?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Linux 6.13-rc4
[linux.git] / fs / bcachefs / btree_iter.c
blobeef9b89c561ddfdd2ef4b886fde90fb012e8e45f
1 // SPDX-License-Identifier: GPL-2.0
2
3 #include "bcachefs.h"
4 #include "bkey_methods.h"
5 #include "bkey_buf.h"
6 #include "btree_cache.h"
7 #include "btree_iter.h"
8 #include "btree_journal_iter.h"
9 #include "btree_key_cache.h"
10 #include "btree_locking.h"
11 #include "btree_update.h"
12 #include "debug.h"
13 #include "error.h"
14 #include "extents.h"
15 #include "journal.h"
16 #include "journal_io.h"
17 #include "replicas.h"
18 #include "snapshot.h"
19 #include "trace.h"
20
21 #include <linux/random.h>
22 #include <linux/prefetch.h>
23
24 static inline void btree_path_list_remove(struct btree_trans *, struct btree_path *);
25 static inline void btree_path_list_add(struct btree_trans *,
26 btree_path_idx_t, btree_path_idx_t);
27
28 static inline unsigned long btree_iter_ip_allocated(struct btree_iter *iter)
29 {
30 #ifdef TRACK_PATH_ALLOCATED
31 return iter->ip_allocated;
32 #else
33 return 0;
34 #endif
35 }
36
37 static btree_path_idx_t btree_path_alloc(struct btree_trans *, btree_path_idx_t);
38 static void bch2_trans_srcu_lock(struct btree_trans *);
39
40 static inline int __btree_path_cmp(const struct btree_path *l,
41 enum btree_id r_btree_id,
42 bool r_cached,
43 struct bpos r_pos,
44 unsigned r_level)
45 {
46 /*
47 * Must match lock ordering as defined by __bch2_btree_node_lock:
48 */
49 return cmp_int(l->btree_id, r_btree_id) ?:
50 cmp_int((int) l->cached, (int) r_cached) ?:
51 bpos_cmp(l->pos, r_pos) ?:
52 -cmp_int(l->level, r_level);
53 }
54
55 static inline int btree_path_cmp(const struct btree_path *l,
56 const struct btree_path *r)
57 {
58 return __btree_path_cmp(l, r->btree_id, r->cached, r->pos, r->level);
59 }
60
61 static inline struct bpos bkey_successor(struct btree_iter *iter, struct bpos p)
62 {
63 /* Are we iterating over keys in all snapshots? */
64 if (iter->flags & BTREE_ITER_all_snapshots) {
65 p = bpos_successor(p);
66 } else {
67 p = bpos_nosnap_successor(p);
68 p.snapshot = iter->snapshot;
69 }
70
71 return p;
72 }
73
74 static inline struct bpos bkey_predecessor(struct btree_iter *iter, struct bpos p)
75 {
76 /* Are we iterating over keys in all snapshots? */
77 if (iter->flags & BTREE_ITER_all_snapshots) {
78 p = bpos_predecessor(p);
79 } else {
80 p = bpos_nosnap_predecessor(p);
81 p.snapshot = iter->snapshot;
82 }
83
84 return p;
85 }
86
87 static inline struct bpos btree_iter_search_key(struct btree_iter *iter)
88 {
89 struct bpos pos = iter->pos;
90
91 if ((iter->flags & BTREE_ITER_is_extents) &&
92 !bkey_eq(pos, POS_MAX))
93 pos = bkey_successor(iter, pos);
94 return pos;
95 }
96
97 static inline bool btree_path_pos_before_node(struct btree_path *path,
98 struct btree *b)
99 {
100 return bpos_lt(path->pos, b->data->min_key);
101 }
102
103 static inline bool btree_path_pos_after_node(struct btree_path *path,
104 struct btree *b)
105 {
106 return bpos_gt(path->pos, b->key.k.p);
107 }
108
109 static inline bool btree_path_pos_in_node(struct btree_path *path,
110 struct btree *b)
111 {
112 return path->btree_id == b->c.btree_id &&
113 !btree_path_pos_before_node(path, b) &&
114 !btree_path_pos_after_node(path, b);
115 }
116
117 /* Btree iterator: */
118
119 #ifdef CONFIG_BCACHEFS_DEBUG
120
121 static void bch2_btree_path_verify_cached(struct btree_trans *trans,
122 struct btree_path *path)
123 {
124 struct bkey_cached *ck;
125 bool locked = btree_node_locked(path, 0);
126
127 if (!bch2_btree_node_relock(trans, path, 0))
128 return;
129
130 ck = (void *) path->l[0].b;
131 BUG_ON(ck->key.btree_id != path->btree_id ||
132 !bkey_eq(ck->key.pos, path->pos));
133
134 if (!locked)
135 btree_node_unlock(trans, path, 0);
136 }
137
138 static void bch2_btree_path_verify_level(struct btree_trans *trans,
139 struct btree_path *path, unsigned level)
140 {
141 struct btree_path_level *l;
142 struct btree_node_iter tmp;
143 bool locked;
144 struct bkey_packed *p, *k;
145 struct printbuf buf1 = PRINTBUF;
146 struct printbuf buf2 = PRINTBUF;
147 struct printbuf buf3 = PRINTBUF;
148 const char *msg;
149
150 if (!bch2_debug_check_iterators)
151 return;
152
153 l = &path->l[level];
154 tmp = l->iter;
155 locked = btree_node_locked(path, level);
156
157 if (path->cached) {
158 if (!level)
159 bch2_btree_path_verify_cached(trans, path);
160 return;
161 }
162
163 if (!btree_path_node(path, level))
164 return;
165
166 if (!bch2_btree_node_relock_notrace(trans, path, level))
167 return;
168
169 BUG_ON(!btree_path_pos_in_node(path, l->b));
170
171 bch2_btree_node_iter_verify(&l->iter, l->b);
172
173 /*
174 * For interior nodes, the iterator will have skipped past deleted keys:
175 */
176 p = level
177 ? bch2_btree_node_iter_prev(&tmp, l->b)
178 : bch2_btree_node_iter_prev_all(&tmp, l->b);
179 k = bch2_btree_node_iter_peek_all(&l->iter, l->b);
180
181 if (p && bkey_iter_pos_cmp(l->b, p, &path->pos) >= 0) {
182 msg = "before";
183 goto err;
184 }
185
186 if (k && bkey_iter_pos_cmp(l->b, k, &path->pos) < 0) {
187 msg = "after";
188 goto err;
189 }
190
191 if (!locked)
192 btree_node_unlock(trans, path, level);
193 return;
194 err:
195 bch2_bpos_to_text(&buf1, path->pos);
196
197 if (p) {
198 struct bkey uk = bkey_unpack_key(l->b, p);
199
200 bch2_bkey_to_text(&buf2, &uk);
201 } else {
202 prt_printf(&buf2, "(none)");
203 }
204
205 if (k) {
206 struct bkey uk = bkey_unpack_key(l->b, k);
207
208 bch2_bkey_to_text(&buf3, &uk);
209 } else {
210 prt_printf(&buf3, "(none)");
211 }
212
213 panic("path should be %s key at level %u:\n"
214 "path pos %s\n"
215 "prev key %s\n"
216 "cur key %s\n",
217 msg, level, buf1.buf, buf2.buf, buf3.buf);
218 }
219
220 static void bch2_btree_path_verify(struct btree_trans *trans,
221 struct btree_path *path)
222 {
223 struct bch_fs *c = trans->c;
224
225 for (unsigned i = 0; i < (!path->cached ? BTREE_MAX_DEPTH : 1); i++) {
226 if (!path->l[i].b) {
227 BUG_ON(!path->cached &&
228 bch2_btree_id_root(c, path->btree_id)->b->c.level > i);
229 break;
230 }
231
232 bch2_btree_path_verify_level(trans, path, i);
233 }
234
235 bch2_btree_path_verify_locks(path);
236 }
237
238 void bch2_trans_verify_paths(struct btree_trans *trans)
239 {
240 struct btree_path *path;
241 unsigned iter;
242
243 trans_for_each_path(trans, path, iter)
244 bch2_btree_path_verify(trans, path);
245 }
246
247 static void bch2_btree_iter_verify(struct btree_iter *iter)
248 {
249 struct btree_trans *trans = iter->trans;
250
251 BUG_ON(!!(iter->flags & BTREE_ITER_cached) != btree_iter_path(trans, iter)->cached);
252
253 BUG_ON((iter->flags & BTREE_ITER_is_extents) &&
254 (iter->flags & BTREE_ITER_all_snapshots));
255
256 BUG_ON(!(iter->flags & BTREE_ITER_snapshot_field) &&
257 (iter->flags & BTREE_ITER_all_snapshots) &&
258 !btree_type_has_snapshot_field(iter->btree_id));
259
260 if (iter->update_path)
261 bch2_btree_path_verify(trans, &trans->paths[iter->update_path]);
262 bch2_btree_path_verify(trans, btree_iter_path(trans, iter));
263 }
264
265 static void bch2_btree_iter_verify_entry_exit(struct btree_iter *iter)
266 {
267 BUG_ON((iter->flags & BTREE_ITER_filter_snapshots) &&
268 !iter->pos.snapshot);
269
270 BUG_ON(!(iter->flags & BTREE_ITER_all_snapshots) &&
271 iter->pos.snapshot != iter->snapshot);
272
273 BUG_ON(bkey_lt(iter->pos, bkey_start_pos(&iter->k)) ||
274 bkey_gt(iter->pos, iter->k.p));
275 }
276
277 static int bch2_btree_iter_verify_ret(struct btree_iter *iter, struct bkey_s_c k)
278 {
279 struct btree_trans *trans = iter->trans;
280 struct btree_iter copy;
281 struct bkey_s_c prev;
282 int ret = 0;
283
284 if (!bch2_debug_check_iterators)
285 return 0;
286
287 if (!(iter->flags & BTREE_ITER_filter_snapshots))
288 return 0;
289
290 if (bkey_err(k) || !k.k)
291 return 0;
292
293 BUG_ON(!bch2_snapshot_is_ancestor(trans->c,
294 iter->snapshot,
295 k.k->p.snapshot));
296
297 bch2_trans_iter_init(trans, &copy, iter->btree_id, iter->pos,
298 BTREE_ITER_nopreserve|
299 BTREE_ITER_all_snapshots);
300 prev = bch2_btree_iter_prev(&copy);
301 if (!prev.k)
302 goto out;
303
304 ret = bkey_err(prev);
305 if (ret)
306 goto out;
307
308 if (bkey_eq(prev.k->p, k.k->p) &&
309 bch2_snapshot_is_ancestor(trans->c, iter->snapshot,
310 prev.k->p.snapshot) > 0) {
311 struct printbuf buf1 = PRINTBUF, buf2 = PRINTBUF;
312
313 bch2_bkey_to_text(&buf1, k.k);
314 bch2_bkey_to_text(&buf2, prev.k);
315
316 panic("iter snap %u\n"
317 "k %s\n"
318 "prev %s\n",
319 iter->snapshot,
320 buf1.buf, buf2.buf);
321 }
322 out:
323 bch2_trans_iter_exit(trans, &copy);
324 return ret;
325 }
326
327 void bch2_assert_pos_locked(struct btree_trans *trans, enum btree_id id,
328 struct bpos pos)
329 {
330 bch2_trans_verify_not_unlocked(trans);
331
332 struct btree_path *path;
333 struct trans_for_each_path_inorder_iter iter;
334 struct printbuf buf = PRINTBUF;
335
336 btree_trans_sort_paths(trans);
337
338 trans_for_each_path_inorder(trans, path, iter) {
339 if (path->btree_id != id ||
340 !btree_node_locked(path, 0) ||
341 !path->should_be_locked)
342 continue;
343
344 if (!path->cached) {
345 if (bkey_ge(pos, path->l[0].b->data->min_key) &&
346 bkey_le(pos, path->l[0].b->key.k.p))
347 return;
348 } else {
349 if (bkey_eq(pos, path->pos))
350 return;
351 }
352 }
353
354 bch2_dump_trans_paths_updates(trans);
355 bch2_bpos_to_text(&buf, pos);
356
357 panic("not locked: %s %s\n", bch2_btree_id_str(id), buf.buf);
358 }
359
360 #else
361
362 static inline void bch2_btree_path_verify_level(struct btree_trans *trans,
363 struct btree_path *path, unsigned l) {}
364 static inline void bch2_btree_path_verify(struct btree_trans *trans,
365 struct btree_path *path) {}
366 static inline void bch2_btree_iter_verify(struct btree_iter *iter) {}
367 static inline void bch2_btree_iter_verify_entry_exit(struct btree_iter *iter) {}
368 static inline int bch2_btree_iter_verify_ret(struct btree_iter *iter, struct bkey_s_c k) { return 0; }
369
370 #endif
371
372 /* Btree path: fixups after btree updates */
373
374 static void btree_node_iter_set_set_pos(struct btree_node_iter *iter,
375 struct btree *b,
376 struct bset_tree *t,
377 struct bkey_packed *k)
378 {
379 struct btree_node_iter_set *set;
380
381 btree_node_iter_for_each(iter, set)
382 if (set->end == t->end_offset) {
383 set->k = __btree_node_key_to_offset(b, k);
384 bch2_btree_node_iter_sort(iter, b);
385 return;
386 }
387
388 bch2_btree_node_iter_push(iter, b, k, btree_bkey_last(b, t));
389 }
390
391 static void __bch2_btree_path_fix_key_modified(struct btree_path *path,
392 struct btree *b,
393 struct bkey_packed *where)
394 {
395 struct btree_path_level *l = &path->l[b->c.level];
396
397 if (where != bch2_btree_node_iter_peek_all(&l->iter, l->b))
398 return;
399
400 if (bkey_iter_pos_cmp(l->b, where, &path->pos) < 0)
401 bch2_btree_node_iter_advance(&l->iter, l->b);
402 }
403
404 void bch2_btree_path_fix_key_modified(struct btree_trans *trans,
405 struct btree *b,
406 struct bkey_packed *where)
407 {
408 struct btree_path *path;
409 unsigned i;
410
411 trans_for_each_path_with_node(trans, b, path, i) {
412 __bch2_btree_path_fix_key_modified(path, b, where);
413 bch2_btree_path_verify_level(trans, path, b->c.level);
414 }
415 }
416
417 static void __bch2_btree_node_iter_fix(struct btree_path *path,
418 struct btree *b,
419 struct btree_node_iter *node_iter,
420 struct bset_tree *t,
421 struct bkey_packed *where,
422 unsigned clobber_u64s,
423 unsigned new_u64s)
424 {
425 const struct bkey_packed *end = btree_bkey_last(b, t);
426 struct btree_node_iter_set *set;
427 unsigned offset = __btree_node_key_to_offset(b, where);
428 int shift = new_u64s - clobber_u64s;
429 unsigned old_end = t->end_offset - shift;
430 unsigned orig_iter_pos = node_iter->data[0].k;
431 bool iter_current_key_modified =
432 orig_iter_pos >= offset &&
433 orig_iter_pos <= offset + clobber_u64s;
434
435 btree_node_iter_for_each(node_iter, set)
436 if (set->end == old_end)
437 goto found;
438
439 /* didn't find the bset in the iterator - might have to readd it: */
440 if (new_u64s &&
441 bkey_iter_pos_cmp(b, where, &path->pos) >= 0) {
442 bch2_btree_node_iter_push(node_iter, b, where, end);
443 goto fixup_done;
444 } else {
445 /* Iterator is after key that changed */
446 return;
447 }
448 found:
449 set->end = t->end_offset;
450
451 /* Iterator hasn't gotten to the key that changed yet: */
452 if (set->k < offset)
453 return;
454
455 if (new_u64s &&
456 bkey_iter_pos_cmp(b, where, &path->pos) >= 0) {
457 set->k = offset;
458 } else if (set->k < offset + clobber_u64s) {
459 set->k = offset + new_u64s;
460 if (set->k == set->end)
461 bch2_btree_node_iter_set_drop(node_iter, set);
462 } else {
463 /* Iterator is after key that changed */
464 set->k = (int) set->k + shift;
465 return;
466 }
467
468 bch2_btree_node_iter_sort(node_iter, b);
469 fixup_done:
470 if (node_iter->data[0].k != orig_iter_pos)
471 iter_current_key_modified = true;
472
473 /*
474 * When a new key is added, and the node iterator now points to that
475 * key, the iterator might have skipped past deleted keys that should
476 * come after the key the iterator now points to. We have to rewind to
477 * before those deleted keys - otherwise
478 * bch2_btree_node_iter_prev_all() breaks:
479 */
480 if (!bch2_btree_node_iter_end(node_iter) &&
481 iter_current_key_modified &&
482 b->c.level) {
483 struct bkey_packed *k, *k2, *p;
484
485 k = bch2_btree_node_iter_peek_all(node_iter, b);
486
487 for_each_bset(b, t) {
488 bool set_pos = false;
489
490 if (node_iter->data[0].end == t->end_offset)
491 continue;
492
493 k2 = bch2_btree_node_iter_bset_pos(node_iter, b, t);
494
495 while ((p = bch2_bkey_prev_all(b, t, k2)) &&
496 bkey_iter_cmp(b, k, p) < 0) {
497 k2 = p;
498 set_pos = true;
499 }
500
501 if (set_pos)
502 btree_node_iter_set_set_pos(node_iter,
503 b, t, k2);
504 }
505 }
506 }
507
508 void bch2_btree_node_iter_fix(struct btree_trans *trans,
509 struct btree_path *path,
510 struct btree *b,
511 struct btree_node_iter *node_iter,
512 struct bkey_packed *where,
513 unsigned clobber_u64s,
514 unsigned new_u64s)
515 {
516 struct bset_tree *t = bch2_bkey_to_bset_inlined(b, where);
517 struct btree_path *linked;
518 unsigned i;
519
520 if (node_iter != &path->l[b->c.level].iter) {
521 __bch2_btree_node_iter_fix(path, b, node_iter, t,
522 where, clobber_u64s, new_u64s);
523
524 if (bch2_debug_check_iterators)
525 bch2_btree_node_iter_verify(node_iter, b);
526 }
527
528 trans_for_each_path_with_node(trans, b, linked, i) {
529 __bch2_btree_node_iter_fix(linked, b,
530 &linked->l[b->c.level].iter, t,
531 where, clobber_u64s, new_u64s);
532 bch2_btree_path_verify_level(trans, linked, b->c.level);
533 }
534 }
535
536 /* Btree path level: pointer to a particular btree node and node iter */
537
538 static inline struct bkey_s_c __btree_iter_unpack(struct bch_fs *c,
539 struct btree_path_level *l,
540 struct bkey *u,
541 struct bkey_packed *k)
542 {
543 if (unlikely(!k)) {
544 /*
545 * signal to bch2_btree_iter_peek_slot() that we're currently at
546 * a hole
547 */
548 u->type = KEY_TYPE_deleted;
549 return bkey_s_c_null;
550 }
551
552 return bkey_disassemble(l->b, k, u);
553 }
554
555 static inline struct bkey_s_c btree_path_level_peek_all(struct bch_fs *c,
556 struct btree_path_level *l,
557 struct bkey *u)
558 {
559 return __btree_iter_unpack(c, l, u,
560 bch2_btree_node_iter_peek_all(&l->iter, l->b));
561 }
562
563 static inline struct bkey_s_c btree_path_level_peek(struct btree_trans *trans,
564 struct btree_path *path,
565 struct btree_path_level *l,
566 struct bkey *u)
567 {
568 struct bkey_s_c k = __btree_iter_unpack(trans->c, l, u,
569 bch2_btree_node_iter_peek(&l->iter, l->b));
570
571 path->pos = k.k ? k.k->p : l->b->key.k.p;
572 trans->paths_sorted = false;
573 bch2_btree_path_verify_level(trans, path, l - path->l);
574 return k;
575 }
576
577 static inline struct bkey_s_c btree_path_level_prev(struct btree_trans *trans,
578 struct btree_path *path,
579 struct btree_path_level *l,
580 struct bkey *u)
581 {
582 struct bkey_s_c k = __btree_iter_unpack(trans->c, l, u,
583 bch2_btree_node_iter_prev(&l->iter, l->b));
584
585 path->pos = k.k ? k.k->p : l->b->data->min_key;
586 trans->paths_sorted = false;
587 bch2_btree_path_verify_level(trans, path, l - path->l);
588 return k;
589 }
590
591 static inline bool btree_path_advance_to_pos(struct btree_path *path,
592 struct btree_path_level *l,
593 int max_advance)
594 {
595 struct bkey_packed *k;
596 int nr_advanced = 0;
597
598 while ((k = bch2_btree_node_iter_peek_all(&l->iter, l->b)) &&
599 bkey_iter_pos_cmp(l->b, k, &path->pos) < 0) {
600 if (max_advance > 0 && nr_advanced >= max_advance)
601 return false;
602
603 bch2_btree_node_iter_advance(&l->iter, l->b);
604 nr_advanced++;
605 }
606
607 return true;
608 }
609
610 static inline void __btree_path_level_init(struct btree_path *path,
611 unsigned level)
612 {
613 struct btree_path_level *l = &path->l[level];
614
615 bch2_btree_node_iter_init(&l->iter, l->b, &path->pos);
616
617 /*
618 * Iterators to interior nodes should always be pointed at the first non
619 * whiteout:
620 */
621 if (level)
622 bch2_btree_node_iter_peek(&l->iter, l->b);
623 }
624
625 void bch2_btree_path_level_init(struct btree_trans *trans,
626 struct btree_path *path,
627 struct btree *b)
628 {
629 BUG_ON(path->cached);
630
631 EBUG_ON(!btree_path_pos_in_node(path, b));
632
633 path->l[b->c.level].lock_seq = six_lock_seq(&b->c.lock);
634 path->l[b->c.level].b = b;
635 __btree_path_level_init(path, b->c.level);
636 }
637
638 /* Btree path: fixups after btree node updates: */
639
640 static void bch2_trans_revalidate_updates_in_node(struct btree_trans *trans, struct btree *b)
641 {
642 struct bch_fs *c = trans->c;
643
644 trans_for_each_update(trans, i)
645 if (!i->cached &&
646 i->level == b->c.level &&
647 i->btree_id == b->c.btree_id &&
648 bpos_cmp(i->k->k.p, b->data->min_key) >= 0 &&
649 bpos_cmp(i->k->k.p, b->data->max_key) <= 0) {
650 i->old_v = bch2_btree_path_peek_slot(trans->paths + i->path, &i->old_k).v;
651
652 if (unlikely(trans->journal_replay_not_finished)) {
653 struct bkey_i *j_k =
654 bch2_journal_keys_peek_slot(c, i->btree_id, i->level,
655 i->k->k.p);
656
657 if (j_k) {
658 i->old_k = j_k->k;
659 i->old_v = &j_k->v;
660 }
661 }
662 }
663 }
664
665 /*
666 * A btree node is being replaced - update the iterator to point to the new
667 * node:
668 */
669 void bch2_trans_node_add(struct btree_trans *trans,
670 struct btree_path *path,
671 struct btree *b)
672 {
673 struct btree_path *prev;
674
675 BUG_ON(!btree_path_pos_in_node(path, b));
676
677 while ((prev = prev_btree_path(trans, path)) &&
678 btree_path_pos_in_node(prev, b))
679 path = prev;
680
681 for (;
682 path && btree_path_pos_in_node(path, b);
683 path = next_btree_path(trans, path))
684 if (path->uptodate == BTREE_ITER_UPTODATE && !path->cached) {
685 enum btree_node_locked_type t =
686 btree_lock_want(path, b->c.level);
687
688 if (t != BTREE_NODE_UNLOCKED) {
689 btree_node_unlock(trans, path, b->c.level);
690 six_lock_increment(&b->c.lock, (enum six_lock_type) t);
691 mark_btree_node_locked(trans, path, b->c.level, t);
692 }
693
694 bch2_btree_path_level_init(trans, path, b);
695 }
696
697 bch2_trans_revalidate_updates_in_node(trans, b);
698 }
699
700 /*
701 * A btree node has been modified in such a way as to invalidate iterators - fix
702 * them:
703 */
704 void bch2_trans_node_reinit_iter(struct btree_trans *trans, struct btree *b)
705 {
706 struct btree_path *path;
707 unsigned i;
708
709 trans_for_each_path_with_node(trans, b, path, i)
710 __btree_path_level_init(path, b->c.level);
711
712 bch2_trans_revalidate_updates_in_node(trans, b);
713 }
714
715 /* Btree path: traverse, set_pos: */
716
717 static inline int btree_path_lock_root(struct btree_trans *trans,
718 struct btree_path *path,
719 unsigned depth_want,
720 unsigned long trace_ip)
721 {
722 struct bch_fs *c = trans->c;
723 struct btree *b, **rootp = &bch2_btree_id_root(c, path->btree_id)->b;
724 enum six_lock_type lock_type;
725 unsigned i;
726 int ret;
727
728 EBUG_ON(path->nodes_locked);
729
730 while (1) {
731 b = READ_ONCE(*rootp);
732 path->level = READ_ONCE(b->c.level);
733
734 if (unlikely(path->level < depth_want)) {
735 /*
736 * the root is at a lower depth than the depth we want:
737 * got to the end of the btree, or we're walking nodes
738 * greater than some depth and there are no nodes >=
739 * that depth
740 */
741 path->level = depth_want;
742 for (i = path->level; i < BTREE_MAX_DEPTH; i++)
743 path->l[i].b = NULL;
744 return 1;
745 }
746
747 lock_type = __btree_lock_want(path, path->level);
748 ret = btree_node_lock(trans, path, &b->c,
749 path->level, lock_type, trace_ip);
750 if (unlikely(ret)) {
751 if (bch2_err_matches(ret, BCH_ERR_lock_fail_root_changed))
752 continue;
753 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
754 return ret;
755 BUG();
756 }
757
758 if (likely(b == READ_ONCE(*rootp) &&
759 b->c.level == path->level &&
760 !race_fault())) {
761 for (i = 0; i < path->level; i++)
762 path->l[i].b = ERR_PTR(-BCH_ERR_no_btree_node_lock_root);
763 path->l[path->level].b = b;
764 for (i = path->level + 1; i < BTREE_MAX_DEPTH; i++)
765 path->l[i].b = NULL;
766
767 mark_btree_node_locked(trans, path, path->level,
768 (enum btree_node_locked_type) lock_type);
769 bch2_btree_path_level_init(trans, path, b);
770 return 0;
771 }
772
773 six_unlock_type(&b->c.lock, lock_type);
774 }
775 }
776
777 noinline
778 static int btree_path_prefetch(struct btree_trans *trans, struct btree_path *path)
779 {
780 struct bch_fs *c = trans->c;
781 struct btree_path_level *l = path_l(path);
782 struct btree_node_iter node_iter = l->iter;
783 struct bkey_packed *k;
784 struct bkey_buf tmp;
785 unsigned nr = test_bit(BCH_FS_started, &c->flags)
786 ? (path->level > 1 ? 0 : 2)
787 : (path->level > 1 ? 1 : 16);
788 bool was_locked = btree_node_locked(path, path->level);
789 int ret = 0;
790
791 bch2_bkey_buf_init(&tmp);
792
793 while (nr-- && !ret) {
794 if (!bch2_btree_node_relock(trans, path, path->level))
795 break;
796
797 bch2_btree_node_iter_advance(&node_iter, l->b);
798 k = bch2_btree_node_iter_peek(&node_iter, l->b);
799 if (!k)
800 break;
801
802 bch2_bkey_buf_unpack(&tmp, c, l->b, k);
803 ret = bch2_btree_node_prefetch(trans, path, tmp.k, path->btree_id,
804 path->level - 1);
805 }
806
807 if (!was_locked)
808 btree_node_unlock(trans, path, path->level);
809
810 bch2_bkey_buf_exit(&tmp, c);
811 return ret;
812 }
813
814 static int btree_path_prefetch_j(struct btree_trans *trans, struct btree_path *path,
815 struct btree_and_journal_iter *jiter)
816 {
817 struct bch_fs *c = trans->c;
818 struct bkey_s_c k;
819 struct bkey_buf tmp;
820 unsigned nr = test_bit(BCH_FS_started, &c->flags)
821 ? (path->level > 1 ? 0 : 2)
822 : (path->level > 1 ? 1 : 16);
823 bool was_locked = btree_node_locked(path, path->level);
824 int ret = 0;
825
826 bch2_bkey_buf_init(&tmp);
827
828 while (nr-- && !ret) {
829 if (!bch2_btree_node_relock(trans, path, path->level))
830 break;
831
832 bch2_btree_and_journal_iter_advance(jiter);
833 k = bch2_btree_and_journal_iter_peek(jiter);
834 if (!k.k)
835 break;
836
837 bch2_bkey_buf_reassemble(&tmp, c, k);
838 ret = bch2_btree_node_prefetch(trans, path, tmp.k, path->btree_id,
839 path->level - 1);
840 }
841
842 if (!was_locked)
843 btree_node_unlock(trans, path, path->level);
844
845 bch2_bkey_buf_exit(&tmp, c);
846 return ret;
847 }
848
849 static noinline void btree_node_mem_ptr_set(struct btree_trans *trans,
850 struct btree_path *path,
851 unsigned plevel, struct btree *b)
852 {
853 struct btree_path_level *l = &path->l[plevel];
854 bool locked = btree_node_locked(path, plevel);
855 struct bkey_packed *k;
856 struct bch_btree_ptr_v2 *bp;
857
858 if (!bch2_btree_node_relock(trans, path, plevel))
859 return;
860
861 k = bch2_btree_node_iter_peek_all(&l->iter, l->b);
862 BUG_ON(k->type != KEY_TYPE_btree_ptr_v2);
863
864 bp = (void *) bkeyp_val(&l->b->format, k);
865 bp->mem_ptr = (unsigned long)b;
866
867 if (!locked)
868 btree_node_unlock(trans, path, plevel);
869 }
870
871 static noinline int btree_node_iter_and_journal_peek(struct btree_trans *trans,
872 struct btree_path *path,
873 unsigned flags,
874 struct bkey_buf *out)
875 {
876 struct bch_fs *c = trans->c;
877 struct btree_path_level *l = path_l(path);
878 struct btree_and_journal_iter jiter;
879 struct bkey_s_c k;
880 int ret = 0;
881
882 __bch2_btree_and_journal_iter_init_node_iter(trans, &jiter, l->b, l->iter, path->pos);
883
884 k = bch2_btree_and_journal_iter_peek(&jiter);
885 if (!k.k) {
886 struct printbuf buf = PRINTBUF;
887
888 prt_str(&buf, "node not found at pos ");
889 bch2_bpos_to_text(&buf, path->pos);
890 prt_str(&buf, " at btree ");
891 bch2_btree_pos_to_text(&buf, c, l->b);
892
893 ret = bch2_fs_topology_error(c, "%s", buf.buf);
894 printbuf_exit(&buf);
895 goto err;
896 }
897
898 bch2_bkey_buf_reassemble(out, c, k);
899
900 if ((flags & BTREE_ITER_prefetch) &&
901 c->opts.btree_node_prefetch)
902 ret = btree_path_prefetch_j(trans, path, &jiter);
903
904 err:
905 bch2_btree_and_journal_iter_exit(&jiter);
906 return ret;
907 }
908
909 static __always_inline int btree_path_down(struct btree_trans *trans,
910 struct btree_path *path,
911 unsigned flags,
912 unsigned long trace_ip)
913 {
914 struct bch_fs *c = trans->c;
915 struct btree_path_level *l = path_l(path);
916 struct btree *b;
917 unsigned level = path->level - 1;
918 enum six_lock_type lock_type = __btree_lock_want(path, level);
919 struct bkey_buf tmp;
920 int ret;
921
922 EBUG_ON(!btree_node_locked(path, path->level));
923
924 bch2_bkey_buf_init(&tmp);
925
926 if (unlikely(trans->journal_replay_not_finished)) {
927 ret = btree_node_iter_and_journal_peek(trans, path, flags, &tmp);
928 if (ret)
929 goto err;
930 } else {
931 struct bkey_packed *k = bch2_btree_node_iter_peek(&l->iter, l->b);
932 if (!k) {
933 struct printbuf buf = PRINTBUF;
934
935 prt_str(&buf, "node not found at pos ");
936 bch2_bpos_to_text(&buf, path->pos);
937 prt_str(&buf, " within parent node ");
938 bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&l->b->key));
939
940 bch2_fs_fatal_error(c, "%s", buf.buf);
941 printbuf_exit(&buf);
942 ret = -BCH_ERR_btree_need_topology_repair;
943 goto err;
944 }
945
946 bch2_bkey_buf_unpack(&tmp, c, l->b, k);
947
948 if ((flags & BTREE_ITER_prefetch) &&
949 c->opts.btree_node_prefetch) {
950 ret = btree_path_prefetch(trans, path);
951 if (ret)
952 goto err;
953 }
954 }
955
956 b = bch2_btree_node_get(trans, path, tmp.k, level, lock_type, trace_ip);
957 ret = PTR_ERR_OR_ZERO(b);
958 if (unlikely(ret))
959 goto err;
960
961 if (likely(!trans->journal_replay_not_finished &&
962 tmp.k->k.type == KEY_TYPE_btree_ptr_v2) &&
963 unlikely(b != btree_node_mem_ptr(tmp.k)))
964 btree_node_mem_ptr_set(trans, path, level + 1, b);
965
966 if (btree_node_read_locked(path, level + 1))
967 btree_node_unlock(trans, path, level + 1);
968
969 mark_btree_node_locked(trans, path, level,
970 (enum btree_node_locked_type) lock_type);
971 path->level = level;
972 bch2_btree_path_level_init(trans, path, b);
973
974 bch2_btree_path_verify_locks(path);
975 err:
976 bch2_bkey_buf_exit(&tmp, c);
977 return ret;
978 }
979
980 static int bch2_btree_path_traverse_all(struct btree_trans *trans)
981 {
982 struct bch_fs *c = trans->c;
983 struct btree_path *path;
984 unsigned long trace_ip = _RET_IP_;
985 unsigned i;
986 int ret = 0;
987
988 if (trans->in_traverse_all)
989 return -BCH_ERR_transaction_restart_in_traverse_all;
990
991 trans->in_traverse_all = true;
992 retry_all:
993 trans->restarted = 0;
994 trans->last_restarted_ip = 0;
995
996 trans_for_each_path(trans, path, i)
997 path->should_be_locked = false;
998
999 btree_trans_sort_paths(trans);
1000
1001 bch2_trans_unlock(trans);
1002 cond_resched();
1003 trans_set_locked(trans);
1004
1005 if (unlikely(trans->memory_allocation_failure)) {
1006 struct closure cl;
1007
1008 closure_init_stack(&cl);
1009
1010 do {
1011 ret = bch2_btree_cache_cannibalize_lock(trans, &cl);
1012 closure_sync(&cl);
1013 } while (ret);
1014 }
1015
1016 /* Now, redo traversals in correct order: */
1017 i = 0;
1018 while (i < trans->nr_sorted) {
1019 btree_path_idx_t idx = trans->sorted[i];
1020
1021 /*
1022 * Traversing a path can cause another path to be added at about
1023 * the same position:
1024 */
1025 if (trans->paths[idx].uptodate) {
1026 __btree_path_get(trans, &trans->paths[idx], false);
1027 ret = bch2_btree_path_traverse_one(trans, idx, 0, _THIS_IP_);
1028 __btree_path_put(trans, &trans->paths[idx], false);
1029
1030 if (bch2_err_matches(ret, BCH_ERR_transaction_restart) ||
1031 bch2_err_matches(ret, ENOMEM))
1032 goto retry_all;
1033 if (ret)
1034 goto err;
1035 } else {
1036 i++;
1037 }
1038 }
1039
1040 /*
1041 * We used to assert that all paths had been traversed here
1042 * (path->uptodate < BTREE_ITER_NEED_TRAVERSE); however, since
1043 * path->should_be_locked is not set yet, we might have unlocked and
1044 * then failed to relock a path - that's fine.
1045 */
1046 err:
1047 bch2_btree_cache_cannibalize_unlock(trans);
1048
1049 trans->in_traverse_all = false;
1050
1051 trace_and_count(c, trans_traverse_all, trans, trace_ip);
1052 return ret;
1053 }
1054
1055 static inline bool btree_path_check_pos_in_node(struct btree_path *path,
1056 unsigned l, int check_pos)
1057 {
1058 if (check_pos < 0 && btree_path_pos_before_node(path, path->l[l].b))
1059 return false;
1060 if (check_pos > 0 && btree_path_pos_after_node(path, path->l[l].b))
1061 return false;
1062 return true;
1063 }
1064
1065 static inline bool btree_path_good_node(struct btree_trans *trans,
1066 struct btree_path *path,
1067 unsigned l, int check_pos)
1068 {
1069 return is_btree_node(path, l) &&
1070 bch2_btree_node_relock(trans, path, l) &&
1071 btree_path_check_pos_in_node(path, l, check_pos);
1072 }
1073
1074 static void btree_path_set_level_down(struct btree_trans *trans,
1075 struct btree_path *path,
1076 unsigned new_level)
1077 {
1078 unsigned l;
1079
1080 path->level = new_level;
1081
1082 for (l = path->level + 1; l < BTREE_MAX_DEPTH; l++)
1083 if (btree_lock_want(path, l) == BTREE_NODE_UNLOCKED)
1084 btree_node_unlock(trans, path, l);
1085
1086 btree_path_set_dirty(path, BTREE_ITER_NEED_TRAVERSE);
1087 bch2_btree_path_verify(trans, path);
1088 }
1089
1090 static noinline unsigned __btree_path_up_until_good_node(struct btree_trans *trans,
1091 struct btree_path *path,
1092 int check_pos)
1093 {
1094 unsigned i, l = path->level;
1095 again:
1096 while (btree_path_node(path, l) &&
1097 !btree_path_good_node(trans, path, l, check_pos))
1098 __btree_path_set_level_up(trans, path, l++);
1099
1100 /* If we need intent locks, take them too: */
1101 for (i = l + 1;
1102 i < path->locks_want && btree_path_node(path, i);
1103 i++)
1104 if (!bch2_btree_node_relock(trans, path, i)) {
1105 while (l <= i)
1106 __btree_path_set_level_up(trans, path, l++);
1107 goto again;
1108 }
1109
1110 return l;
1111 }
1112
1113 static inline unsigned btree_path_up_until_good_node(struct btree_trans *trans,
1114 struct btree_path *path,
1115 int check_pos)
1116 {
1117 return likely(btree_node_locked(path, path->level) &&
1118 btree_path_check_pos_in_node(path, path->level, check_pos))
1119 ? path->level
1120 : __btree_path_up_until_good_node(trans, path, check_pos);
1121 }
1122
1123 /*
1124 * This is the main state machine for walking down the btree - walks down to a
1125 * specified depth
1126 *
1127 * Returns 0 on success, -EIO on error (error reading in a btree node).
1128 *
1129 * On error, caller (peek_node()/peek_key()) must return NULL; the error is
1130 * stashed in the iterator and returned from bch2_trans_exit().
1131 */
1132 int bch2_btree_path_traverse_one(struct btree_trans *trans,
1133 btree_path_idx_t path_idx,
1134 unsigned flags,
1135 unsigned long trace_ip)
1136 {
1137 struct btree_path *path = &trans->paths[path_idx];
1138 unsigned depth_want = path->level;
1139 int ret = -((int) trans->restarted);
1140
1141 if (unlikely(ret))
1142 goto out;
1143
1144 if (unlikely(!trans->srcu_held))
1145 bch2_trans_srcu_lock(trans);
1146
1147 trace_btree_path_traverse_start(trans, path);
1148
1149 /*
1150 * Ensure we obey path->should_be_locked: if it's set, we can't unlock
1151 * and re-traverse the path without a transaction restart:
1152 */
1153 if (path->should_be_locked) {
1154 ret = bch2_btree_path_relock(trans, path, trace_ip);
1155 goto out;
1156 }
1157
1158 if (path->cached) {
1159 ret = bch2_btree_path_traverse_cached(trans, path, flags);
1160 goto out;
1161 }
1162
1163 path = &trans->paths[path_idx];
1164
1165 if (unlikely(path->level >= BTREE_MAX_DEPTH))
1166 goto out_uptodate;
1167
1168 path->level = btree_path_up_until_good_node(trans, path, 0);
1169 unsigned max_level = path->level;
1170
1171 EBUG_ON(btree_path_node(path, path->level) &&
1172 !btree_node_locked(path, path->level));
1173
1174 /*
1175 * Note: path->nodes[path->level] may be temporarily NULL here - that
1176 * would indicate to other code that we got to the end of the btree,
1177 * here it indicates that relocking the root failed - it's critical that
1178 * btree_path_lock_root() comes next and that it can't fail
1179 */
1180 while (path->level > depth_want) {
1181 ret = btree_path_node(path, path->level)
1182 ? btree_path_down(trans, path, flags, trace_ip)
1183 : btree_path_lock_root(trans, path, depth_want, trace_ip);
1184 if (unlikely(ret)) {
1185 if (ret == 1) {
1186 /*
1187 * No nodes at this level - got to the end of
1188 * the btree:
1189 */
1190 ret = 0;
1191 goto out;
1192 }
1193
1194 __bch2_btree_path_unlock(trans, path);
1195 path->level = depth_want;
1196 path->l[path->level].b = ERR_PTR(ret);
1197 goto out;
1198 }
1199 }
1200
1201 if (unlikely(max_level > path->level)) {
1202 struct btree_path *linked;
1203 unsigned iter;
1204
1205 trans_for_each_path_with_node(trans, path_l(path)->b, linked, iter)
1206 for (unsigned j = path->level + 1; j < max_level; j++)
1207 linked->l[j] = path->l[j];
1208 }
1209
1210 out_uptodate:
1211 path->uptodate = BTREE_ITER_UPTODATE;
1212 trace_btree_path_traverse_end(trans, path);
1213 out:
1214 if (bch2_err_matches(ret, BCH_ERR_transaction_restart) != !!trans->restarted)
1215 panic("ret %s (%i) trans->restarted %s (%i)\n",
1216 bch2_err_str(ret), ret,
1217 bch2_err_str(trans->restarted), trans->restarted);
1218 bch2_btree_path_verify(trans, path);
1219 return ret;
1220 }
1221
1222 static inline void btree_path_copy(struct btree_trans *trans, struct btree_path *dst,
1223 struct btree_path *src)
1224 {
1225 unsigned i, offset = offsetof(struct btree_path, pos);
1226
1227 memcpy((void *) dst + offset,
1228 (void *) src + offset,
1229 sizeof(struct btree_path) - offset);
1230
1231 for (i = 0; i < BTREE_MAX_DEPTH; i++) {
1232 unsigned t = btree_node_locked_type(dst, i);
1233
1234 if (t != BTREE_NODE_UNLOCKED)
1235 six_lock_increment(&dst->l[i].b->c.lock, t);
1236 }
1237 }
1238
1239 static btree_path_idx_t btree_path_clone(struct btree_trans *trans, btree_path_idx_t src,
1240 bool intent, unsigned long ip)
1241 {
1242 btree_path_idx_t new = btree_path_alloc(trans, src);
1243 btree_path_copy(trans, trans->paths + new, trans->paths + src);
1244 __btree_path_get(trans, trans->paths + new, intent);
1245 #ifdef TRACK_PATH_ALLOCATED
1246 trans->paths[new].ip_allocated = ip;
1247 #endif
1248 return new;
1249 }
1250
1251 __flatten
1252 btree_path_idx_t __bch2_btree_path_make_mut(struct btree_trans *trans,
1253 btree_path_idx_t path, bool intent, unsigned long ip)
1254 {
1255 struct btree_path *old = trans->paths + path;
1256 __btree_path_put(trans, trans->paths + path, intent);
1257 path = btree_path_clone(trans, path, intent, ip);
1258 trace_btree_path_clone(trans, old, trans->paths + path);
1259 trans->paths[path].preserve = false;
1260 return path;
1261 }
1262
1263 btree_path_idx_t __must_check
1264 __bch2_btree_path_set_pos(struct btree_trans *trans,
1265 btree_path_idx_t path_idx, struct bpos new_pos,
1266 bool intent, unsigned long ip)
1267 {
1268 int cmp = bpos_cmp(new_pos, trans->paths[path_idx].pos);
1269
1270 bch2_trans_verify_not_in_restart(trans);
1271 EBUG_ON(!trans->paths[path_idx].ref);
1272
1273 trace_btree_path_set_pos(trans, trans->paths + path_idx, &new_pos);
1274
1275 path_idx = bch2_btree_path_make_mut(trans, path_idx, intent, ip);
1276
1277 struct btree_path *path = trans->paths + path_idx;
1278 path->pos = new_pos;
1279 trans->paths_sorted = false;
1280
1281 if (unlikely(path->cached)) {
1282 btree_node_unlock(trans, path, 0);
1283 path->l[0].b = ERR_PTR(-BCH_ERR_no_btree_node_up);
1284 btree_path_set_dirty(path, BTREE_ITER_NEED_TRAVERSE);
1285 goto out;
1286 }
1287
1288 unsigned level = btree_path_up_until_good_node(trans, path, cmp);
1289
1290 if (btree_path_node(path, level)) {
1291 struct btree_path_level *l = &path->l[level];
1292
1293 BUG_ON(!btree_node_locked(path, level));
1294 /*
1295 * We might have to skip over many keys, or just a few: try
1296 * advancing the node iterator, and if we have to skip over too
1297 * many keys just reinit it (or if we're rewinding, since that
1298 * is expensive).
1299 */
1300 if (cmp < 0 ||
1301 !btree_path_advance_to_pos(path, l, 8))
1302 bch2_btree_node_iter_init(&l->iter, l->b, &path->pos);
1303
1304 /*
1305 * Iterators to interior nodes should always be pointed at the first non
1306 * whiteout:
1307 */
1308 if (unlikely(level))
1309 bch2_btree_node_iter_peek(&l->iter, l->b);
1310 }
1311
1312 if (unlikely(level != path->level)) {
1313 btree_path_set_dirty(path, BTREE_ITER_NEED_TRAVERSE);
1314 __bch2_btree_path_unlock(trans, path);
1315 }
1316 out:
1317 bch2_btree_path_verify(trans, path);
1318 return path_idx;
1319 }
1320
1321 /* Btree path: main interface: */
1322
1323 static struct btree_path *have_path_at_pos(struct btree_trans *trans, struct btree_path *path)
1324 {
1325 struct btree_path *sib;
1326
1327 sib = prev_btree_path(trans, path);
1328 if (sib && !btree_path_cmp(sib, path))
1329 return sib;
1330
1331 sib = next_btree_path(trans, path);
1332 if (sib && !btree_path_cmp(sib, path))
1333 return sib;
1334
1335 return NULL;
1336 }
1337
1338 static struct btree_path *have_node_at_pos(struct btree_trans *trans, struct btree_path *path)
1339 {
1340 struct btree_path *sib;
1341
1342 sib = prev_btree_path(trans, path);
1343 if (sib && sib->level == path->level && path_l(sib)->b == path_l(path)->b)
1344 return sib;
1345
1346 sib = next_btree_path(trans, path);
1347 if (sib && sib->level == path->level && path_l(sib)->b == path_l(path)->b)
1348 return sib;
1349
1350 return NULL;
1351 }
1352
1353 static inline void __bch2_path_free(struct btree_trans *trans, btree_path_idx_t path)
1354 {
1355 __bch2_btree_path_unlock(trans, trans->paths + path);
1356 btree_path_list_remove(trans, trans->paths + path);
1357 __clear_bit(path, trans->paths_allocated);
1358 }
1359
1360 static bool bch2_btree_path_can_relock(struct btree_trans *trans, struct btree_path *path)
1361 {
1362 unsigned l = path->level;
1363
1364 do {
1365 if (!btree_path_node(path, l))
1366 break;
1367
1368 if (!is_btree_node(path, l))
1369 return false;
1370
1371 if (path->l[l].lock_seq != path->l[l].b->c.lock.seq)
1372 return false;
1373
1374 l++;
1375 } while (l < path->locks_want);
1376
1377 return true;
1378 }
1379
1380 void bch2_path_put(struct btree_trans *trans, btree_path_idx_t path_idx, bool intent)
1381 {
1382 struct btree_path *path = trans->paths + path_idx, *dup;
1383
1384 if (!__btree_path_put(trans, path, intent))
1385 return;
1386
1387 dup = path->preserve
1388 ? have_path_at_pos(trans, path)
1389 : have_node_at_pos(trans, path);
1390
1391 trace_btree_path_free(trans, path_idx, dup);
1392
1393 if (!dup && !(!path->preserve && !is_btree_node(path, path->level)))
1394 return;
1395
1396 if (path->should_be_locked && !trans->restarted) {
1397 if (!dup)
1398 return;
1399
1400 if (!(trans->locked
1401 ? bch2_btree_path_relock_norestart(trans, dup)
1402 : bch2_btree_path_can_relock(trans, dup)))
1403 return;
1404 }
1405
1406 if (dup) {
1407 dup->preserve |= path->preserve;
1408 dup->should_be_locked |= path->should_be_locked;
1409 }
1410
1411 __bch2_path_free(trans, path_idx);
1412 }
1413
1414 static void bch2_path_put_nokeep(struct btree_trans *trans, btree_path_idx_t path,
1415 bool intent)
1416 {
1417 if (!__btree_path_put(trans, trans->paths + path, intent))
1418 return;
1419
1420 __bch2_path_free(trans, path);
1421 }
1422
1423 void __noreturn bch2_trans_restart_error(struct btree_trans *trans, u32 restart_count)
1424 {
1425 panic("trans->restart_count %u, should be %u, last restarted by %pS\n",
1426 trans->restart_count, restart_count,
1427 (void *) trans->last_begin_ip);
1428 }
1429
1430 void __noreturn bch2_trans_in_restart_error(struct btree_trans *trans)
1431 {
1432 panic("in transaction restart: %s, last restarted by %pS\n",
1433 bch2_err_str(trans->restarted),
1434 (void *) trans->last_restarted_ip);
1435 }
1436
1437 void __noreturn bch2_trans_unlocked_error(struct btree_trans *trans)
1438 {
1439 panic("trans should be locked, unlocked by %pS\n",
1440 (void *) trans->last_unlock_ip);
1441 }
1442
1443 noinline __cold
1444 void bch2_trans_updates_to_text(struct printbuf *buf, struct btree_trans *trans)
1445 {
1446 prt_printf(buf, "%u transaction updates for %s journal seq %llu\n",
1447 trans->nr_updates, trans->fn, trans->journal_res.seq);
1448 printbuf_indent_add(buf, 2);
1449
1450 trans_for_each_update(trans, i) {
1451 struct bkey_s_c old = { &i->old_k, i->old_v };
1452
1453 prt_printf(buf, "update: btree=%s cached=%u %pS\n",
1454 bch2_btree_id_str(i->btree_id),
1455 i->cached,
1456 (void *) i->ip_allocated);
1457
1458 prt_printf(buf, " old ");
1459 bch2_bkey_val_to_text(buf, trans->c, old);
1460 prt_newline(buf);
1461
1462 prt_printf(buf, " new ");
1463 bch2_bkey_val_to_text(buf, trans->c, bkey_i_to_s_c(i->k));
1464 prt_newline(buf);
1465 }
1466
1467 for (struct jset_entry *e = trans->journal_entries;
1468 e != btree_trans_journal_entries_top(trans);
1469 e = vstruct_next(e))
1470 bch2_journal_entry_to_text(buf, trans->c, e);
1471
1472 printbuf_indent_sub(buf, 2);
1473 }
1474
1475 noinline __cold
1476 void bch2_dump_trans_updates(struct btree_trans *trans)
1477 {
1478 struct printbuf buf = PRINTBUF;
1479
1480 bch2_trans_updates_to_text(&buf, trans);
1481 bch2_print_str(trans->c, buf.buf);
1482 printbuf_exit(&buf);
1483 }
1484
1485 static void bch2_btree_path_to_text_short(struct printbuf *out, struct btree_trans *trans, btree_path_idx_t path_idx)
1486 {
1487 struct btree_path *path = trans->paths + path_idx;
1488
1489 prt_printf(out, "path: idx %3u ref %u:%u %c %c %c btree=%s l=%u pos ",
1490 path_idx, path->ref, path->intent_ref,
1491 path->preserve ? 'P' : ' ',
1492 path->should_be_locked ? 'S' : ' ',
1493 path->cached ? 'C' : 'B',
1494 bch2_btree_id_str(path->btree_id),
1495 path->level);
1496 bch2_bpos_to_text(out, path->pos);
1497
1498 if (!path->cached && btree_node_locked(path, path->level)) {
1499 prt_char(out, ' ');
1500 struct btree *b = path_l(path)->b;
1501 bch2_bpos_to_text(out, b->data->min_key);
1502 prt_char(out, '-');
1503 bch2_bpos_to_text(out, b->key.k.p);
1504 }
1505
1506 #ifdef TRACK_PATH_ALLOCATED
1507 prt_printf(out, " %pS", (void *) path->ip_allocated);
1508 #endif
1509 }
1510
1511 static const char *btree_node_locked_str(enum btree_node_locked_type t)
1512 {
1513 switch (t) {
1514 case BTREE_NODE_UNLOCKED:
1515 return "unlocked";
1516 case BTREE_NODE_READ_LOCKED:
1517 return "read";
1518 case BTREE_NODE_INTENT_LOCKED:
1519 return "intent";
1520 case BTREE_NODE_WRITE_LOCKED:
1521 return "write";
1522 default:
1523 return NULL;
1524 }
1525 }
1526
1527 void bch2_btree_path_to_text(struct printbuf *out, struct btree_trans *trans, btree_path_idx_t path_idx)
1528 {
1529 bch2_btree_path_to_text_short(out, trans, path_idx);
1530
1531 struct btree_path *path = trans->paths + path_idx;
1532
1533 prt_printf(out, " uptodate %u locks_want %u", path->uptodate, path->locks_want);
1534 prt_newline(out);
1535
1536 printbuf_indent_add(out, 2);
1537 for (unsigned l = 0; l < BTREE_MAX_DEPTH; l++) {
1538 prt_printf(out, "l=%u locks %s seq %u node ", l,
1539 btree_node_locked_str(btree_node_locked_type(path, l)),
1540 path->l[l].lock_seq);
1541
1542 int ret = PTR_ERR_OR_ZERO(path->l[l].b);
1543 if (ret)
1544 prt_str(out, bch2_err_str(ret));
1545 else
1546 prt_printf(out, "%px", path->l[l].b);
1547 prt_newline(out);
1548 }
1549 printbuf_indent_sub(out, 2);
1550 }
1551
1552 static noinline __cold
1553 void __bch2_trans_paths_to_text(struct printbuf *out, struct btree_trans *trans,
1554 bool nosort)
1555 {
1556 struct trans_for_each_path_inorder_iter iter;
1557
1558 if (!nosort)
1559 btree_trans_sort_paths(trans);
1560
1561 trans_for_each_path_idx_inorder(trans, iter) {
1562 bch2_btree_path_to_text_short(out, trans, iter.path_idx);
1563 prt_newline(out);
1564 }
1565 }
1566
1567 noinline __cold
1568 void bch2_trans_paths_to_text(struct printbuf *out, struct btree_trans *trans)
1569 {
1570 __bch2_trans_paths_to_text(out, trans, false);
1571 }
1572
1573 static noinline __cold
1574 void __bch2_dump_trans_paths_updates(struct btree_trans *trans, bool nosort)
1575 {
1576 struct printbuf buf = PRINTBUF;
1577
1578 __bch2_trans_paths_to_text(&buf, trans, nosort);
1579 bch2_trans_updates_to_text(&buf, trans);
1580
1581 bch2_print_str(trans->c, buf.buf);
1582 printbuf_exit(&buf);
1583 }
1584
1585 noinline __cold
1586 void bch2_dump_trans_paths_updates(struct btree_trans *trans)
1587 {
1588 __bch2_dump_trans_paths_updates(trans, false);
1589 }
1590
1591 noinline __cold
1592 static void bch2_trans_update_max_paths(struct btree_trans *trans)
1593 {
1594 struct btree_transaction_stats *s = btree_trans_stats(trans);
1595 struct printbuf buf = PRINTBUF;
1596 size_t nr = bitmap_weight(trans->paths_allocated, trans->nr_paths);
1597
1598 bch2_trans_paths_to_text(&buf, trans);
1599
1600 if (!buf.allocation_failure) {
1601 mutex_lock(&s->lock);
1602 if (nr > s->nr_max_paths) {
1603 s->nr_max_paths = nr;
1604 swap(s->max_paths_text, buf.buf);
1605 }
1606 mutex_unlock(&s->lock);
1607 }
1608
1609 printbuf_exit(&buf);
1610
1611 trans->nr_paths_max = nr;
1612 }
1613
1614 noinline __cold
1615 int __bch2_btree_trans_too_many_iters(struct btree_trans *trans)
1616 {
1617 if (trace_trans_restart_too_many_iters_enabled()) {
1618 struct printbuf buf = PRINTBUF;
1619
1620 bch2_trans_paths_to_text(&buf, trans);
1621 trace_trans_restart_too_many_iters(trans, _THIS_IP_, buf.buf);
1622 printbuf_exit(&buf);
1623 }
1624
1625 count_event(trans->c, trans_restart_too_many_iters);
1626
1627 return btree_trans_restart(trans, BCH_ERR_transaction_restart_too_many_iters);
1628 }
1629
1630 static noinline void btree_path_overflow(struct btree_trans *trans)
1631 {
1632 bch2_dump_trans_paths_updates(trans);
1633 bch_err(trans->c, "trans path overflow");
1634 }
1635
1636 static noinline void btree_paths_realloc(struct btree_trans *trans)
1637 {
1638 unsigned nr = trans->nr_paths * 2;
1639
1640 void *p = kvzalloc(BITS_TO_LONGS(nr) * sizeof(unsigned long) +
1641 sizeof(struct btree_trans_paths) +
1642 nr * sizeof(struct btree_path) +
1643 nr * sizeof(btree_path_idx_t) + 8 +
1644 nr * sizeof(struct btree_insert_entry), GFP_KERNEL|__GFP_NOFAIL);
1645
1646 unsigned long *paths_allocated = p;
1647 memcpy(paths_allocated, trans->paths_allocated, BITS_TO_LONGS(trans->nr_paths) * sizeof(unsigned long));
1648 p += BITS_TO_LONGS(nr) * sizeof(unsigned long);
1649
1650 p += sizeof(struct btree_trans_paths);
1651 struct btree_path *paths = p;
1652 *trans_paths_nr(paths) = nr;
1653 memcpy(paths, trans->paths, trans->nr_paths * sizeof(struct btree_path));
1654 p += nr * sizeof(struct btree_path);
1655
1656 btree_path_idx_t *sorted = p;
1657 memcpy(sorted, trans->sorted, trans->nr_sorted * sizeof(btree_path_idx_t));
1658 p += nr * sizeof(btree_path_idx_t) + 8;
1659
1660 struct btree_insert_entry *updates = p;
1661 memcpy(updates, trans->updates, trans->nr_paths * sizeof(struct btree_insert_entry));
1662
1663 unsigned long *old = trans->paths_allocated;
1664
1665 rcu_assign_pointer(trans->paths_allocated, paths_allocated);
1666 rcu_assign_pointer(trans->paths, paths);
1667 rcu_assign_pointer(trans->sorted, sorted);
1668 rcu_assign_pointer(trans->updates, updates);
1669
1670 trans->nr_paths = nr;
1671
1672 if (old != trans->_paths_allocated)
1673 kfree_rcu_mightsleep(old);
1674 }
1675
1676 static inline btree_path_idx_t btree_path_alloc(struct btree_trans *trans,
1677 btree_path_idx_t pos)
1678 {
1679 btree_path_idx_t idx = find_first_zero_bit(trans->paths_allocated, trans->nr_paths);
1680
1681 if (unlikely(idx == trans->nr_paths)) {
1682 if (trans->nr_paths == BTREE_ITER_MAX) {
1683 btree_path_overflow(trans);
1684 return 0;
1685 }
1686
1687 btree_paths_realloc(trans);
1688 }
1689
1690 /*
1691 * Do this before marking the new path as allocated, since it won't be
1692 * initialized yet:
1693 */
1694 if (unlikely(idx > trans->nr_paths_max))
1695 bch2_trans_update_max_paths(trans);
1696
1697 __set_bit(idx, trans->paths_allocated);
1698
1699 struct btree_path *path = &trans->paths[idx];
1700 path->ref = 0;
1701 path->intent_ref = 0;
1702 path->nodes_locked = 0;
1703
1704 btree_path_list_add(trans, pos, idx);
1705 trans->paths_sorted = false;
1706 return idx;
1707 }
1708
1709 btree_path_idx_t bch2_path_get(struct btree_trans *trans,
1710 enum btree_id btree_id, struct bpos pos,
1711 unsigned locks_want, unsigned level,
1712 unsigned flags, unsigned long ip)
1713 {
1714 struct btree_path *path;
1715 bool cached = flags & BTREE_ITER_cached;
1716 bool intent = flags & BTREE_ITER_intent;
1717 struct trans_for_each_path_inorder_iter iter;
1718 btree_path_idx_t path_pos = 0, path_idx;
1719
1720 bch2_trans_verify_not_unlocked(trans);
1721 bch2_trans_verify_not_in_restart(trans);
1722 bch2_trans_verify_locks(trans);
1723
1724 btree_trans_sort_paths(trans);
1725
1726 trans_for_each_path_inorder(trans, path, iter) {
1727 if (__btree_path_cmp(path,
1728 btree_id,
1729 cached,
1730 pos,
1731 level) > 0)
1732 break;
1733
1734 path_pos = iter.path_idx;
1735 }
1736
1737 if (path_pos &&
1738 trans->paths[path_pos].cached == cached &&
1739 trans->paths[path_pos].btree_id == btree_id &&
1740 trans->paths[path_pos].level == level) {
1741 trace_btree_path_get(trans, trans->paths + path_pos, &pos);
1742
1743 __btree_path_get(trans, trans->paths + path_pos, intent);
1744 path_idx = bch2_btree_path_set_pos(trans, path_pos, pos, intent, ip);
1745 path = trans->paths + path_idx;
1746 } else {
1747 path_idx = btree_path_alloc(trans, path_pos);
1748 path = trans->paths + path_idx;
1749
1750 __btree_path_get(trans, path, intent);
1751 path->pos = pos;
1752 path->btree_id = btree_id;
1753 path->cached = cached;
1754 path->uptodate = BTREE_ITER_NEED_TRAVERSE;
1755 path->should_be_locked = false;
1756 path->level = level;
1757 path->locks_want = locks_want;
1758 path->nodes_locked = 0;
1759 for (unsigned i = 0; i < ARRAY_SIZE(path->l); i++)
1760 path->l[i].b = ERR_PTR(-BCH_ERR_no_btree_node_init);
1761 #ifdef TRACK_PATH_ALLOCATED
1762 path->ip_allocated = ip;
1763 #endif
1764 trans->paths_sorted = false;
1765
1766 trace_btree_path_alloc(trans, path);
1767 }
1768
1769 if (!(flags & BTREE_ITER_nopreserve))
1770 path->preserve = true;
1771
1772 if (path->intent_ref)
1773 locks_want = max(locks_want, level + 1);
1774
1775 /*
1776 * If the path has locks_want greater than requested, we don't downgrade
1777 * it here - on transaction restart because btree node split needs to
1778 * upgrade locks, we might be putting/getting the iterator again.
1779 * Downgrading iterators only happens via bch2_trans_downgrade(), after
1780 * a successful transaction commit.
1781 */
1782
1783 locks_want = min(locks_want, BTREE_MAX_DEPTH);
1784 if (locks_want > path->locks_want)
1785 bch2_btree_path_upgrade_noupgrade_sibs(trans, path, locks_want, NULL);
1786
1787 return path_idx;
1788 }
1789
1790 btree_path_idx_t bch2_path_get_unlocked_mut(struct btree_trans *trans,
1791 enum btree_id btree_id,
1792 unsigned level,
1793 struct bpos pos)
1794 {
1795 btree_path_idx_t path_idx = bch2_path_get(trans, btree_id, pos, level + 1, level,
1796 BTREE_ITER_nopreserve|
1797 BTREE_ITER_intent, _RET_IP_);
1798 path_idx = bch2_btree_path_make_mut(trans, path_idx, true, _RET_IP_);
1799
1800 struct btree_path *path = trans->paths + path_idx;
1801 bch2_btree_path_downgrade(trans, path);
1802 __bch2_btree_path_unlock(trans, path);
1803 return path_idx;
1804 }
1805
1806 struct bkey_s_c bch2_btree_path_peek_slot(struct btree_path *path, struct bkey *u)
1807 {
1808
1809 struct btree_path_level *l = path_l(path);
1810 struct bkey_packed *_k;
1811 struct bkey_s_c k;
1812
1813 if (unlikely(!l->b))
1814 return bkey_s_c_null;
1815
1816 EBUG_ON(path->uptodate != BTREE_ITER_UPTODATE);
1817 EBUG_ON(!btree_node_locked(path, path->level));
1818
1819 if (!path->cached) {
1820 _k = bch2_btree_node_iter_peek_all(&l->iter, l->b);
1821 k = _k ? bkey_disassemble(l->b, _k, u) : bkey_s_c_null;
1822
1823 EBUG_ON(k.k && bkey_deleted(k.k) && bpos_eq(k.k->p, path->pos));
1824
1825 if (!k.k || !bpos_eq(path->pos, k.k->p))
1826 goto hole;
1827 } else {
1828 struct bkey_cached *ck = (void *) path->l[0].b;
1829 if (!ck)
1830 return bkey_s_c_null;
1831
1832 EBUG_ON(path->btree_id != ck->key.btree_id ||
1833 !bkey_eq(path->pos, ck->key.pos));
1834
1835 *u = ck->k->k;
1836 k = bkey_i_to_s_c(ck->k);
1837 }
1838
1839 return k;
1840 hole:
1841 bkey_init(u);
1842 u->p = path->pos;
1843 return (struct bkey_s_c) { u, NULL };
1844 }
1845
1846
1847 void bch2_set_btree_iter_dontneed(struct btree_iter *iter)
1848 {
1849 struct btree_trans *trans = iter->trans;
1850
1851 if (!iter->path || trans->restarted)
1852 return;
1853
1854 struct btree_path *path = btree_iter_path(trans, iter);
1855 path->preserve = false;
1856 if (path->ref == 1)
1857 path->should_be_locked = false;
1858 }
1859 /* Btree iterators: */
1860
1861 int __must_check
1862 __bch2_btree_iter_traverse(struct btree_iter *iter)
1863 {
1864 return bch2_btree_path_traverse(iter->trans, iter->path, iter->flags);
1865 }
1866
1867 int __must_check
1868 bch2_btree_iter_traverse(struct btree_iter *iter)
1869 {
1870 struct btree_trans *trans = iter->trans;
1871 int ret;
1872
1873 bch2_trans_verify_not_unlocked(trans);
1874
1875 iter->path = bch2_btree_path_set_pos(trans, iter->path,
1876 btree_iter_search_key(iter),
1877 iter->flags & BTREE_ITER_intent,
1878 btree_iter_ip_allocated(iter));
1879
1880 ret = bch2_btree_path_traverse(iter->trans, iter->path, iter->flags);
1881 if (ret)
1882 return ret;
1883
1884 struct btree_path *path = btree_iter_path(trans, iter);
1885 if (btree_path_node(path, path->level))
1886 btree_path_set_should_be_locked(trans, path);
1887 return 0;
1888 }
1889
1890 /* Iterate across nodes (leaf and interior nodes) */
1891
1892 struct btree *bch2_btree_iter_peek_node(struct btree_iter *iter)
1893 {
1894 struct btree_trans *trans = iter->trans;
1895 struct btree *b = NULL;
1896 int ret;
1897
1898 EBUG_ON(trans->paths[iter->path].cached);
1899 bch2_btree_iter_verify(iter);
1900
1901 ret = bch2_btree_path_traverse(trans, iter->path, iter->flags);
1902 if (ret)
1903 goto err;
1904
1905 struct btree_path *path = btree_iter_path(trans, iter);
1906 b = btree_path_node(path, path->level);
1907 if (!b)
1908 goto out;
1909
1910 BUG_ON(bpos_lt(b->key.k.p, iter->pos));
1911
1912 bkey_init(&iter->k);
1913 iter->k.p = iter->pos = b->key.k.p;
1914
1915 iter->path = bch2_btree_path_set_pos(trans, iter->path, b->key.k.p,
1916 iter->flags & BTREE_ITER_intent,
1917 btree_iter_ip_allocated(iter));
1918 btree_path_set_should_be_locked(trans, btree_iter_path(trans, iter));
1919 out:
1920 bch2_btree_iter_verify_entry_exit(iter);
1921 bch2_btree_iter_verify(iter);
1922
1923 return b;
1924 err:
1925 b = ERR_PTR(ret);
1926 goto out;
1927 }
1928
1929 /* Only kept for -tools */
1930 struct btree *bch2_btree_iter_peek_node_and_restart(struct btree_iter *iter)
1931 {
1932 struct btree *b;
1933
1934 while (b = bch2_btree_iter_peek_node(iter),
1935 bch2_err_matches(PTR_ERR_OR_ZERO(b), BCH_ERR_transaction_restart))
1936 bch2_trans_begin(iter->trans);
1937
1938 return b;
1939 }
1940
1941 struct btree *bch2_btree_iter_next_node(struct btree_iter *iter)
1942 {
1943 struct btree_trans *trans = iter->trans;
1944 struct btree *b = NULL;
1945 int ret;
1946
1947 EBUG_ON(trans->paths[iter->path].cached);
1948 bch2_trans_verify_not_in_restart(trans);
1949 bch2_btree_iter_verify(iter);
1950
1951 ret = bch2_btree_path_traverse(trans, iter->path, iter->flags);
1952 if (ret)
1953 goto err;
1954
1955
1956 struct btree_path *path = btree_iter_path(trans, iter);
1957
1958 /* already at end? */
1959 if (!btree_path_node(path, path->level))
1960 return NULL;
1961
1962 /* got to end? */
1963 if (!btree_path_node(path, path->level + 1)) {
1964 btree_path_set_level_up(trans, path);
1965 return NULL;
1966 }
1967
1968 if (!bch2_btree_node_relock(trans, path, path->level + 1)) {
1969 __bch2_btree_path_unlock(trans, path);
1970 path->l[path->level].b = ERR_PTR(-BCH_ERR_no_btree_node_relock);
1971 path->l[path->level + 1].b = ERR_PTR(-BCH_ERR_no_btree_node_relock);
1972 btree_path_set_dirty(path, BTREE_ITER_NEED_TRAVERSE);
1973 trace_and_count(trans->c, trans_restart_relock_next_node, trans, _THIS_IP_, path);
1974 ret = btree_trans_restart(trans, BCH_ERR_transaction_restart_relock);
1975 goto err;
1976 }
1977
1978 b = btree_path_node(path, path->level + 1);
1979
1980 if (bpos_eq(iter->pos, b->key.k.p)) {
1981 __btree_path_set_level_up(trans, path, path->level++);
1982 } else {
1983 if (btree_lock_want(path, path->level + 1) == BTREE_NODE_UNLOCKED)
1984 btree_node_unlock(trans, path, path->level + 1);
1985
1986 /*
1987 * Haven't gotten to the end of the parent node: go back down to
1988 * the next child node
1989 */
1990 iter->path = bch2_btree_path_set_pos(trans, iter->path,
1991 bpos_successor(iter->pos),
1992 iter->flags & BTREE_ITER_intent,
1993 btree_iter_ip_allocated(iter));
1994
1995 path = btree_iter_path(trans, iter);
1996 btree_path_set_level_down(trans, path, iter->min_depth);
1997
1998 ret = bch2_btree_path_traverse(trans, iter->path, iter->flags);
1999 if (ret)
2000 goto err;
2001
2002 path = btree_iter_path(trans, iter);
2003 b = path->l[path->level].b;
2004 }
2005
2006 bkey_init(&iter->k);
2007 iter->k.p = iter->pos = b->key.k.p;
2008
2009 iter->path = bch2_btree_path_set_pos(trans, iter->path, b->key.k.p,
2010 iter->flags & BTREE_ITER_intent,
2011 btree_iter_ip_allocated(iter));
2012 btree_path_set_should_be_locked(trans, btree_iter_path(trans, iter));
2013 EBUG_ON(btree_iter_path(trans, iter)->uptodate);
2014 out:
2015 bch2_btree_iter_verify_entry_exit(iter);
2016 bch2_btree_iter_verify(iter);
2017
2018 return b;
2019 err:
2020 b = ERR_PTR(ret);
2021 goto out;
2022 }
2023
2024 /* Iterate across keys (in leaf nodes only) */
2025
2026 inline bool bch2_btree_iter_advance(struct btree_iter *iter)
2027 {
2028 struct bpos pos = iter->k.p;
2029 bool ret = !(iter->flags & BTREE_ITER_all_snapshots
2030 ? bpos_eq(pos, SPOS_MAX)
2031 : bkey_eq(pos, SPOS_MAX));
2032
2033 if (ret && !(iter->flags & BTREE_ITER_is_extents))
2034 pos = bkey_successor(iter, pos);
2035 bch2_btree_iter_set_pos(iter, pos);
2036 return ret;
2037 }
2038
2039 inline bool bch2_btree_iter_rewind(struct btree_iter *iter)
2040 {
2041 struct bpos pos = bkey_start_pos(&iter->k);
2042 bool ret = !(iter->flags & BTREE_ITER_all_snapshots
2043 ? bpos_eq(pos, POS_MIN)
2044 : bkey_eq(pos, POS_MIN));
2045
2046 if (ret && !(iter->flags & BTREE_ITER_is_extents))
2047 pos = bkey_predecessor(iter, pos);
2048 bch2_btree_iter_set_pos(iter, pos);
2049 return ret;
2050 }
2051
2052 static noinline
2053 void bch2_btree_trans_peek_prev_updates(struct btree_trans *trans, struct btree_iter *iter,
2054 struct bkey_s_c *k)
2055 {
2056 struct bpos end = path_l(btree_iter_path(trans, iter))->b->data->min_key;
2057
2058 trans_for_each_update(trans, i)
2059 if (!i->key_cache_already_flushed &&
2060 i->btree_id == iter->btree_id &&
2061 bpos_le(i->k->k.p, iter->pos) &&
2062 bpos_ge(i->k->k.p, k->k ? k->k->p : end)) {
2063 iter->k = i->k->k;
2064 *k = bkey_i_to_s_c(i->k);
2065 }
2066 }
2067
2068 static noinline
2069 void bch2_btree_trans_peek_updates(struct btree_trans *trans, struct btree_iter *iter,
2070 struct bkey_s_c *k)
2071 {
2072 struct btree_path *path = btree_iter_path(trans, iter);
2073 struct bpos end = path_l(path)->b->key.k.p;
2074
2075 trans_for_each_update(trans, i)
2076 if (!i->key_cache_already_flushed &&
2077 i->btree_id == iter->btree_id &&
2078 bpos_ge(i->k->k.p, path->pos) &&
2079 bpos_le(i->k->k.p, k->k ? k->k->p : end)) {
2080 iter->k = i->k->k;
2081 *k = bkey_i_to_s_c(i->k);
2082 }
2083 }
2084
2085 static noinline
2086 void bch2_btree_trans_peek_slot_updates(struct btree_trans *trans, struct btree_iter *iter,
2087 struct bkey_s_c *k)
2088 {
2089 trans_for_each_update(trans, i)
2090 if (!i->key_cache_already_flushed &&
2091 i->btree_id == iter->btree_id &&
2092 bpos_eq(i->k->k.p, iter->pos)) {
2093 iter->k = i->k->k;
2094 *k = bkey_i_to_s_c(i->k);
2095 }
2096 }
2097
2098 static struct bkey_i *bch2_btree_journal_peek(struct btree_trans *trans,
2099 struct btree_iter *iter,
2100 struct bpos end_pos)
2101 {
2102 struct btree_path *path = btree_iter_path(trans, iter);
2103
2104 return bch2_journal_keys_peek_upto(trans->c, iter->btree_id,
2105 path->level,
2106 path->pos,
2107 end_pos,
2108 &iter->journal_idx);
2109 }
2110
2111 static noinline
2112 struct bkey_s_c btree_trans_peek_slot_journal(struct btree_trans *trans,
2113 struct btree_iter *iter)
2114 {
2115 struct btree_path *path = btree_iter_path(trans, iter);
2116 struct bkey_i *k = bch2_btree_journal_peek(trans, iter, path->pos);
2117
2118 if (k) {
2119 iter->k = k->k;
2120 return bkey_i_to_s_c(k);
2121 } else {
2122 return bkey_s_c_null;
2123 }
2124 }
2125
2126 static noinline
2127 struct bkey_s_c btree_trans_peek_journal(struct btree_trans *trans,
2128 struct btree_iter *iter,
2129 struct bkey_s_c k)
2130 {
2131 struct btree_path *path = btree_iter_path(trans, iter);
2132 struct bkey_i *next_journal =
2133 bch2_btree_journal_peek(trans, iter,
2134 k.k ? k.k->p : path_l(path)->b->key.k.p);
2135
2136 if (next_journal) {
2137 iter->k = next_journal->k;
2138 k = bkey_i_to_s_c(next_journal);
2139 }
2140
2141 return k;
2142 }
2143
2144 /*
2145 * Checks btree key cache for key at iter->pos and returns it if present, or
2146 * bkey_s_c_null:
2147 */
2148 static noinline
2149 struct bkey_s_c btree_trans_peek_key_cache(struct btree_iter *iter, struct bpos pos)
2150 {
2151 struct btree_trans *trans = iter->trans;
2152 struct bch_fs *c = trans->c;
2153 struct bkey u;
2154 struct bkey_s_c k;
2155 int ret;
2156
2157 bch2_trans_verify_not_in_restart(trans);
2158 bch2_trans_verify_not_unlocked(trans);
2159
2160 if ((iter->flags & BTREE_ITER_key_cache_fill) &&
2161 bpos_eq(iter->pos, pos))
2162 return bkey_s_c_null;
2163
2164 if (!bch2_btree_key_cache_find(c, iter->btree_id, pos))
2165 return bkey_s_c_null;
2166
2167 if (!iter->key_cache_path)
2168 iter->key_cache_path = bch2_path_get(trans, iter->btree_id, pos,
2169 iter->flags & BTREE_ITER_intent, 0,
2170 iter->flags|BTREE_ITER_cached|
2171 BTREE_ITER_cached_nofill,
2172 _THIS_IP_);
2173
2174 iter->key_cache_path = bch2_btree_path_set_pos(trans, iter->key_cache_path, pos,
2175 iter->flags & BTREE_ITER_intent,
2176 btree_iter_ip_allocated(iter));
2177
2178 ret = bch2_btree_path_traverse(trans, iter->key_cache_path,
2179 iter->flags|BTREE_ITER_cached) ?:
2180 bch2_btree_path_relock(trans, btree_iter_path(trans, iter), _THIS_IP_);
2181 if (unlikely(ret))
2182 return bkey_s_c_err(ret);
2183
2184 btree_path_set_should_be_locked(trans, trans->paths + iter->key_cache_path);
2185
2186 k = bch2_btree_path_peek_slot(trans->paths + iter->key_cache_path, &u);
2187 if (k.k && !bkey_err(k)) {
2188 iter->k = u;
2189 k.k = &iter->k;
2190 }
2191 return k;
2192 }
2193
2194 static struct bkey_s_c __bch2_btree_iter_peek(struct btree_iter *iter, struct bpos search_key)
2195 {
2196 struct btree_trans *trans = iter->trans;
2197 struct bkey_s_c k, k2;
2198 int ret;
2199
2200 EBUG_ON(btree_iter_path(trans, iter)->cached);
2201 bch2_btree_iter_verify(iter);
2202
2203 while (1) {
2204 struct btree_path_level *l;
2205
2206 iter->path = bch2_btree_path_set_pos(trans, iter->path, search_key,
2207 iter->flags & BTREE_ITER_intent,
2208 btree_iter_ip_allocated(iter));
2209
2210 ret = bch2_btree_path_traverse(trans, iter->path, iter->flags);
2211 if (unlikely(ret)) {
2212 /* ensure that iter->k is consistent with iter->pos: */
2213 bch2_btree_iter_set_pos(iter, iter->pos);
2214 k = bkey_s_c_err(ret);
2215 goto out;
2216 }
2217
2218 struct btree_path *path = btree_iter_path(trans, iter);
2219 l = path_l(path);
2220
2221 if (unlikely(!l->b)) {
2222 /* No btree nodes at requested level: */
2223 bch2_btree_iter_set_pos(iter, SPOS_MAX);
2224 k = bkey_s_c_null;
2225 goto out;
2226 }
2227
2228 btree_path_set_should_be_locked(trans, path);
2229
2230 k = btree_path_level_peek_all(trans->c, l, &iter->k);
2231
2232 if (unlikely(iter->flags & BTREE_ITER_with_key_cache) &&
2233 k.k &&
2234 (k2 = btree_trans_peek_key_cache(iter, k.k->p)).k) {
2235 k = k2;
2236 ret = bkey_err(k);
2237 if (ret) {
2238 bch2_btree_iter_set_pos(iter, iter->pos);
2239 goto out;
2240 }
2241 }
2242
2243 if (unlikely(iter->flags & BTREE_ITER_with_journal))
2244 k = btree_trans_peek_journal(trans, iter, k);
2245
2246 if (unlikely((iter->flags & BTREE_ITER_with_updates) &&
2247 trans->nr_updates))
2248 bch2_btree_trans_peek_updates(trans, iter, &k);
2249
2250 if (k.k && bkey_deleted(k.k)) {
2251 /*
2252 * If we've got a whiteout, and it's after the search
2253 * key, advance the search key to the whiteout instead
2254 * of just after the whiteout - it might be a btree
2255 * whiteout, with a real key at the same position, since
2256 * in the btree deleted keys sort before non deleted.
2257 */
2258 search_key = !bpos_eq(search_key, k.k->p)
2259 ? k.k->p
2260 : bpos_successor(k.k->p);
2261 continue;
2262 }
2263
2264 if (likely(k.k)) {
2265 break;
2266 } else if (likely(!bpos_eq(l->b->key.k.p, SPOS_MAX))) {
2267 /* Advance to next leaf node: */
2268 search_key = bpos_successor(l->b->key.k.p);
2269 } else {
2270 /* End of btree: */
2271 bch2_btree_iter_set_pos(iter, SPOS_MAX);
2272 k = bkey_s_c_null;
2273 goto out;
2274 }
2275 }
2276 out:
2277 bch2_btree_iter_verify(iter);
2278
2279 return k;
2280 }
2281
2282 /**
2283 * bch2_btree_iter_peek_upto() - returns first key greater than or equal to
2284 * iterator's current position
2285 * @iter: iterator to peek from
2286 * @end: search limit: returns keys less than or equal to @end
2287 *
2288 * Returns: key if found, or an error extractable with bkey_err().
2289 */
2290 struct bkey_s_c bch2_btree_iter_peek_upto(struct btree_iter *iter, struct bpos end)
2291 {
2292 struct btree_trans *trans = iter->trans;
2293 struct bpos search_key = btree_iter_search_key(iter);
2294 struct bkey_s_c k;
2295 struct bpos iter_pos;
2296 int ret;
2297
2298 bch2_trans_verify_not_unlocked(trans);
2299 EBUG_ON((iter->flags & BTREE_ITER_filter_snapshots) && bkey_eq(end, POS_MAX));
2300
2301 if (iter->update_path) {
2302 bch2_path_put_nokeep(trans, iter->update_path,
2303 iter->flags & BTREE_ITER_intent);
2304 iter->update_path = 0;
2305 }
2306
2307 bch2_btree_iter_verify_entry_exit(iter);
2308
2309 while (1) {
2310 k = __bch2_btree_iter_peek(iter, search_key);
2311 if (unlikely(!k.k))
2312 goto end;
2313 if (unlikely(bkey_err(k)))
2314 goto out_no_locked;
2315
2316 /*
2317 * We need to check against @end before FILTER_SNAPSHOTS because
2318 * if we get to a different inode that requested we might be
2319 * seeing keys for a different snapshot tree that will all be
2320 * filtered out.
2321 *
2322 * But we can't do the full check here, because bkey_start_pos()
2323 * isn't monotonically increasing before FILTER_SNAPSHOTS, and
2324 * that's what we check against in extents mode:
2325 */
2326 if (unlikely(!(iter->flags & BTREE_ITER_is_extents)
2327 ? bkey_gt(k.k->p, end)
2328 : k.k->p.inode > end.inode))
2329 goto end;
2330
2331 if (iter->update_path &&
2332 !bkey_eq(trans->paths[iter->update_path].pos, k.k->p)) {
2333 bch2_path_put_nokeep(trans, iter->update_path,
2334 iter->flags & BTREE_ITER_intent);
2335 iter->update_path = 0;
2336 }
2337
2338 if ((iter->flags & BTREE_ITER_filter_snapshots) &&
2339 (iter->flags & BTREE_ITER_intent) &&
2340 !(iter->flags & BTREE_ITER_is_extents) &&
2341 !iter->update_path) {
2342 struct bpos pos = k.k->p;
2343
2344 if (pos.snapshot < iter->snapshot) {
2345 search_key = bpos_successor(k.k->p);
2346 continue;
2347 }
2348
2349 pos.snapshot = iter->snapshot;
2350
2351 /*
2352 * advance, same as on exit for iter->path, but only up
2353 * to snapshot
2354 */
2355 __btree_path_get(trans, trans->paths + iter->path, iter->flags & BTREE_ITER_intent);
2356 iter->update_path = iter->path;
2357
2358 iter->update_path = bch2_btree_path_set_pos(trans,
2359 iter->update_path, pos,
2360 iter->flags & BTREE_ITER_intent,
2361 _THIS_IP_);
2362 ret = bch2_btree_path_traverse(trans, iter->update_path, iter->flags);
2363 if (unlikely(ret)) {
2364 k = bkey_s_c_err(ret);
2365 goto out_no_locked;
2366 }
2367 }
2368
2369 /*
2370 * We can never have a key in a leaf node at POS_MAX, so
2371 * we don't have to check these successor() calls:
2372 */
2373 if ((iter->flags & BTREE_ITER_filter_snapshots) &&
2374 !bch2_snapshot_is_ancestor(trans->c,
2375 iter->snapshot,
2376 k.k->p.snapshot)) {
2377 search_key = bpos_successor(k.k->p);
2378 continue;
2379 }
2380
2381 if (bkey_whiteout(k.k) &&
2382 !(iter->flags & BTREE_ITER_all_snapshots)) {
2383 search_key = bkey_successor(iter, k.k->p);
2384 continue;
2385 }
2386
2387 /*
2388 * iter->pos should be mononotically increasing, and always be
2389 * equal to the key we just returned - except extents can
2390 * straddle iter->pos:
2391 */
2392 if (!(iter->flags & BTREE_ITER_is_extents))
2393 iter_pos = k.k->p;
2394 else
2395 iter_pos = bkey_max(iter->pos, bkey_start_pos(k.k));
2396
2397 if (unlikely(iter->flags & BTREE_ITER_all_snapshots ? bpos_gt(iter_pos, end) :
2398 iter->flags & BTREE_ITER_is_extents ? bkey_ge(iter_pos, end) :
2399 bkey_gt(iter_pos, end)))
2400 goto end;
2401
2402 break;
2403 }
2404
2405 iter->pos = iter_pos;
2406
2407 iter->path = bch2_btree_path_set_pos(trans, iter->path, k.k->p,
2408 iter->flags & BTREE_ITER_intent,
2409 btree_iter_ip_allocated(iter));
2410
2411 btree_path_set_should_be_locked(trans, btree_iter_path(trans, iter));
2412 out_no_locked:
2413 if (iter->update_path) {
2414 ret = bch2_btree_path_relock(trans, trans->paths + iter->update_path, _THIS_IP_);
2415 if (unlikely(ret))
2416 k = bkey_s_c_err(ret);
2417 else
2418 btree_path_set_should_be_locked(trans, trans->paths + iter->update_path);
2419 }
2420
2421 if (!(iter->flags & BTREE_ITER_all_snapshots))
2422 iter->pos.snapshot = iter->snapshot;
2423
2424 ret = bch2_btree_iter_verify_ret(iter, k);
2425 if (unlikely(ret)) {
2426 bch2_btree_iter_set_pos(iter, iter->pos);
2427 k = bkey_s_c_err(ret);
2428 }
2429
2430 bch2_btree_iter_verify_entry_exit(iter);
2431
2432 return k;
2433 end:
2434 bch2_btree_iter_set_pos(iter, end);
2435 k = bkey_s_c_null;
2436 goto out_no_locked;
2437 }
2438
2439 /**
2440 * bch2_btree_iter_next() - returns first key greater than iterator's current
2441 * position
2442 * @iter: iterator to peek from
2443 *
2444 * Returns: key if found, or an error extractable with bkey_err().
2445 */
2446 struct bkey_s_c bch2_btree_iter_next(struct btree_iter *iter)
2447 {
2448 if (!bch2_btree_iter_advance(iter))
2449 return bkey_s_c_null;
2450
2451 return bch2_btree_iter_peek(iter);
2452 }
2453
2454 /**
2455 * bch2_btree_iter_peek_prev() - returns first key less than or equal to
2456 * iterator's current position
2457 * @iter: iterator to peek from
2458 *
2459 * Returns: key if found, or an error extractable with bkey_err().
2460 */
2461 struct bkey_s_c bch2_btree_iter_peek_prev(struct btree_iter *iter)
2462 {
2463 struct btree_trans *trans = iter->trans;
2464 struct bpos search_key = iter->pos;
2465 struct bkey_s_c k;
2466 struct bkey saved_k;
2467 const struct bch_val *saved_v;
2468 btree_path_idx_t saved_path = 0;
2469 int ret;
2470
2471 bch2_trans_verify_not_unlocked(trans);
2472 EBUG_ON(btree_iter_path(trans, iter)->cached ||
2473 btree_iter_path(trans, iter)->level);
2474
2475 if (iter->flags & BTREE_ITER_with_journal)
2476 return bkey_s_c_err(-BCH_ERR_btree_iter_with_journal_not_supported);
2477
2478 bch2_btree_iter_verify(iter);
2479 bch2_btree_iter_verify_entry_exit(iter);
2480
2481 if (iter->flags & BTREE_ITER_filter_snapshots)
2482 search_key.snapshot = U32_MAX;
2483
2484 while (1) {
2485 iter->path = bch2_btree_path_set_pos(trans, iter->path, search_key,
2486 iter->flags & BTREE_ITER_intent,
2487 btree_iter_ip_allocated(iter));
2488
2489 ret = bch2_btree_path_traverse(trans, iter->path, iter->flags);
2490 if (unlikely(ret)) {
2491 /* ensure that iter->k is consistent with iter->pos: */
2492 bch2_btree_iter_set_pos(iter, iter->pos);
2493 k = bkey_s_c_err(ret);
2494 goto out_no_locked;
2495 }
2496
2497 struct btree_path *path = btree_iter_path(trans, iter);
2498
2499 k = btree_path_level_peek(trans, path, &path->l[0], &iter->k);
2500 if (!k.k ||
2501 ((iter->flags & BTREE_ITER_is_extents)
2502 ? bpos_ge(bkey_start_pos(k.k), search_key)
2503 : bpos_gt(k.k->p, search_key)))
2504 k = btree_path_level_prev(trans, path, &path->l[0], &iter->k);
2505
2506 if (unlikely((iter->flags & BTREE_ITER_with_updates) &&
2507 trans->nr_updates))
2508 bch2_btree_trans_peek_prev_updates(trans, iter, &k);
2509
2510 if (likely(k.k)) {
2511 if (iter->flags & BTREE_ITER_filter_snapshots) {
2512 if (k.k->p.snapshot == iter->snapshot)
2513 goto got_key;
2514
2515 /*
2516 * If we have a saved candidate, and we're no
2517 * longer at the same _key_ (not pos), return
2518 * that candidate
2519 */
2520 if (saved_path && !bkey_eq(k.k->p, saved_k.p)) {
2521 bch2_path_put_nokeep(trans, iter->path,
2522 iter->flags & BTREE_ITER_intent);
2523 iter->path = saved_path;
2524 saved_path = 0;
2525 iter->k = saved_k;
2526 k.v = saved_v;
2527 goto got_key;
2528 }
2529
2530 if (bch2_snapshot_is_ancestor(trans->c,
2531 iter->snapshot,
2532 k.k->p.snapshot)) {
2533 if (saved_path)
2534 bch2_path_put_nokeep(trans, saved_path,
2535 iter->flags & BTREE_ITER_intent);
2536 saved_path = btree_path_clone(trans, iter->path,
2537 iter->flags & BTREE_ITER_intent,
2538 _THIS_IP_);
2539 path = btree_iter_path(trans, iter);
2540 trace_btree_path_save_pos(trans, path, trans->paths + saved_path);
2541 saved_k = *k.k;
2542 saved_v = k.v;
2543 }
2544
2545 search_key = bpos_predecessor(k.k->p);
2546 continue;
2547 }
2548 got_key:
2549 if (bkey_whiteout(k.k) &&
2550 !(iter->flags & BTREE_ITER_all_snapshots)) {
2551 search_key = bkey_predecessor(iter, k.k->p);
2552 if (iter->flags & BTREE_ITER_filter_snapshots)
2553 search_key.snapshot = U32_MAX;
2554 continue;
2555 }
2556
2557 btree_path_set_should_be_locked(trans, path);
2558 break;
2559 } else if (likely(!bpos_eq(path->l[0].b->data->min_key, POS_MIN))) {
2560 /* Advance to previous leaf node: */
2561 search_key = bpos_predecessor(path->l[0].b->data->min_key);
2562 } else {
2563 /* Start of btree: */
2564 bch2_btree_iter_set_pos(iter, POS_MIN);
2565 k = bkey_s_c_null;
2566 goto out_no_locked;
2567 }
2568 }
2569
2570 EBUG_ON(bkey_gt(bkey_start_pos(k.k), iter->pos));
2571
2572 /* Extents can straddle iter->pos: */
2573 if (bkey_lt(k.k->p, iter->pos))
2574 iter->pos = k.k->p;
2575
2576 if (iter->flags & BTREE_ITER_filter_snapshots)
2577 iter->pos.snapshot = iter->snapshot;
2578 out_no_locked:
2579 if (saved_path)
2580 bch2_path_put_nokeep(trans, saved_path, iter->flags & BTREE_ITER_intent);
2581
2582 bch2_btree_iter_verify_entry_exit(iter);
2583 bch2_btree_iter_verify(iter);
2584
2585 return k;
2586 }
2587
2588 /**
2589 * bch2_btree_iter_prev() - returns first key less than iterator's current
2590 * position
2591 * @iter: iterator to peek from
2592 *
2593 * Returns: key if found, or an error extractable with bkey_err().
2594 */
2595 struct bkey_s_c bch2_btree_iter_prev(struct btree_iter *iter)
2596 {
2597 if (!bch2_btree_iter_rewind(iter))
2598 return bkey_s_c_null;
2599
2600 return bch2_btree_iter_peek_prev(iter);
2601 }
2602
2603 struct bkey_s_c bch2_btree_iter_peek_slot(struct btree_iter *iter)
2604 {
2605 struct btree_trans *trans = iter->trans;
2606 struct bpos search_key;
2607 struct bkey_s_c k;
2608 int ret;
2609
2610 bch2_trans_verify_not_unlocked(trans);
2611 bch2_btree_iter_verify(iter);
2612 bch2_btree_iter_verify_entry_exit(iter);
2613 EBUG_ON(btree_iter_path(trans, iter)->level && (iter->flags & BTREE_ITER_with_key_cache));
2614
2615 /* extents can't span inode numbers: */
2616 if ((iter->flags & BTREE_ITER_is_extents) &&
2617 unlikely(iter->pos.offset == KEY_OFFSET_MAX)) {
2618 if (iter->pos.inode == KEY_INODE_MAX)
2619 return bkey_s_c_null;
2620
2621 bch2_btree_iter_set_pos(iter, bpos_nosnap_successor(iter->pos));
2622 }
2623
2624 search_key = btree_iter_search_key(iter);
2625 iter->path = bch2_btree_path_set_pos(trans, iter->path, search_key,
2626 iter->flags & BTREE_ITER_intent,
2627 btree_iter_ip_allocated(iter));
2628
2629 ret = bch2_btree_path_traverse(trans, iter->path, iter->flags);
2630 if (unlikely(ret)) {
2631 k = bkey_s_c_err(ret);
2632 goto out_no_locked;
2633 }
2634
2635 if ((iter->flags & BTREE_ITER_cached) ||
2636 !(iter->flags & (BTREE_ITER_is_extents|BTREE_ITER_filter_snapshots))) {
2637 k = bkey_s_c_null;
2638
2639 if (unlikely((iter->flags & BTREE_ITER_with_updates) &&
2640 trans->nr_updates)) {
2641 bch2_btree_trans_peek_slot_updates(trans, iter, &k);
2642 if (k.k)
2643 goto out;
2644 }
2645
2646 if (unlikely(iter->flags & BTREE_ITER_with_journal) &&
2647 (k = btree_trans_peek_slot_journal(trans, iter)).k)
2648 goto out;
2649
2650 if (unlikely(iter->flags & BTREE_ITER_with_key_cache) &&
2651 (k = btree_trans_peek_key_cache(iter, iter->pos)).k) {
2652 if (!bkey_err(k))
2653 iter->k = *k.k;
2654 /* We're not returning a key from iter->path: */
2655 goto out_no_locked;
2656 }
2657
2658 k = bch2_btree_path_peek_slot(trans->paths + iter->path, &iter->k);
2659 if (unlikely(!k.k))
2660 goto out_no_locked;
2661 } else {
2662 struct bpos next;
2663 struct bpos end = iter->pos;
2664
2665 if (iter->flags & BTREE_ITER_is_extents)
2666 end.offset = U64_MAX;
2667
2668 EBUG_ON(btree_iter_path(trans, iter)->level);
2669
2670 if (iter->flags & BTREE_ITER_intent) {
2671 struct btree_iter iter2;
2672
2673 bch2_trans_copy_iter(&iter2, iter);
2674 k = bch2_btree_iter_peek_upto(&iter2, end);
2675
2676 if (k.k && !bkey_err(k)) {
2677 swap(iter->key_cache_path, iter2.key_cache_path);
2678 iter->k = iter2.k;
2679 k.k = &iter->k;
2680 }
2681 bch2_trans_iter_exit(trans, &iter2);
2682 } else {
2683 struct bpos pos = iter->pos;
2684
2685 k = bch2_btree_iter_peek_upto(iter, end);
2686 if (unlikely(bkey_err(k)))
2687 bch2_btree_iter_set_pos(iter, pos);
2688 else
2689 iter->pos = pos;
2690 }
2691
2692 if (unlikely(bkey_err(k)))
2693 goto out_no_locked;
2694
2695 next = k.k ? bkey_start_pos(k.k) : POS_MAX;
2696
2697 if (bkey_lt(iter->pos, next)) {
2698 bkey_init(&iter->k);
2699 iter->k.p = iter->pos;
2700
2701 if (iter->flags & BTREE_ITER_is_extents) {
2702 bch2_key_resize(&iter->k,
2703 min_t(u64, KEY_SIZE_MAX,
2704 (next.inode == iter->pos.inode
2705 ? next.offset
2706 : KEY_OFFSET_MAX) -
2707 iter->pos.offset));
2708 EBUG_ON(!iter->k.size);
2709 }
2710
2711 k = (struct bkey_s_c) { &iter->k, NULL };
2712 }
2713 }
2714 out:
2715 btree_path_set_should_be_locked(trans, btree_iter_path(trans, iter));
2716 out_no_locked:
2717 bch2_btree_iter_verify_entry_exit(iter);
2718 bch2_btree_iter_verify(iter);
2719 ret = bch2_btree_iter_verify_ret(iter, k);
2720 if (unlikely(ret))
2721 return bkey_s_c_err(ret);
2722
2723 return k;
2724 }
2725
2726 struct bkey_s_c bch2_btree_iter_next_slot(struct btree_iter *iter)
2727 {
2728 if (!bch2_btree_iter_advance(iter))
2729 return bkey_s_c_null;
2730
2731 return bch2_btree_iter_peek_slot(iter);
2732 }
2733
2734 struct bkey_s_c bch2_btree_iter_prev_slot(struct btree_iter *iter)
2735 {
2736 if (!bch2_btree_iter_rewind(iter))
2737 return bkey_s_c_null;
2738
2739 return bch2_btree_iter_peek_slot(iter);
2740 }
2741
2742 /* Obsolete, but still used by rust wrapper in -tools */
2743 struct bkey_s_c bch2_btree_iter_peek_and_restart_outlined(struct btree_iter *iter)
2744 {
2745 struct bkey_s_c k;
2746
2747 while (btree_trans_too_many_iters(iter->trans) ||
2748 (k = bch2_btree_iter_peek_type(iter, iter->flags),
2749 bch2_err_matches(bkey_err(k), BCH_ERR_transaction_restart)))
2750 bch2_trans_begin(iter->trans);
2751
2752 return k;
2753 }
2754
2755 /* new transactional stuff: */
2756
2757 #ifdef CONFIG_BCACHEFS_DEBUG
2758 static void btree_trans_verify_sorted_refs(struct btree_trans *trans)
2759 {
2760 struct btree_path *path;
2761 unsigned i;
2762
2763 BUG_ON(trans->nr_sorted != bitmap_weight(trans->paths_allocated, trans->nr_paths) - 1);
2764
2765 trans_for_each_path(trans, path, i) {
2766 BUG_ON(path->sorted_idx >= trans->nr_sorted);
2767 BUG_ON(trans->sorted[path->sorted_idx] != i);
2768 }
2769
2770 for (i = 0; i < trans->nr_sorted; i++) {
2771 unsigned idx = trans->sorted[i];
2772
2773 BUG_ON(!test_bit(idx, trans->paths_allocated));
2774 BUG_ON(trans->paths[idx].sorted_idx != i);
2775 }
2776 }
2777
2778 static void btree_trans_verify_sorted(struct btree_trans *trans)
2779 {
2780 struct btree_path *path, *prev = NULL;
2781 struct trans_for_each_path_inorder_iter iter;
2782
2783 if (!bch2_debug_check_iterators)
2784 return;
2785
2786 trans_for_each_path_inorder(trans, path, iter) {
2787 if (prev && btree_path_cmp(prev, path) > 0) {
2788 __bch2_dump_trans_paths_updates(trans, true);
2789 panic("trans paths out of order!\n");
2790 }
2791 prev = path;
2792 }
2793 }
2794 #else
2795 static inline void btree_trans_verify_sorted_refs(struct btree_trans *trans) {}
2796 static inline void btree_trans_verify_sorted(struct btree_trans *trans) {}
2797 #endif
2798
2799 void __bch2_btree_trans_sort_paths(struct btree_trans *trans)
2800 {
2801 int i, l = 0, r = trans->nr_sorted, inc = 1;
2802 bool swapped;
2803
2804 btree_trans_verify_sorted_refs(trans);
2805
2806 if (trans->paths_sorted)
2807 goto out;
2808
2809 /*
2810 * Cocktail shaker sort: this is efficient because iterators will be
2811 * mostly sorted.
2812 */
2813 do {
2814 swapped = false;
2815
2816 for (i = inc > 0 ? l : r - 2;
2817 i + 1 < r && i >= l;
2818 i += inc) {
2819 if (btree_path_cmp(trans->paths + trans->sorted[i],
2820 trans->paths + trans->sorted[i + 1]) > 0) {
2821 swap(trans->sorted[i], trans->sorted[i + 1]);
2822 trans->paths[trans->sorted[i]].sorted_idx = i;
2823 trans->paths[trans->sorted[i + 1]].sorted_idx = i + 1;
2824 swapped = true;
2825 }
2826 }
2827
2828 if (inc > 0)
2829 --r;
2830 else
2831 l++;
2832 inc = -inc;
2833 } while (swapped);
2834
2835 trans->paths_sorted = true;
2836 out:
2837 btree_trans_verify_sorted(trans);
2838 }
2839
2840 static inline void btree_path_list_remove(struct btree_trans *trans,
2841 struct btree_path *path)
2842 {
2843 EBUG_ON(path->sorted_idx >= trans->nr_sorted);
2844 #ifdef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
2845 trans->nr_sorted--;
2846 memmove_u64s_down_small(trans->sorted + path->sorted_idx,
2847 trans->sorted + path->sorted_idx + 1,
2848 DIV_ROUND_UP(trans->nr_sorted - path->sorted_idx,
2849 sizeof(u64) / sizeof(btree_path_idx_t)));
2850 #else
2851 array_remove_item(trans->sorted, trans->nr_sorted, path->sorted_idx);
2852 #endif
2853 for (unsigned i = path->sorted_idx; i < trans->nr_sorted; i++)
2854 trans->paths[trans->sorted[i]].sorted_idx = i;
2855 }
2856
2857 static inline void btree_path_list_add(struct btree_trans *trans,
2858 btree_path_idx_t pos,
2859 btree_path_idx_t path_idx)
2860 {
2861 struct btree_path *path = trans->paths + path_idx;
2862
2863 path->sorted_idx = pos ? trans->paths[pos].sorted_idx + 1 : trans->nr_sorted;
2864
2865 #ifdef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
2866 memmove_u64s_up_small(trans->sorted + path->sorted_idx + 1,
2867 trans->sorted + path->sorted_idx,
2868 DIV_ROUND_UP(trans->nr_sorted - path->sorted_idx,
2869 sizeof(u64) / sizeof(btree_path_idx_t)));
2870 trans->nr_sorted++;
2871 trans->sorted[path->sorted_idx] = path_idx;
2872 #else
2873 array_insert_item(trans->sorted, trans->nr_sorted, path->sorted_idx, path_idx);
2874 #endif
2875
2876 for (unsigned i = path->sorted_idx; i < trans->nr_sorted; i++)
2877 trans->paths[trans->sorted[i]].sorted_idx = i;
2878
2879 btree_trans_verify_sorted_refs(trans);
2880 }
2881
2882 void bch2_trans_iter_exit(struct btree_trans *trans, struct btree_iter *iter)
2883 {
2884 if (iter->update_path)
2885 bch2_path_put_nokeep(trans, iter->update_path,
2886 iter->flags & BTREE_ITER_intent);
2887 if (iter->path)
2888 bch2_path_put(trans, iter->path,
2889 iter->flags & BTREE_ITER_intent);
2890 if (iter->key_cache_path)
2891 bch2_path_put(trans, iter->key_cache_path,
2892 iter->flags & BTREE_ITER_intent);
2893 iter->path = 0;
2894 iter->update_path = 0;
2895 iter->key_cache_path = 0;
2896 iter->trans = NULL;
2897 }
2898
2899 void bch2_trans_iter_init_outlined(struct btree_trans *trans,
2900 struct btree_iter *iter,
2901 enum btree_id btree_id, struct bpos pos,
2902 unsigned flags)
2903 {
2904 bch2_trans_iter_init_common(trans, iter, btree_id, pos, 0, 0,
2905 bch2_btree_iter_flags(trans, btree_id, flags),
2906 _RET_IP_);
2907 }
2908
2909 void bch2_trans_node_iter_init(struct btree_trans *trans,
2910 struct btree_iter *iter,
2911 enum btree_id btree_id,
2912 struct bpos pos,
2913 unsigned locks_want,
2914 unsigned depth,
2915 unsigned flags)
2916 {
2917 flags |= BTREE_ITER_not_extents;
2918 flags |= BTREE_ITER_snapshot_field;
2919 flags |= BTREE_ITER_all_snapshots;
2920
2921 bch2_trans_iter_init_common(trans, iter, btree_id, pos, locks_want, depth,
2922 __bch2_btree_iter_flags(trans, btree_id, flags),
2923 _RET_IP_);
2924
2925 iter->min_depth = depth;
2926
2927 struct btree_path *path = btree_iter_path(trans, iter);
2928 BUG_ON(path->locks_want < min(locks_want, BTREE_MAX_DEPTH));
2929 BUG_ON(path->level != depth);
2930 BUG_ON(iter->min_depth != depth);
2931 }
2932
2933 void bch2_trans_copy_iter(struct btree_iter *dst, struct btree_iter *src)
2934 {
2935 struct btree_trans *trans = src->trans;
2936
2937 *dst = *src;
2938 #ifdef TRACK_PATH_ALLOCATED
2939 dst->ip_allocated = _RET_IP_;
2940 #endif
2941 if (src->path)
2942 __btree_path_get(trans, trans->paths + src->path, src->flags & BTREE_ITER_intent);
2943 if (src->update_path)
2944 __btree_path_get(trans, trans->paths + src->update_path, src->flags & BTREE_ITER_intent);
2945 dst->key_cache_path = 0;
2946 }
2947
2948 void *__bch2_trans_kmalloc(struct btree_trans *trans, size_t size)
2949 {
2950 struct bch_fs *c = trans->c;
2951 unsigned new_top = trans->mem_top + size;
2952 unsigned old_bytes = trans->mem_bytes;
2953 unsigned new_bytes = roundup_pow_of_two(new_top);
2954 int ret;
2955 void *new_mem;
2956 void *p;
2957
2958 WARN_ON_ONCE(new_bytes > BTREE_TRANS_MEM_MAX);
2959
2960 struct btree_transaction_stats *s = btree_trans_stats(trans);
2961 s->max_mem = max(s->max_mem, new_bytes);
2962
2963 if (trans->used_mempool) {
2964 if (trans->mem_bytes >= new_bytes)
2965 goto out_change_top;
2966
2967 /* No more space from mempool item, need malloc new one */
2968 new_mem = kmalloc(new_bytes, GFP_NOWAIT|__GFP_NOWARN);
2969 if (unlikely(!new_mem)) {
2970 bch2_trans_unlock(trans);
2971
2972 new_mem = kmalloc(new_bytes, GFP_KERNEL);
2973 if (!new_mem)
2974 return ERR_PTR(-BCH_ERR_ENOMEM_trans_kmalloc);
2975
2976 ret = bch2_trans_relock(trans);
2977 if (ret) {
2978 kfree(new_mem);
2979 return ERR_PTR(ret);
2980 }
2981 }
2982 memcpy(new_mem, trans->mem, trans->mem_top);
2983 trans->used_mempool = false;
2984 mempool_free(trans->mem, &c->btree_trans_mem_pool);
2985 goto out_new_mem;
2986 }
2987
2988 new_mem = krealloc(trans->mem, new_bytes, GFP_NOWAIT|__GFP_NOWARN);
2989 if (unlikely(!new_mem)) {
2990 bch2_trans_unlock(trans);
2991
2992 new_mem = krealloc(trans->mem, new_bytes, GFP_KERNEL);
2993 if (!new_mem && new_bytes <= BTREE_TRANS_MEM_MAX) {
2994 new_mem = mempool_alloc(&c->btree_trans_mem_pool, GFP_KERNEL);
2995 new_bytes = BTREE_TRANS_MEM_MAX;
2996 memcpy(new_mem, trans->mem, trans->mem_top);
2997 trans->used_mempool = true;
2998 kfree(trans->mem);
2999 }
3000
3001 if (!new_mem)
3002 return ERR_PTR(-BCH_ERR_ENOMEM_trans_kmalloc);
3003
3004 trans->mem = new_mem;
3005 trans->mem_bytes = new_bytes;
3006
3007 ret = bch2_trans_relock(trans);
3008 if (ret)
3009 return ERR_PTR(ret);
3010 }
3011 out_new_mem:
3012 trans->mem = new_mem;
3013 trans->mem_bytes = new_bytes;
3014
3015 if (old_bytes) {
3016 trace_and_count(c, trans_restart_mem_realloced, trans, _RET_IP_, new_bytes);
3017 return ERR_PTR(btree_trans_restart(trans, BCH_ERR_transaction_restart_mem_realloced));
3018 }
3019 out_change_top:
3020 p = trans->mem + trans->mem_top;
3021 trans->mem_top += size;
3022 memset(p, 0, size);
3023 return p;
3024 }
3025
3026 static inline void check_srcu_held_too_long(struct btree_trans *trans)
3027 {
3028 WARN(trans->srcu_held && time_after(jiffies, trans->srcu_lock_time + HZ * 10),
3029 "btree trans held srcu lock (delaying memory reclaim) for %lu seconds",
3030 (jiffies - trans->srcu_lock_time) / HZ);
3031 }
3032
3033 void bch2_trans_srcu_unlock(struct btree_trans *trans)
3034 {
3035 if (trans->srcu_held) {
3036 struct bch_fs *c = trans->c;
3037 struct btree_path *path;
3038 unsigned i;
3039
3040 trans_for_each_path(trans, path, i)
3041 if (path->cached && !btree_node_locked(path, 0))
3042 path->l[0].b = ERR_PTR(-BCH_ERR_no_btree_node_srcu_reset);
3043
3044 check_srcu_held_too_long(trans);
3045 srcu_read_unlock(&c->btree_trans_barrier, trans->srcu_idx);
3046 trans->srcu_held = false;
3047 }
3048 }
3049
3050 static void bch2_trans_srcu_lock(struct btree_trans *trans)
3051 {
3052 if (!trans->srcu_held) {
3053 trans->srcu_idx = srcu_read_lock(&trans->c->btree_trans_barrier);
3054 trans->srcu_lock_time = jiffies;
3055 trans->srcu_held = true;
3056 }
3057 }
3058
3059 /**
3060 * bch2_trans_begin() - reset a transaction after a interrupted attempt
3061 * @trans: transaction to reset
3062 *
3063 * Returns: current restart counter, to be used with trans_was_restarted()
3064 *
3065 * While iterating over nodes or updating nodes a attempt to lock a btree node
3066 * may return BCH_ERR_transaction_restart when the trylock fails. When this
3067 * occurs bch2_trans_begin() should be called and the transaction retried.
3068 */
3069 u32 bch2_trans_begin(struct btree_trans *trans)
3070 {
3071 struct btree_path *path;
3072 unsigned i;
3073 u64 now;
3074
3075 bch2_trans_reset_updates(trans);
3076
3077 trans->restart_count++;
3078 trans->mem_top = 0;
3079 trans->journal_entries = NULL;
3080
3081 trans_for_each_path(trans, path, i) {
3082 path->should_be_locked = false;
3083
3084 /*
3085 * If the transaction wasn't restarted, we're presuming to be
3086 * doing something new: dont keep iterators excpt the ones that
3087 * are in use - except for the subvolumes btree:
3088 */
3089 if (!trans->restarted && path->btree_id != BTREE_ID_subvolumes)
3090 path->preserve = false;
3091
3092 /*
3093 * XXX: we probably shouldn't be doing this if the transaction
3094 * was restarted, but currently we still overflow transaction
3095 * iterators if we do that
3096 */
3097 if (!path->ref && !path->preserve)
3098 __bch2_path_free(trans, i);
3099 else
3100 path->preserve = false;
3101 }
3102
3103 now = local_clock();
3104
3105 if (!IS_ENABLED(CONFIG_BCACHEFS_NO_LATENCY_ACCT) &&
3106 time_after64(now, trans->last_begin_time + 10))
3107 __bch2_time_stats_update(&btree_trans_stats(trans)->duration,
3108 trans->last_begin_time, now);
3109
3110 if (!trans->restarted &&
3111 (need_resched() ||
3112 time_after64(now, trans->last_begin_time + BTREE_TRANS_MAX_LOCK_HOLD_TIME_NS))) {
3113 bch2_trans_unlock(trans);
3114 cond_resched();
3115 now = local_clock();
3116 }
3117 trans->last_begin_time = now;
3118
3119 if (unlikely(trans->srcu_held &&
3120 time_after(jiffies, trans->srcu_lock_time + msecs_to_jiffies(10))))
3121 bch2_trans_srcu_unlock(trans);
3122
3123 trans->last_begin_ip = _RET_IP_;
3124
3125 trans_set_locked(trans);
3126
3127 if (trans->restarted) {
3128 bch2_btree_path_traverse_all(trans);
3129 trans->notrace_relock_fail = false;
3130 }
3131
3132 bch2_trans_verify_not_unlocked(trans);
3133 return trans->restart_count;
3134 }
3135
3136 const char *bch2_btree_transaction_fns[BCH_TRANSACTIONS_NR] = { "(unknown)" };
3137
3138 unsigned bch2_trans_get_fn_idx(const char *fn)
3139 {
3140 for (unsigned i = 0; i < ARRAY_SIZE(bch2_btree_transaction_fns); i++)
3141 if (!bch2_btree_transaction_fns[i] ||
3142 bch2_btree_transaction_fns[i] == fn) {
3143 bch2_btree_transaction_fns[i] = fn;
3144 return i;
3145 }
3146
3147 pr_warn_once("BCH_TRANSACTIONS_NR not big enough!");
3148 return 0;
3149 }
3150
3151 struct btree_trans *__bch2_trans_get(struct bch_fs *c, unsigned fn_idx)
3152 __acquires(&c->btree_trans_barrier)
3153 {
3154 struct btree_trans *trans;
3155
3156 if (IS_ENABLED(__KERNEL__)) {
3157 trans = this_cpu_xchg(c->btree_trans_bufs->trans, NULL);
3158 if (trans) {
3159 memset(trans, 0, offsetof(struct btree_trans, list));
3160 goto got_trans;
3161 }
3162 }
3163
3164 trans = mempool_alloc(&c->btree_trans_pool, GFP_NOFS);
3165 memset(trans, 0, sizeof(*trans));
3166
3167 seqmutex_lock(&c->btree_trans_lock);
3168 if (IS_ENABLED(CONFIG_BCACHEFS_DEBUG)) {
3169 struct btree_trans *pos;
3170 pid_t pid = current->pid;
3171
3172 trans->locking_wait.task = current;
3173
3174 list_for_each_entry(pos, &c->btree_trans_list, list) {
3175 struct task_struct *pos_task = READ_ONCE(pos->locking_wait.task);
3176 /*
3177 * We'd much prefer to be stricter here and completely
3178 * disallow multiple btree_trans in the same thread -
3179 * but the data move path calls bch2_write when we
3180 * already have a btree_trans initialized.
3181 */
3182 BUG_ON(pos_task &&
3183 pid == pos_task->pid &&
3184 pos->locked);
3185 }
3186 }
3187
3188 list_add(&trans->list, &c->btree_trans_list);
3189 seqmutex_unlock(&c->btree_trans_lock);
3190 got_trans:
3191 trans->c = c;
3192 trans->last_begin_time = local_clock();
3193 trans->fn_idx = fn_idx;
3194 trans->locking_wait.task = current;
3195 trans->journal_replay_not_finished =
3196 unlikely(!test_bit(JOURNAL_replay_done, &c->journal.flags)) &&
3197 atomic_inc_not_zero(&c->journal_keys.ref);
3198 trans->nr_paths = ARRAY_SIZE(trans->_paths);
3199 trans->paths_allocated = trans->_paths_allocated;
3200 trans->sorted = trans->_sorted;
3201 trans->paths = trans->_paths;
3202 trans->updates = trans->_updates;
3203
3204 *trans_paths_nr(trans->paths) = BTREE_ITER_INITIAL;
3205
3206 trans->paths_allocated[0] = 1;
3207
3208 static struct lock_class_key lockdep_key;
3209 lockdep_init_map(&trans->dep_map, "bcachefs_btree", &lockdep_key, 0);
3210
3211 if (fn_idx < BCH_TRANSACTIONS_NR) {
3212 trans->fn = bch2_btree_transaction_fns[fn_idx];
3213
3214 struct btree_transaction_stats *s = &c->btree_transaction_stats[fn_idx];
3215
3216 if (s->max_mem) {
3217 unsigned expected_mem_bytes = roundup_pow_of_two(s->max_mem);
3218
3219 trans->mem = kmalloc(expected_mem_bytes, GFP_KERNEL);
3220 if (likely(trans->mem))
3221 trans->mem_bytes = expected_mem_bytes;
3222 }
3223
3224 trans->nr_paths_max = s->nr_max_paths;
3225 trans->journal_entries_size = s->journal_entries_size;
3226 }
3227
3228 trans->srcu_idx = srcu_read_lock(&c->btree_trans_barrier);
3229 trans->srcu_lock_time = jiffies;
3230 trans->srcu_held = true;
3231 trans_set_locked(trans);
3232
3233 closure_init_stack_release(&trans->ref);
3234 return trans;
3235 }
3236
3237 static void check_btree_paths_leaked(struct btree_trans *trans)
3238 {
3239 #ifdef CONFIG_BCACHEFS_DEBUG
3240 struct bch_fs *c = trans->c;
3241 struct btree_path *path;
3242 unsigned i;
3243
3244 trans_for_each_path(trans, path, i)
3245 if (path->ref)
3246 goto leaked;
3247 return;
3248 leaked:
3249 bch_err(c, "btree paths leaked from %s!", trans->fn);
3250 trans_for_each_path(trans, path, i)
3251 if (path->ref)
3252 printk(KERN_ERR " btree %s %pS\n",
3253 bch2_btree_id_str(path->btree_id),
3254 (void *) path->ip_allocated);
3255 /* Be noisy about this: */
3256 bch2_fatal_error(c);
3257 #endif
3258 }
3259
3260 void bch2_trans_put(struct btree_trans *trans)
3261 __releases(&c->btree_trans_barrier)
3262 {
3263 struct bch_fs *c = trans->c;
3264
3265 bch2_trans_unlock(trans);
3266
3267 trans_for_each_update(trans, i)
3268 __btree_path_put(trans, trans->paths + i->path, true);
3269 trans->nr_updates = 0;
3270
3271 check_btree_paths_leaked(trans);
3272
3273 if (trans->srcu_held) {
3274 check_srcu_held_too_long(trans);
3275 srcu_read_unlock(&c->btree_trans_barrier, trans->srcu_idx);
3276 }
3277
3278 if (unlikely(trans->journal_replay_not_finished))
3279 bch2_journal_keys_put(c);
3280
3281 /*
3282 * trans->ref protects trans->locking_wait.task, btree_paths array; used
3283 * by cycle detector
3284 */
3285 closure_return_sync(&trans->ref);
3286 trans->locking_wait.task = NULL;
3287
3288 unsigned long *paths_allocated = trans->paths_allocated;
3289 trans->paths_allocated = NULL;
3290 trans->paths = NULL;
3291
3292 if (paths_allocated != trans->_paths_allocated)
3293 kvfree_rcu_mightsleep(paths_allocated);
3294
3295 if (trans->used_mempool)
3296 mempool_free(trans->mem, &c->btree_trans_mem_pool);
3297 else
3298 kfree(trans->mem);
3299
3300 /* Userspace doesn't have a real percpu implementation: */
3301 if (IS_ENABLED(__KERNEL__))
3302 trans = this_cpu_xchg(c->btree_trans_bufs->trans, trans);
3303
3304 if (trans) {
3305 seqmutex_lock(&c->btree_trans_lock);
3306 list_del(&trans->list);
3307 seqmutex_unlock(&c->btree_trans_lock);
3308
3309 mempool_free(trans, &c->btree_trans_pool);
3310 }
3311 }
3312
3313 bool bch2_current_has_btree_trans(struct bch_fs *c)
3314 {
3315 seqmutex_lock(&c->btree_trans_lock);
3316 struct btree_trans *trans;
3317 bool ret = false;
3318 list_for_each_entry(trans, &c->btree_trans_list, list)
3319 if (trans->locking_wait.task == current &&
3320 trans->locked) {
3321 ret = true;
3322 break;
3323 }
3324 seqmutex_unlock(&c->btree_trans_lock);
3325 return ret;
3326 }
3327
3328 static void __maybe_unused
3329 bch2_btree_bkey_cached_common_to_text(struct printbuf *out,
3330 struct btree_bkey_cached_common *b)
3331 {
3332 struct six_lock_count c = six_lock_counts(&b->lock);
3333 struct task_struct *owner;
3334 pid_t pid;
3335
3336 rcu_read_lock();
3337 owner = READ_ONCE(b->lock.owner);
3338 pid = owner ? owner->pid : 0;
3339 rcu_read_unlock();
3340
3341 prt_printf(out, "\t%px %c l=%u %s:", b, b->cached ? 'c' : 'b',
3342 b->level, bch2_btree_id_str(b->btree_id));
3343 bch2_bpos_to_text(out, btree_node_pos(b));
3344
3345 prt_printf(out, "\t locks %u:%u:%u held by pid %u",
3346 c.n[0], c.n[1], c.n[2], pid);
3347 }
3348
3349 void bch2_btree_trans_to_text(struct printbuf *out, struct btree_trans *trans)
3350 {
3351 struct btree_bkey_cached_common *b;
3352 static char lock_types[] = { 'r', 'i', 'w' };
3353 struct task_struct *task = READ_ONCE(trans->locking_wait.task);
3354 unsigned l, idx;
3355
3356 /* before rcu_read_lock(): */
3357 bch2_printbuf_make_room(out, 4096);
3358
3359 if (!out->nr_tabstops) {
3360 printbuf_tabstop_push(out, 16);
3361 printbuf_tabstop_push(out, 32);
3362 }
3363
3364 prt_printf(out, "%i %s\n", task ? task->pid : 0, trans->fn);
3365
3366 /* trans->paths is rcu protected vs. freeing */
3367 rcu_read_lock();
3368 out->atomic++;
3369
3370 struct btree_path *paths = rcu_dereference(trans->paths);
3371 if (!paths)
3372 goto out;
3373
3374 unsigned long *paths_allocated = trans_paths_allocated(paths);
3375
3376 trans_for_each_path_idx_from(paths_allocated, *trans_paths_nr(paths), idx, 1) {
3377 struct btree_path *path = paths + idx;
3378 if (!path->nodes_locked)
3379 continue;
3380
3381 prt_printf(out, " path %u %c l=%u %s:",
3382 idx,
3383 path->cached ? 'c' : 'b',
3384 path->level,
3385 bch2_btree_id_str(path->btree_id));
3386 bch2_bpos_to_text(out, path->pos);
3387 prt_newline(out);
3388
3389 for (l = 0; l < BTREE_MAX_DEPTH; l++) {
3390 if (btree_node_locked(path, l) &&
3391 !IS_ERR_OR_NULL(b = (void *) READ_ONCE(path->l[l].b))) {
3392 prt_printf(out, " %c l=%u ",
3393 lock_types[btree_node_locked_type(path, l)], l);
3394 bch2_btree_bkey_cached_common_to_text(out, b);
3395 prt_newline(out);
3396 }
3397 }
3398 }
3399
3400 b = READ_ONCE(trans->locking);
3401 if (b) {
3402 prt_printf(out, " blocked for %lluus on\n",
3403 div_u64(local_clock() - trans->locking_wait.start_time, 1000));
3404 prt_printf(out, " %c", lock_types[trans->locking_wait.lock_want]);
3405 bch2_btree_bkey_cached_common_to_text(out, b);
3406 prt_newline(out);
3407 }
3408 out:
3409 --out->atomic;
3410 rcu_read_unlock();
3411 }
3412
3413 void bch2_fs_btree_iter_exit(struct bch_fs *c)
3414 {
3415 struct btree_transaction_stats *s;
3416 struct btree_trans *trans;
3417 int cpu;
3418
3419 if (c->btree_trans_bufs)
3420 for_each_possible_cpu(cpu) {
3421 struct btree_trans *trans =
3422 per_cpu_ptr(c->btree_trans_bufs, cpu)->trans;
3423
3424 if (trans) {
3425 seqmutex_lock(&c->btree_trans_lock);
3426 list_del(&trans->list);
3427 seqmutex_unlock(&c->btree_trans_lock);
3428 }
3429 kfree(trans);
3430 }
3431 free_percpu(c->btree_trans_bufs);
3432
3433 trans = list_first_entry_or_null(&c->btree_trans_list, struct btree_trans, list);
3434 if (trans)
3435 panic("%s leaked btree_trans\n", trans->fn);
3436
3437 for (s = c->btree_transaction_stats;
3438 s < c->btree_transaction_stats + ARRAY_SIZE(c->btree_transaction_stats);
3439 s++) {
3440 kfree(s->max_paths_text);
3441 bch2_time_stats_exit(&s->lock_hold_times);
3442 }
3443
3444 if (c->btree_trans_barrier_initialized) {
3445 synchronize_srcu_expedited(&c->btree_trans_barrier);
3446 cleanup_srcu_struct(&c->btree_trans_barrier);
3447 }
3448 mempool_exit(&c->btree_trans_mem_pool);
3449 mempool_exit(&c->btree_trans_pool);
3450 }
3451
3452 void bch2_fs_btree_iter_init_early(struct bch_fs *c)
3453 {
3454 struct btree_transaction_stats *s;
3455
3456 for (s = c->btree_transaction_stats;
3457 s < c->btree_transaction_stats + ARRAY_SIZE(c->btree_transaction_stats);
3458 s++) {
3459 bch2_time_stats_init(&s->duration);
3460 bch2_time_stats_init(&s->lock_hold_times);
3461 mutex_init(&s->lock);
3462 }
3463
3464 INIT_LIST_HEAD(&c->btree_trans_list);
3465 seqmutex_init(&c->btree_trans_lock);
3466 }
3467
3468 int bch2_fs_btree_iter_init(struct bch_fs *c)
3469 {
3470 int ret;
3471
3472 c->btree_trans_bufs = alloc_percpu(struct btree_trans_buf);
3473 if (!c->btree_trans_bufs)
3474 return -ENOMEM;
3475
3476 ret = mempool_init_kmalloc_pool(&c->btree_trans_pool, 1,
3477 sizeof(struct btree_trans)) ?:
3478 mempool_init_kmalloc_pool(&c->btree_trans_mem_pool, 1,
3479 BTREE_TRANS_MEM_MAX) ?:
3480 init_srcu_struct(&c->btree_trans_barrier);
3481 if (ret)
3482 return ret;
3483
3484 /*
3485 * static annotation (hackily done) for lock ordering of reclaim vs.
3486 * btree node locks:
3487 */
3488 #ifdef CONFIG_LOCKDEP
3489 fs_reclaim_acquire(GFP_KERNEL);
3490 struct btree_trans *trans = bch2_trans_get(c);
3491 trans_set_locked(trans);
3492 bch2_trans_put(trans);
3493 fs_reclaim_release(GFP_KERNEL);
3494 #endif
3495
3496 c->btree_trans_barrier_initialized = true;
3497 return 0;
3498
3499 }
Linux Kernel