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/swap_pager.h>
100 #include <vm/vm_kern.h>
101 #include <vm/vm_extern.h>
102 #include <vm/vm_radix.h>
103 #include <vm/vm_reserv.h>
106 static int old_msync;
107 SYSCTL_INT(_vm, OID_AUTO, old_msync, CTLFLAG_RW, &old_msync, 0,
108 "Use old (insecure) msync behavior");
110 static int vm_object_page_collect_flush(vm_object_t object, vm_page_t p,
111 int pagerflags, int flags, boolean_t *clearobjflags,
113 static boolean_t vm_object_page_remove_write(vm_page_t p, int flags,
114 boolean_t *clearobjflags);
115 static void vm_object_qcollapse(vm_object_t object);
116 static void vm_object_vndeallocate(vm_object_t object);
119 * Virtual memory objects maintain the actual data
120 * associated with allocated virtual memory. A given
121 * page of memory exists within exactly one object.
123 * An object is only deallocated when all "references"
124 * are given up. Only one "reference" to a given
125 * region of an object should be writeable.
127 * Associated with each object is a list of all resident
128 * memory pages belonging to that object; this list is
129 * maintained by the "vm_page" module, and locked by the object's
132 * Each object also records a "pager" routine which is
133 * used to retrieve (and store) pages to the proper backing
134 * storage. In addition, objects may be backed by other
135 * objects from which they were virtual-copied.
137 * The only items within the object structure which are
138 * modified after time of creation are:
139 * reference count locked by object's lock
140 * pager routine locked by object's lock
144 struct object_q vm_object_list;
145 struct mtx vm_object_list_mtx; /* lock for object list and count */
147 struct vm_object kernel_object_store;
149 static SYSCTL_NODE(_vm_stats, OID_AUTO, object, CTLFLAG_RD, 0,
152 static counter_u64_t object_collapses = EARLY_COUNTER;
153 SYSCTL_COUNTER_U64(_vm_stats_object, OID_AUTO, collapses, CTLFLAG_RD,
155 "VM object collapses");
157 static counter_u64_t object_bypasses = EARLY_COUNTER;
158 SYSCTL_COUNTER_U64(_vm_stats_object, OID_AUTO, bypasses, CTLFLAG_RD,
160 "VM object bypasses");
163 counter_startup(void)
166 object_collapses = counter_u64_alloc(M_WAITOK);
167 object_bypasses = counter_u64_alloc(M_WAITOK);
169 SYSINIT(object_counters, SI_SUB_CPU, SI_ORDER_ANY, counter_startup, NULL);
171 static uma_zone_t obj_zone;
173 static int vm_object_zinit(void *mem, int size, int flags);
176 static void vm_object_zdtor(void *mem, int size, void *arg);
179 vm_object_zdtor(void *mem, int size, void *arg)
183 object = (vm_object_t)mem;
184 KASSERT(object->ref_count == 0,
185 ("object %p ref_count = %d", object, object->ref_count));
186 KASSERT(TAILQ_EMPTY(&object->memq),
187 ("object %p has resident pages in its memq", object));
188 KASSERT(vm_radix_is_empty(&object->rtree),
189 ("object %p has resident pages in its trie", object));
190 #if VM_NRESERVLEVEL > 0
191 KASSERT(LIST_EMPTY(&object->rvq),
192 ("object %p has reservations",
195 KASSERT(object->paging_in_progress == 0,
196 ("object %p paging_in_progress = %d",
197 object, object->paging_in_progress));
198 KASSERT(object->resident_page_count == 0,
199 ("object %p resident_page_count = %d",
200 object, object->resident_page_count));
201 KASSERT(object->shadow_count == 0,
202 ("object %p shadow_count = %d",
203 object, object->shadow_count));
204 KASSERT(object->type == OBJT_DEAD,
205 ("object %p has non-dead type %d",
206 object, object->type));
211 vm_object_zinit(void *mem, int size, int flags)
215 object = (vm_object_t)mem;
216 rw_init_flags(&object->lock, "vm object", RW_DUPOK | RW_NEW);
218 /* These are true for any object that has been freed */
219 object->type = OBJT_DEAD;
220 object->ref_count = 0;
221 vm_radix_init(&object->rtree);
222 object->paging_in_progress = 0;
223 object->resident_page_count = 0;
224 object->shadow_count = 0;
225 object->flags = OBJ_DEAD;
227 mtx_lock(&vm_object_list_mtx);
228 TAILQ_INSERT_TAIL(&vm_object_list, object, object_list);
229 mtx_unlock(&vm_object_list_mtx);
234 _vm_object_allocate(objtype_t type, vm_pindex_t size, vm_object_t object)
237 TAILQ_INIT(&object->memq);
238 LIST_INIT(&object->shadow_head);
241 if (type == OBJT_SWAP)
242 pctrie_init(&object->un_pager.swp.swp_blks);
245 * Ensure that swap_pager_swapoff() iteration over object_list
246 * sees up to date type and pctrie head if it observed
249 atomic_thread_fence_rel();
253 panic("_vm_object_allocate: can't create OBJT_DEAD");
256 object->flags = OBJ_ONEMAPPING;
260 object->flags = OBJ_FICTITIOUS | OBJ_UNMANAGED;
263 object->flags = OBJ_FICTITIOUS;
266 object->flags = OBJ_UNMANAGED;
272 panic("_vm_object_allocate: type %d is undefined", type);
275 object->generation = 1;
276 object->ref_count = 1;
277 object->memattr = VM_MEMATTR_DEFAULT;
280 object->handle = NULL;
281 object->backing_object = NULL;
282 object->backing_object_offset = (vm_ooffset_t) 0;
283 #if VM_NRESERVLEVEL > 0
284 LIST_INIT(&object->rvq);
286 umtx_shm_object_init(object);
292 * Initialize the VM objects module.
297 TAILQ_INIT(&vm_object_list);
298 mtx_init(&vm_object_list_mtx, "vm object_list", NULL, MTX_DEF);
300 rw_init(&kernel_object->lock, "kernel vm object");
301 _vm_object_allocate(OBJT_PHYS, atop(VM_MAX_KERNEL_ADDRESS -
302 VM_MIN_KERNEL_ADDRESS), kernel_object);
303 #if VM_NRESERVLEVEL > 0
304 kernel_object->flags |= OBJ_COLORED;
305 kernel_object->pg_color = (u_short)atop(VM_MIN_KERNEL_ADDRESS);
309 * The lock portion of struct vm_object must be type stable due
310 * to vm_pageout_fallback_object_lock locking a vm object
311 * without holding any references to it.
313 obj_zone = uma_zcreate("VM OBJECT", sizeof (struct vm_object), NULL,
319 vm_object_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
325 vm_object_clear_flag(vm_object_t object, u_short bits)
328 VM_OBJECT_ASSERT_WLOCKED(object);
329 object->flags &= ~bits;
333 * Sets the default memory attribute for the specified object. Pages
334 * that are allocated to this object are by default assigned this memory
337 * Presently, this function must be called before any pages are allocated
338 * to the object. In the future, this requirement may be relaxed for
339 * "default" and "swap" objects.
342 vm_object_set_memattr(vm_object_t object, vm_memattr_t memattr)
345 VM_OBJECT_ASSERT_WLOCKED(object);
346 switch (object->type) {
354 if (!TAILQ_EMPTY(&object->memq))
355 return (KERN_FAILURE);
358 return (KERN_INVALID_ARGUMENT);
360 panic("vm_object_set_memattr: object %p is of undefined type",
363 object->memattr = memattr;
364 return (KERN_SUCCESS);
368 vm_object_pip_add(vm_object_t object, short i)
371 VM_OBJECT_ASSERT_WLOCKED(object);
372 object->paging_in_progress += i;
376 vm_object_pip_subtract(vm_object_t object, short i)
379 VM_OBJECT_ASSERT_WLOCKED(object);
380 object->paging_in_progress -= i;
384 vm_object_pip_wakeup(vm_object_t object)
387 VM_OBJECT_ASSERT_WLOCKED(object);
388 object->paging_in_progress--;
389 if ((object->flags & OBJ_PIPWNT) && object->paging_in_progress == 0) {
390 vm_object_clear_flag(object, OBJ_PIPWNT);
396 vm_object_pip_wakeupn(vm_object_t object, short i)
399 VM_OBJECT_ASSERT_WLOCKED(object);
401 object->paging_in_progress -= i;
402 if ((object->flags & OBJ_PIPWNT) && object->paging_in_progress == 0) {
403 vm_object_clear_flag(object, OBJ_PIPWNT);
409 vm_object_pip_wait(vm_object_t object, char *waitid)
412 VM_OBJECT_ASSERT_WLOCKED(object);
413 while (object->paging_in_progress) {
414 object->flags |= OBJ_PIPWNT;
415 VM_OBJECT_SLEEP(object, object, PVM, waitid, 0);
420 * vm_object_allocate:
422 * Returns a new object with the given size.
425 vm_object_allocate(objtype_t type, vm_pindex_t size)
429 object = (vm_object_t)uma_zalloc(obj_zone, M_WAITOK);
430 _vm_object_allocate(type, size, object);
436 * vm_object_reference:
438 * Gets another reference to the given object. Note: OBJ_DEAD
439 * objects can be referenced during final cleaning.
442 vm_object_reference(vm_object_t object)
446 VM_OBJECT_WLOCK(object);
447 vm_object_reference_locked(object);
448 VM_OBJECT_WUNLOCK(object);
452 * vm_object_reference_locked:
454 * Gets another reference to the given object.
456 * The object must be locked.
459 vm_object_reference_locked(vm_object_t object)
463 VM_OBJECT_ASSERT_WLOCKED(object);
465 if (object->type == OBJT_VNODE) {
472 * Handle deallocating an object of type OBJT_VNODE.
475 vm_object_vndeallocate(vm_object_t object)
477 struct vnode *vp = (struct vnode *) object->handle;
479 VM_OBJECT_ASSERT_WLOCKED(object);
480 KASSERT(object->type == OBJT_VNODE,
481 ("vm_object_vndeallocate: not a vnode object"));
482 KASSERT(vp != NULL, ("vm_object_vndeallocate: missing vp"));
484 if (object->ref_count == 0) {
485 vn_printf(vp, "vm_object_vndeallocate ");
486 panic("vm_object_vndeallocate: bad object reference count");
490 if (!umtx_shm_vnobj_persistent && object->ref_count == 1)
491 umtx_shm_object_terminated(object);
494 * The test for text of vp vnode does not need a bypass to
495 * reach right VV_TEXT there, since it is obtained from
498 if (object->ref_count > 1 || (vp->v_vflag & VV_TEXT) == 0) {
500 VM_OBJECT_WUNLOCK(object);
501 /* vrele may need the vnode lock. */
505 VM_OBJECT_WUNLOCK(object);
506 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
508 VM_OBJECT_WLOCK(object);
510 if (object->type == OBJT_DEAD) {
511 VM_OBJECT_WUNLOCK(object);
514 if (object->ref_count == 0)
516 VM_OBJECT_WUNLOCK(object);
523 * vm_object_deallocate:
525 * Release a reference to the specified object,
526 * gained either through a vm_object_allocate
527 * or a vm_object_reference call. When all references
528 * are gone, storage associated with this object
529 * may be relinquished.
531 * No object may be locked.
534 vm_object_deallocate(vm_object_t object)
539 while (object != NULL) {
540 VM_OBJECT_WLOCK(object);
541 if (object->type == OBJT_VNODE) {
542 vm_object_vndeallocate(object);
546 KASSERT(object->ref_count != 0,
547 ("vm_object_deallocate: object deallocated too many times: %d", object->type));
550 * If the reference count goes to 0 we start calling
551 * vm_object_terminate() on the object chain.
552 * A ref count of 1 may be a special case depending on the
553 * shadow count being 0 or 1.
556 if (object->ref_count > 1) {
557 VM_OBJECT_WUNLOCK(object);
559 } else if (object->ref_count == 1) {
560 if (object->type == OBJT_SWAP &&
561 (object->flags & OBJ_TMPFS) != 0) {
562 vp = object->un_pager.swp.swp_tmpfs;
564 VM_OBJECT_WUNLOCK(object);
565 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
566 VM_OBJECT_WLOCK(object);
567 if (object->type == OBJT_DEAD ||
568 object->ref_count != 1) {
569 VM_OBJECT_WUNLOCK(object);
574 if ((object->flags & OBJ_TMPFS) != 0)
579 if (object->shadow_count == 0 &&
580 object->handle == NULL &&
581 (object->type == OBJT_DEFAULT ||
582 (object->type == OBJT_SWAP &&
583 (object->flags & OBJ_TMPFS_NODE) == 0))) {
584 vm_object_set_flag(object, OBJ_ONEMAPPING);
585 } else if ((object->shadow_count == 1) &&
586 (object->handle == NULL) &&
587 (object->type == OBJT_DEFAULT ||
588 object->type == OBJT_SWAP)) {
591 robject = LIST_FIRST(&object->shadow_head);
592 KASSERT(robject != NULL,
593 ("vm_object_deallocate: ref_count: %d, shadow_count: %d",
595 object->shadow_count));
596 KASSERT((robject->flags & OBJ_TMPFS_NODE) == 0,
597 ("shadowed tmpfs v_object %p", object));
598 if (!VM_OBJECT_TRYWLOCK(robject)) {
600 * Avoid a potential deadlock.
603 VM_OBJECT_WUNLOCK(object);
605 * More likely than not the thread
606 * holding robject's lock has lower
607 * priority than the current thread.
608 * Let the lower priority thread run.
614 * Collapse object into its shadow unless its
615 * shadow is dead. In that case, object will
616 * be deallocated by the thread that is
617 * deallocating its shadow.
619 if ((robject->flags & OBJ_DEAD) == 0 &&
620 (robject->handle == NULL) &&
621 (robject->type == OBJT_DEFAULT ||
622 robject->type == OBJT_SWAP)) {
624 robject->ref_count++;
626 if (robject->paging_in_progress) {
627 VM_OBJECT_WUNLOCK(object);
628 vm_object_pip_wait(robject,
630 temp = robject->backing_object;
631 if (object == temp) {
632 VM_OBJECT_WLOCK(object);
635 } else if (object->paging_in_progress) {
636 VM_OBJECT_WUNLOCK(robject);
637 object->flags |= OBJ_PIPWNT;
638 VM_OBJECT_SLEEP(object, object,
639 PDROP | PVM, "objde2", 0);
640 VM_OBJECT_WLOCK(robject);
641 temp = robject->backing_object;
642 if (object == temp) {
643 VM_OBJECT_WLOCK(object);
647 VM_OBJECT_WUNLOCK(object);
649 if (robject->ref_count == 1) {
650 robject->ref_count--;
655 vm_object_collapse(object);
656 VM_OBJECT_WUNLOCK(object);
659 VM_OBJECT_WUNLOCK(robject);
661 VM_OBJECT_WUNLOCK(object);
665 umtx_shm_object_terminated(object);
666 temp = object->backing_object;
668 KASSERT((object->flags & OBJ_TMPFS_NODE) == 0,
669 ("shadowed tmpfs v_object 2 %p", object));
670 VM_OBJECT_WLOCK(temp);
671 LIST_REMOVE(object, shadow_list);
672 temp->shadow_count--;
673 VM_OBJECT_WUNLOCK(temp);
674 object->backing_object = NULL;
677 * Don't double-terminate, we could be in a termination
678 * recursion due to the terminate having to sync data
681 if ((object->flags & OBJ_DEAD) == 0)
682 vm_object_terminate(object);
684 VM_OBJECT_WUNLOCK(object);
690 * vm_object_destroy removes the object from the global object list
691 * and frees the space for the object.
694 vm_object_destroy(vm_object_t object)
698 * Release the allocation charge.
700 if (object->cred != NULL) {
701 swap_release_by_cred(object->charge, object->cred);
703 crfree(object->cred);
708 * Free the space for the object.
710 uma_zfree(obj_zone, object);
714 * vm_object_terminate_pages removes any remaining pageable pages
715 * from the object and resets the object to an empty state.
718 vm_object_terminate_pages(vm_object_t object)
721 struct mtx *mtx, *mtx1;
722 struct vm_pagequeue *pq, *pq1;
725 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 mtx1 = vm_page_lockptr(p);
748 vm_pagequeue_cnt_add(pq, dequeued);
749 vm_pagequeue_unlock(pq);
757 if (p->wire_count != 0)
760 p->flags &= ~PG_ZERO;
761 if (p->queue != PQ_NONE) {
762 KASSERT(p->queue < PQ_COUNT, ("vm_object_terminate: "
763 "page %p is not queued", p));
764 pq1 = vm_page_pagequeue(p);
767 vm_pagequeue_cnt_add(pq, dequeued);
768 vm_pagequeue_unlock(pq);
771 vm_pagequeue_lock(pq);
775 TAILQ_REMOVE(&pq->pq_pl, p, plinks.q);
778 if (vm_page_free_prep(p, true))
781 TAILQ_REMOVE(&object->memq, p, listq);
784 vm_pagequeue_cnt_add(pq, dequeued);
785 vm_pagequeue_unlock(pq);
790 vm_page_free_phys_pglist(&object->memq);
793 * If the object contained any pages, then reset it to an empty state.
794 * None of the object's fields, including "resident_page_count", were
795 * modified by the preceding loop.
797 if (object->resident_page_count != 0) {
798 vm_radix_reclaim_allnodes(&object->rtree);
799 TAILQ_INIT(&object->memq);
800 object->resident_page_count = 0;
801 if (object->type == OBJT_VNODE)
802 vdrop(object->handle);
807 * vm_object_terminate actually destroys the specified object, freeing
808 * up all previously used resources.
810 * The object must be locked.
811 * This routine may block.
814 vm_object_terminate(vm_object_t object)
817 VM_OBJECT_ASSERT_WLOCKED(object);
820 * Make sure no one uses us.
822 vm_object_set_flag(object, OBJ_DEAD);
825 * wait for the pageout daemon to be done with the object
827 vm_object_pip_wait(object, "objtrm");
829 KASSERT(!object->paging_in_progress,
830 ("vm_object_terminate: pageout in progress"));
833 * Clean and free the pages, as appropriate. All references to the
834 * object are gone, so we don't need to lock it.
836 if (object->type == OBJT_VNODE) {
837 struct vnode *vp = (struct vnode *)object->handle;
840 * Clean pages and flush buffers.
842 vm_object_page_clean(object, 0, 0, OBJPC_SYNC);
843 VM_OBJECT_WUNLOCK(object);
845 vinvalbuf(vp, V_SAVE, 0, 0);
847 BO_LOCK(&vp->v_bufobj);
848 vp->v_bufobj.bo_flag |= BO_DEAD;
849 BO_UNLOCK(&vp->v_bufobj);
851 VM_OBJECT_WLOCK(object);
854 KASSERT(object->ref_count == 0,
855 ("vm_object_terminate: object with references, ref_count=%d",
858 if ((object->flags & OBJ_PG_DTOR) == 0)
859 vm_object_terminate_pages(object);
861 #if VM_NRESERVLEVEL > 0
862 if (__predict_false(!LIST_EMPTY(&object->rvq)))
863 vm_reserv_break_all(object);
866 KASSERT(object->cred == NULL || object->type == OBJT_DEFAULT ||
867 object->type == OBJT_SWAP,
868 ("%s: non-swap obj %p has cred", __func__, object));
871 * Let the pager know object is dead.
873 vm_pager_deallocate(object);
874 VM_OBJECT_WUNLOCK(object);
876 vm_object_destroy(object);
880 * Make the page read-only so that we can clear the object flags. However, if
881 * this is a nosync mmap then the object is likely to stay dirty so do not
882 * mess with the page and do not clear the object flags. Returns TRUE if the
883 * page should be flushed, and FALSE otherwise.
886 vm_object_page_remove_write(vm_page_t p, int flags, boolean_t *clearobjflags)
890 * If we have been asked to skip nosync pages and this is a
891 * nosync page, skip it. Note that the object flags were not
892 * cleared in this case so we do not have to set them.
894 if ((flags & OBJPC_NOSYNC) != 0 && (p->oflags & VPO_NOSYNC) != 0) {
895 *clearobjflags = FALSE;
898 pmap_remove_write(p);
899 return (p->dirty != 0);
904 * vm_object_page_clean
906 * Clean all dirty pages in the specified range of object. Leaves page
907 * on whatever queue it is currently on. If NOSYNC is set then do not
908 * write out pages with VPO_NOSYNC set (originally comes from MAP_NOSYNC),
909 * leaving the object dirty.
911 * When stuffing pages asynchronously, allow clustering. XXX we need a
912 * synchronous clustering mode implementation.
914 * Odd semantics: if start == end, we clean everything.
916 * The object must be locked.
918 * Returns FALSE if some page from the range was not written, as
919 * reported by the pager, and TRUE otherwise.
922 vm_object_page_clean(vm_object_t object, vm_ooffset_t start, vm_ooffset_t end,
926 vm_pindex_t pi, tend, tstart;
927 int curgeneration, n, pagerflags;
928 boolean_t clearobjflags, eio, res;
930 VM_OBJECT_ASSERT_WLOCKED(object);
933 * The OBJ_MIGHTBEDIRTY flag is only set for OBJT_VNODE
934 * objects. The check below prevents the function from
935 * operating on non-vnode objects.
937 if ((object->flags & OBJ_MIGHTBEDIRTY) == 0 ||
938 object->resident_page_count == 0)
941 pagerflags = (flags & (OBJPC_SYNC | OBJPC_INVAL)) != 0 ?
942 VM_PAGER_PUT_SYNC : VM_PAGER_CLUSTER_OK;
943 pagerflags |= (flags & OBJPC_INVAL) != 0 ? VM_PAGER_PUT_INVAL : 0;
945 tstart = OFF_TO_IDX(start);
946 tend = (end == 0) ? object->size : OFF_TO_IDX(end + PAGE_MASK);
947 clearobjflags = tstart == 0 && tend >= object->size;
951 curgeneration = object->generation;
953 for (p = vm_page_find_least(object, tstart); p != NULL; p = np) {
957 np = TAILQ_NEXT(p, listq);
960 if (vm_page_sleep_if_busy(p, "vpcwai")) {
961 if (object->generation != curgeneration) {
962 if ((flags & OBJPC_SYNC) != 0)
965 clearobjflags = FALSE;
967 np = vm_page_find_least(object, pi);
970 if (!vm_object_page_remove_write(p, flags, &clearobjflags))
973 n = vm_object_page_collect_flush(object, p, pagerflags,
974 flags, &clearobjflags, &eio);
977 clearobjflags = FALSE;
979 if (object->generation != curgeneration) {
980 if ((flags & OBJPC_SYNC) != 0)
983 clearobjflags = FALSE;
987 * If the VOP_PUTPAGES() did a truncated write, so
988 * that even the first page of the run is not fully
989 * written, vm_pageout_flush() returns 0 as the run
990 * length. Since the condition that caused truncated
991 * write may be permanent, e.g. exhausted free space,
992 * accepting n == 0 would cause an infinite loop.
994 * Forwarding the iterator leaves the unwritten page
995 * behind, but there is not much we can do there if
996 * filesystem refuses to write it.
1000 clearobjflags = FALSE;
1002 np = vm_page_find_least(object, pi + n);
1005 VOP_FSYNC(vp, (pagerflags & VM_PAGER_PUT_SYNC) ? MNT_WAIT : 0);
1009 vm_object_clear_flag(object, OBJ_MIGHTBEDIRTY);
1014 vm_object_page_collect_flush(vm_object_t object, vm_page_t p, int pagerflags,
1015 int flags, boolean_t *clearobjflags, boolean_t *eio)
1017 vm_page_t ma[vm_pageout_page_count], p_first, tp;
1018 int count, i, mreq, runlen;
1020 vm_page_lock_assert(p, MA_NOTOWNED);
1021 VM_OBJECT_ASSERT_WLOCKED(object);
1026 for (tp = p; count < vm_pageout_page_count; count++) {
1027 tp = vm_page_next(tp);
1028 if (tp == NULL || vm_page_busied(tp))
1030 if (!vm_object_page_remove_write(tp, flags, clearobjflags))
1034 for (p_first = p; count < vm_pageout_page_count; count++) {
1035 tp = vm_page_prev(p_first);
1036 if (tp == NULL || vm_page_busied(tp))
1038 if (!vm_object_page_remove_write(tp, flags, clearobjflags))
1044 for (tp = p_first, i = 0; i < count; tp = TAILQ_NEXT(tp, listq), i++)
1047 vm_pageout_flush(ma, count, pagerflags, mreq, &runlen, eio);
1052 * Note that there is absolutely no sense in writing out
1053 * anonymous objects, so we track down the vnode object
1055 * We invalidate (remove) all pages from the address space
1056 * for semantic correctness.
1058 * If the backing object is a device object with unmanaged pages, then any
1059 * mappings to the specified range of pages must be removed before this
1060 * function is called.
1062 * Note: certain anonymous maps, such as MAP_NOSYNC maps,
1063 * may start out with a NULL object.
1066 vm_object_sync(vm_object_t object, vm_ooffset_t offset, vm_size_t size,
1067 boolean_t syncio, boolean_t invalidate)
1069 vm_object_t backing_object;
1072 int error, flags, fsync_after;
1079 VM_OBJECT_WLOCK(object);
1080 while ((backing_object = object->backing_object) != NULL) {
1081 VM_OBJECT_WLOCK(backing_object);
1082 offset += object->backing_object_offset;
1083 VM_OBJECT_WUNLOCK(object);
1084 object = backing_object;
1085 if (object->size < OFF_TO_IDX(offset + size))
1086 size = IDX_TO_OFF(object->size) - offset;
1089 * Flush pages if writing is allowed, invalidate them
1090 * if invalidation requested. Pages undergoing I/O
1091 * will be ignored by vm_object_page_remove().
1093 * We cannot lock the vnode and then wait for paging
1094 * to complete without deadlocking against vm_fault.
1095 * Instead we simply call vm_object_page_remove() and
1096 * allow it to block internally on a page-by-page
1097 * basis when it encounters pages undergoing async
1100 if (object->type == OBJT_VNODE &&
1101 (object->flags & OBJ_MIGHTBEDIRTY) != 0 &&
1102 ((vp = object->handle)->v_vflag & VV_NOSYNC) == 0) {
1103 VM_OBJECT_WUNLOCK(object);
1104 (void) vn_start_write(vp, &mp, V_WAIT);
1105 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1106 if (syncio && !invalidate && offset == 0 &&
1107 atop(size) == object->size) {
1109 * If syncing the whole mapping of the file,
1110 * it is faster to schedule all the writes in
1111 * async mode, also allowing the clustering,
1112 * and then wait for i/o to complete.
1117 flags = (syncio || invalidate) ? OBJPC_SYNC : 0;
1118 flags |= invalidate ? (OBJPC_SYNC | OBJPC_INVAL) : 0;
1119 fsync_after = FALSE;
1121 VM_OBJECT_WLOCK(object);
1122 res = vm_object_page_clean(object, offset, offset + size,
1124 VM_OBJECT_WUNLOCK(object);
1126 error = VOP_FSYNC(vp, MNT_WAIT, curthread);
1128 vn_finished_write(mp);
1131 VM_OBJECT_WLOCK(object);
1133 if ((object->type == OBJT_VNODE ||
1134 object->type == OBJT_DEVICE) && invalidate) {
1135 if (object->type == OBJT_DEVICE)
1137 * The option OBJPR_NOTMAPPED must be passed here
1138 * because vm_object_page_remove() cannot remove
1139 * unmanaged mappings.
1141 flags = OBJPR_NOTMAPPED;
1145 flags = OBJPR_CLEANONLY;
1146 vm_object_page_remove(object, OFF_TO_IDX(offset),
1147 OFF_TO_IDX(offset + size + PAGE_MASK), flags);
1149 VM_OBJECT_WUNLOCK(object);
1154 * Determine whether the given advice can be applied to the object. Advice is
1155 * not applied to unmanaged pages since they never belong to page queues, and
1156 * since MADV_FREE is destructive, it can apply only to anonymous pages that
1157 * have been mapped at most once.
1160 vm_object_advice_applies(vm_object_t object, int advice)
1163 if ((object->flags & OBJ_UNMANAGED) != 0)
1165 if (advice != MADV_FREE)
1167 return ((object->type == OBJT_DEFAULT || object->type == OBJT_SWAP) &&
1168 (object->flags & OBJ_ONEMAPPING) != 0);
1172 vm_object_madvise_freespace(vm_object_t object, int advice, vm_pindex_t pindex,
1176 if (advice == MADV_FREE && object->type == OBJT_SWAP)
1177 swap_pager_freespace(object, pindex, size);
1181 * vm_object_madvise:
1183 * Implements the madvise function at the object/page level.
1185 * MADV_WILLNEED (any object)
1187 * Activate the specified pages if they are resident.
1189 * MADV_DONTNEED (any object)
1191 * Deactivate the specified pages if they are resident.
1193 * MADV_FREE (OBJT_DEFAULT/OBJT_SWAP objects,
1194 * OBJ_ONEMAPPING only)
1196 * Deactivate and clean the specified pages if they are
1197 * resident. This permits the process to reuse the pages
1198 * without faulting or the kernel to reclaim the pages
1202 vm_object_madvise(vm_object_t object, vm_pindex_t pindex, vm_pindex_t end,
1205 vm_pindex_t tpindex;
1206 vm_object_t backing_object, tobject;
1213 VM_OBJECT_WLOCK(object);
1214 if (!vm_object_advice_applies(object, advice)) {
1215 VM_OBJECT_WUNLOCK(object);
1218 for (m = vm_page_find_least(object, pindex); pindex < end; pindex++) {
1222 * If the next page isn't resident in the top-level object, we
1223 * need to search the shadow chain. When applying MADV_FREE, we
1224 * take care to release any swap space used to store
1225 * non-resident pages.
1227 if (m == NULL || pindex < m->pindex) {
1229 * Optimize a common case: if the top-level object has
1230 * no backing object, we can skip over the non-resident
1231 * range in constant time.
1233 if (object->backing_object == NULL) {
1234 tpindex = (m != NULL && m->pindex < end) ?
1236 vm_object_madvise_freespace(object, advice,
1237 pindex, tpindex - pindex);
1238 if ((pindex = tpindex) == end)
1245 vm_object_madvise_freespace(tobject, advice,
1248 * Prepare to search the next object in the
1251 backing_object = tobject->backing_object;
1252 if (backing_object == NULL)
1254 VM_OBJECT_WLOCK(backing_object);
1256 OFF_TO_IDX(tobject->backing_object_offset);
1257 if (tobject != object)
1258 VM_OBJECT_WUNLOCK(tobject);
1259 tobject = backing_object;
1260 if (!vm_object_advice_applies(tobject, advice))
1262 } while ((tm = vm_page_lookup(tobject, tpindex)) ==
1267 m = TAILQ_NEXT(m, listq);
1271 * If the page is not in a normal state, skip it.
1273 if (tm->valid != VM_PAGE_BITS_ALL)
1276 if (tm->hold_count != 0 || tm->wire_count != 0) {
1280 KASSERT((tm->flags & PG_FICTITIOUS) == 0,
1281 ("vm_object_madvise: page %p is fictitious", tm));
1282 KASSERT((tm->oflags & VPO_UNMANAGED) == 0,
1283 ("vm_object_madvise: page %p is not managed", tm));
1284 if (vm_page_busied(tm)) {
1285 if (object != tobject)
1286 VM_OBJECT_WUNLOCK(tobject);
1287 VM_OBJECT_WUNLOCK(object);
1288 if (advice == MADV_WILLNEED) {
1290 * Reference the page before unlocking and
1291 * sleeping so that the page daemon is less
1292 * likely to reclaim it.
1294 vm_page_aflag_set(tm, PGA_REFERENCED);
1296 vm_page_busy_sleep(tm, "madvpo", false);
1299 vm_page_advise(tm, advice);
1301 vm_object_madvise_freespace(tobject, advice, tm->pindex, 1);
1303 if (tobject != object)
1304 VM_OBJECT_WUNLOCK(tobject);
1306 VM_OBJECT_WUNLOCK(object);
1312 * Create a new object which is backed by the
1313 * specified existing object range. The source
1314 * object reference is deallocated.
1316 * The new object and offset into that object
1317 * are returned in the source parameters.
1321 vm_object_t *object, /* IN/OUT */
1322 vm_ooffset_t *offset, /* IN/OUT */
1331 * Don't create the new object if the old object isn't shared.
1333 if (source != NULL) {
1334 VM_OBJECT_WLOCK(source);
1335 if (source->ref_count == 1 &&
1336 source->handle == NULL &&
1337 (source->type == OBJT_DEFAULT ||
1338 source->type == OBJT_SWAP)) {
1339 VM_OBJECT_WUNLOCK(source);
1342 VM_OBJECT_WUNLOCK(source);
1346 * Allocate a new object with the given length.
1348 result = vm_object_allocate(OBJT_DEFAULT, atop(length));
1351 * The new object shadows the source object, adding a reference to it.
1352 * Our caller changes his reference to point to the new object,
1353 * removing a reference to the source object. Net result: no change
1354 * of reference count.
1356 * Try to optimize the result object's page color when shadowing
1357 * in order to maintain page coloring consistency in the combined
1360 result->backing_object = source;
1362 * Store the offset into the source object, and fix up the offset into
1365 result->backing_object_offset = *offset;
1366 if (source != NULL) {
1367 VM_OBJECT_WLOCK(source);
1368 result->domain = source->domain;
1369 LIST_INSERT_HEAD(&source->shadow_head, result, shadow_list);
1370 source->shadow_count++;
1371 #if VM_NRESERVLEVEL > 0
1372 result->flags |= source->flags & OBJ_COLORED;
1373 result->pg_color = (source->pg_color + OFF_TO_IDX(*offset)) &
1374 ((1 << (VM_NFREEORDER - 1)) - 1);
1376 VM_OBJECT_WUNLOCK(source);
1381 * Return the new things
1390 * Split the pages in a map entry into a new object. This affords
1391 * easier removal of unused pages, and keeps object inheritance from
1392 * being a negative impact on memory usage.
1395 vm_object_split(vm_map_entry_t entry)
1397 vm_page_t m, m_next;
1398 vm_object_t orig_object, new_object, source;
1399 vm_pindex_t idx, offidxstart;
1402 orig_object = entry->object.vm_object;
1403 if (orig_object->type != OBJT_DEFAULT && orig_object->type != OBJT_SWAP)
1405 if (orig_object->ref_count <= 1)
1407 VM_OBJECT_WUNLOCK(orig_object);
1409 offidxstart = OFF_TO_IDX(entry->offset);
1410 size = atop(entry->end - entry->start);
1413 * If swap_pager_copy() is later called, it will convert new_object
1414 * into a swap object.
1416 new_object = vm_object_allocate(OBJT_DEFAULT, size);
1419 * At this point, the new object is still private, so the order in
1420 * which the original and new objects are locked does not matter.
1422 VM_OBJECT_WLOCK(new_object);
1423 VM_OBJECT_WLOCK(orig_object);
1424 new_object->domain = orig_object->domain;
1425 source = orig_object->backing_object;
1426 if (source != NULL) {
1427 VM_OBJECT_WLOCK(source);
1428 if ((source->flags & OBJ_DEAD) != 0) {
1429 VM_OBJECT_WUNLOCK(source);
1430 VM_OBJECT_WUNLOCK(orig_object);
1431 VM_OBJECT_WUNLOCK(new_object);
1432 vm_object_deallocate(new_object);
1433 VM_OBJECT_WLOCK(orig_object);
1436 LIST_INSERT_HEAD(&source->shadow_head,
1437 new_object, shadow_list);
1438 source->shadow_count++;
1439 vm_object_reference_locked(source); /* for new_object */
1440 vm_object_clear_flag(source, OBJ_ONEMAPPING);
1441 VM_OBJECT_WUNLOCK(source);
1442 new_object->backing_object_offset =
1443 orig_object->backing_object_offset + entry->offset;
1444 new_object->backing_object = source;
1446 if (orig_object->cred != NULL) {
1447 new_object->cred = orig_object->cred;
1448 crhold(orig_object->cred);
1449 new_object->charge = ptoa(size);
1450 KASSERT(orig_object->charge >= ptoa(size),
1451 ("orig_object->charge < 0"));
1452 orig_object->charge -= ptoa(size);
1455 m = vm_page_find_least(orig_object, offidxstart);
1456 for (; m != NULL && (idx = m->pindex - offidxstart) < size;
1458 m_next = TAILQ_NEXT(m, listq);
1461 * We must wait for pending I/O to complete before we can
1464 * We do not have to VM_PROT_NONE the page as mappings should
1465 * not be changed by this operation.
1467 if (vm_page_busied(m)) {
1468 VM_OBJECT_WUNLOCK(new_object);
1470 VM_OBJECT_WUNLOCK(orig_object);
1471 vm_page_busy_sleep(m, "spltwt", false);
1472 VM_OBJECT_WLOCK(orig_object);
1473 VM_OBJECT_WLOCK(new_object);
1477 /* vm_page_rename() will dirty the page. */
1478 if (vm_page_rename(m, new_object, idx)) {
1479 VM_OBJECT_WUNLOCK(new_object);
1480 VM_OBJECT_WUNLOCK(orig_object);
1482 VM_OBJECT_WLOCK(orig_object);
1483 VM_OBJECT_WLOCK(new_object);
1486 #if VM_NRESERVLEVEL > 0
1488 * If some of the reservation's allocated pages remain with
1489 * the original object, then transferring the reservation to
1490 * the new object is neither particularly beneficial nor
1491 * particularly harmful as compared to leaving the reservation
1492 * with the original object. If, however, all of the
1493 * reservation's allocated pages are transferred to the new
1494 * object, then transferring the reservation is typically
1495 * beneficial. Determining which of these two cases applies
1496 * would be more costly than unconditionally renaming the
1499 vm_reserv_rename(m, new_object, orig_object, offidxstart);
1501 if (orig_object->type == OBJT_SWAP)
1504 if (orig_object->type == OBJT_SWAP) {
1506 * swap_pager_copy() can sleep, in which case the orig_object's
1507 * and new_object's locks are released and reacquired.
1509 swap_pager_copy(orig_object, new_object, offidxstart, 0);
1510 TAILQ_FOREACH(m, &new_object->memq, listq)
1513 VM_OBJECT_WUNLOCK(orig_object);
1514 VM_OBJECT_WUNLOCK(new_object);
1515 entry->object.vm_object = new_object;
1516 entry->offset = 0LL;
1517 vm_object_deallocate(orig_object);
1518 VM_OBJECT_WLOCK(new_object);
1521 #define OBSC_COLLAPSE_NOWAIT 0x0002
1522 #define OBSC_COLLAPSE_WAIT 0x0004
1525 vm_object_collapse_scan_wait(vm_object_t object, vm_page_t p, vm_page_t next,
1528 vm_object_t backing_object;
1530 VM_OBJECT_ASSERT_WLOCKED(object);
1531 backing_object = object->backing_object;
1532 VM_OBJECT_ASSERT_WLOCKED(backing_object);
1534 KASSERT(p == NULL || vm_page_busied(p), ("unbusy page %p", p));
1535 KASSERT(p == NULL || p->object == object || p->object == backing_object,
1536 ("invalid ownership %p %p %p", p, object, backing_object));
1537 if ((op & OBSC_COLLAPSE_NOWAIT) != 0)
1541 VM_OBJECT_WUNLOCK(object);
1542 VM_OBJECT_WUNLOCK(backing_object);
1543 /* The page is only NULL when rename fails. */
1547 vm_page_busy_sleep(p, "vmocol", false);
1548 VM_OBJECT_WLOCK(object);
1549 VM_OBJECT_WLOCK(backing_object);
1550 return (TAILQ_FIRST(&backing_object->memq));
1554 vm_object_scan_all_shadowed(vm_object_t object)
1556 vm_object_t backing_object;
1558 vm_pindex_t backing_offset_index, new_pindex, pi, ps;
1560 VM_OBJECT_ASSERT_WLOCKED(object);
1561 VM_OBJECT_ASSERT_WLOCKED(object->backing_object);
1563 backing_object = object->backing_object;
1565 if (backing_object->type != OBJT_DEFAULT &&
1566 backing_object->type != OBJT_SWAP)
1569 pi = backing_offset_index = OFF_TO_IDX(object->backing_object_offset);
1570 p = vm_page_find_least(backing_object, pi);
1571 ps = swap_pager_find_least(backing_object, pi);
1574 * Only check pages inside the parent object's range and
1575 * inside the parent object's mapping of the backing object.
1578 if (p != NULL && p->pindex < pi)
1579 p = TAILQ_NEXT(p, listq);
1581 ps = swap_pager_find_least(backing_object, pi);
1582 if (p == NULL && ps >= backing_object->size)
1587 pi = MIN(p->pindex, ps);
1589 new_pindex = pi - backing_offset_index;
1590 if (new_pindex >= object->size)
1594 * See if the parent has the page or if the parent's object
1595 * pager has the page. If the parent has the page but the page
1596 * is not valid, the parent's object pager must have the page.
1598 * If this fails, the parent does not completely shadow the
1599 * object and we might as well give up now.
1601 pp = vm_page_lookup(object, new_pindex);
1602 if ((pp == NULL || pp->valid == 0) &&
1603 !vm_pager_has_page(object, new_pindex, NULL, NULL))
1610 vm_object_collapse_scan(vm_object_t object, int op)
1612 vm_object_t backing_object;
1613 vm_page_t next, p, pp;
1614 vm_pindex_t backing_offset_index, new_pindex;
1616 VM_OBJECT_ASSERT_WLOCKED(object);
1617 VM_OBJECT_ASSERT_WLOCKED(object->backing_object);
1619 backing_object = object->backing_object;
1620 backing_offset_index = OFF_TO_IDX(object->backing_object_offset);
1623 * Initial conditions
1625 if ((op & OBSC_COLLAPSE_WAIT) != 0)
1626 vm_object_set_flag(backing_object, OBJ_DEAD);
1631 for (p = TAILQ_FIRST(&backing_object->memq); p != NULL; p = next) {
1632 next = TAILQ_NEXT(p, listq);
1633 new_pindex = p->pindex - backing_offset_index;
1636 * Check for busy page
1638 if (vm_page_busied(p)) {
1639 next = vm_object_collapse_scan_wait(object, p, next, op);
1643 KASSERT(p->object == backing_object,
1644 ("vm_object_collapse_scan: object mismatch"));
1646 if (p->pindex < backing_offset_index ||
1647 new_pindex >= object->size) {
1648 if (backing_object->type == OBJT_SWAP)
1649 swap_pager_freespace(backing_object, p->pindex,
1653 * Page is out of the parent object's range, we can
1654 * simply destroy it.
1657 KASSERT(!pmap_page_is_mapped(p),
1658 ("freeing mapped page %p", p));
1659 if (p->wire_count == 0)
1667 pp = vm_page_lookup(object, new_pindex);
1668 if (pp != NULL && vm_page_busied(pp)) {
1670 * The page in the parent is busy and possibly not
1671 * (yet) valid. Until its state is finalized by the
1672 * busy bit owner, we can't tell whether it shadows the
1673 * original page. Therefore, we must either skip it
1674 * and the original (backing_object) page or wait for
1675 * its state to be finalized.
1677 * This is due to a race with vm_fault() where we must
1678 * unbusy the original (backing_obj) page before we can
1679 * (re)lock the parent. Hence we can get here.
1681 next = vm_object_collapse_scan_wait(object, pp, next,
1686 KASSERT(pp == NULL || pp->valid != 0,
1687 ("unbusy invalid page %p", pp));
1689 if (pp != NULL || vm_pager_has_page(object, new_pindex, NULL,
1692 * The page already exists in the parent OR swap exists
1693 * for this location in the parent. Leave the parent's
1694 * page alone. Destroy the original page from the
1697 if (backing_object->type == OBJT_SWAP)
1698 swap_pager_freespace(backing_object, p->pindex,
1701 KASSERT(!pmap_page_is_mapped(p),
1702 ("freeing mapped page %p", p));
1703 if (p->wire_count == 0)
1712 * Page does not exist in parent, rename the page from the
1713 * backing object to the main object.
1715 * If the page was mapped to a process, it can remain mapped
1716 * through the rename. vm_page_rename() will dirty the page.
1718 if (vm_page_rename(p, object, new_pindex)) {
1719 next = vm_object_collapse_scan_wait(object, NULL, next,
1724 /* Use the old pindex to free the right page. */
1725 if (backing_object->type == OBJT_SWAP)
1726 swap_pager_freespace(backing_object,
1727 new_pindex + backing_offset_index, 1);
1729 #if VM_NRESERVLEVEL > 0
1731 * Rename the reservation.
1733 vm_reserv_rename(p, object, backing_object,
1734 backing_offset_index);
1742 * this version of collapse allows the operation to occur earlier and
1743 * when paging_in_progress is true for an object... This is not a complete
1744 * operation, but should plug 99.9% of the rest of the leaks.
1747 vm_object_qcollapse(vm_object_t object)
1749 vm_object_t backing_object = object->backing_object;
1751 VM_OBJECT_ASSERT_WLOCKED(object);
1752 VM_OBJECT_ASSERT_WLOCKED(backing_object);
1754 if (backing_object->ref_count != 1)
1757 vm_object_collapse_scan(object, OBSC_COLLAPSE_NOWAIT);
1761 * vm_object_collapse:
1763 * Collapse an object with the object backing it.
1764 * Pages in the backing object are moved into the
1765 * parent, and the backing object is deallocated.
1768 vm_object_collapse(vm_object_t object)
1770 vm_object_t backing_object, new_backing_object;
1772 VM_OBJECT_ASSERT_WLOCKED(object);
1776 * Verify that the conditions are right for collapse:
1778 * The object exists and the backing object exists.
1780 if ((backing_object = object->backing_object) == NULL)
1784 * we check the backing object first, because it is most likely
1787 VM_OBJECT_WLOCK(backing_object);
1788 if (backing_object->handle != NULL ||
1789 (backing_object->type != OBJT_DEFAULT &&
1790 backing_object->type != OBJT_SWAP) ||
1791 (backing_object->flags & OBJ_DEAD) ||
1792 object->handle != NULL ||
1793 (object->type != OBJT_DEFAULT &&
1794 object->type != OBJT_SWAP) ||
1795 (object->flags & OBJ_DEAD)) {
1796 VM_OBJECT_WUNLOCK(backing_object);
1800 if (object->paging_in_progress != 0 ||
1801 backing_object->paging_in_progress != 0) {
1802 vm_object_qcollapse(object);
1803 VM_OBJECT_WUNLOCK(backing_object);
1808 * We know that we can either collapse the backing object (if
1809 * the parent is the only reference to it) or (perhaps) have
1810 * the parent bypass the object if the parent happens to shadow
1811 * all the resident pages in the entire backing object.
1813 * This is ignoring pager-backed pages such as swap pages.
1814 * vm_object_collapse_scan fails the shadowing test in this
1817 if (backing_object->ref_count == 1) {
1818 vm_object_pip_add(object, 1);
1819 vm_object_pip_add(backing_object, 1);
1822 * If there is exactly one reference to the backing
1823 * object, we can collapse it into the parent.
1825 vm_object_collapse_scan(object, OBSC_COLLAPSE_WAIT);
1827 #if VM_NRESERVLEVEL > 0
1829 * Break any reservations from backing_object.
1831 if (__predict_false(!LIST_EMPTY(&backing_object->rvq)))
1832 vm_reserv_break_all(backing_object);
1836 * Move the pager from backing_object to object.
1838 if (backing_object->type == OBJT_SWAP) {
1840 * swap_pager_copy() can sleep, in which case
1841 * the backing_object's and object's locks are
1842 * released and reacquired.
1843 * Since swap_pager_copy() is being asked to
1844 * destroy the source, it will change the
1845 * backing_object's type to OBJT_DEFAULT.
1850 OFF_TO_IDX(object->backing_object_offset), TRUE);
1853 * Object now shadows whatever backing_object did.
1854 * Note that the reference to
1855 * backing_object->backing_object moves from within
1856 * backing_object to within object.
1858 LIST_REMOVE(object, shadow_list);
1859 backing_object->shadow_count--;
1860 if (backing_object->backing_object) {
1861 VM_OBJECT_WLOCK(backing_object->backing_object);
1862 LIST_REMOVE(backing_object, shadow_list);
1864 &backing_object->backing_object->shadow_head,
1865 object, shadow_list);
1867 * The shadow_count has not changed.
1869 VM_OBJECT_WUNLOCK(backing_object->backing_object);
1871 object->backing_object = backing_object->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 backing_object->ref_count = 0;
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 LIST_REMOVE(object, shadow_list);
1910 backing_object->shadow_count--;
1912 new_backing_object = backing_object->backing_object;
1913 if ((object->backing_object = new_backing_object) != NULL) {
1914 VM_OBJECT_WLOCK(new_backing_object);
1916 &new_backing_object->shadow_head,
1920 new_backing_object->shadow_count++;
1921 vm_object_reference_locked(new_backing_object);
1922 VM_OBJECT_WUNLOCK(new_backing_object);
1923 object->backing_object_offset +=
1924 backing_object->backing_object_offset;
1928 * Drop the reference count on backing_object. Since
1929 * its ref_count was at least 2, it will not vanish.
1931 backing_object->ref_count--;
1932 VM_OBJECT_WUNLOCK(backing_object);
1933 counter_u64_add(object_bypasses, 1);
1937 * Try again with this object's new backing object.
1943 * vm_object_page_remove:
1945 * For the given object, either frees or invalidates each of the
1946 * specified pages. In general, a page is freed. However, if a page is
1947 * wired for any reason other than the existence of a managed, wired
1948 * mapping, then it may be invalidated but not removed from the object.
1949 * Pages are specified by the given range ["start", "end") and the option
1950 * OBJPR_CLEANONLY. As a special case, if "end" is zero, then the range
1951 * extends from "start" to the end of the object. If the option
1952 * OBJPR_CLEANONLY is specified, then only the non-dirty pages within the
1953 * specified range are affected. If the option OBJPR_NOTMAPPED is
1954 * specified, then the pages within the specified range must have no
1955 * mappings. Otherwise, if this option is not specified, any mappings to
1956 * the specified pages are removed before the pages are freed or
1959 * In general, this operation should only be performed on objects that
1960 * contain managed pages. There are, however, two exceptions. First, it
1961 * is performed on the kernel and kmem objects by vm_map_entry_delete().
1962 * Second, it is used by msync(..., MS_INVALIDATE) to invalidate device-
1963 * backed pages. In both of these cases, the option OBJPR_CLEANONLY must
1964 * not be specified and the option OBJPR_NOTMAPPED must be specified.
1966 * The object must be locked.
1969 vm_object_page_remove(vm_object_t object, vm_pindex_t start, vm_pindex_t end,
1976 VM_OBJECT_ASSERT_WLOCKED(object);
1977 KASSERT((object->flags & OBJ_UNMANAGED) == 0 ||
1978 (options & (OBJPR_CLEANONLY | OBJPR_NOTMAPPED)) == OBJPR_NOTMAPPED,
1979 ("vm_object_page_remove: illegal options for object %p", object));
1980 if (object->resident_page_count == 0)
1982 vm_object_pip_add(object, 1);
1985 p = vm_page_find_least(object, start);
1989 * Here, the variable "p" is either (1) the page with the least pindex
1990 * greater than or equal to the parameter "start" or (2) NULL.
1992 for (; p != NULL && (p->pindex < end || end == 0); p = next) {
1993 next = TAILQ_NEXT(p, listq);
1996 * If the page is wired for any reason besides the existence
1997 * of managed, wired mappings, then it cannot be freed. For
1998 * example, fictitious pages, which represent device memory,
1999 * are inherently wired and cannot be freed. They can,
2000 * however, be invalidated if the option OBJPR_CLEANONLY is
2003 vm_page_change_lock(p, &mtx);
2004 if (vm_page_xbusied(p)) {
2005 VM_OBJECT_WUNLOCK(object);
2006 vm_page_busy_sleep(p, "vmopax", true);
2007 VM_OBJECT_WLOCK(object);
2010 if (p->wire_count != 0) {
2011 if ((options & OBJPR_NOTMAPPED) == 0 &&
2012 object->ref_count != 0)
2014 if ((options & OBJPR_CLEANONLY) == 0) {
2020 if (vm_page_busied(p)) {
2021 VM_OBJECT_WUNLOCK(object);
2022 vm_page_busy_sleep(p, "vmopar", false);
2023 VM_OBJECT_WLOCK(object);
2026 KASSERT((p->flags & PG_FICTITIOUS) == 0,
2027 ("vm_object_page_remove: page %p is fictitious", p));
2028 if ((options & OBJPR_CLEANONLY) != 0 && p->valid != 0) {
2029 if ((options & OBJPR_NOTMAPPED) == 0 &&
2030 object->ref_count != 0)
2031 pmap_remove_write(p);
2035 if ((options & OBJPR_NOTMAPPED) == 0 && object->ref_count != 0)
2037 p->flags &= ~PG_ZERO;
2038 if (vm_page_free_prep(p, false))
2039 TAILQ_INSERT_TAIL(&pgl, p, listq);
2043 vm_page_free_phys_pglist(&pgl);
2044 vm_object_pip_wakeup(object);
2048 * vm_object_page_noreuse:
2050 * For the given object, attempt to move the specified pages to
2051 * the head of the inactive queue. This bypasses regular LRU
2052 * operation and allows the pages to be reused quickly under memory
2053 * pressure. If a page is wired for any reason, then it will not
2054 * be queued. Pages are specified by the range ["start", "end").
2055 * As a special case, if "end" is zero, then the range extends from
2056 * "start" to the end of the object.
2058 * This operation should only be performed on objects that
2059 * contain non-fictitious, managed pages.
2061 * The object must be locked.
2064 vm_object_page_noreuse(vm_object_t object, vm_pindex_t start, vm_pindex_t end)
2069 VM_OBJECT_ASSERT_LOCKED(object);
2070 KASSERT((object->flags & (OBJ_FICTITIOUS | OBJ_UNMANAGED)) == 0,
2071 ("vm_object_page_noreuse: illegal object %p", object));
2072 if (object->resident_page_count == 0)
2074 p = vm_page_find_least(object, start);
2077 * Here, the variable "p" is either (1) the page with the least pindex
2078 * greater than or equal to the parameter "start" or (2) NULL.
2081 for (; p != NULL && (p->pindex < end || end == 0); p = next) {
2082 next = TAILQ_NEXT(p, listq);
2083 vm_page_change_lock(p, &mtx);
2084 vm_page_deactivate_noreuse(p);
2091 * Populate the specified range of the object with valid pages. Returns
2092 * TRUE if the range is successfully populated and FALSE otherwise.
2094 * Note: This function should be optimized to pass a larger array of
2095 * pages to vm_pager_get_pages() before it is applied to a non-
2096 * OBJT_DEVICE object.
2098 * The object must be locked.
2101 vm_object_populate(vm_object_t object, vm_pindex_t start, vm_pindex_t end)
2107 VM_OBJECT_ASSERT_WLOCKED(object);
2108 for (pindex = start; pindex < end; pindex++) {
2109 m = vm_page_grab(object, pindex, VM_ALLOC_NORMAL);
2110 if (m->valid != VM_PAGE_BITS_ALL) {
2111 rv = vm_pager_get_pages(object, &m, 1, NULL, NULL);
2112 if (rv != VM_PAGER_OK) {
2120 * Keep "m" busy because a subsequent iteration may unlock
2124 if (pindex > start) {
2125 m = vm_page_lookup(object, start);
2126 while (m != NULL && m->pindex < pindex) {
2128 m = TAILQ_NEXT(m, listq);
2131 return (pindex == end);
2135 * Routine: vm_object_coalesce
2136 * Function: Coalesces two objects backing up adjoining
2137 * regions of memory into a single object.
2139 * returns TRUE if objects were combined.
2141 * NOTE: Only works at the moment if the second object is NULL -
2142 * if it's not, which object do we lock first?
2145 * prev_object First object to coalesce
2146 * prev_offset Offset into prev_object
2147 * prev_size Size of reference to prev_object
2148 * next_size Size of reference to the second object
2149 * reserved Indicator that extension region has
2150 * swap accounted for
2153 * The object must *not* be locked.
2156 vm_object_coalesce(vm_object_t prev_object, vm_ooffset_t prev_offset,
2157 vm_size_t prev_size, vm_size_t next_size, boolean_t reserved)
2159 vm_pindex_t next_pindex;
2161 if (prev_object == NULL)
2163 VM_OBJECT_WLOCK(prev_object);
2164 if ((prev_object->type != OBJT_DEFAULT &&
2165 prev_object->type != OBJT_SWAP) ||
2166 (prev_object->flags & OBJ_TMPFS_NODE) != 0) {
2167 VM_OBJECT_WUNLOCK(prev_object);
2172 * Try to collapse the object first
2174 vm_object_collapse(prev_object);
2177 * Can't coalesce if: . more than one reference . paged out . shadows
2178 * another object . has a copy elsewhere (any of which mean that the
2179 * pages not mapped to prev_entry may be in use anyway)
2181 if (prev_object->backing_object != NULL) {
2182 VM_OBJECT_WUNLOCK(prev_object);
2186 prev_size >>= PAGE_SHIFT;
2187 next_size >>= PAGE_SHIFT;
2188 next_pindex = OFF_TO_IDX(prev_offset) + prev_size;
2190 if ((prev_object->ref_count > 1) &&
2191 (prev_object->size != next_pindex)) {
2192 VM_OBJECT_WUNLOCK(prev_object);
2197 * Account for the charge.
2199 if (prev_object->cred != NULL) {
2202 * If prev_object was charged, then this mapping,
2203 * although not charged now, may become writable
2204 * later. Non-NULL cred in the object would prevent
2205 * swap reservation during enabling of the write
2206 * access, so reserve swap now. Failed reservation
2207 * cause allocation of the separate object for the map
2208 * entry, and swap reservation for this entry is
2209 * managed in appropriate time.
2211 if (!reserved && !swap_reserve_by_cred(ptoa(next_size),
2212 prev_object->cred)) {
2213 VM_OBJECT_WUNLOCK(prev_object);
2216 prev_object->charge += ptoa(next_size);
2220 * Remove any pages that may still be in the object from a previous
2223 if (next_pindex < prev_object->size) {
2224 vm_object_page_remove(prev_object, next_pindex, next_pindex +
2226 if (prev_object->type == OBJT_SWAP)
2227 swap_pager_freespace(prev_object,
2228 next_pindex, next_size);
2230 if (prev_object->cred != NULL) {
2231 KASSERT(prev_object->charge >=
2232 ptoa(prev_object->size - next_pindex),
2233 ("object %p overcharged 1 %jx %jx", prev_object,
2234 (uintmax_t)next_pindex, (uintmax_t)next_size));
2235 prev_object->charge -= ptoa(prev_object->size -
2242 * Extend the object if necessary.
2244 if (next_pindex + next_size > prev_object->size)
2245 prev_object->size = next_pindex + next_size;
2247 VM_OBJECT_WUNLOCK(prev_object);
2252 vm_object_set_writeable_dirty(vm_object_t object)
2255 VM_OBJECT_ASSERT_WLOCKED(object);
2256 if (object->type != OBJT_VNODE) {
2257 if ((object->flags & OBJ_TMPFS_NODE) != 0) {
2258 KASSERT(object->type == OBJT_SWAP, ("non-swap tmpfs"));
2259 vm_object_set_flag(object, OBJ_TMPFS_DIRTY);
2263 object->generation++;
2264 if ((object->flags & OBJ_MIGHTBEDIRTY) != 0)
2266 vm_object_set_flag(object, OBJ_MIGHTBEDIRTY);
2272 * For each page offset within the specified range of the given object,
2273 * find the highest-level page in the shadow chain and unwire it. A page
2274 * must exist at every page offset, and the highest-level page must be
2278 vm_object_unwire(vm_object_t object, vm_ooffset_t offset, vm_size_t length,
2281 vm_object_t tobject;
2283 vm_pindex_t end_pindex, pindex, tpindex;
2284 int depth, locked_depth;
2286 KASSERT((offset & PAGE_MASK) == 0,
2287 ("vm_object_unwire: offset is not page aligned"));
2288 KASSERT((length & PAGE_MASK) == 0,
2289 ("vm_object_unwire: length is not a multiple of PAGE_SIZE"));
2290 /* The wired count of a fictitious page never changes. */
2291 if ((object->flags & OBJ_FICTITIOUS) != 0)
2293 pindex = OFF_TO_IDX(offset);
2294 end_pindex = pindex + atop(length);
2296 VM_OBJECT_RLOCK(object);
2297 m = vm_page_find_least(object, pindex);
2298 while (pindex < end_pindex) {
2299 if (m == NULL || pindex < m->pindex) {
2301 * The first object in the shadow chain doesn't
2302 * contain a page at the current index. Therefore,
2303 * the page must exist in a backing object.
2310 OFF_TO_IDX(tobject->backing_object_offset);
2311 tobject = tobject->backing_object;
2312 KASSERT(tobject != NULL,
2313 ("vm_object_unwire: missing page"));
2314 if ((tobject->flags & OBJ_FICTITIOUS) != 0)
2317 if (depth == locked_depth) {
2319 VM_OBJECT_RLOCK(tobject);
2321 } while ((tm = vm_page_lookup(tobject, tpindex)) ==
2325 m = TAILQ_NEXT(m, listq);
2328 vm_page_unwire(tm, queue);
2333 /* Release the accumulated object locks. */
2334 for (depth = 0; depth < locked_depth; depth++) {
2335 tobject = object->backing_object;
2336 VM_OBJECT_RUNLOCK(object);
2342 vm_object_vnode(vm_object_t object)
2345 VM_OBJECT_ASSERT_LOCKED(object);
2346 if (object->type == OBJT_VNODE)
2347 return (object->handle);
2348 if (object->type == OBJT_SWAP && (object->flags & OBJ_TMPFS) != 0)
2349 return (object->un_pager.swp.swp_tmpfs);
2354 sysctl_vm_object_list(SYSCTL_HANDLER_ARGS)
2356 struct kinfo_vmobject *kvo;
2357 char *fullpath, *freepath;
2364 if (req->oldptr == NULL) {
2366 * If an old buffer has not been provided, generate an
2367 * estimate of the space needed for a subsequent call.
2369 mtx_lock(&vm_object_list_mtx);
2371 TAILQ_FOREACH(obj, &vm_object_list, object_list) {
2372 if (obj->type == OBJT_DEAD)
2376 mtx_unlock(&vm_object_list_mtx);
2377 return (SYSCTL_OUT(req, NULL, sizeof(struct kinfo_vmobject) *
2381 kvo = malloc(sizeof(*kvo), M_TEMP, M_WAITOK);
2385 * VM objects are type stable and are never removed from the
2386 * list once added. This allows us to safely read obj->object_list
2387 * after reacquiring the VM object lock.
2389 mtx_lock(&vm_object_list_mtx);
2390 TAILQ_FOREACH(obj, &vm_object_list, object_list) {
2391 if (obj->type == OBJT_DEAD)
2393 VM_OBJECT_RLOCK(obj);
2394 if (obj->type == OBJT_DEAD) {
2395 VM_OBJECT_RUNLOCK(obj);
2398 mtx_unlock(&vm_object_list_mtx);
2399 kvo->kvo_size = ptoa(obj->size);
2400 kvo->kvo_resident = obj->resident_page_count;
2401 kvo->kvo_ref_count = obj->ref_count;
2402 kvo->kvo_shadow_count = obj->shadow_count;
2403 kvo->kvo_memattr = obj->memattr;
2404 kvo->kvo_active = 0;
2405 kvo->kvo_inactive = 0;
2406 TAILQ_FOREACH(m, &obj->memq, listq) {
2408 * A page may belong to the object but be
2409 * dequeued and set to PQ_NONE while the
2410 * object lock is not held. This makes the
2411 * reads of m->queue below racy, and we do not
2412 * count pages set to PQ_NONE. However, this
2413 * sysctl is only meant to give an
2414 * approximation of the system anyway.
2416 if (vm_page_active(m))
2418 else if (vm_page_inactive(m))
2419 kvo->kvo_inactive++;
2422 kvo->kvo_vn_fileid = 0;
2423 kvo->kvo_vn_fsid = 0;
2424 kvo->kvo_vn_fsid_freebsd11 = 0;
2428 switch (obj->type) {
2430 kvo->kvo_type = KVME_TYPE_DEFAULT;
2433 kvo->kvo_type = KVME_TYPE_VNODE;
2438 kvo->kvo_type = KVME_TYPE_SWAP;
2441 kvo->kvo_type = KVME_TYPE_DEVICE;
2444 kvo->kvo_type = KVME_TYPE_PHYS;
2447 kvo->kvo_type = KVME_TYPE_DEAD;
2450 kvo->kvo_type = KVME_TYPE_SG;
2452 case OBJT_MGTDEVICE:
2453 kvo->kvo_type = KVME_TYPE_MGTDEVICE;
2456 kvo->kvo_type = KVME_TYPE_UNKNOWN;
2459 VM_OBJECT_RUNLOCK(obj);
2461 vn_fullpath(curthread, vp, &fullpath, &freepath);
2462 vn_lock(vp, LK_SHARED | LK_RETRY);
2463 if (VOP_GETATTR(vp, &va, curthread->td_ucred) == 0) {
2464 kvo->kvo_vn_fileid = va.va_fileid;
2465 kvo->kvo_vn_fsid = va.va_fsid;
2466 kvo->kvo_vn_fsid_freebsd11 = va.va_fsid;
2472 strlcpy(kvo->kvo_path, fullpath, sizeof(kvo->kvo_path));
2473 if (freepath != NULL)
2474 free(freepath, M_TEMP);
2476 /* Pack record size down */
2477 kvo->kvo_structsize = offsetof(struct kinfo_vmobject, kvo_path)
2478 + strlen(kvo->kvo_path) + 1;
2479 kvo->kvo_structsize = roundup(kvo->kvo_structsize,
2481 error = SYSCTL_OUT(req, kvo, kvo->kvo_structsize);
2482 mtx_lock(&vm_object_list_mtx);
2486 mtx_unlock(&vm_object_list_mtx);
2490 SYSCTL_PROC(_vm, OID_AUTO, objects, CTLTYPE_STRUCT | CTLFLAG_RW | CTLFLAG_SKIP |
2491 CTLFLAG_MPSAFE, NULL, 0, sysctl_vm_object_list, "S,kinfo_vmobject",
2492 "List of VM objects");
2494 #include "opt_ddb.h"
2496 #include <sys/kernel.h>
2498 #include <sys/cons.h>
2500 #include <ddb/ddb.h>
2503 _vm_object_in_map(vm_map_t map, vm_object_t object, vm_map_entry_t entry)
2506 vm_map_entry_t tmpe;
2514 tmpe = map->header.next;
2515 entcount = map->nentries;
2516 while (entcount-- && (tmpe != &map->header)) {
2517 if (_vm_object_in_map(map, object, tmpe)) {
2522 } else if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
2523 tmpm = entry->object.sub_map;
2524 tmpe = tmpm->header.next;
2525 entcount = tmpm->nentries;
2526 while (entcount-- && tmpe != &tmpm->header) {
2527 if (_vm_object_in_map(tmpm, object, tmpe)) {
2532 } else if ((obj = entry->object.vm_object) != NULL) {
2533 for (; obj; obj = obj->backing_object)
2534 if (obj == object) {
2542 vm_object_in_map(vm_object_t object)
2546 /* sx_slock(&allproc_lock); */
2547 FOREACH_PROC_IN_SYSTEM(p) {
2548 if (!p->p_vmspace /* || (p->p_flag & (P_SYSTEM|P_WEXIT)) */)
2550 if (_vm_object_in_map(&p->p_vmspace->vm_map, object, 0)) {
2551 /* sx_sunlock(&allproc_lock); */
2555 /* sx_sunlock(&allproc_lock); */
2556 if (_vm_object_in_map(kernel_map, object, 0))
2561 DB_SHOW_COMMAND(vmochk, vm_object_check)
2566 * make sure that internal objs are in a map somewhere
2567 * and none have zero ref counts.
2569 TAILQ_FOREACH(object, &vm_object_list, object_list) {
2570 if (object->handle == NULL &&
2571 (object->type == OBJT_DEFAULT || object->type == OBJT_SWAP)) {
2572 if (object->ref_count == 0) {
2573 db_printf("vmochk: internal obj has zero ref count: %ld\n",
2574 (long)object->size);
2576 if (!vm_object_in_map(object)) {
2578 "vmochk: internal obj is not in a map: "
2579 "ref: %d, size: %lu: 0x%lx, backing_object: %p\n",
2580 object->ref_count, (u_long)object->size,
2581 (u_long)object->size,
2582 (void *)object->backing_object);
2589 * vm_object_print: [ debug ]
2591 DB_SHOW_COMMAND(object, vm_object_print_static)
2593 /* XXX convert args. */
2594 vm_object_t object = (vm_object_t)addr;
2595 boolean_t full = have_addr;
2599 /* XXX count is an (unused) arg. Avoid shadowing it. */
2600 #define count was_count
2608 "Object %p: type=%d, size=0x%jx, res=%d, ref=%d, flags=0x%x ruid %d charge %jx\n",
2609 object, (int)object->type, (uintmax_t)object->size,
2610 object->resident_page_count, object->ref_count, object->flags,
2611 object->cred ? object->cred->cr_ruid : -1, (uintmax_t)object->charge);
2612 db_iprintf(" sref=%d, backing_object(%d)=(%p)+0x%jx\n",
2613 object->shadow_count,
2614 object->backing_object ? object->backing_object->ref_count : 0,
2615 object->backing_object, (uintmax_t)object->backing_object_offset);
2622 TAILQ_FOREACH(p, &object->memq, listq) {
2624 db_iprintf("memory:=");
2625 else if (count == 6) {
2633 db_printf("(off=0x%jx,page=0x%jx)",
2634 (uintmax_t)p->pindex, (uintmax_t)VM_PAGE_TO_PHYS(p));
2644 /* XXX need this non-static entry for calling from vm_map_print. */
2647 /* db_expr_t */ long addr,
2648 boolean_t have_addr,
2649 /* db_expr_t */ long count,
2652 vm_object_print_static(addr, have_addr, count, modif);
2655 DB_SHOW_COMMAND(vmopag, vm_object_print_pages)
2660 vm_page_t m, prev_m;
2664 TAILQ_FOREACH(object, &vm_object_list, object_list) {
2665 db_printf("new object: %p\n", (void *)object);
2676 TAILQ_FOREACH(m, &object->memq, listq) {
2677 if (m->pindex > 128)
2679 if ((prev_m = TAILQ_PREV(m, pglist, listq)) != NULL &&
2680 prev_m->pindex + 1 != m->pindex) {
2682 db_printf(" index(%ld)run(%d)pa(0x%lx)\n",
2683 (long)fidx, rcount, (long)pa);
2695 (VM_PAGE_TO_PHYS(m) == pa + rcount * PAGE_SIZE)) {
2700 db_printf(" index(%ld)run(%d)pa(0x%lx)\n",
2701 (long)fidx, rcount, (long)pa);
2711 pa = VM_PAGE_TO_PHYS(m);
2715 db_printf(" index(%ld)run(%d)pa(0x%lx)\n",
2716 (long)fidx, rcount, (long)pa);