2 * Copyright (c) 1991, 1993
3 * The Regents of the University of California. All rights reserved.
5 * This code is derived from software contributed to Berkeley by
6 * The Mach Operating System project at Carnegie-Mellon University.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 4. Neither the name of the University nor the names of its contributors
17 * may be used to endorse or promote products derived from this software
18 * without specific prior written permission.
20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * from: @(#)vm_object.c 8.5 (Berkeley) 3/22/94
35 * Copyright (c) 1987, 1990 Carnegie-Mellon University.
36 * All rights reserved.
38 * Authors: Avadis Tevanian, Jr., Michael Wayne Young
40 * Permission to use, copy, modify and distribute this software and
41 * its documentation is hereby granted, provided that both the copyright
42 * notice and this permission notice appear in all copies of the
43 * software, derivative works or modified versions, and any portions
44 * thereof, and that both notices appear in supporting documentation.
46 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
47 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
48 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
50 * Carnegie Mellon requests users of this software to return to
52 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
53 * School of Computer Science
54 * Carnegie Mellon University
55 * Pittsburgh PA 15213-3890
57 * any improvements or extensions that they make and grant Carnegie the
58 * rights to redistribute these changes.
62 * Virtual memory object module.
65 #include <sys/cdefs.h>
66 __FBSDID("$FreeBSD$");
70 #include <sys/param.h>
71 #include <sys/systm.h>
74 #include <sys/mount.h>
75 #include <sys/kernel.h>
76 #include <sys/pctrie.h>
77 #include <sys/sysctl.h>
78 #include <sys/mutex.h>
79 #include <sys/proc.h> /* for curproc, pageproc */
80 #include <sys/socket.h>
81 #include <sys/resourcevar.h>
82 #include <sys/rwlock.h>
84 #include <sys/vnode.h>
85 #include <sys/vmmeter.h>
89 #include <vm/vm_param.h>
91 #include <vm/vm_map.h>
92 #include <vm/vm_object.h>
93 #include <vm/vm_page.h>
94 #include <vm/vm_pageout.h>
95 #include <vm/vm_pager.h>
96 #include <vm/swap_pager.h>
97 #include <vm/vm_kern.h>
98 #include <vm/vm_extern.h>
99 #include <vm/vm_radix.h>
100 #include <vm/vm_reserv.h>
103 static int old_msync;
104 SYSCTL_INT(_vm, OID_AUTO, old_msync, CTLFLAG_RW, &old_msync, 0,
105 "Use old (insecure) msync behavior");
107 static int vm_object_page_collect_flush(vm_object_t object, vm_page_t p,
108 int pagerflags, int flags, boolean_t *clearobjflags,
110 static boolean_t vm_object_page_remove_write(vm_page_t p, int flags,
111 boolean_t *clearobjflags);
112 static void vm_object_qcollapse(vm_object_t object);
113 static void vm_object_vndeallocate(vm_object_t object);
116 * Virtual memory objects maintain the actual data
117 * associated with allocated virtual memory. A given
118 * page of memory exists within exactly one object.
120 * An object is only deallocated when all "references"
121 * are given up. Only one "reference" to a given
122 * region of an object should be writeable.
124 * Associated with each object is a list of all resident
125 * memory pages belonging to that object; this list is
126 * maintained by the "vm_page" module, and locked by the object's
129 * Each object also records a "pager" routine which is
130 * used to retrieve (and store) pages to the proper backing
131 * storage. In addition, objects may be backed by other
132 * objects from which they were virtual-copied.
134 * The only items within the object structure which are
135 * modified after time of creation are:
136 * reference count locked by object's lock
137 * pager routine locked by object's lock
141 struct object_q vm_object_list;
142 struct mtx vm_object_list_mtx; /* lock for object list and count */
144 struct vm_object kernel_object_store;
145 struct vm_object kmem_object_store;
147 static SYSCTL_NODE(_vm_stats, OID_AUTO, object, CTLFLAG_RD, 0,
150 static long object_collapses;
151 SYSCTL_LONG(_vm_stats_object, OID_AUTO, collapses, CTLFLAG_RD,
152 &object_collapses, 0, "VM object collapses");
154 static long object_bypasses;
155 SYSCTL_LONG(_vm_stats_object, OID_AUTO, bypasses, CTLFLAG_RD,
156 &object_bypasses, 0, "VM object bypasses");
158 static uma_zone_t obj_zone;
160 static int vm_object_zinit(void *mem, int size, int flags);
163 static void vm_object_zdtor(void *mem, int size, void *arg);
166 vm_object_zdtor(void *mem, int size, void *arg)
170 object = (vm_object_t)mem;
171 KASSERT(object->ref_count == 0,
172 ("object %p ref_count = %d", object, object->ref_count));
173 KASSERT(TAILQ_EMPTY(&object->memq),
174 ("object %p has resident pages in its memq", object));
175 KASSERT(vm_radix_is_empty(&object->rtree),
176 ("object %p has resident pages in its trie", object));
177 #if VM_NRESERVLEVEL > 0
178 KASSERT(LIST_EMPTY(&object->rvq),
179 ("object %p has reservations",
182 KASSERT(object->paging_in_progress == 0,
183 ("object %p paging_in_progress = %d",
184 object, object->paging_in_progress));
185 KASSERT(object->resident_page_count == 0,
186 ("object %p resident_page_count = %d",
187 object, object->resident_page_count));
188 KASSERT(object->shadow_count == 0,
189 ("object %p shadow_count = %d",
190 object, object->shadow_count));
191 KASSERT(object->type == OBJT_DEAD,
192 ("object %p has non-dead type %d",
193 object, object->type));
198 vm_object_zinit(void *mem, int size, int flags)
202 object = (vm_object_t)mem;
203 rw_init_flags(&object->lock, "vm object", RW_DUPOK | RW_NEW);
205 /* These are true for any object that has been freed */
206 object->type = OBJT_DEAD;
207 object->ref_count = 0;
208 vm_radix_init(&object->rtree);
209 object->paging_in_progress = 0;
210 object->resident_page_count = 0;
211 object->shadow_count = 0;
212 object->flags = OBJ_DEAD;
214 mtx_lock(&vm_object_list_mtx);
215 TAILQ_INSERT_TAIL(&vm_object_list, object, object_list);
216 mtx_unlock(&vm_object_list_mtx);
221 _vm_object_allocate(objtype_t type, vm_pindex_t size, vm_object_t object)
224 TAILQ_INIT(&object->memq);
225 LIST_INIT(&object->shadow_head);
228 if (type == OBJT_SWAP)
229 pctrie_init(&object->un_pager.swp.swp_blks);
232 * Ensure that swap_pager_swapoff() iteration over object_list
233 * sees up to date type and pctrie head if it observed
236 atomic_thread_fence_rel();
240 panic("_vm_object_allocate: can't create OBJT_DEAD");
243 object->flags = OBJ_ONEMAPPING;
247 object->flags = OBJ_FICTITIOUS | OBJ_UNMANAGED;
250 object->flags = OBJ_FICTITIOUS;
253 object->flags = OBJ_UNMANAGED;
259 panic("_vm_object_allocate: type %d is undefined", type);
262 object->generation = 1;
263 object->ref_count = 1;
264 object->memattr = VM_MEMATTR_DEFAULT;
267 object->handle = NULL;
268 object->backing_object = NULL;
269 object->backing_object_offset = (vm_ooffset_t) 0;
270 #if VM_NRESERVLEVEL > 0
271 LIST_INIT(&object->rvq);
273 umtx_shm_object_init(object);
279 * Initialize the VM objects module.
284 TAILQ_INIT(&vm_object_list);
285 mtx_init(&vm_object_list_mtx, "vm object_list", NULL, MTX_DEF);
287 rw_init(&kernel_object->lock, "kernel vm object");
288 _vm_object_allocate(OBJT_PHYS, atop(VM_MAX_KERNEL_ADDRESS -
289 VM_MIN_KERNEL_ADDRESS), kernel_object);
290 #if VM_NRESERVLEVEL > 0
291 kernel_object->flags |= OBJ_COLORED;
292 kernel_object->pg_color = (u_short)atop(VM_MIN_KERNEL_ADDRESS);
295 rw_init(&kmem_object->lock, "kmem vm object");
296 _vm_object_allocate(OBJT_PHYS, atop(VM_MAX_KERNEL_ADDRESS -
297 VM_MIN_KERNEL_ADDRESS), kmem_object);
298 #if VM_NRESERVLEVEL > 0
299 kmem_object->flags |= OBJ_COLORED;
300 kmem_object->pg_color = (u_short)atop(VM_MIN_KERNEL_ADDRESS);
304 * The lock portion of struct vm_object must be type stable due
305 * to vm_pageout_fallback_object_lock locking a vm object
306 * without holding any references to it.
308 obj_zone = uma_zcreate("VM OBJECT", sizeof (struct vm_object), NULL,
314 vm_object_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
320 vm_object_clear_flag(vm_object_t object, u_short bits)
323 VM_OBJECT_ASSERT_WLOCKED(object);
324 object->flags &= ~bits;
328 * Sets the default memory attribute for the specified object. Pages
329 * that are allocated to this object are by default assigned this memory
332 * Presently, this function must be called before any pages are allocated
333 * to the object. In the future, this requirement may be relaxed for
334 * "default" and "swap" objects.
337 vm_object_set_memattr(vm_object_t object, vm_memattr_t memattr)
340 VM_OBJECT_ASSERT_WLOCKED(object);
341 switch (object->type) {
349 if (!TAILQ_EMPTY(&object->memq))
350 return (KERN_FAILURE);
353 return (KERN_INVALID_ARGUMENT);
355 panic("vm_object_set_memattr: object %p is of undefined type",
358 object->memattr = memattr;
359 return (KERN_SUCCESS);
363 vm_object_pip_add(vm_object_t object, short i)
366 VM_OBJECT_ASSERT_WLOCKED(object);
367 object->paging_in_progress += i;
371 vm_object_pip_subtract(vm_object_t object, short i)
374 VM_OBJECT_ASSERT_WLOCKED(object);
375 object->paging_in_progress -= i;
379 vm_object_pip_wakeup(vm_object_t object)
382 VM_OBJECT_ASSERT_WLOCKED(object);
383 object->paging_in_progress--;
384 if ((object->flags & OBJ_PIPWNT) && object->paging_in_progress == 0) {
385 vm_object_clear_flag(object, OBJ_PIPWNT);
391 vm_object_pip_wakeupn(vm_object_t object, short i)
394 VM_OBJECT_ASSERT_WLOCKED(object);
396 object->paging_in_progress -= i;
397 if ((object->flags & OBJ_PIPWNT) && object->paging_in_progress == 0) {
398 vm_object_clear_flag(object, OBJ_PIPWNT);
404 vm_object_pip_wait(vm_object_t object, char *waitid)
407 VM_OBJECT_ASSERT_WLOCKED(object);
408 while (object->paging_in_progress) {
409 object->flags |= OBJ_PIPWNT;
410 VM_OBJECT_SLEEP(object, object, PVM, waitid, 0);
415 * vm_object_allocate:
417 * Returns a new object with the given size.
420 vm_object_allocate(objtype_t type, vm_pindex_t size)
424 object = (vm_object_t)uma_zalloc(obj_zone, M_WAITOK);
425 _vm_object_allocate(type, size, object);
431 * vm_object_reference:
433 * Gets another reference to the given object. Note: OBJ_DEAD
434 * objects can be referenced during final cleaning.
437 vm_object_reference(vm_object_t object)
441 VM_OBJECT_WLOCK(object);
442 vm_object_reference_locked(object);
443 VM_OBJECT_WUNLOCK(object);
447 * vm_object_reference_locked:
449 * Gets another reference to the given object.
451 * The object must be locked.
454 vm_object_reference_locked(vm_object_t object)
458 VM_OBJECT_ASSERT_WLOCKED(object);
460 if (object->type == OBJT_VNODE) {
467 * Handle deallocating an object of type OBJT_VNODE.
470 vm_object_vndeallocate(vm_object_t object)
472 struct vnode *vp = (struct vnode *) object->handle;
474 VM_OBJECT_ASSERT_WLOCKED(object);
475 KASSERT(object->type == OBJT_VNODE,
476 ("vm_object_vndeallocate: not a vnode object"));
477 KASSERT(vp != NULL, ("vm_object_vndeallocate: missing vp"));
479 if (object->ref_count == 0) {
480 vn_printf(vp, "vm_object_vndeallocate ");
481 panic("vm_object_vndeallocate: bad object reference count");
485 if (!umtx_shm_vnobj_persistent && object->ref_count == 1)
486 umtx_shm_object_terminated(object);
489 * The test for text of vp vnode does not need a bypass to
490 * reach right VV_TEXT there, since it is obtained from
493 if (object->ref_count > 1 || (vp->v_vflag & VV_TEXT) == 0) {
495 VM_OBJECT_WUNLOCK(object);
496 /* vrele may need the vnode lock. */
500 VM_OBJECT_WUNLOCK(object);
501 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
503 VM_OBJECT_WLOCK(object);
505 if (object->type == OBJT_DEAD) {
506 VM_OBJECT_WUNLOCK(object);
509 if (object->ref_count == 0)
511 VM_OBJECT_WUNLOCK(object);
518 * vm_object_deallocate:
520 * Release a reference to the specified object,
521 * gained either through a vm_object_allocate
522 * or a vm_object_reference call. When all references
523 * are gone, storage associated with this object
524 * may be relinquished.
526 * No object may be locked.
529 vm_object_deallocate(vm_object_t object)
534 while (object != NULL) {
535 VM_OBJECT_WLOCK(object);
536 if (object->type == OBJT_VNODE) {
537 vm_object_vndeallocate(object);
541 KASSERT(object->ref_count != 0,
542 ("vm_object_deallocate: object deallocated too many times: %d", object->type));
545 * If the reference count goes to 0 we start calling
546 * vm_object_terminate() on the object chain.
547 * A ref count of 1 may be a special case depending on the
548 * shadow count being 0 or 1.
551 if (object->ref_count > 1) {
552 VM_OBJECT_WUNLOCK(object);
554 } else if (object->ref_count == 1) {
555 if (object->type == OBJT_SWAP &&
556 (object->flags & OBJ_TMPFS) != 0) {
557 vp = object->un_pager.swp.swp_tmpfs;
559 VM_OBJECT_WUNLOCK(object);
560 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
561 VM_OBJECT_WLOCK(object);
562 if (object->type == OBJT_DEAD ||
563 object->ref_count != 1) {
564 VM_OBJECT_WUNLOCK(object);
569 if ((object->flags & OBJ_TMPFS) != 0)
574 if (object->shadow_count == 0 &&
575 object->handle == NULL &&
576 (object->type == OBJT_DEFAULT ||
577 (object->type == OBJT_SWAP &&
578 (object->flags & OBJ_TMPFS_NODE) == 0))) {
579 vm_object_set_flag(object, OBJ_ONEMAPPING);
580 } else if ((object->shadow_count == 1) &&
581 (object->handle == NULL) &&
582 (object->type == OBJT_DEFAULT ||
583 object->type == OBJT_SWAP)) {
586 robject = LIST_FIRST(&object->shadow_head);
587 KASSERT(robject != NULL,
588 ("vm_object_deallocate: ref_count: %d, shadow_count: %d",
590 object->shadow_count));
591 KASSERT((robject->flags & OBJ_TMPFS_NODE) == 0,
592 ("shadowed tmpfs v_object %p", object));
593 if (!VM_OBJECT_TRYWLOCK(robject)) {
595 * Avoid a potential deadlock.
598 VM_OBJECT_WUNLOCK(object);
600 * More likely than not the thread
601 * holding robject's lock has lower
602 * priority than the current thread.
603 * Let the lower priority thread run.
609 * Collapse object into its shadow unless its
610 * shadow is dead. In that case, object will
611 * be deallocated by the thread that is
612 * deallocating its shadow.
614 if ((robject->flags & OBJ_DEAD) == 0 &&
615 (robject->handle == NULL) &&
616 (robject->type == OBJT_DEFAULT ||
617 robject->type == OBJT_SWAP)) {
619 robject->ref_count++;
621 if (robject->paging_in_progress) {
622 VM_OBJECT_WUNLOCK(object);
623 vm_object_pip_wait(robject,
625 temp = robject->backing_object;
626 if (object == temp) {
627 VM_OBJECT_WLOCK(object);
630 } else if (object->paging_in_progress) {
631 VM_OBJECT_WUNLOCK(robject);
632 object->flags |= OBJ_PIPWNT;
633 VM_OBJECT_SLEEP(object, object,
634 PDROP | PVM, "objde2", 0);
635 VM_OBJECT_WLOCK(robject);
636 temp = robject->backing_object;
637 if (object == temp) {
638 VM_OBJECT_WLOCK(object);
642 VM_OBJECT_WUNLOCK(object);
644 if (robject->ref_count == 1) {
645 robject->ref_count--;
650 vm_object_collapse(object);
651 VM_OBJECT_WUNLOCK(object);
654 VM_OBJECT_WUNLOCK(robject);
656 VM_OBJECT_WUNLOCK(object);
660 umtx_shm_object_terminated(object);
661 temp = object->backing_object;
663 KASSERT((object->flags & OBJ_TMPFS_NODE) == 0,
664 ("shadowed tmpfs v_object 2 %p", object));
665 VM_OBJECT_WLOCK(temp);
666 LIST_REMOVE(object, shadow_list);
667 temp->shadow_count--;
668 VM_OBJECT_WUNLOCK(temp);
669 object->backing_object = NULL;
672 * Don't double-terminate, we could be in a termination
673 * recursion due to the terminate having to sync data
676 if ((object->flags & OBJ_DEAD) == 0)
677 vm_object_terminate(object);
679 VM_OBJECT_WUNLOCK(object);
685 * vm_object_destroy removes the object from the global object list
686 * and frees the space for the object.
689 vm_object_destroy(vm_object_t object)
693 * Release the allocation charge.
695 if (object->cred != NULL) {
696 swap_release_by_cred(object->charge, object->cred);
698 crfree(object->cred);
703 * Free the space for the object.
705 uma_zfree(obj_zone, object);
709 * vm_object_terminate_pages removes any remaining pageable pages
710 * from the object and resets the object to an empty state.
713 vm_object_terminate_pages(vm_object_t object)
716 struct mtx *mtx, *mtx1;
717 struct vm_pagequeue *pq, *pq1;
719 VM_OBJECT_ASSERT_WLOCKED(object);
725 * Free any remaining pageable pages. This also removes them from the
726 * paging queues. However, don't free wired pages, just remove them
727 * from the object. Rather than incrementally removing each page from
728 * the object, the page and object are reset to any empty state.
730 TAILQ_FOREACH_SAFE(p, &object->memq, listq, p_next) {
731 vm_page_assert_unbusied(p);
732 if ((object->flags & OBJ_UNMANAGED) == 0) {
734 * vm_page_free_prep() only needs the page
735 * lock for managed pages.
737 mtx1 = vm_page_lockptr(p);
742 vm_pagequeue_unlock(pq);
750 if (p->wire_count != 0)
752 PCPU_INC(cnt.v_pfree);
753 p->flags &= ~PG_ZERO;
754 if (p->queue != PQ_NONE) {
755 KASSERT(p->queue < PQ_COUNT, ("vm_object_terminate: "
756 "page %p is not queued", p));
757 pq1 = vm_page_pagequeue(p);
760 vm_pagequeue_unlock(pq);
762 vm_pagequeue_lock(pq);
765 if (vm_page_free_prep(p, true))
768 TAILQ_REMOVE(&object->memq, p, listq);
771 vm_pagequeue_unlock(pq);
775 vm_page_free_phys_pglist(&object->memq);
778 * If the object contained any pages, then reset it to an empty state.
779 * None of the object's fields, including "resident_page_count", were
780 * modified by the preceding loop.
782 if (object->resident_page_count != 0) {
783 vm_radix_reclaim_allnodes(&object->rtree);
784 TAILQ_INIT(&object->memq);
785 object->resident_page_count = 0;
786 if (object->type == OBJT_VNODE)
787 vdrop(object->handle);
792 * vm_object_terminate actually destroys the specified object, freeing
793 * up all previously used resources.
795 * The object must be locked.
796 * This routine may block.
799 vm_object_terminate(vm_object_t object)
802 VM_OBJECT_ASSERT_WLOCKED(object);
805 * Make sure no one uses us.
807 vm_object_set_flag(object, OBJ_DEAD);
810 * wait for the pageout daemon to be done with the object
812 vm_object_pip_wait(object, "objtrm");
814 KASSERT(!object->paging_in_progress,
815 ("vm_object_terminate: pageout in progress"));
818 * Clean and free the pages, as appropriate. All references to the
819 * object are gone, so we don't need to lock it.
821 if (object->type == OBJT_VNODE) {
822 struct vnode *vp = (struct vnode *)object->handle;
825 * Clean pages and flush buffers.
827 vm_object_page_clean(object, 0, 0, OBJPC_SYNC);
828 VM_OBJECT_WUNLOCK(object);
830 vinvalbuf(vp, V_SAVE, 0, 0);
832 BO_LOCK(&vp->v_bufobj);
833 vp->v_bufobj.bo_flag |= BO_DEAD;
834 BO_UNLOCK(&vp->v_bufobj);
836 VM_OBJECT_WLOCK(object);
839 KASSERT(object->ref_count == 0,
840 ("vm_object_terminate: object with references, ref_count=%d",
843 if ((object->flags & OBJ_PG_DTOR) == 0)
844 vm_object_terminate_pages(object);
846 #if VM_NRESERVLEVEL > 0
847 if (__predict_false(!LIST_EMPTY(&object->rvq)))
848 vm_reserv_break_all(object);
851 KASSERT(object->cred == NULL || object->type == OBJT_DEFAULT ||
852 object->type == OBJT_SWAP,
853 ("%s: non-swap obj %p has cred", __func__, object));
856 * Let the pager know object is dead.
858 vm_pager_deallocate(object);
859 VM_OBJECT_WUNLOCK(object);
861 vm_object_destroy(object);
865 * Make the page read-only so that we can clear the object flags. However, if
866 * this is a nosync mmap then the object is likely to stay dirty so do not
867 * mess with the page and do not clear the object flags. Returns TRUE if the
868 * page should be flushed, and FALSE otherwise.
871 vm_object_page_remove_write(vm_page_t p, int flags, boolean_t *clearobjflags)
875 * If we have been asked to skip nosync pages and this is a
876 * nosync page, skip it. Note that the object flags were not
877 * cleared in this case so we do not have to set them.
879 if ((flags & OBJPC_NOSYNC) != 0 && (p->oflags & VPO_NOSYNC) != 0) {
880 *clearobjflags = FALSE;
883 pmap_remove_write(p);
884 return (p->dirty != 0);
889 * vm_object_page_clean
891 * Clean all dirty pages in the specified range of object. Leaves page
892 * on whatever queue it is currently on. If NOSYNC is set then do not
893 * write out pages with VPO_NOSYNC set (originally comes from MAP_NOSYNC),
894 * leaving the object dirty.
896 * When stuffing pages asynchronously, allow clustering. XXX we need a
897 * synchronous clustering mode implementation.
899 * Odd semantics: if start == end, we clean everything.
901 * The object must be locked.
903 * Returns FALSE if some page from the range was not written, as
904 * reported by the pager, and TRUE otherwise.
907 vm_object_page_clean(vm_object_t object, vm_ooffset_t start, vm_ooffset_t end,
911 vm_pindex_t pi, tend, tstart;
912 int curgeneration, n, pagerflags;
913 boolean_t clearobjflags, eio, res;
915 VM_OBJECT_ASSERT_WLOCKED(object);
918 * The OBJ_MIGHTBEDIRTY flag is only set for OBJT_VNODE
919 * objects. The check below prevents the function from
920 * operating on non-vnode objects.
922 if ((object->flags & OBJ_MIGHTBEDIRTY) == 0 ||
923 object->resident_page_count == 0)
926 pagerflags = (flags & (OBJPC_SYNC | OBJPC_INVAL)) != 0 ?
927 VM_PAGER_PUT_SYNC : VM_PAGER_CLUSTER_OK;
928 pagerflags |= (flags & OBJPC_INVAL) != 0 ? VM_PAGER_PUT_INVAL : 0;
930 tstart = OFF_TO_IDX(start);
931 tend = (end == 0) ? object->size : OFF_TO_IDX(end + PAGE_MASK);
932 clearobjflags = tstart == 0 && tend >= object->size;
936 curgeneration = object->generation;
938 for (p = vm_page_find_least(object, tstart); p != NULL; p = np) {
942 np = TAILQ_NEXT(p, listq);
945 if (vm_page_sleep_if_busy(p, "vpcwai")) {
946 if (object->generation != curgeneration) {
947 if ((flags & OBJPC_SYNC) != 0)
950 clearobjflags = FALSE;
952 np = vm_page_find_least(object, pi);
955 if (!vm_object_page_remove_write(p, flags, &clearobjflags))
958 n = vm_object_page_collect_flush(object, p, pagerflags,
959 flags, &clearobjflags, &eio);
962 clearobjflags = FALSE;
964 if (object->generation != curgeneration) {
965 if ((flags & OBJPC_SYNC) != 0)
968 clearobjflags = FALSE;
972 * If the VOP_PUTPAGES() did a truncated write, so
973 * that even the first page of the run is not fully
974 * written, vm_pageout_flush() returns 0 as the run
975 * length. Since the condition that caused truncated
976 * write may be permanent, e.g. exhausted free space,
977 * accepting n == 0 would cause an infinite loop.
979 * Forwarding the iterator leaves the unwritten page
980 * behind, but there is not much we can do there if
981 * filesystem refuses to write it.
985 clearobjflags = FALSE;
987 np = vm_page_find_least(object, pi + n);
990 VOP_FSYNC(vp, (pagerflags & VM_PAGER_PUT_SYNC) ? MNT_WAIT : 0);
994 vm_object_clear_flag(object, OBJ_MIGHTBEDIRTY);
999 vm_object_page_collect_flush(vm_object_t object, vm_page_t p, int pagerflags,
1000 int flags, boolean_t *clearobjflags, boolean_t *eio)
1002 vm_page_t ma[vm_pageout_page_count], p_first, tp;
1003 int count, i, mreq, runlen;
1005 vm_page_lock_assert(p, MA_NOTOWNED);
1006 VM_OBJECT_ASSERT_WLOCKED(object);
1011 for (tp = p; count < vm_pageout_page_count; count++) {
1012 tp = vm_page_next(tp);
1013 if (tp == NULL || vm_page_busied(tp))
1015 if (!vm_object_page_remove_write(tp, flags, clearobjflags))
1019 for (p_first = p; count < vm_pageout_page_count; count++) {
1020 tp = vm_page_prev(p_first);
1021 if (tp == NULL || vm_page_busied(tp))
1023 if (!vm_object_page_remove_write(tp, flags, clearobjflags))
1029 for (tp = p_first, i = 0; i < count; tp = TAILQ_NEXT(tp, listq), i++)
1032 vm_pageout_flush(ma, count, pagerflags, mreq, &runlen, eio);
1037 * Note that there is absolutely no sense in writing out
1038 * anonymous objects, so we track down the vnode object
1040 * We invalidate (remove) all pages from the address space
1041 * for semantic correctness.
1043 * If the backing object is a device object with unmanaged pages, then any
1044 * mappings to the specified range of pages must be removed before this
1045 * function is called.
1047 * Note: certain anonymous maps, such as MAP_NOSYNC maps,
1048 * may start out with a NULL object.
1051 vm_object_sync(vm_object_t object, vm_ooffset_t offset, vm_size_t size,
1052 boolean_t syncio, boolean_t invalidate)
1054 vm_object_t backing_object;
1057 int error, flags, fsync_after;
1064 VM_OBJECT_WLOCK(object);
1065 while ((backing_object = object->backing_object) != NULL) {
1066 VM_OBJECT_WLOCK(backing_object);
1067 offset += object->backing_object_offset;
1068 VM_OBJECT_WUNLOCK(object);
1069 object = backing_object;
1070 if (object->size < OFF_TO_IDX(offset + size))
1071 size = IDX_TO_OFF(object->size) - offset;
1074 * Flush pages if writing is allowed, invalidate them
1075 * if invalidation requested. Pages undergoing I/O
1076 * will be ignored by vm_object_page_remove().
1078 * We cannot lock the vnode and then wait for paging
1079 * to complete without deadlocking against vm_fault.
1080 * Instead we simply call vm_object_page_remove() and
1081 * allow it to block internally on a page-by-page
1082 * basis when it encounters pages undergoing async
1085 if (object->type == OBJT_VNODE &&
1086 (object->flags & OBJ_MIGHTBEDIRTY) != 0 &&
1087 ((vp = object->handle)->v_vflag & VV_NOSYNC) == 0) {
1088 VM_OBJECT_WUNLOCK(object);
1089 (void) vn_start_write(vp, &mp, V_WAIT);
1090 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1091 if (syncio && !invalidate && offset == 0 &&
1092 atop(size) == object->size) {
1094 * If syncing the whole mapping of the file,
1095 * it is faster to schedule all the writes in
1096 * async mode, also allowing the clustering,
1097 * and then wait for i/o to complete.
1102 flags = (syncio || invalidate) ? OBJPC_SYNC : 0;
1103 flags |= invalidate ? (OBJPC_SYNC | OBJPC_INVAL) : 0;
1104 fsync_after = FALSE;
1106 VM_OBJECT_WLOCK(object);
1107 res = vm_object_page_clean(object, offset, offset + size,
1109 VM_OBJECT_WUNLOCK(object);
1111 error = VOP_FSYNC(vp, MNT_WAIT, curthread);
1113 vn_finished_write(mp);
1116 VM_OBJECT_WLOCK(object);
1118 if ((object->type == OBJT_VNODE ||
1119 object->type == OBJT_DEVICE) && invalidate) {
1120 if (object->type == OBJT_DEVICE)
1122 * The option OBJPR_NOTMAPPED must be passed here
1123 * because vm_object_page_remove() cannot remove
1124 * unmanaged mappings.
1126 flags = OBJPR_NOTMAPPED;
1130 flags = OBJPR_CLEANONLY;
1131 vm_object_page_remove(object, OFF_TO_IDX(offset),
1132 OFF_TO_IDX(offset + size + PAGE_MASK), flags);
1134 VM_OBJECT_WUNLOCK(object);
1139 * Determine whether the given advice can be applied to the object. Advice is
1140 * not applied to unmanaged pages since they never belong to page queues, and
1141 * since MADV_FREE is destructive, it can apply only to anonymous pages that
1142 * have been mapped at most once.
1145 vm_object_advice_applies(vm_object_t object, int advice)
1148 if ((object->flags & OBJ_UNMANAGED) != 0)
1150 if (advice != MADV_FREE)
1152 return ((object->type == OBJT_DEFAULT || object->type == OBJT_SWAP) &&
1153 (object->flags & OBJ_ONEMAPPING) != 0);
1157 vm_object_madvise_freespace(vm_object_t object, int advice, vm_pindex_t pindex,
1161 if (advice == MADV_FREE && object->type == OBJT_SWAP)
1162 swap_pager_freespace(object, pindex, size);
1166 * vm_object_madvise:
1168 * Implements the madvise function at the object/page level.
1170 * MADV_WILLNEED (any object)
1172 * Activate the specified pages if they are resident.
1174 * MADV_DONTNEED (any object)
1176 * Deactivate the specified pages if they are resident.
1178 * MADV_FREE (OBJT_DEFAULT/OBJT_SWAP objects,
1179 * OBJ_ONEMAPPING only)
1181 * Deactivate and clean the specified pages if they are
1182 * resident. This permits the process to reuse the pages
1183 * without faulting or the kernel to reclaim the pages
1187 vm_object_madvise(vm_object_t object, vm_pindex_t pindex, vm_pindex_t end,
1190 vm_pindex_t tpindex;
1191 vm_object_t backing_object, tobject;
1198 VM_OBJECT_WLOCK(object);
1199 if (!vm_object_advice_applies(object, advice)) {
1200 VM_OBJECT_WUNLOCK(object);
1203 for (m = vm_page_find_least(object, pindex); pindex < end; pindex++) {
1207 * If the next page isn't resident in the top-level object, we
1208 * need to search the shadow chain. When applying MADV_FREE, we
1209 * take care to release any swap space used to store
1210 * non-resident pages.
1212 if (m == NULL || pindex < m->pindex) {
1214 * Optimize a common case: if the top-level object has
1215 * no backing object, we can skip over the non-resident
1216 * range in constant time.
1218 if (object->backing_object == NULL) {
1219 tpindex = (m != NULL && m->pindex < end) ?
1221 vm_object_madvise_freespace(object, advice,
1222 pindex, tpindex - pindex);
1223 if ((pindex = tpindex) == end)
1230 vm_object_madvise_freespace(tobject, advice,
1233 * Prepare to search the next object in the
1236 backing_object = tobject->backing_object;
1237 if (backing_object == NULL)
1239 VM_OBJECT_WLOCK(backing_object);
1241 OFF_TO_IDX(tobject->backing_object_offset);
1242 if (tobject != object)
1243 VM_OBJECT_WUNLOCK(tobject);
1244 tobject = backing_object;
1245 if (!vm_object_advice_applies(tobject, advice))
1247 } while ((tm = vm_page_lookup(tobject, tpindex)) ==
1252 m = TAILQ_NEXT(m, listq);
1256 * If the page is not in a normal state, skip it.
1258 if (tm->valid != VM_PAGE_BITS_ALL)
1261 if (tm->hold_count != 0 || tm->wire_count != 0) {
1265 KASSERT((tm->flags & PG_FICTITIOUS) == 0,
1266 ("vm_object_madvise: page %p is fictitious", tm));
1267 KASSERT((tm->oflags & VPO_UNMANAGED) == 0,
1268 ("vm_object_madvise: page %p is not managed", tm));
1269 if (vm_page_busied(tm)) {
1270 if (object != tobject)
1271 VM_OBJECT_WUNLOCK(tobject);
1272 VM_OBJECT_WUNLOCK(object);
1273 if (advice == MADV_WILLNEED) {
1275 * Reference the page before unlocking and
1276 * sleeping so that the page daemon is less
1277 * likely to reclaim it.
1279 vm_page_aflag_set(tm, PGA_REFERENCED);
1281 vm_page_busy_sleep(tm, "madvpo", false);
1284 vm_page_advise(tm, advice);
1286 vm_object_madvise_freespace(tobject, advice, tm->pindex, 1);
1288 if (tobject != object)
1289 VM_OBJECT_WUNLOCK(tobject);
1291 VM_OBJECT_WUNLOCK(object);
1297 * Create a new object which is backed by the
1298 * specified existing object range. The source
1299 * object reference is deallocated.
1301 * The new object and offset into that object
1302 * are returned in the source parameters.
1306 vm_object_t *object, /* IN/OUT */
1307 vm_ooffset_t *offset, /* IN/OUT */
1316 * Don't create the new object if the old object isn't shared.
1318 if (source != NULL) {
1319 VM_OBJECT_WLOCK(source);
1320 if (source->ref_count == 1 &&
1321 source->handle == NULL &&
1322 (source->type == OBJT_DEFAULT ||
1323 source->type == OBJT_SWAP)) {
1324 VM_OBJECT_WUNLOCK(source);
1327 VM_OBJECT_WUNLOCK(source);
1331 * Allocate a new object with the given length.
1333 result = vm_object_allocate(OBJT_DEFAULT, atop(length));
1336 * The new object shadows the source object, adding a reference to it.
1337 * Our caller changes his reference to point to the new object,
1338 * removing a reference to the source object. Net result: no change
1339 * of reference count.
1341 * Try to optimize the result object's page color when shadowing
1342 * in order to maintain page coloring consistency in the combined
1345 result->backing_object = source;
1347 * Store the offset into the source object, and fix up the offset into
1350 result->backing_object_offset = *offset;
1351 if (source != NULL) {
1352 VM_OBJECT_WLOCK(source);
1353 LIST_INSERT_HEAD(&source->shadow_head, result, shadow_list);
1354 source->shadow_count++;
1355 #if VM_NRESERVLEVEL > 0
1356 result->flags |= source->flags & OBJ_COLORED;
1357 result->pg_color = (source->pg_color + OFF_TO_IDX(*offset)) &
1358 ((1 << (VM_NFREEORDER - 1)) - 1);
1360 VM_OBJECT_WUNLOCK(source);
1365 * Return the new things
1374 * Split the pages in a map entry into a new object. This affords
1375 * easier removal of unused pages, and keeps object inheritance from
1376 * being a negative impact on memory usage.
1379 vm_object_split(vm_map_entry_t entry)
1381 vm_page_t m, m_next;
1382 vm_object_t orig_object, new_object, source;
1383 vm_pindex_t idx, offidxstart;
1386 orig_object = entry->object.vm_object;
1387 if (orig_object->type != OBJT_DEFAULT && orig_object->type != OBJT_SWAP)
1389 if (orig_object->ref_count <= 1)
1391 VM_OBJECT_WUNLOCK(orig_object);
1393 offidxstart = OFF_TO_IDX(entry->offset);
1394 size = atop(entry->end - entry->start);
1397 * If swap_pager_copy() is later called, it will convert new_object
1398 * into a swap object.
1400 new_object = vm_object_allocate(OBJT_DEFAULT, size);
1403 * At this point, the new object is still private, so the order in
1404 * which the original and new objects are locked does not matter.
1406 VM_OBJECT_WLOCK(new_object);
1407 VM_OBJECT_WLOCK(orig_object);
1408 source = orig_object->backing_object;
1409 if (source != NULL) {
1410 VM_OBJECT_WLOCK(source);
1411 if ((source->flags & OBJ_DEAD) != 0) {
1412 VM_OBJECT_WUNLOCK(source);
1413 VM_OBJECT_WUNLOCK(orig_object);
1414 VM_OBJECT_WUNLOCK(new_object);
1415 vm_object_deallocate(new_object);
1416 VM_OBJECT_WLOCK(orig_object);
1419 LIST_INSERT_HEAD(&source->shadow_head,
1420 new_object, shadow_list);
1421 source->shadow_count++;
1422 vm_object_reference_locked(source); /* for new_object */
1423 vm_object_clear_flag(source, OBJ_ONEMAPPING);
1424 VM_OBJECT_WUNLOCK(source);
1425 new_object->backing_object_offset =
1426 orig_object->backing_object_offset + entry->offset;
1427 new_object->backing_object = source;
1429 if (orig_object->cred != NULL) {
1430 new_object->cred = orig_object->cred;
1431 crhold(orig_object->cred);
1432 new_object->charge = ptoa(size);
1433 KASSERT(orig_object->charge >= ptoa(size),
1434 ("orig_object->charge < 0"));
1435 orig_object->charge -= ptoa(size);
1438 m = vm_page_find_least(orig_object, offidxstart);
1439 for (; m != NULL && (idx = m->pindex - offidxstart) < size;
1441 m_next = TAILQ_NEXT(m, listq);
1444 * We must wait for pending I/O to complete before we can
1447 * We do not have to VM_PROT_NONE the page as mappings should
1448 * not be changed by this operation.
1450 if (vm_page_busied(m)) {
1451 VM_OBJECT_WUNLOCK(new_object);
1453 VM_OBJECT_WUNLOCK(orig_object);
1454 vm_page_busy_sleep(m, "spltwt", false);
1455 VM_OBJECT_WLOCK(orig_object);
1456 VM_OBJECT_WLOCK(new_object);
1460 /* vm_page_rename() will dirty the page. */
1461 if (vm_page_rename(m, new_object, idx)) {
1462 VM_OBJECT_WUNLOCK(new_object);
1463 VM_OBJECT_WUNLOCK(orig_object);
1465 VM_OBJECT_WLOCK(orig_object);
1466 VM_OBJECT_WLOCK(new_object);
1469 #if VM_NRESERVLEVEL > 0
1471 * If some of the reservation's allocated pages remain with
1472 * the original object, then transferring the reservation to
1473 * the new object is neither particularly beneficial nor
1474 * particularly harmful as compared to leaving the reservation
1475 * with the original object. If, however, all of the
1476 * reservation's allocated pages are transferred to the new
1477 * object, then transferring the reservation is typically
1478 * beneficial. Determining which of these two cases applies
1479 * would be more costly than unconditionally renaming the
1482 vm_reserv_rename(m, new_object, orig_object, offidxstart);
1484 if (orig_object->type == OBJT_SWAP)
1487 if (orig_object->type == OBJT_SWAP) {
1489 * swap_pager_copy() can sleep, in which case the orig_object's
1490 * and new_object's locks are released and reacquired.
1492 swap_pager_copy(orig_object, new_object, offidxstart, 0);
1493 TAILQ_FOREACH(m, &new_object->memq, listq)
1496 VM_OBJECT_WUNLOCK(orig_object);
1497 VM_OBJECT_WUNLOCK(new_object);
1498 entry->object.vm_object = new_object;
1499 entry->offset = 0LL;
1500 vm_object_deallocate(orig_object);
1501 VM_OBJECT_WLOCK(new_object);
1504 #define OBSC_COLLAPSE_NOWAIT 0x0002
1505 #define OBSC_COLLAPSE_WAIT 0x0004
1508 vm_object_collapse_scan_wait(vm_object_t object, vm_page_t p, vm_page_t next,
1511 vm_object_t backing_object;
1513 VM_OBJECT_ASSERT_WLOCKED(object);
1514 backing_object = object->backing_object;
1515 VM_OBJECT_ASSERT_WLOCKED(backing_object);
1517 KASSERT(p == NULL || vm_page_busied(p), ("unbusy page %p", p));
1518 KASSERT(p == NULL || p->object == object || p->object == backing_object,
1519 ("invalid ownership %p %p %p", p, object, backing_object));
1520 if ((op & OBSC_COLLAPSE_NOWAIT) != 0)
1524 VM_OBJECT_WUNLOCK(object);
1525 VM_OBJECT_WUNLOCK(backing_object);
1526 /* The page is only NULL when rename fails. */
1530 vm_page_busy_sleep(p, "vmocol", false);
1531 VM_OBJECT_WLOCK(object);
1532 VM_OBJECT_WLOCK(backing_object);
1533 return (TAILQ_FIRST(&backing_object->memq));
1537 vm_object_scan_all_shadowed(vm_object_t object)
1539 vm_object_t backing_object;
1541 vm_pindex_t backing_offset_index, new_pindex, pi, ps;
1543 VM_OBJECT_ASSERT_WLOCKED(object);
1544 VM_OBJECT_ASSERT_WLOCKED(object->backing_object);
1546 backing_object = object->backing_object;
1549 * Initial conditions:
1551 * We do not want to have to test for the existence of swap
1552 * pages in the backing object. XXX but with the new swapper this
1553 * would be pretty easy to do.
1555 if (backing_object->type != OBJT_DEFAULT &&
1556 backing_object->type != OBJT_SWAP)
1559 pi = backing_offset_index = OFF_TO_IDX(object->backing_object_offset);
1560 p = vm_page_find_least(backing_object, pi);
1561 ps = swap_pager_find_least(backing_object, pi);
1564 * Only check pages inside the parent object's range and
1565 * inside the parent object's mapping of the backing object.
1568 if (p != NULL && p->pindex < pi)
1569 p = TAILQ_NEXT(p, listq);
1571 ps = swap_pager_find_least(backing_object, pi);
1572 if (p == NULL && ps >= backing_object->size)
1577 pi = MIN(p->pindex, ps);
1579 new_pindex = pi - backing_offset_index;
1580 if (new_pindex >= object->size)
1584 * See if the parent has the page or if the parent's object
1585 * pager has the page. If the parent has the page but the page
1586 * is not valid, the parent's object pager must have the page.
1588 * If this fails, the parent does not completely shadow the
1589 * object and we might as well give up now.
1591 pp = vm_page_lookup(object, new_pindex);
1592 if ((pp == NULL || pp->valid == 0) &&
1593 !vm_pager_has_page(object, new_pindex, NULL, NULL))
1600 vm_object_collapse_scan(vm_object_t object, int op)
1602 vm_object_t backing_object;
1603 vm_page_t next, p, pp;
1604 vm_pindex_t backing_offset_index, new_pindex;
1606 VM_OBJECT_ASSERT_WLOCKED(object);
1607 VM_OBJECT_ASSERT_WLOCKED(object->backing_object);
1609 backing_object = object->backing_object;
1610 backing_offset_index = OFF_TO_IDX(object->backing_object_offset);
1613 * Initial conditions
1615 if ((op & OBSC_COLLAPSE_WAIT) != 0)
1616 vm_object_set_flag(backing_object, OBJ_DEAD);
1621 for (p = TAILQ_FIRST(&backing_object->memq); p != NULL; p = next) {
1622 next = TAILQ_NEXT(p, listq);
1623 new_pindex = p->pindex - backing_offset_index;
1626 * Check for busy page
1628 if (vm_page_busied(p)) {
1629 next = vm_object_collapse_scan_wait(object, p, next, op);
1633 KASSERT(p->object == backing_object,
1634 ("vm_object_collapse_scan: object mismatch"));
1636 if (p->pindex < backing_offset_index ||
1637 new_pindex >= object->size) {
1638 if (backing_object->type == OBJT_SWAP)
1639 swap_pager_freespace(backing_object, p->pindex,
1643 * Page is out of the parent object's range, we can
1644 * simply destroy it.
1647 KASSERT(!pmap_page_is_mapped(p),
1648 ("freeing mapped page %p", p));
1649 if (p->wire_count == 0)
1657 pp = vm_page_lookup(object, new_pindex);
1658 if (pp != NULL && vm_page_busied(pp)) {
1660 * The page in the parent is busy and possibly not
1661 * (yet) valid. Until its state is finalized by the
1662 * busy bit owner, we can't tell whether it shadows the
1663 * original page. Therefore, we must either skip it
1664 * and the original (backing_object) page or wait for
1665 * its state to be finalized.
1667 * This is due to a race with vm_fault() where we must
1668 * unbusy the original (backing_obj) page before we can
1669 * (re)lock the parent. Hence we can get here.
1671 next = vm_object_collapse_scan_wait(object, pp, next,
1676 KASSERT(pp == NULL || pp->valid != 0,
1677 ("unbusy invalid page %p", pp));
1679 if (pp != NULL || vm_pager_has_page(object, new_pindex, NULL,
1682 * The page already exists in the parent OR swap exists
1683 * for this location in the parent. Leave the parent's
1684 * page alone. Destroy the original page from the
1687 if (backing_object->type == OBJT_SWAP)
1688 swap_pager_freespace(backing_object, p->pindex,
1691 KASSERT(!pmap_page_is_mapped(p),
1692 ("freeing mapped page %p", p));
1693 if (p->wire_count == 0)
1702 * Page does not exist in parent, rename the page from the
1703 * backing object to the main object.
1705 * If the page was mapped to a process, it can remain mapped
1706 * through the rename. vm_page_rename() will dirty the page.
1708 if (vm_page_rename(p, object, new_pindex)) {
1709 next = vm_object_collapse_scan_wait(object, NULL, next,
1714 /* Use the old pindex to free the right page. */
1715 if (backing_object->type == OBJT_SWAP)
1716 swap_pager_freespace(backing_object,
1717 new_pindex + backing_offset_index, 1);
1719 #if VM_NRESERVLEVEL > 0
1721 * Rename the reservation.
1723 vm_reserv_rename(p, object, backing_object,
1724 backing_offset_index);
1732 * this version of collapse allows the operation to occur earlier and
1733 * when paging_in_progress is true for an object... This is not a complete
1734 * operation, but should plug 99.9% of the rest of the leaks.
1737 vm_object_qcollapse(vm_object_t object)
1739 vm_object_t backing_object = object->backing_object;
1741 VM_OBJECT_ASSERT_WLOCKED(object);
1742 VM_OBJECT_ASSERT_WLOCKED(backing_object);
1744 if (backing_object->ref_count != 1)
1747 vm_object_collapse_scan(object, OBSC_COLLAPSE_NOWAIT);
1751 * vm_object_collapse:
1753 * Collapse an object with the object backing it.
1754 * Pages in the backing object are moved into the
1755 * parent, and the backing object is deallocated.
1758 vm_object_collapse(vm_object_t object)
1760 vm_object_t backing_object, new_backing_object;
1762 VM_OBJECT_ASSERT_WLOCKED(object);
1766 * Verify that the conditions are right for collapse:
1768 * The object exists and the backing object exists.
1770 if ((backing_object = object->backing_object) == NULL)
1774 * we check the backing object first, because it is most likely
1777 VM_OBJECT_WLOCK(backing_object);
1778 if (backing_object->handle != NULL ||
1779 (backing_object->type != OBJT_DEFAULT &&
1780 backing_object->type != OBJT_SWAP) ||
1781 (backing_object->flags & (OBJ_DEAD | OBJ_NOSPLIT)) != 0 ||
1782 object->handle != NULL ||
1783 (object->type != OBJT_DEFAULT &&
1784 object->type != OBJT_SWAP) ||
1785 (object->flags & OBJ_DEAD)) {
1786 VM_OBJECT_WUNLOCK(backing_object);
1790 if (object->paging_in_progress != 0 ||
1791 backing_object->paging_in_progress != 0) {
1792 vm_object_qcollapse(object);
1793 VM_OBJECT_WUNLOCK(backing_object);
1798 * We know that we can either collapse the backing object (if
1799 * the parent is the only reference to it) or (perhaps) have
1800 * the parent bypass the object if the parent happens to shadow
1801 * all the resident pages in the entire backing object.
1803 * This is ignoring pager-backed pages such as swap pages.
1804 * vm_object_collapse_scan fails the shadowing test in this
1807 if (backing_object->ref_count == 1) {
1808 vm_object_pip_add(object, 1);
1809 vm_object_pip_add(backing_object, 1);
1812 * If there is exactly one reference to the backing
1813 * object, we can collapse it into the parent.
1815 vm_object_collapse_scan(object, OBSC_COLLAPSE_WAIT);
1817 #if VM_NRESERVLEVEL > 0
1819 * Break any reservations from backing_object.
1821 if (__predict_false(!LIST_EMPTY(&backing_object->rvq)))
1822 vm_reserv_break_all(backing_object);
1826 * Move the pager from backing_object to object.
1828 if (backing_object->type == OBJT_SWAP) {
1830 * swap_pager_copy() can sleep, in which case
1831 * the backing_object's and object's locks are
1832 * released and reacquired.
1833 * Since swap_pager_copy() is being asked to
1834 * destroy the source, it will change the
1835 * backing_object's type to OBJT_DEFAULT.
1840 OFF_TO_IDX(object->backing_object_offset), TRUE);
1843 * Object now shadows whatever backing_object did.
1844 * Note that the reference to
1845 * backing_object->backing_object moves from within
1846 * backing_object to within object.
1848 LIST_REMOVE(object, shadow_list);
1849 backing_object->shadow_count--;
1850 if (backing_object->backing_object) {
1851 VM_OBJECT_WLOCK(backing_object->backing_object);
1852 LIST_REMOVE(backing_object, shadow_list);
1854 &backing_object->backing_object->shadow_head,
1855 object, shadow_list);
1857 * The shadow_count has not changed.
1859 VM_OBJECT_WUNLOCK(backing_object->backing_object);
1861 object->backing_object = backing_object->backing_object;
1862 object->backing_object_offset +=
1863 backing_object->backing_object_offset;
1866 * Discard backing_object.
1868 * Since the backing object has no pages, no pager left,
1869 * and no object references within it, all that is
1870 * necessary is to dispose of it.
1872 KASSERT(backing_object->ref_count == 1, (
1873 "backing_object %p was somehow re-referenced during collapse!",
1875 vm_object_pip_wakeup(backing_object);
1876 backing_object->type = OBJT_DEAD;
1877 backing_object->ref_count = 0;
1878 VM_OBJECT_WUNLOCK(backing_object);
1879 vm_object_destroy(backing_object);
1881 vm_object_pip_wakeup(object);
1885 * If we do not entirely shadow the backing object,
1886 * there is nothing we can do so we give up.
1888 if (object->resident_page_count != object->size &&
1889 !vm_object_scan_all_shadowed(object)) {
1890 VM_OBJECT_WUNLOCK(backing_object);
1895 * Make the parent shadow the next object in the
1896 * chain. Deallocating backing_object will not remove
1897 * it, since its reference count is at least 2.
1899 LIST_REMOVE(object, shadow_list);
1900 backing_object->shadow_count--;
1902 new_backing_object = backing_object->backing_object;
1903 if ((object->backing_object = new_backing_object) != NULL) {
1904 VM_OBJECT_WLOCK(new_backing_object);
1906 &new_backing_object->shadow_head,
1910 new_backing_object->shadow_count++;
1911 vm_object_reference_locked(new_backing_object);
1912 VM_OBJECT_WUNLOCK(new_backing_object);
1913 object->backing_object_offset +=
1914 backing_object->backing_object_offset;
1918 * Drop the reference count on backing_object. Since
1919 * its ref_count was at least 2, it will not vanish.
1921 backing_object->ref_count--;
1922 VM_OBJECT_WUNLOCK(backing_object);
1927 * Try again with this object's new backing object.
1933 * vm_object_page_remove:
1935 * For the given object, either frees or invalidates each of the
1936 * specified pages. In general, a page is freed. However, if a page is
1937 * wired for any reason other than the existence of a managed, wired
1938 * mapping, then it may be invalidated but not removed from the object.
1939 * Pages are specified by the given range ["start", "end") and the option
1940 * OBJPR_CLEANONLY. As a special case, if "end" is zero, then the range
1941 * extends from "start" to the end of the object. If the option
1942 * OBJPR_CLEANONLY is specified, then only the non-dirty pages within the
1943 * specified range are affected. If the option OBJPR_NOTMAPPED is
1944 * specified, then the pages within the specified range must have no
1945 * mappings. Otherwise, if this option is not specified, any mappings to
1946 * the specified pages are removed before the pages are freed or
1949 * In general, this operation should only be performed on objects that
1950 * contain managed pages. There are, however, two exceptions. First, it
1951 * is performed on the kernel and kmem objects by vm_map_entry_delete().
1952 * Second, it is used by msync(..., MS_INVALIDATE) to invalidate device-
1953 * backed pages. In both of these cases, the option OBJPR_CLEANONLY must
1954 * not be specified and the option OBJPR_NOTMAPPED must be specified.
1956 * The object must be locked.
1959 vm_object_page_remove(vm_object_t object, vm_pindex_t start, vm_pindex_t end,
1966 VM_OBJECT_ASSERT_WLOCKED(object);
1967 KASSERT((object->flags & OBJ_UNMANAGED) == 0 ||
1968 (options & (OBJPR_CLEANONLY | OBJPR_NOTMAPPED)) == OBJPR_NOTMAPPED,
1969 ("vm_object_page_remove: illegal options for object %p", object));
1970 if (object->resident_page_count == 0)
1972 vm_object_pip_add(object, 1);
1975 p = vm_page_find_least(object, start);
1979 * Here, the variable "p" is either (1) the page with the least pindex
1980 * greater than or equal to the parameter "start" or (2) NULL.
1982 for (; p != NULL && (p->pindex < end || end == 0); p = next) {
1983 next = TAILQ_NEXT(p, listq);
1986 * If the page is wired for any reason besides the existence
1987 * of managed, wired mappings, then it cannot be freed. For
1988 * example, fictitious pages, which represent device memory,
1989 * are inherently wired and cannot be freed. They can,
1990 * however, be invalidated if the option OBJPR_CLEANONLY is
1993 vm_page_change_lock(p, &mtx);
1994 if (vm_page_xbusied(p)) {
1995 VM_OBJECT_WUNLOCK(object);
1996 vm_page_busy_sleep(p, "vmopax", true);
1997 VM_OBJECT_WLOCK(object);
2000 if (p->wire_count != 0) {
2001 if ((options & OBJPR_NOTMAPPED) == 0 &&
2002 object->ref_count != 0)
2004 if ((options & OBJPR_CLEANONLY) == 0) {
2010 if (vm_page_busied(p)) {
2011 VM_OBJECT_WUNLOCK(object);
2012 vm_page_busy_sleep(p, "vmopar", false);
2013 VM_OBJECT_WLOCK(object);
2016 KASSERT((p->flags & PG_FICTITIOUS) == 0,
2017 ("vm_object_page_remove: page %p is fictitious", p));
2018 if ((options & OBJPR_CLEANONLY) != 0 && p->valid != 0) {
2019 if ((options & OBJPR_NOTMAPPED) == 0 &&
2020 object->ref_count != 0)
2021 pmap_remove_write(p);
2025 if ((options & OBJPR_NOTMAPPED) == 0 && object->ref_count != 0)
2027 p->flags &= ~PG_ZERO;
2028 if (vm_page_free_prep(p, false))
2029 TAILQ_INSERT_TAIL(&pgl, p, listq);
2033 vm_page_free_phys_pglist(&pgl);
2034 vm_object_pip_wakeup(object);
2038 * vm_object_page_noreuse:
2040 * For the given object, attempt to move the specified pages to
2041 * the head of the inactive queue. This bypasses regular LRU
2042 * operation and allows the pages to be reused quickly under memory
2043 * pressure. If a page is wired for any reason, then it will not
2044 * be queued. Pages are specified by the range ["start", "end").
2045 * As a special case, if "end" is zero, then the range extends from
2046 * "start" to the end of the object.
2048 * This operation should only be performed on objects that
2049 * contain non-fictitious, managed pages.
2051 * The object must be locked.
2054 vm_object_page_noreuse(vm_object_t object, vm_pindex_t start, vm_pindex_t end)
2059 VM_OBJECT_ASSERT_LOCKED(object);
2060 KASSERT((object->flags & (OBJ_FICTITIOUS | OBJ_UNMANAGED)) == 0,
2061 ("vm_object_page_noreuse: illegal object %p", object));
2062 if (object->resident_page_count == 0)
2064 p = vm_page_find_least(object, start);
2067 * Here, the variable "p" is either (1) the page with the least pindex
2068 * greater than or equal to the parameter "start" or (2) NULL.
2071 for (; p != NULL && (p->pindex < end || end == 0); p = next) {
2072 next = TAILQ_NEXT(p, listq);
2073 vm_page_change_lock(p, &mtx);
2074 vm_page_deactivate_noreuse(p);
2081 * Populate the specified range of the object with valid pages. Returns
2082 * TRUE if the range is successfully populated and FALSE otherwise.
2084 * Note: This function should be optimized to pass a larger array of
2085 * pages to vm_pager_get_pages() before it is applied to a non-
2086 * OBJT_DEVICE object.
2088 * The object must be locked.
2091 vm_object_populate(vm_object_t object, vm_pindex_t start, vm_pindex_t end)
2097 VM_OBJECT_ASSERT_WLOCKED(object);
2098 for (pindex = start; pindex < end; pindex++) {
2099 m = vm_page_grab(object, pindex, VM_ALLOC_NORMAL);
2100 if (m->valid != VM_PAGE_BITS_ALL) {
2101 rv = vm_pager_get_pages(object, &m, 1, NULL, NULL);
2102 if (rv != VM_PAGER_OK) {
2110 * Keep "m" busy because a subsequent iteration may unlock
2114 if (pindex > start) {
2115 m = vm_page_lookup(object, start);
2116 while (m != NULL && m->pindex < pindex) {
2118 m = TAILQ_NEXT(m, listq);
2121 return (pindex == end);
2125 * Routine: vm_object_coalesce
2126 * Function: Coalesces two objects backing up adjoining
2127 * regions of memory into a single object.
2129 * returns TRUE if objects were combined.
2131 * NOTE: Only works at the moment if the second object is NULL -
2132 * if it's not, which object do we lock first?
2135 * prev_object First object to coalesce
2136 * prev_offset Offset into prev_object
2137 * prev_size Size of reference to prev_object
2138 * next_size Size of reference to the second object
2139 * reserved Indicator that extension region has
2140 * swap accounted for
2143 * The object must *not* be locked.
2146 vm_object_coalesce(vm_object_t prev_object, vm_ooffset_t prev_offset,
2147 vm_size_t prev_size, vm_size_t next_size, boolean_t reserved)
2149 vm_pindex_t next_pindex;
2151 if (prev_object == NULL)
2153 VM_OBJECT_WLOCK(prev_object);
2154 if ((prev_object->type != OBJT_DEFAULT &&
2155 prev_object->type != OBJT_SWAP) ||
2156 (prev_object->flags & OBJ_TMPFS_NODE) != 0) {
2157 VM_OBJECT_WUNLOCK(prev_object);
2162 * Try to collapse the object first
2164 vm_object_collapse(prev_object);
2167 * Can't coalesce if: . more than one reference . paged out . shadows
2168 * another object . has a copy elsewhere (any of which mean that the
2169 * pages not mapped to prev_entry may be in use anyway)
2171 if (prev_object->backing_object != NULL) {
2172 VM_OBJECT_WUNLOCK(prev_object);
2176 prev_size >>= PAGE_SHIFT;
2177 next_size >>= PAGE_SHIFT;
2178 next_pindex = OFF_TO_IDX(prev_offset) + prev_size;
2180 if ((prev_object->ref_count > 1) &&
2181 (prev_object->size != next_pindex)) {
2182 VM_OBJECT_WUNLOCK(prev_object);
2187 * Account for the charge.
2189 if (prev_object->cred != NULL) {
2192 * If prev_object was charged, then this mapping,
2193 * although not charged now, may become writable
2194 * later. Non-NULL cred in the object would prevent
2195 * swap reservation during enabling of the write
2196 * access, so reserve swap now. Failed reservation
2197 * cause allocation of the separate object for the map
2198 * entry, and swap reservation for this entry is
2199 * managed in appropriate time.
2201 if (!reserved && !swap_reserve_by_cred(ptoa(next_size),
2202 prev_object->cred)) {
2203 VM_OBJECT_WUNLOCK(prev_object);
2206 prev_object->charge += ptoa(next_size);
2210 * Remove any pages that may still be in the object from a previous
2213 if (next_pindex < prev_object->size) {
2214 vm_object_page_remove(prev_object, next_pindex, next_pindex +
2216 if (prev_object->type == OBJT_SWAP)
2217 swap_pager_freespace(prev_object,
2218 next_pindex, next_size);
2220 if (prev_object->cred != NULL) {
2221 KASSERT(prev_object->charge >=
2222 ptoa(prev_object->size - next_pindex),
2223 ("object %p overcharged 1 %jx %jx", prev_object,
2224 (uintmax_t)next_pindex, (uintmax_t)next_size));
2225 prev_object->charge -= ptoa(prev_object->size -
2232 * Extend the object if necessary.
2234 if (next_pindex + next_size > prev_object->size)
2235 prev_object->size = next_pindex + next_size;
2237 VM_OBJECT_WUNLOCK(prev_object);
2242 vm_object_set_writeable_dirty(vm_object_t object)
2245 VM_OBJECT_ASSERT_WLOCKED(object);
2246 if (object->type != OBJT_VNODE) {
2247 if ((object->flags & OBJ_TMPFS_NODE) != 0) {
2248 KASSERT(object->type == OBJT_SWAP, ("non-swap tmpfs"));
2249 vm_object_set_flag(object, OBJ_TMPFS_DIRTY);
2253 object->generation++;
2254 if ((object->flags & OBJ_MIGHTBEDIRTY) != 0)
2256 vm_object_set_flag(object, OBJ_MIGHTBEDIRTY);
2262 * For each page offset within the specified range of the given object,
2263 * find the highest-level page in the shadow chain and unwire it. A page
2264 * must exist at every page offset, and the highest-level page must be
2268 vm_object_unwire(vm_object_t object, vm_ooffset_t offset, vm_size_t length,
2271 vm_object_t tobject, t1object;
2273 vm_pindex_t end_pindex, pindex, tpindex;
2274 int depth, locked_depth;
2276 KASSERT((offset & PAGE_MASK) == 0,
2277 ("vm_object_unwire: offset is not page aligned"));
2278 KASSERT((length & PAGE_MASK) == 0,
2279 ("vm_object_unwire: length is not a multiple of PAGE_SIZE"));
2280 /* The wired count of a fictitious page never changes. */
2281 if ((object->flags & OBJ_FICTITIOUS) != 0)
2283 pindex = OFF_TO_IDX(offset);
2284 end_pindex = pindex + atop(length);
2287 VM_OBJECT_RLOCK(object);
2288 m = vm_page_find_least(object, pindex);
2289 while (pindex < end_pindex) {
2290 if (m == NULL || pindex < m->pindex) {
2292 * The first object in the shadow chain doesn't
2293 * contain a page at the current index. Therefore,
2294 * the page must exist in a backing object.
2301 OFF_TO_IDX(tobject->backing_object_offset);
2302 tobject = tobject->backing_object;
2303 KASSERT(tobject != NULL,
2304 ("vm_object_unwire: missing page"));
2305 if ((tobject->flags & OBJ_FICTITIOUS) != 0)
2308 if (depth == locked_depth) {
2310 VM_OBJECT_RLOCK(tobject);
2312 } while ((tm = vm_page_lookup(tobject, tpindex)) ==
2316 m = TAILQ_NEXT(m, listq);
2319 if (vm_page_xbusied(tm)) {
2320 for (tobject = object; locked_depth >= 1;
2322 t1object = tobject->backing_object;
2323 VM_OBJECT_RUNLOCK(tobject);
2326 vm_page_busy_sleep(tm, "unwbo", true);
2329 vm_page_unwire(tm, queue);
2334 /* Release the accumulated object locks. */
2335 for (tobject = object; locked_depth >= 1; locked_depth--) {
2336 t1object = tobject->backing_object;
2337 VM_OBJECT_RUNLOCK(tobject);
2343 vm_object_vnode(vm_object_t object)
2346 VM_OBJECT_ASSERT_LOCKED(object);
2347 if (object->type == OBJT_VNODE)
2348 return (object->handle);
2349 if (object->type == OBJT_SWAP && (object->flags & OBJ_TMPFS) != 0)
2350 return (object->un_pager.swp.swp_tmpfs);
2355 sysctl_vm_object_list(SYSCTL_HANDLER_ARGS)
2357 struct kinfo_vmobject *kvo;
2358 char *fullpath, *freepath;
2365 if (req->oldptr == NULL) {
2367 * If an old buffer has not been provided, generate an
2368 * estimate of the space needed for a subsequent call.
2370 mtx_lock(&vm_object_list_mtx);
2372 TAILQ_FOREACH(obj, &vm_object_list, object_list) {
2373 if (obj->type == OBJT_DEAD)
2377 mtx_unlock(&vm_object_list_mtx);
2378 return (SYSCTL_OUT(req, NULL, sizeof(struct kinfo_vmobject) *
2382 kvo = malloc(sizeof(*kvo), M_TEMP, M_WAITOK);
2386 * VM objects are type stable and are never removed from the
2387 * list once added. This allows us to safely read obj->object_list
2388 * after reacquiring the VM object lock.
2390 mtx_lock(&vm_object_list_mtx);
2391 TAILQ_FOREACH(obj, &vm_object_list, object_list) {
2392 if (obj->type == OBJT_DEAD)
2394 VM_OBJECT_RLOCK(obj);
2395 if (obj->type == OBJT_DEAD) {
2396 VM_OBJECT_RUNLOCK(obj);
2399 mtx_unlock(&vm_object_list_mtx);
2400 kvo->kvo_size = ptoa(obj->size);
2401 kvo->kvo_resident = obj->resident_page_count;
2402 kvo->kvo_ref_count = obj->ref_count;
2403 kvo->kvo_shadow_count = obj->shadow_count;
2404 kvo->kvo_memattr = obj->memattr;
2405 kvo->kvo_active = 0;
2406 kvo->kvo_inactive = 0;
2407 TAILQ_FOREACH(m, &obj->memq, listq) {
2409 * A page may belong to the object but be
2410 * dequeued and set to PQ_NONE while the
2411 * object lock is not held. This makes the
2412 * reads of m->queue below racy, and we do not
2413 * count pages set to PQ_NONE. However, this
2414 * sysctl is only meant to give an
2415 * approximation of the system anyway.
2417 if (vm_page_active(m))
2419 else if (vm_page_inactive(m))
2420 kvo->kvo_inactive++;
2423 kvo->kvo_vn_fileid = 0;
2424 kvo->kvo_vn_fsid = 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;
2470 strlcpy(kvo->kvo_path, fullpath, sizeof(kvo->kvo_path));
2471 if (freepath != NULL)
2472 free(freepath, M_TEMP);
2474 /* Pack record size down */
2475 kvo->kvo_structsize = offsetof(struct kinfo_vmobject, kvo_path)
2476 + strlen(kvo->kvo_path) + 1;
2477 kvo->kvo_structsize = roundup(kvo->kvo_structsize,
2479 error = SYSCTL_OUT(req, kvo, kvo->kvo_structsize);
2480 mtx_lock(&vm_object_list_mtx);
2484 mtx_unlock(&vm_object_list_mtx);
2488 SYSCTL_PROC(_vm, OID_AUTO, objects, CTLTYPE_STRUCT | CTLFLAG_RW | CTLFLAG_SKIP |
2489 CTLFLAG_MPSAFE, NULL, 0, sysctl_vm_object_list, "S,kinfo_vmobject",
2490 "List of VM objects");
2492 #include "opt_ddb.h"
2494 #include <sys/kernel.h>
2496 #include <sys/cons.h>
2498 #include <ddb/ddb.h>
2501 _vm_object_in_map(vm_map_t map, vm_object_t object, vm_map_entry_t entry)
2504 vm_map_entry_t tmpe;
2512 tmpe = map->header.next;
2513 entcount = map->nentries;
2514 while (entcount-- && (tmpe != &map->header)) {
2515 if (_vm_object_in_map(map, object, tmpe)) {
2520 } else if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
2521 tmpm = entry->object.sub_map;
2522 tmpe = tmpm->header.next;
2523 entcount = tmpm->nentries;
2524 while (entcount-- && tmpe != &tmpm->header) {
2525 if (_vm_object_in_map(tmpm, object, tmpe)) {
2530 } else if ((obj = entry->object.vm_object) != NULL) {
2531 for (; obj; obj = obj->backing_object)
2532 if (obj == object) {
2540 vm_object_in_map(vm_object_t object)
2544 /* sx_slock(&allproc_lock); */
2545 FOREACH_PROC_IN_SYSTEM(p) {
2546 if (!p->p_vmspace /* || (p->p_flag & (P_SYSTEM|P_WEXIT)) */)
2548 if (_vm_object_in_map(&p->p_vmspace->vm_map, object, 0)) {
2549 /* sx_sunlock(&allproc_lock); */
2553 /* sx_sunlock(&allproc_lock); */
2554 if (_vm_object_in_map(kernel_map, object, 0))
2559 DB_SHOW_COMMAND(vmochk, vm_object_check)
2564 * make sure that internal objs are in a map somewhere
2565 * and none have zero ref counts.
2567 TAILQ_FOREACH(object, &vm_object_list, object_list) {
2568 if (object->handle == NULL &&
2569 (object->type == OBJT_DEFAULT || object->type == OBJT_SWAP)) {
2570 if (object->ref_count == 0) {
2571 db_printf("vmochk: internal obj has zero ref count: %ld\n",
2572 (long)object->size);
2574 if (!vm_object_in_map(object)) {
2576 "vmochk: internal obj is not in a map: "
2577 "ref: %d, size: %lu: 0x%lx, backing_object: %p\n",
2578 object->ref_count, (u_long)object->size,
2579 (u_long)object->size,
2580 (void *)object->backing_object);
2587 * vm_object_print: [ debug ]
2589 DB_SHOW_COMMAND(object, vm_object_print_static)
2591 /* XXX convert args. */
2592 vm_object_t object = (vm_object_t)addr;
2593 boolean_t full = have_addr;
2597 /* XXX count is an (unused) arg. Avoid shadowing it. */
2598 #define count was_count
2606 "Object %p: type=%d, size=0x%jx, res=%d, ref=%d, flags=0x%x ruid %d charge %jx\n",
2607 object, (int)object->type, (uintmax_t)object->size,
2608 object->resident_page_count, object->ref_count, object->flags,
2609 object->cred ? object->cred->cr_ruid : -1, (uintmax_t)object->charge);
2610 db_iprintf(" sref=%d, backing_object(%d)=(%p)+0x%jx\n",
2611 object->shadow_count,
2612 object->backing_object ? object->backing_object->ref_count : 0,
2613 object->backing_object, (uintmax_t)object->backing_object_offset);
2620 TAILQ_FOREACH(p, &object->memq, listq) {
2622 db_iprintf("memory:=");
2623 else if (count == 6) {
2631 db_printf("(off=0x%jx,page=0x%jx)",
2632 (uintmax_t)p->pindex, (uintmax_t)VM_PAGE_TO_PHYS(p));
2642 /* XXX need this non-static entry for calling from vm_map_print. */
2645 /* db_expr_t */ long addr,
2646 boolean_t have_addr,
2647 /* db_expr_t */ long count,
2650 vm_object_print_static(addr, have_addr, count, modif);
2653 DB_SHOW_COMMAND(vmopag, vm_object_print_pages)
2658 vm_page_t m, prev_m;
2662 TAILQ_FOREACH(object, &vm_object_list, object_list) {
2663 db_printf("new object: %p\n", (void *)object);
2674 TAILQ_FOREACH(m, &object->memq, listq) {
2675 if (m->pindex > 128)
2677 if ((prev_m = TAILQ_PREV(m, pglist, listq)) != NULL &&
2678 prev_m->pindex + 1 != m->pindex) {
2680 db_printf(" index(%ld)run(%d)pa(0x%lx)\n",
2681 (long)fidx, rcount, (long)pa);
2693 (VM_PAGE_TO_PHYS(m) == pa + rcount * PAGE_SIZE)) {
2698 db_printf(" index(%ld)run(%d)pa(0x%lx)\n",
2699 (long)fidx, rcount, (long)pa);
2709 pa = VM_PAGE_TO_PHYS(m);
2713 db_printf(" index(%ld)run(%d)pa(0x%lx)\n",
2714 (long)fidx, rcount, (long)pa);