Commit | Line | Data |
---|---|---|
cafe5635 KO |
1 | /* |
2 | * bcache journalling code, for btree insertions | |
3 | * | |
4 | * Copyright 2012 Google, Inc. | |
5 | */ | |
6 | ||
7 | #include "bcache.h" | |
8 | #include "btree.h" | |
9 | #include "debug.h" | |
9aa61a99 | 10 | #include "extents.h" |
cafe5635 | 11 | |
c37511b8 KO |
12 | #include <trace/events/bcache.h> |
13 | ||
cafe5635 KO |
14 | /* |
15 | * Journal replay/recovery: | |
16 | * | |
17 | * This code is all driven from run_cache_set(); we first read the journal | |
18 | * entries, do some other stuff, then we mark all the keys in the journal | |
19 | * entries (same as garbage collection would), then we replay them - reinserting | |
20 | * them into the cache in precisely the same order as they appear in the | |
21 | * journal. | |
22 | * | |
23 | * We only journal keys that go in leaf nodes, which simplifies things quite a | |
24 | * bit. | |
25 | */ | |
26 | ||
27 | static void journal_read_endio(struct bio *bio, int error) | |
28 | { | |
29 | struct closure *cl = bio->bi_private; | |
30 | closure_put(cl); | |
31 | } | |
32 | ||
33 | static int journal_read_bucket(struct cache *ca, struct list_head *list, | |
c18536a7 | 34 | unsigned bucket_index) |
cafe5635 KO |
35 | { |
36 | struct journal_device *ja = &ca->journal; | |
37 | struct bio *bio = &ja->bio; | |
38 | ||
39 | struct journal_replay *i; | |
40 | struct jset *j, *data = ca->set->journal.w[0].data; | |
c18536a7 | 41 | struct closure cl; |
cafe5635 KO |
42 | unsigned len, left, offset = 0; |
43 | int ret = 0; | |
44 | sector_t bucket = bucket_to_sector(ca->set, ca->sb.d[bucket_index]); | |
45 | ||
c18536a7 KO |
46 | closure_init_stack(&cl); |
47 | ||
b3fa7e77 | 48 | pr_debug("reading %u", bucket_index); |
cafe5635 KO |
49 | |
50 | while (offset < ca->sb.bucket_size) { | |
51 | reread: left = ca->sb.bucket_size - offset; | |
b3fa7e77 | 52 | len = min_t(unsigned, left, PAGE_SECTORS << JSET_BITS); |
cafe5635 KO |
53 | |
54 | bio_reset(bio); | |
4f024f37 | 55 | bio->bi_iter.bi_sector = bucket + offset; |
cafe5635 KO |
56 | bio->bi_bdev = ca->bdev; |
57 | bio->bi_rw = READ; | |
4f024f37 | 58 | bio->bi_iter.bi_size = len << 9; |
cafe5635 KO |
59 | |
60 | bio->bi_end_io = journal_read_endio; | |
c18536a7 | 61 | bio->bi_private = &cl; |
169ef1cf | 62 | bch_bio_map(bio, data); |
cafe5635 | 63 | |
c18536a7 KO |
64 | closure_bio_submit(bio, &cl, ca); |
65 | closure_sync(&cl); | |
cafe5635 KO |
66 | |
67 | /* This function could be simpler now since we no longer write | |
68 | * journal entries that overlap bucket boundaries; this means | |
69 | * the start of a bucket will always have a valid journal entry | |
70 | * if it has any journal entries at all. | |
71 | */ | |
72 | ||
73 | j = data; | |
74 | while (len) { | |
75 | struct list_head *where; | |
76 | size_t blocks, bytes = set_bytes(j); | |
77 | ||
b3fa7e77 KO |
78 | if (j->magic != jset_magic(&ca->sb)) { |
79 | pr_debug("%u: bad magic", bucket_index); | |
cafe5635 | 80 | return ret; |
b3fa7e77 | 81 | } |
cafe5635 | 82 | |
b3fa7e77 KO |
83 | if (bytes > left << 9 || |
84 | bytes > PAGE_SIZE << JSET_BITS) { | |
85 | pr_info("%u: too big, %zu bytes, offset %u", | |
86 | bucket_index, bytes, offset); | |
cafe5635 | 87 | return ret; |
b3fa7e77 | 88 | } |
cafe5635 KO |
89 | |
90 | if (bytes > len << 9) | |
91 | goto reread; | |
92 | ||
b3fa7e77 KO |
93 | if (j->csum != csum_set(j)) { |
94 | pr_info("%u: bad csum, %zu bytes, offset %u", | |
95 | bucket_index, bytes, offset); | |
cafe5635 | 96 | return ret; |
b3fa7e77 | 97 | } |
cafe5635 | 98 | |
ee811287 | 99 | blocks = set_blocks(j, block_bytes(ca->set)); |
cafe5635 KO |
100 | |
101 | while (!list_empty(list)) { | |
102 | i = list_first_entry(list, | |
103 | struct journal_replay, list); | |
104 | if (i->j.seq >= j->last_seq) | |
105 | break; | |
106 | list_del(&i->list); | |
107 | kfree(i); | |
108 | } | |
109 | ||
110 | list_for_each_entry_reverse(i, list, list) { | |
111 | if (j->seq == i->j.seq) | |
112 | goto next_set; | |
113 | ||
114 | if (j->seq < i->j.last_seq) | |
115 | goto next_set; | |
116 | ||
117 | if (j->seq > i->j.seq) { | |
118 | where = &i->list; | |
119 | goto add; | |
120 | } | |
121 | } | |
122 | ||
123 | where = list; | |
124 | add: | |
125 | i = kmalloc(offsetof(struct journal_replay, j) + | |
126 | bytes, GFP_KERNEL); | |
127 | if (!i) | |
128 | return -ENOMEM; | |
129 | memcpy(&i->j, j, bytes); | |
130 | list_add(&i->list, where); | |
131 | ret = 1; | |
132 | ||
133 | ja->seq[bucket_index] = j->seq; | |
134 | next_set: | |
135 | offset += blocks * ca->sb.block_size; | |
136 | len -= blocks * ca->sb.block_size; | |
137 | j = ((void *) j) + blocks * block_bytes(ca); | |
138 | } | |
139 | } | |
140 | ||
141 | return ret; | |
142 | } | |
143 | ||
c18536a7 | 144 | int bch_journal_read(struct cache_set *c, struct list_head *list) |
cafe5635 KO |
145 | { |
146 | #define read_bucket(b) \ | |
147 | ({ \ | |
c18536a7 | 148 | int ret = journal_read_bucket(ca, list, b); \ |
cafe5635 KO |
149 | __set_bit(b, bitmap); \ |
150 | if (ret < 0) \ | |
151 | return ret; \ | |
152 | ret; \ | |
153 | }) | |
154 | ||
155 | struct cache *ca; | |
156 | unsigned iter; | |
157 | ||
158 | for_each_cache(ca, c, iter) { | |
159 | struct journal_device *ja = &ca->journal; | |
160 | unsigned long bitmap[SB_JOURNAL_BUCKETS / BITS_PER_LONG]; | |
161 | unsigned i, l, r, m; | |
162 | uint64_t seq; | |
163 | ||
164 | bitmap_zero(bitmap, SB_JOURNAL_BUCKETS); | |
165 | pr_debug("%u journal buckets", ca->sb.njournal_buckets); | |
166 | ||
c426c4fd KO |
167 | /* |
168 | * Read journal buckets ordered by golden ratio hash to quickly | |
cafe5635 KO |
169 | * find a sequence of buckets with valid journal entries |
170 | */ | |
171 | for (i = 0; i < ca->sb.njournal_buckets; i++) { | |
172 | l = (i * 2654435769U) % ca->sb.njournal_buckets; | |
173 | ||
174 | if (test_bit(l, bitmap)) | |
175 | break; | |
176 | ||
177 | if (read_bucket(l)) | |
178 | goto bsearch; | |
179 | } | |
180 | ||
c426c4fd KO |
181 | /* |
182 | * If that fails, check all the buckets we haven't checked | |
cafe5635 KO |
183 | * already |
184 | */ | |
185 | pr_debug("falling back to linear search"); | |
186 | ||
c426c4fd KO |
187 | for (l = find_first_zero_bit(bitmap, ca->sb.njournal_buckets); |
188 | l < ca->sb.njournal_buckets; | |
189 | l = find_next_zero_bit(bitmap, ca->sb.njournal_buckets, l + 1)) | |
cafe5635 KO |
190 | if (read_bucket(l)) |
191 | goto bsearch; | |
c426c4fd KO |
192 | |
193 | if (list_empty(list)) | |
194 | continue; | |
cafe5635 KO |
195 | bsearch: |
196 | /* Binary search */ | |
dbd810ab SP |
197 | m = l; |
198 | r = find_next_bit(bitmap, ca->sb.njournal_buckets, l + 1); | |
cafe5635 KO |
199 | pr_debug("starting binary search, l %u r %u", l, r); |
200 | ||
201 | while (l + 1 < r) { | |
faa56736 KO |
202 | seq = list_entry(list->prev, struct journal_replay, |
203 | list)->j.seq; | |
204 | ||
cafe5635 | 205 | m = (l + r) >> 1; |
faa56736 | 206 | read_bucket(m); |
cafe5635 | 207 | |
faa56736 KO |
208 | if (seq != list_entry(list->prev, struct journal_replay, |
209 | list)->j.seq) | |
cafe5635 KO |
210 | l = m; |
211 | else | |
212 | r = m; | |
213 | } | |
214 | ||
c426c4fd KO |
215 | /* |
216 | * Read buckets in reverse order until we stop finding more | |
cafe5635 KO |
217 | * journal entries |
218 | */ | |
c426c4fd KO |
219 | pr_debug("finishing up: m %u njournal_buckets %u", |
220 | m, ca->sb.njournal_buckets); | |
cafe5635 KO |
221 | l = m; |
222 | ||
223 | while (1) { | |
224 | if (!l--) | |
225 | l = ca->sb.njournal_buckets - 1; | |
226 | ||
227 | if (l == m) | |
228 | break; | |
229 | ||
230 | if (test_bit(l, bitmap)) | |
231 | continue; | |
232 | ||
233 | if (!read_bucket(l)) | |
234 | break; | |
235 | } | |
236 | ||
237 | seq = 0; | |
238 | ||
239 | for (i = 0; i < ca->sb.njournal_buckets; i++) | |
240 | if (ja->seq[i] > seq) { | |
241 | seq = ja->seq[i]; | |
27201cfd KO |
242 | /* |
243 | * When journal_reclaim() goes to allocate for | |
244 | * the first time, it'll use the bucket after | |
245 | * ja->cur_idx | |
246 | */ | |
247 | ja->cur_idx = i; | |
248 | ja->last_idx = ja->discard_idx = (i + 1) % | |
249 | ca->sb.njournal_buckets; | |
cafe5635 KO |
250 | |
251 | } | |
252 | } | |
253 | ||
c426c4fd KO |
254 | if (!list_empty(list)) |
255 | c->journal.seq = list_entry(list->prev, | |
256 | struct journal_replay, | |
257 | list)->j.seq; | |
cafe5635 KO |
258 | |
259 | return 0; | |
260 | #undef read_bucket | |
261 | } | |
262 | ||
263 | void bch_journal_mark(struct cache_set *c, struct list_head *list) | |
264 | { | |
265 | atomic_t p = { 0 }; | |
266 | struct bkey *k; | |
267 | struct journal_replay *i; | |
268 | struct journal *j = &c->journal; | |
269 | uint64_t last = j->seq; | |
270 | ||
271 | /* | |
272 | * journal.pin should never fill up - we never write a journal | |
273 | * entry when it would fill up. But if for some reason it does, we | |
274 | * iterate over the list in reverse order so that we can just skip that | |
275 | * refcount instead of bugging. | |
276 | */ | |
277 | ||
278 | list_for_each_entry_reverse(i, list, list) { | |
279 | BUG_ON(last < i->j.seq); | |
280 | i->pin = NULL; | |
281 | ||
282 | while (last-- != i->j.seq) | |
283 | if (fifo_free(&j->pin) > 1) { | |
284 | fifo_push_front(&j->pin, p); | |
285 | atomic_set(&fifo_front(&j->pin), 0); | |
286 | } | |
287 | ||
288 | if (fifo_free(&j->pin) > 1) { | |
289 | fifo_push_front(&j->pin, p); | |
290 | i->pin = &fifo_front(&j->pin); | |
291 | atomic_set(i->pin, 1); | |
292 | } | |
293 | ||
294 | for (k = i->j.start; | |
fafff81c | 295 | k < bset_bkey_last(&i->j); |
9aa61a99 KO |
296 | k = bkey_next(k)) |
297 | if (!__bch_extent_invalid(c, k)) { | |
298 | unsigned j; | |
cafe5635 | 299 | |
9aa61a99 KO |
300 | for (j = 0; j < KEY_PTRS(k); j++) |
301 | if (ptr_available(c, k, j)) | |
302 | atomic_inc(&PTR_BUCKET(c, k, j)->pin); | |
65ddf45a | 303 | |
9aa61a99 KO |
304 | bch_initial_mark_key(c, 0, k); |
305 | } | |
cafe5635 KO |
306 | } |
307 | } | |
308 | ||
c18536a7 | 309 | int bch_journal_replay(struct cache_set *s, struct list_head *list) |
cafe5635 KO |
310 | { |
311 | int ret = 0, keys = 0, entries = 0; | |
312 | struct bkey *k; | |
313 | struct journal_replay *i = | |
314 | list_entry(list->prev, struct journal_replay, list); | |
315 | ||
316 | uint64_t start = i->j.last_seq, end = i->j.seq, n = start; | |
0b93207a KO |
317 | struct keylist keylist; |
318 | ||
cafe5635 KO |
319 | list_for_each_entry(i, list, list) { |
320 | BUG_ON(i->pin && atomic_read(i->pin) != 1); | |
321 | ||
77c320eb KO |
322 | cache_set_err_on(n != i->j.seq, s, |
323 | "bcache: journal entries %llu-%llu missing! (replaying %llu-%llu)", | |
324 | n, i->j.seq - 1, start, end); | |
cafe5635 KO |
325 | |
326 | for (k = i->j.start; | |
fafff81c | 327 | k < bset_bkey_last(&i->j); |
cafe5635 | 328 | k = bkey_next(k)) { |
c37511b8 KO |
329 | trace_bcache_journal_replay_key(k); |
330 | ||
c13f3af9 | 331 | bch_keylist_init_single(&keylist, k); |
cafe5635 | 332 | |
cc7b8819 | 333 | ret = bch_btree_insert(s, &keylist, i->pin, NULL); |
cafe5635 KO |
334 | if (ret) |
335 | goto err; | |
336 | ||
0b93207a | 337 | BUG_ON(!bch_keylist_empty(&keylist)); |
cafe5635 KO |
338 | keys++; |
339 | ||
340 | cond_resched(); | |
341 | } | |
342 | ||
343 | if (i->pin) | |
344 | atomic_dec(i->pin); | |
345 | n = i->j.seq + 1; | |
346 | entries++; | |
347 | } | |
348 | ||
349 | pr_info("journal replay done, %i keys in %i entries, seq %llu", | |
350 | keys, entries, end); | |
b54d6934 | 351 | err: |
cafe5635 KO |
352 | while (!list_empty(list)) { |
353 | i = list_first_entry(list, struct journal_replay, list); | |
354 | list_del(&i->list); | |
355 | kfree(i); | |
356 | } | |
b54d6934 | 357 | |
cafe5635 KO |
358 | return ret; |
359 | } | |
360 | ||
361 | /* Journalling */ | |
362 | ||
363 | static void btree_flush_write(struct cache_set *c) | |
364 | { | |
365 | /* | |
366 | * Try to find the btree node with that references the oldest journal | |
367 | * entry, best is our current candidate and is locked if non NULL: | |
368 | */ | |
a34a8bfd KO |
369 | struct btree *b, *best; |
370 | unsigned i; | |
371 | retry: | |
372 | best = NULL; | |
373 | ||
374 | for_each_cached_btree(b, c, i) | |
375 | if (btree_current_write(b)->journal) { | |
376 | if (!best) | |
377 | best = b; | |
378 | else if (journal_pin_cmp(c, | |
c18536a7 KO |
379 | btree_current_write(best)->journal, |
380 | btree_current_write(b)->journal)) { | |
a34a8bfd KO |
381 | best = b; |
382 | } | |
383 | } | |
cafe5635 | 384 | |
a34a8bfd KO |
385 | b = best; |
386 | if (b) { | |
2a285686 | 387 | mutex_lock(&b->write_lock); |
a34a8bfd | 388 | if (!btree_current_write(b)->journal) { |
2a285686 | 389 | mutex_unlock(&b->write_lock); |
a34a8bfd KO |
390 | /* We raced */ |
391 | goto retry; | |
cafe5635 KO |
392 | } |
393 | ||
2a285686 KO |
394 | __bch_btree_node_write(b, NULL); |
395 | mutex_unlock(&b->write_lock); | |
cafe5635 | 396 | } |
cafe5635 KO |
397 | } |
398 | ||
399 | #define last_seq(j) ((j)->seq - fifo_used(&(j)->pin) + 1) | |
400 | ||
401 | static void journal_discard_endio(struct bio *bio, int error) | |
402 | { | |
403 | struct journal_device *ja = | |
404 | container_of(bio, struct journal_device, discard_bio); | |
405 | struct cache *ca = container_of(ja, struct cache, journal); | |
406 | ||
407 | atomic_set(&ja->discard_in_flight, DISCARD_DONE); | |
408 | ||
409 | closure_wake_up(&ca->set->journal.wait); | |
410 | closure_put(&ca->set->cl); | |
411 | } | |
412 | ||
413 | static void journal_discard_work(struct work_struct *work) | |
414 | { | |
415 | struct journal_device *ja = | |
416 | container_of(work, struct journal_device, discard_work); | |
417 | ||
418 | submit_bio(0, &ja->discard_bio); | |
419 | } | |
420 | ||
421 | static void do_journal_discard(struct cache *ca) | |
422 | { | |
423 | struct journal_device *ja = &ca->journal; | |
424 | struct bio *bio = &ja->discard_bio; | |
425 | ||
426 | if (!ca->discard) { | |
427 | ja->discard_idx = ja->last_idx; | |
428 | return; | |
429 | } | |
430 | ||
6d9d21e3 | 431 | switch (atomic_read(&ja->discard_in_flight)) { |
cafe5635 KO |
432 | case DISCARD_IN_FLIGHT: |
433 | return; | |
434 | ||
435 | case DISCARD_DONE: | |
436 | ja->discard_idx = (ja->discard_idx + 1) % | |
437 | ca->sb.njournal_buckets; | |
438 | ||
439 | atomic_set(&ja->discard_in_flight, DISCARD_READY); | |
440 | /* fallthrough */ | |
441 | ||
442 | case DISCARD_READY: | |
443 | if (ja->discard_idx == ja->last_idx) | |
444 | return; | |
445 | ||
446 | atomic_set(&ja->discard_in_flight, DISCARD_IN_FLIGHT); | |
447 | ||
448 | bio_init(bio); | |
4f024f37 | 449 | bio->bi_iter.bi_sector = bucket_to_sector(ca->set, |
b1a67b0f | 450 | ca->sb.d[ja->discard_idx]); |
cafe5635 KO |
451 | bio->bi_bdev = ca->bdev; |
452 | bio->bi_rw = REQ_WRITE|REQ_DISCARD; | |
453 | bio->bi_max_vecs = 1; | |
454 | bio->bi_io_vec = bio->bi_inline_vecs; | |
4f024f37 | 455 | bio->bi_iter.bi_size = bucket_bytes(ca); |
cafe5635 KO |
456 | bio->bi_end_io = journal_discard_endio; |
457 | ||
458 | closure_get(&ca->set->cl); | |
459 | INIT_WORK(&ja->discard_work, journal_discard_work); | |
460 | schedule_work(&ja->discard_work); | |
461 | } | |
462 | } | |
463 | ||
464 | static void journal_reclaim(struct cache_set *c) | |
465 | { | |
466 | struct bkey *k = &c->journal.key; | |
467 | struct cache *ca; | |
468 | uint64_t last_seq; | |
469 | unsigned iter, n = 0; | |
470 | atomic_t p; | |
471 | ||
472 | while (!atomic_read(&fifo_front(&c->journal.pin))) | |
473 | fifo_pop(&c->journal.pin, p); | |
474 | ||
475 | last_seq = last_seq(&c->journal); | |
476 | ||
477 | /* Update last_idx */ | |
478 | ||
479 | for_each_cache(ca, c, iter) { | |
480 | struct journal_device *ja = &ca->journal; | |
481 | ||
482 | while (ja->last_idx != ja->cur_idx && | |
483 | ja->seq[ja->last_idx] < last_seq) | |
484 | ja->last_idx = (ja->last_idx + 1) % | |
485 | ca->sb.njournal_buckets; | |
486 | } | |
487 | ||
488 | for_each_cache(ca, c, iter) | |
489 | do_journal_discard(ca); | |
490 | ||
491 | if (c->journal.blocks_free) | |
a34a8bfd | 492 | goto out; |
cafe5635 KO |
493 | |
494 | /* | |
495 | * Allocate: | |
496 | * XXX: Sort by free journal space | |
497 | */ | |
498 | ||
499 | for_each_cache(ca, c, iter) { | |
500 | struct journal_device *ja = &ca->journal; | |
501 | unsigned next = (ja->cur_idx + 1) % ca->sb.njournal_buckets; | |
502 | ||
503 | /* No space available on this device */ | |
504 | if (next == ja->discard_idx) | |
505 | continue; | |
506 | ||
507 | ja->cur_idx = next; | |
508 | k->ptr[n++] = PTR(0, | |
509 | bucket_to_sector(c, ca->sb.d[ja->cur_idx]), | |
510 | ca->sb.nr_this_dev); | |
511 | } | |
512 | ||
513 | bkey_init(k); | |
514 | SET_KEY_PTRS(k, n); | |
515 | ||
516 | if (n) | |
517 | c->journal.blocks_free = c->sb.bucket_size >> c->block_bits; | |
a34a8bfd | 518 | out: |
cafe5635 KO |
519 | if (!journal_full(&c->journal)) |
520 | __closure_wake_up(&c->journal.wait); | |
521 | } | |
522 | ||
523 | void bch_journal_next(struct journal *j) | |
524 | { | |
525 | atomic_t p = { 1 }; | |
526 | ||
527 | j->cur = (j->cur == j->w) | |
528 | ? &j->w[1] | |
529 | : &j->w[0]; | |
530 | ||
531 | /* | |
532 | * The fifo_push() needs to happen at the same time as j->seq is | |
533 | * incremented for last_seq() to be calculated correctly | |
534 | */ | |
535 | BUG_ON(!fifo_push(&j->pin, p)); | |
536 | atomic_set(&fifo_back(&j->pin), 1); | |
537 | ||
538 | j->cur->data->seq = ++j->seq; | |
dabb4433 | 539 | j->cur->dirty = false; |
cafe5635 KO |
540 | j->cur->need_write = false; |
541 | j->cur->data->keys = 0; | |
542 | ||
543 | if (fifo_full(&j->pin)) | |
544 | pr_debug("journal_pin full (%zu)", fifo_used(&j->pin)); | |
545 | } | |
546 | ||
547 | static void journal_write_endio(struct bio *bio, int error) | |
548 | { | |
549 | struct journal_write *w = bio->bi_private; | |
550 | ||
551 | cache_set_err_on(error, w->c, "journal io error"); | |
7857d5d4 | 552 | closure_put(&w->c->journal.io); |
cafe5635 KO |
553 | } |
554 | ||
555 | static void journal_write(struct closure *); | |
556 | ||
557 | static void journal_write_done(struct closure *cl) | |
558 | { | |
7857d5d4 | 559 | struct journal *j = container_of(cl, struct journal, io); |
cafe5635 KO |
560 | struct journal_write *w = (j->cur == j->w) |
561 | ? &j->w[1] | |
562 | : &j->w[0]; | |
563 | ||
564 | __closure_wake_up(&w->wait); | |
7857d5d4 | 565 | continue_at_nobarrier(cl, journal_write, system_wq); |
cafe5635 KO |
566 | } |
567 | ||
cb7a583e KO |
568 | static void journal_write_unlock(struct closure *cl) |
569 | { | |
570 | struct cache_set *c = container_of(cl, struct cache_set, journal.io); | |
571 | ||
572 | c->journal.io_in_flight = 0; | |
573 | spin_unlock(&c->journal.lock); | |
574 | } | |
575 | ||
cafe5635 | 576 | static void journal_write_unlocked(struct closure *cl) |
c19ed23a | 577 | __releases(c->journal.lock) |
cafe5635 | 578 | { |
7857d5d4 | 579 | struct cache_set *c = container_of(cl, struct cache_set, journal.io); |
cafe5635 KO |
580 | struct cache *ca; |
581 | struct journal_write *w = c->journal.cur; | |
582 | struct bkey *k = &c->journal.key; | |
ee811287 KO |
583 | unsigned i, sectors = set_blocks(w->data, block_bytes(c)) * |
584 | c->sb.block_size; | |
cafe5635 KO |
585 | |
586 | struct bio *bio; | |
587 | struct bio_list list; | |
588 | bio_list_init(&list); | |
589 | ||
590 | if (!w->need_write) { | |
cb7a583e | 591 | closure_return_with_destructor(cl, journal_write_unlock); |
cafe5635 KO |
592 | } else if (journal_full(&c->journal)) { |
593 | journal_reclaim(c); | |
594 | spin_unlock(&c->journal.lock); | |
595 | ||
596 | btree_flush_write(c); | |
597 | continue_at(cl, journal_write, system_wq); | |
598 | } | |
599 | ||
ee811287 | 600 | c->journal.blocks_free -= set_blocks(w->data, block_bytes(c)); |
cafe5635 KO |
601 | |
602 | w->data->btree_level = c->root->level; | |
603 | ||
604 | bkey_copy(&w->data->btree_root, &c->root->key); | |
605 | bkey_copy(&w->data->uuid_bucket, &c->uuid_bucket); | |
606 | ||
607 | for_each_cache(ca, c, i) | |
608 | w->data->prio_bucket[ca->sb.nr_this_dev] = ca->prio_buckets[0]; | |
609 | ||
81ab4190 | 610 | w->data->magic = jset_magic(&c->sb); |
cafe5635 KO |
611 | w->data->version = BCACHE_JSET_VERSION; |
612 | w->data->last_seq = last_seq(&c->journal); | |
613 | w->data->csum = csum_set(w->data); | |
614 | ||
615 | for (i = 0; i < KEY_PTRS(k); i++) { | |
616 | ca = PTR_CACHE(c, k, i); | |
617 | bio = &ca->journal.bio; | |
618 | ||
619 | atomic_long_add(sectors, &ca->meta_sectors_written); | |
620 | ||
621 | bio_reset(bio); | |
4f024f37 | 622 | bio->bi_iter.bi_sector = PTR_OFFSET(k, i); |
cafe5635 | 623 | bio->bi_bdev = ca->bdev; |
e49c7c37 | 624 | bio->bi_rw = REQ_WRITE|REQ_SYNC|REQ_META|REQ_FLUSH|REQ_FUA; |
4f024f37 | 625 | bio->bi_iter.bi_size = sectors << 9; |
cafe5635 KO |
626 | |
627 | bio->bi_end_io = journal_write_endio; | |
628 | bio->bi_private = w; | |
169ef1cf | 629 | bch_bio_map(bio, w->data); |
cafe5635 KO |
630 | |
631 | trace_bcache_journal_write(bio); | |
632 | bio_list_add(&list, bio); | |
633 | ||
634 | SET_PTR_OFFSET(k, i, PTR_OFFSET(k, i) + sectors); | |
635 | ||
636 | ca->journal.seq[ca->journal.cur_idx] = w->data->seq; | |
637 | } | |
638 | ||
639 | atomic_dec_bug(&fifo_back(&c->journal.pin)); | |
640 | bch_journal_next(&c->journal); | |
641 | journal_reclaim(c); | |
642 | ||
643 | spin_unlock(&c->journal.lock); | |
644 | ||
645 | while ((bio = bio_list_pop(&list))) | |
646 | closure_bio_submit(bio, cl, c->cache[0]); | |
647 | ||
648 | continue_at(cl, journal_write_done, NULL); | |
649 | } | |
650 | ||
651 | static void journal_write(struct closure *cl) | |
652 | { | |
7857d5d4 | 653 | struct cache_set *c = container_of(cl, struct cache_set, journal.io); |
cafe5635 KO |
654 | |
655 | spin_lock(&c->journal.lock); | |
656 | journal_write_unlocked(cl); | |
657 | } | |
658 | ||
a34a8bfd | 659 | static void journal_try_write(struct cache_set *c) |
c19ed23a | 660 | __releases(c->journal.lock) |
cafe5635 | 661 | { |
7857d5d4 KO |
662 | struct closure *cl = &c->journal.io; |
663 | struct journal_write *w = c->journal.cur; | |
664 | ||
665 | w->need_write = true; | |
cafe5635 | 666 | |
cb7a583e KO |
667 | if (!c->journal.io_in_flight) { |
668 | c->journal.io_in_flight = 1; | |
669 | closure_call(cl, journal_write_unlocked, NULL, &c->cl); | |
670 | } else { | |
a34a8bfd | 671 | spin_unlock(&c->journal.lock); |
cb7a583e | 672 | } |
cafe5635 KO |
673 | } |
674 | ||
a34a8bfd KO |
675 | static struct journal_write *journal_wait_for_write(struct cache_set *c, |
676 | unsigned nkeys) | |
cafe5635 | 677 | { |
a34a8bfd KO |
678 | size_t sectors; |
679 | struct closure cl; | |
5775e213 | 680 | bool wait = false; |
cafe5635 | 681 | |
a34a8bfd KO |
682 | closure_init_stack(&cl); |
683 | ||
684 | spin_lock(&c->journal.lock); | |
685 | ||
686 | while (1) { | |
687 | struct journal_write *w = c->journal.cur; | |
688 | ||
689 | sectors = __set_blocks(w->data, w->data->keys + nkeys, | |
ee811287 | 690 | block_bytes(c)) * c->sb.block_size; |
a34a8bfd KO |
691 | |
692 | if (sectors <= min_t(size_t, | |
693 | c->journal.blocks_free * c->sb.block_size, | |
694 | PAGE_SECTORS << JSET_BITS)) | |
695 | return w; | |
696 | ||
5775e213 KO |
697 | if (wait) |
698 | closure_wait(&c->journal.wait, &cl); | |
699 | ||
a34a8bfd | 700 | if (!journal_full(&c->journal)) { |
5775e213 KO |
701 | if (wait) |
702 | trace_bcache_journal_entry_full(c); | |
a34a8bfd KO |
703 | |
704 | /* | |
705 | * XXX: If we were inserting so many keys that they | |
706 | * won't fit in an _empty_ journal write, we'll | |
707 | * deadlock. For now, handle this in | |
708 | * bch_keylist_realloc() - but something to think about. | |
709 | */ | |
710 | BUG_ON(!w->data->keys); | |
711 | ||
a34a8bfd KO |
712 | journal_try_write(c); /* unlocks */ |
713 | } else { | |
5775e213 KO |
714 | if (wait) |
715 | trace_bcache_journal_full(c); | |
a34a8bfd | 716 | |
a34a8bfd KO |
717 | journal_reclaim(c); |
718 | spin_unlock(&c->journal.lock); | |
cafe5635 | 719 | |
a34a8bfd KO |
720 | btree_flush_write(c); |
721 | } | |
cafe5635 | 722 | |
a34a8bfd KO |
723 | closure_sync(&cl); |
724 | spin_lock(&c->journal.lock); | |
5775e213 | 725 | wait = true; |
cafe5635 KO |
726 | } |
727 | } | |
728 | ||
7857d5d4 KO |
729 | static void journal_write_work(struct work_struct *work) |
730 | { | |
731 | struct cache_set *c = container_of(to_delayed_work(work), | |
732 | struct cache_set, | |
733 | journal.work); | |
734 | spin_lock(&c->journal.lock); | |
dabb4433 KO |
735 | if (c->journal.cur->dirty) |
736 | journal_try_write(c); | |
737 | else | |
738 | spin_unlock(&c->journal.lock); | |
7857d5d4 KO |
739 | } |
740 | ||
cafe5635 KO |
741 | /* |
742 | * Entry point to the journalling code - bio_insert() and btree_invalidate() | |
743 | * pass bch_journal() a list of keys to be journalled, and then | |
744 | * bch_journal() hands those same keys off to btree_insert_async() | |
745 | */ | |
746 | ||
a34a8bfd KO |
747 | atomic_t *bch_journal(struct cache_set *c, |
748 | struct keylist *keys, | |
749 | struct closure *parent) | |
cafe5635 | 750 | { |
cafe5635 | 751 | struct journal_write *w; |
a34a8bfd | 752 | atomic_t *ret; |
cafe5635 | 753 | |
a34a8bfd KO |
754 | if (!CACHE_SYNC(&c->sb)) |
755 | return NULL; | |
cafe5635 | 756 | |
a34a8bfd | 757 | w = journal_wait_for_write(c, bch_keylist_nkeys(keys)); |
c37511b8 | 758 | |
fafff81c | 759 | memcpy(bset_bkey_last(w->data), keys->keys, bch_keylist_bytes(keys)); |
a34a8bfd | 760 | w->data->keys += bch_keylist_nkeys(keys); |
cafe5635 | 761 | |
a34a8bfd KO |
762 | ret = &fifo_back(&c->journal.pin); |
763 | atomic_inc(ret); | |
cafe5635 | 764 | |
a34a8bfd KO |
765 | if (parent) { |
766 | closure_wait(&w->wait, parent); | |
7857d5d4 | 767 | journal_try_write(c); |
dabb4433 KO |
768 | } else if (!w->dirty) { |
769 | w->dirty = true; | |
7857d5d4 KO |
770 | schedule_delayed_work(&c->journal.work, |
771 | msecs_to_jiffies(c->journal_delay_ms)); | |
772 | spin_unlock(&c->journal.lock); | |
773 | } else { | |
774 | spin_unlock(&c->journal.lock); | |
cafe5635 | 775 | } |
a34a8bfd KO |
776 | |
777 | ||
778 | return ret; | |
779 | } | |
780 | ||
781 | void bch_journal_meta(struct cache_set *c, struct closure *cl) | |
782 | { | |
783 | struct keylist keys; | |
784 | atomic_t *ref; | |
785 | ||
786 | bch_keylist_init(&keys); | |
787 | ||
788 | ref = bch_journal(c, &keys, cl); | |
789 | if (ref) | |
790 | atomic_dec_bug(ref); | |
cafe5635 KO |
791 | } |
792 | ||
793 | void bch_journal_free(struct cache_set *c) | |
794 | { | |
795 | free_pages((unsigned long) c->journal.w[1].data, JSET_BITS); | |
796 | free_pages((unsigned long) c->journal.w[0].data, JSET_BITS); | |
797 | free_fifo(&c->journal.pin); | |
798 | } | |
799 | ||
800 | int bch_journal_alloc(struct cache_set *c) | |
801 | { | |
802 | struct journal *j = &c->journal; | |
803 | ||
cafe5635 | 804 | spin_lock_init(&j->lock); |
7857d5d4 | 805 | INIT_DELAYED_WORK(&j->work, journal_write_work); |
cafe5635 KO |
806 | |
807 | c->journal_delay_ms = 100; | |
808 | ||
809 | j->w[0].c = c; | |
810 | j->w[1].c = c; | |
811 | ||
812 | if (!(init_fifo(&j->pin, JOURNAL_PIN, GFP_KERNEL)) || | |
813 | !(j->w[0].data = (void *) __get_free_pages(GFP_KERNEL, JSET_BITS)) || | |
814 | !(j->w[1].data = (void *) __get_free_pages(GFP_KERNEL, JSET_BITS))) | |
815 | return -ENOMEM; | |
816 | ||
817 | return 0; | |
818 | } |