Commit | Line | Data |
---|---|---|
175f072e | 1 | /* |
175f072e KM |
2 | * Copyright (c) 1991 Regents of the University of California. |
3 | * 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 | * | |
0e24ad83 | 8 | * %sccs.include.redist.c% |
175f072e | 9 | * |
b5246c21 | 10 | * @(#)vm_map.c 7.10 (Berkeley) %G% |
0e24ad83 KM |
11 | * |
12 | * | |
13 | * Copyright (c) 1987, 1990 Carnegie-Mellon University. | |
14 | * All rights reserved. | |
15 | * | |
16 | * Authors: Avadis Tevanian, Jr., Michael Wayne Young | |
17 | * | |
18 | * Permission to use, copy, modify and distribute this software and | |
19 | * its documentation is hereby granted, provided that both the copyright | |
20 | * notice and this permission notice appear in all copies of the | |
21 | * software, derivative works or modified versions, and any portions | |
22 | * thereof, and that both notices appear in supporting documentation. | |
23 | * | |
24 | * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" | |
25 | * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND | |
26 | * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. | |
27 | * | |
28 | * Carnegie Mellon requests users of this software to return to | |
29 | * | |
30 | * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU | |
31 | * School of Computer Science | |
32 | * Carnegie Mellon University | |
33 | * Pittsburgh PA 15213-3890 | |
34 | * | |
35 | * any improvements or extensions that they make and grant Carnegie the | |
36 | * rights to redistribute these changes. | |
175f072e KM |
37 | */ |
38 | ||
39 | /* | |
40 | * Virtual memory mapping module. | |
41 | */ | |
42 | ||
73506ff8 KB |
43 | #include <sys/param.h> |
44 | #include <sys/systm.h> | |
45 | #include <sys/malloc.h> | |
46 | ||
47 | #include <vm/vm.h> | |
48 | #include <vm/vm_page.h> | |
49 | #include <vm/vm_object.h> | |
175f072e KM |
50 | |
51 | /* | |
52 | * Virtual memory maps provide for the mapping, protection, | |
53 | * and sharing of virtual memory objects. In addition, | |
54 | * this module provides for an efficient virtual copy of | |
55 | * memory from one map to another. | |
56 | * | |
57 | * Synchronization is required prior to most operations. | |
58 | * | |
59 | * Maps consist of an ordered doubly-linked list of simple | |
60 | * entries; a single hint is used to speed up lookups. | |
61 | * | |
62 | * In order to properly represent the sharing of virtual | |
63 | * memory regions among maps, the map structure is bi-level. | |
64 | * Top-level ("address") maps refer to regions of sharable | |
65 | * virtual memory. These regions are implemented as | |
66 | * ("sharing") maps, which then refer to the actual virtual | |
67 | * memory objects. When two address maps "share" memory, | |
68 | * their top-level maps both have references to the same | |
69 | * sharing map. When memory is virtual-copied from one | |
70 | * address map to another, the references in the sharing | |
71 | * maps are actually copied -- no copying occurs at the | |
72 | * virtual memory object level. | |
73 | * | |
74 | * Since portions of maps are specified by start/end addreses, | |
75 | * which may not align with existing map entries, all | |
76 | * routines merely "clip" entries to these start/end values. | |
77 | * [That is, an entry is split into two, bordering at a | |
78 | * start or end value.] Note that these clippings may not | |
79 | * always be necessary (as the two resulting entries are then | |
80 | * not changed); however, the clipping is done for convenience. | |
81 | * No attempt is currently made to "glue back together" two | |
82 | * abutting entries. | |
83 | * | |
84 | * As mentioned above, virtual copy operations are performed | |
85 | * by copying VM object references from one sharing map to | |
86 | * another, and then marking both regions as copy-on-write. | |
87 | * It is important to note that only one writeable reference | |
88 | * to a VM object region exists in any map -- this means that | |
89 | * shadow object creation can be delayed until a write operation | |
90 | * occurs. | |
91 | */ | |
92 | ||
93 | /* | |
5d7b9ad3 | 94 | * vm_map_startup: |
175f072e KM |
95 | * |
96 | * Initialize the vm_map module. Must be called before | |
97 | * any other vm_map routines. | |
98 | * | |
99 | * Map and entry structures are allocated from the general | |
100 | * purpose memory pool with some exceptions: | |
101 | * | |
102 | * - The kernel map and kmem submap are allocated statically. | |
103 | * - Kernel map entries are allocated out of a static pool. | |
104 | * | |
105 | * These restrictions are necessary since malloc() uses the | |
106 | * maps and requires map entries. | |
107 | */ | |
108 | ||
109 | vm_offset_t kentry_data; | |
110 | vm_size_t kentry_data_size; | |
111 | vm_map_entry_t kentry_free; | |
112 | vm_map_t kmap_free; | |
113 | ||
73506ff8 KB |
114 | static void _vm_map_clip_end __P((vm_map_t, vm_map_entry_t, vm_offset_t)); |
115 | static void _vm_map_clip_start __P((vm_map_t, vm_map_entry_t, vm_offset_t)); | |
116 | ||
5d7b9ad3 | 117 | void vm_map_startup() |
175f072e KM |
118 | { |
119 | register int i; | |
120 | register vm_map_entry_t mep; | |
121 | vm_map_t mp; | |
122 | ||
123 | /* | |
124 | * Static map structures for allocation before initialization of | |
125 | * kernel map or kmem map. vm_map_create knows how to deal with them. | |
126 | */ | |
127 | kmap_free = mp = (vm_map_t) kentry_data; | |
128 | i = MAX_KMAP; | |
129 | while (--i > 0) { | |
130 | mp->header.next = (vm_map_entry_t) (mp + 1); | |
131 | mp++; | |
132 | } | |
5d7b9ad3 | 133 | mp++->header.next = NULL; |
175f072e KM |
134 | |
135 | /* | |
136 | * Form a free list of statically allocated kernel map entries | |
137 | * with the rest. | |
138 | */ | |
139 | kentry_free = mep = (vm_map_entry_t) mp; | |
140 | i = (kentry_data_size - MAX_KMAP * sizeof *mp) / sizeof *mep; | |
141 | while (--i > 0) { | |
142 | mep->next = mep + 1; | |
143 | mep++; | |
144 | } | |
5d7b9ad3 MK |
145 | mep->next = NULL; |
146 | } | |
147 | ||
148 | /* | |
149 | * Allocate a vmspace structure, including a vm_map and pmap, | |
150 | * and initialize those structures. The refcnt is set to 1. | |
151 | * The remaining fields must be initialized by the caller. | |
152 | */ | |
153 | struct vmspace * | |
154 | vmspace_alloc(min, max, pageable) | |
155 | vm_offset_t min, max; | |
156 | int pageable; | |
157 | { | |
158 | register struct vmspace *vm; | |
159 | ||
160 | MALLOC(vm, struct vmspace *, sizeof(struct vmspace), M_VMMAP, M_WAITOK); | |
161 | bzero(vm, (caddr_t) &vm->vm_startcopy - (caddr_t) vm); | |
162 | vm_map_init(&vm->vm_map, min, max, pageable); | |
163 | pmap_pinit(&vm->vm_pmap); | |
164 | vm->vm_map.pmap = &vm->vm_pmap; /* XXX */ | |
165 | vm->vm_refcnt = 1; | |
166 | return (vm); | |
167 | } | |
168 | ||
169 | void | |
170 | vmspace_free(vm) | |
171 | register struct vmspace *vm; | |
172 | { | |
173 | ||
174 | if (--vm->vm_refcnt == 0) { | |
175 | /* | |
176 | * Lock the map, to wait out all other references to it. | |
177 | * Delete all of the mappings and pages they hold, | |
178 | * then call the pmap module to reclaim anything left. | |
179 | */ | |
180 | vm_map_lock(&vm->vm_map); | |
181 | (void) vm_map_delete(&vm->vm_map, vm->vm_map.min_offset, | |
182 | vm->vm_map.max_offset); | |
183 | pmap_release(&vm->vm_pmap); | |
184 | FREE(vm, M_VMMAP); | |
185 | } | |
175f072e KM |
186 | } |
187 | ||
188 | /* | |
189 | * vm_map_create: | |
190 | * | |
191 | * Creates and returns a new empty VM map with | |
192 | * the given physical map structure, and having | |
193 | * the given lower and upper address bounds. | |
194 | */ | |
195 | vm_map_t vm_map_create(pmap, min, max, pageable) | |
196 | pmap_t pmap; | |
197 | vm_offset_t min, max; | |
198 | boolean_t pageable; | |
199 | { | |
200 | register vm_map_t result; | |
201 | extern vm_map_t kernel_map, kmem_map; | |
202 | ||
5d7b9ad3 | 203 | if (kmem_map == NULL) { |
175f072e KM |
204 | result = kmap_free; |
205 | kmap_free = (vm_map_t) result->header.next; | |
5d7b9ad3 MK |
206 | if (result == NULL) |
207 | panic("vm_map_create: out of maps"); | |
175f072e KM |
208 | } else |
209 | MALLOC(result, vm_map_t, sizeof(struct vm_map), | |
210 | M_VMMAP, M_WAITOK); | |
211 | ||
5d7b9ad3 | 212 | vm_map_init(result, min, max, pageable); |
175f072e | 213 | result->pmap = pmap; |
175f072e KM |
214 | return(result); |
215 | } | |
216 | ||
5d7b9ad3 MK |
217 | /* |
218 | * Initialize an existing vm_map structure | |
219 | * such as that in the vmspace structure. | |
220 | * The pmap is set elsewhere. | |
221 | */ | |
222 | void | |
223 | vm_map_init(map, min, max, pageable) | |
224 | register struct vm_map *map; | |
225 | vm_offset_t min, max; | |
226 | boolean_t pageable; | |
227 | { | |
228 | map->header.next = map->header.prev = &map->header; | |
229 | map->nentries = 0; | |
230 | map->size = 0; | |
231 | map->ref_count = 1; | |
232 | map->is_main_map = TRUE; | |
233 | map->min_offset = min; | |
234 | map->max_offset = max; | |
235 | map->entries_pageable = pageable; | |
236 | map->first_free = &map->header; | |
237 | map->hint = &map->header; | |
238 | map->timestamp = 0; | |
239 | lock_init(&map->lock, TRUE); | |
240 | simple_lock_init(&map->ref_lock); | |
241 | simple_lock_init(&map->hint_lock); | |
242 | } | |
243 | ||
175f072e KM |
244 | /* |
245 | * vm_map_entry_create: [ internal use only ] | |
246 | * | |
247 | * Allocates a VM map entry for insertion. | |
248 | * No entry fields are filled in. This routine is | |
249 | */ | |
250 | vm_map_entry_t vm_map_entry_create(map) | |
251 | vm_map_t map; | |
252 | { | |
253 | vm_map_entry_t entry; | |
931302a1 | 254 | extern vm_map_t kernel_map, kmem_map, mb_map, pager_map; |
175f072e | 255 | |
931302a1 KM |
256 | if (map == kernel_map || map == kmem_map || map == mb_map || |
257 | map == pager_map) { | |
175f072e KM |
258 | if (entry = kentry_free) |
259 | kentry_free = kentry_free->next; | |
260 | } else | |
261 | MALLOC(entry, vm_map_entry_t, sizeof(struct vm_map_entry), | |
262 | M_VMMAPENT, M_WAITOK); | |
5d7b9ad3 | 263 | if (entry == NULL) |
175f072e KM |
264 | panic("vm_map_entry_create: out of map entries"); |
265 | ||
266 | return(entry); | |
267 | } | |
268 | ||
269 | /* | |
270 | * vm_map_entry_dispose: [ internal use only ] | |
271 | * | |
272 | * Inverse of vm_map_entry_create. | |
273 | */ | |
274 | void vm_map_entry_dispose(map, entry) | |
275 | vm_map_t map; | |
276 | vm_map_entry_t entry; | |
277 | { | |
931302a1 | 278 | extern vm_map_t kernel_map, kmem_map, mb_map, pager_map; |
175f072e | 279 | |
931302a1 KM |
280 | if (map == kernel_map || map == kmem_map || map == mb_map || |
281 | map == pager_map) { | |
175f072e KM |
282 | entry->next = kentry_free; |
283 | kentry_free = entry; | |
284 | } else | |
285 | FREE(entry, M_VMMAPENT); | |
286 | } | |
287 | ||
288 | /* | |
289 | * vm_map_entry_{un,}link: | |
290 | * | |
291 | * Insert/remove entries from maps. | |
292 | */ | |
293 | #define vm_map_entry_link(map, after_where, entry) \ | |
294 | { \ | |
295 | (map)->nentries++; \ | |
296 | (entry)->prev = (after_where); \ | |
297 | (entry)->next = (after_where)->next; \ | |
298 | (entry)->prev->next = (entry); \ | |
299 | (entry)->next->prev = (entry); \ | |
300 | } | |
301 | #define vm_map_entry_unlink(map, entry) \ | |
302 | { \ | |
303 | (map)->nentries--; \ | |
304 | (entry)->next->prev = (entry)->prev; \ | |
305 | (entry)->prev->next = (entry)->next; \ | |
306 | } | |
307 | ||
308 | /* | |
309 | * vm_map_reference: | |
310 | * | |
311 | * Creates another valid reference to the given map. | |
312 | * | |
313 | */ | |
314 | void vm_map_reference(map) | |
315 | register vm_map_t map; | |
316 | { | |
5d7b9ad3 | 317 | if (map == NULL) |
175f072e KM |
318 | return; |
319 | ||
320 | simple_lock(&map->ref_lock); | |
321 | map->ref_count++; | |
322 | simple_unlock(&map->ref_lock); | |
323 | } | |
324 | ||
325 | /* | |
326 | * vm_map_deallocate: | |
327 | * | |
328 | * Removes a reference from the specified map, | |
329 | * destroying it if no references remain. | |
330 | * The map should not be locked. | |
331 | */ | |
332 | void vm_map_deallocate(map) | |
333 | register vm_map_t map; | |
334 | { | |
335 | register int c; | |
336 | ||
5d7b9ad3 | 337 | if (map == NULL) |
175f072e KM |
338 | return; |
339 | ||
340 | simple_lock(&map->ref_lock); | |
341 | c = --map->ref_count; | |
342 | simple_unlock(&map->ref_lock); | |
343 | ||
344 | if (c > 0) { | |
345 | return; | |
346 | } | |
347 | ||
348 | /* | |
349 | * Lock the map, to wait out all other references | |
350 | * to it. | |
351 | */ | |
352 | ||
353 | vm_map_lock(map); | |
354 | ||
355 | (void) vm_map_delete(map, map->min_offset, map->max_offset); | |
356 | ||
357 | pmap_destroy(map->pmap); | |
358 | ||
359 | FREE(map, M_VMMAP); | |
360 | } | |
361 | ||
362 | /* | |
363 | * vm_map_insert: [ internal use only ] | |
364 | * | |
365 | * Inserts the given whole VM object into the target | |
366 | * map at the specified address range. The object's | |
367 | * size should match that of the address range. | |
368 | * | |
369 | * Requires that the map be locked, and leaves it so. | |
370 | */ | |
73506ff8 | 371 | int |
175f072e KM |
372 | vm_map_insert(map, object, offset, start, end) |
373 | vm_map_t map; | |
374 | vm_object_t object; | |
375 | vm_offset_t offset; | |
376 | vm_offset_t start; | |
377 | vm_offset_t end; | |
378 | { | |
379 | register vm_map_entry_t new_entry; | |
380 | register vm_map_entry_t prev_entry; | |
381 | vm_map_entry_t temp_entry; | |
382 | ||
383 | /* | |
384 | * Check that the start and end points are not bogus. | |
385 | */ | |
386 | ||
387 | if ((start < map->min_offset) || (end > map->max_offset) || | |
388 | (start >= end)) | |
389 | return(KERN_INVALID_ADDRESS); | |
390 | ||
391 | /* | |
392 | * Find the entry prior to the proposed | |
393 | * starting address; if it's part of an | |
394 | * existing entry, this range is bogus. | |
395 | */ | |
396 | ||
397 | if (vm_map_lookup_entry(map, start, &temp_entry)) | |
398 | return(KERN_NO_SPACE); | |
399 | ||
400 | prev_entry = temp_entry; | |
401 | ||
402 | /* | |
403 | * Assert that the next entry doesn't overlap the | |
404 | * end point. | |
405 | */ | |
406 | ||
407 | if ((prev_entry->next != &map->header) && | |
408 | (prev_entry->next->start < end)) | |
409 | return(KERN_NO_SPACE); | |
410 | ||
411 | /* | |
412 | * See if we can avoid creating a new entry by | |
413 | * extending one of our neighbors. | |
414 | */ | |
415 | ||
5d7b9ad3 | 416 | if (object == NULL) { |
175f072e KM |
417 | if ((prev_entry != &map->header) && |
418 | (prev_entry->end == start) && | |
419 | (map->is_main_map) && | |
420 | (prev_entry->is_a_map == FALSE) && | |
421 | (prev_entry->is_sub_map == FALSE) && | |
422 | (prev_entry->inheritance == VM_INHERIT_DEFAULT) && | |
423 | (prev_entry->protection == VM_PROT_DEFAULT) && | |
424 | (prev_entry->max_protection == VM_PROT_DEFAULT) && | |
425 | (prev_entry->wired_count == 0)) { | |
426 | ||
427 | if (vm_object_coalesce(prev_entry->object.vm_object, | |
5d7b9ad3 | 428 | NULL, |
175f072e KM |
429 | prev_entry->offset, |
430 | (vm_offset_t) 0, | |
431 | (vm_size_t)(prev_entry->end | |
432 | - prev_entry->start), | |
433 | (vm_size_t)(end - prev_entry->end))) { | |
434 | /* | |
435 | * Coalesced the two objects - can extend | |
436 | * the previous map entry to include the | |
437 | * new range. | |
438 | */ | |
439 | map->size += (end - prev_entry->end); | |
440 | prev_entry->end = end; | |
441 | return(KERN_SUCCESS); | |
442 | } | |
443 | } | |
444 | } | |
445 | ||
446 | /* | |
447 | * Create a new entry | |
448 | */ | |
449 | ||
450 | new_entry = vm_map_entry_create(map); | |
451 | new_entry->start = start; | |
452 | new_entry->end = end; | |
453 | ||
454 | new_entry->is_a_map = FALSE; | |
455 | new_entry->is_sub_map = FALSE; | |
456 | new_entry->object.vm_object = object; | |
457 | new_entry->offset = offset; | |
458 | ||
459 | new_entry->copy_on_write = FALSE; | |
460 | new_entry->needs_copy = FALSE; | |
461 | ||
462 | if (map->is_main_map) { | |
463 | new_entry->inheritance = VM_INHERIT_DEFAULT; | |
464 | new_entry->protection = VM_PROT_DEFAULT; | |
465 | new_entry->max_protection = VM_PROT_DEFAULT; | |
466 | new_entry->wired_count = 0; | |
467 | } | |
468 | ||
469 | /* | |
470 | * Insert the new entry into the list | |
471 | */ | |
472 | ||
473 | vm_map_entry_link(map, prev_entry, new_entry); | |
474 | map->size += new_entry->end - new_entry->start; | |
475 | ||
476 | /* | |
477 | * Update the free space hint | |
478 | */ | |
479 | ||
480 | if ((map->first_free == prev_entry) && (prev_entry->end >= new_entry->start)) | |
481 | map->first_free = new_entry; | |
482 | ||
483 | return(KERN_SUCCESS); | |
484 | } | |
485 | ||
486 | /* | |
487 | * SAVE_HINT: | |
488 | * | |
489 | * Saves the specified entry as the hint for | |
490 | * future lookups. Performs necessary interlocks. | |
491 | */ | |
492 | #define SAVE_HINT(map,value) \ | |
493 | simple_lock(&(map)->hint_lock); \ | |
494 | (map)->hint = (value); \ | |
495 | simple_unlock(&(map)->hint_lock); | |
496 | ||
497 | /* | |
498 | * vm_map_lookup_entry: [ internal use only ] | |
499 | * | |
500 | * Finds the map entry containing (or | |
501 | * immediately preceding) the specified address | |
502 | * in the given map; the entry is returned | |
503 | * in the "entry" parameter. The boolean | |
504 | * result indicates whether the address is | |
505 | * actually contained in the map. | |
506 | */ | |
507 | boolean_t vm_map_lookup_entry(map, address, entry) | |
508 | register vm_map_t map; | |
509 | register vm_offset_t address; | |
510 | vm_map_entry_t *entry; /* OUT */ | |
511 | { | |
512 | register vm_map_entry_t cur; | |
513 | register vm_map_entry_t last; | |
514 | ||
515 | /* | |
516 | * Start looking either from the head of the | |
517 | * list, or from the hint. | |
518 | */ | |
519 | ||
520 | simple_lock(&map->hint_lock); | |
521 | cur = map->hint; | |
522 | simple_unlock(&map->hint_lock); | |
523 | ||
524 | if (cur == &map->header) | |
525 | cur = cur->next; | |
526 | ||
527 | if (address >= cur->start) { | |
528 | /* | |
529 | * Go from hint to end of list. | |
530 | * | |
531 | * But first, make a quick check to see if | |
532 | * we are already looking at the entry we | |
533 | * want (which is usually the case). | |
534 | * Note also that we don't need to save the hint | |
535 | * here... it is the same hint (unless we are | |
536 | * at the header, in which case the hint didn't | |
537 | * buy us anything anyway). | |
538 | */ | |
539 | last = &map->header; | |
540 | if ((cur != last) && (cur->end > address)) { | |
541 | *entry = cur; | |
542 | return(TRUE); | |
543 | } | |
544 | } | |
545 | else { | |
546 | /* | |
547 | * Go from start to hint, *inclusively* | |
548 | */ | |
549 | last = cur->next; | |
550 | cur = map->header.next; | |
551 | } | |
552 | ||
553 | /* | |
554 | * Search linearly | |
555 | */ | |
556 | ||
557 | while (cur != last) { | |
558 | if (cur->end > address) { | |
559 | if (address >= cur->start) { | |
560 | /* | |
561 | * Save this lookup for future | |
562 | * hints, and return | |
563 | */ | |
564 | ||
565 | *entry = cur; | |
566 | SAVE_HINT(map, cur); | |
567 | return(TRUE); | |
568 | } | |
569 | break; | |
570 | } | |
571 | cur = cur->next; | |
572 | } | |
573 | *entry = cur->prev; | |
574 | SAVE_HINT(map, *entry); | |
575 | return(FALSE); | |
576 | } | |
577 | ||
fc8007a4 CT |
578 | /* |
579 | * Find sufficient space for `length' bytes in the given map, starting at | |
580 | * `start'. The map must be locked. Returns 0 on success, 1 on no space. | |
581 | */ | |
582 | int | |
583 | vm_map_findspace(map, start, length, addr) | |
584 | register vm_map_t map; | |
585 | register vm_offset_t start; | |
586 | vm_size_t length; | |
587 | vm_offset_t *addr; | |
588 | { | |
589 | register vm_map_entry_t entry, next; | |
590 | register vm_offset_t end; | |
591 | ||
592 | if (start < map->min_offset) | |
593 | start = map->min_offset; | |
594 | if (start > map->max_offset) | |
595 | return (1); | |
596 | ||
597 | /* | |
598 | * Look for the first possible address; if there's already | |
599 | * something at this address, we have to start after it. | |
600 | */ | |
601 | if (start == map->min_offset) { | |
602 | if ((entry = map->first_free) != &map->header) | |
603 | start = entry->end; | |
604 | } else { | |
605 | vm_map_entry_t tmp; | |
606 | if (vm_map_lookup_entry(map, start, &tmp)) | |
607 | start = tmp->end; | |
608 | entry = tmp; | |
609 | } | |
610 | ||
611 | /* | |
612 | * Look through the rest of the map, trying to fit a new region in | |
613 | * the gap between existing regions, or after the very last region. | |
614 | */ | |
615 | for (;; start = (entry = next)->end) { | |
616 | /* | |
617 | * Find the end of the proposed new region. Be sure we didn't | |
618 | * go beyond the end of the map, or wrap around the address; | |
619 | * if so, we lose. Otherwise, if this is the last entry, or | |
620 | * if the proposed new region fits before the next entry, we | |
621 | * win. | |
622 | */ | |
623 | end = start + length; | |
624 | if (end > map->max_offset || end < start) | |
625 | return (1); | |
626 | next = entry->next; | |
627 | if (next == &map->header || next->start >= end) | |
628 | break; | |
629 | } | |
630 | SAVE_HINT(map, entry); | |
631 | *addr = start; | |
632 | return (0); | |
633 | } | |
634 | ||
175f072e KM |
635 | /* |
636 | * vm_map_find finds an unallocated region in the target address | |
637 | * map with the given length. The search is defined to be | |
638 | * first-fit from the specified address; the region found is | |
639 | * returned in the same parameter. | |
640 | * | |
641 | */ | |
73506ff8 | 642 | int |
175f072e KM |
643 | vm_map_find(map, object, offset, addr, length, find_space) |
644 | vm_map_t map; | |
645 | vm_object_t object; | |
646 | vm_offset_t offset; | |
647 | vm_offset_t *addr; /* IN/OUT */ | |
648 | vm_size_t length; | |
649 | boolean_t find_space; | |
650 | { | |
175f072e | 651 | register vm_offset_t start; |
175f072e KM |
652 | int result; |
653 | ||
654 | start = *addr; | |
175f072e | 655 | vm_map_lock(map); |
175f072e | 656 | if (find_space) { |
fc8007a4 | 657 | if (vm_map_findspace(map, start, length, addr)) { |
175f072e KM |
658 | vm_map_unlock(map); |
659 | return (KERN_NO_SPACE); | |
660 | } | |
fc8007a4 | 661 | start = *addr; |
175f072e | 662 | } |
175f072e | 663 | result = vm_map_insert(map, object, offset, start, start + length); |
175f072e | 664 | vm_map_unlock(map); |
fc8007a4 | 665 | return (result); |
175f072e KM |
666 | } |
667 | ||
668 | /* | |
669 | * vm_map_simplify_entry: [ internal use only ] | |
670 | * | |
671 | * Simplify the given map entry by: | |
672 | * removing extra sharing maps | |
673 | * [XXX maybe later] merging with a neighbor | |
674 | */ | |
675 | void vm_map_simplify_entry(map, entry) | |
676 | vm_map_t map; | |
677 | vm_map_entry_t entry; | |
678 | { | |
679 | #ifdef lint | |
680 | map++; | |
1524bcb8 | 681 | #endif |
175f072e KM |
682 | |
683 | /* | |
684 | * If this entry corresponds to a sharing map, then | |
685 | * see if we can remove the level of indirection. | |
686 | * If it's not a sharing map, then it points to | |
687 | * a VM object, so see if we can merge with either | |
688 | * of our neighbors. | |
689 | */ | |
690 | ||
691 | if (entry->is_sub_map) | |
692 | return; | |
693 | if (entry->is_a_map) { | |
694 | #if 0 | |
695 | vm_map_t my_share_map; | |
696 | int count; | |
697 | ||
698 | my_share_map = entry->object.share_map; | |
699 | simple_lock(&my_share_map->ref_lock); | |
700 | count = my_share_map->ref_count; | |
701 | simple_unlock(&my_share_map->ref_lock); | |
702 | ||
703 | if (count == 1) { | |
704 | /* Can move the region from | |
705 | * entry->start to entry->end (+ entry->offset) | |
706 | * in my_share_map into place of entry. | |
707 | * Later. | |
708 | */ | |
709 | } | |
1524bcb8 | 710 | #endif |
175f072e KM |
711 | } |
712 | else { | |
713 | /* | |
714 | * Try to merge with our neighbors. | |
715 | * | |
716 | * Conditions for merge are: | |
717 | * | |
718 | * 1. entries are adjacent. | |
719 | * 2. both entries point to objects | |
720 | * with null pagers. | |
721 | * | |
722 | * If a merge is possible, we replace the two | |
723 | * entries with a single entry, then merge | |
724 | * the two objects into a single object. | |
725 | * | |
726 | * Now, all that is left to do is write the | |
727 | * code! | |
728 | */ | |
729 | } | |
730 | } | |
731 | ||
732 | /* | |
733 | * vm_map_clip_start: [ internal use only ] | |
734 | * | |
735 | * Asserts that the given entry begins at or after | |
736 | * the specified address; if necessary, | |
737 | * it splits the entry into two. | |
738 | */ | |
739 | #define vm_map_clip_start(map, entry, startaddr) \ | |
740 | { \ | |
741 | if (startaddr > entry->start) \ | |
742 | _vm_map_clip_start(map, entry, startaddr); \ | |
743 | } | |
744 | ||
745 | /* | |
746 | * This routine is called only when it is known that | |
747 | * the entry must be split. | |
748 | */ | |
73506ff8 | 749 | static void _vm_map_clip_start(map, entry, start) |
175f072e KM |
750 | register vm_map_t map; |
751 | register vm_map_entry_t entry; | |
752 | register vm_offset_t start; | |
753 | { | |
754 | register vm_map_entry_t new_entry; | |
755 | ||
756 | /* | |
757 | * See if we can simplify this entry first | |
758 | */ | |
759 | ||
760 | vm_map_simplify_entry(map, entry); | |
761 | ||
762 | /* | |
763 | * Split off the front portion -- | |
764 | * note that we must insert the new | |
765 | * entry BEFORE this one, so that | |
766 | * this entry has the specified starting | |
767 | * address. | |
768 | */ | |
769 | ||
770 | new_entry = vm_map_entry_create(map); | |
771 | *new_entry = *entry; | |
772 | ||
773 | new_entry->end = start; | |
774 | entry->offset += (start - entry->start); | |
775 | entry->start = start; | |
776 | ||
777 | vm_map_entry_link(map, entry->prev, new_entry); | |
778 | ||
779 | if (entry->is_a_map || entry->is_sub_map) | |
780 | vm_map_reference(new_entry->object.share_map); | |
781 | else | |
782 | vm_object_reference(new_entry->object.vm_object); | |
783 | } | |
784 | ||
785 | /* | |
786 | * vm_map_clip_end: [ internal use only ] | |
787 | * | |
788 | * Asserts that the given entry ends at or before | |
789 | * the specified address; if necessary, | |
790 | * it splits the entry into two. | |
791 | */ | |
792 | ||
175f072e KM |
793 | #define vm_map_clip_end(map, entry, endaddr) \ |
794 | { \ | |
795 | if (endaddr < entry->end) \ | |
796 | _vm_map_clip_end(map, entry, endaddr); \ | |
797 | } | |
798 | ||
799 | /* | |
800 | * This routine is called only when it is known that | |
801 | * the entry must be split. | |
802 | */ | |
73506ff8 | 803 | static void _vm_map_clip_end(map, entry, end) |
175f072e KM |
804 | register vm_map_t map; |
805 | register vm_map_entry_t entry; | |
806 | register vm_offset_t end; | |
807 | { | |
808 | register vm_map_entry_t new_entry; | |
809 | ||
810 | /* | |
811 | * Create a new entry and insert it | |
812 | * AFTER the specified entry | |
813 | */ | |
814 | ||
815 | new_entry = vm_map_entry_create(map); | |
816 | *new_entry = *entry; | |
817 | ||
818 | new_entry->start = entry->end = end; | |
819 | new_entry->offset += (end - entry->start); | |
820 | ||
821 | vm_map_entry_link(map, entry, new_entry); | |
822 | ||
823 | if (entry->is_a_map || entry->is_sub_map) | |
824 | vm_map_reference(new_entry->object.share_map); | |
825 | else | |
826 | vm_object_reference(new_entry->object.vm_object); | |
827 | } | |
828 | ||
829 | /* | |
830 | * VM_MAP_RANGE_CHECK: [ internal use only ] | |
831 | * | |
832 | * Asserts that the starting and ending region | |
833 | * addresses fall within the valid range of the map. | |
834 | */ | |
835 | #define VM_MAP_RANGE_CHECK(map, start, end) \ | |
836 | { \ | |
837 | if (start < vm_map_min(map)) \ | |
838 | start = vm_map_min(map); \ | |
839 | if (end > vm_map_max(map)) \ | |
840 | end = vm_map_max(map); \ | |
841 | if (start > end) \ | |
842 | start = end; \ | |
843 | } | |
844 | ||
845 | /* | |
846 | * vm_map_submap: [ kernel use only ] | |
847 | * | |
848 | * Mark the given range as handled by a subordinate map. | |
849 | * | |
850 | * This range must have been created with vm_map_find, | |
851 | * and no other operations may have been performed on this | |
852 | * range prior to calling vm_map_submap. | |
853 | * | |
854 | * Only a limited number of operations can be performed | |
855 | * within this rage after calling vm_map_submap: | |
856 | * vm_fault | |
857 | * [Don't try vm_map_copy!] | |
858 | * | |
859 | * To remove a submapping, one must first remove the | |
860 | * range from the superior map, and then destroy the | |
861 | * submap (if desired). [Better yet, don't try it.] | |
862 | */ | |
73506ff8 | 863 | int |
175f072e KM |
864 | vm_map_submap(map, start, end, submap) |
865 | register vm_map_t map; | |
866 | register vm_offset_t start; | |
867 | register vm_offset_t end; | |
868 | vm_map_t submap; | |
869 | { | |
870 | vm_map_entry_t entry; | |
871 | register int result = KERN_INVALID_ARGUMENT; | |
872 | ||
873 | vm_map_lock(map); | |
874 | ||
875 | VM_MAP_RANGE_CHECK(map, start, end); | |
876 | ||
877 | if (vm_map_lookup_entry(map, start, &entry)) { | |
878 | vm_map_clip_start(map, entry, start); | |
879 | } | |
880 | else | |
881 | entry = entry->next; | |
882 | ||
883 | vm_map_clip_end(map, entry, end); | |
884 | ||
885 | if ((entry->start == start) && (entry->end == end) && | |
886 | (!entry->is_a_map) && | |
5d7b9ad3 | 887 | (entry->object.vm_object == NULL) && |
175f072e KM |
888 | (!entry->copy_on_write)) { |
889 | entry->is_a_map = FALSE; | |
890 | entry->is_sub_map = TRUE; | |
891 | vm_map_reference(entry->object.sub_map = submap); | |
892 | result = KERN_SUCCESS; | |
893 | } | |
894 | vm_map_unlock(map); | |
895 | ||
896 | return(result); | |
897 | } | |
898 | ||
899 | /* | |
900 | * vm_map_protect: | |
901 | * | |
902 | * Sets the protection of the specified address | |
903 | * region in the target map. If "set_max" is | |
904 | * specified, the maximum protection is to be set; | |
905 | * otherwise, only the current protection is affected. | |
906 | */ | |
73506ff8 | 907 | int |
175f072e KM |
908 | vm_map_protect(map, start, end, new_prot, set_max) |
909 | register vm_map_t map; | |
910 | register vm_offset_t start; | |
911 | register vm_offset_t end; | |
912 | register vm_prot_t new_prot; | |
913 | register boolean_t set_max; | |
914 | { | |
915 | register vm_map_entry_t current; | |
916 | vm_map_entry_t entry; | |
917 | ||
918 | vm_map_lock(map); | |
919 | ||
920 | VM_MAP_RANGE_CHECK(map, start, end); | |
921 | ||
922 | if (vm_map_lookup_entry(map, start, &entry)) { | |
923 | vm_map_clip_start(map, entry, start); | |
924 | } | |
925 | else | |
926 | entry = entry->next; | |
927 | ||
928 | /* | |
929 | * Make a first pass to check for protection | |
930 | * violations. | |
931 | */ | |
932 | ||
933 | current = entry; | |
934 | while ((current != &map->header) && (current->start < end)) { | |
935 | if (current->is_sub_map) | |
936 | return(KERN_INVALID_ARGUMENT); | |
937 | if ((new_prot & current->max_protection) != new_prot) { | |
938 | vm_map_unlock(map); | |
939 | return(KERN_PROTECTION_FAILURE); | |
940 | } | |
941 | ||
942 | current = current->next; | |
943 | } | |
944 | ||
945 | /* | |
946 | * Go back and fix up protections. | |
947 | * [Note that clipping is not necessary the second time.] | |
948 | */ | |
949 | ||
950 | current = entry; | |
951 | ||
952 | while ((current != &map->header) && (current->start < end)) { | |
953 | vm_prot_t old_prot; | |
954 | ||
955 | vm_map_clip_end(map, current, end); | |
956 | ||
957 | old_prot = current->protection; | |
958 | if (set_max) | |
959 | current->protection = | |
960 | (current->max_protection = new_prot) & | |
961 | old_prot; | |
962 | else | |
963 | current->protection = new_prot; | |
964 | ||
965 | /* | |
966 | * Update physical map if necessary. | |
967 | * Worry about copy-on-write here -- CHECK THIS XXX | |
968 | */ | |
969 | ||
970 | if (current->protection != old_prot) { | |
971 | ||
972 | #define MASK(entry) ((entry)->copy_on_write ? ~VM_PROT_WRITE : \ | |
973 | VM_PROT_ALL) | |
974 | #define max(a,b) ((a) > (b) ? (a) : (b)) | |
975 | ||
976 | if (current->is_a_map) { | |
977 | vm_map_entry_t share_entry; | |
978 | vm_offset_t share_end; | |
979 | ||
980 | vm_map_lock(current->object.share_map); | |
981 | (void) vm_map_lookup_entry( | |
982 | current->object.share_map, | |
983 | current->offset, | |
984 | &share_entry); | |
985 | share_end = current->offset + | |
986 | (current->end - current->start); | |
987 | while ((share_entry != | |
988 | ¤t->object.share_map->header) && | |
989 | (share_entry->start < share_end)) { | |
990 | ||
991 | pmap_protect(map->pmap, | |
992 | (max(share_entry->start, | |
993 | current->offset) - | |
994 | current->offset + | |
995 | current->start), | |
996 | min(share_entry->end, | |
997 | share_end) - | |
998 | current->offset + | |
999 | current->start, | |
1000 | current->protection & | |
1001 | MASK(share_entry)); | |
1002 | ||
1003 | share_entry = share_entry->next; | |
1004 | } | |
1005 | vm_map_unlock(current->object.share_map); | |
1006 | } | |
1007 | else | |
1008 | pmap_protect(map->pmap, current->start, | |
1009 | current->end, | |
1010 | current->protection & MASK(entry)); | |
1011 | #undef max | |
1012 | #undef MASK | |
1013 | } | |
1014 | current = current->next; | |
1015 | } | |
1016 | ||
1017 | vm_map_unlock(map); | |
1018 | return(KERN_SUCCESS); | |
1019 | } | |
1020 | ||
1021 | /* | |
1022 | * vm_map_inherit: | |
1023 | * | |
1024 | * Sets the inheritance of the specified address | |
1025 | * range in the target map. Inheritance | |
1026 | * affects how the map will be shared with | |
1027 | * child maps at the time of vm_map_fork. | |
1028 | */ | |
73506ff8 | 1029 | int |
175f072e KM |
1030 | vm_map_inherit(map, start, end, new_inheritance) |
1031 | register vm_map_t map; | |
1032 | register vm_offset_t start; | |
1033 | register vm_offset_t end; | |
1034 | register vm_inherit_t new_inheritance; | |
1035 | { | |
1036 | register vm_map_entry_t entry; | |
1037 | vm_map_entry_t temp_entry; | |
1038 | ||
1039 | switch (new_inheritance) { | |
1040 | case VM_INHERIT_NONE: | |
1041 | case VM_INHERIT_COPY: | |
1042 | case VM_INHERIT_SHARE: | |
1043 | break; | |
1044 | default: | |
1045 | return(KERN_INVALID_ARGUMENT); | |
1046 | } | |
1047 | ||
1048 | vm_map_lock(map); | |
1049 | ||
1050 | VM_MAP_RANGE_CHECK(map, start, end); | |
1051 | ||
1052 | if (vm_map_lookup_entry(map, start, &temp_entry)) { | |
1053 | entry = temp_entry; | |
1054 | vm_map_clip_start(map, entry, start); | |
1055 | } | |
1056 | else | |
1057 | entry = temp_entry->next; | |
1058 | ||
1059 | while ((entry != &map->header) && (entry->start < end)) { | |
1060 | vm_map_clip_end(map, entry, end); | |
1061 | ||
1062 | entry->inheritance = new_inheritance; | |
1063 | ||
1064 | entry = entry->next; | |
1065 | } | |
1066 | ||
1067 | vm_map_unlock(map); | |
1068 | return(KERN_SUCCESS); | |
1069 | } | |
1070 | ||
1071 | /* | |
1072 | * vm_map_pageable: | |
1073 | * | |
1074 | * Sets the pageability of the specified address | |
1075 | * range in the target map. Regions specified | |
1076 | * as not pageable require locked-down physical | |
1077 | * memory and physical page maps. | |
1078 | * | |
1079 | * The map must not be locked, but a reference | |
1080 | * must remain to the map throughout the call. | |
1081 | */ | |
73506ff8 | 1082 | int |
175f072e KM |
1083 | vm_map_pageable(map, start, end, new_pageable) |
1084 | register vm_map_t map; | |
1085 | register vm_offset_t start; | |
1086 | register vm_offset_t end; | |
1087 | register boolean_t new_pageable; | |
1088 | { | |
1089 | register vm_map_entry_t entry; | |
b5246c21 | 1090 | vm_map_entry_t start_entry; |
d4f41bb2 MH |
1091 | register vm_offset_t failed; |
1092 | int rv; | |
175f072e KM |
1093 | |
1094 | vm_map_lock(map); | |
1095 | ||
1096 | VM_MAP_RANGE_CHECK(map, start, end); | |
1097 | ||
1098 | /* | |
1099 | * Only one pageability change may take place at one | |
1100 | * time, since vm_fault assumes it will be called | |
1101 | * only once for each wiring/unwiring. Therefore, we | |
1102 | * have to make sure we're actually changing the pageability | |
1103 | * for the entire region. We do so before making any changes. | |
1104 | */ | |
1105 | ||
b5246c21 MH |
1106 | if (vm_map_lookup_entry(map, start, &start_entry) == FALSE) { |
1107 | vm_map_unlock(map); | |
1108 | return(KERN_INVALID_ADDRESS); | |
175f072e | 1109 | } |
b5246c21 | 1110 | entry = start_entry; |
175f072e KM |
1111 | |
1112 | /* | |
1113 | * Actions are rather different for wiring and unwiring, | |
1114 | * so we have two separate cases. | |
1115 | */ | |
1116 | ||
1117 | if (new_pageable) { | |
1118 | ||
b5246c21 MH |
1119 | vm_map_clip_start(map, entry, start); |
1120 | ||
175f072e KM |
1121 | /* |
1122 | * Unwiring. First ensure that the range to be | |
b5246c21 MH |
1123 | * unwired is really wired down and that there |
1124 | * are no holes. | |
175f072e KM |
1125 | */ |
1126 | while ((entry != &map->header) && (entry->start < end)) { | |
1127 | ||
b5246c21 MH |
1128 | if (entry->wired_count == 0 || |
1129 | (entry->end < end && | |
1130 | (entry->next == &map->header || | |
1131 | entry->next->start > entry->end))) { | |
175f072e KM |
1132 | vm_map_unlock(map); |
1133 | return(KERN_INVALID_ARGUMENT); | |
1134 | } | |
1135 | entry = entry->next; | |
1136 | } | |
1137 | ||
1138 | /* | |
1139 | * Now decrement the wiring count for each region. | |
1140 | * If a region becomes completely unwired, | |
1141 | * unwire its physical pages and mappings. | |
1142 | */ | |
1143 | lock_set_recursive(&map->lock); | |
1144 | ||
b5246c21 | 1145 | entry = start_entry; |
175f072e KM |
1146 | while ((entry != &map->header) && (entry->start < end)) { |
1147 | vm_map_clip_end(map, entry, end); | |
1148 | ||
1149 | entry->wired_count--; | |
1150 | if (entry->wired_count == 0) | |
1151 | vm_fault_unwire(map, entry->start, entry->end); | |
1152 | ||
1153 | entry = entry->next; | |
1154 | } | |
1155 | lock_clear_recursive(&map->lock); | |
1156 | } | |
1157 | ||
1158 | else { | |
1159 | /* | |
1160 | * Wiring. We must do this in two passes: | |
1161 | * | |
b5246c21 MH |
1162 | * 1. Holding the write lock, we create any shadow |
1163 | * or zero-fill objects that need to be created. | |
1164 | * Then we clip each map entry to the region to be | |
1165 | * wired and increment its wiring count. We | |
1166 | * create objects before clipping the map entries | |
1167 | * to avoid object proliferation. | |
175f072e KM |
1168 | * |
1169 | * 2. We downgrade to a read lock, and call | |
1170 | * vm_fault_wire to fault in the pages for any | |
1171 | * newly wired area (wired_count is 1). | |
1172 | * | |
1173 | * Downgrading to a read lock for vm_fault_wire avoids | |
1174 | * a possible deadlock with another thread that may have | |
1175 | * faulted on one of the pages to be wired (it would mark | |
1176 | * the page busy, blocking us, then in turn block on the | |
1177 | * map lock that we hold). Because of problems in the | |
1178 | * recursive lock package, we cannot upgrade to a write | |
1179 | * lock in vm_map_lookup. Thus, any actions that require | |
1180 | * the write lock must be done beforehand. Because we | |
1181 | * keep the read lock on the map, the copy-on-write status | |
1182 | * of the entries we modify here cannot change. | |
1183 | */ | |
1184 | ||
1185 | /* | |
1186 | * Pass 1. | |
1187 | */ | |
175f072e | 1188 | while ((entry != &map->header) && (entry->start < end)) { |
b5246c21 | 1189 | #if 0 |
175f072e | 1190 | vm_map_clip_end(map, entry, end); |
b5246c21 MH |
1191 | #endif |
1192 | if (entry->wired_count == 0) { | |
175f072e KM |
1193 | |
1194 | /* | |
1195 | * Perform actions of vm_map_lookup that need | |
1196 | * the write lock on the map: create a shadow | |
1197 | * object for a copy-on-write region, or an | |
1198 | * object for a zero-fill region. | |
1199 | * | |
1200 | * We don't have to do this for entries that | |
1201 | * point to sharing maps, because we won't hold | |
1202 | * the lock on the sharing map. | |
1203 | */ | |
1204 | if (!entry->is_a_map) { | |
1205 | if (entry->needs_copy && | |
1206 | ((entry->protection & VM_PROT_WRITE) != 0)) { | |
1207 | ||
1208 | vm_object_shadow(&entry->object.vm_object, | |
1209 | &entry->offset, | |
1210 | (vm_size_t)(entry->end | |
1211 | - entry->start)); | |
1212 | entry->needs_copy = FALSE; | |
1213 | } | |
5d7b9ad3 | 1214 | else if (entry->object.vm_object == NULL) { |
175f072e KM |
1215 | entry->object.vm_object = |
1216 | vm_object_allocate((vm_size_t)(entry->end | |
1217 | - entry->start)); | |
1218 | entry->offset = (vm_offset_t)0; | |
1219 | } | |
1220 | } | |
1221 | } | |
b5246c21 MH |
1222 | vm_map_clip_start(map, entry, start); |
1223 | vm_map_clip_end(map, entry, end); | |
1224 | entry->wired_count++; | |
175f072e | 1225 | |
b5246c21 MH |
1226 | /* |
1227 | * Check for holes | |
1228 | */ | |
1229 | if (entry->end < end && | |
1230 | (entry->next == &map->header || | |
1231 | entry->next->start > entry->end)) { | |
1232 | /* | |
1233 | * Found one. Object creation actions | |
1234 | * do not need to be undone, but the | |
1235 | * wired counts need to be restored. | |
1236 | */ | |
1237 | while (entry != &map->header && entry->end > start) { | |
1238 | entry->wired_count--; | |
1239 | entry = entry->prev; | |
1240 | } | |
1241 | vm_map_unlock(map); | |
1242 | return(KERN_INVALID_ARGUMENT); | |
1243 | } | |
175f072e KM |
1244 | entry = entry->next; |
1245 | } | |
1246 | ||
1247 | /* | |
1248 | * Pass 2. | |
1249 | */ | |
1250 | ||
1251 | /* | |
1252 | * HACK HACK HACK HACK | |
1253 | * | |
1254 | * If we are wiring in the kernel map or a submap of it, | |
1255 | * unlock the map to avoid deadlocks. We trust that the | |
1256 | * kernel threads are well-behaved, and therefore will | |
1257 | * not do anything destructive to this region of the map | |
1258 | * while we have it unlocked. We cannot trust user threads | |
1259 | * to do the same. | |
1260 | * | |
1261 | * HACK HACK HACK HACK | |
1262 | */ | |
1263 | if (vm_map_pmap(map) == kernel_pmap) { | |
1264 | vm_map_unlock(map); /* trust me ... */ | |
1265 | } | |
1266 | else { | |
1267 | lock_set_recursive(&map->lock); | |
1268 | lock_write_to_read(&map->lock); | |
1269 | } | |
1270 | ||
d4f41bb2 | 1271 | rv = 0; |
b5246c21 | 1272 | entry = start_entry; |
175f072e | 1273 | while (entry != &map->header && entry->start < end) { |
d4f41bb2 MH |
1274 | /* |
1275 | * If vm_fault_wire fails for any page we need to | |
1276 | * undo what has been done. We decrement the wiring | |
1277 | * count for those pages which have not yet been | |
1278 | * wired (now) and unwire those that have (later). | |
1279 | * | |
1280 | * XXX this violates the locking protocol on the map, | |
1281 | * needs to be fixed. | |
1282 | */ | |
1283 | if (rv) | |
1284 | entry->wired_count--; | |
1285 | else if (entry->wired_count == 1) { | |
1286 | rv = vm_fault_wire(map, entry->start, entry->end); | |
1287 | if (rv) { | |
1288 | failed = entry->start; | |
1289 | entry->wired_count--; | |
1290 | } | |
175f072e KM |
1291 | } |
1292 | entry = entry->next; | |
1293 | } | |
1294 | ||
1295 | if (vm_map_pmap(map) == kernel_pmap) { | |
1296 | vm_map_lock(map); | |
1297 | } | |
1298 | else { | |
1299 | lock_clear_recursive(&map->lock); | |
1300 | } | |
d4f41bb2 MH |
1301 | if (rv) { |
1302 | vm_map_unlock(map); | |
1303 | (void) vm_map_pageable(map, start, failed, TRUE); | |
1304 | return(rv); | |
1305 | } | |
175f072e KM |
1306 | } |
1307 | ||
1308 | vm_map_unlock(map); | |
1309 | ||
1310 | return(KERN_SUCCESS); | |
1311 | } | |
1312 | ||
1313 | /* | |
1314 | * vm_map_entry_unwire: [ internal use only ] | |
1315 | * | |
1316 | * Make the region specified by this entry pageable. | |
1317 | * | |
1318 | * The map in question should be locked. | |
1319 | * [This is the reason for this routine's existence.] | |
1320 | */ | |
1321 | void vm_map_entry_unwire(map, entry) | |
1322 | vm_map_t map; | |
1323 | register vm_map_entry_t entry; | |
1324 | { | |
1325 | vm_fault_unwire(map, entry->start, entry->end); | |
1326 | entry->wired_count = 0; | |
1327 | } | |
1328 | ||
1329 | /* | |
1330 | * vm_map_entry_delete: [ internal use only ] | |
1331 | * | |
1332 | * Deallocate the given entry from the target map. | |
1333 | */ | |
1334 | void vm_map_entry_delete(map, entry) | |
1335 | register vm_map_t map; | |
1336 | register vm_map_entry_t entry; | |
1337 | { | |
1338 | if (entry->wired_count != 0) | |
1339 | vm_map_entry_unwire(map, entry); | |
1340 | ||
1341 | vm_map_entry_unlink(map, entry); | |
1342 | map->size -= entry->end - entry->start; | |
1343 | ||
1344 | if (entry->is_a_map || entry->is_sub_map) | |
1345 | vm_map_deallocate(entry->object.share_map); | |
1346 | else | |
1347 | vm_object_deallocate(entry->object.vm_object); | |
1348 | ||
1349 | vm_map_entry_dispose(map, entry); | |
1350 | } | |
1351 | ||
1352 | /* | |
1353 | * vm_map_delete: [ internal use only ] | |
1354 | * | |
1355 | * Deallocates the given address range from the target | |
1356 | * map. | |
1357 | * | |
1358 | * When called with a sharing map, removes pages from | |
1359 | * that region from all physical maps. | |
1360 | */ | |
73506ff8 | 1361 | int |
175f072e KM |
1362 | vm_map_delete(map, start, end) |
1363 | register vm_map_t map; | |
1364 | vm_offset_t start; | |
1365 | register vm_offset_t end; | |
1366 | { | |
1367 | register vm_map_entry_t entry; | |
1368 | vm_map_entry_t first_entry; | |
1369 | ||
1370 | /* | |
1371 | * Find the start of the region, and clip it | |
1372 | */ | |
1373 | ||
1374 | if (!vm_map_lookup_entry(map, start, &first_entry)) | |
1375 | entry = first_entry->next; | |
1376 | else { | |
1377 | entry = first_entry; | |
1378 | vm_map_clip_start(map, entry, start); | |
1379 | ||
1380 | /* | |
1381 | * Fix the lookup hint now, rather than each | |
1382 | * time though the loop. | |
1383 | */ | |
1384 | ||
1385 | SAVE_HINT(map, entry->prev); | |
1386 | } | |
1387 | ||
1388 | /* | |
1389 | * Save the free space hint | |
1390 | */ | |
1391 | ||
1392 | if (map->first_free->start >= start) | |
1393 | map->first_free = entry->prev; | |
1394 | ||
1395 | /* | |
1396 | * Step through all entries in this region | |
1397 | */ | |
1398 | ||
1399 | while ((entry != &map->header) && (entry->start < end)) { | |
1400 | vm_map_entry_t next; | |
1401 | register vm_offset_t s, e; | |
1402 | register vm_object_t object; | |
1403 | ||
1404 | vm_map_clip_end(map, entry, end); | |
1405 | ||
1406 | next = entry->next; | |
1407 | s = entry->start; | |
1408 | e = entry->end; | |
1409 | ||
1410 | /* | |
1411 | * Unwire before removing addresses from the pmap; | |
1412 | * otherwise, unwiring will put the entries back in | |
1413 | * the pmap. | |
1414 | */ | |
1415 | ||
1416 | object = entry->object.vm_object; | |
1417 | if (entry->wired_count != 0) | |
1418 | vm_map_entry_unwire(map, entry); | |
1419 | ||
1420 | /* | |
1421 | * If this is a sharing map, we must remove | |
1422 | * *all* references to this data, since we can't | |
1423 | * find all of the physical maps which are sharing | |
1424 | * it. | |
1425 | */ | |
1426 | ||
1427 | if (object == kernel_object || object == kmem_object) | |
1428 | vm_object_page_remove(object, entry->offset, | |
1429 | entry->offset + (e - s)); | |
1430 | else if (!map->is_main_map) | |
1431 | vm_object_pmap_remove(object, | |
1432 | entry->offset, | |
1433 | entry->offset + (e - s)); | |
1434 | else | |
1435 | pmap_remove(map->pmap, s, e); | |
1436 | ||
1437 | /* | |
1438 | * Delete the entry (which may delete the object) | |
1439 | * only after removing all pmap entries pointing | |
1440 | * to its pages. (Otherwise, its page frames may | |
1441 | * be reallocated, and any modify bits will be | |
1442 | * set in the wrong object!) | |
1443 | */ | |
1444 | ||
1445 | vm_map_entry_delete(map, entry); | |
1446 | entry = next; | |
1447 | } | |
1448 | return(KERN_SUCCESS); | |
1449 | } | |
1450 | ||
1451 | /* | |
1452 | * vm_map_remove: | |
1453 | * | |
1454 | * Remove the given address range from the target map. | |
1455 | * This is the exported form of vm_map_delete. | |
1456 | */ | |
73506ff8 | 1457 | int |
175f072e KM |
1458 | vm_map_remove(map, start, end) |
1459 | register vm_map_t map; | |
1460 | register vm_offset_t start; | |
1461 | register vm_offset_t end; | |
1462 | { | |
1463 | register int result; | |
1464 | ||
1465 | vm_map_lock(map); | |
1466 | VM_MAP_RANGE_CHECK(map, start, end); | |
1467 | result = vm_map_delete(map, start, end); | |
1468 | vm_map_unlock(map); | |
1469 | ||
1470 | return(result); | |
1471 | } | |
1472 | ||
1473 | /* | |
1474 | * vm_map_check_protection: | |
1475 | * | |
1476 | * Assert that the target map allows the specified | |
1477 | * privilege on the entire address region given. | |
1478 | * The entire region must be allocated. | |
1479 | */ | |
1480 | boolean_t vm_map_check_protection(map, start, end, protection) | |
1481 | register vm_map_t map; | |
1482 | register vm_offset_t start; | |
1483 | register vm_offset_t end; | |
1484 | register vm_prot_t protection; | |
1485 | { | |
1486 | register vm_map_entry_t entry; | |
1487 | vm_map_entry_t tmp_entry; | |
1488 | ||
1489 | if (!vm_map_lookup_entry(map, start, &tmp_entry)) { | |
1490 | return(FALSE); | |
1491 | } | |
1492 | ||
1493 | entry = tmp_entry; | |
1494 | ||
1495 | while (start < end) { | |
1496 | if (entry == &map->header) { | |
1497 | return(FALSE); | |
1498 | } | |
1499 | ||
1500 | /* | |
1501 | * No holes allowed! | |
1502 | */ | |
1503 | ||
1504 | if (start < entry->start) { | |
1505 | return(FALSE); | |
1506 | } | |
1507 | ||
1508 | /* | |
1509 | * Check protection associated with entry. | |
1510 | */ | |
1511 | ||
1512 | if ((entry->protection & protection) != protection) { | |
1513 | return(FALSE); | |
1514 | } | |
1515 | ||
1516 | /* go to next entry */ | |
1517 | ||
1518 | start = entry->end; | |
1519 | entry = entry->next; | |
1520 | } | |
1521 | return(TRUE); | |
1522 | } | |
1523 | ||
1524 | /* | |
1525 | * vm_map_copy_entry: | |
1526 | * | |
1527 | * Copies the contents of the source entry to the destination | |
1528 | * entry. The entries *must* be aligned properly. | |
1529 | */ | |
1530 | void vm_map_copy_entry(src_map, dst_map, src_entry, dst_entry) | |
1531 | vm_map_t src_map, dst_map; | |
1532 | register vm_map_entry_t src_entry, dst_entry; | |
1533 | { | |
1534 | vm_object_t temp_object; | |
1535 | ||
1536 | if (src_entry->is_sub_map || dst_entry->is_sub_map) | |
1537 | return; | |
1538 | ||
5d7b9ad3 | 1539 | if (dst_entry->object.vm_object != NULL && |
224765a4 | 1540 | (dst_entry->object.vm_object->flags & OBJ_INTERNAL) == 0) |
175f072e KM |
1541 | printf("vm_map_copy_entry: copying over permanent data!\n"); |
1542 | ||
1543 | /* | |
1544 | * If our destination map was wired down, | |
1545 | * unwire it now. | |
1546 | */ | |
1547 | ||
1548 | if (dst_entry->wired_count != 0) | |
1549 | vm_map_entry_unwire(dst_map, dst_entry); | |
1550 | ||
1551 | /* | |
1552 | * If we're dealing with a sharing map, we | |
1553 | * must remove the destination pages from | |
1554 | * all maps (since we cannot know which maps | |
1555 | * this sharing map belongs in). | |
1556 | */ | |
1557 | ||
1558 | if (dst_map->is_main_map) | |
1559 | pmap_remove(dst_map->pmap, dst_entry->start, dst_entry->end); | |
1560 | else | |
1561 | vm_object_pmap_remove(dst_entry->object.vm_object, | |
1562 | dst_entry->offset, | |
1563 | dst_entry->offset + | |
1564 | (dst_entry->end - dst_entry->start)); | |
1565 | ||
1566 | if (src_entry->wired_count == 0) { | |
1567 | ||
1568 | boolean_t src_needs_copy; | |
1569 | ||
1570 | /* | |
1571 | * If the source entry is marked needs_copy, | |
1572 | * it is already write-protected. | |
1573 | */ | |
1574 | if (!src_entry->needs_copy) { | |
1575 | ||
1576 | boolean_t su; | |
1577 | ||
1578 | /* | |
1579 | * If the source entry has only one mapping, | |
1580 | * we can just protect the virtual address | |
1581 | * range. | |
1582 | */ | |
1583 | if (!(su = src_map->is_main_map)) { | |
1584 | simple_lock(&src_map->ref_lock); | |
1585 | su = (src_map->ref_count == 1); | |
1586 | simple_unlock(&src_map->ref_lock); | |
1587 | } | |
1588 | ||
1589 | if (su) { | |
1590 | pmap_protect(src_map->pmap, | |
1591 | src_entry->start, | |
1592 | src_entry->end, | |
1593 | src_entry->protection & ~VM_PROT_WRITE); | |
1594 | } | |
1595 | else { | |
1596 | vm_object_pmap_copy(src_entry->object.vm_object, | |
1597 | src_entry->offset, | |
1598 | src_entry->offset + (src_entry->end | |
1599 | -src_entry->start)); | |
1600 | } | |
1601 | } | |
1602 | ||
1603 | /* | |
1604 | * Make a copy of the object. | |
1605 | */ | |
1606 | temp_object = dst_entry->object.vm_object; | |
1607 | vm_object_copy(src_entry->object.vm_object, | |
1608 | src_entry->offset, | |
1609 | (vm_size_t)(src_entry->end - | |
1610 | src_entry->start), | |
1611 | &dst_entry->object.vm_object, | |
1612 | &dst_entry->offset, | |
1613 | &src_needs_copy); | |
1614 | /* | |
1615 | * If we didn't get a copy-object now, mark the | |
1616 | * source map entry so that a shadow will be created | |
1617 | * to hold its changed pages. | |
1618 | */ | |
1619 | if (src_needs_copy) | |
1620 | src_entry->needs_copy = TRUE; | |
1621 | ||
1622 | /* | |
1623 | * The destination always needs to have a shadow | |
1624 | * created. | |
1625 | */ | |
1626 | dst_entry->needs_copy = TRUE; | |
1627 | ||
1628 | /* | |
1629 | * Mark the entries copy-on-write, so that write-enabling | |
1630 | * the entry won't make copy-on-write pages writable. | |
1631 | */ | |
1632 | src_entry->copy_on_write = TRUE; | |
1633 | dst_entry->copy_on_write = TRUE; | |
1634 | /* | |
1635 | * Get rid of the old object. | |
1636 | */ | |
1637 | vm_object_deallocate(temp_object); | |
1638 | ||
1639 | pmap_copy(dst_map->pmap, src_map->pmap, dst_entry->start, | |
1640 | dst_entry->end - dst_entry->start, src_entry->start); | |
1641 | } | |
1642 | else { | |
1643 | /* | |
1644 | * Of course, wired down pages can't be set copy-on-write. | |
1645 | * Cause wired pages to be copied into the new | |
1646 | * map by simulating faults (the new pages are | |
1647 | * pageable) | |
1648 | */ | |
1649 | vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry); | |
1650 | } | |
1651 | } | |
1652 | ||
1653 | /* | |
1654 | * vm_map_copy: | |
1655 | * | |
1656 | * Perform a virtual memory copy from the source | |
1657 | * address map/range to the destination map/range. | |
1658 | * | |
1659 | * If src_destroy or dst_alloc is requested, | |
1660 | * the source and destination regions should be | |
1661 | * disjoint, not only in the top-level map, but | |
1662 | * in the sharing maps as well. [The best way | |
1663 | * to guarantee this is to use a new intermediate | |
1664 | * map to make copies. This also reduces map | |
1665 | * fragmentation.] | |
1666 | */ | |
73506ff8 | 1667 | int |
175f072e KM |
1668 | vm_map_copy(dst_map, src_map, |
1669 | dst_addr, len, src_addr, | |
1670 | dst_alloc, src_destroy) | |
1671 | vm_map_t dst_map; | |
1672 | vm_map_t src_map; | |
1673 | vm_offset_t dst_addr; | |
1674 | vm_size_t len; | |
1675 | vm_offset_t src_addr; | |
1676 | boolean_t dst_alloc; | |
1677 | boolean_t src_destroy; | |
1678 | { | |
1679 | register | |
1680 | vm_map_entry_t src_entry; | |
1681 | register | |
1682 | vm_map_entry_t dst_entry; | |
1683 | vm_map_entry_t tmp_entry; | |
1684 | vm_offset_t src_start; | |
1685 | vm_offset_t src_end; | |
1686 | vm_offset_t dst_start; | |
1687 | vm_offset_t dst_end; | |
1688 | vm_offset_t src_clip; | |
1689 | vm_offset_t dst_clip; | |
1690 | int result; | |
1691 | boolean_t old_src_destroy; | |
1692 | ||
1693 | /* | |
1694 | * XXX While we figure out why src_destroy screws up, | |
1695 | * we'll do it by explicitly vm_map_delete'ing at the end. | |
1696 | */ | |
1697 | ||
1698 | old_src_destroy = src_destroy; | |
1699 | src_destroy = FALSE; | |
1700 | ||
1701 | /* | |
1702 | * Compute start and end of region in both maps | |
1703 | */ | |
1704 | ||
1705 | src_start = src_addr; | |
1706 | src_end = src_start + len; | |
1707 | dst_start = dst_addr; | |
1708 | dst_end = dst_start + len; | |
1709 | ||
1710 | /* | |
1711 | * Check that the region can exist in both source | |
1712 | * and destination. | |
1713 | */ | |
1714 | ||
1715 | if ((dst_end < dst_start) || (src_end < src_start)) | |
1716 | return(KERN_NO_SPACE); | |
1717 | ||
1718 | /* | |
1719 | * Lock the maps in question -- we avoid deadlock | |
1720 | * by ordering lock acquisition by map value | |
1721 | */ | |
1722 | ||
1723 | if (src_map == dst_map) { | |
1724 | vm_map_lock(src_map); | |
1725 | } | |
1726 | else if ((int) src_map < (int) dst_map) { | |
1727 | vm_map_lock(src_map); | |
1728 | vm_map_lock(dst_map); | |
1729 | } else { | |
1730 | vm_map_lock(dst_map); | |
1731 | vm_map_lock(src_map); | |
1732 | } | |
1733 | ||
1734 | result = KERN_SUCCESS; | |
1735 | ||
1736 | /* | |
1737 | * Check protections... source must be completely readable and | |
1738 | * destination must be completely writable. [Note that if we're | |
1739 | * allocating the destination region, we don't have to worry | |
1740 | * about protection, but instead about whether the region | |
1741 | * exists.] | |
1742 | */ | |
1743 | ||
1744 | if (src_map->is_main_map && dst_map->is_main_map) { | |
1745 | if (!vm_map_check_protection(src_map, src_start, src_end, | |
1746 | VM_PROT_READ)) { | |
1747 | result = KERN_PROTECTION_FAILURE; | |
1748 | goto Return; | |
1749 | } | |
1750 | ||
1751 | if (dst_alloc) { | |
1752 | /* XXX Consider making this a vm_map_find instead */ | |
5d7b9ad3 | 1753 | if ((result = vm_map_insert(dst_map, NULL, |
175f072e KM |
1754 | (vm_offset_t) 0, dst_start, dst_end)) != KERN_SUCCESS) |
1755 | goto Return; | |
1756 | } | |
1757 | else if (!vm_map_check_protection(dst_map, dst_start, dst_end, | |
1758 | VM_PROT_WRITE)) { | |
1759 | result = KERN_PROTECTION_FAILURE; | |
1760 | goto Return; | |
1761 | } | |
1762 | } | |
1763 | ||
1764 | /* | |
1765 | * Find the start entries and clip. | |
1766 | * | |
1767 | * Note that checking protection asserts that the | |
1768 | * lookup cannot fail. | |
1769 | * | |
1770 | * Also note that we wait to do the second lookup | |
1771 | * until we have done the first clip, as the clip | |
1772 | * may affect which entry we get! | |
1773 | */ | |
1774 | ||
1775 | (void) vm_map_lookup_entry(src_map, src_addr, &tmp_entry); | |
1776 | src_entry = tmp_entry; | |
1777 | vm_map_clip_start(src_map, src_entry, src_start); | |
1778 | ||
1779 | (void) vm_map_lookup_entry(dst_map, dst_addr, &tmp_entry); | |
1780 | dst_entry = tmp_entry; | |
1781 | vm_map_clip_start(dst_map, dst_entry, dst_start); | |
1782 | ||
1783 | /* | |
1784 | * If both source and destination entries are the same, | |
1785 | * retry the first lookup, as it may have changed. | |
1786 | */ | |
1787 | ||
1788 | if (src_entry == dst_entry) { | |
1789 | (void) vm_map_lookup_entry(src_map, src_addr, &tmp_entry); | |
1790 | src_entry = tmp_entry; | |
1791 | } | |
1792 | ||
1793 | /* | |
1794 | * If source and destination entries are still the same, | |
1795 | * a null copy is being performed. | |
1796 | */ | |
1797 | ||
1798 | if (src_entry == dst_entry) | |
1799 | goto Return; | |
1800 | ||
1801 | /* | |
1802 | * Go through entries until we get to the end of the | |
1803 | * region. | |
1804 | */ | |
1805 | ||
1806 | while (src_start < src_end) { | |
1807 | /* | |
1808 | * Clip the entries to the endpoint of the entire region. | |
1809 | */ | |
1810 | ||
1811 | vm_map_clip_end(src_map, src_entry, src_end); | |
1812 | vm_map_clip_end(dst_map, dst_entry, dst_end); | |
1813 | ||
1814 | /* | |
1815 | * Clip each entry to the endpoint of the other entry. | |
1816 | */ | |
1817 | ||
1818 | src_clip = src_entry->start + (dst_entry->end - dst_entry->start); | |
1819 | vm_map_clip_end(src_map, src_entry, src_clip); | |
1820 | ||
1821 | dst_clip = dst_entry->start + (src_entry->end - src_entry->start); | |
1822 | vm_map_clip_end(dst_map, dst_entry, dst_clip); | |
1823 | ||
1824 | /* | |
1825 | * Both entries now match in size and relative endpoints. | |
1826 | * | |
1827 | * If both entries refer to a VM object, we can | |
1828 | * deal with them now. | |
1829 | */ | |
1830 | ||
1831 | if (!src_entry->is_a_map && !dst_entry->is_a_map) { | |
1832 | vm_map_copy_entry(src_map, dst_map, src_entry, | |
1833 | dst_entry); | |
1834 | } | |
1835 | else { | |
1836 | register vm_map_t new_dst_map; | |
1837 | vm_offset_t new_dst_start; | |
1838 | vm_size_t new_size; | |
1839 | vm_map_t new_src_map; | |
1840 | vm_offset_t new_src_start; | |
1841 | ||
1842 | /* | |
1843 | * We have to follow at least one sharing map. | |
1844 | */ | |
1845 | ||
1846 | new_size = (dst_entry->end - dst_entry->start); | |
1847 | ||
1848 | if (src_entry->is_a_map) { | |
1849 | new_src_map = src_entry->object.share_map; | |
1850 | new_src_start = src_entry->offset; | |
1851 | } | |
1852 | else { | |
1853 | new_src_map = src_map; | |
1854 | new_src_start = src_entry->start; | |
1855 | lock_set_recursive(&src_map->lock); | |
1856 | } | |
1857 | ||
1858 | if (dst_entry->is_a_map) { | |
1859 | vm_offset_t new_dst_end; | |
1860 | ||
1861 | new_dst_map = dst_entry->object.share_map; | |
1862 | new_dst_start = dst_entry->offset; | |
1863 | ||
1864 | /* | |
1865 | * Since the destination sharing entries | |
1866 | * will be merely deallocated, we can | |
1867 | * do that now, and replace the region | |
1868 | * with a null object. [This prevents | |
1869 | * splitting the source map to match | |
1870 | * the form of the destination map.] | |
1871 | * Note that we can only do so if the | |
1872 | * source and destination do not overlap. | |
1873 | */ | |
1874 | ||
1875 | new_dst_end = new_dst_start + new_size; | |
1876 | ||
1877 | if (new_dst_map != new_src_map) { | |
1878 | vm_map_lock(new_dst_map); | |
1879 | (void) vm_map_delete(new_dst_map, | |
1880 | new_dst_start, | |
1881 | new_dst_end); | |
1882 | (void) vm_map_insert(new_dst_map, | |
5d7b9ad3 | 1883 | NULL, |
175f072e KM |
1884 | (vm_offset_t) 0, |
1885 | new_dst_start, | |
1886 | new_dst_end); | |
1887 | vm_map_unlock(new_dst_map); | |
1888 | } | |
1889 | } | |
1890 | else { | |
1891 | new_dst_map = dst_map; | |
1892 | new_dst_start = dst_entry->start; | |
1893 | lock_set_recursive(&dst_map->lock); | |
1894 | } | |
1895 | ||
1896 | /* | |
1897 | * Recursively copy the sharing map. | |
1898 | */ | |
1899 | ||
1900 | (void) vm_map_copy(new_dst_map, new_src_map, | |
1901 | new_dst_start, new_size, new_src_start, | |
1902 | FALSE, FALSE); | |
1903 | ||
1904 | if (dst_map == new_dst_map) | |
1905 | lock_clear_recursive(&dst_map->lock); | |
1906 | if (src_map == new_src_map) | |
1907 | lock_clear_recursive(&src_map->lock); | |
1908 | } | |
1909 | ||
1910 | /* | |
1911 | * Update variables for next pass through the loop. | |
1912 | */ | |
1913 | ||
1914 | src_start = src_entry->end; | |
1915 | src_entry = src_entry->next; | |
1916 | dst_start = dst_entry->end; | |
1917 | dst_entry = dst_entry->next; | |
1918 | ||
1919 | /* | |
1920 | * If the source is to be destroyed, here is the | |
1921 | * place to do it. | |
1922 | */ | |
1923 | ||
1924 | if (src_destroy && src_map->is_main_map && | |
1925 | dst_map->is_main_map) | |
1926 | vm_map_entry_delete(src_map, src_entry->prev); | |
1927 | } | |
1928 | ||
1929 | /* | |
1930 | * Update the physical maps as appropriate | |
1931 | */ | |
1932 | ||
1933 | if (src_map->is_main_map && dst_map->is_main_map) { | |
1934 | if (src_destroy) | |
1935 | pmap_remove(src_map->pmap, src_addr, src_addr + len); | |
1936 | } | |
1937 | ||
1938 | /* | |
1939 | * Unlock the maps | |
1940 | */ | |
1941 | ||
1942 | Return: ; | |
1943 | ||
1944 | if (old_src_destroy) | |
1945 | vm_map_delete(src_map, src_addr, src_addr + len); | |
1946 | ||
1947 | vm_map_unlock(src_map); | |
1948 | if (src_map != dst_map) | |
1949 | vm_map_unlock(dst_map); | |
1950 | ||
1951 | return(result); | |
1952 | } | |
1953 | ||
1954 | /* | |
5d7b9ad3 MK |
1955 | * vmspace_fork: |
1956 | * Create a new process vmspace structure and vm_map | |
1957 | * based on those of an existing process. The new map | |
1958 | * is based on the old map, according to the inheritance | |
1959 | * values on the regions in that map. | |
175f072e | 1960 | * |
5d7b9ad3 | 1961 | * The source map must not be locked. |
175f072e | 1962 | */ |
5d7b9ad3 MK |
1963 | struct vmspace * |
1964 | vmspace_fork(vm1) | |
1965 | register struct vmspace *vm1; | |
175f072e | 1966 | { |
5d7b9ad3 MK |
1967 | register struct vmspace *vm2; |
1968 | vm_map_t old_map = &vm1->vm_map; | |
175f072e KM |
1969 | vm_map_t new_map; |
1970 | vm_map_entry_t old_entry; | |
1971 | vm_map_entry_t new_entry; | |
1972 | pmap_t new_pmap; | |
1973 | ||
1974 | vm_map_lock(old_map); | |
1975 | ||
5d7b9ad3 MK |
1976 | vm2 = vmspace_alloc(old_map->min_offset, old_map->max_offset, |
1977 | old_map->entries_pageable); | |
1978 | bcopy(&vm1->vm_startcopy, &vm2->vm_startcopy, | |
1979 | (caddr_t) (vm1 + 1) - (caddr_t) &vm1->vm_startcopy); | |
1980 | new_pmap = &vm2->vm_pmap; /* XXX */ | |
1981 | new_map = &vm2->vm_map; /* XXX */ | |
175f072e KM |
1982 | |
1983 | old_entry = old_map->header.next; | |
1984 | ||
1985 | while (old_entry != &old_map->header) { | |
1986 | if (old_entry->is_sub_map) | |
1987 | panic("vm_map_fork: encountered a submap"); | |
1988 | ||
1989 | switch (old_entry->inheritance) { | |
1990 | case VM_INHERIT_NONE: | |
1991 | break; | |
1992 | ||
1993 | case VM_INHERIT_SHARE: | |
1994 | /* | |
1995 | * If we don't already have a sharing map: | |
1996 | */ | |
1997 | ||
1998 | if (!old_entry->is_a_map) { | |
1999 | vm_map_t new_share_map; | |
2000 | vm_map_entry_t new_share_entry; | |
2001 | ||
2002 | /* | |
2003 | * Create a new sharing map | |
2004 | */ | |
2005 | ||
5d7b9ad3 | 2006 | new_share_map = vm_map_create(NULL, |
175f072e KM |
2007 | old_entry->start, |
2008 | old_entry->end, | |
2009 | TRUE); | |
2010 | new_share_map->is_main_map = FALSE; | |
2011 | ||
2012 | /* | |
2013 | * Create the only sharing entry from the | |
2014 | * old task map entry. | |
2015 | */ | |
2016 | ||
2017 | new_share_entry = | |
2018 | vm_map_entry_create(new_share_map); | |
2019 | *new_share_entry = *old_entry; | |
b5246c21 | 2020 | new_share_entry->wired_count = 0; |
175f072e KM |
2021 | |
2022 | /* | |
2023 | * Insert the entry into the new sharing | |
2024 | * map | |
2025 | */ | |
2026 | ||
2027 | vm_map_entry_link(new_share_map, | |
2028 | new_share_map->header.prev, | |
2029 | new_share_entry); | |
2030 | ||
2031 | /* | |
2032 | * Fix up the task map entry to refer | |
2033 | * to the sharing map now. | |
2034 | */ | |
2035 | ||
2036 | old_entry->is_a_map = TRUE; | |
2037 | old_entry->object.share_map = new_share_map; | |
2038 | old_entry->offset = old_entry->start; | |
2039 | } | |
2040 | ||
2041 | /* | |
2042 | * Clone the entry, referencing the sharing map. | |
2043 | */ | |
2044 | ||
2045 | new_entry = vm_map_entry_create(new_map); | |
2046 | *new_entry = *old_entry; | |
b5246c21 | 2047 | new_entry->wired_count = 0; |
175f072e KM |
2048 | vm_map_reference(new_entry->object.share_map); |
2049 | ||
2050 | /* | |
2051 | * Insert the entry into the new map -- we | |
2052 | * know we're inserting at the end of the new | |
2053 | * map. | |
2054 | */ | |
2055 | ||
2056 | vm_map_entry_link(new_map, new_map->header.prev, | |
2057 | new_entry); | |
2058 | ||
2059 | /* | |
2060 | * Update the physical map | |
2061 | */ | |
2062 | ||
2063 | pmap_copy(new_map->pmap, old_map->pmap, | |
2064 | new_entry->start, | |
2065 | (old_entry->end - old_entry->start), | |
2066 | old_entry->start); | |
2067 | break; | |
2068 | ||
2069 | case VM_INHERIT_COPY: | |
2070 | /* | |
2071 | * Clone the entry and link into the map. | |
2072 | */ | |
2073 | ||
2074 | new_entry = vm_map_entry_create(new_map); | |
2075 | *new_entry = *old_entry; | |
2076 | new_entry->wired_count = 0; | |
5d7b9ad3 | 2077 | new_entry->object.vm_object = NULL; |
175f072e KM |
2078 | new_entry->is_a_map = FALSE; |
2079 | vm_map_entry_link(new_map, new_map->header.prev, | |
2080 | new_entry); | |
2081 | if (old_entry->is_a_map) { | |
2082 | int check; | |
2083 | ||
2084 | check = vm_map_copy(new_map, | |
2085 | old_entry->object.share_map, | |
2086 | new_entry->start, | |
2087 | (vm_size_t)(new_entry->end - | |
2088 | new_entry->start), | |
2089 | old_entry->offset, | |
2090 | FALSE, FALSE); | |
2091 | if (check != KERN_SUCCESS) | |
2092 | printf("vm_map_fork: copy in share_map region failed\n"); | |
2093 | } | |
2094 | else { | |
2095 | vm_map_copy_entry(old_map, new_map, old_entry, | |
2096 | new_entry); | |
2097 | } | |
2098 | break; | |
2099 | } | |
2100 | old_entry = old_entry->next; | |
2101 | } | |
2102 | ||
2103 | new_map->size = old_map->size; | |
2104 | vm_map_unlock(old_map); | |
2105 | ||
5d7b9ad3 | 2106 | return(vm2); |
175f072e KM |
2107 | } |
2108 | ||
2109 | /* | |
2110 | * vm_map_lookup: | |
2111 | * | |
2112 | * Finds the VM object, offset, and | |
2113 | * protection for a given virtual address in the | |
2114 | * specified map, assuming a page fault of the | |
2115 | * type specified. | |
2116 | * | |
2117 | * Leaves the map in question locked for read; return | |
2118 | * values are guaranteed until a vm_map_lookup_done | |
2119 | * call is performed. Note that the map argument | |
2120 | * is in/out; the returned map must be used in | |
2121 | * the call to vm_map_lookup_done. | |
2122 | * | |
2123 | * A handle (out_entry) is returned for use in | |
2124 | * vm_map_lookup_done, to make that fast. | |
2125 | * | |
2126 | * If a lookup is requested with "write protection" | |
2127 | * specified, the map may be changed to perform virtual | |
2128 | * copying operations, although the data referenced will | |
2129 | * remain the same. | |
2130 | */ | |
73506ff8 | 2131 | int |
175f072e KM |
2132 | vm_map_lookup(var_map, vaddr, fault_type, out_entry, |
2133 | object, offset, out_prot, wired, single_use) | |
2134 | vm_map_t *var_map; /* IN/OUT */ | |
2135 | register vm_offset_t vaddr; | |
2136 | register vm_prot_t fault_type; | |
2137 | ||
2138 | vm_map_entry_t *out_entry; /* OUT */ | |
2139 | vm_object_t *object; /* OUT */ | |
2140 | vm_offset_t *offset; /* OUT */ | |
2141 | vm_prot_t *out_prot; /* OUT */ | |
2142 | boolean_t *wired; /* OUT */ | |
2143 | boolean_t *single_use; /* OUT */ | |
2144 | { | |
2145 | vm_map_t share_map; | |
2146 | vm_offset_t share_offset; | |
2147 | register vm_map_entry_t entry; | |
2148 | register vm_map_t map = *var_map; | |
2149 | register vm_prot_t prot; | |
2150 | register boolean_t su; | |
2151 | ||
2152 | RetryLookup: ; | |
2153 | ||
2154 | /* | |
2155 | * Lookup the faulting address. | |
2156 | */ | |
2157 | ||
2158 | vm_map_lock_read(map); | |
2159 | ||
2160 | #define RETURN(why) \ | |
2161 | { \ | |
2162 | vm_map_unlock_read(map); \ | |
2163 | return(why); \ | |
2164 | } | |
2165 | ||
2166 | /* | |
2167 | * If the map has an interesting hint, try it before calling | |
2168 | * full blown lookup routine. | |
2169 | */ | |
2170 | ||
2171 | simple_lock(&map->hint_lock); | |
2172 | entry = map->hint; | |
2173 | simple_unlock(&map->hint_lock); | |
2174 | ||
2175 | *out_entry = entry; | |
2176 | ||
2177 | if ((entry == &map->header) || | |
2178 | (vaddr < entry->start) || (vaddr >= entry->end)) { | |
2179 | vm_map_entry_t tmp_entry; | |
2180 | ||
2181 | /* | |
2182 | * Entry was either not a valid hint, or the vaddr | |
2183 | * was not contained in the entry, so do a full lookup. | |
2184 | */ | |
2185 | if (!vm_map_lookup_entry(map, vaddr, &tmp_entry)) | |
2186 | RETURN(KERN_INVALID_ADDRESS); | |
2187 | ||
2188 | entry = tmp_entry; | |
2189 | *out_entry = entry; | |
2190 | } | |
2191 | ||
2192 | /* | |
2193 | * Handle submaps. | |
2194 | */ | |
2195 | ||
2196 | if (entry->is_sub_map) { | |
2197 | vm_map_t old_map = map; | |
2198 | ||
2199 | *var_map = map = entry->object.sub_map; | |
2200 | vm_map_unlock_read(old_map); | |
2201 | goto RetryLookup; | |
2202 | } | |
2203 | ||
2204 | /* | |
2205 | * Check whether this task is allowed to have | |
2206 | * this page. | |
2207 | */ | |
2208 | ||
2209 | prot = entry->protection; | |
2210 | if ((fault_type & (prot)) != fault_type) | |
2211 | RETURN(KERN_PROTECTION_FAILURE); | |
2212 | ||
2213 | /* | |
2214 | * If this page is not pageable, we have to get | |
2215 | * it for all possible accesses. | |
2216 | */ | |
2217 | ||
2218 | if (*wired = (entry->wired_count != 0)) | |
2219 | prot = fault_type = entry->protection; | |
2220 | ||
2221 | /* | |
2222 | * If we don't already have a VM object, track | |
2223 | * it down. | |
2224 | */ | |
2225 | ||
2226 | if (su = !entry->is_a_map) { | |
2227 | share_map = map; | |
2228 | share_offset = vaddr; | |
2229 | } | |
2230 | else { | |
2231 | vm_map_entry_t share_entry; | |
2232 | ||
2233 | /* | |
2234 | * Compute the sharing map, and offset into it. | |
2235 | */ | |
2236 | ||
2237 | share_map = entry->object.share_map; | |
2238 | share_offset = (vaddr - entry->start) + entry->offset; | |
2239 | ||
2240 | /* | |
2241 | * Look for the backing store object and offset | |
2242 | */ | |
2243 | ||
2244 | vm_map_lock_read(share_map); | |
2245 | ||
2246 | if (!vm_map_lookup_entry(share_map, share_offset, | |
2247 | &share_entry)) { | |
2248 | vm_map_unlock_read(share_map); | |
2249 | RETURN(KERN_INVALID_ADDRESS); | |
2250 | } | |
2251 | entry = share_entry; | |
2252 | } | |
2253 | ||
2254 | /* | |
2255 | * If the entry was copy-on-write, we either ... | |
2256 | */ | |
2257 | ||
2258 | if (entry->needs_copy) { | |
2259 | /* | |
2260 | * If we want to write the page, we may as well | |
2261 | * handle that now since we've got the sharing | |
2262 | * map locked. | |
2263 | * | |
2264 | * If we don't need to write the page, we just | |
2265 | * demote the permissions allowed. | |
2266 | */ | |
2267 | ||
2268 | if (fault_type & VM_PROT_WRITE) { | |
2269 | /* | |
2270 | * Make a new object, and place it in the | |
2271 | * object chain. Note that no new references | |
2272 | * have appeared -- one just moved from the | |
2273 | * share map to the new object. | |
2274 | */ | |
2275 | ||
2276 | if (lock_read_to_write(&share_map->lock)) { | |
2277 | if (share_map != map) | |
2278 | vm_map_unlock_read(map); | |
2279 | goto RetryLookup; | |
2280 | } | |
2281 | ||
2282 | vm_object_shadow( | |
2283 | &entry->object.vm_object, | |
2284 | &entry->offset, | |
2285 | (vm_size_t) (entry->end - entry->start)); | |
2286 | ||
2287 | entry->needs_copy = FALSE; | |
2288 | ||
2289 | lock_write_to_read(&share_map->lock); | |
2290 | } | |
2291 | else { | |
2292 | /* | |
2293 | * We're attempting to read a copy-on-write | |
2294 | * page -- don't allow writes. | |
2295 | */ | |
2296 | ||
2297 | prot &= (~VM_PROT_WRITE); | |
2298 | } | |
2299 | } | |
2300 | ||
2301 | /* | |
2302 | * Create an object if necessary. | |
2303 | */ | |
5d7b9ad3 | 2304 | if (entry->object.vm_object == NULL) { |
175f072e KM |
2305 | |
2306 | if (lock_read_to_write(&share_map->lock)) { | |
2307 | if (share_map != map) | |
2308 | vm_map_unlock_read(map); | |
2309 | goto RetryLookup; | |
2310 | } | |
2311 | ||
2312 | entry->object.vm_object = vm_object_allocate( | |
2313 | (vm_size_t)(entry->end - entry->start)); | |
2314 | entry->offset = 0; | |
2315 | lock_write_to_read(&share_map->lock); | |
2316 | } | |
2317 | ||
2318 | /* | |
2319 | * Return the object/offset from this entry. If the entry | |
2320 | * was copy-on-write or empty, it has been fixed up. | |
2321 | */ | |
2322 | ||
2323 | *offset = (share_offset - entry->start) + entry->offset; | |
2324 | *object = entry->object.vm_object; | |
2325 | ||
2326 | /* | |
2327 | * Return whether this is the only map sharing this data. | |
2328 | */ | |
2329 | ||
2330 | if (!su) { | |
2331 | simple_lock(&share_map->ref_lock); | |
2332 | su = (share_map->ref_count == 1); | |
2333 | simple_unlock(&share_map->ref_lock); | |
2334 | } | |
2335 | ||
2336 | *out_prot = prot; | |
2337 | *single_use = su; | |
2338 | ||
2339 | return(KERN_SUCCESS); | |
2340 | ||
2341 | #undef RETURN | |
2342 | } | |
2343 | ||
2344 | /* | |
2345 | * vm_map_lookup_done: | |
2346 | * | |
2347 | * Releases locks acquired by a vm_map_lookup | |
2348 | * (according to the handle returned by that lookup). | |
2349 | */ | |
2350 | ||
2351 | void vm_map_lookup_done(map, entry) | |
2352 | register vm_map_t map; | |
2353 | vm_map_entry_t entry; | |
2354 | { | |
2355 | /* | |
2356 | * If this entry references a map, unlock it first. | |
2357 | */ | |
2358 | ||
2359 | if (entry->is_a_map) | |
2360 | vm_map_unlock_read(entry->object.share_map); | |
2361 | ||
2362 | /* | |
2363 | * Unlock the main-level map | |
2364 | */ | |
2365 | ||
2366 | vm_map_unlock_read(map); | |
2367 | } | |
2368 | ||
2369 | /* | |
2370 | * Routine: vm_map_simplify | |
2371 | * Purpose: | |
2372 | * Attempt to simplify the map representation in | |
2373 | * the vicinity of the given starting address. | |
2374 | * Note: | |
2375 | * This routine is intended primarily to keep the | |
2376 | * kernel maps more compact -- they generally don't | |
2377 | * benefit from the "expand a map entry" technology | |
2378 | * at allocation time because the adjacent entry | |
2379 | * is often wired down. | |
2380 | */ | |
2381 | void vm_map_simplify(map, start) | |
2382 | vm_map_t map; | |
2383 | vm_offset_t start; | |
2384 | { | |
2385 | vm_map_entry_t this_entry; | |
2386 | vm_map_entry_t prev_entry; | |
2387 | ||
2388 | vm_map_lock(map); | |
2389 | if ( | |
2390 | (vm_map_lookup_entry(map, start, &this_entry)) && | |
2391 | ((prev_entry = this_entry->prev) != &map->header) && | |
2392 | ||
2393 | (prev_entry->end == start) && | |
2394 | (map->is_main_map) && | |
2395 | ||
2396 | (prev_entry->is_a_map == FALSE) && | |
2397 | (prev_entry->is_sub_map == FALSE) && | |
2398 | ||
2399 | (this_entry->is_a_map == FALSE) && | |
2400 | (this_entry->is_sub_map == FALSE) && | |
2401 | ||
2402 | (prev_entry->inheritance == this_entry->inheritance) && | |
2403 | (prev_entry->protection == this_entry->protection) && | |
2404 | (prev_entry->max_protection == this_entry->max_protection) && | |
2405 | (prev_entry->wired_count == this_entry->wired_count) && | |
2406 | ||
2407 | (prev_entry->copy_on_write == this_entry->copy_on_write) && | |
2408 | (prev_entry->needs_copy == this_entry->needs_copy) && | |
2409 | ||
2410 | (prev_entry->object.vm_object == this_entry->object.vm_object) && | |
2411 | ((prev_entry->offset + (prev_entry->end - prev_entry->start)) | |
2412 | == this_entry->offset) | |
2413 | ) { | |
2414 | if (map->first_free == this_entry) | |
2415 | map->first_free = prev_entry; | |
2416 | ||
2417 | SAVE_HINT(map, prev_entry); | |
2418 | vm_map_entry_unlink(map, this_entry); | |
2419 | prev_entry->end = this_entry->end; | |
2420 | vm_object_deallocate(this_entry->object.vm_object); | |
2421 | vm_map_entry_dispose(map, this_entry); | |
2422 | } | |
2423 | vm_map_unlock(map); | |
2424 | } | |
2425 | ||
2426 | /* | |
2427 | * vm_map_print: [ debug ] | |
2428 | */ | |
2429 | void vm_map_print(map, full) | |
2430 | register vm_map_t map; | |
2431 | boolean_t full; | |
2432 | { | |
2433 | register vm_map_entry_t entry; | |
2434 | extern int indent; | |
2435 | ||
2436 | iprintf("%s map 0x%x: pmap=0x%x,ref=%d,nentries=%d,version=%d\n", | |
2437 | (map->is_main_map ? "Task" : "Share"), | |
2438 | (int) map, (int) (map->pmap), map->ref_count, map->nentries, | |
2439 | map->timestamp); | |
2440 | ||
2441 | if (!full && indent) | |
2442 | return; | |
2443 | ||
2444 | indent += 2; | |
2445 | for (entry = map->header.next; entry != &map->header; | |
2446 | entry = entry->next) { | |
2447 | iprintf("map entry 0x%x: start=0x%x, end=0x%x, ", | |
2448 | (int) entry, (int) entry->start, (int) entry->end); | |
2449 | if (map->is_main_map) { | |
2450 | static char *inheritance_name[4] = | |
2451 | { "share", "copy", "none", "donate_copy"}; | |
2452 | printf("prot=%x/%x/%s, ", | |
2453 | entry->protection, | |
2454 | entry->max_protection, | |
2455 | inheritance_name[entry->inheritance]); | |
2456 | if (entry->wired_count != 0) | |
2457 | printf("wired, "); | |
2458 | } | |
2459 | ||
2460 | if (entry->is_a_map || entry->is_sub_map) { | |
2461 | printf("share=0x%x, offset=0x%x\n", | |
2462 | (int) entry->object.share_map, | |
2463 | (int) entry->offset); | |
2464 | if ((entry->prev == &map->header) || | |
2465 | (!entry->prev->is_a_map) || | |
2466 | (entry->prev->object.share_map != | |
2467 | entry->object.share_map)) { | |
2468 | indent += 2; | |
2469 | vm_map_print(entry->object.share_map, full); | |
2470 | indent -= 2; | |
2471 | } | |
2472 | ||
2473 | } | |
2474 | else { | |
2475 | printf("object=0x%x, offset=0x%x", | |
2476 | (int) entry->object.vm_object, | |
2477 | (int) entry->offset); | |
2478 | if (entry->copy_on_write) | |
2479 | printf(", copy (%s)", | |
2480 | entry->needs_copy ? "needed" : "done"); | |
2481 | printf("\n"); | |
2482 | ||
2483 | if ((entry->prev == &map->header) || | |
2484 | (entry->prev->is_a_map) || | |
2485 | (entry->prev->object.vm_object != | |
2486 | entry->object.vm_object)) { | |
2487 | indent += 2; | |
2488 | vm_object_print(entry->object.vm_object, full); | |
2489 | indent -= 2; | |
2490 | } | |
2491 | } | |
2492 | } | |
2493 | indent -= 2; | |
2494 | } |