xref: /freebsd/sys/vm/vm_object.c (revision f0a75d274af375d15b97b830966b99a02b7db911)
1 /*-
2  * Copyright (c) 1991, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * This code is derived from software contributed to Berkeley by
6  * The Mach Operating System project at Carnegie-Mellon University.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 4. Neither the name of the University nor the names of its contributors
17  *    may be used to endorse or promote products derived from this software
18  *    without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  *
32  *	from: @(#)vm_object.c	8.5 (Berkeley) 3/22/94
33  *
34  *
35  * Copyright (c) 1987, 1990 Carnegie-Mellon University.
36  * All rights reserved.
37  *
38  * Authors: Avadis Tevanian, Jr., Michael Wayne Young
39  *
40  * Permission to use, copy, modify and distribute this software and
41  * its documentation is hereby granted, provided that both the copyright
42  * notice and this permission notice appear in all copies of the
43  * software, derivative works or modified versions, and any portions
44  * thereof, and that both notices appear in supporting documentation.
45  *
46  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
47  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
48  * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
49  *
50  * Carnegie Mellon requests users of this software to return to
51  *
52  *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
53  *  School of Computer Science
54  *  Carnegie Mellon University
55  *  Pittsburgh PA 15213-3890
56  *
57  * any improvements or extensions that they make and grant Carnegie the
58  * rights to redistribute these changes.
59  */
60 
61 /*
62  *	Virtual memory object module.
63  */
64 
65 #include <sys/cdefs.h>
66 __FBSDID("$FreeBSD$");
67 
68 #include <sys/param.h>
69 #include <sys/systm.h>
70 #include <sys/lock.h>
71 #include <sys/mman.h>
72 #include <sys/mount.h>
73 #include <sys/kernel.h>
74 #include <sys/sysctl.h>
75 #include <sys/mutex.h>
76 #include <sys/proc.h>		/* for curproc, pageproc */
77 #include <sys/socket.h>
78 #include <sys/vnode.h>
79 #include <sys/vmmeter.h>
80 #include <sys/sx.h>
81 
82 #include <vm/vm.h>
83 #include <vm/vm_param.h>
84 #include <vm/pmap.h>
85 #include <vm/vm_map.h>
86 #include <vm/vm_object.h>
87 #include <vm/vm_page.h>
88 #include <vm/vm_pageout.h>
89 #include <vm/vm_pager.h>
90 #include <vm/swap_pager.h>
91 #include <vm/vm_kern.h>
92 #include <vm/vm_extern.h>
93 #include <vm/uma.h>
94 
95 #define EASY_SCAN_FACTOR       8
96 
97 #define MSYNC_FLUSH_HARDSEQ	0x01
98 #define MSYNC_FLUSH_SOFTSEQ	0x02
99 
100 /*
101  * msync / VM object flushing optimizations
102  */
103 static int msync_flush_flags = MSYNC_FLUSH_HARDSEQ | MSYNC_FLUSH_SOFTSEQ;
104 SYSCTL_INT(_vm, OID_AUTO, msync_flush_flags,
105         CTLFLAG_RW, &msync_flush_flags, 0, "");
106 
107 static int old_msync;
108 SYSCTL_INT(_vm, OID_AUTO, old_msync, CTLFLAG_RW, &old_msync, 0,
109     "Use old (insecure) msync behavior");
110 
111 static void	vm_object_qcollapse(vm_object_t object);
112 static int	vm_object_page_collect_flush(vm_object_t object, vm_page_t p, int curgeneration, int pagerflags);
113 static void	vm_object_vndeallocate(vm_object_t object);
114 
115 /*
116  *	Virtual memory objects maintain the actual data
117  *	associated with allocated virtual memory.  A given
118  *	page of memory exists within exactly one object.
119  *
120  *	An object is only deallocated when all "references"
121  *	are given up.  Only one "reference" to a given
122  *	region of an object should be writeable.
123  *
124  *	Associated with each object is a list of all resident
125  *	memory pages belonging to that object; this list is
126  *	maintained by the "vm_page" module, and locked by the object's
127  *	lock.
128  *
129  *	Each object also records a "pager" routine which is
130  *	used to retrieve (and store) pages to the proper backing
131  *	storage.  In addition, objects may be backed by other
132  *	objects from which they were virtual-copied.
133  *
134  *	The only items within the object structure which are
135  *	modified after time of creation are:
136  *		reference count		locked by object's lock
137  *		pager routine		locked by object's lock
138  *
139  */
140 
141 struct object_q vm_object_list;
142 struct mtx vm_object_list_mtx;	/* lock for object list and count */
143 
144 struct vm_object kernel_object_store;
145 struct vm_object kmem_object_store;
146 
147 SYSCTL_NODE(_vm_stats, OID_AUTO, object, CTLFLAG_RD, 0, "VM object stats");
148 
149 static long object_collapses;
150 SYSCTL_LONG(_vm_stats_object, OID_AUTO, collapses, CTLFLAG_RD,
151     &object_collapses, 0, "VM object collapses");
152 
153 static long object_bypasses;
154 SYSCTL_LONG(_vm_stats_object, OID_AUTO, bypasses, CTLFLAG_RD,
155     &object_bypasses, 0, "VM object bypasses");
156 
157 /*
158  * next_index determines the page color that is assigned to the next
159  * allocated object.  Accesses to next_index are not synchronized
160  * because the effects of two or more object allocations using
161  * next_index simultaneously are inconsequential.  At any given time,
162  * numerous objects have the same page color.
163  */
164 static int next_index;
165 
166 static uma_zone_t obj_zone;
167 
168 static int vm_object_zinit(void *mem, int size, int flags);
169 
170 #ifdef INVARIANTS
171 static void vm_object_zdtor(void *mem, int size, void *arg);
172 
173 static void
174 vm_object_zdtor(void *mem, int size, void *arg)
175 {
176 	vm_object_t object;
177 
178 	object = (vm_object_t)mem;
179 	KASSERT(TAILQ_EMPTY(&object->memq),
180 	    ("object %p has resident pages",
181 	    object));
182 	KASSERT(object->paging_in_progress == 0,
183 	    ("object %p paging_in_progress = %d",
184 	    object, object->paging_in_progress));
185 	KASSERT(object->resident_page_count == 0,
186 	    ("object %p resident_page_count = %d",
187 	    object, object->resident_page_count));
188 	KASSERT(object->shadow_count == 0,
189 	    ("object %p shadow_count = %d",
190 	    object, object->shadow_count));
191 }
192 #endif
193 
194 static int
195 vm_object_zinit(void *mem, int size, int flags)
196 {
197 	vm_object_t object;
198 
199 	object = (vm_object_t)mem;
200 	bzero(&object->mtx, sizeof(object->mtx));
201 	VM_OBJECT_LOCK_INIT(object, "standard object");
202 
203 	/* These are true for any object that has been freed */
204 	object->paging_in_progress = 0;
205 	object->resident_page_count = 0;
206 	object->shadow_count = 0;
207 	return (0);
208 }
209 
210 void
211 _vm_object_allocate(objtype_t type, vm_pindex_t size, vm_object_t object)
212 {
213 	int incr;
214 
215 	TAILQ_INIT(&object->memq);
216 	LIST_INIT(&object->shadow_head);
217 
218 	object->root = NULL;
219 	object->type = type;
220 	object->size = size;
221 	object->generation = 1;
222 	object->ref_count = 1;
223 	object->flags = 0;
224 	if ((object->type == OBJT_DEFAULT) || (object->type == OBJT_SWAP))
225 		object->flags = OBJ_ONEMAPPING;
226 	incr = PQ_MAXLENGTH;
227 	if (size <= incr)
228 		incr = size;
229 	object->pg_color = next_index;
230 	next_index = (object->pg_color + incr) & PQ_COLORMASK;
231 	object->handle = NULL;
232 	object->backing_object = NULL;
233 	object->backing_object_offset = (vm_ooffset_t) 0;
234 
235 	mtx_lock(&vm_object_list_mtx);
236 	TAILQ_INSERT_TAIL(&vm_object_list, object, object_list);
237 	mtx_unlock(&vm_object_list_mtx);
238 }
239 
240 /*
241  *	vm_object_init:
242  *
243  *	Initialize the VM objects module.
244  */
245 void
246 vm_object_init(void)
247 {
248 	TAILQ_INIT(&vm_object_list);
249 	mtx_init(&vm_object_list_mtx, "vm object_list", NULL, MTX_DEF);
250 
251 	VM_OBJECT_LOCK_INIT(&kernel_object_store, "kernel object");
252 	_vm_object_allocate(OBJT_PHYS, OFF_TO_IDX(VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS),
253 	    kernel_object);
254 
255 	VM_OBJECT_LOCK_INIT(&kmem_object_store, "kmem object");
256 	_vm_object_allocate(OBJT_PHYS, OFF_TO_IDX(VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS),
257 	    kmem_object);
258 
259 	/*
260 	 * The lock portion of struct vm_object must be type stable due
261 	 * to vm_pageout_fallback_object_lock locking a vm object
262 	 * without holding any references to it.
263 	 */
264 	obj_zone = uma_zcreate("VM OBJECT", sizeof (struct vm_object), NULL,
265 #ifdef INVARIANTS
266 	    vm_object_zdtor,
267 #else
268 	    NULL,
269 #endif
270 	    vm_object_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_VM|UMA_ZONE_NOFREE);
271 }
272 
273 void
274 vm_object_clear_flag(vm_object_t object, u_short bits)
275 {
276 
277 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
278 	object->flags &= ~bits;
279 }
280 
281 void
282 vm_object_pip_add(vm_object_t object, short i)
283 {
284 
285 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
286 	object->paging_in_progress += i;
287 }
288 
289 void
290 vm_object_pip_subtract(vm_object_t object, short i)
291 {
292 
293 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
294 	object->paging_in_progress -= i;
295 }
296 
297 void
298 vm_object_pip_wakeup(vm_object_t object)
299 {
300 
301 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
302 	object->paging_in_progress--;
303 	if ((object->flags & OBJ_PIPWNT) && object->paging_in_progress == 0) {
304 		vm_object_clear_flag(object, OBJ_PIPWNT);
305 		wakeup(object);
306 	}
307 }
308 
309 void
310 vm_object_pip_wakeupn(vm_object_t object, short i)
311 {
312 
313 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
314 	if (i)
315 		object->paging_in_progress -= i;
316 	if ((object->flags & OBJ_PIPWNT) && object->paging_in_progress == 0) {
317 		vm_object_clear_flag(object, OBJ_PIPWNT);
318 		wakeup(object);
319 	}
320 }
321 
322 void
323 vm_object_pip_wait(vm_object_t object, char *waitid)
324 {
325 
326 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
327 	while (object->paging_in_progress) {
328 		object->flags |= OBJ_PIPWNT;
329 		msleep(object, VM_OBJECT_MTX(object), PVM, waitid, 0);
330 	}
331 }
332 
333 /*
334  *	vm_object_allocate:
335  *
336  *	Returns a new object with the given size.
337  */
338 vm_object_t
339 vm_object_allocate(objtype_t type, vm_pindex_t size)
340 {
341 	vm_object_t object;
342 
343 	object = (vm_object_t)uma_zalloc(obj_zone, M_WAITOK);
344 	_vm_object_allocate(type, size, object);
345 	return (object);
346 }
347 
348 
349 /*
350  *	vm_object_reference:
351  *
352  *	Gets another reference to the given object.  Note: OBJ_DEAD
353  *	objects can be referenced during final cleaning.
354  */
355 void
356 vm_object_reference(vm_object_t object)
357 {
358 	struct vnode *vp;
359 
360 	if (object == NULL)
361 		return;
362 	VM_OBJECT_LOCK(object);
363 	object->ref_count++;
364 	if (object->type == OBJT_VNODE) {
365 		int vfslocked;
366 
367 		vp = object->handle;
368 		VM_OBJECT_UNLOCK(object);
369 		vfslocked = VFS_LOCK_GIANT(vp->v_mount);
370 		vget(vp, LK_RETRY, curthread);
371 		VFS_UNLOCK_GIANT(vfslocked);
372 	} else
373 		VM_OBJECT_UNLOCK(object);
374 }
375 
376 /*
377  *	vm_object_reference_locked:
378  *
379  *	Gets another reference to the given object.
380  *
381  *	The object must be locked.
382  */
383 void
384 vm_object_reference_locked(vm_object_t object)
385 {
386 	struct vnode *vp;
387 
388 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
389 	KASSERT((object->flags & OBJ_DEAD) == 0,
390 	    ("vm_object_reference_locked: dead object referenced"));
391 	object->ref_count++;
392 	if (object->type == OBJT_VNODE) {
393 		vp = object->handle;
394 		vref(vp);
395 	}
396 }
397 
398 /*
399  * Handle deallocating an object of type OBJT_VNODE.
400  */
401 static void
402 vm_object_vndeallocate(vm_object_t object)
403 {
404 	struct vnode *vp = (struct vnode *) object->handle;
405 
406 	VFS_ASSERT_GIANT(vp->v_mount);
407 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
408 	KASSERT(object->type == OBJT_VNODE,
409 	    ("vm_object_vndeallocate: not a vnode object"));
410 	KASSERT(vp != NULL, ("vm_object_vndeallocate: missing vp"));
411 #ifdef INVARIANTS
412 	if (object->ref_count == 0) {
413 		vprint("vm_object_vndeallocate", vp);
414 		panic("vm_object_vndeallocate: bad object reference count");
415 	}
416 #endif
417 
418 	object->ref_count--;
419 	if (object->ref_count == 0) {
420 		mp_fixme("Unlocked vflag access.");
421 		vp->v_vflag &= ~VV_TEXT;
422 	}
423 	VM_OBJECT_UNLOCK(object);
424 	/*
425 	 * vrele may need a vop lock
426 	 */
427 	vrele(vp);
428 }
429 
430 /*
431  *	vm_object_deallocate:
432  *
433  *	Release a reference to the specified object,
434  *	gained either through a vm_object_allocate
435  *	or a vm_object_reference call.  When all references
436  *	are gone, storage associated with this object
437  *	may be relinquished.
438  *
439  *	No object may be locked.
440  */
441 void
442 vm_object_deallocate(vm_object_t object)
443 {
444 	vm_object_t temp;
445 
446 	while (object != NULL) {
447 		int vfslocked;
448 
449 		vfslocked = 0;
450 	restart:
451 		VM_OBJECT_LOCK(object);
452 		if (object->type == OBJT_VNODE) {
453 			struct vnode *vp = (struct vnode *) object->handle;
454 
455 			/*
456 			 * Conditionally acquire Giant for a vnode-backed
457 			 * object.  We have to be careful since the type of
458 			 * a vnode object can change while the object is
459 			 * unlocked.
460 			 */
461 			if (VFS_NEEDSGIANT(vp->v_mount) && !vfslocked) {
462 				vfslocked = 1;
463 				if (!mtx_trylock(&Giant)) {
464 					VM_OBJECT_UNLOCK(object);
465 					mtx_lock(&Giant);
466 					goto restart;
467 				}
468 			}
469 			vm_object_vndeallocate(object);
470 			VFS_UNLOCK_GIANT(vfslocked);
471 			return;
472 		} else
473 			/*
474 			 * This is to handle the case that the object
475 			 * changed type while we dropped its lock to
476 			 * obtain Giant.
477 			 */
478 			VFS_UNLOCK_GIANT(vfslocked);
479 
480 		KASSERT(object->ref_count != 0,
481 			("vm_object_deallocate: object deallocated too many times: %d", object->type));
482 
483 		/*
484 		 * If the reference count goes to 0 we start calling
485 		 * vm_object_terminate() on the object chain.
486 		 * A ref count of 1 may be a special case depending on the
487 		 * shadow count being 0 or 1.
488 		 */
489 		object->ref_count--;
490 		if (object->ref_count > 1) {
491 			VM_OBJECT_UNLOCK(object);
492 			return;
493 		} else if (object->ref_count == 1) {
494 			if (object->shadow_count == 0) {
495 				vm_object_set_flag(object, OBJ_ONEMAPPING);
496 			} else if ((object->shadow_count == 1) &&
497 			    (object->handle == NULL) &&
498 			    (object->type == OBJT_DEFAULT ||
499 			     object->type == OBJT_SWAP)) {
500 				vm_object_t robject;
501 
502 				robject = LIST_FIRST(&object->shadow_head);
503 				KASSERT(robject != NULL,
504 				    ("vm_object_deallocate: ref_count: %d, shadow_count: %d",
505 					 object->ref_count,
506 					 object->shadow_count));
507 				if (!VM_OBJECT_TRYLOCK(robject)) {
508 					/*
509 					 * Avoid a potential deadlock.
510 					 */
511 					object->ref_count++;
512 					VM_OBJECT_UNLOCK(object);
513 					/*
514 					 * More likely than not the thread
515 					 * holding robject's lock has lower
516 					 * priority than the current thread.
517 					 * Let the lower priority thread run.
518 					 */
519 					pause("vmo_de", 1);
520 					continue;
521 				}
522 				/*
523 				 * Collapse object into its shadow unless its
524 				 * shadow is dead.  In that case, object will
525 				 * be deallocated by the thread that is
526 				 * deallocating its shadow.
527 				 */
528 				if ((robject->flags & OBJ_DEAD) == 0 &&
529 				    (robject->handle == NULL) &&
530 				    (robject->type == OBJT_DEFAULT ||
531 				     robject->type == OBJT_SWAP)) {
532 
533 					robject->ref_count++;
534 retry:
535 					if (robject->paging_in_progress) {
536 						VM_OBJECT_UNLOCK(object);
537 						vm_object_pip_wait(robject,
538 						    "objde1");
539 						temp = robject->backing_object;
540 						if (object == temp) {
541 							VM_OBJECT_LOCK(object);
542 							goto retry;
543 						}
544 					} else if (object->paging_in_progress) {
545 						VM_OBJECT_UNLOCK(robject);
546 						object->flags |= OBJ_PIPWNT;
547 						msleep(object,
548 						    VM_OBJECT_MTX(object),
549 						    PDROP | PVM, "objde2", 0);
550 						VM_OBJECT_LOCK(robject);
551 						temp = robject->backing_object;
552 						if (object == temp) {
553 							VM_OBJECT_LOCK(object);
554 							goto retry;
555 						}
556 					} else
557 						VM_OBJECT_UNLOCK(object);
558 
559 					if (robject->ref_count == 1) {
560 						robject->ref_count--;
561 						object = robject;
562 						goto doterm;
563 					}
564 					object = robject;
565 					vm_object_collapse(object);
566 					VM_OBJECT_UNLOCK(object);
567 					continue;
568 				}
569 				VM_OBJECT_UNLOCK(robject);
570 			}
571 			VM_OBJECT_UNLOCK(object);
572 			return;
573 		}
574 doterm:
575 		temp = object->backing_object;
576 		if (temp != NULL) {
577 			VM_OBJECT_LOCK(temp);
578 			LIST_REMOVE(object, shadow_list);
579 			temp->shadow_count--;
580 			temp->generation++;
581 			VM_OBJECT_UNLOCK(temp);
582 			object->backing_object = NULL;
583 		}
584 		/*
585 		 * Don't double-terminate, we could be in a termination
586 		 * recursion due to the terminate having to sync data
587 		 * to disk.
588 		 */
589 		if ((object->flags & OBJ_DEAD) == 0)
590 			vm_object_terminate(object);
591 		else
592 			VM_OBJECT_UNLOCK(object);
593 		object = temp;
594 	}
595 }
596 
597 /*
598  *	vm_object_terminate actually destroys the specified object, freeing
599  *	up all previously used resources.
600  *
601  *	The object must be locked.
602  *	This routine may block.
603  */
604 void
605 vm_object_terminate(vm_object_t object)
606 {
607 	vm_page_t p;
608 
609 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
610 
611 	/*
612 	 * Make sure no one uses us.
613 	 */
614 	vm_object_set_flag(object, OBJ_DEAD);
615 
616 	/*
617 	 * wait for the pageout daemon to be done with the object
618 	 */
619 	vm_object_pip_wait(object, "objtrm");
620 
621 	KASSERT(!object->paging_in_progress,
622 		("vm_object_terminate: pageout in progress"));
623 
624 	/*
625 	 * Clean and free the pages, as appropriate. All references to the
626 	 * object are gone, so we don't need to lock it.
627 	 */
628 	if (object->type == OBJT_VNODE) {
629 		struct vnode *vp = (struct vnode *)object->handle;
630 
631 		/*
632 		 * Clean pages and flush buffers.
633 		 */
634 		vm_object_page_clean(object, 0, 0, OBJPC_SYNC);
635 		VM_OBJECT_UNLOCK(object);
636 
637 		vinvalbuf(vp, V_SAVE, NULL, 0, 0);
638 
639 		VM_OBJECT_LOCK(object);
640 	}
641 
642 	KASSERT(object->ref_count == 0,
643 		("vm_object_terminate: object with references, ref_count=%d",
644 		object->ref_count));
645 
646 	/*
647 	 * Now free any remaining pages. For internal objects, this also
648 	 * removes them from paging queues. Don't free wired pages, just
649 	 * remove them from the object.
650 	 */
651 	vm_page_lock_queues();
652 	while ((p = TAILQ_FIRST(&object->memq)) != NULL) {
653 		KASSERT(!p->busy && (p->oflags & VPO_BUSY) == 0,
654 			("vm_object_terminate: freeing busy page %p "
655 			"p->busy = %d, p->flags %x\n", p, p->busy, p->flags));
656 		if (p->wire_count == 0) {
657 			vm_page_free(p);
658 			cnt.v_pfree++;
659 		} else {
660 			vm_page_remove(p);
661 		}
662 	}
663 	vm_page_unlock_queues();
664 
665 	/*
666 	 * Let the pager know object is dead.
667 	 */
668 	vm_pager_deallocate(object);
669 	VM_OBJECT_UNLOCK(object);
670 
671 	/*
672 	 * Remove the object from the global object list.
673 	 */
674 	mtx_lock(&vm_object_list_mtx);
675 	TAILQ_REMOVE(&vm_object_list, object, object_list);
676 	mtx_unlock(&vm_object_list_mtx);
677 
678 	/*
679 	 * Free the space for the object.
680 	 */
681 	uma_zfree(obj_zone, object);
682 }
683 
684 /*
685  *	vm_object_page_clean
686  *
687  *	Clean all dirty pages in the specified range of object.  Leaves page
688  * 	on whatever queue it is currently on.   If NOSYNC is set then do not
689  *	write out pages with VPO_NOSYNC set (originally comes from MAP_NOSYNC),
690  *	leaving the object dirty.
691  *
692  *	When stuffing pages asynchronously, allow clustering.  XXX we need a
693  *	synchronous clustering mode implementation.
694  *
695  *	Odd semantics: if start == end, we clean everything.
696  *
697  *	The object must be locked.
698  */
699 void
700 vm_object_page_clean(vm_object_t object, vm_pindex_t start, vm_pindex_t end, int flags)
701 {
702 	vm_page_t p, np;
703 	vm_pindex_t tstart, tend;
704 	vm_pindex_t pi;
705 	int clearobjflags;
706 	int pagerflags;
707 	int curgeneration;
708 
709 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
710 	if (object->type != OBJT_VNODE ||
711 		(object->flags & OBJ_MIGHTBEDIRTY) == 0)
712 		return;
713 
714 	pagerflags = (flags & (OBJPC_SYNC | OBJPC_INVAL)) ? VM_PAGER_PUT_SYNC : VM_PAGER_CLUSTER_OK;
715 	pagerflags |= (flags & OBJPC_INVAL) ? VM_PAGER_PUT_INVAL : 0;
716 
717 	vm_object_set_flag(object, OBJ_CLEANING);
718 
719 	tstart = start;
720 	if (end == 0) {
721 		tend = object->size;
722 	} else {
723 		tend = end;
724 	}
725 
726 	vm_page_lock_queues();
727 	/*
728 	 * If the caller is smart and only msync()s a range he knows is
729 	 * dirty, we may be able to avoid an object scan.  This results in
730 	 * a phenominal improvement in performance.  We cannot do this
731 	 * as a matter of course because the object may be huge - e.g.
732 	 * the size might be in the gigabytes or terrabytes.
733 	 */
734 	if (msync_flush_flags & MSYNC_FLUSH_HARDSEQ) {
735 		vm_pindex_t tscan;
736 		int scanlimit;
737 		int scanreset;
738 
739 		scanreset = object->resident_page_count / EASY_SCAN_FACTOR;
740 		if (scanreset < 16)
741 			scanreset = 16;
742 		pagerflags |= VM_PAGER_IGNORE_CLEANCHK;
743 
744 		scanlimit = scanreset;
745 		tscan = tstart;
746 		while (tscan < tend) {
747 			curgeneration = object->generation;
748 			p = vm_page_lookup(object, tscan);
749 			if (p == NULL || p->valid == 0 ||
750 			    VM_PAGE_INQUEUE1(p, PQ_CACHE)) {
751 				if (--scanlimit == 0)
752 					break;
753 				++tscan;
754 				continue;
755 			}
756 			vm_page_test_dirty(p);
757 			if ((p->dirty & p->valid) == 0) {
758 				if (--scanlimit == 0)
759 					break;
760 				++tscan;
761 				continue;
762 			}
763 			/*
764 			 * If we have been asked to skip nosync pages and
765 			 * this is a nosync page, we can't continue.
766 			 */
767 			if ((flags & OBJPC_NOSYNC) && (p->oflags & VPO_NOSYNC)) {
768 				if (--scanlimit == 0)
769 					break;
770 				++tscan;
771 				continue;
772 			}
773 			scanlimit = scanreset;
774 
775 			/*
776 			 * This returns 0 if it was unable to busy the first
777 			 * page (i.e. had to sleep).
778 			 */
779 			tscan += vm_object_page_collect_flush(object, p, curgeneration, pagerflags);
780 		}
781 
782 		/*
783 		 * If everything was dirty and we flushed it successfully,
784 		 * and the requested range is not the entire object, we
785 		 * don't have to mess with CLEANCHK or MIGHTBEDIRTY and can
786 		 * return immediately.
787 		 */
788 		if (tscan >= tend && (tstart || tend < object->size)) {
789 			vm_page_unlock_queues();
790 			vm_object_clear_flag(object, OBJ_CLEANING);
791 			return;
792 		}
793 		pagerflags &= ~VM_PAGER_IGNORE_CLEANCHK;
794 	}
795 
796 	/*
797 	 * Generally set CLEANCHK interlock and make the page read-only so
798 	 * we can then clear the object flags.
799 	 *
800 	 * However, if this is a nosync mmap then the object is likely to
801 	 * stay dirty so do not mess with the page and do not clear the
802 	 * object flags.
803 	 */
804 	clearobjflags = 1;
805 	TAILQ_FOREACH(p, &object->memq, listq) {
806 		p->oflags |= VPO_CLEANCHK;
807 		if ((flags & OBJPC_NOSYNC) && (p->oflags & VPO_NOSYNC))
808 			clearobjflags = 0;
809 		else
810 			pmap_remove_write(p);
811 	}
812 
813 	if (clearobjflags && (tstart == 0) && (tend == object->size)) {
814 		struct vnode *vp;
815 
816 		vm_object_clear_flag(object, OBJ_MIGHTBEDIRTY);
817 		if (object->type == OBJT_VNODE &&
818 		    (vp = (struct vnode *)object->handle) != NULL) {
819 			VI_LOCK(vp);
820 			if (vp->v_iflag & VI_OBJDIRTY)
821 				vp->v_iflag &= ~VI_OBJDIRTY;
822 			VI_UNLOCK(vp);
823 		}
824 	}
825 
826 rescan:
827 	curgeneration = object->generation;
828 
829 	for (p = TAILQ_FIRST(&object->memq); p; p = np) {
830 		int n;
831 
832 		np = TAILQ_NEXT(p, listq);
833 
834 again:
835 		pi = p->pindex;
836 		if ((p->oflags & VPO_CLEANCHK) == 0 ||
837 			(pi < tstart) || (pi >= tend) ||
838 			(p->valid == 0) ||
839 		    VM_PAGE_INQUEUE1(p, PQ_CACHE)) {
840 			p->oflags &= ~VPO_CLEANCHK;
841 			continue;
842 		}
843 
844 		vm_page_test_dirty(p);
845 		if ((p->dirty & p->valid) == 0) {
846 			p->oflags &= ~VPO_CLEANCHK;
847 			continue;
848 		}
849 
850 		/*
851 		 * If we have been asked to skip nosync pages and this is a
852 		 * nosync page, skip it.  Note that the object flags were
853 		 * not cleared in this case so we do not have to set them.
854 		 */
855 		if ((flags & OBJPC_NOSYNC) && (p->oflags & VPO_NOSYNC)) {
856 			p->oflags &= ~VPO_CLEANCHK;
857 			continue;
858 		}
859 
860 		n = vm_object_page_collect_flush(object, p,
861 			curgeneration, pagerflags);
862 		if (n == 0)
863 			goto rescan;
864 
865 		if (object->generation != curgeneration)
866 			goto rescan;
867 
868 		/*
869 		 * Try to optimize the next page.  If we can't we pick up
870 		 * our (random) scan where we left off.
871 		 */
872 		if (msync_flush_flags & MSYNC_FLUSH_SOFTSEQ) {
873 			if ((p = vm_page_lookup(object, pi + n)) != NULL)
874 				goto again;
875 		}
876 	}
877 	vm_page_unlock_queues();
878 #if 0
879 	VOP_FSYNC(vp, (pagerflags & VM_PAGER_PUT_SYNC)?MNT_WAIT:0, curproc);
880 #endif
881 
882 	vm_object_clear_flag(object, OBJ_CLEANING);
883 	return;
884 }
885 
886 static int
887 vm_object_page_collect_flush(vm_object_t object, vm_page_t p, int curgeneration, int pagerflags)
888 {
889 	int runlen;
890 	int maxf;
891 	int chkb;
892 	int maxb;
893 	int i;
894 	vm_pindex_t pi;
895 	vm_page_t maf[vm_pageout_page_count];
896 	vm_page_t mab[vm_pageout_page_count];
897 	vm_page_t ma[vm_pageout_page_count];
898 
899 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
900 	pi = p->pindex;
901 	while (vm_page_sleep_if_busy(p, TRUE, "vpcwai")) {
902 		vm_page_lock_queues();
903 		if (object->generation != curgeneration) {
904 			return(0);
905 		}
906 	}
907 	maxf = 0;
908 	for(i = 1; i < vm_pageout_page_count; i++) {
909 		vm_page_t tp;
910 
911 		if ((tp = vm_page_lookup(object, pi + i)) != NULL) {
912 			if ((tp->oflags & VPO_BUSY) ||
913 				((pagerflags & VM_PAGER_IGNORE_CLEANCHK) == 0 &&
914 				 (tp->oflags & VPO_CLEANCHK) == 0) ||
915 				(tp->busy != 0))
916 				break;
917 			if (VM_PAGE_INQUEUE1(tp, PQ_CACHE)) {
918 				tp->oflags &= ~VPO_CLEANCHK;
919 				break;
920 			}
921 			vm_page_test_dirty(tp);
922 			if ((tp->dirty & tp->valid) == 0) {
923 				tp->oflags &= ~VPO_CLEANCHK;
924 				break;
925 			}
926 			maf[ i - 1 ] = tp;
927 			maxf++;
928 			continue;
929 		}
930 		break;
931 	}
932 
933 	maxb = 0;
934 	chkb = vm_pageout_page_count -  maxf;
935 	if (chkb) {
936 		for(i = 1; i < chkb;i++) {
937 			vm_page_t tp;
938 
939 			if ((tp = vm_page_lookup(object, pi - i)) != NULL) {
940 				if ((tp->oflags & VPO_BUSY) ||
941 					((pagerflags & VM_PAGER_IGNORE_CLEANCHK) == 0 &&
942 					 (tp->oflags & VPO_CLEANCHK) == 0) ||
943 					(tp->busy != 0))
944 					break;
945 				if (VM_PAGE_INQUEUE1(tp, PQ_CACHE)) {
946 					tp->oflags &= ~VPO_CLEANCHK;
947 					break;
948 				}
949 				vm_page_test_dirty(tp);
950 				if ((tp->dirty & tp->valid) == 0) {
951 					tp->oflags &= ~VPO_CLEANCHK;
952 					break;
953 				}
954 				mab[ i - 1 ] = tp;
955 				maxb++;
956 				continue;
957 			}
958 			break;
959 		}
960 	}
961 
962 	for(i = 0; i < maxb; i++) {
963 		int index = (maxb - i) - 1;
964 		ma[index] = mab[i];
965 		ma[index]->oflags &= ~VPO_CLEANCHK;
966 	}
967 	p->oflags &= ~VPO_CLEANCHK;
968 	ma[maxb] = p;
969 	for(i = 0; i < maxf; i++) {
970 		int index = (maxb + i) + 1;
971 		ma[index] = maf[i];
972 		ma[index]->oflags &= ~VPO_CLEANCHK;
973 	}
974 	runlen = maxb + maxf + 1;
975 
976 	vm_pageout_flush(ma, runlen, pagerflags);
977 	for (i = 0; i < runlen; i++) {
978 		if (ma[i]->valid & ma[i]->dirty) {
979 			pmap_remove_write(ma[i]);
980 			ma[i]->oflags |= VPO_CLEANCHK;
981 
982 			/*
983 			 * maxf will end up being the actual number of pages
984 			 * we wrote out contiguously, non-inclusive of the
985 			 * first page.  We do not count look-behind pages.
986 			 */
987 			if (i >= maxb + 1 && (maxf > i - maxb - 1))
988 				maxf = i - maxb - 1;
989 		}
990 	}
991 	return(maxf + 1);
992 }
993 
994 /*
995  * Note that there is absolutely no sense in writing out
996  * anonymous objects, so we track down the vnode object
997  * to write out.
998  * We invalidate (remove) all pages from the address space
999  * for semantic correctness.
1000  *
1001  * Note: certain anonymous maps, such as MAP_NOSYNC maps,
1002  * may start out with a NULL object.
1003  */
1004 void
1005 vm_object_sync(vm_object_t object, vm_ooffset_t offset, vm_size_t size,
1006     boolean_t syncio, boolean_t invalidate)
1007 {
1008 	vm_object_t backing_object;
1009 	struct vnode *vp;
1010 	struct mount *mp;
1011 	int flags;
1012 
1013 	if (object == NULL)
1014 		return;
1015 	VM_OBJECT_LOCK(object);
1016 	while ((backing_object = object->backing_object) != NULL) {
1017 		VM_OBJECT_LOCK(backing_object);
1018 		offset += object->backing_object_offset;
1019 		VM_OBJECT_UNLOCK(object);
1020 		object = backing_object;
1021 		if (object->size < OFF_TO_IDX(offset + size))
1022 			size = IDX_TO_OFF(object->size) - offset;
1023 	}
1024 	/*
1025 	 * Flush pages if writing is allowed, invalidate them
1026 	 * if invalidation requested.  Pages undergoing I/O
1027 	 * will be ignored by vm_object_page_remove().
1028 	 *
1029 	 * We cannot lock the vnode and then wait for paging
1030 	 * to complete without deadlocking against vm_fault.
1031 	 * Instead we simply call vm_object_page_remove() and
1032 	 * allow it to block internally on a page-by-page
1033 	 * basis when it encounters pages undergoing async
1034 	 * I/O.
1035 	 */
1036 	if (object->type == OBJT_VNODE &&
1037 	    (object->flags & OBJ_MIGHTBEDIRTY) != 0) {
1038 		int vfslocked;
1039 		vp = object->handle;
1040 		VM_OBJECT_UNLOCK(object);
1041 		(void) vn_start_write(vp, &mp, V_WAIT);
1042 		vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1043 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, curthread);
1044 		flags = (syncio || invalidate) ? OBJPC_SYNC : 0;
1045 		flags |= invalidate ? OBJPC_INVAL : 0;
1046 		VM_OBJECT_LOCK(object);
1047 		vm_object_page_clean(object,
1048 		    OFF_TO_IDX(offset),
1049 		    OFF_TO_IDX(offset + size + PAGE_MASK),
1050 		    flags);
1051 		VM_OBJECT_UNLOCK(object);
1052 		VOP_UNLOCK(vp, 0, curthread);
1053 		VFS_UNLOCK_GIANT(vfslocked);
1054 		vn_finished_write(mp);
1055 		VM_OBJECT_LOCK(object);
1056 	}
1057 	if ((object->type == OBJT_VNODE ||
1058 	     object->type == OBJT_DEVICE) && invalidate) {
1059 		boolean_t purge;
1060 		purge = old_msync || (object->type == OBJT_DEVICE);
1061 		vm_object_page_remove(object,
1062 		    OFF_TO_IDX(offset),
1063 		    OFF_TO_IDX(offset + size + PAGE_MASK),
1064 		    purge ? FALSE : TRUE);
1065 	}
1066 	VM_OBJECT_UNLOCK(object);
1067 }
1068 
1069 /*
1070  *	vm_object_madvise:
1071  *
1072  *	Implements the madvise function at the object/page level.
1073  *
1074  *	MADV_WILLNEED	(any object)
1075  *
1076  *	    Activate the specified pages if they are resident.
1077  *
1078  *	MADV_DONTNEED	(any object)
1079  *
1080  *	    Deactivate the specified pages if they are resident.
1081  *
1082  *	MADV_FREE	(OBJT_DEFAULT/OBJT_SWAP objects,
1083  *			 OBJ_ONEMAPPING only)
1084  *
1085  *	    Deactivate and clean the specified pages if they are
1086  *	    resident.  This permits the process to reuse the pages
1087  *	    without faulting or the kernel to reclaim the pages
1088  *	    without I/O.
1089  */
1090 void
1091 vm_object_madvise(vm_object_t object, vm_pindex_t pindex, int count, int advise)
1092 {
1093 	vm_pindex_t end, tpindex;
1094 	vm_object_t backing_object, tobject;
1095 	vm_page_t m;
1096 
1097 	if (object == NULL)
1098 		return;
1099 	VM_OBJECT_LOCK(object);
1100 	end = pindex + count;
1101 	/*
1102 	 * Locate and adjust resident pages
1103 	 */
1104 	for (; pindex < end; pindex += 1) {
1105 relookup:
1106 		tobject = object;
1107 		tpindex = pindex;
1108 shadowlookup:
1109 		/*
1110 		 * MADV_FREE only operates on OBJT_DEFAULT or OBJT_SWAP pages
1111 		 * and those pages must be OBJ_ONEMAPPING.
1112 		 */
1113 		if (advise == MADV_FREE) {
1114 			if ((tobject->type != OBJT_DEFAULT &&
1115 			     tobject->type != OBJT_SWAP) ||
1116 			    (tobject->flags & OBJ_ONEMAPPING) == 0) {
1117 				goto unlock_tobject;
1118 			}
1119 		}
1120 		m = vm_page_lookup(tobject, tpindex);
1121 		if (m == NULL) {
1122 			/*
1123 			 * There may be swap even if there is no backing page
1124 			 */
1125 			if (advise == MADV_FREE && tobject->type == OBJT_SWAP)
1126 				swap_pager_freespace(tobject, tpindex, 1);
1127 			/*
1128 			 * next object
1129 			 */
1130 			backing_object = tobject->backing_object;
1131 			if (backing_object == NULL)
1132 				goto unlock_tobject;
1133 			VM_OBJECT_LOCK(backing_object);
1134 			tpindex += OFF_TO_IDX(tobject->backing_object_offset);
1135 			if (tobject != object)
1136 				VM_OBJECT_UNLOCK(tobject);
1137 			tobject = backing_object;
1138 			goto shadowlookup;
1139 		}
1140 		/*
1141 		 * If the page is busy or not in a normal active state,
1142 		 * we skip it.  If the page is not managed there are no
1143 		 * page queues to mess with.  Things can break if we mess
1144 		 * with pages in any of the below states.
1145 		 */
1146 		vm_page_lock_queues();
1147 		if (m->hold_count ||
1148 		    m->wire_count ||
1149 		    (m->flags & PG_UNMANAGED) ||
1150 		    m->valid != VM_PAGE_BITS_ALL) {
1151 			vm_page_unlock_queues();
1152 			goto unlock_tobject;
1153 		}
1154 		if ((m->oflags & VPO_BUSY) || m->busy) {
1155 			vm_page_flag_set(m, PG_REFERENCED);
1156 			vm_page_unlock_queues();
1157 			if (object != tobject)
1158 				VM_OBJECT_UNLOCK(object);
1159 			m->oflags |= VPO_WANTED;
1160 			msleep(m, VM_OBJECT_MTX(tobject), PDROP | PVM, "madvpo", 0);
1161 			VM_OBJECT_LOCK(object);
1162   			goto relookup;
1163 		}
1164 		if (advise == MADV_WILLNEED) {
1165 			vm_page_activate(m);
1166 		} else if (advise == MADV_DONTNEED) {
1167 			vm_page_dontneed(m);
1168 		} else if (advise == MADV_FREE) {
1169 			/*
1170 			 * Mark the page clean.  This will allow the page
1171 			 * to be freed up by the system.  However, such pages
1172 			 * are often reused quickly by malloc()/free()
1173 			 * so we do not do anything that would cause
1174 			 * a page fault if we can help it.
1175 			 *
1176 			 * Specifically, we do not try to actually free
1177 			 * the page now nor do we try to put it in the
1178 			 * cache (which would cause a page fault on reuse).
1179 			 *
1180 			 * But we do make the page is freeable as we
1181 			 * can without actually taking the step of unmapping
1182 			 * it.
1183 			 */
1184 			pmap_clear_modify(m);
1185 			m->dirty = 0;
1186 			m->act_count = 0;
1187 			vm_page_dontneed(m);
1188 		}
1189 		vm_page_unlock_queues();
1190 		if (advise == MADV_FREE && tobject->type == OBJT_SWAP)
1191 			swap_pager_freespace(tobject, tpindex, 1);
1192 unlock_tobject:
1193 		if (tobject != object)
1194 			VM_OBJECT_UNLOCK(tobject);
1195 	}
1196 	VM_OBJECT_UNLOCK(object);
1197 }
1198 
1199 /*
1200  *	vm_object_shadow:
1201  *
1202  *	Create a new object which is backed by the
1203  *	specified existing object range.  The source
1204  *	object reference is deallocated.
1205  *
1206  *	The new object and offset into that object
1207  *	are returned in the source parameters.
1208  */
1209 void
1210 vm_object_shadow(
1211 	vm_object_t *object,	/* IN/OUT */
1212 	vm_ooffset_t *offset,	/* IN/OUT */
1213 	vm_size_t length)
1214 {
1215 	vm_object_t source;
1216 	vm_object_t result;
1217 
1218 	source = *object;
1219 
1220 	/*
1221 	 * Don't create the new object if the old object isn't shared.
1222 	 */
1223 	if (source != NULL) {
1224 		VM_OBJECT_LOCK(source);
1225 		if (source->ref_count == 1 &&
1226 		    source->handle == NULL &&
1227 		    (source->type == OBJT_DEFAULT ||
1228 		     source->type == OBJT_SWAP)) {
1229 			VM_OBJECT_UNLOCK(source);
1230 			return;
1231 		}
1232 		VM_OBJECT_UNLOCK(source);
1233 	}
1234 
1235 	/*
1236 	 * Allocate a new object with the given length.
1237 	 */
1238 	result = vm_object_allocate(OBJT_DEFAULT, length);
1239 
1240 	/*
1241 	 * The new object shadows the source object, adding a reference to it.
1242 	 * Our caller changes his reference to point to the new object,
1243 	 * removing a reference to the source object.  Net result: no change
1244 	 * of reference count.
1245 	 *
1246 	 * Try to optimize the result object's page color when shadowing
1247 	 * in order to maintain page coloring consistency in the combined
1248 	 * shadowed object.
1249 	 */
1250 	result->backing_object = source;
1251 	/*
1252 	 * Store the offset into the source object, and fix up the offset into
1253 	 * the new object.
1254 	 */
1255 	result->backing_object_offset = *offset;
1256 	if (source != NULL) {
1257 		VM_OBJECT_LOCK(source);
1258 		LIST_INSERT_HEAD(&source->shadow_head, result, shadow_list);
1259 		source->shadow_count++;
1260 		source->generation++;
1261 		if (length < source->size)
1262 			length = source->size;
1263 		if (length > PQ_MAXLENGTH || source->generation > 1)
1264 			length = PQ_MAXLENGTH;
1265 		result->pg_color = (source->pg_color +
1266 		    length * source->generation) & PQ_COLORMASK;
1267 		result->flags |= source->flags & OBJ_NEEDGIANT;
1268 		VM_OBJECT_UNLOCK(source);
1269 		next_index = (result->pg_color + PQ_MAXLENGTH) & PQ_COLORMASK;
1270 	}
1271 
1272 
1273 	/*
1274 	 * Return the new things
1275 	 */
1276 	*offset = 0;
1277 	*object = result;
1278 }
1279 
1280 /*
1281  *	vm_object_split:
1282  *
1283  * Split the pages in a map entry into a new object.  This affords
1284  * easier removal of unused pages, and keeps object inheritance from
1285  * being a negative impact on memory usage.
1286  */
1287 void
1288 vm_object_split(vm_map_entry_t entry)
1289 {
1290 	vm_page_t m, m_next;
1291 	vm_object_t orig_object, new_object, source;
1292 	vm_pindex_t idx, offidxstart;
1293 	vm_size_t size;
1294 
1295 	orig_object = entry->object.vm_object;
1296 	if (orig_object->type != OBJT_DEFAULT && orig_object->type != OBJT_SWAP)
1297 		return;
1298 	if (orig_object->ref_count <= 1)
1299 		return;
1300 	VM_OBJECT_UNLOCK(orig_object);
1301 
1302 	offidxstart = OFF_TO_IDX(entry->offset);
1303 	size = atop(entry->end - entry->start);
1304 
1305 	/*
1306 	 * If swap_pager_copy() is later called, it will convert new_object
1307 	 * into a swap object.
1308 	 */
1309 	new_object = vm_object_allocate(OBJT_DEFAULT, size);
1310 
1311 	/*
1312 	 * At this point, the new object is still private, so the order in
1313 	 * which the original and new objects are locked does not matter.
1314 	 */
1315 	VM_OBJECT_LOCK(new_object);
1316 	VM_OBJECT_LOCK(orig_object);
1317 	source = orig_object->backing_object;
1318 	if (source != NULL) {
1319 		VM_OBJECT_LOCK(source);
1320 		if ((source->flags & OBJ_DEAD) != 0) {
1321 			VM_OBJECT_UNLOCK(source);
1322 			VM_OBJECT_UNLOCK(orig_object);
1323 			VM_OBJECT_UNLOCK(new_object);
1324 			vm_object_deallocate(new_object);
1325 			VM_OBJECT_LOCK(orig_object);
1326 			return;
1327 		}
1328 		LIST_INSERT_HEAD(&source->shadow_head,
1329 				  new_object, shadow_list);
1330 		source->shadow_count++;
1331 		source->generation++;
1332 		vm_object_reference_locked(source);	/* for new_object */
1333 		vm_object_clear_flag(source, OBJ_ONEMAPPING);
1334 		VM_OBJECT_UNLOCK(source);
1335 		new_object->backing_object_offset =
1336 			orig_object->backing_object_offset + entry->offset;
1337 		new_object->backing_object = source;
1338 	}
1339 	new_object->flags |= orig_object->flags & OBJ_NEEDGIANT;
1340 retry:
1341 	if ((m = TAILQ_FIRST(&orig_object->memq)) != NULL) {
1342 		if (m->pindex < offidxstart) {
1343 			m = vm_page_splay(offidxstart, orig_object->root);
1344 			if ((orig_object->root = m)->pindex < offidxstart)
1345 				m = TAILQ_NEXT(m, listq);
1346 		}
1347 	}
1348 	vm_page_lock_queues();
1349 	for (; m != NULL && (idx = m->pindex - offidxstart) < size;
1350 	    m = m_next) {
1351 		m_next = TAILQ_NEXT(m, listq);
1352 
1353 		/*
1354 		 * We must wait for pending I/O to complete before we can
1355 		 * rename the page.
1356 		 *
1357 		 * We do not have to VM_PROT_NONE the page as mappings should
1358 		 * not be changed by this operation.
1359 		 */
1360 		if ((m->oflags & VPO_BUSY) || m->busy) {
1361 			vm_page_flag_set(m, PG_REFERENCED);
1362 			vm_page_unlock_queues();
1363 			VM_OBJECT_UNLOCK(new_object);
1364 			m->oflags |= VPO_WANTED;
1365 			msleep(m, VM_OBJECT_MTX(orig_object), PVM, "spltwt", 0);
1366 			VM_OBJECT_LOCK(new_object);
1367 			goto retry;
1368 		}
1369 		vm_page_rename(m, new_object, idx);
1370 		/* page automatically made dirty by rename and cache handled */
1371 		vm_page_busy(m);
1372 	}
1373 	vm_page_unlock_queues();
1374 	if (orig_object->type == OBJT_SWAP) {
1375 		/*
1376 		 * swap_pager_copy() can sleep, in which case the orig_object's
1377 		 * and new_object's locks are released and reacquired.
1378 		 */
1379 		swap_pager_copy(orig_object, new_object, offidxstart, 0);
1380 	}
1381 	VM_OBJECT_UNLOCK(orig_object);
1382 	TAILQ_FOREACH(m, &new_object->memq, listq)
1383 		vm_page_wakeup(m);
1384 	VM_OBJECT_UNLOCK(new_object);
1385 	entry->object.vm_object = new_object;
1386 	entry->offset = 0LL;
1387 	vm_object_deallocate(orig_object);
1388 	VM_OBJECT_LOCK(new_object);
1389 }
1390 
1391 #define	OBSC_TEST_ALL_SHADOWED	0x0001
1392 #define	OBSC_COLLAPSE_NOWAIT	0x0002
1393 #define	OBSC_COLLAPSE_WAIT	0x0004
1394 
1395 static int
1396 vm_object_backing_scan(vm_object_t object, int op)
1397 {
1398 	int r = 1;
1399 	vm_page_t p;
1400 	vm_object_t backing_object;
1401 	vm_pindex_t backing_offset_index;
1402 
1403 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
1404 	VM_OBJECT_LOCK_ASSERT(object->backing_object, MA_OWNED);
1405 
1406 	backing_object = object->backing_object;
1407 	backing_offset_index = OFF_TO_IDX(object->backing_object_offset);
1408 
1409 	/*
1410 	 * Initial conditions
1411 	 */
1412 	if (op & OBSC_TEST_ALL_SHADOWED) {
1413 		/*
1414 		 * We do not want to have to test for the existence of
1415 		 * swap pages in the backing object.  XXX but with the
1416 		 * new swapper this would be pretty easy to do.
1417 		 *
1418 		 * XXX what about anonymous MAP_SHARED memory that hasn't
1419 		 * been ZFOD faulted yet?  If we do not test for this, the
1420 		 * shadow test may succeed! XXX
1421 		 */
1422 		if (backing_object->type != OBJT_DEFAULT) {
1423 			return (0);
1424 		}
1425 	}
1426 	if (op & OBSC_COLLAPSE_WAIT) {
1427 		vm_object_set_flag(backing_object, OBJ_DEAD);
1428 	}
1429 
1430 	/*
1431 	 * Our scan
1432 	 */
1433 	p = TAILQ_FIRST(&backing_object->memq);
1434 	while (p) {
1435 		vm_page_t next = TAILQ_NEXT(p, listq);
1436 		vm_pindex_t new_pindex = p->pindex - backing_offset_index;
1437 
1438 		if (op & OBSC_TEST_ALL_SHADOWED) {
1439 			vm_page_t pp;
1440 
1441 			/*
1442 			 * Ignore pages outside the parent object's range
1443 			 * and outside the parent object's mapping of the
1444 			 * backing object.
1445 			 *
1446 			 * note that we do not busy the backing object's
1447 			 * page.
1448 			 */
1449 			if (
1450 			    p->pindex < backing_offset_index ||
1451 			    new_pindex >= object->size
1452 			) {
1453 				p = next;
1454 				continue;
1455 			}
1456 
1457 			/*
1458 			 * See if the parent has the page or if the parent's
1459 			 * object pager has the page.  If the parent has the
1460 			 * page but the page is not valid, the parent's
1461 			 * object pager must have the page.
1462 			 *
1463 			 * If this fails, the parent does not completely shadow
1464 			 * the object and we might as well give up now.
1465 			 */
1466 
1467 			pp = vm_page_lookup(object, new_pindex);
1468 			if (
1469 			    (pp == NULL || pp->valid == 0) &&
1470 			    !vm_pager_has_page(object, new_pindex, NULL, NULL)
1471 			) {
1472 				r = 0;
1473 				break;
1474 			}
1475 		}
1476 
1477 		/*
1478 		 * Check for busy page
1479 		 */
1480 		if (op & (OBSC_COLLAPSE_WAIT | OBSC_COLLAPSE_NOWAIT)) {
1481 			vm_page_t pp;
1482 
1483 			if (op & OBSC_COLLAPSE_NOWAIT) {
1484 				if ((p->oflags & VPO_BUSY) ||
1485 				    !p->valid ||
1486 				    p->busy) {
1487 					p = next;
1488 					continue;
1489 				}
1490 			} else if (op & OBSC_COLLAPSE_WAIT) {
1491 				if ((p->oflags & VPO_BUSY) || p->busy) {
1492 					vm_page_lock_queues();
1493 					vm_page_flag_set(p, PG_REFERENCED);
1494 					vm_page_unlock_queues();
1495 					VM_OBJECT_UNLOCK(object);
1496 					p->oflags |= VPO_WANTED;
1497 					msleep(p, VM_OBJECT_MTX(backing_object),
1498 					    PDROP | PVM, "vmocol", 0);
1499 					VM_OBJECT_LOCK(object);
1500 					VM_OBJECT_LOCK(backing_object);
1501 					/*
1502 					 * If we slept, anything could have
1503 					 * happened.  Since the object is
1504 					 * marked dead, the backing offset
1505 					 * should not have changed so we
1506 					 * just restart our scan.
1507 					 */
1508 					p = TAILQ_FIRST(&backing_object->memq);
1509 					continue;
1510 				}
1511 			}
1512 
1513 			KASSERT(
1514 			    p->object == backing_object,
1515 			    ("vm_object_backing_scan: object mismatch")
1516 			);
1517 
1518 			/*
1519 			 * Destroy any associated swap
1520 			 */
1521 			if (backing_object->type == OBJT_SWAP) {
1522 				swap_pager_freespace(
1523 				    backing_object,
1524 				    p->pindex,
1525 				    1
1526 				);
1527 			}
1528 
1529 			if (
1530 			    p->pindex < backing_offset_index ||
1531 			    new_pindex >= object->size
1532 			) {
1533 				/*
1534 				 * Page is out of the parent object's range, we
1535 				 * can simply destroy it.
1536 				 */
1537 				vm_page_lock_queues();
1538 				KASSERT(!pmap_page_is_mapped(p),
1539 				    ("freeing mapped page %p", p));
1540 				if (p->wire_count == 0)
1541 					vm_page_free(p);
1542 				else
1543 					vm_page_remove(p);
1544 				vm_page_unlock_queues();
1545 				p = next;
1546 				continue;
1547 			}
1548 
1549 			pp = vm_page_lookup(object, new_pindex);
1550 			if (
1551 			    pp != NULL ||
1552 			    vm_pager_has_page(object, new_pindex, NULL, NULL)
1553 			) {
1554 				/*
1555 				 * page already exists in parent OR swap exists
1556 				 * for this location in the parent.  Destroy
1557 				 * the original page from the backing object.
1558 				 *
1559 				 * Leave the parent's page alone
1560 				 */
1561 				vm_page_lock_queues();
1562 				KASSERT(!pmap_page_is_mapped(p),
1563 				    ("freeing mapped page %p", p));
1564 				if (p->wire_count == 0)
1565 					vm_page_free(p);
1566 				else
1567 					vm_page_remove(p);
1568 				vm_page_unlock_queues();
1569 				p = next;
1570 				continue;
1571 			}
1572 
1573 			/*
1574 			 * Page does not exist in parent, rename the
1575 			 * page from the backing object to the main object.
1576 			 *
1577 			 * If the page was mapped to a process, it can remain
1578 			 * mapped through the rename.
1579 			 */
1580 			vm_page_lock_queues();
1581 			vm_page_rename(p, object, new_pindex);
1582 			vm_page_unlock_queues();
1583 			/* page automatically made dirty by rename */
1584 		}
1585 		p = next;
1586 	}
1587 	return (r);
1588 }
1589 
1590 
1591 /*
1592  * this version of collapse allows the operation to occur earlier and
1593  * when paging_in_progress is true for an object...  This is not a complete
1594  * operation, but should plug 99.9% of the rest of the leaks.
1595  */
1596 static void
1597 vm_object_qcollapse(vm_object_t object)
1598 {
1599 	vm_object_t backing_object = object->backing_object;
1600 
1601 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
1602 	VM_OBJECT_LOCK_ASSERT(backing_object, MA_OWNED);
1603 
1604 	if (backing_object->ref_count != 1)
1605 		return;
1606 
1607 	vm_object_backing_scan(object, OBSC_COLLAPSE_NOWAIT);
1608 }
1609 
1610 /*
1611  *	vm_object_collapse:
1612  *
1613  *	Collapse an object with the object backing it.
1614  *	Pages in the backing object are moved into the
1615  *	parent, and the backing object is deallocated.
1616  */
1617 void
1618 vm_object_collapse(vm_object_t object)
1619 {
1620 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
1621 
1622 	while (TRUE) {
1623 		vm_object_t backing_object;
1624 
1625 		/*
1626 		 * Verify that the conditions are right for collapse:
1627 		 *
1628 		 * The object exists and the backing object exists.
1629 		 */
1630 		if ((backing_object = object->backing_object) == NULL)
1631 			break;
1632 
1633 		/*
1634 		 * we check the backing object first, because it is most likely
1635 		 * not collapsable.
1636 		 */
1637 		VM_OBJECT_LOCK(backing_object);
1638 		if (backing_object->handle != NULL ||
1639 		    (backing_object->type != OBJT_DEFAULT &&
1640 		     backing_object->type != OBJT_SWAP) ||
1641 		    (backing_object->flags & OBJ_DEAD) ||
1642 		    object->handle != NULL ||
1643 		    (object->type != OBJT_DEFAULT &&
1644 		     object->type != OBJT_SWAP) ||
1645 		    (object->flags & OBJ_DEAD)) {
1646 			VM_OBJECT_UNLOCK(backing_object);
1647 			break;
1648 		}
1649 
1650 		if (
1651 		    object->paging_in_progress != 0 ||
1652 		    backing_object->paging_in_progress != 0
1653 		) {
1654 			vm_object_qcollapse(object);
1655 			VM_OBJECT_UNLOCK(backing_object);
1656 			break;
1657 		}
1658 		/*
1659 		 * We know that we can either collapse the backing object (if
1660 		 * the parent is the only reference to it) or (perhaps) have
1661 		 * the parent bypass the object if the parent happens to shadow
1662 		 * all the resident pages in the entire backing object.
1663 		 *
1664 		 * This is ignoring pager-backed pages such as swap pages.
1665 		 * vm_object_backing_scan fails the shadowing test in this
1666 		 * case.
1667 		 */
1668 		if (backing_object->ref_count == 1) {
1669 			/*
1670 			 * If there is exactly one reference to the backing
1671 			 * object, we can collapse it into the parent.
1672 			 */
1673 			vm_object_backing_scan(object, OBSC_COLLAPSE_WAIT);
1674 
1675 			/*
1676 			 * Move the pager from backing_object to object.
1677 			 */
1678 			if (backing_object->type == OBJT_SWAP) {
1679 				/*
1680 				 * swap_pager_copy() can sleep, in which case
1681 				 * the backing_object's and object's locks are
1682 				 * released and reacquired.
1683 				 */
1684 				swap_pager_copy(
1685 				    backing_object,
1686 				    object,
1687 				    OFF_TO_IDX(object->backing_object_offset), TRUE);
1688 			}
1689 			/*
1690 			 * Object now shadows whatever backing_object did.
1691 			 * Note that the reference to
1692 			 * backing_object->backing_object moves from within
1693 			 * backing_object to within object.
1694 			 */
1695 			LIST_REMOVE(object, shadow_list);
1696 			backing_object->shadow_count--;
1697 			backing_object->generation++;
1698 			if (backing_object->backing_object) {
1699 				VM_OBJECT_LOCK(backing_object->backing_object);
1700 				LIST_REMOVE(backing_object, shadow_list);
1701 				LIST_INSERT_HEAD(
1702 				    &backing_object->backing_object->shadow_head,
1703 				    object, shadow_list);
1704 				/*
1705 				 * The shadow_count has not changed.
1706 				 */
1707 				backing_object->backing_object->generation++;
1708 				VM_OBJECT_UNLOCK(backing_object->backing_object);
1709 			}
1710 			object->backing_object = backing_object->backing_object;
1711 			object->backing_object_offset +=
1712 			    backing_object->backing_object_offset;
1713 
1714 			/*
1715 			 * Discard backing_object.
1716 			 *
1717 			 * Since the backing object has no pages, no pager left,
1718 			 * and no object references within it, all that is
1719 			 * necessary is to dispose of it.
1720 			 */
1721 			KASSERT(backing_object->ref_count == 1, ("backing_object %p was somehow re-referenced during collapse!", backing_object));
1722 			VM_OBJECT_UNLOCK(backing_object);
1723 
1724 			mtx_lock(&vm_object_list_mtx);
1725 			TAILQ_REMOVE(
1726 			    &vm_object_list,
1727 			    backing_object,
1728 			    object_list
1729 			);
1730 			mtx_unlock(&vm_object_list_mtx);
1731 
1732 			uma_zfree(obj_zone, backing_object);
1733 
1734 			object_collapses++;
1735 		} else {
1736 			vm_object_t new_backing_object;
1737 
1738 			/*
1739 			 * If we do not entirely shadow the backing object,
1740 			 * there is nothing we can do so we give up.
1741 			 */
1742 			if (object->resident_page_count != object->size &&
1743 			    vm_object_backing_scan(object,
1744 			    OBSC_TEST_ALL_SHADOWED) == 0) {
1745 				VM_OBJECT_UNLOCK(backing_object);
1746 				break;
1747 			}
1748 
1749 			/*
1750 			 * Make the parent shadow the next object in the
1751 			 * chain.  Deallocating backing_object will not remove
1752 			 * it, since its reference count is at least 2.
1753 			 */
1754 			LIST_REMOVE(object, shadow_list);
1755 			backing_object->shadow_count--;
1756 			backing_object->generation++;
1757 
1758 			new_backing_object = backing_object->backing_object;
1759 			if ((object->backing_object = new_backing_object) != NULL) {
1760 				VM_OBJECT_LOCK(new_backing_object);
1761 				LIST_INSERT_HEAD(
1762 				    &new_backing_object->shadow_head,
1763 				    object,
1764 				    shadow_list
1765 				);
1766 				new_backing_object->shadow_count++;
1767 				new_backing_object->generation++;
1768 				vm_object_reference_locked(new_backing_object);
1769 				VM_OBJECT_UNLOCK(new_backing_object);
1770 				object->backing_object_offset +=
1771 					backing_object->backing_object_offset;
1772 			}
1773 
1774 			/*
1775 			 * Drop the reference count on backing_object. Since
1776 			 * its ref_count was at least 2, it will not vanish.
1777 			 */
1778 			backing_object->ref_count--;
1779 			VM_OBJECT_UNLOCK(backing_object);
1780 			object_bypasses++;
1781 		}
1782 
1783 		/*
1784 		 * Try again with this object's new backing object.
1785 		 */
1786 	}
1787 }
1788 
1789 /*
1790  *	vm_object_page_remove:
1791  *
1792  *	Removes all physical pages in the given range from the
1793  *	object's list of pages.  If the range's end is zero, all
1794  *	physical pages from the range's start to the end of the object
1795  *	are deleted.
1796  *
1797  *	The object must be locked.
1798  */
1799 void
1800 vm_object_page_remove(vm_object_t object, vm_pindex_t start, vm_pindex_t end,
1801     boolean_t clean_only)
1802 {
1803 	vm_page_t p, next;
1804 
1805 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
1806 	if (object->resident_page_count == 0)
1807 		return;
1808 
1809 	/*
1810 	 * Since physically-backed objects do not use managed pages, we can't
1811 	 * remove pages from the object (we must instead remove the page
1812 	 * references, and then destroy the object).
1813 	 */
1814 	KASSERT(object->type != OBJT_PHYS || object == kernel_object ||
1815 	    object == kmem_object,
1816 	    ("attempt to remove pages from a physical object"));
1817 
1818 	vm_object_pip_add(object, 1);
1819 again:
1820 	vm_page_lock_queues();
1821 	if ((p = TAILQ_FIRST(&object->memq)) != NULL) {
1822 		if (p->pindex < start) {
1823 			p = vm_page_splay(start, object->root);
1824 			if ((object->root = p)->pindex < start)
1825 				p = TAILQ_NEXT(p, listq);
1826 		}
1827 	}
1828 	/*
1829 	 * Assert: the variable p is either (1) the page with the
1830 	 * least pindex greater than or equal to the parameter pindex
1831 	 * or (2) NULL.
1832 	 */
1833 	for (;
1834 	     p != NULL && (p->pindex < end || end == 0);
1835 	     p = next) {
1836 		next = TAILQ_NEXT(p, listq);
1837 
1838 		if (p->wire_count != 0) {
1839 			pmap_remove_all(p);
1840 			if (!clean_only)
1841 				p->valid = 0;
1842 			continue;
1843 		}
1844 		if (vm_page_sleep_if_busy(p, TRUE, "vmopar"))
1845 			goto again;
1846 		if (clean_only && p->valid) {
1847 			pmap_remove_write(p);
1848 			if (p->valid & p->dirty)
1849 				continue;
1850 		}
1851 		pmap_remove_all(p);
1852 		vm_page_free(p);
1853 	}
1854 	vm_page_unlock_queues();
1855 	vm_object_pip_wakeup(object);
1856 }
1857 
1858 /*
1859  *	Routine:	vm_object_coalesce
1860  *	Function:	Coalesces two objects backing up adjoining
1861  *			regions of memory into a single object.
1862  *
1863  *	returns TRUE if objects were combined.
1864  *
1865  *	NOTE:	Only works at the moment if the second object is NULL -
1866  *		if it's not, which object do we lock first?
1867  *
1868  *	Parameters:
1869  *		prev_object	First object to coalesce
1870  *		prev_offset	Offset into prev_object
1871  *		prev_size	Size of reference to prev_object
1872  *		next_size	Size of reference to the second object
1873  *
1874  *	Conditions:
1875  *	The object must *not* be locked.
1876  */
1877 boolean_t
1878 vm_object_coalesce(vm_object_t prev_object, vm_ooffset_t prev_offset,
1879 	vm_size_t prev_size, vm_size_t next_size)
1880 {
1881 	vm_pindex_t next_pindex;
1882 
1883 	if (prev_object == NULL)
1884 		return (TRUE);
1885 	VM_OBJECT_LOCK(prev_object);
1886 	if (prev_object->type != OBJT_DEFAULT &&
1887 	    prev_object->type != OBJT_SWAP) {
1888 		VM_OBJECT_UNLOCK(prev_object);
1889 		return (FALSE);
1890 	}
1891 
1892 	/*
1893 	 * Try to collapse the object first
1894 	 */
1895 	vm_object_collapse(prev_object);
1896 
1897 	/*
1898 	 * Can't coalesce if: . more than one reference . paged out . shadows
1899 	 * another object . has a copy elsewhere (any of which mean that the
1900 	 * pages not mapped to prev_entry may be in use anyway)
1901 	 */
1902 	if (prev_object->backing_object != NULL) {
1903 		VM_OBJECT_UNLOCK(prev_object);
1904 		return (FALSE);
1905 	}
1906 
1907 	prev_size >>= PAGE_SHIFT;
1908 	next_size >>= PAGE_SHIFT;
1909 	next_pindex = OFF_TO_IDX(prev_offset) + prev_size;
1910 
1911 	if ((prev_object->ref_count > 1) &&
1912 	    (prev_object->size != next_pindex)) {
1913 		VM_OBJECT_UNLOCK(prev_object);
1914 		return (FALSE);
1915 	}
1916 
1917 	/*
1918 	 * Remove any pages that may still be in the object from a previous
1919 	 * deallocation.
1920 	 */
1921 	if (next_pindex < prev_object->size) {
1922 		vm_object_page_remove(prev_object,
1923 				      next_pindex,
1924 				      next_pindex + next_size, FALSE);
1925 		if (prev_object->type == OBJT_SWAP)
1926 			swap_pager_freespace(prev_object,
1927 					     next_pindex, next_size);
1928 	}
1929 
1930 	/*
1931 	 * Extend the object if necessary.
1932 	 */
1933 	if (next_pindex + next_size > prev_object->size)
1934 		prev_object->size = next_pindex + next_size;
1935 
1936 	VM_OBJECT_UNLOCK(prev_object);
1937 	return (TRUE);
1938 }
1939 
1940 void
1941 vm_object_set_writeable_dirty(vm_object_t object)
1942 {
1943 	struct vnode *vp;
1944 
1945 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
1946 	if ((object->flags & OBJ_MIGHTBEDIRTY) != 0)
1947 		return;
1948 	vm_object_set_flag(object, OBJ_MIGHTBEDIRTY);
1949 	if (object->type == OBJT_VNODE &&
1950 	    (vp = (struct vnode *)object->handle) != NULL) {
1951 		VI_LOCK(vp);
1952 		vp->v_iflag |= VI_OBJDIRTY;
1953 		VI_UNLOCK(vp);
1954 	}
1955 }
1956 
1957 #include "opt_ddb.h"
1958 #ifdef DDB
1959 #include <sys/kernel.h>
1960 
1961 #include <sys/cons.h>
1962 
1963 #include <ddb/ddb.h>
1964 
1965 static int
1966 _vm_object_in_map(vm_map_t map, vm_object_t object, vm_map_entry_t entry)
1967 {
1968 	vm_map_t tmpm;
1969 	vm_map_entry_t tmpe;
1970 	vm_object_t obj;
1971 	int entcount;
1972 
1973 	if (map == 0)
1974 		return 0;
1975 
1976 	if (entry == 0) {
1977 		tmpe = map->header.next;
1978 		entcount = map->nentries;
1979 		while (entcount-- && (tmpe != &map->header)) {
1980 			if (_vm_object_in_map(map, object, tmpe)) {
1981 				return 1;
1982 			}
1983 			tmpe = tmpe->next;
1984 		}
1985 	} else if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
1986 		tmpm = entry->object.sub_map;
1987 		tmpe = tmpm->header.next;
1988 		entcount = tmpm->nentries;
1989 		while (entcount-- && tmpe != &tmpm->header) {
1990 			if (_vm_object_in_map(tmpm, object, tmpe)) {
1991 				return 1;
1992 			}
1993 			tmpe = tmpe->next;
1994 		}
1995 	} else if ((obj = entry->object.vm_object) != NULL) {
1996 		for (; obj; obj = obj->backing_object)
1997 			if (obj == object) {
1998 				return 1;
1999 			}
2000 	}
2001 	return 0;
2002 }
2003 
2004 static int
2005 vm_object_in_map(vm_object_t object)
2006 {
2007 	struct proc *p;
2008 
2009 	/* sx_slock(&allproc_lock); */
2010 	FOREACH_PROC_IN_SYSTEM(p) {
2011 		if (!p->p_vmspace /* || (p->p_flag & (P_SYSTEM|P_WEXIT)) */)
2012 			continue;
2013 		if (_vm_object_in_map(&p->p_vmspace->vm_map, object, 0)) {
2014 			/* sx_sunlock(&allproc_lock); */
2015 			return 1;
2016 		}
2017 	}
2018 	/* sx_sunlock(&allproc_lock); */
2019 	if (_vm_object_in_map(kernel_map, object, 0))
2020 		return 1;
2021 	if (_vm_object_in_map(kmem_map, object, 0))
2022 		return 1;
2023 	if (_vm_object_in_map(pager_map, object, 0))
2024 		return 1;
2025 	if (_vm_object_in_map(buffer_map, object, 0))
2026 		return 1;
2027 	return 0;
2028 }
2029 
2030 DB_SHOW_COMMAND(vmochk, vm_object_check)
2031 {
2032 	vm_object_t object;
2033 
2034 	/*
2035 	 * make sure that internal objs are in a map somewhere
2036 	 * and none have zero ref counts.
2037 	 */
2038 	TAILQ_FOREACH(object, &vm_object_list, object_list) {
2039 		if (object->handle == NULL &&
2040 		    (object->type == OBJT_DEFAULT || object->type == OBJT_SWAP)) {
2041 			if (object->ref_count == 0) {
2042 				db_printf("vmochk: internal obj has zero ref count: %ld\n",
2043 					(long)object->size);
2044 			}
2045 			if (!vm_object_in_map(object)) {
2046 				db_printf(
2047 			"vmochk: internal obj is not in a map: "
2048 			"ref: %d, size: %lu: 0x%lx, backing_object: %p\n",
2049 				    object->ref_count, (u_long)object->size,
2050 				    (u_long)object->size,
2051 				    (void *)object->backing_object);
2052 			}
2053 		}
2054 	}
2055 }
2056 
2057 /*
2058  *	vm_object_print:	[ debug ]
2059  */
2060 DB_SHOW_COMMAND(object, vm_object_print_static)
2061 {
2062 	/* XXX convert args. */
2063 	vm_object_t object = (vm_object_t)addr;
2064 	boolean_t full = have_addr;
2065 
2066 	vm_page_t p;
2067 
2068 	/* XXX count is an (unused) arg.  Avoid shadowing it. */
2069 #define	count	was_count
2070 
2071 	int count;
2072 
2073 	if (object == NULL)
2074 		return;
2075 
2076 	db_iprintf(
2077 	    "Object %p: type=%d, size=0x%jx, res=%d, ref=%d, flags=0x%x\n",
2078 	    object, (int)object->type, (uintmax_t)object->size,
2079 	    object->resident_page_count, object->ref_count, object->flags);
2080 	db_iprintf(" sref=%d, backing_object(%d)=(%p)+0x%jx\n",
2081 	    object->shadow_count,
2082 	    object->backing_object ? object->backing_object->ref_count : 0,
2083 	    object->backing_object, (uintmax_t)object->backing_object_offset);
2084 
2085 	if (!full)
2086 		return;
2087 
2088 	db_indent += 2;
2089 	count = 0;
2090 	TAILQ_FOREACH(p, &object->memq, listq) {
2091 		if (count == 0)
2092 			db_iprintf("memory:=");
2093 		else if (count == 6) {
2094 			db_printf("\n");
2095 			db_iprintf(" ...");
2096 			count = 0;
2097 		} else
2098 			db_printf(",");
2099 		count++;
2100 
2101 		db_printf("(off=0x%jx,page=0x%jx)",
2102 		    (uintmax_t)p->pindex, (uintmax_t)VM_PAGE_TO_PHYS(p));
2103 	}
2104 	if (count != 0)
2105 		db_printf("\n");
2106 	db_indent -= 2;
2107 }
2108 
2109 /* XXX. */
2110 #undef count
2111 
2112 /* XXX need this non-static entry for calling from vm_map_print. */
2113 void
2114 vm_object_print(
2115         /* db_expr_t */ long addr,
2116 	boolean_t have_addr,
2117 	/* db_expr_t */ long count,
2118 	char *modif)
2119 {
2120 	vm_object_print_static(addr, have_addr, count, modif);
2121 }
2122 
2123 DB_SHOW_COMMAND(vmopag, vm_object_print_pages)
2124 {
2125 	vm_object_t object;
2126 	int nl = 0;
2127 	int c;
2128 
2129 	TAILQ_FOREACH(object, &vm_object_list, object_list) {
2130 		vm_pindex_t idx, fidx;
2131 		vm_pindex_t osize;
2132 		vm_paddr_t pa = -1, padiff;
2133 		int rcount;
2134 		vm_page_t m;
2135 
2136 		db_printf("new object: %p\n", (void *)object);
2137 		if (nl > 18) {
2138 			c = cngetc();
2139 			if (c != ' ')
2140 				return;
2141 			nl = 0;
2142 		}
2143 		nl++;
2144 		rcount = 0;
2145 		fidx = 0;
2146 		osize = object->size;
2147 		if (osize > 128)
2148 			osize = 128;
2149 		for (idx = 0; idx < osize; idx++) {
2150 			m = vm_page_lookup(object, idx);
2151 			if (m == NULL) {
2152 				if (rcount) {
2153 					db_printf(" index(%ld)run(%d)pa(0x%lx)\n",
2154 						(long)fidx, rcount, (long)pa);
2155 					if (nl > 18) {
2156 						c = cngetc();
2157 						if (c != ' ')
2158 							return;
2159 						nl = 0;
2160 					}
2161 					nl++;
2162 					rcount = 0;
2163 				}
2164 				continue;
2165 			}
2166 
2167 
2168 			if (rcount &&
2169 				(VM_PAGE_TO_PHYS(m) == pa + rcount * PAGE_SIZE)) {
2170 				++rcount;
2171 				continue;
2172 			}
2173 			if (rcount) {
2174 				padiff = pa + rcount * PAGE_SIZE - VM_PAGE_TO_PHYS(m);
2175 				padiff >>= PAGE_SHIFT;
2176 				padiff &= PQ_COLORMASK;
2177 				if (padiff == 0) {
2178 					pa = VM_PAGE_TO_PHYS(m) - rcount * PAGE_SIZE;
2179 					++rcount;
2180 					continue;
2181 				}
2182 				db_printf(" index(%ld)run(%d)pa(0x%lx)",
2183 					(long)fidx, rcount, (long)pa);
2184 				db_printf("pd(%ld)\n", (long)padiff);
2185 				if (nl > 18) {
2186 					c = cngetc();
2187 					if (c != ' ')
2188 						return;
2189 					nl = 0;
2190 				}
2191 				nl++;
2192 			}
2193 			fidx = idx;
2194 			pa = VM_PAGE_TO_PHYS(m);
2195 			rcount = 1;
2196 		}
2197 		if (rcount) {
2198 			db_printf(" index(%ld)run(%d)pa(0x%lx)\n",
2199 				(long)fidx, rcount, (long)pa);
2200 			if (nl > 18) {
2201 				c = cngetc();
2202 				if (c != ' ')
2203 					return;
2204 				nl = 0;
2205 			}
2206 			nl++;
2207 		}
2208 	}
2209 }
2210 #endif /* DDB */
2211