xref: /linux/fs/bcachefs/btree_journal_iter.c (revision eb01fe7abbe2d0b38824d2a93fdb4cc3eaf2ccc1)
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_iters_fix(struct bch_fs *c)
134 {
135 	struct journal_keys *keys = &c->journal_keys;
136 	/* The key we just inserted is immediately before the gap: */
137 	size_t gap_end = keys->gap + (keys->size - keys->nr);
138 	struct btree_and_journal_iter *iter;
139 
140 	/*
141 	 * If an iterator points one after the key we just inserted, decrement
142 	 * the iterator so it points at the key we just inserted - if the
143 	 * decrement was unnecessary, bch2_btree_and_journal_iter_peek() will
144 	 * handle that:
145 	 */
146 	list_for_each_entry(iter, &c->journal_iters, journal.list)
147 		if (iter->journal.idx == gap_end)
148 			iter->journal.idx = keys->gap - 1;
149 }
150 
151 static void journal_iters_move_gap(struct bch_fs *c, size_t old_gap, size_t new_gap)
152 {
153 	struct journal_keys *keys = &c->journal_keys;
154 	struct journal_iter *iter;
155 	size_t gap_size = keys->size - keys->nr;
156 
157 	list_for_each_entry(iter, &c->journal_iters, list) {
158 		if (iter->idx > old_gap)
159 			iter->idx -= gap_size;
160 		if (iter->idx >= new_gap)
161 			iter->idx += gap_size;
162 	}
163 }
164 
165 int bch2_journal_key_insert_take(struct bch_fs *c, enum btree_id id,
166 				 unsigned level, struct bkey_i *k)
167 {
168 	struct journal_key n = {
169 		.btree_id	= id,
170 		.level		= level,
171 		.k		= k,
172 		.allocated	= true,
173 		/*
174 		 * Ensure these keys are done last by journal replay, to unblock
175 		 * journal reclaim:
176 		 */
177 		.journal_seq	= U32_MAX,
178 	};
179 	struct journal_keys *keys = &c->journal_keys;
180 	size_t idx = bch2_journal_key_search(keys, id, level, k->k.p);
181 
182 	BUG_ON(test_bit(BCH_FS_rw, &c->flags));
183 
184 	if (idx < keys->size &&
185 	    journal_key_cmp(&n, &keys->data[idx]) == 0) {
186 		if (keys->data[idx].allocated)
187 			kfree(keys->data[idx].k);
188 		keys->data[idx] = n;
189 		return 0;
190 	}
191 
192 	if (idx > keys->gap)
193 		idx -= keys->size - keys->nr;
194 
195 	if (keys->nr == keys->size) {
196 		struct journal_keys new_keys = {
197 			.nr			= keys->nr,
198 			.size			= max_t(size_t, keys->size, 8) * 2,
199 		};
200 
201 		new_keys.data = kvmalloc_array(new_keys.size, sizeof(new_keys.data[0]), GFP_KERNEL);
202 		if (!new_keys.data) {
203 			bch_err(c, "%s: error allocating new key array (size %zu)",
204 				__func__, new_keys.size);
205 			return -BCH_ERR_ENOMEM_journal_key_insert;
206 		}
207 
208 		/* Since @keys was full, there was no gap: */
209 		memcpy(new_keys.data, keys->data, sizeof(keys->data[0]) * keys->nr);
210 		kvfree(keys->data);
211 		keys->data	= new_keys.data;
212 		keys->nr	= new_keys.nr;
213 		keys->size	= new_keys.size;
214 
215 		/* And now the gap is at the end: */
216 		keys->gap	= keys->nr;
217 	}
218 
219 	journal_iters_move_gap(c, keys->gap, idx);
220 
221 	move_gap(keys, idx);
222 
223 	keys->nr++;
224 	keys->data[keys->gap++] = n;
225 
226 	journal_iters_fix(c);
227 
228 	return 0;
229 }
230 
231 /*
232  * Can only be used from the recovery thread while we're still RO - can't be
233  * used once we've got RW, as journal_keys is at that point used by multiple
234  * threads:
235  */
236 int bch2_journal_key_insert(struct bch_fs *c, enum btree_id id,
237 			    unsigned level, struct bkey_i *k)
238 {
239 	struct bkey_i *n;
240 	int ret;
241 
242 	n = kmalloc(bkey_bytes(&k->k), GFP_KERNEL);
243 	if (!n)
244 		return -BCH_ERR_ENOMEM_journal_key_insert;
245 
246 	bkey_copy(n, k);
247 	ret = bch2_journal_key_insert_take(c, id, level, n);
248 	if (ret)
249 		kfree(n);
250 	return ret;
251 }
252 
253 int bch2_journal_key_delete(struct bch_fs *c, enum btree_id id,
254 			    unsigned level, struct bpos pos)
255 {
256 	struct bkey_i whiteout;
257 
258 	bkey_init(&whiteout.k);
259 	whiteout.k.p = pos;
260 
261 	return bch2_journal_key_insert(c, id, level, &whiteout);
262 }
263 
264 void bch2_journal_key_overwritten(struct bch_fs *c, enum btree_id btree,
265 				  unsigned level, struct bpos pos)
266 {
267 	struct journal_keys *keys = &c->journal_keys;
268 	size_t idx = bch2_journal_key_search(keys, btree, level, pos);
269 
270 	if (idx < keys->size &&
271 	    keys->data[idx].btree_id	== btree &&
272 	    keys->data[idx].level	== level &&
273 	    bpos_eq(keys->data[idx].k->k.p, pos))
274 		keys->data[idx].overwritten = true;
275 }
276 
277 static void bch2_journal_iter_advance(struct journal_iter *iter)
278 {
279 	if (iter->idx < iter->keys->size) {
280 		iter->idx++;
281 		if (iter->idx == iter->keys->gap)
282 			iter->idx += iter->keys->size - iter->keys->nr;
283 	}
284 }
285 
286 static struct bkey_s_c bch2_journal_iter_peek(struct journal_iter *iter)
287 {
288 	struct journal_key *k = iter->keys->data + iter->idx;
289 
290 	while (k < iter->keys->data + iter->keys->size &&
291 	       k->btree_id	== iter->btree_id &&
292 	       k->level		== iter->level) {
293 		if (!k->overwritten)
294 			return bkey_i_to_s_c(k->k);
295 
296 		bch2_journal_iter_advance(iter);
297 		k = iter->keys->data + iter->idx;
298 	}
299 
300 	return bkey_s_c_null;
301 }
302 
303 static void bch2_journal_iter_exit(struct journal_iter *iter)
304 {
305 	list_del(&iter->list);
306 }
307 
308 static void bch2_journal_iter_init(struct bch_fs *c,
309 				   struct journal_iter *iter,
310 				   enum btree_id id, unsigned level,
311 				   struct bpos pos)
312 {
313 	iter->btree_id	= id;
314 	iter->level	= level;
315 	iter->keys	= &c->journal_keys;
316 	iter->idx	= bch2_journal_key_search(&c->journal_keys, id, level, pos);
317 }
318 
319 static struct bkey_s_c bch2_journal_iter_peek_btree(struct btree_and_journal_iter *iter)
320 {
321 	return bch2_btree_node_iter_peek_unpack(&iter->node_iter,
322 						iter->b, &iter->unpacked);
323 }
324 
325 static void bch2_journal_iter_advance_btree(struct btree_and_journal_iter *iter)
326 {
327 	bch2_btree_node_iter_advance(&iter->node_iter, iter->b);
328 }
329 
330 void bch2_btree_and_journal_iter_advance(struct btree_and_journal_iter *iter)
331 {
332 	if (bpos_eq(iter->pos, SPOS_MAX))
333 		iter->at_end = true;
334 	else
335 		iter->pos = bpos_successor(iter->pos);
336 }
337 
338 static void btree_and_journal_iter_prefetch(struct btree_and_journal_iter *_iter)
339 {
340 	struct btree_and_journal_iter iter = *_iter;
341 	struct bch_fs *c = iter.trans->c;
342 	unsigned level = iter.journal.level;
343 	struct bkey_buf tmp;
344 	unsigned nr = test_bit(BCH_FS_started, &c->flags)
345 		? (level > 1 ? 0 :  2)
346 		: (level > 1 ? 1 : 16);
347 
348 	iter.prefetch = false;
349 	bch2_bkey_buf_init(&tmp);
350 
351 	while (nr--) {
352 		bch2_btree_and_journal_iter_advance(&iter);
353 		struct bkey_s_c k = bch2_btree_and_journal_iter_peek(&iter);
354 		if (!k.k)
355 			break;
356 
357 		bch2_bkey_buf_reassemble(&tmp, c, k);
358 		bch2_btree_node_prefetch(iter.trans, NULL, tmp.k, iter.journal.btree_id, level - 1);
359 	}
360 
361 	bch2_bkey_buf_exit(&tmp, c);
362 }
363 
364 struct bkey_s_c bch2_btree_and_journal_iter_peek(struct btree_and_journal_iter *iter)
365 {
366 	struct bkey_s_c btree_k, journal_k, ret;
367 
368 	if (iter->prefetch && iter->journal.level)
369 		btree_and_journal_iter_prefetch(iter);
370 again:
371 	if (iter->at_end)
372 		return bkey_s_c_null;
373 
374 	while ((btree_k = bch2_journal_iter_peek_btree(iter)).k &&
375 	       bpos_lt(btree_k.k->p, iter->pos))
376 		bch2_journal_iter_advance_btree(iter);
377 
378 	while ((journal_k = bch2_journal_iter_peek(&iter->journal)).k &&
379 	       bpos_lt(journal_k.k->p, iter->pos))
380 		bch2_journal_iter_advance(&iter->journal);
381 
382 	ret = journal_k.k &&
383 		(!btree_k.k || bpos_le(journal_k.k->p, btree_k.k->p))
384 		? journal_k
385 		: btree_k;
386 
387 	if (ret.k && iter->b && bpos_gt(ret.k->p, iter->b->data->max_key))
388 		ret = bkey_s_c_null;
389 
390 	if (ret.k) {
391 		iter->pos = ret.k->p;
392 		if (bkey_deleted(ret.k)) {
393 			bch2_btree_and_journal_iter_advance(iter);
394 			goto again;
395 		}
396 	} else {
397 		iter->pos = SPOS_MAX;
398 		iter->at_end = true;
399 	}
400 
401 	return ret;
402 }
403 
404 void bch2_btree_and_journal_iter_exit(struct btree_and_journal_iter *iter)
405 {
406 	bch2_journal_iter_exit(&iter->journal);
407 }
408 
409 void __bch2_btree_and_journal_iter_init_node_iter(struct btree_trans *trans,
410 						  struct btree_and_journal_iter *iter,
411 						  struct btree *b,
412 						  struct btree_node_iter node_iter,
413 						  struct bpos pos)
414 {
415 	memset(iter, 0, sizeof(*iter));
416 
417 	iter->trans = trans;
418 	iter->b = b;
419 	iter->node_iter = node_iter;
420 	bch2_journal_iter_init(trans->c, &iter->journal, b->c.btree_id, b->c.level, pos);
421 	INIT_LIST_HEAD(&iter->journal.list);
422 	iter->pos = b->data->min_key;
423 	iter->at_end = false;
424 }
425 
426 /*
427  * this version is used by btree_gc before filesystem has gone RW and
428  * multithreaded, so uses the journal_iters list:
429  */
430 void bch2_btree_and_journal_iter_init_node_iter(struct btree_trans *trans,
431 						struct btree_and_journal_iter *iter,
432 						struct btree *b)
433 {
434 	struct btree_node_iter node_iter;
435 
436 	bch2_btree_node_iter_init_from_start(&node_iter, b);
437 	__bch2_btree_and_journal_iter_init_node_iter(trans, iter, b, node_iter, b->data->min_key);
438 	list_add(&iter->journal.list, &trans->c->journal_iters);
439 }
440 
441 /* sort and dedup all keys in the journal: */
442 
443 void bch2_journal_entries_free(struct bch_fs *c)
444 {
445 	struct journal_replay **i;
446 	struct genradix_iter iter;
447 
448 	genradix_for_each(&c->journal_entries, iter, i)
449 		kvfree(*i);
450 	genradix_free(&c->journal_entries);
451 }
452 
453 /*
454  * When keys compare equal, oldest compares first:
455  */
456 static int journal_sort_key_cmp(const void *_l, const void *_r)
457 {
458 	const struct journal_key *l = _l;
459 	const struct journal_key *r = _r;
460 
461 	return  journal_key_cmp(l, r) ?:
462 		cmp_int(l->journal_seq, r->journal_seq) ?:
463 		cmp_int(l->journal_offset, r->journal_offset);
464 }
465 
466 void bch2_journal_keys_put(struct bch_fs *c)
467 {
468 	struct journal_keys *keys = &c->journal_keys;
469 
470 	BUG_ON(atomic_read(&keys->ref) <= 0);
471 
472 	if (!atomic_dec_and_test(&keys->ref))
473 		return;
474 
475 	move_gap(keys, keys->nr);
476 
477 	darray_for_each(*keys, i)
478 		if (i->allocated)
479 			kfree(i->k);
480 
481 	kvfree(keys->data);
482 	keys->data = NULL;
483 	keys->nr = keys->gap = keys->size = 0;
484 
485 	bch2_journal_entries_free(c);
486 }
487 
488 static void __journal_keys_sort(struct journal_keys *keys)
489 {
490 	sort(keys->data, keys->nr, sizeof(keys->data[0]), journal_sort_key_cmp, NULL);
491 
492 	struct journal_key *dst = keys->data;
493 
494 	darray_for_each(*keys, src) {
495 		if (src + 1 < &darray_top(*keys) &&
496 		    !journal_key_cmp(src, src + 1))
497 			continue;
498 
499 		*dst++ = *src;
500 	}
501 
502 	keys->nr = dst - keys->data;
503 }
504 
505 int bch2_journal_keys_sort(struct bch_fs *c)
506 {
507 	struct genradix_iter iter;
508 	struct journal_replay *i, **_i;
509 	struct journal_keys *keys = &c->journal_keys;
510 	size_t nr_read = 0;
511 
512 	genradix_for_each(&c->journal_entries, iter, _i) {
513 		i = *_i;
514 
515 		if (journal_replay_ignore(i))
516 			continue;
517 
518 		cond_resched();
519 
520 		for_each_jset_key(k, entry, &i->j) {
521 			struct journal_key n = (struct journal_key) {
522 				.btree_id	= entry->btree_id,
523 				.level		= entry->level,
524 				.k		= k,
525 				.journal_seq	= le64_to_cpu(i->j.seq),
526 				.journal_offset	= k->_data - i->j._data,
527 			};
528 
529 			if (darray_push(keys, n)) {
530 				__journal_keys_sort(keys);
531 
532 				if (keys->nr * 8 > keys->size * 7) {
533 					bch_err(c, "Too many journal keys for slowpath; have %zu compacted, buf size %zu, processed %zu keys at seq %llu",
534 						keys->nr, keys->size, nr_read, le64_to_cpu(i->j.seq));
535 					return -BCH_ERR_ENOMEM_journal_keys_sort;
536 				}
537 
538 				BUG_ON(darray_push(keys, n));
539 			}
540 
541 			nr_read++;
542 		}
543 	}
544 
545 	__journal_keys_sort(keys);
546 	keys->gap = keys->nr;
547 
548 	bch_verbose(c, "Journal keys: %zu read, %zu after sorting and compacting", nr_read, keys->nr);
549 	return 0;
550 }
551