kernel - Fix incorrect dirty/reprotect code in pageout
[dragonfly.git] / sys / vm / vm_object.c
CommitLineData
984263bc 1/*
6846fd23
MD
2 * (MPSAFE)
3 *
984263bc
MD
4 * Copyright (c) 1991, 1993
5 * The Regents of the University of California. All rights reserved.
6 *
7 * This code is derived from software contributed to Berkeley by
8 * The Mach Operating System project at Carnegie-Mellon University.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 3. All advertising materials mentioning features or use of this software
19 * must display the following acknowledgement:
20 * This product includes software developed by the University of
21 * California, Berkeley and its contributors.
22 * 4. Neither the name of the University nor the names of its contributors
23 * may be used to endorse or promote products derived from this software
24 * without specific prior written permission.
25 *
26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 * SUCH DAMAGE.
37 *
38 * from: @(#)vm_object.c 8.5 (Berkeley) 3/22/94
39 *
40 *
41 * Copyright (c) 1987, 1990 Carnegie-Mellon University.
42 * All rights reserved.
43 *
44 * Authors: Avadis Tevanian, Jr., Michael Wayne Young
45 *
46 * Permission to use, copy, modify and distribute this software and
47 * its documentation is hereby granted, provided that both the copyright
48 * notice and this permission notice appear in all copies of the
49 * software, derivative works or modified versions, and any portions
50 * thereof, and that both notices appear in supporting documentation.
51 *
52 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
53 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
54 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
55 *
56 * Carnegie Mellon requests users of this software to return to
57 *
58 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
59 * School of Computer Science
60 * Carnegie Mellon University
61 * Pittsburgh PA 15213-3890
62 *
63 * any improvements or extensions that they make and grant Carnegie the
64 * rights to redistribute these changes.
65 *
66 * $FreeBSD: src/sys/vm/vm_object.c,v 1.171.2.8 2003/05/26 19:17:56 alc Exp $
67 */
68
69/*
70 * Virtual memory object module.
71 */
72
73#include <sys/param.h>
74#include <sys/systm.h>
75#include <sys/proc.h> /* for curproc, pageproc */
e32ad78d 76#include <sys/thread.h>
984263bc
MD
77#include <sys/vnode.h>
78#include <sys/vmmeter.h>
79#include <sys/mman.h>
80#include <sys/mount.h>
81#include <sys/kernel.h>
82#include <sys/sysctl.h>
e1c14c82 83#include <sys/refcount.h>
984263bc
MD
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_zone.h>
97
98#define EASY_SCAN_FACTOR 8
99
b12defdc
MD
100static void vm_object_qcollapse(vm_object_t object,
101 vm_object_t backing_object);
ab3e1edd 102static void vm_object_page_collect_flush(vm_object_t object, vm_page_t p,
1f804340 103 int pagerflags);
e1c14c82 104static void vm_object_lock_init(vm_object_t);
e1c14c82 105
984263bc
MD
106
107/*
108 * Virtual memory objects maintain the actual data
109 * associated with allocated virtual memory. A given
110 * page of memory exists within exactly one object.
111 *
112 * An object is only deallocated when all "references"
113 * are given up. Only one "reference" to a given
114 * region of an object should be writeable.
115 *
116 * Associated with each object is a list of all resident
117 * memory pages belonging to that object; this list is
118 * maintained by the "vm_page" module, and locked by the object's
119 * lock.
120 *
121 * Each object also records a "pager" routine which is
122 * used to retrieve (and store) pages to the proper backing
123 * storage. In addition, objects may be backed by other
124 * objects from which they were virtual-copied.
125 *
126 * The only items within the object structure which are
127 * modified after time of creation are:
128 * reference count locked by object's lock
129 * pager routine locked by object's lock
130 *
131 */
132
2de4f77e 133struct object_q vm_object_list; /* locked by vmobj_token */
c439ad8f
MD
134struct vm_object kernel_object;
135
2de4f77e 136static long vm_object_count; /* locked by vmobj_token */
984263bc
MD
137extern int vm_pageout_page_count;
138
139static long object_collapses;
140static long object_bypasses;
141static int next_index;
142static vm_zone_t obj_zone;
143static struct vm_zone obj_zone_store;
984263bc
MD
144#define VM_OBJECTS_INIT 256
145static struct vm_object vm_objects_init[VM_OBJECTS_INIT];
146
6846fd23 147/*
212f39f5
MD
148 * Misc low level routines
149 */
150static void
151vm_object_lock_init(vm_object_t obj)
152{
153#if defined(DEBUG_LOCKS)
154 int i;
155
156 obj->debug_hold_bitmap = 0;
157 obj->debug_hold_ovfl = 0;
158 for (i = 0; i < VMOBJ_DEBUG_ARRAY_SIZE; i++) {
159 obj->debug_hold_thrs[i] = NULL;
18a4c8dc
VS
160 obj->debug_hold_file[i] = NULL;
161 obj->debug_hold_line[i] = 0;
212f39f5
MD
162 }
163#endif
164}
165
a31129d8 166void
212f39f5
MD
167vm_object_lock_swap(void)
168{
169 lwkt_token_swap();
170}
171
172void
173vm_object_lock(vm_object_t obj)
174{
c9958a5a 175 lwkt_gettoken(&obj->token);
212f39f5
MD
176}
177
609c9aae
MD
178/*
179 * Returns TRUE on sucesss
180 */
181static int
182vm_object_lock_try(vm_object_t obj)
183{
184 return(lwkt_trytoken(&obj->token));
185}
186
212f39f5 187void
54341a3b
MD
188vm_object_lock_shared(vm_object_t obj)
189{
c9958a5a 190 lwkt_gettoken_shared(&obj->token);
54341a3b
MD
191}
192
193void
212f39f5
MD
194vm_object_unlock(vm_object_t obj)
195{
c9958a5a 196 lwkt_reltoken(&obj->token);
212f39f5
MD
197}
198
199static __inline void
200vm_object_assert_held(vm_object_t obj)
201{
c9958a5a 202 ASSERT_LWKT_TOKEN_HELD(&obj->token);
212f39f5
MD
203}
204
205void
18a4c8dc 206#ifndef DEBUG_LOCKS
212f39f5 207vm_object_hold(vm_object_t obj)
18a4c8dc
VS
208#else
209debugvm_object_hold(vm_object_t obj, char *file, int line)
210#endif
212f39f5 211{
b12defdc 212 KKASSERT(obj != NULL);
212f39f5
MD
213
214 /*
215 * Object must be held (object allocation is stable due to callers
216 * context, typically already holding the token on a parent object)
217 * prior to potentially blocking on the lock, otherwise the object
218 * can get ripped away from us.
219 */
220 refcount_acquire(&obj->hold_count);
221 vm_object_lock(obj);
222
223#if defined(DEBUG_LOCKS)
224 int i;
5ef517ef 225 u_int mask;
212f39f5 226
5ef517ef
MD
227 for (;;) {
228 mask = ~obj->debug_hold_bitmap;
229 cpu_ccfence();
230 if (mask == 0xFFFFFFFFU) {
231 if (obj->debug_hold_ovfl == 0)
232 obj->debug_hold_ovfl = 1;
233 break;
234 }
235 i = ffs(mask) - 1;
236 if (atomic_cmpset_int(&obj->debug_hold_bitmap, ~mask,
237 ~mask | (1 << i))) {
238 obj->debug_hold_bitmap |= (1 << i);
239 obj->debug_hold_thrs[i] = curthread;
240 obj->debug_hold_file[i] = file;
241 obj->debug_hold_line[i] = line;
242 break;
243 }
212f39f5 244 }
212f39f5
MD
245#endif
246}
247
609c9aae
MD
248int
249#ifndef DEBUG_LOCKS
250vm_object_hold_try(vm_object_t obj)
251#else
252debugvm_object_hold_try(vm_object_t obj, char *file, int line)
253#endif
254{
255 KKASSERT(obj != NULL);
256
257 /*
258 * Object must be held (object allocation is stable due to callers
259 * context, typically already holding the token on a parent object)
260 * prior to potentially blocking on the lock, otherwise the object
261 * can get ripped away from us.
262 */
263 refcount_acquire(&obj->hold_count);
264 if (vm_object_lock_try(obj) == 0) {
265 if (refcount_release(&obj->hold_count)) {
266 if (obj->ref_count == 0 && (obj->flags & OBJ_DEAD))
267 zfree(obj_zone, obj);
268 }
269 return(0);
270 }
271
272#if defined(DEBUG_LOCKS)
273 int i;
274 u_int mask;
275
276 for (;;) {
277 mask = ~obj->debug_hold_bitmap;
278 cpu_ccfence();
279 if (mask == 0xFFFFFFFFU) {
280 if (obj->debug_hold_ovfl == 0)
281 obj->debug_hold_ovfl = 1;
282 break;
283 }
284 i = ffs(mask) - 1;
285 if (atomic_cmpset_int(&obj->debug_hold_bitmap, ~mask,
286 ~mask | (1 << i))) {
287 obj->debug_hold_bitmap |= (1 << i);
288 obj->debug_hold_thrs[i] = curthread;
289 obj->debug_hold_file[i] = file;
290 obj->debug_hold_line[i] = line;
291 break;
292 }
293 }
294#endif
295 return(1);
296}
297
54341a3b
MD
298void
299#ifndef DEBUG_LOCKS
300vm_object_hold_shared(vm_object_t obj)
301#else
302debugvm_object_hold_shared(vm_object_t obj, char *file, int line)
303#endif
304{
305 KKASSERT(obj != NULL);
306
307 /*
308 * Object must be held (object allocation is stable due to callers
309 * context, typically already holding the token on a parent object)
310 * prior to potentially blocking on the lock, otherwise the object
311 * can get ripped away from us.
312 */
313 refcount_acquire(&obj->hold_count);
314 vm_object_lock_shared(obj);
315
316#if defined(DEBUG_LOCKS)
317 int i;
5ef517ef 318 u_int mask;
54341a3b 319
5ef517ef
MD
320 for (;;) {
321 mask = ~obj->debug_hold_bitmap;
322 cpu_ccfence();
323 if (mask == 0xFFFFFFFFU) {
324 if (obj->debug_hold_ovfl == 0)
325 obj->debug_hold_ovfl = 1;
326 break;
327 }
328 i = ffs(mask) - 1;
329 if (atomic_cmpset_int(&obj->debug_hold_bitmap, ~mask,
330 ~mask | (1 << i))) {
331 obj->debug_hold_bitmap |= (1 << i);
332 obj->debug_hold_thrs[i] = curthread;
333 obj->debug_hold_file[i] = file;
334 obj->debug_hold_line[i] = line;
335 break;
336 }
54341a3b 337 }
54341a3b
MD
338#endif
339}
340
b12defdc
MD
341/*
342 * Drop the token and hold_count on the object.
343 */
212f39f5
MD
344void
345vm_object_drop(vm_object_t obj)
346{
347 if (obj == NULL)
348 return;
349
350#if defined(DEBUG_LOCKS)
351 int found = 0;
352 int i;
353
354 for (i = 0; i < VMOBJ_DEBUG_ARRAY_SIZE; i++) {
355 if ((obj->debug_hold_bitmap & (1 << i)) &&
356 (obj->debug_hold_thrs[i] == curthread)) {
357 obj->debug_hold_bitmap &= ~(1 << i);
358 obj->debug_hold_thrs[i] = NULL;
18a4c8dc
VS
359 obj->debug_hold_file[i] = NULL;
360 obj->debug_hold_line[i] = 0;
212f39f5
MD
361 found = 1;
362 break;
363 }
364 }
365
366 if (found == 0 && obj->debug_hold_ovfl == 0)
367 panic("vm_object: attempt to drop hold on non-self-held obj");
368#endif
369
370 /*
c9958a5a
MD
371 * No new holders should be possible once we drop hold_count 1->0 as
372 * there is no longer any way to reference the object.
212f39f5 373 */
c17a6852 374 KKASSERT(obj->hold_count > 0);
b12defdc 375 if (refcount_release(&obj->hold_count)) {
c9958a5a
MD
376 if (obj->ref_count == 0 && (obj->flags & OBJ_DEAD)) {
377 vm_object_unlock(obj);
b12defdc 378 zfree(obj_zone, obj);
c9958a5a
MD
379 } else {
380 vm_object_unlock(obj);
381 }
382 } else {
383 vm_object_unlock(obj);
212f39f5 384 }
212f39f5
MD
385}
386
212f39f5 387/*
a2ee730d 388 * Initialize a freshly allocated object, returning a held object.
6846fd23
MD
389 *
390 * Used only by vm_object_allocate() and zinitna().
391 *
392 * No requirements.
393 */
984263bc 394void
aecf2182 395_vm_object_allocate(objtype_t type, vm_pindex_t size, vm_object_t object)
984263bc
MD
396{
397 int incr;
6846fd23 398
1f804340 399 RB_INIT(&object->rb_memq);
984263bc 400 LIST_INIT(&object->shadow_head);
c9958a5a 401 lwkt_token_init(&object->token, "vmobj");
984263bc
MD
402
403 object->type = type;
404 object->size = size;
405 object->ref_count = 1;
e1c14c82 406 object->hold_count = 0;
984263bc
MD
407 object->flags = 0;
408 if ((object->type == OBJT_DEFAULT) || (object->type == OBJT_SWAP))
409 vm_object_set_flag(object, OBJ_ONEMAPPING);
410 object->paging_in_progress = 0;
411 object->resident_page_count = 0;
50a55c46 412 object->agg_pv_list_count = 0;
984263bc 413 object->shadow_count = 0;
85946b6c
MD
414#ifdef SMP
415 /* cpu localization twist */
416 object->pg_color = (int)(intptr_t)curthread;
417#else
984263bc 418 object->pg_color = next_index;
85946b6c 419#endif
984263bc
MD
420 if ( size > (PQ_L2_SIZE / 3 + PQ_PRIME1))
421 incr = PQ_L2_SIZE / 3 + PQ_PRIME1;
422 else
423 incr = size;
424 next_index = (next_index + incr) & PQ_L2_MASK;
425 object->handle = NULL;
426 object->backing_object = NULL;
b12defdc 427 object->backing_object_offset = (vm_ooffset_t)0;
984263bc
MD
428
429 object->generation++;
96adc753
MD
430 object->swblock_count = 0;
431 RB_INIT(&object->swblock_root);
e1c14c82 432 vm_object_lock_init(object);
984263bc 433
a2ee730d 434 vm_object_hold(object);
2de4f77e 435 lwkt_gettoken(&vmobj_token);
984263bc
MD
436 TAILQ_INSERT_TAIL(&vm_object_list, object, object_list);
437 vm_object_count++;
2de4f77e 438 lwkt_reltoken(&vmobj_token);
984263bc
MD
439}
440
441/*
6846fd23 442 * Initialize the VM objects module.
984263bc 443 *
6846fd23 444 * Called from the low level boot code only.
984263bc
MD
445 */
446void
57e43348 447vm_object_init(void)
984263bc
MD
448{
449 TAILQ_INIT(&vm_object_list);
984263bc 450
e4846942 451 _vm_object_allocate(OBJT_DEFAULT, OFF_TO_IDX(KvaEnd),
c439ad8f 452 &kernel_object);
a2ee730d 453 vm_object_drop(&kernel_object);
984263bc
MD
454
455 obj_zone = &obj_zone_store;
456 zbootinit(obj_zone, "VM OBJECT", sizeof (struct vm_object),
457 vm_objects_init, VM_OBJECTS_INIT);
458}
459
460void
57e43348 461vm_object_init2(void)
c97ef899
MD
462{
463 zinitna(obj_zone, NULL, NULL, 0, 0, ZONE_PANICFAIL, 1);
984263bc
MD
464}
465
466/*
6846fd23 467 * Allocate and return a new object of the specified type and size.
984263bc 468 *
6846fd23 469 * No requirements.
984263bc 470 */
984263bc 471vm_object_t
aecf2182 472vm_object_allocate(objtype_t type, vm_pindex_t size)
984263bc
MD
473{
474 vm_object_t result;
475
476 result = (vm_object_t) zalloc(obj_zone);
477
478 _vm_object_allocate(type, size, result);
a2ee730d
MD
479 vm_object_drop(result);
480
481 return (result);
482}
483
484/*
485 * This version returns a held object, allowing further atomic initialization
486 * of the object.
487 */
488vm_object_t
489vm_object_allocate_hold(objtype_t type, vm_pindex_t size)
490{
491 vm_object_t result;
492
493 result = (vm_object_t) zalloc(obj_zone);
494
495 _vm_object_allocate(type, size, result);
984263bc
MD
496
497 return (result);
498}
499
984263bc 500/*
b12defdc
MD
501 * Add an additional reference to a vm_object. The object must already be
502 * held. The original non-lock version is no longer supported. The object
503 * must NOT be chain locked by anyone at the time the reference is added.
504 *
505 * Referencing a chain-locked object can blow up the fairly sensitive
506 * ref_count and shadow_count tests in the deallocator. Most callers
507 * will call vm_object_chain_wait() prior to calling
508 * vm_object_reference_locked() to avoid the case.
984263bc 509 *
b12defdc 510 * The object must be held.
984263bc
MD
511 */
512void
b12defdc 513vm_object_reference_locked(vm_object_t object)
984263bc 514{
b12defdc
MD
515 KKASSERT(object != NULL);
516 ASSERT_LWKT_TOKEN_HELD(vm_object_token(object));
517 KKASSERT((object->flags & OBJ_CHAINLOCK) == 0);
518 object->ref_count++;
519 if (object->type == OBJT_VNODE) {
520 vref(object->handle);
521 /* XXX what if the vnode is being destroyed? */
522 }
2de4f77e 523}
984263bc 524
b12defdc
MD
525/*
526 * Object OBJ_CHAINLOCK lock handling.
527 *
528 * The caller can chain-lock backing objects recursively and then
529 * use vm_object_chain_release_all() to undo the whole chain.
530 *
531 * Chain locks are used to prevent collapses and are only applicable
532 * to OBJT_DEFAULT and OBJT_SWAP objects. Chain locking operations
533 * on other object types are ignored. This is also important because
534 * it allows e.g. the vnode underlying a memory mapping to take concurrent
535 * faults.
536 *
537 * The object must usually be held on entry, though intermediate
538 * objects need not be held on release.
539 */
2de4f77e 540void
b12defdc 541vm_object_chain_wait(vm_object_t object)
2de4f77e 542{
b12defdc
MD
543 ASSERT_LWKT_TOKEN_HELD(vm_object_token(object));
544 while (object->flags & OBJ_CHAINLOCK) {
545 vm_object_set_flag(object, OBJ_CHAINWANT);
546 tsleep(object, 0, "objchain", 0);
547 }
548}
549
550void
551vm_object_chain_acquire(vm_object_t object)
552{
553 if (object->type == OBJT_DEFAULT || object->type == OBJT_SWAP) {
554 vm_object_chain_wait(object);
555 vm_object_set_flag(object, OBJ_CHAINLOCK);
556 }
557}
558
559void
560vm_object_chain_release(vm_object_t object)
561{
562 ASSERT_LWKT_TOKEN_HELD(vm_object_token(object));
563 if (object->type == OBJT_DEFAULT || object->type == OBJT_SWAP) {
564 KKASSERT(object->flags & OBJ_CHAINLOCK);
565 if (object->flags & OBJ_CHAINWANT) {
566 vm_object_clear_flag(object,
567 OBJ_CHAINLOCK | OBJ_CHAINWANT);
568 wakeup(object);
569 } else {
570 vm_object_clear_flag(object, OBJ_CHAINLOCK);
2de4f77e 571 }
984263bc
MD
572 }
573}
574
6846fd23 575/*
d2d8515b
MD
576 * This releases the entire chain of objects from first_object to and
577 * including stopobj, flowing through object->backing_object.
578 *
579 * We release stopobj first as an optimization as this object is most
580 * likely to be shared across multiple processes.
b12defdc
MD
581 */
582void
583vm_object_chain_release_all(vm_object_t first_object, vm_object_t stopobj)
584{
585 vm_object_t backing_object;
586 vm_object_t object;
587
588 vm_object_chain_release(stopobj);
589 object = first_object;
590
591 while (object != stopobj) {
592 KKASSERT(object);
593 if (object != first_object)
594 vm_object_hold(object);
595 backing_object = object->backing_object;
596 vm_object_chain_release(object);
597 if (object != first_object)
598 vm_object_drop(object);
599 object = backing_object;
600 }
601}
602
603/*
6846fd23
MD
604 * Dereference an object and its underlying vnode.
605 *
b12defdc 606 * The object must be held and will be held on return.
6846fd23 607 */
b33a2ac2 608static void
57e43348 609vm_object_vndeallocate(vm_object_t object)
984263bc
MD
610{
611 struct vnode *vp = (struct vnode *) object->handle;
612
613 KASSERT(object->type == OBJT_VNODE,
614 ("vm_object_vndeallocate: not a vnode object"));
615 KASSERT(vp != NULL, ("vm_object_vndeallocate: missing vp"));
b12defdc 616 ASSERT_LWKT_TOKEN_HELD(vm_object_token(object));
984263bc
MD
617#ifdef INVARIANTS
618 if (object->ref_count == 0) {
619 vprint("vm_object_vndeallocate", vp);
620 panic("vm_object_vndeallocate: bad object reference count");
621 }
622#endif
984263bc 623 object->ref_count--;
b33a2ac2 624 if (object->ref_count == 0)
2247fe02 625 vclrflags(vp, VTEXT);
984263bc
MD
626 vrele(vp);
627}
628
629/*
6846fd23
MD
630 * Release a reference to the specified object, gained either through a
631 * vm_object_allocate or a vm_object_reference call. When all references
632 * are gone, storage associated with this object may be relinquished.
212f39f5
MD
633 *
634 * The caller does not have to hold the object locked but must have control
635 * over the reference in question in order to guarantee that the object
636 * does not get ripped out from under us.
984263bc
MD
637 */
638void
57e43348 639vm_object_deallocate(vm_object_t object)
984263bc 640{
b12defdc
MD
641 if (object) {
642 vm_object_hold(object);
643 vm_object_deallocate_locked(object);
644 vm_object_drop(object);
645 }
2de4f77e
MD
646}
647
648void
649vm_object_deallocate_locked(vm_object_t object)
650{
e806bedd
MD
651 struct vm_object_dealloc_list *dlist = NULL;
652 struct vm_object_dealloc_list *dtmp;
984263bc 653 vm_object_t temp;
b12defdc 654 int must_drop = 0;
212f39f5 655
e806bedd
MD
656 /*
657 * We may chain deallocate object, but additional objects may
658 * collect on the dlist which also have to be deallocated. We
659 * must avoid a recursion, vm_object chains can get deep.
660 */
661again:
984263bc 662 while (object != NULL) {
b12defdc
MD
663#if 0
664 /*
665 * Don't rip a ref_count out from under an object undergoing
666 * collapse, it will confuse the collapse code.
667 */
668 vm_object_chain_wait(object);
669#endif
984263bc
MD
670 if (object->type == OBJT_VNODE) {
671 vm_object_vndeallocate(object);
6846fd23 672 break;
984263bc
MD
673 }
674
675 if (object->ref_count == 0) {
6846fd23
MD
676 panic("vm_object_deallocate: object deallocated "
677 "too many times: %d", object->type);
678 }
679 if (object->ref_count > 2) {
984263bc 680 object->ref_count--;
2de4f77e
MD
681 break;
682 }
683
684 /*
984263bc
MD
685 * Here on ref_count of one or two, which are special cases for
686 * objects.
212f39f5
MD
687 *
688 * Nominal ref_count > 1 case if the second ref is not from
689 * a shadow.
984263bc 690 */
212f39f5 691 if (object->ref_count == 2 && object->shadow_count == 0) {
984263bc
MD
692 vm_object_set_flag(object, OBJ_ONEMAPPING);
693 object->ref_count--;
6846fd23
MD
694 break;
695 }
212f39f5
MD
696
697 /*
698 * If the second ref is from a shadow we chain along it
b12defdc 699 * upwards if object's handle is exhausted.
1f8fc82a
MD
700 *
701 * We have to decrement object->ref_count before potentially
702 * collapsing the first shadow object or the collapse code
b12defdc
MD
703 * will not be able to handle the degenerate case to remove
704 * object. However, if we do it too early the object can
705 * get ripped out from under us.
212f39f5 706 */
b12defdc
MD
707 if (object->ref_count == 2 && object->shadow_count == 1 &&
708 object->handle == NULL && (object->type == OBJT_DEFAULT ||
709 object->type == OBJT_SWAP)) {
710 temp = LIST_FIRST(&object->shadow_head);
711 KKASSERT(temp != NULL);
712 vm_object_hold(temp);
713
714 /*
715 * Wait for any paging to complete so the collapse
716 * doesn't (or isn't likely to) qcollapse. pip
717 * waiting must occur before we acquire the
718 * chainlock.
719 */
720 while (
721 temp->paging_in_progress ||
722 object->paging_in_progress
723 ) {
724 vm_object_pip_wait(temp, "objde1");
725 vm_object_pip_wait(object, "objde2");
726 }
727
728 /*
729 * If the parent is locked we have to give up, as
730 * otherwise we would be acquiring locks in the
731 * wrong order and potentially deadlock.
732 */
733 if (temp->flags & OBJ_CHAINLOCK) {
734 vm_object_drop(temp);
735 goto skip;
736 }
737 vm_object_chain_acquire(temp);
738
739 /*
740 * Recheck/retry after the hold and the paging
741 * wait, both of which can block us.
742 */
743 if (object->ref_count != 2 ||
744 object->shadow_count != 1 ||
745 object->handle ||
746 LIST_FIRST(&object->shadow_head) != temp ||
747 (object->type != OBJT_DEFAULT &&
748 object->type != OBJT_SWAP)) {
749 vm_object_chain_release(temp);
750 vm_object_drop(temp);
751 continue;
752 }
753
754 /*
755 * We can safely drop object's ref_count now.
756 */
757 KKASSERT(object->ref_count == 2);
1f8fc82a 758 object->ref_count--;
984263bc 759
b12defdc
MD
760 /*
761 * If our single parent is not collapseable just
762 * decrement ref_count (2->1) and stop.
763 */
764 if (temp->handle || (temp->type != OBJT_DEFAULT &&
765 temp->type != OBJT_SWAP)) {
766 vm_object_chain_release(temp);
767 vm_object_drop(temp);
768 break;
984263bc 769 }
b12defdc
MD
770
771 /*
772 * At this point we have already dropped object's
773 * ref_count so it is possible for a race to
774 * deallocate obj out from under us. Any collapse
775 * will re-check the situation. We must not block
776 * until we are able to collapse.
777 *
778 * Bump temp's ref_count to avoid an unwanted
779 * degenerate recursion (can't call
780 * vm_object_reference_locked() because it asserts
781 * that CHAINLOCK is not set).
782 */
783 temp->ref_count++;
784 KKASSERT(temp->ref_count > 1);
785
786 /*
787 * Collapse temp, then deallocate the extra ref
788 * formally.
789 */
e806bedd 790 vm_object_collapse(temp, &dlist);
b12defdc
MD
791 vm_object_chain_release(temp);
792 if (must_drop) {
793 vm_object_lock_swap();
794 vm_object_drop(object);
795 }
796 object = temp;
797 must_drop = 1;
798 continue;
984263bc
MD
799 }
800
6846fd23 801 /*
b12defdc
MD
802 * Drop the ref and handle termination on the 1->0 transition.
803 * We may have blocked above so we have to recheck.
6846fd23 804 */
b12defdc
MD
805skip:
806 KKASSERT(object->ref_count != 0);
807 if (object->ref_count >= 2) {
808 object->ref_count--;
6846fd23 809 break;
2de4f77e 810 }
b12defdc 811 KKASSERT(object->ref_count == 1);
6846fd23
MD
812
813 /*
b12defdc
MD
814 * 1->0 transition. Chain through the backing_object.
815 * Maintain the ref until we've located the backing object,
816 * then re-check.
6846fd23 817 */
212f39f5 818 while ((temp = object->backing_object) != NULL) {
b12defdc 819 vm_object_hold(temp);
212f39f5
MD
820 if (temp == object->backing_object)
821 break;
b12defdc 822 vm_object_drop(temp);
212f39f5 823 }
b12defdc
MD
824
825 /*
826 * 1->0 transition verified, retry if ref_count is no longer
827 * 1. Otherwise disconnect the backing_object (temp) and
828 * clean up.
829 */
830 if (object->ref_count != 1) {
831 vm_object_drop(temp);
832 continue;
833 }
834
835 /*
836 * It shouldn't be possible for the object to be chain locked
837 * if we're removing the last ref on it.
838 */
839 KKASSERT((object->flags & OBJ_CHAINLOCK) == 0);
840
984263bc
MD
841 if (temp) {
842 LIST_REMOVE(object, shadow_list);
843 temp->shadow_count--;
984263bc
MD
844 temp->generation++;
845 object->backing_object = NULL;
846 }
847
b12defdc
MD
848 --object->ref_count;
849 if ((object->flags & OBJ_DEAD) == 0)
984263bc 850 vm_object_terminate(object);
b12defdc
MD
851 if (must_drop && temp)
852 vm_object_lock_swap();
853 if (must_drop)
854 vm_object_drop(object);
984263bc 855 object = temp;
b12defdc 856 must_drop = 1;
984263bc 857 }
b12defdc
MD
858 if (must_drop && object)
859 vm_object_drop(object);
e806bedd
MD
860
861 /*
862 * Additional tail recursion on dlist. Avoid a recursion. Objects
863 * on the dlist have a hold count but are not locked.
864 */
865 if ((dtmp = dlist) != NULL) {
866 dlist = dtmp->next;
867 object = dtmp->object;
868 kfree(dtmp, M_TEMP);
869
870 vm_object_lock(object); /* already held, add lock */
871 must_drop = 1; /* and we're responsible for it */
872 goto again;
873 }
984263bc
MD
874}
875
876/*
6846fd23
MD
877 * Destroy the specified object, freeing up related resources.
878 *
879 * The object must have zero references.
984263bc 880 *
b12defdc
MD
881 * The object must held. The caller is responsible for dropping the object
882 * after terminate returns. Terminate does NOT drop the object.
984263bc 883 */
1f804340
MD
884static int vm_object_terminate_callback(vm_page_t p, void *data);
885
984263bc 886void
57e43348 887vm_object_terminate(vm_object_t object)
984263bc 888{
984263bc 889 /*
2de4f77e
MD
890 * Make sure no one uses us. Once we set OBJ_DEAD we should be
891 * able to safely block.
984263bc 892 */
b12defdc 893 ASSERT_LWKT_TOKEN_HELD(vm_object_token(object));
2de4f77e 894 KKASSERT((object->flags & OBJ_DEAD) == 0);
984263bc
MD
895 vm_object_set_flag(object, OBJ_DEAD);
896
897 /*
2de4f77e 898 * Wait for the pageout daemon to be done with the object
984263bc 899 */
e1c14c82 900 vm_object_pip_wait(object, "objtrm1");
984263bc
MD
901
902 KASSERT(!object->paging_in_progress,
903 ("vm_object_terminate: pageout in progress"));
904
905 /*
906 * Clean and free the pages, as appropriate. All references to the
907 * object are gone, so we don't need to lock it.
908 */
909 if (object->type == OBJT_VNODE) {
910 struct vnode *vp;
911
912 /*
984263bc
MD
913 * Clean pages and flush buffers.
914 */
915 vm_object_page_clean(object, 0, 0, OBJPC_SYNC);
916
917 vp = (struct vnode *) object->handle;
87de5057 918 vinvalbuf(vp, V_SAVE, 0, 0);
984263bc
MD
919 }
920
921 /*
922 * Wait for any I/O to complete, after which there had better not
923 * be any references left on the object.
924 */
e1c14c82 925 vm_object_pip_wait(object, "objtrm2");
984263bc 926
2de4f77e
MD
927 if (object->ref_count != 0) {
928 panic("vm_object_terminate: object with references, "
929 "ref_count=%d", object->ref_count);
930 }
984263bc
MD
931
932 /*
933 * Now free any remaining pages. For internal objects, this also
934 * removes them from paging queues. Don't free wired pages, just
935 * remove them from the object.
936 */
1f804340
MD
937 vm_page_rb_tree_RB_SCAN(&object->rb_memq, NULL,
938 vm_object_terminate_callback, NULL);
984263bc
MD
939
940 /*
398c240d 941 * Let the pager know object is dead.
e1c14c82 942 */
398c240d 943 vm_pager_deallocate(object);
e1c14c82
VS
944
945 /*
b12defdc
MD
946 * Wait for the object hold count to hit 1, clean out pages as
947 * we go. vmobj_token interlocks any race conditions that might
948 * pick the object up from the vm_object_list after we have cleared
949 * rb_memq.
984263bc 950 */
34542daf 951 for (;;) {
34542daf
MD
952 if (RB_ROOT(&object->rb_memq) == NULL)
953 break;
954 kprintf("vm_object_terminate: Warning, object %p "
955 "still has %d pages\n",
956 object, object->resident_page_count);
957 vm_page_rb_tree_RB_SCAN(&object->rb_memq, NULL,
958 vm_object_terminate_callback, NULL);
959 }
34542daf
MD
960
961 /*
962 * There had better not be any pages left
963 */
964 KKASSERT(object->resident_page_count == 0);
984263bc
MD
965
966 /*
967 * Remove the object from the global object list.
968 */
b12defdc 969 lwkt_gettoken(&vmobj_token);
984263bc 970 TAILQ_REMOVE(&vm_object_list, object, object_list);
fad57d0e 971 vm_object_count--;
b12defdc 972 lwkt_reltoken(&vmobj_token);
85946b6c 973 vm_object_dead_wakeup(object);
2de4f77e
MD
974
975 if (object->ref_count != 0) {
976 panic("vm_object_terminate2: object with references, "
977 "ref_count=%d", object->ref_count);
978 }
984263bc
MD
979
980 /*
b12defdc
MD
981 * NOTE: The object hold_count is at least 1, so we cannot zfree()
982 * the object here. See vm_object_drop().
984263bc 983 */
984263bc
MD
984}
985
6846fd23 986/*
b12defdc 987 * The caller must hold the object.
6846fd23 988 */
1f804340
MD
989static int
990vm_object_terminate_callback(vm_page_t p, void *data __unused)
991{
b12defdc
MD
992 vm_object_t object;
993
994 object = p->object;
ac952071 995 vm_page_busy_wait(p, TRUE, "vmpgtrm");
b12defdc
MD
996 if (object != p->object) {
997 kprintf("vm_object_terminate: Warning: Encountered "
998 "busied page %p on queue %d\n", p, p->queue);
999 vm_page_wakeup(p);
1000 } else if (p->wire_count == 0) {
9bf025db
MD
1001 /*
1002 * NOTE: PG_NEED_COMMIT is ignored.
1003 */
1f804340
MD
1004 vm_page_free(p);
1005 mycpu->gd_cnt.v_pfree++;
1006 } else {
c4ba48c4 1007 if (p->queue != PQ_NONE)
b12defdc
MD
1008 kprintf("vm_object_terminate: Warning: Encountered "
1009 "wired page %p on queue %d\n", p, p->queue);
1f804340
MD
1010 vm_page_remove(p);
1011 vm_page_wakeup(p);
1012 }
d2d8515b 1013 lwkt_yield();
1f804340
MD
1014 return(0);
1015}
1016
984263bc 1017/*
9e12ff11
MD
1018 * The object is dead but still has an object<->pager association. Sleep
1019 * and return. The caller typically retests the association in a loop.
6846fd23 1020 *
b12defdc 1021 * The caller must hold the object.
9e12ff11
MD
1022 */
1023void
1024vm_object_dead_sleep(vm_object_t object, const char *wmesg)
1025{
b12defdc 1026 ASSERT_LWKT_TOKEN_HELD(vm_object_token(object));
9e12ff11
MD
1027 if (object->handle) {
1028 vm_object_set_flag(object, OBJ_DEADWNT);
1029 tsleep(object, 0, wmesg, 0);
2de4f77e 1030 /* object may be invalid after this point */
9e12ff11 1031 }
9e12ff11
MD
1032}
1033
1034/*
1035 * Wakeup anyone waiting for the object<->pager disassociation on
1036 * a dead object.
6846fd23 1037 *
b12defdc 1038 * The caller must hold the object.
9e12ff11
MD
1039 */
1040void
1041vm_object_dead_wakeup(vm_object_t object)
1042{
b12defdc 1043 ASSERT_LWKT_TOKEN_HELD(vm_object_token(object));
9e12ff11
MD
1044 if (object->flags & OBJ_DEADWNT) {
1045 vm_object_clear_flag(object, OBJ_DEADWNT);
1046 wakeup(object);
1047 }
1048}
1049
1050/*
6846fd23
MD
1051 * Clean all dirty pages in the specified range of object. Leaves page
1052 * on whatever queue it is currently on. If NOSYNC is set then do not
1053 * write out pages with PG_NOSYNC set (originally comes from MAP_NOSYNC),
1054 * leaving the object dirty.
984263bc 1055 *
6846fd23
MD
1056 * When stuffing pages asynchronously, allow clustering. XXX we need a
1057 * synchronous clustering mode implementation.
984263bc 1058 *
6846fd23 1059 * Odd semantics: if start == end, we clean everything.
984263bc 1060 *
6846fd23 1061 * The object must be locked? XXX
984263bc 1062 */
1f804340
MD
1063static int vm_object_page_clean_pass1(struct vm_page *p, void *data);
1064static int vm_object_page_clean_pass2(struct vm_page *p, void *data);
984263bc
MD
1065
1066void
57e43348 1067vm_object_page_clean(vm_object_t object, vm_pindex_t start, vm_pindex_t end,
1f804340 1068 int flags)
984263bc 1069{
1f804340 1070 struct rb_vm_page_scan_info info;
984263bc 1071 struct vnode *vp;
1f804340 1072 int wholescan;
984263bc 1073 int pagerflags;
b12defdc 1074 int generation;
984263bc 1075
edd6da8e 1076 vm_object_hold(object);
984263bc 1077 if (object->type != OBJT_VNODE ||
6846fd23 1078 (object->flags & OBJ_MIGHTBEDIRTY) == 0) {
edd6da8e 1079 vm_object_drop(object);
984263bc 1080 return;
6846fd23 1081 }
984263bc 1082
1f804340
MD
1083 pagerflags = (flags & (OBJPC_SYNC | OBJPC_INVAL)) ?
1084 VM_PAGER_PUT_SYNC : VM_PAGER_CLUSTER_OK;
984263bc
MD
1085 pagerflags |= (flags & OBJPC_INVAL) ? VM_PAGER_PUT_INVAL : 0;
1086
1087 vp = object->handle;
1088
1f804340
MD
1089 /*
1090 * Interlock other major object operations. This allows us to
1091 * temporarily clear OBJ_WRITEABLE and OBJ_MIGHTBEDIRTY.
1092 */
984263bc
MD
1093 vm_object_set_flag(object, OBJ_CLEANING);
1094
1095 /*
1096 * Handle 'entire object' case
1097 */
1f804340 1098 info.start_pindex = start;
984263bc 1099 if (end == 0) {
1f804340 1100 info.end_pindex = object->size - 1;
984263bc 1101 } else {
1f804340 1102 info.end_pindex = end - 1;
984263bc 1103 }
1f804340
MD
1104 wholescan = (start == 0 && info.end_pindex == object->size - 1);
1105 info.limit = flags;
1106 info.pagerflags = pagerflags;
1107 info.object = object;
984263bc
MD
1108
1109 /*
1f804340
MD
1110 * If cleaning the entire object do a pass to mark the pages read-only.
1111 * If everything worked out ok, clear OBJ_WRITEABLE and
1112 * OBJ_MIGHTBEDIRTY.
984263bc 1113 */
1f804340
MD
1114 if (wholescan) {
1115 info.error = 0;
1116 vm_page_rb_tree_RB_SCAN(&object->rb_memq, rb_vm_page_scancmp,
1117 vm_object_page_clean_pass1, &info);
1118 if (info.error == 0) {
1119 vm_object_clear_flag(object,
1120 OBJ_WRITEABLE|OBJ_MIGHTBEDIRTY);
1121 if (object->type == OBJT_VNODE &&
1122 (vp = (struct vnode *)object->handle) != NULL) {
1123 if (vp->v_flag & VOBJDIRTY)
1124 vclrflags(vp, VOBJDIRTY);
984263bc 1125 }
984263bc 1126 }
984263bc
MD
1127 }
1128
1129 /*
1f804340 1130 * Do a pass to clean all the dirty pages we find.
984263bc 1131 */
1f804340
MD
1132 do {
1133 info.error = 0;
b12defdc 1134 generation = object->generation;
1f804340
MD
1135 vm_page_rb_tree_RB_SCAN(&object->rb_memq, rb_vm_page_scancmp,
1136 vm_object_page_clean_pass2, &info);
b12defdc 1137 } while (info.error || generation != object->generation);
984263bc 1138
1f804340 1139 vm_object_clear_flag(object, OBJ_CLEANING);
edd6da8e 1140 vm_object_drop(object);
1f804340 1141}
984263bc 1142
6846fd23 1143/*
b12defdc 1144 * The caller must hold the object.
6846fd23 1145 */
1f804340
MD
1146static
1147int
1148vm_object_page_clean_pass1(struct vm_page *p, void *data)
1149{
1150 struct rb_vm_page_scan_info *info = data;
984263bc 1151
1f804340 1152 vm_page_flag_set(p, PG_CLEANCHK);
b12defdc 1153 if ((info->limit & OBJPC_NOSYNC) && (p->flags & PG_NOSYNC)) {
1f804340 1154 info->error = 1;
b12defdc 1155 } else if (vm_page_busy_try(p, FALSE) == 0) {
17cde63e 1156 vm_page_protect(p, VM_PROT_READ); /* must not block */
b12defdc
MD
1157 vm_page_wakeup(p);
1158 } else {
1159 info->error = 1;
1160 }
fc9ed34d 1161 lwkt_yield();
1f804340
MD
1162 return(0);
1163}
1164
6846fd23 1165/*
b12defdc 1166 * The caller must hold the object
6846fd23 1167 */
1f804340
MD
1168static
1169int
1170vm_object_page_clean_pass2(struct vm_page *p, void *data)
1171{
1172 struct rb_vm_page_scan_info *info = data;
b12defdc 1173 int generation;
984263bc 1174
06ecca5a 1175 /*
1f804340
MD
1176 * Do not mess with pages that were inserted after we started
1177 * the cleaning pass.
06ecca5a 1178 */
1f804340 1179 if ((p->flags & PG_CLEANCHK) == 0)
fc9ed34d 1180 goto done;
984263bc 1181
b12defdc
MD
1182 generation = info->object->generation;
1183 vm_page_busy_wait(p, TRUE, "vpcwai");
1184 if (p->object != info->object ||
1185 info->object->generation != generation) {
1186 info->error = 1;
1187 vm_page_wakeup(p);
fc9ed34d 1188 goto done;
b12defdc
MD
1189 }
1190
1f804340
MD
1191 /*
1192 * Before wasting time traversing the pmaps, check for trivial
1193 * cases where the page cannot be dirty.
1194 */
1195 if (p->valid == 0 || (p->queue - p->pc) == PQ_CACHE) {
1196 KKASSERT((p->dirty & p->valid) == 0);
b12defdc 1197 vm_page_wakeup(p);
fc9ed34d 1198 goto done;
1f804340 1199 }
984263bc 1200
1f804340
MD
1201 /*
1202 * Check whether the page is dirty or not. The page has been set
1203 * to be read-only so the check will not race a user dirtying the
1204 * page.
1205 */
1206 vm_page_test_dirty(p);
1207 if ((p->dirty & p->valid) == 0) {
1208 vm_page_flag_clear(p, PG_CLEANCHK);
b12defdc 1209 vm_page_wakeup(p);
fc9ed34d 1210 goto done;
1f804340 1211 }
984263bc 1212
1f804340
MD
1213 /*
1214 * If we have been asked to skip nosync pages and this is a
1215 * nosync page, skip it. Note that the object flags were
1216 * not cleared in this case (because pass1 will have returned an
1217 * error), so we do not have to set them.
1218 */
1219 if ((info->limit & OBJPC_NOSYNC) && (p->flags & PG_NOSYNC)) {
1220 vm_page_flag_clear(p, PG_CLEANCHK);
b12defdc 1221 vm_page_wakeup(p);
fc9ed34d 1222 goto done;
984263bc
MD
1223 }
1224
1f804340
MD
1225 /*
1226 * Flush as many pages as we can. PG_CLEANCHK will be cleared on
1227 * the pages that get successfully flushed. Set info->error if
1228 * we raced an object modification.
1229 */
b12defdc 1230 vm_object_page_collect_flush(info->object, p, info->pagerflags);
55b50bd5 1231 vm_wait_nominal();
fc9ed34d
MD
1232done:
1233 lwkt_yield();
1f804340 1234 return(0);
984263bc
MD
1235}
1236
06ecca5a 1237/*
6846fd23 1238 * Collect the specified page and nearby pages and flush them out.
b12defdc
MD
1239 * The number of pages flushed is returned. The passed page is busied
1240 * by the caller and we are responsible for its disposition.
06ecca5a 1241 *
b12defdc 1242 * The caller must hold the object.
06ecca5a 1243 */
ab3e1edd 1244static void
1f804340 1245vm_object_page_collect_flush(vm_object_t object, vm_page_t p, int pagerflags)
984263bc
MD
1246{
1247 int runlen;
b12defdc 1248 int error;
984263bc
MD
1249 int maxf;
1250 int chkb;
1251 int maxb;
1252 int i;
1253 vm_pindex_t pi;
1254 vm_page_t maf[vm_pageout_page_count];
1255 vm_page_t mab[vm_pageout_page_count];
1256 vm_page_t ma[vm_pageout_page_count];
1257
b12defdc 1258 ASSERT_LWKT_TOKEN_HELD(vm_object_token(object));
1f804340 1259
984263bc 1260 pi = p->pindex;
984263bc
MD
1261
1262 maxf = 0;
1263 for(i = 1; i < vm_pageout_page_count; i++) {
1264 vm_page_t tp;
1265
b12defdc
MD
1266 tp = vm_page_lookup_busy_try(object, pi + i, TRUE, &error);
1267 if (error)
1268 break;
1269 if (tp == NULL)
1270 break;
1271 if ((pagerflags & VM_PAGER_IGNORE_CLEANCHK) == 0 &&
1272 (tp->flags & PG_CLEANCHK) == 0) {
1273 vm_page_wakeup(tp);
1274 break;
984263bc 1275 }
b12defdc
MD
1276 if ((tp->queue - tp->pc) == PQ_CACHE) {
1277 vm_page_flag_clear(tp, PG_CLEANCHK);
1278 vm_page_wakeup(tp);
1279 break;
1280 }
1281 vm_page_test_dirty(tp);
1282 if ((tp->dirty & tp->valid) == 0) {
1283 vm_page_flag_clear(tp, PG_CLEANCHK);
1284 vm_page_wakeup(tp);
1285 break;
1286 }
1287 maf[i - 1] = tp;
1288 maxf++;
984263bc
MD
1289 }
1290
1291 maxb = 0;
1292 chkb = vm_pageout_page_count - maxf;
b12defdc
MD
1293 /*
1294 * NOTE: chkb can be 0
1295 */
1296 for(i = 1; chkb && i < chkb; i++) {
1297 vm_page_t tp;
1298
1299 tp = vm_page_lookup_busy_try(object, pi - i, TRUE, &error);
1300 if (error)
1301 break;
1302 if (tp == NULL)
1303 break;
1304 if ((pagerflags & VM_PAGER_IGNORE_CLEANCHK) == 0 &&
1305 (tp->flags & PG_CLEANCHK) == 0) {
1306 vm_page_wakeup(tp);
1307 break;
1308 }
1309 if ((tp->queue - tp->pc) == PQ_CACHE) {
1310 vm_page_flag_clear(tp, PG_CLEANCHK);
1311 vm_page_wakeup(tp);
984263bc
MD
1312 break;
1313 }
b12defdc
MD
1314 vm_page_test_dirty(tp);
1315 if ((tp->dirty & tp->valid) == 0) {
1316 vm_page_flag_clear(tp, PG_CLEANCHK);
1317 vm_page_wakeup(tp);
1318 break;
1319 }
1320 mab[i - 1] = tp;
1321 maxb++;
984263bc
MD
1322 }
1323
b12defdc
MD
1324 /*
1325 * All pages in the maf[] and mab[] array are busied.
1326 */
1327 for (i = 0; i < maxb; i++) {
984263bc
MD
1328 int index = (maxb - i) - 1;
1329 ma[index] = mab[i];
1330 vm_page_flag_clear(ma[index], PG_CLEANCHK);
1331 }
1332 vm_page_flag_clear(p, PG_CLEANCHK);
1333 ma[maxb] = p;
1334 for(i = 0; i < maxf; i++) {
1335 int index = (maxb + i) + 1;
1336 ma[index] = maf[i];
1337 vm_page_flag_clear(ma[index], PG_CLEANCHK);
1338 }
1339 runlen = maxb + maxf + 1;
1340
b12defdc
MD
1341 for (i = 0; i < runlen; i++)
1342 vm_page_hold(ma[i]);
1343
984263bc 1344 vm_pageout_flush(ma, runlen, pagerflags);
b12defdc 1345
ab3e1edd
MD
1346 /*
1347 * WARNING: Related pages are still held but the BUSY was inherited
1348 * by the pageout I/O, so the pages might not be busy any
1349 * more. We cannot re-protect the page without waiting
1350 * for the I/O to complete and then busying it again.
1351 */
984263bc
MD
1352 for (i = 0; i < runlen; i++) {
1353 if (ma[i]->valid & ma[i]->dirty) {
ab3e1edd 1354 /*vm_page_protect(ma[i], VM_PROT_READ);*/
984263bc
MD
1355 vm_page_flag_set(ma[i], PG_CLEANCHK);
1356
1357 /*
1358 * maxf will end up being the actual number of pages
1359 * we wrote out contiguously, non-inclusive of the
1360 * first page. We do not count look-behind pages.
1361 */
1362 if (i >= maxb + 1 && (maxf > i - maxb - 1))
1363 maxf = i - maxb - 1;
1364 }
b12defdc 1365 vm_page_unhold(ma[i]);
984263bc 1366 }
ab3e1edd 1367 /*return(maxf + 1);*/
984263bc
MD
1368}
1369
984263bc
MD
1370/*
1371 * Same as vm_object_pmap_copy, except range checking really
1372 * works, and is meant for small sections of an object.
1373 *
1374 * This code protects resident pages by making them read-only
1375 * and is typically called on a fork or split when a page
1376 * is converted to copy-on-write.
1377 *
1378 * NOTE: If the page is already at VM_PROT_NONE, calling
1379 * vm_page_protect will have no effect.
1380 */
984263bc 1381void
57e43348 1382vm_object_pmap_copy_1(vm_object_t object, vm_pindex_t start, vm_pindex_t end)
984263bc
MD
1383{
1384 vm_pindex_t idx;
1385 vm_page_t p;
1386
1387 if (object == NULL || (object->flags & OBJ_WRITEABLE) == 0)
1388 return;
1389
b12defdc 1390 vm_object_hold(object);
984263bc
MD
1391 for (idx = start; idx < end; idx++) {
1392 p = vm_page_lookup(object, idx);
1393 if (p == NULL)
1394 continue;
1395 vm_page_protect(p, VM_PROT_READ);
1396 }
b12defdc 1397 vm_object_drop(object);
984263bc
MD
1398}
1399
1400/*
6846fd23
MD
1401 * Removes all physical pages in the specified object range from all
1402 * physical maps.
984263bc 1403 *
6846fd23 1404 * The object must *not* be locked.
984263bc 1405 */
1f804340
MD
1406
1407static int vm_object_pmap_remove_callback(vm_page_t p, void *data);
1408
984263bc 1409void
57e43348 1410vm_object_pmap_remove(vm_object_t object, vm_pindex_t start, vm_pindex_t end)
984263bc 1411{
1f804340 1412 struct rb_vm_page_scan_info info;
984263bc
MD
1413
1414 if (object == NULL)
1415 return;
1f804340
MD
1416 info.start_pindex = start;
1417 info.end_pindex = end - 1;
6846fd23 1418
b12defdc 1419 vm_object_hold(object);
1f804340
MD
1420 vm_page_rb_tree_RB_SCAN(&object->rb_memq, rb_vm_page_scancmp,
1421 vm_object_pmap_remove_callback, &info);
1422 if (start == 0 && end == object->size)
984263bc 1423 vm_object_clear_flag(object, OBJ_WRITEABLE);
b12defdc 1424 vm_object_drop(object);
1f804340
MD
1425}
1426
6846fd23 1427/*
b12defdc 1428 * The caller must hold the object
6846fd23 1429 */
1f804340
MD
1430static int
1431vm_object_pmap_remove_callback(vm_page_t p, void *data __unused)
1432{
1433 vm_page_protect(p, VM_PROT_NONE);
1434 return(0);
984263bc
MD
1435}
1436
1437/*
6846fd23 1438 * Implements the madvise function at the object/page level.
984263bc 1439 *
6846fd23 1440 * MADV_WILLNEED (any object)
984263bc 1441 *
6846fd23 1442 * Activate the specified pages if they are resident.
984263bc 1443 *
6846fd23 1444 * MADV_DONTNEED (any object)
984263bc 1445 *
6846fd23 1446 * Deactivate the specified pages if they are resident.
984263bc 1447 *
6846fd23 1448 * MADV_FREE (OBJT_DEFAULT/OBJT_SWAP objects, OBJ_ONEMAPPING only)
984263bc 1449 *
6846fd23
MD
1450 * Deactivate and clean the specified pages if they are
1451 * resident. This permits the process to reuse the pages
1452 * without faulting or the kernel to reclaim the pages
1453 * without I/O.
984263bc 1454 *
6846fd23 1455 * No requirements.
984263bc
MD
1456 */
1457void
57e43348 1458vm_object_madvise(vm_object_t object, vm_pindex_t pindex, int count, int advise)
984263bc
MD
1459{
1460 vm_pindex_t end, tpindex;
1461 vm_object_t tobject;
b12defdc 1462 vm_object_t xobj;
984263bc 1463 vm_page_t m;
b12defdc 1464 int error;
984263bc
MD
1465
1466 if (object == NULL)
1467 return;
1468
1469 end = pindex + count;
1470
b12defdc
MD
1471 vm_object_hold(object);
1472 tobject = object;
6846fd23 1473
984263bc
MD
1474 /*
1475 * Locate and adjust resident pages
1476 */
984263bc
MD
1477 for (; pindex < end; pindex += 1) {
1478relookup:
b12defdc
MD
1479 if (tobject != object)
1480 vm_object_drop(tobject);
984263bc
MD
1481 tobject = object;
1482 tpindex = pindex;
1483shadowlookup:
1484 /*
1485 * MADV_FREE only operates on OBJT_DEFAULT or OBJT_SWAP pages
1486 * and those pages must be OBJ_ONEMAPPING.
1487 */
1488 if (advise == MADV_FREE) {
1489 if ((tobject->type != OBJT_DEFAULT &&
1490 tobject->type != OBJT_SWAP) ||
1491 (tobject->flags & OBJ_ONEMAPPING) == 0) {
1492 continue;
1493 }
1494 }
1495
b12defdc 1496 m = vm_page_lookup_busy_try(tobject, tpindex, TRUE, &error);
984263bc 1497
b12defdc
MD
1498 if (error) {
1499 vm_page_sleep_busy(m, TRUE, "madvpo");
1500 goto relookup;
1501 }
984263bc
MD
1502 if (m == NULL) {
1503 /*
1504 * There may be swap even if there is no backing page
1505 */
1506 if (advise == MADV_FREE && tobject->type == OBJT_SWAP)
1507 swap_pager_freespace(tobject, tpindex, 1);
1508
1509 /*
1510 * next object
1511 */
b12defdc
MD
1512 while ((xobj = tobject->backing_object) != NULL) {
1513 KKASSERT(xobj != object);
1514 vm_object_hold(xobj);
1515 if (xobj == tobject->backing_object)
1516 break;
1517 vm_object_drop(xobj);
1518 }
1519 if (xobj == NULL)
984263bc
MD
1520 continue;
1521 tpindex += OFF_TO_IDX(tobject->backing_object_offset);
b12defdc
MD
1522 if (tobject != object) {
1523 vm_object_lock_swap();
1524 vm_object_drop(tobject);
1525 }
1526 tobject = xobj;
984263bc
MD
1527 goto shadowlookup;
1528 }
1529
1530 /*
b12defdc
MD
1531 * If the page is not in a normal active state, we skip it.
1532 * If the page is not managed there are no page queues to
1533 * mess with. Things can break if we mess with pages in
1534 * any of the below states.
984263bc 1535 */
9bf025db
MD
1536 if (m->wire_count ||
1537 (m->flags & (PG_UNMANAGED | PG_NEED_COMMIT)) ||
984263bc
MD
1538 m->valid != VM_PAGE_BITS_ALL
1539 ) {
b12defdc 1540 vm_page_wakeup(m);
984263bc
MD
1541 continue;
1542 }
1543
06ecca5a
MD
1544 /*
1545 * Theoretically once a page is known not to be busy, an
1546 * interrupt cannot come along and rip it out from under us.
1547 */
984263bc
MD
1548
1549 if (advise == MADV_WILLNEED) {
1550 vm_page_activate(m);
1551 } else if (advise == MADV_DONTNEED) {
1552 vm_page_dontneed(m);
1553 } else if (advise == MADV_FREE) {
1554 /*
1555 * Mark the page clean. This will allow the page
1556 * to be freed up by the system. However, such pages
1557 * are often reused quickly by malloc()/free()
1558 * so we do not do anything that would cause
1559 * a page fault if we can help it.
1560 *
1561 * Specifically, we do not try to actually free
1562 * the page now nor do we try to put it in the
1563 * cache (which would cause a page fault on reuse).
1564 *
1565 * But we do make the page is freeable as we
1566 * can without actually taking the step of unmapping
1567 * it.
1568 */
1569 pmap_clear_modify(m);
1570 m->dirty = 0;
1571 m->act_count = 0;
1572 vm_page_dontneed(m);
1573 if (tobject->type == OBJT_SWAP)
1574 swap_pager_freespace(tobject, tpindex, 1);
1575 }
a491077e 1576 vm_page_wakeup(m);
984263bc 1577 }
b12defdc
MD
1578 if (tobject != object)
1579 vm_object_drop(tobject);
1580 vm_object_drop(object);
984263bc
MD
1581}
1582
1583/*
6846fd23 1584 * Create a new object which is backed by the specified existing object
b12defdc
MD
1585 * range. Replace the pointer and offset that was pointing at the existing
1586 * object with the pointer/offset for the new object.
984263bc 1587 *
6846fd23 1588 * No other requirements.
984263bc 1589 */
984263bc 1590void
b12defdc
MD
1591vm_object_shadow(vm_object_t *objectp, vm_ooffset_t *offset, vm_size_t length,
1592 int addref)
984263bc
MD
1593{
1594 vm_object_t source;
1595 vm_object_t result;
1596
b12defdc 1597 source = *objectp;
984263bc
MD
1598
1599 /*
1600 * Don't create the new object if the old object isn't shared.
b12defdc
MD
1601 * We have to chain wait before adding the reference to avoid
1602 * racing a collapse or deallocation.
1603 *
1604 * Add the additional ref to source here to avoid racing a later
1605 * collapse or deallocation. Clear the ONEMAPPING flag whether
1606 * addref is TRUE or not in this case because the original object
1607 * will be shadowed.
984263bc 1608 */
b12defdc
MD
1609 if (source) {
1610 vm_object_hold(source);
1611 vm_object_chain_wait(source);
1612 if (source->ref_count == 1 &&
1613 source->handle == NULL &&
1614 (source->type == OBJT_DEFAULT ||
1615 source->type == OBJT_SWAP)) {
1616 vm_object_drop(source);
1617 if (addref) {
b12defdc 1618 vm_object_reference_locked(source);
6056eb53 1619 vm_object_clear_flag(source, OBJ_ONEMAPPING);
b12defdc
MD
1620 }
1621 return;
1622 }
1623 vm_object_reference_locked(source);
1624 vm_object_clear_flag(source, OBJ_ONEMAPPING);
6846fd23 1625 }
984263bc
MD
1626
1627 /*
b12defdc
MD
1628 * Allocate a new object with the given length. The new object
1629 * is returned referenced but we may have to add another one.
1630 * If we are adding a second reference we must clear OBJ_ONEMAPPING.
1631 * (typically because the caller is about to clone a vm_map_entry).
1632 *
1633 * The source object currently has an extra reference to prevent
1634 * collapses into it while we mess with its shadow list, which
1635 * we will remove later in this routine.
984263bc 1636 */
984263bc
MD
1637 if ((result = vm_object_allocate(OBJT_DEFAULT, length)) == NULL)
1638 panic("vm_object_shadow: no object for shadowing");
b12defdc
MD
1639 vm_object_hold(result);
1640 if (addref) {
1641 vm_object_reference_locked(result);
1642 vm_object_clear_flag(result, OBJ_ONEMAPPING);
1643 }
984263bc
MD
1644
1645 /*
b12defdc
MD
1646 * The new object shadows the source object. Chain wait before
1647 * adjusting shadow_count or the shadow list to avoid races.
984263bc
MD
1648 *
1649 * Try to optimize the result object's page color when shadowing
1650 * in order to maintain page coloring consistency in the combined
1651 * shadowed object.
1652 */
b12defdc 1653 KKASSERT(result->backing_object == NULL);
984263bc
MD
1654 result->backing_object = source;
1655 if (source) {
b12defdc 1656 vm_object_chain_wait(source);
984263bc
MD
1657 LIST_INSERT_HEAD(&source->shadow_head, result, shadow_list);
1658 source->shadow_count++;
1659 source->generation++;
85946b6c
MD
1660#ifdef SMP
1661 /* cpu localization twist */
1662 result->pg_color = (int)(intptr_t)curthread;
1663#else
b12defdc
MD
1664 result->pg_color = (source->pg_color + OFF_TO_IDX(*offset)) &
1665 PQ_L2_MASK;
85946b6c 1666#endif
984263bc
MD
1667 }
1668
1669 /*
b12defdc
MD
1670 * Adjust the return storage. Drop the ref on source before
1671 * returning.
984263bc 1672 */
984263bc 1673 result->backing_object_offset = *offset;
b12defdc
MD
1674 vm_object_drop(result);
1675 *offset = 0;
1676 if (source) {
1677 vm_object_deallocate_locked(source);
1678 vm_object_drop(source);
1679 }
984263bc
MD
1680
1681 /*
1682 * Return the new things
1683 */
b12defdc 1684 *objectp = result;
984263bc
MD
1685}
1686
1687#define OBSC_TEST_ALL_SHADOWED 0x0001
1688#define OBSC_COLLAPSE_NOWAIT 0x0002
1689#define OBSC_COLLAPSE_WAIT 0x0004
1690
1f804340
MD
1691static int vm_object_backing_scan_callback(vm_page_t p, void *data);
1692
6846fd23 1693/*
b12defdc 1694 * The caller must hold the object.
6846fd23 1695 */
984263bc 1696static __inline int
b12defdc 1697vm_object_backing_scan(vm_object_t object, vm_object_t backing_object, int op)
984263bc 1698{
1f804340 1699 struct rb_vm_page_scan_info info;
984263bc 1700
b12defdc
MD
1701 vm_object_assert_held(object);
1702 vm_object_assert_held(backing_object);
1703
1704 KKASSERT(backing_object == object->backing_object);
1f804340 1705 info.backing_offset_index = OFF_TO_IDX(object->backing_object_offset);
984263bc
MD
1706
1707 /*
1708 * Initial conditions
1709 */
984263bc
MD
1710 if (op & OBSC_TEST_ALL_SHADOWED) {
1711 /*
1712 * We do not want to have to test for the existence of
1713 * swap pages in the backing object. XXX but with the
1714 * new swapper this would be pretty easy to do.
1715 *
1716 * XXX what about anonymous MAP_SHARED memory that hasn't
1717 * been ZFOD faulted yet? If we do not test for this, the
1718 * shadow test may succeed! XXX
1719 */
b12defdc 1720 if (backing_object->type != OBJT_DEFAULT)
984263bc 1721 return(0);
984263bc
MD
1722 }
1723 if (op & OBSC_COLLAPSE_WAIT) {
fad57d0e 1724 KKASSERT((backing_object->flags & OBJ_DEAD) == 0);
984263bc 1725 vm_object_set_flag(backing_object, OBJ_DEAD);
b12defdc
MD
1726 lwkt_gettoken(&vmobj_token);
1727 TAILQ_REMOVE(&vm_object_list, backing_object, object_list);
1728 vm_object_count--;
b12defdc 1729 lwkt_reltoken(&vmobj_token);
85946b6c 1730 vm_object_dead_wakeup(backing_object);
984263bc
MD
1731 }
1732
1733 /*
1f804340
MD
1734 * Our scan. We have to retry if a negative error code is returned,
1735 * otherwise 0 or 1 will be returned in info.error. 0 Indicates that
1736 * the scan had to be stopped because the parent does not completely
1737 * shadow the child.
984263bc 1738 */
1f804340
MD
1739 info.object = object;
1740 info.backing_object = backing_object;
1741 info.limit = op;
1742 do {
1743 info.error = 1;
1744 vm_page_rb_tree_RB_SCAN(&backing_object->rb_memq, NULL,
1745 vm_object_backing_scan_callback,
1746 &info);
1747 } while (info.error < 0);
a5fc46c9 1748
1f804340
MD
1749 return(info.error);
1750}
984263bc 1751
6846fd23 1752/*
b12defdc 1753 * The caller must hold the object.
6846fd23 1754 */
1f804340
MD
1755static int
1756vm_object_backing_scan_callback(vm_page_t p, void *data)
1757{
1758 struct rb_vm_page_scan_info *info = data;
1759 vm_object_t backing_object;
1760 vm_object_t object;
54341a3b 1761 vm_pindex_t pindex;
1f804340
MD
1762 vm_pindex_t new_pindex;
1763 vm_pindex_t backing_offset_index;
1764 int op;
984263bc 1765
54341a3b
MD
1766 pindex = p->pindex;
1767 new_pindex = pindex - info->backing_offset_index;
1f804340
MD
1768 op = info->limit;
1769 object = info->object;
1770 backing_object = info->backing_object;
1771 backing_offset_index = info->backing_offset_index;
984263bc 1772
1f804340
MD
1773 if (op & OBSC_TEST_ALL_SHADOWED) {
1774 vm_page_t pp;
984263bc 1775
1f804340
MD
1776 /*
1777 * Ignore pages outside the parent object's range
1778 * and outside the parent object's mapping of the
1779 * backing object.
1780 *
1781 * note that we do not busy the backing object's
1782 * page.
1783 */
54341a3b 1784 if (pindex < backing_offset_index ||
1f804340
MD
1785 new_pindex >= object->size
1786 ) {
1787 return(0);
984263bc
MD
1788 }
1789
1790 /*
1f804340
MD
1791 * See if the parent has the page or if the parent's
1792 * object pager has the page. If the parent has the
1793 * page but the page is not valid, the parent's
1794 * object pager must have the page.
1795 *
1796 * If this fails, the parent does not completely shadow
1797 * the object and we might as well give up now.
984263bc 1798 */
1f804340 1799 pp = vm_page_lookup(object, new_pindex);
1b9d3514
MD
1800 if ((pp == NULL || pp->valid == 0) &&
1801 !vm_pager_has_page(object, new_pindex)
1f804340
MD
1802 ) {
1803 info->error = 0; /* problemo */
1804 return(-1); /* stop the scan */
1805 }
1806 }
984263bc 1807
1f804340 1808 /*
54341a3b
MD
1809 * Check for busy page. Note that we may have lost (p) when we
1810 * possibly blocked above.
1f804340 1811 */
1f804340
MD
1812 if (op & (OBSC_COLLAPSE_WAIT | OBSC_COLLAPSE_NOWAIT)) {
1813 vm_page_t pp;
984263bc 1814
b12defdc
MD
1815 if (vm_page_busy_try(p, TRUE)) {
1816 if (op & OBSC_COLLAPSE_NOWAIT) {
1f804340 1817 return(0);
b12defdc 1818 } else {
984263bc 1819 /*
1f804340
MD
1820 * If we slept, anything could have
1821 * happened. Ask that the scan be restarted.
984263bc 1822 *
1f804340
MD
1823 * Since the object is marked dead, the
1824 * backing offset should not have changed.
984263bc 1825 */
b12defdc 1826 vm_page_sleep_busy(p, TRUE, "vmocol");
1f804340
MD
1827 info->error = -1;
1828 return(-1);
984263bc 1829 }
1f804340 1830 }
54341a3b
MD
1831
1832 /*
1833 * If (p) is no longer valid restart the scan.
1834 */
1835 if (p->object != backing_object || p->pindex != pindex) {
1836 kprintf("vm_object_backing_scan: Warning: page "
1837 "%p ripped out from under us\n", p);
1838 vm_page_wakeup(p);
1839 info->error = -1;
1840 return(-1);
1841 }
1842
b12defdc 1843 if (op & OBSC_COLLAPSE_NOWAIT) {
9bf025db
MD
1844 if (p->valid == 0 ||
1845 p->wire_count ||
1846 (p->flags & PG_NEED_COMMIT)) {
b12defdc
MD
1847 vm_page_wakeup(p);
1848 return(0);
1849 }
1850 } else {
1851 /* XXX what if p->valid == 0 , hold_count, etc? */
1852 }
984263bc 1853
1f804340
MD
1854 KASSERT(
1855 p->object == backing_object,
1856 ("vm_object_qcollapse(): object mismatch")
1857 );
1858
1859 /*
1860 * Destroy any associated swap
1861 */
8d292090
MD
1862 if (backing_object->type == OBJT_SWAP)
1863 swap_pager_freespace(backing_object, p->pindex, 1);
1f804340
MD
1864
1865 if (
1866 p->pindex < backing_offset_index ||
1867 new_pindex >= object->size
1868 ) {
984263bc 1869 /*
1f804340
MD
1870 * Page is out of the parent object's range, we
1871 * can simply destroy it.
984263bc 1872 */
1f804340
MD
1873 vm_page_protect(p, VM_PROT_NONE);
1874 vm_page_free(p);
1875 return(0);
1876 }
984263bc 1877
1f804340 1878 pp = vm_page_lookup(object, new_pindex);
1b9d3514 1879 if (pp != NULL || vm_pager_has_page(object, new_pindex)) {
1f804340
MD
1880 /*
1881 * page already exists in parent OR swap exists
1882 * for this location in the parent. Destroy
1883 * the original page from the backing object.
1884 *
1885 * Leave the parent's page alone
1886 */
1887 vm_page_protect(p, VM_PROT_NONE);
1888 vm_page_free(p);
1889 return(0);
984263bc 1890 }
1f804340
MD
1891
1892 /*
1893 * Page does not exist in parent, rename the
1894 * page from the backing object to the main object.
1895 *
1896 * If the page was mapped to a process, it can remain
1897 * mapped through the rename.
1898 */
1899 if ((p->queue - p->pc) == PQ_CACHE)
1900 vm_page_deactivate(p);
1901
1902 vm_page_rename(p, object, new_pindex);
b12defdc 1903 vm_page_wakeup(p);
1f804340 1904 /* page automatically made dirty by rename */
984263bc 1905 }
1f804340 1906 return(0);
984263bc
MD
1907}
1908
984263bc 1909/*
6846fd23 1910 * This version of collapse allows the operation to occur earlier and
984263bc
MD
1911 * when paging_in_progress is true for an object... This is not a complete
1912 * operation, but should plug 99.9% of the rest of the leaks.
6846fd23 1913 *
b12defdc
MD
1914 * The caller must hold the object and backing_object and both must be
1915 * chainlocked.
1916 *
2de4f77e 1917 * (only called from vm_object_collapse)
984263bc
MD
1918 */
1919static void
b12defdc 1920vm_object_qcollapse(vm_object_t object, vm_object_t backing_object)
984263bc 1921{
b12defdc
MD
1922 if (backing_object->ref_count == 1) {
1923 backing_object->ref_count += 2;
1924 vm_object_backing_scan(object, backing_object,
1925 OBSC_COLLAPSE_NOWAIT);
1926 backing_object->ref_count -= 2;
1927 }
984263bc
MD
1928}
1929
1930/*
6846fd23
MD
1931 * Collapse an object with the object backing it. Pages in the backing
1932 * object are moved into the parent, and the backing object is deallocated.
e806bedd 1933 * Any conflict is resolved in favor of the parent's existing pages.
212f39f5 1934 *
b12defdc
MD
1935 * object must be held and chain-locked on call.
1936 *
1937 * The caller must have an extra ref on object to prevent a race from
1938 * destroying it during the collapse.
984263bc
MD
1939 */
1940void
e806bedd 1941vm_object_collapse(vm_object_t object, struct vm_object_dealloc_list **dlistp)
984263bc 1942{
e806bedd 1943 struct vm_object_dealloc_list *dlist = NULL;
b12defdc
MD
1944 vm_object_t backing_object;
1945
1946 /*
1947 * Only one thread is attempting a collapse at any given moment.
1948 * There are few restrictions for (object) that callers of this
1949 * function check so reentrancy is likely.
1950 */
1951 KKASSERT(object != NULL);
212f39f5 1952 vm_object_assert_held(object);
b12defdc
MD
1953 KKASSERT(object->flags & OBJ_CHAINLOCK);
1954
1955 for (;;) {
1956 vm_object_t bbobj;
1957 int dodealloc;
6846fd23 1958
b12defdc
MD
1959 /*
1960 * We have to hold the backing object, check races.
1961 */
1962 while ((backing_object = object->backing_object) != NULL) {
1963 vm_object_hold(backing_object);
1964 if (backing_object == object->backing_object)
1965 break;
1966 vm_object_drop(backing_object);
1967 }
984263bc
MD
1968
1969 /*
b12defdc 1970 * No backing object? Nothing to collapse then.
984263bc 1971 */
b12defdc 1972 if (backing_object == NULL)
984263bc
MD
1973 break;
1974
b12defdc
MD
1975 /*
1976 * You can't collapse with a non-default/non-swap object.
1977 */
1978 if (backing_object->type != OBJT_DEFAULT &&
1979 backing_object->type != OBJT_SWAP) {
1980 vm_object_drop(backing_object);
1981 backing_object = NULL;
984263bc 1982 break;
b12defdc 1983 }
984263bc 1984
b12defdc
MD
1985 /*
1986 * Chain-lock the backing object too because if we
1987 * successfully merge its pages into the top object we
1988 * will collapse backing_object->backing_object as the
1989 * new backing_object. Re-check that it is still our
1990 * backing object.
1991 */
1992 vm_object_chain_acquire(backing_object);
212f39f5 1993 if (backing_object != object->backing_object) {
b12defdc 1994 vm_object_chain_release(backing_object);
212f39f5
MD
1995 vm_object_drop(backing_object);
1996 continue;
1997 }
1998
984263bc
MD
1999 /*
2000 * we check the backing object first, because it is most likely
2001 * not collapsable.
2002 */
2003 if (backing_object->handle != NULL ||
2004 (backing_object->type != OBJT_DEFAULT &&
2005 backing_object->type != OBJT_SWAP) ||
2006 (backing_object->flags & OBJ_DEAD) ||
2007 object->handle != NULL ||
2008 (object->type != OBJT_DEFAULT &&
2009 object->type != OBJT_SWAP) ||
2010 (object->flags & OBJ_DEAD)) {
2011 break;
2012 }
2013
b12defdc
MD
2014 /*
2015 * If paging is in progress we can't do a normal collapse.
2016 */
984263bc
MD
2017 if (
2018 object->paging_in_progress != 0 ||
2019 backing_object->paging_in_progress != 0
2020 ) {
b12defdc 2021 vm_object_qcollapse(object, backing_object);
984263bc
MD
2022 break;
2023 }
2024
2025 /*
2026 * We know that we can either collapse the backing object (if
2027 * the parent is the only reference to it) or (perhaps) have
2028 * the parent bypass the object if the parent happens to shadow
2029 * all the resident pages in the entire backing object.
2030 *
2031 * This is ignoring pager-backed pages such as swap pages.
2032 * vm_object_backing_scan fails the shadowing test in this
2033 * case.
2034 */
984263bc
MD
2035 if (backing_object->ref_count == 1) {
2036 /*
2037 * If there is exactly one reference to the backing
2038 * object, we can collapse it into the parent.
2039 */
b12defdc
MD
2040 KKASSERT(object->backing_object == backing_object);
2041 vm_object_backing_scan(object, backing_object,
2042 OBSC_COLLAPSE_WAIT);
984263bc
MD
2043
2044 /*
2045 * Move the pager from backing_object to object.
2046 */
984263bc
MD
2047 if (backing_object->type == OBJT_SWAP) {
2048 vm_object_pip_add(backing_object, 1);
2049
2050 /*
2051 * scrap the paging_offset junk and do a
2052 * discrete copy. This also removes major
2053 * assumptions about how the swap-pager
2054 * works from where it doesn't belong. The
2055 * new swapper is able to optimize the
2056 * destroy-source case.
2057 */
984263bc 2058 vm_object_pip_add(object, 1);
b12defdc
MD
2059 swap_pager_copy(backing_object, object,
2060 OFF_TO_IDX(object->backing_object_offset),
2061 TRUE);
984263bc 2062 vm_object_pip_wakeup(object);
984263bc
MD
2063 vm_object_pip_wakeup(backing_object);
2064 }
b12defdc 2065
984263bc
MD
2066 /*
2067 * Object now shadows whatever backing_object did.
b12defdc 2068 * Remove object from backing_object's shadow_list.
984263bc 2069 */
984263bc 2070 LIST_REMOVE(object, shadow_list);
b12defdc
MD
2071 KKASSERT(object->backing_object == backing_object);
2072 backing_object->shadow_count--;
2073 backing_object->generation++;
2074
e806bedd
MD
2075 /*
2076 * backing_object->backing_object moves from within
2077 * backing_object to within object.
2078 */
b12defdc
MD
2079 while ((bbobj = backing_object->backing_object) != NULL) {
2080 vm_object_hold(bbobj);
2081 if (bbobj == backing_object->backing_object)
2082 break;
2083 vm_object_drop(bbobj);
2084 }
2085 if (bbobj) {
984263bc 2086 LIST_REMOVE(backing_object, shadow_list);
b12defdc
MD
2087 bbobj->shadow_count--;
2088 bbobj->generation++;
e806bedd 2089 backing_object->backing_object = NULL;
984263bc 2090 }
b12defdc
MD
2091 object->backing_object = bbobj;
2092 if (bbobj) {
2093 LIST_INSERT_HEAD(&bbobj->shadow_head,
2094 object, shadow_list);
2095 bbobj->shadow_count++;
2096 bbobj->generation++;
984263bc
MD
2097 }
2098
2099 object->backing_object_offset +=
b12defdc
MD
2100 backing_object->backing_object_offset;
2101
2102 vm_object_drop(bbobj);
984263bc
MD
2103
2104 /*
b12defdc
MD
2105 * Discard the old backing_object. Nothing should be
2106 * able to ref it, other than a vm_map_split(),
2107 * and vm_map_split() will stall on our chain lock.
2108 * And we control the parent so it shouldn't be
2109 * possible for it to go away either.
984263bc 2110 *
b12defdc
MD
2111 * Since the backing object has no pages, no pager
2112 * left, and no object references within it, all
2113 * that is necessary is to dispose of it.
984263bc 2114 */
2de4f77e
MD
2115 KASSERT(backing_object->ref_count == 1,
2116 ("backing_object %p was somehow "
2117 "re-referenced during collapse!",
2118 backing_object));
2119 KASSERT(RB_EMPTY(&backing_object->rb_memq),
2120 ("backing_object %p somehow has left "
2121 "over pages during collapse!",
2122 backing_object));
2123
e1c14c82 2124 /*
b12defdc
MD
2125 * The object can be destroyed.
2126 *
2127 * XXX just fall through and dodealloc instead
2128 * of forcing destruction?
e1c14c82 2129 */
b12defdc
MD
2130 --backing_object->ref_count;
2131 if ((backing_object->flags & OBJ_DEAD) == 0)
2132 vm_object_terminate(backing_object);
984263bc 2133 object_collapses++;
b12defdc 2134 dodealloc = 0;
984263bc 2135 } else {
984263bc
MD
2136 /*
2137 * If we do not entirely shadow the backing object,
2138 * there is nothing we can do so we give up.
2139 */
b12defdc
MD
2140 if (vm_object_backing_scan(object, backing_object,
2141 OBSC_TEST_ALL_SHADOWED) == 0) {
984263bc
MD
2142 break;
2143 }
2144
e806bedd
MD
2145 /*
2146 * bbobj is backing_object->backing_object. Since
2147 * object completely shadows backing_object we can
2148 * bypass it and become backed by bbobj instead.
2149 */
b12defdc
MD
2150 while ((bbobj = backing_object->backing_object) != NULL) {
2151 vm_object_hold(bbobj);
2152 if (bbobj == backing_object->backing_object)
2153 break;
2154 vm_object_drop(bbobj);
2155 }
e806bedd 2156
984263bc 2157 /*
e806bedd
MD
2158 * Make object shadow bbobj instead of backing_object.
2159 * Remove object from backing_object's shadow list.
b12defdc
MD
2160 *
2161 * Deallocating backing_object will not remove
984263bc
MD
2162 * it, since its reference count is at least 2.
2163 */
b12defdc 2164 KKASSERT(object->backing_object == backing_object);
984263bc
MD
2165 LIST_REMOVE(object, shadow_list);
2166 backing_object->shadow_count--;
2167 backing_object->generation++;
2168
b12defdc 2169 /*
e806bedd
MD
2170 * Add a ref to bbobj, bbobj now shadows object.
2171 *
2172 * NOTE: backing_object->backing_object still points
2173 * to bbobj. That relationship remains intact
2174 * because backing_object has > 1 ref, so
2175 * someone else is pointing to it (hence why
2176 * we can't collapse it into object and can
2177 * only handle the all-shadowed bypass case).
b12defdc
MD
2178 */
2179 if (bbobj) {
2180 vm_object_chain_wait(bbobj);
2181 vm_object_reference_locked(bbobj);
2182 LIST_INSERT_HEAD(&bbobj->shadow_head,
2183 object, shadow_list);
2184 bbobj->shadow_count++;
2185 bbobj->generation++;
984263bc
MD
2186 object->backing_object_offset +=
2187 backing_object->backing_object_offset;
b12defdc
MD
2188 object->backing_object = bbobj;
2189 vm_object_drop(bbobj);
2190 } else {
2191 object->backing_object = NULL;
984263bc
MD
2192 }
2193
2194 /*
b12defdc
MD
2195 * Drop the reference count on backing_object. To
2196 * handle ref_count races properly we can't assume
2197 * that the ref_count is still at least 2 so we
2198 * have to actually call vm_object_deallocate()
2199 * (after clearing the chainlock).
984263bc 2200 */
984263bc 2201 object_bypasses++;
b12defdc 2202 dodealloc = 1;
984263bc
MD
2203 }
2204
2205 /*
e806bedd
MD
2206 * Ok, we want to loop on the new object->bbobj association,
2207 * possibly collapsing it further. However if dodealloc is
2208 * non-zero we have to deallocate the backing_object which
2209 * itself can potentially undergo a collapse, creating a
2210 * recursion depth issue with the LWKT token subsystem.
2211 *
2212 * In the case where we must deallocate the backing_object
2213 * it is possible now that the backing_object has a single
2214 * shadow count on some other object (not represented here
2215 * as yet), since it no longer shadows us. Thus when we
2216 * call vm_object_deallocate() it may attempt to collapse
2217 * itself into its remaining parent.
984263bc 2218 */
e806bedd
MD
2219 if (dodealloc) {
2220 struct vm_object_dealloc_list *dtmp;
b12defdc 2221
e806bedd
MD
2222 vm_object_chain_release(backing_object);
2223 vm_object_unlock(backing_object);
2224 /* backing_object remains held */
2225
2226 /*
2227 * Auto-deallocation list for caller convenience.
2228 */
2229 if (dlistp == NULL)
2230 dlistp = &dlist;
2231
2232 dtmp = kmalloc(sizeof(*dtmp), M_TEMP, M_WAITOK);
2233 dtmp->object = backing_object;
2234 dtmp->next = *dlistp;
2235 *dlistp = dtmp;
2236 } else {
2237 vm_object_chain_release(backing_object);
2238 vm_object_drop(backing_object);
2239 }
2240 /* backing_object = NULL; not needed */
b12defdc
MD
2241 /* loop */
2242 }
2243
2244 /*
b59b553b
MD
2245 * Clean up any left over backing_object
2246 */
2247 if (backing_object) {
2248 vm_object_chain_release(backing_object);
2249 vm_object_drop(backing_object);
2250 }
2251
2252 /*
e806bedd
MD
2253 * Clean up any auto-deallocation list. This is a convenience
2254 * for top-level callers so they don't have to pass &dlist.
2255 * Do not clean up any caller-passed dlistp, the caller will
2256 * do that.
2257 */
2258 if (dlist)
2259 vm_object_deallocate_list(&dlist);
2260
984263bc
MD
2261}
2262
2263/*
e806bedd
MD
2264 * vm_object_collapse() may collect additional objects in need of
2265 * deallocation. This routine deallocates these objects. The
2266 * deallocation itself can trigger additional collapses (which the
2267 * deallocate function takes care of). This procedure is used to
2268 * reduce procedural recursion since these vm_object shadow chains
2269 * can become quite long.
2270 */
2271void
2272vm_object_deallocate_list(struct vm_object_dealloc_list **dlistp)
2273{
2274 struct vm_object_dealloc_list *dlist;
2275
2276 while ((dlist = *dlistp) != NULL) {
2277 *dlistp = dlist->next;
2278 vm_object_lock(dlist->object);
2279 vm_object_deallocate_locked(dlist->object);
2280 vm_object_drop(dlist->object);
2281 kfree(dlist, M_TEMP);
2282 }
2283}
2284
2285/*
6846fd23
MD
2286 * Removes all physical pages in the specified object range from the
2287 * object's list of pages.
984263bc 2288 *
6846fd23 2289 * No requirements.
984263bc 2290 */
1f804340
MD
2291static int vm_object_page_remove_callback(vm_page_t p, void *data);
2292
984263bc 2293void
57e43348 2294vm_object_page_remove(vm_object_t object, vm_pindex_t start, vm_pindex_t end,
1f804340 2295 boolean_t clean_only)
984263bc 2296{
1f804340 2297 struct rb_vm_page_scan_info info;
984263bc
MD
2298 int all;
2299
1f804340
MD
2300 /*
2301 * Degenerate cases and assertions
2302 */
b12defdc 2303 vm_object_hold(object);
b8158020
MD
2304 if (object == NULL ||
2305 (object->resident_page_count == 0 && object->swblock_count == 0)) {
b12defdc 2306 vm_object_drop(object);
984263bc 2307 return;
b8158020 2308 }
1f804340
MD
2309 KASSERT(object->type != OBJT_PHYS,
2310 ("attempt to remove pages from a physical object"));
984263bc 2311
1f804340
MD
2312 /*
2313 * Indicate that paging is occuring on the object
2314 */
1f804340 2315 vm_object_pip_add(object, 1);
984263bc
MD
2316
2317 /*
1f804340
MD
2318 * Figure out the actual removal range and whether we are removing
2319 * the entire contents of the object or not. If removing the entire
2320 * contents, be sure to get all pages, even those that might be
2321 * beyond the end of the object.
984263bc 2322 */
1f804340
MD
2323 info.start_pindex = start;
2324 if (end == 0)
2325 info.end_pindex = (vm_pindex_t)-1;
2326 else
2327 info.end_pindex = end - 1;
2328 info.limit = clean_only;
2329 all = (start == 0 && info.end_pindex >= object->size - 1);
984263bc 2330
06ecca5a 2331 /*
1f804340 2332 * Loop until we are sure we have gotten them all.
06ecca5a 2333 */
1f804340
MD
2334 do {
2335 info.error = 0;
2336 vm_page_rb_tree_RB_SCAN(&object->rb_memq, rb_vm_page_scancmp,
2337 vm_object_page_remove_callback, &info);
2338 } while (info.error);
984263bc 2339
1f804340 2340 /*
5ac04117
MD
2341 * Remove any related swap if throwing away pages, or for
2342 * non-swap objects (the swap is a clean copy in that case).
8d292090 2343 */
5ac04117 2344 if (object->type != OBJT_SWAP || clean_only == FALSE) {
8d292090
MD
2345 if (all)
2346 swap_pager_freespace_all(object);
2347 else
2348 swap_pager_freespace(object, info.start_pindex,
2349 info.end_pindex - info.start_pindex + 1);
2350 }
2351
2352 /*
1f804340
MD
2353 * Cleanup
2354 */
2355 vm_object_pip_wakeup(object);
b12defdc 2356 vm_object_drop(object);
1f804340 2357}
984263bc 2358
6846fd23 2359/*
b12defdc 2360 * The caller must hold the object
6846fd23 2361 */
1f804340
MD
2362static int
2363vm_object_page_remove_callback(vm_page_t p, void *data)
2364{
2365 struct rb_vm_page_scan_info *info = data;
984263bc 2366
b12defdc
MD
2367 if (vm_page_busy_try(p, TRUE)) {
2368 vm_page_sleep_busy(p, TRUE, "vmopar");
2369 info->error = 1;
2370 return(0);
2371 }
2372
1f804340
MD
2373 /*
2374 * Wired pages cannot be destroyed, but they can be invalidated
2375 * and we do so if clean_only (limit) is not set.
1c9602b3
MD
2376 *
2377 * WARNING! The page may be wired due to being part of a buffer
2378 * cache buffer, and the buffer might be marked B_CACHE.
2379 * This is fine as part of a truncation but VFSs must be
2380 * sure to fix the buffer up when re-extending the file.
9bf025db
MD
2381 *
2382 * NOTE! PG_NEED_COMMIT is ignored.
1f804340
MD
2383 */
2384 if (p->wire_count != 0) {
2385 vm_page_protect(p, VM_PROT_NONE);
2386 if (info->limit == 0)
2387 p->valid = 0;
b12defdc 2388 vm_page_wakeup(p);
1f804340
MD
2389 return(0);
2390 }
984263bc 2391
1f804340
MD
2392 /*
2393 * limit is our clean_only flag. If set and the page is dirty, do
d309a59d 2394 * not free it. If set and the page is being held by someone, do
1f804340
MD
2395 * not free it.
2396 */
2397 if (info->limit && p->valid) {
2398 vm_page_test_dirty(p);
b12defdc
MD
2399 if (p->valid & p->dirty) {
2400 vm_page_wakeup(p);
1f804340 2401 return(0);
b12defdc
MD
2402 }
2403#if 0
2404 if (p->hold_count) {
2405 vm_page_wakeup(p);
d309a59d 2406 return(0);
b12defdc
MD
2407 }
2408#endif
984263bc 2409 }
1f804340
MD
2410
2411 /*
2412 * Destroy the page
2413 */
1f804340
MD
2414 vm_page_protect(p, VM_PROT_NONE);
2415 vm_page_free(p);
2416 return(0);
984263bc
MD
2417}
2418
2419/*
6846fd23
MD
2420 * Coalesces two objects backing up adjoining regions of memory into a
2421 * single object.
984263bc 2422 *
6846fd23 2423 * returns TRUE if objects were combined.
984263bc 2424 *
6846fd23
MD
2425 * NOTE: Only works at the moment if the second object is NULL -
2426 * if it's not, which object do we lock first?
984263bc 2427 *
6846fd23
MD
2428 * Parameters:
2429 * prev_object First object to coalesce
2430 * prev_offset Offset into prev_object
2431 * next_object Second object into coalesce
2432 * next_offset Offset into next_object
984263bc 2433 *
6846fd23
MD
2434 * prev_size Size of reference to prev_object
2435 * next_size Size of reference to next_object
984263bc 2436 *
212f39f5
MD
2437 * The caller does not need to hold (prev_object) but must have a stable
2438 * pointer to it (typically by holding the vm_map locked).
984263bc
MD
2439 */
2440boolean_t
57e43348 2441vm_object_coalesce(vm_object_t prev_object, vm_pindex_t prev_pindex,
8d292090 2442 vm_size_t prev_size, vm_size_t next_size)
984263bc
MD
2443{
2444 vm_pindex_t next_pindex;
2445
b12defdc 2446 if (prev_object == NULL)
984263bc 2447 return (TRUE);
984263bc 2448
212f39f5
MD
2449 vm_object_hold(prev_object);
2450
984263bc
MD
2451 if (prev_object->type != OBJT_DEFAULT &&
2452 prev_object->type != OBJT_SWAP) {
212f39f5 2453 vm_object_drop(prev_object);
984263bc
MD
2454 return (FALSE);
2455 }
2456
2457 /*
2458 * Try to collapse the object first
2459 */
b12defdc 2460 vm_object_chain_acquire(prev_object);
e806bedd 2461 vm_object_collapse(prev_object, NULL);
984263bc
MD
2462
2463 /*
2464 * Can't coalesce if: . more than one reference . paged out . shadows
2465 * another object . has a copy elsewhere (any of which mean that the
2466 * pages not mapped to prev_entry may be in use anyway)
2467 */
2468
70de66db 2469 if (prev_object->backing_object != NULL) {
b12defdc 2470 vm_object_chain_release(prev_object);
212f39f5 2471 vm_object_drop(prev_object);
984263bc 2472 return (FALSE);
70de66db 2473 }
984263bc
MD
2474
2475 prev_size >>= PAGE_SHIFT;
2476 next_size >>= PAGE_SHIFT;
2477 next_pindex = prev_pindex + prev_size;
2478
2479 if ((prev_object->ref_count > 1) &&
2480 (prev_object->size != next_pindex)) {
b12defdc 2481 vm_object_chain_release(prev_object);
212f39f5 2482 vm_object_drop(prev_object);
984263bc
MD
2483 return (FALSE);
2484 }
2485
2486 /*
2487 * Remove any pages that may still be in the object from a previous
2488 * deallocation.
2489 */
2490 if (next_pindex < prev_object->size) {
2491 vm_object_page_remove(prev_object,
2492 next_pindex,
2493 next_pindex + next_size, FALSE);
2494 if (prev_object->type == OBJT_SWAP)
2495 swap_pager_freespace(prev_object,
2496 next_pindex, next_size);
2497 }
2498
2499 /*
2500 * Extend the object if necessary.
2501 */
2502 if (next_pindex + next_size > prev_object->size)
2503 prev_object->size = next_pindex + next_size;
70de66db 2504
b12defdc 2505 vm_object_chain_release(prev_object);
212f39f5 2506 vm_object_drop(prev_object);
984263bc
MD
2507 return (TRUE);
2508}
2509
6846fd23
MD
2510/*
2511 * Make the object writable and flag is being possibly dirty.
2512 *
b12defdc
MD
2513 * The caller must hold the object. XXX called from vm_page_dirty(),
2514 * There is currently no requirement to hold the object.
6846fd23 2515 */
984263bc
MD
2516void
2517vm_object_set_writeable_dirty(vm_object_t object)
2518{
2519 struct vnode *vp;
2520
b12defdc 2521 /*vm_object_assert_held(object);*/
54341a3b
MD
2522 /*
2523 * Avoid contention in vm fault path by checking the state before
2524 * issuing an atomic op on it.
2525 */
2526 if ((object->flags & (OBJ_WRITEABLE|OBJ_MIGHTBEDIRTY)) !=
2527 (OBJ_WRITEABLE|OBJ_MIGHTBEDIRTY)) {
2528 vm_object_set_flag(object, OBJ_WRITEABLE|OBJ_MIGHTBEDIRTY);
2529 }
984263bc
MD
2530 if (object->type == OBJT_VNODE &&
2531 (vp = (struct vnode *)object->handle) != NULL) {
2532 if ((vp->v_flag & VOBJDIRTY) == 0) {
5fd012e0 2533 vsetflags(vp, VOBJDIRTY);
984263bc
MD
2534 }
2535 }
2536}
2537
984263bc
MD
2538#include "opt_ddb.h"
2539#ifdef DDB
2540#include <sys/kernel.h>
2541
2542#include <sys/cons.h>
2543
2544#include <ddb/ddb.h>
2545
1388df65
RG
2546static int _vm_object_in_map (vm_map_t map, vm_object_t object,
2547 vm_map_entry_t entry);
2548static int vm_object_in_map (vm_object_t object);
984263bc 2549
6846fd23 2550/*
b12defdc 2551 * The caller must hold the object.
6846fd23 2552 */
984263bc 2553static int
57e43348 2554_vm_object_in_map(vm_map_t map, vm_object_t object, vm_map_entry_t entry)
984263bc
MD
2555{
2556 vm_map_t tmpm;
2557 vm_map_entry_t tmpe;
b12defdc 2558 vm_object_t obj, nobj;
984263bc
MD
2559 int entcount;
2560
2561 if (map == 0)
2562 return 0;
984263bc
MD
2563 if (entry == 0) {
2564 tmpe = map->header.next;
2565 entcount = map->nentries;
2566 while (entcount-- && (tmpe != &map->header)) {
2567 if( _vm_object_in_map(map, object, tmpe)) {
2568 return 1;
2569 }
2570 tmpe = tmpe->next;
2571 }
1b874851
MD
2572 return (0);
2573 }
2574 switch(entry->maptype) {
2575 case VM_MAPTYPE_SUBMAP:
984263bc
MD
2576 tmpm = entry->object.sub_map;
2577 tmpe = tmpm->header.next;
2578 entcount = tmpm->nentries;
2579 while (entcount-- && tmpe != &tmpm->header) {
2580 if( _vm_object_in_map(tmpm, object, tmpe)) {
2581 return 1;
2582 }
2583 tmpe = tmpe->next;
2584 }
1b874851
MD
2585 break;
2586 case VM_MAPTYPE_NORMAL:
2587 case VM_MAPTYPE_VPAGETABLE:
2588 obj = entry->object.vm_object;
2589 while (obj) {
b12defdc
MD
2590 if (obj == object) {
2591 if (obj != entry->object.vm_object)
2592 vm_object_drop(obj);
984263bc 2593 return 1;
b12defdc
MD
2594 }
2595 while ((nobj = obj->backing_object) != NULL) {
2596 vm_object_hold(nobj);
2597 if (nobj == obj->backing_object)
2598 break;
2599 vm_object_drop(nobj);
2600 }
2601 if (obj != entry->object.vm_object) {
2602 if (nobj)
2603 vm_object_lock_swap();
2604 vm_object_drop(obj);
2605 }
2606 obj = nobj;
1b874851
MD
2607 }
2608 break;
2609 default:
2610 break;
984263bc
MD
2611 }
2612 return 0;
2613}
2614
8fa76237
MD
2615static int vm_object_in_map_callback(struct proc *p, void *data);
2616
2617struct vm_object_in_map_info {
2618 vm_object_t object;
2619 int rv;
2620};
2621
6846fd23
MD
2622/*
2623 * Debugging only
2624 */
984263bc 2625static int
57e43348 2626vm_object_in_map(vm_object_t object)
984263bc 2627{
8fa76237
MD
2628 struct vm_object_in_map_info info;
2629
2630 info.rv = 0;
2631 info.object = object;
2632
2633 allproc_scan(vm_object_in_map_callback, &info);
2634 if (info.rv)
2635 return 1;
e4846942 2636 if( _vm_object_in_map(&kernel_map, object, 0))
984263bc 2637 return 1;
e4846942 2638 if( _vm_object_in_map(&pager_map, object, 0))
984263bc 2639 return 1;
e4846942 2640 if( _vm_object_in_map(&buffer_map, object, 0))
984263bc 2641 return 1;
984263bc
MD
2642 return 0;
2643}
2644
6846fd23
MD
2645/*
2646 * Debugging only
2647 */
8fa76237
MD
2648static int
2649vm_object_in_map_callback(struct proc *p, void *data)
2650{
2651 struct vm_object_in_map_info *info = data;
2652
2653 if (p->p_vmspace) {
2654 if (_vm_object_in_map(&p->p_vmspace->vm_map, info->object, 0)) {
2655 info->rv = 1;
2656 return -1;
2657 }
2658 }
2659 return (0);
2660}
2661
984263bc
MD
2662DB_SHOW_COMMAND(vmochk, vm_object_check)
2663{
2664 vm_object_t object;
2665
2666 /*
2667 * make sure that internal objs are in a map somewhere
2668 * and none have zero ref counts.
2669 */
2670 for (object = TAILQ_FIRST(&vm_object_list);
2671 object != NULL;
2672 object = TAILQ_NEXT(object, object_list)) {
00a3fdca
MD
2673 if (object->type == OBJT_MARKER)
2674 continue;
984263bc
MD
2675 if (object->handle == NULL &&
2676 (object->type == OBJT_DEFAULT || object->type == OBJT_SWAP)) {
2677 if (object->ref_count == 0) {
2678 db_printf("vmochk: internal obj has zero ref count: %ld\n",
2679 (long)object->size);
2680 }
2681 if (!vm_object_in_map(object)) {
2682 db_printf(
2683 "vmochk: internal obj is not in a map: "
2684 "ref: %d, size: %lu: 0x%lx, backing_object: %p\n",
2685 object->ref_count, (u_long)object->size,
2686 (u_long)object->size,
2687 (void *)object->backing_object);
2688 }
2689 }
2690 }
2691}
2692
2693/*
6846fd23 2694 * Debugging only
984263bc
MD
2695 */
2696DB_SHOW_COMMAND(object, vm_object_print_static)
2697{
2698 /* XXX convert args. */
2699 vm_object_t object = (vm_object_t)addr;
2700 boolean_t full = have_addr;
2701
2702 vm_page_t p;
2703
2704 /* XXX count is an (unused) arg. Avoid shadowing it. */
2705#define count was_count
2706
2707 int count;
2708
2709 if (object == NULL)
2710 return;
2711
2712 db_iprintf(
2713 "Object %p: type=%d, size=0x%lx, res=%d, ref=%d, flags=0x%x\n",
2714 object, (int)object->type, (u_long)object->size,
2715 object->resident_page_count, object->ref_count, object->flags);
2716 /*
2717 * XXX no %qd in kernel. Truncate object->backing_object_offset.
2718 */
2719 db_iprintf(" sref=%d, backing_object(%d)=(%p)+0x%lx\n",
2720 object->shadow_count,
2721 object->backing_object ? object->backing_object->ref_count : 0,
2722 object->backing_object, (long)object->backing_object_offset);
2723
2724 if (!full)
2725 return;
2726
2727 db_indent += 2;
2728 count = 0;
1f804340 2729 RB_FOREACH(p, vm_page_rb_tree, &object->rb_memq) {
984263bc
MD
2730 if (count == 0)
2731 db_iprintf("memory:=");
2732 else if (count == 6) {
2733 db_printf("\n");
2734 db_iprintf(" ...");
2735 count = 0;
2736 } else
2737 db_printf(",");
2738 count++;
2739
2740 db_printf("(off=0x%lx,page=0x%lx)",
2741 (u_long) p->pindex, (u_long) VM_PAGE_TO_PHYS(p));
2742 }
2743 if (count != 0)
2744 db_printf("\n");
2745 db_indent -= 2;
2746}
2747
2748/* XXX. */
2749#undef count
2750
6846fd23
MD
2751/*
2752 * XXX need this non-static entry for calling from vm_map_print.
2753 *
2754 * Debugging only
2755 */
984263bc 2756void
57e43348
MD
2757vm_object_print(/* db_expr_t */ long addr,
2758 boolean_t have_addr,
2759 /* db_expr_t */ long count,
2760 char *modif)
984263bc
MD
2761{
2762 vm_object_print_static(addr, have_addr, count, modif);
2763}
2764
6846fd23
MD
2765/*
2766 * Debugging only
2767 */
984263bc
MD
2768DB_SHOW_COMMAND(vmopag, vm_object_print_pages)
2769{
2770 vm_object_t object;
2771 int nl = 0;
2772 int c;
2773 for (object = TAILQ_FIRST(&vm_object_list);
2774 object != NULL;
2775 object = TAILQ_NEXT(object, object_list)) {
2776 vm_pindex_t idx, fidx;
2777 vm_pindex_t osize;
6ef943a3 2778 vm_paddr_t pa = -1, padiff;
984263bc
MD
2779 int rcount;
2780 vm_page_t m;
2781
00a3fdca
MD
2782 if (object->type == OBJT_MARKER)
2783 continue;
984263bc
MD
2784 db_printf("new object: %p\n", (void *)object);
2785 if ( nl > 18) {
2786 c = cngetc();
2787 if (c != ' ')
2788 return;
2789 nl = 0;
2790 }
2791 nl++;
2792 rcount = 0;
2793 fidx = 0;
2794 osize = object->size;
2795 if (osize > 128)
2796 osize = 128;
06ecca5a 2797 for (idx = 0; idx < osize; idx++) {
984263bc
MD
2798 m = vm_page_lookup(object, idx);
2799 if (m == NULL) {
2800 if (rcount) {
2801 db_printf(" index(%ld)run(%d)pa(0x%lx)\n",
2802 (long)fidx, rcount, (long)pa);
2803 if ( nl > 18) {
2804 c = cngetc();
2805 if (c != ' ')
2806 return;
2807 nl = 0;
2808 }
2809 nl++;
2810 rcount = 0;
2811 }
2812 continue;
2813 }
2814
2815
2816 if (rcount &&
2817 (VM_PAGE_TO_PHYS(m) == pa + rcount * PAGE_SIZE)) {
2818 ++rcount;
2819 continue;
2820 }
2821 if (rcount) {
2822 padiff = pa + rcount * PAGE_SIZE - VM_PAGE_TO_PHYS(m);
2823 padiff >>= PAGE_SHIFT;
2824 padiff &= PQ_L2_MASK;
2825 if (padiff == 0) {
2826 pa = VM_PAGE_TO_PHYS(m) - rcount * PAGE_SIZE;
2827 ++rcount;
2828 continue;
2829 }
2830 db_printf(" index(%ld)run(%d)pa(0x%lx)",
2831 (long)fidx, rcount, (long)pa);
2832 db_printf("pd(%ld)\n", (long)padiff);
2833 if ( nl > 18) {
2834 c = cngetc();
2835 if (c != ' ')
2836 return;
2837 nl = 0;
2838 }
2839 nl++;
2840 }
2841 fidx = idx;
2842 pa = VM_PAGE_TO_PHYS(m);
2843 rcount = 1;
2844 }
2845 if (rcount) {
2846 db_printf(" index(%ld)run(%d)pa(0x%lx)\n",
2847 (long)fidx, rcount, (long)pa);
2848 if ( nl > 18) {
2849 c = cngetc();
2850 if (c != ' ')
2851 return;
2852 nl = 0;
2853 }
2854 nl++;
2855 }
2856 }
2857}
2858#endif /* DDB */