xref: /netbsd-src/sys/uvm/uvm_map.c (revision e89934bbf778a6d6d6894877c4da59d0c7835b0f)
1 /*	$NetBSD: uvm_map.c,v 1.342 2016/12/01 02:09:03 mrg 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 1.342 2016/12/01 02:09:03 mrg Exp $");
70 
71 #include "opt_ddb.h"
72 #include "opt_uvmhist.h"
73 #include "opt_uvm.h"
74 #include "opt_sysv.h"
75 
76 #include <sys/param.h>
77 #include <sys/systm.h>
78 #include <sys/mman.h>
79 #include <sys/proc.h>
80 #include <sys/pool.h>
81 #include <sys/kernel.h>
82 #include <sys/mount.h>
83 #include <sys/vnode.h>
84 #include <sys/filedesc.h>
85 #include <sys/lockdebug.h>
86 #include <sys/atomic.h>
87 #include <sys/sysctl.h>
88 #ifndef __USER_VA0_IS_SAFE
89 #include <sys/kauth.h>
90 #include "opt_user_va0_disable_default.h"
91 #endif
92 
93 #include <sys/shm.h>
94 
95 #include <uvm/uvm.h>
96 #include <uvm/uvm_readahead.h>
97 
98 #if defined(DDB) || defined(DEBUGPRINT)
99 #include <uvm/uvm_ddb.h>
100 #endif
101 
102 #ifdef UVMHIST
103 #ifndef UVMHIST_MAPHIST_SIZE
104 #define UVMHIST_MAPHIST_SIZE 100
105 #endif
106 #ifndef UVMHIST_PDHIST_SIZE
107 #define UVMHIST_PDHIST_SIZE 100
108 #endif
109 static struct kern_history_ent maphistbuf[UVMHIST_MAPHIST_SIZE];
110 UVMHIST_DEFINE(maphist) = UVMHIST_INITIALIZER(maphist, maphistbuf);
111 #endif
112 
113 #if !defined(UVMMAP_COUNTERS)
114 
115 #define	UVMMAP_EVCNT_DEFINE(name)	/* nothing */
116 #define UVMMAP_EVCNT_INCR(ev)		/* nothing */
117 #define UVMMAP_EVCNT_DECR(ev)		/* nothing */
118 
119 #else /* defined(UVMMAP_NOCOUNTERS) */
120 
121 #include <sys/evcnt.h>
122 #define	UVMMAP_EVCNT_DEFINE(name) \
123 struct evcnt uvmmap_evcnt_##name = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, \
124     "uvmmap", #name); \
125 EVCNT_ATTACH_STATIC(uvmmap_evcnt_##name);
126 #define	UVMMAP_EVCNT_INCR(ev)		uvmmap_evcnt_##ev.ev_count++
127 #define	UVMMAP_EVCNT_DECR(ev)		uvmmap_evcnt_##ev.ev_count--
128 
129 #endif /* defined(UVMMAP_NOCOUNTERS) */
130 
131 UVMMAP_EVCNT_DEFINE(ubackmerge)
132 UVMMAP_EVCNT_DEFINE(uforwmerge)
133 UVMMAP_EVCNT_DEFINE(ubimerge)
134 UVMMAP_EVCNT_DEFINE(unomerge)
135 UVMMAP_EVCNT_DEFINE(kbackmerge)
136 UVMMAP_EVCNT_DEFINE(kforwmerge)
137 UVMMAP_EVCNT_DEFINE(kbimerge)
138 UVMMAP_EVCNT_DEFINE(knomerge)
139 UVMMAP_EVCNT_DEFINE(map_call)
140 UVMMAP_EVCNT_DEFINE(mlk_call)
141 UVMMAP_EVCNT_DEFINE(mlk_hint)
142 UVMMAP_EVCNT_DEFINE(mlk_list)
143 UVMMAP_EVCNT_DEFINE(mlk_tree)
144 UVMMAP_EVCNT_DEFINE(mlk_treeloop)
145 UVMMAP_EVCNT_DEFINE(mlk_listloop)
146 
147 const char vmmapbsy[] = "vmmapbsy";
148 
149 /*
150  * cache for vmspace structures.
151  */
152 
153 static struct pool_cache uvm_vmspace_cache;
154 
155 /*
156  * cache for dynamically-allocated map entries.
157  */
158 
159 static struct pool_cache uvm_map_entry_cache;
160 
161 #ifdef PMAP_GROWKERNEL
162 /*
163  * This global represents the end of the kernel virtual address
164  * space.  If we want to exceed this, we must grow the kernel
165  * virtual address space dynamically.
166  *
167  * Note, this variable is locked by kernel_map's lock.
168  */
169 vaddr_t uvm_maxkaddr;
170 #endif
171 
172 #ifndef __USER_VA0_IS_SAFE
173 #ifndef __USER_VA0_DISABLE_DEFAULT
174 #define __USER_VA0_DISABLE_DEFAULT 1
175 #endif
176 #ifdef USER_VA0_DISABLE_DEFAULT /* kernel config option overrides */
177 #undef __USER_VA0_DISABLE_DEFAULT
178 #define __USER_VA0_DISABLE_DEFAULT USER_VA0_DISABLE_DEFAULT
179 #endif
180 int user_va0_disable = __USER_VA0_DISABLE_DEFAULT;
181 #endif
182 
183 /*
184  * macros
185  */
186 
187 /*
188  * UVM_ET_ISCOMPATIBLE: check some requirements for map entry merging
189  */
190 extern struct vm_map *pager_map;
191 
192 #define	UVM_ET_ISCOMPATIBLE(ent, type, uobj, meflags, \
193     prot, maxprot, inh, adv, wire) \
194 	((ent)->etype == (type) && \
195 	(((ent)->flags ^ (meflags)) & (UVM_MAP_NOMERGE)) == 0 && \
196 	(ent)->object.uvm_obj == (uobj) && \
197 	(ent)->protection == (prot) && \
198 	(ent)->max_protection == (maxprot) && \
199 	(ent)->inheritance == (inh) && \
200 	(ent)->advice == (adv) && \
201 	(ent)->wired_count == (wire))
202 
203 /*
204  * uvm_map_entry_link: insert entry into a map
205  *
206  * => map must be locked
207  */
208 #define uvm_map_entry_link(map, after_where, entry) do { \
209 	uvm_mapent_check(entry); \
210 	(map)->nentries++; \
211 	(entry)->prev = (after_where); \
212 	(entry)->next = (after_where)->next; \
213 	(entry)->prev->next = (entry); \
214 	(entry)->next->prev = (entry); \
215 	uvm_rb_insert((map), (entry)); \
216 } while (/*CONSTCOND*/ 0)
217 
218 /*
219  * uvm_map_entry_unlink: remove entry from a map
220  *
221  * => map must be locked
222  */
223 #define uvm_map_entry_unlink(map, entry) do { \
224 	KASSERT((entry) != (map)->first_free); \
225 	KASSERT((entry) != (map)->hint); \
226 	uvm_mapent_check(entry); \
227 	(map)->nentries--; \
228 	(entry)->next->prev = (entry)->prev; \
229 	(entry)->prev->next = (entry)->next; \
230 	uvm_rb_remove((map), (entry)); \
231 } while (/*CONSTCOND*/ 0)
232 
233 /*
234  * SAVE_HINT: saves the specified entry as the hint for future lookups.
235  *
236  * => map need not be locked.
237  */
238 #define SAVE_HINT(map, check, value) do { \
239 	if ((map)->hint == (check)) \
240 		(map)->hint = (value); \
241 } while (/*CONSTCOND*/ 0)
242 
243 /*
244  * clear_hints: ensure that hints don't point to the entry.
245  *
246  * => map must be write-locked.
247  */
248 static void
249 clear_hints(struct vm_map *map, struct vm_map_entry *ent)
250 {
251 
252 	SAVE_HINT(map, ent, ent->prev);
253 	if (map->first_free == ent) {
254 		map->first_free = ent->prev;
255 	}
256 }
257 
258 /*
259  * VM_MAP_RANGE_CHECK: check and correct range
260  *
261  * => map must at least be read locked
262  */
263 
264 #define VM_MAP_RANGE_CHECK(map, start, end) do { \
265 	if (start < vm_map_min(map))		\
266 		start = vm_map_min(map);	\
267 	if (end > vm_map_max(map))		\
268 		end = vm_map_max(map);		\
269 	if (start > end)			\
270 		start = end;			\
271 } while (/*CONSTCOND*/ 0)
272 
273 /*
274  * local prototypes
275  */
276 
277 static struct vm_map_entry *
278 		uvm_mapent_alloc(struct vm_map *, int);
279 static void	uvm_mapent_copy(struct vm_map_entry *, struct vm_map_entry *);
280 static void	uvm_mapent_free(struct vm_map_entry *);
281 #if defined(DEBUG)
282 static void	_uvm_mapent_check(const struct vm_map_entry *, const char *,
283 		    int);
284 #define	uvm_mapent_check(map)	_uvm_mapent_check(map, __FILE__, __LINE__)
285 #else /* defined(DEBUG) */
286 #define	uvm_mapent_check(e)	/* nothing */
287 #endif /* defined(DEBUG) */
288 
289 static void	uvm_map_entry_unwire(struct vm_map *, struct vm_map_entry *);
290 static void	uvm_map_reference_amap(struct vm_map_entry *, int);
291 static int	uvm_map_space_avail(vaddr_t *, vsize_t, voff_t, vsize_t, int,
292 		    int, struct vm_map_entry *);
293 static void	uvm_map_unreference_amap(struct vm_map_entry *, int);
294 
295 int _uvm_map_sanity(struct vm_map *);
296 int _uvm_tree_sanity(struct vm_map *);
297 static vsize_t uvm_rb_maxgap(const struct vm_map_entry *);
298 
299 #define	ROOT_ENTRY(map)		((struct vm_map_entry *)(map)->rb_tree.rbt_root)
300 #define	LEFT_ENTRY(entry)	((struct vm_map_entry *)(entry)->rb_node.rb_left)
301 #define	RIGHT_ENTRY(entry)	((struct vm_map_entry *)(entry)->rb_node.rb_right)
302 #define	PARENT_ENTRY(map, entry) \
303 	(ROOT_ENTRY(map) == (entry) \
304 	    ? NULL : (struct vm_map_entry *)RB_FATHER(&(entry)->rb_node))
305 
306 /*
307  * These get filled in if/when SYSVSHM shared memory code is loaded
308  *
309  * We do this with function pointers rather the #ifdef SYSVSHM so the
310  * SYSVSHM code can be loaded and unloaded
311  */
312 void (*uvm_shmexit)(struct vmspace *) = NULL;
313 void (*uvm_shmfork)(struct vmspace *, struct vmspace *) = NULL;
314 
315 static int
316 uvm_map_compare_nodes(void *ctx, const void *nparent, const void *nkey)
317 {
318 	const struct vm_map_entry *eparent = nparent;
319 	const struct vm_map_entry *ekey = nkey;
320 
321 	KASSERT(eparent->start < ekey->start || eparent->start >= ekey->end);
322 	KASSERT(ekey->start < eparent->start || ekey->start >= eparent->end);
323 
324 	if (eparent->start < ekey->start)
325 		return -1;
326 	if (eparent->end >= ekey->start)
327 		return 1;
328 	return 0;
329 }
330 
331 static int
332 uvm_map_compare_key(void *ctx, const void *nparent, const void *vkey)
333 {
334 	const struct vm_map_entry *eparent = nparent;
335 	const vaddr_t va = *(const vaddr_t *) vkey;
336 
337 	if (eparent->start < va)
338 		return -1;
339 	if (eparent->end >= va)
340 		return 1;
341 	return 0;
342 }
343 
344 static const rb_tree_ops_t uvm_map_tree_ops = {
345 	.rbto_compare_nodes = uvm_map_compare_nodes,
346 	.rbto_compare_key = uvm_map_compare_key,
347 	.rbto_node_offset = offsetof(struct vm_map_entry, rb_node),
348 	.rbto_context = NULL
349 };
350 
351 /*
352  * uvm_rb_gap: return the gap size between our entry and next entry.
353  */
354 static inline vsize_t
355 uvm_rb_gap(const struct vm_map_entry *entry)
356 {
357 
358 	KASSERT(entry->next != NULL);
359 	return entry->next->start - entry->end;
360 }
361 
362 static vsize_t
363 uvm_rb_maxgap(const struct vm_map_entry *entry)
364 {
365 	struct vm_map_entry *child;
366 	vsize_t maxgap = entry->gap;
367 
368 	/*
369 	 * We need maxgap to be the largest gap of us or any of our
370 	 * descendents.  Since each of our children's maxgap is the
371 	 * cached value of their largest gap of themselves or their
372 	 * descendents, we can just use that value and avoid recursing
373 	 * down the tree to calculate it.
374 	 */
375 	if ((child = LEFT_ENTRY(entry)) != NULL && maxgap < child->maxgap)
376 		maxgap = child->maxgap;
377 
378 	if ((child = RIGHT_ENTRY(entry)) != NULL && maxgap < child->maxgap)
379 		maxgap = child->maxgap;
380 
381 	return maxgap;
382 }
383 
384 static void
385 uvm_rb_fixup(struct vm_map *map, struct vm_map_entry *entry)
386 {
387 	struct vm_map_entry *parent;
388 
389 	KASSERT(entry->gap == uvm_rb_gap(entry));
390 	entry->maxgap = uvm_rb_maxgap(entry);
391 
392 	while ((parent = PARENT_ENTRY(map, entry)) != NULL) {
393 		struct vm_map_entry *brother;
394 		vsize_t maxgap = parent->gap;
395 		unsigned int which;
396 
397 		KDASSERT(parent->gap == uvm_rb_gap(parent));
398 		if (maxgap < entry->maxgap)
399 			maxgap = entry->maxgap;
400 		/*
401 		 * Since we work towards the root, we know entry's maxgap
402 		 * value is OK, but its brothers may now be out-of-date due
403 		 * to rebalancing.  So refresh it.
404 		 */
405 		which = RB_POSITION(&entry->rb_node) ^ RB_DIR_OTHER;
406 		brother = (struct vm_map_entry *)parent->rb_node.rb_nodes[which];
407 		if (brother != NULL) {
408 			KDASSERT(brother->gap == uvm_rb_gap(brother));
409 			brother->maxgap = uvm_rb_maxgap(brother);
410 			if (maxgap < brother->maxgap)
411 				maxgap = brother->maxgap;
412 		}
413 
414 		parent->maxgap = maxgap;
415 		entry = parent;
416 	}
417 }
418 
419 static void
420 uvm_rb_insert(struct vm_map *map, struct vm_map_entry *entry)
421 {
422 	struct vm_map_entry *ret __diagused;
423 
424 	entry->gap = entry->maxgap = uvm_rb_gap(entry);
425 	if (entry->prev != &map->header)
426 		entry->prev->gap = uvm_rb_gap(entry->prev);
427 
428 	ret = rb_tree_insert_node(&map->rb_tree, entry);
429 	KASSERTMSG(ret == entry,
430 	    "uvm_rb_insert: map %p: duplicate entry %p", map, ret);
431 
432 	/*
433 	 * If the previous entry is not our immediate left child, then it's an
434 	 * ancestor and will be fixed up on the way to the root.  We don't
435 	 * have to check entry->prev against &map->header since &map->header
436 	 * will never be in the tree.
437 	 */
438 	uvm_rb_fixup(map,
439 	    LEFT_ENTRY(entry) == entry->prev ? entry->prev : entry);
440 }
441 
442 static void
443 uvm_rb_remove(struct vm_map *map, struct vm_map_entry *entry)
444 {
445 	struct vm_map_entry *prev_parent = NULL, *next_parent = NULL;
446 
447 	/*
448 	 * If we are removing an interior node, then an adjacent node will
449 	 * be used to replace its position in the tree.  Therefore we will
450 	 * need to fixup the tree starting at the parent of the replacement
451 	 * node.  So record their parents for later use.
452 	 */
453 	if (entry->prev != &map->header)
454 		prev_parent = PARENT_ENTRY(map, entry->prev);
455 	if (entry->next != &map->header)
456 		next_parent = PARENT_ENTRY(map, entry->next);
457 
458 	rb_tree_remove_node(&map->rb_tree, entry);
459 
460 	/*
461 	 * If the previous node has a new parent, fixup the tree starting
462 	 * at the previous node's old parent.
463 	 */
464 	if (entry->prev != &map->header) {
465 		/*
466 		 * Update the previous entry's gap due to our absence.
467 		 */
468 		entry->prev->gap = uvm_rb_gap(entry->prev);
469 		uvm_rb_fixup(map, entry->prev);
470 		if (prev_parent != NULL
471 		    && prev_parent != entry
472 		    && prev_parent != PARENT_ENTRY(map, entry->prev))
473 			uvm_rb_fixup(map, prev_parent);
474 	}
475 
476 	/*
477 	 * If the next node has a new parent, fixup the tree starting
478 	 * at the next node's old parent.
479 	 */
480 	if (entry->next != &map->header) {
481 		uvm_rb_fixup(map, entry->next);
482 		if (next_parent != NULL
483 		    && next_parent != entry
484 		    && next_parent != PARENT_ENTRY(map, entry->next))
485 			uvm_rb_fixup(map, next_parent);
486 	}
487 }
488 
489 #if defined(DEBUG)
490 int uvm_debug_check_map = 0;
491 int uvm_debug_check_rbtree = 0;
492 #define uvm_map_check(map, name) \
493 	_uvm_map_check((map), (name), __FILE__, __LINE__)
494 static void
495 _uvm_map_check(struct vm_map *map, const char *name,
496     const char *file, int line)
497 {
498 
499 	if ((uvm_debug_check_map && _uvm_map_sanity(map)) ||
500 	    (uvm_debug_check_rbtree && _uvm_tree_sanity(map))) {
501 		panic("uvm_map_check failed: \"%s\" map=%p (%s:%d)",
502 		    name, map, file, line);
503 	}
504 }
505 #else /* defined(DEBUG) */
506 #define uvm_map_check(map, name)	/* nothing */
507 #endif /* defined(DEBUG) */
508 
509 #if defined(DEBUG) || defined(DDB)
510 int
511 _uvm_map_sanity(struct vm_map *map)
512 {
513 	bool first_free_found = false;
514 	bool hint_found = false;
515 	const struct vm_map_entry *e;
516 	struct vm_map_entry *hint = map->hint;
517 
518 	e = &map->header;
519 	for (;;) {
520 		if (map->first_free == e) {
521 			first_free_found = true;
522 		} else if (!first_free_found && e->next->start > e->end) {
523 			printf("first_free %p should be %p\n",
524 			    map->first_free, e);
525 			return -1;
526 		}
527 		if (hint == e) {
528 			hint_found = true;
529 		}
530 
531 		e = e->next;
532 		if (e == &map->header) {
533 			break;
534 		}
535 	}
536 	if (!first_free_found) {
537 		printf("stale first_free\n");
538 		return -1;
539 	}
540 	if (!hint_found) {
541 		printf("stale hint\n");
542 		return -1;
543 	}
544 	return 0;
545 }
546 
547 int
548 _uvm_tree_sanity(struct vm_map *map)
549 {
550 	struct vm_map_entry *tmp, *trtmp;
551 	int n = 0, i = 1;
552 
553 	for (tmp = map->header.next; tmp != &map->header; tmp = tmp->next) {
554 		if (tmp->gap != uvm_rb_gap(tmp)) {
555 			printf("%d/%d gap %#lx != %#lx %s\n",
556 			    n + 1, map->nentries,
557 			    (ulong)tmp->gap, (ulong)uvm_rb_gap(tmp),
558 			    tmp->next == &map->header ? "(last)" : "");
559 			goto error;
560 		}
561 		/*
562 		 * If any entries are out of order, tmp->gap will be unsigned
563 		 * and will likely exceed the size of the map.
564 		 */
565 		if (tmp->gap >= vm_map_max(map) - vm_map_min(map)) {
566 			printf("too large gap %zu\n", (size_t)tmp->gap);
567 			goto error;
568 		}
569 		n++;
570 	}
571 
572 	if (n != map->nentries) {
573 		printf("nentries: %d vs %d\n", n, map->nentries);
574 		goto error;
575 	}
576 
577 	trtmp = NULL;
578 	for (tmp = map->header.next; tmp != &map->header; tmp = tmp->next) {
579 		if (tmp->maxgap != uvm_rb_maxgap(tmp)) {
580 			printf("maxgap %#lx != %#lx\n",
581 			    (ulong)tmp->maxgap,
582 			    (ulong)uvm_rb_maxgap(tmp));
583 			goto error;
584 		}
585 		if (trtmp != NULL && trtmp->start >= tmp->start) {
586 			printf("corrupt: 0x%"PRIxVADDR"x >= 0x%"PRIxVADDR"x\n",
587 			    trtmp->start, tmp->start);
588 			goto error;
589 		}
590 
591 		trtmp = tmp;
592 	}
593 
594 	for (tmp = map->header.next; tmp != &map->header;
595 	    tmp = tmp->next, i++) {
596 		trtmp = rb_tree_iterate(&map->rb_tree, tmp, RB_DIR_LEFT);
597 		if (trtmp == NULL)
598 			trtmp = &map->header;
599 		if (tmp->prev != trtmp) {
600 			printf("lookup: %d: %p->prev=%p: %p\n",
601 			    i, tmp, tmp->prev, trtmp);
602 			goto error;
603 		}
604 		trtmp = rb_tree_iterate(&map->rb_tree, tmp, RB_DIR_RIGHT);
605 		if (trtmp == NULL)
606 			trtmp = &map->header;
607 		if (tmp->next != trtmp) {
608 			printf("lookup: %d: %p->next=%p: %p\n",
609 			    i, tmp, tmp->next, trtmp);
610 			goto error;
611 		}
612 		trtmp = rb_tree_find_node(&map->rb_tree, &tmp->start);
613 		if (trtmp != tmp) {
614 			printf("lookup: %d: %p - %p: %p\n", i, tmp, trtmp,
615 			    PARENT_ENTRY(map, tmp));
616 			goto error;
617 		}
618 	}
619 
620 	return (0);
621  error:
622 	return (-1);
623 }
624 #endif /* defined(DEBUG) || defined(DDB) */
625 
626 /*
627  * vm_map_lock: acquire an exclusive (write) lock on a map.
628  *
629  * => The locking protocol provides for guaranteed upgrade from shared ->
630  *    exclusive by whichever thread currently has the map marked busy.
631  *    See "LOCKING PROTOCOL NOTES" in uvm_map.h.  This is horrible; among
632  *    other problems, it defeats any fairness guarantees provided by RW
633  *    locks.
634  */
635 
636 void
637 vm_map_lock(struct vm_map *map)
638 {
639 
640 	for (;;) {
641 		rw_enter(&map->lock, RW_WRITER);
642 		if (map->busy == NULL || map->busy == curlwp) {
643 			break;
644 		}
645 		mutex_enter(&map->misc_lock);
646 		rw_exit(&map->lock);
647 		if (map->busy != NULL) {
648 			cv_wait(&map->cv, &map->misc_lock);
649 		}
650 		mutex_exit(&map->misc_lock);
651 	}
652 	map->timestamp++;
653 }
654 
655 /*
656  * vm_map_lock_try: try to lock a map, failing if it is already locked.
657  */
658 
659 bool
660 vm_map_lock_try(struct vm_map *map)
661 {
662 
663 	if (!rw_tryenter(&map->lock, RW_WRITER)) {
664 		return false;
665 	}
666 	if (map->busy != NULL) {
667 		rw_exit(&map->lock);
668 		return false;
669 	}
670 	map->timestamp++;
671 	return true;
672 }
673 
674 /*
675  * vm_map_unlock: release an exclusive lock on a map.
676  */
677 
678 void
679 vm_map_unlock(struct vm_map *map)
680 {
681 
682 	KASSERT(rw_write_held(&map->lock));
683 	KASSERT(map->busy == NULL || map->busy == curlwp);
684 	rw_exit(&map->lock);
685 }
686 
687 /*
688  * vm_map_unbusy: mark the map as unbusy, and wake any waiters that
689  *     want an exclusive lock.
690  */
691 
692 void
693 vm_map_unbusy(struct vm_map *map)
694 {
695 
696 	KASSERT(map->busy == curlwp);
697 
698 	/*
699 	 * Safe to clear 'busy' and 'waiters' with only a read lock held:
700 	 *
701 	 * o they can only be set with a write lock held
702 	 * o writers are blocked out with a read or write hold
703 	 * o at any time, only one thread owns the set of values
704 	 */
705 	mutex_enter(&map->misc_lock);
706 	map->busy = NULL;
707 	cv_broadcast(&map->cv);
708 	mutex_exit(&map->misc_lock);
709 }
710 
711 /*
712  * vm_map_lock_read: acquire a shared (read) lock on a map.
713  */
714 
715 void
716 vm_map_lock_read(struct vm_map *map)
717 {
718 
719 	rw_enter(&map->lock, RW_READER);
720 }
721 
722 /*
723  * vm_map_unlock_read: release a shared lock on a map.
724  */
725 
726 void
727 vm_map_unlock_read(struct vm_map *map)
728 {
729 
730 	rw_exit(&map->lock);
731 }
732 
733 /*
734  * vm_map_busy: mark a map as busy.
735  *
736  * => the caller must hold the map write locked
737  */
738 
739 void
740 vm_map_busy(struct vm_map *map)
741 {
742 
743 	KASSERT(rw_write_held(&map->lock));
744 	KASSERT(map->busy == NULL);
745 
746 	map->busy = curlwp;
747 }
748 
749 /*
750  * vm_map_locked_p: return true if the map is write locked.
751  *
752  * => only for debug purposes like KASSERTs.
753  * => should not be used to verify that a map is not locked.
754  */
755 
756 bool
757 vm_map_locked_p(struct vm_map *map)
758 {
759 
760 	return rw_write_held(&map->lock);
761 }
762 
763 /*
764  * uvm_mapent_alloc: allocate a map entry
765  */
766 
767 static struct vm_map_entry *
768 uvm_mapent_alloc(struct vm_map *map, int flags)
769 {
770 	struct vm_map_entry *me;
771 	int pflags = (flags & UVM_FLAG_NOWAIT) ? PR_NOWAIT : PR_WAITOK;
772 	UVMHIST_FUNC("uvm_mapent_alloc"); UVMHIST_CALLED(maphist);
773 
774 	me = pool_cache_get(&uvm_map_entry_cache, pflags);
775 	if (__predict_false(me == NULL)) {
776 		return NULL;
777 	}
778 	me->flags = 0;
779 
780 	UVMHIST_LOG(maphist, "<- new entry=%p [kentry=%d]", me,
781 	    (map == kernel_map), 0, 0);
782 	return me;
783 }
784 
785 /*
786  * uvm_mapent_free: free map entry
787  */
788 
789 static void
790 uvm_mapent_free(struct vm_map_entry *me)
791 {
792 	UVMHIST_FUNC("uvm_mapent_free"); UVMHIST_CALLED(maphist);
793 
794 	UVMHIST_LOG(maphist,"<- freeing map entry=%p [flags=%d]",
795 		me, me->flags, 0, 0);
796 	pool_cache_put(&uvm_map_entry_cache, me);
797 }
798 
799 /*
800  * uvm_mapent_copy: copy a map entry, preserving flags
801  */
802 
803 static inline void
804 uvm_mapent_copy(struct vm_map_entry *src, struct vm_map_entry *dst)
805 {
806 
807 	memcpy(dst, src, ((char *)&src->uvm_map_entry_stop_copy) -
808 	    ((char *)src));
809 }
810 
811 #if defined(DEBUG)
812 static void
813 _uvm_mapent_check(const struct vm_map_entry *entry, const char *file, int line)
814 {
815 
816 	if (entry->start >= entry->end) {
817 		goto bad;
818 	}
819 	if (UVM_ET_ISOBJ(entry)) {
820 		if (entry->object.uvm_obj == NULL) {
821 			goto bad;
822 		}
823 	} else if (UVM_ET_ISSUBMAP(entry)) {
824 		if (entry->object.sub_map == NULL) {
825 			goto bad;
826 		}
827 	} else {
828 		if (entry->object.uvm_obj != NULL ||
829 		    entry->object.sub_map != NULL) {
830 			goto bad;
831 		}
832 	}
833 	if (!UVM_ET_ISOBJ(entry)) {
834 		if (entry->offset != 0) {
835 			goto bad;
836 		}
837 	}
838 
839 	return;
840 
841 bad:
842 	panic("%s: bad entry %p (%s:%d)", __func__, entry, file, line);
843 }
844 #endif /* defined(DEBUG) */
845 
846 /*
847  * uvm_map_entry_unwire: unwire a map entry
848  *
849  * => map should be locked by caller
850  */
851 
852 static inline void
853 uvm_map_entry_unwire(struct vm_map *map, struct vm_map_entry *entry)
854 {
855 
856 	entry->wired_count = 0;
857 	uvm_fault_unwire_locked(map, entry->start, entry->end);
858 }
859 
860 
861 /*
862  * wrapper for calling amap_ref()
863  */
864 static inline void
865 uvm_map_reference_amap(struct vm_map_entry *entry, int flags)
866 {
867 
868 	amap_ref(entry->aref.ar_amap, entry->aref.ar_pageoff,
869 	    (entry->end - entry->start) >> PAGE_SHIFT, flags);
870 }
871 
872 
873 /*
874  * wrapper for calling amap_unref()
875  */
876 static inline void
877 uvm_map_unreference_amap(struct vm_map_entry *entry, int flags)
878 {
879 
880 	amap_unref(entry->aref.ar_amap, entry->aref.ar_pageoff,
881 	    (entry->end - entry->start) >> PAGE_SHIFT, flags);
882 }
883 
884 
885 /*
886  * uvm_map_init: init mapping system at boot time.
887  */
888 
889 void
890 uvm_map_init(void)
891 {
892 #if defined(UVMHIST)
893 	static struct kern_history_ent pdhistbuf[UVMHIST_PDHIST_SIZE];
894 #endif
895 
896 	/*
897 	 * first, init logging system.
898 	 */
899 
900 	UVMHIST_FUNC("uvm_map_init");
901 	UVMHIST_LINK_STATIC(maphist);
902 	UVMHIST_INIT_STATIC(pdhist, pdhistbuf);
903 	UVMHIST_CALLED(maphist);
904 	UVMHIST_LOG(maphist,"<starting uvm map system>", 0, 0, 0, 0);
905 
906 	/*
907 	 * initialize the global lock for kernel map entry.
908 	 */
909 
910 	mutex_init(&uvm_kentry_lock, MUTEX_DRIVER, IPL_VM);
911 }
912 
913 /*
914  * uvm_map_init_caches: init mapping system caches.
915  */
916 void
917 uvm_map_init_caches(void)
918 {
919 	/*
920 	 * initialize caches.
921 	 */
922 
923 	pool_cache_bootstrap(&uvm_map_entry_cache, sizeof(struct vm_map_entry),
924 	    0, 0, 0, "vmmpepl", NULL, IPL_NONE, NULL, NULL, NULL);
925 	pool_cache_bootstrap(&uvm_vmspace_cache, sizeof(struct vmspace),
926 	    0, 0, 0, "vmsppl", NULL, IPL_NONE, NULL, NULL, NULL);
927 }
928 
929 /*
930  * clippers
931  */
932 
933 /*
934  * uvm_mapent_splitadj: adjust map entries for splitting, after uvm_mapent_copy.
935  */
936 
937 static void
938 uvm_mapent_splitadj(struct vm_map_entry *entry1, struct vm_map_entry *entry2,
939     vaddr_t splitat)
940 {
941 	vaddr_t adj;
942 
943 	KASSERT(entry1->start < splitat);
944 	KASSERT(splitat < entry1->end);
945 
946 	adj = splitat - entry1->start;
947 	entry1->end = entry2->start = splitat;
948 
949 	if (entry1->aref.ar_amap) {
950 		amap_splitref(&entry1->aref, &entry2->aref, adj);
951 	}
952 	if (UVM_ET_ISSUBMAP(entry1)) {
953 		/* ... unlikely to happen, but play it safe */
954 		 uvm_map_reference(entry1->object.sub_map);
955 	} else if (UVM_ET_ISOBJ(entry1)) {
956 		KASSERT(entry1->object.uvm_obj != NULL); /* suppress coverity */
957 		entry2->offset += adj;
958 		if (entry1->object.uvm_obj->pgops &&
959 		    entry1->object.uvm_obj->pgops->pgo_reference)
960 			entry1->object.uvm_obj->pgops->pgo_reference(
961 			    entry1->object.uvm_obj);
962 	}
963 }
964 
965 /*
966  * uvm_map_clip_start: ensure that the entry begins at or after
967  *	the starting address, if it doesn't we split the entry.
968  *
969  * => caller should use UVM_MAP_CLIP_START macro rather than calling
970  *    this directly
971  * => map must be locked by caller
972  */
973 
974 void
975 uvm_map_clip_start(struct vm_map *map, struct vm_map_entry *entry,
976     vaddr_t start)
977 {
978 	struct vm_map_entry *new_entry;
979 
980 	/* uvm_map_simplify_entry(map, entry); */ /* XXX */
981 
982 	uvm_map_check(map, "clip_start entry");
983 	uvm_mapent_check(entry);
984 
985 	/*
986 	 * Split off the front portion.  note that we must insert the new
987 	 * entry BEFORE this one, so that this entry has the specified
988 	 * starting address.
989 	 */
990 	new_entry = uvm_mapent_alloc(map, 0);
991 	uvm_mapent_copy(entry, new_entry); /* entry -> new_entry */
992 	uvm_mapent_splitadj(new_entry, entry, start);
993 	uvm_map_entry_link(map, entry->prev, new_entry);
994 
995 	uvm_map_check(map, "clip_start leave");
996 }
997 
998 /*
999  * uvm_map_clip_end: ensure that the entry ends at or before
1000  *	the ending address, if it does't we split the reference
1001  *
1002  * => caller should use UVM_MAP_CLIP_END macro rather than calling
1003  *    this directly
1004  * => map must be locked by caller
1005  */
1006 
1007 void
1008 uvm_map_clip_end(struct vm_map *map, struct vm_map_entry *entry, vaddr_t end)
1009 {
1010 	struct vm_map_entry *new_entry;
1011 
1012 	uvm_map_check(map, "clip_end entry");
1013 	uvm_mapent_check(entry);
1014 
1015 	/*
1016 	 *	Create a new entry and insert it
1017 	 *	AFTER the specified entry
1018 	 */
1019 	new_entry = uvm_mapent_alloc(map, 0);
1020 	uvm_mapent_copy(entry, new_entry); /* entry -> new_entry */
1021 	uvm_mapent_splitadj(entry, new_entry, end);
1022 	uvm_map_entry_link(map, entry, new_entry);
1023 
1024 	uvm_map_check(map, "clip_end leave");
1025 }
1026 
1027 /*
1028  *   M A P   -   m a i n   e n t r y   p o i n t
1029  */
1030 /*
1031  * uvm_map: establish a valid mapping in a map
1032  *
1033  * => assume startp is page aligned.
1034  * => assume size is a multiple of PAGE_SIZE.
1035  * => assume sys_mmap provides enough of a "hint" to have us skip
1036  *	over text/data/bss area.
1037  * => map must be unlocked (we will lock it)
1038  * => <uobj,uoffset> value meanings (4 cases):
1039  *	 [1] <NULL,uoffset>		== uoffset is a hint for PMAP_PREFER
1040  *	 [2] <NULL,UVM_UNKNOWN_OFFSET>	== don't PMAP_PREFER
1041  *	 [3] <uobj,uoffset>		== normal mapping
1042  *	 [4] <uobj,UVM_UNKNOWN_OFFSET>	== uvm_map finds offset based on VA
1043  *
1044  *    case [4] is for kernel mappings where we don't know the offset until
1045  *    we've found a virtual address.   note that kernel object offsets are
1046  *    always relative to vm_map_min(kernel_map).
1047  *
1048  * => if `align' is non-zero, we align the virtual address to the specified
1049  *	alignment.
1050  *	this is provided as a mechanism for large pages.
1051  *
1052  * => XXXCDC: need way to map in external amap?
1053  */
1054 
1055 int
1056 uvm_map(struct vm_map *map, vaddr_t *startp /* IN/OUT */, vsize_t size,
1057     struct uvm_object *uobj, voff_t uoffset, vsize_t align, uvm_flag_t flags)
1058 {
1059 	struct uvm_map_args args;
1060 	struct vm_map_entry *new_entry;
1061 	int error;
1062 
1063 	KASSERT((size & PAGE_MASK) == 0);
1064 
1065 #ifndef __USER_VA0_IS_SAFE
1066 	if ((flags & UVM_FLAG_FIXED) && *startp == 0 &&
1067 	    !VM_MAP_IS_KERNEL(map) && user_va0_disable)
1068 		return EACCES;
1069 #endif
1070 
1071 	/*
1072 	 * for pager_map, allocate the new entry first to avoid sleeping
1073 	 * for memory while we have the map locked.
1074 	 */
1075 
1076 	new_entry = NULL;
1077 	if (map == pager_map) {
1078 		new_entry = uvm_mapent_alloc(map, (flags & UVM_FLAG_NOWAIT));
1079 		if (__predict_false(new_entry == NULL))
1080 			return ENOMEM;
1081 	}
1082 	if (map == pager_map)
1083 		flags |= UVM_FLAG_NOMERGE;
1084 
1085 	error = uvm_map_prepare(map, *startp, size, uobj, uoffset, align,
1086 	    flags, &args);
1087 	if (!error) {
1088 		error = uvm_map_enter(map, &args, new_entry);
1089 		*startp = args.uma_start;
1090 	} else if (new_entry) {
1091 		uvm_mapent_free(new_entry);
1092 	}
1093 
1094 #if defined(DEBUG)
1095 	if (!error && VM_MAP_IS_KERNEL(map) && (flags & UVM_FLAG_NOWAIT) == 0) {
1096 		uvm_km_check_empty(map, *startp, *startp + size);
1097 	}
1098 #endif /* defined(DEBUG) */
1099 
1100 	return error;
1101 }
1102 
1103 /*
1104  * uvm_map_prepare:
1105  *
1106  * called with map unlocked.
1107  * on success, returns the map locked.
1108  */
1109 
1110 int
1111 uvm_map_prepare(struct vm_map *map, vaddr_t start, vsize_t size,
1112     struct uvm_object *uobj, voff_t uoffset, vsize_t align, uvm_flag_t flags,
1113     struct uvm_map_args *args)
1114 {
1115 	struct vm_map_entry *prev_entry;
1116 	vm_prot_t prot = UVM_PROTECTION(flags);
1117 	vm_prot_t maxprot = UVM_MAXPROTECTION(flags);
1118 
1119 	UVMHIST_FUNC("uvm_map_prepare");
1120 	UVMHIST_CALLED(maphist);
1121 
1122 	UVMHIST_LOG(maphist, "(map=%p, start=%#lx, size=%lu, flags=%#x)",
1123 	    map, start, size, flags);
1124 	UVMHIST_LOG(maphist, "  uobj/offset %p/%ld", uobj, uoffset,0,0);
1125 
1126 	/*
1127 	 * detect a popular device driver bug.
1128 	 */
1129 
1130 	KASSERT(doing_shutdown || curlwp != NULL);
1131 
1132 	/*
1133 	 * zero-sized mapping doesn't make any sense.
1134 	 */
1135 	KASSERT(size > 0);
1136 
1137 	KASSERT((~flags & (UVM_FLAG_NOWAIT | UVM_FLAG_WAITVA)) != 0);
1138 
1139 	uvm_map_check(map, "map entry");
1140 
1141 	/*
1142 	 * check sanity of protection code
1143 	 */
1144 
1145 	if ((prot & maxprot) != prot) {
1146 		UVMHIST_LOG(maphist, "<- prot. failure:  prot=%#x, max=%#x",
1147 		prot, maxprot,0,0);
1148 		return EACCES;
1149 	}
1150 
1151 	/*
1152 	 * figure out where to put new VM range
1153 	 */
1154 retry:
1155 	if (vm_map_lock_try(map) == false) {
1156 		if ((flags & UVM_FLAG_TRYLOCK) != 0) {
1157 			return EAGAIN;
1158 		}
1159 		vm_map_lock(map); /* could sleep here */
1160 	}
1161 	prev_entry = uvm_map_findspace(map, start, size, &start,
1162 	    uobj, uoffset, align, flags);
1163 	if (prev_entry == NULL) {
1164 		unsigned int timestamp;
1165 
1166 		timestamp = map->timestamp;
1167 		UVMHIST_LOG(maphist,"waiting va timestamp=%#x",
1168 			    timestamp,0,0,0);
1169 		map->flags |= VM_MAP_WANTVA;
1170 		vm_map_unlock(map);
1171 
1172 		/*
1173 		 * try to reclaim kva and wait until someone does unmap.
1174 		 * fragile locking here, so we awaken every second to
1175 		 * recheck the condition.
1176 		 */
1177 
1178 		mutex_enter(&map->misc_lock);
1179 		while ((map->flags & VM_MAP_WANTVA) != 0 &&
1180 		   map->timestamp == timestamp) {
1181 			if ((flags & UVM_FLAG_WAITVA) == 0) {
1182 				mutex_exit(&map->misc_lock);
1183 				UVMHIST_LOG(maphist,
1184 				    "<- uvm_map_findspace failed!", 0,0,0,0);
1185 				return ENOMEM;
1186 			} else {
1187 				cv_timedwait(&map->cv, &map->misc_lock, hz);
1188 			}
1189 		}
1190 		mutex_exit(&map->misc_lock);
1191 		goto retry;
1192 	}
1193 
1194 #ifdef PMAP_GROWKERNEL
1195 	/*
1196 	 * If the kernel pmap can't map the requested space,
1197 	 * then allocate more resources for it.
1198 	 */
1199 	if (map == kernel_map && uvm_maxkaddr < (start + size))
1200 		uvm_maxkaddr = pmap_growkernel(start + size);
1201 #endif
1202 
1203 	UVMMAP_EVCNT_INCR(map_call);
1204 
1205 	/*
1206 	 * if uobj is null, then uoffset is either a VAC hint for PMAP_PREFER
1207 	 * [typically from uvm_map_reserve] or it is UVM_UNKNOWN_OFFSET.   in
1208 	 * either case we want to zero it  before storing it in the map entry
1209 	 * (because it looks strange and confusing when debugging...)
1210 	 *
1211 	 * if uobj is not null
1212 	 *   if uoffset is not UVM_UNKNOWN_OFFSET then we have a normal mapping
1213 	 *      and we do not need to change uoffset.
1214 	 *   if uoffset is UVM_UNKNOWN_OFFSET then we need to find the offset
1215 	 *      now (based on the starting address of the map).   this case is
1216 	 *      for kernel object mappings where we don't know the offset until
1217 	 *      the virtual address is found (with uvm_map_findspace).   the
1218 	 *      offset is the distance we are from the start of the map.
1219 	 */
1220 
1221 	if (uobj == NULL) {
1222 		uoffset = 0;
1223 	} else {
1224 		if (uoffset == UVM_UNKNOWN_OFFSET) {
1225 			KASSERT(UVM_OBJ_IS_KERN_OBJECT(uobj));
1226 			uoffset = start - vm_map_min(kernel_map);
1227 		}
1228 	}
1229 
1230 	args->uma_flags = flags;
1231 	args->uma_prev = prev_entry;
1232 	args->uma_start = start;
1233 	args->uma_size = size;
1234 	args->uma_uobj = uobj;
1235 	args->uma_uoffset = uoffset;
1236 
1237 	UVMHIST_LOG(maphist, "<- done!", 0,0,0,0);
1238 	return 0;
1239 }
1240 
1241 /*
1242  * uvm_map_enter:
1243  *
1244  * called with map locked.
1245  * unlock the map before returning.
1246  */
1247 
1248 int
1249 uvm_map_enter(struct vm_map *map, const struct uvm_map_args *args,
1250     struct vm_map_entry *new_entry)
1251 {
1252 	struct vm_map_entry *prev_entry = args->uma_prev;
1253 	struct vm_map_entry *dead = NULL;
1254 
1255 	const uvm_flag_t flags = args->uma_flags;
1256 	const vm_prot_t prot = UVM_PROTECTION(flags);
1257 	const vm_prot_t maxprot = UVM_MAXPROTECTION(flags);
1258 	const vm_inherit_t inherit = UVM_INHERIT(flags);
1259 	const int amapwaitflag = (flags & UVM_FLAG_NOWAIT) ?
1260 	    AMAP_EXTEND_NOWAIT : 0;
1261 	const int advice = UVM_ADVICE(flags);
1262 
1263 	vaddr_t start = args->uma_start;
1264 	vsize_t size = args->uma_size;
1265 	struct uvm_object *uobj = args->uma_uobj;
1266 	voff_t uoffset = args->uma_uoffset;
1267 
1268 	const int kmap = (vm_map_pmap(map) == pmap_kernel());
1269 	int merged = 0;
1270 	int error;
1271 	int newetype;
1272 
1273 	UVMHIST_FUNC("uvm_map_enter");
1274 	UVMHIST_CALLED(maphist);
1275 
1276 	UVMHIST_LOG(maphist, "(map=%p, start=%#lx, size=%lu, flags=%#x)",
1277 	    map, start, size, flags);
1278 	UVMHIST_LOG(maphist, "  uobj/offset %p/%ld", uobj, uoffset,0,0);
1279 
1280 	KASSERT(map->hint == prev_entry); /* bimerge case assumes this */
1281 	KASSERT(vm_map_locked_p(map));
1282 
1283 	if (uobj)
1284 		newetype = UVM_ET_OBJ;
1285 	else
1286 		newetype = 0;
1287 
1288 	if (flags & UVM_FLAG_COPYONW) {
1289 		newetype |= UVM_ET_COPYONWRITE;
1290 		if ((flags & UVM_FLAG_OVERLAY) == 0)
1291 			newetype |= UVM_ET_NEEDSCOPY;
1292 	}
1293 
1294 	/*
1295 	 * try and insert in map by extending previous entry, if possible.
1296 	 * XXX: we don't try and pull back the next entry.   might be useful
1297 	 * for a stack, but we are currently allocating our stack in advance.
1298 	 */
1299 
1300 	if (flags & UVM_FLAG_NOMERGE)
1301 		goto nomerge;
1302 
1303 	if (prev_entry->end == start &&
1304 	    prev_entry != &map->header &&
1305 	    UVM_ET_ISCOMPATIBLE(prev_entry, newetype, uobj, 0,
1306 	    prot, maxprot, inherit, advice, 0)) {
1307 
1308 		if (uobj && prev_entry->offset +
1309 		    (prev_entry->end - prev_entry->start) != uoffset)
1310 			goto forwardmerge;
1311 
1312 		/*
1313 		 * can't extend a shared amap.  note: no need to lock amap to
1314 		 * look at refs since we don't care about its exact value.
1315 		 * if it is one (i.e. we have only reference) it will stay there
1316 		 */
1317 
1318 		if (prev_entry->aref.ar_amap &&
1319 		    amap_refs(prev_entry->aref.ar_amap) != 1) {
1320 			goto forwardmerge;
1321 		}
1322 
1323 		if (prev_entry->aref.ar_amap) {
1324 			error = amap_extend(prev_entry, size,
1325 			    amapwaitflag | AMAP_EXTEND_FORWARDS);
1326 			if (error)
1327 				goto nomerge;
1328 		}
1329 
1330 		if (kmap) {
1331 			UVMMAP_EVCNT_INCR(kbackmerge);
1332 		} else {
1333 			UVMMAP_EVCNT_INCR(ubackmerge);
1334 		}
1335 		UVMHIST_LOG(maphist,"  starting back merge", 0, 0, 0, 0);
1336 
1337 		/*
1338 		 * drop our reference to uobj since we are extending a reference
1339 		 * that we already have (the ref count can not drop to zero).
1340 		 */
1341 
1342 		if (uobj && uobj->pgops->pgo_detach)
1343 			uobj->pgops->pgo_detach(uobj);
1344 
1345 		/*
1346 		 * Now that we've merged the entries, note that we've grown
1347 		 * and our gap has shrunk.  Then fix the tree.
1348 		 */
1349 		prev_entry->end += size;
1350 		prev_entry->gap -= size;
1351 		uvm_rb_fixup(map, prev_entry);
1352 
1353 		uvm_map_check(map, "map backmerged");
1354 
1355 		UVMHIST_LOG(maphist,"<- done (via backmerge)!", 0, 0, 0, 0);
1356 		merged++;
1357 	}
1358 
1359 forwardmerge:
1360 	if (prev_entry->next->start == (start + size) &&
1361 	    prev_entry->next != &map->header &&
1362 	    UVM_ET_ISCOMPATIBLE(prev_entry->next, newetype, uobj, 0,
1363 	    prot, maxprot, inherit, advice, 0)) {
1364 
1365 		if (uobj && prev_entry->next->offset != uoffset + size)
1366 			goto nomerge;
1367 
1368 		/*
1369 		 * can't extend a shared amap.  note: no need to lock amap to
1370 		 * look at refs since we don't care about its exact value.
1371 		 * if it is one (i.e. we have only reference) it will stay there.
1372 		 *
1373 		 * note that we also can't merge two amaps, so if we
1374 		 * merged with the previous entry which has an amap,
1375 		 * and the next entry also has an amap, we give up.
1376 		 *
1377 		 * Interesting cases:
1378 		 * amap, new, amap -> give up second merge (single fwd extend)
1379 		 * amap, new, none -> double forward extend (extend again here)
1380 		 * none, new, amap -> double backward extend (done here)
1381 		 * uobj, new, amap -> single backward extend (done here)
1382 		 *
1383 		 * XXX should we attempt to deal with someone refilling
1384 		 * the deallocated region between two entries that are
1385 		 * backed by the same amap (ie, arefs is 2, "prev" and
1386 		 * "next" refer to it, and adding this allocation will
1387 		 * close the hole, thus restoring arefs to 1 and
1388 		 * deallocating the "next" vm_map_entry)?  -- @@@
1389 		 */
1390 
1391 		if (prev_entry->next->aref.ar_amap &&
1392 		    (amap_refs(prev_entry->next->aref.ar_amap) != 1 ||
1393 		     (merged && prev_entry->aref.ar_amap))) {
1394 			goto nomerge;
1395 		}
1396 
1397 		if (merged) {
1398 			/*
1399 			 * Try to extend the amap of the previous entry to
1400 			 * cover the next entry as well.  If it doesn't work
1401 			 * just skip on, don't actually give up, since we've
1402 			 * already completed the back merge.
1403 			 */
1404 			if (prev_entry->aref.ar_amap) {
1405 				if (amap_extend(prev_entry,
1406 				    prev_entry->next->end -
1407 				    prev_entry->next->start,
1408 				    amapwaitflag | AMAP_EXTEND_FORWARDS))
1409 					goto nomerge;
1410 			}
1411 
1412 			/*
1413 			 * Try to extend the amap of the *next* entry
1414 			 * back to cover the new allocation *and* the
1415 			 * previous entry as well (the previous merge
1416 			 * didn't have an amap already otherwise we
1417 			 * wouldn't be checking here for an amap).  If
1418 			 * it doesn't work just skip on, again, don't
1419 			 * actually give up, since we've already
1420 			 * completed the back merge.
1421 			 */
1422 			else if (prev_entry->next->aref.ar_amap) {
1423 				if (amap_extend(prev_entry->next,
1424 				    prev_entry->end -
1425 				    prev_entry->start,
1426 				    amapwaitflag | AMAP_EXTEND_BACKWARDS))
1427 					goto nomerge;
1428 			}
1429 		} else {
1430 			/*
1431 			 * Pull the next entry's amap backwards to cover this
1432 			 * new allocation.
1433 			 */
1434 			if (prev_entry->next->aref.ar_amap) {
1435 				error = amap_extend(prev_entry->next, size,
1436 				    amapwaitflag | AMAP_EXTEND_BACKWARDS);
1437 				if (error)
1438 					goto nomerge;
1439 			}
1440 		}
1441 
1442 		if (merged) {
1443 			if (kmap) {
1444 				UVMMAP_EVCNT_DECR(kbackmerge);
1445 				UVMMAP_EVCNT_INCR(kbimerge);
1446 			} else {
1447 				UVMMAP_EVCNT_DECR(ubackmerge);
1448 				UVMMAP_EVCNT_INCR(ubimerge);
1449 			}
1450 		} else {
1451 			if (kmap) {
1452 				UVMMAP_EVCNT_INCR(kforwmerge);
1453 			} else {
1454 				UVMMAP_EVCNT_INCR(uforwmerge);
1455 			}
1456 		}
1457 		UVMHIST_LOG(maphist,"  starting forward merge", 0, 0, 0, 0);
1458 
1459 		/*
1460 		 * drop our reference to uobj since we are extending a reference
1461 		 * that we already have (the ref count can not drop to zero).
1462 		 */
1463 		if (uobj && uobj->pgops->pgo_detach)
1464 			uobj->pgops->pgo_detach(uobj);
1465 
1466 		if (merged) {
1467 			dead = prev_entry->next;
1468 			prev_entry->end = dead->end;
1469 			uvm_map_entry_unlink(map, dead);
1470 			if (dead->aref.ar_amap != NULL) {
1471 				prev_entry->aref = dead->aref;
1472 				dead->aref.ar_amap = NULL;
1473 			}
1474 		} else {
1475 			prev_entry->next->start -= size;
1476 			if (prev_entry != &map->header) {
1477 				prev_entry->gap -= size;
1478 				KASSERT(prev_entry->gap == uvm_rb_gap(prev_entry));
1479 				uvm_rb_fixup(map, prev_entry);
1480 			}
1481 			if (uobj)
1482 				prev_entry->next->offset = uoffset;
1483 		}
1484 
1485 		uvm_map_check(map, "map forwardmerged");
1486 
1487 		UVMHIST_LOG(maphist,"<- done forwardmerge", 0, 0, 0, 0);
1488 		merged++;
1489 	}
1490 
1491 nomerge:
1492 	if (!merged) {
1493 		UVMHIST_LOG(maphist,"  allocating new map entry", 0, 0, 0, 0);
1494 		if (kmap) {
1495 			UVMMAP_EVCNT_INCR(knomerge);
1496 		} else {
1497 			UVMMAP_EVCNT_INCR(unomerge);
1498 		}
1499 
1500 		/*
1501 		 * allocate new entry and link it in.
1502 		 */
1503 
1504 		if (new_entry == NULL) {
1505 			new_entry = uvm_mapent_alloc(map,
1506 				(flags & UVM_FLAG_NOWAIT));
1507 			if (__predict_false(new_entry == NULL)) {
1508 				error = ENOMEM;
1509 				goto done;
1510 			}
1511 		}
1512 		new_entry->start = start;
1513 		new_entry->end = new_entry->start + size;
1514 		new_entry->object.uvm_obj = uobj;
1515 		new_entry->offset = uoffset;
1516 
1517 		new_entry->etype = newetype;
1518 
1519 		if (flags & UVM_FLAG_NOMERGE) {
1520 			new_entry->flags |= UVM_MAP_NOMERGE;
1521 		}
1522 
1523 		new_entry->protection = prot;
1524 		new_entry->max_protection = maxprot;
1525 		new_entry->inheritance = inherit;
1526 		new_entry->wired_count = 0;
1527 		new_entry->advice = advice;
1528 		if (flags & UVM_FLAG_OVERLAY) {
1529 
1530 			/*
1531 			 * to_add: for BSS we overallocate a little since we
1532 			 * are likely to extend
1533 			 */
1534 
1535 			vaddr_t to_add = (flags & UVM_FLAG_AMAPPAD) ?
1536 				UVM_AMAP_CHUNK << PAGE_SHIFT : 0;
1537 			struct vm_amap *amap = amap_alloc(size, to_add,
1538 			    (flags & UVM_FLAG_NOWAIT));
1539 			if (__predict_false(amap == NULL)) {
1540 				error = ENOMEM;
1541 				goto done;
1542 			}
1543 			new_entry->aref.ar_pageoff = 0;
1544 			new_entry->aref.ar_amap = amap;
1545 		} else {
1546 			new_entry->aref.ar_pageoff = 0;
1547 			new_entry->aref.ar_amap = NULL;
1548 		}
1549 		uvm_map_entry_link(map, prev_entry, new_entry);
1550 
1551 		/*
1552 		 * Update the free space hint
1553 		 */
1554 
1555 		if ((map->first_free == prev_entry) &&
1556 		    (prev_entry->end >= new_entry->start))
1557 			map->first_free = new_entry;
1558 
1559 		new_entry = NULL;
1560 	}
1561 
1562 	map->size += size;
1563 
1564 	UVMHIST_LOG(maphist,"<- done!", 0, 0, 0, 0);
1565 
1566 	error = 0;
1567 done:
1568 	vm_map_unlock(map);
1569 
1570 	if (new_entry) {
1571 		uvm_mapent_free(new_entry);
1572 	}
1573 
1574 	if (dead) {
1575 		KDASSERT(merged);
1576 		uvm_mapent_free(dead);
1577 	}
1578 
1579 	return error;
1580 }
1581 
1582 /*
1583  * uvm_map_lookup_entry_bytree: lookup an entry in tree
1584  */
1585 
1586 static inline bool
1587 uvm_map_lookup_entry_bytree(struct vm_map *map, vaddr_t address,
1588     struct vm_map_entry **entry	/* OUT */)
1589 {
1590 	struct vm_map_entry *prev = &map->header;
1591 	struct vm_map_entry *cur = ROOT_ENTRY(map);
1592 
1593 	while (cur) {
1594 		UVMMAP_EVCNT_INCR(mlk_treeloop);
1595 		if (address >= cur->start) {
1596 			if (address < cur->end) {
1597 				*entry = cur;
1598 				return true;
1599 			}
1600 			prev = cur;
1601 			cur = RIGHT_ENTRY(cur);
1602 		} else
1603 			cur = LEFT_ENTRY(cur);
1604 	}
1605 	*entry = prev;
1606 	return false;
1607 }
1608 
1609 /*
1610  * uvm_map_lookup_entry: find map entry at or before an address
1611  *
1612  * => map must at least be read-locked by caller
1613  * => entry is returned in "entry"
1614  * => return value is true if address is in the returned entry
1615  */
1616 
1617 bool
1618 uvm_map_lookup_entry(struct vm_map *map, vaddr_t address,
1619     struct vm_map_entry **entry	/* OUT */)
1620 {
1621 	struct vm_map_entry *cur;
1622 	bool use_tree = false;
1623 	UVMHIST_FUNC("uvm_map_lookup_entry");
1624 	UVMHIST_CALLED(maphist);
1625 
1626 	UVMHIST_LOG(maphist,"(map=%p,addr=%#lx,ent=%p)",
1627 	    map, address, entry, 0);
1628 
1629 	/*
1630 	 * start looking either from the head of the
1631 	 * list, or from the hint.
1632 	 */
1633 
1634 	cur = map->hint;
1635 
1636 	if (cur == &map->header)
1637 		cur = cur->next;
1638 
1639 	UVMMAP_EVCNT_INCR(mlk_call);
1640 	if (address >= cur->start) {
1641 
1642 		/*
1643 		 * go from hint to end of list.
1644 		 *
1645 		 * but first, make a quick check to see if
1646 		 * we are already looking at the entry we
1647 		 * want (which is usually the case).
1648 		 * note also that we don't need to save the hint
1649 		 * here... it is the same hint (unless we are
1650 		 * at the header, in which case the hint didn't
1651 		 * buy us anything anyway).
1652 		 */
1653 
1654 		if (cur != &map->header && cur->end > address) {
1655 			UVMMAP_EVCNT_INCR(mlk_hint);
1656 			*entry = cur;
1657 			UVMHIST_LOG(maphist,"<- got it via hint (%p)",
1658 			    cur, 0, 0, 0);
1659 			uvm_mapent_check(*entry);
1660 			return (true);
1661 		}
1662 
1663 		if (map->nentries > 15)
1664 			use_tree = true;
1665 	} else {
1666 
1667 		/*
1668 		 * invalid hint.  use tree.
1669 		 */
1670 		use_tree = true;
1671 	}
1672 
1673 	uvm_map_check(map, __func__);
1674 
1675 	if (use_tree) {
1676 		/*
1677 		 * Simple lookup in the tree.  Happens when the hint is
1678 		 * invalid, or nentries reach a threshold.
1679 		 */
1680 		UVMMAP_EVCNT_INCR(mlk_tree);
1681 		if (uvm_map_lookup_entry_bytree(map, address, entry)) {
1682 			goto got;
1683 		} else {
1684 			goto failed;
1685 		}
1686 	}
1687 
1688 	/*
1689 	 * search linearly
1690 	 */
1691 
1692 	UVMMAP_EVCNT_INCR(mlk_list);
1693 	while (cur != &map->header) {
1694 		UVMMAP_EVCNT_INCR(mlk_listloop);
1695 		if (cur->end > address) {
1696 			if (address >= cur->start) {
1697 				/*
1698 				 * save this lookup for future
1699 				 * hints, and return
1700 				 */
1701 
1702 				*entry = cur;
1703 got:
1704 				SAVE_HINT(map, map->hint, *entry);
1705 				UVMHIST_LOG(maphist,"<- search got it (%p)",
1706 					cur, 0, 0, 0);
1707 				KDASSERT((*entry)->start <= address);
1708 				KDASSERT(address < (*entry)->end);
1709 				uvm_mapent_check(*entry);
1710 				return (true);
1711 			}
1712 			break;
1713 		}
1714 		cur = cur->next;
1715 	}
1716 	*entry = cur->prev;
1717 failed:
1718 	SAVE_HINT(map, map->hint, *entry);
1719 	UVMHIST_LOG(maphist,"<- failed!",0,0,0,0);
1720 	KDASSERT((*entry) == &map->header || (*entry)->end <= address);
1721 	KDASSERT((*entry)->next == &map->header ||
1722 	    address < (*entry)->next->start);
1723 	return (false);
1724 }
1725 
1726 /*
1727  * See if the range between start and start + length fits in the gap
1728  * entry->next->start and entry->end.  Returns 1 if fits, 0 if doesn't
1729  * fit, and -1 address wraps around.
1730  */
1731 static int
1732 uvm_map_space_avail(vaddr_t *start, vsize_t length, voff_t uoffset,
1733     vsize_t align, int flags, int topdown, struct vm_map_entry *entry)
1734 {
1735 	vaddr_t end;
1736 
1737 #ifdef PMAP_PREFER
1738 	/*
1739 	 * push start address forward as needed to avoid VAC alias problems.
1740 	 * we only do this if a valid offset is specified.
1741 	 */
1742 
1743 	if (uoffset != UVM_UNKNOWN_OFFSET)
1744 		PMAP_PREFER(uoffset, start, length, topdown);
1745 #endif
1746 	if ((flags & UVM_FLAG_COLORMATCH) != 0) {
1747 		KASSERT(align < uvmexp.ncolors);
1748 		if (uvmexp.ncolors > 1) {
1749 			const u_int colormask = uvmexp.colormask;
1750 			const u_int colorsize = colormask + 1;
1751 			vaddr_t hint = atop(*start);
1752 			const u_int color = hint & colormask;
1753 			if (color != align) {
1754 				hint -= color;	/* adjust to color boundary */
1755 				KASSERT((hint & colormask) == 0);
1756 				if (topdown) {
1757 					if (align > color)
1758 						hint -= colorsize;
1759 				} else {
1760 					if (align < color)
1761 						hint += colorsize;
1762 				}
1763 				*start = ptoa(hint + align); /* adjust to color */
1764 			}
1765 		}
1766 	} else if (align != 0) {
1767 		if ((*start & (align - 1)) != 0) {
1768 			if (topdown)
1769 				*start &= ~(align - 1);
1770 			else
1771 				*start = roundup(*start, align);
1772 		}
1773 		/*
1774 		 * XXX Should we PMAP_PREFER() here again?
1775 		 * eh...i think we're okay
1776 		 */
1777 	}
1778 
1779 	/*
1780 	 * Find the end of the proposed new region.  Be sure we didn't
1781 	 * wrap around the address; if so, we lose.  Otherwise, if the
1782 	 * proposed new region fits before the next entry, we win.
1783 	 */
1784 
1785 	end = *start + length;
1786 	if (end < *start)
1787 		return (-1);
1788 
1789 	if (entry->next->start >= end && *start >= entry->end)
1790 		return (1);
1791 
1792 	return (0);
1793 }
1794 
1795 /*
1796  * uvm_map_findspace: find "length" sized space in "map".
1797  *
1798  * => "hint" is a hint about where we want it, unless UVM_FLAG_FIXED is
1799  *	set in "flags" (in which case we insist on using "hint").
1800  * => "result" is VA returned
1801  * => uobj/uoffset are to be used to handle VAC alignment, if required
1802  * => if "align" is non-zero, we attempt to align to that value.
1803  * => caller must at least have read-locked map
1804  * => returns NULL on failure, or pointer to prev. map entry if success
1805  * => note this is a cross between the old vm_map_findspace and vm_map_find
1806  */
1807 
1808 struct vm_map_entry *
1809 uvm_map_findspace(struct vm_map *map, vaddr_t hint, vsize_t length,
1810     vaddr_t *result /* OUT */, struct uvm_object *uobj, voff_t uoffset,
1811     vsize_t align, int flags)
1812 {
1813 	struct vm_map_entry *entry;
1814 	struct vm_map_entry *child, *prev, *tmp;
1815 	vaddr_t orig_hint __diagused;
1816 	const int topdown = map->flags & VM_MAP_TOPDOWN;
1817 	UVMHIST_FUNC("uvm_map_findspace");
1818 	UVMHIST_CALLED(maphist);
1819 
1820 	UVMHIST_LOG(maphist, "(map=%p, hint=%l#x, len=%lu, flags=%#x)",
1821 	    map, hint, length, flags);
1822 	KASSERT((flags & UVM_FLAG_COLORMATCH) != 0 || (align & (align - 1)) == 0);
1823 	KASSERT((flags & UVM_FLAG_COLORMATCH) == 0 || align < uvmexp.ncolors);
1824 	KASSERT((flags & UVM_FLAG_FIXED) == 0 || align == 0);
1825 
1826 	uvm_map_check(map, "map_findspace entry");
1827 
1828 	/*
1829 	 * remember the original hint.  if we are aligning, then we
1830 	 * may have to try again with no alignment constraint if
1831 	 * we fail the first time.
1832 	 */
1833 
1834 	orig_hint = hint;
1835 	if (hint < vm_map_min(map)) {	/* check ranges ... */
1836 		if (flags & UVM_FLAG_FIXED) {
1837 			UVMHIST_LOG(maphist,"<- VA below map range",0,0,0,0);
1838 			return (NULL);
1839 		}
1840 		hint = vm_map_min(map);
1841 	}
1842 	if (hint > vm_map_max(map)) {
1843 		UVMHIST_LOG(maphist,"<- VA %#lx > range [%#lx->%#lx]",
1844 		    hint, vm_map_min(map), vm_map_max(map), 0);
1845 		return (NULL);
1846 	}
1847 
1848 	/*
1849 	 * Look for the first possible address; if there's already
1850 	 * something at this address, we have to start after it.
1851 	 */
1852 
1853 	/*
1854 	 * @@@: there are four, no, eight cases to consider.
1855 	 *
1856 	 * 0: found,     fixed,     bottom up -> fail
1857 	 * 1: found,     fixed,     top down  -> fail
1858 	 * 2: found,     not fixed, bottom up -> start after entry->end,
1859 	 *                                       loop up
1860 	 * 3: found,     not fixed, top down  -> start before entry->start,
1861 	 *                                       loop down
1862 	 * 4: not found, fixed,     bottom up -> check entry->next->start, fail
1863 	 * 5: not found, fixed,     top down  -> check entry->next->start, fail
1864 	 * 6: not found, not fixed, bottom up -> check entry->next->start,
1865 	 *                                       loop up
1866 	 * 7: not found, not fixed, top down  -> check entry->next->start,
1867 	 *                                       loop down
1868 	 *
1869 	 * as you can see, it reduces to roughly five cases, and that
1870 	 * adding top down mapping only adds one unique case (without
1871 	 * it, there would be four cases).
1872 	 */
1873 
1874 	if ((flags & UVM_FLAG_FIXED) == 0 && hint == vm_map_min(map)) {
1875 		entry = map->first_free;
1876 	} else {
1877 		if (uvm_map_lookup_entry(map, hint, &entry)) {
1878 			/* "hint" address already in use ... */
1879 			if (flags & UVM_FLAG_FIXED) {
1880 				UVMHIST_LOG(maphist, "<- fixed & VA in use",
1881 				    0, 0, 0, 0);
1882 				return (NULL);
1883 			}
1884 			if (topdown)
1885 				/* Start from lower gap. */
1886 				entry = entry->prev;
1887 		} else if (flags & UVM_FLAG_FIXED) {
1888 			if (entry->next->start >= hint + length &&
1889 			    hint + length > hint)
1890 				goto found;
1891 
1892 			/* "hint" address is gap but too small */
1893 			UVMHIST_LOG(maphist, "<- fixed mapping failed",
1894 			    0, 0, 0, 0);
1895 			return (NULL); /* only one shot at it ... */
1896 		} else {
1897 			/*
1898 			 * See if given hint fits in this gap.
1899 			 */
1900 			switch (uvm_map_space_avail(&hint, length,
1901 			    uoffset, align, flags, topdown, entry)) {
1902 			case 1:
1903 				goto found;
1904 			case -1:
1905 				goto wraparound;
1906 			}
1907 
1908 			if (topdown) {
1909 				/*
1910 				 * Still there is a chance to fit
1911 				 * if hint > entry->end.
1912 				 */
1913 			} else {
1914 				/* Start from higher gap. */
1915 				entry = entry->next;
1916 				if (entry == &map->header)
1917 					goto notfound;
1918 				goto nextgap;
1919 			}
1920 		}
1921 	}
1922 
1923 	/*
1924 	 * Note that all UVM_FLAGS_FIXED case is already handled.
1925 	 */
1926 	KDASSERT((flags & UVM_FLAG_FIXED) == 0);
1927 
1928 	/* Try to find the space in the red-black tree */
1929 
1930 	/* Check slot before any entry */
1931 	hint = topdown ? entry->next->start - length : entry->end;
1932 	switch (uvm_map_space_avail(&hint, length, uoffset, align, flags,
1933 	    topdown, entry)) {
1934 	case 1:
1935 		goto found;
1936 	case -1:
1937 		goto wraparound;
1938 	}
1939 
1940 nextgap:
1941 	KDASSERT((flags & UVM_FLAG_FIXED) == 0);
1942 	/* If there is not enough space in the whole tree, we fail */
1943 	tmp = ROOT_ENTRY(map);
1944 	if (tmp == NULL || tmp->maxgap < length)
1945 		goto notfound;
1946 
1947 	prev = NULL; /* previous candidate */
1948 
1949 	/* Find an entry close to hint that has enough space */
1950 	for (; tmp;) {
1951 		KASSERT(tmp->next->start == tmp->end + tmp->gap);
1952 		if (topdown) {
1953 			if (tmp->next->start < hint + length &&
1954 			    (prev == NULL || tmp->end > prev->end)) {
1955 				if (tmp->gap >= length)
1956 					prev = tmp;
1957 				else if ((child = LEFT_ENTRY(tmp)) != NULL
1958 				    && child->maxgap >= length)
1959 					prev = tmp;
1960 			}
1961 		} else {
1962 			if (tmp->end >= hint &&
1963 			    (prev == NULL || tmp->end < prev->end)) {
1964 				if (tmp->gap >= length)
1965 					prev = tmp;
1966 				else if ((child = RIGHT_ENTRY(tmp)) != NULL
1967 				    && child->maxgap >= length)
1968 					prev = tmp;
1969 			}
1970 		}
1971 		if (tmp->next->start < hint + length)
1972 			child = RIGHT_ENTRY(tmp);
1973 		else if (tmp->end > hint)
1974 			child = LEFT_ENTRY(tmp);
1975 		else {
1976 			if (tmp->gap >= length)
1977 				break;
1978 			if (topdown)
1979 				child = LEFT_ENTRY(tmp);
1980 			else
1981 				child = RIGHT_ENTRY(tmp);
1982 		}
1983 		if (child == NULL || child->maxgap < length)
1984 			break;
1985 		tmp = child;
1986 	}
1987 
1988 	if (tmp != NULL && tmp->start < hint && hint < tmp->next->start) {
1989 		/*
1990 		 * Check if the entry that we found satifies the
1991 		 * space requirement
1992 		 */
1993 		if (topdown) {
1994 			if (hint > tmp->next->start - length)
1995 				hint = tmp->next->start - length;
1996 		} else {
1997 			if (hint < tmp->end)
1998 				hint = tmp->end;
1999 		}
2000 		switch (uvm_map_space_avail(&hint, length, uoffset, align,
2001 		    flags, topdown, tmp)) {
2002 		case 1:
2003 			entry = tmp;
2004 			goto found;
2005 		case -1:
2006 			goto wraparound;
2007 		}
2008 		if (tmp->gap >= length)
2009 			goto listsearch;
2010 	}
2011 	if (prev == NULL)
2012 		goto notfound;
2013 
2014 	if (topdown) {
2015 		KASSERT(orig_hint >= prev->next->start - length ||
2016 		    prev->next->start - length > prev->next->start);
2017 		hint = prev->next->start - length;
2018 	} else {
2019 		KASSERT(orig_hint <= prev->end);
2020 		hint = prev->end;
2021 	}
2022 	switch (uvm_map_space_avail(&hint, length, uoffset, align,
2023 	    flags, topdown, prev)) {
2024 	case 1:
2025 		entry = prev;
2026 		goto found;
2027 	case -1:
2028 		goto wraparound;
2029 	}
2030 	if (prev->gap >= length)
2031 		goto listsearch;
2032 
2033 	if (topdown)
2034 		tmp = LEFT_ENTRY(prev);
2035 	else
2036 		tmp = RIGHT_ENTRY(prev);
2037 	for (;;) {
2038 		KASSERT(tmp && tmp->maxgap >= length);
2039 		if (topdown)
2040 			child = RIGHT_ENTRY(tmp);
2041 		else
2042 			child = LEFT_ENTRY(tmp);
2043 		if (child && child->maxgap >= length) {
2044 			tmp = child;
2045 			continue;
2046 		}
2047 		if (tmp->gap >= length)
2048 			break;
2049 		if (topdown)
2050 			tmp = LEFT_ENTRY(tmp);
2051 		else
2052 			tmp = RIGHT_ENTRY(tmp);
2053 	}
2054 
2055 	if (topdown) {
2056 		KASSERT(orig_hint >= tmp->next->start - length ||
2057 		    tmp->next->start - length > tmp->next->start);
2058 		hint = tmp->next->start - length;
2059 	} else {
2060 		KASSERT(orig_hint <= tmp->end);
2061 		hint = tmp->end;
2062 	}
2063 	switch (uvm_map_space_avail(&hint, length, uoffset, align,
2064 	    flags, topdown, tmp)) {
2065 	case 1:
2066 		entry = tmp;
2067 		goto found;
2068 	case -1:
2069 		goto wraparound;
2070 	}
2071 
2072 	/*
2073 	 * The tree fails to find an entry because of offset or alignment
2074 	 * restrictions.  Search the list instead.
2075 	 */
2076  listsearch:
2077 	/*
2078 	 * Look through the rest of the map, trying to fit a new region in
2079 	 * the gap between existing regions, or after the very last region.
2080 	 * note: entry->end = base VA of current gap,
2081 	 *	 entry->next->start = VA of end of current gap
2082 	 */
2083 
2084 	for (;;) {
2085 		/* Update hint for current gap. */
2086 		hint = topdown ? entry->next->start - length : entry->end;
2087 
2088 		/* See if it fits. */
2089 		switch (uvm_map_space_avail(&hint, length, uoffset, align,
2090 		    flags, topdown, entry)) {
2091 		case 1:
2092 			goto found;
2093 		case -1:
2094 			goto wraparound;
2095 		}
2096 
2097 		/* Advance to next/previous gap */
2098 		if (topdown) {
2099 			if (entry == &map->header) {
2100 				UVMHIST_LOG(maphist, "<- failed (off start)",
2101 				    0,0,0,0);
2102 				goto notfound;
2103 			}
2104 			entry = entry->prev;
2105 		} else {
2106 			entry = entry->next;
2107 			if (entry == &map->header) {
2108 				UVMHIST_LOG(maphist, "<- failed (off end)",
2109 				    0,0,0,0);
2110 				goto notfound;
2111 			}
2112 		}
2113 	}
2114 
2115  found:
2116 	SAVE_HINT(map, map->hint, entry);
2117 	*result = hint;
2118 	UVMHIST_LOG(maphist,"<- got it!  (result=%#lx)", hint, 0,0,0);
2119 	KASSERTMSG( topdown || hint >= orig_hint, "hint: %jx, orig_hint: %jx",
2120 	    (uintmax_t)hint, (uintmax_t)orig_hint);
2121 	KASSERTMSG(!topdown || hint <= orig_hint, "hint: %jx, orig_hint: %jx",
2122 	    (uintmax_t)hint, (uintmax_t)orig_hint);
2123 	KASSERT(entry->end <= hint);
2124 	KASSERT(hint + length <= entry->next->start);
2125 	return (entry);
2126 
2127  wraparound:
2128 	UVMHIST_LOG(maphist, "<- failed (wrap around)", 0,0,0,0);
2129 
2130 	return (NULL);
2131 
2132  notfound:
2133 	UVMHIST_LOG(maphist, "<- failed (notfound)", 0,0,0,0);
2134 
2135 	return (NULL);
2136 }
2137 
2138 /*
2139  *   U N M A P   -   m a i n   h e l p e r   f u n c t i o n s
2140  */
2141 
2142 /*
2143  * uvm_unmap_remove: remove mappings from a vm_map (from "start" up to "stop")
2144  *
2145  * => caller must check alignment and size
2146  * => map must be locked by caller
2147  * => we return a list of map entries that we've remove from the map
2148  *    in "entry_list"
2149  */
2150 
2151 void
2152 uvm_unmap_remove(struct vm_map *map, vaddr_t start, vaddr_t end,
2153     struct vm_map_entry **entry_list /* OUT */, int flags)
2154 {
2155 	struct vm_map_entry *entry, *first_entry, *next;
2156 	vaddr_t len;
2157 	UVMHIST_FUNC("uvm_unmap_remove"); UVMHIST_CALLED(maphist);
2158 
2159 	UVMHIST_LOG(maphist,"(map=%p, start=%#lx, end=%#lx)",
2160 	    map, start, end, 0);
2161 	VM_MAP_RANGE_CHECK(map, start, end);
2162 
2163 	uvm_map_check(map, "unmap_remove entry");
2164 
2165 	/*
2166 	 * find first entry
2167 	 */
2168 
2169 	if (uvm_map_lookup_entry(map, start, &first_entry) == true) {
2170 		/* clip and go... */
2171 		entry = first_entry;
2172 		UVM_MAP_CLIP_START(map, entry, start);
2173 		/* critical!  prevents stale hint */
2174 		SAVE_HINT(map, entry, entry->prev);
2175 	} else {
2176 		entry = first_entry->next;
2177 	}
2178 
2179 	/*
2180 	 * Save the free space hint
2181 	 */
2182 
2183 	if (map->first_free != &map->header && map->first_free->start >= start)
2184 		map->first_free = entry->prev;
2185 
2186 	/*
2187 	 * note: we now re-use first_entry for a different task.  we remove
2188 	 * a number of map entries from the map and save them in a linked
2189 	 * list headed by "first_entry".  once we remove them from the map
2190 	 * the caller should unlock the map and drop the references to the
2191 	 * backing objects [c.f. uvm_unmap_detach].  the object is to
2192 	 * separate unmapping from reference dropping.  why?
2193 	 *   [1] the map has to be locked for unmapping
2194 	 *   [2] the map need not be locked for reference dropping
2195 	 *   [3] dropping references may trigger pager I/O, and if we hit
2196 	 *       a pager that does synchronous I/O we may have to wait for it.
2197 	 *   [4] we would like all waiting for I/O to occur with maps unlocked
2198 	 *       so that we don't block other threads.
2199 	 */
2200 
2201 	first_entry = NULL;
2202 	*entry_list = NULL;
2203 
2204 	/*
2205 	 * break up the area into map entry sized regions and unmap.  note
2206 	 * that all mappings have to be removed before we can even consider
2207 	 * dropping references to amaps or VM objects (otherwise we could end
2208 	 * up with a mapping to a page on the free list which would be very bad)
2209 	 */
2210 
2211 	while ((entry != &map->header) && (entry->start < end)) {
2212 		KASSERT((entry->flags & UVM_MAP_STATIC) == 0);
2213 
2214 		UVM_MAP_CLIP_END(map, entry, end);
2215 		next = entry->next;
2216 		len = entry->end - entry->start;
2217 
2218 		/*
2219 		 * unwire before removing addresses from the pmap; otherwise
2220 		 * unwiring will put the entries back into the pmap (XXX).
2221 		 */
2222 
2223 		if (VM_MAPENT_ISWIRED(entry)) {
2224 			uvm_map_entry_unwire(map, entry);
2225 		}
2226 		if (flags & UVM_FLAG_VAONLY) {
2227 
2228 			/* nothing */
2229 
2230 		} else if ((map->flags & VM_MAP_PAGEABLE) == 0) {
2231 
2232 			/*
2233 			 * if the map is non-pageable, any pages mapped there
2234 			 * must be wired and entered with pmap_kenter_pa(),
2235 			 * and we should free any such pages immediately.
2236 			 * this is mostly used for kmem_map.
2237 			 */
2238 			KASSERT(vm_map_pmap(map) == pmap_kernel());
2239 
2240 			uvm_km_pgremove_intrsafe(map, entry->start, entry->end);
2241 		} else if (UVM_ET_ISOBJ(entry) &&
2242 			   UVM_OBJ_IS_KERN_OBJECT(entry->object.uvm_obj)) {
2243 			panic("%s: kernel object %p %p\n",
2244 			    __func__, map, entry);
2245 		} else if (UVM_ET_ISOBJ(entry) || entry->aref.ar_amap) {
2246 			/*
2247 			 * remove mappings the standard way.  lock object
2248 			 * and/or amap to ensure vm_page state does not
2249 			 * change while in pmap_remove().
2250 			 */
2251 
2252 			uvm_map_lock_entry(entry);
2253 			pmap_remove(map->pmap, entry->start, entry->end);
2254 			uvm_map_unlock_entry(entry);
2255 		}
2256 
2257 #if defined(UVMDEBUG)
2258 		/*
2259 		 * check if there's remaining mapping,
2260 		 * which is a bug in caller.
2261 		 */
2262 
2263 		vaddr_t va;
2264 		for (va = entry->start; va < entry->end;
2265 		    va += PAGE_SIZE) {
2266 			if (pmap_extract(vm_map_pmap(map), va, NULL)) {
2267 				panic("%s: %#"PRIxVADDR" has mapping",
2268 				    __func__, va);
2269 			}
2270 		}
2271 
2272 		if (VM_MAP_IS_KERNEL(map) && (flags & UVM_FLAG_NOWAIT) == 0) {
2273 			uvm_km_check_empty(map, entry->start,
2274 			    entry->end);
2275 		}
2276 #endif /* defined(UVMDEBUG) */
2277 
2278 		/*
2279 		 * remove entry from map and put it on our list of entries
2280 		 * that we've nuked.  then go to next entry.
2281 		 */
2282 
2283 		UVMHIST_LOG(maphist, "  removed map entry %p", entry, 0, 0,0);
2284 
2285 		/* critical!  prevents stale hint */
2286 		SAVE_HINT(map, entry, entry->prev);
2287 
2288 		uvm_map_entry_unlink(map, entry);
2289 		KASSERT(map->size >= len);
2290 		map->size -= len;
2291 		entry->prev = NULL;
2292 		entry->next = first_entry;
2293 		first_entry = entry;
2294 		entry = next;
2295 	}
2296 
2297 	/*
2298 	 * Note: if map is dying, leave pmap_update() for pmap_destroy(),
2299 	 * which will be called later.
2300 	 */
2301 	if ((map->flags & VM_MAP_DYING) == 0) {
2302 		pmap_update(vm_map_pmap(map));
2303 	} else {
2304 		KASSERT(vm_map_pmap(map) != pmap_kernel());
2305 	}
2306 
2307 	uvm_map_check(map, "unmap_remove leave");
2308 
2309 	/*
2310 	 * now we've cleaned up the map and are ready for the caller to drop
2311 	 * references to the mapped objects.
2312 	 */
2313 
2314 	*entry_list = first_entry;
2315 	UVMHIST_LOG(maphist,"<- done!", 0, 0, 0, 0);
2316 
2317 	if (map->flags & VM_MAP_WANTVA) {
2318 		mutex_enter(&map->misc_lock);
2319 		map->flags &= ~VM_MAP_WANTVA;
2320 		cv_broadcast(&map->cv);
2321 		mutex_exit(&map->misc_lock);
2322 	}
2323 }
2324 
2325 /*
2326  * uvm_unmap_detach: drop references in a chain of map entries
2327  *
2328  * => we will free the map entries as we traverse the list.
2329  */
2330 
2331 void
2332 uvm_unmap_detach(struct vm_map_entry *first_entry, int flags)
2333 {
2334 	struct vm_map_entry *next_entry;
2335 	UVMHIST_FUNC("uvm_unmap_detach"); UVMHIST_CALLED(maphist);
2336 
2337 	while (first_entry) {
2338 		KASSERT(!VM_MAPENT_ISWIRED(first_entry));
2339 		UVMHIST_LOG(maphist,
2340 		    "  detach %p: amap=%p, obj=%p, submap?=%d",
2341 		    first_entry, first_entry->aref.ar_amap,
2342 		    first_entry->object.uvm_obj,
2343 		    UVM_ET_ISSUBMAP(first_entry));
2344 
2345 		/*
2346 		 * drop reference to amap, if we've got one
2347 		 */
2348 
2349 		if (first_entry->aref.ar_amap)
2350 			uvm_map_unreference_amap(first_entry, flags);
2351 
2352 		/*
2353 		 * drop reference to our backing object, if we've got one
2354 		 */
2355 
2356 		KASSERT(!UVM_ET_ISSUBMAP(first_entry));
2357 		if (UVM_ET_ISOBJ(first_entry) &&
2358 		    first_entry->object.uvm_obj->pgops->pgo_detach) {
2359 			(*first_entry->object.uvm_obj->pgops->pgo_detach)
2360 				(first_entry->object.uvm_obj);
2361 		}
2362 		next_entry = first_entry->next;
2363 		uvm_mapent_free(first_entry);
2364 		first_entry = next_entry;
2365 	}
2366 	UVMHIST_LOG(maphist, "<- done", 0,0,0,0);
2367 }
2368 
2369 /*
2370  *   E X T R A C T I O N   F U N C T I O N S
2371  */
2372 
2373 /*
2374  * uvm_map_reserve: reserve space in a vm_map for future use.
2375  *
2376  * => we reserve space in a map by putting a dummy map entry in the
2377  *    map (dummy means obj=NULL, amap=NULL, prot=VM_PROT_NONE)
2378  * => map should be unlocked (we will write lock it)
2379  * => we return true if we were able to reserve space
2380  * => XXXCDC: should be inline?
2381  */
2382 
2383 int
2384 uvm_map_reserve(struct vm_map *map, vsize_t size,
2385     vaddr_t offset	/* hint for pmap_prefer */,
2386     vsize_t align	/* alignment */,
2387     vaddr_t *raddr	/* IN:hint, OUT: reserved VA */,
2388     uvm_flag_t flags	/* UVM_FLAG_FIXED or UVM_FLAG_COLORMATCH or 0 */)
2389 {
2390 	UVMHIST_FUNC("uvm_map_reserve"); UVMHIST_CALLED(maphist);
2391 
2392 	UVMHIST_LOG(maphist, "(map=%p, size=%#lx, offset=%#lx, addr=%p)",
2393 	    map,size,offset,raddr);
2394 
2395 	size = round_page(size);
2396 
2397 	/*
2398 	 * reserve some virtual space.
2399 	 */
2400 
2401 	if (uvm_map(map, raddr, size, NULL, offset, align,
2402 	    UVM_MAPFLAG(UVM_PROT_NONE, UVM_PROT_NONE, UVM_INH_NONE,
2403 	    UVM_ADV_RANDOM, UVM_FLAG_NOMERGE|flags)) != 0) {
2404 	    UVMHIST_LOG(maphist, "<- done (no VM)", 0,0,0,0);
2405 		return (false);
2406 	}
2407 
2408 	UVMHIST_LOG(maphist, "<- done (*raddr=%#lx)", *raddr,0,0,0);
2409 	return (true);
2410 }
2411 
2412 /*
2413  * uvm_map_replace: replace a reserved (blank) area of memory with
2414  * real mappings.
2415  *
2416  * => caller must WRITE-LOCK the map
2417  * => we return true if replacement was a success
2418  * => we expect the newents chain to have nnewents entrys on it and
2419  *    we expect newents->prev to point to the last entry on the list
2420  * => note newents is allowed to be NULL
2421  */
2422 
2423 static int
2424 uvm_map_replace(struct vm_map *map, vaddr_t start, vaddr_t end,
2425     struct vm_map_entry *newents, int nnewents, vsize_t nsize,
2426     struct vm_map_entry **oldentryp)
2427 {
2428 	struct vm_map_entry *oldent, *last;
2429 
2430 	uvm_map_check(map, "map_replace entry");
2431 
2432 	/*
2433 	 * first find the blank map entry at the specified address
2434 	 */
2435 
2436 	if (!uvm_map_lookup_entry(map, start, &oldent)) {
2437 		return (false);
2438 	}
2439 
2440 	/*
2441 	 * check to make sure we have a proper blank entry
2442 	 */
2443 
2444 	if (end < oldent->end) {
2445 		UVM_MAP_CLIP_END(map, oldent, end);
2446 	}
2447 	if (oldent->start != start || oldent->end != end ||
2448 	    oldent->object.uvm_obj != NULL || oldent->aref.ar_amap != NULL) {
2449 		return (false);
2450 	}
2451 
2452 #ifdef DIAGNOSTIC
2453 
2454 	/*
2455 	 * sanity check the newents chain
2456 	 */
2457 
2458 	{
2459 		struct vm_map_entry *tmpent = newents;
2460 		int nent = 0;
2461 		vsize_t sz = 0;
2462 		vaddr_t cur = start;
2463 
2464 		while (tmpent) {
2465 			nent++;
2466 			sz += tmpent->end - tmpent->start;
2467 			if (tmpent->start < cur)
2468 				panic("uvm_map_replace1");
2469 			if (tmpent->start >= tmpent->end || tmpent->end > end) {
2470 				panic("uvm_map_replace2: "
2471 				    "tmpent->start=%#"PRIxVADDR
2472 				    ", tmpent->end=%#"PRIxVADDR
2473 				    ", end=%#"PRIxVADDR,
2474 				    tmpent->start, tmpent->end, end);
2475 			}
2476 			cur = tmpent->end;
2477 			if (tmpent->next) {
2478 				if (tmpent->next->prev != tmpent)
2479 					panic("uvm_map_replace3");
2480 			} else {
2481 				if (newents->prev != tmpent)
2482 					panic("uvm_map_replace4");
2483 			}
2484 			tmpent = tmpent->next;
2485 		}
2486 		if (nent != nnewents)
2487 			panic("uvm_map_replace5");
2488 		if (sz != nsize)
2489 			panic("uvm_map_replace6");
2490 	}
2491 #endif
2492 
2493 	/*
2494 	 * map entry is a valid blank!   replace it.   (this does all the
2495 	 * work of map entry link/unlink...).
2496 	 */
2497 
2498 	if (newents) {
2499 		last = newents->prev;
2500 
2501 		/* critical: flush stale hints out of map */
2502 		SAVE_HINT(map, map->hint, newents);
2503 		if (map->first_free == oldent)
2504 			map->first_free = last;
2505 
2506 		last->next = oldent->next;
2507 		last->next->prev = last;
2508 
2509 		/* Fix RB tree */
2510 		uvm_rb_remove(map, oldent);
2511 
2512 		newents->prev = oldent->prev;
2513 		newents->prev->next = newents;
2514 		map->nentries = map->nentries + (nnewents - 1);
2515 
2516 		/* Fixup the RB tree */
2517 		{
2518 			int i;
2519 			struct vm_map_entry *tmp;
2520 
2521 			tmp = newents;
2522 			for (i = 0; i < nnewents && tmp; i++) {
2523 				uvm_rb_insert(map, tmp);
2524 				tmp = tmp->next;
2525 			}
2526 		}
2527 	} else {
2528 		/* NULL list of new entries: just remove the old one */
2529 		clear_hints(map, oldent);
2530 		uvm_map_entry_unlink(map, oldent);
2531 	}
2532 	map->size -= end - start - nsize;
2533 
2534 	uvm_map_check(map, "map_replace leave");
2535 
2536 	/*
2537 	 * now we can free the old blank entry and return.
2538 	 */
2539 
2540 	*oldentryp = oldent;
2541 	return (true);
2542 }
2543 
2544 /*
2545  * uvm_map_extract: extract a mapping from a map and put it somewhere
2546  *	(maybe removing the old mapping)
2547  *
2548  * => maps should be unlocked (we will write lock them)
2549  * => returns 0 on success, error code otherwise
2550  * => start must be page aligned
2551  * => len must be page sized
2552  * => flags:
2553  *      UVM_EXTRACT_REMOVE: remove mappings from srcmap
2554  *      UVM_EXTRACT_CONTIG: abort if unmapped area (advisory only)
2555  *      UVM_EXTRACT_QREF: for a temporary extraction do quick obj refs
2556  *      UVM_EXTRACT_FIXPROT: set prot to maxprot as we go
2557  *      UVM_EXTRACT_PROT_ALL: set prot to UVM_PROT_ALL as we go
2558  *    >>>NOTE: if you set REMOVE, you are not allowed to use CONTIG or QREF!<<<
2559  *    >>>NOTE: QREF's must be unmapped via the QREF path, thus should only
2560  *             be used from within the kernel in a kernel level map <<<
2561  */
2562 
2563 int
2564 uvm_map_extract(struct vm_map *srcmap, vaddr_t start, vsize_t len,
2565     struct vm_map *dstmap, vaddr_t *dstaddrp, int flags)
2566 {
2567 	vaddr_t dstaddr, end, newend, oldoffset, fudge, orig_fudge;
2568 	struct vm_map_entry *chain, *endchain, *entry, *orig_entry, *newentry,
2569 	    *deadentry, *oldentry;
2570 	struct vm_map_entry *resentry = NULL; /* a dummy reservation entry */
2571 	vsize_t elen __unused;
2572 	int nchain, error, copy_ok;
2573 	vsize_t nsize;
2574 	UVMHIST_FUNC("uvm_map_extract"); UVMHIST_CALLED(maphist);
2575 
2576 	UVMHIST_LOG(maphist,"(srcmap=%p,start=%#lx, len=%#lx", srcmap, start,
2577 	    len,0);
2578 	UVMHIST_LOG(maphist," ...,dstmap=%p, flags=%#x)", dstmap,flags,0,0);
2579 
2580 	/*
2581 	 * step 0: sanity check: start must be on a page boundary, length
2582 	 * must be page sized.  can't ask for CONTIG/QREF if you asked for
2583 	 * REMOVE.
2584 	 */
2585 
2586 	KASSERT((start & PAGE_MASK) == 0 && (len & PAGE_MASK) == 0);
2587 	KASSERT((flags & UVM_EXTRACT_REMOVE) == 0 ||
2588 		(flags & (UVM_EXTRACT_CONTIG|UVM_EXTRACT_QREF)) == 0);
2589 
2590 	/*
2591 	 * step 1: reserve space in the target map for the extracted area
2592 	 */
2593 
2594 	if ((flags & UVM_EXTRACT_RESERVED) == 0) {
2595 		dstaddr = vm_map_min(dstmap);
2596 		if (!uvm_map_reserve(dstmap, len, start,
2597 		    atop(start) & uvmexp.colormask, &dstaddr,
2598 		    UVM_FLAG_COLORMATCH))
2599 			return (ENOMEM);
2600 		KASSERT((atop(start ^ dstaddr) & uvmexp.colormask) == 0);
2601 		*dstaddrp = dstaddr;	/* pass address back to caller */
2602 		UVMHIST_LOG(maphist, "  dstaddr=%#lx", dstaddr,0,0,0);
2603 	} else {
2604 		dstaddr = *dstaddrp;
2605 	}
2606 
2607 	/*
2608 	 * step 2: setup for the extraction process loop by init'ing the
2609 	 * map entry chain, locking src map, and looking up the first useful
2610 	 * entry in the map.
2611 	 */
2612 
2613 	end = start + len;
2614 	newend = dstaddr + len;
2615 	chain = endchain = NULL;
2616 	nchain = 0;
2617 	nsize = 0;
2618 	vm_map_lock(srcmap);
2619 
2620 	if (uvm_map_lookup_entry(srcmap, start, &entry)) {
2621 
2622 		/* "start" is within an entry */
2623 		if (flags & UVM_EXTRACT_QREF) {
2624 
2625 			/*
2626 			 * for quick references we don't clip the entry, so
2627 			 * the entry may map space "before" the starting
2628 			 * virtual address... this is the "fudge" factor
2629 			 * (which can be non-zero only the first time
2630 			 * through the "while" loop in step 3).
2631 			 */
2632 
2633 			fudge = start - entry->start;
2634 		} else {
2635 
2636 			/*
2637 			 * normal reference: we clip the map to fit (thus
2638 			 * fudge is zero)
2639 			 */
2640 
2641 			UVM_MAP_CLIP_START(srcmap, entry, start);
2642 			SAVE_HINT(srcmap, srcmap->hint, entry->prev);
2643 			fudge = 0;
2644 		}
2645 	} else {
2646 
2647 		/* "start" is not within an entry ... skip to next entry */
2648 		if (flags & UVM_EXTRACT_CONTIG) {
2649 			error = EINVAL;
2650 			goto bad;    /* definite hole here ... */
2651 		}
2652 
2653 		entry = entry->next;
2654 		fudge = 0;
2655 	}
2656 
2657 	/* save values from srcmap for step 6 */
2658 	orig_entry = entry;
2659 	orig_fudge = fudge;
2660 
2661 	/*
2662 	 * step 3: now start looping through the map entries, extracting
2663 	 * as we go.
2664 	 */
2665 
2666 	while (entry->start < end && entry != &srcmap->header) {
2667 
2668 		/* if we are not doing a quick reference, clip it */
2669 		if ((flags & UVM_EXTRACT_QREF) == 0)
2670 			UVM_MAP_CLIP_END(srcmap, entry, end);
2671 
2672 		/* clear needs_copy (allow chunking) */
2673 		if (UVM_ET_ISNEEDSCOPY(entry)) {
2674 			amap_copy(srcmap, entry,
2675 			    AMAP_COPY_NOWAIT|AMAP_COPY_NOMERGE, start, end);
2676 			if (UVM_ET_ISNEEDSCOPY(entry)) {  /* failed? */
2677 				error = ENOMEM;
2678 				goto bad;
2679 			}
2680 
2681 			/* amap_copy could clip (during chunk)!  update fudge */
2682 			if (fudge) {
2683 				fudge = start - entry->start;
2684 				orig_fudge = fudge;
2685 			}
2686 		}
2687 
2688 		/* calculate the offset of this from "start" */
2689 		oldoffset = (entry->start + fudge) - start;
2690 
2691 		/* allocate a new map entry */
2692 		newentry = uvm_mapent_alloc(dstmap, 0);
2693 		if (newentry == NULL) {
2694 			error = ENOMEM;
2695 			goto bad;
2696 		}
2697 
2698 		/* set up new map entry */
2699 		newentry->next = NULL;
2700 		newentry->prev = endchain;
2701 		newentry->start = dstaddr + oldoffset;
2702 		newentry->end =
2703 		    newentry->start + (entry->end - (entry->start + fudge));
2704 		if (newentry->end > newend || newentry->end < newentry->start)
2705 			newentry->end = newend;
2706 		newentry->object.uvm_obj = entry->object.uvm_obj;
2707 		if (newentry->object.uvm_obj) {
2708 			if (newentry->object.uvm_obj->pgops->pgo_reference)
2709 				newentry->object.uvm_obj->pgops->
2710 				    pgo_reference(newentry->object.uvm_obj);
2711 				newentry->offset = entry->offset + fudge;
2712 		} else {
2713 			newentry->offset = 0;
2714 		}
2715 		newentry->etype = entry->etype;
2716 		if (flags & UVM_EXTRACT_PROT_ALL) {
2717 			newentry->protection = newentry->max_protection =
2718 			    UVM_PROT_ALL;
2719 		} else {
2720 			newentry->protection = (flags & UVM_EXTRACT_FIXPROT) ?
2721 			    entry->max_protection : entry->protection;
2722 			newentry->max_protection = entry->max_protection;
2723 		}
2724 		newentry->inheritance = entry->inheritance;
2725 		newentry->wired_count = 0;
2726 		newentry->aref.ar_amap = entry->aref.ar_amap;
2727 		if (newentry->aref.ar_amap) {
2728 			newentry->aref.ar_pageoff =
2729 			    entry->aref.ar_pageoff + (fudge >> PAGE_SHIFT);
2730 			uvm_map_reference_amap(newentry, AMAP_SHARED |
2731 			    ((flags & UVM_EXTRACT_QREF) ? AMAP_REFALL : 0));
2732 		} else {
2733 			newentry->aref.ar_pageoff = 0;
2734 		}
2735 		newentry->advice = entry->advice;
2736 		if ((flags & UVM_EXTRACT_QREF) != 0) {
2737 			newentry->flags |= UVM_MAP_NOMERGE;
2738 		}
2739 
2740 		/* now link it on the chain */
2741 		nchain++;
2742 		nsize += newentry->end - newentry->start;
2743 		if (endchain == NULL) {
2744 			chain = endchain = newentry;
2745 		} else {
2746 			endchain->next = newentry;
2747 			endchain = newentry;
2748 		}
2749 
2750 		/* end of 'while' loop! */
2751 		if ((flags & UVM_EXTRACT_CONTIG) && entry->end < end &&
2752 		    (entry->next == &srcmap->header ||
2753 		    entry->next->start != entry->end)) {
2754 			error = EINVAL;
2755 			goto bad;
2756 		}
2757 		entry = entry->next;
2758 		fudge = 0;
2759 	}
2760 
2761 	/*
2762 	 * step 4: close off chain (in format expected by uvm_map_replace)
2763 	 */
2764 
2765 	if (chain)
2766 		chain->prev = endchain;
2767 
2768 	/*
2769 	 * step 5: attempt to lock the dest map so we can pmap_copy.
2770 	 * note usage of copy_ok:
2771 	 *   1 => dstmap locked, pmap_copy ok, and we "replace" here (step 5)
2772 	 *   0 => dstmap unlocked, NO pmap_copy, and we will "replace" in step 7
2773 	 */
2774 
2775 	if (srcmap == dstmap || vm_map_lock_try(dstmap) == true) {
2776 		copy_ok = 1;
2777 		if (!uvm_map_replace(dstmap, dstaddr, dstaddr+len, chain,
2778 		    nchain, nsize, &resentry)) {
2779 			if (srcmap != dstmap)
2780 				vm_map_unlock(dstmap);
2781 			error = EIO;
2782 			goto bad;
2783 		}
2784 	} else {
2785 		copy_ok = 0;
2786 		/* replace defered until step 7 */
2787 	}
2788 
2789 	/*
2790 	 * step 6: traverse the srcmap a second time to do the following:
2791 	 *  - if we got a lock on the dstmap do pmap_copy
2792 	 *  - if UVM_EXTRACT_REMOVE remove the entries
2793 	 * we make use of orig_entry and orig_fudge (saved in step 2)
2794 	 */
2795 
2796 	if (copy_ok || (flags & UVM_EXTRACT_REMOVE)) {
2797 
2798 		/* purge possible stale hints from srcmap */
2799 		if (flags & UVM_EXTRACT_REMOVE) {
2800 			SAVE_HINT(srcmap, srcmap->hint, orig_entry->prev);
2801 			if (srcmap->first_free != &srcmap->header &&
2802 			    srcmap->first_free->start >= start)
2803 				srcmap->first_free = orig_entry->prev;
2804 		}
2805 
2806 		entry = orig_entry;
2807 		fudge = orig_fudge;
2808 		deadentry = NULL;	/* for UVM_EXTRACT_REMOVE */
2809 
2810 		while (entry->start < end && entry != &srcmap->header) {
2811 			if (copy_ok) {
2812 				oldoffset = (entry->start + fudge) - start;
2813 				elen = MIN(end, entry->end) -
2814 				    (entry->start + fudge);
2815 				pmap_copy(dstmap->pmap, srcmap->pmap,
2816 				    dstaddr + oldoffset, elen,
2817 				    entry->start + fudge);
2818 			}
2819 
2820 			/* we advance "entry" in the following if statement */
2821 			if (flags & UVM_EXTRACT_REMOVE) {
2822 				uvm_map_lock_entry(entry);
2823 				pmap_remove(srcmap->pmap, entry->start,
2824 						entry->end);
2825 				uvm_map_unlock_entry(entry);
2826 				oldentry = entry;	/* save entry */
2827 				entry = entry->next;	/* advance */
2828 				uvm_map_entry_unlink(srcmap, oldentry);
2829 							/* add to dead list */
2830 				oldentry->next = deadentry;
2831 				deadentry = oldentry;
2832 			} else {
2833 				entry = entry->next;		/* advance */
2834 			}
2835 
2836 			/* end of 'while' loop */
2837 			fudge = 0;
2838 		}
2839 		pmap_update(srcmap->pmap);
2840 
2841 		/*
2842 		 * unlock dstmap.  we will dispose of deadentry in
2843 		 * step 7 if needed
2844 		 */
2845 
2846 		if (copy_ok && srcmap != dstmap)
2847 			vm_map_unlock(dstmap);
2848 
2849 	} else {
2850 		deadentry = NULL;
2851 	}
2852 
2853 	/*
2854 	 * step 7: we are done with the source map, unlock.   if copy_ok
2855 	 * is 0 then we have not replaced the dummy mapping in dstmap yet
2856 	 * and we need to do so now.
2857 	 */
2858 
2859 	vm_map_unlock(srcmap);
2860 	if ((flags & UVM_EXTRACT_REMOVE) && deadentry)
2861 		uvm_unmap_detach(deadentry, 0);   /* dispose of old entries */
2862 
2863 	/* now do the replacement if we didn't do it in step 5 */
2864 	if (copy_ok == 0) {
2865 		vm_map_lock(dstmap);
2866 		error = uvm_map_replace(dstmap, dstaddr, dstaddr+len, chain,
2867 		    nchain, nsize, &resentry);
2868 		vm_map_unlock(dstmap);
2869 
2870 		if (error == false) {
2871 			error = EIO;
2872 			goto bad2;
2873 		}
2874 	}
2875 
2876 	if (resentry != NULL)
2877 		uvm_mapent_free(resentry);
2878 
2879 	return (0);
2880 
2881 	/*
2882 	 * bad: failure recovery
2883 	 */
2884 bad:
2885 	vm_map_unlock(srcmap);
2886 bad2:			/* src already unlocked */
2887 	if (chain)
2888 		uvm_unmap_detach(chain,
2889 		    (flags & UVM_EXTRACT_QREF) ? AMAP_REFALL : 0);
2890 
2891 	if (resentry != NULL)
2892 		uvm_mapent_free(resentry);
2893 
2894 	if ((flags & UVM_EXTRACT_RESERVED) == 0) {
2895 		uvm_unmap(dstmap, dstaddr, dstaddr+len);   /* ??? */
2896 	}
2897 	return (error);
2898 }
2899 
2900 /* end of extraction functions */
2901 
2902 /*
2903  * uvm_map_submap: punch down part of a map into a submap
2904  *
2905  * => only the kernel_map is allowed to be submapped
2906  * => the purpose of submapping is to break up the locking granularity
2907  *	of a larger map
2908  * => the range specified must have been mapped previously with a uvm_map()
2909  *	call [with uobj==NULL] to create a blank map entry in the main map.
2910  *	[And it had better still be blank!]
2911  * => maps which contain submaps should never be copied or forked.
2912  * => to remove a submap, use uvm_unmap() on the main map
2913  *	and then uvm_map_deallocate() the submap.
2914  * => main map must be unlocked.
2915  * => submap must have been init'd and have a zero reference count.
2916  *	[need not be locked as we don't actually reference it]
2917  */
2918 
2919 int
2920 uvm_map_submap(struct vm_map *map, vaddr_t start, vaddr_t end,
2921     struct vm_map *submap)
2922 {
2923 	struct vm_map_entry *entry;
2924 	int error;
2925 
2926 	vm_map_lock(map);
2927 	VM_MAP_RANGE_CHECK(map, start, end);
2928 
2929 	if (uvm_map_lookup_entry(map, start, &entry)) {
2930 		UVM_MAP_CLIP_START(map, entry, start);
2931 		UVM_MAP_CLIP_END(map, entry, end);	/* to be safe */
2932 	} else {
2933 		entry = NULL;
2934 	}
2935 
2936 	if (entry != NULL &&
2937 	    entry->start == start && entry->end == end &&
2938 	    entry->object.uvm_obj == NULL && entry->aref.ar_amap == NULL &&
2939 	    !UVM_ET_ISCOPYONWRITE(entry) && !UVM_ET_ISNEEDSCOPY(entry)) {
2940 		entry->etype |= UVM_ET_SUBMAP;
2941 		entry->object.sub_map = submap;
2942 		entry->offset = 0;
2943 		uvm_map_reference(submap);
2944 		error = 0;
2945 	} else {
2946 		error = EINVAL;
2947 	}
2948 	vm_map_unlock(map);
2949 
2950 	return error;
2951 }
2952 
2953 /*
2954  * uvm_map_protect: change map protection
2955  *
2956  * => set_max means set max_protection.
2957  * => map must be unlocked.
2958  */
2959 
2960 #define MASK(entry)	(UVM_ET_ISCOPYONWRITE(entry) ? \
2961 			 ~VM_PROT_WRITE : VM_PROT_ALL)
2962 
2963 int
2964 uvm_map_protect(struct vm_map *map, vaddr_t start, vaddr_t end,
2965     vm_prot_t new_prot, bool set_max)
2966 {
2967 	struct vm_map_entry *current, *entry;
2968 	int error = 0;
2969 	UVMHIST_FUNC("uvm_map_protect"); UVMHIST_CALLED(maphist);
2970 	UVMHIST_LOG(maphist,"(map=%p,start=%#lx,end=%#lx,new_prot=%#x)",
2971 		    map, start, end, new_prot);
2972 
2973 	vm_map_lock(map);
2974 	VM_MAP_RANGE_CHECK(map, start, end);
2975 	if (uvm_map_lookup_entry(map, start, &entry)) {
2976 		UVM_MAP_CLIP_START(map, entry, start);
2977 	} else {
2978 		entry = entry->next;
2979 	}
2980 
2981 	/*
2982 	 * make a first pass to check for protection violations.
2983 	 */
2984 
2985 	current = entry;
2986 	while ((current != &map->header) && (current->start < end)) {
2987 		if (UVM_ET_ISSUBMAP(current)) {
2988 			error = EINVAL;
2989 			goto out;
2990 		}
2991 		if ((new_prot & current->max_protection) != new_prot) {
2992 			error = EACCES;
2993 			goto out;
2994 		}
2995 		/*
2996 		 * Don't allow VM_PROT_EXECUTE to be set on entries that
2997 		 * point to vnodes that are associated with a NOEXEC file
2998 		 * system.
2999 		 */
3000 		if (UVM_ET_ISOBJ(current) &&
3001 		    UVM_OBJ_IS_VNODE(current->object.uvm_obj)) {
3002 			struct vnode *vp =
3003 			    (struct vnode *) current->object.uvm_obj;
3004 
3005 			if ((new_prot & VM_PROT_EXECUTE) != 0 &&
3006 			    (vp->v_mount->mnt_flag & MNT_NOEXEC) != 0) {
3007 				error = EACCES;
3008 				goto out;
3009 			}
3010 		}
3011 
3012 		current = current->next;
3013 	}
3014 
3015 	/* go back and fix up protections (no need to clip this time). */
3016 
3017 	current = entry;
3018 	while ((current != &map->header) && (current->start < end)) {
3019 		vm_prot_t old_prot;
3020 
3021 		UVM_MAP_CLIP_END(map, current, end);
3022 		old_prot = current->protection;
3023 		if (set_max)
3024 			current->protection =
3025 			    (current->max_protection = new_prot) & old_prot;
3026 		else
3027 			current->protection = new_prot;
3028 
3029 		/*
3030 		 * update physical map if necessary.  worry about copy-on-write
3031 		 * here -- CHECK THIS XXX
3032 		 */
3033 
3034 		if (current->protection != old_prot) {
3035 			/* update pmap! */
3036 			uvm_map_lock_entry(current);
3037 			pmap_protect(map->pmap, current->start, current->end,
3038 			    current->protection & MASK(entry));
3039 			uvm_map_unlock_entry(current);
3040 
3041 			/*
3042 			 * If this entry points at a vnode, and the
3043 			 * protection includes VM_PROT_EXECUTE, mark
3044 			 * the vnode as VEXECMAP.
3045 			 */
3046 			if (UVM_ET_ISOBJ(current)) {
3047 				struct uvm_object *uobj =
3048 				    current->object.uvm_obj;
3049 
3050 				if (UVM_OBJ_IS_VNODE(uobj) &&
3051 				    (current->protection & VM_PROT_EXECUTE)) {
3052 					vn_markexec((struct vnode *) uobj);
3053 				}
3054 			}
3055 		}
3056 
3057 		/*
3058 		 * If the map is configured to lock any future mappings,
3059 		 * wire this entry now if the old protection was VM_PROT_NONE
3060 		 * and the new protection is not VM_PROT_NONE.
3061 		 */
3062 
3063 		if ((map->flags & VM_MAP_WIREFUTURE) != 0 &&
3064 		    VM_MAPENT_ISWIRED(entry) == 0 &&
3065 		    old_prot == VM_PROT_NONE &&
3066 		    new_prot != VM_PROT_NONE) {
3067 			if (uvm_map_pageable(map, entry->start,
3068 			    entry->end, false,
3069 			    UVM_LK_ENTER|UVM_LK_EXIT) != 0) {
3070 
3071 				/*
3072 				 * If locking the entry fails, remember the
3073 				 * error if it's the first one.  Note we
3074 				 * still continue setting the protection in
3075 				 * the map, but will return the error
3076 				 * condition regardless.
3077 				 *
3078 				 * XXX Ignore what the actual error is,
3079 				 * XXX just call it a resource shortage
3080 				 * XXX so that it doesn't get confused
3081 				 * XXX what uvm_map_protect() itself would
3082 				 * XXX normally return.
3083 				 */
3084 
3085 				error = ENOMEM;
3086 			}
3087 		}
3088 		current = current->next;
3089 	}
3090 	pmap_update(map->pmap);
3091 
3092  out:
3093 	vm_map_unlock(map);
3094 
3095 	UVMHIST_LOG(maphist, "<- done, error=%d",error,0,0,0);
3096 	return error;
3097 }
3098 
3099 #undef  MASK
3100 
3101 /*
3102  * uvm_map_inherit: set inheritance code for range of addrs in map.
3103  *
3104  * => map must be unlocked
3105  * => note that the inherit code is used during a "fork".  see fork
3106  *	code for details.
3107  */
3108 
3109 int
3110 uvm_map_inherit(struct vm_map *map, vaddr_t start, vaddr_t end,
3111     vm_inherit_t new_inheritance)
3112 {
3113 	struct vm_map_entry *entry, *temp_entry;
3114 	UVMHIST_FUNC("uvm_map_inherit"); UVMHIST_CALLED(maphist);
3115 	UVMHIST_LOG(maphist,"(map=%p,start=%#lx,end=%#lx,new_inh=%#x)",
3116 	    map, start, end, new_inheritance);
3117 
3118 	switch (new_inheritance) {
3119 	case MAP_INHERIT_NONE:
3120 	case MAP_INHERIT_COPY:
3121 	case MAP_INHERIT_SHARE:
3122 	case MAP_INHERIT_ZERO:
3123 		break;
3124 	default:
3125 		UVMHIST_LOG(maphist,"<- done (INVALID ARG)",0,0,0,0);
3126 		return EINVAL;
3127 	}
3128 
3129 	vm_map_lock(map);
3130 	VM_MAP_RANGE_CHECK(map, start, end);
3131 	if (uvm_map_lookup_entry(map, start, &temp_entry)) {
3132 		entry = temp_entry;
3133 		UVM_MAP_CLIP_START(map, entry, start);
3134 	}  else {
3135 		entry = temp_entry->next;
3136 	}
3137 	while ((entry != &map->header) && (entry->start < end)) {
3138 		UVM_MAP_CLIP_END(map, entry, end);
3139 		entry->inheritance = new_inheritance;
3140 		entry = entry->next;
3141 	}
3142 	vm_map_unlock(map);
3143 	UVMHIST_LOG(maphist,"<- done (OK)",0,0,0,0);
3144 	return 0;
3145 }
3146 
3147 /*
3148  * uvm_map_advice: set advice code for range of addrs in map.
3149  *
3150  * => map must be unlocked
3151  */
3152 
3153 int
3154 uvm_map_advice(struct vm_map *map, vaddr_t start, vaddr_t end, int new_advice)
3155 {
3156 	struct vm_map_entry *entry, *temp_entry;
3157 	UVMHIST_FUNC("uvm_map_advice"); UVMHIST_CALLED(maphist);
3158 	UVMHIST_LOG(maphist,"(map=%p,start=%#lx,end=%#lx,new_adv=%#x)",
3159 	    map, start, end, new_advice);
3160 
3161 	vm_map_lock(map);
3162 	VM_MAP_RANGE_CHECK(map, start, end);
3163 	if (uvm_map_lookup_entry(map, start, &temp_entry)) {
3164 		entry = temp_entry;
3165 		UVM_MAP_CLIP_START(map, entry, start);
3166 	} else {
3167 		entry = temp_entry->next;
3168 	}
3169 
3170 	/*
3171 	 * XXXJRT: disallow holes?
3172 	 */
3173 
3174 	while ((entry != &map->header) && (entry->start < end)) {
3175 		UVM_MAP_CLIP_END(map, entry, end);
3176 
3177 		switch (new_advice) {
3178 		case MADV_NORMAL:
3179 		case MADV_RANDOM:
3180 		case MADV_SEQUENTIAL:
3181 			/* nothing special here */
3182 			break;
3183 
3184 		default:
3185 			vm_map_unlock(map);
3186 			UVMHIST_LOG(maphist,"<- done (INVALID ARG)",0,0,0,0);
3187 			return EINVAL;
3188 		}
3189 		entry->advice = new_advice;
3190 		entry = entry->next;
3191 	}
3192 
3193 	vm_map_unlock(map);
3194 	UVMHIST_LOG(maphist,"<- done (OK)",0,0,0,0);
3195 	return 0;
3196 }
3197 
3198 /*
3199  * uvm_map_willneed: apply MADV_WILLNEED
3200  */
3201 
3202 int
3203 uvm_map_willneed(struct vm_map *map, vaddr_t start, vaddr_t end)
3204 {
3205 	struct vm_map_entry *entry;
3206 	UVMHIST_FUNC("uvm_map_willneed"); UVMHIST_CALLED(maphist);
3207 	UVMHIST_LOG(maphist,"(map=%p,start=%#lx,end=%#lx)",
3208 	    map, start, end, 0);
3209 
3210 	vm_map_lock_read(map);
3211 	VM_MAP_RANGE_CHECK(map, start, end);
3212 	if (!uvm_map_lookup_entry(map, start, &entry)) {
3213 		entry = entry->next;
3214 	}
3215 	while (entry->start < end) {
3216 		struct vm_amap * const amap = entry->aref.ar_amap;
3217 		struct uvm_object * const uobj = entry->object.uvm_obj;
3218 
3219 		KASSERT(entry != &map->header);
3220 		KASSERT(start < entry->end);
3221 		/*
3222 		 * For now, we handle only the easy but commonly-requested case.
3223 		 * ie. start prefetching of backing uobj pages.
3224 		 *
3225 		 * XXX It might be useful to pmap_enter() the already-in-core
3226 		 * pages by inventing a "weak" mode for uvm_fault() which would
3227 		 * only do the PGO_LOCKED pgo_get().
3228 		 */
3229 		if (UVM_ET_ISOBJ(entry) && amap == NULL && uobj != NULL) {
3230 			off_t offset;
3231 			off_t size;
3232 
3233 			offset = entry->offset;
3234 			if (start < entry->start) {
3235 				offset += entry->start - start;
3236 			}
3237 			size = entry->offset + (entry->end - entry->start);
3238 			if (entry->end < end) {
3239 				size -= end - entry->end;
3240 			}
3241 			uvm_readahead(uobj, offset, size);
3242 		}
3243 		entry = entry->next;
3244 	}
3245 	vm_map_unlock_read(map);
3246 	UVMHIST_LOG(maphist,"<- done (OK)",0,0,0,0);
3247 	return 0;
3248 }
3249 
3250 /*
3251  * uvm_map_pageable: sets the pageability of a range in a map.
3252  *
3253  * => wires map entries.  should not be used for transient page locking.
3254  *	for that, use uvm_fault_wire()/uvm_fault_unwire() (see uvm_vslock()).
3255  * => regions specified as not pageable require lock-down (wired) memory
3256  *	and page tables.
3257  * => map must never be read-locked
3258  * => if islocked is true, map is already write-locked
3259  * => we always unlock the map, since we must downgrade to a read-lock
3260  *	to call uvm_fault_wire()
3261  * => XXXCDC: check this and try and clean it up.
3262  */
3263 
3264 int
3265 uvm_map_pageable(struct vm_map *map, vaddr_t start, vaddr_t end,
3266     bool new_pageable, int lockflags)
3267 {
3268 	struct vm_map_entry *entry, *start_entry, *failed_entry;
3269 	int rv;
3270 #ifdef DIAGNOSTIC
3271 	u_int timestamp_save;
3272 #endif
3273 	UVMHIST_FUNC("uvm_map_pageable"); UVMHIST_CALLED(maphist);
3274 	UVMHIST_LOG(maphist,"(map=%p,start=%#lx,end=%#lx,new_pageable=%u)",
3275 		    map, start, end, new_pageable);
3276 	KASSERT(map->flags & VM_MAP_PAGEABLE);
3277 
3278 	if ((lockflags & UVM_LK_ENTER) == 0)
3279 		vm_map_lock(map);
3280 	VM_MAP_RANGE_CHECK(map, start, end);
3281 
3282 	/*
3283 	 * only one pageability change may take place at one time, since
3284 	 * uvm_fault_wire assumes it will be called only once for each
3285 	 * wiring/unwiring.  therefore, we have to make sure we're actually
3286 	 * changing the pageability for the entire region.  we do so before
3287 	 * making any changes.
3288 	 */
3289 
3290 	if (uvm_map_lookup_entry(map, start, &start_entry) == false) {
3291 		if ((lockflags & UVM_LK_EXIT) == 0)
3292 			vm_map_unlock(map);
3293 
3294 		UVMHIST_LOG(maphist,"<- done (fault)",0,0,0,0);
3295 		return EFAULT;
3296 	}
3297 	entry = start_entry;
3298 
3299 	/*
3300 	 * handle wiring and unwiring separately.
3301 	 */
3302 
3303 	if (new_pageable) {		/* unwire */
3304 		UVM_MAP_CLIP_START(map, entry, start);
3305 
3306 		/*
3307 		 * unwiring.  first ensure that the range to be unwired is
3308 		 * really wired down and that there are no holes.
3309 		 */
3310 
3311 		while ((entry != &map->header) && (entry->start < end)) {
3312 			if (entry->wired_count == 0 ||
3313 			    (entry->end < end &&
3314 			     (entry->next == &map->header ||
3315 			      entry->next->start > entry->end))) {
3316 				if ((lockflags & UVM_LK_EXIT) == 0)
3317 					vm_map_unlock(map);
3318 				UVMHIST_LOG(maphist, "<- done (INVAL)",0,0,0,0);
3319 				return EINVAL;
3320 			}
3321 			entry = entry->next;
3322 		}
3323 
3324 		/*
3325 		 * POSIX 1003.1b - a single munlock call unlocks a region,
3326 		 * regardless of the number of mlock calls made on that
3327 		 * region.
3328 		 */
3329 
3330 		entry = start_entry;
3331 		while ((entry != &map->header) && (entry->start < end)) {
3332 			UVM_MAP_CLIP_END(map, entry, end);
3333 			if (VM_MAPENT_ISWIRED(entry))
3334 				uvm_map_entry_unwire(map, entry);
3335 			entry = entry->next;
3336 		}
3337 		if ((lockflags & UVM_LK_EXIT) == 0)
3338 			vm_map_unlock(map);
3339 		UVMHIST_LOG(maphist,"<- done (OK UNWIRE)",0,0,0,0);
3340 		return 0;
3341 	}
3342 
3343 	/*
3344 	 * wire case: in two passes [XXXCDC: ugly block of code here]
3345 	 *
3346 	 * 1: holding the write lock, we create any anonymous maps that need
3347 	 *    to be created.  then we clip each map entry to the region to
3348 	 *    be wired and increment its wiring count.
3349 	 *
3350 	 * 2: we downgrade to a read lock, and call uvm_fault_wire to fault
3351 	 *    in the pages for any newly wired area (wired_count == 1).
3352 	 *
3353 	 *    downgrading to a read lock for uvm_fault_wire avoids a possible
3354 	 *    deadlock with another thread that may have faulted on one of
3355 	 *    the pages to be wired (it would mark the page busy, blocking
3356 	 *    us, then in turn block on the map lock that we hold).  because
3357 	 *    of problems in the recursive lock package, we cannot upgrade
3358 	 *    to a write lock in vm_map_lookup.  thus, any actions that
3359 	 *    require the write lock must be done beforehand.  because we
3360 	 *    keep the read lock on the map, the copy-on-write status of the
3361 	 *    entries we modify here cannot change.
3362 	 */
3363 
3364 	while ((entry != &map->header) && (entry->start < end)) {
3365 		if (VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */
3366 
3367 			/*
3368 			 * perform actions of vm_map_lookup that need the
3369 			 * write lock on the map: create an anonymous map
3370 			 * for a copy-on-write region, or an anonymous map
3371 			 * for a zero-fill region.  (XXXCDC: submap case
3372 			 * ok?)
3373 			 */
3374 
3375 			if (!UVM_ET_ISSUBMAP(entry)) {  /* not submap */
3376 				if (UVM_ET_ISNEEDSCOPY(entry) &&
3377 				    ((entry->max_protection & VM_PROT_WRITE) ||
3378 				     (entry->object.uvm_obj == NULL))) {
3379 					amap_copy(map, entry, 0, start, end);
3380 					/* XXXCDC: wait OK? */
3381 				}
3382 			}
3383 		}
3384 		UVM_MAP_CLIP_START(map, entry, start);
3385 		UVM_MAP_CLIP_END(map, entry, end);
3386 		entry->wired_count++;
3387 
3388 		/*
3389 		 * Check for holes
3390 		 */
3391 
3392 		if (entry->protection == VM_PROT_NONE ||
3393 		    (entry->end < end &&
3394 		     (entry->next == &map->header ||
3395 		      entry->next->start > entry->end))) {
3396 
3397 			/*
3398 			 * found one.  amap creation actions do not need to
3399 			 * be undone, but the wired counts need to be restored.
3400 			 */
3401 
3402 			while (entry != &map->header && entry->end > start) {
3403 				entry->wired_count--;
3404 				entry = entry->prev;
3405 			}
3406 			if ((lockflags & UVM_LK_EXIT) == 0)
3407 				vm_map_unlock(map);
3408 			UVMHIST_LOG(maphist,"<- done (INVALID WIRE)",0,0,0,0);
3409 			return EINVAL;
3410 		}
3411 		entry = entry->next;
3412 	}
3413 
3414 	/*
3415 	 * Pass 2.
3416 	 */
3417 
3418 #ifdef DIAGNOSTIC
3419 	timestamp_save = map->timestamp;
3420 #endif
3421 	vm_map_busy(map);
3422 	vm_map_unlock(map);
3423 
3424 	rv = 0;
3425 	entry = start_entry;
3426 	while (entry != &map->header && entry->start < end) {
3427 		if (entry->wired_count == 1) {
3428 			rv = uvm_fault_wire(map, entry->start, entry->end,
3429 			    entry->max_protection, 1);
3430 			if (rv) {
3431 
3432 				/*
3433 				 * wiring failed.  break out of the loop.
3434 				 * we'll clean up the map below, once we
3435 				 * have a write lock again.
3436 				 */
3437 
3438 				break;
3439 			}
3440 		}
3441 		entry = entry->next;
3442 	}
3443 
3444 	if (rv) {	/* failed? */
3445 
3446 		/*
3447 		 * Get back to an exclusive (write) lock.
3448 		 */
3449 
3450 		vm_map_lock(map);
3451 		vm_map_unbusy(map);
3452 
3453 #ifdef DIAGNOSTIC
3454 		if (timestamp_save + 1 != map->timestamp)
3455 			panic("uvm_map_pageable: stale map");
3456 #endif
3457 
3458 		/*
3459 		 * first drop the wiring count on all the entries
3460 		 * which haven't actually been wired yet.
3461 		 */
3462 
3463 		failed_entry = entry;
3464 		while (entry != &map->header && entry->start < end) {
3465 			entry->wired_count--;
3466 			entry = entry->next;
3467 		}
3468 
3469 		/*
3470 		 * now, unwire all the entries that were successfully
3471 		 * wired above.
3472 		 */
3473 
3474 		entry = start_entry;
3475 		while (entry != failed_entry) {
3476 			entry->wired_count--;
3477 			if (VM_MAPENT_ISWIRED(entry) == 0)
3478 				uvm_map_entry_unwire(map, entry);
3479 			entry = entry->next;
3480 		}
3481 		if ((lockflags & UVM_LK_EXIT) == 0)
3482 			vm_map_unlock(map);
3483 		UVMHIST_LOG(maphist, "<- done (RV=%d)", rv,0,0,0);
3484 		return (rv);
3485 	}
3486 
3487 	if ((lockflags & UVM_LK_EXIT) == 0) {
3488 		vm_map_unbusy(map);
3489 	} else {
3490 
3491 		/*
3492 		 * Get back to an exclusive (write) lock.
3493 		 */
3494 
3495 		vm_map_lock(map);
3496 		vm_map_unbusy(map);
3497 	}
3498 
3499 	UVMHIST_LOG(maphist,"<- done (OK WIRE)",0,0,0,0);
3500 	return 0;
3501 }
3502 
3503 /*
3504  * uvm_map_pageable_all: special case of uvm_map_pageable - affects
3505  * all mapped regions.
3506  *
3507  * => map must not be locked.
3508  * => if no flags are specified, all regions are unwired.
3509  * => XXXJRT: has some of the same problems as uvm_map_pageable() above.
3510  */
3511 
3512 int
3513 uvm_map_pageable_all(struct vm_map *map, int flags, vsize_t limit)
3514 {
3515 	struct vm_map_entry *entry, *failed_entry;
3516 	vsize_t size;
3517 	int rv;
3518 #ifdef DIAGNOSTIC
3519 	u_int timestamp_save;
3520 #endif
3521 	UVMHIST_FUNC("uvm_map_pageable_all"); UVMHIST_CALLED(maphist);
3522 	UVMHIST_LOG(maphist,"(map=%p,flags=%#x)", map, flags, 0, 0);
3523 
3524 	KASSERT(map->flags & VM_MAP_PAGEABLE);
3525 
3526 	vm_map_lock(map);
3527 
3528 	/*
3529 	 * handle wiring and unwiring separately.
3530 	 */
3531 
3532 	if (flags == 0) {			/* unwire */
3533 
3534 		/*
3535 		 * POSIX 1003.1b -- munlockall unlocks all regions,
3536 		 * regardless of how many times mlockall has been called.
3537 		 */
3538 
3539 		for (entry = map->header.next; entry != &map->header;
3540 		     entry = entry->next) {
3541 			if (VM_MAPENT_ISWIRED(entry))
3542 				uvm_map_entry_unwire(map, entry);
3543 		}
3544 		map->flags &= ~VM_MAP_WIREFUTURE;
3545 		vm_map_unlock(map);
3546 		UVMHIST_LOG(maphist,"<- done (OK UNWIRE)",0,0,0,0);
3547 		return 0;
3548 	}
3549 
3550 	if (flags & MCL_FUTURE) {
3551 
3552 		/*
3553 		 * must wire all future mappings; remember this.
3554 		 */
3555 
3556 		map->flags |= VM_MAP_WIREFUTURE;
3557 	}
3558 
3559 	if ((flags & MCL_CURRENT) == 0) {
3560 
3561 		/*
3562 		 * no more work to do!
3563 		 */
3564 
3565 		UVMHIST_LOG(maphist,"<- done (OK no wire)",0,0,0,0);
3566 		vm_map_unlock(map);
3567 		return 0;
3568 	}
3569 
3570 	/*
3571 	 * wire case: in three passes [XXXCDC: ugly block of code here]
3572 	 *
3573 	 * 1: holding the write lock, count all pages mapped by non-wired
3574 	 *    entries.  if this would cause us to go over our limit, we fail.
3575 	 *
3576 	 * 2: still holding the write lock, we create any anonymous maps that
3577 	 *    need to be created.  then we increment its wiring count.
3578 	 *
3579 	 * 3: we downgrade to a read lock, and call uvm_fault_wire to fault
3580 	 *    in the pages for any newly wired area (wired_count == 1).
3581 	 *
3582 	 *    downgrading to a read lock for uvm_fault_wire avoids a possible
3583 	 *    deadlock with another thread that may have faulted on one of
3584 	 *    the pages to be wired (it would mark the page busy, blocking
3585 	 *    us, then in turn block on the map lock that we hold).  because
3586 	 *    of problems in the recursive lock package, we cannot upgrade
3587 	 *    to a write lock in vm_map_lookup.  thus, any actions that
3588 	 *    require the write lock must be done beforehand.  because we
3589 	 *    keep the read lock on the map, the copy-on-write status of the
3590 	 *    entries we modify here cannot change.
3591 	 */
3592 
3593 	for (size = 0, entry = map->header.next; entry != &map->header;
3594 	     entry = entry->next) {
3595 		if (entry->protection != VM_PROT_NONE &&
3596 		    VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */
3597 			size += entry->end - entry->start;
3598 		}
3599 	}
3600 
3601 	if (atop(size) + uvmexp.wired > uvmexp.wiredmax) {
3602 		vm_map_unlock(map);
3603 		return ENOMEM;
3604 	}
3605 
3606 	if (limit != 0 &&
3607 	    (size + ptoa(pmap_wired_count(vm_map_pmap(map))) > limit)) {
3608 		vm_map_unlock(map);
3609 		return ENOMEM;
3610 	}
3611 
3612 	/*
3613 	 * Pass 2.
3614 	 */
3615 
3616 	for (entry = map->header.next; entry != &map->header;
3617 	     entry = entry->next) {
3618 		if (entry->protection == VM_PROT_NONE)
3619 			continue;
3620 		if (VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */
3621 
3622 			/*
3623 			 * perform actions of vm_map_lookup that need the
3624 			 * write lock on the map: create an anonymous map
3625 			 * for a copy-on-write region, or an anonymous map
3626 			 * for a zero-fill region.  (XXXCDC: submap case
3627 			 * ok?)
3628 			 */
3629 
3630 			if (!UVM_ET_ISSUBMAP(entry)) {	/* not submap */
3631 				if (UVM_ET_ISNEEDSCOPY(entry) &&
3632 				    ((entry->max_protection & VM_PROT_WRITE) ||
3633 				     (entry->object.uvm_obj == NULL))) {
3634 					amap_copy(map, entry, 0, entry->start,
3635 					    entry->end);
3636 					/* XXXCDC: wait OK? */
3637 				}
3638 			}
3639 		}
3640 		entry->wired_count++;
3641 	}
3642 
3643 	/*
3644 	 * Pass 3.
3645 	 */
3646 
3647 #ifdef DIAGNOSTIC
3648 	timestamp_save = map->timestamp;
3649 #endif
3650 	vm_map_busy(map);
3651 	vm_map_unlock(map);
3652 
3653 	rv = 0;
3654 	for (entry = map->header.next; entry != &map->header;
3655 	     entry = entry->next) {
3656 		if (entry->wired_count == 1) {
3657 			rv = uvm_fault_wire(map, entry->start, entry->end,
3658 			    entry->max_protection, 1);
3659 			if (rv) {
3660 
3661 				/*
3662 				 * wiring failed.  break out of the loop.
3663 				 * we'll clean up the map below, once we
3664 				 * have a write lock again.
3665 				 */
3666 
3667 				break;
3668 			}
3669 		}
3670 	}
3671 
3672 	if (rv) {
3673 
3674 		/*
3675 		 * Get back an exclusive (write) lock.
3676 		 */
3677 
3678 		vm_map_lock(map);
3679 		vm_map_unbusy(map);
3680 
3681 #ifdef DIAGNOSTIC
3682 		if (timestamp_save + 1 != map->timestamp)
3683 			panic("uvm_map_pageable_all: stale map");
3684 #endif
3685 
3686 		/*
3687 		 * first drop the wiring count on all the entries
3688 		 * which haven't actually been wired yet.
3689 		 *
3690 		 * Skip VM_PROT_NONE entries like we did above.
3691 		 */
3692 
3693 		failed_entry = entry;
3694 		for (/* nothing */; entry != &map->header;
3695 		     entry = entry->next) {
3696 			if (entry->protection == VM_PROT_NONE)
3697 				continue;
3698 			entry->wired_count--;
3699 		}
3700 
3701 		/*
3702 		 * now, unwire all the entries that were successfully
3703 		 * wired above.
3704 		 *
3705 		 * Skip VM_PROT_NONE entries like we did above.
3706 		 */
3707 
3708 		for (entry = map->header.next; entry != failed_entry;
3709 		     entry = entry->next) {
3710 			if (entry->protection == VM_PROT_NONE)
3711 				continue;
3712 			entry->wired_count--;
3713 			if (VM_MAPENT_ISWIRED(entry))
3714 				uvm_map_entry_unwire(map, entry);
3715 		}
3716 		vm_map_unlock(map);
3717 		UVMHIST_LOG(maphist,"<- done (RV=%d)", rv,0,0,0);
3718 		return (rv);
3719 	}
3720 
3721 	vm_map_unbusy(map);
3722 
3723 	UVMHIST_LOG(maphist,"<- done (OK WIRE)",0,0,0,0);
3724 	return 0;
3725 }
3726 
3727 /*
3728  * uvm_map_clean: clean out a map range
3729  *
3730  * => valid flags:
3731  *   if (flags & PGO_CLEANIT): dirty pages are cleaned first
3732  *   if (flags & PGO_SYNCIO): dirty pages are written synchronously
3733  *   if (flags & PGO_DEACTIVATE): any cached pages are deactivated after clean
3734  *   if (flags & PGO_FREE): any cached pages are freed after clean
3735  * => returns an error if any part of the specified range isn't mapped
3736  * => never a need to flush amap layer since the anonymous memory has
3737  *	no permanent home, but may deactivate pages there
3738  * => called from sys_msync() and sys_madvise()
3739  * => caller must not write-lock map (read OK).
3740  * => we may sleep while cleaning if SYNCIO [with map read-locked]
3741  */
3742 
3743 int
3744 uvm_map_clean(struct vm_map *map, vaddr_t start, vaddr_t end, int flags)
3745 {
3746 	struct vm_map_entry *current, *entry;
3747 	struct uvm_object *uobj;
3748 	struct vm_amap *amap;
3749 	struct vm_anon *anon, *anon_tofree;
3750 	struct vm_page *pg;
3751 	vaddr_t offset;
3752 	vsize_t size;
3753 	voff_t uoff;
3754 	int error, refs;
3755 	UVMHIST_FUNC("uvm_map_clean"); UVMHIST_CALLED(maphist);
3756 
3757 	UVMHIST_LOG(maphist,"(map=%p,start=%#lx,end=%#lx,flags=%#x)",
3758 		    map, start, end, flags);
3759 	KASSERT((flags & (PGO_FREE|PGO_DEACTIVATE)) !=
3760 		(PGO_FREE|PGO_DEACTIVATE));
3761 
3762 	vm_map_lock_read(map);
3763 	VM_MAP_RANGE_CHECK(map, start, end);
3764 	if (uvm_map_lookup_entry(map, start, &entry) == false) {
3765 		vm_map_unlock_read(map);
3766 		return EFAULT;
3767 	}
3768 
3769 	/*
3770 	 * Make a first pass to check for holes and wiring problems.
3771 	 */
3772 
3773 	for (current = entry; current->start < end; current = current->next) {
3774 		if (UVM_ET_ISSUBMAP(current)) {
3775 			vm_map_unlock_read(map);
3776 			return EINVAL;
3777 		}
3778 		if ((flags & PGO_FREE) != 0 && VM_MAPENT_ISWIRED(entry)) {
3779 			vm_map_unlock_read(map);
3780 			return EBUSY;
3781 		}
3782 		if (end <= current->end) {
3783 			break;
3784 		}
3785 		if (current->end != current->next->start) {
3786 			vm_map_unlock_read(map);
3787 			return EFAULT;
3788 		}
3789 	}
3790 
3791 	error = 0;
3792 	for (current = entry; start < end; current = current->next) {
3793 		amap = current->aref.ar_amap;	/* upper layer */
3794 		uobj = current->object.uvm_obj;	/* lower layer */
3795 		KASSERT(start >= current->start);
3796 
3797 		/*
3798 		 * No amap cleaning necessary if:
3799 		 *
3800 		 *	(1) There's no amap.
3801 		 *
3802 		 *	(2) We're not deactivating or freeing pages.
3803 		 */
3804 
3805 		if (amap == NULL || (flags & (PGO_DEACTIVATE|PGO_FREE)) == 0)
3806 			goto flush_object;
3807 
3808 		offset = start - current->start;
3809 		size = MIN(end, current->end) - start;
3810 		anon_tofree = NULL;
3811 
3812 		amap_lock(amap);
3813 		for ( ; size != 0; size -= PAGE_SIZE, offset += PAGE_SIZE) {
3814 			anon = amap_lookup(&current->aref, offset);
3815 			if (anon == NULL)
3816 				continue;
3817 
3818 			KASSERT(anon->an_lock == amap->am_lock);
3819 			pg = anon->an_page;
3820 			if (pg == NULL) {
3821 				continue;
3822 			}
3823 			if (pg->flags & PG_BUSY) {
3824 				continue;
3825 			}
3826 
3827 			switch (flags & (PGO_CLEANIT|PGO_FREE|PGO_DEACTIVATE)) {
3828 
3829 			/*
3830 			 * In these first 3 cases, we just deactivate the page.
3831 			 */
3832 
3833 			case PGO_CLEANIT|PGO_FREE:
3834 			case PGO_CLEANIT|PGO_DEACTIVATE:
3835 			case PGO_DEACTIVATE:
3836  deactivate_it:
3837 				/*
3838 				 * skip the page if it's loaned or wired,
3839 				 * since it shouldn't be on a paging queue
3840 				 * at all in these cases.
3841 				 */
3842 
3843 				mutex_enter(&uvm_pageqlock);
3844 				if (pg->loan_count != 0 ||
3845 				    pg->wire_count != 0) {
3846 					mutex_exit(&uvm_pageqlock);
3847 					continue;
3848 				}
3849 				KASSERT(pg->uanon == anon);
3850 				uvm_pagedeactivate(pg);
3851 				mutex_exit(&uvm_pageqlock);
3852 				continue;
3853 
3854 			case PGO_FREE:
3855 
3856 				/*
3857 				 * If there are multiple references to
3858 				 * the amap, just deactivate the page.
3859 				 */
3860 
3861 				if (amap_refs(amap) > 1)
3862 					goto deactivate_it;
3863 
3864 				/* skip the page if it's wired */
3865 				if (pg->wire_count != 0) {
3866 					continue;
3867 				}
3868 				amap_unadd(&current->aref, offset);
3869 				refs = --anon->an_ref;
3870 				if (refs == 0) {
3871 					anon->an_link = anon_tofree;
3872 					anon_tofree = anon;
3873 				}
3874 				continue;
3875 			}
3876 		}
3877 		uvm_anon_freelst(amap, anon_tofree);
3878 
3879  flush_object:
3880 		/*
3881 		 * flush pages if we've got a valid backing object.
3882 		 * note that we must always clean object pages before
3883 		 * freeing them since otherwise we could reveal stale
3884 		 * data from files.
3885 		 */
3886 
3887 		uoff = current->offset + (start - current->start);
3888 		size = MIN(end, current->end) - start;
3889 		if (uobj != NULL) {
3890 			mutex_enter(uobj->vmobjlock);
3891 			if (uobj->pgops->pgo_put != NULL)
3892 				error = (uobj->pgops->pgo_put)(uobj, uoff,
3893 				    uoff + size, flags | PGO_CLEANIT);
3894 			else
3895 				error = 0;
3896 		}
3897 		start += size;
3898 	}
3899 	vm_map_unlock_read(map);
3900 	return (error);
3901 }
3902 
3903 
3904 /*
3905  * uvm_map_checkprot: check protection in map
3906  *
3907  * => must allow specified protection in a fully allocated region.
3908  * => map must be read or write locked by caller.
3909  */
3910 
3911 bool
3912 uvm_map_checkprot(struct vm_map *map, vaddr_t start, vaddr_t end,
3913     vm_prot_t protection)
3914 {
3915 	struct vm_map_entry *entry;
3916 	struct vm_map_entry *tmp_entry;
3917 
3918 	if (!uvm_map_lookup_entry(map, start, &tmp_entry)) {
3919 		return (false);
3920 	}
3921 	entry = tmp_entry;
3922 	while (start < end) {
3923 		if (entry == &map->header) {
3924 			return (false);
3925 		}
3926 
3927 		/*
3928 		 * no holes allowed
3929 		 */
3930 
3931 		if (start < entry->start) {
3932 			return (false);
3933 		}
3934 
3935 		/*
3936 		 * check protection associated with entry
3937 		 */
3938 
3939 		if ((entry->protection & protection) != protection) {
3940 			return (false);
3941 		}
3942 		start = entry->end;
3943 		entry = entry->next;
3944 	}
3945 	return (true);
3946 }
3947 
3948 /*
3949  * uvmspace_alloc: allocate a vmspace structure.
3950  *
3951  * - structure includes vm_map and pmap
3952  * - XXX: no locking on this structure
3953  * - refcnt set to 1, rest must be init'd by caller
3954  */
3955 struct vmspace *
3956 uvmspace_alloc(vaddr_t vmin, vaddr_t vmax, bool topdown)
3957 {
3958 	struct vmspace *vm;
3959 	UVMHIST_FUNC("uvmspace_alloc"); UVMHIST_CALLED(maphist);
3960 
3961 	vm = pool_cache_get(&uvm_vmspace_cache, PR_WAITOK);
3962 	uvmspace_init(vm, NULL, vmin, vmax, topdown);
3963 	UVMHIST_LOG(maphist,"<- done (vm=%p)", vm,0,0,0);
3964 	return (vm);
3965 }
3966 
3967 /*
3968  * uvmspace_init: initialize a vmspace structure.
3969  *
3970  * - XXX: no locking on this structure
3971  * - refcnt set to 1, rest must be init'd by caller
3972  */
3973 void
3974 uvmspace_init(struct vmspace *vm, struct pmap *pmap, vaddr_t vmin,
3975     vaddr_t vmax, bool topdown)
3976 {
3977 	UVMHIST_FUNC("uvmspace_init"); UVMHIST_CALLED(maphist);
3978 
3979 	UVMHIST_LOG(maphist, "(vm=%p, pmap=%p, vmin=%#lx, vmax=%#lx",
3980 	    vm, pmap, vmin, vmax);
3981 	UVMHIST_LOG(maphist, "   topdown=%u)", topdown, 0, 0, 0);
3982 
3983 	memset(vm, 0, sizeof(*vm));
3984 	uvm_map_setup(&vm->vm_map, vmin, vmax, VM_MAP_PAGEABLE
3985 	    | (topdown ? VM_MAP_TOPDOWN : 0)
3986 	    );
3987 	if (pmap)
3988 		pmap_reference(pmap);
3989 	else
3990 		pmap = pmap_create();
3991 	vm->vm_map.pmap = pmap;
3992 	vm->vm_refcnt = 1;
3993 	UVMHIST_LOG(maphist,"<- done",0,0,0,0);
3994 }
3995 
3996 /*
3997  * uvmspace_share: share a vmspace between two processes
3998  *
3999  * - used for vfork, threads(?)
4000  */
4001 
4002 void
4003 uvmspace_share(struct proc *p1, struct proc *p2)
4004 {
4005 
4006 	uvmspace_addref(p1->p_vmspace);
4007 	p2->p_vmspace = p1->p_vmspace;
4008 }
4009 
4010 #if 0
4011 
4012 /*
4013  * uvmspace_unshare: ensure that process "p" has its own, unshared, vmspace
4014  *
4015  * - XXX: no locking on vmspace
4016  */
4017 
4018 void
4019 uvmspace_unshare(struct lwp *l)
4020 {
4021 	struct proc *p = l->l_proc;
4022 	struct vmspace *nvm, *ovm = p->p_vmspace;
4023 
4024 	if (ovm->vm_refcnt == 1)
4025 		/* nothing to do: vmspace isn't shared in the first place */
4026 		return;
4027 
4028 	/* make a new vmspace, still holding old one */
4029 	nvm = uvmspace_fork(ovm);
4030 
4031 	kpreempt_disable();
4032 	pmap_deactivate(l);		/* unbind old vmspace */
4033 	p->p_vmspace = nvm;
4034 	pmap_activate(l);		/* switch to new vmspace */
4035 	kpreempt_enable();
4036 
4037 	uvmspace_free(ovm);		/* drop reference to old vmspace */
4038 }
4039 
4040 #endif
4041 
4042 
4043 /*
4044  * uvmspace_spawn: a new process has been spawned and needs a vmspace
4045  */
4046 
4047 void
4048 uvmspace_spawn(struct lwp *l, vaddr_t start, vaddr_t end, bool topdown)
4049 {
4050 	struct proc *p = l->l_proc;
4051 	struct vmspace *nvm;
4052 
4053 #ifdef __HAVE_CPU_VMSPACE_EXEC
4054 	cpu_vmspace_exec(l, start, end);
4055 #endif
4056 
4057 	nvm = uvmspace_alloc(start, end, topdown);
4058 	kpreempt_disable();
4059 	p->p_vmspace = nvm;
4060 	pmap_activate(l);
4061 	kpreempt_enable();
4062 }
4063 
4064 /*
4065  * uvmspace_exec: the process wants to exec a new program
4066  */
4067 
4068 void
4069 uvmspace_exec(struct lwp *l, vaddr_t start, vaddr_t end, bool topdown)
4070 {
4071 	struct proc *p = l->l_proc;
4072 	struct vmspace *nvm, *ovm = p->p_vmspace;
4073 	struct vm_map *map;
4074 
4075 	KASSERT(ovm != NULL);
4076 #ifdef __HAVE_CPU_VMSPACE_EXEC
4077 	cpu_vmspace_exec(l, start, end);
4078 #endif
4079 
4080 	map = &ovm->vm_map;
4081 	/*
4082 	 * see if more than one process is using this vmspace...
4083 	 */
4084 
4085 	if (ovm->vm_refcnt == 1
4086 	    && topdown == ((ovm->vm_map.flags & VM_MAP_TOPDOWN) != 0)) {
4087 
4088 		/*
4089 		 * if p is the only process using its vmspace then we can safely
4090 		 * recycle that vmspace for the program that is being exec'd.
4091 		 * But only if TOPDOWN matches the requested value for the new
4092 		 * vm space!
4093 		 */
4094 
4095 		/*
4096 		 * SYSV SHM semantics require us to kill all segments on an exec
4097 		 */
4098 		if (uvm_shmexit && ovm->vm_shm)
4099 			(*uvm_shmexit)(ovm);
4100 
4101 		/*
4102 		 * POSIX 1003.1b -- "lock future mappings" is revoked
4103 		 * when a process execs another program image.
4104 		 */
4105 
4106 		map->flags &= ~VM_MAP_WIREFUTURE;
4107 
4108 		/*
4109 		 * now unmap the old program
4110 		 */
4111 
4112 		pmap_remove_all(map->pmap);
4113 		uvm_unmap(map, vm_map_min(map), vm_map_max(map));
4114 		KASSERT(map->header.prev == &map->header);
4115 		KASSERT(map->nentries == 0);
4116 
4117 		/*
4118 		 * resize the map
4119 		 */
4120 
4121 		vm_map_setmin(map, start);
4122 		vm_map_setmax(map, end);
4123 	} else {
4124 
4125 		/*
4126 		 * p's vmspace is being shared, so we can't reuse it for p since
4127 		 * it is still being used for others.   allocate a new vmspace
4128 		 * for p
4129 		 */
4130 
4131 		nvm = uvmspace_alloc(start, end, topdown);
4132 
4133 		/*
4134 		 * install new vmspace and drop our ref to the old one.
4135 		 */
4136 
4137 		kpreempt_disable();
4138 		pmap_deactivate(l);
4139 		p->p_vmspace = nvm;
4140 		pmap_activate(l);
4141 		kpreempt_enable();
4142 
4143 		uvmspace_free(ovm);
4144 	}
4145 }
4146 
4147 /*
4148  * uvmspace_addref: add a referece to a vmspace.
4149  */
4150 
4151 void
4152 uvmspace_addref(struct vmspace *vm)
4153 {
4154 	struct vm_map *map = &vm->vm_map;
4155 
4156 	KASSERT((map->flags & VM_MAP_DYING) == 0);
4157 
4158 	mutex_enter(&map->misc_lock);
4159 	KASSERT(vm->vm_refcnt > 0);
4160 	vm->vm_refcnt++;
4161 	mutex_exit(&map->misc_lock);
4162 }
4163 
4164 /*
4165  * uvmspace_free: free a vmspace data structure
4166  */
4167 
4168 void
4169 uvmspace_free(struct vmspace *vm)
4170 {
4171 	struct vm_map_entry *dead_entries;
4172 	struct vm_map *map = &vm->vm_map;
4173 	int n;
4174 
4175 	UVMHIST_FUNC("uvmspace_free"); UVMHIST_CALLED(maphist);
4176 
4177 	UVMHIST_LOG(maphist,"(vm=%p) ref=%d", vm, vm->vm_refcnt,0,0);
4178 	mutex_enter(&map->misc_lock);
4179 	n = --vm->vm_refcnt;
4180 	mutex_exit(&map->misc_lock);
4181 	if (n > 0)
4182 		return;
4183 
4184 	/*
4185 	 * at this point, there should be no other references to the map.
4186 	 * delete all of the mappings, then destroy the pmap.
4187 	 */
4188 
4189 	map->flags |= VM_MAP_DYING;
4190 	pmap_remove_all(map->pmap);
4191 
4192 	/* Get rid of any SYSV shared memory segments. */
4193 	if (uvm_shmexit && vm->vm_shm != NULL)
4194 		(*uvm_shmexit)(vm);
4195 
4196 	if (map->nentries) {
4197 		uvm_unmap_remove(map, vm_map_min(map), vm_map_max(map),
4198 		    &dead_entries, 0);
4199 		if (dead_entries != NULL)
4200 			uvm_unmap_detach(dead_entries, 0);
4201 	}
4202 	KASSERT(map->nentries == 0);
4203 	KASSERT(map->size == 0);
4204 
4205 	mutex_destroy(&map->misc_lock);
4206 	rw_destroy(&map->lock);
4207 	cv_destroy(&map->cv);
4208 	pmap_destroy(map->pmap);
4209 	pool_cache_put(&uvm_vmspace_cache, vm);
4210 }
4211 
4212 static struct vm_map_entry *
4213 uvm_mapent_clone(struct vm_map *new_map, struct vm_map_entry *old_entry,
4214     int flags)
4215 {
4216 	struct vm_map_entry *new_entry;
4217 
4218 	new_entry = uvm_mapent_alloc(new_map, 0);
4219 	/* old_entry -> new_entry */
4220 	uvm_mapent_copy(old_entry, new_entry);
4221 
4222 	/* new pmap has nothing wired in it */
4223 	new_entry->wired_count = 0;
4224 
4225 	/*
4226 	 * gain reference to object backing the map (can't
4227 	 * be a submap, already checked this case).
4228 	 */
4229 
4230 	if (new_entry->aref.ar_amap)
4231 		uvm_map_reference_amap(new_entry, flags);
4232 
4233 	if (new_entry->object.uvm_obj &&
4234 	    new_entry->object.uvm_obj->pgops->pgo_reference)
4235 		new_entry->object.uvm_obj->pgops->pgo_reference(
4236 			new_entry->object.uvm_obj);
4237 
4238 	/* insert entry at end of new_map's entry list */
4239 	uvm_map_entry_link(new_map, new_map->header.prev,
4240 	    new_entry);
4241 
4242 	return new_entry;
4243 }
4244 
4245 /*
4246  * share the mapping: this means we want the old and
4247  * new entries to share amaps and backing objects.
4248  */
4249 static void
4250 uvm_mapent_forkshared(struct vm_map *new_map, struct vm_map *old_map,
4251     struct vm_map_entry *old_entry)
4252 {
4253 	/*
4254 	 * if the old_entry needs a new amap (due to prev fork)
4255 	 * then we need to allocate it now so that we have
4256 	 * something we own to share with the new_entry.   [in
4257 	 * other words, we need to clear needs_copy]
4258 	 */
4259 
4260 	if (UVM_ET_ISNEEDSCOPY(old_entry)) {
4261 		/* get our own amap, clears needs_copy */
4262 		amap_copy(old_map, old_entry, AMAP_COPY_NOCHUNK,
4263 		    0, 0);
4264 		/* XXXCDC: WAITOK??? */
4265 	}
4266 
4267 	uvm_mapent_clone(new_map, old_entry, AMAP_SHARED);
4268 }
4269 
4270 
4271 static void
4272 uvm_mapent_forkcopy(struct vm_map *new_map, struct vm_map *old_map,
4273     struct vm_map_entry *old_entry)
4274 {
4275 	struct vm_map_entry *new_entry;
4276 
4277 	/*
4278 	 * copy-on-write the mapping (using mmap's
4279 	 * MAP_PRIVATE semantics)
4280 	 *
4281 	 * allocate new_entry, adjust reference counts.
4282 	 * (note that new references are read-only).
4283 	 */
4284 
4285 	new_entry = uvm_mapent_clone(new_map, old_entry, 0);
4286 
4287 	new_entry->etype |=
4288 	    (UVM_ET_COPYONWRITE|UVM_ET_NEEDSCOPY);
4289 
4290 	/*
4291 	 * the new entry will need an amap.  it will either
4292 	 * need to be copied from the old entry or created
4293 	 * from scratch (if the old entry does not have an
4294 	 * amap).  can we defer this process until later
4295 	 * (by setting "needs_copy") or do we need to copy
4296 	 * the amap now?
4297 	 *
4298 	 * we must copy the amap now if any of the following
4299 	 * conditions hold:
4300 	 * 1. the old entry has an amap and that amap is
4301 	 *    being shared.  this means that the old (parent)
4302 	 *    process is sharing the amap with another
4303 	 *    process.  if we do not clear needs_copy here
4304 	 *    we will end up in a situation where both the
4305 	 *    parent and child process are refering to the
4306 	 *    same amap with "needs_copy" set.  if the
4307 	 *    parent write-faults, the fault routine will
4308 	 *    clear "needs_copy" in the parent by allocating
4309 	 *    a new amap.   this is wrong because the
4310 	 *    parent is supposed to be sharing the old amap
4311 	 *    and the new amap will break that.
4312 	 *
4313 	 * 2. if the old entry has an amap and a non-zero
4314 	 *    wire count then we are going to have to call
4315 	 *    amap_cow_now to avoid page faults in the
4316 	 *    parent process.   since amap_cow_now requires
4317 	 *    "needs_copy" to be clear we might as well
4318 	 *    clear it here as well.
4319 	 *
4320 	 */
4321 
4322 	if (old_entry->aref.ar_amap != NULL) {
4323 		if ((amap_flags(old_entry->aref.ar_amap) & AMAP_SHARED) != 0 ||
4324 		    VM_MAPENT_ISWIRED(old_entry)) {
4325 
4326 			amap_copy(new_map, new_entry,
4327 			    AMAP_COPY_NOCHUNK, 0, 0);
4328 			/* XXXCDC: M_WAITOK ... ok? */
4329 		}
4330 	}
4331 
4332 	/*
4333 	 * if the parent's entry is wired down, then the
4334 	 * parent process does not want page faults on
4335 	 * access to that memory.  this means that we
4336 	 * cannot do copy-on-write because we can't write
4337 	 * protect the old entry.   in this case we
4338 	 * resolve all copy-on-write faults now, using
4339 	 * amap_cow_now.   note that we have already
4340 	 * allocated any needed amap (above).
4341 	 */
4342 
4343 	if (VM_MAPENT_ISWIRED(old_entry)) {
4344 
4345 		/*
4346 		 * resolve all copy-on-write faults now
4347 		 * (note that there is nothing to do if
4348 		 * the old mapping does not have an amap).
4349 		 */
4350 		if (old_entry->aref.ar_amap)
4351 			amap_cow_now(new_map, new_entry);
4352 
4353 	} else {
4354 		/*
4355 		 * setup mappings to trigger copy-on-write faults
4356 		 * we must write-protect the parent if it has
4357 		 * an amap and it is not already "needs_copy"...
4358 		 * if it is already "needs_copy" then the parent
4359 		 * has already been write-protected by a previous
4360 		 * fork operation.
4361 		 */
4362 		if (old_entry->aref.ar_amap &&
4363 		    !UVM_ET_ISNEEDSCOPY(old_entry)) {
4364 			if (old_entry->max_protection & VM_PROT_WRITE) {
4365 				pmap_protect(old_map->pmap,
4366 				    old_entry->start, old_entry->end,
4367 				    old_entry->protection & ~VM_PROT_WRITE);
4368 			}
4369 			old_entry->etype |= UVM_ET_NEEDSCOPY;
4370 		}
4371 	}
4372 }
4373 
4374 /*
4375  * zero the mapping: the new entry will be zero initialized
4376  */
4377 static void
4378 uvm_mapent_forkzero(struct vm_map *new_map, struct vm_map *old_map,
4379     struct vm_map_entry *old_entry)
4380 {
4381 	struct vm_map_entry *new_entry;
4382 
4383 	new_entry = uvm_mapent_clone(new_map, old_entry, 0);
4384 
4385 	new_entry->etype |=
4386 	    (UVM_ET_COPYONWRITE|UVM_ET_NEEDSCOPY);
4387 
4388 	if (new_entry->aref.ar_amap) {
4389 		uvm_map_unreference_amap(new_entry, 0);
4390 		new_entry->aref.ar_pageoff = 0;
4391 		new_entry->aref.ar_amap = NULL;
4392 	}
4393 
4394 	if (UVM_ET_ISOBJ(new_entry)) {
4395 		if (new_entry->object.uvm_obj->pgops->pgo_detach)
4396 			new_entry->object.uvm_obj->pgops->pgo_detach(
4397 			    new_entry->object.uvm_obj);
4398 		new_entry->object.uvm_obj = NULL;
4399 		new_entry->etype &= ~UVM_ET_OBJ;
4400 	}
4401 }
4402 
4403 /*
4404  *   F O R K   -   m a i n   e n t r y   p o i n t
4405  */
4406 /*
4407  * uvmspace_fork: fork a process' main map
4408  *
4409  * => create a new vmspace for child process from parent.
4410  * => parent's map must not be locked.
4411  */
4412 
4413 struct vmspace *
4414 uvmspace_fork(struct vmspace *vm1)
4415 {
4416 	struct vmspace *vm2;
4417 	struct vm_map *old_map = &vm1->vm_map;
4418 	struct vm_map *new_map;
4419 	struct vm_map_entry *old_entry;
4420 	UVMHIST_FUNC("uvmspace_fork"); UVMHIST_CALLED(maphist);
4421 
4422 	vm_map_lock(old_map);
4423 
4424 	vm2 = uvmspace_alloc(vm_map_min(old_map), vm_map_max(old_map),
4425 	    vm1->vm_map.flags & VM_MAP_TOPDOWN);
4426 	memcpy(&vm2->vm_startcopy, &vm1->vm_startcopy,
4427 	    (char *) (vm1 + 1) - (char *) &vm1->vm_startcopy);
4428 	new_map = &vm2->vm_map;		  /* XXX */
4429 
4430 	old_entry = old_map->header.next;
4431 	new_map->size = old_map->size;
4432 
4433 	/*
4434 	 * go entry-by-entry
4435 	 */
4436 
4437 	while (old_entry != &old_map->header) {
4438 
4439 		/*
4440 		 * first, some sanity checks on the old entry
4441 		 */
4442 
4443 		KASSERT(!UVM_ET_ISSUBMAP(old_entry));
4444 		KASSERT(UVM_ET_ISCOPYONWRITE(old_entry) ||
4445 			!UVM_ET_ISNEEDSCOPY(old_entry));
4446 
4447 		switch (old_entry->inheritance) {
4448 		case MAP_INHERIT_NONE:
4449 			/*
4450 			 * drop the mapping, modify size
4451 			 */
4452 			new_map->size -= old_entry->end - old_entry->start;
4453 			break;
4454 
4455 		case MAP_INHERIT_SHARE:
4456 			uvm_mapent_forkshared(new_map, old_map, old_entry);
4457 			break;
4458 
4459 		case MAP_INHERIT_COPY:
4460 			uvm_mapent_forkcopy(new_map, old_map, old_entry);
4461 			break;
4462 
4463 		case MAP_INHERIT_ZERO:
4464 			uvm_mapent_forkzero(new_map, old_map, old_entry);
4465 			break;
4466 		default:
4467 			KASSERT(0);
4468 			break;
4469 		}
4470 		old_entry = old_entry->next;
4471 	}
4472 
4473 	pmap_update(old_map->pmap);
4474 	vm_map_unlock(old_map);
4475 
4476 	if (uvm_shmfork && vm1->vm_shm)
4477 		(*uvm_shmfork)(vm1, vm2);
4478 
4479 #ifdef PMAP_FORK
4480 	pmap_fork(vm1->vm_map.pmap, vm2->vm_map.pmap);
4481 #endif
4482 
4483 	UVMHIST_LOG(maphist,"<- done",0,0,0,0);
4484 	return (vm2);
4485 }
4486 
4487 
4488 /*
4489  * uvm_mapent_trymerge: try to merge an entry with its neighbors.
4490  *
4491  * => called with map locked.
4492  * => return non zero if successfully merged.
4493  */
4494 
4495 int
4496 uvm_mapent_trymerge(struct vm_map *map, struct vm_map_entry *entry, int flags)
4497 {
4498 	struct uvm_object *uobj;
4499 	struct vm_map_entry *next;
4500 	struct vm_map_entry *prev;
4501 	vsize_t size;
4502 	int merged = 0;
4503 	bool copying;
4504 	int newetype;
4505 
4506 	if (entry->aref.ar_amap != NULL) {
4507 		return 0;
4508 	}
4509 	if ((entry->flags & UVM_MAP_NOMERGE) != 0) {
4510 		return 0;
4511 	}
4512 
4513 	uobj = entry->object.uvm_obj;
4514 	size = entry->end - entry->start;
4515 	copying = (flags & UVM_MERGE_COPYING) != 0;
4516 	newetype = copying ? (entry->etype & ~UVM_ET_NEEDSCOPY) : entry->etype;
4517 
4518 	next = entry->next;
4519 	if (next != &map->header &&
4520 	    next->start == entry->end &&
4521 	    ((copying && next->aref.ar_amap != NULL &&
4522 	    amap_refs(next->aref.ar_amap) == 1) ||
4523 	    (!copying && next->aref.ar_amap == NULL)) &&
4524 	    UVM_ET_ISCOMPATIBLE(next, newetype,
4525 	    uobj, entry->flags, entry->protection,
4526 	    entry->max_protection, entry->inheritance, entry->advice,
4527 	    entry->wired_count) &&
4528 	    (uobj == NULL || entry->offset + size == next->offset)) {
4529 		int error;
4530 
4531 		if (copying) {
4532 			error = amap_extend(next, size,
4533 			    AMAP_EXTEND_NOWAIT|AMAP_EXTEND_BACKWARDS);
4534 		} else {
4535 			error = 0;
4536 		}
4537 		if (error == 0) {
4538 			if (uobj) {
4539 				if (uobj->pgops->pgo_detach) {
4540 					uobj->pgops->pgo_detach(uobj);
4541 				}
4542 			}
4543 
4544 			entry->end = next->end;
4545 			clear_hints(map, next);
4546 			uvm_map_entry_unlink(map, next);
4547 			if (copying) {
4548 				entry->aref = next->aref;
4549 				entry->etype &= ~UVM_ET_NEEDSCOPY;
4550 			}
4551 			uvm_map_check(map, "trymerge forwardmerge");
4552 			uvm_mapent_free(next);
4553 			merged++;
4554 		}
4555 	}
4556 
4557 	prev = entry->prev;
4558 	if (prev != &map->header &&
4559 	    prev->end == entry->start &&
4560 	    ((copying && !merged && prev->aref.ar_amap != NULL &&
4561 	    amap_refs(prev->aref.ar_amap) == 1) ||
4562 	    (!copying && prev->aref.ar_amap == NULL)) &&
4563 	    UVM_ET_ISCOMPATIBLE(prev, newetype,
4564 	    uobj, entry->flags, entry->protection,
4565 	    entry->max_protection, entry->inheritance, entry->advice,
4566 	    entry->wired_count) &&
4567 	    (uobj == NULL ||
4568 	    prev->offset + prev->end - prev->start == entry->offset)) {
4569 		int error;
4570 
4571 		if (copying) {
4572 			error = amap_extend(prev, size,
4573 			    AMAP_EXTEND_NOWAIT|AMAP_EXTEND_FORWARDS);
4574 		} else {
4575 			error = 0;
4576 		}
4577 		if (error == 0) {
4578 			if (uobj) {
4579 				if (uobj->pgops->pgo_detach) {
4580 					uobj->pgops->pgo_detach(uobj);
4581 				}
4582 				entry->offset = prev->offset;
4583 			}
4584 
4585 			entry->start = prev->start;
4586 			clear_hints(map, prev);
4587 			uvm_map_entry_unlink(map, prev);
4588 			if (copying) {
4589 				entry->aref = prev->aref;
4590 				entry->etype &= ~UVM_ET_NEEDSCOPY;
4591 			}
4592 			uvm_map_check(map, "trymerge backmerge");
4593 			uvm_mapent_free(prev);
4594 			merged++;
4595 		}
4596 	}
4597 
4598 	return merged;
4599 }
4600 
4601 /*
4602  * uvm_map_setup: init map
4603  *
4604  * => map must not be in service yet.
4605  */
4606 
4607 void
4608 uvm_map_setup(struct vm_map *map, vaddr_t vmin, vaddr_t vmax, int flags)
4609 {
4610 
4611 	rb_tree_init(&map->rb_tree, &uvm_map_tree_ops);
4612 	map->header.next = map->header.prev = &map->header;
4613 	map->nentries = 0;
4614 	map->size = 0;
4615 	map->ref_count = 1;
4616 	vm_map_setmin(map, vmin);
4617 	vm_map_setmax(map, vmax);
4618 	map->flags = flags;
4619 	map->first_free = &map->header;
4620 	map->hint = &map->header;
4621 	map->timestamp = 0;
4622 	map->busy = NULL;
4623 
4624 	rw_init(&map->lock);
4625 	cv_init(&map->cv, "vm_map");
4626 	mutex_init(&map->misc_lock, MUTEX_DRIVER, IPL_NONE);
4627 }
4628 
4629 /*
4630  *   U N M A P   -   m a i n   e n t r y   p o i n t
4631  */
4632 
4633 /*
4634  * uvm_unmap1: remove mappings from a vm_map (from "start" up to "stop")
4635  *
4636  * => caller must check alignment and size
4637  * => map must be unlocked (we will lock it)
4638  * => flags is UVM_FLAG_QUANTUM or 0.
4639  */
4640 
4641 void
4642 uvm_unmap1(struct vm_map *map, vaddr_t start, vaddr_t end, int flags)
4643 {
4644 	struct vm_map_entry *dead_entries;
4645 	UVMHIST_FUNC("uvm_unmap"); UVMHIST_CALLED(maphist);
4646 
4647 	UVMHIST_LOG(maphist, "  (map=%p, start=%#lx, end=%#lx)",
4648 	    map, start, end, 0);
4649 	if (map == kernel_map) {
4650 		LOCKDEBUG_MEM_CHECK((void *)start, end - start);
4651 	}
4652 	/*
4653 	 * work now done by helper functions.   wipe the pmap's and then
4654 	 * detach from the dead entries...
4655 	 */
4656 	vm_map_lock(map);
4657 	uvm_unmap_remove(map, start, end, &dead_entries, flags);
4658 	vm_map_unlock(map);
4659 
4660 	if (dead_entries != NULL)
4661 		uvm_unmap_detach(dead_entries, 0);
4662 
4663 	UVMHIST_LOG(maphist, "<- done", 0,0,0,0);
4664 }
4665 
4666 
4667 /*
4668  * uvm_map_reference: add reference to a map
4669  *
4670  * => map need not be locked (we use misc_lock).
4671  */
4672 
4673 void
4674 uvm_map_reference(struct vm_map *map)
4675 {
4676 	mutex_enter(&map->misc_lock);
4677 	map->ref_count++;
4678 	mutex_exit(&map->misc_lock);
4679 }
4680 
4681 bool
4682 vm_map_starved_p(struct vm_map *map)
4683 {
4684 
4685 	if ((map->flags & VM_MAP_WANTVA) != 0) {
4686 		return true;
4687 	}
4688 	/* XXX */
4689 	if ((vm_map_max(map) - vm_map_min(map)) / 16 * 15 < map->size) {
4690 		return true;
4691 	}
4692 	return false;
4693 }
4694 
4695 void
4696 uvm_map_lock_entry(struct vm_map_entry *entry)
4697 {
4698 
4699 	if (entry->aref.ar_amap != NULL) {
4700 		amap_lock(entry->aref.ar_amap);
4701 	}
4702 	if (UVM_ET_ISOBJ(entry)) {
4703 		mutex_enter(entry->object.uvm_obj->vmobjlock);
4704 	}
4705 }
4706 
4707 void
4708 uvm_map_unlock_entry(struct vm_map_entry *entry)
4709 {
4710 
4711 	if (UVM_ET_ISOBJ(entry)) {
4712 		mutex_exit(entry->object.uvm_obj->vmobjlock);
4713 	}
4714 	if (entry->aref.ar_amap != NULL) {
4715 		amap_unlock(entry->aref.ar_amap);
4716 	}
4717 }
4718 
4719 #if defined(DDB) || defined(DEBUGPRINT)
4720 
4721 /*
4722  * uvm_map_printit: actually prints the map
4723  */
4724 
4725 void
4726 uvm_map_printit(struct vm_map *map, bool full,
4727     void (*pr)(const char *, ...))
4728 {
4729 	struct vm_map_entry *entry;
4730 
4731 	(*pr)("MAP %p: [%#lx->%#lx]\n", map, vm_map_min(map),
4732 	    vm_map_max(map));
4733 	(*pr)("\t#ent=%d, sz=%d, ref=%d, version=%d, flags=%#x\n",
4734 	    map->nentries, map->size, map->ref_count, map->timestamp,
4735 	    map->flags);
4736 	(*pr)("\tpmap=%p(resident=%ld, wired=%ld)\n", map->pmap,
4737 	    pmap_resident_count(map->pmap), pmap_wired_count(map->pmap));
4738 	if (!full)
4739 		return;
4740 	for (entry = map->header.next; entry != &map->header;
4741 	    entry = entry->next) {
4742 		(*pr)(" - %p: %#lx->%#lx: obj=%p/%#llx, amap=%p/%d\n",
4743 		    entry, entry->start, entry->end, entry->object.uvm_obj,
4744 		    (long long)entry->offset, entry->aref.ar_amap,
4745 		    entry->aref.ar_pageoff);
4746 		(*pr)(
4747 		    "\tsubmap=%c, cow=%c, nc=%c, prot(max)=%d/%d, inh=%d, "
4748 		    "wc=%d, adv=%d\n",
4749 		    (entry->etype & UVM_ET_SUBMAP) ? 'T' : 'F',
4750 		    (entry->etype & UVM_ET_COPYONWRITE) ? 'T' : 'F',
4751 		    (entry->etype & UVM_ET_NEEDSCOPY) ? 'T' : 'F',
4752 		    entry->protection, entry->max_protection,
4753 		    entry->inheritance, entry->wired_count, entry->advice);
4754 	}
4755 }
4756 
4757 void
4758 uvm_whatis(uintptr_t addr, void (*pr)(const char *, ...))
4759 {
4760 	struct vm_map *map;
4761 
4762 	for (map = kernel_map;;) {
4763 		struct vm_map_entry *entry;
4764 
4765 		if (!uvm_map_lookup_entry_bytree(map, (vaddr_t)addr, &entry)) {
4766 			break;
4767 		}
4768 		(*pr)("%p is %p+%zu from VMMAP %p\n",
4769 		    (void *)addr, (void *)entry->start,
4770 		    (size_t)(addr - (uintptr_t)entry->start), map);
4771 		if (!UVM_ET_ISSUBMAP(entry)) {
4772 			break;
4773 		}
4774 		map = entry->object.sub_map;
4775 	}
4776 }
4777 
4778 #endif /* DDB || DEBUGPRINT */
4779 
4780 #ifndef __USER_VA0_IS_SAFE
4781 static int
4782 sysctl_user_va0_disable(SYSCTLFN_ARGS)
4783 {
4784 	struct sysctlnode node;
4785 	int t, error;
4786 
4787 	node = *rnode;
4788 	node.sysctl_data = &t;
4789 	t = user_va0_disable;
4790 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
4791 	if (error || newp == NULL)
4792 		return (error);
4793 
4794 	if (!t && user_va0_disable &&
4795 	    kauth_authorize_system(l->l_cred, KAUTH_SYSTEM_MAP_VA_ZERO, 0,
4796 	    NULL, NULL, NULL))
4797 		return EPERM;
4798 
4799 	user_va0_disable = !!t;
4800 	return 0;
4801 }
4802 #endif
4803 
4804 static int
4805 fill_vmentry(struct lwp *l, struct proc *p, struct kinfo_vmentry *kve,
4806     struct vm_map *m, struct vm_map_entry *e)
4807 {
4808 #ifndef _RUMPKERNEL
4809 	int error;
4810 
4811 	memset(kve, 0, sizeof(*kve));
4812 	KASSERT(e != NULL);
4813 	if (UVM_ET_ISOBJ(e)) {
4814 		struct uvm_object *uobj = e->object.uvm_obj;
4815 		KASSERT(uobj != NULL);
4816 		kve->kve_ref_count = uobj->uo_refs;
4817 		kve->kve_count = uobj->uo_npages;
4818 		if (UVM_OBJ_IS_VNODE(uobj)) {
4819 			struct vattr va;
4820 			struct vnode *vp = (struct vnode *)uobj;
4821 			vn_lock(vp, LK_SHARED | LK_RETRY);
4822 			error = VOP_GETATTR(vp, &va, l->l_cred);
4823 			VOP_UNLOCK(vp);
4824 			kve->kve_type = KVME_TYPE_VNODE;
4825 			if (error == 0) {
4826 				kve->kve_vn_size = vp->v_size;
4827 				kve->kve_vn_type = (int)vp->v_type;
4828 				kve->kve_vn_mode = va.va_mode;
4829 				kve->kve_vn_rdev = va.va_rdev;
4830 				kve->kve_vn_fileid = va.va_fileid;
4831 				kve->kve_vn_fsid = va.va_fsid;
4832 				error = vnode_to_path(kve->kve_path,
4833 				    sizeof(kve->kve_path) / 2, vp, l, p);
4834 #ifdef DIAGNOSTIC
4835 				if (error)
4836 					printf("%s: vp %p error %d\n", __func__,
4837 						vp, error);
4838 #endif
4839 			}
4840 		} else if (UVM_OBJ_IS_KERN_OBJECT(uobj)) {
4841 			kve->kve_type = KVME_TYPE_KERN;
4842 		} else if (UVM_OBJ_IS_DEVICE(uobj)) {
4843 			kve->kve_type = KVME_TYPE_DEVICE;
4844 		} else if (UVM_OBJ_IS_AOBJ(uobj)) {
4845 			kve->kve_type = KVME_TYPE_ANON;
4846 		} else {
4847 			kve->kve_type = KVME_TYPE_OBJECT;
4848 		}
4849 	} else if (UVM_ET_ISSUBMAP(e)) {
4850 		struct vm_map *map = e->object.sub_map;
4851 		KASSERT(map != NULL);
4852 		kve->kve_ref_count = map->ref_count;
4853 		kve->kve_count = map->nentries;
4854 		kve->kve_type = KVME_TYPE_SUBMAP;
4855 	} else
4856 		kve->kve_type = KVME_TYPE_UNKNOWN;
4857 
4858 	kve->kve_start = e->start;
4859 	kve->kve_end = e->end;
4860 	kve->kve_offset = e->offset;
4861 	kve->kve_wired_count = e->wired_count;
4862 	kve->kve_inheritance = e->inheritance;
4863 	kve->kve_attributes = e->map_attrib;
4864 	kve->kve_advice = e->advice;
4865 #define PROT(p) (((p) & VM_PROT_READ) ? KVME_PROT_READ : 0) | \
4866 	(((p) & VM_PROT_WRITE) ? KVME_PROT_WRITE : 0) | \
4867 	(((p) & VM_PROT_EXECUTE) ? KVME_PROT_EXEC : 0)
4868 	kve->kve_protection = PROT(e->protection);
4869 	kve->kve_max_protection = PROT(e->max_protection);
4870 	kve->kve_flags |= (e->etype & UVM_ET_COPYONWRITE)
4871 	    ? KVME_FLAG_COW : 0;
4872 	kve->kve_flags |= (e->etype & UVM_ET_NEEDSCOPY)
4873 	    ? KVME_FLAG_NEEDS_COPY : 0;
4874 	kve->kve_flags |= (m->flags & VM_MAP_TOPDOWN)
4875 	    ? KVME_FLAG_GROWS_DOWN : KVME_FLAG_GROWS_UP;
4876 	kve->kve_flags |= (m->flags & VM_MAP_PAGEABLE)
4877 	    ? KVME_FLAG_PAGEABLE : 0;
4878 #endif
4879 	return 0;
4880 }
4881 
4882 static int
4883 fill_vmentries(struct lwp *l, pid_t pid, u_int elem_size, void *oldp,
4884     size_t *oldlenp)
4885 {
4886 	int error;
4887 	struct proc *p;
4888 	struct kinfo_vmentry *vme;
4889 	struct vmspace *vm;
4890 	struct vm_map *map;
4891 	struct vm_map_entry *entry;
4892 	char *dp;
4893 	size_t count, vmesize;
4894 
4895 	vme = NULL;
4896 	vmesize = *oldlenp;
4897 	count = 0;
4898 	if (oldp && *oldlenp > 1024 * 1024)
4899 		return E2BIG;
4900 
4901 	if ((error = proc_find_locked(l, &p, pid)) != 0)
4902 		return error;
4903 
4904 	if ((error = proc_vmspace_getref(p, &vm)) != 0)
4905 		goto out;
4906 
4907 	map = &vm->vm_map;
4908 	vm_map_lock_read(map);
4909 
4910 	dp = oldp;
4911 	if (oldp)
4912 		vme = kmem_alloc(vmesize, KM_SLEEP);
4913 	for (entry = map->header.next; entry != &map->header;
4914 	    entry = entry->next) {
4915 		if (oldp && (dp - (char *)oldp) < *oldlenp + elem_size) {
4916 			error = fill_vmentry(l, p, &vme[count], map, entry);
4917 			if (error)
4918 				goto out;
4919 			dp += elem_size;
4920 		}
4921 		count++;
4922 	}
4923 	vm_map_unlock_read(map);
4924 	uvmspace_free(vm);
4925 
4926 out:
4927 	if (pid != -1)
4928 		mutex_exit(p->p_lock);
4929 	if (error == 0) {
4930 		const u_int esize = min(sizeof(*vme), elem_size);
4931 		dp = oldp;
4932 		for (size_t i = 0; i < count; i++) {
4933 			if (oldp && (dp - (char *)oldp) < *oldlenp + elem_size)
4934 			{
4935 				error = sysctl_copyout(l, &vme[i], dp, esize);
4936 				if (error)
4937 					break;
4938 				dp += elem_size;
4939 			} else
4940 				break;
4941 		}
4942 		count *= elem_size;
4943 		if (oldp != NULL && *oldlenp < count)
4944 			error = ENOSPC;
4945 		*oldlenp = count;
4946 	}
4947 	if (vme)
4948 		kmem_free(vme, vmesize);
4949 	return error;
4950 }
4951 
4952 static int
4953 sysctl_vmproc(SYSCTLFN_ARGS)
4954 {
4955 	int error;
4956 
4957 	if (namelen == 1 && name[0] == CTL_QUERY)
4958 		return (sysctl_query(SYSCTLFN_CALL(rnode)));
4959 
4960 	if (namelen == 0)
4961 		return EINVAL;
4962 
4963 	switch (name[0]) {
4964 	case VM_PROC_MAP:
4965 		if (namelen != 3)
4966 			return EINVAL;
4967 		sysctl_unlock();
4968 		error = fill_vmentries(l, name[1], name[2],
4969 		    oldp, oldlenp);
4970 		sysctl_relock();
4971 		return error;
4972 	default:
4973 		return EINVAL;
4974 	}
4975 }
4976 
4977 SYSCTL_SETUP(sysctl_uvmmap_setup, "sysctl uvmmap setup")
4978 {
4979 
4980 	sysctl_createv(clog, 0, NULL, NULL,
4981 		       CTLFLAG_PERMANENT,
4982 		       CTLTYPE_STRUCT, "proc",
4983 		       SYSCTL_DESCR("Process vm information"),
4984 		       sysctl_vmproc, 0, NULL, 0,
4985 		       CTL_VM, VM_PROC, CTL_EOL);
4986 #ifndef __USER_VA0_IS_SAFE
4987         sysctl_createv(clog, 0, NULL, NULL,
4988                        CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
4989                        CTLTYPE_INT, "user_va0_disable",
4990                        SYSCTL_DESCR("Disable VA 0"),
4991                        sysctl_user_va0_disable, 0, &user_va0_disable, 0,
4992                        CTL_VM, CTL_CREATE, CTL_EOL);
4993 #endif
4994 }
4995