xref: /freebsd/sys/vm/vm_map.c (revision c17d43407fe04133a94055b0dbc7ea8965654a9f)
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  * $FreeBSD$
65  */
66 
67 /*
68  *	Virtual memory mapping module.
69  */
70 
71 #include <sys/param.h>
72 #include <sys/systm.h>
73 #include <sys/ktr.h>
74 #include <sys/lock.h>
75 #include <sys/mutex.h>
76 #include <sys/proc.h>
77 #include <sys/vmmeter.h>
78 #include <sys/mman.h>
79 #include <sys/vnode.h>
80 #include <sys/resourcevar.h>
81 
82 #include <vm/vm.h>
83 #include <vm/vm_param.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_pager.h>
89 #include <vm/vm_kern.h>
90 #include <vm/vm_extern.h>
91 #include <vm/swap_pager.h>
92 #include <vm/uma.h>
93 
94 /*
95  *	Virtual memory maps provide for the mapping, protection,
96  *	and sharing of virtual memory objects.  In addition,
97  *	this module provides for an efficient virtual copy of
98  *	memory from one map to another.
99  *
100  *	Synchronization is required prior to most operations.
101  *
102  *	Maps consist of an ordered doubly-linked list of simple
103  *	entries; a single hint is used to speed up lookups.
104  *
105  *	Since portions of maps are specified by start/end addresses,
106  *	which may not align with existing map entries, all
107  *	routines merely "clip" entries to these start/end values.
108  *	[That is, an entry is split into two, bordering at a
109  *	start or end value.]  Note that these clippings may not
110  *	always be necessary (as the two resulting entries are then
111  *	not changed); however, the clipping is done for convenience.
112  *
113  *	As mentioned above, virtual copy operations are performed
114  *	by copying VM object references from one map to
115  *	another, and then marking both regions as copy-on-write.
116  */
117 
118 /*
119  *	vm_map_startup:
120  *
121  *	Initialize the vm_map module.  Must be called before
122  *	any other vm_map routines.
123  *
124  *	Map and entry structures are allocated from the general
125  *	purpose memory pool with some exceptions:
126  *
127  *	- The kernel map and kmem submap are allocated statically.
128  *	- Kernel map entries are allocated out of a static pool.
129  *
130  *	These restrictions are necessary since malloc() uses the
131  *	maps and requires map entries.
132  */
133 
134 static uma_zone_t mapentzone;
135 static uma_zone_t kmapentzone;
136 static uma_zone_t mapzone;
137 static uma_zone_t vmspace_zone;
138 static struct vm_object kmapentobj;
139 static void vmspace_zinit(void *mem, int size);
140 static void vmspace_zfini(void *mem, int size);
141 static void vm_map_zinit(void *mem, int size);
142 static void vm_map_zfini(void *mem, int size);
143 static void _vm_map_init(vm_map_t map, vm_offset_t min, vm_offset_t max);
144 
145 #ifdef INVARIANTS
146 static void vm_map_zdtor(void *mem, int size, void *arg);
147 static void vmspace_zdtor(void *mem, int size, void *arg);
148 #endif
149 
150 void
151 vm_map_startup(void)
152 {
153 	mapzone = uma_zcreate("MAP", sizeof(struct vm_map), NULL,
154 #ifdef INVARIANTS
155 	    vm_map_zdtor,
156 #else
157 	    NULL,
158 #endif
159 	    vm_map_zinit, vm_map_zfini, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
160 	uma_prealloc(mapzone, MAX_KMAP);
161 	kmapentzone = uma_zcreate("KMAP ENTRY", sizeof(struct vm_map_entry),
162 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
163 	uma_prealloc(kmapentzone, MAX_KMAPENT);
164 	mapentzone = uma_zcreate("MAP ENTRY", sizeof(struct vm_map_entry),
165 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
166 	uma_prealloc(mapentzone, MAX_MAPENT);
167 }
168 
169 static void
170 vmspace_zfini(void *mem, int size)
171 {
172 	struct vmspace *vm;
173 
174 	vm = (struct vmspace *)mem;
175 
176 	vm_map_zfini(&vm->vm_map, sizeof(vm->vm_map));
177 }
178 
179 static void
180 vmspace_zinit(void *mem, int size)
181 {
182 	struct vmspace *vm;
183 
184 	vm = (struct vmspace *)mem;
185 
186 	vm_map_zinit(&vm->vm_map, sizeof(vm->vm_map));
187 }
188 
189 static void
190 vm_map_zfini(void *mem, int size)
191 {
192 	vm_map_t map;
193 
194 	GIANT_REQUIRED;
195 	map = (vm_map_t)mem;
196 
197 	lockdestroy(&map->lock);
198 }
199 
200 static void
201 vm_map_zinit(void *mem, int size)
202 {
203 	vm_map_t map;
204 
205 	GIANT_REQUIRED;
206 
207 	map = (vm_map_t)mem;
208 	map->nentries = 0;
209 	map->size = 0;
210 	map->infork = 0;
211 	lockinit(&map->lock, PVM, "thrd_sleep", 0, LK_NOPAUSE);
212 }
213 
214 #ifdef INVARIANTS
215 static void
216 vmspace_zdtor(void *mem, int size, void *arg)
217 {
218 	struct vmspace *vm;
219 
220 	vm = (struct vmspace *)mem;
221 
222 	vm_map_zdtor(&vm->vm_map, sizeof(vm->vm_map), arg);
223 }
224 static void
225 vm_map_zdtor(void *mem, int size, void *arg)
226 {
227 	vm_map_t map;
228 
229 	map = (vm_map_t)mem;
230 	KASSERT(map->nentries == 0,
231 	    ("map %p nentries == %d on free.",
232 	    map, map->nentries));
233 	KASSERT(map->size == 0,
234 	    ("map %p size == %lu on free.",
235 	    map, (unsigned long)map->size));
236 	KASSERT(map->infork == 0,
237 	    ("map %p infork == %d on free.",
238 	    map, map->infork));
239 }
240 #endif	/* INVARIANTS */
241 
242 /*
243  * Allocate a vmspace structure, including a vm_map and pmap,
244  * and initialize those structures.  The refcnt is set to 1.
245  * The remaining fields must be initialized by the caller.
246  */
247 struct vmspace *
248 vmspace_alloc(min, max)
249 	vm_offset_t min, max;
250 {
251 	struct vmspace *vm;
252 
253 	GIANT_REQUIRED;
254 	vm = uma_zalloc(vmspace_zone, M_WAITOK);
255 	CTR1(KTR_VM, "vmspace_alloc: %p", vm);
256 	_vm_map_init(&vm->vm_map, min, max);
257 	pmap_pinit(vmspace_pmap(vm));
258 	vm->vm_map.pmap = vmspace_pmap(vm);		/* XXX */
259 	vm->vm_refcnt = 1;
260 	vm->vm_shm = NULL;
261 	vm->vm_freer = NULL;
262 	return (vm);
263 }
264 
265 void
266 vm_init2(void)
267 {
268 	uma_zone_set_obj(kmapentzone, &kmapentobj, cnt.v_page_count / 4);
269 	vmspace_zone = uma_zcreate("VMSPACE", sizeof(struct vmspace), NULL,
270 #ifdef INVARIANTS
271 	    vmspace_zdtor,
272 #else
273 	    NULL,
274 #endif
275 	    vmspace_zinit, vmspace_zfini, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
276 	pmap_init2();
277 	vm_object_init2();
278 }
279 
280 static __inline void
281 vmspace_dofree(struct vmspace *vm)
282 {
283 	CTR1(KTR_VM, "vmspace_free: %p", vm);
284 	/*
285 	 * Lock the map, to wait out all other references to it.
286 	 * Delete all of the mappings and pages they hold, then call
287 	 * the pmap module to reclaim anything left.
288 	 */
289 	vm_map_lock(&vm->vm_map);
290 	(void) vm_map_delete(&vm->vm_map, vm->vm_map.min_offset,
291 	    vm->vm_map.max_offset);
292 	vm_map_unlock(&vm->vm_map);
293 
294 	pmap_release(vmspace_pmap(vm));
295 	uma_zfree(vmspace_zone, vm);
296 }
297 
298 void
299 vmspace_free(struct vmspace *vm)
300 {
301 	GIANT_REQUIRED;
302 
303 	if (vm->vm_refcnt == 0)
304 		panic("vmspace_free: attempt to free already freed vmspace");
305 
306 	if (--vm->vm_refcnt == 0)
307 		vmspace_dofree(vm);
308 }
309 
310 void
311 vmspace_exitfree(struct proc *p)
312 {
313 	GIANT_REQUIRED;
314 
315 	if (p == p->p_vmspace->vm_freer)
316 		vmspace_dofree(p->p_vmspace);
317 }
318 
319 /*
320  * vmspace_swap_count() - count the approximate swap useage in pages for a
321  *			  vmspace.
322  *
323  *	Swap useage is determined by taking the proportional swap used by
324  *	VM objects backing the VM map.  To make up for fractional losses,
325  *	if the VM object has any swap use at all the associated map entries
326  *	count for at least 1 swap page.
327  */
328 int
329 vmspace_swap_count(struct vmspace *vmspace)
330 {
331 	vm_map_t map = &vmspace->vm_map;
332 	vm_map_entry_t cur;
333 	int count = 0;
334 
335 	for (cur = map->header.next; cur != &map->header; cur = cur->next) {
336 		vm_object_t object;
337 
338 		if ((cur->eflags & MAP_ENTRY_IS_SUB_MAP) == 0 &&
339 		    (object = cur->object.vm_object) != NULL &&
340 		    object->type == OBJT_SWAP
341 		) {
342 			int n = (cur->end - cur->start) / PAGE_SIZE;
343 
344 			if (object->un_pager.swp.swp_bcount) {
345 				count += object->un_pager.swp.swp_bcount *
346 				    SWAP_META_PAGES * n / object->size + 1;
347 			}
348 		}
349 	}
350 	return (count);
351 }
352 
353 u_char
354 vm_map_entry_behavior(struct vm_map_entry *entry)
355 {
356 	return entry->eflags & MAP_ENTRY_BEHAV_MASK;
357 }
358 
359 void
360 vm_map_entry_set_behavior(struct vm_map_entry *entry, u_char behavior)
361 {
362 	entry->eflags = (entry->eflags & ~MAP_ENTRY_BEHAV_MASK) |
363 		(behavior & MAP_ENTRY_BEHAV_MASK);
364 }
365 
366 void
367 vm_map_lock(vm_map_t map)
368 {
369 	vm_map_printf("locking map LK_EXCLUSIVE: %p\n", map);
370 	if (lockmgr(&map->lock, LK_EXCLUSIVE, NULL, curthread) != 0)
371 		panic("vm_map_lock: failed to get lock");
372 	map->timestamp++;
373 }
374 
375 void
376 vm_map_unlock(vm_map_t map)
377 {
378 	vm_map_printf("locking map LK_RELEASE: %p\n", map);
379 	lockmgr(&(map)->lock, LK_RELEASE, NULL, curthread);
380 }
381 
382 void
383 vm_map_lock_read(vm_map_t map)
384 {
385 	vm_map_printf("locking map LK_SHARED: %p\n", map);
386 	lockmgr(&(map)->lock, LK_SHARED, NULL, curthread);
387 }
388 
389 void
390 vm_map_unlock_read(vm_map_t map)
391 {
392 	vm_map_printf("locking map LK_RELEASE: %p\n", map);
393 	lockmgr(&(map)->lock, LK_RELEASE, NULL, curthread);
394 }
395 
396 static __inline__ int
397 _vm_map_lock_upgrade(vm_map_t map, struct thread *td) {
398 	int error;
399 
400 	vm_map_printf("locking map LK_EXCLUPGRADE: %p\n", map);
401 	error = lockmgr(&map->lock, LK_EXCLUPGRADE, NULL, td);
402 	if (error == 0)
403 		map->timestamp++;
404 	return error;
405 }
406 
407 int
408 vm_map_lock_upgrade(vm_map_t map)
409 {
410     return (_vm_map_lock_upgrade(map, curthread));
411 }
412 
413 void
414 vm_map_lock_downgrade(vm_map_t map)
415 {
416 	vm_map_printf("locking map LK_DOWNGRADE: %p\n", map);
417 	lockmgr(&map->lock, LK_DOWNGRADE, NULL, curthread);
418 }
419 
420 void
421 vm_map_set_recursive(vm_map_t map)
422 {
423 	mtx_lock((map)->lock.lk_interlock);
424 	map->lock.lk_flags |= LK_CANRECURSE;
425 	mtx_unlock((map)->lock.lk_interlock);
426 }
427 
428 void
429 vm_map_clear_recursive(vm_map_t map)
430 {
431 	mtx_lock((map)->lock.lk_interlock);
432 	map->lock.lk_flags &= ~LK_CANRECURSE;
433 	mtx_unlock((map)->lock.lk_interlock);
434 }
435 
436 vm_offset_t
437 vm_map_min(vm_map_t map)
438 {
439 	return (map->min_offset);
440 }
441 
442 vm_offset_t
443 vm_map_max(vm_map_t map)
444 {
445 	return (map->max_offset);
446 }
447 
448 struct pmap *
449 vm_map_pmap(vm_map_t map)
450 {
451 	return (map->pmap);
452 }
453 
454 struct pmap *
455 vmspace_pmap(struct vmspace *vmspace)
456 {
457 	return &vmspace->vm_pmap;
458 }
459 
460 long
461 vmspace_resident_count(struct vmspace *vmspace)
462 {
463 	return pmap_resident_count(vmspace_pmap(vmspace));
464 }
465 
466 /*
467  *	vm_map_create:
468  *
469  *	Creates and returns a new empty VM map with
470  *	the given physical map structure, and having
471  *	the given lower and upper address bounds.
472  */
473 vm_map_t
474 vm_map_create(pmap_t pmap, vm_offset_t min, vm_offset_t max)
475 {
476 	vm_map_t result;
477 
478 	GIANT_REQUIRED;
479 
480 	result = uma_zalloc(mapzone, M_WAITOK);
481 	CTR1(KTR_VM, "vm_map_create: %p", result);
482 	_vm_map_init(result, min, max);
483 	result->pmap = pmap;
484 	return (result);
485 }
486 
487 /*
488  * Initialize an existing vm_map structure
489  * such as that in the vmspace structure.
490  * The pmap is set elsewhere.
491  */
492 static void
493 _vm_map_init(vm_map_t map, vm_offset_t min, vm_offset_t max)
494 {
495 	GIANT_REQUIRED;
496 
497 	map->header.next = map->header.prev = &map->header;
498 	map->system_map = 0;
499 	map->min_offset = min;
500 	map->max_offset = max;
501 	map->first_free = &map->header;
502 	map->hint = &map->header;
503 	map->timestamp = 0;
504 }
505 
506 void
507 vm_map_init(vm_map_t map, vm_offset_t min, vm_offset_t max)
508 {
509 	_vm_map_init(map, min, max);
510 	lockinit(&map->lock, PVM, "thrd_sleep", 0, LK_NOPAUSE);
511 }
512 
513 /*
514  *	vm_map_entry_dispose:	[ internal use only ]
515  *
516  *	Inverse of vm_map_entry_create.
517  */
518 static void
519 vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry)
520 {
521 	uma_zfree((map->system_map || !mapentzone)
522 	    ? kmapentzone : mapentzone, entry);
523 }
524 
525 /*
526  *	vm_map_entry_create:	[ internal use only ]
527  *
528  *	Allocates a VM map entry for insertion.
529  *	No entry fields are filled in.
530  */
531 static vm_map_entry_t
532 vm_map_entry_create(vm_map_t map)
533 {
534 	vm_map_entry_t new_entry;
535 
536 	new_entry = uma_zalloc((map->system_map || !mapentzone) ?
537 		kmapentzone : mapentzone, M_WAITOK);
538 	if (new_entry == NULL)
539 	    panic("vm_map_entry_create: kernel resources exhausted");
540 	return (new_entry);
541 }
542 
543 /*
544  *	vm_map_entry_{un,}link:
545  *
546  *	Insert/remove entries from maps.
547  */
548 static __inline void
549 vm_map_entry_link(vm_map_t map,
550 		  vm_map_entry_t after_where,
551 		  vm_map_entry_t entry)
552 {
553 
554 	CTR4(KTR_VM,
555 	    "vm_map_entry_link: map %p, nentries %d, entry %p, after %p", map,
556 	    map->nentries, entry, after_where);
557 	map->nentries++;
558 	entry->prev = after_where;
559 	entry->next = after_where->next;
560 	entry->next->prev = entry;
561 	after_where->next = entry;
562 }
563 
564 static __inline void
565 vm_map_entry_unlink(vm_map_t map,
566 		    vm_map_entry_t entry)
567 {
568 	vm_map_entry_t prev = entry->prev;
569 	vm_map_entry_t next = entry->next;
570 
571 	next->prev = prev;
572 	prev->next = next;
573 	map->nentries--;
574 	CTR3(KTR_VM, "vm_map_entry_unlink: map %p, nentries %d, entry %p", map,
575 	    map->nentries, entry);
576 }
577 
578 /*
579  *	SAVE_HINT:
580  *
581  *	Saves the specified entry as the hint for
582  *	future lookups.
583  */
584 #define	SAVE_HINT(map,value) \
585 		(map)->hint = (value);
586 
587 /*
588  *	vm_map_lookup_entry:	[ internal use only ]
589  *
590  *	Finds the map entry containing (or
591  *	immediately preceding) the specified address
592  *	in the given map; the entry is returned
593  *	in the "entry" parameter.  The boolean
594  *	result indicates whether the address is
595  *	actually contained in the map.
596  */
597 boolean_t
598 vm_map_lookup_entry(
599 	vm_map_t map,
600 	vm_offset_t address,
601 	vm_map_entry_t *entry)	/* OUT */
602 {
603 	vm_map_entry_t cur;
604 	vm_map_entry_t last;
605 
606 	GIANT_REQUIRED;
607 	/*
608 	 * Start looking either from the head of the list, or from the hint.
609 	 */
610 	cur = map->hint;
611 
612 	if (cur == &map->header)
613 		cur = cur->next;
614 
615 	if (address >= cur->start) {
616 		/*
617 		 * Go from hint to end of list.
618 		 *
619 		 * But first, make a quick check to see if we are already looking
620 		 * at the entry we want (which is usually the case). Note also
621 		 * that we don't need to save the hint here... it is the same
622 		 * hint (unless we are at the header, in which case the hint
623 		 * didn't buy us anything anyway).
624 		 */
625 		last = &map->header;
626 		if ((cur != last) && (cur->end > address)) {
627 			*entry = cur;
628 			return (TRUE);
629 		}
630 	} else {
631 		/*
632 		 * Go from start to hint, *inclusively*
633 		 */
634 		last = cur->next;
635 		cur = map->header.next;
636 	}
637 
638 	/*
639 	 * Search linearly
640 	 */
641 	while (cur != last) {
642 		if (cur->end > address) {
643 			if (address >= cur->start) {
644 				/*
645 				 * Save this lookup for future hints, and
646 				 * return
647 				 */
648 				*entry = cur;
649 				SAVE_HINT(map, cur);
650 				return (TRUE);
651 			}
652 			break;
653 		}
654 		cur = cur->next;
655 	}
656 	*entry = cur->prev;
657 	SAVE_HINT(map, *entry);
658 	return (FALSE);
659 }
660 
661 /*
662  *	vm_map_insert:
663  *
664  *	Inserts the given whole VM object into the target
665  *	map at the specified address range.  The object's
666  *	size should match that of the address range.
667  *
668  *	Requires that the map be locked, and leaves it so.
669  *
670  *	If object is non-NULL, ref count must be bumped by caller
671  *	prior to making call to account for the new entry.
672  */
673 int
674 vm_map_insert(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
675 	      vm_offset_t start, vm_offset_t end, vm_prot_t prot, vm_prot_t max,
676 	      int cow)
677 {
678 	vm_map_entry_t new_entry;
679 	vm_map_entry_t prev_entry;
680 	vm_map_entry_t temp_entry;
681 	vm_eflags_t protoeflags;
682 
683 	GIANT_REQUIRED;
684 
685 	/*
686 	 * Check that the start and end points are not bogus.
687 	 */
688 	if ((start < map->min_offset) || (end > map->max_offset) ||
689 	    (start >= end))
690 		return (KERN_INVALID_ADDRESS);
691 
692 	/*
693 	 * Find the entry prior to the proposed starting address; if it's part
694 	 * of an existing entry, this range is bogus.
695 	 */
696 	if (vm_map_lookup_entry(map, start, &temp_entry))
697 		return (KERN_NO_SPACE);
698 
699 	prev_entry = temp_entry;
700 
701 	/*
702 	 * Assert that the next entry doesn't overlap the end point.
703 	 */
704 	if ((prev_entry->next != &map->header) &&
705 	    (prev_entry->next->start < end))
706 		return (KERN_NO_SPACE);
707 
708 	protoeflags = 0;
709 
710 	if (cow & MAP_COPY_ON_WRITE)
711 		protoeflags |= MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY;
712 
713 	if (cow & MAP_NOFAULT) {
714 		protoeflags |= MAP_ENTRY_NOFAULT;
715 
716 		KASSERT(object == NULL,
717 			("vm_map_insert: paradoxical MAP_NOFAULT request"));
718 	}
719 	if (cow & MAP_DISABLE_SYNCER)
720 		protoeflags |= MAP_ENTRY_NOSYNC;
721 	if (cow & MAP_DISABLE_COREDUMP)
722 		protoeflags |= MAP_ENTRY_NOCOREDUMP;
723 
724 	if (object) {
725 		/*
726 		 * When object is non-NULL, it could be shared with another
727 		 * process.  We have to set or clear OBJ_ONEMAPPING
728 		 * appropriately.
729 		 */
730 		if ((object->ref_count > 1) || (object->shadow_count != 0)) {
731 			vm_object_clear_flag(object, OBJ_ONEMAPPING);
732 		}
733 	}
734 	else if ((prev_entry != &map->header) &&
735 		 (prev_entry->eflags == protoeflags) &&
736 		 (prev_entry->end == start) &&
737 		 (prev_entry->wired_count == 0) &&
738 		 ((prev_entry->object.vm_object == NULL) ||
739 		  vm_object_coalesce(prev_entry->object.vm_object,
740 				     OFF_TO_IDX(prev_entry->offset),
741 				     (vm_size_t)(prev_entry->end - prev_entry->start),
742 				     (vm_size_t)(end - prev_entry->end)))) {
743 		/*
744 		 * We were able to extend the object.  Determine if we
745 		 * can extend the previous map entry to include the
746 		 * new range as well.
747 		 */
748 		if ((prev_entry->inheritance == VM_INHERIT_DEFAULT) &&
749 		    (prev_entry->protection == prot) &&
750 		    (prev_entry->max_protection == max)) {
751 			map->size += (end - prev_entry->end);
752 			prev_entry->end = end;
753 			vm_map_simplify_entry(map, prev_entry);
754 			return (KERN_SUCCESS);
755 		}
756 
757 		/*
758 		 * If we can extend the object but cannot extend the
759 		 * map entry, we have to create a new map entry.  We
760 		 * must bump the ref count on the extended object to
761 		 * account for it.  object may be NULL.
762 		 */
763 		object = prev_entry->object.vm_object;
764 		offset = prev_entry->offset +
765 			(prev_entry->end - prev_entry->start);
766 		vm_object_reference(object);
767 	}
768 
769 	/*
770 	 * NOTE: if conditionals fail, object can be NULL here.  This occurs
771 	 * in things like the buffer map where we manage kva but do not manage
772 	 * backing objects.
773 	 */
774 
775 	/*
776 	 * Create a new entry
777 	 */
778 	new_entry = vm_map_entry_create(map);
779 	new_entry->start = start;
780 	new_entry->end = end;
781 
782 	new_entry->eflags = protoeflags;
783 	new_entry->object.vm_object = object;
784 	new_entry->offset = offset;
785 	new_entry->avail_ssize = 0;
786 
787 	new_entry->inheritance = VM_INHERIT_DEFAULT;
788 	new_entry->protection = prot;
789 	new_entry->max_protection = max;
790 	new_entry->wired_count = 0;
791 
792 	/*
793 	 * Insert the new entry into the list
794 	 */
795 	vm_map_entry_link(map, prev_entry, new_entry);
796 	map->size += new_entry->end - new_entry->start;
797 
798 	/*
799 	 * Update the free space hint
800 	 */
801 	if ((map->first_free == prev_entry) &&
802 	    (prev_entry->end >= new_entry->start)) {
803 		map->first_free = new_entry;
804 	}
805 
806 #if 0
807 	/*
808 	 * Temporarily removed to avoid MAP_STACK panic, due to
809 	 * MAP_STACK being a huge hack.  Will be added back in
810 	 * when MAP_STACK (and the user stack mapping) is fixed.
811 	 */
812 	/*
813 	 * It may be possible to simplify the entry
814 	 */
815 	vm_map_simplify_entry(map, new_entry);
816 #endif
817 
818 	if (cow & (MAP_PREFAULT|MAP_PREFAULT_PARTIAL)) {
819 		pmap_object_init_pt(map->pmap, start,
820 				    object, OFF_TO_IDX(offset), end - start,
821 				    cow & MAP_PREFAULT_PARTIAL);
822 	}
823 
824 	return (KERN_SUCCESS);
825 }
826 
827 /*
828  * Find sufficient space for `length' bytes in the given map, starting at
829  * `start'.  The map must be locked.  Returns 0 on success, 1 on no space.
830  */
831 int
832 vm_map_findspace(
833 	vm_map_t map,
834 	vm_offset_t start,
835 	vm_size_t length,
836 	vm_offset_t *addr)
837 {
838 	vm_map_entry_t entry, next;
839 	vm_offset_t end;
840 
841 	GIANT_REQUIRED;
842 	if (start < map->min_offset)
843 		start = map->min_offset;
844 	if (start > map->max_offset)
845 		return (1);
846 
847 	/*
848 	 * Look for the first possible address; if there's already something
849 	 * at this address, we have to start after it.
850 	 */
851 	if (start == map->min_offset) {
852 		if ((entry = map->first_free) != &map->header)
853 			start = entry->end;
854 	} else {
855 		vm_map_entry_t tmp;
856 
857 		if (vm_map_lookup_entry(map, start, &tmp))
858 			start = tmp->end;
859 		entry = tmp;
860 	}
861 
862 	/*
863 	 * Look through the rest of the map, trying to fit a new region in the
864 	 * gap between existing regions, or after the very last region.
865 	 */
866 	for (;; start = (entry = next)->end) {
867 		/*
868 		 * Find the end of the proposed new region.  Be sure we didn't
869 		 * go beyond the end of the map, or wrap around the address;
870 		 * if so, we lose.  Otherwise, if this is the last entry, or
871 		 * if the proposed new region fits before the next entry, we
872 		 * win.
873 		 */
874 		end = start + length;
875 		if (end > map->max_offset || end < start)
876 			return (1);
877 		next = entry->next;
878 		if (next == &map->header || next->start >= end)
879 			break;
880 	}
881 	SAVE_HINT(map, entry);
882 	*addr = start;
883 	if (map == kernel_map) {
884 		vm_offset_t ksize;
885 		if ((ksize = round_page(start + length)) > kernel_vm_end) {
886 			pmap_growkernel(ksize);
887 		}
888 	}
889 	return (0);
890 }
891 
892 /*
893  *	vm_map_find finds an unallocated region in the target address
894  *	map with the given length.  The search is defined to be
895  *	first-fit from the specified address; the region found is
896  *	returned in the same parameter.
897  *
898  *	If object is non-NULL, ref count must be bumped by caller
899  *	prior to making call to account for the new entry.
900  */
901 int
902 vm_map_find(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
903 	    vm_offset_t *addr,	/* IN/OUT */
904 	    vm_size_t length, boolean_t find_space, vm_prot_t prot,
905 	    vm_prot_t max, int cow)
906 {
907 	vm_offset_t start;
908 	int result, s = 0;
909 
910 	GIANT_REQUIRED;
911 
912 	start = *addr;
913 
914 	if (map == kmem_map)
915 		s = splvm();
916 
917 	vm_map_lock(map);
918 	if (find_space) {
919 		if (vm_map_findspace(map, start, length, addr)) {
920 			vm_map_unlock(map);
921 			if (map == kmem_map)
922 				splx(s);
923 			return (KERN_NO_SPACE);
924 		}
925 		start = *addr;
926 	}
927 	result = vm_map_insert(map, object, offset,
928 		start, start + length, prot, max, cow);
929 	vm_map_unlock(map);
930 
931 	if (map == kmem_map)
932 		splx(s);
933 
934 	return (result);
935 }
936 
937 /*
938  *	vm_map_simplify_entry:
939  *
940  *	Simplify the given map entry by merging with either neighbor.  This
941  *	routine also has the ability to merge with both neighbors.
942  *
943  *	The map must be locked.
944  *
945  *	This routine guarentees that the passed entry remains valid (though
946  *	possibly extended).  When merging, this routine may delete one or
947  *	both neighbors.
948  */
949 void
950 vm_map_simplify_entry(vm_map_t map, vm_map_entry_t entry)
951 {
952 	vm_map_entry_t next, prev;
953 	vm_size_t prevsize, esize;
954 
955 	GIANT_REQUIRED;
956 
957 	if (entry->eflags & MAP_ENTRY_IS_SUB_MAP)
958 		return;
959 
960 	prev = entry->prev;
961 	if (prev != &map->header) {
962 		prevsize = prev->end - prev->start;
963 		if ( (prev->end == entry->start) &&
964 		     (prev->object.vm_object == entry->object.vm_object) &&
965 		     (!prev->object.vm_object ||
966 			(prev->offset + prevsize == entry->offset)) &&
967 		     (prev->eflags == entry->eflags) &&
968 		     (prev->protection == entry->protection) &&
969 		     (prev->max_protection == entry->max_protection) &&
970 		     (prev->inheritance == entry->inheritance) &&
971 		     (prev->wired_count == entry->wired_count)) {
972 			if (map->first_free == prev)
973 				map->first_free = entry;
974 			if (map->hint == prev)
975 				map->hint = entry;
976 			vm_map_entry_unlink(map, prev);
977 			entry->start = prev->start;
978 			entry->offset = prev->offset;
979 			if (prev->object.vm_object)
980 				vm_object_deallocate(prev->object.vm_object);
981 			vm_map_entry_dispose(map, prev);
982 		}
983 	}
984 
985 	next = entry->next;
986 	if (next != &map->header) {
987 		esize = entry->end - entry->start;
988 		if ((entry->end == next->start) &&
989 		    (next->object.vm_object == entry->object.vm_object) &&
990 		     (!entry->object.vm_object ||
991 			(entry->offset + esize == next->offset)) &&
992 		    (next->eflags == entry->eflags) &&
993 		    (next->protection == entry->protection) &&
994 		    (next->max_protection == entry->max_protection) &&
995 		    (next->inheritance == entry->inheritance) &&
996 		    (next->wired_count == entry->wired_count)) {
997 			if (map->first_free == next)
998 				map->first_free = entry;
999 			if (map->hint == next)
1000 				map->hint = entry;
1001 			vm_map_entry_unlink(map, next);
1002 			entry->end = next->end;
1003 			if (next->object.vm_object)
1004 				vm_object_deallocate(next->object.vm_object);
1005 			vm_map_entry_dispose(map, next);
1006 	        }
1007 	}
1008 }
1009 /*
1010  *	vm_map_clip_start:	[ internal use only ]
1011  *
1012  *	Asserts that the given entry begins at or after
1013  *	the specified address; if necessary,
1014  *	it splits the entry into two.
1015  */
1016 #define vm_map_clip_start(map, entry, startaddr) \
1017 { \
1018 	if (startaddr > entry->start) \
1019 		_vm_map_clip_start(map, entry, startaddr); \
1020 }
1021 
1022 /*
1023  *	This routine is called only when it is known that
1024  *	the entry must be split.
1025  */
1026 static void
1027 _vm_map_clip_start(vm_map_t map, vm_map_entry_t entry, vm_offset_t start)
1028 {
1029 	vm_map_entry_t new_entry;
1030 
1031 	/*
1032 	 * Split off the front portion -- note that we must insert the new
1033 	 * entry BEFORE this one, so that this entry has the specified
1034 	 * starting address.
1035 	 */
1036 	vm_map_simplify_entry(map, entry);
1037 
1038 	/*
1039 	 * If there is no object backing this entry, we might as well create
1040 	 * one now.  If we defer it, an object can get created after the map
1041 	 * is clipped, and individual objects will be created for the split-up
1042 	 * map.  This is a bit of a hack, but is also about the best place to
1043 	 * put this improvement.
1044 	 */
1045 	if (entry->object.vm_object == NULL && !map->system_map) {
1046 		vm_object_t object;
1047 		object = vm_object_allocate(OBJT_DEFAULT,
1048 				atop(entry->end - entry->start));
1049 		entry->object.vm_object = object;
1050 		entry->offset = 0;
1051 	}
1052 
1053 	new_entry = vm_map_entry_create(map);
1054 	*new_entry = *entry;
1055 
1056 	new_entry->end = start;
1057 	entry->offset += (start - entry->start);
1058 	entry->start = start;
1059 
1060 	vm_map_entry_link(map, entry->prev, new_entry);
1061 
1062 	if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1063 		vm_object_reference(new_entry->object.vm_object);
1064 	}
1065 }
1066 
1067 /*
1068  *	vm_map_clip_end:	[ internal use only ]
1069  *
1070  *	Asserts that the given entry ends at or before
1071  *	the specified address; if necessary,
1072  *	it splits the entry into two.
1073  */
1074 #define vm_map_clip_end(map, entry, endaddr) \
1075 { \
1076 	if (endaddr < entry->end) \
1077 		_vm_map_clip_end(map, entry, endaddr); \
1078 }
1079 
1080 /*
1081  *	This routine is called only when it is known that
1082  *	the entry must be split.
1083  */
1084 static void
1085 _vm_map_clip_end(vm_map_t map, vm_map_entry_t entry, vm_offset_t end)
1086 {
1087 	vm_map_entry_t new_entry;
1088 
1089 	/*
1090 	 * If there is no object backing this entry, we might as well create
1091 	 * one now.  If we defer it, an object can get created after the map
1092 	 * is clipped, and individual objects will be created for the split-up
1093 	 * map.  This is a bit of a hack, but is also about the best place to
1094 	 * put this improvement.
1095 	 */
1096 	if (entry->object.vm_object == NULL && !map->system_map) {
1097 		vm_object_t object;
1098 		object = vm_object_allocate(OBJT_DEFAULT,
1099 				atop(entry->end - entry->start));
1100 		entry->object.vm_object = object;
1101 		entry->offset = 0;
1102 	}
1103 
1104 	/*
1105 	 * Create a new entry and insert it AFTER the specified entry
1106 	 */
1107 	new_entry = vm_map_entry_create(map);
1108 	*new_entry = *entry;
1109 
1110 	new_entry->start = entry->end = end;
1111 	new_entry->offset += (end - entry->start);
1112 
1113 	vm_map_entry_link(map, entry, new_entry);
1114 
1115 	if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1116 		vm_object_reference(new_entry->object.vm_object);
1117 	}
1118 }
1119 
1120 /*
1121  *	VM_MAP_RANGE_CHECK:	[ internal use only ]
1122  *
1123  *	Asserts that the starting and ending region
1124  *	addresses fall within the valid range of the map.
1125  */
1126 #define	VM_MAP_RANGE_CHECK(map, start, end)		\
1127 		{					\
1128 		if (start < vm_map_min(map))		\
1129 			start = vm_map_min(map);	\
1130 		if (end > vm_map_max(map))		\
1131 			end = vm_map_max(map);		\
1132 		if (start > end)			\
1133 			start = end;			\
1134 		}
1135 
1136 /*
1137  *	vm_map_submap:		[ kernel use only ]
1138  *
1139  *	Mark the given range as handled by a subordinate map.
1140  *
1141  *	This range must have been created with vm_map_find,
1142  *	and no other operations may have been performed on this
1143  *	range prior to calling vm_map_submap.
1144  *
1145  *	Only a limited number of operations can be performed
1146  *	within this rage after calling vm_map_submap:
1147  *		vm_fault
1148  *	[Don't try vm_map_copy!]
1149  *
1150  *	To remove a submapping, one must first remove the
1151  *	range from the superior map, and then destroy the
1152  *	submap (if desired).  [Better yet, don't try it.]
1153  */
1154 int
1155 vm_map_submap(
1156 	vm_map_t map,
1157 	vm_offset_t start,
1158 	vm_offset_t end,
1159 	vm_map_t submap)
1160 {
1161 	vm_map_entry_t entry;
1162 	int result = KERN_INVALID_ARGUMENT;
1163 
1164 	GIANT_REQUIRED;
1165 
1166 	vm_map_lock(map);
1167 
1168 	VM_MAP_RANGE_CHECK(map, start, end);
1169 
1170 	if (vm_map_lookup_entry(map, start, &entry)) {
1171 		vm_map_clip_start(map, entry, start);
1172 	} else
1173 		entry = entry->next;
1174 
1175 	vm_map_clip_end(map, entry, end);
1176 
1177 	if ((entry->start == start) && (entry->end == end) &&
1178 	    ((entry->eflags & MAP_ENTRY_COW) == 0) &&
1179 	    (entry->object.vm_object == NULL)) {
1180 		entry->object.sub_map = submap;
1181 		entry->eflags |= MAP_ENTRY_IS_SUB_MAP;
1182 		result = KERN_SUCCESS;
1183 	}
1184 	vm_map_unlock(map);
1185 
1186 	return (result);
1187 }
1188 
1189 /*
1190  *	vm_map_protect:
1191  *
1192  *	Sets the protection of the specified address
1193  *	region in the target map.  If "set_max" is
1194  *	specified, the maximum protection is to be set;
1195  *	otherwise, only the current protection is affected.
1196  */
1197 int
1198 vm_map_protect(vm_map_t map, vm_offset_t start, vm_offset_t end,
1199 	       vm_prot_t new_prot, boolean_t set_max)
1200 {
1201 	vm_map_entry_t current;
1202 	vm_map_entry_t entry;
1203 
1204 	GIANT_REQUIRED;
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 
1215 	/*
1216 	 * Make a first pass to check for protection violations.
1217 	 */
1218 	current = entry;
1219 	while ((current != &map->header) && (current->start < end)) {
1220 		if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
1221 			vm_map_unlock(map);
1222 			return (KERN_INVALID_ARGUMENT);
1223 		}
1224 		if ((new_prot & current->max_protection) != new_prot) {
1225 			vm_map_unlock(map);
1226 			return (KERN_PROTECTION_FAILURE);
1227 		}
1228 		current = current->next;
1229 	}
1230 
1231 	/*
1232 	 * Go back and fix up protections. [Note that clipping is not
1233 	 * necessary the second time.]
1234 	 */
1235 	current = entry;
1236 	while ((current != &map->header) && (current->start < end)) {
1237 		vm_prot_t old_prot;
1238 
1239 		vm_map_clip_end(map, current, end);
1240 
1241 		old_prot = current->protection;
1242 		if (set_max)
1243 			current->protection =
1244 			    (current->max_protection = new_prot) &
1245 			    old_prot;
1246 		else
1247 			current->protection = new_prot;
1248 
1249 		/*
1250 		 * Update physical map if necessary. Worry about copy-on-write
1251 		 * here -- CHECK THIS XXX
1252 		 */
1253 		if (current->protection != old_prot) {
1254 #define MASK(entry)	(((entry)->eflags & MAP_ENTRY_COW) ? ~VM_PROT_WRITE : \
1255 							VM_PROT_ALL)
1256 			pmap_protect(map->pmap, current->start,
1257 			    current->end,
1258 			    current->protection & MASK(current));
1259 #undef	MASK
1260 		}
1261 		vm_map_simplify_entry(map, current);
1262 		current = current->next;
1263 	}
1264 	vm_map_unlock(map);
1265 	return (KERN_SUCCESS);
1266 }
1267 
1268 /*
1269  *	vm_map_madvise:
1270  *
1271  * 	This routine traverses a processes map handling the madvise
1272  *	system call.  Advisories are classified as either those effecting
1273  *	the vm_map_entry structure, or those effecting the underlying
1274  *	objects.
1275  */
1276 int
1277 vm_map_madvise(
1278 	vm_map_t map,
1279 	vm_offset_t start,
1280 	vm_offset_t end,
1281 	int behav)
1282 {
1283 	vm_map_entry_t current, entry;
1284 	int modify_map = 0;
1285 
1286 	GIANT_REQUIRED;
1287 
1288 	/*
1289 	 * Some madvise calls directly modify the vm_map_entry, in which case
1290 	 * we need to use an exclusive lock on the map and we need to perform
1291 	 * various clipping operations.  Otherwise we only need a read-lock
1292 	 * on the map.
1293 	 */
1294 	switch(behav) {
1295 	case MADV_NORMAL:
1296 	case MADV_SEQUENTIAL:
1297 	case MADV_RANDOM:
1298 	case MADV_NOSYNC:
1299 	case MADV_AUTOSYNC:
1300 	case MADV_NOCORE:
1301 	case MADV_CORE:
1302 		modify_map = 1;
1303 		vm_map_lock(map);
1304 		break;
1305 	case MADV_WILLNEED:
1306 	case MADV_DONTNEED:
1307 	case MADV_FREE:
1308 		vm_map_lock_read(map);
1309 		break;
1310 	default:
1311 		return (KERN_INVALID_ARGUMENT);
1312 	}
1313 
1314 	/*
1315 	 * Locate starting entry and clip if necessary.
1316 	 */
1317 	VM_MAP_RANGE_CHECK(map, start, end);
1318 
1319 	if (vm_map_lookup_entry(map, start, &entry)) {
1320 		if (modify_map)
1321 			vm_map_clip_start(map, entry, start);
1322 	} else {
1323 		entry = entry->next;
1324 	}
1325 
1326 	if (modify_map) {
1327 		/*
1328 		 * madvise behaviors that are implemented in the vm_map_entry.
1329 		 *
1330 		 * We clip the vm_map_entry so that behavioral changes are
1331 		 * limited to the specified address range.
1332 		 */
1333 		for (current = entry;
1334 		     (current != &map->header) && (current->start < end);
1335 		     current = current->next
1336 		) {
1337 			if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
1338 				continue;
1339 
1340 			vm_map_clip_end(map, current, end);
1341 
1342 			switch (behav) {
1343 			case MADV_NORMAL:
1344 				vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_NORMAL);
1345 				break;
1346 			case MADV_SEQUENTIAL:
1347 				vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_SEQUENTIAL);
1348 				break;
1349 			case MADV_RANDOM:
1350 				vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_RANDOM);
1351 				break;
1352 			case MADV_NOSYNC:
1353 				current->eflags |= MAP_ENTRY_NOSYNC;
1354 				break;
1355 			case MADV_AUTOSYNC:
1356 				current->eflags &= ~MAP_ENTRY_NOSYNC;
1357 				break;
1358 			case MADV_NOCORE:
1359 				current->eflags |= MAP_ENTRY_NOCOREDUMP;
1360 				break;
1361 			case MADV_CORE:
1362 				current->eflags &= ~MAP_ENTRY_NOCOREDUMP;
1363 				break;
1364 			default:
1365 				break;
1366 			}
1367 			vm_map_simplify_entry(map, current);
1368 		}
1369 		vm_map_unlock(map);
1370 	} else {
1371 		vm_pindex_t pindex;
1372 		int count;
1373 
1374 		/*
1375 		 * madvise behaviors that are implemented in the underlying
1376 		 * vm_object.
1377 		 *
1378 		 * Since we don't clip the vm_map_entry, we have to clip
1379 		 * the vm_object pindex and count.
1380 		 */
1381 		for (current = entry;
1382 		     (current != &map->header) && (current->start < end);
1383 		     current = current->next
1384 		) {
1385 			vm_offset_t useStart;
1386 
1387 			if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
1388 				continue;
1389 
1390 			pindex = OFF_TO_IDX(current->offset);
1391 			count = atop(current->end - current->start);
1392 			useStart = current->start;
1393 
1394 			if (current->start < start) {
1395 				pindex += atop(start - current->start);
1396 				count -= atop(start - current->start);
1397 				useStart = start;
1398 			}
1399 			if (current->end > end)
1400 				count -= atop(current->end - end);
1401 
1402 			if (count <= 0)
1403 				continue;
1404 
1405 			vm_object_madvise(current->object.vm_object,
1406 					  pindex, count, behav);
1407 			if (behav == MADV_WILLNEED) {
1408 				pmap_object_init_pt(
1409 				    map->pmap,
1410 				    useStart,
1411 				    current->object.vm_object,
1412 				    pindex,
1413 				    (count << PAGE_SHIFT),
1414 				    MAP_PREFAULT_MADVISE
1415 				);
1416 			}
1417 		}
1418 		vm_map_unlock_read(map);
1419 	}
1420 	return (0);
1421 }
1422 
1423 
1424 /*
1425  *	vm_map_inherit:
1426  *
1427  *	Sets the inheritance of the specified address
1428  *	range in the target map.  Inheritance
1429  *	affects how the map will be shared with
1430  *	child maps at the time of vm_map_fork.
1431  */
1432 int
1433 vm_map_inherit(vm_map_t map, vm_offset_t start, vm_offset_t end,
1434 	       vm_inherit_t new_inheritance)
1435 {
1436 	vm_map_entry_t entry;
1437 	vm_map_entry_t temp_entry;
1438 
1439 	GIANT_REQUIRED;
1440 
1441 	switch (new_inheritance) {
1442 	case VM_INHERIT_NONE:
1443 	case VM_INHERIT_COPY:
1444 	case VM_INHERIT_SHARE:
1445 		break;
1446 	default:
1447 		return (KERN_INVALID_ARGUMENT);
1448 	}
1449 
1450 	vm_map_lock(map);
1451 
1452 	VM_MAP_RANGE_CHECK(map, start, end);
1453 
1454 	if (vm_map_lookup_entry(map, start, &temp_entry)) {
1455 		entry = temp_entry;
1456 		vm_map_clip_start(map, entry, start);
1457 	} else
1458 		entry = temp_entry->next;
1459 
1460 	while ((entry != &map->header) && (entry->start < end)) {
1461 		vm_map_clip_end(map, entry, end);
1462 
1463 		entry->inheritance = new_inheritance;
1464 
1465 		vm_map_simplify_entry(map, entry);
1466 
1467 		entry = entry->next;
1468 	}
1469 
1470 	vm_map_unlock(map);
1471 	return (KERN_SUCCESS);
1472 }
1473 
1474 /*
1475  * Implement the semantics of mlock
1476  */
1477 int
1478 vm_map_user_pageable(
1479 	vm_map_t map,
1480 	vm_offset_t start,
1481 	vm_offset_t end,
1482 	boolean_t new_pageable)
1483 {
1484 	vm_map_entry_t entry;
1485 	vm_map_entry_t start_entry;
1486 	vm_offset_t estart;
1487 	vm_offset_t eend;
1488 	int rv;
1489 
1490 	vm_map_lock(map);
1491 	VM_MAP_RANGE_CHECK(map, start, end);
1492 
1493 	if (vm_map_lookup_entry(map, start, &start_entry) == FALSE) {
1494 		vm_map_unlock(map);
1495 		return (KERN_INVALID_ADDRESS);
1496 	}
1497 
1498 	if (new_pageable) {
1499 
1500 		entry = start_entry;
1501 		vm_map_clip_start(map, entry, start);
1502 
1503 		/*
1504 		 * Now decrement the wiring count for each region. If a region
1505 		 * becomes completely unwired, unwire its physical pages and
1506 		 * mappings.
1507 		 */
1508 		while ((entry != &map->header) && (entry->start < end)) {
1509 			if (entry->eflags & MAP_ENTRY_USER_WIRED) {
1510 				vm_map_clip_end(map, entry, end);
1511 				entry->eflags &= ~MAP_ENTRY_USER_WIRED;
1512 				entry->wired_count--;
1513 				if (entry->wired_count == 0)
1514 					vm_fault_unwire(map, entry->start, entry->end);
1515 			}
1516 			vm_map_simplify_entry(map,entry);
1517 			entry = entry->next;
1518 		}
1519 	} else {
1520 
1521 		entry = start_entry;
1522 
1523 		while ((entry != &map->header) && (entry->start < end)) {
1524 
1525 			if (entry->eflags & MAP_ENTRY_USER_WIRED) {
1526 				entry = entry->next;
1527 				continue;
1528 			}
1529 
1530 			if (entry->wired_count != 0) {
1531 				entry->wired_count++;
1532 				entry->eflags |= MAP_ENTRY_USER_WIRED;
1533 				entry = entry->next;
1534 				continue;
1535 			}
1536 
1537 			/* Here on entry being newly wired */
1538 
1539 			if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1540 				int copyflag = entry->eflags & MAP_ENTRY_NEEDS_COPY;
1541 				if (copyflag && ((entry->protection & VM_PROT_WRITE) != 0)) {
1542 
1543 					vm_object_shadow(&entry->object.vm_object,
1544 					    &entry->offset,
1545 					    atop(entry->end - entry->start));
1546 					entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
1547 
1548 				} else if (entry->object.vm_object == NULL &&
1549 					   !map->system_map) {
1550 
1551 					entry->object.vm_object =
1552 					    vm_object_allocate(OBJT_DEFAULT,
1553 						atop(entry->end - entry->start));
1554 					entry->offset = (vm_offset_t) 0;
1555 
1556 				}
1557 			}
1558 
1559 			vm_map_clip_start(map, entry, start);
1560 			vm_map_clip_end(map, entry, end);
1561 
1562 			entry->wired_count++;
1563 			entry->eflags |= MAP_ENTRY_USER_WIRED;
1564 			estart = entry->start;
1565 			eend = entry->end;
1566 
1567 			/* First we need to allow map modifications */
1568 			vm_map_set_recursive(map);
1569 			vm_map_lock_downgrade(map);
1570 			map->timestamp++;
1571 
1572 			rv = vm_fault_user_wire(map, entry->start, entry->end);
1573 			if (rv) {
1574 
1575 				entry->wired_count--;
1576 				entry->eflags &= ~MAP_ENTRY_USER_WIRED;
1577 
1578 				vm_map_clear_recursive(map);
1579 				vm_map_unlock(map);
1580 
1581 				/*
1582 				 * At this point, the map is unlocked, and
1583 				 * entry might no longer be valid.  Use copy
1584 				 * of entry start value obtained while entry
1585 				 * was valid.
1586 				 */
1587 				(void) vm_map_user_pageable(map, start, estart,
1588 							    TRUE);
1589 				return rv;
1590 			}
1591 
1592 			vm_map_clear_recursive(map);
1593 			if (vm_map_lock_upgrade(map)) {
1594 				vm_map_lock(map);
1595 				if (vm_map_lookup_entry(map, estart, &entry)
1596 				    == FALSE) {
1597 					vm_map_unlock(map);
1598 					/*
1599 					 * vm_fault_user_wire succeded, thus
1600 					 * the area between start and eend
1601 					 * is wired and has to be unwired
1602 					 * here as part of the cleanup.
1603 					 */
1604 					(void) vm_map_user_pageable(map,
1605 								    start,
1606 								    eend,
1607 								    TRUE);
1608 					return (KERN_INVALID_ADDRESS);
1609 				}
1610 			}
1611 			vm_map_simplify_entry(map,entry);
1612 		}
1613 	}
1614 	map->timestamp++;
1615 	vm_map_unlock(map);
1616 	return KERN_SUCCESS;
1617 }
1618 
1619 /*
1620  *	vm_map_pageable:
1621  *
1622  *	Sets the pageability of the specified address
1623  *	range in the target map.  Regions specified
1624  *	as not pageable require locked-down physical
1625  *	memory and physical page maps.
1626  *
1627  *	The map must not be locked, but a reference
1628  *	must remain to the map throughout the call.
1629  */
1630 int
1631 vm_map_pageable(
1632 	vm_map_t map,
1633 	vm_offset_t start,
1634 	vm_offset_t end,
1635 	boolean_t new_pageable)
1636 {
1637 	vm_map_entry_t entry;
1638 	vm_map_entry_t start_entry;
1639 	vm_offset_t failed = 0;
1640 	int rv;
1641 
1642 	GIANT_REQUIRED;
1643 
1644 	vm_map_lock(map);
1645 
1646 	VM_MAP_RANGE_CHECK(map, start, end);
1647 
1648 	/*
1649 	 * Only one pageability change may take place at one time, since
1650 	 * vm_fault assumes it will be called only once for each
1651 	 * wiring/unwiring.  Therefore, we have to make sure we're actually
1652 	 * changing the pageability for the entire region.  We do so before
1653 	 * making any changes.
1654 	 */
1655 	if (vm_map_lookup_entry(map, start, &start_entry) == FALSE) {
1656 		vm_map_unlock(map);
1657 		return (KERN_INVALID_ADDRESS);
1658 	}
1659 	entry = start_entry;
1660 
1661 	/*
1662 	 * Actions are rather different for wiring and unwiring, so we have
1663 	 * two separate cases.
1664 	 */
1665 	if (new_pageable) {
1666 		vm_map_clip_start(map, entry, start);
1667 
1668 		/*
1669 		 * Unwiring.  First ensure that the range to be unwired is
1670 		 * really wired down and that there are no holes.
1671 		 */
1672 		while ((entry != &map->header) && (entry->start < end)) {
1673 			if (entry->wired_count == 0 ||
1674 			    (entry->end < end &&
1675 				(entry->next == &map->header ||
1676 				    entry->next->start > entry->end))) {
1677 				vm_map_unlock(map);
1678 				return (KERN_INVALID_ARGUMENT);
1679 			}
1680 			entry = entry->next;
1681 		}
1682 
1683 		/*
1684 		 * Now decrement the wiring count for each region. If a region
1685 		 * becomes completely unwired, unwire its physical pages and
1686 		 * mappings.
1687 		 */
1688 		entry = start_entry;
1689 		while ((entry != &map->header) && (entry->start < end)) {
1690 			vm_map_clip_end(map, entry, end);
1691 
1692 			entry->wired_count--;
1693 			if (entry->wired_count == 0)
1694 				vm_fault_unwire(map, entry->start, entry->end);
1695 
1696 			vm_map_simplify_entry(map, entry);
1697 
1698 			entry = entry->next;
1699 		}
1700 	} else {
1701 		/*
1702 		 * Wiring.  We must do this in two passes:
1703 		 *
1704 		 * 1.  Holding the write lock, we create any shadow or zero-fill
1705 		 * objects that need to be created. Then we clip each map
1706 		 * entry to the region to be wired and increment its wiring
1707 		 * count.  We create objects before clipping the map entries
1708 		 * to avoid object proliferation.
1709 		 *
1710 		 * 2.  We downgrade to a read lock, and call vm_fault_wire to
1711 		 * fault in the pages for any newly wired area (wired_count is
1712 		 * 1).
1713 		 *
1714 		 * Downgrading to a read lock for vm_fault_wire avoids a possible
1715 		 * deadlock with another process that may have faulted on one
1716 		 * of the pages to be wired (it would mark the page busy,
1717 		 * blocking us, then in turn block on the map lock that we
1718 		 * hold).  Because of problems in the recursive lock package,
1719 		 * we cannot upgrade to a write lock in vm_map_lookup.  Thus,
1720 		 * any actions that require the write lock must be done
1721 		 * beforehand.  Because we keep the read lock on the map, the
1722 		 * copy-on-write status of the entries we modify here cannot
1723 		 * change.
1724 		 */
1725 
1726 		/*
1727 		 * Pass 1.
1728 		 */
1729 		while ((entry != &map->header) && (entry->start < end)) {
1730 			if (entry->wired_count == 0) {
1731 
1732 				/*
1733 				 * Perform actions of vm_map_lookup that need
1734 				 * the write lock on the map: create a shadow
1735 				 * object for a copy-on-write region, or an
1736 				 * object for a zero-fill region.
1737 				 *
1738 				 * We don't have to do this for entries that
1739 				 * point to sub maps, because we won't
1740 				 * hold the lock on the sub map.
1741 				 */
1742 				if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1743 					int copyflag = entry->eflags & MAP_ENTRY_NEEDS_COPY;
1744 					if (copyflag &&
1745 					    ((entry->protection & VM_PROT_WRITE) != 0)) {
1746 
1747 						vm_object_shadow(&entry->object.vm_object,
1748 						    &entry->offset,
1749 						    atop(entry->end - entry->start));
1750 						entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
1751 					} else if (entry->object.vm_object == NULL &&
1752 						   !map->system_map) {
1753 						entry->object.vm_object =
1754 						    vm_object_allocate(OBJT_DEFAULT,
1755 							atop(entry->end - entry->start));
1756 						entry->offset = (vm_offset_t) 0;
1757 					}
1758 				}
1759 			}
1760 			vm_map_clip_start(map, entry, start);
1761 			vm_map_clip_end(map, entry, end);
1762 			entry->wired_count++;
1763 
1764 			/*
1765 			 * Check for holes
1766 			 */
1767 			if (entry->end < end &&
1768 			    (entry->next == &map->header ||
1769 				entry->next->start > entry->end)) {
1770 				/*
1771 				 * Found one.  Object creation actions do not
1772 				 * need to be undone, but the wired counts
1773 				 * need to be restored.
1774 				 */
1775 				while (entry != &map->header && entry->end > start) {
1776 					entry->wired_count--;
1777 					entry = entry->prev;
1778 				}
1779 				vm_map_unlock(map);
1780 				return (KERN_INVALID_ARGUMENT);
1781 			}
1782 			entry = entry->next;
1783 		}
1784 
1785 		/*
1786 		 * Pass 2.
1787 		 */
1788 
1789 		/*
1790 		 * HACK HACK HACK HACK
1791 		 *
1792 		 * If we are wiring in the kernel map or a submap of it,
1793 		 * unlock the map to avoid deadlocks.  We trust that the
1794 		 * kernel is well-behaved, and therefore will not do
1795 		 * anything destructive to this region of the map while
1796 		 * we have it unlocked.  We cannot trust user processes
1797 		 * to do the same.
1798 		 *
1799 		 * HACK HACK HACK HACK
1800 		 */
1801 		if (vm_map_pmap(map) == kernel_pmap) {
1802 			vm_map_unlock(map);	/* trust me ... */
1803 		} else {
1804 			vm_map_lock_downgrade(map);
1805 		}
1806 
1807 		rv = 0;
1808 		entry = start_entry;
1809 		while (entry != &map->header && entry->start < end) {
1810 			/*
1811 			 * If vm_fault_wire fails for any page we need to undo
1812 			 * what has been done.  We decrement the wiring count
1813 			 * for those pages which have not yet been wired (now)
1814 			 * and unwire those that have (later).
1815 			 *
1816 			 * XXX this violates the locking protocol on the map,
1817 			 * needs to be fixed.
1818 			 */
1819 			if (rv)
1820 				entry->wired_count--;
1821 			else if (entry->wired_count == 1) {
1822 				rv = vm_fault_wire(map, entry->start, entry->end);
1823 				if (rv) {
1824 					failed = entry->start;
1825 					entry->wired_count--;
1826 				}
1827 			}
1828 			entry = entry->next;
1829 		}
1830 
1831 		if (vm_map_pmap(map) == kernel_pmap) {
1832 			vm_map_lock(map);
1833 		}
1834 		if (rv) {
1835 			vm_map_unlock(map);
1836 			(void) vm_map_pageable(map, start, failed, TRUE);
1837 			return (rv);
1838 		}
1839 		/*
1840 		 * An exclusive lock on the map is needed in order to call
1841 		 * vm_map_simplify_entry().  If the current lock on the map
1842 		 * is only a shared lock, an upgrade is needed.
1843 		 */
1844 		if (vm_map_pmap(map) != kernel_pmap &&
1845 		    vm_map_lock_upgrade(map)) {
1846 			vm_map_lock(map);
1847 			if (vm_map_lookup_entry(map, start, &start_entry) ==
1848 			    FALSE) {
1849 				vm_map_unlock(map);
1850 				return KERN_SUCCESS;
1851 			}
1852 		}
1853 		vm_map_simplify_entry(map, start_entry);
1854 	}
1855 
1856 	vm_map_unlock(map);
1857 
1858 	return (KERN_SUCCESS);
1859 }
1860 
1861 /*
1862  * vm_map_clean
1863  *
1864  * Push any dirty cached pages in the address range to their pager.
1865  * If syncio is TRUE, dirty pages are written synchronously.
1866  * If invalidate is TRUE, any cached pages are freed as well.
1867  *
1868  * Returns an error if any part of the specified range is not mapped.
1869  */
1870 int
1871 vm_map_clean(
1872 	vm_map_t map,
1873 	vm_offset_t start,
1874 	vm_offset_t end,
1875 	boolean_t syncio,
1876 	boolean_t invalidate)
1877 {
1878 	vm_map_entry_t current;
1879 	vm_map_entry_t entry;
1880 	vm_size_t size;
1881 	vm_object_t object;
1882 	vm_ooffset_t offset;
1883 
1884 	GIANT_REQUIRED;
1885 
1886 	vm_map_lock_read(map);
1887 	VM_MAP_RANGE_CHECK(map, start, end);
1888 	if (!vm_map_lookup_entry(map, start, &entry)) {
1889 		vm_map_unlock_read(map);
1890 		return (KERN_INVALID_ADDRESS);
1891 	}
1892 	/*
1893 	 * Make a first pass to check for holes.
1894 	 */
1895 	for (current = entry; current->start < end; current = current->next) {
1896 		if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
1897 			vm_map_unlock_read(map);
1898 			return (KERN_INVALID_ARGUMENT);
1899 		}
1900 		if (end > current->end &&
1901 		    (current->next == &map->header ||
1902 			current->end != current->next->start)) {
1903 			vm_map_unlock_read(map);
1904 			return (KERN_INVALID_ADDRESS);
1905 		}
1906 	}
1907 
1908 	if (invalidate)
1909 		pmap_remove(vm_map_pmap(map), start, end);
1910 	/*
1911 	 * Make a second pass, cleaning/uncaching pages from the indicated
1912 	 * objects as we go.
1913 	 */
1914 	for (current = entry; current->start < end; current = current->next) {
1915 		offset = current->offset + (start - current->start);
1916 		size = (end <= current->end ? end : current->end) - start;
1917 		if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
1918 			vm_map_t smap;
1919 			vm_map_entry_t tentry;
1920 			vm_size_t tsize;
1921 
1922 			smap = current->object.sub_map;
1923 			vm_map_lock_read(smap);
1924 			(void) vm_map_lookup_entry(smap, offset, &tentry);
1925 			tsize = tentry->end - offset;
1926 			if (tsize < size)
1927 				size = tsize;
1928 			object = tentry->object.vm_object;
1929 			offset = tentry->offset + (offset - tentry->start);
1930 			vm_map_unlock_read(smap);
1931 		} else {
1932 			object = current->object.vm_object;
1933 		}
1934 		/*
1935 		 * Note that there is absolutely no sense in writing out
1936 		 * anonymous objects, so we track down the vnode object
1937 		 * to write out.
1938 		 * We invalidate (remove) all pages from the address space
1939 		 * anyway, for semantic correctness.
1940 		 *
1941 		 * note: certain anonymous maps, such as MAP_NOSYNC maps,
1942 		 * may start out with a NULL object.
1943 		 */
1944 		while (object && object->backing_object) {
1945 			object = object->backing_object;
1946 			offset += object->backing_object_offset;
1947 			if (object->size < OFF_TO_IDX(offset + size))
1948 				size = IDX_TO_OFF(object->size) - offset;
1949 		}
1950 		if (object && (object->type == OBJT_VNODE) &&
1951 		    (current->protection & VM_PROT_WRITE)) {
1952 			/*
1953 			 * Flush pages if writing is allowed, invalidate them
1954 			 * if invalidation requested.  Pages undergoing I/O
1955 			 * will be ignored by vm_object_page_remove().
1956 			 *
1957 			 * We cannot lock the vnode and then wait for paging
1958 			 * to complete without deadlocking against vm_fault.
1959 			 * Instead we simply call vm_object_page_remove() and
1960 			 * allow it to block internally on a page-by-page
1961 			 * basis when it encounters pages undergoing async
1962 			 * I/O.
1963 			 */
1964 			int flags;
1965 
1966 			vm_object_reference(object);
1967 			vn_lock(object->handle, LK_EXCLUSIVE | LK_RETRY, curthread);
1968 			flags = (syncio || invalidate) ? OBJPC_SYNC : 0;
1969 			flags |= invalidate ? OBJPC_INVAL : 0;
1970 			vm_object_page_clean(object,
1971 			    OFF_TO_IDX(offset),
1972 			    OFF_TO_IDX(offset + size + PAGE_MASK),
1973 			    flags);
1974 			if (invalidate) {
1975 				/*vm_object_pip_wait(object, "objmcl");*/
1976 				vm_object_page_remove(object,
1977 				    OFF_TO_IDX(offset),
1978 				    OFF_TO_IDX(offset + size + PAGE_MASK),
1979 				    FALSE);
1980 			}
1981 			VOP_UNLOCK(object->handle, 0, curthread);
1982 			vm_object_deallocate(object);
1983 		}
1984 		start += size;
1985 	}
1986 
1987 	vm_map_unlock_read(map);
1988 	return (KERN_SUCCESS);
1989 }
1990 
1991 /*
1992  *	vm_map_entry_unwire:	[ internal use only ]
1993  *
1994  *	Make the region specified by this entry pageable.
1995  *
1996  *	The map in question should be locked.
1997  *	[This is the reason for this routine's existence.]
1998  */
1999 static void
2000 vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry)
2001 {
2002 	vm_fault_unwire(map, entry->start, entry->end);
2003 	entry->wired_count = 0;
2004 }
2005 
2006 /*
2007  *	vm_map_entry_delete:	[ internal use only ]
2008  *
2009  *	Deallocate the given entry from the target map.
2010  */
2011 static void
2012 vm_map_entry_delete(vm_map_t map, vm_map_entry_t entry)
2013 {
2014 	vm_map_entry_unlink(map, entry);
2015 	map->size -= entry->end - entry->start;
2016 
2017 	if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
2018 		vm_object_deallocate(entry->object.vm_object);
2019 	}
2020 
2021 	vm_map_entry_dispose(map, entry);
2022 }
2023 
2024 /*
2025  *	vm_map_delete:	[ internal use only ]
2026  *
2027  *	Deallocates the given address range from the target
2028  *	map.
2029  */
2030 int
2031 vm_map_delete(vm_map_t map, vm_offset_t start, vm_offset_t end)
2032 {
2033 	vm_object_t object;
2034 	vm_map_entry_t entry;
2035 	vm_map_entry_t first_entry;
2036 
2037 	GIANT_REQUIRED;
2038 
2039 	/*
2040 	 * Find the start of the region, and clip it
2041 	 */
2042 	if (!vm_map_lookup_entry(map, start, &first_entry))
2043 		entry = first_entry->next;
2044 	else {
2045 		entry = first_entry;
2046 		vm_map_clip_start(map, entry, start);
2047 		/*
2048 		 * Fix the lookup hint now, rather than each time though the
2049 		 * loop.
2050 		 */
2051 		SAVE_HINT(map, entry->prev);
2052 	}
2053 
2054 	/*
2055 	 * Save the free space hint
2056 	 */
2057 	if (entry == &map->header) {
2058 		map->first_free = &map->header;
2059 	} else if (map->first_free->start >= start) {
2060 		map->first_free = entry->prev;
2061 	}
2062 
2063 	/*
2064 	 * Step through all entries in this region
2065 	 */
2066 	while ((entry != &map->header) && (entry->start < end)) {
2067 		vm_map_entry_t next;
2068 		vm_offset_t s, e;
2069 		vm_pindex_t offidxstart, offidxend, count;
2070 
2071 		vm_map_clip_end(map, entry, end);
2072 
2073 		s = entry->start;
2074 		e = entry->end;
2075 		next = entry->next;
2076 
2077 		offidxstart = OFF_TO_IDX(entry->offset);
2078 		count = OFF_TO_IDX(e - s);
2079 		object = entry->object.vm_object;
2080 
2081 		/*
2082 		 * Unwire before removing addresses from the pmap; otherwise,
2083 		 * unwiring will put the entries back in the pmap.
2084 		 */
2085 		if (entry->wired_count != 0) {
2086 			vm_map_entry_unwire(map, entry);
2087 		}
2088 
2089 		offidxend = offidxstart + count;
2090 
2091 		if ((object == kernel_object) || (object == kmem_object)) {
2092 			vm_object_page_remove(object, offidxstart, offidxend, FALSE);
2093 		} else {
2094 			pmap_remove(map->pmap, s, e);
2095 			if (object != NULL &&
2096 			    object->ref_count != 1 &&
2097 			    (object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING &&
2098 			    (object->type == OBJT_DEFAULT || object->type == OBJT_SWAP)) {
2099 				vm_object_collapse(object);
2100 				vm_object_page_remove(object, offidxstart, offidxend, FALSE);
2101 				if (object->type == OBJT_SWAP) {
2102 					swap_pager_freespace(object, offidxstart, count);
2103 				}
2104 				if (offidxend >= object->size &&
2105 				    offidxstart < object->size) {
2106 					object->size = offidxstart;
2107 				}
2108 			}
2109 		}
2110 
2111 		/*
2112 		 * Delete the entry (which may delete the object) only after
2113 		 * removing all pmap entries pointing to its pages.
2114 		 * (Otherwise, its page frames may be reallocated, and any
2115 		 * modify bits will be set in the wrong object!)
2116 		 */
2117 		vm_map_entry_delete(map, entry);
2118 		entry = next;
2119 	}
2120 	return (KERN_SUCCESS);
2121 }
2122 
2123 /*
2124  *	vm_map_remove:
2125  *
2126  *	Remove the given address range from the target map.
2127  *	This is the exported form of vm_map_delete.
2128  */
2129 int
2130 vm_map_remove(vm_map_t map, vm_offset_t start, vm_offset_t end)
2131 {
2132 	int result, s = 0;
2133 
2134 	GIANT_REQUIRED;
2135 
2136 	if (map == kmem_map)
2137 		s = splvm();
2138 
2139 	vm_map_lock(map);
2140 	VM_MAP_RANGE_CHECK(map, start, end);
2141 	result = vm_map_delete(map, start, end);
2142 	vm_map_unlock(map);
2143 
2144 	if (map == kmem_map)
2145 		splx(s);
2146 
2147 	return (result);
2148 }
2149 
2150 /*
2151  *	vm_map_check_protection:
2152  *
2153  *	Assert that the target map allows the specified
2154  *	privilege on the entire address region given.
2155  *	The entire region must be allocated.
2156  */
2157 boolean_t
2158 vm_map_check_protection(vm_map_t map, vm_offset_t start, vm_offset_t end,
2159 			vm_prot_t protection)
2160 {
2161 	vm_map_entry_t entry;
2162 	vm_map_entry_t tmp_entry;
2163 
2164 	GIANT_REQUIRED;
2165 
2166 	vm_map_lock_read(map);
2167 	if (!vm_map_lookup_entry(map, start, &tmp_entry)) {
2168 		vm_map_unlock_read(map);
2169 		return (FALSE);
2170 	}
2171 	entry = tmp_entry;
2172 
2173 	while (start < end) {
2174 		if (entry == &map->header) {
2175 			vm_map_unlock_read(map);
2176 			return (FALSE);
2177 		}
2178 		/*
2179 		 * No holes allowed!
2180 		 */
2181 		if (start < entry->start) {
2182 			vm_map_unlock_read(map);
2183 			return (FALSE);
2184 		}
2185 		/*
2186 		 * Check protection associated with entry.
2187 		 */
2188 		if ((entry->protection & protection) != protection) {
2189 			vm_map_unlock_read(map);
2190 			return (FALSE);
2191 		}
2192 		/* go to next entry */
2193 		start = entry->end;
2194 		entry = entry->next;
2195 	}
2196 	vm_map_unlock_read(map);
2197 	return (TRUE);
2198 }
2199 
2200 /*
2201  * Split the pages in a map entry into a new object.  This affords
2202  * easier removal of unused pages, and keeps object inheritance from
2203  * being a negative impact on memory usage.
2204  */
2205 static void
2206 vm_map_split(vm_map_entry_t entry)
2207 {
2208 	vm_page_t m;
2209 	vm_object_t orig_object, new_object, source;
2210 	vm_offset_t s, e;
2211 	vm_pindex_t offidxstart, offidxend, idx;
2212 	vm_size_t size;
2213 	vm_ooffset_t offset;
2214 
2215 	GIANT_REQUIRED;
2216 
2217 	orig_object = entry->object.vm_object;
2218 	if (orig_object->type != OBJT_DEFAULT && orig_object->type != OBJT_SWAP)
2219 		return;
2220 	if (orig_object->ref_count <= 1)
2221 		return;
2222 
2223 	offset = entry->offset;
2224 	s = entry->start;
2225 	e = entry->end;
2226 
2227 	offidxstart = OFF_TO_IDX(offset);
2228 	offidxend = offidxstart + OFF_TO_IDX(e - s);
2229 	size = offidxend - offidxstart;
2230 
2231 	new_object = vm_pager_allocate(orig_object->type,
2232 		NULL, IDX_TO_OFF(size), VM_PROT_ALL, 0LL);
2233 	if (new_object == NULL)
2234 		return;
2235 
2236 	source = orig_object->backing_object;
2237 	if (source != NULL) {
2238 		vm_object_reference(source);	/* Referenced by new_object */
2239 		TAILQ_INSERT_TAIL(&source->shadow_head,
2240 				  new_object, shadow_list);
2241 		vm_object_clear_flag(source, OBJ_ONEMAPPING);
2242 		new_object->backing_object_offset =
2243 			orig_object->backing_object_offset + IDX_TO_OFF(offidxstart);
2244 		new_object->backing_object = source;
2245 		source->shadow_count++;
2246 		source->generation++;
2247 	}
2248 
2249 	for (idx = 0; idx < size; idx++) {
2250 		vm_page_t m;
2251 
2252 	retry:
2253 		m = vm_page_lookup(orig_object, offidxstart + idx);
2254 		if (m == NULL)
2255 			continue;
2256 
2257 		/*
2258 		 * We must wait for pending I/O to complete before we can
2259 		 * rename the page.
2260 		 *
2261 		 * We do not have to VM_PROT_NONE the page as mappings should
2262 		 * not be changed by this operation.
2263 		 */
2264 		if (vm_page_sleep_busy(m, TRUE, "spltwt"))
2265 			goto retry;
2266 
2267 		vm_page_busy(m);
2268 		vm_page_rename(m, new_object, idx);
2269 		/* page automatically made dirty by rename and cache handled */
2270 		vm_page_busy(m);
2271 	}
2272 
2273 	if (orig_object->type == OBJT_SWAP) {
2274 		vm_object_pip_add(orig_object, 1);
2275 		/*
2276 		 * copy orig_object pages into new_object
2277 		 * and destroy unneeded pages in
2278 		 * shadow object.
2279 		 */
2280 		swap_pager_copy(orig_object, new_object, offidxstart, 0);
2281 		vm_object_pip_wakeup(orig_object);
2282 	}
2283 
2284 	for (idx = 0; idx < size; idx++) {
2285 		m = vm_page_lookup(new_object, idx);
2286 		if (m) {
2287 			vm_page_wakeup(m);
2288 		}
2289 	}
2290 
2291 	entry->object.vm_object = new_object;
2292 	entry->offset = 0LL;
2293 	vm_object_deallocate(orig_object);
2294 }
2295 
2296 /*
2297  *	vm_map_copy_entry:
2298  *
2299  *	Copies the contents of the source entry to the destination
2300  *	entry.  The entries *must* be aligned properly.
2301  */
2302 static void
2303 vm_map_copy_entry(
2304 	vm_map_t src_map,
2305 	vm_map_t dst_map,
2306 	vm_map_entry_t src_entry,
2307 	vm_map_entry_t dst_entry)
2308 {
2309 	vm_object_t src_object;
2310 
2311 	if ((dst_entry->eflags|src_entry->eflags) & MAP_ENTRY_IS_SUB_MAP)
2312 		return;
2313 
2314 	if (src_entry->wired_count == 0) {
2315 
2316 		/*
2317 		 * If the source entry is marked needs_copy, it is already
2318 		 * write-protected.
2319 		 */
2320 		if ((src_entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) {
2321 			pmap_protect(src_map->pmap,
2322 			    src_entry->start,
2323 			    src_entry->end,
2324 			    src_entry->protection & ~VM_PROT_WRITE);
2325 		}
2326 
2327 		/*
2328 		 * Make a copy of the object.
2329 		 */
2330 		if ((src_object = src_entry->object.vm_object) != NULL) {
2331 
2332 			if ((src_object->handle == NULL) &&
2333 				(src_object->type == OBJT_DEFAULT ||
2334 				 src_object->type == OBJT_SWAP)) {
2335 				vm_object_collapse(src_object);
2336 				if ((src_object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING) {
2337 					vm_map_split(src_entry);
2338 					src_object = src_entry->object.vm_object;
2339 				}
2340 			}
2341 
2342 			vm_object_reference(src_object);
2343 			vm_object_clear_flag(src_object, OBJ_ONEMAPPING);
2344 			dst_entry->object.vm_object = src_object;
2345 			src_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY);
2346 			dst_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY);
2347 			dst_entry->offset = src_entry->offset;
2348 		} else {
2349 			dst_entry->object.vm_object = NULL;
2350 			dst_entry->offset = 0;
2351 		}
2352 
2353 		pmap_copy(dst_map->pmap, src_map->pmap, dst_entry->start,
2354 		    dst_entry->end - dst_entry->start, src_entry->start);
2355 	} else {
2356 		/*
2357 		 * Of course, wired down pages can't be set copy-on-write.
2358 		 * Cause wired pages to be copied into the new map by
2359 		 * simulating faults (the new pages are pageable)
2360 		 */
2361 		vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry);
2362 	}
2363 }
2364 
2365 /*
2366  * vmspace_fork:
2367  * Create a new process vmspace structure and vm_map
2368  * based on those of an existing process.  The new map
2369  * is based on the old map, according to the inheritance
2370  * values on the regions in that map.
2371  *
2372  * The source map must not be locked.
2373  */
2374 struct vmspace *
2375 vmspace_fork(struct vmspace *vm1)
2376 {
2377 	struct vmspace *vm2;
2378 	vm_map_t old_map = &vm1->vm_map;
2379 	vm_map_t new_map;
2380 	vm_map_entry_t old_entry;
2381 	vm_map_entry_t new_entry;
2382 	vm_object_t object;
2383 
2384 	GIANT_REQUIRED;
2385 
2386 	vm_map_lock(old_map);
2387 	old_map->infork = 1;
2388 
2389 	vm2 = vmspace_alloc(old_map->min_offset, old_map->max_offset);
2390 	bcopy(&vm1->vm_startcopy, &vm2->vm_startcopy,
2391 	    (caddr_t) &vm1->vm_endcopy - (caddr_t) &vm1->vm_startcopy);
2392 	new_map = &vm2->vm_map;	/* XXX */
2393 	new_map->timestamp = 1;
2394 
2395 	old_entry = old_map->header.next;
2396 
2397 	while (old_entry != &old_map->header) {
2398 		if (old_entry->eflags & MAP_ENTRY_IS_SUB_MAP)
2399 			panic("vm_map_fork: encountered a submap");
2400 
2401 		switch (old_entry->inheritance) {
2402 		case VM_INHERIT_NONE:
2403 			break;
2404 
2405 		case VM_INHERIT_SHARE:
2406 			/*
2407 			 * Clone the entry, creating the shared object if necessary.
2408 			 */
2409 			object = old_entry->object.vm_object;
2410 			if (object == NULL) {
2411 				object = vm_object_allocate(OBJT_DEFAULT,
2412 					atop(old_entry->end - old_entry->start));
2413 				old_entry->object.vm_object = object;
2414 				old_entry->offset = (vm_offset_t) 0;
2415 			}
2416 
2417 			/*
2418 			 * Add the reference before calling vm_object_shadow
2419 			 * to insure that a shadow object is created.
2420 			 */
2421 			vm_object_reference(object);
2422 			if (old_entry->eflags & MAP_ENTRY_NEEDS_COPY) {
2423 				vm_object_shadow(&old_entry->object.vm_object,
2424 					&old_entry->offset,
2425 					atop(old_entry->end - old_entry->start));
2426 				old_entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
2427 				/* Transfer the second reference too. */
2428 				vm_object_reference(
2429 				    old_entry->object.vm_object);
2430 				vm_object_deallocate(object);
2431 				object = old_entry->object.vm_object;
2432 			}
2433 			vm_object_clear_flag(object, OBJ_ONEMAPPING);
2434 
2435 			/*
2436 			 * Clone the entry, referencing the shared object.
2437 			 */
2438 			new_entry = vm_map_entry_create(new_map);
2439 			*new_entry = *old_entry;
2440 			new_entry->eflags &= ~MAP_ENTRY_USER_WIRED;
2441 			new_entry->wired_count = 0;
2442 
2443 			/*
2444 			 * Insert the entry into the new map -- we know we're
2445 			 * inserting at the end of the new map.
2446 			 */
2447 			vm_map_entry_link(new_map, new_map->header.prev,
2448 			    new_entry);
2449 
2450 			/*
2451 			 * Update the physical map
2452 			 */
2453 			pmap_copy(new_map->pmap, old_map->pmap,
2454 			    new_entry->start,
2455 			    (old_entry->end - old_entry->start),
2456 			    old_entry->start);
2457 			break;
2458 
2459 		case VM_INHERIT_COPY:
2460 			/*
2461 			 * Clone the entry and link into the map.
2462 			 */
2463 			new_entry = vm_map_entry_create(new_map);
2464 			*new_entry = *old_entry;
2465 			new_entry->eflags &= ~MAP_ENTRY_USER_WIRED;
2466 			new_entry->wired_count = 0;
2467 			new_entry->object.vm_object = NULL;
2468 			vm_map_entry_link(new_map, new_map->header.prev,
2469 			    new_entry);
2470 			vm_map_copy_entry(old_map, new_map, old_entry,
2471 			    new_entry);
2472 			break;
2473 		}
2474 		old_entry = old_entry->next;
2475 	}
2476 
2477 	new_map->size = old_map->size;
2478 	old_map->infork = 0;
2479 	vm_map_unlock(old_map);
2480 
2481 	return (vm2);
2482 }
2483 
2484 int
2485 vm_map_stack (vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
2486 	      vm_prot_t prot, vm_prot_t max, int cow)
2487 {
2488 	vm_map_entry_t prev_entry;
2489 	vm_map_entry_t new_stack_entry;
2490 	vm_size_t      init_ssize;
2491 	int            rv;
2492 
2493 	GIANT_REQUIRED;
2494 
2495 	if (VM_MIN_ADDRESS > 0 && addrbos < VM_MIN_ADDRESS)
2496 		return (KERN_NO_SPACE);
2497 
2498 	if (max_ssize < sgrowsiz)
2499 		init_ssize = max_ssize;
2500 	else
2501 		init_ssize = sgrowsiz;
2502 
2503 	vm_map_lock(map);
2504 
2505 	/* If addr is already mapped, no go */
2506 	if (vm_map_lookup_entry(map, addrbos, &prev_entry)) {
2507 		vm_map_unlock(map);
2508 		return (KERN_NO_SPACE);
2509 	}
2510 
2511 	/* If we can't accomodate max_ssize in the current mapping,
2512 	 * no go.  However, we need to be aware that subsequent user
2513 	 * mappings might map into the space we have reserved for
2514 	 * stack, and currently this space is not protected.
2515 	 *
2516 	 * Hopefully we will at least detect this condition
2517 	 * when we try to grow the stack.
2518 	 */
2519 	if ((prev_entry->next != &map->header) &&
2520 	    (prev_entry->next->start < addrbos + max_ssize)) {
2521 		vm_map_unlock(map);
2522 		return (KERN_NO_SPACE);
2523 	}
2524 
2525 	/* We initially map a stack of only init_ssize.  We will
2526 	 * grow as needed later.  Since this is to be a grow
2527 	 * down stack, we map at the top of the range.
2528 	 *
2529 	 * Note: we would normally expect prot and max to be
2530 	 * VM_PROT_ALL, and cow to be 0.  Possibly we should
2531 	 * eliminate these as input parameters, and just
2532 	 * pass these values here in the insert call.
2533 	 */
2534 	rv = vm_map_insert(map, NULL, 0, addrbos + max_ssize - init_ssize,
2535 	                   addrbos + max_ssize, prot, max, cow);
2536 
2537 	/* Now set the avail_ssize amount */
2538 	if (rv == KERN_SUCCESS){
2539 		if (prev_entry != &map->header)
2540 			vm_map_clip_end(map, prev_entry, addrbos + max_ssize - init_ssize);
2541 		new_stack_entry = prev_entry->next;
2542 		if (new_stack_entry->end   != addrbos + max_ssize ||
2543 		    new_stack_entry->start != addrbos + max_ssize - init_ssize)
2544 			panic ("Bad entry start/end for new stack entry");
2545 		else
2546 			new_stack_entry->avail_ssize = max_ssize - init_ssize;
2547 	}
2548 
2549 	vm_map_unlock(map);
2550 	return (rv);
2551 }
2552 
2553 /* Attempts to grow a vm stack entry.  Returns KERN_SUCCESS if the
2554  * desired address is already mapped, or if we successfully grow
2555  * the stack.  Also returns KERN_SUCCESS if addr is outside the
2556  * stack range (this is strange, but preserves compatibility with
2557  * the grow function in vm_machdep.c).
2558  */
2559 int
2560 vm_map_growstack (struct proc *p, vm_offset_t addr)
2561 {
2562 	vm_map_entry_t prev_entry;
2563 	vm_map_entry_t stack_entry;
2564 	vm_map_entry_t new_stack_entry;
2565 	struct vmspace *vm = p->p_vmspace;
2566 	vm_map_t map = &vm->vm_map;
2567 	vm_offset_t    end;
2568 	int      grow_amount;
2569 	int      rv;
2570 	int      is_procstack;
2571 
2572 	GIANT_REQUIRED;
2573 
2574 Retry:
2575 	vm_map_lock_read(map);
2576 
2577 	/* If addr is already in the entry range, no need to grow.*/
2578 	if (vm_map_lookup_entry(map, addr, &prev_entry)) {
2579 		vm_map_unlock_read(map);
2580 		return (KERN_SUCCESS);
2581 	}
2582 
2583 	if ((stack_entry = prev_entry->next) == &map->header) {
2584 		vm_map_unlock_read(map);
2585 		return (KERN_SUCCESS);
2586 	}
2587 	if (prev_entry == &map->header)
2588 		end = stack_entry->start - stack_entry->avail_ssize;
2589 	else
2590 		end = prev_entry->end;
2591 
2592 	/* This next test mimics the old grow function in vm_machdep.c.
2593 	 * It really doesn't quite make sense, but we do it anyway
2594 	 * for compatibility.
2595 	 *
2596 	 * If not growable stack, return success.  This signals the
2597 	 * caller to proceed as he would normally with normal vm.
2598 	 */
2599 	if (stack_entry->avail_ssize < 1 ||
2600 	    addr >= stack_entry->start ||
2601 	    addr <  stack_entry->start - stack_entry->avail_ssize) {
2602 		vm_map_unlock_read(map);
2603 		return (KERN_SUCCESS);
2604 	}
2605 
2606 	/* Find the minimum grow amount */
2607 	grow_amount = roundup (stack_entry->start - addr, PAGE_SIZE);
2608 	if (grow_amount > stack_entry->avail_ssize) {
2609 		vm_map_unlock_read(map);
2610 		return (KERN_NO_SPACE);
2611 	}
2612 
2613 	/* If there is no longer enough space between the entries
2614 	 * nogo, and adjust the available space.  Note: this
2615 	 * should only happen if the user has mapped into the
2616 	 * stack area after the stack was created, and is
2617 	 * probably an error.
2618 	 *
2619 	 * This also effectively destroys any guard page the user
2620 	 * might have intended by limiting the stack size.
2621 	 */
2622 	if (grow_amount > stack_entry->start - end) {
2623 		if (vm_map_lock_upgrade(map))
2624 			goto Retry;
2625 
2626 		stack_entry->avail_ssize = stack_entry->start - end;
2627 
2628 		vm_map_unlock(map);
2629 		return (KERN_NO_SPACE);
2630 	}
2631 
2632 	is_procstack = addr >= (vm_offset_t)vm->vm_maxsaddr;
2633 
2634 	/* If this is the main process stack, see if we're over the
2635 	 * stack limit.
2636 	 */
2637 	if (is_procstack && (ctob(vm->vm_ssize) + grow_amount >
2638 			     p->p_rlimit[RLIMIT_STACK].rlim_cur)) {
2639 		vm_map_unlock_read(map);
2640 		return (KERN_NO_SPACE);
2641 	}
2642 
2643 	/* Round up the grow amount modulo SGROWSIZ */
2644 	grow_amount = roundup (grow_amount, sgrowsiz);
2645 	if (grow_amount > stack_entry->avail_ssize) {
2646 		grow_amount = stack_entry->avail_ssize;
2647 	}
2648 	if (is_procstack && (ctob(vm->vm_ssize) + grow_amount >
2649 	                     p->p_rlimit[RLIMIT_STACK].rlim_cur)) {
2650 		grow_amount = p->p_rlimit[RLIMIT_STACK].rlim_cur -
2651 		              ctob(vm->vm_ssize);
2652 	}
2653 
2654 	if (vm_map_lock_upgrade(map))
2655 		goto Retry;
2656 
2657 	/* Get the preliminary new entry start value */
2658 	addr = stack_entry->start - grow_amount;
2659 
2660 	/* If this puts us into the previous entry, cut back our growth
2661 	 * to the available space.  Also, see the note above.
2662 	 */
2663 	if (addr < end) {
2664 		stack_entry->avail_ssize = stack_entry->start - end;
2665 		addr = end;
2666 	}
2667 
2668 	rv = vm_map_insert(map, NULL, 0, addr, stack_entry->start,
2669 			   VM_PROT_ALL,
2670 			   VM_PROT_ALL,
2671 			   0);
2672 
2673 	/* Adjust the available stack space by the amount we grew. */
2674 	if (rv == KERN_SUCCESS) {
2675 		if (prev_entry != &map->header)
2676 			vm_map_clip_end(map, prev_entry, addr);
2677 		new_stack_entry = prev_entry->next;
2678 		if (new_stack_entry->end   != stack_entry->start  ||
2679 		    new_stack_entry->start != addr)
2680 			panic ("Bad stack grow start/end in new stack entry");
2681 		else {
2682 			new_stack_entry->avail_ssize = stack_entry->avail_ssize -
2683 							(new_stack_entry->end -
2684 							 new_stack_entry->start);
2685 			if (is_procstack)
2686 				vm->vm_ssize += btoc(new_stack_entry->end -
2687 						     new_stack_entry->start);
2688 		}
2689 	}
2690 
2691 	vm_map_unlock(map);
2692 	return (rv);
2693 }
2694 
2695 /*
2696  * Unshare the specified VM space for exec.  If other processes are
2697  * mapped to it, then create a new one.  The new vmspace is null.
2698  */
2699 void
2700 vmspace_exec(struct proc *p)
2701 {
2702 	struct vmspace *oldvmspace = p->p_vmspace;
2703 	struct vmspace *newvmspace;
2704 	vm_map_t map = &p->p_vmspace->vm_map;
2705 
2706 	GIANT_REQUIRED;
2707 	newvmspace = vmspace_alloc(map->min_offset, map->max_offset);
2708 	bcopy(&oldvmspace->vm_startcopy, &newvmspace->vm_startcopy,
2709 	    (caddr_t) (newvmspace + 1) - (caddr_t) &newvmspace->vm_startcopy);
2710 	/*
2711 	 * This code is written like this for prototype purposes.  The
2712 	 * goal is to avoid running down the vmspace here, but let the
2713 	 * other process's that are still using the vmspace to finally
2714 	 * run it down.  Even though there is little or no chance of blocking
2715 	 * here, it is a good idea to keep this form for future mods.
2716 	 */
2717 	p->p_vmspace = newvmspace;
2718 	pmap_pinit2(vmspace_pmap(newvmspace));
2719 	vmspace_free(oldvmspace);
2720 	if (p == curthread->td_proc)		/* XXXKSE ? */
2721 		pmap_activate(curthread);
2722 }
2723 
2724 /*
2725  * Unshare the specified VM space for forcing COW.  This
2726  * is called by rfork, for the (RFMEM|RFPROC) == 0 case.
2727  */
2728 void
2729 vmspace_unshare(struct proc *p)
2730 {
2731 	struct vmspace *oldvmspace = p->p_vmspace;
2732 	struct vmspace *newvmspace;
2733 
2734 	GIANT_REQUIRED;
2735 	if (oldvmspace->vm_refcnt == 1)
2736 		return;
2737 	newvmspace = vmspace_fork(oldvmspace);
2738 	p->p_vmspace = newvmspace;
2739 	pmap_pinit2(vmspace_pmap(newvmspace));
2740 	vmspace_free(oldvmspace);
2741 	if (p == curthread->td_proc)		/* XXXKSE ? */
2742 		pmap_activate(curthread);
2743 }
2744 
2745 /*
2746  *	vm_map_lookup:
2747  *
2748  *	Finds the VM object, offset, and
2749  *	protection for a given virtual address in the
2750  *	specified map, assuming a page fault of the
2751  *	type specified.
2752  *
2753  *	Leaves the map in question locked for read; return
2754  *	values are guaranteed until a vm_map_lookup_done
2755  *	call is performed.  Note that the map argument
2756  *	is in/out; the returned map must be used in
2757  *	the call to vm_map_lookup_done.
2758  *
2759  *	A handle (out_entry) is returned for use in
2760  *	vm_map_lookup_done, to make that fast.
2761  *
2762  *	If a lookup is requested with "write protection"
2763  *	specified, the map may be changed to perform virtual
2764  *	copying operations, although the data referenced will
2765  *	remain the same.
2766  */
2767 int
2768 vm_map_lookup(vm_map_t *var_map,		/* IN/OUT */
2769 	      vm_offset_t vaddr,
2770 	      vm_prot_t fault_typea,
2771 	      vm_map_entry_t *out_entry,	/* OUT */
2772 	      vm_object_t *object,		/* OUT */
2773 	      vm_pindex_t *pindex,		/* OUT */
2774 	      vm_prot_t *out_prot,		/* OUT */
2775 	      boolean_t *wired)			/* OUT */
2776 {
2777 	vm_map_entry_t entry;
2778 	vm_map_t map = *var_map;
2779 	vm_prot_t prot;
2780 	vm_prot_t fault_type = fault_typea;
2781 
2782 	GIANT_REQUIRED;
2783 RetryLookup:;
2784 	/*
2785 	 * Lookup the faulting address.
2786 	 */
2787 
2788 	vm_map_lock_read(map);
2789 #define	RETURN(why) \
2790 		{ \
2791 		vm_map_unlock_read(map); \
2792 		return (why); \
2793 		}
2794 
2795 	/*
2796 	 * If the map has an interesting hint, try it before calling full
2797 	 * blown lookup routine.
2798 	 */
2799 	entry = map->hint;
2800 	*out_entry = entry;
2801 	if ((entry == &map->header) ||
2802 	    (vaddr < entry->start) || (vaddr >= entry->end)) {
2803 		vm_map_entry_t tmp_entry;
2804 
2805 		/*
2806 		 * Entry was either not a valid hint, or the vaddr was not
2807 		 * contained in the entry, so do a full lookup.
2808 		 */
2809 		if (!vm_map_lookup_entry(map, vaddr, &tmp_entry))
2810 			RETURN(KERN_INVALID_ADDRESS);
2811 
2812 		entry = tmp_entry;
2813 		*out_entry = entry;
2814 	}
2815 
2816 	/*
2817 	 * Handle submaps.
2818 	 */
2819 	if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
2820 		vm_map_t old_map = map;
2821 
2822 		*var_map = map = entry->object.sub_map;
2823 		vm_map_unlock_read(old_map);
2824 		goto RetryLookup;
2825 	}
2826 
2827 	/*
2828 	 * Check whether this task is allowed to have this page.
2829 	 * Note the special case for MAP_ENTRY_COW
2830 	 * pages with an override.  This is to implement a forced
2831 	 * COW for debuggers.
2832 	 */
2833 	if (fault_type & VM_PROT_OVERRIDE_WRITE)
2834 		prot = entry->max_protection;
2835 	else
2836 		prot = entry->protection;
2837 	fault_type &= (VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE);
2838 	if ((fault_type & prot) != fault_type) {
2839 			RETURN(KERN_PROTECTION_FAILURE);
2840 	}
2841 	if ((entry->eflags & MAP_ENTRY_USER_WIRED) &&
2842 	    (entry->eflags & MAP_ENTRY_COW) &&
2843 	    (fault_type & VM_PROT_WRITE) &&
2844 	    (fault_typea & VM_PROT_OVERRIDE_WRITE) == 0) {
2845 		RETURN(KERN_PROTECTION_FAILURE);
2846 	}
2847 
2848 	/*
2849 	 * If this page is not pageable, we have to get it for all possible
2850 	 * accesses.
2851 	 */
2852 	*wired = (entry->wired_count != 0);
2853 	if (*wired)
2854 		prot = fault_type = entry->protection;
2855 
2856 	/*
2857 	 * If the entry was copy-on-write, we either ...
2858 	 */
2859 	if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
2860 		/*
2861 		 * If we want to write the page, we may as well handle that
2862 		 * now since we've got the map locked.
2863 		 *
2864 		 * If we don't need to write the page, we just demote the
2865 		 * permissions allowed.
2866 		 */
2867 		if (fault_type & VM_PROT_WRITE) {
2868 			/*
2869 			 * Make a new object, and place it in the object
2870 			 * chain.  Note that no new references have appeared
2871 			 * -- one just moved from the map to the new
2872 			 * object.
2873 			 */
2874 			if (vm_map_lock_upgrade(map))
2875 				goto RetryLookup;
2876 			vm_object_shadow(
2877 			    &entry->object.vm_object,
2878 			    &entry->offset,
2879 			    atop(entry->end - entry->start));
2880 			entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
2881 			vm_map_lock_downgrade(map);
2882 		} else {
2883 			/*
2884 			 * We're attempting to read a copy-on-write page --
2885 			 * don't allow writes.
2886 			 */
2887 			prot &= ~VM_PROT_WRITE;
2888 		}
2889 	}
2890 
2891 	/*
2892 	 * Create an object if necessary.
2893 	 */
2894 	if (entry->object.vm_object == NULL &&
2895 	    !map->system_map) {
2896 		if (vm_map_lock_upgrade(map))
2897 			goto RetryLookup;
2898 		entry->object.vm_object = vm_object_allocate(OBJT_DEFAULT,
2899 		    atop(entry->end - entry->start));
2900 		entry->offset = 0;
2901 		vm_map_lock_downgrade(map);
2902 	}
2903 
2904 	/*
2905 	 * Return the object/offset from this entry.  If the entry was
2906 	 * copy-on-write or empty, it has been fixed up.
2907 	 */
2908 	*pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
2909 	*object = entry->object.vm_object;
2910 
2911 	/*
2912 	 * Return whether this is the only map sharing this data.
2913 	 */
2914 	*out_prot = prot;
2915 	return (KERN_SUCCESS);
2916 
2917 #undef	RETURN
2918 }
2919 
2920 /*
2921  *	vm_map_lookup_done:
2922  *
2923  *	Releases locks acquired by a vm_map_lookup
2924  *	(according to the handle returned by that lookup).
2925  */
2926 void
2927 vm_map_lookup_done(vm_map_t map, vm_map_entry_t entry)
2928 {
2929 	/*
2930 	 * Unlock the main-level map
2931 	 */
2932 	GIANT_REQUIRED;
2933 	vm_map_unlock_read(map);
2934 }
2935 
2936 /*
2937  * Implement uiomove with VM operations.  This handles (and collateral changes)
2938  * support every combination of source object modification, and COW type
2939  * operations.
2940  */
2941 int
2942 vm_uiomove(
2943 	vm_map_t mapa,
2944 	vm_object_t srcobject,
2945 	off_t cp,
2946 	int cnta,
2947 	vm_offset_t uaddra,
2948 	int *npages)
2949 {
2950 	vm_map_t map;
2951 	vm_object_t first_object, oldobject, object;
2952 	vm_map_entry_t entry;
2953 	vm_prot_t prot;
2954 	boolean_t wired;
2955 	int tcnt, rv;
2956 	vm_offset_t uaddr, start, end, tend;
2957 	vm_pindex_t first_pindex, osize, oindex;
2958 	off_t ooffset;
2959 	int cnt;
2960 
2961 	GIANT_REQUIRED;
2962 
2963 	if (npages)
2964 		*npages = 0;
2965 
2966 	cnt = cnta;
2967 	uaddr = uaddra;
2968 
2969 	while (cnt > 0) {
2970 		map = mapa;
2971 
2972 		if ((vm_map_lookup(&map, uaddr,
2973 			VM_PROT_READ, &entry, &first_object,
2974 			&first_pindex, &prot, &wired)) != KERN_SUCCESS) {
2975 			return EFAULT;
2976 		}
2977 
2978 		vm_map_clip_start(map, entry, uaddr);
2979 
2980 		tcnt = cnt;
2981 		tend = uaddr + tcnt;
2982 		if (tend > entry->end) {
2983 			tcnt = entry->end - uaddr;
2984 			tend = entry->end;
2985 		}
2986 
2987 		vm_map_clip_end(map, entry, tend);
2988 
2989 		start = entry->start;
2990 		end = entry->end;
2991 
2992 		osize = atop(tcnt);
2993 
2994 		oindex = OFF_TO_IDX(cp);
2995 		if (npages) {
2996 			vm_pindex_t idx;
2997 			for (idx = 0; idx < osize; idx++) {
2998 				vm_page_t m;
2999 				if ((m = vm_page_lookup(srcobject, oindex + idx)) == NULL) {
3000 					vm_map_lookup_done(map, entry);
3001 					return 0;
3002 				}
3003 				/*
3004 				 * disallow busy or invalid pages, but allow
3005 				 * m->busy pages if they are entirely valid.
3006 				 */
3007 				if ((m->flags & PG_BUSY) ||
3008 					((m->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL)) {
3009 					vm_map_lookup_done(map, entry);
3010 					return 0;
3011 				}
3012 			}
3013 		}
3014 
3015 /*
3016  * If we are changing an existing map entry, just redirect
3017  * the object, and change mappings.
3018  */
3019 		if ((first_object->type == OBJT_VNODE) &&
3020 			((oldobject = entry->object.vm_object) == first_object)) {
3021 
3022 			if ((entry->offset != cp) || (oldobject != srcobject)) {
3023 				/*
3024    				* Remove old window into the file
3025    				*/
3026 				pmap_remove (map->pmap, uaddr, tend);
3027 
3028 				/*
3029    				* Force copy on write for mmaped regions
3030    				*/
3031 				vm_object_pmap_copy_1 (srcobject, oindex, oindex + osize);
3032 
3033 				/*
3034    				* Point the object appropriately
3035    				*/
3036 				if (oldobject != srcobject) {
3037 
3038 				/*
3039    				* Set the object optimization hint flag
3040    				*/
3041 					vm_object_set_flag(srcobject, OBJ_OPT);
3042 					vm_object_reference(srcobject);
3043 					entry->object.vm_object = srcobject;
3044 
3045 					if (oldobject) {
3046 						vm_object_deallocate(oldobject);
3047 					}
3048 				}
3049 
3050 				entry->offset = cp;
3051 				map->timestamp++;
3052 			} else {
3053 				pmap_remove (map->pmap, uaddr, tend);
3054 			}
3055 
3056 		} else if ((first_object->ref_count == 1) &&
3057 			(first_object->size == osize) &&
3058 			((first_object->type == OBJT_DEFAULT) ||
3059 				(first_object->type == OBJT_SWAP)) ) {
3060 
3061 			oldobject = first_object->backing_object;
3062 
3063 			if ((first_object->backing_object_offset != cp) ||
3064 				(oldobject != srcobject)) {
3065 				/*
3066    				* Remove old window into the file
3067    				*/
3068 				pmap_remove (map->pmap, uaddr, tend);
3069 
3070 				/*
3071 				 * Remove unneeded old pages
3072 				 */
3073 				vm_object_page_remove(first_object, 0, 0, 0);
3074 
3075 				/*
3076 				 * Invalidate swap space
3077 				 */
3078 				if (first_object->type == OBJT_SWAP) {
3079 					swap_pager_freespace(first_object,
3080 						0,
3081 						first_object->size);
3082 				}
3083 
3084 				/*
3085    				 * Force copy on write for mmaped regions
3086    				 */
3087 				vm_object_pmap_copy_1 (srcobject, oindex, oindex + osize);
3088 
3089 				/*
3090    				 * Point the object appropriately
3091    				 */
3092 				if (oldobject != srcobject) {
3093 					/*
3094    					 * Set the object optimization hint flag
3095    					 */
3096 					vm_object_set_flag(srcobject, OBJ_OPT);
3097 					vm_object_reference(srcobject);
3098 
3099 					if (oldobject) {
3100 						TAILQ_REMOVE(&oldobject->shadow_head,
3101 							first_object, shadow_list);
3102 						oldobject->shadow_count--;
3103 						/* XXX bump generation? */
3104 						vm_object_deallocate(oldobject);
3105 					}
3106 
3107 					TAILQ_INSERT_TAIL(&srcobject->shadow_head,
3108 						first_object, shadow_list);
3109 					srcobject->shadow_count++;
3110 					/* XXX bump generation? */
3111 
3112 					first_object->backing_object = srcobject;
3113 				}
3114 				first_object->backing_object_offset = cp;
3115 				map->timestamp++;
3116 			} else {
3117 				pmap_remove (map->pmap, uaddr, tend);
3118 			}
3119 /*
3120  * Otherwise, we have to do a logical mmap.
3121  */
3122 		} else {
3123 
3124 			vm_object_set_flag(srcobject, OBJ_OPT);
3125 			vm_object_reference(srcobject);
3126 
3127 			pmap_remove (map->pmap, uaddr, tend);
3128 
3129 			vm_object_pmap_copy_1 (srcobject, oindex, oindex + osize);
3130 			vm_map_lock_upgrade(map);
3131 
3132 			if (entry == &map->header) {
3133 				map->first_free = &map->header;
3134 			} else if (map->first_free->start >= start) {
3135 				map->first_free = entry->prev;
3136 			}
3137 
3138 			SAVE_HINT(map, entry->prev);
3139 			vm_map_entry_delete(map, entry);
3140 
3141 			object = srcobject;
3142 			ooffset = cp;
3143 
3144 			rv = vm_map_insert(map, object, ooffset, start, tend,
3145 				VM_PROT_ALL, VM_PROT_ALL, MAP_COPY_ON_WRITE);
3146 
3147 			if (rv != KERN_SUCCESS)
3148 				panic("vm_uiomove: could not insert new entry: %d", rv);
3149 		}
3150 
3151 /*
3152  * Map the window directly, if it is already in memory
3153  */
3154 		pmap_object_init_pt(map->pmap, uaddr,
3155 			srcobject, oindex, tcnt, 0);
3156 
3157 		map->timestamp++;
3158 		vm_map_unlock(map);
3159 
3160 		cnt -= tcnt;
3161 		uaddr += tcnt;
3162 		cp += tcnt;
3163 		if (npages)
3164 			*npages += osize;
3165 	}
3166 	return 0;
3167 }
3168 
3169 /*
3170  * Performs the copy_on_write operations necessary to allow the virtual copies
3171  * into user space to work.  This has to be called for write(2) system calls
3172  * from other processes, file unlinking, and file size shrinkage.
3173  */
3174 void
3175 vm_freeze_copyopts(vm_object_t object, vm_pindex_t froma, vm_pindex_t toa)
3176 {
3177 	int rv;
3178 	vm_object_t robject;
3179 	vm_pindex_t idx;
3180 
3181 	GIANT_REQUIRED;
3182 	if ((object == NULL) ||
3183 		((object->flags & OBJ_OPT) == 0))
3184 		return;
3185 
3186 	if (object->shadow_count > object->ref_count)
3187 		panic("vm_freeze_copyopts: sc > rc");
3188 
3189 	while ((robject = TAILQ_FIRST(&object->shadow_head)) != NULL) {
3190 		vm_pindex_t bo_pindex;
3191 		vm_page_t m_in, m_out;
3192 
3193 		bo_pindex = OFF_TO_IDX(robject->backing_object_offset);
3194 
3195 		vm_object_reference(robject);
3196 
3197 		vm_object_pip_wait(robject, "objfrz");
3198 
3199 		if (robject->ref_count == 1) {
3200 			vm_object_deallocate(robject);
3201 			continue;
3202 		}
3203 
3204 		vm_object_pip_add(robject, 1);
3205 
3206 		for (idx = 0; idx < robject->size; idx++) {
3207 
3208 			m_out = vm_page_grab(robject, idx,
3209 						VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
3210 
3211 			if (m_out->valid == 0) {
3212 				m_in = vm_page_grab(object, bo_pindex + idx,
3213 						VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
3214 				if (m_in->valid == 0) {
3215 					rv = vm_pager_get_pages(object, &m_in, 1, 0);
3216 					if (rv != VM_PAGER_OK) {
3217 						printf("vm_freeze_copyopts: cannot read page from file: %lx\n", (long)m_in->pindex);
3218 						continue;
3219 					}
3220 					vm_page_deactivate(m_in);
3221 				}
3222 
3223 				vm_page_protect(m_in, VM_PROT_NONE);
3224 				pmap_copy_page(VM_PAGE_TO_PHYS(m_in), VM_PAGE_TO_PHYS(m_out));
3225 				m_out->valid = m_in->valid;
3226 				vm_page_dirty(m_out);
3227 				vm_page_activate(m_out);
3228 				vm_page_wakeup(m_in);
3229 			}
3230 			vm_page_wakeup(m_out);
3231 		}
3232 
3233 		object->shadow_count--;
3234 		object->ref_count--;
3235 		TAILQ_REMOVE(&object->shadow_head, robject, shadow_list);
3236 		robject->backing_object = NULL;
3237 		robject->backing_object_offset = 0;
3238 
3239 		vm_object_pip_wakeup(robject);
3240 		vm_object_deallocate(robject);
3241 	}
3242 
3243 	vm_object_clear_flag(object, OBJ_OPT);
3244 }
3245 
3246 #include "opt_ddb.h"
3247 #ifdef DDB
3248 #include <sys/kernel.h>
3249 
3250 #include <ddb/ddb.h>
3251 
3252 /*
3253  *	vm_map_print:	[ debug ]
3254  */
3255 DB_SHOW_COMMAND(map, vm_map_print)
3256 {
3257 	static int nlines;
3258 	/* XXX convert args. */
3259 	vm_map_t map = (vm_map_t)addr;
3260 	boolean_t full = have_addr;
3261 
3262 	vm_map_entry_t entry;
3263 
3264 	db_iprintf("Task map %p: pmap=%p, nentries=%d, version=%u\n",
3265 	    (void *)map,
3266 	    (void *)map->pmap, map->nentries, map->timestamp);
3267 	nlines++;
3268 
3269 	if (!full && db_indent)
3270 		return;
3271 
3272 	db_indent += 2;
3273 	for (entry = map->header.next; entry != &map->header;
3274 	    entry = entry->next) {
3275 		db_iprintf("map entry %p: start=%p, end=%p\n",
3276 		    (void *)entry, (void *)entry->start, (void *)entry->end);
3277 		nlines++;
3278 		{
3279 			static char *inheritance_name[4] =
3280 			{"share", "copy", "none", "donate_copy"};
3281 
3282 			db_iprintf(" prot=%x/%x/%s",
3283 			    entry->protection,
3284 			    entry->max_protection,
3285 			    inheritance_name[(int)(unsigned char)entry->inheritance]);
3286 			if (entry->wired_count != 0)
3287 				db_printf(", wired");
3288 		}
3289 		if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
3290 			/* XXX no %qd in kernel.  Truncate entry->offset. */
3291 			db_printf(", share=%p, offset=0x%lx\n",
3292 			    (void *)entry->object.sub_map,
3293 			    (long)entry->offset);
3294 			nlines++;
3295 			if ((entry->prev == &map->header) ||
3296 			    (entry->prev->object.sub_map !=
3297 				entry->object.sub_map)) {
3298 				db_indent += 2;
3299 				vm_map_print((db_expr_t)(intptr_t)
3300 					     entry->object.sub_map,
3301 					     full, 0, (char *)0);
3302 				db_indent -= 2;
3303 			}
3304 		} else {
3305 			/* XXX no %qd in kernel.  Truncate entry->offset. */
3306 			db_printf(", object=%p, offset=0x%lx",
3307 			    (void *)entry->object.vm_object,
3308 			    (long)entry->offset);
3309 			if (entry->eflags & MAP_ENTRY_COW)
3310 				db_printf(", copy (%s)",
3311 				    (entry->eflags & MAP_ENTRY_NEEDS_COPY) ? "needed" : "done");
3312 			db_printf("\n");
3313 			nlines++;
3314 
3315 			if ((entry->prev == &map->header) ||
3316 			    (entry->prev->object.vm_object !=
3317 				entry->object.vm_object)) {
3318 				db_indent += 2;
3319 				vm_object_print((db_expr_t)(intptr_t)
3320 						entry->object.vm_object,
3321 						full, 0, (char *)0);
3322 				nlines += 4;
3323 				db_indent -= 2;
3324 			}
3325 		}
3326 	}
3327 	db_indent -= 2;
3328 	if (db_indent == 0)
3329 		nlines = 0;
3330 }
3331 
3332 
3333 DB_SHOW_COMMAND(procvm, procvm)
3334 {
3335 	struct proc *p;
3336 
3337 	if (have_addr) {
3338 		p = (struct proc *) addr;
3339 	} else {
3340 		p = curproc;
3341 	}
3342 
3343 	db_printf("p = %p, vmspace = %p, map = %p, pmap = %p\n",
3344 	    (void *)p, (void *)p->p_vmspace, (void *)&p->p_vmspace->vm_map,
3345 	    (void *)vmspace_pmap(p->p_vmspace));
3346 
3347 	vm_map_print((db_expr_t)(intptr_t)&p->p_vmspace->vm_map, 1, 0, NULL);
3348 }
3349 
3350 #endif /* DDB */
3351