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 * 3. All advertising materials mentioning features or use of this software
17 * must display the following acknowledgement:
18 * This product includes software developed by the University of
19 * California, Berkeley and its contributors.
20 * 4. Neither the name of the University nor the names of its contributors
21 * may be used to endorse or promote products derived from this software
22 * without specific prior written permission.
23 *
24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * SUCH DAMAGE.
35 *
36 * from: @(#)vm_map.c 8.3 (Berkeley) 1/12/94
37 *
38 *
39 * Copyright (c) 1987, 1990 Carnegie-Mellon University.
40 * All rights reserved.
41 *
42 * Authors: Avadis Tevanian, Jr., Michael Wayne Young
43 *
44 * Permission to use, copy, modify and distribute this software and
45 * its documentation is hereby granted, provided that both the copyright
46 * notice and this permission notice appear in all copies of the
47 * software, derivative works or modified versions, and any portions
48 * thereof, and that both notices appear in supporting documentation.
49 *
50 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
51 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
52 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
53 *
54 * Carnegie Mellon requests users of this software to return to
55 *
56 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
57 * School of Computer Science
58 * Carnegie Mellon University
59 * Pittsburgh PA 15213-3890
60 *
61 * any improvements or extensions that they make and grant Carnegie the
62 * rights to redistribute these changes.
63 *
64 * $FreeBSD$
65 */
66
67 /*
68 * Virtual memory mapping module.
69 */
70
71 #include <sys/param.h>
72 #include <sys/systm.h>
73 #include <sys/kernel.h>
74 #include <sys/sysctl.h>
75 #include <sys/proc.h>
76 #include <sys/vmmeter.h>
77 #include <sys/mman.h>
78 #include <sys/vnode.h>
79 #include <sys/resourcevar.h>
80 #include <sys/file.h>
81
82 #include <vm/vm.h>
83 #include <vm/vm_param.h>
84 #include <sys/lock.h>
85 #include <vm/pmap.h>
86 #include <vm/vm_map.h>
87 #include <vm/vm_page.h>
88 #include <vm/vm_object.h>
89 #include <vm/vm_pager.h>
90 #include <vm/vm_kern.h>
91 #include <vm/vm_extern.h>
92 #include <vm/swap_pager.h>
93 #include <vm/vm_zone.h>
94
95 /*
96 * Virtual memory maps provide for the mapping, protection,
97 * and sharing of virtual memory objects. In addition,
98 * this module provides for an efficient virtual copy of
99 * memory from one map to another.
100 *
101 * Synchronization is required prior to most operations.
102 *
103 * Maps consist of an ordered doubly-linked list of simple
104 * entries; a single hint is used to speed up lookups.
105 *
106 * Since portions of maps are specified by start/end addresses,
107 * which may not align with existing map entries, all
108 * routines merely "clip" entries to these start/end values.
109 * [That is, an entry is split into two, bordering at a
110 * start or end value.] Note that these clippings may not
111 * always be necessary (as the two resulting entries are then
112 * not changed); however, the clipping is done for convenience.
113 *
114 * As mentioned above, virtual copy operations are performed
115 * by copying VM object references from one map to
116 * another, and then marking both regions as copy-on-write.
117 */
118
119 /*
120 * vm_map_startup:
121 *
122 * Initialize the vm_map module. Must be called before
123 * any other vm_map routines.
124 *
125 * Map and entry structures are allocated from the general
126 * purpose memory pool with some exceptions:
127 *
128 * - The kernel map and kmem submap are allocated statically.
129 * - Kernel map entries are allocated out of a static pool.
130 *
131 * These restrictions are necessary since malloc() uses the
132 * maps and requires map entries.
133 */
134
135 static struct vm_zone kmapentzone_store, mapentzone_store, mapzone_store;
136 static vm_zone_t mapentzone, kmapentzone, mapzone, vmspace_zone;
137 static struct vm_object kmapentobj, mapentobj, mapobj;
138
139 static struct vm_map_entry map_entry_init[MAX_MAPENT];
140 static struct vm_map_entry kmap_entry_init[MAX_KMAPENT];
141 static struct vm_map map_init[MAX_KMAP];
142
143 static void _vm_map_clip_end __P((vm_map_t, vm_map_entry_t, vm_offset_t));
144 static void _vm_map_clip_start __P((vm_map_t, vm_map_entry_t, vm_offset_t));
145 static vm_map_entry_t vm_map_entry_create __P((vm_map_t));
146 static void vm_map_entry_delete __P((vm_map_t, vm_map_entry_t));
147 static void vm_map_entry_dispose __P((vm_map_t, vm_map_entry_t));
148 static void vm_map_entry_unwire __P((vm_map_t, vm_map_entry_t));
149 static void vm_map_copy_entry __P((vm_map_t, vm_map_t, vm_map_entry_t,
150 vm_map_entry_t));
151 static void vm_map_split __P((vm_map_entry_t));
152 static void vm_map_unclip_range __P((vm_map_t map, vm_map_entry_t start_entry, vm_offset_t start, vm_offset_t end, int flags));
153
154 static int old_msync;
155 SYSCTL_INT(_vm, OID_AUTO, old_msync, CTLFLAG_RW, &old_msync, 0,
156 "Use old (insecure) msync behavior");
157
158 void
159 vm_map_startup()
160 {
161 mapzone = &mapzone_store;
162 zbootinit(mapzone, "MAP", sizeof (struct vm_map),
163 map_init, MAX_KMAP);
164 kmapentzone = &kmapentzone_store;
165 zbootinit(kmapentzone, "KMAP ENTRY", sizeof (struct vm_map_entry),
166 kmap_entry_init, MAX_KMAPENT);
167 mapentzone = &mapentzone_store;
168 zbootinit(mapentzone, "MAP ENTRY", sizeof (struct vm_map_entry),
169 map_entry_init, MAX_MAPENT);
170 }
171
172 /*
173 * Allocate a vmspace structure, including a vm_map and pmap,
174 * and initialize those structures. The refcnt is set to 1.
175 * The remaining fields must be initialized by the caller.
176 */
177 struct vmspace *
178 vmspace_alloc(min, max)
179 vm_offset_t min, max;
180 {
181 struct vmspace *vm;
182
183 vm = zalloc(vmspace_zone);
184 vm_map_init(&vm->vm_map, min, max);
185 pmap_pinit(vmspace_pmap(vm));
186 vm->vm_map.pmap = vmspace_pmap(vm); /* XXX */
187 vm->vm_refcnt = 1;
188 vm->vm_shm = NULL;
189 vm->vm_exitingcnt = 0;
190 return (vm);
191 }
192
193 void
194 vm_init2(void) {
195 zinitna(kmapentzone, &kmapentobj,
196 NULL, 0, lmin((VM_MAX_KERNEL_ADDRESS - KERNBASE) / PAGE_SIZE,
197 cnt.v_page_count) / 8 + maxproc * 2 + maxfiles,
198 ZONE_INTERRUPT, 1);
199 zinitna(mapentzone, &mapentobj,
200 NULL, 0, 0, 0, 1);
201 zinitna(mapzone, &mapobj,
202 NULL, 0, 0, 0, 1);
203 vmspace_zone = zinit("VMSPACE", sizeof (struct vmspace), 0, 0, 3);
204 pmap_init2();
205 vm_object_init2();
206 }
207
208 static __inline void
209 vmspace_dofree(struct vmspace *vm)
210 {
211 /*
212 * Lock the map, to wait out all other references to it.
213 * Delete all of the mappings and pages they hold, then call
214 * the pmap module to reclaim anything left.
215 */
216 vm_map_lock(&vm->vm_map);
217 (void) vm_map_delete(&vm->vm_map, vm->vm_map.min_offset,
218 vm->vm_map.max_offset);
219 vm_map_unlock(&vm->vm_map);
220
221 pmap_release(vmspace_pmap(vm));
222 zfree(vmspace_zone, vm);
223 }
224
225 void
226 vmspace_free(struct vmspace *vm)
227 {
228 if (vm->vm_refcnt == 0)
229 panic("vmspace_free: attempt to free already freed vmspace");
230
231 if (--vm->vm_refcnt == 0 && vm->vm_exitingcnt == 0)
232 vmspace_dofree(vm);
233 }
234
235 void
236 vmspace_exitfree(struct proc *p)
237 {
238 struct vmspace *vm;
239
240 vm = p->p_vmspace;
241 p->p_vmspace = NULL;
242
243 /*
244 * cleanup by parent process wait()ing on exiting child. vm_refcnt
245 * may not be 0 (e.g. fork() and child exits without exec()ing).
246 * exitingcnt may increment above 0 and drop back down to zero
247 * several times while vm_refcnt is held non-zero. vm_refcnt
248 * may also increment above 0 and drop back down to zero several
249 * times while vm_exitingcnt is held non-zero.
250 *
251 * The last wait on the exiting child's vmspace will clean up
252 * the remainder of the vmspace.
253 */
254 if (--vm->vm_exitingcnt == 0 && vm->vm_refcnt == 0)
255 vmspace_dofree(vm);
256 }
257
258 /*
259 * vmspace_swap_count() - count the approximate swap useage in pages for a
260 * vmspace.
261 *
262 * Swap useage is determined by taking the proportional swap used by
263 * VM objects backing the VM map. To make up for fractional losses,
264 * if the VM object has any swap use at all the associated map entries
265 * count for at least 1 swap page.
266 */
267 int
268 vmspace_swap_count(struct vmspace *vmspace)
269 {
270 vm_map_t map = &vmspace->vm_map;
271 vm_map_entry_t cur;
272 int count = 0;
273
274 for (cur = map->header.next; cur != &map->header; cur = cur->next) {
275 vm_object_t object;
276
277 if ((cur->eflags & MAP_ENTRY_IS_SUB_MAP) == 0 &&
278 (object = cur->object.vm_object) != NULL &&
279 object->type == OBJT_SWAP
280 ) {
281 int n = (cur->end - cur->start) / PAGE_SIZE;
282
283 if (object->un_pager.swp.swp_bcount) {
284 count += object->un_pager.swp.swp_bcount *
285 SWAP_META_PAGES * n / object->size + 1;
286 }
287 }
288 }
289 return(count);
290 }
291
292
293 /*
294 * vm_map_create:
295 *
296 * Creates and returns a new empty VM map with
297 * the given physical map structure, and having
298 * the given lower and upper address bounds.
299 */
300 vm_map_t
301 vm_map_create(pmap, min, max)
302 pmap_t pmap;
303 vm_offset_t min, max;
304 {
305 vm_map_t result;
306
307 result = zalloc(mapzone);
308 vm_map_init(result, min, max);
309 result->pmap = pmap;
310 return (result);
311 }
312
313 /*
314 * Initialize an existing vm_map structure
315 * such as that in the vmspace structure.
316 * The pmap is set elsewhere.
317 */
318 void
319 vm_map_init(map, min, max)
320 struct vm_map *map;
321 vm_offset_t min, max;
322 {
323 map->header.next = map->header.prev = &map->header;
324 map->nentries = 0;
325 map->size = 0;
326 map->system_map = 0;
327 map->infork = 0;
328 map->min_offset = min;
329 map->max_offset = max;
330 map->first_free = &map->header;
331 map->hint = &map->header;
332 map->timestamp = 0;
333 lockinit(&map->lock, PVM, "thrd_sleep", 0, LK_NOPAUSE);
334 }
335
336 /*
337 * vm_map_entry_dispose: [ internal use only ]
338 *
339 * Inverse of vm_map_entry_create.
340 */
341 static void
342 vm_map_entry_dispose(map, entry)
343 vm_map_t map;
344 vm_map_entry_t entry;
345 {
346 if (map->system_map || !mapentzone)
347 zfreei(kmapentzone, entry);
348 else
349 zfree(mapentzone, entry);
350 }
351
352 /*
353 * vm_map_entry_create: [ internal use only ]
354 *
355 * Allocates a VM map entry for insertion.
356 * No entry fields are filled in. This routine is
357 */
358 static vm_map_entry_t
359 vm_map_entry_create(map)
360 vm_map_t map;
361 {
362 vm_map_entry_t new_entry;
363
364 if (map->system_map || !mapentzone)
365 new_entry = zalloci(kmapentzone);
366 else
367 new_entry = zalloc(mapentzone);
368 if (new_entry == NULL)
369 panic("vm_map_entry_create: kernel resources exhausted");
370 return(new_entry);
371 }
372
373 /*
374 * vm_map_entry_{un,}link:
375 *
376 * Insert/remove entries from maps.
377 */
378 static __inline void
379 vm_map_entry_link(vm_map_t map,
380 vm_map_entry_t after_where,
381 vm_map_entry_t entry)
382 {
383 map->nentries++;
384 entry->prev = after_where;
385 entry->next = after_where->next;
386 entry->next->prev = entry;
387 after_where->next = entry;
388 }
389
390 static __inline void
391 vm_map_entry_unlink(vm_map_t map,
392 vm_map_entry_t entry)
393 {
394 vm_map_entry_t prev;
395 vm_map_entry_t next;
396
397 if (entry->eflags & MAP_ENTRY_IN_TRANSITION)
398 panic("vm_map_entry_unlink: attempt to mess with locked entry! %p", entry);
399 prev = entry->prev;
400 next = entry->next;
401 next->prev = prev;
402 prev->next = next;
403 map->nentries--;
404 }
405
406 /*
407 * SAVE_HINT:
408 *
409 * Saves the specified entry as the hint for
410 * future lookups.
411 */
412 #define SAVE_HINT(map,value) \
413 (map)->hint = (value);
414
415 /*
416 * vm_map_lookup_entry: [ internal use only ]
417 *
418 * Finds the map entry containing (or
419 * immediately preceding) the specified address
420 * in the given map; the entry is returned
421 * in the "entry" parameter. The boolean
422 * result indicates whether the address is
423 * actually contained in the map.
424 */
425 boolean_t
426 vm_map_lookup_entry(map, address, entry)
427 vm_map_t map;
428 vm_offset_t address;
429 vm_map_entry_t *entry; /* OUT */
430 {
431 vm_map_entry_t cur;
432 vm_map_entry_t last;
433
434 /*
435 * Start looking either from the head of the list, or from the hint.
436 */
437
438 cur = map->hint;
439
440 if (cur == &map->header)
441 cur = cur->next;
442
443 if (address >= cur->start) {
444 /*
445 * Go from hint to end of list.
446 *
447 * But first, make a quick check to see if we are already looking
448 * at the entry we want (which is usually the case). Note also
449 * that we don't need to save the hint here... it is the same
450 * hint (unless we are at the header, in which case the hint
451 * didn't buy us anything anyway).
452 */
453 last = &map->header;
454 if ((cur != last) && (cur->end > address)) {
455 *entry = cur;
456 return (TRUE);
457 }
458 } else {
459 /*
460 * Go from start to hint, *inclusively*
461 */
462 last = cur->next;
463 cur = map->header.next;
464 }
465
466 /*
467 * Search linearly
468 */
469
470 while (cur != last) {
471 if (cur->end > address) {
472 if (address >= cur->start) {
473 /*
474 * Save this lookup for future hints, and
475 * return
476 */
477
478 *entry = cur;
479 SAVE_HINT(map, cur);
480 return (TRUE);
481 }
482 break;
483 }
484 cur = cur->next;
485 }
486 *entry = cur->prev;
487 SAVE_HINT(map, *entry);
488 return (FALSE);
489 }
490
491 /*
492 * vm_map_insert:
493 *
494 * Inserts the given whole VM object into the target
495 * map at the specified address range. The object's
496 * size should match that of the address range.
497 *
498 * Requires that the map be locked, and leaves it so.
499 *
500 * If object is non-NULL, ref count must be bumped by caller
501 * prior to making call to account for the new entry.
502 */
503 int
504 vm_map_insert(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
505 vm_offset_t start, vm_offset_t end, vm_prot_t prot, vm_prot_t max,
506 int cow)
507 {
508 vm_map_entry_t new_entry;
509 vm_map_entry_t prev_entry;
510 vm_map_entry_t temp_entry;
511 vm_eflags_t protoeflags;
512
513 /*
514 * Check that the start and end points are not bogus.
515 */
516
517 if ((start < map->min_offset) || (end > map->max_offset) ||
518 (start >= end))
519 return (KERN_INVALID_ADDRESS);
520
521 /*
522 * Find the entry prior to the proposed starting address; if it's part
523 * of an existing entry, this range is bogus.
524 */
525
526 if (vm_map_lookup_entry(map, start, &temp_entry))
527 return (KERN_NO_SPACE);
528
529 prev_entry = temp_entry;
530
531 /*
532 * Assert that the next entry doesn't overlap the end point.
533 */
534
535 if ((prev_entry->next != &map->header) &&
536 (prev_entry->next->start < end))
537 return (KERN_NO_SPACE);
538
539 protoeflags = 0;
540
541 if (cow & MAP_COPY_ON_WRITE)
542 protoeflags |= MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY;
543
544 if (cow & MAP_NOFAULT) {
545 protoeflags |= MAP_ENTRY_NOFAULT;
546
547 KASSERT(object == NULL,
548 ("vm_map_insert: paradoxical MAP_NOFAULT request"));
549 }
550 if (cow & MAP_DISABLE_SYNCER)
551 protoeflags |= MAP_ENTRY_NOSYNC;
552 if (cow & MAP_DISABLE_COREDUMP)
553 protoeflags |= MAP_ENTRY_NOCOREDUMP;
554
555 if (object) {
556 /*
557 * When object is non-NULL, it could be shared with another
558 * process. We have to set or clear OBJ_ONEMAPPING
559 * appropriately.
560 */
561 if ((object->ref_count > 1) || (object->shadow_count != 0)) {
562 vm_object_clear_flag(object, OBJ_ONEMAPPING);
563 }
564 }
565 else if ((prev_entry != &map->header) &&
566 (prev_entry->eflags == protoeflags) &&
567 (prev_entry->end == start) &&
568 (prev_entry->wired_count == 0) &&
569 ((prev_entry->object.vm_object == NULL) ||
570 vm_object_coalesce(prev_entry->object.vm_object,
571 OFF_TO_IDX(prev_entry->offset),
572 (vm_size_t)(prev_entry->end - prev_entry->start),
573 (vm_size_t)(end - prev_entry->end)))) {
574 /*
575 * We were able to extend the object. Determine if we
576 * can extend the previous map entry to include the
577 * new range as well.
578 */
579 if ((prev_entry->inheritance == VM_INHERIT_DEFAULT) &&
580 (prev_entry->protection == prot) &&
581 (prev_entry->max_protection == max)) {
582 map->size += (end - prev_entry->end);
583 prev_entry->end = end;
584 vm_map_simplify_entry(map, prev_entry);
585 return (KERN_SUCCESS);
586 }
587
588 /*
589 * If we can extend the object but cannot extend the
590 * map entry, we have to create a new map entry. We
591 * must bump the ref count on the extended object to
592 * account for it. object may be NULL.
593 */
594 object = prev_entry->object.vm_object;
595 offset = prev_entry->offset +
596 (prev_entry->end - prev_entry->start);
597 vm_object_reference(object);
598 }
599
600 /*
601 * NOTE: if conditionals fail, object can be NULL here. This occurs
602 * in things like the buffer map where we manage kva but do not manage
603 * backing objects.
604 */
605
606 /*
607 * Create a new entry
608 */
609
610 new_entry = vm_map_entry_create(map);
611 new_entry->start = start;
612 new_entry->end = end;
613
614 new_entry->eflags = protoeflags;
615 new_entry->object.vm_object = object;
616 new_entry->offset = offset;
617 new_entry->avail_ssize = 0;
618
619 new_entry->inheritance = VM_INHERIT_DEFAULT;
620 new_entry->protection = prot;
621 new_entry->max_protection = max;
622 new_entry->wired_count = 0;
623
624 /*
625 * Insert the new entry into the list
626 */
627
628 vm_map_entry_link(map, prev_entry, new_entry);
629 map->size += new_entry->end - new_entry->start;
630
631 /*
632 * Update the free space hint
633 */
634 if ((map->first_free == prev_entry) &&
635 (prev_entry->end >= new_entry->start)) {
636 map->first_free = new_entry;
637 }
638
639 #if 0
640 /*
641 * Temporarily removed to avoid MAP_STACK panic, due to
642 * MAP_STACK being a huge hack. Will be added back in
643 * when MAP_STACK (and the user stack mapping) is fixed.
644 */
645 /*
646 * It may be possible to simplify the entry
647 */
648 vm_map_simplify_entry(map, new_entry);
649 #endif
650
651 if (cow & (MAP_PREFAULT|MAP_PREFAULT_PARTIAL)) {
652 pmap_object_init_pt(map->pmap, start, prot,
653 object, OFF_TO_IDX(offset), end - start,
654 cow & MAP_PREFAULT_PARTIAL);
655 }
656
657 return (KERN_SUCCESS);
658 }
659
660 /*
661 * Find sufficient space for `length' bytes in the given map, starting at
662 * `start'. The map must be locked. Returns 0 on success, 1 on no space.
663 */
664 int
665 vm_map_findspace(map, start, length, addr)
666 vm_map_t map;
667 vm_offset_t start;
668 vm_size_t length;
669 vm_offset_t *addr;
670 {
671 vm_map_entry_t entry, next;
672 vm_offset_t end;
673
674 if (start < map->min_offset)
675 start = map->min_offset;
676 if (start > map->max_offset)
677 return (1);
678
679 /*
680 * Look for the first possible address; if there's already something
681 * at this address, we have to start after it.
682 */
683 if (start == map->min_offset) {
684 if ((entry = map->first_free) != &map->header)
685 start = entry->end;
686 } else {
687 vm_map_entry_t tmp;
688
689 if (vm_map_lookup_entry(map, start, &tmp))
690 start = tmp->end;
691 entry = tmp;
692 }
693
694 /*
695 * Look through the rest of the map, trying to fit a new region in the
696 * gap between existing regions, or after the very last region.
697 */
698 for (;; start = (entry = next)->end) {
699 /*
700 * Find the end of the proposed new region. Be sure we didn't
701 * go beyond the end of the map, or wrap around the address;
702 * if so, we lose. Otherwise, if this is the last entry, or
703 * if the proposed new region fits before the next entry, we
704 * win.
705 */
706 end = start + length;
707 if (end > map->max_offset || end < start)
708 return (1);
709 next = entry->next;
710 if (next == &map->header || next->start >= end)
711 break;
712 }
713 SAVE_HINT(map, entry);
714 *addr = start;
715 if (map == kernel_map) {
716 vm_offset_t ksize;
717 if ((ksize = round_page(start + length)) > kernel_vm_end) {
718 pmap_growkernel(ksize);
719 }
720 }
721 return (0);
722 }
723
724 /*
725 * vm_map_find finds an unallocated region in the target address
726 * map with the given length. The search is defined to be
727 * first-fit from the specified address; the region found is
728 * returned in the same parameter.
729 *
730 * If object is non-NULL, ref count must be bumped by caller
731 * prior to making call to account for the new entry.
732 */
733 int
734 vm_map_find(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
735 vm_offset_t *addr, /* IN/OUT */
736 vm_size_t length, boolean_t find_space, vm_prot_t prot,
737 vm_prot_t max, int cow)
738 {
739 vm_offset_t start;
740 int result, s = 0;
741
742 start = *addr;
743
744 if (map == kmem_map || map == mb_map)
745 s = splvm();
746
747 vm_map_lock(map);
748 if (find_space) {
749 if (vm_map_findspace(map, start, length, addr)) {
750 vm_map_unlock(map);
751 if (map == kmem_map || map == mb_map)
752 splx(s);
753 return (KERN_NO_SPACE);
754 }
755 start = *addr;
756 }
757 result = vm_map_insert(map, object, offset,
758 start, start + length, prot, max, cow);
759 vm_map_unlock(map);
760
761 if (map == kmem_map || map == mb_map)
762 splx(s);
763
764 return (result);
765 }
766
767 /*
768 * vm_map_simplify_entry:
769 *
770 * Simplify the given map entry by merging with either neighbor. This
771 * routine also has the ability to merge with both neighbors.
772 *
773 * The map must be locked.
774 *
775 * This routine guarentees that the passed entry remains valid (though
776 * possibly extended). When merging, this routine may delete one or
777 * both neighbors. No action is taken on entries which have their
778 * in-transition flag set.
779 */
780 void
781 vm_map_simplify_entry(map, entry)
782 vm_map_t map;
783 vm_map_entry_t entry;
784 {
785 vm_map_entry_t next, prev;
786 vm_size_t prevsize, esize;
787
788 if (entry->eflags & (MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_IS_SUB_MAP)) {
789 ++cnt.v_intrans_coll;
790 return;
791 }
792
793 prev = entry->prev;
794 if (prev != &map->header) {
795 prevsize = prev->end - prev->start;
796 if ( (prev->end == entry->start) &&
797 (prev->object.vm_object == entry->object.vm_object) &&
798 (!prev->object.vm_object ||
799 (prev->offset + prevsize == entry->offset)) &&
800 (prev->eflags == entry->eflags) &&
801 (prev->protection == entry->protection) &&
802 (prev->max_protection == entry->max_protection) &&
803 (prev->inheritance == entry->inheritance) &&
804 (prev->wired_count == entry->wired_count)) {
805 if (map->first_free == prev)
806 map->first_free = entry;
807 if (map->hint == prev)
808 map->hint = entry;
809 vm_map_entry_unlink(map, prev);
810 entry->start = prev->start;
811 entry->offset = prev->offset;
812 if (prev->object.vm_object)
813 vm_object_deallocate(prev->object.vm_object);
814 vm_map_entry_dispose(map, prev);
815 }
816 }
817
818 next = entry->next;
819 if (next != &map->header) {
820 esize = entry->end - entry->start;
821 if ((entry->end == next->start) &&
822 (next->object.vm_object == entry->object.vm_object) &&
823 (!entry->object.vm_object ||
824 (entry->offset + esize == next->offset)) &&
825 (next->eflags == entry->eflags) &&
826 (next->protection == entry->protection) &&
827 (next->max_protection == entry->max_protection) &&
828 (next->inheritance == entry->inheritance) &&
829 (next->wired_count == entry->wired_count)) {
830 if (map->first_free == next)
831 map->first_free = entry;
832 if (map->hint == next)
833 map->hint = entry;
834 vm_map_entry_unlink(map, next);
835 entry->end = next->end;
836 if (next->object.vm_object)
837 vm_object_deallocate(next->object.vm_object);
838 vm_map_entry_dispose(map, next);
839 }
840 }
841 }
842 /*
843 * vm_map_clip_start: [ internal use only ]
844 *
845 * Asserts that the given entry begins at or after
846 * the specified address; if necessary,
847 * it splits the entry into two.
848 */
849 #define vm_map_clip_start(map, entry, startaddr) \
850 { \
851 if (startaddr > entry->start) \
852 _vm_map_clip_start(map, entry, startaddr); \
853 }
854
855 /*
856 * This routine is called only when it is known that
857 * the entry must be split.
858 */
859 static void
860 _vm_map_clip_start(map, entry, start)
861 vm_map_t map;
862 vm_map_entry_t entry;
863 vm_offset_t start;
864 {
865 vm_map_entry_t new_entry;
866
867 /*
868 * Split off the front portion -- note that we must insert the new
869 * entry BEFORE this one, so that this entry has the specified
870 * starting address.
871 */
872
873 vm_map_simplify_entry(map, entry);
874
875 /*
876 * If there is no object backing this entry, we might as well create
877 * one now. If we defer it, an object can get created after the map
878 * is clipped, and individual objects will be created for the split-up
879 * map. This is a bit of a hack, but is also about the best place to
880 * put this improvement.
881 */
882
883 if (entry->object.vm_object == NULL && !map->system_map) {
884 vm_object_t object;
885 object = vm_object_allocate(OBJT_DEFAULT,
886 atop(entry->end - entry->start));
887 entry->object.vm_object = object;
888 entry->offset = 0;
889 }
890
891 new_entry = vm_map_entry_create(map);
892 *new_entry = *entry;
893
894 new_entry->end = start;
895 entry->offset += (start - entry->start);
896 entry->start = start;
897
898 vm_map_entry_link(map, entry->prev, new_entry);
899
900 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
901 vm_object_reference(new_entry->object.vm_object);
902 }
903 }
904
905 /*
906 * vm_map_clip_end: [ internal use only ]
907 *
908 * Asserts that the given entry ends at or before
909 * the specified address; if necessary,
910 * it splits the entry into two.
911 */
912
913 #define vm_map_clip_end(map, entry, endaddr) \
914 { \
915 if (endaddr < entry->end) \
916 _vm_map_clip_end(map, entry, endaddr); \
917 }
918
919 /*
920 * This routine is called only when it is known that
921 * the entry must be split.
922 */
923 static void
924 _vm_map_clip_end(map, entry, end)
925 vm_map_t map;
926 vm_map_entry_t entry;
927 vm_offset_t end;
928 {
929 vm_map_entry_t new_entry;
930
931 /*
932 * If there is no object backing this entry, we might as well create
933 * one now. If we defer it, an object can get created after the map
934 * is clipped, and individual objects will be created for the split-up
935 * map. This is a bit of a hack, but is also about the best place to
936 * put this improvement.
937 */
938
939 if (entry->object.vm_object == NULL && !map->system_map) {
940 vm_object_t object;
941 object = vm_object_allocate(OBJT_DEFAULT,
942 atop(entry->end - entry->start));
943 entry->object.vm_object = object;
944 entry->offset = 0;
945 }
946
947 /*
948 * Create a new entry and insert it AFTER the specified entry
949 */
950
951 new_entry = vm_map_entry_create(map);
952 *new_entry = *entry;
953
954 new_entry->start = entry->end = end;
955 new_entry->offset += (end - entry->start);
956
957 vm_map_entry_link(map, entry, new_entry);
958
959 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
960 vm_object_reference(new_entry->object.vm_object);
961 }
962 }
963
964 /*
965 * VM_MAP_RANGE_CHECK: [ internal use only ]
966 *
967 * Asserts that the starting and ending region
968 * addresses fall within the valid range of the map.
969 */
970 #define VM_MAP_RANGE_CHECK(map, start, end) \
971 { \
972 if (start < vm_map_min(map)) \
973 start = vm_map_min(map); \
974 if (end > vm_map_max(map)) \
975 end = vm_map_max(map); \
976 if (start > end) \
977 start = end; \
978 }
979
980 /*
981 * vm_map_transition_wait: [ kernel use only ]
982 *
983 * Used to block when an in-transition collison occurs. The map
984 * is unlocked for the sleep and relocked before the return.
985 */
986 static
987 void
988 vm_map_transition_wait(vm_map_t map)
989 {
990 vm_map_unlock(map);
991 tsleep(map, PVM, "vment", 0);
992 vm_map_lock(map);
993 }
994
995 /*
996 * CLIP_CHECK_BACK
997 * CLIP_CHECK_FWD
998 *
999 * When we do blocking operations with the map lock held it is
1000 * possible that a clip might have occured on our in-transit entry,
1001 * requiring an adjustment to the entry in our loop. These macros
1002 * help the pageable and clip_range code deal with the case. The
1003 * conditional costs virtually nothing if no clipping has occured.
1004 */
1005
1006 #define CLIP_CHECK_BACK(entry, save_start) \
1007 do { \
1008 while (entry->start != save_start) { \
1009 entry = entry->prev; \
1010 KASSERT(entry != &map->header, ("bad entry clip")); \
1011 } \
1012 } while(0)
1013
1014 #define CLIP_CHECK_FWD(entry, save_end) \
1015 do { \
1016 while (entry->end != save_end) { \
1017 entry = entry->next; \
1018 KASSERT(entry != &map->header, ("bad entry clip")); \
1019 } \
1020 } while(0)
1021
1022
1023 /*
1024 * vm_map_clip_range: [ kernel use only ]
1025 *
1026 * Clip the specified range and return the base entry. The
1027 * range may cover several entries starting at the returned base
1028 * and the first and last entry in the covering sequence will be
1029 * properly clipped to the requested start and end address.
1030 *
1031 * If no holes are allowed you should pass the MAP_CLIP_NO_HOLES
1032 * flag.
1033 *
1034 * The MAP_ENTRY_IN_TRANSITION flag will be set for the entries
1035 * covered by the requested range.
1036 *
1037 * The map must be exclusively locked on entry and will remain locked
1038 * on return. If no range exists or the range contains holes and you
1039 * specified that no holes were allowed, NULL will be returned. This
1040 * routine may temporarily unlock the map in order avoid a deadlock when
1041 * sleeping.
1042 */
1043 static
1044 vm_map_entry_t
1045 vm_map_clip_range(vm_map_t map, vm_offset_t start, vm_offset_t end, int flags)
1046 {
1047 vm_map_entry_t start_entry;
1048 vm_map_entry_t entry;
1049
1050 /*
1051 * Locate the entry and effect initial clipping. The in-transition
1052 * case does not occur very often so do not try to optimize it.
1053 */
1054 again:
1055 if (vm_map_lookup_entry(map, start, &start_entry) == FALSE)
1056 return (NULL);
1057 entry = start_entry;
1058 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
1059 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
1060 ++cnt.v_intrans_coll;
1061 ++cnt.v_intrans_wait;
1062 vm_map_transition_wait(map);
1063 /*
1064 * entry and/or start_entry may have been clipped while
1065 * we slept, or may have gone away entirely. We have
1066 * to restart from the lookup.
1067 */
1068 goto again;
1069 }
1070 /*
1071 * Since we hold an exclusive map lock we do not have to restart
1072 * after clipping, even though clipping may block in zalloc.
1073 */
1074 vm_map_clip_start(map, entry, start);
1075 vm_map_clip_end(map, entry, end);
1076 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
1077
1078 /*
1079 * Scan entries covered by the range. When working on the next
1080 * entry a restart need only re-loop on the current entry which
1081 * we have already locked, since 'next' may have changed. Also,
1082 * even though entry is safe, it may have been clipped so we
1083 * have to iterate forwards through the clip after sleeping.
1084 */
1085 while (entry->next != &map->header && entry->next->start < end) {
1086 vm_map_entry_t next = entry->next;
1087
1088 if (flags & MAP_CLIP_NO_HOLES) {
1089 if (next->start > entry->end) {
1090 vm_map_unclip_range(map, start_entry,
1091 start, entry->end, flags);
1092 return(NULL);
1093 }
1094 }
1095
1096 if (next->eflags & MAP_ENTRY_IN_TRANSITION) {
1097 vm_offset_t save_end = entry->end;
1098 next->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
1099 ++cnt.v_intrans_coll;
1100 ++cnt.v_intrans_wait;
1101 vm_map_transition_wait(map);
1102
1103 /*
1104 * clips might have occured while we blocked.
1105 */
1106 CLIP_CHECK_FWD(entry, save_end);
1107 CLIP_CHECK_BACK(start_entry, start);
1108 continue;
1109 }
1110 /*
1111 * No restart necessary even though clip_end may block, we
1112 * are holding the map lock.
1113 */
1114 vm_map_clip_end(map, next, end);
1115 next->eflags |= MAP_ENTRY_IN_TRANSITION;
1116 entry = next;
1117 }
1118 if (flags & MAP_CLIP_NO_HOLES) {
1119 if (entry->end != end) {
1120 vm_map_unclip_range(map, start_entry,
1121 start, entry->end, flags);
1122 return(NULL);
1123 }
1124 }
1125 return(start_entry);
1126 }
1127
1128 /*
1129 * vm_map_unclip_range: [ kernel use only ]
1130 *
1131 * Undo the effect of vm_map_clip_range(). You should pass the same
1132 * flags and the same range that you passed to vm_map_clip_range().
1133 * This code will clear the in-transition flag on the entries and
1134 * wake up anyone waiting. This code will also simplify the sequence
1135 * and attempt to merge it with entries before and after the sequence.
1136 *
1137 * The map must be locked on entry and will remain locked on return.
1138 *
1139 * Note that you should also pass the start_entry returned by
1140 * vm_map_clip_range(). However, if you block between the two calls
1141 * with the map unlocked please be aware that the start_entry may
1142 * have been clipped and you may need to scan it backwards to find
1143 * the entry corresponding with the original start address. You are
1144 * responsible for this, vm_map_unclip_range() expects the correct
1145 * start_entry to be passed to it and will KASSERT otherwise.
1146 */
1147 static
1148 void
1149 vm_map_unclip_range(
1150 vm_map_t map,
1151 vm_map_entry_t start_entry,
1152 vm_offset_t start,
1153 vm_offset_t end,
1154 int flags)
1155 {
1156 vm_map_entry_t entry;
1157
1158 entry = start_entry;
1159
1160 KASSERT(entry->start == start, ("unclip_range: illegal base entry"));
1161 while (entry != &map->header && entry->start < end) {
1162 KASSERT(entry->eflags & MAP_ENTRY_IN_TRANSITION, ("in-transition flag not set during unclip on: %p", entry));
1163 KASSERT(entry->end <= end, ("unclip_range: tail wasn't clipped"));
1164 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION;
1165 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
1166 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
1167 wakeup(map);
1168 }
1169 entry = entry->next;
1170 }
1171
1172 /*
1173 * Simplification does not block so there is no restart case.
1174 */
1175 entry = start_entry;
1176 while (entry != &map->header && entry->start < end) {
1177 vm_map_simplify_entry(map, entry);
1178 entry = entry->next;
1179 }
1180 }
1181
1182 /*
1183 * vm_map_submap: [ kernel use only ]
1184 *
1185 * Mark the given range as handled by a subordinate map.
1186 *
1187 * This range must have been created with vm_map_find,
1188 * and no other operations may have been performed on this
1189 * range prior to calling vm_map_submap.
1190 *
1191 * Only a limited number of operations can be performed
1192 * within this rage after calling vm_map_submap:
1193 * vm_fault
1194 * [Don't try vm_map_copy!]
1195 *
1196 * To remove a submapping, one must first remove the
1197 * range from the superior map, and then destroy the
1198 * submap (if desired). [Better yet, don't try it.]
1199 */
1200 int
1201 vm_map_submap(map, start, end, submap)
1202 vm_map_t map;
1203 vm_offset_t start;
1204 vm_offset_t end;
1205 vm_map_t submap;
1206 {
1207 vm_map_entry_t entry;
1208 int result = KERN_INVALID_ARGUMENT;
1209
1210 vm_map_lock(map);
1211
1212 VM_MAP_RANGE_CHECK(map, start, end);
1213
1214 if (vm_map_lookup_entry(map, start, &entry)) {
1215 vm_map_clip_start(map, entry, start);
1216 } else {
1217 entry = entry->next;
1218 }
1219
1220 vm_map_clip_end(map, entry, end);
1221
1222 if ((entry->start == start) && (entry->end == end) &&
1223 ((entry->eflags & MAP_ENTRY_COW) == 0) &&
1224 (entry->object.vm_object == NULL)) {
1225 entry->object.sub_map = submap;
1226 entry->eflags |= MAP_ENTRY_IS_SUB_MAP;
1227 result = KERN_SUCCESS;
1228 }
1229 vm_map_unlock(map);
1230
1231 return (result);
1232 }
1233
1234 /*
1235 * vm_map_protect:
1236 *
1237 * Sets the protection of the specified address
1238 * region in the target map. If "set_max" is
1239 * specified, the maximum protection is to be set;
1240 * otherwise, only the current protection is affected.
1241 */
1242 int
1243 vm_map_protect(vm_map_t map, vm_offset_t start, vm_offset_t end,
1244 vm_prot_t new_prot, boolean_t set_max)
1245 {
1246 vm_map_entry_t current;
1247 vm_map_entry_t entry;
1248
1249 vm_map_lock(map);
1250
1251 VM_MAP_RANGE_CHECK(map, start, end);
1252
1253 if (vm_map_lookup_entry(map, start, &entry)) {
1254 vm_map_clip_start(map, entry, start);
1255 } else {
1256 entry = entry->next;
1257 }
1258
1259 /*
1260 * Make a first pass to check for protection violations.
1261 */
1262
1263 current = entry;
1264 while ((current != &map->header) && (current->start < end)) {
1265 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
1266 vm_map_unlock(map);
1267 return (KERN_INVALID_ARGUMENT);
1268 }
1269 if ((new_prot & current->max_protection) != new_prot) {
1270 vm_map_unlock(map);
1271 return (KERN_PROTECTION_FAILURE);
1272 }
1273 current = current->next;
1274 }
1275
1276 /*
1277 * Go back and fix up protections. [Note that clipping is not
1278 * necessary the second time.]
1279 */
1280
1281 current = entry;
1282
1283 while ((current != &map->header) && (current->start < end)) {
1284 vm_prot_t old_prot;
1285
1286 vm_map_clip_end(map, current, end);
1287
1288 old_prot = current->protection;
1289 if (set_max)
1290 current->protection =
1291 (current->max_protection = new_prot) &
1292 old_prot;
1293 else
1294 current->protection = new_prot;
1295
1296 /*
1297 * Update physical map if necessary. Worry about copy-on-write
1298 * here -- CHECK THIS XXX
1299 */
1300
1301 if (current->protection != old_prot) {
1302 #define MASK(entry) (((entry)->eflags & MAP_ENTRY_COW) ? ~VM_PROT_WRITE : \
1303 VM_PROT_ALL)
1304
1305 pmap_protect(map->pmap, current->start,
1306 current->end,
1307 current->protection & MASK(current));
1308 #undef MASK
1309 }
1310
1311 vm_map_simplify_entry(map, current);
1312
1313 current = current->next;
1314 }
1315
1316 vm_map_unlock(map);
1317 return (KERN_SUCCESS);
1318 }
1319
1320 /*
1321 * vm_map_madvise:
1322 *
1323 * This routine traverses a processes map handling the madvise
1324 * system call. Advisories are classified as either those effecting
1325 * the vm_map_entry structure, or those effecting the underlying
1326 * objects.
1327 */
1328
1329 int
1330 vm_map_madvise(map, start, end, behav)
1331 vm_map_t map;
1332 vm_offset_t start, end;
1333 int behav;
1334 {
1335 vm_map_entry_t current, entry;
1336 int modify_map = 0;
1337
1338 /*
1339 * Some madvise calls directly modify the vm_map_entry, in which case
1340 * we need to use an exclusive lock on the map and we need to perform
1341 * various clipping operations. Otherwise we only need a read-lock
1342 * on the map.
1343 */
1344
1345 switch(behav) {
1346 case MADV_NORMAL:
1347 case MADV_SEQUENTIAL:
1348 case MADV_RANDOM:
1349 case MADV_NOSYNC:
1350 case MADV_AUTOSYNC:
1351 case MADV_NOCORE:
1352 case MADV_CORE:
1353 modify_map = 1;
1354 vm_map_lock(map);
1355 break;
1356 case MADV_WILLNEED:
1357 case MADV_DONTNEED:
1358 case MADV_FREE:
1359 vm_map_lock_read(map);
1360 break;
1361 default:
1362 return (KERN_INVALID_ARGUMENT);
1363 }
1364
1365 /*
1366 * Locate starting entry and clip if necessary.
1367 */
1368
1369 VM_MAP_RANGE_CHECK(map, start, end);
1370
1371 if (vm_map_lookup_entry(map, start, &entry)) {
1372 if (modify_map)
1373 vm_map_clip_start(map, entry, start);
1374 } else {
1375 entry = entry->next;
1376 }
1377
1378 if (modify_map) {
1379 /*
1380 * madvise behaviors that are implemented in the vm_map_entry.
1381 *
1382 * We clip the vm_map_entry so that behavioral changes are
1383 * limited to the specified address range.
1384 */
1385 for (current = entry;
1386 (current != &map->header) && (current->start < end);
1387 current = current->next
1388 ) {
1389 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
1390 continue;
1391
1392 vm_map_clip_end(map, current, end);
1393
1394 switch (behav) {
1395 case MADV_NORMAL:
1396 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_NORMAL);
1397 break;
1398 case MADV_SEQUENTIAL:
1399 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_SEQUENTIAL);
1400 break;
1401 case MADV_RANDOM:
1402 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_RANDOM);
1403 break;
1404 case MADV_NOSYNC:
1405 current->eflags |= MAP_ENTRY_NOSYNC;
1406 break;
1407 case MADV_AUTOSYNC:
1408 current->eflags &= ~MAP_ENTRY_NOSYNC;
1409 break;
1410 case MADV_NOCORE:
1411 current->eflags |= MAP_ENTRY_NOCOREDUMP;
1412 break;
1413 case MADV_CORE:
1414 current->eflags &= ~MAP_ENTRY_NOCOREDUMP;
1415 break;
1416 default:
1417 break;
1418 }
1419 vm_map_simplify_entry(map, current);
1420 }
1421 vm_map_unlock(map);
1422 } else {
1423 vm_pindex_t pindex;
1424 int count;
1425
1426 /*
1427 * madvise behaviors that are implemented in the underlying
1428 * vm_object.
1429 *
1430 * Since we don't clip the vm_map_entry, we have to clip
1431 * the vm_object pindex and count.
1432 */
1433 for (current = entry;
1434 (current != &map->header) && (current->start < end);
1435 current = current->next
1436 ) {
1437 vm_offset_t useStart;
1438
1439 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
1440 continue;
1441
1442 pindex = OFF_TO_IDX(current->offset);
1443 count = atop(current->end - current->start);
1444 useStart = current->start;
1445
1446 if (current->start < start) {
1447 pindex += atop(start - current->start);
1448 count -= atop(start - current->start);
1449 useStart = start;
1450 }
1451 if (current->end > end)
1452 count -= atop(current->end - end);
1453
1454 if (count <= 0)
1455 continue;
1456
1457 vm_object_madvise(current->object.vm_object,
1458 pindex, count, behav);
1459 if (behav == MADV_WILLNEED) {
1460 pmap_object_init_pt(
1461 map->pmap,
1462 useStart,
1463 current->protection,
1464 current->object.vm_object,
1465 pindex,
1466 (count << PAGE_SHIFT),
1467 MAP_PREFAULT_MADVISE
1468 );
1469 }
1470 }
1471 vm_map_unlock_read(map);
1472 }
1473 return(0);
1474 }
1475
1476
1477 /*
1478 * vm_map_inherit:
1479 *
1480 * Sets the inheritance of the specified address
1481 * range in the target map. Inheritance
1482 * affects how the map will be shared with
1483 * child maps at the time of vm_map_fork.
1484 */
1485 int
1486 vm_map_inherit(vm_map_t map, vm_offset_t start, vm_offset_t end,
1487 vm_inherit_t new_inheritance)
1488 {
1489 vm_map_entry_t entry;
1490 vm_map_entry_t temp_entry;
1491
1492 switch (new_inheritance) {
1493 case VM_INHERIT_NONE:
1494 case VM_INHERIT_COPY:
1495 case VM_INHERIT_SHARE:
1496 break;
1497 default:
1498 return (KERN_INVALID_ARGUMENT);
1499 }
1500
1501 vm_map_lock(map);
1502
1503 VM_MAP_RANGE_CHECK(map, start, end);
1504
1505 if (vm_map_lookup_entry(map, start, &temp_entry)) {
1506 entry = temp_entry;
1507 vm_map_clip_start(map, entry, start);
1508 } else
1509 entry = temp_entry->next;
1510
1511 while ((entry != &map->header) && (entry->start < end)) {
1512 vm_map_clip_end(map, entry, end);
1513
1514 entry->inheritance = new_inheritance;
1515
1516 vm_map_simplify_entry(map, entry);
1517
1518 entry = entry->next;
1519 }
1520
1521 vm_map_unlock(map);
1522 return (KERN_SUCCESS);
1523 }
1524
1525 /*
1526 * Implement the semantics of mlock
1527 */
1528 int
1529 vm_map_user_pageable(map, start, real_end, new_pageable)
1530 vm_map_t map;
1531 vm_offset_t start;
1532 vm_offset_t real_end;
1533 boolean_t new_pageable;
1534 {
1535 vm_map_entry_t entry;
1536 vm_map_entry_t start_entry;
1537 vm_offset_t end;
1538 boolean_t fictitious;
1539 int rv = KERN_SUCCESS;
1540
1541 vm_map_lock(map);
1542 VM_MAP_RANGE_CHECK(map, start, real_end);
1543 end = real_end;
1544
1545 start_entry = vm_map_clip_range(map, start, end, MAP_CLIP_NO_HOLES);
1546 if (start_entry == NULL) {
1547 vm_map_unlock(map);
1548 return (KERN_INVALID_ADDRESS);
1549 }
1550
1551 if (new_pageable == 0) {
1552 entry = start_entry;
1553 while ((entry != &map->header) && (entry->start < end)) {
1554 vm_offset_t save_start;
1555 vm_offset_t save_end;
1556
1557 /*
1558 * Already user wired or hard wired (trivial cases)
1559 */
1560 if (entry->eflags & MAP_ENTRY_USER_WIRED) {
1561 entry = entry->next;
1562 continue;
1563 }
1564 if (entry->wired_count != 0) {
1565 entry->wired_count++;
1566 entry->eflags |= MAP_ENTRY_USER_WIRED;
1567 entry = entry->next;
1568 continue;
1569 }
1570
1571 /*
1572 * A new wiring requires instantiation of appropriate
1573 * management structures and the faulting in of the
1574 * page.
1575 */
1576 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1577 int copyflag = entry->eflags & MAP_ENTRY_NEEDS_COPY;
1578 if (copyflag && ((entry->protection & VM_PROT_WRITE) != 0)) {
1579
1580 vm_object_shadow(&entry->object.vm_object,
1581 &entry->offset,
1582 atop(entry->end - entry->start));
1583 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
1584
1585 } else if (entry->object.vm_object == NULL &&
1586 !map->system_map) {
1587
1588 entry->object.vm_object =
1589 vm_object_allocate(OBJT_DEFAULT,
1590 atop(entry->end - entry->start));
1591 entry->offset = (vm_offset_t) 0;
1592
1593 }
1594 }
1595 entry->wired_count++;
1596 entry->eflags |= MAP_ENTRY_USER_WIRED;
1597
1598 /*
1599 * Now fault in the area. The map lock needs to be
1600 * manipulated to avoid deadlocks. The in-transition
1601 * flag protects the entries.
1602 */
1603 fictitious = entry->object.vm_object != NULL &&
1604 entry->object.vm_object->type == OBJT_DEVICE;
1605 save_start = entry->start;
1606 save_end = entry->end;
1607 vm_map_unlock(map);
1608 map->timestamp++;
1609 rv = vm_fault_wire(map, save_start, save_end, 1, fictitious);
1610 vm_map_lock(map);
1611 if (rv) {
1612 CLIP_CHECK_BACK(entry, save_start);
1613 for (;;) {
1614 KASSERT(entry->wired_count == 1, ("bad wired_count on entry"));
1615 entry->eflags &= ~MAP_ENTRY_USER_WIRED;
1616 entry->wired_count = 0;
1617 if (entry->end == save_end)
1618 break;
1619 entry = entry->next;
1620 KASSERT(entry != &map->header, ("bad entry clip during backout"));
1621 }
1622 end = save_start; /* unwire the rest */
1623 break;
1624 }
1625 /*
1626 * note that even though the entry might have been
1627 * clipped, the USER_WIRED flag we set prevents
1628 * duplication so we do not have to do a
1629 * clip check.
1630 */
1631 entry = entry->next;
1632 }
1633
1634 /*
1635 * If we failed fall through to the unwiring section to
1636 * unwire what we had wired so far. 'end' has already
1637 * been adjusted.
1638 */
1639 if (rv)
1640 new_pageable = 1;
1641
1642 /*
1643 * start_entry might have been clipped if we unlocked the
1644 * map and blocked. No matter how clipped it has gotten
1645 * there should be a fragment that is on our start boundary.
1646 */
1647 CLIP_CHECK_BACK(start_entry, start);
1648 }
1649
1650 /*
1651 * Deal with the unwiring case.
1652 */
1653 if (new_pageable) {
1654 /*
1655 * This is the unwiring case. We must first ensure that the
1656 * range to be unwired is really wired down. We know there
1657 * are no holes.
1658 */
1659 entry = start_entry;
1660 while ((entry != &map->header) && (entry->start < end)) {
1661 if ((entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
1662 rv = KERN_INVALID_ARGUMENT;
1663 goto done;
1664 }
1665 KASSERT(entry->wired_count != 0, ("wired count was 0 with USER_WIRED set! %p", entry));
1666 entry = entry->next;
1667 }
1668
1669 /*
1670 * Now decrement the wiring count for each region. If a region
1671 * becomes completely unwired, unwire its physical pages and
1672 * mappings.
1673 */
1674 entry = start_entry;
1675 while ((entry != &map->header) && (entry->start < end)) {
1676 KASSERT(entry->eflags & MAP_ENTRY_USER_WIRED, ("expected USER_WIRED on entry %p", entry));
1677 entry->eflags &= ~MAP_ENTRY_USER_WIRED;
1678 entry->wired_count--;
1679 if (entry->wired_count == 0)
1680 vm_fault_unwire(map, entry->start, entry->end,
1681 entry->object.vm_object != NULL &&
1682 entry->object.vm_object->type == OBJT_DEVICE);
1683 entry = entry->next;
1684 }
1685 }
1686 done:
1687 vm_map_unclip_range(map, start_entry, start, real_end,
1688 MAP_CLIP_NO_HOLES);
1689 map->timestamp++;
1690 vm_map_unlock(map);
1691 return (rv);
1692 }
1693
1694 /*
1695 * vm_map_pageable:
1696 *
1697 * Sets the pageability of the specified address
1698 * range in the target map. Regions specified
1699 * as not pageable require locked-down physical
1700 * memory and physical page maps.
1701 *
1702 * The map must not be locked, but a reference
1703 * must remain to the map throughout the call.
1704 */
1705 int
1706 vm_map_pageable(map, start, real_end, new_pageable)
1707 vm_map_t map;
1708 vm_offset_t start;
1709 vm_offset_t real_end;
1710 boolean_t new_pageable;
1711 {
1712 vm_map_entry_t entry;
1713 vm_map_entry_t start_entry;
1714 vm_offset_t end;
1715 boolean_t fictitious;
1716 int rv = KERN_SUCCESS;
1717 int s;
1718
1719 vm_map_lock(map);
1720 VM_MAP_RANGE_CHECK(map, start, real_end);
1721 end = real_end;
1722
1723 start_entry = vm_map_clip_range(map, start, end, MAP_CLIP_NO_HOLES);
1724 if (start_entry == NULL) {
1725 vm_map_unlock(map);
1726 return (KERN_INVALID_ADDRESS);
1727 }
1728 if (new_pageable == 0) {
1729 /*
1730 * Wiring.
1731 *
1732 * 1. Holding the write lock, we create any shadow or zero-fill
1733 * objects that need to be created. Then we clip each map
1734 * entry to the region to be wired and increment its wiring
1735 * count. We create objects before clipping the map entries
1736 * to avoid object proliferation.
1737 *
1738 * 2. We downgrade to a read lock, and call vm_fault_wire to
1739 * fault in the pages for any newly wired area (wired_count is
1740 * 1).
1741 *
1742 * Downgrading to a read lock for vm_fault_wire avoids a
1743 * possible deadlock with another process that may have faulted
1744 * on one of the pages to be wired (it would mark the page busy,
1745 * blocking us, then in turn block on the map lock that we
1746 * hold). Because of problems in the recursive lock package,
1747 * we cannot upgrade to a write lock in vm_map_lookup. Thus,
1748 * any actions that require the write lock must be done
1749 * beforehand. Because we keep the read lock on the map, the
1750 * copy-on-write status of the entries we modify here cannot
1751 * change.
1752 */
1753
1754 entry = start_entry;
1755 while ((entry != &map->header) && (entry->start < end)) {
1756 /*
1757 * Trivial case if the entry is already wired
1758 */
1759 if (entry->wired_count) {
1760 entry->wired_count++;
1761 entry = entry->next;
1762 continue;
1763 }
1764
1765 /*
1766 * The entry is being newly wired, we have to setup
1767 * appropriate management structures. A shadow
1768 * object is required for a copy-on-write region,
1769 * or a normal object for a zero-fill region. We
1770 * do not have to do this for entries that point to sub
1771 * maps because we won't hold the lock on the sub map.
1772 */
1773 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1774 int copyflag = entry->eflags & MAP_ENTRY_NEEDS_COPY;
1775 if (copyflag &&
1776 ((entry->protection & VM_PROT_WRITE) != 0)) {
1777
1778 vm_object_shadow(&entry->object.vm_object,
1779 &entry->offset,
1780 atop(entry->end - entry->start));
1781 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
1782 } else if (entry->object.vm_object == NULL &&
1783 !map->system_map) {
1784 entry->object.vm_object =
1785 vm_object_allocate(OBJT_DEFAULT,
1786 atop(entry->end - entry->start));
1787 entry->offset = (vm_offset_t) 0;
1788 }
1789 }
1790
1791 entry->wired_count++;
1792 entry = entry->next;
1793 }
1794
1795 /*
1796 * Pass 2.
1797 */
1798
1799 /*
1800 * HACK HACK HACK HACK
1801 *
1802 * Unlock the map to avoid deadlocks. The in-transit flag
1803 * protects us from most changes but note that
1804 * clipping may still occur. To prevent clipping from
1805 * occuring after the unlock, except for when we are
1806 * blocking in vm_fault_wire, we must run at splvm().
1807 * Otherwise our accesses to entry->start and entry->end
1808 * could be corrupted. We have to set splvm() prior to
1809 * unlocking so start_entry does not change out from
1810 * under us at the very beginning of the loop.
1811 *
1812 * HACK HACK HACK HACK
1813 */
1814
1815 s = splvm();
1816 vm_map_unlock(map);
1817
1818 entry = start_entry;
1819 while (entry != &map->header && entry->start < end) {
1820 /*
1821 * If vm_fault_wire fails for any page we need to undo
1822 * what has been done. We decrement the wiring count
1823 * for those pages which have not yet been wired (now)
1824 * and unwire those that have (later).
1825 */
1826 vm_offset_t save_start = entry->start;
1827 vm_offset_t save_end = entry->end;
1828
1829 fictitious = entry->object.vm_object != NULL &&
1830 entry->object.vm_object->type == OBJT_DEVICE;
1831 if (entry->wired_count == 1)
1832 rv = vm_fault_wire(map, entry->start, entry->end, 0, fictitious);
1833 if (rv) {
1834 CLIP_CHECK_BACK(entry, save_start);
1835 for (;;) {
1836 KASSERT(entry->wired_count == 1, ("wired_count changed unexpectedly"));
1837 entry->wired_count = 0;
1838 if (entry->end == save_end)
1839 break;
1840 entry = entry->next;
1841 KASSERT(entry != &map->header, ("bad entry clip during backout"));
1842 }
1843 end = save_start;
1844 break;
1845 }
1846 CLIP_CHECK_FWD(entry, save_end);
1847 entry = entry->next;
1848 }
1849 splx(s);
1850
1851 /*
1852 * relock. start_entry is still IN_TRANSITION and must
1853 * still exist, but may have been clipped (handled just
1854 * below).
1855 */
1856 vm_map_lock(map);
1857
1858 /*
1859 * If a failure occured undo everything by falling through
1860 * to the unwiring code. 'end' has already been adjusted
1861 * appropriately.
1862 */
1863 if (rv)
1864 new_pageable = 1;
1865
1866 /*
1867 * start_entry might have been clipped if we unlocked the
1868 * map and blocked. No matter how clipped it has gotten
1869 * there should be a fragment that is on our start boundary.
1870 */
1871 CLIP_CHECK_BACK(start_entry, start);
1872 }
1873
1874 if (new_pageable) {
1875 /*
1876 * This is the unwiring case. We must first ensure that the
1877 * range to be unwired is really wired down. We know there
1878 * are no holes.
1879 */
1880 entry = start_entry;
1881 while ((entry != &map->header) && (entry->start < end)) {
1882 if (entry->wired_count == 0) {
1883 rv = KERN_INVALID_ARGUMENT;
1884 goto done;
1885 }
1886 entry = entry->next;
1887 }
1888
1889 /*
1890 * Now decrement the wiring count for each region. If a region
1891 * becomes completely unwired, unwire its physical pages and
1892 * mappings.
1893 */
1894 entry = start_entry;
1895 while ((entry != &map->header) && (entry->start < end)) {
1896 entry->wired_count--;
1897 if (entry->wired_count == 0)
1898 vm_fault_unwire(map, entry->start, entry->end,
1899 entry->object.vm_object != NULL &&
1900 entry->object.vm_object->type == OBJT_DEVICE);
1901 entry = entry->next;
1902 }
1903 }
1904 done:
1905 vm_map_unclip_range(map, start_entry, start, real_end,
1906 MAP_CLIP_NO_HOLES);
1907 map->timestamp++;
1908 vm_map_unlock(map);
1909 return (rv);
1910 }
1911
1912 /*
1913 * vm_map_clean
1914 *
1915 * Push any dirty cached pages in the address range to their pager.
1916 * If syncio is TRUE, dirty pages are written synchronously.
1917 * If invalidate is TRUE, any cached pages are freed as well.
1918 *
1919 * Returns an error if any part of the specified range is not mapped.
1920 */
1921 int
1922 vm_map_clean(map, start, end, syncio, invalidate)
1923 vm_map_t map;
1924 vm_offset_t start;
1925 vm_offset_t end;
1926 boolean_t syncio;
1927 boolean_t invalidate;
1928 {
1929 vm_map_entry_t current;
1930 vm_map_entry_t entry;
1931 vm_size_t size;
1932 vm_object_t object;
1933 vm_ooffset_t offset;
1934
1935 vm_map_lock_read(map);
1936 VM_MAP_RANGE_CHECK(map, start, end);
1937 if (!vm_map_lookup_entry(map, start, &entry)) {
1938 vm_map_unlock_read(map);
1939 return (KERN_INVALID_ADDRESS);
1940 }
1941 /*
1942 * Make a first pass to check for holes.
1943 */
1944 for (current = entry; current->start < end; current = current->next) {
1945 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
1946 vm_map_unlock_read(map);
1947 return (KERN_INVALID_ARGUMENT);
1948 }
1949 if (end > current->end &&
1950 (current->next == &map->header ||
1951 current->end != current->next->start)) {
1952 vm_map_unlock_read(map);
1953 return (KERN_INVALID_ADDRESS);
1954 }
1955 }
1956
1957 if (invalidate)
1958 pmap_remove(vm_map_pmap(map), start, end);
1959 /*
1960 * Make a second pass, cleaning/uncaching pages from the indicated
1961 * objects as we go.
1962 */
1963 for (current = entry; current->start < end; current = current->next) {
1964 offset = current->offset + (start - current->start);
1965 size = (end <= current->end ? end : current->end) - start;
1966 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
1967 vm_map_t smap;
1968 vm_map_entry_t tentry;
1969 vm_size_t tsize;
1970
1971 smap = current->object.sub_map;
1972 vm_map_lock_read(smap);
1973 (void) vm_map_lookup_entry(smap, offset, &tentry);
1974 tsize = tentry->end - offset;
1975 if (tsize < size)
1976 size = tsize;
1977 object = tentry->object.vm_object;
1978 offset = tentry->offset + (offset - tentry->start);
1979 vm_map_unlock_read(smap);
1980 } else {
1981 object = current->object.vm_object;
1982 }
1983 /*
1984 * Note that there is absolutely no sense in writing out
1985 * anonymous objects, so we track down the vnode object
1986 * to write out.
1987 * We invalidate (remove) all pages from the address space
1988 * anyway, for semantic correctness.
1989 *
1990 * note: certain anonymous maps, such as MAP_NOSYNC maps,
1991 * may start out with a NULL object.
1992 */
1993 while (object && object->backing_object) {
1994 offset += object->backing_object_offset;
1995 object = object->backing_object;
1996 if (object->size < OFF_TO_IDX( offset + size))
1997 size = IDX_TO_OFF(object->size) - offset;
1998 }
1999 if (object && (object->type == OBJT_VNODE) &&
2000 (current->protection & VM_PROT_WRITE)) {
2001 /*
2002 * Flush pages if writing is allowed, invalidate them
2003 * if invalidation requested. Pages undergoing I/O
2004 * will be ignored by vm_object_page_remove().
2005 *
2006 * We cannot lock the vnode and then wait for paging
2007 * to complete without deadlocking against vm_fault.
2008 * Instead we simply call vm_object_page_remove() and
2009 * allow it to block internally on a page-by-page
2010 * basis when it encounters pages undergoing async
2011 * I/O.
2012 */
2013 int flags;
2014
2015 vm_object_reference(object);
2016 vn_lock(object->handle, LK_EXCLUSIVE | LK_RETRY, curproc);
2017 flags = (syncio || invalidate) ? OBJPC_SYNC : 0;
2018 flags |= invalidate ? OBJPC_INVAL : 0;
2019 vm_object_page_clean(object,
2020 OFF_TO_IDX(offset),
2021 OFF_TO_IDX(offset + size + PAGE_MASK),
2022 flags);
2023 VOP_UNLOCK(object->handle, 0, curproc);
2024 vm_object_deallocate(object);
2025 }
2026 if (object && invalidate &&
2027 ((object->type == OBJT_VNODE) ||
2028 (object->type == OBJT_DEVICE))) {
2029 vm_object_reference(object);
2030 vm_object_page_remove(object,
2031 OFF_TO_IDX(offset),
2032 OFF_TO_IDX(offset + size + PAGE_MASK),
2033 old_msync ? FALSE : TRUE);
2034 vm_object_deallocate(object);
2035 }
2036 start += size;
2037 }
2038
2039 vm_map_unlock_read(map);
2040 return (KERN_SUCCESS);
2041 }
2042
2043 /*
2044 * vm_map_entry_unwire: [ internal use only ]
2045 *
2046 * Make the region specified by this entry pageable.
2047 *
2048 * The map in question should be locked.
2049 * [This is the reason for this routine's existence.]
2050 */
2051 static void
2052 vm_map_entry_unwire(map, entry)
2053 vm_map_t map;
2054 vm_map_entry_t entry;
2055 {
2056 vm_fault_unwire(map, entry->start, entry->end,
2057 entry->object.vm_object != NULL &&
2058 entry->object.vm_object->type == OBJT_DEVICE);
2059 entry->wired_count = 0;
2060 }
2061
2062 /*
2063 * vm_map_entry_delete: [ internal use only ]
2064 *
2065 * Deallocate the given entry from the target map.
2066 */
2067 static void
2068 vm_map_entry_delete(map, entry)
2069 vm_map_t map;
2070 vm_map_entry_t entry;
2071 {
2072 vm_map_entry_unlink(map, entry);
2073 map->size -= entry->end - entry->start;
2074
2075 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
2076 vm_object_deallocate(entry->object.vm_object);
2077 }
2078
2079 vm_map_entry_dispose(map, entry);
2080 }
2081
2082 /*
2083 * vm_map_delete: [ internal use only ]
2084 *
2085 * Deallocates the given address range from the target
2086 * map.
2087 */
2088 int
2089 vm_map_delete(map, start, end)
2090 vm_map_t map;
2091 vm_offset_t start;
2092 vm_offset_t end;
2093 {
2094 vm_object_t object;
2095 vm_map_entry_t entry;
2096 vm_map_entry_t first_entry;
2097
2098 /*
2099 * Find the start of the region, and clip it
2100 */
2101
2102 again:
2103 if (!vm_map_lookup_entry(map, start, &first_entry))
2104 entry = first_entry->next;
2105 else {
2106 entry = first_entry;
2107 vm_map_clip_start(map, entry, start);
2108 /*
2109 * Fix the lookup hint now, rather than each time though the
2110 * loop.
2111 */
2112 SAVE_HINT(map, entry->prev);
2113 }
2114
2115 /*
2116 * Save the free space hint
2117 */
2118
2119 if (entry == &map->header) {
2120 map->first_free = &map->header;
2121 } else if (map->first_free->start >= start) {
2122 map->first_free = entry->prev;
2123 }
2124
2125 /*
2126 * Step through all entries in this region
2127 */
2128
2129 while ((entry != &map->header) && (entry->start < end)) {
2130 vm_map_entry_t next;
2131 vm_offset_t s, e;
2132 vm_pindex_t offidxstart, offidxend, count;
2133
2134 /*
2135 * If we hit an in-transition entry we have to sleep and
2136 * retry. It's easier (and not really slower) to just retry
2137 * since this case occurs so rarely and the hint is already
2138 * pointing at the right place. We have to reset the
2139 * start offset so as not to accidently delete an entry
2140 * another process just created in vacated space.
2141 */
2142 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
2143 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2144 start = entry->start;
2145 ++cnt.v_intrans_coll;
2146 ++cnt.v_intrans_wait;
2147 vm_map_transition_wait(map);
2148 goto again;
2149 }
2150 vm_map_clip_end(map, entry, end);
2151
2152 s = entry->start;
2153 e = entry->end;
2154 next = entry->next;
2155
2156 offidxstart = OFF_TO_IDX(entry->offset);
2157 count = OFF_TO_IDX(e - s);
2158 object = entry->object.vm_object;
2159
2160 /*
2161 * Unwire before removing addresses from the pmap; otherwise,
2162 * unwiring will put the entries back in the pmap.
2163 */
2164 if (entry->wired_count != 0) {
2165 vm_map_entry_unwire(map, entry);
2166 }
2167
2168 offidxend = offidxstart + count;
2169
2170 if ((object == kernel_object) || (object == kmem_object)) {
2171 vm_object_page_remove(object, offidxstart, offidxend, FALSE);
2172 } else {
2173 pmap_remove(map->pmap, s, e);
2174 if (object != NULL &&
2175 object->ref_count != 1 &&
2176 (object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING &&
2177 (object->type == OBJT_DEFAULT || object->type == OBJT_SWAP)) {
2178 vm_object_collapse(object);
2179 vm_object_page_remove(object, offidxstart, offidxend, FALSE);
2180 if (object->type == OBJT_SWAP) {
2181 swap_pager_freespace(object, offidxstart, count);
2182 }
2183 if (offidxend >= object->size &&
2184 offidxstart < object->size) {
2185 object->size = offidxstart;
2186 }
2187 }
2188 }
2189
2190 /*
2191 * Delete the entry (which may delete the object) only after
2192 * removing all pmap entries pointing to its pages.
2193 * (Otherwise, its page frames may be reallocated, and any
2194 * modify bits will be set in the wrong object!)
2195 */
2196 vm_map_entry_delete(map, entry);
2197 entry = next;
2198 }
2199 return (KERN_SUCCESS);
2200 }
2201
2202 /*
2203 * vm_map_remove:
2204 *
2205 * Remove the given address range from the target map.
2206 * This is the exported form of vm_map_delete.
2207 */
2208 int
2209 vm_map_remove(map, start, end)
2210 vm_map_t map;
2211 vm_offset_t start;
2212 vm_offset_t end;
2213 {
2214 int result, s = 0;
2215
2216 if (map == kmem_map || map == mb_map)
2217 s = splvm();
2218
2219 vm_map_lock(map);
2220 VM_MAP_RANGE_CHECK(map, start, end);
2221 result = vm_map_delete(map, start, end);
2222 vm_map_unlock(map);
2223
2224 if (map == kmem_map || map == mb_map)
2225 splx(s);
2226
2227 return (result);
2228 }
2229
2230 /*
2231 * vm_map_check_protection:
2232 *
2233 * Assert that the target map allows the specified
2234 * privilege on the entire address region given.
2235 * The entire region must be allocated.
2236 */
2237 boolean_t
2238 vm_map_check_protection(vm_map_t map, vm_offset_t start, vm_offset_t end,
2239 vm_prot_t protection)
2240 {
2241 vm_map_entry_t entry;
2242 vm_map_entry_t tmp_entry;
2243
2244 if (!vm_map_lookup_entry(map, start, &tmp_entry)) {
2245 return (FALSE);
2246 }
2247 entry = tmp_entry;
2248
2249 while (start < end) {
2250 if (entry == &map->header) {
2251 return (FALSE);
2252 }
2253 /*
2254 * No holes allowed!
2255 */
2256
2257 if (start < entry->start) {
2258 return (FALSE);
2259 }
2260 /*
2261 * Check protection associated with entry.
2262 */
2263
2264 if ((entry->protection & protection) != protection) {
2265 return (FALSE);
2266 }
2267 /* go to next entry */
2268
2269 start = entry->end;
2270 entry = entry->next;
2271 }
2272 return (TRUE);
2273 }
2274
2275 /*
2276 * Split the pages in a map entry into a new object. This affords
2277 * easier removal of unused pages, and keeps object inheritance from
2278 * being a negative impact on memory usage.
2279 */
2280 static void
2281 vm_map_split(entry)
2282 vm_map_entry_t entry;
2283 {
2284 vm_page_t m;
2285 vm_object_t orig_object, new_object, source;
2286 vm_offset_t s, e;
2287 vm_pindex_t offidxstart, offidxend, idx;
2288 vm_size_t size;
2289 vm_ooffset_t offset;
2290
2291 orig_object = entry->object.vm_object;
2292 if (orig_object->type != OBJT_DEFAULT && orig_object->type != OBJT_SWAP)
2293 return;
2294 if (orig_object->ref_count <= 1)
2295 return;
2296
2297 offset = entry->offset;
2298 s = entry->start;
2299 e = entry->end;
2300
2301 offidxstart = OFF_TO_IDX(offset);
2302 offidxend = offidxstart + OFF_TO_IDX(e - s);
2303 size = offidxend - offidxstart;
2304
2305 new_object = vm_pager_allocate(orig_object->type,
2306 NULL, IDX_TO_OFF(size), VM_PROT_ALL, 0LL);
2307 if (new_object == NULL)
2308 return;
2309
2310 source = orig_object->backing_object;
2311 if (source != NULL) {
2312 vm_object_reference(source); /* Referenced by new_object */
2313 LIST_INSERT_HEAD(&source->shadow_head,
2314 new_object, shadow_list);
2315 vm_object_clear_flag(source, OBJ_ONEMAPPING);
2316 new_object->backing_object_offset =
2317 orig_object->backing_object_offset + IDX_TO_OFF(offidxstart);
2318 new_object->backing_object = source;
2319 source->shadow_count++;
2320 source->generation++;
2321 }
2322
2323 for (idx = 0; idx < size; idx++) {
2324 vm_page_t m;
2325
2326 retry:
2327 m = vm_page_lookup(orig_object, offidxstart + idx);
2328 if (m == NULL)
2329 continue;
2330
2331 /*
2332 * We must wait for pending I/O to complete before we can
2333 * rename the page.
2334 *
2335 * We do not have to VM_PROT_NONE the page as mappings should
2336 * not be changed by this operation.
2337 */
2338 if (vm_page_sleep_busy(m, TRUE, "spltwt"))
2339 goto retry;
2340
2341 vm_page_busy(m);
2342 vm_page_rename(m, new_object, idx);
2343 /* page automatically made dirty by rename and cache handled */
2344 vm_page_busy(m);
2345 }
2346
2347 if (orig_object->type == OBJT_SWAP) {
2348 vm_object_pip_add(orig_object, 1);
2349 /*
2350 * copy orig_object pages into new_object
2351 * and destroy unneeded pages in
2352 * shadow object.
2353 */
2354 swap_pager_copy(orig_object, new_object, offidxstart, 0);
2355 vm_object_pip_wakeup(orig_object);
2356 }
2357
2358 for (idx = 0; idx < size; idx++) {
2359 m = vm_page_lookup(new_object, idx);
2360 if (m) {
2361 vm_page_wakeup(m);
2362 }
2363 }
2364
2365 entry->object.vm_object = new_object;
2366 entry->offset = 0LL;
2367 vm_object_deallocate(orig_object);
2368 }
2369
2370 /*
2371 * vm_map_copy_entry:
2372 *
2373 * Copies the contents of the source entry to the destination
2374 * entry. The entries *must* be aligned properly.
2375 */
2376 static void
2377 vm_map_copy_entry(src_map, dst_map, src_entry, dst_entry)
2378 vm_map_t src_map, dst_map;
2379 vm_map_entry_t src_entry, dst_entry;
2380 {
2381 vm_object_t src_object;
2382
2383 if ((dst_entry->eflags|src_entry->eflags) & MAP_ENTRY_IS_SUB_MAP)
2384 return;
2385
2386 if (src_entry->wired_count == 0) {
2387
2388 /*
2389 * If the source entry is marked needs_copy, it is already
2390 * write-protected.
2391 */
2392 if ((src_entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) {
2393 pmap_protect(src_map->pmap,
2394 src_entry->start,
2395 src_entry->end,
2396 src_entry->protection & ~VM_PROT_WRITE);
2397 }
2398
2399 /*
2400 * Make a copy of the object.
2401 */
2402 if ((src_object = src_entry->object.vm_object) != NULL) {
2403
2404 if ((src_object->handle == NULL) &&
2405 (src_object->type == OBJT_DEFAULT ||
2406 src_object->type == OBJT_SWAP)) {
2407 vm_object_collapse(src_object);
2408 if ((src_object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING) {
2409 vm_map_split(src_entry);
2410 src_object = src_entry->object.vm_object;
2411 }
2412 }
2413
2414 vm_object_reference(src_object);
2415 vm_object_clear_flag(src_object, OBJ_ONEMAPPING);
2416 dst_entry->object.vm_object = src_object;
2417 src_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY);
2418 dst_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY);
2419 dst_entry->offset = src_entry->offset;
2420 } else {
2421 dst_entry->object.vm_object = NULL;
2422 dst_entry->offset = 0;
2423 }
2424
2425 pmap_copy(dst_map->pmap, src_map->pmap, dst_entry->start,
2426 dst_entry->end - dst_entry->start, src_entry->start);
2427 } else {
2428 /*
2429 * Of course, wired down pages can't be set copy-on-write.
2430 * Cause wired pages to be copied into the new map by
2431 * simulating faults (the new pages are pageable)
2432 */
2433 vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry);
2434 }
2435 }
2436
2437 /*
2438 * vmspace_fork:
2439 * Create a new process vmspace structure and vm_map
2440 * based on those of an existing process. The new map
2441 * is based on the old map, according to the inheritance
2442 * values on the regions in that map.
2443 *
2444 * The source map must not be locked.
2445 */
2446 struct vmspace *
2447 vmspace_fork(vm1)
2448 struct vmspace *vm1;
2449 {
2450 struct vmspace *vm2;
2451 vm_map_t old_map = &vm1->vm_map;
2452 vm_map_t new_map;
2453 vm_map_entry_t old_entry;
2454 vm_map_entry_t new_entry;
2455 vm_object_t object;
2456
2457 vm_map_lock(old_map);
2458 old_map->infork = 1;
2459
2460 vm2 = vmspace_alloc(old_map->min_offset, old_map->max_offset);
2461 bcopy(&vm1->vm_startcopy, &vm2->vm_startcopy,
2462 (caddr_t) &vm1->vm_endcopy - (caddr_t) &vm1->vm_startcopy);
2463 new_map = &vm2->vm_map; /* XXX */
2464 new_map->timestamp = 1;
2465
2466 old_entry = old_map->header.next;
2467
2468 while (old_entry != &old_map->header) {
2469 if (old_entry->eflags & MAP_ENTRY_IS_SUB_MAP)
2470 panic("vm_map_fork: encountered a submap");
2471
2472 switch (old_entry->inheritance) {
2473 case VM_INHERIT_NONE:
2474 break;
2475
2476 case VM_INHERIT_SHARE:
2477 /*
2478 * Clone the entry, creating the shared object if necessary.
2479 */
2480 object = old_entry->object.vm_object;
2481 if (object == NULL) {
2482 object = vm_object_allocate(OBJT_DEFAULT,
2483 atop(old_entry->end - old_entry->start));
2484 old_entry->object.vm_object = object;
2485 old_entry->offset = (vm_offset_t) 0;
2486 }
2487
2488 /*
2489 * Add the reference before calling vm_object_shadow
2490 * to insure that a shadow object is created.
2491 */
2492 vm_object_reference(object);
2493 if (old_entry->eflags & MAP_ENTRY_NEEDS_COPY) {
2494 vm_object_shadow(&old_entry->object.vm_object,
2495 &old_entry->offset,
2496 atop(old_entry->end - old_entry->start));
2497 old_entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
2498 /* Transfer the second reference too. */
2499 vm_object_reference(
2500 old_entry->object.vm_object);
2501 vm_object_deallocate(object);
2502 object = old_entry->object.vm_object;
2503 }
2504 vm_object_clear_flag(object, OBJ_ONEMAPPING);
2505
2506 /*
2507 * Clone the entry, referencing the shared object.
2508 */
2509 new_entry = vm_map_entry_create(new_map);
2510 *new_entry = *old_entry;
2511 new_entry->eflags &= ~MAP_ENTRY_USER_WIRED;
2512 new_entry->wired_count = 0;
2513
2514 /*
2515 * Insert the entry into the new map -- we know we're
2516 * inserting at the end of the new map.
2517 */
2518
2519 vm_map_entry_link(new_map, new_map->header.prev,
2520 new_entry);
2521
2522 /*
2523 * Update the physical map
2524 */
2525
2526 pmap_copy(new_map->pmap, old_map->pmap,
2527 new_entry->start,
2528 (old_entry->end - old_entry->start),
2529 old_entry->start);
2530 break;
2531
2532 case VM_INHERIT_COPY:
2533 /*
2534 * Clone the entry and link into the map.
2535 */
2536 new_entry = vm_map_entry_create(new_map);
2537 *new_entry = *old_entry;
2538 new_entry->eflags &= ~MAP_ENTRY_USER_WIRED;
2539 new_entry->wired_count = 0;
2540 new_entry->object.vm_object = NULL;
2541 vm_map_entry_link(new_map, new_map->header.prev,
2542 new_entry);
2543 vm_map_copy_entry(old_map, new_map, old_entry,
2544 new_entry);
2545 break;
2546 }
2547 old_entry = old_entry->next;
2548 }
2549
2550 new_map->size = old_map->size;
2551 old_map->infork = 0;
2552 vm_map_unlock(old_map);
2553
2554 return (vm2);
2555 }
2556
2557 int
2558 vm_map_stack (vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
2559 vm_prot_t prot, vm_prot_t max, int cow)
2560 {
2561 vm_map_entry_t prev_entry;
2562 vm_map_entry_t new_stack_entry;
2563 vm_size_t init_ssize;
2564 int rv;
2565
2566 if (VM_MIN_ADDRESS > 0 && addrbos < VM_MIN_ADDRESS)
2567 return (KERN_NO_SPACE);
2568 if (addrbos > map->max_offset)
2569 return (KERN_NO_SPACE);
2570 if (max_ssize < sgrowsiz)
2571 init_ssize = max_ssize;
2572 else
2573 init_ssize = sgrowsiz;
2574
2575 vm_map_lock(map);
2576
2577 /* If addr is already mapped, no go */
2578 if (vm_map_lookup_entry(map, addrbos, &prev_entry)) {
2579 vm_map_unlock(map);
2580 return (KERN_NO_SPACE);
2581 }
2582
2583 /* If we would blow our VMEM resource limit, no go */
2584 if (map->size + init_ssize >
2585 curproc->p_rlimit[RLIMIT_VMEM].rlim_cur) {
2586 vm_map_unlock(map);
2587 return (KERN_NO_SPACE);
2588 }
2589
2590 /* If we can't accomodate max_ssize in the current mapping,
2591 * no go. However, we need to be aware that subsequent user
2592 * mappings might map into the space we have reserved for
2593 * stack, and currently this space is not protected.
2594 *
2595 * Hopefully we will at least detect this condition
2596 * when we try to grow the stack.
2597 */
2598 if ((prev_entry->next != &map->header) &&
2599 (prev_entry->next->start < addrbos + max_ssize)) {
2600 vm_map_unlock(map);
2601 return (KERN_NO_SPACE);
2602 }
2603
2604 /* We initially map a stack of only init_ssize. We will
2605 * grow as needed later. Since this is to be a grow
2606 * down stack, we map at the top of the range.
2607 *
2608 * Note: we would normally expect prot and max to be
2609 * VM_PROT_ALL, and cow to be 0. Possibly we should
2610 * eliminate these as input parameters, and just
2611 * pass these values here in the insert call.
2612 */
2613 rv = vm_map_insert(map, NULL, 0, addrbos + max_ssize - init_ssize,
2614 addrbos + max_ssize, prot, max, cow);
2615
2616 /* Now set the avail_ssize amount */
2617 if (rv == KERN_SUCCESS){
2618 if (prev_entry != &map->header)
2619 vm_map_clip_end(map, prev_entry, addrbos + max_ssize - init_ssize);
2620 new_stack_entry = prev_entry->next;
2621 if (new_stack_entry->end != addrbos + max_ssize ||
2622 new_stack_entry->start != addrbos + max_ssize - init_ssize)
2623 panic ("Bad entry start/end for new stack entry");
2624 else
2625 new_stack_entry->avail_ssize = max_ssize - init_ssize;
2626 }
2627
2628 vm_map_unlock(map);
2629 return (rv);
2630 }
2631
2632 /* Attempts to grow a vm stack entry. Returns KERN_SUCCESS if the
2633 * desired address is already mapped, or if we successfully grow
2634 * the stack. Also returns KERN_SUCCESS if addr is outside the
2635 * stack range (this is strange, but preserves compatibility with
2636 * the grow function in vm_machdep.c).
2637 */
2638 int
2639 vm_map_growstack (struct proc *p, vm_offset_t addr)
2640 {
2641 vm_map_entry_t prev_entry;
2642 vm_map_entry_t stack_entry;
2643 vm_map_entry_t new_stack_entry;
2644 struct vmspace *vm = p->p_vmspace;
2645 vm_map_t map = &vm->vm_map;
2646 vm_offset_t end;
2647 int grow_amount;
2648 int rv = KERN_SUCCESS;
2649 int is_procstack;
2650 int use_read_lock = 1;
2651
2652 Retry:
2653 if (use_read_lock)
2654 vm_map_lock_read(map);
2655 else
2656 vm_map_lock(map);
2657
2658 /* If addr is already in the entry range, no need to grow.*/
2659 if (vm_map_lookup_entry(map, addr, &prev_entry))
2660 goto done;
2661
2662 if ((stack_entry = prev_entry->next) == &map->header)
2663 goto done;
2664 if (prev_entry == &map->header)
2665 end = stack_entry->start - stack_entry->avail_ssize;
2666 else
2667 end = prev_entry->end;
2668
2669 /* This next test mimics the old grow function in vm_machdep.c.
2670 * It really doesn't quite make sense, but we do it anyway
2671 * for compatibility.
2672 *
2673 * If not growable stack, return success. This signals the
2674 * caller to proceed as he would normally with normal vm.
2675 */
2676 if (stack_entry->avail_ssize < 1 ||
2677 addr >= stack_entry->start ||
2678 addr < stack_entry->start - stack_entry->avail_ssize) {
2679 goto done;
2680 }
2681
2682 /* Find the minimum grow amount */
2683 grow_amount = roundup (stack_entry->start - addr, PAGE_SIZE);
2684 if (grow_amount > stack_entry->avail_ssize) {
2685 rv = KERN_NO_SPACE;
2686 goto done;
2687 }
2688
2689 /* If there is no longer enough space between the entries
2690 * nogo, and adjust the available space. Note: this
2691 * should only happen if the user has mapped into the
2692 * stack area after the stack was created, and is
2693 * probably an error.
2694 *
2695 * This also effectively destroys any guard page the user
2696 * might have intended by limiting the stack size.
2697 */
2698 if (grow_amount > stack_entry->start - end) {
2699 if (use_read_lock && vm_map_lock_upgrade(map)) {
2700 use_read_lock = 0;
2701 goto Retry;
2702 }
2703 use_read_lock = 0;
2704 stack_entry->avail_ssize = stack_entry->start - end;
2705 rv = KERN_NO_SPACE;
2706 goto done;
2707 }
2708
2709 is_procstack = addr >= (vm_offset_t)vm->vm_maxsaddr;
2710
2711 /* If this is the main process stack, see if we're over the
2712 * stack limit.
2713 */
2714 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount >
2715 p->p_rlimit[RLIMIT_STACK].rlim_cur)) {
2716 rv = KERN_NO_SPACE;
2717 goto done;
2718 }
2719
2720 /* Round up the grow amount modulo SGROWSIZ */
2721 grow_amount = roundup (grow_amount, sgrowsiz);
2722 if (grow_amount > stack_entry->avail_ssize) {
2723 grow_amount = stack_entry->avail_ssize;
2724 }
2725 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount >
2726 p->p_rlimit[RLIMIT_STACK].rlim_cur)) {
2727 grow_amount = p->p_rlimit[RLIMIT_STACK].rlim_cur -
2728 ctob(vm->vm_ssize);
2729 }
2730
2731 /* If we would blow our VMEM resource limit, no go */
2732 if (map->size + grow_amount >
2733 curproc->p_rlimit[RLIMIT_VMEM].rlim_cur) {
2734 rv = KERN_NO_SPACE;
2735 goto done;
2736 }
2737
2738 if (use_read_lock && vm_map_lock_upgrade(map)) {
2739 use_read_lock = 0;
2740 goto Retry;
2741 }
2742 use_read_lock = 0;
2743
2744 /* Get the preliminary new entry start value */
2745 addr = stack_entry->start - grow_amount;
2746
2747 /* If this puts us into the previous entry, cut back our growth
2748 * to the available space. Also, see the note above.
2749 */
2750 if (addr < end) {
2751 stack_entry->avail_ssize = stack_entry->start - end;
2752 addr = end;
2753 }
2754
2755 rv = vm_map_insert(map, NULL, 0, addr, stack_entry->start,
2756 VM_PROT_ALL,
2757 VM_PROT_ALL,
2758 0);
2759
2760 /* Adjust the available stack space by the amount we grew. */
2761 if (rv == KERN_SUCCESS) {
2762 if (prev_entry != &map->header)
2763 vm_map_clip_end(map, prev_entry, addr);
2764 new_stack_entry = prev_entry->next;
2765 if (new_stack_entry->end != stack_entry->start ||
2766 new_stack_entry->start != addr)
2767 panic ("Bad stack grow start/end in new stack entry");
2768 else {
2769 new_stack_entry->avail_ssize = stack_entry->avail_ssize -
2770 (new_stack_entry->end -
2771 new_stack_entry->start);
2772 if (is_procstack)
2773 vm->vm_ssize += btoc(new_stack_entry->end -
2774 new_stack_entry->start);
2775 }
2776 }
2777
2778 done:
2779 if (use_read_lock)
2780 vm_map_unlock_read(map);
2781 else
2782 vm_map_unlock(map);
2783 return (rv);
2784 }
2785
2786 /*
2787 * Unshare the specified VM space for exec. If other processes are
2788 * mapped to it, then create a new one. The new vmspace is null.
2789 */
2790
2791 void
2792 vmspace_exec(struct proc *p) {
2793 struct vmspace *oldvmspace = p->p_vmspace;
2794 struct vmspace *newvmspace;
2795 vm_map_t map = &p->p_vmspace->vm_map;
2796
2797 newvmspace = vmspace_alloc(map->min_offset, map->max_offset);
2798 bcopy(&oldvmspace->vm_startcopy, &newvmspace->vm_startcopy,
2799 (caddr_t) &newvmspace->vm_endcopy -
2800 (caddr_t) &newvmspace->vm_startcopy);
2801 /*
2802 * This code is written like this for prototype purposes. The
2803 * goal is to avoid running down the vmspace here, but let the
2804 * other process's that are still using the vmspace to finally
2805 * run it down. Even though there is little or no chance of blocking
2806 * here, it is a good idea to keep this form for future mods.
2807 */
2808 vmspace_free(oldvmspace);
2809 p->p_vmspace = newvmspace;
2810 pmap_pinit2(vmspace_pmap(newvmspace));
2811 if (p == curproc)
2812 pmap_activate(p);
2813 }
2814
2815 /*
2816 * Unshare the specified VM space for forcing COW. This
2817 * is called by rfork, for the (RFMEM|RFPROC) == 0 case.
2818 */
2819
2820 void
2821 vmspace_unshare(struct proc *p) {
2822 struct vmspace *oldvmspace = p->p_vmspace;
2823 struct vmspace *newvmspace;
2824
2825 if (oldvmspace->vm_refcnt == 1)
2826 return;
2827 newvmspace = vmspace_fork(oldvmspace);
2828 vmspace_free(oldvmspace);
2829 p->p_vmspace = newvmspace;
2830 pmap_pinit2(vmspace_pmap(newvmspace));
2831 if (p == curproc)
2832 pmap_activate(p);
2833 }
2834
2835
2836 /*
2837 * vm_map_lookup:
2838 *
2839 * Finds the VM object, offset, and
2840 * protection for a given virtual address in the
2841 * specified map, assuming a page fault of the
2842 * type specified.
2843 *
2844 * Leaves the map in question locked for read; return
2845 * values are guaranteed until a vm_map_lookup_done
2846 * call is performed. Note that the map argument
2847 * is in/out; the returned map must be used in
2848 * the call to vm_map_lookup_done.
2849 *
2850 * A handle (out_entry) is returned for use in
2851 * vm_map_lookup_done, to make that fast.
2852 *
2853 * If a lookup is requested with "write protection"
2854 * specified, the map may be changed to perform virtual
2855 * copying operations, although the data referenced will
2856 * remain the same.
2857 */
2858 int
2859 vm_map_lookup(vm_map_t *var_map, /* IN/OUT */
2860 vm_offset_t vaddr,
2861 vm_prot_t fault_typea,
2862 vm_map_entry_t *out_entry, /* OUT */
2863 vm_object_t *object, /* OUT */
2864 vm_pindex_t *pindex, /* OUT */
2865 vm_prot_t *out_prot, /* OUT */
2866 boolean_t *wired) /* OUT */
2867 {
2868 vm_map_entry_t entry;
2869 vm_map_t map = *var_map;
2870 vm_prot_t prot;
2871 vm_prot_t fault_type = fault_typea;
2872 int use_read_lock = 1;
2873 int rv = KERN_SUCCESS;
2874
2875 RetryLookup:
2876 if (use_read_lock)
2877 vm_map_lock_read(map);
2878 else
2879 vm_map_lock(map);
2880
2881 /*
2882 * If the map has an interesting hint, try it before calling full
2883 * blown lookup routine.
2884 */
2885 entry = map->hint;
2886 *out_entry = entry;
2887
2888 if ((entry == &map->header) ||
2889 (vaddr < entry->start) || (vaddr >= entry->end)) {
2890 vm_map_entry_t tmp_entry;
2891
2892 /*
2893 * Entry was either not a valid hint, or the vaddr was not
2894 * contained in the entry, so do a full lookup.
2895 */
2896 if (!vm_map_lookup_entry(map, vaddr, &tmp_entry)) {
2897 rv = KERN_INVALID_ADDRESS;
2898 goto done;
2899 }
2900
2901 entry = tmp_entry;
2902 *out_entry = entry;
2903 }
2904
2905 /*
2906 * Handle submaps.
2907 */
2908
2909 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
2910 vm_map_t old_map = map;
2911
2912 *var_map = map = entry->object.sub_map;
2913 if (use_read_lock)
2914 vm_map_unlock_read(old_map);
2915 else
2916 vm_map_unlock(old_map);
2917 use_read_lock = 1;
2918 goto RetryLookup;
2919 }
2920
2921 /*
2922 * Check whether this task is allowed to have this page.
2923 * Note the special case for MAP_ENTRY_COW
2924 * pages with an override. This is to implement a forced
2925 * COW for debuggers.
2926 */
2927
2928 if (fault_type & VM_PROT_OVERRIDE_WRITE)
2929 prot = entry->max_protection;
2930 else
2931 prot = entry->protection;
2932
2933 fault_type &= (VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE);
2934 if ((fault_type & prot) != fault_type) {
2935 rv = KERN_PROTECTION_FAILURE;
2936 goto done;
2937 }
2938
2939 if ((entry->eflags & MAP_ENTRY_USER_WIRED) &&
2940 (entry->eflags & MAP_ENTRY_COW) &&
2941 (fault_type & VM_PROT_WRITE) &&
2942 (fault_typea & VM_PROT_OVERRIDE_WRITE) == 0) {
2943 rv = KERN_PROTECTION_FAILURE;
2944 goto done;
2945 }
2946
2947 /*
2948 * If this page is not pageable, we have to get it for all possible
2949 * accesses.
2950 */
2951
2952 *wired = (entry->wired_count != 0);
2953 if (*wired)
2954 prot = fault_type = entry->protection;
2955
2956 /*
2957 * If the entry was copy-on-write, we either ...
2958 */
2959
2960 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
2961 /*
2962 * If we want to write the page, we may as well handle that
2963 * now since we've got the map locked.
2964 *
2965 * If we don't need to write the page, we just demote the
2966 * permissions allowed.
2967 */
2968
2969 if (fault_type & VM_PROT_WRITE) {
2970 /*
2971 * Make a new object, and place it in the object
2972 * chain. Note that no new references have appeared
2973 * -- one just moved from the map to the new
2974 * object.
2975 */
2976
2977 if (use_read_lock && vm_map_lock_upgrade(map)) {
2978 use_read_lock = 0;
2979 goto RetryLookup;
2980 }
2981 use_read_lock = 0;
2982
2983 vm_object_shadow(
2984 &entry->object.vm_object,
2985 &entry->offset,
2986 atop(entry->end - entry->start));
2987
2988 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
2989 } else {
2990 /*
2991 * We're attempting to read a copy-on-write page --
2992 * don't allow writes.
2993 */
2994
2995 prot &= ~VM_PROT_WRITE;
2996 }
2997 }
2998
2999 /*
3000 * Create an object if necessary.
3001 */
3002 if (entry->object.vm_object == NULL &&
3003 !map->system_map) {
3004 if (use_read_lock && vm_map_lock_upgrade(map)) {
3005 use_read_lock = 0;
3006 goto RetryLookup;
3007 }
3008 use_read_lock = 0;
3009 entry->object.vm_object = vm_object_allocate(OBJT_DEFAULT,
3010 atop(entry->end - entry->start));
3011 entry->offset = 0;
3012 }
3013
3014 /*
3015 * Return the object/offset from this entry. If the entry was
3016 * copy-on-write or empty, it has been fixed up.
3017 */
3018
3019 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
3020 *object = entry->object.vm_object;
3021
3022 /*
3023 * Return whether this is the only map sharing this data. On
3024 * success we return with a read lock held on the map. On failure
3025 * we return with the map unlocked.
3026 */
3027 *out_prot = prot;
3028 done:
3029 if (rv == KERN_SUCCESS) {
3030 if (use_read_lock == 0)
3031 vm_map_lock_downgrade(map);
3032 } else if (use_read_lock) {
3033 vm_map_unlock_read(map);
3034 } else {
3035 vm_map_unlock(map);
3036 }
3037 return (rv);
3038 }
3039
3040 /*
3041 * vm_map_lookup_done:
3042 *
3043 * Releases locks acquired by a vm_map_lookup
3044 * (according to the handle returned by that lookup).
3045 */
3046
3047 void
3048 vm_map_lookup_done(map, entry)
3049 vm_map_t map;
3050 vm_map_entry_t entry;
3051 {
3052 /*
3053 * Unlock the main-level map
3054 */
3055
3056 vm_map_unlock_read(map);
3057 }
3058
3059 /*
3060 * Implement uiomove with VM operations. This handles (and collateral changes)
3061 * support every combination of source object modification, and COW type
3062 * operations.
3063 */
3064 int
3065 vm_uiomove(mapa, srcobject, cp, cnta, uaddra, npages)
3066 vm_map_t mapa;
3067 vm_object_t srcobject;
3068 off_t cp;
3069 int cnta;
3070 vm_offset_t uaddra;
3071 int *npages;
3072 {
3073 vm_map_t map;
3074 vm_object_t first_object, oldobject, object;
3075 vm_map_entry_t entry;
3076 vm_prot_t prot;
3077 boolean_t wired;
3078 int tcnt, rv;
3079 vm_offset_t uaddr, start, end, tend;
3080 vm_pindex_t first_pindex, osize, oindex;
3081 off_t ooffset;
3082 int cnt;
3083
3084 if (npages)
3085 *npages = 0;
3086
3087 cnt = cnta;
3088 uaddr = uaddra;
3089
3090 while (cnt > 0) {
3091 map = mapa;
3092
3093 if ((vm_map_lookup(&map, uaddr,
3094 VM_PROT_READ, &entry, &first_object,
3095 &first_pindex, &prot, &wired)) != KERN_SUCCESS) {
3096 return EFAULT;
3097 }
3098
3099 vm_map_clip_start(map, entry, uaddr);
3100
3101 tcnt = cnt;
3102 tend = uaddr + tcnt;
3103 if (tend > entry->end) {
3104 tcnt = entry->end - uaddr;
3105 tend = entry->end;
3106 }
3107
3108 vm_map_clip_end(map, entry, tend);
3109
3110 start = entry->start;
3111 end = entry->end;
3112
3113 osize = atop(tcnt);
3114
3115 oindex = OFF_TO_IDX(cp);
3116 if (npages) {
3117 vm_pindex_t idx;
3118 for (idx = 0; idx < osize; idx++) {
3119 vm_page_t m;
3120 if ((m = vm_page_lookup(srcobject, oindex + idx)) == NULL) {
3121 vm_map_lookup_done(map, entry);
3122 return 0;
3123 }
3124 /*
3125 * disallow busy or invalid pages, but allow
3126 * m->busy pages if they are entirely valid.
3127 */
3128 if ((m->flags & PG_BUSY) ||
3129 ((m->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL)) {
3130 vm_map_lookup_done(map, entry);
3131 return 0;
3132 }
3133 }
3134 }
3135
3136 /*
3137 * If we are changing an existing map entry, just redirect
3138 * the object, and change mappings.
3139 */
3140 if ((first_object->type == OBJT_VNODE) &&
3141 ((oldobject = entry->object.vm_object) == first_object)) {
3142
3143 if ((entry->offset != cp) || (oldobject != srcobject)) {
3144 /*
3145 * Remove old window into the file
3146 */
3147 pmap_remove (map->pmap, uaddr, tend);
3148
3149 /*
3150 * Force copy on write for mmaped regions
3151 */
3152 vm_object_pmap_copy_1 (srcobject, oindex, oindex + osize);
3153
3154 /*
3155 * Point the object appropriately
3156 */
3157 if (oldobject != srcobject) {
3158
3159 /*
3160 * Set the object optimization hint flag
3161 */
3162 vm_object_set_flag(srcobject, OBJ_OPT);
3163 vm_object_reference(srcobject);
3164 entry->object.vm_object = srcobject;
3165
3166 if (oldobject) {
3167 vm_object_deallocate(oldobject);
3168 }
3169 }
3170
3171 entry->offset = cp;
3172 map->timestamp++;
3173 } else {
3174 pmap_remove (map->pmap, uaddr, tend);
3175 }
3176
3177 } else if ((first_object->ref_count == 1) &&
3178 (first_object->size == osize) &&
3179 ((first_object->type == OBJT_DEFAULT) ||
3180 (first_object->type == OBJT_SWAP)) ) {
3181
3182 oldobject = first_object->backing_object;
3183
3184 if ((first_object->backing_object_offset != cp) ||
3185 (oldobject != srcobject)) {
3186 /*
3187 * Remove old window into the file
3188 */
3189 pmap_remove (map->pmap, uaddr, tend);
3190
3191 /*
3192 * Remove unneeded old pages
3193 */
3194 vm_object_page_remove(first_object, 0, 0, 0);
3195
3196 /*
3197 * Invalidate swap space
3198 */
3199 if (first_object->type == OBJT_SWAP) {
3200 swap_pager_freespace(first_object,
3201 0,
3202 first_object->size);
3203 }
3204
3205 /*
3206 * Force copy on write for mmaped regions
3207 */
3208 vm_object_pmap_copy_1 (srcobject, oindex, oindex + osize);
3209
3210 /*
3211 * Point the object appropriately
3212 */
3213 if (oldobject != srcobject) {
3214
3215 /*
3216 * Set the object optimization hint flag
3217 */
3218 vm_object_set_flag(srcobject, OBJ_OPT);
3219 vm_object_reference(srcobject);
3220
3221 if (oldobject) {
3222 LIST_REMOVE(
3223 first_object, shadow_list);
3224 oldobject->shadow_count--;
3225 /* XXX bump generation? */
3226 vm_object_deallocate(oldobject);
3227 }
3228
3229 LIST_INSERT_HEAD(&srcobject->shadow_head,
3230 first_object, shadow_list);
3231 srcobject->shadow_count++;
3232 /* XXX bump generation? */
3233
3234 first_object->backing_object = srcobject;
3235 }
3236 first_object->backing_object_offset = cp;
3237 map->timestamp++;
3238 } else {
3239 pmap_remove (map->pmap, uaddr, tend);
3240 }
3241 /*
3242 * Otherwise, we have to do a logical mmap.
3243 */
3244 } else {
3245
3246 vm_object_set_flag(srcobject, OBJ_OPT);
3247 vm_object_reference(srcobject);
3248
3249 pmap_remove (map->pmap, uaddr, tend);
3250
3251 vm_object_pmap_copy_1 (srcobject, oindex, oindex + osize);
3252 vm_map_lock_upgrade(map);
3253
3254 if (entry == &map->header) {
3255 map->first_free = &map->header;
3256 } else if (map->first_free->start >= start) {
3257 map->first_free = entry->prev;
3258 }
3259
3260 SAVE_HINT(map, entry->prev);
3261 vm_map_entry_delete(map, entry);
3262
3263 object = srcobject;
3264 ooffset = cp;
3265
3266 rv = vm_map_insert(map, object, ooffset, start, tend,
3267 VM_PROT_ALL, VM_PROT_ALL, MAP_COPY_ON_WRITE);
3268
3269 if (rv != KERN_SUCCESS)
3270 panic("vm_uiomove: could not insert new entry: %d", rv);
3271 }
3272
3273 /*
3274 * Map the window directly, if it is already in memory
3275 */
3276 pmap_object_init_pt(map->pmap, uaddr, entry->protection,
3277 srcobject, oindex, tcnt, 0);
3278
3279 map->timestamp++;
3280 vm_map_unlock(map);
3281
3282 cnt -= tcnt;
3283 uaddr += tcnt;
3284 cp += tcnt;
3285 if (npages)
3286 *npages += osize;
3287 }
3288 return 0;
3289 }
3290
3291 /*
3292 * Performs the copy_on_write operations necessary to allow the virtual copies
3293 * into user space to work. This has to be called for write(2) system calls
3294 * from other processes, file unlinking, and file size shrinkage.
3295 */
3296 void
3297 vm_freeze_copyopts(object, froma, toa)
3298 vm_object_t object;
3299 vm_pindex_t froma, toa;
3300 {
3301 int rv;
3302 vm_object_t robject;
3303 vm_pindex_t idx;
3304
3305 if ((object == NULL) ||
3306 ((object->flags & OBJ_OPT) == 0))
3307 return;
3308
3309 if (object->shadow_count > object->ref_count)
3310 panic("vm_freeze_copyopts: sc > rc");
3311
3312 while((robject = LIST_FIRST(&object->shadow_head)) != NULL) {
3313 vm_pindex_t bo_pindex;
3314 vm_page_t m_in, m_out;
3315
3316 bo_pindex = OFF_TO_IDX(robject->backing_object_offset);
3317
3318 vm_object_reference(robject);
3319
3320 vm_object_pip_wait(robject, "objfrz");
3321
3322 if (robject->ref_count == 1) {
3323 vm_object_deallocate(robject);
3324 continue;
3325 }
3326
3327 vm_object_pip_add(robject, 1);
3328
3329 for (idx = 0; idx < robject->size; idx++) {
3330
3331 m_out = vm_page_grab(robject, idx,
3332 VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
3333
3334 if (m_out->valid == 0) {
3335 m_in = vm_page_grab(object, bo_pindex + idx,
3336 VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
3337 if (m_in->valid == 0) {
3338 rv = vm_pager_get_pages(object, &m_in, 1, 0);
3339 if (rv != VM_PAGER_OK) {
3340 printf("vm_freeze_copyopts: cannot read page from file: %lx\n", (long)m_in->pindex);
3341 continue;
3342 }
3343 vm_page_deactivate(m_in);
3344 }
3345
3346 vm_page_protect(m_in, VM_PROT_NONE);
3347 pmap_copy_page(VM_PAGE_TO_PHYS(m_in), VM_PAGE_TO_PHYS(m_out));
3348 m_out->valid = m_in->valid;
3349 vm_page_dirty(m_out);
3350 vm_page_activate(m_out);
3351 vm_page_wakeup(m_in);
3352 }
3353 vm_page_wakeup(m_out);
3354 }
3355
3356 object->shadow_count--;
3357 object->ref_count--;
3358 LIST_REMOVE(robject, shadow_list);
3359 robject->backing_object = NULL;
3360 robject->backing_object_offset = 0;
3361
3362 vm_object_pip_wakeup(robject);
3363 vm_object_deallocate(robject);
3364 }
3365
3366 vm_object_clear_flag(object, OBJ_OPT);
3367 }
3368
3369 #include "opt_ddb.h"
3370 #ifdef DDB
3371 #include <sys/kernel.h>
3372
3373 #include <ddb/ddb.h>
3374
3375 /*
3376 * vm_map_print: [ debug ]
3377 */
3378 DB_SHOW_COMMAND(map, vm_map_print)
3379 {
3380 static int nlines;
3381 /* XXX convert args. */
3382 vm_map_t map = (vm_map_t)addr;
3383 boolean_t full = have_addr;
3384
3385 vm_map_entry_t entry;
3386
3387 db_iprintf("Task map %p: pmap=%p, nentries=%d, version=%u\n",
3388 (void *)map,
3389 (void *)map->pmap, map->nentries, map->timestamp);
3390 nlines++;
3391
3392 if (!full && db_indent)
3393 return;
3394
3395 db_indent += 2;
3396 for (entry = map->header.next; entry != &map->header;
3397 entry = entry->next) {
3398 db_iprintf("map entry %p: start=%p, end=%p\n",
3399 (void *)entry, (void *)entry->start, (void *)entry->end);
3400 nlines++;
3401 {
3402 static char *inheritance_name[4] =
3403 {"share", "copy", "none", "donate_copy"};
3404
3405 db_iprintf(" prot=%x/%x/%s",
3406 entry->protection,
3407 entry->max_protection,
3408 inheritance_name[(int)(unsigned char)entry->inheritance]);
3409 if (entry->wired_count != 0)
3410 db_printf(", wired");
3411 }
3412 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
3413 /* XXX no %qd in kernel. Truncate entry->offset. */
3414 db_printf(", share=%p, offset=0x%lx\n",
3415 (void *)entry->object.sub_map,
3416 (long)entry->offset);
3417 nlines++;
3418 if ((entry->prev == &map->header) ||
3419 (entry->prev->object.sub_map !=
3420 entry->object.sub_map)) {
3421 db_indent += 2;
3422 vm_map_print((db_expr_t)(intptr_t)
3423 entry->object.sub_map,
3424 full, 0, (char *)0);
3425 db_indent -= 2;
3426 }
3427 } else {
3428 /* XXX no %qd in kernel. Truncate entry->offset. */
3429 db_printf(", object=%p, offset=0x%lx",
3430 (void *)entry->object.vm_object,
3431 (long)entry->offset);
3432 if (entry->eflags & MAP_ENTRY_COW)
3433 db_printf(", copy (%s)",
3434 (entry->eflags & MAP_ENTRY_NEEDS_COPY) ? "needed" : "done");
3435 db_printf("\n");
3436 nlines++;
3437
3438 if ((entry->prev == &map->header) ||
3439 (entry->prev->object.vm_object !=
3440 entry->object.vm_object)) {
3441 db_indent += 2;
3442 vm_object_print((db_expr_t)(intptr_t)
3443 entry->object.vm_object,
3444 full, 0, (char *)0);
3445 nlines += 4;
3446 db_indent -= 2;
3447 }
3448 }
3449 }
3450 db_indent -= 2;
3451 if (db_indent == 0)
3452 nlines = 0;
3453 }
3454
3455
3456 DB_SHOW_COMMAND(procvm, procvm)
3457 {
3458 struct proc *p;
3459
3460 if (have_addr) {
3461 p = (struct proc *) addr;
3462 } else {
3463 p = curproc;
3464 }
3465
3466 db_printf("p = %p, vmspace = %p, map = %p, pmap = %p\n",
3467 (void *)p, (void *)p->p_vmspace, (void *)&p->p_vmspace->vm_map,
3468 (void *)vmspace_pmap(p->p_vmspace));
3469
3470 vm_map_print((db_expr_t)(intptr_t)&p->p_vmspace->vm_map, 1, 0, NULL);
3471 }
3472
3473 #endif /* DDB */
Cache object: d0993de21bf03391d018e62aad07e71d
|