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. All advertising materials mentioning features or use of this software
19 * must display the following acknowledgement:
20 * This product includes software developed by the University of
21 * California, Berkeley and its contributors.
22 * 4. Neither the name of the University nor the names of its contributors
23 * may be used to endorse or promote products derived from this software
24 * without specific prior written permission.
26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
38 * from: @(#)vm_object.c 8.5 (Berkeley) 3/22/94
41 * Copyright (c) 1987, 1990 Carnegie-Mellon University.
42 * All rights reserved.
44 * Authors: Avadis Tevanian, Jr., Michael Wayne Young
46 * Permission to use, copy, modify and distribute this software and
47 * its documentation is hereby granted, provided that both the copyright
48 * notice and this permission notice appear in all copies of the
49 * software, derivative works or modified versions, and any portions
50 * thereof, and that both notices appear in supporting documentation.
52 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
53 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
54 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
56 * Carnegie Mellon requests users of this software to return to
58 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
59 * School of Computer Science
60 * Carnegie Mellon University
61 * Pittsburgh PA 15213-3890
63 * any improvements or extensions that they make and grant Carnegie the
64 * rights to redistribute these changes.
66 * $FreeBSD: src/sys/vm/vm_object.c,v 1.171.2.8 2003/05/26 19:17:56 alc Exp $
70 * Virtual memory object module.
73 #include <sys/param.h>
74 #include <sys/systm.h>
75 #include <sys/proc.h> /* for curproc, pageproc */
76 #include <sys/thread.h>
77 #include <sys/vnode.h>
78 #include <sys/vmmeter.h>
80 #include <sys/mount.h>
81 #include <sys/kernel.h>
82 #include <sys/sysctl.h>
83 #include <sys/refcount.h>
86 #include <vm/vm_param.h>
88 #include <vm/vm_map.h>
89 #include <vm/vm_object.h>
90 #include <vm/vm_page.h>
91 #include <vm/vm_pageout.h>
92 #include <vm/vm_pager.h>
93 #include <vm/swap_pager.h>
94 #include <vm/vm_kern.h>
95 #include <vm/vm_extern.h>
96 #include <vm/vm_zone.h>
98 #define EASY_SCAN_FACTOR 8
100 static void vm_object_qcollapse(vm_object_t object);
101 static int vm_object_page_collect_flush(vm_object_t object, vm_page_t p,
103 static void vm_object_lock_init(vm_object_t);
104 static void vm_object_hold_wait(vm_object_t);
108 * Virtual memory objects maintain the actual data
109 * associated with allocated virtual memory. A given
110 * page of memory exists within exactly one object.
112 * An object is only deallocated when all "references"
113 * are given up. Only one "reference" to a given
114 * region of an object should be writeable.
116 * Associated with each object is a list of all resident
117 * memory pages belonging to that object; this list is
118 * maintained by the "vm_page" module, and locked by the object's
121 * Each object also records a "pager" routine which is
122 * used to retrieve (and store) pages to the proper backing
123 * storage. In addition, objects may be backed by other
124 * objects from which they were virtual-copied.
126 * The only items within the object structure which are
127 * modified after time of creation are:
128 * reference count locked by object's lock
129 * pager routine locked by object's lock
133 struct object_q vm_object_list; /* locked by vmobj_token */
134 struct vm_object kernel_object;
136 static long vm_object_count; /* locked by vmobj_token */
137 extern int vm_pageout_page_count;
139 static long object_collapses;
140 static long object_bypasses;
141 static int next_index;
142 static vm_zone_t obj_zone;
143 static struct vm_zone obj_zone_store;
144 #define VM_OBJECTS_INIT 256
145 static struct vm_object vm_objects_init[VM_OBJECTS_INIT];
148 * Misc low level routines
151 vm_object_lock_init(vm_object_t obj)
153 #if defined(DEBUG_LOCKS)
156 obj->debug_hold_bitmap = 0;
157 obj->debug_hold_ovfl = 0;
158 for (i = 0; i < VMOBJ_DEBUG_ARRAY_SIZE; i++) {
159 obj->debug_hold_thrs[i] = NULL;
165 vm_object_lock_swap(void)
171 vm_object_lock(vm_object_t obj)
173 lwkt_getpooltoken(obj);
177 vm_object_unlock(vm_object_t obj)
179 lwkt_relpooltoken(obj);
183 vm_object_assert_held(vm_object_t obj)
185 ASSERT_LWKT_TOKEN_HELD(lwkt_token_pool_lookup(obj));
189 vm_object_hold(vm_object_t obj)
195 * Object must be held (object allocation is stable due to callers
196 * context, typically already holding the token on a parent object)
197 * prior to potentially blocking on the lock, otherwise the object
198 * can get ripped away from us.
200 refcount_acquire(&obj->hold_count);
203 #if defined(DEBUG_LOCKS)
206 i = ffs(~obj->debug_hold_bitmap) - 1;
208 kprintf("vm_object hold count > VMOBJ_DEBUG_ARRAY_SIZE");
209 obj->debug_hold_ovfl = 1;
212 obj->debug_hold_bitmap |= (1 << i);
213 obj->debug_hold_thrs[i] = curthread;
218 vm_object_drop(vm_object_t obj)
223 #if defined(DEBUG_LOCKS)
227 for (i = 0; i < VMOBJ_DEBUG_ARRAY_SIZE; i++) {
228 if ((obj->debug_hold_bitmap & (1 << i)) &&
229 (obj->debug_hold_thrs[i] == curthread)) {
230 obj->debug_hold_bitmap &= ~(1 << i);
231 obj->debug_hold_thrs[i] = NULL;
237 if (found == 0 && obj->debug_hold_ovfl == 0)
238 panic("vm_object: attempt to drop hold on non-self-held obj");
242 * The lock is a pool token, keep holding it across potential
243 * wakeups to interlock the tsleep/wakeup.
245 if (refcount_release(&obj->hold_count))
247 vm_object_unlock(obj);
251 * This can only be called while the caller holds the object
252 * with the OBJ_DEAD interlock. Since there are no refs this
253 * is the only thing preventing an object destruction race.
256 vm_object_hold_wait(vm_object_t obj)
260 #if defined(DEBUG_LOCKS)
263 for (i = 0; i < VMOBJ_DEBUG_ARRAY_SIZE; i++) {
264 if ((obj->debug_hold_bitmap & (1 << i)) &&
265 (obj->debug_hold_thrs[i] == curthread)) {
266 panic("vm_object: self-hold in terminate or collapse");
271 while (obj->hold_count)
272 tsleep(obj, 0, "vmobjhld", 0);
274 vm_object_unlock(obj);
279 * Initialize a freshly allocated object
281 * Used only by vm_object_allocate() and zinitna().
286 _vm_object_allocate(objtype_t type, vm_pindex_t size, vm_object_t object)
290 RB_INIT(&object->rb_memq);
291 LIST_INIT(&object->shadow_head);
295 object->ref_count = 1;
296 object->hold_count = 0;
298 if ((object->type == OBJT_DEFAULT) || (object->type == OBJT_SWAP))
299 vm_object_set_flag(object, OBJ_ONEMAPPING);
300 object->paging_in_progress = 0;
301 object->resident_page_count = 0;
302 object->agg_pv_list_count = 0;
303 object->shadow_count = 0;
304 object->pg_color = next_index;
305 if ( size > (PQ_L2_SIZE / 3 + PQ_PRIME1))
306 incr = PQ_L2_SIZE / 3 + PQ_PRIME1;
309 next_index = (next_index + incr) & PQ_L2_MASK;
310 object->handle = NULL;
311 object->backing_object = NULL;
312 object->backing_object_offset = (vm_ooffset_t) 0;
314 object->generation++;
315 object->swblock_count = 0;
316 RB_INIT(&object->swblock_root);
317 vm_object_lock_init(object);
319 lwkt_gettoken(&vmobj_token);
320 TAILQ_INSERT_TAIL(&vm_object_list, object, object_list);
322 lwkt_reltoken(&vmobj_token);
326 * Initialize the VM objects module.
328 * Called from the low level boot code only.
333 TAILQ_INIT(&vm_object_list);
335 _vm_object_allocate(OBJT_DEFAULT, OFF_TO_IDX(KvaEnd),
338 obj_zone = &obj_zone_store;
339 zbootinit(obj_zone, "VM OBJECT", sizeof (struct vm_object),
340 vm_objects_init, VM_OBJECTS_INIT);
344 vm_object_init2(void)
346 zinitna(obj_zone, NULL, NULL, 0, 0, ZONE_PANICFAIL, 1);
350 * Allocate and return a new object of the specified type and size.
355 vm_object_allocate(objtype_t type, vm_pindex_t size)
359 result = (vm_object_t) zalloc(obj_zone);
361 _vm_object_allocate(type, size, result);
367 * Add an additional reference to a vm_object.
369 * Object passed by caller must be stable or caller must already
370 * hold vmobj_token to avoid races.
373 vm_object_reference(vm_object_t object)
375 lwkt_gettoken(&vmobj_token);
376 vm_object_hold(object);
377 vm_object_reference_locked(object);
378 vm_object_drop(object);
379 lwkt_reltoken(&vmobj_token);
383 vm_object_reference_locked(vm_object_t object)
386 ASSERT_LWKT_TOKEN_HELD(&vmobj_token);
388 /*ASSERT_LWKT_TOKEN_HELD(vm_object_token(object));*/
390 if (object->type == OBJT_VNODE) {
391 vref(object->handle);
392 /* XXX what if the vnode is being destroyed? */
398 * Dereference an object and its underlying vnode.
400 * The caller must hold vmobj_token.
401 * The object must be locked but not held. This function will eat the lock.
404 vm_object_vndeallocate(vm_object_t object)
406 struct vnode *vp = (struct vnode *) object->handle;
408 KASSERT(object->type == OBJT_VNODE,
409 ("vm_object_vndeallocate: not a vnode object"));
410 KASSERT(vp != NULL, ("vm_object_vndeallocate: missing vp"));
411 ASSERT_LWKT_TOKEN_HELD(&vmobj_token);
413 if (object->ref_count == 0) {
414 vprint("vm_object_vndeallocate", vp);
415 panic("vm_object_vndeallocate: bad object reference count");
420 if (object->ref_count == 0)
421 vclrflags(vp, VTEXT);
422 vm_object_unlock(object);
427 * Release a reference to the specified object, gained either through a
428 * vm_object_allocate or a vm_object_reference call. When all references
429 * are gone, storage associated with this object may be relinquished.
431 * The caller does not have to hold the object locked but must have control
432 * over the reference in question in order to guarantee that the object
433 * does not get ripped out from under us.
436 vm_object_deallocate(vm_object_t object)
438 lwkt_gettoken(&vmobj_token);
439 vm_object_deallocate_locked(object);
440 lwkt_reltoken(&vmobj_token);
444 vm_object_deallocate_locked(vm_object_t object)
448 ASSERT_LWKT_TOKEN_HELD(&vmobj_token);
451 vm_object_lock(object);
453 while (object != NULL) {
454 if (object->type == OBJT_VNODE) {
455 vm_object_vndeallocate(object);
456 /* vndeallocate ate the lock */
460 if (object->ref_count == 0) {
461 panic("vm_object_deallocate: object deallocated "
462 "too many times: %d", object->type);
464 if (object->ref_count > 2) {
466 vm_object_unlock(object);
471 * We currently need the vm_token from this point on, and
472 * we must recheck ref_count after acquiring it.
474 lwkt_gettoken(&vm_token);
476 if (object->ref_count > 2) {
478 lwkt_reltoken(&vm_token);
479 vm_object_unlock(object);
484 * Here on ref_count of one or two, which are special cases for
487 * Nominal ref_count > 1 case if the second ref is not from
490 if (object->ref_count == 2 && object->shadow_count == 0) {
491 vm_object_set_flag(object, OBJ_ONEMAPPING);
493 lwkt_reltoken(&vm_token);
494 vm_object_unlock(object);
499 * If the second ref is from a shadow we chain along it
500 * if object's handle is exhausted.
502 if (object->ref_count == 2 && object->shadow_count == 1) {
503 if (object->handle == NULL &&
504 (object->type == OBJT_DEFAULT ||
505 object->type == OBJT_SWAP)) {
506 temp = LIST_FIRST(&object->shadow_head);
507 KASSERT(temp != NULL,
508 ("vm_object_deallocate: ref_count: "
509 "%d, shadow_count: %d",
511 object->shadow_count));
512 lwkt_reltoken(&vm_token);
513 vm_object_lock(temp);
515 if ((temp->handle == NULL) &&
516 (temp->type == OBJT_DEFAULT ||
517 temp->type == OBJT_SWAP)) {
519 * Special case, must handle ref_count
520 * manually to avoid recursion.
523 vm_object_lock_swap();
526 temp->paging_in_progress ||
527 object->paging_in_progress
529 vm_object_pip_sleep(temp,
531 vm_object_pip_sleep(object,
535 if (temp->ref_count == 1) {
538 vm_object_unlock(object);
543 lwkt_gettoken(&vm_token);
544 vm_object_collapse(temp);
545 lwkt_reltoken(&vm_token);
547 vm_object_unlock(object);
551 vm_object_unlock(temp);
553 lwkt_reltoken(&vm_token);
556 vm_object_unlock(object);
561 * Normal dereferencing path
564 if (object->ref_count != 0) {
565 lwkt_reltoken(&vm_token);
566 vm_object_unlock(object);
573 * We may have to loop to resolve races if we block getting
574 * temp's lock. If temp is non NULL we have to swap the
575 * lock order so the original object lock as at the top
578 lwkt_reltoken(&vm_token);
580 while ((temp = object->backing_object) != NULL) {
581 vm_object_lock(temp);
582 if (temp == object->backing_object)
584 vm_object_unlock(temp);
587 LIST_REMOVE(object, shadow_list);
588 temp->shadow_count--;
590 object->backing_object = NULL;
591 vm_object_lock_swap();
595 * Don't double-terminate, we could be in a termination
596 * recursion due to the terminate having to sync data
599 if ((object->flags & OBJ_DEAD) == 0) {
600 vm_object_terminate(object);
601 /* termination ate the object lock */
603 vm_object_unlock(object);
610 * Destroy the specified object, freeing up related resources.
612 * The object must have zero references.
614 * The caller must be holding vmobj_token and properly interlock with
615 * OBJ_DEAD (at the moment).
617 * The caller must have locked the object only, and not be holding it.
618 * This function will eat the caller's lock on the object.
620 static int vm_object_terminate_callback(vm_page_t p, void *data);
623 vm_object_terminate(vm_object_t object)
626 * Make sure no one uses us. Once we set OBJ_DEAD we should be
627 * able to safely block.
629 KKASSERT((object->flags & OBJ_DEAD) == 0);
630 ASSERT_LWKT_TOKEN_HELD(&vmobj_token);
631 vm_object_set_flag(object, OBJ_DEAD);
634 * Wait for the pageout daemon to be done with the object
636 vm_object_pip_wait(object, "objtrm1");
638 KASSERT(!object->paging_in_progress,
639 ("vm_object_terminate: pageout in progress"));
642 * Clean and free the pages, as appropriate. All references to the
643 * object are gone, so we don't need to lock it.
645 if (object->type == OBJT_VNODE) {
649 * Clean pages and flush buffers.
651 vm_object_page_clean(object, 0, 0, OBJPC_SYNC);
653 vp = (struct vnode *) object->handle;
654 vinvalbuf(vp, V_SAVE, 0, 0);
658 * Wait for any I/O to complete, after which there had better not
659 * be any references left on the object.
661 vm_object_pip_wait(object, "objtrm2");
663 if (object->ref_count != 0) {
664 panic("vm_object_terminate: object with references, "
665 "ref_count=%d", object->ref_count);
669 * Now free any remaining pages. For internal objects, this also
670 * removes them from paging queues. Don't free wired pages, just
671 * remove them from the object.
673 lwkt_gettoken(&vm_token);
674 vm_page_rb_tree_RB_SCAN(&object->rb_memq, NULL,
675 vm_object_terminate_callback, NULL);
676 lwkt_reltoken(&vm_token);
679 * Let the pager know object is dead.
681 vm_pager_deallocate(object);
684 * Wait for the object hold count to hit zero, clean out pages as
687 lwkt_gettoken(&vm_token);
689 vm_object_hold_wait(object);
690 if (RB_ROOT(&object->rb_memq) == NULL)
692 kprintf("vm_object_terminate: Warning, object %p "
693 "still has %d pages\n",
694 object, object->resident_page_count);
695 vm_page_rb_tree_RB_SCAN(&object->rb_memq, NULL,
696 vm_object_terminate_callback, NULL);
698 lwkt_reltoken(&vm_token);
701 * There had better not be any pages left
703 KKASSERT(object->resident_page_count == 0);
706 * Remove the object from the global object list.
708 * (we are holding vmobj_token)
710 TAILQ_REMOVE(&vm_object_list, object, object_list);
712 vm_object_dead_wakeup(object);
713 vm_object_unlock(object);
715 if (object->ref_count != 0) {
716 panic("vm_object_terminate2: object with references, "
717 "ref_count=%d", object->ref_count);
721 * Free the space for the object.
723 zfree(obj_zone, object);
727 * The caller must hold vm_token.
730 vm_object_terminate_callback(vm_page_t p, void *data __unused)
732 if (p->busy || (p->flags & PG_BUSY))
733 panic("vm_object_terminate: freeing busy page %p", p);
734 if (p->wire_count == 0) {
737 mycpu->gd_cnt.v_pfree++;
739 if (p->queue != PQ_NONE)
740 kprintf("vm_object_terminate: Warning: Encountered wired page %p on queue %d\n", p, p->queue);
749 * The object is dead but still has an object<->pager association. Sleep
750 * and return. The caller typically retests the association in a loop.
752 * Must be called with the vmobj_token held.
755 vm_object_dead_sleep(vm_object_t object, const char *wmesg)
757 ASSERT_LWKT_TOKEN_HELD(&vmobj_token);
758 if (object->handle) {
759 vm_object_set_flag(object, OBJ_DEADWNT);
760 tsleep(object, 0, wmesg, 0);
761 /* object may be invalid after this point */
766 * Wakeup anyone waiting for the object<->pager disassociation on
769 * Must be called with the vmobj_token held.
772 vm_object_dead_wakeup(vm_object_t object)
774 ASSERT_LWKT_TOKEN_HELD(&vmobj_token);
775 if (object->flags & OBJ_DEADWNT) {
776 vm_object_clear_flag(object, OBJ_DEADWNT);
782 * Clean all dirty pages in the specified range of object. Leaves page
783 * on whatever queue it is currently on. If NOSYNC is set then do not
784 * write out pages with PG_NOSYNC set (originally comes from MAP_NOSYNC),
785 * leaving the object dirty.
787 * When stuffing pages asynchronously, allow clustering. XXX we need a
788 * synchronous clustering mode implementation.
790 * Odd semantics: if start == end, we clean everything.
792 * The object must be locked? XXX
794 static int vm_object_page_clean_pass1(struct vm_page *p, void *data);
795 static int vm_object_page_clean_pass2(struct vm_page *p, void *data);
798 vm_object_page_clean(vm_object_t object, vm_pindex_t start, vm_pindex_t end,
801 struct rb_vm_page_scan_info info;
807 vm_object_hold(object);
808 if (object->type != OBJT_VNODE ||
809 (object->flags & OBJ_MIGHTBEDIRTY) == 0) {
810 vm_object_drop(object);
814 pagerflags = (flags & (OBJPC_SYNC | OBJPC_INVAL)) ?
815 VM_PAGER_PUT_SYNC : VM_PAGER_CLUSTER_OK;
816 pagerflags |= (flags & OBJPC_INVAL) ? VM_PAGER_PUT_INVAL : 0;
821 * Interlock other major object operations. This allows us to
822 * temporarily clear OBJ_WRITEABLE and OBJ_MIGHTBEDIRTY.
825 vm_object_set_flag(object, OBJ_CLEANING);
828 * Handle 'entire object' case
830 info.start_pindex = start;
832 info.end_pindex = object->size - 1;
834 info.end_pindex = end - 1;
836 wholescan = (start == 0 && info.end_pindex == object->size - 1);
838 info.pagerflags = pagerflags;
839 info.object = object;
842 * If cleaning the entire object do a pass to mark the pages read-only.
843 * If everything worked out ok, clear OBJ_WRITEABLE and
848 lwkt_gettoken(&vm_token);
849 vm_page_rb_tree_RB_SCAN(&object->rb_memq, rb_vm_page_scancmp,
850 vm_object_page_clean_pass1, &info);
851 lwkt_reltoken(&vm_token);
852 if (info.error == 0) {
853 vm_object_clear_flag(object,
854 OBJ_WRITEABLE|OBJ_MIGHTBEDIRTY);
855 if (object->type == OBJT_VNODE &&
856 (vp = (struct vnode *)object->handle) != NULL) {
857 if (vp->v_flag & VOBJDIRTY)
858 vclrflags(vp, VOBJDIRTY);
864 * Do a pass to clean all the dirty pages we find.
868 curgeneration = object->generation;
869 lwkt_gettoken(&vm_token);
870 vm_page_rb_tree_RB_SCAN(&object->rb_memq, rb_vm_page_scancmp,
871 vm_object_page_clean_pass2, &info);
872 lwkt_reltoken(&vm_token);
873 } while (info.error || curgeneration != object->generation);
875 vm_object_clear_flag(object, OBJ_CLEANING);
877 vm_object_drop(object);
881 * The caller must hold vm_token.
885 vm_object_page_clean_pass1(struct vm_page *p, void *data)
887 struct rb_vm_page_scan_info *info = data;
889 vm_page_flag_set(p, PG_CLEANCHK);
890 if ((info->limit & OBJPC_NOSYNC) && (p->flags & PG_NOSYNC))
893 vm_page_protect(p, VM_PROT_READ); /* must not block */
898 * The caller must hold vm_token.
902 vm_object_page_clean_pass2(struct vm_page *p, void *data)
904 struct rb_vm_page_scan_info *info = data;
908 * Do not mess with pages that were inserted after we started
911 if ((p->flags & PG_CLEANCHK) == 0)
915 * Before wasting time traversing the pmaps, check for trivial
916 * cases where the page cannot be dirty.
918 if (p->valid == 0 || (p->queue - p->pc) == PQ_CACHE) {
919 KKASSERT((p->dirty & p->valid) == 0);
924 * Check whether the page is dirty or not. The page has been set
925 * to be read-only so the check will not race a user dirtying the
928 vm_page_test_dirty(p);
929 if ((p->dirty & p->valid) == 0) {
930 vm_page_flag_clear(p, PG_CLEANCHK);
935 * If we have been asked to skip nosync pages and this is a
936 * nosync page, skip it. Note that the object flags were
937 * not cleared in this case (because pass1 will have returned an
938 * error), so we do not have to set them.
940 if ((info->limit & OBJPC_NOSYNC) && (p->flags & PG_NOSYNC)) {
941 vm_page_flag_clear(p, PG_CLEANCHK);
946 * Flush as many pages as we can. PG_CLEANCHK will be cleared on
947 * the pages that get successfully flushed. Set info->error if
948 * we raced an object modification.
950 n = vm_object_page_collect_flush(info->object, p, info->pagerflags);
957 * Collect the specified page and nearby pages and flush them out.
958 * The number of pages flushed is returned.
960 * The caller must hold vm_token.
963 vm_object_page_collect_flush(vm_object_t object, vm_page_t p, int pagerflags)
972 vm_page_t maf[vm_pageout_page_count];
973 vm_page_t mab[vm_pageout_page_count];
974 vm_page_t ma[vm_pageout_page_count];
976 curgeneration = object->generation;
979 while (vm_page_sleep_busy(p, TRUE, "vpcwai")) {
980 if (object->generation != curgeneration) {
984 KKASSERT(p->object == object && p->pindex == pi);
987 for(i = 1; i < vm_pageout_page_count; i++) {
990 if ((tp = vm_page_lookup(object, pi + i)) != NULL) {
991 if ((tp->flags & PG_BUSY) ||
992 ((pagerflags & VM_PAGER_IGNORE_CLEANCHK) == 0 &&
993 (tp->flags & PG_CLEANCHK) == 0) ||
996 if((tp->queue - tp->pc) == PQ_CACHE) {
997 vm_page_flag_clear(tp, PG_CLEANCHK);
1000 vm_page_test_dirty(tp);
1001 if ((tp->dirty & tp->valid) == 0) {
1002 vm_page_flag_clear(tp, PG_CLEANCHK);
1013 chkb = vm_pageout_page_count - maxf;
1015 for(i = 1; i < chkb;i++) {
1018 if ((tp = vm_page_lookup(object, pi - i)) != NULL) {
1019 if ((tp->flags & PG_BUSY) ||
1020 ((pagerflags & VM_PAGER_IGNORE_CLEANCHK) == 0 &&
1021 (tp->flags & PG_CLEANCHK) == 0) ||
1024 if((tp->queue - tp->pc) == PQ_CACHE) {
1025 vm_page_flag_clear(tp, PG_CLEANCHK);
1028 vm_page_test_dirty(tp);
1029 if ((tp->dirty & tp->valid) == 0) {
1030 vm_page_flag_clear(tp, PG_CLEANCHK);
1041 for(i = 0; i < maxb; i++) {
1042 int index = (maxb - i) - 1;
1044 vm_page_flag_clear(ma[index], PG_CLEANCHK);
1046 vm_page_flag_clear(p, PG_CLEANCHK);
1048 for(i = 0; i < maxf; i++) {
1049 int index = (maxb + i) + 1;
1051 vm_page_flag_clear(ma[index], PG_CLEANCHK);
1053 runlen = maxb + maxf + 1;
1055 vm_pageout_flush(ma, runlen, pagerflags);
1056 for (i = 0; i < runlen; i++) {
1057 if (ma[i]->valid & ma[i]->dirty) {
1058 vm_page_protect(ma[i], VM_PROT_READ);
1059 vm_page_flag_set(ma[i], PG_CLEANCHK);
1062 * maxf will end up being the actual number of pages
1063 * we wrote out contiguously, non-inclusive of the
1064 * first page. We do not count look-behind pages.
1066 if (i >= maxb + 1 && (maxf > i - maxb - 1))
1067 maxf = i - maxb - 1;
1074 * Same as vm_object_pmap_copy, except range checking really
1075 * works, and is meant for small sections of an object.
1077 * This code protects resident pages by making them read-only
1078 * and is typically called on a fork or split when a page
1079 * is converted to copy-on-write.
1081 * NOTE: If the page is already at VM_PROT_NONE, calling
1082 * vm_page_protect will have no effect.
1085 vm_object_pmap_copy_1(vm_object_t object, vm_pindex_t start, vm_pindex_t end)
1090 if (object == NULL || (object->flags & OBJ_WRITEABLE) == 0)
1094 * spl protection needed to prevent races between the lookup,
1095 * an interrupt unbusy/free, and our protect call.
1098 lwkt_gettoken(&vm_token);
1099 for (idx = start; idx < end; idx++) {
1100 p = vm_page_lookup(object, idx);
1103 vm_page_protect(p, VM_PROT_READ);
1105 lwkt_reltoken(&vm_token);
1110 * Removes all physical pages in the specified object range from all
1113 * The object must *not* be locked.
1116 static int vm_object_pmap_remove_callback(vm_page_t p, void *data);
1119 vm_object_pmap_remove(vm_object_t object, vm_pindex_t start, vm_pindex_t end)
1121 struct rb_vm_page_scan_info info;
1125 info.start_pindex = start;
1126 info.end_pindex = end - 1;
1129 lwkt_gettoken(&vm_token);
1130 vm_page_rb_tree_RB_SCAN(&object->rb_memq, rb_vm_page_scancmp,
1131 vm_object_pmap_remove_callback, &info);
1132 if (start == 0 && end == object->size)
1133 vm_object_clear_flag(object, OBJ_WRITEABLE);
1134 lwkt_reltoken(&vm_token);
1139 * The caller must hold vm_token.
1142 vm_object_pmap_remove_callback(vm_page_t p, void *data __unused)
1144 vm_page_protect(p, VM_PROT_NONE);
1149 * Implements the madvise function at the object/page level.
1151 * MADV_WILLNEED (any object)
1153 * Activate the specified pages if they are resident.
1155 * MADV_DONTNEED (any object)
1157 * Deactivate the specified pages if they are resident.
1159 * MADV_FREE (OBJT_DEFAULT/OBJT_SWAP objects, OBJ_ONEMAPPING only)
1161 * Deactivate and clean the specified pages if they are
1162 * resident. This permits the process to reuse the pages
1163 * without faulting or the kernel to reclaim the pages
1169 vm_object_madvise(vm_object_t object, vm_pindex_t pindex, int count, int advise)
1171 vm_pindex_t end, tpindex;
1172 vm_object_t tobject;
1178 end = pindex + count;
1180 lwkt_gettoken(&vm_token);
1183 * Locate and adjust resident pages
1185 for (; pindex < end; pindex += 1) {
1191 * MADV_FREE only operates on OBJT_DEFAULT or OBJT_SWAP pages
1192 * and those pages must be OBJ_ONEMAPPING.
1194 if (advise == MADV_FREE) {
1195 if ((tobject->type != OBJT_DEFAULT &&
1196 tobject->type != OBJT_SWAP) ||
1197 (tobject->flags & OBJ_ONEMAPPING) == 0) {
1203 * spl protection is required to avoid a race between the
1204 * lookup, an interrupt unbusy/free, and our busy check.
1208 m = vm_page_lookup(tobject, tpindex);
1212 * There may be swap even if there is no backing page
1214 if (advise == MADV_FREE && tobject->type == OBJT_SWAP)
1215 swap_pager_freespace(tobject, tpindex, 1);
1221 if (tobject->backing_object == NULL)
1223 tpindex += OFF_TO_IDX(tobject->backing_object_offset);
1224 tobject = tobject->backing_object;
1229 * If the page is busy or not in a normal active state,
1230 * we skip it. If the page is not managed there are no
1231 * page queues to mess with. Things can break if we mess
1232 * with pages in any of the below states.
1237 (m->flags & PG_UNMANAGED) ||
1238 m->valid != VM_PAGE_BITS_ALL
1244 if (vm_page_sleep_busy(m, TRUE, "madvpo")) {
1252 * Theoretically once a page is known not to be busy, an
1253 * interrupt cannot come along and rip it out from under us.
1256 if (advise == MADV_WILLNEED) {
1257 vm_page_activate(m);
1258 } else if (advise == MADV_DONTNEED) {
1259 vm_page_dontneed(m);
1260 } else if (advise == MADV_FREE) {
1262 * Mark the page clean. This will allow the page
1263 * to be freed up by the system. However, such pages
1264 * are often reused quickly by malloc()/free()
1265 * so we do not do anything that would cause
1266 * a page fault if we can help it.
1268 * Specifically, we do not try to actually free
1269 * the page now nor do we try to put it in the
1270 * cache (which would cause a page fault on reuse).
1272 * But we do make the page is freeable as we
1273 * can without actually taking the step of unmapping
1276 pmap_clear_modify(m);
1279 vm_page_dontneed(m);
1280 if (tobject->type == OBJT_SWAP)
1281 swap_pager_freespace(tobject, tpindex, 1);
1285 lwkt_reltoken(&vm_token);
1289 * Create a new object which is backed by the specified existing object
1290 * range. The source object reference is deallocated.
1292 * The new object and offset into that object are returned in the source
1295 * No other requirements.
1298 vm_object_shadow(vm_object_t *object, vm_ooffset_t *offset, vm_size_t length)
1306 * Don't create the new object if the old object isn't shared.
1308 lwkt_gettoken(&vm_token);
1310 if (source != NULL &&
1311 source->ref_count == 1 &&
1312 source->handle == NULL &&
1313 (source->type == OBJT_DEFAULT ||
1314 source->type == OBJT_SWAP)) {
1315 lwkt_reltoken(&vm_token);
1320 * Allocate a new object with the given length
1323 if ((result = vm_object_allocate(OBJT_DEFAULT, length)) == NULL)
1324 panic("vm_object_shadow: no object for shadowing");
1327 * The new object shadows the source object, adding a reference to it.
1328 * Our caller changes his reference to point to the new object,
1329 * removing a reference to the source object. Net result: no change
1330 * of reference count.
1332 * Try to optimize the result object's page color when shadowing
1333 * in order to maintain page coloring consistency in the combined
1336 result->backing_object = source;
1338 LIST_INSERT_HEAD(&source->shadow_head, result, shadow_list);
1339 source->shadow_count++;
1340 source->generation++;
1341 result->pg_color = (source->pg_color + OFF_TO_IDX(*offset)) & PQ_L2_MASK;
1345 * Store the offset into the source object, and fix up the offset into
1348 result->backing_object_offset = *offset;
1349 lwkt_reltoken(&vm_token);
1352 * Return the new things
1358 #define OBSC_TEST_ALL_SHADOWED 0x0001
1359 #define OBSC_COLLAPSE_NOWAIT 0x0002
1360 #define OBSC_COLLAPSE_WAIT 0x0004
1362 static int vm_object_backing_scan_callback(vm_page_t p, void *data);
1365 * The caller must hold vm_token.
1368 vm_object_backing_scan(vm_object_t object, int op)
1370 struct rb_vm_page_scan_info info;
1371 vm_object_t backing_object;
1375 backing_object = object->backing_object;
1376 info.backing_offset_index = OFF_TO_IDX(object->backing_object_offset);
1379 * Initial conditions
1382 if (op & OBSC_TEST_ALL_SHADOWED) {
1384 * We do not want to have to test for the existence of
1385 * swap pages in the backing object. XXX but with the
1386 * new swapper this would be pretty easy to do.
1388 * XXX what about anonymous MAP_SHARED memory that hasn't
1389 * been ZFOD faulted yet? If we do not test for this, the
1390 * shadow test may succeed! XXX
1392 if (backing_object->type != OBJT_DEFAULT) {
1397 if (op & OBSC_COLLAPSE_WAIT) {
1398 KKASSERT((backing_object->flags & OBJ_DEAD) == 0);
1399 vm_object_set_flag(backing_object, OBJ_DEAD);
1403 * Our scan. We have to retry if a negative error code is returned,
1404 * otherwise 0 or 1 will be returned in info.error. 0 Indicates that
1405 * the scan had to be stopped because the parent does not completely
1408 info.object = object;
1409 info.backing_object = backing_object;
1413 vm_page_rb_tree_RB_SCAN(&backing_object->rb_memq, NULL,
1414 vm_object_backing_scan_callback,
1416 } while (info.error < 0);
1422 * The caller must hold vm_token.
1425 vm_object_backing_scan_callback(vm_page_t p, void *data)
1427 struct rb_vm_page_scan_info *info = data;
1428 vm_object_t backing_object;
1430 vm_pindex_t new_pindex;
1431 vm_pindex_t backing_offset_index;
1434 new_pindex = p->pindex - info->backing_offset_index;
1436 object = info->object;
1437 backing_object = info->backing_object;
1438 backing_offset_index = info->backing_offset_index;
1440 if (op & OBSC_TEST_ALL_SHADOWED) {
1444 * Ignore pages outside the parent object's range
1445 * and outside the parent object's mapping of the
1448 * note that we do not busy the backing object's
1452 p->pindex < backing_offset_index ||
1453 new_pindex >= object->size
1459 * See if the parent has the page or if the parent's
1460 * object pager has the page. If the parent has the
1461 * page but the page is not valid, the parent's
1462 * object pager must have the page.
1464 * If this fails, the parent does not completely shadow
1465 * the object and we might as well give up now.
1468 pp = vm_page_lookup(object, new_pindex);
1469 if ((pp == NULL || pp->valid == 0) &&
1470 !vm_pager_has_page(object, new_pindex)
1472 info->error = 0; /* problemo */
1473 return(-1); /* stop the scan */
1478 * Check for busy page
1481 if (op & (OBSC_COLLAPSE_WAIT | OBSC_COLLAPSE_NOWAIT)) {
1484 if (op & OBSC_COLLAPSE_NOWAIT) {
1486 (p->flags & PG_BUSY) ||
1494 } else if (op & OBSC_COLLAPSE_WAIT) {
1495 if (vm_page_sleep_busy(p, TRUE, "vmocol")) {
1497 * If we slept, anything could have
1498 * happened. Ask that the scan be restarted.
1500 * Since the object is marked dead, the
1501 * backing offset should not have changed.
1514 p->object == backing_object,
1515 ("vm_object_qcollapse(): object mismatch")
1519 * Destroy any associated swap
1521 if (backing_object->type == OBJT_SWAP)
1522 swap_pager_freespace(backing_object, p->pindex, 1);
1525 p->pindex < backing_offset_index ||
1526 new_pindex >= object->size
1529 * Page is out of the parent object's range, we
1530 * can simply destroy it.
1532 vm_page_protect(p, VM_PROT_NONE);
1537 pp = vm_page_lookup(object, new_pindex);
1538 if (pp != NULL || vm_pager_has_page(object, new_pindex)) {
1540 * page already exists in parent OR swap exists
1541 * for this location in the parent. Destroy
1542 * the original page from the backing object.
1544 * Leave the parent's page alone
1546 vm_page_protect(p, VM_PROT_NONE);
1552 * Page does not exist in parent, rename the
1553 * page from the backing object to the main object.
1555 * If the page was mapped to a process, it can remain
1556 * mapped through the rename.
1558 if ((p->queue - p->pc) == PQ_CACHE)
1559 vm_page_deactivate(p);
1561 vm_page_rename(p, object, new_pindex);
1562 /* page automatically made dirty by rename */
1568 * This version of collapse allows the operation to occur earlier and
1569 * when paging_in_progress is true for an object... This is not a complete
1570 * operation, but should plug 99.9% of the rest of the leaks.
1572 * The caller must hold vm_token and vmobj_token.
1573 * (only called from vm_object_collapse)
1576 vm_object_qcollapse(vm_object_t object)
1578 vm_object_t backing_object = object->backing_object;
1580 if (backing_object->ref_count != 1)
1583 backing_object->ref_count += 2;
1585 vm_object_backing_scan(object, OBSC_COLLAPSE_NOWAIT);
1587 backing_object->ref_count -= 2;
1591 * Collapse an object with the object backing it. Pages in the backing
1592 * object are moved into the parent, and the backing object is deallocated.
1594 * The caller must hold (object).
1597 vm_object_collapse(vm_object_t object)
1599 ASSERT_LWKT_TOKEN_HELD(&vm_token);
1600 ASSERT_LWKT_TOKEN_HELD(&vmobj_token);
1601 vm_object_assert_held(object);
1604 vm_object_t backing_object;
1607 * Verify that the conditions are right for collapse:
1609 * The object exists and the backing object exists.
1614 if ((backing_object = object->backing_object) == NULL)
1617 vm_object_hold(backing_object);
1618 if (backing_object != object->backing_object) {
1619 vm_object_drop(backing_object);
1624 * we check the backing object first, because it is most likely
1627 if (backing_object->handle != NULL ||
1628 (backing_object->type != OBJT_DEFAULT &&
1629 backing_object->type != OBJT_SWAP) ||
1630 (backing_object->flags & OBJ_DEAD) ||
1631 object->handle != NULL ||
1632 (object->type != OBJT_DEFAULT &&
1633 object->type != OBJT_SWAP) ||
1634 (object->flags & OBJ_DEAD)) {
1635 vm_object_drop(backing_object);
1640 object->paging_in_progress != 0 ||
1641 backing_object->paging_in_progress != 0
1643 vm_object_drop(backing_object);
1644 vm_object_qcollapse(object);
1649 * We know that we can either collapse the backing object (if
1650 * the parent is the only reference to it) or (perhaps) have
1651 * the parent bypass the object if the parent happens to shadow
1652 * all the resident pages in the entire backing object.
1654 * This is ignoring pager-backed pages such as swap pages.
1655 * vm_object_backing_scan fails the shadowing test in this
1659 if (backing_object->ref_count == 1) {
1661 * If there is exactly one reference to the backing
1662 * object, we can collapse it into the parent.
1664 vm_object_backing_scan(object, OBSC_COLLAPSE_WAIT);
1667 * Move the pager from backing_object to object.
1670 if (backing_object->type == OBJT_SWAP) {
1671 vm_object_pip_add(backing_object, 1);
1674 * scrap the paging_offset junk and do a
1675 * discrete copy. This also removes major
1676 * assumptions about how the swap-pager
1677 * works from where it doesn't belong. The
1678 * new swapper is able to optimize the
1679 * destroy-source case.
1682 vm_object_pip_add(object, 1);
1686 OFF_TO_IDX(object->backing_object_offset), TRUE);
1687 vm_object_pip_wakeup(object);
1689 vm_object_pip_wakeup(backing_object);
1692 * Object now shadows whatever backing_object did.
1693 * Note that the reference to
1694 * backing_object->backing_object moves from within
1695 * backing_object to within object.
1698 LIST_REMOVE(object, shadow_list);
1699 object->backing_object->shadow_count--;
1700 object->backing_object->generation++;
1701 if (backing_object->backing_object) {
1702 LIST_REMOVE(backing_object, shadow_list);
1703 backing_object->backing_object->shadow_count--;
1704 backing_object->backing_object->generation++;
1706 object->backing_object = backing_object->backing_object;
1707 if (object->backing_object) {
1709 &object->backing_object->shadow_head,
1713 object->backing_object->shadow_count++;
1714 object->backing_object->generation++;
1717 object->backing_object_offset +=
1718 backing_object->backing_object_offset;
1721 * Discard backing_object.
1723 * Since the backing object has no pages, no pager left,
1724 * and no object references within it, all that is
1725 * necessary is to dispose of it.
1728 KASSERT(backing_object->ref_count == 1,
1729 ("backing_object %p was somehow "
1730 "re-referenced during collapse!",
1732 KASSERT(RB_EMPTY(&backing_object->rb_memq),
1733 ("backing_object %p somehow has left "
1734 "over pages during collapse!",
1738 * Wait for hold count to hit zero
1740 vm_object_drop(backing_object);
1741 vm_object_hold_wait(backing_object);
1743 /* (we are holding vmobj_token) */
1744 TAILQ_REMOVE(&vm_object_list, backing_object,
1748 zfree(obj_zone, backing_object);
1752 vm_object_t new_backing_object;
1755 * If we do not entirely shadow the backing object,
1756 * there is nothing we can do so we give up.
1759 if (vm_object_backing_scan(object, OBSC_TEST_ALL_SHADOWED) == 0) {
1760 vm_object_drop(backing_object);
1765 * Make the parent shadow the next object in the
1766 * chain. Deallocating backing_object will not remove
1767 * it, since its reference count is at least 2.
1770 LIST_REMOVE(object, shadow_list);
1771 backing_object->shadow_count--;
1772 backing_object->generation++;
1774 new_backing_object = backing_object->backing_object;
1775 if ((object->backing_object = new_backing_object) != NULL) {
1776 vm_object_reference(new_backing_object);
1778 &new_backing_object->shadow_head,
1782 new_backing_object->shadow_count++;
1783 new_backing_object->generation++;
1784 object->backing_object_offset +=
1785 backing_object->backing_object_offset;
1789 * Drop the reference count on backing_object. Since
1790 * its ref_count was at least 2, it will not vanish;
1791 * so we don't need to call vm_object_deallocate, but
1794 vm_object_drop(backing_object);
1795 vm_object_deallocate_locked(backing_object);
1800 * Try again with this object's new backing object.
1806 * Removes all physical pages in the specified object range from the
1807 * object's list of pages.
1811 static int vm_object_page_remove_callback(vm_page_t p, void *data);
1814 vm_object_page_remove(vm_object_t object, vm_pindex_t start, vm_pindex_t end,
1815 boolean_t clean_only)
1817 struct rb_vm_page_scan_info info;
1821 * Degenerate cases and assertions
1823 lwkt_gettoken(&vm_token);
1824 if (object == NULL ||
1825 (object->resident_page_count == 0 && object->swblock_count == 0)) {
1826 lwkt_reltoken(&vm_token);
1829 KASSERT(object->type != OBJT_PHYS,
1830 ("attempt to remove pages from a physical object"));
1833 * Indicate that paging is occuring on the object
1836 vm_object_pip_add(object, 1);
1839 * Figure out the actual removal range and whether we are removing
1840 * the entire contents of the object or not. If removing the entire
1841 * contents, be sure to get all pages, even those that might be
1842 * beyond the end of the object.
1844 info.start_pindex = start;
1846 info.end_pindex = (vm_pindex_t)-1;
1848 info.end_pindex = end - 1;
1849 info.limit = clean_only;
1850 all = (start == 0 && info.end_pindex >= object->size - 1);
1853 * Loop until we are sure we have gotten them all.
1857 vm_page_rb_tree_RB_SCAN(&object->rb_memq, rb_vm_page_scancmp,
1858 vm_object_page_remove_callback, &info);
1859 } while (info.error);
1862 * Remove any related swap if throwing away pages, or for
1863 * non-swap objects (the swap is a clean copy in that case).
1865 if (object->type != OBJT_SWAP || clean_only == FALSE) {
1867 swap_pager_freespace_all(object);
1869 swap_pager_freespace(object, info.start_pindex,
1870 info.end_pindex - info.start_pindex + 1);
1876 vm_object_pip_wakeup(object);
1878 lwkt_reltoken(&vm_token);
1882 * The caller must hold vm_token.
1885 vm_object_page_remove_callback(vm_page_t p, void *data)
1887 struct rb_vm_page_scan_info *info = data;
1890 * Wired pages cannot be destroyed, but they can be invalidated
1891 * and we do so if clean_only (limit) is not set.
1893 * WARNING! The page may be wired due to being part of a buffer
1894 * cache buffer, and the buffer might be marked B_CACHE.
1895 * This is fine as part of a truncation but VFSs must be
1896 * sure to fix the buffer up when re-extending the file.
1898 if (p->wire_count != 0) {
1899 vm_page_protect(p, VM_PROT_NONE);
1900 if (info->limit == 0)
1906 * The busy flags are only cleared at
1907 * interrupt -- minimize the spl transitions
1910 if (vm_page_sleep_busy(p, TRUE, "vmopar")) {
1916 * limit is our clean_only flag. If set and the page is dirty, do
1917 * not free it. If set and the page is being held by someone, do
1920 if (info->limit && p->valid) {
1921 vm_page_test_dirty(p);
1922 if (p->valid & p->dirty)
1932 vm_page_protect(p, VM_PROT_NONE);
1938 * Coalesces two objects backing up adjoining regions of memory into a
1941 * returns TRUE if objects were combined.
1943 * NOTE: Only works at the moment if the second object is NULL -
1944 * if it's not, which object do we lock first?
1947 * prev_object First object to coalesce
1948 * prev_offset Offset into prev_object
1949 * next_object Second object into coalesce
1950 * next_offset Offset into next_object
1952 * prev_size Size of reference to prev_object
1953 * next_size Size of reference to next_object
1955 * The caller must hold vm_token and vmobj_token.
1957 * The caller does not need to hold (prev_object) but must have a stable
1958 * pointer to it (typically by holding the vm_map locked).
1961 vm_object_coalesce(vm_object_t prev_object, vm_pindex_t prev_pindex,
1962 vm_size_t prev_size, vm_size_t next_size)
1964 vm_pindex_t next_pindex;
1966 ASSERT_LWKT_TOKEN_HELD(&vm_token);
1967 ASSERT_LWKT_TOKEN_HELD(&vmobj_token);
1969 if (prev_object == NULL) {
1973 vm_object_hold(prev_object);
1975 if (prev_object->type != OBJT_DEFAULT &&
1976 prev_object->type != OBJT_SWAP) {
1977 vm_object_drop(prev_object);
1982 * Try to collapse the object first
1984 vm_object_collapse(prev_object);
1987 * Can't coalesce if: . more than one reference . paged out . shadows
1988 * another object . has a copy elsewhere (any of which mean that the
1989 * pages not mapped to prev_entry may be in use anyway)
1992 if (prev_object->backing_object != NULL) {
1993 vm_object_drop(prev_object);
1997 prev_size >>= PAGE_SHIFT;
1998 next_size >>= PAGE_SHIFT;
1999 next_pindex = prev_pindex + prev_size;
2001 if ((prev_object->ref_count > 1) &&
2002 (prev_object->size != next_pindex)) {
2003 vm_object_drop(prev_object);
2008 * Remove any pages that may still be in the object from a previous
2011 if (next_pindex < prev_object->size) {
2012 vm_object_page_remove(prev_object,
2014 next_pindex + next_size, FALSE);
2015 if (prev_object->type == OBJT_SWAP)
2016 swap_pager_freespace(prev_object,
2017 next_pindex, next_size);
2021 * Extend the object if necessary.
2023 if (next_pindex + next_size > prev_object->size)
2024 prev_object->size = next_pindex + next_size;
2026 vm_object_drop(prev_object);
2031 * Make the object writable and flag is being possibly dirty.
2036 vm_object_set_writeable_dirty(vm_object_t object)
2040 lwkt_gettoken(&vm_token);
2041 vm_object_set_flag(object, OBJ_WRITEABLE|OBJ_MIGHTBEDIRTY);
2042 if (object->type == OBJT_VNODE &&
2043 (vp = (struct vnode *)object->handle) != NULL) {
2044 if ((vp->v_flag & VOBJDIRTY) == 0) {
2045 vsetflags(vp, VOBJDIRTY);
2048 lwkt_reltoken(&vm_token);
2051 #include "opt_ddb.h"
2053 #include <sys/kernel.h>
2055 #include <sys/cons.h>
2057 #include <ddb/ddb.h>
2059 static int _vm_object_in_map (vm_map_t map, vm_object_t object,
2060 vm_map_entry_t entry);
2061 static int vm_object_in_map (vm_object_t object);
2064 * The caller must hold vm_token.
2067 _vm_object_in_map(vm_map_t map, vm_object_t object, vm_map_entry_t entry)
2070 vm_map_entry_t tmpe;
2077 tmpe = map->header.next;
2078 entcount = map->nentries;
2079 while (entcount-- && (tmpe != &map->header)) {
2080 if( _vm_object_in_map(map, object, tmpe)) {
2087 switch(entry->maptype) {
2088 case VM_MAPTYPE_SUBMAP:
2089 tmpm = entry->object.sub_map;
2090 tmpe = tmpm->header.next;
2091 entcount = tmpm->nentries;
2092 while (entcount-- && tmpe != &tmpm->header) {
2093 if( _vm_object_in_map(tmpm, object, tmpe)) {
2099 case VM_MAPTYPE_NORMAL:
2100 case VM_MAPTYPE_VPAGETABLE:
2101 obj = entry->object.vm_object;
2105 obj = obj->backing_object;
2114 static int vm_object_in_map_callback(struct proc *p, void *data);
2116 struct vm_object_in_map_info {
2125 vm_object_in_map(vm_object_t object)
2127 struct vm_object_in_map_info info;
2130 info.object = object;
2132 allproc_scan(vm_object_in_map_callback, &info);
2135 if( _vm_object_in_map(&kernel_map, object, 0))
2137 if( _vm_object_in_map(&pager_map, object, 0))
2139 if( _vm_object_in_map(&buffer_map, object, 0))
2148 vm_object_in_map_callback(struct proc *p, void *data)
2150 struct vm_object_in_map_info *info = data;
2153 if (_vm_object_in_map(&p->p_vmspace->vm_map, info->object, 0)) {
2161 DB_SHOW_COMMAND(vmochk, vm_object_check)
2166 * make sure that internal objs are in a map somewhere
2167 * and none have zero ref counts.
2169 for (object = TAILQ_FIRST(&vm_object_list);
2171 object = TAILQ_NEXT(object, object_list)) {
2172 if (object->type == OBJT_MARKER)
2174 if (object->handle == NULL &&
2175 (object->type == OBJT_DEFAULT || object->type == OBJT_SWAP)) {
2176 if (object->ref_count == 0) {
2177 db_printf("vmochk: internal obj has zero ref count: %ld\n",
2178 (long)object->size);
2180 if (!vm_object_in_map(object)) {
2182 "vmochk: internal obj is not in a map: "
2183 "ref: %d, size: %lu: 0x%lx, backing_object: %p\n",
2184 object->ref_count, (u_long)object->size,
2185 (u_long)object->size,
2186 (void *)object->backing_object);
2195 DB_SHOW_COMMAND(object, vm_object_print_static)
2197 /* XXX convert args. */
2198 vm_object_t object = (vm_object_t)addr;
2199 boolean_t full = have_addr;
2203 /* XXX count is an (unused) arg. Avoid shadowing it. */
2204 #define count was_count
2212 "Object %p: type=%d, size=0x%lx, res=%d, ref=%d, flags=0x%x\n",
2213 object, (int)object->type, (u_long)object->size,
2214 object->resident_page_count, object->ref_count, object->flags);
2216 * XXX no %qd in kernel. Truncate object->backing_object_offset.
2218 db_iprintf(" sref=%d, backing_object(%d)=(%p)+0x%lx\n",
2219 object->shadow_count,
2220 object->backing_object ? object->backing_object->ref_count : 0,
2221 object->backing_object, (long)object->backing_object_offset);
2228 RB_FOREACH(p, vm_page_rb_tree, &object->rb_memq) {
2230 db_iprintf("memory:=");
2231 else if (count == 6) {
2239 db_printf("(off=0x%lx,page=0x%lx)",
2240 (u_long) p->pindex, (u_long) VM_PAGE_TO_PHYS(p));
2251 * XXX need this non-static entry for calling from vm_map_print.
2256 vm_object_print(/* db_expr_t */ long addr,
2257 boolean_t have_addr,
2258 /* db_expr_t */ long count,
2261 vm_object_print_static(addr, have_addr, count, modif);
2267 DB_SHOW_COMMAND(vmopag, vm_object_print_pages)
2272 for (object = TAILQ_FIRST(&vm_object_list);
2274 object = TAILQ_NEXT(object, object_list)) {
2275 vm_pindex_t idx, fidx;
2277 vm_paddr_t pa = -1, padiff;
2281 if (object->type == OBJT_MARKER)
2283 db_printf("new object: %p\n", (void *)object);
2293 osize = object->size;
2296 for (idx = 0; idx < osize; idx++) {
2297 m = vm_page_lookup(object, idx);
2300 db_printf(" index(%ld)run(%d)pa(0x%lx)\n",
2301 (long)fidx, rcount, (long)pa);
2316 (VM_PAGE_TO_PHYS(m) == pa + rcount * PAGE_SIZE)) {
2321 padiff = pa + rcount * PAGE_SIZE - VM_PAGE_TO_PHYS(m);
2322 padiff >>= PAGE_SHIFT;
2323 padiff &= PQ_L2_MASK;
2325 pa = VM_PAGE_TO_PHYS(m) - rcount * PAGE_SIZE;
2329 db_printf(" index(%ld)run(%d)pa(0x%lx)",
2330 (long)fidx, rcount, (long)pa);
2331 db_printf("pd(%ld)\n", (long)padiff);
2341 pa = VM_PAGE_TO_PHYS(m);
2345 db_printf(" index(%ld)run(%d)pa(0x%lx)\n",
2346 (long)fidx, rcount, (long)pa);