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/cpuset.h>
77 #include <sys/mount.h>
78 #include <sys/kernel.h>
79 #include <sys/pctrie.h>
80 #include <sys/sysctl.h>
81 #include <sys/mutex.h>
82 #include <sys/proc.h> /* for curproc, pageproc */
83 #include <sys/refcount.h>
84 #include <sys/socket.h>
85 #include <sys/resourcevar.h>
86 #include <sys/refcount.h>
87 #include <sys/rwlock.h>
89 #include <sys/vnode.h>
90 #include <sys/vmmeter.h>
94 #include <vm/vm_param.h>
96 #include <vm/vm_map.h>
97 #include <vm/vm_object.h>
98 #include <vm/vm_page.h>
99 #include <vm/vm_pageout.h>
100 #include <vm/vm_pager.h>
101 #include <vm/vm_phys.h>
102 #include <vm/vm_pagequeue.h>
103 #include <vm/swap_pager.h>
104 #include <vm/vm_kern.h>
105 #include <vm/vm_extern.h>
106 #include <vm/vm_radix.h>
107 #include <vm/vm_reserv.h>
110 static int old_msync;
111 SYSCTL_INT(_vm, OID_AUTO, old_msync, CTLFLAG_RW, &old_msync, 0,
112 "Use old (insecure) msync behavior");
114 static int vm_object_page_collect_flush(vm_object_t object, vm_page_t p,
115 int pagerflags, int flags, boolean_t *allclean,
117 static boolean_t vm_object_page_remove_write(vm_page_t p, int flags,
118 boolean_t *allclean);
119 static void vm_object_qcollapse(vm_object_t object);
120 static void vm_object_vndeallocate(vm_object_t object);
121 static void vm_object_backing_remove(vm_object_t object);
124 * Virtual memory objects maintain the actual data
125 * associated with allocated virtual memory. A given
126 * page of memory exists within exactly one object.
128 * An object is only deallocated when all "references"
129 * are given up. Only one "reference" to a given
130 * region of an object should be writeable.
132 * Associated with each object is a list of all resident
133 * memory pages belonging to that object; this list is
134 * maintained by the "vm_page" module, and locked by the object's
137 * Each object also records a "pager" routine which is
138 * used to retrieve (and store) pages to the proper backing
139 * storage. In addition, objects may be backed by other
140 * objects from which they were virtual-copied.
142 * The only items within the object structure which are
143 * modified after time of creation are:
144 * reference count locked by object's lock
145 * pager routine locked by object's lock
149 struct object_q vm_object_list;
150 struct mtx vm_object_list_mtx; /* lock for object list and count */
152 struct vm_object kernel_object_store;
154 static SYSCTL_NODE(_vm_stats, OID_AUTO, object, CTLFLAG_RD, 0,
157 static counter_u64_t object_collapses = EARLY_COUNTER;
158 SYSCTL_COUNTER_U64(_vm_stats_object, OID_AUTO, collapses, CTLFLAG_RD,
160 "VM object collapses");
162 static counter_u64_t object_bypasses = EARLY_COUNTER;
163 SYSCTL_COUNTER_U64(_vm_stats_object, OID_AUTO, bypasses, CTLFLAG_RD,
165 "VM object bypasses");
168 counter_startup(void)
171 object_collapses = counter_u64_alloc(M_WAITOK);
172 object_bypasses = counter_u64_alloc(M_WAITOK);
174 SYSINIT(object_counters, SI_SUB_CPU, SI_ORDER_ANY, counter_startup, NULL);
176 static uma_zone_t obj_zone;
178 static int vm_object_zinit(void *mem, int size, int flags);
181 static void vm_object_zdtor(void *mem, int size, void *arg);
184 vm_object_zdtor(void *mem, int size, void *arg)
188 object = (vm_object_t)mem;
189 KASSERT(object->ref_count == 0,
190 ("object %p ref_count = %d", object, object->ref_count));
191 KASSERT(TAILQ_EMPTY(&object->memq),
192 ("object %p has resident pages in its memq", object));
193 KASSERT(vm_radix_is_empty(&object->rtree),
194 ("object %p has resident pages in its trie", object));
195 #if VM_NRESERVLEVEL > 0
196 KASSERT(LIST_EMPTY(&object->rvq),
197 ("object %p has reservations",
200 KASSERT(REFCOUNT_COUNT(object->paging_in_progress) == 0,
201 ("object %p paging_in_progress = %d",
202 object, REFCOUNT_COUNT(object->paging_in_progress)));
203 KASSERT(object->busy == 0,
204 ("object %p busy = %d",
205 object, object->busy));
206 KASSERT(object->resident_page_count == 0,
207 ("object %p resident_page_count = %d",
208 object, object->resident_page_count));
209 KASSERT(object->shadow_count == 0,
210 ("object %p shadow_count = %d",
211 object, object->shadow_count));
212 KASSERT(object->type == OBJT_DEAD,
213 ("object %p has non-dead type %d",
214 object, object->type));
219 vm_object_zinit(void *mem, int size, int flags)
223 object = (vm_object_t)mem;
224 rw_init_flags(&object->lock, "vm object", RW_DUPOK | RW_NEW);
226 /* These are true for any object that has been freed */
227 object->type = OBJT_DEAD;
228 vm_radix_init(&object->rtree);
229 refcount_init(&object->ref_count, 0);
230 refcount_init(&object->paging_in_progress, 0);
231 refcount_init(&object->busy, 0);
232 object->resident_page_count = 0;
233 object->shadow_count = 0;
234 object->flags = OBJ_DEAD;
236 mtx_lock(&vm_object_list_mtx);
237 TAILQ_INSERT_TAIL(&vm_object_list, object, object_list);
238 mtx_unlock(&vm_object_list_mtx);
243 _vm_object_allocate(objtype_t type, vm_pindex_t size, u_short flags,
244 vm_object_t object, void *handle)
247 TAILQ_INIT(&object->memq);
248 LIST_INIT(&object->shadow_head);
251 if (type == OBJT_SWAP)
252 pctrie_init(&object->un_pager.swp.swp_blks);
255 * Ensure that swap_pager_swapoff() iteration over object_list
256 * sees up to date type and pctrie head if it observed
259 atomic_thread_fence_rel();
261 object->pg_color = 0;
262 object->flags = flags;
264 object->domain.dr_policy = NULL;
265 object->generation = 1;
266 object->cleangeneration = 1;
267 refcount_init(&object->ref_count, 1);
268 object->memattr = VM_MEMATTR_DEFAULT;
271 object->handle = handle;
272 object->backing_object = NULL;
273 object->backing_object_offset = (vm_ooffset_t) 0;
274 #if VM_NRESERVLEVEL > 0
275 LIST_INIT(&object->rvq);
277 umtx_shm_object_init(object);
283 * Initialize the VM objects module.
288 TAILQ_INIT(&vm_object_list);
289 mtx_init(&vm_object_list_mtx, "vm object_list", NULL, MTX_DEF);
291 rw_init(&kernel_object->lock, "kernel vm object");
292 _vm_object_allocate(OBJT_PHYS, atop(VM_MAX_KERNEL_ADDRESS -
293 VM_MIN_KERNEL_ADDRESS), OBJ_UNMANAGED, kernel_object, NULL);
294 #if VM_NRESERVLEVEL > 0
295 kernel_object->flags |= OBJ_COLORED;
296 kernel_object->pg_color = (u_short)atop(VM_MIN_KERNEL_ADDRESS);
300 * The lock portion of struct vm_object must be type stable due
301 * to vm_pageout_fallback_object_lock locking a vm object
302 * without holding any references to it.
304 obj_zone = uma_zcreate("VM OBJECT", sizeof (struct vm_object), NULL,
310 vm_object_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
316 vm_object_clear_flag(vm_object_t object, u_short bits)
319 VM_OBJECT_ASSERT_WLOCKED(object);
320 object->flags &= ~bits;
324 * Sets the default memory attribute for the specified object. Pages
325 * that are allocated to this object are by default assigned this memory
328 * Presently, this function must be called before any pages are allocated
329 * to the object. In the future, this requirement may be relaxed for
330 * "default" and "swap" objects.
333 vm_object_set_memattr(vm_object_t object, vm_memattr_t memattr)
336 VM_OBJECT_ASSERT_WLOCKED(object);
337 switch (object->type) {
345 if (!TAILQ_EMPTY(&object->memq))
346 return (KERN_FAILURE);
349 return (KERN_INVALID_ARGUMENT);
351 panic("vm_object_set_memattr: object %p is of undefined type",
354 object->memattr = memattr;
355 return (KERN_SUCCESS);
359 vm_object_pip_add(vm_object_t object, short i)
362 refcount_acquiren(&object->paging_in_progress, i);
366 vm_object_pip_wakeup(vm_object_t object)
369 refcount_release(&object->paging_in_progress);
373 vm_object_pip_wakeupn(vm_object_t object, short i)
376 refcount_releasen(&object->paging_in_progress, i);
380 vm_object_pip_wait(vm_object_t object, char *waitid)
383 VM_OBJECT_ASSERT_WLOCKED(object);
385 while (REFCOUNT_COUNT(object->paging_in_progress) > 0) {
386 VM_OBJECT_WUNLOCK(object);
387 refcount_wait(&object->paging_in_progress, waitid, PVM);
388 VM_OBJECT_WLOCK(object);
393 vm_object_pip_wait_unlocked(vm_object_t object, char *waitid)
396 VM_OBJECT_ASSERT_UNLOCKED(object);
398 while (REFCOUNT_COUNT(object->paging_in_progress) > 0)
399 refcount_wait(&object->paging_in_progress, waitid, PVM);
403 * vm_object_allocate:
405 * Returns a new object with the given size.
408 vm_object_allocate(objtype_t type, vm_pindex_t size)
415 panic("vm_object_allocate: can't create OBJT_DEAD");
422 flags = OBJ_FICTITIOUS | OBJ_UNMANAGED;
425 flags = OBJ_FICTITIOUS;
428 flags = OBJ_UNMANAGED;
434 panic("vm_object_allocate: type %d is undefined", type);
436 object = (vm_object_t)uma_zalloc(obj_zone, M_WAITOK);
437 _vm_object_allocate(type, size, flags, object, NULL);
443 * vm_object_allocate_anon:
445 * Returns a new default object of the given size and marked as
446 * anonymous memory for special split/collapse handling. Color
447 * to be initialized by the caller.
450 vm_object_allocate_anon(vm_pindex_t size, vm_object_t backing_object,
451 struct ucred *cred, vm_size_t charge)
453 vm_object_t handle, object;
455 if (backing_object == NULL)
457 else if ((backing_object->flags & OBJ_ANON) != 0)
458 handle = backing_object->handle;
460 handle = backing_object;
461 object = uma_zalloc(obj_zone, M_WAITOK);
462 _vm_object_allocate(OBJT_DEFAULT, size, OBJ_ANON | OBJ_ONEMAPPING,
465 object->charge = cred != NULL ? charge : 0;
471 * vm_object_reference:
473 * Gets another reference to the given object. Note: OBJ_DEAD
474 * objects can be referenced during final cleaning.
477 vm_object_reference(vm_object_t object)
486 * Many places assume exclusive access to objects with a single
487 * ref. vm_object_collapse() in particular will directly mainpulate
488 * references for objects in this state. vnode objects only need
489 * the lock for the first ref to reference the vnode.
491 if (!refcount_acquire_if_gt(&object->ref_count,
492 object->type == OBJT_VNODE ? 0 : 1)) {
493 VM_OBJECT_RLOCK(object);
494 old = refcount_acquire(&object->ref_count);
495 if (object->type == OBJT_VNODE && old == 0) {
499 VM_OBJECT_RUNLOCK(object);
504 * vm_object_reference_locked:
506 * Gets another reference to the given object.
508 * The object must be locked.
511 vm_object_reference_locked(vm_object_t object)
516 VM_OBJECT_ASSERT_LOCKED(object);
517 old = refcount_acquire(&object->ref_count);
518 if (object->type == OBJT_VNODE && old == 0) {
525 * Handle deallocating an object of type OBJT_VNODE.
528 vm_object_vndeallocate(vm_object_t object)
530 struct vnode *vp = (struct vnode *) object->handle;
533 KASSERT(object->type == OBJT_VNODE,
534 ("vm_object_vndeallocate: not a vnode object"));
535 KASSERT(vp != NULL, ("vm_object_vndeallocate: missing vp"));
537 /* Object lock to protect handle lookup. */
538 last = refcount_release(&object->ref_count);
539 VM_OBJECT_RUNLOCK(object);
544 if (!umtx_shm_vnobj_persistent)
545 umtx_shm_object_terminated(object);
547 /* vrele may need the vnode lock. */
552 * vm_object_deallocate:
554 * Release a reference to the specified object,
555 * gained either through a vm_object_allocate
556 * or a vm_object_reference call. When all references
557 * are gone, storage associated with this object
558 * may be relinquished.
560 * No object may be locked.
563 vm_object_deallocate(vm_object_t object)
565 vm_object_t robject, temp;
568 while (object != NULL) {
570 * If the reference count goes to 0 we start calling
571 * vm_object_terminate() on the object chain. A ref count
572 * of 1 may be a special case depending on the shadow count
573 * being 0 or 1. These cases require a write lock on the
576 if ((object->flags & OBJ_ANON) == 0)
577 released = refcount_release_if_gt(&object->ref_count, 1);
579 released = refcount_release_if_gt(&object->ref_count, 2);
583 if (object->type == OBJT_VNODE) {
584 VM_OBJECT_RLOCK(object);
585 if (object->type == OBJT_VNODE) {
586 vm_object_vndeallocate(object);
589 VM_OBJECT_RUNLOCK(object);
592 VM_OBJECT_WLOCK(object);
593 KASSERT(object->ref_count > 0,
594 ("vm_object_deallocate: object deallocated too many times: %d",
597 if (refcount_release(&object->ref_count))
599 if (object->ref_count > 1) {
600 VM_OBJECT_WUNLOCK(object);
602 } else if (object->ref_count == 1) {
603 if (object->shadow_count == 0 &&
604 (object->flags & OBJ_ANON) != 0) {
605 vm_object_set_flag(object, OBJ_ONEMAPPING);
606 } else if (object->shadow_count == 1) {
607 KASSERT((object->flags & OBJ_ANON) != 0,
608 ("obj %p with shadow_count > 0 is not anon",
610 robject = LIST_FIRST(&object->shadow_head);
611 KASSERT(robject != NULL,
612 ("vm_object_deallocate: ref_count: %d, "
613 "shadow_count: %d", object->ref_count,
614 object->shadow_count));
615 KASSERT((robject->flags & OBJ_TMPFS_NODE) == 0,
616 ("shadowed tmpfs v_object %p", object));
617 if (!VM_OBJECT_TRYWLOCK(robject)) {
619 * Avoid a potential deadlock.
621 refcount_acquire(&object->ref_count);
622 VM_OBJECT_WUNLOCK(object);
624 * More likely than not the thread
625 * holding robject's lock has lower
626 * priority than the current thread.
627 * Let the lower priority thread run.
633 * Collapse object into its shadow unless its
634 * shadow is dead. In that case, object will
635 * be deallocated by the thread that is
636 * deallocating its shadow.
638 if ((robject->flags &
639 (OBJ_DEAD | OBJ_ANON)) == OBJ_ANON) {
641 refcount_acquire(&robject->ref_count);
643 if (REFCOUNT_COUNT(robject->paging_in_progress) > 0) {
644 VM_OBJECT_WUNLOCK(object);
645 vm_object_pip_wait(robject,
647 temp = robject->backing_object;
648 if (object == temp) {
649 VM_OBJECT_WLOCK(object);
652 } else if (REFCOUNT_COUNT(object->paging_in_progress) > 0) {
653 VM_OBJECT_WUNLOCK(robject);
654 VM_OBJECT_WUNLOCK(object);
656 &object->paging_in_progress,
658 VM_OBJECT_WLOCK(robject);
659 temp = robject->backing_object;
660 if (object == temp) {
661 VM_OBJECT_WLOCK(object);
665 VM_OBJECT_WUNLOCK(object);
667 if (robject->ref_count == 1) {
668 refcount_release(&robject->ref_count);
673 vm_object_collapse(object);
674 VM_OBJECT_WUNLOCK(object);
677 VM_OBJECT_WUNLOCK(robject);
679 VM_OBJECT_WUNLOCK(object);
683 umtx_shm_object_terminated(object);
684 temp = object->backing_object;
686 KASSERT((object->flags & OBJ_TMPFS_NODE) == 0,
687 ("shadowed tmpfs v_object 2 %p", object));
688 vm_object_backing_remove(object);
691 * Don't double-terminate, we could be in a termination
692 * recursion due to the terminate having to sync data
695 if ((object->flags & OBJ_DEAD) == 0) {
696 vm_object_set_flag(object, OBJ_DEAD);
697 vm_object_terminate(object);
699 VM_OBJECT_WUNLOCK(object);
705 * vm_object_destroy removes the object from the global object list
706 * and frees the space for the object.
709 vm_object_destroy(vm_object_t object)
713 * Release the allocation charge.
715 if (object->cred != NULL) {
716 swap_release_by_cred(object->charge, object->cred);
718 crfree(object->cred);
723 * Free the space for the object.
725 uma_zfree(obj_zone, object);
729 vm_object_backing_remove_locked(vm_object_t object)
731 vm_object_t backing_object;
733 backing_object = object->backing_object;
734 VM_OBJECT_ASSERT_WLOCKED(object);
735 VM_OBJECT_ASSERT_WLOCKED(backing_object);
737 if ((object->flags & OBJ_SHADOWLIST) != 0) {
738 LIST_REMOVE(object, shadow_list);
739 backing_object->shadow_count--;
740 object->flags &= ~OBJ_SHADOWLIST;
742 object->backing_object = NULL;
746 vm_object_backing_remove(vm_object_t object)
748 vm_object_t backing_object;
750 VM_OBJECT_ASSERT_WLOCKED(object);
752 if ((object->flags & OBJ_SHADOWLIST) != 0) {
753 backing_object = object->backing_object;
754 VM_OBJECT_WLOCK(backing_object);
755 vm_object_backing_remove_locked(object);
756 VM_OBJECT_WUNLOCK(backing_object);
758 object->backing_object = NULL;
762 vm_object_backing_insert_locked(vm_object_t object, vm_object_t backing_object)
765 VM_OBJECT_ASSERT_WLOCKED(object);
767 if ((backing_object->flags & OBJ_ANON) != 0) {
768 VM_OBJECT_ASSERT_WLOCKED(backing_object);
769 LIST_INSERT_HEAD(&backing_object->shadow_head, object,
771 backing_object->shadow_count++;
772 object->flags |= OBJ_SHADOWLIST;
774 object->backing_object = backing_object;
778 vm_object_backing_insert(vm_object_t object, vm_object_t backing_object)
781 VM_OBJECT_ASSERT_WLOCKED(object);
783 if ((backing_object->flags & OBJ_ANON) != 0) {
784 VM_OBJECT_WLOCK(backing_object);
785 vm_object_backing_insert_locked(object, backing_object);
786 VM_OBJECT_WUNLOCK(backing_object);
788 object->backing_object = backing_object;
793 * vm_object_terminate_pages removes any remaining pageable pages
794 * from the object and resets the object to an empty state.
797 vm_object_terminate_pages(vm_object_t object)
801 VM_OBJECT_ASSERT_WLOCKED(object);
804 * Free any remaining pageable pages. This also removes them from the
805 * paging queues. However, don't free wired pages, just remove them
806 * from the object. Rather than incrementally removing each page from
807 * the object, the page and object are reset to any empty state.
809 TAILQ_FOREACH_SAFE(p, &object->memq, listq, p_next) {
810 vm_page_assert_unbusied(p);
811 KASSERT(p->object == object &&
812 (p->ref_count & VPRC_OBJREF) != 0,
813 ("vm_object_terminate_pages: page %p is inconsistent", p));
816 if (vm_page_drop(p, VPRC_OBJREF) == VPRC_OBJREF) {
823 * If the object contained any pages, then reset it to an empty state.
824 * None of the object's fields, including "resident_page_count", were
825 * modified by the preceding loop.
827 if (object->resident_page_count != 0) {
828 vm_radix_reclaim_allnodes(&object->rtree);
829 TAILQ_INIT(&object->memq);
830 object->resident_page_count = 0;
831 if (object->type == OBJT_VNODE)
832 vdrop(object->handle);
837 * vm_object_terminate actually destroys the specified object, freeing
838 * up all previously used resources.
840 * The object must be locked.
841 * This routine may block.
844 vm_object_terminate(vm_object_t object)
846 VM_OBJECT_ASSERT_WLOCKED(object);
847 KASSERT((object->flags & OBJ_DEAD) != 0,
848 ("terminating non-dead obj %p", object));
851 * wait for the pageout daemon to be done with the object
853 vm_object_pip_wait(object, "objtrm");
855 KASSERT(!REFCOUNT_COUNT(object->paging_in_progress),
856 ("vm_object_terminate: pageout in progress"));
858 KASSERT(object->ref_count == 0,
859 ("vm_object_terminate: object with references, ref_count=%d",
862 if ((object->flags & OBJ_PG_DTOR) == 0)
863 vm_object_terminate_pages(object);
865 #if VM_NRESERVLEVEL > 0
866 if (__predict_false(!LIST_EMPTY(&object->rvq)))
867 vm_reserv_break_all(object);
870 KASSERT(object->cred == NULL || object->type == OBJT_DEFAULT ||
871 object->type == OBJT_SWAP,
872 ("%s: non-swap obj %p has cred", __func__, object));
875 * Let the pager know object is dead.
877 vm_pager_deallocate(object);
878 VM_OBJECT_WUNLOCK(object);
880 vm_object_destroy(object);
884 * Make the page read-only so that we can clear the object flags. However, if
885 * this is a nosync mmap then the object is likely to stay dirty so do not
886 * mess with the page and do not clear the object flags. Returns TRUE if the
887 * page should be flushed, and FALSE otherwise.
890 vm_object_page_remove_write(vm_page_t p, int flags, boolean_t *allclean)
893 vm_page_assert_busied(p);
896 * If we have been asked to skip nosync pages and this is a
897 * nosync page, skip it. Note that the object flags were not
898 * cleared in this case so we do not have to set them.
900 if ((flags & OBJPC_NOSYNC) != 0 && (p->a.flags & PGA_NOSYNC) != 0) {
904 pmap_remove_write(p);
905 return (p->dirty != 0);
910 * vm_object_page_clean
912 * Clean all dirty pages in the specified range of object. Leaves page
913 * on whatever queue it is currently on. If NOSYNC is set then do not
914 * write out pages with PGA_NOSYNC set (originally comes from MAP_NOSYNC),
915 * leaving the object dirty.
917 * When stuffing pages asynchronously, allow clustering. XXX we need a
918 * synchronous clustering mode implementation.
920 * Odd semantics: if start == end, we clean everything.
922 * The object must be locked.
924 * Returns FALSE if some page from the range was not written, as
925 * reported by the pager, and TRUE otherwise.
928 vm_object_page_clean(vm_object_t object, vm_ooffset_t start, vm_ooffset_t end,
932 vm_pindex_t pi, tend, tstart;
933 int curgeneration, n, pagerflags;
934 boolean_t eio, res, allclean;
936 VM_OBJECT_ASSERT_WLOCKED(object);
938 if (object->type != OBJT_VNODE || !vm_object_mightbedirty(object) ||
939 object->resident_page_count == 0)
942 pagerflags = (flags & (OBJPC_SYNC | OBJPC_INVAL)) != 0 ?
943 VM_PAGER_PUT_SYNC : VM_PAGER_CLUSTER_OK;
944 pagerflags |= (flags & OBJPC_INVAL) != 0 ? VM_PAGER_PUT_INVAL : 0;
946 tstart = OFF_TO_IDX(start);
947 tend = (end == 0) ? object->size : OFF_TO_IDX(end + PAGE_MASK);
948 allclean = tstart == 0 && tend >= object->size;
952 curgeneration = object->generation;
954 for (p = vm_page_find_least(object, tstart); p != NULL; p = np) {
958 np = TAILQ_NEXT(p, listq);
959 if (vm_page_none_valid(p))
961 if (vm_page_busy_acquire(p, VM_ALLOC_WAITFAIL) == 0) {
962 if (object->generation != curgeneration &&
963 (flags & OBJPC_SYNC) != 0)
965 np = vm_page_find_least(object, pi);
968 if (!vm_object_page_remove_write(p, flags, &allclean)) {
973 n = vm_object_page_collect_flush(object, p, pagerflags,
974 flags, &allclean, &eio);
979 if (object->generation != curgeneration &&
980 (flags & OBJPC_SYNC) != 0)
984 * If the VOP_PUTPAGES() did a truncated write, so
985 * that even the first page of the run is not fully
986 * written, vm_pageout_flush() returns 0 as the run
987 * length. Since the condition that caused truncated
988 * write may be permanent, e.g. exhausted free space,
989 * accepting n == 0 would cause an infinite loop.
991 * Forwarding the iterator leaves the unwritten page
992 * behind, but there is not much we can do there if
993 * filesystem refuses to write it.
999 np = vm_page_find_least(object, pi + n);
1002 VOP_FSYNC(vp, (pagerflags & VM_PAGER_PUT_SYNC) ? MNT_WAIT : 0);
1006 object->cleangeneration = curgeneration;
1011 vm_object_page_collect_flush(vm_object_t object, vm_page_t p, int pagerflags,
1012 int flags, boolean_t *allclean, boolean_t *eio)
1014 vm_page_t ma[vm_pageout_page_count], p_first, tp;
1015 int count, i, mreq, runlen;
1017 vm_page_lock_assert(p, MA_NOTOWNED);
1018 vm_page_assert_xbusied(p);
1019 VM_OBJECT_ASSERT_WLOCKED(object);
1024 for (tp = p; count < vm_pageout_page_count; count++) {
1025 tp = vm_page_next(tp);
1026 if (tp == NULL || vm_page_tryxbusy(tp) == 0)
1028 if (!vm_object_page_remove_write(tp, flags, allclean)) {
1029 vm_page_xunbusy(tp);
1034 for (p_first = p; count < vm_pageout_page_count; count++) {
1035 tp = vm_page_prev(p_first);
1036 if (tp == NULL || vm_page_tryxbusy(tp) == 0)
1038 if (!vm_object_page_remove_write(tp, flags, allclean)) {
1039 vm_page_xunbusy(tp);
1046 for (tp = p_first, i = 0; i < count; tp = TAILQ_NEXT(tp, listq), i++)
1049 vm_pageout_flush(ma, count, pagerflags, mreq, &runlen, eio);
1054 * Note that there is absolutely no sense in writing out
1055 * anonymous objects, so we track down the vnode object
1057 * We invalidate (remove) all pages from the address space
1058 * for semantic correctness.
1060 * If the backing object is a device object with unmanaged pages, then any
1061 * mappings to the specified range of pages must be removed before this
1062 * function is called.
1064 * Note: certain anonymous maps, such as MAP_NOSYNC maps,
1065 * may start out with a NULL object.
1068 vm_object_sync(vm_object_t object, vm_ooffset_t offset, vm_size_t size,
1069 boolean_t syncio, boolean_t invalidate)
1071 vm_object_t backing_object;
1074 int error, flags, fsync_after;
1081 VM_OBJECT_WLOCK(object);
1082 while ((backing_object = object->backing_object) != NULL) {
1083 VM_OBJECT_WLOCK(backing_object);
1084 offset += object->backing_object_offset;
1085 VM_OBJECT_WUNLOCK(object);
1086 object = backing_object;
1087 if (object->size < OFF_TO_IDX(offset + size))
1088 size = IDX_TO_OFF(object->size) - offset;
1091 * Flush pages if writing is allowed, invalidate them
1092 * if invalidation requested. Pages undergoing I/O
1093 * will be ignored by vm_object_page_remove().
1095 * We cannot lock the vnode and then wait for paging
1096 * to complete without deadlocking against vm_fault.
1097 * Instead we simply call vm_object_page_remove() and
1098 * allow it to block internally on a page-by-page
1099 * basis when it encounters pages undergoing async
1102 if (object->type == OBJT_VNODE &&
1103 vm_object_mightbedirty(object) != 0 &&
1104 ((vp = object->handle)->v_vflag & VV_NOSYNC) == 0) {
1105 VM_OBJECT_WUNLOCK(object);
1106 (void) vn_start_write(vp, &mp, V_WAIT);
1107 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1108 if (syncio && !invalidate && offset == 0 &&
1109 atop(size) == object->size) {
1111 * If syncing the whole mapping of the file,
1112 * it is faster to schedule all the writes in
1113 * async mode, also allowing the clustering,
1114 * and then wait for i/o to complete.
1119 flags = (syncio || invalidate) ? OBJPC_SYNC : 0;
1120 flags |= invalidate ? (OBJPC_SYNC | OBJPC_INVAL) : 0;
1121 fsync_after = FALSE;
1123 VM_OBJECT_WLOCK(object);
1124 res = vm_object_page_clean(object, offset, offset + size,
1126 VM_OBJECT_WUNLOCK(object);
1128 error = VOP_FSYNC(vp, MNT_WAIT, curthread);
1130 vn_finished_write(mp);
1133 VM_OBJECT_WLOCK(object);
1135 if ((object->type == OBJT_VNODE ||
1136 object->type == OBJT_DEVICE) && invalidate) {
1137 if (object->type == OBJT_DEVICE)
1139 * The option OBJPR_NOTMAPPED must be passed here
1140 * because vm_object_page_remove() cannot remove
1141 * unmanaged mappings.
1143 flags = OBJPR_NOTMAPPED;
1147 flags = OBJPR_CLEANONLY;
1148 vm_object_page_remove(object, OFF_TO_IDX(offset),
1149 OFF_TO_IDX(offset + size + PAGE_MASK), flags);
1151 VM_OBJECT_WUNLOCK(object);
1156 * Determine whether the given advice can be applied to the object. Advice is
1157 * not applied to unmanaged pages since they never belong to page queues, and
1158 * since MADV_FREE is destructive, it can apply only to anonymous pages that
1159 * have been mapped at most once.
1162 vm_object_advice_applies(vm_object_t object, int advice)
1165 if ((object->flags & OBJ_UNMANAGED) != 0)
1167 if (advice != MADV_FREE)
1169 return ((object->flags & (OBJ_ONEMAPPING | OBJ_ANON)) ==
1170 (OBJ_ONEMAPPING | OBJ_ANON));
1174 vm_object_madvise_freespace(vm_object_t object, int advice, vm_pindex_t pindex,
1178 if (advice == MADV_FREE && object->type == OBJT_SWAP)
1179 swap_pager_freespace(object, pindex, size);
1183 * vm_object_madvise:
1185 * Implements the madvise function at the object/page level.
1187 * MADV_WILLNEED (any object)
1189 * Activate the specified pages if they are resident.
1191 * MADV_DONTNEED (any object)
1193 * Deactivate the specified pages if they are resident.
1195 * MADV_FREE (OBJT_DEFAULT/OBJT_SWAP objects,
1196 * OBJ_ONEMAPPING only)
1198 * Deactivate and clean the specified pages if they are
1199 * resident. This permits the process to reuse the pages
1200 * without faulting or the kernel to reclaim the pages
1204 vm_object_madvise(vm_object_t object, vm_pindex_t pindex, vm_pindex_t end,
1207 vm_pindex_t tpindex;
1208 vm_object_t backing_object, tobject;
1215 VM_OBJECT_WLOCK(object);
1216 if (!vm_object_advice_applies(object, advice)) {
1217 VM_OBJECT_WUNLOCK(object);
1220 for (m = vm_page_find_least(object, pindex); pindex < end; pindex++) {
1224 * If the next page isn't resident in the top-level object, we
1225 * need to search the shadow chain. When applying MADV_FREE, we
1226 * take care to release any swap space used to store
1227 * non-resident pages.
1229 if (m == NULL || pindex < m->pindex) {
1231 * Optimize a common case: if the top-level object has
1232 * no backing object, we can skip over the non-resident
1233 * range in constant time.
1235 if (object->backing_object == NULL) {
1236 tpindex = (m != NULL && m->pindex < end) ?
1238 vm_object_madvise_freespace(object, advice,
1239 pindex, tpindex - pindex);
1240 if ((pindex = tpindex) == end)
1247 vm_object_madvise_freespace(tobject, advice,
1250 * Prepare to search the next object in the
1253 backing_object = tobject->backing_object;
1254 if (backing_object == NULL)
1256 VM_OBJECT_WLOCK(backing_object);
1258 OFF_TO_IDX(tobject->backing_object_offset);
1259 if (tobject != object)
1260 VM_OBJECT_WUNLOCK(tobject);
1261 tobject = backing_object;
1262 if (!vm_object_advice_applies(tobject, advice))
1264 } while ((tm = vm_page_lookup(tobject, tpindex)) ==
1269 m = TAILQ_NEXT(m, listq);
1273 * If the page is not in a normal state, skip it. The page
1274 * can not be invalidated while the object lock is held.
1276 if (!vm_page_all_valid(tm) || vm_page_wired(tm))
1278 KASSERT((tm->flags & PG_FICTITIOUS) == 0,
1279 ("vm_object_madvise: page %p is fictitious", tm));
1280 KASSERT((tm->oflags & VPO_UNMANAGED) == 0,
1281 ("vm_object_madvise: page %p is not managed", tm));
1282 if (vm_page_tryxbusy(tm) == 0) {
1283 if (object != tobject)
1284 VM_OBJECT_WUNLOCK(object);
1285 if (advice == MADV_WILLNEED) {
1287 * Reference the page before unlocking and
1288 * sleeping so that the page daemon is less
1289 * likely to reclaim it.
1291 vm_page_aflag_set(tm, PGA_REFERENCED);
1293 vm_page_busy_sleep(tm, "madvpo", false);
1297 vm_page_advise(tm, advice);
1299 vm_page_xunbusy(tm);
1300 vm_object_madvise_freespace(tobject, advice, tm->pindex, 1);
1302 if (tobject != object)
1303 VM_OBJECT_WUNLOCK(tobject);
1305 VM_OBJECT_WUNLOCK(object);
1311 * Create a new object which is backed by the
1312 * specified existing object range. The source
1313 * object reference is deallocated.
1315 * The new object and offset into that object
1316 * are returned in the source parameters.
1319 vm_object_shadow(vm_object_t *object, vm_ooffset_t *offset, vm_size_t length,
1320 struct ucred *cred, bool shared)
1328 * Don't create the new object if the old object isn't shared.
1330 * If we hold the only reference we can guarantee that it won't
1331 * increase while we have the map locked. Otherwise the race is
1332 * harmless and we will end up with an extra shadow object that
1333 * will be collapsed later.
1335 if (source != NULL && source->ref_count == 1 &&
1336 (source->flags & OBJ_ANON) != 0)
1340 * Allocate a new object with the given length.
1342 result = vm_object_allocate_anon(atop(length), source, cred, length);
1345 * Store the offset into the source object, and fix up the offset into
1348 result->backing_object_offset = *offset;
1350 if (shared || source != NULL) {
1351 VM_OBJECT_WLOCK(result);
1354 * The new object shadows the source object, adding a
1355 * reference to it. Our caller changes his reference
1356 * to point to the new object, removing a reference to
1357 * the source object. Net result: no change of
1358 * reference count, unless the caller needs to add one
1359 * more reference due to forking a shared map entry.
1362 vm_object_reference_locked(result);
1363 vm_object_clear_flag(result, OBJ_ONEMAPPING);
1367 * Try to optimize the result object's page color when
1368 * shadowing in order to maintain page coloring
1369 * consistency in the combined shadowed object.
1371 if (source != NULL) {
1372 vm_object_backing_insert(result, source);
1373 result->domain = source->domain;
1374 #if VM_NRESERVLEVEL > 0
1375 result->flags |= source->flags & OBJ_COLORED;
1376 result->pg_color = (source->pg_color +
1377 OFF_TO_IDX(*offset)) & ((1 << (VM_NFREEORDER -
1381 VM_OBJECT_WUNLOCK(result);
1385 * Return the new things
1394 * Split the pages in a map entry into a new object. This affords
1395 * easier removal of unused pages, and keeps object inheritance from
1396 * being a negative impact on memory usage.
1399 vm_object_split(vm_map_entry_t entry)
1401 vm_page_t m, m_next;
1402 vm_object_t orig_object, new_object, source;
1403 vm_pindex_t idx, offidxstart;
1406 orig_object = entry->object.vm_object;
1407 if ((orig_object->flags & OBJ_ANON) == 0)
1409 if (orig_object->ref_count <= 1)
1411 VM_OBJECT_WUNLOCK(orig_object);
1413 offidxstart = OFF_TO_IDX(entry->offset);
1414 size = atop(entry->end - entry->start);
1417 * If swap_pager_copy() is later called, it will convert new_object
1418 * into a swap object.
1420 new_object = vm_object_allocate_anon(size, orig_object,
1421 orig_object->cred, ptoa(size));
1424 * At this point, the new object is still private, so the order in
1425 * which the original and new objects are locked does not matter.
1427 VM_OBJECT_WLOCK(new_object);
1428 VM_OBJECT_WLOCK(orig_object);
1429 new_object->domain = orig_object->domain;
1430 source = orig_object->backing_object;
1431 if (source != NULL) {
1432 if ((source->flags & (OBJ_ANON | OBJ_DEAD)) != 0) {
1433 VM_OBJECT_WLOCK(source);
1434 if ((source->flags & OBJ_DEAD) != 0) {
1435 VM_OBJECT_WUNLOCK(source);
1436 VM_OBJECT_WUNLOCK(orig_object);
1437 VM_OBJECT_WUNLOCK(new_object);
1438 new_object->cred = NULL;
1439 vm_object_deallocate(new_object);
1440 VM_OBJECT_WLOCK(orig_object);
1443 vm_object_backing_insert_locked(new_object, source);
1444 vm_object_reference_locked(source); /* for new_object */
1445 vm_object_clear_flag(source, OBJ_ONEMAPPING);
1446 VM_OBJECT_WUNLOCK(source);
1448 vm_object_backing_insert(new_object, source);
1449 vm_object_reference(source);
1451 new_object->backing_object_offset =
1452 orig_object->backing_object_offset + entry->offset;
1454 if (orig_object->cred != NULL) {
1455 crhold(orig_object->cred);
1456 KASSERT(orig_object->charge >= ptoa(size),
1457 ("orig_object->charge < 0"));
1458 orig_object->charge -= ptoa(size);
1461 m = vm_page_find_least(orig_object, offidxstart);
1462 for (; m != NULL && (idx = m->pindex - offidxstart) < size;
1464 m_next = TAILQ_NEXT(m, listq);
1467 * We must wait for pending I/O to complete before we can
1470 * We do not have to VM_PROT_NONE the page as mappings should
1471 * not be changed by this operation.
1473 if (vm_page_tryxbusy(m) == 0) {
1474 VM_OBJECT_WUNLOCK(new_object);
1475 vm_page_sleep_if_busy(m, "spltwt");
1476 VM_OBJECT_WLOCK(new_object);
1480 /* vm_page_rename() will dirty the page. */
1481 if (vm_page_rename(m, new_object, idx)) {
1483 VM_OBJECT_WUNLOCK(new_object);
1484 VM_OBJECT_WUNLOCK(orig_object);
1486 VM_OBJECT_WLOCK(orig_object);
1487 VM_OBJECT_WLOCK(new_object);
1491 #if VM_NRESERVLEVEL > 0
1493 * If some of the reservation's allocated pages remain with
1494 * the original object, then transferring the reservation to
1495 * the new object is neither particularly beneficial nor
1496 * particularly harmful as compared to leaving the reservation
1497 * with the original object. If, however, all of the
1498 * reservation's allocated pages are transferred to the new
1499 * object, then transferring the reservation is typically
1500 * beneficial. Determining which of these two cases applies
1501 * would be more costly than unconditionally renaming the
1504 vm_reserv_rename(m, new_object, orig_object, offidxstart);
1506 if (orig_object->type != OBJT_SWAP)
1509 if (orig_object->type == OBJT_SWAP) {
1511 * swap_pager_copy() can sleep, in which case the orig_object's
1512 * and new_object's locks are released and reacquired.
1514 swap_pager_copy(orig_object, new_object, offidxstart, 0);
1515 TAILQ_FOREACH(m, &new_object->memq, listq)
1518 VM_OBJECT_WUNLOCK(orig_object);
1519 VM_OBJECT_WUNLOCK(new_object);
1520 entry->object.vm_object = new_object;
1521 entry->offset = 0LL;
1522 vm_object_deallocate(orig_object);
1523 VM_OBJECT_WLOCK(new_object);
1526 #define OBSC_COLLAPSE_NOWAIT 0x0002
1527 #define OBSC_COLLAPSE_WAIT 0x0004
1530 vm_object_collapse_scan_wait(vm_object_t object, vm_page_t p, vm_page_t next,
1533 vm_object_t backing_object;
1535 VM_OBJECT_ASSERT_WLOCKED(object);
1536 backing_object = object->backing_object;
1537 VM_OBJECT_ASSERT_WLOCKED(backing_object);
1539 KASSERT(p == NULL || p->object == object || p->object == backing_object,
1540 ("invalid ownership %p %p %p", p, object, backing_object));
1541 if ((op & OBSC_COLLAPSE_NOWAIT) != 0)
1543 /* The page is only NULL when rename fails. */
1545 VM_OBJECT_WUNLOCK(object);
1546 VM_OBJECT_WUNLOCK(backing_object);
1549 if (p->object == object)
1550 VM_OBJECT_WUNLOCK(backing_object);
1552 VM_OBJECT_WUNLOCK(object);
1553 vm_page_busy_sleep(p, "vmocol", false);
1555 VM_OBJECT_WLOCK(object);
1556 VM_OBJECT_WLOCK(backing_object);
1557 return (TAILQ_FIRST(&backing_object->memq));
1561 vm_object_scan_all_shadowed(vm_object_t object)
1563 vm_object_t backing_object;
1565 vm_pindex_t backing_offset_index, new_pindex, pi, ps;
1567 VM_OBJECT_ASSERT_WLOCKED(object);
1568 VM_OBJECT_ASSERT_WLOCKED(object->backing_object);
1570 backing_object = object->backing_object;
1572 if ((backing_object->flags & OBJ_ANON) == 0)
1575 pi = backing_offset_index = OFF_TO_IDX(object->backing_object_offset);
1576 p = vm_page_find_least(backing_object, pi);
1577 ps = swap_pager_find_least(backing_object, pi);
1580 * Only check pages inside the parent object's range and
1581 * inside the parent object's mapping of the backing object.
1584 if (p != NULL && p->pindex < pi)
1585 p = TAILQ_NEXT(p, listq);
1587 ps = swap_pager_find_least(backing_object, pi);
1588 if (p == NULL && ps >= backing_object->size)
1593 pi = MIN(p->pindex, ps);
1595 new_pindex = pi - backing_offset_index;
1596 if (new_pindex >= object->size)
1600 * See if the parent has the page or if the parent's object
1601 * pager has the page. If the parent has the page but the page
1602 * is not valid, the parent's object pager must have the page.
1604 * If this fails, the parent does not completely shadow the
1605 * object and we might as well give up now.
1607 pp = vm_page_lookup(object, new_pindex);
1609 * The valid check here is stable due to object lock being
1610 * required to clear valid and initiate paging.
1612 if ((pp == NULL || vm_page_none_valid(pp)) &&
1613 !vm_pager_has_page(object, new_pindex, NULL, NULL))
1620 vm_object_collapse_scan(vm_object_t object, int op)
1622 vm_object_t backing_object;
1623 vm_page_t next, p, pp;
1624 vm_pindex_t backing_offset_index, new_pindex;
1626 VM_OBJECT_ASSERT_WLOCKED(object);
1627 VM_OBJECT_ASSERT_WLOCKED(object->backing_object);
1629 backing_object = object->backing_object;
1630 backing_offset_index = OFF_TO_IDX(object->backing_object_offset);
1633 * Initial conditions
1635 if ((op & OBSC_COLLAPSE_WAIT) != 0)
1636 vm_object_set_flag(backing_object, OBJ_DEAD);
1641 for (p = TAILQ_FIRST(&backing_object->memq); p != NULL; p = next) {
1642 next = TAILQ_NEXT(p, listq);
1643 new_pindex = p->pindex - backing_offset_index;
1646 * Check for busy page
1648 if (vm_page_tryxbusy(p) == 0) {
1649 next = vm_object_collapse_scan_wait(object, p, next, op);
1653 KASSERT(p->object == backing_object,
1654 ("vm_object_collapse_scan: object mismatch"));
1656 if (p->pindex < backing_offset_index ||
1657 new_pindex >= object->size) {
1658 if (backing_object->type == OBJT_SWAP)
1659 swap_pager_freespace(backing_object, p->pindex,
1662 KASSERT(!pmap_page_is_mapped(p),
1663 ("freeing mapped page %p", p));
1664 if (vm_page_remove(p))
1671 pp = vm_page_lookup(object, new_pindex);
1672 if (pp != NULL && vm_page_tryxbusy(pp) == 0) {
1675 * The page in the parent is busy and possibly not
1676 * (yet) valid. Until its state is finalized by the
1677 * busy bit owner, we can't tell whether it shadows the
1678 * original page. Therefore, we must either skip it
1679 * and the original (backing_object) page or wait for
1680 * its state to be finalized.
1682 * This is due to a race with vm_fault() where we must
1683 * unbusy the original (backing_obj) page before we can
1684 * (re)lock the parent. Hence we can get here.
1686 next = vm_object_collapse_scan_wait(object, pp, next,
1691 KASSERT(pp == NULL || !vm_page_none_valid(pp),
1692 ("unbusy invalid page %p", pp));
1694 if (pp != NULL || vm_pager_has_page(object, new_pindex, NULL,
1697 * The page already exists in the parent OR swap exists
1698 * for this location in the parent. Leave the parent's
1699 * page alone. Destroy the original page from the
1702 if (backing_object->type == OBJT_SWAP)
1703 swap_pager_freespace(backing_object, p->pindex,
1705 KASSERT(!pmap_page_is_mapped(p),
1706 ("freeing mapped page %p", p));
1707 if (vm_page_remove(p))
1712 vm_page_xunbusy(pp);
1717 * Page does not exist in parent, rename the page from the
1718 * backing object to the main object.
1720 * If the page was mapped to a process, it can remain mapped
1721 * through the rename. vm_page_rename() will dirty the page.
1723 if (vm_page_rename(p, object, new_pindex)) {
1726 vm_page_xunbusy(pp);
1727 next = vm_object_collapse_scan_wait(object, NULL, next,
1732 /* Use the old pindex to free the right page. */
1733 if (backing_object->type == OBJT_SWAP)
1734 swap_pager_freespace(backing_object,
1735 new_pindex + backing_offset_index, 1);
1737 #if VM_NRESERVLEVEL > 0
1739 * Rename the reservation.
1741 vm_reserv_rename(p, object, backing_object,
1742 backing_offset_index);
1751 * this version of collapse allows the operation to occur earlier and
1752 * when paging_in_progress is true for an object... This is not a complete
1753 * operation, but should plug 99.9% of the rest of the leaks.
1756 vm_object_qcollapse(vm_object_t object)
1758 vm_object_t backing_object = object->backing_object;
1760 VM_OBJECT_ASSERT_WLOCKED(object);
1761 VM_OBJECT_ASSERT_WLOCKED(backing_object);
1763 if (backing_object->ref_count != 1)
1766 vm_object_collapse_scan(object, OBSC_COLLAPSE_NOWAIT);
1770 * vm_object_collapse:
1772 * Collapse an object with the object backing it.
1773 * Pages in the backing object are moved into the
1774 * parent, and the backing object is deallocated.
1777 vm_object_collapse(vm_object_t object)
1779 vm_object_t backing_object, new_backing_object;
1781 VM_OBJECT_ASSERT_WLOCKED(object);
1785 * Verify that the conditions are right for collapse:
1787 * The object exists and the backing object exists.
1789 if ((backing_object = object->backing_object) == NULL)
1793 * we check the backing object first, because it is most likely
1796 if ((backing_object->flags & OBJ_ANON) == 0)
1798 VM_OBJECT_WLOCK(backing_object);
1799 if ((backing_object->flags & OBJ_DEAD) != 0 ||
1800 (object->flags & (OBJ_DEAD | OBJ_ANON)) != OBJ_ANON) {
1801 VM_OBJECT_WUNLOCK(backing_object);
1805 if (REFCOUNT_COUNT(object->paging_in_progress) > 0 ||
1806 REFCOUNT_COUNT(backing_object->paging_in_progress) > 0) {
1807 vm_object_qcollapse(object);
1808 VM_OBJECT_WUNLOCK(backing_object);
1813 * We know that we can either collapse the backing object (if
1814 * the parent is the only reference to it) or (perhaps) have
1815 * the parent bypass the object if the parent happens to shadow
1816 * all the resident pages in the entire backing object.
1818 * This is ignoring pager-backed pages such as swap pages.
1819 * vm_object_collapse_scan fails the shadowing test in this
1822 if (backing_object->ref_count == 1) {
1823 vm_object_pip_add(object, 1);
1824 vm_object_pip_add(backing_object, 1);
1827 * If there is exactly one reference to the backing
1828 * object, we can collapse it into the parent.
1830 vm_object_collapse_scan(object, OBSC_COLLAPSE_WAIT);
1832 #if VM_NRESERVLEVEL > 0
1834 * Break any reservations from backing_object.
1836 if (__predict_false(!LIST_EMPTY(&backing_object->rvq)))
1837 vm_reserv_break_all(backing_object);
1841 * Move the pager from backing_object to object.
1843 if (backing_object->type == OBJT_SWAP) {
1845 * swap_pager_copy() can sleep, in which case
1846 * the backing_object's and object's locks are
1847 * released and reacquired.
1848 * Since swap_pager_copy() is being asked to
1849 * destroy the source, it will change the
1850 * backing_object's type to OBJT_DEFAULT.
1855 OFF_TO_IDX(object->backing_object_offset), TRUE);
1858 * Object now shadows whatever backing_object did.
1859 * Note that the reference to
1860 * backing_object->backing_object moves from within
1861 * backing_object to within object.
1863 vm_object_backing_remove_locked(object);
1864 new_backing_object = backing_object->backing_object;
1865 if (new_backing_object != NULL) {
1866 VM_OBJECT_WLOCK(new_backing_object);
1867 vm_object_backing_remove_locked(backing_object);
1868 vm_object_backing_insert_locked(object,
1869 new_backing_object);
1870 VM_OBJECT_WUNLOCK(new_backing_object);
1872 object->backing_object_offset +=
1873 backing_object->backing_object_offset;
1876 * Discard backing_object.
1878 * Since the backing object has no pages, no pager left,
1879 * and no object references within it, all that is
1880 * necessary is to dispose of it.
1882 KASSERT(backing_object->ref_count == 1, (
1883 "backing_object %p was somehow re-referenced during collapse!",
1885 vm_object_pip_wakeup(backing_object);
1886 backing_object->type = OBJT_DEAD;
1887 refcount_release(&backing_object->ref_count);
1888 VM_OBJECT_WUNLOCK(backing_object);
1889 vm_object_destroy(backing_object);
1891 vm_object_pip_wakeup(object);
1892 counter_u64_add(object_collapses, 1);
1895 * If we do not entirely shadow the backing object,
1896 * there is nothing we can do so we give up.
1898 if (object->resident_page_count != object->size &&
1899 !vm_object_scan_all_shadowed(object)) {
1900 VM_OBJECT_WUNLOCK(backing_object);
1905 * Make the parent shadow the next object in the
1906 * chain. Deallocating backing_object will not remove
1907 * it, since its reference count is at least 2.
1909 vm_object_backing_remove_locked(object);
1911 new_backing_object = backing_object->backing_object;
1912 if (new_backing_object != NULL) {
1913 vm_object_backing_insert(object,
1914 new_backing_object);
1915 vm_object_reference(new_backing_object);
1916 object->backing_object_offset +=
1917 backing_object->backing_object_offset;
1921 * Drop the reference count on backing_object. Since
1922 * its ref_count was at least 2, it will not vanish.
1924 refcount_release(&backing_object->ref_count);
1925 VM_OBJECT_WUNLOCK(backing_object);
1926 counter_u64_add(object_bypasses, 1);
1930 * Try again with this object's new backing object.
1936 * vm_object_page_remove:
1938 * For the given object, either frees or invalidates each of the
1939 * specified pages. In general, a page is freed. However, if a page is
1940 * wired for any reason other than the existence of a managed, wired
1941 * mapping, then it may be invalidated but not removed from the object.
1942 * Pages are specified by the given range ["start", "end") and the option
1943 * OBJPR_CLEANONLY. As a special case, if "end" is zero, then the range
1944 * extends from "start" to the end of the object. If the option
1945 * OBJPR_CLEANONLY is specified, then only the non-dirty pages within the
1946 * specified range are affected. If the option OBJPR_NOTMAPPED is
1947 * specified, then the pages within the specified range must have no
1948 * mappings. Otherwise, if this option is not specified, any mappings to
1949 * the specified pages are removed before the pages are freed or
1952 * In general, this operation should only be performed on objects that
1953 * contain managed pages. There are, however, two exceptions. First, it
1954 * is performed on the kernel and kmem objects by vm_map_entry_delete().
1955 * Second, it is used by msync(..., MS_INVALIDATE) to invalidate device-
1956 * backed pages. In both of these cases, the option OBJPR_CLEANONLY must
1957 * not be specified and the option OBJPR_NOTMAPPED must be specified.
1959 * The object must be locked.
1962 vm_object_page_remove(vm_object_t object, vm_pindex_t start, vm_pindex_t end,
1967 VM_OBJECT_ASSERT_WLOCKED(object);
1968 KASSERT((object->flags & OBJ_UNMANAGED) == 0 ||
1969 (options & (OBJPR_CLEANONLY | OBJPR_NOTMAPPED)) == OBJPR_NOTMAPPED,
1970 ("vm_object_page_remove: illegal options for object %p", object));
1971 if (object->resident_page_count == 0)
1973 vm_object_pip_add(object, 1);
1975 p = vm_page_find_least(object, start);
1978 * Here, the variable "p" is either (1) the page with the least pindex
1979 * greater than or equal to the parameter "start" or (2) NULL.
1981 for (; p != NULL && (p->pindex < end || end == 0); p = next) {
1982 next = TAILQ_NEXT(p, listq);
1985 * If the page is wired for any reason besides the existence
1986 * of managed, wired mappings, then it cannot be freed. For
1987 * example, fictitious pages, which represent device memory,
1988 * are inherently wired and cannot be freed. They can,
1989 * however, be invalidated if the option OBJPR_CLEANONLY is
1992 if (vm_page_tryxbusy(p) == 0) {
1993 vm_page_sleep_if_busy(p, "vmopar");
1996 if (vm_page_wired(p)) {
1998 if ((options & OBJPR_NOTMAPPED) == 0 &&
1999 object->ref_count != 0)
2001 if ((options & OBJPR_CLEANONLY) == 0) {
2008 KASSERT((p->flags & PG_FICTITIOUS) == 0,
2009 ("vm_object_page_remove: page %p is fictitious", p));
2010 if ((options & OBJPR_CLEANONLY) != 0 &&
2011 !vm_page_none_valid(p)) {
2012 if ((options & OBJPR_NOTMAPPED) == 0 &&
2013 object->ref_count != 0 &&
2014 !vm_page_try_remove_write(p))
2016 if (p->dirty != 0) {
2021 if ((options & OBJPR_NOTMAPPED) == 0 &&
2022 object->ref_count != 0 && !vm_page_try_remove_all(p))
2026 vm_object_pip_wakeup(object);
2030 * vm_object_page_noreuse:
2032 * For the given object, attempt to move the specified pages to
2033 * the head of the inactive queue. This bypasses regular LRU
2034 * operation and allows the pages to be reused quickly under memory
2035 * pressure. If a page is wired for any reason, then it will not
2036 * be queued. Pages are specified by the range ["start", "end").
2037 * As a special case, if "end" is zero, then the range extends from
2038 * "start" to the end of the object.
2040 * This operation should only be performed on objects that
2041 * contain non-fictitious, managed pages.
2043 * The object must be locked.
2046 vm_object_page_noreuse(vm_object_t object, vm_pindex_t start, vm_pindex_t end)
2051 VM_OBJECT_ASSERT_LOCKED(object);
2052 KASSERT((object->flags & (OBJ_FICTITIOUS | OBJ_UNMANAGED)) == 0,
2053 ("vm_object_page_noreuse: illegal object %p", object));
2054 if (object->resident_page_count == 0)
2056 p = vm_page_find_least(object, start);
2059 * Here, the variable "p" is either (1) the page with the least pindex
2060 * greater than or equal to the parameter "start" or (2) NULL.
2063 for (; p != NULL && (p->pindex < end || end == 0); p = next) {
2064 next = TAILQ_NEXT(p, listq);
2065 vm_page_change_lock(p, &mtx);
2066 vm_page_deactivate_noreuse(p);
2073 * Populate the specified range of the object with valid pages. Returns
2074 * TRUE if the range is successfully populated and FALSE otherwise.
2076 * Note: This function should be optimized to pass a larger array of
2077 * pages to vm_pager_get_pages() before it is applied to a non-
2078 * OBJT_DEVICE object.
2080 * The object must be locked.
2083 vm_object_populate(vm_object_t object, vm_pindex_t start, vm_pindex_t end)
2089 VM_OBJECT_ASSERT_WLOCKED(object);
2090 for (pindex = start; pindex < end; pindex++) {
2091 rv = vm_page_grab_valid(&m, object, pindex, VM_ALLOC_NORMAL);
2092 if (rv != VM_PAGER_OK)
2096 * Keep "m" busy because a subsequent iteration may unlock
2100 if (pindex > start) {
2101 m = vm_page_lookup(object, start);
2102 while (m != NULL && m->pindex < pindex) {
2104 m = TAILQ_NEXT(m, listq);
2107 return (pindex == end);
2111 * Routine: vm_object_coalesce
2112 * Function: Coalesces two objects backing up adjoining
2113 * regions of memory into a single object.
2115 * returns TRUE if objects were combined.
2117 * NOTE: Only works at the moment if the second object is NULL -
2118 * if it's not, which object do we lock first?
2121 * prev_object First object to coalesce
2122 * prev_offset Offset into prev_object
2123 * prev_size Size of reference to prev_object
2124 * next_size Size of reference to the second object
2125 * reserved Indicator that extension region has
2126 * swap accounted for
2129 * The object must *not* be locked.
2132 vm_object_coalesce(vm_object_t prev_object, vm_ooffset_t prev_offset,
2133 vm_size_t prev_size, vm_size_t next_size, boolean_t reserved)
2135 vm_pindex_t next_pindex;
2137 if (prev_object == NULL)
2139 if ((prev_object->flags & OBJ_ANON) == 0)
2142 VM_OBJECT_WLOCK(prev_object);
2144 * Try to collapse the object first
2146 vm_object_collapse(prev_object);
2149 * Can't coalesce if: . more than one reference . paged out . shadows
2150 * another object . has a copy elsewhere (any of which mean that the
2151 * pages not mapped to prev_entry may be in use anyway)
2153 if (prev_object->backing_object != NULL) {
2154 VM_OBJECT_WUNLOCK(prev_object);
2158 prev_size >>= PAGE_SHIFT;
2159 next_size >>= PAGE_SHIFT;
2160 next_pindex = OFF_TO_IDX(prev_offset) + prev_size;
2162 if (prev_object->ref_count > 1 &&
2163 prev_object->size != next_pindex &&
2164 (prev_object->flags & OBJ_ONEMAPPING) == 0) {
2165 VM_OBJECT_WUNLOCK(prev_object);
2170 * Account for the charge.
2172 if (prev_object->cred != NULL) {
2175 * If prev_object was charged, then this mapping,
2176 * although not charged now, may become writable
2177 * later. Non-NULL cred in the object would prevent
2178 * swap reservation during enabling of the write
2179 * access, so reserve swap now. Failed reservation
2180 * cause allocation of the separate object for the map
2181 * entry, and swap reservation for this entry is
2182 * managed in appropriate time.
2184 if (!reserved && !swap_reserve_by_cred(ptoa(next_size),
2185 prev_object->cred)) {
2186 VM_OBJECT_WUNLOCK(prev_object);
2189 prev_object->charge += ptoa(next_size);
2193 * Remove any pages that may still be in the object from a previous
2196 if (next_pindex < prev_object->size) {
2197 vm_object_page_remove(prev_object, next_pindex, next_pindex +
2199 if (prev_object->type == OBJT_SWAP)
2200 swap_pager_freespace(prev_object,
2201 next_pindex, next_size);
2203 if (prev_object->cred != NULL) {
2204 KASSERT(prev_object->charge >=
2205 ptoa(prev_object->size - next_pindex),
2206 ("object %p overcharged 1 %jx %jx", prev_object,
2207 (uintmax_t)next_pindex, (uintmax_t)next_size));
2208 prev_object->charge -= ptoa(prev_object->size -
2215 * Extend the object if necessary.
2217 if (next_pindex + next_size > prev_object->size)
2218 prev_object->size = next_pindex + next_size;
2220 VM_OBJECT_WUNLOCK(prev_object);
2225 vm_object_set_writeable_dirty(vm_object_t object)
2228 /* Only set for vnodes & tmpfs */
2229 if (object->type != OBJT_VNODE &&
2230 (object->flags & OBJ_TMPFS_NODE) == 0)
2232 atomic_add_int(&object->generation, 1);
2238 * For each page offset within the specified range of the given object,
2239 * find the highest-level page in the shadow chain and unwire it. A page
2240 * must exist at every page offset, and the highest-level page must be
2244 vm_object_unwire(vm_object_t object, vm_ooffset_t offset, vm_size_t length,
2247 vm_object_t tobject, t1object;
2249 vm_pindex_t end_pindex, pindex, tpindex;
2250 int depth, locked_depth;
2252 KASSERT((offset & PAGE_MASK) == 0,
2253 ("vm_object_unwire: offset is not page aligned"));
2254 KASSERT((length & PAGE_MASK) == 0,
2255 ("vm_object_unwire: length is not a multiple of PAGE_SIZE"));
2256 /* The wired count of a fictitious page never changes. */
2257 if ((object->flags & OBJ_FICTITIOUS) != 0)
2259 pindex = OFF_TO_IDX(offset);
2260 end_pindex = pindex + atop(length);
2263 VM_OBJECT_RLOCK(object);
2264 m = vm_page_find_least(object, pindex);
2265 while (pindex < end_pindex) {
2266 if (m == NULL || pindex < m->pindex) {
2268 * The first object in the shadow chain doesn't
2269 * contain a page at the current index. Therefore,
2270 * the page must exist in a backing object.
2277 OFF_TO_IDX(tobject->backing_object_offset);
2278 tobject = tobject->backing_object;
2279 KASSERT(tobject != NULL,
2280 ("vm_object_unwire: missing page"));
2281 if ((tobject->flags & OBJ_FICTITIOUS) != 0)
2284 if (depth == locked_depth) {
2286 VM_OBJECT_RLOCK(tobject);
2288 } while ((tm = vm_page_lookup(tobject, tpindex)) ==
2292 m = TAILQ_NEXT(m, listq);
2294 if (vm_page_trysbusy(tm) == 0) {
2295 for (tobject = object; locked_depth >= 1;
2297 t1object = tobject->backing_object;
2298 if (tm->object != tobject)
2299 VM_OBJECT_RUNLOCK(tobject);
2302 vm_page_busy_sleep(tm, "unwbo", true);
2305 vm_page_unwire(tm, queue);
2306 vm_page_sunbusy(tm);
2310 /* Release the accumulated object locks. */
2311 for (tobject = object; locked_depth >= 1; locked_depth--) {
2312 t1object = tobject->backing_object;
2313 VM_OBJECT_RUNLOCK(tobject);
2319 * Return the vnode for the given object, or NULL if none exists.
2320 * For tmpfs objects, the function may return NULL if there is
2321 * no vnode allocated at the time of the call.
2324 vm_object_vnode(vm_object_t object)
2328 VM_OBJECT_ASSERT_LOCKED(object);
2329 if (object->type == OBJT_VNODE) {
2330 vp = object->handle;
2331 KASSERT(vp != NULL, ("%s: OBJT_VNODE has no vnode", __func__));
2332 } else if (object->type == OBJT_SWAP &&
2333 (object->flags & OBJ_TMPFS) != 0) {
2334 vp = object->un_pager.swp.swp_tmpfs;
2335 KASSERT(vp != NULL, ("%s: OBJT_TMPFS has no vnode", __func__));
2344 * Busy the vm object. This prevents new pages belonging to the object from
2345 * becoming busy. Existing pages persist as busy. Callers are responsible
2346 * for checking page state before proceeding.
2349 vm_object_busy(vm_object_t obj)
2352 VM_OBJECT_ASSERT_LOCKED(obj);
2354 refcount_acquire(&obj->busy);
2355 /* The fence is required to order loads of page busy. */
2356 atomic_thread_fence_acq_rel();
2360 vm_object_unbusy(vm_object_t obj)
2364 refcount_release(&obj->busy);
2368 vm_object_busy_wait(vm_object_t obj, const char *wmesg)
2371 VM_OBJECT_ASSERT_UNLOCKED(obj);
2374 refcount_sleep(&obj->busy, wmesg, PVM);
2378 * Return the kvme type of the given object.
2379 * If vpp is not NULL, set it to the object's vm_object_vnode() or NULL.
2382 vm_object_kvme_type(vm_object_t object, struct vnode **vpp)
2385 VM_OBJECT_ASSERT_LOCKED(object);
2387 *vpp = vm_object_vnode(object);
2388 switch (object->type) {
2390 return (KVME_TYPE_DEFAULT);
2392 return (KVME_TYPE_VNODE);
2394 if ((object->flags & OBJ_TMPFS_NODE) != 0)
2395 return (KVME_TYPE_VNODE);
2396 return (KVME_TYPE_SWAP);
2398 return (KVME_TYPE_DEVICE);
2400 return (KVME_TYPE_PHYS);
2402 return (KVME_TYPE_DEAD);
2404 return (KVME_TYPE_SG);
2405 case OBJT_MGTDEVICE:
2406 return (KVME_TYPE_MGTDEVICE);
2408 return (KVME_TYPE_UNKNOWN);
2413 sysctl_vm_object_list(SYSCTL_HANDLER_ARGS)
2415 struct kinfo_vmobject *kvo;
2416 char *fullpath, *freepath;
2423 if (req->oldptr == NULL) {
2425 * If an old buffer has not been provided, generate an
2426 * estimate of the space needed for a subsequent call.
2428 mtx_lock(&vm_object_list_mtx);
2430 TAILQ_FOREACH(obj, &vm_object_list, object_list) {
2431 if (obj->type == OBJT_DEAD)
2435 mtx_unlock(&vm_object_list_mtx);
2436 return (SYSCTL_OUT(req, NULL, sizeof(struct kinfo_vmobject) *
2440 kvo = malloc(sizeof(*kvo), M_TEMP, M_WAITOK);
2444 * VM objects are type stable and are never removed from the
2445 * list once added. This allows us to safely read obj->object_list
2446 * after reacquiring the VM object lock.
2448 mtx_lock(&vm_object_list_mtx);
2449 TAILQ_FOREACH(obj, &vm_object_list, object_list) {
2450 if (obj->type == OBJT_DEAD)
2452 VM_OBJECT_RLOCK(obj);
2453 if (obj->type == OBJT_DEAD) {
2454 VM_OBJECT_RUNLOCK(obj);
2457 mtx_unlock(&vm_object_list_mtx);
2458 kvo->kvo_size = ptoa(obj->size);
2459 kvo->kvo_resident = obj->resident_page_count;
2460 kvo->kvo_ref_count = obj->ref_count;
2461 kvo->kvo_shadow_count = obj->shadow_count;
2462 kvo->kvo_memattr = obj->memattr;
2463 kvo->kvo_active = 0;
2464 kvo->kvo_inactive = 0;
2465 TAILQ_FOREACH(m, &obj->memq, listq) {
2467 * A page may belong to the object but be
2468 * dequeued and set to PQ_NONE while the
2469 * object lock is not held. This makes the
2470 * reads of m->queue below racy, and we do not
2471 * count pages set to PQ_NONE. However, this
2472 * sysctl is only meant to give an
2473 * approximation of the system anyway.
2475 if (m->a.queue == PQ_ACTIVE)
2477 else if (m->a.queue == PQ_INACTIVE)
2478 kvo->kvo_inactive++;
2481 kvo->kvo_vn_fileid = 0;
2482 kvo->kvo_vn_fsid = 0;
2483 kvo->kvo_vn_fsid_freebsd11 = 0;
2486 kvo->kvo_type = vm_object_kvme_type(obj, &vp);
2489 VM_OBJECT_RUNLOCK(obj);
2491 vn_fullpath(curthread, vp, &fullpath, &freepath);
2492 vn_lock(vp, LK_SHARED | LK_RETRY);
2493 if (VOP_GETATTR(vp, &va, curthread->td_ucred) == 0) {
2494 kvo->kvo_vn_fileid = va.va_fileid;
2495 kvo->kvo_vn_fsid = va.va_fsid;
2496 kvo->kvo_vn_fsid_freebsd11 = va.va_fsid;
2502 strlcpy(kvo->kvo_path, fullpath, sizeof(kvo->kvo_path));
2503 if (freepath != NULL)
2504 free(freepath, M_TEMP);
2506 /* Pack record size down */
2507 kvo->kvo_structsize = offsetof(struct kinfo_vmobject, kvo_path)
2508 + strlen(kvo->kvo_path) + 1;
2509 kvo->kvo_structsize = roundup(kvo->kvo_structsize,
2511 error = SYSCTL_OUT(req, kvo, kvo->kvo_structsize);
2512 mtx_lock(&vm_object_list_mtx);
2516 mtx_unlock(&vm_object_list_mtx);
2520 SYSCTL_PROC(_vm, OID_AUTO, objects, CTLTYPE_STRUCT | CTLFLAG_RW | CTLFLAG_SKIP |
2521 CTLFLAG_MPSAFE, NULL, 0, sysctl_vm_object_list, "S,kinfo_vmobject",
2522 "List of VM objects");
2524 #include "opt_ddb.h"
2526 #include <sys/kernel.h>
2528 #include <sys/cons.h>
2530 #include <ddb/ddb.h>
2533 _vm_object_in_map(vm_map_t map, vm_object_t object, vm_map_entry_t entry)
2536 vm_map_entry_t tmpe;
2543 VM_MAP_ENTRY_FOREACH(tmpe, map) {
2544 if (_vm_object_in_map(map, object, tmpe)) {
2548 } else if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
2549 tmpm = entry->object.sub_map;
2550 VM_MAP_ENTRY_FOREACH(tmpe, tmpm) {
2551 if (_vm_object_in_map(tmpm, object, tmpe)) {
2555 } else if ((obj = entry->object.vm_object) != NULL) {
2556 for (; obj; obj = obj->backing_object)
2557 if (obj == object) {
2565 vm_object_in_map(vm_object_t object)
2569 /* sx_slock(&allproc_lock); */
2570 FOREACH_PROC_IN_SYSTEM(p) {
2571 if (!p->p_vmspace /* || (p->p_flag & (P_SYSTEM|P_WEXIT)) */)
2573 if (_vm_object_in_map(&p->p_vmspace->vm_map, object, 0)) {
2574 /* sx_sunlock(&allproc_lock); */
2578 /* sx_sunlock(&allproc_lock); */
2579 if (_vm_object_in_map(kernel_map, object, 0))
2584 DB_SHOW_COMMAND(vmochk, vm_object_check)
2589 * make sure that internal objs are in a map somewhere
2590 * and none have zero ref counts.
2592 TAILQ_FOREACH(object, &vm_object_list, object_list) {
2593 if ((object->flags & OBJ_ANON) != 0) {
2594 if (object->ref_count == 0) {
2595 db_printf("vmochk: internal obj has zero ref count: %ld\n",
2596 (long)object->size);
2598 if (!vm_object_in_map(object)) {
2600 "vmochk: internal obj is not in a map: "
2601 "ref: %d, size: %lu: 0x%lx, backing_object: %p\n",
2602 object->ref_count, (u_long)object->size,
2603 (u_long)object->size,
2604 (void *)object->backing_object);
2611 * vm_object_print: [ debug ]
2613 DB_SHOW_COMMAND(object, vm_object_print_static)
2615 /* XXX convert args. */
2616 vm_object_t object = (vm_object_t)addr;
2617 boolean_t full = have_addr;
2621 /* XXX count is an (unused) arg. Avoid shadowing it. */
2622 #define count was_count
2630 "Object %p: type=%d, size=0x%jx, res=%d, ref=%d, flags=0x%x ruid %d charge %jx\n",
2631 object, (int)object->type, (uintmax_t)object->size,
2632 object->resident_page_count, object->ref_count, object->flags,
2633 object->cred ? object->cred->cr_ruid : -1, (uintmax_t)object->charge);
2634 db_iprintf(" sref=%d, backing_object(%d)=(%p)+0x%jx\n",
2635 object->shadow_count,
2636 object->backing_object ? object->backing_object->ref_count : 0,
2637 object->backing_object, (uintmax_t)object->backing_object_offset);
2644 TAILQ_FOREACH(p, &object->memq, listq) {
2646 db_iprintf("memory:=");
2647 else if (count == 6) {
2655 db_printf("(off=0x%jx,page=0x%jx)",
2656 (uintmax_t)p->pindex, (uintmax_t)VM_PAGE_TO_PHYS(p));
2666 /* XXX need this non-static entry for calling from vm_map_print. */
2669 /* db_expr_t */ long addr,
2670 boolean_t have_addr,
2671 /* db_expr_t */ long count,
2674 vm_object_print_static(addr, have_addr, count, modif);
2677 DB_SHOW_COMMAND(vmopag, vm_object_print_pages)
2682 vm_page_t m, prev_m;
2686 TAILQ_FOREACH(object, &vm_object_list, object_list) {
2687 db_printf("new object: %p\n", (void *)object);
2698 TAILQ_FOREACH(m, &object->memq, listq) {
2699 if (m->pindex > 128)
2701 if ((prev_m = TAILQ_PREV(m, pglist, listq)) != NULL &&
2702 prev_m->pindex + 1 != m->pindex) {
2704 db_printf(" index(%ld)run(%d)pa(0x%lx)\n",
2705 (long)fidx, rcount, (long)pa);
2717 (VM_PAGE_TO_PHYS(m) == pa + rcount * PAGE_SIZE)) {
2722 db_printf(" index(%ld)run(%d)pa(0x%lx)\n",
2723 (long)fidx, rcount, (long)pa);
2733 pa = VM_PAGE_TO_PHYS(m);
2737 db_printf(" index(%ld)run(%d)pa(0x%lx)\n",
2738 (long)fidx, rcount, (long)pa);