xref: /openbmc/linux/drivers/md/bcache/journal.c (revision 275876e2)
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"
10 #include "extents.h"
11 
12 #include <trace/events/bcache.h>
13 
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,
34 			       unsigned bucket_index)
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;
41 	struct closure cl;
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 
46 	closure_init_stack(&cl);
47 
48 	pr_debug("reading %u", bucket_index);
49 
50 	while (offset < ca->sb.bucket_size) {
51 reread:		left = ca->sb.bucket_size - offset;
52 		len = min_t(unsigned, left, PAGE_SECTORS << JSET_BITS);
53 
54 		bio_reset(bio);
55 		bio->bi_iter.bi_sector	= bucket + offset;
56 		bio->bi_bdev	= ca->bdev;
57 		bio->bi_rw	= READ;
58 		bio->bi_iter.bi_size	= len << 9;
59 
60 		bio->bi_end_io	= journal_read_endio;
61 		bio->bi_private = &cl;
62 		bch_bio_map(bio, data);
63 
64 		closure_bio_submit(bio, &cl, ca);
65 		closure_sync(&cl);
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 
78 			if (j->magic != jset_magic(&ca->sb)) {
79 				pr_debug("%u: bad magic", bucket_index);
80 				return ret;
81 			}
82 
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);
87 				return ret;
88 			}
89 
90 			if (bytes > len << 9)
91 				goto reread;
92 
93 			if (j->csum != csum_set(j)) {
94 				pr_info("%u: bad csum, %zu bytes, offset %u",
95 					bucket_index, bytes, offset);
96 				return ret;
97 			}
98 
99 			blocks = set_blocks(j, block_bytes(ca->set));
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 
144 int bch_journal_read(struct cache_set *c, struct list_head *list)
145 {
146 #define read_bucket(b)							\
147 	({								\
148 		int ret = journal_read_bucket(ca, list, b);		\
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 
167 		/*
168 		 * Read journal buckets ordered by golden ratio hash to quickly
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 
181 		/*
182 		 * If that fails, check all the buckets we haven't checked
183 		 * already
184 		 */
185 		pr_debug("falling back to linear search");
186 
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))
190 			if (read_bucket(l))
191 				goto bsearch;
192 
193 		/* no journal entries on this device? */
194 		if (l == ca->sb.njournal_buckets)
195 			continue;
196 bsearch:
197 		BUG_ON(list_empty(list));
198 
199 		/* Binary search */
200 		m = l;
201 		r = find_next_bit(bitmap, ca->sb.njournal_buckets, l + 1);
202 		pr_debug("starting binary search, l %u r %u", l, r);
203 
204 		while (l + 1 < r) {
205 			seq = list_entry(list->prev, struct journal_replay,
206 					 list)->j.seq;
207 
208 			m = (l + r) >> 1;
209 			read_bucket(m);
210 
211 			if (seq != list_entry(list->prev, struct journal_replay,
212 					      list)->j.seq)
213 				l = m;
214 			else
215 				r = m;
216 		}
217 
218 		/*
219 		 * Read buckets in reverse order until we stop finding more
220 		 * journal entries
221 		 */
222 		pr_debug("finishing up: m %u njournal_buckets %u",
223 			 m, ca->sb.njournal_buckets);
224 		l = m;
225 
226 		while (1) {
227 			if (!l--)
228 				l = ca->sb.njournal_buckets - 1;
229 
230 			if (l == m)
231 				break;
232 
233 			if (test_bit(l, bitmap))
234 				continue;
235 
236 			if (!read_bucket(l))
237 				break;
238 		}
239 
240 		seq = 0;
241 
242 		for (i = 0; i < ca->sb.njournal_buckets; i++)
243 			if (ja->seq[i] > seq) {
244 				seq = ja->seq[i];
245 				/*
246 				 * When journal_reclaim() goes to allocate for
247 				 * the first time, it'll use the bucket after
248 				 * ja->cur_idx
249 				 */
250 				ja->cur_idx = i;
251 				ja->last_idx = ja->discard_idx = (i + 1) %
252 					ca->sb.njournal_buckets;
253 
254 			}
255 	}
256 
257 	if (!list_empty(list))
258 		c->journal.seq = list_entry(list->prev,
259 					    struct journal_replay,
260 					    list)->j.seq;
261 
262 	return 0;
263 #undef read_bucket
264 }
265 
266 void bch_journal_mark(struct cache_set *c, struct list_head *list)
267 {
268 	atomic_t p = { 0 };
269 	struct bkey *k;
270 	struct journal_replay *i;
271 	struct journal *j = &c->journal;
272 	uint64_t last = j->seq;
273 
274 	/*
275 	 * journal.pin should never fill up - we never write a journal
276 	 * entry when it would fill up. But if for some reason it does, we
277 	 * iterate over the list in reverse order so that we can just skip that
278 	 * refcount instead of bugging.
279 	 */
280 
281 	list_for_each_entry_reverse(i, list, list) {
282 		BUG_ON(last < i->j.seq);
283 		i->pin = NULL;
284 
285 		while (last-- != i->j.seq)
286 			if (fifo_free(&j->pin) > 1) {
287 				fifo_push_front(&j->pin, p);
288 				atomic_set(&fifo_front(&j->pin), 0);
289 			}
290 
291 		if (fifo_free(&j->pin) > 1) {
292 			fifo_push_front(&j->pin, p);
293 			i->pin = &fifo_front(&j->pin);
294 			atomic_set(i->pin, 1);
295 		}
296 
297 		for (k = i->j.start;
298 		     k < bset_bkey_last(&i->j);
299 		     k = bkey_next(k))
300 			if (!__bch_extent_invalid(c, k)) {
301 				unsigned j;
302 
303 				for (j = 0; j < KEY_PTRS(k); j++)
304 					if (ptr_available(c, k, j))
305 						atomic_inc(&PTR_BUCKET(c, k, j)->pin);
306 
307 				bch_initial_mark_key(c, 0, k);
308 			}
309 	}
310 }
311 
312 int bch_journal_replay(struct cache_set *s, struct list_head *list)
313 {
314 	int ret = 0, keys = 0, entries = 0;
315 	struct bkey *k;
316 	struct journal_replay *i =
317 		list_entry(list->prev, struct journal_replay, list);
318 
319 	uint64_t start = i->j.last_seq, end = i->j.seq, n = start;
320 	struct keylist keylist;
321 
322 	list_for_each_entry(i, list, list) {
323 		BUG_ON(i->pin && atomic_read(i->pin) != 1);
324 
325 		cache_set_err_on(n != i->j.seq, s,
326 "bcache: journal entries %llu-%llu missing! (replaying %llu-%llu)",
327 				 n, i->j.seq - 1, start, end);
328 
329 		for (k = i->j.start;
330 		     k < bset_bkey_last(&i->j);
331 		     k = bkey_next(k)) {
332 			trace_bcache_journal_replay_key(k);
333 
334 			bch_keylist_init_single(&keylist, k);
335 
336 			ret = bch_btree_insert(s, &keylist, i->pin, NULL);
337 			if (ret)
338 				goto err;
339 
340 			BUG_ON(!bch_keylist_empty(&keylist));
341 			keys++;
342 
343 			cond_resched();
344 		}
345 
346 		if (i->pin)
347 			atomic_dec(i->pin);
348 		n = i->j.seq + 1;
349 		entries++;
350 	}
351 
352 	pr_info("journal replay done, %i keys in %i entries, seq %llu",
353 		keys, entries, end);
354 err:
355 	while (!list_empty(list)) {
356 		i = list_first_entry(list, struct journal_replay, list);
357 		list_del(&i->list);
358 		kfree(i);
359 	}
360 
361 	return ret;
362 }
363 
364 /* Journalling */
365 
366 static void btree_flush_write(struct cache_set *c)
367 {
368 	/*
369 	 * Try to find the btree node with that references the oldest journal
370 	 * entry, best is our current candidate and is locked if non NULL:
371 	 */
372 	struct btree *b, *best;
373 	unsigned i;
374 retry:
375 	best = NULL;
376 
377 	for_each_cached_btree(b, c, i)
378 		if (btree_current_write(b)->journal) {
379 			if (!best)
380 				best = b;
381 			else if (journal_pin_cmp(c,
382 					btree_current_write(best)->journal,
383 					btree_current_write(b)->journal)) {
384 				best = b;
385 			}
386 		}
387 
388 	b = best;
389 	if (b) {
390 		mutex_lock(&b->write_lock);
391 		if (!btree_current_write(b)->journal) {
392 			mutex_unlock(&b->write_lock);
393 			/* We raced */
394 			goto retry;
395 		}
396 
397 		__bch_btree_node_write(b, NULL);
398 		mutex_unlock(&b->write_lock);
399 	}
400 }
401 
402 #define last_seq(j)	((j)->seq - fifo_used(&(j)->pin) + 1)
403 
404 static void journal_discard_endio(struct bio *bio, int error)
405 {
406 	struct journal_device *ja =
407 		container_of(bio, struct journal_device, discard_bio);
408 	struct cache *ca = container_of(ja, struct cache, journal);
409 
410 	atomic_set(&ja->discard_in_flight, DISCARD_DONE);
411 
412 	closure_wake_up(&ca->set->journal.wait);
413 	closure_put(&ca->set->cl);
414 }
415 
416 static void journal_discard_work(struct work_struct *work)
417 {
418 	struct journal_device *ja =
419 		container_of(work, struct journal_device, discard_work);
420 
421 	submit_bio(0, &ja->discard_bio);
422 }
423 
424 static void do_journal_discard(struct cache *ca)
425 {
426 	struct journal_device *ja = &ca->journal;
427 	struct bio *bio = &ja->discard_bio;
428 
429 	if (!ca->discard) {
430 		ja->discard_idx = ja->last_idx;
431 		return;
432 	}
433 
434 	switch (atomic_read(&ja->discard_in_flight)) {
435 	case DISCARD_IN_FLIGHT:
436 		return;
437 
438 	case DISCARD_DONE:
439 		ja->discard_idx = (ja->discard_idx + 1) %
440 			ca->sb.njournal_buckets;
441 
442 		atomic_set(&ja->discard_in_flight, DISCARD_READY);
443 		/* fallthrough */
444 
445 	case DISCARD_READY:
446 		if (ja->discard_idx == ja->last_idx)
447 			return;
448 
449 		atomic_set(&ja->discard_in_flight, DISCARD_IN_FLIGHT);
450 
451 		bio_init(bio);
452 		bio->bi_iter.bi_sector	= bucket_to_sector(ca->set,
453 						ca->sb.d[ja->discard_idx]);
454 		bio->bi_bdev		= ca->bdev;
455 		bio->bi_rw		= REQ_WRITE|REQ_DISCARD;
456 		bio->bi_max_vecs	= 1;
457 		bio->bi_io_vec		= bio->bi_inline_vecs;
458 		bio->bi_iter.bi_size	= bucket_bytes(ca);
459 		bio->bi_end_io		= journal_discard_endio;
460 
461 		closure_get(&ca->set->cl);
462 		INIT_WORK(&ja->discard_work, journal_discard_work);
463 		schedule_work(&ja->discard_work);
464 	}
465 }
466 
467 static void journal_reclaim(struct cache_set *c)
468 {
469 	struct bkey *k = &c->journal.key;
470 	struct cache *ca;
471 	uint64_t last_seq;
472 	unsigned iter, n = 0;
473 	atomic_t p;
474 
475 	while (!atomic_read(&fifo_front(&c->journal.pin)))
476 		fifo_pop(&c->journal.pin, p);
477 
478 	last_seq = last_seq(&c->journal);
479 
480 	/* Update last_idx */
481 
482 	for_each_cache(ca, c, iter) {
483 		struct journal_device *ja = &ca->journal;
484 
485 		while (ja->last_idx != ja->cur_idx &&
486 		       ja->seq[ja->last_idx] < last_seq)
487 			ja->last_idx = (ja->last_idx + 1) %
488 				ca->sb.njournal_buckets;
489 	}
490 
491 	for_each_cache(ca, c, iter)
492 		do_journal_discard(ca);
493 
494 	if (c->journal.blocks_free)
495 		goto out;
496 
497 	/*
498 	 * Allocate:
499 	 * XXX: Sort by free journal space
500 	 */
501 
502 	for_each_cache(ca, c, iter) {
503 		struct journal_device *ja = &ca->journal;
504 		unsigned next = (ja->cur_idx + 1) % ca->sb.njournal_buckets;
505 
506 		/* No space available on this device */
507 		if (next == ja->discard_idx)
508 			continue;
509 
510 		ja->cur_idx = next;
511 		k->ptr[n++] = PTR(0,
512 				  bucket_to_sector(c, ca->sb.d[ja->cur_idx]),
513 				  ca->sb.nr_this_dev);
514 	}
515 
516 	bkey_init(k);
517 	SET_KEY_PTRS(k, n);
518 
519 	if (n)
520 		c->journal.blocks_free = c->sb.bucket_size >> c->block_bits;
521 out:
522 	if (!journal_full(&c->journal))
523 		__closure_wake_up(&c->journal.wait);
524 }
525 
526 void bch_journal_next(struct journal *j)
527 {
528 	atomic_t p = { 1 };
529 
530 	j->cur = (j->cur == j->w)
531 		? &j->w[1]
532 		: &j->w[0];
533 
534 	/*
535 	 * The fifo_push() needs to happen at the same time as j->seq is
536 	 * incremented for last_seq() to be calculated correctly
537 	 */
538 	BUG_ON(!fifo_push(&j->pin, p));
539 	atomic_set(&fifo_back(&j->pin), 1);
540 
541 	j->cur->data->seq	= ++j->seq;
542 	j->cur->dirty		= false;
543 	j->cur->need_write	= false;
544 	j->cur->data->keys	= 0;
545 
546 	if (fifo_full(&j->pin))
547 		pr_debug("journal_pin full (%zu)", fifo_used(&j->pin));
548 }
549 
550 static void journal_write_endio(struct bio *bio, int error)
551 {
552 	struct journal_write *w = bio->bi_private;
553 
554 	cache_set_err_on(error, w->c, "journal io error");
555 	closure_put(&w->c->journal.io);
556 }
557 
558 static void journal_write(struct closure *);
559 
560 static void journal_write_done(struct closure *cl)
561 {
562 	struct journal *j = container_of(cl, struct journal, io);
563 	struct journal_write *w = (j->cur == j->w)
564 		? &j->w[1]
565 		: &j->w[0];
566 
567 	__closure_wake_up(&w->wait);
568 	continue_at_nobarrier(cl, journal_write, system_wq);
569 }
570 
571 static void journal_write_unlock(struct closure *cl)
572 {
573 	struct cache_set *c = container_of(cl, struct cache_set, journal.io);
574 
575 	c->journal.io_in_flight = 0;
576 	spin_unlock(&c->journal.lock);
577 }
578 
579 static void journal_write_unlocked(struct closure *cl)
580 	__releases(c->journal.lock)
581 {
582 	struct cache_set *c = container_of(cl, struct cache_set, journal.io);
583 	struct cache *ca;
584 	struct journal_write *w = c->journal.cur;
585 	struct bkey *k = &c->journal.key;
586 	unsigned i, sectors = set_blocks(w->data, block_bytes(c)) *
587 		c->sb.block_size;
588 
589 	struct bio *bio;
590 	struct bio_list list;
591 	bio_list_init(&list);
592 
593 	if (!w->need_write) {
594 		closure_return_with_destructor(cl, journal_write_unlock);
595 	} else if (journal_full(&c->journal)) {
596 		journal_reclaim(c);
597 		spin_unlock(&c->journal.lock);
598 
599 		btree_flush_write(c);
600 		continue_at(cl, journal_write, system_wq);
601 	}
602 
603 	c->journal.blocks_free -= set_blocks(w->data, block_bytes(c));
604 
605 	w->data->btree_level = c->root->level;
606 
607 	bkey_copy(&w->data->btree_root, &c->root->key);
608 	bkey_copy(&w->data->uuid_bucket, &c->uuid_bucket);
609 
610 	for_each_cache(ca, c, i)
611 		w->data->prio_bucket[ca->sb.nr_this_dev] = ca->prio_buckets[0];
612 
613 	w->data->magic		= jset_magic(&c->sb);
614 	w->data->version	= BCACHE_JSET_VERSION;
615 	w->data->last_seq	= last_seq(&c->journal);
616 	w->data->csum		= csum_set(w->data);
617 
618 	for (i = 0; i < KEY_PTRS(k); i++) {
619 		ca = PTR_CACHE(c, k, i);
620 		bio = &ca->journal.bio;
621 
622 		atomic_long_add(sectors, &ca->meta_sectors_written);
623 
624 		bio_reset(bio);
625 		bio->bi_iter.bi_sector	= PTR_OFFSET(k, i);
626 		bio->bi_bdev	= ca->bdev;
627 		bio->bi_rw	= REQ_WRITE|REQ_SYNC|REQ_META|REQ_FLUSH|REQ_FUA;
628 		bio->bi_iter.bi_size = sectors << 9;
629 
630 		bio->bi_end_io	= journal_write_endio;
631 		bio->bi_private = w;
632 		bch_bio_map(bio, w->data);
633 
634 		trace_bcache_journal_write(bio);
635 		bio_list_add(&list, bio);
636 
637 		SET_PTR_OFFSET(k, i, PTR_OFFSET(k, i) + sectors);
638 
639 		ca->journal.seq[ca->journal.cur_idx] = w->data->seq;
640 	}
641 
642 	atomic_dec_bug(&fifo_back(&c->journal.pin));
643 	bch_journal_next(&c->journal);
644 	journal_reclaim(c);
645 
646 	spin_unlock(&c->journal.lock);
647 
648 	while ((bio = bio_list_pop(&list)))
649 		closure_bio_submit(bio, cl, c->cache[0]);
650 
651 	continue_at(cl, journal_write_done, NULL);
652 }
653 
654 static void journal_write(struct closure *cl)
655 {
656 	struct cache_set *c = container_of(cl, struct cache_set, journal.io);
657 
658 	spin_lock(&c->journal.lock);
659 	journal_write_unlocked(cl);
660 }
661 
662 static void journal_try_write(struct cache_set *c)
663 	__releases(c->journal.lock)
664 {
665 	struct closure *cl = &c->journal.io;
666 	struct journal_write *w = c->journal.cur;
667 
668 	w->need_write = true;
669 
670 	if (!c->journal.io_in_flight) {
671 		c->journal.io_in_flight = 1;
672 		closure_call(cl, journal_write_unlocked, NULL, &c->cl);
673 	} else {
674 		spin_unlock(&c->journal.lock);
675 	}
676 }
677 
678 static struct journal_write *journal_wait_for_write(struct cache_set *c,
679 						    unsigned nkeys)
680 {
681 	size_t sectors;
682 	struct closure cl;
683 	bool wait = false;
684 
685 	closure_init_stack(&cl);
686 
687 	spin_lock(&c->journal.lock);
688 
689 	while (1) {
690 		struct journal_write *w = c->journal.cur;
691 
692 		sectors = __set_blocks(w->data, w->data->keys + nkeys,
693 				       block_bytes(c)) * c->sb.block_size;
694 
695 		if (sectors <= min_t(size_t,
696 				     c->journal.blocks_free * c->sb.block_size,
697 				     PAGE_SECTORS << JSET_BITS))
698 			return w;
699 
700 		if (wait)
701 			closure_wait(&c->journal.wait, &cl);
702 
703 		if (!journal_full(&c->journal)) {
704 			if (wait)
705 				trace_bcache_journal_entry_full(c);
706 
707 			/*
708 			 * XXX: If we were inserting so many keys that they
709 			 * won't fit in an _empty_ journal write, we'll
710 			 * deadlock. For now, handle this in
711 			 * bch_keylist_realloc() - but something to think about.
712 			 */
713 			BUG_ON(!w->data->keys);
714 
715 			journal_try_write(c); /* unlocks */
716 		} else {
717 			if (wait)
718 				trace_bcache_journal_full(c);
719 
720 			journal_reclaim(c);
721 			spin_unlock(&c->journal.lock);
722 
723 			btree_flush_write(c);
724 		}
725 
726 		closure_sync(&cl);
727 		spin_lock(&c->journal.lock);
728 		wait = true;
729 	}
730 }
731 
732 static void journal_write_work(struct work_struct *work)
733 {
734 	struct cache_set *c = container_of(to_delayed_work(work),
735 					   struct cache_set,
736 					   journal.work);
737 	spin_lock(&c->journal.lock);
738 	if (c->journal.cur->dirty)
739 		journal_try_write(c);
740 	else
741 		spin_unlock(&c->journal.lock);
742 }
743 
744 /*
745  * Entry point to the journalling code - bio_insert() and btree_invalidate()
746  * pass bch_journal() a list of keys to be journalled, and then
747  * bch_journal() hands those same keys off to btree_insert_async()
748  */
749 
750 atomic_t *bch_journal(struct cache_set *c,
751 		      struct keylist *keys,
752 		      struct closure *parent)
753 {
754 	struct journal_write *w;
755 	atomic_t *ret;
756 
757 	if (!CACHE_SYNC(&c->sb))
758 		return NULL;
759 
760 	w = journal_wait_for_write(c, bch_keylist_nkeys(keys));
761 
762 	memcpy(bset_bkey_last(w->data), keys->keys, bch_keylist_bytes(keys));
763 	w->data->keys += bch_keylist_nkeys(keys);
764 
765 	ret = &fifo_back(&c->journal.pin);
766 	atomic_inc(ret);
767 
768 	if (parent) {
769 		closure_wait(&w->wait, parent);
770 		journal_try_write(c);
771 	} else if (!w->dirty) {
772 		w->dirty = true;
773 		schedule_delayed_work(&c->journal.work,
774 				      msecs_to_jiffies(c->journal_delay_ms));
775 		spin_unlock(&c->journal.lock);
776 	} else {
777 		spin_unlock(&c->journal.lock);
778 	}
779 
780 
781 	return ret;
782 }
783 
784 void bch_journal_meta(struct cache_set *c, struct closure *cl)
785 {
786 	struct keylist keys;
787 	atomic_t *ref;
788 
789 	bch_keylist_init(&keys);
790 
791 	ref = bch_journal(c, &keys, cl);
792 	if (ref)
793 		atomic_dec_bug(ref);
794 }
795 
796 void bch_journal_free(struct cache_set *c)
797 {
798 	free_pages((unsigned long) c->journal.w[1].data, JSET_BITS);
799 	free_pages((unsigned long) c->journal.w[0].data, JSET_BITS);
800 	free_fifo(&c->journal.pin);
801 }
802 
803 int bch_journal_alloc(struct cache_set *c)
804 {
805 	struct journal *j = &c->journal;
806 
807 	spin_lock_init(&j->lock);
808 	INIT_DELAYED_WORK(&j->work, journal_write_work);
809 
810 	c->journal_delay_ms = 100;
811 
812 	j->w[0].c = c;
813 	j->w[1].c = c;
814 
815 	if (!(init_fifo(&j->pin, JOURNAL_PIN, GFP_KERNEL)) ||
816 	    !(j->w[0].data = (void *) __get_free_pages(GFP_KERNEL, JSET_BITS)) ||
817 	    !(j->w[1].data = (void *) __get_free_pages(GFP_KERNEL, JSET_BITS)))
818 		return -ENOMEM;
819 
820 	return 0;
821 }
822