xref: /freebsd/sys/vm/vm_map.c (revision f856af0466c076beef4ea9b15d088e1119a945b8)
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  * 4. Neither the name of the University nor the names of its contributors
17  *    may be used to endorse or promote products derived from this software
18  *    without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  *
32  *	from: @(#)vm_map.c	8.3 (Berkeley) 1/12/94
33  *
34  *
35  * Copyright (c) 1987, 1990 Carnegie-Mellon University.
36  * All rights reserved.
37  *
38  * Authors: Avadis Tevanian, Jr., Michael Wayne Young
39  *
40  * Permission to use, copy, modify and distribute this software and
41  * its documentation is hereby granted, provided that both the copyright
42  * notice and this permission notice appear in all copies of the
43  * software, derivative works or modified versions, and any portions
44  * thereof, and that both notices appear in supporting documentation.
45  *
46  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
47  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
48  * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
49  *
50  * Carnegie Mellon requests users of this software to return to
51  *
52  *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
53  *  School of Computer Science
54  *  Carnegie Mellon University
55  *  Pittsburgh PA 15213-3890
56  *
57  * any improvements or extensions that they make and grant Carnegie the
58  * rights to redistribute these changes.
59  */
60 
61 /*
62  *	Virtual memory mapping module.
63  */
64 
65 #include <sys/cdefs.h>
66 __FBSDID("$FreeBSD$");
67 
68 #include <sys/param.h>
69 #include <sys/systm.h>
70 #include <sys/ktr.h>
71 #include <sys/lock.h>
72 #include <sys/mutex.h>
73 #include <sys/proc.h>
74 #include <sys/vmmeter.h>
75 #include <sys/mman.h>
76 #include <sys/vnode.h>
77 #include <sys/resourcevar.h>
78 #include <sys/file.h>
79 #include <sys/sysent.h>
80 #include <sys/shm.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 struct mtx map_sleep_mtx;
135 static uma_zone_t mapentzone;
136 static uma_zone_t kmapentzone;
137 static uma_zone_t mapzone;
138 static uma_zone_t vmspace_zone;
139 static struct vm_object kmapentobj;
140 static int vmspace_zinit(void *mem, int size, int flags);
141 static void vmspace_zfini(void *mem, int size);
142 static int vm_map_zinit(void *mem, int ize, int flags);
143 static void vm_map_zfini(void *mem, int size);
144 static void _vm_map_init(vm_map_t map, vm_offset_t min, vm_offset_t max);
145 
146 #ifdef INVARIANTS
147 static void vm_map_zdtor(void *mem, int size, void *arg);
148 static void vmspace_zdtor(void *mem, int size, void *arg);
149 #endif
150 
151 /*
152  * PROC_VMSPACE_{UN,}LOCK() can be a noop as long as vmspaces are type
153  * stable.
154  */
155 #define PROC_VMSPACE_LOCK(p) do { } while (0)
156 #define PROC_VMSPACE_UNLOCK(p) do { } while (0)
157 
158 void
159 vm_map_startup(void)
160 {
161 	mtx_init(&map_sleep_mtx, "vm map sleep mutex", NULL, MTX_DEF);
162 	mapzone = uma_zcreate("MAP", sizeof(struct vm_map), NULL,
163 #ifdef INVARIANTS
164 	    vm_map_zdtor,
165 #else
166 	    NULL,
167 #endif
168 	    vm_map_zinit, vm_map_zfini, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
169 	uma_prealloc(mapzone, MAX_KMAP);
170 	kmapentzone = uma_zcreate("KMAP ENTRY", sizeof(struct vm_map_entry),
171 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR,
172 	    UMA_ZONE_MTXCLASS | UMA_ZONE_VM);
173 	uma_prealloc(kmapentzone, MAX_KMAPENT);
174 	mapentzone = uma_zcreate("MAP ENTRY", sizeof(struct vm_map_entry),
175 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
176 }
177 
178 static void
179 vmspace_zfini(void *mem, int size)
180 {
181 	struct vmspace *vm;
182 
183 	vm = (struct vmspace *)mem;
184 	pmap_release(vmspace_pmap(vm));
185 	vm_map_zfini(&vm->vm_map, sizeof(vm->vm_map));
186 }
187 
188 static int
189 vmspace_zinit(void *mem, int size, int flags)
190 {
191 	struct vmspace *vm;
192 
193 	vm = (struct vmspace *)mem;
194 
195 	(void)vm_map_zinit(&vm->vm_map, sizeof(vm->vm_map), flags);
196 	pmap_pinit(vmspace_pmap(vm));
197 	return (0);
198 }
199 
200 static void
201 vm_map_zfini(void *mem, int size)
202 {
203 	vm_map_t map;
204 
205 	map = (vm_map_t)mem;
206 	mtx_destroy(&map->system_mtx);
207 	sx_destroy(&map->lock);
208 }
209 
210 static int
211 vm_map_zinit(void *mem, int size, int flags)
212 {
213 	vm_map_t map;
214 
215 	map = (vm_map_t)mem;
216 	map->nentries = 0;
217 	map->size = 0;
218 	mtx_init(&map->system_mtx, "system map", NULL, MTX_DEF | MTX_DUPOK);
219 	sx_init(&map->lock, "user map");
220 	return (0);
221 }
222 
223 #ifdef INVARIANTS
224 static void
225 vmspace_zdtor(void *mem, int size, void *arg)
226 {
227 	struct vmspace *vm;
228 
229 	vm = (struct vmspace *)mem;
230 
231 	vm_map_zdtor(&vm->vm_map, sizeof(vm->vm_map), arg);
232 }
233 static void
234 vm_map_zdtor(void *mem, int size, void *arg)
235 {
236 	vm_map_t map;
237 
238 	map = (vm_map_t)mem;
239 	KASSERT(map->nentries == 0,
240 	    ("map %p nentries == %d on free.",
241 	    map, map->nentries));
242 	KASSERT(map->size == 0,
243 	    ("map %p size == %lu on free.",
244 	    map, (unsigned long)map->size));
245 }
246 #endif	/* INVARIANTS */
247 
248 /*
249  * Allocate a vmspace structure, including a vm_map and pmap,
250  * and initialize those structures.  The refcnt is set to 1.
251  */
252 struct vmspace *
253 vmspace_alloc(min, max)
254 	vm_offset_t min, max;
255 {
256 	struct vmspace *vm;
257 
258 	vm = uma_zalloc(vmspace_zone, M_WAITOK);
259 	CTR1(KTR_VM, "vmspace_alloc: %p", vm);
260 	_vm_map_init(&vm->vm_map, min, max);
261 	vm->vm_map.pmap = vmspace_pmap(vm);		/* XXX */
262 	vm->vm_refcnt = 1;
263 	vm->vm_shm = NULL;
264 	vm->vm_swrss = 0;
265 	vm->vm_tsize = 0;
266 	vm->vm_dsize = 0;
267 	vm->vm_ssize = 0;
268 	vm->vm_taddr = 0;
269 	vm->vm_daddr = 0;
270 	vm->vm_maxsaddr = 0;
271 	return (vm);
272 }
273 
274 void
275 vm_init2(void)
276 {
277 	uma_zone_set_obj(kmapentzone, &kmapentobj, lmin(cnt.v_page_count,
278 	    (VM_MAX_KERNEL_ADDRESS - KERNBASE) / PAGE_SIZE) / 8 +
279 	     maxproc * 2 + maxfiles);
280 	vmspace_zone = uma_zcreate("VMSPACE", sizeof(struct vmspace), NULL,
281 #ifdef INVARIANTS
282 	    vmspace_zdtor,
283 #else
284 	    NULL,
285 #endif
286 	    vmspace_zinit, vmspace_zfini, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
287 }
288 
289 static inline void
290 vmspace_dofree(struct vmspace *vm)
291 {
292 	CTR1(KTR_VM, "vmspace_free: %p", vm);
293 
294 	/*
295 	 * Make sure any SysV shm is freed, it might not have been in
296 	 * exit1().
297 	 */
298 	shmexit(vm);
299 
300 	/*
301 	 * Lock the map, to wait out all other references to it.
302 	 * Delete all of the mappings and pages they hold, then call
303 	 * the pmap module to reclaim anything left.
304 	 */
305 	(void)vm_map_remove(&vm->vm_map, vm->vm_map.min_offset,
306 	    vm->vm_map.max_offset);
307 
308 	uma_zfree(vmspace_zone, vm);
309 }
310 
311 void
312 vmspace_free(struct vmspace *vm)
313 {
314 	int refcnt;
315 
316 	if (vm->vm_refcnt == 0)
317 		panic("vmspace_free: attempt to free already freed vmspace");
318 
319 	do
320 		refcnt = vm->vm_refcnt;
321 	while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt - 1));
322 	if (refcnt == 1)
323 		vmspace_dofree(vm);
324 }
325 
326 void
327 vmspace_exitfree(struct proc *p)
328 {
329 	struct vmspace *vm;
330 
331 	PROC_VMSPACE_LOCK(p);
332 	vm = p->p_vmspace;
333 	p->p_vmspace = NULL;
334 	PROC_VMSPACE_UNLOCK(p);
335 	KASSERT(vm == &vmspace0, ("vmspace_exitfree: wrong vmspace"));
336 	vmspace_free(vm);
337 }
338 
339 void
340 vmspace_exit(struct thread *td)
341 {
342 	int refcnt;
343 	struct vmspace *vm;
344 	struct proc *p;
345 
346 	/*
347 	 * Release user portion of address space.
348 	 * This releases references to vnodes,
349 	 * which could cause I/O if the file has been unlinked.
350 	 * Need to do this early enough that we can still sleep.
351 	 *
352 	 * The last exiting process to reach this point releases as
353 	 * much of the environment as it can. vmspace_dofree() is the
354 	 * slower fallback in case another process had a temporary
355 	 * reference to the vmspace.
356 	 */
357 
358 	p = td->td_proc;
359 	vm = p->p_vmspace;
360 	atomic_add_int(&vmspace0.vm_refcnt, 1);
361 	do {
362 		refcnt = vm->vm_refcnt;
363 		if (refcnt > 1 && p->p_vmspace != &vmspace0) {
364 			/* Switch now since other proc might free vmspace */
365 			PROC_VMSPACE_LOCK(p);
366 			p->p_vmspace = &vmspace0;
367 			PROC_VMSPACE_UNLOCK(p);
368 			pmap_activate(td);
369 		}
370 	} while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt - 1));
371 	if (refcnt == 1) {
372 		if (p->p_vmspace != vm) {
373 			/* vmspace not yet freed, switch back */
374 			PROC_VMSPACE_LOCK(p);
375 			p->p_vmspace = vm;
376 			PROC_VMSPACE_UNLOCK(p);
377 			pmap_activate(td);
378 		}
379 		pmap_remove_pages(vmspace_pmap(vm));
380 		/* Switch now since this proc will free vmspace */
381 		PROC_VMSPACE_LOCK(p);
382 		p->p_vmspace = &vmspace0;
383 		PROC_VMSPACE_UNLOCK(p);
384 		pmap_activate(td);
385 		vmspace_dofree(vm);
386 	}
387 }
388 
389 /* Acquire reference to vmspace owned by another process. */
390 
391 struct vmspace *
392 vmspace_acquire_ref(struct proc *p)
393 {
394 	struct vmspace *vm;
395 	int refcnt;
396 
397 	PROC_VMSPACE_LOCK(p);
398 	vm = p->p_vmspace;
399 	if (vm == NULL) {
400 		PROC_VMSPACE_UNLOCK(p);
401 		return (NULL);
402 	}
403 	do {
404 		refcnt = vm->vm_refcnt;
405 		if (refcnt <= 0) { 	/* Avoid 0->1 transition */
406 			PROC_VMSPACE_UNLOCK(p);
407 			return (NULL);
408 		}
409 	} while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt + 1));
410 	if (vm != p->p_vmspace) {
411 		PROC_VMSPACE_UNLOCK(p);
412 		vmspace_free(vm);
413 		return (NULL);
414 	}
415 	PROC_VMSPACE_UNLOCK(p);
416 	return (vm);
417 }
418 
419 void
420 _vm_map_lock(vm_map_t map, const char *file, int line)
421 {
422 
423 	if (map->system_map)
424 		_mtx_lock_flags(&map->system_mtx, 0, file, line);
425 	else
426 		_sx_xlock(&map->lock, file, line);
427 	map->timestamp++;
428 }
429 
430 void
431 _vm_map_unlock(vm_map_t map, const char *file, int line)
432 {
433 
434 	if (map->system_map)
435 		_mtx_unlock_flags(&map->system_mtx, 0, file, line);
436 	else
437 		_sx_xunlock(&map->lock, file, line);
438 }
439 
440 void
441 _vm_map_lock_read(vm_map_t map, const char *file, int line)
442 {
443 
444 	if (map->system_map)
445 		_mtx_lock_flags(&map->system_mtx, 0, file, line);
446 	else
447 		_sx_xlock(&map->lock, file, line);
448 }
449 
450 void
451 _vm_map_unlock_read(vm_map_t map, const char *file, int line)
452 {
453 
454 	if (map->system_map)
455 		_mtx_unlock_flags(&map->system_mtx, 0, file, line);
456 	else
457 		_sx_xunlock(&map->lock, file, line);
458 }
459 
460 int
461 _vm_map_trylock(vm_map_t map, const char *file, int line)
462 {
463 	int error;
464 
465 	error = map->system_map ?
466 	    !_mtx_trylock(&map->system_mtx, 0, file, line) :
467 	    !_sx_try_xlock(&map->lock, file, line);
468 	if (error == 0)
469 		map->timestamp++;
470 	return (error == 0);
471 }
472 
473 int
474 _vm_map_trylock_read(vm_map_t map, const char *file, int line)
475 {
476 	int error;
477 
478 	error = map->system_map ?
479 	    !_mtx_trylock(&map->system_mtx, 0, file, line) :
480 	    !_sx_try_xlock(&map->lock, file, line);
481 	return (error == 0);
482 }
483 
484 int
485 _vm_map_lock_upgrade(vm_map_t map, const char *file, int line)
486 {
487 
488 #ifdef INVARIANTS
489 	if (map->system_map) {
490 		_mtx_assert(&map->system_mtx, MA_OWNED, file, line);
491 	} else
492 		_sx_assert(&map->lock, SX_XLOCKED, file, line);
493 #endif
494 	map->timestamp++;
495 	return (0);
496 }
497 
498 void
499 _vm_map_lock_downgrade(vm_map_t map, const char *file, int line)
500 {
501 
502 #ifdef INVARIANTS
503 	if (map->system_map) {
504 		_mtx_assert(&map->system_mtx, MA_OWNED, file, line);
505 	} else
506 		_sx_assert(&map->lock, SX_XLOCKED, file, line);
507 #endif
508 }
509 
510 /*
511  *	vm_map_unlock_and_wait:
512  */
513 int
514 vm_map_unlock_and_wait(vm_map_t map, boolean_t user_wait)
515 {
516 
517 	mtx_lock(&map_sleep_mtx);
518 	vm_map_unlock(map);
519 	return (msleep(&map->root, &map_sleep_mtx, PDROP | PVM, "vmmaps", 0));
520 }
521 
522 /*
523  *	vm_map_wakeup:
524  */
525 void
526 vm_map_wakeup(vm_map_t map)
527 {
528 
529 	/*
530 	 * Acquire and release map_sleep_mtx to prevent a wakeup()
531 	 * from being performed (and lost) between the vm_map_unlock()
532 	 * and the msleep() in vm_map_unlock_and_wait().
533 	 */
534 	mtx_lock(&map_sleep_mtx);
535 	mtx_unlock(&map_sleep_mtx);
536 	wakeup(&map->root);
537 }
538 
539 long
540 vmspace_resident_count(struct vmspace *vmspace)
541 {
542 	return pmap_resident_count(vmspace_pmap(vmspace));
543 }
544 
545 long
546 vmspace_wired_count(struct vmspace *vmspace)
547 {
548 	return pmap_wired_count(vmspace_pmap(vmspace));
549 }
550 
551 /*
552  *	vm_map_create:
553  *
554  *	Creates and returns a new empty VM map with
555  *	the given physical map structure, and having
556  *	the given lower and upper address bounds.
557  */
558 vm_map_t
559 vm_map_create(pmap_t pmap, vm_offset_t min, vm_offset_t max)
560 {
561 	vm_map_t result;
562 
563 	result = uma_zalloc(mapzone, M_WAITOK);
564 	CTR1(KTR_VM, "vm_map_create: %p", result);
565 	_vm_map_init(result, min, max);
566 	result->pmap = pmap;
567 	return (result);
568 }
569 
570 /*
571  * Initialize an existing vm_map structure
572  * such as that in the vmspace structure.
573  * The pmap is set elsewhere.
574  */
575 static void
576 _vm_map_init(vm_map_t map, vm_offset_t min, vm_offset_t max)
577 {
578 
579 	map->header.next = map->header.prev = &map->header;
580 	map->needs_wakeup = FALSE;
581 	map->system_map = 0;
582 	map->min_offset = min;
583 	map->max_offset = max;
584 	map->flags = 0;
585 	map->root = NULL;
586 	map->timestamp = 0;
587 }
588 
589 void
590 vm_map_init(vm_map_t map, vm_offset_t min, vm_offset_t max)
591 {
592 	_vm_map_init(map, min, max);
593 	mtx_init(&map->system_mtx, "system map", NULL, MTX_DEF | MTX_DUPOK);
594 	sx_init(&map->lock, "user map");
595 }
596 
597 /*
598  *	vm_map_entry_dispose:	[ internal use only ]
599  *
600  *	Inverse of vm_map_entry_create.
601  */
602 static void
603 vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry)
604 {
605 	uma_zfree(map->system_map ? kmapentzone : mapentzone, entry);
606 }
607 
608 /*
609  *	vm_map_entry_create:	[ internal use only ]
610  *
611  *	Allocates a VM map entry for insertion.
612  *	No entry fields are filled in.
613  */
614 static vm_map_entry_t
615 vm_map_entry_create(vm_map_t map)
616 {
617 	vm_map_entry_t new_entry;
618 
619 	if (map->system_map)
620 		new_entry = uma_zalloc(kmapentzone, M_NOWAIT);
621 	else
622 		new_entry = uma_zalloc(mapentzone, M_WAITOK);
623 	if (new_entry == NULL)
624 		panic("vm_map_entry_create: kernel resources exhausted");
625 	return (new_entry);
626 }
627 
628 /*
629  *	vm_map_entry_set_behavior:
630  *
631  *	Set the expected access behavior, either normal, random, or
632  *	sequential.
633  */
634 static inline void
635 vm_map_entry_set_behavior(vm_map_entry_t entry, u_char behavior)
636 {
637 	entry->eflags = (entry->eflags & ~MAP_ENTRY_BEHAV_MASK) |
638 	    (behavior & MAP_ENTRY_BEHAV_MASK);
639 }
640 
641 /*
642  *	vm_map_entry_set_max_free:
643  *
644  *	Set the max_free field in a vm_map_entry.
645  */
646 static inline void
647 vm_map_entry_set_max_free(vm_map_entry_t entry)
648 {
649 
650 	entry->max_free = entry->adj_free;
651 	if (entry->left != NULL && entry->left->max_free > entry->max_free)
652 		entry->max_free = entry->left->max_free;
653 	if (entry->right != NULL && entry->right->max_free > entry->max_free)
654 		entry->max_free = entry->right->max_free;
655 }
656 
657 /*
658  *	vm_map_entry_splay:
659  *
660  *	The Sleator and Tarjan top-down splay algorithm with the
661  *	following variation.  Max_free must be computed bottom-up, so
662  *	on the downward pass, maintain the left and right spines in
663  *	reverse order.  Then, make a second pass up each side to fix
664  *	the pointers and compute max_free.  The time bound is O(log n)
665  *	amortized.
666  *
667  *	The new root is the vm_map_entry containing "addr", or else an
668  *	adjacent entry (lower or higher) if addr is not in the tree.
669  *
670  *	The map must be locked, and leaves it so.
671  *
672  *	Returns: the new root.
673  */
674 static vm_map_entry_t
675 vm_map_entry_splay(vm_offset_t addr, vm_map_entry_t root)
676 {
677 	vm_map_entry_t llist, rlist;
678 	vm_map_entry_t ltree, rtree;
679 	vm_map_entry_t y;
680 
681 	/* Special case of empty tree. */
682 	if (root == NULL)
683 		return (root);
684 
685 	/*
686 	 * Pass One: Splay down the tree until we find addr or a NULL
687 	 * pointer where addr would go.  llist and rlist are the two
688 	 * sides in reverse order (bottom-up), with llist linked by
689 	 * the right pointer and rlist linked by the left pointer in
690 	 * the vm_map_entry.  Wait until Pass Two to set max_free on
691 	 * the two spines.
692 	 */
693 	llist = NULL;
694 	rlist = NULL;
695 	for (;;) {
696 		/* root is never NULL in here. */
697 		if (addr < root->start) {
698 			y = root->left;
699 			if (y == NULL)
700 				break;
701 			if (addr < y->start && y->left != NULL) {
702 				/* Rotate right and put y on rlist. */
703 				root->left = y->right;
704 				y->right = root;
705 				vm_map_entry_set_max_free(root);
706 				root = y->left;
707 				y->left = rlist;
708 				rlist = y;
709 			} else {
710 				/* Put root on rlist. */
711 				root->left = rlist;
712 				rlist = root;
713 				root = y;
714 			}
715 		} else {
716 			y = root->right;
717 			if (addr < root->end || y == NULL)
718 				break;
719 			if (addr >= y->end && y->right != NULL) {
720 				/* Rotate left and put y on llist. */
721 				root->right = y->left;
722 				y->left = root;
723 				vm_map_entry_set_max_free(root);
724 				root = y->right;
725 				y->right = llist;
726 				llist = y;
727 			} else {
728 				/* Put root on llist. */
729 				root->right = llist;
730 				llist = root;
731 				root = y;
732 			}
733 		}
734 	}
735 
736 	/*
737 	 * Pass Two: Walk back up the two spines, flip the pointers
738 	 * and set max_free.  The subtrees of the root go at the
739 	 * bottom of llist and rlist.
740 	 */
741 	ltree = root->left;
742 	while (llist != NULL) {
743 		y = llist->right;
744 		llist->right = ltree;
745 		vm_map_entry_set_max_free(llist);
746 		ltree = llist;
747 		llist = y;
748 	}
749 	rtree = root->right;
750 	while (rlist != NULL) {
751 		y = rlist->left;
752 		rlist->left = rtree;
753 		vm_map_entry_set_max_free(rlist);
754 		rtree = rlist;
755 		rlist = y;
756 	}
757 
758 	/*
759 	 * Final assembly: add ltree and rtree as subtrees of root.
760 	 */
761 	root->left = ltree;
762 	root->right = rtree;
763 	vm_map_entry_set_max_free(root);
764 
765 	return (root);
766 }
767 
768 /*
769  *	vm_map_entry_{un,}link:
770  *
771  *	Insert/remove entries from maps.
772  */
773 static void
774 vm_map_entry_link(vm_map_t map,
775 		  vm_map_entry_t after_where,
776 		  vm_map_entry_t entry)
777 {
778 
779 	CTR4(KTR_VM,
780 	    "vm_map_entry_link: map %p, nentries %d, entry %p, after %p", map,
781 	    map->nentries, entry, after_where);
782 	map->nentries++;
783 	entry->prev = after_where;
784 	entry->next = after_where->next;
785 	entry->next->prev = entry;
786 	after_where->next = entry;
787 
788 	if (after_where != &map->header) {
789 		if (after_where != map->root)
790 			vm_map_entry_splay(after_where->start, map->root);
791 		entry->right = after_where->right;
792 		entry->left = after_where;
793 		after_where->right = NULL;
794 		after_where->adj_free = entry->start - after_where->end;
795 		vm_map_entry_set_max_free(after_where);
796 	} else {
797 		entry->right = map->root;
798 		entry->left = NULL;
799 	}
800 	entry->adj_free = (entry->next == &map->header ? map->max_offset :
801 	    entry->next->start) - entry->end;
802 	vm_map_entry_set_max_free(entry);
803 	map->root = entry;
804 }
805 
806 static void
807 vm_map_entry_unlink(vm_map_t map,
808 		    vm_map_entry_t entry)
809 {
810 	vm_map_entry_t next, prev, root;
811 
812 	if (entry != map->root)
813 		vm_map_entry_splay(entry->start, map->root);
814 	if (entry->left == NULL)
815 		root = entry->right;
816 	else {
817 		root = vm_map_entry_splay(entry->start, entry->left);
818 		root->right = entry->right;
819 		root->adj_free = (entry->next == &map->header ? map->max_offset :
820 		    entry->next->start) - root->end;
821 		vm_map_entry_set_max_free(root);
822 	}
823 	map->root = root;
824 
825 	prev = entry->prev;
826 	next = entry->next;
827 	next->prev = prev;
828 	prev->next = next;
829 	map->nentries--;
830 	CTR3(KTR_VM, "vm_map_entry_unlink: map %p, nentries %d, entry %p", map,
831 	    map->nentries, entry);
832 }
833 
834 /*
835  *	vm_map_entry_resize_free:
836  *
837  *	Recompute the amount of free space following a vm_map_entry
838  *	and propagate that value up the tree.  Call this function after
839  *	resizing a map entry in-place, that is, without a call to
840  *	vm_map_entry_link() or _unlink().
841  *
842  *	The map must be locked, and leaves it so.
843  */
844 static void
845 vm_map_entry_resize_free(vm_map_t map, vm_map_entry_t entry)
846 {
847 
848 	/*
849 	 * Using splay trees without parent pointers, propagating
850 	 * max_free up the tree is done by moving the entry to the
851 	 * root and making the change there.
852 	 */
853 	if (entry != map->root)
854 		map->root = vm_map_entry_splay(entry->start, map->root);
855 
856 	entry->adj_free = (entry->next == &map->header ? map->max_offset :
857 	    entry->next->start) - entry->end;
858 	vm_map_entry_set_max_free(entry);
859 }
860 
861 /*
862  *	vm_map_lookup_entry:	[ internal use only ]
863  *
864  *	Finds the map entry containing (or
865  *	immediately preceding) the specified address
866  *	in the given map; the entry is returned
867  *	in the "entry" parameter.  The boolean
868  *	result indicates whether the address is
869  *	actually contained in the map.
870  */
871 boolean_t
872 vm_map_lookup_entry(
873 	vm_map_t map,
874 	vm_offset_t address,
875 	vm_map_entry_t *entry)	/* OUT */
876 {
877 	vm_map_entry_t cur;
878 
879 	cur = vm_map_entry_splay(address, map->root);
880 	if (cur == NULL)
881 		*entry = &map->header;
882 	else {
883 		map->root = cur;
884 
885 		if (address >= cur->start) {
886 			*entry = cur;
887 			if (cur->end > address)
888 				return (TRUE);
889 		} else
890 			*entry = cur->prev;
891 	}
892 	return (FALSE);
893 }
894 
895 /*
896  *	vm_map_insert:
897  *
898  *	Inserts the given whole VM object into the target
899  *	map at the specified address range.  The object's
900  *	size should match that of the address range.
901  *
902  *	Requires that the map be locked, and leaves it so.
903  *
904  *	If object is non-NULL, ref count must be bumped by caller
905  *	prior to making call to account for the new entry.
906  */
907 int
908 vm_map_insert(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
909 	      vm_offset_t start, vm_offset_t end, vm_prot_t prot, vm_prot_t max,
910 	      int cow)
911 {
912 	vm_map_entry_t new_entry;
913 	vm_map_entry_t prev_entry;
914 	vm_map_entry_t temp_entry;
915 	vm_eflags_t protoeflags;
916 
917 	/*
918 	 * Check that the start and end points are not bogus.
919 	 */
920 	if ((start < map->min_offset) || (end > map->max_offset) ||
921 	    (start >= end))
922 		return (KERN_INVALID_ADDRESS);
923 
924 	/*
925 	 * Find the entry prior to the proposed starting address; if it's part
926 	 * of an existing entry, this range is bogus.
927 	 */
928 	if (vm_map_lookup_entry(map, start, &temp_entry))
929 		return (KERN_NO_SPACE);
930 
931 	prev_entry = temp_entry;
932 
933 	/*
934 	 * Assert that the next entry doesn't overlap the end point.
935 	 */
936 	if ((prev_entry->next != &map->header) &&
937 	    (prev_entry->next->start < end))
938 		return (KERN_NO_SPACE);
939 
940 	protoeflags = 0;
941 
942 	if (cow & MAP_COPY_ON_WRITE)
943 		protoeflags |= MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY;
944 
945 	if (cow & MAP_NOFAULT) {
946 		protoeflags |= MAP_ENTRY_NOFAULT;
947 
948 		KASSERT(object == NULL,
949 			("vm_map_insert: paradoxical MAP_NOFAULT request"));
950 	}
951 	if (cow & MAP_DISABLE_SYNCER)
952 		protoeflags |= MAP_ENTRY_NOSYNC;
953 	if (cow & MAP_DISABLE_COREDUMP)
954 		protoeflags |= MAP_ENTRY_NOCOREDUMP;
955 
956 	if (object != NULL) {
957 		/*
958 		 * OBJ_ONEMAPPING must be cleared unless this mapping
959 		 * is trivially proven to be the only mapping for any
960 		 * of the object's pages.  (Object granularity
961 		 * reference counting is insufficient to recognize
962 		 * aliases with precision.)
963 		 */
964 		VM_OBJECT_LOCK(object);
965 		if (object->ref_count > 1 || object->shadow_count != 0)
966 			vm_object_clear_flag(object, OBJ_ONEMAPPING);
967 		VM_OBJECT_UNLOCK(object);
968 	}
969 	else if ((prev_entry != &map->header) &&
970 		 (prev_entry->eflags == protoeflags) &&
971 		 (prev_entry->end == start) &&
972 		 (prev_entry->wired_count == 0) &&
973 		 ((prev_entry->object.vm_object == NULL) ||
974 		  vm_object_coalesce(prev_entry->object.vm_object,
975 				     prev_entry->offset,
976 				     (vm_size_t)(prev_entry->end - prev_entry->start),
977 				     (vm_size_t)(end - prev_entry->end)))) {
978 		/*
979 		 * We were able to extend the object.  Determine if we
980 		 * can extend the previous map entry to include the
981 		 * new range as well.
982 		 */
983 		if ((prev_entry->inheritance == VM_INHERIT_DEFAULT) &&
984 		    (prev_entry->protection == prot) &&
985 		    (prev_entry->max_protection == max)) {
986 			map->size += (end - prev_entry->end);
987 			prev_entry->end = end;
988 			vm_map_entry_resize_free(map, prev_entry);
989 			vm_map_simplify_entry(map, prev_entry);
990 			return (KERN_SUCCESS);
991 		}
992 
993 		/*
994 		 * If we can extend the object but cannot extend the
995 		 * map entry, we have to create a new map entry.  We
996 		 * must bump the ref count on the extended object to
997 		 * account for it.  object may be NULL.
998 		 */
999 		object = prev_entry->object.vm_object;
1000 		offset = prev_entry->offset +
1001 			(prev_entry->end - prev_entry->start);
1002 		vm_object_reference(object);
1003 	}
1004 
1005 	/*
1006 	 * NOTE: if conditionals fail, object can be NULL here.  This occurs
1007 	 * in things like the buffer map where we manage kva but do not manage
1008 	 * backing objects.
1009 	 */
1010 
1011 	/*
1012 	 * Create a new entry
1013 	 */
1014 	new_entry = vm_map_entry_create(map);
1015 	new_entry->start = start;
1016 	new_entry->end = end;
1017 
1018 	new_entry->eflags = protoeflags;
1019 	new_entry->object.vm_object = object;
1020 	new_entry->offset = offset;
1021 	new_entry->avail_ssize = 0;
1022 
1023 	new_entry->inheritance = VM_INHERIT_DEFAULT;
1024 	new_entry->protection = prot;
1025 	new_entry->max_protection = max;
1026 	new_entry->wired_count = 0;
1027 
1028 	/*
1029 	 * Insert the new entry into the list
1030 	 */
1031 	vm_map_entry_link(map, prev_entry, new_entry);
1032 	map->size += new_entry->end - new_entry->start;
1033 
1034 #if 0
1035 	/*
1036 	 * Temporarily removed to avoid MAP_STACK panic, due to
1037 	 * MAP_STACK being a huge hack.  Will be added back in
1038 	 * when MAP_STACK (and the user stack mapping) is fixed.
1039 	 */
1040 	/*
1041 	 * It may be possible to simplify the entry
1042 	 */
1043 	vm_map_simplify_entry(map, new_entry);
1044 #endif
1045 
1046 	if (cow & (MAP_PREFAULT|MAP_PREFAULT_PARTIAL)) {
1047 		vm_map_pmap_enter(map, start, prot,
1048 				    object, OFF_TO_IDX(offset), end - start,
1049 				    cow & MAP_PREFAULT_PARTIAL);
1050 	}
1051 
1052 	return (KERN_SUCCESS);
1053 }
1054 
1055 /*
1056  *	vm_map_findspace:
1057  *
1058  *	Find the first fit (lowest VM address) for "length" free bytes
1059  *	beginning at address >= start in the given map.
1060  *
1061  *	In a vm_map_entry, "adj_free" is the amount of free space
1062  *	adjacent (higher address) to this entry, and "max_free" is the
1063  *	maximum amount of contiguous free space in its subtree.  This
1064  *	allows finding a free region in one path down the tree, so
1065  *	O(log n) amortized with splay trees.
1066  *
1067  *	The map must be locked, and leaves it so.
1068  *
1069  *	Returns: 0 on success, and starting address in *addr,
1070  *		 1 if insufficient space.
1071  */
1072 int
1073 vm_map_findspace(vm_map_t map, vm_offset_t start, vm_size_t length,
1074     vm_offset_t *addr)	/* OUT */
1075 {
1076 	vm_map_entry_t entry;
1077 	vm_offset_t end, st;
1078 
1079 	/*
1080 	 * Request must fit within min/max VM address and must avoid
1081 	 * address wrap.
1082 	 */
1083 	if (start < map->min_offset)
1084 		start = map->min_offset;
1085 	if (start + length > map->max_offset || start + length < start)
1086 		return (1);
1087 
1088 	/* Empty tree means wide open address space. */
1089 	if (map->root == NULL) {
1090 		*addr = start;
1091 		goto found;
1092 	}
1093 
1094 	/*
1095 	 * After splay, if start comes before root node, then there
1096 	 * must be a gap from start to the root.
1097 	 */
1098 	map->root = vm_map_entry_splay(start, map->root);
1099 	if (start + length <= map->root->start) {
1100 		*addr = start;
1101 		goto found;
1102 	}
1103 
1104 	/*
1105 	 * Root is the last node that might begin its gap before
1106 	 * start, and this is the last comparison where address
1107 	 * wrap might be a problem.
1108 	 */
1109 	st = (start > map->root->end) ? start : map->root->end;
1110 	if (length <= map->root->end + map->root->adj_free - st) {
1111 		*addr = st;
1112 		goto found;
1113 	}
1114 
1115 	/* With max_free, can immediately tell if no solution. */
1116 	entry = map->root->right;
1117 	if (entry == NULL || length > entry->max_free)
1118 		return (1);
1119 
1120 	/*
1121 	 * Search the right subtree in the order: left subtree, root,
1122 	 * right subtree (first fit).  The previous splay implies that
1123 	 * all regions in the right subtree have addresses > start.
1124 	 */
1125 	while (entry != NULL) {
1126 		if (entry->left != NULL && entry->left->max_free >= length)
1127 			entry = entry->left;
1128 		else if (entry->adj_free >= length) {
1129 			*addr = entry->end;
1130 			goto found;
1131 		} else
1132 			entry = entry->right;
1133 	}
1134 
1135 	/* Can't get here, so panic if we do. */
1136 	panic("vm_map_findspace: max_free corrupt");
1137 
1138 found:
1139 	/* Expand the kernel pmap, if necessary. */
1140 	if (map == kernel_map) {
1141 		end = round_page(*addr + length);
1142 		if (end > kernel_vm_end)
1143 			pmap_growkernel(end);
1144 	}
1145 	return (0);
1146 }
1147 
1148 /*
1149  *	vm_map_find finds an unallocated region in the target address
1150  *	map with the given length.  The search is defined to be
1151  *	first-fit from the specified address; the region found is
1152  *	returned in the same parameter.
1153  *
1154  *	If object is non-NULL, ref count must be bumped by caller
1155  *	prior to making call to account for the new entry.
1156  */
1157 int
1158 vm_map_find(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1159 	    vm_offset_t *addr,	/* IN/OUT */
1160 	    vm_size_t length, boolean_t find_space, vm_prot_t prot,
1161 	    vm_prot_t max, int cow)
1162 {
1163 	vm_offset_t start;
1164 	int result;
1165 
1166 	start = *addr;
1167 	vm_map_lock(map);
1168 	if (find_space) {
1169 		if (vm_map_findspace(map, start, length, addr)) {
1170 			vm_map_unlock(map);
1171 			return (KERN_NO_SPACE);
1172 		}
1173 		start = *addr;
1174 	}
1175 	result = vm_map_insert(map, object, offset,
1176 		start, start + length, prot, max, cow);
1177 	vm_map_unlock(map);
1178 	return (result);
1179 }
1180 
1181 /*
1182  *	vm_map_simplify_entry:
1183  *
1184  *	Simplify the given map entry by merging with either neighbor.  This
1185  *	routine also has the ability to merge with both neighbors.
1186  *
1187  *	The map must be locked.
1188  *
1189  *	This routine guarentees that the passed entry remains valid (though
1190  *	possibly extended).  When merging, this routine may delete one or
1191  *	both neighbors.
1192  */
1193 void
1194 vm_map_simplify_entry(vm_map_t map, vm_map_entry_t entry)
1195 {
1196 	vm_map_entry_t next, prev;
1197 	vm_size_t prevsize, esize;
1198 
1199 	if (entry->eflags & (MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_IS_SUB_MAP))
1200 		return;
1201 
1202 	prev = entry->prev;
1203 	if (prev != &map->header) {
1204 		prevsize = prev->end - prev->start;
1205 		if ( (prev->end == entry->start) &&
1206 		     (prev->object.vm_object == entry->object.vm_object) &&
1207 		     (!prev->object.vm_object ||
1208 			(prev->offset + prevsize == entry->offset)) &&
1209 		     (prev->eflags == entry->eflags) &&
1210 		     (prev->protection == entry->protection) &&
1211 		     (prev->max_protection == entry->max_protection) &&
1212 		     (prev->inheritance == entry->inheritance) &&
1213 		     (prev->wired_count == entry->wired_count)) {
1214 			vm_map_entry_unlink(map, prev);
1215 			entry->start = prev->start;
1216 			entry->offset = prev->offset;
1217 			if (entry->prev != &map->header)
1218 				vm_map_entry_resize_free(map, entry->prev);
1219 			if (prev->object.vm_object)
1220 				vm_object_deallocate(prev->object.vm_object);
1221 			vm_map_entry_dispose(map, prev);
1222 		}
1223 	}
1224 
1225 	next = entry->next;
1226 	if (next != &map->header) {
1227 		esize = entry->end - entry->start;
1228 		if ((entry->end == next->start) &&
1229 		    (next->object.vm_object == entry->object.vm_object) &&
1230 		     (!entry->object.vm_object ||
1231 			(entry->offset + esize == next->offset)) &&
1232 		    (next->eflags == entry->eflags) &&
1233 		    (next->protection == entry->protection) &&
1234 		    (next->max_protection == entry->max_protection) &&
1235 		    (next->inheritance == entry->inheritance) &&
1236 		    (next->wired_count == entry->wired_count)) {
1237 			vm_map_entry_unlink(map, next);
1238 			entry->end = next->end;
1239 			vm_map_entry_resize_free(map, entry);
1240 			if (next->object.vm_object)
1241 				vm_object_deallocate(next->object.vm_object);
1242 			vm_map_entry_dispose(map, next);
1243 		}
1244 	}
1245 }
1246 /*
1247  *	vm_map_clip_start:	[ internal use only ]
1248  *
1249  *	Asserts that the given entry begins at or after
1250  *	the specified address; if necessary,
1251  *	it splits the entry into two.
1252  */
1253 #define vm_map_clip_start(map, entry, startaddr) \
1254 { \
1255 	if (startaddr > entry->start) \
1256 		_vm_map_clip_start(map, entry, startaddr); \
1257 }
1258 
1259 /*
1260  *	This routine is called only when it is known that
1261  *	the entry must be split.
1262  */
1263 static void
1264 _vm_map_clip_start(vm_map_t map, vm_map_entry_t entry, vm_offset_t start)
1265 {
1266 	vm_map_entry_t new_entry;
1267 
1268 	/*
1269 	 * Split off the front portion -- note that we must insert the new
1270 	 * entry BEFORE this one, so that this entry has the specified
1271 	 * starting address.
1272 	 */
1273 	vm_map_simplify_entry(map, entry);
1274 
1275 	/*
1276 	 * If there is no object backing this entry, we might as well create
1277 	 * one now.  If we defer it, an object can get created after the map
1278 	 * is clipped, and individual objects will be created for the split-up
1279 	 * map.  This is a bit of a hack, but is also about the best place to
1280 	 * put this improvement.
1281 	 */
1282 	if (entry->object.vm_object == NULL && !map->system_map) {
1283 		vm_object_t object;
1284 		object = vm_object_allocate(OBJT_DEFAULT,
1285 				atop(entry->end - entry->start));
1286 		entry->object.vm_object = object;
1287 		entry->offset = 0;
1288 	}
1289 
1290 	new_entry = vm_map_entry_create(map);
1291 	*new_entry = *entry;
1292 
1293 	new_entry->end = start;
1294 	entry->offset += (start - entry->start);
1295 	entry->start = start;
1296 
1297 	vm_map_entry_link(map, entry->prev, new_entry);
1298 
1299 	if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1300 		vm_object_reference(new_entry->object.vm_object);
1301 	}
1302 }
1303 
1304 /*
1305  *	vm_map_clip_end:	[ internal use only ]
1306  *
1307  *	Asserts that the given entry ends at or before
1308  *	the specified address; if necessary,
1309  *	it splits the entry into two.
1310  */
1311 #define vm_map_clip_end(map, entry, endaddr) \
1312 { \
1313 	if ((endaddr) < (entry->end)) \
1314 		_vm_map_clip_end((map), (entry), (endaddr)); \
1315 }
1316 
1317 /*
1318  *	This routine is called only when it is known that
1319  *	the entry must be split.
1320  */
1321 static void
1322 _vm_map_clip_end(vm_map_t map, vm_map_entry_t entry, vm_offset_t end)
1323 {
1324 	vm_map_entry_t new_entry;
1325 
1326 	/*
1327 	 * If there is no object backing this entry, we might as well create
1328 	 * one now.  If we defer it, an object can get created after the map
1329 	 * is clipped, and individual objects will be created for the split-up
1330 	 * map.  This is a bit of a hack, but is also about the best place to
1331 	 * put this improvement.
1332 	 */
1333 	if (entry->object.vm_object == NULL && !map->system_map) {
1334 		vm_object_t object;
1335 		object = vm_object_allocate(OBJT_DEFAULT,
1336 				atop(entry->end - entry->start));
1337 		entry->object.vm_object = object;
1338 		entry->offset = 0;
1339 	}
1340 
1341 	/*
1342 	 * Create a new entry and insert it AFTER the specified entry
1343 	 */
1344 	new_entry = vm_map_entry_create(map);
1345 	*new_entry = *entry;
1346 
1347 	new_entry->start = entry->end = end;
1348 	new_entry->offset += (end - entry->start);
1349 
1350 	vm_map_entry_link(map, entry, new_entry);
1351 
1352 	if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1353 		vm_object_reference(new_entry->object.vm_object);
1354 	}
1355 }
1356 
1357 /*
1358  *	VM_MAP_RANGE_CHECK:	[ internal use only ]
1359  *
1360  *	Asserts that the starting and ending region
1361  *	addresses fall within the valid range of the map.
1362  */
1363 #define	VM_MAP_RANGE_CHECK(map, start, end)		\
1364 		{					\
1365 		if (start < vm_map_min(map))		\
1366 			start = vm_map_min(map);	\
1367 		if (end > vm_map_max(map))		\
1368 			end = vm_map_max(map);		\
1369 		if (start > end)			\
1370 			start = end;			\
1371 		}
1372 
1373 /*
1374  *	vm_map_submap:		[ kernel use only ]
1375  *
1376  *	Mark the given range as handled by a subordinate map.
1377  *
1378  *	This range must have been created with vm_map_find,
1379  *	and no other operations may have been performed on this
1380  *	range prior to calling vm_map_submap.
1381  *
1382  *	Only a limited number of operations can be performed
1383  *	within this rage after calling vm_map_submap:
1384  *		vm_fault
1385  *	[Don't try vm_map_copy!]
1386  *
1387  *	To remove a submapping, one must first remove the
1388  *	range from the superior map, and then destroy the
1389  *	submap (if desired).  [Better yet, don't try it.]
1390  */
1391 int
1392 vm_map_submap(
1393 	vm_map_t map,
1394 	vm_offset_t start,
1395 	vm_offset_t end,
1396 	vm_map_t submap)
1397 {
1398 	vm_map_entry_t entry;
1399 	int result = KERN_INVALID_ARGUMENT;
1400 
1401 	vm_map_lock(map);
1402 
1403 	VM_MAP_RANGE_CHECK(map, start, end);
1404 
1405 	if (vm_map_lookup_entry(map, start, &entry)) {
1406 		vm_map_clip_start(map, entry, start);
1407 	} else
1408 		entry = entry->next;
1409 
1410 	vm_map_clip_end(map, entry, end);
1411 
1412 	if ((entry->start == start) && (entry->end == end) &&
1413 	    ((entry->eflags & MAP_ENTRY_COW) == 0) &&
1414 	    (entry->object.vm_object == NULL)) {
1415 		entry->object.sub_map = submap;
1416 		entry->eflags |= MAP_ENTRY_IS_SUB_MAP;
1417 		result = KERN_SUCCESS;
1418 	}
1419 	vm_map_unlock(map);
1420 
1421 	return (result);
1422 }
1423 
1424 /*
1425  * The maximum number of pages to map
1426  */
1427 #define	MAX_INIT_PT	96
1428 
1429 /*
1430  *	vm_map_pmap_enter:
1431  *
1432  *	Preload read-only mappings for the given object into the specified
1433  *	map.  This eliminates the soft faults on process startup and
1434  *	immediately after an mmap(2).
1435  */
1436 void
1437 vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot,
1438     vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags)
1439 {
1440 	vm_offset_t start, tmpidx;
1441 	int psize;
1442 	vm_page_t p, p_start;
1443 	boolean_t are_queues_locked;
1444 
1445 	if ((prot & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0 || object == NULL)
1446 		return;
1447 	VM_OBJECT_LOCK(object);
1448 	if (object->type == OBJT_DEVICE) {
1449 		pmap_object_init_pt(map->pmap, addr, object, pindex, size);
1450 		goto unlock_return;
1451 	}
1452 
1453 	psize = atop(size);
1454 
1455 	if (object->type != OBJT_VNODE ||
1456 	    ((flags & MAP_PREFAULT_PARTIAL) && (psize > MAX_INIT_PT) &&
1457 	     (object->resident_page_count > MAX_INIT_PT))) {
1458 		goto unlock_return;
1459 	}
1460 
1461 	if (psize + pindex > object->size) {
1462 		if (object->size < pindex)
1463 			goto unlock_return;
1464 		psize = object->size - pindex;
1465 	}
1466 
1467 	are_queues_locked = FALSE;
1468 	start = 0;
1469 	p_start = NULL;
1470 
1471 	if ((p = TAILQ_FIRST(&object->memq)) != NULL) {
1472 		if (p->pindex < pindex) {
1473 			p = vm_page_splay(pindex, object->root);
1474 			if ((object->root = p)->pindex < pindex)
1475 				p = TAILQ_NEXT(p, listq);
1476 		}
1477 	}
1478 	/*
1479 	 * Assert: the variable p is either (1) the page with the
1480 	 * least pindex greater than or equal to the parameter pindex
1481 	 * or (2) NULL.
1482 	 */
1483 	for (;
1484 	     p != NULL && (tmpidx = p->pindex - pindex) < psize;
1485 	     p = TAILQ_NEXT(p, listq)) {
1486 		/*
1487 		 * don't allow an madvise to blow away our really
1488 		 * free pages allocating pv entries.
1489 		 */
1490 		if ((flags & MAP_PREFAULT_MADVISE) &&
1491 		    cnt.v_free_count < cnt.v_free_reserved) {
1492 			psize = tmpidx;
1493 			break;
1494 		}
1495 		if ((p->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL &&
1496 		    (p->busy == 0) &&
1497 		    (p->flags & PG_FICTITIOUS) == 0) {
1498 			if (p_start == NULL) {
1499 				start = addr + ptoa(tmpidx);
1500 				p_start = p;
1501 			}
1502 			if (!are_queues_locked) {
1503 				are_queues_locked = TRUE;
1504 				vm_page_lock_queues();
1505 			}
1506 			if (VM_PAGE_INQUEUE1(p, PQ_CACHE))
1507 				vm_page_deactivate(p);
1508 		} else if (p_start != NULL) {
1509 			pmap_enter_object(map->pmap, start, addr +
1510 			    ptoa(tmpidx), p_start, prot);
1511 			p_start = NULL;
1512 		}
1513 	}
1514 	if (p_start != NULL)
1515 		pmap_enter_object(map->pmap, start, addr + ptoa(psize),
1516 		    p_start, prot);
1517 	if (are_queues_locked)
1518 		vm_page_unlock_queues();
1519 unlock_return:
1520 	VM_OBJECT_UNLOCK(object);
1521 }
1522 
1523 /*
1524  *	vm_map_protect:
1525  *
1526  *	Sets the protection of the specified address
1527  *	region in the target map.  If "set_max" is
1528  *	specified, the maximum protection is to be set;
1529  *	otherwise, only the current protection is affected.
1530  */
1531 int
1532 vm_map_protect(vm_map_t map, vm_offset_t start, vm_offset_t end,
1533 	       vm_prot_t new_prot, boolean_t set_max)
1534 {
1535 	vm_map_entry_t current;
1536 	vm_map_entry_t entry;
1537 
1538 	vm_map_lock(map);
1539 
1540 	VM_MAP_RANGE_CHECK(map, start, end);
1541 
1542 	if (vm_map_lookup_entry(map, start, &entry)) {
1543 		vm_map_clip_start(map, entry, start);
1544 	} else {
1545 		entry = entry->next;
1546 	}
1547 
1548 	/*
1549 	 * Make a first pass to check for protection violations.
1550 	 */
1551 	current = entry;
1552 	while ((current != &map->header) && (current->start < end)) {
1553 		if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
1554 			vm_map_unlock(map);
1555 			return (KERN_INVALID_ARGUMENT);
1556 		}
1557 		if ((new_prot & current->max_protection) != new_prot) {
1558 			vm_map_unlock(map);
1559 			return (KERN_PROTECTION_FAILURE);
1560 		}
1561 		current = current->next;
1562 	}
1563 
1564 	/*
1565 	 * Go back and fix up protections. [Note that clipping is not
1566 	 * necessary the second time.]
1567 	 */
1568 	current = entry;
1569 	while ((current != &map->header) && (current->start < end)) {
1570 		vm_prot_t old_prot;
1571 
1572 		vm_map_clip_end(map, current, end);
1573 
1574 		old_prot = current->protection;
1575 		if (set_max)
1576 			current->protection =
1577 			    (current->max_protection = new_prot) &
1578 			    old_prot;
1579 		else
1580 			current->protection = new_prot;
1581 
1582 		/*
1583 		 * Update physical map if necessary. Worry about copy-on-write
1584 		 * here -- CHECK THIS XXX
1585 		 */
1586 		if (current->protection != old_prot) {
1587 #define MASK(entry)	(((entry)->eflags & MAP_ENTRY_COW) ? ~VM_PROT_WRITE : \
1588 							VM_PROT_ALL)
1589 			pmap_protect(map->pmap, current->start,
1590 			    current->end,
1591 			    current->protection & MASK(current));
1592 #undef	MASK
1593 		}
1594 		vm_map_simplify_entry(map, current);
1595 		current = current->next;
1596 	}
1597 	vm_map_unlock(map);
1598 	return (KERN_SUCCESS);
1599 }
1600 
1601 /*
1602  *	vm_map_madvise:
1603  *
1604  *	This routine traverses a processes map handling the madvise
1605  *	system call.  Advisories are classified as either those effecting
1606  *	the vm_map_entry structure, or those effecting the underlying
1607  *	objects.
1608  */
1609 int
1610 vm_map_madvise(
1611 	vm_map_t map,
1612 	vm_offset_t start,
1613 	vm_offset_t end,
1614 	int behav)
1615 {
1616 	vm_map_entry_t current, entry;
1617 	int modify_map = 0;
1618 
1619 	/*
1620 	 * Some madvise calls directly modify the vm_map_entry, in which case
1621 	 * we need to use an exclusive lock on the map and we need to perform
1622 	 * various clipping operations.  Otherwise we only need a read-lock
1623 	 * on the map.
1624 	 */
1625 	switch(behav) {
1626 	case MADV_NORMAL:
1627 	case MADV_SEQUENTIAL:
1628 	case MADV_RANDOM:
1629 	case MADV_NOSYNC:
1630 	case MADV_AUTOSYNC:
1631 	case MADV_NOCORE:
1632 	case MADV_CORE:
1633 		modify_map = 1;
1634 		vm_map_lock(map);
1635 		break;
1636 	case MADV_WILLNEED:
1637 	case MADV_DONTNEED:
1638 	case MADV_FREE:
1639 		vm_map_lock_read(map);
1640 		break;
1641 	default:
1642 		return (KERN_INVALID_ARGUMENT);
1643 	}
1644 
1645 	/*
1646 	 * Locate starting entry and clip if necessary.
1647 	 */
1648 	VM_MAP_RANGE_CHECK(map, start, end);
1649 
1650 	if (vm_map_lookup_entry(map, start, &entry)) {
1651 		if (modify_map)
1652 			vm_map_clip_start(map, entry, start);
1653 	} else {
1654 		entry = entry->next;
1655 	}
1656 
1657 	if (modify_map) {
1658 		/*
1659 		 * madvise behaviors that are implemented in the vm_map_entry.
1660 		 *
1661 		 * We clip the vm_map_entry so that behavioral changes are
1662 		 * limited to the specified address range.
1663 		 */
1664 		for (current = entry;
1665 		     (current != &map->header) && (current->start < end);
1666 		     current = current->next
1667 		) {
1668 			if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
1669 				continue;
1670 
1671 			vm_map_clip_end(map, current, end);
1672 
1673 			switch (behav) {
1674 			case MADV_NORMAL:
1675 				vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_NORMAL);
1676 				break;
1677 			case MADV_SEQUENTIAL:
1678 				vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_SEQUENTIAL);
1679 				break;
1680 			case MADV_RANDOM:
1681 				vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_RANDOM);
1682 				break;
1683 			case MADV_NOSYNC:
1684 				current->eflags |= MAP_ENTRY_NOSYNC;
1685 				break;
1686 			case MADV_AUTOSYNC:
1687 				current->eflags &= ~MAP_ENTRY_NOSYNC;
1688 				break;
1689 			case MADV_NOCORE:
1690 				current->eflags |= MAP_ENTRY_NOCOREDUMP;
1691 				break;
1692 			case MADV_CORE:
1693 				current->eflags &= ~MAP_ENTRY_NOCOREDUMP;
1694 				break;
1695 			default:
1696 				break;
1697 			}
1698 			vm_map_simplify_entry(map, current);
1699 		}
1700 		vm_map_unlock(map);
1701 	} else {
1702 		vm_pindex_t pindex;
1703 		int count;
1704 
1705 		/*
1706 		 * madvise behaviors that are implemented in the underlying
1707 		 * vm_object.
1708 		 *
1709 		 * Since we don't clip the vm_map_entry, we have to clip
1710 		 * the vm_object pindex and count.
1711 		 */
1712 		for (current = entry;
1713 		     (current != &map->header) && (current->start < end);
1714 		     current = current->next
1715 		) {
1716 			vm_offset_t useStart;
1717 
1718 			if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
1719 				continue;
1720 
1721 			pindex = OFF_TO_IDX(current->offset);
1722 			count = atop(current->end - current->start);
1723 			useStart = current->start;
1724 
1725 			if (current->start < start) {
1726 				pindex += atop(start - current->start);
1727 				count -= atop(start - current->start);
1728 				useStart = start;
1729 			}
1730 			if (current->end > end)
1731 				count -= atop(current->end - end);
1732 
1733 			if (count <= 0)
1734 				continue;
1735 
1736 			vm_object_madvise(current->object.vm_object,
1737 					  pindex, count, behav);
1738 			if (behav == MADV_WILLNEED) {
1739 				vm_map_pmap_enter(map,
1740 				    useStart,
1741 				    current->protection,
1742 				    current->object.vm_object,
1743 				    pindex,
1744 				    (count << PAGE_SHIFT),
1745 				    MAP_PREFAULT_MADVISE
1746 				);
1747 			}
1748 		}
1749 		vm_map_unlock_read(map);
1750 	}
1751 	return (0);
1752 }
1753 
1754 
1755 /*
1756  *	vm_map_inherit:
1757  *
1758  *	Sets the inheritance of the specified address
1759  *	range in the target map.  Inheritance
1760  *	affects how the map will be shared with
1761  *	child maps at the time of vm_map_fork.
1762  */
1763 int
1764 vm_map_inherit(vm_map_t map, vm_offset_t start, vm_offset_t end,
1765 	       vm_inherit_t new_inheritance)
1766 {
1767 	vm_map_entry_t entry;
1768 	vm_map_entry_t temp_entry;
1769 
1770 	switch (new_inheritance) {
1771 	case VM_INHERIT_NONE:
1772 	case VM_INHERIT_COPY:
1773 	case VM_INHERIT_SHARE:
1774 		break;
1775 	default:
1776 		return (KERN_INVALID_ARGUMENT);
1777 	}
1778 	vm_map_lock(map);
1779 	VM_MAP_RANGE_CHECK(map, start, end);
1780 	if (vm_map_lookup_entry(map, start, &temp_entry)) {
1781 		entry = temp_entry;
1782 		vm_map_clip_start(map, entry, start);
1783 	} else
1784 		entry = temp_entry->next;
1785 	while ((entry != &map->header) && (entry->start < end)) {
1786 		vm_map_clip_end(map, entry, end);
1787 		entry->inheritance = new_inheritance;
1788 		vm_map_simplify_entry(map, entry);
1789 		entry = entry->next;
1790 	}
1791 	vm_map_unlock(map);
1792 	return (KERN_SUCCESS);
1793 }
1794 
1795 /*
1796  *	vm_map_unwire:
1797  *
1798  *	Implements both kernel and user unwiring.
1799  */
1800 int
1801 vm_map_unwire(vm_map_t map, vm_offset_t start, vm_offset_t end,
1802     int flags)
1803 {
1804 	vm_map_entry_t entry, first_entry, tmp_entry;
1805 	vm_offset_t saved_start;
1806 	unsigned int last_timestamp;
1807 	int rv;
1808 	boolean_t need_wakeup, result, user_unwire;
1809 
1810 	user_unwire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE;
1811 	vm_map_lock(map);
1812 	VM_MAP_RANGE_CHECK(map, start, end);
1813 	if (!vm_map_lookup_entry(map, start, &first_entry)) {
1814 		if (flags & VM_MAP_WIRE_HOLESOK)
1815 			first_entry = first_entry->next;
1816 		else {
1817 			vm_map_unlock(map);
1818 			return (KERN_INVALID_ADDRESS);
1819 		}
1820 	}
1821 	last_timestamp = map->timestamp;
1822 	entry = first_entry;
1823 	while (entry != &map->header && entry->start < end) {
1824 		if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
1825 			/*
1826 			 * We have not yet clipped the entry.
1827 			 */
1828 			saved_start = (start >= entry->start) ? start :
1829 			    entry->start;
1830 			entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
1831 			if (vm_map_unlock_and_wait(map, user_unwire)) {
1832 				/*
1833 				 * Allow interruption of user unwiring?
1834 				 */
1835 			}
1836 			vm_map_lock(map);
1837 			if (last_timestamp+1 != map->timestamp) {
1838 				/*
1839 				 * Look again for the entry because the map was
1840 				 * modified while it was unlocked.
1841 				 * Specifically, the entry may have been
1842 				 * clipped, merged, or deleted.
1843 				 */
1844 				if (!vm_map_lookup_entry(map, saved_start,
1845 				    &tmp_entry)) {
1846 					if (flags & VM_MAP_WIRE_HOLESOK)
1847 						tmp_entry = tmp_entry->next;
1848 					else {
1849 						if (saved_start == start) {
1850 							/*
1851 							 * First_entry has been deleted.
1852 							 */
1853 							vm_map_unlock(map);
1854 							return (KERN_INVALID_ADDRESS);
1855 						}
1856 						end = saved_start;
1857 						rv = KERN_INVALID_ADDRESS;
1858 						goto done;
1859 					}
1860 				}
1861 				if (entry == first_entry)
1862 					first_entry = tmp_entry;
1863 				else
1864 					first_entry = NULL;
1865 				entry = tmp_entry;
1866 			}
1867 			last_timestamp = map->timestamp;
1868 			continue;
1869 		}
1870 		vm_map_clip_start(map, entry, start);
1871 		vm_map_clip_end(map, entry, end);
1872 		/*
1873 		 * Mark the entry in case the map lock is released.  (See
1874 		 * above.)
1875 		 */
1876 		entry->eflags |= MAP_ENTRY_IN_TRANSITION;
1877 		/*
1878 		 * Check the map for holes in the specified region.
1879 		 * If VM_MAP_WIRE_HOLESOK was specified, skip this check.
1880 		 */
1881 		if (((flags & VM_MAP_WIRE_HOLESOK) == 0) &&
1882 		    (entry->end < end && (entry->next == &map->header ||
1883 		    entry->next->start > entry->end))) {
1884 			end = entry->end;
1885 			rv = KERN_INVALID_ADDRESS;
1886 			goto done;
1887 		}
1888 		/*
1889 		 * If system unwiring, require that the entry is system wired.
1890 		 */
1891 		if (!user_unwire &&
1892 		    vm_map_entry_system_wired_count(entry) == 0) {
1893 			end = entry->end;
1894 			rv = KERN_INVALID_ARGUMENT;
1895 			goto done;
1896 		}
1897 		entry = entry->next;
1898 	}
1899 	rv = KERN_SUCCESS;
1900 done:
1901 	need_wakeup = FALSE;
1902 	if (first_entry == NULL) {
1903 		result = vm_map_lookup_entry(map, start, &first_entry);
1904 		if (!result && (flags & VM_MAP_WIRE_HOLESOK))
1905 			first_entry = first_entry->next;
1906 		else
1907 			KASSERT(result, ("vm_map_unwire: lookup failed"));
1908 	}
1909 	entry = first_entry;
1910 	while (entry != &map->header && entry->start < end) {
1911 		if (rv == KERN_SUCCESS && (!user_unwire ||
1912 		    (entry->eflags & MAP_ENTRY_USER_WIRED))) {
1913 			if (user_unwire)
1914 				entry->eflags &= ~MAP_ENTRY_USER_WIRED;
1915 			entry->wired_count--;
1916 			if (entry->wired_count == 0) {
1917 				/*
1918 				 * Retain the map lock.
1919 				 */
1920 				vm_fault_unwire(map, entry->start, entry->end,
1921 				    entry->object.vm_object != NULL &&
1922 				    entry->object.vm_object->type == OBJT_DEVICE);
1923 			}
1924 		}
1925 		KASSERT(entry->eflags & MAP_ENTRY_IN_TRANSITION,
1926 			("vm_map_unwire: in-transition flag missing"));
1927 		entry->eflags &= ~MAP_ENTRY_IN_TRANSITION;
1928 		if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
1929 			entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
1930 			need_wakeup = TRUE;
1931 		}
1932 		vm_map_simplify_entry(map, entry);
1933 		entry = entry->next;
1934 	}
1935 	vm_map_unlock(map);
1936 	if (need_wakeup)
1937 		vm_map_wakeup(map);
1938 	return (rv);
1939 }
1940 
1941 /*
1942  *	vm_map_wire:
1943  *
1944  *	Implements both kernel and user wiring.
1945  */
1946 int
1947 vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t end,
1948     int flags)
1949 {
1950 	vm_map_entry_t entry, first_entry, tmp_entry;
1951 	vm_offset_t saved_end, saved_start;
1952 	unsigned int last_timestamp;
1953 	int rv;
1954 	boolean_t fictitious, need_wakeup, result, user_wire;
1955 
1956 	user_wire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE;
1957 	vm_map_lock(map);
1958 	VM_MAP_RANGE_CHECK(map, start, end);
1959 	if (!vm_map_lookup_entry(map, start, &first_entry)) {
1960 		if (flags & VM_MAP_WIRE_HOLESOK)
1961 			first_entry = first_entry->next;
1962 		else {
1963 			vm_map_unlock(map);
1964 			return (KERN_INVALID_ADDRESS);
1965 		}
1966 	}
1967 	last_timestamp = map->timestamp;
1968 	entry = first_entry;
1969 	while (entry != &map->header && entry->start < end) {
1970 		if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
1971 			/*
1972 			 * We have not yet clipped the entry.
1973 			 */
1974 			saved_start = (start >= entry->start) ? start :
1975 			    entry->start;
1976 			entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
1977 			if (vm_map_unlock_and_wait(map, user_wire)) {
1978 				/*
1979 				 * Allow interruption of user wiring?
1980 				 */
1981 			}
1982 			vm_map_lock(map);
1983 			if (last_timestamp + 1 != map->timestamp) {
1984 				/*
1985 				 * Look again for the entry because the map was
1986 				 * modified while it was unlocked.
1987 				 * Specifically, the entry may have been
1988 				 * clipped, merged, or deleted.
1989 				 */
1990 				if (!vm_map_lookup_entry(map, saved_start,
1991 				    &tmp_entry)) {
1992 					if (flags & VM_MAP_WIRE_HOLESOK)
1993 						tmp_entry = tmp_entry->next;
1994 					else {
1995 						if (saved_start == start) {
1996 							/*
1997 							 * first_entry has been deleted.
1998 							 */
1999 							vm_map_unlock(map);
2000 							return (KERN_INVALID_ADDRESS);
2001 						}
2002 						end = saved_start;
2003 						rv = KERN_INVALID_ADDRESS;
2004 						goto done;
2005 					}
2006 				}
2007 				if (entry == first_entry)
2008 					first_entry = tmp_entry;
2009 				else
2010 					first_entry = NULL;
2011 				entry = tmp_entry;
2012 			}
2013 			last_timestamp = map->timestamp;
2014 			continue;
2015 		}
2016 		vm_map_clip_start(map, entry, start);
2017 		vm_map_clip_end(map, entry, end);
2018 		/*
2019 		 * Mark the entry in case the map lock is released.  (See
2020 		 * above.)
2021 		 */
2022 		entry->eflags |= MAP_ENTRY_IN_TRANSITION;
2023 		/*
2024 		 *
2025 		 */
2026 		if (entry->wired_count == 0) {
2027 			entry->wired_count++;
2028 			saved_start = entry->start;
2029 			saved_end = entry->end;
2030 			fictitious = entry->object.vm_object != NULL &&
2031 			    entry->object.vm_object->type == OBJT_DEVICE;
2032 			/*
2033 			 * Release the map lock, relying on the in-transition
2034 			 * mark.
2035 			 */
2036 			vm_map_unlock(map);
2037 			rv = vm_fault_wire(map, saved_start, saved_end,
2038 			    user_wire, fictitious);
2039 			vm_map_lock(map);
2040 			if (last_timestamp + 1 != map->timestamp) {
2041 				/*
2042 				 * Look again for the entry because the map was
2043 				 * modified while it was unlocked.  The entry
2044 				 * may have been clipped, but NOT merged or
2045 				 * deleted.
2046 				 */
2047 				result = vm_map_lookup_entry(map, saved_start,
2048 				    &tmp_entry);
2049 				KASSERT(result, ("vm_map_wire: lookup failed"));
2050 				if (entry == first_entry)
2051 					first_entry = tmp_entry;
2052 				else
2053 					first_entry = NULL;
2054 				entry = tmp_entry;
2055 				while (entry->end < saved_end) {
2056 					if (rv != KERN_SUCCESS) {
2057 						KASSERT(entry->wired_count == 1,
2058 						    ("vm_map_wire: bad count"));
2059 						entry->wired_count = -1;
2060 					}
2061 					entry = entry->next;
2062 				}
2063 			}
2064 			last_timestamp = map->timestamp;
2065 			if (rv != KERN_SUCCESS) {
2066 				KASSERT(entry->wired_count == 1,
2067 				    ("vm_map_wire: bad count"));
2068 				/*
2069 				 * Assign an out-of-range value to represent
2070 				 * the failure to wire this entry.
2071 				 */
2072 				entry->wired_count = -1;
2073 				end = entry->end;
2074 				goto done;
2075 			}
2076 		} else if (!user_wire ||
2077 			   (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
2078 			entry->wired_count++;
2079 		}
2080 		/*
2081 		 * Check the map for holes in the specified region.
2082 		 * If VM_MAP_WIRE_HOLESOK was specified, skip this check.
2083 		 */
2084 		if (((flags & VM_MAP_WIRE_HOLESOK) == 0) &&
2085 		    (entry->end < end && (entry->next == &map->header ||
2086 		    entry->next->start > entry->end))) {
2087 			end = entry->end;
2088 			rv = KERN_INVALID_ADDRESS;
2089 			goto done;
2090 		}
2091 		entry = entry->next;
2092 	}
2093 	rv = KERN_SUCCESS;
2094 done:
2095 	need_wakeup = FALSE;
2096 	if (first_entry == NULL) {
2097 		result = vm_map_lookup_entry(map, start, &first_entry);
2098 		if (!result && (flags & VM_MAP_WIRE_HOLESOK))
2099 			first_entry = first_entry->next;
2100 		else
2101 			KASSERT(result, ("vm_map_wire: lookup failed"));
2102 	}
2103 	entry = first_entry;
2104 	while (entry != &map->header && entry->start < end) {
2105 		if (rv == KERN_SUCCESS) {
2106 			if (user_wire)
2107 				entry->eflags |= MAP_ENTRY_USER_WIRED;
2108 		} else if (entry->wired_count == -1) {
2109 			/*
2110 			 * Wiring failed on this entry.  Thus, unwiring is
2111 			 * unnecessary.
2112 			 */
2113 			entry->wired_count = 0;
2114 		} else {
2115 			if (!user_wire ||
2116 			    (entry->eflags & MAP_ENTRY_USER_WIRED) == 0)
2117 				entry->wired_count--;
2118 			if (entry->wired_count == 0) {
2119 				/*
2120 				 * Retain the map lock.
2121 				 */
2122 				vm_fault_unwire(map, entry->start, entry->end,
2123 				    entry->object.vm_object != NULL &&
2124 				    entry->object.vm_object->type == OBJT_DEVICE);
2125 			}
2126 		}
2127 		KASSERT(entry->eflags & MAP_ENTRY_IN_TRANSITION,
2128 			("vm_map_wire: in-transition flag missing"));
2129 		entry->eflags &= ~MAP_ENTRY_IN_TRANSITION;
2130 		if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
2131 			entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
2132 			need_wakeup = TRUE;
2133 		}
2134 		vm_map_simplify_entry(map, entry);
2135 		entry = entry->next;
2136 	}
2137 	vm_map_unlock(map);
2138 	if (need_wakeup)
2139 		vm_map_wakeup(map);
2140 	return (rv);
2141 }
2142 
2143 /*
2144  * vm_map_sync
2145  *
2146  * Push any dirty cached pages in the address range to their pager.
2147  * If syncio is TRUE, dirty pages are written synchronously.
2148  * If invalidate is TRUE, any cached pages are freed as well.
2149  *
2150  * If the size of the region from start to end is zero, we are
2151  * supposed to flush all modified pages within the region containing
2152  * start.  Unfortunately, a region can be split or coalesced with
2153  * neighboring regions, making it difficult to determine what the
2154  * original region was.  Therefore, we approximate this requirement by
2155  * flushing the current region containing start.
2156  *
2157  * Returns an error if any part of the specified range is not mapped.
2158  */
2159 int
2160 vm_map_sync(
2161 	vm_map_t map,
2162 	vm_offset_t start,
2163 	vm_offset_t end,
2164 	boolean_t syncio,
2165 	boolean_t invalidate)
2166 {
2167 	vm_map_entry_t current;
2168 	vm_map_entry_t entry;
2169 	vm_size_t size;
2170 	vm_object_t object;
2171 	vm_ooffset_t offset;
2172 
2173 	vm_map_lock_read(map);
2174 	VM_MAP_RANGE_CHECK(map, start, end);
2175 	if (!vm_map_lookup_entry(map, start, &entry)) {
2176 		vm_map_unlock_read(map);
2177 		return (KERN_INVALID_ADDRESS);
2178 	} else if (start == end) {
2179 		start = entry->start;
2180 		end = entry->end;
2181 	}
2182 	/*
2183 	 * Make a first pass to check for user-wired memory and holes.
2184 	 */
2185 	for (current = entry; current->start < end; current = current->next) {
2186 		if (invalidate && (current->eflags & MAP_ENTRY_USER_WIRED)) {
2187 			vm_map_unlock_read(map);
2188 			return (KERN_INVALID_ARGUMENT);
2189 		}
2190 		if (end > current->end &&
2191 		    (current->next == &map->header ||
2192 			current->end != current->next->start)) {
2193 			vm_map_unlock_read(map);
2194 			return (KERN_INVALID_ADDRESS);
2195 		}
2196 	}
2197 
2198 	if (invalidate)
2199 		pmap_remove(map->pmap, start, end);
2200 
2201 	/*
2202 	 * Make a second pass, cleaning/uncaching pages from the indicated
2203 	 * objects as we go.
2204 	 */
2205 	for (current = entry; current->start < end; current = current->next) {
2206 		offset = current->offset + (start - current->start);
2207 		size = (end <= current->end ? end : current->end) - start;
2208 		if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
2209 			vm_map_t smap;
2210 			vm_map_entry_t tentry;
2211 			vm_size_t tsize;
2212 
2213 			smap = current->object.sub_map;
2214 			vm_map_lock_read(smap);
2215 			(void) vm_map_lookup_entry(smap, offset, &tentry);
2216 			tsize = tentry->end - offset;
2217 			if (tsize < size)
2218 				size = tsize;
2219 			object = tentry->object.vm_object;
2220 			offset = tentry->offset + (offset - tentry->start);
2221 			vm_map_unlock_read(smap);
2222 		} else {
2223 			object = current->object.vm_object;
2224 		}
2225 		vm_object_sync(object, offset, size, syncio, invalidate);
2226 		start += size;
2227 	}
2228 
2229 	vm_map_unlock_read(map);
2230 	return (KERN_SUCCESS);
2231 }
2232 
2233 /*
2234  *	vm_map_entry_unwire:	[ internal use only ]
2235  *
2236  *	Make the region specified by this entry pageable.
2237  *
2238  *	The map in question should be locked.
2239  *	[This is the reason for this routine's existence.]
2240  */
2241 static void
2242 vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry)
2243 {
2244 	vm_fault_unwire(map, entry->start, entry->end,
2245 	    entry->object.vm_object != NULL &&
2246 	    entry->object.vm_object->type == OBJT_DEVICE);
2247 	entry->wired_count = 0;
2248 }
2249 
2250 /*
2251  *	vm_map_entry_delete:	[ internal use only ]
2252  *
2253  *	Deallocate the given entry from the target map.
2254  */
2255 static void
2256 vm_map_entry_delete(vm_map_t map, vm_map_entry_t entry)
2257 {
2258 	vm_object_t object;
2259 	vm_pindex_t offidxstart, offidxend, count;
2260 
2261 	vm_map_entry_unlink(map, entry);
2262 	map->size -= entry->end - entry->start;
2263 
2264 	if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0 &&
2265 	    (object = entry->object.vm_object) != NULL) {
2266 		count = OFF_TO_IDX(entry->end - entry->start);
2267 		offidxstart = OFF_TO_IDX(entry->offset);
2268 		offidxend = offidxstart + count;
2269 		VM_OBJECT_LOCK(object);
2270 		if (object->ref_count != 1 &&
2271 		    ((object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING ||
2272 		     object == kernel_object || object == kmem_object) &&
2273 		    (object->type == OBJT_DEFAULT || object->type == OBJT_SWAP)) {
2274 			vm_object_collapse(object);
2275 			vm_object_page_remove(object, offidxstart, offidxend, FALSE);
2276 			if (object->type == OBJT_SWAP)
2277 				swap_pager_freespace(object, offidxstart, count);
2278 			if (offidxend >= object->size &&
2279 			    offidxstart < object->size)
2280 				object->size = offidxstart;
2281 		}
2282 		VM_OBJECT_UNLOCK(object);
2283 		vm_object_deallocate(object);
2284 	}
2285 
2286 	vm_map_entry_dispose(map, entry);
2287 }
2288 
2289 /*
2290  *	vm_map_delete:	[ internal use only ]
2291  *
2292  *	Deallocates the given address range from the target
2293  *	map.
2294  */
2295 int
2296 vm_map_delete(vm_map_t map, vm_offset_t start, vm_offset_t end)
2297 {
2298 	vm_map_entry_t entry;
2299 	vm_map_entry_t first_entry;
2300 
2301 	/*
2302 	 * Find the start of the region, and clip it
2303 	 */
2304 	if (!vm_map_lookup_entry(map, start, &first_entry))
2305 		entry = first_entry->next;
2306 	else {
2307 		entry = first_entry;
2308 		vm_map_clip_start(map, entry, start);
2309 	}
2310 
2311 	/*
2312 	 * Step through all entries in this region
2313 	 */
2314 	while ((entry != &map->header) && (entry->start < end)) {
2315 		vm_map_entry_t next;
2316 
2317 		/*
2318 		 * Wait for wiring or unwiring of an entry to complete.
2319 		 * Also wait for any system wirings to disappear on
2320 		 * user maps.
2321 		 */
2322 		if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 ||
2323 		    (vm_map_pmap(map) != kernel_pmap &&
2324 		    vm_map_entry_system_wired_count(entry) != 0)) {
2325 			unsigned int last_timestamp;
2326 			vm_offset_t saved_start;
2327 			vm_map_entry_t tmp_entry;
2328 
2329 			saved_start = entry->start;
2330 			entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2331 			last_timestamp = map->timestamp;
2332 			(void) vm_map_unlock_and_wait(map, FALSE);
2333 			vm_map_lock(map);
2334 			if (last_timestamp + 1 != map->timestamp) {
2335 				/*
2336 				 * Look again for the entry because the map was
2337 				 * modified while it was unlocked.
2338 				 * Specifically, the entry may have been
2339 				 * clipped, merged, or deleted.
2340 				 */
2341 				if (!vm_map_lookup_entry(map, saved_start,
2342 							 &tmp_entry))
2343 					entry = tmp_entry->next;
2344 				else {
2345 					entry = tmp_entry;
2346 					vm_map_clip_start(map, entry,
2347 							  saved_start);
2348 				}
2349 			}
2350 			continue;
2351 		}
2352 		vm_map_clip_end(map, entry, end);
2353 
2354 		next = entry->next;
2355 
2356 		/*
2357 		 * Unwire before removing addresses from the pmap; otherwise,
2358 		 * unwiring will put the entries back in the pmap.
2359 		 */
2360 		if (entry->wired_count != 0) {
2361 			vm_map_entry_unwire(map, entry);
2362 		}
2363 
2364 		pmap_remove(map->pmap, entry->start, entry->end);
2365 
2366 		/*
2367 		 * Delete the entry (which may delete the object) only after
2368 		 * removing all pmap entries pointing to its pages.
2369 		 * (Otherwise, its page frames may be reallocated, and any
2370 		 * modify bits will be set in the wrong object!)
2371 		 */
2372 		vm_map_entry_delete(map, entry);
2373 		entry = next;
2374 	}
2375 	return (KERN_SUCCESS);
2376 }
2377 
2378 /*
2379  *	vm_map_remove:
2380  *
2381  *	Remove the given address range from the target map.
2382  *	This is the exported form of vm_map_delete.
2383  */
2384 int
2385 vm_map_remove(vm_map_t map, vm_offset_t start, vm_offset_t end)
2386 {
2387 	int result;
2388 
2389 	vm_map_lock(map);
2390 	VM_MAP_RANGE_CHECK(map, start, end);
2391 	result = vm_map_delete(map, start, end);
2392 	vm_map_unlock(map);
2393 	return (result);
2394 }
2395 
2396 /*
2397  *	vm_map_check_protection:
2398  *
2399  *	Assert that the target map allows the specified privilege on the
2400  *	entire address region given.  The entire region must be allocated.
2401  *
2402  *	WARNING!  This code does not and should not check whether the
2403  *	contents of the region is accessible.  For example a smaller file
2404  *	might be mapped into a larger address space.
2405  *
2406  *	NOTE!  This code is also called by munmap().
2407  *
2408  *	The map must be locked.  A read lock is sufficient.
2409  */
2410 boolean_t
2411 vm_map_check_protection(vm_map_t map, vm_offset_t start, vm_offset_t end,
2412 			vm_prot_t protection)
2413 {
2414 	vm_map_entry_t entry;
2415 	vm_map_entry_t tmp_entry;
2416 
2417 	if (!vm_map_lookup_entry(map, start, &tmp_entry))
2418 		return (FALSE);
2419 	entry = tmp_entry;
2420 
2421 	while (start < end) {
2422 		if (entry == &map->header)
2423 			return (FALSE);
2424 		/*
2425 		 * No holes allowed!
2426 		 */
2427 		if (start < entry->start)
2428 			return (FALSE);
2429 		/*
2430 		 * Check protection associated with entry.
2431 		 */
2432 		if ((entry->protection & protection) != protection)
2433 			return (FALSE);
2434 		/* go to next entry */
2435 		start = entry->end;
2436 		entry = entry->next;
2437 	}
2438 	return (TRUE);
2439 }
2440 
2441 /*
2442  *	vm_map_copy_entry:
2443  *
2444  *	Copies the contents of the source entry to the destination
2445  *	entry.  The entries *must* be aligned properly.
2446  */
2447 static void
2448 vm_map_copy_entry(
2449 	vm_map_t src_map,
2450 	vm_map_t dst_map,
2451 	vm_map_entry_t src_entry,
2452 	vm_map_entry_t dst_entry)
2453 {
2454 	vm_object_t src_object;
2455 
2456 	if ((dst_entry->eflags|src_entry->eflags) & MAP_ENTRY_IS_SUB_MAP)
2457 		return;
2458 
2459 	if (src_entry->wired_count == 0) {
2460 
2461 		/*
2462 		 * If the source entry is marked needs_copy, it is already
2463 		 * write-protected.
2464 		 */
2465 		if ((src_entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) {
2466 			pmap_protect(src_map->pmap,
2467 			    src_entry->start,
2468 			    src_entry->end,
2469 			    src_entry->protection & ~VM_PROT_WRITE);
2470 		}
2471 
2472 		/*
2473 		 * Make a copy of the object.
2474 		 */
2475 		if ((src_object = src_entry->object.vm_object) != NULL) {
2476 			VM_OBJECT_LOCK(src_object);
2477 			if ((src_object->handle == NULL) &&
2478 				(src_object->type == OBJT_DEFAULT ||
2479 				 src_object->type == OBJT_SWAP)) {
2480 				vm_object_collapse(src_object);
2481 				if ((src_object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING) {
2482 					vm_object_split(src_entry);
2483 					src_object = src_entry->object.vm_object;
2484 				}
2485 			}
2486 			vm_object_reference_locked(src_object);
2487 			vm_object_clear_flag(src_object, OBJ_ONEMAPPING);
2488 			VM_OBJECT_UNLOCK(src_object);
2489 			dst_entry->object.vm_object = src_object;
2490 			src_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY);
2491 			dst_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY);
2492 			dst_entry->offset = src_entry->offset;
2493 		} else {
2494 			dst_entry->object.vm_object = NULL;
2495 			dst_entry->offset = 0;
2496 		}
2497 
2498 		pmap_copy(dst_map->pmap, src_map->pmap, dst_entry->start,
2499 		    dst_entry->end - dst_entry->start, src_entry->start);
2500 	} else {
2501 		/*
2502 		 * Of course, wired down pages can't be set copy-on-write.
2503 		 * Cause wired pages to be copied into the new map by
2504 		 * simulating faults (the new pages are pageable)
2505 		 */
2506 		vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry);
2507 	}
2508 }
2509 
2510 /*
2511  * vmspace_map_entry_forked:
2512  * Update the newly-forked vmspace each time a map entry is inherited
2513  * or copied.  The values for vm_dsize and vm_tsize are approximate
2514  * (and mostly-obsolete ideas in the face of mmap(2) et al.)
2515  */
2516 static void
2517 vmspace_map_entry_forked(const struct vmspace *vm1, struct vmspace *vm2,
2518     vm_map_entry_t entry)
2519 {
2520 	vm_size_t entrysize;
2521 	vm_offset_t newend;
2522 
2523 	entrysize = entry->end - entry->start;
2524 	vm2->vm_map.size += entrysize;
2525 	if (entry->eflags & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP)) {
2526 		vm2->vm_ssize += btoc(entrysize);
2527 	} else if (entry->start >= (vm_offset_t)vm1->vm_daddr &&
2528 	    entry->start < (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)) {
2529 		newend = MIN(entry->end,
2530 		    (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize));
2531 		vm2->vm_dsize += btoc(newend - entry->start);
2532 	} else if (entry->start >= (vm_offset_t)vm1->vm_taddr &&
2533 	    entry->start < (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)) {
2534 		newend = MIN(entry->end,
2535 		    (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize));
2536 		vm2->vm_tsize += btoc(newend - entry->start);
2537 	}
2538 }
2539 
2540 /*
2541  * vmspace_fork:
2542  * Create a new process vmspace structure and vm_map
2543  * based on those of an existing process.  The new map
2544  * is based on the old map, according to the inheritance
2545  * values on the regions in that map.
2546  *
2547  * XXX It might be worth coalescing the entries added to the new vmspace.
2548  *
2549  * The source map must not be locked.
2550  */
2551 struct vmspace *
2552 vmspace_fork(struct vmspace *vm1)
2553 {
2554 	struct vmspace *vm2;
2555 	vm_map_t old_map = &vm1->vm_map;
2556 	vm_map_t new_map;
2557 	vm_map_entry_t old_entry;
2558 	vm_map_entry_t new_entry;
2559 	vm_object_t object;
2560 
2561 	vm_map_lock(old_map);
2562 
2563 	vm2 = vmspace_alloc(old_map->min_offset, old_map->max_offset);
2564 	vm2->vm_taddr = vm1->vm_taddr;
2565 	vm2->vm_daddr = vm1->vm_daddr;
2566 	vm2->vm_maxsaddr = vm1->vm_maxsaddr;
2567 	new_map = &vm2->vm_map;	/* XXX */
2568 	new_map->timestamp = 1;
2569 
2570 	/* Do not inherit the MAP_WIREFUTURE property. */
2571 	if ((new_map->flags & MAP_WIREFUTURE) == MAP_WIREFUTURE)
2572 		new_map->flags &= ~MAP_WIREFUTURE;
2573 
2574 	old_entry = old_map->header.next;
2575 
2576 	while (old_entry != &old_map->header) {
2577 		if (old_entry->eflags & MAP_ENTRY_IS_SUB_MAP)
2578 			panic("vm_map_fork: encountered a submap");
2579 
2580 		switch (old_entry->inheritance) {
2581 		case VM_INHERIT_NONE:
2582 			break;
2583 
2584 		case VM_INHERIT_SHARE:
2585 			/*
2586 			 * Clone the entry, creating the shared object if necessary.
2587 			 */
2588 			object = old_entry->object.vm_object;
2589 			if (object == NULL) {
2590 				object = vm_object_allocate(OBJT_DEFAULT,
2591 					atop(old_entry->end - old_entry->start));
2592 				old_entry->object.vm_object = object;
2593 				old_entry->offset = 0;
2594 			}
2595 
2596 			/*
2597 			 * Add the reference before calling vm_object_shadow
2598 			 * to insure that a shadow object is created.
2599 			 */
2600 			vm_object_reference(object);
2601 			if (old_entry->eflags & MAP_ENTRY_NEEDS_COPY) {
2602 				vm_object_shadow(&old_entry->object.vm_object,
2603 					&old_entry->offset,
2604 					atop(old_entry->end - old_entry->start));
2605 				old_entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
2606 				/* Transfer the second reference too. */
2607 				vm_object_reference(
2608 				    old_entry->object.vm_object);
2609 				vm_object_deallocate(object);
2610 				object = old_entry->object.vm_object;
2611 			}
2612 			VM_OBJECT_LOCK(object);
2613 			vm_object_clear_flag(object, OBJ_ONEMAPPING);
2614 			VM_OBJECT_UNLOCK(object);
2615 
2616 			/*
2617 			 * Clone the entry, referencing the shared object.
2618 			 */
2619 			new_entry = vm_map_entry_create(new_map);
2620 			*new_entry = *old_entry;
2621 			new_entry->eflags &= ~MAP_ENTRY_USER_WIRED;
2622 			new_entry->wired_count = 0;
2623 
2624 			/*
2625 			 * Insert the entry into the new map -- we know we're
2626 			 * inserting at the end of the new map.
2627 			 */
2628 			vm_map_entry_link(new_map, new_map->header.prev,
2629 			    new_entry);
2630 			vmspace_map_entry_forked(vm1, vm2, new_entry);
2631 
2632 			/*
2633 			 * Update the physical map
2634 			 */
2635 			pmap_copy(new_map->pmap, old_map->pmap,
2636 			    new_entry->start,
2637 			    (old_entry->end - old_entry->start),
2638 			    old_entry->start);
2639 			break;
2640 
2641 		case VM_INHERIT_COPY:
2642 			/*
2643 			 * Clone the entry and link into the map.
2644 			 */
2645 			new_entry = vm_map_entry_create(new_map);
2646 			*new_entry = *old_entry;
2647 			new_entry->eflags &= ~MAP_ENTRY_USER_WIRED;
2648 			new_entry->wired_count = 0;
2649 			new_entry->object.vm_object = NULL;
2650 			vm_map_entry_link(new_map, new_map->header.prev,
2651 			    new_entry);
2652 			vmspace_map_entry_forked(vm1, vm2, new_entry);
2653 			vm_map_copy_entry(old_map, new_map, old_entry,
2654 			    new_entry);
2655 			break;
2656 		}
2657 		old_entry = old_entry->next;
2658 	}
2659 
2660 	vm_map_unlock(old_map);
2661 
2662 	return (vm2);
2663 }
2664 
2665 int
2666 vm_map_stack(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
2667     vm_prot_t prot, vm_prot_t max, int cow)
2668 {
2669 	vm_map_entry_t new_entry, prev_entry;
2670 	vm_offset_t bot, top;
2671 	vm_size_t init_ssize;
2672 	int orient, rv;
2673 	rlim_t vmemlim;
2674 
2675 	/*
2676 	 * The stack orientation is piggybacked with the cow argument.
2677 	 * Extract it into orient and mask the cow argument so that we
2678 	 * don't pass it around further.
2679 	 * NOTE: We explicitly allow bi-directional stacks.
2680 	 */
2681 	orient = cow & (MAP_STACK_GROWS_DOWN|MAP_STACK_GROWS_UP);
2682 	cow &= ~orient;
2683 	KASSERT(orient != 0, ("No stack grow direction"));
2684 
2685 	if (addrbos < vm_map_min(map) || addrbos > map->max_offset)
2686 		return (KERN_NO_SPACE);
2687 
2688 	init_ssize = (max_ssize < sgrowsiz) ? max_ssize : sgrowsiz;
2689 
2690 	PROC_LOCK(curthread->td_proc);
2691 	vmemlim = lim_cur(curthread->td_proc, RLIMIT_VMEM);
2692 	PROC_UNLOCK(curthread->td_proc);
2693 
2694 	vm_map_lock(map);
2695 
2696 	/* If addr is already mapped, no go */
2697 	if (vm_map_lookup_entry(map, addrbos, &prev_entry)) {
2698 		vm_map_unlock(map);
2699 		return (KERN_NO_SPACE);
2700 	}
2701 
2702 	/* If we would blow our VMEM resource limit, no go */
2703 	if (map->size + init_ssize > vmemlim) {
2704 		vm_map_unlock(map);
2705 		return (KERN_NO_SPACE);
2706 	}
2707 
2708 	/*
2709 	 * If we can't accomodate max_ssize in the current mapping, no go.
2710 	 * However, we need to be aware that subsequent user mappings might
2711 	 * map into the space we have reserved for stack, and currently this
2712 	 * space is not protected.
2713 	 *
2714 	 * Hopefully we will at least detect this condition when we try to
2715 	 * grow the stack.
2716 	 */
2717 	if ((prev_entry->next != &map->header) &&
2718 	    (prev_entry->next->start < addrbos + max_ssize)) {
2719 		vm_map_unlock(map);
2720 		return (KERN_NO_SPACE);
2721 	}
2722 
2723 	/*
2724 	 * We initially map a stack of only init_ssize.  We will grow as
2725 	 * needed later.  Depending on the orientation of the stack (i.e.
2726 	 * the grow direction) we either map at the top of the range, the
2727 	 * bottom of the range or in the middle.
2728 	 *
2729 	 * Note: we would normally expect prot and max to be VM_PROT_ALL,
2730 	 * and cow to be 0.  Possibly we should eliminate these as input
2731 	 * parameters, and just pass these values here in the insert call.
2732 	 */
2733 	if (orient == MAP_STACK_GROWS_DOWN)
2734 		bot = addrbos + max_ssize - init_ssize;
2735 	else if (orient == MAP_STACK_GROWS_UP)
2736 		bot = addrbos;
2737 	else
2738 		bot = round_page(addrbos + max_ssize/2 - init_ssize/2);
2739 	top = bot + init_ssize;
2740 	rv = vm_map_insert(map, NULL, 0, bot, top, prot, max, cow);
2741 
2742 	/* Now set the avail_ssize amount. */
2743 	if (rv == KERN_SUCCESS) {
2744 		if (prev_entry != &map->header)
2745 			vm_map_clip_end(map, prev_entry, bot);
2746 		new_entry = prev_entry->next;
2747 		if (new_entry->end != top || new_entry->start != bot)
2748 			panic("Bad entry start/end for new stack entry");
2749 
2750 		new_entry->avail_ssize = max_ssize - init_ssize;
2751 		if (orient & MAP_STACK_GROWS_DOWN)
2752 			new_entry->eflags |= MAP_ENTRY_GROWS_DOWN;
2753 		if (orient & MAP_STACK_GROWS_UP)
2754 			new_entry->eflags |= MAP_ENTRY_GROWS_UP;
2755 	}
2756 
2757 	vm_map_unlock(map);
2758 	return (rv);
2759 }
2760 
2761 /* Attempts to grow a vm stack entry.  Returns KERN_SUCCESS if the
2762  * desired address is already mapped, or if we successfully grow
2763  * the stack.  Also returns KERN_SUCCESS if addr is outside the
2764  * stack range (this is strange, but preserves compatibility with
2765  * the grow function in vm_machdep.c).
2766  */
2767 int
2768 vm_map_growstack(struct proc *p, vm_offset_t addr)
2769 {
2770 	vm_map_entry_t next_entry, prev_entry;
2771 	vm_map_entry_t new_entry, stack_entry;
2772 	struct vmspace *vm = p->p_vmspace;
2773 	vm_map_t map = &vm->vm_map;
2774 	vm_offset_t end;
2775 	size_t grow_amount, max_grow;
2776 	rlim_t stacklim, vmemlim;
2777 	int is_procstack, rv;
2778 
2779 Retry:
2780 	PROC_LOCK(p);
2781 	stacklim = lim_cur(p, RLIMIT_STACK);
2782 	vmemlim = lim_cur(p, RLIMIT_VMEM);
2783 	PROC_UNLOCK(p);
2784 
2785 	vm_map_lock_read(map);
2786 
2787 	/* If addr is already in the entry range, no need to grow.*/
2788 	if (vm_map_lookup_entry(map, addr, &prev_entry)) {
2789 		vm_map_unlock_read(map);
2790 		return (KERN_SUCCESS);
2791 	}
2792 
2793 	next_entry = prev_entry->next;
2794 	if (!(prev_entry->eflags & MAP_ENTRY_GROWS_UP)) {
2795 		/*
2796 		 * This entry does not grow upwards. Since the address lies
2797 		 * beyond this entry, the next entry (if one exists) has to
2798 		 * be a downward growable entry. The entry list header is
2799 		 * never a growable entry, so it suffices to check the flags.
2800 		 */
2801 		if (!(next_entry->eflags & MAP_ENTRY_GROWS_DOWN)) {
2802 			vm_map_unlock_read(map);
2803 			return (KERN_SUCCESS);
2804 		}
2805 		stack_entry = next_entry;
2806 	} else {
2807 		/*
2808 		 * This entry grows upward. If the next entry does not at
2809 		 * least grow downwards, this is the entry we need to grow.
2810 		 * otherwise we have two possible choices and we have to
2811 		 * select one.
2812 		 */
2813 		if (next_entry->eflags & MAP_ENTRY_GROWS_DOWN) {
2814 			/*
2815 			 * We have two choices; grow the entry closest to
2816 			 * the address to minimize the amount of growth.
2817 			 */
2818 			if (addr - prev_entry->end <= next_entry->start - addr)
2819 				stack_entry = prev_entry;
2820 			else
2821 				stack_entry = next_entry;
2822 		} else
2823 			stack_entry = prev_entry;
2824 	}
2825 
2826 	if (stack_entry == next_entry) {
2827 		KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_DOWN, ("foo"));
2828 		KASSERT(addr < stack_entry->start, ("foo"));
2829 		end = (prev_entry != &map->header) ? prev_entry->end :
2830 		    stack_entry->start - stack_entry->avail_ssize;
2831 		grow_amount = roundup(stack_entry->start - addr, PAGE_SIZE);
2832 		max_grow = stack_entry->start - end;
2833 	} else {
2834 		KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_UP, ("foo"));
2835 		KASSERT(addr >= stack_entry->end, ("foo"));
2836 		end = (next_entry != &map->header) ? next_entry->start :
2837 		    stack_entry->end + stack_entry->avail_ssize;
2838 		grow_amount = roundup(addr + 1 - stack_entry->end, PAGE_SIZE);
2839 		max_grow = end - stack_entry->end;
2840 	}
2841 
2842 	if (grow_amount > stack_entry->avail_ssize) {
2843 		vm_map_unlock_read(map);
2844 		return (KERN_NO_SPACE);
2845 	}
2846 
2847 	/*
2848 	 * If there is no longer enough space between the entries nogo, and
2849 	 * adjust the available space.  Note: this  should only happen if the
2850 	 * user has mapped into the stack area after the stack was created,
2851 	 * and is probably an error.
2852 	 *
2853 	 * This also effectively destroys any guard page the user might have
2854 	 * intended by limiting the stack size.
2855 	 */
2856 	if (grow_amount > max_grow) {
2857 		if (vm_map_lock_upgrade(map))
2858 			goto Retry;
2859 
2860 		stack_entry->avail_ssize = max_grow;
2861 
2862 		vm_map_unlock(map);
2863 		return (KERN_NO_SPACE);
2864 	}
2865 
2866 	is_procstack = (addr >= (vm_offset_t)vm->vm_maxsaddr) ? 1 : 0;
2867 
2868 	/*
2869 	 * If this is the main process stack, see if we're over the stack
2870 	 * limit.
2871 	 */
2872 	if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) {
2873 		vm_map_unlock_read(map);
2874 		return (KERN_NO_SPACE);
2875 	}
2876 
2877 	/* Round up the grow amount modulo SGROWSIZ */
2878 	grow_amount = roundup (grow_amount, sgrowsiz);
2879 	if (grow_amount > stack_entry->avail_ssize)
2880 		grow_amount = stack_entry->avail_ssize;
2881 	if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) {
2882 		grow_amount = stacklim - ctob(vm->vm_ssize);
2883 	}
2884 
2885 	/* If we would blow our VMEM resource limit, no go */
2886 	if (map->size + grow_amount > vmemlim) {
2887 		vm_map_unlock_read(map);
2888 		return (KERN_NO_SPACE);
2889 	}
2890 
2891 	if (vm_map_lock_upgrade(map))
2892 		goto Retry;
2893 
2894 	if (stack_entry == next_entry) {
2895 		/*
2896 		 * Growing downward.
2897 		 */
2898 		/* Get the preliminary new entry start value */
2899 		addr = stack_entry->start - grow_amount;
2900 
2901 		/*
2902 		 * If this puts us into the previous entry, cut back our
2903 		 * growth to the available space. Also, see the note above.
2904 		 */
2905 		if (addr < end) {
2906 			stack_entry->avail_ssize = max_grow;
2907 			addr = end;
2908 		}
2909 
2910 		rv = vm_map_insert(map, NULL, 0, addr, stack_entry->start,
2911 		    p->p_sysent->sv_stackprot, VM_PROT_ALL, 0);
2912 
2913 		/* Adjust the available stack space by the amount we grew. */
2914 		if (rv == KERN_SUCCESS) {
2915 			if (prev_entry != &map->header)
2916 				vm_map_clip_end(map, prev_entry, addr);
2917 			new_entry = prev_entry->next;
2918 			KASSERT(new_entry == stack_entry->prev, ("foo"));
2919 			KASSERT(new_entry->end == stack_entry->start, ("foo"));
2920 			KASSERT(new_entry->start == addr, ("foo"));
2921 			grow_amount = new_entry->end - new_entry->start;
2922 			new_entry->avail_ssize = stack_entry->avail_ssize -
2923 			    grow_amount;
2924 			stack_entry->eflags &= ~MAP_ENTRY_GROWS_DOWN;
2925 			new_entry->eflags |= MAP_ENTRY_GROWS_DOWN;
2926 		}
2927 	} else {
2928 		/*
2929 		 * Growing upward.
2930 		 */
2931 		addr = stack_entry->end + grow_amount;
2932 
2933 		/*
2934 		 * If this puts us into the next entry, cut back our growth
2935 		 * to the available space. Also, see the note above.
2936 		 */
2937 		if (addr > end) {
2938 			stack_entry->avail_ssize = end - stack_entry->end;
2939 			addr = end;
2940 		}
2941 
2942 		grow_amount = addr - stack_entry->end;
2943 
2944 		/* Grow the underlying object if applicable. */
2945 		if (stack_entry->object.vm_object == NULL ||
2946 		    vm_object_coalesce(stack_entry->object.vm_object,
2947 		    stack_entry->offset,
2948 		    (vm_size_t)(stack_entry->end - stack_entry->start),
2949 		    (vm_size_t)grow_amount)) {
2950 			map->size += (addr - stack_entry->end);
2951 			/* Update the current entry. */
2952 			stack_entry->end = addr;
2953 			stack_entry->avail_ssize -= grow_amount;
2954 			vm_map_entry_resize_free(map, stack_entry);
2955 			rv = KERN_SUCCESS;
2956 
2957 			if (next_entry != &map->header)
2958 				vm_map_clip_start(map, next_entry, addr);
2959 		} else
2960 			rv = KERN_FAILURE;
2961 	}
2962 
2963 	if (rv == KERN_SUCCESS && is_procstack)
2964 		vm->vm_ssize += btoc(grow_amount);
2965 
2966 	vm_map_unlock(map);
2967 
2968 	/*
2969 	 * Heed the MAP_WIREFUTURE flag if it was set for this process.
2970 	 */
2971 	if (rv == KERN_SUCCESS && (map->flags & MAP_WIREFUTURE)) {
2972 		vm_map_wire(map,
2973 		    (stack_entry == next_entry) ? addr : addr - grow_amount,
2974 		    (stack_entry == next_entry) ? stack_entry->start : addr,
2975 		    (p->p_flag & P_SYSTEM)
2976 		    ? VM_MAP_WIRE_SYSTEM|VM_MAP_WIRE_NOHOLES
2977 		    : VM_MAP_WIRE_USER|VM_MAP_WIRE_NOHOLES);
2978 	}
2979 
2980 	return (rv);
2981 }
2982 
2983 /*
2984  * Unshare the specified VM space for exec.  If other processes are
2985  * mapped to it, then create a new one.  The new vmspace is null.
2986  */
2987 void
2988 vmspace_exec(struct proc *p, vm_offset_t minuser, vm_offset_t maxuser)
2989 {
2990 	struct vmspace *oldvmspace = p->p_vmspace;
2991 	struct vmspace *newvmspace;
2992 
2993 	newvmspace = vmspace_alloc(minuser, maxuser);
2994 	newvmspace->vm_swrss = oldvmspace->vm_swrss;
2995 	/*
2996 	 * This code is written like this for prototype purposes.  The
2997 	 * goal is to avoid running down the vmspace here, but let the
2998 	 * other process's that are still using the vmspace to finally
2999 	 * run it down.  Even though there is little or no chance of blocking
3000 	 * here, it is a good idea to keep this form for future mods.
3001 	 */
3002 	PROC_VMSPACE_LOCK(p);
3003 	p->p_vmspace = newvmspace;
3004 	PROC_VMSPACE_UNLOCK(p);
3005 	if (p == curthread->td_proc)		/* XXXKSE ? */
3006 		pmap_activate(curthread);
3007 	vmspace_free(oldvmspace);
3008 }
3009 
3010 /*
3011  * Unshare the specified VM space for forcing COW.  This
3012  * is called by rfork, for the (RFMEM|RFPROC) == 0 case.
3013  */
3014 void
3015 vmspace_unshare(struct proc *p)
3016 {
3017 	struct vmspace *oldvmspace = p->p_vmspace;
3018 	struct vmspace *newvmspace;
3019 
3020 	if (oldvmspace->vm_refcnt == 1)
3021 		return;
3022 	newvmspace = vmspace_fork(oldvmspace);
3023 	PROC_VMSPACE_LOCK(p);
3024 	p->p_vmspace = newvmspace;
3025 	PROC_VMSPACE_UNLOCK(p);
3026 	if (p == curthread->td_proc)		/* XXXKSE ? */
3027 		pmap_activate(curthread);
3028 	vmspace_free(oldvmspace);
3029 }
3030 
3031 /*
3032  *	vm_map_lookup:
3033  *
3034  *	Finds the VM object, offset, and
3035  *	protection for a given virtual address in the
3036  *	specified map, assuming a page fault of the
3037  *	type specified.
3038  *
3039  *	Leaves the map in question locked for read; return
3040  *	values are guaranteed until a vm_map_lookup_done
3041  *	call is performed.  Note that the map argument
3042  *	is in/out; the returned map must be used in
3043  *	the call to vm_map_lookup_done.
3044  *
3045  *	A handle (out_entry) is returned for use in
3046  *	vm_map_lookup_done, to make that fast.
3047  *
3048  *	If a lookup is requested with "write protection"
3049  *	specified, the map may be changed to perform virtual
3050  *	copying operations, although the data referenced will
3051  *	remain the same.
3052  */
3053 int
3054 vm_map_lookup(vm_map_t *var_map,		/* IN/OUT */
3055 	      vm_offset_t vaddr,
3056 	      vm_prot_t fault_typea,
3057 	      vm_map_entry_t *out_entry,	/* OUT */
3058 	      vm_object_t *object,		/* OUT */
3059 	      vm_pindex_t *pindex,		/* OUT */
3060 	      vm_prot_t *out_prot,		/* OUT */
3061 	      boolean_t *wired)			/* OUT */
3062 {
3063 	vm_map_entry_t entry;
3064 	vm_map_t map = *var_map;
3065 	vm_prot_t prot;
3066 	vm_prot_t fault_type = fault_typea;
3067 
3068 RetryLookup:;
3069 	/*
3070 	 * Lookup the faulting address.
3071 	 */
3072 
3073 	vm_map_lock_read(map);
3074 #define	RETURN(why) \
3075 		{ \
3076 		vm_map_unlock_read(map); \
3077 		return (why); \
3078 		}
3079 
3080 	/*
3081 	 * If the map has an interesting hint, try it before calling full
3082 	 * blown lookup routine.
3083 	 */
3084 	entry = map->root;
3085 	*out_entry = entry;
3086 	if (entry == NULL ||
3087 	    (vaddr < entry->start) || (vaddr >= entry->end)) {
3088 		/*
3089 		 * Entry was either not a valid hint, or the vaddr was not
3090 		 * contained in the entry, so do a full lookup.
3091 		 */
3092 		if (!vm_map_lookup_entry(map, vaddr, out_entry))
3093 			RETURN(KERN_INVALID_ADDRESS);
3094 
3095 		entry = *out_entry;
3096 	}
3097 
3098 	/*
3099 	 * Handle submaps.
3100 	 */
3101 	if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
3102 		vm_map_t old_map = map;
3103 
3104 		*var_map = map = entry->object.sub_map;
3105 		vm_map_unlock_read(old_map);
3106 		goto RetryLookup;
3107 	}
3108 
3109 	/*
3110 	 * Check whether this task is allowed to have this page.
3111 	 * Note the special case for MAP_ENTRY_COW
3112 	 * pages with an override.  This is to implement a forced
3113 	 * COW for debuggers.
3114 	 */
3115 	if (fault_type & VM_PROT_OVERRIDE_WRITE)
3116 		prot = entry->max_protection;
3117 	else
3118 		prot = entry->protection;
3119 	fault_type &= (VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE);
3120 	if ((fault_type & prot) != fault_type) {
3121 			RETURN(KERN_PROTECTION_FAILURE);
3122 	}
3123 	if ((entry->eflags & MAP_ENTRY_USER_WIRED) &&
3124 	    (entry->eflags & MAP_ENTRY_COW) &&
3125 	    (fault_type & VM_PROT_WRITE) &&
3126 	    (fault_typea & VM_PROT_OVERRIDE_WRITE) == 0) {
3127 		RETURN(KERN_PROTECTION_FAILURE);
3128 	}
3129 
3130 	/*
3131 	 * If this page is not pageable, we have to get it for all possible
3132 	 * accesses.
3133 	 */
3134 	*wired = (entry->wired_count != 0);
3135 	if (*wired)
3136 		prot = fault_type = entry->protection;
3137 
3138 	/*
3139 	 * If the entry was copy-on-write, we either ...
3140 	 */
3141 	if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
3142 		/*
3143 		 * If we want to write the page, we may as well handle that
3144 		 * now since we've got the map locked.
3145 		 *
3146 		 * If we don't need to write the page, we just demote the
3147 		 * permissions allowed.
3148 		 */
3149 		if (fault_type & VM_PROT_WRITE) {
3150 			/*
3151 			 * Make a new object, and place it in the object
3152 			 * chain.  Note that no new references have appeared
3153 			 * -- one just moved from the map to the new
3154 			 * object.
3155 			 */
3156 			if (vm_map_lock_upgrade(map))
3157 				goto RetryLookup;
3158 
3159 			vm_object_shadow(
3160 			    &entry->object.vm_object,
3161 			    &entry->offset,
3162 			    atop(entry->end - entry->start));
3163 			entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
3164 
3165 			vm_map_lock_downgrade(map);
3166 		} else {
3167 			/*
3168 			 * We're attempting to read a copy-on-write page --
3169 			 * don't allow writes.
3170 			 */
3171 			prot &= ~VM_PROT_WRITE;
3172 		}
3173 	}
3174 
3175 	/*
3176 	 * Create an object if necessary.
3177 	 */
3178 	if (entry->object.vm_object == NULL &&
3179 	    !map->system_map) {
3180 		if (vm_map_lock_upgrade(map))
3181 			goto RetryLookup;
3182 		entry->object.vm_object = vm_object_allocate(OBJT_DEFAULT,
3183 		    atop(entry->end - entry->start));
3184 		entry->offset = 0;
3185 		vm_map_lock_downgrade(map);
3186 	}
3187 
3188 	/*
3189 	 * Return the object/offset from this entry.  If the entry was
3190 	 * copy-on-write or empty, it has been fixed up.
3191 	 */
3192 	*pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
3193 	*object = entry->object.vm_object;
3194 
3195 	*out_prot = prot;
3196 	return (KERN_SUCCESS);
3197 
3198 #undef	RETURN
3199 }
3200 
3201 /*
3202  *	vm_map_lookup_locked:
3203  *
3204  *	Lookup the faulting address.  A version of vm_map_lookup that returns
3205  *      KERN_FAILURE instead of blocking on map lock or memory allocation.
3206  */
3207 int
3208 vm_map_lookup_locked(vm_map_t *var_map,		/* IN/OUT */
3209 		     vm_offset_t vaddr,
3210 		     vm_prot_t fault_typea,
3211 		     vm_map_entry_t *out_entry,	/* OUT */
3212 		     vm_object_t *object,	/* OUT */
3213 		     vm_pindex_t *pindex,	/* OUT */
3214 		     vm_prot_t *out_prot,	/* OUT */
3215 		     boolean_t *wired)		/* OUT */
3216 {
3217 	vm_map_entry_t entry;
3218 	vm_map_t map = *var_map;
3219 	vm_prot_t prot;
3220 	vm_prot_t fault_type = fault_typea;
3221 
3222 	/*
3223 	 * If the map has an interesting hint, try it before calling full
3224 	 * blown lookup routine.
3225 	 */
3226 	entry = map->root;
3227 	*out_entry = entry;
3228 	if (entry == NULL ||
3229 	    (vaddr < entry->start) || (vaddr >= entry->end)) {
3230 		/*
3231 		 * Entry was either not a valid hint, or the vaddr was not
3232 		 * contained in the entry, so do a full lookup.
3233 		 */
3234 		if (!vm_map_lookup_entry(map, vaddr, out_entry))
3235 			return (KERN_INVALID_ADDRESS);
3236 
3237 		entry = *out_entry;
3238 	}
3239 
3240 	/*
3241 	 * Fail if the entry refers to a submap.
3242 	 */
3243 	if (entry->eflags & MAP_ENTRY_IS_SUB_MAP)
3244 		return (KERN_FAILURE);
3245 
3246 	/*
3247 	 * Check whether this task is allowed to have this page.
3248 	 * Note the special case for MAP_ENTRY_COW
3249 	 * pages with an override.  This is to implement a forced
3250 	 * COW for debuggers.
3251 	 */
3252 	if (fault_type & VM_PROT_OVERRIDE_WRITE)
3253 		prot = entry->max_protection;
3254 	else
3255 		prot = entry->protection;
3256 	fault_type &= VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE;
3257 	if ((fault_type & prot) != fault_type)
3258 		return (KERN_PROTECTION_FAILURE);
3259 	if ((entry->eflags & MAP_ENTRY_USER_WIRED) &&
3260 	    (entry->eflags & MAP_ENTRY_COW) &&
3261 	    (fault_type & VM_PROT_WRITE) &&
3262 	    (fault_typea & VM_PROT_OVERRIDE_WRITE) == 0)
3263 		return (KERN_PROTECTION_FAILURE);
3264 
3265 	/*
3266 	 * If this page is not pageable, we have to get it for all possible
3267 	 * accesses.
3268 	 */
3269 	*wired = (entry->wired_count != 0);
3270 	if (*wired)
3271 		prot = fault_type = entry->protection;
3272 
3273 	if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
3274 		/*
3275 		 * Fail if the entry was copy-on-write for a write fault.
3276 		 */
3277 		if (fault_type & VM_PROT_WRITE)
3278 			return (KERN_FAILURE);
3279 		/*
3280 		 * We're attempting to read a copy-on-write page --
3281 		 * don't allow writes.
3282 		 */
3283 		prot &= ~VM_PROT_WRITE;
3284 	}
3285 
3286 	/*
3287 	 * Fail if an object should be created.
3288 	 */
3289 	if (entry->object.vm_object == NULL && !map->system_map)
3290 		return (KERN_FAILURE);
3291 
3292 	/*
3293 	 * Return the object/offset from this entry.  If the entry was
3294 	 * copy-on-write or empty, it has been fixed up.
3295 	 */
3296 	*pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
3297 	*object = entry->object.vm_object;
3298 
3299 	*out_prot = prot;
3300 	return (KERN_SUCCESS);
3301 }
3302 
3303 /*
3304  *	vm_map_lookup_done:
3305  *
3306  *	Releases locks acquired by a vm_map_lookup
3307  *	(according to the handle returned by that lookup).
3308  */
3309 void
3310 vm_map_lookup_done(vm_map_t map, vm_map_entry_t entry)
3311 {
3312 	/*
3313 	 * Unlock the main-level map
3314 	 */
3315 	vm_map_unlock_read(map);
3316 }
3317 
3318 #include "opt_ddb.h"
3319 #ifdef DDB
3320 #include <sys/kernel.h>
3321 
3322 #include <ddb/ddb.h>
3323 
3324 /*
3325  *	vm_map_print:	[ debug ]
3326  */
3327 DB_SHOW_COMMAND(map, vm_map_print)
3328 {
3329 	static int nlines;
3330 	/* XXX convert args. */
3331 	vm_map_t map = (vm_map_t)addr;
3332 	boolean_t full = have_addr;
3333 
3334 	vm_map_entry_t entry;
3335 
3336 	db_iprintf("Task map %p: pmap=%p, nentries=%d, version=%u\n",
3337 	    (void *)map,
3338 	    (void *)map->pmap, map->nentries, map->timestamp);
3339 	nlines++;
3340 
3341 	if (!full && db_indent)
3342 		return;
3343 
3344 	db_indent += 2;
3345 	for (entry = map->header.next; entry != &map->header;
3346 	    entry = entry->next) {
3347 		db_iprintf("map entry %p: start=%p, end=%p\n",
3348 		    (void *)entry, (void *)entry->start, (void *)entry->end);
3349 		nlines++;
3350 		{
3351 			static char *inheritance_name[4] =
3352 			{"share", "copy", "none", "donate_copy"};
3353 
3354 			db_iprintf(" prot=%x/%x/%s",
3355 			    entry->protection,
3356 			    entry->max_protection,
3357 			    inheritance_name[(int)(unsigned char)entry->inheritance]);
3358 			if (entry->wired_count != 0)
3359 				db_printf(", wired");
3360 		}
3361 		if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
3362 			db_printf(", share=%p, offset=0x%jx\n",
3363 			    (void *)entry->object.sub_map,
3364 			    (uintmax_t)entry->offset);
3365 			nlines++;
3366 			if ((entry->prev == &map->header) ||
3367 			    (entry->prev->object.sub_map !=
3368 				entry->object.sub_map)) {
3369 				db_indent += 2;
3370 				vm_map_print((db_expr_t)(intptr_t)
3371 					     entry->object.sub_map,
3372 					     full, 0, (char *)0);
3373 				db_indent -= 2;
3374 			}
3375 		} else {
3376 			db_printf(", object=%p, offset=0x%jx",
3377 			    (void *)entry->object.vm_object,
3378 			    (uintmax_t)entry->offset);
3379 			if (entry->eflags & MAP_ENTRY_COW)
3380 				db_printf(", copy (%s)",
3381 				    (entry->eflags & MAP_ENTRY_NEEDS_COPY) ? "needed" : "done");
3382 			db_printf("\n");
3383 			nlines++;
3384 
3385 			if ((entry->prev == &map->header) ||
3386 			    (entry->prev->object.vm_object !=
3387 				entry->object.vm_object)) {
3388 				db_indent += 2;
3389 				vm_object_print((db_expr_t)(intptr_t)
3390 						entry->object.vm_object,
3391 						full, 0, (char *)0);
3392 				nlines += 4;
3393 				db_indent -= 2;
3394 			}
3395 		}
3396 	}
3397 	db_indent -= 2;
3398 	if (db_indent == 0)
3399 		nlines = 0;
3400 }
3401 
3402 
3403 DB_SHOW_COMMAND(procvm, procvm)
3404 {
3405 	struct proc *p;
3406 
3407 	if (have_addr) {
3408 		p = (struct proc *) addr;
3409 	} else {
3410 		p = curproc;
3411 	}
3412 
3413 	db_printf("p = %p, vmspace = %p, map = %p, pmap = %p\n",
3414 	    (void *)p, (void *)p->p_vmspace, (void *)&p->p_vmspace->vm_map,
3415 	    (void *)vmspace_pmap(p->p_vmspace));
3416 
3417 	vm_map_print((db_expr_t)(intptr_t)&p->p_vmspace->vm_map, 1, 0, NULL);
3418 }
3419 
3420 #endif /* DDB */
3421