kernel - Adjust vm_object->paging_in_progress to use refcount API
[dragonfly.git] / sys / vm / vm_object.c
1 /*
2  * (MPSAFE)
3  *
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 */
76 #include <sys/thread.h>
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>
83 #include <sys/refcount.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_zone.h>
97
98 #define EASY_SCAN_FACTOR        8
99
100 static void     vm_object_qcollapse(vm_object_t object);
101 static int      vm_object_page_collect_flush(vm_object_t object, vm_page_t p,
102                                              int pagerflags);
103 static void     vm_object_lock_init(vm_object_t);
104 static void     vm_object_hold_wait(vm_object_t);
105
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
133 struct object_q vm_object_list;         /* locked by vmobj_token */
134 struct vm_object kernel_object;
135
136 static long vm_object_count;            /* locked by vmobj_token */
137 extern int vm_pageout_page_count;
138
139 static long object_collapses;
140 static long object_bypasses;
141 static int next_index;
142 static vm_zone_t obj_zone;
143 static struct vm_zone obj_zone_store;
144 #define VM_OBJECTS_INIT 256
145 static struct vm_object vm_objects_init[VM_OBJECTS_INIT];
146
147 /*
148  * Misc low level routines
149  */
150 static void
151 vm_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;
160                 obj->debug_hold_file[i] = NULL;
161                 obj->debug_hold_line[i] = 0;
162         }
163 #endif
164 }
165
166 static void
167 vm_object_lock_swap(void)
168 {
169         lwkt_token_swap();
170 }
171
172 void
173 vm_object_lock(vm_object_t obj)
174 {
175         lwkt_getpooltoken(obj);
176 }
177
178 void
179 vm_object_unlock(vm_object_t obj)
180 {
181         lwkt_relpooltoken(obj);
182 }
183
184 static __inline void
185 vm_object_assert_held(vm_object_t obj)
186 {
187         ASSERT_LWKT_TOKEN_HELD(lwkt_token_pool_lookup(obj));
188 }
189
190 void
191 #ifndef DEBUG_LOCKS
192 vm_object_hold(vm_object_t obj)
193 #else
194 debugvm_object_hold(vm_object_t obj, char *file, int line)
195 #endif
196 {
197         if (obj == NULL)
198                 return;
199
200         /*
201          * Object must be held (object allocation is stable due to callers
202          * context, typically already holding the token on a parent object)
203          * prior to potentially blocking on the lock, otherwise the object
204          * can get ripped away from us.
205          */
206         refcount_acquire(&obj->hold_count);
207         vm_object_lock(obj);
208
209 #if defined(DEBUG_LOCKS)
210         int i;
211
212         i = ffs(~obj->debug_hold_bitmap) - 1;
213         if (i == -1) {
214                 kprintf("vm_object hold count > VMOBJ_DEBUG_ARRAY_SIZE");
215                 obj->debug_hold_ovfl = 1;
216         }
217
218         obj->debug_hold_bitmap |= (1 << i);
219         obj->debug_hold_thrs[i] = curthread;
220         obj->debug_hold_file[i] = file;
221         obj->debug_hold_line[i] = line;
222 #endif
223 }
224
225 void
226 vm_object_drop(vm_object_t obj)
227 {
228         if (obj == NULL)
229                 return;
230
231 #if defined(DEBUG_LOCKS)
232         int found = 0;
233         int i;
234
235         for (i = 0; i < VMOBJ_DEBUG_ARRAY_SIZE; i++) {
236                 if ((obj->debug_hold_bitmap & (1 << i)) &&
237                     (obj->debug_hold_thrs[i] == curthread)) {
238                         obj->debug_hold_bitmap &= ~(1 << i);
239                         obj->debug_hold_thrs[i] = NULL;
240                         obj->debug_hold_file[i] = NULL;
241                         obj->debug_hold_line[i] = 0;
242                         found = 1;
243                         break;
244                 }
245         }
246
247         if (found == 0 && obj->debug_hold_ovfl == 0)
248                 panic("vm_object: attempt to drop hold on non-self-held obj");
249 #endif
250
251         /*
252          * The lock is a pool token, keep holding it across potential
253          * wakeups to interlock the tsleep/wakeup.
254          */
255         if (refcount_release(&obj->hold_count))
256                 wakeup(obj);
257         vm_object_unlock(obj);
258 }
259
260 /*
261  * This can only be called while the caller holds the object
262  * with the OBJ_DEAD interlock.  Since there are no refs this
263  * is the only thing preventing an object destruction race.
264  */
265 static void
266 vm_object_hold_wait(vm_object_t obj)
267 {
268         vm_object_lock(obj);
269
270 #if defined(DEBUG_LOCKS)
271         int i;
272
273         for (i = 0; i < VMOBJ_DEBUG_ARRAY_SIZE; i++) {
274                 if ((obj->debug_hold_bitmap & (1 << i)) &&
275                     (obj->debug_hold_thrs[i] == curthread))  {
276                         kprintf("vm_object %p: self-hold in at %s:%d\n", obj,
277                                 obj->debug_hold_file[i], obj->debug_hold_line[i]);
278                         panic("vm_object: self-hold in terminate or collapse");
279                 }
280         }
281 #endif
282
283         while (obj->hold_count)
284                 tsleep(obj, 0, "vmobjhld", 0);
285
286         vm_object_unlock(obj);
287 }
288
289
290 /*
291  * Initialize a freshly allocated object
292  *
293  * Used only by vm_object_allocate() and zinitna().
294  *
295  * No requirements.
296  */
297 void
298 _vm_object_allocate(objtype_t type, vm_pindex_t size, vm_object_t object)
299 {
300         int incr;
301
302         RB_INIT(&object->rb_memq);
303         LIST_INIT(&object->shadow_head);
304
305         object->type = type;
306         object->size = size;
307         object->ref_count = 1;
308         object->hold_count = 0;
309         object->flags = 0;
310         if ((object->type == OBJT_DEFAULT) || (object->type == OBJT_SWAP))
311                 vm_object_set_flag(object, OBJ_ONEMAPPING);
312         object->paging_in_progress = 0;
313         object->resident_page_count = 0;
314         object->agg_pv_list_count = 0;
315         object->shadow_count = 0;
316         object->pg_color = next_index;
317         if ( size > (PQ_L2_SIZE / 3 + PQ_PRIME1))
318                 incr = PQ_L2_SIZE / 3 + PQ_PRIME1;
319         else
320                 incr = size;
321         next_index = (next_index + incr) & PQ_L2_MASK;
322         object->handle = NULL;
323         object->backing_object = NULL;
324         object->backing_object_offset = (vm_ooffset_t) 0;
325
326         object->generation++;
327         object->swblock_count = 0;
328         RB_INIT(&object->swblock_root);
329         vm_object_lock_init(object);
330
331         lwkt_gettoken(&vmobj_token);
332         TAILQ_INSERT_TAIL(&vm_object_list, object, object_list);
333         vm_object_count++;
334         lwkt_reltoken(&vmobj_token);
335 }
336
337 /*
338  * Initialize the VM objects module.
339  *
340  * Called from the low level boot code only.
341  */
342 void
343 vm_object_init(void)
344 {
345         TAILQ_INIT(&vm_object_list);
346         
347         _vm_object_allocate(OBJT_DEFAULT, OFF_TO_IDX(KvaEnd),
348                             &kernel_object);
349
350         obj_zone = &obj_zone_store;
351         zbootinit(obj_zone, "VM OBJECT", sizeof (struct vm_object),
352                 vm_objects_init, VM_OBJECTS_INIT);
353 }
354
355 void
356 vm_object_init2(void)
357 {
358         zinitna(obj_zone, NULL, NULL, 0, 0, ZONE_PANICFAIL, 1);
359 }
360
361 /*
362  * Allocate and return a new object of the specified type and size.
363  *
364  * No requirements.
365  */
366 vm_object_t
367 vm_object_allocate(objtype_t type, vm_pindex_t size)
368 {
369         vm_object_t result;
370
371         result = (vm_object_t) zalloc(obj_zone);
372
373         _vm_object_allocate(type, size, result);
374
375         return (result);
376 }
377
378 /*
379  * Add an additional reference to a vm_object.
380  *
381  * Object passed by caller must be stable or caller must already
382  * hold vmobj_token to avoid races.
383  */
384 void
385 vm_object_reference(vm_object_t object)
386 {
387         lwkt_gettoken(&vmobj_token);
388         vm_object_hold(object);
389         vm_object_reference_locked(object);
390         vm_object_drop(object);
391         lwkt_reltoken(&vmobj_token);
392 }
393
394 void
395 vm_object_reference_locked(vm_object_t object)
396 {
397         if (object) {
398                 ASSERT_LWKT_TOKEN_HELD(&vmobj_token);
399                 /*NOTYET*/
400                 /*ASSERT_LWKT_TOKEN_HELD(vm_object_token(object));*/
401                 object->ref_count++;
402                 if (object->type == OBJT_VNODE) {
403                         vref(object->handle);
404                         /* XXX what if the vnode is being destroyed? */
405                 }
406         }
407 }
408
409 /*
410  * Dereference an object and its underlying vnode.
411  *
412  * The caller must hold vmobj_token.
413  * The object must be locked but not held.  This function will eat the lock.
414  */
415 static void
416 vm_object_vndeallocate(vm_object_t object)
417 {
418         struct vnode *vp = (struct vnode *) object->handle;
419
420         KASSERT(object->type == OBJT_VNODE,
421             ("vm_object_vndeallocate: not a vnode object"));
422         KASSERT(vp != NULL, ("vm_object_vndeallocate: missing vp"));
423         ASSERT_LWKT_TOKEN_HELD(&vmobj_token);
424 #ifdef INVARIANTS
425         if (object->ref_count == 0) {
426                 vprint("vm_object_vndeallocate", vp);
427                 panic("vm_object_vndeallocate: bad object reference count");
428         }
429 #endif
430
431         object->ref_count--;
432         if (object->ref_count == 0)
433                 vclrflags(vp, VTEXT);
434         vm_object_unlock(object);
435         vrele(vp);
436 }
437
438 /*
439  * Release a reference to the specified object, gained either through a
440  * vm_object_allocate or a vm_object_reference call.  When all references
441  * are gone, storage associated with this object may be relinquished.
442  *
443  * The caller does not have to hold the object locked but must have control
444  * over the reference in question in order to guarantee that the object
445  * does not get ripped out from under us.
446  */
447 void
448 vm_object_deallocate(vm_object_t object)
449 {
450         lwkt_gettoken(&vmobj_token);
451         vm_object_deallocate_locked(object);
452         lwkt_reltoken(&vmobj_token);
453 }
454
455 void
456 vm_object_deallocate_locked(vm_object_t object)
457 {
458         vm_object_t temp;
459
460         ASSERT_LWKT_TOKEN_HELD(&vmobj_token);
461
462         if (object)
463                 vm_object_lock(object);
464
465         while (object != NULL) {
466                 if (object->type == OBJT_VNODE) {
467                         vm_object_vndeallocate(object);
468                         /* vndeallocate ate the lock */
469                         break;
470                 }
471
472                 if (object->ref_count == 0) {
473                         panic("vm_object_deallocate: object deallocated "
474                               "too many times: %d", object->type);
475                 }
476                 if (object->ref_count > 2) {
477                         object->ref_count--;
478                         vm_object_unlock(object);
479                         break;
480                 }
481
482                 /*
483                  * We currently need the vm_token from this point on, and
484                  * we must recheck ref_count after acquiring it.
485                  */
486                 lwkt_gettoken(&vm_token);
487
488                 if (object->ref_count > 2) {
489                         object->ref_count--;
490                         lwkt_reltoken(&vm_token);
491                         vm_object_unlock(object);
492                         break;
493                 }
494
495                 /*
496                  * Here on ref_count of one or two, which are special cases for
497                  * objects.
498                  *
499                  * Nominal ref_count > 1 case if the second ref is not from
500                  * a shadow.
501                  */
502                 if (object->ref_count == 2 && object->shadow_count == 0) {
503                         vm_object_set_flag(object, OBJ_ONEMAPPING);
504                         object->ref_count--;
505                         lwkt_reltoken(&vm_token);
506                         vm_object_unlock(object);
507                         break;
508                 }
509
510                 /*
511                  * If the second ref is from a shadow we chain along it
512                  * if object's handle is exhausted.
513                  */
514                 if (object->ref_count == 2 && object->shadow_count == 1) {
515                         if (object->handle == NULL &&
516                             (object->type == OBJT_DEFAULT ||
517                              object->type == OBJT_SWAP)) {
518                                 temp = LIST_FIRST(&object->shadow_head);
519                                 KASSERT(temp != NULL,
520                                         ("vm_object_deallocate: ref_count: "
521                                         "%d, shadow_count: %d",
522                                         object->ref_count,
523                                         object->shadow_count));
524                                 lwkt_reltoken(&vm_token);
525                                 vm_object_lock(temp);
526
527                                 if ((temp->handle == NULL) &&
528                                     (temp->type == OBJT_DEFAULT ||
529                                      temp->type == OBJT_SWAP)) {
530                                         /*
531                                          * Special case, must handle ref_count
532                                          * manually to avoid recursion.
533                                          */
534                                         temp->ref_count++;
535                                         vm_object_lock_swap();
536
537                                         while (
538                                                 temp->paging_in_progress ||
539                                                 object->paging_in_progress
540                                         ) {
541                                                 vm_object_pip_wait(temp,
542                                                                    "objde1");
543                                                 vm_object_pip_wait(object,
544                                                                    "objde2");
545                                         }
546
547                                         if (temp->ref_count == 1) {
548                                                 object->ref_count--;
549                                                 temp->ref_count--;
550                                                 vm_object_unlock(object);
551                                                 object = temp;
552                                                 goto doterm;
553                                         }
554
555                                         lwkt_gettoken(&vm_token);
556                                         vm_object_collapse(temp);
557                                         lwkt_reltoken(&vm_token);
558                                         object->ref_count--;
559                                         vm_object_unlock(object);
560                                         object = temp;
561                                         continue;
562                                 }
563                                 vm_object_unlock(temp);
564                         } else {
565                                 lwkt_reltoken(&vm_token);
566                         }
567                         object->ref_count--;
568                         vm_object_unlock(object);
569                         break;
570                 }
571
572                 /*
573                  * Normal dereferencing path
574                  */
575                 object->ref_count--;
576                 if (object->ref_count != 0) {
577                         lwkt_reltoken(&vm_token);
578                         vm_object_unlock(object);
579                         break;
580                 }
581
582                 /*
583                  * Termination path
584                  *
585                  * We may have to loop to resolve races if we block getting
586                  * temp's lock.  If temp is non NULL we have to swap the
587                  * lock order so the original object lock as at the top
588                  * of the lock heap.
589                  */
590                 lwkt_reltoken(&vm_token);
591 doterm:
592                 while ((temp = object->backing_object) != NULL) {
593                         vm_object_lock(temp);
594                         if (temp == object->backing_object)
595                                 break;
596                         vm_object_unlock(temp);
597                 }
598                 if (temp) {
599                         LIST_REMOVE(object, shadow_list);
600                         temp->shadow_count--;
601                         temp->generation++;
602                         object->backing_object = NULL;
603                         vm_object_lock_swap();
604                 }
605
606                 /*
607                  * Don't double-terminate, we could be in a termination
608                  * recursion due to the terminate having to sync data
609                  * to disk.
610                  */
611                 if ((object->flags & OBJ_DEAD) == 0) {
612                         vm_object_terminate(object);
613                         /* termination ate the object lock */
614                 } else {
615                         vm_object_unlock(object);
616                 }
617                 object = temp;
618         }
619 }
620
621 /*
622  * Destroy the specified object, freeing up related resources.
623  *
624  * The object must have zero references.
625  *
626  * The caller must be holding vmobj_token and properly interlock with
627  * OBJ_DEAD (at the moment).
628  *
629  * The caller must have locked the object only, and not be holding it.
630  * This function will eat the caller's lock on the object.
631  */
632 static int vm_object_terminate_callback(vm_page_t p, void *data);
633
634 void
635 vm_object_terminate(vm_object_t object)
636 {
637         /*
638          * Make sure no one uses us.  Once we set OBJ_DEAD we should be
639          * able to safely block.
640          */
641         KKASSERT((object->flags & OBJ_DEAD) == 0);
642         ASSERT_LWKT_TOKEN_HELD(&vmobj_token);
643         vm_object_set_flag(object, OBJ_DEAD);
644
645         /*
646          * Wait for the pageout daemon to be done with the object
647          */
648         vm_object_pip_wait(object, "objtrm1");
649
650         KASSERT(!object->paging_in_progress,
651                 ("vm_object_terminate: pageout in progress"));
652
653         /*
654          * Clean and free the pages, as appropriate. All references to the
655          * object are gone, so we don't need to lock it.
656          */
657         if (object->type == OBJT_VNODE) {
658                 struct vnode *vp;
659
660                 /*
661                  * Clean pages and flush buffers.
662                  */
663                 vm_object_page_clean(object, 0, 0, OBJPC_SYNC);
664
665                 vp = (struct vnode *) object->handle;
666                 vinvalbuf(vp, V_SAVE, 0, 0);
667         }
668
669         /*
670          * Wait for any I/O to complete, after which there had better not
671          * be any references left on the object.
672          */
673         vm_object_pip_wait(object, "objtrm2");
674
675         if (object->ref_count != 0) {
676                 panic("vm_object_terminate: object with references, "
677                       "ref_count=%d", object->ref_count);
678         }
679
680         /*
681          * Now free any remaining pages. For internal objects, this also
682          * removes them from paging queues. Don't free wired pages, just
683          * remove them from the object. 
684          */
685         lwkt_gettoken(&vm_token);
686         vm_page_rb_tree_RB_SCAN(&object->rb_memq, NULL,
687                                 vm_object_terminate_callback, NULL);
688         lwkt_reltoken(&vm_token);
689
690         /*
691          * Let the pager know object is dead.
692          */
693         vm_pager_deallocate(object);
694
695         /*
696          * Wait for the object hold count to hit zero, clean out pages as
697          * we go.
698          */
699         lwkt_gettoken(&vm_token);
700         for (;;) {
701                 vm_object_hold_wait(object);
702                 if (RB_ROOT(&object->rb_memq) == NULL)
703                         break;
704                 kprintf("vm_object_terminate: Warning, object %p "
705                         "still has %d pages\n",
706                         object, object->resident_page_count);
707                 vm_page_rb_tree_RB_SCAN(&object->rb_memq, NULL,
708                                         vm_object_terminate_callback, NULL);
709         }
710         lwkt_reltoken(&vm_token);
711
712         /*
713          * There had better not be any pages left
714          */
715         KKASSERT(object->resident_page_count == 0);
716
717         /*
718          * Remove the object from the global object list.
719          *
720          * (we are holding vmobj_token)
721          */
722         TAILQ_REMOVE(&vm_object_list, object, object_list);
723         vm_object_count--;
724         vm_object_dead_wakeup(object);
725         vm_object_unlock(object);
726
727         if (object->ref_count != 0) {
728                 panic("vm_object_terminate2: object with references, "
729                       "ref_count=%d", object->ref_count);
730         }
731
732         /*
733          * Free the space for the object.
734          */
735         zfree(obj_zone, object);
736 }
737
738 /*
739  * The caller must hold vm_token.
740  */
741 static int
742 vm_object_terminate_callback(vm_page_t p, void *data __unused)
743 {
744         if (p->busy || (p->flags & PG_BUSY))
745                 panic("vm_object_terminate: freeing busy page %p", p);
746         if (p->wire_count == 0) {
747                 vm_page_busy(p);
748                 vm_page_free(p);
749                 mycpu->gd_cnt.v_pfree++;
750         } else {
751                 if (p->queue != PQ_NONE)
752                         kprintf("vm_object_terminate: Warning: Encountered wired page %p on queue %d\n", p, p->queue);
753                 vm_page_busy(p);
754                 vm_page_remove(p);
755                 vm_page_wakeup(p);
756         }
757         return(0);
758 }
759
760 /*
761  * The object is dead but still has an object<->pager association.  Sleep
762  * and return.  The caller typically retests the association in a loop.
763  *
764  * Must be called with the vmobj_token held.
765  */
766 void
767 vm_object_dead_sleep(vm_object_t object, const char *wmesg)
768 {
769         ASSERT_LWKT_TOKEN_HELD(&vmobj_token);
770         if (object->handle) {
771                 vm_object_set_flag(object, OBJ_DEADWNT);
772                 tsleep(object, 0, wmesg, 0);
773                 /* object may be invalid after this point */
774         }
775 }
776
777 /*
778  * Wakeup anyone waiting for the object<->pager disassociation on
779  * a dead object.
780  *
781  * Must be called with the vmobj_token held.
782  */
783 void
784 vm_object_dead_wakeup(vm_object_t object)
785 {
786         ASSERT_LWKT_TOKEN_HELD(&vmobj_token);
787         if (object->flags & OBJ_DEADWNT) {
788                 vm_object_clear_flag(object, OBJ_DEADWNT);
789                 wakeup(object);
790         }
791 }
792
793 /*
794  * Clean all dirty pages in the specified range of object.  Leaves page
795  * on whatever queue it is currently on.   If NOSYNC is set then do not
796  * write out pages with PG_NOSYNC set (originally comes from MAP_NOSYNC),
797  * leaving the object dirty.
798  *
799  * When stuffing pages asynchronously, allow clustering.  XXX we need a
800  * synchronous clustering mode implementation.
801  *
802  * Odd semantics: if start == end, we clean everything.
803  *
804  * The object must be locked? XXX
805  */
806 static int vm_object_page_clean_pass1(struct vm_page *p, void *data);
807 static int vm_object_page_clean_pass2(struct vm_page *p, void *data);
808
809 void
810 vm_object_page_clean(vm_object_t object, vm_pindex_t start, vm_pindex_t end,
811                      int flags)
812 {
813         struct rb_vm_page_scan_info info;
814         struct vnode *vp;
815         int wholescan;
816         int pagerflags;
817         int curgeneration;
818
819         vm_object_hold(object);
820         if (object->type != OBJT_VNODE ||
821             (object->flags & OBJ_MIGHTBEDIRTY) == 0) {
822                 vm_object_drop(object);
823                 return;
824         }
825
826         pagerflags = (flags & (OBJPC_SYNC | OBJPC_INVAL)) ? 
827                         VM_PAGER_PUT_SYNC : VM_PAGER_CLUSTER_OK;
828         pagerflags |= (flags & OBJPC_INVAL) ? VM_PAGER_PUT_INVAL : 0;
829
830         vp = object->handle;
831
832         /*
833          * Interlock other major object operations.  This allows us to 
834          * temporarily clear OBJ_WRITEABLE and OBJ_MIGHTBEDIRTY.
835          */
836         crit_enter();
837         vm_object_set_flag(object, OBJ_CLEANING);
838
839         /*
840          * Handle 'entire object' case
841          */
842         info.start_pindex = start;
843         if (end == 0) {
844                 info.end_pindex = object->size - 1;
845         } else {
846                 info.end_pindex = end - 1;
847         }
848         wholescan = (start == 0 && info.end_pindex == object->size - 1);
849         info.limit = flags;
850         info.pagerflags = pagerflags;
851         info.object = object;
852
853         /*
854          * If cleaning the entire object do a pass to mark the pages read-only.
855          * If everything worked out ok, clear OBJ_WRITEABLE and
856          * OBJ_MIGHTBEDIRTY.
857          */
858         if (wholescan) {
859                 info.error = 0;
860                 lwkt_gettoken(&vm_token);
861                 vm_page_rb_tree_RB_SCAN(&object->rb_memq, rb_vm_page_scancmp,
862                                         vm_object_page_clean_pass1, &info);
863                 lwkt_reltoken(&vm_token);
864                 if (info.error == 0) {
865                         vm_object_clear_flag(object,
866                                              OBJ_WRITEABLE|OBJ_MIGHTBEDIRTY);
867                         if (object->type == OBJT_VNODE &&
868                             (vp = (struct vnode *)object->handle) != NULL) {
869                                 if (vp->v_flag & VOBJDIRTY) 
870                                         vclrflags(vp, VOBJDIRTY);
871                         }
872                 }
873         }
874
875         /*
876          * Do a pass to clean all the dirty pages we find.
877          */
878         do {
879                 info.error = 0;
880                 curgeneration = object->generation;
881                 lwkt_gettoken(&vm_token);
882                 vm_page_rb_tree_RB_SCAN(&object->rb_memq, rb_vm_page_scancmp,
883                                         vm_object_page_clean_pass2, &info);
884                 lwkt_reltoken(&vm_token);
885         } while (info.error || curgeneration != object->generation);
886
887         vm_object_clear_flag(object, OBJ_CLEANING);
888         crit_exit();
889         vm_object_drop(object);
890 }
891
892 /*
893  * The caller must hold vm_token.
894  */
895 static 
896 int
897 vm_object_page_clean_pass1(struct vm_page *p, void *data)
898 {
899         struct rb_vm_page_scan_info *info = data;
900
901         vm_page_flag_set(p, PG_CLEANCHK);
902         if ((info->limit & OBJPC_NOSYNC) && (p->flags & PG_NOSYNC))
903                 info->error = 1;
904         else
905                 vm_page_protect(p, VM_PROT_READ);       /* must not block */
906         return(0);
907 }
908
909 /*
910  * The caller must hold vm_token.
911  */
912 static 
913 int
914 vm_object_page_clean_pass2(struct vm_page *p, void *data)
915 {
916         struct rb_vm_page_scan_info *info = data;
917         int n;
918
919         /*
920          * Do not mess with pages that were inserted after we started
921          * the cleaning pass.
922          */
923         if ((p->flags & PG_CLEANCHK) == 0)
924                 return(0);
925
926         /*
927          * Before wasting time traversing the pmaps, check for trivial
928          * cases where the page cannot be dirty.
929          */
930         if (p->valid == 0 || (p->queue - p->pc) == PQ_CACHE) {
931                 KKASSERT((p->dirty & p->valid) == 0);
932                 return(0);
933         }
934
935         /*
936          * Check whether the page is dirty or not.  The page has been set
937          * to be read-only so the check will not race a user dirtying the
938          * page.
939          */
940         vm_page_test_dirty(p);
941         if ((p->dirty & p->valid) == 0) {
942                 vm_page_flag_clear(p, PG_CLEANCHK);
943                 return(0);
944         }
945
946         /*
947          * If we have been asked to skip nosync pages and this is a
948          * nosync page, skip it.  Note that the object flags were
949          * not cleared in this case (because pass1 will have returned an
950          * error), so we do not have to set them.
951          */
952         if ((info->limit & OBJPC_NOSYNC) && (p->flags & PG_NOSYNC)) {
953                 vm_page_flag_clear(p, PG_CLEANCHK);
954                 return(0);
955         }
956
957         /*
958          * Flush as many pages as we can.  PG_CLEANCHK will be cleared on
959          * the pages that get successfully flushed.  Set info->error if
960          * we raced an object modification.
961          */
962         n = vm_object_page_collect_flush(info->object, p, info->pagerflags);
963         if (n == 0)
964                 info->error = 1;
965         return(0);
966 }
967
968 /*
969  * Collect the specified page and nearby pages and flush them out.
970  * The number of pages flushed is returned.
971  *
972  * The caller must hold vm_token.
973  */
974 static int
975 vm_object_page_collect_flush(vm_object_t object, vm_page_t p, int pagerflags)
976 {
977         int runlen;
978         int maxf;
979         int chkb;
980         int maxb;
981         int i;
982         int curgeneration;
983         vm_pindex_t pi;
984         vm_page_t maf[vm_pageout_page_count];
985         vm_page_t mab[vm_pageout_page_count];
986         vm_page_t ma[vm_pageout_page_count];
987
988         curgeneration = object->generation;
989
990         pi = p->pindex;
991         while (vm_page_sleep_busy(p, TRUE, "vpcwai")) {
992                 if (object->generation != curgeneration) {
993                         return(0);
994                 }
995         }
996         KKASSERT(p->object == object && p->pindex == pi);
997
998         maxf = 0;
999         for(i = 1; i < vm_pageout_page_count; i++) {
1000                 vm_page_t tp;
1001
1002                 if ((tp = vm_page_lookup(object, pi + i)) != NULL) {
1003                         if ((tp->flags & PG_BUSY) ||
1004                                 ((pagerflags & VM_PAGER_IGNORE_CLEANCHK) == 0 && 
1005                                  (tp->flags & PG_CLEANCHK) == 0) ||
1006                                 (tp->busy != 0))
1007                                 break;
1008                         if((tp->queue - tp->pc) == PQ_CACHE) {
1009                                 vm_page_flag_clear(tp, PG_CLEANCHK);
1010                                 break;
1011                         }
1012                         vm_page_test_dirty(tp);
1013                         if ((tp->dirty & tp->valid) == 0) {
1014                                 vm_page_flag_clear(tp, PG_CLEANCHK);
1015                                 break;
1016                         }
1017                         maf[ i - 1 ] = tp;
1018                         maxf++;
1019                         continue;
1020                 }
1021                 break;
1022         }
1023
1024         maxb = 0;
1025         chkb = vm_pageout_page_count -  maxf;
1026         if (chkb) {
1027                 for(i = 1; i < chkb;i++) {
1028                         vm_page_t tp;
1029
1030                         if ((tp = vm_page_lookup(object, pi - i)) != NULL) {
1031                                 if ((tp->flags & PG_BUSY) ||
1032                                         ((pagerflags & VM_PAGER_IGNORE_CLEANCHK) == 0 && 
1033                                          (tp->flags & PG_CLEANCHK) == 0) ||
1034                                         (tp->busy != 0))
1035                                         break;
1036                                 if((tp->queue - tp->pc) == PQ_CACHE) {
1037                                         vm_page_flag_clear(tp, PG_CLEANCHK);
1038                                         break;
1039                                 }
1040                                 vm_page_test_dirty(tp);
1041                                 if ((tp->dirty & tp->valid) == 0) {
1042                                         vm_page_flag_clear(tp, PG_CLEANCHK);
1043                                         break;
1044                                 }
1045                                 mab[ i - 1 ] = tp;
1046                                 maxb++;
1047                                 continue;
1048                         }
1049                         break;
1050                 }
1051         }
1052
1053         for(i = 0; i < maxb; i++) {
1054                 int index = (maxb - i) - 1;
1055                 ma[index] = mab[i];
1056                 vm_page_flag_clear(ma[index], PG_CLEANCHK);
1057         }
1058         vm_page_flag_clear(p, PG_CLEANCHK);
1059         ma[maxb] = p;
1060         for(i = 0; i < maxf; i++) {
1061                 int index = (maxb + i) + 1;
1062                 ma[index] = maf[i];
1063                 vm_page_flag_clear(ma[index], PG_CLEANCHK);
1064         }
1065         runlen = maxb + maxf + 1;
1066
1067         vm_pageout_flush(ma, runlen, pagerflags);
1068         for (i = 0; i < runlen; i++) {
1069                 if (ma[i]->valid & ma[i]->dirty) {
1070                         vm_page_protect(ma[i], VM_PROT_READ);
1071                         vm_page_flag_set(ma[i], PG_CLEANCHK);
1072
1073                         /*
1074                          * maxf will end up being the actual number of pages
1075                          * we wrote out contiguously, non-inclusive of the
1076                          * first page.  We do not count look-behind pages.
1077                          */
1078                         if (i >= maxb + 1 && (maxf > i - maxb - 1))
1079                                 maxf = i - maxb - 1;
1080                 }
1081         }
1082         return(maxf + 1);
1083 }
1084
1085 /*
1086  * Same as vm_object_pmap_copy, except range checking really
1087  * works, and is meant for small sections of an object.
1088  *
1089  * This code protects resident pages by making them read-only
1090  * and is typically called on a fork or split when a page
1091  * is converted to copy-on-write.  
1092  *
1093  * NOTE: If the page is already at VM_PROT_NONE, calling
1094  * vm_page_protect will have no effect.
1095  */
1096 void
1097 vm_object_pmap_copy_1(vm_object_t object, vm_pindex_t start, vm_pindex_t end)
1098 {
1099         vm_pindex_t idx;
1100         vm_page_t p;
1101
1102         if (object == NULL || (object->flags & OBJ_WRITEABLE) == 0)
1103                 return;
1104
1105         /*
1106          * spl protection needed to prevent races between the lookup,
1107          * an interrupt unbusy/free, and our protect call.
1108          */
1109         crit_enter();
1110         lwkt_gettoken(&vm_token);
1111         for (idx = start; idx < end; idx++) {
1112                 p = vm_page_lookup(object, idx);
1113                 if (p == NULL)
1114                         continue;
1115                 vm_page_protect(p, VM_PROT_READ);
1116         }
1117         lwkt_reltoken(&vm_token);
1118         crit_exit();
1119 }
1120
1121 /*
1122  * Removes all physical pages in the specified object range from all
1123  * physical maps.
1124  *
1125  * The object must *not* be locked.
1126  */
1127
1128 static int vm_object_pmap_remove_callback(vm_page_t p, void *data);
1129
1130 void
1131 vm_object_pmap_remove(vm_object_t object, vm_pindex_t start, vm_pindex_t end)
1132 {
1133         struct rb_vm_page_scan_info info;
1134
1135         if (object == NULL)
1136                 return;
1137         info.start_pindex = start;
1138         info.end_pindex = end - 1;
1139
1140         crit_enter();
1141         lwkt_gettoken(&vm_token);
1142         vm_page_rb_tree_RB_SCAN(&object->rb_memq, rb_vm_page_scancmp,
1143                                 vm_object_pmap_remove_callback, &info);
1144         if (start == 0 && end == object->size)
1145                 vm_object_clear_flag(object, OBJ_WRITEABLE);
1146         lwkt_reltoken(&vm_token);
1147         crit_exit();
1148 }
1149
1150 /*
1151  * The caller must hold vm_token.
1152  */
1153 static int
1154 vm_object_pmap_remove_callback(vm_page_t p, void *data __unused)
1155 {
1156         vm_page_protect(p, VM_PROT_NONE);
1157         return(0);
1158 }
1159
1160 /*
1161  * Implements the madvise function at the object/page level.
1162  *
1163  * MADV_WILLNEED        (any object)
1164  *
1165  *      Activate the specified pages if they are resident.
1166  *
1167  * MADV_DONTNEED        (any object)
1168  *
1169  *      Deactivate the specified pages if they are resident.
1170  *
1171  * MADV_FREE    (OBJT_DEFAULT/OBJT_SWAP objects, OBJ_ONEMAPPING only)
1172  *
1173  *      Deactivate and clean the specified pages if they are
1174  *      resident.  This permits the process to reuse the pages
1175  *      without faulting or the kernel to reclaim the pages
1176  *      without I/O.
1177  *
1178  * No requirements.
1179  */
1180 void
1181 vm_object_madvise(vm_object_t object, vm_pindex_t pindex, int count, int advise)
1182 {
1183         vm_pindex_t end, tpindex;
1184         vm_object_t tobject;
1185         vm_page_t m;
1186
1187         if (object == NULL)
1188                 return;
1189
1190         end = pindex + count;
1191
1192         lwkt_gettoken(&vm_token);
1193
1194         /*
1195          * Locate and adjust resident pages
1196          */
1197         for (; pindex < end; pindex += 1) {
1198 relookup:
1199                 tobject = object;
1200                 tpindex = pindex;
1201 shadowlookup:
1202                 /*
1203                  * MADV_FREE only operates on OBJT_DEFAULT or OBJT_SWAP pages
1204                  * and those pages must be OBJ_ONEMAPPING.
1205                  */
1206                 if (advise == MADV_FREE) {
1207                         if ((tobject->type != OBJT_DEFAULT &&
1208                              tobject->type != OBJT_SWAP) ||
1209                             (tobject->flags & OBJ_ONEMAPPING) == 0) {
1210                                 continue;
1211                         }
1212                 }
1213
1214                 /*
1215                  * spl protection is required to avoid a race between the
1216                  * lookup, an interrupt unbusy/free, and our busy check.
1217                  */
1218
1219                 crit_enter();
1220                 m = vm_page_lookup(tobject, tpindex);
1221
1222                 if (m == NULL) {
1223                         /*
1224                          * There may be swap even if there is no backing page
1225                          */
1226                         if (advise == MADV_FREE && tobject->type == OBJT_SWAP)
1227                                 swap_pager_freespace(tobject, tpindex, 1);
1228
1229                         /*
1230                          * next object
1231                          */
1232                         crit_exit();
1233                         if (tobject->backing_object == NULL)
1234                                 continue;
1235                         tpindex += OFF_TO_IDX(tobject->backing_object_offset);
1236                         tobject = tobject->backing_object;
1237                         goto shadowlookup;
1238                 }
1239
1240                 /*
1241                  * If the page is busy or not in a normal active state,
1242                  * we skip it.  If the page is not managed there are no
1243                  * page queues to mess with.  Things can break if we mess
1244                  * with pages in any of the below states.
1245                  */
1246                 if (
1247                     m->hold_count ||
1248                     m->wire_count ||
1249                     (m->flags & PG_UNMANAGED) ||
1250                     m->valid != VM_PAGE_BITS_ALL
1251                 ) {
1252                         crit_exit();
1253                         continue;
1254                 }
1255
1256                 if (vm_page_sleep_busy(m, TRUE, "madvpo")) {
1257                         crit_exit();
1258                         goto relookup;
1259                 }
1260                 vm_page_busy(m);
1261                 crit_exit();
1262
1263                 /*
1264                  * Theoretically once a page is known not to be busy, an
1265                  * interrupt cannot come along and rip it out from under us.
1266                  */
1267
1268                 if (advise == MADV_WILLNEED) {
1269                         vm_page_activate(m);
1270                 } else if (advise == MADV_DONTNEED) {
1271                         vm_page_dontneed(m);
1272                 } else if (advise == MADV_FREE) {
1273                         /*
1274                          * Mark the page clean.  This will allow the page
1275                          * to be freed up by the system.  However, such pages
1276                          * are often reused quickly by malloc()/free()
1277                          * so we do not do anything that would cause
1278                          * a page fault if we can help it.
1279                          *
1280                          * Specifically, we do not try to actually free
1281                          * the page now nor do we try to put it in the
1282                          * cache (which would cause a page fault on reuse).
1283                          *
1284                          * But we do make the page is freeable as we
1285                          * can without actually taking the step of unmapping
1286                          * it.
1287                          */
1288                         pmap_clear_modify(m);
1289                         m->dirty = 0;
1290                         m->act_count = 0;
1291                         vm_page_dontneed(m);
1292                         if (tobject->type == OBJT_SWAP)
1293                                 swap_pager_freespace(tobject, tpindex, 1);
1294                 }
1295                 vm_page_wakeup(m);
1296         }       
1297         lwkt_reltoken(&vm_token);
1298 }
1299
1300 /*
1301  * Create a new object which is backed by the specified existing object
1302  * range.  The source object reference is deallocated.
1303  *
1304  * The new object and offset into that object are returned in the source
1305  * parameters.
1306  *
1307  * No other requirements.
1308  */
1309 void
1310 vm_object_shadow(vm_object_t *object, vm_ooffset_t *offset, vm_size_t length)
1311 {
1312         vm_object_t source;
1313         vm_object_t result;
1314
1315         source = *object;
1316
1317         /*
1318          * Don't create the new object if the old object isn't shared.
1319          */
1320         lwkt_gettoken(&vm_token);
1321
1322         if (source != NULL &&
1323             source->ref_count == 1 &&
1324             source->handle == NULL &&
1325             (source->type == OBJT_DEFAULT ||
1326              source->type == OBJT_SWAP)) {
1327                 lwkt_reltoken(&vm_token);
1328                 return;
1329         }
1330
1331         /*
1332          * Allocate a new object with the given length
1333          */
1334
1335         if ((result = vm_object_allocate(OBJT_DEFAULT, length)) == NULL)
1336                 panic("vm_object_shadow: no object for shadowing");
1337
1338         /*
1339          * The new object shadows the source object, adding a reference to it.
1340          * Our caller changes his reference to point to the new object,
1341          * removing a reference to the source object.  Net result: no change
1342          * of reference count.
1343          *
1344          * Try to optimize the result object's page color when shadowing
1345          * in order to maintain page coloring consistency in the combined 
1346          * shadowed object.
1347          */
1348         result->backing_object = source;
1349         if (source) {
1350                 LIST_INSERT_HEAD(&source->shadow_head, result, shadow_list);
1351                 source->shadow_count++;
1352                 source->generation++;
1353                 result->pg_color = (source->pg_color + OFF_TO_IDX(*offset)) & PQ_L2_MASK;
1354         }
1355
1356         /*
1357          * Store the offset into the source object, and fix up the offset into
1358          * the new object.
1359          */
1360         result->backing_object_offset = *offset;
1361         lwkt_reltoken(&vm_token);
1362
1363         /*
1364          * Return the new things
1365          */
1366         *offset = 0;
1367         *object = result;
1368 }
1369
1370 #define OBSC_TEST_ALL_SHADOWED  0x0001
1371 #define OBSC_COLLAPSE_NOWAIT    0x0002
1372 #define OBSC_COLLAPSE_WAIT      0x0004
1373
1374 static int vm_object_backing_scan_callback(vm_page_t p, void *data);
1375
1376 /*
1377  * The caller must hold vm_token.
1378  */
1379 static __inline int
1380 vm_object_backing_scan(vm_object_t object, int op)
1381 {
1382         struct rb_vm_page_scan_info info;
1383         vm_object_t backing_object;
1384
1385         crit_enter();
1386
1387         backing_object = object->backing_object;
1388         info.backing_offset_index = OFF_TO_IDX(object->backing_object_offset);
1389
1390         /*
1391          * Initial conditions
1392          */
1393
1394         if (op & OBSC_TEST_ALL_SHADOWED) {
1395                 /*
1396                  * We do not want to have to test for the existence of
1397                  * swap pages in the backing object.  XXX but with the
1398                  * new swapper this would be pretty easy to do.
1399                  *
1400                  * XXX what about anonymous MAP_SHARED memory that hasn't
1401                  * been ZFOD faulted yet?  If we do not test for this, the
1402                  * shadow test may succeed! XXX
1403                  */
1404                 if (backing_object->type != OBJT_DEFAULT) {
1405                         crit_exit();
1406                         return(0);
1407                 }
1408         }
1409         if (op & OBSC_COLLAPSE_WAIT) {
1410                 KKASSERT((backing_object->flags & OBJ_DEAD) == 0);
1411                 vm_object_set_flag(backing_object, OBJ_DEAD);
1412         }
1413
1414         /*
1415          * Our scan.   We have to retry if a negative error code is returned,
1416          * otherwise 0 or 1 will be returned in info.error.  0 Indicates that
1417          * the scan had to be stopped because the parent does not completely
1418          * shadow the child.
1419          */
1420         info.object = object;
1421         info.backing_object = backing_object;
1422         info.limit = op;
1423         do {
1424                 info.error = 1;
1425                 vm_page_rb_tree_RB_SCAN(&backing_object->rb_memq, NULL,
1426                                         vm_object_backing_scan_callback,
1427                                         &info);
1428         } while (info.error < 0);
1429         crit_exit();
1430         return(info.error);
1431 }
1432
1433 /*
1434  * The caller must hold vm_token.
1435  */
1436 static int
1437 vm_object_backing_scan_callback(vm_page_t p, void *data)
1438 {
1439         struct rb_vm_page_scan_info *info = data;
1440         vm_object_t backing_object;
1441         vm_object_t object;
1442         vm_pindex_t new_pindex;
1443         vm_pindex_t backing_offset_index;
1444         int op;
1445
1446         new_pindex = p->pindex - info->backing_offset_index;
1447         op = info->limit;
1448         object = info->object;
1449         backing_object = info->backing_object;
1450         backing_offset_index = info->backing_offset_index;
1451
1452         if (op & OBSC_TEST_ALL_SHADOWED) {
1453                 vm_page_t pp;
1454
1455                 /*
1456                  * Ignore pages outside the parent object's range
1457                  * and outside the parent object's mapping of the 
1458                  * backing object.
1459                  *
1460                  * note that we do not busy the backing object's
1461                  * page.
1462                  */
1463                 if (
1464                     p->pindex < backing_offset_index ||
1465                     new_pindex >= object->size
1466                 ) {
1467                         return(0);
1468                 }
1469
1470                 /*
1471                  * See if the parent has the page or if the parent's
1472                  * object pager has the page.  If the parent has the
1473                  * page but the page is not valid, the parent's
1474                  * object pager must have the page.
1475                  *
1476                  * If this fails, the parent does not completely shadow
1477                  * the object and we might as well give up now.
1478                  */
1479
1480                 pp = vm_page_lookup(object, new_pindex);
1481                 if ((pp == NULL || pp->valid == 0) &&
1482                     !vm_pager_has_page(object, new_pindex)
1483                 ) {
1484                         info->error = 0;        /* problemo */
1485                         return(-1);             /* stop the scan */
1486                 }
1487         }
1488
1489         /*
1490          * Check for busy page
1491          */
1492
1493         if (op & (OBSC_COLLAPSE_WAIT | OBSC_COLLAPSE_NOWAIT)) {
1494                 vm_page_t pp;
1495
1496                 if (op & OBSC_COLLAPSE_NOWAIT) {
1497                         if (
1498                             (p->flags & PG_BUSY) ||
1499                             !p->valid || 
1500                             p->hold_count || 
1501                             p->wire_count ||
1502                             p->busy
1503                         ) {
1504                                 return(0);
1505                         }
1506                 } else if (op & OBSC_COLLAPSE_WAIT) {
1507                         if (vm_page_sleep_busy(p, TRUE, "vmocol")) {
1508                                 /*
1509                                  * If we slept, anything could have
1510                                  * happened.   Ask that the scan be restarted.
1511                                  *
1512                                  * Since the object is marked dead, the
1513                                  * backing offset should not have changed.  
1514                                  */
1515                                 info->error = -1;
1516                                 return(-1);
1517                         }
1518                 }
1519
1520                 /* 
1521                  * Busy the page
1522                  */
1523                 vm_page_busy(p);
1524
1525                 KASSERT(
1526                     p->object == backing_object,
1527                     ("vm_object_qcollapse(): object mismatch")
1528                 );
1529
1530                 /*
1531                  * Destroy any associated swap
1532                  */
1533                 if (backing_object->type == OBJT_SWAP)
1534                         swap_pager_freespace(backing_object, p->pindex, 1);
1535
1536                 if (
1537                     p->pindex < backing_offset_index ||
1538                     new_pindex >= object->size
1539                 ) {
1540                         /*
1541                          * Page is out of the parent object's range, we 
1542                          * can simply destroy it. 
1543                          */
1544                         vm_page_protect(p, VM_PROT_NONE);
1545                         vm_page_free(p);
1546                         return(0);
1547                 }
1548
1549                 pp = vm_page_lookup(object, new_pindex);
1550                 if (pp != NULL || vm_pager_has_page(object, new_pindex)) {
1551                         /*
1552                          * page already exists in parent OR swap exists
1553                          * for this location in the parent.  Destroy 
1554                          * the original page from the backing object.
1555                          *
1556                          * Leave the parent's page alone
1557                          */
1558                         vm_page_protect(p, VM_PROT_NONE);
1559                         vm_page_free(p);
1560                         return(0);
1561                 }
1562
1563                 /*
1564                  * Page does not exist in parent, rename the
1565                  * page from the backing object to the main object. 
1566                  *
1567                  * If the page was mapped to a process, it can remain 
1568                  * mapped through the rename.
1569                  */
1570                 if ((p->queue - p->pc) == PQ_CACHE)
1571                         vm_page_deactivate(p);
1572
1573                 vm_page_rename(p, object, new_pindex);
1574                 /* page automatically made dirty by rename */
1575         }
1576         return(0);
1577 }
1578
1579 /*
1580  * This version of collapse allows the operation to occur earlier and
1581  * when paging_in_progress is true for an object...  This is not a complete
1582  * operation, but should plug 99.9% of the rest of the leaks.
1583  *
1584  * The caller must hold vm_token and vmobj_token.
1585  * (only called from vm_object_collapse)
1586  */
1587 static void
1588 vm_object_qcollapse(vm_object_t object)
1589 {
1590         vm_object_t backing_object = object->backing_object;
1591
1592         if (backing_object->ref_count != 1)
1593                 return;
1594
1595         backing_object->ref_count += 2;
1596
1597         vm_object_backing_scan(object, OBSC_COLLAPSE_NOWAIT);
1598
1599         backing_object->ref_count -= 2;
1600 }
1601
1602 /*
1603  * Collapse an object with the object backing it.  Pages in the backing
1604  * object are moved into the parent, and the backing object is deallocated.
1605  *
1606  * The caller must hold (object).
1607  */
1608 void
1609 vm_object_collapse(vm_object_t object)
1610 {
1611         ASSERT_LWKT_TOKEN_HELD(&vm_token);
1612         ASSERT_LWKT_TOKEN_HELD(&vmobj_token);
1613         vm_object_assert_held(object);
1614
1615         while (TRUE) {
1616                 vm_object_t backing_object;
1617
1618                 /*
1619                  * Verify that the conditions are right for collapse:
1620                  *
1621                  * The object exists and the backing object exists.
1622                  */
1623                 if (object == NULL)
1624                         break;
1625
1626                 if ((backing_object = object->backing_object) == NULL)
1627                         break;
1628
1629                 vm_object_hold(backing_object);
1630                 if (backing_object != object->backing_object) {
1631                         vm_object_drop(backing_object);
1632                         continue;
1633                 }
1634
1635                 /*
1636                  * we check the backing object first, because it is most likely
1637                  * not collapsable.
1638                  */
1639                 if (backing_object->handle != NULL ||
1640                     (backing_object->type != OBJT_DEFAULT &&
1641                      backing_object->type != OBJT_SWAP) ||
1642                     (backing_object->flags & OBJ_DEAD) ||
1643                     object->handle != NULL ||
1644                     (object->type != OBJT_DEFAULT &&
1645                      object->type != OBJT_SWAP) ||
1646                     (object->flags & OBJ_DEAD)) {
1647                         vm_object_drop(backing_object);
1648                         break;
1649                 }
1650
1651                 if (
1652                     object->paging_in_progress != 0 ||
1653                     backing_object->paging_in_progress != 0
1654                 ) {
1655                         vm_object_drop(backing_object);
1656                         vm_object_qcollapse(object);
1657                         break;
1658                 }
1659
1660                 /*
1661                  * We know that we can either collapse the backing object (if
1662                  * the parent is the only reference to it) or (perhaps) have
1663                  * the parent bypass the object if the parent happens to shadow
1664                  * all the resident pages in the entire backing object.
1665                  *
1666                  * This is ignoring pager-backed pages such as swap pages.
1667                  * vm_object_backing_scan fails the shadowing test in this
1668                  * case.
1669                  */
1670
1671                 if (backing_object->ref_count == 1) {
1672                         /*
1673                          * If there is exactly one reference to the backing
1674                          * object, we can collapse it into the parent.  
1675                          */
1676                         vm_object_backing_scan(object, OBSC_COLLAPSE_WAIT);
1677
1678                         /*
1679                          * Move the pager from backing_object to object.
1680                          */
1681
1682                         if (backing_object->type == OBJT_SWAP) {
1683                                 vm_object_pip_add(backing_object, 1);
1684
1685                                 /*
1686                                  * scrap the paging_offset junk and do a 
1687                                  * discrete copy.  This also removes major 
1688                                  * assumptions about how the swap-pager 
1689                                  * works from where it doesn't belong.  The
1690                                  * new swapper is able to optimize the
1691                                  * destroy-source case.
1692                                  */
1693
1694                                 vm_object_pip_add(object, 1);
1695                                 swap_pager_copy(
1696                                     backing_object,
1697                                     object,
1698                                     OFF_TO_IDX(object->backing_object_offset), TRUE);
1699                                 vm_object_pip_wakeup(object);
1700
1701                                 vm_object_pip_wakeup(backing_object);
1702                         }
1703                         /*
1704                          * Object now shadows whatever backing_object did.
1705                          * Note that the reference to 
1706                          * backing_object->backing_object moves from within 
1707                          * backing_object to within object.
1708                          */
1709
1710                         LIST_REMOVE(object, shadow_list);
1711                         object->backing_object->shadow_count--;
1712                         object->backing_object->generation++;
1713                         if (backing_object->backing_object) {
1714                                 LIST_REMOVE(backing_object, shadow_list);
1715                                 backing_object->backing_object->shadow_count--;
1716                                 backing_object->backing_object->generation++;
1717                         }
1718                         object->backing_object = backing_object->backing_object;
1719                         if (object->backing_object) {
1720                                 LIST_INSERT_HEAD(
1721                                     &object->backing_object->shadow_head,
1722                                     object, 
1723                                     shadow_list
1724                                 );
1725                                 object->backing_object->shadow_count++;
1726                                 object->backing_object->generation++;
1727                         }
1728
1729                         object->backing_object_offset +=
1730                             backing_object->backing_object_offset;
1731
1732                         /*
1733                          * Discard backing_object.
1734                          *
1735                          * Since the backing object has no pages, no pager left,
1736                          * and no object references within it, all that is
1737                          * necessary is to dispose of it.
1738                          */
1739
1740                         KASSERT(backing_object->ref_count == 1,
1741                                 ("backing_object %p was somehow "
1742                                  "re-referenced during collapse!",
1743                                  backing_object));
1744                         KASSERT(RB_EMPTY(&backing_object->rb_memq),
1745                                 ("backing_object %p somehow has left "
1746                                  "over pages during collapse!",
1747                                  backing_object));
1748
1749                         /*
1750                          * Wait for hold count to hit zero
1751                          */
1752                         vm_object_drop(backing_object);
1753                         vm_object_hold_wait(backing_object);
1754
1755                         /* (we are holding vmobj_token) */
1756                         TAILQ_REMOVE(&vm_object_list, backing_object,
1757                                      object_list);
1758                         vm_object_count--;
1759
1760                         zfree(obj_zone, backing_object);
1761
1762                         object_collapses++;
1763                 } else {
1764                         vm_object_t new_backing_object;
1765
1766                         /*
1767                          * If we do not entirely shadow the backing object,
1768                          * there is nothing we can do so we give up.
1769                          */
1770
1771                         if (vm_object_backing_scan(object, OBSC_TEST_ALL_SHADOWED) == 0) {
1772                                 vm_object_drop(backing_object);
1773                                 break;
1774                         }
1775
1776                         /*
1777                          * Make the parent shadow the next object in the
1778                          * chain.  Deallocating backing_object will not remove
1779                          * it, since its reference count is at least 2.
1780                          */
1781
1782                         LIST_REMOVE(object, shadow_list);
1783                         backing_object->shadow_count--;
1784                         backing_object->generation++;
1785
1786                         new_backing_object = backing_object->backing_object;
1787                         if ((object->backing_object = new_backing_object) != NULL) {
1788                                 vm_object_reference(new_backing_object);
1789                                 LIST_INSERT_HEAD(
1790                                     &new_backing_object->shadow_head,
1791                                     object,
1792                                     shadow_list
1793                                 );
1794                                 new_backing_object->shadow_count++;
1795                                 new_backing_object->generation++;
1796                                 object->backing_object_offset +=
1797                                         backing_object->backing_object_offset;
1798                         }
1799
1800                         /*
1801                          * Drop the reference count on backing_object. Since
1802                          * its ref_count was at least 2, it will not vanish;
1803                          * so we don't need to call vm_object_deallocate, but
1804                          * we do anyway.
1805                          */
1806                         vm_object_drop(backing_object);
1807                         vm_object_deallocate_locked(backing_object);
1808                         object_bypasses++;
1809                 }
1810
1811                 /*
1812                  * Try again with this object's new backing object.
1813                  */
1814         }
1815 }
1816
1817 /*
1818  * Removes all physical pages in the specified object range from the
1819  * object's list of pages.
1820  *
1821  * No requirements.
1822  */
1823 static int vm_object_page_remove_callback(vm_page_t p, void *data);
1824
1825 void
1826 vm_object_page_remove(vm_object_t object, vm_pindex_t start, vm_pindex_t end,
1827                       boolean_t clean_only)
1828 {
1829         struct rb_vm_page_scan_info info;
1830         int all;
1831
1832         /*
1833          * Degenerate cases and assertions
1834          */
1835         lwkt_gettoken(&vm_token);
1836         if (object == NULL ||
1837             (object->resident_page_count == 0 && object->swblock_count == 0)) {
1838                 lwkt_reltoken(&vm_token);
1839                 return;
1840         }
1841         KASSERT(object->type != OBJT_PHYS, 
1842                 ("attempt to remove pages from a physical object"));
1843
1844         /*
1845          * Indicate that paging is occuring on the object
1846          */
1847         crit_enter();
1848         vm_object_pip_add(object, 1);
1849
1850         /*
1851          * Figure out the actual removal range and whether we are removing
1852          * the entire contents of the object or not.  If removing the entire
1853          * contents, be sure to get all pages, even those that might be 
1854          * beyond the end of the object.
1855          */
1856         info.start_pindex = start;
1857         if (end == 0)
1858                 info.end_pindex = (vm_pindex_t)-1;
1859         else
1860                 info.end_pindex = end - 1;
1861         info.limit = clean_only;
1862         all = (start == 0 && info.end_pindex >= object->size - 1);
1863
1864         /*
1865          * Loop until we are sure we have gotten them all.
1866          */
1867         do {
1868                 info.error = 0;
1869                 vm_page_rb_tree_RB_SCAN(&object->rb_memq, rb_vm_page_scancmp,
1870                                         vm_object_page_remove_callback, &info);
1871         } while (info.error);
1872
1873         /*
1874          * Remove any related swap if throwing away pages, or for
1875          * non-swap objects (the swap is a clean copy in that case).
1876          */
1877         if (object->type != OBJT_SWAP || clean_only == FALSE) {
1878                 if (all)
1879                         swap_pager_freespace_all(object);
1880                 else
1881                         swap_pager_freespace(object, info.start_pindex,
1882                              info.end_pindex - info.start_pindex + 1);
1883         }
1884
1885         /*
1886          * Cleanup
1887          */
1888         vm_object_pip_wakeup(object);
1889         crit_exit();
1890         lwkt_reltoken(&vm_token);
1891 }
1892
1893 /*
1894  * The caller must hold vm_token.
1895  */
1896 static int
1897 vm_object_page_remove_callback(vm_page_t p, void *data)
1898 {
1899         struct rb_vm_page_scan_info *info = data;
1900
1901         /*
1902          * Wired pages cannot be destroyed, but they can be invalidated
1903          * and we do so if clean_only (limit) is not set.
1904          *
1905          * WARNING!  The page may be wired due to being part of a buffer
1906          *           cache buffer, and the buffer might be marked B_CACHE.
1907          *           This is fine as part of a truncation but VFSs must be
1908          *           sure to fix the buffer up when re-extending the file.
1909          */
1910         if (p->wire_count != 0) {
1911                 vm_page_protect(p, VM_PROT_NONE);
1912                 if (info->limit == 0)
1913                         p->valid = 0;
1914                 return(0);
1915         }
1916
1917         /*
1918          * The busy flags are only cleared at
1919          * interrupt -- minimize the spl transitions
1920          */
1921
1922         if (vm_page_sleep_busy(p, TRUE, "vmopar")) {
1923                 info->error = 1;
1924                 return(0);
1925         }
1926
1927         /*
1928          * limit is our clean_only flag.  If set and the page is dirty, do
1929          * not free it.  If set and the page is being held by someone, do
1930          * not free it.
1931          */
1932         if (info->limit && p->valid) {
1933                 vm_page_test_dirty(p);
1934                 if (p->valid & p->dirty)
1935                         return(0);
1936                 if (p->hold_count)
1937                         return(0);
1938         }
1939
1940         /*
1941          * Destroy the page
1942          */
1943         vm_page_busy(p);
1944         vm_page_protect(p, VM_PROT_NONE);
1945         vm_page_free(p);
1946         return(0);
1947 }
1948
1949 /*
1950  * Coalesces two objects backing up adjoining regions of memory into a
1951  * single object.
1952  *
1953  * returns TRUE if objects were combined.
1954  *
1955  * NOTE: Only works at the moment if the second object is NULL -
1956  *       if it's not, which object do we lock first?
1957  *
1958  * Parameters:
1959  *      prev_object     First object to coalesce
1960  *      prev_offset     Offset into prev_object
1961  *      next_object     Second object into coalesce
1962  *      next_offset     Offset into next_object
1963  *
1964  *      prev_size       Size of reference to prev_object
1965  *      next_size       Size of reference to next_object
1966  *
1967  * The caller must hold vm_token and vmobj_token.
1968  *
1969  * The caller does not need to hold (prev_object) but must have a stable
1970  * pointer to it (typically by holding the vm_map locked).
1971  */
1972 boolean_t
1973 vm_object_coalesce(vm_object_t prev_object, vm_pindex_t prev_pindex,
1974                    vm_size_t prev_size, vm_size_t next_size)
1975 {
1976         vm_pindex_t next_pindex;
1977
1978         ASSERT_LWKT_TOKEN_HELD(&vm_token);
1979         ASSERT_LWKT_TOKEN_HELD(&vmobj_token);
1980
1981         if (prev_object == NULL) {
1982                 return (TRUE);
1983         }
1984
1985         vm_object_hold(prev_object);
1986
1987         if (prev_object->type != OBJT_DEFAULT &&
1988             prev_object->type != OBJT_SWAP) {
1989                 vm_object_drop(prev_object);
1990                 return (FALSE);
1991         }
1992
1993         /*
1994          * Try to collapse the object first
1995          */
1996         vm_object_collapse(prev_object);
1997
1998         /*
1999          * Can't coalesce if: . more than one reference . paged out . shadows
2000          * another object . has a copy elsewhere (any of which mean that the
2001          * pages not mapped to prev_entry may be in use anyway)
2002          */
2003
2004         if (prev_object->backing_object != NULL) {
2005                 vm_object_drop(prev_object);
2006                 return (FALSE);
2007         }
2008
2009         prev_size >>= PAGE_SHIFT;
2010         next_size >>= PAGE_SHIFT;
2011         next_pindex = prev_pindex + prev_size;
2012
2013         if ((prev_object->ref_count > 1) &&
2014             (prev_object->size != next_pindex)) {
2015                 vm_object_drop(prev_object);
2016                 return (FALSE);
2017         }
2018
2019         /*
2020          * Remove any pages that may still be in the object from a previous
2021          * deallocation.
2022          */
2023         if (next_pindex < prev_object->size) {
2024                 vm_object_page_remove(prev_object,
2025                                       next_pindex,
2026                                       next_pindex + next_size, FALSE);
2027                 if (prev_object->type == OBJT_SWAP)
2028                         swap_pager_freespace(prev_object,
2029                                              next_pindex, next_size);
2030         }
2031
2032         /*
2033          * Extend the object if necessary.
2034          */
2035         if (next_pindex + next_size > prev_object->size)
2036                 prev_object->size = next_pindex + next_size;
2037
2038         vm_object_drop(prev_object);
2039         return (TRUE);
2040 }
2041
2042 /*
2043  * Make the object writable and flag is being possibly dirty.
2044  *
2045  * No requirements.
2046  */
2047 void
2048 vm_object_set_writeable_dirty(vm_object_t object)
2049 {
2050         struct vnode *vp;
2051
2052         lwkt_gettoken(&vm_token);
2053         vm_object_set_flag(object, OBJ_WRITEABLE|OBJ_MIGHTBEDIRTY);
2054         if (object->type == OBJT_VNODE &&
2055             (vp = (struct vnode *)object->handle) != NULL) {
2056                 if ((vp->v_flag & VOBJDIRTY) == 0) {
2057                         vsetflags(vp, VOBJDIRTY);
2058                 }
2059         }
2060         lwkt_reltoken(&vm_token);
2061 }
2062
2063 #include "opt_ddb.h"
2064 #ifdef DDB
2065 #include <sys/kernel.h>
2066
2067 #include <sys/cons.h>
2068
2069 #include <ddb/ddb.h>
2070
2071 static int      _vm_object_in_map (vm_map_t map, vm_object_t object,
2072                                        vm_map_entry_t entry);
2073 static int      vm_object_in_map (vm_object_t object);
2074
2075 /*
2076  * The caller must hold vm_token.
2077  */
2078 static int
2079 _vm_object_in_map(vm_map_t map, vm_object_t object, vm_map_entry_t entry)
2080 {
2081         vm_map_t tmpm;
2082         vm_map_entry_t tmpe;
2083         vm_object_t obj;
2084         int entcount;
2085
2086         if (map == 0)
2087                 return 0;
2088         if (entry == 0) {
2089                 tmpe = map->header.next;
2090                 entcount = map->nentries;
2091                 while (entcount-- && (tmpe != &map->header)) {
2092                         if( _vm_object_in_map(map, object, tmpe)) {
2093                                 return 1;
2094                         }
2095                         tmpe = tmpe->next;
2096                 }
2097                 return (0);
2098         }
2099         switch(entry->maptype) {
2100         case VM_MAPTYPE_SUBMAP:
2101                 tmpm = entry->object.sub_map;
2102                 tmpe = tmpm->header.next;
2103                 entcount = tmpm->nentries;
2104                 while (entcount-- && tmpe != &tmpm->header) {
2105                         if( _vm_object_in_map(tmpm, object, tmpe)) {
2106                                 return 1;
2107                         }
2108                         tmpe = tmpe->next;
2109                 }
2110                 break;
2111         case VM_MAPTYPE_NORMAL:
2112         case VM_MAPTYPE_VPAGETABLE:
2113                 obj = entry->object.vm_object;
2114                 while (obj) {
2115                         if (obj == object)
2116                                 return 1;
2117                         obj = obj->backing_object;
2118                 }
2119                 break;
2120         default:
2121                 break;
2122         }
2123         return 0;
2124 }
2125
2126 static int vm_object_in_map_callback(struct proc *p, void *data);
2127
2128 struct vm_object_in_map_info {
2129         vm_object_t object;
2130         int rv;
2131 };
2132
2133 /*
2134  * Debugging only
2135  */
2136 static int
2137 vm_object_in_map(vm_object_t object)
2138 {
2139         struct vm_object_in_map_info info;
2140
2141         info.rv = 0;
2142         info.object = object;
2143
2144         allproc_scan(vm_object_in_map_callback, &info);
2145         if (info.rv)
2146                 return 1;
2147         if( _vm_object_in_map(&kernel_map, object, 0))
2148                 return 1;
2149         if( _vm_object_in_map(&pager_map, object, 0))
2150                 return 1;
2151         if( _vm_object_in_map(&buffer_map, object, 0))
2152                 return 1;
2153         return 0;
2154 }
2155
2156 /*
2157  * Debugging only
2158  */
2159 static int
2160 vm_object_in_map_callback(struct proc *p, void *data)
2161 {
2162         struct vm_object_in_map_info *info = data;
2163
2164         if (p->p_vmspace) {
2165                 if (_vm_object_in_map(&p->p_vmspace->vm_map, info->object, 0)) {
2166                         info->rv = 1;
2167                         return -1;
2168                 }
2169         }
2170         return (0);
2171 }
2172
2173 DB_SHOW_COMMAND(vmochk, vm_object_check)
2174 {
2175         vm_object_t object;
2176
2177         /*
2178          * make sure that internal objs are in a map somewhere
2179          * and none have zero ref counts.
2180          */
2181         for (object = TAILQ_FIRST(&vm_object_list);
2182                         object != NULL;
2183                         object = TAILQ_NEXT(object, object_list)) {
2184                 if (object->type == OBJT_MARKER)
2185                         continue;
2186                 if (object->handle == NULL &&
2187                     (object->type == OBJT_DEFAULT || object->type == OBJT_SWAP)) {
2188                         if (object->ref_count == 0) {
2189                                 db_printf("vmochk: internal obj has zero ref count: %ld\n",
2190                                         (long)object->size);
2191                         }
2192                         if (!vm_object_in_map(object)) {
2193                                 db_printf(
2194                         "vmochk: internal obj is not in a map: "
2195                         "ref: %d, size: %lu: 0x%lx, backing_object: %p\n",
2196                                     object->ref_count, (u_long)object->size, 
2197                                     (u_long)object->size,
2198                                     (void *)object->backing_object);
2199                         }
2200                 }
2201         }
2202 }
2203
2204 /*
2205  * Debugging only
2206  */
2207 DB_SHOW_COMMAND(object, vm_object_print_static)
2208 {
2209         /* XXX convert args. */
2210         vm_object_t object = (vm_object_t)addr;
2211         boolean_t full = have_addr;
2212
2213         vm_page_t p;
2214
2215         /* XXX count is an (unused) arg.  Avoid shadowing it. */
2216 #define count   was_count
2217
2218         int count;
2219
2220         if (object == NULL)
2221                 return;
2222
2223         db_iprintf(
2224             "Object %p: type=%d, size=0x%lx, res=%d, ref=%d, flags=0x%x\n",
2225             object, (int)object->type, (u_long)object->size,
2226             object->resident_page_count, object->ref_count, object->flags);
2227         /*
2228          * XXX no %qd in kernel.  Truncate object->backing_object_offset.
2229          */
2230         db_iprintf(" sref=%d, backing_object(%d)=(%p)+0x%lx\n",
2231             object->shadow_count, 
2232             object->backing_object ? object->backing_object->ref_count : 0,
2233             object->backing_object, (long)object->backing_object_offset);
2234
2235         if (!full)
2236                 return;
2237
2238         db_indent += 2;
2239         count = 0;
2240         RB_FOREACH(p, vm_page_rb_tree, &object->rb_memq) {
2241                 if (count == 0)
2242                         db_iprintf("memory:=");
2243                 else if (count == 6) {
2244                         db_printf("\n");
2245                         db_iprintf(" ...");
2246                         count = 0;
2247                 } else
2248                         db_printf(",");
2249                 count++;
2250
2251                 db_printf("(off=0x%lx,page=0x%lx)",
2252                     (u_long) p->pindex, (u_long) VM_PAGE_TO_PHYS(p));
2253         }
2254         if (count != 0)
2255                 db_printf("\n");
2256         db_indent -= 2;
2257 }
2258
2259 /* XXX. */
2260 #undef count
2261
2262 /*
2263  * XXX need this non-static entry for calling from vm_map_print.
2264  *
2265  * Debugging only
2266  */
2267 void
2268 vm_object_print(/* db_expr_t */ long addr,
2269                 boolean_t have_addr,
2270                 /* db_expr_t */ long count,
2271                 char *modif)
2272 {
2273         vm_object_print_static(addr, have_addr, count, modif);
2274 }
2275
2276 /*
2277  * Debugging only
2278  */
2279 DB_SHOW_COMMAND(vmopag, vm_object_print_pages)
2280 {
2281         vm_object_t object;
2282         int nl = 0;
2283         int c;
2284         for (object = TAILQ_FIRST(&vm_object_list);
2285                         object != NULL;
2286                         object = TAILQ_NEXT(object, object_list)) {
2287                 vm_pindex_t idx, fidx;
2288                 vm_pindex_t osize;
2289                 vm_paddr_t pa = -1, padiff;
2290                 int rcount;
2291                 vm_page_t m;
2292
2293                 if (object->type == OBJT_MARKER)
2294                         continue;
2295                 db_printf("new object: %p\n", (void *)object);
2296                 if ( nl > 18) {
2297                         c = cngetc();
2298                         if (c != ' ')
2299                                 return;
2300                         nl = 0;
2301                 }
2302                 nl++;
2303                 rcount = 0;
2304                 fidx = 0;
2305                 osize = object->size;
2306                 if (osize > 128)
2307                         osize = 128;
2308                 for (idx = 0; idx < osize; idx++) {
2309                         m = vm_page_lookup(object, idx);
2310                         if (m == NULL) {
2311                                 if (rcount) {
2312                                         db_printf(" index(%ld)run(%d)pa(0x%lx)\n",
2313                                                 (long)fidx, rcount, (long)pa);
2314                                         if ( nl > 18) {
2315                                                 c = cngetc();
2316                                                 if (c != ' ')
2317                                                         return;
2318                                                 nl = 0;
2319                                         }
2320                                         nl++;
2321                                         rcount = 0;
2322                                 }
2323                                 continue;
2324                         }
2325
2326                                 
2327                         if (rcount &&
2328                                 (VM_PAGE_TO_PHYS(m) == pa + rcount * PAGE_SIZE)) {
2329                                 ++rcount;
2330                                 continue;
2331                         }
2332                         if (rcount) {
2333                                 padiff = pa + rcount * PAGE_SIZE - VM_PAGE_TO_PHYS(m);
2334                                 padiff >>= PAGE_SHIFT;
2335                                 padiff &= PQ_L2_MASK;
2336                                 if (padiff == 0) {
2337                                         pa = VM_PAGE_TO_PHYS(m) - rcount * PAGE_SIZE;
2338                                         ++rcount;
2339                                         continue;
2340                                 }
2341                                 db_printf(" index(%ld)run(%d)pa(0x%lx)",
2342                                         (long)fidx, rcount, (long)pa);
2343                                 db_printf("pd(%ld)\n", (long)padiff);
2344                                 if ( nl > 18) {
2345                                         c = cngetc();
2346                                         if (c != ' ')
2347                                                 return;
2348                                         nl = 0;
2349                                 }
2350                                 nl++;
2351                         }
2352                         fidx = idx;
2353                         pa = VM_PAGE_TO_PHYS(m);
2354                         rcount = 1;
2355                 }
2356                 if (rcount) {
2357                         db_printf(" index(%ld)run(%d)pa(0x%lx)\n",
2358                                 (long)fidx, rcount, (long)pa);
2359                         if ( nl > 18) {
2360                                 c = cngetc();
2361                                 if (c != ' ')
2362                                         return;
2363                                 nl = 0;
2364                         }
2365                         nl++;
2366                 }
2367         }
2368 }
2369 #endif /* DDB */