?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Linux 4.19.133
[linux/fpc-iii.git] / drivers / md / bcache / journal.c
blob7bb15cddca5ecb6dbd25d3f21b2ca201e9d106f8
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * bcache journalling code, for btree insertions
4 *
5 * Copyright 2012 Google, Inc.
6 */
7
8 #include "bcache.h"
9 #include "btree.h"
10 #include "debug.h"
11 #include "extents.h"
12
13 #include <trace/events/bcache.h>
14
15 /*
16 * Journal replay/recovery:
17 *
18 * This code is all driven from run_cache_set(); we first read the journal
19 * entries, do some other stuff, then we mark all the keys in the journal
20 * entries (same as garbage collection would), then we replay them - reinserting
21 * them into the cache in precisely the same order as they appear in the
22 * journal.
23 *
24 * We only journal keys that go in leaf nodes, which simplifies things quite a
25 * bit.
26 */
27
28 static void journal_read_endio(struct bio *bio)
29 {
30 struct closure *cl = bio->bi_private;
31
32 closure_put(cl);
33 }
34
35 static int journal_read_bucket(struct cache *ca, struct list_head *list,
36 unsigned int bucket_index)
37 {
38 struct journal_device *ja = &ca->journal;
39 struct bio *bio = &ja->bio;
40
41 struct journal_replay *i;
42 struct jset *j, *data = ca->set->journal.w[0].data;
43 struct closure cl;
44 unsigned int len, left, offset = 0;
45 int ret = 0;
46 sector_t bucket = bucket_to_sector(ca->set, ca->sb.d[bucket_index]);
47
48 closure_init_stack(&cl);
49
50 pr_debug("reading %u", bucket_index);
51
52 while (offset < ca->sb.bucket_size) {
53 reread: left = ca->sb.bucket_size - offset;
54 len = min_t(unsigned int, left, PAGE_SECTORS << JSET_BITS);
55
56 bio_reset(bio);
57 bio->bi_iter.bi_sector = bucket + offset;
58 bio_set_dev(bio, ca->bdev);
59 bio->bi_iter.bi_size = len << 9;
60
61 bio->bi_end_io = journal_read_endio;
62 bio->bi_private = &cl;
63 bio_set_op_attrs(bio, REQ_OP_READ, 0);
64 bch_bio_map(bio, data);
65
66 closure_bio_submit(ca->set, bio, &cl);
67 closure_sync(&cl);
68
69 /* This function could be simpler now since we no longer write
70 * journal entries that overlap bucket boundaries; this means
71 * the start of a bucket will always have a valid journal entry
72 * if it has any journal entries at all.
73 */
74
75 j = data;
76 while (len) {
77 struct list_head *where;
78 size_t blocks, bytes = set_bytes(j);
79
80 if (j->magic != jset_magic(&ca->sb)) {
81 pr_debug("%u: bad magic", bucket_index);
82 return ret;
83 }
84
85 if (bytes > left << 9 ||
86 bytes > PAGE_SIZE << JSET_BITS) {
87 pr_info("%u: too big, %zu bytes, offset %u",
88 bucket_index, bytes, offset);
89 return ret;
90 }
91
92 if (bytes > len << 9)
93 goto reread;
94
95 if (j->csum != csum_set(j)) {
96 pr_info("%u: bad csum, %zu bytes, offset %u",
97 bucket_index, bytes, offset);
98 return ret;
99 }
100
101 blocks = set_blocks(j, block_bytes(ca->set));
102
103 while (!list_empty(list)) {
104 i = list_first_entry(list,
105 struct journal_replay, list);
106 if (i->j.seq >= j->last_seq)
107 break;
108 list_del(&i->list);
109 kfree(i);
110 }
111
112 list_for_each_entry_reverse(i, list, list) {
113 if (j->seq == i->j.seq)
114 goto next_set;
115
116 if (j->seq < i->j.last_seq)
117 goto next_set;
118
119 if (j->seq > i->j.seq) {
120 where = &i->list;
121 goto add;
122 }
123 }
124
125 where = list;
126 add:
127 i = kmalloc(offsetof(struct journal_replay, j) +
128 bytes, GFP_KERNEL);
129 if (!i)
130 return -ENOMEM;
131 memcpy(&i->j, j, bytes);
132 list_add(&i->list, where);
133 ret = 1;
134
135 ja->seq[bucket_index] = j->seq;
136 next_set:
137 offset += blocks * ca->sb.block_size;
138 len -= blocks * ca->sb.block_size;
139 j = ((void *) j) + blocks * block_bytes(ca);
140 }
141 }
142
143 return ret;
144 }
145
146 int bch_journal_read(struct cache_set *c, struct list_head *list)
147 {
148 #define read_bucket(b) \
149 ({ \
150 int ret = journal_read_bucket(ca, list, b); \
151 __set_bit(b, bitmap); \
152 if (ret < 0) \
153 return ret; \
154 ret; \
155 })
156
157 struct cache *ca;
158 unsigned int iter;
159
160 for_each_cache(ca, c, iter) {
161 struct journal_device *ja = &ca->journal;
162 DECLARE_BITMAP(bitmap, SB_JOURNAL_BUCKETS);
163 unsigned int i, l, r, m;
164 uint64_t seq;
165
166 bitmap_zero(bitmap, SB_JOURNAL_BUCKETS);
167 pr_debug("%u journal buckets", ca->sb.njournal_buckets);
168
169 /*
170 * Read journal buckets ordered by golden ratio hash to quickly
171 * find a sequence of buckets with valid journal entries
172 */
173 for (i = 0; i < ca->sb.njournal_buckets; i++) {
174 /*
175 * We must try the index l with ZERO first for
176 * correctness due to the scenario that the journal
177 * bucket is circular buffer which might have wrapped
178 */
179 l = (i * 2654435769U) % ca->sb.njournal_buckets;
180
181 if (test_bit(l, bitmap))
182 break;
183
184 if (read_bucket(l))
185 goto bsearch;
186 }
187
188 /*
189 * If that fails, check all the buckets we haven't checked
190 * already
191 */
192 pr_debug("falling back to linear search");
193
194 for (l = find_first_zero_bit(bitmap, ca->sb.njournal_buckets);
195 l < ca->sb.njournal_buckets;
196 l = find_next_zero_bit(bitmap, ca->sb.njournal_buckets,
197 l + 1))
198 if (read_bucket(l))
199 goto bsearch;
200
201 /* no journal entries on this device? */
202 if (l == ca->sb.njournal_buckets)
203 continue;
204 bsearch:
205 BUG_ON(list_empty(list));
206
207 /* Binary search */
208 m = l;
209 r = find_next_bit(bitmap, ca->sb.njournal_buckets, l + 1);
210 pr_debug("starting binary search, l %u r %u", l, r);
211
212 while (l + 1 < r) {
213 seq = list_entry(list->prev, struct journal_replay,
214 list)->j.seq;
215
216 m = (l + r) >> 1;
217 read_bucket(m);
218
219 if (seq != list_entry(list->prev, struct journal_replay,
220 list)->j.seq)
221 l = m;
222 else
223 r = m;
224 }
225
226 /*
227 * Read buckets in reverse order until we stop finding more
228 * journal entries
229 */
230 pr_debug("finishing up: m %u njournal_buckets %u",
231 m, ca->sb.njournal_buckets);
232 l = m;
233
234 while (1) {
235 if (!l--)
236 l = ca->sb.njournal_buckets - 1;
237
238 if (l == m)
239 break;
240
241 if (test_bit(l, bitmap))
242 continue;
243
244 if (!read_bucket(l))
245 break;
246 }
247
248 seq = 0;
249
250 for (i = 0; i < ca->sb.njournal_buckets; i++)
251 if (ja->seq[i] > seq) {
252 seq = ja->seq[i];
253 /*
254 * When journal_reclaim() goes to allocate for
255 * the first time, it'll use the bucket after
256 * ja->cur_idx
257 */
258 ja->cur_idx = i;
259 ja->last_idx = ja->discard_idx = (i + 1) %
260 ca->sb.njournal_buckets;
261
262 }
263 }
264
265 if (!list_empty(list))
266 c->journal.seq = list_entry(list->prev,
267 struct journal_replay,
268 list)->j.seq;
269
270 return 0;
271 #undef read_bucket
272 }
273
274 void bch_journal_mark(struct cache_set *c, struct list_head *list)
275 {
276 atomic_t p = { 0 };
277 struct bkey *k;
278 struct journal_replay *i;
279 struct journal *j = &c->journal;
280 uint64_t last = j->seq;
281
282 /*
283 * journal.pin should never fill up - we never write a journal
284 * entry when it would fill up. But if for some reason it does, we
285 * iterate over the list in reverse order so that we can just skip that
286 * refcount instead of bugging.
287 */
288
289 list_for_each_entry_reverse(i, list, list) {
290 BUG_ON(last < i->j.seq);
291 i->pin = NULL;
292
293 while (last-- != i->j.seq)
294 if (fifo_free(&j->pin) > 1) {
295 fifo_push_front(&j->pin, p);
296 atomic_set(&fifo_front(&j->pin), 0);
297 }
298
299 if (fifo_free(&j->pin) > 1) {
300 fifo_push_front(&j->pin, p);
301 i->pin = &fifo_front(&j->pin);
302 atomic_set(i->pin, 1);
303 }
304
305 for (k = i->j.start;
306 k < bset_bkey_last(&i->j);
307 k = bkey_next(k))
308 if (!__bch_extent_invalid(c, k)) {
309 unsigned int j;
310
311 for (j = 0; j < KEY_PTRS(k); j++)
312 if (ptr_available(c, k, j))
313 atomic_inc(&PTR_BUCKET(c, k, j)->pin);
314
315 bch_initial_mark_key(c, 0, k);
316 }
317 }
318 }
319
320 bool is_discard_enabled(struct cache_set *s)
321 {
322 struct cache *ca;
323 unsigned int i;
324
325 for_each_cache(ca, s, i)
326 if (ca->discard)
327 return true;
328
329 return false;
330 }
331
332 int bch_journal_replay(struct cache_set *s, struct list_head *list)
333 {
334 int ret = 0, keys = 0, entries = 0;
335 struct bkey *k;
336 struct journal_replay *i =
337 list_entry(list->prev, struct journal_replay, list);
338
339 uint64_t start = i->j.last_seq, end = i->j.seq, n = start;
340 struct keylist keylist;
341
342 list_for_each_entry(i, list, list) {
343 BUG_ON(i->pin && atomic_read(i->pin) != 1);
344
345 if (n != i->j.seq) {
346 if (n == start && is_discard_enabled(s))
347 pr_info("bcache: journal entries %llu-%llu may be discarded! (replaying %llu-%llu)",
348 n, i->j.seq - 1, start, end);
349 else {
350 pr_err("bcache: journal entries %llu-%llu missing! (replaying %llu-%llu)",
351 n, i->j.seq - 1, start, end);
352 ret = -EIO;
353 goto err;
354 }
355 }
356
357 for (k = i->j.start;
358 k < bset_bkey_last(&i->j);
359 k = bkey_next(k)) {
360 trace_bcache_journal_replay_key(k);
361
362 bch_keylist_init_single(&keylist, k);
363
364 ret = bch_btree_insert(s, &keylist, i->pin, NULL);
365 if (ret)
366 goto err;
367
368 BUG_ON(!bch_keylist_empty(&keylist));
369 keys++;
370
371 cond_resched();
372 }
373
374 if (i->pin)
375 atomic_dec(i->pin);
376 n = i->j.seq + 1;
377 entries++;
378 }
379
380 pr_info("journal replay done, %i keys in %i entries, seq %llu",
381 keys, entries, end);
382 err:
383 while (!list_empty(list)) {
384 i = list_first_entry(list, struct journal_replay, list);
385 list_del(&i->list);
386 kfree(i);
387 }
388
389 return ret;
390 }
391
392 /* Journalling */
393
394 static void btree_flush_write(struct cache_set *c)
395 {
396 /*
397 * Try to find the btree node with that references the oldest journal
398 * entry, best is our current candidate and is locked if non NULL:
399 */
400 struct btree *b, *best;
401 unsigned int i;
402
403 atomic_long_inc(&c->flush_write);
404 retry:
405 best = NULL;
406
407 mutex_lock(&c->bucket_lock);
408 for_each_cached_btree(b, c, i)
409 if (btree_current_write(b)->journal) {
410 if (!best)
411 best = b;
412 else if (journal_pin_cmp(c,
413 btree_current_write(best)->journal,
414 btree_current_write(b)->journal)) {
415 best = b;
416 }
417 }
418
419 b = best;
420 if (b)
421 set_btree_node_journal_flush(b);
422 mutex_unlock(&c->bucket_lock);
423
424 if (b) {
425 mutex_lock(&b->write_lock);
426 if (!btree_current_write(b)->journal) {
427 clear_bit(BTREE_NODE_journal_flush, &b->flags);
428 mutex_unlock(&b->write_lock);
429 /* We raced */
430 atomic_long_inc(&c->retry_flush_write);
431 goto retry;
432 }
433
434 __bch_btree_node_write(b, NULL);
435 clear_bit(BTREE_NODE_journal_flush, &b->flags);
436 mutex_unlock(&b->write_lock);
437 }
438 }
439
440 #define last_seq(j) ((j)->seq - fifo_used(&(j)->pin) + 1)
441
442 static void journal_discard_endio(struct bio *bio)
443 {
444 struct journal_device *ja =
445 container_of(bio, struct journal_device, discard_bio);
446 struct cache *ca = container_of(ja, struct cache, journal);
447
448 atomic_set(&ja->discard_in_flight, DISCARD_DONE);
449
450 closure_wake_up(&ca->set->journal.wait);
451 closure_put(&ca->set->cl);
452 }
453
454 static void journal_discard_work(struct work_struct *work)
455 {
456 struct journal_device *ja =
457 container_of(work, struct journal_device, discard_work);
458
459 submit_bio(&ja->discard_bio);
460 }
461
462 static void do_journal_discard(struct cache *ca)
463 {
464 struct journal_device *ja = &ca->journal;
465 struct bio *bio = &ja->discard_bio;
466
467 if (!ca->discard) {
468 ja->discard_idx = ja->last_idx;
469 return;
470 }
471
472 switch (atomic_read(&ja->discard_in_flight)) {
473 case DISCARD_IN_FLIGHT:
474 return;
475
476 case DISCARD_DONE:
477 ja->discard_idx = (ja->discard_idx + 1) %
478 ca->sb.njournal_buckets;
479
480 atomic_set(&ja->discard_in_flight, DISCARD_READY);
481 /* fallthrough */
482
483 case DISCARD_READY:
484 if (ja->discard_idx == ja->last_idx)
485 return;
486
487 atomic_set(&ja->discard_in_flight, DISCARD_IN_FLIGHT);
488
489 bio_init(bio, bio->bi_inline_vecs, 1);
490 bio_set_op_attrs(bio, REQ_OP_DISCARD, 0);
491 bio->bi_iter.bi_sector = bucket_to_sector(ca->set,
492 ca->sb.d[ja->discard_idx]);
493 bio_set_dev(bio, ca->bdev);
494 bio->bi_iter.bi_size = bucket_bytes(ca);
495 bio->bi_end_io = journal_discard_endio;
496
497 closure_get(&ca->set->cl);
498 INIT_WORK(&ja->discard_work, journal_discard_work);
499 queue_work(bch_journal_wq, &ja->discard_work);
500 }
501 }
502
503 static void journal_reclaim(struct cache_set *c)
504 {
505 struct bkey *k = &c->journal.key;
506 struct cache *ca;
507 uint64_t last_seq;
508 unsigned int iter, n = 0;
509 atomic_t p __maybe_unused;
510
511 atomic_long_inc(&c->reclaim);
512
513 while (!atomic_read(&fifo_front(&c->journal.pin)))
514 fifo_pop(&c->journal.pin, p);
515
516 last_seq = last_seq(&c->journal);
517
518 /* Update last_idx */
519
520 for_each_cache(ca, c, iter) {
521 struct journal_device *ja = &ca->journal;
522
523 while (ja->last_idx != ja->cur_idx &&
524 ja->seq[ja->last_idx] < last_seq)
525 ja->last_idx = (ja->last_idx + 1) %
526 ca->sb.njournal_buckets;
527 }
528
529 for_each_cache(ca, c, iter)
530 do_journal_discard(ca);
531
532 if (c->journal.blocks_free)
533 goto out;
534
535 /*
536 * Allocate:
537 * XXX: Sort by free journal space
538 */
539
540 for_each_cache(ca, c, iter) {
541 struct journal_device *ja = &ca->journal;
542 unsigned int next = (ja->cur_idx + 1) % ca->sb.njournal_buckets;
543
544 /* No space available on this device */
545 if (next == ja->discard_idx)
546 continue;
547
548 ja->cur_idx = next;
549 k->ptr[n++] = MAKE_PTR(0,
550 bucket_to_sector(c, ca->sb.d[ja->cur_idx]),
551 ca->sb.nr_this_dev);
552 }
553
554 if (n) {
555 bkey_init(k);
556 SET_KEY_PTRS(k, n);
557 c->journal.blocks_free = c->sb.bucket_size >> c->block_bits;
558 }
559 out:
560 if (!journal_full(&c->journal))
561 __closure_wake_up(&c->journal.wait);
562 }
563
564 void bch_journal_next(struct journal *j)
565 {
566 atomic_t p = { 1 };
567
568 j->cur = (j->cur == j->w)
569 ? &j->w[1]
570 : &j->w[0];
571
572 /*
573 * The fifo_push() needs to happen at the same time as j->seq is
574 * incremented for last_seq() to be calculated correctly
575 */
576 BUG_ON(!fifo_push(&j->pin, p));
577 atomic_set(&fifo_back(&j->pin), 1);
578
579 j->cur->data->seq = ++j->seq;
580 j->cur->dirty = false;
581 j->cur->need_write = false;
582 j->cur->data->keys = 0;
583
584 if (fifo_full(&j->pin))
585 pr_debug("journal_pin full (%zu)", fifo_used(&j->pin));
586 }
587
588 static void journal_write_endio(struct bio *bio)
589 {
590 struct journal_write *w = bio->bi_private;
591
592 cache_set_err_on(bio->bi_status, w->c, "journal io error");
593 closure_put(&w->c->journal.io);
594 }
595
596 static void journal_write(struct closure *cl);
597
598 static void journal_write_done(struct closure *cl)
599 {
600 struct journal *j = container_of(cl, struct journal, io);
601 struct journal_write *w = (j->cur == j->w)
602 ? &j->w[1]
603 : &j->w[0];
604
605 __closure_wake_up(&w->wait);
606 continue_at_nobarrier(cl, journal_write, bch_journal_wq);
607 }
608
609 static void journal_write_unlock(struct closure *cl)
610 __releases(&c->journal.lock)
611 {
612 struct cache_set *c = container_of(cl, struct cache_set, journal.io);
613
614 c->journal.io_in_flight = 0;
615 spin_unlock(&c->journal.lock);
616 }
617
618 static void journal_write_unlocked(struct closure *cl)
619 __releases(c->journal.lock)
620 {
621 struct cache_set *c = container_of(cl, struct cache_set, journal.io);
622 struct cache *ca;
623 struct journal_write *w = c->journal.cur;
624 struct bkey *k = &c->journal.key;
625 unsigned int i, sectors = set_blocks(w->data, block_bytes(c)) *
626 c->sb.block_size;
627
628 struct bio *bio;
629 struct bio_list list;
630
631 bio_list_init(&list);
632
633 if (!w->need_write) {
634 closure_return_with_destructor(cl, journal_write_unlock);
635 return;
636 } else if (journal_full(&c->journal)) {
637 journal_reclaim(c);
638 spin_unlock(&c->journal.lock);
639
640 btree_flush_write(c);
641 continue_at(cl, journal_write, bch_journal_wq);
642 return;
643 }
644
645 c->journal.blocks_free -= set_blocks(w->data, block_bytes(c));
646
647 w->data->btree_level = c->root->level;
648
649 bkey_copy(&w->data->btree_root, &c->root->key);
650 bkey_copy(&w->data->uuid_bucket, &c->uuid_bucket);
651
652 for_each_cache(ca, c, i)
653 w->data->prio_bucket[ca->sb.nr_this_dev] = ca->prio_buckets[0];
654
655 w->data->magic = jset_magic(&c->sb);
656 w->data->version = BCACHE_JSET_VERSION;
657 w->data->last_seq = last_seq(&c->journal);
658 w->data->csum = csum_set(w->data);
659
660 for (i = 0; i < KEY_PTRS(k); i++) {
661 ca = PTR_CACHE(c, k, i);
662 bio = &ca->journal.bio;
663
664 atomic_long_add(sectors, &ca->meta_sectors_written);
665
666 bio_reset(bio);
667 bio->bi_iter.bi_sector = PTR_OFFSET(k, i);
668 bio_set_dev(bio, ca->bdev);
669 bio->bi_iter.bi_size = sectors << 9;
670
671 bio->bi_end_io = journal_write_endio;
672 bio->bi_private = w;
673 bio_set_op_attrs(bio, REQ_OP_WRITE,
674 REQ_SYNC|REQ_META|REQ_PREFLUSH|REQ_FUA);
675 bch_bio_map(bio, w->data);
676
677 trace_bcache_journal_write(bio);
678 bio_list_add(&list, bio);
679
680 SET_PTR_OFFSET(k, i, PTR_OFFSET(k, i) + sectors);
681
682 ca->journal.seq[ca->journal.cur_idx] = w->data->seq;
683 }
684
685 /* If KEY_PTRS(k) == 0, this jset gets lost in air */
686 BUG_ON(i == 0);
687
688 atomic_dec_bug(&fifo_back(&c->journal.pin));
689 bch_journal_next(&c->journal);
690 journal_reclaim(c);
691
692 spin_unlock(&c->journal.lock);
693
694 while ((bio = bio_list_pop(&list)))
695 closure_bio_submit(c, bio, cl);
696
697 continue_at(cl, journal_write_done, NULL);
698 }
699
700 static void journal_write(struct closure *cl)
701 {
702 struct cache_set *c = container_of(cl, struct cache_set, journal.io);
703
704 spin_lock(&c->journal.lock);
705 journal_write_unlocked(cl);
706 }
707
708 static void journal_try_write(struct cache_set *c)
709 __releases(c->journal.lock)
710 {
711 struct closure *cl = &c->journal.io;
712 struct journal_write *w = c->journal.cur;
713
714 w->need_write = true;
715
716 if (!c->journal.io_in_flight) {
717 c->journal.io_in_flight = 1;
718 closure_call(cl, journal_write_unlocked, NULL, &c->cl);
719 } else {
720 spin_unlock(&c->journal.lock);
721 }
722 }
723
724 static struct journal_write *journal_wait_for_write(struct cache_set *c,
725 unsigned int nkeys)
726 __acquires(&c->journal.lock)
727 {
728 size_t sectors;
729 struct closure cl;
730 bool wait = false;
731
732 closure_init_stack(&cl);
733
734 spin_lock(&c->journal.lock);
735
736 while (1) {
737 struct journal_write *w = c->journal.cur;
738
739 sectors = __set_blocks(w->data, w->data->keys + nkeys,
740 block_bytes(c)) * c->sb.block_size;
741
742 if (sectors <= min_t(size_t,
743 c->journal.blocks_free * c->sb.block_size,
744 PAGE_SECTORS << JSET_BITS))
745 return w;
746
747 if (wait)
748 closure_wait(&c->journal.wait, &cl);
749
750 if (!journal_full(&c->journal)) {
751 if (wait)
752 trace_bcache_journal_entry_full(c);
753
754 /*
755 * XXX: If we were inserting so many keys that they
756 * won't fit in an _empty_ journal write, we'll
757 * deadlock. For now, handle this in
758 * bch_keylist_realloc() - but something to think about.
759 */
760 BUG_ON(!w->data->keys);
761
762 journal_try_write(c); /* unlocks */
763 } else {
764 if (wait)
765 trace_bcache_journal_full(c);
766
767 journal_reclaim(c);
768 spin_unlock(&c->journal.lock);
769
770 btree_flush_write(c);
771 }
772
773 closure_sync(&cl);
774 spin_lock(&c->journal.lock);
775 wait = true;
776 }
777 }
778
779 static void journal_write_work(struct work_struct *work)
780 {
781 struct cache_set *c = container_of(to_delayed_work(work),
782 struct cache_set,
783 journal.work);
784 spin_lock(&c->journal.lock);
785 if (c->journal.cur->dirty)
786 journal_try_write(c);
787 else
788 spin_unlock(&c->journal.lock);
789 }
790
791 /*
792 * Entry point to the journalling code - bio_insert() and btree_invalidate()
793 * pass bch_journal() a list of keys to be journalled, and then
794 * bch_journal() hands those same keys off to btree_insert_async()
795 */
796
797 atomic_t *bch_journal(struct cache_set *c,
798 struct keylist *keys,
799 struct closure *parent)
800 {
801 struct journal_write *w;
802 atomic_t *ret;
803
804 /* No journaling if CACHE_SET_IO_DISABLE set already */
805 if (unlikely(test_bit(CACHE_SET_IO_DISABLE, &c->flags)))
806 return NULL;
807
808 if (!CACHE_SYNC(&c->sb))
809 return NULL;
810
811 w = journal_wait_for_write(c, bch_keylist_nkeys(keys));
812
813 memcpy(bset_bkey_last(w->data), keys->keys, bch_keylist_bytes(keys));
814 w->data->keys += bch_keylist_nkeys(keys);
815
816 ret = &fifo_back(&c->journal.pin);
817 atomic_inc(ret);
818
819 if (parent) {
820 closure_wait(&w->wait, parent);
821 journal_try_write(c);
822 } else if (!w->dirty) {
823 w->dirty = true;
824 schedule_delayed_work(&c->journal.work,
825 msecs_to_jiffies(c->journal_delay_ms));
826 spin_unlock(&c->journal.lock);
827 } else {
828 spin_unlock(&c->journal.lock);
829 }
830
831
832 return ret;
833 }
834
835 void bch_journal_meta(struct cache_set *c, struct closure *cl)
836 {
837 struct keylist keys;
838 atomic_t *ref;
839
840 bch_keylist_init(&keys);
841
842 ref = bch_journal(c, &keys, cl);
843 if (ref)
844 atomic_dec_bug(ref);
845 }
846
847 void bch_journal_free(struct cache_set *c)
848 {
849 free_pages((unsigned long) c->journal.w[1].data, JSET_BITS);
850 free_pages((unsigned long) c->journal.w[0].data, JSET_BITS);
851 free_fifo(&c->journal.pin);
852 }
853
854 int bch_journal_alloc(struct cache_set *c)
855 {
856 struct journal *j = &c->journal;
857
858 spin_lock_init(&j->lock);
859 INIT_DELAYED_WORK(&j->work, journal_write_work);
860
861 c->journal_delay_ms = 100;
862
863 j->w[0].c = c;
864 j->w[1].c = c;
865
866 if (!(init_fifo(&j->pin, JOURNAL_PIN, GFP_KERNEL)) ||
867 !(j->w[0].data = (void *) __get_free_pages(GFP_KERNEL, JSET_BITS)) ||
868 !(j->w[1].data = (void *) __get_free_pages(GFP_KERNEL, JSET_BITS)))
869 return -ENOMEM;
870
871 return 0;
872 }
Linux with added FPC-III support