xref: /linux/fs/bcachefs/journal_reclaim.c (revision ff0905bbf991f4337b5ebc19c0d43525ebb0d96b)
1 // SPDX-License-Identifier: GPL-2.0
2 
3 #include "bcachefs.h"
4 #include "btree_key_cache.h"
5 #include "btree_update.h"
6 #include "btree_write_buffer.h"
7 #include "buckets.h"
8 #include "errcode.h"
9 #include "error.h"
10 #include "journal.h"
11 #include "journal_io.h"
12 #include "journal_reclaim.h"
13 #include "replicas.h"
14 #include "sb-members.h"
15 #include "trace.h"
16 
17 #include <linux/kthread.h>
18 #include <linux/sched/mm.h>
19 
20 static bool __should_discard_bucket(struct journal *, struct journal_device *);
21 
22 /* Free space calculations: */
23 
journal_space_from(struct journal_device * ja,enum journal_space_from from)24 static unsigned journal_space_from(struct journal_device *ja,
25 				   enum journal_space_from from)
26 {
27 	switch (from) {
28 	case journal_space_discarded:
29 		return ja->discard_idx;
30 	case journal_space_clean_ondisk:
31 		return ja->dirty_idx_ondisk;
32 	case journal_space_clean:
33 		return ja->dirty_idx;
34 	default:
35 		BUG();
36 	}
37 }
38 
bch2_journal_dev_buckets_available(struct journal * j,struct journal_device * ja,enum journal_space_from from)39 unsigned bch2_journal_dev_buckets_available(struct journal *j,
40 					    struct journal_device *ja,
41 					    enum journal_space_from from)
42 {
43 	if (!ja->nr)
44 		return 0;
45 
46 	unsigned available = (journal_space_from(ja, from) -
47 			      ja->cur_idx - 1 + ja->nr) % ja->nr;
48 
49 	/*
50 	 * Don't use the last bucket unless writing the new last_seq
51 	 * will make another bucket available:
52 	 */
53 	if (available && ja->dirty_idx_ondisk == ja->dirty_idx)
54 		--available;
55 
56 	return available;
57 }
58 
bch2_journal_set_watermark(struct journal * j)59 void bch2_journal_set_watermark(struct journal *j)
60 {
61 	struct bch_fs *c = container_of(j, struct bch_fs, journal);
62 	bool low_on_space = j->space[journal_space_clean].total * 4 <=
63 		j->space[journal_space_total].total;
64 	bool low_on_pin = fifo_free(&j->pin) < j->pin.size / 4;
65 	bool low_on_wb = bch2_btree_write_buffer_must_wait(c);
66 	unsigned watermark = low_on_space || low_on_pin || low_on_wb
67 		? BCH_WATERMARK_reclaim
68 		: BCH_WATERMARK_stripe;
69 
70 	if (track_event_change(&c->times[BCH_TIME_blocked_journal_low_on_space], low_on_space) ||
71 	    track_event_change(&c->times[BCH_TIME_blocked_journal_low_on_pin], low_on_pin) ||
72 	    track_event_change(&c->times[BCH_TIME_blocked_write_buffer_full], low_on_wb))
73 		trace_and_count(c, journal_full, c);
74 
75 	mod_bit(JOURNAL_space_low, &j->flags, low_on_space || low_on_pin);
76 
77 	swap(watermark, j->watermark);
78 	if (watermark > j->watermark)
79 		journal_wake(j);
80 }
81 
82 static struct journal_space
journal_dev_space_available(struct journal * j,struct bch_dev * ca,enum journal_space_from from)83 journal_dev_space_available(struct journal *j, struct bch_dev *ca,
84 			    enum journal_space_from from)
85 {
86 	struct bch_fs *c = container_of(j, struct bch_fs, journal);
87 	struct journal_device *ja = &ca->journal;
88 	unsigned sectors, buckets, unwritten;
89 	unsigned bucket_size_aligned = round_down(ca->mi.bucket_size, block_sectors(c));
90 	u64 seq;
91 
92 	if (from == journal_space_total)
93 		return (struct journal_space) {
94 			.next_entry	= bucket_size_aligned,
95 			.total		= bucket_size_aligned * ja->nr,
96 		};
97 
98 	buckets = bch2_journal_dev_buckets_available(j, ja, from);
99 	sectors = round_down(ja->sectors_free, block_sectors(c));
100 
101 	/*
102 	 * We that we don't allocate the space for a journal entry
103 	 * until we write it out - thus, account for it here:
104 	 */
105 	for (seq = journal_last_unwritten_seq(j);
106 	     seq <= journal_cur_seq(j);
107 	     seq++) {
108 		unwritten = j->buf[seq & JOURNAL_BUF_MASK].sectors;
109 
110 		if (!unwritten)
111 			continue;
112 
113 		/* entry won't fit on this device, skip: */
114 		if (unwritten > bucket_size_aligned)
115 			continue;
116 
117 		if (unwritten >= sectors) {
118 			if (!buckets) {
119 				sectors = 0;
120 				break;
121 			}
122 
123 			buckets--;
124 			sectors = bucket_size_aligned;
125 		}
126 
127 		sectors -= unwritten;
128 	}
129 
130 	if (sectors < ca->mi.bucket_size && buckets) {
131 		buckets--;
132 		sectors = bucket_size_aligned;
133 	}
134 
135 	return (struct journal_space) {
136 		.next_entry	= sectors,
137 		.total		= sectors + buckets * bucket_size_aligned,
138 	};
139 }
140 
__journal_space_available(struct journal * j,unsigned nr_devs_want,enum journal_space_from from)141 static struct journal_space __journal_space_available(struct journal *j, unsigned nr_devs_want,
142 			    enum journal_space_from from)
143 {
144 	struct bch_fs *c = container_of(j, struct bch_fs, journal);
145 	unsigned pos, nr_devs = 0;
146 	struct journal_space space, dev_space[BCH_SB_MEMBERS_MAX];
147 	unsigned min_bucket_size = U32_MAX;
148 
149 	BUG_ON(nr_devs_want > ARRAY_SIZE(dev_space));
150 
151 	for_each_member_device_rcu(c, ca, &c->rw_devs[BCH_DATA_journal]) {
152 		if (!ca->journal.nr ||
153 		    !ca->mi.durability)
154 			continue;
155 
156 		min_bucket_size = min(min_bucket_size, ca->mi.bucket_size);
157 
158 		space = journal_dev_space_available(j, ca, from);
159 		if (!space.next_entry)
160 			continue;
161 
162 		for (pos = 0; pos < nr_devs; pos++)
163 			if (space.total > dev_space[pos].total)
164 				break;
165 
166 		array_insert_item(dev_space, nr_devs, pos, space);
167 	}
168 
169 	if (nr_devs < nr_devs_want)
170 		return (struct journal_space) { 0, 0 };
171 
172 	/*
173 	 * We sorted largest to smallest, and we want the smallest out of the
174 	 * @nr_devs_want largest devices:
175 	 */
176 	space = dev_space[nr_devs_want - 1];
177 	space.next_entry = min(space.next_entry, min_bucket_size);
178 	return space;
179 }
180 
bch2_journal_space_available(struct journal * j)181 void bch2_journal_space_available(struct journal *j)
182 {
183 	struct bch_fs *c = container_of(j, struct bch_fs, journal);
184 	unsigned clean, clean_ondisk, total;
185 	unsigned max_entry_size	 = min(j->buf[0].buf_size >> 9,
186 				       j->buf[1].buf_size >> 9);
187 	unsigned nr_online = 0, nr_devs_want;
188 	bool can_discard = false;
189 	int ret = 0;
190 
191 	lockdep_assert_held(&j->lock);
192 	guard(rcu)();
193 
194 	for_each_member_device_rcu(c, ca, &c->rw_devs[BCH_DATA_journal]) {
195 		struct journal_device *ja = &ca->journal;
196 
197 		if (!ja->nr)
198 			continue;
199 
200 		while (ja->dirty_idx != ja->cur_idx &&
201 		       ja->bucket_seq[ja->dirty_idx] < journal_last_seq(j))
202 			ja->dirty_idx = (ja->dirty_idx + 1) % ja->nr;
203 
204 		while (ja->dirty_idx_ondisk != ja->dirty_idx &&
205 		       ja->bucket_seq[ja->dirty_idx_ondisk] < j->last_seq_ondisk)
206 			ja->dirty_idx_ondisk = (ja->dirty_idx_ondisk + 1) % ja->nr;
207 
208 		can_discard |= __should_discard_bucket(j, ja);
209 
210 		max_entry_size = min_t(unsigned, max_entry_size, ca->mi.bucket_size);
211 		nr_online++;
212 	}
213 
214 	j->can_discard = can_discard;
215 
216 	if (nr_online < metadata_replicas_required(c)) {
217 		if (!(c->sb.features & BIT_ULL(BCH_FEATURE_small_image))) {
218 			struct printbuf buf = PRINTBUF;
219 			buf.atomic++;
220 			prt_printf(&buf, "insufficient writeable journal devices available: have %u, need %u\n"
221 				   "rw journal devs:", nr_online, metadata_replicas_required(c));
222 
223 			for_each_member_device_rcu(c, ca, &c->rw_devs[BCH_DATA_journal])
224 				prt_printf(&buf, " %s", ca->name);
225 
226 			bch_err(c, "%s", buf.buf);
227 			printbuf_exit(&buf);
228 		}
229 		ret = bch_err_throw(c, insufficient_journal_devices);
230 		goto out;
231 	}
232 
233 	nr_devs_want = min_t(unsigned, nr_online, c->opts.metadata_replicas);
234 
235 	for (unsigned i = 0; i < journal_space_nr; i++)
236 		j->space[i] = __journal_space_available(j, nr_devs_want, i);
237 
238 	clean_ondisk	= j->space[journal_space_clean_ondisk].total;
239 	clean		= j->space[journal_space_clean].total;
240 	total		= j->space[journal_space_total].total;
241 
242 	if (!j->space[journal_space_discarded].next_entry)
243 		ret = bch_err_throw(c, journal_full);
244 
245 	if ((j->space[journal_space_clean_ondisk].next_entry <
246 	     j->space[journal_space_clean_ondisk].total) &&
247 	    (clean - clean_ondisk <= total / 8) &&
248 	    (clean_ondisk * 2 > clean))
249 		set_bit(JOURNAL_may_skip_flush, &j->flags);
250 	else
251 		clear_bit(JOURNAL_may_skip_flush, &j->flags);
252 
253 	bch2_journal_set_watermark(j);
254 out:
255 	j->cur_entry_sectors	= !ret
256 		? j->space[journal_space_discarded].next_entry
257 		: 0;
258 	j->cur_entry_error	= ret;
259 
260 	if (!ret)
261 		journal_wake(j);
262 }
263 
264 /* Discards - last part of journal reclaim: */
265 
__should_discard_bucket(struct journal * j,struct journal_device * ja)266 static bool __should_discard_bucket(struct journal *j, struct journal_device *ja)
267 {
268 	unsigned min_free = max(4, ja->nr / 8);
269 
270 	return bch2_journal_dev_buckets_available(j, ja, journal_space_discarded) <
271 		min_free &&
272 		ja->discard_idx != ja->dirty_idx_ondisk;
273 }
274 
should_discard_bucket(struct journal * j,struct journal_device * ja)275 static bool should_discard_bucket(struct journal *j, struct journal_device *ja)
276 {
277 	spin_lock(&j->lock);
278 	bool ret = __should_discard_bucket(j, ja);
279 	spin_unlock(&j->lock);
280 
281 	return ret;
282 }
283 
284 /*
285  * Advance ja->discard_idx as long as it points to buckets that are no longer
286  * dirty, issuing discards if necessary:
287  */
bch2_journal_do_discards(struct journal * j)288 void bch2_journal_do_discards(struct journal *j)
289 {
290 	struct bch_fs *c = container_of(j, struct bch_fs, journal);
291 
292 	mutex_lock(&j->discard_lock);
293 
294 	for_each_rw_member(c, ca, BCH_DEV_WRITE_REF_journal_do_discards) {
295 		struct journal_device *ja = &ca->journal;
296 
297 		while (should_discard_bucket(j, ja)) {
298 			if (!c->opts.nochanges &&
299 			    bch2_discard_opt_enabled(c, ca) &&
300 			    bdev_max_discard_sectors(ca->disk_sb.bdev))
301 				blkdev_issue_discard(ca->disk_sb.bdev,
302 					bucket_to_sector(ca,
303 						ja->buckets[ja->discard_idx]),
304 					ca->mi.bucket_size, GFP_NOFS);
305 
306 			spin_lock(&j->lock);
307 			ja->discard_idx = (ja->discard_idx + 1) % ja->nr;
308 
309 			bch2_journal_space_available(j);
310 			spin_unlock(&j->lock);
311 		}
312 	}
313 
314 	mutex_unlock(&j->discard_lock);
315 }
316 
317 /*
318  * Journal entry pinning - machinery for holding a reference on a given journal
319  * entry, holding it open to ensure it gets replayed during recovery:
320  */
321 
bch2_journal_reclaim_fast(struct journal * j)322 void bch2_journal_reclaim_fast(struct journal *j)
323 {
324 	bool popped = false;
325 
326 	lockdep_assert_held(&j->lock);
327 
328 	/*
329 	 * Unpin journal entries whose reference counts reached zero, meaning
330 	 * all btree nodes got written out
331 	 */
332 	while (!fifo_empty(&j->pin) &&
333 	       j->pin.front <= j->seq_ondisk &&
334 	       !atomic_read(&fifo_peek_front(&j->pin).count)) {
335 		j->pin.front++;
336 		popped = true;
337 	}
338 
339 	if (popped) {
340 		bch2_journal_space_available(j);
341 		__closure_wake_up(&j->reclaim_flush_wait);
342 	}
343 }
344 
__bch2_journal_pin_put(struct journal * j,u64 seq)345 bool __bch2_journal_pin_put(struct journal *j, u64 seq)
346 {
347 	struct journal_entry_pin_list *pin_list = journal_seq_pin(j, seq);
348 
349 	return atomic_dec_and_test(&pin_list->count);
350 }
351 
bch2_journal_pin_put(struct journal * j,u64 seq)352 void bch2_journal_pin_put(struct journal *j, u64 seq)
353 {
354 	if (__bch2_journal_pin_put(j, seq)) {
355 		spin_lock(&j->lock);
356 		bch2_journal_reclaim_fast(j);
357 		spin_unlock(&j->lock);
358 	}
359 }
360 
__journal_pin_drop(struct journal * j,struct journal_entry_pin * pin)361 static inline bool __journal_pin_drop(struct journal *j,
362 				      struct journal_entry_pin *pin)
363 {
364 	struct journal_entry_pin_list *pin_list;
365 
366 	if (!journal_pin_active(pin))
367 		return false;
368 
369 	if (j->flush_in_progress == pin)
370 		j->flush_in_progress_dropped = true;
371 
372 	pin_list = journal_seq_pin(j, pin->seq);
373 	pin->seq = 0;
374 	list_del_init(&pin->list);
375 
376 	if (j->reclaim_flush_wait.list.first)
377 		__closure_wake_up(&j->reclaim_flush_wait);
378 
379 	/*
380 	 * Unpinning a journal entry may make journal_next_bucket() succeed, if
381 	 * writing a new last_seq will now make another bucket available:
382 	 */
383 	return atomic_dec_and_test(&pin_list->count) &&
384 		pin_list == &fifo_peek_front(&j->pin);
385 }
386 
bch2_journal_pin_drop(struct journal * j,struct journal_entry_pin * pin)387 void bch2_journal_pin_drop(struct journal *j,
388 			   struct journal_entry_pin *pin)
389 {
390 	spin_lock(&j->lock);
391 	if (__journal_pin_drop(j, pin))
392 		bch2_journal_reclaim_fast(j);
393 	spin_unlock(&j->lock);
394 }
395 
journal_pin_type(struct journal_entry_pin * pin,journal_pin_flush_fn fn)396 static enum journal_pin_type journal_pin_type(struct journal_entry_pin *pin,
397 					      journal_pin_flush_fn fn)
398 {
399 	if (fn == bch2_btree_node_flush0 ||
400 	    fn == bch2_btree_node_flush1) {
401 		unsigned idx = fn == bch2_btree_node_flush1;
402 		struct btree *b = container_of(pin, struct btree, writes[idx].journal);
403 
404 		return JOURNAL_PIN_TYPE_btree0 - b->c.level;
405 	} else if (fn == bch2_btree_key_cache_journal_flush)
406 		return JOURNAL_PIN_TYPE_key_cache;
407 	else
408 		return JOURNAL_PIN_TYPE_other;
409 }
410 
bch2_journal_pin_set_locked(struct journal * j,u64 seq,struct journal_entry_pin * pin,journal_pin_flush_fn flush_fn,enum journal_pin_type type)411 static inline void bch2_journal_pin_set_locked(struct journal *j, u64 seq,
412 			  struct journal_entry_pin *pin,
413 			  journal_pin_flush_fn flush_fn,
414 			  enum journal_pin_type type)
415 {
416 	struct journal_entry_pin_list *pin_list = journal_seq_pin(j, seq);
417 
418 	/*
419 	 * flush_fn is how we identify journal pins in debugfs, so must always
420 	 * exist, even if it doesn't do anything:
421 	 */
422 	BUG_ON(!flush_fn);
423 
424 	atomic_inc(&pin_list->count);
425 	pin->seq	= seq;
426 	pin->flush	= flush_fn;
427 
428 	if (list_empty(&pin_list->unflushed[type]) &&
429 	    j->reclaim_flush_wait.list.first)
430 		__closure_wake_up(&j->reclaim_flush_wait);
431 
432 	list_add(&pin->list, &pin_list->unflushed[type]);
433 }
434 
bch2_journal_pin_copy(struct journal * j,struct journal_entry_pin * dst,struct journal_entry_pin * src,journal_pin_flush_fn flush_fn)435 void bch2_journal_pin_copy(struct journal *j,
436 			   struct journal_entry_pin *dst,
437 			   struct journal_entry_pin *src,
438 			   journal_pin_flush_fn flush_fn)
439 {
440 	spin_lock(&j->lock);
441 
442 	u64 seq = READ_ONCE(src->seq);
443 
444 	if (seq < journal_last_seq(j)) {
445 		/*
446 		 * bch2_journal_pin_copy() raced with bch2_journal_pin_drop() on
447 		 * the src pin - with the pin dropped, the entry to pin might no
448 		 * longer to exist, but that means there's no longer anything to
449 		 * copy and we can bail out here:
450 		 */
451 		spin_unlock(&j->lock);
452 		return;
453 	}
454 
455 	bool reclaim = __journal_pin_drop(j, dst);
456 
457 	bch2_journal_pin_set_locked(j, seq, dst, flush_fn, journal_pin_type(dst, flush_fn));
458 
459 	if (reclaim)
460 		bch2_journal_reclaim_fast(j);
461 
462 	/*
463 	 * If the journal is currently full,  we might want to call flush_fn
464 	 * immediately:
465 	 */
466 	if (seq == journal_last_seq(j))
467 		journal_wake(j);
468 	spin_unlock(&j->lock);
469 }
470 
bch2_journal_pin_set(struct journal * j,u64 seq,struct journal_entry_pin * pin,journal_pin_flush_fn flush_fn)471 void bch2_journal_pin_set(struct journal *j, u64 seq,
472 			  struct journal_entry_pin *pin,
473 			  journal_pin_flush_fn flush_fn)
474 {
475 	spin_lock(&j->lock);
476 
477 	BUG_ON(seq < journal_last_seq(j));
478 
479 	bool reclaim = __journal_pin_drop(j, pin);
480 
481 	bch2_journal_pin_set_locked(j, seq, pin, flush_fn, journal_pin_type(pin, flush_fn));
482 
483 	if (reclaim)
484 		bch2_journal_reclaim_fast(j);
485 	/*
486 	 * If the journal is currently full,  we might want to call flush_fn
487 	 * immediately:
488 	 */
489 	if (seq == journal_last_seq(j))
490 		journal_wake(j);
491 
492 	spin_unlock(&j->lock);
493 }
494 
495 /**
496  * bch2_journal_pin_flush: ensure journal pin callback is no longer running
497  * @j:		journal object
498  * @pin:	pin to flush
499  */
bch2_journal_pin_flush(struct journal * j,struct journal_entry_pin * pin)500 void bch2_journal_pin_flush(struct journal *j, struct journal_entry_pin *pin)
501 {
502 	BUG_ON(journal_pin_active(pin));
503 
504 	wait_event(j->pin_flush_wait, j->flush_in_progress != pin);
505 }
506 
507 /*
508  * Journal reclaim: flush references to open journal entries to reclaim space in
509  * the journal
510  *
511  * May be done by the journal code in the background as needed to free up space
512  * for more journal entries, or as part of doing a clean shutdown, or to migrate
513  * data off of a specific device:
514  */
515 
516 static struct journal_entry_pin *
journal_get_next_pin(struct journal * j,u64 seq_to_flush,unsigned allowed_below_seq,unsigned allowed_above_seq,u64 * seq)517 journal_get_next_pin(struct journal *j,
518 		     u64 seq_to_flush,
519 		     unsigned allowed_below_seq,
520 		     unsigned allowed_above_seq,
521 		     u64 *seq)
522 {
523 	struct journal_entry_pin_list *pin_list;
524 	struct journal_entry_pin *ret = NULL;
525 
526 	fifo_for_each_entry_ptr(pin_list, &j->pin, *seq) {
527 		if (*seq > seq_to_flush && !allowed_above_seq)
528 			break;
529 
530 		for (unsigned i = 0; i < JOURNAL_PIN_TYPE_NR; i++)
531 			if (((BIT(i) & allowed_below_seq) && *seq <= seq_to_flush) ||
532 			    (BIT(i) & allowed_above_seq)) {
533 				ret = list_first_entry_or_null(&pin_list->unflushed[i],
534 					struct journal_entry_pin, list);
535 				if (ret)
536 					return ret;
537 			}
538 	}
539 
540 	return NULL;
541 }
542 
543 /* returns true if we did work */
journal_flush_pins(struct journal * j,u64 seq_to_flush,unsigned allowed_below_seq,unsigned allowed_above_seq,unsigned min_any,unsigned min_key_cache)544 static size_t journal_flush_pins(struct journal *j,
545 				 u64 seq_to_flush,
546 				 unsigned allowed_below_seq,
547 				 unsigned allowed_above_seq,
548 				 unsigned min_any,
549 				 unsigned min_key_cache)
550 {
551 	struct journal_entry_pin *pin;
552 	size_t nr_flushed = 0;
553 	journal_pin_flush_fn flush_fn;
554 	u64 seq;
555 	int err;
556 
557 	lockdep_assert_held(&j->reclaim_lock);
558 
559 	while (1) {
560 		unsigned allowed_above = allowed_above_seq;
561 		unsigned allowed_below = allowed_below_seq;
562 
563 		if (min_any) {
564 			allowed_above |= ~0;
565 			allowed_below |= ~0;
566 		}
567 
568 		if (min_key_cache) {
569 			allowed_above |= BIT(JOURNAL_PIN_TYPE_key_cache);
570 			allowed_below |= BIT(JOURNAL_PIN_TYPE_key_cache);
571 		}
572 
573 		cond_resched();
574 
575 		j->last_flushed = jiffies;
576 
577 		spin_lock(&j->lock);
578 		pin = journal_get_next_pin(j, seq_to_flush,
579 					   allowed_below,
580 					   allowed_above, &seq);
581 		if (pin) {
582 			BUG_ON(j->flush_in_progress);
583 			j->flush_in_progress = pin;
584 			j->flush_in_progress_dropped = false;
585 			flush_fn = pin->flush;
586 		}
587 		spin_unlock(&j->lock);
588 
589 		if (!pin)
590 			break;
591 
592 		if (min_key_cache && pin->flush == bch2_btree_key_cache_journal_flush)
593 			min_key_cache--;
594 
595 		if (min_any)
596 			min_any--;
597 
598 		err = flush_fn(j, pin, seq);
599 
600 		spin_lock(&j->lock);
601 		/* Pin might have been dropped or rearmed: */
602 		if (likely(!err && !j->flush_in_progress_dropped))
603 			list_move(&pin->list, &journal_seq_pin(j, seq)->flushed[journal_pin_type(pin, flush_fn)]);
604 		j->flush_in_progress = NULL;
605 		j->flush_in_progress_dropped = false;
606 		spin_unlock(&j->lock);
607 
608 		wake_up(&j->pin_flush_wait);
609 
610 		if (err)
611 			break;
612 
613 		nr_flushed++;
614 	}
615 
616 	return nr_flushed;
617 }
618 
journal_seq_to_flush(struct journal * j)619 static u64 journal_seq_to_flush(struct journal *j)
620 {
621 	struct bch_fs *c = container_of(j, struct bch_fs, journal);
622 	u64 seq_to_flush = 0;
623 
624 	guard(spinlock)(&j->lock);
625 	guard(rcu)();
626 
627 	for_each_rw_member_rcu(c, ca) {
628 		struct journal_device *ja = &ca->journal;
629 		unsigned nr_buckets, bucket_to_flush;
630 
631 		if (!ja->nr)
632 			continue;
633 
634 		/* Try to keep the journal at most half full: */
635 		nr_buckets = ja->nr / 2;
636 
637 		bucket_to_flush = (ja->cur_idx + nr_buckets) % ja->nr;
638 		seq_to_flush = max(seq_to_flush,
639 				   ja->bucket_seq[bucket_to_flush]);
640 	}
641 
642 	/* Also flush if the pin fifo is more than half full */
643 	return max_t(s64, seq_to_flush,
644 		     (s64) journal_cur_seq(j) -
645 		     (j->pin.size >> 1));
646 }
647 
648 /**
649  * __bch2_journal_reclaim - free up journal buckets
650  * @j:		journal object
651  * @direct:	direct or background reclaim?
652  * @kicked:	requested to run since we last ran?
653  *
654  * Background journal reclaim writes out btree nodes. It should be run
655  * early enough so that we never completely run out of journal buckets.
656  *
657  * High watermarks for triggering background reclaim:
658  * - FIFO has fewer than 512 entries left
659  * - fewer than 25% journal buckets free
660  *
661  * Background reclaim runs until low watermarks are reached:
662  * - FIFO has more than 1024 entries left
663  * - more than 50% journal buckets free
664  *
665  * As long as a reclaim can complete in the time it takes to fill up
666  * 512 journal entries or 25% of all journal buckets, then
667  * journal_next_bucket() should not stall.
668  */
__bch2_journal_reclaim(struct journal * j,bool direct,bool kicked)669 static int __bch2_journal_reclaim(struct journal *j, bool direct, bool kicked)
670 {
671 	struct bch_fs *c = container_of(j, struct bch_fs, journal);
672 	struct btree_cache *bc = &c->btree_cache;
673 	bool kthread = (current->flags & PF_KTHREAD) != 0;
674 	u64 seq_to_flush;
675 	size_t min_nr, min_key_cache, nr_flushed;
676 	unsigned flags;
677 	int ret = 0;
678 
679 	/*
680 	 * We can't invoke memory reclaim while holding the reclaim_lock -
681 	 * journal reclaim is required to make progress for memory reclaim
682 	 * (cleaning the caches), so we can't get stuck in memory reclaim while
683 	 * we're holding the reclaim lock:
684 	 */
685 	lockdep_assert_held(&j->reclaim_lock);
686 	flags = memalloc_noreclaim_save();
687 
688 	do {
689 		if (kthread && kthread_should_stop())
690 			break;
691 
692 		ret = bch2_journal_error(j);
693 		if (ret)
694 			break;
695 
696 		/* XXX shove journal discards off to another thread */
697 		bch2_journal_do_discards(j);
698 
699 		seq_to_flush = journal_seq_to_flush(j);
700 		min_nr = 0;
701 
702 		/*
703 		 * If it's been longer than j->reclaim_delay_ms since we last flushed,
704 		 * make sure to flush at least one journal pin:
705 		 */
706 		if (time_after(jiffies, j->last_flushed +
707 			       msecs_to_jiffies(c->opts.journal_reclaim_delay)))
708 			min_nr = 1;
709 
710 		if (j->watermark != BCH_WATERMARK_stripe)
711 			min_nr = 1;
712 
713 		size_t btree_cache_live = bc->live[0].nr + bc->live[1].nr;
714 		if (atomic_long_read(&bc->nr_dirty) * 2 > btree_cache_live)
715 			min_nr = 1;
716 
717 		min_key_cache = min(bch2_nr_btree_keys_need_flush(c), (size_t) 128);
718 
719 		trace_and_count(c, journal_reclaim_start, c,
720 				direct, kicked,
721 				min_nr, min_key_cache,
722 				atomic_long_read(&bc->nr_dirty), btree_cache_live,
723 				atomic_long_read(&c->btree_key_cache.nr_dirty),
724 				atomic_long_read(&c->btree_key_cache.nr_keys));
725 
726 		nr_flushed = journal_flush_pins(j, seq_to_flush,
727 						~0, 0,
728 						min_nr, min_key_cache);
729 
730 		if (direct)
731 			j->nr_direct_reclaim += nr_flushed;
732 		else
733 			j->nr_background_reclaim += nr_flushed;
734 		trace_and_count(c, journal_reclaim_finish, c, nr_flushed);
735 
736 		if (nr_flushed)
737 			wake_up(&j->reclaim_wait);
738 	} while ((min_nr || min_key_cache) && nr_flushed && !direct);
739 
740 	memalloc_noreclaim_restore(flags);
741 
742 	return ret;
743 }
744 
bch2_journal_reclaim(struct journal * j)745 int bch2_journal_reclaim(struct journal *j)
746 {
747 	return __bch2_journal_reclaim(j, true, true);
748 }
749 
bch2_journal_reclaim_thread(void * arg)750 static int bch2_journal_reclaim_thread(void *arg)
751 {
752 	struct journal *j = arg;
753 	struct bch_fs *c = container_of(j, struct bch_fs, journal);
754 	unsigned long delay, now;
755 	bool journal_empty;
756 	int ret = 0;
757 
758 	set_freezable();
759 
760 	j->last_flushed = jiffies;
761 
762 	while (!ret && !kthread_should_stop()) {
763 		bool kicked = j->reclaim_kicked;
764 
765 		j->reclaim_kicked = false;
766 
767 		mutex_lock(&j->reclaim_lock);
768 		ret = __bch2_journal_reclaim(j, false, kicked);
769 		mutex_unlock(&j->reclaim_lock);
770 
771 		now = jiffies;
772 		delay = msecs_to_jiffies(c->opts.journal_reclaim_delay);
773 		j->next_reclaim = j->last_flushed + delay;
774 
775 		if (!time_in_range(j->next_reclaim, now, now + delay))
776 			j->next_reclaim = now + delay;
777 
778 		while (1) {
779 			set_current_state(TASK_INTERRUPTIBLE|TASK_FREEZABLE);
780 			if (kthread_should_stop())
781 				break;
782 			if (j->reclaim_kicked)
783 				break;
784 
785 			spin_lock(&j->lock);
786 			journal_empty = fifo_empty(&j->pin);
787 			spin_unlock(&j->lock);
788 
789 			long timeout = j->next_reclaim - jiffies;
790 
791 			if (journal_empty)
792 				schedule();
793 			else if (timeout > 0)
794 				schedule_timeout(timeout);
795 			else
796 				break;
797 		}
798 		__set_current_state(TASK_RUNNING);
799 	}
800 
801 	return 0;
802 }
803 
bch2_journal_reclaim_stop(struct journal * j)804 void bch2_journal_reclaim_stop(struct journal *j)
805 {
806 	struct task_struct *p = j->reclaim_thread;
807 
808 	j->reclaim_thread = NULL;
809 
810 	if (p) {
811 		kthread_stop(p);
812 		put_task_struct(p);
813 	}
814 }
815 
bch2_journal_reclaim_start(struct journal * j)816 int bch2_journal_reclaim_start(struct journal *j)
817 {
818 	struct bch_fs *c = container_of(j, struct bch_fs, journal);
819 	struct task_struct *p;
820 	int ret;
821 
822 	if (j->reclaim_thread)
823 		return 0;
824 
825 	p = kthread_create(bch2_journal_reclaim_thread, j,
826 			   "bch-reclaim/%s", c->name);
827 	ret = PTR_ERR_OR_ZERO(p);
828 	bch_err_msg(c, ret, "creating journal reclaim thread");
829 	if (ret)
830 		return ret;
831 
832 	get_task_struct(p);
833 	j->reclaim_thread = p;
834 	wake_up_process(p);
835 	return 0;
836 }
837 
journal_pins_still_flushing(struct journal * j,u64 seq_to_flush,unsigned types)838 static bool journal_pins_still_flushing(struct journal *j, u64 seq_to_flush,
839 					unsigned types)
840 {
841 	struct journal_entry_pin_list *pin_list;
842 	u64 seq;
843 
844 	spin_lock(&j->lock);
845 	fifo_for_each_entry_ptr(pin_list, &j->pin, seq) {
846 		if (seq > seq_to_flush)
847 			break;
848 
849 		for (unsigned i = 0; i < JOURNAL_PIN_TYPE_NR; i++)
850 			if ((BIT(i) & types) &&
851 			    (!list_empty(&pin_list->unflushed[i]) ||
852 			     !list_empty(&pin_list->flushed[i]))) {
853 				spin_unlock(&j->lock);
854 				return true;
855 			}
856 	}
857 	spin_unlock(&j->lock);
858 
859 	return false;
860 }
861 
journal_flush_pins_or_still_flushing(struct journal * j,u64 seq_to_flush,unsigned types)862 static bool journal_flush_pins_or_still_flushing(struct journal *j, u64 seq_to_flush,
863 						 unsigned types)
864 {
865 	return  journal_flush_pins(j, seq_to_flush, types, 0, 0, 0) ||
866 		journal_pins_still_flushing(j, seq_to_flush, types);
867 }
868 
journal_flush_done(struct journal * j,u64 seq_to_flush,bool * did_work)869 static int journal_flush_done(struct journal *j, u64 seq_to_flush,
870 			      bool *did_work)
871 {
872 	int ret = 0;
873 
874 	ret = bch2_journal_error(j);
875 	if (ret)
876 		return ret;
877 
878 	mutex_lock(&j->reclaim_lock);
879 
880 	for (int type = JOURNAL_PIN_TYPE_NR - 1;
881 	     type >= 0;
882 	     --type)
883 		if (journal_flush_pins_or_still_flushing(j, seq_to_flush, BIT(type))) {
884 			*did_work = true;
885 			goto unlock;
886 		}
887 
888 	if (seq_to_flush > journal_cur_seq(j))
889 		bch2_journal_entry_close(j);
890 
891 	spin_lock(&j->lock);
892 	/*
893 	 * If journal replay hasn't completed, the unreplayed journal entries
894 	 * hold refs on their corresponding sequence numbers
895 	 */
896 	ret = !test_bit(JOURNAL_replay_done, &j->flags) ||
897 		journal_last_seq(j) > seq_to_flush ||
898 		!fifo_used(&j->pin);
899 
900 	spin_unlock(&j->lock);
901 unlock:
902 	mutex_unlock(&j->reclaim_lock);
903 
904 	return ret;
905 }
906 
bch2_journal_flush_pins(struct journal * j,u64 seq_to_flush)907 bool bch2_journal_flush_pins(struct journal *j, u64 seq_to_flush)
908 {
909 	/* time_stats this */
910 	bool did_work = false;
911 
912 	if (!test_bit(JOURNAL_running, &j->flags))
913 		return false;
914 
915 	closure_wait_event(&j->reclaim_flush_wait,
916 		journal_flush_done(j, seq_to_flush, &did_work));
917 
918 	return did_work;
919 }
920 
bch2_journal_flush_device_pins(struct journal * j,int dev_idx)921 int bch2_journal_flush_device_pins(struct journal *j, int dev_idx)
922 {
923 	struct bch_fs *c = container_of(j, struct bch_fs, journal);
924 	struct journal_entry_pin_list *p;
925 	u64 iter, seq = 0;
926 	int ret = 0;
927 
928 	spin_lock(&j->lock);
929 	fifo_for_each_entry_ptr(p, &j->pin, iter)
930 		if (dev_idx >= 0
931 		    ? bch2_dev_list_has_dev(p->devs, dev_idx)
932 		    : p->devs.nr < c->opts.metadata_replicas)
933 			seq = iter;
934 	spin_unlock(&j->lock);
935 
936 	bch2_journal_flush_pins(j, seq);
937 
938 	ret = bch2_journal_error(j);
939 	if (ret)
940 		return ret;
941 
942 	mutex_lock(&c->replicas_gc_lock);
943 	bch2_replicas_gc_start(c, 1 << BCH_DATA_journal);
944 
945 	/*
946 	 * Now that we've populated replicas_gc, write to the journal to mark
947 	 * active journal devices. This handles the case where the journal might
948 	 * be empty. Otherwise we could clear all journal replicas and
949 	 * temporarily put the fs into an unrecoverable state. Journal recovery
950 	 * expects to find devices marked for journal data on unclean mount.
951 	 */
952 	ret = bch2_journal_meta(&c->journal);
953 	if (ret)
954 		goto err;
955 
956 	seq = 0;
957 	spin_lock(&j->lock);
958 	while (!ret) {
959 		union bch_replicas_padded replicas;
960 
961 		seq = max(seq, journal_last_seq(j));
962 		if (seq >= j->pin.back)
963 			break;
964 		bch2_devlist_to_replicas(&replicas.e, BCH_DATA_journal,
965 					 journal_seq_pin(j, seq)->devs);
966 		seq++;
967 
968 		if (replicas.e.nr_devs) {
969 			spin_unlock(&j->lock);
970 			ret = bch2_mark_replicas(c, &replicas.e);
971 			spin_lock(&j->lock);
972 		}
973 	}
974 	spin_unlock(&j->lock);
975 err:
976 	ret = bch2_replicas_gc_end(c, ret);
977 	mutex_unlock(&c->replicas_gc_lock);
978 
979 	return ret;
980 }
981 
bch2_journal_seq_pins_to_text(struct printbuf * out,struct journal * j,u64 * seq)982 bool bch2_journal_seq_pins_to_text(struct printbuf *out, struct journal *j, u64 *seq)
983 {
984 	struct journal_entry_pin_list *pin_list;
985 	struct journal_entry_pin *pin;
986 
987 	spin_lock(&j->lock);
988 	if (!test_bit(JOURNAL_running, &j->flags)) {
989 		spin_unlock(&j->lock);
990 		return true;
991 	}
992 
993 	*seq = max(*seq, j->pin.front);
994 
995 	if (*seq >= j->pin.back) {
996 		spin_unlock(&j->lock);
997 		return true;
998 	}
999 
1000 	out->atomic++;
1001 
1002 	pin_list = journal_seq_pin(j, *seq);
1003 
1004 	prt_printf(out, "%llu: count %u\n", *seq, atomic_read(&pin_list->count));
1005 	printbuf_indent_add(out, 2);
1006 
1007 	prt_printf(out, "unflushed:\n");
1008 	for (unsigned i = 0; i < ARRAY_SIZE(pin_list->unflushed); i++)
1009 		list_for_each_entry(pin, &pin_list->unflushed[i], list)
1010 			prt_printf(out, "\t%px %ps\n", pin, pin->flush);
1011 
1012 	prt_printf(out, "flushed:\n");
1013 	for (unsigned i = 0; i < ARRAY_SIZE(pin_list->flushed); i++)
1014 		list_for_each_entry(pin, &pin_list->flushed[i], list)
1015 			prt_printf(out, "\t%px %ps\n", pin, pin->flush);
1016 
1017 	printbuf_indent_sub(out, 2);
1018 
1019 	--out->atomic;
1020 	spin_unlock(&j->lock);
1021 
1022 	return false;
1023 }
1024 
bch2_journal_pins_to_text(struct printbuf * out,struct journal * j)1025 void bch2_journal_pins_to_text(struct printbuf *out, struct journal *j)
1026 {
1027 	u64 seq = 0;
1028 
1029 	while (!bch2_journal_seq_pins_to_text(out, j, &seq))
1030 		seq++;
1031 }
1032