xref: /freebsd/sys/vm/vm_object.c (revision 4cf49a43559ed9fdad601bdcccd2c55963008675)
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  * 3. All advertising materials mentioning features or use of this software
17  *    must display the following acknowledgement:
18  *	This product includes software developed by the University of
19  *	California, Berkeley and its contributors.
20  * 4. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	from: @(#)vm_object.c	8.5 (Berkeley) 3/22/94
37  *
38  *
39  * Copyright (c) 1987, 1990 Carnegie-Mellon University.
40  * All rights reserved.
41  *
42  * Authors: Avadis Tevanian, Jr., Michael Wayne Young
43  *
44  * Permission to use, copy, modify and distribute this software and
45  * its documentation is hereby granted, provided that both the copyright
46  * notice and this permission notice appear in all copies of the
47  * software, derivative works or modified versions, and any portions
48  * thereof, and that both notices appear in supporting documentation.
49  *
50  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
51  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
52  * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
53  *
54  * Carnegie Mellon requests users of this software to return to
55  *
56  *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
57  *  School of Computer Science
58  *  Carnegie Mellon University
59  *  Pittsburgh PA 15213-3890
60  *
61  * any improvements or extensions that they make and grant Carnegie the
62  * rights to redistribute these changes.
63  *
64  * $FreeBSD$
65  */
66 
67 /*
68  *	Virtual memory object module.
69  */
70 
71 #include <sys/param.h>
72 #include <sys/systm.h>
73 #include <sys/proc.h>		/* for curproc, pageproc */
74 #include <sys/vnode.h>
75 #include <sys/vmmeter.h>
76 #include <sys/mman.h>
77 #include <sys/mount.h>
78 
79 #include <vm/vm.h>
80 #include <vm/vm_param.h>
81 #include <vm/vm_prot.h>
82 #include <vm/pmap.h>
83 #include <vm/vm_map.h>
84 #include <vm/vm_object.h>
85 #include <vm/vm_page.h>
86 #include <vm/vm_pageout.h>
87 #include <vm/vm_pager.h>
88 #include <vm/swap_pager.h>
89 #include <vm/vm_kern.h>
90 #include <vm/vm_extern.h>
91 #include <vm/vm_zone.h>
92 
93 static void	vm_object_qcollapse __P((vm_object_t object));
94 
95 /*
96  *	Virtual memory objects maintain the actual data
97  *	associated with allocated virtual memory.  A given
98  *	page of memory exists within exactly one object.
99  *
100  *	An object is only deallocated when all "references"
101  *	are given up.  Only one "reference" to a given
102  *	region of an object should be writeable.
103  *
104  *	Associated with each object is a list of all resident
105  *	memory pages belonging to that object; this list is
106  *	maintained by the "vm_page" module, and locked by the object's
107  *	lock.
108  *
109  *	Each object also records a "pager" routine which is
110  *	used to retrieve (and store) pages to the proper backing
111  *	storage.  In addition, objects may be backed by other
112  *	objects from which they were virtual-copied.
113  *
114  *	The only items within the object structure which are
115  *	modified after time of creation are:
116  *		reference count		locked by object's lock
117  *		pager routine		locked by object's lock
118  *
119  */
120 
121 struct object_q vm_object_list;
122 #ifndef NULL_SIMPLELOCKS
123 static struct simplelock vm_object_list_lock;
124 #endif
125 static long vm_object_count;		/* count of all objects */
126 vm_object_t kernel_object;
127 vm_object_t kmem_object;
128 static struct vm_object kernel_object_store;
129 static struct vm_object kmem_object_store;
130 extern int vm_pageout_page_count;
131 
132 static long object_collapses;
133 static long object_bypasses;
134 static int next_index;
135 static vm_zone_t obj_zone;
136 static struct vm_zone obj_zone_store;
137 static int object_hash_rand;
138 #define VM_OBJECTS_INIT 256
139 static struct vm_object vm_objects_init[VM_OBJECTS_INIT];
140 
141 void
142 _vm_object_allocate(type, size, object)
143 	objtype_t type;
144 	vm_size_t size;
145 	vm_object_t object;
146 {
147 	int incr;
148 	TAILQ_INIT(&object->memq);
149 	TAILQ_INIT(&object->shadow_head);
150 
151 	object->type = type;
152 	object->size = size;
153 	object->ref_count = 1;
154 	object->flags = 0;
155 	if ((object->type == OBJT_DEFAULT) || (object->type == OBJT_SWAP))
156 		vm_object_set_flag(object, OBJ_ONEMAPPING);
157 	object->paging_in_progress = 0;
158 	object->resident_page_count = 0;
159 	object->shadow_count = 0;
160 	object->pg_color = next_index;
161 	if ( size > (PQ_L2_SIZE / 3 + PQ_PRIME1))
162 		incr = PQ_L2_SIZE / 3 + PQ_PRIME1;
163 	else
164 		incr = size;
165 	next_index = (next_index + incr) & PQ_L2_MASK;
166 	object->handle = NULL;
167 	object->backing_object = NULL;
168 	object->backing_object_offset = (vm_ooffset_t) 0;
169 	/*
170 	 * Try to generate a number that will spread objects out in the
171 	 * hash table.  We 'wipe' new objects across the hash in 128 page
172 	 * increments plus 1 more to offset it a little more by the time
173 	 * it wraps around.
174 	 */
175 	object->hash_rand = object_hash_rand - 129;
176 
177 	object->generation++;
178 
179 	TAILQ_INSERT_TAIL(&vm_object_list, object, object_list);
180 	vm_object_count++;
181 	object_hash_rand = object->hash_rand;
182 }
183 
184 /*
185  *	vm_object_init:
186  *
187  *	Initialize the VM objects module.
188  */
189 void
190 vm_object_init()
191 {
192 	TAILQ_INIT(&vm_object_list);
193 	simple_lock_init(&vm_object_list_lock);
194 	vm_object_count = 0;
195 
196 	kernel_object = &kernel_object_store;
197 	_vm_object_allocate(OBJT_DEFAULT, OFF_TO_IDX(VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS),
198 	    kernel_object);
199 
200 	kmem_object = &kmem_object_store;
201 	_vm_object_allocate(OBJT_DEFAULT, OFF_TO_IDX(VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS),
202 	    kmem_object);
203 
204 	obj_zone = &obj_zone_store;
205 	zbootinit(obj_zone, "VM OBJECT", sizeof (struct vm_object),
206 		vm_objects_init, VM_OBJECTS_INIT);
207 }
208 
209 void
210 vm_object_init2() {
211 	zinitna(obj_zone, NULL, NULL, 0, 0, 0, 1);
212 }
213 
214 /*
215  *	vm_object_allocate:
216  *
217  *	Returns a new object with the given size.
218  */
219 
220 vm_object_t
221 vm_object_allocate(type, size)
222 	objtype_t type;
223 	vm_size_t size;
224 {
225 	vm_object_t result;
226 
227 	result = (vm_object_t) zalloc(obj_zone);
228 
229 	_vm_object_allocate(type, size, result);
230 
231 	return (result);
232 }
233 
234 
235 /*
236  *	vm_object_reference:
237  *
238  *	Gets another reference to the given object.
239  */
240 void
241 vm_object_reference(object)
242 	vm_object_t object;
243 {
244 	if (object == NULL)
245 		return;
246 
247 	KASSERT(!(object->flags & OBJ_DEAD),
248 	    ("vm_object_reference: attempting to reference dead obj"));
249 
250 	object->ref_count++;
251 	if (object->type == OBJT_VNODE) {
252 		while (vget((struct vnode *) object->handle, LK_RETRY|LK_NOOBJ, curproc)) {
253 #if !defined(MAX_PERF)
254 			printf("vm_object_reference: delay in getting object\n");
255 #endif
256 		}
257 	}
258 }
259 
260 void
261 vm_object_vndeallocate(object)
262 	vm_object_t object;
263 {
264 	struct vnode *vp = (struct vnode *) object->handle;
265 
266 	KASSERT(object->type == OBJT_VNODE,
267 	    ("vm_object_vndeallocate: not a vnode object"));
268 	KASSERT(vp != NULL, ("vm_object_vndeallocate: missing vp"));
269 #ifdef INVARIANTS
270 	if (object->ref_count == 0) {
271 		vprint("vm_object_vndeallocate", vp);
272 		panic("vm_object_vndeallocate: bad object reference count");
273 	}
274 #endif
275 
276 	object->ref_count--;
277 	if (object->ref_count == 0) {
278 		vp->v_flag &= ~VTEXT;
279 		vm_object_clear_flag(object, OBJ_OPT);
280 	}
281 	vrele(vp);
282 }
283 
284 /*
285  *	vm_object_deallocate:
286  *
287  *	Release a reference to the specified object,
288  *	gained either through a vm_object_allocate
289  *	or a vm_object_reference call.  When all references
290  *	are gone, storage associated with this object
291  *	may be relinquished.
292  *
293  *	No object may be locked.
294  */
295 void
296 vm_object_deallocate(object)
297 	vm_object_t object;
298 {
299 	vm_object_t temp;
300 
301 	while (object != NULL) {
302 
303 		if (object->type == OBJT_VNODE) {
304 			vm_object_vndeallocate(object);
305 			return;
306 		}
307 
308 		if (object->ref_count == 0) {
309 			panic("vm_object_deallocate: object deallocated too many times: %d", object->type);
310 		} else if (object->ref_count > 2) {
311 			object->ref_count--;
312 			return;
313 		}
314 
315 		/*
316 		 * Here on ref_count of one or two, which are special cases for
317 		 * objects.
318 		 */
319 		if ((object->ref_count == 2) && (object->shadow_count == 0)) {
320 			vm_object_set_flag(object, OBJ_ONEMAPPING);
321 			object->ref_count--;
322 			return;
323 		} else if ((object->ref_count == 2) && (object->shadow_count == 1)) {
324 			object->ref_count--;
325 			if ((object->handle == NULL) &&
326 			    (object->type == OBJT_DEFAULT ||
327 			     object->type == OBJT_SWAP)) {
328 				vm_object_t robject;
329 
330 				robject = TAILQ_FIRST(&object->shadow_head);
331 				KASSERT(robject != NULL,
332 				    ("vm_object_deallocate: ref_count: %d, shadow_count: %d",
333 					 object->ref_count,
334 					 object->shadow_count));
335 				if ((robject->handle == NULL) &&
336 				    (robject->type == OBJT_DEFAULT ||
337 				     robject->type == OBJT_SWAP)) {
338 
339 					robject->ref_count++;
340 
341 					while (
342 						robject->paging_in_progress ||
343 						object->paging_in_progress
344 					) {
345 						vm_object_pip_sleep(robject, "objde1");
346 						vm_object_pip_sleep(object, "objde2");
347 					}
348 
349 					if (robject->ref_count == 1) {
350 						robject->ref_count--;
351 						object = robject;
352 						goto doterm;
353 					}
354 
355 					object = robject;
356 					vm_object_collapse(object);
357 					continue;
358 				}
359 			}
360 
361 			return;
362 
363 		} else {
364 			object->ref_count--;
365 			if (object->ref_count != 0)
366 				return;
367 		}
368 
369 doterm:
370 
371 		temp = object->backing_object;
372 		if (temp) {
373 			TAILQ_REMOVE(&temp->shadow_head, object, shadow_list);
374 			temp->shadow_count--;
375 			if (temp->ref_count == 0)
376 				vm_object_clear_flag(temp, OBJ_OPT);
377 			temp->generation++;
378 			object->backing_object = NULL;
379 		}
380 		vm_object_terminate(object);
381 		/* unlocks and deallocates object */
382 		object = temp;
383 	}
384 }
385 
386 /*
387  *	vm_object_terminate actually destroys the specified object, freeing
388  *	up all previously used resources.
389  *
390  *	The object must be locked.
391  *	This routine may block.
392  */
393 void
394 vm_object_terminate(object)
395 	vm_object_t object;
396 {
397 	vm_page_t p;
398 	int s;
399 
400 	/*
401 	 * Make sure no one uses us.
402 	 */
403 	vm_object_set_flag(object, OBJ_DEAD);
404 
405 	/*
406 	 * wait for the pageout daemon to be done with the object
407 	 */
408 	vm_object_pip_wait(object, "objtrm");
409 
410 	KASSERT(!object->paging_in_progress,
411 		("vm_object_terminate: pageout in progress"));
412 
413 	/*
414 	 * Clean and free the pages, as appropriate. All references to the
415 	 * object are gone, so we don't need to lock it.
416 	 */
417 	if (object->type == OBJT_VNODE) {
418 		struct vnode *vp;
419 
420 		/*
421 		 * Freeze optimized copies.
422 		 */
423 		vm_freeze_copyopts(object, 0, object->size);
424 
425 		/*
426 		 * Clean pages and flush buffers.
427 		 */
428 		vm_object_page_clean(object, 0, 0, OBJPC_SYNC);
429 
430 		vp = (struct vnode *) object->handle;
431 		vinvalbuf(vp, V_SAVE, NOCRED, NULL, 0, 0);
432 	}
433 
434 	if (object->ref_count != 0)
435 		panic("vm_object_terminate: object with references, ref_count=%d", object->ref_count);
436 
437 	/*
438 	 * Now free any remaining pages. For internal objects, this also
439 	 * removes them from paging queues. Don't free wired pages, just
440 	 * remove them from the object.
441 	 */
442 	s = splvm();
443 	while ((p = TAILQ_FIRST(&object->memq)) != NULL) {
444 #if !defined(MAX_PERF)
445 		if (p->busy || (p->flags & PG_BUSY))
446 			panic("vm_object_terminate: freeing busy page %p\n", p);
447 #endif
448 		if (p->wire_count == 0) {
449 			vm_page_busy(p);
450 			vm_page_free(p);
451 			cnt.v_pfree++;
452 		} else {
453 			vm_page_busy(p);
454 			vm_page_remove(p);
455 		}
456 	}
457 	splx(s);
458 
459 	/*
460 	 * Let the pager know object is dead.
461 	 */
462 	vm_pager_deallocate(object);
463 
464 	/*
465 	 * Remove the object from the global object list.
466 	 */
467 	simple_lock(&vm_object_list_lock);
468 	TAILQ_REMOVE(&vm_object_list, object, object_list);
469 	simple_unlock(&vm_object_list_lock);
470 
471 	wakeup(object);
472 
473 	/*
474 	 * Free the space for the object.
475 	 */
476 	zfree(obj_zone, object);
477 }
478 
479 /*
480  *	vm_object_page_clean
481  *
482  *	Clean all dirty pages in the specified range of object.
483  *	Leaves page on whatever queue it is currently on.
484  *
485  *	Odd semantics: if start == end, we clean everything.
486  *
487  *	The object must be locked.
488  */
489 
490 void
491 vm_object_page_clean(object, start, end, flags)
492 	vm_object_t object;
493 	vm_pindex_t start;
494 	vm_pindex_t end;
495 	int flags;
496 {
497 	vm_page_t p, np, tp;
498 	vm_offset_t tstart, tend;
499 	vm_pindex_t pi;
500 	int s;
501 	struct vnode *vp;
502 	int runlen;
503 	int maxf;
504 	int chkb;
505 	int maxb;
506 	int i;
507 	int pagerflags;
508 	vm_page_t maf[vm_pageout_page_count];
509 	vm_page_t mab[vm_pageout_page_count];
510 	vm_page_t ma[vm_pageout_page_count];
511 	int curgeneration;
512 
513 	if (object->type != OBJT_VNODE ||
514 		(object->flags & OBJ_MIGHTBEDIRTY) == 0)
515 		return;
516 
517 	pagerflags = (flags & (OBJPC_SYNC | OBJPC_INVAL)) ? VM_PAGER_PUT_SYNC : 0;
518 	pagerflags |= (flags & OBJPC_INVAL) ? VM_PAGER_PUT_INVAL : 0;
519 
520 	vp = object->handle;
521 
522 	vm_object_set_flag(object, OBJ_CLEANING);
523 
524 	tstart = start;
525 	if (end == 0) {
526 		tend = object->size;
527 	} else {
528 		tend = end;
529 	}
530 
531 	for(p = TAILQ_FIRST(&object->memq); p; p = TAILQ_NEXT(p, listq)) {
532 		vm_page_flag_set(p, PG_CLEANCHK);
533 		vm_page_protect(p, VM_PROT_READ);
534 	}
535 
536 	if ((tstart == 0) && (tend == object->size)) {
537 		vm_object_clear_flag(object, OBJ_WRITEABLE|OBJ_MIGHTBEDIRTY);
538 	}
539 
540 rescan:
541 	curgeneration = object->generation;
542 
543 	for(p = TAILQ_FIRST(&object->memq); p; p = np) {
544 		np = TAILQ_NEXT(p, listq);
545 
546 		pi = p->pindex;
547 		if (((p->flags & PG_CLEANCHK) == 0) ||
548 			(pi < tstart) || (pi >= tend) ||
549 			(p->valid == 0) ||
550 			((p->queue - p->pc) == PQ_CACHE)) {
551 			vm_page_flag_clear(p, PG_CLEANCHK);
552 			continue;
553 		}
554 
555 		vm_page_test_dirty(p);
556 		if ((p->dirty & p->valid) == 0) {
557 			vm_page_flag_clear(p, PG_CLEANCHK);
558 			continue;
559 		}
560 
561 		s = splvm();
562 		while (vm_page_sleep_busy(p, TRUE, "vpcwai")) {
563 			if (object->generation != curgeneration) {
564 				splx(s);
565 				goto rescan;
566 			}
567 		}
568 
569 		maxf = 0;
570 		for(i=1;i<vm_pageout_page_count;i++) {
571 			if ((tp = vm_page_lookup(object, pi + i)) != NULL) {
572 				if ((tp->flags & PG_BUSY) ||
573 					(tp->flags & PG_CLEANCHK) == 0 ||
574 					(tp->busy != 0))
575 					break;
576 				if((tp->queue - tp->pc) == PQ_CACHE) {
577 					vm_page_flag_clear(tp, PG_CLEANCHK);
578 					break;
579 				}
580 				vm_page_test_dirty(tp);
581 				if ((tp->dirty & tp->valid) == 0) {
582 					vm_page_flag_clear(tp, PG_CLEANCHK);
583 					break;
584 				}
585 				maf[ i - 1 ] = tp;
586 				maxf++;
587 				continue;
588 			}
589 			break;
590 		}
591 
592 		maxb = 0;
593 		chkb = vm_pageout_page_count -  maxf;
594 		if (chkb) {
595 			for(i = 1; i < chkb;i++) {
596 				if ((tp = vm_page_lookup(object, pi - i)) != NULL) {
597 					if ((tp->flags & PG_BUSY) ||
598 						(tp->flags & PG_CLEANCHK) == 0 ||
599 						(tp->busy != 0))
600 						break;
601 					if((tp->queue - tp->pc) == PQ_CACHE) {
602 						vm_page_flag_clear(tp, PG_CLEANCHK);
603 						break;
604 					}
605 					vm_page_test_dirty(tp);
606 					if ((tp->dirty & tp->valid) == 0) {
607 						vm_page_flag_clear(tp, PG_CLEANCHK);
608 						break;
609 					}
610 					mab[ i - 1 ] = tp;
611 					maxb++;
612 					continue;
613 				}
614 				break;
615 			}
616 		}
617 
618 		for(i=0;i<maxb;i++) {
619 			int index = (maxb - i) - 1;
620 			ma[index] = mab[i];
621 			vm_page_flag_clear(ma[index], PG_CLEANCHK);
622 		}
623 		vm_page_flag_clear(p, PG_CLEANCHK);
624 		ma[maxb] = p;
625 		for(i=0;i<maxf;i++) {
626 			int index = (maxb + i) + 1;
627 			ma[index] = maf[i];
628 			vm_page_flag_clear(ma[index], PG_CLEANCHK);
629 		}
630 		runlen = maxb + maxf + 1;
631 
632 		splx(s);
633 		vm_pageout_flush(ma, runlen, pagerflags);
634 		for (i = 0; i<runlen; i++) {
635 			if (ma[i]->valid & ma[i]->dirty) {
636 				vm_page_protect(ma[i], VM_PROT_READ);
637 				vm_page_flag_set(ma[i], PG_CLEANCHK);
638 			}
639 		}
640 		if (object->generation != curgeneration)
641 			goto rescan;
642 	}
643 
644 #if 0
645 	VOP_FSYNC(vp, NULL, (pagerflags & VM_PAGER_PUT_SYNC)?MNT_WAIT:0, curproc);
646 #endif
647 
648 	vm_object_clear_flag(object, OBJ_CLEANING);
649 	return;
650 }
651 
652 #ifdef not_used
653 /* XXX I cannot tell if this should be an exported symbol */
654 /*
655  *	vm_object_deactivate_pages
656  *
657  *	Deactivate all pages in the specified object.  (Keep its pages
658  *	in memory even though it is no longer referenced.)
659  *
660  *	The object must be locked.
661  */
662 static void
663 vm_object_deactivate_pages(object)
664 	vm_object_t object;
665 {
666 	vm_page_t p, next;
667 
668 	for (p = TAILQ_FIRST(&object->memq); p != NULL; p = next) {
669 		next = TAILQ_NEXT(p, listq);
670 		vm_page_deactivate(p);
671 	}
672 }
673 #endif
674 
675 /*
676  * Same as vm_object_pmap_copy, except range checking really
677  * works, and is meant for small sections of an object.
678  *
679  * This code protects resident pages by making them read-only
680  * and is typically called on a fork or split when a page
681  * is converted to copy-on-write.
682  *
683  * NOTE: If the page is already at VM_PROT_NONE, calling
684  * vm_page_protect will have no effect.
685  */
686 
687 void
688 vm_object_pmap_copy_1(object, start, end)
689 	vm_object_t object;
690 	vm_pindex_t start;
691 	vm_pindex_t end;
692 {
693 	vm_pindex_t idx;
694 	vm_page_t p;
695 
696 	if (object == NULL || (object->flags & OBJ_WRITEABLE) == 0)
697 		return;
698 
699 	for (idx = start; idx < end; idx++) {
700 		p = vm_page_lookup(object, idx);
701 		if (p == NULL)
702 			continue;
703 		vm_page_protect(p, VM_PROT_READ);
704 	}
705 }
706 
707 /*
708  *	vm_object_pmap_remove:
709  *
710  *	Removes all physical pages in the specified
711  *	object range from all physical maps.
712  *
713  *	The object must *not* be locked.
714  */
715 void
716 vm_object_pmap_remove(object, start, end)
717 	vm_object_t object;
718 	vm_pindex_t start;
719 	vm_pindex_t end;
720 {
721 	vm_page_t p;
722 
723 	if (object == NULL)
724 		return;
725 	for (p = TAILQ_FIRST(&object->memq);
726 		p != NULL;
727 		p = TAILQ_NEXT(p, listq)) {
728 		if (p->pindex >= start && p->pindex < end)
729 			vm_page_protect(p, VM_PROT_NONE);
730 	}
731 	if ((start == 0) && (object->size == end))
732 		vm_object_clear_flag(object, OBJ_WRITEABLE);
733 }
734 
735 /*
736  *	vm_object_madvise:
737  *
738  *	Implements the madvise function at the object/page level.
739  *
740  *	MADV_WILLNEED	(any object)
741  *
742  *	    Activate the specified pages if they are resident.
743  *
744  *	MADV_DONTNEED	(any object)
745  *
746  *	    Deactivate the specified pages if they are resident.
747  *
748  *	MADV_FREE	(OBJT_DEFAULT/OBJT_SWAP objects,
749  *			 OBJ_ONEMAPPING only)
750  *
751  *	    Deactivate and clean the specified pages if they are
752  *	    resident.  This permits the process to reuse the pages
753  *	    without faulting or the kernel to reclaim the pages
754  *	    without I/O.
755  */
756 void
757 vm_object_madvise(object, pindex, count, advise)
758 	vm_object_t object;
759 	vm_pindex_t pindex;
760 	int count;
761 	int advise;
762 {
763 	vm_pindex_t end, tpindex;
764 	vm_object_t tobject;
765 	vm_page_t m;
766 
767 	if (object == NULL)
768 		return;
769 
770 	end = pindex + count;
771 
772 	/*
773 	 * Locate and adjust resident pages
774 	 */
775 
776 	for (; pindex < end; pindex += 1) {
777 relookup:
778 		tobject = object;
779 		tpindex = pindex;
780 shadowlookup:
781 		/*
782 		 * MADV_FREE only operates on OBJT_DEFAULT or OBJT_SWAP pages
783 		 * and those pages must be OBJ_ONEMAPPING.
784 		 */
785 		if (advise == MADV_FREE) {
786 			if ((tobject->type != OBJT_DEFAULT &&
787 			     tobject->type != OBJT_SWAP) ||
788 			    (tobject->flags & OBJ_ONEMAPPING) == 0) {
789 				continue;
790 			}
791 		}
792 
793 		m = vm_page_lookup(tobject, tpindex);
794 
795 		if (m == NULL) {
796 			/*
797 			 * There may be swap even if there is no backing page
798 			 */
799 			if (advise == MADV_FREE && tobject->type == OBJT_SWAP)
800 				swap_pager_freespace(tobject, tpindex, 1);
801 
802 			/*
803 			 * next object
804 			 */
805 			tobject = tobject->backing_object;
806 			if (tobject == NULL)
807 				continue;
808 			tpindex += OFF_TO_IDX(tobject->backing_object_offset);
809 			goto shadowlookup;
810 		}
811 
812 		/*
813 		 * If the page is busy or not in a normal active state,
814 		 * we skip it.  Things can break if we mess with pages
815 		 * in any of the below states.
816 		 */
817 		if (
818 		    m->hold_count ||
819 		    m->wire_count ||
820 		    m->valid != VM_PAGE_BITS_ALL
821 		) {
822 			continue;
823 		}
824 
825  		if (vm_page_sleep_busy(m, TRUE, "madvpo"))
826   			goto relookup;
827 
828 		if (advise == MADV_WILLNEED) {
829 			vm_page_activate(m);
830 		} else if (advise == MADV_DONTNEED) {
831 			vm_page_dontneed(m);
832 		} else if (advise == MADV_FREE) {
833 			/*
834 			 * Mark the page clean.  This will allow the page
835 			 * to be freed up by the system.  However, such pages
836 			 * are often reused quickly by malloc()/free()
837 			 * so we do not do anything that would cause
838 			 * a page fault if we can help it.
839 			 *
840 			 * Specifically, we do not try to actually free
841 			 * the page now nor do we try to put it in the
842 			 * cache (which would cause a page fault on reuse).
843 			 *
844 			 * But we do make the page is freeable as we
845 			 * can without actually taking the step of unmapping
846 			 * it.
847 			 */
848 			pmap_clear_modify(VM_PAGE_TO_PHYS(m));
849 			m->dirty = 0;
850 			m->act_count = 0;
851 			vm_page_dontneed(m);
852 			if (tobject->type == OBJT_SWAP)
853 				swap_pager_freespace(tobject, tpindex, 1);
854 		}
855 	}
856 }
857 
858 /*
859  *	vm_object_shadow:
860  *
861  *	Create a new object which is backed by the
862  *	specified existing object range.  The source
863  *	object reference is deallocated.
864  *
865  *	The new object and offset into that object
866  *	are returned in the source parameters.
867  */
868 
869 void
870 vm_object_shadow(object, offset, length)
871 	vm_object_t *object;	/* IN/OUT */
872 	vm_ooffset_t *offset;	/* IN/OUT */
873 	vm_size_t length;
874 {
875 	vm_object_t source;
876 	vm_object_t result;
877 
878 	source = *object;
879 
880 	/*
881 	 * Don't create the new object if the old object isn't shared.
882 	 */
883 
884 	if (source != NULL &&
885 	    source->ref_count == 1 &&
886 	    source->handle == NULL &&
887 	    (source->type == OBJT_DEFAULT ||
888 	     source->type == OBJT_SWAP))
889 		return;
890 
891 	KASSERT((source->flags & OBJ_ONEMAPPING) == 0,
892 		("vm_object_shadow: source object has OBJ_ONEMAPPING set.\n"));
893 
894 	/*
895 	 * Allocate a new object with the given length
896 	 */
897 
898 	if ((result = vm_object_allocate(OBJT_DEFAULT, length)) == NULL)
899 		panic("vm_object_shadow: no object for shadowing");
900 
901 	/*
902 	 * The new object shadows the source object, adding a reference to it.
903 	 * Our caller changes his reference to point to the new object,
904 	 * removing a reference to the source object.  Net result: no change
905 	 * of reference count.
906 	 *
907 	 * Try to optimize the result object's page color when shadowing
908 	 * in order to maintain page coloring consistancy in the combined
909 	 * shadowed object.
910 	 */
911 	result->backing_object = source;
912 	if (source) {
913 		TAILQ_INSERT_TAIL(&source->shadow_head, result, shadow_list);
914 		source->shadow_count++;
915 		source->generation++;
916 		result->pg_color = (source->pg_color + OFF_TO_IDX(*offset)) & PQ_L2_MASK;
917 	}
918 
919 	/*
920 	 * Store the offset into the source object, and fix up the offset into
921 	 * the new object.
922 	 */
923 
924 	result->backing_object_offset = *offset;
925 
926 	/*
927 	 * Return the new things
928 	 */
929 
930 	*offset = 0;
931 	*object = result;
932 }
933 
934 #define	OBSC_TEST_ALL_SHADOWED	0x0001
935 #define	OBSC_COLLAPSE_NOWAIT	0x0002
936 #define	OBSC_COLLAPSE_WAIT	0x0004
937 
938 static __inline int
939 vm_object_backing_scan(vm_object_t object, int op)
940 {
941 	int s;
942 	int r = 1;
943 	vm_page_t p;
944 	vm_object_t backing_object;
945 	vm_pindex_t backing_offset_index;
946 
947 	s = splvm();
948 
949 	backing_object = object->backing_object;
950 	backing_offset_index = OFF_TO_IDX(object->backing_object_offset);
951 
952 	/*
953 	 * Initial conditions
954 	 */
955 
956 	if (op & OBSC_TEST_ALL_SHADOWED) {
957 		/*
958 		 * We do not want to have to test for the existance of
959 		 * swap pages in the backing object.  XXX but with the
960 		 * new swapper this would be pretty easy to do.
961 		 *
962 		 * XXX what about anonymous MAP_SHARED memory that hasn't
963 		 * been ZFOD faulted yet?  If we do not test for this, the
964 		 * shadow test may succeed! XXX
965 		 */
966 		if (backing_object->type != OBJT_DEFAULT) {
967 			splx(s);
968 			return(0);
969 		}
970 	}
971 	if (op & OBSC_COLLAPSE_WAIT) {
972 		vm_object_set_flag(backing_object, OBJ_DEAD);
973 	}
974 
975 	/*
976 	 * Our scan
977 	 */
978 
979 	p = TAILQ_FIRST(&backing_object->memq);
980 	while (p) {
981 		vm_page_t next = TAILQ_NEXT(p, listq);
982 		vm_pindex_t new_pindex = p->pindex - backing_offset_index;
983 
984 		if (op & OBSC_TEST_ALL_SHADOWED) {
985 			vm_page_t pp;
986 
987 			/*
988 			 * Ignore pages outside the parent object's range
989 			 * and outside the parent object's mapping of the
990 			 * backing object.
991 			 *
992 			 * note that we do not busy the backing object's
993 			 * page.
994 			 */
995 
996 			if (
997 			    p->pindex < backing_offset_index ||
998 			    new_pindex >= object->size
999 			) {
1000 				p = next;
1001 				continue;
1002 			}
1003 
1004 			/*
1005 			 * See if the parent has the page or if the parent's
1006 			 * object pager has the page.  If the parent has the
1007 			 * page but the page is not valid, the parent's
1008 			 * object pager must have the page.
1009 			 *
1010 			 * If this fails, the parent does not completely shadow
1011 			 * the object and we might as well give up now.
1012 			 */
1013 
1014 			pp = vm_page_lookup(object, new_pindex);
1015 			if (
1016 			    (pp == NULL || pp->valid == 0) &&
1017 			    !vm_pager_has_page(object, new_pindex, NULL, NULL)
1018 			) {
1019 				r = 0;
1020 				break;
1021 			}
1022 		}
1023 
1024 		/*
1025 		 * Check for busy page
1026 		 */
1027 
1028 		if (op & (OBSC_COLLAPSE_WAIT | OBSC_COLLAPSE_NOWAIT)) {
1029 			vm_page_t pp;
1030 
1031 			if (op & OBSC_COLLAPSE_NOWAIT) {
1032 				if (
1033 				    (p->flags & PG_BUSY) ||
1034 				    !p->valid ||
1035 				    p->hold_count ||
1036 				    p->wire_count ||
1037 				    p->busy
1038 				) {
1039 					p = next;
1040 					continue;
1041 				}
1042 			} else if (op & OBSC_COLLAPSE_WAIT) {
1043 				if (vm_page_sleep_busy(p, TRUE, "vmocol")) {
1044 					/*
1045 					 * If we slept, anything could have
1046 					 * happened.  Since the object is
1047 					 * marked dead, the backing offset
1048 					 * should not have changed so we
1049 					 * just restart our scan.
1050 					 */
1051 					p = TAILQ_FIRST(&backing_object->memq);
1052 					continue;
1053 				}
1054 			}
1055 
1056 			/*
1057 			 * Busy the page
1058 			 */
1059 			vm_page_busy(p);
1060 
1061 			KASSERT(
1062 			    p->object == backing_object,
1063 			    ("vm_object_qcollapse(): object mismatch")
1064 			);
1065 
1066 			/*
1067 			 * Destroy any associated swap
1068 			 */
1069 			if (backing_object->type == OBJT_SWAP) {
1070 				swap_pager_freespace(
1071 				    backing_object,
1072 				    p->pindex,
1073 				    1
1074 				);
1075 			}
1076 
1077 			if (
1078 			    p->pindex < backing_offset_index ||
1079 			    new_pindex >= object->size
1080 			) {
1081 				/*
1082 				 * Page is out of the parent object's range, we
1083 				 * can simply destroy it.
1084 				 */
1085 				vm_page_protect(p, VM_PROT_NONE);
1086 				vm_page_free(p);
1087 				p = next;
1088 				continue;
1089 			}
1090 
1091 			pp = vm_page_lookup(object, new_pindex);
1092 			if (
1093 			    pp != NULL ||
1094 			    vm_pager_has_page(object, new_pindex, NULL, NULL)
1095 			) {
1096 				/*
1097 				 * page already exists in parent OR swap exists
1098 				 * for this location in the parent.  Destroy
1099 				 * the original page from the backing object.
1100 				 *
1101 				 * Leave the parent's page alone
1102 				 */
1103 				vm_page_protect(p, VM_PROT_NONE);
1104 				vm_page_free(p);
1105 				p = next;
1106 				continue;
1107 			}
1108 
1109 			/*
1110 			 * Page does not exist in parent, rename the
1111 			 * page from the backing object to the main object.
1112 			 *
1113 			 * If the page was mapped to a process, it can remain
1114 			 * mapped through the rename.
1115 			 */
1116 			if ((p->queue - p->pc) == PQ_CACHE)
1117 				vm_page_deactivate(p);
1118 
1119 			vm_page_rename(p, object, new_pindex);
1120 			/* page automatically made dirty by rename */
1121 		}
1122 		p = next;
1123 	}
1124 	splx(s);
1125 	return(r);
1126 }
1127 
1128 
1129 /*
1130  * this version of collapse allows the operation to occur earlier and
1131  * when paging_in_progress is true for an object...  This is not a complete
1132  * operation, but should plug 99.9% of the rest of the leaks.
1133  */
1134 static void
1135 vm_object_qcollapse(object)
1136 	vm_object_t object;
1137 {
1138 	vm_object_t backing_object = object->backing_object;
1139 
1140 	if (backing_object->ref_count != 1)
1141 		return;
1142 
1143 	backing_object->ref_count += 2;
1144 
1145 	vm_object_backing_scan(object, OBSC_COLLAPSE_NOWAIT);
1146 
1147 	backing_object->ref_count -= 2;
1148 }
1149 
1150 /*
1151  *	vm_object_collapse:
1152  *
1153  *	Collapse an object with the object backing it.
1154  *	Pages in the backing object are moved into the
1155  *	parent, and the backing object is deallocated.
1156  */
1157 void
1158 vm_object_collapse(object)
1159 	vm_object_t object;
1160 {
1161 	while (TRUE) {
1162 		vm_object_t backing_object;
1163 
1164 		/*
1165 		 * Verify that the conditions are right for collapse:
1166 		 *
1167 		 * The object exists and the backing object exists.
1168 		 */
1169 		if (object == NULL)
1170 			break;
1171 
1172 		if ((backing_object = object->backing_object) == NULL)
1173 			break;
1174 
1175 		/*
1176 		 * we check the backing object first, because it is most likely
1177 		 * not collapsable.
1178 		 */
1179 		if (backing_object->handle != NULL ||
1180 		    (backing_object->type != OBJT_DEFAULT &&
1181 		     backing_object->type != OBJT_SWAP) ||
1182 		    (backing_object->flags & OBJ_DEAD) ||
1183 		    object->handle != NULL ||
1184 		    (object->type != OBJT_DEFAULT &&
1185 		     object->type != OBJT_SWAP) ||
1186 		    (object->flags & OBJ_DEAD)) {
1187 			break;
1188 		}
1189 
1190 		if (
1191 		    object->paging_in_progress != 0 ||
1192 		    backing_object->paging_in_progress != 0
1193 		) {
1194 			vm_object_qcollapse(object);
1195 			break;
1196 		}
1197 
1198 		/*
1199 		 * We know that we can either collapse the backing object (if
1200 		 * the parent is the only reference to it) or (perhaps) have
1201 		 * the parent bypass the object if the parent happens to shadow
1202 		 * all the resident pages in the entire backing object.
1203 		 *
1204 		 * This is ignoring pager-backed pages such as swap pages.
1205 		 * vm_object_backing_scan fails the shadowing test in this
1206 		 * case.
1207 		 */
1208 
1209 		if (backing_object->ref_count == 1) {
1210 			/*
1211 			 * If there is exactly one reference to the backing
1212 			 * object, we can collapse it into the parent.
1213 			 */
1214 
1215 			vm_object_backing_scan(object, OBSC_COLLAPSE_WAIT);
1216 
1217 			/*
1218 			 * Move the pager from backing_object to object.
1219 			 */
1220 
1221 			if (backing_object->type == OBJT_SWAP) {
1222 				vm_object_pip_add(backing_object, 1);
1223 
1224 				/*
1225 				 * scrap the paging_offset junk and do a
1226 				 * discrete copy.  This also removes major
1227 				 * assumptions about how the swap-pager
1228 				 * works from where it doesn't belong.  The
1229 				 * new swapper is able to optimize the
1230 				 * destroy-source case.
1231 				 */
1232 
1233 				vm_object_pip_add(object, 1);
1234 				swap_pager_copy(
1235 				    backing_object,
1236 				    object,
1237 				    OFF_TO_IDX(object->backing_object_offset), TRUE);
1238 				vm_object_pip_wakeup(object);
1239 
1240 				vm_object_pip_wakeup(backing_object);
1241 			}
1242 			/*
1243 			 * Object now shadows whatever backing_object did.
1244 			 * Note that the reference to
1245 			 * backing_object->backing_object moves from within
1246 			 * backing_object to within object.
1247 			 */
1248 
1249 			TAILQ_REMOVE(
1250 			    &object->backing_object->shadow_head,
1251 			    object,
1252 			    shadow_list
1253 			);
1254 			object->backing_object->shadow_count--;
1255 			object->backing_object->generation++;
1256 			if (backing_object->backing_object) {
1257 				TAILQ_REMOVE(
1258 				    &backing_object->backing_object->shadow_head,
1259 				    backing_object,
1260 				    shadow_list
1261 				);
1262 				backing_object->backing_object->shadow_count--;
1263 				backing_object->backing_object->generation++;
1264 			}
1265 			object->backing_object = backing_object->backing_object;
1266 			if (object->backing_object) {
1267 				TAILQ_INSERT_TAIL(
1268 				    &object->backing_object->shadow_head,
1269 				    object,
1270 				    shadow_list
1271 				);
1272 				object->backing_object->shadow_count++;
1273 				object->backing_object->generation++;
1274 			}
1275 
1276 			object->backing_object_offset +=
1277 			    backing_object->backing_object_offset;
1278 
1279 			/*
1280 			 * Discard backing_object.
1281 			 *
1282 			 * Since the backing object has no pages, no pager left,
1283 			 * and no object references within it, all that is
1284 			 * necessary is to dispose of it.
1285 			 */
1286 
1287 			TAILQ_REMOVE(
1288 			    &vm_object_list,
1289 			    backing_object,
1290 			    object_list
1291 			);
1292 			vm_object_count--;
1293 
1294 			zfree(obj_zone, backing_object);
1295 
1296 			object_collapses++;
1297 		} else {
1298 			vm_object_t new_backing_object;
1299 
1300 			/*
1301 			 * If we do not entirely shadow the backing object,
1302 			 * there is nothing we can do so we give up.
1303 			 */
1304 
1305 			if (vm_object_backing_scan(object, OBSC_TEST_ALL_SHADOWED) == 0) {
1306 				break;
1307 			}
1308 
1309 			/*
1310 			 * Make the parent shadow the next object in the
1311 			 * chain.  Deallocating backing_object will not remove
1312 			 * it, since its reference count is at least 2.
1313 			 */
1314 
1315 			TAILQ_REMOVE(
1316 			    &backing_object->shadow_head,
1317 			    object,
1318 			    shadow_list
1319 			);
1320 			backing_object->shadow_count--;
1321 			backing_object->generation++;
1322 
1323 			new_backing_object = backing_object->backing_object;
1324 			if ((object->backing_object = new_backing_object) != NULL) {
1325 				vm_object_reference(new_backing_object);
1326 				TAILQ_INSERT_TAIL(
1327 				    &new_backing_object->shadow_head,
1328 				    object,
1329 				    shadow_list
1330 				);
1331 				new_backing_object->shadow_count++;
1332 				new_backing_object->generation++;
1333 				object->backing_object_offset +=
1334 					backing_object->backing_object_offset;
1335 			}
1336 
1337 			/*
1338 			 * Drop the reference count on backing_object. Since
1339 			 * its ref_count was at least 2, it will not vanish;
1340 			 * so we don't need to call vm_object_deallocate, but
1341 			 * we do anyway.
1342 			 */
1343 			vm_object_deallocate(backing_object);
1344 			object_bypasses++;
1345 		}
1346 
1347 		/*
1348 		 * Try again with this object's new backing object.
1349 		 */
1350 	}
1351 }
1352 
1353 /*
1354  *	vm_object_page_remove: [internal]
1355  *
1356  *	Removes all physical pages in the specified
1357  *	object range from the object's list of pages.
1358  *
1359  *	The object must be locked.
1360  */
1361 void
1362 vm_object_page_remove(object, start, end, clean_only)
1363 	vm_object_t object;
1364 	vm_pindex_t start;
1365 	vm_pindex_t end;
1366 	boolean_t clean_only;
1367 {
1368 	vm_page_t p, next;
1369 	unsigned int size;
1370 	int all;
1371 
1372 	if (object == NULL ||
1373 	    object->resident_page_count == 0)
1374 		return;
1375 
1376 	all = ((end == 0) && (start == 0));
1377 
1378 	vm_object_pip_add(object, 1);
1379 again:
1380 	size = end - start;
1381 	if (all || size > object->resident_page_count / 4) {
1382 		for (p = TAILQ_FIRST(&object->memq); p != NULL; p = next) {
1383 			next = TAILQ_NEXT(p, listq);
1384 			if (all || ((start <= p->pindex) && (p->pindex < end))) {
1385 				if (p->wire_count != 0) {
1386 					vm_page_protect(p, VM_PROT_NONE);
1387 					if (!clean_only)
1388 						p->valid = 0;
1389 					continue;
1390 				}
1391 
1392 				/*
1393 				 * The busy flags are only cleared at
1394 				 * interrupt -- minimize the spl transitions
1395 				 */
1396 
1397  				if (vm_page_sleep_busy(p, TRUE, "vmopar"))
1398  					goto again;
1399 
1400 				if (clean_only && p->valid) {
1401 					vm_page_test_dirty(p);
1402 					if (p->valid & p->dirty)
1403 						continue;
1404 				}
1405 
1406 				vm_page_busy(p);
1407 				vm_page_protect(p, VM_PROT_NONE);
1408 				vm_page_free(p);
1409 			}
1410 		}
1411 	} else {
1412 		while (size > 0) {
1413 			if ((p = vm_page_lookup(object, start)) != 0) {
1414 
1415 				if (p->wire_count != 0) {
1416 					vm_page_protect(p, VM_PROT_NONE);
1417 					if (!clean_only)
1418 						p->valid = 0;
1419 					start += 1;
1420 					size -= 1;
1421 					continue;
1422 				}
1423 
1424 				/*
1425 				 * The busy flags are only cleared at
1426 				 * interrupt -- minimize the spl transitions
1427 				 */
1428  				if (vm_page_sleep_busy(p, TRUE, "vmopar"))
1429 					goto again;
1430 
1431 				if (clean_only && p->valid) {
1432 					vm_page_test_dirty(p);
1433 					if (p->valid & p->dirty) {
1434 						start += 1;
1435 						size -= 1;
1436 						continue;
1437 					}
1438 				}
1439 
1440 				vm_page_busy(p);
1441 				vm_page_protect(p, VM_PROT_NONE);
1442 				vm_page_free(p);
1443 			}
1444 			start += 1;
1445 			size -= 1;
1446 		}
1447 	}
1448 	vm_object_pip_wakeup(object);
1449 }
1450 
1451 /*
1452  *	Routine:	vm_object_coalesce
1453  *	Function:	Coalesces two objects backing up adjoining
1454  *			regions of memory into a single object.
1455  *
1456  *	returns TRUE if objects were combined.
1457  *
1458  *	NOTE:	Only works at the moment if the second object is NULL -
1459  *		if it's not, which object do we lock first?
1460  *
1461  *	Parameters:
1462  *		prev_object	First object to coalesce
1463  *		prev_offset	Offset into prev_object
1464  *		next_object	Second object into coalesce
1465  *		next_offset	Offset into next_object
1466  *
1467  *		prev_size	Size of reference to prev_object
1468  *		next_size	Size of reference to next_object
1469  *
1470  *	Conditions:
1471  *	The object must *not* be locked.
1472  */
1473 boolean_t
1474 vm_object_coalesce(prev_object, prev_pindex, prev_size, next_size)
1475 	vm_object_t prev_object;
1476 	vm_pindex_t prev_pindex;
1477 	vm_size_t prev_size, next_size;
1478 {
1479 	vm_pindex_t next_pindex;
1480 
1481 	if (prev_object == NULL) {
1482 		return (TRUE);
1483 	}
1484 
1485 	if (prev_object->type != OBJT_DEFAULT &&
1486 	    prev_object->type != OBJT_SWAP) {
1487 		return (FALSE);
1488 	}
1489 
1490 	/*
1491 	 * Try to collapse the object first
1492 	 */
1493 	vm_object_collapse(prev_object);
1494 
1495 	/*
1496 	 * Can't coalesce if: . more than one reference . paged out . shadows
1497 	 * another object . has a copy elsewhere (any of which mean that the
1498 	 * pages not mapped to prev_entry may be in use anyway)
1499 	 */
1500 
1501 	if (prev_object->backing_object != NULL) {
1502 		return (FALSE);
1503 	}
1504 
1505 	prev_size >>= PAGE_SHIFT;
1506 	next_size >>= PAGE_SHIFT;
1507 	next_pindex = prev_pindex + prev_size;
1508 
1509 	if ((prev_object->ref_count > 1) &&
1510 	    (prev_object->size != next_pindex)) {
1511 		return (FALSE);
1512 	}
1513 
1514 	/*
1515 	 * Remove any pages that may still be in the object from a previous
1516 	 * deallocation.
1517 	 */
1518 	if (next_pindex < prev_object->size) {
1519 		vm_object_page_remove(prev_object,
1520 				      next_pindex,
1521 				      next_pindex + next_size, FALSE);
1522 		if (prev_object->type == OBJT_SWAP)
1523 			swap_pager_freespace(prev_object,
1524 					     next_pindex, next_size);
1525 	}
1526 
1527 	/*
1528 	 * Extend the object if necessary.
1529 	 */
1530 	if (next_pindex + next_size > prev_object->size)
1531 		prev_object->size = next_pindex + next_size;
1532 
1533 	return (TRUE);
1534 }
1535 
1536 #include "opt_ddb.h"
1537 #ifdef DDB
1538 #include <sys/kernel.h>
1539 
1540 #include <sys/cons.h>
1541 
1542 #include <ddb/ddb.h>
1543 
1544 static int	_vm_object_in_map __P((vm_map_t map, vm_object_t object,
1545 				       vm_map_entry_t entry));
1546 static int	vm_object_in_map __P((vm_object_t object));
1547 
1548 static int
1549 _vm_object_in_map(map, object, entry)
1550 	vm_map_t map;
1551 	vm_object_t object;
1552 	vm_map_entry_t entry;
1553 {
1554 	vm_map_t tmpm;
1555 	vm_map_entry_t tmpe;
1556 	vm_object_t obj;
1557 	int entcount;
1558 
1559 	if (map == 0)
1560 		return 0;
1561 
1562 	if (entry == 0) {
1563 		tmpe = map->header.next;
1564 		entcount = map->nentries;
1565 		while (entcount-- && (tmpe != &map->header)) {
1566 			if( _vm_object_in_map(map, object, tmpe)) {
1567 				return 1;
1568 			}
1569 			tmpe = tmpe->next;
1570 		}
1571 	} else if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
1572 		tmpm = entry->object.sub_map;
1573 		tmpe = tmpm->header.next;
1574 		entcount = tmpm->nentries;
1575 		while (entcount-- && tmpe != &tmpm->header) {
1576 			if( _vm_object_in_map(tmpm, object, tmpe)) {
1577 				return 1;
1578 			}
1579 			tmpe = tmpe->next;
1580 		}
1581 	} else if ((obj = entry->object.vm_object) != NULL) {
1582 		for(; obj; obj=obj->backing_object)
1583 			if( obj == object) {
1584 				return 1;
1585 			}
1586 	}
1587 	return 0;
1588 }
1589 
1590 static int
1591 vm_object_in_map( object)
1592 	vm_object_t object;
1593 {
1594 	struct proc *p;
1595 	for (p = allproc.lh_first; p != 0; p = p->p_list.le_next) {
1596 		if( !p->p_vmspace /* || (p->p_flag & (P_SYSTEM|P_WEXIT)) */)
1597 			continue;
1598 		if( _vm_object_in_map(&p->p_vmspace->vm_map, object, 0))
1599 			return 1;
1600 	}
1601 	if( _vm_object_in_map( kernel_map, object, 0))
1602 		return 1;
1603 	if( _vm_object_in_map( kmem_map, object, 0))
1604 		return 1;
1605 	if( _vm_object_in_map( pager_map, object, 0))
1606 		return 1;
1607 	if( _vm_object_in_map( buffer_map, object, 0))
1608 		return 1;
1609 	if( _vm_object_in_map( phys_map, object, 0))
1610 		return 1;
1611 	if( _vm_object_in_map( mb_map, object, 0))
1612 		return 1;
1613 	return 0;
1614 }
1615 
1616 DB_SHOW_COMMAND(vmochk, vm_object_check)
1617 {
1618 	vm_object_t object;
1619 
1620 	/*
1621 	 * make sure that internal objs are in a map somewhere
1622 	 * and none have zero ref counts.
1623 	 */
1624 	for (object = TAILQ_FIRST(&vm_object_list);
1625 			object != NULL;
1626 			object = TAILQ_NEXT(object, object_list)) {
1627 		if (object->handle == NULL &&
1628 		    (object->type == OBJT_DEFAULT || object->type == OBJT_SWAP)) {
1629 			if (object->ref_count == 0) {
1630 				db_printf("vmochk: internal obj has zero ref count: %ld\n",
1631 					(long)object->size);
1632 			}
1633 			if (!vm_object_in_map(object)) {
1634 				db_printf(
1635 			"vmochk: internal obj is not in a map: "
1636 			"ref: %d, size: %lu: 0x%lx, backing_object: %p\n",
1637 				    object->ref_count, (u_long)object->size,
1638 				    (u_long)object->size,
1639 				    (void *)object->backing_object);
1640 			}
1641 		}
1642 	}
1643 }
1644 
1645 /*
1646  *	vm_object_print:	[ debug ]
1647  */
1648 DB_SHOW_COMMAND(object, vm_object_print_static)
1649 {
1650 	/* XXX convert args. */
1651 	vm_object_t object = (vm_object_t)addr;
1652 	boolean_t full = have_addr;
1653 
1654 	vm_page_t p;
1655 
1656 	/* XXX count is an (unused) arg.  Avoid shadowing it. */
1657 #define	count	was_count
1658 
1659 	int count;
1660 
1661 	if (object == NULL)
1662 		return;
1663 
1664 	db_iprintf(
1665 	    "Object %p: type=%d, size=0x%lx, res=%d, ref=%d, flags=0x%x\n",
1666 	    object, (int)object->type, (u_long)object->size,
1667 	    object->resident_page_count, object->ref_count, object->flags);
1668 	/*
1669 	 * XXX no %qd in kernel.  Truncate object->backing_object_offset.
1670 	 */
1671 	db_iprintf(" sref=%d, backing_object(%d)=(%p)+0x%lx\n",
1672 	    object->shadow_count,
1673 	    object->backing_object ? object->backing_object->ref_count : 0,
1674 	    object->backing_object, (long)object->backing_object_offset);
1675 
1676 	if (!full)
1677 		return;
1678 
1679 	db_indent += 2;
1680 	count = 0;
1681 	for (p = TAILQ_FIRST(&object->memq); p != NULL; p = TAILQ_NEXT(p, listq)) {
1682 		if (count == 0)
1683 			db_iprintf("memory:=");
1684 		else if (count == 6) {
1685 			db_printf("\n");
1686 			db_iprintf(" ...");
1687 			count = 0;
1688 		} else
1689 			db_printf(",");
1690 		count++;
1691 
1692 		db_printf("(off=0x%lx,page=0x%lx)",
1693 		    (u_long) p->pindex, (u_long) VM_PAGE_TO_PHYS(p));
1694 	}
1695 	if (count != 0)
1696 		db_printf("\n");
1697 	db_indent -= 2;
1698 }
1699 
1700 /* XXX. */
1701 #undef count
1702 
1703 /* XXX need this non-static entry for calling from vm_map_print. */
1704 void
1705 vm_object_print(addr, have_addr, count, modif)
1706         /* db_expr_t */ long addr;
1707 	boolean_t have_addr;
1708 	/* db_expr_t */ long count;
1709 	char *modif;
1710 {
1711 	vm_object_print_static(addr, have_addr, count, modif);
1712 }
1713 
1714 DB_SHOW_COMMAND(vmopag, vm_object_print_pages)
1715 {
1716 	vm_object_t object;
1717 	int nl = 0;
1718 	int c;
1719 	for (object = TAILQ_FIRST(&vm_object_list);
1720 			object != NULL;
1721 			object = TAILQ_NEXT(object, object_list)) {
1722 		vm_pindex_t idx, fidx;
1723 		vm_pindex_t osize;
1724 		vm_offset_t pa = -1, padiff;
1725 		int rcount;
1726 		vm_page_t m;
1727 
1728 		db_printf("new object: %p\n", (void *)object);
1729 		if ( nl > 18) {
1730 			c = cngetc();
1731 			if (c != ' ')
1732 				return;
1733 			nl = 0;
1734 		}
1735 		nl++;
1736 		rcount = 0;
1737 		fidx = 0;
1738 		osize = object->size;
1739 		if (osize > 128)
1740 			osize = 128;
1741 		for(idx=0;idx<osize;idx++) {
1742 			m = vm_page_lookup(object, idx);
1743 			if (m == NULL) {
1744 				if (rcount) {
1745 					db_printf(" index(%ld)run(%d)pa(0x%lx)\n",
1746 						(long)fidx, rcount, (long)pa);
1747 					if ( nl > 18) {
1748 						c = cngetc();
1749 						if (c != ' ')
1750 							return;
1751 						nl = 0;
1752 					}
1753 					nl++;
1754 					rcount = 0;
1755 				}
1756 				continue;
1757 			}
1758 
1759 
1760 			if (rcount &&
1761 				(VM_PAGE_TO_PHYS(m) == pa + rcount * PAGE_SIZE)) {
1762 				++rcount;
1763 				continue;
1764 			}
1765 			if (rcount) {
1766 				padiff = pa + rcount * PAGE_SIZE - VM_PAGE_TO_PHYS(m);
1767 				padiff >>= PAGE_SHIFT;
1768 				padiff &= PQ_L2_MASK;
1769 				if (padiff == 0) {
1770 					pa = VM_PAGE_TO_PHYS(m) - rcount * PAGE_SIZE;
1771 					++rcount;
1772 					continue;
1773 				}
1774 				db_printf(" index(%ld)run(%d)pa(0x%lx)",
1775 					(long)fidx, rcount, (long)pa);
1776 				db_printf("pd(%ld)\n", (long)padiff);
1777 				if ( nl > 18) {
1778 					c = cngetc();
1779 					if (c != ' ')
1780 						return;
1781 					nl = 0;
1782 				}
1783 				nl++;
1784 			}
1785 			fidx = idx;
1786 			pa = VM_PAGE_TO_PHYS(m);
1787 			rcount = 1;
1788 		}
1789 		if (rcount) {
1790 			db_printf(" index(%ld)run(%d)pa(0x%lx)\n",
1791 				(long)fidx, rcount, (long)pa);
1792 			if ( nl > 18) {
1793 				c = cngetc();
1794 				if (c != ' ')
1795 					return;
1796 				nl = 0;
1797 			}
1798 			nl++;
1799 		}
1800 	}
1801 }
1802 #endif /* DDB */
1803