xref: /freebsd/sys/vm/vm_object.c (revision 3ff369fed2a08f32dda232c10470b949bef9489f)
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/kernel.h>
77 #include <sys/sysctl.h>
78 #include <sys/mutex.h>
79 #include <sys/proc.h>		/* for curproc, pageproc */
80 #include <sys/socket.h>
81 #include <sys/vnode.h>
82 #include <sys/vmmeter.h>
83 #include <sys/sx.h>
84 
85 #include <vm/vm.h>
86 #include <vm/vm_param.h>
87 #include <vm/pmap.h>
88 #include <vm/vm_map.h>
89 #include <vm/vm_object.h>
90 #include <vm/vm_page.h>
91 #include <vm/vm_pageout.h>
92 #include <vm/vm_pager.h>
93 #include <vm/swap_pager.h>
94 #include <vm/vm_kern.h>
95 #include <vm/vm_extern.h>
96 #include <vm/uma.h>
97 
98 #define EASY_SCAN_FACTOR       8
99 
100 #define MSYNC_FLUSH_HARDSEQ	0x01
101 #define MSYNC_FLUSH_SOFTSEQ	0x02
102 
103 /*
104  * msync / VM object flushing optimizations
105  */
106 static int msync_flush_flags = MSYNC_FLUSH_HARDSEQ | MSYNC_FLUSH_SOFTSEQ;
107 SYSCTL_INT(_vm, OID_AUTO, msync_flush_flags,
108         CTLFLAG_RW, &msync_flush_flags, 0, "");
109 
110 static void	vm_object_qcollapse(vm_object_t object);
111 static int	vm_object_page_collect_flush(vm_object_t object, vm_page_t p, int curgeneration, int pagerflags);
112 
113 /*
114  *	Virtual memory objects maintain the actual data
115  *	associated with allocated virtual memory.  A given
116  *	page of memory exists within exactly one object.
117  *
118  *	An object is only deallocated when all "references"
119  *	are given up.  Only one "reference" to a given
120  *	region of an object should be writeable.
121  *
122  *	Associated with each object is a list of all resident
123  *	memory pages belonging to that object; this list is
124  *	maintained by the "vm_page" module, and locked by the object's
125  *	lock.
126  *
127  *	Each object also records a "pager" routine which is
128  *	used to retrieve (and store) pages to the proper backing
129  *	storage.  In addition, objects may be backed by other
130  *	objects from which they were virtual-copied.
131  *
132  *	The only items within the object structure which are
133  *	modified after time of creation are:
134  *		reference count		locked by object's lock
135  *		pager routine		locked by object's lock
136  *
137  */
138 
139 struct object_q vm_object_list;
140 struct mtx vm_object_list_mtx;	/* lock for object list and count */
141 vm_object_t kernel_object;
142 vm_object_t kmem_object;
143 static struct vm_object kernel_object_store;
144 static struct vm_object kmem_object_store;
145 extern int vm_pageout_page_count;
146 
147 static long object_collapses;
148 static long object_bypasses;
149 static int next_index;
150 static uma_zone_t obj_zone;
151 #define VM_OBJECTS_INIT 256
152 
153 static void vm_object_zinit(void *mem, int size);
154 
155 #ifdef INVARIANTS
156 static void vm_object_zdtor(void *mem, int size, void *arg);
157 
158 static void
159 vm_object_zdtor(void *mem, int size, void *arg)
160 {
161 	vm_object_t object;
162 
163 	object = (vm_object_t)mem;
164 	KASSERT(object->paging_in_progress == 0,
165 	    ("object %p paging_in_progress = %d",
166 	    object, object->paging_in_progress));
167 	KASSERT(object->resident_page_count == 0,
168 	    ("object %p resident_page_count = %d",
169 	    object, object->resident_page_count));
170 	KASSERT(object->shadow_count == 0,
171 	    ("object %p shadow_count = %d",
172 	    object, object->shadow_count));
173 }
174 #endif
175 
176 static void
177 vm_object_zinit(void *mem, int size)
178 {
179 	vm_object_t object;
180 
181 	object = (vm_object_t)mem;
182 
183 	/* These are true for any object that has been freed */
184 	object->paging_in_progress = 0;
185 	object->resident_page_count = 0;
186 	object->shadow_count = 0;
187 }
188 
189 void
190 _vm_object_allocate(objtype_t type, vm_size_t size, vm_object_t object)
191 {
192 	static int object_hash_rand;
193 	int exp, incr;
194 
195 	TAILQ_INIT(&object->memq);
196 	TAILQ_INIT(&object->shadow_head);
197 
198 	object->type = type;
199 	object->size = size;
200 	object->ref_count = 1;
201 	object->flags = 0;
202 	if ((object->type == OBJT_DEFAULT) || (object->type == OBJT_SWAP))
203 		vm_object_set_flag(object, OBJ_ONEMAPPING);
204 	if (size > (PQ_L2_SIZE / 3 + PQ_PRIME1))
205 		incr = PQ_L2_SIZE / 3 + PQ_PRIME1;
206 	else
207 		incr = size;
208 	do
209 		object->pg_color = next_index;
210 	while (!atomic_cmpset_int(&next_index, object->pg_color,
211 				  (object->pg_color + incr) & PQ_L2_MASK));
212 	object->handle = NULL;
213 	object->backing_object = NULL;
214 	object->backing_object_offset = (vm_ooffset_t) 0;
215 	/*
216 	 * Try to generate a number that will spread objects out in the
217 	 * hash table.  We 'wipe' new objects across the hash in 128 page
218 	 * increments plus 1 more to offset it a little more by the time
219 	 * it wraps around.
220 	 */
221 	do {
222 		exp = object_hash_rand;
223 		object->hash_rand = exp - 129;
224 	} while (!atomic_cmpset_int(&object_hash_rand, exp, object->hash_rand));
225 
226 	object->generation++;		/* atomicity needed? XXX */
227 
228 	mtx_lock(&vm_object_list_mtx);
229 	TAILQ_INSERT_TAIL(&vm_object_list, object, object_list);
230 	mtx_unlock(&vm_object_list_mtx);
231 }
232 
233 /*
234  *	vm_object_init:
235  *
236  *	Initialize the VM objects module.
237  */
238 void
239 vm_object_init(void)
240 {
241 	TAILQ_INIT(&vm_object_list);
242 	mtx_init(&vm_object_list_mtx, "vm object_list", NULL, MTX_DEF);
243 
244 	kernel_object = &kernel_object_store;
245 	_vm_object_allocate(OBJT_DEFAULT, OFF_TO_IDX(VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS),
246 	    kernel_object);
247 
248 	kmem_object = &kmem_object_store;
249 	_vm_object_allocate(OBJT_DEFAULT, OFF_TO_IDX(VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS),
250 	    kmem_object);
251 	obj_zone = uma_zcreate("VM OBJECT", sizeof (struct vm_object), NULL,
252 #ifdef INVARIANTS
253 	    vm_object_zdtor,
254 #else
255 	    NULL,
256 #endif
257 	    vm_object_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
258 	uma_prealloc(obj_zone, VM_OBJECTS_INIT);
259 }
260 
261 void
262 vm_object_init2(void)
263 {
264 }
265 
266 void
267 vm_object_set_flag(vm_object_t object, u_short bits)
268 {
269 	object->flags |= bits;
270 }
271 
272 void
273 vm_object_clear_flag(vm_object_t object, u_short bits)
274 {
275 	GIANT_REQUIRED;
276 	object->flags &= ~bits;
277 }
278 
279 void
280 vm_object_pip_add(vm_object_t object, short i)
281 {
282 	GIANT_REQUIRED;
283 	object->paging_in_progress += i;
284 }
285 
286 void
287 vm_object_pip_subtract(vm_object_t object, short i)
288 {
289 	GIANT_REQUIRED;
290 	object->paging_in_progress -= i;
291 }
292 
293 void
294 vm_object_pip_wakeup(vm_object_t object)
295 {
296 	GIANT_REQUIRED;
297 	object->paging_in_progress--;
298 	if ((object->flags & OBJ_PIPWNT) && object->paging_in_progress == 0) {
299 		vm_object_clear_flag(object, OBJ_PIPWNT);
300 		wakeup(object);
301 	}
302 }
303 
304 void
305 vm_object_pip_wakeupn(vm_object_t object, short i)
306 {
307 	GIANT_REQUIRED;
308 	if (i)
309 		object->paging_in_progress -= i;
310 	if ((object->flags & OBJ_PIPWNT) && object->paging_in_progress == 0) {
311 		vm_object_clear_flag(object, OBJ_PIPWNT);
312 		wakeup(object);
313 	}
314 }
315 
316 void
317 vm_object_pip_sleep(vm_object_t object, char *waitid)
318 {
319 	GIANT_REQUIRED;
320 	if (object->paging_in_progress) {
321 		int s = splvm();
322 		if (object->paging_in_progress) {
323 			vm_object_set_flag(object, OBJ_PIPWNT);
324 			tsleep(object, PVM, waitid, 0);
325 		}
326 		splx(s);
327 	}
328 }
329 
330 void
331 vm_object_pip_wait(vm_object_t object, char *waitid)
332 {
333 	GIANT_REQUIRED;
334 	while (object->paging_in_progress)
335 		vm_object_pip_sleep(object, waitid);
336 }
337 
338 /*
339  *	vm_object_allocate:
340  *
341  *	Returns a new object with the given size.
342  */
343 vm_object_t
344 vm_object_allocate(objtype_t type, vm_size_t size)
345 {
346 	vm_object_t result;
347 
348 	result = (vm_object_t) uma_zalloc(obj_zone, M_WAITOK);
349 	_vm_object_allocate(type, size, result);
350 
351 	return (result);
352 }
353 
354 
355 /*
356  *	vm_object_reference:
357  *
358  *	Gets another reference to the given object.
359  */
360 void
361 vm_object_reference(vm_object_t object)
362 {
363 	if (object == NULL)
364 		return;
365 
366 	mtx_lock(&Giant);
367 #if 0
368 	/* object can be re-referenced during final cleaning */
369 	KASSERT(!(object->flags & OBJ_DEAD),
370 	    ("vm_object_reference: attempting to reference dead obj"));
371 #endif
372 
373 	object->ref_count++;
374 	if (object->type == OBJT_VNODE) {
375 		while (vget((struct vnode *) object->handle, LK_RETRY|LK_NOOBJ, curthread)) {
376 			printf("vm_object_reference: delay in getting object\n");
377 		}
378 	}
379 	mtx_unlock(&Giant);
380 }
381 
382 /*
383  * handle deallocating a object of type OBJT_VNODE
384  */
385 void
386 vm_object_vndeallocate(vm_object_t object)
387 {
388 	struct vnode *vp = (struct vnode *) object->handle;
389 
390 	GIANT_REQUIRED;
391 	KASSERT(object->type == OBJT_VNODE,
392 	    ("vm_object_vndeallocate: not a vnode object"));
393 	KASSERT(vp != NULL, ("vm_object_vndeallocate: missing vp"));
394 #ifdef INVARIANTS
395 	if (object->ref_count == 0) {
396 		vprint("vm_object_vndeallocate", vp);
397 		panic("vm_object_vndeallocate: bad object reference count");
398 	}
399 #endif
400 
401 	object->ref_count--;
402 	if (object->ref_count == 0) {
403 		vp->v_flag &= ~VTEXT;
404 #ifdef ENABLE_VFS_IOOPT
405 		vm_object_clear_flag(object, OBJ_OPT);
406 #endif
407 	}
408 	/*
409 	 * vrele may need a vop lock
410 	 */
411 	vrele(vp);
412 }
413 
414 /*
415  *	vm_object_deallocate:
416  *
417  *	Release a reference to the specified object,
418  *	gained either through a vm_object_allocate
419  *	or a vm_object_reference call.  When all references
420  *	are gone, storage associated with this object
421  *	may be relinquished.
422  *
423  *	No object may be locked.
424  */
425 void
426 vm_object_deallocate(vm_object_t object)
427 {
428 	vm_object_t temp;
429 
430 	mtx_lock(&Giant);
431 	while (object != NULL) {
432 
433 		if (object->type == OBJT_VNODE) {
434 			vm_object_vndeallocate(object);
435 			mtx_unlock(&Giant);
436 			return;
437 		}
438 
439 		KASSERT(object->ref_count != 0,
440 			("vm_object_deallocate: object deallocated too many times: %d", object->type));
441 
442 		/*
443 		 * If the reference count goes to 0 we start calling
444 		 * vm_object_terminate() on the object chain.
445 		 * A ref count of 1 may be a special case depending on the
446 		 * shadow count being 0 or 1.
447 		 */
448 		object->ref_count--;
449 		if (object->ref_count > 1) {
450 			mtx_unlock(&Giant);
451 			return;
452 		} else if (object->ref_count == 1) {
453 			if (object->shadow_count == 0) {
454 				vm_object_set_flag(object, OBJ_ONEMAPPING);
455 			} else if ((object->shadow_count == 1) &&
456 			    (object->handle == NULL) &&
457 			    (object->type == OBJT_DEFAULT ||
458 			     object->type == OBJT_SWAP)) {
459 				vm_object_t robject;
460 
461 				robject = TAILQ_FIRST(&object->shadow_head);
462 				KASSERT(robject != NULL,
463 				    ("vm_object_deallocate: ref_count: %d, shadow_count: %d",
464 					 object->ref_count,
465 					 object->shadow_count));
466 				if ((robject->handle == NULL) &&
467 				    (robject->type == OBJT_DEFAULT ||
468 				     robject->type == OBJT_SWAP)) {
469 
470 					robject->ref_count++;
471 
472 					while (
473 						robject->paging_in_progress ||
474 						object->paging_in_progress
475 					) {
476 						vm_object_pip_sleep(robject, "objde1");
477 						vm_object_pip_sleep(object, "objde2");
478 					}
479 
480 					if (robject->ref_count == 1) {
481 						robject->ref_count--;
482 						object = robject;
483 						goto doterm;
484 					}
485 
486 					object = robject;
487 					vm_object_collapse(object);
488 					continue;
489 				}
490 			}
491 			mtx_unlock(&Giant);
492 			return;
493 		}
494 doterm:
495 		temp = object->backing_object;
496 		if (temp) {
497 			TAILQ_REMOVE(&temp->shadow_head, object, shadow_list);
498 			temp->shadow_count--;
499 #ifdef ENABLE_VFS_IOOPT
500 			if (temp->ref_count == 0)
501 				vm_object_clear_flag(temp, OBJ_OPT);
502 #endif
503 			temp->generation++;
504 			object->backing_object = NULL;
505 		}
506 		/*
507 		 * Don't double-terminate, we could be in a termination
508 		 * recursion due to the terminate having to sync data
509 		 * to disk.
510 		 */
511 		if ((object->flags & OBJ_DEAD) == 0)
512 			vm_object_terminate(object);
513 		object = temp;
514 	}
515 	mtx_unlock(&Giant);
516 }
517 
518 /*
519  *	vm_object_terminate actually destroys the specified object, freeing
520  *	up all previously used resources.
521  *
522  *	The object must be locked.
523  *	This routine may block.
524  */
525 void
526 vm_object_terminate(vm_object_t object)
527 {
528 	vm_page_t p;
529 	int s;
530 
531 	GIANT_REQUIRED;
532 
533 	/*
534 	 * Make sure no one uses us.
535 	 */
536 	vm_object_set_flag(object, OBJ_DEAD);
537 
538 	/*
539 	 * wait for the pageout daemon to be done with the object
540 	 */
541 	vm_object_pip_wait(object, "objtrm");
542 
543 	KASSERT(!object->paging_in_progress,
544 		("vm_object_terminate: pageout in progress"));
545 
546 	/*
547 	 * Clean and free the pages, as appropriate. All references to the
548 	 * object are gone, so we don't need to lock it.
549 	 */
550 	if (object->type == OBJT_VNODE) {
551 		struct vnode *vp;
552 
553 #ifdef ENABLE_VFS_IOOPT
554 		/*
555 		 * Freeze optimized copies.
556 		 */
557 		vm_freeze_copyopts(object, 0, object->size);
558 #endif
559 		/*
560 		 * Clean pages and flush buffers.
561 		 */
562 		vm_object_page_clean(object, 0, 0, OBJPC_SYNC);
563 
564 		vp = (struct vnode *) object->handle;
565 		vinvalbuf(vp, V_SAVE, NOCRED, NULL, 0, 0);
566 	}
567 
568 	KASSERT(object->ref_count == 0,
569 		("vm_object_terminate: object with references, ref_count=%d",
570 		object->ref_count));
571 
572 	/*
573 	 * Now free any remaining pages. For internal objects, this also
574 	 * removes them from paging queues. Don't free wired pages, just
575 	 * remove them from the object.
576 	 */
577 	s = splvm();
578 	while ((p = TAILQ_FIRST(&object->memq)) != NULL) {
579 		KASSERT(!p->busy && (p->flags & PG_BUSY) == 0,
580 			("vm_object_terminate: freeing busy page %p "
581 			"p->busy = %d, p->flags %x\n", p, p->busy, p->flags));
582 		if (p->wire_count == 0) {
583 			vm_page_busy(p);
584 			vm_page_free(p);
585 			cnt.v_pfree++;
586 		} else {
587 			vm_page_busy(p);
588 			vm_page_remove(p);
589 		}
590 	}
591 	splx(s);
592 
593 	/*
594 	 * Let the pager know object is dead.
595 	 */
596 	vm_pager_deallocate(object);
597 
598 	/*
599 	 * Remove the object from the global object list.
600 	 */
601 	mtx_lock(&vm_object_list_mtx);
602 	TAILQ_REMOVE(&vm_object_list, object, object_list);
603 	mtx_unlock(&vm_object_list_mtx);
604 
605 	wakeup(object);
606 
607 	/*
608 	 * Free the space for the object.
609 	 */
610 	uma_zfree(obj_zone, object);
611 }
612 
613 /*
614  *	vm_object_page_clean
615  *
616  *	Clean all dirty pages in the specified range of object.  Leaves page
617  * 	on whatever queue it is currently on.   If NOSYNC is set then do not
618  *	write out pages with PG_NOSYNC set (originally comes from MAP_NOSYNC),
619  *	leaving the object dirty.
620  *
621  *	Odd semantics: if start == end, we clean everything.
622  *
623  *	The object must be locked.
624  */
625 void
626 vm_object_page_clean(vm_object_t object, vm_pindex_t start, vm_pindex_t end, int flags)
627 {
628 	vm_page_t p, np;
629 	vm_offset_t tstart, tend;
630 	vm_pindex_t pi;
631 	struct vnode *vp;
632 	int clearobjflags;
633 	int pagerflags;
634 	int curgeneration;
635 
636 	GIANT_REQUIRED;
637 
638 	if (object->type != OBJT_VNODE ||
639 		(object->flags & OBJ_MIGHTBEDIRTY) == 0)
640 		return;
641 
642 	pagerflags = (flags & (OBJPC_SYNC | OBJPC_INVAL)) ? VM_PAGER_PUT_SYNC : 0;
643 	pagerflags |= (flags & OBJPC_INVAL) ? VM_PAGER_PUT_INVAL : 0;
644 
645 	vp = object->handle;
646 
647 	vm_object_set_flag(object, OBJ_CLEANING);
648 
649 	tstart = start;
650 	if (end == 0) {
651 		tend = object->size;
652 	} else {
653 		tend = end;
654 	}
655 
656 	/*
657 	 * If the caller is smart and only msync()s a range he knows is
658 	 * dirty, we may be able to avoid an object scan.  This results in
659 	 * a phenominal improvement in performance.  We cannot do this
660 	 * as a matter of course because the object may be huge - e.g.
661 	 * the size might be in the gigabytes or terrabytes.
662 	 */
663 	if (msync_flush_flags & MSYNC_FLUSH_HARDSEQ) {
664 		vm_offset_t tscan;
665 		int scanlimit;
666 		int scanreset;
667 
668 		scanreset = object->resident_page_count / EASY_SCAN_FACTOR;
669 		if (scanreset < 16)
670 			scanreset = 16;
671 
672 		scanlimit = scanreset;
673 		tscan = tstart;
674 		while (tscan < tend) {
675 			curgeneration = object->generation;
676 			p = vm_page_lookup(object, tscan);
677 			if (p == NULL || p->valid == 0 ||
678 			    (p->queue - p->pc) == PQ_CACHE) {
679 				if (--scanlimit == 0)
680 					break;
681 				++tscan;
682 				continue;
683 			}
684 			vm_page_test_dirty(p);
685 			if ((p->dirty & p->valid) == 0) {
686 				if (--scanlimit == 0)
687 					break;
688 				++tscan;
689 				continue;
690 			}
691 			/*
692 			 * If we have been asked to skip nosync pages and
693 			 * this is a nosync page, we can't continue.
694 			 */
695 			if ((flags & OBJPC_NOSYNC) && (p->flags & PG_NOSYNC)) {
696 				if (--scanlimit == 0)
697 					break;
698 				++tscan;
699 				continue;
700 			}
701 			scanlimit = scanreset;
702 
703 			/*
704 			 * This returns 0 if it was unable to busy the first
705 			 * page (i.e. had to sleep).
706 			 */
707 			tscan += vm_object_page_collect_flush(object, p, curgeneration, pagerflags);
708 		}
709 
710 		/*
711 		 * If everything was dirty and we flushed it successfully,
712 		 * and the requested range is not the entire object, we
713 		 * don't have to mess with CLEANCHK or MIGHTBEDIRTY and can
714 		 * return immediately.
715 		 */
716 		if (tscan >= tend && (tstart || tend < object->size)) {
717 			vm_object_clear_flag(object, OBJ_CLEANING);
718 			return;
719 		}
720 	}
721 
722 	/*
723 	 * Generally set CLEANCHK interlock and make the page read-only so
724 	 * we can then clear the object flags.
725 	 *
726 	 * However, if this is a nosync mmap then the object is likely to
727 	 * stay dirty so do not mess with the page and do not clear the
728 	 * object flags.
729 	 */
730 	clearobjflags = 1;
731 
732 	TAILQ_FOREACH(p, &object->memq, listq) {
733 		vm_page_flag_set(p, PG_CLEANCHK);
734 		if ((flags & OBJPC_NOSYNC) && (p->flags & PG_NOSYNC))
735 			clearobjflags = 0;
736 		else
737 			vm_page_protect(p, VM_PROT_READ);
738 	}
739 
740 	if (clearobjflags && (tstart == 0) && (tend == object->size)) {
741 		struct vnode *vp;
742 
743 		vm_object_clear_flag(object, OBJ_WRITEABLE|OBJ_MIGHTBEDIRTY);
744 		if (object->type == OBJT_VNODE &&
745 		    (vp = (struct vnode *)object->handle) != NULL) {
746 			if (vp->v_flag & VOBJDIRTY) {
747 				mtx_lock(&vp->v_interlock);
748 				vp->v_flag &= ~VOBJDIRTY;
749 				mtx_unlock(&vp->v_interlock);
750 			}
751 		}
752 	}
753 
754 rescan:
755 	curgeneration = object->generation;
756 
757 	for (p = TAILQ_FIRST(&object->memq); p; p = np) {
758 		int n;
759 
760 		np = TAILQ_NEXT(p, listq);
761 
762 again:
763 		pi = p->pindex;
764 		if (((p->flags & PG_CLEANCHK) == 0) ||
765 			(pi < tstart) || (pi >= tend) ||
766 			(p->valid == 0) ||
767 			((p->queue - p->pc) == PQ_CACHE)) {
768 			vm_page_flag_clear(p, PG_CLEANCHK);
769 			continue;
770 		}
771 
772 		vm_page_test_dirty(p);
773 		if ((p->dirty & p->valid) == 0) {
774 			vm_page_flag_clear(p, PG_CLEANCHK);
775 			continue;
776 		}
777 
778 		/*
779 		 * If we have been asked to skip nosync pages and this is a
780 		 * nosync page, skip it.  Note that the object flags were
781 		 * not cleared in this case so we do not have to set them.
782 		 */
783 		if ((flags & OBJPC_NOSYNC) && (p->flags & PG_NOSYNC)) {
784 			vm_page_flag_clear(p, PG_CLEANCHK);
785 			continue;
786 		}
787 
788 		n = vm_object_page_collect_flush(object, p,
789 			curgeneration, pagerflags);
790 		if (n == 0)
791 			goto rescan;
792 
793 		if (object->generation != curgeneration)
794 			goto rescan;
795 
796 		/*
797 		 * Try to optimize the next page.  If we can't we pick up
798 		 * our (random) scan where we left off.
799 		 */
800 		if (msync_flush_flags & MSYNC_FLUSH_SOFTSEQ) {
801 			if ((p = vm_page_lookup(object, pi + n)) != NULL)
802 				goto again;
803 		}
804 	}
805 
806 #if 0
807 	VOP_FSYNC(vp, NULL, (pagerflags & VM_PAGER_PUT_SYNC)?MNT_WAIT:0, curproc);
808 #endif
809 
810 	vm_object_clear_flag(object, OBJ_CLEANING);
811 	return;
812 }
813 
814 static int
815 vm_object_page_collect_flush(vm_object_t object, vm_page_t p, int curgeneration, int pagerflags)
816 {
817 	int runlen;
818 	int s;
819 	int maxf;
820 	int chkb;
821 	int maxb;
822 	int i;
823 	vm_pindex_t pi;
824 	vm_page_t maf[vm_pageout_page_count];
825 	vm_page_t mab[vm_pageout_page_count];
826 	vm_page_t ma[vm_pageout_page_count];
827 
828 	s = splvm();
829 	pi = p->pindex;
830 	while (vm_page_sleep_busy(p, TRUE, "vpcwai")) {
831 		if (object->generation != curgeneration) {
832 			splx(s);
833 			return(0);
834 		}
835 	}
836 
837 	maxf = 0;
838 	for(i = 1; i < vm_pageout_page_count; i++) {
839 		vm_page_t tp;
840 
841 		if ((tp = vm_page_lookup(object, pi + i)) != NULL) {
842 			if ((tp->flags & PG_BUSY) ||
843 				(tp->flags & PG_CLEANCHK) == 0 ||
844 				(tp->busy != 0))
845 				break;
846 			if((tp->queue - tp->pc) == PQ_CACHE) {
847 				vm_page_flag_clear(tp, PG_CLEANCHK);
848 				break;
849 			}
850 			vm_page_test_dirty(tp);
851 			if ((tp->dirty & tp->valid) == 0) {
852 				vm_page_flag_clear(tp, PG_CLEANCHK);
853 				break;
854 			}
855 			maf[ i - 1 ] = tp;
856 			maxf++;
857 			continue;
858 		}
859 		break;
860 	}
861 
862 	maxb = 0;
863 	chkb = vm_pageout_page_count -  maxf;
864 	if (chkb) {
865 		for(i = 1; i < chkb;i++) {
866 			vm_page_t tp;
867 
868 			if ((tp = vm_page_lookup(object, pi - i)) != NULL) {
869 				if ((tp->flags & PG_BUSY) ||
870 					(tp->flags & PG_CLEANCHK) == 0 ||
871 					(tp->busy != 0))
872 					break;
873 				if ((tp->queue - tp->pc) == PQ_CACHE) {
874 					vm_page_flag_clear(tp, PG_CLEANCHK);
875 					break;
876 				}
877 				vm_page_test_dirty(tp);
878 				if ((tp->dirty & tp->valid) == 0) {
879 					vm_page_flag_clear(tp, PG_CLEANCHK);
880 					break;
881 				}
882 				mab[ i - 1 ] = tp;
883 				maxb++;
884 				continue;
885 			}
886 			break;
887 		}
888 	}
889 
890 	for(i = 0; i < maxb; i++) {
891 		int index = (maxb - i) - 1;
892 		ma[index] = mab[i];
893 		vm_page_flag_clear(ma[index], PG_CLEANCHK);
894 	}
895 	vm_page_flag_clear(p, PG_CLEANCHK);
896 	ma[maxb] = p;
897 	for(i = 0; i < maxf; i++) {
898 		int index = (maxb + i) + 1;
899 		ma[index] = maf[i];
900 		vm_page_flag_clear(ma[index], PG_CLEANCHK);
901 	}
902 	runlen = maxb + maxf + 1;
903 
904 	splx(s);
905 	vm_pageout_flush(ma, runlen, pagerflags);
906 	for (i = 0; i < runlen; i++) {
907 		if (ma[i]->valid & ma[i]->dirty) {
908 			vm_page_protect(ma[i], VM_PROT_READ);
909 			vm_page_flag_set(ma[i], PG_CLEANCHK);
910 
911 			/*
912 			 * maxf will end up being the actual number of pages
913 			 * we wrote out contiguously, non-inclusive of the
914 			 * first page.  We do not count look-behind pages.
915 			 */
916 			if (i >= maxb + 1 && (maxf > i - maxb - 1))
917 				maxf = i - maxb - 1;
918 		}
919 	}
920 	return(maxf + 1);
921 }
922 
923 #ifdef ENABLE_VFS_IOOPT
924 /*
925  * Same as vm_object_pmap_copy, except range checking really
926  * works, and is meant for small sections of an object.
927  *
928  * This code protects resident pages by making them read-only
929  * and is typically called on a fork or split when a page
930  * is converted to copy-on-write.
931  *
932  * NOTE: If the page is already at VM_PROT_NONE, calling
933  * vm_page_protect will have no effect.
934  */
935 void
936 vm_object_pmap_copy_1(vm_object_t object, vm_pindex_t start, vm_pindex_t end)
937 {
938 	vm_pindex_t idx;
939 	vm_page_t p;
940 
941 	GIANT_REQUIRED;
942 
943 	if (object == NULL || (object->flags & OBJ_WRITEABLE) == 0)
944 		return;
945 
946 	for (idx = start; idx < end; idx++) {
947 		p = vm_page_lookup(object, idx);
948 		if (p == NULL)
949 			continue;
950 		vm_page_protect(p, VM_PROT_READ);
951 	}
952 }
953 #endif
954 
955 /*
956  *	vm_object_pmap_remove:
957  *
958  *	Removes all physical pages in the specified
959  *	object range from all physical maps.
960  *
961  *	The object must *not* be locked.
962  */
963 void
964 vm_object_pmap_remove(vm_object_t object, vm_pindex_t start, vm_pindex_t end)
965 {
966 	vm_page_t p;
967 
968 	GIANT_REQUIRED;
969 	if (object == NULL)
970 		return;
971 	TAILQ_FOREACH(p, &object->memq, listq) {
972 		if (p->pindex >= start && p->pindex < end)
973 			vm_page_protect(p, VM_PROT_NONE);
974 	}
975 	if ((start == 0) && (object->size == end))
976 		vm_object_clear_flag(object, OBJ_WRITEABLE);
977 }
978 
979 /*
980  *	vm_object_madvise:
981  *
982  *	Implements the madvise function at the object/page level.
983  *
984  *	MADV_WILLNEED	(any object)
985  *
986  *	    Activate the specified pages if they are resident.
987  *
988  *	MADV_DONTNEED	(any object)
989  *
990  *	    Deactivate the specified pages if they are resident.
991  *
992  *	MADV_FREE	(OBJT_DEFAULT/OBJT_SWAP objects,
993  *			 OBJ_ONEMAPPING only)
994  *
995  *	    Deactivate and clean the specified pages if they are
996  *	    resident.  This permits the process to reuse the pages
997  *	    without faulting or the kernel to reclaim the pages
998  *	    without I/O.
999  */
1000 void
1001 vm_object_madvise(vm_object_t object, vm_pindex_t pindex, int count, int advise)
1002 {
1003 	vm_pindex_t end, tpindex;
1004 	vm_object_t tobject;
1005 	vm_page_t m;
1006 
1007 	if (object == NULL)
1008 		return;
1009 
1010 	mtx_lock(&Giant);
1011 
1012 	end = pindex + count;
1013 
1014 	/*
1015 	 * Locate and adjust resident pages
1016 	 */
1017 	for (; pindex < end; pindex += 1) {
1018 relookup:
1019 		tobject = object;
1020 		tpindex = pindex;
1021 shadowlookup:
1022 		/*
1023 		 * MADV_FREE only operates on OBJT_DEFAULT or OBJT_SWAP pages
1024 		 * and those pages must be OBJ_ONEMAPPING.
1025 		 */
1026 		if (advise == MADV_FREE) {
1027 			if ((tobject->type != OBJT_DEFAULT &&
1028 			     tobject->type != OBJT_SWAP) ||
1029 			    (tobject->flags & OBJ_ONEMAPPING) == 0) {
1030 				continue;
1031 			}
1032 		}
1033 
1034 		m = vm_page_lookup(tobject, tpindex);
1035 
1036 		if (m == NULL) {
1037 			/*
1038 			 * There may be swap even if there is no backing page
1039 			 */
1040 			if (advise == MADV_FREE && tobject->type == OBJT_SWAP)
1041 				swap_pager_freespace(tobject, tpindex, 1);
1042 
1043 			/*
1044 			 * next object
1045 			 */
1046 			tobject = tobject->backing_object;
1047 			if (tobject == NULL)
1048 				continue;
1049 			tpindex += OFF_TO_IDX(tobject->backing_object_offset);
1050 			goto shadowlookup;
1051 		}
1052 
1053 		/*
1054 		 * If the page is busy or not in a normal active state,
1055 		 * we skip it.  If the page is not managed there are no
1056 		 * page queues to mess with.  Things can break if we mess
1057 		 * with pages in any of the below states.
1058 		 */
1059 		if (
1060 		    m->hold_count ||
1061 		    m->wire_count ||
1062 		    (m->flags & PG_UNMANAGED) ||
1063 		    m->valid != VM_PAGE_BITS_ALL
1064 		) {
1065 			continue;
1066 		}
1067 
1068  		if (vm_page_sleep_busy(m, TRUE, "madvpo"))
1069   			goto relookup;
1070 
1071 		if (advise == MADV_WILLNEED) {
1072 			vm_page_activate(m);
1073 		} else if (advise == MADV_DONTNEED) {
1074 			vm_page_dontneed(m);
1075 		} else if (advise == MADV_FREE) {
1076 			/*
1077 			 * Mark the page clean.  This will allow the page
1078 			 * to be freed up by the system.  However, such pages
1079 			 * are often reused quickly by malloc()/free()
1080 			 * so we do not do anything that would cause
1081 			 * a page fault if we can help it.
1082 			 *
1083 			 * Specifically, we do not try to actually free
1084 			 * the page now nor do we try to put it in the
1085 			 * cache (which would cause a page fault on reuse).
1086 			 *
1087 			 * But we do make the page is freeable as we
1088 			 * can without actually taking the step of unmapping
1089 			 * it.
1090 			 */
1091 			pmap_clear_modify(m);
1092 			m->dirty = 0;
1093 			m->act_count = 0;
1094 			vm_page_dontneed(m);
1095 			if (tobject->type == OBJT_SWAP)
1096 				swap_pager_freespace(tobject, tpindex, 1);
1097 		}
1098 	}
1099 	mtx_unlock(&Giant);
1100 }
1101 
1102 /*
1103  *	vm_object_shadow:
1104  *
1105  *	Create a new object which is backed by the
1106  *	specified existing object range.  The source
1107  *	object reference is deallocated.
1108  *
1109  *	The new object and offset into that object
1110  *	are returned in the source parameters.
1111  */
1112 void
1113 vm_object_shadow(
1114 	vm_object_t *object,	/* IN/OUT */
1115 	vm_ooffset_t *offset,	/* IN/OUT */
1116 	vm_size_t length)
1117 {
1118 	vm_object_t source;
1119 	vm_object_t result;
1120 
1121 	source = *object;
1122 
1123 	mtx_lock(&Giant);
1124 	/*
1125 	 * Don't create the new object if the old object isn't shared.
1126 	 */
1127 	if (source != NULL &&
1128 	    source->ref_count == 1 &&
1129 	    source->handle == NULL &&
1130 	    (source->type == OBJT_DEFAULT ||
1131 	     source->type == OBJT_SWAP)) {
1132 		mtx_unlock(&Giant);
1133 		return;
1134 	}
1135 
1136 	/*
1137 	 * Allocate a new object with the given length
1138 	 */
1139 	result = vm_object_allocate(OBJT_DEFAULT, length);
1140 	KASSERT(result != NULL, ("vm_object_shadow: no object for shadowing"));
1141 
1142 	/*
1143 	 * The new object shadows the source object, adding a reference to it.
1144 	 * Our caller changes his reference to point to the new object,
1145 	 * removing a reference to the source object.  Net result: no change
1146 	 * of reference count.
1147 	 *
1148 	 * Try to optimize the result object's page color when shadowing
1149 	 * in order to maintain page coloring consistency in the combined
1150 	 * shadowed object.
1151 	 */
1152 	result->backing_object = source;
1153 	if (source) {
1154 		TAILQ_INSERT_TAIL(&source->shadow_head, result, shadow_list);
1155 		source->shadow_count++;
1156 		source->generation++;
1157 		if (length < source->size)
1158 			length = source->size;
1159 		if (length > PQ_L2_SIZE / 3 + PQ_PRIME1 ||
1160 		    source->generation > 1)
1161 			length = PQ_L2_SIZE / 3 + PQ_PRIME1;
1162 		result->pg_color = (source->pg_color +
1163 		    length * source->generation) & PQ_L2_MASK;
1164 		next_index = (result->pg_color + PQ_L2_SIZE / 3 + PQ_PRIME1) &
1165 		    PQ_L2_MASK;
1166 	}
1167 
1168 	/*
1169 	 * Store the offset into the source object, and fix up the offset into
1170 	 * the new object.
1171 	 */
1172 	result->backing_object_offset = *offset;
1173 
1174 	/*
1175 	 * Return the new things
1176 	 */
1177 	*offset = 0;
1178 	*object = result;
1179 
1180 	mtx_unlock(&Giant);
1181 }
1182 
1183 /*
1184  *	vm_object_split:
1185  *
1186  * Split the pages in a map entry into a new object.  This affords
1187  * easier removal of unused pages, and keeps object inheritance from
1188  * being a negative impact on memory usage.
1189  */
1190 void
1191 vm_object_split(vm_map_entry_t entry)
1192 {
1193 	vm_page_t m;
1194 	vm_object_t orig_object, new_object, source;
1195 	vm_offset_t s, e;
1196 	vm_pindex_t offidxstart, offidxend, idx;
1197 	vm_size_t size;
1198 	vm_ooffset_t offset;
1199 
1200 	GIANT_REQUIRED;
1201 
1202 	orig_object = entry->object.vm_object;
1203 	if (orig_object->type != OBJT_DEFAULT && orig_object->type != OBJT_SWAP)
1204 		return;
1205 	if (orig_object->ref_count <= 1)
1206 		return;
1207 
1208 	offset = entry->offset;
1209 	s = entry->start;
1210 	e = entry->end;
1211 
1212 	offidxstart = OFF_TO_IDX(offset);
1213 	offidxend = offidxstart + OFF_TO_IDX(e - s);
1214 	size = offidxend - offidxstart;
1215 
1216 	new_object = vm_pager_allocate(orig_object->type,
1217 		NULL, IDX_TO_OFF(size), VM_PROT_ALL, 0LL);
1218 	if (new_object == NULL)
1219 		return;
1220 
1221 	source = orig_object->backing_object;
1222 	if (source != NULL) {
1223 		vm_object_reference(source);	/* Referenced by new_object */
1224 		TAILQ_INSERT_TAIL(&source->shadow_head,
1225 				  new_object, shadow_list);
1226 		vm_object_clear_flag(source, OBJ_ONEMAPPING);
1227 		new_object->backing_object_offset =
1228 			orig_object->backing_object_offset + IDX_TO_OFF(offidxstart);
1229 		new_object->backing_object = source;
1230 		source->shadow_count++;
1231 		source->generation++;
1232 	}
1233 	for (idx = 0; idx < size; idx++) {
1234 	retry:
1235 		m = vm_page_lookup(orig_object, offidxstart + idx);
1236 		if (m == NULL)
1237 			continue;
1238 
1239 		/*
1240 		 * We must wait for pending I/O to complete before we can
1241 		 * rename the page.
1242 		 *
1243 		 * We do not have to VM_PROT_NONE the page as mappings should
1244 		 * not be changed by this operation.
1245 		 */
1246 		if (vm_page_sleep_busy(m, TRUE, "spltwt"))
1247 			goto retry;
1248 
1249 		vm_page_busy(m);
1250 		vm_page_rename(m, new_object, idx);
1251 		/* page automatically made dirty by rename and cache handled */
1252 		vm_page_busy(m);
1253 	}
1254 	if (orig_object->type == OBJT_SWAP) {
1255 		vm_object_pip_add(orig_object, 1);
1256 		/*
1257 		 * copy orig_object pages into new_object
1258 		 * and destroy unneeded pages in
1259 		 * shadow object.
1260 		 */
1261 		swap_pager_copy(orig_object, new_object, offidxstart, 0);
1262 		vm_object_pip_wakeup(orig_object);
1263 	}
1264 	for (idx = 0; idx < size; idx++) {
1265 		m = vm_page_lookup(new_object, idx);
1266 		if (m != NULL)
1267 			vm_page_wakeup(m);
1268 	}
1269 	entry->object.vm_object = new_object;
1270 	entry->offset = 0LL;
1271 	vm_object_deallocate(orig_object);
1272 }
1273 
1274 #define	OBSC_TEST_ALL_SHADOWED	0x0001
1275 #define	OBSC_COLLAPSE_NOWAIT	0x0002
1276 #define	OBSC_COLLAPSE_WAIT	0x0004
1277 
1278 static __inline int
1279 vm_object_backing_scan(vm_object_t object, int op)
1280 {
1281 	int s;
1282 	int r = 1;
1283 	vm_page_t p;
1284 	vm_object_t backing_object;
1285 	vm_pindex_t backing_offset_index;
1286 
1287 	s = splvm();
1288 	GIANT_REQUIRED;
1289 
1290 	backing_object = object->backing_object;
1291 	backing_offset_index = OFF_TO_IDX(object->backing_object_offset);
1292 
1293 	/*
1294 	 * Initial conditions
1295 	 */
1296 	if (op & OBSC_TEST_ALL_SHADOWED) {
1297 		/*
1298 		 * We do not want to have to test for the existence of
1299 		 * swap pages in the backing object.  XXX but with the
1300 		 * new swapper this would be pretty easy to do.
1301 		 *
1302 		 * XXX what about anonymous MAP_SHARED memory that hasn't
1303 		 * been ZFOD faulted yet?  If we do not test for this, the
1304 		 * shadow test may succeed! XXX
1305 		 */
1306 		if (backing_object->type != OBJT_DEFAULT) {
1307 			splx(s);
1308 			return (0);
1309 		}
1310 	}
1311 	if (op & OBSC_COLLAPSE_WAIT) {
1312 		vm_object_set_flag(backing_object, OBJ_DEAD);
1313 	}
1314 
1315 	/*
1316 	 * Our scan
1317 	 */
1318 	p = TAILQ_FIRST(&backing_object->memq);
1319 	while (p) {
1320 		vm_page_t next = TAILQ_NEXT(p, listq);
1321 		vm_pindex_t new_pindex = p->pindex - backing_offset_index;
1322 
1323 		if (op & OBSC_TEST_ALL_SHADOWED) {
1324 			vm_page_t pp;
1325 
1326 			/*
1327 			 * Ignore pages outside the parent object's range
1328 			 * and outside the parent object's mapping of the
1329 			 * backing object.
1330 			 *
1331 			 * note that we do not busy the backing object's
1332 			 * page.
1333 			 */
1334 			if (
1335 			    p->pindex < backing_offset_index ||
1336 			    new_pindex >= object->size
1337 			) {
1338 				p = next;
1339 				continue;
1340 			}
1341 
1342 			/*
1343 			 * See if the parent has the page or if the parent's
1344 			 * object pager has the page.  If the parent has the
1345 			 * page but the page is not valid, the parent's
1346 			 * object pager must have the page.
1347 			 *
1348 			 * If this fails, the parent does not completely shadow
1349 			 * the object and we might as well give up now.
1350 			 */
1351 
1352 			pp = vm_page_lookup(object, new_pindex);
1353 			if (
1354 			    (pp == NULL || pp->valid == 0) &&
1355 			    !vm_pager_has_page(object, new_pindex, NULL, NULL)
1356 			) {
1357 				r = 0;
1358 				break;
1359 			}
1360 		}
1361 
1362 		/*
1363 		 * Check for busy page
1364 		 */
1365 		if (op & (OBSC_COLLAPSE_WAIT | OBSC_COLLAPSE_NOWAIT)) {
1366 			vm_page_t pp;
1367 
1368 			if (op & OBSC_COLLAPSE_NOWAIT) {
1369 				if (
1370 				    (p->flags & PG_BUSY) ||
1371 				    !p->valid ||
1372 				    p->hold_count ||
1373 				    p->wire_count ||
1374 				    p->busy
1375 				) {
1376 					p = next;
1377 					continue;
1378 				}
1379 			} else if (op & OBSC_COLLAPSE_WAIT) {
1380 				if (vm_page_sleep_busy(p, TRUE, "vmocol")) {
1381 					/*
1382 					 * If we slept, anything could have
1383 					 * happened.  Since the object is
1384 					 * marked dead, the backing offset
1385 					 * should not have changed so we
1386 					 * just restart our scan.
1387 					 */
1388 					p = TAILQ_FIRST(&backing_object->memq);
1389 					continue;
1390 				}
1391 			}
1392 
1393 			/*
1394 			 * Busy the page
1395 			 */
1396 			vm_page_busy(p);
1397 
1398 			KASSERT(
1399 			    p->object == backing_object,
1400 			    ("vm_object_qcollapse(): object mismatch")
1401 			);
1402 
1403 			/*
1404 			 * Destroy any associated swap
1405 			 */
1406 			if (backing_object->type == OBJT_SWAP) {
1407 				swap_pager_freespace(
1408 				    backing_object,
1409 				    p->pindex,
1410 				    1
1411 				);
1412 			}
1413 
1414 			if (
1415 			    p->pindex < backing_offset_index ||
1416 			    new_pindex >= object->size
1417 			) {
1418 				/*
1419 				 * Page is out of the parent object's range, we
1420 				 * can simply destroy it.
1421 				 */
1422 				vm_page_protect(p, VM_PROT_NONE);
1423 				vm_page_free(p);
1424 				p = next;
1425 				continue;
1426 			}
1427 
1428 			pp = vm_page_lookup(object, new_pindex);
1429 			if (
1430 			    pp != NULL ||
1431 			    vm_pager_has_page(object, new_pindex, NULL, NULL)
1432 			) {
1433 				/*
1434 				 * page already exists in parent OR swap exists
1435 				 * for this location in the parent.  Destroy
1436 				 * the original page from the backing object.
1437 				 *
1438 				 * Leave the parent's page alone
1439 				 */
1440 				vm_page_protect(p, VM_PROT_NONE);
1441 				vm_page_free(p);
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 			if ((p->queue - p->pc) == PQ_CACHE)
1454 				vm_page_deactivate(p);
1455 
1456 			vm_page_rename(p, object, new_pindex);
1457 			/* page automatically made dirty by rename */
1458 		}
1459 		p = next;
1460 	}
1461 	splx(s);
1462 	return (r);
1463 }
1464 
1465 
1466 /*
1467  * this version of collapse allows the operation to occur earlier and
1468  * when paging_in_progress is true for an object...  This is not a complete
1469  * operation, but should plug 99.9% of the rest of the leaks.
1470  */
1471 static void
1472 vm_object_qcollapse(vm_object_t object)
1473 {
1474 	vm_object_t backing_object = object->backing_object;
1475 
1476 	GIANT_REQUIRED;
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 	GIANT_REQUIRED;
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 (object == NULL)
1509 			break;
1510 
1511 		if ((backing_object = object->backing_object) == NULL)
1512 			break;
1513 
1514 		/*
1515 		 * we check the backing object first, because it is most likely
1516 		 * not collapsable.
1517 		 */
1518 		if (backing_object->handle != NULL ||
1519 		    (backing_object->type != OBJT_DEFAULT &&
1520 		     backing_object->type != OBJT_SWAP) ||
1521 		    (backing_object->flags & OBJ_DEAD) ||
1522 		    object->handle != NULL ||
1523 		    (object->type != OBJT_DEFAULT &&
1524 		     object->type != OBJT_SWAP) ||
1525 		    (object->flags & OBJ_DEAD)) {
1526 			break;
1527 		}
1528 
1529 		if (
1530 		    object->paging_in_progress != 0 ||
1531 		    backing_object->paging_in_progress != 0
1532 		) {
1533 			vm_object_qcollapse(object);
1534 			break;
1535 		}
1536 
1537 		/*
1538 		 * We know that we can either collapse the backing object (if
1539 		 * the parent is the only reference to it) or (perhaps) have
1540 		 * the parent bypass the object if the parent happens to shadow
1541 		 * all the resident pages in the entire backing object.
1542 		 *
1543 		 * This is ignoring pager-backed pages such as swap pages.
1544 		 * vm_object_backing_scan fails the shadowing test in this
1545 		 * case.
1546 		 */
1547 		if (backing_object->ref_count == 1) {
1548 			/*
1549 			 * If there is exactly one reference to the backing
1550 			 * object, we can collapse it into the parent.
1551 			 */
1552 			vm_object_backing_scan(object, OBSC_COLLAPSE_WAIT);
1553 
1554 			/*
1555 			 * Move the pager from backing_object to object.
1556 			 */
1557 			if (backing_object->type == OBJT_SWAP) {
1558 				vm_object_pip_add(backing_object, 1);
1559 
1560 				/*
1561 				 * scrap the paging_offset junk and do a
1562 				 * discrete copy.  This also removes major
1563 				 * assumptions about how the swap-pager
1564 				 * works from where it doesn't belong.  The
1565 				 * new swapper is able to optimize the
1566 				 * destroy-source case.
1567 				 */
1568 				vm_object_pip_add(object, 1);
1569 				swap_pager_copy(
1570 				    backing_object,
1571 				    object,
1572 				    OFF_TO_IDX(object->backing_object_offset), TRUE);
1573 				vm_object_pip_wakeup(object);
1574 
1575 				vm_object_pip_wakeup(backing_object);
1576 			}
1577 			/*
1578 			 * Object now shadows whatever backing_object did.
1579 			 * Note that the reference to
1580 			 * backing_object->backing_object moves from within
1581 			 * backing_object to within object.
1582 			 */
1583 			TAILQ_REMOVE(
1584 			    &object->backing_object->shadow_head,
1585 			    object,
1586 			    shadow_list
1587 			);
1588 			object->backing_object->shadow_count--;
1589 			object->backing_object->generation++;
1590 			if (backing_object->backing_object) {
1591 				TAILQ_REMOVE(
1592 				    &backing_object->backing_object->shadow_head,
1593 				    backing_object,
1594 				    shadow_list
1595 				);
1596 				backing_object->backing_object->shadow_count--;
1597 				backing_object->backing_object->generation++;
1598 			}
1599 			object->backing_object = backing_object->backing_object;
1600 			if (object->backing_object) {
1601 				TAILQ_INSERT_TAIL(
1602 				    &object->backing_object->shadow_head,
1603 				    object,
1604 				    shadow_list
1605 				);
1606 				object->backing_object->shadow_count++;
1607 				object->backing_object->generation++;
1608 			}
1609 
1610 			object->backing_object_offset +=
1611 			    backing_object->backing_object_offset;
1612 
1613 			/*
1614 			 * Discard backing_object.
1615 			 *
1616 			 * Since the backing object has no pages, no pager left,
1617 			 * and no object references within it, all that is
1618 			 * necessary is to dispose of it.
1619 			 */
1620 			KASSERT(backing_object->ref_count == 1, ("backing_object %p was somehow re-referenced during collapse!", backing_object));
1621 			KASSERT(TAILQ_FIRST(&backing_object->memq) == NULL, ("backing_object %p somehow has left over pages during collapse!", 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 				break;
1643 			}
1644 
1645 			/*
1646 			 * Make the parent shadow the next object in the
1647 			 * chain.  Deallocating backing_object will not remove
1648 			 * it, since its reference count is at least 2.
1649 			 */
1650 			TAILQ_REMOVE(
1651 			    &backing_object->shadow_head,
1652 			    object,
1653 			    shadow_list
1654 			);
1655 			backing_object->shadow_count--;
1656 			backing_object->generation++;
1657 
1658 			new_backing_object = backing_object->backing_object;
1659 			if ((object->backing_object = new_backing_object) != NULL) {
1660 				vm_object_reference(new_backing_object);
1661 				TAILQ_INSERT_TAIL(
1662 				    &new_backing_object->shadow_head,
1663 				    object,
1664 				    shadow_list
1665 				);
1666 				new_backing_object->shadow_count++;
1667 				new_backing_object->generation++;
1668 				object->backing_object_offset +=
1669 					backing_object->backing_object_offset;
1670 			}
1671 
1672 			/*
1673 			 * Drop the reference count on backing_object. Since
1674 			 * its ref_count was at least 2, it will not vanish;
1675 			 * so we don't need to call vm_object_deallocate, but
1676 			 * we do anyway.
1677 			 */
1678 			vm_object_deallocate(backing_object);
1679 			object_bypasses++;
1680 		}
1681 
1682 		/*
1683 		 * Try again with this object's new backing object.
1684 		 */
1685 	}
1686 }
1687 
1688 /*
1689  *	vm_object_page_remove: [internal]
1690  *
1691  *	Removes all physical pages in the specified
1692  *	object range from the object's list of pages.
1693  *
1694  *	The object must be locked.
1695  */
1696 void
1697 vm_object_page_remove(vm_object_t object, vm_pindex_t start, vm_pindex_t end, boolean_t clean_only)
1698 {
1699 	vm_page_t p, next;
1700 	unsigned int size;
1701 	int all;
1702 
1703 	if (object == NULL)
1704 		return;
1705 
1706 	mtx_lock(&Giant);
1707 	if (object->resident_page_count == 0) {
1708 		mtx_unlock(&Giant);
1709 		return;
1710 	}
1711 	all = ((end == 0) && (start == 0));
1712 
1713 	/*
1714 	 * Since physically-backed objects do not use managed pages, we can't
1715 	 * remove pages from the object (we must instead remove the page
1716 	 * references, and then destroy the object).
1717 	 */
1718 	KASSERT(object->type != OBJT_PHYS, ("attempt to remove pages from a physical object"));
1719 
1720 	vm_object_pip_add(object, 1);
1721 again:
1722 	size = end - start;
1723 	if (all || size > object->resident_page_count / 4) {
1724 		for (p = TAILQ_FIRST(&object->memq); p != NULL; p = next) {
1725 			next = TAILQ_NEXT(p, listq);
1726 			if (all || ((start <= p->pindex) && (p->pindex < end))) {
1727 				if (p->wire_count != 0) {
1728 					vm_page_protect(p, VM_PROT_NONE);
1729 					if (!clean_only)
1730 						p->valid = 0;
1731 					continue;
1732 				}
1733 
1734 				/*
1735 				 * The busy flags are only cleared at
1736 				 * interrupt -- minimize the spl transitions
1737 				 */
1738  				if (vm_page_sleep_busy(p, TRUE, "vmopar"))
1739  					goto again;
1740 
1741 				if (clean_only && p->valid) {
1742 					vm_page_test_dirty(p);
1743 					if (p->valid & p->dirty)
1744 						continue;
1745 				}
1746 
1747 				vm_page_busy(p);
1748 				vm_page_protect(p, VM_PROT_NONE);
1749 				vm_page_free(p);
1750 			}
1751 		}
1752 	} else {
1753 		while (size > 0) {
1754 			if ((p = vm_page_lookup(object, start)) != 0) {
1755 
1756 				if (p->wire_count != 0) {
1757 					vm_page_protect(p, VM_PROT_NONE);
1758 					if (!clean_only)
1759 						p->valid = 0;
1760 					start += 1;
1761 					size -= 1;
1762 					continue;
1763 				}
1764 
1765 				/*
1766 				 * The busy flags are only cleared at
1767 				 * interrupt -- minimize the spl transitions
1768 				 */
1769  				if (vm_page_sleep_busy(p, TRUE, "vmopar"))
1770 					goto again;
1771 
1772 				if (clean_only && p->valid) {
1773 					vm_page_test_dirty(p);
1774 					if (p->valid & p->dirty) {
1775 						start += 1;
1776 						size -= 1;
1777 						continue;
1778 					}
1779 				}
1780 
1781 				vm_page_busy(p);
1782 				vm_page_protect(p, VM_PROT_NONE);
1783 				vm_page_free(p);
1784 			}
1785 			start += 1;
1786 			size -= 1;
1787 		}
1788 	}
1789 	vm_object_pip_wakeup(object);
1790 	mtx_unlock(&Giant);
1791 }
1792 
1793 /*
1794  *	Routine:	vm_object_coalesce
1795  *	Function:	Coalesces two objects backing up adjoining
1796  *			regions of memory into a single object.
1797  *
1798  *	returns TRUE if objects were combined.
1799  *
1800  *	NOTE:	Only works at the moment if the second object is NULL -
1801  *		if it's not, which object do we lock first?
1802  *
1803  *	Parameters:
1804  *		prev_object	First object to coalesce
1805  *		prev_offset	Offset into prev_object
1806  *		next_object	Second object into coalesce
1807  *		next_offset	Offset into next_object
1808  *
1809  *		prev_size	Size of reference to prev_object
1810  *		next_size	Size of reference to next_object
1811  *
1812  *	Conditions:
1813  *	The object must *not* be locked.
1814  */
1815 boolean_t
1816 vm_object_coalesce(vm_object_t prev_object, vm_pindex_t prev_pindex,
1817 	vm_size_t prev_size, vm_size_t next_size)
1818 {
1819 	vm_pindex_t next_pindex;
1820 
1821 	if (prev_object == NULL)
1822 		return (TRUE);
1823 	mtx_lock(&Giant);
1824 	if (prev_object->type != OBJT_DEFAULT &&
1825 	    prev_object->type != OBJT_SWAP) {
1826 		mtx_unlock(&Giant);
1827 		return (FALSE);
1828 	}
1829 
1830 	/*
1831 	 * Try to collapse the object first
1832 	 */
1833 	vm_object_collapse(prev_object);
1834 
1835 	/*
1836 	 * Can't coalesce if: . more than one reference . paged out . shadows
1837 	 * another object . has a copy elsewhere (any of which mean that the
1838 	 * pages not mapped to prev_entry may be in use anyway)
1839 	 */
1840 	if (prev_object->backing_object != NULL) {
1841 		mtx_unlock(&Giant);
1842 		return (FALSE);
1843 	}
1844 
1845 	prev_size >>= PAGE_SHIFT;
1846 	next_size >>= PAGE_SHIFT;
1847 	next_pindex = prev_pindex + prev_size;
1848 
1849 	if ((prev_object->ref_count > 1) &&
1850 	    (prev_object->size != next_pindex)) {
1851 		mtx_unlock(&Giant);
1852 		return (FALSE);
1853 	}
1854 
1855 	/*
1856 	 * Remove any pages that may still be in the object from a previous
1857 	 * deallocation.
1858 	 */
1859 	if (next_pindex < prev_object->size) {
1860 		vm_object_page_remove(prev_object,
1861 				      next_pindex,
1862 				      next_pindex + next_size, FALSE);
1863 		if (prev_object->type == OBJT_SWAP)
1864 			swap_pager_freespace(prev_object,
1865 					     next_pindex, next_size);
1866 	}
1867 
1868 	/*
1869 	 * Extend the object if necessary.
1870 	 */
1871 	if (next_pindex + next_size > prev_object->size)
1872 		prev_object->size = next_pindex + next_size;
1873 
1874 	mtx_unlock(&Giant);
1875 	return (TRUE);
1876 }
1877 
1878 void
1879 vm_object_set_writeable_dirty(vm_object_t object)
1880 {
1881 	struct vnode *vp;
1882 
1883 	vm_object_set_flag(object, OBJ_WRITEABLE|OBJ_MIGHTBEDIRTY);
1884 	if (object->type == OBJT_VNODE &&
1885 	    (vp = (struct vnode *)object->handle) != NULL) {
1886 		if ((vp->v_flag & VOBJDIRTY) == 0) {
1887 			mtx_lock(&vp->v_interlock);
1888 			vp->v_flag |= VOBJDIRTY;
1889 			mtx_unlock(&vp->v_interlock);
1890 		}
1891 	}
1892 }
1893 
1894 #ifdef ENABLE_VFS_IOOPT
1895 /*
1896  * Experimental support for zero-copy I/O
1897  *
1898  * Performs the copy_on_write operations necessary to allow the virtual copies
1899  * into user space to work.  This has to be called for write(2) system calls
1900  * from other processes, file unlinking, and file size shrinkage.
1901  */
1902 void
1903 vm_freeze_copyopts(vm_object_t object, vm_pindex_t froma, vm_pindex_t toa)
1904 {
1905 	int rv;
1906 	vm_object_t robject;
1907 	vm_pindex_t idx;
1908 
1909 	GIANT_REQUIRED;
1910 	if ((object == NULL) ||
1911 		((object->flags & OBJ_OPT) == 0))
1912 		return;
1913 
1914 	if (object->shadow_count > object->ref_count)
1915 		panic("vm_freeze_copyopts: sc > rc");
1916 
1917 	while ((robject = TAILQ_FIRST(&object->shadow_head)) != NULL) {
1918 		vm_pindex_t bo_pindex;
1919 		vm_page_t m_in, m_out;
1920 
1921 		bo_pindex = OFF_TO_IDX(robject->backing_object_offset);
1922 
1923 		vm_object_reference(robject);
1924 
1925 		vm_object_pip_wait(robject, "objfrz");
1926 
1927 		if (robject->ref_count == 1) {
1928 			vm_object_deallocate(robject);
1929 			continue;
1930 		}
1931 
1932 		vm_object_pip_add(robject, 1);
1933 
1934 		for (idx = 0; idx < robject->size; idx++) {
1935 
1936 			m_out = vm_page_grab(robject, idx,
1937 						VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
1938 
1939 			if (m_out->valid == 0) {
1940 				m_in = vm_page_grab(object, bo_pindex + idx,
1941 						VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
1942 				if (m_in->valid == 0) {
1943 					rv = vm_pager_get_pages(object, &m_in, 1, 0);
1944 					if (rv != VM_PAGER_OK) {
1945 						printf("vm_freeze_copyopts: cannot read page from file: %lx\n", (long)m_in->pindex);
1946 						continue;
1947 					}
1948 					vm_page_deactivate(m_in);
1949 				}
1950 
1951 				vm_page_protect(m_in, VM_PROT_NONE);
1952 				pmap_copy_page(m_in, m_out);
1953 				m_out->valid = m_in->valid;
1954 				vm_page_dirty(m_out);
1955 				vm_page_activate(m_out);
1956 				vm_page_wakeup(m_in);
1957 			}
1958 			vm_page_wakeup(m_out);
1959 		}
1960 
1961 		object->shadow_count--;
1962 		object->ref_count--;
1963 		TAILQ_REMOVE(&object->shadow_head, robject, shadow_list);
1964 		robject->backing_object = NULL;
1965 		robject->backing_object_offset = 0;
1966 
1967 		vm_object_pip_wakeup(robject);
1968 		vm_object_deallocate(robject);
1969 	}
1970 
1971 	vm_object_clear_flag(object, OBJ_OPT);
1972 }
1973 #endif
1974 
1975 #include "opt_ddb.h"
1976 #ifdef DDB
1977 #include <sys/kernel.h>
1978 
1979 #include <sys/cons.h>
1980 
1981 #include <ddb/ddb.h>
1982 
1983 static int
1984 _vm_object_in_map(vm_map_t map, vm_object_t object, vm_map_entry_t entry)
1985 {
1986 	vm_map_t tmpm;
1987 	vm_map_entry_t tmpe;
1988 	vm_object_t obj;
1989 	int entcount;
1990 
1991 	if (map == 0)
1992 		return 0;
1993 
1994 	if (entry == 0) {
1995 		tmpe = map->header.next;
1996 		entcount = map->nentries;
1997 		while (entcount-- && (tmpe != &map->header)) {
1998 			if (_vm_object_in_map(map, object, tmpe)) {
1999 				return 1;
2000 			}
2001 			tmpe = tmpe->next;
2002 		}
2003 	} else if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
2004 		tmpm = entry->object.sub_map;
2005 		tmpe = tmpm->header.next;
2006 		entcount = tmpm->nentries;
2007 		while (entcount-- && tmpe != &tmpm->header) {
2008 			if (_vm_object_in_map(tmpm, object, tmpe)) {
2009 				return 1;
2010 			}
2011 			tmpe = tmpe->next;
2012 		}
2013 	} else if ((obj = entry->object.vm_object) != NULL) {
2014 		for (; obj; obj = obj->backing_object)
2015 			if (obj == object) {
2016 				return 1;
2017 			}
2018 	}
2019 	return 0;
2020 }
2021 
2022 static int
2023 vm_object_in_map(vm_object_t object)
2024 {
2025 	struct proc *p;
2026 
2027 	/* sx_slock(&allproc_lock); */
2028 	LIST_FOREACH(p, &allproc, p_list) {
2029 		if (!p->p_vmspace /* || (p->p_flag & (P_SYSTEM|P_WEXIT)) */)
2030 			continue;
2031 		if (_vm_object_in_map(&p->p_vmspace->vm_map, object, 0)) {
2032 			/* sx_sunlock(&allproc_lock); */
2033 			return 1;
2034 		}
2035 	}
2036 	/* sx_sunlock(&allproc_lock); */
2037 	if (_vm_object_in_map(kernel_map, object, 0))
2038 		return 1;
2039 	if (_vm_object_in_map(kmem_map, object, 0))
2040 		return 1;
2041 	if (_vm_object_in_map(pager_map, object, 0))
2042 		return 1;
2043 	if (_vm_object_in_map(buffer_map, object, 0))
2044 		return 1;
2045 	return 0;
2046 }
2047 
2048 DB_SHOW_COMMAND(vmochk, vm_object_check)
2049 {
2050 	vm_object_t object;
2051 
2052 	/*
2053 	 * make sure that internal objs are in a map somewhere
2054 	 * and none have zero ref counts.
2055 	 */
2056 	TAILQ_FOREACH(object, &vm_object_list, object_list) {
2057 		if (object->handle == NULL &&
2058 		    (object->type == OBJT_DEFAULT || object->type == OBJT_SWAP)) {
2059 			if (object->ref_count == 0) {
2060 				db_printf("vmochk: internal obj has zero ref count: %ld\n",
2061 					(long)object->size);
2062 			}
2063 			if (!vm_object_in_map(object)) {
2064 				db_printf(
2065 			"vmochk: internal obj is not in a map: "
2066 			"ref: %d, size: %lu: 0x%lx, backing_object: %p\n",
2067 				    object->ref_count, (u_long)object->size,
2068 				    (u_long)object->size,
2069 				    (void *)object->backing_object);
2070 			}
2071 		}
2072 	}
2073 }
2074 
2075 /*
2076  *	vm_object_print:	[ debug ]
2077  */
2078 DB_SHOW_COMMAND(object, vm_object_print_static)
2079 {
2080 	/* XXX convert args. */
2081 	vm_object_t object = (vm_object_t)addr;
2082 	boolean_t full = have_addr;
2083 
2084 	vm_page_t p;
2085 
2086 	/* XXX count is an (unused) arg.  Avoid shadowing it. */
2087 #define	count	was_count
2088 
2089 	int count;
2090 
2091 	if (object == NULL)
2092 		return;
2093 
2094 	db_iprintf(
2095 	    "Object %p: type=%d, size=0x%lx, res=%d, ref=%d, flags=0x%x\n",
2096 	    object, (int)object->type, (u_long)object->size,
2097 	    object->resident_page_count, object->ref_count, object->flags);
2098 	/*
2099 	 * XXX no %qd in kernel.  Truncate object->backing_object_offset.
2100 	 */
2101 	db_iprintf(" sref=%d, backing_object(%d)=(%p)+0x%lx\n",
2102 	    object->shadow_count,
2103 	    object->backing_object ? object->backing_object->ref_count : 0,
2104 	    object->backing_object, (long)object->backing_object_offset);
2105 
2106 	if (!full)
2107 		return;
2108 
2109 	db_indent += 2;
2110 	count = 0;
2111 	TAILQ_FOREACH(p, &object->memq, listq) {
2112 		if (count == 0)
2113 			db_iprintf("memory:=");
2114 		else if (count == 6) {
2115 			db_printf("\n");
2116 			db_iprintf(" ...");
2117 			count = 0;
2118 		} else
2119 			db_printf(",");
2120 		count++;
2121 
2122 		db_printf("(off=0x%lx,page=0x%lx)",
2123 		    (u_long) p->pindex, (u_long) VM_PAGE_TO_PHYS(p));
2124 	}
2125 	if (count != 0)
2126 		db_printf("\n");
2127 	db_indent -= 2;
2128 }
2129 
2130 /* XXX. */
2131 #undef count
2132 
2133 /* XXX need this non-static entry for calling from vm_map_print. */
2134 void
2135 vm_object_print(
2136         /* db_expr_t */ long addr,
2137 	boolean_t have_addr,
2138 	/* db_expr_t */ long count,
2139 	char *modif)
2140 {
2141 	vm_object_print_static(addr, have_addr, count, modif);
2142 }
2143 
2144 DB_SHOW_COMMAND(vmopag, vm_object_print_pages)
2145 {
2146 	vm_object_t object;
2147 	int nl = 0;
2148 	int c;
2149 
2150 	TAILQ_FOREACH(object, &vm_object_list, object_list) {
2151 		vm_pindex_t idx, fidx;
2152 		vm_pindex_t osize;
2153 		vm_offset_t pa = -1, padiff;
2154 		int rcount;
2155 		vm_page_t m;
2156 
2157 		db_printf("new object: %p\n", (void *)object);
2158 		if (nl > 18) {
2159 			c = cngetc();
2160 			if (c != ' ')
2161 				return;
2162 			nl = 0;
2163 		}
2164 		nl++;
2165 		rcount = 0;
2166 		fidx = 0;
2167 		osize = object->size;
2168 		if (osize > 128)
2169 			osize = 128;
2170 		for (idx = 0; idx < osize; idx++) {
2171 			m = vm_page_lookup(object, idx);
2172 			if (m == NULL) {
2173 				if (rcount) {
2174 					db_printf(" index(%ld)run(%d)pa(0x%lx)\n",
2175 						(long)fidx, rcount, (long)pa);
2176 					if (nl > 18) {
2177 						c = cngetc();
2178 						if (c != ' ')
2179 							return;
2180 						nl = 0;
2181 					}
2182 					nl++;
2183 					rcount = 0;
2184 				}
2185 				continue;
2186 			}
2187 
2188 
2189 			if (rcount &&
2190 				(VM_PAGE_TO_PHYS(m) == pa + rcount * PAGE_SIZE)) {
2191 				++rcount;
2192 				continue;
2193 			}
2194 			if (rcount) {
2195 				padiff = pa + rcount * PAGE_SIZE - VM_PAGE_TO_PHYS(m);
2196 				padiff >>= PAGE_SHIFT;
2197 				padiff &= PQ_L2_MASK;
2198 				if (padiff == 0) {
2199 					pa = VM_PAGE_TO_PHYS(m) - rcount * PAGE_SIZE;
2200 					++rcount;
2201 					continue;
2202 				}
2203 				db_printf(" index(%ld)run(%d)pa(0x%lx)",
2204 					(long)fidx, rcount, (long)pa);
2205 				db_printf("pd(%ld)\n", (long)padiff);
2206 				if (nl > 18) {
2207 					c = cngetc();
2208 					if (c != ' ')
2209 						return;
2210 					nl = 0;
2211 				}
2212 				nl++;
2213 			}
2214 			fidx = idx;
2215 			pa = VM_PAGE_TO_PHYS(m);
2216 			rcount = 1;
2217 		}
2218 		if (rcount) {
2219 			db_printf(" index(%ld)run(%d)pa(0x%lx)\n",
2220 				(long)fidx, rcount, (long)pa);
2221 			if (nl > 18) {
2222 				c = cngetc();
2223 				if (c != ' ')
2224 					return;
2225 				nl = 0;
2226 			}
2227 			nl++;
2228 		}
2229 	}
2230 }
2231 #endif /* DDB */
2232