159 * for semantic correctness.
1161 * If the backing object is a device object with unmanaged pages, then any
1162 * mappings to the specified range of pages must be removed before this
1163 * function is called.
1165 * Note: certain anonymous maps, such as MAP_NOSYNC maps,
1166 * may start out with a NULL object.
1169 vm_object_sync(vm_object_t object, vm_ooffset_t offset, vm_size_t size,
1170 boolean_t syncio, boolean_t invalidate)
1172 vm_object_t backing_object;
1175 int error, flags, fsync_after;
1182 VM_OBJECT_WLOCK(object);
1183 while ((backing_object = object->backing_object) != NULL) {
1184 VM_OBJECT_WLOCK(backing_object);
1185 offset += object->backing_object_offset;
1186 VM_OBJECT_WUNLOCK(object);
1187 object = backing_object;
1188 if (object->size < OFF_TO_IDX(offset + size))
1189 size = IDX_TO_OFF(object->size) - offset;
1192 * Flush pages if writing is allowed, invalidate them
1193 * if invalidation requested. Pages undergoing I/O
1194 * will be ignored by vm_object_page_remove().
1196 * We cannot lock the vnode and then wait for paging
1197 * to complete without deadlocking against vm_fault.
1198 * Instead we simply call vm_object_page_remove() and
1199 * allow it to block internally on a page-by-page
1200 * basis when it encounters pages undergoing async
1203 if (object->type == OBJT_VNODE &&
1204 vm_object_mightbedirty(object) != 0 &&
1205 ((vp = object->handle)->v_vflag & VV_NOSYNC) == 0) {
1206 VM_OBJECT_WUNLOCK(object);
1207 (void) vn_start_write(vp, &mp, V_WAIT);
1208 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1209 if (syncio && !invalidate && offset == 0 &&
1210 atop(size) == object->size) {
1212 * If syncing the whole mapping of the file,
1213 * it is faster to schedule all the writes in
1214 * async mode, also allowing the clustering,
1215 * and then wait for i/o to complete.
1220 flags = (syncio || invalidate) ? OBJPC_SYNC : 0;
1221 flags |= invalidate ? (OBJPC_SYNC | OBJPC_INVAL) : 0;
1222 fsync_after = FALSE;
1224 VM_OBJECT_WLOCK(object);
1225 res = vm_object_page_clean(object, offset, offset + size,
1227 VM_OBJECT_WUNLOCK(object);
1229 error = VOP_FSYNC(vp, MNT_WAIT, curthread);
1231 vn_finished_write(mp);
1234 VM_OBJECT_WLOCK(object);
1236 if ((object->type == OBJT_VNODE ||
1237 object->type == OBJT_DEVICE) && invalidate) {
1238 if (object->type == OBJT_DEVICE)
1240 * The option OBJPR_NOTMAPPED must be passed here
1241 * because vm_object_page_remove() cannot remove
1242 * unmanaged mappings.
1244 flags = OBJPR_NOTMAPPED;
1248 flags = OBJPR_CLEANONLY;
1249 vm_object_page_remove(object, OFF_TO_IDX(offset),
1250 OFF_TO_IDX(offset + size + PAGE_MASK), flags);
1252 VM_OBJECT_WUNLOCK(object);
1257 * Determine whether the given advice can be applied to the object. Advice is
1258 * not applied to unmanaged pages since they never belong to page queues, and
1259 * since MADV_FREE is destructive, it can apply only to anonymous pages that
1260 * have been mapped at most once.
1263 vm_object_advice_applies(vm_object_t object, int advice)
1266 if ((object->flags & OBJ_UNMANAGED) != 0)
1268 if (advice != MADV_FREE)
1270 return ((object->flags & (OBJ_ONEMAPPING | OBJ_ANON)) ==
1271 (OBJ_ONEMAPPING | OBJ_ANON));
1275 vm_object_madvise_freespace(vm_object_t object, int advice, vm_pindex_t pindex,
1279 if (advice == MADV_FREE && object->type == OBJT_SWAP)
1280 swap_pager_freespace(object, pindex, size);
1284 * vm_object_madvise:
1286 * Implements the madvise function at the object/page level.
1288 * MADV_WILLNEED (any object)
1290 * Activate the specified pages if they are resident.
1292 * MADV_DONTNEED (any object)
1294 * Deactivate the specified pages if they are resident.
1296 * MADV_FREE (OBJT_DEFAULT/OBJT_SWAP objects,
1297 * OBJ_ONEMAPPING only)
1299 * Deactivate and clean the specified pages if they are
1300 * resident. This permits the process to reuse the pages
1301 * without faulting or the kernel to reclaim the pages
1305 vm_object_madvise(vm_object_t object, vm_pindex_t pindex, vm_pindex_t end,
1308 vm_pindex_t tpindex;
1309 vm_object_t backing_object, tobject;
1316 VM_OBJECT_WLOCK(object);
1317 if (!vm_object_advice_applies(object, advice)) {
1318 VM_OBJECT_WUNLOCK(object);
1321 for (m = vm_page_find_least(object, pindex); pindex < end; pindex++) {
1325 * If the next page isn't resident in the top-level object, we
1326 * need to search the shadow chain. When applying MADV_FREE, we
1327 * take care to release any swap space used to store
1328 * non-resident pages.
1330 if (m == NULL || pindex < m->pindex) {
1332 * Optimize a common case: if the top-level object has
1333 * no backing object, we can skip over the non-resident
1334 * range in constant time.
1336 if (object->backing_object == NULL) {
1337 tpindex = (m != NULL && m->pindex < end) ?
1339 vm_object_madvise_freespace(object, advice,
1340 pindex, tpindex - pindex);
1341 if ((pindex = tpindex) == end)
1348 vm_object_madvise_freespace(tobject, advice,
1351 * Prepare to search the next object in the
1354 backing_object = tobject->backing_object;
1355 if (backing_object == NULL)
1357 VM_OBJECT_WLOCK(backing_object);
1359 OFF_TO_IDX(tobject->backing_object_offset);
1360 if (tobject != object)
1361 VM_OBJECT_WUNLOCK(tobject);
1362 tobject = backing_object;
1363 if (!vm_object_advice_applies(tobject, advice))
1365 } while ((tm = vm_page_lookup(tobject, tpindex)) ==
1370 m = TAILQ_NEXT(m, listq);
1374 * If the page is not in a normal state, skip it. The page
1375 * can not be invalidated while the object lock is held.
1377 if (!vm_page_all_valid(tm) || vm_page_wired(tm))
1379 KASSERT((tm->flags & PG_FICTITIOUS) == 0,
1380 ("vm_object_madvise: page %p is fictitious", tm));
1381 KASSERT((tm->oflags & VPO_UNMANAGED) == 0,
1382 ("vm_object_madvise: page %p is not managed", tm));
1383 if (vm_page_tryxbusy(tm) == 0) {
1384 if (object != tobject)
1385 VM_OBJECT_WUNLOCK(object);
1386 if (advice == MADV_WILLNEED) {
1388 * Reference the page before unlocking and
1389 * sleeping so that the page daemon is less
1390 * likely to reclaim it.
1392 vm_page_aflag_set(tm, PGA_REFERENCED);
1394 vm_page_busy_sleep(tm, "madvpo", false);
1397 vm_page_advise(tm, advice);
1398 vm_page_xunbusy(tm);
1399 vm_object_madvise_freespace(tobject, advice, tm->pindex, 1);
1401 if (tobject != object)
1402 VM_OBJECT_WUNLOCK(tobject);
1404 VM_OBJECT_WUNLOCK(object);
1410 * Create a new object which is backed by the
1411 * specified existing object range. The source
1412 * object reference is deallocated.
1414 * The new object and offset into that object
1415 * are returned in the source parameters.
1418 vm_object_shadow(vm_object_t *object, vm_ooffset_t *offset, vm_size_t length,
1419 struct ucred *cred, bool shared)
1427 * Don't create the new object if the old object isn't shared.
1429 * If we hold the only reference we can guarantee that it won't
1430 * increase while we have the map locked. Otherwise the race is
1431 * harmless and we will end up with an extra shadow object that
1432 * will be collapsed later.
1434 if (source != NULL && source->ref_count == 1 &&
1435 (source->flags & OBJ_ANON) != 0)
1439 * Allocate a new object with the given length.
1441 result = vm_object_allocate_anon(atop(length), source, cred, length);
1444 * Store the offset into the source object, and fix up the offset into
1447 result->backing_object_offset = *offset;
1449 if (shared || source != NULL) {
1450 VM_OBJECT_WLOCK(result);
1453 * The new object shadows the source object, adding a
1454 * reference to it. Our caller changes his reference
1455 * to point to the new object, removing a reference to
1456 * the source object. Net result: no change of
1457 * reference count, unless the caller needs to add one
1458 * more reference due to forking a shared map entry.
1461 vm_object_reference_locked(result);
1462 vm_object_clear_flag(result, OBJ_ONEMAPPING);
1466 * Try to optimize the result object's page color when
1467 * shadowing in order to maintain page coloring
1468 * consistency in the combined shadowed object.
1470 if (source != NULL) {
1471 vm_object_backing_insert(result, source);
1472 result->domain = source->domain;
1473 #if VM_NRESERVLEVEL > 0
1474 result->flags |= source->flags & OBJ_COLORED;
1475 result->pg_color = (source->pg_color +
1476 OFF_TO_IDX(*offset)) & ((1 << (VM_NFREEORDER -
1480 VM_OBJECT_WUNLOCK(result);
1484 * Return the new things
1493 * Split the pages in a map entry into a new object. This affords
1494 * easier removal of unused pages, and keeps object inheritance from
1495 * being a negative impact on memory usage.
1498 vm_object_split(vm_map_entry_t entry)
1500 vm_page_t m, m_next;
1501 vm_object_t orig_object, new_object, backing_object;
1502 vm_pindex_t idx, offidxstart;
1505 orig_object = entry->object.vm_object;
1506 KASSERT((orig_object->flags & OBJ_ONEMAPPING) != 0,
1507 ("vm_object_split: Splitting object with multiple mappings."));
1508 if ((orig_object->flags & OBJ_ANON) == 0)
1510 if (orig_object->ref_count <= 1)
1512 VM_OBJECT_WUNLOCK(orig_object);
1514 offidxstart = OFF_TO_IDX(entry->offset);
1515 size = atop(entry->end - entry->start);
1518 * If swap_pager_copy() is later called, it will convert new_object
1519 * into a swap object.
1521 new_object = vm_object_allocate_anon(size, orig_object,
1522 orig_object->cred, ptoa(size));
1525 * We must wait for the orig_object to complete any in-progress
1526 * collapse so that the swap blocks are stable below. The
1527 * additional reference on backing_object by new object will
1528 * prevent further collapse operations until split completes.
1530 VM_OBJECT_WLOCK(orig_object);
1531 vm_object_collapse_wait(orig_object);
1534 * At this point, the new object is still private, so the order in
1535 * which the original and new objects are locked does not matter.
1537 VM_OBJECT_WLOCK(new_object);
1538 new_object->domain = orig_object->domain;
1539 backing_object = orig_object->backing_object;
1540 if (backing_object != NULL) {
1541 vm_object_backing_insert_ref(new_object, backing_object);
1542 new_object->backing_object_offset =
1543 orig_object->backing_object_offset + entry->offset;
1545 if (orig_object->cred != NULL) {
1546 crhold(orig_object->cred);
1547 KASSERT(orig_object->charge >= ptoa(size),
1548 ("orig_object->charge < 0"));
1549 orig_object->charge -= ptoa(size);
1553 * Mark the split operation so that swap_pager_getpages() knows
1554 * that the object is in transition.
1556 vm_object_set_flag(orig_object, OBJ_SPLIT);
1558 m = vm_page_find_least(orig_object, offidxstart);
1559 for (; m != NULL && (idx = m->pindex - offidxstart) < size;
1561 m_next = TAILQ_NEXT(m, listq);
1564 * We must wait for pending I/O to complete before we can
1567 * We do not have to VM_PROT_NONE the page as mappings should
1568 * not be changed by this operation.
1570 if (vm_page_tryxbusy(m) == 0) {
1571 VM_OBJECT_WUNLOCK(new_object);
1572 vm_page_sleep_if_busy(m, "spltwt");
1573 VM_OBJECT_WLOCK(new_object);
1578 * The page was left invalid. Likely placed there by
1579 * an incomplete fault. Just remove and ignore.
1581 if (vm_page_none_valid(m)) {
1582 if (vm_page_remove(m))
1587 /* vm_page_rename() will dirty the page. */
1588 if (vm_page_rename(m, new_object, idx)) {
1590 VM_OBJECT_WUNLOCK(new_object);
1591 VM_OBJECT_WUNLOCK(orig_object);
1593 VM_OBJECT_WLOCK(orig_object);
1594 VM_OBJECT_WLOCK(new_object);
1598 #if VM_NRESERVLEVEL > 0
1600 * If some of the reservation's allocated pages remain with
1601 * the original object, then transferring the reservation to
1602 * the new object is neither particularly beneficial nor
1603 * particularly harmful as compared to leaving the reservation
1604 * with the original object. If, however, all of the
1605 * reservation's allocated pages are transferred to the new
1606 * object, then transferring the reservation is typically
1607 * beneficial. Determining which of these two cases applies
1608 * would be more costly than unconditionally renaming the
1611 vm_reserv_rename(m, new_object, orig_object, offidxstart);
1613 if (orig_object->type != OBJT_SWAP)
1616 if (orig_object->type == OBJT_SWAP) {
1618 * swap_pager_copy() can sleep, in which case the orig_object's
1619 * and new_object's locks are released and reacquired.
1621 swap_pager_copy(orig_object, new_object, offidxstart, 0);
1622 TAILQ_FOREACH(m, &new_object->memq, listq)
1625 vm_object_clear_flag(orig_object, OBJ_SPLIT);
1626 VM_OBJECT_WUNLOCK(orig_object);
1627 VM_OBJECT_WUNLOCK(new_object);
1628 entry->object.vm_object = new_object;
1629 entry->offset = 0LL;
1630 vm_object_deallocate(orig_object);
1631 VM_OBJECT_WLOCK(new_object);
1635 vm_object_collapse_scan_wait(vm_object_t object, vm_page_t p)
1637 vm_object_t backing_object;
1639 VM_OBJECT_ASSERT_WLOCKED(object);
1640 backing_object = object->backing_object;
1641 VM_OBJECT_ASSERT_WLOCKED(backing_object);
1643 KASSERT(p == NULL || p->object == object || p->object == backing_object,
1644 ("invalid ownership %p %p %p", p, object, backing_object));
1645 /* The page is only NULL when rename fails. */
1647 VM_OBJECT_WUNLOCK(object);
1648 VM_OBJECT_WUNLOCK(backing_object);
1651 if (p->object == object)
1652 VM_OBJECT_WUNLOCK(backing_object);
1654 VM_OBJECT_WUNLOCK(object);
1655 vm_page_busy_sleep(p, "vmocol", false);
1657 VM_OBJECT_WLOCK(object);
1658 VM_OBJECT_WLOCK(backing_object);
1659 return (TAILQ_FIRST(&backing_object->memq));
1663 vm_object_scan_all_shadowed(vm_object_t object)
1665 vm_object_t backing_object;
1667 vm_pindex_t backing_offset_index, new_pindex, pi, ps;
1669 VM_OBJECT_ASSERT_WLOCKED(object);
1670 VM_OBJECT_ASSERT_WLOCKED(object->backing_object);
1672 backing_object = object->backing_object;
1674 if ((backing_object->flags & OBJ_ANON) == 0)
1677 pi = backing_offset_index = OFF_TO_IDX(object->backing_object_offset);
1678 p = vm_page_find_least(backing_object, pi);
1679 ps = swap_pager_find_least(backing_object, pi);
1682 * Only check pages inside the parent object's range and
1683 * inside the parent object's mapping of the backing object.
1686 if (p != NULL && p->pindex < pi)
1687 p = TAILQ_NEXT(p, listq);
1689 ps = swap_pager_find_least(backing_object, pi);
1690 if (p == NULL && ps >= backing_object->size)
1695 pi = MIN(p->pindex, ps);
1697 new_pindex = pi - backing_offset_index;
1698 if (new_pindex >= object->size)
1703 * If the backing object page is busy a
1704 * grandparent or older page may still be
1705 * undergoing CoW. It is not safe to collapse
1706 * the backing object until it is quiesced.
1708 if (vm_page_tryxbusy(p) == 0)
1712 * We raced with the fault handler that left
1713 * newly allocated invalid page on the object
1714 * queue and retried.
1716 if (!vm_page_all_valid(p))
1721 * See if the parent has the page or if the parent's object
1722 * pager has the page. If the parent has the page but the page
1723 * is not valid, the parent's object pager must have the page.
1725 * If this fails, the parent does not completely shadow the
1726 * object and we might as well give up now.
1728 pp = vm_page_lookup(object, new_pindex);
1731 * The valid check here is stable due to object lock
1732 * being required to clear valid and initiate paging.
1733 * Busy of p disallows fault handler to validate pp.
1735 if ((pp == NULL || vm_page_none_valid(pp)) &&
1736 !vm_pager_has_page(object, new_pindex, NULL, NULL))
1750 vm_object_collapse_scan(vm_object_t object)
1752 vm_object_t backing_object;
1753 vm_page_t next, p, pp;
1754 vm_pindex_t backing_offset_index, new_pindex;
1756 VM_OBJECT_ASSERT_WLOCKED(object);
1757 VM_OBJECT_ASSERT_WLOCKED(object->backing_object);
1759 backing_object = object->backing_object;
1760 backing_offset_index = OFF_TO_IDX(object->backing_object_offset);
1765 for (p = TAILQ_FIRST(&backing_object->memq); p != NULL; p = next) {
1766 next = TAILQ_NEXT(p, listq);
1767 new_pindex = p->pindex - backing_offset_index;
1770 * Check for busy page
1772 if (vm_page_tryxbusy(p) == 0) {
1773 next = vm_object_collapse_scan_wait(object, p);
1777 KASSERT(object->backing_object == backing_object,
1778 ("vm_object_collapse_scan: backing object mismatch %p != %p",
1779 object->backing_object, backing_object));
1780 KASSERT(p->object == backing_object,
1781 ("vm_object_collapse_scan: object mismatch %p != %p",
1782 p->object, backing_object));
1784 if (p->pindex < backing_offset_index ||
1785 new_pindex >= object->size) {
1786 if (backing_object->type == OBJT_SWAP)
1787 swap_pager_freespace(backing_object, p->pindex,
1790 KASSERT(!pmap_page_is_mapped(p),
1791 ("freeing mapped page %p", p));
1792 if (vm_page_remove(p))
1797 if (!vm_page_all_valid(p)) {
1798 KASSERT(!pmap_page_is_mapped(p),
1799 ("freeing mapped page %p", p));
1800 if (vm_page_remove(p))
1805 pp = vm_page_lookup(object, new_pindex);
1806 if (pp != NULL && vm_page_tryxbusy(pp) == 0) {
1809 * The page in the parent is busy and possibly not
1810 * (yet) valid. Until its state is finalized by the
1811 * busy bit owner, we can't tell whether it shadows the
1814 next = vm_object_collapse_scan_wait(object, pp);
1818 if (pp != NULL && vm_page_none_valid(pp)) {
1820 * The page was invalid in the parent. Likely placed
1821 * there by an incomplete fault. Just remove and
1822 * ignore. p can replace it.
1824 if (vm_page_remove(pp))
1829 if (pp != NULL || vm_pager_has_page(object, new_pindex, NULL,
1832 * The page already exists in the parent OR swap exists
1833 * for this location in the parent. Leave the parent's
1834 * page alone. Destroy the original page from the
1837 if (backing_object->type == OBJT_SWAP)
1838 swap_pager_freespace(backing_object, p->pindex,
1840 KASSERT(!pmap_page_is_mapped(p),
1841 ("freeing mapped page %p", p));
1842 if (vm_page_remove(p))
1845 vm_page_xunbusy(pp);
1850 * Page does not exist in parent, rename the page from the
1851 * backing object to the main object.
1853 * If the page was mapped to a process, it can remain mapped
1854 * through the rename. vm_page_rename() will dirty the page.
1856 if (vm_page_rename(p, object, new_pindex)) {
1858 next = vm_object_collapse_scan_wait(object, NULL);
1862 /* Use the old pindex to free the right page. */
1863 if (backing_object->type == OBJT_SWAP)
1864 swap_pager_freespace(backing_object,
1865 new_pindex + backing_offset_index, 1);
1867 #if VM_NRESERVLEVEL > 0
1869 * Rename the reservation.
1871 vm_reserv_rename(p, object, backing_object,
1872 backing_offset_index);
1880 * vm_object_collapse:
1882 * Collapse an object with the object backing it.
1883 * Pages in the backing object are moved into the
1884 * parent, and the backing object is deallocated.
1887 vm_object_collapse(vm_object_t object)
1889 vm_object_t backing_object, new_backing_object;
1891 VM_OBJECT_ASSERT_WLOCKED(object);
1894 KASSERT((object->flags & (OBJ_DEAD | OBJ_ANON)) == OBJ_ANON,
1895 ("collapsing invalid object"));
1898 * Wait for the backing_object to finish any pending
1899 * collapse so that the caller sees the shortest possible
1902 backing_object = vm_object_backing_collapse_wait(object);
1903 if (backing_object == NULL)
1906 KASSERT(object->ref_count > 0 &&
1907 object->ref_count > object->shadow_count,
1908 ("collapse with invalid ref %d or shadow %d count.",
1909 object->ref_count, object->shadow_count));
1910 KASSERT((backing_object->flags &
1911 (OBJ_COLLAPSING | OBJ_DEAD)) == 0,
1912 ("vm_object_collapse: Backing object already collapsing."));
1913 KASSERT((object->flags & (OBJ_COLLAPSING | OBJ_DEAD)) == 0,
1914 ("vm_object_collapse: object is already collapsing."));
1917 * We know that we can either collapse the backing object if
1918 * the parent is the only reference to it, or (perhaps) have
1919 * the parent bypass the object if the parent happens to shadow
1920 * all the resident pages in the entire backing object.
1922 if (backing_object->ref_count == 1) {
1923 KASSERT(backing_object->shadow_count == 1,
1924 ("vm_object_collapse: shadow_count: %d",
1925 backing_object->shadow_count));
1926 vm_object_pip_add(object, 1);
1927 vm_object_set_flag(object, OBJ_COLLAPSING);
1928 vm_object_pip_add(backing_object, 1);
1929 vm_object_set_flag(backing_object, OBJ_DEAD);
1932 * If there is exactly one reference to the backing
1933 * object, we can collapse it into the parent.
1935 vm_object_collapse_scan(object);
1937 #if VM_NRESERVLEVEL > 0
1939 * Break any reservations from backing_object.
1941 if (__predict_false(!LIST_EMPTY(&backing_object->rvq)))
1942 vm_reserv_break_all(backing_object);
1946 * Move the pager from backing_object to object.
1948 if (backing_object->type == OBJT_SWAP) {
1950 * swap_pager_copy() can sleep, in which case
1951 * the backing_object's and object's locks are
1952 * released and reacquired.
1953 * Since swap_pager_copy() is being asked to
1954 * destroy backing_object, it will change the
1955 * type to OBJT_DEFAULT.
1960 OFF_TO_IDX(object->backing_object_offset), TRUE);
1964 * Object now shadows whatever backing_object did.
1966 vm_object_clear_flag(object, OBJ_COLLAPSING);
1967 vm_object_backing_transfer(object, backing_object);
1968 object->backing_object_offset +=
1969 backing_object->backing_object_offset;
1970 VM_OBJECT_WUNLOCK(object);
1971 vm_object_pip_wakeup(object);
1974 * Discard backing_object.
1976 * Since the backing object has no pages, no pager left,
1977 * and no object references within it, all that is
1978 * necessary is to dispose of it.
1980 KASSERT(backing_object->ref_count == 1, (
1981 "backing_object %p was somehow re-referenced during collapse!",
1983 vm_object_pip_wakeup(backing_object);
1984 (void)refcount_release(&backing_object->ref_count);
1985 vm_object_terminate(backing_object);
1986 counter_u64_add(object_collapses, 1);
1987 VM_OBJECT_WLOCK(object);
1990 * If we do not entirely shadow the backing object,
1991 * there is nothing we can do so we give up.
1993 * The object lock and backing_object lock must not
1994 * be dropped during this sequence.
1996 if (!vm_object_scan_all_shadowed(object)) {
1997 VM_OBJECT_WUNLOCK(backing_object);
2002 * Make the parent shadow the next object in the
2003 * chain. Deallocating backing_object will not remove
2004 * it, since its reference count is at least 2.
2006 vm_object_backing_remove_locked(object);
2007 new_backing_object = backing_object->backing_object;
2008 if (new_backing_object != NULL) {
2009 vm_object_backing_insert_ref(object,
2010 new_backing_object);
2011 object->backing_object_offset +=
2012 backing_object->backing_object_offset;
2016 * Drop the reference count on backing_object. Since
2017 * its ref_count was at least 2, it will not vanish.
2019 (void)refcount_release(&backing_object->ref_count);
2020 KASSERT(backing_object->ref_count >= 1, (
2021 "backing_object %p was somehow dereferenced during collapse!",
2023 VM_OBJECT_WUNLOCK(backing_object);
2024 counter_u64_add(object_bypasses, 1);
2028 * Try again with this object's new backing object.
2034 * vm_object_page_remove:
2036 * For the given object, either frees or invalidates each of the
2037 * specified pages. In general, a page is freed. However, if a page is
2038 * wired for any reason other than the existence of a managed, wired
2039 * mapping, then it may be invalidated but not removed from the object.
2040 * Pages are specified by the given range ["start", "end") and the option
2041 * OBJPR_CLEANONLY. As a special case, if "end" is zero, then the range
2042 * extends from "start" to the end of the object. If the option
2043 * OBJPR_CLEANONLY is specified, then only the non-dirty pages within the
2044 * specified range are affected. If the option OBJPR_NOTMAPPED is
2045 * specified, then the pages within the specified range must have no
2046 * mappings. Otherwise, if this option is not specified, any mappings to
2047 * the specified pages are removed before the pages are freed or
2050 * In general, this operation should only be performed on objects that
2051 * contain managed pages. There are, however, two exceptions. First, it
2052 * is performed on the kernel and kmem objects by vm_map_entry_delete().
2053 * Second, it is used by msync(..., MS_INVALIDATE) to invalidate device-
2054 * backed pages. In both of these cases, the option OBJPR_CLEANONLY must
2055 * not be specified and the option OBJPR_NOTMAPPED must be specified.
2057 * The object must be locked.
2060 vm_object_page_remove(vm_object_t object, vm_pindex_t start, vm_pindex_t end,
2065 VM_OBJECT_ASSERT_WLOCKED(object);
2066 KASSERT((object->flags & OBJ_UNMANAGED) == 0 ||
2067 (options & (OBJPR_CLEANONLY | OBJPR_NOTMAPPED)) == OBJPR_NOTMAPPED,
2068 ("vm_object_page_remove: illegal options for object %p", object));
2069 if (object->resident_page_count == 0)
2071 vm_object_pip_add(object, 1);
2073 p = vm_page_find_least(object, start);
2076 * Here, the variable "p" is either (1) the page with the least pindex
2077 * greater than or equal to the parameter "start" or (2) NULL.
2079 for (; p != NULL && (p->pindex < end || end == 0); p = next) {
2080 next = TAILQ_NEXT(p, listq);
2083 * If the page is wired for any reason besides the existence
2084 * of managed, wired mappings, then it cannot be freed. For
2085 * example, fictitious pages, which represent device memory,
2086 * are inherently wired and cannot be freed. They can,
2087 * however, be invalidated if the option OBJPR_CLEANONLY is
2090 if (vm_page_tryxbusy(p) == 0) {
2091 vm_page_sleep_if_busy(p, "vmopar");
2094 if (vm_page_wired(p)) {
2096 if ((options & OBJPR_NOTMAPPED) == 0 &&
2097 object->ref_count != 0)
2099 if ((options & OBJPR_CLEANONLY) == 0) {
2106 KASSERT((p->flags & PG_FICTITIOUS) == 0,
2107 ("vm_object_page_remove: page %p is fictitious", p));
2108 if ((options & OBJPR_CLEANONLY) != 0 &&
2109 !vm_page_none_valid(p)) {
2110 if ((options & OBJPR_NOTMAPPED) == 0 &&
2111 object->ref_count != 0 &&
2112 !vm_page_try_remove_write(p))
2114 if (p->dirty != 0) {
2119 if ((options & OBJPR_NOTMAPPED) == 0 &&
2120 object->ref_count != 0 && !vm_page_try_remove_all(p))
2124 vm_object_pip_wakeup(object);
2126 if (object->type == OBJT_SWAP) {
2129 swap_pager_freespace(object, start, end - start);
2134 * vm_object_page_noreuse:
2136 * For the given object, attempt to move the specified pages to
2137 * the head of the inactive queue. This bypasses regular LRU
2138 * operation and allows the pages to be reused quickly under memory
2139 * pressure. If a page is wired for any reason, then it will not
2140 * be queued. Pages are specified by the range ["start", "end").
2141 * As a special case, if "end" is zero, then the range extends from
2142 * "start" to the end of the object.
2144 * This operation should only be performed on objects that
2145 * contain non-fictitious, managed pages.
2147 * The object must be locked.
2150 vm_object_page_noreuse(vm_object_t object, vm_pindex_t start, vm_pindex_t end)
2154 VM_OBJECT_ASSERT_LOCKED(object);
2155 KASSERT((object->flags & (OBJ_FICTITIOUS | OBJ_UNMANAGED)) == 0,
2156 ("vm_object_page_noreuse: illegal object %p", object));
2157 if (object->resident_page_count == 0)
2159 p = vm_page_find_least(object, start);
2162 * Here, the variable "p" is either (1) the page with the least pindex
2163 * greater than or equal to the parameter "start" or (2) NULL.
2165 for (; p != NULL && (p->pindex < end || end == 0); p = next) {
2166 next = TAILQ_NEXT(p, listq);
2167 vm_page_deactivate_noreuse(p);
2172 * Populate the specified range of the object with valid pages. Returns
2173 * TRUE if the range is successfully populated and FALSE otherwise.
2175 * Note: This function should be optimized to pass a larger array of
2176 * pages to vm_pager_get_pages() before it is applied to a non-
2177 * OBJT_DEVICE object.
2179 * The object must be locked.
2182 vm_object_populate(vm_object_t object, vm_pindex_t start, vm_pindex_t end)
2188 VM_OBJECT_ASSERT_WLOCKED(object);
2189 for (pindex = start; pindex < end; pindex++) {
2190 rv = vm_page_grab_valid(&m, object, pindex, VM_ALLOC_NORMAL);
2191 if (rv != VM_PAGER_OK)
2195 * Keep "m" busy because a subsequent iteration may unlock
2199 if (pindex > start) {
2200 m = vm_page_lookup(object, start);
2201 while (m != NULL && m->pindex < pindex) {
2203 m = TAILQ_NEXT(m, listq);
2206 return (pindex == end);
2210 * Routine: vm_object_coalesce
2211 * Function: Coalesces two objects backing up adjoining
2212 * regions of memory into a single object.
2214 * returns TRUE if objects were combined.
2216 * NOTE: Only works at the moment if the second object is NULL -
2217 * if it's not, which object do we lock first?
2220 * prev_object First object to coalesce
2221 * prev_offset Offset into prev_object
2222 * prev_size Size of reference to prev_object
2223 * next_size Size of reference to the second object
2224 * reserved Indicator that extension region has
2225 * swap accounted for
2228 * The object must *not* be locked.
2231 vm_object_coalesce(vm_object_t prev_object, vm_ooffset_t prev_offset,
2232 vm_size_t prev_size, vm_size_t next_size, boolean_t reserved)
2234 vm_pindex_t next_pindex;
2236 if (prev_object == NULL)
2238 if ((prev_object->flags & OBJ_ANON) == 0)
2241 VM_OBJECT_WLOCK(prev_object);
2243 * Try to collapse the object first.
2245 vm_object_collapse(prev_object);
2248 * Can't coalesce if: . more than one reference . paged out . shadows
2249 * another object . has a copy elsewhere (any of which mean that the
2250 * pages not mapped to prev_entry may be in use anyway)
2252 if (prev_object->backing_object != NULL) {
2253 VM_OBJECT_WUNLOCK(prev_object);
2257 prev_size >>= PAGE_SHIFT;
2258 next_size >>= PAGE_SHIFT;
2259 next_pindex = OFF_TO_IDX(prev_offset) + prev_size;
2261 if (prev_object->ref_count > 1 &&
2262 prev_object->size != next_pindex &&
2263 (prev_object->flags & OBJ_ONEMAPPING) == 0) {
2264 VM_OBJECT_WUNLOCK(prev_object);
2269 * Account for the charge.
2271 if (prev_object->cred != NULL) {
2274 * If prev_object was charged, then this mapping,
2275 * although not charged now, may become writable
2276 * later. Non-NULL cred in the object would prevent
2277 * swap reservation during enabling of the write
2278 * access, so reserve swap now. Failed reservation
2279 * cause allocation of the separate object for the map
2280 * entry, and swap reservation for this entry is
2281 * managed in appropriate time.
2283 if (!reserved && !swap_reserve_by_cred(ptoa(next_size),
2284 prev_object->cred)) {
2285 VM_OBJECT_WUNLOCK(prev_object);
2288 prev_object->charge += ptoa(next_size);
2292 * Remove any pages that may still be in the object from a previous
2295 if (next_pindex < prev_object->size) {
2296 vm_object_page_remove(prev_object, next_pindex, next_pindex +
2299 if (prev_object->cred != NULL) {
2300 KASSERT(prev_object->charge >=
2301 ptoa(prev_object->size - next_pindex),
2302 ("object %p overcharged 1 %jx %jx", prev_object,
2303 (uintmax_t)next_pindex, (uintmax_t)next_size));
2304 prev_object->charge -= ptoa(prev_object->size -
2311 * Extend the object if necessary.
2313 if (next_pindex + next_size > prev_object->size)
2314 prev_object->size = next_pindex + next_size;
2316 VM_OBJECT_WUNLOCK(prev_object);
2321 vm_object_set_writeable_dirty(vm_object_t object)
2324 /* Only set for vnodes & tmpfs */
2325 if (object->type != OBJT_VNODE &&
2326 (object->flags & OBJ_TMPFS_NODE) == 0)
2328 atomic_add_int(&object->generation, 1);
2334 * For each page offset within the specified range of the given object,
2335 * find the highest-level page in the shadow chain and unwire it. A page
2336 * must exist at every page offset, and the highest-level page must be
2340 vm_object_unwire(vm_object_t object, vm_ooffset_t offset, vm_size_t length,
2343 vm_object_t tobject, t1object;
2345 vm_pindex_t end_pindex, pindex, tpindex;
2346 int depth, locked_depth;
2348 KASSERT((offset & PAGE_MASK) == 0,
2349 ("vm_object_unwire: offset is not page aligned"));
2350 KASSERT((length & PAGE_MASK) == 0,
2351 ("vm_object_unwire: length is not a multiple of PAGE_SIZE"));
2352 /* The wired count of a fictitious page never changes. */
2353 if ((object->flags & OBJ_FICTITIOUS) != 0)
2355 pindex = OFF_TO_IDX(offset);
2356 end_pindex = pindex + atop(length);
2359 VM_OBJECT_RLOCK(object);
2360 m = vm_page_find_least(object, pindex);
2361 while (pindex < end_pindex) {
2362 if (m == NULL || pindex < m->pindex) {
2364 * The first object in the shadow chain doesn't
2365 * contain a page at the current index. Therefore,
2366 * the page must exist in a backing object.
2373 OFF_TO_IDX(tobject->backing_object_offset);
2374 tobject = tobject->backing_object;
2375 KASSERT(tobject != NULL,
2376 ("vm_object_unwire: missing page"));
2377 if ((tobject->flags & OBJ_FICTITIOUS) != 0)
2380 if (depth == locked_depth) {
2382 VM_OBJECT_RLOCK(tobject);
2384 } while ((tm = vm_page_lookup(tobject, tpindex)) ==
2388 m = TAILQ_NEXT(m, listq);
2390 if (vm_page_trysbusy(tm) == 0) {
2391 for (tobject = object; locked_depth >= 1;
2393 t1object = tobject->backing_object;
2394 if (tm->object != tobject)
2395 VM_OBJECT_RUNLOCK(tobject);
2398 vm_page_busy_sleep(tm, "unwbo", true);
2401 vm_page_unwire(tm, queue);
2402 vm_page_sunbusy(tm);
2406 /* Release the accumulated object locks. */
2407 for (tobject = object; locked_depth >= 1; locked_depth--) {
2408 t1object = tobject->backing_object;
2409 VM_OBJECT_RUNLOCK(tobject);
2415 * Return the vnode for the given object, or NULL if none exists.
2416 * For tmpfs objects, the function may return NULL if there is
2417 * no vnode allocated at the time of the call.
2420 vm_object_vnode(vm_object_t object)
2424 VM_OBJECT_ASSERT_LOCKED(object);
2425 if (object->type == OBJT_VNODE) {
2426 vp = object->handle;
2427 KASSERT(vp != NULL, ("%s: OBJT_VNODE has no vnode", __func__));
2428 } else if (object->type == OBJT_SWAP &&
2429 (object->flags & OBJ_TMPFS) != 0) {
2430 vp = object->un_pager.swp.swp_tmpfs;
2431 KASSERT(vp != NULL, ("%s: OBJT_TMPFS has no vnode", __func__));
2440 * Busy the vm object. This prevents new pages belonging to the object from
2441 * becoming busy. Existing pages persist as busy. Callers are responsible
2442 * for checking page state before proceeding.
2445 vm_object_busy(vm_object_t obj)
2448 VM_OBJECT_ASSERT_LOCKED(obj);
2450 blockcount_acquire(&obj->busy, 1);
2451 /* The fence is required to order loads of page busy. */
2452 atomic_thread_fence_acq_rel();
2456 vm_object_unbusy(vm_object_t obj)
2459 blockcount_release(&obj->busy, 1);
2463 vm_object_busy_wait(vm_object_t obj, const char *wmesg)
2466 VM_OBJECT_ASSERT_UNLOCKED(obj);
2468 (void)blockcount_sleep(&obj->busy, NULL, wmesg, PVM);
2472 * Return the kvme type of the given object.
2473 * If vpp is not NULL, set it to the object's vm_object_vnode() or NULL.
2476 vm_object_kvme_type(vm_object_t object, struct vnode **vpp)
2479 VM_OBJECT_ASSERT_LOCKED(object);
2481 *vpp = vm_object_vnode(object);
2482 switch (object->type) {
2484 return (KVME_TYPE_DEFAULT);
2486 return (KVME_TYPE_VNODE);
2488 if ((object->flags & OBJ_TMPFS_NODE) != 0)
2489 return (KVME_TYPE_VNODE);
2490 return (KVME_TYPE_SWAP);
2492 return (KVME_TYPE_DEVICE);
2494 return (KVME_TYPE_PHYS);
2496 return (KVME_TYPE_DEAD);
2498 return (KVME_TYPE_SG);
2499 case OBJT_MGTDEVICE:
2500 return (KVME_TYPE_MGTDEVICE);
2502 return (KVME_TYPE_UNKNOWN);
2507 sysctl_vm_object_list(SYSCTL_HANDLER_ARGS)
2509 struct kinfo_vmobject *kvo;
2510 char *fullpath, *freepath;
2517 if (req->oldptr == NULL) {
2519 * If an old buffer has not been provided, generate an
2520 * estimate of the space needed for a subsequent call.
2522 mtx_lock(&vm_object_list_mtx);
2524 TAILQ_FOREACH(obj, &vm_object_list, object_list) {
2525 if (obj->type == OBJT_DEAD)
2529 mtx_unlock(&vm_object_list_mtx);
2530 return (SYSCTL_OUT(req, NULL, sizeof(struct kinfo_vmobject) *
2534 kvo = malloc(sizeof(*kvo), M_TEMP, M_WAITOK);
2538 * VM objects are type stable and are never removed from the
2539 * list once added. This allows us to safely read obj->object_list
2540 * after reacquiring the VM object lock.
2542 mtx_lock(&vm_object_list_mtx);
2543 TAILQ_FOREACH(obj, &vm_object_list, object_list) {
2544 if (obj->type == OBJT_DEAD)
2546 VM_OBJECT_RLOCK(obj);
2547 if (obj->type == OBJT_DEAD) {
2548 VM_OBJECT_RUNLOCK(obj);
2551 mtx_unlock(&vm_object_list_mtx);
2552 kvo->kvo_size = ptoa(obj->size);
2553 kvo->kvo_resident = obj->resident_page_count;
2554 kvo->kvo_ref_count = obj->ref_count;
2555 kvo->kvo_shadow_count = obj->shadow_count;
2556 kvo->kvo_memattr = obj->memattr;
2557 kvo->kvo_active = 0;
2558 kvo->kvo_inactive = 0;
2559 TAILQ_FOREACH(m, &obj->memq, listq) {
2561 * A page may belong to the object but be
2562 * dequeued and set to PQ_NONE while the
2563 * object lock is not held. This makes the
2564 * reads of m->queue below racy, and we do not
2565 * count pages set to PQ_NONE. However, this
2566 * sysctl is only meant to give an
2567 * approximation of the system anyway.
2569 if (m->a.queue == PQ_ACTIVE)
2571 else if (m->a.queue == PQ_INACTIVE)
2572 kvo->kvo_inactive++;
2575 kvo->kvo_vn_fileid = 0;
2576 kvo->kvo_vn_fsid = 0;
2577 kvo->kvo_vn_fsid_freebsd11 = 0;
2580 kvo->kvo_type = vm_object_kvme_type(obj, &vp);
2583 VM_OBJECT_RUNLOCK(obj);
2585 vn_fullpath(vp, &fullpath, &freepath);
2586 vn_lock(vp, LK_SHARED | LK_RETRY);
2587 if (VOP_GETATTR(vp, &va, curthread->td_ucred) == 0) {
2588 kvo->kvo_vn_fileid = va.va_fileid;
2589 kvo->kvo_vn_fsid = va.va_fsid;
2590 kvo->kvo_vn_fsid_freebsd11 = va.va_fsid;
2596 strlcpy(kvo->kvo_path, fullpath, sizeof(kvo->kvo_path));
2597 if (freepath != NULL)
2598 free(freepath, M_TEMP);
2600 /* Pack record size down */
2601 kvo->kvo_structsize = offsetof(struct kinfo_vmobject, kvo_path)
2602 + strlen(kvo->kvo_path) + 1;
2603 kvo->kvo_structsize = roundup(kvo->kvo_structsize,
2605 error = SYSCTL_OUT(req, kvo, kvo->kvo_structsize);
2606 mtx_lock(&vm_object_list_mtx);
2610 mtx_unlock(&vm_object_list_mtx);
2614 SYSCTL_PROC(_vm, OID_AUTO, objects, CTLTYPE_STRUCT | CTLFLAG_RW | CTLFLAG_SKIP |
2615 CTLFLAG_MPSAFE, NULL, 0, sysctl_vm_object_list, "S,kinfo_vmobject",
2616 "List of VM objects");
2618 #include "opt_ddb.h"
2620 #include <sys/kernel.h>
2622 #include <sys/cons.h>
2624 #include <ddb/ddb.h>
2627 _vm_object_in_map(vm_map_t map, vm_object_t object, vm_map_entry_t entry)
2630 vm_map_entry_t tmpe;
2637 VM_MAP_ENTRY_FOREACH(tmpe, map) {
2638 if (_vm_object_in_map(map, object, tmpe)) {
2642 } else if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
2643 tmpm = entry->object.sub_map;
2644 VM_MAP_ENTRY_FOREACH(tmpe, tmpm) {
2645 if (_vm_object_in_map(tmpm, object, tmpe)) {
2649 } else if ((obj = entry->object.vm_object) != NULL) {
2650 for (; obj; obj = obj->backing_object)
2651 if (obj == object) {
2659 vm_object_in_map(vm_object_t object)
2663 /* sx_slock(&allproc_lock); */
2664 FOREACH_PROC_IN_SYSTEM(p) {
2665 if (!p->p_vmspace /* || (p->p_flag & (P_SYSTEM|P_WEXIT)) */)
2667 if (_vm_object_in_map(&p->p_vmspace->vm_map, object, 0)) {
2668 /* sx_sunlock(&allproc_lock); */
2672 /* sx_sunlock(&allproc_lock); */
2673 if (_vm_object_in_map(kernel_map, object, 0))
2678 DB_SHOW_COMMAND(vmochk, vm_object_check)
2683 * make sure that internal objs are in a map somewhere
2684 * and none have zero ref counts.
2686 TAILQ_FOREACH(object, &vm_object_list, object_list) {
2687 if ((object->flags & OBJ_ANON) != 0) {
2688 if (object->ref_count == 0) {
2689 db_printf("vmochk: internal obj has zero ref count: %ld\n",
2690 (long)object->size);
2692 if (!vm_object_in_map(object)) {
2694 "vmochk: internal obj is not in a map: "
2695 "ref: %d, size: %lu: 0x%lx, backing_object: %p\n",
2696 object->ref_count, (u_long)object->size,
2697 (u_long)object->size,
2698 (void *)object->backing_object);
2707 * vm_object_print: [ debug ]
2709 DB_SHOW_COMMAND(object, vm_object_print_static)
2711 /* XXX convert args. */
2712 vm_object_t object = (vm_object_t)addr;
2713 boolean_t full = have_addr;
2717 /* XXX count is an (unused) arg. Avoid shadowing it. */
2718 #define count was_count
2726 "Object %p: type=%d, size=0x%jx, res=%d, ref=%d, flags=0x%x ruid %d charge %jx\n",
2727 object, (int)object->type, (uintmax_t)object->size,
2728 object->resident_page_count, object->ref_count, object->flags,
2729 object->cred ? object->cred->cr_ruid : -1, (uintmax_t)object->charge);
2730 db_iprintf(" sref=%d, backing_object(%d)=(%p)+0x%jx\n",
2731 object->shadow_count,
2732 object->backing_object ? object->backing_object->ref_count : 0,
2733 object->backing_object, (uintmax_t)object->backing_object_offset);
2740 TAILQ_FOREACH(p, &object->memq, listq) {
2742 db_iprintf("memory:=");
2743 else if (count == 6) {
2751 db_printf("(off=0x%jx,page=0x%jx)",
2752 (uintmax_t)p->pindex, (uintmax_t)VM_PAGE_TO_PHYS(p));
2765 /* XXX need this non-static entry for calling from vm_map_print. */
2768 /* db_expr_t */ long addr,
2769 boolean_t have_addr,
2770 /* db_expr_t */ long count,
2773 vm_object_print_static(addr, have_addr, count, modif);
2776 DB_SHOW_COMMAND(vmopag, vm_object_print_pages)
2781 vm_page_t m, prev_m;
2785 TAILQ_FOREACH(object, &vm_object_list, object_list) {
2786 db_printf("new object: %p\n", (void *)object);
2797 TAILQ_FOREACH(m, &object->memq, listq) {
2798 if (m->pindex > 128)
2800 if ((prev_m = TAILQ_PREV(m, pglist, listq)) != NULL &&
2801 prev_m->pindex + 1 != m->pindex) {
2803 db_printf(" index(%ld)run(%d)pa(0x%lx)\n",
2804 (long)fidx, rcount, (long)pa);
2816 (VM_PAGE_TO_PHYS(m) == pa + rcount * PAGE_SIZE)) {
2821 db_printf(" index(%ld)run(%d)pa(0x%lx)\n",
2822 (long)fidx, rcount, (long)pa);
2832 pa = VM_PAGE_TO_PHYS(m);
2836 db_printf(" index(%ld)run(%d)pa(0x%lx)\n",
2837 (long)fidx, rcount, (long)pa);