xref: /dflybsd-src/sys/vm/vm_object.c (revision 3bcb6e5e80cb05cecaddd866d4b15b428334b9aa)
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 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 		 * We have to decrement object->ref_count before potentially
515 		 * collapsing the first shadow object or the collapse code
516 		 * will not be able to handle the degenerate case.
517 		 */
518 		if (object->ref_count == 2 && object->shadow_count == 1) {
519 			object->ref_count--;
520 			if (object->handle == NULL &&
521 			    (object->type == OBJT_DEFAULT ||
522 			     object->type == OBJT_SWAP)) {
523 				temp = LIST_FIRST(&object->shadow_head);
524 				KASSERT(temp != NULL,
525 					("vm_object_deallocate: ref_count: "
526 					"%d, shadow_count: %d",
527 					object->ref_count,
528 					object->shadow_count));
529 				lwkt_reltoken(&vm_token);
530 				vm_object_lock(temp);
531 
532 				if ((temp->handle == NULL) &&
533 				    (temp->type == OBJT_DEFAULT ||
534 				     temp->type == OBJT_SWAP)) {
535 					/*
536 					 * Special case, must handle ref_count
537 					 * manually to avoid recursion.
538 					 */
539 					temp->ref_count++;
540 					vm_object_lock_swap();
541 
542 					while (
543 						temp->paging_in_progress ||
544 						object->paging_in_progress
545 					) {
546 						vm_object_pip_wait(temp,
547 								   "objde1");
548 						vm_object_pip_wait(object,
549 								   "objde2");
550 					}
551 
552 					if (temp->ref_count == 1) {
553 						temp->ref_count--;
554 						vm_object_unlock(object);
555 						object = temp;
556 						goto doterm;
557 					}
558 
559 					lwkt_gettoken(&vm_token);
560 					vm_object_collapse(temp);
561 					lwkt_reltoken(&vm_token);
562 					vm_object_unlock(object);
563 					object = temp;
564 					continue;
565 				}
566 				vm_object_unlock(temp);
567 			} else {
568 				lwkt_reltoken(&vm_token);
569 			}
570 			vm_object_unlock(object);
571 			break;
572 		}
573 
574 		/*
575 		 * Normal dereferencing path
576 		 */
577 		object->ref_count--;
578 		if (object->ref_count != 0) {
579 			lwkt_reltoken(&vm_token);
580 			vm_object_unlock(object);
581 			break;
582 		}
583 
584 		/*
585 		 * Termination path
586 		 *
587 		 * We may have to loop to resolve races if we block getting
588 		 * temp's lock.  If temp is non NULL we have to swap the
589 		 * lock order so the original object lock as at the top
590 		 * of the lock heap.
591 		 */
592 		lwkt_reltoken(&vm_token);
593 doterm:
594 		while ((temp = object->backing_object) != NULL) {
595 			vm_object_lock(temp);
596 			if (temp == object->backing_object)
597 				break;
598 			vm_object_unlock(temp);
599 		}
600 		if (temp) {
601 			LIST_REMOVE(object, shadow_list);
602 			temp->shadow_count--;
603 			temp->generation++;
604 			object->backing_object = NULL;
605 			vm_object_lock_swap();
606 		}
607 
608 		/*
609 		 * Don't double-terminate, we could be in a termination
610 		 * recursion due to the terminate having to sync data
611 		 * to disk.
612 		 */
613 		if ((object->flags & OBJ_DEAD) == 0) {
614 			vm_object_terminate(object);
615 			/* termination ate the object lock */
616 		} else {
617 			vm_object_unlock(object);
618 		}
619 		object = temp;
620 	}
621 }
622 
623 /*
624  * Destroy the specified object, freeing up related resources.
625  *
626  * The object must have zero references.
627  *
628  * The caller must be holding vmobj_token and properly interlock with
629  * OBJ_DEAD (at the moment).
630  *
631  * The caller must have locked the object only, and not be holding it.
632  * This function will eat the caller's lock on the object.
633  */
634 static int vm_object_terminate_callback(vm_page_t p, void *data);
635 
636 void
637 vm_object_terminate(vm_object_t object)
638 {
639 	/*
640 	 * Make sure no one uses us.  Once we set OBJ_DEAD we should be
641 	 * able to safely block.
642 	 */
643 	KKASSERT((object->flags & OBJ_DEAD) == 0);
644 	ASSERT_LWKT_TOKEN_HELD(&vmobj_token);
645 	vm_object_set_flag(object, OBJ_DEAD);
646 
647 	/*
648 	 * Wait for the pageout daemon to be done with the object
649 	 */
650 	vm_object_pip_wait(object, "objtrm1");
651 
652 	KASSERT(!object->paging_in_progress,
653 		("vm_object_terminate: pageout in progress"));
654 
655 	/*
656 	 * Clean and free the pages, as appropriate. All references to the
657 	 * object are gone, so we don't need to lock it.
658 	 */
659 	if (object->type == OBJT_VNODE) {
660 		struct vnode *vp;
661 
662 		/*
663 		 * Clean pages and flush buffers.
664 		 */
665 		vm_object_page_clean(object, 0, 0, OBJPC_SYNC);
666 
667 		vp = (struct vnode *) object->handle;
668 		vinvalbuf(vp, V_SAVE, 0, 0);
669 	}
670 
671 	/*
672 	 * Wait for any I/O to complete, after which there had better not
673 	 * be any references left on the object.
674 	 */
675 	vm_object_pip_wait(object, "objtrm2");
676 
677 	if (object->ref_count != 0) {
678 		panic("vm_object_terminate: object with references, "
679 		      "ref_count=%d", object->ref_count);
680 	}
681 
682 	/*
683 	 * Now free any remaining pages. For internal objects, this also
684 	 * removes them from paging queues. Don't free wired pages, just
685 	 * remove them from the object.
686 	 */
687 	lwkt_gettoken(&vm_token);
688 	vm_page_rb_tree_RB_SCAN(&object->rb_memq, NULL,
689 				vm_object_terminate_callback, NULL);
690 	lwkt_reltoken(&vm_token);
691 
692 	/*
693 	 * Let the pager know object is dead.
694 	 */
695 	vm_pager_deallocate(object);
696 
697 	/*
698 	 * Wait for the object hold count to hit zero, clean out pages as
699 	 * we go.
700 	 */
701 	lwkt_gettoken(&vm_token);
702 	for (;;) {
703 		vm_object_hold_wait(object);
704 		if (RB_ROOT(&object->rb_memq) == NULL)
705 			break;
706 		kprintf("vm_object_terminate: Warning, object %p "
707 			"still has %d pages\n",
708 			object, object->resident_page_count);
709 		vm_page_rb_tree_RB_SCAN(&object->rb_memq, NULL,
710 					vm_object_terminate_callback, NULL);
711 	}
712 	lwkt_reltoken(&vm_token);
713 
714 	/*
715 	 * There had better not be any pages left
716 	 */
717 	KKASSERT(object->resident_page_count == 0);
718 
719 	/*
720 	 * Remove the object from the global object list.
721 	 *
722 	 * (we are holding vmobj_token)
723 	 */
724 	TAILQ_REMOVE(&vm_object_list, object, object_list);
725 	vm_object_count--;
726 	vm_object_dead_wakeup(object);
727 	vm_object_unlock(object);
728 
729 	if (object->ref_count != 0) {
730 		panic("vm_object_terminate2: object with references, "
731 		      "ref_count=%d", object->ref_count);
732 	}
733 
734 	/*
735 	 * Free the space for the object.
736 	 */
737 	zfree(obj_zone, object);
738 }
739 
740 /*
741  * The caller must hold vm_token.
742  */
743 static int
744 vm_object_terminate_callback(vm_page_t p, void *data __unused)
745 {
746 	if (p->busy || (p->flags & PG_BUSY))
747 		panic("vm_object_terminate: freeing busy page %p", p);
748 	if (p->wire_count == 0) {
749 		vm_page_busy(p);
750 		vm_page_free(p);
751 		mycpu->gd_cnt.v_pfree++;
752 	} else {
753 		if (p->queue != PQ_NONE)
754 			kprintf("vm_object_terminate: Warning: Encountered wired page %p on queue %d\n", p, p->queue);
755 		vm_page_busy(p);
756 		vm_page_remove(p);
757 		vm_page_wakeup(p);
758 	}
759 	return(0);
760 }
761 
762 /*
763  * The object is dead but still has an object<->pager association.  Sleep
764  * and return.  The caller typically retests the association in a loop.
765  *
766  * Must be called with the vmobj_token held.
767  */
768 void
769 vm_object_dead_sleep(vm_object_t object, const char *wmesg)
770 {
771 	ASSERT_LWKT_TOKEN_HELD(&vmobj_token);
772 	if (object->handle) {
773 		vm_object_set_flag(object, OBJ_DEADWNT);
774 		tsleep(object, 0, wmesg, 0);
775 		/* object may be invalid after this point */
776 	}
777 }
778 
779 /*
780  * Wakeup anyone waiting for the object<->pager disassociation on
781  * a dead object.
782  *
783  * Must be called with the vmobj_token held.
784  */
785 void
786 vm_object_dead_wakeup(vm_object_t object)
787 {
788 	ASSERT_LWKT_TOKEN_HELD(&vmobj_token);
789 	if (object->flags & OBJ_DEADWNT) {
790 		vm_object_clear_flag(object, OBJ_DEADWNT);
791 		wakeup(object);
792 	}
793 }
794 
795 /*
796  * Clean all dirty pages in the specified range of object.  Leaves page
797  * on whatever queue it is currently on.   If NOSYNC is set then do not
798  * write out pages with PG_NOSYNC set (originally comes from MAP_NOSYNC),
799  * leaving the object dirty.
800  *
801  * When stuffing pages asynchronously, allow clustering.  XXX we need a
802  * synchronous clustering mode implementation.
803  *
804  * Odd semantics: if start == end, we clean everything.
805  *
806  * The object must be locked? XXX
807  */
808 static int vm_object_page_clean_pass1(struct vm_page *p, void *data);
809 static int vm_object_page_clean_pass2(struct vm_page *p, void *data);
810 
811 void
812 vm_object_page_clean(vm_object_t object, vm_pindex_t start, vm_pindex_t end,
813 		     int flags)
814 {
815 	struct rb_vm_page_scan_info info;
816 	struct vnode *vp;
817 	int wholescan;
818 	int pagerflags;
819 	int curgeneration;
820 
821 	vm_object_hold(object);
822 	if (object->type != OBJT_VNODE ||
823 	    (object->flags & OBJ_MIGHTBEDIRTY) == 0) {
824 		vm_object_drop(object);
825 		return;
826 	}
827 
828 	pagerflags = (flags & (OBJPC_SYNC | OBJPC_INVAL)) ?
829 			VM_PAGER_PUT_SYNC : VM_PAGER_CLUSTER_OK;
830 	pagerflags |= (flags & OBJPC_INVAL) ? VM_PAGER_PUT_INVAL : 0;
831 
832 	vp = object->handle;
833 
834 	/*
835 	 * Interlock other major object operations.  This allows us to
836 	 * temporarily clear OBJ_WRITEABLE and OBJ_MIGHTBEDIRTY.
837 	 */
838 	crit_enter();
839 	vm_object_set_flag(object, OBJ_CLEANING);
840 
841 	/*
842 	 * Handle 'entire object' case
843 	 */
844 	info.start_pindex = start;
845 	if (end == 0) {
846 		info.end_pindex = object->size - 1;
847 	} else {
848 		info.end_pindex = end - 1;
849 	}
850 	wholescan = (start == 0 && info.end_pindex == object->size - 1);
851 	info.limit = flags;
852 	info.pagerflags = pagerflags;
853 	info.object = object;
854 
855 	/*
856 	 * If cleaning the entire object do a pass to mark the pages read-only.
857 	 * If everything worked out ok, clear OBJ_WRITEABLE and
858 	 * OBJ_MIGHTBEDIRTY.
859 	 */
860 	if (wholescan) {
861 		info.error = 0;
862 		lwkt_gettoken(&vm_token);
863 		vm_page_rb_tree_RB_SCAN(&object->rb_memq, rb_vm_page_scancmp,
864 					vm_object_page_clean_pass1, &info);
865 		lwkt_reltoken(&vm_token);
866 		if (info.error == 0) {
867 			vm_object_clear_flag(object,
868 					     OBJ_WRITEABLE|OBJ_MIGHTBEDIRTY);
869 			if (object->type == OBJT_VNODE &&
870 			    (vp = (struct vnode *)object->handle) != NULL) {
871 				if (vp->v_flag & VOBJDIRTY)
872 					vclrflags(vp, VOBJDIRTY);
873 			}
874 		}
875 	}
876 
877 	/*
878 	 * Do a pass to clean all the dirty pages we find.
879 	 */
880 	do {
881 		info.error = 0;
882 		curgeneration = object->generation;
883 		lwkt_gettoken(&vm_token);
884 		vm_page_rb_tree_RB_SCAN(&object->rb_memq, rb_vm_page_scancmp,
885 					vm_object_page_clean_pass2, &info);
886 		lwkt_reltoken(&vm_token);
887 	} while (info.error || curgeneration != object->generation);
888 
889 	vm_object_clear_flag(object, OBJ_CLEANING);
890 	crit_exit();
891 	vm_object_drop(object);
892 }
893 
894 /*
895  * The caller must hold vm_token.
896  */
897 static
898 int
899 vm_object_page_clean_pass1(struct vm_page *p, void *data)
900 {
901 	struct rb_vm_page_scan_info *info = data;
902 
903 	vm_page_flag_set(p, PG_CLEANCHK);
904 	if ((info->limit & OBJPC_NOSYNC) && (p->flags & PG_NOSYNC))
905 		info->error = 1;
906 	else
907 		vm_page_protect(p, VM_PROT_READ);	/* must not block */
908 	return(0);
909 }
910 
911 /*
912  * The caller must hold vm_token.
913  */
914 static
915 int
916 vm_object_page_clean_pass2(struct vm_page *p, void *data)
917 {
918 	struct rb_vm_page_scan_info *info = data;
919 	int n;
920 
921 	/*
922 	 * Do not mess with pages that were inserted after we started
923 	 * the cleaning pass.
924 	 */
925 	if ((p->flags & PG_CLEANCHK) == 0)
926 		return(0);
927 
928 	/*
929 	 * Before wasting time traversing the pmaps, check for trivial
930 	 * cases where the page cannot be dirty.
931 	 */
932 	if (p->valid == 0 || (p->queue - p->pc) == PQ_CACHE) {
933 		KKASSERT((p->dirty & p->valid) == 0);
934 		return(0);
935 	}
936 
937 	/*
938 	 * Check whether the page is dirty or not.  The page has been set
939 	 * to be read-only so the check will not race a user dirtying the
940 	 * page.
941 	 */
942 	vm_page_test_dirty(p);
943 	if ((p->dirty & p->valid) == 0) {
944 		vm_page_flag_clear(p, PG_CLEANCHK);
945 		return(0);
946 	}
947 
948 	/*
949 	 * If we have been asked to skip nosync pages and this is a
950 	 * nosync page, skip it.  Note that the object flags were
951 	 * not cleared in this case (because pass1 will have returned an
952 	 * error), so we do not have to set them.
953 	 */
954 	if ((info->limit & OBJPC_NOSYNC) && (p->flags & PG_NOSYNC)) {
955 		vm_page_flag_clear(p, PG_CLEANCHK);
956 		return(0);
957 	}
958 
959 	/*
960 	 * Flush as many pages as we can.  PG_CLEANCHK will be cleared on
961 	 * the pages that get successfully flushed.  Set info->error if
962 	 * we raced an object modification.
963 	 */
964 	n = vm_object_page_collect_flush(info->object, p, info->pagerflags);
965 	if (n == 0)
966 		info->error = 1;
967 	return(0);
968 }
969 
970 /*
971  * Collect the specified page and nearby pages and flush them out.
972  * The number of pages flushed is returned.
973  *
974  * The caller must hold vm_token.
975  */
976 static int
977 vm_object_page_collect_flush(vm_object_t object, vm_page_t p, int pagerflags)
978 {
979 	int runlen;
980 	int maxf;
981 	int chkb;
982 	int maxb;
983 	int i;
984 	int curgeneration;
985 	vm_pindex_t pi;
986 	vm_page_t maf[vm_pageout_page_count];
987 	vm_page_t mab[vm_pageout_page_count];
988 	vm_page_t ma[vm_pageout_page_count];
989 
990 	curgeneration = object->generation;
991 
992 	pi = p->pindex;
993 	while (vm_page_sleep_busy(p, TRUE, "vpcwai")) {
994 		if (object->generation != curgeneration) {
995 			return(0);
996 		}
997 	}
998 	KKASSERT(p->object == object && p->pindex == pi);
999 
1000 	maxf = 0;
1001 	for(i = 1; i < vm_pageout_page_count; i++) {
1002 		vm_page_t tp;
1003 
1004 		if ((tp = vm_page_lookup(object, pi + i)) != NULL) {
1005 			if ((tp->flags & PG_BUSY) ||
1006 				((pagerflags & VM_PAGER_IGNORE_CLEANCHK) == 0 &&
1007 				 (tp->flags & PG_CLEANCHK) == 0) ||
1008 				(tp->busy != 0))
1009 				break;
1010 			if((tp->queue - tp->pc) == PQ_CACHE) {
1011 				vm_page_flag_clear(tp, PG_CLEANCHK);
1012 				break;
1013 			}
1014 			vm_page_test_dirty(tp);
1015 			if ((tp->dirty & tp->valid) == 0) {
1016 				vm_page_flag_clear(tp, PG_CLEANCHK);
1017 				break;
1018 			}
1019 			maf[ i - 1 ] = tp;
1020 			maxf++;
1021 			continue;
1022 		}
1023 		break;
1024 	}
1025 
1026 	maxb = 0;
1027 	chkb = vm_pageout_page_count -  maxf;
1028 	if (chkb) {
1029 		for(i = 1; i < chkb;i++) {
1030 			vm_page_t tp;
1031 
1032 			if ((tp = vm_page_lookup(object, pi - i)) != NULL) {
1033 				if ((tp->flags & PG_BUSY) ||
1034 					((pagerflags & VM_PAGER_IGNORE_CLEANCHK) == 0 &&
1035 					 (tp->flags & PG_CLEANCHK) == 0) ||
1036 					(tp->busy != 0))
1037 					break;
1038 				if((tp->queue - tp->pc) == PQ_CACHE) {
1039 					vm_page_flag_clear(tp, PG_CLEANCHK);
1040 					break;
1041 				}
1042 				vm_page_test_dirty(tp);
1043 				if ((tp->dirty & tp->valid) == 0) {
1044 					vm_page_flag_clear(tp, PG_CLEANCHK);
1045 					break;
1046 				}
1047 				mab[ i - 1 ] = tp;
1048 				maxb++;
1049 				continue;
1050 			}
1051 			break;
1052 		}
1053 	}
1054 
1055 	for(i = 0; i < maxb; i++) {
1056 		int index = (maxb - i) - 1;
1057 		ma[index] = mab[i];
1058 		vm_page_flag_clear(ma[index], PG_CLEANCHK);
1059 	}
1060 	vm_page_flag_clear(p, PG_CLEANCHK);
1061 	ma[maxb] = p;
1062 	for(i = 0; i < maxf; i++) {
1063 		int index = (maxb + i) + 1;
1064 		ma[index] = maf[i];
1065 		vm_page_flag_clear(ma[index], PG_CLEANCHK);
1066 	}
1067 	runlen = maxb + maxf + 1;
1068 
1069 	vm_pageout_flush(ma, runlen, pagerflags);
1070 	for (i = 0; i < runlen; i++) {
1071 		if (ma[i]->valid & ma[i]->dirty) {
1072 			vm_page_protect(ma[i], VM_PROT_READ);
1073 			vm_page_flag_set(ma[i], PG_CLEANCHK);
1074 
1075 			/*
1076 			 * maxf will end up being the actual number of pages
1077 			 * we wrote out contiguously, non-inclusive of the
1078 			 * first page.  We do not count look-behind pages.
1079 			 */
1080 			if (i >= maxb + 1 && (maxf > i - maxb - 1))
1081 				maxf = i - maxb - 1;
1082 		}
1083 	}
1084 	return(maxf + 1);
1085 }
1086 
1087 /*
1088  * Same as vm_object_pmap_copy, except range checking really
1089  * works, and is meant for small sections of an object.
1090  *
1091  * This code protects resident pages by making them read-only
1092  * and is typically called on a fork or split when a page
1093  * is converted to copy-on-write.
1094  *
1095  * NOTE: If the page is already at VM_PROT_NONE, calling
1096  * vm_page_protect will have no effect.
1097  */
1098 void
1099 vm_object_pmap_copy_1(vm_object_t object, vm_pindex_t start, vm_pindex_t end)
1100 {
1101 	vm_pindex_t idx;
1102 	vm_page_t p;
1103 
1104 	if (object == NULL || (object->flags & OBJ_WRITEABLE) == 0)
1105 		return;
1106 
1107 	/*
1108 	 * spl protection needed to prevent races between the lookup,
1109 	 * an interrupt unbusy/free, and our protect call.
1110 	 */
1111 	crit_enter();
1112 	lwkt_gettoken(&vm_token);
1113 	for (idx = start; idx < end; idx++) {
1114 		p = vm_page_lookup(object, idx);
1115 		if (p == NULL)
1116 			continue;
1117 		vm_page_protect(p, VM_PROT_READ);
1118 	}
1119 	lwkt_reltoken(&vm_token);
1120 	crit_exit();
1121 }
1122 
1123 /*
1124  * Removes all physical pages in the specified object range from all
1125  * physical maps.
1126  *
1127  * The object must *not* be locked.
1128  */
1129 
1130 static int vm_object_pmap_remove_callback(vm_page_t p, void *data);
1131 
1132 void
1133 vm_object_pmap_remove(vm_object_t object, vm_pindex_t start, vm_pindex_t end)
1134 {
1135 	struct rb_vm_page_scan_info info;
1136 
1137 	if (object == NULL)
1138 		return;
1139 	info.start_pindex = start;
1140 	info.end_pindex = end - 1;
1141 
1142 	crit_enter();
1143 	lwkt_gettoken(&vm_token);
1144 	vm_page_rb_tree_RB_SCAN(&object->rb_memq, rb_vm_page_scancmp,
1145 				vm_object_pmap_remove_callback, &info);
1146 	if (start == 0 && end == object->size)
1147 		vm_object_clear_flag(object, OBJ_WRITEABLE);
1148 	lwkt_reltoken(&vm_token);
1149 	crit_exit();
1150 }
1151 
1152 /*
1153  * The caller must hold vm_token.
1154  */
1155 static int
1156 vm_object_pmap_remove_callback(vm_page_t p, void *data __unused)
1157 {
1158 	vm_page_protect(p, VM_PROT_NONE);
1159 	return(0);
1160 }
1161 
1162 /*
1163  * Implements the madvise function at the object/page level.
1164  *
1165  * MADV_WILLNEED	(any object)
1166  *
1167  *	Activate the specified pages if they are resident.
1168  *
1169  * MADV_DONTNEED	(any object)
1170  *
1171  *	Deactivate the specified pages if they are resident.
1172  *
1173  * MADV_FREE	(OBJT_DEFAULT/OBJT_SWAP objects, OBJ_ONEMAPPING only)
1174  *
1175  *	Deactivate and clean the specified pages if they are
1176  *	resident.  This permits the process to reuse the pages
1177  *	without faulting or the kernel to reclaim the pages
1178  *	without I/O.
1179  *
1180  * No requirements.
1181  */
1182 void
1183 vm_object_madvise(vm_object_t object, vm_pindex_t pindex, int count, int advise)
1184 {
1185 	vm_pindex_t end, tpindex;
1186 	vm_object_t tobject;
1187 	vm_page_t m;
1188 
1189 	if (object == NULL)
1190 		return;
1191 
1192 	end = pindex + count;
1193 
1194 	lwkt_gettoken(&vm_token);
1195 
1196 	/*
1197 	 * Locate and adjust resident pages
1198 	 */
1199 	for (; pindex < end; pindex += 1) {
1200 relookup:
1201 		tobject = object;
1202 		tpindex = pindex;
1203 shadowlookup:
1204 		/*
1205 		 * MADV_FREE only operates on OBJT_DEFAULT or OBJT_SWAP pages
1206 		 * and those pages must be OBJ_ONEMAPPING.
1207 		 */
1208 		if (advise == MADV_FREE) {
1209 			if ((tobject->type != OBJT_DEFAULT &&
1210 			     tobject->type != OBJT_SWAP) ||
1211 			    (tobject->flags & OBJ_ONEMAPPING) == 0) {
1212 				continue;
1213 			}
1214 		}
1215 
1216 		/*
1217 		 * spl protection is required to avoid a race between the
1218 		 * lookup, an interrupt unbusy/free, and our busy check.
1219 		 */
1220 
1221 		crit_enter();
1222 		m = vm_page_lookup(tobject, tpindex);
1223 
1224 		if (m == NULL) {
1225 			/*
1226 			 * There may be swap even if there is no backing page
1227 			 */
1228 			if (advise == MADV_FREE && tobject->type == OBJT_SWAP)
1229 				swap_pager_freespace(tobject, tpindex, 1);
1230 
1231 			/*
1232 			 * next object
1233 			 */
1234 			crit_exit();
1235 			if (tobject->backing_object == NULL)
1236 				continue;
1237 			tpindex += OFF_TO_IDX(tobject->backing_object_offset);
1238 			tobject = tobject->backing_object;
1239 			goto shadowlookup;
1240 		}
1241 
1242 		/*
1243 		 * If the page is busy or not in a normal active state,
1244 		 * we skip it.  If the page is not managed there are no
1245 		 * page queues to mess with.  Things can break if we mess
1246 		 * with pages in any of the below states.
1247 		 */
1248 		if (
1249 		    m->hold_count ||
1250 		    m->wire_count ||
1251 		    (m->flags & PG_UNMANAGED) ||
1252 		    m->valid != VM_PAGE_BITS_ALL
1253 		) {
1254 			crit_exit();
1255 			continue;
1256 		}
1257 
1258  		if (vm_page_sleep_busy(m, TRUE, "madvpo")) {
1259 			crit_exit();
1260   			goto relookup;
1261 		}
1262 		vm_page_busy(m);
1263 		crit_exit();
1264 
1265 		/*
1266 		 * Theoretically once a page is known not to be busy, an
1267 		 * interrupt cannot come along and rip it out from under us.
1268 		 */
1269 
1270 		if (advise == MADV_WILLNEED) {
1271 			vm_page_activate(m);
1272 		} else if (advise == MADV_DONTNEED) {
1273 			vm_page_dontneed(m);
1274 		} else if (advise == MADV_FREE) {
1275 			/*
1276 			 * Mark the page clean.  This will allow the page
1277 			 * to be freed up by the system.  However, such pages
1278 			 * are often reused quickly by malloc()/free()
1279 			 * so we do not do anything that would cause
1280 			 * a page fault if we can help it.
1281 			 *
1282 			 * Specifically, we do not try to actually free
1283 			 * the page now nor do we try to put it in the
1284 			 * cache (which would cause a page fault on reuse).
1285 			 *
1286 			 * But we do make the page is freeable as we
1287 			 * can without actually taking the step of unmapping
1288 			 * it.
1289 			 */
1290 			pmap_clear_modify(m);
1291 			m->dirty = 0;
1292 			m->act_count = 0;
1293 			vm_page_dontneed(m);
1294 			if (tobject->type == OBJT_SWAP)
1295 				swap_pager_freespace(tobject, tpindex, 1);
1296 		}
1297 		vm_page_wakeup(m);
1298 	}
1299 	lwkt_reltoken(&vm_token);
1300 }
1301 
1302 /*
1303  * Create a new object which is backed by the specified existing object
1304  * range.  The source object reference is deallocated.
1305  *
1306  * The new object and offset into that object are returned in the source
1307  * parameters.
1308  *
1309  * No other requirements.
1310  */
1311 void
1312 vm_object_shadow(vm_object_t *object, vm_ooffset_t *offset, vm_size_t length)
1313 {
1314 	vm_object_t source;
1315 	vm_object_t result;
1316 
1317 	source = *object;
1318 
1319 	/*
1320 	 * Don't create the new object if the old object isn't shared.
1321 	 */
1322 	lwkt_gettoken(&vm_token);
1323 
1324 	if (source != NULL &&
1325 	    source->ref_count == 1 &&
1326 	    source->handle == NULL &&
1327 	    (source->type == OBJT_DEFAULT ||
1328 	     source->type == OBJT_SWAP)) {
1329 		lwkt_reltoken(&vm_token);
1330 		return;
1331 	}
1332 
1333 	/*
1334 	 * Allocate a new object with the given length
1335 	 */
1336 
1337 	if ((result = vm_object_allocate(OBJT_DEFAULT, length)) == NULL)
1338 		panic("vm_object_shadow: no object for shadowing");
1339 
1340 	/*
1341 	 * The new object shadows the source object, adding a reference to it.
1342 	 * Our caller changes his reference to point to the new object,
1343 	 * removing a reference to the source object.  Net result: no change
1344 	 * of reference count.
1345 	 *
1346 	 * Try to optimize the result object's page color when shadowing
1347 	 * in order to maintain page coloring consistency in the combined
1348 	 * shadowed object.
1349 	 */
1350 	result->backing_object = source;
1351 	if (source) {
1352 		LIST_INSERT_HEAD(&source->shadow_head, result, shadow_list);
1353 		source->shadow_count++;
1354 		source->generation++;
1355 		result->pg_color = (source->pg_color + OFF_TO_IDX(*offset)) & PQ_L2_MASK;
1356 	}
1357 
1358 	/*
1359 	 * Store the offset into the source object, and fix up the offset into
1360 	 * the new object.
1361 	 */
1362 	result->backing_object_offset = *offset;
1363 	lwkt_reltoken(&vm_token);
1364 
1365 	/*
1366 	 * Return the new things
1367 	 */
1368 	*offset = 0;
1369 	*object = result;
1370 }
1371 
1372 #define	OBSC_TEST_ALL_SHADOWED	0x0001
1373 #define	OBSC_COLLAPSE_NOWAIT	0x0002
1374 #define	OBSC_COLLAPSE_WAIT	0x0004
1375 
1376 static int vm_object_backing_scan_callback(vm_page_t p, void *data);
1377 
1378 /*
1379  * The caller must hold vm_token.
1380  */
1381 static __inline int
1382 vm_object_backing_scan(vm_object_t object, int op)
1383 {
1384 	struct rb_vm_page_scan_info info;
1385 	vm_object_t backing_object;
1386 
1387 	crit_enter();
1388 
1389 	backing_object = object->backing_object;
1390 	info.backing_offset_index = OFF_TO_IDX(object->backing_object_offset);
1391 
1392 	/*
1393 	 * Initial conditions
1394 	 */
1395 
1396 	if (op & OBSC_TEST_ALL_SHADOWED) {
1397 		/*
1398 		 * We do not want to have to test for the existence of
1399 		 * swap pages in the backing object.  XXX but with the
1400 		 * new swapper this would be pretty easy to do.
1401 		 *
1402 		 * XXX what about anonymous MAP_SHARED memory that hasn't
1403 		 * been ZFOD faulted yet?  If we do not test for this, the
1404 		 * shadow test may succeed! XXX
1405 		 */
1406 		if (backing_object->type != OBJT_DEFAULT) {
1407 			crit_exit();
1408 			return(0);
1409 		}
1410 	}
1411 	if (op & OBSC_COLLAPSE_WAIT) {
1412 		KKASSERT((backing_object->flags & OBJ_DEAD) == 0);
1413 		vm_object_set_flag(backing_object, OBJ_DEAD);
1414 	}
1415 
1416 	/*
1417 	 * Our scan.   We have to retry if a negative error code is returned,
1418 	 * otherwise 0 or 1 will be returned in info.error.  0 Indicates that
1419 	 * the scan had to be stopped because the parent does not completely
1420 	 * shadow the child.
1421 	 */
1422 	info.object = object;
1423 	info.backing_object = backing_object;
1424 	info.limit = op;
1425 	do {
1426 		info.error = 1;
1427 		vm_page_rb_tree_RB_SCAN(&backing_object->rb_memq, NULL,
1428 					vm_object_backing_scan_callback,
1429 					&info);
1430 	} while (info.error < 0);
1431 	crit_exit();
1432 	return(info.error);
1433 }
1434 
1435 /*
1436  * The caller must hold vm_token.
1437  */
1438 static int
1439 vm_object_backing_scan_callback(vm_page_t p, void *data)
1440 {
1441 	struct rb_vm_page_scan_info *info = data;
1442 	vm_object_t backing_object;
1443 	vm_object_t object;
1444 	vm_pindex_t new_pindex;
1445 	vm_pindex_t backing_offset_index;
1446 	int op;
1447 
1448 	new_pindex = p->pindex - info->backing_offset_index;
1449 	op = info->limit;
1450 	object = info->object;
1451 	backing_object = info->backing_object;
1452 	backing_offset_index = info->backing_offset_index;
1453 
1454 	if (op & OBSC_TEST_ALL_SHADOWED) {
1455 		vm_page_t pp;
1456 
1457 		/*
1458 		 * Ignore pages outside the parent object's range
1459 		 * and outside the parent object's mapping of the
1460 		 * backing object.
1461 		 *
1462 		 * note that we do not busy the backing object's
1463 		 * page.
1464 		 */
1465 		if (
1466 		    p->pindex < backing_offset_index ||
1467 		    new_pindex >= object->size
1468 		) {
1469 			return(0);
1470 		}
1471 
1472 		/*
1473 		 * See if the parent has the page or if the parent's
1474 		 * object pager has the page.  If the parent has the
1475 		 * page but the page is not valid, the parent's
1476 		 * object pager must have the page.
1477 		 *
1478 		 * If this fails, the parent does not completely shadow
1479 		 * the object and we might as well give up now.
1480 		 */
1481 
1482 		pp = vm_page_lookup(object, new_pindex);
1483 		if ((pp == NULL || pp->valid == 0) &&
1484 		    !vm_pager_has_page(object, new_pindex)
1485 		) {
1486 			info->error = 0;	/* problemo */
1487 			return(-1);		/* stop the scan */
1488 		}
1489 	}
1490 
1491 	/*
1492 	 * Check for busy page
1493 	 */
1494 
1495 	if (op & (OBSC_COLLAPSE_WAIT | OBSC_COLLAPSE_NOWAIT)) {
1496 		vm_page_t pp;
1497 
1498 		if (op & OBSC_COLLAPSE_NOWAIT) {
1499 			if (
1500 			    (p->flags & PG_BUSY) ||
1501 			    !p->valid ||
1502 			    p->hold_count ||
1503 			    p->wire_count ||
1504 			    p->busy
1505 			) {
1506 				return(0);
1507 			}
1508 		} else if (op & OBSC_COLLAPSE_WAIT) {
1509 			if (vm_page_sleep_busy(p, TRUE, "vmocol")) {
1510 				/*
1511 				 * If we slept, anything could have
1512 				 * happened.   Ask that the scan be restarted.
1513 				 *
1514 				 * Since the object is marked dead, the
1515 				 * backing offset should not have changed.
1516 				 */
1517 				info->error = -1;
1518 				return(-1);
1519 			}
1520 		}
1521 
1522 		/*
1523 		 * Busy the page
1524 		 */
1525 		vm_page_busy(p);
1526 
1527 		KASSERT(
1528 		    p->object == backing_object,
1529 		    ("vm_object_qcollapse(): object mismatch")
1530 		);
1531 
1532 		/*
1533 		 * Destroy any associated swap
1534 		 */
1535 		if (backing_object->type == OBJT_SWAP)
1536 			swap_pager_freespace(backing_object, p->pindex, 1);
1537 
1538 		if (
1539 		    p->pindex < backing_offset_index ||
1540 		    new_pindex >= object->size
1541 		) {
1542 			/*
1543 			 * Page is out of the parent object's range, we
1544 			 * can simply destroy it.
1545 			 */
1546 			vm_page_protect(p, VM_PROT_NONE);
1547 			vm_page_free(p);
1548 			return(0);
1549 		}
1550 
1551 		pp = vm_page_lookup(object, new_pindex);
1552 		if (pp != NULL || vm_pager_has_page(object, new_pindex)) {
1553 			/*
1554 			 * page already exists in parent OR swap exists
1555 			 * for this location in the parent.  Destroy
1556 			 * the original page from the backing object.
1557 			 *
1558 			 * Leave the parent's page alone
1559 			 */
1560 			vm_page_protect(p, VM_PROT_NONE);
1561 			vm_page_free(p);
1562 			return(0);
1563 		}
1564 
1565 		/*
1566 		 * Page does not exist in parent, rename the
1567 		 * page from the backing object to the main object.
1568 		 *
1569 		 * If the page was mapped to a process, it can remain
1570 		 * mapped through the rename.
1571 		 */
1572 		if ((p->queue - p->pc) == PQ_CACHE)
1573 			vm_page_deactivate(p);
1574 
1575 		vm_page_rename(p, object, new_pindex);
1576 		/* page automatically made dirty by rename */
1577 	}
1578 	return(0);
1579 }
1580 
1581 /*
1582  * This version of collapse allows the operation to occur earlier and
1583  * when paging_in_progress is true for an object...  This is not a complete
1584  * operation, but should plug 99.9% of the rest of the leaks.
1585  *
1586  * The caller must hold vm_token and vmobj_token.
1587  * (only called from vm_object_collapse)
1588  */
1589 static void
1590 vm_object_qcollapse(vm_object_t object)
1591 {
1592 	vm_object_t backing_object = object->backing_object;
1593 
1594 	if (backing_object->ref_count != 1)
1595 		return;
1596 
1597 	backing_object->ref_count += 2;
1598 
1599 	vm_object_backing_scan(object, OBSC_COLLAPSE_NOWAIT);
1600 
1601 	backing_object->ref_count -= 2;
1602 }
1603 
1604 /*
1605  * Collapse an object with the object backing it.  Pages in the backing
1606  * object are moved into the parent, and the backing object is deallocated.
1607  *
1608  * The caller must hold (object).
1609  */
1610 void
1611 vm_object_collapse(vm_object_t object)
1612 {
1613 	ASSERT_LWKT_TOKEN_HELD(&vm_token);
1614 	ASSERT_LWKT_TOKEN_HELD(&vmobj_token);
1615 	vm_object_assert_held(object);
1616 
1617 	while (TRUE) {
1618 		vm_object_t backing_object;
1619 
1620 		/*
1621 		 * Verify that the conditions are right for collapse:
1622 		 *
1623 		 * The object exists and the backing object exists.
1624 		 */
1625 		if (object == NULL)
1626 			break;
1627 
1628 		if ((backing_object = object->backing_object) == NULL)
1629 			break;
1630 
1631 		vm_object_hold(backing_object);
1632 		if (backing_object != object->backing_object) {
1633 			vm_object_drop(backing_object);
1634 			continue;
1635 		}
1636 
1637 		/*
1638 		 * we check the backing object first, because it is most likely
1639 		 * not collapsable.
1640 		 */
1641 		if (backing_object->handle != NULL ||
1642 		    (backing_object->type != OBJT_DEFAULT &&
1643 		     backing_object->type != OBJT_SWAP) ||
1644 		    (backing_object->flags & OBJ_DEAD) ||
1645 		    object->handle != NULL ||
1646 		    (object->type != OBJT_DEFAULT &&
1647 		     object->type != OBJT_SWAP) ||
1648 		    (object->flags & OBJ_DEAD)) {
1649 			vm_object_drop(backing_object);
1650 			break;
1651 		}
1652 
1653 		if (
1654 		    object->paging_in_progress != 0 ||
1655 		    backing_object->paging_in_progress != 0
1656 		) {
1657 			vm_object_drop(backing_object);
1658 			vm_object_qcollapse(object);
1659 			break;
1660 		}
1661 
1662 		/*
1663 		 * We know that we can either collapse the backing object (if
1664 		 * the parent is the only reference to it) or (perhaps) have
1665 		 * the parent bypass the object if the parent happens to shadow
1666 		 * all the resident pages in the entire backing object.
1667 		 *
1668 		 * This is ignoring pager-backed pages such as swap pages.
1669 		 * vm_object_backing_scan fails the shadowing test in this
1670 		 * case.
1671 		 */
1672 
1673 		if (backing_object->ref_count == 1) {
1674 			/*
1675 			 * If there is exactly one reference to the backing
1676 			 * object, we can collapse it into the parent.
1677 			 */
1678 			vm_object_backing_scan(object, OBSC_COLLAPSE_WAIT);
1679 
1680 			/*
1681 			 * Move the pager from backing_object to object.
1682 			 */
1683 
1684 			if (backing_object->type == OBJT_SWAP) {
1685 				vm_object_pip_add(backing_object, 1);
1686 
1687 				/*
1688 				 * scrap the paging_offset junk and do a
1689 				 * discrete copy.  This also removes major
1690 				 * assumptions about how the swap-pager
1691 				 * works from where it doesn't belong.  The
1692 				 * new swapper is able to optimize the
1693 				 * destroy-source case.
1694 				 */
1695 
1696 				vm_object_pip_add(object, 1);
1697 				swap_pager_copy(
1698 				    backing_object,
1699 				    object,
1700 				    OFF_TO_IDX(object->backing_object_offset), TRUE);
1701 				vm_object_pip_wakeup(object);
1702 
1703 				vm_object_pip_wakeup(backing_object);
1704 			}
1705 			/*
1706 			 * Object now shadows whatever backing_object did.
1707 			 * Note that the reference to
1708 			 * backing_object->backing_object moves from within
1709 			 * backing_object to within object.
1710 			 */
1711 
1712 			LIST_REMOVE(object, shadow_list);
1713 			object->backing_object->shadow_count--;
1714 			object->backing_object->generation++;
1715 			if (backing_object->backing_object) {
1716 				LIST_REMOVE(backing_object, shadow_list);
1717 				backing_object->backing_object->shadow_count--;
1718 				backing_object->backing_object->generation++;
1719 			}
1720 			object->backing_object = backing_object->backing_object;
1721 			if (object->backing_object) {
1722 				LIST_INSERT_HEAD(
1723 				    &object->backing_object->shadow_head,
1724 				    object,
1725 				    shadow_list
1726 				);
1727 				object->backing_object->shadow_count++;
1728 				object->backing_object->generation++;
1729 			}
1730 
1731 			object->backing_object_offset +=
1732 			    backing_object->backing_object_offset;
1733 
1734 			/*
1735 			 * Discard backing_object.
1736 			 *
1737 			 * Since the backing object has no pages, no pager left,
1738 			 * and no object references within it, all that is
1739 			 * necessary is to dispose of it.
1740 			 */
1741 
1742 			KASSERT(backing_object->ref_count == 1,
1743 				("backing_object %p was somehow "
1744 				 "re-referenced during collapse!",
1745 				 backing_object));
1746 			KASSERT(RB_EMPTY(&backing_object->rb_memq),
1747 				("backing_object %p somehow has left "
1748 				 "over pages during collapse!",
1749 				 backing_object));
1750 
1751 			/*
1752 			 * Wait for hold count to hit zero
1753 			 */
1754 			vm_object_drop(backing_object);
1755 			vm_object_hold_wait(backing_object);
1756 
1757 			/* (we are holding vmobj_token) */
1758 			TAILQ_REMOVE(&vm_object_list, backing_object,
1759 				     object_list);
1760 			--backing_object->ref_count;	/* safety/debug */
1761 			vm_object_count--;
1762 
1763 			zfree(obj_zone, backing_object);
1764 
1765 			object_collapses++;
1766 		} else {
1767 			vm_object_t new_backing_object;
1768 
1769 			/*
1770 			 * If we do not entirely shadow the backing object,
1771 			 * there is nothing we can do so we give up.
1772 			 */
1773 
1774 			if (vm_object_backing_scan(object, OBSC_TEST_ALL_SHADOWED) == 0) {
1775 				vm_object_drop(backing_object);
1776 				break;
1777 			}
1778 
1779 			/*
1780 			 * Make the parent shadow the next object in the
1781 			 * chain.  Deallocating backing_object will not remove
1782 			 * it, since its reference count is at least 2.
1783 			 */
1784 
1785 			LIST_REMOVE(object, shadow_list);
1786 			backing_object->shadow_count--;
1787 			backing_object->generation++;
1788 
1789 			new_backing_object = backing_object->backing_object;
1790 			if ((object->backing_object = new_backing_object) != NULL) {
1791 				vm_object_reference(new_backing_object);
1792 				LIST_INSERT_HEAD(
1793 				    &new_backing_object->shadow_head,
1794 				    object,
1795 				    shadow_list
1796 				);
1797 				new_backing_object->shadow_count++;
1798 				new_backing_object->generation++;
1799 				object->backing_object_offset +=
1800 					backing_object->backing_object_offset;
1801 			}
1802 
1803 			/*
1804 			 * Drop the reference count on backing_object. Since
1805 			 * its ref_count was at least 2, it will not vanish;
1806 			 * so we don't need to call vm_object_deallocate, but
1807 			 * we do anyway.
1808 			 */
1809 			vm_object_drop(backing_object);
1810 			vm_object_deallocate_locked(backing_object);
1811 			object_bypasses++;
1812 		}
1813 
1814 		/*
1815 		 * Try again with this object's new backing object.
1816 		 */
1817 	}
1818 }
1819 
1820 /*
1821  * Removes all physical pages in the specified object range from the
1822  * object's list of pages.
1823  *
1824  * No requirements.
1825  */
1826 static int vm_object_page_remove_callback(vm_page_t p, void *data);
1827 
1828 void
1829 vm_object_page_remove(vm_object_t object, vm_pindex_t start, vm_pindex_t end,
1830 		      boolean_t clean_only)
1831 {
1832 	struct rb_vm_page_scan_info info;
1833 	int all;
1834 
1835 	/*
1836 	 * Degenerate cases and assertions
1837 	 */
1838 	lwkt_gettoken(&vm_token);
1839 	if (object == NULL ||
1840 	    (object->resident_page_count == 0 && object->swblock_count == 0)) {
1841 		lwkt_reltoken(&vm_token);
1842 		return;
1843 	}
1844 	KASSERT(object->type != OBJT_PHYS,
1845 		("attempt to remove pages from a physical object"));
1846 
1847 	/*
1848 	 * Indicate that paging is occuring on the object
1849 	 */
1850 	crit_enter();
1851 	vm_object_pip_add(object, 1);
1852 
1853 	/*
1854 	 * Figure out the actual removal range and whether we are removing
1855 	 * the entire contents of the object or not.  If removing the entire
1856 	 * contents, be sure to get all pages, even those that might be
1857 	 * beyond the end of the object.
1858 	 */
1859 	info.start_pindex = start;
1860 	if (end == 0)
1861 		info.end_pindex = (vm_pindex_t)-1;
1862 	else
1863 		info.end_pindex = end - 1;
1864 	info.limit = clean_only;
1865 	all = (start == 0 && info.end_pindex >= object->size - 1);
1866 
1867 	/*
1868 	 * Loop until we are sure we have gotten them all.
1869 	 */
1870 	do {
1871 		info.error = 0;
1872 		vm_page_rb_tree_RB_SCAN(&object->rb_memq, rb_vm_page_scancmp,
1873 					vm_object_page_remove_callback, &info);
1874 	} while (info.error);
1875 
1876 	/*
1877 	 * Remove any related swap if throwing away pages, or for
1878 	 * non-swap objects (the swap is a clean copy in that case).
1879 	 */
1880 	if (object->type != OBJT_SWAP || clean_only == FALSE) {
1881 		if (all)
1882 			swap_pager_freespace_all(object);
1883 		else
1884 			swap_pager_freespace(object, info.start_pindex,
1885 			     info.end_pindex - info.start_pindex + 1);
1886 	}
1887 
1888 	/*
1889 	 * Cleanup
1890 	 */
1891 	vm_object_pip_wakeup(object);
1892 	crit_exit();
1893 	lwkt_reltoken(&vm_token);
1894 }
1895 
1896 /*
1897  * The caller must hold vm_token.
1898  */
1899 static int
1900 vm_object_page_remove_callback(vm_page_t p, void *data)
1901 {
1902 	struct rb_vm_page_scan_info *info = data;
1903 
1904 	/*
1905 	 * Wired pages cannot be destroyed, but they can be invalidated
1906 	 * and we do so if clean_only (limit) is not set.
1907 	 *
1908 	 * WARNING!  The page may be wired due to being part of a buffer
1909 	 *	     cache buffer, and the buffer might be marked B_CACHE.
1910 	 *	     This is fine as part of a truncation but VFSs must be
1911 	 *	     sure to fix the buffer up when re-extending the file.
1912 	 */
1913 	if (p->wire_count != 0) {
1914 		vm_page_protect(p, VM_PROT_NONE);
1915 		if (info->limit == 0)
1916 			p->valid = 0;
1917 		return(0);
1918 	}
1919 
1920 	/*
1921 	 * The busy flags are only cleared at
1922 	 * interrupt -- minimize the spl transitions
1923 	 */
1924 
1925 	if (vm_page_sleep_busy(p, TRUE, "vmopar")) {
1926 		info->error = 1;
1927 		return(0);
1928 	}
1929 
1930 	/*
1931 	 * limit is our clean_only flag.  If set and the page is dirty, do
1932 	 * not free it.  If set and the page is being held by someone, do
1933 	 * not free it.
1934 	 */
1935 	if (info->limit && p->valid) {
1936 		vm_page_test_dirty(p);
1937 		if (p->valid & p->dirty)
1938 			return(0);
1939 		if (p->hold_count)
1940 			return(0);
1941 	}
1942 
1943 	/*
1944 	 * Destroy the page
1945 	 */
1946 	vm_page_busy(p);
1947 	vm_page_protect(p, VM_PROT_NONE);
1948 	vm_page_free(p);
1949 	return(0);
1950 }
1951 
1952 /*
1953  * Coalesces two objects backing up adjoining regions of memory into a
1954  * single object.
1955  *
1956  * returns TRUE if objects were combined.
1957  *
1958  * NOTE: Only works at the moment if the second object is NULL -
1959  *	 if it's not, which object do we lock first?
1960  *
1961  * Parameters:
1962  *	prev_object	First object to coalesce
1963  *	prev_offset	Offset into prev_object
1964  *	next_object	Second object into coalesce
1965  *	next_offset	Offset into next_object
1966  *
1967  *	prev_size	Size of reference to prev_object
1968  *	next_size	Size of reference to next_object
1969  *
1970  * The caller must hold vm_token and vmobj_token.
1971  *
1972  * The caller does not need to hold (prev_object) but must have a stable
1973  * pointer to it (typically by holding the vm_map locked).
1974  */
1975 boolean_t
1976 vm_object_coalesce(vm_object_t prev_object, vm_pindex_t prev_pindex,
1977 		   vm_size_t prev_size, vm_size_t next_size)
1978 {
1979 	vm_pindex_t next_pindex;
1980 
1981 	ASSERT_LWKT_TOKEN_HELD(&vm_token);
1982 	ASSERT_LWKT_TOKEN_HELD(&vmobj_token);
1983 
1984 	if (prev_object == NULL) {
1985 		return (TRUE);
1986 	}
1987 
1988 	vm_object_hold(prev_object);
1989 
1990 	if (prev_object->type != OBJT_DEFAULT &&
1991 	    prev_object->type != OBJT_SWAP) {
1992 		vm_object_drop(prev_object);
1993 		return (FALSE);
1994 	}
1995 
1996 	/*
1997 	 * Try to collapse the object first
1998 	 */
1999 	vm_object_collapse(prev_object);
2000 
2001 	/*
2002 	 * Can't coalesce if: . more than one reference . paged out . shadows
2003 	 * another object . has a copy elsewhere (any of which mean that the
2004 	 * pages not mapped to prev_entry may be in use anyway)
2005 	 */
2006 
2007 	if (prev_object->backing_object != NULL) {
2008 		vm_object_drop(prev_object);
2009 		return (FALSE);
2010 	}
2011 
2012 	prev_size >>= PAGE_SHIFT;
2013 	next_size >>= PAGE_SHIFT;
2014 	next_pindex = prev_pindex + prev_size;
2015 
2016 	if ((prev_object->ref_count > 1) &&
2017 	    (prev_object->size != next_pindex)) {
2018 		vm_object_drop(prev_object);
2019 		return (FALSE);
2020 	}
2021 
2022 	/*
2023 	 * Remove any pages that may still be in the object from a previous
2024 	 * deallocation.
2025 	 */
2026 	if (next_pindex < prev_object->size) {
2027 		vm_object_page_remove(prev_object,
2028 				      next_pindex,
2029 				      next_pindex + next_size, FALSE);
2030 		if (prev_object->type == OBJT_SWAP)
2031 			swap_pager_freespace(prev_object,
2032 					     next_pindex, next_size);
2033 	}
2034 
2035 	/*
2036 	 * Extend the object if necessary.
2037 	 */
2038 	if (next_pindex + next_size > prev_object->size)
2039 		prev_object->size = next_pindex + next_size;
2040 
2041 	vm_object_drop(prev_object);
2042 	return (TRUE);
2043 }
2044 
2045 /*
2046  * Make the object writable and flag is being possibly dirty.
2047  *
2048  * No requirements.
2049  */
2050 void
2051 vm_object_set_writeable_dirty(vm_object_t object)
2052 {
2053 	struct vnode *vp;
2054 
2055 	lwkt_gettoken(&vm_token);
2056 	vm_object_set_flag(object, OBJ_WRITEABLE|OBJ_MIGHTBEDIRTY);
2057 	if (object->type == OBJT_VNODE &&
2058 	    (vp = (struct vnode *)object->handle) != NULL) {
2059 		if ((vp->v_flag & VOBJDIRTY) == 0) {
2060 			vsetflags(vp, VOBJDIRTY);
2061 		}
2062 	}
2063 	lwkt_reltoken(&vm_token);
2064 }
2065 
2066 #include "opt_ddb.h"
2067 #ifdef DDB
2068 #include <sys/kernel.h>
2069 
2070 #include <sys/cons.h>
2071 
2072 #include <ddb/ddb.h>
2073 
2074 static int	_vm_object_in_map (vm_map_t map, vm_object_t object,
2075 				       vm_map_entry_t entry);
2076 static int	vm_object_in_map (vm_object_t object);
2077 
2078 /*
2079  * The caller must hold vm_token.
2080  */
2081 static int
2082 _vm_object_in_map(vm_map_t map, vm_object_t object, vm_map_entry_t entry)
2083 {
2084 	vm_map_t tmpm;
2085 	vm_map_entry_t tmpe;
2086 	vm_object_t obj;
2087 	int entcount;
2088 
2089 	if (map == 0)
2090 		return 0;
2091 	if (entry == 0) {
2092 		tmpe = map->header.next;
2093 		entcount = map->nentries;
2094 		while (entcount-- && (tmpe != &map->header)) {
2095 			if( _vm_object_in_map(map, object, tmpe)) {
2096 				return 1;
2097 			}
2098 			tmpe = tmpe->next;
2099 		}
2100 		return (0);
2101 	}
2102 	switch(entry->maptype) {
2103 	case VM_MAPTYPE_SUBMAP:
2104 		tmpm = entry->object.sub_map;
2105 		tmpe = tmpm->header.next;
2106 		entcount = tmpm->nentries;
2107 		while (entcount-- && tmpe != &tmpm->header) {
2108 			if( _vm_object_in_map(tmpm, object, tmpe)) {
2109 				return 1;
2110 			}
2111 			tmpe = tmpe->next;
2112 		}
2113 		break;
2114 	case VM_MAPTYPE_NORMAL:
2115 	case VM_MAPTYPE_VPAGETABLE:
2116 		obj = entry->object.vm_object;
2117 		while (obj) {
2118 			if (obj == object)
2119 				return 1;
2120 			obj = obj->backing_object;
2121 		}
2122 		break;
2123 	default:
2124 		break;
2125 	}
2126 	return 0;
2127 }
2128 
2129 static int vm_object_in_map_callback(struct proc *p, void *data);
2130 
2131 struct vm_object_in_map_info {
2132 	vm_object_t object;
2133 	int rv;
2134 };
2135 
2136 /*
2137  * Debugging only
2138  */
2139 static int
2140 vm_object_in_map(vm_object_t object)
2141 {
2142 	struct vm_object_in_map_info info;
2143 
2144 	info.rv = 0;
2145 	info.object = object;
2146 
2147 	allproc_scan(vm_object_in_map_callback, &info);
2148 	if (info.rv)
2149 		return 1;
2150 	if( _vm_object_in_map(&kernel_map, object, 0))
2151 		return 1;
2152 	if( _vm_object_in_map(&pager_map, object, 0))
2153 		return 1;
2154 	if( _vm_object_in_map(&buffer_map, object, 0))
2155 		return 1;
2156 	return 0;
2157 }
2158 
2159 /*
2160  * Debugging only
2161  */
2162 static int
2163 vm_object_in_map_callback(struct proc *p, void *data)
2164 {
2165 	struct vm_object_in_map_info *info = data;
2166 
2167 	if (p->p_vmspace) {
2168 		if (_vm_object_in_map(&p->p_vmspace->vm_map, info->object, 0)) {
2169 			info->rv = 1;
2170 			return -1;
2171 		}
2172 	}
2173 	return (0);
2174 }
2175 
2176 DB_SHOW_COMMAND(vmochk, vm_object_check)
2177 {
2178 	vm_object_t object;
2179 
2180 	/*
2181 	 * make sure that internal objs are in a map somewhere
2182 	 * and none have zero ref counts.
2183 	 */
2184 	for (object = TAILQ_FIRST(&vm_object_list);
2185 			object != NULL;
2186 			object = TAILQ_NEXT(object, object_list)) {
2187 		if (object->type == OBJT_MARKER)
2188 			continue;
2189 		if (object->handle == NULL &&
2190 		    (object->type == OBJT_DEFAULT || object->type == OBJT_SWAP)) {
2191 			if (object->ref_count == 0) {
2192 				db_printf("vmochk: internal obj has zero ref count: %ld\n",
2193 					(long)object->size);
2194 			}
2195 			if (!vm_object_in_map(object)) {
2196 				db_printf(
2197 			"vmochk: internal obj is not in a map: "
2198 			"ref: %d, size: %lu: 0x%lx, backing_object: %p\n",
2199 				    object->ref_count, (u_long)object->size,
2200 				    (u_long)object->size,
2201 				    (void *)object->backing_object);
2202 			}
2203 		}
2204 	}
2205 }
2206 
2207 /*
2208  * Debugging only
2209  */
2210 DB_SHOW_COMMAND(object, vm_object_print_static)
2211 {
2212 	/* XXX convert args. */
2213 	vm_object_t object = (vm_object_t)addr;
2214 	boolean_t full = have_addr;
2215 
2216 	vm_page_t p;
2217 
2218 	/* XXX count is an (unused) arg.  Avoid shadowing it. */
2219 #define	count	was_count
2220 
2221 	int count;
2222 
2223 	if (object == NULL)
2224 		return;
2225 
2226 	db_iprintf(
2227 	    "Object %p: type=%d, size=0x%lx, res=%d, ref=%d, flags=0x%x\n",
2228 	    object, (int)object->type, (u_long)object->size,
2229 	    object->resident_page_count, object->ref_count, object->flags);
2230 	/*
2231 	 * XXX no %qd in kernel.  Truncate object->backing_object_offset.
2232 	 */
2233 	db_iprintf(" sref=%d, backing_object(%d)=(%p)+0x%lx\n",
2234 	    object->shadow_count,
2235 	    object->backing_object ? object->backing_object->ref_count : 0,
2236 	    object->backing_object, (long)object->backing_object_offset);
2237 
2238 	if (!full)
2239 		return;
2240 
2241 	db_indent += 2;
2242 	count = 0;
2243 	RB_FOREACH(p, vm_page_rb_tree, &object->rb_memq) {
2244 		if (count == 0)
2245 			db_iprintf("memory:=");
2246 		else if (count == 6) {
2247 			db_printf("\n");
2248 			db_iprintf(" ...");
2249 			count = 0;
2250 		} else
2251 			db_printf(",");
2252 		count++;
2253 
2254 		db_printf("(off=0x%lx,page=0x%lx)",
2255 		    (u_long) p->pindex, (u_long) VM_PAGE_TO_PHYS(p));
2256 	}
2257 	if (count != 0)
2258 		db_printf("\n");
2259 	db_indent -= 2;
2260 }
2261 
2262 /* XXX. */
2263 #undef count
2264 
2265 /*
2266  * XXX need this non-static entry for calling from vm_map_print.
2267  *
2268  * Debugging only
2269  */
2270 void
2271 vm_object_print(/* db_expr_t */ long addr,
2272 		boolean_t have_addr,
2273 		/* db_expr_t */ long count,
2274 		char *modif)
2275 {
2276 	vm_object_print_static(addr, have_addr, count, modif);
2277 }
2278 
2279 /*
2280  * Debugging only
2281  */
2282 DB_SHOW_COMMAND(vmopag, vm_object_print_pages)
2283 {
2284 	vm_object_t object;
2285 	int nl = 0;
2286 	int c;
2287 	for (object = TAILQ_FIRST(&vm_object_list);
2288 			object != NULL;
2289 			object = TAILQ_NEXT(object, object_list)) {
2290 		vm_pindex_t idx, fidx;
2291 		vm_pindex_t osize;
2292 		vm_paddr_t pa = -1, padiff;
2293 		int rcount;
2294 		vm_page_t m;
2295 
2296 		if (object->type == OBJT_MARKER)
2297 			continue;
2298 		db_printf("new object: %p\n", (void *)object);
2299 		if ( nl > 18) {
2300 			c = cngetc();
2301 			if (c != ' ')
2302 				return;
2303 			nl = 0;
2304 		}
2305 		nl++;
2306 		rcount = 0;
2307 		fidx = 0;
2308 		osize = object->size;
2309 		if (osize > 128)
2310 			osize = 128;
2311 		for (idx = 0; idx < osize; idx++) {
2312 			m = vm_page_lookup(object, idx);
2313 			if (m == NULL) {
2314 				if (rcount) {
2315 					db_printf(" index(%ld)run(%d)pa(0x%lx)\n",
2316 						(long)fidx, rcount, (long)pa);
2317 					if ( nl > 18) {
2318 						c = cngetc();
2319 						if (c != ' ')
2320 							return;
2321 						nl = 0;
2322 					}
2323 					nl++;
2324 					rcount = 0;
2325 				}
2326 				continue;
2327 			}
2328 
2329 
2330 			if (rcount &&
2331 				(VM_PAGE_TO_PHYS(m) == pa + rcount * PAGE_SIZE)) {
2332 				++rcount;
2333 				continue;
2334 			}
2335 			if (rcount) {
2336 				padiff = pa + rcount * PAGE_SIZE - VM_PAGE_TO_PHYS(m);
2337 				padiff >>= PAGE_SHIFT;
2338 				padiff &= PQ_L2_MASK;
2339 				if (padiff == 0) {
2340 					pa = VM_PAGE_TO_PHYS(m) - rcount * PAGE_SIZE;
2341 					++rcount;
2342 					continue;
2343 				}
2344 				db_printf(" index(%ld)run(%d)pa(0x%lx)",
2345 					(long)fidx, rcount, (long)pa);
2346 				db_printf("pd(%ld)\n", (long)padiff);
2347 				if ( nl > 18) {
2348 					c = cngetc();
2349 					if (c != ' ')
2350 						return;
2351 					nl = 0;
2352 				}
2353 				nl++;
2354 			}
2355 			fidx = idx;
2356 			pa = VM_PAGE_TO_PHYS(m);
2357 			rcount = 1;
2358 		}
2359 		if (rcount) {
2360 			db_printf(" index(%ld)run(%d)pa(0x%lx)\n",
2361 				(long)fidx, rcount, (long)pa);
2362 			if ( nl > 18) {
2363 				c = cngetc();
2364 				if (c != ' ')
2365 					return;
2366 				nl = 0;
2367 			}
2368 			nl++;
2369 		}
2370 	}
2371 }
2372 #endif /* DDB */
2373