Commit | Line | Data |
---|---|---|
175f072e KM |
1 | /* |
2 | * Copyright (c) 1985, Avadis Tevanian, Jr., Michael Wayne Young | |
3 | * Copyright (c) 1987 Carnegie-Mellon University | |
4 | * Copyright (c) 1991 Regents of the University of California. | |
5 | * All rights reserved. | |
6 | * | |
7 | * This code is derived from software contributed to Berkeley by | |
8 | * The Mach Operating System project at Carnegie-Mellon University. | |
9 | * | |
10 | * The CMU software License Agreement specifies the terms and conditions | |
11 | * for use and redistribution. | |
12 | * | |
5d7b9ad3 | 13 | * @(#)vm_map.c 7.2 (Berkeley) %G% |
175f072e KM |
14 | */ |
15 | ||
16 | /* | |
17 | * Virtual memory mapping module. | |
18 | */ | |
19 | ||
5d7b9ad3 | 20 | #include "param.h" |
175f072e | 21 | #include "malloc.h" |
5d7b9ad3 MK |
22 | #include "vm.h" |
23 | #include "vm_page.h" | |
24 | #include "vm_object.h" | |
175f072e KM |
25 | |
26 | /* | |
27 | * Virtual memory maps provide for the mapping, protection, | |
28 | * and sharing of virtual memory objects. In addition, | |
29 | * this module provides for an efficient virtual copy of | |
30 | * memory from one map to another. | |
31 | * | |
32 | * Synchronization is required prior to most operations. | |
33 | * | |
34 | * Maps consist of an ordered doubly-linked list of simple | |
35 | * entries; a single hint is used to speed up lookups. | |
36 | * | |
37 | * In order to properly represent the sharing of virtual | |
38 | * memory regions among maps, the map structure is bi-level. | |
39 | * Top-level ("address") maps refer to regions of sharable | |
40 | * virtual memory. These regions are implemented as | |
41 | * ("sharing") maps, which then refer to the actual virtual | |
42 | * memory objects. When two address maps "share" memory, | |
43 | * their top-level maps both have references to the same | |
44 | * sharing map. When memory is virtual-copied from one | |
45 | * address map to another, the references in the sharing | |
46 | * maps are actually copied -- no copying occurs at the | |
47 | * virtual memory object level. | |
48 | * | |
49 | * Since portions of maps are specified by start/end addreses, | |
50 | * which may not align with existing map entries, all | |
51 | * routines merely "clip" entries to these start/end values. | |
52 | * [That is, an entry is split into two, bordering at a | |
53 | * start or end value.] Note that these clippings may not | |
54 | * always be necessary (as the two resulting entries are then | |
55 | * not changed); however, the clipping is done for convenience. | |
56 | * No attempt is currently made to "glue back together" two | |
57 | * abutting entries. | |
58 | * | |
59 | * As mentioned above, virtual copy operations are performed | |
60 | * by copying VM object references from one sharing map to | |
61 | * another, and then marking both regions as copy-on-write. | |
62 | * It is important to note that only one writeable reference | |
63 | * to a VM object region exists in any map -- this means that | |
64 | * shadow object creation can be delayed until a write operation | |
65 | * occurs. | |
66 | */ | |
67 | ||
68 | /* | |
5d7b9ad3 | 69 | * vm_map_startup: |
175f072e KM |
70 | * |
71 | * Initialize the vm_map module. Must be called before | |
72 | * any other vm_map routines. | |
73 | * | |
74 | * Map and entry structures are allocated from the general | |
75 | * purpose memory pool with some exceptions: | |
76 | * | |
77 | * - The kernel map and kmem submap are allocated statically. | |
78 | * - Kernel map entries are allocated out of a static pool. | |
79 | * | |
80 | * These restrictions are necessary since malloc() uses the | |
81 | * maps and requires map entries. | |
82 | */ | |
83 | ||
84 | vm_offset_t kentry_data; | |
85 | vm_size_t kentry_data_size; | |
86 | vm_map_entry_t kentry_free; | |
87 | vm_map_t kmap_free; | |
88 | ||
5d7b9ad3 | 89 | void vm_map_startup() |
175f072e KM |
90 | { |
91 | register int i; | |
92 | register vm_map_entry_t mep; | |
93 | vm_map_t mp; | |
94 | ||
95 | /* | |
96 | * Static map structures for allocation before initialization of | |
97 | * kernel map or kmem map. vm_map_create knows how to deal with them. | |
98 | */ | |
99 | kmap_free = mp = (vm_map_t) kentry_data; | |
100 | i = MAX_KMAP; | |
101 | while (--i > 0) { | |
102 | mp->header.next = (vm_map_entry_t) (mp + 1); | |
103 | mp++; | |
104 | } | |
5d7b9ad3 | 105 | mp++->header.next = NULL; |
175f072e KM |
106 | |
107 | /* | |
108 | * Form a free list of statically allocated kernel map entries | |
109 | * with the rest. | |
110 | */ | |
111 | kentry_free = mep = (vm_map_entry_t) mp; | |
112 | i = (kentry_data_size - MAX_KMAP * sizeof *mp) / sizeof *mep; | |
113 | while (--i > 0) { | |
114 | mep->next = mep + 1; | |
115 | mep++; | |
116 | } | |
5d7b9ad3 MK |
117 | mep->next = NULL; |
118 | } | |
119 | ||
120 | /* | |
121 | * Allocate a vmspace structure, including a vm_map and pmap, | |
122 | * and initialize those structures. The refcnt is set to 1. | |
123 | * The remaining fields must be initialized by the caller. | |
124 | */ | |
125 | struct vmspace * | |
126 | vmspace_alloc(min, max, pageable) | |
127 | vm_offset_t min, max; | |
128 | int pageable; | |
129 | { | |
130 | register struct vmspace *vm; | |
131 | ||
132 | MALLOC(vm, struct vmspace *, sizeof(struct vmspace), M_VMMAP, M_WAITOK); | |
133 | bzero(vm, (caddr_t) &vm->vm_startcopy - (caddr_t) vm); | |
134 | vm_map_init(&vm->vm_map, min, max, pageable); | |
135 | pmap_pinit(&vm->vm_pmap); | |
136 | vm->vm_map.pmap = &vm->vm_pmap; /* XXX */ | |
137 | vm->vm_refcnt = 1; | |
138 | return (vm); | |
139 | } | |
140 | ||
141 | void | |
142 | vmspace_free(vm) | |
143 | register struct vmspace *vm; | |
144 | { | |
145 | ||
146 | if (--vm->vm_refcnt == 0) { | |
147 | /* | |
148 | * Lock the map, to wait out all other references to it. | |
149 | * Delete all of the mappings and pages they hold, | |
150 | * then call the pmap module to reclaim anything left. | |
151 | */ | |
152 | vm_map_lock(&vm->vm_map); | |
153 | (void) vm_map_delete(&vm->vm_map, vm->vm_map.min_offset, | |
154 | vm->vm_map.max_offset); | |
155 | pmap_release(&vm->vm_pmap); | |
156 | FREE(vm, M_VMMAP); | |
157 | } | |
175f072e KM |
158 | } |
159 | ||
160 | /* | |
161 | * vm_map_create: | |
162 | * | |
163 | * Creates and returns a new empty VM map with | |
164 | * the given physical map structure, and having | |
165 | * the given lower and upper address bounds. | |
166 | */ | |
167 | vm_map_t vm_map_create(pmap, min, max, pageable) | |
168 | pmap_t pmap; | |
169 | vm_offset_t min, max; | |
170 | boolean_t pageable; | |
171 | { | |
172 | register vm_map_t result; | |
173 | extern vm_map_t kernel_map, kmem_map; | |
174 | ||
5d7b9ad3 | 175 | if (kmem_map == NULL) { |
175f072e KM |
176 | result = kmap_free; |
177 | kmap_free = (vm_map_t) result->header.next; | |
5d7b9ad3 MK |
178 | if (result == NULL) |
179 | panic("vm_map_create: out of maps"); | |
175f072e KM |
180 | } else |
181 | MALLOC(result, vm_map_t, sizeof(struct vm_map), | |
182 | M_VMMAP, M_WAITOK); | |
183 | ||
5d7b9ad3 | 184 | vm_map_init(result, min, max, pageable); |
175f072e | 185 | result->pmap = pmap; |
175f072e KM |
186 | return(result); |
187 | } | |
188 | ||
5d7b9ad3 MK |
189 | /* |
190 | * Initialize an existing vm_map structure | |
191 | * such as that in the vmspace structure. | |
192 | * The pmap is set elsewhere. | |
193 | */ | |
194 | void | |
195 | vm_map_init(map, min, max, pageable) | |
196 | register struct vm_map *map; | |
197 | vm_offset_t min, max; | |
198 | boolean_t pageable; | |
199 | { | |
200 | map->header.next = map->header.prev = &map->header; | |
201 | map->nentries = 0; | |
202 | map->size = 0; | |
203 | map->ref_count = 1; | |
204 | map->is_main_map = TRUE; | |
205 | map->min_offset = min; | |
206 | map->max_offset = max; | |
207 | map->entries_pageable = pageable; | |
208 | map->first_free = &map->header; | |
209 | map->hint = &map->header; | |
210 | map->timestamp = 0; | |
211 | lock_init(&map->lock, TRUE); | |
212 | simple_lock_init(&map->ref_lock); | |
213 | simple_lock_init(&map->hint_lock); | |
214 | } | |
215 | ||
175f072e KM |
216 | /* |
217 | * vm_map_entry_create: [ internal use only ] | |
218 | * | |
219 | * Allocates a VM map entry for insertion. | |
220 | * No entry fields are filled in. This routine is | |
221 | */ | |
222 | vm_map_entry_t vm_map_entry_create(map) | |
223 | vm_map_t map; | |
224 | { | |
225 | vm_map_entry_t entry; | |
226 | extern vm_map_t kernel_map, kmem_map, mb_map; | |
227 | ||
228 | if (map == kernel_map || map == kmem_map || map == mb_map) { | |
229 | if (entry = kentry_free) | |
230 | kentry_free = kentry_free->next; | |
231 | } else | |
232 | MALLOC(entry, vm_map_entry_t, sizeof(struct vm_map_entry), | |
233 | M_VMMAPENT, M_WAITOK); | |
5d7b9ad3 | 234 | if (entry == NULL) |
175f072e KM |
235 | panic("vm_map_entry_create: out of map entries"); |
236 | ||
237 | return(entry); | |
238 | } | |
239 | ||
240 | /* | |
241 | * vm_map_entry_dispose: [ internal use only ] | |
242 | * | |
243 | * Inverse of vm_map_entry_create. | |
244 | */ | |
245 | void vm_map_entry_dispose(map, entry) | |
246 | vm_map_t map; | |
247 | vm_map_entry_t entry; | |
248 | { | |
249 | extern vm_map_t kernel_map, kmem_map, mb_map; | |
250 | ||
251 | if (map == kernel_map || map == kmem_map || map == mb_map) { | |
252 | entry->next = kentry_free; | |
253 | kentry_free = entry; | |
254 | } else | |
255 | FREE(entry, M_VMMAPENT); | |
256 | } | |
257 | ||
258 | /* | |
259 | * vm_map_entry_{un,}link: | |
260 | * | |
261 | * Insert/remove entries from maps. | |
262 | */ | |
263 | #define vm_map_entry_link(map, after_where, entry) \ | |
264 | { \ | |
265 | (map)->nentries++; \ | |
266 | (entry)->prev = (after_where); \ | |
267 | (entry)->next = (after_where)->next; \ | |
268 | (entry)->prev->next = (entry); \ | |
269 | (entry)->next->prev = (entry); \ | |
270 | } | |
271 | #define vm_map_entry_unlink(map, entry) \ | |
272 | { \ | |
273 | (map)->nentries--; \ | |
274 | (entry)->next->prev = (entry)->prev; \ | |
275 | (entry)->prev->next = (entry)->next; \ | |
276 | } | |
277 | ||
278 | /* | |
279 | * vm_map_reference: | |
280 | * | |
281 | * Creates another valid reference to the given map. | |
282 | * | |
283 | */ | |
284 | void vm_map_reference(map) | |
285 | register vm_map_t map; | |
286 | { | |
5d7b9ad3 | 287 | if (map == NULL) |
175f072e KM |
288 | return; |
289 | ||
290 | simple_lock(&map->ref_lock); | |
291 | map->ref_count++; | |
292 | simple_unlock(&map->ref_lock); | |
293 | } | |
294 | ||
295 | /* | |
296 | * vm_map_deallocate: | |
297 | * | |
298 | * Removes a reference from the specified map, | |
299 | * destroying it if no references remain. | |
300 | * The map should not be locked. | |
301 | */ | |
302 | void vm_map_deallocate(map) | |
303 | register vm_map_t map; | |
304 | { | |
305 | register int c; | |
306 | ||
5d7b9ad3 | 307 | if (map == NULL) |
175f072e KM |
308 | return; |
309 | ||
310 | simple_lock(&map->ref_lock); | |
311 | c = --map->ref_count; | |
312 | simple_unlock(&map->ref_lock); | |
313 | ||
314 | if (c > 0) { | |
315 | return; | |
316 | } | |
317 | ||
318 | /* | |
319 | * Lock the map, to wait out all other references | |
320 | * to it. | |
321 | */ | |
322 | ||
323 | vm_map_lock(map); | |
324 | ||
325 | (void) vm_map_delete(map, map->min_offset, map->max_offset); | |
326 | ||
327 | pmap_destroy(map->pmap); | |
328 | ||
329 | FREE(map, M_VMMAP); | |
330 | } | |
331 | ||
332 | /* | |
333 | * vm_map_insert: [ internal use only ] | |
334 | * | |
335 | * Inserts the given whole VM object into the target | |
336 | * map at the specified address range. The object's | |
337 | * size should match that of the address range. | |
338 | * | |
339 | * Requires that the map be locked, and leaves it so. | |
340 | */ | |
341 | vm_map_insert(map, object, offset, start, end) | |
342 | vm_map_t map; | |
343 | vm_object_t object; | |
344 | vm_offset_t offset; | |
345 | vm_offset_t start; | |
346 | vm_offset_t end; | |
347 | { | |
348 | register vm_map_entry_t new_entry; | |
349 | register vm_map_entry_t prev_entry; | |
350 | vm_map_entry_t temp_entry; | |
351 | ||
352 | /* | |
353 | * Check that the start and end points are not bogus. | |
354 | */ | |
355 | ||
356 | if ((start < map->min_offset) || (end > map->max_offset) || | |
357 | (start >= end)) | |
358 | return(KERN_INVALID_ADDRESS); | |
359 | ||
360 | /* | |
361 | * Find the entry prior to the proposed | |
362 | * starting address; if it's part of an | |
363 | * existing entry, this range is bogus. | |
364 | */ | |
365 | ||
366 | if (vm_map_lookup_entry(map, start, &temp_entry)) | |
367 | return(KERN_NO_SPACE); | |
368 | ||
369 | prev_entry = temp_entry; | |
370 | ||
371 | /* | |
372 | * Assert that the next entry doesn't overlap the | |
373 | * end point. | |
374 | */ | |
375 | ||
376 | if ((prev_entry->next != &map->header) && | |
377 | (prev_entry->next->start < end)) | |
378 | return(KERN_NO_SPACE); | |
379 | ||
380 | /* | |
381 | * See if we can avoid creating a new entry by | |
382 | * extending one of our neighbors. | |
383 | */ | |
384 | ||
5d7b9ad3 | 385 | if (object == NULL) { |
175f072e KM |
386 | if ((prev_entry != &map->header) && |
387 | (prev_entry->end == start) && | |
388 | (map->is_main_map) && | |
389 | (prev_entry->is_a_map == FALSE) && | |
390 | (prev_entry->is_sub_map == FALSE) && | |
391 | (prev_entry->inheritance == VM_INHERIT_DEFAULT) && | |
392 | (prev_entry->protection == VM_PROT_DEFAULT) && | |
393 | (prev_entry->max_protection == VM_PROT_DEFAULT) && | |
394 | (prev_entry->wired_count == 0)) { | |
395 | ||
396 | if (vm_object_coalesce(prev_entry->object.vm_object, | |
5d7b9ad3 | 397 | NULL, |
175f072e KM |
398 | prev_entry->offset, |
399 | (vm_offset_t) 0, | |
400 | (vm_size_t)(prev_entry->end | |
401 | - prev_entry->start), | |
402 | (vm_size_t)(end - prev_entry->end))) { | |
403 | /* | |
404 | * Coalesced the two objects - can extend | |
405 | * the previous map entry to include the | |
406 | * new range. | |
407 | */ | |
408 | map->size += (end - prev_entry->end); | |
409 | prev_entry->end = end; | |
410 | return(KERN_SUCCESS); | |
411 | } | |
412 | } | |
413 | } | |
414 | ||
415 | /* | |
416 | * Create a new entry | |
417 | */ | |
418 | ||
419 | new_entry = vm_map_entry_create(map); | |
420 | new_entry->start = start; | |
421 | new_entry->end = end; | |
422 | ||
423 | new_entry->is_a_map = FALSE; | |
424 | new_entry->is_sub_map = FALSE; | |
425 | new_entry->object.vm_object = object; | |
426 | new_entry->offset = offset; | |
427 | ||
428 | new_entry->copy_on_write = FALSE; | |
429 | new_entry->needs_copy = FALSE; | |
430 | ||
431 | if (map->is_main_map) { | |
432 | new_entry->inheritance = VM_INHERIT_DEFAULT; | |
433 | new_entry->protection = VM_PROT_DEFAULT; | |
434 | new_entry->max_protection = VM_PROT_DEFAULT; | |
435 | new_entry->wired_count = 0; | |
436 | } | |
437 | ||
438 | /* | |
439 | * Insert the new entry into the list | |
440 | */ | |
441 | ||
442 | vm_map_entry_link(map, prev_entry, new_entry); | |
443 | map->size += new_entry->end - new_entry->start; | |
444 | ||
445 | /* | |
446 | * Update the free space hint | |
447 | */ | |
448 | ||
449 | if ((map->first_free == prev_entry) && (prev_entry->end >= new_entry->start)) | |
450 | map->first_free = new_entry; | |
451 | ||
452 | return(KERN_SUCCESS); | |
453 | } | |
454 | ||
455 | /* | |
456 | * SAVE_HINT: | |
457 | * | |
458 | * Saves the specified entry as the hint for | |
459 | * future lookups. Performs necessary interlocks. | |
460 | */ | |
461 | #define SAVE_HINT(map,value) \ | |
462 | simple_lock(&(map)->hint_lock); \ | |
463 | (map)->hint = (value); \ | |
464 | simple_unlock(&(map)->hint_lock); | |
465 | ||
466 | /* | |
467 | * vm_map_lookup_entry: [ internal use only ] | |
468 | * | |
469 | * Finds the map entry containing (or | |
470 | * immediately preceding) the specified address | |
471 | * in the given map; the entry is returned | |
472 | * in the "entry" parameter. The boolean | |
473 | * result indicates whether the address is | |
474 | * actually contained in the map. | |
475 | */ | |
476 | boolean_t vm_map_lookup_entry(map, address, entry) | |
477 | register vm_map_t map; | |
478 | register vm_offset_t address; | |
479 | vm_map_entry_t *entry; /* OUT */ | |
480 | { | |
481 | register vm_map_entry_t cur; | |
482 | register vm_map_entry_t last; | |
483 | ||
484 | /* | |
485 | * Start looking either from the head of the | |
486 | * list, or from the hint. | |
487 | */ | |
488 | ||
489 | simple_lock(&map->hint_lock); | |
490 | cur = map->hint; | |
491 | simple_unlock(&map->hint_lock); | |
492 | ||
493 | if (cur == &map->header) | |
494 | cur = cur->next; | |
495 | ||
496 | if (address >= cur->start) { | |
497 | /* | |
498 | * Go from hint to end of list. | |
499 | * | |
500 | * But first, make a quick check to see if | |
501 | * we are already looking at the entry we | |
502 | * want (which is usually the case). | |
503 | * Note also that we don't need to save the hint | |
504 | * here... it is the same hint (unless we are | |
505 | * at the header, in which case the hint didn't | |
506 | * buy us anything anyway). | |
507 | */ | |
508 | last = &map->header; | |
509 | if ((cur != last) && (cur->end > address)) { | |
510 | *entry = cur; | |
511 | return(TRUE); | |
512 | } | |
513 | } | |
514 | else { | |
515 | /* | |
516 | * Go from start to hint, *inclusively* | |
517 | */ | |
518 | last = cur->next; | |
519 | cur = map->header.next; | |
520 | } | |
521 | ||
522 | /* | |
523 | * Search linearly | |
524 | */ | |
525 | ||
526 | while (cur != last) { | |
527 | if (cur->end > address) { | |
528 | if (address >= cur->start) { | |
529 | /* | |
530 | * Save this lookup for future | |
531 | * hints, and return | |
532 | */ | |
533 | ||
534 | *entry = cur; | |
535 | SAVE_HINT(map, cur); | |
536 | return(TRUE); | |
537 | } | |
538 | break; | |
539 | } | |
540 | cur = cur->next; | |
541 | } | |
542 | *entry = cur->prev; | |
543 | SAVE_HINT(map, *entry); | |
544 | return(FALSE); | |
545 | } | |
546 | ||
547 | /* | |
548 | * vm_map_find finds an unallocated region in the target address | |
549 | * map with the given length. The search is defined to be | |
550 | * first-fit from the specified address; the region found is | |
551 | * returned in the same parameter. | |
552 | * | |
553 | */ | |
554 | vm_map_find(map, object, offset, addr, length, find_space) | |
555 | vm_map_t map; | |
556 | vm_object_t object; | |
557 | vm_offset_t offset; | |
558 | vm_offset_t *addr; /* IN/OUT */ | |
559 | vm_size_t length; | |
560 | boolean_t find_space; | |
561 | { | |
562 | register vm_map_entry_t entry; | |
563 | register vm_offset_t start; | |
564 | register vm_offset_t end; | |
565 | int result; | |
566 | ||
567 | start = *addr; | |
568 | ||
569 | vm_map_lock(map); | |
570 | ||
571 | if (find_space) { | |
572 | /* | |
573 | * Calculate the first possible address. | |
574 | */ | |
575 | ||
576 | if (start < map->min_offset) | |
577 | start = map->min_offset; | |
578 | if (start > map->max_offset) { | |
579 | vm_map_unlock(map); | |
580 | return (KERN_NO_SPACE); | |
581 | } | |
582 | ||
583 | /* | |
584 | * Look for the first possible address; | |
585 | * if there's already something at this | |
586 | * address, we have to start after it. | |
587 | */ | |
588 | ||
589 | if (start == map->min_offset) { | |
590 | if ((entry = map->first_free) != &map->header) | |
591 | start = entry->end; | |
592 | } else { | |
593 | vm_map_entry_t tmp_entry; | |
594 | if (vm_map_lookup_entry(map, start, &tmp_entry)) | |
595 | start = tmp_entry->end; | |
596 | entry = tmp_entry; | |
597 | } | |
598 | ||
599 | /* | |
600 | * In any case, the "entry" always precedes | |
601 | * the proposed new region throughout the | |
602 | * loop: | |
603 | */ | |
604 | ||
605 | while (TRUE) { | |
606 | register vm_map_entry_t next; | |
607 | ||
608 | /* | |
609 | * Find the end of the proposed new region. | |
610 | * Be sure we didn't go beyond the end, or | |
611 | * wrap around the address. | |
612 | */ | |
613 | ||
614 | end = start + length; | |
615 | ||
616 | if ((end > map->max_offset) || (end < start)) { | |
617 | vm_map_unlock(map); | |
618 | return (KERN_NO_SPACE); | |
619 | } | |
620 | ||
621 | /* | |
622 | * If there are no more entries, we must win. | |
623 | */ | |
624 | ||
625 | next = entry->next; | |
626 | if (next == &map->header) | |
627 | break; | |
628 | ||
629 | /* | |
630 | * If there is another entry, it must be | |
631 | * after the end of the potential new region. | |
632 | */ | |
633 | ||
634 | if (next->start >= end) | |
635 | break; | |
636 | ||
637 | /* | |
638 | * Didn't fit -- move to the next entry. | |
639 | */ | |
640 | ||
641 | entry = next; | |
642 | start = entry->end; | |
643 | } | |
644 | *addr = start; | |
645 | ||
646 | SAVE_HINT(map, entry); | |
647 | } | |
648 | ||
649 | result = vm_map_insert(map, object, offset, start, start + length); | |
650 | ||
651 | vm_map_unlock(map); | |
652 | return(result); | |
653 | } | |
654 | ||
655 | /* | |
656 | * vm_map_simplify_entry: [ internal use only ] | |
657 | * | |
658 | * Simplify the given map entry by: | |
659 | * removing extra sharing maps | |
660 | * [XXX maybe later] merging with a neighbor | |
661 | */ | |
662 | void vm_map_simplify_entry(map, entry) | |
663 | vm_map_t map; | |
664 | vm_map_entry_t entry; | |
665 | { | |
666 | #ifdef lint | |
667 | map++; | |
668 | #endif lint | |
669 | ||
670 | /* | |
671 | * If this entry corresponds to a sharing map, then | |
672 | * see if we can remove the level of indirection. | |
673 | * If it's not a sharing map, then it points to | |
674 | * a VM object, so see if we can merge with either | |
675 | * of our neighbors. | |
676 | */ | |
677 | ||
678 | if (entry->is_sub_map) | |
679 | return; | |
680 | if (entry->is_a_map) { | |
681 | #if 0 | |
682 | vm_map_t my_share_map; | |
683 | int count; | |
684 | ||
685 | my_share_map = entry->object.share_map; | |
686 | simple_lock(&my_share_map->ref_lock); | |
687 | count = my_share_map->ref_count; | |
688 | simple_unlock(&my_share_map->ref_lock); | |
689 | ||
690 | if (count == 1) { | |
691 | /* Can move the region from | |
692 | * entry->start to entry->end (+ entry->offset) | |
693 | * in my_share_map into place of entry. | |
694 | * Later. | |
695 | */ | |
696 | } | |
697 | #endif 0 | |
698 | } | |
699 | else { | |
700 | /* | |
701 | * Try to merge with our neighbors. | |
702 | * | |
703 | * Conditions for merge are: | |
704 | * | |
705 | * 1. entries are adjacent. | |
706 | * 2. both entries point to objects | |
707 | * with null pagers. | |
708 | * | |
709 | * If a merge is possible, we replace the two | |
710 | * entries with a single entry, then merge | |
711 | * the two objects into a single object. | |
712 | * | |
713 | * Now, all that is left to do is write the | |
714 | * code! | |
715 | */ | |
716 | } | |
717 | } | |
718 | ||
719 | /* | |
720 | * vm_map_clip_start: [ internal use only ] | |
721 | * | |
722 | * Asserts that the given entry begins at or after | |
723 | * the specified address; if necessary, | |
724 | * it splits the entry into two. | |
725 | */ | |
726 | #define vm_map_clip_start(map, entry, startaddr) \ | |
727 | { \ | |
728 | if (startaddr > entry->start) \ | |
729 | _vm_map_clip_start(map, entry, startaddr); \ | |
730 | } | |
731 | ||
732 | /* | |
733 | * This routine is called only when it is known that | |
734 | * the entry must be split. | |
735 | */ | |
736 | void _vm_map_clip_start(map, entry, start) | |
737 | register vm_map_t map; | |
738 | register vm_map_entry_t entry; | |
739 | register vm_offset_t start; | |
740 | { | |
741 | register vm_map_entry_t new_entry; | |
742 | ||
743 | /* | |
744 | * See if we can simplify this entry first | |
745 | */ | |
746 | ||
747 | vm_map_simplify_entry(map, entry); | |
748 | ||
749 | /* | |
750 | * Split off the front portion -- | |
751 | * note that we must insert the new | |
752 | * entry BEFORE this one, so that | |
753 | * this entry has the specified starting | |
754 | * address. | |
755 | */ | |
756 | ||
757 | new_entry = vm_map_entry_create(map); | |
758 | *new_entry = *entry; | |
759 | ||
760 | new_entry->end = start; | |
761 | entry->offset += (start - entry->start); | |
762 | entry->start = start; | |
763 | ||
764 | vm_map_entry_link(map, entry->prev, new_entry); | |
765 | ||
766 | if (entry->is_a_map || entry->is_sub_map) | |
767 | vm_map_reference(new_entry->object.share_map); | |
768 | else | |
769 | vm_object_reference(new_entry->object.vm_object); | |
770 | } | |
771 | ||
772 | /* | |
773 | * vm_map_clip_end: [ internal use only ] | |
774 | * | |
775 | * Asserts that the given entry ends at or before | |
776 | * the specified address; if necessary, | |
777 | * it splits the entry into two. | |
778 | */ | |
779 | ||
780 | void _vm_map_clip_end(); | |
781 | #define vm_map_clip_end(map, entry, endaddr) \ | |
782 | { \ | |
783 | if (endaddr < entry->end) \ | |
784 | _vm_map_clip_end(map, entry, endaddr); \ | |
785 | } | |
786 | ||
787 | /* | |
788 | * This routine is called only when it is known that | |
789 | * the entry must be split. | |
790 | */ | |
791 | void _vm_map_clip_end(map, entry, end) | |
792 | register vm_map_t map; | |
793 | register vm_map_entry_t entry; | |
794 | register vm_offset_t end; | |
795 | { | |
796 | register vm_map_entry_t new_entry; | |
797 | ||
798 | /* | |
799 | * Create a new entry and insert it | |
800 | * AFTER the specified entry | |
801 | */ | |
802 | ||
803 | new_entry = vm_map_entry_create(map); | |
804 | *new_entry = *entry; | |
805 | ||
806 | new_entry->start = entry->end = end; | |
807 | new_entry->offset += (end - entry->start); | |
808 | ||
809 | vm_map_entry_link(map, entry, new_entry); | |
810 | ||
811 | if (entry->is_a_map || entry->is_sub_map) | |
812 | vm_map_reference(new_entry->object.share_map); | |
813 | else | |
814 | vm_object_reference(new_entry->object.vm_object); | |
815 | } | |
816 | ||
817 | /* | |
818 | * VM_MAP_RANGE_CHECK: [ internal use only ] | |
819 | * | |
820 | * Asserts that the starting and ending region | |
821 | * addresses fall within the valid range of the map. | |
822 | */ | |
823 | #define VM_MAP_RANGE_CHECK(map, start, end) \ | |
824 | { \ | |
825 | if (start < vm_map_min(map)) \ | |
826 | start = vm_map_min(map); \ | |
827 | if (end > vm_map_max(map)) \ | |
828 | end = vm_map_max(map); \ | |
829 | if (start > end) \ | |
830 | start = end; \ | |
831 | } | |
832 | ||
833 | /* | |
834 | * vm_map_submap: [ kernel use only ] | |
835 | * | |
836 | * Mark the given range as handled by a subordinate map. | |
837 | * | |
838 | * This range must have been created with vm_map_find, | |
839 | * and no other operations may have been performed on this | |
840 | * range prior to calling vm_map_submap. | |
841 | * | |
842 | * Only a limited number of operations can be performed | |
843 | * within this rage after calling vm_map_submap: | |
844 | * vm_fault | |
845 | * [Don't try vm_map_copy!] | |
846 | * | |
847 | * To remove a submapping, one must first remove the | |
848 | * range from the superior map, and then destroy the | |
849 | * submap (if desired). [Better yet, don't try it.] | |
850 | */ | |
851 | vm_map_submap(map, start, end, submap) | |
852 | register vm_map_t map; | |
853 | register vm_offset_t start; | |
854 | register vm_offset_t end; | |
855 | vm_map_t submap; | |
856 | { | |
857 | vm_map_entry_t entry; | |
858 | register int result = KERN_INVALID_ARGUMENT; | |
859 | ||
860 | vm_map_lock(map); | |
861 | ||
862 | VM_MAP_RANGE_CHECK(map, start, end); | |
863 | ||
864 | if (vm_map_lookup_entry(map, start, &entry)) { | |
865 | vm_map_clip_start(map, entry, start); | |
866 | } | |
867 | else | |
868 | entry = entry->next; | |
869 | ||
870 | vm_map_clip_end(map, entry, end); | |
871 | ||
872 | if ((entry->start == start) && (entry->end == end) && | |
873 | (!entry->is_a_map) && | |
5d7b9ad3 | 874 | (entry->object.vm_object == NULL) && |
175f072e KM |
875 | (!entry->copy_on_write)) { |
876 | entry->is_a_map = FALSE; | |
877 | entry->is_sub_map = TRUE; | |
878 | vm_map_reference(entry->object.sub_map = submap); | |
879 | result = KERN_SUCCESS; | |
880 | } | |
881 | vm_map_unlock(map); | |
882 | ||
883 | return(result); | |
884 | } | |
885 | ||
886 | /* | |
887 | * vm_map_protect: | |
888 | * | |
889 | * Sets the protection of the specified address | |
890 | * region in the target map. If "set_max" is | |
891 | * specified, the maximum protection is to be set; | |
892 | * otherwise, only the current protection is affected. | |
893 | */ | |
894 | vm_map_protect(map, start, end, new_prot, set_max) | |
895 | register vm_map_t map; | |
896 | register vm_offset_t start; | |
897 | register vm_offset_t end; | |
898 | register vm_prot_t new_prot; | |
899 | register boolean_t set_max; | |
900 | { | |
901 | register vm_map_entry_t current; | |
902 | vm_map_entry_t entry; | |
903 | ||
904 | vm_map_lock(map); | |
905 | ||
906 | VM_MAP_RANGE_CHECK(map, start, end); | |
907 | ||
908 | if (vm_map_lookup_entry(map, start, &entry)) { | |
909 | vm_map_clip_start(map, entry, start); | |
910 | } | |
911 | else | |
912 | entry = entry->next; | |
913 | ||
914 | /* | |
915 | * Make a first pass to check for protection | |
916 | * violations. | |
917 | */ | |
918 | ||
919 | current = entry; | |
920 | while ((current != &map->header) && (current->start < end)) { | |
921 | if (current->is_sub_map) | |
922 | return(KERN_INVALID_ARGUMENT); | |
923 | if ((new_prot & current->max_protection) != new_prot) { | |
924 | vm_map_unlock(map); | |
925 | return(KERN_PROTECTION_FAILURE); | |
926 | } | |
927 | ||
928 | current = current->next; | |
929 | } | |
930 | ||
931 | /* | |
932 | * Go back and fix up protections. | |
933 | * [Note that clipping is not necessary the second time.] | |
934 | */ | |
935 | ||
936 | current = entry; | |
937 | ||
938 | while ((current != &map->header) && (current->start < end)) { | |
939 | vm_prot_t old_prot; | |
940 | ||
941 | vm_map_clip_end(map, current, end); | |
942 | ||
943 | old_prot = current->protection; | |
944 | if (set_max) | |
945 | current->protection = | |
946 | (current->max_protection = new_prot) & | |
947 | old_prot; | |
948 | else | |
949 | current->protection = new_prot; | |
950 | ||
951 | /* | |
952 | * Update physical map if necessary. | |
953 | * Worry about copy-on-write here -- CHECK THIS XXX | |
954 | */ | |
955 | ||
956 | if (current->protection != old_prot) { | |
957 | ||
958 | #define MASK(entry) ((entry)->copy_on_write ? ~VM_PROT_WRITE : \ | |
959 | VM_PROT_ALL) | |
960 | #define max(a,b) ((a) > (b) ? (a) : (b)) | |
961 | ||
962 | if (current->is_a_map) { | |
963 | vm_map_entry_t share_entry; | |
964 | vm_offset_t share_end; | |
965 | ||
966 | vm_map_lock(current->object.share_map); | |
967 | (void) vm_map_lookup_entry( | |
968 | current->object.share_map, | |
969 | current->offset, | |
970 | &share_entry); | |
971 | share_end = current->offset + | |
972 | (current->end - current->start); | |
973 | while ((share_entry != | |
974 | ¤t->object.share_map->header) && | |
975 | (share_entry->start < share_end)) { | |
976 | ||
977 | pmap_protect(map->pmap, | |
978 | (max(share_entry->start, | |
979 | current->offset) - | |
980 | current->offset + | |
981 | current->start), | |
982 | min(share_entry->end, | |
983 | share_end) - | |
984 | current->offset + | |
985 | current->start, | |
986 | current->protection & | |
987 | MASK(share_entry)); | |
988 | ||
989 | share_entry = share_entry->next; | |
990 | } | |
991 | vm_map_unlock(current->object.share_map); | |
992 | } | |
993 | else | |
994 | pmap_protect(map->pmap, current->start, | |
995 | current->end, | |
996 | current->protection & MASK(entry)); | |
997 | #undef max | |
998 | #undef MASK | |
999 | } | |
1000 | current = current->next; | |
1001 | } | |
1002 | ||
1003 | vm_map_unlock(map); | |
1004 | return(KERN_SUCCESS); | |
1005 | } | |
1006 | ||
1007 | /* | |
1008 | * vm_map_inherit: | |
1009 | * | |
1010 | * Sets the inheritance of the specified address | |
1011 | * range in the target map. Inheritance | |
1012 | * affects how the map will be shared with | |
1013 | * child maps at the time of vm_map_fork. | |
1014 | */ | |
1015 | vm_map_inherit(map, start, end, new_inheritance) | |
1016 | register vm_map_t map; | |
1017 | register vm_offset_t start; | |
1018 | register vm_offset_t end; | |
1019 | register vm_inherit_t new_inheritance; | |
1020 | { | |
1021 | register vm_map_entry_t entry; | |
1022 | vm_map_entry_t temp_entry; | |
1023 | ||
1024 | switch (new_inheritance) { | |
1025 | case VM_INHERIT_NONE: | |
1026 | case VM_INHERIT_COPY: | |
1027 | case VM_INHERIT_SHARE: | |
1028 | break; | |
1029 | default: | |
1030 | return(KERN_INVALID_ARGUMENT); | |
1031 | } | |
1032 | ||
1033 | vm_map_lock(map); | |
1034 | ||
1035 | VM_MAP_RANGE_CHECK(map, start, end); | |
1036 | ||
1037 | if (vm_map_lookup_entry(map, start, &temp_entry)) { | |
1038 | entry = temp_entry; | |
1039 | vm_map_clip_start(map, entry, start); | |
1040 | } | |
1041 | else | |
1042 | entry = temp_entry->next; | |
1043 | ||
1044 | while ((entry != &map->header) && (entry->start < end)) { | |
1045 | vm_map_clip_end(map, entry, end); | |
1046 | ||
1047 | entry->inheritance = new_inheritance; | |
1048 | ||
1049 | entry = entry->next; | |
1050 | } | |
1051 | ||
1052 | vm_map_unlock(map); | |
1053 | return(KERN_SUCCESS); | |
1054 | } | |
1055 | ||
1056 | /* | |
1057 | * vm_map_pageable: | |
1058 | * | |
1059 | * Sets the pageability of the specified address | |
1060 | * range in the target map. Regions specified | |
1061 | * as not pageable require locked-down physical | |
1062 | * memory and physical page maps. | |
1063 | * | |
1064 | * The map must not be locked, but a reference | |
1065 | * must remain to the map throughout the call. | |
1066 | */ | |
1067 | vm_map_pageable(map, start, end, new_pageable) | |
1068 | register vm_map_t map; | |
1069 | register vm_offset_t start; | |
1070 | register vm_offset_t end; | |
1071 | register boolean_t new_pageable; | |
1072 | { | |
1073 | register vm_map_entry_t entry; | |
1074 | vm_map_entry_t temp_entry; | |
1075 | ||
1076 | vm_map_lock(map); | |
1077 | ||
1078 | VM_MAP_RANGE_CHECK(map, start, end); | |
1079 | ||
1080 | /* | |
1081 | * Only one pageability change may take place at one | |
1082 | * time, since vm_fault assumes it will be called | |
1083 | * only once for each wiring/unwiring. Therefore, we | |
1084 | * have to make sure we're actually changing the pageability | |
1085 | * for the entire region. We do so before making any changes. | |
1086 | */ | |
1087 | ||
1088 | if (vm_map_lookup_entry(map, start, &temp_entry)) { | |
1089 | entry = temp_entry; | |
1090 | vm_map_clip_start(map, entry, start); | |
1091 | } | |
1092 | else | |
1093 | entry = temp_entry->next; | |
1094 | temp_entry = entry; | |
1095 | ||
1096 | /* | |
1097 | * Actions are rather different for wiring and unwiring, | |
1098 | * so we have two separate cases. | |
1099 | */ | |
1100 | ||
1101 | if (new_pageable) { | |
1102 | ||
1103 | /* | |
1104 | * Unwiring. First ensure that the range to be | |
1105 | * unwired is really wired down. | |
1106 | */ | |
1107 | while ((entry != &map->header) && (entry->start < end)) { | |
1108 | ||
1109 | if (entry->wired_count == 0) { | |
1110 | vm_map_unlock(map); | |
1111 | return(KERN_INVALID_ARGUMENT); | |
1112 | } | |
1113 | entry = entry->next; | |
1114 | } | |
1115 | ||
1116 | /* | |
1117 | * Now decrement the wiring count for each region. | |
1118 | * If a region becomes completely unwired, | |
1119 | * unwire its physical pages and mappings. | |
1120 | */ | |
1121 | lock_set_recursive(&map->lock); | |
1122 | ||
1123 | entry = temp_entry; | |
1124 | while ((entry != &map->header) && (entry->start < end)) { | |
1125 | vm_map_clip_end(map, entry, end); | |
1126 | ||
1127 | entry->wired_count--; | |
1128 | if (entry->wired_count == 0) | |
1129 | vm_fault_unwire(map, entry->start, entry->end); | |
1130 | ||
1131 | entry = entry->next; | |
1132 | } | |
1133 | lock_clear_recursive(&map->lock); | |
1134 | } | |
1135 | ||
1136 | else { | |
1137 | /* | |
1138 | * Wiring. We must do this in two passes: | |
1139 | * | |
1140 | * 1. Holding the write lock, we increment the | |
1141 | * wiring count. For any area that is not already | |
1142 | * wired, we create any shadow objects that need | |
1143 | * to be created. | |
1144 | * | |
1145 | * 2. We downgrade to a read lock, and call | |
1146 | * vm_fault_wire to fault in the pages for any | |
1147 | * newly wired area (wired_count is 1). | |
1148 | * | |
1149 | * Downgrading to a read lock for vm_fault_wire avoids | |
1150 | * a possible deadlock with another thread that may have | |
1151 | * faulted on one of the pages to be wired (it would mark | |
1152 | * the page busy, blocking us, then in turn block on the | |
1153 | * map lock that we hold). Because of problems in the | |
1154 | * recursive lock package, we cannot upgrade to a write | |
1155 | * lock in vm_map_lookup. Thus, any actions that require | |
1156 | * the write lock must be done beforehand. Because we | |
1157 | * keep the read lock on the map, the copy-on-write status | |
1158 | * of the entries we modify here cannot change. | |
1159 | */ | |
1160 | ||
1161 | /* | |
1162 | * Pass 1. | |
1163 | */ | |
1164 | entry = temp_entry; | |
1165 | while ((entry != &map->header) && (entry->start < end)) { | |
1166 | vm_map_clip_end(map, entry, end); | |
1167 | ||
1168 | entry->wired_count++; | |
1169 | if (entry->wired_count == 1) { | |
1170 | ||
1171 | /* | |
1172 | * Perform actions of vm_map_lookup that need | |
1173 | * the write lock on the map: create a shadow | |
1174 | * object for a copy-on-write region, or an | |
1175 | * object for a zero-fill region. | |
1176 | * | |
1177 | * We don't have to do this for entries that | |
1178 | * point to sharing maps, because we won't hold | |
1179 | * the lock on the sharing map. | |
1180 | */ | |
1181 | if (!entry->is_a_map) { | |
1182 | if (entry->needs_copy && | |
1183 | ((entry->protection & VM_PROT_WRITE) != 0)) { | |
1184 | ||
1185 | vm_object_shadow(&entry->object.vm_object, | |
1186 | &entry->offset, | |
1187 | (vm_size_t)(entry->end | |
1188 | - entry->start)); | |
1189 | entry->needs_copy = FALSE; | |
1190 | } | |
5d7b9ad3 | 1191 | else if (entry->object.vm_object == NULL) { |
175f072e KM |
1192 | entry->object.vm_object = |
1193 | vm_object_allocate((vm_size_t)(entry->end | |
1194 | - entry->start)); | |
1195 | entry->offset = (vm_offset_t)0; | |
1196 | } | |
1197 | } | |
1198 | } | |
1199 | ||
1200 | entry = entry->next; | |
1201 | } | |
1202 | ||
1203 | /* | |
1204 | * Pass 2. | |
1205 | */ | |
1206 | ||
1207 | /* | |
1208 | * HACK HACK HACK HACK | |
1209 | * | |
1210 | * If we are wiring in the kernel map or a submap of it, | |
1211 | * unlock the map to avoid deadlocks. We trust that the | |
1212 | * kernel threads are well-behaved, and therefore will | |
1213 | * not do anything destructive to this region of the map | |
1214 | * while we have it unlocked. We cannot trust user threads | |
1215 | * to do the same. | |
1216 | * | |
1217 | * HACK HACK HACK HACK | |
1218 | */ | |
1219 | if (vm_map_pmap(map) == kernel_pmap) { | |
1220 | vm_map_unlock(map); /* trust me ... */ | |
1221 | } | |
1222 | else { | |
1223 | lock_set_recursive(&map->lock); | |
1224 | lock_write_to_read(&map->lock); | |
1225 | } | |
1226 | ||
1227 | entry = temp_entry; | |
1228 | while (entry != &map->header && entry->start < end) { | |
1229 | if (entry->wired_count == 1) { | |
1230 | vm_fault_wire(map, entry->start, entry->end); | |
1231 | } | |
1232 | entry = entry->next; | |
1233 | } | |
1234 | ||
1235 | if (vm_map_pmap(map) == kernel_pmap) { | |
1236 | vm_map_lock(map); | |
1237 | } | |
1238 | else { | |
1239 | lock_clear_recursive(&map->lock); | |
1240 | } | |
1241 | } | |
1242 | ||
1243 | vm_map_unlock(map); | |
1244 | ||
1245 | return(KERN_SUCCESS); | |
1246 | } | |
1247 | ||
1248 | /* | |
1249 | * vm_map_entry_unwire: [ internal use only ] | |
1250 | * | |
1251 | * Make the region specified by this entry pageable. | |
1252 | * | |
1253 | * The map in question should be locked. | |
1254 | * [This is the reason for this routine's existence.] | |
1255 | */ | |
1256 | void vm_map_entry_unwire(map, entry) | |
1257 | vm_map_t map; | |
1258 | register vm_map_entry_t entry; | |
1259 | { | |
1260 | vm_fault_unwire(map, entry->start, entry->end); | |
1261 | entry->wired_count = 0; | |
1262 | } | |
1263 | ||
1264 | /* | |
1265 | * vm_map_entry_delete: [ internal use only ] | |
1266 | * | |
1267 | * Deallocate the given entry from the target map. | |
1268 | */ | |
1269 | void vm_map_entry_delete(map, entry) | |
1270 | register vm_map_t map; | |
1271 | register vm_map_entry_t entry; | |
1272 | { | |
1273 | if (entry->wired_count != 0) | |
1274 | vm_map_entry_unwire(map, entry); | |
1275 | ||
1276 | vm_map_entry_unlink(map, entry); | |
1277 | map->size -= entry->end - entry->start; | |
1278 | ||
1279 | if (entry->is_a_map || entry->is_sub_map) | |
1280 | vm_map_deallocate(entry->object.share_map); | |
1281 | else | |
1282 | vm_object_deallocate(entry->object.vm_object); | |
1283 | ||
1284 | vm_map_entry_dispose(map, entry); | |
1285 | } | |
1286 | ||
1287 | /* | |
1288 | * vm_map_delete: [ internal use only ] | |
1289 | * | |
1290 | * Deallocates the given address range from the target | |
1291 | * map. | |
1292 | * | |
1293 | * When called with a sharing map, removes pages from | |
1294 | * that region from all physical maps. | |
1295 | */ | |
1296 | vm_map_delete(map, start, end) | |
1297 | register vm_map_t map; | |
1298 | vm_offset_t start; | |
1299 | register vm_offset_t end; | |
1300 | { | |
1301 | register vm_map_entry_t entry; | |
1302 | vm_map_entry_t first_entry; | |
1303 | ||
1304 | /* | |
1305 | * Find the start of the region, and clip it | |
1306 | */ | |
1307 | ||
1308 | if (!vm_map_lookup_entry(map, start, &first_entry)) | |
1309 | entry = first_entry->next; | |
1310 | else { | |
1311 | entry = first_entry; | |
1312 | vm_map_clip_start(map, entry, start); | |
1313 | ||
1314 | /* | |
1315 | * Fix the lookup hint now, rather than each | |
1316 | * time though the loop. | |
1317 | */ | |
1318 | ||
1319 | SAVE_HINT(map, entry->prev); | |
1320 | } | |
1321 | ||
1322 | /* | |
1323 | * Save the free space hint | |
1324 | */ | |
1325 | ||
1326 | if (map->first_free->start >= start) | |
1327 | map->first_free = entry->prev; | |
1328 | ||
1329 | /* | |
1330 | * Step through all entries in this region | |
1331 | */ | |
1332 | ||
1333 | while ((entry != &map->header) && (entry->start < end)) { | |
1334 | vm_map_entry_t next; | |
1335 | register vm_offset_t s, e; | |
1336 | register vm_object_t object; | |
1337 | ||
1338 | vm_map_clip_end(map, entry, end); | |
1339 | ||
1340 | next = entry->next; | |
1341 | s = entry->start; | |
1342 | e = entry->end; | |
1343 | ||
1344 | /* | |
1345 | * Unwire before removing addresses from the pmap; | |
1346 | * otherwise, unwiring will put the entries back in | |
1347 | * the pmap. | |
1348 | */ | |
1349 | ||
1350 | object = entry->object.vm_object; | |
1351 | if (entry->wired_count != 0) | |
1352 | vm_map_entry_unwire(map, entry); | |
1353 | ||
1354 | /* | |
1355 | * If this is a sharing map, we must remove | |
1356 | * *all* references to this data, since we can't | |
1357 | * find all of the physical maps which are sharing | |
1358 | * it. | |
1359 | */ | |
1360 | ||
1361 | if (object == kernel_object || object == kmem_object) | |
1362 | vm_object_page_remove(object, entry->offset, | |
1363 | entry->offset + (e - s)); | |
1364 | else if (!map->is_main_map) | |
1365 | vm_object_pmap_remove(object, | |
1366 | entry->offset, | |
1367 | entry->offset + (e - s)); | |
1368 | else | |
1369 | pmap_remove(map->pmap, s, e); | |
1370 | ||
1371 | /* | |
1372 | * Delete the entry (which may delete the object) | |
1373 | * only after removing all pmap entries pointing | |
1374 | * to its pages. (Otherwise, its page frames may | |
1375 | * be reallocated, and any modify bits will be | |
1376 | * set in the wrong object!) | |
1377 | */ | |
1378 | ||
1379 | vm_map_entry_delete(map, entry); | |
1380 | entry = next; | |
1381 | } | |
1382 | return(KERN_SUCCESS); | |
1383 | } | |
1384 | ||
1385 | /* | |
1386 | * vm_map_remove: | |
1387 | * | |
1388 | * Remove the given address range from the target map. | |
1389 | * This is the exported form of vm_map_delete. | |
1390 | */ | |
1391 | vm_map_remove(map, start, end) | |
1392 | register vm_map_t map; | |
1393 | register vm_offset_t start; | |
1394 | register vm_offset_t end; | |
1395 | { | |
1396 | register int result; | |
1397 | ||
1398 | vm_map_lock(map); | |
1399 | VM_MAP_RANGE_CHECK(map, start, end); | |
1400 | result = vm_map_delete(map, start, end); | |
1401 | vm_map_unlock(map); | |
1402 | ||
1403 | return(result); | |
1404 | } | |
1405 | ||
1406 | /* | |
1407 | * vm_map_check_protection: | |
1408 | * | |
1409 | * Assert that the target map allows the specified | |
1410 | * privilege on the entire address region given. | |
1411 | * The entire region must be allocated. | |
1412 | */ | |
1413 | boolean_t vm_map_check_protection(map, start, end, protection) | |
1414 | register vm_map_t map; | |
1415 | register vm_offset_t start; | |
1416 | register vm_offset_t end; | |
1417 | register vm_prot_t protection; | |
1418 | { | |
1419 | register vm_map_entry_t entry; | |
1420 | vm_map_entry_t tmp_entry; | |
1421 | ||
1422 | if (!vm_map_lookup_entry(map, start, &tmp_entry)) { | |
1423 | return(FALSE); | |
1424 | } | |
1425 | ||
1426 | entry = tmp_entry; | |
1427 | ||
1428 | while (start < end) { | |
1429 | if (entry == &map->header) { | |
1430 | return(FALSE); | |
1431 | } | |
1432 | ||
1433 | /* | |
1434 | * No holes allowed! | |
1435 | */ | |
1436 | ||
1437 | if (start < entry->start) { | |
1438 | return(FALSE); | |
1439 | } | |
1440 | ||
1441 | /* | |
1442 | * Check protection associated with entry. | |
1443 | */ | |
1444 | ||
1445 | if ((entry->protection & protection) != protection) { | |
1446 | return(FALSE); | |
1447 | } | |
1448 | ||
1449 | /* go to next entry */ | |
1450 | ||
1451 | start = entry->end; | |
1452 | entry = entry->next; | |
1453 | } | |
1454 | return(TRUE); | |
1455 | } | |
1456 | ||
1457 | /* | |
1458 | * vm_map_copy_entry: | |
1459 | * | |
1460 | * Copies the contents of the source entry to the destination | |
1461 | * entry. The entries *must* be aligned properly. | |
1462 | */ | |
1463 | void vm_map_copy_entry(src_map, dst_map, src_entry, dst_entry) | |
1464 | vm_map_t src_map, dst_map; | |
1465 | register vm_map_entry_t src_entry, dst_entry; | |
1466 | { | |
1467 | vm_object_t temp_object; | |
1468 | ||
1469 | if (src_entry->is_sub_map || dst_entry->is_sub_map) | |
1470 | return; | |
1471 | ||
5d7b9ad3 | 1472 | if (dst_entry->object.vm_object != NULL && |
175f072e KM |
1473 | !dst_entry->object.vm_object->internal) |
1474 | printf("vm_map_copy_entry: copying over permanent data!\n"); | |
1475 | ||
1476 | /* | |
1477 | * If our destination map was wired down, | |
1478 | * unwire it now. | |
1479 | */ | |
1480 | ||
1481 | if (dst_entry->wired_count != 0) | |
1482 | vm_map_entry_unwire(dst_map, dst_entry); | |
1483 | ||
1484 | /* | |
1485 | * If we're dealing with a sharing map, we | |
1486 | * must remove the destination pages from | |
1487 | * all maps (since we cannot know which maps | |
1488 | * this sharing map belongs in). | |
1489 | */ | |
1490 | ||
1491 | if (dst_map->is_main_map) | |
1492 | pmap_remove(dst_map->pmap, dst_entry->start, dst_entry->end); | |
1493 | else | |
1494 | vm_object_pmap_remove(dst_entry->object.vm_object, | |
1495 | dst_entry->offset, | |
1496 | dst_entry->offset + | |
1497 | (dst_entry->end - dst_entry->start)); | |
1498 | ||
1499 | if (src_entry->wired_count == 0) { | |
1500 | ||
1501 | boolean_t src_needs_copy; | |
1502 | ||
1503 | /* | |
1504 | * If the source entry is marked needs_copy, | |
1505 | * it is already write-protected. | |
1506 | */ | |
1507 | if (!src_entry->needs_copy) { | |
1508 | ||
1509 | boolean_t su; | |
1510 | ||
1511 | /* | |
1512 | * If the source entry has only one mapping, | |
1513 | * we can just protect the virtual address | |
1514 | * range. | |
1515 | */ | |
1516 | if (!(su = src_map->is_main_map)) { | |
1517 | simple_lock(&src_map->ref_lock); | |
1518 | su = (src_map->ref_count == 1); | |
1519 | simple_unlock(&src_map->ref_lock); | |
1520 | } | |
1521 | ||
1522 | if (su) { | |
1523 | pmap_protect(src_map->pmap, | |
1524 | src_entry->start, | |
1525 | src_entry->end, | |
1526 | src_entry->protection & ~VM_PROT_WRITE); | |
1527 | } | |
1528 | else { | |
1529 | vm_object_pmap_copy(src_entry->object.vm_object, | |
1530 | src_entry->offset, | |
1531 | src_entry->offset + (src_entry->end | |
1532 | -src_entry->start)); | |
1533 | } | |
1534 | } | |
1535 | ||
1536 | /* | |
1537 | * Make a copy of the object. | |
1538 | */ | |
1539 | temp_object = dst_entry->object.vm_object; | |
1540 | vm_object_copy(src_entry->object.vm_object, | |
1541 | src_entry->offset, | |
1542 | (vm_size_t)(src_entry->end - | |
1543 | src_entry->start), | |
1544 | &dst_entry->object.vm_object, | |
1545 | &dst_entry->offset, | |
1546 | &src_needs_copy); | |
1547 | /* | |
1548 | * If we didn't get a copy-object now, mark the | |
1549 | * source map entry so that a shadow will be created | |
1550 | * to hold its changed pages. | |
1551 | */ | |
1552 | if (src_needs_copy) | |
1553 | src_entry->needs_copy = TRUE; | |
1554 | ||
1555 | /* | |
1556 | * The destination always needs to have a shadow | |
1557 | * created. | |
1558 | */ | |
1559 | dst_entry->needs_copy = TRUE; | |
1560 | ||
1561 | /* | |
1562 | * Mark the entries copy-on-write, so that write-enabling | |
1563 | * the entry won't make copy-on-write pages writable. | |
1564 | */ | |
1565 | src_entry->copy_on_write = TRUE; | |
1566 | dst_entry->copy_on_write = TRUE; | |
1567 | /* | |
1568 | * Get rid of the old object. | |
1569 | */ | |
1570 | vm_object_deallocate(temp_object); | |
1571 | ||
1572 | pmap_copy(dst_map->pmap, src_map->pmap, dst_entry->start, | |
1573 | dst_entry->end - dst_entry->start, src_entry->start); | |
1574 | } | |
1575 | else { | |
1576 | /* | |
1577 | * Of course, wired down pages can't be set copy-on-write. | |
1578 | * Cause wired pages to be copied into the new | |
1579 | * map by simulating faults (the new pages are | |
1580 | * pageable) | |
1581 | */ | |
1582 | vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry); | |
1583 | } | |
1584 | } | |
1585 | ||
1586 | /* | |
1587 | * vm_map_copy: | |
1588 | * | |
1589 | * Perform a virtual memory copy from the source | |
1590 | * address map/range to the destination map/range. | |
1591 | * | |
1592 | * If src_destroy or dst_alloc is requested, | |
1593 | * the source and destination regions should be | |
1594 | * disjoint, not only in the top-level map, but | |
1595 | * in the sharing maps as well. [The best way | |
1596 | * to guarantee this is to use a new intermediate | |
1597 | * map to make copies. This also reduces map | |
1598 | * fragmentation.] | |
1599 | */ | |
1600 | vm_map_copy(dst_map, src_map, | |
1601 | dst_addr, len, src_addr, | |
1602 | dst_alloc, src_destroy) | |
1603 | vm_map_t dst_map; | |
1604 | vm_map_t src_map; | |
1605 | vm_offset_t dst_addr; | |
1606 | vm_size_t len; | |
1607 | vm_offset_t src_addr; | |
1608 | boolean_t dst_alloc; | |
1609 | boolean_t src_destroy; | |
1610 | { | |
1611 | register | |
1612 | vm_map_entry_t src_entry; | |
1613 | register | |
1614 | vm_map_entry_t dst_entry; | |
1615 | vm_map_entry_t tmp_entry; | |
1616 | vm_offset_t src_start; | |
1617 | vm_offset_t src_end; | |
1618 | vm_offset_t dst_start; | |
1619 | vm_offset_t dst_end; | |
1620 | vm_offset_t src_clip; | |
1621 | vm_offset_t dst_clip; | |
1622 | int result; | |
1623 | boolean_t old_src_destroy; | |
1624 | ||
1625 | /* | |
1626 | * XXX While we figure out why src_destroy screws up, | |
1627 | * we'll do it by explicitly vm_map_delete'ing at the end. | |
1628 | */ | |
1629 | ||
1630 | old_src_destroy = src_destroy; | |
1631 | src_destroy = FALSE; | |
1632 | ||
1633 | /* | |
1634 | * Compute start and end of region in both maps | |
1635 | */ | |
1636 | ||
1637 | src_start = src_addr; | |
1638 | src_end = src_start + len; | |
1639 | dst_start = dst_addr; | |
1640 | dst_end = dst_start + len; | |
1641 | ||
1642 | /* | |
1643 | * Check that the region can exist in both source | |
1644 | * and destination. | |
1645 | */ | |
1646 | ||
1647 | if ((dst_end < dst_start) || (src_end < src_start)) | |
1648 | return(KERN_NO_SPACE); | |
1649 | ||
1650 | /* | |
1651 | * Lock the maps in question -- we avoid deadlock | |
1652 | * by ordering lock acquisition by map value | |
1653 | */ | |
1654 | ||
1655 | if (src_map == dst_map) { | |
1656 | vm_map_lock(src_map); | |
1657 | } | |
1658 | else if ((int) src_map < (int) dst_map) { | |
1659 | vm_map_lock(src_map); | |
1660 | vm_map_lock(dst_map); | |
1661 | } else { | |
1662 | vm_map_lock(dst_map); | |
1663 | vm_map_lock(src_map); | |
1664 | } | |
1665 | ||
1666 | result = KERN_SUCCESS; | |
1667 | ||
1668 | /* | |
1669 | * Check protections... source must be completely readable and | |
1670 | * destination must be completely writable. [Note that if we're | |
1671 | * allocating the destination region, we don't have to worry | |
1672 | * about protection, but instead about whether the region | |
1673 | * exists.] | |
1674 | */ | |
1675 | ||
1676 | if (src_map->is_main_map && dst_map->is_main_map) { | |
1677 | if (!vm_map_check_protection(src_map, src_start, src_end, | |
1678 | VM_PROT_READ)) { | |
1679 | result = KERN_PROTECTION_FAILURE; | |
1680 | goto Return; | |
1681 | } | |
1682 | ||
1683 | if (dst_alloc) { | |
1684 | /* XXX Consider making this a vm_map_find instead */ | |
5d7b9ad3 | 1685 | if ((result = vm_map_insert(dst_map, NULL, |
175f072e KM |
1686 | (vm_offset_t) 0, dst_start, dst_end)) != KERN_SUCCESS) |
1687 | goto Return; | |
1688 | } | |
1689 | else if (!vm_map_check_protection(dst_map, dst_start, dst_end, | |
1690 | VM_PROT_WRITE)) { | |
1691 | result = KERN_PROTECTION_FAILURE; | |
1692 | goto Return; | |
1693 | } | |
1694 | } | |
1695 | ||
1696 | /* | |
1697 | * Find the start entries and clip. | |
1698 | * | |
1699 | * Note that checking protection asserts that the | |
1700 | * lookup cannot fail. | |
1701 | * | |
1702 | * Also note that we wait to do the second lookup | |
1703 | * until we have done the first clip, as the clip | |
1704 | * may affect which entry we get! | |
1705 | */ | |
1706 | ||
1707 | (void) vm_map_lookup_entry(src_map, src_addr, &tmp_entry); | |
1708 | src_entry = tmp_entry; | |
1709 | vm_map_clip_start(src_map, src_entry, src_start); | |
1710 | ||
1711 | (void) vm_map_lookup_entry(dst_map, dst_addr, &tmp_entry); | |
1712 | dst_entry = tmp_entry; | |
1713 | vm_map_clip_start(dst_map, dst_entry, dst_start); | |
1714 | ||
1715 | /* | |
1716 | * If both source and destination entries are the same, | |
1717 | * retry the first lookup, as it may have changed. | |
1718 | */ | |
1719 | ||
1720 | if (src_entry == dst_entry) { | |
1721 | (void) vm_map_lookup_entry(src_map, src_addr, &tmp_entry); | |
1722 | src_entry = tmp_entry; | |
1723 | } | |
1724 | ||
1725 | /* | |
1726 | * If source and destination entries are still the same, | |
1727 | * a null copy is being performed. | |
1728 | */ | |
1729 | ||
1730 | if (src_entry == dst_entry) | |
1731 | goto Return; | |
1732 | ||
1733 | /* | |
1734 | * Go through entries until we get to the end of the | |
1735 | * region. | |
1736 | */ | |
1737 | ||
1738 | while (src_start < src_end) { | |
1739 | /* | |
1740 | * Clip the entries to the endpoint of the entire region. | |
1741 | */ | |
1742 | ||
1743 | vm_map_clip_end(src_map, src_entry, src_end); | |
1744 | vm_map_clip_end(dst_map, dst_entry, dst_end); | |
1745 | ||
1746 | /* | |
1747 | * Clip each entry to the endpoint of the other entry. | |
1748 | */ | |
1749 | ||
1750 | src_clip = src_entry->start + (dst_entry->end - dst_entry->start); | |
1751 | vm_map_clip_end(src_map, src_entry, src_clip); | |
1752 | ||
1753 | dst_clip = dst_entry->start + (src_entry->end - src_entry->start); | |
1754 | vm_map_clip_end(dst_map, dst_entry, dst_clip); | |
1755 | ||
1756 | /* | |
1757 | * Both entries now match in size and relative endpoints. | |
1758 | * | |
1759 | * If both entries refer to a VM object, we can | |
1760 | * deal with them now. | |
1761 | */ | |
1762 | ||
1763 | if (!src_entry->is_a_map && !dst_entry->is_a_map) { | |
1764 | vm_map_copy_entry(src_map, dst_map, src_entry, | |
1765 | dst_entry); | |
1766 | } | |
1767 | else { | |
1768 | register vm_map_t new_dst_map; | |
1769 | vm_offset_t new_dst_start; | |
1770 | vm_size_t new_size; | |
1771 | vm_map_t new_src_map; | |
1772 | vm_offset_t new_src_start; | |
1773 | ||
1774 | /* | |
1775 | * We have to follow at least one sharing map. | |
1776 | */ | |
1777 | ||
1778 | new_size = (dst_entry->end - dst_entry->start); | |
1779 | ||
1780 | if (src_entry->is_a_map) { | |
1781 | new_src_map = src_entry->object.share_map; | |
1782 | new_src_start = src_entry->offset; | |
1783 | } | |
1784 | else { | |
1785 | new_src_map = src_map; | |
1786 | new_src_start = src_entry->start; | |
1787 | lock_set_recursive(&src_map->lock); | |
1788 | } | |
1789 | ||
1790 | if (dst_entry->is_a_map) { | |
1791 | vm_offset_t new_dst_end; | |
1792 | ||
1793 | new_dst_map = dst_entry->object.share_map; | |
1794 | new_dst_start = dst_entry->offset; | |
1795 | ||
1796 | /* | |
1797 | * Since the destination sharing entries | |
1798 | * will be merely deallocated, we can | |
1799 | * do that now, and replace the region | |
1800 | * with a null object. [This prevents | |
1801 | * splitting the source map to match | |
1802 | * the form of the destination map.] | |
1803 | * Note that we can only do so if the | |
1804 | * source and destination do not overlap. | |
1805 | */ | |
1806 | ||
1807 | new_dst_end = new_dst_start + new_size; | |
1808 | ||
1809 | if (new_dst_map != new_src_map) { | |
1810 | vm_map_lock(new_dst_map); | |
1811 | (void) vm_map_delete(new_dst_map, | |
1812 | new_dst_start, | |
1813 | new_dst_end); | |
1814 | (void) vm_map_insert(new_dst_map, | |
5d7b9ad3 | 1815 | NULL, |
175f072e KM |
1816 | (vm_offset_t) 0, |
1817 | new_dst_start, | |
1818 | new_dst_end); | |
1819 | vm_map_unlock(new_dst_map); | |
1820 | } | |
1821 | } | |
1822 | else { | |
1823 | new_dst_map = dst_map; | |
1824 | new_dst_start = dst_entry->start; | |
1825 | lock_set_recursive(&dst_map->lock); | |
1826 | } | |
1827 | ||
1828 | /* | |
1829 | * Recursively copy the sharing map. | |
1830 | */ | |
1831 | ||
1832 | (void) vm_map_copy(new_dst_map, new_src_map, | |
1833 | new_dst_start, new_size, new_src_start, | |
1834 | FALSE, FALSE); | |
1835 | ||
1836 | if (dst_map == new_dst_map) | |
1837 | lock_clear_recursive(&dst_map->lock); | |
1838 | if (src_map == new_src_map) | |
1839 | lock_clear_recursive(&src_map->lock); | |
1840 | } | |
1841 | ||
1842 | /* | |
1843 | * Update variables for next pass through the loop. | |
1844 | */ | |
1845 | ||
1846 | src_start = src_entry->end; | |
1847 | src_entry = src_entry->next; | |
1848 | dst_start = dst_entry->end; | |
1849 | dst_entry = dst_entry->next; | |
1850 | ||
1851 | /* | |
1852 | * If the source is to be destroyed, here is the | |
1853 | * place to do it. | |
1854 | */ | |
1855 | ||
1856 | if (src_destroy && src_map->is_main_map && | |
1857 | dst_map->is_main_map) | |
1858 | vm_map_entry_delete(src_map, src_entry->prev); | |
1859 | } | |
1860 | ||
1861 | /* | |
1862 | * Update the physical maps as appropriate | |
1863 | */ | |
1864 | ||
1865 | if (src_map->is_main_map && dst_map->is_main_map) { | |
1866 | if (src_destroy) | |
1867 | pmap_remove(src_map->pmap, src_addr, src_addr + len); | |
1868 | } | |
1869 | ||
1870 | /* | |
1871 | * Unlock the maps | |
1872 | */ | |
1873 | ||
1874 | Return: ; | |
1875 | ||
1876 | if (old_src_destroy) | |
1877 | vm_map_delete(src_map, src_addr, src_addr + len); | |
1878 | ||
1879 | vm_map_unlock(src_map); | |
1880 | if (src_map != dst_map) | |
1881 | vm_map_unlock(dst_map); | |
1882 | ||
1883 | return(result); | |
1884 | } | |
1885 | ||
1886 | /* | |
5d7b9ad3 MK |
1887 | * vmspace_fork: |
1888 | * Create a new process vmspace structure and vm_map | |
1889 | * based on those of an existing process. The new map | |
1890 | * is based on the old map, according to the inheritance | |
1891 | * values on the regions in that map. | |
175f072e | 1892 | * |
5d7b9ad3 | 1893 | * The source map must not be locked. |
175f072e | 1894 | */ |
5d7b9ad3 MK |
1895 | struct vmspace * |
1896 | vmspace_fork(vm1) | |
1897 | register struct vmspace *vm1; | |
175f072e | 1898 | { |
5d7b9ad3 MK |
1899 | register struct vmspace *vm2; |
1900 | vm_map_t old_map = &vm1->vm_map; | |
175f072e KM |
1901 | vm_map_t new_map; |
1902 | vm_map_entry_t old_entry; | |
1903 | vm_map_entry_t new_entry; | |
1904 | pmap_t new_pmap; | |
1905 | ||
1906 | vm_map_lock(old_map); | |
1907 | ||
5d7b9ad3 MK |
1908 | vm2 = vmspace_alloc(old_map->min_offset, old_map->max_offset, |
1909 | old_map->entries_pageable); | |
1910 | bcopy(&vm1->vm_startcopy, &vm2->vm_startcopy, | |
1911 | (caddr_t) (vm1 + 1) - (caddr_t) &vm1->vm_startcopy); | |
1912 | new_pmap = &vm2->vm_pmap; /* XXX */ | |
1913 | new_map = &vm2->vm_map; /* XXX */ | |
175f072e KM |
1914 | |
1915 | old_entry = old_map->header.next; | |
1916 | ||
1917 | while (old_entry != &old_map->header) { | |
1918 | if (old_entry->is_sub_map) | |
1919 | panic("vm_map_fork: encountered a submap"); | |
1920 | ||
1921 | switch (old_entry->inheritance) { | |
1922 | case VM_INHERIT_NONE: | |
1923 | break; | |
1924 | ||
1925 | case VM_INHERIT_SHARE: | |
1926 | /* | |
1927 | * If we don't already have a sharing map: | |
1928 | */ | |
1929 | ||
1930 | if (!old_entry->is_a_map) { | |
1931 | vm_map_t new_share_map; | |
1932 | vm_map_entry_t new_share_entry; | |
1933 | ||
1934 | /* | |
1935 | * Create a new sharing map | |
1936 | */ | |
1937 | ||
5d7b9ad3 | 1938 | new_share_map = vm_map_create(NULL, |
175f072e KM |
1939 | old_entry->start, |
1940 | old_entry->end, | |
1941 | TRUE); | |
1942 | new_share_map->is_main_map = FALSE; | |
1943 | ||
1944 | /* | |
1945 | * Create the only sharing entry from the | |
1946 | * old task map entry. | |
1947 | */ | |
1948 | ||
1949 | new_share_entry = | |
1950 | vm_map_entry_create(new_share_map); | |
1951 | *new_share_entry = *old_entry; | |
1952 | ||
1953 | /* | |
1954 | * Insert the entry into the new sharing | |
1955 | * map | |
1956 | */ | |
1957 | ||
1958 | vm_map_entry_link(new_share_map, | |
1959 | new_share_map->header.prev, | |
1960 | new_share_entry); | |
1961 | ||
1962 | /* | |
1963 | * Fix up the task map entry to refer | |
1964 | * to the sharing map now. | |
1965 | */ | |
1966 | ||
1967 | old_entry->is_a_map = TRUE; | |
1968 | old_entry->object.share_map = new_share_map; | |
1969 | old_entry->offset = old_entry->start; | |
1970 | } | |
1971 | ||
1972 | /* | |
1973 | * Clone the entry, referencing the sharing map. | |
1974 | */ | |
1975 | ||
1976 | new_entry = vm_map_entry_create(new_map); | |
1977 | *new_entry = *old_entry; | |
1978 | vm_map_reference(new_entry->object.share_map); | |
1979 | ||
1980 | /* | |
1981 | * Insert the entry into the new map -- we | |
1982 | * know we're inserting at the end of the new | |
1983 | * map. | |
1984 | */ | |
1985 | ||
1986 | vm_map_entry_link(new_map, new_map->header.prev, | |
1987 | new_entry); | |
1988 | ||
1989 | /* | |
1990 | * Update the physical map | |
1991 | */ | |
1992 | ||
1993 | pmap_copy(new_map->pmap, old_map->pmap, | |
1994 | new_entry->start, | |
1995 | (old_entry->end - old_entry->start), | |
1996 | old_entry->start); | |
1997 | break; | |
1998 | ||
1999 | case VM_INHERIT_COPY: | |
2000 | /* | |
2001 | * Clone the entry and link into the map. | |
2002 | */ | |
2003 | ||
2004 | new_entry = vm_map_entry_create(new_map); | |
2005 | *new_entry = *old_entry; | |
2006 | new_entry->wired_count = 0; | |
5d7b9ad3 | 2007 | new_entry->object.vm_object = NULL; |
175f072e KM |
2008 | new_entry->is_a_map = FALSE; |
2009 | vm_map_entry_link(new_map, new_map->header.prev, | |
2010 | new_entry); | |
2011 | if (old_entry->is_a_map) { | |
2012 | int check; | |
2013 | ||
2014 | check = vm_map_copy(new_map, | |
2015 | old_entry->object.share_map, | |
2016 | new_entry->start, | |
2017 | (vm_size_t)(new_entry->end - | |
2018 | new_entry->start), | |
2019 | old_entry->offset, | |
2020 | FALSE, FALSE); | |
2021 | if (check != KERN_SUCCESS) | |
2022 | printf("vm_map_fork: copy in share_map region failed\n"); | |
2023 | } | |
2024 | else { | |
2025 | vm_map_copy_entry(old_map, new_map, old_entry, | |
2026 | new_entry); | |
2027 | } | |
2028 | break; | |
2029 | } | |
2030 | old_entry = old_entry->next; | |
2031 | } | |
2032 | ||
2033 | new_map->size = old_map->size; | |
2034 | vm_map_unlock(old_map); | |
2035 | ||
5d7b9ad3 | 2036 | return(vm2); |
175f072e KM |
2037 | } |
2038 | ||
2039 | /* | |
2040 | * vm_map_lookup: | |
2041 | * | |
2042 | * Finds the VM object, offset, and | |
2043 | * protection for a given virtual address in the | |
2044 | * specified map, assuming a page fault of the | |
2045 | * type specified. | |
2046 | * | |
2047 | * Leaves the map in question locked for read; return | |
2048 | * values are guaranteed until a vm_map_lookup_done | |
2049 | * call is performed. Note that the map argument | |
2050 | * is in/out; the returned map must be used in | |
2051 | * the call to vm_map_lookup_done. | |
2052 | * | |
2053 | * A handle (out_entry) is returned for use in | |
2054 | * vm_map_lookup_done, to make that fast. | |
2055 | * | |
2056 | * If a lookup is requested with "write protection" | |
2057 | * specified, the map may be changed to perform virtual | |
2058 | * copying operations, although the data referenced will | |
2059 | * remain the same. | |
2060 | */ | |
2061 | vm_map_lookup(var_map, vaddr, fault_type, out_entry, | |
2062 | object, offset, out_prot, wired, single_use) | |
2063 | vm_map_t *var_map; /* IN/OUT */ | |
2064 | register vm_offset_t vaddr; | |
2065 | register vm_prot_t fault_type; | |
2066 | ||
2067 | vm_map_entry_t *out_entry; /* OUT */ | |
2068 | vm_object_t *object; /* OUT */ | |
2069 | vm_offset_t *offset; /* OUT */ | |
2070 | vm_prot_t *out_prot; /* OUT */ | |
2071 | boolean_t *wired; /* OUT */ | |
2072 | boolean_t *single_use; /* OUT */ | |
2073 | { | |
2074 | vm_map_t share_map; | |
2075 | vm_offset_t share_offset; | |
2076 | register vm_map_entry_t entry; | |
2077 | register vm_map_t map = *var_map; | |
2078 | register vm_prot_t prot; | |
2079 | register boolean_t su; | |
2080 | ||
2081 | RetryLookup: ; | |
2082 | ||
2083 | /* | |
2084 | * Lookup the faulting address. | |
2085 | */ | |
2086 | ||
2087 | vm_map_lock_read(map); | |
2088 | ||
2089 | #define RETURN(why) \ | |
2090 | { \ | |
2091 | vm_map_unlock_read(map); \ | |
2092 | return(why); \ | |
2093 | } | |
2094 | ||
2095 | /* | |
2096 | * If the map has an interesting hint, try it before calling | |
2097 | * full blown lookup routine. | |
2098 | */ | |
2099 | ||
2100 | simple_lock(&map->hint_lock); | |
2101 | entry = map->hint; | |
2102 | simple_unlock(&map->hint_lock); | |
2103 | ||
2104 | *out_entry = entry; | |
2105 | ||
2106 | if ((entry == &map->header) || | |
2107 | (vaddr < entry->start) || (vaddr >= entry->end)) { | |
2108 | vm_map_entry_t tmp_entry; | |
2109 | ||
2110 | /* | |
2111 | * Entry was either not a valid hint, or the vaddr | |
2112 | * was not contained in the entry, so do a full lookup. | |
2113 | */ | |
2114 | if (!vm_map_lookup_entry(map, vaddr, &tmp_entry)) | |
2115 | RETURN(KERN_INVALID_ADDRESS); | |
2116 | ||
2117 | entry = tmp_entry; | |
2118 | *out_entry = entry; | |
2119 | } | |
2120 | ||
2121 | /* | |
2122 | * Handle submaps. | |
2123 | */ | |
2124 | ||
2125 | if (entry->is_sub_map) { | |
2126 | vm_map_t old_map = map; | |
2127 | ||
2128 | *var_map = map = entry->object.sub_map; | |
2129 | vm_map_unlock_read(old_map); | |
2130 | goto RetryLookup; | |
2131 | } | |
2132 | ||
2133 | /* | |
2134 | * Check whether this task is allowed to have | |
2135 | * this page. | |
2136 | */ | |
2137 | ||
2138 | prot = entry->protection; | |
2139 | if ((fault_type & (prot)) != fault_type) | |
2140 | RETURN(KERN_PROTECTION_FAILURE); | |
2141 | ||
2142 | /* | |
2143 | * If this page is not pageable, we have to get | |
2144 | * it for all possible accesses. | |
2145 | */ | |
2146 | ||
2147 | if (*wired = (entry->wired_count != 0)) | |
2148 | prot = fault_type = entry->protection; | |
2149 | ||
2150 | /* | |
2151 | * If we don't already have a VM object, track | |
2152 | * it down. | |
2153 | */ | |
2154 | ||
2155 | if (su = !entry->is_a_map) { | |
2156 | share_map = map; | |
2157 | share_offset = vaddr; | |
2158 | } | |
2159 | else { | |
2160 | vm_map_entry_t share_entry; | |
2161 | ||
2162 | /* | |
2163 | * Compute the sharing map, and offset into it. | |
2164 | */ | |
2165 | ||
2166 | share_map = entry->object.share_map; | |
2167 | share_offset = (vaddr - entry->start) + entry->offset; | |
2168 | ||
2169 | /* | |
2170 | * Look for the backing store object and offset | |
2171 | */ | |
2172 | ||
2173 | vm_map_lock_read(share_map); | |
2174 | ||
2175 | if (!vm_map_lookup_entry(share_map, share_offset, | |
2176 | &share_entry)) { | |
2177 | vm_map_unlock_read(share_map); | |
2178 | RETURN(KERN_INVALID_ADDRESS); | |
2179 | } | |
2180 | entry = share_entry; | |
2181 | } | |
2182 | ||
2183 | /* | |
2184 | * If the entry was copy-on-write, we either ... | |
2185 | */ | |
2186 | ||
2187 | if (entry->needs_copy) { | |
2188 | /* | |
2189 | * If we want to write the page, we may as well | |
2190 | * handle that now since we've got the sharing | |
2191 | * map locked. | |
2192 | * | |
2193 | * If we don't need to write the page, we just | |
2194 | * demote the permissions allowed. | |
2195 | */ | |
2196 | ||
2197 | if (fault_type & VM_PROT_WRITE) { | |
2198 | /* | |
2199 | * Make a new object, and place it in the | |
2200 | * object chain. Note that no new references | |
2201 | * have appeared -- one just moved from the | |
2202 | * share map to the new object. | |
2203 | */ | |
2204 | ||
2205 | if (lock_read_to_write(&share_map->lock)) { | |
2206 | if (share_map != map) | |
2207 | vm_map_unlock_read(map); | |
2208 | goto RetryLookup; | |
2209 | } | |
2210 | ||
2211 | vm_object_shadow( | |
2212 | &entry->object.vm_object, | |
2213 | &entry->offset, | |
2214 | (vm_size_t) (entry->end - entry->start)); | |
2215 | ||
2216 | entry->needs_copy = FALSE; | |
2217 | ||
2218 | lock_write_to_read(&share_map->lock); | |
2219 | } | |
2220 | else { | |
2221 | /* | |
2222 | * We're attempting to read a copy-on-write | |
2223 | * page -- don't allow writes. | |
2224 | */ | |
2225 | ||
2226 | prot &= (~VM_PROT_WRITE); | |
2227 | } | |
2228 | } | |
2229 | ||
2230 | /* | |
2231 | * Create an object if necessary. | |
2232 | */ | |
5d7b9ad3 | 2233 | if (entry->object.vm_object == NULL) { |
175f072e KM |
2234 | |
2235 | if (lock_read_to_write(&share_map->lock)) { | |
2236 | if (share_map != map) | |
2237 | vm_map_unlock_read(map); | |
2238 | goto RetryLookup; | |
2239 | } | |
2240 | ||
2241 | entry->object.vm_object = vm_object_allocate( | |
2242 | (vm_size_t)(entry->end - entry->start)); | |
2243 | entry->offset = 0; | |
2244 | lock_write_to_read(&share_map->lock); | |
2245 | } | |
2246 | ||
2247 | /* | |
2248 | * Return the object/offset from this entry. If the entry | |
2249 | * was copy-on-write or empty, it has been fixed up. | |
2250 | */ | |
2251 | ||
2252 | *offset = (share_offset - entry->start) + entry->offset; | |
2253 | *object = entry->object.vm_object; | |
2254 | ||
2255 | /* | |
2256 | * Return whether this is the only map sharing this data. | |
2257 | */ | |
2258 | ||
2259 | if (!su) { | |
2260 | simple_lock(&share_map->ref_lock); | |
2261 | su = (share_map->ref_count == 1); | |
2262 | simple_unlock(&share_map->ref_lock); | |
2263 | } | |
2264 | ||
2265 | *out_prot = prot; | |
2266 | *single_use = su; | |
2267 | ||
2268 | return(KERN_SUCCESS); | |
2269 | ||
2270 | #undef RETURN | |
2271 | } | |
2272 | ||
2273 | /* | |
2274 | * vm_map_lookup_done: | |
2275 | * | |
2276 | * Releases locks acquired by a vm_map_lookup | |
2277 | * (according to the handle returned by that lookup). | |
2278 | */ | |
2279 | ||
2280 | void vm_map_lookup_done(map, entry) | |
2281 | register vm_map_t map; | |
2282 | vm_map_entry_t entry; | |
2283 | { | |
2284 | /* | |
2285 | * If this entry references a map, unlock it first. | |
2286 | */ | |
2287 | ||
2288 | if (entry->is_a_map) | |
2289 | vm_map_unlock_read(entry->object.share_map); | |
2290 | ||
2291 | /* | |
2292 | * Unlock the main-level map | |
2293 | */ | |
2294 | ||
2295 | vm_map_unlock_read(map); | |
2296 | } | |
2297 | ||
2298 | /* | |
2299 | * Routine: vm_map_simplify | |
2300 | * Purpose: | |
2301 | * Attempt to simplify the map representation in | |
2302 | * the vicinity of the given starting address. | |
2303 | * Note: | |
2304 | * This routine is intended primarily to keep the | |
2305 | * kernel maps more compact -- they generally don't | |
2306 | * benefit from the "expand a map entry" technology | |
2307 | * at allocation time because the adjacent entry | |
2308 | * is often wired down. | |
2309 | */ | |
2310 | void vm_map_simplify(map, start) | |
2311 | vm_map_t map; | |
2312 | vm_offset_t start; | |
2313 | { | |
2314 | vm_map_entry_t this_entry; | |
2315 | vm_map_entry_t prev_entry; | |
2316 | ||
2317 | vm_map_lock(map); | |
2318 | if ( | |
2319 | (vm_map_lookup_entry(map, start, &this_entry)) && | |
2320 | ((prev_entry = this_entry->prev) != &map->header) && | |
2321 | ||
2322 | (prev_entry->end == start) && | |
2323 | (map->is_main_map) && | |
2324 | ||
2325 | (prev_entry->is_a_map == FALSE) && | |
2326 | (prev_entry->is_sub_map == FALSE) && | |
2327 | ||
2328 | (this_entry->is_a_map == FALSE) && | |
2329 | (this_entry->is_sub_map == FALSE) && | |
2330 | ||
2331 | (prev_entry->inheritance == this_entry->inheritance) && | |
2332 | (prev_entry->protection == this_entry->protection) && | |
2333 | (prev_entry->max_protection == this_entry->max_protection) && | |
2334 | (prev_entry->wired_count == this_entry->wired_count) && | |
2335 | ||
2336 | (prev_entry->copy_on_write == this_entry->copy_on_write) && | |
2337 | (prev_entry->needs_copy == this_entry->needs_copy) && | |
2338 | ||
2339 | (prev_entry->object.vm_object == this_entry->object.vm_object) && | |
2340 | ((prev_entry->offset + (prev_entry->end - prev_entry->start)) | |
2341 | == this_entry->offset) | |
2342 | ) { | |
2343 | if (map->first_free == this_entry) | |
2344 | map->first_free = prev_entry; | |
2345 | ||
2346 | SAVE_HINT(map, prev_entry); | |
2347 | vm_map_entry_unlink(map, this_entry); | |
2348 | prev_entry->end = this_entry->end; | |
2349 | vm_object_deallocate(this_entry->object.vm_object); | |
2350 | vm_map_entry_dispose(map, this_entry); | |
2351 | } | |
2352 | vm_map_unlock(map); | |
2353 | } | |
2354 | ||
2355 | /* | |
2356 | * vm_map_print: [ debug ] | |
2357 | */ | |
2358 | void vm_map_print(map, full) | |
2359 | register vm_map_t map; | |
2360 | boolean_t full; | |
2361 | { | |
2362 | register vm_map_entry_t entry; | |
2363 | extern int indent; | |
2364 | ||
2365 | iprintf("%s map 0x%x: pmap=0x%x,ref=%d,nentries=%d,version=%d\n", | |
2366 | (map->is_main_map ? "Task" : "Share"), | |
2367 | (int) map, (int) (map->pmap), map->ref_count, map->nentries, | |
2368 | map->timestamp); | |
2369 | ||
2370 | if (!full && indent) | |
2371 | return; | |
2372 | ||
2373 | indent += 2; | |
2374 | for (entry = map->header.next; entry != &map->header; | |
2375 | entry = entry->next) { | |
2376 | iprintf("map entry 0x%x: start=0x%x, end=0x%x, ", | |
2377 | (int) entry, (int) entry->start, (int) entry->end); | |
2378 | if (map->is_main_map) { | |
2379 | static char *inheritance_name[4] = | |
2380 | { "share", "copy", "none", "donate_copy"}; | |
2381 | printf("prot=%x/%x/%s, ", | |
2382 | entry->protection, | |
2383 | entry->max_protection, | |
2384 | inheritance_name[entry->inheritance]); | |
2385 | if (entry->wired_count != 0) | |
2386 | printf("wired, "); | |
2387 | } | |
2388 | ||
2389 | if (entry->is_a_map || entry->is_sub_map) { | |
2390 | printf("share=0x%x, offset=0x%x\n", | |
2391 | (int) entry->object.share_map, | |
2392 | (int) entry->offset); | |
2393 | if ((entry->prev == &map->header) || | |
2394 | (!entry->prev->is_a_map) || | |
2395 | (entry->prev->object.share_map != | |
2396 | entry->object.share_map)) { | |
2397 | indent += 2; | |
2398 | vm_map_print(entry->object.share_map, full); | |
2399 | indent -= 2; | |
2400 | } | |
2401 | ||
2402 | } | |
2403 | else { | |
2404 | printf("object=0x%x, offset=0x%x", | |
2405 | (int) entry->object.vm_object, | |
2406 | (int) entry->offset); | |
2407 | if (entry->copy_on_write) | |
2408 | printf(", copy (%s)", | |
2409 | entry->needs_copy ? "needed" : "done"); | |
2410 | printf("\n"); | |
2411 | ||
2412 | if ((entry->prev == &map->header) || | |
2413 | (entry->prev->is_a_map) || | |
2414 | (entry->prev->object.vm_object != | |
2415 | entry->object.vm_object)) { | |
2416 | indent += 2; | |
2417 | vm_object_print(entry->object.vm_object, full); | |
2418 | indent -= 2; | |
2419 | } | |
2420 | } | |
2421 | } | |
2422 | indent -= 2; | |
2423 | } |