Commit | Line | Data |
---|---|---|
3241b1d3 JT |
1 | /* |
2 | * Copyright (C) 2011 Red Hat, Inc. | |
3 | * | |
4 | * This file is released under the GPL. | |
5 | */ | |
6 | #include "dm-block-manager.h" | |
7 | #include "dm-persistent-data-internal.h" | |
8 | #include "../dm-bufio.h" | |
9 | ||
10 | #include <linux/crc32c.h> | |
11 | #include <linux/module.h> | |
12 | #include <linux/slab.h> | |
13 | #include <linux/rwsem.h> | |
14 | #include <linux/device-mapper.h> | |
15 | #include <linux/stacktrace.h> | |
16 | ||
17 | #define DM_MSG_PREFIX "block manager" | |
18 | ||
19 | /*----------------------------------------------------------------*/ | |
20 | ||
21 | /* | |
22 | * This is a read/write semaphore with a couple of differences. | |
23 | * | |
24 | * i) There is a restriction on the number of concurrent read locks that | |
25 | * may be held at once. This is just an implementation detail. | |
26 | * | |
27 | * ii) Recursive locking attempts are detected and return EINVAL. A stack | |
28 | * trace is also emitted for the previous lock aquisition. | |
29 | * | |
30 | * iii) Priority is given to write locks. | |
31 | */ | |
32 | #define MAX_HOLDERS 4 | |
33 | #define MAX_STACK 10 | |
34 | ||
35 | typedef unsigned long stack_entries[MAX_STACK]; | |
36 | ||
37 | struct block_lock { | |
38 | spinlock_t lock; | |
39 | __s32 count; | |
40 | struct list_head waiters; | |
41 | struct task_struct *holders[MAX_HOLDERS]; | |
42 | ||
43 | #ifdef CONFIG_DM_DEBUG_BLOCK_STACK_TRACING | |
44 | struct stack_trace traces[MAX_HOLDERS]; | |
45 | stack_entries entries[MAX_HOLDERS]; | |
46 | #endif | |
47 | }; | |
48 | ||
49 | struct waiter { | |
50 | struct list_head list; | |
51 | struct task_struct *task; | |
52 | int wants_write; | |
53 | }; | |
54 | ||
55 | static unsigned __find_holder(struct block_lock *lock, | |
56 | struct task_struct *task) | |
57 | { | |
58 | unsigned i; | |
59 | ||
60 | for (i = 0; i < MAX_HOLDERS; i++) | |
61 | if (lock->holders[i] == task) | |
62 | break; | |
63 | ||
64 | BUG_ON(i == MAX_HOLDERS); | |
65 | return i; | |
66 | } | |
67 | ||
68 | /* call this *after* you increment lock->count */ | |
69 | static void __add_holder(struct block_lock *lock, struct task_struct *task) | |
70 | { | |
71 | unsigned h = __find_holder(lock, NULL); | |
72 | #ifdef CONFIG_DM_DEBUG_BLOCK_STACK_TRACING | |
73 | struct stack_trace *t; | |
74 | #endif | |
75 | ||
76 | get_task_struct(task); | |
77 | lock->holders[h] = task; | |
78 | ||
79 | #ifdef CONFIG_DM_DEBUG_BLOCK_STACK_TRACING | |
80 | t = lock->traces + h; | |
81 | t->nr_entries = 0; | |
82 | t->max_entries = MAX_STACK; | |
83 | t->entries = lock->entries[h]; | |
84 | t->skip = 2; | |
85 | save_stack_trace(t); | |
86 | #endif | |
87 | } | |
88 | ||
89 | /* call this *before* you decrement lock->count */ | |
90 | static void __del_holder(struct block_lock *lock, struct task_struct *task) | |
91 | { | |
92 | unsigned h = __find_holder(lock, task); | |
93 | lock->holders[h] = NULL; | |
94 | put_task_struct(task); | |
95 | } | |
96 | ||
97 | static int __check_holder(struct block_lock *lock) | |
98 | { | |
99 | unsigned i; | |
100 | #ifdef CONFIG_DM_DEBUG_BLOCK_STACK_TRACING | |
101 | static struct stack_trace t; | |
102 | static stack_entries entries; | |
103 | #endif | |
104 | ||
105 | for (i = 0; i < MAX_HOLDERS; i++) { | |
106 | if (lock->holders[i] == current) { | |
107 | DMERR("recursive lock detected in pool metadata"); | |
108 | #ifdef CONFIG_DM_DEBUG_BLOCK_STACK_TRACING | |
109 | DMERR("previously held here:"); | |
110 | print_stack_trace(lock->traces + i, 4); | |
111 | ||
112 | DMERR("subsequent aquisition attempted here:"); | |
113 | t.nr_entries = 0; | |
114 | t.max_entries = MAX_STACK; | |
115 | t.entries = entries; | |
116 | t.skip = 3; | |
117 | save_stack_trace(&t); | |
118 | print_stack_trace(&t, 4); | |
119 | #endif | |
120 | return -EINVAL; | |
121 | } | |
122 | } | |
123 | ||
124 | return 0; | |
125 | } | |
126 | ||
127 | static void __wait(struct waiter *w) | |
128 | { | |
129 | for (;;) { | |
130 | set_task_state(current, TASK_UNINTERRUPTIBLE); | |
131 | ||
132 | if (!w->task) | |
133 | break; | |
134 | ||
135 | schedule(); | |
136 | } | |
137 | ||
138 | set_task_state(current, TASK_RUNNING); | |
139 | } | |
140 | ||
141 | static void __wake_waiter(struct waiter *w) | |
142 | { | |
143 | struct task_struct *task; | |
144 | ||
145 | list_del(&w->list); | |
146 | task = w->task; | |
147 | smp_mb(); | |
148 | w->task = NULL; | |
149 | wake_up_process(task); | |
150 | } | |
151 | ||
152 | /* | |
153 | * We either wake a few readers or a single writer. | |
154 | */ | |
155 | static void __wake_many(struct block_lock *lock) | |
156 | { | |
157 | struct waiter *w, *tmp; | |
158 | ||
159 | BUG_ON(lock->count < 0); | |
160 | list_for_each_entry_safe(w, tmp, &lock->waiters, list) { | |
161 | if (lock->count >= MAX_HOLDERS) | |
162 | return; | |
163 | ||
164 | if (w->wants_write) { | |
165 | if (lock->count > 0) | |
166 | return; /* still read locked */ | |
167 | ||
168 | lock->count = -1; | |
169 | __add_holder(lock, w->task); | |
170 | __wake_waiter(w); | |
171 | return; | |
172 | } | |
173 | ||
174 | lock->count++; | |
175 | __add_holder(lock, w->task); | |
176 | __wake_waiter(w); | |
177 | } | |
178 | } | |
179 | ||
180 | static void bl_init(struct block_lock *lock) | |
181 | { | |
182 | int i; | |
183 | ||
184 | spin_lock_init(&lock->lock); | |
185 | lock->count = 0; | |
186 | INIT_LIST_HEAD(&lock->waiters); | |
187 | for (i = 0; i < MAX_HOLDERS; i++) | |
188 | lock->holders[i] = NULL; | |
189 | } | |
190 | ||
191 | static int __available_for_read(struct block_lock *lock) | |
192 | { | |
193 | return lock->count >= 0 && | |
194 | lock->count < MAX_HOLDERS && | |
195 | list_empty(&lock->waiters); | |
196 | } | |
197 | ||
198 | static int bl_down_read(struct block_lock *lock) | |
199 | { | |
200 | int r; | |
201 | struct waiter w; | |
202 | ||
203 | spin_lock(&lock->lock); | |
204 | r = __check_holder(lock); | |
205 | if (r) { | |
206 | spin_unlock(&lock->lock); | |
207 | return r; | |
208 | } | |
209 | ||
210 | if (__available_for_read(lock)) { | |
211 | lock->count++; | |
212 | __add_holder(lock, current); | |
213 | spin_unlock(&lock->lock); | |
214 | return 0; | |
215 | } | |
216 | ||
217 | get_task_struct(current); | |
218 | ||
219 | w.task = current; | |
220 | w.wants_write = 0; | |
221 | list_add_tail(&w.list, &lock->waiters); | |
222 | spin_unlock(&lock->lock); | |
223 | ||
224 | __wait(&w); | |
225 | put_task_struct(current); | |
226 | return 0; | |
227 | } | |
228 | ||
229 | static int bl_down_read_nonblock(struct block_lock *lock) | |
230 | { | |
231 | int r; | |
232 | ||
233 | spin_lock(&lock->lock); | |
234 | r = __check_holder(lock); | |
235 | if (r) | |
236 | goto out; | |
237 | ||
238 | if (__available_for_read(lock)) { | |
239 | lock->count++; | |
240 | __add_holder(lock, current); | |
241 | r = 0; | |
242 | } else | |
243 | r = -EWOULDBLOCK; | |
244 | ||
245 | out: | |
246 | spin_unlock(&lock->lock); | |
247 | return r; | |
248 | } | |
249 | ||
250 | static void bl_up_read(struct block_lock *lock) | |
251 | { | |
252 | spin_lock(&lock->lock); | |
253 | BUG_ON(lock->count <= 0); | |
254 | __del_holder(lock, current); | |
255 | --lock->count; | |
256 | if (!list_empty(&lock->waiters)) | |
257 | __wake_many(lock); | |
258 | spin_unlock(&lock->lock); | |
259 | } | |
260 | ||
261 | static int bl_down_write(struct block_lock *lock) | |
262 | { | |
263 | int r; | |
264 | struct waiter w; | |
265 | ||
266 | spin_lock(&lock->lock); | |
267 | r = __check_holder(lock); | |
268 | if (r) { | |
269 | spin_unlock(&lock->lock); | |
270 | return r; | |
271 | } | |
272 | ||
273 | if (lock->count == 0 && list_empty(&lock->waiters)) { | |
274 | lock->count = -1; | |
275 | __add_holder(lock, current); | |
276 | spin_unlock(&lock->lock); | |
277 | return 0; | |
278 | } | |
279 | ||
280 | get_task_struct(current); | |
281 | w.task = current; | |
282 | w.wants_write = 1; | |
283 | ||
284 | /* | |
285 | * Writers given priority. We know there's only one mutator in the | |
286 | * system, so ignoring the ordering reversal. | |
287 | */ | |
288 | list_add(&w.list, &lock->waiters); | |
289 | spin_unlock(&lock->lock); | |
290 | ||
291 | __wait(&w); | |
292 | put_task_struct(current); | |
293 | ||
294 | return 0; | |
295 | } | |
296 | ||
297 | static void bl_up_write(struct block_lock *lock) | |
298 | { | |
299 | spin_lock(&lock->lock); | |
300 | __del_holder(lock, current); | |
301 | lock->count = 0; | |
302 | if (!list_empty(&lock->waiters)) | |
303 | __wake_many(lock); | |
304 | spin_unlock(&lock->lock); | |
305 | } | |
306 | ||
307 | static void report_recursive_bug(dm_block_t b, int r) | |
308 | { | |
309 | if (r == -EINVAL) | |
310 | DMERR("recursive acquisition of block %llu requested.", | |
311 | (unsigned long long) b); | |
312 | } | |
313 | ||
314 | /*----------------------------------------------------------------*/ | |
315 | ||
316 | /* | |
317 | * Block manager is currently implemented using dm-bufio. struct | |
318 | * dm_block_manager and struct dm_block map directly onto a couple of | |
319 | * structs in the bufio interface. I want to retain the freedom to move | |
320 | * away from bufio in the future. So these structs are just cast within | |
321 | * this .c file, rather than making it through to the public interface. | |
322 | */ | |
323 | static struct dm_buffer *to_buffer(struct dm_block *b) | |
324 | { | |
325 | return (struct dm_buffer *) b; | |
326 | } | |
327 | ||
3241b1d3 JT |
328 | dm_block_t dm_block_location(struct dm_block *b) |
329 | { | |
330 | return dm_bufio_get_block_number(to_buffer(b)); | |
331 | } | |
332 | EXPORT_SYMBOL_GPL(dm_block_location); | |
333 | ||
334 | void *dm_block_data(struct dm_block *b) | |
335 | { | |
336 | return dm_bufio_get_block_data(to_buffer(b)); | |
337 | } | |
338 | EXPORT_SYMBOL_GPL(dm_block_data); | |
339 | ||
340 | struct buffer_aux { | |
341 | struct dm_block_validator *validator; | |
342 | struct block_lock lock; | |
343 | int write_locked; | |
344 | }; | |
345 | ||
346 | static void dm_block_manager_alloc_callback(struct dm_buffer *buf) | |
347 | { | |
348 | struct buffer_aux *aux = dm_bufio_get_aux_data(buf); | |
349 | aux->validator = NULL; | |
350 | bl_init(&aux->lock); | |
351 | } | |
352 | ||
353 | static void dm_block_manager_write_callback(struct dm_buffer *buf) | |
354 | { | |
355 | struct buffer_aux *aux = dm_bufio_get_aux_data(buf); | |
356 | if (aux->validator) { | |
357 | aux->validator->prepare_for_write(aux->validator, (struct dm_block *) buf, | |
358 | dm_bufio_get_block_size(dm_bufio_get_client(buf))); | |
359 | } | |
360 | } | |
361 | ||
362 | /*---------------------------------------------------------------- | |
363 | * Public interface | |
364 | *--------------------------------------------------------------*/ | |
51a0f659 JT |
365 | struct dm_block_manager { |
366 | struct dm_bufio_client *bufio; | |
367 | }; | |
368 | ||
3241b1d3 JT |
369 | struct dm_block_manager *dm_block_manager_create(struct block_device *bdev, |
370 | unsigned block_size, | |
371 | unsigned cache_size, | |
372 | unsigned max_held_per_thread) | |
373 | { | |
51a0f659 JT |
374 | int r; |
375 | struct dm_block_manager *bm; | |
376 | ||
377 | bm = kmalloc(sizeof(*bm), GFP_KERNEL); | |
378 | if (!bm) { | |
379 | r = -ENOMEM; | |
380 | goto bad; | |
381 | } | |
382 | ||
383 | bm->bufio = dm_bufio_client_create(bdev, block_size, max_held_per_thread, | |
384 | sizeof(struct buffer_aux), | |
385 | dm_block_manager_alloc_callback, | |
386 | dm_block_manager_write_callback); | |
387 | if (IS_ERR(bm->bufio)) { | |
388 | r = PTR_ERR(bm->bufio); | |
389 | kfree(bm); | |
390 | goto bad; | |
391 | } | |
392 | ||
393 | return bm; | |
394 | ||
395 | bad: | |
396 | return ERR_PTR(r); | |
3241b1d3 JT |
397 | } |
398 | EXPORT_SYMBOL_GPL(dm_block_manager_create); | |
399 | ||
400 | void dm_block_manager_destroy(struct dm_block_manager *bm) | |
401 | { | |
51a0f659 JT |
402 | dm_bufio_client_destroy(bm->bufio); |
403 | kfree(bm); | |
3241b1d3 JT |
404 | } |
405 | EXPORT_SYMBOL_GPL(dm_block_manager_destroy); | |
406 | ||
407 | unsigned dm_bm_block_size(struct dm_block_manager *bm) | |
408 | { | |
51a0f659 | 409 | return dm_bufio_get_block_size(bm->bufio); |
3241b1d3 JT |
410 | } |
411 | EXPORT_SYMBOL_GPL(dm_bm_block_size); | |
412 | ||
413 | dm_block_t dm_bm_nr_blocks(struct dm_block_manager *bm) | |
414 | { | |
51a0f659 | 415 | return dm_bufio_get_device_size(bm->bufio); |
3241b1d3 JT |
416 | } |
417 | ||
418 | static int dm_bm_validate_buffer(struct dm_block_manager *bm, | |
419 | struct dm_buffer *buf, | |
420 | struct buffer_aux *aux, | |
421 | struct dm_block_validator *v) | |
422 | { | |
423 | if (unlikely(!aux->validator)) { | |
424 | int r; | |
425 | if (!v) | |
426 | return 0; | |
51a0f659 | 427 | r = v->check(v, (struct dm_block *) buf, dm_bufio_get_block_size(bm->bufio)); |
3241b1d3 JT |
428 | if (unlikely(r)) |
429 | return r; | |
430 | aux->validator = v; | |
431 | } else { | |
432 | if (unlikely(aux->validator != v)) { | |
433 | DMERR("validator mismatch (old=%s vs new=%s) for block %llu", | |
434 | aux->validator->name, v ? v->name : "NULL", | |
435 | (unsigned long long) | |
436 | dm_bufio_get_block_number(buf)); | |
437 | return -EINVAL; | |
438 | } | |
439 | } | |
440 | ||
441 | return 0; | |
442 | } | |
443 | int dm_bm_read_lock(struct dm_block_manager *bm, dm_block_t b, | |
444 | struct dm_block_validator *v, | |
445 | struct dm_block **result) | |
446 | { | |
447 | struct buffer_aux *aux; | |
448 | void *p; | |
449 | int r; | |
450 | ||
51a0f659 | 451 | p = dm_bufio_read(bm->bufio, b, (struct dm_buffer **) result); |
3241b1d3 JT |
452 | if (unlikely(IS_ERR(p))) |
453 | return PTR_ERR(p); | |
454 | ||
455 | aux = dm_bufio_get_aux_data(to_buffer(*result)); | |
456 | r = bl_down_read(&aux->lock); | |
457 | if (unlikely(r)) { | |
458 | dm_bufio_release(to_buffer(*result)); | |
459 | report_recursive_bug(b, r); | |
460 | return r; | |
461 | } | |
462 | ||
463 | aux->write_locked = 0; | |
464 | ||
465 | r = dm_bm_validate_buffer(bm, to_buffer(*result), aux, v); | |
466 | if (unlikely(r)) { | |
467 | bl_up_read(&aux->lock); | |
468 | dm_bufio_release(to_buffer(*result)); | |
469 | return r; | |
470 | } | |
471 | ||
472 | return 0; | |
473 | } | |
474 | EXPORT_SYMBOL_GPL(dm_bm_read_lock); | |
475 | ||
476 | int dm_bm_write_lock(struct dm_block_manager *bm, | |
477 | dm_block_t b, struct dm_block_validator *v, | |
478 | struct dm_block **result) | |
479 | { | |
480 | struct buffer_aux *aux; | |
481 | void *p; | |
482 | int r; | |
483 | ||
51a0f659 | 484 | p = dm_bufio_read(bm->bufio, b, (struct dm_buffer **) result); |
3241b1d3 JT |
485 | if (unlikely(IS_ERR(p))) |
486 | return PTR_ERR(p); | |
487 | ||
488 | aux = dm_bufio_get_aux_data(to_buffer(*result)); | |
489 | r = bl_down_write(&aux->lock); | |
490 | if (r) { | |
491 | dm_bufio_release(to_buffer(*result)); | |
492 | report_recursive_bug(b, r); | |
493 | return r; | |
494 | } | |
495 | ||
496 | aux->write_locked = 1; | |
497 | ||
498 | r = dm_bm_validate_buffer(bm, to_buffer(*result), aux, v); | |
499 | if (unlikely(r)) { | |
500 | bl_up_write(&aux->lock); | |
501 | dm_bufio_release(to_buffer(*result)); | |
502 | return r; | |
503 | } | |
504 | ||
505 | return 0; | |
506 | } | |
507 | EXPORT_SYMBOL_GPL(dm_bm_write_lock); | |
508 | ||
509 | int dm_bm_read_try_lock(struct dm_block_manager *bm, | |
510 | dm_block_t b, struct dm_block_validator *v, | |
511 | struct dm_block **result) | |
512 | { | |
513 | struct buffer_aux *aux; | |
514 | void *p; | |
515 | int r; | |
516 | ||
51a0f659 | 517 | p = dm_bufio_get(bm->bufio, b, (struct dm_buffer **) result); |
3241b1d3 JT |
518 | if (unlikely(IS_ERR(p))) |
519 | return PTR_ERR(p); | |
520 | if (unlikely(!p)) | |
521 | return -EWOULDBLOCK; | |
522 | ||
523 | aux = dm_bufio_get_aux_data(to_buffer(*result)); | |
524 | r = bl_down_read_nonblock(&aux->lock); | |
525 | if (r < 0) { | |
526 | dm_bufio_release(to_buffer(*result)); | |
527 | report_recursive_bug(b, r); | |
528 | return r; | |
529 | } | |
530 | aux->write_locked = 0; | |
531 | ||
532 | r = dm_bm_validate_buffer(bm, to_buffer(*result), aux, v); | |
533 | if (unlikely(r)) { | |
534 | bl_up_read(&aux->lock); | |
535 | dm_bufio_release(to_buffer(*result)); | |
536 | return r; | |
537 | } | |
538 | ||
539 | return 0; | |
540 | } | |
541 | ||
542 | int dm_bm_write_lock_zero(struct dm_block_manager *bm, | |
543 | dm_block_t b, struct dm_block_validator *v, | |
544 | struct dm_block **result) | |
545 | { | |
546 | int r; | |
547 | struct buffer_aux *aux; | |
548 | void *p; | |
549 | ||
51a0f659 | 550 | p = dm_bufio_new(bm->bufio, b, (struct dm_buffer **) result); |
3241b1d3 JT |
551 | if (unlikely(IS_ERR(p))) |
552 | return PTR_ERR(p); | |
553 | ||
554 | memset(p, 0, dm_bm_block_size(bm)); | |
555 | ||
556 | aux = dm_bufio_get_aux_data(to_buffer(*result)); | |
557 | r = bl_down_write(&aux->lock); | |
558 | if (r) { | |
559 | dm_bufio_release(to_buffer(*result)); | |
560 | return r; | |
561 | } | |
562 | ||
563 | aux->write_locked = 1; | |
564 | aux->validator = v; | |
565 | ||
566 | return 0; | |
567 | } | |
568 | ||
569 | int dm_bm_unlock(struct dm_block *b) | |
570 | { | |
571 | struct buffer_aux *aux; | |
572 | aux = dm_bufio_get_aux_data(to_buffer(b)); | |
573 | ||
574 | if (aux->write_locked) { | |
575 | dm_bufio_mark_buffer_dirty(to_buffer(b)); | |
576 | bl_up_write(&aux->lock); | |
577 | } else | |
578 | bl_up_read(&aux->lock); | |
579 | ||
580 | dm_bufio_release(to_buffer(b)); | |
581 | ||
582 | return 0; | |
583 | } | |
584 | EXPORT_SYMBOL_GPL(dm_bm_unlock); | |
585 | ||
586 | int dm_bm_unlock_move(struct dm_block *b, dm_block_t n) | |
587 | { | |
588 | struct buffer_aux *aux; | |
589 | ||
590 | aux = dm_bufio_get_aux_data(to_buffer(b)); | |
591 | ||
592 | if (aux->write_locked) { | |
593 | dm_bufio_mark_buffer_dirty(to_buffer(b)); | |
594 | bl_up_write(&aux->lock); | |
595 | } else | |
596 | bl_up_read(&aux->lock); | |
597 | ||
598 | dm_bufio_release_move(to_buffer(b), n); | |
599 | return 0; | |
600 | } | |
601 | ||
602 | int dm_bm_flush_and_unlock(struct dm_block_manager *bm, | |
603 | struct dm_block *superblock) | |
604 | { | |
605 | int r; | |
606 | ||
51a0f659 | 607 | r = dm_bufio_write_dirty_buffers(bm->bufio); |
8d44c98a JT |
608 | if (unlikely(r)) { |
609 | dm_bm_unlock(superblock); | |
3241b1d3 | 610 | return r; |
8d44c98a | 611 | } |
3241b1d3 JT |
612 | |
613 | dm_bm_unlock(superblock); | |
614 | ||
51a0f659 | 615 | return dm_bufio_write_dirty_buffers(bm->bufio); |
3241b1d3 JT |
616 | } |
617 | ||
618 | u32 dm_bm_checksum(const void *data, size_t len, u32 init_xor) | |
619 | { | |
620 | return crc32c(~(u32) 0, data, len) ^ init_xor; | |
621 | } | |
622 | EXPORT_SYMBOL_GPL(dm_bm_checksum); | |
623 | ||
624 | /*----------------------------------------------------------------*/ | |
625 | ||
626 | MODULE_LICENSE("GPL"); | |
627 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); | |
628 | MODULE_DESCRIPTION("Immutable metadata library for dm"); | |
629 | ||
630 | /*----------------------------------------------------------------*/ |