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/9.0/sys/vm/vm_object.c 225418 2011-09-06 10:30:11Z kib $");
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 old_msync;
100 SYSCTL_INT(_vm, OID_AUTO, old_msync, CTLFLAG_RW, &old_msync, 0,
101 "Use old (insecure) msync behavior");
102
103 static int vm_object_page_collect_flush(vm_object_t object, vm_page_t p,
104 int pagerflags, int flags, int *clearobjflags);
105 static boolean_t vm_object_page_remove_write(vm_page_t p, int flags,
106 int *clearobjflags);
107 static void vm_object_qcollapse(vm_object_t object);
108 static void vm_object_vndeallocate(vm_object_t object);
109
110 /*
111 * Virtual memory objects maintain the actual data
112 * associated with allocated virtual memory. A given
113 * page of memory exists within exactly one object.
114 *
115 * An object is only deallocated when all "references"
116 * are given up. Only one "reference" to a given
117 * region of an object should be writeable.
118 *
119 * Associated with each object is a list of all resident
120 * memory pages belonging to that object; this list is
121 * maintained by the "vm_page" module, and locked by the object's
122 * lock.
123 *
124 * Each object also records a "pager" routine which is
125 * used to retrieve (and store) pages to the proper backing
126 * storage. In addition, objects may be backed by other
127 * objects from which they were virtual-copied.
128 *
129 * The only items within the object structure which are
130 * modified after time of creation are:
131 * reference count locked by object's lock
132 * pager routine locked by object's lock
133 *
134 */
135
136 struct object_q vm_object_list;
137 struct mtx vm_object_list_mtx; /* lock for object list and count */
138
139 struct vm_object kernel_object_store;
140 struct vm_object kmem_object_store;
141
142 SYSCTL_NODE(_vm_stats, OID_AUTO, object, CTLFLAG_RD, 0, "VM object stats");
143
144 static long object_collapses;
145 SYSCTL_LONG(_vm_stats_object, OID_AUTO, collapses, CTLFLAG_RD,
146 &object_collapses, 0, "VM object collapses");
147
148 static long object_bypasses;
149 SYSCTL_LONG(_vm_stats_object, OID_AUTO, bypasses, CTLFLAG_RD,
150 &object_bypasses, 0, "VM object bypasses");
151
152 static uma_zone_t obj_zone;
153
154 static int vm_object_zinit(void *mem, int size, int flags);
155
156 #ifdef INVARIANTS
157 static void vm_object_zdtor(void *mem, int size, void *arg);
158
159 static void
160 vm_object_zdtor(void *mem, int size, void *arg)
161 {
162 vm_object_t object;
163
164 object = (vm_object_t)mem;
165 KASSERT(TAILQ_EMPTY(&object->memq),
166 ("object %p has resident pages",
167 object));
168 #if VM_NRESERVLEVEL > 0
169 KASSERT(LIST_EMPTY(&object->rvq),
170 ("object %p has reservations",
171 object));
172 #endif
173 KASSERT(object->cache == NULL,
174 ("object %p has cached pages",
175 object));
176 KASSERT(object->paging_in_progress == 0,
177 ("object %p paging_in_progress = %d",
178 object, object->paging_in_progress));
179 KASSERT(object->resident_page_count == 0,
180 ("object %p resident_page_count = %d",
181 object, object->resident_page_count));
182 KASSERT(object->shadow_count == 0,
183 ("object %p shadow_count = %d",
184 object, object->shadow_count));
185 }
186 #endif
187
188 static int
189 vm_object_zinit(void *mem, int size, int flags)
190 {
191 vm_object_t object;
192
193 object = (vm_object_t)mem;
194 bzero(&object->mtx, sizeof(object->mtx));
195 VM_OBJECT_LOCK_INIT(object, "standard object");
196
197 /* These are true for any object that has been freed */
198 object->paging_in_progress = 0;
199 object->resident_page_count = 0;
200 object->shadow_count = 0;
201 return (0);
202 }
203
204 void
205 _vm_object_allocate(objtype_t type, vm_pindex_t size, vm_object_t object)
206 {
207
208 TAILQ_INIT(&object->memq);
209 LIST_INIT(&object->shadow_head);
210
211 object->root = NULL;
212 object->type = type;
213 object->size = size;
214 object->generation = 1;
215 object->ref_count = 1;
216 object->memattr = VM_MEMATTR_DEFAULT;
217 object->flags = 0;
218 object->cred = NULL;
219 object->charge = 0;
220 if ((object->type == OBJT_DEFAULT) || (object->type == OBJT_SWAP))
221 object->flags = OBJ_ONEMAPPING;
222 object->pg_color = 0;
223 object->handle = NULL;
224 object->backing_object = NULL;
225 object->backing_object_offset = (vm_ooffset_t) 0;
226 #if VM_NRESERVLEVEL > 0
227 LIST_INIT(&object->rvq);
228 #endif
229 object->cache = NULL;
230
231 mtx_lock(&vm_object_list_mtx);
232 TAILQ_INSERT_TAIL(&vm_object_list, object, object_list);
233 mtx_unlock(&vm_object_list_mtx);
234 }
235
236 /*
237 * vm_object_init:
238 *
239 * Initialize the VM objects module.
240 */
241 void
242 vm_object_init(void)
243 {
244 TAILQ_INIT(&vm_object_list);
245 mtx_init(&vm_object_list_mtx, "vm object_list", NULL, MTX_DEF);
246
247 VM_OBJECT_LOCK_INIT(kernel_object, "kernel object");
248 _vm_object_allocate(OBJT_PHYS, OFF_TO_IDX(VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS),
249 kernel_object);
250 #if VM_NRESERVLEVEL > 0
251 kernel_object->flags |= OBJ_COLORED;
252 kernel_object->pg_color = (u_short)atop(VM_MIN_KERNEL_ADDRESS);
253 #endif
254
255 VM_OBJECT_LOCK_INIT(kmem_object, "kmem object");
256 _vm_object_allocate(OBJT_PHYS, OFF_TO_IDX(VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS),
257 kmem_object);
258 #if VM_NRESERVLEVEL > 0
259 kmem_object->flags |= OBJ_COLORED;
260 kmem_object->pg_color = (u_short)atop(VM_MIN_KERNEL_ADDRESS);
261 #endif
262
263 /*
264 * The lock portion of struct vm_object must be type stable due
265 * to vm_pageout_fallback_object_lock locking a vm object
266 * without holding any references to it.
267 */
268 obj_zone = uma_zcreate("VM OBJECT", sizeof (struct vm_object), NULL,
269 #ifdef INVARIANTS
270 vm_object_zdtor,
271 #else
272 NULL,
273 #endif
274 vm_object_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_VM|UMA_ZONE_NOFREE);
275 }
276
277 void
278 vm_object_clear_flag(vm_object_t object, u_short bits)
279 {
280
281 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
282 object->flags &= ~bits;
283 }
284
285 /*
286 * Sets the default memory attribute for the specified object. Pages
287 * that are allocated to this object are by default assigned this memory
288 * attribute.
289 *
290 * Presently, this function must be called before any pages are allocated
291 * to the object. In the future, this requirement may be relaxed for
292 * "default" and "swap" objects.
293 */
294 int
295 vm_object_set_memattr(vm_object_t object, vm_memattr_t memattr)
296 {
297
298 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
299 switch (object->type) {
300 case OBJT_DEFAULT:
301 case OBJT_DEVICE:
302 case OBJT_PHYS:
303 case OBJT_SG:
304 case OBJT_SWAP:
305 case OBJT_VNODE:
306 if (!TAILQ_EMPTY(&object->memq))
307 return (KERN_FAILURE);
308 break;
309 case OBJT_DEAD:
310 return (KERN_INVALID_ARGUMENT);
311 }
312 object->memattr = memattr;
313 return (KERN_SUCCESS);
314 }
315
316 void
317 vm_object_pip_add(vm_object_t object, short i)
318 {
319
320 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
321 object->paging_in_progress += i;
322 }
323
324 void
325 vm_object_pip_subtract(vm_object_t object, short i)
326 {
327
328 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
329 object->paging_in_progress -= i;
330 }
331
332 void
333 vm_object_pip_wakeup(vm_object_t object)
334 {
335
336 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
337 object->paging_in_progress--;
338 if ((object->flags & OBJ_PIPWNT) && object->paging_in_progress == 0) {
339 vm_object_clear_flag(object, OBJ_PIPWNT);
340 wakeup(object);
341 }
342 }
343
344 void
345 vm_object_pip_wakeupn(vm_object_t object, short i)
346 {
347
348 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
349 if (i)
350 object->paging_in_progress -= i;
351 if ((object->flags & OBJ_PIPWNT) && object->paging_in_progress == 0) {
352 vm_object_clear_flag(object, OBJ_PIPWNT);
353 wakeup(object);
354 }
355 }
356
357 void
358 vm_object_pip_wait(vm_object_t object, char *waitid)
359 {
360
361 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
362 while (object->paging_in_progress) {
363 object->flags |= OBJ_PIPWNT;
364 msleep(object, VM_OBJECT_MTX(object), PVM, waitid, 0);
365 }
366 }
367
368 /*
369 * vm_object_allocate:
370 *
371 * Returns a new object with the given size.
372 */
373 vm_object_t
374 vm_object_allocate(objtype_t type, vm_pindex_t size)
375 {
376 vm_object_t object;
377
378 object = (vm_object_t)uma_zalloc(obj_zone, M_WAITOK);
379 _vm_object_allocate(type, size, object);
380 return (object);
381 }
382
383
384 /*
385 * vm_object_reference:
386 *
387 * Gets another reference to the given object. Note: OBJ_DEAD
388 * objects can be referenced during final cleaning.
389 */
390 void
391 vm_object_reference(vm_object_t object)
392 {
393 if (object == NULL)
394 return;
395 VM_OBJECT_LOCK(object);
396 vm_object_reference_locked(object);
397 VM_OBJECT_UNLOCK(object);
398 }
399
400 /*
401 * vm_object_reference_locked:
402 *
403 * Gets another reference to the given object.
404 *
405 * The object must be locked.
406 */
407 void
408 vm_object_reference_locked(vm_object_t object)
409 {
410 struct vnode *vp;
411
412 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
413 object->ref_count++;
414 if (object->type == OBJT_VNODE) {
415 vp = object->handle;
416 vref(vp);
417 }
418 }
419
420 /*
421 * Handle deallocating an object of type OBJT_VNODE.
422 */
423 static void
424 vm_object_vndeallocate(vm_object_t object)
425 {
426 struct vnode *vp = (struct vnode *) object->handle;
427
428 VFS_ASSERT_GIANT(vp->v_mount);
429 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
430 KASSERT(object->type == OBJT_VNODE,
431 ("vm_object_vndeallocate: not a vnode object"));
432 KASSERT(vp != NULL, ("vm_object_vndeallocate: missing vp"));
433 #ifdef INVARIANTS
434 if (object->ref_count == 0) {
435 vprint("vm_object_vndeallocate", vp);
436 panic("vm_object_vndeallocate: bad object reference count");
437 }
438 #endif
439
440 if (object->ref_count > 1) {
441 object->ref_count--;
442 VM_OBJECT_UNLOCK(object);
443 /* vrele may need the vnode lock. */
444 vrele(vp);
445 } else {
446 vhold(vp);
447 VM_OBJECT_UNLOCK(object);
448 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
449 vdrop(vp);
450 VM_OBJECT_LOCK(object);
451 object->ref_count--;
452 if (object->type == OBJT_DEAD) {
453 VM_OBJECT_UNLOCK(object);
454 VOP_UNLOCK(vp, 0);
455 } else {
456 if (object->ref_count == 0)
457 vp->v_vflag &= ~VV_TEXT;
458 VM_OBJECT_UNLOCK(object);
459 vput(vp);
460 }
461 }
462 }
463
464 /*
465 * vm_object_deallocate:
466 *
467 * Release a reference to the specified object,
468 * gained either through a vm_object_allocate
469 * or a vm_object_reference call. When all references
470 * are gone, storage associated with this object
471 * may be relinquished.
472 *
473 * No object may be locked.
474 */
475 void
476 vm_object_deallocate(vm_object_t object)
477 {
478 vm_object_t temp;
479
480 while (object != NULL) {
481 int vfslocked;
482
483 vfslocked = 0;
484 restart:
485 VM_OBJECT_LOCK(object);
486 if (object->type == OBJT_VNODE) {
487 struct vnode *vp = (struct vnode *) object->handle;
488
489 /*
490 * Conditionally acquire Giant for a vnode-backed
491 * object. We have to be careful since the type of
492 * a vnode object can change while the object is
493 * unlocked.
494 */
495 if (VFS_NEEDSGIANT(vp->v_mount) && !vfslocked) {
496 vfslocked = 1;
497 if (!mtx_trylock(&Giant)) {
498 VM_OBJECT_UNLOCK(object);
499 mtx_lock(&Giant);
500 goto restart;
501 }
502 }
503 vm_object_vndeallocate(object);
504 VFS_UNLOCK_GIANT(vfslocked);
505 return;
506 } else
507 /*
508 * This is to handle the case that the object
509 * changed type while we dropped its lock to
510 * obtain Giant.
511 */
512 VFS_UNLOCK_GIANT(vfslocked);
513
514 KASSERT(object->ref_count != 0,
515 ("vm_object_deallocate: object deallocated too many times: %d", object->type));
516
517 /*
518 * If the reference count goes to 0 we start calling
519 * vm_object_terminate() on the object chain.
520 * A ref count of 1 may be a special case depending on the
521 * shadow count being 0 or 1.
522 */
523 object->ref_count--;
524 if (object->ref_count > 1) {
525 VM_OBJECT_UNLOCK(object);
526 return;
527 } else if (object->ref_count == 1) {
528 if (object->shadow_count == 0 &&
529 object->handle == NULL &&
530 (object->type == OBJT_DEFAULT ||
531 object->type == OBJT_SWAP)) {
532 vm_object_set_flag(object, OBJ_ONEMAPPING);
533 } else if ((object->shadow_count == 1) &&
534 (object->handle == NULL) &&
535 (object->type == OBJT_DEFAULT ||
536 object->type == OBJT_SWAP)) {
537 vm_object_t robject;
538
539 robject = LIST_FIRST(&object->shadow_head);
540 KASSERT(robject != NULL,
541 ("vm_object_deallocate: ref_count: %d, shadow_count: %d",
542 object->ref_count,
543 object->shadow_count));
544 if (!VM_OBJECT_TRYLOCK(robject)) {
545 /*
546 * Avoid a potential deadlock.
547 */
548 object->ref_count++;
549 VM_OBJECT_UNLOCK(object);
550 /*
551 * More likely than not the thread
552 * holding robject's lock has lower
553 * priority than the current thread.
554 * Let the lower priority thread run.
555 */
556 pause("vmo_de", 1);
557 continue;
558 }
559 /*
560 * Collapse object into its shadow unless its
561 * shadow is dead. In that case, object will
562 * be deallocated by the thread that is
563 * deallocating its shadow.
564 */
565 if ((robject->flags & OBJ_DEAD) == 0 &&
566 (robject->handle == NULL) &&
567 (robject->type == OBJT_DEFAULT ||
568 robject->type == OBJT_SWAP)) {
569
570 robject->ref_count++;
571 retry:
572 if (robject->paging_in_progress) {
573 VM_OBJECT_UNLOCK(object);
574 vm_object_pip_wait(robject,
575 "objde1");
576 temp = robject->backing_object;
577 if (object == temp) {
578 VM_OBJECT_LOCK(object);
579 goto retry;
580 }
581 } else if (object->paging_in_progress) {
582 VM_OBJECT_UNLOCK(robject);
583 object->flags |= OBJ_PIPWNT;
584 msleep(object,
585 VM_OBJECT_MTX(object),
586 PDROP | PVM, "objde2", 0);
587 VM_OBJECT_LOCK(robject);
588 temp = robject->backing_object;
589 if (object == temp) {
590 VM_OBJECT_LOCK(object);
591 goto retry;
592 }
593 } else
594 VM_OBJECT_UNLOCK(object);
595
596 if (robject->ref_count == 1) {
597 robject->ref_count--;
598 object = robject;
599 goto doterm;
600 }
601 object = robject;
602 vm_object_collapse(object);
603 VM_OBJECT_UNLOCK(object);
604 continue;
605 }
606 VM_OBJECT_UNLOCK(robject);
607 }
608 VM_OBJECT_UNLOCK(object);
609 return;
610 }
611 doterm:
612 temp = object->backing_object;
613 if (temp != NULL) {
614 VM_OBJECT_LOCK(temp);
615 LIST_REMOVE(object, shadow_list);
616 temp->shadow_count--;
617 VM_OBJECT_UNLOCK(temp);
618 object->backing_object = NULL;
619 }
620 /*
621 * Don't double-terminate, we could be in a termination
622 * recursion due to the terminate having to sync data
623 * to disk.
624 */
625 if ((object->flags & OBJ_DEAD) == 0)
626 vm_object_terminate(object);
627 else
628 VM_OBJECT_UNLOCK(object);
629 object = temp;
630 }
631 }
632
633 /*
634 * vm_object_destroy removes the object from the global object list
635 * and frees the space for the object.
636 */
637 void
638 vm_object_destroy(vm_object_t object)
639 {
640
641 /*
642 * Remove the object from the global object list.
643 */
644 mtx_lock(&vm_object_list_mtx);
645 TAILQ_REMOVE(&vm_object_list, object, object_list);
646 mtx_unlock(&vm_object_list_mtx);
647
648 /*
649 * Release the allocation charge.
650 */
651 if (object->cred != NULL) {
652 KASSERT(object->type == OBJT_DEFAULT ||
653 object->type == OBJT_SWAP,
654 ("vm_object_terminate: non-swap obj %p has cred",
655 object));
656 swap_release_by_cred(object->charge, object->cred);
657 object->charge = 0;
658 crfree(object->cred);
659 object->cred = NULL;
660 }
661
662 /*
663 * Free the space for the object.
664 */
665 uma_zfree(obj_zone, object);
666 }
667
668 /*
669 * vm_object_terminate actually destroys the specified object, freeing
670 * up all previously used resources.
671 *
672 * The object must be locked.
673 * This routine may block.
674 */
675 void
676 vm_object_terminate(vm_object_t object)
677 {
678 vm_page_t p, p_next;
679
680 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
681
682 /*
683 * Make sure no one uses us.
684 */
685 vm_object_set_flag(object, OBJ_DEAD);
686
687 /*
688 * wait for the pageout daemon to be done with the object
689 */
690 vm_object_pip_wait(object, "objtrm");
691
692 KASSERT(!object->paging_in_progress,
693 ("vm_object_terminate: pageout in progress"));
694
695 /*
696 * Clean and free the pages, as appropriate. All references to the
697 * object are gone, so we don't need to lock it.
698 */
699 if (object->type == OBJT_VNODE) {
700 struct vnode *vp = (struct vnode *)object->handle;
701
702 /*
703 * Clean pages and flush buffers.
704 */
705 vm_object_page_clean(object, 0, 0, OBJPC_SYNC);
706 VM_OBJECT_UNLOCK(object);
707
708 vinvalbuf(vp, V_SAVE, 0, 0);
709
710 VM_OBJECT_LOCK(object);
711 }
712
713 KASSERT(object->ref_count == 0,
714 ("vm_object_terminate: object with references, ref_count=%d",
715 object->ref_count));
716
717 /*
718 * Free any remaining pageable pages. This also removes them from the
719 * paging queues. However, don't free wired pages, just remove them
720 * from the object. Rather than incrementally removing each page from
721 * the object, the page and object are reset to any empty state.
722 */
723 TAILQ_FOREACH_SAFE(p, &object->memq, listq, p_next) {
724 KASSERT(!p->busy && (p->oflags & VPO_BUSY) == 0,
725 ("vm_object_terminate: freeing busy page %p", p));
726 vm_page_lock(p);
727 /*
728 * Optimize the page's removal from the object by resetting
729 * its "object" field. Specifically, if the page is not
730 * wired, then the effect of this assignment is that
731 * vm_page_free()'s call to vm_page_remove() will return
732 * immediately without modifying the page or the object.
733 */
734 p->object = NULL;
735 if (p->wire_count == 0) {
736 vm_page_free(p);
737 PCPU_INC(cnt.v_pfree);
738 }
739 vm_page_unlock(p);
740 }
741 /*
742 * If the object contained any pages, then reset it to an empty state.
743 * None of the object's fields, including "resident_page_count", were
744 * modified by the preceding loop.
745 */
746 if (object->resident_page_count != 0) {
747 object->root = NULL;
748 TAILQ_INIT(&object->memq);
749 object->resident_page_count = 0;
750 if (object->type == OBJT_VNODE)
751 vdrop(object->handle);
752 }
753
754 #if VM_NRESERVLEVEL > 0
755 if (__predict_false(!LIST_EMPTY(&object->rvq)))
756 vm_reserv_break_all(object);
757 #endif
758 if (__predict_false(object->cache != NULL))
759 vm_page_cache_free(object, 0, 0);
760
761 /*
762 * Let the pager know object is dead.
763 */
764 vm_pager_deallocate(object);
765 VM_OBJECT_UNLOCK(object);
766
767 vm_object_destroy(object);
768 }
769
770 /*
771 * Make the page read-only so that we can clear the object flags. However, if
772 * this is a nosync mmap then the object is likely to stay dirty so do not
773 * mess with the page and do not clear the object flags. Returns TRUE if the
774 * page should be flushed, and FALSE otherwise.
775 */
776 static boolean_t
777 vm_object_page_remove_write(vm_page_t p, int flags, int *clearobjflags)
778 {
779
780 /*
781 * If we have been asked to skip nosync pages and this is a
782 * nosync page, skip it. Note that the object flags were not
783 * cleared in this case so we do not have to set them.
784 */
785 if ((flags & OBJPC_NOSYNC) != 0 && (p->oflags & VPO_NOSYNC) != 0) {
786 *clearobjflags = 0;
787 return (FALSE);
788 } else {
789 pmap_remove_write(p);
790 return (p->dirty != 0);
791 }
792 }
793
794 /*
795 * vm_object_page_clean
796 *
797 * Clean all dirty pages in the specified range of object. Leaves page
798 * on whatever queue it is currently on. If NOSYNC is set then do not
799 * write out pages with VPO_NOSYNC set (originally comes from MAP_NOSYNC),
800 * leaving the object dirty.
801 *
802 * When stuffing pages asynchronously, allow clustering. XXX we need a
803 * synchronous clustering mode implementation.
804 *
805 * Odd semantics: if start == end, we clean everything.
806 *
807 * The object must be locked.
808 */
809 void
810 vm_object_page_clean(vm_object_t object, vm_ooffset_t start, vm_ooffset_t end,
811 int flags)
812 {
813 vm_page_t np, p;
814 vm_pindex_t pi, tend, tstart;
815 int clearobjflags, curgeneration, n, pagerflags;
816
817 mtx_assert(&vm_page_queue_mtx, MA_NOTOWNED);
818 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
819 KASSERT(object->type == OBJT_VNODE, ("Not a vnode object"));
820 if ((object->flags & OBJ_MIGHTBEDIRTY) == 0 ||
821 object->resident_page_count == 0)
822 return;
823
824 pagerflags = (flags & (OBJPC_SYNC | OBJPC_INVAL)) != 0 ?
825 VM_PAGER_PUT_SYNC : VM_PAGER_CLUSTER_OK;
826 pagerflags |= (flags & OBJPC_INVAL) != 0 ? VM_PAGER_PUT_INVAL : 0;
827
828 tstart = OFF_TO_IDX(start);
829 tend = (end == 0) ? object->size : OFF_TO_IDX(end + PAGE_MASK);
830 clearobjflags = tstart == 0 && tend >= object->size;
831
832 rescan:
833 curgeneration = object->generation;
834
835 for (p = vm_page_find_least(object, tstart); p != NULL; p = np) {
836 pi = p->pindex;
837 if (pi >= tend)
838 break;
839 np = TAILQ_NEXT(p, listq);
840 if (p->valid == 0)
841 continue;
842 if (vm_page_sleep_if_busy(p, TRUE, "vpcwai")) {
843 if (object->generation != curgeneration)
844 goto rescan;
845 np = vm_page_find_least(object, pi);
846 continue;
847 }
848 if (!vm_object_page_remove_write(p, flags, &clearobjflags))
849 continue;
850
851 n = vm_object_page_collect_flush(object, p, pagerflags,
852 flags, &clearobjflags);
853 if (object->generation != curgeneration)
854 goto rescan;
855
856 /*
857 * If the VOP_PUTPAGES() did a truncated write, so
858 * that even the first page of the run is not fully
859 * written, vm_pageout_flush() returns 0 as the run
860 * length. Since the condition that caused truncated
861 * write may be permanent, e.g. exhausted free space,
862 * accepting n == 0 would cause an infinite loop.
863 *
864 * Forwarding the iterator leaves the unwritten page
865 * behind, but there is not much we can do there if
866 * filesystem refuses to write it.
867 */
868 if (n == 0)
869 n = 1;
870 np = vm_page_find_least(object, pi + n);
871 }
872 #if 0
873 VOP_FSYNC(vp, (pagerflags & VM_PAGER_PUT_SYNC) ? MNT_WAIT : 0);
874 #endif
875
876 if (clearobjflags)
877 vm_object_clear_flag(object, OBJ_MIGHTBEDIRTY);
878 }
879
880 static int
881 vm_object_page_collect_flush(vm_object_t object, vm_page_t p, int pagerflags,
882 int flags, int *clearobjflags)
883 {
884 vm_page_t ma[vm_pageout_page_count], p_first, tp;
885 int count, i, mreq, runlen;
886
887 mtx_assert(&vm_page_queue_mtx, MA_NOTOWNED);
888 vm_page_lock_assert(p, MA_NOTOWNED);
889 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
890
891 count = 1;
892 mreq = 0;
893
894 for (tp = p; count < vm_pageout_page_count; count++) {
895 tp = vm_page_next(tp);
896 if (tp == NULL || tp->busy != 0 || (tp->oflags & VPO_BUSY) != 0)
897 break;
898 if (!vm_object_page_remove_write(tp, flags, clearobjflags))
899 break;
900 }
901
902 for (p_first = p; count < vm_pageout_page_count; count++) {
903 tp = vm_page_prev(p_first);
904 if (tp == NULL || tp->busy != 0 || (tp->oflags & VPO_BUSY) != 0)
905 break;
906 if (!vm_object_page_remove_write(tp, flags, clearobjflags))
907 break;
908 p_first = tp;
909 mreq++;
910 }
911
912 for (tp = p_first, i = 0; i < count; tp = TAILQ_NEXT(tp, listq), i++)
913 ma[i] = tp;
914
915 vm_pageout_flush(ma, count, pagerflags, mreq, &runlen);
916 return (runlen);
917 }
918
919 /*
920 * Note that there is absolutely no sense in writing out
921 * anonymous objects, so we track down the vnode object
922 * to write out.
923 * We invalidate (remove) all pages from the address space
924 * for semantic correctness.
925 *
926 * If the backing object is a device object with unmanaged pages, then any
927 * mappings to the specified range of pages must be removed before this
928 * function is called.
929 *
930 * Note: certain anonymous maps, such as MAP_NOSYNC maps,
931 * may start out with a NULL object.
932 */
933 void
934 vm_object_sync(vm_object_t object, vm_ooffset_t offset, vm_size_t size,
935 boolean_t syncio, boolean_t invalidate)
936 {
937 vm_object_t backing_object;
938 struct vnode *vp;
939 struct mount *mp;
940 int flags;
941
942 if (object == NULL)
943 return;
944 VM_OBJECT_LOCK(object);
945 while ((backing_object = object->backing_object) != NULL) {
946 VM_OBJECT_LOCK(backing_object);
947 offset += object->backing_object_offset;
948 VM_OBJECT_UNLOCK(object);
949 object = backing_object;
950 if (object->size < OFF_TO_IDX(offset + size))
951 size = IDX_TO_OFF(object->size) - offset;
952 }
953 /*
954 * Flush pages if writing is allowed, invalidate them
955 * if invalidation requested. Pages undergoing I/O
956 * will be ignored by vm_object_page_remove().
957 *
958 * We cannot lock the vnode and then wait for paging
959 * to complete without deadlocking against vm_fault.
960 * Instead we simply call vm_object_page_remove() and
961 * allow it to block internally on a page-by-page
962 * basis when it encounters pages undergoing async
963 * I/O.
964 */
965 if (object->type == OBJT_VNODE &&
966 (object->flags & OBJ_MIGHTBEDIRTY) != 0) {
967 int vfslocked;
968 vp = object->handle;
969 VM_OBJECT_UNLOCK(object);
970 (void) vn_start_write(vp, &mp, V_WAIT);
971 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
972 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
973 flags = (syncio || invalidate) ? OBJPC_SYNC : 0;
974 flags |= invalidate ? OBJPC_INVAL : 0;
975 VM_OBJECT_LOCK(object);
976 vm_object_page_clean(object, offset, offset + size, flags);
977 VM_OBJECT_UNLOCK(object);
978 VOP_UNLOCK(vp, 0);
979 VFS_UNLOCK_GIANT(vfslocked);
980 vn_finished_write(mp);
981 VM_OBJECT_LOCK(object);
982 }
983 if ((object->type == OBJT_VNODE ||
984 object->type == OBJT_DEVICE) && invalidate) {
985 if (object->type == OBJT_DEVICE)
986 /*
987 * The option OBJPR_NOTMAPPED must be passed here
988 * because vm_object_page_remove() cannot remove
989 * unmanaged mappings.
990 */
991 flags = OBJPR_NOTMAPPED;
992 else if (old_msync)
993 flags = 0;
994 else
995 flags = OBJPR_CLEANONLY;
996 vm_object_page_remove(object, OFF_TO_IDX(offset),
997 OFF_TO_IDX(offset + size + PAGE_MASK), flags);
998 }
999 VM_OBJECT_UNLOCK(object);
1000 }
1001
1002 /*
1003 * vm_object_madvise:
1004 *
1005 * Implements the madvise function at the object/page level.
1006 *
1007 * MADV_WILLNEED (any object)
1008 *
1009 * Activate the specified pages if they are resident.
1010 *
1011 * MADV_DONTNEED (any object)
1012 *
1013 * Deactivate the specified pages if they are resident.
1014 *
1015 * MADV_FREE (OBJT_DEFAULT/OBJT_SWAP objects,
1016 * OBJ_ONEMAPPING only)
1017 *
1018 * Deactivate and clean the specified pages if they are
1019 * resident. This permits the process to reuse the pages
1020 * without faulting or the kernel to reclaim the pages
1021 * without I/O.
1022 */
1023 void
1024 vm_object_madvise(vm_object_t object, vm_pindex_t pindex, int count, int advise)
1025 {
1026 vm_pindex_t end, tpindex;
1027 vm_object_t backing_object, tobject;
1028 vm_page_t m;
1029
1030 if (object == NULL)
1031 return;
1032 VM_OBJECT_LOCK(object);
1033 end = pindex + count;
1034 /*
1035 * Locate and adjust resident pages
1036 */
1037 for (; pindex < end; pindex += 1) {
1038 relookup:
1039 tobject = object;
1040 tpindex = pindex;
1041 shadowlookup:
1042 /*
1043 * MADV_FREE only operates on OBJT_DEFAULT or OBJT_SWAP pages
1044 * and those pages must be OBJ_ONEMAPPING.
1045 */
1046 if (advise == MADV_FREE) {
1047 if ((tobject->type != OBJT_DEFAULT &&
1048 tobject->type != OBJT_SWAP) ||
1049 (tobject->flags & OBJ_ONEMAPPING) == 0) {
1050 goto unlock_tobject;
1051 }
1052 } else if (tobject->type == OBJT_PHYS)
1053 goto unlock_tobject;
1054 m = vm_page_lookup(tobject, tpindex);
1055 if (m == NULL && advise == MADV_WILLNEED) {
1056 /*
1057 * If the page is cached, reactivate it.
1058 */
1059 m = vm_page_alloc(tobject, tpindex, VM_ALLOC_IFCACHED |
1060 VM_ALLOC_NOBUSY);
1061 }
1062 if (m == NULL) {
1063 /*
1064 * There may be swap even if there is no backing page
1065 */
1066 if (advise == MADV_FREE && tobject->type == OBJT_SWAP)
1067 swap_pager_freespace(tobject, tpindex, 1);
1068 /*
1069 * next object
1070 */
1071 backing_object = tobject->backing_object;
1072 if (backing_object == NULL)
1073 goto unlock_tobject;
1074 VM_OBJECT_LOCK(backing_object);
1075 tpindex += OFF_TO_IDX(tobject->backing_object_offset);
1076 if (tobject != object)
1077 VM_OBJECT_UNLOCK(tobject);
1078 tobject = backing_object;
1079 goto shadowlookup;
1080 } else if (m->valid != VM_PAGE_BITS_ALL)
1081 goto unlock_tobject;
1082 /*
1083 * If the page is not in a normal state, skip it.
1084 */
1085 vm_page_lock(m);
1086 if (m->hold_count != 0 || m->wire_count != 0) {
1087 vm_page_unlock(m);
1088 goto unlock_tobject;
1089 }
1090 KASSERT((m->flags & PG_FICTITIOUS) == 0,
1091 ("vm_object_madvise: page %p is fictitious", m));
1092 KASSERT((m->oflags & VPO_UNMANAGED) == 0,
1093 ("vm_object_madvise: page %p is not managed", m));
1094 if ((m->oflags & VPO_BUSY) || m->busy) {
1095 if (advise == MADV_WILLNEED) {
1096 /*
1097 * Reference the page before unlocking and
1098 * sleeping so that the page daemon is less
1099 * likely to reclaim it.
1100 */
1101 vm_page_aflag_set(m, PGA_REFERENCED);
1102 }
1103 vm_page_unlock(m);
1104 if (object != tobject)
1105 VM_OBJECT_UNLOCK(object);
1106 m->oflags |= VPO_WANTED;
1107 msleep(m, VM_OBJECT_MTX(tobject), PDROP | PVM, "madvpo",
1108 0);
1109 VM_OBJECT_LOCK(object);
1110 goto relookup;
1111 }
1112 if (advise == MADV_WILLNEED) {
1113 vm_page_activate(m);
1114 } else if (advise == MADV_DONTNEED) {
1115 vm_page_dontneed(m);
1116 } else if (advise == MADV_FREE) {
1117 /*
1118 * Mark the page clean. This will allow the page
1119 * to be freed up by the system. However, such pages
1120 * are often reused quickly by malloc()/free()
1121 * so we do not do anything that would cause
1122 * a page fault if we can help it.
1123 *
1124 * Specifically, we do not try to actually free
1125 * the page now nor do we try to put it in the
1126 * cache (which would cause a page fault on reuse).
1127 *
1128 * But we do make the page is freeable as we
1129 * can without actually taking the step of unmapping
1130 * it.
1131 */
1132 pmap_clear_modify(m);
1133 m->dirty = 0;
1134 m->act_count = 0;
1135 vm_page_dontneed(m);
1136 }
1137 vm_page_unlock(m);
1138 if (advise == MADV_FREE && tobject->type == OBJT_SWAP)
1139 swap_pager_freespace(tobject, tpindex, 1);
1140 unlock_tobject:
1141 if (tobject != object)
1142 VM_OBJECT_UNLOCK(tobject);
1143 }
1144 VM_OBJECT_UNLOCK(object);
1145 }
1146
1147 /*
1148 * vm_object_shadow:
1149 *
1150 * Create a new object which is backed by the
1151 * specified existing object range. The source
1152 * object reference is deallocated.
1153 *
1154 * The new object and offset into that object
1155 * are returned in the source parameters.
1156 */
1157 void
1158 vm_object_shadow(
1159 vm_object_t *object, /* IN/OUT */
1160 vm_ooffset_t *offset, /* IN/OUT */
1161 vm_size_t length)
1162 {
1163 vm_object_t source;
1164 vm_object_t result;
1165
1166 source = *object;
1167
1168 /*
1169 * Don't create the new object if the old object isn't shared.
1170 */
1171 if (source != NULL) {
1172 VM_OBJECT_LOCK(source);
1173 if (source->ref_count == 1 &&
1174 source->handle == NULL &&
1175 (source->type == OBJT_DEFAULT ||
1176 source->type == OBJT_SWAP)) {
1177 VM_OBJECT_UNLOCK(source);
1178 return;
1179 }
1180 VM_OBJECT_UNLOCK(source);
1181 }
1182
1183 /*
1184 * Allocate a new object with the given length.
1185 */
1186 result = vm_object_allocate(OBJT_DEFAULT, atop(length));
1187
1188 /*
1189 * The new object shadows the source object, adding a reference to it.
1190 * Our caller changes his reference to point to the new object,
1191 * removing a reference to the source object. Net result: no change
1192 * of reference count.
1193 *
1194 * Try to optimize the result object's page color when shadowing
1195 * in order to maintain page coloring consistency in the combined
1196 * shadowed object.
1197 */
1198 result->backing_object = source;
1199 /*
1200 * Store the offset into the source object, and fix up the offset into
1201 * the new object.
1202 */
1203 result->backing_object_offset = *offset;
1204 if (source != NULL) {
1205 VM_OBJECT_LOCK(source);
1206 LIST_INSERT_HEAD(&source->shadow_head, result, shadow_list);
1207 source->shadow_count++;
1208 #if VM_NRESERVLEVEL > 0
1209 result->flags |= source->flags & OBJ_COLORED;
1210 result->pg_color = (source->pg_color + OFF_TO_IDX(*offset)) &
1211 ((1 << (VM_NFREEORDER - 1)) - 1);
1212 #endif
1213 VM_OBJECT_UNLOCK(source);
1214 }
1215
1216
1217 /*
1218 * Return the new things
1219 */
1220 *offset = 0;
1221 *object = result;
1222 }
1223
1224 /*
1225 * vm_object_split:
1226 *
1227 * Split the pages in a map entry into a new object. This affords
1228 * easier removal of unused pages, and keeps object inheritance from
1229 * being a negative impact on memory usage.
1230 */
1231 void
1232 vm_object_split(vm_map_entry_t entry)
1233 {
1234 vm_page_t m, m_next;
1235 vm_object_t orig_object, new_object, source;
1236 vm_pindex_t idx, offidxstart;
1237 vm_size_t size;
1238
1239 orig_object = entry->object.vm_object;
1240 if (orig_object->type != OBJT_DEFAULT && orig_object->type != OBJT_SWAP)
1241 return;
1242 if (orig_object->ref_count <= 1)
1243 return;
1244 VM_OBJECT_UNLOCK(orig_object);
1245
1246 offidxstart = OFF_TO_IDX(entry->offset);
1247 size = atop(entry->end - entry->start);
1248
1249 /*
1250 * If swap_pager_copy() is later called, it will convert new_object
1251 * into a swap object.
1252 */
1253 new_object = vm_object_allocate(OBJT_DEFAULT, size);
1254
1255 /*
1256 * At this point, the new object is still private, so the order in
1257 * which the original and new objects are locked does not matter.
1258 */
1259 VM_OBJECT_LOCK(new_object);
1260 VM_OBJECT_LOCK(orig_object);
1261 source = orig_object->backing_object;
1262 if (source != NULL) {
1263 VM_OBJECT_LOCK(source);
1264 if ((source->flags & OBJ_DEAD) != 0) {
1265 VM_OBJECT_UNLOCK(source);
1266 VM_OBJECT_UNLOCK(orig_object);
1267 VM_OBJECT_UNLOCK(new_object);
1268 vm_object_deallocate(new_object);
1269 VM_OBJECT_LOCK(orig_object);
1270 return;
1271 }
1272 LIST_INSERT_HEAD(&source->shadow_head,
1273 new_object, shadow_list);
1274 source->shadow_count++;
1275 vm_object_reference_locked(source); /* for new_object */
1276 vm_object_clear_flag(source, OBJ_ONEMAPPING);
1277 VM_OBJECT_UNLOCK(source);
1278 new_object->backing_object_offset =
1279 orig_object->backing_object_offset + entry->offset;
1280 new_object->backing_object = source;
1281 }
1282 if (orig_object->cred != NULL) {
1283 new_object->cred = orig_object->cred;
1284 crhold(orig_object->cred);
1285 new_object->charge = ptoa(size);
1286 KASSERT(orig_object->charge >= ptoa(size),
1287 ("orig_object->charge < 0"));
1288 orig_object->charge -= ptoa(size);
1289 }
1290 retry:
1291 m = vm_page_find_least(orig_object, offidxstart);
1292 for (; m != NULL && (idx = m->pindex - offidxstart) < size;
1293 m = m_next) {
1294 m_next = TAILQ_NEXT(m, listq);
1295
1296 /*
1297 * We must wait for pending I/O to complete before we can
1298 * rename the page.
1299 *
1300 * We do not have to VM_PROT_NONE the page as mappings should
1301 * not be changed by this operation.
1302 */
1303 if ((m->oflags & VPO_BUSY) || m->busy) {
1304 VM_OBJECT_UNLOCK(new_object);
1305 m->oflags |= VPO_WANTED;
1306 msleep(m, VM_OBJECT_MTX(orig_object), PVM, "spltwt", 0);
1307 VM_OBJECT_LOCK(new_object);
1308 goto retry;
1309 }
1310 vm_page_lock(m);
1311 vm_page_rename(m, new_object, idx);
1312 vm_page_unlock(m);
1313 /* page automatically made dirty by rename and cache handled */
1314 vm_page_busy(m);
1315 }
1316 if (orig_object->type == OBJT_SWAP) {
1317 /*
1318 * swap_pager_copy() can sleep, in which case the orig_object's
1319 * and new_object's locks are released and reacquired.
1320 */
1321 swap_pager_copy(orig_object, new_object, offidxstart, 0);
1322
1323 /*
1324 * Transfer any cached pages from orig_object to new_object.
1325 */
1326 if (__predict_false(orig_object->cache != NULL))
1327 vm_page_cache_transfer(orig_object, offidxstart,
1328 new_object);
1329 }
1330 VM_OBJECT_UNLOCK(orig_object);
1331 TAILQ_FOREACH(m, &new_object->memq, listq)
1332 vm_page_wakeup(m);
1333 VM_OBJECT_UNLOCK(new_object);
1334 entry->object.vm_object = new_object;
1335 entry->offset = 0LL;
1336 vm_object_deallocate(orig_object);
1337 VM_OBJECT_LOCK(new_object);
1338 }
1339
1340 #define OBSC_TEST_ALL_SHADOWED 0x0001
1341 #define OBSC_COLLAPSE_NOWAIT 0x0002
1342 #define OBSC_COLLAPSE_WAIT 0x0004
1343
1344 static int
1345 vm_object_backing_scan(vm_object_t object, int op)
1346 {
1347 int r = 1;
1348 vm_page_t p;
1349 vm_object_t backing_object;
1350 vm_pindex_t backing_offset_index;
1351
1352 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
1353 VM_OBJECT_LOCK_ASSERT(object->backing_object, MA_OWNED);
1354
1355 backing_object = object->backing_object;
1356 backing_offset_index = OFF_TO_IDX(object->backing_object_offset);
1357
1358 /*
1359 * Initial conditions
1360 */
1361 if (op & OBSC_TEST_ALL_SHADOWED) {
1362 /*
1363 * We do not want to have to test for the existence of cache
1364 * or swap pages in the backing object. XXX but with the
1365 * new swapper this would be pretty easy to do.
1366 *
1367 * XXX what about anonymous MAP_SHARED memory that hasn't
1368 * been ZFOD faulted yet? If we do not test for this, the
1369 * shadow test may succeed! XXX
1370 */
1371 if (backing_object->type != OBJT_DEFAULT) {
1372 return (0);
1373 }
1374 }
1375 if (op & OBSC_COLLAPSE_WAIT) {
1376 vm_object_set_flag(backing_object, OBJ_DEAD);
1377 }
1378
1379 /*
1380 * Our scan
1381 */
1382 p = TAILQ_FIRST(&backing_object->memq);
1383 while (p) {
1384 vm_page_t next = TAILQ_NEXT(p, listq);
1385 vm_pindex_t new_pindex = p->pindex - backing_offset_index;
1386
1387 if (op & OBSC_TEST_ALL_SHADOWED) {
1388 vm_page_t pp;
1389
1390 /*
1391 * Ignore pages outside the parent object's range
1392 * and outside the parent object's mapping of the
1393 * backing object.
1394 *
1395 * note that we do not busy the backing object's
1396 * page.
1397 */
1398 if (
1399 p->pindex < backing_offset_index ||
1400 new_pindex >= object->size
1401 ) {
1402 p = next;
1403 continue;
1404 }
1405
1406 /*
1407 * See if the parent has the page or if the parent's
1408 * object pager has the page. If the parent has the
1409 * page but the page is not valid, the parent's
1410 * object pager must have the page.
1411 *
1412 * If this fails, the parent does not completely shadow
1413 * the object and we might as well give up now.
1414 */
1415
1416 pp = vm_page_lookup(object, new_pindex);
1417 if (
1418 (pp == NULL || pp->valid == 0) &&
1419 !vm_pager_has_page(object, new_pindex, NULL, NULL)
1420 ) {
1421 r = 0;
1422 break;
1423 }
1424 }
1425
1426 /*
1427 * Check for busy page
1428 */
1429 if (op & (OBSC_COLLAPSE_WAIT | OBSC_COLLAPSE_NOWAIT)) {
1430 vm_page_t pp;
1431
1432 if (op & OBSC_COLLAPSE_NOWAIT) {
1433 if ((p->oflags & VPO_BUSY) ||
1434 !p->valid ||
1435 p->busy) {
1436 p = next;
1437 continue;
1438 }
1439 } else if (op & OBSC_COLLAPSE_WAIT) {
1440 if ((p->oflags & VPO_BUSY) || p->busy) {
1441 VM_OBJECT_UNLOCK(object);
1442 p->oflags |= VPO_WANTED;
1443 msleep(p, VM_OBJECT_MTX(backing_object),
1444 PDROP | PVM, "vmocol", 0);
1445 VM_OBJECT_LOCK(object);
1446 VM_OBJECT_LOCK(backing_object);
1447 /*
1448 * If we slept, anything could have
1449 * happened. Since the object is
1450 * marked dead, the backing offset
1451 * should not have changed so we
1452 * just restart our scan.
1453 */
1454 p = TAILQ_FIRST(&backing_object->memq);
1455 continue;
1456 }
1457 }
1458
1459 KASSERT(
1460 p->object == backing_object,
1461 ("vm_object_backing_scan: object mismatch")
1462 );
1463
1464 /*
1465 * Destroy any associated swap
1466 */
1467 if (backing_object->type == OBJT_SWAP) {
1468 swap_pager_freespace(
1469 backing_object,
1470 p->pindex,
1471 1
1472 );
1473 }
1474
1475 if (
1476 p->pindex < backing_offset_index ||
1477 new_pindex >= object->size
1478 ) {
1479 /*
1480 * Page is out of the parent object's range, we
1481 * can simply destroy it.
1482 */
1483 vm_page_lock(p);
1484 KASSERT(!pmap_page_is_mapped(p),
1485 ("freeing mapped page %p", p));
1486 if (p->wire_count == 0)
1487 vm_page_free(p);
1488 else
1489 vm_page_remove(p);
1490 vm_page_unlock(p);
1491 p = next;
1492 continue;
1493 }
1494
1495 pp = vm_page_lookup(object, new_pindex);
1496 if (
1497 (op & OBSC_COLLAPSE_NOWAIT) != 0 &&
1498 (pp != NULL && pp->valid == 0)
1499 ) {
1500 /*
1501 * The page in the parent is not (yet) valid.
1502 * We don't know anything about the state of
1503 * the original page. It might be mapped,
1504 * so we must avoid the next if here.
1505 *
1506 * This is due to a race in vm_fault() where
1507 * we must unbusy the original (backing_obj)
1508 * page before we can (re)lock the parent.
1509 * Hence we can get here.
1510 */
1511 p = next;
1512 continue;
1513 }
1514 if (
1515 pp != NULL ||
1516 vm_pager_has_page(object, new_pindex, NULL, NULL)
1517 ) {
1518 /*
1519 * page already exists in parent OR swap exists
1520 * for this location in the parent. Destroy
1521 * the original page from the backing object.
1522 *
1523 * Leave the parent's page alone
1524 */
1525 vm_page_lock(p);
1526 KASSERT(!pmap_page_is_mapped(p),
1527 ("freeing mapped page %p", p));
1528 if (p->wire_count == 0)
1529 vm_page_free(p);
1530 else
1531 vm_page_remove(p);
1532 vm_page_unlock(p);
1533 p = next;
1534 continue;
1535 }
1536
1537 #if VM_NRESERVLEVEL > 0
1538 /*
1539 * Rename the reservation.
1540 */
1541 vm_reserv_rename(p, object, backing_object,
1542 backing_offset_index);
1543 #endif
1544
1545 /*
1546 * Page does not exist in parent, rename the
1547 * page from the backing object to the main object.
1548 *
1549 * If the page was mapped to a process, it can remain
1550 * mapped through the rename.
1551 */
1552 vm_page_lock(p);
1553 vm_page_rename(p, object, new_pindex);
1554 vm_page_unlock(p);
1555 /* page automatically made dirty by rename */
1556 }
1557 p = next;
1558 }
1559 return (r);
1560 }
1561
1562
1563 /*
1564 * this version of collapse allows the operation to occur earlier and
1565 * when paging_in_progress is true for an object... This is not a complete
1566 * operation, but should plug 99.9% of the rest of the leaks.
1567 */
1568 static void
1569 vm_object_qcollapse(vm_object_t object)
1570 {
1571 vm_object_t backing_object = object->backing_object;
1572
1573 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
1574 VM_OBJECT_LOCK_ASSERT(backing_object, MA_OWNED);
1575
1576 if (backing_object->ref_count != 1)
1577 return;
1578
1579 vm_object_backing_scan(object, OBSC_COLLAPSE_NOWAIT);
1580 }
1581
1582 /*
1583 * vm_object_collapse:
1584 *
1585 * Collapse an object with the object backing it.
1586 * Pages in the backing object are moved into the
1587 * parent, and the backing object is deallocated.
1588 */
1589 void
1590 vm_object_collapse(vm_object_t object)
1591 {
1592 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
1593
1594 while (TRUE) {
1595 vm_object_t backing_object;
1596
1597 /*
1598 * Verify that the conditions are right for collapse:
1599 *
1600 * The object exists and the backing object exists.
1601 */
1602 if ((backing_object = object->backing_object) == NULL)
1603 break;
1604
1605 /*
1606 * we check the backing object first, because it is most likely
1607 * not collapsable.
1608 */
1609 VM_OBJECT_LOCK(backing_object);
1610 if (backing_object->handle != NULL ||
1611 (backing_object->type != OBJT_DEFAULT &&
1612 backing_object->type != OBJT_SWAP) ||
1613 (backing_object->flags & OBJ_DEAD) ||
1614 object->handle != NULL ||
1615 (object->type != OBJT_DEFAULT &&
1616 object->type != OBJT_SWAP) ||
1617 (object->flags & OBJ_DEAD)) {
1618 VM_OBJECT_UNLOCK(backing_object);
1619 break;
1620 }
1621
1622 if (
1623 object->paging_in_progress != 0 ||
1624 backing_object->paging_in_progress != 0
1625 ) {
1626 vm_object_qcollapse(object);
1627 VM_OBJECT_UNLOCK(backing_object);
1628 break;
1629 }
1630 /*
1631 * We know that we can either collapse the backing object (if
1632 * the parent is the only reference to it) or (perhaps) have
1633 * the parent bypass the object if the parent happens to shadow
1634 * all the resident pages in the entire backing object.
1635 *
1636 * This is ignoring pager-backed pages such as swap pages.
1637 * vm_object_backing_scan fails the shadowing test in this
1638 * case.
1639 */
1640 if (backing_object->ref_count == 1) {
1641 /*
1642 * If there is exactly one reference to the backing
1643 * object, we can collapse it into the parent.
1644 */
1645 vm_object_backing_scan(object, OBSC_COLLAPSE_WAIT);
1646
1647 #if VM_NRESERVLEVEL > 0
1648 /*
1649 * Break any reservations from backing_object.
1650 */
1651 if (__predict_false(!LIST_EMPTY(&backing_object->rvq)))
1652 vm_reserv_break_all(backing_object);
1653 #endif
1654
1655 /*
1656 * Move the pager from backing_object to object.
1657 */
1658 if (backing_object->type == OBJT_SWAP) {
1659 /*
1660 * swap_pager_copy() can sleep, in which case
1661 * the backing_object's and object's locks are
1662 * released and reacquired.
1663 */
1664 swap_pager_copy(
1665 backing_object,
1666 object,
1667 OFF_TO_IDX(object->backing_object_offset), TRUE);
1668
1669 /*
1670 * Free any cached pages from backing_object.
1671 */
1672 if (__predict_false(backing_object->cache != NULL))
1673 vm_page_cache_free(backing_object, 0, 0);
1674 }
1675 /*
1676 * Object now shadows whatever backing_object did.
1677 * Note that the reference to
1678 * backing_object->backing_object moves from within
1679 * backing_object to within object.
1680 */
1681 LIST_REMOVE(object, shadow_list);
1682 backing_object->shadow_count--;
1683 if (backing_object->backing_object) {
1684 VM_OBJECT_LOCK(backing_object->backing_object);
1685 LIST_REMOVE(backing_object, shadow_list);
1686 LIST_INSERT_HEAD(
1687 &backing_object->backing_object->shadow_head,
1688 object, shadow_list);
1689 /*
1690 * The shadow_count has not changed.
1691 */
1692 VM_OBJECT_UNLOCK(backing_object->backing_object);
1693 }
1694 object->backing_object = backing_object->backing_object;
1695 object->backing_object_offset +=
1696 backing_object->backing_object_offset;
1697
1698 /*
1699 * Discard backing_object.
1700 *
1701 * Since the backing object has no pages, no pager left,
1702 * and no object references within it, all that is
1703 * necessary is to dispose of it.
1704 */
1705 KASSERT(backing_object->ref_count == 1, (
1706 "backing_object %p was somehow re-referenced during collapse!",
1707 backing_object));
1708 VM_OBJECT_UNLOCK(backing_object);
1709 vm_object_destroy(backing_object);
1710
1711 object_collapses++;
1712 } else {
1713 vm_object_t new_backing_object;
1714
1715 /*
1716 * If we do not entirely shadow the backing object,
1717 * there is nothing we can do so we give up.
1718 */
1719 if (object->resident_page_count != object->size &&
1720 vm_object_backing_scan(object,
1721 OBSC_TEST_ALL_SHADOWED) == 0) {
1722 VM_OBJECT_UNLOCK(backing_object);
1723 break;
1724 }
1725
1726 /*
1727 * Make the parent shadow the next object in the
1728 * chain. Deallocating backing_object will not remove
1729 * it, since its reference count is at least 2.
1730 */
1731 LIST_REMOVE(object, shadow_list);
1732 backing_object->shadow_count--;
1733
1734 new_backing_object = backing_object->backing_object;
1735 if ((object->backing_object = new_backing_object) != NULL) {
1736 VM_OBJECT_LOCK(new_backing_object);
1737 LIST_INSERT_HEAD(
1738 &new_backing_object->shadow_head,
1739 object,
1740 shadow_list
1741 );
1742 new_backing_object->shadow_count++;
1743 vm_object_reference_locked(new_backing_object);
1744 VM_OBJECT_UNLOCK(new_backing_object);
1745 object->backing_object_offset +=
1746 backing_object->backing_object_offset;
1747 }
1748
1749 /*
1750 * Drop the reference count on backing_object. Since
1751 * its ref_count was at least 2, it will not vanish.
1752 */
1753 backing_object->ref_count--;
1754 VM_OBJECT_UNLOCK(backing_object);
1755 object_bypasses++;
1756 }
1757
1758 /*
1759 * Try again with this object's new backing object.
1760 */
1761 }
1762 }
1763
1764 /*
1765 * vm_object_page_remove:
1766 *
1767 * For the given object, either frees or invalidates each of the
1768 * specified pages. In general, a page is freed. However, if a page is
1769 * wired for any reason other than the existence of a managed, wired
1770 * mapping, then it may be invalidated but not removed from the object.
1771 * Pages are specified by the given range ["start", "end") and the option
1772 * OBJPR_CLEANONLY. As a special case, if "end" is zero, then the range
1773 * extends from "start" to the end of the object. If the option
1774 * OBJPR_CLEANONLY is specified, then only the non-dirty pages within the
1775 * specified range are affected. If the option OBJPR_NOTMAPPED is
1776 * specified, then the pages within the specified range must have no
1777 * mappings. Otherwise, if this option is not specified, any mappings to
1778 * the specified pages are removed before the pages are freed or
1779 * invalidated.
1780 *
1781 * In general, this operation should only be performed on objects that
1782 * contain managed pages. There are, however, two exceptions. First, it
1783 * is performed on the kernel and kmem objects by vm_map_entry_delete().
1784 * Second, it is used by msync(..., MS_INVALIDATE) to invalidate device-
1785 * backed pages. In both of these cases, the option OBJPR_CLEANONLY must
1786 * not be specified and the option OBJPR_NOTMAPPED must be specified.
1787 *
1788 * The object must be locked.
1789 */
1790 void
1791 vm_object_page_remove(vm_object_t object, vm_pindex_t start, vm_pindex_t end,
1792 int options)
1793 {
1794 vm_page_t p, next;
1795 int wirings;
1796
1797 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
1798 KASSERT((object->type != OBJT_DEVICE && object->type != OBJT_PHYS) ||
1799 (options & (OBJPR_CLEANONLY | OBJPR_NOTMAPPED)) == OBJPR_NOTMAPPED,
1800 ("vm_object_page_remove: illegal options for object %p", object));
1801 if (object->resident_page_count == 0)
1802 goto skipmemq;
1803 vm_object_pip_add(object, 1);
1804 again:
1805 p = vm_page_find_least(object, start);
1806
1807 /*
1808 * Here, the variable "p" is either (1) the page with the least pindex
1809 * greater than or equal to the parameter "start" or (2) NULL.
1810 */
1811 for (; p != NULL && (p->pindex < end || end == 0); p = next) {
1812 next = TAILQ_NEXT(p, listq);
1813
1814 /*
1815 * If the page is wired for any reason besides the existence
1816 * of managed, wired mappings, then it cannot be freed. For
1817 * example, fictitious pages, which represent device memory,
1818 * are inherently wired and cannot be freed. They can,
1819 * however, be invalidated if the option OBJPR_CLEANONLY is
1820 * not specified.
1821 */
1822 vm_page_lock(p);
1823 if ((wirings = p->wire_count) != 0 &&
1824 (wirings = pmap_page_wired_mappings(p)) != p->wire_count) {
1825 if ((options & OBJPR_NOTMAPPED) == 0) {
1826 pmap_remove_all(p);
1827 /* Account for removal of wired mappings. */
1828 if (wirings != 0)
1829 p->wire_count -= wirings;
1830 }
1831 if ((options & OBJPR_CLEANONLY) == 0) {
1832 p->valid = 0;
1833 vm_page_undirty(p);
1834 }
1835 vm_page_unlock(p);
1836 continue;
1837 }
1838 if (vm_page_sleep_if_busy(p, TRUE, "vmopar"))
1839 goto again;
1840 KASSERT((p->flags & PG_FICTITIOUS) == 0,
1841 ("vm_object_page_remove: page %p is fictitious", p));
1842 if ((options & OBJPR_CLEANONLY) != 0 && p->valid != 0) {
1843 if ((options & OBJPR_NOTMAPPED) == 0)
1844 pmap_remove_write(p);
1845 if (p->dirty) {
1846 vm_page_unlock(p);
1847 continue;
1848 }
1849 }
1850 if ((options & OBJPR_NOTMAPPED) == 0) {
1851 pmap_remove_all(p);
1852 /* Account for removal of wired mappings. */
1853 if (wirings != 0)
1854 p->wire_count -= wirings;
1855 }
1856 vm_page_free(p);
1857 vm_page_unlock(p);
1858 }
1859 vm_object_pip_wakeup(object);
1860 skipmemq:
1861 if (__predict_false(object->cache != NULL))
1862 vm_page_cache_free(object, start, end);
1863 }
1864
1865 /*
1866 * Populate the specified range of the object with valid pages. Returns
1867 * TRUE if the range is successfully populated and FALSE otherwise.
1868 *
1869 * Note: This function should be optimized to pass a larger array of
1870 * pages to vm_pager_get_pages() before it is applied to a non-
1871 * OBJT_DEVICE object.
1872 *
1873 * The object must be locked.
1874 */
1875 boolean_t
1876 vm_object_populate(vm_object_t object, vm_pindex_t start, vm_pindex_t end)
1877 {
1878 vm_page_t m, ma[1];
1879 vm_pindex_t pindex;
1880 int rv;
1881
1882 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
1883 for (pindex = start; pindex < end; pindex++) {
1884 m = vm_page_grab(object, pindex, VM_ALLOC_NORMAL |
1885 VM_ALLOC_RETRY);
1886 if (m->valid != VM_PAGE_BITS_ALL) {
1887 ma[0] = m;
1888 rv = vm_pager_get_pages(object, ma, 1, 0);
1889 m = vm_page_lookup(object, pindex);
1890 if (m == NULL)
1891 break;
1892 if (rv != VM_PAGER_OK) {
1893 vm_page_lock(m);
1894 vm_page_free(m);
1895 vm_page_unlock(m);
1896 break;
1897 }
1898 }
1899 /*
1900 * Keep "m" busy because a subsequent iteration may unlock
1901 * the object.
1902 */
1903 }
1904 if (pindex > start) {
1905 m = vm_page_lookup(object, start);
1906 while (m != NULL && m->pindex < pindex) {
1907 vm_page_wakeup(m);
1908 m = TAILQ_NEXT(m, listq);
1909 }
1910 }
1911 return (pindex == end);
1912 }
1913
1914 /*
1915 * Routine: vm_object_coalesce
1916 * Function: Coalesces two objects backing up adjoining
1917 * regions of memory into a single object.
1918 *
1919 * returns TRUE if objects were combined.
1920 *
1921 * NOTE: Only works at the moment if the second object is NULL -
1922 * if it's not, which object do we lock first?
1923 *
1924 * Parameters:
1925 * prev_object First object to coalesce
1926 * prev_offset Offset into prev_object
1927 * prev_size Size of reference to prev_object
1928 * next_size Size of reference to the second object
1929 * reserved Indicator that extension region has
1930 * swap accounted for
1931 *
1932 * Conditions:
1933 * The object must *not* be locked.
1934 */
1935 boolean_t
1936 vm_object_coalesce(vm_object_t prev_object, vm_ooffset_t prev_offset,
1937 vm_size_t prev_size, vm_size_t next_size, boolean_t reserved)
1938 {
1939 vm_pindex_t next_pindex;
1940
1941 if (prev_object == NULL)
1942 return (TRUE);
1943 VM_OBJECT_LOCK(prev_object);
1944 if (prev_object->type != OBJT_DEFAULT &&
1945 prev_object->type != OBJT_SWAP) {
1946 VM_OBJECT_UNLOCK(prev_object);
1947 return (FALSE);
1948 }
1949
1950 /*
1951 * Try to collapse the object first
1952 */
1953 vm_object_collapse(prev_object);
1954
1955 /*
1956 * Can't coalesce if: . more than one reference . paged out . shadows
1957 * another object . has a copy elsewhere (any of which mean that the
1958 * pages not mapped to prev_entry may be in use anyway)
1959 */
1960 if (prev_object->backing_object != NULL) {
1961 VM_OBJECT_UNLOCK(prev_object);
1962 return (FALSE);
1963 }
1964
1965 prev_size >>= PAGE_SHIFT;
1966 next_size >>= PAGE_SHIFT;
1967 next_pindex = OFF_TO_IDX(prev_offset) + prev_size;
1968
1969 if ((prev_object->ref_count > 1) &&
1970 (prev_object->size != next_pindex)) {
1971 VM_OBJECT_UNLOCK(prev_object);
1972 return (FALSE);
1973 }
1974
1975 /*
1976 * Account for the charge.
1977 */
1978 if (prev_object->cred != NULL) {
1979
1980 /*
1981 * If prev_object was charged, then this mapping,
1982 * althought not charged now, may become writable
1983 * later. Non-NULL cred in the object would prevent
1984 * swap reservation during enabling of the write
1985 * access, so reserve swap now. Failed reservation
1986 * cause allocation of the separate object for the map
1987 * entry, and swap reservation for this entry is
1988 * managed in appropriate time.
1989 */
1990 if (!reserved && !swap_reserve_by_cred(ptoa(next_size),
1991 prev_object->cred)) {
1992 return (FALSE);
1993 }
1994 prev_object->charge += ptoa(next_size);
1995 }
1996
1997 /*
1998 * Remove any pages that may still be in the object from a previous
1999 * deallocation.
2000 */
2001 if (next_pindex < prev_object->size) {
2002 vm_object_page_remove(prev_object, next_pindex, next_pindex +
2003 next_size, 0);
2004 if (prev_object->type == OBJT_SWAP)
2005 swap_pager_freespace(prev_object,
2006 next_pindex, next_size);
2007 #if 0
2008 if (prev_object->cred != NULL) {
2009 KASSERT(prev_object->charge >=
2010 ptoa(prev_object->size - next_pindex),
2011 ("object %p overcharged 1 %jx %jx", prev_object,
2012 (uintmax_t)next_pindex, (uintmax_t)next_size));
2013 prev_object->charge -= ptoa(prev_object->size -
2014 next_pindex);
2015 }
2016 #endif
2017 }
2018
2019 /*
2020 * Extend the object if necessary.
2021 */
2022 if (next_pindex + next_size > prev_object->size)
2023 prev_object->size = next_pindex + next_size;
2024
2025 VM_OBJECT_UNLOCK(prev_object);
2026 return (TRUE);
2027 }
2028
2029 void
2030 vm_object_set_writeable_dirty(vm_object_t object)
2031 {
2032
2033 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
2034 if (object->type != OBJT_VNODE)
2035 return;
2036 object->generation++;
2037 if ((object->flags & OBJ_MIGHTBEDIRTY) != 0)
2038 return;
2039 vm_object_set_flag(object, OBJ_MIGHTBEDIRTY);
2040 }
2041
2042 #include "opt_ddb.h"
2043 #ifdef DDB
2044 #include <sys/kernel.h>
2045
2046 #include <sys/cons.h>
2047
2048 #include <ddb/ddb.h>
2049
2050 static int
2051 _vm_object_in_map(vm_map_t map, vm_object_t object, vm_map_entry_t entry)
2052 {
2053 vm_map_t tmpm;
2054 vm_map_entry_t tmpe;
2055 vm_object_t obj;
2056 int entcount;
2057
2058 if (map == 0)
2059 return 0;
2060
2061 if (entry == 0) {
2062 tmpe = map->header.next;
2063 entcount = map->nentries;
2064 while (entcount-- && (tmpe != &map->header)) {
2065 if (_vm_object_in_map(map, object, tmpe)) {
2066 return 1;
2067 }
2068 tmpe = tmpe->next;
2069 }
2070 } else if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
2071 tmpm = entry->object.sub_map;
2072 tmpe = tmpm->header.next;
2073 entcount = tmpm->nentries;
2074 while (entcount-- && tmpe != &tmpm->header) {
2075 if (_vm_object_in_map(tmpm, object, tmpe)) {
2076 return 1;
2077 }
2078 tmpe = tmpe->next;
2079 }
2080 } else if ((obj = entry->object.vm_object) != NULL) {
2081 for (; obj; obj = obj->backing_object)
2082 if (obj == object) {
2083 return 1;
2084 }
2085 }
2086 return 0;
2087 }
2088
2089 static int
2090 vm_object_in_map(vm_object_t object)
2091 {
2092 struct proc *p;
2093
2094 /* sx_slock(&allproc_lock); */
2095 FOREACH_PROC_IN_SYSTEM(p) {
2096 if (!p->p_vmspace /* || (p->p_flag & (P_SYSTEM|P_WEXIT)) */)
2097 continue;
2098 if (_vm_object_in_map(&p->p_vmspace->vm_map, object, 0)) {
2099 /* sx_sunlock(&allproc_lock); */
2100 return 1;
2101 }
2102 }
2103 /* sx_sunlock(&allproc_lock); */
2104 if (_vm_object_in_map(kernel_map, object, 0))
2105 return 1;
2106 if (_vm_object_in_map(kmem_map, object, 0))
2107 return 1;
2108 if (_vm_object_in_map(pager_map, object, 0))
2109 return 1;
2110 if (_vm_object_in_map(buffer_map, object, 0))
2111 return 1;
2112 return 0;
2113 }
2114
2115 DB_SHOW_COMMAND(vmochk, vm_object_check)
2116 {
2117 vm_object_t object;
2118
2119 /*
2120 * make sure that internal objs are in a map somewhere
2121 * and none have zero ref counts.
2122 */
2123 TAILQ_FOREACH(object, &vm_object_list, object_list) {
2124 if (object->handle == NULL &&
2125 (object->type == OBJT_DEFAULT || object->type == OBJT_SWAP)) {
2126 if (object->ref_count == 0) {
2127 db_printf("vmochk: internal obj has zero ref count: %ld\n",
2128 (long)object->size);
2129 }
2130 if (!vm_object_in_map(object)) {
2131 db_printf(
2132 "vmochk: internal obj is not in a map: "
2133 "ref: %d, size: %lu: 0x%lx, backing_object: %p\n",
2134 object->ref_count, (u_long)object->size,
2135 (u_long)object->size,
2136 (void *)object->backing_object);
2137 }
2138 }
2139 }
2140 }
2141
2142 /*
2143 * vm_object_print: [ debug ]
2144 */
2145 DB_SHOW_COMMAND(object, vm_object_print_static)
2146 {
2147 /* XXX convert args. */
2148 vm_object_t object = (vm_object_t)addr;
2149 boolean_t full = have_addr;
2150
2151 vm_page_t p;
2152
2153 /* XXX count is an (unused) arg. Avoid shadowing it. */
2154 #define count was_count
2155
2156 int count;
2157
2158 if (object == NULL)
2159 return;
2160
2161 db_iprintf(
2162 "Object %p: type=%d, size=0x%jx, res=%d, ref=%d, flags=0x%x ruid %d charge %jx\n",
2163 object, (int)object->type, (uintmax_t)object->size,
2164 object->resident_page_count, object->ref_count, object->flags,
2165 object->cred ? object->cred->cr_ruid : -1, (uintmax_t)object->charge);
2166 db_iprintf(" sref=%d, backing_object(%d)=(%p)+0x%jx\n",
2167 object->shadow_count,
2168 object->backing_object ? object->backing_object->ref_count : 0,
2169 object->backing_object, (uintmax_t)object->backing_object_offset);
2170
2171 if (!full)
2172 return;
2173
2174 db_indent += 2;
2175 count = 0;
2176 TAILQ_FOREACH(p, &object->memq, listq) {
2177 if (count == 0)
2178 db_iprintf("memory:=");
2179 else if (count == 6) {
2180 db_printf("\n");
2181 db_iprintf(" ...");
2182 count = 0;
2183 } else
2184 db_printf(",");
2185 count++;
2186
2187 db_printf("(off=0x%jx,page=0x%jx)",
2188 (uintmax_t)p->pindex, (uintmax_t)VM_PAGE_TO_PHYS(p));
2189 }
2190 if (count != 0)
2191 db_printf("\n");
2192 db_indent -= 2;
2193 }
2194
2195 /* XXX. */
2196 #undef count
2197
2198 /* XXX need this non-static entry for calling from vm_map_print. */
2199 void
2200 vm_object_print(
2201 /* db_expr_t */ long addr,
2202 boolean_t have_addr,
2203 /* db_expr_t */ long count,
2204 char *modif)
2205 {
2206 vm_object_print_static(addr, have_addr, count, modif);
2207 }
2208
2209 DB_SHOW_COMMAND(vmopag, vm_object_print_pages)
2210 {
2211 vm_object_t object;
2212 vm_pindex_t fidx;
2213 vm_paddr_t pa;
2214 vm_page_t m, prev_m;
2215 int rcount, nl, c;
2216
2217 nl = 0;
2218 TAILQ_FOREACH(object, &vm_object_list, object_list) {
2219 db_printf("new object: %p\n", (void *)object);
2220 if (nl > 18) {
2221 c = cngetc();
2222 if (c != ' ')
2223 return;
2224 nl = 0;
2225 }
2226 nl++;
2227 rcount = 0;
2228 fidx = 0;
2229 pa = -1;
2230 TAILQ_FOREACH(m, &object->memq, listq) {
2231 if (m->pindex > 128)
2232 break;
2233 if ((prev_m = TAILQ_PREV(m, pglist, listq)) != NULL &&
2234 prev_m->pindex + 1 != m->pindex) {
2235 if (rcount) {
2236 db_printf(" index(%ld)run(%d)pa(0x%lx)\n",
2237 (long)fidx, rcount, (long)pa);
2238 if (nl > 18) {
2239 c = cngetc();
2240 if (c != ' ')
2241 return;
2242 nl = 0;
2243 }
2244 nl++;
2245 rcount = 0;
2246 }
2247 }
2248 if (rcount &&
2249 (VM_PAGE_TO_PHYS(m) == pa + rcount * PAGE_SIZE)) {
2250 ++rcount;
2251 continue;
2252 }
2253 if (rcount) {
2254 db_printf(" index(%ld)run(%d)pa(0x%lx)\n",
2255 (long)fidx, rcount, (long)pa);
2256 if (nl > 18) {
2257 c = cngetc();
2258 if (c != ' ')
2259 return;
2260 nl = 0;
2261 }
2262 nl++;
2263 }
2264 fidx = m->pindex;
2265 pa = VM_PAGE_TO_PHYS(m);
2266 rcount = 1;
2267 }
2268 if (rcount) {
2269 db_printf(" index(%ld)run(%d)pa(0x%lx)\n",
2270 (long)fidx, rcount, (long)pa);
2271 if (nl > 18) {
2272 c = cngetc();
2273 if (c != ' ')
2274 return;
2275 nl = 0;
2276 }
2277 nl++;
2278 }
2279 }
2280 }
2281 #endif /* DDB */
Cache object: 80971f6276e47f54e0728e0f33bd59a7
|