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/socket.h>
84 #include <sys/resourcevar.h>
85 #include <sys/rwlock.h>
87 #include <sys/vnode.h>
88 #include <sys/vmmeter.h>
92 #include <vm/vm_param.h>
94 #include <vm/vm_map.h>
95 #include <vm/vm_object.h>
96 #include <vm/vm_page.h>
97 #include <vm/vm_pageout.h>
98 #include <vm/vm_pager.h>
99 #include <vm/vm_phys.h>
100 #include <vm/vm_pagequeue.h>
101 #include <vm/swap_pager.h>
102 #include <vm/vm_kern.h>
103 #include <vm/vm_extern.h>
104 #include <vm/vm_radix.h>
105 #include <vm/vm_reserv.h>
108 static int old_msync;
109 SYSCTL_INT(_vm, OID_AUTO, old_msync, CTLFLAG_RW, &old_msync, 0,
110 "Use old (insecure) msync behavior");
112 static int vm_object_page_collect_flush(vm_object_t object, vm_page_t p,
113 int pagerflags, int flags, boolean_t *clearobjflags,
115 static boolean_t vm_object_page_remove_write(vm_page_t p, int flags,
116 boolean_t *clearobjflags);
117 static void vm_object_qcollapse(vm_object_t object);
118 static void vm_object_vndeallocate(vm_object_t object);
121 * Virtual memory objects maintain the actual data
122 * associated with allocated virtual memory. A given
123 * page of memory exists within exactly one object.
125 * An object is only deallocated when all "references"
126 * are given up. Only one "reference" to a given
127 * region of an object should be writeable.
129 * Associated with each object is a list of all resident
130 * memory pages belonging to that object; this list is
131 * maintained by the "vm_page" module, and locked by the object's
134 * Each object also records a "pager" routine which is
135 * used to retrieve (and store) pages to the proper backing
136 * storage. In addition, objects may be backed by other
137 * objects from which they were virtual-copied.
139 * The only items within the object structure which are
140 * modified after time of creation are:
141 * reference count locked by object's lock
142 * pager routine locked by object's lock
146 struct object_q vm_object_list;
147 struct mtx vm_object_list_mtx; /* lock for object list and count */
149 struct vm_object kernel_object_store;
151 static SYSCTL_NODE(_vm_stats, OID_AUTO, object, CTLFLAG_RD, 0,
154 static counter_u64_t object_collapses = EARLY_COUNTER;
155 SYSCTL_COUNTER_U64(_vm_stats_object, OID_AUTO, collapses, CTLFLAG_RD,
157 "VM object collapses");
159 static counter_u64_t object_bypasses = EARLY_COUNTER;
160 SYSCTL_COUNTER_U64(_vm_stats_object, OID_AUTO, bypasses, CTLFLAG_RD,
162 "VM object bypasses");
165 counter_startup(void)
168 object_collapses = counter_u64_alloc(M_WAITOK);
169 object_bypasses = counter_u64_alloc(M_WAITOK);
171 SYSINIT(object_counters, SI_SUB_CPU, SI_ORDER_ANY, counter_startup, NULL);
173 static uma_zone_t obj_zone;
175 static int vm_object_zinit(void *mem, int size, int flags);
178 static void vm_object_zdtor(void *mem, int size, void *arg);
181 vm_object_zdtor(void *mem, int size, void *arg)
185 object = (vm_object_t)mem;
186 KASSERT(object->ref_count == 0,
187 ("object %p ref_count = %d", object, object->ref_count));
188 KASSERT(TAILQ_EMPTY(&object->memq),
189 ("object %p has resident pages in its memq", object));
190 KASSERT(vm_radix_is_empty(&object->rtree),
191 ("object %p has resident pages in its trie", object));
192 #if VM_NRESERVLEVEL > 0
193 KASSERT(LIST_EMPTY(&object->rvq),
194 ("object %p has reservations",
197 KASSERT(object->paging_in_progress == 0,
198 ("object %p paging_in_progress = %d",
199 object, object->paging_in_progress));
200 KASSERT(object->resident_page_count == 0,
201 ("object %p resident_page_count = %d",
202 object, object->resident_page_count));
203 KASSERT(object->shadow_count == 0,
204 ("object %p shadow_count = %d",
205 object, object->shadow_count));
206 KASSERT(object->type == OBJT_DEAD,
207 ("object %p has non-dead type %d",
208 object, object->type));
213 vm_object_zinit(void *mem, int size, int flags)
217 object = (vm_object_t)mem;
218 rw_init_flags(&object->lock, "vm object", RW_DUPOK | RW_NEW);
220 /* These are true for any object that has been freed */
221 object->type = OBJT_DEAD;
222 object->ref_count = 0;
223 vm_radix_init(&object->rtree);
224 object->paging_in_progress = 0;
225 object->resident_page_count = 0;
226 object->shadow_count = 0;
227 object->flags = OBJ_DEAD;
229 mtx_lock(&vm_object_list_mtx);
230 TAILQ_INSERT_TAIL(&vm_object_list, object, object_list);
231 mtx_unlock(&vm_object_list_mtx);
236 _vm_object_allocate(objtype_t type, vm_pindex_t size, vm_object_t object)
239 TAILQ_INIT(&object->memq);
240 LIST_INIT(&object->shadow_head);
243 if (type == OBJT_SWAP)
244 pctrie_init(&object->un_pager.swp.swp_blks);
247 * Ensure that swap_pager_swapoff() iteration over object_list
248 * sees up to date type and pctrie head if it observed
251 atomic_thread_fence_rel();
255 panic("_vm_object_allocate: can't create OBJT_DEAD");
258 object->flags = OBJ_ONEMAPPING;
262 object->flags = OBJ_FICTITIOUS | OBJ_UNMANAGED;
265 object->flags = OBJ_FICTITIOUS;
268 object->flags = OBJ_UNMANAGED;
274 panic("_vm_object_allocate: type %d is undefined", type);
277 object->domain.dr_policy = NULL;
278 object->generation = 1;
279 object->ref_count = 1;
280 object->memattr = VM_MEMATTR_DEFAULT;
283 object->handle = NULL;
284 object->backing_object = NULL;
285 object->backing_object_offset = (vm_ooffset_t) 0;
286 #if VM_NRESERVLEVEL > 0
287 LIST_INIT(&object->rvq);
289 umtx_shm_object_init(object);
295 * Initialize the VM objects module.
300 TAILQ_INIT(&vm_object_list);
301 mtx_init(&vm_object_list_mtx, "vm object_list", NULL, MTX_DEF);
303 rw_init(&kernel_object->lock, "kernel vm object");
304 _vm_object_allocate(OBJT_PHYS, atop(VM_MAX_KERNEL_ADDRESS -
305 VM_MIN_KERNEL_ADDRESS), kernel_object);
306 #if VM_NRESERVLEVEL > 0
307 kernel_object->flags |= OBJ_COLORED;
308 kernel_object->pg_color = (u_short)atop(VM_MIN_KERNEL_ADDRESS);
312 * The lock portion of struct vm_object must be type stable due
313 * to vm_pageout_fallback_object_lock locking a vm object
314 * without holding any references to it.
316 obj_zone = uma_zcreate("VM OBJECT", sizeof (struct vm_object), NULL,
322 vm_object_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
328 vm_object_clear_flag(vm_object_t object, u_short bits)
331 VM_OBJECT_ASSERT_WLOCKED(object);
332 object->flags &= ~bits;
336 * Sets the default memory attribute for the specified object. Pages
337 * that are allocated to this object are by default assigned this memory
340 * Presently, this function must be called before any pages are allocated
341 * to the object. In the future, this requirement may be relaxed for
342 * "default" and "swap" objects.
345 vm_object_set_memattr(vm_object_t object, vm_memattr_t memattr)
348 VM_OBJECT_ASSERT_WLOCKED(object);
349 switch (object->type) {
357 if (!TAILQ_EMPTY(&object->memq))
358 return (KERN_FAILURE);
361 return (KERN_INVALID_ARGUMENT);
363 panic("vm_object_set_memattr: object %p is of undefined type",
366 object->memattr = memattr;
367 return (KERN_SUCCESS);
371 vm_object_pip_add(vm_object_t object, short i)
374 VM_OBJECT_ASSERT_WLOCKED(object);
375 object->paging_in_progress += i;
379 vm_object_pip_subtract(vm_object_t object, short i)
382 VM_OBJECT_ASSERT_WLOCKED(object);
383 object->paging_in_progress -= i;
387 vm_object_pip_wakeup(vm_object_t object)
390 VM_OBJECT_ASSERT_WLOCKED(object);
391 object->paging_in_progress--;
392 if ((object->flags & OBJ_PIPWNT) && object->paging_in_progress == 0) {
393 vm_object_clear_flag(object, OBJ_PIPWNT);
399 vm_object_pip_wakeupn(vm_object_t object, short i)
402 VM_OBJECT_ASSERT_WLOCKED(object);
404 object->paging_in_progress -= i;
405 if ((object->flags & OBJ_PIPWNT) && object->paging_in_progress == 0) {
406 vm_object_clear_flag(object, OBJ_PIPWNT);
412 vm_object_pip_wait(vm_object_t object, char *waitid)
415 VM_OBJECT_ASSERT_WLOCKED(object);
416 while (object->paging_in_progress) {
417 object->flags |= OBJ_PIPWNT;
418 VM_OBJECT_SLEEP(object, object, PVM, waitid, 0);
423 * vm_object_allocate:
425 * Returns a new object with the given size.
428 vm_object_allocate(objtype_t type, vm_pindex_t size)
432 object = (vm_object_t)uma_zalloc(obj_zone, M_WAITOK);
433 _vm_object_allocate(type, size, object);
439 * vm_object_reference:
441 * Gets another reference to the given object. Note: OBJ_DEAD
442 * objects can be referenced during final cleaning.
445 vm_object_reference(vm_object_t object)
449 VM_OBJECT_WLOCK(object);
450 vm_object_reference_locked(object);
451 VM_OBJECT_WUNLOCK(object);
455 * vm_object_reference_locked:
457 * Gets another reference to the given object.
459 * The object must be locked.
462 vm_object_reference_locked(vm_object_t object)
466 VM_OBJECT_ASSERT_WLOCKED(object);
468 if (object->type == OBJT_VNODE) {
475 * Handle deallocating an object of type OBJT_VNODE.
478 vm_object_vndeallocate(vm_object_t object)
480 struct vnode *vp = (struct vnode *) object->handle;
482 VM_OBJECT_ASSERT_WLOCKED(object);
483 KASSERT(object->type == OBJT_VNODE,
484 ("vm_object_vndeallocate: not a vnode object"));
485 KASSERT(vp != NULL, ("vm_object_vndeallocate: missing vp"));
487 if (object->ref_count == 0) {
488 vn_printf(vp, "vm_object_vndeallocate ");
489 panic("vm_object_vndeallocate: bad object reference count");
493 if (!umtx_shm_vnobj_persistent && object->ref_count == 1)
494 umtx_shm_object_terminated(object);
497 * The test for text of vp vnode does not need a bypass to
498 * reach right VV_TEXT there, since it is obtained from
501 if (object->ref_count > 1 || (vp->v_vflag & VV_TEXT) == 0) {
503 VM_OBJECT_WUNLOCK(object);
504 /* vrele may need the vnode lock. */
508 VM_OBJECT_WUNLOCK(object);
509 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
511 VM_OBJECT_WLOCK(object);
513 if (object->type == OBJT_DEAD) {
514 VM_OBJECT_WUNLOCK(object);
517 if (object->ref_count == 0)
519 VM_OBJECT_WUNLOCK(object);
526 * vm_object_deallocate:
528 * Release a reference to the specified object,
529 * gained either through a vm_object_allocate
530 * or a vm_object_reference call. When all references
531 * are gone, storage associated with this object
532 * may be relinquished.
534 * No object may be locked.
537 vm_object_deallocate(vm_object_t object)
542 while (object != NULL) {
543 VM_OBJECT_WLOCK(object);
544 if (object->type == OBJT_VNODE) {
545 vm_object_vndeallocate(object);
549 KASSERT(object->ref_count != 0,
550 ("vm_object_deallocate: object deallocated too many times: %d", object->type));
553 * If the reference count goes to 0 we start calling
554 * vm_object_terminate() on the object chain.
555 * A ref count of 1 may be a special case depending on the
556 * shadow count being 0 or 1.
559 if (object->ref_count > 1) {
560 VM_OBJECT_WUNLOCK(object);
562 } else if (object->ref_count == 1) {
563 if (object->type == OBJT_SWAP &&
564 (object->flags & OBJ_TMPFS) != 0) {
565 vp = object->un_pager.swp.swp_tmpfs;
567 VM_OBJECT_WUNLOCK(object);
568 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
569 VM_OBJECT_WLOCK(object);
570 if (object->type == OBJT_DEAD ||
571 object->ref_count != 1) {
572 VM_OBJECT_WUNLOCK(object);
577 if ((object->flags & OBJ_TMPFS) != 0)
582 if (object->shadow_count == 0 &&
583 object->handle == NULL &&
584 (object->type == OBJT_DEFAULT ||
585 (object->type == OBJT_SWAP &&
586 (object->flags & OBJ_TMPFS_NODE) == 0))) {
587 vm_object_set_flag(object, OBJ_ONEMAPPING);
588 } else if ((object->shadow_count == 1) &&
589 (object->handle == NULL) &&
590 (object->type == OBJT_DEFAULT ||
591 object->type == OBJT_SWAP)) {
594 robject = LIST_FIRST(&object->shadow_head);
595 KASSERT(robject != NULL,
596 ("vm_object_deallocate: ref_count: %d, shadow_count: %d",
598 object->shadow_count));
599 KASSERT((robject->flags & OBJ_TMPFS_NODE) == 0,
600 ("shadowed tmpfs v_object %p", object));
601 if (!VM_OBJECT_TRYWLOCK(robject)) {
603 * Avoid a potential deadlock.
606 VM_OBJECT_WUNLOCK(object);
608 * More likely than not the thread
609 * holding robject's lock has lower
610 * priority than the current thread.
611 * Let the lower priority thread run.
617 * Collapse object into its shadow unless its
618 * shadow is dead. In that case, object will
619 * be deallocated by the thread that is
620 * deallocating its shadow.
622 if ((robject->flags & OBJ_DEAD) == 0 &&
623 (robject->handle == NULL) &&
624 (robject->type == OBJT_DEFAULT ||
625 robject->type == OBJT_SWAP)) {
627 robject->ref_count++;
629 if (robject->paging_in_progress) {
630 VM_OBJECT_WUNLOCK(object);
631 vm_object_pip_wait(robject,
633 temp = robject->backing_object;
634 if (object == temp) {
635 VM_OBJECT_WLOCK(object);
638 } else if (object->paging_in_progress) {
639 VM_OBJECT_WUNLOCK(robject);
640 object->flags |= OBJ_PIPWNT;
641 VM_OBJECT_SLEEP(object, object,
642 PDROP | PVM, "objde2", 0);
643 VM_OBJECT_WLOCK(robject);
644 temp = robject->backing_object;
645 if (object == temp) {
646 VM_OBJECT_WLOCK(object);
650 VM_OBJECT_WUNLOCK(object);
652 if (robject->ref_count == 1) {
653 robject->ref_count--;
658 vm_object_collapse(object);
659 VM_OBJECT_WUNLOCK(object);
662 VM_OBJECT_WUNLOCK(robject);
664 VM_OBJECT_WUNLOCK(object);
668 umtx_shm_object_terminated(object);
669 temp = object->backing_object;
671 KASSERT((object->flags & OBJ_TMPFS_NODE) == 0,
672 ("shadowed tmpfs v_object 2 %p", object));
673 VM_OBJECT_WLOCK(temp);
674 LIST_REMOVE(object, shadow_list);
675 temp->shadow_count--;
676 VM_OBJECT_WUNLOCK(temp);
677 object->backing_object = NULL;
680 * Don't double-terminate, we could be in a termination
681 * recursion due to the terminate having to sync data
684 if ((object->flags & OBJ_DEAD) == 0)
685 vm_object_terminate(object);
687 VM_OBJECT_WUNLOCK(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_terminate_pages removes any remaining pageable pages
718 * from the object and resets the object to an empty state.
721 vm_object_terminate_pages(vm_object_t object)
726 VM_OBJECT_ASSERT_WLOCKED(object);
731 * Free any remaining pageable pages. This also removes them from the
732 * paging queues. However, don't free wired pages, just remove them
733 * from the object. Rather than incrementally removing each page from
734 * the object, the page and object are reset to any empty state.
736 TAILQ_FOREACH_SAFE(p, &object->memq, listq, p_next) {
737 vm_page_assert_unbusied(p);
738 if ((object->flags & OBJ_UNMANAGED) == 0)
740 * vm_page_free_prep() only needs the page
741 * lock for managed pages.
743 vm_page_change_lock(p, &mtx);
745 if (p->wire_count != 0)
754 * If the object contained any pages, then reset it to an empty state.
755 * None of the object's fields, including "resident_page_count", were
756 * modified by the preceding loop.
758 if (object->resident_page_count != 0) {
759 vm_radix_reclaim_allnodes(&object->rtree);
760 TAILQ_INIT(&object->memq);
761 object->resident_page_count = 0;
762 if (object->type == OBJT_VNODE)
763 vdrop(object->handle);
768 * vm_object_terminate actually destroys the specified object, freeing
769 * up all previously used resources.
771 * The object must be locked.
772 * This routine may block.
775 vm_object_terminate(vm_object_t object)
778 VM_OBJECT_ASSERT_WLOCKED(object);
781 * Make sure no one uses us.
783 vm_object_set_flag(object, OBJ_DEAD);
786 * wait for the pageout daemon to be done with the object
788 vm_object_pip_wait(object, "objtrm");
790 KASSERT(!object->paging_in_progress,
791 ("vm_object_terminate: pageout in progress"));
794 * Clean and free the pages, as appropriate. All references to the
795 * object are gone, so we don't need to lock it.
797 if (object->type == OBJT_VNODE) {
798 struct vnode *vp = (struct vnode *)object->handle;
801 * Clean pages and flush buffers.
803 vm_object_page_clean(object, 0, 0, OBJPC_SYNC);
804 VM_OBJECT_WUNLOCK(object);
806 vinvalbuf(vp, V_SAVE, 0, 0);
808 BO_LOCK(&vp->v_bufobj);
809 vp->v_bufobj.bo_flag |= BO_DEAD;
810 BO_UNLOCK(&vp->v_bufobj);
812 VM_OBJECT_WLOCK(object);
815 KASSERT(object->ref_count == 0,
816 ("vm_object_terminate: object with references, ref_count=%d",
819 if ((object->flags & OBJ_PG_DTOR) == 0)
820 vm_object_terminate_pages(object);
822 #if VM_NRESERVLEVEL > 0
823 if (__predict_false(!LIST_EMPTY(&object->rvq)))
824 vm_reserv_break_all(object);
827 KASSERT(object->cred == NULL || object->type == OBJT_DEFAULT ||
828 object->type == OBJT_SWAP,
829 ("%s: non-swap obj %p has cred", __func__, object));
832 * Let the pager know object is dead.
834 vm_pager_deallocate(object);
835 VM_OBJECT_WUNLOCK(object);
837 vm_object_destroy(object);
841 * Make the page read-only so that we can clear the object flags. However, if
842 * this is a nosync mmap then the object is likely to stay dirty so do not
843 * mess with the page and do not clear the object flags. Returns TRUE if the
844 * page should be flushed, and FALSE otherwise.
847 vm_object_page_remove_write(vm_page_t p, int flags, boolean_t *clearobjflags)
851 * If we have been asked to skip nosync pages and this is a
852 * nosync page, skip it. Note that the object flags were not
853 * cleared in this case so we do not have to set them.
855 if ((flags & OBJPC_NOSYNC) != 0 && (p->oflags & VPO_NOSYNC) != 0) {
856 *clearobjflags = FALSE;
859 pmap_remove_write(p);
860 return (p->dirty != 0);
865 * vm_object_page_clean
867 * Clean all dirty pages in the specified range of object. Leaves page
868 * on whatever queue it is currently on. If NOSYNC is set then do not
869 * write out pages with VPO_NOSYNC set (originally comes from MAP_NOSYNC),
870 * leaving the object dirty.
872 * When stuffing pages asynchronously, allow clustering. XXX we need a
873 * synchronous clustering mode implementation.
875 * Odd semantics: if start == end, we clean everything.
877 * The object must be locked.
879 * Returns FALSE if some page from the range was not written, as
880 * reported by the pager, and TRUE otherwise.
883 vm_object_page_clean(vm_object_t object, vm_ooffset_t start, vm_ooffset_t end,
887 vm_pindex_t pi, tend, tstart;
888 int curgeneration, n, pagerflags;
889 boolean_t clearobjflags, eio, res;
891 VM_OBJECT_ASSERT_WLOCKED(object);
894 * The OBJ_MIGHTBEDIRTY flag is only set for OBJT_VNODE
895 * objects. The check below prevents the function from
896 * operating on non-vnode objects.
898 if ((object->flags & OBJ_MIGHTBEDIRTY) == 0 ||
899 object->resident_page_count == 0)
902 pagerflags = (flags & (OBJPC_SYNC | OBJPC_INVAL)) != 0 ?
903 VM_PAGER_PUT_SYNC : VM_PAGER_CLUSTER_OK;
904 pagerflags |= (flags & OBJPC_INVAL) != 0 ? VM_PAGER_PUT_INVAL : 0;
906 tstart = OFF_TO_IDX(start);
907 tend = (end == 0) ? object->size : OFF_TO_IDX(end + PAGE_MASK);
908 clearobjflags = tstart == 0 && tend >= object->size;
912 curgeneration = object->generation;
914 for (p = vm_page_find_least(object, tstart); p != NULL; p = np) {
918 np = TAILQ_NEXT(p, listq);
921 if (vm_page_sleep_if_busy(p, "vpcwai")) {
922 if (object->generation != curgeneration) {
923 if ((flags & OBJPC_SYNC) != 0)
926 clearobjflags = FALSE;
928 np = vm_page_find_least(object, pi);
931 if (!vm_object_page_remove_write(p, flags, &clearobjflags))
934 n = vm_object_page_collect_flush(object, p, pagerflags,
935 flags, &clearobjflags, &eio);
938 clearobjflags = FALSE;
940 if (object->generation != curgeneration) {
941 if ((flags & OBJPC_SYNC) != 0)
944 clearobjflags = FALSE;
948 * If the VOP_PUTPAGES() did a truncated write, so
949 * that even the first page of the run is not fully
950 * written, vm_pageout_flush() returns 0 as the run
951 * length. Since the condition that caused truncated
952 * write may be permanent, e.g. exhausted free space,
953 * accepting n == 0 would cause an infinite loop.
955 * Forwarding the iterator leaves the unwritten page
956 * behind, but there is not much we can do there if
957 * filesystem refuses to write it.
961 clearobjflags = FALSE;
963 np = vm_page_find_least(object, pi + n);
966 VOP_FSYNC(vp, (pagerflags & VM_PAGER_PUT_SYNC) ? MNT_WAIT : 0);
970 vm_object_clear_flag(object, OBJ_MIGHTBEDIRTY);
975 vm_object_page_collect_flush(vm_object_t object, vm_page_t p, int pagerflags,
976 int flags, boolean_t *clearobjflags, boolean_t *eio)
978 vm_page_t ma[vm_pageout_page_count], p_first, tp;
979 int count, i, mreq, runlen;
981 vm_page_lock_assert(p, MA_NOTOWNED);
982 VM_OBJECT_ASSERT_WLOCKED(object);
987 for (tp = p; count < vm_pageout_page_count; count++) {
988 tp = vm_page_next(tp);
989 if (tp == NULL || vm_page_busied(tp))
991 if (!vm_object_page_remove_write(tp, flags, clearobjflags))
995 for (p_first = p; count < vm_pageout_page_count; count++) {
996 tp = vm_page_prev(p_first);
997 if (tp == NULL || vm_page_busied(tp))
999 if (!vm_object_page_remove_write(tp, flags, clearobjflags))
1005 for (tp = p_first, i = 0; i < count; tp = TAILQ_NEXT(tp, listq), i++)
1008 vm_pageout_flush(ma, count, pagerflags, mreq, &runlen, eio);
1013 * Note that there is absolutely no sense in writing out
1014 * anonymous objects, so we track down the vnode object
1016 * We invalidate (remove) all pages from the address space
1017 * for semantic correctness.
1019 * If the backing object is a device object with unmanaged pages, then any
1020 * mappings to the specified range of pages must be removed before this
1021 * function is called.
1023 * Note: certain anonymous maps, such as MAP_NOSYNC maps,
1024 * may start out with a NULL object.
1027 vm_object_sync(vm_object_t object, vm_ooffset_t offset, vm_size_t size,
1028 boolean_t syncio, boolean_t invalidate)
1030 vm_object_t backing_object;
1033 int error, flags, fsync_after;
1040 VM_OBJECT_WLOCK(object);
1041 while ((backing_object = object->backing_object) != NULL) {
1042 VM_OBJECT_WLOCK(backing_object);
1043 offset += object->backing_object_offset;
1044 VM_OBJECT_WUNLOCK(object);
1045 object = backing_object;
1046 if (object->size < OFF_TO_IDX(offset + size))
1047 size = IDX_TO_OFF(object->size) - offset;
1050 * Flush pages if writing is allowed, invalidate them
1051 * if invalidation requested. Pages undergoing I/O
1052 * will be ignored by vm_object_page_remove().
1054 * We cannot lock the vnode and then wait for paging
1055 * to complete without deadlocking against vm_fault.
1056 * Instead we simply call vm_object_page_remove() and
1057 * allow it to block internally on a page-by-page
1058 * basis when it encounters pages undergoing async
1061 if (object->type == OBJT_VNODE &&
1062 (object->flags & OBJ_MIGHTBEDIRTY) != 0 &&
1063 ((vp = object->handle)->v_vflag & VV_NOSYNC) == 0) {
1064 VM_OBJECT_WUNLOCK(object);
1065 (void) vn_start_write(vp, &mp, V_WAIT);
1066 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1067 if (syncio && !invalidate && offset == 0 &&
1068 atop(size) == object->size) {
1070 * If syncing the whole mapping of the file,
1071 * it is faster to schedule all the writes in
1072 * async mode, also allowing the clustering,
1073 * and then wait for i/o to complete.
1078 flags = (syncio || invalidate) ? OBJPC_SYNC : 0;
1079 flags |= invalidate ? (OBJPC_SYNC | OBJPC_INVAL) : 0;
1080 fsync_after = FALSE;
1082 VM_OBJECT_WLOCK(object);
1083 res = vm_object_page_clean(object, offset, offset + size,
1085 VM_OBJECT_WUNLOCK(object);
1087 error = VOP_FSYNC(vp, MNT_WAIT, curthread);
1089 vn_finished_write(mp);
1092 VM_OBJECT_WLOCK(object);
1094 if ((object->type == OBJT_VNODE ||
1095 object->type == OBJT_DEVICE) && invalidate) {
1096 if (object->type == OBJT_DEVICE)
1098 * The option OBJPR_NOTMAPPED must be passed here
1099 * because vm_object_page_remove() cannot remove
1100 * unmanaged mappings.
1102 flags = OBJPR_NOTMAPPED;
1106 flags = OBJPR_CLEANONLY;
1107 vm_object_page_remove(object, OFF_TO_IDX(offset),
1108 OFF_TO_IDX(offset + size + PAGE_MASK), flags);
1110 VM_OBJECT_WUNLOCK(object);
1115 * Determine whether the given advice can be applied to the object. Advice is
1116 * not applied to unmanaged pages since they never belong to page queues, and
1117 * since MADV_FREE is destructive, it can apply only to anonymous pages that
1118 * have been mapped at most once.
1121 vm_object_advice_applies(vm_object_t object, int advice)
1124 if ((object->flags & OBJ_UNMANAGED) != 0)
1126 if (advice != MADV_FREE)
1128 return ((object->type == OBJT_DEFAULT || object->type == OBJT_SWAP) &&
1129 (object->flags & OBJ_ONEMAPPING) != 0);
1133 vm_object_madvise_freespace(vm_object_t object, int advice, vm_pindex_t pindex,
1137 if (advice == MADV_FREE && object->type == OBJT_SWAP)
1138 swap_pager_freespace(object, pindex, size);
1142 * vm_object_madvise:
1144 * Implements the madvise function at the object/page level.
1146 * MADV_WILLNEED (any object)
1148 * Activate the specified pages if they are resident.
1150 * MADV_DONTNEED (any object)
1152 * Deactivate the specified pages if they are resident.
1154 * MADV_FREE (OBJT_DEFAULT/OBJT_SWAP objects,
1155 * OBJ_ONEMAPPING only)
1157 * Deactivate and clean the specified pages if they are
1158 * resident. This permits the process to reuse the pages
1159 * without faulting or the kernel to reclaim the pages
1163 vm_object_madvise(vm_object_t object, vm_pindex_t pindex, vm_pindex_t end,
1166 vm_pindex_t tpindex;
1167 vm_object_t backing_object, tobject;
1174 VM_OBJECT_WLOCK(object);
1175 if (!vm_object_advice_applies(object, advice)) {
1176 VM_OBJECT_WUNLOCK(object);
1179 for (m = vm_page_find_least(object, pindex); pindex < end; pindex++) {
1183 * If the next page isn't resident in the top-level object, we
1184 * need to search the shadow chain. When applying MADV_FREE, we
1185 * take care to release any swap space used to store
1186 * non-resident pages.
1188 if (m == NULL || pindex < m->pindex) {
1190 * Optimize a common case: if the top-level object has
1191 * no backing object, we can skip over the non-resident
1192 * range in constant time.
1194 if (object->backing_object == NULL) {
1195 tpindex = (m != NULL && m->pindex < end) ?
1197 vm_object_madvise_freespace(object, advice,
1198 pindex, tpindex - pindex);
1199 if ((pindex = tpindex) == end)
1206 vm_object_madvise_freespace(tobject, advice,
1209 * Prepare to search the next object in the
1212 backing_object = tobject->backing_object;
1213 if (backing_object == NULL)
1215 VM_OBJECT_WLOCK(backing_object);
1217 OFF_TO_IDX(tobject->backing_object_offset);
1218 if (tobject != object)
1219 VM_OBJECT_WUNLOCK(tobject);
1220 tobject = backing_object;
1221 if (!vm_object_advice_applies(tobject, advice))
1223 } while ((tm = vm_page_lookup(tobject, tpindex)) ==
1228 m = TAILQ_NEXT(m, listq);
1232 * If the page is not in a normal state, skip it.
1234 if (tm->valid != VM_PAGE_BITS_ALL)
1237 if (vm_page_held(tm)) {
1241 KASSERT((tm->flags & PG_FICTITIOUS) == 0,
1242 ("vm_object_madvise: page %p is fictitious", tm));
1243 KASSERT((tm->oflags & VPO_UNMANAGED) == 0,
1244 ("vm_object_madvise: page %p is not managed", tm));
1245 if (vm_page_busied(tm)) {
1246 if (object != tobject)
1247 VM_OBJECT_WUNLOCK(tobject);
1248 VM_OBJECT_WUNLOCK(object);
1249 if (advice == MADV_WILLNEED) {
1251 * Reference the page before unlocking and
1252 * sleeping so that the page daemon is less
1253 * likely to reclaim it.
1255 vm_page_aflag_set(tm, PGA_REFERENCED);
1257 vm_page_busy_sleep(tm, "madvpo", false);
1260 vm_page_advise(tm, advice);
1262 vm_object_madvise_freespace(tobject, advice, tm->pindex, 1);
1264 if (tobject != object)
1265 VM_OBJECT_WUNLOCK(tobject);
1267 VM_OBJECT_WUNLOCK(object);
1273 * Create a new object which is backed by the
1274 * specified existing object range. The source
1275 * object reference is deallocated.
1277 * The new object and offset into that object
1278 * are returned in the source parameters.
1282 vm_object_t *object, /* IN/OUT */
1283 vm_ooffset_t *offset, /* IN/OUT */
1292 * Don't create the new object if the old object isn't shared.
1294 if (source != NULL) {
1295 VM_OBJECT_WLOCK(source);
1296 if (source->ref_count == 1 &&
1297 source->handle == NULL &&
1298 (source->type == OBJT_DEFAULT ||
1299 source->type == OBJT_SWAP)) {
1300 VM_OBJECT_WUNLOCK(source);
1303 VM_OBJECT_WUNLOCK(source);
1307 * Allocate a new object with the given length.
1309 result = vm_object_allocate(OBJT_DEFAULT, atop(length));
1312 * The new object shadows the source object, adding a reference to it.
1313 * Our caller changes his reference to point to the new object,
1314 * removing a reference to the source object. Net result: no change
1315 * of reference count.
1317 * Try to optimize the result object's page color when shadowing
1318 * in order to maintain page coloring consistency in the combined
1321 result->backing_object = source;
1323 * Store the offset into the source object, and fix up the offset into
1326 result->backing_object_offset = *offset;
1327 if (source != NULL) {
1328 VM_OBJECT_WLOCK(source);
1329 result->domain = source->domain;
1330 LIST_INSERT_HEAD(&source->shadow_head, result, shadow_list);
1331 source->shadow_count++;
1332 #if VM_NRESERVLEVEL > 0
1333 result->flags |= source->flags & OBJ_COLORED;
1334 result->pg_color = (source->pg_color + OFF_TO_IDX(*offset)) &
1335 ((1 << (VM_NFREEORDER - 1)) - 1);
1337 VM_OBJECT_WUNLOCK(source);
1342 * Return the new things
1351 * Split the pages in a map entry into a new object. This affords
1352 * easier removal of unused pages, and keeps object inheritance from
1353 * being a negative impact on memory usage.
1356 vm_object_split(vm_map_entry_t entry)
1358 vm_page_t m, m_next;
1359 vm_object_t orig_object, new_object, source;
1360 vm_pindex_t idx, offidxstart;
1363 orig_object = entry->object.vm_object;
1364 if (orig_object->type != OBJT_DEFAULT && orig_object->type != OBJT_SWAP)
1366 if (orig_object->ref_count <= 1)
1368 VM_OBJECT_WUNLOCK(orig_object);
1370 offidxstart = OFF_TO_IDX(entry->offset);
1371 size = atop(entry->end - entry->start);
1374 * If swap_pager_copy() is later called, it will convert new_object
1375 * into a swap object.
1377 new_object = vm_object_allocate(OBJT_DEFAULT, size);
1380 * At this point, the new object is still private, so the order in
1381 * which the original and new objects are locked does not matter.
1383 VM_OBJECT_WLOCK(new_object);
1384 VM_OBJECT_WLOCK(orig_object);
1385 new_object->domain = orig_object->domain;
1386 source = orig_object->backing_object;
1387 if (source != NULL) {
1388 VM_OBJECT_WLOCK(source);
1389 if ((source->flags & OBJ_DEAD) != 0) {
1390 VM_OBJECT_WUNLOCK(source);
1391 VM_OBJECT_WUNLOCK(orig_object);
1392 VM_OBJECT_WUNLOCK(new_object);
1393 vm_object_deallocate(new_object);
1394 VM_OBJECT_WLOCK(orig_object);
1397 LIST_INSERT_HEAD(&source->shadow_head,
1398 new_object, shadow_list);
1399 source->shadow_count++;
1400 vm_object_reference_locked(source); /* for new_object */
1401 vm_object_clear_flag(source, OBJ_ONEMAPPING);
1402 VM_OBJECT_WUNLOCK(source);
1403 new_object->backing_object_offset =
1404 orig_object->backing_object_offset + entry->offset;
1405 new_object->backing_object = source;
1407 if (orig_object->cred != NULL) {
1408 new_object->cred = orig_object->cred;
1409 crhold(orig_object->cred);
1410 new_object->charge = ptoa(size);
1411 KASSERT(orig_object->charge >= ptoa(size),
1412 ("orig_object->charge < 0"));
1413 orig_object->charge -= ptoa(size);
1416 m = vm_page_find_least(orig_object, offidxstart);
1417 for (; m != NULL && (idx = m->pindex - offidxstart) < size;
1419 m_next = TAILQ_NEXT(m, listq);
1422 * We must wait for pending I/O to complete before we can
1425 * We do not have to VM_PROT_NONE the page as mappings should
1426 * not be changed by this operation.
1428 if (vm_page_busied(m)) {
1429 VM_OBJECT_WUNLOCK(new_object);
1431 VM_OBJECT_WUNLOCK(orig_object);
1432 vm_page_busy_sleep(m, "spltwt", false);
1433 VM_OBJECT_WLOCK(orig_object);
1434 VM_OBJECT_WLOCK(new_object);
1438 /* vm_page_rename() will dirty the page. */
1439 if (vm_page_rename(m, new_object, idx)) {
1440 VM_OBJECT_WUNLOCK(new_object);
1441 VM_OBJECT_WUNLOCK(orig_object);
1443 VM_OBJECT_WLOCK(orig_object);
1444 VM_OBJECT_WLOCK(new_object);
1447 #if VM_NRESERVLEVEL > 0
1449 * If some of the reservation's allocated pages remain with
1450 * the original object, then transferring the reservation to
1451 * the new object is neither particularly beneficial nor
1452 * particularly harmful as compared to leaving the reservation
1453 * with the original object. If, however, all of the
1454 * reservation's allocated pages are transferred to the new
1455 * object, then transferring the reservation is typically
1456 * beneficial. Determining which of these two cases applies
1457 * would be more costly than unconditionally renaming the
1460 vm_reserv_rename(m, new_object, orig_object, offidxstart);
1462 if (orig_object->type == OBJT_SWAP)
1465 if (orig_object->type == OBJT_SWAP) {
1467 * swap_pager_copy() can sleep, in which case the orig_object's
1468 * and new_object's locks are released and reacquired.
1470 swap_pager_copy(orig_object, new_object, offidxstart, 0);
1471 TAILQ_FOREACH(m, &new_object->memq, listq)
1474 VM_OBJECT_WUNLOCK(orig_object);
1475 VM_OBJECT_WUNLOCK(new_object);
1476 entry->object.vm_object = new_object;
1477 entry->offset = 0LL;
1478 vm_object_deallocate(orig_object);
1479 VM_OBJECT_WLOCK(new_object);
1482 #define OBSC_COLLAPSE_NOWAIT 0x0002
1483 #define OBSC_COLLAPSE_WAIT 0x0004
1486 vm_object_collapse_scan_wait(vm_object_t object, vm_page_t p, vm_page_t next,
1489 vm_object_t backing_object;
1491 VM_OBJECT_ASSERT_WLOCKED(object);
1492 backing_object = object->backing_object;
1493 VM_OBJECT_ASSERT_WLOCKED(backing_object);
1495 KASSERT(p == NULL || vm_page_busied(p), ("unbusy page %p", p));
1496 KASSERT(p == NULL || p->object == object || p->object == backing_object,
1497 ("invalid ownership %p %p %p", p, object, backing_object));
1498 if ((op & OBSC_COLLAPSE_NOWAIT) != 0)
1502 VM_OBJECT_WUNLOCK(object);
1503 VM_OBJECT_WUNLOCK(backing_object);
1504 /* The page is only NULL when rename fails. */
1508 vm_page_busy_sleep(p, "vmocol", false);
1509 VM_OBJECT_WLOCK(object);
1510 VM_OBJECT_WLOCK(backing_object);
1511 return (TAILQ_FIRST(&backing_object->memq));
1515 vm_object_scan_all_shadowed(vm_object_t object)
1517 vm_object_t backing_object;
1519 vm_pindex_t backing_offset_index, new_pindex, pi, ps;
1521 VM_OBJECT_ASSERT_WLOCKED(object);
1522 VM_OBJECT_ASSERT_WLOCKED(object->backing_object);
1524 backing_object = object->backing_object;
1526 if (backing_object->type != OBJT_DEFAULT &&
1527 backing_object->type != OBJT_SWAP)
1530 pi = backing_offset_index = OFF_TO_IDX(object->backing_object_offset);
1531 p = vm_page_find_least(backing_object, pi);
1532 ps = swap_pager_find_least(backing_object, pi);
1535 * Only check pages inside the parent object's range and
1536 * inside the parent object's mapping of the backing object.
1539 if (p != NULL && p->pindex < pi)
1540 p = TAILQ_NEXT(p, listq);
1542 ps = swap_pager_find_least(backing_object, pi);
1543 if (p == NULL && ps >= backing_object->size)
1548 pi = MIN(p->pindex, ps);
1550 new_pindex = pi - backing_offset_index;
1551 if (new_pindex >= object->size)
1555 * See if the parent has the page or if the parent's object
1556 * pager has the page. If the parent has the page but the page
1557 * is not valid, the parent's object pager must have the page.
1559 * If this fails, the parent does not completely shadow the
1560 * object and we might as well give up now.
1562 pp = vm_page_lookup(object, new_pindex);
1563 if ((pp == NULL || pp->valid == 0) &&
1564 !vm_pager_has_page(object, new_pindex, NULL, NULL))
1571 vm_object_collapse_scan(vm_object_t object, int op)
1573 vm_object_t backing_object;
1574 vm_page_t next, p, pp;
1575 vm_pindex_t backing_offset_index, new_pindex;
1577 VM_OBJECT_ASSERT_WLOCKED(object);
1578 VM_OBJECT_ASSERT_WLOCKED(object->backing_object);
1580 backing_object = object->backing_object;
1581 backing_offset_index = OFF_TO_IDX(object->backing_object_offset);
1584 * Initial conditions
1586 if ((op & OBSC_COLLAPSE_WAIT) != 0)
1587 vm_object_set_flag(backing_object, OBJ_DEAD);
1592 for (p = TAILQ_FIRST(&backing_object->memq); p != NULL; p = next) {
1593 next = TAILQ_NEXT(p, listq);
1594 new_pindex = p->pindex - backing_offset_index;
1597 * Check for busy page
1599 if (vm_page_busied(p)) {
1600 next = vm_object_collapse_scan_wait(object, p, next, op);
1604 KASSERT(p->object == backing_object,
1605 ("vm_object_collapse_scan: object mismatch"));
1607 if (p->pindex < backing_offset_index ||
1608 new_pindex >= object->size) {
1609 if (backing_object->type == OBJT_SWAP)
1610 swap_pager_freespace(backing_object, p->pindex,
1614 * Page is out of the parent object's range, we can
1615 * simply destroy it.
1618 KASSERT(!pmap_page_is_mapped(p),
1619 ("freeing mapped page %p", p));
1620 if (p->wire_count == 0)
1628 pp = vm_page_lookup(object, new_pindex);
1629 if (pp != NULL && vm_page_busied(pp)) {
1631 * The page in the parent is busy and possibly not
1632 * (yet) valid. Until its state is finalized by the
1633 * busy bit owner, we can't tell whether it shadows the
1634 * original page. Therefore, we must either skip it
1635 * and the original (backing_object) page or wait for
1636 * its state to be finalized.
1638 * This is due to a race with vm_fault() where we must
1639 * unbusy the original (backing_obj) page before we can
1640 * (re)lock the parent. Hence we can get here.
1642 next = vm_object_collapse_scan_wait(object, pp, next,
1647 KASSERT(pp == NULL || pp->valid != 0,
1648 ("unbusy invalid page %p", pp));
1650 if (pp != NULL || vm_pager_has_page(object, new_pindex, NULL,
1653 * The page already exists in the parent OR swap exists
1654 * for this location in the parent. Leave the parent's
1655 * page alone. Destroy the original page from the
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 (p->wire_count == 0)
1673 * Page does not exist in parent, rename the page from the
1674 * backing object to the main object.
1676 * If the page was mapped to a process, it can remain mapped
1677 * through the rename. vm_page_rename() will dirty the page.
1679 if (vm_page_rename(p, object, new_pindex)) {
1680 next = vm_object_collapse_scan_wait(object, NULL, next,
1685 /* Use the old pindex to free the right page. */
1686 if (backing_object->type == OBJT_SWAP)
1687 swap_pager_freespace(backing_object,
1688 new_pindex + backing_offset_index, 1);
1690 #if VM_NRESERVLEVEL > 0
1692 * Rename the reservation.
1694 vm_reserv_rename(p, object, backing_object,
1695 backing_offset_index);
1703 * this version of collapse allows the operation to occur earlier and
1704 * when paging_in_progress is true for an object... This is not a complete
1705 * operation, but should plug 99.9% of the rest of the leaks.
1708 vm_object_qcollapse(vm_object_t object)
1710 vm_object_t backing_object = object->backing_object;
1712 VM_OBJECT_ASSERT_WLOCKED(object);
1713 VM_OBJECT_ASSERT_WLOCKED(backing_object);
1715 if (backing_object->ref_count != 1)
1718 vm_object_collapse_scan(object, OBSC_COLLAPSE_NOWAIT);
1722 * vm_object_collapse:
1724 * Collapse an object with the object backing it.
1725 * Pages in the backing object are moved into the
1726 * parent, and the backing object is deallocated.
1729 vm_object_collapse(vm_object_t object)
1731 vm_object_t backing_object, new_backing_object;
1733 VM_OBJECT_ASSERT_WLOCKED(object);
1737 * Verify that the conditions are right for collapse:
1739 * The object exists and the backing object exists.
1741 if ((backing_object = object->backing_object) == NULL)
1745 * we check the backing object first, because it is most likely
1748 VM_OBJECT_WLOCK(backing_object);
1749 if (backing_object->handle != NULL ||
1750 (backing_object->type != OBJT_DEFAULT &&
1751 backing_object->type != OBJT_SWAP) ||
1752 (backing_object->flags & OBJ_DEAD) ||
1753 object->handle != NULL ||
1754 (object->type != OBJT_DEFAULT &&
1755 object->type != OBJT_SWAP) ||
1756 (object->flags & OBJ_DEAD)) {
1757 VM_OBJECT_WUNLOCK(backing_object);
1761 if (object->paging_in_progress != 0 ||
1762 backing_object->paging_in_progress != 0) {
1763 vm_object_qcollapse(object);
1764 VM_OBJECT_WUNLOCK(backing_object);
1769 * We know that we can either collapse the backing object (if
1770 * the parent is the only reference to it) or (perhaps) have
1771 * the parent bypass the object if the parent happens to shadow
1772 * all the resident pages in the entire backing object.
1774 * This is ignoring pager-backed pages such as swap pages.
1775 * vm_object_collapse_scan fails the shadowing test in this
1778 if (backing_object->ref_count == 1) {
1779 vm_object_pip_add(object, 1);
1780 vm_object_pip_add(backing_object, 1);
1783 * If there is exactly one reference to the backing
1784 * object, we can collapse it into the parent.
1786 vm_object_collapse_scan(object, OBSC_COLLAPSE_WAIT);
1788 #if VM_NRESERVLEVEL > 0
1790 * Break any reservations from backing_object.
1792 if (__predict_false(!LIST_EMPTY(&backing_object->rvq)))
1793 vm_reserv_break_all(backing_object);
1797 * Move the pager from backing_object to object.
1799 if (backing_object->type == OBJT_SWAP) {
1801 * swap_pager_copy() can sleep, in which case
1802 * the backing_object's and object's locks are
1803 * released and reacquired.
1804 * Since swap_pager_copy() is being asked to
1805 * destroy the source, it will change the
1806 * backing_object's type to OBJT_DEFAULT.
1811 OFF_TO_IDX(object->backing_object_offset), TRUE);
1814 * Object now shadows whatever backing_object did.
1815 * Note that the reference to
1816 * backing_object->backing_object moves from within
1817 * backing_object to within object.
1819 LIST_REMOVE(object, shadow_list);
1820 backing_object->shadow_count--;
1821 if (backing_object->backing_object) {
1822 VM_OBJECT_WLOCK(backing_object->backing_object);
1823 LIST_REMOVE(backing_object, shadow_list);
1825 &backing_object->backing_object->shadow_head,
1826 object, shadow_list);
1828 * The shadow_count has not changed.
1830 VM_OBJECT_WUNLOCK(backing_object->backing_object);
1832 object->backing_object = backing_object->backing_object;
1833 object->backing_object_offset +=
1834 backing_object->backing_object_offset;
1837 * Discard backing_object.
1839 * Since the backing object has no pages, no pager left,
1840 * and no object references within it, all that is
1841 * necessary is to dispose of it.
1843 KASSERT(backing_object->ref_count == 1, (
1844 "backing_object %p was somehow re-referenced during collapse!",
1846 vm_object_pip_wakeup(backing_object);
1847 backing_object->type = OBJT_DEAD;
1848 backing_object->ref_count = 0;
1849 VM_OBJECT_WUNLOCK(backing_object);
1850 vm_object_destroy(backing_object);
1852 vm_object_pip_wakeup(object);
1853 counter_u64_add(object_collapses, 1);
1856 * If we do not entirely shadow the backing object,
1857 * there is nothing we can do so we give up.
1859 if (object->resident_page_count != object->size &&
1860 !vm_object_scan_all_shadowed(object)) {
1861 VM_OBJECT_WUNLOCK(backing_object);
1866 * Make the parent shadow the next object in the
1867 * chain. Deallocating backing_object will not remove
1868 * it, since its reference count is at least 2.
1870 LIST_REMOVE(object, shadow_list);
1871 backing_object->shadow_count--;
1873 new_backing_object = backing_object->backing_object;
1874 if ((object->backing_object = new_backing_object) != NULL) {
1875 VM_OBJECT_WLOCK(new_backing_object);
1877 &new_backing_object->shadow_head,
1881 new_backing_object->shadow_count++;
1882 vm_object_reference_locked(new_backing_object);
1883 VM_OBJECT_WUNLOCK(new_backing_object);
1884 object->backing_object_offset +=
1885 backing_object->backing_object_offset;
1889 * Drop the reference count on backing_object. Since
1890 * its ref_count was at least 2, it will not vanish.
1892 backing_object->ref_count--;
1893 VM_OBJECT_WUNLOCK(backing_object);
1894 counter_u64_add(object_bypasses, 1);
1898 * Try again with this object's new backing object.
1904 * vm_object_page_remove:
1906 * For the given object, either frees or invalidates each of the
1907 * specified pages. In general, a page is freed. However, if a page is
1908 * wired for any reason other than the existence of a managed, wired
1909 * mapping, then it may be invalidated but not removed from the object.
1910 * Pages are specified by the given range ["start", "end") and the option
1911 * OBJPR_CLEANONLY. As a special case, if "end" is zero, then the range
1912 * extends from "start" to the end of the object. If the option
1913 * OBJPR_CLEANONLY is specified, then only the non-dirty pages within the
1914 * specified range are affected. If the option OBJPR_NOTMAPPED is
1915 * specified, then the pages within the specified range must have no
1916 * mappings. Otherwise, if this option is not specified, any mappings to
1917 * the specified pages are removed before the pages are freed or
1920 * In general, this operation should only be performed on objects that
1921 * contain managed pages. There are, however, two exceptions. First, it
1922 * is performed on the kernel and kmem objects by vm_map_entry_delete().
1923 * Second, it is used by msync(..., MS_INVALIDATE) to invalidate device-
1924 * backed pages. In both of these cases, the option OBJPR_CLEANONLY must
1925 * not be specified and the option OBJPR_NOTMAPPED must be specified.
1927 * The object must be locked.
1930 vm_object_page_remove(vm_object_t object, vm_pindex_t start, vm_pindex_t end,
1936 VM_OBJECT_ASSERT_WLOCKED(object);
1937 KASSERT((object->flags & OBJ_UNMANAGED) == 0 ||
1938 (options & (OBJPR_CLEANONLY | OBJPR_NOTMAPPED)) == OBJPR_NOTMAPPED,
1939 ("vm_object_page_remove: illegal options for object %p", object));
1940 if (object->resident_page_count == 0)
1942 vm_object_pip_add(object, 1);
1944 p = vm_page_find_least(object, start);
1948 * Here, the variable "p" is either (1) the page with the least pindex
1949 * greater than or equal to the parameter "start" or (2) NULL.
1951 for (; p != NULL && (p->pindex < end || end == 0); p = next) {
1952 next = TAILQ_NEXT(p, listq);
1955 * If the page is wired for any reason besides the existence
1956 * of managed, wired mappings, then it cannot be freed. For
1957 * example, fictitious pages, which represent device memory,
1958 * are inherently wired and cannot be freed. They can,
1959 * however, be invalidated if the option OBJPR_CLEANONLY is
1962 vm_page_change_lock(p, &mtx);
1963 if (vm_page_xbusied(p)) {
1964 VM_OBJECT_WUNLOCK(object);
1965 vm_page_busy_sleep(p, "vmopax", true);
1966 VM_OBJECT_WLOCK(object);
1969 if (p->wire_count != 0) {
1970 if ((options & OBJPR_NOTMAPPED) == 0 &&
1971 object->ref_count != 0)
1973 if ((options & OBJPR_CLEANONLY) == 0) {
1979 if (vm_page_busied(p)) {
1980 VM_OBJECT_WUNLOCK(object);
1981 vm_page_busy_sleep(p, "vmopar", false);
1982 VM_OBJECT_WLOCK(object);
1985 KASSERT((p->flags & PG_FICTITIOUS) == 0,
1986 ("vm_object_page_remove: page %p is fictitious", p));
1987 if ((options & OBJPR_CLEANONLY) != 0 && p->valid != 0) {
1988 if ((options & OBJPR_NOTMAPPED) == 0 &&
1989 object->ref_count != 0)
1990 pmap_remove_write(p);
1994 if ((options & OBJPR_NOTMAPPED) == 0 && object->ref_count != 0)
2000 vm_object_pip_wakeup(object);
2004 * vm_object_page_noreuse:
2006 * For the given object, attempt to move the specified pages to
2007 * the head of the inactive queue. This bypasses regular LRU
2008 * operation and allows the pages to be reused quickly under memory
2009 * pressure. If a page is wired for any reason, then it will not
2010 * be queued. Pages are specified by the range ["start", "end").
2011 * As a special case, if "end" is zero, then the range extends from
2012 * "start" to the end of the object.
2014 * This operation should only be performed on objects that
2015 * contain non-fictitious, managed pages.
2017 * The object must be locked.
2020 vm_object_page_noreuse(vm_object_t object, vm_pindex_t start, vm_pindex_t end)
2025 VM_OBJECT_ASSERT_LOCKED(object);
2026 KASSERT((object->flags & (OBJ_FICTITIOUS | OBJ_UNMANAGED)) == 0,
2027 ("vm_object_page_noreuse: illegal object %p", object));
2028 if (object->resident_page_count == 0)
2030 p = vm_page_find_least(object, start);
2033 * Here, the variable "p" is either (1) the page with the least pindex
2034 * greater than or equal to the parameter "start" or (2) NULL.
2037 for (; p != NULL && (p->pindex < end || end == 0); p = next) {
2038 next = TAILQ_NEXT(p, listq);
2039 vm_page_change_lock(p, &mtx);
2040 vm_page_deactivate_noreuse(p);
2047 * Populate the specified range of the object with valid pages. Returns
2048 * TRUE if the range is successfully populated and FALSE otherwise.
2050 * Note: This function should be optimized to pass a larger array of
2051 * pages to vm_pager_get_pages() before it is applied to a non-
2052 * OBJT_DEVICE object.
2054 * The object must be locked.
2057 vm_object_populate(vm_object_t object, vm_pindex_t start, vm_pindex_t end)
2063 VM_OBJECT_ASSERT_WLOCKED(object);
2064 for (pindex = start; pindex < end; pindex++) {
2065 m = vm_page_grab(object, pindex, VM_ALLOC_NORMAL);
2066 if (m->valid != VM_PAGE_BITS_ALL) {
2067 rv = vm_pager_get_pages(object, &m, 1, NULL, NULL);
2068 if (rv != VM_PAGER_OK) {
2076 * Keep "m" busy because a subsequent iteration may unlock
2080 if (pindex > start) {
2081 m = vm_page_lookup(object, start);
2082 while (m != NULL && m->pindex < pindex) {
2084 m = TAILQ_NEXT(m, listq);
2087 return (pindex == end);
2091 * Routine: vm_object_coalesce
2092 * Function: Coalesces two objects backing up adjoining
2093 * regions of memory into a single object.
2095 * returns TRUE if objects were combined.
2097 * NOTE: Only works at the moment if the second object is NULL -
2098 * if it's not, which object do we lock first?
2101 * prev_object First object to coalesce
2102 * prev_offset Offset into prev_object
2103 * prev_size Size of reference to prev_object
2104 * next_size Size of reference to the second object
2105 * reserved Indicator that extension region has
2106 * swap accounted for
2109 * The object must *not* be locked.
2112 vm_object_coalesce(vm_object_t prev_object, vm_ooffset_t prev_offset,
2113 vm_size_t prev_size, vm_size_t next_size, boolean_t reserved)
2115 vm_pindex_t next_pindex;
2117 if (prev_object == NULL)
2119 VM_OBJECT_WLOCK(prev_object);
2120 if ((prev_object->type != OBJT_DEFAULT &&
2121 prev_object->type != OBJT_SWAP) ||
2122 (prev_object->flags & OBJ_TMPFS_NODE) != 0) {
2123 VM_OBJECT_WUNLOCK(prev_object);
2128 * Try to collapse the object first
2130 vm_object_collapse(prev_object);
2133 * Can't coalesce if: . more than one reference . paged out . shadows
2134 * another object . has a copy elsewhere (any of which mean that the
2135 * pages not mapped to prev_entry may be in use anyway)
2137 if (prev_object->backing_object != NULL) {
2138 VM_OBJECT_WUNLOCK(prev_object);
2142 prev_size >>= PAGE_SHIFT;
2143 next_size >>= PAGE_SHIFT;
2144 next_pindex = OFF_TO_IDX(prev_offset) + prev_size;
2146 if (prev_object->ref_count > 1 &&
2147 prev_object->size != next_pindex &&
2148 (prev_object->flags & OBJ_ONEMAPPING) == 0) {
2149 VM_OBJECT_WUNLOCK(prev_object);
2154 * Account for the charge.
2156 if (prev_object->cred != NULL) {
2159 * If prev_object was charged, then this mapping,
2160 * although not charged now, may become writable
2161 * later. Non-NULL cred in the object would prevent
2162 * swap reservation during enabling of the write
2163 * access, so reserve swap now. Failed reservation
2164 * cause allocation of the separate object for the map
2165 * entry, and swap reservation for this entry is
2166 * managed in appropriate time.
2168 if (!reserved && !swap_reserve_by_cred(ptoa(next_size),
2169 prev_object->cred)) {
2170 VM_OBJECT_WUNLOCK(prev_object);
2173 prev_object->charge += ptoa(next_size);
2177 * Remove any pages that may still be in the object from a previous
2180 if (next_pindex < prev_object->size) {
2181 vm_object_page_remove(prev_object, next_pindex, next_pindex +
2183 if (prev_object->type == OBJT_SWAP)
2184 swap_pager_freespace(prev_object,
2185 next_pindex, next_size);
2187 if (prev_object->cred != NULL) {
2188 KASSERT(prev_object->charge >=
2189 ptoa(prev_object->size - next_pindex),
2190 ("object %p overcharged 1 %jx %jx", prev_object,
2191 (uintmax_t)next_pindex, (uintmax_t)next_size));
2192 prev_object->charge -= ptoa(prev_object->size -
2199 * Extend the object if necessary.
2201 if (next_pindex + next_size > prev_object->size)
2202 prev_object->size = next_pindex + next_size;
2204 VM_OBJECT_WUNLOCK(prev_object);
2209 vm_object_set_writeable_dirty(vm_object_t object)
2212 VM_OBJECT_ASSERT_WLOCKED(object);
2213 if (object->type != OBJT_VNODE) {
2214 if ((object->flags & OBJ_TMPFS_NODE) != 0) {
2215 KASSERT(object->type == OBJT_SWAP, ("non-swap tmpfs"));
2216 vm_object_set_flag(object, OBJ_TMPFS_DIRTY);
2220 object->generation++;
2221 if ((object->flags & OBJ_MIGHTBEDIRTY) != 0)
2223 vm_object_set_flag(object, OBJ_MIGHTBEDIRTY);
2229 * For each page offset within the specified range of the given object,
2230 * find the highest-level page in the shadow chain and unwire it. A page
2231 * must exist at every page offset, and the highest-level page must be
2235 vm_object_unwire(vm_object_t object, vm_ooffset_t offset, vm_size_t length,
2238 vm_object_t tobject, t1object;
2240 vm_pindex_t end_pindex, pindex, tpindex;
2241 int depth, locked_depth;
2243 KASSERT((offset & PAGE_MASK) == 0,
2244 ("vm_object_unwire: offset is not page aligned"));
2245 KASSERT((length & PAGE_MASK) == 0,
2246 ("vm_object_unwire: length is not a multiple of PAGE_SIZE"));
2247 /* The wired count of a fictitious page never changes. */
2248 if ((object->flags & OBJ_FICTITIOUS) != 0)
2250 pindex = OFF_TO_IDX(offset);
2251 end_pindex = pindex + atop(length);
2254 VM_OBJECT_RLOCK(object);
2255 m = vm_page_find_least(object, pindex);
2256 while (pindex < end_pindex) {
2257 if (m == NULL || pindex < m->pindex) {
2259 * The first object in the shadow chain doesn't
2260 * contain a page at the current index. Therefore,
2261 * the page must exist in a backing object.
2268 OFF_TO_IDX(tobject->backing_object_offset);
2269 tobject = tobject->backing_object;
2270 KASSERT(tobject != NULL,
2271 ("vm_object_unwire: missing page"));
2272 if ((tobject->flags & OBJ_FICTITIOUS) != 0)
2275 if (depth == locked_depth) {
2277 VM_OBJECT_RLOCK(tobject);
2279 } while ((tm = vm_page_lookup(tobject, tpindex)) ==
2283 m = TAILQ_NEXT(m, listq);
2286 if (vm_page_xbusied(tm)) {
2287 for (tobject = object; locked_depth >= 1;
2289 t1object = tobject->backing_object;
2290 VM_OBJECT_RUNLOCK(tobject);
2293 vm_page_busy_sleep(tm, "unwbo", true);
2296 vm_page_unwire(tm, queue);
2301 /* Release the accumulated object locks. */
2302 for (tobject = object; locked_depth >= 1; locked_depth--) {
2303 t1object = tobject->backing_object;
2304 VM_OBJECT_RUNLOCK(tobject);
2310 * Return the vnode for the given object, or NULL if none exists.
2311 * For tmpfs objects, the function may return NULL if there is
2312 * no vnode allocated at the time of the call.
2315 vm_object_vnode(vm_object_t object)
2319 VM_OBJECT_ASSERT_LOCKED(object);
2320 if (object->type == OBJT_VNODE) {
2321 vp = object->handle;
2322 KASSERT(vp != NULL, ("%s: OBJT_VNODE has no vnode", __func__));
2323 } else if (object->type == OBJT_SWAP &&
2324 (object->flags & OBJ_TMPFS) != 0) {
2325 vp = object->un_pager.swp.swp_tmpfs;
2326 KASSERT(vp != NULL, ("%s: OBJT_TMPFS has no vnode", __func__));
2334 * Return the kvme type of the given object.
2335 * If vpp is not NULL, set it to the object's vm_object_vnode() or NULL.
2338 vm_object_kvme_type(vm_object_t object, struct vnode **vpp)
2341 VM_OBJECT_ASSERT_LOCKED(object);
2343 *vpp = vm_object_vnode(object);
2344 switch (object->type) {
2346 return (KVME_TYPE_DEFAULT);
2348 return (KVME_TYPE_VNODE);
2350 if ((object->flags & OBJ_TMPFS_NODE) != 0)
2351 return (KVME_TYPE_VNODE);
2352 return (KVME_TYPE_SWAP);
2354 return (KVME_TYPE_DEVICE);
2356 return (KVME_TYPE_PHYS);
2358 return (KVME_TYPE_DEAD);
2360 return (KVME_TYPE_SG);
2361 case OBJT_MGTDEVICE:
2362 return (KVME_TYPE_MGTDEVICE);
2364 return (KVME_TYPE_UNKNOWN);
2369 sysctl_vm_object_list(SYSCTL_HANDLER_ARGS)
2371 struct kinfo_vmobject *kvo;
2372 char *fullpath, *freepath;
2379 if (req->oldptr == NULL) {
2381 * If an old buffer has not been provided, generate an
2382 * estimate of the space needed for a subsequent call.
2384 mtx_lock(&vm_object_list_mtx);
2386 TAILQ_FOREACH(obj, &vm_object_list, object_list) {
2387 if (obj->type == OBJT_DEAD)
2391 mtx_unlock(&vm_object_list_mtx);
2392 return (SYSCTL_OUT(req, NULL, sizeof(struct kinfo_vmobject) *
2396 kvo = malloc(sizeof(*kvo), M_TEMP, M_WAITOK);
2400 * VM objects are type stable and are never removed from the
2401 * list once added. This allows us to safely read obj->object_list
2402 * after reacquiring the VM object lock.
2404 mtx_lock(&vm_object_list_mtx);
2405 TAILQ_FOREACH(obj, &vm_object_list, object_list) {
2406 if (obj->type == OBJT_DEAD)
2408 VM_OBJECT_RLOCK(obj);
2409 if (obj->type == OBJT_DEAD) {
2410 VM_OBJECT_RUNLOCK(obj);
2413 mtx_unlock(&vm_object_list_mtx);
2414 kvo->kvo_size = ptoa(obj->size);
2415 kvo->kvo_resident = obj->resident_page_count;
2416 kvo->kvo_ref_count = obj->ref_count;
2417 kvo->kvo_shadow_count = obj->shadow_count;
2418 kvo->kvo_memattr = obj->memattr;
2419 kvo->kvo_active = 0;
2420 kvo->kvo_inactive = 0;
2421 TAILQ_FOREACH(m, &obj->memq, listq) {
2423 * A page may belong to the object but be
2424 * dequeued and set to PQ_NONE while the
2425 * object lock is not held. This makes the
2426 * reads of m->queue below racy, and we do not
2427 * count pages set to PQ_NONE. However, this
2428 * sysctl is only meant to give an
2429 * approximation of the system anyway.
2431 if (m->queue == PQ_ACTIVE)
2433 else if (m->queue == PQ_INACTIVE)
2434 kvo->kvo_inactive++;
2437 kvo->kvo_vn_fileid = 0;
2438 kvo->kvo_vn_fsid = 0;
2439 kvo->kvo_vn_fsid_freebsd11 = 0;
2442 kvo->kvo_type = vm_object_kvme_type(obj, &vp);
2445 VM_OBJECT_RUNLOCK(obj);
2447 vn_fullpath(curthread, vp, &fullpath, &freepath);
2448 vn_lock(vp, LK_SHARED | LK_RETRY);
2449 if (VOP_GETATTR(vp, &va, curthread->td_ucred) == 0) {
2450 kvo->kvo_vn_fileid = va.va_fileid;
2451 kvo->kvo_vn_fsid = va.va_fsid;
2452 kvo->kvo_vn_fsid_freebsd11 = va.va_fsid;
2458 strlcpy(kvo->kvo_path, fullpath, sizeof(kvo->kvo_path));
2459 if (freepath != NULL)
2460 free(freepath, M_TEMP);
2462 /* Pack record size down */
2463 kvo->kvo_structsize = offsetof(struct kinfo_vmobject, kvo_path)
2464 + strlen(kvo->kvo_path) + 1;
2465 kvo->kvo_structsize = roundup(kvo->kvo_structsize,
2467 error = SYSCTL_OUT(req, kvo, kvo->kvo_structsize);
2468 mtx_lock(&vm_object_list_mtx);
2472 mtx_unlock(&vm_object_list_mtx);
2476 SYSCTL_PROC(_vm, OID_AUTO, objects, CTLTYPE_STRUCT | CTLFLAG_RW | CTLFLAG_SKIP |
2477 CTLFLAG_MPSAFE, NULL, 0, sysctl_vm_object_list, "S,kinfo_vmobject",
2478 "List of VM objects");
2480 #include "opt_ddb.h"
2482 #include <sys/kernel.h>
2484 #include <sys/cons.h>
2486 #include <ddb/ddb.h>
2489 _vm_object_in_map(vm_map_t map, vm_object_t object, vm_map_entry_t entry)
2492 vm_map_entry_t tmpe;
2500 tmpe = map->header.next;
2501 entcount = map->nentries;
2502 while (entcount-- && (tmpe != &map->header)) {
2503 if (_vm_object_in_map(map, object, tmpe)) {
2508 } else if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
2509 tmpm = entry->object.sub_map;
2510 tmpe = tmpm->header.next;
2511 entcount = tmpm->nentries;
2512 while (entcount-- && tmpe != &tmpm->header) {
2513 if (_vm_object_in_map(tmpm, object, tmpe)) {
2518 } else if ((obj = entry->object.vm_object) != NULL) {
2519 for (; obj; obj = obj->backing_object)
2520 if (obj == object) {
2528 vm_object_in_map(vm_object_t object)
2532 /* sx_slock(&allproc_lock); */
2533 FOREACH_PROC_IN_SYSTEM(p) {
2534 if (!p->p_vmspace /* || (p->p_flag & (P_SYSTEM|P_WEXIT)) */)
2536 if (_vm_object_in_map(&p->p_vmspace->vm_map, object, 0)) {
2537 /* sx_sunlock(&allproc_lock); */
2541 /* sx_sunlock(&allproc_lock); */
2542 if (_vm_object_in_map(kernel_map, object, 0))
2547 DB_SHOW_COMMAND(vmochk, vm_object_check)
2552 * make sure that internal objs are in a map somewhere
2553 * and none have zero ref counts.
2555 TAILQ_FOREACH(object, &vm_object_list, object_list) {
2556 if (object->handle == NULL &&
2557 (object->type == OBJT_DEFAULT || object->type == OBJT_SWAP)) {
2558 if (object->ref_count == 0) {
2559 db_printf("vmochk: internal obj has zero ref count: %ld\n",
2560 (long)object->size);
2562 if (!vm_object_in_map(object)) {
2564 "vmochk: internal obj is not in a map: "
2565 "ref: %d, size: %lu: 0x%lx, backing_object: %p\n",
2566 object->ref_count, (u_long)object->size,
2567 (u_long)object->size,
2568 (void *)object->backing_object);
2575 * vm_object_print: [ debug ]
2577 DB_SHOW_COMMAND(object, vm_object_print_static)
2579 /* XXX convert args. */
2580 vm_object_t object = (vm_object_t)addr;
2581 boolean_t full = have_addr;
2585 /* XXX count is an (unused) arg. Avoid shadowing it. */
2586 #define count was_count
2594 "Object %p: type=%d, size=0x%jx, res=%d, ref=%d, flags=0x%x ruid %d charge %jx\n",
2595 object, (int)object->type, (uintmax_t)object->size,
2596 object->resident_page_count, object->ref_count, object->flags,
2597 object->cred ? object->cred->cr_ruid : -1, (uintmax_t)object->charge);
2598 db_iprintf(" sref=%d, backing_object(%d)=(%p)+0x%jx\n",
2599 object->shadow_count,
2600 object->backing_object ? object->backing_object->ref_count : 0,
2601 object->backing_object, (uintmax_t)object->backing_object_offset);
2608 TAILQ_FOREACH(p, &object->memq, listq) {
2610 db_iprintf("memory:=");
2611 else if (count == 6) {
2619 db_printf("(off=0x%jx,page=0x%jx)",
2620 (uintmax_t)p->pindex, (uintmax_t)VM_PAGE_TO_PHYS(p));
2630 /* XXX need this non-static entry for calling from vm_map_print. */
2633 /* db_expr_t */ long addr,
2634 boolean_t have_addr,
2635 /* db_expr_t */ long count,
2638 vm_object_print_static(addr, have_addr, count, modif);
2641 DB_SHOW_COMMAND(vmopag, vm_object_print_pages)
2646 vm_page_t m, prev_m;
2650 TAILQ_FOREACH(object, &vm_object_list, object_list) {
2651 db_printf("new object: %p\n", (void *)object);
2662 TAILQ_FOREACH(m, &object->memq, listq) {
2663 if (m->pindex > 128)
2665 if ((prev_m = TAILQ_PREV(m, pglist, listq)) != NULL &&
2666 prev_m->pindex + 1 != m->pindex) {
2668 db_printf(" index(%ld)run(%d)pa(0x%lx)\n",
2669 (long)fidx, rcount, (long)pa);
2681 (VM_PAGE_TO_PHYS(m) == pa + rcount * PAGE_SIZE)) {
2686 db_printf(" index(%ld)run(%d)pa(0x%lx)\n",
2687 (long)fidx, rcount, (long)pa);
2697 pa = VM_PAGE_TO_PHYS(m);
2701 db_printf(" index(%ld)run(%d)pa(0x%lx)\n",
2702 (long)fidx, rcount, (long)pa);