xref: /freebsd/sys/vm/vm_map.c (revision 4a0f765fbf09711e612e86fce8bb09ec43f482d9)
1 /*
2  * Copyright (c) 1991, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * This code is derived from software contributed to Berkeley by
6  * The Mach Operating System project at Carnegie-Mellon University.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. All advertising materials mentioning features or use of this software
17  *    must display the following acknowledgement:
18  *	This product includes software developed by the University of
19  *	California, Berkeley and its contributors.
20  * 4. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	from: @(#)vm_map.c	8.3 (Berkeley) 1/12/94
37  *
38  *
39  * Copyright (c) 1987, 1990 Carnegie-Mellon University.
40  * All rights reserved.
41  *
42  * Authors: Avadis Tevanian, Jr., Michael Wayne Young
43  *
44  * Permission to use, copy, modify and distribute this software and
45  * its documentation is hereby granted, provided that both the copyright
46  * notice and this permission notice appear in all copies of the
47  * software, derivative works or modified versions, and any portions
48  * thereof, and that both notices appear in supporting documentation.
49  *
50  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
51  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
52  * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
53  *
54  * Carnegie Mellon requests users of this software to return to
55  *
56  *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
57  *  School of Computer Science
58  *  Carnegie Mellon University
59  *  Pittsburgh PA 15213-3890
60  *
61  * any improvements or extensions that they make and grant Carnegie the
62  * rights to redistribute these changes.
63  *
64  * $Id$
65  */
66 
67 /*
68  *	Virtual memory mapping module.
69  */
70 
71 #include <sys/param.h>
72 #include <sys/systm.h>
73 #include <sys/malloc.h>
74 #include <sys/proc.h>
75 #include <sys/queue.h>
76 #include <sys/vmmeter.h>
77 #include <sys/mman.h>
78 
79 #include <vm/vm.h>
80 #include <vm/vm_param.h>
81 #include <vm/vm_prot.h>
82 #include <vm/vm_inherit.h>
83 #include <sys/lock.h>
84 #include <vm/pmap.h>
85 #include <vm/vm_map.h>
86 #include <vm/vm_page.h>
87 #include <vm/vm_object.h>
88 #include <vm/vm_kern.h>
89 #include <vm/vm_pager.h>
90 #include <vm/vm_extern.h>
91 #include <vm/default_pager.h>
92 
93 /*
94  *	Virtual memory maps provide for the mapping, protection,
95  *	and sharing of virtual memory objects.  In addition,
96  *	this module provides for an efficient virtual copy of
97  *	memory from one map to another.
98  *
99  *	Synchronization is required prior to most operations.
100  *
101  *	Maps consist of an ordered doubly-linked list of simple
102  *	entries; a single hint is used to speed up lookups.
103  *
104  *	In order to properly represent the sharing of virtual
105  *	memory regions among maps, the map structure is bi-level.
106  *	Top-level ("address") maps refer to regions of sharable
107  *	virtual memory.  These regions are implemented as
108  *	("sharing") maps, which then refer to the actual virtual
109  *	memory objects.  When two address maps "share" memory,
110  *	their top-level maps both have references to the same
111  *	sharing map.  When memory is virtual-copied from one
112  *	address map to another, the references in the sharing
113  *	maps are actually copied -- no copying occurs at the
114  *	virtual memory object level.
115  *
116  *	Since portions of maps are specified by start/end addreses,
117  *	which may not align with existing map entries, all
118  *	routines merely "clip" entries to these start/end values.
119  *	[That is, an entry is split into two, bordering at a
120  *	start or end value.]  Note that these clippings may not
121  *	always be necessary (as the two resulting entries are then
122  *	not changed); however, the clipping is done for convenience.
123  *	No attempt is currently made to "glue back together" two
124  *	abutting entries.
125  *
126  *	As mentioned above, virtual copy operations are performed
127  *	by copying VM object references from one sharing map to
128  *	another, and then marking both regions as copy-on-write.
129  *	It is important to note that only one writeable reference
130  *	to a VM object region exists in any map -- this means that
131  *	shadow object creation can be delayed until a write operation
132  *	occurs.
133  */
134 
135 /*
136  *	vm_map_startup:
137  *
138  *	Initialize the vm_map module.  Must be called before
139  *	any other vm_map routines.
140  *
141  *	Map and entry structures are allocated from the general
142  *	purpose memory pool with some exceptions:
143  *
144  *	- The kernel map and kmem submap are allocated statically.
145  *	- Kernel map entries are allocated out of a static pool.
146  *
147  *	These restrictions are necessary since malloc() uses the
148  *	maps and requires map entries.
149  */
150 
151 vm_offset_t kentry_data;
152 vm_size_t kentry_data_size;
153 static vm_map_entry_t kentry_free;
154 static vm_map_t kmap_free;
155 extern char kstack[];
156 extern int inmprotect;
157 
158 static int kentry_count;
159 static vm_offset_t mapvm_start, mapvm, mapvmmax;
160 static int mapvmpgcnt;
161 
162 static struct vm_map_entry *mappool;
163 static int mappoolcnt;
164 #define KENTRY_LOW_WATER 128
165 
166 static void _vm_map_clip_end __P((vm_map_t, vm_map_entry_t, vm_offset_t));
167 static void _vm_map_clip_start __P((vm_map_t, vm_map_entry_t, vm_offset_t));
168 static vm_map_entry_t vm_map_entry_create __P((vm_map_t));
169 static void vm_map_entry_delete __P((vm_map_t, vm_map_entry_t));
170 static void vm_map_entry_dispose __P((vm_map_t, vm_map_entry_t));
171 static void vm_map_entry_unwire __P((vm_map_t, vm_map_entry_t));
172 static void vm_map_copy_entry __P((vm_map_t, vm_map_t, vm_map_entry_t,
173 		vm_map_entry_t));
174 
175 void
176 vm_map_startup()
177 {
178 	register int i;
179 	register vm_map_entry_t mep;
180 	vm_map_t mp;
181 
182 	/*
183 	 * Static map structures for allocation before initialization of
184 	 * kernel map or kmem map.  vm_map_create knows how to deal with them.
185 	 */
186 	kmap_free = mp = (vm_map_t) kentry_data;
187 	i = MAX_KMAP;
188 	while (--i > 0) {
189 		mp->header.next = (vm_map_entry_t) (mp + 1);
190 		mp++;
191 	}
192 	mp++->header.next = NULL;
193 
194 	/*
195 	 * Form a free list of statically allocated kernel map entries with
196 	 * the rest.
197 	 */
198 	kentry_free = mep = (vm_map_entry_t) mp;
199 	kentry_count = i = (kentry_data_size - MAX_KMAP * sizeof *mp) / sizeof *mep;
200 	while (--i > 0) {
201 		mep->next = mep + 1;
202 		mep++;
203 	}
204 	mep->next = NULL;
205 }
206 
207 /*
208  * Allocate a vmspace structure, including a vm_map and pmap,
209  * and initialize those structures.  The refcnt is set to 1.
210  * The remaining fields must be initialized by the caller.
211  */
212 struct vmspace *
213 vmspace_alloc(min, max, pageable)
214 	vm_offset_t min, max;
215 	int pageable;
216 {
217 	register struct vmspace *vm;
218 
219 	if (mapvmpgcnt == 0 && mapvm == 0) {
220 		mapvmpgcnt = (cnt.v_page_count * sizeof(struct vm_map_entry) + PAGE_SIZE - 1) / PAGE_SIZE;
221 		mapvm_start = mapvm = kmem_alloc_pageable(kernel_map,
222 			mapvmpgcnt * PAGE_SIZE);
223 		mapvmmax = mapvm_start + mapvmpgcnt * PAGE_SIZE;
224 		if (!mapvm)
225 			mapvmpgcnt = 0;
226 	}
227 	MALLOC(vm, struct vmspace *, sizeof(struct vmspace), M_VMMAP, M_WAITOK);
228 	bzero(vm, (caddr_t) &vm->vm_startcopy - (caddr_t) vm);
229 	vm_map_init(&vm->vm_map, min, max, pageable);
230 	pmap_pinit(&vm->vm_pmap);
231 	vm->vm_map.pmap = &vm->vm_pmap;		/* XXX */
232 	vm->vm_refcnt = 1;
233 	return (vm);
234 }
235 
236 void
237 vmspace_free(vm)
238 	register struct vmspace *vm;
239 {
240 
241 	if (vm->vm_refcnt == 0)
242 		panic("vmspace_free: attempt to free already freed vmspace");
243 
244 	if (--vm->vm_refcnt == 0) {
245 
246 		/*
247 		 * Lock the map, to wait out all other references to it.
248 		 * Delete all of the mappings and pages they hold, then call
249 		 * the pmap module to reclaim anything left.
250 		 */
251 		vm_map_lock(&vm->vm_map);
252 		(void) vm_map_delete(&vm->vm_map, vm->vm_map.min_offset,
253 		    vm->vm_map.max_offset);
254 		vm_map_unlock(&vm->vm_map);
255 
256 		while( vm->vm_map.ref_count != 1)
257 			tsleep(&vm->vm_map.ref_count, PVM, "vmsfre", 0);
258 		--vm->vm_map.ref_count;
259 		vm_object_pmap_remove(vm->vm_upages_obj,
260 			0, vm->vm_upages_obj->size);
261 		vm_object_deallocate(vm->vm_upages_obj);
262 		pmap_release(&vm->vm_pmap);
263 		FREE(vm, M_VMMAP);
264 	} else {
265 		wakeup(&vm->vm_map.ref_count);
266 	}
267 }
268 
269 /*
270  *	vm_map_create:
271  *
272  *	Creates and returns a new empty VM map with
273  *	the given physical map structure, and having
274  *	the given lower and upper address bounds.
275  */
276 vm_map_t
277 vm_map_create(pmap, min, max, pageable)
278 	pmap_t pmap;
279 	vm_offset_t min, max;
280 	boolean_t pageable;
281 {
282 	register vm_map_t result;
283 
284 	if (kmem_map == NULL) {
285 		result = kmap_free;
286 		if (result == NULL)
287 			panic("vm_map_create: out of maps");
288 		kmap_free = (vm_map_t) result->header.next;
289 	} else
290 		MALLOC(result, vm_map_t, sizeof(struct vm_map),
291 		    M_VMMAP, M_WAITOK);
292 
293 	vm_map_init(result, min, max, pageable);
294 	result->pmap = pmap;
295 	return (result);
296 }
297 
298 /*
299  * Initialize an existing vm_map structure
300  * such as that in the vmspace structure.
301  * The pmap is set elsewhere.
302  */
303 void
304 vm_map_init(map, min, max, pageable)
305 	register struct vm_map *map;
306 	vm_offset_t min, max;
307 	boolean_t pageable;
308 {
309 	map->header.next = map->header.prev = &map->header;
310 	map->nentries = 0;
311 	map->size = 0;
312 	map->ref_count = 1;
313 	map->is_main_map = TRUE;
314 	map->min_offset = min;
315 	map->max_offset = max;
316 	map->entries_pageable = pageable;
317 	map->first_free = &map->header;
318 	map->hint = &map->header;
319 	map->timestamp = 0;
320 	lockinit(&map->lock, PVM, "thrd_sleep", 0, 0);
321 	simple_lock_init(&map->ref_lock);
322 }
323 
324 /*
325  *	vm_map_entry_dispose:	[ internal use only ]
326  *
327  *	Inverse of vm_map_entry_create.
328  */
329 static void
330 vm_map_entry_dispose(map, entry)
331 	vm_map_t map;
332 	vm_map_entry_t entry;
333 {
334 	int s;
335 
336 	if (map == kernel_map || map == kmem_map ||
337 		map == mb_map || map == pager_map) {
338 		s = splvm();
339 		entry->next = kentry_free;
340 		kentry_free = entry;
341 		++kentry_count;
342 		splx(s);
343 	} else {
344 		entry->next = mappool;
345 		mappool = entry;
346 		++mappoolcnt;
347 	}
348 }
349 
350 /*
351  *	vm_map_entry_create:	[ internal use only ]
352  *
353  *	Allocates a VM map entry for insertion.
354  *	No entry fields are filled in.  This routine is
355  */
356 static vm_map_entry_t
357 vm_map_entry_create(map)
358 	vm_map_t map;
359 {
360 	vm_map_entry_t entry;
361 	int i;
362 	int s;
363 
364 	/*
365 	 * This is a *very* nasty (and sort of incomplete) hack!!!!
366 	 */
367 	if (kentry_count < KENTRY_LOW_WATER) {
368 		s = splvm();
369 		if (mapvmpgcnt && mapvm) {
370 			vm_page_t m;
371 
372 			m = vm_page_alloc(kernel_object,
373 			        OFF_TO_IDX(mapvm - VM_MIN_KERNEL_ADDRESS),
374 				    (map == kmem_map || map == mb_map) ? VM_ALLOC_INTERRUPT : VM_ALLOC_NORMAL);
375 
376 			if (m) {
377 				int newentries;
378 
379 				newentries = (PAGE_SIZE / sizeof(struct vm_map_entry));
380 				vm_page_wire(m);
381 				PAGE_WAKEUP(m);
382 				m->valid = VM_PAGE_BITS_ALL;
383 				pmap_kenter(mapvm, VM_PAGE_TO_PHYS(m));
384 				m->flags |= PG_WRITEABLE;
385 
386 				entry = (vm_map_entry_t) mapvm;
387 				mapvm += PAGE_SIZE;
388 				--mapvmpgcnt;
389 
390 				for (i = 0; i < newentries; i++) {
391 					vm_map_entry_dispose(kernel_map, entry);
392 					entry++;
393 				}
394 			}
395 		}
396 		splx(s);
397 	}
398 
399 	if (map == kernel_map || map == kmem_map ||
400 		map == mb_map || map == pager_map) {
401 		s = splvm();
402 		entry = kentry_free;
403 		if (entry) {
404 			kentry_free = entry->next;
405 			--kentry_count;
406 		} else {
407 			panic("vm_map_entry_create: out of map entries for kernel");
408 		}
409 		splx(s);
410 	} else {
411 		entry = mappool;
412 		if (entry) {
413 			mappool = entry->next;
414 			--mappoolcnt;
415 		} else {
416 			MALLOC(entry, vm_map_entry_t, sizeof(struct vm_map_entry),
417 			    M_VMMAPENT, M_WAITOK);
418 		}
419 	}
420 
421 	return (entry);
422 }
423 
424 /*
425  *	vm_map_entry_{un,}link:
426  *
427  *	Insert/remove entries from maps.
428  */
429 #define	vm_map_entry_link(map, after_where, entry) \
430 		{ \
431 		(map)->nentries++; \
432 		(entry)->prev = (after_where); \
433 		(entry)->next = (after_where)->next; \
434 		(entry)->prev->next = (entry); \
435 		(entry)->next->prev = (entry); \
436 		}
437 #define	vm_map_entry_unlink(map, entry) \
438 		{ \
439 		(map)->nentries--; \
440 		(entry)->next->prev = (entry)->prev; \
441 		(entry)->prev->next = (entry)->next; \
442 		}
443 
444 /*
445  *	vm_map_reference:
446  *
447  *	Creates another valid reference to the given map.
448  *
449  */
450 void
451 vm_map_reference(map)
452 	register vm_map_t map;
453 {
454 	if (map == NULL)
455 		return;
456 
457 	map->ref_count++;
458 }
459 
460 /*
461  *	vm_map_deallocate:
462  *
463  *	Removes a reference from the specified map,
464  *	destroying it if no references remain.
465  *	The map should not be locked.
466  */
467 void
468 vm_map_deallocate(map)
469 	register vm_map_t map;
470 {
471 	register int c;
472 
473 	if (map == NULL)
474 		return;
475 
476 	c = map->ref_count;
477 
478 	if (c == 0)
479 		panic("vm_map_deallocate: deallocating already freed map");
480 
481 	if (c != 1) {
482 		--map->ref_count;
483 		wakeup(&map->ref_count);
484 		return;
485 	}
486 	/*
487 	 * Lock the map, to wait out all other references to it.
488 	 */
489 
490 	vm_map_lock_drain_interlock(map);
491 	(void) vm_map_delete(map, map->min_offset, map->max_offset);
492 	--map->ref_count;
493 	if( map->ref_count != 0) {
494 		vm_map_unlock(map);
495 		return;
496 	}
497 
498 	pmap_destroy(map->pmap);
499 
500 	vm_map_unlock(map);
501 
502 	FREE(map, M_VMMAP);
503 }
504 
505 /*
506  *	SAVE_HINT:
507  *
508  *	Saves the specified entry as the hint for
509  *	future lookups.
510  */
511 #define	SAVE_HINT(map,value) \
512 		(map)->hint = (value);
513 
514 /*
515  *	vm_map_lookup_entry:	[ internal use only ]
516  *
517  *	Finds the map entry containing (or
518  *	immediately preceding) the specified address
519  *	in the given map; the entry is returned
520  *	in the "entry" parameter.  The boolean
521  *	result indicates whether the address is
522  *	actually contained in the map.
523  */
524 boolean_t
525 vm_map_lookup_entry(map, address, entry)
526 	register vm_map_t map;
527 	register vm_offset_t address;
528 	vm_map_entry_t *entry;	/* OUT */
529 {
530 	register vm_map_entry_t cur;
531 	register vm_map_entry_t last;
532 
533 	/*
534 	 * Start looking either from the head of the list, or from the hint.
535 	 */
536 
537 	cur = map->hint;
538 
539 	if (cur == &map->header)
540 		cur = cur->next;
541 
542 	if (address >= cur->start) {
543 		/*
544 		 * Go from hint to end of list.
545 		 *
546 		 * But first, make a quick check to see if we are already looking
547 		 * at the entry we want (which is usually the case). Note also
548 		 * that we don't need to save the hint here... it is the same
549 		 * hint (unless we are at the header, in which case the hint
550 		 * didn't buy us anything anyway).
551 		 */
552 		last = &map->header;
553 		if ((cur != last) && (cur->end > address)) {
554 			*entry = cur;
555 			return (TRUE);
556 		}
557 	} else {
558 		/*
559 		 * Go from start to hint, *inclusively*
560 		 */
561 		last = cur->next;
562 		cur = map->header.next;
563 	}
564 
565 	/*
566 	 * Search linearly
567 	 */
568 
569 	while (cur != last) {
570 		if (cur->end > address) {
571 			if (address >= cur->start) {
572 				/*
573 				 * Save this lookup for future hints, and
574 				 * return
575 				 */
576 
577 				*entry = cur;
578 				SAVE_HINT(map, cur);
579 				return (TRUE);
580 			}
581 			break;
582 		}
583 		cur = cur->next;
584 	}
585 	*entry = cur->prev;
586 	SAVE_HINT(map, *entry);
587 	return (FALSE);
588 }
589 
590 /*
591  *	vm_map_insert:
592  *
593  *	Inserts the given whole VM object into the target
594  *	map at the specified address range.  The object's
595  *	size should match that of the address range.
596  *
597  *	Requires that the map be locked, and leaves it so.
598  */
599 int
600 vm_map_insert(map, object, offset, start, end, prot, max, cow)
601 	vm_map_t map;
602 	vm_object_t object;
603 	vm_ooffset_t offset;
604 	vm_offset_t start;
605 	vm_offset_t end;
606 	vm_prot_t prot, max;
607 	int cow;
608 {
609 	register vm_map_entry_t new_entry;
610 	register vm_map_entry_t prev_entry;
611 	vm_map_entry_t temp_entry;
612 	vm_object_t prev_object;
613 	u_char protoeflags;
614 
615 	if ((object != NULL) && (cow & MAP_NOFAULT)) {
616 		panic("vm_map_insert: paradoxical MAP_NOFAULT request");
617 	}
618 
619 	/*
620 	 * Check that the start and end points are not bogus.
621 	 */
622 
623 	if ((start < map->min_offset) || (end > map->max_offset) ||
624 	    (start >= end))
625 		return (KERN_INVALID_ADDRESS);
626 
627 	/*
628 	 * Find the entry prior to the proposed starting address; if it's part
629 	 * of an existing entry, this range is bogus.
630 	 */
631 
632 	if (vm_map_lookup_entry(map, start, &temp_entry))
633 		return (KERN_NO_SPACE);
634 
635 	prev_entry = temp_entry;
636 
637 	/*
638 	 * Assert that the next entry doesn't overlap the end point.
639 	 */
640 
641 	if ((prev_entry->next != &map->header) &&
642 	    (prev_entry->next->start < end))
643 		return (KERN_NO_SPACE);
644 
645 	protoeflags = 0;
646 	if (cow & MAP_COPY_NEEDED)
647 		protoeflags |= MAP_ENTRY_NEEDS_COPY;
648 
649 	if (cow & MAP_COPY_ON_WRITE)
650 		protoeflags |= MAP_ENTRY_COW;
651 
652 	if (cow & MAP_NOFAULT)
653 		protoeflags |= MAP_ENTRY_NOFAULT;
654 
655 	/*
656 	 * See if we can avoid creating a new entry by extending one of our
657 	 * neighbors.  Or at least extend the object.
658 	 */
659 
660 	if ((object == NULL) &&
661 	    (prev_entry != &map->header) &&
662 	    (( prev_entry->eflags & (MAP_ENTRY_IS_A_MAP | MAP_ENTRY_IS_SUB_MAP)) == 0) &&
663 	    (prev_entry->end == start) &&
664 	    (prev_entry->wired_count == 0)) {
665 
666 
667 		if ((protoeflags == prev_entry->eflags) &&
668 		    ((cow & MAP_NOFAULT) ||
669 		     vm_object_coalesce(prev_entry->object.vm_object,
670 					OFF_TO_IDX(prev_entry->offset),
671 					(vm_size_t) (prev_entry->end - prev_entry->start),
672 					(vm_size_t) (end - prev_entry->end)))) {
673 
674 			/*
675 			 * Coalesced the two objects.  Can we extend the
676 			 * previous map entry to include the new range?
677 			 */
678 			if ((prev_entry->inheritance == VM_INHERIT_DEFAULT) &&
679 			    (prev_entry->protection == prot) &&
680 			    (prev_entry->max_protection == max)) {
681 
682 				map->size += (end - prev_entry->end);
683 				prev_entry->end = end;
684 				if ((cow & MAP_NOFAULT) == 0) {
685 					prev_object = prev_entry->object.vm_object;
686 					default_pager_convert_to_swapq(prev_object);
687 				}
688 				return (KERN_SUCCESS);
689 			}
690 			else {
691 				object = prev_entry->object.vm_object;
692 				offset = prev_entry->offset + (prev_entry->end -
693 							       prev_entry->start);
694 
695 				vm_object_reference(object);
696 			}
697 		}
698 	}
699 
700 	/*
701 	 * Create a new entry
702 	 */
703 
704 	new_entry = vm_map_entry_create(map);
705 	new_entry->start = start;
706 	new_entry->end = end;
707 
708 	new_entry->eflags = protoeflags;
709 	new_entry->object.vm_object = object;
710 	new_entry->offset = offset;
711 
712 	if (map->is_main_map) {
713 		new_entry->inheritance = VM_INHERIT_DEFAULT;
714 		new_entry->protection = prot;
715 		new_entry->max_protection = max;
716 		new_entry->wired_count = 0;
717 	}
718 	/*
719 	 * Insert the new entry into the list
720 	 */
721 
722 	vm_map_entry_link(map, prev_entry, new_entry);
723 	map->size += new_entry->end - new_entry->start;
724 
725 	/*
726 	 * Update the free space hint
727 	 */
728 	if ((map->first_free == prev_entry) &&
729 		(prev_entry->end >= new_entry->start))
730 		map->first_free = new_entry;
731 
732 	default_pager_convert_to_swapq(object);
733 	return (KERN_SUCCESS);
734 }
735 
736 /*
737  * Find sufficient space for `length' bytes in the given map, starting at
738  * `start'.  The map must be locked.  Returns 0 on success, 1 on no space.
739  */
740 int
741 vm_map_findspace(map, start, length, addr)
742 	register vm_map_t map;
743 	register vm_offset_t start;
744 	vm_size_t length;
745 	vm_offset_t *addr;
746 {
747 	register vm_map_entry_t entry, next;
748 	register vm_offset_t end;
749 
750 	if (start < map->min_offset)
751 		start = map->min_offset;
752 	if (start > map->max_offset)
753 		return (1);
754 
755 	/*
756 	 * Look for the first possible address; if there's already something
757 	 * at this address, we have to start after it.
758 	 */
759 	if (start == map->min_offset) {
760 		if ((entry = map->first_free) != &map->header)
761 			start = entry->end;
762 	} else {
763 		vm_map_entry_t tmp;
764 
765 		if (vm_map_lookup_entry(map, start, &tmp))
766 			start = tmp->end;
767 		entry = tmp;
768 	}
769 
770 	/*
771 	 * Look through the rest of the map, trying to fit a new region in the
772 	 * gap between existing regions, or after the very last region.
773 	 */
774 	for (;; start = (entry = next)->end) {
775 		/*
776 		 * Find the end of the proposed new region.  Be sure we didn't
777 		 * go beyond the end of the map, or wrap around the address;
778 		 * if so, we lose.  Otherwise, if this is the last entry, or
779 		 * if the proposed new region fits before the next entry, we
780 		 * win.
781 		 */
782 		end = start + length;
783 		if (end > map->max_offset || end < start)
784 			return (1);
785 		next = entry->next;
786 		if (next == &map->header || next->start >= end)
787 			break;
788 	}
789 	SAVE_HINT(map, entry);
790 	*addr = start;
791 	if (map == kernel_map && round_page(start + length) > kernel_vm_end)
792 		pmap_growkernel(round_page(start + length));
793 	return (0);
794 }
795 
796 /*
797  *	vm_map_find finds an unallocated region in the target address
798  *	map with the given length.  The search is defined to be
799  *	first-fit from the specified address; the region found is
800  *	returned in the same parameter.
801  *
802  */
803 int
804 vm_map_find(map, object, offset, addr, length, find_space, prot, max, cow)
805 	vm_map_t map;
806 	vm_object_t object;
807 	vm_ooffset_t offset;
808 	vm_offset_t *addr;	/* IN/OUT */
809 	vm_size_t length;
810 	boolean_t find_space;
811 	vm_prot_t prot, max;
812 	int cow;
813 {
814 	register vm_offset_t start;
815 	int result, s = 0;
816 
817 	start = *addr;
818 
819 	if (map == kmem_map || map == mb_map)
820 		s = splvm();
821 
822 	vm_map_lock(map);
823 	if (find_space) {
824 		if (vm_map_findspace(map, start, length, addr)) {
825 			vm_map_unlock(map);
826 			if (map == kmem_map || map == mb_map)
827 				splx(s);
828 			return (KERN_NO_SPACE);
829 		}
830 		start = *addr;
831 	}
832 	result = vm_map_insert(map, object, offset,
833 		start, start + length, prot, max, cow);
834 	vm_map_unlock(map);
835 
836 	if (map == kmem_map || map == mb_map)
837 		splx(s);
838 
839 	return (result);
840 }
841 
842 /*
843  *	vm_map_simplify_entry:
844  *
845  *	Simplify the given map entry by merging with either neighbor.
846  */
847 void
848 vm_map_simplify_entry(map, entry)
849 	vm_map_t map;
850 	vm_map_entry_t entry;
851 {
852 	vm_map_entry_t next, prev;
853 	vm_size_t prevsize, esize;
854 
855 	if (entry->eflags & (MAP_ENTRY_IS_SUB_MAP|MAP_ENTRY_IS_A_MAP))
856 		return;
857 
858 	prev = entry->prev;
859 	if (prev != &map->header) {
860 		prevsize = prev->end - prev->start;
861 		if ( (prev->end == entry->start) &&
862 		     (prev->object.vm_object == entry->object.vm_object) &&
863 		     (!prev->object.vm_object || (prev->object.vm_object->behavior == entry->object.vm_object->behavior)) &&
864 		     (!prev->object.vm_object ||
865 			(prev->offset + prevsize == entry->offset)) &&
866 		     (prev->eflags == entry->eflags) &&
867 		     (prev->protection == entry->protection) &&
868 		     (prev->max_protection == entry->max_protection) &&
869 		     (prev->inheritance == entry->inheritance) &&
870 		     (prev->wired_count == entry->wired_count)) {
871 			if (map->first_free == prev)
872 				map->first_free = entry;
873 			if (map->hint == prev)
874 				map->hint = entry;
875 			vm_map_entry_unlink(map, prev);
876 			entry->start = prev->start;
877 			entry->offset = prev->offset;
878 			if (prev->object.vm_object)
879 				vm_object_deallocate(prev->object.vm_object);
880 			vm_map_entry_dispose(map, prev);
881 		}
882 	}
883 
884 	next = entry->next;
885 	if (next != &map->header) {
886 		esize = entry->end - entry->start;
887 		if ((entry->end == next->start) &&
888 		    (next->object.vm_object == entry->object.vm_object) &&
889 		    (!next->object.vm_object || (next->object.vm_object->behavior == entry->object.vm_object->behavior)) &&
890 		     (!entry->object.vm_object ||
891 			(entry->offset + esize == next->offset)) &&
892 		    (next->eflags == entry->eflags) &&
893 		    (next->protection == entry->protection) &&
894 		    (next->max_protection == entry->max_protection) &&
895 		    (next->inheritance == entry->inheritance) &&
896 		    (next->wired_count == entry->wired_count)) {
897 			if (map->first_free == next)
898 				map->first_free = entry;
899 			if (map->hint == next)
900 				map->hint = entry;
901 			vm_map_entry_unlink(map, next);
902 			entry->end = next->end;
903 			if (next->object.vm_object)
904 				vm_object_deallocate(next->object.vm_object);
905 			vm_map_entry_dispose(map, next);
906 	        }
907 	}
908 }
909 /*
910  *	vm_map_clip_start:	[ internal use only ]
911  *
912  *	Asserts that the given entry begins at or after
913  *	the specified address; if necessary,
914  *	it splits the entry into two.
915  */
916 #define vm_map_clip_start(map, entry, startaddr) \
917 { \
918 	if (startaddr > entry->start) \
919 		_vm_map_clip_start(map, entry, startaddr); \
920 }
921 
922 /*
923  *	This routine is called only when it is known that
924  *	the entry must be split.
925  */
926 static void
927 _vm_map_clip_start(map, entry, start)
928 	register vm_map_t map;
929 	register vm_map_entry_t entry;
930 	register vm_offset_t start;
931 {
932 	register vm_map_entry_t new_entry;
933 
934 	/*
935 	 * Split off the front portion -- note that we must insert the new
936 	 * entry BEFORE this one, so that this entry has the specified
937 	 * starting address.
938 	 */
939 
940 	vm_map_simplify_entry(map, entry);
941 
942 	new_entry = vm_map_entry_create(map);
943 	*new_entry = *entry;
944 
945 	new_entry->end = start;
946 	entry->offset += (start - entry->start);
947 	entry->start = start;
948 
949 	vm_map_entry_link(map, entry->prev, new_entry);
950 
951 	if (entry->eflags & (MAP_ENTRY_IS_A_MAP|MAP_ENTRY_IS_SUB_MAP))
952 		vm_map_reference(new_entry->object.share_map);
953 	else
954 		vm_object_reference(new_entry->object.vm_object);
955 }
956 
957 /*
958  *	vm_map_clip_end:	[ internal use only ]
959  *
960  *	Asserts that the given entry ends at or before
961  *	the specified address; if necessary,
962  *	it splits the entry into two.
963  */
964 
965 #define vm_map_clip_end(map, entry, endaddr) \
966 { \
967 	if (endaddr < entry->end) \
968 		_vm_map_clip_end(map, entry, endaddr); \
969 }
970 
971 /*
972  *	This routine is called only when it is known that
973  *	the entry must be split.
974  */
975 static void
976 _vm_map_clip_end(map, entry, end)
977 	register vm_map_t map;
978 	register vm_map_entry_t entry;
979 	register vm_offset_t end;
980 {
981 	register vm_map_entry_t new_entry;
982 
983 	/*
984 	 * Create a new entry and insert it AFTER the specified entry
985 	 */
986 
987 	new_entry = vm_map_entry_create(map);
988 	*new_entry = *entry;
989 
990 	new_entry->start = entry->end = end;
991 	new_entry->offset += (end - entry->start);
992 
993 	vm_map_entry_link(map, entry, new_entry);
994 
995 	if (entry->eflags & (MAP_ENTRY_IS_A_MAP|MAP_ENTRY_IS_SUB_MAP))
996 		vm_map_reference(new_entry->object.share_map);
997 	else
998 		vm_object_reference(new_entry->object.vm_object);
999 }
1000 
1001 /*
1002  *	VM_MAP_RANGE_CHECK:	[ internal use only ]
1003  *
1004  *	Asserts that the starting and ending region
1005  *	addresses fall within the valid range of the map.
1006  */
1007 #define	VM_MAP_RANGE_CHECK(map, start, end)		\
1008 		{					\
1009 		if (start < vm_map_min(map))		\
1010 			start = vm_map_min(map);	\
1011 		if (end > vm_map_max(map))		\
1012 			end = vm_map_max(map);		\
1013 		if (start > end)			\
1014 			start = end;			\
1015 		}
1016 
1017 /*
1018  *	vm_map_submap:		[ kernel use only ]
1019  *
1020  *	Mark the given range as handled by a subordinate map.
1021  *
1022  *	This range must have been created with vm_map_find,
1023  *	and no other operations may have been performed on this
1024  *	range prior to calling vm_map_submap.
1025  *
1026  *	Only a limited number of operations can be performed
1027  *	within this rage after calling vm_map_submap:
1028  *		vm_fault
1029  *	[Don't try vm_map_copy!]
1030  *
1031  *	To remove a submapping, one must first remove the
1032  *	range from the superior map, and then destroy the
1033  *	submap (if desired).  [Better yet, don't try it.]
1034  */
1035 int
1036 vm_map_submap(map, start, end, submap)
1037 	register vm_map_t map;
1038 	register vm_offset_t start;
1039 	register vm_offset_t end;
1040 	vm_map_t submap;
1041 {
1042 	vm_map_entry_t entry;
1043 	register int result = KERN_INVALID_ARGUMENT;
1044 
1045 	vm_map_lock(map);
1046 
1047 	VM_MAP_RANGE_CHECK(map, start, end);
1048 
1049 	if (vm_map_lookup_entry(map, start, &entry)) {
1050 		vm_map_clip_start(map, entry, start);
1051 	} else
1052 		entry = entry->next;
1053 
1054 	vm_map_clip_end(map, entry, end);
1055 
1056 	if ((entry->start == start) && (entry->end == end) &&
1057 	    ((entry->eflags & (MAP_ENTRY_IS_A_MAP|MAP_ENTRY_COW)) == 0) &&
1058 	    (entry->object.vm_object == NULL)) {
1059 		entry->eflags |= MAP_ENTRY_IS_SUB_MAP;
1060 		vm_map_reference(entry->object.sub_map = submap);
1061 		result = KERN_SUCCESS;
1062 	}
1063 	vm_map_unlock(map);
1064 
1065 	return (result);
1066 }
1067 
1068 /*
1069  *	vm_map_protect:
1070  *
1071  *	Sets the protection of the specified address
1072  *	region in the target map.  If "set_max" is
1073  *	specified, the maximum protection is to be set;
1074  *	otherwise, only the current protection is affected.
1075  */
1076 int
1077 vm_map_protect(map, start, end, new_prot, set_max)
1078 	register vm_map_t map;
1079 	register vm_offset_t start;
1080 	register vm_offset_t end;
1081 	register vm_prot_t new_prot;
1082 	register boolean_t set_max;
1083 {
1084 	register vm_map_entry_t current;
1085 	vm_map_entry_t entry;
1086 
1087 	vm_map_lock(map);
1088 
1089 	VM_MAP_RANGE_CHECK(map, start, end);
1090 
1091 	if (vm_map_lookup_entry(map, start, &entry)) {
1092 		vm_map_clip_start(map, entry, start);
1093 	} else {
1094 		entry = entry->next;
1095 	}
1096 
1097 	/*
1098 	 * Make a first pass to check for protection violations.
1099 	 */
1100 
1101 	current = entry;
1102 	while ((current != &map->header) && (current->start < end)) {
1103 		if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
1104 			vm_map_unlock(map);
1105 			return (KERN_INVALID_ARGUMENT);
1106 		}
1107 		if ((new_prot & current->max_protection) != new_prot) {
1108 			vm_map_unlock(map);
1109 			return (KERN_PROTECTION_FAILURE);
1110 		}
1111 		current = current->next;
1112 	}
1113 
1114 	/*
1115 	 * Go back and fix up protections. [Note that clipping is not
1116 	 * necessary the second time.]
1117 	 */
1118 
1119 	current = entry;
1120 
1121 	while ((current != &map->header) && (current->start < end)) {
1122 		vm_prot_t old_prot;
1123 
1124 		vm_map_clip_end(map, current, end);
1125 
1126 		old_prot = current->protection;
1127 		if (set_max)
1128 			current->protection =
1129 			    (current->max_protection = new_prot) &
1130 			    old_prot;
1131 		else
1132 			current->protection = new_prot;
1133 
1134 		/*
1135 		 * Update physical map if necessary. Worry about copy-on-write
1136 		 * here -- CHECK THIS XXX
1137 		 */
1138 
1139 		if (current->protection != old_prot) {
1140 #define MASK(entry)	(((entry)->eflags & MAP_ENTRY_COW) ? ~VM_PROT_WRITE : \
1141 							VM_PROT_ALL)
1142 #define	max(a,b)	((a) > (b) ? (a) : (b))
1143 
1144 			if (current->eflags & MAP_ENTRY_IS_A_MAP) {
1145 				vm_map_entry_t share_entry;
1146 				vm_offset_t share_end;
1147 
1148 				vm_map_lock(current->object.share_map);
1149 				(void) vm_map_lookup_entry(
1150 				    current->object.share_map,
1151 				    current->offset,
1152 				    &share_entry);
1153 				share_end = current->offset +
1154 				    (current->end - current->start);
1155 				while ((share_entry !=
1156 					&current->object.share_map->header) &&
1157 				    (share_entry->start < share_end)) {
1158 
1159 					pmap_protect(map->pmap,
1160 					    (max(share_entry->start,
1161 						    current->offset) -
1162 						current->offset +
1163 						current->start),
1164 					    min(share_entry->end,
1165 						share_end) -
1166 					    current->offset +
1167 					    current->start,
1168 					    current->protection &
1169 					    MASK(share_entry));
1170 
1171 					share_entry = share_entry->next;
1172 				}
1173 				vm_map_unlock(current->object.share_map);
1174 			} else
1175 				pmap_protect(map->pmap, current->start,
1176 				    current->end,
1177 				    current->protection & MASK(entry));
1178 #undef	max
1179 #undef	MASK
1180 		}
1181 		current = current->next;
1182 	}
1183 
1184 	vm_map_simplify_entry(map, entry);
1185 	vm_map_unlock(map);
1186 	return (KERN_SUCCESS);
1187 }
1188 
1189 /*
1190  *	vm_map_madvise:
1191  *
1192  * 	This routine traverses a processes map handling the madvise
1193  *	system call.
1194  */
1195 void
1196 vm_map_madvise(map, pmap, start, end, advise)
1197 	vm_map_t map;
1198 	pmap_t pmap;
1199 	vm_offset_t start, end;
1200 	int advise;
1201 {
1202 	register vm_map_entry_t current;
1203 	vm_map_entry_t entry;
1204 
1205 	vm_map_lock(map);
1206 
1207 	VM_MAP_RANGE_CHECK(map, start, end);
1208 
1209 	if (vm_map_lookup_entry(map, start, &entry)) {
1210 		vm_map_clip_start(map, entry, start);
1211 	} else
1212 		entry = entry->next;
1213 
1214 	for(current = entry;
1215 		(current != &map->header) && (current->start < end);
1216 		current = current->next) {
1217 		vm_size_t size = current->end - current->start;
1218 
1219 		if (current->eflags & (MAP_ENTRY_IS_A_MAP|MAP_ENTRY_IS_SUB_MAP)) {
1220 			continue;
1221 		}
1222 
1223 		/*
1224 		 * Create an object if needed
1225 		 */
1226 		if (current->object.vm_object == NULL) {
1227 			vm_object_t object;
1228 			object = vm_object_allocate(OBJT_DEFAULT, OFF_TO_IDX(size));
1229 			current->object.vm_object = object;
1230 			current->offset = 0;
1231 		}
1232 
1233 		vm_map_clip_end(map, current, end);
1234 		switch (advise) {
1235 	case MADV_NORMAL:
1236 			current->object.vm_object->behavior = OBJ_NORMAL;
1237 			break;
1238 	case MADV_SEQUENTIAL:
1239 			current->object.vm_object->behavior = OBJ_SEQUENTIAL;
1240 			break;
1241 	case MADV_RANDOM:
1242 			current->object.vm_object->behavior = OBJ_RANDOM;
1243 			break;
1244 	/*
1245 	 * Right now, we could handle DONTNEED and WILLNEED with common code.
1246 	 * They are mostly the same, except for the potential async reads (NYI).
1247 	 */
1248 	case MADV_FREE:
1249 	case MADV_DONTNEED:
1250 			{
1251 				vm_pindex_t pindex;
1252 				int count;
1253 				size = current->end - current->start;
1254 				pindex = OFF_TO_IDX(entry->offset);
1255 				count = OFF_TO_IDX(size);
1256 				/*
1257 				 * MADV_DONTNEED removes the page from all
1258 				 * pmaps, so pmap_remove is not necessary.
1259 				 */
1260 				vm_object_madvise(current->object.vm_object,
1261 					pindex, count, advise);
1262 			}
1263 			break;
1264 
1265 	case MADV_WILLNEED:
1266 			{
1267 				vm_pindex_t pindex;
1268 				int count;
1269 				size = current->end - current->start;
1270 				pindex = OFF_TO_IDX(current->offset);
1271 				count = OFF_TO_IDX(size);
1272 				vm_object_madvise(current->object.vm_object,
1273 					pindex, count, advise);
1274 				pmap_object_init_pt(pmap, current->start,
1275 					current->object.vm_object, pindex,
1276 					(count << PAGE_SHIFT), 0);
1277 			}
1278 			break;
1279 
1280 	default:
1281 			break;
1282 		}
1283 	}
1284 
1285 	vm_map_simplify_entry(map, entry);
1286 	vm_map_unlock(map);
1287 	return;
1288 }
1289 
1290 
1291 /*
1292  *	vm_map_inherit:
1293  *
1294  *	Sets the inheritance of the specified address
1295  *	range in the target map.  Inheritance
1296  *	affects how the map will be shared with
1297  *	child maps at the time of vm_map_fork.
1298  */
1299 int
1300 vm_map_inherit(map, start, end, new_inheritance)
1301 	register vm_map_t map;
1302 	register vm_offset_t start;
1303 	register vm_offset_t end;
1304 	register vm_inherit_t new_inheritance;
1305 {
1306 	register vm_map_entry_t entry;
1307 	vm_map_entry_t temp_entry;
1308 
1309 	switch (new_inheritance) {
1310 	case VM_INHERIT_NONE:
1311 	case VM_INHERIT_COPY:
1312 	case VM_INHERIT_SHARE:
1313 		break;
1314 	default:
1315 		return (KERN_INVALID_ARGUMENT);
1316 	}
1317 
1318 	vm_map_lock(map);
1319 
1320 	VM_MAP_RANGE_CHECK(map, start, end);
1321 
1322 	if (vm_map_lookup_entry(map, start, &temp_entry)) {
1323 		entry = temp_entry;
1324 		vm_map_clip_start(map, entry, start);
1325 	} else
1326 		entry = temp_entry->next;
1327 
1328 	while ((entry != &map->header) && (entry->start < end)) {
1329 		vm_map_clip_end(map, entry, end);
1330 
1331 		entry->inheritance = new_inheritance;
1332 
1333 		entry = entry->next;
1334 	}
1335 
1336 	vm_map_simplify_entry(map, temp_entry);
1337 	vm_map_unlock(map);
1338 	return (KERN_SUCCESS);
1339 }
1340 
1341 /*
1342  * Implement the semantics of mlock
1343  */
1344 int
1345 vm_map_user_pageable(map, start, end, new_pageable)
1346 	register vm_map_t map;
1347 	register vm_offset_t start;
1348 	register vm_offset_t end;
1349 	register boolean_t new_pageable;
1350 {
1351 	register vm_map_entry_t entry;
1352 	vm_map_entry_t start_entry;
1353 	register vm_offset_t failed = 0;
1354 	int rv;
1355 
1356 	vm_map_lock(map);
1357 	VM_MAP_RANGE_CHECK(map, start, end);
1358 
1359 	if (vm_map_lookup_entry(map, start, &start_entry) == FALSE) {
1360 		vm_map_unlock(map);
1361 		return (KERN_INVALID_ADDRESS);
1362 	}
1363 
1364 	if (new_pageable) {
1365 
1366 		entry = start_entry;
1367 		vm_map_clip_start(map, entry, start);
1368 
1369 		/*
1370 		 * Now decrement the wiring count for each region. If a region
1371 		 * becomes completely unwired, unwire its physical pages and
1372 		 * mappings.
1373 		 */
1374 		vm_map_set_recursive(map);
1375 
1376 		entry = start_entry;
1377 		while ((entry != &map->header) && (entry->start < end)) {
1378 			if (entry->eflags & MAP_ENTRY_USER_WIRED) {
1379 				vm_map_clip_end(map, entry, end);
1380 				entry->eflags &= ~MAP_ENTRY_USER_WIRED;
1381 				entry->wired_count--;
1382 				if (entry->wired_count == 0)
1383 					vm_fault_unwire(map, entry->start, entry->end);
1384 			}
1385 			entry = entry->next;
1386 		}
1387 		vm_map_simplify_entry(map, start_entry);
1388 		vm_map_clear_recursive(map);
1389 	} else {
1390 
1391 		/*
1392 		 * Because of the possiblity of blocking, etc.  We restart
1393 		 * through the process's map entries from beginning so that
1394 		 * we don't end up depending on a map entry that could have
1395 		 * changed.
1396 		 */
1397 	rescan:
1398 
1399 		entry = start_entry;
1400 
1401 		while ((entry != &map->header) && (entry->start < end)) {
1402 
1403 			if (entry->eflags & MAP_ENTRY_USER_WIRED) {
1404 				entry = entry->next;
1405 				continue;
1406 			}
1407 
1408 			if (entry->wired_count != 0) {
1409 				entry->wired_count++;
1410 				entry->eflags |= MAP_ENTRY_USER_WIRED;
1411 				entry = entry->next;
1412 				continue;
1413 			}
1414 
1415 			/* Here on entry being newly wired */
1416 
1417 			if ((entry->eflags & (MAP_ENTRY_IS_A_MAP|MAP_ENTRY_IS_SUB_MAP)) == 0) {
1418 				int copyflag = entry->eflags & MAP_ENTRY_NEEDS_COPY;
1419 				if (copyflag && ((entry->protection & VM_PROT_WRITE) != 0)) {
1420 
1421 					vm_object_shadow(&entry->object.vm_object,
1422 					    &entry->offset,
1423 					    OFF_TO_IDX(entry->end
1424 						- entry->start));
1425 					entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
1426 
1427 				} else if (entry->object.vm_object == NULL) {
1428 
1429 					entry->object.vm_object =
1430 					    vm_object_allocate(OBJT_DEFAULT,
1431 						OFF_TO_IDX(entry->end - entry->start));
1432 					entry->offset = (vm_offset_t) 0;
1433 
1434 				}
1435 				default_pager_convert_to_swapq(entry->object.vm_object);
1436 			}
1437 
1438 			vm_map_clip_start(map, entry, start);
1439 			vm_map_clip_end(map, entry, end);
1440 
1441 			entry->wired_count++;
1442 			entry->eflags |= MAP_ENTRY_USER_WIRED;
1443 
1444 			/* First we need to allow map modifications */
1445 			vm_map_set_recursive(map);
1446 			if (lockmgr(&map->lock, LK_EXCLUPGRADE,
1447 				(void *)0, curproc)) {
1448 				entry->wired_count--;
1449 				entry->eflags &= ~MAP_ENTRY_USER_WIRED;
1450 
1451 				vm_map_clear_recursive(map);
1452 				vm_map_unlock(map);
1453 
1454 				(void) vm_map_user_pageable(map, start, entry->start, TRUE);
1455 				return rv;
1456 			}
1457 
1458 
1459 			rv = vm_fault_user_wire(map, entry->start, entry->end);
1460 			if (rv) {
1461 
1462 				entry->wired_count--;
1463 				entry->eflags &= ~MAP_ENTRY_USER_WIRED;
1464 
1465 				vm_map_clear_recursive(map);
1466 				vm_map_unlock(map);
1467 
1468 				(void) vm_map_user_pageable(map, start, entry->start, TRUE);
1469 				return rv;
1470 			}
1471 
1472 			vm_map_clear_recursive(map);
1473 			lockmgr(&map->lock, LK_DOWNGRADE, (void *)0, curproc);
1474 
1475 			goto rescan;
1476 		}
1477 	}
1478 	vm_map_unlock(map);
1479 	return KERN_SUCCESS;
1480 }
1481 
1482 /*
1483  *	vm_map_pageable:
1484  *
1485  *	Sets the pageability of the specified address
1486  *	range in the target map.  Regions specified
1487  *	as not pageable require locked-down physical
1488  *	memory and physical page maps.
1489  *
1490  *	The map must not be locked, but a reference
1491  *	must remain to the map throughout the call.
1492  */
1493 int
1494 vm_map_pageable(map, start, end, new_pageable)
1495 	register vm_map_t map;
1496 	register vm_offset_t start;
1497 	register vm_offset_t end;
1498 	register boolean_t new_pageable;
1499 {
1500 	register vm_map_entry_t entry;
1501 	vm_map_entry_t start_entry;
1502 	register vm_offset_t failed = 0;
1503 	int rv;
1504 
1505 	vm_map_lock(map);
1506 
1507 	VM_MAP_RANGE_CHECK(map, start, end);
1508 
1509 	/*
1510 	 * Only one pageability change may take place at one time, since
1511 	 * vm_fault assumes it will be called only once for each
1512 	 * wiring/unwiring.  Therefore, we have to make sure we're actually
1513 	 * changing the pageability for the entire region.  We do so before
1514 	 * making any changes.
1515 	 */
1516 
1517 	if (vm_map_lookup_entry(map, start, &start_entry) == FALSE) {
1518 		vm_map_unlock(map);
1519 		return (KERN_INVALID_ADDRESS);
1520 	}
1521 	entry = start_entry;
1522 
1523 	/*
1524 	 * Actions are rather different for wiring and unwiring, so we have
1525 	 * two separate cases.
1526 	 */
1527 
1528 	if (new_pageable) {
1529 
1530 		vm_map_clip_start(map, entry, start);
1531 
1532 		/*
1533 		 * Unwiring.  First ensure that the range to be unwired is
1534 		 * really wired down and that there are no holes.
1535 		 */
1536 		while ((entry != &map->header) && (entry->start < end)) {
1537 
1538 			if (entry->wired_count == 0 ||
1539 			    (entry->end < end &&
1540 				(entry->next == &map->header ||
1541 				    entry->next->start > entry->end))) {
1542 				vm_map_unlock(map);
1543 				return (KERN_INVALID_ARGUMENT);
1544 			}
1545 			entry = entry->next;
1546 		}
1547 
1548 		/*
1549 		 * Now decrement the wiring count for each region. If a region
1550 		 * becomes completely unwired, unwire its physical pages and
1551 		 * mappings.
1552 		 */
1553 		vm_map_set_recursive(map);
1554 
1555 		entry = start_entry;
1556 		while ((entry != &map->header) && (entry->start < end)) {
1557 			vm_map_clip_end(map, entry, end);
1558 
1559 			entry->wired_count--;
1560 			if (entry->wired_count == 0)
1561 				vm_fault_unwire(map, entry->start, entry->end);
1562 
1563 			entry = entry->next;
1564 		}
1565 		vm_map_simplify_entry(map, start_entry);
1566 		vm_map_clear_recursive(map);
1567 	} else {
1568 		/*
1569 		 * Wiring.  We must do this in two passes:
1570 		 *
1571 		 * 1.  Holding the write lock, we create any shadow or zero-fill
1572 		 * objects that need to be created. Then we clip each map
1573 		 * entry to the region to be wired and increment its wiring
1574 		 * count.  We create objects before clipping the map entries
1575 		 * to avoid object proliferation.
1576 		 *
1577 		 * 2.  We downgrade to a read lock, and call vm_fault_wire to
1578 		 * fault in the pages for any newly wired area (wired_count is
1579 		 * 1).
1580 		 *
1581 		 * Downgrading to a read lock for vm_fault_wire avoids a possible
1582 		 * deadlock with another process that may have faulted on one
1583 		 * of the pages to be wired (it would mark the page busy,
1584 		 * blocking us, then in turn block on the map lock that we
1585 		 * hold).  Because of problems in the recursive lock package,
1586 		 * we cannot upgrade to a write lock in vm_map_lookup.  Thus,
1587 		 * any actions that require the write lock must be done
1588 		 * beforehand.  Because we keep the read lock on the map, the
1589 		 * copy-on-write status of the entries we modify here cannot
1590 		 * change.
1591 		 */
1592 
1593 		/*
1594 		 * Pass 1.
1595 		 */
1596 		while ((entry != &map->header) && (entry->start < end)) {
1597 			if (entry->wired_count == 0) {
1598 
1599 				/*
1600 				 * Perform actions of vm_map_lookup that need
1601 				 * the write lock on the map: create a shadow
1602 				 * object for a copy-on-write region, or an
1603 				 * object for a zero-fill region.
1604 				 *
1605 				 * We don't have to do this for entries that
1606 				 * point to sharing maps, because we won't
1607 				 * hold the lock on the sharing map.
1608 				 */
1609 				if ((entry->eflags & (MAP_ENTRY_IS_A_MAP|MAP_ENTRY_IS_SUB_MAP)) == 0) {
1610 					int copyflag = entry->eflags & MAP_ENTRY_NEEDS_COPY;
1611 					if (copyflag &&
1612 					    ((entry->protection & VM_PROT_WRITE) != 0)) {
1613 
1614 						vm_object_shadow(&entry->object.vm_object,
1615 						    &entry->offset,
1616 						    OFF_TO_IDX(entry->end
1617 							- entry->start));
1618 						entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
1619 					} else if (entry->object.vm_object == NULL) {
1620 						entry->object.vm_object =
1621 						    vm_object_allocate(OBJT_DEFAULT,
1622 							OFF_TO_IDX(entry->end - entry->start));
1623 						entry->offset = (vm_offset_t) 0;
1624 					}
1625 					default_pager_convert_to_swapq(entry->object.vm_object);
1626 				}
1627 			}
1628 			vm_map_clip_start(map, entry, start);
1629 			vm_map_clip_end(map, entry, end);
1630 			entry->wired_count++;
1631 
1632 			/*
1633 			 * Check for holes
1634 			 */
1635 			if (entry->end < end &&
1636 			    (entry->next == &map->header ||
1637 				entry->next->start > entry->end)) {
1638 				/*
1639 				 * Found one.  Object creation actions do not
1640 				 * need to be undone, but the wired counts
1641 				 * need to be restored.
1642 				 */
1643 				while (entry != &map->header && entry->end > start) {
1644 					entry->wired_count--;
1645 					entry = entry->prev;
1646 				}
1647 				vm_map_unlock(map);
1648 				return (KERN_INVALID_ARGUMENT);
1649 			}
1650 			entry = entry->next;
1651 		}
1652 
1653 		/*
1654 		 * Pass 2.
1655 		 */
1656 
1657 		/*
1658 		 * HACK HACK HACK HACK
1659 		 *
1660 		 * If we are wiring in the kernel map or a submap of it,
1661 		 * unlock the map to avoid deadlocks.  We trust that the
1662 		 * kernel is well-behaved, and therefore will not do
1663 		 * anything destructive to this region of the map while
1664 		 * we have it unlocked.  We cannot trust user processes
1665 		 * to do the same.
1666 		 *
1667 		 * HACK HACK HACK HACK
1668 		 */
1669 		if (vm_map_pmap(map) == kernel_pmap) {
1670 			vm_map_unlock(map);	/* trust me ... */
1671 		} else {
1672 			vm_map_set_recursive(map);
1673 			lockmgr(&map->lock, LK_DOWNGRADE, (void*)0, curproc);
1674 		}
1675 
1676 		rv = 0;
1677 		entry = start_entry;
1678 		while (entry != &map->header && entry->start < end) {
1679 			/*
1680 			 * If vm_fault_wire fails for any page we need to undo
1681 			 * what has been done.  We decrement the wiring count
1682 			 * for those pages which have not yet been wired (now)
1683 			 * and unwire those that have (later).
1684 			 *
1685 			 * XXX this violates the locking protocol on the map,
1686 			 * needs to be fixed.
1687 			 */
1688 			if (rv)
1689 				entry->wired_count--;
1690 			else if (entry->wired_count == 1) {
1691 				rv = vm_fault_wire(map, entry->start, entry->end);
1692 				if (rv) {
1693 					failed = entry->start;
1694 					entry->wired_count--;
1695 				}
1696 			}
1697 			entry = entry->next;
1698 		}
1699 
1700 		if (vm_map_pmap(map) == kernel_pmap) {
1701 			vm_map_lock(map);
1702 		} else {
1703 			vm_map_clear_recursive(map);
1704 		}
1705 		if (rv) {
1706 			vm_map_unlock(map);
1707 			(void) vm_map_pageable(map, start, failed, TRUE);
1708 			return (rv);
1709 		}
1710 		vm_map_simplify_entry(map, start_entry);
1711 	}
1712 
1713 	vm_map_unlock(map);
1714 
1715 	return (KERN_SUCCESS);
1716 }
1717 
1718 /*
1719  * vm_map_clean
1720  *
1721  * Push any dirty cached pages in the address range to their pager.
1722  * If syncio is TRUE, dirty pages are written synchronously.
1723  * If invalidate is TRUE, any cached pages are freed as well.
1724  *
1725  * Returns an error if any part of the specified range is not mapped.
1726  */
1727 int
1728 vm_map_clean(map, start, end, syncio, invalidate)
1729 	vm_map_t map;
1730 	vm_offset_t start;
1731 	vm_offset_t end;
1732 	boolean_t syncio;
1733 	boolean_t invalidate;
1734 {
1735 	register vm_map_entry_t current;
1736 	vm_map_entry_t entry;
1737 	vm_size_t size;
1738 	vm_object_t object;
1739 	vm_ooffset_t offset;
1740 
1741 	vm_map_lock_read(map);
1742 	VM_MAP_RANGE_CHECK(map, start, end);
1743 	if (!vm_map_lookup_entry(map, start, &entry)) {
1744 		vm_map_unlock_read(map);
1745 		return (KERN_INVALID_ADDRESS);
1746 	}
1747 	/*
1748 	 * Make a first pass to check for holes.
1749 	 */
1750 	for (current = entry; current->start < end; current = current->next) {
1751 		if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
1752 			vm_map_unlock_read(map);
1753 			return (KERN_INVALID_ARGUMENT);
1754 		}
1755 		if (end > current->end &&
1756 		    (current->next == &map->header ||
1757 			current->end != current->next->start)) {
1758 			vm_map_unlock_read(map);
1759 			return (KERN_INVALID_ADDRESS);
1760 		}
1761 	}
1762 
1763 	/*
1764 	 * Make a second pass, cleaning/uncaching pages from the indicated
1765 	 * objects as we go.
1766 	 */
1767 	for (current = entry; current->start < end; current = current->next) {
1768 		offset = current->offset + (start - current->start);
1769 		size = (end <= current->end ? end : current->end) - start;
1770 		if (current->eflags & (MAP_ENTRY_IS_A_MAP|MAP_ENTRY_IS_SUB_MAP)) {
1771 			register vm_map_t smap;
1772 			vm_map_entry_t tentry;
1773 			vm_size_t tsize;
1774 
1775 			smap = current->object.share_map;
1776 			vm_map_lock_read(smap);
1777 			(void) vm_map_lookup_entry(smap, offset, &tentry);
1778 			tsize = tentry->end - offset;
1779 			if (tsize < size)
1780 				size = tsize;
1781 			object = tentry->object.vm_object;
1782 			offset = tentry->offset + (offset - tentry->start);
1783 			vm_map_unlock_read(smap);
1784 		} else {
1785 			object = current->object.vm_object;
1786 		}
1787 		/*
1788 		 * Note that there is absolutely no sense in writing out
1789 		 * anonymous objects, so we track down the vnode object
1790 		 * to write out.
1791 		 * We invalidate (remove) all pages from the address space
1792 		 * anyway, for semantic correctness.
1793 		 */
1794 		while (object->backing_object) {
1795 			object = object->backing_object;
1796 			offset += object->backing_object_offset;
1797 			if (object->size < OFF_TO_IDX( offset + size))
1798 				size = IDX_TO_OFF(object->size) - offset;
1799 		}
1800 		if (invalidate)
1801 			pmap_remove(vm_map_pmap(map), current->start,
1802 				current->start + size);
1803 		if (object && (object->type == OBJT_VNODE)) {
1804 			/*
1805 			 * Flush pages if writing is allowed. XXX should we continue
1806 			 * on an error?
1807 			 *
1808 			 * XXX Doing async I/O and then removing all the pages from
1809 			 *     the object before it completes is probably a very bad
1810 			 *     idea.
1811 			 */
1812 			if (current->protection & VM_PROT_WRITE) {
1813 		   	    	vm_object_page_clean(object,
1814 					OFF_TO_IDX(offset),
1815 					OFF_TO_IDX(offset + size),
1816 					(syncio||invalidate)?1:0, TRUE);
1817 				if (invalidate)
1818 					vm_object_page_remove(object,
1819 						OFF_TO_IDX(offset),
1820 						OFF_TO_IDX(offset + size),
1821 						FALSE);
1822 			}
1823 		}
1824 		start += size;
1825 	}
1826 
1827 	vm_map_unlock_read(map);
1828 	return (KERN_SUCCESS);
1829 }
1830 
1831 /*
1832  *	vm_map_entry_unwire:	[ internal use only ]
1833  *
1834  *	Make the region specified by this entry pageable.
1835  *
1836  *	The map in question should be locked.
1837  *	[This is the reason for this routine's existence.]
1838  */
1839 static void
1840 vm_map_entry_unwire(map, entry)
1841 	vm_map_t map;
1842 	register vm_map_entry_t entry;
1843 {
1844 	vm_fault_unwire(map, entry->start, entry->end);
1845 	entry->wired_count = 0;
1846 }
1847 
1848 /*
1849  *	vm_map_entry_delete:	[ internal use only ]
1850  *
1851  *	Deallocate the given entry from the target map.
1852  */
1853 static void
1854 vm_map_entry_delete(map, entry)
1855 	register vm_map_t map;
1856 	register vm_map_entry_t entry;
1857 {
1858 	vm_map_entry_unlink(map, entry);
1859 	map->size -= entry->end - entry->start;
1860 
1861 	if (entry->eflags & (MAP_ENTRY_IS_A_MAP|MAP_ENTRY_IS_SUB_MAP)) {
1862 		vm_map_deallocate(entry->object.share_map);
1863 	} else {
1864 		vm_object_deallocate(entry->object.vm_object);
1865 	}
1866 
1867 	vm_map_entry_dispose(map, entry);
1868 }
1869 
1870 /*
1871  *	vm_map_delete:	[ internal use only ]
1872  *
1873  *	Deallocates the given address range from the target
1874  *	map.
1875  *
1876  *	When called with a sharing map, removes pages from
1877  *	that region from all physical maps.
1878  */
1879 int
1880 vm_map_delete(map, start, end)
1881 	register vm_map_t map;
1882 	vm_offset_t start;
1883 	register vm_offset_t end;
1884 {
1885 	register vm_map_entry_t entry;
1886 	vm_map_entry_t first_entry;
1887 
1888 	/*
1889 	 * Find the start of the region, and clip it
1890 	 */
1891 
1892 	if (!vm_map_lookup_entry(map, start, &first_entry))
1893 		entry = first_entry->next;
1894 	else {
1895 		entry = first_entry;
1896 		vm_map_clip_start(map, entry, start);
1897 
1898 		/*
1899 		 * Fix the lookup hint now, rather than each time though the
1900 		 * loop.
1901 		 */
1902 
1903 		SAVE_HINT(map, entry->prev);
1904 	}
1905 
1906 	/*
1907 	 * Save the free space hint
1908 	 */
1909 
1910 	if (entry == &map->header) {
1911 		map->first_free = &map->header;
1912 	} else if (map->first_free->start >= start)
1913 		map->first_free = entry->prev;
1914 
1915 	/*
1916 	 * Step through all entries in this region
1917 	 */
1918 
1919 	while ((entry != &map->header) && (entry->start < end)) {
1920 		vm_map_entry_t next;
1921 		vm_offset_t s, e;
1922 		vm_object_t object;
1923 		vm_ooffset_t offset;
1924 
1925 		vm_map_clip_end(map, entry, end);
1926 
1927 		next = entry->next;
1928 		s = entry->start;
1929 		e = entry->end;
1930 		offset = entry->offset;
1931 
1932 		/*
1933 		 * Unwire before removing addresses from the pmap; otherwise,
1934 		 * unwiring will put the entries back in the pmap.
1935 		 */
1936 
1937 		object = entry->object.vm_object;
1938 		if (entry->wired_count != 0)
1939 			vm_map_entry_unwire(map, entry);
1940 
1941 		/*
1942 		 * If this is a sharing map, we must remove *all* references
1943 		 * to this data, since we can't find all of the physical maps
1944 		 * which are sharing it.
1945 		 */
1946 
1947 		if (object == kernel_object || object == kmem_object) {
1948 			vm_object_page_remove(object, OFF_TO_IDX(offset),
1949 			    OFF_TO_IDX(offset + (e - s)), FALSE);
1950 		} else if (!map->is_main_map) {
1951 			vm_object_pmap_remove(object,
1952 			    OFF_TO_IDX(offset),
1953 			    OFF_TO_IDX(offset + (e - s)));
1954 		} else {
1955 			pmap_remove(map->pmap, s, e);
1956 		}
1957 
1958 		/*
1959 		 * Delete the entry (which may delete the object) only after
1960 		 * removing all pmap entries pointing to its pages.
1961 		 * (Otherwise, its page frames may be reallocated, and any
1962 		 * modify bits will be set in the wrong object!)
1963 		 */
1964 
1965 		vm_map_entry_delete(map, entry);
1966 		entry = next;
1967 	}
1968 	return (KERN_SUCCESS);
1969 }
1970 
1971 /*
1972  *	vm_map_remove:
1973  *
1974  *	Remove the given address range from the target map.
1975  *	This is the exported form of vm_map_delete.
1976  */
1977 int
1978 vm_map_remove(map, start, end)
1979 	register vm_map_t map;
1980 	register vm_offset_t start;
1981 	register vm_offset_t end;
1982 {
1983 	register int result, s = 0;
1984 
1985 	if (map == kmem_map || map == mb_map)
1986 		s = splvm();
1987 
1988 	vm_map_lock(map);
1989 	VM_MAP_RANGE_CHECK(map, start, end);
1990 	result = vm_map_delete(map, start, end);
1991 	vm_map_unlock(map);
1992 
1993 	if (map == kmem_map || map == mb_map)
1994 		splx(s);
1995 
1996 	return (result);
1997 }
1998 
1999 /*
2000  *	vm_map_check_protection:
2001  *
2002  *	Assert that the target map allows the specified
2003  *	privilege on the entire address region given.
2004  *	The entire region must be allocated.
2005  */
2006 boolean_t
2007 vm_map_check_protection(map, start, end, protection)
2008 	register vm_map_t map;
2009 	register vm_offset_t start;
2010 	register vm_offset_t end;
2011 	register vm_prot_t protection;
2012 {
2013 	register vm_map_entry_t entry;
2014 	vm_map_entry_t tmp_entry;
2015 
2016 	if (!vm_map_lookup_entry(map, start, &tmp_entry)) {
2017 		return (FALSE);
2018 	}
2019 	entry = tmp_entry;
2020 
2021 	while (start < end) {
2022 		if (entry == &map->header) {
2023 			return (FALSE);
2024 		}
2025 		/*
2026 		 * No holes allowed!
2027 		 */
2028 
2029 		if (start < entry->start) {
2030 			return (FALSE);
2031 		}
2032 		/*
2033 		 * Check protection associated with entry.
2034 		 */
2035 
2036 		if ((entry->protection & protection) != protection) {
2037 			return (FALSE);
2038 		}
2039 		/* go to next entry */
2040 
2041 		start = entry->end;
2042 		entry = entry->next;
2043 	}
2044 	return (TRUE);
2045 }
2046 
2047 /*
2048  *	vm_map_copy_entry:
2049  *
2050  *	Copies the contents of the source entry to the destination
2051  *	entry.  The entries *must* be aligned properly.
2052  */
2053 static void
2054 vm_map_copy_entry(src_map, dst_map, src_entry, dst_entry)
2055 	vm_map_t src_map, dst_map;
2056 	register vm_map_entry_t src_entry, dst_entry;
2057 {
2058 	if ((dst_entry->eflags|src_entry->eflags) &
2059 		(MAP_ENTRY_IS_A_MAP|MAP_ENTRY_IS_SUB_MAP))
2060 		return;
2061 
2062 	if (src_entry->wired_count == 0) {
2063 
2064 		/*
2065 		 * If the source entry is marked needs_copy, it is already
2066 		 * write-protected.
2067 		 */
2068 		if ((src_entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) {
2069 
2070 			boolean_t su;
2071 
2072 			/*
2073 			 * If the source entry has only one mapping, we can
2074 			 * just protect the virtual address range.
2075 			 */
2076 			if (!(su = src_map->is_main_map)) {
2077 				su = (src_map->ref_count == 1);
2078 			}
2079 			if (su) {
2080 				pmap_protect(src_map->pmap,
2081 				    src_entry->start,
2082 				    src_entry->end,
2083 				    src_entry->protection & ~VM_PROT_WRITE);
2084 			} else {
2085 				vm_object_pmap_copy(src_entry->object.vm_object,
2086 				    OFF_TO_IDX(src_entry->offset),
2087 				    OFF_TO_IDX(src_entry->offset + (src_entry->end
2088 					- src_entry->start)));
2089 			}
2090 		}
2091 
2092 		/*
2093 		 * Make a copy of the object.
2094 		 */
2095 		if (src_entry->object.vm_object) {
2096 			if ((src_entry->object.vm_object->handle == NULL) &&
2097 				(src_entry->object.vm_object->type == OBJT_DEFAULT ||
2098 				 src_entry->object.vm_object->type == OBJT_SWAP))
2099 				vm_object_collapse(src_entry->object.vm_object);
2100 			++src_entry->object.vm_object->ref_count;
2101 			src_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY);
2102 			dst_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY);
2103 			dst_entry->object.vm_object =
2104 				src_entry->object.vm_object;
2105 			dst_entry->offset = src_entry->offset;
2106 		} else {
2107 			dst_entry->object.vm_object = NULL;
2108 			dst_entry->offset = 0;
2109 		}
2110 
2111 		pmap_copy(dst_map->pmap, src_map->pmap, dst_entry->start,
2112 		    dst_entry->end - dst_entry->start, src_entry->start);
2113 	} else {
2114 		/*
2115 		 * Of course, wired down pages can't be set copy-on-write.
2116 		 * Cause wired pages to be copied into the new map by
2117 		 * simulating faults (the new pages are pageable)
2118 		 */
2119 		vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry);
2120 	}
2121 }
2122 
2123 /*
2124  * vmspace_fork:
2125  * Create a new process vmspace structure and vm_map
2126  * based on those of an existing process.  The new map
2127  * is based on the old map, according to the inheritance
2128  * values on the regions in that map.
2129  *
2130  * The source map must not be locked.
2131  */
2132 struct vmspace *
2133 vmspace_fork(vm1)
2134 	register struct vmspace *vm1;
2135 {
2136 	register struct vmspace *vm2;
2137 	vm_map_t old_map = &vm1->vm_map;
2138 	vm_map_t new_map;
2139 	vm_map_entry_t old_entry;
2140 	vm_map_entry_t new_entry;
2141 	pmap_t new_pmap;
2142 	vm_object_t object;
2143 
2144 	vm_map_lock(old_map);
2145 
2146 	vm2 = vmspace_alloc(old_map->min_offset, old_map->max_offset,
2147 	    old_map->entries_pageable);
2148 	bcopy(&vm1->vm_startcopy, &vm2->vm_startcopy,
2149 	    (caddr_t) (vm1 + 1) - (caddr_t) &vm1->vm_startcopy);
2150 	new_pmap = &vm2->vm_pmap;	/* XXX */
2151 	new_map = &vm2->vm_map;	/* XXX */
2152 
2153 	old_entry = old_map->header.next;
2154 
2155 	while (old_entry != &old_map->header) {
2156 		if (old_entry->eflags & MAP_ENTRY_IS_SUB_MAP)
2157 			panic("vm_map_fork: encountered a submap");
2158 
2159 		switch (old_entry->inheritance) {
2160 		case VM_INHERIT_NONE:
2161 			break;
2162 
2163                 case VM_INHERIT_SHARE:
2164                         /*
2165                          * Clone the entry, creating the shared object if necessary.
2166                          */
2167                         object = old_entry->object.vm_object;
2168                         if (object == NULL) {
2169                                 object = vm_object_allocate(OBJT_DEFAULT,
2170                                                             OFF_TO_IDX(old_entry->end -
2171                                                                        old_entry->start));
2172                                 old_entry->object.vm_object = object;
2173                                 old_entry->offset = (vm_offset_t) 0;
2174                         } else if (old_entry->eflags & MAP_ENTRY_NEEDS_COPY) {
2175 				vm_object_shadow(&old_entry->object.vm_object,
2176 						 &old_entry->offset,
2177 						 OFF_TO_IDX(old_entry->end -
2178 							old_entry->start));
2179 
2180 				old_entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
2181 				object = old_entry->object.vm_object;
2182 			}
2183 
2184 			/*
2185 			 * Clone the entry, referencing the sharing map.
2186 			 */
2187 			new_entry = vm_map_entry_create(new_map);
2188 			*new_entry = *old_entry;
2189 			new_entry->wired_count = 0;
2190 			++object->ref_count;
2191 
2192 			/*
2193 			 * Insert the entry into the new map -- we know we're
2194 			 * inserting at the end of the new map.
2195 			 */
2196 
2197 			vm_map_entry_link(new_map, new_map->header.prev,
2198 			    new_entry);
2199 
2200 			/*
2201 			 * Update the physical map
2202 			 */
2203 
2204 			pmap_copy(new_map->pmap, old_map->pmap,
2205 			    new_entry->start,
2206 			    (old_entry->end - old_entry->start),
2207 			    old_entry->start);
2208 			break;
2209 
2210 		case VM_INHERIT_COPY:
2211 			/*
2212 			 * Clone the entry and link into the map.
2213 			 */
2214 			new_entry = vm_map_entry_create(new_map);
2215 			*new_entry = *old_entry;
2216 			new_entry->wired_count = 0;
2217 			new_entry->object.vm_object = NULL;
2218 			new_entry->eflags &= ~MAP_ENTRY_IS_A_MAP;
2219 			vm_map_entry_link(new_map, new_map->header.prev,
2220 			    new_entry);
2221 			vm_map_copy_entry(old_map, new_map, old_entry,
2222 			    new_entry);
2223 			break;
2224 		}
2225 		old_entry = old_entry->next;
2226 	}
2227 
2228 	new_map->size = old_map->size;
2229 	vm_map_unlock(old_map);
2230 
2231 	return (vm2);
2232 }
2233 
2234 /*
2235  *	vm_map_lookup:
2236  *
2237  *	Finds the VM object, offset, and
2238  *	protection for a given virtual address in the
2239  *	specified map, assuming a page fault of the
2240  *	type specified.
2241  *
2242  *	Leaves the map in question locked for read; return
2243  *	values are guaranteed until a vm_map_lookup_done
2244  *	call is performed.  Note that the map argument
2245  *	is in/out; the returned map must be used in
2246  *	the call to vm_map_lookup_done.
2247  *
2248  *	A handle (out_entry) is returned for use in
2249  *	vm_map_lookup_done, to make that fast.
2250  *
2251  *	If a lookup is requested with "write protection"
2252  *	specified, the map may be changed to perform virtual
2253  *	copying operations, although the data referenced will
2254  *	remain the same.
2255  */
2256 int
2257 vm_map_lookup(var_map, vaddr, fault_type, out_entry,
2258     object, pindex, out_prot, wired, single_use)
2259 	vm_map_t *var_map;	/* IN/OUT */
2260 	register vm_offset_t vaddr;
2261 	register vm_prot_t fault_type;
2262 
2263 	vm_map_entry_t *out_entry;	/* OUT */
2264 	vm_object_t *object;	/* OUT */
2265 	vm_pindex_t *pindex;	/* OUT */
2266 	vm_prot_t *out_prot;	/* OUT */
2267 	boolean_t *wired;	/* OUT */
2268 	boolean_t *single_use;	/* OUT */
2269 {
2270 	vm_map_t share_map;
2271 	vm_offset_t share_offset;
2272 	register vm_map_entry_t entry;
2273 	register vm_map_t map = *var_map;
2274 	register vm_prot_t prot;
2275 	register boolean_t su;
2276 
2277 RetryLookup:;
2278 
2279 	/*
2280 	 * Lookup the faulting address.
2281 	 */
2282 
2283 	vm_map_lock_read(map);
2284 
2285 #define	RETURN(why) \
2286 		{ \
2287 		vm_map_unlock_read(map); \
2288 		return(why); \
2289 		}
2290 
2291 	/*
2292 	 * If the map has an interesting hint, try it before calling full
2293 	 * blown lookup routine.
2294 	 */
2295 
2296 	entry = map->hint;
2297 
2298 	*out_entry = entry;
2299 
2300 	if ((entry == &map->header) ||
2301 	    (vaddr < entry->start) || (vaddr >= entry->end)) {
2302 		vm_map_entry_t tmp_entry;
2303 
2304 		/*
2305 		 * Entry was either not a valid hint, or the vaddr was not
2306 		 * contained in the entry, so do a full lookup.
2307 		 */
2308 		if (!vm_map_lookup_entry(map, vaddr, &tmp_entry))
2309 			RETURN(KERN_INVALID_ADDRESS);
2310 
2311 		entry = tmp_entry;
2312 		*out_entry = entry;
2313 	}
2314 
2315 	/*
2316 	 * Handle submaps.
2317 	 */
2318 
2319 	if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
2320 		vm_map_t old_map = map;
2321 
2322 		*var_map = map = entry->object.sub_map;
2323 		vm_map_unlock_read(old_map);
2324 		goto RetryLookup;
2325 	}
2326 	/*
2327 	 * Check whether this task is allowed to have this page.
2328 	 */
2329 
2330 	prot = entry->protection;
2331 	if ((fault_type & (prot)) != fault_type)
2332 		RETURN(KERN_PROTECTION_FAILURE);
2333 
2334 	/*
2335 	 * If this page is not pageable, we have to get it for all possible
2336 	 * accesses.
2337 	 */
2338 
2339 	*wired = (entry->wired_count != 0);
2340 	if (*wired)
2341 		prot = fault_type = entry->protection;
2342 
2343 	/*
2344 	 * If we don't already have a VM object, track it down.
2345 	 */
2346 
2347 	su = (entry->eflags & MAP_ENTRY_IS_A_MAP) == 0;
2348 	if (su) {
2349 		share_map = map;
2350 		share_offset = vaddr;
2351 	} else {
2352 		vm_map_entry_t share_entry;
2353 
2354 		/*
2355 		 * Compute the sharing map, and offset into it.
2356 		 */
2357 
2358 		share_map = entry->object.share_map;
2359 		share_offset = (vaddr - entry->start) + entry->offset;
2360 
2361 		/*
2362 		 * Look for the backing store object and offset
2363 		 */
2364 
2365 		vm_map_lock_read(share_map);
2366 
2367 		if (!vm_map_lookup_entry(share_map, share_offset,
2368 			&share_entry)) {
2369 			vm_map_unlock_read(share_map);
2370 			RETURN(KERN_INVALID_ADDRESS);
2371 		}
2372 		entry = share_entry;
2373 	}
2374 
2375 	/*
2376 	 * If the entry was copy-on-write, we either ...
2377 	 */
2378 
2379 	if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
2380 		/*
2381 		 * If we want to write the page, we may as well handle that
2382 		 * now since we've got the sharing map locked.
2383 		 *
2384 		 * If we don't need to write the page, we just demote the
2385 		 * permissions allowed.
2386 		 */
2387 
2388 		if (fault_type & VM_PROT_WRITE) {
2389 			/*
2390 			 * Make a new object, and place it in the object
2391 			 * chain.  Note that no new references have appeared
2392 			 * -- one just moved from the share map to the new
2393 			 * object.
2394 			 */
2395 
2396 			if (lockmgr(&share_map->lock, LK_EXCLUPGRADE,
2397 					(void *)0, curproc)) {
2398 				if (share_map != map)
2399 					vm_map_unlock_read(map);
2400 				goto RetryLookup;
2401 			}
2402 			vm_object_shadow(
2403 			    &entry->object.vm_object,
2404 			    &entry->offset,
2405 			    OFF_TO_IDX(entry->end - entry->start));
2406 
2407 			entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
2408 
2409 			lockmgr(&share_map->lock, LK_DOWNGRADE,
2410 				(void *)0, curproc);
2411 		} else {
2412 			/*
2413 			 * We're attempting to read a copy-on-write page --
2414 			 * don't allow writes.
2415 			 */
2416 
2417 			prot &= (~VM_PROT_WRITE);
2418 		}
2419 	}
2420 	/*
2421 	 * Create an object if necessary.
2422 	 */
2423 	if (entry->object.vm_object == NULL) {
2424 
2425 		if (lockmgr(&share_map->lock, LK_EXCLUPGRADE,
2426 				(void *)0, curproc)) {
2427 			if (share_map != map)
2428 				vm_map_unlock_read(map);
2429 			goto RetryLookup;
2430 		}
2431 		entry->object.vm_object = vm_object_allocate(OBJT_DEFAULT,
2432 		    OFF_TO_IDX(entry->end - entry->start));
2433 		entry->offset = 0;
2434 		lockmgr(&share_map->lock, LK_DOWNGRADE, (void *)0, curproc);
2435 	}
2436 
2437 	if (entry->object.vm_object != NULL)
2438 		default_pager_convert_to_swapq(entry->object.vm_object);
2439 	/*
2440 	 * Return the object/offset from this entry.  If the entry was
2441 	 * copy-on-write or empty, it has been fixed up.
2442 	 */
2443 
2444 	*pindex = OFF_TO_IDX((share_offset - entry->start) + entry->offset);
2445 	*object = entry->object.vm_object;
2446 
2447 	/*
2448 	 * Return whether this is the only map sharing this data.
2449 	 */
2450 
2451 	if (!su) {
2452 		su = (share_map->ref_count == 1);
2453 	}
2454 	*out_prot = prot;
2455 	*single_use = su;
2456 
2457 	return (KERN_SUCCESS);
2458 
2459 #undef	RETURN
2460 }
2461 
2462 /*
2463  *	vm_map_lookup_done:
2464  *
2465  *	Releases locks acquired by a vm_map_lookup
2466  *	(according to the handle returned by that lookup).
2467  */
2468 
2469 void
2470 vm_map_lookup_done(map, entry)
2471 	register vm_map_t map;
2472 	vm_map_entry_t entry;
2473 {
2474 	/*
2475 	 * If this entry references a map, unlock it first.
2476 	 */
2477 
2478 	if (entry->eflags & MAP_ENTRY_IS_A_MAP)
2479 		vm_map_unlock_read(entry->object.share_map);
2480 
2481 	/*
2482 	 * Unlock the main-level map
2483 	 */
2484 
2485 	vm_map_unlock_read(map);
2486 }
2487 
2488 #include "opt_ddb.h"
2489 #ifdef DDB
2490 #include <sys/kernel.h>
2491 
2492 #include <ddb/ddb.h>
2493 
2494 /*
2495  *	vm_map_print:	[ debug ]
2496  */
2497 DB_SHOW_COMMAND(map, vm_map_print)
2498 {
2499 	/* XXX convert args. */
2500 	register vm_map_t map = (vm_map_t)addr;
2501 	boolean_t full = have_addr;
2502 
2503 	register vm_map_entry_t entry;
2504 
2505 	db_iprintf("%s map 0x%x: pmap=0x%x,ref=%d,nentries=%d,version=%d\n",
2506 	    (map->is_main_map ? "Task" : "Share"),
2507 	    (int) map, (int) (map->pmap), map->ref_count, map->nentries,
2508 	    map->timestamp);
2509 
2510 	if (!full && db_indent)
2511 		return;
2512 
2513 	db_indent += 2;
2514 	for (entry = map->header.next; entry != &map->header;
2515 	    entry = entry->next) {
2516 		db_iprintf("map entry 0x%x: start=0x%x, end=0x%x, ",
2517 		    (int) entry, (int) entry->start, (int) entry->end);
2518 		if (map->is_main_map) {
2519 			static char *inheritance_name[4] =
2520 			{"share", "copy", "none", "donate_copy"};
2521 
2522 			db_printf("prot=%x/%x/%s, ",
2523 			    entry->protection,
2524 			    entry->max_protection,
2525 			    inheritance_name[entry->inheritance]);
2526 			if (entry->wired_count != 0)
2527 				db_printf("wired, ");
2528 		}
2529 		if (entry->eflags & (MAP_ENTRY_IS_A_MAP|MAP_ENTRY_IS_SUB_MAP)) {
2530 			db_printf("share=0x%x, offset=0x%x\n",
2531 			    (int) entry->object.share_map,
2532 			    (int) entry->offset);
2533 			if ((entry->prev == &map->header) ||
2534 			    ((entry->prev->eflags & MAP_ENTRY_IS_A_MAP) == 0) ||
2535 			    (entry->prev->object.share_map !=
2536 				entry->object.share_map)) {
2537 				db_indent += 2;
2538 				vm_map_print((int)entry->object.share_map,
2539 					     full, 0, (char *)0);
2540 				db_indent -= 2;
2541 			}
2542 		} else {
2543 			db_printf("object=0x%x, offset=0x%x",
2544 			    (int) entry->object.vm_object,
2545 			    (int) entry->offset);
2546 			if (entry->eflags & MAP_ENTRY_COW)
2547 				db_printf(", copy (%s)",
2548 				    (entry->eflags & MAP_ENTRY_NEEDS_COPY) ? "needed" : "done");
2549 			db_printf("\n");
2550 
2551 			if ((entry->prev == &map->header) ||
2552 			    (entry->prev->eflags & MAP_ENTRY_IS_A_MAP) ||
2553 			    (entry->prev->object.vm_object !=
2554 				entry->object.vm_object)) {
2555 				db_indent += 2;
2556 				vm_object_print((int)entry->object.vm_object,
2557 						full, 0, (char *)0);
2558 				db_indent -= 2;
2559 			}
2560 		}
2561 	}
2562 	db_indent -= 2;
2563 }
2564 #endif /* DDB */
2565