2 * SPDX-License-Identifier: (BSD-3-Clause AND MIT-CMU)
4 * Copyright (c) 1991, 1993
5 * The Regents of the University of California. All rights reserved.
7 * This code is derived from software contributed to Berkeley by
8 * The Mach Operating System project at Carnegie-Mellon University.
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 3. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * from: @(#)vm_object.c 8.5 (Berkeley) 3/22/94
37 * Copyright (c) 1987, 1990 Carnegie-Mellon University.
38 * All rights reserved.
40 * Authors: Avadis Tevanian, Jr., Michael Wayne Young
42 * Permission to use, copy, modify and distribute this software and
43 * its documentation is hereby granted, provided that both the copyright
44 * notice and this permission notice appear in all copies of the
45 * software, derivative works or modified versions, and any portions
46 * thereof, and that both notices appear in supporting documentation.
48 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
49 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
50 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
52 * Carnegie Mellon requests users of this software to return to
54 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
55 * School of Computer Science
56 * Carnegie Mellon University
57 * Pittsburgh PA 15213-3890
59 * any improvements or extensions that they make and grant Carnegie the
60 * rights to redistribute these changes.
64 * Virtual memory object module.
67 #include <sys/cdefs.h>
68 __FBSDID("$FreeBSD$");
72 #include <sys/param.h>
73 #include <sys/systm.h>
74 #include <sys/cpuset.h>
77 #include <sys/mount.h>
78 #include <sys/kernel.h>
79 #include <sys/pctrie.h>
80 #include <sys/sysctl.h>
81 #include <sys/mutex.h>
82 #include <sys/proc.h> /* for curproc, pageproc */
83 #include <sys/socket.h>
84 #include <sys/resourcevar.h>
85 #include <sys/rwlock.h>
87 #include <sys/vnode.h>
88 #include <sys/vmmeter.h>
92 #include <vm/vm_param.h>
94 #include <vm/vm_map.h>
95 #include <vm/vm_object.h>
96 #include <vm/vm_page.h>
97 #include <vm/vm_pageout.h>
98 #include <vm/vm_pager.h>
99 #include <vm/vm_phys.h>
100 #include <vm/vm_pagequeue.h>
101 #include <vm/swap_pager.h>
102 #include <vm/vm_kern.h>
103 #include <vm/vm_extern.h>
104 #include <vm/vm_radix.h>
105 #include <vm/vm_reserv.h>
108 static int old_msync;
109 SYSCTL_INT(_vm, OID_AUTO, old_msync, CTLFLAG_RW, &old_msync, 0,
110 "Use old (insecure) msync behavior");
112 static int vm_object_page_collect_flush(vm_object_t object, vm_page_t p,
113 int pagerflags, int flags, boolean_t *clearobjflags,
115 static boolean_t vm_object_page_remove_write(vm_page_t p, int flags,
116 boolean_t *clearobjflags);
117 static void vm_object_qcollapse(vm_object_t object);
118 static void vm_object_vndeallocate(vm_object_t object);
121 * Virtual memory objects maintain the actual data
122 * associated with allocated virtual memory. A given
123 * page of memory exists within exactly one object.
125 * An object is only deallocated when all "references"
126 * are given up. Only one "reference" to a given
127 * region of an object should be writeable.
129 * Associated with each object is a list of all resident
130 * memory pages belonging to that object; this list is
131 * maintained by the "vm_page" module, and locked by the object's
134 * Each object also records a "pager" routine which is
135 * used to retrieve (and store) pages to the proper backing
136 * storage. In addition, objects may be backed by other
137 * objects from which they were virtual-copied.
139 * The only items within the object structure which are
140 * modified after time of creation are:
141 * reference count locked by object's lock
142 * pager routine locked by object's lock
146 struct object_q vm_object_list;
147 struct mtx vm_object_list_mtx; /* lock for object list and count */
149 struct vm_object kernel_object_store;
151 static SYSCTL_NODE(_vm_stats, OID_AUTO, object, CTLFLAG_RD, 0,
154 static counter_u64_t object_collapses = EARLY_COUNTER;
155 SYSCTL_COUNTER_U64(_vm_stats_object, OID_AUTO, collapses, CTLFLAG_RD,
157 "VM object collapses");
159 static counter_u64_t object_bypasses = EARLY_COUNTER;
160 SYSCTL_COUNTER_U64(_vm_stats_object, OID_AUTO, bypasses, CTLFLAG_RD,
162 "VM object bypasses");
165 counter_startup(void)
168 object_collapses = counter_u64_alloc(M_WAITOK);
169 object_bypasses = counter_u64_alloc(M_WAITOK);
171 SYSINIT(object_counters, SI_SUB_CPU, SI_ORDER_ANY, counter_startup, NULL);
173 static uma_zone_t obj_zone;
175 static int vm_object_zinit(void *mem, int size, int flags);
178 static void vm_object_zdtor(void *mem, int size, void *arg);
181 vm_object_zdtor(void *mem, int size, void *arg)
185 object = (vm_object_t)mem;
186 KASSERT(object->ref_count == 0,
187 ("object %p ref_count = %d", object, object->ref_count));
188 KASSERT(TAILQ_EMPTY(&object->memq),
189 ("object %p has resident pages in its memq", object));
190 KASSERT(vm_radix_is_empty(&object->rtree),
191 ("object %p has resident pages in its trie", object));
192 #if VM_NRESERVLEVEL > 0
193 KASSERT(LIST_EMPTY(&object->rvq),
194 ("object %p has reservations",
197 KASSERT(object->paging_in_progress == 0,
198 ("object %p paging_in_progress = %d",
199 object, object->paging_in_progress));
200 KASSERT(object->resident_page_count == 0,
201 ("object %p resident_page_count = %d",
202 object, object->resident_page_count));
203 KASSERT(object->shadow_count == 0,
204 ("object %p shadow_count = %d",
205 object, object->shadow_count));
206 KASSERT(object->type == OBJT_DEAD,
207 ("object %p has non-dead type %d",
208 object, object->type));
213 vm_object_zinit(void *mem, int size, int flags)
217 object = (vm_object_t)mem;
218 rw_init_flags(&object->lock, "vm object", RW_DUPOK | RW_NEW);
220 /* These are true for any object that has been freed */
221 object->type = OBJT_DEAD;
222 object->ref_count = 0;
223 vm_radix_init(&object->rtree);
224 object->paging_in_progress = 0;
225 object->resident_page_count = 0;
226 object->shadow_count = 0;
227 object->flags = OBJ_DEAD;
229 mtx_lock(&vm_object_list_mtx);
230 TAILQ_INSERT_TAIL(&vm_object_list, object, object_list);
231 mtx_unlock(&vm_object_list_mtx);
236 _vm_object_allocate(objtype_t type, vm_pindex_t size, vm_object_t object)
239 TAILQ_INIT(&object->memq);
240 LIST_INIT(&object->shadow_head);
243 if (type == OBJT_SWAP)
244 pctrie_init(&object->un_pager.swp.swp_blks);
247 * Ensure that swap_pager_swapoff() iteration over object_list
248 * sees up to date type and pctrie head if it observed
251 atomic_thread_fence_rel();
255 panic("_vm_object_allocate: can't create OBJT_DEAD");
258 object->flags = OBJ_ONEMAPPING;
262 object->flags = OBJ_FICTITIOUS | OBJ_UNMANAGED;
265 object->flags = OBJ_FICTITIOUS;
268 object->flags = OBJ_UNMANAGED;
274 panic("_vm_object_allocate: type %d is undefined", type);
277 object->generation = 1;
278 object->ref_count = 1;
279 object->memattr = VM_MEMATTR_DEFAULT;
282 object->handle = NULL;
283 object->backing_object = NULL;
284 object->backing_object_offset = (vm_ooffset_t) 0;
285 #if VM_NRESERVLEVEL > 0
286 LIST_INIT(&object->rvq);
288 umtx_shm_object_init(object);
294 * Initialize the VM objects module.
299 TAILQ_INIT(&vm_object_list);
300 mtx_init(&vm_object_list_mtx, "vm object_list", NULL, MTX_DEF);
302 rw_init(&kernel_object->lock, "kernel vm object");
303 _vm_object_allocate(OBJT_PHYS, atop(VM_MAX_KERNEL_ADDRESS -
304 VM_MIN_KERNEL_ADDRESS), kernel_object);
305 #if VM_NRESERVLEVEL > 0
306 kernel_object->flags |= OBJ_COLORED;
307 kernel_object->pg_color = (u_short)atop(VM_MIN_KERNEL_ADDRESS);
311 * The lock portion of struct vm_object must be type stable due
312 * to vm_pageout_fallback_object_lock locking a vm object
313 * without holding any references to it.
315 obj_zone = uma_zcreate("VM OBJECT", sizeof (struct vm_object), NULL,
321 vm_object_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
327 vm_object_clear_flag(vm_object_t object, u_short bits)
330 VM_OBJECT_ASSERT_WLOCKED(object);
331 object->flags &= ~bits;
335 * Sets the default memory attribute for the specified object. Pages
336 * that are allocated to this object are by default assigned this memory
339 * Presently, this function must be called before any pages are allocated
340 * to the object. In the future, this requirement may be relaxed for
341 * "default" and "swap" objects.
344 vm_object_set_memattr(vm_object_t object, vm_memattr_t memattr)
347 VM_OBJECT_ASSERT_WLOCKED(object);
348 switch (object->type) {
356 if (!TAILQ_EMPTY(&object->memq))
357 return (KERN_FAILURE);
360 return (KERN_INVALID_ARGUMENT);
362 panic("vm_object_set_memattr: object %p is of undefined type",
365 object->memattr = memattr;
366 return (KERN_SUCCESS);
370 vm_object_pip_add(vm_object_t object, short i)
373 VM_OBJECT_ASSERT_WLOCKED(object);
374 object->paging_in_progress += i;
378 vm_object_pip_subtract(vm_object_t object, short i)
381 VM_OBJECT_ASSERT_WLOCKED(object);
382 object->paging_in_progress -= i;
386 vm_object_pip_wakeup(vm_object_t object)
389 VM_OBJECT_ASSERT_WLOCKED(object);
390 object->paging_in_progress--;
391 if ((object->flags & OBJ_PIPWNT) && object->paging_in_progress == 0) {
392 vm_object_clear_flag(object, OBJ_PIPWNT);
398 vm_object_pip_wakeupn(vm_object_t object, short i)
401 VM_OBJECT_ASSERT_WLOCKED(object);
403 object->paging_in_progress -= i;
404 if ((object->flags & OBJ_PIPWNT) && object->paging_in_progress == 0) {
405 vm_object_clear_flag(object, OBJ_PIPWNT);
411 vm_object_pip_wait(vm_object_t object, char *waitid)
414 VM_OBJECT_ASSERT_WLOCKED(object);
415 while (object->paging_in_progress) {
416 object->flags |= OBJ_PIPWNT;
417 VM_OBJECT_SLEEP(object, object, PVM, waitid, 0);
422 * vm_object_allocate:
424 * Returns a new object with the given size.
427 vm_object_allocate(objtype_t type, vm_pindex_t size)
431 object = (vm_object_t)uma_zalloc(obj_zone, M_WAITOK);
432 _vm_object_allocate(type, size, object);
438 * vm_object_reference:
440 * Gets another reference to the given object. Note: OBJ_DEAD
441 * objects can be referenced during final cleaning.
444 vm_object_reference(vm_object_t object)
448 VM_OBJECT_WLOCK(object);
449 vm_object_reference_locked(object);
450 VM_OBJECT_WUNLOCK(object);
454 * vm_object_reference_locked:
456 * Gets another reference to the given object.
458 * The object must be locked.
461 vm_object_reference_locked(vm_object_t object)
465 VM_OBJECT_ASSERT_WLOCKED(object);
467 if (object->type == OBJT_VNODE) {
474 * Handle deallocating an object of type OBJT_VNODE.
477 vm_object_vndeallocate(vm_object_t object)
479 struct vnode *vp = (struct vnode *) object->handle;
481 VM_OBJECT_ASSERT_WLOCKED(object);
482 KASSERT(object->type == OBJT_VNODE,
483 ("vm_object_vndeallocate: not a vnode object"));
484 KASSERT(vp != NULL, ("vm_object_vndeallocate: missing vp"));
486 if (object->ref_count == 0) {
487 vn_printf(vp, "vm_object_vndeallocate ");
488 panic("vm_object_vndeallocate: bad object reference count");
492 if (!umtx_shm_vnobj_persistent && object->ref_count == 1)
493 umtx_shm_object_terminated(object);
496 * The test for text of vp vnode does not need a bypass to
497 * reach right VV_TEXT there, since it is obtained from
500 if (object->ref_count > 1 || (vp->v_vflag & VV_TEXT) == 0) {
502 VM_OBJECT_WUNLOCK(object);
503 /* vrele may need the vnode lock. */
507 VM_OBJECT_WUNLOCK(object);
508 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
510 VM_OBJECT_WLOCK(object);
512 if (object->type == OBJT_DEAD) {
513 VM_OBJECT_WUNLOCK(object);
516 if (object->ref_count == 0)
518 VM_OBJECT_WUNLOCK(object);
525 * vm_object_deallocate:
527 * Release a reference to the specified object,
528 * gained either through a vm_object_allocate
529 * or a vm_object_reference call. When all references
530 * are gone, storage associated with this object
531 * may be relinquished.
533 * No object may be locked.
536 vm_object_deallocate(vm_object_t object)
541 while (object != NULL) {
542 VM_OBJECT_WLOCK(object);
543 if (object->type == OBJT_VNODE) {
544 vm_object_vndeallocate(object);
548 KASSERT(object->ref_count != 0,
549 ("vm_object_deallocate: object deallocated too many times: %d", object->type));
552 * If the reference count goes to 0 we start calling
553 * vm_object_terminate() on the object chain.
554 * A ref count of 1 may be a special case depending on the
555 * shadow count being 0 or 1.
558 if (object->ref_count > 1) {
559 VM_OBJECT_WUNLOCK(object);
561 } else if (object->ref_count == 1) {
562 if (object->type == OBJT_SWAP &&
563 (object->flags & OBJ_TMPFS) != 0) {
564 vp = object->un_pager.swp.swp_tmpfs;
566 VM_OBJECT_WUNLOCK(object);
567 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
568 VM_OBJECT_WLOCK(object);
569 if (object->type == OBJT_DEAD ||
570 object->ref_count != 1) {
571 VM_OBJECT_WUNLOCK(object);
576 if ((object->flags & OBJ_TMPFS) != 0)
581 if (object->shadow_count == 0 &&
582 object->handle == NULL &&
583 (object->type == OBJT_DEFAULT ||
584 (object->type == OBJT_SWAP &&
585 (object->flags & OBJ_TMPFS_NODE) == 0))) {
586 vm_object_set_flag(object, OBJ_ONEMAPPING);
587 } else if ((object->shadow_count == 1) &&
588 (object->handle == NULL) &&
589 (object->type == OBJT_DEFAULT ||
590 object->type == OBJT_SWAP)) {
593 robject = LIST_FIRST(&object->shadow_head);
594 KASSERT(robject != NULL,
595 ("vm_object_deallocate: ref_count: %d, shadow_count: %d",
597 object->shadow_count));
598 KASSERT((robject->flags & OBJ_TMPFS_NODE) == 0,
599 ("shadowed tmpfs v_object %p", object));
600 if (!VM_OBJECT_TRYWLOCK(robject)) {
602 * Avoid a potential deadlock.
605 VM_OBJECT_WUNLOCK(object);
607 * More likely than not the thread
608 * holding robject's lock has lower
609 * priority than the current thread.
610 * Let the lower priority thread run.
616 * Collapse object into its shadow unless its
617 * shadow is dead. In that case, object will
618 * be deallocated by the thread that is
619 * deallocating its shadow.
621 if ((robject->flags & OBJ_DEAD) == 0 &&
622 (robject->handle == NULL) &&
623 (robject->type == OBJT_DEFAULT ||
624 robject->type == OBJT_SWAP)) {
626 robject->ref_count++;
628 if (robject->paging_in_progress) {
629 VM_OBJECT_WUNLOCK(object);
630 vm_object_pip_wait(robject,
632 temp = robject->backing_object;
633 if (object == temp) {
634 VM_OBJECT_WLOCK(object);
637 } else if (object->paging_in_progress) {
638 VM_OBJECT_WUNLOCK(robject);
639 object->flags |= OBJ_PIPWNT;
640 VM_OBJECT_SLEEP(object, object,
641 PDROP | PVM, "objde2", 0);
642 VM_OBJECT_WLOCK(robject);
643 temp = robject->backing_object;
644 if (object == temp) {
645 VM_OBJECT_WLOCK(object);
649 VM_OBJECT_WUNLOCK(object);
651 if (robject->ref_count == 1) {
652 robject->ref_count--;
657 vm_object_collapse(object);
658 VM_OBJECT_WUNLOCK(object);
661 VM_OBJECT_WUNLOCK(robject);
663 VM_OBJECT_WUNLOCK(object);
667 umtx_shm_object_terminated(object);
668 temp = object->backing_object;
670 KASSERT((object->flags & OBJ_TMPFS_NODE) == 0,
671 ("shadowed tmpfs v_object 2 %p", object));
672 VM_OBJECT_WLOCK(temp);
673 LIST_REMOVE(object, shadow_list);
674 temp->shadow_count--;
675 VM_OBJECT_WUNLOCK(temp);
676 object->backing_object = NULL;
679 * Don't double-terminate, we could be in a termination
680 * recursion due to the terminate having to sync data
683 if ((object->flags & OBJ_DEAD) == 0)
684 vm_object_terminate(object);
686 VM_OBJECT_WUNLOCK(object);
692 * vm_object_destroy removes the object from the global object list
693 * and frees the space for the object.
696 vm_object_destroy(vm_object_t object)
700 * Release the allocation charge.
702 if (object->cred != NULL) {
703 swap_release_by_cred(object->charge, object->cred);
705 crfree(object->cred);
710 * Free the space for the object.
712 uma_zfree(obj_zone, object);
716 * vm_object_terminate_pages removes any remaining pageable pages
717 * from the object and resets the object to an empty state.
720 vm_object_terminate_pages(vm_object_t object)
725 VM_OBJECT_ASSERT_WLOCKED(object);
730 * Free any remaining pageable pages. This also removes them from the
731 * paging queues. However, don't free wired pages, just remove them
732 * from the object. Rather than incrementally removing each page from
733 * the object, the page and object are reset to any empty state.
735 TAILQ_FOREACH_SAFE(p, &object->memq, listq, p_next) {
736 vm_page_assert_unbusied(p);
737 if ((object->flags & OBJ_UNMANAGED) == 0)
739 * vm_page_free_prep() only needs the page
740 * lock for managed pages.
742 vm_page_change_lock(p, &mtx);
744 if (p->wire_count != 0)
753 * If the object contained any pages, then reset it to an empty state.
754 * None of the object's fields, including "resident_page_count", were
755 * modified by the preceding loop.
757 if (object->resident_page_count != 0) {
758 vm_radix_reclaim_allnodes(&object->rtree);
759 TAILQ_INIT(&object->memq);
760 object->resident_page_count = 0;
761 if (object->type == OBJT_VNODE)
762 vdrop(object->handle);
767 * vm_object_terminate actually destroys the specified object, freeing
768 * up all previously used resources.
770 * The object must be locked.
771 * This routine may block.
774 vm_object_terminate(vm_object_t object)
777 VM_OBJECT_ASSERT_WLOCKED(object);
780 * Make sure no one uses us.
782 vm_object_set_flag(object, OBJ_DEAD);
785 * wait for the pageout daemon to be done with the object
787 vm_object_pip_wait(object, "objtrm");
789 KASSERT(!object->paging_in_progress,
790 ("vm_object_terminate: pageout in progress"));
793 * Clean and free the pages, as appropriate. All references to the
794 * object are gone, so we don't need to lock it.
796 if (object->type == OBJT_VNODE) {
797 struct vnode *vp = (struct vnode *)object->handle;
800 * Clean pages and flush buffers.
802 vm_object_page_clean(object, 0, 0, OBJPC_SYNC);
803 VM_OBJECT_WUNLOCK(object);
805 vinvalbuf(vp, V_SAVE, 0, 0);
807 BO_LOCK(&vp->v_bufobj);
808 vp->v_bufobj.bo_flag |= BO_DEAD;
809 BO_UNLOCK(&vp->v_bufobj);
811 VM_OBJECT_WLOCK(object);
814 KASSERT(object->ref_count == 0,
815 ("vm_object_terminate: object with references, ref_count=%d",
818 if ((object->flags & OBJ_PG_DTOR) == 0)
819 vm_object_terminate_pages(object);
821 #if VM_NRESERVLEVEL > 0
822 if (__predict_false(!LIST_EMPTY(&object->rvq)))
823 vm_reserv_break_all(object);
826 KASSERT(object->cred == NULL || object->type == OBJT_DEFAULT ||
827 object->type == OBJT_SWAP,
828 ("%s: non-swap obj %p has cred", __func__, object));
831 * Let the pager know object is dead.
833 vm_pager_deallocate(object);
834 VM_OBJECT_WUNLOCK(object);
836 vm_object_destroy(object);
840 * Make the page read-only so that we can clear the object flags. However, if
841 * this is a nosync mmap then the object is likely to stay dirty so do not
842 * mess with the page and do not clear the object flags. Returns TRUE if the
843 * page should be flushed, and FALSE otherwise.
846 vm_object_page_remove_write(vm_page_t p, int flags, boolean_t *clearobjflags)
850 * If we have been asked to skip nosync pages and this is a
851 * nosync page, skip it. Note that the object flags were not
852 * cleared in this case so we do not have to set them.
854 if ((flags & OBJPC_NOSYNC) != 0 && (p->oflags & VPO_NOSYNC) != 0) {
855 *clearobjflags = FALSE;
858 pmap_remove_write(p);
859 return (p->dirty != 0);
864 * vm_object_page_clean
866 * Clean all dirty pages in the specified range of object. Leaves page
867 * on whatever queue it is currently on. If NOSYNC is set then do not
868 * write out pages with VPO_NOSYNC set (originally comes from MAP_NOSYNC),
869 * leaving the object dirty.
871 * When stuffing pages asynchronously, allow clustering. XXX we need a
872 * synchronous clustering mode implementation.
874 * Odd semantics: if start == end, we clean everything.
876 * The object must be locked.
878 * Returns FALSE if some page from the range was not written, as
879 * reported by the pager, and TRUE otherwise.
882 vm_object_page_clean(vm_object_t object, vm_ooffset_t start, vm_ooffset_t end,
886 vm_pindex_t pi, tend, tstart;
887 int curgeneration, n, pagerflags;
888 boolean_t clearobjflags, eio, res;
890 VM_OBJECT_ASSERT_WLOCKED(object);
893 * The OBJ_MIGHTBEDIRTY flag is only set for OBJT_VNODE
894 * objects. The check below prevents the function from
895 * operating on non-vnode objects.
897 if ((object->flags & OBJ_MIGHTBEDIRTY) == 0 ||
898 object->resident_page_count == 0)
901 pagerflags = (flags & (OBJPC_SYNC | OBJPC_INVAL)) != 0 ?
902 VM_PAGER_PUT_SYNC : VM_PAGER_CLUSTER_OK;
903 pagerflags |= (flags & OBJPC_INVAL) != 0 ? VM_PAGER_PUT_INVAL : 0;
905 tstart = OFF_TO_IDX(start);
906 tend = (end == 0) ? object->size : OFF_TO_IDX(end + PAGE_MASK);
907 clearobjflags = tstart == 0 && tend >= object->size;
911 curgeneration = object->generation;
913 for (p = vm_page_find_least(object, tstart); p != NULL; p = np) {
917 np = TAILQ_NEXT(p, listq);
920 if (vm_page_sleep_if_busy(p, "vpcwai")) {
921 if (object->generation != curgeneration) {
922 if ((flags & OBJPC_SYNC) != 0)
925 clearobjflags = FALSE;
927 np = vm_page_find_least(object, pi);
930 if (!vm_object_page_remove_write(p, flags, &clearobjflags))
933 n = vm_object_page_collect_flush(object, p, pagerflags,
934 flags, &clearobjflags, &eio);
937 clearobjflags = FALSE;
939 if (object->generation != curgeneration) {
940 if ((flags & OBJPC_SYNC) != 0)
943 clearobjflags = FALSE;
947 * If the VOP_PUTPAGES() did a truncated write, so
948 * that even the first page of the run is not fully
949 * written, vm_pageout_flush() returns 0 as the run
950 * length. Since the condition that caused truncated
951 * write may be permanent, e.g. exhausted free space,
952 * accepting n == 0 would cause an infinite loop.
954 * Forwarding the iterator leaves the unwritten page
955 * behind, but there is not much we can do there if
956 * filesystem refuses to write it.
960 clearobjflags = FALSE;
962 np = vm_page_find_least(object, pi + n);
965 VOP_FSYNC(vp, (pagerflags & VM_PAGER_PUT_SYNC) ? MNT_WAIT : 0);
969 vm_object_clear_flag(object, OBJ_MIGHTBEDIRTY);
974 vm_object_page_collect_flush(vm_object_t object, vm_page_t p, int pagerflags,
975 int flags, boolean_t *clearobjflags, boolean_t *eio)
977 vm_page_t ma[vm_pageout_page_count], p_first, tp;
978 int count, i, mreq, runlen;
980 vm_page_lock_assert(p, MA_NOTOWNED);
981 VM_OBJECT_ASSERT_WLOCKED(object);
986 for (tp = p; count < vm_pageout_page_count; count++) {
987 tp = vm_page_next(tp);
988 if (tp == NULL || vm_page_busied(tp))
990 if (!vm_object_page_remove_write(tp, flags, clearobjflags))
994 for (p_first = p; count < vm_pageout_page_count; count++) {
995 tp = vm_page_prev(p_first);
996 if (tp == NULL || vm_page_busied(tp))
998 if (!vm_object_page_remove_write(tp, flags, clearobjflags))
1004 for (tp = p_first, i = 0; i < count; tp = TAILQ_NEXT(tp, listq), i++)
1007 vm_pageout_flush(ma, count, pagerflags, mreq, &runlen, eio);
1012 * Note that there is absolutely no sense in writing out
1013 * anonymous objects, so we track down the vnode object
1015 * We invalidate (remove) all pages from the address space
1016 * for semantic correctness.
1018 * If the backing object is a device object with unmanaged pages, then any
1019 * mappings to the specified range of pages must be removed before this
1020 * function is called.
1022 * Note: certain anonymous maps, such as MAP_NOSYNC maps,
1023 * may start out with a NULL object.
1026 vm_object_sync(vm_object_t object, vm_ooffset_t offset, vm_size_t size,
1027 boolean_t syncio, boolean_t invalidate)
1029 vm_object_t backing_object;
1032 int error, flags, fsync_after;
1039 VM_OBJECT_WLOCK(object);
1040 while ((backing_object = object->backing_object) != NULL) {
1041 VM_OBJECT_WLOCK(backing_object);
1042 offset += object->backing_object_offset;
1043 VM_OBJECT_WUNLOCK(object);
1044 object = backing_object;
1045 if (object->size < OFF_TO_IDX(offset + size))
1046 size = IDX_TO_OFF(object->size) - offset;
1049 * Flush pages if writing is allowed, invalidate them
1050 * if invalidation requested. Pages undergoing I/O
1051 * will be ignored by vm_object_page_remove().
1053 * We cannot lock the vnode and then wait for paging
1054 * to complete without deadlocking against vm_fault.
1055 * Instead we simply call vm_object_page_remove() and
1056 * allow it to block internally on a page-by-page
1057 * basis when it encounters pages undergoing async
1060 if (object->type == OBJT_VNODE &&
1061 (object->flags & OBJ_MIGHTBEDIRTY) != 0 &&
1062 ((vp = object->handle)->v_vflag & VV_NOSYNC) == 0) {
1063 VM_OBJECT_WUNLOCK(object);
1064 (void) vn_start_write(vp, &mp, V_WAIT);
1065 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1066 if (syncio && !invalidate && offset == 0 &&
1067 atop(size) == object->size) {
1069 * If syncing the whole mapping of the file,
1070 * it is faster to schedule all the writes in
1071 * async mode, also allowing the clustering,
1072 * and then wait for i/o to complete.
1077 flags = (syncio || invalidate) ? OBJPC_SYNC : 0;
1078 flags |= invalidate ? (OBJPC_SYNC | OBJPC_INVAL) : 0;
1079 fsync_after = FALSE;
1081 VM_OBJECT_WLOCK(object);
1082 res = vm_object_page_clean(object, offset, offset + size,
1084 VM_OBJECT_WUNLOCK(object);
1086 error = VOP_FSYNC(vp, MNT_WAIT, curthread);
1088 vn_finished_write(mp);
1091 VM_OBJECT_WLOCK(object);
1093 if ((object->type == OBJT_VNODE ||
1094 object->type == OBJT_DEVICE) && invalidate) {
1095 if (object->type == OBJT_DEVICE)
1097 * The option OBJPR_NOTMAPPED must be passed here
1098 * because vm_object_page_remove() cannot remove
1099 * unmanaged mappings.
1101 flags = OBJPR_NOTMAPPED;
1105 flags = OBJPR_CLEANONLY;
1106 vm_object_page_remove(object, OFF_TO_IDX(offset),
1107 OFF_TO_IDX(offset + size + PAGE_MASK), flags);
1109 VM_OBJECT_WUNLOCK(object);
1114 * Determine whether the given advice can be applied to the object. Advice is
1115 * not applied to unmanaged pages since they never belong to page queues, and
1116 * since MADV_FREE is destructive, it can apply only to anonymous pages that
1117 * have been mapped at most once.
1120 vm_object_advice_applies(vm_object_t object, int advice)
1123 if ((object->flags & OBJ_UNMANAGED) != 0)
1125 if (advice != MADV_FREE)
1127 return ((object->type == OBJT_DEFAULT || object->type == OBJT_SWAP) &&
1128 (object->flags & OBJ_ONEMAPPING) != 0);
1132 vm_object_madvise_freespace(vm_object_t object, int advice, vm_pindex_t pindex,
1136 if (advice == MADV_FREE && object->type == OBJT_SWAP)
1137 swap_pager_freespace(object, pindex, size);
1141 * vm_object_madvise:
1143 * Implements the madvise function at the object/page level.
1145 * MADV_WILLNEED (any object)
1147 * Activate the specified pages if they are resident.
1149 * MADV_DONTNEED (any object)
1151 * Deactivate the specified pages if they are resident.
1153 * MADV_FREE (OBJT_DEFAULT/OBJT_SWAP objects,
1154 * OBJ_ONEMAPPING only)
1156 * Deactivate and clean the specified pages if they are
1157 * resident. This permits the process to reuse the pages
1158 * without faulting or the kernel to reclaim the pages
1162 vm_object_madvise(vm_object_t object, vm_pindex_t pindex, vm_pindex_t end,
1165 vm_pindex_t tpindex;
1166 vm_object_t backing_object, tobject;
1173 VM_OBJECT_WLOCK(object);
1174 if (!vm_object_advice_applies(object, advice)) {
1175 VM_OBJECT_WUNLOCK(object);
1178 for (m = vm_page_find_least(object, pindex); pindex < end; pindex++) {
1182 * If the next page isn't resident in the top-level object, we
1183 * need to search the shadow chain. When applying MADV_FREE, we
1184 * take care to release any swap space used to store
1185 * non-resident pages.
1187 if (m == NULL || pindex < m->pindex) {
1189 * Optimize a common case: if the top-level object has
1190 * no backing object, we can skip over the non-resident
1191 * range in constant time.
1193 if (object->backing_object == NULL) {
1194 tpindex = (m != NULL && m->pindex < end) ?
1196 vm_object_madvise_freespace(object, advice,
1197 pindex, tpindex - pindex);
1198 if ((pindex = tpindex) == end)
1205 vm_object_madvise_freespace(tobject, advice,
1208 * Prepare to search the next object in the
1211 backing_object = tobject->backing_object;
1212 if (backing_object == NULL)
1214 VM_OBJECT_WLOCK(backing_object);
1216 OFF_TO_IDX(tobject->backing_object_offset);
1217 if (tobject != object)
1218 VM_OBJECT_WUNLOCK(tobject);
1219 tobject = backing_object;
1220 if (!vm_object_advice_applies(tobject, advice))
1222 } while ((tm = vm_page_lookup(tobject, tpindex)) ==
1227 m = TAILQ_NEXT(m, listq);
1231 * If the page is not in a normal state, skip it.
1233 if (tm->valid != VM_PAGE_BITS_ALL)
1236 if (vm_page_held(tm)) {
1240 KASSERT((tm->flags & PG_FICTITIOUS) == 0,
1241 ("vm_object_madvise: page %p is fictitious", tm));
1242 KASSERT((tm->oflags & VPO_UNMANAGED) == 0,
1243 ("vm_object_madvise: page %p is not managed", tm));
1244 if (vm_page_busied(tm)) {
1245 if (object != tobject)
1246 VM_OBJECT_WUNLOCK(tobject);
1247 VM_OBJECT_WUNLOCK(object);
1248 if (advice == MADV_WILLNEED) {
1250 * Reference the page before unlocking and
1251 * sleeping so that the page daemon is less
1252 * likely to reclaim it.
1254 vm_page_aflag_set(tm, PGA_REFERENCED);
1256 vm_page_busy_sleep(tm, "madvpo", false);
1259 vm_page_advise(tm, advice);
1261 vm_object_madvise_freespace(tobject, advice, tm->pindex, 1);
1263 if (tobject != object)
1264 VM_OBJECT_WUNLOCK(tobject);
1266 VM_OBJECT_WUNLOCK(object);
1272 * Create a new object which is backed by the
1273 * specified existing object range. The source
1274 * object reference is deallocated.
1276 * The new object and offset into that object
1277 * are returned in the source parameters.
1281 vm_object_t *object, /* IN/OUT */
1282 vm_ooffset_t *offset, /* IN/OUT */
1291 * Don't create the new object if the old object isn't shared.
1293 if (source != NULL) {
1294 VM_OBJECT_WLOCK(source);
1295 if (source->ref_count == 1 &&
1296 source->handle == NULL &&
1297 (source->type == OBJT_DEFAULT ||
1298 source->type == OBJT_SWAP)) {
1299 VM_OBJECT_WUNLOCK(source);
1302 VM_OBJECT_WUNLOCK(source);
1306 * Allocate a new object with the given length.
1308 result = vm_object_allocate(OBJT_DEFAULT, atop(length));
1311 * The new object shadows the source object, adding a reference to it.
1312 * Our caller changes his reference to point to the new object,
1313 * removing a reference to the source object. Net result: no change
1314 * of reference count.
1316 * Try to optimize the result object's page color when shadowing
1317 * in order to maintain page coloring consistency in the combined
1320 result->backing_object = source;
1322 * Store the offset into the source object, and fix up the offset into
1325 result->backing_object_offset = *offset;
1326 if (source != NULL) {
1327 VM_OBJECT_WLOCK(source);
1328 result->domain = source->domain;
1329 LIST_INSERT_HEAD(&source->shadow_head, result, shadow_list);
1330 source->shadow_count++;
1331 #if VM_NRESERVLEVEL > 0
1332 result->flags |= source->flags & OBJ_COLORED;
1333 result->pg_color = (source->pg_color + OFF_TO_IDX(*offset)) &
1334 ((1 << (VM_NFREEORDER - 1)) - 1);
1336 VM_OBJECT_WUNLOCK(source);
1341 * Return the new things
1350 * Split the pages in a map entry into a new object. This affords
1351 * easier removal of unused pages, and keeps object inheritance from
1352 * being a negative impact on memory usage.
1355 vm_object_split(vm_map_entry_t entry)
1357 vm_page_t m, m_next;
1358 vm_object_t orig_object, new_object, source;
1359 vm_pindex_t idx, offidxstart;
1362 orig_object = entry->object.vm_object;
1363 if (orig_object->type != OBJT_DEFAULT && orig_object->type != OBJT_SWAP)
1365 if (orig_object->ref_count <= 1)
1367 VM_OBJECT_WUNLOCK(orig_object);
1369 offidxstart = OFF_TO_IDX(entry->offset);
1370 size = atop(entry->end - entry->start);
1373 * If swap_pager_copy() is later called, it will convert new_object
1374 * into a swap object.
1376 new_object = vm_object_allocate(OBJT_DEFAULT, size);
1379 * At this point, the new object is still private, so the order in
1380 * which the original and new objects are locked does not matter.
1382 VM_OBJECT_WLOCK(new_object);
1383 VM_OBJECT_WLOCK(orig_object);
1384 new_object->domain = orig_object->domain;
1385 source = orig_object->backing_object;
1386 if (source != NULL) {
1387 VM_OBJECT_WLOCK(source);
1388 if ((source->flags & OBJ_DEAD) != 0) {
1389 VM_OBJECT_WUNLOCK(source);
1390 VM_OBJECT_WUNLOCK(orig_object);
1391 VM_OBJECT_WUNLOCK(new_object);
1392 vm_object_deallocate(new_object);
1393 VM_OBJECT_WLOCK(orig_object);
1396 LIST_INSERT_HEAD(&source->shadow_head,
1397 new_object, shadow_list);
1398 source->shadow_count++;
1399 vm_object_reference_locked(source); /* for new_object */
1400 vm_object_clear_flag(source, OBJ_ONEMAPPING);
1401 VM_OBJECT_WUNLOCK(source);
1402 new_object->backing_object_offset =
1403 orig_object->backing_object_offset + entry->offset;
1404 new_object->backing_object = source;
1406 if (orig_object->cred != NULL) {
1407 new_object->cred = orig_object->cred;
1408 crhold(orig_object->cred);
1409 new_object->charge = ptoa(size);
1410 KASSERT(orig_object->charge >= ptoa(size),
1411 ("orig_object->charge < 0"));
1412 orig_object->charge -= ptoa(size);
1415 m = vm_page_find_least(orig_object, offidxstart);
1416 for (; m != NULL && (idx = m->pindex - offidxstart) < size;
1418 m_next = TAILQ_NEXT(m, listq);
1421 * We must wait for pending I/O to complete before we can
1424 * We do not have to VM_PROT_NONE the page as mappings should
1425 * not be changed by this operation.
1427 if (vm_page_busied(m)) {
1428 VM_OBJECT_WUNLOCK(new_object);
1430 VM_OBJECT_WUNLOCK(orig_object);
1431 vm_page_busy_sleep(m, "spltwt", false);
1432 VM_OBJECT_WLOCK(orig_object);
1433 VM_OBJECT_WLOCK(new_object);
1437 /* vm_page_rename() will dirty the page. */
1438 if (vm_page_rename(m, new_object, idx)) {
1439 VM_OBJECT_WUNLOCK(new_object);
1440 VM_OBJECT_WUNLOCK(orig_object);
1442 VM_OBJECT_WLOCK(orig_object);
1443 VM_OBJECT_WLOCK(new_object);
1446 #if VM_NRESERVLEVEL > 0
1448 * If some of the reservation's allocated pages remain with
1449 * the original object, then transferring the reservation to
1450 * the new object is neither particularly beneficial nor
1451 * particularly harmful as compared to leaving the reservation
1452 * with the original object. If, however, all of the
1453 * reservation's allocated pages are transferred to the new
1454 * object, then transferring the reservation is typically
1455 * beneficial. Determining which of these two cases applies
1456 * would be more costly than unconditionally renaming the
1459 vm_reserv_rename(m, new_object, orig_object, offidxstart);
1461 if (orig_object->type == OBJT_SWAP)
1464 if (orig_object->type == OBJT_SWAP) {
1466 * swap_pager_copy() can sleep, in which case the orig_object's
1467 * and new_object's locks are released and reacquired.
1469 swap_pager_copy(orig_object, new_object, offidxstart, 0);
1470 TAILQ_FOREACH(m, &new_object->memq, listq)
1473 VM_OBJECT_WUNLOCK(orig_object);
1474 VM_OBJECT_WUNLOCK(new_object);
1475 entry->object.vm_object = new_object;
1476 entry->offset = 0LL;
1477 vm_object_deallocate(orig_object);
1478 VM_OBJECT_WLOCK(new_object);
1481 #define OBSC_COLLAPSE_NOWAIT 0x0002
1482 #define OBSC_COLLAPSE_WAIT 0x0004
1485 vm_object_collapse_scan_wait(vm_object_t object, vm_page_t p, vm_page_t next,
1488 vm_object_t backing_object;
1490 VM_OBJECT_ASSERT_WLOCKED(object);
1491 backing_object = object->backing_object;
1492 VM_OBJECT_ASSERT_WLOCKED(backing_object);
1494 KASSERT(p == NULL || vm_page_busied(p), ("unbusy page %p", p));
1495 KASSERT(p == NULL || p->object == object || p->object == backing_object,
1496 ("invalid ownership %p %p %p", p, object, backing_object));
1497 if ((op & OBSC_COLLAPSE_NOWAIT) != 0)
1501 VM_OBJECT_WUNLOCK(object);
1502 VM_OBJECT_WUNLOCK(backing_object);
1503 /* The page is only NULL when rename fails. */
1507 vm_page_busy_sleep(p, "vmocol", false);
1508 VM_OBJECT_WLOCK(object);
1509 VM_OBJECT_WLOCK(backing_object);
1510 return (TAILQ_FIRST(&backing_object->memq));
1514 vm_object_scan_all_shadowed(vm_object_t object)
1516 vm_object_t backing_object;
1518 vm_pindex_t backing_offset_index, new_pindex, pi, ps;
1520 VM_OBJECT_ASSERT_WLOCKED(object);
1521 VM_OBJECT_ASSERT_WLOCKED(object->backing_object);
1523 backing_object = object->backing_object;
1525 if (backing_object->type != OBJT_DEFAULT &&
1526 backing_object->type != OBJT_SWAP)
1529 pi = backing_offset_index = OFF_TO_IDX(object->backing_object_offset);
1530 p = vm_page_find_least(backing_object, pi);
1531 ps = swap_pager_find_least(backing_object, pi);
1534 * Only check pages inside the parent object's range and
1535 * inside the parent object's mapping of the backing object.
1538 if (p != NULL && p->pindex < pi)
1539 p = TAILQ_NEXT(p, listq);
1541 ps = swap_pager_find_least(backing_object, pi);
1542 if (p == NULL && ps >= backing_object->size)
1547 pi = MIN(p->pindex, ps);
1549 new_pindex = pi - backing_offset_index;
1550 if (new_pindex >= object->size)
1554 * See if the parent has the page or if the parent's object
1555 * pager has the page. If the parent has the page but the page
1556 * is not valid, the parent's object pager must have the page.
1558 * If this fails, the parent does not completely shadow the
1559 * object and we might as well give up now.
1561 pp = vm_page_lookup(object, new_pindex);
1562 if ((pp == NULL || pp->valid == 0) &&
1563 !vm_pager_has_page(object, new_pindex, NULL, NULL))
1570 vm_object_collapse_scan(vm_object_t object, int op)
1572 vm_object_t backing_object;
1573 vm_page_t next, p, pp;
1574 vm_pindex_t backing_offset_index, new_pindex;
1576 VM_OBJECT_ASSERT_WLOCKED(object);
1577 VM_OBJECT_ASSERT_WLOCKED(object->backing_object);
1579 backing_object = object->backing_object;
1580 backing_offset_index = OFF_TO_IDX(object->backing_object_offset);
1583 * Initial conditions
1585 if ((op & OBSC_COLLAPSE_WAIT) != 0)
1586 vm_object_set_flag(backing_object, OBJ_DEAD);
1591 for (p = TAILQ_FIRST(&backing_object->memq); p != NULL; p = next) {
1592 next = TAILQ_NEXT(p, listq);
1593 new_pindex = p->pindex - backing_offset_index;
1596 * Check for busy page
1598 if (vm_page_busied(p)) {
1599 next = vm_object_collapse_scan_wait(object, p, next, op);
1603 KASSERT(p->object == backing_object,
1604 ("vm_object_collapse_scan: object mismatch"));
1606 if (p->pindex < backing_offset_index ||
1607 new_pindex >= object->size) {
1608 if (backing_object->type == OBJT_SWAP)
1609 swap_pager_freespace(backing_object, p->pindex,
1613 * Page is out of the parent object's range, we can
1614 * simply destroy it.
1617 KASSERT(!pmap_page_is_mapped(p),
1618 ("freeing mapped page %p", p));
1619 if (p->wire_count == 0)
1627 pp = vm_page_lookup(object, new_pindex);
1628 if (pp != NULL && vm_page_busied(pp)) {
1630 * The page in the parent is busy and possibly not
1631 * (yet) valid. Until its state is finalized by the
1632 * busy bit owner, we can't tell whether it shadows the
1633 * original page. Therefore, we must either skip it
1634 * and the original (backing_object) page or wait for
1635 * its state to be finalized.
1637 * This is due to a race with vm_fault() where we must
1638 * unbusy the original (backing_obj) page before we can
1639 * (re)lock the parent. Hence we can get here.
1641 next = vm_object_collapse_scan_wait(object, pp, next,
1646 KASSERT(pp == NULL || pp->valid != 0,
1647 ("unbusy invalid page %p", pp));
1649 if (pp != NULL || vm_pager_has_page(object, new_pindex, NULL,
1652 * The page already exists in the parent OR swap exists
1653 * for this location in the parent. Leave the parent's
1654 * page alone. Destroy the original page from the
1657 if (backing_object->type == OBJT_SWAP)
1658 swap_pager_freespace(backing_object, p->pindex,
1661 KASSERT(!pmap_page_is_mapped(p),
1662 ("freeing mapped page %p", p));
1663 if (p->wire_count == 0)
1672 * Page does not exist in parent, rename the page from the
1673 * backing object to the main object.
1675 * If the page was mapped to a process, it can remain mapped
1676 * through the rename. vm_page_rename() will dirty the page.
1678 if (vm_page_rename(p, object, new_pindex)) {
1679 next = vm_object_collapse_scan_wait(object, NULL, next,
1684 /* Use the old pindex to free the right page. */
1685 if (backing_object->type == OBJT_SWAP)
1686 swap_pager_freespace(backing_object,
1687 new_pindex + backing_offset_index, 1);
1689 #if VM_NRESERVLEVEL > 0
1691 * Rename the reservation.
1693 vm_reserv_rename(p, object, backing_object,
1694 backing_offset_index);
1702 * this version of collapse allows the operation to occur earlier and
1703 * when paging_in_progress is true for an object... This is not a complete
1704 * operation, but should plug 99.9% of the rest of the leaks.
1707 vm_object_qcollapse(vm_object_t object)
1709 vm_object_t backing_object = object->backing_object;
1711 VM_OBJECT_ASSERT_WLOCKED(object);
1712 VM_OBJECT_ASSERT_WLOCKED(backing_object);
1714 if (backing_object->ref_count != 1)
1717 vm_object_collapse_scan(object, OBSC_COLLAPSE_NOWAIT);
1721 * vm_object_collapse:
1723 * Collapse an object with the object backing it.
1724 * Pages in the backing object are moved into the
1725 * parent, and the backing object is deallocated.
1728 vm_object_collapse(vm_object_t object)
1730 vm_object_t backing_object, new_backing_object;
1732 VM_OBJECT_ASSERT_WLOCKED(object);
1736 * Verify that the conditions are right for collapse:
1738 * The object exists and the backing object exists.
1740 if ((backing_object = object->backing_object) == NULL)
1744 * we check the backing object first, because it is most likely
1747 VM_OBJECT_WLOCK(backing_object);
1748 if (backing_object->handle != NULL ||
1749 (backing_object->type != OBJT_DEFAULT &&
1750 backing_object->type != OBJT_SWAP) ||
1751 (backing_object->flags & OBJ_DEAD) ||
1752 object->handle != NULL ||
1753 (object->type != OBJT_DEFAULT &&
1754 object->type != OBJT_SWAP) ||
1755 (object->flags & OBJ_DEAD)) {
1756 VM_OBJECT_WUNLOCK(backing_object);
1760 if (object->paging_in_progress != 0 ||
1761 backing_object->paging_in_progress != 0) {
1762 vm_object_qcollapse(object);
1763 VM_OBJECT_WUNLOCK(backing_object);
1768 * We know that we can either collapse the backing object (if
1769 * the parent is the only reference to it) or (perhaps) have
1770 * the parent bypass the object if the parent happens to shadow
1771 * all the resident pages in the entire backing object.
1773 * This is ignoring pager-backed pages such as swap pages.
1774 * vm_object_collapse_scan fails the shadowing test in this
1777 if (backing_object->ref_count == 1) {
1778 vm_object_pip_add(object, 1);
1779 vm_object_pip_add(backing_object, 1);
1782 * If there is exactly one reference to the backing
1783 * object, we can collapse it into the parent.
1785 vm_object_collapse_scan(object, OBSC_COLLAPSE_WAIT);
1787 #if VM_NRESERVLEVEL > 0
1789 * Break any reservations from backing_object.
1791 if (__predict_false(!LIST_EMPTY(&backing_object->rvq)))
1792 vm_reserv_break_all(backing_object);
1796 * Move the pager from backing_object to object.
1798 if (backing_object->type == OBJT_SWAP) {
1800 * swap_pager_copy() can sleep, in which case
1801 * the backing_object's and object's locks are
1802 * released and reacquired.
1803 * Since swap_pager_copy() is being asked to
1804 * destroy the source, it will change the
1805 * backing_object's type to OBJT_DEFAULT.
1810 OFF_TO_IDX(object->backing_object_offset), TRUE);
1813 * Object now shadows whatever backing_object did.
1814 * Note that the reference to
1815 * backing_object->backing_object moves from within
1816 * backing_object to within object.
1818 LIST_REMOVE(object, shadow_list);
1819 backing_object->shadow_count--;
1820 if (backing_object->backing_object) {
1821 VM_OBJECT_WLOCK(backing_object->backing_object);
1822 LIST_REMOVE(backing_object, shadow_list);
1824 &backing_object->backing_object->shadow_head,
1825 object, shadow_list);
1827 * The shadow_count has not changed.
1829 VM_OBJECT_WUNLOCK(backing_object->backing_object);
1831 object->backing_object = backing_object->backing_object;
1832 object->backing_object_offset +=
1833 backing_object->backing_object_offset;
1836 * Discard backing_object.
1838 * Since the backing object has no pages, no pager left,
1839 * and no object references within it, all that is
1840 * necessary is to dispose of it.
1842 KASSERT(backing_object->ref_count == 1, (
1843 "backing_object %p was somehow re-referenced during collapse!",
1845 vm_object_pip_wakeup(backing_object);
1846 backing_object->type = OBJT_DEAD;
1847 backing_object->ref_count = 0;
1848 VM_OBJECT_WUNLOCK(backing_object);
1849 vm_object_destroy(backing_object);
1851 vm_object_pip_wakeup(object);
1852 counter_u64_add(object_collapses, 1);
1855 * If we do not entirely shadow the backing object,
1856 * there is nothing we can do so we give up.
1858 if (object->resident_page_count != object->size &&
1859 !vm_object_scan_all_shadowed(object)) {
1860 VM_OBJECT_WUNLOCK(backing_object);
1865 * Make the parent shadow the next object in the
1866 * chain. Deallocating backing_object will not remove
1867 * it, since its reference count is at least 2.
1869 LIST_REMOVE(object, shadow_list);
1870 backing_object->shadow_count--;
1872 new_backing_object = backing_object->backing_object;
1873 if ((object->backing_object = new_backing_object) != NULL) {
1874 VM_OBJECT_WLOCK(new_backing_object);
1876 &new_backing_object->shadow_head,
1880 new_backing_object->shadow_count++;
1881 vm_object_reference_locked(new_backing_object);
1882 VM_OBJECT_WUNLOCK(new_backing_object);
1883 object->backing_object_offset +=
1884 backing_object->backing_object_offset;
1888 * Drop the reference count on backing_object. Since
1889 * its ref_count was at least 2, it will not vanish.
1891 backing_object->ref_count--;
1892 VM_OBJECT_WUNLOCK(backing_object);
1893 counter_u64_add(object_bypasses, 1);
1897 * Try again with this object's new backing object.
1903 * vm_object_page_remove:
1905 * For the given object, either frees or invalidates each of the
1906 * specified pages. In general, a page is freed. However, if a page is
1907 * wired for any reason other than the existence of a managed, wired
1908 * mapping, then it may be invalidated but not removed from the object.
1909 * Pages are specified by the given range ["start", "end") and the option
1910 * OBJPR_CLEANONLY. As a special case, if "end" is zero, then the range
1911 * extends from "start" to the end of the object. If the option
1912 * OBJPR_CLEANONLY is specified, then only the non-dirty pages within the
1913 * specified range are affected. If the option OBJPR_NOTMAPPED is
1914 * specified, then the pages within the specified range must have no
1915 * mappings. Otherwise, if this option is not specified, any mappings to
1916 * the specified pages are removed before the pages are freed or
1919 * In general, this operation should only be performed on objects that
1920 * contain managed pages. There are, however, two exceptions. First, it
1921 * is performed on the kernel and kmem objects by vm_map_entry_delete().
1922 * Second, it is used by msync(..., MS_INVALIDATE) to invalidate device-
1923 * backed pages. In both of these cases, the option OBJPR_CLEANONLY must
1924 * not be specified and the option OBJPR_NOTMAPPED must be specified.
1926 * The object must be locked.
1929 vm_object_page_remove(vm_object_t object, vm_pindex_t start, vm_pindex_t end,
1935 VM_OBJECT_ASSERT_WLOCKED(object);
1936 KASSERT((object->flags & OBJ_UNMANAGED) == 0 ||
1937 (options & (OBJPR_CLEANONLY | OBJPR_NOTMAPPED)) == OBJPR_NOTMAPPED,
1938 ("vm_object_page_remove: illegal options for object %p", object));
1939 if (object->resident_page_count == 0)
1941 vm_object_pip_add(object, 1);
1943 p = vm_page_find_least(object, start);
1947 * Here, the variable "p" is either (1) the page with the least pindex
1948 * greater than or equal to the parameter "start" or (2) NULL.
1950 for (; p != NULL && (p->pindex < end || end == 0); p = next) {
1951 next = TAILQ_NEXT(p, listq);
1954 * If the page is wired for any reason besides the existence
1955 * of managed, wired mappings, then it cannot be freed. For
1956 * example, fictitious pages, which represent device memory,
1957 * are inherently wired and cannot be freed. They can,
1958 * however, be invalidated if the option OBJPR_CLEANONLY is
1961 vm_page_change_lock(p, &mtx);
1962 if (vm_page_xbusied(p)) {
1963 VM_OBJECT_WUNLOCK(object);
1964 vm_page_busy_sleep(p, "vmopax", true);
1965 VM_OBJECT_WLOCK(object);
1968 if (p->wire_count != 0) {
1969 if ((options & OBJPR_NOTMAPPED) == 0 &&
1970 object->ref_count != 0)
1972 if ((options & OBJPR_CLEANONLY) == 0) {
1978 if (vm_page_busied(p)) {
1979 VM_OBJECT_WUNLOCK(object);
1980 vm_page_busy_sleep(p, "vmopar", false);
1981 VM_OBJECT_WLOCK(object);
1984 KASSERT((p->flags & PG_FICTITIOUS) == 0,
1985 ("vm_object_page_remove: page %p is fictitious", p));
1986 if ((options & OBJPR_CLEANONLY) != 0 && p->valid != 0) {
1987 if ((options & OBJPR_NOTMAPPED) == 0 &&
1988 object->ref_count != 0)
1989 pmap_remove_write(p);
1993 if ((options & OBJPR_NOTMAPPED) == 0 && object->ref_count != 0)
1999 vm_object_pip_wakeup(object);
2003 * vm_object_page_noreuse:
2005 * For the given object, attempt to move the specified pages to
2006 * the head of the inactive queue. This bypasses regular LRU
2007 * operation and allows the pages to be reused quickly under memory
2008 * pressure. If a page is wired for any reason, then it will not
2009 * be queued. Pages are specified by the range ["start", "end").
2010 * As a special case, if "end" is zero, then the range extends from
2011 * "start" to the end of the object.
2013 * This operation should only be performed on objects that
2014 * contain non-fictitious, managed pages.
2016 * The object must be locked.
2019 vm_object_page_noreuse(vm_object_t object, vm_pindex_t start, vm_pindex_t end)
2024 VM_OBJECT_ASSERT_LOCKED(object);
2025 KASSERT((object->flags & (OBJ_FICTITIOUS | OBJ_UNMANAGED)) == 0,
2026 ("vm_object_page_noreuse: illegal object %p", object));
2027 if (object->resident_page_count == 0)
2029 p = vm_page_find_least(object, start);
2032 * Here, the variable "p" is either (1) the page with the least pindex
2033 * greater than or equal to the parameter "start" or (2) NULL.
2036 for (; p != NULL && (p->pindex < end || end == 0); p = next) {
2037 next = TAILQ_NEXT(p, listq);
2038 vm_page_change_lock(p, &mtx);
2039 vm_page_deactivate_noreuse(p);
2046 * Populate the specified range of the object with valid pages. Returns
2047 * TRUE if the range is successfully populated and FALSE otherwise.
2049 * Note: This function should be optimized to pass a larger array of
2050 * pages to vm_pager_get_pages() before it is applied to a non-
2051 * OBJT_DEVICE object.
2053 * The object must be locked.
2056 vm_object_populate(vm_object_t object, vm_pindex_t start, vm_pindex_t end)
2062 VM_OBJECT_ASSERT_WLOCKED(object);
2063 for (pindex = start; pindex < end; pindex++) {
2064 m = vm_page_grab(object, pindex, VM_ALLOC_NORMAL);
2065 if (m->valid != VM_PAGE_BITS_ALL) {
2066 rv = vm_pager_get_pages(object, &m, 1, NULL, NULL);
2067 if (rv != VM_PAGER_OK) {
2075 * Keep "m" busy because a subsequent iteration may unlock
2079 if (pindex > start) {
2080 m = vm_page_lookup(object, start);
2081 while (m != NULL && m->pindex < pindex) {
2083 m = TAILQ_NEXT(m, listq);
2086 return (pindex == end);
2090 * Routine: vm_object_coalesce
2091 * Function: Coalesces two objects backing up adjoining
2092 * regions of memory into a single object.
2094 * returns TRUE if objects were combined.
2096 * NOTE: Only works at the moment if the second object is NULL -
2097 * if it's not, which object do we lock first?
2100 * prev_object First object to coalesce
2101 * prev_offset Offset into prev_object
2102 * prev_size Size of reference to prev_object
2103 * next_size Size of reference to the second object
2104 * reserved Indicator that extension region has
2105 * swap accounted for
2108 * The object must *not* be locked.
2111 vm_object_coalesce(vm_object_t prev_object, vm_ooffset_t prev_offset,
2112 vm_size_t prev_size, vm_size_t next_size, boolean_t reserved)
2114 vm_pindex_t next_pindex;
2116 if (prev_object == NULL)
2118 VM_OBJECT_WLOCK(prev_object);
2119 if ((prev_object->type != OBJT_DEFAULT &&
2120 prev_object->type != OBJT_SWAP) ||
2121 (prev_object->flags & OBJ_TMPFS_NODE) != 0) {
2122 VM_OBJECT_WUNLOCK(prev_object);
2127 * Try to collapse the object first
2129 vm_object_collapse(prev_object);
2132 * Can't coalesce if: . more than one reference . paged out . shadows
2133 * another object . has a copy elsewhere (any of which mean that the
2134 * pages not mapped to prev_entry may be in use anyway)
2136 if (prev_object->backing_object != NULL) {
2137 VM_OBJECT_WUNLOCK(prev_object);
2141 prev_size >>= PAGE_SHIFT;
2142 next_size >>= PAGE_SHIFT;
2143 next_pindex = OFF_TO_IDX(prev_offset) + prev_size;
2145 if (prev_object->ref_count > 1 &&
2146 prev_object->size != next_pindex &&
2147 (prev_object->flags & OBJ_ONEMAPPING) == 0) {
2148 VM_OBJECT_WUNLOCK(prev_object);
2153 * Account for the charge.
2155 if (prev_object->cred != NULL) {
2158 * If prev_object was charged, then this mapping,
2159 * although not charged now, may become writable
2160 * later. Non-NULL cred in the object would prevent
2161 * swap reservation during enabling of the write
2162 * access, so reserve swap now. Failed reservation
2163 * cause allocation of the separate object for the map
2164 * entry, and swap reservation for this entry is
2165 * managed in appropriate time.
2167 if (!reserved && !swap_reserve_by_cred(ptoa(next_size),
2168 prev_object->cred)) {
2169 VM_OBJECT_WUNLOCK(prev_object);
2172 prev_object->charge += ptoa(next_size);
2176 * Remove any pages that may still be in the object from a previous
2179 if (next_pindex < prev_object->size) {
2180 vm_object_page_remove(prev_object, next_pindex, next_pindex +
2182 if (prev_object->type == OBJT_SWAP)
2183 swap_pager_freespace(prev_object,
2184 next_pindex, next_size);
2186 if (prev_object->cred != NULL) {
2187 KASSERT(prev_object->charge >=
2188 ptoa(prev_object->size - next_pindex),
2189 ("object %p overcharged 1 %jx %jx", prev_object,
2190 (uintmax_t)next_pindex, (uintmax_t)next_size));
2191 prev_object->charge -= ptoa(prev_object->size -
2198 * Extend the object if necessary.
2200 if (next_pindex + next_size > prev_object->size)
2201 prev_object->size = next_pindex + next_size;
2203 VM_OBJECT_WUNLOCK(prev_object);
2208 vm_object_set_writeable_dirty(vm_object_t object)
2211 VM_OBJECT_ASSERT_WLOCKED(object);
2212 if (object->type != OBJT_VNODE) {
2213 if ((object->flags & OBJ_TMPFS_NODE) != 0) {
2214 KASSERT(object->type == OBJT_SWAP, ("non-swap tmpfs"));
2215 vm_object_set_flag(object, OBJ_TMPFS_DIRTY);
2219 object->generation++;
2220 if ((object->flags & OBJ_MIGHTBEDIRTY) != 0)
2222 vm_object_set_flag(object, OBJ_MIGHTBEDIRTY);
2228 * For each page offset within the specified range of the given object,
2229 * find the highest-level page in the shadow chain and unwire it. A page
2230 * must exist at every page offset, and the highest-level page must be
2234 vm_object_unwire(vm_object_t object, vm_ooffset_t offset, vm_size_t length,
2237 vm_object_t tobject, t1object;
2239 vm_pindex_t end_pindex, pindex, tpindex;
2240 int depth, locked_depth;
2242 KASSERT((offset & PAGE_MASK) == 0,
2243 ("vm_object_unwire: offset is not page aligned"));
2244 KASSERT((length & PAGE_MASK) == 0,
2245 ("vm_object_unwire: length is not a multiple of PAGE_SIZE"));
2246 /* The wired count of a fictitious page never changes. */
2247 if ((object->flags & OBJ_FICTITIOUS) != 0)
2249 pindex = OFF_TO_IDX(offset);
2250 end_pindex = pindex + atop(length);
2253 VM_OBJECT_RLOCK(object);
2254 m = vm_page_find_least(object, pindex);
2255 while (pindex < end_pindex) {
2256 if (m == NULL || pindex < m->pindex) {
2258 * The first object in the shadow chain doesn't
2259 * contain a page at the current index. Therefore,
2260 * the page must exist in a backing object.
2267 OFF_TO_IDX(tobject->backing_object_offset);
2268 tobject = tobject->backing_object;
2269 KASSERT(tobject != NULL,
2270 ("vm_object_unwire: missing page"));
2271 if ((tobject->flags & OBJ_FICTITIOUS) != 0)
2274 if (depth == locked_depth) {
2276 VM_OBJECT_RLOCK(tobject);
2278 } while ((tm = vm_page_lookup(tobject, tpindex)) ==
2282 m = TAILQ_NEXT(m, listq);
2285 if (vm_page_xbusied(tm)) {
2286 for (tobject = object; locked_depth >= 1;
2288 t1object = tobject->backing_object;
2289 VM_OBJECT_RUNLOCK(tobject);
2292 vm_page_busy_sleep(tm, "unwbo", true);
2295 vm_page_unwire(tm, queue);
2300 /* Release the accumulated object locks. */
2301 for (tobject = object; locked_depth >= 1; locked_depth--) {
2302 t1object = tobject->backing_object;
2303 VM_OBJECT_RUNLOCK(tobject);
2309 vm_object_vnode(vm_object_t object)
2312 VM_OBJECT_ASSERT_LOCKED(object);
2313 if (object->type == OBJT_VNODE)
2314 return (object->handle);
2315 if (object->type == OBJT_SWAP && (object->flags & OBJ_TMPFS) != 0)
2316 return (object->un_pager.swp.swp_tmpfs);
2321 sysctl_vm_object_list(SYSCTL_HANDLER_ARGS)
2323 struct kinfo_vmobject *kvo;
2324 char *fullpath, *freepath;
2331 if (req->oldptr == NULL) {
2333 * If an old buffer has not been provided, generate an
2334 * estimate of the space needed for a subsequent call.
2336 mtx_lock(&vm_object_list_mtx);
2338 TAILQ_FOREACH(obj, &vm_object_list, object_list) {
2339 if (obj->type == OBJT_DEAD)
2343 mtx_unlock(&vm_object_list_mtx);
2344 return (SYSCTL_OUT(req, NULL, sizeof(struct kinfo_vmobject) *
2348 kvo = malloc(sizeof(*kvo), M_TEMP, M_WAITOK);
2352 * VM objects are type stable and are never removed from the
2353 * list once added. This allows us to safely read obj->object_list
2354 * after reacquiring the VM object lock.
2356 mtx_lock(&vm_object_list_mtx);
2357 TAILQ_FOREACH(obj, &vm_object_list, object_list) {
2358 if (obj->type == OBJT_DEAD)
2360 VM_OBJECT_RLOCK(obj);
2361 if (obj->type == OBJT_DEAD) {
2362 VM_OBJECT_RUNLOCK(obj);
2365 mtx_unlock(&vm_object_list_mtx);
2366 kvo->kvo_size = ptoa(obj->size);
2367 kvo->kvo_resident = obj->resident_page_count;
2368 kvo->kvo_ref_count = obj->ref_count;
2369 kvo->kvo_shadow_count = obj->shadow_count;
2370 kvo->kvo_memattr = obj->memattr;
2371 kvo->kvo_active = 0;
2372 kvo->kvo_inactive = 0;
2373 TAILQ_FOREACH(m, &obj->memq, listq) {
2375 * A page may belong to the object but be
2376 * dequeued and set to PQ_NONE while the
2377 * object lock is not held. This makes the
2378 * reads of m->queue below racy, and we do not
2379 * count pages set to PQ_NONE. However, this
2380 * sysctl is only meant to give an
2381 * approximation of the system anyway.
2383 if (m->queue == PQ_ACTIVE)
2385 else if (m->queue == PQ_INACTIVE)
2386 kvo->kvo_inactive++;
2389 kvo->kvo_vn_fileid = 0;
2390 kvo->kvo_vn_fsid = 0;
2391 kvo->kvo_vn_fsid_freebsd11 = 0;
2395 switch (obj->type) {
2397 kvo->kvo_type = KVME_TYPE_DEFAULT;
2400 kvo->kvo_type = KVME_TYPE_VNODE;
2405 kvo->kvo_type = KVME_TYPE_SWAP;
2408 kvo->kvo_type = KVME_TYPE_DEVICE;
2411 kvo->kvo_type = KVME_TYPE_PHYS;
2414 kvo->kvo_type = KVME_TYPE_DEAD;
2417 kvo->kvo_type = KVME_TYPE_SG;
2419 case OBJT_MGTDEVICE:
2420 kvo->kvo_type = KVME_TYPE_MGTDEVICE;
2423 kvo->kvo_type = KVME_TYPE_UNKNOWN;
2426 VM_OBJECT_RUNLOCK(obj);
2428 vn_fullpath(curthread, vp, &fullpath, &freepath);
2429 vn_lock(vp, LK_SHARED | LK_RETRY);
2430 if (VOP_GETATTR(vp, &va, curthread->td_ucred) == 0) {
2431 kvo->kvo_vn_fileid = va.va_fileid;
2432 kvo->kvo_vn_fsid = va.va_fsid;
2433 kvo->kvo_vn_fsid_freebsd11 = va.va_fsid;
2439 strlcpy(kvo->kvo_path, fullpath, sizeof(kvo->kvo_path));
2440 if (freepath != NULL)
2441 free(freepath, M_TEMP);
2443 /* Pack record size down */
2444 kvo->kvo_structsize = offsetof(struct kinfo_vmobject, kvo_path)
2445 + strlen(kvo->kvo_path) + 1;
2446 kvo->kvo_structsize = roundup(kvo->kvo_structsize,
2448 error = SYSCTL_OUT(req, kvo, kvo->kvo_structsize);
2449 mtx_lock(&vm_object_list_mtx);
2453 mtx_unlock(&vm_object_list_mtx);
2457 SYSCTL_PROC(_vm, OID_AUTO, objects, CTLTYPE_STRUCT | CTLFLAG_RW | CTLFLAG_SKIP |
2458 CTLFLAG_MPSAFE, NULL, 0, sysctl_vm_object_list, "S,kinfo_vmobject",
2459 "List of VM objects");
2461 #include "opt_ddb.h"
2463 #include <sys/kernel.h>
2465 #include <sys/cons.h>
2467 #include <ddb/ddb.h>
2470 _vm_object_in_map(vm_map_t map, vm_object_t object, vm_map_entry_t entry)
2473 vm_map_entry_t tmpe;
2481 tmpe = map->header.next;
2482 entcount = map->nentries;
2483 while (entcount-- && (tmpe != &map->header)) {
2484 if (_vm_object_in_map(map, object, tmpe)) {
2489 } else if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
2490 tmpm = entry->object.sub_map;
2491 tmpe = tmpm->header.next;
2492 entcount = tmpm->nentries;
2493 while (entcount-- && tmpe != &tmpm->header) {
2494 if (_vm_object_in_map(tmpm, object, tmpe)) {
2499 } else if ((obj = entry->object.vm_object) != NULL) {
2500 for (; obj; obj = obj->backing_object)
2501 if (obj == object) {
2509 vm_object_in_map(vm_object_t object)
2513 /* sx_slock(&allproc_lock); */
2514 FOREACH_PROC_IN_SYSTEM(p) {
2515 if (!p->p_vmspace /* || (p->p_flag & (P_SYSTEM|P_WEXIT)) */)
2517 if (_vm_object_in_map(&p->p_vmspace->vm_map, object, 0)) {
2518 /* sx_sunlock(&allproc_lock); */
2522 /* sx_sunlock(&allproc_lock); */
2523 if (_vm_object_in_map(kernel_map, object, 0))
2528 DB_SHOW_COMMAND(vmochk, vm_object_check)
2533 * make sure that internal objs are in a map somewhere
2534 * and none have zero ref counts.
2536 TAILQ_FOREACH(object, &vm_object_list, object_list) {
2537 if (object->handle == NULL &&
2538 (object->type == OBJT_DEFAULT || object->type == OBJT_SWAP)) {
2539 if (object->ref_count == 0) {
2540 db_printf("vmochk: internal obj has zero ref count: %ld\n",
2541 (long)object->size);
2543 if (!vm_object_in_map(object)) {
2545 "vmochk: internal obj is not in a map: "
2546 "ref: %d, size: %lu: 0x%lx, backing_object: %p\n",
2547 object->ref_count, (u_long)object->size,
2548 (u_long)object->size,
2549 (void *)object->backing_object);
2556 * vm_object_print: [ debug ]
2558 DB_SHOW_COMMAND(object, vm_object_print_static)
2560 /* XXX convert args. */
2561 vm_object_t object = (vm_object_t)addr;
2562 boolean_t full = have_addr;
2566 /* XXX count is an (unused) arg. Avoid shadowing it. */
2567 #define count was_count
2575 "Object %p: type=%d, size=0x%jx, res=%d, ref=%d, flags=0x%x ruid %d charge %jx\n",
2576 object, (int)object->type, (uintmax_t)object->size,
2577 object->resident_page_count, object->ref_count, object->flags,
2578 object->cred ? object->cred->cr_ruid : -1, (uintmax_t)object->charge);
2579 db_iprintf(" sref=%d, backing_object(%d)=(%p)+0x%jx\n",
2580 object->shadow_count,
2581 object->backing_object ? object->backing_object->ref_count : 0,
2582 object->backing_object, (uintmax_t)object->backing_object_offset);
2589 TAILQ_FOREACH(p, &object->memq, listq) {
2591 db_iprintf("memory:=");
2592 else if (count == 6) {
2600 db_printf("(off=0x%jx,page=0x%jx)",
2601 (uintmax_t)p->pindex, (uintmax_t)VM_PAGE_TO_PHYS(p));
2611 /* XXX need this non-static entry for calling from vm_map_print. */
2614 /* db_expr_t */ long addr,
2615 boolean_t have_addr,
2616 /* db_expr_t */ long count,
2619 vm_object_print_static(addr, have_addr, count, modif);
2622 DB_SHOW_COMMAND(vmopag, vm_object_print_pages)
2627 vm_page_t m, prev_m;
2631 TAILQ_FOREACH(object, &vm_object_list, object_list) {
2632 db_printf("new object: %p\n", (void *)object);
2643 TAILQ_FOREACH(m, &object->memq, listq) {
2644 if (m->pindex > 128)
2646 if ((prev_m = TAILQ_PREV(m, pglist, listq)) != NULL &&
2647 prev_m->pindex + 1 != m->pindex) {
2649 db_printf(" index(%ld)run(%d)pa(0x%lx)\n",
2650 (long)fidx, rcount, (long)pa);
2662 (VM_PAGE_TO_PHYS(m) == pa + rcount * PAGE_SIZE)) {
2667 db_printf(" index(%ld)run(%d)pa(0x%lx)\n",
2668 (long)fidx, rcount, (long)pa);
2678 pa = VM_PAGE_TO_PHYS(m);
2682 db_printf(" index(%ld)run(%d)pa(0x%lx)\n",
2683 (long)fidx, rcount, (long)pa);