xref: /freebsd/sys/vm/vm_object.c (revision 0e1497aefd602cea581d2380d22e67dfdcac6b4e)
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 "opt_vm.h"
69 
70 #include <sys/param.h>
71 #include <sys/systm.h>
72 #include <sys/lock.h>
73 #include <sys/mman.h>
74 #include <sys/mount.h>
75 #include <sys/kernel.h>
76 #include <sys/sysctl.h>
77 #include <sys/mutex.h>
78 #include <sys/proc.h>		/* for curproc, pageproc */
79 #include <sys/socket.h>
80 #include <sys/resourcevar.h>
81 #include <sys/vnode.h>
82 #include <sys/vmmeter.h>
83 #include <sys/sx.h>
84 
85 #include <vm/vm.h>
86 #include <vm/vm_param.h>
87 #include <vm/pmap.h>
88 #include <vm/vm_map.h>
89 #include <vm/vm_object.h>
90 #include <vm/vm_page.h>
91 #include <vm/vm_pageout.h>
92 #include <vm/vm_pager.h>
93 #include <vm/swap_pager.h>
94 #include <vm/vm_kern.h>
95 #include <vm/vm_extern.h>
96 #include <vm/vm_reserv.h>
97 #include <vm/uma.h>
98 
99 static int old_msync;
100 SYSCTL_INT(_vm, OID_AUTO, old_msync, CTLFLAG_RW, &old_msync, 0,
101     "Use old (insecure) msync behavior");
102 
103 static int	vm_object_page_collect_flush(vm_object_t object, vm_page_t p,
104 		    int pagerflags);
105 static void	vm_object_qcollapse(vm_object_t object);
106 static void	vm_object_vndeallocate(vm_object_t object);
107 
108 /*
109  *	Virtual memory objects maintain the actual data
110  *	associated with allocated virtual memory.  A given
111  *	page of memory exists within exactly one object.
112  *
113  *	An object is only deallocated when all "references"
114  *	are given up.  Only one "reference" to a given
115  *	region of an object should be writeable.
116  *
117  *	Associated with each object is a list of all resident
118  *	memory pages belonging to that object; this list is
119  *	maintained by the "vm_page" module, and locked by the object's
120  *	lock.
121  *
122  *	Each object also records a "pager" routine which is
123  *	used to retrieve (and store) pages to the proper backing
124  *	storage.  In addition, objects may be backed by other
125  *	objects from which they were virtual-copied.
126  *
127  *	The only items within the object structure which are
128  *	modified after time of creation are:
129  *		reference count		locked by object's lock
130  *		pager routine		locked by object's lock
131  *
132  */
133 
134 struct object_q vm_object_list;
135 struct mtx vm_object_list_mtx;	/* lock for object list and count */
136 
137 struct vm_object kernel_object_store;
138 struct vm_object kmem_object_store;
139 
140 SYSCTL_NODE(_vm_stats, OID_AUTO, object, CTLFLAG_RD, 0, "VM object stats");
141 
142 static long object_collapses;
143 SYSCTL_LONG(_vm_stats_object, OID_AUTO, collapses, CTLFLAG_RD,
144     &object_collapses, 0, "VM object collapses");
145 
146 static long object_bypasses;
147 SYSCTL_LONG(_vm_stats_object, OID_AUTO, bypasses, CTLFLAG_RD,
148     &object_bypasses, 0, "VM object bypasses");
149 
150 static uma_zone_t obj_zone;
151 
152 static int vm_object_zinit(void *mem, int size, int flags);
153 
154 #ifdef INVARIANTS
155 static void vm_object_zdtor(void *mem, int size, void *arg);
156 
157 static void
158 vm_object_zdtor(void *mem, int size, void *arg)
159 {
160 	vm_object_t object;
161 
162 	object = (vm_object_t)mem;
163 	KASSERT(TAILQ_EMPTY(&object->memq),
164 	    ("object %p has resident pages",
165 	    object));
166 #if VM_NRESERVLEVEL > 0
167 	KASSERT(LIST_EMPTY(&object->rvq),
168 	    ("object %p has reservations",
169 	    object));
170 #endif
171 	KASSERT(object->cache == NULL,
172 	    ("object %p has cached pages",
173 	    object));
174 	KASSERT(object->paging_in_progress == 0,
175 	    ("object %p paging_in_progress = %d",
176 	    object, object->paging_in_progress));
177 	KASSERT(object->resident_page_count == 0,
178 	    ("object %p resident_page_count = %d",
179 	    object, object->resident_page_count));
180 	KASSERT(object->shadow_count == 0,
181 	    ("object %p shadow_count = %d",
182 	    object, object->shadow_count));
183 }
184 #endif
185 
186 static int
187 vm_object_zinit(void *mem, int size, int flags)
188 {
189 	vm_object_t object;
190 
191 	object = (vm_object_t)mem;
192 	bzero(&object->mtx, sizeof(object->mtx));
193 	VM_OBJECT_LOCK_INIT(object, "standard object");
194 
195 	/* These are true for any object that has been freed */
196 	object->paging_in_progress = 0;
197 	object->resident_page_count = 0;
198 	object->shadow_count = 0;
199 	return (0);
200 }
201 
202 void
203 _vm_object_allocate(objtype_t type, vm_pindex_t size, vm_object_t object)
204 {
205 
206 	TAILQ_INIT(&object->memq);
207 	LIST_INIT(&object->shadow_head);
208 
209 	object->root = NULL;
210 	object->type = type;
211 	object->size = size;
212 	object->generation = 1;
213 	object->ref_count = 1;
214 	object->memattr = VM_MEMATTR_DEFAULT;
215 	object->flags = 0;
216 	object->cred = NULL;
217 	object->charge = 0;
218 	if ((object->type == OBJT_DEFAULT) || (object->type == OBJT_SWAP))
219 		object->flags = OBJ_ONEMAPPING;
220 	object->pg_color = 0;
221 	object->handle = NULL;
222 	object->backing_object = NULL;
223 	object->backing_object_offset = (vm_ooffset_t) 0;
224 #if VM_NRESERVLEVEL > 0
225 	LIST_INIT(&object->rvq);
226 #endif
227 	object->cache = NULL;
228 
229 	mtx_lock(&vm_object_list_mtx);
230 	TAILQ_INSERT_TAIL(&vm_object_list, object, object_list);
231 	mtx_unlock(&vm_object_list_mtx);
232 }
233 
234 /*
235  *	vm_object_init:
236  *
237  *	Initialize the VM objects module.
238  */
239 void
240 vm_object_init(void)
241 {
242 	TAILQ_INIT(&vm_object_list);
243 	mtx_init(&vm_object_list_mtx, "vm object_list", NULL, MTX_DEF);
244 
245 	VM_OBJECT_LOCK_INIT(&kernel_object_store, "kernel object");
246 	_vm_object_allocate(OBJT_PHYS, OFF_TO_IDX(VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS),
247 	    kernel_object);
248 #if VM_NRESERVLEVEL > 0
249 	kernel_object->flags |= OBJ_COLORED;
250 	kernel_object->pg_color = (u_short)atop(VM_MIN_KERNEL_ADDRESS);
251 #endif
252 
253 	VM_OBJECT_LOCK_INIT(&kmem_object_store, "kmem object");
254 	_vm_object_allocate(OBJT_PHYS, OFF_TO_IDX(VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS),
255 	    kmem_object);
256 #if VM_NRESERVLEVEL > 0
257 	kmem_object->flags |= OBJ_COLORED;
258 	kmem_object->pg_color = (u_short)atop(VM_MIN_KERNEL_ADDRESS);
259 #endif
260 
261 	/*
262 	 * The lock portion of struct vm_object must be type stable due
263 	 * to vm_pageout_fallback_object_lock locking a vm object
264 	 * without holding any references to it.
265 	 */
266 	obj_zone = uma_zcreate("VM OBJECT", sizeof (struct vm_object), NULL,
267 #ifdef INVARIANTS
268 	    vm_object_zdtor,
269 #else
270 	    NULL,
271 #endif
272 	    vm_object_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_VM|UMA_ZONE_NOFREE);
273 }
274 
275 void
276 vm_object_clear_flag(vm_object_t object, u_short bits)
277 {
278 
279 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
280 	object->flags &= ~bits;
281 }
282 
283 /*
284  *	Sets the default memory attribute for the specified object.  Pages
285  *	that are allocated to this object are by default assigned this memory
286  *	attribute.
287  *
288  *	Presently, this function must be called before any pages are allocated
289  *	to the object.  In the future, this requirement may be relaxed for
290  *	"default" and "swap" objects.
291  */
292 int
293 vm_object_set_memattr(vm_object_t object, vm_memattr_t memattr)
294 {
295 
296 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
297 	switch (object->type) {
298 	case OBJT_DEFAULT:
299 	case OBJT_DEVICE:
300 	case OBJT_PHYS:
301 	case OBJT_SG:
302 	case OBJT_SWAP:
303 	case OBJT_VNODE:
304 		if (!TAILQ_EMPTY(&object->memq))
305 			return (KERN_FAILURE);
306 		break;
307 	case OBJT_DEAD:
308 		return (KERN_INVALID_ARGUMENT);
309 	}
310 	object->memattr = memattr;
311 	return (KERN_SUCCESS);
312 }
313 
314 void
315 vm_object_pip_add(vm_object_t object, short i)
316 {
317 
318 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
319 	object->paging_in_progress += i;
320 }
321 
322 void
323 vm_object_pip_subtract(vm_object_t object, short i)
324 {
325 
326 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
327 	object->paging_in_progress -= i;
328 }
329 
330 void
331 vm_object_pip_wakeup(vm_object_t object)
332 {
333 
334 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
335 	object->paging_in_progress--;
336 	if ((object->flags & OBJ_PIPWNT) && object->paging_in_progress == 0) {
337 		vm_object_clear_flag(object, OBJ_PIPWNT);
338 		wakeup(object);
339 	}
340 }
341 
342 void
343 vm_object_pip_wakeupn(vm_object_t object, short i)
344 {
345 
346 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
347 	if (i)
348 		object->paging_in_progress -= i;
349 	if ((object->flags & OBJ_PIPWNT) && object->paging_in_progress == 0) {
350 		vm_object_clear_flag(object, OBJ_PIPWNT);
351 		wakeup(object);
352 	}
353 }
354 
355 void
356 vm_object_pip_wait(vm_object_t object, char *waitid)
357 {
358 
359 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
360 	while (object->paging_in_progress) {
361 		object->flags |= OBJ_PIPWNT;
362 		msleep(object, VM_OBJECT_MTX(object), PVM, waitid, 0);
363 	}
364 }
365 
366 /*
367  *	vm_object_allocate:
368  *
369  *	Returns a new object with the given size.
370  */
371 vm_object_t
372 vm_object_allocate(objtype_t type, vm_pindex_t size)
373 {
374 	vm_object_t object;
375 
376 	object = (vm_object_t)uma_zalloc(obj_zone, M_WAITOK);
377 	_vm_object_allocate(type, size, object);
378 	return (object);
379 }
380 
381 
382 /*
383  *	vm_object_reference:
384  *
385  *	Gets another reference to the given object.  Note: OBJ_DEAD
386  *	objects can be referenced during final cleaning.
387  */
388 void
389 vm_object_reference(vm_object_t object)
390 {
391 	if (object == NULL)
392 		return;
393 	VM_OBJECT_LOCK(object);
394 	vm_object_reference_locked(object);
395 	VM_OBJECT_UNLOCK(object);
396 }
397 
398 /*
399  *	vm_object_reference_locked:
400  *
401  *	Gets another reference to the given object.
402  *
403  *	The object must be locked.
404  */
405 void
406 vm_object_reference_locked(vm_object_t object)
407 {
408 	struct vnode *vp;
409 
410 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
411 	object->ref_count++;
412 	if (object->type == OBJT_VNODE) {
413 		vp = object->handle;
414 		vref(vp);
415 	}
416 }
417 
418 /*
419  * Handle deallocating an object of type OBJT_VNODE.
420  */
421 static void
422 vm_object_vndeallocate(vm_object_t object)
423 {
424 	struct vnode *vp = (struct vnode *) object->handle;
425 
426 	VFS_ASSERT_GIANT(vp->v_mount);
427 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
428 	KASSERT(object->type == OBJT_VNODE,
429 	    ("vm_object_vndeallocate: not a vnode object"));
430 	KASSERT(vp != NULL, ("vm_object_vndeallocate: missing vp"));
431 #ifdef INVARIANTS
432 	if (object->ref_count == 0) {
433 		vprint("vm_object_vndeallocate", vp);
434 		panic("vm_object_vndeallocate: bad object reference count");
435 	}
436 #endif
437 
438 	object->ref_count--;
439 	if (object->ref_count == 0) {
440 		mp_fixme("Unlocked vflag access.");
441 		vp->v_vflag &= ~VV_TEXT;
442 	}
443 	VM_OBJECT_UNLOCK(object);
444 	/*
445 	 * vrele may need a vop lock
446 	 */
447 	vrele(vp);
448 }
449 
450 /*
451  *	vm_object_deallocate:
452  *
453  *	Release a reference to the specified object,
454  *	gained either through a vm_object_allocate
455  *	or a vm_object_reference call.  When all references
456  *	are gone, storage associated with this object
457  *	may be relinquished.
458  *
459  *	No object may be locked.
460  */
461 void
462 vm_object_deallocate(vm_object_t object)
463 {
464 	vm_object_t temp;
465 
466 	while (object != NULL) {
467 		int vfslocked;
468 
469 		vfslocked = 0;
470 	restart:
471 		VM_OBJECT_LOCK(object);
472 		if (object->type == OBJT_VNODE) {
473 			struct vnode *vp = (struct vnode *) object->handle;
474 
475 			/*
476 			 * Conditionally acquire Giant for a vnode-backed
477 			 * object.  We have to be careful since the type of
478 			 * a vnode object can change while the object is
479 			 * unlocked.
480 			 */
481 			if (VFS_NEEDSGIANT(vp->v_mount) && !vfslocked) {
482 				vfslocked = 1;
483 				if (!mtx_trylock(&Giant)) {
484 					VM_OBJECT_UNLOCK(object);
485 					mtx_lock(&Giant);
486 					goto restart;
487 				}
488 			}
489 			vm_object_vndeallocate(object);
490 			VFS_UNLOCK_GIANT(vfslocked);
491 			return;
492 		} else
493 			/*
494 			 * This is to handle the case that the object
495 			 * changed type while we dropped its lock to
496 			 * obtain Giant.
497 			 */
498 			VFS_UNLOCK_GIANT(vfslocked);
499 
500 		KASSERT(object->ref_count != 0,
501 			("vm_object_deallocate: object deallocated too many times: %d", object->type));
502 
503 		/*
504 		 * If the reference count goes to 0 we start calling
505 		 * vm_object_terminate() on the object chain.
506 		 * A ref count of 1 may be a special case depending on the
507 		 * shadow count being 0 or 1.
508 		 */
509 		object->ref_count--;
510 		if (object->ref_count > 1) {
511 			VM_OBJECT_UNLOCK(object);
512 			return;
513 		} else if (object->ref_count == 1) {
514 			if (object->shadow_count == 0 &&
515 			    object->handle == NULL &&
516 			    (object->type == OBJT_DEFAULT ||
517 			     object->type == OBJT_SWAP)) {
518 				vm_object_set_flag(object, OBJ_ONEMAPPING);
519 			} else if ((object->shadow_count == 1) &&
520 			    (object->handle == NULL) &&
521 			    (object->type == OBJT_DEFAULT ||
522 			     object->type == OBJT_SWAP)) {
523 				vm_object_t robject;
524 
525 				robject = LIST_FIRST(&object->shadow_head);
526 				KASSERT(robject != NULL,
527 				    ("vm_object_deallocate: ref_count: %d, shadow_count: %d",
528 					 object->ref_count,
529 					 object->shadow_count));
530 				if (!VM_OBJECT_TRYLOCK(robject)) {
531 					/*
532 					 * Avoid a potential deadlock.
533 					 */
534 					object->ref_count++;
535 					VM_OBJECT_UNLOCK(object);
536 					/*
537 					 * More likely than not the thread
538 					 * holding robject's lock has lower
539 					 * priority than the current thread.
540 					 * Let the lower priority thread run.
541 					 */
542 					pause("vmo_de", 1);
543 					continue;
544 				}
545 				/*
546 				 * Collapse object into its shadow unless its
547 				 * shadow is dead.  In that case, object will
548 				 * be deallocated by the thread that is
549 				 * deallocating its shadow.
550 				 */
551 				if ((robject->flags & OBJ_DEAD) == 0 &&
552 				    (robject->handle == NULL) &&
553 				    (robject->type == OBJT_DEFAULT ||
554 				     robject->type == OBJT_SWAP)) {
555 
556 					robject->ref_count++;
557 retry:
558 					if (robject->paging_in_progress) {
559 						VM_OBJECT_UNLOCK(object);
560 						vm_object_pip_wait(robject,
561 						    "objde1");
562 						temp = robject->backing_object;
563 						if (object == temp) {
564 							VM_OBJECT_LOCK(object);
565 							goto retry;
566 						}
567 					} else if (object->paging_in_progress) {
568 						VM_OBJECT_UNLOCK(robject);
569 						object->flags |= OBJ_PIPWNT;
570 						msleep(object,
571 						    VM_OBJECT_MTX(object),
572 						    PDROP | PVM, "objde2", 0);
573 						VM_OBJECT_LOCK(robject);
574 						temp = robject->backing_object;
575 						if (object == temp) {
576 							VM_OBJECT_LOCK(object);
577 							goto retry;
578 						}
579 					} else
580 						VM_OBJECT_UNLOCK(object);
581 
582 					if (robject->ref_count == 1) {
583 						robject->ref_count--;
584 						object = robject;
585 						goto doterm;
586 					}
587 					object = robject;
588 					vm_object_collapse(object);
589 					VM_OBJECT_UNLOCK(object);
590 					continue;
591 				}
592 				VM_OBJECT_UNLOCK(robject);
593 			}
594 			VM_OBJECT_UNLOCK(object);
595 			return;
596 		}
597 doterm:
598 		temp = object->backing_object;
599 		if (temp != NULL) {
600 			VM_OBJECT_LOCK(temp);
601 			LIST_REMOVE(object, shadow_list);
602 			temp->shadow_count--;
603 			VM_OBJECT_UNLOCK(temp);
604 			object->backing_object = NULL;
605 		}
606 		/*
607 		 * Don't double-terminate, we could be in a termination
608 		 * recursion due to the terminate having to sync data
609 		 * to disk.
610 		 */
611 		if ((object->flags & OBJ_DEAD) == 0)
612 			vm_object_terminate(object);
613 		else
614 			VM_OBJECT_UNLOCK(object);
615 		object = temp;
616 	}
617 }
618 
619 /*
620  *	vm_object_destroy removes the object from the global object list
621  *      and frees the space for the object.
622  */
623 void
624 vm_object_destroy(vm_object_t object)
625 {
626 
627 	/*
628 	 * Remove the object from the global object list.
629 	 */
630 	mtx_lock(&vm_object_list_mtx);
631 	TAILQ_REMOVE(&vm_object_list, object, object_list);
632 	mtx_unlock(&vm_object_list_mtx);
633 
634 	/*
635 	 * Release the allocation charge.
636 	 */
637 	if (object->cred != NULL) {
638 		KASSERT(object->type == OBJT_DEFAULT ||
639 		    object->type == OBJT_SWAP,
640 		    ("vm_object_terminate: non-swap obj %p has cred",
641 		     object));
642 		swap_release_by_cred(object->charge, object->cred);
643 		object->charge = 0;
644 		crfree(object->cred);
645 		object->cred = NULL;
646 	}
647 
648 	/*
649 	 * Free the space for the object.
650 	 */
651 	uma_zfree(obj_zone, object);
652 }
653 
654 /*
655  *	vm_object_terminate actually destroys the specified object, freeing
656  *	up all previously used resources.
657  *
658  *	The object must be locked.
659  *	This routine may block.
660  */
661 void
662 vm_object_terminate(vm_object_t object)
663 {
664 	vm_page_t p, p_next;
665 
666 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
667 
668 	/*
669 	 * Make sure no one uses us.
670 	 */
671 	vm_object_set_flag(object, OBJ_DEAD);
672 
673 	/*
674 	 * wait for the pageout daemon to be done with the object
675 	 */
676 	vm_object_pip_wait(object, "objtrm");
677 
678 	KASSERT(!object->paging_in_progress,
679 		("vm_object_terminate: pageout in progress"));
680 
681 	/*
682 	 * Clean and free the pages, as appropriate. All references to the
683 	 * object are gone, so we don't need to lock it.
684 	 */
685 	if (object->type == OBJT_VNODE) {
686 		struct vnode *vp = (struct vnode *)object->handle;
687 
688 		/*
689 		 * Clean pages and flush buffers.
690 		 */
691 		vm_object_page_clean(object, 0, 0, OBJPC_SYNC);
692 		VM_OBJECT_UNLOCK(object);
693 
694 		vinvalbuf(vp, V_SAVE, 0, 0);
695 
696 		VM_OBJECT_LOCK(object);
697 	}
698 
699 	KASSERT(object->ref_count == 0,
700 		("vm_object_terminate: object with references, ref_count=%d",
701 		object->ref_count));
702 
703 	/*
704 	 * Free any remaining pageable pages.  This also removes them from the
705 	 * paging queues.  However, don't free wired pages, just remove them
706 	 * from the object.  Rather than incrementally removing each page from
707 	 * the object, the page and object are reset to any empty state.
708 	 */
709 	TAILQ_FOREACH_SAFE(p, &object->memq, listq, p_next) {
710 		KASSERT(!p->busy && (p->oflags & VPO_BUSY) == 0,
711 		    ("vm_object_terminate: freeing busy page %p", p));
712 		vm_page_lock(p);
713 		/*
714 		 * Optimize the page's removal from the object by resetting
715 		 * its "object" field.  Specifically, if the page is not
716 		 * wired, then the effect of this assignment is that
717 		 * vm_page_free()'s call to vm_page_remove() will return
718 		 * immediately without modifying the page or the object.
719 		 */
720 		p->object = NULL;
721 		if (p->wire_count == 0) {
722 			vm_page_free(p);
723 			PCPU_INC(cnt.v_pfree);
724 		}
725 		vm_page_unlock(p);
726 	}
727 	/*
728 	 * If the object contained any pages, then reset it to an empty state.
729 	 * None of the object's fields, including "resident_page_count", were
730 	 * modified by the preceding loop.
731 	 */
732 	if (object->resident_page_count != 0) {
733 		object->root = NULL;
734 		TAILQ_INIT(&object->memq);
735 		object->resident_page_count = 0;
736 		if (object->type == OBJT_VNODE)
737 			vdrop(object->handle);
738 	}
739 
740 #if VM_NRESERVLEVEL > 0
741 	if (__predict_false(!LIST_EMPTY(&object->rvq)))
742 		vm_reserv_break_all(object);
743 #endif
744 	if (__predict_false(object->cache != NULL))
745 		vm_page_cache_free(object, 0, 0);
746 
747 	/*
748 	 * Let the pager know object is dead.
749 	 */
750 	vm_pager_deallocate(object);
751 	VM_OBJECT_UNLOCK(object);
752 
753 	vm_object_destroy(object);
754 }
755 
756 /*
757  *	vm_object_page_clean
758  *
759  *	Clean all dirty pages in the specified range of object.  Leaves page
760  * 	on whatever queue it is currently on.   If NOSYNC is set then do not
761  *	write out pages with VPO_NOSYNC set (originally comes from MAP_NOSYNC),
762  *	leaving the object dirty.
763  *
764  *	When stuffing pages asynchronously, allow clustering.  XXX we need a
765  *	synchronous clustering mode implementation.
766  *
767  *	Odd semantics: if start == end, we clean everything.
768  *
769  *	The object must be locked.
770  */
771 void
772 vm_object_page_clean(vm_object_t object, vm_pindex_t start, vm_pindex_t end,
773     int flags)
774 {
775 	vm_page_t np, p;
776 	vm_pindex_t pi, tend;
777 	int clearobjflags, curgeneration, n, pagerflags;
778 
779 	mtx_assert(&vm_page_queue_mtx, MA_NOTOWNED);
780 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
781 	KASSERT(object->type == OBJT_VNODE, ("Not a vnode object"));
782 	if ((object->flags & OBJ_MIGHTBEDIRTY) == 0 ||
783 	    object->resident_page_count == 0)
784 		return;
785 
786 	pagerflags = (flags & (OBJPC_SYNC | OBJPC_INVAL)) != 0 ?
787 	    VM_PAGER_PUT_SYNC : VM_PAGER_CLUSTER_OK;
788 	pagerflags |= (flags & OBJPC_INVAL) != 0 ? VM_PAGER_PUT_INVAL : 0;
789 
790 	tend = (end == 0) ? object->size : end;
791 
792 	vm_object_set_flag(object, OBJ_CLEANING);
793 
794 	/*
795 	 * Make the page read-only so we can then clear the object flags.
796 	 *
797 	 * However, if this is a nosync mmap then the object is likely to
798 	 * stay dirty so do not mess with the page and do not clear the
799 	 * object flags.
800 	 */
801 	clearobjflags = 1;
802 	for (p = vm_page_find_least(object, start);
803 	    p != NULL && p->pindex < tend; p = TAILQ_NEXT(p, listq)) {
804 		if ((flags & OBJPC_NOSYNC) != 0 &&
805 		    (p->oflags & VPO_NOSYNC) != 0)
806 			clearobjflags = 0;
807 		else
808 			pmap_remove_write(p);
809 	}
810 
811 	if (clearobjflags && (start == 0) && (tend == object->size))
812 		vm_object_clear_flag(object, OBJ_MIGHTBEDIRTY);
813 
814 rescan:
815 	curgeneration = object->generation;
816 
817 	for (p = vm_page_find_least(object, start); p != NULL; p = np) {
818 		pi = p->pindex;
819 		if (pi >= tend)
820 			break;
821 		np = TAILQ_NEXT(p, listq);
822 		if (p->valid == 0)
823 			continue;
824 		if (vm_page_sleep_if_busy(p, TRUE, "vpcwai")) {
825 			if (object->generation != curgeneration)
826 				goto rescan;
827 			np = vm_page_find_least(object, pi);
828 			continue;
829 		}
830 		vm_page_test_dirty(p);
831 		if (p->dirty == 0)
832 			continue;
833 
834 		/*
835 		 * If we have been asked to skip nosync pages and this is a
836 		 * nosync page, skip it.  Note that the object flags were
837 		 * not cleared in this case so we do not have to set them.
838 		 */
839 		if ((flags & OBJPC_NOSYNC) != 0 &&
840 		    (p->oflags & VPO_NOSYNC) != 0)
841 			continue;
842 
843 		n = vm_object_page_collect_flush(object, p, pagerflags);
844 		if (object->generation != curgeneration)
845 			goto rescan;
846 		np = vm_page_find_least(object, pi + n);
847 	}
848 #if 0
849 	VOP_FSYNC(vp, (pagerflags & VM_PAGER_PUT_SYNC) ? MNT_WAIT : 0);
850 #endif
851 
852 	vm_object_clear_flag(object, OBJ_CLEANING);
853 }
854 
855 static int
856 vm_object_page_collect_flush(vm_object_t object, vm_page_t p, int pagerflags)
857 {
858 	vm_page_t ma[vm_pageout_page_count], p_first, tp;
859 	int count, i, mreq, runlen;
860 
861 	mtx_assert(&vm_page_queue_mtx, MA_NOTOWNED);
862 	vm_page_lock_assert(p, MA_NOTOWNED);
863 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
864 
865 	count = 1;
866 	mreq = 0;
867 
868 	for (tp = p; count < vm_pageout_page_count; count++) {
869 		tp = vm_page_next(tp);
870 		if (tp == NULL || tp->busy != 0 || (tp->oflags & VPO_BUSY) != 0)
871 			break;
872 		vm_page_test_dirty(tp);
873 		if (tp->dirty == 0)
874 			break;
875 	}
876 
877 	for (p_first = p; count < vm_pageout_page_count; count++) {
878 		tp = vm_page_prev(p_first);
879 		if (tp == NULL || tp->busy != 0 || (tp->oflags & VPO_BUSY) != 0)
880 			break;
881 		vm_page_test_dirty(tp);
882 		if (tp->dirty == 0)
883 			break;
884 		p_first = tp;
885 		mreq++;
886 	}
887 
888 	for (tp = p_first, i = 0; i < count; tp = TAILQ_NEXT(tp, listq), i++)
889 		ma[i] = tp;
890 
891 	vm_pageout_flush(ma, count, pagerflags, mreq, &runlen);
892 	return (runlen);
893 }
894 
895 /*
896  * Note that there is absolutely no sense in writing out
897  * anonymous objects, so we track down the vnode object
898  * to write out.
899  * We invalidate (remove) all pages from the address space
900  * for semantic correctness.
901  *
902  * Note: certain anonymous maps, such as MAP_NOSYNC maps,
903  * may start out with a NULL object.
904  */
905 void
906 vm_object_sync(vm_object_t object, vm_ooffset_t offset, vm_size_t size,
907     boolean_t syncio, boolean_t invalidate)
908 {
909 	vm_object_t backing_object;
910 	struct vnode *vp;
911 	struct mount *mp;
912 	int flags;
913 
914 	if (object == NULL)
915 		return;
916 	VM_OBJECT_LOCK(object);
917 	while ((backing_object = object->backing_object) != NULL) {
918 		VM_OBJECT_LOCK(backing_object);
919 		offset += object->backing_object_offset;
920 		VM_OBJECT_UNLOCK(object);
921 		object = backing_object;
922 		if (object->size < OFF_TO_IDX(offset + size))
923 			size = IDX_TO_OFF(object->size) - offset;
924 	}
925 	/*
926 	 * Flush pages if writing is allowed, invalidate them
927 	 * if invalidation requested.  Pages undergoing I/O
928 	 * will be ignored by vm_object_page_remove().
929 	 *
930 	 * We cannot lock the vnode and then wait for paging
931 	 * to complete without deadlocking against vm_fault.
932 	 * Instead we simply call vm_object_page_remove() and
933 	 * allow it to block internally on a page-by-page
934 	 * basis when it encounters pages undergoing async
935 	 * I/O.
936 	 */
937 	if (object->type == OBJT_VNODE &&
938 	    (object->flags & OBJ_MIGHTBEDIRTY) != 0) {
939 		int vfslocked;
940 		vp = object->handle;
941 		VM_OBJECT_UNLOCK(object);
942 		(void) vn_start_write(vp, &mp, V_WAIT);
943 		vfslocked = VFS_LOCK_GIANT(vp->v_mount);
944 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
945 		flags = (syncio || invalidate) ? OBJPC_SYNC : 0;
946 		flags |= invalidate ? OBJPC_INVAL : 0;
947 		VM_OBJECT_LOCK(object);
948 		vm_object_page_clean(object,
949 		    OFF_TO_IDX(offset),
950 		    OFF_TO_IDX(offset + size + PAGE_MASK),
951 		    flags);
952 		VM_OBJECT_UNLOCK(object);
953 		VOP_UNLOCK(vp, 0);
954 		VFS_UNLOCK_GIANT(vfslocked);
955 		vn_finished_write(mp);
956 		VM_OBJECT_LOCK(object);
957 	}
958 	if ((object->type == OBJT_VNODE ||
959 	     object->type == OBJT_DEVICE) && invalidate) {
960 		boolean_t purge;
961 		purge = old_msync || (object->type == OBJT_DEVICE);
962 		vm_object_page_remove(object,
963 		    OFF_TO_IDX(offset),
964 		    OFF_TO_IDX(offset + size + PAGE_MASK),
965 		    purge ? FALSE : TRUE);
966 	}
967 	VM_OBJECT_UNLOCK(object);
968 }
969 
970 /*
971  *	vm_object_madvise:
972  *
973  *	Implements the madvise function at the object/page level.
974  *
975  *	MADV_WILLNEED	(any object)
976  *
977  *	    Activate the specified pages if they are resident.
978  *
979  *	MADV_DONTNEED	(any object)
980  *
981  *	    Deactivate the specified pages if they are resident.
982  *
983  *	MADV_FREE	(OBJT_DEFAULT/OBJT_SWAP objects,
984  *			 OBJ_ONEMAPPING only)
985  *
986  *	    Deactivate and clean the specified pages if they are
987  *	    resident.  This permits the process to reuse the pages
988  *	    without faulting or the kernel to reclaim the pages
989  *	    without I/O.
990  */
991 void
992 vm_object_madvise(vm_object_t object, vm_pindex_t pindex, int count, int advise)
993 {
994 	vm_pindex_t end, tpindex;
995 	vm_object_t backing_object, tobject;
996 	vm_page_t m;
997 
998 	if (object == NULL)
999 		return;
1000 	VM_OBJECT_LOCK(object);
1001 	end = pindex + count;
1002 	/*
1003 	 * Locate and adjust resident pages
1004 	 */
1005 	for (; pindex < end; pindex += 1) {
1006 relookup:
1007 		tobject = object;
1008 		tpindex = pindex;
1009 shadowlookup:
1010 		/*
1011 		 * MADV_FREE only operates on OBJT_DEFAULT or OBJT_SWAP pages
1012 		 * and those pages must be OBJ_ONEMAPPING.
1013 		 */
1014 		if (advise == MADV_FREE) {
1015 			if ((tobject->type != OBJT_DEFAULT &&
1016 			     tobject->type != OBJT_SWAP) ||
1017 			    (tobject->flags & OBJ_ONEMAPPING) == 0) {
1018 				goto unlock_tobject;
1019 			}
1020 		} else if (tobject->type == OBJT_PHYS)
1021 			goto unlock_tobject;
1022 		m = vm_page_lookup(tobject, tpindex);
1023 		if (m == NULL && advise == MADV_WILLNEED) {
1024 			/*
1025 			 * If the page is cached, reactivate it.
1026 			 */
1027 			m = vm_page_alloc(tobject, tpindex, VM_ALLOC_IFCACHED |
1028 			    VM_ALLOC_NOBUSY);
1029 		}
1030 		if (m == NULL) {
1031 			/*
1032 			 * There may be swap even if there is no backing page
1033 			 */
1034 			if (advise == MADV_FREE && tobject->type == OBJT_SWAP)
1035 				swap_pager_freespace(tobject, tpindex, 1);
1036 			/*
1037 			 * next object
1038 			 */
1039 			backing_object = tobject->backing_object;
1040 			if (backing_object == NULL)
1041 				goto unlock_tobject;
1042 			VM_OBJECT_LOCK(backing_object);
1043 			tpindex += OFF_TO_IDX(tobject->backing_object_offset);
1044 			if (tobject != object)
1045 				VM_OBJECT_UNLOCK(tobject);
1046 			tobject = backing_object;
1047 			goto shadowlookup;
1048 		} else if (m->valid != VM_PAGE_BITS_ALL)
1049 			goto unlock_tobject;
1050 		/*
1051 		 * If the page is not in a normal state, skip it.
1052 		 */
1053 		vm_page_lock(m);
1054 		if (m->hold_count != 0 || m->wire_count != 0) {
1055 			vm_page_unlock(m);
1056 			goto unlock_tobject;
1057 		}
1058 		KASSERT((m->flags & (PG_FICTITIOUS | PG_UNMANAGED)) == 0,
1059 		    ("vm_object_madvise: page %p is not managed", m));
1060 		if ((m->oflags & VPO_BUSY) || m->busy) {
1061 			if (advise == MADV_WILLNEED) {
1062 				/*
1063 				 * Reference the page before unlocking and
1064 				 * sleeping so that the page daemon is less
1065 				 * likely to reclaim it.
1066 				 */
1067 				vm_page_lock_queues();
1068 				vm_page_flag_set(m, PG_REFERENCED);
1069 				vm_page_unlock_queues();
1070 			}
1071 			vm_page_unlock(m);
1072 			if (object != tobject)
1073 				VM_OBJECT_UNLOCK(object);
1074 			m->oflags |= VPO_WANTED;
1075 			msleep(m, VM_OBJECT_MTX(tobject), PDROP | PVM, "madvpo",
1076 			    0);
1077 			VM_OBJECT_LOCK(object);
1078   			goto relookup;
1079 		}
1080 		if (advise == MADV_WILLNEED) {
1081 			vm_page_activate(m);
1082 		} else if (advise == MADV_DONTNEED) {
1083 			vm_page_dontneed(m);
1084 		} else if (advise == MADV_FREE) {
1085 			/*
1086 			 * Mark the page clean.  This will allow the page
1087 			 * to be freed up by the system.  However, such pages
1088 			 * are often reused quickly by malloc()/free()
1089 			 * so we do not do anything that would cause
1090 			 * a page fault if we can help it.
1091 			 *
1092 			 * Specifically, we do not try to actually free
1093 			 * the page now nor do we try to put it in the
1094 			 * cache (which would cause a page fault on reuse).
1095 			 *
1096 			 * But we do make the page is freeable as we
1097 			 * can without actually taking the step of unmapping
1098 			 * it.
1099 			 */
1100 			pmap_clear_modify(m);
1101 			m->dirty = 0;
1102 			m->act_count = 0;
1103 			vm_page_dontneed(m);
1104 		}
1105 		vm_page_unlock(m);
1106 		if (advise == MADV_FREE && tobject->type == OBJT_SWAP)
1107 			swap_pager_freespace(tobject, tpindex, 1);
1108 unlock_tobject:
1109 		if (tobject != object)
1110 			VM_OBJECT_UNLOCK(tobject);
1111 	}
1112 	VM_OBJECT_UNLOCK(object);
1113 }
1114 
1115 /*
1116  *	vm_object_shadow:
1117  *
1118  *	Create a new object which is backed by the
1119  *	specified existing object range.  The source
1120  *	object reference is deallocated.
1121  *
1122  *	The new object and offset into that object
1123  *	are returned in the source parameters.
1124  */
1125 void
1126 vm_object_shadow(
1127 	vm_object_t *object,	/* IN/OUT */
1128 	vm_ooffset_t *offset,	/* IN/OUT */
1129 	vm_size_t length)
1130 {
1131 	vm_object_t source;
1132 	vm_object_t result;
1133 
1134 	source = *object;
1135 
1136 	/*
1137 	 * Don't create the new object if the old object isn't shared.
1138 	 */
1139 	if (source != NULL) {
1140 		VM_OBJECT_LOCK(source);
1141 		if (source->ref_count == 1 &&
1142 		    source->handle == NULL &&
1143 		    (source->type == OBJT_DEFAULT ||
1144 		     source->type == OBJT_SWAP)) {
1145 			VM_OBJECT_UNLOCK(source);
1146 			return;
1147 		}
1148 		VM_OBJECT_UNLOCK(source);
1149 	}
1150 
1151 	/*
1152 	 * Allocate a new object with the given length.
1153 	 */
1154 	result = vm_object_allocate(OBJT_DEFAULT, length);
1155 
1156 	/*
1157 	 * The new object shadows the source object, adding a reference to it.
1158 	 * Our caller changes his reference to point to the new object,
1159 	 * removing a reference to the source object.  Net result: no change
1160 	 * of reference count.
1161 	 *
1162 	 * Try to optimize the result object's page color when shadowing
1163 	 * in order to maintain page coloring consistency in the combined
1164 	 * shadowed object.
1165 	 */
1166 	result->backing_object = source;
1167 	/*
1168 	 * Store the offset into the source object, and fix up the offset into
1169 	 * the new object.
1170 	 */
1171 	result->backing_object_offset = *offset;
1172 	if (source != NULL) {
1173 		VM_OBJECT_LOCK(source);
1174 		LIST_INSERT_HEAD(&source->shadow_head, result, shadow_list);
1175 		source->shadow_count++;
1176 #if VM_NRESERVLEVEL > 0
1177 		result->flags |= source->flags & OBJ_COLORED;
1178 		result->pg_color = (source->pg_color + OFF_TO_IDX(*offset)) &
1179 		    ((1 << (VM_NFREEORDER - 1)) - 1);
1180 #endif
1181 		VM_OBJECT_UNLOCK(source);
1182 	}
1183 
1184 
1185 	/*
1186 	 * Return the new things
1187 	 */
1188 	*offset = 0;
1189 	*object = result;
1190 }
1191 
1192 /*
1193  *	vm_object_split:
1194  *
1195  * Split the pages in a map entry into a new object.  This affords
1196  * easier removal of unused pages, and keeps object inheritance from
1197  * being a negative impact on memory usage.
1198  */
1199 void
1200 vm_object_split(vm_map_entry_t entry)
1201 {
1202 	vm_page_t m, m_next;
1203 	vm_object_t orig_object, new_object, source;
1204 	vm_pindex_t idx, offidxstart;
1205 	vm_size_t size;
1206 
1207 	orig_object = entry->object.vm_object;
1208 	if (orig_object->type != OBJT_DEFAULT && orig_object->type != OBJT_SWAP)
1209 		return;
1210 	if (orig_object->ref_count <= 1)
1211 		return;
1212 	VM_OBJECT_UNLOCK(orig_object);
1213 
1214 	offidxstart = OFF_TO_IDX(entry->offset);
1215 	size = atop(entry->end - entry->start);
1216 
1217 	/*
1218 	 * If swap_pager_copy() is later called, it will convert new_object
1219 	 * into a swap object.
1220 	 */
1221 	new_object = vm_object_allocate(OBJT_DEFAULT, size);
1222 
1223 	/*
1224 	 * At this point, the new object is still private, so the order in
1225 	 * which the original and new objects are locked does not matter.
1226 	 */
1227 	VM_OBJECT_LOCK(new_object);
1228 	VM_OBJECT_LOCK(orig_object);
1229 	source = orig_object->backing_object;
1230 	if (source != NULL) {
1231 		VM_OBJECT_LOCK(source);
1232 		if ((source->flags & OBJ_DEAD) != 0) {
1233 			VM_OBJECT_UNLOCK(source);
1234 			VM_OBJECT_UNLOCK(orig_object);
1235 			VM_OBJECT_UNLOCK(new_object);
1236 			vm_object_deallocate(new_object);
1237 			VM_OBJECT_LOCK(orig_object);
1238 			return;
1239 		}
1240 		LIST_INSERT_HEAD(&source->shadow_head,
1241 				  new_object, shadow_list);
1242 		source->shadow_count++;
1243 		vm_object_reference_locked(source);	/* for new_object */
1244 		vm_object_clear_flag(source, OBJ_ONEMAPPING);
1245 		VM_OBJECT_UNLOCK(source);
1246 		new_object->backing_object_offset =
1247 			orig_object->backing_object_offset + entry->offset;
1248 		new_object->backing_object = source;
1249 	}
1250 	if (orig_object->cred != NULL) {
1251 		new_object->cred = orig_object->cred;
1252 		crhold(orig_object->cred);
1253 		new_object->charge = ptoa(size);
1254 		KASSERT(orig_object->charge >= ptoa(size),
1255 		    ("orig_object->charge < 0"));
1256 		orig_object->charge -= ptoa(size);
1257 	}
1258 retry:
1259 	m = vm_page_find_least(orig_object, offidxstart);
1260 	for (; m != NULL && (idx = m->pindex - offidxstart) < size;
1261 	    m = m_next) {
1262 		m_next = TAILQ_NEXT(m, listq);
1263 
1264 		/*
1265 		 * We must wait for pending I/O to complete before we can
1266 		 * rename the page.
1267 		 *
1268 		 * We do not have to VM_PROT_NONE the page as mappings should
1269 		 * not be changed by this operation.
1270 		 */
1271 		if ((m->oflags & VPO_BUSY) || m->busy) {
1272 			VM_OBJECT_UNLOCK(new_object);
1273 			m->oflags |= VPO_WANTED;
1274 			msleep(m, VM_OBJECT_MTX(orig_object), PVM, "spltwt", 0);
1275 			VM_OBJECT_LOCK(new_object);
1276 			goto retry;
1277 		}
1278 		vm_page_lock(m);
1279 		vm_page_rename(m, new_object, idx);
1280 		vm_page_unlock(m);
1281 		/* page automatically made dirty by rename and cache handled */
1282 		vm_page_busy(m);
1283 	}
1284 	if (orig_object->type == OBJT_SWAP) {
1285 		/*
1286 		 * swap_pager_copy() can sleep, in which case the orig_object's
1287 		 * and new_object's locks are released and reacquired.
1288 		 */
1289 		swap_pager_copy(orig_object, new_object, offidxstart, 0);
1290 
1291 		/*
1292 		 * Transfer any cached pages from orig_object to new_object.
1293 		 */
1294 		if (__predict_false(orig_object->cache != NULL))
1295 			vm_page_cache_transfer(orig_object, offidxstart,
1296 			    new_object);
1297 	}
1298 	VM_OBJECT_UNLOCK(orig_object);
1299 	TAILQ_FOREACH(m, &new_object->memq, listq)
1300 		vm_page_wakeup(m);
1301 	VM_OBJECT_UNLOCK(new_object);
1302 	entry->object.vm_object = new_object;
1303 	entry->offset = 0LL;
1304 	vm_object_deallocate(orig_object);
1305 	VM_OBJECT_LOCK(new_object);
1306 }
1307 
1308 #define	OBSC_TEST_ALL_SHADOWED	0x0001
1309 #define	OBSC_COLLAPSE_NOWAIT	0x0002
1310 #define	OBSC_COLLAPSE_WAIT	0x0004
1311 
1312 static int
1313 vm_object_backing_scan(vm_object_t object, int op)
1314 {
1315 	int r = 1;
1316 	vm_page_t p;
1317 	vm_object_t backing_object;
1318 	vm_pindex_t backing_offset_index;
1319 
1320 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
1321 	VM_OBJECT_LOCK_ASSERT(object->backing_object, MA_OWNED);
1322 
1323 	backing_object = object->backing_object;
1324 	backing_offset_index = OFF_TO_IDX(object->backing_object_offset);
1325 
1326 	/*
1327 	 * Initial conditions
1328 	 */
1329 	if (op & OBSC_TEST_ALL_SHADOWED) {
1330 		/*
1331 		 * We do not want to have to test for the existence of cache
1332 		 * or swap pages in the backing object.  XXX but with the
1333 		 * new swapper this would be pretty easy to do.
1334 		 *
1335 		 * XXX what about anonymous MAP_SHARED memory that hasn't
1336 		 * been ZFOD faulted yet?  If we do not test for this, the
1337 		 * shadow test may succeed! XXX
1338 		 */
1339 		if (backing_object->type != OBJT_DEFAULT) {
1340 			return (0);
1341 		}
1342 	}
1343 	if (op & OBSC_COLLAPSE_WAIT) {
1344 		vm_object_set_flag(backing_object, OBJ_DEAD);
1345 	}
1346 
1347 	/*
1348 	 * Our scan
1349 	 */
1350 	p = TAILQ_FIRST(&backing_object->memq);
1351 	while (p) {
1352 		vm_page_t next = TAILQ_NEXT(p, listq);
1353 		vm_pindex_t new_pindex = p->pindex - backing_offset_index;
1354 
1355 		if (op & OBSC_TEST_ALL_SHADOWED) {
1356 			vm_page_t pp;
1357 
1358 			/*
1359 			 * Ignore pages outside the parent object's range
1360 			 * and outside the parent object's mapping of the
1361 			 * backing object.
1362 			 *
1363 			 * note that we do not busy the backing object's
1364 			 * page.
1365 			 */
1366 			if (
1367 			    p->pindex < backing_offset_index ||
1368 			    new_pindex >= object->size
1369 			) {
1370 				p = next;
1371 				continue;
1372 			}
1373 
1374 			/*
1375 			 * See if the parent has the page or if the parent's
1376 			 * object pager has the page.  If the parent has the
1377 			 * page but the page is not valid, the parent's
1378 			 * object pager must have the page.
1379 			 *
1380 			 * If this fails, the parent does not completely shadow
1381 			 * the object and we might as well give up now.
1382 			 */
1383 
1384 			pp = vm_page_lookup(object, new_pindex);
1385 			if (
1386 			    (pp == NULL || pp->valid == 0) &&
1387 			    !vm_pager_has_page(object, new_pindex, NULL, NULL)
1388 			) {
1389 				r = 0;
1390 				break;
1391 			}
1392 		}
1393 
1394 		/*
1395 		 * Check for busy page
1396 		 */
1397 		if (op & (OBSC_COLLAPSE_WAIT | OBSC_COLLAPSE_NOWAIT)) {
1398 			vm_page_t pp;
1399 
1400 			if (op & OBSC_COLLAPSE_NOWAIT) {
1401 				if ((p->oflags & VPO_BUSY) ||
1402 				    !p->valid ||
1403 				    p->busy) {
1404 					p = next;
1405 					continue;
1406 				}
1407 			} else if (op & OBSC_COLLAPSE_WAIT) {
1408 				if ((p->oflags & VPO_BUSY) || p->busy) {
1409 					VM_OBJECT_UNLOCK(object);
1410 					p->oflags |= VPO_WANTED;
1411 					msleep(p, VM_OBJECT_MTX(backing_object),
1412 					    PDROP | PVM, "vmocol", 0);
1413 					VM_OBJECT_LOCK(object);
1414 					VM_OBJECT_LOCK(backing_object);
1415 					/*
1416 					 * If we slept, anything could have
1417 					 * happened.  Since the object is
1418 					 * marked dead, the backing offset
1419 					 * should not have changed so we
1420 					 * just restart our scan.
1421 					 */
1422 					p = TAILQ_FIRST(&backing_object->memq);
1423 					continue;
1424 				}
1425 			}
1426 
1427 			KASSERT(
1428 			    p->object == backing_object,
1429 			    ("vm_object_backing_scan: object mismatch")
1430 			);
1431 
1432 			/*
1433 			 * Destroy any associated swap
1434 			 */
1435 			if (backing_object->type == OBJT_SWAP) {
1436 				swap_pager_freespace(
1437 				    backing_object,
1438 				    p->pindex,
1439 				    1
1440 				);
1441 			}
1442 
1443 			if (
1444 			    p->pindex < backing_offset_index ||
1445 			    new_pindex >= object->size
1446 			) {
1447 				/*
1448 				 * Page is out of the parent object's range, we
1449 				 * can simply destroy it.
1450 				 */
1451 				vm_page_lock(p);
1452 				KASSERT(!pmap_page_is_mapped(p),
1453 				    ("freeing mapped page %p", p));
1454 				if (p->wire_count == 0)
1455 					vm_page_free(p);
1456 				else
1457 					vm_page_remove(p);
1458 				vm_page_unlock(p);
1459 				p = next;
1460 				continue;
1461 			}
1462 
1463 			pp = vm_page_lookup(object, new_pindex);
1464 			if (
1465 			    pp != NULL ||
1466 			    vm_pager_has_page(object, new_pindex, NULL, NULL)
1467 			) {
1468 				/*
1469 				 * page already exists in parent OR swap exists
1470 				 * for this location in the parent.  Destroy
1471 				 * the original page from the backing object.
1472 				 *
1473 				 * Leave the parent's page alone
1474 				 */
1475 				vm_page_lock(p);
1476 				KASSERT(!pmap_page_is_mapped(p),
1477 				    ("freeing mapped page %p", p));
1478 				if (p->wire_count == 0)
1479 					vm_page_free(p);
1480 				else
1481 					vm_page_remove(p);
1482 				vm_page_unlock(p);
1483 				p = next;
1484 				continue;
1485 			}
1486 
1487 #if VM_NRESERVLEVEL > 0
1488 			/*
1489 			 * Rename the reservation.
1490 			 */
1491 			vm_reserv_rename(p, object, backing_object,
1492 			    backing_offset_index);
1493 #endif
1494 
1495 			/*
1496 			 * Page does not exist in parent, rename the
1497 			 * page from the backing object to the main object.
1498 			 *
1499 			 * If the page was mapped to a process, it can remain
1500 			 * mapped through the rename.
1501 			 */
1502 			vm_page_lock(p);
1503 			vm_page_rename(p, object, new_pindex);
1504 			vm_page_unlock(p);
1505 			/* page automatically made dirty by rename */
1506 		}
1507 		p = next;
1508 	}
1509 	return (r);
1510 }
1511 
1512 
1513 /*
1514  * this version of collapse allows the operation to occur earlier and
1515  * when paging_in_progress is true for an object...  This is not a complete
1516  * operation, but should plug 99.9% of the rest of the leaks.
1517  */
1518 static void
1519 vm_object_qcollapse(vm_object_t object)
1520 {
1521 	vm_object_t backing_object = object->backing_object;
1522 
1523 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
1524 	VM_OBJECT_LOCK_ASSERT(backing_object, MA_OWNED);
1525 
1526 	if (backing_object->ref_count != 1)
1527 		return;
1528 
1529 	vm_object_backing_scan(object, OBSC_COLLAPSE_NOWAIT);
1530 }
1531 
1532 /*
1533  *	vm_object_collapse:
1534  *
1535  *	Collapse an object with the object backing it.
1536  *	Pages in the backing object are moved into the
1537  *	parent, and the backing object is deallocated.
1538  */
1539 void
1540 vm_object_collapse(vm_object_t object)
1541 {
1542 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
1543 
1544 	while (TRUE) {
1545 		vm_object_t backing_object;
1546 
1547 		/*
1548 		 * Verify that the conditions are right for collapse:
1549 		 *
1550 		 * The object exists and the backing object exists.
1551 		 */
1552 		if ((backing_object = object->backing_object) == NULL)
1553 			break;
1554 
1555 		/*
1556 		 * we check the backing object first, because it is most likely
1557 		 * not collapsable.
1558 		 */
1559 		VM_OBJECT_LOCK(backing_object);
1560 		if (backing_object->handle != NULL ||
1561 		    (backing_object->type != OBJT_DEFAULT &&
1562 		     backing_object->type != OBJT_SWAP) ||
1563 		    (backing_object->flags & OBJ_DEAD) ||
1564 		    object->handle != NULL ||
1565 		    (object->type != OBJT_DEFAULT &&
1566 		     object->type != OBJT_SWAP) ||
1567 		    (object->flags & OBJ_DEAD)) {
1568 			VM_OBJECT_UNLOCK(backing_object);
1569 			break;
1570 		}
1571 
1572 		if (
1573 		    object->paging_in_progress != 0 ||
1574 		    backing_object->paging_in_progress != 0
1575 		) {
1576 			vm_object_qcollapse(object);
1577 			VM_OBJECT_UNLOCK(backing_object);
1578 			break;
1579 		}
1580 		/*
1581 		 * We know that we can either collapse the backing object (if
1582 		 * the parent is the only reference to it) or (perhaps) have
1583 		 * the parent bypass the object if the parent happens to shadow
1584 		 * all the resident pages in the entire backing object.
1585 		 *
1586 		 * This is ignoring pager-backed pages such as swap pages.
1587 		 * vm_object_backing_scan fails the shadowing test in this
1588 		 * case.
1589 		 */
1590 		if (backing_object->ref_count == 1) {
1591 			/*
1592 			 * If there is exactly one reference to the backing
1593 			 * object, we can collapse it into the parent.
1594 			 */
1595 			vm_object_backing_scan(object, OBSC_COLLAPSE_WAIT);
1596 
1597 #if VM_NRESERVLEVEL > 0
1598 			/*
1599 			 * Break any reservations from backing_object.
1600 			 */
1601 			if (__predict_false(!LIST_EMPTY(&backing_object->rvq)))
1602 				vm_reserv_break_all(backing_object);
1603 #endif
1604 
1605 			/*
1606 			 * Move the pager from backing_object to object.
1607 			 */
1608 			if (backing_object->type == OBJT_SWAP) {
1609 				/*
1610 				 * swap_pager_copy() can sleep, in which case
1611 				 * the backing_object's and object's locks are
1612 				 * released and reacquired.
1613 				 */
1614 				swap_pager_copy(
1615 				    backing_object,
1616 				    object,
1617 				    OFF_TO_IDX(object->backing_object_offset), TRUE);
1618 
1619 				/*
1620 				 * Free any cached pages from backing_object.
1621 				 */
1622 				if (__predict_false(backing_object->cache != NULL))
1623 					vm_page_cache_free(backing_object, 0, 0);
1624 			}
1625 			/*
1626 			 * Object now shadows whatever backing_object did.
1627 			 * Note that the reference to
1628 			 * backing_object->backing_object moves from within
1629 			 * backing_object to within object.
1630 			 */
1631 			LIST_REMOVE(object, shadow_list);
1632 			backing_object->shadow_count--;
1633 			if (backing_object->backing_object) {
1634 				VM_OBJECT_LOCK(backing_object->backing_object);
1635 				LIST_REMOVE(backing_object, shadow_list);
1636 				LIST_INSERT_HEAD(
1637 				    &backing_object->backing_object->shadow_head,
1638 				    object, shadow_list);
1639 				/*
1640 				 * The shadow_count has not changed.
1641 				 */
1642 				VM_OBJECT_UNLOCK(backing_object->backing_object);
1643 			}
1644 			object->backing_object = backing_object->backing_object;
1645 			object->backing_object_offset +=
1646 			    backing_object->backing_object_offset;
1647 
1648 			/*
1649 			 * Discard backing_object.
1650 			 *
1651 			 * Since the backing object has no pages, no pager left,
1652 			 * and no object references within it, all that is
1653 			 * necessary is to dispose of it.
1654 			 */
1655 			KASSERT(backing_object->ref_count == 1, (
1656 "backing_object %p was somehow re-referenced during collapse!",
1657 			    backing_object));
1658 			VM_OBJECT_UNLOCK(backing_object);
1659 			vm_object_destroy(backing_object);
1660 
1661 			object_collapses++;
1662 		} else {
1663 			vm_object_t new_backing_object;
1664 
1665 			/*
1666 			 * If we do not entirely shadow the backing object,
1667 			 * there is nothing we can do so we give up.
1668 			 */
1669 			if (object->resident_page_count != object->size &&
1670 			    vm_object_backing_scan(object,
1671 			    OBSC_TEST_ALL_SHADOWED) == 0) {
1672 				VM_OBJECT_UNLOCK(backing_object);
1673 				break;
1674 			}
1675 
1676 			/*
1677 			 * Make the parent shadow the next object in the
1678 			 * chain.  Deallocating backing_object will not remove
1679 			 * it, since its reference count is at least 2.
1680 			 */
1681 			LIST_REMOVE(object, shadow_list);
1682 			backing_object->shadow_count--;
1683 
1684 			new_backing_object = backing_object->backing_object;
1685 			if ((object->backing_object = new_backing_object) != NULL) {
1686 				VM_OBJECT_LOCK(new_backing_object);
1687 				LIST_INSERT_HEAD(
1688 				    &new_backing_object->shadow_head,
1689 				    object,
1690 				    shadow_list
1691 				);
1692 				new_backing_object->shadow_count++;
1693 				vm_object_reference_locked(new_backing_object);
1694 				VM_OBJECT_UNLOCK(new_backing_object);
1695 				object->backing_object_offset +=
1696 					backing_object->backing_object_offset;
1697 			}
1698 
1699 			/*
1700 			 * Drop the reference count on backing_object. Since
1701 			 * its ref_count was at least 2, it will not vanish.
1702 			 */
1703 			backing_object->ref_count--;
1704 			VM_OBJECT_UNLOCK(backing_object);
1705 			object_bypasses++;
1706 		}
1707 
1708 		/*
1709 		 * Try again with this object's new backing object.
1710 		 */
1711 	}
1712 }
1713 
1714 /*
1715  *	vm_object_page_remove:
1716  *
1717  *	For the given object, either frees or invalidates each of the
1718  *	specified pages.  In general, a page is freed.  However, if a
1719  *	page is wired for any reason other than the existence of a
1720  *	managed, wired mapping, then it may be invalidated but not
1721  *	removed from the object.  Pages are specified by the given
1722  *	range ["start", "end") and Boolean "clean_only".  As a
1723  *	special case, if "end" is zero, then the range extends from
1724  *	"start" to the end of the object.  If "clean_only" is TRUE,
1725  *	then only the non-dirty pages within the specified range are
1726  *	affected.
1727  *
1728  *	In general, this operation should only be performed on objects
1729  *	that contain managed pages.  There are two exceptions.  First,
1730  *	it may be performed on the kernel and kmem objects.  Second,
1731  *	it may be used by msync(..., MS_INVALIDATE) to invalidate
1732  *	device-backed pages.  In both of these cases, "clean_only"
1733  *	must be FALSE.
1734  *
1735  *	The object must be locked.
1736  */
1737 void
1738 vm_object_page_remove(vm_object_t object, vm_pindex_t start, vm_pindex_t end,
1739     boolean_t clean_only)
1740 {
1741 	vm_page_t p, next;
1742 	int wirings;
1743 
1744 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
1745 	if (object->resident_page_count == 0)
1746 		goto skipmemq;
1747 
1748 	/*
1749 	 * Since physically-backed objects do not use managed pages, we can't
1750 	 * remove pages from the object (we must instead remove the page
1751 	 * references, and then destroy the object).
1752 	 */
1753 	KASSERT(object->type != OBJT_PHYS || object == kernel_object ||
1754 	    object == kmem_object,
1755 	    ("attempt to remove pages from a physical object"));
1756 
1757 	vm_object_pip_add(object, 1);
1758 again:
1759 	p = vm_page_find_least(object, start);
1760 
1761 	/*
1762 	 * Assert: the variable p is either (1) the page with the
1763 	 * least pindex greater than or equal to the parameter pindex
1764 	 * or (2) NULL.
1765 	 */
1766 	for (;
1767 	     p != NULL && (p->pindex < end || end == 0);
1768 	     p = next) {
1769 		next = TAILQ_NEXT(p, listq);
1770 
1771 		/*
1772 		 * If the page is wired for any reason besides the
1773 		 * existence of managed, wired mappings, then it cannot
1774 		 * be freed.  For example, fictitious pages, which
1775 		 * represent device memory, are inherently wired and
1776 		 * cannot be freed.  They can, however, be invalidated
1777 		 * if "clean_only" is FALSE.
1778 		 */
1779 		vm_page_lock(p);
1780 		if ((wirings = p->wire_count) != 0 &&
1781 		    (wirings = pmap_page_wired_mappings(p)) != p->wire_count) {
1782 			/* Fictitious pages do not have managed mappings. */
1783 			if ((p->flags & PG_FICTITIOUS) == 0)
1784 				pmap_remove_all(p);
1785 			/* Account for removal of managed, wired mappings. */
1786 			p->wire_count -= wirings;
1787 			if (!clean_only) {
1788 				p->valid = 0;
1789 				vm_page_undirty(p);
1790 			}
1791 			vm_page_unlock(p);
1792 			continue;
1793 		}
1794 		if (vm_page_sleep_if_busy(p, TRUE, "vmopar"))
1795 			goto again;
1796 		KASSERT((p->flags & PG_FICTITIOUS) == 0,
1797 		    ("vm_object_page_remove: page %p is fictitious", p));
1798 		if (clean_only && p->valid) {
1799 			pmap_remove_write(p);
1800 			if (p->dirty) {
1801 				vm_page_unlock(p);
1802 				continue;
1803 			}
1804 		}
1805 		pmap_remove_all(p);
1806 		/* Account for removal of managed, wired mappings. */
1807 		if (wirings != 0)
1808 			p->wire_count -= wirings;
1809 		vm_page_free(p);
1810 		vm_page_unlock(p);
1811 	}
1812 	vm_object_pip_wakeup(object);
1813 skipmemq:
1814 	if (__predict_false(object->cache != NULL))
1815 		vm_page_cache_free(object, start, end);
1816 }
1817 
1818 /*
1819  *	Populate the specified range of the object with valid pages.  Returns
1820  *	TRUE if the range is successfully populated and FALSE otherwise.
1821  *
1822  *	Note: This function should be optimized to pass a larger array of
1823  *	pages to vm_pager_get_pages() before it is applied to a non-
1824  *	OBJT_DEVICE object.
1825  *
1826  *	The object must be locked.
1827  */
1828 boolean_t
1829 vm_object_populate(vm_object_t object, vm_pindex_t start, vm_pindex_t end)
1830 {
1831 	vm_page_t m, ma[1];
1832 	vm_pindex_t pindex;
1833 	int rv;
1834 
1835 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
1836 	for (pindex = start; pindex < end; pindex++) {
1837 		m = vm_page_grab(object, pindex, VM_ALLOC_NORMAL |
1838 		    VM_ALLOC_RETRY);
1839 		if (m->valid != VM_PAGE_BITS_ALL) {
1840 			ma[0] = m;
1841 			rv = vm_pager_get_pages(object, ma, 1, 0);
1842 			m = vm_page_lookup(object, pindex);
1843 			if (m == NULL)
1844 				break;
1845 			if (rv != VM_PAGER_OK) {
1846 				vm_page_lock(m);
1847 				vm_page_free(m);
1848 				vm_page_unlock(m);
1849 				break;
1850 			}
1851 		}
1852 		/*
1853 		 * Keep "m" busy because a subsequent iteration may unlock
1854 		 * the object.
1855 		 */
1856 	}
1857 	if (pindex > start) {
1858 		m = vm_page_lookup(object, start);
1859 		while (m != NULL && m->pindex < pindex) {
1860 			vm_page_wakeup(m);
1861 			m = TAILQ_NEXT(m, listq);
1862 		}
1863 	}
1864 	return (pindex == end);
1865 }
1866 
1867 /*
1868  *	Routine:	vm_object_coalesce
1869  *	Function:	Coalesces two objects backing up adjoining
1870  *			regions of memory into a single object.
1871  *
1872  *	returns TRUE if objects were combined.
1873  *
1874  *	NOTE:	Only works at the moment if the second object is NULL -
1875  *		if it's not, which object do we lock first?
1876  *
1877  *	Parameters:
1878  *		prev_object	First object to coalesce
1879  *		prev_offset	Offset into prev_object
1880  *		prev_size	Size of reference to prev_object
1881  *		next_size	Size of reference to the second object
1882  *		reserved	Indicator that extension region has
1883  *				swap accounted for
1884  *
1885  *	Conditions:
1886  *	The object must *not* be locked.
1887  */
1888 boolean_t
1889 vm_object_coalesce(vm_object_t prev_object, vm_ooffset_t prev_offset,
1890     vm_size_t prev_size, vm_size_t next_size, boolean_t reserved)
1891 {
1892 	vm_pindex_t next_pindex;
1893 
1894 	if (prev_object == NULL)
1895 		return (TRUE);
1896 	VM_OBJECT_LOCK(prev_object);
1897 	if (prev_object->type != OBJT_DEFAULT &&
1898 	    prev_object->type != OBJT_SWAP) {
1899 		VM_OBJECT_UNLOCK(prev_object);
1900 		return (FALSE);
1901 	}
1902 
1903 	/*
1904 	 * Try to collapse the object first
1905 	 */
1906 	vm_object_collapse(prev_object);
1907 
1908 	/*
1909 	 * Can't coalesce if: . more than one reference . paged out . shadows
1910 	 * another object . has a copy elsewhere (any of which mean that the
1911 	 * pages not mapped to prev_entry may be in use anyway)
1912 	 */
1913 	if (prev_object->backing_object != NULL) {
1914 		VM_OBJECT_UNLOCK(prev_object);
1915 		return (FALSE);
1916 	}
1917 
1918 	prev_size >>= PAGE_SHIFT;
1919 	next_size >>= PAGE_SHIFT;
1920 	next_pindex = OFF_TO_IDX(prev_offset) + prev_size;
1921 
1922 	if ((prev_object->ref_count > 1) &&
1923 	    (prev_object->size != next_pindex)) {
1924 		VM_OBJECT_UNLOCK(prev_object);
1925 		return (FALSE);
1926 	}
1927 
1928 	/*
1929 	 * Account for the charge.
1930 	 */
1931 	if (prev_object->cred != NULL) {
1932 
1933 		/*
1934 		 * If prev_object was charged, then this mapping,
1935 		 * althought not charged now, may become writable
1936 		 * later. Non-NULL cred in the object would prevent
1937 		 * swap reservation during enabling of the write
1938 		 * access, so reserve swap now. Failed reservation
1939 		 * cause allocation of the separate object for the map
1940 		 * entry, and swap reservation for this entry is
1941 		 * managed in appropriate time.
1942 		 */
1943 		if (!reserved && !swap_reserve_by_cred(ptoa(next_size),
1944 		    prev_object->cred)) {
1945 			return (FALSE);
1946 		}
1947 		prev_object->charge += ptoa(next_size);
1948 	}
1949 
1950 	/*
1951 	 * Remove any pages that may still be in the object from a previous
1952 	 * deallocation.
1953 	 */
1954 	if (next_pindex < prev_object->size) {
1955 		vm_object_page_remove(prev_object,
1956 				      next_pindex,
1957 				      next_pindex + next_size, FALSE);
1958 		if (prev_object->type == OBJT_SWAP)
1959 			swap_pager_freespace(prev_object,
1960 					     next_pindex, next_size);
1961 #if 0
1962 		if (prev_object->cred != NULL) {
1963 			KASSERT(prev_object->charge >=
1964 			    ptoa(prev_object->size - next_pindex),
1965 			    ("object %p overcharged 1 %jx %jx", prev_object,
1966 				(uintmax_t)next_pindex, (uintmax_t)next_size));
1967 			prev_object->charge -= ptoa(prev_object->size -
1968 			    next_pindex);
1969 		}
1970 #endif
1971 	}
1972 
1973 	/*
1974 	 * Extend the object if necessary.
1975 	 */
1976 	if (next_pindex + next_size > prev_object->size)
1977 		prev_object->size = next_pindex + next_size;
1978 
1979 	VM_OBJECT_UNLOCK(prev_object);
1980 	return (TRUE);
1981 }
1982 
1983 void
1984 vm_object_set_writeable_dirty(vm_object_t object)
1985 {
1986 
1987 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
1988 	if (object->type != OBJT_VNODE ||
1989 	    (object->flags & OBJ_MIGHTBEDIRTY) != 0)
1990 		return;
1991 	vm_object_set_flag(object, OBJ_MIGHTBEDIRTY);
1992 }
1993 
1994 #include "opt_ddb.h"
1995 #ifdef DDB
1996 #include <sys/kernel.h>
1997 
1998 #include <sys/cons.h>
1999 
2000 #include <ddb/ddb.h>
2001 
2002 static int
2003 _vm_object_in_map(vm_map_t map, vm_object_t object, vm_map_entry_t entry)
2004 {
2005 	vm_map_t tmpm;
2006 	vm_map_entry_t tmpe;
2007 	vm_object_t obj;
2008 	int entcount;
2009 
2010 	if (map == 0)
2011 		return 0;
2012 
2013 	if (entry == 0) {
2014 		tmpe = map->header.next;
2015 		entcount = map->nentries;
2016 		while (entcount-- && (tmpe != &map->header)) {
2017 			if (_vm_object_in_map(map, object, tmpe)) {
2018 				return 1;
2019 			}
2020 			tmpe = tmpe->next;
2021 		}
2022 	} else if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
2023 		tmpm = entry->object.sub_map;
2024 		tmpe = tmpm->header.next;
2025 		entcount = tmpm->nentries;
2026 		while (entcount-- && tmpe != &tmpm->header) {
2027 			if (_vm_object_in_map(tmpm, object, tmpe)) {
2028 				return 1;
2029 			}
2030 			tmpe = tmpe->next;
2031 		}
2032 	} else if ((obj = entry->object.vm_object) != NULL) {
2033 		for (; obj; obj = obj->backing_object)
2034 			if (obj == object) {
2035 				return 1;
2036 			}
2037 	}
2038 	return 0;
2039 }
2040 
2041 static int
2042 vm_object_in_map(vm_object_t object)
2043 {
2044 	struct proc *p;
2045 
2046 	/* sx_slock(&allproc_lock); */
2047 	FOREACH_PROC_IN_SYSTEM(p) {
2048 		if (!p->p_vmspace /* || (p->p_flag & (P_SYSTEM|P_WEXIT)) */)
2049 			continue;
2050 		if (_vm_object_in_map(&p->p_vmspace->vm_map, object, 0)) {
2051 			/* sx_sunlock(&allproc_lock); */
2052 			return 1;
2053 		}
2054 	}
2055 	/* sx_sunlock(&allproc_lock); */
2056 	if (_vm_object_in_map(kernel_map, object, 0))
2057 		return 1;
2058 	if (_vm_object_in_map(kmem_map, object, 0))
2059 		return 1;
2060 	if (_vm_object_in_map(pager_map, object, 0))
2061 		return 1;
2062 	if (_vm_object_in_map(buffer_map, object, 0))
2063 		return 1;
2064 	return 0;
2065 }
2066 
2067 DB_SHOW_COMMAND(vmochk, vm_object_check)
2068 {
2069 	vm_object_t object;
2070 
2071 	/*
2072 	 * make sure that internal objs are in a map somewhere
2073 	 * and none have zero ref counts.
2074 	 */
2075 	TAILQ_FOREACH(object, &vm_object_list, object_list) {
2076 		if (object->handle == NULL &&
2077 		    (object->type == OBJT_DEFAULT || object->type == OBJT_SWAP)) {
2078 			if (object->ref_count == 0) {
2079 				db_printf("vmochk: internal obj has zero ref count: %ld\n",
2080 					(long)object->size);
2081 			}
2082 			if (!vm_object_in_map(object)) {
2083 				db_printf(
2084 			"vmochk: internal obj is not in a map: "
2085 			"ref: %d, size: %lu: 0x%lx, backing_object: %p\n",
2086 				    object->ref_count, (u_long)object->size,
2087 				    (u_long)object->size,
2088 				    (void *)object->backing_object);
2089 			}
2090 		}
2091 	}
2092 }
2093 
2094 /*
2095  *	vm_object_print:	[ debug ]
2096  */
2097 DB_SHOW_COMMAND(object, vm_object_print_static)
2098 {
2099 	/* XXX convert args. */
2100 	vm_object_t object = (vm_object_t)addr;
2101 	boolean_t full = have_addr;
2102 
2103 	vm_page_t p;
2104 
2105 	/* XXX count is an (unused) arg.  Avoid shadowing it. */
2106 #define	count	was_count
2107 
2108 	int count;
2109 
2110 	if (object == NULL)
2111 		return;
2112 
2113 	db_iprintf(
2114 	    "Object %p: type=%d, size=0x%jx, res=%d, ref=%d, flags=0x%x ruid %d charge %jx\n",
2115 	    object, (int)object->type, (uintmax_t)object->size,
2116 	    object->resident_page_count, object->ref_count, object->flags,
2117 	    object->cred ? object->cred->cr_ruid : -1, (uintmax_t)object->charge);
2118 	db_iprintf(" sref=%d, backing_object(%d)=(%p)+0x%jx\n",
2119 	    object->shadow_count,
2120 	    object->backing_object ? object->backing_object->ref_count : 0,
2121 	    object->backing_object, (uintmax_t)object->backing_object_offset);
2122 
2123 	if (!full)
2124 		return;
2125 
2126 	db_indent += 2;
2127 	count = 0;
2128 	TAILQ_FOREACH(p, &object->memq, listq) {
2129 		if (count == 0)
2130 			db_iprintf("memory:=");
2131 		else if (count == 6) {
2132 			db_printf("\n");
2133 			db_iprintf(" ...");
2134 			count = 0;
2135 		} else
2136 			db_printf(",");
2137 		count++;
2138 
2139 		db_printf("(off=0x%jx,page=0x%jx)",
2140 		    (uintmax_t)p->pindex, (uintmax_t)VM_PAGE_TO_PHYS(p));
2141 	}
2142 	if (count != 0)
2143 		db_printf("\n");
2144 	db_indent -= 2;
2145 }
2146 
2147 /* XXX. */
2148 #undef count
2149 
2150 /* XXX need this non-static entry for calling from vm_map_print. */
2151 void
2152 vm_object_print(
2153         /* db_expr_t */ long addr,
2154 	boolean_t have_addr,
2155 	/* db_expr_t */ long count,
2156 	char *modif)
2157 {
2158 	vm_object_print_static(addr, have_addr, count, modif);
2159 }
2160 
2161 DB_SHOW_COMMAND(vmopag, vm_object_print_pages)
2162 {
2163 	vm_object_t object;
2164 	vm_pindex_t fidx;
2165 	vm_paddr_t pa;
2166 	vm_page_t m, prev_m;
2167 	int rcount, nl, c;
2168 
2169 	nl = 0;
2170 	TAILQ_FOREACH(object, &vm_object_list, object_list) {
2171 		db_printf("new object: %p\n", (void *)object);
2172 		if (nl > 18) {
2173 			c = cngetc();
2174 			if (c != ' ')
2175 				return;
2176 			nl = 0;
2177 		}
2178 		nl++;
2179 		rcount = 0;
2180 		fidx = 0;
2181 		pa = -1;
2182 		TAILQ_FOREACH(m, &object->memq, listq) {
2183 			if (m->pindex > 128)
2184 				break;
2185 			if ((prev_m = TAILQ_PREV(m, pglist, listq)) != NULL &&
2186 			    prev_m->pindex + 1 != m->pindex) {
2187 				if (rcount) {
2188 					db_printf(" index(%ld)run(%d)pa(0x%lx)\n",
2189 						(long)fidx, rcount, (long)pa);
2190 					if (nl > 18) {
2191 						c = cngetc();
2192 						if (c != ' ')
2193 							return;
2194 						nl = 0;
2195 					}
2196 					nl++;
2197 					rcount = 0;
2198 				}
2199 			}
2200 			if (rcount &&
2201 				(VM_PAGE_TO_PHYS(m) == pa + rcount * PAGE_SIZE)) {
2202 				++rcount;
2203 				continue;
2204 			}
2205 			if (rcount) {
2206 				db_printf(" index(%ld)run(%d)pa(0x%lx)\n",
2207 					(long)fidx, rcount, (long)pa);
2208 				if (nl > 18) {
2209 					c = cngetc();
2210 					if (c != ' ')
2211 						return;
2212 					nl = 0;
2213 				}
2214 				nl++;
2215 			}
2216 			fidx = m->pindex;
2217 			pa = VM_PAGE_TO_PHYS(m);
2218 			rcount = 1;
2219 		}
2220 		if (rcount) {
2221 			db_printf(" index(%ld)run(%d)pa(0x%lx)\n",
2222 				(long)fidx, rcount, (long)pa);
2223 			if (nl > 18) {
2224 				c = cngetc();
2225 				if (c != ' ')
2226 					return;
2227 				nl = 0;
2228 			}
2229 			nl++;
2230 		}
2231 	}
2232 }
2233 #endif /* DDB */
2234