xref: /netbsd-src/sys/uvm/uvm_map.c (revision f3cfa6f6ce31685c6c4a758bc430e69eb99f50a4)
1 /*	$NetBSD: uvm_map.c,v 1.359 2019/03/14 19:10:04 kre Exp $	*/
2 
3 /*
4  * Copyright (c) 1997 Charles D. Cranor and Washington University.
5  * Copyright (c) 1991, 1993, The Regents of the University of California.
6  *
7  * All rights reserved.
8  *
9  * This code is derived from software contributed to Berkeley by
10  * The Mach Operating System project at Carnegie-Mellon University.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	@(#)vm_map.c    8.3 (Berkeley) 1/12/94
37  * from: Id: uvm_map.c,v 1.1.2.27 1998/02/07 01:16:54 chs Exp
38  *
39  *
40  * Copyright (c) 1987, 1990 Carnegie-Mellon University.
41  * All rights reserved.
42  *
43  * Permission to use, copy, modify and distribute this software and
44  * its documentation is hereby granted, provided that both the copyright
45  * notice and this permission notice appear in all copies of the
46  * software, derivative works or modified versions, and any portions
47  * thereof, and that both notices appear in supporting documentation.
48  *
49  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
50  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
51  * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
52  *
53  * Carnegie Mellon requests users of this software to return to
54  *
55  *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
56  *  School of Computer Science
57  *  Carnegie Mellon University
58  *  Pittsburgh PA 15213-3890
59  *
60  * any improvements or extensions that they make and grant Carnegie the
61  * rights to redistribute these changes.
62  */
63 
64 /*
65  * uvm_map.c: uvm map operations
66  */
67 
68 #include <sys/cdefs.h>
69 __KERNEL_RCSID(0, "$NetBSD: uvm_map.c,v 1.359 2019/03/14 19:10:04 kre Exp $");
70 
71 #include "opt_ddb.h"
72 #include "opt_pax.h"
73 #include "opt_uvmhist.h"
74 #include "opt_uvm.h"
75 #include "opt_sysv.h"
76 
77 #include <sys/param.h>
78 #include <sys/systm.h>
79 #include <sys/mman.h>
80 #include <sys/proc.h>
81 #include <sys/pool.h>
82 #include <sys/kernel.h>
83 #include <sys/mount.h>
84 #include <sys/pax.h>
85 #include <sys/vnode.h>
86 #include <sys/filedesc.h>
87 #include <sys/lockdebug.h>
88 #include <sys/atomic.h>
89 #include <sys/sysctl.h>
90 #ifndef __USER_VA0_IS_SAFE
91 #include <sys/kauth.h>
92 #include "opt_user_va0_disable_default.h"
93 #endif
94 
95 #include <sys/shm.h>
96 
97 #include <uvm/uvm.h>
98 #include <uvm/uvm_readahead.h>
99 
100 #if defined(DDB) || defined(DEBUGPRINT)
101 #include <uvm/uvm_ddb.h>
102 #endif
103 
104 #ifdef UVMHIST
105 #ifndef UVMHIST_MAPHIST_SIZE
106 #define UVMHIST_MAPHIST_SIZE 100
107 #endif
108 #ifndef UVMHIST_PDHIST_SIZE
109 #define UVMHIST_PDHIST_SIZE 100
110 #endif
111 static struct kern_history_ent maphistbuf[UVMHIST_MAPHIST_SIZE];
112 UVMHIST_DEFINE(maphist) = UVMHIST_INITIALIZER(maphist, maphistbuf);
113 #endif
114 
115 #if !defined(UVMMAP_COUNTERS)
116 
117 #define	UVMMAP_EVCNT_DEFINE(name)	/* nothing */
118 #define UVMMAP_EVCNT_INCR(ev)		/* nothing */
119 #define UVMMAP_EVCNT_DECR(ev)		/* nothing */
120 
121 #else /* defined(UVMMAP_NOCOUNTERS) */
122 
123 #include <sys/evcnt.h>
124 #define	UVMMAP_EVCNT_DEFINE(name) \
125 struct evcnt uvmmap_evcnt_##name = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, \
126     "uvmmap", #name); \
127 EVCNT_ATTACH_STATIC(uvmmap_evcnt_##name);
128 #define	UVMMAP_EVCNT_INCR(ev)		uvmmap_evcnt_##ev.ev_count++
129 #define	UVMMAP_EVCNT_DECR(ev)		uvmmap_evcnt_##ev.ev_count--
130 
131 #endif /* defined(UVMMAP_NOCOUNTERS) */
132 
133 UVMMAP_EVCNT_DEFINE(ubackmerge)
134 UVMMAP_EVCNT_DEFINE(uforwmerge)
135 UVMMAP_EVCNT_DEFINE(ubimerge)
136 UVMMAP_EVCNT_DEFINE(unomerge)
137 UVMMAP_EVCNT_DEFINE(kbackmerge)
138 UVMMAP_EVCNT_DEFINE(kforwmerge)
139 UVMMAP_EVCNT_DEFINE(kbimerge)
140 UVMMAP_EVCNT_DEFINE(knomerge)
141 UVMMAP_EVCNT_DEFINE(map_call)
142 UVMMAP_EVCNT_DEFINE(mlk_call)
143 UVMMAP_EVCNT_DEFINE(mlk_hint)
144 UVMMAP_EVCNT_DEFINE(mlk_list)
145 UVMMAP_EVCNT_DEFINE(mlk_tree)
146 UVMMAP_EVCNT_DEFINE(mlk_treeloop)
147 UVMMAP_EVCNT_DEFINE(mlk_listloop)
148 
149 const char vmmapbsy[] = "vmmapbsy";
150 
151 /*
152  * cache for vmspace structures.
153  */
154 
155 static struct pool_cache uvm_vmspace_cache;
156 
157 /*
158  * cache for dynamically-allocated map entries.
159  */
160 
161 static struct pool_cache uvm_map_entry_cache;
162 
163 #ifdef PMAP_GROWKERNEL
164 /*
165  * This global represents the end of the kernel virtual address
166  * space.  If we want to exceed this, we must grow the kernel
167  * virtual address space dynamically.
168  *
169  * Note, this variable is locked by kernel_map's lock.
170  */
171 vaddr_t uvm_maxkaddr;
172 #endif
173 
174 #ifndef __USER_VA0_IS_SAFE
175 #ifndef __USER_VA0_DISABLE_DEFAULT
176 #define __USER_VA0_DISABLE_DEFAULT 1
177 #endif
178 #ifdef USER_VA0_DISABLE_DEFAULT /* kernel config option overrides */
179 #undef __USER_VA0_DISABLE_DEFAULT
180 #define __USER_VA0_DISABLE_DEFAULT USER_VA0_DISABLE_DEFAULT
181 #endif
182 int user_va0_disable = __USER_VA0_DISABLE_DEFAULT;
183 #endif
184 
185 /*
186  * macros
187  */
188 
189 /*
190  * UVM_ET_ISCOMPATIBLE: check some requirements for map entry merging
191  */
192 extern struct vm_map *pager_map;
193 
194 #define	UVM_ET_ISCOMPATIBLE(ent, type, uobj, meflags, \
195     prot, maxprot, inh, adv, wire) \
196 	((ent)->etype == (type) && \
197 	(((ent)->flags ^ (meflags)) & (UVM_MAP_NOMERGE)) == 0 && \
198 	(ent)->object.uvm_obj == (uobj) && \
199 	(ent)->protection == (prot) && \
200 	(ent)->max_protection == (maxprot) && \
201 	(ent)->inheritance == (inh) && \
202 	(ent)->advice == (adv) && \
203 	(ent)->wired_count == (wire))
204 
205 /*
206  * uvm_map_entry_link: insert entry into a map
207  *
208  * => map must be locked
209  */
210 #define uvm_map_entry_link(map, after_where, entry) do { \
211 	uvm_mapent_check(entry); \
212 	(map)->nentries++; \
213 	(entry)->prev = (after_where); \
214 	(entry)->next = (after_where)->next; \
215 	(entry)->prev->next = (entry); \
216 	(entry)->next->prev = (entry); \
217 	uvm_rb_insert((map), (entry)); \
218 } while (/*CONSTCOND*/ 0)
219 
220 /*
221  * uvm_map_entry_unlink: remove entry from a map
222  *
223  * => map must be locked
224  */
225 #define uvm_map_entry_unlink(map, entry) do { \
226 	KASSERT((entry) != (map)->first_free); \
227 	KASSERT((entry) != (map)->hint); \
228 	uvm_mapent_check(entry); \
229 	(map)->nentries--; \
230 	(entry)->next->prev = (entry)->prev; \
231 	(entry)->prev->next = (entry)->next; \
232 	uvm_rb_remove((map), (entry)); \
233 } while (/*CONSTCOND*/ 0)
234 
235 /*
236  * SAVE_HINT: saves the specified entry as the hint for future lookups.
237  *
238  * => map need not be locked.
239  */
240 #define SAVE_HINT(map, check, value) do { \
241 	if ((map)->hint == (check)) \
242 		(map)->hint = (value); \
243 } while (/*CONSTCOND*/ 0)
244 
245 /*
246  * clear_hints: ensure that hints don't point to the entry.
247  *
248  * => map must be write-locked.
249  */
250 static void
251 clear_hints(struct vm_map *map, struct vm_map_entry *ent)
252 {
253 
254 	SAVE_HINT(map, ent, ent->prev);
255 	if (map->first_free == ent) {
256 		map->first_free = ent->prev;
257 	}
258 }
259 
260 /*
261  * VM_MAP_RANGE_CHECK: check and correct range
262  *
263  * => map must at least be read locked
264  */
265 
266 #define VM_MAP_RANGE_CHECK(map, start, end) do { \
267 	if (start < vm_map_min(map))		\
268 		start = vm_map_min(map);	\
269 	if (end > vm_map_max(map))		\
270 		end = vm_map_max(map);		\
271 	if (start > end)			\
272 		start = end;			\
273 } while (/*CONSTCOND*/ 0)
274 
275 /*
276  * local prototypes
277  */
278 
279 static struct vm_map_entry *
280 		uvm_mapent_alloc(struct vm_map *, int);
281 static void	uvm_mapent_copy(struct vm_map_entry *, struct vm_map_entry *);
282 static void	uvm_mapent_free(struct vm_map_entry *);
283 #if defined(DEBUG)
284 static void	_uvm_mapent_check(const struct vm_map_entry *, const char *,
285 		    int);
286 #define	uvm_mapent_check(map)	_uvm_mapent_check(map, __FILE__, __LINE__)
287 #else /* defined(DEBUG) */
288 #define	uvm_mapent_check(e)	/* nothing */
289 #endif /* defined(DEBUG) */
290 
291 static void	uvm_map_entry_unwire(struct vm_map *, struct vm_map_entry *);
292 static void	uvm_map_reference_amap(struct vm_map_entry *, int);
293 static int	uvm_map_space_avail(vaddr_t *, vsize_t, voff_t, vsize_t, int,
294 		    int, struct vm_map_entry *);
295 static void	uvm_map_unreference_amap(struct vm_map_entry *, int);
296 
297 int _uvm_map_sanity(struct vm_map *);
298 int _uvm_tree_sanity(struct vm_map *);
299 static vsize_t uvm_rb_maxgap(const struct vm_map_entry *);
300 
301 #define	ROOT_ENTRY(map)		((struct vm_map_entry *)(map)->rb_tree.rbt_root)
302 #define	LEFT_ENTRY(entry)	((struct vm_map_entry *)(entry)->rb_node.rb_left)
303 #define	RIGHT_ENTRY(entry)	((struct vm_map_entry *)(entry)->rb_node.rb_right)
304 #define	PARENT_ENTRY(map, entry) \
305 	(ROOT_ENTRY(map) == (entry) \
306 	    ? NULL : (struct vm_map_entry *)RB_FATHER(&(entry)->rb_node))
307 
308 /*
309  * These get filled in if/when SYSVSHM shared memory code is loaded
310  *
311  * We do this with function pointers rather the #ifdef SYSVSHM so the
312  * SYSVSHM code can be loaded and unloaded
313  */
314 void (*uvm_shmexit)(struct vmspace *) = NULL;
315 void (*uvm_shmfork)(struct vmspace *, struct vmspace *) = NULL;
316 
317 static int
318 uvm_map_compare_nodes(void *ctx, const void *nparent, const void *nkey)
319 {
320 	const struct vm_map_entry *eparent = nparent;
321 	const struct vm_map_entry *ekey = nkey;
322 
323 	KASSERT(eparent->start < ekey->start || eparent->start >= ekey->end);
324 	KASSERT(ekey->start < eparent->start || ekey->start >= eparent->end);
325 
326 	if (eparent->start < ekey->start)
327 		return -1;
328 	if (eparent->end >= ekey->start)
329 		return 1;
330 	return 0;
331 }
332 
333 static int
334 uvm_map_compare_key(void *ctx, const void *nparent, const void *vkey)
335 {
336 	const struct vm_map_entry *eparent = nparent;
337 	const vaddr_t va = *(const vaddr_t *) vkey;
338 
339 	if (eparent->start < va)
340 		return -1;
341 	if (eparent->end >= va)
342 		return 1;
343 	return 0;
344 }
345 
346 static const rb_tree_ops_t uvm_map_tree_ops = {
347 	.rbto_compare_nodes = uvm_map_compare_nodes,
348 	.rbto_compare_key = uvm_map_compare_key,
349 	.rbto_node_offset = offsetof(struct vm_map_entry, rb_node),
350 	.rbto_context = NULL
351 };
352 
353 /*
354  * uvm_rb_gap: return the gap size between our entry and next entry.
355  */
356 static inline vsize_t
357 uvm_rb_gap(const struct vm_map_entry *entry)
358 {
359 
360 	KASSERT(entry->next != NULL);
361 	return entry->next->start - entry->end;
362 }
363 
364 static vsize_t
365 uvm_rb_maxgap(const struct vm_map_entry *entry)
366 {
367 	struct vm_map_entry *child;
368 	vsize_t maxgap = entry->gap;
369 
370 	/*
371 	 * We need maxgap to be the largest gap of us or any of our
372 	 * descendents.  Since each of our children's maxgap is the
373 	 * cached value of their largest gap of themselves or their
374 	 * descendents, we can just use that value and avoid recursing
375 	 * down the tree to calculate it.
376 	 */
377 	if ((child = LEFT_ENTRY(entry)) != NULL && maxgap < child->maxgap)
378 		maxgap = child->maxgap;
379 
380 	if ((child = RIGHT_ENTRY(entry)) != NULL && maxgap < child->maxgap)
381 		maxgap = child->maxgap;
382 
383 	return maxgap;
384 }
385 
386 static void
387 uvm_rb_fixup(struct vm_map *map, struct vm_map_entry *entry)
388 {
389 	struct vm_map_entry *parent;
390 
391 	KASSERT(entry->gap == uvm_rb_gap(entry));
392 	entry->maxgap = uvm_rb_maxgap(entry);
393 
394 	while ((parent = PARENT_ENTRY(map, entry)) != NULL) {
395 		struct vm_map_entry *brother;
396 		vsize_t maxgap = parent->gap;
397 		unsigned int which;
398 
399 		KDASSERT(parent->gap == uvm_rb_gap(parent));
400 		if (maxgap < entry->maxgap)
401 			maxgap = entry->maxgap;
402 		/*
403 		 * Since we work towards the root, we know entry's maxgap
404 		 * value is OK, but its brothers may now be out-of-date due
405 		 * to rebalancing.  So refresh it.
406 		 */
407 		which = RB_POSITION(&entry->rb_node) ^ RB_DIR_OTHER;
408 		brother = (struct vm_map_entry *)parent->rb_node.rb_nodes[which];
409 		if (brother != NULL) {
410 			KDASSERT(brother->gap == uvm_rb_gap(brother));
411 			brother->maxgap = uvm_rb_maxgap(brother);
412 			if (maxgap < brother->maxgap)
413 				maxgap = brother->maxgap;
414 		}
415 
416 		parent->maxgap = maxgap;
417 		entry = parent;
418 	}
419 }
420 
421 static void
422 uvm_rb_insert(struct vm_map *map, struct vm_map_entry *entry)
423 {
424 	struct vm_map_entry *ret __diagused;
425 
426 	entry->gap = entry->maxgap = uvm_rb_gap(entry);
427 	if (entry->prev != &map->header)
428 		entry->prev->gap = uvm_rb_gap(entry->prev);
429 
430 	ret = rb_tree_insert_node(&map->rb_tree, entry);
431 	KASSERTMSG(ret == entry,
432 	    "uvm_rb_insert: map %p: duplicate entry %p", map, ret);
433 
434 	/*
435 	 * If the previous entry is not our immediate left child, then it's an
436 	 * ancestor and will be fixed up on the way to the root.  We don't
437 	 * have to check entry->prev against &map->header since &map->header
438 	 * will never be in the tree.
439 	 */
440 	uvm_rb_fixup(map,
441 	    LEFT_ENTRY(entry) == entry->prev ? entry->prev : entry);
442 }
443 
444 static void
445 uvm_rb_remove(struct vm_map *map, struct vm_map_entry *entry)
446 {
447 	struct vm_map_entry *prev_parent = NULL, *next_parent = NULL;
448 
449 	/*
450 	 * If we are removing an interior node, then an adjacent node will
451 	 * be used to replace its position in the tree.  Therefore we will
452 	 * need to fixup the tree starting at the parent of the replacement
453 	 * node.  So record their parents for later use.
454 	 */
455 	if (entry->prev != &map->header)
456 		prev_parent = PARENT_ENTRY(map, entry->prev);
457 	if (entry->next != &map->header)
458 		next_parent = PARENT_ENTRY(map, entry->next);
459 
460 	rb_tree_remove_node(&map->rb_tree, entry);
461 
462 	/*
463 	 * If the previous node has a new parent, fixup the tree starting
464 	 * at the previous node's old parent.
465 	 */
466 	if (entry->prev != &map->header) {
467 		/*
468 		 * Update the previous entry's gap due to our absence.
469 		 */
470 		entry->prev->gap = uvm_rb_gap(entry->prev);
471 		uvm_rb_fixup(map, entry->prev);
472 		if (prev_parent != NULL
473 		    && prev_parent != entry
474 		    && prev_parent != PARENT_ENTRY(map, entry->prev))
475 			uvm_rb_fixup(map, prev_parent);
476 	}
477 
478 	/*
479 	 * If the next node has a new parent, fixup the tree starting
480 	 * at the next node's old parent.
481 	 */
482 	if (entry->next != &map->header) {
483 		uvm_rb_fixup(map, entry->next);
484 		if (next_parent != NULL
485 		    && next_parent != entry
486 		    && next_parent != PARENT_ENTRY(map, entry->next))
487 			uvm_rb_fixup(map, next_parent);
488 	}
489 }
490 
491 #if defined(DEBUG)
492 int uvm_debug_check_map = 0;
493 int uvm_debug_check_rbtree = 0;
494 #define uvm_map_check(map, name) \
495 	_uvm_map_check((map), (name), __FILE__, __LINE__)
496 static void
497 _uvm_map_check(struct vm_map *map, const char *name,
498     const char *file, int line)
499 {
500 
501 	if ((uvm_debug_check_map && _uvm_map_sanity(map)) ||
502 	    (uvm_debug_check_rbtree && _uvm_tree_sanity(map))) {
503 		panic("uvm_map_check failed: \"%s\" map=%p (%s:%d)",
504 		    name, map, file, line);
505 	}
506 }
507 #else /* defined(DEBUG) */
508 #define uvm_map_check(map, name)	/* nothing */
509 #endif /* defined(DEBUG) */
510 
511 #if defined(DEBUG) || defined(DDB)
512 int
513 _uvm_map_sanity(struct vm_map *map)
514 {
515 	bool first_free_found = false;
516 	bool hint_found = false;
517 	const struct vm_map_entry *e;
518 	struct vm_map_entry *hint = map->hint;
519 
520 	e = &map->header;
521 	for (;;) {
522 		if (map->first_free == e) {
523 			first_free_found = true;
524 		} else if (!first_free_found && e->next->start > e->end) {
525 			printf("first_free %p should be %p\n",
526 			    map->first_free, e);
527 			return -1;
528 		}
529 		if (hint == e) {
530 			hint_found = true;
531 		}
532 
533 		e = e->next;
534 		if (e == &map->header) {
535 			break;
536 		}
537 	}
538 	if (!first_free_found) {
539 		printf("stale first_free\n");
540 		return -1;
541 	}
542 	if (!hint_found) {
543 		printf("stale hint\n");
544 		return -1;
545 	}
546 	return 0;
547 }
548 
549 int
550 _uvm_tree_sanity(struct vm_map *map)
551 {
552 	struct vm_map_entry *tmp, *trtmp;
553 	int n = 0, i = 1;
554 
555 	for (tmp = map->header.next; tmp != &map->header; tmp = tmp->next) {
556 		if (tmp->gap != uvm_rb_gap(tmp)) {
557 			printf("%d/%d gap %#lx != %#lx %s\n",
558 			    n + 1, map->nentries,
559 			    (ulong)tmp->gap, (ulong)uvm_rb_gap(tmp),
560 			    tmp->next == &map->header ? "(last)" : "");
561 			goto error;
562 		}
563 		/*
564 		 * If any entries are out of order, tmp->gap will be unsigned
565 		 * and will likely exceed the size of the map.
566 		 */
567 		if (tmp->gap >= vm_map_max(map) - vm_map_min(map)) {
568 			printf("too large gap %zu\n", (size_t)tmp->gap);
569 			goto error;
570 		}
571 		n++;
572 	}
573 
574 	if (n != map->nentries) {
575 		printf("nentries: %d vs %d\n", n, map->nentries);
576 		goto error;
577 	}
578 
579 	trtmp = NULL;
580 	for (tmp = map->header.next; tmp != &map->header; tmp = tmp->next) {
581 		if (tmp->maxgap != uvm_rb_maxgap(tmp)) {
582 			printf("maxgap %#lx != %#lx\n",
583 			    (ulong)tmp->maxgap,
584 			    (ulong)uvm_rb_maxgap(tmp));
585 			goto error;
586 		}
587 		if (trtmp != NULL && trtmp->start >= tmp->start) {
588 			printf("corrupt: 0x%"PRIxVADDR"x >= 0x%"PRIxVADDR"x\n",
589 			    trtmp->start, tmp->start);
590 			goto error;
591 		}
592 
593 		trtmp = tmp;
594 	}
595 
596 	for (tmp = map->header.next; tmp != &map->header;
597 	    tmp = tmp->next, i++) {
598 		trtmp = rb_tree_iterate(&map->rb_tree, tmp, RB_DIR_LEFT);
599 		if (trtmp == NULL)
600 			trtmp = &map->header;
601 		if (tmp->prev != trtmp) {
602 			printf("lookup: %d: %p->prev=%p: %p\n",
603 			    i, tmp, tmp->prev, trtmp);
604 			goto error;
605 		}
606 		trtmp = rb_tree_iterate(&map->rb_tree, tmp, RB_DIR_RIGHT);
607 		if (trtmp == NULL)
608 			trtmp = &map->header;
609 		if (tmp->next != trtmp) {
610 			printf("lookup: %d: %p->next=%p: %p\n",
611 			    i, tmp, tmp->next, trtmp);
612 			goto error;
613 		}
614 		trtmp = rb_tree_find_node(&map->rb_tree, &tmp->start);
615 		if (trtmp != tmp) {
616 			printf("lookup: %d: %p - %p: %p\n", i, tmp, trtmp,
617 			    PARENT_ENTRY(map, tmp));
618 			goto error;
619 		}
620 	}
621 
622 	return (0);
623  error:
624 	return (-1);
625 }
626 #endif /* defined(DEBUG) || defined(DDB) */
627 
628 /*
629  * vm_map_lock: acquire an exclusive (write) lock on a map.
630  *
631  * => The locking protocol provides for guaranteed upgrade from shared ->
632  *    exclusive by whichever thread currently has the map marked busy.
633  *    See "LOCKING PROTOCOL NOTES" in uvm_map.h.  This is horrible; among
634  *    other problems, it defeats any fairness guarantees provided by RW
635  *    locks.
636  */
637 
638 void
639 vm_map_lock(struct vm_map *map)
640 {
641 
642 	for (;;) {
643 		rw_enter(&map->lock, RW_WRITER);
644 		if (map->busy == NULL || map->busy == curlwp) {
645 			break;
646 		}
647 		mutex_enter(&map->misc_lock);
648 		rw_exit(&map->lock);
649 		if (map->busy != NULL) {
650 			cv_wait(&map->cv, &map->misc_lock);
651 		}
652 		mutex_exit(&map->misc_lock);
653 	}
654 	map->timestamp++;
655 }
656 
657 /*
658  * vm_map_lock_try: try to lock a map, failing if it is already locked.
659  */
660 
661 bool
662 vm_map_lock_try(struct vm_map *map)
663 {
664 
665 	if (!rw_tryenter(&map->lock, RW_WRITER)) {
666 		return false;
667 	}
668 	if (map->busy != NULL) {
669 		rw_exit(&map->lock);
670 		return false;
671 	}
672 	map->timestamp++;
673 	return true;
674 }
675 
676 /*
677  * vm_map_unlock: release an exclusive lock on a map.
678  */
679 
680 void
681 vm_map_unlock(struct vm_map *map)
682 {
683 
684 	KASSERT(rw_write_held(&map->lock));
685 	KASSERT(map->busy == NULL || map->busy == curlwp);
686 	rw_exit(&map->lock);
687 }
688 
689 /*
690  * vm_map_unbusy: mark the map as unbusy, and wake any waiters that
691  *     want an exclusive lock.
692  */
693 
694 void
695 vm_map_unbusy(struct vm_map *map)
696 {
697 
698 	KASSERT(map->busy == curlwp);
699 
700 	/*
701 	 * Safe to clear 'busy' and 'waiters' with only a read lock held:
702 	 *
703 	 * o they can only be set with a write lock held
704 	 * o writers are blocked out with a read or write hold
705 	 * o at any time, only one thread owns the set of values
706 	 */
707 	mutex_enter(&map->misc_lock);
708 	map->busy = NULL;
709 	cv_broadcast(&map->cv);
710 	mutex_exit(&map->misc_lock);
711 }
712 
713 /*
714  * vm_map_lock_read: acquire a shared (read) lock on a map.
715  */
716 
717 void
718 vm_map_lock_read(struct vm_map *map)
719 {
720 
721 	rw_enter(&map->lock, RW_READER);
722 }
723 
724 /*
725  * vm_map_unlock_read: release a shared lock on a map.
726  */
727 
728 void
729 vm_map_unlock_read(struct vm_map *map)
730 {
731 
732 	rw_exit(&map->lock);
733 }
734 
735 /*
736  * vm_map_busy: mark a map as busy.
737  *
738  * => the caller must hold the map write locked
739  */
740 
741 void
742 vm_map_busy(struct vm_map *map)
743 {
744 
745 	KASSERT(rw_write_held(&map->lock));
746 	KASSERT(map->busy == NULL);
747 
748 	map->busy = curlwp;
749 }
750 
751 /*
752  * vm_map_locked_p: return true if the map is write locked.
753  *
754  * => only for debug purposes like KASSERTs.
755  * => should not be used to verify that a map is not locked.
756  */
757 
758 bool
759 vm_map_locked_p(struct vm_map *map)
760 {
761 
762 	return rw_write_held(&map->lock);
763 }
764 
765 /*
766  * uvm_mapent_alloc: allocate a map entry
767  */
768 
769 static struct vm_map_entry *
770 uvm_mapent_alloc(struct vm_map *map, int flags)
771 {
772 	struct vm_map_entry *me;
773 	int pflags = (flags & UVM_FLAG_NOWAIT) ? PR_NOWAIT : PR_WAITOK;
774 	UVMHIST_FUNC("uvm_mapent_alloc"); UVMHIST_CALLED(maphist);
775 
776 	me = pool_cache_get(&uvm_map_entry_cache, pflags);
777 	if (__predict_false(me == NULL)) {
778 		return NULL;
779 	}
780 	me->flags = 0;
781 
782 	UVMHIST_LOG(maphist, "<- new entry=%#jx [kentry=%jd]", (uintptr_t)me,
783 	    (map == kernel_map), 0, 0);
784 	return me;
785 }
786 
787 /*
788  * uvm_mapent_free: free map entry
789  */
790 
791 static void
792 uvm_mapent_free(struct vm_map_entry *me)
793 {
794 	UVMHIST_FUNC("uvm_mapent_free"); UVMHIST_CALLED(maphist);
795 
796 	UVMHIST_LOG(maphist,"<- freeing map entry=%#jx [flags=%jd]",
797 		(uintptr_t)me, me->flags, 0, 0);
798 	pool_cache_put(&uvm_map_entry_cache, me);
799 }
800 
801 /*
802  * uvm_mapent_copy: copy a map entry, preserving flags
803  */
804 
805 static inline void
806 uvm_mapent_copy(struct vm_map_entry *src, struct vm_map_entry *dst)
807 {
808 
809 	memcpy(dst, src, ((char *)&src->uvm_map_entry_stop_copy) -
810 	    ((char *)src));
811 }
812 
813 #if defined(DEBUG)
814 static void
815 _uvm_mapent_check(const struct vm_map_entry *entry, const char *file, int line)
816 {
817 
818 	if (entry->start >= entry->end) {
819 		goto bad;
820 	}
821 	if (UVM_ET_ISOBJ(entry)) {
822 		if (entry->object.uvm_obj == NULL) {
823 			goto bad;
824 		}
825 	} else if (UVM_ET_ISSUBMAP(entry)) {
826 		if (entry->object.sub_map == NULL) {
827 			goto bad;
828 		}
829 	} else {
830 		if (entry->object.uvm_obj != NULL ||
831 		    entry->object.sub_map != NULL) {
832 			goto bad;
833 		}
834 	}
835 	if (!UVM_ET_ISOBJ(entry)) {
836 		if (entry->offset != 0) {
837 			goto bad;
838 		}
839 	}
840 
841 	return;
842 
843 bad:
844 	panic("%s: bad entry %p (%s:%d)", __func__, entry, file, line);
845 }
846 #endif /* defined(DEBUG) */
847 
848 /*
849  * uvm_map_entry_unwire: unwire a map entry
850  *
851  * => map should be locked by caller
852  */
853 
854 static inline void
855 uvm_map_entry_unwire(struct vm_map *map, struct vm_map_entry *entry)
856 {
857 
858 	entry->wired_count = 0;
859 	uvm_fault_unwire_locked(map, entry->start, entry->end);
860 }
861 
862 
863 /*
864  * wrapper for calling amap_ref()
865  */
866 static inline void
867 uvm_map_reference_amap(struct vm_map_entry *entry, int flags)
868 {
869 
870 	amap_ref(entry->aref.ar_amap, entry->aref.ar_pageoff,
871 	    (entry->end - entry->start) >> PAGE_SHIFT, flags);
872 }
873 
874 
875 /*
876  * wrapper for calling amap_unref()
877  */
878 static inline void
879 uvm_map_unreference_amap(struct vm_map_entry *entry, int flags)
880 {
881 
882 	amap_unref(entry->aref.ar_amap, entry->aref.ar_pageoff,
883 	    (entry->end - entry->start) >> PAGE_SHIFT, flags);
884 }
885 
886 
887 /*
888  * uvm_map_init: init mapping system at boot time.
889  */
890 
891 void
892 uvm_map_init(void)
893 {
894 #if defined(UVMHIST)
895 	static struct kern_history_ent pdhistbuf[UVMHIST_PDHIST_SIZE];
896 #endif
897 
898 	/*
899 	 * first, init logging system.
900 	 */
901 
902 	UVMHIST_FUNC("uvm_map_init");
903 	UVMHIST_LINK_STATIC(maphist);
904 	UVMHIST_INIT_STATIC(pdhist, pdhistbuf);
905 	UVMHIST_CALLED(maphist);
906 	UVMHIST_LOG(maphist,"<starting uvm map system>", 0, 0, 0, 0);
907 
908 	/*
909 	 * initialize the global lock for kernel map entry.
910 	 */
911 
912 	mutex_init(&uvm_kentry_lock, MUTEX_DRIVER, IPL_VM);
913 }
914 
915 /*
916  * uvm_map_init_caches: init mapping system caches.
917  */
918 void
919 uvm_map_init_caches(void)
920 {
921 	/*
922 	 * initialize caches.
923 	 */
924 
925 	pool_cache_bootstrap(&uvm_map_entry_cache, sizeof(struct vm_map_entry),
926 	    0, 0, 0, "vmmpepl", NULL, IPL_NONE, NULL, NULL, NULL);
927 	pool_cache_bootstrap(&uvm_vmspace_cache, sizeof(struct vmspace),
928 	    0, 0, 0, "vmsppl", NULL, IPL_NONE, NULL, NULL, NULL);
929 }
930 
931 /*
932  * clippers
933  */
934 
935 /*
936  * uvm_mapent_splitadj: adjust map entries for splitting, after uvm_mapent_copy.
937  */
938 
939 static void
940 uvm_mapent_splitadj(struct vm_map_entry *entry1, struct vm_map_entry *entry2,
941     vaddr_t splitat)
942 {
943 	vaddr_t adj;
944 
945 	KASSERT(entry1->start < splitat);
946 	KASSERT(splitat < entry1->end);
947 
948 	adj = splitat - entry1->start;
949 	entry1->end = entry2->start = splitat;
950 
951 	if (entry1->aref.ar_amap) {
952 		amap_splitref(&entry1->aref, &entry2->aref, adj);
953 	}
954 	if (UVM_ET_ISSUBMAP(entry1)) {
955 		/* ... unlikely to happen, but play it safe */
956 		 uvm_map_reference(entry1->object.sub_map);
957 	} else if (UVM_ET_ISOBJ(entry1)) {
958 		KASSERT(entry1->object.uvm_obj != NULL); /* suppress coverity */
959 		entry2->offset += adj;
960 		if (entry1->object.uvm_obj->pgops &&
961 		    entry1->object.uvm_obj->pgops->pgo_reference)
962 			entry1->object.uvm_obj->pgops->pgo_reference(
963 			    entry1->object.uvm_obj);
964 	}
965 }
966 
967 /*
968  * uvm_map_clip_start: ensure that the entry begins at or after
969  *	the starting address, if it doesn't we split the entry.
970  *
971  * => caller should use UVM_MAP_CLIP_START macro rather than calling
972  *    this directly
973  * => map must be locked by caller
974  */
975 
976 void
977 uvm_map_clip_start(struct vm_map *map, struct vm_map_entry *entry,
978     vaddr_t start)
979 {
980 	struct vm_map_entry *new_entry;
981 
982 	/* uvm_map_simplify_entry(map, entry); */ /* XXX */
983 
984 	uvm_map_check(map, "clip_start entry");
985 	uvm_mapent_check(entry);
986 
987 	/*
988 	 * Split off the front portion.  note that we must insert the new
989 	 * entry BEFORE this one, so that this entry has the specified
990 	 * starting address.
991 	 */
992 	new_entry = uvm_mapent_alloc(map, 0);
993 	uvm_mapent_copy(entry, new_entry); /* entry -> new_entry */
994 	uvm_mapent_splitadj(new_entry, entry, start);
995 	uvm_map_entry_link(map, entry->prev, new_entry);
996 
997 	uvm_map_check(map, "clip_start leave");
998 }
999 
1000 /*
1001  * uvm_map_clip_end: ensure that the entry ends at or before
1002  *	the ending address, if it does't we split the reference
1003  *
1004  * => caller should use UVM_MAP_CLIP_END macro rather than calling
1005  *    this directly
1006  * => map must be locked by caller
1007  */
1008 
1009 void
1010 uvm_map_clip_end(struct vm_map *map, struct vm_map_entry *entry, vaddr_t end)
1011 {
1012 	struct vm_map_entry *new_entry;
1013 
1014 	uvm_map_check(map, "clip_end entry");
1015 	uvm_mapent_check(entry);
1016 
1017 	/*
1018 	 *	Create a new entry and insert it
1019 	 *	AFTER the specified entry
1020 	 */
1021 	new_entry = uvm_mapent_alloc(map, 0);
1022 	uvm_mapent_copy(entry, new_entry); /* entry -> new_entry */
1023 	uvm_mapent_splitadj(entry, new_entry, end);
1024 	uvm_map_entry_link(map, entry, new_entry);
1025 
1026 	uvm_map_check(map, "clip_end leave");
1027 }
1028 
1029 /*
1030  *   M A P   -   m a i n   e n t r y   p o i n t
1031  */
1032 /*
1033  * uvm_map: establish a valid mapping in a map
1034  *
1035  * => assume startp is page aligned.
1036  * => assume size is a multiple of PAGE_SIZE.
1037  * => assume sys_mmap provides enough of a "hint" to have us skip
1038  *	over text/data/bss area.
1039  * => map must be unlocked (we will lock it)
1040  * => <uobj,uoffset> value meanings (4 cases):
1041  *	 [1] <NULL,uoffset>		== uoffset is a hint for PMAP_PREFER
1042  *	 [2] <NULL,UVM_UNKNOWN_OFFSET>	== don't PMAP_PREFER
1043  *	 [3] <uobj,uoffset>		== normal mapping
1044  *	 [4] <uobj,UVM_UNKNOWN_OFFSET>	== uvm_map finds offset based on VA
1045  *
1046  *    case [4] is for kernel mappings where we don't know the offset until
1047  *    we've found a virtual address.   note that kernel object offsets are
1048  *    always relative to vm_map_min(kernel_map).
1049  *
1050  * => if `align' is non-zero, we align the virtual address to the specified
1051  *	alignment.
1052  *	this is provided as a mechanism for large pages.
1053  *
1054  * => XXXCDC: need way to map in external amap?
1055  */
1056 
1057 int
1058 uvm_map(struct vm_map *map, vaddr_t *startp /* IN/OUT */, vsize_t size,
1059     struct uvm_object *uobj, voff_t uoffset, vsize_t align, uvm_flag_t flags)
1060 {
1061 	struct uvm_map_args args;
1062 	struct vm_map_entry *new_entry;
1063 	int error;
1064 
1065 	KASSERT((size & PAGE_MASK) == 0);
1066 
1067 	/*
1068 	 * for pager_map, allocate the new entry first to avoid sleeping
1069 	 * for memory while we have the map locked.
1070 	 */
1071 
1072 	new_entry = NULL;
1073 	if (map == pager_map) {
1074 		new_entry = uvm_mapent_alloc(map, (flags & UVM_FLAG_NOWAIT));
1075 		if (__predict_false(new_entry == NULL))
1076 			return ENOMEM;
1077 	}
1078 	if (map == pager_map)
1079 		flags |= UVM_FLAG_NOMERGE;
1080 
1081 	error = uvm_map_prepare(map, *startp, size, uobj, uoffset, align,
1082 	    flags, &args);
1083 	if (!error) {
1084 		error = uvm_map_enter(map, &args, new_entry);
1085 		*startp = args.uma_start;
1086 	} else if (new_entry) {
1087 		uvm_mapent_free(new_entry);
1088 	}
1089 
1090 #if defined(DEBUG)
1091 	if (!error && VM_MAP_IS_KERNEL(map) && (flags & UVM_FLAG_NOWAIT) == 0) {
1092 		uvm_km_check_empty(map, *startp, *startp + size);
1093 	}
1094 #endif /* defined(DEBUG) */
1095 
1096 	return error;
1097 }
1098 
1099 /*
1100  * uvm_map_prepare:
1101  *
1102  * called with map unlocked.
1103  * on success, returns the map locked.
1104  */
1105 
1106 int
1107 uvm_map_prepare(struct vm_map *map, vaddr_t start, vsize_t size,
1108     struct uvm_object *uobj, voff_t uoffset, vsize_t align, uvm_flag_t flags,
1109     struct uvm_map_args *args)
1110 {
1111 	struct vm_map_entry *prev_entry;
1112 	vm_prot_t prot = UVM_PROTECTION(flags);
1113 	vm_prot_t maxprot = UVM_MAXPROTECTION(flags);
1114 
1115 	UVMHIST_FUNC("uvm_map_prepare");
1116 	UVMHIST_CALLED(maphist);
1117 
1118 	UVMHIST_LOG(maphist, "(map=%#jx, start=%#jx, size=%ju, flags=%#jx)",
1119 	    (uintptr_t)map, start, size, flags);
1120 	UVMHIST_LOG(maphist, "  uobj/offset %#jx/%jd", (uintptr_t)uobj,
1121 	    uoffset,0,0);
1122 
1123 	/*
1124 	 * detect a popular device driver bug.
1125 	 */
1126 
1127 	KASSERT(doing_shutdown || curlwp != NULL);
1128 
1129 	/*
1130 	 * zero-sized mapping doesn't make any sense.
1131 	 */
1132 	KASSERT(size > 0);
1133 
1134 	KASSERT((~flags & (UVM_FLAG_NOWAIT | UVM_FLAG_WAITVA)) != 0);
1135 
1136 	uvm_map_check(map, "map entry");
1137 
1138 	/*
1139 	 * check sanity of protection code
1140 	 */
1141 
1142 	if ((prot & maxprot) != prot) {
1143 		UVMHIST_LOG(maphist, "<- prot. failure:  prot=%#jx, max=%#jx",
1144 		prot, maxprot,0,0);
1145 		return EACCES;
1146 	}
1147 
1148 	/*
1149 	 * figure out where to put new VM range
1150 	 */
1151 retry:
1152 	if (vm_map_lock_try(map) == false) {
1153 		if ((flags & UVM_FLAG_TRYLOCK) != 0) {
1154 			return EAGAIN;
1155 		}
1156 		vm_map_lock(map); /* could sleep here */
1157 	}
1158 	if (flags & UVM_FLAG_UNMAP) {
1159 		KASSERT(flags & UVM_FLAG_FIXED);
1160 		KASSERT((flags & UVM_FLAG_NOWAIT) == 0);
1161 
1162 		/*
1163 		 * Set prev_entry to what it will need to be after any existing
1164 		 * entries are removed later in uvm_map_enter().
1165 		 */
1166 
1167 		if (uvm_map_lookup_entry(map, start, &prev_entry)) {
1168 			if (start == prev_entry->start)
1169 				prev_entry = prev_entry->prev;
1170 			else
1171 				UVM_MAP_CLIP_END(map, prev_entry, start);
1172 			SAVE_HINT(map, map->hint, prev_entry);
1173 		}
1174 	} else {
1175 		prev_entry = uvm_map_findspace(map, start, size, &start,
1176 		    uobj, uoffset, align, flags);
1177 	}
1178 	if (prev_entry == NULL) {
1179 		unsigned int timestamp;
1180 
1181 		timestamp = map->timestamp;
1182 		UVMHIST_LOG(maphist,"waiting va timestamp=%#jx",
1183 			    timestamp,0,0,0);
1184 		map->flags |= VM_MAP_WANTVA;
1185 		vm_map_unlock(map);
1186 
1187 		/*
1188 		 * try to reclaim kva and wait until someone does unmap.
1189 		 * fragile locking here, so we awaken every second to
1190 		 * recheck the condition.
1191 		 */
1192 
1193 		mutex_enter(&map->misc_lock);
1194 		while ((map->flags & VM_MAP_WANTVA) != 0 &&
1195 		   map->timestamp == timestamp) {
1196 			if ((flags & UVM_FLAG_WAITVA) == 0) {
1197 				mutex_exit(&map->misc_lock);
1198 				UVMHIST_LOG(maphist,
1199 				    "<- uvm_map_findspace failed!", 0,0,0,0);
1200 				return ENOMEM;
1201 			} else {
1202 				cv_timedwait(&map->cv, &map->misc_lock, hz);
1203 			}
1204 		}
1205 		mutex_exit(&map->misc_lock);
1206 		goto retry;
1207 	}
1208 
1209 #ifdef PMAP_GROWKERNEL
1210 	/*
1211 	 * If the kernel pmap can't map the requested space,
1212 	 * then allocate more resources for it.
1213 	 */
1214 	if (map == kernel_map && uvm_maxkaddr < (start + size))
1215 		uvm_maxkaddr = pmap_growkernel(start + size);
1216 #endif
1217 
1218 	UVMMAP_EVCNT_INCR(map_call);
1219 
1220 	/*
1221 	 * if uobj is null, then uoffset is either a VAC hint for PMAP_PREFER
1222 	 * [typically from uvm_map_reserve] or it is UVM_UNKNOWN_OFFSET.   in
1223 	 * either case we want to zero it  before storing it in the map entry
1224 	 * (because it looks strange and confusing when debugging...)
1225 	 *
1226 	 * if uobj is not null
1227 	 *   if uoffset is not UVM_UNKNOWN_OFFSET then we have a normal mapping
1228 	 *      and we do not need to change uoffset.
1229 	 *   if uoffset is UVM_UNKNOWN_OFFSET then we need to find the offset
1230 	 *      now (based on the starting address of the map).   this case is
1231 	 *      for kernel object mappings where we don't know the offset until
1232 	 *      the virtual address is found (with uvm_map_findspace).   the
1233 	 *      offset is the distance we are from the start of the map.
1234 	 */
1235 
1236 	if (uobj == NULL) {
1237 		uoffset = 0;
1238 	} else {
1239 		if (uoffset == UVM_UNKNOWN_OFFSET) {
1240 			KASSERT(UVM_OBJ_IS_KERN_OBJECT(uobj));
1241 			uoffset = start - vm_map_min(kernel_map);
1242 		}
1243 	}
1244 
1245 	args->uma_flags = flags;
1246 	args->uma_prev = prev_entry;
1247 	args->uma_start = start;
1248 	args->uma_size = size;
1249 	args->uma_uobj = uobj;
1250 	args->uma_uoffset = uoffset;
1251 
1252 	UVMHIST_LOG(maphist, "<- done!", 0,0,0,0);
1253 	return 0;
1254 }
1255 
1256 /*
1257  * uvm_map_enter:
1258  *
1259  * called with map locked.
1260  * unlock the map before returning.
1261  */
1262 
1263 int
1264 uvm_map_enter(struct vm_map *map, const struct uvm_map_args *args,
1265     struct vm_map_entry *new_entry)
1266 {
1267 	struct vm_map_entry *prev_entry = args->uma_prev;
1268 	struct vm_map_entry *dead = NULL, *dead_entries = NULL;
1269 
1270 	const uvm_flag_t flags = args->uma_flags;
1271 	const vm_prot_t prot = UVM_PROTECTION(flags);
1272 	const vm_prot_t maxprot = UVM_MAXPROTECTION(flags);
1273 	const vm_inherit_t inherit = UVM_INHERIT(flags);
1274 	const int amapwaitflag = (flags & UVM_FLAG_NOWAIT) ?
1275 	    AMAP_EXTEND_NOWAIT : 0;
1276 	const int advice = UVM_ADVICE(flags);
1277 
1278 	vaddr_t start = args->uma_start;
1279 	vsize_t size = args->uma_size;
1280 	struct uvm_object *uobj = args->uma_uobj;
1281 	voff_t uoffset = args->uma_uoffset;
1282 
1283 	const int kmap = (vm_map_pmap(map) == pmap_kernel());
1284 	int merged = 0;
1285 	int error;
1286 	int newetype;
1287 
1288 	UVMHIST_FUNC("uvm_map_enter");
1289 	UVMHIST_CALLED(maphist);
1290 
1291 	UVMHIST_LOG(maphist, "(map=%#jx, start=%#jx, size=%ju, flags=%#jx)",
1292 	    (uintptr_t)map, start, size, flags);
1293 	UVMHIST_LOG(maphist, "  uobj/offset %#jx/%jd", (uintptr_t)uobj,
1294 	    uoffset,0,0);
1295 
1296 	KASSERT(map->hint == prev_entry); /* bimerge case assumes this */
1297 	KASSERT(vm_map_locked_p(map));
1298 	KASSERT((flags & (UVM_FLAG_NOWAIT | UVM_FLAG_UNMAP)) !=
1299 		(UVM_FLAG_NOWAIT | UVM_FLAG_UNMAP));
1300 
1301 	if (uobj)
1302 		newetype = UVM_ET_OBJ;
1303 	else
1304 		newetype = 0;
1305 
1306 	if (flags & UVM_FLAG_COPYONW) {
1307 		newetype |= UVM_ET_COPYONWRITE;
1308 		if ((flags & UVM_FLAG_OVERLAY) == 0)
1309 			newetype |= UVM_ET_NEEDSCOPY;
1310 	}
1311 
1312 	/*
1313 	 * For mappings with unmap, remove any old entries now.  Adding the new
1314 	 * entry cannot fail because that can only happen if UVM_FLAG_NOWAIT
1315 	 * is set, and we do not support nowait and unmap together.
1316 	 */
1317 
1318 	if (flags & UVM_FLAG_UNMAP) {
1319 		KASSERT(flags & UVM_FLAG_FIXED);
1320 		uvm_unmap_remove(map, start, start + size, &dead_entries, 0);
1321 #ifdef DEBUG
1322 		struct vm_map_entry *tmp_entry __diagused;
1323 		bool rv __diagused;
1324 
1325 		rv = uvm_map_lookup_entry(map, start, &tmp_entry);
1326 		KASSERT(!rv);
1327 		KASSERTMSG(prev_entry == tmp_entry,
1328 			   "args %p prev_entry %p tmp_entry %p",
1329 			   args, prev_entry, tmp_entry);
1330 #endif
1331 		SAVE_HINT(map, map->hint, prev_entry);
1332 	}
1333 
1334 	/*
1335 	 * try and insert in map by extending previous entry, if possible.
1336 	 * XXX: we don't try and pull back the next entry.   might be useful
1337 	 * for a stack, but we are currently allocating our stack in advance.
1338 	 */
1339 
1340 	if (flags & UVM_FLAG_NOMERGE)
1341 		goto nomerge;
1342 
1343 	if (prev_entry->end == start &&
1344 	    prev_entry != &map->header &&
1345 	    UVM_ET_ISCOMPATIBLE(prev_entry, newetype, uobj, 0,
1346 	    prot, maxprot, inherit, advice, 0)) {
1347 
1348 		if (uobj && prev_entry->offset +
1349 		    (prev_entry->end - prev_entry->start) != uoffset)
1350 			goto forwardmerge;
1351 
1352 		/*
1353 		 * can't extend a shared amap.  note: no need to lock amap to
1354 		 * look at refs since we don't care about its exact value.
1355 		 * if it is one (i.e. we have only reference) it will stay there
1356 		 */
1357 
1358 		if (prev_entry->aref.ar_amap &&
1359 		    amap_refs(prev_entry->aref.ar_amap) != 1) {
1360 			goto forwardmerge;
1361 		}
1362 
1363 		if (prev_entry->aref.ar_amap) {
1364 			error = amap_extend(prev_entry, size,
1365 			    amapwaitflag | AMAP_EXTEND_FORWARDS);
1366 			if (error)
1367 				goto nomerge;
1368 		}
1369 
1370 		if (kmap) {
1371 			UVMMAP_EVCNT_INCR(kbackmerge);
1372 		} else {
1373 			UVMMAP_EVCNT_INCR(ubackmerge);
1374 		}
1375 		UVMHIST_LOG(maphist,"  starting back merge", 0, 0, 0, 0);
1376 
1377 		/*
1378 		 * drop our reference to uobj since we are extending a reference
1379 		 * that we already have (the ref count can not drop to zero).
1380 		 */
1381 
1382 		if (uobj && uobj->pgops->pgo_detach)
1383 			uobj->pgops->pgo_detach(uobj);
1384 
1385 		/*
1386 		 * Now that we've merged the entries, note that we've grown
1387 		 * and our gap has shrunk.  Then fix the tree.
1388 		 */
1389 		prev_entry->end += size;
1390 		prev_entry->gap -= size;
1391 		uvm_rb_fixup(map, prev_entry);
1392 
1393 		uvm_map_check(map, "map backmerged");
1394 
1395 		UVMHIST_LOG(maphist,"<- done (via backmerge)!", 0, 0, 0, 0);
1396 		merged++;
1397 	}
1398 
1399 forwardmerge:
1400 	if (prev_entry->next->start == (start + size) &&
1401 	    prev_entry->next != &map->header &&
1402 	    UVM_ET_ISCOMPATIBLE(prev_entry->next, newetype, uobj, 0,
1403 	    prot, maxprot, inherit, advice, 0)) {
1404 
1405 		if (uobj && prev_entry->next->offset != uoffset + size)
1406 			goto nomerge;
1407 
1408 		/*
1409 		 * can't extend a shared amap.  note: no need to lock amap to
1410 		 * look at refs since we don't care about its exact value.
1411 		 * if it is one (i.e. we have only reference) it will stay there.
1412 		 *
1413 		 * note that we also can't merge two amaps, so if we
1414 		 * merged with the previous entry which has an amap,
1415 		 * and the next entry also has an amap, we give up.
1416 		 *
1417 		 * Interesting cases:
1418 		 * amap, new, amap -> give up second merge (single fwd extend)
1419 		 * amap, new, none -> double forward extend (extend again here)
1420 		 * none, new, amap -> double backward extend (done here)
1421 		 * uobj, new, amap -> single backward extend (done here)
1422 		 *
1423 		 * XXX should we attempt to deal with someone refilling
1424 		 * the deallocated region between two entries that are
1425 		 * backed by the same amap (ie, arefs is 2, "prev" and
1426 		 * "next" refer to it, and adding this allocation will
1427 		 * close the hole, thus restoring arefs to 1 and
1428 		 * deallocating the "next" vm_map_entry)?  -- @@@
1429 		 */
1430 
1431 		if (prev_entry->next->aref.ar_amap &&
1432 		    (amap_refs(prev_entry->next->aref.ar_amap) != 1 ||
1433 		     (merged && prev_entry->aref.ar_amap))) {
1434 			goto nomerge;
1435 		}
1436 
1437 		if (merged) {
1438 			/*
1439 			 * Try to extend the amap of the previous entry to
1440 			 * cover the next entry as well.  If it doesn't work
1441 			 * just skip on, don't actually give up, since we've
1442 			 * already completed the back merge.
1443 			 */
1444 			if (prev_entry->aref.ar_amap) {
1445 				if (amap_extend(prev_entry,
1446 				    prev_entry->next->end -
1447 				    prev_entry->next->start,
1448 				    amapwaitflag | AMAP_EXTEND_FORWARDS))
1449 					goto nomerge;
1450 			}
1451 
1452 			/*
1453 			 * Try to extend the amap of the *next* entry
1454 			 * back to cover the new allocation *and* the
1455 			 * previous entry as well (the previous merge
1456 			 * didn't have an amap already otherwise we
1457 			 * wouldn't be checking here for an amap).  If
1458 			 * it doesn't work just skip on, again, don't
1459 			 * actually give up, since we've already
1460 			 * completed the back merge.
1461 			 */
1462 			else if (prev_entry->next->aref.ar_amap) {
1463 				if (amap_extend(prev_entry->next,
1464 				    prev_entry->end -
1465 				    prev_entry->start,
1466 				    amapwaitflag | AMAP_EXTEND_BACKWARDS))
1467 					goto nomerge;
1468 			}
1469 		} else {
1470 			/*
1471 			 * Pull the next entry's amap backwards to cover this
1472 			 * new allocation.
1473 			 */
1474 			if (prev_entry->next->aref.ar_amap) {
1475 				error = amap_extend(prev_entry->next, size,
1476 				    amapwaitflag | AMAP_EXTEND_BACKWARDS);
1477 				if (error)
1478 					goto nomerge;
1479 			}
1480 		}
1481 
1482 		if (merged) {
1483 			if (kmap) {
1484 				UVMMAP_EVCNT_DECR(kbackmerge);
1485 				UVMMAP_EVCNT_INCR(kbimerge);
1486 			} else {
1487 				UVMMAP_EVCNT_DECR(ubackmerge);
1488 				UVMMAP_EVCNT_INCR(ubimerge);
1489 			}
1490 		} else {
1491 			if (kmap) {
1492 				UVMMAP_EVCNT_INCR(kforwmerge);
1493 			} else {
1494 				UVMMAP_EVCNT_INCR(uforwmerge);
1495 			}
1496 		}
1497 		UVMHIST_LOG(maphist,"  starting forward merge", 0, 0, 0, 0);
1498 
1499 		/*
1500 		 * drop our reference to uobj since we are extending a reference
1501 		 * that we already have (the ref count can not drop to zero).
1502 		 */
1503 		if (uobj && uobj->pgops->pgo_detach)
1504 			uobj->pgops->pgo_detach(uobj);
1505 
1506 		if (merged) {
1507 			dead = prev_entry->next;
1508 			prev_entry->end = dead->end;
1509 			uvm_map_entry_unlink(map, dead);
1510 			if (dead->aref.ar_amap != NULL) {
1511 				prev_entry->aref = dead->aref;
1512 				dead->aref.ar_amap = NULL;
1513 			}
1514 		} else {
1515 			prev_entry->next->start -= size;
1516 			if (prev_entry != &map->header) {
1517 				prev_entry->gap -= size;
1518 				KASSERT(prev_entry->gap == uvm_rb_gap(prev_entry));
1519 				uvm_rb_fixup(map, prev_entry);
1520 			}
1521 			if (uobj)
1522 				prev_entry->next->offset = uoffset;
1523 		}
1524 
1525 		uvm_map_check(map, "map forwardmerged");
1526 
1527 		UVMHIST_LOG(maphist,"<- done forwardmerge", 0, 0, 0, 0);
1528 		merged++;
1529 	}
1530 
1531 nomerge:
1532 	if (!merged) {
1533 		UVMHIST_LOG(maphist,"  allocating new map entry", 0, 0, 0, 0);
1534 		if (kmap) {
1535 			UVMMAP_EVCNT_INCR(knomerge);
1536 		} else {
1537 			UVMMAP_EVCNT_INCR(unomerge);
1538 		}
1539 
1540 		/*
1541 		 * allocate new entry and link it in.
1542 		 */
1543 
1544 		if (new_entry == NULL) {
1545 			new_entry = uvm_mapent_alloc(map,
1546 				(flags & UVM_FLAG_NOWAIT));
1547 			if (__predict_false(new_entry == NULL)) {
1548 				error = ENOMEM;
1549 				goto done;
1550 			}
1551 		}
1552 		new_entry->start = start;
1553 		new_entry->end = new_entry->start + size;
1554 		new_entry->object.uvm_obj = uobj;
1555 		new_entry->offset = uoffset;
1556 
1557 		new_entry->etype = newetype;
1558 
1559 		if (flags & UVM_FLAG_NOMERGE) {
1560 			new_entry->flags |= UVM_MAP_NOMERGE;
1561 		}
1562 
1563 		new_entry->protection = prot;
1564 		new_entry->max_protection = maxprot;
1565 		new_entry->inheritance = inherit;
1566 		new_entry->wired_count = 0;
1567 		new_entry->advice = advice;
1568 		if (flags & UVM_FLAG_OVERLAY) {
1569 
1570 			/*
1571 			 * to_add: for BSS we overallocate a little since we
1572 			 * are likely to extend
1573 			 */
1574 
1575 			vaddr_t to_add = (flags & UVM_FLAG_AMAPPAD) ?
1576 				UVM_AMAP_CHUNK << PAGE_SHIFT : 0;
1577 			struct vm_amap *amap = amap_alloc(size, to_add,
1578 			    (flags & UVM_FLAG_NOWAIT));
1579 			if (__predict_false(amap == NULL)) {
1580 				error = ENOMEM;
1581 				goto done;
1582 			}
1583 			new_entry->aref.ar_pageoff = 0;
1584 			new_entry->aref.ar_amap = amap;
1585 		} else {
1586 			new_entry->aref.ar_pageoff = 0;
1587 			new_entry->aref.ar_amap = NULL;
1588 		}
1589 		uvm_map_entry_link(map, prev_entry, new_entry);
1590 
1591 		/*
1592 		 * Update the free space hint
1593 		 */
1594 
1595 		if ((map->first_free == prev_entry) &&
1596 		    (prev_entry->end >= new_entry->start))
1597 			map->first_free = new_entry;
1598 
1599 		new_entry = NULL;
1600 	}
1601 
1602 	map->size += size;
1603 
1604 	UVMHIST_LOG(maphist,"<- done!", 0, 0, 0, 0);
1605 
1606 	error = 0;
1607 
1608 done:
1609 	vm_map_unlock(map);
1610 
1611 	if (new_entry) {
1612 		uvm_mapent_free(new_entry);
1613 	}
1614 	if (dead) {
1615 		KDASSERT(merged);
1616 		uvm_mapent_free(dead);
1617 	}
1618 	if (dead_entries)
1619 		uvm_unmap_detach(dead_entries, 0);
1620 
1621 	return error;
1622 }
1623 
1624 /*
1625  * uvm_map_lookup_entry_bytree: lookup an entry in tree
1626  */
1627 
1628 static inline bool
1629 uvm_map_lookup_entry_bytree(struct vm_map *map, vaddr_t address,
1630     struct vm_map_entry **entry	/* OUT */)
1631 {
1632 	struct vm_map_entry *prev = &map->header;
1633 	struct vm_map_entry *cur = ROOT_ENTRY(map);
1634 
1635 	while (cur) {
1636 		UVMMAP_EVCNT_INCR(mlk_treeloop);
1637 		if (address >= cur->start) {
1638 			if (address < cur->end) {
1639 				*entry = cur;
1640 				return true;
1641 			}
1642 			prev = cur;
1643 			cur = RIGHT_ENTRY(cur);
1644 		} else
1645 			cur = LEFT_ENTRY(cur);
1646 	}
1647 	*entry = prev;
1648 	return false;
1649 }
1650 
1651 /*
1652  * uvm_map_lookup_entry: find map entry at or before an address
1653  *
1654  * => map must at least be read-locked by caller
1655  * => entry is returned in "entry"
1656  * => return value is true if address is in the returned entry
1657  */
1658 
1659 bool
1660 uvm_map_lookup_entry(struct vm_map *map, vaddr_t address,
1661     struct vm_map_entry **entry	/* OUT */)
1662 {
1663 	struct vm_map_entry *cur;
1664 	bool use_tree = false;
1665 	UVMHIST_FUNC("uvm_map_lookup_entry");
1666 	UVMHIST_CALLED(maphist);
1667 
1668 	UVMHIST_LOG(maphist,"(map=%#jx,addr=%#jx,ent=%#jx)",
1669 	    (uintptr_t)map, address, (uintptr_t)entry, 0);
1670 
1671 	/*
1672 	 * start looking either from the head of the
1673 	 * list, or from the hint.
1674 	 */
1675 
1676 	cur = map->hint;
1677 
1678 	if (cur == &map->header)
1679 		cur = cur->next;
1680 
1681 	UVMMAP_EVCNT_INCR(mlk_call);
1682 	if (address >= cur->start) {
1683 
1684 		/*
1685 		 * go from hint to end of list.
1686 		 *
1687 		 * but first, make a quick check to see if
1688 		 * we are already looking at the entry we
1689 		 * want (which is usually the case).
1690 		 * note also that we don't need to save the hint
1691 		 * here... it is the same hint (unless we are
1692 		 * at the header, in which case the hint didn't
1693 		 * buy us anything anyway).
1694 		 */
1695 
1696 		if (cur != &map->header && cur->end > address) {
1697 			UVMMAP_EVCNT_INCR(mlk_hint);
1698 			*entry = cur;
1699 			UVMHIST_LOG(maphist,"<- got it via hint (%#jx)",
1700 			    (uintptr_t)cur, 0, 0, 0);
1701 			uvm_mapent_check(*entry);
1702 			return (true);
1703 		}
1704 
1705 		if (map->nentries > 15)
1706 			use_tree = true;
1707 	} else {
1708 
1709 		/*
1710 		 * invalid hint.  use tree.
1711 		 */
1712 		use_tree = true;
1713 	}
1714 
1715 	uvm_map_check(map, __func__);
1716 
1717 	if (use_tree) {
1718 		/*
1719 		 * Simple lookup in the tree.  Happens when the hint is
1720 		 * invalid, or nentries reach a threshold.
1721 		 */
1722 		UVMMAP_EVCNT_INCR(mlk_tree);
1723 		if (uvm_map_lookup_entry_bytree(map, address, entry)) {
1724 			goto got;
1725 		} else {
1726 			goto failed;
1727 		}
1728 	}
1729 
1730 	/*
1731 	 * search linearly
1732 	 */
1733 
1734 	UVMMAP_EVCNT_INCR(mlk_list);
1735 	while (cur != &map->header) {
1736 		UVMMAP_EVCNT_INCR(mlk_listloop);
1737 		if (cur->end > address) {
1738 			if (address >= cur->start) {
1739 				/*
1740 				 * save this lookup for future
1741 				 * hints, and return
1742 				 */
1743 
1744 				*entry = cur;
1745 got:
1746 				SAVE_HINT(map, map->hint, *entry);
1747 				UVMHIST_LOG(maphist,"<- search got it (%#jx)",
1748 					(uintptr_t)cur, 0, 0, 0);
1749 				KDASSERT((*entry)->start <= address);
1750 				KDASSERT(address < (*entry)->end);
1751 				uvm_mapent_check(*entry);
1752 				return (true);
1753 			}
1754 			break;
1755 		}
1756 		cur = cur->next;
1757 	}
1758 	*entry = cur->prev;
1759 failed:
1760 	SAVE_HINT(map, map->hint, *entry);
1761 	UVMHIST_LOG(maphist,"<- failed!",0,0,0,0);
1762 	KDASSERT((*entry) == &map->header || (*entry)->end <= address);
1763 	KDASSERT((*entry)->next == &map->header ||
1764 	    address < (*entry)->next->start);
1765 	return (false);
1766 }
1767 
1768 /*
1769  * See if the range between start and start + length fits in the gap
1770  * entry->next->start and entry->end.  Returns 1 if fits, 0 if doesn't
1771  * fit, and -1 address wraps around.
1772  */
1773 static int
1774 uvm_map_space_avail(vaddr_t *start, vsize_t length, voff_t uoffset,
1775     vsize_t align, int flags, int topdown, struct vm_map_entry *entry)
1776 {
1777 	vaddr_t end;
1778 
1779 #ifdef PMAP_PREFER
1780 	/*
1781 	 * push start address forward as needed to avoid VAC alias problems.
1782 	 * we only do this if a valid offset is specified.
1783 	 */
1784 
1785 	if (uoffset != UVM_UNKNOWN_OFFSET)
1786 		PMAP_PREFER(uoffset, start, length, topdown);
1787 #endif
1788 	if ((flags & UVM_FLAG_COLORMATCH) != 0) {
1789 		KASSERT(align < uvmexp.ncolors);
1790 		if (uvmexp.ncolors > 1) {
1791 			const u_int colormask = uvmexp.colormask;
1792 			const u_int colorsize = colormask + 1;
1793 			vaddr_t hint = atop(*start);
1794 			const u_int color = hint & colormask;
1795 			if (color != align) {
1796 				hint -= color;	/* adjust to color boundary */
1797 				KASSERT((hint & colormask) == 0);
1798 				if (topdown) {
1799 					if (align > color)
1800 						hint -= colorsize;
1801 				} else {
1802 					if (align < color)
1803 						hint += colorsize;
1804 				}
1805 				*start = ptoa(hint + align); /* adjust to color */
1806 			}
1807 		}
1808 	} else if (align != 0) {
1809 		if ((*start & (align - 1)) != 0) {
1810 			if (topdown)
1811 				*start &= ~(align - 1);
1812 			else
1813 				*start = roundup(*start, align);
1814 		}
1815 		/*
1816 		 * XXX Should we PMAP_PREFER() here again?
1817 		 * eh...i think we're okay
1818 		 */
1819 	}
1820 
1821 	/*
1822 	 * Find the end of the proposed new region.  Be sure we didn't
1823 	 * wrap around the address; if so, we lose.  Otherwise, if the
1824 	 * proposed new region fits before the next entry, we win.
1825 	 */
1826 
1827 	end = *start + length;
1828 	if (end < *start)
1829 		return (-1);
1830 
1831 	if (entry->next->start >= end && *start >= entry->end)
1832 		return (1);
1833 
1834 	return (0);
1835 }
1836 
1837 /*
1838  * uvm_map_findspace: find "length" sized space in "map".
1839  *
1840  * => "hint" is a hint about where we want it, unless UVM_FLAG_FIXED is
1841  *	set in "flags" (in which case we insist on using "hint").
1842  * => "result" is VA returned
1843  * => uobj/uoffset are to be used to handle VAC alignment, if required
1844  * => if "align" is non-zero, we attempt to align to that value.
1845  * => caller must at least have read-locked map
1846  * => returns NULL on failure, or pointer to prev. map entry if success
1847  * => note this is a cross between the old vm_map_findspace and vm_map_find
1848  */
1849 
1850 struct vm_map_entry *
1851 uvm_map_findspace(struct vm_map *map, vaddr_t hint, vsize_t length,
1852     vaddr_t *result /* OUT */, struct uvm_object *uobj, voff_t uoffset,
1853     vsize_t align, int flags)
1854 {
1855 	struct vm_map_entry *entry;
1856 	struct vm_map_entry *child, *prev, *tmp;
1857 	vaddr_t orig_hint __diagused;
1858 	const int topdown = map->flags & VM_MAP_TOPDOWN;
1859 	UVMHIST_FUNC("uvm_map_findspace");
1860 	UVMHIST_CALLED(maphist);
1861 
1862 	UVMHIST_LOG(maphist, "(map=%#jx, hint=%#jx, len=%ju, flags=%#jx)",
1863 	    (uintptr_t)map, hint, length, flags);
1864 	KASSERT((flags & UVM_FLAG_COLORMATCH) != 0 || (align & (align - 1)) == 0);
1865 	KASSERT((flags & UVM_FLAG_COLORMATCH) == 0 || align < uvmexp.ncolors);
1866 	KASSERT((flags & UVM_FLAG_FIXED) == 0 || align == 0);
1867 
1868 	uvm_map_check(map, "map_findspace entry");
1869 
1870 	/*
1871 	 * remember the original hint.  if we are aligning, then we
1872 	 * may have to try again with no alignment constraint if
1873 	 * we fail the first time.
1874 	 */
1875 
1876 	orig_hint = hint;
1877 	if (hint < vm_map_min(map)) {	/* check ranges ... */
1878 		if (flags & UVM_FLAG_FIXED) {
1879 			UVMHIST_LOG(maphist,"<- VA below map range",0,0,0,0);
1880 			return (NULL);
1881 		}
1882 		hint = vm_map_min(map);
1883 	}
1884 	if (hint > vm_map_max(map)) {
1885 		UVMHIST_LOG(maphist,"<- VA %#jx > range [%#jx->%#jx]",
1886 		    hint, vm_map_min(map), vm_map_max(map), 0);
1887 		return (NULL);
1888 	}
1889 
1890 	/*
1891 	 * Look for the first possible address; if there's already
1892 	 * something at this address, we have to start after it.
1893 	 */
1894 
1895 	/*
1896 	 * @@@: there are four, no, eight cases to consider.
1897 	 *
1898 	 * 0: found,     fixed,     bottom up -> fail
1899 	 * 1: found,     fixed,     top down  -> fail
1900 	 * 2: found,     not fixed, bottom up -> start after entry->end,
1901 	 *                                       loop up
1902 	 * 3: found,     not fixed, top down  -> start before entry->start,
1903 	 *                                       loop down
1904 	 * 4: not found, fixed,     bottom up -> check entry->next->start, fail
1905 	 * 5: not found, fixed,     top down  -> check entry->next->start, fail
1906 	 * 6: not found, not fixed, bottom up -> check entry->next->start,
1907 	 *                                       loop up
1908 	 * 7: not found, not fixed, top down  -> check entry->next->start,
1909 	 *                                       loop down
1910 	 *
1911 	 * as you can see, it reduces to roughly five cases, and that
1912 	 * adding top down mapping only adds one unique case (without
1913 	 * it, there would be four cases).
1914 	 */
1915 
1916 	if ((flags & UVM_FLAG_FIXED) == 0 && hint == vm_map_min(map)) {
1917 		entry = map->first_free;
1918 	} else {
1919 		if (uvm_map_lookup_entry(map, hint, &entry)) {
1920 			/* "hint" address already in use ... */
1921 			if (flags & UVM_FLAG_FIXED) {
1922 				UVMHIST_LOG(maphist, "<- fixed & VA in use",
1923 				    0, 0, 0, 0);
1924 				return (NULL);
1925 			}
1926 			if (topdown)
1927 				/* Start from lower gap. */
1928 				entry = entry->prev;
1929 		} else if (flags & UVM_FLAG_FIXED) {
1930 			if (entry->next->start >= hint + length &&
1931 			    hint + length > hint)
1932 				goto found;
1933 
1934 			/* "hint" address is gap but too small */
1935 			UVMHIST_LOG(maphist, "<- fixed mapping failed",
1936 			    0, 0, 0, 0);
1937 			return (NULL); /* only one shot at it ... */
1938 		} else {
1939 			/*
1940 			 * See if given hint fits in this gap.
1941 			 */
1942 			switch (uvm_map_space_avail(&hint, length,
1943 			    uoffset, align, flags, topdown, entry)) {
1944 			case 1:
1945 				goto found;
1946 			case -1:
1947 				goto wraparound;
1948 			}
1949 
1950 			if (topdown) {
1951 				/*
1952 				 * Still there is a chance to fit
1953 				 * if hint > entry->end.
1954 				 */
1955 			} else {
1956 				/* Start from higher gap. */
1957 				entry = entry->next;
1958 				if (entry == &map->header)
1959 					goto notfound;
1960 				goto nextgap;
1961 			}
1962 		}
1963 	}
1964 
1965 	/*
1966 	 * Note that all UVM_FLAGS_FIXED case is already handled.
1967 	 */
1968 	KDASSERT((flags & UVM_FLAG_FIXED) == 0);
1969 
1970 	/* Try to find the space in the red-black tree */
1971 
1972 	/* Check slot before any entry */
1973 	hint = topdown ? entry->next->start - length : entry->end;
1974 	switch (uvm_map_space_avail(&hint, length, uoffset, align, flags,
1975 	    topdown, entry)) {
1976 	case 1:
1977 		goto found;
1978 	case -1:
1979 		goto wraparound;
1980 	}
1981 
1982 nextgap:
1983 	KDASSERT((flags & UVM_FLAG_FIXED) == 0);
1984 	/* If there is not enough space in the whole tree, we fail */
1985 	tmp = ROOT_ENTRY(map);
1986 	if (tmp == NULL || tmp->maxgap < length)
1987 		goto notfound;
1988 
1989 	prev = NULL; /* previous candidate */
1990 
1991 	/* Find an entry close to hint that has enough space */
1992 	for (; tmp;) {
1993 		KASSERT(tmp->next->start == tmp->end + tmp->gap);
1994 		if (topdown) {
1995 			if (tmp->next->start < hint + length &&
1996 			    (prev == NULL || tmp->end > prev->end)) {
1997 				if (tmp->gap >= length)
1998 					prev = tmp;
1999 				else if ((child = LEFT_ENTRY(tmp)) != NULL
2000 				    && child->maxgap >= length)
2001 					prev = tmp;
2002 			}
2003 		} else {
2004 			if (tmp->end >= hint &&
2005 			    (prev == NULL || tmp->end < prev->end)) {
2006 				if (tmp->gap >= length)
2007 					prev = tmp;
2008 				else if ((child = RIGHT_ENTRY(tmp)) != NULL
2009 				    && child->maxgap >= length)
2010 					prev = tmp;
2011 			}
2012 		}
2013 		if (tmp->next->start < hint + length)
2014 			child = RIGHT_ENTRY(tmp);
2015 		else if (tmp->end > hint)
2016 			child = LEFT_ENTRY(tmp);
2017 		else {
2018 			if (tmp->gap >= length)
2019 				break;
2020 			if (topdown)
2021 				child = LEFT_ENTRY(tmp);
2022 			else
2023 				child = RIGHT_ENTRY(tmp);
2024 		}
2025 		if (child == NULL || child->maxgap < length)
2026 			break;
2027 		tmp = child;
2028 	}
2029 
2030 	if (tmp != NULL && tmp->start < hint && hint < tmp->next->start) {
2031 		/*
2032 		 * Check if the entry that we found satifies the
2033 		 * space requirement
2034 		 */
2035 		if (topdown) {
2036 			if (hint > tmp->next->start - length)
2037 				hint = tmp->next->start - length;
2038 		} else {
2039 			if (hint < tmp->end)
2040 				hint = tmp->end;
2041 		}
2042 		switch (uvm_map_space_avail(&hint, length, uoffset, align,
2043 		    flags, topdown, tmp)) {
2044 		case 1:
2045 			entry = tmp;
2046 			goto found;
2047 		case -1:
2048 			goto wraparound;
2049 		}
2050 		if (tmp->gap >= length)
2051 			goto listsearch;
2052 	}
2053 	if (prev == NULL)
2054 		goto notfound;
2055 
2056 	if (topdown) {
2057 		KASSERT(orig_hint >= prev->next->start - length ||
2058 		    prev->next->start - length > prev->next->start);
2059 		hint = prev->next->start - length;
2060 	} else {
2061 		KASSERT(orig_hint <= prev->end);
2062 		hint = prev->end;
2063 	}
2064 	switch (uvm_map_space_avail(&hint, length, uoffset, align,
2065 	    flags, topdown, prev)) {
2066 	case 1:
2067 		entry = prev;
2068 		goto found;
2069 	case -1:
2070 		goto wraparound;
2071 	}
2072 	if (prev->gap >= length)
2073 		goto listsearch;
2074 
2075 	if (topdown)
2076 		tmp = LEFT_ENTRY(prev);
2077 	else
2078 		tmp = RIGHT_ENTRY(prev);
2079 	for (;;) {
2080 		KASSERT(tmp && tmp->maxgap >= length);
2081 		if (topdown)
2082 			child = RIGHT_ENTRY(tmp);
2083 		else
2084 			child = LEFT_ENTRY(tmp);
2085 		if (child && child->maxgap >= length) {
2086 			tmp = child;
2087 			continue;
2088 		}
2089 		if (tmp->gap >= length)
2090 			break;
2091 		if (topdown)
2092 			tmp = LEFT_ENTRY(tmp);
2093 		else
2094 			tmp = RIGHT_ENTRY(tmp);
2095 	}
2096 
2097 	if (topdown) {
2098 		KASSERT(orig_hint >= tmp->next->start - length ||
2099 		    tmp->next->start - length > tmp->next->start);
2100 		hint = tmp->next->start - length;
2101 	} else {
2102 		KASSERT(orig_hint <= tmp->end);
2103 		hint = tmp->end;
2104 	}
2105 	switch (uvm_map_space_avail(&hint, length, uoffset, align,
2106 	    flags, topdown, tmp)) {
2107 	case 1:
2108 		entry = tmp;
2109 		goto found;
2110 	case -1:
2111 		goto wraparound;
2112 	}
2113 
2114 	/*
2115 	 * The tree fails to find an entry because of offset or alignment
2116 	 * restrictions.  Search the list instead.
2117 	 */
2118  listsearch:
2119 	/*
2120 	 * Look through the rest of the map, trying to fit a new region in
2121 	 * the gap between existing regions, or after the very last region.
2122 	 * note: entry->end = base VA of current gap,
2123 	 *	 entry->next->start = VA of end of current gap
2124 	 */
2125 
2126 	for (;;) {
2127 		/* Update hint for current gap. */
2128 		hint = topdown ? entry->next->start - length : entry->end;
2129 
2130 		/* See if it fits. */
2131 		switch (uvm_map_space_avail(&hint, length, uoffset, align,
2132 		    flags, topdown, entry)) {
2133 		case 1:
2134 			goto found;
2135 		case -1:
2136 			goto wraparound;
2137 		}
2138 
2139 		/* Advance to next/previous gap */
2140 		if (topdown) {
2141 			if (entry == &map->header) {
2142 				UVMHIST_LOG(maphist, "<- failed (off start)",
2143 				    0,0,0,0);
2144 				goto notfound;
2145 			}
2146 			entry = entry->prev;
2147 		} else {
2148 			entry = entry->next;
2149 			if (entry == &map->header) {
2150 				UVMHIST_LOG(maphist, "<- failed (off end)",
2151 				    0,0,0,0);
2152 				goto notfound;
2153 			}
2154 		}
2155 	}
2156 
2157  found:
2158 	SAVE_HINT(map, map->hint, entry);
2159 	*result = hint;
2160 	UVMHIST_LOG(maphist,"<- got it!  (result=%#jx)", hint, 0,0,0);
2161 	KASSERTMSG( topdown || hint >= orig_hint, "hint: %jx, orig_hint: %jx",
2162 	    (uintmax_t)hint, (uintmax_t)orig_hint);
2163 	KASSERTMSG(!topdown || hint <= orig_hint, "hint: %jx, orig_hint: %jx",
2164 	    (uintmax_t)hint, (uintmax_t)orig_hint);
2165 	KASSERT(entry->end <= hint);
2166 	KASSERT(hint + length <= entry->next->start);
2167 	return (entry);
2168 
2169  wraparound:
2170 	UVMHIST_LOG(maphist, "<- failed (wrap around)", 0,0,0,0);
2171 
2172 	return (NULL);
2173 
2174  notfound:
2175 	UVMHIST_LOG(maphist, "<- failed (notfound)", 0,0,0,0);
2176 
2177 	return (NULL);
2178 }
2179 
2180 /*
2181  *   U N M A P   -   m a i n   h e l p e r   f u n c t i o n s
2182  */
2183 
2184 /*
2185  * uvm_unmap_remove: remove mappings from a vm_map (from "start" up to "stop")
2186  *
2187  * => caller must check alignment and size
2188  * => map must be locked by caller
2189  * => we return a list of map entries that we've remove from the map
2190  *    in "entry_list"
2191  */
2192 
2193 void
2194 uvm_unmap_remove(struct vm_map *map, vaddr_t start, vaddr_t end,
2195     struct vm_map_entry **entry_list /* OUT */, int flags)
2196 {
2197 	struct vm_map_entry *entry, *first_entry, *next;
2198 	vaddr_t len;
2199 	UVMHIST_FUNC("uvm_unmap_remove"); UVMHIST_CALLED(maphist);
2200 
2201 	UVMHIST_LOG(maphist,"(map=%#jx, start=%#jx, end=%#jx)",
2202 	    (uintptr_t)map, start, end, 0);
2203 	VM_MAP_RANGE_CHECK(map, start, end);
2204 
2205 	uvm_map_check(map, "unmap_remove entry");
2206 
2207 	/*
2208 	 * find first entry
2209 	 */
2210 
2211 	if (uvm_map_lookup_entry(map, start, &first_entry) == true) {
2212 		/* clip and go... */
2213 		entry = first_entry;
2214 		UVM_MAP_CLIP_START(map, entry, start);
2215 		/* critical!  prevents stale hint */
2216 		SAVE_HINT(map, entry, entry->prev);
2217 	} else {
2218 		entry = first_entry->next;
2219 	}
2220 
2221 	/*
2222 	 * Save the free space hint
2223 	 */
2224 
2225 	if (map->first_free != &map->header && map->first_free->start >= start)
2226 		map->first_free = entry->prev;
2227 
2228 	/*
2229 	 * note: we now re-use first_entry for a different task.  we remove
2230 	 * a number of map entries from the map and save them in a linked
2231 	 * list headed by "first_entry".  once we remove them from the map
2232 	 * the caller should unlock the map and drop the references to the
2233 	 * backing objects [c.f. uvm_unmap_detach].  the object is to
2234 	 * separate unmapping from reference dropping.  why?
2235 	 *   [1] the map has to be locked for unmapping
2236 	 *   [2] the map need not be locked for reference dropping
2237 	 *   [3] dropping references may trigger pager I/O, and if we hit
2238 	 *       a pager that does synchronous I/O we may have to wait for it.
2239 	 *   [4] we would like all waiting for I/O to occur with maps unlocked
2240 	 *       so that we don't block other threads.
2241 	 */
2242 
2243 	first_entry = NULL;
2244 	*entry_list = NULL;
2245 
2246 	/*
2247 	 * break up the area into map entry sized regions and unmap.  note
2248 	 * that all mappings have to be removed before we can even consider
2249 	 * dropping references to amaps or VM objects (otherwise we could end
2250 	 * up with a mapping to a page on the free list which would be very bad)
2251 	 */
2252 
2253 	while ((entry != &map->header) && (entry->start < end)) {
2254 		KASSERT((entry->flags & UVM_MAP_STATIC) == 0);
2255 
2256 		UVM_MAP_CLIP_END(map, entry, end);
2257 		next = entry->next;
2258 		len = entry->end - entry->start;
2259 
2260 		/*
2261 		 * unwire before removing addresses from the pmap; otherwise
2262 		 * unwiring will put the entries back into the pmap (XXX).
2263 		 */
2264 
2265 		if (VM_MAPENT_ISWIRED(entry)) {
2266 			uvm_map_entry_unwire(map, entry);
2267 		}
2268 		if (flags & UVM_FLAG_VAONLY) {
2269 
2270 			/* nothing */
2271 
2272 		} else if ((map->flags & VM_MAP_PAGEABLE) == 0) {
2273 
2274 			/*
2275 			 * if the map is non-pageable, any pages mapped there
2276 			 * must be wired and entered with pmap_kenter_pa(),
2277 			 * and we should free any such pages immediately.
2278 			 * this is mostly used for kmem_map.
2279 			 */
2280 			KASSERT(vm_map_pmap(map) == pmap_kernel());
2281 
2282 			uvm_km_pgremove_intrsafe(map, entry->start, entry->end);
2283 		} else if (UVM_ET_ISOBJ(entry) &&
2284 			   UVM_OBJ_IS_KERN_OBJECT(entry->object.uvm_obj)) {
2285 			panic("%s: kernel object %p %p\n",
2286 			    __func__, map, entry);
2287 		} else if (UVM_ET_ISOBJ(entry) || entry->aref.ar_amap) {
2288 			/*
2289 			 * remove mappings the standard way.  lock object
2290 			 * and/or amap to ensure vm_page state does not
2291 			 * change while in pmap_remove().
2292 			 */
2293 
2294 			uvm_map_lock_entry(entry);
2295 			pmap_remove(map->pmap, entry->start, entry->end);
2296 			uvm_map_unlock_entry(entry);
2297 		}
2298 
2299 #if defined(UVMDEBUG)
2300 		/*
2301 		 * check if there's remaining mapping,
2302 		 * which is a bug in caller.
2303 		 */
2304 
2305 		vaddr_t va;
2306 		for (va = entry->start; va < entry->end;
2307 		    va += PAGE_SIZE) {
2308 			if (pmap_extract(vm_map_pmap(map), va, NULL)) {
2309 				panic("%s: %#"PRIxVADDR" has mapping",
2310 				    __func__, va);
2311 			}
2312 		}
2313 
2314 		if (VM_MAP_IS_KERNEL(map) && (flags & UVM_FLAG_NOWAIT) == 0) {
2315 			uvm_km_check_empty(map, entry->start,
2316 			    entry->end);
2317 		}
2318 #endif /* defined(UVMDEBUG) */
2319 
2320 		/*
2321 		 * remove entry from map and put it on our list of entries
2322 		 * that we've nuked.  then go to next entry.
2323 		 */
2324 
2325 		UVMHIST_LOG(maphist, "  removed map entry %#jx",
2326 		    (uintptr_t)entry, 0, 0, 0);
2327 
2328 		/* critical!  prevents stale hint */
2329 		SAVE_HINT(map, entry, entry->prev);
2330 
2331 		uvm_map_entry_unlink(map, entry);
2332 		KASSERT(map->size >= len);
2333 		map->size -= len;
2334 		entry->prev = NULL;
2335 		entry->next = first_entry;
2336 		first_entry = entry;
2337 		entry = next;
2338 	}
2339 
2340 	/*
2341 	 * Note: if map is dying, leave pmap_update() for pmap_destroy(),
2342 	 * which will be called later.
2343 	 */
2344 	if ((map->flags & VM_MAP_DYING) == 0) {
2345 		pmap_update(vm_map_pmap(map));
2346 	} else {
2347 		KASSERT(vm_map_pmap(map) != pmap_kernel());
2348 	}
2349 
2350 	uvm_map_check(map, "unmap_remove leave");
2351 
2352 	/*
2353 	 * now we've cleaned up the map and are ready for the caller to drop
2354 	 * references to the mapped objects.
2355 	 */
2356 
2357 	*entry_list = first_entry;
2358 	UVMHIST_LOG(maphist,"<- done!", 0, 0, 0, 0);
2359 
2360 	if (map->flags & VM_MAP_WANTVA) {
2361 		mutex_enter(&map->misc_lock);
2362 		map->flags &= ~VM_MAP_WANTVA;
2363 		cv_broadcast(&map->cv);
2364 		mutex_exit(&map->misc_lock);
2365 	}
2366 }
2367 
2368 /*
2369  * uvm_unmap_detach: drop references in a chain of map entries
2370  *
2371  * => we will free the map entries as we traverse the list.
2372  */
2373 
2374 void
2375 uvm_unmap_detach(struct vm_map_entry *first_entry, int flags)
2376 {
2377 	struct vm_map_entry *next_entry;
2378 	UVMHIST_FUNC("uvm_unmap_detach"); UVMHIST_CALLED(maphist);
2379 
2380 	while (first_entry) {
2381 		KASSERT(!VM_MAPENT_ISWIRED(first_entry));
2382 		UVMHIST_LOG(maphist,
2383 		    "  detach %#jx: amap=%#jx, obj=%#jx, submap?=%jd",
2384 		    (uintptr_t)first_entry,
2385 		    (uintptr_t)first_entry->aref.ar_amap,
2386 		    (uintptr_t)first_entry->object.uvm_obj,
2387 		    UVM_ET_ISSUBMAP(first_entry));
2388 
2389 		/*
2390 		 * drop reference to amap, if we've got one
2391 		 */
2392 
2393 		if (first_entry->aref.ar_amap)
2394 			uvm_map_unreference_amap(first_entry, flags);
2395 
2396 		/*
2397 		 * drop reference to our backing object, if we've got one
2398 		 */
2399 
2400 		KASSERT(!UVM_ET_ISSUBMAP(first_entry));
2401 		if (UVM_ET_ISOBJ(first_entry) &&
2402 		    first_entry->object.uvm_obj->pgops->pgo_detach) {
2403 			(*first_entry->object.uvm_obj->pgops->pgo_detach)
2404 				(first_entry->object.uvm_obj);
2405 		}
2406 		next_entry = first_entry->next;
2407 		uvm_mapent_free(first_entry);
2408 		first_entry = next_entry;
2409 	}
2410 	UVMHIST_LOG(maphist, "<- done", 0,0,0,0);
2411 }
2412 
2413 /*
2414  *   E X T R A C T I O N   F U N C T I O N S
2415  */
2416 
2417 /*
2418  * uvm_map_reserve: reserve space in a vm_map for future use.
2419  *
2420  * => we reserve space in a map by putting a dummy map entry in the
2421  *    map (dummy means obj=NULL, amap=NULL, prot=VM_PROT_NONE)
2422  * => map should be unlocked (we will write lock it)
2423  * => we return true if we were able to reserve space
2424  * => XXXCDC: should be inline?
2425  */
2426 
2427 int
2428 uvm_map_reserve(struct vm_map *map, vsize_t size,
2429     vaddr_t offset	/* hint for pmap_prefer */,
2430     vsize_t align	/* alignment */,
2431     vaddr_t *raddr	/* IN:hint, OUT: reserved VA */,
2432     uvm_flag_t flags	/* UVM_FLAG_FIXED or UVM_FLAG_COLORMATCH or 0 */)
2433 {
2434 	UVMHIST_FUNC("uvm_map_reserve"); UVMHIST_CALLED(maphist);
2435 
2436 	UVMHIST_LOG(maphist, "(map=%#jx, size=%#jx, offset=%#jx, addr=%#jx)",
2437 	    (uintptr_t)map, size, offset, (uintptr_t)raddr);
2438 
2439 	size = round_page(size);
2440 
2441 	/*
2442 	 * reserve some virtual space.
2443 	 */
2444 
2445 	if (uvm_map(map, raddr, size, NULL, offset, align,
2446 	    UVM_MAPFLAG(UVM_PROT_NONE, UVM_PROT_NONE, UVM_INH_NONE,
2447 	    UVM_ADV_RANDOM, UVM_FLAG_NOMERGE|flags)) != 0) {
2448 	    UVMHIST_LOG(maphist, "<- done (no VM)", 0,0,0,0);
2449 		return (false);
2450 	}
2451 
2452 	UVMHIST_LOG(maphist, "<- done (*raddr=%#jx)", *raddr,0,0,0);
2453 	return (true);
2454 }
2455 
2456 /*
2457  * uvm_map_replace: replace a reserved (blank) area of memory with
2458  * real mappings.
2459  *
2460  * => caller must WRITE-LOCK the map
2461  * => we return true if replacement was a success
2462  * => we expect the newents chain to have nnewents entrys on it and
2463  *    we expect newents->prev to point to the last entry on the list
2464  * => note newents is allowed to be NULL
2465  */
2466 
2467 static int
2468 uvm_map_replace(struct vm_map *map, vaddr_t start, vaddr_t end,
2469     struct vm_map_entry *newents, int nnewents, vsize_t nsize,
2470     struct vm_map_entry **oldentryp)
2471 {
2472 	struct vm_map_entry *oldent, *last;
2473 
2474 	uvm_map_check(map, "map_replace entry");
2475 
2476 	/*
2477 	 * first find the blank map entry at the specified address
2478 	 */
2479 
2480 	if (!uvm_map_lookup_entry(map, start, &oldent)) {
2481 		return (false);
2482 	}
2483 
2484 	/*
2485 	 * check to make sure we have a proper blank entry
2486 	 */
2487 
2488 	if (end < oldent->end) {
2489 		UVM_MAP_CLIP_END(map, oldent, end);
2490 	}
2491 	if (oldent->start != start || oldent->end != end ||
2492 	    oldent->object.uvm_obj != NULL || oldent->aref.ar_amap != NULL) {
2493 		return (false);
2494 	}
2495 
2496 #ifdef DIAGNOSTIC
2497 
2498 	/*
2499 	 * sanity check the newents chain
2500 	 */
2501 
2502 	{
2503 		struct vm_map_entry *tmpent = newents;
2504 		int nent = 0;
2505 		vsize_t sz = 0;
2506 		vaddr_t cur = start;
2507 
2508 		while (tmpent) {
2509 			nent++;
2510 			sz += tmpent->end - tmpent->start;
2511 			if (tmpent->start < cur)
2512 				panic("uvm_map_replace1");
2513 			if (tmpent->start >= tmpent->end || tmpent->end > end) {
2514 				panic("uvm_map_replace2: "
2515 				    "tmpent->start=%#"PRIxVADDR
2516 				    ", tmpent->end=%#"PRIxVADDR
2517 				    ", end=%#"PRIxVADDR,
2518 				    tmpent->start, tmpent->end, end);
2519 			}
2520 			cur = tmpent->end;
2521 			if (tmpent->next) {
2522 				if (tmpent->next->prev != tmpent)
2523 					panic("uvm_map_replace3");
2524 			} else {
2525 				if (newents->prev != tmpent)
2526 					panic("uvm_map_replace4");
2527 			}
2528 			tmpent = tmpent->next;
2529 		}
2530 		if (nent != nnewents)
2531 			panic("uvm_map_replace5");
2532 		if (sz != nsize)
2533 			panic("uvm_map_replace6");
2534 	}
2535 #endif
2536 
2537 	/*
2538 	 * map entry is a valid blank!   replace it.   (this does all the
2539 	 * work of map entry link/unlink...).
2540 	 */
2541 
2542 	if (newents) {
2543 		last = newents->prev;
2544 
2545 		/* critical: flush stale hints out of map */
2546 		SAVE_HINT(map, map->hint, newents);
2547 		if (map->first_free == oldent)
2548 			map->first_free = last;
2549 
2550 		last->next = oldent->next;
2551 		last->next->prev = last;
2552 
2553 		/* Fix RB tree */
2554 		uvm_rb_remove(map, oldent);
2555 
2556 		newents->prev = oldent->prev;
2557 		newents->prev->next = newents;
2558 		map->nentries = map->nentries + (nnewents - 1);
2559 
2560 		/* Fixup the RB tree */
2561 		{
2562 			int i;
2563 			struct vm_map_entry *tmp;
2564 
2565 			tmp = newents;
2566 			for (i = 0; i < nnewents && tmp; i++) {
2567 				uvm_rb_insert(map, tmp);
2568 				tmp = tmp->next;
2569 			}
2570 		}
2571 	} else {
2572 		/* NULL list of new entries: just remove the old one */
2573 		clear_hints(map, oldent);
2574 		uvm_map_entry_unlink(map, oldent);
2575 	}
2576 	map->size -= end - start - nsize;
2577 
2578 	uvm_map_check(map, "map_replace leave");
2579 
2580 	/*
2581 	 * now we can free the old blank entry and return.
2582 	 */
2583 
2584 	*oldentryp = oldent;
2585 	return (true);
2586 }
2587 
2588 /*
2589  * uvm_map_extract: extract a mapping from a map and put it somewhere
2590  *	(maybe removing the old mapping)
2591  *
2592  * => maps should be unlocked (we will write lock them)
2593  * => returns 0 on success, error code otherwise
2594  * => start must be page aligned
2595  * => len must be page sized
2596  * => flags:
2597  *      UVM_EXTRACT_REMOVE: remove mappings from srcmap
2598  *      UVM_EXTRACT_CONTIG: abort if unmapped area (advisory only)
2599  *      UVM_EXTRACT_QREF: for a temporary extraction do quick obj refs
2600  *      UVM_EXTRACT_FIXPROT: set prot to maxprot as we go
2601  *      UVM_EXTRACT_PROT_ALL: set prot to UVM_PROT_ALL as we go
2602  *    >>>NOTE: if you set REMOVE, you are not allowed to use CONTIG or QREF!<<<
2603  *    >>>NOTE: QREF's must be unmapped via the QREF path, thus should only
2604  *             be used from within the kernel in a kernel level map <<<
2605  */
2606 
2607 int
2608 uvm_map_extract(struct vm_map *srcmap, vaddr_t start, vsize_t len,
2609     struct vm_map *dstmap, vaddr_t *dstaddrp, int flags)
2610 {
2611 	vaddr_t dstaddr, end, newend, oldoffset, fudge, orig_fudge;
2612 	struct vm_map_entry *chain, *endchain, *entry, *orig_entry, *newentry,
2613 	    *deadentry, *oldentry;
2614 	struct vm_map_entry *resentry = NULL; /* a dummy reservation entry */
2615 	vsize_t elen __unused;
2616 	int nchain, error, copy_ok;
2617 	vsize_t nsize;
2618 	UVMHIST_FUNC("uvm_map_extract"); UVMHIST_CALLED(maphist);
2619 
2620 	UVMHIST_LOG(maphist,"(srcmap=%#jx,start=%#jx, len=%#jx",
2621 	    (uintptr_t)srcmap, start, len, 0);
2622 	UVMHIST_LOG(maphist," ...,dstmap=%#jx, flags=%#jx)",
2623 	    (uintptr_t)dstmap, flags, 0, 0);
2624 
2625 	/*
2626 	 * step 0: sanity check: start must be on a page boundary, length
2627 	 * must be page sized.  can't ask for CONTIG/QREF if you asked for
2628 	 * REMOVE.
2629 	 */
2630 
2631 	KASSERT((start & PAGE_MASK) == 0 && (len & PAGE_MASK) == 0);
2632 	KASSERT((flags & UVM_EXTRACT_REMOVE) == 0 ||
2633 		(flags & (UVM_EXTRACT_CONTIG|UVM_EXTRACT_QREF)) == 0);
2634 
2635 	/*
2636 	 * step 1: reserve space in the target map for the extracted area
2637 	 */
2638 
2639 	if ((flags & UVM_EXTRACT_RESERVED) == 0) {
2640 		dstaddr = vm_map_min(dstmap);
2641 		if (!uvm_map_reserve(dstmap, len, start,
2642 		    atop(start) & uvmexp.colormask, &dstaddr,
2643 		    UVM_FLAG_COLORMATCH))
2644 			return (ENOMEM);
2645 		KASSERT((atop(start ^ dstaddr) & uvmexp.colormask) == 0);
2646 		*dstaddrp = dstaddr;	/* pass address back to caller */
2647 		UVMHIST_LOG(maphist, "  dstaddr=%#jx", dstaddr,0,0,0);
2648 	} else {
2649 		dstaddr = *dstaddrp;
2650 	}
2651 
2652 	/*
2653 	 * step 2: setup for the extraction process loop by init'ing the
2654 	 * map entry chain, locking src map, and looking up the first useful
2655 	 * entry in the map.
2656 	 */
2657 
2658 	end = start + len;
2659 	newend = dstaddr + len;
2660 	chain = endchain = NULL;
2661 	nchain = 0;
2662 	nsize = 0;
2663 	vm_map_lock(srcmap);
2664 
2665 	if (uvm_map_lookup_entry(srcmap, start, &entry)) {
2666 
2667 		/* "start" is within an entry */
2668 		if (flags & UVM_EXTRACT_QREF) {
2669 
2670 			/*
2671 			 * for quick references we don't clip the entry, so
2672 			 * the entry may map space "before" the starting
2673 			 * virtual address... this is the "fudge" factor
2674 			 * (which can be non-zero only the first time
2675 			 * through the "while" loop in step 3).
2676 			 */
2677 
2678 			fudge = start - entry->start;
2679 		} else {
2680 
2681 			/*
2682 			 * normal reference: we clip the map to fit (thus
2683 			 * fudge is zero)
2684 			 */
2685 
2686 			UVM_MAP_CLIP_START(srcmap, entry, start);
2687 			SAVE_HINT(srcmap, srcmap->hint, entry->prev);
2688 			fudge = 0;
2689 		}
2690 	} else {
2691 
2692 		/* "start" is not within an entry ... skip to next entry */
2693 		if (flags & UVM_EXTRACT_CONTIG) {
2694 			error = EINVAL;
2695 			goto bad;    /* definite hole here ... */
2696 		}
2697 
2698 		entry = entry->next;
2699 		fudge = 0;
2700 	}
2701 
2702 	/* save values from srcmap for step 6 */
2703 	orig_entry = entry;
2704 	orig_fudge = fudge;
2705 
2706 	/*
2707 	 * step 3: now start looping through the map entries, extracting
2708 	 * as we go.
2709 	 */
2710 
2711 	while (entry->start < end && entry != &srcmap->header) {
2712 
2713 		/* if we are not doing a quick reference, clip it */
2714 		if ((flags & UVM_EXTRACT_QREF) == 0)
2715 			UVM_MAP_CLIP_END(srcmap, entry, end);
2716 
2717 		/* clear needs_copy (allow chunking) */
2718 		if (UVM_ET_ISNEEDSCOPY(entry)) {
2719 			amap_copy(srcmap, entry,
2720 			    AMAP_COPY_NOWAIT|AMAP_COPY_NOMERGE, start, end);
2721 			if (UVM_ET_ISNEEDSCOPY(entry)) {  /* failed? */
2722 				error = ENOMEM;
2723 				goto bad;
2724 			}
2725 
2726 			/* amap_copy could clip (during chunk)!  update fudge */
2727 			if (fudge) {
2728 				fudge = start - entry->start;
2729 				orig_fudge = fudge;
2730 			}
2731 		}
2732 
2733 		/* calculate the offset of this from "start" */
2734 		oldoffset = (entry->start + fudge) - start;
2735 
2736 		/* allocate a new map entry */
2737 		newentry = uvm_mapent_alloc(dstmap, 0);
2738 		if (newentry == NULL) {
2739 			error = ENOMEM;
2740 			goto bad;
2741 		}
2742 
2743 		/* set up new map entry */
2744 		newentry->next = NULL;
2745 		newentry->prev = endchain;
2746 		newentry->start = dstaddr + oldoffset;
2747 		newentry->end =
2748 		    newentry->start + (entry->end - (entry->start + fudge));
2749 		if (newentry->end > newend || newentry->end < newentry->start)
2750 			newentry->end = newend;
2751 		newentry->object.uvm_obj = entry->object.uvm_obj;
2752 		if (newentry->object.uvm_obj) {
2753 			if (newentry->object.uvm_obj->pgops->pgo_reference)
2754 				newentry->object.uvm_obj->pgops->
2755 				    pgo_reference(newentry->object.uvm_obj);
2756 			newentry->offset = entry->offset + fudge;
2757 		} else {
2758 			newentry->offset = 0;
2759 		}
2760 		newentry->etype = entry->etype;
2761 		if (flags & UVM_EXTRACT_PROT_ALL) {
2762 			newentry->protection = newentry->max_protection =
2763 			    UVM_PROT_ALL;
2764 		} else {
2765 			newentry->protection = (flags & UVM_EXTRACT_FIXPROT) ?
2766 			    entry->max_protection : entry->protection;
2767 			newentry->max_protection = entry->max_protection;
2768 		}
2769 		newentry->inheritance = entry->inheritance;
2770 		newentry->wired_count = 0;
2771 		newentry->aref.ar_amap = entry->aref.ar_amap;
2772 		if (newentry->aref.ar_amap) {
2773 			newentry->aref.ar_pageoff =
2774 			    entry->aref.ar_pageoff + (fudge >> PAGE_SHIFT);
2775 			uvm_map_reference_amap(newentry, AMAP_SHARED |
2776 			    ((flags & UVM_EXTRACT_QREF) ? AMAP_REFALL : 0));
2777 		} else {
2778 			newentry->aref.ar_pageoff = 0;
2779 		}
2780 		newentry->advice = entry->advice;
2781 		if ((flags & UVM_EXTRACT_QREF) != 0) {
2782 			newentry->flags |= UVM_MAP_NOMERGE;
2783 		}
2784 
2785 		/* now link it on the chain */
2786 		nchain++;
2787 		nsize += newentry->end - newentry->start;
2788 		if (endchain == NULL) {
2789 			chain = endchain = newentry;
2790 		} else {
2791 			endchain->next = newentry;
2792 			endchain = newentry;
2793 		}
2794 
2795 		/* end of 'while' loop! */
2796 		if ((flags & UVM_EXTRACT_CONTIG) && entry->end < end &&
2797 		    (entry->next == &srcmap->header ||
2798 		    entry->next->start != entry->end)) {
2799 			error = EINVAL;
2800 			goto bad;
2801 		}
2802 		entry = entry->next;
2803 		fudge = 0;
2804 	}
2805 
2806 	/*
2807 	 * step 4: close off chain (in format expected by uvm_map_replace)
2808 	 */
2809 
2810 	if (chain)
2811 		chain->prev = endchain;
2812 
2813 	/*
2814 	 * step 5: attempt to lock the dest map so we can pmap_copy.
2815 	 * note usage of copy_ok:
2816 	 *   1 => dstmap locked, pmap_copy ok, and we "replace" here (step 5)
2817 	 *   0 => dstmap unlocked, NO pmap_copy, and we will "replace" in step 7
2818 	 */
2819 
2820 	if (srcmap == dstmap || vm_map_lock_try(dstmap) == true) {
2821 		copy_ok = 1;
2822 		if (!uvm_map_replace(dstmap, dstaddr, dstaddr+len, chain,
2823 		    nchain, nsize, &resentry)) {
2824 			if (srcmap != dstmap)
2825 				vm_map_unlock(dstmap);
2826 			error = EIO;
2827 			goto bad;
2828 		}
2829 	} else {
2830 		copy_ok = 0;
2831 		/* replace defered until step 7 */
2832 	}
2833 
2834 	/*
2835 	 * step 6: traverse the srcmap a second time to do the following:
2836 	 *  - if we got a lock on the dstmap do pmap_copy
2837 	 *  - if UVM_EXTRACT_REMOVE remove the entries
2838 	 * we make use of orig_entry and orig_fudge (saved in step 2)
2839 	 */
2840 
2841 	if (copy_ok || (flags & UVM_EXTRACT_REMOVE)) {
2842 
2843 		/* purge possible stale hints from srcmap */
2844 		if (flags & UVM_EXTRACT_REMOVE) {
2845 			SAVE_HINT(srcmap, srcmap->hint, orig_entry->prev);
2846 			if (srcmap->first_free != &srcmap->header &&
2847 			    srcmap->first_free->start >= start)
2848 				srcmap->first_free = orig_entry->prev;
2849 		}
2850 
2851 		entry = orig_entry;
2852 		fudge = orig_fudge;
2853 		deadentry = NULL;	/* for UVM_EXTRACT_REMOVE */
2854 
2855 		while (entry->start < end && entry != &srcmap->header) {
2856 			if (copy_ok) {
2857 				oldoffset = (entry->start + fudge) - start;
2858 				elen = MIN(end, entry->end) -
2859 				    (entry->start + fudge);
2860 				pmap_copy(dstmap->pmap, srcmap->pmap,
2861 				    dstaddr + oldoffset, elen,
2862 				    entry->start + fudge);
2863 			}
2864 
2865 			/* we advance "entry" in the following if statement */
2866 			if (flags & UVM_EXTRACT_REMOVE) {
2867 				uvm_map_lock_entry(entry);
2868 				pmap_remove(srcmap->pmap, entry->start,
2869 						entry->end);
2870 				uvm_map_unlock_entry(entry);
2871 				oldentry = entry;	/* save entry */
2872 				entry = entry->next;	/* advance */
2873 				uvm_map_entry_unlink(srcmap, oldentry);
2874 							/* add to dead list */
2875 				oldentry->next = deadentry;
2876 				deadentry = oldentry;
2877 			} else {
2878 				entry = entry->next;		/* advance */
2879 			}
2880 
2881 			/* end of 'while' loop */
2882 			fudge = 0;
2883 		}
2884 		pmap_update(srcmap->pmap);
2885 
2886 		/*
2887 		 * unlock dstmap.  we will dispose of deadentry in
2888 		 * step 7 if needed
2889 		 */
2890 
2891 		if (copy_ok && srcmap != dstmap)
2892 			vm_map_unlock(dstmap);
2893 
2894 	} else {
2895 		deadentry = NULL;
2896 	}
2897 
2898 	/*
2899 	 * step 7: we are done with the source map, unlock.   if copy_ok
2900 	 * is 0 then we have not replaced the dummy mapping in dstmap yet
2901 	 * and we need to do so now.
2902 	 */
2903 
2904 	vm_map_unlock(srcmap);
2905 	if ((flags & UVM_EXTRACT_REMOVE) && deadentry)
2906 		uvm_unmap_detach(deadentry, 0);   /* dispose of old entries */
2907 
2908 	/* now do the replacement if we didn't do it in step 5 */
2909 	if (copy_ok == 0) {
2910 		vm_map_lock(dstmap);
2911 		error = uvm_map_replace(dstmap, dstaddr, dstaddr+len, chain,
2912 		    nchain, nsize, &resentry);
2913 		vm_map_unlock(dstmap);
2914 
2915 		if (error == false) {
2916 			error = EIO;
2917 			goto bad2;
2918 		}
2919 	}
2920 
2921 	if (resentry != NULL)
2922 		uvm_mapent_free(resentry);
2923 
2924 	return (0);
2925 
2926 	/*
2927 	 * bad: failure recovery
2928 	 */
2929 bad:
2930 	vm_map_unlock(srcmap);
2931 bad2:			/* src already unlocked */
2932 	if (chain)
2933 		uvm_unmap_detach(chain,
2934 		    (flags & UVM_EXTRACT_QREF) ? AMAP_REFALL : 0);
2935 
2936 	if (resentry != NULL)
2937 		uvm_mapent_free(resentry);
2938 
2939 	if ((flags & UVM_EXTRACT_RESERVED) == 0) {
2940 		uvm_unmap(dstmap, dstaddr, dstaddr+len);   /* ??? */
2941 	}
2942 	return (error);
2943 }
2944 
2945 /* end of extraction functions */
2946 
2947 /*
2948  * uvm_map_submap: punch down part of a map into a submap
2949  *
2950  * => only the kernel_map is allowed to be submapped
2951  * => the purpose of submapping is to break up the locking granularity
2952  *	of a larger map
2953  * => the range specified must have been mapped previously with a uvm_map()
2954  *	call [with uobj==NULL] to create a blank map entry in the main map.
2955  *	[And it had better still be blank!]
2956  * => maps which contain submaps should never be copied or forked.
2957  * => to remove a submap, use uvm_unmap() on the main map
2958  *	and then uvm_map_deallocate() the submap.
2959  * => main map must be unlocked.
2960  * => submap must have been init'd and have a zero reference count.
2961  *	[need not be locked as we don't actually reference it]
2962  */
2963 
2964 int
2965 uvm_map_submap(struct vm_map *map, vaddr_t start, vaddr_t end,
2966     struct vm_map *submap)
2967 {
2968 	struct vm_map_entry *entry;
2969 	int error;
2970 
2971 	vm_map_lock(map);
2972 	VM_MAP_RANGE_CHECK(map, start, end);
2973 
2974 	if (uvm_map_lookup_entry(map, start, &entry)) {
2975 		UVM_MAP_CLIP_START(map, entry, start);
2976 		UVM_MAP_CLIP_END(map, entry, end);	/* to be safe */
2977 	} else {
2978 		entry = NULL;
2979 	}
2980 
2981 	if (entry != NULL &&
2982 	    entry->start == start && entry->end == end &&
2983 	    entry->object.uvm_obj == NULL && entry->aref.ar_amap == NULL &&
2984 	    !UVM_ET_ISCOPYONWRITE(entry) && !UVM_ET_ISNEEDSCOPY(entry)) {
2985 		entry->etype |= UVM_ET_SUBMAP;
2986 		entry->object.sub_map = submap;
2987 		entry->offset = 0;
2988 		uvm_map_reference(submap);
2989 		error = 0;
2990 	} else {
2991 		error = EINVAL;
2992 	}
2993 	vm_map_unlock(map);
2994 
2995 	return error;
2996 }
2997 
2998 /*
2999  * uvm_map_protect_user: change map protection on behalf of the user.
3000  * Enforces PAX settings as necessary.
3001  */
3002 int
3003 uvm_map_protect_user(struct lwp *l, vaddr_t start, vaddr_t end,
3004     vm_prot_t new_prot)
3005 {
3006 	int error;
3007 
3008 	if ((error = PAX_MPROTECT_VALIDATE(l, new_prot)))
3009 		return error;
3010 
3011 	return uvm_map_protect(&l->l_proc->p_vmspace->vm_map, start, end,
3012 	    new_prot, false);
3013 }
3014 
3015 
3016 /*
3017  * uvm_map_protect: change map protection
3018  *
3019  * => set_max means set max_protection.
3020  * => map must be unlocked.
3021  */
3022 
3023 #define MASK(entry)	(UVM_ET_ISCOPYONWRITE(entry) ? \
3024 			 ~VM_PROT_WRITE : VM_PROT_ALL)
3025 
3026 int
3027 uvm_map_protect(struct vm_map *map, vaddr_t start, vaddr_t end,
3028     vm_prot_t new_prot, bool set_max)
3029 {
3030 	struct vm_map_entry *current, *entry;
3031 	int error = 0;
3032 	UVMHIST_FUNC("uvm_map_protect"); UVMHIST_CALLED(maphist);
3033 	UVMHIST_LOG(maphist,"(map=%#jx,start=%#jx,end=%#jx,new_prot=%#jx)",
3034 	    (uintptr_t)map, start, end, new_prot);
3035 
3036 	vm_map_lock(map);
3037 	VM_MAP_RANGE_CHECK(map, start, end);
3038 	if (uvm_map_lookup_entry(map, start, &entry)) {
3039 		UVM_MAP_CLIP_START(map, entry, start);
3040 	} else {
3041 		entry = entry->next;
3042 	}
3043 
3044 	/*
3045 	 * make a first pass to check for protection violations.
3046 	 */
3047 
3048 	current = entry;
3049 	while ((current != &map->header) && (current->start < end)) {
3050 		if (UVM_ET_ISSUBMAP(current)) {
3051 			error = EINVAL;
3052 			goto out;
3053 		}
3054 		if ((new_prot & current->max_protection) != new_prot) {
3055 			error = EACCES;
3056 			goto out;
3057 		}
3058 		/*
3059 		 * Don't allow VM_PROT_EXECUTE to be set on entries that
3060 		 * point to vnodes that are associated with a NOEXEC file
3061 		 * system.
3062 		 */
3063 		if (UVM_ET_ISOBJ(current) &&
3064 		    UVM_OBJ_IS_VNODE(current->object.uvm_obj)) {
3065 			struct vnode *vp =
3066 			    (struct vnode *) current->object.uvm_obj;
3067 
3068 			if ((new_prot & VM_PROT_EXECUTE) != 0 &&
3069 			    (vp->v_mount->mnt_flag & MNT_NOEXEC) != 0) {
3070 				error = EACCES;
3071 				goto out;
3072 			}
3073 		}
3074 
3075 		current = current->next;
3076 	}
3077 
3078 	/* go back and fix up protections (no need to clip this time). */
3079 
3080 	current = entry;
3081 	while ((current != &map->header) && (current->start < end)) {
3082 		vm_prot_t old_prot;
3083 
3084 		UVM_MAP_CLIP_END(map, current, end);
3085 		old_prot = current->protection;
3086 		if (set_max)
3087 			current->protection =
3088 			    (current->max_protection = new_prot) & old_prot;
3089 		else
3090 			current->protection = new_prot;
3091 
3092 		/*
3093 		 * update physical map if necessary.  worry about copy-on-write
3094 		 * here -- CHECK THIS XXX
3095 		 */
3096 
3097 		if (current->protection != old_prot) {
3098 			/* update pmap! */
3099 			uvm_map_lock_entry(current);
3100 			pmap_protect(map->pmap, current->start, current->end,
3101 			    current->protection & MASK(current));
3102 			uvm_map_unlock_entry(current);
3103 
3104 			/*
3105 			 * If this entry points at a vnode, and the
3106 			 * protection includes VM_PROT_EXECUTE, mark
3107 			 * the vnode as VEXECMAP.
3108 			 */
3109 			if (UVM_ET_ISOBJ(current)) {
3110 				struct uvm_object *uobj =
3111 				    current->object.uvm_obj;
3112 
3113 				if (UVM_OBJ_IS_VNODE(uobj) &&
3114 				    (current->protection & VM_PROT_EXECUTE)) {
3115 					vn_markexec((struct vnode *) uobj);
3116 				}
3117 			}
3118 		}
3119 
3120 		/*
3121 		 * If the map is configured to lock any future mappings,
3122 		 * wire this entry now if the old protection was VM_PROT_NONE
3123 		 * and the new protection is not VM_PROT_NONE.
3124 		 */
3125 
3126 		if ((map->flags & VM_MAP_WIREFUTURE) != 0 &&
3127 		    VM_MAPENT_ISWIRED(current) == 0 &&
3128 		    old_prot == VM_PROT_NONE &&
3129 		    new_prot != VM_PROT_NONE) {
3130 			if (uvm_map_pageable(map, current->start,
3131 			    current->end, false,
3132 			    UVM_LK_ENTER|UVM_LK_EXIT) != 0) {
3133 
3134 				/*
3135 				 * If locking the entry fails, remember the
3136 				 * error if it's the first one.  Note we
3137 				 * still continue setting the protection in
3138 				 * the map, but will return the error
3139 				 * condition regardless.
3140 				 *
3141 				 * XXX Ignore what the actual error is,
3142 				 * XXX just call it a resource shortage
3143 				 * XXX so that it doesn't get confused
3144 				 * XXX what uvm_map_protect() itself would
3145 				 * XXX normally return.
3146 				 */
3147 
3148 				error = ENOMEM;
3149 			}
3150 		}
3151 		current = current->next;
3152 	}
3153 	pmap_update(map->pmap);
3154 
3155  out:
3156 	vm_map_unlock(map);
3157 
3158 	UVMHIST_LOG(maphist, "<- done, error=%jd",error,0,0,0);
3159 	return error;
3160 }
3161 
3162 #undef  MASK
3163 
3164 /*
3165  * uvm_map_inherit: set inheritance code for range of addrs in map.
3166  *
3167  * => map must be unlocked
3168  * => note that the inherit code is used during a "fork".  see fork
3169  *	code for details.
3170  */
3171 
3172 int
3173 uvm_map_inherit(struct vm_map *map, vaddr_t start, vaddr_t end,
3174     vm_inherit_t new_inheritance)
3175 {
3176 	struct vm_map_entry *entry, *temp_entry;
3177 	UVMHIST_FUNC("uvm_map_inherit"); UVMHIST_CALLED(maphist);
3178 	UVMHIST_LOG(maphist,"(map=%#jx,start=%#jx,end=%#jx,new_inh=%#jx)",
3179 	    (uintptr_t)map, start, end, new_inheritance);
3180 
3181 	switch (new_inheritance) {
3182 	case MAP_INHERIT_NONE:
3183 	case MAP_INHERIT_COPY:
3184 	case MAP_INHERIT_SHARE:
3185 	case MAP_INHERIT_ZERO:
3186 		break;
3187 	default:
3188 		UVMHIST_LOG(maphist,"<- done (INVALID ARG)",0,0,0,0);
3189 		return EINVAL;
3190 	}
3191 
3192 	vm_map_lock(map);
3193 	VM_MAP_RANGE_CHECK(map, start, end);
3194 	if (uvm_map_lookup_entry(map, start, &temp_entry)) {
3195 		entry = temp_entry;
3196 		UVM_MAP_CLIP_START(map, entry, start);
3197 	}  else {
3198 		entry = temp_entry->next;
3199 	}
3200 	while ((entry != &map->header) && (entry->start < end)) {
3201 		UVM_MAP_CLIP_END(map, entry, end);
3202 		entry->inheritance = new_inheritance;
3203 		entry = entry->next;
3204 	}
3205 	vm_map_unlock(map);
3206 	UVMHIST_LOG(maphist,"<- done (OK)",0,0,0,0);
3207 	return 0;
3208 }
3209 
3210 /*
3211  * uvm_map_advice: set advice code for range of addrs in map.
3212  *
3213  * => map must be unlocked
3214  */
3215 
3216 int
3217 uvm_map_advice(struct vm_map *map, vaddr_t start, vaddr_t end, int new_advice)
3218 {
3219 	struct vm_map_entry *entry, *temp_entry;
3220 	UVMHIST_FUNC("uvm_map_advice"); UVMHIST_CALLED(maphist);
3221 	UVMHIST_LOG(maphist,"(map=%#jx,start=%#jx,end=%#jx,new_adv=%#jx)",
3222 	    (uintptr_t)map, start, end, new_advice);
3223 
3224 	vm_map_lock(map);
3225 	VM_MAP_RANGE_CHECK(map, start, end);
3226 	if (uvm_map_lookup_entry(map, start, &temp_entry)) {
3227 		entry = temp_entry;
3228 		UVM_MAP_CLIP_START(map, entry, start);
3229 	} else {
3230 		entry = temp_entry->next;
3231 	}
3232 
3233 	/*
3234 	 * XXXJRT: disallow holes?
3235 	 */
3236 
3237 	while ((entry != &map->header) && (entry->start < end)) {
3238 		UVM_MAP_CLIP_END(map, entry, end);
3239 
3240 		switch (new_advice) {
3241 		case MADV_NORMAL:
3242 		case MADV_RANDOM:
3243 		case MADV_SEQUENTIAL:
3244 			/* nothing special here */
3245 			break;
3246 
3247 		default:
3248 			vm_map_unlock(map);
3249 			UVMHIST_LOG(maphist,"<- done (INVALID ARG)",0,0,0,0);
3250 			return EINVAL;
3251 		}
3252 		entry->advice = new_advice;
3253 		entry = entry->next;
3254 	}
3255 
3256 	vm_map_unlock(map);
3257 	UVMHIST_LOG(maphist,"<- done (OK)",0,0,0,0);
3258 	return 0;
3259 }
3260 
3261 /*
3262  * uvm_map_willneed: apply MADV_WILLNEED
3263  */
3264 
3265 int
3266 uvm_map_willneed(struct vm_map *map, vaddr_t start, vaddr_t end)
3267 {
3268 	struct vm_map_entry *entry;
3269 	UVMHIST_FUNC("uvm_map_willneed"); UVMHIST_CALLED(maphist);
3270 	UVMHIST_LOG(maphist,"(map=%#jx,start=%#jx,end=%#jx)",
3271 	    (uintptr_t)map, start, end, 0);
3272 
3273 	vm_map_lock_read(map);
3274 	VM_MAP_RANGE_CHECK(map, start, end);
3275 	if (!uvm_map_lookup_entry(map, start, &entry)) {
3276 		entry = entry->next;
3277 	}
3278 	while (entry->start < end) {
3279 		struct vm_amap * const amap = entry->aref.ar_amap;
3280 		struct uvm_object * const uobj = entry->object.uvm_obj;
3281 
3282 		KASSERT(entry != &map->header);
3283 		KASSERT(start < entry->end);
3284 		/*
3285 		 * For now, we handle only the easy but commonly-requested case.
3286 		 * ie. start prefetching of backing uobj pages.
3287 		 *
3288 		 * XXX It might be useful to pmap_enter() the already-in-core
3289 		 * pages by inventing a "weak" mode for uvm_fault() which would
3290 		 * only do the PGO_LOCKED pgo_get().
3291 		 */
3292 		if (UVM_ET_ISOBJ(entry) && amap == NULL && uobj != NULL) {
3293 			off_t offset;
3294 			off_t size;
3295 
3296 			offset = entry->offset;
3297 			if (start < entry->start) {
3298 				offset += entry->start - start;
3299 			}
3300 			size = entry->offset + (entry->end - entry->start);
3301 			if (entry->end < end) {
3302 				size -= end - entry->end;
3303 			}
3304 			uvm_readahead(uobj, offset, size);
3305 		}
3306 		entry = entry->next;
3307 	}
3308 	vm_map_unlock_read(map);
3309 	UVMHIST_LOG(maphist,"<- done (OK)",0,0,0,0);
3310 	return 0;
3311 }
3312 
3313 /*
3314  * uvm_map_pageable: sets the pageability of a range in a map.
3315  *
3316  * => wires map entries.  should not be used for transient page locking.
3317  *	for that, use uvm_fault_wire()/uvm_fault_unwire() (see uvm_vslock()).
3318  * => regions specified as not pageable require lock-down (wired) memory
3319  *	and page tables.
3320  * => map must never be read-locked
3321  * => if islocked is true, map is already write-locked
3322  * => we always unlock the map, since we must downgrade to a read-lock
3323  *	to call uvm_fault_wire()
3324  * => XXXCDC: check this and try and clean it up.
3325  */
3326 
3327 int
3328 uvm_map_pageable(struct vm_map *map, vaddr_t start, vaddr_t end,
3329     bool new_pageable, int lockflags)
3330 {
3331 	struct vm_map_entry *entry, *start_entry, *failed_entry;
3332 	int rv;
3333 #ifdef DIAGNOSTIC
3334 	u_int timestamp_save;
3335 #endif
3336 	UVMHIST_FUNC("uvm_map_pageable"); UVMHIST_CALLED(maphist);
3337 	UVMHIST_LOG(maphist,"(map=%#jx,start=%#jx,end=%#jx,new_pageable=%ju)",
3338 	    (uintptr_t)map, start, end, new_pageable);
3339 	KASSERT(map->flags & VM_MAP_PAGEABLE);
3340 
3341 	if ((lockflags & UVM_LK_ENTER) == 0)
3342 		vm_map_lock(map);
3343 	VM_MAP_RANGE_CHECK(map, start, end);
3344 
3345 	/*
3346 	 * only one pageability change may take place at one time, since
3347 	 * uvm_fault_wire assumes it will be called only once for each
3348 	 * wiring/unwiring.  therefore, we have to make sure we're actually
3349 	 * changing the pageability for the entire region.  we do so before
3350 	 * making any changes.
3351 	 */
3352 
3353 	if (uvm_map_lookup_entry(map, start, &start_entry) == false) {
3354 		if ((lockflags & UVM_LK_EXIT) == 0)
3355 			vm_map_unlock(map);
3356 
3357 		UVMHIST_LOG(maphist,"<- done (fault)",0,0,0,0);
3358 		return EFAULT;
3359 	}
3360 	entry = start_entry;
3361 
3362 	if (start == end) {		/* nothing required */
3363 		if ((lockflags & UVM_LK_EXIT) == 0)
3364 			vm_map_unlock(map);
3365 
3366 		UVMHIST_LOG(maphist,"<- done (nothing)",0,0,0,0);
3367 		return 0;
3368 	}
3369 
3370 	/*
3371 	 * handle wiring and unwiring separately.
3372 	 */
3373 
3374 	if (new_pageable) {		/* unwire */
3375 		UVM_MAP_CLIP_START(map, entry, start);
3376 
3377 		/*
3378 		 * unwiring.  first ensure that the range to be unwired is
3379 		 * really wired down and that there are no holes.
3380 		 */
3381 
3382 		while ((entry != &map->header) && (entry->start < end)) {
3383 			if (entry->wired_count == 0 ||
3384 			    (entry->end < end &&
3385 			     (entry->next == &map->header ||
3386 			      entry->next->start > entry->end))) {
3387 				if ((lockflags & UVM_LK_EXIT) == 0)
3388 					vm_map_unlock(map);
3389 				UVMHIST_LOG(maphist, "<- done (INVAL)",0,0,0,0);
3390 				return EINVAL;
3391 			}
3392 			entry = entry->next;
3393 		}
3394 
3395 		/*
3396 		 * POSIX 1003.1b - a single munlock call unlocks a region,
3397 		 * regardless of the number of mlock calls made on that
3398 		 * region.
3399 		 */
3400 
3401 		entry = start_entry;
3402 		while ((entry != &map->header) && (entry->start < end)) {
3403 			UVM_MAP_CLIP_END(map, entry, end);
3404 			if (VM_MAPENT_ISWIRED(entry))
3405 				uvm_map_entry_unwire(map, entry);
3406 			entry = entry->next;
3407 		}
3408 		if ((lockflags & UVM_LK_EXIT) == 0)
3409 			vm_map_unlock(map);
3410 		UVMHIST_LOG(maphist,"<- done (OK UNWIRE)",0,0,0,0);
3411 		return 0;
3412 	}
3413 
3414 	/*
3415 	 * wire case: in two passes [XXXCDC: ugly block of code here]
3416 	 *
3417 	 * 1: holding the write lock, we create any anonymous maps that need
3418 	 *    to be created.  then we clip each map entry to the region to
3419 	 *    be wired and increment its wiring count.
3420 	 *
3421 	 * 2: we downgrade to a read lock, and call uvm_fault_wire to fault
3422 	 *    in the pages for any newly wired area (wired_count == 1).
3423 	 *
3424 	 *    downgrading to a read lock for uvm_fault_wire avoids a possible
3425 	 *    deadlock with another thread that may have faulted on one of
3426 	 *    the pages to be wired (it would mark the page busy, blocking
3427 	 *    us, then in turn block on the map lock that we hold).  because
3428 	 *    of problems in the recursive lock package, we cannot upgrade
3429 	 *    to a write lock in vm_map_lookup.  thus, any actions that
3430 	 *    require the write lock must be done beforehand.  because we
3431 	 *    keep the read lock on the map, the copy-on-write status of the
3432 	 *    entries we modify here cannot change.
3433 	 */
3434 
3435 	while ((entry != &map->header) && (entry->start < end)) {
3436 		if (VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */
3437 
3438 			/*
3439 			 * perform actions of vm_map_lookup that need the
3440 			 * write lock on the map: create an anonymous map
3441 			 * for a copy-on-write region, or an anonymous map
3442 			 * for a zero-fill region.  (XXXCDC: submap case
3443 			 * ok?)
3444 			 */
3445 
3446 			if (!UVM_ET_ISSUBMAP(entry)) {  /* not submap */
3447 				if (UVM_ET_ISNEEDSCOPY(entry) &&
3448 				    ((entry->max_protection & VM_PROT_WRITE) ||
3449 				     (entry->object.uvm_obj == NULL))) {
3450 					amap_copy(map, entry, 0, start, end);
3451 					/* XXXCDC: wait OK? */
3452 				}
3453 			}
3454 		}
3455 		UVM_MAP_CLIP_START(map, entry, start);
3456 		UVM_MAP_CLIP_END(map, entry, end);
3457 		entry->wired_count++;
3458 
3459 		/*
3460 		 * Check for holes
3461 		 */
3462 
3463 		if (entry->protection == VM_PROT_NONE ||
3464 		    (entry->end < end &&
3465 		     (entry->next == &map->header ||
3466 		      entry->next->start > entry->end))) {
3467 
3468 			/*
3469 			 * found one.  amap creation actions do not need to
3470 			 * be undone, but the wired counts need to be restored.
3471 			 */
3472 
3473 			while (entry != &map->header && entry->end > start) {
3474 				entry->wired_count--;
3475 				entry = entry->prev;
3476 			}
3477 			if ((lockflags & UVM_LK_EXIT) == 0)
3478 				vm_map_unlock(map);
3479 			UVMHIST_LOG(maphist,"<- done (INVALID WIRE)",0,0,0,0);
3480 			return EINVAL;
3481 		}
3482 		entry = entry->next;
3483 	}
3484 
3485 	/*
3486 	 * Pass 2.
3487 	 */
3488 
3489 #ifdef DIAGNOSTIC
3490 	timestamp_save = map->timestamp;
3491 #endif
3492 	vm_map_busy(map);
3493 	vm_map_unlock(map);
3494 
3495 	rv = 0;
3496 	entry = start_entry;
3497 	while (entry != &map->header && entry->start < end) {
3498 		if (entry->wired_count == 1) {
3499 			rv = uvm_fault_wire(map, entry->start, entry->end,
3500 			    entry->max_protection, 1);
3501 			if (rv) {
3502 
3503 				/*
3504 				 * wiring failed.  break out of the loop.
3505 				 * we'll clean up the map below, once we
3506 				 * have a write lock again.
3507 				 */
3508 
3509 				break;
3510 			}
3511 		}
3512 		entry = entry->next;
3513 	}
3514 
3515 	if (rv) {	/* failed? */
3516 
3517 		/*
3518 		 * Get back to an exclusive (write) lock.
3519 		 */
3520 
3521 		vm_map_lock(map);
3522 		vm_map_unbusy(map);
3523 
3524 #ifdef DIAGNOSTIC
3525 		if (timestamp_save + 1 != map->timestamp)
3526 			panic("uvm_map_pageable: stale map");
3527 #endif
3528 
3529 		/*
3530 		 * first drop the wiring count on all the entries
3531 		 * which haven't actually been wired yet.
3532 		 */
3533 
3534 		failed_entry = entry;
3535 		while (entry != &map->header && entry->start < end) {
3536 			entry->wired_count--;
3537 			entry = entry->next;
3538 		}
3539 
3540 		/*
3541 		 * now, unwire all the entries that were successfully
3542 		 * wired above.
3543 		 */
3544 
3545 		entry = start_entry;
3546 		while (entry != failed_entry) {
3547 			entry->wired_count--;
3548 			if (VM_MAPENT_ISWIRED(entry) == 0)
3549 				uvm_map_entry_unwire(map, entry);
3550 			entry = entry->next;
3551 		}
3552 		if ((lockflags & UVM_LK_EXIT) == 0)
3553 			vm_map_unlock(map);
3554 		UVMHIST_LOG(maphist, "<- done (RV=%jd)", rv,0,0,0);
3555 		return (rv);
3556 	}
3557 
3558 	if ((lockflags & UVM_LK_EXIT) == 0) {
3559 		vm_map_unbusy(map);
3560 	} else {
3561 
3562 		/*
3563 		 * Get back to an exclusive (write) lock.
3564 		 */
3565 
3566 		vm_map_lock(map);
3567 		vm_map_unbusy(map);
3568 	}
3569 
3570 	UVMHIST_LOG(maphist,"<- done (OK WIRE)",0,0,0,0);
3571 	return 0;
3572 }
3573 
3574 /*
3575  * uvm_map_pageable_all: special case of uvm_map_pageable - affects
3576  * all mapped regions.
3577  *
3578  * => map must not be locked.
3579  * => if no flags are specified, all regions are unwired.
3580  * => XXXJRT: has some of the same problems as uvm_map_pageable() above.
3581  */
3582 
3583 int
3584 uvm_map_pageable_all(struct vm_map *map, int flags, vsize_t limit)
3585 {
3586 	struct vm_map_entry *entry, *failed_entry;
3587 	vsize_t size;
3588 	int rv;
3589 #ifdef DIAGNOSTIC
3590 	u_int timestamp_save;
3591 #endif
3592 	UVMHIST_FUNC("uvm_map_pageable_all"); UVMHIST_CALLED(maphist);
3593 	UVMHIST_LOG(maphist,"(map=%#jx,flags=%#jx)", (uintptr_t)map, flags,
3594 	    0, 0);
3595 
3596 	KASSERT(map->flags & VM_MAP_PAGEABLE);
3597 
3598 	vm_map_lock(map);
3599 
3600 	/*
3601 	 * handle wiring and unwiring separately.
3602 	 */
3603 
3604 	if (flags == 0) {			/* unwire */
3605 
3606 		/*
3607 		 * POSIX 1003.1b -- munlockall unlocks all regions,
3608 		 * regardless of how many times mlockall has been called.
3609 		 */
3610 
3611 		for (entry = map->header.next; entry != &map->header;
3612 		     entry = entry->next) {
3613 			if (VM_MAPENT_ISWIRED(entry))
3614 				uvm_map_entry_unwire(map, entry);
3615 		}
3616 		map->flags &= ~VM_MAP_WIREFUTURE;
3617 		vm_map_unlock(map);
3618 		UVMHIST_LOG(maphist,"<- done (OK UNWIRE)",0,0,0,0);
3619 		return 0;
3620 	}
3621 
3622 	if (flags & MCL_FUTURE) {
3623 
3624 		/*
3625 		 * must wire all future mappings; remember this.
3626 		 */
3627 
3628 		map->flags |= VM_MAP_WIREFUTURE;
3629 	}
3630 
3631 	if ((flags & MCL_CURRENT) == 0) {
3632 
3633 		/*
3634 		 * no more work to do!
3635 		 */
3636 
3637 		UVMHIST_LOG(maphist,"<- done (OK no wire)",0,0,0,0);
3638 		vm_map_unlock(map);
3639 		return 0;
3640 	}
3641 
3642 	/*
3643 	 * wire case: in three passes [XXXCDC: ugly block of code here]
3644 	 *
3645 	 * 1: holding the write lock, count all pages mapped by non-wired
3646 	 *    entries.  if this would cause us to go over our limit, we fail.
3647 	 *
3648 	 * 2: still holding the write lock, we create any anonymous maps that
3649 	 *    need to be created.  then we increment its wiring count.
3650 	 *
3651 	 * 3: we downgrade to a read lock, and call uvm_fault_wire to fault
3652 	 *    in the pages for any newly wired area (wired_count == 1).
3653 	 *
3654 	 *    downgrading to a read lock for uvm_fault_wire avoids a possible
3655 	 *    deadlock with another thread that may have faulted on one of
3656 	 *    the pages to be wired (it would mark the page busy, blocking
3657 	 *    us, then in turn block on the map lock that we hold).  because
3658 	 *    of problems in the recursive lock package, we cannot upgrade
3659 	 *    to a write lock in vm_map_lookup.  thus, any actions that
3660 	 *    require the write lock must be done beforehand.  because we
3661 	 *    keep the read lock on the map, the copy-on-write status of the
3662 	 *    entries we modify here cannot change.
3663 	 */
3664 
3665 	for (size = 0, entry = map->header.next; entry != &map->header;
3666 	     entry = entry->next) {
3667 		if (entry->protection != VM_PROT_NONE &&
3668 		    VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */
3669 			size += entry->end - entry->start;
3670 		}
3671 	}
3672 
3673 	if (atop(size) + uvmexp.wired > uvmexp.wiredmax) {
3674 		vm_map_unlock(map);
3675 		return ENOMEM;
3676 	}
3677 
3678 	if (limit != 0 &&
3679 	    (size + ptoa(pmap_wired_count(vm_map_pmap(map))) > limit)) {
3680 		vm_map_unlock(map);
3681 		return ENOMEM;
3682 	}
3683 
3684 	/*
3685 	 * Pass 2.
3686 	 */
3687 
3688 	for (entry = map->header.next; entry != &map->header;
3689 	     entry = entry->next) {
3690 		if (entry->protection == VM_PROT_NONE)
3691 			continue;
3692 		if (VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */
3693 
3694 			/*
3695 			 * perform actions of vm_map_lookup that need the
3696 			 * write lock on the map: create an anonymous map
3697 			 * for a copy-on-write region, or an anonymous map
3698 			 * for a zero-fill region.  (XXXCDC: submap case
3699 			 * ok?)
3700 			 */
3701 
3702 			if (!UVM_ET_ISSUBMAP(entry)) {	/* not submap */
3703 				if (UVM_ET_ISNEEDSCOPY(entry) &&
3704 				    ((entry->max_protection & VM_PROT_WRITE) ||
3705 				     (entry->object.uvm_obj == NULL))) {
3706 					amap_copy(map, entry, 0, entry->start,
3707 					    entry->end);
3708 					/* XXXCDC: wait OK? */
3709 				}
3710 			}
3711 		}
3712 		entry->wired_count++;
3713 	}
3714 
3715 	/*
3716 	 * Pass 3.
3717 	 */
3718 
3719 #ifdef DIAGNOSTIC
3720 	timestamp_save = map->timestamp;
3721 #endif
3722 	vm_map_busy(map);
3723 	vm_map_unlock(map);
3724 
3725 	rv = 0;
3726 	for (entry = map->header.next; entry != &map->header;
3727 	     entry = entry->next) {
3728 		if (entry->wired_count == 1) {
3729 			rv = uvm_fault_wire(map, entry->start, entry->end,
3730 			    entry->max_protection, 1);
3731 			if (rv) {
3732 
3733 				/*
3734 				 * wiring failed.  break out of the loop.
3735 				 * we'll clean up the map below, once we
3736 				 * have a write lock again.
3737 				 */
3738 
3739 				break;
3740 			}
3741 		}
3742 	}
3743 
3744 	if (rv) {
3745 
3746 		/*
3747 		 * Get back an exclusive (write) lock.
3748 		 */
3749 
3750 		vm_map_lock(map);
3751 		vm_map_unbusy(map);
3752 
3753 #ifdef DIAGNOSTIC
3754 		if (timestamp_save + 1 != map->timestamp)
3755 			panic("uvm_map_pageable_all: stale map");
3756 #endif
3757 
3758 		/*
3759 		 * first drop the wiring count on all the entries
3760 		 * which haven't actually been wired yet.
3761 		 *
3762 		 * Skip VM_PROT_NONE entries like we did above.
3763 		 */
3764 
3765 		failed_entry = entry;
3766 		for (/* nothing */; entry != &map->header;
3767 		     entry = entry->next) {
3768 			if (entry->protection == VM_PROT_NONE)
3769 				continue;
3770 			entry->wired_count--;
3771 		}
3772 
3773 		/*
3774 		 * now, unwire all the entries that were successfully
3775 		 * wired above.
3776 		 *
3777 		 * Skip VM_PROT_NONE entries like we did above.
3778 		 */
3779 
3780 		for (entry = map->header.next; entry != failed_entry;
3781 		     entry = entry->next) {
3782 			if (entry->protection == VM_PROT_NONE)
3783 				continue;
3784 			entry->wired_count--;
3785 			if (VM_MAPENT_ISWIRED(entry))
3786 				uvm_map_entry_unwire(map, entry);
3787 		}
3788 		vm_map_unlock(map);
3789 		UVMHIST_LOG(maphist,"<- done (RV=%jd)", rv,0,0,0);
3790 		return (rv);
3791 	}
3792 
3793 	vm_map_unbusy(map);
3794 
3795 	UVMHIST_LOG(maphist,"<- done (OK WIRE)",0,0,0,0);
3796 	return 0;
3797 }
3798 
3799 /*
3800  * uvm_map_clean: clean out a map range
3801  *
3802  * => valid flags:
3803  *   if (flags & PGO_CLEANIT): dirty pages are cleaned first
3804  *   if (flags & PGO_SYNCIO): dirty pages are written synchronously
3805  *   if (flags & PGO_DEACTIVATE): any cached pages are deactivated after clean
3806  *   if (flags & PGO_FREE): any cached pages are freed after clean
3807  * => returns an error if any part of the specified range isn't mapped
3808  * => never a need to flush amap layer since the anonymous memory has
3809  *	no permanent home, but may deactivate pages there
3810  * => called from sys_msync() and sys_madvise()
3811  * => caller must not write-lock map (read OK).
3812  * => we may sleep while cleaning if SYNCIO [with map read-locked]
3813  */
3814 
3815 int
3816 uvm_map_clean(struct vm_map *map, vaddr_t start, vaddr_t end, int flags)
3817 {
3818 	struct vm_map_entry *current, *entry;
3819 	struct uvm_object *uobj;
3820 	struct vm_amap *amap;
3821 	struct vm_anon *anon, *anon_tofree;
3822 	struct vm_page *pg;
3823 	vaddr_t offset;
3824 	vsize_t size;
3825 	voff_t uoff;
3826 	int error, refs;
3827 	UVMHIST_FUNC("uvm_map_clean"); UVMHIST_CALLED(maphist);
3828 
3829 	UVMHIST_LOG(maphist,"(map=%#jx,start=%#jx,end=%#jx,flags=%#jx)",
3830 	    (uintptr_t)map, start, end, flags);
3831 	KASSERT((flags & (PGO_FREE|PGO_DEACTIVATE)) !=
3832 		(PGO_FREE|PGO_DEACTIVATE));
3833 
3834 	vm_map_lock_read(map);
3835 	VM_MAP_RANGE_CHECK(map, start, end);
3836 	if (uvm_map_lookup_entry(map, start, &entry) == false) {
3837 		vm_map_unlock_read(map);
3838 		return EFAULT;
3839 	}
3840 
3841 	/*
3842 	 * Make a first pass to check for holes and wiring problems.
3843 	 */
3844 
3845 	for (current = entry; current->start < end; current = current->next) {
3846 		if (UVM_ET_ISSUBMAP(current)) {
3847 			vm_map_unlock_read(map);
3848 			return EINVAL;
3849 		}
3850 		if ((flags & PGO_FREE) != 0 && VM_MAPENT_ISWIRED(entry)) {
3851 			vm_map_unlock_read(map);
3852 			return EBUSY;
3853 		}
3854 		if (end <= current->end) {
3855 			break;
3856 		}
3857 		if (current->end != current->next->start) {
3858 			vm_map_unlock_read(map);
3859 			return EFAULT;
3860 		}
3861 	}
3862 
3863 	error = 0;
3864 	for (current = entry; start < end; current = current->next) {
3865 		amap = current->aref.ar_amap;	/* upper layer */
3866 		uobj = current->object.uvm_obj;	/* lower layer */
3867 		KASSERT(start >= current->start);
3868 
3869 		/*
3870 		 * No amap cleaning necessary if:
3871 		 *
3872 		 *	(1) There's no amap.
3873 		 *
3874 		 *	(2) We're not deactivating or freeing pages.
3875 		 */
3876 
3877 		if (amap == NULL || (flags & (PGO_DEACTIVATE|PGO_FREE)) == 0)
3878 			goto flush_object;
3879 
3880 		offset = start - current->start;
3881 		size = MIN(end, current->end) - start;
3882 		anon_tofree = NULL;
3883 
3884 		amap_lock(amap);
3885 		for ( ; size != 0; size -= PAGE_SIZE, offset += PAGE_SIZE) {
3886 			anon = amap_lookup(&current->aref, offset);
3887 			if (anon == NULL)
3888 				continue;
3889 
3890 			KASSERT(anon->an_lock == amap->am_lock);
3891 			pg = anon->an_page;
3892 			if (pg == NULL) {
3893 				continue;
3894 			}
3895 			if (pg->flags & PG_BUSY) {
3896 				continue;
3897 			}
3898 
3899 			switch (flags & (PGO_CLEANIT|PGO_FREE|PGO_DEACTIVATE)) {
3900 
3901 			/*
3902 			 * In these first 3 cases, we just deactivate the page.
3903 			 */
3904 
3905 			case PGO_CLEANIT|PGO_FREE:
3906 			case PGO_CLEANIT|PGO_DEACTIVATE:
3907 			case PGO_DEACTIVATE:
3908  deactivate_it:
3909 				/*
3910 				 * skip the page if it's loaned or wired,
3911 				 * since it shouldn't be on a paging queue
3912 				 * at all in these cases.
3913 				 */
3914 
3915 				mutex_enter(&uvm_pageqlock);
3916 				if (pg->loan_count != 0 ||
3917 				    pg->wire_count != 0) {
3918 					mutex_exit(&uvm_pageqlock);
3919 					continue;
3920 				}
3921 				KASSERT(pg->uanon == anon);
3922 				uvm_pagedeactivate(pg);
3923 				mutex_exit(&uvm_pageqlock);
3924 				continue;
3925 
3926 			case PGO_FREE:
3927 
3928 				/*
3929 				 * If there are multiple references to
3930 				 * the amap, just deactivate the page.
3931 				 */
3932 
3933 				if (amap_refs(amap) > 1)
3934 					goto deactivate_it;
3935 
3936 				/* skip the page if it's wired */
3937 				if (pg->wire_count != 0) {
3938 					continue;
3939 				}
3940 				amap_unadd(&current->aref, offset);
3941 				refs = --anon->an_ref;
3942 				if (refs == 0) {
3943 					anon->an_link = anon_tofree;
3944 					anon_tofree = anon;
3945 				}
3946 				continue;
3947 			}
3948 		}
3949 		uvm_anon_freelst(amap, anon_tofree);
3950 
3951  flush_object:
3952 		/*
3953 		 * flush pages if we've got a valid backing object.
3954 		 * note that we must always clean object pages before
3955 		 * freeing them since otherwise we could reveal stale
3956 		 * data from files.
3957 		 */
3958 
3959 		uoff = current->offset + (start - current->start);
3960 		size = MIN(end, current->end) - start;
3961 		if (uobj != NULL) {
3962 			mutex_enter(uobj->vmobjlock);
3963 			if (uobj->pgops->pgo_put != NULL)
3964 				error = (uobj->pgops->pgo_put)(uobj, uoff,
3965 				    uoff + size, flags | PGO_CLEANIT);
3966 			else
3967 				error = 0;
3968 		}
3969 		start += size;
3970 	}
3971 	vm_map_unlock_read(map);
3972 	return (error);
3973 }
3974 
3975 
3976 /*
3977  * uvm_map_checkprot: check protection in map
3978  *
3979  * => must allow specified protection in a fully allocated region.
3980  * => map must be read or write locked by caller.
3981  */
3982 
3983 bool
3984 uvm_map_checkprot(struct vm_map *map, vaddr_t start, vaddr_t end,
3985     vm_prot_t protection)
3986 {
3987 	struct vm_map_entry *entry;
3988 	struct vm_map_entry *tmp_entry;
3989 
3990 	if (!uvm_map_lookup_entry(map, start, &tmp_entry)) {
3991 		return (false);
3992 	}
3993 	entry = tmp_entry;
3994 	while (start < end) {
3995 		if (entry == &map->header) {
3996 			return (false);
3997 		}
3998 
3999 		/*
4000 		 * no holes allowed
4001 		 */
4002 
4003 		if (start < entry->start) {
4004 			return (false);
4005 		}
4006 
4007 		/*
4008 		 * check protection associated with entry
4009 		 */
4010 
4011 		if ((entry->protection & protection) != protection) {
4012 			return (false);
4013 		}
4014 		start = entry->end;
4015 		entry = entry->next;
4016 	}
4017 	return (true);
4018 }
4019 
4020 /*
4021  * uvmspace_alloc: allocate a vmspace structure.
4022  *
4023  * - structure includes vm_map and pmap
4024  * - XXX: no locking on this structure
4025  * - refcnt set to 1, rest must be init'd by caller
4026  */
4027 struct vmspace *
4028 uvmspace_alloc(vaddr_t vmin, vaddr_t vmax, bool topdown)
4029 {
4030 	struct vmspace *vm;
4031 	UVMHIST_FUNC("uvmspace_alloc"); UVMHIST_CALLED(maphist);
4032 
4033 	vm = pool_cache_get(&uvm_vmspace_cache, PR_WAITOK);
4034 	uvmspace_init(vm, NULL, vmin, vmax, topdown);
4035 	UVMHIST_LOG(maphist,"<- done (vm=%#jx)", (uintptr_t)vm, 0, 0, 0);
4036 	return (vm);
4037 }
4038 
4039 /*
4040  * uvmspace_init: initialize a vmspace structure.
4041  *
4042  * - XXX: no locking on this structure
4043  * - refcnt set to 1, rest must be init'd by caller
4044  */
4045 void
4046 uvmspace_init(struct vmspace *vm, struct pmap *pmap, vaddr_t vmin,
4047     vaddr_t vmax, bool topdown)
4048 {
4049 	UVMHIST_FUNC("uvmspace_init"); UVMHIST_CALLED(maphist);
4050 
4051 	UVMHIST_LOG(maphist, "(vm=%#jx, pmap=%#jx, vmin=%#jx, vmax=%#jx",
4052 	    (uintptr_t)vm, (uintptr_t)pmap, vmin, vmax);
4053 	UVMHIST_LOG(maphist, "   topdown=%ju)", topdown, 0, 0, 0);
4054 
4055 	memset(vm, 0, sizeof(*vm));
4056 	uvm_map_setup(&vm->vm_map, vmin, vmax, VM_MAP_PAGEABLE
4057 	    | (topdown ? VM_MAP_TOPDOWN : 0)
4058 	    );
4059 	if (pmap)
4060 		pmap_reference(pmap);
4061 	else
4062 		pmap = pmap_create();
4063 	vm->vm_map.pmap = pmap;
4064 	vm->vm_refcnt = 1;
4065 	UVMHIST_LOG(maphist,"<- done",0,0,0,0);
4066 }
4067 
4068 /*
4069  * uvmspace_share: share a vmspace between two processes
4070  *
4071  * - used for vfork, threads(?)
4072  */
4073 
4074 void
4075 uvmspace_share(struct proc *p1, struct proc *p2)
4076 {
4077 
4078 	uvmspace_addref(p1->p_vmspace);
4079 	p2->p_vmspace = p1->p_vmspace;
4080 }
4081 
4082 #if 0
4083 
4084 /*
4085  * uvmspace_unshare: ensure that process "p" has its own, unshared, vmspace
4086  *
4087  * - XXX: no locking on vmspace
4088  */
4089 
4090 void
4091 uvmspace_unshare(struct lwp *l)
4092 {
4093 	struct proc *p = l->l_proc;
4094 	struct vmspace *nvm, *ovm = p->p_vmspace;
4095 
4096 	if (ovm->vm_refcnt == 1)
4097 		/* nothing to do: vmspace isn't shared in the first place */
4098 		return;
4099 
4100 	/* make a new vmspace, still holding old one */
4101 	nvm = uvmspace_fork(ovm);
4102 
4103 	kpreempt_disable();
4104 	pmap_deactivate(l);		/* unbind old vmspace */
4105 	p->p_vmspace = nvm;
4106 	pmap_activate(l);		/* switch to new vmspace */
4107 	kpreempt_enable();
4108 
4109 	uvmspace_free(ovm);		/* drop reference to old vmspace */
4110 }
4111 
4112 #endif
4113 
4114 
4115 /*
4116  * uvmspace_spawn: a new process has been spawned and needs a vmspace
4117  */
4118 
4119 void
4120 uvmspace_spawn(struct lwp *l, vaddr_t start, vaddr_t end, bool topdown)
4121 {
4122 	struct proc *p = l->l_proc;
4123 	struct vmspace *nvm;
4124 
4125 #ifdef __HAVE_CPU_VMSPACE_EXEC
4126 	cpu_vmspace_exec(l, start, end);
4127 #endif
4128 
4129 	nvm = uvmspace_alloc(start, end, topdown);
4130 	kpreempt_disable();
4131 	p->p_vmspace = nvm;
4132 	pmap_activate(l);
4133 	kpreempt_enable();
4134 }
4135 
4136 /*
4137  * uvmspace_exec: the process wants to exec a new program
4138  */
4139 
4140 void
4141 uvmspace_exec(struct lwp *l, vaddr_t start, vaddr_t end, bool topdown)
4142 {
4143 	struct proc *p = l->l_proc;
4144 	struct vmspace *nvm, *ovm = p->p_vmspace;
4145 	struct vm_map *map;
4146 
4147 	KASSERT(ovm != NULL);
4148 #ifdef __HAVE_CPU_VMSPACE_EXEC
4149 	cpu_vmspace_exec(l, start, end);
4150 #endif
4151 
4152 	map = &ovm->vm_map;
4153 	/*
4154 	 * see if more than one process is using this vmspace...
4155 	 */
4156 
4157 	if (ovm->vm_refcnt == 1
4158 	    && topdown == ((ovm->vm_map.flags & VM_MAP_TOPDOWN) != 0)) {
4159 
4160 		/*
4161 		 * if p is the only process using its vmspace then we can safely
4162 		 * recycle that vmspace for the program that is being exec'd.
4163 		 * But only if TOPDOWN matches the requested value for the new
4164 		 * vm space!
4165 		 */
4166 
4167 		/*
4168 		 * SYSV SHM semantics require us to kill all segments on an exec
4169 		 */
4170 		if (uvm_shmexit && ovm->vm_shm)
4171 			(*uvm_shmexit)(ovm);
4172 
4173 		/*
4174 		 * POSIX 1003.1b -- "lock future mappings" is revoked
4175 		 * when a process execs another program image.
4176 		 */
4177 
4178 		map->flags &= ~VM_MAP_WIREFUTURE;
4179 
4180 		/*
4181 		 * now unmap the old program
4182 		 */
4183 
4184 		pmap_remove_all(map->pmap);
4185 		uvm_unmap(map, vm_map_min(map), vm_map_max(map));
4186 		KASSERT(map->header.prev == &map->header);
4187 		KASSERT(map->nentries == 0);
4188 
4189 		/*
4190 		 * resize the map
4191 		 */
4192 
4193 		vm_map_setmin(map, start);
4194 		vm_map_setmax(map, end);
4195 	} else {
4196 
4197 		/*
4198 		 * p's vmspace is being shared, so we can't reuse it for p since
4199 		 * it is still being used for others.   allocate a new vmspace
4200 		 * for p
4201 		 */
4202 
4203 		nvm = uvmspace_alloc(start, end, topdown);
4204 
4205 		/*
4206 		 * install new vmspace and drop our ref to the old one.
4207 		 */
4208 
4209 		kpreempt_disable();
4210 		pmap_deactivate(l);
4211 		p->p_vmspace = nvm;
4212 		pmap_activate(l);
4213 		kpreempt_enable();
4214 
4215 		uvmspace_free(ovm);
4216 	}
4217 }
4218 
4219 /*
4220  * uvmspace_addref: add a referece to a vmspace.
4221  */
4222 
4223 void
4224 uvmspace_addref(struct vmspace *vm)
4225 {
4226 	struct vm_map *map = &vm->vm_map;
4227 
4228 	KASSERT((map->flags & VM_MAP_DYING) == 0);
4229 
4230 	mutex_enter(&map->misc_lock);
4231 	KASSERT(vm->vm_refcnt > 0);
4232 	vm->vm_refcnt++;
4233 	mutex_exit(&map->misc_lock);
4234 }
4235 
4236 /*
4237  * uvmspace_free: free a vmspace data structure
4238  */
4239 
4240 void
4241 uvmspace_free(struct vmspace *vm)
4242 {
4243 	struct vm_map_entry *dead_entries;
4244 	struct vm_map *map = &vm->vm_map;
4245 	int n;
4246 
4247 	UVMHIST_FUNC("uvmspace_free"); UVMHIST_CALLED(maphist);
4248 
4249 	UVMHIST_LOG(maphist,"(vm=%#jx) ref=%jd", (uintptr_t)vm, vm->vm_refcnt,
4250 	    0, 0);
4251 	mutex_enter(&map->misc_lock);
4252 	n = --vm->vm_refcnt;
4253 	mutex_exit(&map->misc_lock);
4254 	if (n > 0)
4255 		return;
4256 
4257 	/*
4258 	 * at this point, there should be no other references to the map.
4259 	 * delete all of the mappings, then destroy the pmap.
4260 	 */
4261 
4262 	map->flags |= VM_MAP_DYING;
4263 	pmap_remove_all(map->pmap);
4264 
4265 	/* Get rid of any SYSV shared memory segments. */
4266 	if (uvm_shmexit && vm->vm_shm != NULL)
4267 		(*uvm_shmexit)(vm);
4268 
4269 	if (map->nentries) {
4270 		uvm_unmap_remove(map, vm_map_min(map), vm_map_max(map),
4271 		    &dead_entries, 0);
4272 		if (dead_entries != NULL)
4273 			uvm_unmap_detach(dead_entries, 0);
4274 	}
4275 	KASSERT(map->nentries == 0);
4276 	KASSERT(map->size == 0);
4277 
4278 	mutex_destroy(&map->misc_lock);
4279 	rw_destroy(&map->lock);
4280 	cv_destroy(&map->cv);
4281 	pmap_destroy(map->pmap);
4282 	pool_cache_put(&uvm_vmspace_cache, vm);
4283 }
4284 
4285 static struct vm_map_entry *
4286 uvm_mapent_clone(struct vm_map *new_map, struct vm_map_entry *old_entry,
4287     int flags)
4288 {
4289 	struct vm_map_entry *new_entry;
4290 
4291 	new_entry = uvm_mapent_alloc(new_map, 0);
4292 	/* old_entry -> new_entry */
4293 	uvm_mapent_copy(old_entry, new_entry);
4294 
4295 	/* new pmap has nothing wired in it */
4296 	new_entry->wired_count = 0;
4297 
4298 	/*
4299 	 * gain reference to object backing the map (can't
4300 	 * be a submap, already checked this case).
4301 	 */
4302 
4303 	if (new_entry->aref.ar_amap)
4304 		uvm_map_reference_amap(new_entry, flags);
4305 
4306 	if (new_entry->object.uvm_obj &&
4307 	    new_entry->object.uvm_obj->pgops->pgo_reference)
4308 		new_entry->object.uvm_obj->pgops->pgo_reference(
4309 			new_entry->object.uvm_obj);
4310 
4311 	/* insert entry at end of new_map's entry list */
4312 	uvm_map_entry_link(new_map, new_map->header.prev,
4313 	    new_entry);
4314 
4315 	return new_entry;
4316 }
4317 
4318 /*
4319  * share the mapping: this means we want the old and
4320  * new entries to share amaps and backing objects.
4321  */
4322 static void
4323 uvm_mapent_forkshared(struct vm_map *new_map, struct vm_map *old_map,
4324     struct vm_map_entry *old_entry)
4325 {
4326 	/*
4327 	 * if the old_entry needs a new amap (due to prev fork)
4328 	 * then we need to allocate it now so that we have
4329 	 * something we own to share with the new_entry.   [in
4330 	 * other words, we need to clear needs_copy]
4331 	 */
4332 
4333 	if (UVM_ET_ISNEEDSCOPY(old_entry)) {
4334 		/* get our own amap, clears needs_copy */
4335 		amap_copy(old_map, old_entry, AMAP_COPY_NOCHUNK,
4336 		    0, 0);
4337 		/* XXXCDC: WAITOK??? */
4338 	}
4339 
4340 	uvm_mapent_clone(new_map, old_entry, AMAP_SHARED);
4341 }
4342 
4343 
4344 static void
4345 uvm_mapent_forkcopy(struct vm_map *new_map, struct vm_map *old_map,
4346     struct vm_map_entry *old_entry)
4347 {
4348 	struct vm_map_entry *new_entry;
4349 
4350 	/*
4351 	 * copy-on-write the mapping (using mmap's
4352 	 * MAP_PRIVATE semantics)
4353 	 *
4354 	 * allocate new_entry, adjust reference counts.
4355 	 * (note that new references are read-only).
4356 	 */
4357 
4358 	new_entry = uvm_mapent_clone(new_map, old_entry, 0);
4359 
4360 	new_entry->etype |=
4361 	    (UVM_ET_COPYONWRITE|UVM_ET_NEEDSCOPY);
4362 
4363 	/*
4364 	 * the new entry will need an amap.  it will either
4365 	 * need to be copied from the old entry or created
4366 	 * from scratch (if the old entry does not have an
4367 	 * amap).  can we defer this process until later
4368 	 * (by setting "needs_copy") or do we need to copy
4369 	 * the amap now?
4370 	 *
4371 	 * we must copy the amap now if any of the following
4372 	 * conditions hold:
4373 	 * 1. the old entry has an amap and that amap is
4374 	 *    being shared.  this means that the old (parent)
4375 	 *    process is sharing the amap with another
4376 	 *    process.  if we do not clear needs_copy here
4377 	 *    we will end up in a situation where both the
4378 	 *    parent and child process are refering to the
4379 	 *    same amap with "needs_copy" set.  if the
4380 	 *    parent write-faults, the fault routine will
4381 	 *    clear "needs_copy" in the parent by allocating
4382 	 *    a new amap.   this is wrong because the
4383 	 *    parent is supposed to be sharing the old amap
4384 	 *    and the new amap will break that.
4385 	 *
4386 	 * 2. if the old entry has an amap and a non-zero
4387 	 *    wire count then we are going to have to call
4388 	 *    amap_cow_now to avoid page faults in the
4389 	 *    parent process.   since amap_cow_now requires
4390 	 *    "needs_copy" to be clear we might as well
4391 	 *    clear it here as well.
4392 	 *
4393 	 */
4394 
4395 	if (old_entry->aref.ar_amap != NULL) {
4396 		if ((amap_flags(old_entry->aref.ar_amap) & AMAP_SHARED) != 0 ||
4397 		    VM_MAPENT_ISWIRED(old_entry)) {
4398 
4399 			amap_copy(new_map, new_entry,
4400 			    AMAP_COPY_NOCHUNK, 0, 0);
4401 			/* XXXCDC: M_WAITOK ... ok? */
4402 		}
4403 	}
4404 
4405 	/*
4406 	 * if the parent's entry is wired down, then the
4407 	 * parent process does not want page faults on
4408 	 * access to that memory.  this means that we
4409 	 * cannot do copy-on-write because we can't write
4410 	 * protect the old entry.   in this case we
4411 	 * resolve all copy-on-write faults now, using
4412 	 * amap_cow_now.   note that we have already
4413 	 * allocated any needed amap (above).
4414 	 */
4415 
4416 	if (VM_MAPENT_ISWIRED(old_entry)) {
4417 
4418 		/*
4419 		 * resolve all copy-on-write faults now
4420 		 * (note that there is nothing to do if
4421 		 * the old mapping does not have an amap).
4422 		 */
4423 		if (old_entry->aref.ar_amap)
4424 			amap_cow_now(new_map, new_entry);
4425 
4426 	} else {
4427 		/*
4428 		 * setup mappings to trigger copy-on-write faults
4429 		 * we must write-protect the parent if it has
4430 		 * an amap and it is not already "needs_copy"...
4431 		 * if it is already "needs_copy" then the parent
4432 		 * has already been write-protected by a previous
4433 		 * fork operation.
4434 		 */
4435 		if (old_entry->aref.ar_amap &&
4436 		    !UVM_ET_ISNEEDSCOPY(old_entry)) {
4437 			if (old_entry->max_protection & VM_PROT_WRITE) {
4438 				pmap_protect(old_map->pmap,
4439 				    old_entry->start, old_entry->end,
4440 				    old_entry->protection & ~VM_PROT_WRITE);
4441 			}
4442 			old_entry->etype |= UVM_ET_NEEDSCOPY;
4443 		}
4444 	}
4445 }
4446 
4447 /*
4448  * zero the mapping: the new entry will be zero initialized
4449  */
4450 static void
4451 uvm_mapent_forkzero(struct vm_map *new_map, struct vm_map *old_map,
4452     struct vm_map_entry *old_entry)
4453 {
4454 	struct vm_map_entry *new_entry;
4455 
4456 	new_entry = uvm_mapent_clone(new_map, old_entry, 0);
4457 
4458 	new_entry->etype |=
4459 	    (UVM_ET_COPYONWRITE|UVM_ET_NEEDSCOPY);
4460 
4461 	if (new_entry->aref.ar_amap) {
4462 		uvm_map_unreference_amap(new_entry, 0);
4463 		new_entry->aref.ar_pageoff = 0;
4464 		new_entry->aref.ar_amap = NULL;
4465 	}
4466 
4467 	if (UVM_ET_ISOBJ(new_entry)) {
4468 		if (new_entry->object.uvm_obj->pgops->pgo_detach)
4469 			new_entry->object.uvm_obj->pgops->pgo_detach(
4470 			    new_entry->object.uvm_obj);
4471 		new_entry->object.uvm_obj = NULL;
4472 		new_entry->etype &= ~UVM_ET_OBJ;
4473 	}
4474 }
4475 
4476 /*
4477  *   F O R K   -   m a i n   e n t r y   p o i n t
4478  */
4479 /*
4480  * uvmspace_fork: fork a process' main map
4481  *
4482  * => create a new vmspace for child process from parent.
4483  * => parent's map must not be locked.
4484  */
4485 
4486 struct vmspace *
4487 uvmspace_fork(struct vmspace *vm1)
4488 {
4489 	struct vmspace *vm2;
4490 	struct vm_map *old_map = &vm1->vm_map;
4491 	struct vm_map *new_map;
4492 	struct vm_map_entry *old_entry;
4493 	UVMHIST_FUNC("uvmspace_fork"); UVMHIST_CALLED(maphist);
4494 
4495 	vm_map_lock(old_map);
4496 
4497 	vm2 = uvmspace_alloc(vm_map_min(old_map), vm_map_max(old_map),
4498 	    vm1->vm_map.flags & VM_MAP_TOPDOWN);
4499 	memcpy(&vm2->vm_startcopy, &vm1->vm_startcopy,
4500 	    (char *) (vm1 + 1) - (char *) &vm1->vm_startcopy);
4501 	new_map = &vm2->vm_map;		  /* XXX */
4502 
4503 	old_entry = old_map->header.next;
4504 	new_map->size = old_map->size;
4505 
4506 	/*
4507 	 * go entry-by-entry
4508 	 */
4509 
4510 	while (old_entry != &old_map->header) {
4511 
4512 		/*
4513 		 * first, some sanity checks on the old entry
4514 		 */
4515 
4516 		KASSERT(!UVM_ET_ISSUBMAP(old_entry));
4517 		KASSERT(UVM_ET_ISCOPYONWRITE(old_entry) ||
4518 			!UVM_ET_ISNEEDSCOPY(old_entry));
4519 
4520 		switch (old_entry->inheritance) {
4521 		case MAP_INHERIT_NONE:
4522 			/*
4523 			 * drop the mapping, modify size
4524 			 */
4525 			new_map->size -= old_entry->end - old_entry->start;
4526 			break;
4527 
4528 		case MAP_INHERIT_SHARE:
4529 			uvm_mapent_forkshared(new_map, old_map, old_entry);
4530 			break;
4531 
4532 		case MAP_INHERIT_COPY:
4533 			uvm_mapent_forkcopy(new_map, old_map, old_entry);
4534 			break;
4535 
4536 		case MAP_INHERIT_ZERO:
4537 			uvm_mapent_forkzero(new_map, old_map, old_entry);
4538 			break;
4539 		default:
4540 			KASSERT(0);
4541 			break;
4542 		}
4543 		old_entry = old_entry->next;
4544 	}
4545 
4546 	pmap_update(old_map->pmap);
4547 	vm_map_unlock(old_map);
4548 
4549 	if (uvm_shmfork && vm1->vm_shm)
4550 		(*uvm_shmfork)(vm1, vm2);
4551 
4552 #ifdef PMAP_FORK
4553 	pmap_fork(vm1->vm_map.pmap, vm2->vm_map.pmap);
4554 #endif
4555 
4556 	UVMHIST_LOG(maphist,"<- done",0,0,0,0);
4557 	return (vm2);
4558 }
4559 
4560 
4561 /*
4562  * uvm_mapent_trymerge: try to merge an entry with its neighbors.
4563  *
4564  * => called with map locked.
4565  * => return non zero if successfully merged.
4566  */
4567 
4568 int
4569 uvm_mapent_trymerge(struct vm_map *map, struct vm_map_entry *entry, int flags)
4570 {
4571 	struct uvm_object *uobj;
4572 	struct vm_map_entry *next;
4573 	struct vm_map_entry *prev;
4574 	vsize_t size;
4575 	int merged = 0;
4576 	bool copying;
4577 	int newetype;
4578 
4579 	if (entry->aref.ar_amap != NULL) {
4580 		return 0;
4581 	}
4582 	if ((entry->flags & UVM_MAP_NOMERGE) != 0) {
4583 		return 0;
4584 	}
4585 
4586 	uobj = entry->object.uvm_obj;
4587 	size = entry->end - entry->start;
4588 	copying = (flags & UVM_MERGE_COPYING) != 0;
4589 	newetype = copying ? (entry->etype & ~UVM_ET_NEEDSCOPY) : entry->etype;
4590 
4591 	next = entry->next;
4592 	if (next != &map->header &&
4593 	    next->start == entry->end &&
4594 	    ((copying && next->aref.ar_amap != NULL &&
4595 	    amap_refs(next->aref.ar_amap) == 1) ||
4596 	    (!copying && next->aref.ar_amap == NULL)) &&
4597 	    UVM_ET_ISCOMPATIBLE(next, newetype,
4598 	    uobj, entry->flags, entry->protection,
4599 	    entry->max_protection, entry->inheritance, entry->advice,
4600 	    entry->wired_count) &&
4601 	    (uobj == NULL || entry->offset + size == next->offset)) {
4602 		int error;
4603 
4604 		if (copying) {
4605 			error = amap_extend(next, size,
4606 			    AMAP_EXTEND_NOWAIT|AMAP_EXTEND_BACKWARDS);
4607 		} else {
4608 			error = 0;
4609 		}
4610 		if (error == 0) {
4611 			if (uobj) {
4612 				if (uobj->pgops->pgo_detach) {
4613 					uobj->pgops->pgo_detach(uobj);
4614 				}
4615 			}
4616 
4617 			entry->end = next->end;
4618 			clear_hints(map, next);
4619 			uvm_map_entry_unlink(map, next);
4620 			if (copying) {
4621 				entry->aref = next->aref;
4622 				entry->etype &= ~UVM_ET_NEEDSCOPY;
4623 			}
4624 			uvm_map_check(map, "trymerge forwardmerge");
4625 			uvm_mapent_free(next);
4626 			merged++;
4627 		}
4628 	}
4629 
4630 	prev = entry->prev;
4631 	if (prev != &map->header &&
4632 	    prev->end == entry->start &&
4633 	    ((copying && !merged && prev->aref.ar_amap != NULL &&
4634 	    amap_refs(prev->aref.ar_amap) == 1) ||
4635 	    (!copying && prev->aref.ar_amap == NULL)) &&
4636 	    UVM_ET_ISCOMPATIBLE(prev, newetype,
4637 	    uobj, entry->flags, entry->protection,
4638 	    entry->max_protection, entry->inheritance, entry->advice,
4639 	    entry->wired_count) &&
4640 	    (uobj == NULL ||
4641 	    prev->offset + prev->end - prev->start == entry->offset)) {
4642 		int error;
4643 
4644 		if (copying) {
4645 			error = amap_extend(prev, size,
4646 			    AMAP_EXTEND_NOWAIT|AMAP_EXTEND_FORWARDS);
4647 		} else {
4648 			error = 0;
4649 		}
4650 		if (error == 0) {
4651 			if (uobj) {
4652 				if (uobj->pgops->pgo_detach) {
4653 					uobj->pgops->pgo_detach(uobj);
4654 				}
4655 				entry->offset = prev->offset;
4656 			}
4657 
4658 			entry->start = prev->start;
4659 			clear_hints(map, prev);
4660 			uvm_map_entry_unlink(map, prev);
4661 			if (copying) {
4662 				entry->aref = prev->aref;
4663 				entry->etype &= ~UVM_ET_NEEDSCOPY;
4664 			}
4665 			uvm_map_check(map, "trymerge backmerge");
4666 			uvm_mapent_free(prev);
4667 			merged++;
4668 		}
4669 	}
4670 
4671 	return merged;
4672 }
4673 
4674 /*
4675  * uvm_map_setup: init map
4676  *
4677  * => map must not be in service yet.
4678  */
4679 
4680 void
4681 uvm_map_setup(struct vm_map *map, vaddr_t vmin, vaddr_t vmax, int flags)
4682 {
4683 
4684 	rb_tree_init(&map->rb_tree, &uvm_map_tree_ops);
4685 	map->header.next = map->header.prev = &map->header;
4686 	map->nentries = 0;
4687 	map->size = 0;
4688 	map->ref_count = 1;
4689 	vm_map_setmin(map, vmin);
4690 	vm_map_setmax(map, vmax);
4691 	map->flags = flags;
4692 	map->first_free = &map->header;
4693 	map->hint = &map->header;
4694 	map->timestamp = 0;
4695 	map->busy = NULL;
4696 
4697 	rw_init(&map->lock);
4698 	cv_init(&map->cv, "vm_map");
4699 	mutex_init(&map->misc_lock, MUTEX_DRIVER, IPL_NONE);
4700 }
4701 
4702 /*
4703  *   U N M A P   -   m a i n   e n t r y   p o i n t
4704  */
4705 
4706 /*
4707  * uvm_unmap1: remove mappings from a vm_map (from "start" up to "stop")
4708  *
4709  * => caller must check alignment and size
4710  * => map must be unlocked (we will lock it)
4711  * => flags is UVM_FLAG_QUANTUM or 0.
4712  */
4713 
4714 void
4715 uvm_unmap1(struct vm_map *map, vaddr_t start, vaddr_t end, int flags)
4716 {
4717 	struct vm_map_entry *dead_entries;
4718 	UVMHIST_FUNC("uvm_unmap"); UVMHIST_CALLED(maphist);
4719 
4720 	KASSERT(start < end);
4721 	UVMHIST_LOG(maphist, "  (map=%#jx, start=%#jx, end=%#jx)",
4722 	    (uintptr_t)map, start, end, 0);
4723 	if (map == kernel_map) {
4724 		LOCKDEBUG_MEM_CHECK((void *)start, end - start);
4725 	}
4726 
4727 	/*
4728 	 * work now done by helper functions.   wipe the pmap's and then
4729 	 * detach from the dead entries...
4730 	 */
4731 	vm_map_lock(map);
4732 	uvm_unmap_remove(map, start, end, &dead_entries, flags);
4733 	vm_map_unlock(map);
4734 
4735 	if (dead_entries != NULL)
4736 		uvm_unmap_detach(dead_entries, 0);
4737 
4738 	UVMHIST_LOG(maphist, "<- done", 0,0,0,0);
4739 }
4740 
4741 
4742 /*
4743  * uvm_map_reference: add reference to a map
4744  *
4745  * => map need not be locked (we use misc_lock).
4746  */
4747 
4748 void
4749 uvm_map_reference(struct vm_map *map)
4750 {
4751 	mutex_enter(&map->misc_lock);
4752 	map->ref_count++;
4753 	mutex_exit(&map->misc_lock);
4754 }
4755 
4756 bool
4757 vm_map_starved_p(struct vm_map *map)
4758 {
4759 
4760 	if ((map->flags & VM_MAP_WANTVA) != 0) {
4761 		return true;
4762 	}
4763 	/* XXX */
4764 	if ((vm_map_max(map) - vm_map_min(map)) / 16 * 15 < map->size) {
4765 		return true;
4766 	}
4767 	return false;
4768 }
4769 
4770 void
4771 uvm_map_lock_entry(struct vm_map_entry *entry)
4772 {
4773 
4774 	if (entry->aref.ar_amap != NULL) {
4775 		amap_lock(entry->aref.ar_amap);
4776 	}
4777 	if (UVM_ET_ISOBJ(entry)) {
4778 		mutex_enter(entry->object.uvm_obj->vmobjlock);
4779 	}
4780 }
4781 
4782 void
4783 uvm_map_unlock_entry(struct vm_map_entry *entry)
4784 {
4785 
4786 	if (UVM_ET_ISOBJ(entry)) {
4787 		mutex_exit(entry->object.uvm_obj->vmobjlock);
4788 	}
4789 	if (entry->aref.ar_amap != NULL) {
4790 		amap_unlock(entry->aref.ar_amap);
4791 	}
4792 }
4793 
4794 #if defined(DDB) || defined(DEBUGPRINT)
4795 
4796 /*
4797  * uvm_map_printit: actually prints the map
4798  */
4799 
4800 void
4801 uvm_map_printit(struct vm_map *map, bool full,
4802     void (*pr)(const char *, ...))
4803 {
4804 	struct vm_map_entry *entry;
4805 
4806 	(*pr)("MAP %p: [%#lx->%#lx]\n", map, vm_map_min(map),
4807 	    vm_map_max(map));
4808 	(*pr)("\t#ent=%d, sz=%d, ref=%d, version=%d, flags=%#x\n",
4809 	    map->nentries, map->size, map->ref_count, map->timestamp,
4810 	    map->flags);
4811 	(*pr)("\tpmap=%p(resident=%ld, wired=%ld)\n", map->pmap,
4812 	    pmap_resident_count(map->pmap), pmap_wired_count(map->pmap));
4813 	if (!full)
4814 		return;
4815 	for (entry = map->header.next; entry != &map->header;
4816 	    entry = entry->next) {
4817 		(*pr)(" - %p: %#lx->%#lx: obj=%p/%#llx, amap=%p/%d\n",
4818 		    entry, entry->start, entry->end, entry->object.uvm_obj,
4819 		    (long long)entry->offset, entry->aref.ar_amap,
4820 		    entry->aref.ar_pageoff);
4821 		(*pr)(
4822 		    "\tsubmap=%c, cow=%c, nc=%c, prot(max)=%d/%d, inh=%d, "
4823 		    "wc=%d, adv=%d\n",
4824 		    (entry->etype & UVM_ET_SUBMAP) ? 'T' : 'F',
4825 		    (entry->etype & UVM_ET_COPYONWRITE) ? 'T' : 'F',
4826 		    (entry->etype & UVM_ET_NEEDSCOPY) ? 'T' : 'F',
4827 		    entry->protection, entry->max_protection,
4828 		    entry->inheritance, entry->wired_count, entry->advice);
4829 	}
4830 }
4831 
4832 void
4833 uvm_whatis(uintptr_t addr, void (*pr)(const char *, ...))
4834 {
4835 	struct vm_map *map;
4836 
4837 	for (map = kernel_map;;) {
4838 		struct vm_map_entry *entry;
4839 
4840 		if (!uvm_map_lookup_entry_bytree(map, (vaddr_t)addr, &entry)) {
4841 			break;
4842 		}
4843 		(*pr)("%p is %p+%zu from VMMAP %p\n",
4844 		    (void *)addr, (void *)entry->start,
4845 		    (size_t)(addr - (uintptr_t)entry->start), map);
4846 		if (!UVM_ET_ISSUBMAP(entry)) {
4847 			break;
4848 		}
4849 		map = entry->object.sub_map;
4850 	}
4851 }
4852 
4853 #endif /* DDB || DEBUGPRINT */
4854 
4855 #ifndef __USER_VA0_IS_SAFE
4856 static int
4857 sysctl_user_va0_disable(SYSCTLFN_ARGS)
4858 {
4859 	struct sysctlnode node;
4860 	int t, error;
4861 
4862 	node = *rnode;
4863 	node.sysctl_data = &t;
4864 	t = user_va0_disable;
4865 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
4866 	if (error || newp == NULL)
4867 		return (error);
4868 
4869 	if (!t && user_va0_disable &&
4870 	    kauth_authorize_system(l->l_cred, KAUTH_SYSTEM_MAP_VA_ZERO, 0,
4871 	    NULL, NULL, NULL))
4872 		return EPERM;
4873 
4874 	user_va0_disable = !!t;
4875 	return 0;
4876 }
4877 #endif
4878 
4879 static int
4880 fill_vmentry(struct lwp *l, struct proc *p, struct kinfo_vmentry *kve,
4881     struct vm_map *m, struct vm_map_entry *e)
4882 {
4883 #ifndef _RUMPKERNEL
4884 	int error;
4885 
4886 	memset(kve, 0, sizeof(*kve));
4887 	KASSERT(e != NULL);
4888 	if (UVM_ET_ISOBJ(e)) {
4889 		struct uvm_object *uobj = e->object.uvm_obj;
4890 		KASSERT(uobj != NULL);
4891 		kve->kve_ref_count = uobj->uo_refs;
4892 		kve->kve_count = uobj->uo_npages;
4893 		if (UVM_OBJ_IS_VNODE(uobj)) {
4894 			struct vattr va;
4895 			struct vnode *vp = (struct vnode *)uobj;
4896 			vn_lock(vp, LK_SHARED | LK_RETRY);
4897 			error = VOP_GETATTR(vp, &va, l->l_cred);
4898 			VOP_UNLOCK(vp);
4899 			kve->kve_type = KVME_TYPE_VNODE;
4900 			if (error == 0) {
4901 				kve->kve_vn_size = vp->v_size;
4902 				kve->kve_vn_type = (int)vp->v_type;
4903 				kve->kve_vn_mode = va.va_mode;
4904 				kve->kve_vn_rdev = va.va_rdev;
4905 				kve->kve_vn_fileid = va.va_fileid;
4906 				kve->kve_vn_fsid = va.va_fsid;
4907 				error = vnode_to_path(kve->kve_path,
4908 				    sizeof(kve->kve_path) / 2, vp, l, p);
4909 #ifdef DIAGNOSTIC
4910 				if (error)
4911 					printf("%s: vp %p error %d\n", __func__,
4912 						vp, error);
4913 #endif
4914 			}
4915 		} else if (UVM_OBJ_IS_KERN_OBJECT(uobj)) {
4916 			kve->kve_type = KVME_TYPE_KERN;
4917 		} else if (UVM_OBJ_IS_DEVICE(uobj)) {
4918 			kve->kve_type = KVME_TYPE_DEVICE;
4919 		} else if (UVM_OBJ_IS_AOBJ(uobj)) {
4920 			kve->kve_type = KVME_TYPE_ANON;
4921 		} else {
4922 			kve->kve_type = KVME_TYPE_OBJECT;
4923 		}
4924 	} else if (UVM_ET_ISSUBMAP(e)) {
4925 		struct vm_map *map = e->object.sub_map;
4926 		KASSERT(map != NULL);
4927 		kve->kve_ref_count = map->ref_count;
4928 		kve->kve_count = map->nentries;
4929 		kve->kve_type = KVME_TYPE_SUBMAP;
4930 	} else
4931 		kve->kve_type = KVME_TYPE_UNKNOWN;
4932 
4933 	kve->kve_start = e->start;
4934 	kve->kve_end = e->end;
4935 	kve->kve_offset = e->offset;
4936 	kve->kve_wired_count = e->wired_count;
4937 	kve->kve_inheritance = e->inheritance;
4938 	kve->kve_attributes = e->map_attrib;
4939 	kve->kve_advice = e->advice;
4940 #define PROT(p) (((p) & VM_PROT_READ) ? KVME_PROT_READ : 0) | \
4941 	(((p) & VM_PROT_WRITE) ? KVME_PROT_WRITE : 0) | \
4942 	(((p) & VM_PROT_EXECUTE) ? KVME_PROT_EXEC : 0)
4943 	kve->kve_protection = PROT(e->protection);
4944 	kve->kve_max_protection = PROT(e->max_protection);
4945 	kve->kve_flags |= (e->etype & UVM_ET_COPYONWRITE)
4946 	    ? KVME_FLAG_COW : 0;
4947 	kve->kve_flags |= (e->etype & UVM_ET_NEEDSCOPY)
4948 	    ? KVME_FLAG_NEEDS_COPY : 0;
4949 	kve->kve_flags |= (m->flags & VM_MAP_TOPDOWN)
4950 	    ? KVME_FLAG_GROWS_DOWN : KVME_FLAG_GROWS_UP;
4951 	kve->kve_flags |= (m->flags & VM_MAP_PAGEABLE)
4952 	    ? KVME_FLAG_PAGEABLE : 0;
4953 #endif
4954 	return 0;
4955 }
4956 
4957 static int
4958 fill_vmentries(struct lwp *l, pid_t pid, u_int elem_size, void *oldp,
4959     size_t *oldlenp)
4960 {
4961 	int error;
4962 	struct proc *p;
4963 	struct kinfo_vmentry *vme;
4964 	struct vmspace *vm;
4965 	struct vm_map *map;
4966 	struct vm_map_entry *entry;
4967 	char *dp;
4968 	size_t count, vmesize;
4969 
4970 	if (elem_size == 0 || elem_size > 2 * sizeof(*vme))
4971 		return EINVAL;
4972 
4973 	if (oldp) {
4974 		if (*oldlenp > 10UL * 1024UL * 1024UL)
4975 			return E2BIG;
4976 		count = *oldlenp / elem_size;
4977 		if (count == 0)
4978 			return ENOMEM;
4979 		vmesize = count * sizeof(*vme);
4980 	} else
4981 		vmesize = 0;
4982 
4983 	if ((error = proc_find_locked(l, &p, pid)) != 0)
4984 		return error;
4985 
4986 	vme = NULL;
4987 	count = 0;
4988 
4989 	if ((error = proc_vmspace_getref(p, &vm)) != 0)
4990 		goto out;
4991 
4992 	map = &vm->vm_map;
4993 	vm_map_lock_read(map);
4994 
4995 	dp = oldp;
4996 	if (oldp)
4997 		vme = kmem_alloc(vmesize, KM_SLEEP);
4998 	for (entry = map->header.next; entry != &map->header;
4999 	    entry = entry->next) {
5000 		if (oldp && (dp - (char *)oldp) < vmesize) {
5001 			error = fill_vmentry(l, p, &vme[count], map, entry);
5002 			if (error)
5003 				goto out;
5004 			dp += elem_size;
5005 		}
5006 		count++;
5007 	}
5008 	vm_map_unlock_read(map);
5009 	uvmspace_free(vm);
5010 
5011 out:
5012 	if (pid != -1)
5013 		mutex_exit(p->p_lock);
5014 	if (error == 0) {
5015 		const u_int esize = uimin(sizeof(*vme), elem_size);
5016 		dp = oldp;
5017 		for (size_t i = 0; i < count; i++) {
5018 			if (oldp && (dp - (char *)oldp) < vmesize) {
5019 				error = sysctl_copyout(l, &vme[i], dp, esize);
5020 				if (error)
5021 					break;
5022 				dp += elem_size;
5023 			} else
5024 				break;
5025 		}
5026 		count *= elem_size;
5027 		if (oldp != NULL && *oldlenp < count)
5028 			error = ENOSPC;
5029 		*oldlenp = count;
5030 	}
5031 	if (vme)
5032 		kmem_free(vme, vmesize);
5033 	return error;
5034 }
5035 
5036 static int
5037 sysctl_vmproc(SYSCTLFN_ARGS)
5038 {
5039 	int error;
5040 
5041 	if (namelen == 1 && name[0] == CTL_QUERY)
5042 		return (sysctl_query(SYSCTLFN_CALL(rnode)));
5043 
5044 	if (namelen == 0)
5045 		return EINVAL;
5046 
5047 	switch (name[0]) {
5048 	case VM_PROC_MAP:
5049 		if (namelen != 3)
5050 			return EINVAL;
5051 		sysctl_unlock();
5052 		error = fill_vmentries(l, name[1], name[2], oldp, oldlenp);
5053 		sysctl_relock();
5054 		return error;
5055 	default:
5056 		return EINVAL;
5057 	}
5058 }
5059 
5060 SYSCTL_SETUP(sysctl_uvmmap_setup, "sysctl uvmmap setup")
5061 {
5062 
5063 	sysctl_createv(clog, 0, NULL, NULL,
5064 		       CTLFLAG_PERMANENT,
5065 		       CTLTYPE_STRUCT, "proc",
5066 		       SYSCTL_DESCR("Process vm information"),
5067 		       sysctl_vmproc, 0, NULL, 0,
5068 		       CTL_VM, VM_PROC, CTL_EOL);
5069 #ifndef __USER_VA0_IS_SAFE
5070         sysctl_createv(clog, 0, NULL, NULL,
5071                        CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
5072                        CTLTYPE_INT, "user_va0_disable",
5073                        SYSCTL_DESCR("Disable VA 0"),
5074                        sysctl_user_va0_disable, 0, &user_va0_disable, 0,
5075                        CTL_VM, CTL_CREATE, CTL_EOL);
5076 #endif
5077 }
5078