xref: /freebsd/sys/vm/vm_object.c (revision a3cf0ef5a295c885c895fabfd56470c0d1db322d)
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->uip = 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 			temp->generation++;
604 			VM_OBJECT_UNLOCK(temp);
605 			object->backing_object = NULL;
606 		}
607 		/*
608 		 * Don't double-terminate, we could be in a termination
609 		 * recursion due to the terminate having to sync data
610 		 * to disk.
611 		 */
612 		if ((object->flags & OBJ_DEAD) == 0)
613 			vm_object_terminate(object);
614 		else
615 			VM_OBJECT_UNLOCK(object);
616 		object = temp;
617 	}
618 }
619 
620 /*
621  *	vm_object_destroy removes the object from the global object list
622  *      and frees the space for the object.
623  */
624 void
625 vm_object_destroy(vm_object_t object)
626 {
627 
628 	/*
629 	 * Remove the object from the global object list.
630 	 */
631 	mtx_lock(&vm_object_list_mtx);
632 	TAILQ_REMOVE(&vm_object_list, object, object_list);
633 	mtx_unlock(&vm_object_list_mtx);
634 
635 	/*
636 	 * Release the allocation charge.
637 	 */
638 	if (object->uip != NULL) {
639 		KASSERT(object->type == OBJT_DEFAULT ||
640 		    object->type == OBJT_SWAP,
641 		    ("vm_object_terminate: non-swap obj %p has uip",
642 		     object));
643 		swap_release_by_uid(object->charge, object->uip);
644 		object->charge = 0;
645 		uifree(object->uip);
646 		object->uip = NULL;
647 	}
648 
649 	/*
650 	 * Free the space for the object.
651 	 */
652 	uma_zfree(obj_zone, object);
653 }
654 
655 /*
656  *	vm_object_terminate actually destroys the specified object, freeing
657  *	up all previously used resources.
658  *
659  *	The object must be locked.
660  *	This routine may block.
661  */
662 void
663 vm_object_terminate(vm_object_t object)
664 {
665 	vm_page_t p;
666 
667 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
668 
669 	/*
670 	 * Make sure no one uses us.
671 	 */
672 	vm_object_set_flag(object, OBJ_DEAD);
673 
674 	/*
675 	 * wait for the pageout daemon to be done with the object
676 	 */
677 	vm_object_pip_wait(object, "objtrm");
678 
679 	KASSERT(!object->paging_in_progress,
680 		("vm_object_terminate: pageout in progress"));
681 
682 	/*
683 	 * Clean and free the pages, as appropriate. All references to the
684 	 * object are gone, so we don't need to lock it.
685 	 */
686 	if (object->type == OBJT_VNODE) {
687 		struct vnode *vp = (struct vnode *)object->handle;
688 
689 		/*
690 		 * Clean pages and flush buffers.
691 		 */
692 		vm_object_page_clean(object, 0, 0, OBJPC_SYNC);
693 		VM_OBJECT_UNLOCK(object);
694 
695 		vinvalbuf(vp, V_SAVE, 0, 0);
696 
697 		VM_OBJECT_LOCK(object);
698 	}
699 
700 	KASSERT(object->ref_count == 0,
701 		("vm_object_terminate: object with references, ref_count=%d",
702 		object->ref_count));
703 
704 	/*
705 	 * Now free any remaining pages. For internal objects, this also
706 	 * removes them from paging queues. Don't free wired pages, just
707 	 * remove them from the object.
708 	 */
709 	while ((p = TAILQ_FIRST(&object->memq)) != NULL) {
710 		KASSERT(!p->busy && (p->oflags & VPO_BUSY) == 0,
711 			("vm_object_terminate: freeing busy page %p "
712 			"p->busy = %d, p->oflags %x\n", p, p->busy, p->oflags));
713 		vm_page_lock(p);
714 		if (p->wire_count == 0) {
715 			vm_page_free(p);
716 			PCPU_INC(cnt.v_pfree);
717 		} else
718 			vm_page_remove(p);
719 		vm_page_unlock(p);
720 	}
721 
722 #if VM_NRESERVLEVEL > 0
723 	if (__predict_false(!LIST_EMPTY(&object->rvq)))
724 		vm_reserv_break_all(object);
725 #endif
726 	if (__predict_false(object->cache != NULL))
727 		vm_page_cache_free(object, 0, 0);
728 
729 	/*
730 	 * Let the pager know object is dead.
731 	 */
732 	vm_pager_deallocate(object);
733 	VM_OBJECT_UNLOCK(object);
734 
735 	vm_object_destroy(object);
736 }
737 
738 /*
739  *	vm_object_page_clean
740  *
741  *	Clean all dirty pages in the specified range of object.  Leaves page
742  * 	on whatever queue it is currently on.   If NOSYNC is set then do not
743  *	write out pages with VPO_NOSYNC set (originally comes from MAP_NOSYNC),
744  *	leaving the object dirty.
745  *
746  *	When stuffing pages asynchronously, allow clustering.  XXX we need a
747  *	synchronous clustering mode implementation.
748  *
749  *	Odd semantics: if start == end, we clean everything.
750  *
751  *	The object must be locked.
752  */
753 void
754 vm_object_page_clean(vm_object_t object, vm_pindex_t start, vm_pindex_t end,
755     int flags)
756 {
757 	vm_page_t np, p;
758 	vm_pindex_t pi, tend;
759 	int clearobjflags, curgeneration, n, pagerflags;
760 
761 	mtx_assert(&vm_page_queue_mtx, MA_NOTOWNED);
762 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
763 	KASSERT(object->type == OBJT_VNODE, ("Not a vnode object"));
764 	if ((object->flags & OBJ_MIGHTBEDIRTY) == 0 ||
765 	    object->resident_page_count == 0)
766 		return;
767 
768 	pagerflags = (flags & (OBJPC_SYNC | OBJPC_INVAL)) != 0 ?
769 	    VM_PAGER_PUT_SYNC : VM_PAGER_CLUSTER_OK;
770 	pagerflags |= (flags & OBJPC_INVAL) != 0 ? VM_PAGER_PUT_INVAL : 0;
771 
772 	tend = (end == 0) ? object->size : end;
773 
774 	vm_object_set_flag(object, OBJ_CLEANING);
775 
776 	/*
777 	 * Make the page read-only so we can then clear the object flags.
778 	 *
779 	 * However, if this is a nosync mmap then the object is likely to
780 	 * stay dirty so do not mess with the page and do not clear the
781 	 * object flags.
782 	 */
783 	clearobjflags = 1;
784 	for (p = vm_page_find_least(object, start);
785 	    p != NULL && p->pindex < tend; p = TAILQ_NEXT(p, listq)) {
786 		if ((flags & OBJPC_NOSYNC) != 0 &&
787 		    (p->oflags & VPO_NOSYNC) != 0)
788 			clearobjflags = 0;
789 		else
790 			pmap_remove_write(p);
791 	}
792 
793 	if (clearobjflags && (start == 0) && (tend == object->size))
794 		vm_object_clear_flag(object, OBJ_MIGHTBEDIRTY);
795 
796 rescan:
797 	curgeneration = object->generation;
798 
799 	for (p = vm_page_find_least(object, start); p != NULL; p = np) {
800 		pi = p->pindex;
801 		if (pi >= tend)
802 			break;
803 		np = TAILQ_NEXT(p, listq);
804 		if (p->valid == 0)
805 			continue;
806 		while (vm_page_sleep_if_busy(p, TRUE, "vpcwai")) {
807 			if (object->generation != curgeneration)
808 				goto rescan;
809 		}
810 		vm_page_test_dirty(p);
811 		if (p->dirty == 0)
812 			continue;
813 
814 		/*
815 		 * If we have been asked to skip nosync pages and this is a
816 		 * nosync page, skip it.  Note that the object flags were
817 		 * not cleared in this case so we do not have to set them.
818 		 */
819 		if ((flags & OBJPC_NOSYNC) != 0 &&
820 		    (p->oflags & VPO_NOSYNC) != 0)
821 			continue;
822 
823 		n = vm_object_page_collect_flush(object, p, pagerflags);
824 		KASSERT(n > 0, ("vm_object_page_collect_flush failed"));
825 		if (object->generation != curgeneration)
826 			goto rescan;
827 		np = vm_page_find_least(object, pi + n);
828 	}
829 #if 0
830 	VOP_FSYNC(vp, (pagerflags & VM_PAGER_PUT_SYNC) ? MNT_WAIT : 0);
831 #endif
832 
833 	vm_object_clear_flag(object, OBJ_CLEANING);
834 }
835 
836 static int
837 vm_object_page_collect_flush(vm_object_t object, vm_page_t p, int pagerflags)
838 {
839 	int runlen;
840 	int maxf;
841 	int chkb;
842 	int maxb;
843 	int i, index;
844 	vm_pindex_t pi;
845 	vm_page_t maf[vm_pageout_page_count];
846 	vm_page_t mab[vm_pageout_page_count];
847 	vm_page_t ma[vm_pageout_page_count];
848 	vm_page_t tp, p1;
849 
850 	mtx_assert(&vm_page_queue_mtx, MA_NOTOWNED);
851 	vm_page_lock_assert(p, MA_NOTOWNED);
852 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
853 	pi = p->pindex;
854 	maxf = 0;
855 	for (i = 1, p1 = p; i < vm_pageout_page_count; i++) {
856 		tp = vm_page_next(p1);
857 		if (tp == NULL || tp->busy != 0 || (tp->oflags & VPO_BUSY) != 0)
858 			break;
859 		vm_page_test_dirty(tp);
860 		if (tp->dirty == 0)
861 			break;
862 		maf[i - 1] = p1 = tp;
863 		maxf++;
864 	}
865 
866 	maxb = 0;
867 	chkb = vm_pageout_page_count -  maxf;
868 	for (i = 1, p1 = p; i < chkb; i++) {
869 		tp = vm_page_prev(p1);
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 		mab[i - 1] = p1 = tp;
876 		maxb++;
877 	}
878 
879 	for (i = 0; i < maxb; i++) {
880 		index = (maxb - i) - 1;
881 		ma[index] = mab[i];
882 	}
883 	ma[maxb] = p;
884 	for (i = 0; i < maxf; i++) {
885 		index = (maxb + i) + 1;
886 		ma[index] = maf[i];
887 	}
888 	runlen = maxb + maxf + 1;
889 
890 	vm_pageout_flush(ma, runlen, pagerflags);
891 	for (i = 0; i < runlen; i++) {
892 		if (ma[i]->dirty != 0) {
893 			KASSERT((ma[i]->flags & PG_WRITEABLE) == 0,
894 	("vm_object_page_collect_flush: page %p is not write protected",
895 			    ma[i]));
896 		}
897 	}
898 	for (i = 0; i < maxf; i++) {
899 		if (ma[i + maxb + 1]->dirty != 0) {
900 			/*
901 			 * maxf will end up being the actual number of pages
902 			 * we wrote out contiguously, non-inclusive of the
903 			 * first page.  We do not count look-behind pages.
904 			 */
905 			if (maxf > i) {
906 				maxf = i;
907 				break;
908 			}
909 		}
910 	}
911 	return (maxf + 1);
912 }
913 
914 /*
915  * Note that there is absolutely no sense in writing out
916  * anonymous objects, so we track down the vnode object
917  * to write out.
918  * We invalidate (remove) all pages from the address space
919  * for semantic correctness.
920  *
921  * Note: certain anonymous maps, such as MAP_NOSYNC maps,
922  * may start out with a NULL object.
923  */
924 void
925 vm_object_sync(vm_object_t object, vm_ooffset_t offset, vm_size_t size,
926     boolean_t syncio, boolean_t invalidate)
927 {
928 	vm_object_t backing_object;
929 	struct vnode *vp;
930 	struct mount *mp;
931 	int flags;
932 
933 	if (object == NULL)
934 		return;
935 	VM_OBJECT_LOCK(object);
936 	while ((backing_object = object->backing_object) != NULL) {
937 		VM_OBJECT_LOCK(backing_object);
938 		offset += object->backing_object_offset;
939 		VM_OBJECT_UNLOCK(object);
940 		object = backing_object;
941 		if (object->size < OFF_TO_IDX(offset + size))
942 			size = IDX_TO_OFF(object->size) - offset;
943 	}
944 	/*
945 	 * Flush pages if writing is allowed, invalidate them
946 	 * if invalidation requested.  Pages undergoing I/O
947 	 * will be ignored by vm_object_page_remove().
948 	 *
949 	 * We cannot lock the vnode and then wait for paging
950 	 * to complete without deadlocking against vm_fault.
951 	 * Instead we simply call vm_object_page_remove() and
952 	 * allow it to block internally on a page-by-page
953 	 * basis when it encounters pages undergoing async
954 	 * I/O.
955 	 */
956 	if (object->type == OBJT_VNODE &&
957 	    (object->flags & OBJ_MIGHTBEDIRTY) != 0) {
958 		int vfslocked;
959 		vp = object->handle;
960 		VM_OBJECT_UNLOCK(object);
961 		(void) vn_start_write(vp, &mp, V_WAIT);
962 		vfslocked = VFS_LOCK_GIANT(vp->v_mount);
963 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
964 		flags = (syncio || invalidate) ? OBJPC_SYNC : 0;
965 		flags |= invalidate ? OBJPC_INVAL : 0;
966 		VM_OBJECT_LOCK(object);
967 		vm_object_page_clean(object,
968 		    OFF_TO_IDX(offset),
969 		    OFF_TO_IDX(offset + size + PAGE_MASK),
970 		    flags);
971 		VM_OBJECT_UNLOCK(object);
972 		VOP_UNLOCK(vp, 0);
973 		VFS_UNLOCK_GIANT(vfslocked);
974 		vn_finished_write(mp);
975 		VM_OBJECT_LOCK(object);
976 	}
977 	if ((object->type == OBJT_VNODE ||
978 	     object->type == OBJT_DEVICE) && invalidate) {
979 		boolean_t purge;
980 		purge = old_msync || (object->type == OBJT_DEVICE);
981 		vm_object_page_remove(object,
982 		    OFF_TO_IDX(offset),
983 		    OFF_TO_IDX(offset + size + PAGE_MASK),
984 		    purge ? FALSE : TRUE);
985 	}
986 	VM_OBJECT_UNLOCK(object);
987 }
988 
989 /*
990  *	vm_object_madvise:
991  *
992  *	Implements the madvise function at the object/page level.
993  *
994  *	MADV_WILLNEED	(any object)
995  *
996  *	    Activate the specified pages if they are resident.
997  *
998  *	MADV_DONTNEED	(any object)
999  *
1000  *	    Deactivate the specified pages if they are resident.
1001  *
1002  *	MADV_FREE	(OBJT_DEFAULT/OBJT_SWAP objects,
1003  *			 OBJ_ONEMAPPING only)
1004  *
1005  *	    Deactivate and clean the specified pages if they are
1006  *	    resident.  This permits the process to reuse the pages
1007  *	    without faulting or the kernel to reclaim the pages
1008  *	    without I/O.
1009  */
1010 void
1011 vm_object_madvise(vm_object_t object, vm_pindex_t pindex, int count, int advise)
1012 {
1013 	vm_pindex_t end, tpindex;
1014 	vm_object_t backing_object, tobject;
1015 	vm_page_t m;
1016 
1017 	if (object == NULL)
1018 		return;
1019 	VM_OBJECT_LOCK(object);
1020 	end = pindex + count;
1021 	/*
1022 	 * Locate and adjust resident pages
1023 	 */
1024 	for (; pindex < end; pindex += 1) {
1025 relookup:
1026 		tobject = object;
1027 		tpindex = pindex;
1028 shadowlookup:
1029 		/*
1030 		 * MADV_FREE only operates on OBJT_DEFAULT or OBJT_SWAP pages
1031 		 * and those pages must be OBJ_ONEMAPPING.
1032 		 */
1033 		if (advise == MADV_FREE) {
1034 			if ((tobject->type != OBJT_DEFAULT &&
1035 			     tobject->type != OBJT_SWAP) ||
1036 			    (tobject->flags & OBJ_ONEMAPPING) == 0) {
1037 				goto unlock_tobject;
1038 			}
1039 		} else if (tobject->type == OBJT_PHYS)
1040 			goto unlock_tobject;
1041 		m = vm_page_lookup(tobject, tpindex);
1042 		if (m == NULL && advise == MADV_WILLNEED) {
1043 			/*
1044 			 * If the page is cached, reactivate it.
1045 			 */
1046 			m = vm_page_alloc(tobject, tpindex, VM_ALLOC_IFCACHED |
1047 			    VM_ALLOC_NOBUSY);
1048 		}
1049 		if (m == NULL) {
1050 			/*
1051 			 * There may be swap even if there is no backing page
1052 			 */
1053 			if (advise == MADV_FREE && tobject->type == OBJT_SWAP)
1054 				swap_pager_freespace(tobject, tpindex, 1);
1055 			/*
1056 			 * next object
1057 			 */
1058 			backing_object = tobject->backing_object;
1059 			if (backing_object == NULL)
1060 				goto unlock_tobject;
1061 			VM_OBJECT_LOCK(backing_object);
1062 			tpindex += OFF_TO_IDX(tobject->backing_object_offset);
1063 			if (tobject != object)
1064 				VM_OBJECT_UNLOCK(tobject);
1065 			tobject = backing_object;
1066 			goto shadowlookup;
1067 		} else if (m->valid != VM_PAGE_BITS_ALL)
1068 			goto unlock_tobject;
1069 		/*
1070 		 * If the page is not in a normal state, skip it.
1071 		 */
1072 		vm_page_lock(m);
1073 		if (m->hold_count != 0 || m->wire_count != 0) {
1074 			vm_page_unlock(m);
1075 			goto unlock_tobject;
1076 		}
1077 		KASSERT((m->flags & (PG_FICTITIOUS | PG_UNMANAGED)) == 0,
1078 		    ("vm_object_madvise: page %p is not managed", m));
1079 		if ((m->oflags & VPO_BUSY) || m->busy) {
1080 			if (advise == MADV_WILLNEED) {
1081 				/*
1082 				 * Reference the page before unlocking and
1083 				 * sleeping so that the page daemon is less
1084 				 * likely to reclaim it.
1085 				 */
1086 				vm_page_lock_queues();
1087 				vm_page_flag_set(m, PG_REFERENCED);
1088 				vm_page_unlock_queues();
1089 			}
1090 			vm_page_unlock(m);
1091 			if (object != tobject)
1092 				VM_OBJECT_UNLOCK(object);
1093 			m->oflags |= VPO_WANTED;
1094 			msleep(m, VM_OBJECT_MTX(tobject), PDROP | PVM, "madvpo",
1095 			    0);
1096 			VM_OBJECT_LOCK(object);
1097   			goto relookup;
1098 		}
1099 		if (advise == MADV_WILLNEED) {
1100 			vm_page_activate(m);
1101 		} else if (advise == MADV_DONTNEED) {
1102 			vm_page_dontneed(m);
1103 		} else if (advise == MADV_FREE) {
1104 			/*
1105 			 * Mark the page clean.  This will allow the page
1106 			 * to be freed up by the system.  However, such pages
1107 			 * are often reused quickly by malloc()/free()
1108 			 * so we do not do anything that would cause
1109 			 * a page fault if we can help it.
1110 			 *
1111 			 * Specifically, we do not try to actually free
1112 			 * the page now nor do we try to put it in the
1113 			 * cache (which would cause a page fault on reuse).
1114 			 *
1115 			 * But we do make the page is freeable as we
1116 			 * can without actually taking the step of unmapping
1117 			 * it.
1118 			 */
1119 			pmap_clear_modify(m);
1120 			m->dirty = 0;
1121 			m->act_count = 0;
1122 			vm_page_dontneed(m);
1123 		}
1124 		vm_page_unlock(m);
1125 		if (advise == MADV_FREE && tobject->type == OBJT_SWAP)
1126 			swap_pager_freespace(tobject, tpindex, 1);
1127 unlock_tobject:
1128 		if (tobject != object)
1129 			VM_OBJECT_UNLOCK(tobject);
1130 	}
1131 	VM_OBJECT_UNLOCK(object);
1132 }
1133 
1134 /*
1135  *	vm_object_shadow:
1136  *
1137  *	Create a new object which is backed by the
1138  *	specified existing object range.  The source
1139  *	object reference is deallocated.
1140  *
1141  *	The new object and offset into that object
1142  *	are returned in the source parameters.
1143  */
1144 void
1145 vm_object_shadow(
1146 	vm_object_t *object,	/* IN/OUT */
1147 	vm_ooffset_t *offset,	/* IN/OUT */
1148 	vm_size_t length)
1149 {
1150 	vm_object_t source;
1151 	vm_object_t result;
1152 
1153 	source = *object;
1154 
1155 	/*
1156 	 * Don't create the new object if the old object isn't shared.
1157 	 */
1158 	if (source != NULL) {
1159 		VM_OBJECT_LOCK(source);
1160 		if (source->ref_count == 1 &&
1161 		    source->handle == NULL &&
1162 		    (source->type == OBJT_DEFAULT ||
1163 		     source->type == OBJT_SWAP)) {
1164 			VM_OBJECT_UNLOCK(source);
1165 			return;
1166 		}
1167 		VM_OBJECT_UNLOCK(source);
1168 	}
1169 
1170 	/*
1171 	 * Allocate a new object with the given length.
1172 	 */
1173 	result = vm_object_allocate(OBJT_DEFAULT, length);
1174 
1175 	/*
1176 	 * The new object shadows the source object, adding a reference to it.
1177 	 * Our caller changes his reference to point to the new object,
1178 	 * removing a reference to the source object.  Net result: no change
1179 	 * of reference count.
1180 	 *
1181 	 * Try to optimize the result object's page color when shadowing
1182 	 * in order to maintain page coloring consistency in the combined
1183 	 * shadowed object.
1184 	 */
1185 	result->backing_object = source;
1186 	/*
1187 	 * Store the offset into the source object, and fix up the offset into
1188 	 * the new object.
1189 	 */
1190 	result->backing_object_offset = *offset;
1191 	if (source != NULL) {
1192 		VM_OBJECT_LOCK(source);
1193 		LIST_INSERT_HEAD(&source->shadow_head, result, shadow_list);
1194 		source->shadow_count++;
1195 		source->generation++;
1196 #if VM_NRESERVLEVEL > 0
1197 		result->flags |= source->flags & OBJ_COLORED;
1198 		result->pg_color = (source->pg_color + OFF_TO_IDX(*offset)) &
1199 		    ((1 << (VM_NFREEORDER - 1)) - 1);
1200 #endif
1201 		VM_OBJECT_UNLOCK(source);
1202 	}
1203 
1204 
1205 	/*
1206 	 * Return the new things
1207 	 */
1208 	*offset = 0;
1209 	*object = result;
1210 }
1211 
1212 /*
1213  *	vm_object_split:
1214  *
1215  * Split the pages in a map entry into a new object.  This affords
1216  * easier removal of unused pages, and keeps object inheritance from
1217  * being a negative impact on memory usage.
1218  */
1219 void
1220 vm_object_split(vm_map_entry_t entry)
1221 {
1222 	vm_page_t m, m_next;
1223 	vm_object_t orig_object, new_object, source;
1224 	vm_pindex_t idx, offidxstart;
1225 	vm_size_t size;
1226 
1227 	orig_object = entry->object.vm_object;
1228 	if (orig_object->type != OBJT_DEFAULT && orig_object->type != OBJT_SWAP)
1229 		return;
1230 	if (orig_object->ref_count <= 1)
1231 		return;
1232 	VM_OBJECT_UNLOCK(orig_object);
1233 
1234 	offidxstart = OFF_TO_IDX(entry->offset);
1235 	size = atop(entry->end - entry->start);
1236 
1237 	/*
1238 	 * If swap_pager_copy() is later called, it will convert new_object
1239 	 * into a swap object.
1240 	 */
1241 	new_object = vm_object_allocate(OBJT_DEFAULT, size);
1242 
1243 	/*
1244 	 * At this point, the new object is still private, so the order in
1245 	 * which the original and new objects are locked does not matter.
1246 	 */
1247 	VM_OBJECT_LOCK(new_object);
1248 	VM_OBJECT_LOCK(orig_object);
1249 	source = orig_object->backing_object;
1250 	if (source != NULL) {
1251 		VM_OBJECT_LOCK(source);
1252 		if ((source->flags & OBJ_DEAD) != 0) {
1253 			VM_OBJECT_UNLOCK(source);
1254 			VM_OBJECT_UNLOCK(orig_object);
1255 			VM_OBJECT_UNLOCK(new_object);
1256 			vm_object_deallocate(new_object);
1257 			VM_OBJECT_LOCK(orig_object);
1258 			return;
1259 		}
1260 		LIST_INSERT_HEAD(&source->shadow_head,
1261 				  new_object, shadow_list);
1262 		source->shadow_count++;
1263 		source->generation++;
1264 		vm_object_reference_locked(source);	/* for new_object */
1265 		vm_object_clear_flag(source, OBJ_ONEMAPPING);
1266 		VM_OBJECT_UNLOCK(source);
1267 		new_object->backing_object_offset =
1268 			orig_object->backing_object_offset + entry->offset;
1269 		new_object->backing_object = source;
1270 	}
1271 	if (orig_object->uip != NULL) {
1272 		new_object->uip = orig_object->uip;
1273 		uihold(orig_object->uip);
1274 		new_object->charge = ptoa(size);
1275 		KASSERT(orig_object->charge >= ptoa(size),
1276 		    ("orig_object->charge < 0"));
1277 		orig_object->charge -= ptoa(size);
1278 	}
1279 retry:
1280 	m = vm_page_find_least(orig_object, offidxstart);
1281 	for (; m != NULL && (idx = m->pindex - offidxstart) < size;
1282 	    m = m_next) {
1283 		m_next = TAILQ_NEXT(m, listq);
1284 
1285 		/*
1286 		 * We must wait for pending I/O to complete before we can
1287 		 * rename the page.
1288 		 *
1289 		 * We do not have to VM_PROT_NONE the page as mappings should
1290 		 * not be changed by this operation.
1291 		 */
1292 		if ((m->oflags & VPO_BUSY) || m->busy) {
1293 			VM_OBJECT_UNLOCK(new_object);
1294 			m->oflags |= VPO_WANTED;
1295 			msleep(m, VM_OBJECT_MTX(orig_object), PVM, "spltwt", 0);
1296 			VM_OBJECT_LOCK(new_object);
1297 			goto retry;
1298 		}
1299 		vm_page_lock(m);
1300 		vm_page_rename(m, new_object, idx);
1301 		vm_page_unlock(m);
1302 		/* page automatically made dirty by rename and cache handled */
1303 		vm_page_busy(m);
1304 	}
1305 	if (orig_object->type == OBJT_SWAP) {
1306 		/*
1307 		 * swap_pager_copy() can sleep, in which case the orig_object's
1308 		 * and new_object's locks are released and reacquired.
1309 		 */
1310 		swap_pager_copy(orig_object, new_object, offidxstart, 0);
1311 
1312 		/*
1313 		 * Transfer any cached pages from orig_object to new_object.
1314 		 */
1315 		if (__predict_false(orig_object->cache != NULL))
1316 			vm_page_cache_transfer(orig_object, offidxstart,
1317 			    new_object);
1318 	}
1319 	VM_OBJECT_UNLOCK(orig_object);
1320 	TAILQ_FOREACH(m, &new_object->memq, listq)
1321 		vm_page_wakeup(m);
1322 	VM_OBJECT_UNLOCK(new_object);
1323 	entry->object.vm_object = new_object;
1324 	entry->offset = 0LL;
1325 	vm_object_deallocate(orig_object);
1326 	VM_OBJECT_LOCK(new_object);
1327 }
1328 
1329 #define	OBSC_TEST_ALL_SHADOWED	0x0001
1330 #define	OBSC_COLLAPSE_NOWAIT	0x0002
1331 #define	OBSC_COLLAPSE_WAIT	0x0004
1332 
1333 static int
1334 vm_object_backing_scan(vm_object_t object, int op)
1335 {
1336 	int r = 1;
1337 	vm_page_t p;
1338 	vm_object_t backing_object;
1339 	vm_pindex_t backing_offset_index;
1340 
1341 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
1342 	VM_OBJECT_LOCK_ASSERT(object->backing_object, MA_OWNED);
1343 
1344 	backing_object = object->backing_object;
1345 	backing_offset_index = OFF_TO_IDX(object->backing_object_offset);
1346 
1347 	/*
1348 	 * Initial conditions
1349 	 */
1350 	if (op & OBSC_TEST_ALL_SHADOWED) {
1351 		/*
1352 		 * We do not want to have to test for the existence of cache
1353 		 * or swap pages in the backing object.  XXX but with the
1354 		 * new swapper this would be pretty easy to do.
1355 		 *
1356 		 * XXX what about anonymous MAP_SHARED memory that hasn't
1357 		 * been ZFOD faulted yet?  If we do not test for this, the
1358 		 * shadow test may succeed! XXX
1359 		 */
1360 		if (backing_object->type != OBJT_DEFAULT) {
1361 			return (0);
1362 		}
1363 	}
1364 	if (op & OBSC_COLLAPSE_WAIT) {
1365 		vm_object_set_flag(backing_object, OBJ_DEAD);
1366 	}
1367 
1368 	/*
1369 	 * Our scan
1370 	 */
1371 	p = TAILQ_FIRST(&backing_object->memq);
1372 	while (p) {
1373 		vm_page_t next = TAILQ_NEXT(p, listq);
1374 		vm_pindex_t new_pindex = p->pindex - backing_offset_index;
1375 
1376 		if (op & OBSC_TEST_ALL_SHADOWED) {
1377 			vm_page_t pp;
1378 
1379 			/*
1380 			 * Ignore pages outside the parent object's range
1381 			 * and outside the parent object's mapping of the
1382 			 * backing object.
1383 			 *
1384 			 * note that we do not busy the backing object's
1385 			 * page.
1386 			 */
1387 			if (
1388 			    p->pindex < backing_offset_index ||
1389 			    new_pindex >= object->size
1390 			) {
1391 				p = next;
1392 				continue;
1393 			}
1394 
1395 			/*
1396 			 * See if the parent has the page or if the parent's
1397 			 * object pager has the page.  If the parent has the
1398 			 * page but the page is not valid, the parent's
1399 			 * object pager must have the page.
1400 			 *
1401 			 * If this fails, the parent does not completely shadow
1402 			 * the object and we might as well give up now.
1403 			 */
1404 
1405 			pp = vm_page_lookup(object, new_pindex);
1406 			if (
1407 			    (pp == NULL || pp->valid == 0) &&
1408 			    !vm_pager_has_page(object, new_pindex, NULL, NULL)
1409 			) {
1410 				r = 0;
1411 				break;
1412 			}
1413 		}
1414 
1415 		/*
1416 		 * Check for busy page
1417 		 */
1418 		if (op & (OBSC_COLLAPSE_WAIT | OBSC_COLLAPSE_NOWAIT)) {
1419 			vm_page_t pp;
1420 
1421 			if (op & OBSC_COLLAPSE_NOWAIT) {
1422 				if ((p->oflags & VPO_BUSY) ||
1423 				    !p->valid ||
1424 				    p->busy) {
1425 					p = next;
1426 					continue;
1427 				}
1428 			} else if (op & OBSC_COLLAPSE_WAIT) {
1429 				if ((p->oflags & VPO_BUSY) || p->busy) {
1430 					VM_OBJECT_UNLOCK(object);
1431 					p->oflags |= VPO_WANTED;
1432 					msleep(p, VM_OBJECT_MTX(backing_object),
1433 					    PDROP | PVM, "vmocol", 0);
1434 					VM_OBJECT_LOCK(object);
1435 					VM_OBJECT_LOCK(backing_object);
1436 					/*
1437 					 * If we slept, anything could have
1438 					 * happened.  Since the object is
1439 					 * marked dead, the backing offset
1440 					 * should not have changed so we
1441 					 * just restart our scan.
1442 					 */
1443 					p = TAILQ_FIRST(&backing_object->memq);
1444 					continue;
1445 				}
1446 			}
1447 
1448 			KASSERT(
1449 			    p->object == backing_object,
1450 			    ("vm_object_backing_scan: object mismatch")
1451 			);
1452 
1453 			/*
1454 			 * Destroy any associated swap
1455 			 */
1456 			if (backing_object->type == OBJT_SWAP) {
1457 				swap_pager_freespace(
1458 				    backing_object,
1459 				    p->pindex,
1460 				    1
1461 				);
1462 			}
1463 
1464 			if (
1465 			    p->pindex < backing_offset_index ||
1466 			    new_pindex >= object->size
1467 			) {
1468 				/*
1469 				 * Page is out of the parent object's range, we
1470 				 * can simply destroy it.
1471 				 */
1472 				vm_page_lock(p);
1473 				KASSERT(!pmap_page_is_mapped(p),
1474 				    ("freeing mapped page %p", p));
1475 				if (p->wire_count == 0)
1476 					vm_page_free(p);
1477 				else
1478 					vm_page_remove(p);
1479 				vm_page_unlock(p);
1480 				p = next;
1481 				continue;
1482 			}
1483 
1484 			pp = vm_page_lookup(object, new_pindex);
1485 			if (
1486 			    pp != NULL ||
1487 			    vm_pager_has_page(object, new_pindex, NULL, NULL)
1488 			) {
1489 				/*
1490 				 * page already exists in parent OR swap exists
1491 				 * for this location in the parent.  Destroy
1492 				 * the original page from the backing object.
1493 				 *
1494 				 * Leave the parent's page alone
1495 				 */
1496 				vm_page_lock(p);
1497 				KASSERT(!pmap_page_is_mapped(p),
1498 				    ("freeing mapped page %p", p));
1499 				if (p->wire_count == 0)
1500 					vm_page_free(p);
1501 				else
1502 					vm_page_remove(p);
1503 				vm_page_unlock(p);
1504 				p = next;
1505 				continue;
1506 			}
1507 
1508 #if VM_NRESERVLEVEL > 0
1509 			/*
1510 			 * Rename the reservation.
1511 			 */
1512 			vm_reserv_rename(p, object, backing_object,
1513 			    backing_offset_index);
1514 #endif
1515 
1516 			/*
1517 			 * Page does not exist in parent, rename the
1518 			 * page from the backing object to the main object.
1519 			 *
1520 			 * If the page was mapped to a process, it can remain
1521 			 * mapped through the rename.
1522 			 */
1523 			vm_page_lock(p);
1524 			vm_page_rename(p, object, new_pindex);
1525 			vm_page_unlock(p);
1526 			/* page automatically made dirty by rename */
1527 		}
1528 		p = next;
1529 	}
1530 	return (r);
1531 }
1532 
1533 
1534 /*
1535  * this version of collapse allows the operation to occur earlier and
1536  * when paging_in_progress is true for an object...  This is not a complete
1537  * operation, but should plug 99.9% of the rest of the leaks.
1538  */
1539 static void
1540 vm_object_qcollapse(vm_object_t object)
1541 {
1542 	vm_object_t backing_object = object->backing_object;
1543 
1544 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
1545 	VM_OBJECT_LOCK_ASSERT(backing_object, MA_OWNED);
1546 
1547 	if (backing_object->ref_count != 1)
1548 		return;
1549 
1550 	vm_object_backing_scan(object, OBSC_COLLAPSE_NOWAIT);
1551 }
1552 
1553 /*
1554  *	vm_object_collapse:
1555  *
1556  *	Collapse an object with the object backing it.
1557  *	Pages in the backing object are moved into the
1558  *	parent, and the backing object is deallocated.
1559  */
1560 void
1561 vm_object_collapse(vm_object_t object)
1562 {
1563 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
1564 
1565 	while (TRUE) {
1566 		vm_object_t backing_object;
1567 
1568 		/*
1569 		 * Verify that the conditions are right for collapse:
1570 		 *
1571 		 * The object exists and the backing object exists.
1572 		 */
1573 		if ((backing_object = object->backing_object) == NULL)
1574 			break;
1575 
1576 		/*
1577 		 * we check the backing object first, because it is most likely
1578 		 * not collapsable.
1579 		 */
1580 		VM_OBJECT_LOCK(backing_object);
1581 		if (backing_object->handle != NULL ||
1582 		    (backing_object->type != OBJT_DEFAULT &&
1583 		     backing_object->type != OBJT_SWAP) ||
1584 		    (backing_object->flags & OBJ_DEAD) ||
1585 		    object->handle != NULL ||
1586 		    (object->type != OBJT_DEFAULT &&
1587 		     object->type != OBJT_SWAP) ||
1588 		    (object->flags & OBJ_DEAD)) {
1589 			VM_OBJECT_UNLOCK(backing_object);
1590 			break;
1591 		}
1592 
1593 		if (
1594 		    object->paging_in_progress != 0 ||
1595 		    backing_object->paging_in_progress != 0
1596 		) {
1597 			vm_object_qcollapse(object);
1598 			VM_OBJECT_UNLOCK(backing_object);
1599 			break;
1600 		}
1601 		/*
1602 		 * We know that we can either collapse the backing object (if
1603 		 * the parent is the only reference to it) or (perhaps) have
1604 		 * the parent bypass the object if the parent happens to shadow
1605 		 * all the resident pages in the entire backing object.
1606 		 *
1607 		 * This is ignoring pager-backed pages such as swap pages.
1608 		 * vm_object_backing_scan fails the shadowing test in this
1609 		 * case.
1610 		 */
1611 		if (backing_object->ref_count == 1) {
1612 			/*
1613 			 * If there is exactly one reference to the backing
1614 			 * object, we can collapse it into the parent.
1615 			 */
1616 			vm_object_backing_scan(object, OBSC_COLLAPSE_WAIT);
1617 
1618 #if VM_NRESERVLEVEL > 0
1619 			/*
1620 			 * Break any reservations from backing_object.
1621 			 */
1622 			if (__predict_false(!LIST_EMPTY(&backing_object->rvq)))
1623 				vm_reserv_break_all(backing_object);
1624 #endif
1625 
1626 			/*
1627 			 * Move the pager from backing_object to object.
1628 			 */
1629 			if (backing_object->type == OBJT_SWAP) {
1630 				/*
1631 				 * swap_pager_copy() can sleep, in which case
1632 				 * the backing_object's and object's locks are
1633 				 * released and reacquired.
1634 				 */
1635 				swap_pager_copy(
1636 				    backing_object,
1637 				    object,
1638 				    OFF_TO_IDX(object->backing_object_offset), TRUE);
1639 
1640 				/*
1641 				 * Free any cached pages from backing_object.
1642 				 */
1643 				if (__predict_false(backing_object->cache != NULL))
1644 					vm_page_cache_free(backing_object, 0, 0);
1645 			}
1646 			/*
1647 			 * Object now shadows whatever backing_object did.
1648 			 * Note that the reference to
1649 			 * backing_object->backing_object moves from within
1650 			 * backing_object to within object.
1651 			 */
1652 			LIST_REMOVE(object, shadow_list);
1653 			backing_object->shadow_count--;
1654 			backing_object->generation++;
1655 			if (backing_object->backing_object) {
1656 				VM_OBJECT_LOCK(backing_object->backing_object);
1657 				LIST_REMOVE(backing_object, shadow_list);
1658 				LIST_INSERT_HEAD(
1659 				    &backing_object->backing_object->shadow_head,
1660 				    object, shadow_list);
1661 				/*
1662 				 * The shadow_count has not changed.
1663 				 */
1664 				backing_object->backing_object->generation++;
1665 				VM_OBJECT_UNLOCK(backing_object->backing_object);
1666 			}
1667 			object->backing_object = backing_object->backing_object;
1668 			object->backing_object_offset +=
1669 			    backing_object->backing_object_offset;
1670 
1671 			/*
1672 			 * Discard backing_object.
1673 			 *
1674 			 * Since the backing object has no pages, no pager left,
1675 			 * and no object references within it, all that is
1676 			 * necessary is to dispose of it.
1677 			 */
1678 			KASSERT(backing_object->ref_count == 1, (
1679 "backing_object %p was somehow re-referenced during collapse!",
1680 			    backing_object));
1681 			VM_OBJECT_UNLOCK(backing_object);
1682 			vm_object_destroy(backing_object);
1683 
1684 			object_collapses++;
1685 		} else {
1686 			vm_object_t new_backing_object;
1687 
1688 			/*
1689 			 * If we do not entirely shadow the backing object,
1690 			 * there is nothing we can do so we give up.
1691 			 */
1692 			if (object->resident_page_count != object->size &&
1693 			    vm_object_backing_scan(object,
1694 			    OBSC_TEST_ALL_SHADOWED) == 0) {
1695 				VM_OBJECT_UNLOCK(backing_object);
1696 				break;
1697 			}
1698 
1699 			/*
1700 			 * Make the parent shadow the next object in the
1701 			 * chain.  Deallocating backing_object will not remove
1702 			 * it, since its reference count is at least 2.
1703 			 */
1704 			LIST_REMOVE(object, shadow_list);
1705 			backing_object->shadow_count--;
1706 			backing_object->generation++;
1707 
1708 			new_backing_object = backing_object->backing_object;
1709 			if ((object->backing_object = new_backing_object) != NULL) {
1710 				VM_OBJECT_LOCK(new_backing_object);
1711 				LIST_INSERT_HEAD(
1712 				    &new_backing_object->shadow_head,
1713 				    object,
1714 				    shadow_list
1715 				);
1716 				new_backing_object->shadow_count++;
1717 				new_backing_object->generation++;
1718 				vm_object_reference_locked(new_backing_object);
1719 				VM_OBJECT_UNLOCK(new_backing_object);
1720 				object->backing_object_offset +=
1721 					backing_object->backing_object_offset;
1722 			}
1723 
1724 			/*
1725 			 * Drop the reference count on backing_object. Since
1726 			 * its ref_count was at least 2, it will not vanish.
1727 			 */
1728 			backing_object->ref_count--;
1729 			VM_OBJECT_UNLOCK(backing_object);
1730 			object_bypasses++;
1731 		}
1732 
1733 		/*
1734 		 * Try again with this object's new backing object.
1735 		 */
1736 	}
1737 }
1738 
1739 /*
1740  *	vm_object_page_remove:
1741  *
1742  *	For the given object, either frees or invalidates each of the
1743  *	specified pages.  In general, a page is freed.  However, if a
1744  *	page is wired for any reason other than the existence of a
1745  *	managed, wired mapping, then it may be invalidated but not
1746  *	removed from the object.  Pages are specified by the given
1747  *	range ["start", "end") and Boolean "clean_only".  As a
1748  *	special case, if "end" is zero, then the range extends from
1749  *	"start" to the end of the object.  If "clean_only" is TRUE,
1750  *	then only the non-dirty pages within the specified range are
1751  *	affected.
1752  *
1753  *	In general, this operation should only be performed on objects
1754  *	that contain managed pages.  There are two exceptions.  First,
1755  *	it may be performed on the kernel and kmem objects.  Second,
1756  *	it may be used by msync(..., MS_INVALIDATE) to invalidate
1757  *	device-backed pages.  In both of these cases, "clean_only"
1758  *	must be FALSE.
1759  *
1760  *	The object must be locked.
1761  */
1762 void
1763 vm_object_page_remove(vm_object_t object, vm_pindex_t start, vm_pindex_t end,
1764     boolean_t clean_only)
1765 {
1766 	vm_page_t p, next;
1767 	int wirings;
1768 
1769 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
1770 	if (object->resident_page_count == 0)
1771 		goto skipmemq;
1772 
1773 	/*
1774 	 * Since physically-backed objects do not use managed pages, we can't
1775 	 * remove pages from the object (we must instead remove the page
1776 	 * references, and then destroy the object).
1777 	 */
1778 	KASSERT(object->type != OBJT_PHYS || object == kernel_object ||
1779 	    object == kmem_object,
1780 	    ("attempt to remove pages from a physical object"));
1781 
1782 	vm_object_pip_add(object, 1);
1783 again:
1784 	p = vm_page_find_least(object, start);
1785 
1786 	/*
1787 	 * Assert: the variable p is either (1) the page with the
1788 	 * least pindex greater than or equal to the parameter pindex
1789 	 * or (2) NULL.
1790 	 */
1791 	for (;
1792 	     p != NULL && (p->pindex < end || end == 0);
1793 	     p = next) {
1794 		next = TAILQ_NEXT(p, listq);
1795 
1796 		/*
1797 		 * If the page is wired for any reason besides the
1798 		 * existence of managed, wired mappings, then it cannot
1799 		 * be freed.  For example, fictitious pages, which
1800 		 * represent device memory, are inherently wired and
1801 		 * cannot be freed.  They can, however, be invalidated
1802 		 * if "clean_only" is FALSE.
1803 		 */
1804 		vm_page_lock(p);
1805 		if ((wirings = p->wire_count) != 0 &&
1806 		    (wirings = pmap_page_wired_mappings(p)) != p->wire_count) {
1807 			/* Fictitious pages do not have managed mappings. */
1808 			if ((p->flags & PG_FICTITIOUS) == 0)
1809 				pmap_remove_all(p);
1810 			/* Account for removal of managed, wired mappings. */
1811 			p->wire_count -= wirings;
1812 			if (!clean_only) {
1813 				p->valid = 0;
1814 				vm_page_undirty(p);
1815 			}
1816 			vm_page_unlock(p);
1817 			continue;
1818 		}
1819 		if (vm_page_sleep_if_busy(p, TRUE, "vmopar"))
1820 			goto again;
1821 		KASSERT((p->flags & PG_FICTITIOUS) == 0,
1822 		    ("vm_object_page_remove: page %p is fictitious", p));
1823 		if (clean_only && p->valid) {
1824 			pmap_remove_write(p);
1825 			if (p->dirty) {
1826 				vm_page_unlock(p);
1827 				continue;
1828 			}
1829 		}
1830 		pmap_remove_all(p);
1831 		/* Account for removal of managed, wired mappings. */
1832 		if (wirings != 0)
1833 			p->wire_count -= wirings;
1834 		vm_page_free(p);
1835 		vm_page_unlock(p);
1836 	}
1837 	vm_object_pip_wakeup(object);
1838 skipmemq:
1839 	if (__predict_false(object->cache != NULL))
1840 		vm_page_cache_free(object, start, end);
1841 }
1842 
1843 /*
1844  *	Populate the specified range of the object with valid pages.  Returns
1845  *	TRUE if the range is successfully populated and FALSE otherwise.
1846  *
1847  *	Note: This function should be optimized to pass a larger array of
1848  *	pages to vm_pager_get_pages() before it is applied to a non-
1849  *	OBJT_DEVICE object.
1850  *
1851  *	The object must be locked.
1852  */
1853 boolean_t
1854 vm_object_populate(vm_object_t object, vm_pindex_t start, vm_pindex_t end)
1855 {
1856 	vm_page_t m, ma[1];
1857 	vm_pindex_t pindex;
1858 	int rv;
1859 
1860 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
1861 	for (pindex = start; pindex < end; pindex++) {
1862 		m = vm_page_grab(object, pindex, VM_ALLOC_NORMAL |
1863 		    VM_ALLOC_RETRY);
1864 		if (m->valid != VM_PAGE_BITS_ALL) {
1865 			ma[0] = m;
1866 			rv = vm_pager_get_pages(object, ma, 1, 0);
1867 			m = vm_page_lookup(object, pindex);
1868 			if (m == NULL)
1869 				break;
1870 			if (rv != VM_PAGER_OK) {
1871 				vm_page_lock(m);
1872 				vm_page_free(m);
1873 				vm_page_unlock(m);
1874 				break;
1875 			}
1876 		}
1877 		/*
1878 		 * Keep "m" busy because a subsequent iteration may unlock
1879 		 * the object.
1880 		 */
1881 	}
1882 	if (pindex > start) {
1883 		m = vm_page_lookup(object, start);
1884 		while (m != NULL && m->pindex < pindex) {
1885 			vm_page_wakeup(m);
1886 			m = TAILQ_NEXT(m, listq);
1887 		}
1888 	}
1889 	return (pindex == end);
1890 }
1891 
1892 /*
1893  *	Routine:	vm_object_coalesce
1894  *	Function:	Coalesces two objects backing up adjoining
1895  *			regions of memory into a single object.
1896  *
1897  *	returns TRUE if objects were combined.
1898  *
1899  *	NOTE:	Only works at the moment if the second object is NULL -
1900  *		if it's not, which object do we lock first?
1901  *
1902  *	Parameters:
1903  *		prev_object	First object to coalesce
1904  *		prev_offset	Offset into prev_object
1905  *		prev_size	Size of reference to prev_object
1906  *		next_size	Size of reference to the second object
1907  *		reserved	Indicator that extension region has
1908  *				swap accounted for
1909  *
1910  *	Conditions:
1911  *	The object must *not* be locked.
1912  */
1913 boolean_t
1914 vm_object_coalesce(vm_object_t prev_object, vm_ooffset_t prev_offset,
1915     vm_size_t prev_size, vm_size_t next_size, boolean_t reserved)
1916 {
1917 	vm_pindex_t next_pindex;
1918 
1919 	if (prev_object == NULL)
1920 		return (TRUE);
1921 	VM_OBJECT_LOCK(prev_object);
1922 	if (prev_object->type != OBJT_DEFAULT &&
1923 	    prev_object->type != OBJT_SWAP) {
1924 		VM_OBJECT_UNLOCK(prev_object);
1925 		return (FALSE);
1926 	}
1927 
1928 	/*
1929 	 * Try to collapse the object first
1930 	 */
1931 	vm_object_collapse(prev_object);
1932 
1933 	/*
1934 	 * Can't coalesce if: . more than one reference . paged out . shadows
1935 	 * another object . has a copy elsewhere (any of which mean that the
1936 	 * pages not mapped to prev_entry may be in use anyway)
1937 	 */
1938 	if (prev_object->backing_object != NULL) {
1939 		VM_OBJECT_UNLOCK(prev_object);
1940 		return (FALSE);
1941 	}
1942 
1943 	prev_size >>= PAGE_SHIFT;
1944 	next_size >>= PAGE_SHIFT;
1945 	next_pindex = OFF_TO_IDX(prev_offset) + prev_size;
1946 
1947 	if ((prev_object->ref_count > 1) &&
1948 	    (prev_object->size != next_pindex)) {
1949 		VM_OBJECT_UNLOCK(prev_object);
1950 		return (FALSE);
1951 	}
1952 
1953 	/*
1954 	 * Account for the charge.
1955 	 */
1956 	if (prev_object->uip != NULL) {
1957 
1958 		/*
1959 		 * If prev_object was charged, then this mapping,
1960 		 * althought not charged now, may become writable
1961 		 * later. Non-NULL uip in the object would prevent
1962 		 * swap reservation during enabling of the write
1963 		 * access, so reserve swap now. Failed reservation
1964 		 * cause allocation of the separate object for the map
1965 		 * entry, and swap reservation for this entry is
1966 		 * managed in appropriate time.
1967 		 */
1968 		if (!reserved && !swap_reserve_by_uid(ptoa(next_size),
1969 		    prev_object->uip)) {
1970 			return (FALSE);
1971 		}
1972 		prev_object->charge += ptoa(next_size);
1973 	}
1974 
1975 	/*
1976 	 * Remove any pages that may still be in the object from a previous
1977 	 * deallocation.
1978 	 */
1979 	if (next_pindex < prev_object->size) {
1980 		vm_object_page_remove(prev_object,
1981 				      next_pindex,
1982 				      next_pindex + next_size, FALSE);
1983 		if (prev_object->type == OBJT_SWAP)
1984 			swap_pager_freespace(prev_object,
1985 					     next_pindex, next_size);
1986 #if 0
1987 		if (prev_object->uip != NULL) {
1988 			KASSERT(prev_object->charge >=
1989 			    ptoa(prev_object->size - next_pindex),
1990 			    ("object %p overcharged 1 %jx %jx", prev_object,
1991 				(uintmax_t)next_pindex, (uintmax_t)next_size));
1992 			prev_object->charge -= ptoa(prev_object->size -
1993 			    next_pindex);
1994 		}
1995 #endif
1996 	}
1997 
1998 	/*
1999 	 * Extend the object if necessary.
2000 	 */
2001 	if (next_pindex + next_size > prev_object->size)
2002 		prev_object->size = next_pindex + next_size;
2003 
2004 	VM_OBJECT_UNLOCK(prev_object);
2005 	return (TRUE);
2006 }
2007 
2008 void
2009 vm_object_set_writeable_dirty(vm_object_t object)
2010 {
2011 
2012 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
2013 	if (object->type != OBJT_VNODE ||
2014 	    (object->flags & OBJ_MIGHTBEDIRTY) != 0)
2015 		return;
2016 	vm_object_set_flag(object, OBJ_MIGHTBEDIRTY);
2017 }
2018 
2019 #include "opt_ddb.h"
2020 #ifdef DDB
2021 #include <sys/kernel.h>
2022 
2023 #include <sys/cons.h>
2024 
2025 #include <ddb/ddb.h>
2026 
2027 static int
2028 _vm_object_in_map(vm_map_t map, vm_object_t object, vm_map_entry_t entry)
2029 {
2030 	vm_map_t tmpm;
2031 	vm_map_entry_t tmpe;
2032 	vm_object_t obj;
2033 	int entcount;
2034 
2035 	if (map == 0)
2036 		return 0;
2037 
2038 	if (entry == 0) {
2039 		tmpe = map->header.next;
2040 		entcount = map->nentries;
2041 		while (entcount-- && (tmpe != &map->header)) {
2042 			if (_vm_object_in_map(map, object, tmpe)) {
2043 				return 1;
2044 			}
2045 			tmpe = tmpe->next;
2046 		}
2047 	} else if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
2048 		tmpm = entry->object.sub_map;
2049 		tmpe = tmpm->header.next;
2050 		entcount = tmpm->nentries;
2051 		while (entcount-- && tmpe != &tmpm->header) {
2052 			if (_vm_object_in_map(tmpm, object, tmpe)) {
2053 				return 1;
2054 			}
2055 			tmpe = tmpe->next;
2056 		}
2057 	} else if ((obj = entry->object.vm_object) != NULL) {
2058 		for (; obj; obj = obj->backing_object)
2059 			if (obj == object) {
2060 				return 1;
2061 			}
2062 	}
2063 	return 0;
2064 }
2065 
2066 static int
2067 vm_object_in_map(vm_object_t object)
2068 {
2069 	struct proc *p;
2070 
2071 	/* sx_slock(&allproc_lock); */
2072 	FOREACH_PROC_IN_SYSTEM(p) {
2073 		if (!p->p_vmspace /* || (p->p_flag & (P_SYSTEM|P_WEXIT)) */)
2074 			continue;
2075 		if (_vm_object_in_map(&p->p_vmspace->vm_map, object, 0)) {
2076 			/* sx_sunlock(&allproc_lock); */
2077 			return 1;
2078 		}
2079 	}
2080 	/* sx_sunlock(&allproc_lock); */
2081 	if (_vm_object_in_map(kernel_map, object, 0))
2082 		return 1;
2083 	if (_vm_object_in_map(kmem_map, object, 0))
2084 		return 1;
2085 	if (_vm_object_in_map(pager_map, object, 0))
2086 		return 1;
2087 	if (_vm_object_in_map(buffer_map, object, 0))
2088 		return 1;
2089 	return 0;
2090 }
2091 
2092 DB_SHOW_COMMAND(vmochk, vm_object_check)
2093 {
2094 	vm_object_t object;
2095 
2096 	/*
2097 	 * make sure that internal objs are in a map somewhere
2098 	 * and none have zero ref counts.
2099 	 */
2100 	TAILQ_FOREACH(object, &vm_object_list, object_list) {
2101 		if (object->handle == NULL &&
2102 		    (object->type == OBJT_DEFAULT || object->type == OBJT_SWAP)) {
2103 			if (object->ref_count == 0) {
2104 				db_printf("vmochk: internal obj has zero ref count: %ld\n",
2105 					(long)object->size);
2106 			}
2107 			if (!vm_object_in_map(object)) {
2108 				db_printf(
2109 			"vmochk: internal obj is not in a map: "
2110 			"ref: %d, size: %lu: 0x%lx, backing_object: %p\n",
2111 				    object->ref_count, (u_long)object->size,
2112 				    (u_long)object->size,
2113 				    (void *)object->backing_object);
2114 			}
2115 		}
2116 	}
2117 }
2118 
2119 /*
2120  *	vm_object_print:	[ debug ]
2121  */
2122 DB_SHOW_COMMAND(object, vm_object_print_static)
2123 {
2124 	/* XXX convert args. */
2125 	vm_object_t object = (vm_object_t)addr;
2126 	boolean_t full = have_addr;
2127 
2128 	vm_page_t p;
2129 
2130 	/* XXX count is an (unused) arg.  Avoid shadowing it. */
2131 #define	count	was_count
2132 
2133 	int count;
2134 
2135 	if (object == NULL)
2136 		return;
2137 
2138 	db_iprintf(
2139 	    "Object %p: type=%d, size=0x%jx, res=%d, ref=%d, flags=0x%x uip %d charge %jx\n",
2140 	    object, (int)object->type, (uintmax_t)object->size,
2141 	    object->resident_page_count, object->ref_count, object->flags,
2142 	    object->uip ? object->uip->ui_uid : -1, (uintmax_t)object->charge);
2143 	db_iprintf(" sref=%d, backing_object(%d)=(%p)+0x%jx\n",
2144 	    object->shadow_count,
2145 	    object->backing_object ? object->backing_object->ref_count : 0,
2146 	    object->backing_object, (uintmax_t)object->backing_object_offset);
2147 
2148 	if (!full)
2149 		return;
2150 
2151 	db_indent += 2;
2152 	count = 0;
2153 	TAILQ_FOREACH(p, &object->memq, listq) {
2154 		if (count == 0)
2155 			db_iprintf("memory:=");
2156 		else if (count == 6) {
2157 			db_printf("\n");
2158 			db_iprintf(" ...");
2159 			count = 0;
2160 		} else
2161 			db_printf(",");
2162 		count++;
2163 
2164 		db_printf("(off=0x%jx,page=0x%jx)",
2165 		    (uintmax_t)p->pindex, (uintmax_t)VM_PAGE_TO_PHYS(p));
2166 	}
2167 	if (count != 0)
2168 		db_printf("\n");
2169 	db_indent -= 2;
2170 }
2171 
2172 /* XXX. */
2173 #undef count
2174 
2175 /* XXX need this non-static entry for calling from vm_map_print. */
2176 void
2177 vm_object_print(
2178         /* db_expr_t */ long addr,
2179 	boolean_t have_addr,
2180 	/* db_expr_t */ long count,
2181 	char *modif)
2182 {
2183 	vm_object_print_static(addr, have_addr, count, modif);
2184 }
2185 
2186 DB_SHOW_COMMAND(vmopag, vm_object_print_pages)
2187 {
2188 	vm_object_t object;
2189 	vm_pindex_t fidx;
2190 	vm_paddr_t pa;
2191 	vm_page_t m, prev_m;
2192 	int rcount, nl, c;
2193 
2194 	nl = 0;
2195 	TAILQ_FOREACH(object, &vm_object_list, object_list) {
2196 		db_printf("new object: %p\n", (void *)object);
2197 		if (nl > 18) {
2198 			c = cngetc();
2199 			if (c != ' ')
2200 				return;
2201 			nl = 0;
2202 		}
2203 		nl++;
2204 		rcount = 0;
2205 		fidx = 0;
2206 		pa = -1;
2207 		TAILQ_FOREACH(m, &object->memq, listq) {
2208 			if (m->pindex > 128)
2209 				break;
2210 			if ((prev_m = TAILQ_PREV(m, pglist, listq)) != NULL &&
2211 			    prev_m->pindex + 1 != m->pindex) {
2212 				if (rcount) {
2213 					db_printf(" index(%ld)run(%d)pa(0x%lx)\n",
2214 						(long)fidx, rcount, (long)pa);
2215 					if (nl > 18) {
2216 						c = cngetc();
2217 						if (c != ' ')
2218 							return;
2219 						nl = 0;
2220 					}
2221 					nl++;
2222 					rcount = 0;
2223 				}
2224 			}
2225 			if (rcount &&
2226 				(VM_PAGE_TO_PHYS(m) == pa + rcount * PAGE_SIZE)) {
2227 				++rcount;
2228 				continue;
2229 			}
2230 			if (rcount) {
2231 				db_printf(" index(%ld)run(%d)pa(0x%lx)\n",
2232 					(long)fidx, rcount, (long)pa);
2233 				if (nl > 18) {
2234 					c = cngetc();
2235 					if (c != ' ')
2236 						return;
2237 					nl = 0;
2238 				}
2239 				nl++;
2240 			}
2241 			fidx = m->pindex;
2242 			pa = VM_PAGE_TO_PHYS(m);
2243 			rcount = 1;
2244 		}
2245 		if (rcount) {
2246 			db_printf(" index(%ld)run(%d)pa(0x%lx)\n",
2247 				(long)fidx, rcount, (long)pa);
2248 			if (nl > 18) {
2249 				c = cngetc();
2250 				if (c != ' ')
2251 					return;
2252 				nl = 0;
2253 			}
2254 			nl++;
2255 		}
2256 	}
2257 }
2258 #endif /* DDB */
2259