xref: /freebsd/sys/vm/vm_object.c (revision ce4946daa5ce852d28008dac492029500ab2ee95)
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/lock.h>
74 #include <sys/mman.h>
75 #include <sys/mount.h>
76 #include <sys/mutex.h>
77 #include <sys/proc.h>		/* for curproc, pageproc */
78 #include <sys/socket.h>
79 #include <sys/vnode.h>
80 #include <sys/vmmeter.h>
81 #include <sys/sx.h>
82 
83 #include <vm/vm.h>
84 #include <vm/vm_param.h>
85 #include <vm/pmap.h>
86 #include <vm/vm_map.h>
87 #include <vm/vm_object.h>
88 #include <vm/vm_page.h>
89 #include <vm/vm_pageout.h>
90 #include <vm/vm_pager.h>
91 #include <vm/vm_zone.h>
92 #include <vm/swap_pager.h>
93 #include <vm/vm_kern.h>
94 #include <vm/vm_extern.h>
95 
96 static void	vm_object_qcollapse __P((vm_object_t object));
97 
98 /*
99  *	Virtual memory objects maintain the actual data
100  *	associated with allocated virtual memory.  A given
101  *	page of memory exists within exactly one object.
102  *
103  *	An object is only deallocated when all "references"
104  *	are given up.  Only one "reference" to a given
105  *	region of an object should be writeable.
106  *
107  *	Associated with each object is a list of all resident
108  *	memory pages belonging to that object; this list is
109  *	maintained by the "vm_page" module, and locked by the object's
110  *	lock.
111  *
112  *	Each object also records a "pager" routine which is
113  *	used to retrieve (and store) pages to the proper backing
114  *	storage.  In addition, objects may be backed by other
115  *	objects from which they were virtual-copied.
116  *
117  *	The only items within the object structure which are
118  *	modified after time of creation are:
119  *		reference count		locked by object's lock
120  *		pager routine		locked by object's lock
121  *
122  */
123 
124 struct object_q vm_object_list;
125 static struct mtx vm_object_list_mtx;	/* lock for object list and count */
126 static long vm_object_count;		/* count of all objects */
127 vm_object_t kernel_object;
128 vm_object_t kmem_object;
129 static struct vm_object kernel_object_store;
130 static struct vm_object kmem_object_store;
131 extern int vm_pageout_page_count;
132 
133 static long object_collapses;
134 static long object_bypasses;
135 static int next_index;
136 static vm_zone_t obj_zone;
137 static struct vm_zone obj_zone_store;
138 static int object_hash_rand;
139 #define VM_OBJECTS_INIT 256
140 static struct vm_object vm_objects_init[VM_OBJECTS_INIT];
141 
142 void
143 _vm_object_allocate(type, size, object)
144 	objtype_t type;
145 	vm_size_t size;
146 	vm_object_t object;
147 {
148 	int incr;
149 	TAILQ_INIT(&object->memq);
150 	TAILQ_INIT(&object->shadow_head);
151 
152 	object->type = type;
153 	object->size = size;
154 	object->ref_count = 1;
155 	object->flags = 0;
156 	if ((object->type == OBJT_DEFAULT) || (object->type == OBJT_SWAP))
157 		vm_object_set_flag(object, OBJ_ONEMAPPING);
158 	object->paging_in_progress = 0;
159 	object->resident_page_count = 0;
160 	object->shadow_count = 0;
161 	object->pg_color = next_index;
162 	if ( size > (PQ_L2_SIZE / 3 + PQ_PRIME1))
163 		incr = PQ_L2_SIZE / 3 + PQ_PRIME1;
164 	else
165 		incr = size;
166 	next_index = (next_index + incr) & PQ_L2_MASK;
167 	object->handle = NULL;
168 	object->backing_object = NULL;
169 	object->backing_object_offset = (vm_ooffset_t) 0;
170 	/*
171 	 * Try to generate a number that will spread objects out in the
172 	 * hash table.  We 'wipe' new objects across the hash in 128 page
173 	 * increments plus 1 more to offset it a little more by the time
174 	 * it wraps around.
175 	 */
176 	object->hash_rand = object_hash_rand - 129;
177 
178 	object->generation++;
179 
180 	TAILQ_INSERT_TAIL(&vm_object_list, object, object_list);
181 	vm_object_count++;
182 	object_hash_rand = object->hash_rand;
183 }
184 
185 /*
186  *	vm_object_init:
187  *
188  *	Initialize the VM objects module.
189  */
190 void
191 vm_object_init()
192 {
193 	TAILQ_INIT(&vm_object_list);
194 	mtx_init(&vm_object_list_mtx, "vm object_list", MTX_DEF);
195 	vm_object_count = 0;
196 
197 	kernel_object = &kernel_object_store;
198 	_vm_object_allocate(OBJT_DEFAULT, OFF_TO_IDX(VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS),
199 	    kernel_object);
200 
201 	kmem_object = &kmem_object_store;
202 	_vm_object_allocate(OBJT_DEFAULT, OFF_TO_IDX(VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS),
203 	    kmem_object);
204 
205 	obj_zone = &obj_zone_store;
206 	zbootinit(obj_zone, "VM OBJECT", sizeof (struct vm_object),
207 		vm_objects_init, VM_OBJECTS_INIT);
208 }
209 
210 void
211 vm_object_init2() {
212 	zinitna(obj_zone, NULL, NULL, 0, 0, 0, 1);
213 }
214 
215 /*
216  *	vm_object_allocate:
217  *
218  *	Returns a new object with the given size.
219  */
220 
221 vm_object_t
222 vm_object_allocate(type, size)
223 	objtype_t type;
224 	vm_size_t size;
225 {
226 	vm_object_t result;
227 
228 	result = (vm_object_t) zalloc(obj_zone);
229 
230 	_vm_object_allocate(type, size, result);
231 
232 	return (result);
233 }
234 
235 
236 /*
237  *	vm_object_reference:
238  *
239  *	Gets another reference to the given object.
240  */
241 void
242 vm_object_reference(object)
243 	vm_object_t object;
244 {
245 	if (object == NULL)
246 		return;
247 
248 	KASSERT(!(object->flags & OBJ_DEAD),
249 	    ("vm_object_reference: attempting to reference dead obj"));
250 
251 	object->ref_count++;
252 	if (object->type == OBJT_VNODE) {
253 		while (vget((struct vnode *) object->handle, LK_RETRY|LK_NOOBJ, curproc)) {
254 			printf("vm_object_reference: delay in getting object\n");
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 
265 	KASSERT(object->type == OBJT_VNODE,
266 	    ("vm_object_vndeallocate: not a vnode object"));
267 	KASSERT(vp != NULL, ("vm_object_vndeallocate: missing vp"));
268 #ifdef INVARIANTS
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 	vm_object_t temp;
299 
300 	while (object != NULL) {
301 
302 		if (object->type == OBJT_VNODE) {
303 			vm_object_vndeallocate(object);
304 			return;
305 		}
306 
307 		KASSERT(object->ref_count != 0,
308 			("vm_object_deallocate: object deallocated too many times: %d", object->type));
309 
310 		/*
311 		 * If the reference count goes to 0 we start calling
312 		 * vm_object_terminate() on the object chain.
313 		 * A ref count of 1 may be a special case depending on the
314 		 * shadow count being 0 or 1.
315 		 */
316 		object->ref_count--;
317 		if (object->ref_count > 1) {
318 			return;
319 		} else if (object->ref_count == 1) {
320 			if (object->shadow_count == 0) {
321 				vm_object_set_flag(object, OBJ_ONEMAPPING);
322 			} else if ((object->shadow_count == 1) &&
323 			    (object->handle == NULL) &&
324 			    (object->type == OBJT_DEFAULT ||
325 			     object->type == OBJT_SWAP)) {
326 				vm_object_t robject;
327 
328 				robject = TAILQ_FIRST(&object->shadow_head);
329 				KASSERT(robject != NULL,
330 				    ("vm_object_deallocate: ref_count: %d, shadow_count: %d",
331 					 object->ref_count,
332 					 object->shadow_count));
333 				if ((robject->handle == NULL) &&
334 				    (robject->type == OBJT_DEFAULT ||
335 				     robject->type == OBJT_SWAP)) {
336 
337 					robject->ref_count++;
338 
339 					while (
340 						robject->paging_in_progress ||
341 						object->paging_in_progress
342 					) {
343 						vm_object_pip_sleep(robject, "objde1");
344 						vm_object_pip_sleep(object, "objde2");
345 					}
346 
347 					if (robject->ref_count == 1) {
348 						robject->ref_count--;
349 						object = robject;
350 						goto doterm;
351 					}
352 
353 					object = robject;
354 					vm_object_collapse(object);
355 					continue;
356 				}
357 			}
358 
359 			return;
360 
361 		}
362 
363 doterm:
364 
365 		temp = object->backing_object;
366 		if (temp) {
367 			TAILQ_REMOVE(&temp->shadow_head, object, shadow_list);
368 			temp->shadow_count--;
369 			if (temp->ref_count == 0)
370 				vm_object_clear_flag(temp, OBJ_OPT);
371 			temp->generation++;
372 			object->backing_object = NULL;
373 		}
374 		vm_object_terminate(object);
375 		/* unlocks and deallocates object */
376 		object = temp;
377 	}
378 }
379 
380 /*
381  *	vm_object_terminate actually destroys the specified object, freeing
382  *	up all previously used resources.
383  *
384  *	The object must be locked.
385  *	This routine may block.
386  */
387 void
388 vm_object_terminate(object)
389 	vm_object_t object;
390 {
391 	vm_page_t p;
392 	int s;
393 
394 	/*
395 	 * Make sure no one uses us.
396 	 */
397 	vm_object_set_flag(object, OBJ_DEAD);
398 
399 	/*
400 	 * wait for the pageout daemon to be done with the object
401 	 */
402 	vm_object_pip_wait(object, "objtrm");
403 
404 	KASSERT(!object->paging_in_progress,
405 		("vm_object_terminate: pageout in progress"));
406 
407 	/*
408 	 * Clean and free the pages, as appropriate. All references to the
409 	 * object are gone, so we don't need to lock it.
410 	 */
411 	if (object->type == OBJT_VNODE) {
412 		struct vnode *vp;
413 
414 		/*
415 		 * Freeze optimized copies.
416 		 */
417 		vm_freeze_copyopts(object, 0, object->size);
418 
419 		/*
420 		 * Clean pages and flush buffers.
421 		 */
422 		vm_object_page_clean(object, 0, 0, OBJPC_SYNC);
423 
424 		vp = (struct vnode *) object->handle;
425 		vinvalbuf(vp, V_SAVE, NOCRED, NULL, 0, 0);
426 	}
427 
428 	KASSERT(object->ref_count == 0,
429 		("vm_object_terminate: object with references, ref_count=%d",
430 		object->ref_count));
431 
432 	/*
433 	 * Now free any remaining pages. For internal objects, this also
434 	 * removes them from paging queues. Don't free wired pages, just
435 	 * remove them from the object.
436 	 */
437 	s = splvm();
438 	while ((p = TAILQ_FIRST(&object->memq)) != NULL) {
439 		KASSERT(!p->busy && (p->flags & PG_BUSY) == 0,
440 			("vm_object_terminate: freeing busy page %p "
441 			"p->busy = %d, p->flags %x\n", p, p->busy, p->flags));
442 		if (p->wire_count == 0) {
443 			vm_page_busy(p);
444 			vm_page_free(p);
445 			cnt.v_pfree++;
446 		} else {
447 			vm_page_busy(p);
448 			vm_page_remove(p);
449 		}
450 	}
451 	splx(s);
452 
453 	/*
454 	 * Let the pager know object is dead.
455 	 */
456 	vm_pager_deallocate(object);
457 
458 	/*
459 	 * Remove the object from the global object list.
460 	 */
461 	mtx_lock(&vm_object_list_mtx);
462 	TAILQ_REMOVE(&vm_object_list, object, object_list);
463 	mtx_unlock(&vm_object_list_mtx);
464 
465 	wakeup(object);
466 
467 	/*
468 	 * Free the space for the object.
469 	 */
470 	zfree(obj_zone, object);
471 }
472 
473 /*
474  *	vm_object_page_clean
475  *
476  *	Clean all dirty pages in the specified range of object.  Leaves page
477  * 	on whatever queue it is currently on.   If NOSYNC is set then do not
478  *	write out pages with PG_NOSYNC set (originally comes from MAP_NOSYNC),
479  *	leaving the object dirty.
480  *
481  *	Odd semantics: if start == end, we clean everything.
482  *
483  *	The object must be locked.
484  */
485 
486 void
487 vm_object_page_clean(object, start, end, flags)
488 	vm_object_t object;
489 	vm_pindex_t start;
490 	vm_pindex_t end;
491 	int flags;
492 {
493 	vm_page_t p, np, tp;
494 	vm_offset_t tstart, tend;
495 	vm_pindex_t pi;
496 	int s;
497 	struct vnode *vp;
498 	int runlen;
499 	int maxf;
500 	int chkb;
501 	int maxb;
502 	int i;
503 	int clearobjflags;
504 	int pagerflags;
505 	vm_page_t maf[vm_pageout_page_count];
506 	vm_page_t mab[vm_pageout_page_count];
507 	vm_page_t ma[vm_pageout_page_count];
508 	int curgeneration;
509 
510 	if (object->type != OBJT_VNODE ||
511 		(object->flags & OBJ_MIGHTBEDIRTY) == 0)
512 		return;
513 
514 	pagerflags = (flags & (OBJPC_SYNC | OBJPC_INVAL)) ? VM_PAGER_PUT_SYNC : 0;
515 	pagerflags |= (flags & OBJPC_INVAL) ? VM_PAGER_PUT_INVAL : 0;
516 
517 	vp = object->handle;
518 
519 	vm_object_set_flag(object, OBJ_CLEANING);
520 
521 	tstart = start;
522 	if (end == 0) {
523 		tend = object->size;
524 	} else {
525 		tend = end;
526 	}
527 
528 	/*
529 	 * Generally set CLEANCHK interlock and make the page read-only so
530 	 * we can then clear the object flags.
531 	 *
532 	 * However, if this is a nosync mmap then the object is likely to
533 	 * stay dirty so do not mess with the page and do not clear the
534 	 * object flags.
535 	 */
536 
537 	clearobjflags = 1;
538 
539 	TAILQ_FOREACH(p, &object->memq, listq) {
540 		vm_page_flag_set(p, PG_CLEANCHK);
541 		if ((flags & OBJPC_NOSYNC) && (p->flags & PG_NOSYNC))
542 			clearobjflags = 0;
543 		else
544 			vm_page_protect(p, VM_PROT_READ);
545 	}
546 
547 	if (clearobjflags && (tstart == 0) && (tend == object->size)) {
548 		vm_object_clear_flag(object, OBJ_WRITEABLE|OBJ_MIGHTBEDIRTY);
549 	}
550 
551 rescan:
552 	curgeneration = object->generation;
553 
554 	for(p = TAILQ_FIRST(&object->memq); p; p = np) {
555 		np = TAILQ_NEXT(p, listq);
556 
557 		pi = p->pindex;
558 		if (((p->flags & PG_CLEANCHK) == 0) ||
559 			(pi < tstart) || (pi >= tend) ||
560 			(p->valid == 0) ||
561 			((p->queue - p->pc) == PQ_CACHE)) {
562 			vm_page_flag_clear(p, PG_CLEANCHK);
563 			continue;
564 		}
565 
566 		vm_page_test_dirty(p);
567 		if ((p->dirty & p->valid) == 0) {
568 			vm_page_flag_clear(p, PG_CLEANCHK);
569 			continue;
570 		}
571 
572 		/*
573 		 * If we have been asked to skip nosync pages and this is a
574 		 * nosync page, skip it.  Note that the object flags were
575 		 * not cleared in this case so we do not have to set them.
576 		 */
577 		if ((flags & OBJPC_NOSYNC) && (p->flags & PG_NOSYNC)) {
578 			vm_page_flag_clear(p, PG_CLEANCHK);
579 			continue;
580 		}
581 
582 		s = splvm();
583 		while (vm_page_sleep_busy(p, TRUE, "vpcwai")) {
584 			if (object->generation != curgeneration) {
585 				splx(s);
586 				goto rescan;
587 			}
588 		}
589 
590 		maxf = 0;
591 		for(i=1;i<vm_pageout_page_count;i++) {
592 			if ((tp = vm_page_lookup(object, pi + i)) != NULL) {
593 				if ((tp->flags & PG_BUSY) ||
594 					(tp->flags & PG_CLEANCHK) == 0 ||
595 					(tp->busy != 0))
596 					break;
597 				if((tp->queue - tp->pc) == PQ_CACHE) {
598 					vm_page_flag_clear(tp, PG_CLEANCHK);
599 					break;
600 				}
601 				vm_page_test_dirty(tp);
602 				if ((tp->dirty & tp->valid) == 0) {
603 					vm_page_flag_clear(tp, PG_CLEANCHK);
604 					break;
605 				}
606 				maf[ i - 1 ] = tp;
607 				maxf++;
608 				continue;
609 			}
610 			break;
611 		}
612 
613 		maxb = 0;
614 		chkb = vm_pageout_page_count -  maxf;
615 		if (chkb) {
616 			for(i = 1; i < chkb;i++) {
617 				if ((tp = vm_page_lookup(object, pi - i)) != NULL) {
618 					if ((tp->flags & PG_BUSY) ||
619 						(tp->flags & PG_CLEANCHK) == 0 ||
620 						(tp->busy != 0))
621 						break;
622 					if((tp->queue - tp->pc) == PQ_CACHE) {
623 						vm_page_flag_clear(tp, PG_CLEANCHK);
624 						break;
625 					}
626 					vm_page_test_dirty(tp);
627 					if ((tp->dirty & tp->valid) == 0) {
628 						vm_page_flag_clear(tp, PG_CLEANCHK);
629 						break;
630 					}
631 					mab[ i - 1 ] = tp;
632 					maxb++;
633 					continue;
634 				}
635 				break;
636 			}
637 		}
638 
639 		for(i=0;i<maxb;i++) {
640 			int index = (maxb - i) - 1;
641 			ma[index] = mab[i];
642 			vm_page_flag_clear(ma[index], PG_CLEANCHK);
643 		}
644 		vm_page_flag_clear(p, PG_CLEANCHK);
645 		ma[maxb] = p;
646 		for(i=0;i<maxf;i++) {
647 			int index = (maxb + i) + 1;
648 			ma[index] = maf[i];
649 			vm_page_flag_clear(ma[index], PG_CLEANCHK);
650 		}
651 		runlen = maxb + maxf + 1;
652 
653 		splx(s);
654 		vm_pageout_flush(ma, runlen, pagerflags);
655 		for (i = 0; i<runlen; i++) {
656 			if (ma[i]->valid & ma[i]->dirty) {
657 				vm_page_protect(ma[i], VM_PROT_READ);
658 				vm_page_flag_set(ma[i], PG_CLEANCHK);
659 			}
660 		}
661 		if (object->generation != curgeneration)
662 			goto rescan;
663 	}
664 
665 #if 0
666 	VOP_FSYNC(vp, NULL, (pagerflags & VM_PAGER_PUT_SYNC)?MNT_WAIT:0, curproc);
667 #endif
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 	vm_object_t object;
686 {
687 	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  * Same as vm_object_pmap_copy, except range checking really
698  * works, and is meant for small sections of an object.
699  *
700  * This code protects resident pages by making them read-only
701  * and is typically called on a fork or split when a page
702  * is converted to copy-on-write.
703  *
704  * NOTE: If the page is already at VM_PROT_NONE, calling
705  * vm_page_protect will have no effect.
706  */
707 
708 void
709 vm_object_pmap_copy_1(object, start, end)
710 	vm_object_t object;
711 	vm_pindex_t start;
712 	vm_pindex_t end;
713 {
714 	vm_pindex_t idx;
715 	vm_page_t p;
716 
717 	if (object == NULL || (object->flags & OBJ_WRITEABLE) == 0)
718 		return;
719 
720 	for (idx = start; idx < end; idx++) {
721 		p = vm_page_lookup(object, idx);
722 		if (p == NULL)
723 			continue;
724 		vm_page_protect(p, VM_PROT_READ);
725 	}
726 }
727 
728 /*
729  *	vm_object_pmap_remove:
730  *
731  *	Removes all physical pages in the specified
732  *	object range from all physical maps.
733  *
734  *	The object must *not* be locked.
735  */
736 void
737 vm_object_pmap_remove(object, start, end)
738 	vm_object_t object;
739 	vm_pindex_t start;
740 	vm_pindex_t end;
741 {
742 	vm_page_t p;
743 
744 	if (object == NULL)
745 		return;
746 	TAILQ_FOREACH(p, &object->memq, listq) {
747 		if (p->pindex >= start && p->pindex < end)
748 			vm_page_protect(p, VM_PROT_NONE);
749 	}
750 	if ((start == 0) && (object->size == end))
751 		vm_object_clear_flag(object, OBJ_WRITEABLE);
752 }
753 
754 /*
755  *	vm_object_madvise:
756  *
757  *	Implements the madvise function at the object/page level.
758  *
759  *	MADV_WILLNEED	(any object)
760  *
761  *	    Activate the specified pages if they are resident.
762  *
763  *	MADV_DONTNEED	(any object)
764  *
765  *	    Deactivate the specified pages if they are resident.
766  *
767  *	MADV_FREE	(OBJT_DEFAULT/OBJT_SWAP objects,
768  *			 OBJ_ONEMAPPING only)
769  *
770  *	    Deactivate and clean the specified pages if they are
771  *	    resident.  This permits the process to reuse the pages
772  *	    without faulting or the kernel to reclaim the pages
773  *	    without I/O.
774  */
775 void
776 vm_object_madvise(object, pindex, count, advise)
777 	vm_object_t object;
778 	vm_pindex_t pindex;
779 	int count;
780 	int advise;
781 {
782 	vm_pindex_t end, tpindex;
783 	vm_object_t tobject;
784 	vm_page_t m;
785 
786 	if (object == NULL)
787 		return;
788 
789 	end = pindex + count;
790 
791 	/*
792 	 * Locate and adjust resident pages
793 	 */
794 
795 	for (; pindex < end; pindex += 1) {
796 relookup:
797 		tobject = object;
798 		tpindex = pindex;
799 shadowlookup:
800 		/*
801 		 * MADV_FREE only operates on OBJT_DEFAULT or OBJT_SWAP pages
802 		 * and those pages must be OBJ_ONEMAPPING.
803 		 */
804 		if (advise == MADV_FREE) {
805 			if ((tobject->type != OBJT_DEFAULT &&
806 			     tobject->type != OBJT_SWAP) ||
807 			    (tobject->flags & OBJ_ONEMAPPING) == 0) {
808 				continue;
809 			}
810 		}
811 
812 		m = vm_page_lookup(tobject, tpindex);
813 
814 		if (m == NULL) {
815 			/*
816 			 * There may be swap even if there is no backing page
817 			 */
818 			if (advise == MADV_FREE && tobject->type == OBJT_SWAP)
819 				swap_pager_freespace(tobject, tpindex, 1);
820 
821 			/*
822 			 * next object
823 			 */
824 			tobject = tobject->backing_object;
825 			if (tobject == NULL)
826 				continue;
827 			tpindex += OFF_TO_IDX(tobject->backing_object_offset);
828 			goto shadowlookup;
829 		}
830 
831 		/*
832 		 * If the page is busy or not in a normal active state,
833 		 * we skip it.  If the page is not managed there are no
834 		 * page queues to mess with.  Things can break if we mess
835 		 * with pages in any of the below states.
836 		 */
837 		if (
838 		    m->hold_count ||
839 		    m->wire_count ||
840 		    (m->flags & PG_UNMANAGED) ||
841 		    m->valid != VM_PAGE_BITS_ALL
842 		) {
843 			continue;
844 		}
845 
846  		if (vm_page_sleep_busy(m, TRUE, "madvpo"))
847   			goto relookup;
848 
849 		if (advise == MADV_WILLNEED) {
850 			vm_page_activate(m);
851 		} else if (advise == MADV_DONTNEED) {
852 			vm_page_dontneed(m);
853 		} else if (advise == MADV_FREE) {
854 			/*
855 			 * Mark the page clean.  This will allow the page
856 			 * to be freed up by the system.  However, such pages
857 			 * are often reused quickly by malloc()/free()
858 			 * so we do not do anything that would cause
859 			 * a page fault if we can help it.
860 			 *
861 			 * Specifically, we do not try to actually free
862 			 * the page now nor do we try to put it in the
863 			 * cache (which would cause a page fault on reuse).
864 			 *
865 			 * But we do make the page is freeable as we
866 			 * can without actually taking the step of unmapping
867 			 * it.
868 			 */
869 			pmap_clear_modify(m);
870 			m->dirty = 0;
871 			m->act_count = 0;
872 			vm_page_dontneed(m);
873 			if (tobject->type == OBJT_SWAP)
874 				swap_pager_freespace(tobject, tpindex, 1);
875 		}
876 	}
877 }
878 
879 /*
880  *	vm_object_shadow:
881  *
882  *	Create a new object which is backed by the
883  *	specified existing object range.  The source
884  *	object reference is deallocated.
885  *
886  *	The new object and offset into that object
887  *	are returned in the source parameters.
888  */
889 
890 void
891 vm_object_shadow(object, offset, length)
892 	vm_object_t *object;	/* IN/OUT */
893 	vm_ooffset_t *offset;	/* IN/OUT */
894 	vm_size_t length;
895 {
896 	vm_object_t source;
897 	vm_object_t result;
898 
899 	source = *object;
900 
901 	/*
902 	 * Don't create the new object if the old object isn't shared.
903 	 */
904 
905 	if (source != NULL &&
906 	    source->ref_count == 1 &&
907 	    source->handle == NULL &&
908 	    (source->type == OBJT_DEFAULT ||
909 	     source->type == OBJT_SWAP))
910 		return;
911 
912 	/*
913 	 * Allocate a new object with the given length
914 	 */
915 	result = vm_object_allocate(OBJT_DEFAULT, length);
916 	KASSERT(result != NULL, ("vm_object_shadow: no object for shadowing"));
917 
918 	/*
919 	 * The new object shadows the source object, adding a reference to it.
920 	 * Our caller changes his reference to point to the new object,
921 	 * removing a reference to the source object.  Net result: no change
922 	 * of reference count.
923 	 *
924 	 * Try to optimize the result object's page color when shadowing
925 	 * in order to maintain page coloring consistency in the combined
926 	 * shadowed object.
927 	 */
928 	result->backing_object = source;
929 	if (source) {
930 		TAILQ_INSERT_TAIL(&source->shadow_head, result, shadow_list);
931 		source->shadow_count++;
932 		source->generation++;
933 		result->pg_color = (source->pg_color + OFF_TO_IDX(*offset)) & PQ_L2_MASK;
934 	}
935 
936 	/*
937 	 * Store the offset into the source object, and fix up the offset into
938 	 * the new object.
939 	 */
940 
941 	result->backing_object_offset = *offset;
942 
943 	/*
944 	 * Return the new things
945 	 */
946 
947 	*offset = 0;
948 	*object = result;
949 }
950 
951 #define	OBSC_TEST_ALL_SHADOWED	0x0001
952 #define	OBSC_COLLAPSE_NOWAIT	0x0002
953 #define	OBSC_COLLAPSE_WAIT	0x0004
954 
955 static __inline int
956 vm_object_backing_scan(vm_object_t object, int op)
957 {
958 	int s;
959 	int r = 1;
960 	vm_page_t p;
961 	vm_object_t backing_object;
962 	vm_pindex_t backing_offset_index;
963 
964 	s = splvm();
965 
966 	backing_object = object->backing_object;
967 	backing_offset_index = OFF_TO_IDX(object->backing_object_offset);
968 
969 	/*
970 	 * Initial conditions
971 	 */
972 
973 	if (op & OBSC_TEST_ALL_SHADOWED) {
974 		/*
975 		 * We do not want to have to test for the existence of
976 		 * swap pages in the backing object.  XXX but with the
977 		 * new swapper this would be pretty easy to do.
978 		 *
979 		 * XXX what about anonymous MAP_SHARED memory that hasn't
980 		 * been ZFOD faulted yet?  If we do not test for this, the
981 		 * shadow test may succeed! XXX
982 		 */
983 		if (backing_object->type != OBJT_DEFAULT) {
984 			splx(s);
985 			return(0);
986 		}
987 	}
988 	if (op & OBSC_COLLAPSE_WAIT) {
989 		vm_object_set_flag(backing_object, OBJ_DEAD);
990 	}
991 
992 	/*
993 	 * Our scan
994 	 */
995 
996 	p = TAILQ_FIRST(&backing_object->memq);
997 	while (p) {
998 		vm_page_t next = TAILQ_NEXT(p, listq);
999 		vm_pindex_t new_pindex = p->pindex - backing_offset_index;
1000 
1001 		if (op & OBSC_TEST_ALL_SHADOWED) {
1002 			vm_page_t pp;
1003 
1004 			/*
1005 			 * Ignore pages outside the parent object's range
1006 			 * and outside the parent object's mapping of the
1007 			 * backing object.
1008 			 *
1009 			 * note that we do not busy the backing object's
1010 			 * page.
1011 			 */
1012 
1013 			if (
1014 			    p->pindex < backing_offset_index ||
1015 			    new_pindex >= object->size
1016 			) {
1017 				p = next;
1018 				continue;
1019 			}
1020 
1021 			/*
1022 			 * See if the parent has the page or if the parent's
1023 			 * object pager has the page.  If the parent has the
1024 			 * page but the page is not valid, the parent's
1025 			 * object pager must have the page.
1026 			 *
1027 			 * If this fails, the parent does not completely shadow
1028 			 * the object and we might as well give up now.
1029 			 */
1030 
1031 			pp = vm_page_lookup(object, new_pindex);
1032 			if (
1033 			    (pp == NULL || pp->valid == 0) &&
1034 			    !vm_pager_has_page(object, new_pindex, NULL, NULL)
1035 			) {
1036 				r = 0;
1037 				break;
1038 			}
1039 		}
1040 
1041 		/*
1042 		 * Check for busy page
1043 		 */
1044 
1045 		if (op & (OBSC_COLLAPSE_WAIT | OBSC_COLLAPSE_NOWAIT)) {
1046 			vm_page_t pp;
1047 
1048 			if (op & OBSC_COLLAPSE_NOWAIT) {
1049 				if (
1050 				    (p->flags & PG_BUSY) ||
1051 				    !p->valid ||
1052 				    p->hold_count ||
1053 				    p->wire_count ||
1054 				    p->busy
1055 				) {
1056 					p = next;
1057 					continue;
1058 				}
1059 			} else if (op & OBSC_COLLAPSE_WAIT) {
1060 				if (vm_page_sleep_busy(p, TRUE, "vmocol")) {
1061 					/*
1062 					 * If we slept, anything could have
1063 					 * happened.  Since the object is
1064 					 * marked dead, the backing offset
1065 					 * should not have changed so we
1066 					 * just restart our scan.
1067 					 */
1068 					p = TAILQ_FIRST(&backing_object->memq);
1069 					continue;
1070 				}
1071 			}
1072 
1073 			/*
1074 			 * Busy the page
1075 			 */
1076 			vm_page_busy(p);
1077 
1078 			KASSERT(
1079 			    p->object == backing_object,
1080 			    ("vm_object_qcollapse(): object mismatch")
1081 			);
1082 
1083 			/*
1084 			 * Destroy any associated swap
1085 			 */
1086 			if (backing_object->type == OBJT_SWAP) {
1087 				swap_pager_freespace(
1088 				    backing_object,
1089 				    p->pindex,
1090 				    1
1091 				);
1092 			}
1093 
1094 			if (
1095 			    p->pindex < backing_offset_index ||
1096 			    new_pindex >= object->size
1097 			) {
1098 				/*
1099 				 * Page is out of the parent object's range, we
1100 				 * can simply destroy it.
1101 				 */
1102 				vm_page_protect(p, VM_PROT_NONE);
1103 				vm_page_free(p);
1104 				p = next;
1105 				continue;
1106 			}
1107 
1108 			pp = vm_page_lookup(object, new_pindex);
1109 			if (
1110 			    pp != NULL ||
1111 			    vm_pager_has_page(object, new_pindex, NULL, NULL)
1112 			) {
1113 				/*
1114 				 * page already exists in parent OR swap exists
1115 				 * for this location in the parent.  Destroy
1116 				 * the original page from the backing object.
1117 				 *
1118 				 * Leave the parent's page alone
1119 				 */
1120 				vm_page_protect(p, VM_PROT_NONE);
1121 				vm_page_free(p);
1122 				p = next;
1123 				continue;
1124 			}
1125 
1126 			/*
1127 			 * Page does not exist in parent, rename the
1128 			 * page from the backing object to the main object.
1129 			 *
1130 			 * If the page was mapped to a process, it can remain
1131 			 * mapped through the rename.
1132 			 */
1133 			if ((p->queue - p->pc) == PQ_CACHE)
1134 				vm_page_deactivate(p);
1135 
1136 			vm_page_rename(p, object, new_pindex);
1137 			/* page automatically made dirty by rename */
1138 		}
1139 		p = next;
1140 	}
1141 	splx(s);
1142 	return(r);
1143 }
1144 
1145 
1146 /*
1147  * this version of collapse allows the operation to occur earlier and
1148  * when paging_in_progress is true for an object...  This is not a complete
1149  * operation, but should plug 99.9% of the rest of the leaks.
1150  */
1151 static void
1152 vm_object_qcollapse(object)
1153 	vm_object_t object;
1154 {
1155 	vm_object_t backing_object = object->backing_object;
1156 
1157 	if (backing_object->ref_count != 1)
1158 		return;
1159 
1160 	backing_object->ref_count += 2;
1161 
1162 	vm_object_backing_scan(object, OBSC_COLLAPSE_NOWAIT);
1163 
1164 	backing_object->ref_count -= 2;
1165 }
1166 
1167 /*
1168  *	vm_object_collapse:
1169  *
1170  *	Collapse an object with the object backing it.
1171  *	Pages in the backing object are moved into the
1172  *	parent, and the backing object is deallocated.
1173  */
1174 void
1175 vm_object_collapse(object)
1176 	vm_object_t object;
1177 {
1178 	while (TRUE) {
1179 		vm_object_t backing_object;
1180 
1181 		/*
1182 		 * Verify that the conditions are right for collapse:
1183 		 *
1184 		 * The object exists and the backing object exists.
1185 		 */
1186 		if (object == NULL)
1187 			break;
1188 
1189 		if ((backing_object = object->backing_object) == NULL)
1190 			break;
1191 
1192 		/*
1193 		 * we check the backing object first, because it is most likely
1194 		 * not collapsable.
1195 		 */
1196 		if (backing_object->handle != NULL ||
1197 		    (backing_object->type != OBJT_DEFAULT &&
1198 		     backing_object->type != OBJT_SWAP) ||
1199 		    (backing_object->flags & OBJ_DEAD) ||
1200 		    object->handle != NULL ||
1201 		    (object->type != OBJT_DEFAULT &&
1202 		     object->type != OBJT_SWAP) ||
1203 		    (object->flags & OBJ_DEAD)) {
1204 			break;
1205 		}
1206 
1207 		if (
1208 		    object->paging_in_progress != 0 ||
1209 		    backing_object->paging_in_progress != 0
1210 		) {
1211 			vm_object_qcollapse(object);
1212 			break;
1213 		}
1214 
1215 		/*
1216 		 * We know that we can either collapse the backing object (if
1217 		 * the parent is the only reference to it) or (perhaps) have
1218 		 * the parent bypass the object if the parent happens to shadow
1219 		 * all the resident pages in the entire backing object.
1220 		 *
1221 		 * This is ignoring pager-backed pages such as swap pages.
1222 		 * vm_object_backing_scan fails the shadowing test in this
1223 		 * case.
1224 		 */
1225 
1226 		if (backing_object->ref_count == 1) {
1227 			/*
1228 			 * If there is exactly one reference to the backing
1229 			 * object, we can collapse it into the parent.
1230 			 */
1231 
1232 			vm_object_backing_scan(object, OBSC_COLLAPSE_WAIT);
1233 
1234 			/*
1235 			 * Move the pager from backing_object to object.
1236 			 */
1237 
1238 			if (backing_object->type == OBJT_SWAP) {
1239 				vm_object_pip_add(backing_object, 1);
1240 
1241 				/*
1242 				 * scrap the paging_offset junk and do a
1243 				 * discrete copy.  This also removes major
1244 				 * assumptions about how the swap-pager
1245 				 * works from where it doesn't belong.  The
1246 				 * new swapper is able to optimize the
1247 				 * destroy-source case.
1248 				 */
1249 
1250 				vm_object_pip_add(object, 1);
1251 				swap_pager_copy(
1252 				    backing_object,
1253 				    object,
1254 				    OFF_TO_IDX(object->backing_object_offset), TRUE);
1255 				vm_object_pip_wakeup(object);
1256 
1257 				vm_object_pip_wakeup(backing_object);
1258 			}
1259 			/*
1260 			 * Object now shadows whatever backing_object did.
1261 			 * Note that the reference to
1262 			 * backing_object->backing_object moves from within
1263 			 * backing_object to within object.
1264 			 */
1265 
1266 			TAILQ_REMOVE(
1267 			    &object->backing_object->shadow_head,
1268 			    object,
1269 			    shadow_list
1270 			);
1271 			object->backing_object->shadow_count--;
1272 			object->backing_object->generation++;
1273 			if (backing_object->backing_object) {
1274 				TAILQ_REMOVE(
1275 				    &backing_object->backing_object->shadow_head,
1276 				    backing_object,
1277 				    shadow_list
1278 				);
1279 				backing_object->backing_object->shadow_count--;
1280 				backing_object->backing_object->generation++;
1281 			}
1282 			object->backing_object = backing_object->backing_object;
1283 			if (object->backing_object) {
1284 				TAILQ_INSERT_TAIL(
1285 				    &object->backing_object->shadow_head,
1286 				    object,
1287 				    shadow_list
1288 				);
1289 				object->backing_object->shadow_count++;
1290 				object->backing_object->generation++;
1291 			}
1292 
1293 			object->backing_object_offset +=
1294 			    backing_object->backing_object_offset;
1295 
1296 			/*
1297 			 * Discard backing_object.
1298 			 *
1299 			 * Since the backing object has no pages, no pager left,
1300 			 * and no object references within it, all that is
1301 			 * necessary is to dispose of it.
1302 			 */
1303 
1304 			TAILQ_REMOVE(
1305 			    &vm_object_list,
1306 			    backing_object,
1307 			    object_list
1308 			);
1309 			vm_object_count--;
1310 
1311 			zfree(obj_zone, backing_object);
1312 
1313 			object_collapses++;
1314 		} else {
1315 			vm_object_t new_backing_object;
1316 
1317 			/*
1318 			 * If we do not entirely shadow the backing object,
1319 			 * there is nothing we can do so we give up.
1320 			 */
1321 
1322 			if (vm_object_backing_scan(object, OBSC_TEST_ALL_SHADOWED) == 0) {
1323 				break;
1324 			}
1325 
1326 			/*
1327 			 * Make the parent shadow the next object in the
1328 			 * chain.  Deallocating backing_object will not remove
1329 			 * it, since its reference count is at least 2.
1330 			 */
1331 
1332 			TAILQ_REMOVE(
1333 			    &backing_object->shadow_head,
1334 			    object,
1335 			    shadow_list
1336 			);
1337 			backing_object->shadow_count--;
1338 			backing_object->generation++;
1339 
1340 			new_backing_object = backing_object->backing_object;
1341 			if ((object->backing_object = new_backing_object) != NULL) {
1342 				vm_object_reference(new_backing_object);
1343 				TAILQ_INSERT_TAIL(
1344 				    &new_backing_object->shadow_head,
1345 				    object,
1346 				    shadow_list
1347 				);
1348 				new_backing_object->shadow_count++;
1349 				new_backing_object->generation++;
1350 				object->backing_object_offset +=
1351 					backing_object->backing_object_offset;
1352 			}
1353 
1354 			/*
1355 			 * Drop the reference count on backing_object. Since
1356 			 * its ref_count was at least 2, it will not vanish;
1357 			 * so we don't need to call vm_object_deallocate, but
1358 			 * we do anyway.
1359 			 */
1360 			vm_object_deallocate(backing_object);
1361 			object_bypasses++;
1362 		}
1363 
1364 		/*
1365 		 * Try again with this object's new backing object.
1366 		 */
1367 	}
1368 }
1369 
1370 /*
1371  *	vm_object_page_remove: [internal]
1372  *
1373  *	Removes all physical pages in the specified
1374  *	object range from the object's list of pages.
1375  *
1376  *	The object must be locked.
1377  */
1378 void
1379 vm_object_page_remove(object, start, end, clean_only)
1380 	vm_object_t object;
1381 	vm_pindex_t start;
1382 	vm_pindex_t end;
1383 	boolean_t clean_only;
1384 {
1385 	vm_page_t p, next;
1386 	unsigned int size;
1387 	int all;
1388 
1389 	if (object == NULL ||
1390 	    object->resident_page_count == 0)
1391 		return;
1392 
1393 	all = ((end == 0) && (start == 0));
1394 
1395 	/*
1396 	 * Since physically-backed objects do not use managed pages, we can't
1397 	 * remove pages from the object (we must instead remove the page
1398 	 * references, and then destroy the object).
1399 	 */
1400 	KASSERT(object->type != OBJT_PHYS, ("attempt to remove pages from a physical object"));
1401 
1402 	vm_object_pip_add(object, 1);
1403 again:
1404 	size = end - start;
1405 	if (all || size > object->resident_page_count / 4) {
1406 		for (p = TAILQ_FIRST(&object->memq); p != NULL; p = next) {
1407 			next = TAILQ_NEXT(p, listq);
1408 			if (all || ((start <= p->pindex) && (p->pindex < end))) {
1409 				if (p->wire_count != 0) {
1410 					vm_page_protect(p, VM_PROT_NONE);
1411 					if (!clean_only)
1412 						p->valid = 0;
1413 					continue;
1414 				}
1415 
1416 				/*
1417 				 * The busy flags are only cleared at
1418 				 * interrupt -- minimize the spl transitions
1419 				 */
1420 
1421  				if (vm_page_sleep_busy(p, TRUE, "vmopar"))
1422  					goto again;
1423 
1424 				if (clean_only && p->valid) {
1425 					vm_page_test_dirty(p);
1426 					if (p->valid & p->dirty)
1427 						continue;
1428 				}
1429 
1430 				vm_page_busy(p);
1431 				vm_page_protect(p, VM_PROT_NONE);
1432 				vm_page_free(p);
1433 			}
1434 		}
1435 	} else {
1436 		while (size > 0) {
1437 			if ((p = vm_page_lookup(object, start)) != 0) {
1438 
1439 				if (p->wire_count != 0) {
1440 					vm_page_protect(p, VM_PROT_NONE);
1441 					if (!clean_only)
1442 						p->valid = 0;
1443 					start += 1;
1444 					size -= 1;
1445 					continue;
1446 				}
1447 
1448 				/*
1449 				 * The busy flags are only cleared at
1450 				 * interrupt -- minimize the spl transitions
1451 				 */
1452  				if (vm_page_sleep_busy(p, TRUE, "vmopar"))
1453 					goto again;
1454 
1455 				if (clean_only && p->valid) {
1456 					vm_page_test_dirty(p);
1457 					if (p->valid & p->dirty) {
1458 						start += 1;
1459 						size -= 1;
1460 						continue;
1461 					}
1462 				}
1463 
1464 				vm_page_busy(p);
1465 				vm_page_protect(p, VM_PROT_NONE);
1466 				vm_page_free(p);
1467 			}
1468 			start += 1;
1469 			size -= 1;
1470 		}
1471 	}
1472 	vm_object_pip_wakeup(object);
1473 }
1474 
1475 /*
1476  *	Routine:	vm_object_coalesce
1477  *	Function:	Coalesces two objects backing up adjoining
1478  *			regions of memory into a single object.
1479  *
1480  *	returns TRUE if objects were combined.
1481  *
1482  *	NOTE:	Only works at the moment if the second object is NULL -
1483  *		if it's not, which object do we lock first?
1484  *
1485  *	Parameters:
1486  *		prev_object	First object to coalesce
1487  *		prev_offset	Offset into prev_object
1488  *		next_object	Second object into coalesce
1489  *		next_offset	Offset into next_object
1490  *
1491  *		prev_size	Size of reference to prev_object
1492  *		next_size	Size of reference to next_object
1493  *
1494  *	Conditions:
1495  *	The object must *not* be locked.
1496  */
1497 boolean_t
1498 vm_object_coalesce(prev_object, prev_pindex, prev_size, next_size)
1499 	vm_object_t prev_object;
1500 	vm_pindex_t prev_pindex;
1501 	vm_size_t prev_size, next_size;
1502 {
1503 	vm_pindex_t next_pindex;
1504 
1505 	if (prev_object == NULL) {
1506 		return (TRUE);
1507 	}
1508 
1509 	if (prev_object->type != OBJT_DEFAULT &&
1510 	    prev_object->type != OBJT_SWAP) {
1511 		return (FALSE);
1512 	}
1513 
1514 	/*
1515 	 * Try to collapse the object first
1516 	 */
1517 	vm_object_collapse(prev_object);
1518 
1519 	/*
1520 	 * Can't coalesce if: . more than one reference . paged out . shadows
1521 	 * another object . has a copy elsewhere (any of which mean that the
1522 	 * pages not mapped to prev_entry may be in use anyway)
1523 	 */
1524 
1525 	if (prev_object->backing_object != NULL) {
1526 		return (FALSE);
1527 	}
1528 
1529 	prev_size >>= PAGE_SHIFT;
1530 	next_size >>= PAGE_SHIFT;
1531 	next_pindex = prev_pindex + prev_size;
1532 
1533 	if ((prev_object->ref_count > 1) &&
1534 	    (prev_object->size != next_pindex)) {
1535 		return (FALSE);
1536 	}
1537 
1538 	/*
1539 	 * Remove any pages that may still be in the object from a previous
1540 	 * deallocation.
1541 	 */
1542 	if (next_pindex < prev_object->size) {
1543 		vm_object_page_remove(prev_object,
1544 				      next_pindex,
1545 				      next_pindex + next_size, FALSE);
1546 		if (prev_object->type == OBJT_SWAP)
1547 			swap_pager_freespace(prev_object,
1548 					     next_pindex, next_size);
1549 	}
1550 
1551 	/*
1552 	 * Extend the object if necessary.
1553 	 */
1554 	if (next_pindex + next_size > prev_object->size)
1555 		prev_object->size = next_pindex + next_size;
1556 
1557 	return (TRUE);
1558 }
1559 
1560 #include "opt_ddb.h"
1561 #ifdef DDB
1562 #include <sys/kernel.h>
1563 
1564 #include <sys/cons.h>
1565 
1566 #include <ddb/ddb.h>
1567 
1568 static int	_vm_object_in_map __P((vm_map_t map, vm_object_t object,
1569 				       vm_map_entry_t entry));
1570 static int	vm_object_in_map __P((vm_object_t object));
1571 
1572 static int
1573 _vm_object_in_map(map, object, entry)
1574 	vm_map_t map;
1575 	vm_object_t object;
1576 	vm_map_entry_t entry;
1577 {
1578 	vm_map_t tmpm;
1579 	vm_map_entry_t tmpe;
1580 	vm_object_t obj;
1581 	int entcount;
1582 
1583 	if (map == 0)
1584 		return 0;
1585 
1586 	if (entry == 0) {
1587 		tmpe = map->header.next;
1588 		entcount = map->nentries;
1589 		while (entcount-- && (tmpe != &map->header)) {
1590 			if( _vm_object_in_map(map, object, tmpe)) {
1591 				return 1;
1592 			}
1593 			tmpe = tmpe->next;
1594 		}
1595 	} else if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
1596 		tmpm = entry->object.sub_map;
1597 		tmpe = tmpm->header.next;
1598 		entcount = tmpm->nentries;
1599 		while (entcount-- && tmpe != &tmpm->header) {
1600 			if( _vm_object_in_map(tmpm, object, tmpe)) {
1601 				return 1;
1602 			}
1603 			tmpe = tmpe->next;
1604 		}
1605 	} else if ((obj = entry->object.vm_object) != NULL) {
1606 		for(; obj; obj=obj->backing_object)
1607 			if( obj == object) {
1608 				return 1;
1609 			}
1610 	}
1611 	return 0;
1612 }
1613 
1614 static int
1615 vm_object_in_map( object)
1616 	vm_object_t object;
1617 {
1618 	struct proc *p;
1619 
1620 	sx_slock(&allproc_lock);
1621 	LIST_FOREACH(p, &allproc, p_list) {
1622 		if( !p->p_vmspace /* || (p->p_flag & (P_SYSTEM|P_WEXIT)) */)
1623 			continue;
1624 		if( _vm_object_in_map(&p->p_vmspace->vm_map, object, 0)) {
1625 			sx_sunlock(&allproc_lock);
1626 			return 1;
1627 		}
1628 	}
1629 	sx_sunlock(&allproc_lock);
1630 	if( _vm_object_in_map( kernel_map, object, 0))
1631 		return 1;
1632 	if( _vm_object_in_map( kmem_map, object, 0))
1633 		return 1;
1634 	if( _vm_object_in_map( pager_map, object, 0))
1635 		return 1;
1636 	if( _vm_object_in_map( buffer_map, object, 0))
1637 		return 1;
1638 	if( _vm_object_in_map( mb_map, object, 0))
1639 		return 1;
1640 	return 0;
1641 }
1642 
1643 DB_SHOW_COMMAND(vmochk, vm_object_check)
1644 {
1645 	vm_object_t object;
1646 
1647 	/*
1648 	 * make sure that internal objs are in a map somewhere
1649 	 * and none have zero ref counts.
1650 	 */
1651 	TAILQ_FOREACH(object, &vm_object_list, object_list) {
1652 		if (object->handle == NULL &&
1653 		    (object->type == OBJT_DEFAULT || object->type == OBJT_SWAP)) {
1654 			if (object->ref_count == 0) {
1655 				db_printf("vmochk: internal obj has zero ref count: %ld\n",
1656 					(long)object->size);
1657 			}
1658 			if (!vm_object_in_map(object)) {
1659 				db_printf(
1660 			"vmochk: internal obj is not in a map: "
1661 			"ref: %d, size: %lu: 0x%lx, backing_object: %p\n",
1662 				    object->ref_count, (u_long)object->size,
1663 				    (u_long)object->size,
1664 				    (void *)object->backing_object);
1665 			}
1666 		}
1667 	}
1668 }
1669 
1670 /*
1671  *	vm_object_print:	[ debug ]
1672  */
1673 DB_SHOW_COMMAND(object, vm_object_print_static)
1674 {
1675 	/* XXX convert args. */
1676 	vm_object_t object = (vm_object_t)addr;
1677 	boolean_t full = have_addr;
1678 
1679 	vm_page_t p;
1680 
1681 	/* XXX count is an (unused) arg.  Avoid shadowing it. */
1682 #define	count	was_count
1683 
1684 	int count;
1685 
1686 	if (object == NULL)
1687 		return;
1688 
1689 	db_iprintf(
1690 	    "Object %p: type=%d, size=0x%lx, res=%d, ref=%d, flags=0x%x\n",
1691 	    object, (int)object->type, (u_long)object->size,
1692 	    object->resident_page_count, object->ref_count, object->flags);
1693 	/*
1694 	 * XXX no %qd in kernel.  Truncate object->backing_object_offset.
1695 	 */
1696 	db_iprintf(" sref=%d, backing_object(%d)=(%p)+0x%lx\n",
1697 	    object->shadow_count,
1698 	    object->backing_object ? object->backing_object->ref_count : 0,
1699 	    object->backing_object, (long)object->backing_object_offset);
1700 
1701 	if (!full)
1702 		return;
1703 
1704 	db_indent += 2;
1705 	count = 0;
1706 	TAILQ_FOREACH(p, &object->memq, listq) {
1707 		if (count == 0)
1708 			db_iprintf("memory:=");
1709 		else if (count == 6) {
1710 			db_printf("\n");
1711 			db_iprintf(" ...");
1712 			count = 0;
1713 		} else
1714 			db_printf(",");
1715 		count++;
1716 
1717 		db_printf("(off=0x%lx,page=0x%lx)",
1718 		    (u_long) p->pindex, (u_long) VM_PAGE_TO_PHYS(p));
1719 	}
1720 	if (count != 0)
1721 		db_printf("\n");
1722 	db_indent -= 2;
1723 }
1724 
1725 /* XXX. */
1726 #undef count
1727 
1728 /* XXX need this non-static entry for calling from vm_map_print. */
1729 void
1730 vm_object_print(addr, have_addr, count, modif)
1731         /* db_expr_t */ long addr;
1732 	boolean_t have_addr;
1733 	/* db_expr_t */ long count;
1734 	char *modif;
1735 {
1736 	vm_object_print_static(addr, have_addr, count, modif);
1737 }
1738 
1739 DB_SHOW_COMMAND(vmopag, vm_object_print_pages)
1740 {
1741 	vm_object_t object;
1742 	int nl = 0;
1743 	int c;
1744 
1745 	TAILQ_FOREACH(object, &vm_object_list, object_list) {
1746 		vm_pindex_t idx, fidx;
1747 		vm_pindex_t osize;
1748 		vm_offset_t pa = -1, padiff;
1749 		int rcount;
1750 		vm_page_t m;
1751 
1752 		db_printf("new object: %p\n", (void *)object);
1753 		if ( nl > 18) {
1754 			c = cngetc();
1755 			if (c != ' ')
1756 				return;
1757 			nl = 0;
1758 		}
1759 		nl++;
1760 		rcount = 0;
1761 		fidx = 0;
1762 		osize = object->size;
1763 		if (osize > 128)
1764 			osize = 128;
1765 		for(idx=0;idx<osize;idx++) {
1766 			m = vm_page_lookup(object, idx);
1767 			if (m == NULL) {
1768 				if (rcount) {
1769 					db_printf(" index(%ld)run(%d)pa(0x%lx)\n",
1770 						(long)fidx, rcount, (long)pa);
1771 					if ( nl > 18) {
1772 						c = cngetc();
1773 						if (c != ' ')
1774 							return;
1775 						nl = 0;
1776 					}
1777 					nl++;
1778 					rcount = 0;
1779 				}
1780 				continue;
1781 			}
1782 
1783 
1784 			if (rcount &&
1785 				(VM_PAGE_TO_PHYS(m) == pa + rcount * PAGE_SIZE)) {
1786 				++rcount;
1787 				continue;
1788 			}
1789 			if (rcount) {
1790 				padiff = pa + rcount * PAGE_SIZE - VM_PAGE_TO_PHYS(m);
1791 				padiff >>= PAGE_SHIFT;
1792 				padiff &= PQ_L2_MASK;
1793 				if (padiff == 0) {
1794 					pa = VM_PAGE_TO_PHYS(m) - rcount * PAGE_SIZE;
1795 					++rcount;
1796 					continue;
1797 				}
1798 				db_printf(" index(%ld)run(%d)pa(0x%lx)",
1799 					(long)fidx, rcount, (long)pa);
1800 				db_printf("pd(%ld)\n", (long)padiff);
1801 				if ( nl > 18) {
1802 					c = cngetc();
1803 					if (c != ' ')
1804 						return;
1805 					nl = 0;
1806 				}
1807 				nl++;
1808 			}
1809 			fidx = idx;
1810 			pa = VM_PAGE_TO_PHYS(m);
1811 			rcount = 1;
1812 		}
1813 		if (rcount) {
1814 			db_printf(" index(%ld)run(%d)pa(0x%lx)\n",
1815 				(long)fidx, rcount, (long)pa);
1816 			if ( nl > 18) {
1817 				c = cngetc();
1818 				if (c != ' ')
1819 					return;
1820 				nl = 0;
1821 			}
1822 			nl++;
1823 		}
1824 	}
1825 }
1826 #endif /* DDB */
1827