xref: /freebsd/sys/vm/vm_object.c (revision ea0b1bdbe42f99b52924480e8252689d3981d0b7)
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 
65 /*
66  *	Virtual memory object module.
67  */
68 
69 #include <sys/cdefs.h>
70 __FBSDID("$FreeBSD$");
71 
72 #include <sys/param.h>
73 #include <sys/systm.h>
74 #include <sys/lock.h>
75 #include <sys/mman.h>
76 #include <sys/mount.h>
77 #include <sys/kernel.h>
78 #include <sys/sysctl.h>
79 #include <sys/mutex.h>
80 #include <sys/proc.h>		/* for curproc, pageproc */
81 #include <sys/socket.h>
82 #include <sys/vnode.h>
83 #include <sys/vmmeter.h>
84 #include <sys/sx.h>
85 
86 #include <vm/vm.h>
87 #include <vm/vm_param.h>
88 #include <vm/pmap.h>
89 #include <vm/vm_map.h>
90 #include <vm/vm_object.h>
91 #include <vm/vm_page.h>
92 #include <vm/vm_pageout.h>
93 #include <vm/vm_pager.h>
94 #include <vm/swap_pager.h>
95 #include <vm/vm_kern.h>
96 #include <vm/vm_extern.h>
97 #include <vm/uma.h>
98 
99 #define EASY_SCAN_FACTOR       8
100 
101 #define MSYNC_FLUSH_HARDSEQ	0x01
102 #define MSYNC_FLUSH_SOFTSEQ	0x02
103 
104 /*
105  * msync / VM object flushing optimizations
106  */
107 static int msync_flush_flags = MSYNC_FLUSH_HARDSEQ | MSYNC_FLUSH_SOFTSEQ;
108 SYSCTL_INT(_vm, OID_AUTO, msync_flush_flags,
109         CTLFLAG_RW, &msync_flush_flags, 0, "");
110 
111 static void	vm_object_qcollapse(vm_object_t object);
112 static int	vm_object_page_collect_flush(vm_object_t object, vm_page_t p, int curgeneration, int pagerflags);
113 
114 /*
115  *	Virtual memory objects maintain the actual data
116  *	associated with allocated virtual memory.  A given
117  *	page of memory exists within exactly one object.
118  *
119  *	An object is only deallocated when all "references"
120  *	are given up.  Only one "reference" to a given
121  *	region of an object should be writeable.
122  *
123  *	Associated with each object is a list of all resident
124  *	memory pages belonging to that object; this list is
125  *	maintained by the "vm_page" module, and locked by the object's
126  *	lock.
127  *
128  *	Each object also records a "pager" routine which is
129  *	used to retrieve (and store) pages to the proper backing
130  *	storage.  In addition, objects may be backed by other
131  *	objects from which they were virtual-copied.
132  *
133  *	The only items within the object structure which are
134  *	modified after time of creation are:
135  *		reference count		locked by object's lock
136  *		pager routine		locked by object's lock
137  *
138  */
139 
140 struct object_q vm_object_list;
141 struct mtx vm_object_list_mtx;	/* lock for object list and count */
142 
143 struct vm_object kernel_object_store;
144 struct vm_object kmem_object_store;
145 
146 static long object_collapses;
147 static long object_bypasses;
148 static int next_index;
149 static uma_zone_t obj_zone;
150 #define VM_OBJECTS_INIT 256
151 
152 static void vm_object_zinit(void *mem, int size);
153 
154 #ifdef INVARIANTS
155 static void vm_object_zdtor(void *mem, int size, void *arg);
156 
157 static void
158 vm_object_zdtor(void *mem, int size, void *arg)
159 {
160 	vm_object_t object;
161 
162 	object = (vm_object_t)mem;
163 	KASSERT(object->paging_in_progress == 0,
164 	    ("object %p paging_in_progress = %d",
165 	    object, object->paging_in_progress));
166 	KASSERT(object->resident_page_count == 0,
167 	    ("object %p resident_page_count = %d",
168 	    object, object->resident_page_count));
169 	KASSERT(object->shadow_count == 0,
170 	    ("object %p shadow_count = %d",
171 	    object, object->shadow_count));
172 }
173 #endif
174 
175 static void
176 vm_object_zinit(void *mem, int size)
177 {
178 	vm_object_t object;
179 
180 	object = (vm_object_t)mem;
181 	bzero(&object->mtx, sizeof(object->mtx));
182 	VM_OBJECT_LOCK_INIT(object);
183 
184 	/* These are true for any object that has been freed */
185 	object->paging_in_progress = 0;
186 	object->resident_page_count = 0;
187 	object->shadow_count = 0;
188 }
189 
190 void
191 _vm_object_allocate(objtype_t type, vm_pindex_t size, vm_object_t object)
192 {
193 	int incr;
194 
195 	TAILQ_INIT(&object->memq);
196 	LIST_INIT(&object->shadow_head);
197 
198 	object->root = NULL;
199 	object->type = type;
200 	object->size = size;
201 	object->ref_count = 1;
202 	object->flags = 0;
203 	if ((object->type == OBJT_DEFAULT) || (object->type == OBJT_SWAP))
204 		object->flags = OBJ_ONEMAPPING;
205 	if (size > (PQ_L2_SIZE / 3 + PQ_PRIME1))
206 		incr = PQ_L2_SIZE / 3 + PQ_PRIME1;
207 	else
208 		incr = size;
209 	do
210 		object->pg_color = next_index;
211 	while (!atomic_cmpset_int(&next_index, object->pg_color,
212 				  (object->pg_color + incr) & PQ_L2_MASK));
213 	object->handle = NULL;
214 	object->backing_object = NULL;
215 	object->backing_object_offset = (vm_ooffset_t) 0;
216 
217 	atomic_add_int(&object->generation, 1);
218 
219 	mtx_lock(&vm_object_list_mtx);
220 	TAILQ_INSERT_TAIL(&vm_object_list, object, object_list);
221 	mtx_unlock(&vm_object_list_mtx);
222 }
223 
224 /*
225  *	vm_object_init:
226  *
227  *	Initialize the VM objects module.
228  */
229 void
230 vm_object_init(void)
231 {
232 	TAILQ_INIT(&vm_object_list);
233 	mtx_init(&vm_object_list_mtx, "vm object_list", NULL, MTX_DEF);
234 
235 	VM_OBJECT_LOCK_INIT(&kernel_object_store);
236 	_vm_object_allocate(OBJT_DEFAULT, OFF_TO_IDX(VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS),
237 	    kernel_object);
238 
239 	/*
240 	 * The kmem object's mutex is given a unique name, instead of
241 	 * "vm object", to avoid false reports of lock-order reversal
242 	 * with a system map mutex.
243 	 */
244 	mtx_init(VM_OBJECT_MTX(kmem_object), "kmem object", NULL, MTX_DEF);
245 	_vm_object_allocate(OBJT_DEFAULT, OFF_TO_IDX(VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS),
246 	    kmem_object);
247 
248 	obj_zone = uma_zcreate("VM OBJECT", sizeof (struct vm_object), NULL,
249 #ifdef INVARIANTS
250 	    vm_object_zdtor,
251 #else
252 	    NULL,
253 #endif
254 	    vm_object_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
255 	uma_prealloc(obj_zone, VM_OBJECTS_INIT);
256 }
257 
258 void
259 vm_object_set_flag(vm_object_t object, u_short bits)
260 {
261 	object->flags |= bits;
262 }
263 
264 void
265 vm_object_clear_flag(vm_object_t object, u_short bits)
266 {
267 
268 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
269 	object->flags &= ~bits;
270 }
271 
272 void
273 vm_object_pip_add(vm_object_t object, short i)
274 {
275 
276 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
277 	object->paging_in_progress += i;
278 }
279 
280 void
281 vm_object_pip_subtract(vm_object_t object, short i)
282 {
283 
284 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
285 	object->paging_in_progress -= i;
286 }
287 
288 void
289 vm_object_pip_wakeup(vm_object_t object)
290 {
291 
292 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
293 	object->paging_in_progress--;
294 	if ((object->flags & OBJ_PIPWNT) && object->paging_in_progress == 0) {
295 		vm_object_clear_flag(object, OBJ_PIPWNT);
296 		wakeup(object);
297 	}
298 }
299 
300 void
301 vm_object_pip_wakeupn(vm_object_t object, short i)
302 {
303 
304 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
305 	if (i)
306 		object->paging_in_progress -= i;
307 	if ((object->flags & OBJ_PIPWNT) && object->paging_in_progress == 0) {
308 		vm_object_clear_flag(object, OBJ_PIPWNT);
309 		wakeup(object);
310 	}
311 }
312 
313 void
314 vm_object_pip_wait(vm_object_t object, char *waitid)
315 {
316 
317 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
318 	while (object->paging_in_progress) {
319 		object->flags |= OBJ_PIPWNT;
320 		msleep(object, VM_OBJECT_MTX(object), PVM, waitid, 0);
321 	}
322 }
323 
324 /*
325  *	vm_object_allocate_wait
326  *
327  *	Return a new object with the given size, and give the user the
328  *	option of waiting for it to complete or failing if the needed
329  *	memory isn't available.
330  */
331 vm_object_t
332 vm_object_allocate_wait(objtype_t type, vm_pindex_t size, int flags)
333 {
334 	vm_object_t result;
335 
336 	result = (vm_object_t) uma_zalloc(obj_zone, flags);
337 
338 	if (result != NULL)
339 		_vm_object_allocate(type, size, result);
340 
341 	return (result);
342 }
343 
344 /*
345  *	vm_object_allocate:
346  *
347  *	Returns a new object with the given size.
348  */
349 vm_object_t
350 vm_object_allocate(objtype_t type, vm_pindex_t size)
351 {
352 	return(vm_object_allocate_wait(type, size, M_WAITOK));
353 }
354 
355 
356 /*
357  *	vm_object_reference:
358  *
359  *	Gets another reference to the given object.  Note: OBJ_DEAD
360  *	objects can be referenced during final cleaning.
361  */
362 void
363 vm_object_reference(vm_object_t object)
364 {
365 	if (object == NULL)
366 		return;
367 	if (object != kernel_object &&
368 	    object != kmem_object)
369 		mtx_lock(&Giant);
370 	VM_OBJECT_LOCK(object);
371 	object->ref_count++;
372 	VM_OBJECT_UNLOCK(object);
373 	if (object->type == OBJT_VNODE) {
374 		while (vget((struct vnode *) object->handle, LK_RETRY, curthread)) {
375 			printf("vm_object_reference: delay in getting object\n");
376 		}
377 	}
378 	if (object != kernel_object &&
379 	    object != kmem_object)
380 		mtx_unlock(&Giant);
381 }
382 
383 /*
384  * Handle deallocating an object of type OBJT_VNODE.
385  */
386 void
387 vm_object_vndeallocate(vm_object_t object)
388 {
389 	struct vnode *vp = (struct vnode *) object->handle;
390 
391 	GIANT_REQUIRED;
392 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
393 	KASSERT(object->type == OBJT_VNODE,
394 	    ("vm_object_vndeallocate: not a vnode object"));
395 	KASSERT(vp != NULL, ("vm_object_vndeallocate: missing vp"));
396 #ifdef INVARIANTS
397 	if (object->ref_count == 0) {
398 		vprint("vm_object_vndeallocate", vp);
399 		panic("vm_object_vndeallocate: bad object reference count");
400 	}
401 #endif
402 
403 	object->ref_count--;
404 	if (object->ref_count == 0) {
405 		mp_fixme("Unlocked vflag access.");
406 		vp->v_vflag &= ~VV_TEXT;
407 	}
408 	VM_OBJECT_UNLOCK(object);
409 	/*
410 	 * vrele may need a vop lock
411 	 */
412 	vrele(vp);
413 }
414 
415 /*
416  *	vm_object_deallocate:
417  *
418  *	Release a reference to the specified object,
419  *	gained either through a vm_object_allocate
420  *	or a vm_object_reference call.  When all references
421  *	are gone, storage associated with this object
422  *	may be relinquished.
423  *
424  *	No object may be locked.
425  */
426 void
427 vm_object_deallocate(vm_object_t object)
428 {
429 	vm_object_t temp;
430 
431 	if (object != kmem_object)
432 		mtx_lock(&Giant);
433 	while (object != NULL) {
434 		VM_OBJECT_LOCK(object);
435 		if (object->type == OBJT_VNODE) {
436 			vm_object_vndeallocate(object);
437 			goto done;
438 		}
439 
440 		KASSERT(object->ref_count != 0,
441 			("vm_object_deallocate: object deallocated too many times: %d", object->type));
442 
443 		/*
444 		 * If the reference count goes to 0 we start calling
445 		 * vm_object_terminate() on the object chain.
446 		 * A ref count of 1 may be a special case depending on the
447 		 * shadow count being 0 or 1.
448 		 */
449 		object->ref_count--;
450 		if (object->ref_count > 1) {
451 			VM_OBJECT_UNLOCK(object);
452 			goto done;
453 		} else if (object->ref_count == 1) {
454 			if (object->shadow_count == 0) {
455 				vm_object_set_flag(object, OBJ_ONEMAPPING);
456 			} else if ((object->shadow_count == 1) &&
457 			    (object->handle == NULL) &&
458 			    (object->type == OBJT_DEFAULT ||
459 			     object->type == OBJT_SWAP)) {
460 				vm_object_t robject;
461 
462 				robject = LIST_FIRST(&object->shadow_head);
463 				KASSERT(robject != NULL,
464 				    ("vm_object_deallocate: ref_count: %d, shadow_count: %d",
465 					 object->ref_count,
466 					 object->shadow_count));
467 				if (!VM_OBJECT_TRYLOCK(robject)) {
468 					/*
469 					 * Avoid a potential deadlock.
470 					 */
471 					object->ref_count++;
472 					VM_OBJECT_UNLOCK(object);
473 					continue;
474 				}
475 				if ((robject->handle == NULL) &&
476 				    (robject->type == OBJT_DEFAULT ||
477 				     robject->type == OBJT_SWAP)) {
478 
479 					robject->ref_count++;
480 retry:
481 					if (robject->paging_in_progress) {
482 						VM_OBJECT_UNLOCK(object);
483 						vm_object_pip_wait(robject,
484 						    "objde1");
485 						VM_OBJECT_LOCK(object);
486 						goto retry;
487 					} else if (object->paging_in_progress) {
488 						VM_OBJECT_UNLOCK(robject);
489 						object->flags |= OBJ_PIPWNT;
490 						msleep(object,
491 						    VM_OBJECT_MTX(object),
492 						    PDROP | PVM, "objde2", 0);
493 						VM_OBJECT_LOCK(robject);
494 						VM_OBJECT_LOCK(object);
495 						goto retry;
496 					}
497 					VM_OBJECT_UNLOCK(object);
498 					if (robject->ref_count == 1) {
499 						robject->ref_count--;
500 						object = robject;
501 						goto doterm;
502 					}
503 					object = robject;
504 					vm_object_collapse(object);
505 					VM_OBJECT_UNLOCK(object);
506 					continue;
507 				}
508 				VM_OBJECT_UNLOCK(robject);
509 			}
510 			VM_OBJECT_UNLOCK(object);
511 			goto done;
512 		}
513 doterm:
514 		temp = object->backing_object;
515 		if (temp != NULL) {
516 			VM_OBJECT_LOCK(temp);
517 			LIST_REMOVE(object, shadow_list);
518 			temp->shadow_count--;
519 			temp->generation++;
520 			VM_OBJECT_UNLOCK(temp);
521 			object->backing_object = NULL;
522 		}
523 		/*
524 		 * Don't double-terminate, we could be in a termination
525 		 * recursion due to the terminate having to sync data
526 		 * to disk.
527 		 */
528 		if ((object->flags & OBJ_DEAD) == 0)
529 			vm_object_terminate(object);
530 		else
531 			VM_OBJECT_UNLOCK(object);
532 		object = temp;
533 	}
534 done:
535 	if (object != kmem_object)
536 		mtx_unlock(&Giant);
537 }
538 
539 /*
540  *	vm_object_terminate actually destroys the specified object, freeing
541  *	up all previously used resources.
542  *
543  *	The object must be locked.
544  *	This routine may block.
545  */
546 void
547 vm_object_terminate(vm_object_t object)
548 {
549 	vm_page_t p;
550 	int s;
551 
552 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
553 
554 	/*
555 	 * Make sure no one uses us.
556 	 */
557 	vm_object_set_flag(object, OBJ_DEAD);
558 
559 	/*
560 	 * wait for the pageout daemon to be done with the object
561 	 */
562 	vm_object_pip_wait(object, "objtrm");
563 
564 	KASSERT(!object->paging_in_progress,
565 		("vm_object_terminate: pageout in progress"));
566 
567 	/*
568 	 * Clean and free the pages, as appropriate. All references to the
569 	 * object are gone, so we don't need to lock it.
570 	 */
571 	if (object->type == OBJT_VNODE) {
572 		struct vnode *vp = (struct vnode *)object->handle;
573 
574 		/*
575 		 * Clean pages and flush buffers.
576 		 */
577 		vm_object_page_clean(object, 0, 0, OBJPC_SYNC);
578 		VM_OBJECT_UNLOCK(object);
579 
580 		vinvalbuf(vp, V_SAVE, NOCRED, NULL, 0, 0);
581 
582 		VM_OBJECT_LOCK(object);
583 	}
584 
585 	KASSERT(object->ref_count == 0,
586 		("vm_object_terminate: object with references, ref_count=%d",
587 		object->ref_count));
588 
589 	/*
590 	 * Now free any remaining pages. For internal objects, this also
591 	 * removes them from paging queues. Don't free wired pages, just
592 	 * remove them from the object.
593 	 */
594 	s = splvm();
595 	vm_page_lock_queues();
596 	while ((p = TAILQ_FIRST(&object->memq)) != NULL) {
597 		KASSERT(!p->busy && (p->flags & PG_BUSY) == 0,
598 			("vm_object_terminate: freeing busy page %p "
599 			"p->busy = %d, p->flags %x\n", p, p->busy, p->flags));
600 		if (p->wire_count == 0) {
601 			vm_page_busy(p);
602 			vm_page_free(p);
603 			cnt.v_pfree++;
604 		} else {
605 			vm_page_busy(p);
606 			vm_page_remove(p);
607 		}
608 	}
609 	vm_page_unlock_queues();
610 	splx(s);
611 
612 	/*
613 	 * Let the pager know object is dead.
614 	 */
615 	vm_pager_deallocate(object);
616 	VM_OBJECT_UNLOCK(object);
617 
618 	/*
619 	 * Remove the object from the global object list.
620 	 */
621 	mtx_lock(&vm_object_list_mtx);
622 	TAILQ_REMOVE(&vm_object_list, object, object_list);
623 	mtx_unlock(&vm_object_list_mtx);
624 
625 	wakeup(object);
626 
627 	/*
628 	 * Free the space for the object.
629 	 */
630 	uma_zfree(obj_zone, object);
631 }
632 
633 /*
634  *	vm_object_page_clean
635  *
636  *	Clean all dirty pages in the specified range of object.  Leaves page
637  * 	on whatever queue it is currently on.   If NOSYNC is set then do not
638  *	write out pages with PG_NOSYNC set (originally comes from MAP_NOSYNC),
639  *	leaving the object dirty.
640  *
641  *	When stuffing pages asynchronously, allow clustering.  XXX we need a
642  *	synchronous clustering mode implementation.
643  *
644  *	Odd semantics: if start == end, we clean everything.
645  *
646  *	The object must be locked.
647  */
648 void
649 vm_object_page_clean(vm_object_t object, vm_pindex_t start, vm_pindex_t end, int flags)
650 {
651 	vm_page_t p, np;
652 	vm_pindex_t tstart, tend;
653 	vm_pindex_t pi;
654 	int clearobjflags;
655 	int pagerflags;
656 	int curgeneration;
657 
658 	GIANT_REQUIRED;
659 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
660 	if (object->type != OBJT_VNODE ||
661 		(object->flags & OBJ_MIGHTBEDIRTY) == 0)
662 		return;
663 
664 	pagerflags = (flags & (OBJPC_SYNC | OBJPC_INVAL)) ? VM_PAGER_PUT_SYNC : VM_PAGER_CLUSTER_OK;
665 	pagerflags |= (flags & OBJPC_INVAL) ? VM_PAGER_PUT_INVAL : 0;
666 
667 	vm_object_set_flag(object, OBJ_CLEANING);
668 
669 	tstart = start;
670 	if (end == 0) {
671 		tend = object->size;
672 	} else {
673 		tend = end;
674 	}
675 
676 	vm_page_lock_queues();
677 	/*
678 	 * If the caller is smart and only msync()s a range he knows is
679 	 * dirty, we may be able to avoid an object scan.  This results in
680 	 * a phenominal improvement in performance.  We cannot do this
681 	 * as a matter of course because the object may be huge - e.g.
682 	 * the size might be in the gigabytes or terrabytes.
683 	 */
684 	if (msync_flush_flags & MSYNC_FLUSH_HARDSEQ) {
685 		vm_pindex_t tscan;
686 		int scanlimit;
687 		int scanreset;
688 
689 		scanreset = object->resident_page_count / EASY_SCAN_FACTOR;
690 		if (scanreset < 16)
691 			scanreset = 16;
692 		pagerflags |= VM_PAGER_IGNORE_CLEANCHK;
693 
694 		scanlimit = scanreset;
695 		tscan = tstart;
696 		while (tscan < tend) {
697 			curgeneration = object->generation;
698 			p = vm_page_lookup(object, tscan);
699 			if (p == NULL || p->valid == 0 ||
700 			    (p->queue - p->pc) == PQ_CACHE) {
701 				if (--scanlimit == 0)
702 					break;
703 				++tscan;
704 				continue;
705 			}
706 			vm_page_test_dirty(p);
707 			if ((p->dirty & p->valid) == 0) {
708 				if (--scanlimit == 0)
709 					break;
710 				++tscan;
711 				continue;
712 			}
713 			/*
714 			 * If we have been asked to skip nosync pages and
715 			 * this is a nosync page, we can't continue.
716 			 */
717 			if ((flags & OBJPC_NOSYNC) && (p->flags & PG_NOSYNC)) {
718 				if (--scanlimit == 0)
719 					break;
720 				++tscan;
721 				continue;
722 			}
723 			scanlimit = scanreset;
724 
725 			/*
726 			 * This returns 0 if it was unable to busy the first
727 			 * page (i.e. had to sleep).
728 			 */
729 			tscan += vm_object_page_collect_flush(object, p, curgeneration, pagerflags);
730 		}
731 
732 		/*
733 		 * If everything was dirty and we flushed it successfully,
734 		 * and the requested range is not the entire object, we
735 		 * don't have to mess with CLEANCHK or MIGHTBEDIRTY and can
736 		 * return immediately.
737 		 */
738 		if (tscan >= tend && (tstart || tend < object->size)) {
739 			vm_page_unlock_queues();
740 			vm_object_clear_flag(object, OBJ_CLEANING);
741 			return;
742 		}
743 		pagerflags &= ~VM_PAGER_IGNORE_CLEANCHK;
744 	}
745 
746 	/*
747 	 * Generally set CLEANCHK interlock and make the page read-only so
748 	 * we can then clear the object flags.
749 	 *
750 	 * However, if this is a nosync mmap then the object is likely to
751 	 * stay dirty so do not mess with the page and do not clear the
752 	 * object flags.
753 	 */
754 	clearobjflags = 1;
755 	TAILQ_FOREACH(p, &object->memq, listq) {
756 		vm_page_flag_set(p, PG_CLEANCHK);
757 		if ((flags & OBJPC_NOSYNC) && (p->flags & PG_NOSYNC))
758 			clearobjflags = 0;
759 		else
760 			pmap_page_protect(p, VM_PROT_READ);
761 	}
762 
763 	if (clearobjflags && (tstart == 0) && (tend == object->size)) {
764 		struct vnode *vp;
765 
766 		vm_object_clear_flag(object, OBJ_WRITEABLE|OBJ_MIGHTBEDIRTY);
767 		if (object->type == OBJT_VNODE &&
768 		    (vp = (struct vnode *)object->handle) != NULL) {
769 			VI_LOCK(vp);
770 			if (vp->v_iflag & VI_OBJDIRTY)
771 				vp->v_iflag &= ~VI_OBJDIRTY;
772 			VI_UNLOCK(vp);
773 		}
774 	}
775 
776 rescan:
777 	curgeneration = object->generation;
778 
779 	for (p = TAILQ_FIRST(&object->memq); p; p = np) {
780 		int n;
781 
782 		np = TAILQ_NEXT(p, listq);
783 
784 again:
785 		pi = p->pindex;
786 		if (((p->flags & PG_CLEANCHK) == 0) ||
787 			(pi < tstart) || (pi >= tend) ||
788 			(p->valid == 0) ||
789 			((p->queue - p->pc) == PQ_CACHE)) {
790 			vm_page_flag_clear(p, PG_CLEANCHK);
791 			continue;
792 		}
793 
794 		vm_page_test_dirty(p);
795 		if ((p->dirty & p->valid) == 0) {
796 			vm_page_flag_clear(p, PG_CLEANCHK);
797 			continue;
798 		}
799 
800 		/*
801 		 * If we have been asked to skip nosync pages and this is a
802 		 * nosync page, skip it.  Note that the object flags were
803 		 * not cleared in this case so we do not have to set them.
804 		 */
805 		if ((flags & OBJPC_NOSYNC) && (p->flags & PG_NOSYNC)) {
806 			vm_page_flag_clear(p, PG_CLEANCHK);
807 			continue;
808 		}
809 
810 		n = vm_object_page_collect_flush(object, p,
811 			curgeneration, pagerflags);
812 		if (n == 0)
813 			goto rescan;
814 
815 		if (object->generation != curgeneration)
816 			goto rescan;
817 
818 		/*
819 		 * Try to optimize the next page.  If we can't we pick up
820 		 * our (random) scan where we left off.
821 		 */
822 		if (msync_flush_flags & MSYNC_FLUSH_SOFTSEQ) {
823 			if ((p = vm_page_lookup(object, pi + n)) != NULL)
824 				goto again;
825 		}
826 	}
827 	vm_page_unlock_queues();
828 #if 0
829 	VOP_FSYNC(vp, NULL, (pagerflags & VM_PAGER_PUT_SYNC)?MNT_WAIT:0, curproc);
830 #endif
831 
832 	vm_object_clear_flag(object, OBJ_CLEANING);
833 	return;
834 }
835 
836 static int
837 vm_object_page_collect_flush(vm_object_t object, vm_page_t p, int curgeneration, int pagerflags)
838 {
839 	int runlen;
840 	int s;
841 	int maxf;
842 	int chkb;
843 	int maxb;
844 	int i;
845 	vm_pindex_t pi;
846 	vm_page_t maf[vm_pageout_page_count];
847 	vm_page_t mab[vm_pageout_page_count];
848 	vm_page_t ma[vm_pageout_page_count];
849 
850 	s = splvm();
851 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
852 	pi = p->pindex;
853 	while (vm_page_sleep_if_busy(p, TRUE, "vpcwai")) {
854 		vm_page_lock_queues();
855 		if (object->generation != curgeneration) {
856 			splx(s);
857 			return(0);
858 		}
859 	}
860 	maxf = 0;
861 	for(i = 1; i < vm_pageout_page_count; i++) {
862 		vm_page_t tp;
863 
864 		if ((tp = vm_page_lookup(object, pi + i)) != NULL) {
865 			if ((tp->flags & PG_BUSY) ||
866 				((pagerflags & VM_PAGER_IGNORE_CLEANCHK) == 0 &&
867 				 (tp->flags & PG_CLEANCHK) == 0) ||
868 				(tp->busy != 0))
869 				break;
870 			if((tp->queue - tp->pc) == PQ_CACHE) {
871 				vm_page_flag_clear(tp, PG_CLEANCHK);
872 				break;
873 			}
874 			vm_page_test_dirty(tp);
875 			if ((tp->dirty & tp->valid) == 0) {
876 				vm_page_flag_clear(tp, PG_CLEANCHK);
877 				break;
878 			}
879 			maf[ i - 1 ] = tp;
880 			maxf++;
881 			continue;
882 		}
883 		break;
884 	}
885 
886 	maxb = 0;
887 	chkb = vm_pageout_page_count -  maxf;
888 	if (chkb) {
889 		for(i = 1; i < chkb;i++) {
890 			vm_page_t tp;
891 
892 			if ((tp = vm_page_lookup(object, pi - i)) != NULL) {
893 				if ((tp->flags & PG_BUSY) ||
894 					((pagerflags & VM_PAGER_IGNORE_CLEANCHK) == 0 &&
895 					 (tp->flags & PG_CLEANCHK) == 0) ||
896 					(tp->busy != 0))
897 					break;
898 				if ((tp->queue - tp->pc) == PQ_CACHE) {
899 					vm_page_flag_clear(tp, PG_CLEANCHK);
900 					break;
901 				}
902 				vm_page_test_dirty(tp);
903 				if ((tp->dirty & tp->valid) == 0) {
904 					vm_page_flag_clear(tp, PG_CLEANCHK);
905 					break;
906 				}
907 				mab[ i - 1 ] = tp;
908 				maxb++;
909 				continue;
910 			}
911 			break;
912 		}
913 	}
914 
915 	for(i = 0; i < maxb; i++) {
916 		int index = (maxb - i) - 1;
917 		ma[index] = mab[i];
918 		vm_page_flag_clear(ma[index], PG_CLEANCHK);
919 	}
920 	vm_page_flag_clear(p, PG_CLEANCHK);
921 	ma[maxb] = p;
922 	for(i = 0; i < maxf; i++) {
923 		int index = (maxb + i) + 1;
924 		ma[index] = maf[i];
925 		vm_page_flag_clear(ma[index], PG_CLEANCHK);
926 	}
927 	runlen = maxb + maxf + 1;
928 
929 	splx(s);
930 	vm_pageout_flush(ma, runlen, pagerflags, TRUE);
931 	for (i = 0; i < runlen; i++) {
932 		if (ma[i]->valid & ma[i]->dirty) {
933 			pmap_page_protect(ma[i], VM_PROT_READ);
934 			vm_page_flag_set(ma[i], PG_CLEANCHK);
935 
936 			/*
937 			 * maxf will end up being the actual number of pages
938 			 * we wrote out contiguously, non-inclusive of the
939 			 * first page.  We do not count look-behind pages.
940 			 */
941 			if (i >= maxb + 1 && (maxf > i - maxb - 1))
942 				maxf = i - maxb - 1;
943 		}
944 	}
945 	return(maxf + 1);
946 }
947 
948 /*
949  *	vm_object_madvise:
950  *
951  *	Implements the madvise function at the object/page level.
952  *
953  *	MADV_WILLNEED	(any object)
954  *
955  *	    Activate the specified pages if they are resident.
956  *
957  *	MADV_DONTNEED	(any object)
958  *
959  *	    Deactivate the specified pages if they are resident.
960  *
961  *	MADV_FREE	(OBJT_DEFAULT/OBJT_SWAP objects,
962  *			 OBJ_ONEMAPPING only)
963  *
964  *	    Deactivate and clean the specified pages if they are
965  *	    resident.  This permits the process to reuse the pages
966  *	    without faulting or the kernel to reclaim the pages
967  *	    without I/O.
968  */
969 void
970 vm_object_madvise(vm_object_t object, vm_pindex_t pindex, int count, int advise)
971 {
972 	vm_pindex_t end, tpindex;
973 	vm_object_t backing_object, tobject;
974 	vm_page_t m;
975 
976 	if (object == NULL)
977 		return;
978 
979 	mtx_lock(&Giant);
980 
981 	end = pindex + count;
982 
983 	/*
984 	 * Locate and adjust resident pages
985 	 */
986 	for (; pindex < end; pindex += 1) {
987 relookup:
988 		tobject = object;
989 		tpindex = pindex;
990 		VM_OBJECT_LOCK(tobject);
991 shadowlookup:
992 		/*
993 		 * MADV_FREE only operates on OBJT_DEFAULT or OBJT_SWAP pages
994 		 * and those pages must be OBJ_ONEMAPPING.
995 		 */
996 		if (advise == MADV_FREE) {
997 			if ((tobject->type != OBJT_DEFAULT &&
998 			     tobject->type != OBJT_SWAP) ||
999 			    (tobject->flags & OBJ_ONEMAPPING) == 0) {
1000 				goto unlock_tobject;
1001 			}
1002 		}
1003 
1004 		m = vm_page_lookup(tobject, tpindex);
1005 
1006 		if (m == NULL) {
1007 			/*
1008 			 * There may be swap even if there is no backing page
1009 			 */
1010 			if (advise == MADV_FREE && tobject->type == OBJT_SWAP)
1011 				swap_pager_freespace(tobject, tpindex, 1);
1012 
1013 			/*
1014 			 * next object
1015 			 */
1016 			backing_object = tobject->backing_object;
1017 			if (backing_object == NULL)
1018 				goto unlock_tobject;
1019 			VM_OBJECT_LOCK(backing_object);
1020 			VM_OBJECT_UNLOCK(tobject);
1021 			tobject = backing_object;
1022 			tpindex += OFF_TO_IDX(tobject->backing_object_offset);
1023 			goto shadowlookup;
1024 		}
1025 
1026 		/*
1027 		 * If the page is busy or not in a normal active state,
1028 		 * we skip it.  If the page is not managed there are no
1029 		 * page queues to mess with.  Things can break if we mess
1030 		 * with pages in any of the below states.
1031 		 */
1032 		vm_page_lock_queues();
1033 		if (m->hold_count ||
1034 		    m->wire_count ||
1035 		    (m->flags & PG_UNMANAGED) ||
1036 		    m->valid != VM_PAGE_BITS_ALL) {
1037 			vm_page_unlock_queues();
1038 			goto unlock_tobject;
1039 		}
1040  		if (vm_page_sleep_if_busy(m, TRUE, "madvpo")) {
1041 			VM_OBJECT_UNLOCK(tobject);
1042   			goto relookup;
1043 		}
1044 		if (advise == MADV_WILLNEED) {
1045 			vm_page_activate(m);
1046 		} else if (advise == MADV_DONTNEED) {
1047 			vm_page_dontneed(m);
1048 		} else if (advise == MADV_FREE) {
1049 			/*
1050 			 * Mark the page clean.  This will allow the page
1051 			 * to be freed up by the system.  However, such pages
1052 			 * are often reused quickly by malloc()/free()
1053 			 * so we do not do anything that would cause
1054 			 * a page fault if we can help it.
1055 			 *
1056 			 * Specifically, we do not try to actually free
1057 			 * the page now nor do we try to put it in the
1058 			 * cache (which would cause a page fault on reuse).
1059 			 *
1060 			 * But we do make the page is freeable as we
1061 			 * can without actually taking the step of unmapping
1062 			 * it.
1063 			 */
1064 			pmap_clear_modify(m);
1065 			m->dirty = 0;
1066 			m->act_count = 0;
1067 			vm_page_dontneed(m);
1068 		}
1069 		vm_page_unlock_queues();
1070 		if (advise == MADV_FREE && tobject->type == OBJT_SWAP)
1071 			swap_pager_freespace(tobject, tpindex, 1);
1072 unlock_tobject:
1073 		VM_OBJECT_UNLOCK(tobject);
1074 	}
1075 	mtx_unlock(&Giant);
1076 }
1077 
1078 /*
1079  *	vm_object_shadow:
1080  *
1081  *	Create a new object which is backed by the
1082  *	specified existing object range.  The source
1083  *	object reference is deallocated.
1084  *
1085  *	The new object and offset into that object
1086  *	are returned in the source parameters.
1087  */
1088 void
1089 vm_object_shadow(
1090 	vm_object_t *object,	/* IN/OUT */
1091 	vm_ooffset_t *offset,	/* IN/OUT */
1092 	vm_size_t length)
1093 {
1094 	vm_object_t source;
1095 	vm_object_t result;
1096 
1097 	GIANT_REQUIRED;
1098 
1099 	source = *object;
1100 
1101 	/*
1102 	 * Don't create the new object if the old object isn't shared.
1103 	 */
1104 	if (source != NULL) {
1105 		VM_OBJECT_LOCK(source);
1106 		if (source->ref_count == 1 &&
1107 		    source->handle == NULL &&
1108 		    (source->type == OBJT_DEFAULT ||
1109 		     source->type == OBJT_SWAP)) {
1110 			VM_OBJECT_UNLOCK(source);
1111 			return;
1112 		}
1113 		VM_OBJECT_UNLOCK(source);
1114 	}
1115 
1116 	/*
1117 	 * Allocate a new object with the given length.
1118 	 */
1119 	result = vm_object_allocate(OBJT_DEFAULT, length);
1120 
1121 	/*
1122 	 * The new object shadows the source object, adding a reference to it.
1123 	 * Our caller changes his reference to point to the new object,
1124 	 * removing a reference to the source object.  Net result: no change
1125 	 * of reference count.
1126 	 *
1127 	 * Try to optimize the result object's page color when shadowing
1128 	 * in order to maintain page coloring consistency in the combined
1129 	 * shadowed object.
1130 	 */
1131 	result->backing_object = source;
1132 	if (source != NULL) {
1133 		VM_OBJECT_LOCK(source);
1134 		LIST_INSERT_HEAD(&source->shadow_head, result, shadow_list);
1135 		source->shadow_count++;
1136 		source->generation++;
1137 		if (length < source->size)
1138 			length = source->size;
1139 		if (length > PQ_L2_SIZE / 3 + PQ_PRIME1 ||
1140 		    source->generation > 1)
1141 			length = PQ_L2_SIZE / 3 + PQ_PRIME1;
1142 		result->pg_color = (source->pg_color +
1143 		    length * source->generation) & PQ_L2_MASK;
1144 		VM_OBJECT_UNLOCK(source);
1145 		next_index = (result->pg_color + PQ_L2_SIZE / 3 + PQ_PRIME1) &
1146 		    PQ_L2_MASK;
1147 	}
1148 
1149 	/*
1150 	 * Store the offset into the source object, and fix up the offset into
1151 	 * the new object.
1152 	 */
1153 	result->backing_object_offset = *offset;
1154 
1155 	/*
1156 	 * Return the new things
1157 	 */
1158 	*offset = 0;
1159 	*object = result;
1160 }
1161 
1162 /*
1163  *	vm_object_split:
1164  *
1165  * Split the pages in a map entry into a new object.  This affords
1166  * easier removal of unused pages, and keeps object inheritance from
1167  * being a negative impact on memory usage.
1168  */
1169 void
1170 vm_object_split(vm_map_entry_t entry)
1171 {
1172 	vm_page_t m;
1173 	vm_object_t orig_object, new_object, source;
1174 	vm_offset_t s, e;
1175 	vm_pindex_t offidxstart, offidxend;
1176 	vm_size_t idx, size;
1177 	vm_ooffset_t offset;
1178 
1179 	GIANT_REQUIRED;
1180 
1181 	orig_object = entry->object.vm_object;
1182 	if (orig_object->type != OBJT_DEFAULT && orig_object->type != OBJT_SWAP)
1183 		return;
1184 	if (orig_object->ref_count <= 1)
1185 		return;
1186 
1187 	offset = entry->offset;
1188 	s = entry->start;
1189 	e = entry->end;
1190 
1191 	offidxstart = OFF_TO_IDX(offset);
1192 	offidxend = offidxstart + OFF_TO_IDX(e - s);
1193 	size = offidxend - offidxstart;
1194 
1195 	new_object = vm_pager_allocate(orig_object->type,
1196 		NULL, IDX_TO_OFF(size), VM_PROT_ALL, 0LL);
1197 	if (new_object == NULL)
1198 		return;
1199 
1200 	source = orig_object->backing_object;
1201 	if (source != NULL) {
1202 		vm_object_reference(source);	/* Referenced by new_object */
1203 		VM_OBJECT_LOCK(source);
1204 		LIST_INSERT_HEAD(&source->shadow_head,
1205 				  new_object, shadow_list);
1206 		source->shadow_count++;
1207 		source->generation++;
1208 		vm_object_clear_flag(source, OBJ_ONEMAPPING);
1209 		VM_OBJECT_UNLOCK(source);
1210 		new_object->backing_object_offset =
1211 			orig_object->backing_object_offset + offset;
1212 		new_object->backing_object = source;
1213 	}
1214 	VM_OBJECT_LOCK(orig_object);
1215 	for (idx = 0; idx < size; idx++) {
1216 	retry:
1217 		m = vm_page_lookup(orig_object, offidxstart + idx);
1218 		if (m == NULL)
1219 			continue;
1220 
1221 		/*
1222 		 * We must wait for pending I/O to complete before we can
1223 		 * rename the page.
1224 		 *
1225 		 * We do not have to VM_PROT_NONE the page as mappings should
1226 		 * not be changed by this operation.
1227 		 */
1228 		vm_page_lock_queues();
1229 		if (vm_page_sleep_if_busy(m, TRUE, "spltwt"))
1230 			goto retry;
1231 
1232 		vm_page_busy(m);
1233 		vm_page_rename(m, new_object, idx);
1234 		/* page automatically made dirty by rename and cache handled */
1235 		vm_page_busy(m);
1236 		vm_page_unlock_queues();
1237 	}
1238 	if (orig_object->type == OBJT_SWAP) {
1239 		vm_object_pip_add(orig_object, 1);
1240 		VM_OBJECT_UNLOCK(orig_object);
1241 		/*
1242 		 * copy orig_object pages into new_object
1243 		 * and destroy unneeded pages in
1244 		 * shadow object.
1245 		 */
1246 		swap_pager_copy(orig_object, new_object, offidxstart, 0);
1247 		VM_OBJECT_LOCK(orig_object);
1248 		vm_object_pip_wakeup(orig_object);
1249 	}
1250 	VM_OBJECT_UNLOCK(orig_object);
1251 	vm_page_lock_queues();
1252 	TAILQ_FOREACH(m, &new_object->memq, listq)
1253 		vm_page_wakeup(m);
1254 	vm_page_unlock_queues();
1255 	entry->object.vm_object = new_object;
1256 	entry->offset = 0LL;
1257 	vm_object_deallocate(orig_object);
1258 }
1259 
1260 #define	OBSC_TEST_ALL_SHADOWED	0x0001
1261 #define	OBSC_COLLAPSE_NOWAIT	0x0002
1262 #define	OBSC_COLLAPSE_WAIT	0x0004
1263 
1264 static int
1265 vm_object_backing_scan(vm_object_t object, int op)
1266 {
1267 	int s;
1268 	int r = 1;
1269 	vm_page_t p;
1270 	vm_object_t backing_object;
1271 	vm_pindex_t backing_offset_index;
1272 
1273 	s = splvm();
1274 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
1275 	VM_OBJECT_LOCK_ASSERT(object->backing_object, MA_OWNED);
1276 
1277 	backing_object = object->backing_object;
1278 	backing_offset_index = OFF_TO_IDX(object->backing_object_offset);
1279 
1280 	/*
1281 	 * Initial conditions
1282 	 */
1283 	if (op & OBSC_TEST_ALL_SHADOWED) {
1284 		/*
1285 		 * We do not want to have to test for the existence of
1286 		 * swap pages in the backing object.  XXX but with the
1287 		 * new swapper this would be pretty easy to do.
1288 		 *
1289 		 * XXX what about anonymous MAP_SHARED memory that hasn't
1290 		 * been ZFOD faulted yet?  If we do not test for this, the
1291 		 * shadow test may succeed! XXX
1292 		 */
1293 		if (backing_object->type != OBJT_DEFAULT) {
1294 			splx(s);
1295 			return (0);
1296 		}
1297 	}
1298 	if (op & OBSC_COLLAPSE_WAIT) {
1299 		vm_object_set_flag(backing_object, OBJ_DEAD);
1300 	}
1301 
1302 	/*
1303 	 * Our scan
1304 	 */
1305 	p = TAILQ_FIRST(&backing_object->memq);
1306 	while (p) {
1307 		vm_page_t next = TAILQ_NEXT(p, listq);
1308 		vm_pindex_t new_pindex = p->pindex - backing_offset_index;
1309 
1310 		if (op & OBSC_TEST_ALL_SHADOWED) {
1311 			vm_page_t pp;
1312 
1313 			/*
1314 			 * Ignore pages outside the parent object's range
1315 			 * and outside the parent object's mapping of the
1316 			 * backing object.
1317 			 *
1318 			 * note that we do not busy the backing object's
1319 			 * page.
1320 			 */
1321 			if (
1322 			    p->pindex < backing_offset_index ||
1323 			    new_pindex >= object->size
1324 			) {
1325 				p = next;
1326 				continue;
1327 			}
1328 
1329 			/*
1330 			 * See if the parent has the page or if the parent's
1331 			 * object pager has the page.  If the parent has the
1332 			 * page but the page is not valid, the parent's
1333 			 * object pager must have the page.
1334 			 *
1335 			 * If this fails, the parent does not completely shadow
1336 			 * the object and we might as well give up now.
1337 			 */
1338 
1339 			pp = vm_page_lookup(object, new_pindex);
1340 			if (
1341 			    (pp == NULL || pp->valid == 0) &&
1342 			    !vm_pager_has_page(object, new_pindex, NULL, NULL)
1343 			) {
1344 				r = 0;
1345 				break;
1346 			}
1347 		}
1348 
1349 		/*
1350 		 * Check for busy page
1351 		 */
1352 		if (op & (OBSC_COLLAPSE_WAIT | OBSC_COLLAPSE_NOWAIT)) {
1353 			vm_page_t pp;
1354 
1355 			vm_page_lock_queues();
1356 			if (op & OBSC_COLLAPSE_NOWAIT) {
1357 				if ((p->flags & PG_BUSY) ||
1358 				    !p->valid ||
1359 				    p->hold_count ||
1360 				    p->wire_count ||
1361 				    p->busy) {
1362 					vm_page_unlock_queues();
1363 					p = next;
1364 					continue;
1365 				}
1366 			} else if (op & OBSC_COLLAPSE_WAIT) {
1367 				if ((p->flags & PG_BUSY) || p->busy) {
1368 					vm_page_flag_set(p,
1369 					    PG_WANTED | PG_REFERENCED);
1370 					VM_OBJECT_UNLOCK(backing_object);
1371 					VM_OBJECT_UNLOCK(object);
1372 					msleep(p, &vm_page_queue_mtx,
1373 					    PDROP | PVM, "vmocol", 0);
1374 					VM_OBJECT_LOCK(object);
1375 					VM_OBJECT_LOCK(backing_object);
1376 					/*
1377 					 * If we slept, anything could have
1378 					 * happened.  Since the object is
1379 					 * marked dead, the backing offset
1380 					 * should not have changed so we
1381 					 * just restart our scan.
1382 					 */
1383 					p = TAILQ_FIRST(&backing_object->memq);
1384 					continue;
1385 				}
1386 			}
1387 
1388 			/*
1389 			 * Busy the page
1390 			 */
1391 			vm_page_busy(p);
1392 			vm_page_unlock_queues();
1393 
1394 			KASSERT(
1395 			    p->object == backing_object,
1396 			    ("vm_object_qcollapse(): object mismatch")
1397 			);
1398 
1399 			/*
1400 			 * Destroy any associated swap
1401 			 */
1402 			if (backing_object->type == OBJT_SWAP) {
1403 				swap_pager_freespace(
1404 				    backing_object,
1405 				    p->pindex,
1406 				    1
1407 				);
1408 			}
1409 
1410 			if (
1411 			    p->pindex < backing_offset_index ||
1412 			    new_pindex >= object->size
1413 			) {
1414 				/*
1415 				 * Page is out of the parent object's range, we
1416 				 * can simply destroy it.
1417 				 */
1418 				vm_page_lock_queues();
1419 				pmap_remove_all(p);
1420 				vm_page_free(p);
1421 				vm_page_unlock_queues();
1422 				p = next;
1423 				continue;
1424 			}
1425 
1426 			pp = vm_page_lookup(object, new_pindex);
1427 			if (
1428 			    pp != NULL ||
1429 			    vm_pager_has_page(object, new_pindex, NULL, NULL)
1430 			) {
1431 				/*
1432 				 * page already exists in parent OR swap exists
1433 				 * for this location in the parent.  Destroy
1434 				 * the original page from the backing object.
1435 				 *
1436 				 * Leave the parent's page alone
1437 				 */
1438 				vm_page_lock_queues();
1439 				pmap_remove_all(p);
1440 				vm_page_free(p);
1441 				vm_page_unlock_queues();
1442 				p = next;
1443 				continue;
1444 			}
1445 
1446 			/*
1447 			 * Page does not exist in parent, rename the
1448 			 * page from the backing object to the main object.
1449 			 *
1450 			 * If the page was mapped to a process, it can remain
1451 			 * mapped through the rename.
1452 			 */
1453 			vm_page_lock_queues();
1454 			vm_page_rename(p, object, new_pindex);
1455 			vm_page_unlock_queues();
1456 			/* page automatically made dirty by rename */
1457 		}
1458 		p = next;
1459 	}
1460 	splx(s);
1461 	return (r);
1462 }
1463 
1464 
1465 /*
1466  * this version of collapse allows the operation to occur earlier and
1467  * when paging_in_progress is true for an object...  This is not a complete
1468  * operation, but should plug 99.9% of the rest of the leaks.
1469  */
1470 static void
1471 vm_object_qcollapse(vm_object_t object)
1472 {
1473 	vm_object_t backing_object = object->backing_object;
1474 
1475 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
1476 	VM_OBJECT_LOCK_ASSERT(backing_object, MA_OWNED);
1477 
1478 	if (backing_object->ref_count != 1)
1479 		return;
1480 
1481 	backing_object->ref_count += 2;
1482 
1483 	vm_object_backing_scan(object, OBSC_COLLAPSE_NOWAIT);
1484 
1485 	backing_object->ref_count -= 2;
1486 }
1487 
1488 /*
1489  *	vm_object_collapse:
1490  *
1491  *	Collapse an object with the object backing it.
1492  *	Pages in the backing object are moved into the
1493  *	parent, and the backing object is deallocated.
1494  */
1495 void
1496 vm_object_collapse(vm_object_t object)
1497 {
1498 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
1499 
1500 	while (TRUE) {
1501 		vm_object_t backing_object;
1502 
1503 		/*
1504 		 * Verify that the conditions are right for collapse:
1505 		 *
1506 		 * The object exists and the backing object exists.
1507 		 */
1508 		if ((backing_object = object->backing_object) == NULL)
1509 			break;
1510 
1511 		/*
1512 		 * we check the backing object first, because it is most likely
1513 		 * not collapsable.
1514 		 */
1515 		VM_OBJECT_LOCK(backing_object);
1516 		if (backing_object->handle != NULL ||
1517 		    (backing_object->type != OBJT_DEFAULT &&
1518 		     backing_object->type != OBJT_SWAP) ||
1519 		    (backing_object->flags & OBJ_DEAD) ||
1520 		    object->handle != NULL ||
1521 		    (object->type != OBJT_DEFAULT &&
1522 		     object->type != OBJT_SWAP) ||
1523 		    (object->flags & OBJ_DEAD)) {
1524 			VM_OBJECT_UNLOCK(backing_object);
1525 			break;
1526 		}
1527 
1528 		if (
1529 		    object->paging_in_progress != 0 ||
1530 		    backing_object->paging_in_progress != 0
1531 		) {
1532 			vm_object_qcollapse(object);
1533 			VM_OBJECT_UNLOCK(backing_object);
1534 			break;
1535 		}
1536 		/*
1537 		 * We know that we can either collapse the backing object (if
1538 		 * the parent is the only reference to it) or (perhaps) have
1539 		 * the parent bypass the object if the parent happens to shadow
1540 		 * all the resident pages in the entire backing object.
1541 		 *
1542 		 * This is ignoring pager-backed pages such as swap pages.
1543 		 * vm_object_backing_scan fails the shadowing test in this
1544 		 * case.
1545 		 */
1546 		if (backing_object->ref_count == 1) {
1547 			/*
1548 			 * If there is exactly one reference to the backing
1549 			 * object, we can collapse it into the parent.
1550 			 */
1551 			vm_object_backing_scan(object, OBSC_COLLAPSE_WAIT);
1552 
1553 			/*
1554 			 * Move the pager from backing_object to object.
1555 			 */
1556 			if (backing_object->type == OBJT_SWAP) {
1557 				vm_object_pip_add(backing_object, 1);
1558 				VM_OBJECT_UNLOCK(backing_object);
1559 				/*
1560 				 * scrap the paging_offset junk and do a
1561 				 * discrete copy.  This also removes major
1562 				 * assumptions about how the swap-pager
1563 				 * works from where it doesn't belong.  The
1564 				 * new swapper is able to optimize the
1565 				 * destroy-source case.
1566 				 */
1567 				vm_object_pip_add(object, 1);
1568 				VM_OBJECT_UNLOCK(object);
1569 				swap_pager_copy(
1570 				    backing_object,
1571 				    object,
1572 				    OFF_TO_IDX(object->backing_object_offset), TRUE);
1573 				VM_OBJECT_LOCK(object);
1574 				vm_object_pip_wakeup(object);
1575 
1576 				VM_OBJECT_LOCK(backing_object);
1577 				vm_object_pip_wakeup(backing_object);
1578 			}
1579 			/*
1580 			 * Object now shadows whatever backing_object did.
1581 			 * Note that the reference to
1582 			 * backing_object->backing_object moves from within
1583 			 * backing_object to within object.
1584 			 */
1585 			LIST_REMOVE(object, shadow_list);
1586 			backing_object->shadow_count--;
1587 			backing_object->generation++;
1588 			if (backing_object->backing_object) {
1589 				VM_OBJECT_LOCK(backing_object->backing_object);
1590 				LIST_REMOVE(backing_object, shadow_list);
1591 				backing_object->backing_object->shadow_count--;
1592 				backing_object->backing_object->generation++;
1593 				VM_OBJECT_UNLOCK(backing_object->backing_object);
1594 			}
1595 			object->backing_object = backing_object->backing_object;
1596 			if (object->backing_object) {
1597 				VM_OBJECT_LOCK(object->backing_object);
1598 				LIST_INSERT_HEAD(
1599 				    &object->backing_object->shadow_head,
1600 				    object,
1601 				    shadow_list
1602 				);
1603 				object->backing_object->shadow_count++;
1604 				object->backing_object->generation++;
1605 				VM_OBJECT_UNLOCK(object->backing_object);
1606 			}
1607 
1608 			object->backing_object_offset +=
1609 			    backing_object->backing_object_offset;
1610 /* XXX */		VM_OBJECT_UNLOCK(object);
1611 
1612 			/*
1613 			 * Discard backing_object.
1614 			 *
1615 			 * Since the backing object has no pages, no pager left,
1616 			 * and no object references within it, all that is
1617 			 * necessary is to dispose of it.
1618 			 */
1619 			KASSERT(backing_object->ref_count == 1, ("backing_object %p was somehow re-referenced during collapse!", backing_object));
1620 			KASSERT(TAILQ_FIRST(&backing_object->memq) == NULL, ("backing_object %p somehow has left over pages during collapse!", backing_object));
1621 			VM_OBJECT_UNLOCK(backing_object);
1622 
1623 			mtx_lock(&vm_object_list_mtx);
1624 			TAILQ_REMOVE(
1625 			    &vm_object_list,
1626 			    backing_object,
1627 			    object_list
1628 			);
1629 			mtx_unlock(&vm_object_list_mtx);
1630 
1631 			uma_zfree(obj_zone, backing_object);
1632 
1633 			object_collapses++;
1634 		} else {
1635 			vm_object_t new_backing_object;
1636 
1637 			/*
1638 			 * If we do not entirely shadow the backing object,
1639 			 * there is nothing we can do so we give up.
1640 			 */
1641 			if (vm_object_backing_scan(object, OBSC_TEST_ALL_SHADOWED) == 0) {
1642 				VM_OBJECT_UNLOCK(backing_object);
1643 				break;
1644 			}
1645 
1646 			/*
1647 			 * Make the parent shadow the next object in the
1648 			 * chain.  Deallocating backing_object will not remove
1649 			 * it, since its reference count is at least 2.
1650 			 */
1651 			LIST_REMOVE(object, shadow_list);
1652 			backing_object->shadow_count--;
1653 			backing_object->generation++;
1654 			VM_OBJECT_UNLOCK(backing_object);
1655 /* XXX */		VM_OBJECT_UNLOCK(object);
1656 
1657 			new_backing_object = backing_object->backing_object;
1658 			if ((object->backing_object = new_backing_object) != NULL) {
1659 				vm_object_reference(new_backing_object);
1660 				VM_OBJECT_LOCK(new_backing_object);
1661 				LIST_INSERT_HEAD(
1662 				    &new_backing_object->shadow_head,
1663 				    object,
1664 				    shadow_list
1665 				);
1666 				new_backing_object->shadow_count++;
1667 				new_backing_object->generation++;
1668 				VM_OBJECT_UNLOCK(new_backing_object);
1669 				object->backing_object_offset +=
1670 					backing_object->backing_object_offset;
1671 			}
1672 
1673 			/*
1674 			 * Drop the reference count on backing_object. Since
1675 			 * its ref_count was at least 2, it will not vanish;
1676 			 * so we don't need to call vm_object_deallocate, but
1677 			 * we do anyway.
1678 			 */
1679 			vm_object_deallocate(backing_object);
1680 			object_bypasses++;
1681 		}
1682 
1683 		/*
1684 		 * Try again with this object's new backing object.
1685 		 */
1686 /* XXX */	VM_OBJECT_LOCK(object);
1687 	}
1688 }
1689 
1690 /*
1691  *	vm_object_page_remove:
1692  *
1693  *	Removes all physical pages in the given range from the
1694  *	object's list of pages.  If the range's end is zero, all
1695  *	physical pages from the range's start to the end of the object
1696  *	are deleted.
1697  *
1698  *	The object must be locked.
1699  */
1700 void
1701 vm_object_page_remove(vm_object_t object, vm_pindex_t start, vm_pindex_t end,
1702     boolean_t clean_only)
1703 {
1704 	vm_page_t p, next;
1705 
1706 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
1707 	if (object->resident_page_count == 0)
1708 		return;
1709 
1710 	/*
1711 	 * Since physically-backed objects do not use managed pages, we can't
1712 	 * remove pages from the object (we must instead remove the page
1713 	 * references, and then destroy the object).
1714 	 */
1715 	KASSERT(object->type != OBJT_PHYS,
1716 	    ("attempt to remove pages from a physical object"));
1717 
1718 	vm_object_pip_add(object, 1);
1719 again:
1720 	vm_page_lock_queues();
1721 	if ((p = TAILQ_FIRST(&object->memq)) != NULL) {
1722 		if (p->pindex < start) {
1723 			p = vm_page_splay(start, object->root);
1724 			if ((object->root = p)->pindex < start)
1725 				p = TAILQ_NEXT(p, listq);
1726 		}
1727 	}
1728 	/*
1729 	 * Assert: the variable p is either (1) the page with the
1730 	 * least pindex greater than or equal to the parameter pindex
1731 	 * or (2) NULL.
1732 	 */
1733 	for (;
1734 	     p != NULL && (p->pindex < end || end == 0);
1735 	     p = next) {
1736 		next = TAILQ_NEXT(p, listq);
1737 
1738 		if (p->wire_count != 0) {
1739 			pmap_remove_all(p);
1740 			if (!clean_only)
1741 				p->valid = 0;
1742 			continue;
1743 		}
1744 		if (vm_page_sleep_if_busy(p, TRUE, "vmopar"))
1745 			goto again;
1746 		if (clean_only && p->valid) {
1747 			vm_page_test_dirty(p);
1748 			if (p->valid & p->dirty)
1749 				continue;
1750 		}
1751 		vm_page_busy(p);
1752 		pmap_remove_all(p);
1753 		vm_page_free(p);
1754 	}
1755 	vm_page_unlock_queues();
1756 	vm_object_pip_wakeup(object);
1757 }
1758 
1759 /*
1760  *	Routine:	vm_object_coalesce
1761  *	Function:	Coalesces two objects backing up adjoining
1762  *			regions of memory into a single object.
1763  *
1764  *	returns TRUE if objects were combined.
1765  *
1766  *	NOTE:	Only works at the moment if the second object is NULL -
1767  *		if it's not, which object do we lock first?
1768  *
1769  *	Parameters:
1770  *		prev_object	First object to coalesce
1771  *		prev_offset	Offset into prev_object
1772  *		next_object	Second object into coalesce
1773  *		next_offset	Offset into next_object
1774  *
1775  *		prev_size	Size of reference to prev_object
1776  *		next_size	Size of reference to next_object
1777  *
1778  *	Conditions:
1779  *	The object must *not* be locked.
1780  */
1781 boolean_t
1782 vm_object_coalesce(vm_object_t prev_object, vm_pindex_t prev_pindex,
1783 	vm_size_t prev_size, vm_size_t next_size)
1784 {
1785 	vm_pindex_t next_pindex;
1786 
1787 	if (prev_object == NULL)
1788 		return (TRUE);
1789 	mtx_lock(&Giant);
1790 	VM_OBJECT_LOCK(prev_object);
1791 	if (prev_object->type != OBJT_DEFAULT &&
1792 	    prev_object->type != OBJT_SWAP) {
1793 		VM_OBJECT_UNLOCK(prev_object);
1794 		mtx_unlock(&Giant);
1795 		return (FALSE);
1796 	}
1797 
1798 	/*
1799 	 * Try to collapse the object first
1800 	 */
1801 	vm_object_collapse(prev_object);
1802 
1803 	/*
1804 	 * Can't coalesce if: . more than one reference . paged out . shadows
1805 	 * another object . has a copy elsewhere (any of which mean that the
1806 	 * pages not mapped to prev_entry may be in use anyway)
1807 	 */
1808 	if (prev_object->backing_object != NULL) {
1809 		VM_OBJECT_UNLOCK(prev_object);
1810 		mtx_unlock(&Giant);
1811 		return (FALSE);
1812 	}
1813 
1814 	prev_size >>= PAGE_SHIFT;
1815 	next_size >>= PAGE_SHIFT;
1816 	next_pindex = prev_pindex + prev_size;
1817 
1818 	if ((prev_object->ref_count > 1) &&
1819 	    (prev_object->size != next_pindex)) {
1820 		VM_OBJECT_UNLOCK(prev_object);
1821 		mtx_unlock(&Giant);
1822 		return (FALSE);
1823 	}
1824 
1825 	/*
1826 	 * Remove any pages that may still be in the object from a previous
1827 	 * deallocation.
1828 	 */
1829 	if (next_pindex < prev_object->size) {
1830 		vm_object_page_remove(prev_object,
1831 				      next_pindex,
1832 				      next_pindex + next_size, FALSE);
1833 		if (prev_object->type == OBJT_SWAP)
1834 			swap_pager_freespace(prev_object,
1835 					     next_pindex, next_size);
1836 	}
1837 
1838 	/*
1839 	 * Extend the object if necessary.
1840 	 */
1841 	if (next_pindex + next_size > prev_object->size)
1842 		prev_object->size = next_pindex + next_size;
1843 
1844 	VM_OBJECT_UNLOCK(prev_object);
1845 	mtx_unlock(&Giant);
1846 	return (TRUE);
1847 }
1848 
1849 void
1850 vm_object_set_writeable_dirty(vm_object_t object)
1851 {
1852 	struct vnode *vp;
1853 
1854 	vm_object_set_flag(object, OBJ_WRITEABLE|OBJ_MIGHTBEDIRTY);
1855 	if (object->type == OBJT_VNODE &&
1856 	    (vp = (struct vnode *)object->handle) != NULL) {
1857 		VI_LOCK(vp);
1858 		if ((vp->v_iflag & VI_OBJDIRTY) == 0)
1859 			vp->v_iflag |= VI_OBJDIRTY;
1860 		VI_UNLOCK(vp);
1861 	}
1862 }
1863 
1864 #include "opt_ddb.h"
1865 #ifdef DDB
1866 #include <sys/kernel.h>
1867 
1868 #include <sys/cons.h>
1869 
1870 #include <ddb/ddb.h>
1871 
1872 static int
1873 _vm_object_in_map(vm_map_t map, vm_object_t object, vm_map_entry_t entry)
1874 {
1875 	vm_map_t tmpm;
1876 	vm_map_entry_t tmpe;
1877 	vm_object_t obj;
1878 	int entcount;
1879 
1880 	if (map == 0)
1881 		return 0;
1882 
1883 	if (entry == 0) {
1884 		tmpe = map->header.next;
1885 		entcount = map->nentries;
1886 		while (entcount-- && (tmpe != &map->header)) {
1887 			if (_vm_object_in_map(map, object, tmpe)) {
1888 				return 1;
1889 			}
1890 			tmpe = tmpe->next;
1891 		}
1892 	} else if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
1893 		tmpm = entry->object.sub_map;
1894 		tmpe = tmpm->header.next;
1895 		entcount = tmpm->nentries;
1896 		while (entcount-- && tmpe != &tmpm->header) {
1897 			if (_vm_object_in_map(tmpm, object, tmpe)) {
1898 				return 1;
1899 			}
1900 			tmpe = tmpe->next;
1901 		}
1902 	} else if ((obj = entry->object.vm_object) != NULL) {
1903 		for (; obj; obj = obj->backing_object)
1904 			if (obj == object) {
1905 				return 1;
1906 			}
1907 	}
1908 	return 0;
1909 }
1910 
1911 static int
1912 vm_object_in_map(vm_object_t object)
1913 {
1914 	struct proc *p;
1915 
1916 	/* sx_slock(&allproc_lock); */
1917 	LIST_FOREACH(p, &allproc, p_list) {
1918 		if (!p->p_vmspace /* || (p->p_flag & (P_SYSTEM|P_WEXIT)) */)
1919 			continue;
1920 		if (_vm_object_in_map(&p->p_vmspace->vm_map, object, 0)) {
1921 			/* sx_sunlock(&allproc_lock); */
1922 			return 1;
1923 		}
1924 	}
1925 	/* sx_sunlock(&allproc_lock); */
1926 	if (_vm_object_in_map(kernel_map, object, 0))
1927 		return 1;
1928 	if (_vm_object_in_map(kmem_map, object, 0))
1929 		return 1;
1930 	if (_vm_object_in_map(pager_map, object, 0))
1931 		return 1;
1932 	if (_vm_object_in_map(buffer_map, object, 0))
1933 		return 1;
1934 	return 0;
1935 }
1936 
1937 DB_SHOW_COMMAND(vmochk, vm_object_check)
1938 {
1939 	vm_object_t object;
1940 
1941 	/*
1942 	 * make sure that internal objs are in a map somewhere
1943 	 * and none have zero ref counts.
1944 	 */
1945 	TAILQ_FOREACH(object, &vm_object_list, object_list) {
1946 		if (object->handle == NULL &&
1947 		    (object->type == OBJT_DEFAULT || object->type == OBJT_SWAP)) {
1948 			if (object->ref_count == 0) {
1949 				db_printf("vmochk: internal obj has zero ref count: %ld\n",
1950 					(long)object->size);
1951 			}
1952 			if (!vm_object_in_map(object)) {
1953 				db_printf(
1954 			"vmochk: internal obj is not in a map: "
1955 			"ref: %d, size: %lu: 0x%lx, backing_object: %p\n",
1956 				    object->ref_count, (u_long)object->size,
1957 				    (u_long)object->size,
1958 				    (void *)object->backing_object);
1959 			}
1960 		}
1961 	}
1962 }
1963 
1964 /*
1965  *	vm_object_print:	[ debug ]
1966  */
1967 DB_SHOW_COMMAND(object, vm_object_print_static)
1968 {
1969 	/* XXX convert args. */
1970 	vm_object_t object = (vm_object_t)addr;
1971 	boolean_t full = have_addr;
1972 
1973 	vm_page_t p;
1974 
1975 	/* XXX count is an (unused) arg.  Avoid shadowing it. */
1976 #define	count	was_count
1977 
1978 	int count;
1979 
1980 	if (object == NULL)
1981 		return;
1982 
1983 	db_iprintf(
1984 	    "Object %p: type=%d, size=0x%jx, res=%d, ref=%d, flags=0x%x\n",
1985 	    object, (int)object->type, (uintmax_t)object->size,
1986 	    object->resident_page_count, object->ref_count, object->flags);
1987 	db_iprintf(" sref=%d, backing_object(%d)=(%p)+0x%jx\n",
1988 	    object->shadow_count,
1989 	    object->backing_object ? object->backing_object->ref_count : 0,
1990 	    object->backing_object, (uintmax_t)object->backing_object_offset);
1991 
1992 	if (!full)
1993 		return;
1994 
1995 	db_indent += 2;
1996 	count = 0;
1997 	TAILQ_FOREACH(p, &object->memq, listq) {
1998 		if (count == 0)
1999 			db_iprintf("memory:=");
2000 		else if (count == 6) {
2001 			db_printf("\n");
2002 			db_iprintf(" ...");
2003 			count = 0;
2004 		} else
2005 			db_printf(",");
2006 		count++;
2007 
2008 		db_printf("(off=0x%jx,page=0x%jx)",
2009 		    (uintmax_t)p->pindex, (uintmax_t)VM_PAGE_TO_PHYS(p));
2010 	}
2011 	if (count != 0)
2012 		db_printf("\n");
2013 	db_indent -= 2;
2014 }
2015 
2016 /* XXX. */
2017 #undef count
2018 
2019 /* XXX need this non-static entry for calling from vm_map_print. */
2020 void
2021 vm_object_print(
2022         /* db_expr_t */ long addr,
2023 	boolean_t have_addr,
2024 	/* db_expr_t */ long count,
2025 	char *modif)
2026 {
2027 	vm_object_print_static(addr, have_addr, count, modif);
2028 }
2029 
2030 DB_SHOW_COMMAND(vmopag, vm_object_print_pages)
2031 {
2032 	vm_object_t object;
2033 	int nl = 0;
2034 	int c;
2035 
2036 	TAILQ_FOREACH(object, &vm_object_list, object_list) {
2037 		vm_pindex_t idx, fidx;
2038 		vm_pindex_t osize;
2039 		vm_paddr_t pa = -1, padiff;
2040 		int rcount;
2041 		vm_page_t m;
2042 
2043 		db_printf("new object: %p\n", (void *)object);
2044 		if (nl > 18) {
2045 			c = cngetc();
2046 			if (c != ' ')
2047 				return;
2048 			nl = 0;
2049 		}
2050 		nl++;
2051 		rcount = 0;
2052 		fidx = 0;
2053 		osize = object->size;
2054 		if (osize > 128)
2055 			osize = 128;
2056 		for (idx = 0; idx < osize; idx++) {
2057 			m = vm_page_lookup(object, idx);
2058 			if (m == NULL) {
2059 				if (rcount) {
2060 					db_printf(" index(%ld)run(%d)pa(0x%lx)\n",
2061 						(long)fidx, rcount, (long)pa);
2062 					if (nl > 18) {
2063 						c = cngetc();
2064 						if (c != ' ')
2065 							return;
2066 						nl = 0;
2067 					}
2068 					nl++;
2069 					rcount = 0;
2070 				}
2071 				continue;
2072 			}
2073 
2074 
2075 			if (rcount &&
2076 				(VM_PAGE_TO_PHYS(m) == pa + rcount * PAGE_SIZE)) {
2077 				++rcount;
2078 				continue;
2079 			}
2080 			if (rcount) {
2081 				padiff = pa + rcount * PAGE_SIZE - VM_PAGE_TO_PHYS(m);
2082 				padiff >>= PAGE_SHIFT;
2083 				padiff &= PQ_L2_MASK;
2084 				if (padiff == 0) {
2085 					pa = VM_PAGE_TO_PHYS(m) - rcount * PAGE_SIZE;
2086 					++rcount;
2087 					continue;
2088 				}
2089 				db_printf(" index(%ld)run(%d)pa(0x%lx)",
2090 					(long)fidx, rcount, (long)pa);
2091 				db_printf("pd(%ld)\n", (long)padiff);
2092 				if (nl > 18) {
2093 					c = cngetc();
2094 					if (c != ' ')
2095 						return;
2096 					nl = 0;
2097 				}
2098 				nl++;
2099 			}
2100 			fidx = idx;
2101 			pa = VM_PAGE_TO_PHYS(m);
2102 			rcount = 1;
2103 		}
2104 		if (rcount) {
2105 			db_printf(" index(%ld)run(%d)pa(0x%lx)\n",
2106 				(long)fidx, rcount, (long)pa);
2107 			if (nl > 18) {
2108 				c = cngetc();
2109 				if (c != ' ')
2110 					return;
2111 				nl = 0;
2112 			}
2113 			nl++;
2114 		}
2115 	}
2116 }
2117 #endif /* DDB */
2118