FreeBSD/Linux Kernel Cross Reference
sys/vm/vm_map.c
1 /*
2 * Copyright (c) 1991, 1993
3 * The Regents of the University of California. All rights reserved.
4 *
5 * This code is derived from software contributed to Berkeley by
6 * The Mach Operating System project at Carnegie-Mellon University.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 4. Neither the name of the University nor the names of its contributors
17 * may be used to endorse or promote products derived from this software
18 * without specific prior written permission.
19 *
20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30 * SUCH DAMAGE.
31 *
32 * from: @(#)vm_map.c 8.3 (Berkeley) 1/12/94
33 *
34 *
35 * Copyright (c) 1987, 1990 Carnegie-Mellon University.
36 * All rights reserved.
37 *
38 * Authors: Avadis Tevanian, Jr., Michael Wayne Young
39 *
40 * Permission to use, copy, modify and distribute this software and
41 * its documentation is hereby granted, provided that both the copyright
42 * notice and this permission notice appear in all copies of the
43 * software, derivative works or modified versions, and any portions
44 * thereof, and that both notices appear in supporting documentation.
45 *
46 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
47 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
48 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
49 *
50 * Carnegie Mellon requests users of this software to return to
51 *
52 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
53 * School of Computer Science
54 * Carnegie Mellon University
55 * Pittsburgh PA 15213-3890
56 *
57 * any improvements or extensions that they make and grant Carnegie the
58 * rights to redistribute these changes.
59 */
60
61 /*
62 * Virtual memory mapping module.
63 */
64
65 #include <sys/cdefs.h>
66 __FBSDID("$FreeBSD: releng/5.3/sys/vm/vm_map.c 133807 2004-08-16 06:16:12Z alc $");
67
68 #include <sys/param.h>
69 #include <sys/systm.h>
70 #include <sys/ktr.h>
71 #include <sys/lock.h>
72 #include <sys/mutex.h>
73 #include <sys/proc.h>
74 #include <sys/vmmeter.h>
75 #include <sys/mman.h>
76 #include <sys/vnode.h>
77 #include <sys/resourcevar.h>
78 #include <sys/file.h>
79 #include <sys/sysent.h>
80 #include <sys/shm.h>
81
82 #include <vm/vm.h>
83 #include <vm/vm_param.h>
84 #include <vm/pmap.h>
85 #include <vm/vm_map.h>
86 #include <vm/vm_page.h>
87 #include <vm/vm_object.h>
88 #include <vm/vm_pager.h>
89 #include <vm/vm_kern.h>
90 #include <vm/vm_extern.h>
91 #include <vm/swap_pager.h>
92 #include <vm/uma.h>
93
94 /*
95 * Virtual memory maps provide for the mapping, protection,
96 * and sharing of virtual memory objects. In addition,
97 * this module provides for an efficient virtual copy of
98 * memory from one map to another.
99 *
100 * Synchronization is required prior to most operations.
101 *
102 * Maps consist of an ordered doubly-linked list of simple
103 * entries; a single hint is used to speed up lookups.
104 *
105 * Since portions of maps are specified by start/end addresses,
106 * which may not align with existing map entries, all
107 * routines merely "clip" entries to these start/end values.
108 * [That is, an entry is split into two, bordering at a
109 * start or end value.] Note that these clippings may not
110 * always be necessary (as the two resulting entries are then
111 * not changed); however, the clipping is done for convenience.
112 *
113 * As mentioned above, virtual copy operations are performed
114 * by copying VM object references from one map to
115 * another, and then marking both regions as copy-on-write.
116 */
117
118 /*
119 * vm_map_startup:
120 *
121 * Initialize the vm_map module. Must be called before
122 * any other vm_map routines.
123 *
124 * Map and entry structures are allocated from the general
125 * purpose memory pool with some exceptions:
126 *
127 * - The kernel map and kmem submap are allocated statically.
128 * - Kernel map entries are allocated out of a static pool.
129 *
130 * These restrictions are necessary since malloc() uses the
131 * maps and requires map entries.
132 */
133
134 static struct mtx map_sleep_mtx;
135 static uma_zone_t mapentzone;
136 static uma_zone_t kmapentzone;
137 static uma_zone_t mapzone;
138 static uma_zone_t vmspace_zone;
139 static struct vm_object kmapentobj;
140 static int vmspace_zinit(void *mem, int size, int flags);
141 static void vmspace_zfini(void *mem, int size);
142 static int vm_map_zinit(void *mem, int ize, int flags);
143 static void vm_map_zfini(void *mem, int size);
144 static void _vm_map_init(vm_map_t map, vm_offset_t min, vm_offset_t max);
145
146 #ifdef INVARIANTS
147 static void vm_map_zdtor(void *mem, int size, void *arg);
148 static void vmspace_zdtor(void *mem, int size, void *arg);
149 #endif
150
151 void
152 vm_map_startup(void)
153 {
154 mtx_init(&map_sleep_mtx, "vm map sleep mutex", NULL, MTX_DEF);
155 mapzone = uma_zcreate("MAP", sizeof(struct vm_map), NULL,
156 #ifdef INVARIANTS
157 vm_map_zdtor,
158 #else
159 NULL,
160 #endif
161 vm_map_zinit, vm_map_zfini, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
162 uma_prealloc(mapzone, MAX_KMAP);
163 kmapentzone = uma_zcreate("KMAP ENTRY", sizeof(struct vm_map_entry),
164 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR,
165 UMA_ZONE_MTXCLASS | UMA_ZONE_VM);
166 uma_prealloc(kmapentzone, MAX_KMAPENT);
167 mapentzone = uma_zcreate("MAP ENTRY", sizeof(struct vm_map_entry),
168 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
169 uma_prealloc(mapentzone, MAX_MAPENT);
170 }
171
172 static void
173 vmspace_zfini(void *mem, int size)
174 {
175 struct vmspace *vm;
176
177 vm = (struct vmspace *)mem;
178 pmap_release(vmspace_pmap(vm));
179 vm_map_zfini(&vm->vm_map, sizeof(vm->vm_map));
180 }
181
182 static int
183 vmspace_zinit(void *mem, int size, int flags)
184 {
185 struct vmspace *vm;
186
187 vm = (struct vmspace *)mem;
188
189 (void)vm_map_zinit(&vm->vm_map, sizeof(vm->vm_map), flags);
190 pmap_pinit(vmspace_pmap(vm));
191 return (0);
192 }
193
194 static void
195 vm_map_zfini(void *mem, int size)
196 {
197 vm_map_t map;
198
199 map = (vm_map_t)mem;
200 mtx_destroy(&map->system_mtx);
201 sx_destroy(&map->lock);
202 }
203
204 static int
205 vm_map_zinit(void *mem, int size, int flags)
206 {
207 vm_map_t map;
208
209 map = (vm_map_t)mem;
210 map->nentries = 0;
211 map->size = 0;
212 mtx_init(&map->system_mtx, "system map", NULL, MTX_DEF | MTX_DUPOK);
213 sx_init(&map->lock, "user map");
214 return (0);
215 }
216
217 #ifdef INVARIANTS
218 static void
219 vmspace_zdtor(void *mem, int size, void *arg)
220 {
221 struct vmspace *vm;
222
223 vm = (struct vmspace *)mem;
224
225 vm_map_zdtor(&vm->vm_map, sizeof(vm->vm_map), arg);
226 }
227 static void
228 vm_map_zdtor(void *mem, int size, void *arg)
229 {
230 vm_map_t map;
231
232 map = (vm_map_t)mem;
233 KASSERT(map->nentries == 0,
234 ("map %p nentries == %d on free.",
235 map, map->nentries));
236 KASSERT(map->size == 0,
237 ("map %p size == %lu on free.",
238 map, (unsigned long)map->size));
239 }
240 #endif /* INVARIANTS */
241
242 /*
243 * Allocate a vmspace structure, including a vm_map and pmap,
244 * and initialize those structures. The refcnt is set to 1.
245 */
246 struct vmspace *
247 vmspace_alloc(min, max)
248 vm_offset_t min, max;
249 {
250 struct vmspace *vm;
251
252 vm = uma_zalloc(vmspace_zone, M_WAITOK);
253 CTR1(KTR_VM, "vmspace_alloc: %p", vm);
254 _vm_map_init(&vm->vm_map, min, max);
255 vm->vm_map.pmap = vmspace_pmap(vm); /* XXX */
256 vm->vm_refcnt = 1;
257 vm->vm_shm = NULL;
258 vm->vm_swrss = 0;
259 vm->vm_tsize = 0;
260 vm->vm_dsize = 0;
261 vm->vm_ssize = 0;
262 vm->vm_taddr = 0;
263 vm->vm_daddr = 0;
264 vm->vm_maxsaddr = 0;
265 vm->vm_exitingcnt = 0;
266 return (vm);
267 }
268
269 void
270 vm_init2(void)
271 {
272 uma_zone_set_obj(kmapentzone, &kmapentobj, lmin(cnt.v_page_count,
273 (VM_MAX_KERNEL_ADDRESS - KERNBASE) / PAGE_SIZE) / 8 +
274 maxproc * 2 + maxfiles);
275 vmspace_zone = uma_zcreate("VMSPACE", sizeof(struct vmspace), NULL,
276 #ifdef INVARIANTS
277 vmspace_zdtor,
278 #else
279 NULL,
280 #endif
281 vmspace_zinit, vmspace_zfini, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
282 pmap_init2();
283 }
284
285 static __inline void
286 vmspace_dofree(struct vmspace *vm)
287 {
288 CTR1(KTR_VM, "vmspace_free: %p", vm);
289
290 /*
291 * Make sure any SysV shm is freed, it might not have been in
292 * exit1().
293 */
294 shmexit(vm);
295
296 /*
297 * Lock the map, to wait out all other references to it.
298 * Delete all of the mappings and pages they hold, then call
299 * the pmap module to reclaim anything left.
300 */
301 vm_map_lock(&vm->vm_map);
302 (void) vm_map_delete(&vm->vm_map, vm->vm_map.min_offset,
303 vm->vm_map.max_offset);
304 vm_map_unlock(&vm->vm_map);
305
306 uma_zfree(vmspace_zone, vm);
307 }
308
309 void
310 vmspace_free(struct vmspace *vm)
311 {
312 int refcnt;
313
314 if (vm->vm_refcnt == 0)
315 panic("vmspace_free: attempt to free already freed vmspace");
316
317 do
318 refcnt = vm->vm_refcnt;
319 while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt - 1));
320 if (refcnt == 1 && vm->vm_exitingcnt == 0)
321 vmspace_dofree(vm);
322 }
323
324 void
325 vmspace_exitfree(struct proc *p)
326 {
327 struct vmspace *vm;
328 int exitingcnt;
329
330 vm = p->p_vmspace;
331 p->p_vmspace = NULL;
332
333 /*
334 * cleanup by parent process wait()ing on exiting child. vm_refcnt
335 * may not be 0 (e.g. fork() and child exits without exec()ing).
336 * exitingcnt may increment above 0 and drop back down to zero
337 * several times while vm_refcnt is held non-zero. vm_refcnt
338 * may also increment above 0 and drop back down to zero several
339 * times while vm_exitingcnt is held non-zero.
340 *
341 * The last wait on the exiting child's vmspace will clean up
342 * the remainder of the vmspace.
343 */
344 do
345 exitingcnt = vm->vm_exitingcnt;
346 while (!atomic_cmpset_int(&vm->vm_exitingcnt, exitingcnt,
347 exitingcnt - 1));
348 if (vm->vm_refcnt == 0 && exitingcnt == 1)
349 vmspace_dofree(vm);
350 }
351
352 void
353 _vm_map_lock(vm_map_t map, const char *file, int line)
354 {
355
356 if (map->system_map)
357 _mtx_lock_flags(&map->system_mtx, 0, file, line);
358 else
359 _sx_xlock(&map->lock, file, line);
360 map->timestamp++;
361 }
362
363 void
364 _vm_map_unlock(vm_map_t map, const char *file, int line)
365 {
366
367 if (map->system_map)
368 _mtx_unlock_flags(&map->system_mtx, 0, file, line);
369 else
370 _sx_xunlock(&map->lock, file, line);
371 }
372
373 void
374 _vm_map_lock_read(vm_map_t map, const char *file, int line)
375 {
376
377 if (map->system_map)
378 _mtx_lock_flags(&map->system_mtx, 0, file, line);
379 else
380 _sx_xlock(&map->lock, file, line);
381 }
382
383 void
384 _vm_map_unlock_read(vm_map_t map, const char *file, int line)
385 {
386
387 if (map->system_map)
388 _mtx_unlock_flags(&map->system_mtx, 0, file, line);
389 else
390 _sx_xunlock(&map->lock, file, line);
391 }
392
393 int
394 _vm_map_trylock(vm_map_t map, const char *file, int line)
395 {
396 int error;
397
398 error = map->system_map ?
399 !_mtx_trylock(&map->system_mtx, 0, file, line) :
400 !_sx_try_xlock(&map->lock, file, line);
401 if (error == 0)
402 map->timestamp++;
403 return (error == 0);
404 }
405
406 int
407 _vm_map_trylock_read(vm_map_t map, const char *file, int line)
408 {
409 int error;
410
411 error = map->system_map ?
412 !_mtx_trylock(&map->system_mtx, 0, file, line) :
413 !_sx_try_xlock(&map->lock, file, line);
414 return (error == 0);
415 }
416
417 int
418 _vm_map_lock_upgrade(vm_map_t map, const char *file, int line)
419 {
420
421 #ifdef INVARIANTS
422 if (map->system_map) {
423 _mtx_assert(&map->system_mtx, MA_OWNED, file, line);
424 } else
425 _sx_assert(&map->lock, SX_XLOCKED, file, line);
426 #endif
427 map->timestamp++;
428 return (0);
429 }
430
431 void
432 _vm_map_lock_downgrade(vm_map_t map, const char *file, int line)
433 {
434
435 #ifdef INVARIANTS
436 if (map->system_map) {
437 _mtx_assert(&map->system_mtx, MA_OWNED, file, line);
438 } else
439 _sx_assert(&map->lock, SX_XLOCKED, file, line);
440 #endif
441 }
442
443 /*
444 * vm_map_unlock_and_wait:
445 */
446 int
447 vm_map_unlock_and_wait(vm_map_t map, boolean_t user_wait)
448 {
449
450 mtx_lock(&map_sleep_mtx);
451 vm_map_unlock(map);
452 return (msleep(&map->root, &map_sleep_mtx, PDROP | PVM, "vmmaps", 0));
453 }
454
455 /*
456 * vm_map_wakeup:
457 */
458 void
459 vm_map_wakeup(vm_map_t map)
460 {
461
462 /*
463 * Acquire and release map_sleep_mtx to prevent a wakeup()
464 * from being performed (and lost) between the vm_map_unlock()
465 * and the msleep() in vm_map_unlock_and_wait().
466 */
467 mtx_lock(&map_sleep_mtx);
468 mtx_unlock(&map_sleep_mtx);
469 wakeup(&map->root);
470 }
471
472 long
473 vmspace_resident_count(struct vmspace *vmspace)
474 {
475 return pmap_resident_count(vmspace_pmap(vmspace));
476 }
477
478 long
479 vmspace_wired_count(struct vmspace *vmspace)
480 {
481 return pmap_wired_count(vmspace_pmap(vmspace));
482 }
483
484 /*
485 * vm_map_create:
486 *
487 * Creates and returns a new empty VM map with
488 * the given physical map structure, and having
489 * the given lower and upper address bounds.
490 */
491 vm_map_t
492 vm_map_create(pmap_t pmap, vm_offset_t min, vm_offset_t max)
493 {
494 vm_map_t result;
495
496 result = uma_zalloc(mapzone, M_WAITOK);
497 CTR1(KTR_VM, "vm_map_create: %p", result);
498 _vm_map_init(result, min, max);
499 result->pmap = pmap;
500 return (result);
501 }
502
503 /*
504 * Initialize an existing vm_map structure
505 * such as that in the vmspace structure.
506 * The pmap is set elsewhere.
507 */
508 static void
509 _vm_map_init(vm_map_t map, vm_offset_t min, vm_offset_t max)
510 {
511
512 map->header.next = map->header.prev = &map->header;
513 map->needs_wakeup = FALSE;
514 map->system_map = 0;
515 map->min_offset = min;
516 map->max_offset = max;
517 map->flags = 0;
518 map->root = NULL;
519 map->timestamp = 0;
520 }
521
522 void
523 vm_map_init(vm_map_t map, vm_offset_t min, vm_offset_t max)
524 {
525 _vm_map_init(map, min, max);
526 mtx_init(&map->system_mtx, "system map", NULL, MTX_DEF | MTX_DUPOK);
527 sx_init(&map->lock, "user map");
528 }
529
530 /*
531 * vm_map_entry_dispose: [ internal use only ]
532 *
533 * Inverse of vm_map_entry_create.
534 */
535 static void
536 vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry)
537 {
538 uma_zfree(map->system_map ? kmapentzone : mapentzone, entry);
539 }
540
541 /*
542 * vm_map_entry_create: [ internal use only ]
543 *
544 * Allocates a VM map entry for insertion.
545 * No entry fields are filled in.
546 */
547 static vm_map_entry_t
548 vm_map_entry_create(vm_map_t map)
549 {
550 vm_map_entry_t new_entry;
551
552 if (map->system_map)
553 new_entry = uma_zalloc(kmapentzone, M_NOWAIT);
554 else
555 new_entry = uma_zalloc(mapentzone, M_WAITOK);
556 if (new_entry == NULL)
557 panic("vm_map_entry_create: kernel resources exhausted");
558 return (new_entry);
559 }
560
561 /*
562 * vm_map_entry_set_behavior:
563 *
564 * Set the expected access behavior, either normal, random, or
565 * sequential.
566 */
567 static __inline void
568 vm_map_entry_set_behavior(vm_map_entry_t entry, u_char behavior)
569 {
570 entry->eflags = (entry->eflags & ~MAP_ENTRY_BEHAV_MASK) |
571 (behavior & MAP_ENTRY_BEHAV_MASK);
572 }
573
574 /*
575 * vm_map_entry_set_max_free:
576 *
577 * Set the max_free field in a vm_map_entry.
578 */
579 static __inline void
580 vm_map_entry_set_max_free(vm_map_entry_t entry)
581 {
582
583 entry->max_free = entry->adj_free;
584 if (entry->left != NULL && entry->left->max_free > entry->max_free)
585 entry->max_free = entry->left->max_free;
586 if (entry->right != NULL && entry->right->max_free > entry->max_free)
587 entry->max_free = entry->right->max_free;
588 }
589
590 /*
591 * vm_map_entry_splay:
592 *
593 * The Sleator and Tarjan top-down splay algorithm with the
594 * following variation. Max_free must be computed bottom-up, so
595 * on the downward pass, maintain the left and right spines in
596 * reverse order. Then, make a second pass up each side to fix
597 * the pointers and compute max_free. The time bound is O(log n)
598 * amortized.
599 *
600 * The new root is the vm_map_entry containing "addr", or else an
601 * adjacent entry (lower or higher) if addr is not in the tree.
602 *
603 * The map must be locked, and leaves it so.
604 *
605 * Returns: the new root.
606 */
607 static vm_map_entry_t
608 vm_map_entry_splay(vm_offset_t addr, vm_map_entry_t root)
609 {
610 vm_map_entry_t llist, rlist;
611 vm_map_entry_t ltree, rtree;
612 vm_map_entry_t y;
613
614 /* Special case of empty tree. */
615 if (root == NULL)
616 return (root);
617
618 /*
619 * Pass One: Splay down the tree until we find addr or a NULL
620 * pointer where addr would go. llist and rlist are the two
621 * sides in reverse order (bottom-up), with llist linked by
622 * the right pointer and rlist linked by the left pointer in
623 * the vm_map_entry. Wait until Pass Two to set max_free on
624 * the two spines.
625 */
626 llist = NULL;
627 rlist = NULL;
628 for (;;) {
629 /* root is never NULL in here. */
630 if (addr < root->start) {
631 y = root->left;
632 if (y == NULL)
633 break;
634 if (addr < y->start && y->left != NULL) {
635 /* Rotate right and put y on rlist. */
636 root->left = y->right;
637 y->right = root;
638 vm_map_entry_set_max_free(root);
639 root = y->left;
640 y->left = rlist;
641 rlist = y;
642 } else {
643 /* Put root on rlist. */
644 root->left = rlist;
645 rlist = root;
646 root = y;
647 }
648 } else {
649 y = root->right;
650 if (addr < root->end || y == NULL)
651 break;
652 if (addr >= y->end && y->right != NULL) {
653 /* Rotate left and put y on llist. */
654 root->right = y->left;
655 y->left = root;
656 vm_map_entry_set_max_free(root);
657 root = y->right;
658 y->right = llist;
659 llist = y;
660 } else {
661 /* Put root on llist. */
662 root->right = llist;
663 llist = root;
664 root = y;
665 }
666 }
667 }
668
669 /*
670 * Pass Two: Walk back up the two spines, flip the pointers
671 * and set max_free. The subtrees of the root go at the
672 * bottom of llist and rlist.
673 */
674 ltree = root->left;
675 while (llist != NULL) {
676 y = llist->right;
677 llist->right = ltree;
678 vm_map_entry_set_max_free(llist);
679 ltree = llist;
680 llist = y;
681 }
682 rtree = root->right;
683 while (rlist != NULL) {
684 y = rlist->left;
685 rlist->left = rtree;
686 vm_map_entry_set_max_free(rlist);
687 rtree = rlist;
688 rlist = y;
689 }
690
691 /*
692 * Final assembly: add ltree and rtree as subtrees of root.
693 */
694 root->left = ltree;
695 root->right = rtree;
696 vm_map_entry_set_max_free(root);
697
698 return (root);
699 }
700
701 /*
702 * vm_map_entry_{un,}link:
703 *
704 * Insert/remove entries from maps.
705 */
706 static void
707 vm_map_entry_link(vm_map_t map,
708 vm_map_entry_t after_where,
709 vm_map_entry_t entry)
710 {
711
712 CTR4(KTR_VM,
713 "vm_map_entry_link: map %p, nentries %d, entry %p, after %p", map,
714 map->nentries, entry, after_where);
715 map->nentries++;
716 entry->prev = after_where;
717 entry->next = after_where->next;
718 entry->next->prev = entry;
719 after_where->next = entry;
720
721 if (after_where != &map->header) {
722 if (after_where != map->root)
723 vm_map_entry_splay(after_where->start, map->root);
724 entry->right = after_where->right;
725 entry->left = after_where;
726 after_where->right = NULL;
727 after_where->adj_free = entry->start - after_where->end;
728 vm_map_entry_set_max_free(after_where);
729 } else {
730 entry->right = map->root;
731 entry->left = NULL;
732 }
733 entry->adj_free = (entry->next == &map->header ? map->max_offset :
734 entry->next->start) - entry->end;
735 vm_map_entry_set_max_free(entry);
736 map->root = entry;
737 }
738
739 static void
740 vm_map_entry_unlink(vm_map_t map,
741 vm_map_entry_t entry)
742 {
743 vm_map_entry_t next, prev, root;
744
745 if (entry != map->root)
746 vm_map_entry_splay(entry->start, map->root);
747 if (entry->left == NULL)
748 root = entry->right;
749 else {
750 root = vm_map_entry_splay(entry->start, entry->left);
751 root->right = entry->right;
752 root->adj_free = (entry->next == &map->header ? map->max_offset :
753 entry->next->start) - root->end;
754 vm_map_entry_set_max_free(root);
755 }
756 map->root = root;
757
758 prev = entry->prev;
759 next = entry->next;
760 next->prev = prev;
761 prev->next = next;
762 map->nentries--;
763 CTR3(KTR_VM, "vm_map_entry_unlink: map %p, nentries %d, entry %p", map,
764 map->nentries, entry);
765 }
766
767 /*
768 * vm_map_entry_resize_free:
769 *
770 * Recompute the amount of free space following a vm_map_entry
771 * and propagate that value up the tree. Call this function after
772 * resizing a map entry in-place, that is, without a call to
773 * vm_map_entry_link() or _unlink().
774 *
775 * The map must be locked, and leaves it so.
776 */
777 static void
778 vm_map_entry_resize_free(vm_map_t map, vm_map_entry_t entry)
779 {
780
781 /*
782 * Using splay trees without parent pointers, propagating
783 * max_free up the tree is done by moving the entry to the
784 * root and making the change there.
785 */
786 if (entry != map->root)
787 map->root = vm_map_entry_splay(entry->start, map->root);
788
789 entry->adj_free = (entry->next == &map->header ? map->max_offset :
790 entry->next->start) - entry->end;
791 vm_map_entry_set_max_free(entry);
792 }
793
794 /*
795 * vm_map_lookup_entry: [ internal use only ]
796 *
797 * Finds the map entry containing (or
798 * immediately preceding) the specified address
799 * in the given map; the entry is returned
800 * in the "entry" parameter. The boolean
801 * result indicates whether the address is
802 * actually contained in the map.
803 */
804 boolean_t
805 vm_map_lookup_entry(
806 vm_map_t map,
807 vm_offset_t address,
808 vm_map_entry_t *entry) /* OUT */
809 {
810 vm_map_entry_t cur;
811
812 cur = vm_map_entry_splay(address, map->root);
813 if (cur == NULL)
814 *entry = &map->header;
815 else {
816 map->root = cur;
817
818 if (address >= cur->start) {
819 *entry = cur;
820 if (cur->end > address)
821 return (TRUE);
822 } else
823 *entry = cur->prev;
824 }
825 return (FALSE);
826 }
827
828 /*
829 * vm_map_insert:
830 *
831 * Inserts the given whole VM object into the target
832 * map at the specified address range. The object's
833 * size should match that of the address range.
834 *
835 * Requires that the map be locked, and leaves it so.
836 *
837 * If object is non-NULL, ref count must be bumped by caller
838 * prior to making call to account for the new entry.
839 */
840 int
841 vm_map_insert(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
842 vm_offset_t start, vm_offset_t end, vm_prot_t prot, vm_prot_t max,
843 int cow)
844 {
845 vm_map_entry_t new_entry;
846 vm_map_entry_t prev_entry;
847 vm_map_entry_t temp_entry;
848 vm_eflags_t protoeflags;
849
850 /*
851 * Check that the start and end points are not bogus.
852 */
853 if ((start < map->min_offset) || (end > map->max_offset) ||
854 (start >= end))
855 return (KERN_INVALID_ADDRESS);
856
857 /*
858 * Find the entry prior to the proposed starting address; if it's part
859 * of an existing entry, this range is bogus.
860 */
861 if (vm_map_lookup_entry(map, start, &temp_entry))
862 return (KERN_NO_SPACE);
863
864 prev_entry = temp_entry;
865
866 /*
867 * Assert that the next entry doesn't overlap the end point.
868 */
869 if ((prev_entry->next != &map->header) &&
870 (prev_entry->next->start < end))
871 return (KERN_NO_SPACE);
872
873 protoeflags = 0;
874
875 if (cow & MAP_COPY_ON_WRITE)
876 protoeflags |= MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY;
877
878 if (cow & MAP_NOFAULT) {
879 protoeflags |= MAP_ENTRY_NOFAULT;
880
881 KASSERT(object == NULL,
882 ("vm_map_insert: paradoxical MAP_NOFAULT request"));
883 }
884 if (cow & MAP_DISABLE_SYNCER)
885 protoeflags |= MAP_ENTRY_NOSYNC;
886 if (cow & MAP_DISABLE_COREDUMP)
887 protoeflags |= MAP_ENTRY_NOCOREDUMP;
888
889 if (object != NULL) {
890 /*
891 * OBJ_ONEMAPPING must be cleared unless this mapping
892 * is trivially proven to be the only mapping for any
893 * of the object's pages. (Object granularity
894 * reference counting is insufficient to recognize
895 * aliases with precision.)
896 */
897 VM_OBJECT_LOCK(object);
898 if (object->ref_count > 1 || object->shadow_count != 0)
899 vm_object_clear_flag(object, OBJ_ONEMAPPING);
900 VM_OBJECT_UNLOCK(object);
901 }
902 else if ((prev_entry != &map->header) &&
903 (prev_entry->eflags == protoeflags) &&
904 (prev_entry->end == start) &&
905 (prev_entry->wired_count == 0) &&
906 ((prev_entry->object.vm_object == NULL) ||
907 vm_object_coalesce(prev_entry->object.vm_object,
908 prev_entry->offset,
909 (vm_size_t)(prev_entry->end - prev_entry->start),
910 (vm_size_t)(end - prev_entry->end)))) {
911 /*
912 * We were able to extend the object. Determine if we
913 * can extend the previous map entry to include the
914 * new range as well.
915 */
916 if ((prev_entry->inheritance == VM_INHERIT_DEFAULT) &&
917 (prev_entry->protection == prot) &&
918 (prev_entry->max_protection == max)) {
919 map->size += (end - prev_entry->end);
920 prev_entry->end = end;
921 vm_map_entry_resize_free(map, prev_entry);
922 vm_map_simplify_entry(map, prev_entry);
923 return (KERN_SUCCESS);
924 }
925
926 /*
927 * If we can extend the object but cannot extend the
928 * map entry, we have to create a new map entry. We
929 * must bump the ref count on the extended object to
930 * account for it. object may be NULL.
931 */
932 object = prev_entry->object.vm_object;
933 offset = prev_entry->offset +
934 (prev_entry->end - prev_entry->start);
935 vm_object_reference(object);
936 }
937
938 /*
939 * NOTE: if conditionals fail, object can be NULL here. This occurs
940 * in things like the buffer map where we manage kva but do not manage
941 * backing objects.
942 */
943
944 /*
945 * Create a new entry
946 */
947 new_entry = vm_map_entry_create(map);
948 new_entry->start = start;
949 new_entry->end = end;
950
951 new_entry->eflags = protoeflags;
952 new_entry->object.vm_object = object;
953 new_entry->offset = offset;
954 new_entry->avail_ssize = 0;
955
956 new_entry->inheritance = VM_INHERIT_DEFAULT;
957 new_entry->protection = prot;
958 new_entry->max_protection = max;
959 new_entry->wired_count = 0;
960
961 /*
962 * Insert the new entry into the list
963 */
964 vm_map_entry_link(map, prev_entry, new_entry);
965 map->size += new_entry->end - new_entry->start;
966
967 #if 0
968 /*
969 * Temporarily removed to avoid MAP_STACK panic, due to
970 * MAP_STACK being a huge hack. Will be added back in
971 * when MAP_STACK (and the user stack mapping) is fixed.
972 */
973 /*
974 * It may be possible to simplify the entry
975 */
976 vm_map_simplify_entry(map, new_entry);
977 #endif
978
979 if (cow & (MAP_PREFAULT|MAP_PREFAULT_PARTIAL)) {
980 vm_map_pmap_enter(map, start, prot,
981 object, OFF_TO_IDX(offset), end - start,
982 cow & MAP_PREFAULT_PARTIAL);
983 }
984
985 return (KERN_SUCCESS);
986 }
987
988 /*
989 * vm_map_findspace:
990 *
991 * Find the first fit (lowest VM address) for "length" free bytes
992 * beginning at address >= start in the given map.
993 *
994 * In a vm_map_entry, "adj_free" is the amount of free space
995 * adjacent (higher address) to this entry, and "max_free" is the
996 * maximum amount of contiguous free space in its subtree. This
997 * allows finding a free region in one path down the tree, so
998 * O(log n) amortized with splay trees.
999 *
1000 * The map must be locked, and leaves it so.
1001 *
1002 * Returns: 0 on success, and starting address in *addr,
1003 * 1 if insufficient space.
1004 */
1005 int
1006 vm_map_findspace(vm_map_t map, vm_offset_t start, vm_size_t length,
1007 vm_offset_t *addr) /* OUT */
1008 {
1009 vm_map_entry_t entry;
1010 vm_offset_t end, st;
1011
1012 /* Request must fit within min/max VM address. */
1013 if (start < map->min_offset)
1014 start = map->min_offset;
1015 if (start + length > map->max_offset)
1016 return (1);
1017
1018 /* Empty tree means wide open address space. */
1019 if (map->root == NULL) {
1020 *addr = start;
1021 goto found;
1022 }
1023
1024 /*
1025 * After splay, if start comes before root node, then there
1026 * must be a gap from start to the root.
1027 */
1028 map->root = vm_map_entry_splay(start, map->root);
1029 if (start + length <= map->root->start) {
1030 *addr = start;
1031 goto found;
1032 }
1033
1034 /*
1035 * Root is the last node that might begin its gap before
1036 * start.
1037 */
1038 st = (start > map->root->end) ? start : map->root->end;
1039 if (st + length <= map->root->end + map->root->adj_free) {
1040 *addr = st;
1041 goto found;
1042 }
1043
1044 /* With max_free, can immediately tell if no solution. */
1045 entry = map->root->right;
1046 if (entry == NULL || length > entry->max_free)
1047 return (1);
1048
1049 /*
1050 * Search the right subtree in the order: left subtree, root,
1051 * right subtree (first fit). The previous splay implies that
1052 * all regions in the right subtree have addresses > start.
1053 */
1054 while (entry != NULL) {
1055 if (entry->left != NULL && entry->left->max_free >= length)
1056 entry = entry->left;
1057 else if (entry->adj_free >= length) {
1058 *addr = entry->end;
1059 goto found;
1060 } else
1061 entry = entry->right;
1062 }
1063
1064 /* Can't get here, so panic if we do. */
1065 panic("vm_map_findspace: max_free corrupt");
1066
1067 found:
1068 /* Expand the kernel pmap, if necessary. */
1069 if (map == kernel_map) {
1070 end = round_page(*addr + length);
1071 if (end > kernel_vm_end)
1072 pmap_growkernel(end);
1073 }
1074 return (0);
1075 }
1076
1077 /*
1078 * vm_map_find finds an unallocated region in the target address
1079 * map with the given length. The search is defined to be
1080 * first-fit from the specified address; the region found is
1081 * returned in the same parameter.
1082 *
1083 * If object is non-NULL, ref count must be bumped by caller
1084 * prior to making call to account for the new entry.
1085 */
1086 int
1087 vm_map_find(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1088 vm_offset_t *addr, /* IN/OUT */
1089 vm_size_t length, boolean_t find_space, vm_prot_t prot,
1090 vm_prot_t max, int cow)
1091 {
1092 vm_offset_t start;
1093 int result;
1094
1095 start = *addr;
1096 vm_map_lock(map);
1097 if (find_space) {
1098 if (vm_map_findspace(map, start, length, addr)) {
1099 vm_map_unlock(map);
1100 return (KERN_NO_SPACE);
1101 }
1102 start = *addr;
1103 }
1104 result = vm_map_insert(map, object, offset,
1105 start, start + length, prot, max, cow);
1106 vm_map_unlock(map);
1107 return (result);
1108 }
1109
1110 /*
1111 * vm_map_simplify_entry:
1112 *
1113 * Simplify the given map entry by merging with either neighbor. This
1114 * routine also has the ability to merge with both neighbors.
1115 *
1116 * The map must be locked.
1117 *
1118 * This routine guarentees that the passed entry remains valid (though
1119 * possibly extended). When merging, this routine may delete one or
1120 * both neighbors.
1121 */
1122 void
1123 vm_map_simplify_entry(vm_map_t map, vm_map_entry_t entry)
1124 {
1125 vm_map_entry_t next, prev;
1126 vm_size_t prevsize, esize;
1127
1128 if (entry->eflags & (MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_IS_SUB_MAP))
1129 return;
1130
1131 prev = entry->prev;
1132 if (prev != &map->header) {
1133 prevsize = prev->end - prev->start;
1134 if ( (prev->end == entry->start) &&
1135 (prev->object.vm_object == entry->object.vm_object) &&
1136 (!prev->object.vm_object ||
1137 (prev->offset + prevsize == entry->offset)) &&
1138 (prev->eflags == entry->eflags) &&
1139 (prev->protection == entry->protection) &&
1140 (prev->max_protection == entry->max_protection) &&
1141 (prev->inheritance == entry->inheritance) &&
1142 (prev->wired_count == entry->wired_count)) {
1143 vm_map_entry_unlink(map, prev);
1144 entry->start = prev->start;
1145 entry->offset = prev->offset;
1146 if (entry->prev != &map->header)
1147 vm_map_entry_resize_free(map, entry->prev);
1148 if (prev->object.vm_object)
1149 vm_object_deallocate(prev->object.vm_object);
1150 vm_map_entry_dispose(map, prev);
1151 }
1152 }
1153
1154 next = entry->next;
1155 if (next != &map->header) {
1156 esize = entry->end - entry->start;
1157 if ((entry->end == next->start) &&
1158 (next->object.vm_object == entry->object.vm_object) &&
1159 (!entry->object.vm_object ||
1160 (entry->offset + esize == next->offset)) &&
1161 (next->eflags == entry->eflags) &&
1162 (next->protection == entry->protection) &&
1163 (next->max_protection == entry->max_protection) &&
1164 (next->inheritance == entry->inheritance) &&
1165 (next->wired_count == entry->wired_count)) {
1166 vm_map_entry_unlink(map, next);
1167 entry->end = next->end;
1168 vm_map_entry_resize_free(map, entry);
1169 if (next->object.vm_object)
1170 vm_object_deallocate(next->object.vm_object);
1171 vm_map_entry_dispose(map, next);
1172 }
1173 }
1174 }
1175 /*
1176 * vm_map_clip_start: [ internal use only ]
1177 *
1178 * Asserts that the given entry begins at or after
1179 * the specified address; if necessary,
1180 * it splits the entry into two.
1181 */
1182 #define vm_map_clip_start(map, entry, startaddr) \
1183 { \
1184 if (startaddr > entry->start) \
1185 _vm_map_clip_start(map, entry, startaddr); \
1186 }
1187
1188 /*
1189 * This routine is called only when it is known that
1190 * the entry must be split.
1191 */
1192 static void
1193 _vm_map_clip_start(vm_map_t map, vm_map_entry_t entry, vm_offset_t start)
1194 {
1195 vm_map_entry_t new_entry;
1196
1197 /*
1198 * Split off the front portion -- note that we must insert the new
1199 * entry BEFORE this one, so that this entry has the specified
1200 * starting address.
1201 */
1202 vm_map_simplify_entry(map, entry);
1203
1204 /*
1205 * If there is no object backing this entry, we might as well create
1206 * one now. If we defer it, an object can get created after the map
1207 * is clipped, and individual objects will be created for the split-up
1208 * map. This is a bit of a hack, but is also about the best place to
1209 * put this improvement.
1210 */
1211 if (entry->object.vm_object == NULL && !map->system_map) {
1212 vm_object_t object;
1213 object = vm_object_allocate(OBJT_DEFAULT,
1214 atop(entry->end - entry->start));
1215 entry->object.vm_object = object;
1216 entry->offset = 0;
1217 }
1218
1219 new_entry = vm_map_entry_create(map);
1220 *new_entry = *entry;
1221
1222 new_entry->end = start;
1223 entry->offset += (start - entry->start);
1224 entry->start = start;
1225
1226 vm_map_entry_link(map, entry->prev, new_entry);
1227
1228 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1229 vm_object_reference(new_entry->object.vm_object);
1230 }
1231 }
1232
1233 /*
1234 * vm_map_clip_end: [ internal use only ]
1235 *
1236 * Asserts that the given entry ends at or before
1237 * the specified address; if necessary,
1238 * it splits the entry into two.
1239 */
1240 #define vm_map_clip_end(map, entry, endaddr) \
1241 { \
1242 if ((endaddr) < (entry->end)) \
1243 _vm_map_clip_end((map), (entry), (endaddr)); \
1244 }
1245
1246 /*
1247 * This routine is called only when it is known that
1248 * the entry must be split.
1249 */
1250 static void
1251 _vm_map_clip_end(vm_map_t map, vm_map_entry_t entry, vm_offset_t end)
1252 {
1253 vm_map_entry_t new_entry;
1254
1255 /*
1256 * If there is no object backing this entry, we might as well create
1257 * one now. If we defer it, an object can get created after the map
1258 * is clipped, and individual objects will be created for the split-up
1259 * map. This is a bit of a hack, but is also about the best place to
1260 * put this improvement.
1261 */
1262 if (entry->object.vm_object == NULL && !map->system_map) {
1263 vm_object_t object;
1264 object = vm_object_allocate(OBJT_DEFAULT,
1265 atop(entry->end - entry->start));
1266 entry->object.vm_object = object;
1267 entry->offset = 0;
1268 }
1269
1270 /*
1271 * Create a new entry and insert it AFTER the specified entry
1272 */
1273 new_entry = vm_map_entry_create(map);
1274 *new_entry = *entry;
1275
1276 new_entry->start = entry->end = end;
1277 new_entry->offset += (end - entry->start);
1278
1279 vm_map_entry_link(map, entry, new_entry);
1280
1281 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1282 vm_object_reference(new_entry->object.vm_object);
1283 }
1284 }
1285
1286 /*
1287 * VM_MAP_RANGE_CHECK: [ internal use only ]
1288 *
1289 * Asserts that the starting and ending region
1290 * addresses fall within the valid range of the map.
1291 */
1292 #define VM_MAP_RANGE_CHECK(map, start, end) \
1293 { \
1294 if (start < vm_map_min(map)) \
1295 start = vm_map_min(map); \
1296 if (end > vm_map_max(map)) \
1297 end = vm_map_max(map); \
1298 if (start > end) \
1299 start = end; \
1300 }
1301
1302 /*
1303 * vm_map_submap: [ kernel use only ]
1304 *
1305 * Mark the given range as handled by a subordinate map.
1306 *
1307 * This range must have been created with vm_map_find,
1308 * and no other operations may have been performed on this
1309 * range prior to calling vm_map_submap.
1310 *
1311 * Only a limited number of operations can be performed
1312 * within this rage after calling vm_map_submap:
1313 * vm_fault
1314 * [Don't try vm_map_copy!]
1315 *
1316 * To remove a submapping, one must first remove the
1317 * range from the superior map, and then destroy the
1318 * submap (if desired). [Better yet, don't try it.]
1319 */
1320 int
1321 vm_map_submap(
1322 vm_map_t map,
1323 vm_offset_t start,
1324 vm_offset_t end,
1325 vm_map_t submap)
1326 {
1327 vm_map_entry_t entry;
1328 int result = KERN_INVALID_ARGUMENT;
1329
1330 vm_map_lock(map);
1331
1332 VM_MAP_RANGE_CHECK(map, start, end);
1333
1334 if (vm_map_lookup_entry(map, start, &entry)) {
1335 vm_map_clip_start(map, entry, start);
1336 } else
1337 entry = entry->next;
1338
1339 vm_map_clip_end(map, entry, end);
1340
1341 if ((entry->start == start) && (entry->end == end) &&
1342 ((entry->eflags & MAP_ENTRY_COW) == 0) &&
1343 (entry->object.vm_object == NULL)) {
1344 entry->object.sub_map = submap;
1345 entry->eflags |= MAP_ENTRY_IS_SUB_MAP;
1346 result = KERN_SUCCESS;
1347 }
1348 vm_map_unlock(map);
1349
1350 return (result);
1351 }
1352
1353 /*
1354 * The maximum number of pages to map
1355 */
1356 #define MAX_INIT_PT 96
1357
1358 /*
1359 * vm_map_pmap_enter:
1360 *
1361 * Preload read-only mappings for the given object into the specified
1362 * map. This eliminates the soft faults on process startup and
1363 * immediately after an mmap(2).
1364 */
1365 void
1366 vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot,
1367 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags)
1368 {
1369 vm_offset_t tmpidx;
1370 int psize;
1371 vm_page_t p, mpte;
1372
1373 if ((prot & VM_PROT_READ) == 0 || object == NULL)
1374 return;
1375 VM_OBJECT_LOCK(object);
1376 if (object->type == OBJT_DEVICE) {
1377 pmap_object_init_pt(map->pmap, addr, object, pindex, size);
1378 goto unlock_return;
1379 }
1380
1381 psize = atop(size);
1382
1383 if (object->type != OBJT_VNODE ||
1384 ((flags & MAP_PREFAULT_PARTIAL) && (psize > MAX_INIT_PT) &&
1385 (object->resident_page_count > MAX_INIT_PT))) {
1386 goto unlock_return;
1387 }
1388
1389 if (psize + pindex > object->size) {
1390 if (object->size < pindex)
1391 goto unlock_return;
1392 psize = object->size - pindex;
1393 }
1394
1395 mpte = NULL;
1396
1397 if ((p = TAILQ_FIRST(&object->memq)) != NULL) {
1398 if (p->pindex < pindex) {
1399 p = vm_page_splay(pindex, object->root);
1400 if ((object->root = p)->pindex < pindex)
1401 p = TAILQ_NEXT(p, listq);
1402 }
1403 }
1404 /*
1405 * Assert: the variable p is either (1) the page with the
1406 * least pindex greater than or equal to the parameter pindex
1407 * or (2) NULL.
1408 */
1409 for (;
1410 p != NULL && (tmpidx = p->pindex - pindex) < psize;
1411 p = TAILQ_NEXT(p, listq)) {
1412 /*
1413 * don't allow an madvise to blow away our really
1414 * free pages allocating pv entries.
1415 */
1416 if ((flags & MAP_PREFAULT_MADVISE) &&
1417 cnt.v_free_count < cnt.v_free_reserved) {
1418 break;
1419 }
1420 vm_page_lock_queues();
1421 if ((p->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL &&
1422 (p->busy == 0) &&
1423 (p->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) {
1424 if ((p->queue - p->pc) == PQ_CACHE)
1425 vm_page_deactivate(p);
1426 vm_page_busy(p);
1427 vm_page_unlock_queues();
1428 VM_OBJECT_UNLOCK(object);
1429 mpte = pmap_enter_quick(map->pmap,
1430 addr + ptoa(tmpidx), p, mpte);
1431 VM_OBJECT_LOCK(object);
1432 vm_page_lock_queues();
1433 vm_page_wakeup(p);
1434 }
1435 vm_page_unlock_queues();
1436 }
1437 unlock_return:
1438 VM_OBJECT_UNLOCK(object);
1439 }
1440
1441 /*
1442 * vm_map_protect:
1443 *
1444 * Sets the protection of the specified address
1445 * region in the target map. If "set_max" is
1446 * specified, the maximum protection is to be set;
1447 * otherwise, only the current protection is affected.
1448 */
1449 int
1450 vm_map_protect(vm_map_t map, vm_offset_t start, vm_offset_t end,
1451 vm_prot_t new_prot, boolean_t set_max)
1452 {
1453 vm_map_entry_t current;
1454 vm_map_entry_t entry;
1455
1456 vm_map_lock(map);
1457
1458 VM_MAP_RANGE_CHECK(map, start, end);
1459
1460 if (vm_map_lookup_entry(map, start, &entry)) {
1461 vm_map_clip_start(map, entry, start);
1462 } else {
1463 entry = entry->next;
1464 }
1465
1466 /*
1467 * Make a first pass to check for protection violations.
1468 */
1469 current = entry;
1470 while ((current != &map->header) && (current->start < end)) {
1471 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
1472 vm_map_unlock(map);
1473 return (KERN_INVALID_ARGUMENT);
1474 }
1475 if ((new_prot & current->max_protection) != new_prot) {
1476 vm_map_unlock(map);
1477 return (KERN_PROTECTION_FAILURE);
1478 }
1479 current = current->next;
1480 }
1481
1482 /*
1483 * Go back and fix up protections. [Note that clipping is not
1484 * necessary the second time.]
1485 */
1486 current = entry;
1487 while ((current != &map->header) && (current->start < end)) {
1488 vm_prot_t old_prot;
1489
1490 vm_map_clip_end(map, current, end);
1491
1492 old_prot = current->protection;
1493 if (set_max)
1494 current->protection =
1495 (current->max_protection = new_prot) &
1496 old_prot;
1497 else
1498 current->protection = new_prot;
1499
1500 /*
1501 * Update physical map if necessary. Worry about copy-on-write
1502 * here -- CHECK THIS XXX
1503 */
1504 if (current->protection != old_prot) {
1505 #define MASK(entry) (((entry)->eflags & MAP_ENTRY_COW) ? ~VM_PROT_WRITE : \
1506 VM_PROT_ALL)
1507 pmap_protect(map->pmap, current->start,
1508 current->end,
1509 current->protection & MASK(current));
1510 #undef MASK
1511 }
1512 vm_map_simplify_entry(map, current);
1513 current = current->next;
1514 }
1515 vm_map_unlock(map);
1516 return (KERN_SUCCESS);
1517 }
1518
1519 /*
1520 * vm_map_madvise:
1521 *
1522 * This routine traverses a processes map handling the madvise
1523 * system call. Advisories are classified as either those effecting
1524 * the vm_map_entry structure, or those effecting the underlying
1525 * objects.
1526 */
1527 int
1528 vm_map_madvise(
1529 vm_map_t map,
1530 vm_offset_t start,
1531 vm_offset_t end,
1532 int behav)
1533 {
1534 vm_map_entry_t current, entry;
1535 int modify_map = 0;
1536
1537 /*
1538 * Some madvise calls directly modify the vm_map_entry, in which case
1539 * we need to use an exclusive lock on the map and we need to perform
1540 * various clipping operations. Otherwise we only need a read-lock
1541 * on the map.
1542 */
1543 switch(behav) {
1544 case MADV_NORMAL:
1545 case MADV_SEQUENTIAL:
1546 case MADV_RANDOM:
1547 case MADV_NOSYNC:
1548 case MADV_AUTOSYNC:
1549 case MADV_NOCORE:
1550 case MADV_CORE:
1551 modify_map = 1;
1552 vm_map_lock(map);
1553 break;
1554 case MADV_WILLNEED:
1555 case MADV_DONTNEED:
1556 case MADV_FREE:
1557 vm_map_lock_read(map);
1558 break;
1559 default:
1560 return (KERN_INVALID_ARGUMENT);
1561 }
1562
1563 /*
1564 * Locate starting entry and clip if necessary.
1565 */
1566 VM_MAP_RANGE_CHECK(map, start, end);
1567
1568 if (vm_map_lookup_entry(map, start, &entry)) {
1569 if (modify_map)
1570 vm_map_clip_start(map, entry, start);
1571 } else {
1572 entry = entry->next;
1573 }
1574
1575 if (modify_map) {
1576 /*
1577 * madvise behaviors that are implemented in the vm_map_entry.
1578 *
1579 * We clip the vm_map_entry so that behavioral changes are
1580 * limited to the specified address range.
1581 */
1582 for (current = entry;
1583 (current != &map->header) && (current->start < end);
1584 current = current->next
1585 ) {
1586 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
1587 continue;
1588
1589 vm_map_clip_end(map, current, end);
1590
1591 switch (behav) {
1592 case MADV_NORMAL:
1593 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_NORMAL);
1594 break;
1595 case MADV_SEQUENTIAL:
1596 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_SEQUENTIAL);
1597 break;
1598 case MADV_RANDOM:
1599 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_RANDOM);
1600 break;
1601 case MADV_NOSYNC:
1602 current->eflags |= MAP_ENTRY_NOSYNC;
1603 break;
1604 case MADV_AUTOSYNC:
1605 current->eflags &= ~MAP_ENTRY_NOSYNC;
1606 break;
1607 case MADV_NOCORE:
1608 current->eflags |= MAP_ENTRY_NOCOREDUMP;
1609 break;
1610 case MADV_CORE:
1611 current->eflags &= ~MAP_ENTRY_NOCOREDUMP;
1612 break;
1613 default:
1614 break;
1615 }
1616 vm_map_simplify_entry(map, current);
1617 }
1618 vm_map_unlock(map);
1619 } else {
1620 vm_pindex_t pindex;
1621 int count;
1622
1623 /*
1624 * madvise behaviors that are implemented in the underlying
1625 * vm_object.
1626 *
1627 * Since we don't clip the vm_map_entry, we have to clip
1628 * the vm_object pindex and count.
1629 */
1630 for (current = entry;
1631 (current != &map->header) && (current->start < end);
1632 current = current->next
1633 ) {
1634 vm_offset_t useStart;
1635
1636 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
1637 continue;
1638
1639 pindex = OFF_TO_IDX(current->offset);
1640 count = atop(current->end - current->start);
1641 useStart = current->start;
1642
1643 if (current->start < start) {
1644 pindex += atop(start - current->start);
1645 count -= atop(start - current->start);
1646 useStart = start;
1647 }
1648 if (current->end > end)
1649 count -= atop(current->end - end);
1650
1651 if (count <= 0)
1652 continue;
1653
1654 vm_object_madvise(current->object.vm_object,
1655 pindex, count, behav);
1656 if (behav == MADV_WILLNEED) {
1657 vm_map_pmap_enter(map,
1658 useStart,
1659 current->protection,
1660 current->object.vm_object,
1661 pindex,
1662 (count << PAGE_SHIFT),
1663 MAP_PREFAULT_MADVISE
1664 );
1665 }
1666 }
1667 vm_map_unlock_read(map);
1668 }
1669 return (0);
1670 }
1671
1672
1673 /*
1674 * vm_map_inherit:
1675 *
1676 * Sets the inheritance of the specified address
1677 * range in the target map. Inheritance
1678 * affects how the map will be shared with
1679 * child maps at the time of vm_map_fork.
1680 */
1681 int
1682 vm_map_inherit(vm_map_t map, vm_offset_t start, vm_offset_t end,
1683 vm_inherit_t new_inheritance)
1684 {
1685 vm_map_entry_t entry;
1686 vm_map_entry_t temp_entry;
1687
1688 switch (new_inheritance) {
1689 case VM_INHERIT_NONE:
1690 case VM_INHERIT_COPY:
1691 case VM_INHERIT_SHARE:
1692 break;
1693 default:
1694 return (KERN_INVALID_ARGUMENT);
1695 }
1696 vm_map_lock(map);
1697 VM_MAP_RANGE_CHECK(map, start, end);
1698 if (vm_map_lookup_entry(map, start, &temp_entry)) {
1699 entry = temp_entry;
1700 vm_map_clip_start(map, entry, start);
1701 } else
1702 entry = temp_entry->next;
1703 while ((entry != &map->header) && (entry->start < end)) {
1704 vm_map_clip_end(map, entry, end);
1705 entry->inheritance = new_inheritance;
1706 vm_map_simplify_entry(map, entry);
1707 entry = entry->next;
1708 }
1709 vm_map_unlock(map);
1710 return (KERN_SUCCESS);
1711 }
1712
1713 /*
1714 * vm_map_unwire:
1715 *
1716 * Implements both kernel and user unwiring.
1717 */
1718 int
1719 vm_map_unwire(vm_map_t map, vm_offset_t start, vm_offset_t end,
1720 int flags)
1721 {
1722 vm_map_entry_t entry, first_entry, tmp_entry;
1723 vm_offset_t saved_start;
1724 unsigned int last_timestamp;
1725 int rv;
1726 boolean_t need_wakeup, result, user_unwire;
1727
1728 user_unwire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE;
1729 vm_map_lock(map);
1730 VM_MAP_RANGE_CHECK(map, start, end);
1731 if (!vm_map_lookup_entry(map, start, &first_entry)) {
1732 if (flags & VM_MAP_WIRE_HOLESOK)
1733 first_entry = first_entry->next;
1734 else {
1735 vm_map_unlock(map);
1736 return (KERN_INVALID_ADDRESS);
1737 }
1738 }
1739 last_timestamp = map->timestamp;
1740 entry = first_entry;
1741 while (entry != &map->header && entry->start < end) {
1742 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
1743 /*
1744 * We have not yet clipped the entry.
1745 */
1746 saved_start = (start >= entry->start) ? start :
1747 entry->start;
1748 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
1749 if (vm_map_unlock_and_wait(map, user_unwire)) {
1750 /*
1751 * Allow interruption of user unwiring?
1752 */
1753 }
1754 vm_map_lock(map);
1755 if (last_timestamp+1 != map->timestamp) {
1756 /*
1757 * Look again for the entry because the map was
1758 * modified while it was unlocked.
1759 * Specifically, the entry may have been
1760 * clipped, merged, or deleted.
1761 */
1762 if (!vm_map_lookup_entry(map, saved_start,
1763 &tmp_entry)) {
1764 if (flags & VM_MAP_WIRE_HOLESOK)
1765 tmp_entry = tmp_entry->next;
1766 else {
1767 if (saved_start == start) {
1768 /*
1769 * First_entry has been deleted.
1770 */
1771 vm_map_unlock(map);
1772 return (KERN_INVALID_ADDRESS);
1773 }
1774 end = saved_start;
1775 rv = KERN_INVALID_ADDRESS;
1776 goto done;
1777 }
1778 }
1779 if (entry == first_entry)
1780 first_entry = tmp_entry;
1781 else
1782 first_entry = NULL;
1783 entry = tmp_entry;
1784 }
1785 last_timestamp = map->timestamp;
1786 continue;
1787 }
1788 vm_map_clip_start(map, entry, start);
1789 vm_map_clip_end(map, entry, end);
1790 /*
1791 * Mark the entry in case the map lock is released. (See
1792 * above.)
1793 */
1794 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
1795 /*
1796 * Check the map for holes in the specified region.
1797 * If VM_MAP_WIRE_HOLESOK was specified, skip this check.
1798 */
1799 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) &&
1800 (entry->end < end && (entry->next == &map->header ||
1801 entry->next->start > entry->end))) {
1802 end = entry->end;
1803 rv = KERN_INVALID_ADDRESS;
1804 goto done;
1805 }
1806 /*
1807 * If system unwiring, require that the entry is system wired.
1808 */
1809 if (!user_unwire &&
1810 vm_map_entry_system_wired_count(entry) == 0) {
1811 end = entry->end;
1812 rv = KERN_INVALID_ARGUMENT;
1813 goto done;
1814 }
1815 entry = entry->next;
1816 }
1817 rv = KERN_SUCCESS;
1818 done:
1819 need_wakeup = FALSE;
1820 if (first_entry == NULL) {
1821 result = vm_map_lookup_entry(map, start, &first_entry);
1822 if (!result && (flags & VM_MAP_WIRE_HOLESOK))
1823 first_entry = first_entry->next;
1824 else
1825 KASSERT(result, ("vm_map_unwire: lookup failed"));
1826 }
1827 entry = first_entry;
1828 while (entry != &map->header && entry->start < end) {
1829 if (rv == KERN_SUCCESS && (!user_unwire ||
1830 (entry->eflags & MAP_ENTRY_USER_WIRED))) {
1831 if (user_unwire)
1832 entry->eflags &= ~MAP_ENTRY_USER_WIRED;
1833 entry->wired_count--;
1834 if (entry->wired_count == 0) {
1835 /*
1836 * Retain the map lock.
1837 */
1838 vm_fault_unwire(map, entry->start, entry->end,
1839 entry->object.vm_object != NULL &&
1840 entry->object.vm_object->type == OBJT_DEVICE);
1841 }
1842 }
1843 KASSERT(entry->eflags & MAP_ENTRY_IN_TRANSITION,
1844 ("vm_map_unwire: in-transition flag missing"));
1845 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION;
1846 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
1847 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
1848 need_wakeup = TRUE;
1849 }
1850 vm_map_simplify_entry(map, entry);
1851 entry = entry->next;
1852 }
1853 vm_map_unlock(map);
1854 if (need_wakeup)
1855 vm_map_wakeup(map);
1856 return (rv);
1857 }
1858
1859 /*
1860 * vm_map_wire:
1861 *
1862 * Implements both kernel and user wiring.
1863 */
1864 int
1865 vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t end,
1866 int flags)
1867 {
1868 vm_map_entry_t entry, first_entry, tmp_entry;
1869 vm_offset_t saved_end, saved_start;
1870 unsigned int last_timestamp;
1871 int rv;
1872 boolean_t fictitious, need_wakeup, result, user_wire;
1873
1874 user_wire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE;
1875 vm_map_lock(map);
1876 VM_MAP_RANGE_CHECK(map, start, end);
1877 if (!vm_map_lookup_entry(map, start, &first_entry)) {
1878 if (flags & VM_MAP_WIRE_HOLESOK)
1879 first_entry = first_entry->next;
1880 else {
1881 vm_map_unlock(map);
1882 return (KERN_INVALID_ADDRESS);
1883 }
1884 }
1885 last_timestamp = map->timestamp;
1886 entry = first_entry;
1887 while (entry != &map->header && entry->start < end) {
1888 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
1889 /*
1890 * We have not yet clipped the entry.
1891 */
1892 saved_start = (start >= entry->start) ? start :
1893 entry->start;
1894 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
1895 if (vm_map_unlock_and_wait(map, user_wire)) {
1896 /*
1897 * Allow interruption of user wiring?
1898 */
1899 }
1900 vm_map_lock(map);
1901 if (last_timestamp + 1 != map->timestamp) {
1902 /*
1903 * Look again for the entry because the map was
1904 * modified while it was unlocked.
1905 * Specifically, the entry may have been
1906 * clipped, merged, or deleted.
1907 */
1908 if (!vm_map_lookup_entry(map, saved_start,
1909 &tmp_entry)) {
1910 if (flags & VM_MAP_WIRE_HOLESOK)
1911 tmp_entry = tmp_entry->next;
1912 else {
1913 if (saved_start == start) {
1914 /*
1915 * first_entry has been deleted.
1916 */
1917 vm_map_unlock(map);
1918 return (KERN_INVALID_ADDRESS);
1919 }
1920 end = saved_start;
1921 rv = KERN_INVALID_ADDRESS;
1922 goto done;
1923 }
1924 }
1925 if (entry == first_entry)
1926 first_entry = tmp_entry;
1927 else
1928 first_entry = NULL;
1929 entry = tmp_entry;
1930 }
1931 last_timestamp = map->timestamp;
1932 continue;
1933 }
1934 vm_map_clip_start(map, entry, start);
1935 vm_map_clip_end(map, entry, end);
1936 /*
1937 * Mark the entry in case the map lock is released. (See
1938 * above.)
1939 */
1940 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
1941 /*
1942 *
1943 */
1944 if (entry->wired_count == 0) {
1945 entry->wired_count++;
1946 saved_start = entry->start;
1947 saved_end = entry->end;
1948 fictitious = entry->object.vm_object != NULL &&
1949 entry->object.vm_object->type == OBJT_DEVICE;
1950 /*
1951 * Release the map lock, relying on the in-transition
1952 * mark.
1953 */
1954 vm_map_unlock(map);
1955 rv = vm_fault_wire(map, saved_start, saved_end,
1956 user_wire, fictitious);
1957 vm_map_lock(map);
1958 if (last_timestamp + 1 != map->timestamp) {
1959 /*
1960 * Look again for the entry because the map was
1961 * modified while it was unlocked. The entry
1962 * may have been clipped, but NOT merged or
1963 * deleted.
1964 */
1965 result = vm_map_lookup_entry(map, saved_start,
1966 &tmp_entry);
1967 KASSERT(result, ("vm_map_wire: lookup failed"));
1968 if (entry == first_entry)
1969 first_entry = tmp_entry;
1970 else
1971 first_entry = NULL;
1972 entry = tmp_entry;
1973 while (entry->end < saved_end) {
1974 if (rv != KERN_SUCCESS) {
1975 KASSERT(entry->wired_count == 1,
1976 ("vm_map_wire: bad count"));
1977 entry->wired_count = -1;
1978 }
1979 entry = entry->next;
1980 }
1981 }
1982 last_timestamp = map->timestamp;
1983 if (rv != KERN_SUCCESS) {
1984 KASSERT(entry->wired_count == 1,
1985 ("vm_map_wire: bad count"));
1986 /*
1987 * Assign an out-of-range value to represent
1988 * the failure to wire this entry.
1989 */
1990 entry->wired_count = -1;
1991 end = entry->end;
1992 goto done;
1993 }
1994 } else if (!user_wire ||
1995 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
1996 entry->wired_count++;
1997 }
1998 /*
1999 * Check the map for holes in the specified region.
2000 * If VM_MAP_WIRE_HOLESOK was specified, skip this check.
2001 */
2002 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) &&
2003 (entry->end < end && (entry->next == &map->header ||
2004 entry->next->start > entry->end))) {
2005 end = entry->end;
2006 rv = KERN_INVALID_ADDRESS;
2007 goto done;
2008 }
2009 entry = entry->next;
2010 }
2011 rv = KERN_SUCCESS;
2012 done:
2013 need_wakeup = FALSE;
2014 if (first_entry == NULL) {
2015 result = vm_map_lookup_entry(map, start, &first_entry);
2016 if (!result && (flags & VM_MAP_WIRE_HOLESOK))
2017 first_entry = first_entry->next;
2018 else
2019 KASSERT(result, ("vm_map_wire: lookup failed"));
2020 }
2021 entry = first_entry;
2022 while (entry != &map->header && entry->start < end) {
2023 if (rv == KERN_SUCCESS) {
2024 if (user_wire)
2025 entry->eflags |= MAP_ENTRY_USER_WIRED;
2026 } else if (entry->wired_count == -1) {
2027 /*
2028 * Wiring failed on this entry. Thus, unwiring is
2029 * unnecessary.
2030 */
2031 entry->wired_count = 0;
2032 } else {
2033 if (!user_wire ||
2034 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0)
2035 entry->wired_count--;
2036 if (entry->wired_count == 0) {
2037 /*
2038 * Retain the map lock.
2039 */
2040 vm_fault_unwire(map, entry->start, entry->end,
2041 entry->object.vm_object != NULL &&
2042 entry->object.vm_object->type == OBJT_DEVICE);
2043 }
2044 }
2045 KASSERT(entry->eflags & MAP_ENTRY_IN_TRANSITION,
2046 ("vm_map_wire: in-transition flag missing"));
2047 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION;
2048 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
2049 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
2050 need_wakeup = TRUE;
2051 }
2052 vm_map_simplify_entry(map, entry);
2053 entry = entry->next;
2054 }
2055 vm_map_unlock(map);
2056 if (need_wakeup)
2057 vm_map_wakeup(map);
2058 return (rv);
2059 }
2060
2061 /*
2062 * vm_map_sync
2063 *
2064 * Push any dirty cached pages in the address range to their pager.
2065 * If syncio is TRUE, dirty pages are written synchronously.
2066 * If invalidate is TRUE, any cached pages are freed as well.
2067 *
2068 * If the size of the region from start to end is zero, we are
2069 * supposed to flush all modified pages within the region containing
2070 * start. Unfortunately, a region can be split or coalesced with
2071 * neighboring regions, making it difficult to determine what the
2072 * original region was. Therefore, we approximate this requirement by
2073 * flushing the current region containing start.
2074 *
2075 * Returns an error if any part of the specified range is not mapped.
2076 */
2077 int
2078 vm_map_sync(
2079 vm_map_t map,
2080 vm_offset_t start,
2081 vm_offset_t end,
2082 boolean_t syncio,
2083 boolean_t invalidate)
2084 {
2085 vm_map_entry_t current;
2086 vm_map_entry_t entry;
2087 vm_size_t size;
2088 vm_object_t object;
2089 vm_ooffset_t offset;
2090
2091 vm_map_lock_read(map);
2092 VM_MAP_RANGE_CHECK(map, start, end);
2093 if (!vm_map_lookup_entry(map, start, &entry)) {
2094 vm_map_unlock_read(map);
2095 return (KERN_INVALID_ADDRESS);
2096 } else if (start == end) {
2097 start = entry->start;
2098 end = entry->end;
2099 }
2100 /*
2101 * Make a first pass to check for user-wired memory and holes.
2102 */
2103 for (current = entry; current->start < end; current = current->next) {
2104 if (invalidate && (current->eflags & MAP_ENTRY_USER_WIRED)) {
2105 vm_map_unlock_read(map);
2106 return (KERN_INVALID_ARGUMENT);
2107 }
2108 if (end > current->end &&
2109 (current->next == &map->header ||
2110 current->end != current->next->start)) {
2111 vm_map_unlock_read(map);
2112 return (KERN_INVALID_ADDRESS);
2113 }
2114 }
2115
2116 if (invalidate) {
2117 VM_LOCK_GIANT();
2118 pmap_remove(map->pmap, start, end);
2119 VM_UNLOCK_GIANT();
2120 }
2121 /*
2122 * Make a second pass, cleaning/uncaching pages from the indicated
2123 * objects as we go.
2124 */
2125 for (current = entry; current->start < end; current = current->next) {
2126 offset = current->offset + (start - current->start);
2127 size = (end <= current->end ? end : current->end) - start;
2128 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
2129 vm_map_t smap;
2130 vm_map_entry_t tentry;
2131 vm_size_t tsize;
2132
2133 smap = current->object.sub_map;
2134 vm_map_lock_read(smap);
2135 (void) vm_map_lookup_entry(smap, offset, &tentry);
2136 tsize = tentry->end - offset;
2137 if (tsize < size)
2138 size = tsize;
2139 object = tentry->object.vm_object;
2140 offset = tentry->offset + (offset - tentry->start);
2141 vm_map_unlock_read(smap);
2142 } else {
2143 object = current->object.vm_object;
2144 }
2145 vm_object_sync(object, offset, size, syncio, invalidate);
2146 start += size;
2147 }
2148
2149 vm_map_unlock_read(map);
2150 return (KERN_SUCCESS);
2151 }
2152
2153 /*
2154 * vm_map_entry_unwire: [ internal use only ]
2155 *
2156 * Make the region specified by this entry pageable.
2157 *
2158 * The map in question should be locked.
2159 * [This is the reason for this routine's existence.]
2160 */
2161 static void
2162 vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry)
2163 {
2164 vm_fault_unwire(map, entry->start, entry->end,
2165 entry->object.vm_object != NULL &&
2166 entry->object.vm_object->type == OBJT_DEVICE);
2167 entry->wired_count = 0;
2168 }
2169
2170 /*
2171 * vm_map_entry_delete: [ internal use only ]
2172 *
2173 * Deallocate the given entry from the target map.
2174 */
2175 static void
2176 vm_map_entry_delete(vm_map_t map, vm_map_entry_t entry)
2177 {
2178 vm_object_t object;
2179 vm_pindex_t offidxstart, offidxend, count;
2180
2181 vm_map_entry_unlink(map, entry);
2182 map->size -= entry->end - entry->start;
2183
2184 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0 &&
2185 (object = entry->object.vm_object) != NULL) {
2186 count = OFF_TO_IDX(entry->end - entry->start);
2187 offidxstart = OFF_TO_IDX(entry->offset);
2188 offidxend = offidxstart + count;
2189 VM_OBJECT_LOCK(object);
2190 if (object->ref_count != 1 &&
2191 ((object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING ||
2192 object == kernel_object || object == kmem_object) &&
2193 (object->type == OBJT_DEFAULT || object->type == OBJT_SWAP)) {
2194 vm_object_collapse(object);
2195 vm_object_page_remove(object, offidxstart, offidxend, FALSE);
2196 if (object->type == OBJT_SWAP)
2197 swap_pager_freespace(object, offidxstart, count);
2198 if (offidxend >= object->size &&
2199 offidxstart < object->size)
2200 object->size = offidxstart;
2201 }
2202 VM_OBJECT_UNLOCK(object);
2203 vm_object_deallocate(object);
2204 }
2205
2206 vm_map_entry_dispose(map, entry);
2207 }
2208
2209 /*
2210 * vm_map_delete: [ internal use only ]
2211 *
2212 * Deallocates the given address range from the target
2213 * map.
2214 */
2215 int
2216 vm_map_delete(vm_map_t map, vm_offset_t start, vm_offset_t end)
2217 {
2218 vm_map_entry_t entry;
2219 vm_map_entry_t first_entry;
2220
2221 /*
2222 * Find the start of the region, and clip it
2223 */
2224 if (!vm_map_lookup_entry(map, start, &first_entry))
2225 entry = first_entry->next;
2226 else {
2227 entry = first_entry;
2228 vm_map_clip_start(map, entry, start);
2229 }
2230
2231 /*
2232 * Step through all entries in this region
2233 */
2234 while ((entry != &map->header) && (entry->start < end)) {
2235 vm_map_entry_t next;
2236
2237 /*
2238 * Wait for wiring or unwiring of an entry to complete.
2239 * Also wait for any system wirings to disappear on
2240 * user maps.
2241 */
2242 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 ||
2243 (vm_map_pmap(map) != kernel_pmap &&
2244 vm_map_entry_system_wired_count(entry) != 0)) {
2245 unsigned int last_timestamp;
2246 vm_offset_t saved_start;
2247 vm_map_entry_t tmp_entry;
2248
2249 saved_start = entry->start;
2250 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2251 last_timestamp = map->timestamp;
2252 (void) vm_map_unlock_and_wait(map, FALSE);
2253 vm_map_lock(map);
2254 if (last_timestamp + 1 != map->timestamp) {
2255 /*
2256 * Look again for the entry because the map was
2257 * modified while it was unlocked.
2258 * Specifically, the entry may have been
2259 * clipped, merged, or deleted.
2260 */
2261 if (!vm_map_lookup_entry(map, saved_start,
2262 &tmp_entry))
2263 entry = tmp_entry->next;
2264 else {
2265 entry = tmp_entry;
2266 vm_map_clip_start(map, entry,
2267 saved_start);
2268 }
2269 }
2270 continue;
2271 }
2272 vm_map_clip_end(map, entry, end);
2273
2274 next = entry->next;
2275
2276 /*
2277 * Unwire before removing addresses from the pmap; otherwise,
2278 * unwiring will put the entries back in the pmap.
2279 */
2280 if (entry->wired_count != 0) {
2281 vm_map_entry_unwire(map, entry);
2282 }
2283
2284 if (!map->system_map)
2285 VM_LOCK_GIANT();
2286 pmap_remove(map->pmap, entry->start, entry->end);
2287 if (!map->system_map)
2288 VM_UNLOCK_GIANT();
2289
2290 /*
2291 * Delete the entry (which may delete the object) only after
2292 * removing all pmap entries pointing to its pages.
2293 * (Otherwise, its page frames may be reallocated, and any
2294 * modify bits will be set in the wrong object!)
2295 */
2296 vm_map_entry_delete(map, entry);
2297 entry = next;
2298 }
2299 return (KERN_SUCCESS);
2300 }
2301
2302 /*
2303 * vm_map_remove:
2304 *
2305 * Remove the given address range from the target map.
2306 * This is the exported form of vm_map_delete.
2307 */
2308 int
2309 vm_map_remove(vm_map_t map, vm_offset_t start, vm_offset_t end)
2310 {
2311 int result;
2312
2313 vm_map_lock(map);
2314 VM_MAP_RANGE_CHECK(map, start, end);
2315 result = vm_map_delete(map, start, end);
2316 vm_map_unlock(map);
2317 return (result);
2318 }
2319
2320 /*
2321 * vm_map_check_protection:
2322 *
2323 * Assert that the target map allows the specified privilege on the
2324 * entire address region given. The entire region must be allocated.
2325 *
2326 * WARNING! This code does not and should not check whether the
2327 * contents of the region is accessible. For example a smaller file
2328 * might be mapped into a larger address space.
2329 *
2330 * NOTE! This code is also called by munmap().
2331 *
2332 * The map must be locked. A read lock is sufficient.
2333 */
2334 boolean_t
2335 vm_map_check_protection(vm_map_t map, vm_offset_t start, vm_offset_t end,
2336 vm_prot_t protection)
2337 {
2338 vm_map_entry_t entry;
2339 vm_map_entry_t tmp_entry;
2340
2341 if (!vm_map_lookup_entry(map, start, &tmp_entry))
2342 return (FALSE);
2343 entry = tmp_entry;
2344
2345 while (start < end) {
2346 if (entry == &map->header)
2347 return (FALSE);
2348 /*
2349 * No holes allowed!
2350 */
2351 if (start < entry->start)
2352 return (FALSE);
2353 /*
2354 * Check protection associated with entry.
2355 */
2356 if ((entry->protection & protection) != protection)
2357 return (FALSE);
2358 /* go to next entry */
2359 start = entry->end;
2360 entry = entry->next;
2361 }
2362 return (TRUE);
2363 }
2364
2365 /*
2366 * vm_map_copy_entry:
2367 *
2368 * Copies the contents of the source entry to the destination
2369 * entry. The entries *must* be aligned properly.
2370 */
2371 static void
2372 vm_map_copy_entry(
2373 vm_map_t src_map,
2374 vm_map_t dst_map,
2375 vm_map_entry_t src_entry,
2376 vm_map_entry_t dst_entry)
2377 {
2378 vm_object_t src_object;
2379
2380 if ((dst_entry->eflags|src_entry->eflags) & MAP_ENTRY_IS_SUB_MAP)
2381 return;
2382
2383 if (src_entry->wired_count == 0) {
2384
2385 /*
2386 * If the source entry is marked needs_copy, it is already
2387 * write-protected.
2388 */
2389 if ((src_entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) {
2390 pmap_protect(src_map->pmap,
2391 src_entry->start,
2392 src_entry->end,
2393 src_entry->protection & ~VM_PROT_WRITE);
2394 }
2395
2396 /*
2397 * Make a copy of the object.
2398 */
2399 if ((src_object = src_entry->object.vm_object) != NULL) {
2400 VM_OBJECT_LOCK(src_object);
2401 if ((src_object->handle == NULL) &&
2402 (src_object->type == OBJT_DEFAULT ||
2403 src_object->type == OBJT_SWAP)) {
2404 vm_object_collapse(src_object);
2405 if ((src_object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING) {
2406 vm_object_split(src_entry);
2407 src_object = src_entry->object.vm_object;
2408 }
2409 }
2410 vm_object_reference_locked(src_object);
2411 vm_object_clear_flag(src_object, OBJ_ONEMAPPING);
2412 VM_OBJECT_UNLOCK(src_object);
2413 dst_entry->object.vm_object = src_object;
2414 src_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY);
2415 dst_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY);
2416 dst_entry->offset = src_entry->offset;
2417 } else {
2418 dst_entry->object.vm_object = NULL;
2419 dst_entry->offset = 0;
2420 }
2421
2422 pmap_copy(dst_map->pmap, src_map->pmap, dst_entry->start,
2423 dst_entry->end - dst_entry->start, src_entry->start);
2424 } else {
2425 /*
2426 * Of course, wired down pages can't be set copy-on-write.
2427 * Cause wired pages to be copied into the new map by
2428 * simulating faults (the new pages are pageable)
2429 */
2430 vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry);
2431 }
2432 }
2433
2434 /*
2435 * vmspace_map_entry_forked:
2436 * Update the newly-forked vmspace each time a map entry is inherited
2437 * or copied. The values for vm_dsize and vm_tsize are approximate
2438 * (and mostly-obsolete ideas in the face of mmap(2) et al.)
2439 */
2440 static void
2441 vmspace_map_entry_forked(const struct vmspace *vm1, struct vmspace *vm2,
2442 vm_map_entry_t entry)
2443 {
2444 vm_size_t entrysize;
2445 vm_offset_t newend;
2446
2447 entrysize = entry->end - entry->start;
2448 vm2->vm_map.size += entrysize;
2449 if (entry->eflags & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP)) {
2450 vm2->vm_ssize += btoc(entrysize);
2451 } else if (entry->start >= (vm_offset_t)vm1->vm_daddr &&
2452 entry->start < (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)) {
2453 newend = MIN(entry->end,
2454 (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize));
2455 vm2->vm_dsize += btoc(newend - entry->start);
2456 } else if (entry->start >= (vm_offset_t)vm1->vm_taddr &&
2457 entry->start < (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)) {
2458 newend = MIN(entry->end,
2459 (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize));
2460 vm2->vm_tsize += btoc(newend - entry->start);
2461 }
2462 }
2463
2464 /*
2465 * vmspace_fork:
2466 * Create a new process vmspace structure and vm_map
2467 * based on those of an existing process. The new map
2468 * is based on the old map, according to the inheritance
2469 * values on the regions in that map.
2470 *
2471 * XXX It might be worth coalescing the entries added to the new vmspace.
2472 *
2473 * The source map must not be locked.
2474 */
2475 struct vmspace *
2476 vmspace_fork(struct vmspace *vm1)
2477 {
2478 struct vmspace *vm2;
2479 vm_map_t old_map = &vm1->vm_map;
2480 vm_map_t new_map;
2481 vm_map_entry_t old_entry;
2482 vm_map_entry_t new_entry;
2483 vm_object_t object;
2484
2485 GIANT_REQUIRED;
2486
2487 vm_map_lock(old_map);
2488
2489 vm2 = vmspace_alloc(old_map->min_offset, old_map->max_offset);
2490 vm2->vm_taddr = vm1->vm_taddr;
2491 vm2->vm_daddr = vm1->vm_daddr;
2492 vm2->vm_maxsaddr = vm1->vm_maxsaddr;
2493 new_map = &vm2->vm_map; /* XXX */
2494 new_map->timestamp = 1;
2495
2496 /* Do not inherit the MAP_WIREFUTURE property. */
2497 if ((new_map->flags & MAP_WIREFUTURE) == MAP_WIREFUTURE)
2498 new_map->flags &= ~MAP_WIREFUTURE;
2499
2500 old_entry = old_map->header.next;
2501
2502 while (old_entry != &old_map->header) {
2503 if (old_entry->eflags & MAP_ENTRY_IS_SUB_MAP)
2504 panic("vm_map_fork: encountered a submap");
2505
2506 switch (old_entry->inheritance) {
2507 case VM_INHERIT_NONE:
2508 break;
2509
2510 case VM_INHERIT_SHARE:
2511 /*
2512 * Clone the entry, creating the shared object if necessary.
2513 */
2514 object = old_entry->object.vm_object;
2515 if (object == NULL) {
2516 object = vm_object_allocate(OBJT_DEFAULT,
2517 atop(old_entry->end - old_entry->start));
2518 old_entry->object.vm_object = object;
2519 old_entry->offset = (vm_offset_t) 0;
2520 }
2521
2522 /*
2523 * Add the reference before calling vm_object_shadow
2524 * to insure that a shadow object is created.
2525 */
2526 vm_object_reference(object);
2527 if (old_entry->eflags & MAP_ENTRY_NEEDS_COPY) {
2528 vm_object_shadow(&old_entry->object.vm_object,
2529 &old_entry->offset,
2530 atop(old_entry->end - old_entry->start));
2531 old_entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
2532 /* Transfer the second reference too. */
2533 vm_object_reference(
2534 old_entry->object.vm_object);
2535 vm_object_deallocate(object);
2536 object = old_entry->object.vm_object;
2537 }
2538 VM_OBJECT_LOCK(object);
2539 vm_object_clear_flag(object, OBJ_ONEMAPPING);
2540 VM_OBJECT_UNLOCK(object);
2541
2542 /*
2543 * Clone the entry, referencing the shared object.
2544 */
2545 new_entry = vm_map_entry_create(new_map);
2546 *new_entry = *old_entry;
2547 new_entry->eflags &= ~MAP_ENTRY_USER_WIRED;
2548 new_entry->wired_count = 0;
2549
2550 /*
2551 * Insert the entry into the new map -- we know we're
2552 * inserting at the end of the new map.
2553 */
2554 vm_map_entry_link(new_map, new_map->header.prev,
2555 new_entry);
2556 vmspace_map_entry_forked(vm1, vm2, new_entry);
2557
2558 /*
2559 * Update the physical map
2560 */
2561 pmap_copy(new_map->pmap, old_map->pmap,
2562 new_entry->start,
2563 (old_entry->end - old_entry->start),
2564 old_entry->start);
2565 break;
2566
2567 case VM_INHERIT_COPY:
2568 /*
2569 * Clone the entry and link into the map.
2570 */
2571 new_entry = vm_map_entry_create(new_map);
2572 *new_entry = *old_entry;
2573 new_entry->eflags &= ~MAP_ENTRY_USER_WIRED;
2574 new_entry->wired_count = 0;
2575 new_entry->object.vm_object = NULL;
2576 vm_map_entry_link(new_map, new_map->header.prev,
2577 new_entry);
2578 vmspace_map_entry_forked(vm1, vm2, new_entry);
2579 vm_map_copy_entry(old_map, new_map, old_entry,
2580 new_entry);
2581 break;
2582 }
2583 old_entry = old_entry->next;
2584 }
2585
2586 vm_map_unlock(old_map);
2587
2588 return (vm2);
2589 }
2590
2591 int
2592 vm_map_stack(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
2593 vm_prot_t prot, vm_prot_t max, int cow)
2594 {
2595 vm_map_entry_t new_entry, prev_entry;
2596 vm_offset_t bot, top;
2597 vm_size_t init_ssize;
2598 int orient, rv;
2599 rlim_t vmemlim;
2600
2601 /*
2602 * The stack orientation is piggybacked with the cow argument.
2603 * Extract it into orient and mask the cow argument so that we
2604 * don't pass it around further.
2605 * NOTE: We explicitly allow bi-directional stacks.
2606 */
2607 orient = cow & (MAP_STACK_GROWS_DOWN|MAP_STACK_GROWS_UP);
2608 cow &= ~orient;
2609 KASSERT(orient != 0, ("No stack grow direction"));
2610
2611 if (addrbos < vm_map_min(map) || addrbos > map->max_offset)
2612 return (KERN_NO_SPACE);
2613
2614 init_ssize = (max_ssize < sgrowsiz) ? max_ssize : sgrowsiz;
2615
2616 PROC_LOCK(curthread->td_proc);
2617 vmemlim = lim_cur(curthread->td_proc, RLIMIT_VMEM);
2618 PROC_UNLOCK(curthread->td_proc);
2619
2620 vm_map_lock(map);
2621
2622 /* If addr is already mapped, no go */
2623 if (vm_map_lookup_entry(map, addrbos, &prev_entry)) {
2624 vm_map_unlock(map);
2625 return (KERN_NO_SPACE);
2626 }
2627
2628 /* If we would blow our VMEM resource limit, no go */
2629 if (map->size + init_ssize > vmemlim) {
2630 vm_map_unlock(map);
2631 return (KERN_NO_SPACE);
2632 }
2633
2634 /*
2635 * If we can't accomodate max_ssize in the current mapping, no go.
2636 * However, we need to be aware that subsequent user mappings might
2637 * map into the space we have reserved for stack, and currently this
2638 * space is not protected.
2639 *
2640 * Hopefully we will at least detect this condition when we try to
2641 * grow the stack.
2642 */
2643 if ((prev_entry->next != &map->header) &&
2644 (prev_entry->next->start < addrbos + max_ssize)) {
2645 vm_map_unlock(map);
2646 return (KERN_NO_SPACE);
2647 }
2648
2649 /*
2650 * We initially map a stack of only init_ssize. We will grow as
2651 * needed later. Depending on the orientation of the stack (i.e.
2652 * the grow direction) we either map at the top of the range, the
2653 * bottom of the range or in the middle.
2654 *
2655 * Note: we would normally expect prot and max to be VM_PROT_ALL,
2656 * and cow to be 0. Possibly we should eliminate these as input
2657 * parameters, and just pass these values here in the insert call.
2658 */
2659 if (orient == MAP_STACK_GROWS_DOWN)
2660 bot = addrbos + max_ssize - init_ssize;
2661 else if (orient == MAP_STACK_GROWS_UP)
2662 bot = addrbos;
2663 else
2664 bot = round_page(addrbos + max_ssize/2 - init_ssize/2);
2665 top = bot + init_ssize;
2666 rv = vm_map_insert(map, NULL, 0, bot, top, prot, max, cow);
2667
2668 /* Now set the avail_ssize amount. */
2669 if (rv == KERN_SUCCESS) {
2670 if (prev_entry != &map->header)
2671 vm_map_clip_end(map, prev_entry, bot);
2672 new_entry = prev_entry->next;
2673 if (new_entry->end != top || new_entry->start != bot)
2674 panic("Bad entry start/end for new stack entry");
2675
2676 new_entry->avail_ssize = max_ssize - init_ssize;
2677 if (orient & MAP_STACK_GROWS_DOWN)
2678 new_entry->eflags |= MAP_ENTRY_GROWS_DOWN;
2679 if (orient & MAP_STACK_GROWS_UP)
2680 new_entry->eflags |= MAP_ENTRY_GROWS_UP;
2681 }
2682
2683 vm_map_unlock(map);
2684 return (rv);
2685 }
2686
2687 /* Attempts to grow a vm stack entry. Returns KERN_SUCCESS if the
2688 * desired address is already mapped, or if we successfully grow
2689 * the stack. Also returns KERN_SUCCESS if addr is outside the
2690 * stack range (this is strange, but preserves compatibility with
2691 * the grow function in vm_machdep.c).
2692 */
2693 int
2694 vm_map_growstack(struct proc *p, vm_offset_t addr)
2695 {
2696 vm_map_entry_t next_entry, prev_entry;
2697 vm_map_entry_t new_entry, stack_entry;
2698 struct vmspace *vm = p->p_vmspace;
2699 vm_map_t map = &vm->vm_map;
2700 vm_offset_t end;
2701 size_t grow_amount, max_grow;
2702 rlim_t stacklim, vmemlim;
2703 int is_procstack, rv;
2704
2705 Retry:
2706 PROC_LOCK(p);
2707 stacklim = lim_cur(p, RLIMIT_STACK);
2708 vmemlim = lim_cur(p, RLIMIT_VMEM);
2709 PROC_UNLOCK(p);
2710
2711 vm_map_lock_read(map);
2712
2713 /* If addr is already in the entry range, no need to grow.*/
2714 if (vm_map_lookup_entry(map, addr, &prev_entry)) {
2715 vm_map_unlock_read(map);
2716 return (KERN_SUCCESS);
2717 }
2718
2719 next_entry = prev_entry->next;
2720 if (!(prev_entry->eflags & MAP_ENTRY_GROWS_UP)) {
2721 /*
2722 * This entry does not grow upwards. Since the address lies
2723 * beyond this entry, the next entry (if one exists) has to
2724 * be a downward growable entry. The entry list header is
2725 * never a growable entry, so it suffices to check the flags.
2726 */
2727 if (!(next_entry->eflags & MAP_ENTRY_GROWS_DOWN)) {
2728 vm_map_unlock_read(map);
2729 return (KERN_SUCCESS);
2730 }
2731 stack_entry = next_entry;
2732 } else {
2733 /*
2734 * This entry grows upward. If the next entry does not at
2735 * least grow downwards, this is the entry we need to grow.
2736 * otherwise we have two possible choices and we have to
2737 * select one.
2738 */
2739 if (next_entry->eflags & MAP_ENTRY_GROWS_DOWN) {
2740 /*
2741 * We have two choices; grow the entry closest to
2742 * the address to minimize the amount of growth.
2743 */
2744 if (addr - prev_entry->end <= next_entry->start - addr)
2745 stack_entry = prev_entry;
2746 else
2747 stack_entry = next_entry;
2748 } else
2749 stack_entry = prev_entry;
2750 }
2751
2752 if (stack_entry == next_entry) {
2753 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_DOWN, ("foo"));
2754 KASSERT(addr < stack_entry->start, ("foo"));
2755 end = (prev_entry != &map->header) ? prev_entry->end :
2756 stack_entry->start - stack_entry->avail_ssize;
2757 grow_amount = roundup(stack_entry->start - addr, PAGE_SIZE);
2758 max_grow = stack_entry->start - end;
2759 } else {
2760 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_UP, ("foo"));
2761 KASSERT(addr >= stack_entry->end, ("foo"));
2762 end = (next_entry != &map->header) ? next_entry->start :
2763 stack_entry->end + stack_entry->avail_ssize;
2764 grow_amount = roundup(addr + 1 - stack_entry->end, PAGE_SIZE);
2765 max_grow = end - stack_entry->end;
2766 }
2767
2768 if (grow_amount > stack_entry->avail_ssize) {
2769 vm_map_unlock_read(map);
2770 return (KERN_NO_SPACE);
2771 }
2772
2773 /*
2774 * If there is no longer enough space between the entries nogo, and
2775 * adjust the available space. Note: this should only happen if the
2776 * user has mapped into the stack area after the stack was created,
2777 * and is probably an error.
2778 *
2779 * This also effectively destroys any guard page the user might have
2780 * intended by limiting the stack size.
2781 */
2782 if (grow_amount > max_grow) {
2783 if (vm_map_lock_upgrade(map))
2784 goto Retry;
2785
2786 stack_entry->avail_ssize = max_grow;
2787
2788 vm_map_unlock(map);
2789 return (KERN_NO_SPACE);
2790 }
2791
2792 is_procstack = (addr >= (vm_offset_t)vm->vm_maxsaddr) ? 1 : 0;
2793
2794 /*
2795 * If this is the main process stack, see if we're over the stack
2796 * limit.
2797 */
2798 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) {
2799 vm_map_unlock_read(map);
2800 return (KERN_NO_SPACE);
2801 }
2802
2803 /* Round up the grow amount modulo SGROWSIZ */
2804 grow_amount = roundup (grow_amount, sgrowsiz);
2805 if (grow_amount > stack_entry->avail_ssize)
2806 grow_amount = stack_entry->avail_ssize;
2807 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) {
2808 grow_amount = stacklim - ctob(vm->vm_ssize);
2809 }
2810
2811 /* If we would blow our VMEM resource limit, no go */
2812 if (map->size + grow_amount > vmemlim) {
2813 vm_map_unlock_read(map);
2814 return (KERN_NO_SPACE);
2815 }
2816
2817 if (vm_map_lock_upgrade(map))
2818 goto Retry;
2819
2820 if (stack_entry == next_entry) {
2821 /*
2822 * Growing downward.
2823 */
2824 /* Get the preliminary new entry start value */
2825 addr = stack_entry->start - grow_amount;
2826
2827 /*
2828 * If this puts us into the previous entry, cut back our
2829 * growth to the available space. Also, see the note above.
2830 */
2831 if (addr < end) {
2832 stack_entry->avail_ssize = max_grow;
2833 addr = end;
2834 }
2835
2836 rv = vm_map_insert(map, NULL, 0, addr, stack_entry->start,
2837 p->p_sysent->sv_stackprot, VM_PROT_ALL, 0);
2838
2839 /* Adjust the available stack space by the amount we grew. */
2840 if (rv == KERN_SUCCESS) {
2841 if (prev_entry != &map->header)
2842 vm_map_clip_end(map, prev_entry, addr);
2843 new_entry = prev_entry->next;
2844 KASSERT(new_entry == stack_entry->prev, ("foo"));
2845 KASSERT(new_entry->end == stack_entry->start, ("foo"));
2846 KASSERT(new_entry->start == addr, ("foo"));
2847 grow_amount = new_entry->end - new_entry->start;
2848 new_entry->avail_ssize = stack_entry->avail_ssize -
2849 grow_amount;
2850 stack_entry->eflags &= ~MAP_ENTRY_GROWS_DOWN;
2851 new_entry->eflags |= MAP_ENTRY_GROWS_DOWN;
2852 }
2853 } else {
2854 /*
2855 * Growing upward.
2856 */
2857 addr = stack_entry->end + grow_amount;
2858
2859 /*
2860 * If this puts us into the next entry, cut back our growth
2861 * to the available space. Also, see the note above.
2862 */
2863 if (addr > end) {
2864 stack_entry->avail_ssize = end - stack_entry->end;
2865 addr = end;
2866 }
2867
2868 grow_amount = addr - stack_entry->end;
2869
2870 /* Grow the underlying object if applicable. */
2871 if (stack_entry->object.vm_object == NULL ||
2872 vm_object_coalesce(stack_entry->object.vm_object,
2873 stack_entry->offset,
2874 (vm_size_t)(stack_entry->end - stack_entry->start),
2875 (vm_size_t)grow_amount)) {
2876 map->size += (addr - stack_entry->end);
2877 /* Update the current entry. */
2878 stack_entry->end = addr;
2879 stack_entry->avail_ssize -= grow_amount;
2880 vm_map_entry_resize_free(map, stack_entry);
2881 rv = KERN_SUCCESS;
2882
2883 if (next_entry != &map->header)
2884 vm_map_clip_start(map, next_entry, addr);
2885 } else
2886 rv = KERN_FAILURE;
2887 }
2888
2889 if (rv == KERN_SUCCESS && is_procstack)
2890 vm->vm_ssize += btoc(grow_amount);
2891
2892 vm_map_unlock(map);
2893
2894 /*
2895 * Heed the MAP_WIREFUTURE flag if it was set for this process.
2896 */
2897 if (rv == KERN_SUCCESS && (map->flags & MAP_WIREFUTURE)) {
2898 vm_map_wire(map,
2899 (stack_entry == next_entry) ? addr : addr - grow_amount,
2900 (stack_entry == next_entry) ? stack_entry->start : addr,
2901 (p->p_flag & P_SYSTEM)
2902 ? VM_MAP_WIRE_SYSTEM|VM_MAP_WIRE_NOHOLES
2903 : VM_MAP_WIRE_USER|VM_MAP_WIRE_NOHOLES);
2904 }
2905
2906 return (rv);
2907 }
2908
2909 /*
2910 * Unshare the specified VM space for exec. If other processes are
2911 * mapped to it, then create a new one. The new vmspace is null.
2912 */
2913 void
2914 vmspace_exec(struct proc *p, vm_offset_t minuser, vm_offset_t maxuser)
2915 {
2916 struct vmspace *oldvmspace = p->p_vmspace;
2917 struct vmspace *newvmspace;
2918
2919 GIANT_REQUIRED;
2920 newvmspace = vmspace_alloc(minuser, maxuser);
2921 newvmspace->vm_swrss = oldvmspace->vm_swrss;
2922 /*
2923 * This code is written like this for prototype purposes. The
2924 * goal is to avoid running down the vmspace here, but let the
2925 * other process's that are still using the vmspace to finally
2926 * run it down. Even though there is little or no chance of blocking
2927 * here, it is a good idea to keep this form for future mods.
2928 */
2929 p->p_vmspace = newvmspace;
2930 if (p == curthread->td_proc) /* XXXKSE ? */
2931 pmap_activate(curthread);
2932 vmspace_free(oldvmspace);
2933 }
2934
2935 /*
2936 * Unshare the specified VM space for forcing COW. This
2937 * is called by rfork, for the (RFMEM|RFPROC) == 0 case.
2938 */
2939 void
2940 vmspace_unshare(struct proc *p)
2941 {
2942 struct vmspace *oldvmspace = p->p_vmspace;
2943 struct vmspace *newvmspace;
2944
2945 GIANT_REQUIRED;
2946 if (oldvmspace->vm_refcnt == 1)
2947 return;
2948 newvmspace = vmspace_fork(oldvmspace);
2949 p->p_vmspace = newvmspace;
2950 if (p == curthread->td_proc) /* XXXKSE ? */
2951 pmap_activate(curthread);
2952 vmspace_free(oldvmspace);
2953 }
2954
2955 /*
2956 * vm_map_lookup:
2957 *
2958 * Finds the VM object, offset, and
2959 * protection for a given virtual address in the
2960 * specified map, assuming a page fault of the
2961 * type specified.
2962 *
2963 * Leaves the map in question locked for read; return
2964 * values are guaranteed until a vm_map_lookup_done
2965 * call is performed. Note that the map argument
2966 * is in/out; the returned map must be used in
2967 * the call to vm_map_lookup_done.
2968 *
2969 * A handle (out_entry) is returned for use in
2970 * vm_map_lookup_done, to make that fast.
2971 *
2972 * If a lookup is requested with "write protection"
2973 * specified, the map may be changed to perform virtual
2974 * copying operations, although the data referenced will
2975 * remain the same.
2976 */
2977 int
2978 vm_map_lookup(vm_map_t *var_map, /* IN/OUT */
2979 vm_offset_t vaddr,
2980 vm_prot_t fault_typea,
2981 vm_map_entry_t *out_entry, /* OUT */
2982 vm_object_t *object, /* OUT */
2983 vm_pindex_t *pindex, /* OUT */
2984 vm_prot_t *out_prot, /* OUT */
2985 boolean_t *wired) /* OUT */
2986 {
2987 vm_map_entry_t entry;
2988 vm_map_t map = *var_map;
2989 vm_prot_t prot;
2990 vm_prot_t fault_type = fault_typea;
2991
2992 RetryLookup:;
2993 /*
2994 * Lookup the faulting address.
2995 */
2996
2997 vm_map_lock_read(map);
2998 #define RETURN(why) \
2999 { \
3000 vm_map_unlock_read(map); \
3001 return (why); \
3002 }
3003
3004 /*
3005 * If the map has an interesting hint, try it before calling full
3006 * blown lookup routine.
3007 */
3008 entry = map->root;
3009 *out_entry = entry;
3010 if (entry == NULL ||
3011 (vaddr < entry->start) || (vaddr >= entry->end)) {
3012 /*
3013 * Entry was either not a valid hint, or the vaddr was not
3014 * contained in the entry, so do a full lookup.
3015 */
3016 if (!vm_map_lookup_entry(map, vaddr, out_entry))
3017 RETURN(KERN_INVALID_ADDRESS);
3018
3019 entry = *out_entry;
3020 }
3021
3022 /*
3023 * Handle submaps.
3024 */
3025 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
3026 vm_map_t old_map = map;
3027
3028 *var_map = map = entry->object.sub_map;
3029 vm_map_unlock_read(old_map);
3030 goto RetryLookup;
3031 }
3032
3033 /*
3034 * Check whether this task is allowed to have this page.
3035 * Note the special case for MAP_ENTRY_COW
3036 * pages with an override. This is to implement a forced
3037 * COW for debuggers.
3038 */
3039 if (fault_type & VM_PROT_OVERRIDE_WRITE)
3040 prot = entry->max_protection;
3041 else
3042 prot = entry->protection;
3043 fault_type &= (VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE);
3044 if ((fault_type & prot) != fault_type) {
3045 RETURN(KERN_PROTECTION_FAILURE);
3046 }
3047 if ((entry->eflags & MAP_ENTRY_USER_WIRED) &&
3048 (entry->eflags & MAP_ENTRY_COW) &&
3049 (fault_type & VM_PROT_WRITE) &&
3050 (fault_typea & VM_PROT_OVERRIDE_WRITE) == 0) {
3051 RETURN(KERN_PROTECTION_FAILURE);
3052 }
3053
3054 /*
3055 * If this page is not pageable, we have to get it for all possible
3056 * accesses.
3057 */
3058 *wired = (entry->wired_count != 0);
3059 if (*wired)
3060 prot = fault_type = entry->protection;
3061
3062 /*
3063 * If the entry was copy-on-write, we either ...
3064 */
3065 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
3066 /*
3067 * If we want to write the page, we may as well handle that
3068 * now since we've got the map locked.
3069 *
3070 * If we don't need to write the page, we just demote the
3071 * permissions allowed.
3072 */
3073 if (fault_type & VM_PROT_WRITE) {
3074 /*
3075 * Make a new object, and place it in the object
3076 * chain. Note that no new references have appeared
3077 * -- one just moved from the map to the new
3078 * object.
3079 */
3080 if (vm_map_lock_upgrade(map))
3081 goto RetryLookup;
3082
3083 vm_object_shadow(
3084 &entry->object.vm_object,
3085 &entry->offset,
3086 atop(entry->end - entry->start));
3087 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
3088
3089 vm_map_lock_downgrade(map);
3090 } else {
3091 /*
3092 * We're attempting to read a copy-on-write page --
3093 * don't allow writes.
3094 */
3095 prot &= ~VM_PROT_WRITE;
3096 }
3097 }
3098
3099 /*
3100 * Create an object if necessary.
3101 */
3102 if (entry->object.vm_object == NULL &&
3103 !map->system_map) {
3104 if (vm_map_lock_upgrade(map))
3105 goto RetryLookup;
3106 entry->object.vm_object = vm_object_allocate(OBJT_DEFAULT,
3107 atop(entry->end - entry->start));
3108 entry->offset = 0;
3109 vm_map_lock_downgrade(map);
3110 }
3111
3112 /*
3113 * Return the object/offset from this entry. If the entry was
3114 * copy-on-write or empty, it has been fixed up.
3115 */
3116 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
3117 *object = entry->object.vm_object;
3118
3119 *out_prot = prot;
3120 return (KERN_SUCCESS);
3121
3122 #undef RETURN
3123 }
3124
3125 /*
3126 * vm_map_lookup_locked:
3127 *
3128 * Lookup the faulting address. A version of vm_map_lookup that returns
3129 * KERN_FAILURE instead of blocking on map lock or memory allocation.
3130 */
3131 int
3132 vm_map_lookup_locked(vm_map_t *var_map, /* IN/OUT */
3133 vm_offset_t vaddr,
3134 vm_prot_t fault_typea,
3135 vm_map_entry_t *out_entry, /* OUT */
3136 vm_object_t *object, /* OUT */
3137 vm_pindex_t *pindex, /* OUT */
3138 vm_prot_t *out_prot, /* OUT */
3139 boolean_t *wired) /* OUT */
3140 {
3141 vm_map_entry_t entry;
3142 vm_map_t map = *var_map;
3143 vm_prot_t prot;
3144 vm_prot_t fault_type = fault_typea;
3145
3146 /*
3147 * If the map has an interesting hint, try it before calling full
3148 * blown lookup routine.
3149 */
3150 entry = map->root;
3151 *out_entry = entry;
3152 if (entry == NULL ||
3153 (vaddr < entry->start) || (vaddr >= entry->end)) {
3154 /*
3155 * Entry was either not a valid hint, or the vaddr was not
3156 * contained in the entry, so do a full lookup.
3157 */
3158 if (!vm_map_lookup_entry(map, vaddr, out_entry))
3159 return (KERN_INVALID_ADDRESS);
3160
3161 entry = *out_entry;
3162 }
3163
3164 /*
3165 * Fail if the entry refers to a submap.
3166 */
3167 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP)
3168 return (KERN_FAILURE);
3169
3170 /*
3171 * Check whether this task is allowed to have this page.
3172 * Note the special case for MAP_ENTRY_COW
3173 * pages with an override. This is to implement a forced
3174 * COW for debuggers.
3175 */
3176 if (fault_type & VM_PROT_OVERRIDE_WRITE)
3177 prot = entry->max_protection;
3178 else
3179 prot = entry->protection;
3180 fault_type &= VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE;
3181 if ((fault_type & prot) != fault_type)
3182 return (KERN_PROTECTION_FAILURE);
3183 if ((entry->eflags & MAP_ENTRY_USER_WIRED) &&
3184 (entry->eflags & MAP_ENTRY_COW) &&
3185 (fault_type & VM_PROT_WRITE) &&
3186 (fault_typea & VM_PROT_OVERRIDE_WRITE) == 0)
3187 return (KERN_PROTECTION_FAILURE);
3188
3189 /*
3190 * If this page is not pageable, we have to get it for all possible
3191 * accesses.
3192 */
3193 *wired = (entry->wired_count != 0);
3194 if (*wired)
3195 prot = fault_type = entry->protection;
3196
3197 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
3198 /*
3199 * Fail if the entry was copy-on-write for a write fault.
3200 */
3201 if (fault_type & VM_PROT_WRITE)
3202 return (KERN_FAILURE);
3203 /*
3204 * We're attempting to read a copy-on-write page --
3205 * don't allow writes.
3206 */
3207 prot &= ~VM_PROT_WRITE;
3208 }
3209
3210 /*
3211 * Fail if an object should be created.
3212 */
3213 if (entry->object.vm_object == NULL && !map->system_map)
3214 return (KERN_FAILURE);
3215
3216 /*
3217 * Return the object/offset from this entry. If the entry was
3218 * copy-on-write or empty, it has been fixed up.
3219 */
3220 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
3221 *object = entry->object.vm_object;
3222
3223 *out_prot = prot;
3224 return (KERN_SUCCESS);
3225 }
3226
3227 /*
3228 * vm_map_lookup_done:
3229 *
3230 * Releases locks acquired by a vm_map_lookup
3231 * (according to the handle returned by that lookup).
3232 */
3233 void
3234 vm_map_lookup_done(vm_map_t map, vm_map_entry_t entry)
3235 {
3236 /*
3237 * Unlock the main-level map
3238 */
3239 vm_map_unlock_read(map);
3240 }
3241
3242 #include "opt_ddb.h"
3243 #ifdef DDB
3244 #include <sys/kernel.h>
3245
3246 #include <ddb/ddb.h>
3247
3248 /*
3249 * vm_map_print: [ debug ]
3250 */
3251 DB_SHOW_COMMAND(map, vm_map_print)
3252 {
3253 static int nlines;
3254 /* XXX convert args. */
3255 vm_map_t map = (vm_map_t)addr;
3256 boolean_t full = have_addr;
3257
3258 vm_map_entry_t entry;
3259
3260 db_iprintf("Task map %p: pmap=%p, nentries=%d, version=%u\n",
3261 (void *)map,
3262 (void *)map->pmap, map->nentries, map->timestamp);
3263 nlines++;
3264
3265 if (!full && db_indent)
3266 return;
3267
3268 db_indent += 2;
3269 for (entry = map->header.next; entry != &map->header;
3270 entry = entry->next) {
3271 db_iprintf("map entry %p: start=%p, end=%p\n",
3272 (void *)entry, (void *)entry->start, (void *)entry->end);
3273 nlines++;
3274 {
3275 static char *inheritance_name[4] =
3276 {"share", "copy", "none", "donate_copy"};
3277
3278 db_iprintf(" prot=%x/%x/%s",
3279 entry->protection,
3280 entry->max_protection,
3281 inheritance_name[(int)(unsigned char)entry->inheritance]);
3282 if (entry->wired_count != 0)
3283 db_printf(", wired");
3284 }
3285 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
3286 db_printf(", share=%p, offset=0x%jx\n",
3287 (void *)entry->object.sub_map,
3288 (uintmax_t)entry->offset);
3289 nlines++;
3290 if ((entry->prev == &map->header) ||
3291 (entry->prev->object.sub_map !=
3292 entry->object.sub_map)) {
3293 db_indent += 2;
3294 vm_map_print((db_expr_t)(intptr_t)
3295 entry->object.sub_map,
3296 full, 0, (char *)0);
3297 db_indent -= 2;
3298 }
3299 } else {
3300 db_printf(", object=%p, offset=0x%jx",
3301 (void *)entry->object.vm_object,
3302 (uintmax_t)entry->offset);
3303 if (entry->eflags & MAP_ENTRY_COW)
3304 db_printf(", copy (%s)",
3305 (entry->eflags & MAP_ENTRY_NEEDS_COPY) ? "needed" : "done");
3306 db_printf("\n");
3307 nlines++;
3308
3309 if ((entry->prev == &map->header) ||
3310 (entry->prev->object.vm_object !=
3311 entry->object.vm_object)) {
3312 db_indent += 2;
3313 vm_object_print((db_expr_t)(intptr_t)
3314 entry->object.vm_object,
3315 full, 0, (char *)0);
3316 nlines += 4;
3317 db_indent -= 2;
3318 }
3319 }
3320 }
3321 db_indent -= 2;
3322 if (db_indent == 0)
3323 nlines = 0;
3324 }
3325
3326
3327 DB_SHOW_COMMAND(procvm, procvm)
3328 {
3329 struct proc *p;
3330
3331 if (have_addr) {
3332 p = (struct proc *) addr;
3333 } else {
3334 p = curproc;
3335 }
3336
3337 db_printf("p = %p, vmspace = %p, map = %p, pmap = %p\n",
3338 (void *)p, (void *)p->p_vmspace, (void *)&p->p_vmspace->vm_map,
3339 (void *)vmspace_pmap(p->p_vmspace));
3340
3341 vm_map_print((db_expr_t)(intptr_t)&p->p_vmspace->vm_map, 1, 0, NULL);
3342 }
3343
3344 #endif /* DDB */
Cache object: 49de03da141e3b21b51e9800da630671
|