xref: /freebsd/sys/vm/vm_fault.c (revision 05c7a37afb48ddd5ee1bd921a5d46fe59cc70b15)
1 /*
2  * Copyright (c) 1991, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  * Copyright (c) 1994 John S. Dyson
5  * All rights reserved.
6  * Copyright (c) 1994 David Greenman
7  * All rights reserved.
8  *
9  *
10  * This code is derived from software contributed to Berkeley by
11  * The Mach Operating System project at Carnegie-Mellon University.
12  *
13  * Redistribution and use in source and binary forms, with or without
14  * modification, are permitted provided that the following conditions
15  * are met:
16  * 1. Redistributions of source code must retain the above copyright
17  *    notice, this list of conditions and the following disclaimer.
18  * 2. Redistributions in binary form must reproduce the above copyright
19  *    notice, this list of conditions and the following disclaimer in the
20  *    documentation and/or other materials provided with the distribution.
21  * 3. All advertising materials mentioning features or use of this software
22  *    must display the following acknowledgement:
23  *	This product includes software developed by the University of
24  *	California, Berkeley and its contributors.
25  * 4. Neither the name of the University nor the names of its contributors
26  *    may be used to endorse or promote products derived from this software
27  *    without specific prior written permission.
28  *
29  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
30  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
31  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
32  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
33  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
34  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
35  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
36  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
37  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
38  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
39  * SUCH DAMAGE.
40  *
41  *	from: @(#)vm_fault.c	8.4 (Berkeley) 1/12/94
42  *
43  *
44  * Copyright (c) 1987, 1990 Carnegie-Mellon University.
45  * All rights reserved.
46  *
47  * Authors: Avadis Tevanian, Jr., Michael Wayne Young
48  *
49  * Permission to use, copy, modify and distribute this software and
50  * its documentation is hereby granted, provided that both the copyright
51  * notice and this permission notice appear in all copies of the
52  * software, derivative works or modified versions, and any portions
53  * thereof, and that both notices appear in supporting documentation.
54  *
55  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
56  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
57  * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
58  *
59  * Carnegie Mellon requests users of this software to return to
60  *
61  *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
62  *  School of Computer Science
63  *  Carnegie Mellon University
64  *  Pittsburgh PA 15213-3890
65  *
66  * any improvements or extensions that they make and grant Carnegie the
67  * rights to redistribute these changes.
68  *
69  * $Id: vm_fault.c,v 1.42 1996/03/09 06:48:26 dyson Exp $
70  */
71 
72 /*
73  *	Page fault handling module.
74  */
75 
76 #include <sys/param.h>
77 #include <sys/systm.h>
78 #include <sys/proc.h>
79 #include <sys/vnode.h>
80 #include <sys/resource.h>
81 #include <sys/signalvar.h>
82 #include <sys/resourcevar.h>
83 #include <sys/vmmeter.h>
84 
85 #include <vm/vm.h>
86 #include <vm/vm_param.h>
87 #include <vm/vm_prot.h>
88 #include <vm/lock.h>
89 #include <vm/pmap.h>
90 #include <vm/vm_map.h>
91 #include <vm/vm_object.h>
92 #include <vm/vm_page.h>
93 #include <vm/vm_pageout.h>
94 #include <vm/vm_kern.h>
95 #include <vm/vm_pager.h>
96 #include <vm/vnode_pager.h>
97 #include <vm/swap_pager.h>
98 #include <vm/vm_extern.h>
99 
100 int vm_fault_additional_pages __P((vm_page_t, int, int, vm_page_t *, int *));
101 
102 #define VM_FAULT_READ_AHEAD 4
103 #define VM_FAULT_READ_BEHIND 3
104 #define VM_FAULT_READ (VM_FAULT_READ_AHEAD+VM_FAULT_READ_BEHIND+1)
105 
106 int vm_fault_free_1;
107 int vm_fault_copy_save_1;
108 int vm_fault_copy_save_2;
109 
110 /*
111  *	vm_fault:
112  *
113  *	Handle a page fault occuring at the given address,
114  *	requiring the given permissions, in the map specified.
115  *	If successful, the page is inserted into the
116  *	associated physical map.
117  *
118  *	NOTE: the given address should be truncated to the
119  *	proper page address.
120  *
121  *	KERN_SUCCESS is returned if the page fault is handled; otherwise,
122  *	a standard error specifying why the fault is fatal is returned.
123  *
124  *
125  *	The map in question must be referenced, and remains so.
126  *	Caller may hold no locks.
127  */
128 int
129 vm_fault(map, vaddr, fault_type, change_wiring)
130 	vm_map_t map;
131 	vm_offset_t vaddr;
132 	vm_prot_t fault_type;
133 	boolean_t change_wiring;
134 {
135 	vm_object_t first_object;
136 	vm_pindex_t first_pindex;
137 	vm_map_entry_t entry;
138 	register vm_object_t object;
139 	register vm_pindex_t pindex;
140 	vm_page_t m;
141 	vm_page_t first_m;
142 	vm_prot_t prot;
143 	int result;
144 	boolean_t wired;
145 	boolean_t su;
146 	boolean_t lookup_still_valid;
147 	vm_page_t old_m;
148 	vm_object_t next_object;
149 	vm_page_t marray[VM_FAULT_READ];
150 	int hardfault = 0;
151 	struct vnode *vp = NULL;
152 
153 	cnt.v_vm_faults++;	/* needs lock XXX */
154 /*
155  *	Recovery actions
156  */
157 #define	FREE_PAGE(m)	{				\
158 	PAGE_WAKEUP(m);					\
159 	vm_page_free(m);				\
160 }
161 
162 #define	RELEASE_PAGE(m)	{				\
163 	PAGE_WAKEUP(m);					\
164 	if (m->queue != PQ_ACTIVE) vm_page_activate(m);		\
165 }
166 
167 #define	UNLOCK_MAP	{				\
168 	if (lookup_still_valid) {			\
169 		vm_map_lookup_done(map, entry);		\
170 		lookup_still_valid = FALSE;		\
171 	}						\
172 }
173 
174 #define	UNLOCK_THINGS	{				\
175 	vm_object_pip_wakeup(object); \
176 	if (object != first_object) {			\
177 		FREE_PAGE(first_m);			\
178 		vm_object_pip_wakeup(first_object); \
179 	}						\
180 	UNLOCK_MAP;					\
181 	if (vp != NULL) VOP_UNLOCK(vp);			\
182 }
183 
184 #define	UNLOCK_AND_DEALLOCATE	{			\
185 	UNLOCK_THINGS;					\
186 	vm_object_deallocate(first_object);		\
187 }
188 
189 
190 RetryFault:;
191 
192 	/*
193 	 * Find the backing store object and offset into it to begin the
194 	 * search.
195 	 */
196 
197 	if ((result = vm_map_lookup(&map, vaddr,
198 		fault_type, &entry, &first_object,
199 		&first_pindex, &prot, &wired, &su)) != KERN_SUCCESS) {
200 		return (result);
201 	}
202 
203 	vp = vnode_pager_lock(first_object);
204 
205 	lookup_still_valid = TRUE;
206 
207 	if (wired)
208 		fault_type = prot;
209 
210 	first_m = NULL;
211 
212 	/*
213 	 * Make a reference to this object to prevent its disposal while we
214 	 * are messing with it.  Once we have the reference, the map is free
215 	 * to be diddled.  Since objects reference their shadows (and copies),
216 	 * they will stay around as well.
217 	 */
218 
219 	first_object->ref_count++;
220 	first_object->paging_in_progress++;
221 
222 	/*
223 	 * INVARIANTS (through entire routine):
224 	 *
225 	 * 1)	At all times, we must either have the object lock or a busy
226 	 * page in some object to prevent some other process from trying to
227 	 * bring in the same page.
228 	 *
229 	 * Note that we cannot hold any locks during the pager access or when
230 	 * waiting for memory, so we use a busy page then.
231 	 *
232 	 * Note also that we aren't as concerned about more than one thead
233 	 * attempting to pager_data_unlock the same page at once, so we don't
234 	 * hold the page as busy then, but do record the highest unlock value
235 	 * so far.  [Unlock requests may also be delivered out of order.]
236 	 *
237 	 * 2)	Once we have a busy page, we must remove it from the pageout
238 	 * queues, so that the pageout daemon will not grab it away.
239 	 *
240 	 * 3)	To prevent another process from racing us down the shadow chain
241 	 * and entering a new page in the top object before we do, we must
242 	 * keep a busy page in the top object while following the shadow
243 	 * chain.
244 	 *
245 	 * 4)	We must increment paging_in_progress on any object for which
246 	 * we have a busy page, to prevent vm_object_collapse from removing
247 	 * the busy page without our noticing.
248 	 */
249 
250 	/*
251 	 * Search for the page at object/offset.
252 	 */
253 
254 	object = first_object;
255 	pindex = first_pindex;
256 
257 	/*
258 	 * See whether this page is resident
259 	 */
260 
261 	while (TRUE) {
262 		m = vm_page_lookup(object, pindex);
263 		if (m != NULL) {
264 			/*
265 			 * If the page is being brought in, wait for it and
266 			 * then retry.
267 			 */
268 			if ((m->flags & PG_BUSY) || m->busy) {
269 				int s;
270 
271 				UNLOCK_THINGS;
272 				s = splhigh();
273 				if ((m->flags & PG_BUSY) || m->busy) {
274 					m->flags |= PG_WANTED | PG_REFERENCED;
275 					cnt.v_intrans++;
276 					tsleep(m, PSWP, "vmpfw", 0);
277 				}
278 				splx(s);
279 				vm_object_deallocate(first_object);
280 				goto RetryFault;
281 			}
282 
283 			/*
284 			 * Mark page busy for other processes, and the pagedaemon.
285 			 */
286 			m->flags |= PG_BUSY;
287 			if ((m->queue == PQ_CACHE) &&
288 			    (cnt.v_free_count + cnt.v_cache_count) < cnt.v_free_reserved) {
289 				UNLOCK_AND_DEALLOCATE;
290 				VM_WAIT;
291 				PAGE_WAKEUP(m);
292 				goto RetryFault;
293 			}
294 
295 			if (m->valid &&
296 				((m->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) &&
297 				m->object != kernel_object && m->object != kmem_object) {
298 				goto readrest;
299 			}
300 			break;
301 		}
302 		if (((object->type != OBJT_DEFAULT) && (!change_wiring || wired))
303 		    || (object == first_object)) {
304 
305 			if (pindex >= object->size) {
306 				UNLOCK_AND_DEALLOCATE;
307 				return (KERN_PROTECTION_FAILURE);
308 			}
309 
310 			/*
311 			 * Allocate a new page for this object/offset pair.
312 			 */
313 			m = vm_page_alloc(object, pindex,
314 				vp?VM_ALLOC_NORMAL:VM_ALLOC_ZERO);
315 
316 			if (m == NULL) {
317 				UNLOCK_AND_DEALLOCATE;
318 				VM_WAIT;
319 				goto RetryFault;
320 			}
321 		}
322 readrest:
323 		if (object->type != OBJT_DEFAULT && (!change_wiring || wired)) {
324 			int rv;
325 			int faultcount;
326 			int reqpage;
327 
328 			/*
329 			 * now we find out if any other pages should be paged
330 			 * in at this time this routine checks to see if the
331 			 * pages surrounding this fault reside in the same
332 			 * object as the page for this fault.  If they do,
333 			 * then they are faulted in also into the object.  The
334 			 * array "marray" returned contains an array of
335 			 * vm_page_t structs where one of them is the
336 			 * vm_page_t passed to the routine.  The reqpage
337 			 * return value is the index into the marray for the
338 			 * vm_page_t passed to the routine.
339 			 */
340 			faultcount = vm_fault_additional_pages(
341 			    m, VM_FAULT_READ_BEHIND, VM_FAULT_READ_AHEAD,
342 			    marray, &reqpage);
343 
344 			/*
345 			 * Call the pager to retrieve the data, if any, after
346 			 * releasing the lock on the map.
347 			 */
348 			UNLOCK_MAP;
349 
350 			rv = faultcount ?
351 			    vm_pager_get_pages(object, marray, faultcount,
352 				reqpage) : VM_PAGER_FAIL;
353 
354 			if (rv == VM_PAGER_OK) {
355 				/*
356 				 * Found the page. Leave it busy while we play
357 				 * with it.
358 				 */
359 
360 				/*
361 				 * Relookup in case pager changed page. Pager
362 				 * is responsible for disposition of old page
363 				 * if moved.
364 				 */
365 				m = vm_page_lookup(object, pindex);
366 				if( !m) {
367 					UNLOCK_AND_DEALLOCATE;
368 					goto RetryFault;
369 				}
370 
371 				hardfault++;
372 				break;
373 			}
374 			/*
375 			 * Remove the bogus page (which does not exist at this
376 			 * object/offset); before doing so, we must get back
377 			 * our object lock to preserve our invariant.
378 			 *
379 			 * Also wake up any other process that may want to bring
380 			 * in this page.
381 			 *
382 			 * If this is the top-level object, we must leave the
383 			 * busy page to prevent another process from rushing
384 			 * past us, and inserting the page in that object at
385 			 * the same time that we are.
386 			 */
387 
388 			if (rv == VM_PAGER_ERROR)
389 				printf("vm_fault: pager input (probably hardware) error, PID %d failure\n",
390 				    curproc->p_pid);
391 			/*
392 			 * Data outside the range of the pager or an I/O error
393 			 */
394 			/*
395 			 * XXX - the check for kernel_map is a kludge to work
396 			 * around having the machine panic on a kernel space
397 			 * fault w/ I/O error.
398 			 */
399 			if (((map != kernel_map) && (rv == VM_PAGER_ERROR)) || (rv == VM_PAGER_BAD)) {
400 				FREE_PAGE(m);
401 				UNLOCK_AND_DEALLOCATE;
402 				return ((rv == VM_PAGER_ERROR) ? KERN_FAILURE : KERN_PROTECTION_FAILURE);
403 			}
404 			if (object != first_object) {
405 				FREE_PAGE(m);
406 				/*
407 				 * XXX - we cannot just fall out at this
408 				 * point, m has been freed and is invalid!
409 				 */
410 			}
411 		}
412 		/*
413 		 * We get here if the object has default pager (or unwiring) or the
414 		 * pager doesn't have the page.
415 		 */
416 		if (object == first_object)
417 			first_m = m;
418 
419 		/*
420 		 * Move on to the next object.  Lock the next object before
421 		 * unlocking the current one.
422 		 */
423 
424 		pindex += OFF_TO_IDX(object->backing_object_offset);
425 		next_object = object->backing_object;
426 		if (next_object == NULL) {
427 			/*
428 			 * If there's no object left, fill the page in the top
429 			 * object with zeros.
430 			 */
431 			if (object != first_object) {
432 				vm_object_pip_wakeup(object);
433 
434 				object = first_object;
435 				pindex = first_pindex;
436 				m = first_m;
437 			}
438 			first_m = NULL;
439 
440 			if ((m->flags & PG_ZERO) == 0)
441 				vm_page_zero_fill(m);
442 			cnt.v_zfod++;
443 			break;
444 		} else {
445 			if (object != first_object) {
446 				vm_object_pip_wakeup(object);
447 			}
448 			object = next_object;
449 			object->paging_in_progress++;
450 		}
451 	}
452 
453 	if ((m->flags & PG_BUSY) == 0)
454 		panic("vm_fault: not busy after main loop");
455 
456 	/*
457 	 * PAGE HAS BEEN FOUND. [Loop invariant still holds -- the object lock
458 	 * is held.]
459 	 */
460 
461 	old_m = m;	/* save page that would be copied */
462 
463 	/*
464 	 * If the page is being written, but isn't already owned by the
465 	 * top-level object, we have to copy it into a new page owned by the
466 	 * top-level object.
467 	 */
468 
469 	if (object != first_object) {
470 		/*
471 		 * We only really need to copy if we want to write it.
472 		 */
473 
474 		if (fault_type & VM_PROT_WRITE) {
475 
476 			/*
477 			 * We already have an empty page in first_object - use
478 			 * it.
479 			 */
480 
481 			if (lookup_still_valid &&
482 				/*
483 				 * Only one shadow object
484 				 */
485 				(object->shadow_count == 1) &&
486 				/*
487 				 * No COW refs, except us
488 				 */
489 				(object->ref_count == 1) &&
490 				/*
491 				 * Noone else can look this object up
492 				 */
493 				(object->handle == NULL) &&
494 				/*
495 				 * No other ways to look the object up
496 				 */
497 				((object->type == OBJT_DEFAULT) ||
498 				 (object->type == OBJT_SWAP)) &&
499 				/*
500 				 * We don't chase down the shadow chain
501 				 */
502 				(object == first_object->backing_object)) {
503 
504 				/*
505 				 * get rid of the unnecessary page
506 				 */
507 				vm_page_protect(first_m, VM_PROT_NONE);
508 				PAGE_WAKEUP(first_m);
509 				vm_page_free(first_m);
510 				/*
511 				 * grab the page and put it into the process'es object
512 				 */
513 				vm_page_rename(m, first_object, first_pindex);
514 				first_m = m;
515 				m->dirty = VM_PAGE_BITS_ALL;
516 				m = NULL;
517 				++vm_fault_copy_save_1;
518 			} else {
519 				/*
520 				 * Oh, well, lets copy it.
521 				 */
522 				vm_page_copy(m, first_m);
523 			}
524 
525 			if (lookup_still_valid &&
526 				/*
527 				 * make sure that we have two shadow objs
528 				 */
529 				(object->shadow_count == 2) &&
530 				/*
531 				 * And no COW refs -- note that there are sometimes
532 				 * temp refs to objs, but ignore that case -- we just
533 				 * punt.
534 				 */
535 				(object->ref_count == 2) &&
536 				/*
537 				 * Noone else can look us up
538 				 */
539 				(object->handle == NULL) &&
540 				/*
541 				 * Not something that can be referenced elsewhere
542 				 */
543 				((object->type == OBJT_DEFAULT) ||
544 				 (object->type == OBJT_SWAP)) &&
545 				/*
546 				 * We don't bother chasing down object chain
547 				 */
548 				(object == first_object->backing_object)) {
549 
550 				vm_object_t other_object;
551 				vm_pindex_t other_pindex, other_pindex_offset;
552 				vm_page_t tm;
553 
554 				other_object = object->shadow_head.tqh_first;
555 				if (other_object == first_object)
556 					other_object = other_object->shadow_list.tqe_next;
557 				if (!other_object)
558 					panic("vm_fault: other object missing");
559 				if (other_object &&
560 					(other_object->type == OBJT_DEFAULT) &&
561 					(other_object->paging_in_progress == 0)) {
562 					other_pindex_offset =
563 						OFF_TO_IDX(other_object->backing_object_offset);
564 					if (pindex >= other_pindex_offset) {
565 						other_pindex = pindex - other_pindex_offset;
566 						/*
567 						 * If the other object has the page, just free it.
568 						 */
569 						if ((tm = vm_page_lookup(other_object, other_pindex))) {
570 							if ((tm->flags & PG_BUSY) == 0 &&
571 								tm->busy == 0 &&
572 								tm->valid == VM_PAGE_BITS_ALL) {
573 								/*
574 								 * get rid of the unnecessary page
575 								 */
576 								vm_page_protect(m, VM_PROT_NONE);
577 								PAGE_WAKEUP(m);
578 								vm_page_free(m);
579 								m = NULL;
580 								++vm_fault_free_1;
581 								tm->dirty = VM_PAGE_BITS_ALL;
582 								first_m->dirty = VM_PAGE_BITS_ALL;
583 							}
584 						} else {
585 							/*
586 							 * If the other object doesn't have the page,
587 							 * then we move it there.
588 							 */
589 							vm_page_rename(m, other_object, other_pindex);
590 							m->dirty = VM_PAGE_BITS_ALL;
591 							m->valid = VM_PAGE_BITS_ALL;
592 							++vm_fault_copy_save_2;
593 						}
594 					}
595 				}
596 			}
597 
598 			if (m) {
599 				if (m->queue != PQ_ACTIVE)
600 					vm_page_activate(m);
601 			/*
602 			 * We no longer need the old page or object.
603 			 */
604 				PAGE_WAKEUP(m);
605 			}
606 
607 			vm_object_pip_wakeup(object);
608 			/*
609 			 * Only use the new page below...
610 			 */
611 
612 			cnt.v_cow_faults++;
613 			m = first_m;
614 			object = first_object;
615 			pindex = first_pindex;
616 
617 			/*
618 			 * Now that we've gotten the copy out of the way,
619 			 * let's try to collapse the top object.
620 			 *
621 			 * But we have to play ugly games with
622 			 * paging_in_progress to do that...
623 			 */
624 			vm_object_pip_wakeup(object);
625 			vm_object_collapse(object);
626 			object->paging_in_progress++;
627 		} else {
628 			prot &= ~VM_PROT_WRITE;
629 		}
630 	}
631 
632 	/*
633 	 * We must verify that the maps have not changed since our last
634 	 * lookup.
635 	 */
636 
637 	if (!lookup_still_valid) {
638 		vm_object_t retry_object;
639 		vm_pindex_t retry_pindex;
640 		vm_prot_t retry_prot;
641 
642 		/*
643 		 * Since map entries may be pageable, make sure we can take a
644 		 * page fault on them.
645 		 */
646 
647 		/*
648 		 * To avoid trying to write_lock the map while another process
649 		 * has it read_locked (in vm_map_pageable), we do not try for
650 		 * write permission.  If the page is still writable, we will
651 		 * get write permission.  If it is not, or has been marked
652 		 * needs_copy, we enter the mapping without write permission,
653 		 * and will merely take another fault.
654 		 */
655 		result = vm_map_lookup(&map, vaddr, fault_type & ~VM_PROT_WRITE,
656 		    &entry, &retry_object, &retry_pindex, &retry_prot, &wired, &su);
657 
658 		/*
659 		 * If we don't need the page any longer, put it on the active
660 		 * list (the easiest thing to do here).  If no one needs it,
661 		 * pageout will grab it eventually.
662 		 */
663 
664 		if (result != KERN_SUCCESS) {
665 			RELEASE_PAGE(m);
666 			UNLOCK_AND_DEALLOCATE;
667 			return (result);
668 		}
669 		lookup_still_valid = TRUE;
670 
671 		if ((retry_object != first_object) ||
672 		    (retry_pindex != first_pindex)) {
673 			RELEASE_PAGE(m);
674 			UNLOCK_AND_DEALLOCATE;
675 			goto RetryFault;
676 		}
677 		/*
678 		 * Check whether the protection has changed or the object has
679 		 * been copied while we left the map unlocked. Changing from
680 		 * read to write permission is OK - we leave the page
681 		 * write-protected, and catch the write fault. Changing from
682 		 * write to read permission means that we can't mark the page
683 		 * write-enabled after all.
684 		 */
685 		prot &= retry_prot;
686 	}
687 
688 	/*
689 	 * Put this page into the physical map. We had to do the unlock above
690 	 * because pmap_enter may cause other faults.   We don't put the page
691 	 * back on the active queue until later so that the page-out daemon
692 	 * won't find us (yet).
693 	 */
694 
695 	if (prot & VM_PROT_WRITE) {
696 		m->flags |= PG_WRITEABLE;
697 		m->object->flags |= OBJ_WRITEABLE|OBJ_MIGHTBEDIRTY;
698 		/*
699 		 * If the fault is a write, we know that this page is being
700 		 * written NOW. This will save on the pmap_is_modified() calls
701 		 * later.
702 		 */
703 		if (fault_type & VM_PROT_WRITE) {
704 			m->dirty = VM_PAGE_BITS_ALL;
705 		}
706 	}
707 
708 	UNLOCK_THINGS;
709 
710 	m->flags |= PG_MAPPED|PG_REFERENCED;
711 	m->flags &= ~PG_ZERO;
712 	m->valid = VM_PAGE_BITS_ALL;
713 
714 	pmap_enter(map->pmap, vaddr, VM_PAGE_TO_PHYS(m), prot, wired);
715 	if (vp && (change_wiring == 0) && (wired == 0))
716 		pmap_prefault(map->pmap, vaddr, entry, first_object);
717 
718 	/*
719 	 * If the page is not wired down, then put it where the pageout daemon
720 	 * can find it.
721 	 */
722 	if (change_wiring) {
723 		if (wired)
724 			vm_page_wire(m);
725 		else
726 			vm_page_unwire(m);
727 	} else {
728 		if (m->queue != PQ_ACTIVE)
729 			vm_page_activate(m);
730 	}
731 
732 	if (curproc && (curproc->p_flag & P_INMEM) && curproc->p_stats) {
733 		if (hardfault) {
734 			curproc->p_stats->p_ru.ru_majflt++;
735 		} else {
736 			curproc->p_stats->p_ru.ru_minflt++;
737 		}
738 	}
739 
740 	/*
741 	 * Unlock everything, and return
742 	 */
743 
744 	PAGE_WAKEUP(m);
745 	vm_object_deallocate(first_object);
746 
747 	return (KERN_SUCCESS);
748 
749 }
750 
751 /*
752  *	vm_fault_wire:
753  *
754  *	Wire down a range of virtual addresses in a map.
755  */
756 int
757 vm_fault_wire(map, start, end)
758 	vm_map_t map;
759 	vm_offset_t start, end;
760 {
761 
762 	register vm_offset_t va;
763 	register pmap_t pmap;
764 	int rv;
765 
766 	pmap = vm_map_pmap(map);
767 
768 	/*
769 	 * Inform the physical mapping system that the range of addresses may
770 	 * not fault, so that page tables and such can be locked down as well.
771 	 */
772 
773 	pmap_pageable(pmap, start, end, FALSE);
774 
775 	/*
776 	 * We simulate a fault to get the page and enter it in the physical
777 	 * map.
778 	 */
779 
780 	for (va = start; va < end; va += PAGE_SIZE) {
781 
782 		while( curproc != pageproc &&
783 			(cnt.v_free_count <= cnt.v_pageout_free_min))
784 			VM_WAIT;
785 
786 		rv = vm_fault(map, va, VM_PROT_READ|VM_PROT_WRITE, TRUE);
787 		if (rv) {
788 			if (va != start)
789 				vm_fault_unwire(map, start, va);
790 			return (rv);
791 		}
792 	}
793 	return (KERN_SUCCESS);
794 }
795 
796 
797 /*
798  *	vm_fault_unwire:
799  *
800  *	Unwire a range of virtual addresses in a map.
801  */
802 void
803 vm_fault_unwire(map, start, end)
804 	vm_map_t map;
805 	vm_offset_t start, end;
806 {
807 
808 	register vm_offset_t va, pa;
809 	register pmap_t pmap;
810 
811 	pmap = vm_map_pmap(map);
812 
813 	/*
814 	 * Since the pages are wired down, we must be able to get their
815 	 * mappings from the physical map system.
816 	 */
817 
818 	for (va = start; va < end; va += PAGE_SIZE) {
819 		pa = pmap_extract(pmap, va);
820 		if (pa == (vm_offset_t) 0) {
821 			panic("unwire: page not in pmap");
822 		}
823 		pmap_change_wiring(pmap, va, FALSE);
824 		vm_page_unwire(PHYS_TO_VM_PAGE(pa));
825 	}
826 
827 	/*
828 	 * Inform the physical mapping system that the range of addresses may
829 	 * fault, so that page tables and such may be unwired themselves.
830 	 */
831 
832 	pmap_pageable(pmap, start, end, TRUE);
833 
834 }
835 
836 /*
837  *	Routine:
838  *		vm_fault_copy_entry
839  *	Function:
840  *		Copy all of the pages from a wired-down map entry to another.
841  *
842  *	In/out conditions:
843  *		The source and destination maps must be locked for write.
844  *		The source map entry must be wired down (or be a sharing map
845  *		entry corresponding to a main map entry that is wired down).
846  */
847 
848 void
849 vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry)
850 	vm_map_t dst_map;
851 	vm_map_t src_map;
852 	vm_map_entry_t dst_entry;
853 	vm_map_entry_t src_entry;
854 {
855 	vm_object_t dst_object;
856 	vm_object_t src_object;
857 	vm_ooffset_t dst_offset;
858 	vm_ooffset_t src_offset;
859 	vm_prot_t prot;
860 	vm_offset_t vaddr;
861 	vm_page_t dst_m;
862 	vm_page_t src_m;
863 
864 #ifdef	lint
865 	src_map++;
866 #endif	/* lint */
867 
868 	src_object = src_entry->object.vm_object;
869 	src_offset = src_entry->offset;
870 
871 	/*
872 	 * Create the top-level object for the destination entry. (Doesn't
873 	 * actually shadow anything - we copy the pages directly.)
874 	 */
875 	dst_object = vm_object_allocate(OBJT_DEFAULT,
876 	    (vm_size_t) OFF_TO_IDX(dst_entry->end - dst_entry->start));
877 
878 	dst_entry->object.vm_object = dst_object;
879 	dst_entry->offset = 0;
880 
881 	prot = dst_entry->max_protection;
882 
883 	/*
884 	 * Loop through all of the pages in the entry's range, copying each
885 	 * one from the source object (it should be there) to the destination
886 	 * object.
887 	 */
888 	for (vaddr = dst_entry->start, dst_offset = 0;
889 	    vaddr < dst_entry->end;
890 	    vaddr += PAGE_SIZE, dst_offset += PAGE_SIZE) {
891 
892 		/*
893 		 * Allocate a page in the destination object
894 		 */
895 		do {
896 			dst_m = vm_page_alloc(dst_object,
897 				OFF_TO_IDX(dst_offset), VM_ALLOC_NORMAL);
898 			if (dst_m == NULL) {
899 				VM_WAIT;
900 			}
901 		} while (dst_m == NULL);
902 
903 		/*
904 		 * Find the page in the source object, and copy it in.
905 		 * (Because the source is wired down, the page will be in
906 		 * memory.)
907 		 */
908 		src_m = vm_page_lookup(src_object,
909 			OFF_TO_IDX(dst_offset + src_offset));
910 		if (src_m == NULL)
911 			panic("vm_fault_copy_wired: page missing");
912 
913 		vm_page_copy(src_m, dst_m);
914 
915 		/*
916 		 * Enter it in the pmap...
917 		 */
918 
919 		dst_m->flags |= PG_WRITEABLE|PG_MAPPED;
920 		dst_m->flags &= ~PG_ZERO;
921 		pmap_enter(dst_map->pmap, vaddr, VM_PAGE_TO_PHYS(dst_m),
922 		    prot, FALSE);
923 
924 		/*
925 		 * Mark it no longer busy, and put it on the active list.
926 		 */
927 		vm_page_activate(dst_m);
928 		PAGE_WAKEUP(dst_m);
929 	}
930 }
931 
932 
933 /*
934  * This routine checks around the requested page for other pages that
935  * might be able to be faulted in.  This routine brackets the viable
936  * pages for the pages to be paged in.
937  *
938  * Inputs:
939  *	m, rbehind, rahead
940  *
941  * Outputs:
942  *  marray (array of vm_page_t), reqpage (index of requested page)
943  *
944  * Return value:
945  *  number of pages in marray
946  */
947 int
948 vm_fault_additional_pages(m, rbehind, rahead, marray, reqpage)
949 	vm_page_t m;
950 	int rbehind;
951 	int rahead;
952 	vm_page_t *marray;
953 	int *reqpage;
954 {
955 	int i;
956 	vm_object_t object;
957 	vm_pindex_t pindex, startpindex, endpindex, tpindex;
958 	vm_offset_t size;
959 	vm_page_t rtm;
960 	int treqpage;
961 	int cbehind, cahead;
962 
963 	object = m->object;
964 	pindex = m->pindex;
965 
966 	/*
967 	 * if the requested page is not available, then give up now
968 	 */
969 
970 	if (!vm_pager_has_page(object,
971 		OFF_TO_IDX(object->paging_offset) + pindex, &cbehind, &cahead))
972 		return 0;
973 
974 	if ((cbehind == 0) && (cahead == 0)) {
975 		*reqpage = 0;
976 		marray[0] = m;
977 		return 1;
978 	}
979 
980 	if (rahead > cahead) {
981 		rahead = cahead;
982 	}
983 
984 	if (rbehind > cbehind) {
985 		rbehind = cbehind;
986 	}
987 
988 	/*
989 	 * try to do any readahead that we might have free pages for.
990 	 */
991 	if ((rahead + rbehind) >
992 		((cnt.v_free_count + cnt.v_cache_count) - cnt.v_free_reserved)) {
993 		pagedaemon_wakeup();
994 		*reqpage = 0;
995 		marray[0] = m;
996 		return 1;
997 	}
998 
999 	/*
1000 	 * scan backward for the read behind pages -- in memory or on disk not
1001 	 * in same object
1002 	 */
1003 	tpindex = pindex - 1;
1004 	if (tpindex < pindex) {
1005 		if (rbehind > pindex)
1006 			rbehind = pindex;
1007 		startpindex = pindex - rbehind;
1008 		while (tpindex >= startpindex) {
1009 			if (vm_page_lookup( object, tpindex)) {
1010 				startpindex = tpindex + 1;
1011 				break;
1012 			}
1013 			if (tpindex == 0)
1014 				break;
1015 			tpindex -= 1;
1016 		}
1017 	} else {
1018 		startpindex = pindex;
1019 	}
1020 
1021 	/*
1022 	 * scan forward for the read ahead pages -- in memory or on disk not
1023 	 * in same object
1024 	 */
1025 	tpindex = pindex + 1;
1026 	endpindex = pindex + (rahead + 1);
1027 	if (endpindex > object->size)
1028 		endpindex = object->size;
1029 	while (tpindex <  endpindex) {
1030 		if ( vm_page_lookup(object, tpindex)) {
1031 			break;
1032 		}
1033 		tpindex += 1;
1034 	}
1035 	endpindex = tpindex;
1036 
1037 	/* calculate number of bytes of pages */
1038 	size = endpindex - startpindex;
1039 
1040 	/* calculate the page offset of the required page */
1041 	treqpage = pindex - startpindex;
1042 
1043 	/* see if we have space (again) */
1044 	if ((cnt.v_free_count + cnt.v_cache_count) >
1045 		(cnt.v_free_reserved + size)) {
1046 		/*
1047 		 * get our pages and don't block for them
1048 		 */
1049 		for (i = 0; i < size; i++) {
1050 			if (i != treqpage) {
1051 				rtm = vm_page_alloc(object,
1052 					startpindex + i,
1053 					VM_ALLOC_NORMAL);
1054 				if (rtm == NULL) {
1055 					if (i < treqpage) {
1056 						int j;
1057 						for (j = 0; j < i; j++) {
1058 							FREE_PAGE(marray[j]);
1059 						}
1060 						*reqpage = 0;
1061 						marray[0] = m;
1062 						return 1;
1063 					} else {
1064 						size = i;
1065 						*reqpage = treqpage;
1066 						return size;
1067 					}
1068 				}
1069 				marray[i] = rtm;
1070 			} else {
1071 				marray[i] = m;
1072 			}
1073 		}
1074 
1075 		*reqpage = treqpage;
1076 		return size;
1077 	}
1078 	*reqpage = 0;
1079 	marray[0] = m;
1080 	return 1;
1081 }
1082