FreeBSD/Linux Kernel Cross Reference
sys/vm/vm_object.c
1 /*-
2 * Copyright (c) 1991, 1993
3 * The Regents of the University of California. All rights reserved.
4 *
5 * This code is derived from software contributed to Berkeley by
6 * The Mach Operating System project at Carnegie-Mellon University.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 4. Neither the name of the University nor the names of its contributors
17 * may be used to endorse or promote products derived from this software
18 * without specific prior written permission.
19 *
20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30 * SUCH DAMAGE.
31 *
32 * from: @(#)vm_object.c 8.5 (Berkeley) 3/22/94
33 *
34 *
35 * Copyright (c) 1987, 1990 Carnegie-Mellon University.
36 * All rights reserved.
37 *
38 * Authors: Avadis Tevanian, Jr., Michael Wayne Young
39 *
40 * Permission to use, copy, modify and distribute this software and
41 * its documentation is hereby granted, provided that both the copyright
42 * notice and this permission notice appear in all copies of the
43 * software, derivative works or modified versions, and any portions
44 * thereof, and that both notices appear in supporting documentation.
45 *
46 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
47 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
48 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
49 *
50 * Carnegie Mellon requests users of this software to return to
51 *
52 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
53 * School of Computer Science
54 * Carnegie Mellon University
55 * Pittsburgh PA 15213-3890
56 *
57 * any improvements or extensions that they make and grant Carnegie the
58 * rights to redistribute these changes.
59 */
60
61 /*
62 * Virtual memory object module.
63 */
64
65 #include <sys/cdefs.h>
66 __FBSDID("$FreeBSD: releng/8.3/sys/vm/vm_object.c 229725 2012-01-06 19:32:39Z jhb $");
67
68 #include "opt_vm.h"
69
70 #include <sys/param.h>
71 #include <sys/systm.h>
72 #include <sys/lock.h>
73 #include <sys/mman.h>
74 #include <sys/mount.h>
75 #include <sys/kernel.h>
76 #include <sys/sysctl.h>
77 #include <sys/mutex.h>
78 #include <sys/proc.h> /* for curproc, pageproc */
79 #include <sys/socket.h>
80 #include <sys/resourcevar.h>
81 #include <sys/vnode.h>
82 #include <sys/vmmeter.h>
83 #include <sys/sx.h>
84
85 #include <vm/vm.h>
86 #include <vm/vm_param.h>
87 #include <vm/pmap.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_reserv.h>
97 #include <vm/uma.h>
98
99 static int msync_flush_flags = 0;
100 SYSCTL_INT(_vm, OID_AUTO, msync_flush_flags, CTLFLAG_RW, &msync_flush_flags, 0,
101 "Does nothing; kept for backward compatibility");
102
103 static int old_msync;
104 SYSCTL_INT(_vm, OID_AUTO, old_msync, CTLFLAG_RW, &old_msync, 0,
105 "Use old (insecure) msync behavior");
106
107 static int vm_object_page_collect_flush(vm_object_t object, vm_page_t p,
108 int pagerflags, int flags, int *clearobjflags);
109 static boolean_t vm_object_page_remove_write(vm_page_t p, int flags,
110 int *clearobjflags);
111 static void vm_object_qcollapse(vm_object_t object);
112 static void vm_object_vndeallocate(vm_object_t object);
113
114 /*
115 * Virtual memory objects maintain the actual data
116 * associated with allocated virtual memory. A given
117 * page of memory exists within exactly one object.
118 *
119 * An object is only deallocated when all "references"
120 * are given up. Only one "reference" to a given
121 * region of an object should be writeable.
122 *
123 * Associated with each object is a list of all resident
124 * memory pages belonging to that object; this list is
125 * maintained by the "vm_page" module, and locked by the object's
126 * lock.
127 *
128 * Each object also records a "pager" routine which is
129 * used to retrieve (and store) pages to the proper backing
130 * storage. In addition, objects may be backed by other
131 * objects from which they were virtual-copied.
132 *
133 * The only items within the object structure which are
134 * modified after time of creation are:
135 * reference count locked by object's lock
136 * pager routine locked by object's lock
137 *
138 */
139
140 struct object_q vm_object_list;
141 struct mtx vm_object_list_mtx; /* lock for object list and count */
142
143 struct vm_object kernel_object_store;
144 struct vm_object kmem_object_store;
145
146 SYSCTL_NODE(_vm_stats, OID_AUTO, object, CTLFLAG_RD, 0, "VM object stats");
147
148 static long object_collapses;
149 SYSCTL_LONG(_vm_stats_object, OID_AUTO, collapses, CTLFLAG_RD,
150 &object_collapses, 0, "VM object collapses");
151
152 static long object_bypasses;
153 SYSCTL_LONG(_vm_stats_object, OID_AUTO, bypasses, CTLFLAG_RD,
154 &object_bypasses, 0, "VM object bypasses");
155
156 static uma_zone_t obj_zone;
157
158 static int vm_object_zinit(void *mem, int size, int flags);
159
160 #ifdef INVARIANTS
161 static void vm_object_zdtor(void *mem, int size, void *arg);
162
163 static void
164 vm_object_zdtor(void *mem, int size, void *arg)
165 {
166 vm_object_t object;
167
168 object = (vm_object_t)mem;
169 KASSERT(TAILQ_EMPTY(&object->memq),
170 ("object %p has resident pages",
171 object));
172 #if VM_NRESERVLEVEL > 0
173 KASSERT(LIST_EMPTY(&object->rvq),
174 ("object %p has reservations",
175 object));
176 #endif
177 KASSERT(object->cache == NULL,
178 ("object %p has cached pages",
179 object));
180 KASSERT(object->paging_in_progress == 0,
181 ("object %p paging_in_progress = %d",
182 object, object->paging_in_progress));
183 KASSERT(object->resident_page_count == 0,
184 ("object %p resident_page_count = %d",
185 object, object->resident_page_count));
186 KASSERT(object->shadow_count == 0,
187 ("object %p shadow_count = %d",
188 object, object->shadow_count));
189 }
190 #endif
191
192 static int
193 vm_object_zinit(void *mem, int size, int flags)
194 {
195 vm_object_t object;
196
197 object = (vm_object_t)mem;
198 bzero(&object->mtx, sizeof(object->mtx));
199 VM_OBJECT_LOCK_INIT(object, "standard object");
200
201 /* These are true for any object that has been freed */
202 object->paging_in_progress = 0;
203 object->resident_page_count = 0;
204 object->shadow_count = 0;
205 return (0);
206 }
207
208 void
209 _vm_object_allocate(objtype_t type, vm_pindex_t size, vm_object_t object)
210 {
211
212 TAILQ_INIT(&object->memq);
213 LIST_INIT(&object->shadow_head);
214
215 object->root = NULL;
216 object->type = type;
217 object->size = size;
218 object->generation = 1;
219 object->ref_count = 1;
220 object->memattr = VM_MEMATTR_DEFAULT;
221 object->flags = 0;
222 object->uip = NULL;
223 object->charge = 0;
224 if ((object->type == OBJT_DEFAULT) || (object->type == OBJT_SWAP))
225 object->flags = OBJ_ONEMAPPING;
226 object->pg_color = 0;
227 object->handle = NULL;
228 object->backing_object = NULL;
229 object->backing_object_offset = (vm_ooffset_t) 0;
230 #if VM_NRESERVLEVEL > 0
231 LIST_INIT(&object->rvq);
232 #endif
233 object->cache = NULL;
234
235 mtx_lock(&vm_object_list_mtx);
236 TAILQ_INSERT_TAIL(&vm_object_list, object, object_list);
237 mtx_unlock(&vm_object_list_mtx);
238 }
239
240 /*
241 * vm_object_init:
242 *
243 * Initialize the VM objects module.
244 */
245 void
246 vm_object_init(void)
247 {
248 TAILQ_INIT(&vm_object_list);
249 mtx_init(&vm_object_list_mtx, "vm object_list", NULL, MTX_DEF);
250
251 VM_OBJECT_LOCK_INIT(kernel_object, "kernel object");
252 _vm_object_allocate(OBJT_PHYS, OFF_TO_IDX(VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS),
253 kernel_object);
254 #if VM_NRESERVLEVEL > 0
255 kernel_object->flags |= OBJ_COLORED;
256 kernel_object->pg_color = (u_short)atop(VM_MIN_KERNEL_ADDRESS);
257 #endif
258
259 VM_OBJECT_LOCK_INIT(kmem_object, "kmem object");
260 _vm_object_allocate(OBJT_PHYS, OFF_TO_IDX(VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS),
261 kmem_object);
262 #if VM_NRESERVLEVEL > 0
263 kmem_object->flags |= OBJ_COLORED;
264 kmem_object->pg_color = (u_short)atop(VM_MIN_KERNEL_ADDRESS);
265 #endif
266
267 /*
268 * The lock portion of struct vm_object must be type stable due
269 * to vm_pageout_fallback_object_lock locking a vm object
270 * without holding any references to it.
271 */
272 obj_zone = uma_zcreate("VM OBJECT", sizeof (struct vm_object), NULL,
273 #ifdef INVARIANTS
274 vm_object_zdtor,
275 #else
276 NULL,
277 #endif
278 vm_object_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_VM|UMA_ZONE_NOFREE);
279 }
280
281 void
282 vm_object_clear_flag(vm_object_t object, u_short bits)
283 {
284
285 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
286 object->flags &= ~bits;
287 }
288
289 /*
290 * Sets the default memory attribute for the specified object. Pages
291 * that are allocated to this object are by default assigned this memory
292 * attribute.
293 *
294 * Presently, this function must be called before any pages are allocated
295 * to the object. In the future, this requirement may be relaxed for
296 * "default" and "swap" objects.
297 */
298 int
299 vm_object_set_memattr(vm_object_t object, vm_memattr_t memattr)
300 {
301
302 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
303 switch (object->type) {
304 case OBJT_DEFAULT:
305 case OBJT_DEVICE:
306 case OBJT_PHYS:
307 case OBJT_SG:
308 case OBJT_SWAP:
309 case OBJT_VNODE:
310 if (!TAILQ_EMPTY(&object->memq))
311 return (KERN_FAILURE);
312 break;
313 case OBJT_DEAD:
314 return (KERN_INVALID_ARGUMENT);
315 }
316 object->memattr = memattr;
317 return (KERN_SUCCESS);
318 }
319
320 void
321 vm_object_pip_add(vm_object_t object, short i)
322 {
323
324 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
325 object->paging_in_progress += i;
326 }
327
328 void
329 vm_object_pip_subtract(vm_object_t object, short i)
330 {
331
332 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
333 object->paging_in_progress -= i;
334 }
335
336 void
337 vm_object_pip_wakeup(vm_object_t object)
338 {
339
340 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
341 object->paging_in_progress--;
342 if ((object->flags & OBJ_PIPWNT) && object->paging_in_progress == 0) {
343 vm_object_clear_flag(object, OBJ_PIPWNT);
344 wakeup(object);
345 }
346 }
347
348 void
349 vm_object_pip_wakeupn(vm_object_t object, short i)
350 {
351
352 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
353 if (i)
354 object->paging_in_progress -= i;
355 if ((object->flags & OBJ_PIPWNT) && object->paging_in_progress == 0) {
356 vm_object_clear_flag(object, OBJ_PIPWNT);
357 wakeup(object);
358 }
359 }
360
361 void
362 vm_object_pip_wait(vm_object_t object, char *waitid)
363 {
364
365 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
366 while (object->paging_in_progress) {
367 object->flags |= OBJ_PIPWNT;
368 msleep(object, VM_OBJECT_MTX(object), PVM, waitid, 0);
369 }
370 }
371
372 /*
373 * vm_object_allocate:
374 *
375 * Returns a new object with the given size.
376 */
377 vm_object_t
378 vm_object_allocate(objtype_t type, vm_pindex_t size)
379 {
380 vm_object_t object;
381
382 object = (vm_object_t)uma_zalloc(obj_zone, M_WAITOK);
383 _vm_object_allocate(type, size, object);
384 return (object);
385 }
386
387
388 /*
389 * vm_object_reference:
390 *
391 * Gets another reference to the given object. Note: OBJ_DEAD
392 * objects can be referenced during final cleaning.
393 */
394 void
395 vm_object_reference(vm_object_t object)
396 {
397 if (object == NULL)
398 return;
399 VM_OBJECT_LOCK(object);
400 vm_object_reference_locked(object);
401 VM_OBJECT_UNLOCK(object);
402 }
403
404 /*
405 * vm_object_reference_locked:
406 *
407 * Gets another reference to the given object.
408 *
409 * The object must be locked.
410 */
411 void
412 vm_object_reference_locked(vm_object_t object)
413 {
414 struct vnode *vp;
415
416 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
417 object->ref_count++;
418 if (object->type == OBJT_VNODE) {
419 vp = object->handle;
420 vref(vp);
421 }
422 }
423
424 /*
425 * Handle deallocating an object of type OBJT_VNODE.
426 */
427 static void
428 vm_object_vndeallocate(vm_object_t object)
429 {
430 struct vnode *vp = (struct vnode *) object->handle;
431
432 VFS_ASSERT_GIANT(vp->v_mount);
433 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
434 KASSERT(object->type == OBJT_VNODE,
435 ("vm_object_vndeallocate: not a vnode object"));
436 KASSERT(vp != NULL, ("vm_object_vndeallocate: missing vp"));
437 #ifdef INVARIANTS
438 if (object->ref_count == 0) {
439 vprint("vm_object_vndeallocate", vp);
440 panic("vm_object_vndeallocate: bad object reference count");
441 }
442 #endif
443
444 if (object->ref_count > 1) {
445 object->ref_count--;
446 VM_OBJECT_UNLOCK(object);
447 /* vrele may need the vnode lock. */
448 vrele(vp);
449 } else {
450 vhold(vp);
451 VM_OBJECT_UNLOCK(object);
452 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
453 vdrop(vp);
454 VM_OBJECT_LOCK(object);
455 object->ref_count--;
456 if (object->type == OBJT_DEAD) {
457 VM_OBJECT_UNLOCK(object);
458 VOP_UNLOCK(vp, 0);
459 } else {
460 if (object->ref_count == 0)
461 vp->v_vflag &= ~VV_TEXT;
462 VM_OBJECT_UNLOCK(object);
463 vput(vp);
464 }
465 }
466 }
467
468 /*
469 * vm_object_deallocate:
470 *
471 * Release a reference to the specified object,
472 * gained either through a vm_object_allocate
473 * or a vm_object_reference call. When all references
474 * are gone, storage associated with this object
475 * may be relinquished.
476 *
477 * No object may be locked.
478 */
479 void
480 vm_object_deallocate(vm_object_t object)
481 {
482 vm_object_t temp;
483
484 while (object != NULL) {
485 int vfslocked;
486
487 vfslocked = 0;
488 restart:
489 VM_OBJECT_LOCK(object);
490 if (object->type == OBJT_VNODE) {
491 struct vnode *vp = (struct vnode *) object->handle;
492
493 /*
494 * Conditionally acquire Giant for a vnode-backed
495 * object. We have to be careful since the type of
496 * a vnode object can change while the object is
497 * unlocked.
498 */
499 if (VFS_NEEDSGIANT(vp->v_mount) && !vfslocked) {
500 vfslocked = 1;
501 if (!mtx_trylock(&Giant)) {
502 VM_OBJECT_UNLOCK(object);
503 mtx_lock(&Giant);
504 goto restart;
505 }
506 }
507 vm_object_vndeallocate(object);
508 VFS_UNLOCK_GIANT(vfslocked);
509 return;
510 } else
511 /*
512 * This is to handle the case that the object
513 * changed type while we dropped its lock to
514 * obtain Giant.
515 */
516 VFS_UNLOCK_GIANT(vfslocked);
517
518 KASSERT(object->ref_count != 0,
519 ("vm_object_deallocate: object deallocated too many times: %d", object->type));
520
521 /*
522 * If the reference count goes to 0 we start calling
523 * vm_object_terminate() on the object chain.
524 * A ref count of 1 may be a special case depending on the
525 * shadow count being 0 or 1.
526 */
527 object->ref_count--;
528 if (object->ref_count > 1) {
529 VM_OBJECT_UNLOCK(object);
530 return;
531 } else if (object->ref_count == 1) {
532 if (object->shadow_count == 0 &&
533 object->handle == NULL &&
534 (object->type == OBJT_DEFAULT ||
535 object->type == OBJT_SWAP)) {
536 vm_object_set_flag(object, OBJ_ONEMAPPING);
537 } else if ((object->shadow_count == 1) &&
538 (object->handle == NULL) &&
539 (object->type == OBJT_DEFAULT ||
540 object->type == OBJT_SWAP)) {
541 vm_object_t robject;
542
543 robject = LIST_FIRST(&object->shadow_head);
544 KASSERT(robject != NULL,
545 ("vm_object_deallocate: ref_count: %d, shadow_count: %d",
546 object->ref_count,
547 object->shadow_count));
548 if (!VM_OBJECT_TRYLOCK(robject)) {
549 /*
550 * Avoid a potential deadlock.
551 */
552 object->ref_count++;
553 VM_OBJECT_UNLOCK(object);
554 /*
555 * More likely than not the thread
556 * holding robject's lock has lower
557 * priority than the current thread.
558 * Let the lower priority thread run.
559 */
560 pause("vmo_de", 1);
561 continue;
562 }
563 /*
564 * Collapse object into its shadow unless its
565 * shadow is dead. In that case, object will
566 * be deallocated by the thread that is
567 * deallocating its shadow.
568 */
569 if ((robject->flags & OBJ_DEAD) == 0 &&
570 (robject->handle == NULL) &&
571 (robject->type == OBJT_DEFAULT ||
572 robject->type == OBJT_SWAP)) {
573
574 robject->ref_count++;
575 retry:
576 if (robject->paging_in_progress) {
577 VM_OBJECT_UNLOCK(object);
578 vm_object_pip_wait(robject,
579 "objde1");
580 temp = robject->backing_object;
581 if (object == temp) {
582 VM_OBJECT_LOCK(object);
583 goto retry;
584 }
585 } else if (object->paging_in_progress) {
586 VM_OBJECT_UNLOCK(robject);
587 object->flags |= OBJ_PIPWNT;
588 msleep(object,
589 VM_OBJECT_MTX(object),
590 PDROP | PVM, "objde2", 0);
591 VM_OBJECT_LOCK(robject);
592 temp = robject->backing_object;
593 if (object == temp) {
594 VM_OBJECT_LOCK(object);
595 goto retry;
596 }
597 } else
598 VM_OBJECT_UNLOCK(object);
599
600 if (robject->ref_count == 1) {
601 robject->ref_count--;
602 object = robject;
603 goto doterm;
604 }
605 object = robject;
606 vm_object_collapse(object);
607 VM_OBJECT_UNLOCK(object);
608 continue;
609 }
610 VM_OBJECT_UNLOCK(robject);
611 }
612 VM_OBJECT_UNLOCK(object);
613 return;
614 }
615 doterm:
616 temp = object->backing_object;
617 if (temp != NULL) {
618 VM_OBJECT_LOCK(temp);
619 LIST_REMOVE(object, shadow_list);
620 temp->shadow_count--;
621 VM_OBJECT_UNLOCK(temp);
622 object->backing_object = NULL;
623 }
624 /*
625 * Don't double-terminate, we could be in a termination
626 * recursion due to the terminate having to sync data
627 * to disk.
628 */
629 if ((object->flags & OBJ_DEAD) == 0)
630 vm_object_terminate(object);
631 else
632 VM_OBJECT_UNLOCK(object);
633 object = temp;
634 }
635 }
636
637 /*
638 * vm_object_destroy removes the object from the global object list
639 * and frees the space for the object.
640 */
641 void
642 vm_object_destroy(vm_object_t object)
643 {
644
645 /*
646 * Remove the object from the global object list.
647 */
648 mtx_lock(&vm_object_list_mtx);
649 TAILQ_REMOVE(&vm_object_list, object, object_list);
650 mtx_unlock(&vm_object_list_mtx);
651
652 /*
653 * Release the allocation charge.
654 */
655 if (object->uip != NULL) {
656 KASSERT(object->type == OBJT_DEFAULT ||
657 object->type == OBJT_SWAP,
658 ("vm_object_terminate: non-swap obj %p has uip",
659 object));
660 swap_release_by_uid(object->charge, object->uip);
661 object->charge = 0;
662 uifree(object->uip);
663 object->uip = NULL;
664 }
665
666 /*
667 * Free the space for the object.
668 */
669 uma_zfree(obj_zone, object);
670 }
671
672 /*
673 * vm_object_terminate actually destroys the specified object, freeing
674 * up all previously used resources.
675 *
676 * The object must be locked.
677 * This routine may block.
678 */
679 void
680 vm_object_terminate(vm_object_t object)
681 {
682 vm_page_t p, p_next;
683
684 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
685
686 /*
687 * Make sure no one uses us.
688 */
689 vm_object_set_flag(object, OBJ_DEAD);
690
691 /*
692 * wait for the pageout daemon to be done with the object
693 */
694 vm_object_pip_wait(object, "objtrm");
695
696 KASSERT(!object->paging_in_progress,
697 ("vm_object_terminate: pageout in progress"));
698
699 /*
700 * Clean and free the pages, as appropriate. All references to the
701 * object are gone, so we don't need to lock it.
702 */
703 if (object->type == OBJT_VNODE) {
704 struct vnode *vp = (struct vnode *)object->handle;
705
706 /*
707 * Clean pages and flush buffers.
708 */
709 vm_object_page_clean(object, 0, 0, OBJPC_SYNC);
710 VM_OBJECT_UNLOCK(object);
711
712 vinvalbuf(vp, V_SAVE, 0, 0);
713
714 VM_OBJECT_LOCK(object);
715 }
716
717 KASSERT(object->ref_count == 0,
718 ("vm_object_terminate: object with references, ref_count=%d",
719 object->ref_count));
720
721 /*
722 * Free any remaining pageable pages. This also removes them from the
723 * paging queues. However, don't free wired pages, just remove them
724 * from the object. Rather than incrementally removing each page from
725 * the object, the page and object are reset to any empty state.
726 */
727 vm_page_lock_queues();
728 TAILQ_FOREACH_SAFE(p, &object->memq, listq, p_next) {
729 KASSERT(!p->busy && (p->oflags & VPO_BUSY) == 0,
730 ("vm_object_terminate: freeing busy page %p", p));
731 /*
732 * Optimize the page's removal from the object by resetting
733 * its "object" field. Specifically, if the page is not
734 * wired, then the effect of this assignment is that
735 * vm_page_free()'s call to vm_page_remove() will return
736 * immediately without modifying the page or the object.
737 */
738 p->object = NULL;
739 if (p->wire_count == 0) {
740 vm_page_free(p);
741 cnt.v_pfree++;
742 }
743 }
744 vm_page_unlock_queues();
745 /*
746 * If the object contained any pages, then reset it to an empty state.
747 * None of the object's fields, including "resident_page_count", were
748 * modified by the preceding loop.
749 */
750 if (object->resident_page_count != 0) {
751 object->root = NULL;
752 TAILQ_INIT(&object->memq);
753 object->resident_page_count = 0;
754 if (object->type == OBJT_VNODE)
755 vdrop(object->handle);
756 }
757
758 #if VM_NRESERVLEVEL > 0
759 if (__predict_false(!LIST_EMPTY(&object->rvq)))
760 vm_reserv_break_all(object);
761 #endif
762 if (__predict_false(object->cache != NULL))
763 vm_page_cache_free(object, 0, 0);
764
765 /*
766 * Let the pager know object is dead.
767 */
768 vm_pager_deallocate(object);
769 VM_OBJECT_UNLOCK(object);
770
771 vm_object_destroy(object);
772 }
773
774 static boolean_t
775 vm_object_page_remove_write(vm_page_t p, int flags, int *clearobjflags)
776 {
777
778 /*
779 * If we have been asked to skip nosync pages and this is a
780 * nosync page, skip it. Note that the object flags were not
781 * cleared in this case so we do not have to set them.
782 */
783 if ((flags & OBJPC_NOSYNC) != 0 && (p->oflags & VPO_NOSYNC) != 0) {
784 *clearobjflags = 0;
785 return (FALSE);
786 } else {
787 pmap_remove_write(p);
788 return (p->dirty != 0);
789 }
790 }
791
792 /*
793 * vm_object_page_clean
794 *
795 * Clean all dirty pages in the specified range of object. Leaves page
796 * on whatever queue it is currently on. If NOSYNC is set then do not
797 * write out pages with VPO_NOSYNC set (originally comes from MAP_NOSYNC),
798 * leaving the object dirty.
799 *
800 * When stuffing pages asynchronously, allow clustering. XXX we need a
801 * synchronous clustering mode implementation.
802 *
803 * Odd semantics: if start == end, we clean everything.
804 *
805 * The object must be locked.
806 */
807 void
808 vm_object_page_clean(vm_object_t object, vm_pindex_t start, vm_pindex_t end,
809 int flags)
810 {
811 vm_page_t np, p;
812 vm_pindex_t pi, tend;
813 int clearobjflags, curgeneration, n, pagerflags;
814
815 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
816 KASSERT(object->type == OBJT_VNODE, ("Not a vnode object"));
817 if ((object->flags & OBJ_MIGHTBEDIRTY) == 0 ||
818 object->resident_page_count == 0)
819 return;
820
821 pagerflags = (flags & (OBJPC_SYNC | OBJPC_INVAL)) != 0 ?
822 VM_PAGER_PUT_SYNC : VM_PAGER_CLUSTER_OK;
823 pagerflags |= (flags & OBJPC_INVAL) != 0 ? VM_PAGER_PUT_INVAL : 0;
824
825 tend = (end == 0) ? object->size : end;
826
827 vm_object_set_flag(object, OBJ_CLEANING);
828
829 vm_page_lock_queues();
830
831 /*
832 * Make the page read-only so we can then clear the object flags.
833 *
834 * However, if this is a nosync mmap then the object is likely to
835 * stay dirty so do not mess with the page and do not clear the
836 * object flags.
837 */
838 clearobjflags = 1;
839
840 rescan:
841 curgeneration = object->generation;
842
843 for (p = vm_page_find_least(object, start); p != NULL; p = np) {
844 pi = p->pindex;
845 if (pi >= tend)
846 break;
847 np = TAILQ_NEXT(p, listq);
848 if (p->valid == 0)
849 continue;
850 if (vm_page_sleep_if_busy(p, TRUE, "vpcwai")) {
851 vm_page_lock_queues();
852 if (object->generation != curgeneration)
853 goto rescan;
854 np = vm_page_find_least(object, pi);
855 continue;
856 }
857 if (!vm_object_page_remove_write(p, flags, &clearobjflags))
858 continue;
859
860 n = vm_object_page_collect_flush(object, p, pagerflags,
861 flags, &clearobjflags);
862 if (object->generation != curgeneration)
863 goto rescan;
864
865 /*
866 * If the VOP_PUTPAGES() did a truncated write, so
867 * that even the first page of the run is not fully
868 * written, vm_pageout_flush() returns 0 as the run
869 * length. Since the condition that caused truncated
870 * write may be permanent, e.g. exhausted free space,
871 * accepting n == 0 would cause an infinite loop.
872 *
873 * Forwarding the iterator leaves the unwritten page
874 * behind, but there is not much we can do there if
875 * filesystem refuses to write it.
876 */
877 if (n == 0)
878 n = 1;
879 np = vm_page_find_least(object, pi + n);
880 }
881 vm_page_unlock_queues();
882 #if 0
883 VOP_FSYNC(vp, (pagerflags & VM_PAGER_PUT_SYNC) ? MNT_WAIT : 0);
884 #endif
885
886 vm_object_clear_flag(object, OBJ_CLEANING);
887 if (clearobjflags && start == 0 && tend == object->size)
888 vm_object_clear_flag(object, OBJ_MIGHTBEDIRTY);
889 }
890
891 static int
892 vm_object_page_collect_flush(vm_object_t object, vm_page_t p, int pagerflags,
893 int flags, int *clearobjflags)
894 {
895 vm_page_t ma[vm_pageout_page_count], p_first, tp;
896 int count, i, mreq, runlen;
897
898 mtx_assert(&vm_page_queue_mtx, MA_OWNED);
899
900 count = 1;
901 mreq = 0;
902
903 for (tp = p; count < vm_pageout_page_count; count++) {
904 tp = vm_page_next(tp);
905 if (tp == NULL || tp->busy != 0 || (tp->oflags & VPO_BUSY) != 0)
906 break;
907 if (!vm_object_page_remove_write(tp, flags, clearobjflags))
908 break;
909 }
910
911 for (p_first = p; count < vm_pageout_page_count; count++) {
912 tp = vm_page_prev(p_first);
913 if (tp == NULL || tp->busy != 0 || (tp->oflags & VPO_BUSY) != 0)
914 break;
915 if (!vm_object_page_remove_write(tp, flags, clearobjflags))
916 break;
917 p_first = tp;
918 mreq++;
919 }
920
921 for (tp = p_first, i = 0; i < count; tp = TAILQ_NEXT(tp, listq), i++)
922 ma[i] = tp;
923
924 vm_pageout_flush(ma, count, pagerflags, mreq, &runlen);
925 return (runlen);
926 }
927
928 /*
929 * Note that there is absolutely no sense in writing out
930 * anonymous objects, so we track down the vnode object
931 * to write out.
932 * We invalidate (remove) all pages from the address space
933 * for semantic correctness.
934 *
935 * Note: certain anonymous maps, such as MAP_NOSYNC maps,
936 * may start out with a NULL object.
937 */
938 void
939 vm_object_sync(vm_object_t object, vm_ooffset_t offset, vm_size_t size,
940 boolean_t syncio, boolean_t invalidate)
941 {
942 vm_object_t backing_object;
943 struct vnode *vp;
944 struct mount *mp;
945 int flags, fsync_after;
946
947 if (object == NULL)
948 return;
949 VM_OBJECT_LOCK(object);
950 while ((backing_object = object->backing_object) != NULL) {
951 VM_OBJECT_LOCK(backing_object);
952 offset += object->backing_object_offset;
953 VM_OBJECT_UNLOCK(object);
954 object = backing_object;
955 if (object->size < OFF_TO_IDX(offset + size))
956 size = IDX_TO_OFF(object->size) - offset;
957 }
958 /*
959 * Flush pages if writing is allowed, invalidate them
960 * if invalidation requested. Pages undergoing I/O
961 * will be ignored by vm_object_page_remove().
962 *
963 * We cannot lock the vnode and then wait for paging
964 * to complete without deadlocking against vm_fault.
965 * Instead we simply call vm_object_page_remove() and
966 * allow it to block internally on a page-by-page
967 * basis when it encounters pages undergoing async
968 * I/O.
969 */
970 if (object->type == OBJT_VNODE &&
971 (object->flags & OBJ_MIGHTBEDIRTY) != 0) {
972 int vfslocked;
973 vp = object->handle;
974 VM_OBJECT_UNLOCK(object);
975 (void) vn_start_write(vp, &mp, V_WAIT);
976 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
977 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
978 if (syncio && !invalidate && offset == 0 &&
979 OFF_TO_IDX(size) == object->size) {
980 /*
981 * If syncing the whole mapping of the file,
982 * it is faster to schedule all the writes in
983 * async mode, also allowing the clustering,
984 * and then wait for i/o to complete.
985 */
986 flags = 0;
987 fsync_after = TRUE;
988 } else {
989 flags = (syncio || invalidate) ? OBJPC_SYNC : 0;
990 flags |= invalidate ? (OBJPC_SYNC | OBJPC_INVAL) : 0;
991 fsync_after = FALSE;
992 }
993 VM_OBJECT_LOCK(object);
994 vm_object_page_clean(object,
995 OFF_TO_IDX(offset),
996 OFF_TO_IDX(offset + size + PAGE_MASK),
997 flags);
998 VM_OBJECT_UNLOCK(object);
999 if (fsync_after)
1000 (void) VOP_FSYNC(vp, MNT_WAIT, curthread);
1001 VOP_UNLOCK(vp, 0);
1002 VFS_UNLOCK_GIANT(vfslocked);
1003 vn_finished_write(mp);
1004 VM_OBJECT_LOCK(object);
1005 }
1006 if ((object->type == OBJT_VNODE ||
1007 object->type == OBJT_DEVICE) && invalidate) {
1008 boolean_t purge;
1009 purge = old_msync || (object->type == OBJT_DEVICE);
1010 vm_object_page_remove(object,
1011 OFF_TO_IDX(offset),
1012 OFF_TO_IDX(offset + size + PAGE_MASK),
1013 purge ? FALSE : TRUE);
1014 }
1015 VM_OBJECT_UNLOCK(object);
1016 }
1017
1018 /*
1019 * vm_object_madvise:
1020 *
1021 * Implements the madvise function at the object/page level.
1022 *
1023 * MADV_WILLNEED (any object)
1024 *
1025 * Activate the specified pages if they are resident.
1026 *
1027 * MADV_DONTNEED (any object)
1028 *
1029 * Deactivate the specified pages if they are resident.
1030 *
1031 * MADV_FREE (OBJT_DEFAULT/OBJT_SWAP objects,
1032 * OBJ_ONEMAPPING only)
1033 *
1034 * Deactivate and clean the specified pages if they are
1035 * resident. This permits the process to reuse the pages
1036 * without faulting or the kernel to reclaim the pages
1037 * without I/O.
1038 */
1039 void
1040 vm_object_madvise(vm_object_t object, vm_pindex_t pindex, int count, int advise)
1041 {
1042 vm_pindex_t end, tpindex;
1043 vm_object_t backing_object, tobject;
1044 vm_page_t m;
1045
1046 if (object == NULL)
1047 return;
1048 VM_OBJECT_LOCK(object);
1049 end = pindex + count;
1050 /*
1051 * Locate and adjust resident pages
1052 */
1053 for (; pindex < end; pindex += 1) {
1054 relookup:
1055 tobject = object;
1056 tpindex = pindex;
1057 shadowlookup:
1058 /*
1059 * MADV_FREE only operates on OBJT_DEFAULT or OBJT_SWAP pages
1060 * and those pages must be OBJ_ONEMAPPING.
1061 */
1062 if (advise == MADV_FREE) {
1063 if ((tobject->type != OBJT_DEFAULT &&
1064 tobject->type != OBJT_SWAP) ||
1065 (tobject->flags & OBJ_ONEMAPPING) == 0) {
1066 goto unlock_tobject;
1067 }
1068 } else if (tobject->type == OBJT_PHYS)
1069 goto unlock_tobject;
1070 m = vm_page_lookup(tobject, tpindex);
1071 if (m == NULL && advise == MADV_WILLNEED) {
1072 /*
1073 * If the page is cached, reactivate it.
1074 */
1075 m = vm_page_alloc(tobject, tpindex, VM_ALLOC_IFCACHED |
1076 VM_ALLOC_NOBUSY);
1077 }
1078 if (m == NULL) {
1079 /*
1080 * There may be swap even if there is no backing page
1081 */
1082 if (advise == MADV_FREE && tobject->type == OBJT_SWAP)
1083 swap_pager_freespace(tobject, tpindex, 1);
1084 /*
1085 * next object
1086 */
1087 backing_object = tobject->backing_object;
1088 if (backing_object == NULL)
1089 goto unlock_tobject;
1090 VM_OBJECT_LOCK(backing_object);
1091 tpindex += OFF_TO_IDX(tobject->backing_object_offset);
1092 if (tobject != object)
1093 VM_OBJECT_UNLOCK(tobject);
1094 tobject = backing_object;
1095 goto shadowlookup;
1096 } else if (m->valid != VM_PAGE_BITS_ALL)
1097 goto unlock_tobject;
1098 /*
1099 * If the page is not in a normal state, skip it.
1100 */
1101 vm_page_lock_queues();
1102 if (m->hold_count != 0 || m->wire_count != 0) {
1103 vm_page_unlock_queues();
1104 goto unlock_tobject;
1105 }
1106 if ((m->oflags & VPO_BUSY) || m->busy) {
1107 if (advise == MADV_WILLNEED)
1108 /*
1109 * Reference the page before unlocking and
1110 * sleeping so that the page daemon is less
1111 * likely to reclaim it.
1112 */
1113 vm_page_flag_set(m, PG_REFERENCED);
1114 vm_page_unlock_queues();
1115 if (object != tobject)
1116 VM_OBJECT_UNLOCK(object);
1117 m->oflags |= VPO_WANTED;
1118 msleep(m, VM_OBJECT_MTX(tobject), PDROP | PVM, "madvpo",
1119 0);
1120 VM_OBJECT_LOCK(object);
1121 goto relookup;
1122 }
1123 if (advise == MADV_WILLNEED) {
1124 vm_page_activate(m);
1125 } else if (advise == MADV_DONTNEED) {
1126 vm_page_dontneed(m);
1127 } else if (advise == MADV_FREE) {
1128 /*
1129 * Mark the page clean. This will allow the page
1130 * to be freed up by the system. However, such pages
1131 * are often reused quickly by malloc()/free()
1132 * so we do not do anything that would cause
1133 * a page fault if we can help it.
1134 *
1135 * Specifically, we do not try to actually free
1136 * the page now nor do we try to put it in the
1137 * cache (which would cause a page fault on reuse).
1138 *
1139 * But we do make the page is freeable as we
1140 * can without actually taking the step of unmapping
1141 * it.
1142 */
1143 pmap_clear_modify(m);
1144 m->dirty = 0;
1145 m->act_count = 0;
1146 vm_page_dontneed(m);
1147 }
1148 vm_page_unlock_queues();
1149 if (advise == MADV_FREE && tobject->type == OBJT_SWAP)
1150 swap_pager_freespace(tobject, tpindex, 1);
1151 unlock_tobject:
1152 if (tobject != object)
1153 VM_OBJECT_UNLOCK(tobject);
1154 }
1155 VM_OBJECT_UNLOCK(object);
1156 }
1157
1158 /*
1159 * vm_object_shadow:
1160 *
1161 * Create a new object which is backed by the
1162 * specified existing object range. The source
1163 * object reference is deallocated.
1164 *
1165 * The new object and offset into that object
1166 * are returned in the source parameters.
1167 */
1168 void
1169 vm_object_shadow(
1170 vm_object_t *object, /* IN/OUT */
1171 vm_ooffset_t *offset, /* IN/OUT */
1172 vm_size_t length)
1173 {
1174 vm_object_t source;
1175 vm_object_t result;
1176
1177 source = *object;
1178
1179 /*
1180 * Don't create the new object if the old object isn't shared.
1181 */
1182 if (source != NULL) {
1183 VM_OBJECT_LOCK(source);
1184 if (source->ref_count == 1 &&
1185 source->handle == NULL &&
1186 (source->type == OBJT_DEFAULT ||
1187 source->type == OBJT_SWAP)) {
1188 VM_OBJECT_UNLOCK(source);
1189 return;
1190 }
1191 VM_OBJECT_UNLOCK(source);
1192 }
1193
1194 /*
1195 * Allocate a new object with the given length.
1196 */
1197 result = vm_object_allocate(OBJT_DEFAULT, length);
1198
1199 /*
1200 * The new object shadows the source object, adding a reference to it.
1201 * Our caller changes his reference to point to the new object,
1202 * removing a reference to the source object. Net result: no change
1203 * of reference count.
1204 *
1205 * Try to optimize the result object's page color when shadowing
1206 * in order to maintain page coloring consistency in the combined
1207 * shadowed object.
1208 */
1209 result->backing_object = source;
1210 /*
1211 * Store the offset into the source object, and fix up the offset into
1212 * the new object.
1213 */
1214 result->backing_object_offset = *offset;
1215 if (source != NULL) {
1216 VM_OBJECT_LOCK(source);
1217 LIST_INSERT_HEAD(&source->shadow_head, result, shadow_list);
1218 source->shadow_count++;
1219 #if VM_NRESERVLEVEL > 0
1220 result->flags |= source->flags & OBJ_COLORED;
1221 result->pg_color = (source->pg_color + OFF_TO_IDX(*offset)) &
1222 ((1 << (VM_NFREEORDER - 1)) - 1);
1223 #endif
1224 VM_OBJECT_UNLOCK(source);
1225 }
1226
1227
1228 /*
1229 * Return the new things
1230 */
1231 *offset = 0;
1232 *object = result;
1233 }
1234
1235 /*
1236 * vm_object_split:
1237 *
1238 * Split the pages in a map entry into a new object. This affords
1239 * easier removal of unused pages, and keeps object inheritance from
1240 * being a negative impact on memory usage.
1241 */
1242 void
1243 vm_object_split(vm_map_entry_t entry)
1244 {
1245 vm_page_t m, m_next;
1246 vm_object_t orig_object, new_object, source;
1247 vm_pindex_t idx, offidxstart;
1248 vm_size_t size;
1249
1250 orig_object = entry->object.vm_object;
1251 if (orig_object->type != OBJT_DEFAULT && orig_object->type != OBJT_SWAP)
1252 return;
1253 if (orig_object->ref_count <= 1)
1254 return;
1255 VM_OBJECT_UNLOCK(orig_object);
1256
1257 offidxstart = OFF_TO_IDX(entry->offset);
1258 size = atop(entry->end - entry->start);
1259
1260 /*
1261 * If swap_pager_copy() is later called, it will convert new_object
1262 * into a swap object.
1263 */
1264 new_object = vm_object_allocate(OBJT_DEFAULT, size);
1265
1266 /*
1267 * At this point, the new object is still private, so the order in
1268 * which the original and new objects are locked does not matter.
1269 */
1270 VM_OBJECT_LOCK(new_object);
1271 VM_OBJECT_LOCK(orig_object);
1272 source = orig_object->backing_object;
1273 if (source != NULL) {
1274 VM_OBJECT_LOCK(source);
1275 if ((source->flags & OBJ_DEAD) != 0) {
1276 VM_OBJECT_UNLOCK(source);
1277 VM_OBJECT_UNLOCK(orig_object);
1278 VM_OBJECT_UNLOCK(new_object);
1279 vm_object_deallocate(new_object);
1280 VM_OBJECT_LOCK(orig_object);
1281 return;
1282 }
1283 LIST_INSERT_HEAD(&source->shadow_head,
1284 new_object, shadow_list);
1285 source->shadow_count++;
1286 vm_object_reference_locked(source); /* for new_object */
1287 vm_object_clear_flag(source, OBJ_ONEMAPPING);
1288 VM_OBJECT_UNLOCK(source);
1289 new_object->backing_object_offset =
1290 orig_object->backing_object_offset + entry->offset;
1291 new_object->backing_object = source;
1292 }
1293 if (orig_object->uip != NULL) {
1294 new_object->uip = orig_object->uip;
1295 uihold(orig_object->uip);
1296 new_object->charge = ptoa(size);
1297 KASSERT(orig_object->charge >= ptoa(size),
1298 ("orig_object->charge < 0"));
1299 orig_object->charge -= ptoa(size);
1300 }
1301 retry:
1302 m = vm_page_find_least(orig_object, offidxstart);
1303 vm_page_lock_queues();
1304 for (; m != NULL && (idx = m->pindex - offidxstart) < size;
1305 m = m_next) {
1306 m_next = TAILQ_NEXT(m, listq);
1307
1308 /*
1309 * We must wait for pending I/O to complete before we can
1310 * rename the page.
1311 *
1312 * We do not have to VM_PROT_NONE the page as mappings should
1313 * not be changed by this operation.
1314 */
1315 if ((m->oflags & VPO_BUSY) || m->busy) {
1316 vm_page_unlock_queues();
1317 VM_OBJECT_UNLOCK(new_object);
1318 m->oflags |= VPO_WANTED;
1319 msleep(m, VM_OBJECT_MTX(orig_object), PVM, "spltwt", 0);
1320 VM_OBJECT_LOCK(new_object);
1321 goto retry;
1322 }
1323 vm_page_rename(m, new_object, idx);
1324 /* page automatically made dirty by rename and cache handled */
1325 vm_page_busy(m);
1326 }
1327 vm_page_unlock_queues();
1328 if (orig_object->type == OBJT_SWAP) {
1329 /*
1330 * swap_pager_copy() can sleep, in which case the orig_object's
1331 * and new_object's locks are released and reacquired.
1332 */
1333 swap_pager_copy(orig_object, new_object, offidxstart, 0);
1334
1335 /*
1336 * Transfer any cached pages from orig_object to new_object.
1337 */
1338 if (__predict_false(orig_object->cache != NULL))
1339 vm_page_cache_transfer(orig_object, offidxstart,
1340 new_object);
1341 }
1342 VM_OBJECT_UNLOCK(orig_object);
1343 TAILQ_FOREACH(m, &new_object->memq, listq)
1344 vm_page_wakeup(m);
1345 VM_OBJECT_UNLOCK(new_object);
1346 entry->object.vm_object = new_object;
1347 entry->offset = 0LL;
1348 vm_object_deallocate(orig_object);
1349 VM_OBJECT_LOCK(new_object);
1350 }
1351
1352 #define OBSC_TEST_ALL_SHADOWED 0x0001
1353 #define OBSC_COLLAPSE_NOWAIT 0x0002
1354 #define OBSC_COLLAPSE_WAIT 0x0004
1355
1356 static int
1357 vm_object_backing_scan(vm_object_t object, int op)
1358 {
1359 int r = 1;
1360 vm_page_t p;
1361 vm_object_t backing_object;
1362 vm_pindex_t backing_offset_index;
1363
1364 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
1365 VM_OBJECT_LOCK_ASSERT(object->backing_object, MA_OWNED);
1366
1367 backing_object = object->backing_object;
1368 backing_offset_index = OFF_TO_IDX(object->backing_object_offset);
1369
1370 /*
1371 * Initial conditions
1372 */
1373 if (op & OBSC_TEST_ALL_SHADOWED) {
1374 /*
1375 * We do not want to have to test for the existence of cache
1376 * or swap pages in the backing object. XXX but with the
1377 * new swapper this would be pretty easy to do.
1378 *
1379 * XXX what about anonymous MAP_SHARED memory that hasn't
1380 * been ZFOD faulted yet? If we do not test for this, the
1381 * shadow test may succeed! XXX
1382 */
1383 if (backing_object->type != OBJT_DEFAULT) {
1384 return (0);
1385 }
1386 }
1387 if (op & OBSC_COLLAPSE_WAIT) {
1388 vm_object_set_flag(backing_object, OBJ_DEAD);
1389 }
1390
1391 /*
1392 * Our scan
1393 */
1394 p = TAILQ_FIRST(&backing_object->memq);
1395 while (p) {
1396 vm_page_t next = TAILQ_NEXT(p, listq);
1397 vm_pindex_t new_pindex = p->pindex - backing_offset_index;
1398
1399 if (op & OBSC_TEST_ALL_SHADOWED) {
1400 vm_page_t pp;
1401
1402 /*
1403 * Ignore pages outside the parent object's range
1404 * and outside the parent object's mapping of the
1405 * backing object.
1406 *
1407 * note that we do not busy the backing object's
1408 * page.
1409 */
1410 if (
1411 p->pindex < backing_offset_index ||
1412 new_pindex >= object->size
1413 ) {
1414 p = next;
1415 continue;
1416 }
1417
1418 /*
1419 * See if the parent has the page or if the parent's
1420 * object pager has the page. If the parent has the
1421 * page but the page is not valid, the parent's
1422 * object pager must have the page.
1423 *
1424 * If this fails, the parent does not completely shadow
1425 * the object and we might as well give up now.
1426 */
1427
1428 pp = vm_page_lookup(object, new_pindex);
1429 if (
1430 (pp == NULL || pp->valid == 0) &&
1431 !vm_pager_has_page(object, new_pindex, NULL, NULL)
1432 ) {
1433 r = 0;
1434 break;
1435 }
1436 }
1437
1438 /*
1439 * Check for busy page
1440 */
1441 if (op & (OBSC_COLLAPSE_WAIT | OBSC_COLLAPSE_NOWAIT)) {
1442 vm_page_t pp;
1443
1444 if (op & OBSC_COLLAPSE_NOWAIT) {
1445 if ((p->oflags & VPO_BUSY) ||
1446 !p->valid ||
1447 p->busy) {
1448 p = next;
1449 continue;
1450 }
1451 } else if (op & OBSC_COLLAPSE_WAIT) {
1452 if ((p->oflags & VPO_BUSY) || p->busy) {
1453 VM_OBJECT_UNLOCK(object);
1454 p->oflags |= VPO_WANTED;
1455 msleep(p, VM_OBJECT_MTX(backing_object),
1456 PDROP | PVM, "vmocol", 0);
1457 VM_OBJECT_LOCK(object);
1458 VM_OBJECT_LOCK(backing_object);
1459 /*
1460 * If we slept, anything could have
1461 * happened. Since the object is
1462 * marked dead, the backing offset
1463 * should not have changed so we
1464 * just restart our scan.
1465 */
1466 p = TAILQ_FIRST(&backing_object->memq);
1467 continue;
1468 }
1469 }
1470
1471 KASSERT(
1472 p->object == backing_object,
1473 ("vm_object_backing_scan: object mismatch")
1474 );
1475
1476 /*
1477 * Destroy any associated swap
1478 */
1479 if (backing_object->type == OBJT_SWAP) {
1480 swap_pager_freespace(
1481 backing_object,
1482 p->pindex,
1483 1
1484 );
1485 }
1486
1487 if (
1488 p->pindex < backing_offset_index ||
1489 new_pindex >= object->size
1490 ) {
1491 /*
1492 * Page is out of the parent object's range, we
1493 * can simply destroy it.
1494 */
1495 vm_page_lock_queues();
1496 KASSERT(!pmap_page_is_mapped(p),
1497 ("freeing mapped page %p", p));
1498 if (p->wire_count == 0)
1499 vm_page_free(p);
1500 else
1501 vm_page_remove(p);
1502 vm_page_unlock_queues();
1503 p = next;
1504 continue;
1505 }
1506
1507 pp = vm_page_lookup(object, new_pindex);
1508 if (
1509 (op & OBSC_COLLAPSE_NOWAIT) != 0 &&
1510 (pp != NULL && pp->valid == 0)
1511 ) {
1512 /*
1513 * The page in the parent is not (yet) valid.
1514 * We don't know anything about the state of
1515 * the original page. It might be mapped,
1516 * so we must avoid the next if here.
1517 *
1518 * This is due to a race in vm_fault() where
1519 * we must unbusy the original (backing_obj)
1520 * page before we can (re)lock the parent.
1521 * Hence we can get here.
1522 */
1523 p = next;
1524 continue;
1525 }
1526 if (
1527 pp != NULL ||
1528 vm_pager_has_page(object, new_pindex, NULL, NULL)
1529 ) {
1530 /*
1531 * page already exists in parent OR swap exists
1532 * for this location in the parent. Destroy
1533 * the original page from the backing object.
1534 *
1535 * Leave the parent's page alone
1536 */
1537 vm_page_lock_queues();
1538 KASSERT(!pmap_page_is_mapped(p),
1539 ("freeing mapped page %p", p));
1540 if (p->wire_count == 0)
1541 vm_page_free(p);
1542 else
1543 vm_page_remove(p);
1544 vm_page_unlock_queues();
1545 p = next;
1546 continue;
1547 }
1548
1549 #if VM_NRESERVLEVEL > 0
1550 /*
1551 * Rename the reservation.
1552 */
1553 vm_reserv_rename(p, object, backing_object,
1554 backing_offset_index);
1555 #endif
1556
1557 /*
1558 * Page does not exist in parent, rename the
1559 * page from the backing object to the main object.
1560 *
1561 * If the page was mapped to a process, it can remain
1562 * mapped through the rename.
1563 */
1564 vm_page_lock_queues();
1565 vm_page_rename(p, object, new_pindex);
1566 vm_page_unlock_queues();
1567 /* page automatically made dirty by rename */
1568 }
1569 p = next;
1570 }
1571 return (r);
1572 }
1573
1574
1575 /*
1576 * this version of collapse allows the operation to occur earlier and
1577 * when paging_in_progress is true for an object... This is not a complete
1578 * operation, but should plug 99.9% of the rest of the leaks.
1579 */
1580 static void
1581 vm_object_qcollapse(vm_object_t object)
1582 {
1583 vm_object_t backing_object = object->backing_object;
1584
1585 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
1586 VM_OBJECT_LOCK_ASSERT(backing_object, MA_OWNED);
1587
1588 if (backing_object->ref_count != 1)
1589 return;
1590
1591 vm_object_backing_scan(object, OBSC_COLLAPSE_NOWAIT);
1592 }
1593
1594 /*
1595 * vm_object_collapse:
1596 *
1597 * Collapse an object with the object backing it.
1598 * Pages in the backing object are moved into the
1599 * parent, and the backing object is deallocated.
1600 */
1601 void
1602 vm_object_collapse(vm_object_t object)
1603 {
1604 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
1605
1606 while (TRUE) {
1607 vm_object_t backing_object;
1608
1609 /*
1610 * Verify that the conditions are right for collapse:
1611 *
1612 * The object exists and the backing object exists.
1613 */
1614 if ((backing_object = object->backing_object) == NULL)
1615 break;
1616
1617 /*
1618 * we check the backing object first, because it is most likely
1619 * not collapsable.
1620 */
1621 VM_OBJECT_LOCK(backing_object);
1622 if (backing_object->handle != NULL ||
1623 (backing_object->type != OBJT_DEFAULT &&
1624 backing_object->type != OBJT_SWAP) ||
1625 (backing_object->flags & OBJ_DEAD) ||
1626 object->handle != NULL ||
1627 (object->type != OBJT_DEFAULT &&
1628 object->type != OBJT_SWAP) ||
1629 (object->flags & OBJ_DEAD)) {
1630 VM_OBJECT_UNLOCK(backing_object);
1631 break;
1632 }
1633
1634 if (
1635 object->paging_in_progress != 0 ||
1636 backing_object->paging_in_progress != 0
1637 ) {
1638 vm_object_qcollapse(object);
1639 VM_OBJECT_UNLOCK(backing_object);
1640 break;
1641 }
1642 /*
1643 * We know that we can either collapse the backing object (if
1644 * the parent is the only reference to it) or (perhaps) have
1645 * the parent bypass the object if the parent happens to shadow
1646 * all the resident pages in the entire backing object.
1647 *
1648 * This is ignoring pager-backed pages such as swap pages.
1649 * vm_object_backing_scan fails the shadowing test in this
1650 * case.
1651 */
1652 if (backing_object->ref_count == 1) {
1653 /*
1654 * If there is exactly one reference to the backing
1655 * object, we can collapse it into the parent.
1656 */
1657 vm_object_backing_scan(object, OBSC_COLLAPSE_WAIT);
1658
1659 #if VM_NRESERVLEVEL > 0
1660 /*
1661 * Break any reservations from backing_object.
1662 */
1663 if (__predict_false(!LIST_EMPTY(&backing_object->rvq)))
1664 vm_reserv_break_all(backing_object);
1665 #endif
1666
1667 /*
1668 * Move the pager from backing_object to object.
1669 */
1670 if (backing_object->type == OBJT_SWAP) {
1671 /*
1672 * swap_pager_copy() can sleep, in which case
1673 * the backing_object's and object's locks are
1674 * released and reacquired.
1675 */
1676 swap_pager_copy(
1677 backing_object,
1678 object,
1679 OFF_TO_IDX(object->backing_object_offset), TRUE);
1680
1681 /*
1682 * Free any cached pages from backing_object.
1683 */
1684 if (__predict_false(backing_object->cache != NULL))
1685 vm_page_cache_free(backing_object, 0, 0);
1686 }
1687 /*
1688 * Object now shadows whatever backing_object did.
1689 * Note that the reference to
1690 * backing_object->backing_object moves from within
1691 * backing_object to within object.
1692 */
1693 LIST_REMOVE(object, shadow_list);
1694 backing_object->shadow_count--;
1695 if (backing_object->backing_object) {
1696 VM_OBJECT_LOCK(backing_object->backing_object);
1697 LIST_REMOVE(backing_object, shadow_list);
1698 LIST_INSERT_HEAD(
1699 &backing_object->backing_object->shadow_head,
1700 object, shadow_list);
1701 /*
1702 * The shadow_count has not changed.
1703 */
1704 VM_OBJECT_UNLOCK(backing_object->backing_object);
1705 }
1706 object->backing_object = backing_object->backing_object;
1707 object->backing_object_offset +=
1708 backing_object->backing_object_offset;
1709
1710 /*
1711 * Discard backing_object.
1712 *
1713 * Since the backing object has no pages, no pager left,
1714 * and no object references within it, all that is
1715 * necessary is to dispose of it.
1716 */
1717 KASSERT(backing_object->ref_count == 1, (
1718 "backing_object %p was somehow re-referenced during collapse!",
1719 backing_object));
1720 VM_OBJECT_UNLOCK(backing_object);
1721 vm_object_destroy(backing_object);
1722
1723 object_collapses++;
1724 } else {
1725 vm_object_t new_backing_object;
1726
1727 /*
1728 * If we do not entirely shadow the backing object,
1729 * there is nothing we can do so we give up.
1730 */
1731 if (object->resident_page_count != object->size &&
1732 vm_object_backing_scan(object,
1733 OBSC_TEST_ALL_SHADOWED) == 0) {
1734 VM_OBJECT_UNLOCK(backing_object);
1735 break;
1736 }
1737
1738 /*
1739 * Make the parent shadow the next object in the
1740 * chain. Deallocating backing_object will not remove
1741 * it, since its reference count is at least 2.
1742 */
1743 LIST_REMOVE(object, shadow_list);
1744 backing_object->shadow_count--;
1745
1746 new_backing_object = backing_object->backing_object;
1747 if ((object->backing_object = new_backing_object) != NULL) {
1748 VM_OBJECT_LOCK(new_backing_object);
1749 LIST_INSERT_HEAD(
1750 &new_backing_object->shadow_head,
1751 object,
1752 shadow_list
1753 );
1754 new_backing_object->shadow_count++;
1755 vm_object_reference_locked(new_backing_object);
1756 VM_OBJECT_UNLOCK(new_backing_object);
1757 object->backing_object_offset +=
1758 backing_object->backing_object_offset;
1759 }
1760
1761 /*
1762 * Drop the reference count on backing_object. Since
1763 * its ref_count was at least 2, it will not vanish.
1764 */
1765 backing_object->ref_count--;
1766 VM_OBJECT_UNLOCK(backing_object);
1767 object_bypasses++;
1768 }
1769
1770 /*
1771 * Try again with this object's new backing object.
1772 */
1773 }
1774 }
1775
1776 /*
1777 * vm_object_page_remove:
1778 *
1779 * For the given object, either frees or invalidates each of the
1780 * specified pages. In general, a page is freed. However, if a
1781 * page is wired for any reason other than the existence of a
1782 * managed, wired mapping, then it may be invalidated but not
1783 * removed from the object. Pages are specified by the given
1784 * range ["start", "end") and Boolean "clean_only". As a
1785 * special case, if "end" is zero, then the range extends from
1786 * "start" to the end of the object. If "clean_only" is TRUE,
1787 * then only the non-dirty pages within the specified range are
1788 * affected.
1789 *
1790 * In general, this operation should only be performed on objects
1791 * that contain managed pages. There are two exceptions. First,
1792 * it may be performed on the kernel and kmem objects. Second,
1793 * it may be used by msync(..., MS_INVALIDATE) to invalidate
1794 * device-backed pages. In both of these cases, "clean_only"
1795 * must be FALSE.
1796 *
1797 * The object must be locked.
1798 */
1799 void
1800 vm_object_page_remove(vm_object_t object, vm_pindex_t start, vm_pindex_t end,
1801 boolean_t clean_only)
1802 {
1803 vm_page_t p, next;
1804 int wirings;
1805
1806 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
1807 if (object->resident_page_count == 0)
1808 goto skipmemq;
1809
1810 /*
1811 * Since physically-backed objects do not use managed pages, we can't
1812 * remove pages from the object (we must instead remove the page
1813 * references, and then destroy the object).
1814 */
1815 KASSERT(object->type != OBJT_PHYS || object == kernel_object ||
1816 object == kmem_object,
1817 ("attempt to remove pages from a physical object"));
1818
1819 vm_object_pip_add(object, 1);
1820 again:
1821 p = vm_page_find_least(object, start);
1822 vm_page_lock_queues();
1823 /*
1824 * Assert: the variable p is either (1) the page with the
1825 * least pindex greater than or equal to the parameter pindex
1826 * or (2) NULL.
1827 */
1828 for (;
1829 p != NULL && (p->pindex < end || end == 0);
1830 p = next) {
1831 next = TAILQ_NEXT(p, listq);
1832
1833 /*
1834 * If the page is wired for any reason besides the
1835 * existence of managed, wired mappings, then it cannot
1836 * be freed. For example, fictitious pages, which
1837 * represent device memory, are inherently wired and
1838 * cannot be freed. They can, however, be invalidated
1839 * if "clean_only" is FALSE.
1840 */
1841 if ((wirings = p->wire_count) != 0 &&
1842 (wirings = pmap_page_wired_mappings(p)) != p->wire_count) {
1843 /* Fictitious pages do not have managed mappings. */
1844 if ((p->flags & PG_FICTITIOUS) == 0)
1845 pmap_remove_all(p);
1846 /* Account for removal of managed, wired mappings. */
1847 p->wire_count -= wirings;
1848 if (!clean_only) {
1849 p->valid = 0;
1850 vm_page_undirty(p);
1851 }
1852 continue;
1853 }
1854 if (vm_page_sleep_if_busy(p, TRUE, "vmopar"))
1855 goto again;
1856 KASSERT((p->flags & PG_FICTITIOUS) == 0,
1857 ("vm_object_page_remove: page %p is fictitious", p));
1858 if (clean_only && p->valid) {
1859 pmap_remove_write(p);
1860 if (p->dirty)
1861 continue;
1862 }
1863 pmap_remove_all(p);
1864 /* Account for removal of managed, wired mappings. */
1865 if (wirings != 0)
1866 p->wire_count -= wirings;
1867 vm_page_free(p);
1868 }
1869 vm_page_unlock_queues();
1870 vm_object_pip_wakeup(object);
1871 skipmemq:
1872 if (__predict_false(object->cache != NULL))
1873 vm_page_cache_free(object, start, end);
1874 }
1875
1876 /*
1877 * vm_object_page_cache:
1878 *
1879 * For the given object, attempt to move the specified clean
1880 * pages to the cache queue. If a page is wired for any reason,
1881 * then it will not be changed. Pages are specified by the given
1882 * range ["start", "end"). As a special case, if "end" is zero,
1883 * then the range extends from "start" to the end of the object.
1884 * Any mappings to the specified pages are removed before the
1885 * pages are moved to the cache queue.
1886 *
1887 * This operation should only be performed on objects that
1888 * contain managed pages.
1889 *
1890 * The object must be locked.
1891 */
1892 void
1893 vm_object_page_cache(vm_object_t object, vm_pindex_t start, vm_pindex_t end)
1894 {
1895 vm_page_t p, next;
1896
1897 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
1898 KASSERT((object->type != OBJT_DEVICE && object->type != OBJT_SG &&
1899 object->type != OBJT_PHYS),
1900 ("vm_object_page_cache: illegal object %p", object));
1901 if (object->resident_page_count == 0)
1902 return;
1903 p = vm_page_find_least(object, start);
1904
1905 /*
1906 * Here, the variable "p" is either (1) the page with the least pindex
1907 * greater than or equal to the parameter "start" or (2) NULL.
1908 */
1909 vm_page_lock_queues();
1910 for (; p != NULL && (p->pindex < end || end == 0); p = next) {
1911 next = TAILQ_NEXT(p, listq);
1912
1913 vm_page_try_to_cache(p);
1914 }
1915 vm_page_unlock_queues();
1916 }
1917
1918 /*
1919 * Populate the specified range of the object with valid pages. Returns
1920 * TRUE if the range is successfully populated and FALSE otherwise.
1921 *
1922 * Note: This function should be optimized to pass a larger array of
1923 * pages to vm_pager_get_pages() before it is applied to a non-
1924 * OBJT_DEVICE object.
1925 *
1926 * The object must be locked.
1927 */
1928 boolean_t
1929 vm_object_populate(vm_object_t object, vm_pindex_t start, vm_pindex_t end)
1930 {
1931 vm_page_t m, ma[1];
1932 vm_pindex_t pindex;
1933 int rv;
1934
1935 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
1936 for (pindex = start; pindex < end; pindex++) {
1937 m = vm_page_grab(object, pindex, VM_ALLOC_NORMAL |
1938 VM_ALLOC_RETRY);
1939 if (m->valid != VM_PAGE_BITS_ALL) {
1940 ma[0] = m;
1941 rv = vm_pager_get_pages(object, ma, 1, 0);
1942 m = vm_page_lookup(object, pindex);
1943 if (m == NULL)
1944 break;
1945 if (rv != VM_PAGER_OK) {
1946 vm_page_lock_queues();
1947 vm_page_free(m);
1948 vm_page_unlock_queues();
1949 break;
1950 }
1951 }
1952 /*
1953 * Keep "m" busy because a subsequent iteration may unlock
1954 * the object.
1955 */
1956 }
1957 if (pindex > start) {
1958 m = vm_page_lookup(object, start);
1959 while (m != NULL && m->pindex < pindex) {
1960 vm_page_wakeup(m);
1961 m = TAILQ_NEXT(m, listq);
1962 }
1963 }
1964 return (pindex == end);
1965 }
1966
1967 /*
1968 * Routine: vm_object_coalesce
1969 * Function: Coalesces two objects backing up adjoining
1970 * regions of memory into a single object.
1971 *
1972 * returns TRUE if objects were combined.
1973 *
1974 * NOTE: Only works at the moment if the second object is NULL -
1975 * if it's not, which object do we lock first?
1976 *
1977 * Parameters:
1978 * prev_object First object to coalesce
1979 * prev_offset Offset into prev_object
1980 * prev_size Size of reference to prev_object
1981 * next_size Size of reference to the second object
1982 * reserved Indicator that extension region has
1983 * swap accounted for
1984 *
1985 * Conditions:
1986 * The object must *not* be locked.
1987 */
1988 boolean_t
1989 vm_object_coalesce(vm_object_t prev_object, vm_ooffset_t prev_offset,
1990 vm_size_t prev_size, vm_size_t next_size, boolean_t reserved)
1991 {
1992 vm_pindex_t next_pindex;
1993
1994 if (prev_object == NULL)
1995 return (TRUE);
1996 VM_OBJECT_LOCK(prev_object);
1997 if (prev_object->type != OBJT_DEFAULT &&
1998 prev_object->type != OBJT_SWAP) {
1999 VM_OBJECT_UNLOCK(prev_object);
2000 return (FALSE);
2001 }
2002
2003 /*
2004 * Try to collapse the object first
2005 */
2006 vm_object_collapse(prev_object);
2007
2008 /*
2009 * Can't coalesce if: . more than one reference . paged out . shadows
2010 * another object . has a copy elsewhere (any of which mean that the
2011 * pages not mapped to prev_entry may be in use anyway)
2012 */
2013 if (prev_object->backing_object != NULL) {
2014 VM_OBJECT_UNLOCK(prev_object);
2015 return (FALSE);
2016 }
2017
2018 prev_size >>= PAGE_SHIFT;
2019 next_size >>= PAGE_SHIFT;
2020 next_pindex = OFF_TO_IDX(prev_offset) + prev_size;
2021
2022 if ((prev_object->ref_count > 1) &&
2023 (prev_object->size != next_pindex)) {
2024 VM_OBJECT_UNLOCK(prev_object);
2025 return (FALSE);
2026 }
2027
2028 /*
2029 * Account for the charge.
2030 */
2031 if (prev_object->uip != NULL) {
2032
2033 /*
2034 * If prev_object was charged, then this mapping,
2035 * althought not charged now, may become writable
2036 * later. Non-NULL uip in the object would prevent
2037 * swap reservation during enabling of the write
2038 * access, so reserve swap now. Failed reservation
2039 * cause allocation of the separate object for the map
2040 * entry, and swap reservation for this entry is
2041 * managed in appropriate time.
2042 */
2043 if (!reserved && !swap_reserve_by_uid(ptoa(next_size),
2044 prev_object->uip)) {
2045 return (FALSE);
2046 }
2047 prev_object->charge += ptoa(next_size);
2048 }
2049
2050 /*
2051 * Remove any pages that may still be in the object from a previous
2052 * deallocation.
2053 */
2054 if (next_pindex < prev_object->size) {
2055 vm_object_page_remove(prev_object,
2056 next_pindex,
2057 next_pindex + next_size, FALSE);
2058 if (prev_object->type == OBJT_SWAP)
2059 swap_pager_freespace(prev_object,
2060 next_pindex, next_size);
2061 #if 0
2062 if (prev_object->uip != NULL) {
2063 KASSERT(prev_object->charge >=
2064 ptoa(prev_object->size - next_pindex),
2065 ("object %p overcharged 1 %jx %jx", prev_object,
2066 (uintmax_t)next_pindex, (uintmax_t)next_size));
2067 prev_object->charge -= ptoa(prev_object->size -
2068 next_pindex);
2069 }
2070 #endif
2071 }
2072
2073 /*
2074 * Extend the object if necessary.
2075 */
2076 if (next_pindex + next_size > prev_object->size)
2077 prev_object->size = next_pindex + next_size;
2078
2079 VM_OBJECT_UNLOCK(prev_object);
2080 return (TRUE);
2081 }
2082
2083 void
2084 vm_object_set_writeable_dirty(vm_object_t object)
2085 {
2086
2087 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
2088 if (object->type != OBJT_VNODE)
2089 return;
2090 object->generation++;
2091 if ((object->flags & OBJ_MIGHTBEDIRTY) != 0)
2092 return;
2093 vm_object_set_flag(object, OBJ_MIGHTBEDIRTY);
2094 }
2095
2096 #include "opt_ddb.h"
2097 #ifdef DDB
2098 #include <sys/kernel.h>
2099
2100 #include <sys/cons.h>
2101
2102 #include <ddb/ddb.h>
2103
2104 static int
2105 _vm_object_in_map(vm_map_t map, vm_object_t object, vm_map_entry_t entry)
2106 {
2107 vm_map_t tmpm;
2108 vm_map_entry_t tmpe;
2109 vm_object_t obj;
2110 int entcount;
2111
2112 if (map == 0)
2113 return 0;
2114
2115 if (entry == 0) {
2116 tmpe = map->header.next;
2117 entcount = map->nentries;
2118 while (entcount-- && (tmpe != &map->header)) {
2119 if (_vm_object_in_map(map, object, tmpe)) {
2120 return 1;
2121 }
2122 tmpe = tmpe->next;
2123 }
2124 } else if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
2125 tmpm = entry->object.sub_map;
2126 tmpe = tmpm->header.next;
2127 entcount = tmpm->nentries;
2128 while (entcount-- && tmpe != &tmpm->header) {
2129 if (_vm_object_in_map(tmpm, object, tmpe)) {
2130 return 1;
2131 }
2132 tmpe = tmpe->next;
2133 }
2134 } else if ((obj = entry->object.vm_object) != NULL) {
2135 for (; obj; obj = obj->backing_object)
2136 if (obj == object) {
2137 return 1;
2138 }
2139 }
2140 return 0;
2141 }
2142
2143 static int
2144 vm_object_in_map(vm_object_t object)
2145 {
2146 struct proc *p;
2147
2148 /* sx_slock(&allproc_lock); */
2149 FOREACH_PROC_IN_SYSTEM(p) {
2150 if (!p->p_vmspace /* || (p->p_flag & (P_SYSTEM|P_WEXIT)) */)
2151 continue;
2152 if (_vm_object_in_map(&p->p_vmspace->vm_map, object, 0)) {
2153 /* sx_sunlock(&allproc_lock); */
2154 return 1;
2155 }
2156 }
2157 /* sx_sunlock(&allproc_lock); */
2158 if (_vm_object_in_map(kernel_map, object, 0))
2159 return 1;
2160 if (_vm_object_in_map(kmem_map, object, 0))
2161 return 1;
2162 if (_vm_object_in_map(pager_map, object, 0))
2163 return 1;
2164 if (_vm_object_in_map(buffer_map, object, 0))
2165 return 1;
2166 return 0;
2167 }
2168
2169 DB_SHOW_COMMAND(vmochk, vm_object_check)
2170 {
2171 vm_object_t object;
2172
2173 /*
2174 * make sure that internal objs are in a map somewhere
2175 * and none have zero ref counts.
2176 */
2177 TAILQ_FOREACH(object, &vm_object_list, object_list) {
2178 if (object->handle == NULL &&
2179 (object->type == OBJT_DEFAULT || object->type == OBJT_SWAP)) {
2180 if (object->ref_count == 0) {
2181 db_printf("vmochk: internal obj has zero ref count: %ld\n",
2182 (long)object->size);
2183 }
2184 if (!vm_object_in_map(object)) {
2185 db_printf(
2186 "vmochk: internal obj is not in a map: "
2187 "ref: %d, size: %lu: 0x%lx, backing_object: %p\n",
2188 object->ref_count, (u_long)object->size,
2189 (u_long)object->size,
2190 (void *)object->backing_object);
2191 }
2192 }
2193 }
2194 }
2195
2196 /*
2197 * vm_object_print: [ debug ]
2198 */
2199 DB_SHOW_COMMAND(object, vm_object_print_static)
2200 {
2201 /* XXX convert args. */
2202 vm_object_t object = (vm_object_t)addr;
2203 boolean_t full = have_addr;
2204
2205 vm_page_t p;
2206
2207 /* XXX count is an (unused) arg. Avoid shadowing it. */
2208 #define count was_count
2209
2210 int count;
2211
2212 if (object == NULL)
2213 return;
2214
2215 db_iprintf(
2216 "Object %p: type=%d, size=0x%jx, res=%d, ref=%d, flags=0x%x uip %d charge %jx\n",
2217 object, (int)object->type, (uintmax_t)object->size,
2218 object->resident_page_count, object->ref_count, object->flags,
2219 object->uip ? object->uip->ui_uid : -1, (uintmax_t)object->charge);
2220 db_iprintf(" sref=%d, backing_object(%d)=(%p)+0x%jx\n",
2221 object->shadow_count,
2222 object->backing_object ? object->backing_object->ref_count : 0,
2223 object->backing_object, (uintmax_t)object->backing_object_offset);
2224
2225 if (!full)
2226 return;
2227
2228 db_indent += 2;
2229 count = 0;
2230 TAILQ_FOREACH(p, &object->memq, listq) {
2231 if (count == 0)
2232 db_iprintf("memory:=");
2233 else if (count == 6) {
2234 db_printf("\n");
2235 db_iprintf(" ...");
2236 count = 0;
2237 } else
2238 db_printf(",");
2239 count++;
2240
2241 db_printf("(off=0x%jx,page=0x%jx)",
2242 (uintmax_t)p->pindex, (uintmax_t)VM_PAGE_TO_PHYS(p));
2243 }
2244 if (count != 0)
2245 db_printf("\n");
2246 db_indent -= 2;
2247 }
2248
2249 /* XXX. */
2250 #undef count
2251
2252 /* XXX need this non-static entry for calling from vm_map_print. */
2253 void
2254 vm_object_print(
2255 /* db_expr_t */ long addr,
2256 boolean_t have_addr,
2257 /* db_expr_t */ long count,
2258 char *modif)
2259 {
2260 vm_object_print_static(addr, have_addr, count, modif);
2261 }
2262
2263 DB_SHOW_COMMAND(vmopag, vm_object_print_pages)
2264 {
2265 vm_object_t object;
2266 vm_pindex_t fidx;
2267 vm_paddr_t pa;
2268 vm_page_t m, prev_m;
2269 int rcount, nl, c;
2270
2271 nl = 0;
2272 TAILQ_FOREACH(object, &vm_object_list, object_list) {
2273 db_printf("new object: %p\n", (void *)object);
2274 if (nl > 18) {
2275 c = cngetc();
2276 if (c != ' ')
2277 return;
2278 nl = 0;
2279 }
2280 nl++;
2281 rcount = 0;
2282 fidx = 0;
2283 pa = -1;
2284 TAILQ_FOREACH(m, &object->memq, listq) {
2285 if (m->pindex > 128)
2286 break;
2287 if ((prev_m = TAILQ_PREV(m, pglist, listq)) != NULL &&
2288 prev_m->pindex + 1 != m->pindex) {
2289 if (rcount) {
2290 db_printf(" index(%ld)run(%d)pa(0x%lx)\n",
2291 (long)fidx, rcount, (long)pa);
2292 if (nl > 18) {
2293 c = cngetc();
2294 if (c != ' ')
2295 return;
2296 nl = 0;
2297 }
2298 nl++;
2299 rcount = 0;
2300 }
2301 }
2302 if (rcount &&
2303 (VM_PAGE_TO_PHYS(m) == pa + rcount * PAGE_SIZE)) {
2304 ++rcount;
2305 continue;
2306 }
2307 if (rcount) {
2308 db_printf(" index(%ld)run(%d)pa(0x%lx)\n",
2309 (long)fidx, rcount, (long)pa);
2310 if (nl > 18) {
2311 c = cngetc();
2312 if (c != ' ')
2313 return;
2314 nl = 0;
2315 }
2316 nl++;
2317 }
2318 fidx = m->pindex;
2319 pa = VM_PAGE_TO_PHYS(m);
2320 rcount = 1;
2321 }
2322 if (rcount) {
2323 db_printf(" index(%ld)run(%d)pa(0x%lx)\n",
2324 (long)fidx, rcount, (long)pa);
2325 if (nl > 18) {
2326 c = cngetc();
2327 if (c != ' ')
2328 return;
2329 nl = 0;
2330 }
2331 nl++;
2332 }
2333 }
2334 }
2335 #endif /* DDB */
Cache object: 9e59f75a20c9c807c1d8030223242834
|