xref: /freebsd/sys/vm/vm_object.c (revision a8445737e740901f5f2c8d24c12ef7fc8b00134e)
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  * $Id: vm_object.c,v 1.127 1998/08/24 08:39:37 dfr Exp $
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 static void vm_object_dispose __P((vm_object_t));
95 
96 /*
97  *	Virtual memory objects maintain the actual data
98  *	associated with allocated virtual memory.  A given
99  *	page of memory exists within exactly one object.
100  *
101  *	An object is only deallocated when all "references"
102  *	are given up.  Only one "reference" to a given
103  *	region of an object should be writeable.
104  *
105  *	Associated with each object is a list of all resident
106  *	memory pages belonging to that object; this list is
107  *	maintained by the "vm_page" module, and locked by the object's
108  *	lock.
109  *
110  *	Each object also records a "pager" routine which is
111  *	used to retrieve (and store) pages to the proper backing
112  *	storage.  In addition, objects may be backed by other
113  *	objects from which they were virtual-copied.
114  *
115  *	The only items within the object structure which are
116  *	modified after time of creation are:
117  *		reference count		locked by object's lock
118  *		pager routine		locked by object's lock
119  *
120  */
121 
122 struct object_q vm_object_list;
123 static struct simplelock vm_object_list_lock;
124 static long vm_object_count;		/* count of all objects */
125 vm_object_t kernel_object;
126 vm_object_t kmem_object;
127 static struct vm_object kernel_object_store;
128 static struct vm_object kmem_object_store;
129 extern int vm_pageout_page_count;
130 
131 static long object_collapses;
132 static long object_bypasses;
133 static int next_index;
134 static vm_zone_t obj_zone;
135 static struct vm_zone obj_zone_store;
136 #define VM_OBJECTS_INIT 256
137 static struct vm_object vm_objects_init[VM_OBJECTS_INIT];
138 static int objidnumber;
139 
140 void
141 _vm_object_allocate(type, size, object)
142 	objtype_t type;
143 	vm_size_t size;
144 	register vm_object_t object;
145 {
146 	int incr;
147 	TAILQ_INIT(&object->memq);
148 	TAILQ_INIT(&object->shadow_head);
149 
150 	object->type = type;
151 	object->size = size;
152 	object->ref_count = 1;
153 	object->flags = 0;
154 	object->id = ++objidnumber;
155 	if ((object->type == OBJT_DEFAULT) || (object->type == OBJT_SWAP))
156 		vm_object_set_flag(object, OBJ_ONEMAPPING);
157 	object->behavior = OBJ_NORMAL;
158 	object->paging_in_progress = 0;
159 	object->resident_page_count = 0;
160 	object->cache_count = 0;
161 	object->wire_count = 0;
162 	object->shadow_count = 0;
163 	object->pg_color = next_index;
164 	if ( size > (PQ_L2_SIZE / 3 + PQ_PRIME1))
165 		incr = PQ_L2_SIZE / 3 + PQ_PRIME1;
166 	else
167 		incr = size;
168 	next_index = (next_index + incr) & PQ_L2_MASK;
169 	object->handle = NULL;
170 	object->paging_offset = (vm_ooffset_t) 0;
171 	object->backing_object = NULL;
172 	object->backing_object_offset = (vm_ooffset_t) 0;
173 	object->page_hint = NULL;
174 
175 	object->last_read = 0;
176 	object->generation++;
177 
178 	TAILQ_INSERT_TAIL(&vm_object_list, object, object_list);
179 	vm_object_count++;
180 }
181 
182 /*
183  *	vm_object_init:
184  *
185  *	Initialize the VM objects module.
186  */
187 void
188 vm_object_init()
189 {
190 	TAILQ_INIT(&vm_object_list);
191 	simple_lock_init(&vm_object_list_lock);
192 	vm_object_count = 0;
193 
194 	kernel_object = &kernel_object_store;
195 	_vm_object_allocate(OBJT_DEFAULT, OFF_TO_IDX(VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS),
196 	    kernel_object);
197 
198 	kmem_object = &kmem_object_store;
199 	_vm_object_allocate(OBJT_DEFAULT, OFF_TO_IDX(VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS),
200 	    kmem_object);
201 
202 	obj_zone = &obj_zone_store;
203 	zbootinit(obj_zone, "VM OBJECT", sizeof (struct vm_object),
204 		vm_objects_init, VM_OBJECTS_INIT);
205 }
206 
207 void
208 vm_object_init2() {
209 	zinitna(obj_zone, NULL, NULL, 0, 0, 0, 1);
210 }
211 
212 /*
213  *	vm_object_allocate:
214  *
215  *	Returns a new object with the given size.
216  */
217 
218 vm_object_t
219 vm_object_allocate(type, size)
220 	objtype_t type;
221 	vm_size_t size;
222 {
223 	register vm_object_t result;
224 	result = (vm_object_t) zalloc(obj_zone);
225 
226 	_vm_object_allocate(type, size, result);
227 
228 	return (result);
229 }
230 
231 
232 /*
233  *	vm_object_reference:
234  *
235  *	Gets another reference to the given object.
236  */
237 void
238 vm_object_reference(object)
239 	register vm_object_t object;
240 {
241 	if (object == NULL)
242 		return;
243 
244 #if defined(DIAGNOSTIC)
245 	if (object->flags & OBJ_DEAD)
246 		panic("vm_object_reference: attempting to reference dead obj");
247 #endif
248 
249 	object->ref_count++;
250 	if (object->type == OBJT_VNODE) {
251 		while (vget((struct vnode *) object->handle, LK_RETRY|LK_NOOBJ, curproc)) {
252 #if !defined(MAX_PERF)
253 			printf("vm_object_reference: delay in getting object\n");
254 #endif
255 		}
256 	}
257 }
258 
259 void
260 vm_object_vndeallocate(object)
261 	vm_object_t object;
262 {
263 	struct vnode *vp = (struct vnode *) object->handle;
264 #if defined(DIAGNOSTIC)
265 	if (object->type != OBJT_VNODE)
266 		panic("vm_object_vndeallocate: not a vnode object");
267 	if (vp == NULL)
268 		panic("vm_object_vndeallocate: missing vp");
269 	if (object->ref_count == 0) {
270 		vprint("vm_object_vndeallocate", vp);
271 		panic("vm_object_vndeallocate: bad object reference count");
272 	}
273 #endif
274 
275 	object->ref_count--;
276 	if (object->ref_count == 0) {
277 		vp->v_flag &= ~VTEXT;
278 		vm_object_clear_flag(object, OBJ_OPT);
279 	}
280 	vrele(vp);
281 }
282 
283 /*
284  *	vm_object_deallocate:
285  *
286  *	Release a reference to the specified object,
287  *	gained either through a vm_object_allocate
288  *	or a vm_object_reference call.  When all references
289  *	are gone, storage associated with this object
290  *	may be relinquished.
291  *
292  *	No object may be locked.
293  */
294 void
295 vm_object_deallocate(object)
296 	vm_object_t object;
297 {
298 	int s;
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 #if defined(DIAGNOSTIC)
332 				if (robject == NULL)
333 					panic("vm_object_deallocate: ref_count: %d,"
334 						  " shadow_count: %d",
335 						  object->ref_count, object->shadow_count);
336 #endif
337 				if ((robject->handle == NULL) &&
338 				    (robject->type == OBJT_DEFAULT ||
339 				     robject->type == OBJT_SWAP)) {
340 
341 					robject->ref_count++;
342 
343 			retry:
344 					if (robject->paging_in_progress ||
345 							object->paging_in_progress) {
346 						vm_object_pip_sleep(robject, "objde1");
347 						if (robject->paging_in_progress &&
348 							robject->type == OBJT_SWAP) {
349 							swap_pager_sync();
350 							goto retry;
351 						}
352 
353 						vm_object_pip_sleep(object, "objde2");
354 						if (object->paging_in_progress &&
355 							object->type == OBJT_SWAP) {
356 							swap_pager_sync();
357 						}
358 						goto retry;
359 					}
360 
361 					if( robject->ref_count == 1) {
362 						robject->ref_count--;
363 						object = robject;
364 						goto doterm;
365 					}
366 
367 					object = robject;
368 					vm_object_collapse(object);
369 					continue;
370 				}
371 			}
372 
373 			return;
374 
375 		} else {
376 			object->ref_count--;
377 			if (object->ref_count != 0)
378 				return;
379 		}
380 
381 doterm:
382 
383 		temp = object->backing_object;
384 		if (temp) {
385 			TAILQ_REMOVE(&temp->shadow_head, object, shadow_list);
386 			temp->shadow_count--;
387 			if (temp->ref_count == 0)
388 				vm_object_clear_flag(temp, OBJ_OPT);
389 			temp->generation++;
390 			object->backing_object = NULL;
391 		}
392 		vm_object_terminate(object);
393 		/* unlocks and deallocates object */
394 		object = temp;
395 	}
396 }
397 
398 /*
399  *	vm_object_terminate actually destroys the specified object, freeing
400  *	up all previously used resources.
401  *
402  *	The object must be locked.
403  */
404 void
405 vm_object_terminate(object)
406 	register vm_object_t object;
407 {
408 	register vm_page_t p;
409 	int s;
410 
411 	/*
412 	 * Make sure no one uses us.
413 	 */
414 	vm_object_set_flag(object, OBJ_DEAD);
415 
416 	/*
417 	 * wait for the pageout daemon to be done with the object
418 	 */
419 	vm_object_pip_wait(object, "objtrm");
420 
421 #if defined(DIAGNOSTIC)
422 	if (object->paging_in_progress != 0)
423 		panic("vm_object_terminate: pageout in progress");
424 #endif
425 
426 	/*
427 	 * Clean and free the pages, as appropriate. All references to the
428 	 * object are gone, so we don't need to lock it.
429 	 */
430 	if (object->type == OBJT_VNODE) {
431 		struct vnode *vp;
432 
433 		/*
434 		 * Freeze optimized copies.
435 		 */
436 		vm_freeze_copyopts(object, 0, object->size);
437 
438 		/*
439 		 * Clean pages and flush buffers.
440 		 */
441 		vm_object_page_clean(object, 0, 0, OBJPC_SYNC);
442 
443 		vp = (struct vnode *) object->handle;
444 		vinvalbuf(vp, V_SAVE, NOCRED, NULL, 0, 0);
445 
446 		/*
447 		 * Let the pager know object is dead.
448 		 */
449 		vm_pager_deallocate(object);
450 
451 	}
452 
453 	if ((object->type != OBJT_VNODE) && (object->ref_count == 0)) {
454 
455 		/*
456 		 * Now free the pages. For internal objects, this also removes them
457 		 * from paging queues.
458 		 */
459 		while ((p = TAILQ_FIRST(&object->memq)) != NULL) {
460 #if !defined(MAX_PERF)
461 			if (p->busy || (p->flags & PG_BUSY))
462 				printf("vm_object_terminate: freeing busy page\n");
463 #endif
464 			vm_page_busy(p);
465 			vm_page_free(p);
466 			cnt.v_pfree++;
467 		}
468 		/*
469 		 * Let the pager know object is dead.
470 		 */
471 		vm_pager_deallocate(object);
472 
473 	}
474 
475 	if ((object->ref_count == 0) && (object->resident_page_count == 0))
476 		vm_object_dispose(object);
477 }
478 
479 /*
480  * vm_object_dispose
481  *
482  * Dispose the object.
483  */
484 static void
485 vm_object_dispose(object)
486 	vm_object_t object;
487 {
488 		simple_lock(&vm_object_list_lock);
489 		TAILQ_REMOVE(&vm_object_list, object, object_list);
490 		vm_object_count--;
491 		simple_unlock(&vm_object_list_lock);
492 		/*
493    		* Free the space for the object.
494    		*/
495 		zfree(obj_zone, object);
496 		wakeup(object);
497 }
498 
499 /*
500  *	vm_object_page_clean
501  *
502  *	Clean all dirty pages in the specified range of object.
503  *	Leaves page on whatever queue it is currently on.
504  *
505  *	Odd semantics: if start == end, we clean everything.
506  *
507  *	The object must be locked.
508  */
509 
510 void
511 vm_object_page_clean(object, start, end, flags)
512 	vm_object_t object;
513 	vm_pindex_t start;
514 	vm_pindex_t end;
515 	int flags;
516 {
517 	register vm_page_t p, np, tp;
518 	register vm_offset_t tstart, tend;
519 	vm_pindex_t pi;
520 	int s;
521 	struct vnode *vp;
522 	int runlen;
523 	int maxf;
524 	int chkb;
525 	int maxb;
526 	int i;
527 	int pagerflags;
528 	vm_page_t maf[vm_pageout_page_count];
529 	vm_page_t mab[vm_pageout_page_count];
530 	vm_page_t ma[vm_pageout_page_count];
531 	int curgeneration;
532 	struct proc *pproc = curproc;	/* XXX */
533 
534 	if (object->type != OBJT_VNODE ||
535 		(object->flags & OBJ_MIGHTBEDIRTY) == 0)
536 		return;
537 
538 	pagerflags = (flags & (OBJPC_SYNC | OBJPC_INVAL)) ? VM_PAGER_PUT_SYNC : 0;
539 	pagerflags |= (flags & OBJPC_INVAL) ? VM_PAGER_PUT_INVAL : 0;
540 
541 	vp = object->handle;
542 
543 	vm_object_set_flag(object, OBJ_CLEANING);
544 
545 	tstart = start;
546 	if (end == 0) {
547 		tend = object->size;
548 	} else {
549 		tend = end;
550 	}
551 
552 	for(p = TAILQ_FIRST(&object->memq); p; p = TAILQ_NEXT(p, listq)) {
553 		vm_page_flag_set(p, PG_CLEANCHK);
554 		vm_page_protect(p, VM_PROT_READ);
555 	}
556 
557 	if ((tstart == 0) && (tend == object->size)) {
558 		vm_object_clear_flag(object, OBJ_WRITEABLE|OBJ_MIGHTBEDIRTY);
559 	}
560 
561 rescan:
562 	curgeneration = object->generation;
563 
564 	for(p = TAILQ_FIRST(&object->memq); p; p = np) {
565 		np = TAILQ_NEXT(p, listq);
566 
567 		pi = p->pindex;
568 		if (((p->flags & PG_CLEANCHK) == 0) ||
569 			(pi < tstart) || (pi >= tend) ||
570 			(p->valid == 0) ||
571 			((p->queue - p->pc) == PQ_CACHE)) {
572 			vm_page_flag_clear(p, PG_CLEANCHK);
573 			continue;
574 		}
575 
576 		vm_page_test_dirty(p);
577 		if ((p->dirty & p->valid) == 0) {
578 			vm_page_flag_clear(p, PG_CLEANCHK);
579 			continue;
580 		}
581 
582 		s = splvm();
583 		while ((p->flags & PG_BUSY) || p->busy) {
584 			vm_page_flag_set(p, PG_WANTED | PG_REFERENCED);
585 			tsleep(p, PVM, "vpcwai", 0);
586 			if (object->generation != curgeneration) {
587 				splx(s);
588 				goto rescan;
589 			}
590 		}
591 
592 		maxf = 0;
593 		for(i=1;i<vm_pageout_page_count;i++) {
594 			if (tp = vm_page_lookup(object, pi + i)) {
595 				if ((tp->flags & PG_BUSY) ||
596 					(tp->flags & PG_CLEANCHK) == 0 ||
597 					(tp->busy != 0))
598 					break;
599 				if((tp->queue - tp->pc) == PQ_CACHE) {
600 					vm_page_flag_clear(tp, PG_CLEANCHK);
601 					break;
602 				}
603 				vm_page_test_dirty(tp);
604 				if ((tp->dirty & tp->valid) == 0) {
605 					vm_page_flag_clear(tp, PG_CLEANCHK);
606 					break;
607 				}
608 				maf[ i - 1 ] = tp;
609 				maxf++;
610 				continue;
611 			}
612 			break;
613 		}
614 
615 		maxb = 0;
616 		chkb = vm_pageout_page_count -  maxf;
617 		if (chkb) {
618 			for(i = 1; i < chkb;i++) {
619 				if (tp = vm_page_lookup(object, pi - i)) {
620 					if ((tp->flags & PG_BUSY) ||
621 						(tp->flags & PG_CLEANCHK) == 0 ||
622 						(tp->busy != 0))
623 						break;
624 					if((tp->queue - tp->pc) == PQ_CACHE) {
625 						vm_page_flag_clear(tp, PG_CLEANCHK);
626 						break;
627 					}
628 					vm_page_test_dirty(tp);
629 					if ((tp->dirty & tp->valid) == 0) {
630 						vm_page_flag_clear(tp, PG_CLEANCHK);
631 						break;
632 					}
633 					mab[ i - 1 ] = tp;
634 					maxb++;
635 					continue;
636 				}
637 				break;
638 			}
639 		}
640 
641 		for(i=0;i<maxb;i++) {
642 			int index = (maxb - i) - 1;
643 			ma[index] = mab[i];
644 			vm_page_flag_clear(ma[index], PG_CLEANCHK);
645 		}
646 		vm_page_flag_clear(p, PG_CLEANCHK);
647 		ma[maxb] = p;
648 		for(i=0;i<maxf;i++) {
649 			int index = (maxb + i) + 1;
650 			ma[index] = maf[i];
651 			vm_page_flag_clear(ma[index], PG_CLEANCHK);
652 		}
653 		runlen = maxb + maxf + 1;
654 
655 		splx(s);
656 		vm_pageout_flush(ma, runlen, pagerflags);
657 		for (i = 0; i<runlen; i++) {
658 			if (ma[i]->valid & ma[i]->dirty) {
659 				vm_page_protect(ma[i], VM_PROT_READ);
660 				vm_page_flag_set(ma[i], PG_CLEANCHK);
661 			}
662 		}
663 		if (object->generation != curgeneration)
664 			goto rescan;
665 	}
666 
667 	VOP_FSYNC(vp, NULL, (pagerflags & VM_PAGER_PUT_SYNC)?MNT_WAIT:0, curproc);
668 
669 	vm_object_clear_flag(object, OBJ_CLEANING);
670 	return;
671 }
672 
673 #ifdef not_used
674 /* XXX I cannot tell if this should be an exported symbol */
675 /*
676  *	vm_object_deactivate_pages
677  *
678  *	Deactivate all pages in the specified object.  (Keep its pages
679  *	in memory even though it is no longer referenced.)
680  *
681  *	The object must be locked.
682  */
683 static void
684 vm_object_deactivate_pages(object)
685 	register vm_object_t object;
686 {
687 	register vm_page_t p, next;
688 
689 	for (p = TAILQ_FIRST(&object->memq); p != NULL; p = next) {
690 		next = TAILQ_NEXT(p, listq);
691 		vm_page_deactivate(p);
692 	}
693 }
694 #endif
695 
696 /*
697  *	vm_object_pmap_copy:
698  *
699  *	Makes all physical pages in the specified
700  *	object range copy-on-write.  No writeable
701  *	references to these pages should remain.
702  *
703  *	The object must *not* be locked.
704  */
705 void
706 vm_object_pmap_copy(object, start, end)
707 	register vm_object_t object;
708 	register vm_pindex_t start;
709 	register vm_pindex_t end;
710 {
711 	register vm_page_t p;
712 
713 	if (object == NULL || (object->flags & OBJ_WRITEABLE) == 0)
714 		return;
715 
716 	for (p = TAILQ_FIRST(&object->memq);
717 		p != NULL;
718 		p = TAILQ_NEXT(p, listq)) {
719 		vm_page_protect(p, VM_PROT_READ);
720 	}
721 
722 	vm_object_clear_flag(object, OBJ_WRITEABLE);
723 }
724 
725 /*
726  * Same as vm_object_pmap_copy_1, except range checking really
727  * works, and is meant for small sections of an object.
728  */
729 void
730 vm_object_pmap_copy_1(object, start, end)
731 	register vm_object_t object;
732 	register vm_pindex_t start;
733 	register vm_pindex_t end;
734 {
735 	vm_pindex_t idx;
736 	register vm_page_t p;
737 
738 	if (object == NULL || (object->flags & OBJ_WRITEABLE) == 0)
739 		return;
740 
741 	for (idx = start; idx < end; idx++) {
742 		p = vm_page_lookup(object, idx);
743 		if (p == NULL)
744 			continue;
745 		vm_page_protect(p, VM_PROT_READ);
746 	}
747 }
748 
749 /*
750  *	vm_object_pmap_remove:
751  *
752  *	Removes all physical pages in the specified
753  *	object range from all physical maps.
754  *
755  *	The object must *not* be locked.
756  */
757 void
758 vm_object_pmap_remove(object, start, end)
759 	register vm_object_t object;
760 	register vm_pindex_t start;
761 	register vm_pindex_t end;
762 {
763 	register vm_page_t p;
764 	if (object == NULL)
765 		return;
766 	for (p = TAILQ_FIRST(&object->memq);
767 		p != NULL;
768 		p = TAILQ_NEXT(p, listq)) {
769 		if (p->pindex >= start && p->pindex < end)
770 			vm_page_protect(p, VM_PROT_NONE);
771 	}
772 	if ((start == 0) && (object->size == end))
773 		vm_object_clear_flag(object, OBJ_WRITEABLE);
774 }
775 
776 /*
777  *	vm_object_madvise:
778  *
779  *	Implements the madvise function at the object/page level.
780  */
781 void
782 vm_object_madvise(object, pindex, count, advise)
783 	vm_object_t object;
784 	vm_pindex_t pindex;
785 	int count;
786 	int advise;
787 {
788 	int s;
789 	vm_pindex_t end, tpindex;
790 	vm_object_t tobject;
791 	vm_page_t m;
792 
793 	if (object == NULL)
794 		return;
795 
796 	end = pindex + count;
797 
798 	for (; pindex < end; pindex += 1) {
799 
800 relookup:
801 		tobject = object;
802 		tpindex = pindex;
803 shadowlookup:
804 		m = vm_page_lookup(tobject, tpindex);
805 		if (m == NULL) {
806 			if (tobject->type != OBJT_DEFAULT) {
807 				continue;
808 			}
809 
810 			tobject = tobject->backing_object;
811 			if ((tobject == NULL) || (tobject->ref_count != 1)) {
812 				continue;
813 			}
814 			tpindex += OFF_TO_IDX(tobject->backing_object_offset);
815 			goto shadowlookup;
816 		}
817 
818 		/*
819 		 * If the page is busy or not in a normal active state,
820 		 * we skip it.  Things can break if we mess with pages
821 		 * in any of the below states.
822 		 */
823 		if (m->hold_count || m->wire_count ||
824 			m->valid != VM_PAGE_BITS_ALL) {
825 			continue;
826 		}
827 
828  		if (vm_page_sleep(m, "madvpo", &m->busy))
829   			goto relookup;
830 
831 		if (advise == MADV_WILLNEED) {
832 			vm_page_activate(m);
833 		} else if (advise == MADV_DONTNEED) {
834 			vm_page_deactivate(m);
835 		} else if (advise == MADV_FREE) {
836 			pmap_clear_modify(VM_PAGE_TO_PHYS(m));
837 			m->dirty = 0;
838 			/*
839 			 * Force a demand zero if attempt to read from swap.
840 			 * We currently don't handle vnode files correctly,
841 			 * and will reread stale contents unnecessarily.
842 			 */
843 			if (object->type == OBJT_SWAP)
844 				swap_pager_dmzspace(tobject, m->pindex, 1);
845 		}
846 	}
847 }
848 
849 /*
850  *	vm_object_shadow:
851  *
852  *	Create a new object which is backed by the
853  *	specified existing object range.  The source
854  *	object reference is deallocated.
855  *
856  *	The new object and offset into that object
857  *	are returned in the source parameters.
858  */
859 
860 void
861 vm_object_shadow(object, offset, length)
862 	vm_object_t *object;	/* IN/OUT */
863 	vm_ooffset_t *offset;	/* IN/OUT */
864 	vm_size_t length;
865 {
866 	register vm_object_t source;
867 	register vm_object_t result;
868 
869 	source = *object;
870 
871 	/*
872 	 * Allocate a new object with the given length
873 	 */
874 
875 	if ((result = vm_object_allocate(OBJT_DEFAULT, length)) == NULL)
876 		panic("vm_object_shadow: no object for shadowing");
877 
878 	/*
879 	 * The new object shadows the source object, adding a reference to it.
880 	 * Our caller changes his reference to point to the new object,
881 	 * removing a reference to the source object.  Net result: no change
882 	 * of reference count.
883 	 */
884 	result->backing_object = source;
885 	if (source) {
886 		TAILQ_INSERT_TAIL(&source->shadow_head, result, shadow_list);
887 		vm_object_clear_flag(source, OBJ_ONEMAPPING);
888 		source->shadow_count++;
889 		source->generation++;
890 	}
891 
892 	/*
893 	 * Store the offset into the source object, and fix up the offset into
894 	 * the new object.
895 	 */
896 
897 	result->backing_object_offset = *offset;
898 
899 	/*
900 	 * Return the new things
901 	 */
902 
903 	*offset = 0;
904 	*object = result;
905 }
906 
907 
908 /*
909  * this version of collapse allows the operation to occur earlier and
910  * when paging_in_progress is true for an object...  This is not a complete
911  * operation, but should plug 99.9% of the rest of the leaks.
912  */
913 static void
914 vm_object_qcollapse(object)
915 	register vm_object_t object;
916 {
917 	register vm_object_t backing_object;
918 	register vm_pindex_t backing_offset_index, paging_offset_index;
919 	vm_pindex_t backing_object_paging_offset_index;
920 	vm_pindex_t new_pindex;
921 	register vm_page_t p, pp;
922 	register vm_size_t size;
923 
924 	backing_object = object->backing_object;
925 	if (backing_object->ref_count != 1)
926 		return;
927 
928 	backing_object->ref_count += 2;
929 
930 	backing_offset_index = OFF_TO_IDX(object->backing_object_offset);
931 	backing_object_paging_offset_index = OFF_TO_IDX(backing_object->paging_offset);
932 	paging_offset_index = OFF_TO_IDX(object->paging_offset);
933 	size = object->size;
934 	p = TAILQ_FIRST(&backing_object->memq);
935 	while (p) {
936 		vm_page_t next;
937 
938 		next = TAILQ_NEXT(p, listq);
939 		if ((p->flags & (PG_BUSY | PG_FICTITIOUS)) ||
940 		    !p->valid || p->hold_count || p->wire_count || p->busy) {
941 			p = next;
942 			continue;
943 		}
944 		vm_page_busy(p);
945 
946 		new_pindex = p->pindex - backing_offset_index;
947 		if (p->pindex < backing_offset_index ||
948 		    new_pindex >= size) {
949 			if (backing_object->type == OBJT_SWAP)
950 				swap_pager_freespace(backing_object,
951 				    backing_object_paging_offset_index+p->pindex,
952 				    1);
953 			vm_page_protect(p, VM_PROT_NONE);
954 			vm_page_free(p);
955 		} else {
956 			pp = vm_page_lookup(object, new_pindex);
957 			if (pp != NULL ||
958 				(object->type == OBJT_SWAP && vm_pager_has_page(object,
959 				    paging_offset_index + new_pindex, NULL, NULL))) {
960 				if (backing_object->type == OBJT_SWAP)
961 					swap_pager_freespace(backing_object,
962 					    backing_object_paging_offset_index + p->pindex, 1);
963 				vm_page_protect(p, VM_PROT_NONE);
964 				vm_page_free(p);
965 			} else {
966 				if (backing_object->type == OBJT_SWAP)
967 					swap_pager_freespace(backing_object,
968 					    backing_object_paging_offset_index + p->pindex, 1);
969 
970 				if ((p->queue - p->pc) == PQ_CACHE)
971 					vm_page_deactivate(p);
972 				else
973 					vm_page_protect(p, VM_PROT_NONE);
974 
975 				vm_page_rename(p, object, new_pindex);
976 				p->dirty = VM_PAGE_BITS_ALL;
977 			}
978 		}
979 		p = next;
980 	}
981 	backing_object->ref_count -= 2;
982 }
983 
984 /*
985  *	vm_object_collapse:
986  *
987  *	Collapse an object with the object backing it.
988  *	Pages in the backing object are moved into the
989  *	parent, and the backing object is deallocated.
990  */
991 void
992 vm_object_collapse(object)
993 	vm_object_t object;
994 
995 {
996 	vm_object_t backing_object;
997 	vm_ooffset_t backing_offset;
998 	vm_size_t size;
999 	vm_pindex_t new_pindex, backing_offset_index;
1000 	vm_page_t p, pp;
1001 
1002 	while (TRUE) {
1003 		/*
1004 		 * Verify that the conditions are right for collapse:
1005 		 *
1006 		 * The object exists and no pages in it are currently being paged
1007 		 * out.
1008 		 */
1009 		if (object == NULL)
1010 			return;
1011 
1012 		/*
1013 		 * Make sure there is a backing object.
1014 		 */
1015 		if ((backing_object = object->backing_object) == NULL)
1016 			return;
1017 
1018 		/*
1019 		 * we check the backing object first, because it is most likely
1020 		 * not collapsable.
1021 		 */
1022 		if (backing_object->handle != NULL ||
1023 		    (backing_object->type != OBJT_DEFAULT &&
1024 		     backing_object->type != OBJT_SWAP) ||
1025 		    (backing_object->flags & OBJ_DEAD) ||
1026 		    object->handle != NULL ||
1027 		    (object->type != OBJT_DEFAULT &&
1028 		     object->type != OBJT_SWAP) ||
1029 		    (object->flags & OBJ_DEAD)) {
1030 			return;
1031 		}
1032 
1033 		if (object->paging_in_progress != 0 ||
1034 		    backing_object->paging_in_progress != 0) {
1035 			vm_object_qcollapse(object);
1036 			return;
1037 		}
1038 
1039 		/*
1040 		 * We know that we can either collapse the backing object (if
1041 		 * the parent is the only reference to it) or (perhaps) remove
1042 		 * the parent's reference to it.
1043 		 */
1044 
1045 		backing_offset = object->backing_object_offset;
1046 		backing_offset_index = OFF_TO_IDX(backing_offset);
1047 		size = object->size;
1048 
1049 		/*
1050 		 * If there is exactly one reference to the backing object, we
1051 		 * can collapse it into the parent.
1052 		 */
1053 
1054 		if (backing_object->ref_count == 1) {
1055 
1056 			vm_object_set_flag(backing_object, OBJ_DEAD);
1057 			/*
1058 			 * We can collapse the backing object.
1059 			 *
1060 			 * Move all in-memory pages from backing_object to the
1061 			 * parent.  Pages that have been paged out will be
1062 			 * overwritten by any of the parent's pages that
1063 			 * shadow them.
1064 			 */
1065 
1066 			while ((p = TAILQ_FIRST(&backing_object->memq)) != 0) {
1067 
1068 				new_pindex = p->pindex - backing_offset_index;
1069 				vm_page_busy(p);
1070 
1071 				/*
1072 				 * If the parent has a page here, or if this
1073 				 * page falls outside the parent, dispose of
1074 				 * it.
1075 				 *
1076 				 * Otherwise, move it as planned.
1077 				 */
1078 
1079 				if (p->pindex < backing_offset_index ||
1080 				    new_pindex >= size) {
1081 					vm_page_protect(p, VM_PROT_NONE);
1082 					vm_page_free(p);
1083 				} else {
1084 					pp = vm_page_lookup(object, new_pindex);
1085 					if (pp != NULL || (object->type == OBJT_SWAP && vm_pager_has_page(object,
1086 					    OFF_TO_IDX(object->paging_offset) + new_pindex, NULL, NULL))) {
1087 						vm_page_protect(p, VM_PROT_NONE);
1088 						vm_page_free(p);
1089 					} else {
1090 						if ((p->queue - p->pc) == PQ_CACHE)
1091 							vm_page_deactivate(p);
1092 						else
1093 							vm_page_protect(p, VM_PROT_NONE);
1094 						vm_page_rename(p, object, new_pindex);
1095 						p->dirty = VM_PAGE_BITS_ALL;
1096 					}
1097 				}
1098 			}
1099 
1100 			/*
1101 			 * Move the pager from backing_object to object.
1102 			 */
1103 
1104 			if (backing_object->type == OBJT_SWAP) {
1105 				vm_object_pip_add(backing_object, 1);
1106 				if (object->type == OBJT_SWAP) {
1107 					vm_object_pip_add(object, 1);
1108 					/*
1109 					 * copy shadow object pages into ours
1110 					 * and destroy unneeded pages in
1111 					 * shadow object.
1112 					 */
1113 					swap_pager_copy(
1114 					    backing_object,
1115 					    OFF_TO_IDX(backing_object->paging_offset),
1116 					    object,
1117 					    OFF_TO_IDX(object->paging_offset),
1118 					    OFF_TO_IDX(object->backing_object_offset), TRUE);
1119 					vm_object_pip_wakeup(object);
1120 				} else {
1121 					vm_object_pip_add(object, 1);
1122 					/*
1123 					 * move the shadow backing_object's pager data to
1124 					 * "object" and convert "object" type to OBJT_SWAP.
1125 					 */
1126 					object->type = OBJT_SWAP;
1127 					object->un_pager.swp.swp_nblocks =
1128 					    backing_object->un_pager.swp.swp_nblocks;
1129 					object->un_pager.swp.swp_allocsize =
1130 					    backing_object->un_pager.swp.swp_allocsize;
1131 					object->un_pager.swp.swp_blocks =
1132 					    backing_object->un_pager.swp.swp_blocks;
1133 					object->un_pager.swp.swp_poip =		/* XXX */
1134 					    backing_object->un_pager.swp.swp_poip;
1135 					object->paging_offset = backing_object->paging_offset + backing_offset;
1136 					TAILQ_INSERT_TAIL(&swap_pager_un_object_list, object, pager_object_list);
1137 
1138 					/*
1139 					 * Convert backing object from OBJT_SWAP to
1140 					 * OBJT_DEFAULT. XXX - only the TAILQ_REMOVE is
1141 					 * actually necessary.
1142 					 */
1143 					backing_object->type = OBJT_DEFAULT;
1144 					TAILQ_REMOVE(&swap_pager_un_object_list, backing_object, pager_object_list);
1145 					/*
1146 					 * free unnecessary blocks
1147 					 */
1148 					swap_pager_freespace(object, 0,
1149 						OFF_TO_IDX(object->paging_offset));
1150 					vm_object_pip_wakeup(object);
1151 				}
1152 
1153 				vm_object_pip_wakeup(backing_object);
1154 			}
1155 			/*
1156 			 * Object now shadows whatever backing_object did.
1157 			 * Note that the reference to backing_object->backing_object
1158 			 * moves from within backing_object to within object.
1159 			 */
1160 
1161 			TAILQ_REMOVE(&object->backing_object->shadow_head, object,
1162 			    shadow_list);
1163 			object->backing_object->shadow_count--;
1164 			object->backing_object->generation++;
1165 			if (backing_object->backing_object) {
1166 				TAILQ_REMOVE(&backing_object->backing_object->shadow_head,
1167 				    backing_object, shadow_list);
1168 				backing_object->backing_object->shadow_count--;
1169 				backing_object->backing_object->generation++;
1170 			}
1171 			object->backing_object = backing_object->backing_object;
1172 			if (object->backing_object) {
1173 				TAILQ_INSERT_TAIL(&object->backing_object->shadow_head,
1174 				    object, shadow_list);
1175 				object->backing_object->shadow_count++;
1176 				object->backing_object->generation++;
1177 			}
1178 
1179 			object->backing_object_offset += backing_object->backing_object_offset;
1180 			/*
1181 			 * Discard backing_object.
1182 			 *
1183 			 * Since the backing object has no pages, no pager left,
1184 			 * and no object references within it, all that is
1185 			 * necessary is to dispose of it.
1186 			 */
1187 
1188 			TAILQ_REMOVE(&vm_object_list, backing_object,
1189 			    object_list);
1190 			vm_object_count--;
1191 
1192 			zfree(obj_zone, backing_object);
1193 
1194 			object_collapses++;
1195 		} else {
1196 			vm_object_t new_backing_object;
1197 			/*
1198 			 * If all of the pages in the backing object are
1199 			 * shadowed by the parent object, the parent object no
1200 			 * longer has to shadow the backing object; it can
1201 			 * shadow the next one in the chain.
1202 			 *
1203 			 * The backing object must not be paged out - we'd have
1204 			 * to check all of the paged-out pages, as well.
1205 			 */
1206 
1207 			if (backing_object->type != OBJT_DEFAULT) {
1208 				return;
1209 			}
1210 			/*
1211 			 * Should have a check for a 'small' number of pages
1212 			 * here.
1213 			 */
1214 
1215 			for (p = TAILQ_FIRST(&backing_object->memq); p;
1216 					p = TAILQ_NEXT(p, listq)) {
1217 
1218 				new_pindex = p->pindex - backing_offset_index;
1219 				vm_page_busy(p);
1220 
1221 				/*
1222 				 * If the parent has a page here, or if this
1223 				 * page falls outside the parent, keep going.
1224 				 *
1225 				 * Otherwise, the backing_object must be left in
1226 				 * the chain.
1227 				 */
1228 
1229 				if (p->pindex >= backing_offset_index &&
1230 					new_pindex <= size) {
1231 
1232 					pp = vm_page_lookup(object, new_pindex);
1233 
1234 					if ((pp == NULL) || (pp->flags & PG_BUSY) || pp->busy) {
1235 						vm_page_wakeup(p);
1236 						return;
1237 					}
1238 
1239 					vm_page_busy(pp);
1240 					if ((pp->valid == 0) &&
1241 				   	    !vm_pager_has_page(object, OFF_TO_IDX(object->paging_offset) + new_pindex, NULL, NULL)) {
1242 						/*
1243 						 * Page still needed. Can't go any
1244 						 * further.
1245 						 */
1246 						vm_page_wakeup(pp);
1247 						vm_page_wakeup(p);
1248 						return;
1249 					}
1250 					vm_page_wakeup(pp);
1251 				}
1252 				vm_page_wakeup(p);
1253 			}
1254 
1255 			/*
1256 			 * Make the parent shadow the next object in the
1257 			 * chain.  Deallocating backing_object will not remove
1258 			 * it, since its reference count is at least 2.
1259 			 */
1260 
1261 			TAILQ_REMOVE(&backing_object->shadow_head,
1262 			    object, shadow_list);
1263 			backing_object->shadow_count--;
1264 			backing_object->generation++;
1265 
1266 			new_backing_object = backing_object->backing_object;
1267 			if (object->backing_object = new_backing_object) {
1268 				vm_object_reference(new_backing_object);
1269 				TAILQ_INSERT_TAIL(&new_backing_object->shadow_head,
1270 				    object, shadow_list);
1271 				new_backing_object->shadow_count++;
1272 				new_backing_object->generation++;
1273 				object->backing_object_offset +=
1274 					backing_object->backing_object_offset;
1275 			}
1276 
1277 			/*
1278 			 * Drop the reference count on backing_object. Since
1279 			 * its ref_count was at least 2, it will not vanish;
1280 			 * so we don't need to call vm_object_deallocate, but
1281 			 * we do anyway.
1282 			 */
1283 			vm_object_deallocate(backing_object);
1284 			object_bypasses++;
1285 		}
1286 
1287 		/*
1288 		 * Try again with this object's new backing object.
1289 		 */
1290 	}
1291 }
1292 
1293 /*
1294  *	vm_object_page_remove: [internal]
1295  *
1296  *	Removes all physical pages in the specified
1297  *	object range from the object's list of pages.
1298  *
1299  *	The object must be locked.
1300  */
1301 void
1302 vm_object_page_remove(object, start, end, clean_only)
1303 	register vm_object_t object;
1304 	register vm_pindex_t start;
1305 	register vm_pindex_t end;
1306 	boolean_t clean_only;
1307 {
1308 	register vm_page_t p, next;
1309 	unsigned int size;
1310 	int s, all;
1311 
1312 	if (object == NULL)
1313 		return;
1314 
1315 	all = ((end == 0) && (start == 0));
1316 
1317 	vm_object_pip_add(object, 1);
1318 again:
1319 	size = end - start;
1320 	if (all || size > 4 || size >= object->size / 4) {
1321 		for (p = TAILQ_FIRST(&object->memq); p != NULL; p = next) {
1322 			next = TAILQ_NEXT(p, listq);
1323 			if (all || ((start <= p->pindex) && (p->pindex < end))) {
1324 				if (p->wire_count != 0) {
1325 					vm_page_protect(p, VM_PROT_NONE);
1326 					p->valid = 0;
1327 					continue;
1328 				}
1329 
1330 				/*
1331 				 * The busy flags are only cleared at
1332 				 * interrupt -- minimize the spl transitions
1333 				 */
1334 
1335  				if (vm_page_sleep(p, "vmopar", &p->busy))
1336  					goto again;
1337 
1338 				if (clean_only && p->valid) {
1339 					vm_page_test_dirty(p);
1340 					if (p->valid & p->dirty)
1341 						continue;
1342 				}
1343 
1344 				vm_page_busy(p);
1345 				vm_page_protect(p, VM_PROT_NONE);
1346 				vm_page_free(p);
1347 			}
1348 		}
1349 	} else {
1350 		while (size > 0) {
1351 			if ((p = vm_page_lookup(object, start)) != 0) {
1352 
1353 				if (p->wire_count != 0) {
1354 					p->valid = 0;
1355 					vm_page_protect(p, VM_PROT_NONE);
1356 					start += 1;
1357 					size -= 1;
1358 					continue;
1359 				}
1360 
1361 				/*
1362 				 * The busy flags are only cleared at
1363 				 * interrupt -- minimize the spl transitions
1364 				 */
1365  				if (vm_page_sleep(p, "vmopar", &p->busy))
1366 					goto again;
1367 
1368 				if (clean_only && p->valid) {
1369 					vm_page_test_dirty(p);
1370 					if (p->valid & p->dirty) {
1371 						start += 1;
1372 						size -= 1;
1373 						continue;
1374 					}
1375 				}
1376 
1377 				vm_page_busy(p);
1378 				vm_page_protect(p, VM_PROT_NONE);
1379 				vm_page_free(p);
1380 			}
1381 			start += 1;
1382 			size -= 1;
1383 		}
1384 	}
1385 	vm_object_pip_wakeup(object);
1386 }
1387 
1388 /*
1389  *	Routine:	vm_object_coalesce
1390  *	Function:	Coalesces two objects backing up adjoining
1391  *			regions of memory into a single object.
1392  *
1393  *	returns TRUE if objects were combined.
1394  *
1395  *	NOTE:	Only works at the moment if the second object is NULL -
1396  *		if it's not, which object do we lock first?
1397  *
1398  *	Parameters:
1399  *		prev_object	First object to coalesce
1400  *		prev_offset	Offset into prev_object
1401  *		next_object	Second object into coalesce
1402  *		next_offset	Offset into next_object
1403  *
1404  *		prev_size	Size of reference to prev_object
1405  *		next_size	Size of reference to next_object
1406  *
1407  *	Conditions:
1408  *	The object must *not* be locked.
1409  */
1410 boolean_t
1411 vm_object_coalesce(prev_object, prev_pindex, prev_size, next_size)
1412 	register vm_object_t prev_object;
1413 	vm_pindex_t prev_pindex;
1414 	vm_size_t prev_size, next_size;
1415 {
1416 	vm_size_t newsize;
1417 
1418 	if (prev_object == NULL) {
1419 		return (TRUE);
1420 	}
1421 
1422 	if (prev_object->type != OBJT_DEFAULT) {
1423 		return (FALSE);
1424 	}
1425 
1426 	/*
1427 	 * Try to collapse the object first
1428 	 */
1429 	vm_object_collapse(prev_object);
1430 
1431 	/*
1432 	 * Can't coalesce if: . more than one reference . paged out . shadows
1433 	 * another object . has a copy elsewhere (any of which mean that the
1434 	 * pages not mapped to prev_entry may be in use anyway)
1435 	 */
1436 
1437 	if (prev_object->backing_object != NULL) {
1438 		return (FALSE);
1439 	}
1440 
1441 	prev_size >>= PAGE_SHIFT;
1442 	next_size >>= PAGE_SHIFT;
1443 
1444 	if ((prev_object->ref_count > 1) &&
1445 	    (prev_object->size != prev_pindex + prev_size)) {
1446 		return (FALSE);
1447 	}
1448 
1449 	/*
1450 	 * Remove any pages that may still be in the object from a previous
1451 	 * deallocation.
1452 	 */
1453 
1454 	vm_object_page_remove(prev_object,
1455 	    prev_pindex + prev_size,
1456 	    prev_pindex + prev_size + next_size, FALSE);
1457 
1458 	/*
1459 	 * Extend the object if necessary.
1460 	 */
1461 	newsize = prev_pindex + prev_size + next_size;
1462 	if (newsize > prev_object->size)
1463 		prev_object->size = newsize;
1464 
1465 	return (TRUE);
1466 }
1467 
1468 #include "opt_ddb.h"
1469 #ifdef DDB
1470 #include <sys/kernel.h>
1471 
1472 #include <machine/cons.h>
1473 
1474 #include <ddb/ddb.h>
1475 
1476 static int	_vm_object_in_map __P((vm_map_t map, vm_object_t object,
1477 				       vm_map_entry_t entry));
1478 static int	vm_object_in_map __P((vm_object_t object));
1479 
1480 static int
1481 _vm_object_in_map(map, object, entry)
1482 	vm_map_t map;
1483 	vm_object_t object;
1484 	vm_map_entry_t entry;
1485 {
1486 	vm_map_t tmpm;
1487 	vm_map_entry_t tmpe;
1488 	vm_object_t obj;
1489 	int entcount;
1490 
1491 	if (map == 0)
1492 		return 0;
1493 
1494 	if (entry == 0) {
1495 		tmpe = map->header.next;
1496 		entcount = map->nentries;
1497 		while (entcount-- && (tmpe != &map->header)) {
1498 			if( _vm_object_in_map(map, object, tmpe)) {
1499 				return 1;
1500 			}
1501 			tmpe = tmpe->next;
1502 		}
1503 	} else if (entry->eflags & (MAP_ENTRY_IS_A_MAP|MAP_ENTRY_IS_SUB_MAP)) {
1504 		tmpm = entry->object.share_map;
1505 		tmpe = tmpm->header.next;
1506 		entcount = tmpm->nentries;
1507 		while (entcount-- && tmpe != &tmpm->header) {
1508 			if( _vm_object_in_map(tmpm, object, tmpe)) {
1509 				return 1;
1510 			}
1511 			tmpe = tmpe->next;
1512 		}
1513 	} else if (obj = entry->object.vm_object) {
1514 		for(; obj; obj=obj->backing_object)
1515 			if( obj == object) {
1516 				return 1;
1517 			}
1518 	}
1519 	return 0;
1520 }
1521 
1522 static int
1523 vm_object_in_map( object)
1524 	vm_object_t object;
1525 {
1526 	struct proc *p;
1527 	for (p = allproc.lh_first; p != 0; p = p->p_list.le_next) {
1528 		if( !p->p_vmspace /* || (p->p_flag & (P_SYSTEM|P_WEXIT)) */)
1529 			continue;
1530 		if( _vm_object_in_map(&p->p_vmspace->vm_map, object, 0))
1531 			return 1;
1532 	}
1533 	if( _vm_object_in_map( kernel_map, object, 0))
1534 		return 1;
1535 	if( _vm_object_in_map( kmem_map, object, 0))
1536 		return 1;
1537 	if( _vm_object_in_map( pager_map, object, 0))
1538 		return 1;
1539 	if( _vm_object_in_map( buffer_map, object, 0))
1540 		return 1;
1541 	if( _vm_object_in_map( io_map, object, 0))
1542 		return 1;
1543 	if( _vm_object_in_map( phys_map, object, 0))
1544 		return 1;
1545 	if( _vm_object_in_map( mb_map, object, 0))
1546 		return 1;
1547 	if( _vm_object_in_map( u_map, object, 0))
1548 		return 1;
1549 	return 0;
1550 }
1551 
1552 DB_SHOW_COMMAND(vmochk, vm_object_check)
1553 {
1554 	vm_object_t object;
1555 
1556 	/*
1557 	 * make sure that internal objs are in a map somewhere
1558 	 * and none have zero ref counts.
1559 	 */
1560 	for (object = TAILQ_FIRST(&vm_object_list);
1561 			object != NULL;
1562 			object = TAILQ_NEXT(object, object_list)) {
1563 		if (object->handle == NULL &&
1564 		    (object->type == OBJT_DEFAULT || object->type == OBJT_SWAP)) {
1565 			if (object->ref_count == 0) {
1566 				db_printf("vmochk: internal obj has zero ref count: %d\n",
1567 					object->size);
1568 			}
1569 			if (!vm_object_in_map(object)) {
1570 				db_printf(
1571 			"vmochk: internal obj is not in a map: "
1572 			"ref: %d, size: %lu: 0x%lx, backing_object: %p\n",
1573 				    object->ref_count, (u_long)object->size,
1574 				    (u_long)object->size,
1575 				    (void *)object->backing_object);
1576 			}
1577 		}
1578 	}
1579 }
1580 
1581 /*
1582  *	vm_object_print:	[ debug ]
1583  */
1584 DB_SHOW_COMMAND(object, vm_object_print_static)
1585 {
1586 	/* XXX convert args. */
1587 	vm_object_t object = (vm_object_t)addr;
1588 	boolean_t full = have_addr;
1589 
1590 	register vm_page_t p;
1591 
1592 	/* XXX count is an (unused) arg.  Avoid shadowing it. */
1593 #define	count	was_count
1594 
1595 	register int count;
1596 
1597 	if (object == NULL)
1598 		return;
1599 
1600 	db_iprintf(
1601 	    "Object %p: type=%d, size=0x%lx, res=%d, ref=%d, flags=0x%x\n",
1602 	    object, (int)object->type, (u_long)object->size,
1603 	    object->resident_page_count, object->ref_count, object->flags);
1604 	/*
1605 	 * XXX no %qd in kernel.  Truncate object->paging_offset and
1606 	 * object->backing_object_offset.
1607 	 */
1608 	db_iprintf(" sref=%d, offset=0x%lx, backing_object(%d)=(%p)+0x%lx\n",
1609 	    object->shadow_count, (long)object->paging_offset,
1610 	    object->backing_object ? object->backing_object->ref_count : 0,
1611 	    object->backing_object, (long)object->backing_object_offset);
1612 
1613 	if (!full)
1614 		return;
1615 
1616 	db_indent += 2;
1617 	count = 0;
1618 	for (p = TAILQ_FIRST(&object->memq); p != NULL; p = TAILQ_NEXT(p, listq)) {
1619 		if (count == 0)
1620 			db_iprintf("memory:=");
1621 		else if (count == 6) {
1622 			db_printf("\n");
1623 			db_iprintf(" ...");
1624 			count = 0;
1625 		} else
1626 			db_printf(",");
1627 		count++;
1628 
1629 		db_printf("(off=0x%lx,page=0x%lx)",
1630 		    (u_long) p->pindex, (u_long) VM_PAGE_TO_PHYS(p));
1631 	}
1632 	if (count != 0)
1633 		db_printf("\n");
1634 	db_indent -= 2;
1635 }
1636 
1637 /* XXX. */
1638 #undef count
1639 
1640 /* XXX need this non-static entry for calling from vm_map_print. */
1641 void
1642 vm_object_print(addr, have_addr, count, modif)
1643         /* db_expr_t */ long addr;
1644 	boolean_t have_addr;
1645 	/* db_expr_t */ long count;
1646 	char *modif;
1647 {
1648 	vm_object_print_static(addr, have_addr, count, modif);
1649 }
1650 
1651 DB_SHOW_COMMAND(vmopag, vm_object_print_pages)
1652 {
1653 	vm_object_t object;
1654 	int nl = 0;
1655 	int c;
1656 	for (object = TAILQ_FIRST(&vm_object_list);
1657 			object != NULL;
1658 			object = TAILQ_NEXT(object, object_list)) {
1659 		vm_pindex_t idx, fidx;
1660 		vm_pindex_t osize;
1661 		vm_offset_t pa = -1, padiff;
1662 		int rcount;
1663 		vm_page_t m;
1664 
1665 		db_printf("new object: %p\n", (void *)object);
1666 		if ( nl > 18) {
1667 			c = cngetc();
1668 			if (c != ' ')
1669 				return;
1670 			nl = 0;
1671 		}
1672 		nl++;
1673 		rcount = 0;
1674 		fidx = 0;
1675 		osize = object->size;
1676 		if (osize > 128)
1677 			osize = 128;
1678 		for(idx=0;idx<osize;idx++) {
1679 			m = vm_page_lookup(object, idx);
1680 			if (m == NULL) {
1681 				if (rcount) {
1682 					db_printf(" index(%d)run(%d)pa(0x%x)\n",
1683 						fidx, rcount, pa);
1684 					if ( nl > 18) {
1685 						c = cngetc();
1686 						if (c != ' ')
1687 							return;
1688 						nl = 0;
1689 					}
1690 					nl++;
1691 					rcount = 0;
1692 				}
1693 				continue;
1694 			}
1695 
1696 
1697 			if (rcount &&
1698 				(VM_PAGE_TO_PHYS(m) == pa + rcount * PAGE_SIZE)) {
1699 				++rcount;
1700 				continue;
1701 			}
1702 			if (rcount) {
1703 				padiff = pa + rcount * PAGE_SIZE - VM_PAGE_TO_PHYS(m);
1704 				padiff >>= PAGE_SHIFT;
1705 				padiff &= PQ_L2_MASK;
1706 				if (padiff == 0) {
1707 					pa = VM_PAGE_TO_PHYS(m) - rcount * PAGE_SIZE;
1708 					++rcount;
1709 					continue;
1710 				}
1711 				db_printf(" index(%d)run(%d)pa(0x%x)", fidx, rcount, pa);
1712 				db_printf("pd(%d)\n", padiff);
1713 				if ( nl > 18) {
1714 					c = cngetc();
1715 					if (c != ' ')
1716 						return;
1717 					nl = 0;
1718 				}
1719 				nl++;
1720 			}
1721 			fidx = idx;
1722 			pa = VM_PAGE_TO_PHYS(m);
1723 			rcount = 1;
1724 		}
1725 		if (rcount) {
1726 			db_printf(" index(%d)run(%d)pa(0x%x)\n", fidx, rcount, pa);
1727 			if ( nl > 18) {
1728 				c = cngetc();
1729 				if (c != ' ')
1730 					return;
1731 				nl = 0;
1732 			}
1733 			nl++;
1734 		}
1735 	}
1736 }
1737 #endif /* DDB */
1738