xref: /freebsd/sys/vm/vm_object.c (revision 52f72944b8f5abb2386eae924357dee8aea17d5b)
1 /*-
2  * SPDX-License-Identifier: (BSD-3-Clause AND MIT-CMU)
3  *
4  * Copyright (c) 1991, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * The Mach Operating System project at Carnegie-Mellon University.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	from: @(#)vm_object.c	8.5 (Berkeley) 3/22/94
35  *
36  *
37  * Copyright (c) 1987, 1990 Carnegie-Mellon University.
38  * All rights reserved.
39  *
40  * Authors: Avadis Tevanian, Jr., Michael Wayne Young
41  *
42  * Permission to use, copy, modify and distribute this software and
43  * its documentation is hereby granted, provided that both the copyright
44  * notice and this permission notice appear in all copies of the
45  * software, derivative works or modified versions, and any portions
46  * thereof, and that both notices appear in supporting documentation.
47  *
48  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
49  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
50  * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
51  *
52  * Carnegie Mellon requests users of this software to return to
53  *
54  *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
55  *  School of Computer Science
56  *  Carnegie Mellon University
57  *  Pittsburgh PA 15213-3890
58  *
59  * any improvements or extensions that they make and grant Carnegie the
60  * rights to redistribute these changes.
61  */
62 
63 /*
64  *	Virtual memory object module.
65  */
66 
67 #include <sys/cdefs.h>
68 __FBSDID("$FreeBSD$");
69 
70 #include "opt_vm.h"
71 
72 #include <sys/param.h>
73 #include <sys/systm.h>
74 #include <sys/cpuset.h>
75 #include <sys/lock.h>
76 #include <sys/mman.h>
77 #include <sys/mount.h>
78 #include <sys/kernel.h>
79 #include <sys/pctrie.h>
80 #include <sys/sysctl.h>
81 #include <sys/mutex.h>
82 #include <sys/proc.h>		/* for curproc, pageproc */
83 #include <sys/socket.h>
84 #include <sys/resourcevar.h>
85 #include <sys/rwlock.h>
86 #include <sys/user.h>
87 #include <sys/vnode.h>
88 #include <sys/vmmeter.h>
89 #include <sys/sx.h>
90 
91 #include <vm/vm.h>
92 #include <vm/vm_param.h>
93 #include <vm/pmap.h>
94 #include <vm/vm_map.h>
95 #include <vm/vm_object.h>
96 #include <vm/vm_page.h>
97 #include <vm/vm_pageout.h>
98 #include <vm/vm_pager.h>
99 #include <vm/vm_phys.h>
100 #include <vm/vm_pagequeue.h>
101 #include <vm/swap_pager.h>
102 #include <vm/vm_kern.h>
103 #include <vm/vm_extern.h>
104 #include <vm/vm_radix.h>
105 #include <vm/vm_reserv.h>
106 #include <vm/uma.h>
107 
108 static int old_msync;
109 SYSCTL_INT(_vm, OID_AUTO, old_msync, CTLFLAG_RW, &old_msync, 0,
110     "Use old (insecure) msync behavior");
111 
112 static int	vm_object_page_collect_flush(vm_object_t object, vm_page_t p,
113 		    int pagerflags, int flags, boolean_t *clearobjflags,
114 		    boolean_t *eio);
115 static boolean_t vm_object_page_remove_write(vm_page_t p, int flags,
116 		    boolean_t *clearobjflags);
117 static void	vm_object_qcollapse(vm_object_t object);
118 static void	vm_object_vndeallocate(vm_object_t object);
119 
120 /*
121  *	Virtual memory objects maintain the actual data
122  *	associated with allocated virtual memory.  A given
123  *	page of memory exists within exactly one object.
124  *
125  *	An object is only deallocated when all "references"
126  *	are given up.  Only one "reference" to a given
127  *	region of an object should be writeable.
128  *
129  *	Associated with each object is a list of all resident
130  *	memory pages belonging to that object; this list is
131  *	maintained by the "vm_page" module, and locked by the object's
132  *	lock.
133  *
134  *	Each object also records a "pager" routine which is
135  *	used to retrieve (and store) pages to the proper backing
136  *	storage.  In addition, objects may be backed by other
137  *	objects from which they were virtual-copied.
138  *
139  *	The only items within the object structure which are
140  *	modified after time of creation are:
141  *		reference count		locked by object's lock
142  *		pager routine		locked by object's lock
143  *
144  */
145 
146 struct object_q vm_object_list;
147 struct mtx vm_object_list_mtx;	/* lock for object list and count */
148 
149 struct vm_object kernel_object_store;
150 
151 static SYSCTL_NODE(_vm_stats, OID_AUTO, object, CTLFLAG_RD, 0,
152     "VM object stats");
153 
154 static counter_u64_t object_collapses = EARLY_COUNTER;
155 SYSCTL_COUNTER_U64(_vm_stats_object, OID_AUTO, collapses, CTLFLAG_RD,
156     &object_collapses,
157     "VM object collapses");
158 
159 static counter_u64_t object_bypasses = EARLY_COUNTER;
160 SYSCTL_COUNTER_U64(_vm_stats_object, OID_AUTO, bypasses, CTLFLAG_RD,
161     &object_bypasses,
162     "VM object bypasses");
163 
164 static void
165 counter_startup(void)
166 {
167 
168 	object_collapses = counter_u64_alloc(M_WAITOK);
169 	object_bypasses = counter_u64_alloc(M_WAITOK);
170 }
171 SYSINIT(object_counters, SI_SUB_CPU, SI_ORDER_ANY, counter_startup, NULL);
172 
173 static uma_zone_t obj_zone;
174 
175 static int vm_object_zinit(void *mem, int size, int flags);
176 
177 #ifdef INVARIANTS
178 static void vm_object_zdtor(void *mem, int size, void *arg);
179 
180 static void
181 vm_object_zdtor(void *mem, int size, void *arg)
182 {
183 	vm_object_t object;
184 
185 	object = (vm_object_t)mem;
186 	KASSERT(object->ref_count == 0,
187 	    ("object %p ref_count = %d", object, object->ref_count));
188 	KASSERT(TAILQ_EMPTY(&object->memq),
189 	    ("object %p has resident pages in its memq", object));
190 	KASSERT(vm_radix_is_empty(&object->rtree),
191 	    ("object %p has resident pages in its trie", object));
192 #if VM_NRESERVLEVEL > 0
193 	KASSERT(LIST_EMPTY(&object->rvq),
194 	    ("object %p has reservations",
195 	    object));
196 #endif
197 	KASSERT(object->paging_in_progress == 0,
198 	    ("object %p paging_in_progress = %d",
199 	    object, object->paging_in_progress));
200 	KASSERT(object->resident_page_count == 0,
201 	    ("object %p resident_page_count = %d",
202 	    object, object->resident_page_count));
203 	KASSERT(object->shadow_count == 0,
204 	    ("object %p shadow_count = %d",
205 	    object, object->shadow_count));
206 	KASSERT(object->type == OBJT_DEAD,
207 	    ("object %p has non-dead type %d",
208 	    object, object->type));
209 }
210 #endif
211 
212 static int
213 vm_object_zinit(void *mem, int size, int flags)
214 {
215 	vm_object_t object;
216 
217 	object = (vm_object_t)mem;
218 	rw_init_flags(&object->lock, "vm object", RW_DUPOK | RW_NEW);
219 
220 	/* These are true for any object that has been freed */
221 	object->type = OBJT_DEAD;
222 	object->ref_count = 0;
223 	vm_radix_init(&object->rtree);
224 	object->paging_in_progress = 0;
225 	object->resident_page_count = 0;
226 	object->shadow_count = 0;
227 	object->flags = OBJ_DEAD;
228 
229 	mtx_lock(&vm_object_list_mtx);
230 	TAILQ_INSERT_TAIL(&vm_object_list, object, object_list);
231 	mtx_unlock(&vm_object_list_mtx);
232 	return (0);
233 }
234 
235 static void
236 _vm_object_allocate(objtype_t type, vm_pindex_t size, vm_object_t object)
237 {
238 
239 	TAILQ_INIT(&object->memq);
240 	LIST_INIT(&object->shadow_head);
241 
242 	object->type = type;
243 	if (type == OBJT_SWAP)
244 		pctrie_init(&object->un_pager.swp.swp_blks);
245 
246 	/*
247 	 * Ensure that swap_pager_swapoff() iteration over object_list
248 	 * sees up to date type and pctrie head if it observed
249 	 * non-dead object.
250 	 */
251 	atomic_thread_fence_rel();
252 
253 	switch (type) {
254 	case OBJT_DEAD:
255 		panic("_vm_object_allocate: can't create OBJT_DEAD");
256 	case OBJT_DEFAULT:
257 	case OBJT_SWAP:
258 		object->flags = OBJ_ONEMAPPING;
259 		break;
260 	case OBJT_DEVICE:
261 	case OBJT_SG:
262 		object->flags = OBJ_FICTITIOUS | OBJ_UNMANAGED;
263 		break;
264 	case OBJT_MGTDEVICE:
265 		object->flags = OBJ_FICTITIOUS;
266 		break;
267 	case OBJT_PHYS:
268 		object->flags = OBJ_UNMANAGED;
269 		break;
270 	case OBJT_VNODE:
271 		object->flags = 0;
272 		break;
273 	default:
274 		panic("_vm_object_allocate: type %d is undefined", type);
275 	}
276 	object->size = size;
277 	object->generation = 1;
278 	object->ref_count = 1;
279 	object->memattr = VM_MEMATTR_DEFAULT;
280 	object->cred = NULL;
281 	object->charge = 0;
282 	object->handle = NULL;
283 	object->backing_object = NULL;
284 	object->backing_object_offset = (vm_ooffset_t) 0;
285 #if VM_NRESERVLEVEL > 0
286 	LIST_INIT(&object->rvq);
287 #endif
288 	umtx_shm_object_init(object);
289 }
290 
291 /*
292  *	vm_object_init:
293  *
294  *	Initialize the VM objects module.
295  */
296 void
297 vm_object_init(void)
298 {
299 	TAILQ_INIT(&vm_object_list);
300 	mtx_init(&vm_object_list_mtx, "vm object_list", NULL, MTX_DEF);
301 
302 	rw_init(&kernel_object->lock, "kernel vm object");
303 	_vm_object_allocate(OBJT_PHYS, atop(VM_MAX_KERNEL_ADDRESS -
304 	    VM_MIN_KERNEL_ADDRESS), kernel_object);
305 #if VM_NRESERVLEVEL > 0
306 	kernel_object->flags |= OBJ_COLORED;
307 	kernel_object->pg_color = (u_short)atop(VM_MIN_KERNEL_ADDRESS);
308 #endif
309 
310 	/*
311 	 * The lock portion of struct vm_object must be type stable due
312 	 * to vm_pageout_fallback_object_lock locking a vm object
313 	 * without holding any references to it.
314 	 */
315 	obj_zone = uma_zcreate("VM OBJECT", sizeof (struct vm_object), NULL,
316 #ifdef INVARIANTS
317 	    vm_object_zdtor,
318 #else
319 	    NULL,
320 #endif
321 	    vm_object_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
322 
323 	vm_radix_zinit();
324 }
325 
326 void
327 vm_object_clear_flag(vm_object_t object, u_short bits)
328 {
329 
330 	VM_OBJECT_ASSERT_WLOCKED(object);
331 	object->flags &= ~bits;
332 }
333 
334 /*
335  *	Sets the default memory attribute for the specified object.  Pages
336  *	that are allocated to this object are by default assigned this memory
337  *	attribute.
338  *
339  *	Presently, this function must be called before any pages are allocated
340  *	to the object.  In the future, this requirement may be relaxed for
341  *	"default" and "swap" objects.
342  */
343 int
344 vm_object_set_memattr(vm_object_t object, vm_memattr_t memattr)
345 {
346 
347 	VM_OBJECT_ASSERT_WLOCKED(object);
348 	switch (object->type) {
349 	case OBJT_DEFAULT:
350 	case OBJT_DEVICE:
351 	case OBJT_MGTDEVICE:
352 	case OBJT_PHYS:
353 	case OBJT_SG:
354 	case OBJT_SWAP:
355 	case OBJT_VNODE:
356 		if (!TAILQ_EMPTY(&object->memq))
357 			return (KERN_FAILURE);
358 		break;
359 	case OBJT_DEAD:
360 		return (KERN_INVALID_ARGUMENT);
361 	default:
362 		panic("vm_object_set_memattr: object %p is of undefined type",
363 		    object);
364 	}
365 	object->memattr = memattr;
366 	return (KERN_SUCCESS);
367 }
368 
369 void
370 vm_object_pip_add(vm_object_t object, short i)
371 {
372 
373 	VM_OBJECT_ASSERT_WLOCKED(object);
374 	object->paging_in_progress += i;
375 }
376 
377 void
378 vm_object_pip_subtract(vm_object_t object, short i)
379 {
380 
381 	VM_OBJECT_ASSERT_WLOCKED(object);
382 	object->paging_in_progress -= i;
383 }
384 
385 void
386 vm_object_pip_wakeup(vm_object_t object)
387 {
388 
389 	VM_OBJECT_ASSERT_WLOCKED(object);
390 	object->paging_in_progress--;
391 	if ((object->flags & OBJ_PIPWNT) && object->paging_in_progress == 0) {
392 		vm_object_clear_flag(object, OBJ_PIPWNT);
393 		wakeup(object);
394 	}
395 }
396 
397 void
398 vm_object_pip_wakeupn(vm_object_t object, short i)
399 {
400 
401 	VM_OBJECT_ASSERT_WLOCKED(object);
402 	if (i)
403 		object->paging_in_progress -= i;
404 	if ((object->flags & OBJ_PIPWNT) && object->paging_in_progress == 0) {
405 		vm_object_clear_flag(object, OBJ_PIPWNT);
406 		wakeup(object);
407 	}
408 }
409 
410 void
411 vm_object_pip_wait(vm_object_t object, char *waitid)
412 {
413 
414 	VM_OBJECT_ASSERT_WLOCKED(object);
415 	while (object->paging_in_progress) {
416 		object->flags |= OBJ_PIPWNT;
417 		VM_OBJECT_SLEEP(object, object, PVM, waitid, 0);
418 	}
419 }
420 
421 /*
422  *	vm_object_allocate:
423  *
424  *	Returns a new object with the given size.
425  */
426 vm_object_t
427 vm_object_allocate(objtype_t type, vm_pindex_t size)
428 {
429 	vm_object_t object;
430 
431 	object = (vm_object_t)uma_zalloc(obj_zone, M_WAITOK);
432 	_vm_object_allocate(type, size, object);
433 	return (object);
434 }
435 
436 
437 /*
438  *	vm_object_reference:
439  *
440  *	Gets another reference to the given object.  Note: OBJ_DEAD
441  *	objects can be referenced during final cleaning.
442  */
443 void
444 vm_object_reference(vm_object_t object)
445 {
446 	if (object == NULL)
447 		return;
448 	VM_OBJECT_WLOCK(object);
449 	vm_object_reference_locked(object);
450 	VM_OBJECT_WUNLOCK(object);
451 }
452 
453 /*
454  *	vm_object_reference_locked:
455  *
456  *	Gets another reference to the given object.
457  *
458  *	The object must be locked.
459  */
460 void
461 vm_object_reference_locked(vm_object_t object)
462 {
463 	struct vnode *vp;
464 
465 	VM_OBJECT_ASSERT_WLOCKED(object);
466 	object->ref_count++;
467 	if (object->type == OBJT_VNODE) {
468 		vp = object->handle;
469 		vref(vp);
470 	}
471 }
472 
473 /*
474  * Handle deallocating an object of type OBJT_VNODE.
475  */
476 static void
477 vm_object_vndeallocate(vm_object_t object)
478 {
479 	struct vnode *vp = (struct vnode *) object->handle;
480 
481 	VM_OBJECT_ASSERT_WLOCKED(object);
482 	KASSERT(object->type == OBJT_VNODE,
483 	    ("vm_object_vndeallocate: not a vnode object"));
484 	KASSERT(vp != NULL, ("vm_object_vndeallocate: missing vp"));
485 #ifdef INVARIANTS
486 	if (object->ref_count == 0) {
487 		vn_printf(vp, "vm_object_vndeallocate ");
488 		panic("vm_object_vndeallocate: bad object reference count");
489 	}
490 #endif
491 
492 	if (!umtx_shm_vnobj_persistent && object->ref_count == 1)
493 		umtx_shm_object_terminated(object);
494 
495 	/*
496 	 * The test for text of vp vnode does not need a bypass to
497 	 * reach right VV_TEXT there, since it is obtained from
498 	 * object->handle.
499 	 */
500 	if (object->ref_count > 1 || (vp->v_vflag & VV_TEXT) == 0) {
501 		object->ref_count--;
502 		VM_OBJECT_WUNLOCK(object);
503 		/* vrele may need the vnode lock. */
504 		vrele(vp);
505 	} else {
506 		vhold(vp);
507 		VM_OBJECT_WUNLOCK(object);
508 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
509 		vdrop(vp);
510 		VM_OBJECT_WLOCK(object);
511 		object->ref_count--;
512 		if (object->type == OBJT_DEAD) {
513 			VM_OBJECT_WUNLOCK(object);
514 			VOP_UNLOCK(vp, 0);
515 		} else {
516 			if (object->ref_count == 0)
517 				VOP_UNSET_TEXT(vp);
518 			VM_OBJECT_WUNLOCK(object);
519 			vput(vp);
520 		}
521 	}
522 }
523 
524 /*
525  *	vm_object_deallocate:
526  *
527  *	Release a reference to the specified object,
528  *	gained either through a vm_object_allocate
529  *	or a vm_object_reference call.  When all references
530  *	are gone, storage associated with this object
531  *	may be relinquished.
532  *
533  *	No object may be locked.
534  */
535 void
536 vm_object_deallocate(vm_object_t object)
537 {
538 	vm_object_t temp;
539 	struct vnode *vp;
540 
541 	while (object != NULL) {
542 		VM_OBJECT_WLOCK(object);
543 		if (object->type == OBJT_VNODE) {
544 			vm_object_vndeallocate(object);
545 			return;
546 		}
547 
548 		KASSERT(object->ref_count != 0,
549 			("vm_object_deallocate: object deallocated too many times: %d", object->type));
550 
551 		/*
552 		 * If the reference count goes to 0 we start calling
553 		 * vm_object_terminate() on the object chain.
554 		 * A ref count of 1 may be a special case depending on the
555 		 * shadow count being 0 or 1.
556 		 */
557 		object->ref_count--;
558 		if (object->ref_count > 1) {
559 			VM_OBJECT_WUNLOCK(object);
560 			return;
561 		} else if (object->ref_count == 1) {
562 			if (object->type == OBJT_SWAP &&
563 			    (object->flags & OBJ_TMPFS) != 0) {
564 				vp = object->un_pager.swp.swp_tmpfs;
565 				vhold(vp);
566 				VM_OBJECT_WUNLOCK(object);
567 				vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
568 				VM_OBJECT_WLOCK(object);
569 				if (object->type == OBJT_DEAD ||
570 				    object->ref_count != 1) {
571 					VM_OBJECT_WUNLOCK(object);
572 					VOP_UNLOCK(vp, 0);
573 					vdrop(vp);
574 					return;
575 				}
576 				if ((object->flags & OBJ_TMPFS) != 0)
577 					VOP_UNSET_TEXT(vp);
578 				VOP_UNLOCK(vp, 0);
579 				vdrop(vp);
580 			}
581 			if (object->shadow_count == 0 &&
582 			    object->handle == NULL &&
583 			    (object->type == OBJT_DEFAULT ||
584 			    (object->type == OBJT_SWAP &&
585 			    (object->flags & OBJ_TMPFS_NODE) == 0))) {
586 				vm_object_set_flag(object, OBJ_ONEMAPPING);
587 			} else if ((object->shadow_count == 1) &&
588 			    (object->handle == NULL) &&
589 			    (object->type == OBJT_DEFAULT ||
590 			     object->type == OBJT_SWAP)) {
591 				vm_object_t robject;
592 
593 				robject = LIST_FIRST(&object->shadow_head);
594 				KASSERT(robject != NULL,
595 				    ("vm_object_deallocate: ref_count: %d, shadow_count: %d",
596 					 object->ref_count,
597 					 object->shadow_count));
598 				KASSERT((robject->flags & OBJ_TMPFS_NODE) == 0,
599 				    ("shadowed tmpfs v_object %p", object));
600 				if (!VM_OBJECT_TRYWLOCK(robject)) {
601 					/*
602 					 * Avoid a potential deadlock.
603 					 */
604 					object->ref_count++;
605 					VM_OBJECT_WUNLOCK(object);
606 					/*
607 					 * More likely than not the thread
608 					 * holding robject's lock has lower
609 					 * priority than the current thread.
610 					 * Let the lower priority thread run.
611 					 */
612 					pause("vmo_de", 1);
613 					continue;
614 				}
615 				/*
616 				 * Collapse object into its shadow unless its
617 				 * shadow is dead.  In that case, object will
618 				 * be deallocated by the thread that is
619 				 * deallocating its shadow.
620 				 */
621 				if ((robject->flags & OBJ_DEAD) == 0 &&
622 				    (robject->handle == NULL) &&
623 				    (robject->type == OBJT_DEFAULT ||
624 				     robject->type == OBJT_SWAP)) {
625 
626 					robject->ref_count++;
627 retry:
628 					if (robject->paging_in_progress) {
629 						VM_OBJECT_WUNLOCK(object);
630 						vm_object_pip_wait(robject,
631 						    "objde1");
632 						temp = robject->backing_object;
633 						if (object == temp) {
634 							VM_OBJECT_WLOCK(object);
635 							goto retry;
636 						}
637 					} else if (object->paging_in_progress) {
638 						VM_OBJECT_WUNLOCK(robject);
639 						object->flags |= OBJ_PIPWNT;
640 						VM_OBJECT_SLEEP(object, object,
641 						    PDROP | PVM, "objde2", 0);
642 						VM_OBJECT_WLOCK(robject);
643 						temp = robject->backing_object;
644 						if (object == temp) {
645 							VM_OBJECT_WLOCK(object);
646 							goto retry;
647 						}
648 					} else
649 						VM_OBJECT_WUNLOCK(object);
650 
651 					if (robject->ref_count == 1) {
652 						robject->ref_count--;
653 						object = robject;
654 						goto doterm;
655 					}
656 					object = robject;
657 					vm_object_collapse(object);
658 					VM_OBJECT_WUNLOCK(object);
659 					continue;
660 				}
661 				VM_OBJECT_WUNLOCK(robject);
662 			}
663 			VM_OBJECT_WUNLOCK(object);
664 			return;
665 		}
666 doterm:
667 		umtx_shm_object_terminated(object);
668 		temp = object->backing_object;
669 		if (temp != NULL) {
670 			KASSERT((object->flags & OBJ_TMPFS_NODE) == 0,
671 			    ("shadowed tmpfs v_object 2 %p", object));
672 			VM_OBJECT_WLOCK(temp);
673 			LIST_REMOVE(object, shadow_list);
674 			temp->shadow_count--;
675 			VM_OBJECT_WUNLOCK(temp);
676 			object->backing_object = NULL;
677 		}
678 		/*
679 		 * Don't double-terminate, we could be in a termination
680 		 * recursion due to the terminate having to sync data
681 		 * to disk.
682 		 */
683 		if ((object->flags & OBJ_DEAD) == 0)
684 			vm_object_terminate(object);
685 		else
686 			VM_OBJECT_WUNLOCK(object);
687 		object = temp;
688 	}
689 }
690 
691 /*
692  *	vm_object_destroy removes the object from the global object list
693  *      and frees the space for the object.
694  */
695 void
696 vm_object_destroy(vm_object_t object)
697 {
698 
699 	/*
700 	 * Release the allocation charge.
701 	 */
702 	if (object->cred != NULL) {
703 		swap_release_by_cred(object->charge, object->cred);
704 		object->charge = 0;
705 		crfree(object->cred);
706 		object->cred = NULL;
707 	}
708 
709 	/*
710 	 * Free the space for the object.
711 	 */
712 	uma_zfree(obj_zone, object);
713 }
714 
715 /*
716  *	vm_object_terminate_pages removes any remaining pageable pages
717  *	from the object and resets the object to an empty state.
718  */
719 static void
720 vm_object_terminate_pages(vm_object_t object)
721 {
722 	vm_page_t p, p_next;
723 	struct mtx *mtx, *mtx1;
724 	struct vm_pagequeue *pq, *pq1;
725 	int dequeued;
726 
727 	VM_OBJECT_ASSERT_WLOCKED(object);
728 
729 	mtx = NULL;
730 	pq = NULL;
731 
732 	/*
733 	 * Free any remaining pageable pages.  This also removes them from the
734 	 * paging queues.  However, don't free wired pages, just remove them
735 	 * from the object.  Rather than incrementally removing each page from
736 	 * the object, the page and object are reset to any empty state.
737 	 */
738 	TAILQ_FOREACH_SAFE(p, &object->memq, listq, p_next) {
739 		vm_page_assert_unbusied(p);
740 		if ((object->flags & OBJ_UNMANAGED) == 0) {
741 			/*
742 			 * vm_page_free_prep() only needs the page
743 			 * lock for managed pages.
744 			 */
745 			mtx1 = vm_page_lockptr(p);
746 			if (mtx1 != mtx) {
747 				if (mtx != NULL)
748 					mtx_unlock(mtx);
749 				if (pq != NULL) {
750 					vm_pagequeue_cnt_add(pq, dequeued);
751 					vm_pagequeue_unlock(pq);
752 					pq = NULL;
753 				}
754 				mtx = mtx1;
755 				mtx_lock(mtx);
756 			}
757 		}
758 		p->object = NULL;
759 		if (p->wire_count != 0)
760 			goto unlist;
761 		VM_CNT_INC(v_pfree);
762 		p->flags &= ~PG_ZERO;
763 		if (p->queue != PQ_NONE) {
764 			KASSERT(p->queue < PQ_COUNT, ("vm_object_terminate: "
765 			    "page %p is not queued", p));
766 			pq1 = vm_page_pagequeue(p);
767 			if (pq != pq1) {
768 				if (pq != NULL) {
769 					vm_pagequeue_cnt_add(pq, dequeued);
770 					vm_pagequeue_unlock(pq);
771 				}
772 				pq = pq1;
773 				vm_pagequeue_lock(pq);
774 				dequeued = 0;
775 			}
776 			p->queue = PQ_NONE;
777 			TAILQ_REMOVE(&pq->pq_pl, p, plinks.q);
778 			dequeued--;
779 		}
780 		if (vm_page_free_prep(p, true))
781 			continue;
782 unlist:
783 		TAILQ_REMOVE(&object->memq, p, listq);
784 	}
785 	if (pq != NULL) {
786 		vm_pagequeue_cnt_add(pq, dequeued);
787 		vm_pagequeue_unlock(pq);
788 	}
789 	if (mtx != NULL)
790 		mtx_unlock(mtx);
791 
792 	vm_page_free_phys_pglist(&object->memq);
793 
794 	/*
795 	 * If the object contained any pages, then reset it to an empty state.
796 	 * None of the object's fields, including "resident_page_count", were
797 	 * modified by the preceding loop.
798 	 */
799 	if (object->resident_page_count != 0) {
800 		vm_radix_reclaim_allnodes(&object->rtree);
801 		TAILQ_INIT(&object->memq);
802 		object->resident_page_count = 0;
803 		if (object->type == OBJT_VNODE)
804 			vdrop(object->handle);
805 	}
806 }
807 
808 /*
809  *	vm_object_terminate actually destroys the specified object, freeing
810  *	up all previously used resources.
811  *
812  *	The object must be locked.
813  *	This routine may block.
814  */
815 void
816 vm_object_terminate(vm_object_t object)
817 {
818 
819 	VM_OBJECT_ASSERT_WLOCKED(object);
820 
821 	/*
822 	 * Make sure no one uses us.
823 	 */
824 	vm_object_set_flag(object, OBJ_DEAD);
825 
826 	/*
827 	 * wait for the pageout daemon to be done with the object
828 	 */
829 	vm_object_pip_wait(object, "objtrm");
830 
831 	KASSERT(!object->paging_in_progress,
832 		("vm_object_terminate: pageout in progress"));
833 
834 	/*
835 	 * Clean and free the pages, as appropriate. All references to the
836 	 * object are gone, so we don't need to lock it.
837 	 */
838 	if (object->type == OBJT_VNODE) {
839 		struct vnode *vp = (struct vnode *)object->handle;
840 
841 		/*
842 		 * Clean pages and flush buffers.
843 		 */
844 		vm_object_page_clean(object, 0, 0, OBJPC_SYNC);
845 		VM_OBJECT_WUNLOCK(object);
846 
847 		vinvalbuf(vp, V_SAVE, 0, 0);
848 
849 		BO_LOCK(&vp->v_bufobj);
850 		vp->v_bufobj.bo_flag |= BO_DEAD;
851 		BO_UNLOCK(&vp->v_bufobj);
852 
853 		VM_OBJECT_WLOCK(object);
854 	}
855 
856 	KASSERT(object->ref_count == 0,
857 		("vm_object_terminate: object with references, ref_count=%d",
858 		object->ref_count));
859 
860 	if ((object->flags & OBJ_PG_DTOR) == 0)
861 		vm_object_terminate_pages(object);
862 
863 #if VM_NRESERVLEVEL > 0
864 	if (__predict_false(!LIST_EMPTY(&object->rvq)))
865 		vm_reserv_break_all(object);
866 #endif
867 
868 	KASSERT(object->cred == NULL || object->type == OBJT_DEFAULT ||
869 	    object->type == OBJT_SWAP,
870 	    ("%s: non-swap obj %p has cred", __func__, object));
871 
872 	/*
873 	 * Let the pager know object is dead.
874 	 */
875 	vm_pager_deallocate(object);
876 	VM_OBJECT_WUNLOCK(object);
877 
878 	vm_object_destroy(object);
879 }
880 
881 /*
882  * Make the page read-only so that we can clear the object flags.  However, if
883  * this is a nosync mmap then the object is likely to stay dirty so do not
884  * mess with the page and do not clear the object flags.  Returns TRUE if the
885  * page should be flushed, and FALSE otherwise.
886  */
887 static boolean_t
888 vm_object_page_remove_write(vm_page_t p, int flags, boolean_t *clearobjflags)
889 {
890 
891 	/*
892 	 * If we have been asked to skip nosync pages and this is a
893 	 * nosync page, skip it.  Note that the object flags were not
894 	 * cleared in this case so we do not have to set them.
895 	 */
896 	if ((flags & OBJPC_NOSYNC) != 0 && (p->oflags & VPO_NOSYNC) != 0) {
897 		*clearobjflags = FALSE;
898 		return (FALSE);
899 	} else {
900 		pmap_remove_write(p);
901 		return (p->dirty != 0);
902 	}
903 }
904 
905 /*
906  *	vm_object_page_clean
907  *
908  *	Clean all dirty pages in the specified range of object.  Leaves page
909  * 	on whatever queue it is currently on.   If NOSYNC is set then do not
910  *	write out pages with VPO_NOSYNC set (originally comes from MAP_NOSYNC),
911  *	leaving the object dirty.
912  *
913  *	When stuffing pages asynchronously, allow clustering.  XXX we need a
914  *	synchronous clustering mode implementation.
915  *
916  *	Odd semantics: if start == end, we clean everything.
917  *
918  *	The object must be locked.
919  *
920  *	Returns FALSE if some page from the range was not written, as
921  *	reported by the pager, and TRUE otherwise.
922  */
923 boolean_t
924 vm_object_page_clean(vm_object_t object, vm_ooffset_t start, vm_ooffset_t end,
925     int flags)
926 {
927 	vm_page_t np, p;
928 	vm_pindex_t pi, tend, tstart;
929 	int curgeneration, n, pagerflags;
930 	boolean_t clearobjflags, eio, res;
931 
932 	VM_OBJECT_ASSERT_WLOCKED(object);
933 
934 	/*
935 	 * The OBJ_MIGHTBEDIRTY flag is only set for OBJT_VNODE
936 	 * objects.  The check below prevents the function from
937 	 * operating on non-vnode objects.
938 	 */
939 	if ((object->flags & OBJ_MIGHTBEDIRTY) == 0 ||
940 	    object->resident_page_count == 0)
941 		return (TRUE);
942 
943 	pagerflags = (flags & (OBJPC_SYNC | OBJPC_INVAL)) != 0 ?
944 	    VM_PAGER_PUT_SYNC : VM_PAGER_CLUSTER_OK;
945 	pagerflags |= (flags & OBJPC_INVAL) != 0 ? VM_PAGER_PUT_INVAL : 0;
946 
947 	tstart = OFF_TO_IDX(start);
948 	tend = (end == 0) ? object->size : OFF_TO_IDX(end + PAGE_MASK);
949 	clearobjflags = tstart == 0 && tend >= object->size;
950 	res = TRUE;
951 
952 rescan:
953 	curgeneration = object->generation;
954 
955 	for (p = vm_page_find_least(object, tstart); p != NULL; p = np) {
956 		pi = p->pindex;
957 		if (pi >= tend)
958 			break;
959 		np = TAILQ_NEXT(p, listq);
960 		if (p->valid == 0)
961 			continue;
962 		if (vm_page_sleep_if_busy(p, "vpcwai")) {
963 			if (object->generation != curgeneration) {
964 				if ((flags & OBJPC_SYNC) != 0)
965 					goto rescan;
966 				else
967 					clearobjflags = FALSE;
968 			}
969 			np = vm_page_find_least(object, pi);
970 			continue;
971 		}
972 		if (!vm_object_page_remove_write(p, flags, &clearobjflags))
973 			continue;
974 
975 		n = vm_object_page_collect_flush(object, p, pagerflags,
976 		    flags, &clearobjflags, &eio);
977 		if (eio) {
978 			res = FALSE;
979 			clearobjflags = FALSE;
980 		}
981 		if (object->generation != curgeneration) {
982 			if ((flags & OBJPC_SYNC) != 0)
983 				goto rescan;
984 			else
985 				clearobjflags = FALSE;
986 		}
987 
988 		/*
989 		 * If the VOP_PUTPAGES() did a truncated write, so
990 		 * that even the first page of the run is not fully
991 		 * written, vm_pageout_flush() returns 0 as the run
992 		 * length.  Since the condition that caused truncated
993 		 * write may be permanent, e.g. exhausted free space,
994 		 * accepting n == 0 would cause an infinite loop.
995 		 *
996 		 * Forwarding the iterator leaves the unwritten page
997 		 * behind, but there is not much we can do there if
998 		 * filesystem refuses to write it.
999 		 */
1000 		if (n == 0) {
1001 			n = 1;
1002 			clearobjflags = FALSE;
1003 		}
1004 		np = vm_page_find_least(object, pi + n);
1005 	}
1006 #if 0
1007 	VOP_FSYNC(vp, (pagerflags & VM_PAGER_PUT_SYNC) ? MNT_WAIT : 0);
1008 #endif
1009 
1010 	if (clearobjflags)
1011 		vm_object_clear_flag(object, OBJ_MIGHTBEDIRTY);
1012 	return (res);
1013 }
1014 
1015 static int
1016 vm_object_page_collect_flush(vm_object_t object, vm_page_t p, int pagerflags,
1017     int flags, boolean_t *clearobjflags, boolean_t *eio)
1018 {
1019 	vm_page_t ma[vm_pageout_page_count], p_first, tp;
1020 	int count, i, mreq, runlen;
1021 
1022 	vm_page_lock_assert(p, MA_NOTOWNED);
1023 	VM_OBJECT_ASSERT_WLOCKED(object);
1024 
1025 	count = 1;
1026 	mreq = 0;
1027 
1028 	for (tp = p; count < vm_pageout_page_count; count++) {
1029 		tp = vm_page_next(tp);
1030 		if (tp == NULL || vm_page_busied(tp))
1031 			break;
1032 		if (!vm_object_page_remove_write(tp, flags, clearobjflags))
1033 			break;
1034 	}
1035 
1036 	for (p_first = p; count < vm_pageout_page_count; count++) {
1037 		tp = vm_page_prev(p_first);
1038 		if (tp == NULL || vm_page_busied(tp))
1039 			break;
1040 		if (!vm_object_page_remove_write(tp, flags, clearobjflags))
1041 			break;
1042 		p_first = tp;
1043 		mreq++;
1044 	}
1045 
1046 	for (tp = p_first, i = 0; i < count; tp = TAILQ_NEXT(tp, listq), i++)
1047 		ma[i] = tp;
1048 
1049 	vm_pageout_flush(ma, count, pagerflags, mreq, &runlen, eio);
1050 	return (runlen);
1051 }
1052 
1053 /*
1054  * Note that there is absolutely no sense in writing out
1055  * anonymous objects, so we track down the vnode object
1056  * to write out.
1057  * We invalidate (remove) all pages from the address space
1058  * for semantic correctness.
1059  *
1060  * If the backing object is a device object with unmanaged pages, then any
1061  * mappings to the specified range of pages must be removed before this
1062  * function is called.
1063  *
1064  * Note: certain anonymous maps, such as MAP_NOSYNC maps,
1065  * may start out with a NULL object.
1066  */
1067 boolean_t
1068 vm_object_sync(vm_object_t object, vm_ooffset_t offset, vm_size_t size,
1069     boolean_t syncio, boolean_t invalidate)
1070 {
1071 	vm_object_t backing_object;
1072 	struct vnode *vp;
1073 	struct mount *mp;
1074 	int error, flags, fsync_after;
1075 	boolean_t res;
1076 
1077 	if (object == NULL)
1078 		return (TRUE);
1079 	res = TRUE;
1080 	error = 0;
1081 	VM_OBJECT_WLOCK(object);
1082 	while ((backing_object = object->backing_object) != NULL) {
1083 		VM_OBJECT_WLOCK(backing_object);
1084 		offset += object->backing_object_offset;
1085 		VM_OBJECT_WUNLOCK(object);
1086 		object = backing_object;
1087 		if (object->size < OFF_TO_IDX(offset + size))
1088 			size = IDX_TO_OFF(object->size) - offset;
1089 	}
1090 	/*
1091 	 * Flush pages if writing is allowed, invalidate them
1092 	 * if invalidation requested.  Pages undergoing I/O
1093 	 * will be ignored by vm_object_page_remove().
1094 	 *
1095 	 * We cannot lock the vnode and then wait for paging
1096 	 * to complete without deadlocking against vm_fault.
1097 	 * Instead we simply call vm_object_page_remove() and
1098 	 * allow it to block internally on a page-by-page
1099 	 * basis when it encounters pages undergoing async
1100 	 * I/O.
1101 	 */
1102 	if (object->type == OBJT_VNODE &&
1103 	    (object->flags & OBJ_MIGHTBEDIRTY) != 0 &&
1104 	    ((vp = object->handle)->v_vflag & VV_NOSYNC) == 0) {
1105 		VM_OBJECT_WUNLOCK(object);
1106 		(void) vn_start_write(vp, &mp, V_WAIT);
1107 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1108 		if (syncio && !invalidate && offset == 0 &&
1109 		    atop(size) == object->size) {
1110 			/*
1111 			 * If syncing the whole mapping of the file,
1112 			 * it is faster to schedule all the writes in
1113 			 * async mode, also allowing the clustering,
1114 			 * and then wait for i/o to complete.
1115 			 */
1116 			flags = 0;
1117 			fsync_after = TRUE;
1118 		} else {
1119 			flags = (syncio || invalidate) ? OBJPC_SYNC : 0;
1120 			flags |= invalidate ? (OBJPC_SYNC | OBJPC_INVAL) : 0;
1121 			fsync_after = FALSE;
1122 		}
1123 		VM_OBJECT_WLOCK(object);
1124 		res = vm_object_page_clean(object, offset, offset + size,
1125 		    flags);
1126 		VM_OBJECT_WUNLOCK(object);
1127 		if (fsync_after)
1128 			error = VOP_FSYNC(vp, MNT_WAIT, curthread);
1129 		VOP_UNLOCK(vp, 0);
1130 		vn_finished_write(mp);
1131 		if (error != 0)
1132 			res = FALSE;
1133 		VM_OBJECT_WLOCK(object);
1134 	}
1135 	if ((object->type == OBJT_VNODE ||
1136 	     object->type == OBJT_DEVICE) && invalidate) {
1137 		if (object->type == OBJT_DEVICE)
1138 			/*
1139 			 * The option OBJPR_NOTMAPPED must be passed here
1140 			 * because vm_object_page_remove() cannot remove
1141 			 * unmanaged mappings.
1142 			 */
1143 			flags = OBJPR_NOTMAPPED;
1144 		else if (old_msync)
1145 			flags = 0;
1146 		else
1147 			flags = OBJPR_CLEANONLY;
1148 		vm_object_page_remove(object, OFF_TO_IDX(offset),
1149 		    OFF_TO_IDX(offset + size + PAGE_MASK), flags);
1150 	}
1151 	VM_OBJECT_WUNLOCK(object);
1152 	return (res);
1153 }
1154 
1155 /*
1156  * Determine whether the given advice can be applied to the object.  Advice is
1157  * not applied to unmanaged pages since they never belong to page queues, and
1158  * since MADV_FREE is destructive, it can apply only to anonymous pages that
1159  * have been mapped at most once.
1160  */
1161 static bool
1162 vm_object_advice_applies(vm_object_t object, int advice)
1163 {
1164 
1165 	if ((object->flags & OBJ_UNMANAGED) != 0)
1166 		return (false);
1167 	if (advice != MADV_FREE)
1168 		return (true);
1169 	return ((object->type == OBJT_DEFAULT || object->type == OBJT_SWAP) &&
1170 	    (object->flags & OBJ_ONEMAPPING) != 0);
1171 }
1172 
1173 static void
1174 vm_object_madvise_freespace(vm_object_t object, int advice, vm_pindex_t pindex,
1175     vm_size_t size)
1176 {
1177 
1178 	if (advice == MADV_FREE && object->type == OBJT_SWAP)
1179 		swap_pager_freespace(object, pindex, size);
1180 }
1181 
1182 /*
1183  *	vm_object_madvise:
1184  *
1185  *	Implements the madvise function at the object/page level.
1186  *
1187  *	MADV_WILLNEED	(any object)
1188  *
1189  *	    Activate the specified pages if they are resident.
1190  *
1191  *	MADV_DONTNEED	(any object)
1192  *
1193  *	    Deactivate the specified pages if they are resident.
1194  *
1195  *	MADV_FREE	(OBJT_DEFAULT/OBJT_SWAP objects,
1196  *			 OBJ_ONEMAPPING only)
1197  *
1198  *	    Deactivate and clean the specified pages if they are
1199  *	    resident.  This permits the process to reuse the pages
1200  *	    without faulting or the kernel to reclaim the pages
1201  *	    without I/O.
1202  */
1203 void
1204 vm_object_madvise(vm_object_t object, vm_pindex_t pindex, vm_pindex_t end,
1205     int advice)
1206 {
1207 	vm_pindex_t tpindex;
1208 	vm_object_t backing_object, tobject;
1209 	vm_page_t m, tm;
1210 
1211 	if (object == NULL)
1212 		return;
1213 
1214 relookup:
1215 	VM_OBJECT_WLOCK(object);
1216 	if (!vm_object_advice_applies(object, advice)) {
1217 		VM_OBJECT_WUNLOCK(object);
1218 		return;
1219 	}
1220 	for (m = vm_page_find_least(object, pindex); pindex < end; pindex++) {
1221 		tobject = object;
1222 
1223 		/*
1224 		 * If the next page isn't resident in the top-level object, we
1225 		 * need to search the shadow chain.  When applying MADV_FREE, we
1226 		 * take care to release any swap space used to store
1227 		 * non-resident pages.
1228 		 */
1229 		if (m == NULL || pindex < m->pindex) {
1230 			/*
1231 			 * Optimize a common case: if the top-level object has
1232 			 * no backing object, we can skip over the non-resident
1233 			 * range in constant time.
1234 			 */
1235 			if (object->backing_object == NULL) {
1236 				tpindex = (m != NULL && m->pindex < end) ?
1237 				    m->pindex : end;
1238 				vm_object_madvise_freespace(object, advice,
1239 				    pindex, tpindex - pindex);
1240 				if ((pindex = tpindex) == end)
1241 					break;
1242 				goto next_page;
1243 			}
1244 
1245 			tpindex = pindex;
1246 			do {
1247 				vm_object_madvise_freespace(tobject, advice,
1248 				    tpindex, 1);
1249 				/*
1250 				 * Prepare to search the next object in the
1251 				 * chain.
1252 				 */
1253 				backing_object = tobject->backing_object;
1254 				if (backing_object == NULL)
1255 					goto next_pindex;
1256 				VM_OBJECT_WLOCK(backing_object);
1257 				tpindex +=
1258 				    OFF_TO_IDX(tobject->backing_object_offset);
1259 				if (tobject != object)
1260 					VM_OBJECT_WUNLOCK(tobject);
1261 				tobject = backing_object;
1262 				if (!vm_object_advice_applies(tobject, advice))
1263 					goto next_pindex;
1264 			} while ((tm = vm_page_lookup(tobject, tpindex)) ==
1265 			    NULL);
1266 		} else {
1267 next_page:
1268 			tm = m;
1269 			m = TAILQ_NEXT(m, listq);
1270 		}
1271 
1272 		/*
1273 		 * If the page is not in a normal state, skip it.
1274 		 */
1275 		if (tm->valid != VM_PAGE_BITS_ALL)
1276 			goto next_pindex;
1277 		vm_page_lock(tm);
1278 		if (vm_page_held(tm)) {
1279 			vm_page_unlock(tm);
1280 			goto next_pindex;
1281 		}
1282 		KASSERT((tm->flags & PG_FICTITIOUS) == 0,
1283 		    ("vm_object_madvise: page %p is fictitious", tm));
1284 		KASSERT((tm->oflags & VPO_UNMANAGED) == 0,
1285 		    ("vm_object_madvise: page %p is not managed", tm));
1286 		if (vm_page_busied(tm)) {
1287 			if (object != tobject)
1288 				VM_OBJECT_WUNLOCK(tobject);
1289 			VM_OBJECT_WUNLOCK(object);
1290 			if (advice == MADV_WILLNEED) {
1291 				/*
1292 				 * Reference the page before unlocking and
1293 				 * sleeping so that the page daemon is less
1294 				 * likely to reclaim it.
1295 				 */
1296 				vm_page_aflag_set(tm, PGA_REFERENCED);
1297 			}
1298 			vm_page_busy_sleep(tm, "madvpo", false);
1299   			goto relookup;
1300 		}
1301 		vm_page_advise(tm, advice);
1302 		vm_page_unlock(tm);
1303 		vm_object_madvise_freespace(tobject, advice, tm->pindex, 1);
1304 next_pindex:
1305 		if (tobject != object)
1306 			VM_OBJECT_WUNLOCK(tobject);
1307 	}
1308 	VM_OBJECT_WUNLOCK(object);
1309 }
1310 
1311 /*
1312  *	vm_object_shadow:
1313  *
1314  *	Create a new object which is backed by the
1315  *	specified existing object range.  The source
1316  *	object reference is deallocated.
1317  *
1318  *	The new object and offset into that object
1319  *	are returned in the source parameters.
1320  */
1321 void
1322 vm_object_shadow(
1323 	vm_object_t *object,	/* IN/OUT */
1324 	vm_ooffset_t *offset,	/* IN/OUT */
1325 	vm_size_t length)
1326 {
1327 	vm_object_t source;
1328 	vm_object_t result;
1329 
1330 	source = *object;
1331 
1332 	/*
1333 	 * Don't create the new object if the old object isn't shared.
1334 	 */
1335 	if (source != NULL) {
1336 		VM_OBJECT_WLOCK(source);
1337 		if (source->ref_count == 1 &&
1338 		    source->handle == NULL &&
1339 		    (source->type == OBJT_DEFAULT ||
1340 		     source->type == OBJT_SWAP)) {
1341 			VM_OBJECT_WUNLOCK(source);
1342 			return;
1343 		}
1344 		VM_OBJECT_WUNLOCK(source);
1345 	}
1346 
1347 	/*
1348 	 * Allocate a new object with the given length.
1349 	 */
1350 	result = vm_object_allocate(OBJT_DEFAULT, atop(length));
1351 
1352 	/*
1353 	 * The new object shadows the source object, adding a reference to it.
1354 	 * Our caller changes his reference to point to the new object,
1355 	 * removing a reference to the source object.  Net result: no change
1356 	 * of reference count.
1357 	 *
1358 	 * Try to optimize the result object's page color when shadowing
1359 	 * in order to maintain page coloring consistency in the combined
1360 	 * shadowed object.
1361 	 */
1362 	result->backing_object = source;
1363 	/*
1364 	 * Store the offset into the source object, and fix up the offset into
1365 	 * the new object.
1366 	 */
1367 	result->backing_object_offset = *offset;
1368 	if (source != NULL) {
1369 		VM_OBJECT_WLOCK(source);
1370 		result->domain = source->domain;
1371 		LIST_INSERT_HEAD(&source->shadow_head, result, shadow_list);
1372 		source->shadow_count++;
1373 #if VM_NRESERVLEVEL > 0
1374 		result->flags |= source->flags & OBJ_COLORED;
1375 		result->pg_color = (source->pg_color + OFF_TO_IDX(*offset)) &
1376 		    ((1 << (VM_NFREEORDER - 1)) - 1);
1377 #endif
1378 		VM_OBJECT_WUNLOCK(source);
1379 	}
1380 
1381 
1382 	/*
1383 	 * Return the new things
1384 	 */
1385 	*offset = 0;
1386 	*object = result;
1387 }
1388 
1389 /*
1390  *	vm_object_split:
1391  *
1392  * Split the pages in a map entry into a new object.  This affords
1393  * easier removal of unused pages, and keeps object inheritance from
1394  * being a negative impact on memory usage.
1395  */
1396 void
1397 vm_object_split(vm_map_entry_t entry)
1398 {
1399 	vm_page_t m, m_next;
1400 	vm_object_t orig_object, new_object, source;
1401 	vm_pindex_t idx, offidxstart;
1402 	vm_size_t size;
1403 
1404 	orig_object = entry->object.vm_object;
1405 	if (orig_object->type != OBJT_DEFAULT && orig_object->type != OBJT_SWAP)
1406 		return;
1407 	if (orig_object->ref_count <= 1)
1408 		return;
1409 	VM_OBJECT_WUNLOCK(orig_object);
1410 
1411 	offidxstart = OFF_TO_IDX(entry->offset);
1412 	size = atop(entry->end - entry->start);
1413 
1414 	/*
1415 	 * If swap_pager_copy() is later called, it will convert new_object
1416 	 * into a swap object.
1417 	 */
1418 	new_object = vm_object_allocate(OBJT_DEFAULT, size);
1419 
1420 	/*
1421 	 * At this point, the new object is still private, so the order in
1422 	 * which the original and new objects are locked does not matter.
1423 	 */
1424 	VM_OBJECT_WLOCK(new_object);
1425 	VM_OBJECT_WLOCK(orig_object);
1426 	new_object->domain = orig_object->domain;
1427 	source = orig_object->backing_object;
1428 	if (source != NULL) {
1429 		VM_OBJECT_WLOCK(source);
1430 		if ((source->flags & OBJ_DEAD) != 0) {
1431 			VM_OBJECT_WUNLOCK(source);
1432 			VM_OBJECT_WUNLOCK(orig_object);
1433 			VM_OBJECT_WUNLOCK(new_object);
1434 			vm_object_deallocate(new_object);
1435 			VM_OBJECT_WLOCK(orig_object);
1436 			return;
1437 		}
1438 		LIST_INSERT_HEAD(&source->shadow_head,
1439 				  new_object, shadow_list);
1440 		source->shadow_count++;
1441 		vm_object_reference_locked(source);	/* for new_object */
1442 		vm_object_clear_flag(source, OBJ_ONEMAPPING);
1443 		VM_OBJECT_WUNLOCK(source);
1444 		new_object->backing_object_offset =
1445 			orig_object->backing_object_offset + entry->offset;
1446 		new_object->backing_object = source;
1447 	}
1448 	if (orig_object->cred != NULL) {
1449 		new_object->cred = orig_object->cred;
1450 		crhold(orig_object->cred);
1451 		new_object->charge = ptoa(size);
1452 		KASSERT(orig_object->charge >= ptoa(size),
1453 		    ("orig_object->charge < 0"));
1454 		orig_object->charge -= ptoa(size);
1455 	}
1456 retry:
1457 	m = vm_page_find_least(orig_object, offidxstart);
1458 	for (; m != NULL && (idx = m->pindex - offidxstart) < size;
1459 	    m = m_next) {
1460 		m_next = TAILQ_NEXT(m, listq);
1461 
1462 		/*
1463 		 * We must wait for pending I/O to complete before we can
1464 		 * rename the page.
1465 		 *
1466 		 * We do not have to VM_PROT_NONE the page as mappings should
1467 		 * not be changed by this operation.
1468 		 */
1469 		if (vm_page_busied(m)) {
1470 			VM_OBJECT_WUNLOCK(new_object);
1471 			vm_page_lock(m);
1472 			VM_OBJECT_WUNLOCK(orig_object);
1473 			vm_page_busy_sleep(m, "spltwt", false);
1474 			VM_OBJECT_WLOCK(orig_object);
1475 			VM_OBJECT_WLOCK(new_object);
1476 			goto retry;
1477 		}
1478 
1479 		/* vm_page_rename() will dirty the page. */
1480 		if (vm_page_rename(m, new_object, idx)) {
1481 			VM_OBJECT_WUNLOCK(new_object);
1482 			VM_OBJECT_WUNLOCK(orig_object);
1483 			vm_radix_wait();
1484 			VM_OBJECT_WLOCK(orig_object);
1485 			VM_OBJECT_WLOCK(new_object);
1486 			goto retry;
1487 		}
1488 #if VM_NRESERVLEVEL > 0
1489 		/*
1490 		 * If some of the reservation's allocated pages remain with
1491 		 * the original object, then transferring the reservation to
1492 		 * the new object is neither particularly beneficial nor
1493 		 * particularly harmful as compared to leaving the reservation
1494 		 * with the original object.  If, however, all of the
1495 		 * reservation's allocated pages are transferred to the new
1496 		 * object, then transferring the reservation is typically
1497 		 * beneficial.  Determining which of these two cases applies
1498 		 * would be more costly than unconditionally renaming the
1499 		 * reservation.
1500 		 */
1501 		vm_reserv_rename(m, new_object, orig_object, offidxstart);
1502 #endif
1503 		if (orig_object->type == OBJT_SWAP)
1504 			vm_page_xbusy(m);
1505 	}
1506 	if (orig_object->type == OBJT_SWAP) {
1507 		/*
1508 		 * swap_pager_copy() can sleep, in which case the orig_object's
1509 		 * and new_object's locks are released and reacquired.
1510 		 */
1511 		swap_pager_copy(orig_object, new_object, offidxstart, 0);
1512 		TAILQ_FOREACH(m, &new_object->memq, listq)
1513 			vm_page_xunbusy(m);
1514 	}
1515 	VM_OBJECT_WUNLOCK(orig_object);
1516 	VM_OBJECT_WUNLOCK(new_object);
1517 	entry->object.vm_object = new_object;
1518 	entry->offset = 0LL;
1519 	vm_object_deallocate(orig_object);
1520 	VM_OBJECT_WLOCK(new_object);
1521 }
1522 
1523 #define	OBSC_COLLAPSE_NOWAIT	0x0002
1524 #define	OBSC_COLLAPSE_WAIT	0x0004
1525 
1526 static vm_page_t
1527 vm_object_collapse_scan_wait(vm_object_t object, vm_page_t p, vm_page_t next,
1528     int op)
1529 {
1530 	vm_object_t backing_object;
1531 
1532 	VM_OBJECT_ASSERT_WLOCKED(object);
1533 	backing_object = object->backing_object;
1534 	VM_OBJECT_ASSERT_WLOCKED(backing_object);
1535 
1536 	KASSERT(p == NULL || vm_page_busied(p), ("unbusy page %p", p));
1537 	KASSERT(p == NULL || p->object == object || p->object == backing_object,
1538 	    ("invalid ownership %p %p %p", p, object, backing_object));
1539 	if ((op & OBSC_COLLAPSE_NOWAIT) != 0)
1540 		return (next);
1541 	if (p != NULL)
1542 		vm_page_lock(p);
1543 	VM_OBJECT_WUNLOCK(object);
1544 	VM_OBJECT_WUNLOCK(backing_object);
1545 	/* The page is only NULL when rename fails. */
1546 	if (p == NULL)
1547 		vm_radix_wait();
1548 	else
1549 		vm_page_busy_sleep(p, "vmocol", false);
1550 	VM_OBJECT_WLOCK(object);
1551 	VM_OBJECT_WLOCK(backing_object);
1552 	return (TAILQ_FIRST(&backing_object->memq));
1553 }
1554 
1555 static bool
1556 vm_object_scan_all_shadowed(vm_object_t object)
1557 {
1558 	vm_object_t backing_object;
1559 	vm_page_t p, pp;
1560 	vm_pindex_t backing_offset_index, new_pindex, pi, ps;
1561 
1562 	VM_OBJECT_ASSERT_WLOCKED(object);
1563 	VM_OBJECT_ASSERT_WLOCKED(object->backing_object);
1564 
1565 	backing_object = object->backing_object;
1566 
1567 	if (backing_object->type != OBJT_DEFAULT &&
1568 	    backing_object->type != OBJT_SWAP)
1569 		return (false);
1570 
1571 	pi = backing_offset_index = OFF_TO_IDX(object->backing_object_offset);
1572 	p = vm_page_find_least(backing_object, pi);
1573 	ps = swap_pager_find_least(backing_object, pi);
1574 
1575 	/*
1576 	 * Only check pages inside the parent object's range and
1577 	 * inside the parent object's mapping of the backing object.
1578 	 */
1579 	for (;; pi++) {
1580 		if (p != NULL && p->pindex < pi)
1581 			p = TAILQ_NEXT(p, listq);
1582 		if (ps < pi)
1583 			ps = swap_pager_find_least(backing_object, pi);
1584 		if (p == NULL && ps >= backing_object->size)
1585 			break;
1586 		else if (p == NULL)
1587 			pi = ps;
1588 		else
1589 			pi = MIN(p->pindex, ps);
1590 
1591 		new_pindex = pi - backing_offset_index;
1592 		if (new_pindex >= object->size)
1593 			break;
1594 
1595 		/*
1596 		 * See if the parent has the page or if the parent's object
1597 		 * pager has the page.  If the parent has the page but the page
1598 		 * is not valid, the parent's object pager must have the page.
1599 		 *
1600 		 * If this fails, the parent does not completely shadow the
1601 		 * object and we might as well give up now.
1602 		 */
1603 		pp = vm_page_lookup(object, new_pindex);
1604 		if ((pp == NULL || pp->valid == 0) &&
1605 		    !vm_pager_has_page(object, new_pindex, NULL, NULL))
1606 			return (false);
1607 	}
1608 	return (true);
1609 }
1610 
1611 static bool
1612 vm_object_collapse_scan(vm_object_t object, int op)
1613 {
1614 	vm_object_t backing_object;
1615 	vm_page_t next, p, pp;
1616 	vm_pindex_t backing_offset_index, new_pindex;
1617 
1618 	VM_OBJECT_ASSERT_WLOCKED(object);
1619 	VM_OBJECT_ASSERT_WLOCKED(object->backing_object);
1620 
1621 	backing_object = object->backing_object;
1622 	backing_offset_index = OFF_TO_IDX(object->backing_object_offset);
1623 
1624 	/*
1625 	 * Initial conditions
1626 	 */
1627 	if ((op & OBSC_COLLAPSE_WAIT) != 0)
1628 		vm_object_set_flag(backing_object, OBJ_DEAD);
1629 
1630 	/*
1631 	 * Our scan
1632 	 */
1633 	for (p = TAILQ_FIRST(&backing_object->memq); p != NULL; p = next) {
1634 		next = TAILQ_NEXT(p, listq);
1635 		new_pindex = p->pindex - backing_offset_index;
1636 
1637 		/*
1638 		 * Check for busy page
1639 		 */
1640 		if (vm_page_busied(p)) {
1641 			next = vm_object_collapse_scan_wait(object, p, next, op);
1642 			continue;
1643 		}
1644 
1645 		KASSERT(p->object == backing_object,
1646 		    ("vm_object_collapse_scan: object mismatch"));
1647 
1648 		if (p->pindex < backing_offset_index ||
1649 		    new_pindex >= object->size) {
1650 			if (backing_object->type == OBJT_SWAP)
1651 				swap_pager_freespace(backing_object, p->pindex,
1652 				    1);
1653 
1654 			/*
1655 			 * Page is out of the parent object's range, we can
1656 			 * simply destroy it.
1657 			 */
1658 			vm_page_lock(p);
1659 			KASSERT(!pmap_page_is_mapped(p),
1660 			    ("freeing mapped page %p", p));
1661 			if (p->wire_count == 0)
1662 				vm_page_free(p);
1663 			else
1664 				vm_page_remove(p);
1665 			vm_page_unlock(p);
1666 			continue;
1667 		}
1668 
1669 		pp = vm_page_lookup(object, new_pindex);
1670 		if (pp != NULL && vm_page_busied(pp)) {
1671 			/*
1672 			 * The page in the parent is busy and possibly not
1673 			 * (yet) valid.  Until its state is finalized by the
1674 			 * busy bit owner, we can't tell whether it shadows the
1675 			 * original page.  Therefore, we must either skip it
1676 			 * and the original (backing_object) page or wait for
1677 			 * its state to be finalized.
1678 			 *
1679 			 * This is due to a race with vm_fault() where we must
1680 			 * unbusy the original (backing_obj) page before we can
1681 			 * (re)lock the parent.  Hence we can get here.
1682 			 */
1683 			next = vm_object_collapse_scan_wait(object, pp, next,
1684 			    op);
1685 			continue;
1686 		}
1687 
1688 		KASSERT(pp == NULL || pp->valid != 0,
1689 		    ("unbusy invalid page %p", pp));
1690 
1691 		if (pp != NULL || vm_pager_has_page(object, new_pindex, NULL,
1692 			NULL)) {
1693 			/*
1694 			 * The page already exists in the parent OR swap exists
1695 			 * for this location in the parent.  Leave the parent's
1696 			 * page alone.  Destroy the original page from the
1697 			 * backing object.
1698 			 */
1699 			if (backing_object->type == OBJT_SWAP)
1700 				swap_pager_freespace(backing_object, p->pindex,
1701 				    1);
1702 			vm_page_lock(p);
1703 			KASSERT(!pmap_page_is_mapped(p),
1704 			    ("freeing mapped page %p", p));
1705 			if (p->wire_count == 0)
1706 				vm_page_free(p);
1707 			else
1708 				vm_page_remove(p);
1709 			vm_page_unlock(p);
1710 			continue;
1711 		}
1712 
1713 		/*
1714 		 * Page does not exist in parent, rename the page from the
1715 		 * backing object to the main object.
1716 		 *
1717 		 * If the page was mapped to a process, it can remain mapped
1718 		 * through the rename.  vm_page_rename() will dirty the page.
1719 		 */
1720 		if (vm_page_rename(p, object, new_pindex)) {
1721 			next = vm_object_collapse_scan_wait(object, NULL, next,
1722 			    op);
1723 			continue;
1724 		}
1725 
1726 		/* Use the old pindex to free the right page. */
1727 		if (backing_object->type == OBJT_SWAP)
1728 			swap_pager_freespace(backing_object,
1729 			    new_pindex + backing_offset_index, 1);
1730 
1731 #if VM_NRESERVLEVEL > 0
1732 		/*
1733 		 * Rename the reservation.
1734 		 */
1735 		vm_reserv_rename(p, object, backing_object,
1736 		    backing_offset_index);
1737 #endif
1738 	}
1739 	return (true);
1740 }
1741 
1742 
1743 /*
1744  * this version of collapse allows the operation to occur earlier and
1745  * when paging_in_progress is true for an object...  This is not a complete
1746  * operation, but should plug 99.9% of the rest of the leaks.
1747  */
1748 static void
1749 vm_object_qcollapse(vm_object_t object)
1750 {
1751 	vm_object_t backing_object = object->backing_object;
1752 
1753 	VM_OBJECT_ASSERT_WLOCKED(object);
1754 	VM_OBJECT_ASSERT_WLOCKED(backing_object);
1755 
1756 	if (backing_object->ref_count != 1)
1757 		return;
1758 
1759 	vm_object_collapse_scan(object, OBSC_COLLAPSE_NOWAIT);
1760 }
1761 
1762 /*
1763  *	vm_object_collapse:
1764  *
1765  *	Collapse an object with the object backing it.
1766  *	Pages in the backing object are moved into the
1767  *	parent, and the backing object is deallocated.
1768  */
1769 void
1770 vm_object_collapse(vm_object_t object)
1771 {
1772 	vm_object_t backing_object, new_backing_object;
1773 
1774 	VM_OBJECT_ASSERT_WLOCKED(object);
1775 
1776 	while (TRUE) {
1777 		/*
1778 		 * Verify that the conditions are right for collapse:
1779 		 *
1780 		 * The object exists and the backing object exists.
1781 		 */
1782 		if ((backing_object = object->backing_object) == NULL)
1783 			break;
1784 
1785 		/*
1786 		 * we check the backing object first, because it is most likely
1787 		 * not collapsable.
1788 		 */
1789 		VM_OBJECT_WLOCK(backing_object);
1790 		if (backing_object->handle != NULL ||
1791 		    (backing_object->type != OBJT_DEFAULT &&
1792 		     backing_object->type != OBJT_SWAP) ||
1793 		    (backing_object->flags & OBJ_DEAD) ||
1794 		    object->handle != NULL ||
1795 		    (object->type != OBJT_DEFAULT &&
1796 		     object->type != OBJT_SWAP) ||
1797 		    (object->flags & OBJ_DEAD)) {
1798 			VM_OBJECT_WUNLOCK(backing_object);
1799 			break;
1800 		}
1801 
1802 		if (object->paging_in_progress != 0 ||
1803 		    backing_object->paging_in_progress != 0) {
1804 			vm_object_qcollapse(object);
1805 			VM_OBJECT_WUNLOCK(backing_object);
1806 			break;
1807 		}
1808 
1809 		/*
1810 		 * We know that we can either collapse the backing object (if
1811 		 * the parent is the only reference to it) or (perhaps) have
1812 		 * the parent bypass the object if the parent happens to shadow
1813 		 * all the resident pages in the entire backing object.
1814 		 *
1815 		 * This is ignoring pager-backed pages such as swap pages.
1816 		 * vm_object_collapse_scan fails the shadowing test in this
1817 		 * case.
1818 		 */
1819 		if (backing_object->ref_count == 1) {
1820 			vm_object_pip_add(object, 1);
1821 			vm_object_pip_add(backing_object, 1);
1822 
1823 			/*
1824 			 * If there is exactly one reference to the backing
1825 			 * object, we can collapse it into the parent.
1826 			 */
1827 			vm_object_collapse_scan(object, OBSC_COLLAPSE_WAIT);
1828 
1829 #if VM_NRESERVLEVEL > 0
1830 			/*
1831 			 * Break any reservations from backing_object.
1832 			 */
1833 			if (__predict_false(!LIST_EMPTY(&backing_object->rvq)))
1834 				vm_reserv_break_all(backing_object);
1835 #endif
1836 
1837 			/*
1838 			 * Move the pager from backing_object to object.
1839 			 */
1840 			if (backing_object->type == OBJT_SWAP) {
1841 				/*
1842 				 * swap_pager_copy() can sleep, in which case
1843 				 * the backing_object's and object's locks are
1844 				 * released and reacquired.
1845 				 * Since swap_pager_copy() is being asked to
1846 				 * destroy the source, it will change the
1847 				 * backing_object's type to OBJT_DEFAULT.
1848 				 */
1849 				swap_pager_copy(
1850 				    backing_object,
1851 				    object,
1852 				    OFF_TO_IDX(object->backing_object_offset), TRUE);
1853 			}
1854 			/*
1855 			 * Object now shadows whatever backing_object did.
1856 			 * Note that the reference to
1857 			 * backing_object->backing_object moves from within
1858 			 * backing_object to within object.
1859 			 */
1860 			LIST_REMOVE(object, shadow_list);
1861 			backing_object->shadow_count--;
1862 			if (backing_object->backing_object) {
1863 				VM_OBJECT_WLOCK(backing_object->backing_object);
1864 				LIST_REMOVE(backing_object, shadow_list);
1865 				LIST_INSERT_HEAD(
1866 				    &backing_object->backing_object->shadow_head,
1867 				    object, shadow_list);
1868 				/*
1869 				 * The shadow_count has not changed.
1870 				 */
1871 				VM_OBJECT_WUNLOCK(backing_object->backing_object);
1872 			}
1873 			object->backing_object = backing_object->backing_object;
1874 			object->backing_object_offset +=
1875 			    backing_object->backing_object_offset;
1876 
1877 			/*
1878 			 * Discard backing_object.
1879 			 *
1880 			 * Since the backing object has no pages, no pager left,
1881 			 * and no object references within it, all that is
1882 			 * necessary is to dispose of it.
1883 			 */
1884 			KASSERT(backing_object->ref_count == 1, (
1885 "backing_object %p was somehow re-referenced during collapse!",
1886 			    backing_object));
1887 			vm_object_pip_wakeup(backing_object);
1888 			backing_object->type = OBJT_DEAD;
1889 			backing_object->ref_count = 0;
1890 			VM_OBJECT_WUNLOCK(backing_object);
1891 			vm_object_destroy(backing_object);
1892 
1893 			vm_object_pip_wakeup(object);
1894 			counter_u64_add(object_collapses, 1);
1895 		} else {
1896 			/*
1897 			 * If we do not entirely shadow the backing object,
1898 			 * there is nothing we can do so we give up.
1899 			 */
1900 			if (object->resident_page_count != object->size &&
1901 			    !vm_object_scan_all_shadowed(object)) {
1902 				VM_OBJECT_WUNLOCK(backing_object);
1903 				break;
1904 			}
1905 
1906 			/*
1907 			 * Make the parent shadow the next object in the
1908 			 * chain.  Deallocating backing_object will not remove
1909 			 * it, since its reference count is at least 2.
1910 			 */
1911 			LIST_REMOVE(object, shadow_list);
1912 			backing_object->shadow_count--;
1913 
1914 			new_backing_object = backing_object->backing_object;
1915 			if ((object->backing_object = new_backing_object) != NULL) {
1916 				VM_OBJECT_WLOCK(new_backing_object);
1917 				LIST_INSERT_HEAD(
1918 				    &new_backing_object->shadow_head,
1919 				    object,
1920 				    shadow_list
1921 				);
1922 				new_backing_object->shadow_count++;
1923 				vm_object_reference_locked(new_backing_object);
1924 				VM_OBJECT_WUNLOCK(new_backing_object);
1925 				object->backing_object_offset +=
1926 					backing_object->backing_object_offset;
1927 			}
1928 
1929 			/*
1930 			 * Drop the reference count on backing_object. Since
1931 			 * its ref_count was at least 2, it will not vanish.
1932 			 */
1933 			backing_object->ref_count--;
1934 			VM_OBJECT_WUNLOCK(backing_object);
1935 			counter_u64_add(object_bypasses, 1);
1936 		}
1937 
1938 		/*
1939 		 * Try again with this object's new backing object.
1940 		 */
1941 	}
1942 }
1943 
1944 /*
1945  *	vm_object_page_remove:
1946  *
1947  *	For the given object, either frees or invalidates each of the
1948  *	specified pages.  In general, a page is freed.  However, if a page is
1949  *	wired for any reason other than the existence of a managed, wired
1950  *	mapping, then it may be invalidated but not removed from the object.
1951  *	Pages are specified by the given range ["start", "end") and the option
1952  *	OBJPR_CLEANONLY.  As a special case, if "end" is zero, then the range
1953  *	extends from "start" to the end of the object.  If the option
1954  *	OBJPR_CLEANONLY is specified, then only the non-dirty pages within the
1955  *	specified range are affected.  If the option OBJPR_NOTMAPPED is
1956  *	specified, then the pages within the specified range must have no
1957  *	mappings.  Otherwise, if this option is not specified, any mappings to
1958  *	the specified pages are removed before the pages are freed or
1959  *	invalidated.
1960  *
1961  *	In general, this operation should only be performed on objects that
1962  *	contain managed pages.  There are, however, two exceptions.  First, it
1963  *	is performed on the kernel and kmem objects by vm_map_entry_delete().
1964  *	Second, it is used by msync(..., MS_INVALIDATE) to invalidate device-
1965  *	backed pages.  In both of these cases, the option OBJPR_CLEANONLY must
1966  *	not be specified and the option OBJPR_NOTMAPPED must be specified.
1967  *
1968  *	The object must be locked.
1969  */
1970 void
1971 vm_object_page_remove(vm_object_t object, vm_pindex_t start, vm_pindex_t end,
1972     int options)
1973 {
1974 	vm_page_t p, next;
1975 	struct mtx *mtx;
1976 	struct pglist pgl;
1977 
1978 	VM_OBJECT_ASSERT_WLOCKED(object);
1979 	KASSERT((object->flags & OBJ_UNMANAGED) == 0 ||
1980 	    (options & (OBJPR_CLEANONLY | OBJPR_NOTMAPPED)) == OBJPR_NOTMAPPED,
1981 	    ("vm_object_page_remove: illegal options for object %p", object));
1982 	if (object->resident_page_count == 0)
1983 		return;
1984 	vm_object_pip_add(object, 1);
1985 	TAILQ_INIT(&pgl);
1986 again:
1987 	p = vm_page_find_least(object, start);
1988 	mtx = NULL;
1989 
1990 	/*
1991 	 * Here, the variable "p" is either (1) the page with the least pindex
1992 	 * greater than or equal to the parameter "start" or (2) NULL.
1993 	 */
1994 	for (; p != NULL && (p->pindex < end || end == 0); p = next) {
1995 		next = TAILQ_NEXT(p, listq);
1996 
1997 		/*
1998 		 * If the page is wired for any reason besides the existence
1999 		 * of managed, wired mappings, then it cannot be freed.  For
2000 		 * example, fictitious pages, which represent device memory,
2001 		 * are inherently wired and cannot be freed.  They can,
2002 		 * however, be invalidated if the option OBJPR_CLEANONLY is
2003 		 * not specified.
2004 		 */
2005 		vm_page_change_lock(p, &mtx);
2006 		if (vm_page_xbusied(p)) {
2007 			VM_OBJECT_WUNLOCK(object);
2008 			vm_page_busy_sleep(p, "vmopax", true);
2009 			VM_OBJECT_WLOCK(object);
2010 			goto again;
2011 		}
2012 		if (p->wire_count != 0) {
2013 			if ((options & OBJPR_NOTMAPPED) == 0 &&
2014 			    object->ref_count != 0)
2015 				pmap_remove_all(p);
2016 			if ((options & OBJPR_CLEANONLY) == 0) {
2017 				p->valid = 0;
2018 				vm_page_undirty(p);
2019 			}
2020 			continue;
2021 		}
2022 		if (vm_page_busied(p)) {
2023 			VM_OBJECT_WUNLOCK(object);
2024 			vm_page_busy_sleep(p, "vmopar", false);
2025 			VM_OBJECT_WLOCK(object);
2026 			goto again;
2027 		}
2028 		KASSERT((p->flags & PG_FICTITIOUS) == 0,
2029 		    ("vm_object_page_remove: page %p is fictitious", p));
2030 		if ((options & OBJPR_CLEANONLY) != 0 && p->valid != 0) {
2031 			if ((options & OBJPR_NOTMAPPED) == 0 &&
2032 			    object->ref_count != 0)
2033 				pmap_remove_write(p);
2034 			if (p->dirty != 0)
2035 				continue;
2036 		}
2037 		if ((options & OBJPR_NOTMAPPED) == 0 && object->ref_count != 0)
2038 			pmap_remove_all(p);
2039 		p->flags &= ~PG_ZERO;
2040 		if (vm_page_free_prep(p, false))
2041 			TAILQ_INSERT_TAIL(&pgl, p, listq);
2042 	}
2043 	if (mtx != NULL)
2044 		mtx_unlock(mtx);
2045 	vm_page_free_phys_pglist(&pgl);
2046 	vm_object_pip_wakeup(object);
2047 }
2048 
2049 /*
2050  *	vm_object_page_noreuse:
2051  *
2052  *	For the given object, attempt to move the specified pages to
2053  *	the head of the inactive queue.  This bypasses regular LRU
2054  *	operation and allows the pages to be reused quickly under memory
2055  *	pressure.  If a page is wired for any reason, then it will not
2056  *	be queued.  Pages are specified by the range ["start", "end").
2057  *	As a special case, if "end" is zero, then the range extends from
2058  *	"start" to the end of the object.
2059  *
2060  *	This operation should only be performed on objects that
2061  *	contain non-fictitious, managed pages.
2062  *
2063  *	The object must be locked.
2064  */
2065 void
2066 vm_object_page_noreuse(vm_object_t object, vm_pindex_t start, vm_pindex_t end)
2067 {
2068 	struct mtx *mtx;
2069 	vm_page_t p, next;
2070 
2071 	VM_OBJECT_ASSERT_LOCKED(object);
2072 	KASSERT((object->flags & (OBJ_FICTITIOUS | OBJ_UNMANAGED)) == 0,
2073 	    ("vm_object_page_noreuse: illegal object %p", object));
2074 	if (object->resident_page_count == 0)
2075 		return;
2076 	p = vm_page_find_least(object, start);
2077 
2078 	/*
2079 	 * Here, the variable "p" is either (1) the page with the least pindex
2080 	 * greater than or equal to the parameter "start" or (2) NULL.
2081 	 */
2082 	mtx = NULL;
2083 	for (; p != NULL && (p->pindex < end || end == 0); p = next) {
2084 		next = TAILQ_NEXT(p, listq);
2085 		vm_page_change_lock(p, &mtx);
2086 		vm_page_deactivate_noreuse(p);
2087 	}
2088 	if (mtx != NULL)
2089 		mtx_unlock(mtx);
2090 }
2091 
2092 /*
2093  *	Populate the specified range of the object with valid pages.  Returns
2094  *	TRUE if the range is successfully populated and FALSE otherwise.
2095  *
2096  *	Note: This function should be optimized to pass a larger array of
2097  *	pages to vm_pager_get_pages() before it is applied to a non-
2098  *	OBJT_DEVICE object.
2099  *
2100  *	The object must be locked.
2101  */
2102 boolean_t
2103 vm_object_populate(vm_object_t object, vm_pindex_t start, vm_pindex_t end)
2104 {
2105 	vm_page_t m;
2106 	vm_pindex_t pindex;
2107 	int rv;
2108 
2109 	VM_OBJECT_ASSERT_WLOCKED(object);
2110 	for (pindex = start; pindex < end; pindex++) {
2111 		m = vm_page_grab(object, pindex, VM_ALLOC_NORMAL);
2112 		if (m->valid != VM_PAGE_BITS_ALL) {
2113 			rv = vm_pager_get_pages(object, &m, 1, NULL, NULL);
2114 			if (rv != VM_PAGER_OK) {
2115 				vm_page_lock(m);
2116 				vm_page_free(m);
2117 				vm_page_unlock(m);
2118 				break;
2119 			}
2120 		}
2121 		/*
2122 		 * Keep "m" busy because a subsequent iteration may unlock
2123 		 * the object.
2124 		 */
2125 	}
2126 	if (pindex > start) {
2127 		m = vm_page_lookup(object, start);
2128 		while (m != NULL && m->pindex < pindex) {
2129 			vm_page_xunbusy(m);
2130 			m = TAILQ_NEXT(m, listq);
2131 		}
2132 	}
2133 	return (pindex == end);
2134 }
2135 
2136 /*
2137  *	Routine:	vm_object_coalesce
2138  *	Function:	Coalesces two objects backing up adjoining
2139  *			regions of memory into a single object.
2140  *
2141  *	returns TRUE if objects were combined.
2142  *
2143  *	NOTE:	Only works at the moment if the second object is NULL -
2144  *		if it's not, which object do we lock first?
2145  *
2146  *	Parameters:
2147  *		prev_object	First object to coalesce
2148  *		prev_offset	Offset into prev_object
2149  *		prev_size	Size of reference to prev_object
2150  *		next_size	Size of reference to the second object
2151  *		reserved	Indicator that extension region has
2152  *				swap accounted for
2153  *
2154  *	Conditions:
2155  *	The object must *not* be locked.
2156  */
2157 boolean_t
2158 vm_object_coalesce(vm_object_t prev_object, vm_ooffset_t prev_offset,
2159     vm_size_t prev_size, vm_size_t next_size, boolean_t reserved)
2160 {
2161 	vm_pindex_t next_pindex;
2162 
2163 	if (prev_object == NULL)
2164 		return (TRUE);
2165 	VM_OBJECT_WLOCK(prev_object);
2166 	if ((prev_object->type != OBJT_DEFAULT &&
2167 	    prev_object->type != OBJT_SWAP) ||
2168 	    (prev_object->flags & OBJ_TMPFS_NODE) != 0) {
2169 		VM_OBJECT_WUNLOCK(prev_object);
2170 		return (FALSE);
2171 	}
2172 
2173 	/*
2174 	 * Try to collapse the object first
2175 	 */
2176 	vm_object_collapse(prev_object);
2177 
2178 	/*
2179 	 * Can't coalesce if: . more than one reference . paged out . shadows
2180 	 * another object . has a copy elsewhere (any of which mean that the
2181 	 * pages not mapped to prev_entry may be in use anyway)
2182 	 */
2183 	if (prev_object->backing_object != NULL) {
2184 		VM_OBJECT_WUNLOCK(prev_object);
2185 		return (FALSE);
2186 	}
2187 
2188 	prev_size >>= PAGE_SHIFT;
2189 	next_size >>= PAGE_SHIFT;
2190 	next_pindex = OFF_TO_IDX(prev_offset) + prev_size;
2191 
2192 	if ((prev_object->ref_count > 1) &&
2193 	    (prev_object->size != next_pindex)) {
2194 		VM_OBJECT_WUNLOCK(prev_object);
2195 		return (FALSE);
2196 	}
2197 
2198 	/*
2199 	 * Account for the charge.
2200 	 */
2201 	if (prev_object->cred != NULL) {
2202 
2203 		/*
2204 		 * If prev_object was charged, then this mapping,
2205 		 * although not charged now, may become writable
2206 		 * later. Non-NULL cred in the object would prevent
2207 		 * swap reservation during enabling of the write
2208 		 * access, so reserve swap now. Failed reservation
2209 		 * cause allocation of the separate object for the map
2210 		 * entry, and swap reservation for this entry is
2211 		 * managed in appropriate time.
2212 		 */
2213 		if (!reserved && !swap_reserve_by_cred(ptoa(next_size),
2214 		    prev_object->cred)) {
2215 			VM_OBJECT_WUNLOCK(prev_object);
2216 			return (FALSE);
2217 		}
2218 		prev_object->charge += ptoa(next_size);
2219 	}
2220 
2221 	/*
2222 	 * Remove any pages that may still be in the object from a previous
2223 	 * deallocation.
2224 	 */
2225 	if (next_pindex < prev_object->size) {
2226 		vm_object_page_remove(prev_object, next_pindex, next_pindex +
2227 		    next_size, 0);
2228 		if (prev_object->type == OBJT_SWAP)
2229 			swap_pager_freespace(prev_object,
2230 					     next_pindex, next_size);
2231 #if 0
2232 		if (prev_object->cred != NULL) {
2233 			KASSERT(prev_object->charge >=
2234 			    ptoa(prev_object->size - next_pindex),
2235 			    ("object %p overcharged 1 %jx %jx", prev_object,
2236 				(uintmax_t)next_pindex, (uintmax_t)next_size));
2237 			prev_object->charge -= ptoa(prev_object->size -
2238 			    next_pindex);
2239 		}
2240 #endif
2241 	}
2242 
2243 	/*
2244 	 * Extend the object if necessary.
2245 	 */
2246 	if (next_pindex + next_size > prev_object->size)
2247 		prev_object->size = next_pindex + next_size;
2248 
2249 	VM_OBJECT_WUNLOCK(prev_object);
2250 	return (TRUE);
2251 }
2252 
2253 void
2254 vm_object_set_writeable_dirty(vm_object_t object)
2255 {
2256 
2257 	VM_OBJECT_ASSERT_WLOCKED(object);
2258 	if (object->type != OBJT_VNODE) {
2259 		if ((object->flags & OBJ_TMPFS_NODE) != 0) {
2260 			KASSERT(object->type == OBJT_SWAP, ("non-swap tmpfs"));
2261 			vm_object_set_flag(object, OBJ_TMPFS_DIRTY);
2262 		}
2263 		return;
2264 	}
2265 	object->generation++;
2266 	if ((object->flags & OBJ_MIGHTBEDIRTY) != 0)
2267 		return;
2268 	vm_object_set_flag(object, OBJ_MIGHTBEDIRTY);
2269 }
2270 
2271 /*
2272  *	vm_object_unwire:
2273  *
2274  *	For each page offset within the specified range of the given object,
2275  *	find the highest-level page in the shadow chain and unwire it.  A page
2276  *	must exist at every page offset, and the highest-level page must be
2277  *	wired.
2278  */
2279 void
2280 vm_object_unwire(vm_object_t object, vm_ooffset_t offset, vm_size_t length,
2281     uint8_t queue)
2282 {
2283 	vm_object_t tobject, t1object;
2284 	vm_page_t m, tm;
2285 	vm_pindex_t end_pindex, pindex, tpindex;
2286 	int depth, locked_depth;
2287 
2288 	KASSERT((offset & PAGE_MASK) == 0,
2289 	    ("vm_object_unwire: offset is not page aligned"));
2290 	KASSERT((length & PAGE_MASK) == 0,
2291 	    ("vm_object_unwire: length is not a multiple of PAGE_SIZE"));
2292 	/* The wired count of a fictitious page never changes. */
2293 	if ((object->flags & OBJ_FICTITIOUS) != 0)
2294 		return;
2295 	pindex = OFF_TO_IDX(offset);
2296 	end_pindex = pindex + atop(length);
2297 again:
2298 	locked_depth = 1;
2299 	VM_OBJECT_RLOCK(object);
2300 	m = vm_page_find_least(object, pindex);
2301 	while (pindex < end_pindex) {
2302 		if (m == NULL || pindex < m->pindex) {
2303 			/*
2304 			 * The first object in the shadow chain doesn't
2305 			 * contain a page at the current index.  Therefore,
2306 			 * the page must exist in a backing object.
2307 			 */
2308 			tobject = object;
2309 			tpindex = pindex;
2310 			depth = 0;
2311 			do {
2312 				tpindex +=
2313 				    OFF_TO_IDX(tobject->backing_object_offset);
2314 				tobject = tobject->backing_object;
2315 				KASSERT(tobject != NULL,
2316 				    ("vm_object_unwire: missing page"));
2317 				if ((tobject->flags & OBJ_FICTITIOUS) != 0)
2318 					goto next_page;
2319 				depth++;
2320 				if (depth == locked_depth) {
2321 					locked_depth++;
2322 					VM_OBJECT_RLOCK(tobject);
2323 				}
2324 			} while ((tm = vm_page_lookup(tobject, tpindex)) ==
2325 			    NULL);
2326 		} else {
2327 			tm = m;
2328 			m = TAILQ_NEXT(m, listq);
2329 		}
2330 		vm_page_lock(tm);
2331 		if (vm_page_xbusied(tm)) {
2332 			for (tobject = object; locked_depth >= 1;
2333 			    locked_depth--) {
2334 				t1object = tobject->backing_object;
2335 				VM_OBJECT_RUNLOCK(tobject);
2336 				tobject = t1object;
2337 			}
2338 			vm_page_busy_sleep(tm, "unwbo", true);
2339 			goto again;
2340 		}
2341 		vm_page_unwire(tm, queue);
2342 		vm_page_unlock(tm);
2343 next_page:
2344 		pindex++;
2345 	}
2346 	/* Release the accumulated object locks. */
2347 	for (tobject = object; locked_depth >= 1; locked_depth--) {
2348 		t1object = tobject->backing_object;
2349 		VM_OBJECT_RUNLOCK(tobject);
2350 		tobject = t1object;
2351 	}
2352 }
2353 
2354 struct vnode *
2355 vm_object_vnode(vm_object_t object)
2356 {
2357 
2358 	VM_OBJECT_ASSERT_LOCKED(object);
2359 	if (object->type == OBJT_VNODE)
2360 		return (object->handle);
2361 	if (object->type == OBJT_SWAP && (object->flags & OBJ_TMPFS) != 0)
2362 		return (object->un_pager.swp.swp_tmpfs);
2363 	return (NULL);
2364 }
2365 
2366 static int
2367 sysctl_vm_object_list(SYSCTL_HANDLER_ARGS)
2368 {
2369 	struct kinfo_vmobject *kvo;
2370 	char *fullpath, *freepath;
2371 	struct vnode *vp;
2372 	struct vattr va;
2373 	vm_object_t obj;
2374 	vm_page_t m;
2375 	int count, error;
2376 
2377 	if (req->oldptr == NULL) {
2378 		/*
2379 		 * If an old buffer has not been provided, generate an
2380 		 * estimate of the space needed for a subsequent call.
2381 		 */
2382 		mtx_lock(&vm_object_list_mtx);
2383 		count = 0;
2384 		TAILQ_FOREACH(obj, &vm_object_list, object_list) {
2385 			if (obj->type == OBJT_DEAD)
2386 				continue;
2387 			count++;
2388 		}
2389 		mtx_unlock(&vm_object_list_mtx);
2390 		return (SYSCTL_OUT(req, NULL, sizeof(struct kinfo_vmobject) *
2391 		    count * 11 / 10));
2392 	}
2393 
2394 	kvo = malloc(sizeof(*kvo), M_TEMP, M_WAITOK);
2395 	error = 0;
2396 
2397 	/*
2398 	 * VM objects are type stable and are never removed from the
2399 	 * list once added.  This allows us to safely read obj->object_list
2400 	 * after reacquiring the VM object lock.
2401 	 */
2402 	mtx_lock(&vm_object_list_mtx);
2403 	TAILQ_FOREACH(obj, &vm_object_list, object_list) {
2404 		if (obj->type == OBJT_DEAD)
2405 			continue;
2406 		VM_OBJECT_RLOCK(obj);
2407 		if (obj->type == OBJT_DEAD) {
2408 			VM_OBJECT_RUNLOCK(obj);
2409 			continue;
2410 		}
2411 		mtx_unlock(&vm_object_list_mtx);
2412 		kvo->kvo_size = ptoa(obj->size);
2413 		kvo->kvo_resident = obj->resident_page_count;
2414 		kvo->kvo_ref_count = obj->ref_count;
2415 		kvo->kvo_shadow_count = obj->shadow_count;
2416 		kvo->kvo_memattr = obj->memattr;
2417 		kvo->kvo_active = 0;
2418 		kvo->kvo_inactive = 0;
2419 		TAILQ_FOREACH(m, &obj->memq, listq) {
2420 			/*
2421 			 * A page may belong to the object but be
2422 			 * dequeued and set to PQ_NONE while the
2423 			 * object lock is not held.  This makes the
2424 			 * reads of m->queue below racy, and we do not
2425 			 * count pages set to PQ_NONE.  However, this
2426 			 * sysctl is only meant to give an
2427 			 * approximation of the system anyway.
2428 			 */
2429 			if (vm_page_active(m))
2430 				kvo->kvo_active++;
2431 			else if (vm_page_inactive(m))
2432 				kvo->kvo_inactive++;
2433 		}
2434 
2435 		kvo->kvo_vn_fileid = 0;
2436 		kvo->kvo_vn_fsid = 0;
2437 		kvo->kvo_vn_fsid_freebsd11 = 0;
2438 		freepath = NULL;
2439 		fullpath = "";
2440 		vp = NULL;
2441 		switch (obj->type) {
2442 		case OBJT_DEFAULT:
2443 			kvo->kvo_type = KVME_TYPE_DEFAULT;
2444 			break;
2445 		case OBJT_VNODE:
2446 			kvo->kvo_type = KVME_TYPE_VNODE;
2447 			vp = obj->handle;
2448 			vref(vp);
2449 			break;
2450 		case OBJT_SWAP:
2451 			kvo->kvo_type = KVME_TYPE_SWAP;
2452 			break;
2453 		case OBJT_DEVICE:
2454 			kvo->kvo_type = KVME_TYPE_DEVICE;
2455 			break;
2456 		case OBJT_PHYS:
2457 			kvo->kvo_type = KVME_TYPE_PHYS;
2458 			break;
2459 		case OBJT_DEAD:
2460 			kvo->kvo_type = KVME_TYPE_DEAD;
2461 			break;
2462 		case OBJT_SG:
2463 			kvo->kvo_type = KVME_TYPE_SG;
2464 			break;
2465 		case OBJT_MGTDEVICE:
2466 			kvo->kvo_type = KVME_TYPE_MGTDEVICE;
2467 			break;
2468 		default:
2469 			kvo->kvo_type = KVME_TYPE_UNKNOWN;
2470 			break;
2471 		}
2472 		VM_OBJECT_RUNLOCK(obj);
2473 		if (vp != NULL) {
2474 			vn_fullpath(curthread, vp, &fullpath, &freepath);
2475 			vn_lock(vp, LK_SHARED | LK_RETRY);
2476 			if (VOP_GETATTR(vp, &va, curthread->td_ucred) == 0) {
2477 				kvo->kvo_vn_fileid = va.va_fileid;
2478 				kvo->kvo_vn_fsid = va.va_fsid;
2479 				kvo->kvo_vn_fsid_freebsd11 = va.va_fsid;
2480 								/* truncate */
2481 			}
2482 			vput(vp);
2483 		}
2484 
2485 		strlcpy(kvo->kvo_path, fullpath, sizeof(kvo->kvo_path));
2486 		if (freepath != NULL)
2487 			free(freepath, M_TEMP);
2488 
2489 		/* Pack record size down */
2490 		kvo->kvo_structsize = offsetof(struct kinfo_vmobject, kvo_path)
2491 		    + strlen(kvo->kvo_path) + 1;
2492 		kvo->kvo_structsize = roundup(kvo->kvo_structsize,
2493 		    sizeof(uint64_t));
2494 		error = SYSCTL_OUT(req, kvo, kvo->kvo_structsize);
2495 		mtx_lock(&vm_object_list_mtx);
2496 		if (error)
2497 			break;
2498 	}
2499 	mtx_unlock(&vm_object_list_mtx);
2500 	free(kvo, M_TEMP);
2501 	return (error);
2502 }
2503 SYSCTL_PROC(_vm, OID_AUTO, objects, CTLTYPE_STRUCT | CTLFLAG_RW | CTLFLAG_SKIP |
2504     CTLFLAG_MPSAFE, NULL, 0, sysctl_vm_object_list, "S,kinfo_vmobject",
2505     "List of VM objects");
2506 
2507 #include "opt_ddb.h"
2508 #ifdef DDB
2509 #include <sys/kernel.h>
2510 
2511 #include <sys/cons.h>
2512 
2513 #include <ddb/ddb.h>
2514 
2515 static int
2516 _vm_object_in_map(vm_map_t map, vm_object_t object, vm_map_entry_t entry)
2517 {
2518 	vm_map_t tmpm;
2519 	vm_map_entry_t tmpe;
2520 	vm_object_t obj;
2521 	int entcount;
2522 
2523 	if (map == 0)
2524 		return 0;
2525 
2526 	if (entry == 0) {
2527 		tmpe = map->header.next;
2528 		entcount = map->nentries;
2529 		while (entcount-- && (tmpe != &map->header)) {
2530 			if (_vm_object_in_map(map, object, tmpe)) {
2531 				return 1;
2532 			}
2533 			tmpe = tmpe->next;
2534 		}
2535 	} else if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
2536 		tmpm = entry->object.sub_map;
2537 		tmpe = tmpm->header.next;
2538 		entcount = tmpm->nentries;
2539 		while (entcount-- && tmpe != &tmpm->header) {
2540 			if (_vm_object_in_map(tmpm, object, tmpe)) {
2541 				return 1;
2542 			}
2543 			tmpe = tmpe->next;
2544 		}
2545 	} else if ((obj = entry->object.vm_object) != NULL) {
2546 		for (; obj; obj = obj->backing_object)
2547 			if (obj == object) {
2548 				return 1;
2549 			}
2550 	}
2551 	return 0;
2552 }
2553 
2554 static int
2555 vm_object_in_map(vm_object_t object)
2556 {
2557 	struct proc *p;
2558 
2559 	/* sx_slock(&allproc_lock); */
2560 	FOREACH_PROC_IN_SYSTEM(p) {
2561 		if (!p->p_vmspace /* || (p->p_flag & (P_SYSTEM|P_WEXIT)) */)
2562 			continue;
2563 		if (_vm_object_in_map(&p->p_vmspace->vm_map, object, 0)) {
2564 			/* sx_sunlock(&allproc_lock); */
2565 			return 1;
2566 		}
2567 	}
2568 	/* sx_sunlock(&allproc_lock); */
2569 	if (_vm_object_in_map(kernel_map, object, 0))
2570 		return 1;
2571 	return 0;
2572 }
2573 
2574 DB_SHOW_COMMAND(vmochk, vm_object_check)
2575 {
2576 	vm_object_t object;
2577 
2578 	/*
2579 	 * make sure that internal objs are in a map somewhere
2580 	 * and none have zero ref counts.
2581 	 */
2582 	TAILQ_FOREACH(object, &vm_object_list, object_list) {
2583 		if (object->handle == NULL &&
2584 		    (object->type == OBJT_DEFAULT || object->type == OBJT_SWAP)) {
2585 			if (object->ref_count == 0) {
2586 				db_printf("vmochk: internal obj has zero ref count: %ld\n",
2587 					(long)object->size);
2588 			}
2589 			if (!vm_object_in_map(object)) {
2590 				db_printf(
2591 			"vmochk: internal obj is not in a map: "
2592 			"ref: %d, size: %lu: 0x%lx, backing_object: %p\n",
2593 				    object->ref_count, (u_long)object->size,
2594 				    (u_long)object->size,
2595 				    (void *)object->backing_object);
2596 			}
2597 		}
2598 	}
2599 }
2600 
2601 /*
2602  *	vm_object_print:	[ debug ]
2603  */
2604 DB_SHOW_COMMAND(object, vm_object_print_static)
2605 {
2606 	/* XXX convert args. */
2607 	vm_object_t object = (vm_object_t)addr;
2608 	boolean_t full = have_addr;
2609 
2610 	vm_page_t p;
2611 
2612 	/* XXX count is an (unused) arg.  Avoid shadowing it. */
2613 #define	count	was_count
2614 
2615 	int count;
2616 
2617 	if (object == NULL)
2618 		return;
2619 
2620 	db_iprintf(
2621 	    "Object %p: type=%d, size=0x%jx, res=%d, ref=%d, flags=0x%x ruid %d charge %jx\n",
2622 	    object, (int)object->type, (uintmax_t)object->size,
2623 	    object->resident_page_count, object->ref_count, object->flags,
2624 	    object->cred ? object->cred->cr_ruid : -1, (uintmax_t)object->charge);
2625 	db_iprintf(" sref=%d, backing_object(%d)=(%p)+0x%jx\n",
2626 	    object->shadow_count,
2627 	    object->backing_object ? object->backing_object->ref_count : 0,
2628 	    object->backing_object, (uintmax_t)object->backing_object_offset);
2629 
2630 	if (!full)
2631 		return;
2632 
2633 	db_indent += 2;
2634 	count = 0;
2635 	TAILQ_FOREACH(p, &object->memq, listq) {
2636 		if (count == 0)
2637 			db_iprintf("memory:=");
2638 		else if (count == 6) {
2639 			db_printf("\n");
2640 			db_iprintf(" ...");
2641 			count = 0;
2642 		} else
2643 			db_printf(",");
2644 		count++;
2645 
2646 		db_printf("(off=0x%jx,page=0x%jx)",
2647 		    (uintmax_t)p->pindex, (uintmax_t)VM_PAGE_TO_PHYS(p));
2648 	}
2649 	if (count != 0)
2650 		db_printf("\n");
2651 	db_indent -= 2;
2652 }
2653 
2654 /* XXX. */
2655 #undef count
2656 
2657 /* XXX need this non-static entry for calling from vm_map_print. */
2658 void
2659 vm_object_print(
2660         /* db_expr_t */ long addr,
2661 	boolean_t have_addr,
2662 	/* db_expr_t */ long count,
2663 	char *modif)
2664 {
2665 	vm_object_print_static(addr, have_addr, count, modif);
2666 }
2667 
2668 DB_SHOW_COMMAND(vmopag, vm_object_print_pages)
2669 {
2670 	vm_object_t object;
2671 	vm_pindex_t fidx;
2672 	vm_paddr_t pa;
2673 	vm_page_t m, prev_m;
2674 	int rcount, nl, c;
2675 
2676 	nl = 0;
2677 	TAILQ_FOREACH(object, &vm_object_list, object_list) {
2678 		db_printf("new object: %p\n", (void *)object);
2679 		if (nl > 18) {
2680 			c = cngetc();
2681 			if (c != ' ')
2682 				return;
2683 			nl = 0;
2684 		}
2685 		nl++;
2686 		rcount = 0;
2687 		fidx = 0;
2688 		pa = -1;
2689 		TAILQ_FOREACH(m, &object->memq, listq) {
2690 			if (m->pindex > 128)
2691 				break;
2692 			if ((prev_m = TAILQ_PREV(m, pglist, listq)) != NULL &&
2693 			    prev_m->pindex + 1 != m->pindex) {
2694 				if (rcount) {
2695 					db_printf(" index(%ld)run(%d)pa(0x%lx)\n",
2696 						(long)fidx, rcount, (long)pa);
2697 					if (nl > 18) {
2698 						c = cngetc();
2699 						if (c != ' ')
2700 							return;
2701 						nl = 0;
2702 					}
2703 					nl++;
2704 					rcount = 0;
2705 				}
2706 			}
2707 			if (rcount &&
2708 				(VM_PAGE_TO_PHYS(m) == pa + rcount * PAGE_SIZE)) {
2709 				++rcount;
2710 				continue;
2711 			}
2712 			if (rcount) {
2713 				db_printf(" index(%ld)run(%d)pa(0x%lx)\n",
2714 					(long)fidx, rcount, (long)pa);
2715 				if (nl > 18) {
2716 					c = cngetc();
2717 					if (c != ' ')
2718 						return;
2719 					nl = 0;
2720 				}
2721 				nl++;
2722 			}
2723 			fidx = m->pindex;
2724 			pa = VM_PAGE_TO_PHYS(m);
2725 			rcount = 1;
2726 		}
2727 		if (rcount) {
2728 			db_printf(" index(%ld)run(%d)pa(0x%lx)\n",
2729 				(long)fidx, rcount, (long)pa);
2730 			if (nl > 18) {
2731 				c = cngetc();
2732 				if (c != ' ')
2733 					return;
2734 				nl = 0;
2735 			}
2736 			nl++;
2737 		}
2738 	}
2739 }
2740 #endif /* DDB */
2741