xref: /linux/fs/bcachefs/btree_journal_iter.c (revision 249ebf3f65f8530beb2cbfb91bff1d83ba88d23c)
1 // SPDX-License-Identifier: GPL-2.0
2 
3 #include "bcachefs.h"
4 #include "bkey_buf.h"
5 #include "bset.h"
6 #include "btree_cache.h"
7 #include "btree_journal_iter.h"
8 #include "journal_io.h"
9 
10 #include <linux/sort.h>
11 
12 /*
13  * For managing keys we read from the journal: until journal replay works normal
14  * btree lookups need to be able to find and return keys from the journal where
15  * they overwrite what's in the btree, so we have a special iterator and
16  * operations for the regular btree iter code to use:
17  */
18 
19 static int __journal_key_cmp(enum btree_id	l_btree_id,
20 			     unsigned		l_level,
21 			     struct bpos	l_pos,
22 			     const struct journal_key *r)
23 {
24 	return (cmp_int(l_btree_id,	r->btree_id) ?:
25 		cmp_int(l_level,	r->level) ?:
26 		bpos_cmp(l_pos,	r->k->k.p));
27 }
28 
29 static int journal_key_cmp(const struct journal_key *l, const struct journal_key *r)
30 {
31 	return __journal_key_cmp(l->btree_id, l->level, l->k->k.p, r);
32 }
33 
34 static inline size_t idx_to_pos(struct journal_keys *keys, size_t idx)
35 {
36 	size_t gap_size = keys->size - keys->nr;
37 
38 	if (idx >= keys->gap)
39 		idx += gap_size;
40 	return idx;
41 }
42 
43 static inline struct journal_key *idx_to_key(struct journal_keys *keys, size_t idx)
44 {
45 	return keys->data + idx_to_pos(keys, idx);
46 }
47 
48 static size_t __bch2_journal_key_search(struct journal_keys *keys,
49 					enum btree_id id, unsigned level,
50 					struct bpos pos)
51 {
52 	size_t l = 0, r = keys->nr, m;
53 
54 	while (l < r) {
55 		m = l + ((r - l) >> 1);
56 		if (__journal_key_cmp(id, level, pos, idx_to_key(keys, m)) > 0)
57 			l = m + 1;
58 		else
59 			r = m;
60 	}
61 
62 	BUG_ON(l < keys->nr &&
63 	       __journal_key_cmp(id, level, pos, idx_to_key(keys, l)) > 0);
64 
65 	BUG_ON(l &&
66 	       __journal_key_cmp(id, level, pos, idx_to_key(keys, l - 1)) <= 0);
67 
68 	return l;
69 }
70 
71 static size_t bch2_journal_key_search(struct journal_keys *keys,
72 				      enum btree_id id, unsigned level,
73 				      struct bpos pos)
74 {
75 	return idx_to_pos(keys, __bch2_journal_key_search(keys, id, level, pos));
76 }
77 
78 /* Returns first non-overwritten key >= search key: */
79 struct bkey_i *bch2_journal_keys_peek_upto(struct bch_fs *c, enum btree_id btree_id,
80 					   unsigned level, struct bpos pos,
81 					   struct bpos end_pos, size_t *idx)
82 {
83 	struct journal_keys *keys = &c->journal_keys;
84 	unsigned iters = 0;
85 	struct journal_key *k;
86 
87 	BUG_ON(*idx > keys->nr);
88 search:
89 	if (!*idx)
90 		*idx = __bch2_journal_key_search(keys, btree_id, level, pos);
91 
92 	while (*idx &&
93 	       __journal_key_cmp(btree_id, level, end_pos, idx_to_key(keys, *idx - 1)) <= 0) {
94 		--(*idx);
95 		iters++;
96 		if (iters == 10) {
97 			*idx = 0;
98 			goto search;
99 		}
100 	}
101 
102 	while ((k = *idx < keys->nr ? idx_to_key(keys, *idx) : NULL)) {
103 		if (__journal_key_cmp(btree_id, level, end_pos, k) < 0)
104 			return NULL;
105 
106 		if (k->overwritten) {
107 			(*idx)++;
108 			continue;
109 		}
110 
111 		if (__journal_key_cmp(btree_id, level, pos, k) <= 0)
112 			return k->k;
113 
114 		(*idx)++;
115 		iters++;
116 		if (iters == 10) {
117 			*idx = 0;
118 			goto search;
119 		}
120 	}
121 
122 	return NULL;
123 }
124 
125 struct bkey_i *bch2_journal_keys_peek_slot(struct bch_fs *c, enum btree_id btree_id,
126 					   unsigned level, struct bpos pos)
127 {
128 	size_t idx = 0;
129 
130 	return bch2_journal_keys_peek_upto(c, btree_id, level, pos, pos, &idx);
131 }
132 
133 static void journal_iter_verify(struct journal_iter *iter)
134 {
135 	struct journal_keys *keys = iter->keys;
136 	size_t gap_size = keys->size - keys->nr;
137 
138 	BUG_ON(iter->idx >= keys->gap &&
139 	       iter->idx <  keys->gap + gap_size);
140 
141 	if (iter->idx < keys->size) {
142 		struct journal_key *k = keys->data + iter->idx;
143 
144 		int cmp = cmp_int(k->btree_id,	iter->btree_id) ?:
145 			  cmp_int(k->level,	iter->level);
146 		BUG_ON(cmp < 0);
147 	}
148 }
149 
150 static void journal_iters_fix(struct bch_fs *c)
151 {
152 	struct journal_keys *keys = &c->journal_keys;
153 	/* The key we just inserted is immediately before the gap: */
154 	size_t gap_end = keys->gap + (keys->size - keys->nr);
155 	struct journal_key *new_key = &keys->data[keys->gap - 1];
156 	struct journal_iter *iter;
157 
158 	/*
159 	 * If an iterator points one after the key we just inserted, decrement
160 	 * the iterator so it points at the key we just inserted - if the
161 	 * decrement was unnecessary, bch2_btree_and_journal_iter_peek() will
162 	 * handle that:
163 	 */
164 	list_for_each_entry(iter, &c->journal_iters, list) {
165 		journal_iter_verify(iter);
166 		if (iter->idx		== gap_end &&
167 		    new_key->btree_id	== iter->btree_id &&
168 		    new_key->level	== iter->level)
169 			iter->idx = keys->gap - 1;
170 		journal_iter_verify(iter);
171 	}
172 }
173 
174 static void journal_iters_move_gap(struct bch_fs *c, size_t old_gap, size_t new_gap)
175 {
176 	struct journal_keys *keys = &c->journal_keys;
177 	struct journal_iter *iter;
178 	size_t gap_size = keys->size - keys->nr;
179 
180 	list_for_each_entry(iter, &c->journal_iters, list) {
181 		if (iter->idx > old_gap)
182 			iter->idx -= gap_size;
183 		if (iter->idx >= new_gap)
184 			iter->idx += gap_size;
185 	}
186 }
187 
188 int bch2_journal_key_insert_take(struct bch_fs *c, enum btree_id id,
189 				 unsigned level, struct bkey_i *k)
190 {
191 	struct journal_key n = {
192 		.btree_id	= id,
193 		.level		= level,
194 		.k		= k,
195 		.allocated	= true,
196 		/*
197 		 * Ensure these keys are done last by journal replay, to unblock
198 		 * journal reclaim:
199 		 */
200 		.journal_seq	= U32_MAX,
201 	};
202 	struct journal_keys *keys = &c->journal_keys;
203 	size_t idx = bch2_journal_key_search(keys, id, level, k->k.p);
204 
205 	BUG_ON(test_bit(BCH_FS_rw, &c->flags));
206 
207 	if (idx < keys->size &&
208 	    journal_key_cmp(&n, &keys->data[idx]) == 0) {
209 		if (keys->data[idx].allocated)
210 			kfree(keys->data[idx].k);
211 		keys->data[idx] = n;
212 		return 0;
213 	}
214 
215 	if (idx > keys->gap)
216 		idx -= keys->size - keys->nr;
217 
218 	size_t old_gap = keys->gap;
219 
220 	if (keys->nr == keys->size) {
221 		journal_iters_move_gap(c, old_gap, keys->size);
222 		old_gap = keys->size;
223 
224 		struct journal_keys new_keys = {
225 			.nr			= keys->nr,
226 			.size			= max_t(size_t, keys->size, 8) * 2,
227 		};
228 
229 		new_keys.data = kvmalloc_array(new_keys.size, sizeof(new_keys.data[0]), GFP_KERNEL);
230 		if (!new_keys.data) {
231 			bch_err(c, "%s: error allocating new key array (size %zu)",
232 				__func__, new_keys.size);
233 			return -BCH_ERR_ENOMEM_journal_key_insert;
234 		}
235 
236 		/* Since @keys was full, there was no gap: */
237 		memcpy(new_keys.data, keys->data, sizeof(keys->data[0]) * keys->nr);
238 		kvfree(keys->data);
239 		keys->data	= new_keys.data;
240 		keys->nr	= new_keys.nr;
241 		keys->size	= new_keys.size;
242 
243 		/* And now the gap is at the end: */
244 		keys->gap	= keys->nr;
245 	}
246 
247 	journal_iters_move_gap(c, old_gap, idx);
248 
249 	move_gap(keys, idx);
250 
251 	keys->nr++;
252 	keys->data[keys->gap++] = n;
253 
254 	journal_iters_fix(c);
255 
256 	return 0;
257 }
258 
259 /*
260  * Can only be used from the recovery thread while we're still RO - can't be
261  * used once we've got RW, as journal_keys is at that point used by multiple
262  * threads:
263  */
264 int bch2_journal_key_insert(struct bch_fs *c, enum btree_id id,
265 			    unsigned level, struct bkey_i *k)
266 {
267 	struct bkey_i *n;
268 	int ret;
269 
270 	n = kmalloc(bkey_bytes(&k->k), GFP_KERNEL);
271 	if (!n)
272 		return -BCH_ERR_ENOMEM_journal_key_insert;
273 
274 	bkey_copy(n, k);
275 	ret = bch2_journal_key_insert_take(c, id, level, n);
276 	if (ret)
277 		kfree(n);
278 	return ret;
279 }
280 
281 int bch2_journal_key_delete(struct bch_fs *c, enum btree_id id,
282 			    unsigned level, struct bpos pos)
283 {
284 	struct bkey_i whiteout;
285 
286 	bkey_init(&whiteout.k);
287 	whiteout.k.p = pos;
288 
289 	return bch2_journal_key_insert(c, id, level, &whiteout);
290 }
291 
292 bool bch2_key_deleted_in_journal(struct btree_trans *trans, enum btree_id btree,
293 				 unsigned level, struct bpos pos)
294 {
295 	struct journal_keys *keys = &trans->c->journal_keys;
296 	size_t idx = bch2_journal_key_search(keys, btree, level, pos);
297 
298 	if (!trans->journal_replay_not_finished)
299 		return false;
300 
301 	return (idx < keys->size &&
302 		keys->data[idx].btree_id	== btree &&
303 		keys->data[idx].level		== level &&
304 		bpos_eq(keys->data[idx].k->k.p, pos) &&
305 		bkey_deleted(&keys->data[idx].k->k));
306 }
307 
308 void bch2_journal_key_overwritten(struct bch_fs *c, enum btree_id btree,
309 				  unsigned level, struct bpos pos)
310 {
311 	struct journal_keys *keys = &c->journal_keys;
312 	size_t idx = bch2_journal_key_search(keys, btree, level, pos);
313 
314 	if (idx < keys->size &&
315 	    keys->data[idx].btree_id	== btree &&
316 	    keys->data[idx].level	== level &&
317 	    bpos_eq(keys->data[idx].k->k.p, pos))
318 		keys->data[idx].overwritten = true;
319 }
320 
321 static void bch2_journal_iter_advance(struct journal_iter *iter)
322 {
323 	if (iter->idx < iter->keys->size) {
324 		iter->idx++;
325 		if (iter->idx == iter->keys->gap)
326 			iter->idx += iter->keys->size - iter->keys->nr;
327 	}
328 }
329 
330 static struct bkey_s_c bch2_journal_iter_peek(struct journal_iter *iter)
331 {
332 	journal_iter_verify(iter);
333 
334 	while (iter->idx < iter->keys->size) {
335 		struct journal_key *k = iter->keys->data + iter->idx;
336 
337 		int cmp = cmp_int(k->btree_id,	iter->btree_id) ?:
338 			  cmp_int(k->level,	iter->level);
339 		if (cmp > 0)
340 			break;
341 		BUG_ON(cmp);
342 
343 		if (!k->overwritten)
344 			return bkey_i_to_s_c(k->k);
345 
346 		bch2_journal_iter_advance(iter);
347 	}
348 
349 	return bkey_s_c_null;
350 }
351 
352 static void bch2_journal_iter_exit(struct journal_iter *iter)
353 {
354 	list_del(&iter->list);
355 }
356 
357 static void bch2_journal_iter_init(struct bch_fs *c,
358 				   struct journal_iter *iter,
359 				   enum btree_id id, unsigned level,
360 				   struct bpos pos)
361 {
362 	iter->btree_id	= id;
363 	iter->level	= level;
364 	iter->keys	= &c->journal_keys;
365 	iter->idx	= bch2_journal_key_search(&c->journal_keys, id, level, pos);
366 
367 	journal_iter_verify(iter);
368 }
369 
370 static struct bkey_s_c bch2_journal_iter_peek_btree(struct btree_and_journal_iter *iter)
371 {
372 	return bch2_btree_node_iter_peek_unpack(&iter->node_iter,
373 						iter->b, &iter->unpacked);
374 }
375 
376 static void bch2_journal_iter_advance_btree(struct btree_and_journal_iter *iter)
377 {
378 	bch2_btree_node_iter_advance(&iter->node_iter, iter->b);
379 }
380 
381 void bch2_btree_and_journal_iter_advance(struct btree_and_journal_iter *iter)
382 {
383 	if (bpos_eq(iter->pos, SPOS_MAX))
384 		iter->at_end = true;
385 	else
386 		iter->pos = bpos_successor(iter->pos);
387 }
388 
389 static void btree_and_journal_iter_prefetch(struct btree_and_journal_iter *_iter)
390 {
391 	struct btree_and_journal_iter iter = *_iter;
392 	struct bch_fs *c = iter.trans->c;
393 	unsigned level = iter.journal.level;
394 	struct bkey_buf tmp;
395 	unsigned nr = test_bit(BCH_FS_started, &c->flags)
396 		? (level > 1 ? 0 :  2)
397 		: (level > 1 ? 1 : 16);
398 
399 	iter.prefetch = false;
400 	bch2_bkey_buf_init(&tmp);
401 
402 	while (nr--) {
403 		bch2_btree_and_journal_iter_advance(&iter);
404 		struct bkey_s_c k = bch2_btree_and_journal_iter_peek(&iter);
405 		if (!k.k)
406 			break;
407 
408 		bch2_bkey_buf_reassemble(&tmp, c, k);
409 		bch2_btree_node_prefetch(iter.trans, NULL, tmp.k, iter.journal.btree_id, level - 1);
410 	}
411 
412 	bch2_bkey_buf_exit(&tmp, c);
413 }
414 
415 struct bkey_s_c bch2_btree_and_journal_iter_peek(struct btree_and_journal_iter *iter)
416 {
417 	struct bkey_s_c btree_k, journal_k = bkey_s_c_null, ret;
418 
419 	if (iter->prefetch && iter->journal.level)
420 		btree_and_journal_iter_prefetch(iter);
421 again:
422 	if (iter->at_end)
423 		return bkey_s_c_null;
424 
425 	while ((btree_k = bch2_journal_iter_peek_btree(iter)).k &&
426 	       bpos_lt(btree_k.k->p, iter->pos))
427 		bch2_journal_iter_advance_btree(iter);
428 
429 	if (iter->trans->journal_replay_not_finished)
430 		while ((journal_k = bch2_journal_iter_peek(&iter->journal)).k &&
431 		       bpos_lt(journal_k.k->p, iter->pos))
432 			bch2_journal_iter_advance(&iter->journal);
433 
434 	ret = journal_k.k &&
435 		(!btree_k.k || bpos_le(journal_k.k->p, btree_k.k->p))
436 		? journal_k
437 		: btree_k;
438 
439 	if (ret.k && iter->b && bpos_gt(ret.k->p, iter->b->data->max_key))
440 		ret = bkey_s_c_null;
441 
442 	if (ret.k) {
443 		iter->pos = ret.k->p;
444 		if (bkey_deleted(ret.k)) {
445 			bch2_btree_and_journal_iter_advance(iter);
446 			goto again;
447 		}
448 	} else {
449 		iter->pos = SPOS_MAX;
450 		iter->at_end = true;
451 	}
452 
453 	return ret;
454 }
455 
456 void bch2_btree_and_journal_iter_exit(struct btree_and_journal_iter *iter)
457 {
458 	bch2_journal_iter_exit(&iter->journal);
459 }
460 
461 void __bch2_btree_and_journal_iter_init_node_iter(struct btree_trans *trans,
462 						  struct btree_and_journal_iter *iter,
463 						  struct btree *b,
464 						  struct btree_node_iter node_iter,
465 						  struct bpos pos)
466 {
467 	memset(iter, 0, sizeof(*iter));
468 
469 	iter->trans = trans;
470 	iter->b = b;
471 	iter->node_iter = node_iter;
472 	iter->pos = b->data->min_key;
473 	iter->at_end = false;
474 	INIT_LIST_HEAD(&iter->journal.list);
475 
476 	if (trans->journal_replay_not_finished) {
477 		bch2_journal_iter_init(trans->c, &iter->journal, b->c.btree_id, b->c.level, pos);
478 		if (!test_bit(BCH_FS_may_go_rw, &trans->c->flags))
479 			list_add(&iter->journal.list, &trans->c->journal_iters);
480 	}
481 }
482 
483 /*
484  * this version is used by btree_gc before filesystem has gone RW and
485  * multithreaded, so uses the journal_iters list:
486  */
487 void bch2_btree_and_journal_iter_init_node_iter(struct btree_trans *trans,
488 						struct btree_and_journal_iter *iter,
489 						struct btree *b)
490 {
491 	struct btree_node_iter node_iter;
492 
493 	bch2_btree_node_iter_init_from_start(&node_iter, b);
494 	__bch2_btree_and_journal_iter_init_node_iter(trans, iter, b, node_iter, b->data->min_key);
495 }
496 
497 /* sort and dedup all keys in the journal: */
498 
499 void bch2_journal_entries_free(struct bch_fs *c)
500 {
501 	struct journal_replay **i;
502 	struct genradix_iter iter;
503 
504 	genradix_for_each(&c->journal_entries, iter, i)
505 		kvfree(*i);
506 	genradix_free(&c->journal_entries);
507 }
508 
509 /*
510  * When keys compare equal, oldest compares first:
511  */
512 static int journal_sort_key_cmp(const void *_l, const void *_r)
513 {
514 	const struct journal_key *l = _l;
515 	const struct journal_key *r = _r;
516 
517 	return  journal_key_cmp(l, r) ?:
518 		cmp_int(l->journal_seq, r->journal_seq) ?:
519 		cmp_int(l->journal_offset, r->journal_offset);
520 }
521 
522 void bch2_journal_keys_put(struct bch_fs *c)
523 {
524 	struct journal_keys *keys = &c->journal_keys;
525 
526 	BUG_ON(atomic_read(&keys->ref) <= 0);
527 
528 	if (!atomic_dec_and_test(&keys->ref))
529 		return;
530 
531 	move_gap(keys, keys->nr);
532 
533 	darray_for_each(*keys, i)
534 		if (i->allocated)
535 			kfree(i->k);
536 
537 	kvfree(keys->data);
538 	keys->data = NULL;
539 	keys->nr = keys->gap = keys->size = 0;
540 
541 	bch2_journal_entries_free(c);
542 }
543 
544 static void __journal_keys_sort(struct journal_keys *keys)
545 {
546 	sort(keys->data, keys->nr, sizeof(keys->data[0]), journal_sort_key_cmp, NULL);
547 
548 	struct journal_key *dst = keys->data;
549 
550 	darray_for_each(*keys, src) {
551 		if (src + 1 < &darray_top(*keys) &&
552 		    !journal_key_cmp(src, src + 1))
553 			continue;
554 
555 		*dst++ = *src;
556 	}
557 
558 	keys->nr = dst - keys->data;
559 }
560 
561 int bch2_journal_keys_sort(struct bch_fs *c)
562 {
563 	struct genradix_iter iter;
564 	struct journal_replay *i, **_i;
565 	struct journal_keys *keys = &c->journal_keys;
566 	size_t nr_read = 0;
567 
568 	genradix_for_each(&c->journal_entries, iter, _i) {
569 		i = *_i;
570 
571 		if (journal_replay_ignore(i))
572 			continue;
573 
574 		cond_resched();
575 
576 		for_each_jset_key(k, entry, &i->j) {
577 			struct journal_key n = (struct journal_key) {
578 				.btree_id	= entry->btree_id,
579 				.level		= entry->level,
580 				.k		= k,
581 				.journal_seq	= le64_to_cpu(i->j.seq),
582 				.journal_offset	= k->_data - i->j._data,
583 			};
584 
585 			if (darray_push(keys, n)) {
586 				__journal_keys_sort(keys);
587 
588 				if (keys->nr * 8 > keys->size * 7) {
589 					bch_err(c, "Too many journal keys for slowpath; have %zu compacted, buf size %zu, processed %zu keys at seq %llu",
590 						keys->nr, keys->size, nr_read, le64_to_cpu(i->j.seq));
591 					return -BCH_ERR_ENOMEM_journal_keys_sort;
592 				}
593 
594 				BUG_ON(darray_push(keys, n));
595 			}
596 
597 			nr_read++;
598 		}
599 	}
600 
601 	__journal_keys_sort(keys);
602 	keys->gap = keys->nr;
603 
604 	bch_verbose(c, "Journal keys: %zu read, %zu after sorting and compacting", nr_read, keys->nr);
605 	return 0;
606 }
607 
608 void bch2_shoot_down_journal_keys(struct bch_fs *c, enum btree_id btree,
609 				  unsigned level_min, unsigned level_max,
610 				  struct bpos start, struct bpos end)
611 {
612 	struct journal_keys *keys = &c->journal_keys;
613 	size_t dst = 0;
614 
615 	move_gap(keys, keys->nr);
616 
617 	darray_for_each(*keys, i)
618 		if (!(i->btree_id == btree &&
619 		      i->level >= level_min &&
620 		      i->level <= level_max &&
621 		      bpos_ge(i->k->k.p, start) &&
622 		      bpos_le(i->k->k.p, end)))
623 			keys->data[dst++] = *i;
624 	keys->nr = keys->gap = dst;
625 }
626 
627 void bch2_journal_keys_dump(struct bch_fs *c)
628 {
629 	struct journal_keys *keys = &c->journal_keys;
630 	struct printbuf buf = PRINTBUF;
631 
632 	pr_info("%zu keys:", keys->nr);
633 
634 	move_gap(keys, keys->nr);
635 
636 	darray_for_each(*keys, i) {
637 		printbuf_reset(&buf);
638 		bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(i->k));
639 		pr_err("%s l=%u %s", bch2_btree_id_str(i->btree_id), i->level, buf.buf);
640 	}
641 	printbuf_exit(&buf);
642 }
643