xref: /linux/fs/bcachefs/btree_locking.c (revision 71dfa617ea9f18e4585fe78364217cd32b1fc382)
1 // SPDX-License-Identifier: GPL-2.0
2 
3 #include "bcachefs.h"
4 #include "btree_locking.h"
5 #include "btree_types.h"
6 
7 static struct lock_class_key bch2_btree_node_lock_key;
8 
9 void bch2_btree_lock_init(struct btree_bkey_cached_common *b,
10 			  enum six_lock_init_flags flags)
11 {
12 	__six_lock_init(&b->lock, "b->c.lock", &bch2_btree_node_lock_key, flags);
13 	lockdep_set_novalidate_class(&b->lock);
14 }
15 
16 #ifdef CONFIG_LOCKDEP
17 void bch2_assert_btree_nodes_not_locked(void)
18 {
19 #if 0
20 	//Re-enable when lock_class_is_held() is merged:
21 	BUG_ON(lock_class_is_held(&bch2_btree_node_lock_key));
22 #endif
23 }
24 #endif
25 
26 /* Btree node locking: */
27 
28 struct six_lock_count bch2_btree_node_lock_counts(struct btree_trans *trans,
29 						  struct btree_path *skip,
30 						  struct btree_bkey_cached_common *b,
31 						  unsigned level)
32 {
33 	struct btree_path *path;
34 	struct six_lock_count ret;
35 	unsigned i;
36 
37 	memset(&ret, 0, sizeof(ret));
38 
39 	if (IS_ERR_OR_NULL(b))
40 		return ret;
41 
42 	trans_for_each_path(trans, path, i)
43 		if (path != skip && &path->l[level].b->c == b) {
44 			int t = btree_node_locked_type(path, level);
45 
46 			if (t != BTREE_NODE_UNLOCKED)
47 				ret.n[t]++;
48 		}
49 
50 	return ret;
51 }
52 
53 /* unlock */
54 
55 void bch2_btree_node_unlock_write(struct btree_trans *trans,
56 			struct btree_path *path, struct btree *b)
57 {
58 	bch2_btree_node_unlock_write_inlined(trans, path, b);
59 }
60 
61 /* lock */
62 
63 /*
64  * @trans wants to lock @b with type @type
65  */
66 struct trans_waiting_for_lock {
67 	struct btree_trans		*trans;
68 	struct btree_bkey_cached_common	*node_want;
69 	enum six_lock_type		lock_want;
70 
71 	/* for iterating over held locks :*/
72 	u8				path_idx;
73 	u8				level;
74 	u64				lock_start_time;
75 };
76 
77 struct lock_graph {
78 	struct trans_waiting_for_lock	g[8];
79 	unsigned			nr;
80 };
81 
82 static noinline void print_cycle(struct printbuf *out, struct lock_graph *g)
83 {
84 	struct trans_waiting_for_lock *i;
85 
86 	prt_printf(out, "Found lock cycle (%u entries):", g->nr);
87 	prt_newline(out);
88 
89 	for (i = g->g; i < g->g + g->nr; i++) {
90 		struct task_struct *task = READ_ONCE(i->trans->locking_wait.task);
91 		if (!task)
92 			continue;
93 
94 		bch2_btree_trans_to_text(out, i->trans);
95 		bch2_prt_task_backtrace(out, task, i == g->g ? 5 : 1, GFP_NOWAIT);
96 	}
97 }
98 
99 static noinline void print_chain(struct printbuf *out, struct lock_graph *g)
100 {
101 	struct trans_waiting_for_lock *i;
102 
103 	for (i = g->g; i != g->g + g->nr; i++) {
104 		struct task_struct *task = i->trans->locking_wait.task;
105 		if (i != g->g)
106 			prt_str(out, "<- ");
107 		prt_printf(out, "%u ", task ?task->pid : 0);
108 	}
109 	prt_newline(out);
110 }
111 
112 static void lock_graph_up(struct lock_graph *g)
113 {
114 	closure_put(&g->g[--g->nr].trans->ref);
115 }
116 
117 static noinline void lock_graph_pop_all(struct lock_graph *g)
118 {
119 	while (g->nr)
120 		lock_graph_up(g);
121 }
122 
123 static void __lock_graph_down(struct lock_graph *g, struct btree_trans *trans)
124 {
125 	g->g[g->nr++] = (struct trans_waiting_for_lock) {
126 		.trans		= trans,
127 		.node_want	= trans->locking,
128 		.lock_want	= trans->locking_wait.lock_want,
129 	};
130 }
131 
132 static void lock_graph_down(struct lock_graph *g, struct btree_trans *trans)
133 {
134 	closure_get(&trans->ref);
135 	__lock_graph_down(g, trans);
136 }
137 
138 static bool lock_graph_remove_non_waiters(struct lock_graph *g)
139 {
140 	struct trans_waiting_for_lock *i;
141 
142 	for (i = g->g + 1; i < g->g + g->nr; i++)
143 		if (i->trans->locking != i->node_want ||
144 		    i->trans->locking_wait.start_time != i[-1].lock_start_time) {
145 			while (g->g + g->nr > i)
146 				lock_graph_up(g);
147 			return true;
148 		}
149 
150 	return false;
151 }
152 
153 static void trace_would_deadlock(struct lock_graph *g, struct btree_trans *trans)
154 {
155 	struct bch_fs *c = trans->c;
156 
157 	count_event(c, trans_restart_would_deadlock);
158 
159 	if (trace_trans_restart_would_deadlock_enabled()) {
160 		struct printbuf buf = PRINTBUF;
161 
162 		buf.atomic++;
163 		print_cycle(&buf, g);
164 
165 		trace_trans_restart_would_deadlock(trans, buf.buf);
166 		printbuf_exit(&buf);
167 	}
168 }
169 
170 static int abort_lock(struct lock_graph *g, struct trans_waiting_for_lock *i)
171 {
172 	if (i == g->g) {
173 		trace_would_deadlock(g, i->trans);
174 		return btree_trans_restart(i->trans, BCH_ERR_transaction_restart_would_deadlock);
175 	} else {
176 		i->trans->lock_must_abort = true;
177 		wake_up_process(i->trans->locking_wait.task);
178 		return 0;
179 	}
180 }
181 
182 static int btree_trans_abort_preference(struct btree_trans *trans)
183 {
184 	if (trans->lock_may_not_fail)
185 		return 0;
186 	if (trans->locking_wait.lock_want == SIX_LOCK_write)
187 		return 1;
188 	if (!trans->in_traverse_all)
189 		return 2;
190 	return 3;
191 }
192 
193 static noinline int break_cycle(struct lock_graph *g, struct printbuf *cycle)
194 {
195 	struct trans_waiting_for_lock *i, *abort = NULL;
196 	unsigned best = 0, pref;
197 	int ret;
198 
199 	if (lock_graph_remove_non_waiters(g))
200 		return 0;
201 
202 	/* Only checking, for debugfs: */
203 	if (cycle) {
204 		print_cycle(cycle, g);
205 		ret = -1;
206 		goto out;
207 	}
208 
209 	for (i = g->g; i < g->g + g->nr; i++) {
210 		pref = btree_trans_abort_preference(i->trans);
211 		if (pref > best) {
212 			abort = i;
213 			best = pref;
214 		}
215 	}
216 
217 	if (unlikely(!best)) {
218 		struct printbuf buf = PRINTBUF;
219 
220 		prt_printf(&buf, bch2_fmt(g->g->trans->c, "cycle of nofail locks"));
221 
222 		for (i = g->g; i < g->g + g->nr; i++) {
223 			struct btree_trans *trans = i->trans;
224 
225 			bch2_btree_trans_to_text(&buf, trans);
226 
227 			prt_printf(&buf, "backtrace:");
228 			prt_newline(&buf);
229 			printbuf_indent_add(&buf, 2);
230 			bch2_prt_task_backtrace(&buf, trans->locking_wait.task, 2, GFP_NOWAIT);
231 			printbuf_indent_sub(&buf, 2);
232 			prt_newline(&buf);
233 		}
234 
235 		bch2_print_string_as_lines(KERN_ERR, buf.buf);
236 		printbuf_exit(&buf);
237 		BUG();
238 	}
239 
240 	ret = abort_lock(g, abort);
241 out:
242 	if (ret)
243 		while (g->nr)
244 			lock_graph_up(g);
245 	return ret;
246 }
247 
248 static int lock_graph_descend(struct lock_graph *g, struct btree_trans *trans,
249 			      struct printbuf *cycle)
250 {
251 	struct btree_trans *orig_trans = g->g->trans;
252 	struct trans_waiting_for_lock *i;
253 
254 	for (i = g->g; i < g->g + g->nr; i++)
255 		if (i->trans == trans) {
256 			closure_put(&trans->ref);
257 			return break_cycle(g, cycle);
258 		}
259 
260 	if (g->nr == ARRAY_SIZE(g->g)) {
261 		closure_put(&trans->ref);
262 
263 		if (orig_trans->lock_may_not_fail)
264 			return 0;
265 
266 		while (g->nr)
267 			lock_graph_up(g);
268 
269 		if (cycle)
270 			return 0;
271 
272 		trace_and_count(trans->c, trans_restart_would_deadlock_recursion_limit, trans, _RET_IP_);
273 		return btree_trans_restart(orig_trans, BCH_ERR_transaction_restart_deadlock_recursion_limit);
274 	}
275 
276 	__lock_graph_down(g, trans);
277 	return 0;
278 }
279 
280 static bool lock_type_conflicts(enum six_lock_type t1, enum six_lock_type t2)
281 {
282 	return t1 + t2 > 1;
283 }
284 
285 int bch2_check_for_deadlock(struct btree_trans *trans, struct printbuf *cycle)
286 {
287 	struct lock_graph g;
288 	struct trans_waiting_for_lock *top;
289 	struct btree_bkey_cached_common *b;
290 	btree_path_idx_t path_idx;
291 	int ret = 0;
292 
293 	g.nr = 0;
294 
295 	if (trans->lock_must_abort) {
296 		if (cycle)
297 			return -1;
298 
299 		trace_would_deadlock(&g, trans);
300 		return btree_trans_restart(trans, BCH_ERR_transaction_restart_would_deadlock);
301 	}
302 
303 	lock_graph_down(&g, trans);
304 
305 	/* trans->paths is rcu protected vs. freeing */
306 	rcu_read_lock();
307 	if (cycle)
308 		cycle->atomic++;
309 next:
310 	if (!g.nr)
311 		goto out;
312 
313 	top = &g.g[g.nr - 1];
314 
315 	struct btree_path *paths = rcu_dereference(top->trans->paths);
316 	if (!paths)
317 		goto up;
318 
319 	unsigned long *paths_allocated = trans_paths_allocated(paths);
320 
321 	trans_for_each_path_idx_from(paths_allocated, *trans_paths_nr(paths),
322 				     path_idx, top->path_idx) {
323 		struct btree_path *path = paths + path_idx;
324 		if (!path->nodes_locked)
325 			continue;
326 
327 		if (path_idx != top->path_idx) {
328 			top->path_idx		= path_idx;
329 			top->level		= 0;
330 			top->lock_start_time	= 0;
331 		}
332 
333 		for (;
334 		     top->level < BTREE_MAX_DEPTH;
335 		     top->level++, top->lock_start_time = 0) {
336 			int lock_held = btree_node_locked_type(path, top->level);
337 
338 			if (lock_held == BTREE_NODE_UNLOCKED)
339 				continue;
340 
341 			b = &READ_ONCE(path->l[top->level].b)->c;
342 
343 			if (IS_ERR_OR_NULL(b)) {
344 				/*
345 				 * If we get here, it means we raced with the
346 				 * other thread updating its btree_path
347 				 * structures - which means it can't be blocked
348 				 * waiting on a lock:
349 				 */
350 				if (!lock_graph_remove_non_waiters(&g)) {
351 					/*
352 					 * If lock_graph_remove_non_waiters()
353 					 * didn't do anything, it must be
354 					 * because we're being called by debugfs
355 					 * checking for lock cycles, which
356 					 * invokes us on btree_transactions that
357 					 * aren't actually waiting on anything.
358 					 * Just bail out:
359 					 */
360 					lock_graph_pop_all(&g);
361 				}
362 
363 				goto next;
364 			}
365 
366 			if (list_empty_careful(&b->lock.wait_list))
367 				continue;
368 
369 			raw_spin_lock(&b->lock.wait_lock);
370 			list_for_each_entry(trans, &b->lock.wait_list, locking_wait.list) {
371 				BUG_ON(b != trans->locking);
372 
373 				if (top->lock_start_time &&
374 				    time_after_eq64(top->lock_start_time, trans->locking_wait.start_time))
375 					continue;
376 
377 				top->lock_start_time = trans->locking_wait.start_time;
378 
379 				/* Don't check for self deadlock: */
380 				if (trans == top->trans ||
381 				    !lock_type_conflicts(lock_held, trans->locking_wait.lock_want))
382 					continue;
383 
384 				closure_get(&trans->ref);
385 				raw_spin_unlock(&b->lock.wait_lock);
386 
387 				ret = lock_graph_descend(&g, trans, cycle);
388 				if (ret)
389 					goto out;
390 				goto next;
391 
392 			}
393 			raw_spin_unlock(&b->lock.wait_lock);
394 		}
395 	}
396 up:
397 	if (g.nr > 1 && cycle)
398 		print_chain(cycle, &g);
399 	lock_graph_up(&g);
400 	goto next;
401 out:
402 	if (cycle)
403 		--cycle->atomic;
404 	rcu_read_unlock();
405 	return ret;
406 }
407 
408 int bch2_six_check_for_deadlock(struct six_lock *lock, void *p)
409 {
410 	struct btree_trans *trans = p;
411 
412 	return bch2_check_for_deadlock(trans, NULL);
413 }
414 
415 int __bch2_btree_node_lock_write(struct btree_trans *trans, struct btree_path *path,
416 				 struct btree_bkey_cached_common *b,
417 				 bool lock_may_not_fail)
418 {
419 	int readers = bch2_btree_node_lock_counts(trans, NULL, b, b->level).n[SIX_LOCK_read];
420 	int ret;
421 
422 	/*
423 	 * Must drop our read locks before calling six_lock_write() -
424 	 * six_unlock() won't do wakeups until the reader count
425 	 * goes to 0, and it's safe because we have the node intent
426 	 * locked:
427 	 */
428 	six_lock_readers_add(&b->lock, -readers);
429 	ret = __btree_node_lock_nopath(trans, b, SIX_LOCK_write,
430 				       lock_may_not_fail, _RET_IP_);
431 	six_lock_readers_add(&b->lock, readers);
432 
433 	if (ret)
434 		mark_btree_node_locked_noreset(path, b->level, BTREE_NODE_INTENT_LOCKED);
435 
436 	return ret;
437 }
438 
439 void bch2_btree_node_lock_write_nofail(struct btree_trans *trans,
440 				       struct btree_path *path,
441 				       struct btree_bkey_cached_common *b)
442 {
443 	int ret = __btree_node_lock_write(trans, path, b, true);
444 	BUG_ON(ret);
445 }
446 
447 /* relock */
448 
449 static inline bool btree_path_get_locks(struct btree_trans *trans,
450 					struct btree_path *path,
451 					bool upgrade,
452 					struct get_locks_fail *f)
453 {
454 	unsigned l = path->level;
455 	int fail_idx = -1;
456 
457 	do {
458 		if (!btree_path_node(path, l))
459 			break;
460 
461 		if (!(upgrade
462 		      ? bch2_btree_node_upgrade(trans, path, l)
463 		      : bch2_btree_node_relock(trans, path, l))) {
464 			fail_idx	= l;
465 
466 			if (f) {
467 				f->l	= l;
468 				f->b	= path->l[l].b;
469 			}
470 		}
471 
472 		l++;
473 	} while (l < path->locks_want);
474 
475 	/*
476 	 * When we fail to get a lock, we have to ensure that any child nodes
477 	 * can't be relocked so bch2_btree_path_traverse has to walk back up to
478 	 * the node that we failed to relock:
479 	 */
480 	if (fail_idx >= 0) {
481 		__bch2_btree_path_unlock(trans, path);
482 		btree_path_set_dirty(path, BTREE_ITER_NEED_TRAVERSE);
483 
484 		do {
485 			path->l[fail_idx].b = upgrade
486 				? ERR_PTR(-BCH_ERR_no_btree_node_upgrade)
487 				: ERR_PTR(-BCH_ERR_no_btree_node_relock);
488 			--fail_idx;
489 		} while (fail_idx >= 0);
490 	}
491 
492 	if (path->uptodate == BTREE_ITER_NEED_RELOCK)
493 		path->uptodate = BTREE_ITER_UPTODATE;
494 
495 	bch2_trans_verify_locks(trans);
496 
497 	return path->uptodate < BTREE_ITER_NEED_RELOCK;
498 }
499 
500 bool __bch2_btree_node_relock(struct btree_trans *trans,
501 			      struct btree_path *path, unsigned level,
502 			      bool trace)
503 {
504 	struct btree *b = btree_path_node(path, level);
505 	int want = __btree_lock_want(path, level);
506 
507 	if (race_fault())
508 		goto fail;
509 
510 	if (six_relock_type(&b->c.lock, want, path->l[level].lock_seq) ||
511 	    (btree_node_lock_seq_matches(path, b, level) &&
512 	     btree_node_lock_increment(trans, &b->c, level, want))) {
513 		mark_btree_node_locked(trans, path, level, want);
514 		return true;
515 	}
516 fail:
517 	if (trace && !trans->notrace_relock_fail)
518 		trace_and_count(trans->c, btree_path_relock_fail, trans, _RET_IP_, path, level);
519 	return false;
520 }
521 
522 /* upgrade */
523 
524 bool bch2_btree_node_upgrade(struct btree_trans *trans,
525 			     struct btree_path *path, unsigned level)
526 {
527 	struct btree *b = path->l[level].b;
528 	struct six_lock_count count = bch2_btree_node_lock_counts(trans, path, &b->c, level);
529 
530 	if (!is_btree_node(path, level))
531 		return false;
532 
533 	switch (btree_lock_want(path, level)) {
534 	case BTREE_NODE_UNLOCKED:
535 		BUG_ON(btree_node_locked(path, level));
536 		return true;
537 	case BTREE_NODE_READ_LOCKED:
538 		BUG_ON(btree_node_intent_locked(path, level));
539 		return bch2_btree_node_relock(trans, path, level);
540 	case BTREE_NODE_INTENT_LOCKED:
541 		break;
542 	case BTREE_NODE_WRITE_LOCKED:
543 		BUG();
544 	}
545 
546 	if (btree_node_intent_locked(path, level))
547 		return true;
548 
549 	if (race_fault())
550 		return false;
551 
552 	if (btree_node_locked(path, level)) {
553 		bool ret;
554 
555 		six_lock_readers_add(&b->c.lock, -count.n[SIX_LOCK_read]);
556 		ret = six_lock_tryupgrade(&b->c.lock);
557 		six_lock_readers_add(&b->c.lock, count.n[SIX_LOCK_read]);
558 
559 		if (ret)
560 			goto success;
561 	} else {
562 		if (six_relock_type(&b->c.lock, SIX_LOCK_intent, path->l[level].lock_seq))
563 			goto success;
564 	}
565 
566 	/*
567 	 * Do we already have an intent lock via another path? If so, just bump
568 	 * lock count:
569 	 */
570 	if (btree_node_lock_seq_matches(path, b, level) &&
571 	    btree_node_lock_increment(trans, &b->c, level, BTREE_NODE_INTENT_LOCKED)) {
572 		btree_node_unlock(trans, path, level);
573 		goto success;
574 	}
575 
576 	trace_and_count(trans->c, btree_path_upgrade_fail, trans, _RET_IP_, path, level);
577 	return false;
578 success:
579 	mark_btree_node_locked_noreset(path, level, BTREE_NODE_INTENT_LOCKED);
580 	return true;
581 }
582 
583 /* Btree path locking: */
584 
585 /*
586  * Only for btree_cache.c - only relocks intent locks
587  */
588 int bch2_btree_path_relock_intent(struct btree_trans *trans,
589 				  struct btree_path *path)
590 {
591 	unsigned l;
592 
593 	for (l = path->level;
594 	     l < path->locks_want && btree_path_node(path, l);
595 	     l++) {
596 		if (!bch2_btree_node_relock(trans, path, l)) {
597 			__bch2_btree_path_unlock(trans, path);
598 			btree_path_set_dirty(path, BTREE_ITER_NEED_TRAVERSE);
599 			trace_and_count(trans->c, trans_restart_relock_path_intent, trans, _RET_IP_, path);
600 			return btree_trans_restart(trans, BCH_ERR_transaction_restart_relock_path_intent);
601 		}
602 	}
603 
604 	return 0;
605 }
606 
607 __flatten
608 bool bch2_btree_path_relock_norestart(struct btree_trans *trans, struct btree_path *path)
609 {
610 	struct get_locks_fail f;
611 
612 	return btree_path_get_locks(trans, path, false, &f);
613 }
614 
615 int __bch2_btree_path_relock(struct btree_trans *trans,
616 			struct btree_path *path, unsigned long trace_ip)
617 {
618 	if (!bch2_btree_path_relock_norestart(trans, path)) {
619 		trace_and_count(trans->c, trans_restart_relock_path, trans, trace_ip, path);
620 		return btree_trans_restart(trans, BCH_ERR_transaction_restart_relock_path);
621 	}
622 
623 	return 0;
624 }
625 
626 bool bch2_btree_path_upgrade_noupgrade_sibs(struct btree_trans *trans,
627 			       struct btree_path *path,
628 			       unsigned new_locks_want,
629 			       struct get_locks_fail *f)
630 {
631 	EBUG_ON(path->locks_want >= new_locks_want);
632 
633 	path->locks_want = new_locks_want;
634 
635 	return btree_path_get_locks(trans, path, true, f);
636 }
637 
638 bool __bch2_btree_path_upgrade(struct btree_trans *trans,
639 			       struct btree_path *path,
640 			       unsigned new_locks_want,
641 			       struct get_locks_fail *f)
642 {
643 	if (bch2_btree_path_upgrade_noupgrade_sibs(trans, path, new_locks_want, f))
644 		return true;
645 
646 	/*
647 	 * XXX: this is ugly - we'd prefer to not be mucking with other
648 	 * iterators in the btree_trans here.
649 	 *
650 	 * On failure to upgrade the iterator, setting iter->locks_want and
651 	 * calling get_locks() is sufficient to make bch2_btree_path_traverse()
652 	 * get the locks we want on transaction restart.
653 	 *
654 	 * But if this iterator was a clone, on transaction restart what we did
655 	 * to this iterator isn't going to be preserved.
656 	 *
657 	 * Possibly we could add an iterator field for the parent iterator when
658 	 * an iterator is a copy - for now, we'll just upgrade any other
659 	 * iterators with the same btree id.
660 	 *
661 	 * The code below used to be needed to ensure ancestor nodes get locked
662 	 * before interior nodes - now that's handled by
663 	 * bch2_btree_path_traverse_all().
664 	 */
665 	if (!path->cached && !trans->in_traverse_all) {
666 		struct btree_path *linked;
667 		unsigned i;
668 
669 		trans_for_each_path(trans, linked, i)
670 			if (linked != path &&
671 			    linked->cached == path->cached &&
672 			    linked->btree_id == path->btree_id &&
673 			    linked->locks_want < new_locks_want) {
674 				linked->locks_want = new_locks_want;
675 				btree_path_get_locks(trans, linked, true, NULL);
676 			}
677 	}
678 
679 	return false;
680 }
681 
682 void __bch2_btree_path_downgrade(struct btree_trans *trans,
683 				 struct btree_path *path,
684 				 unsigned new_locks_want)
685 {
686 	unsigned l, old_locks_want = path->locks_want;
687 
688 	if (trans->restarted)
689 		return;
690 
691 	EBUG_ON(path->locks_want < new_locks_want);
692 
693 	path->locks_want = new_locks_want;
694 
695 	while (path->nodes_locked &&
696 	       (l = btree_path_highest_level_locked(path)) >= path->locks_want) {
697 		if (l > path->level) {
698 			btree_node_unlock(trans, path, l);
699 		} else {
700 			if (btree_node_intent_locked(path, l)) {
701 				six_lock_downgrade(&path->l[l].b->c.lock);
702 				mark_btree_node_locked_noreset(path, l, BTREE_NODE_READ_LOCKED);
703 			}
704 			break;
705 		}
706 	}
707 
708 	bch2_btree_path_verify_locks(path);
709 
710 	trace_path_downgrade(trans, _RET_IP_, path, old_locks_want);
711 }
712 
713 /* Btree transaction locking: */
714 
715 void bch2_trans_downgrade(struct btree_trans *trans)
716 {
717 	struct btree_path *path;
718 	unsigned i;
719 
720 	if (trans->restarted)
721 		return;
722 
723 	trans_for_each_path(trans, path, i)
724 		if (path->ref)
725 			bch2_btree_path_downgrade(trans, path);
726 }
727 
728 int bch2_trans_relock(struct btree_trans *trans)
729 {
730 	struct btree_path *path;
731 	unsigned i;
732 
733 	if (unlikely(trans->restarted))
734 		return -((int) trans->restarted);
735 
736 	trans_for_each_path(trans, path, i) {
737 		struct get_locks_fail f;
738 
739 		if (path->should_be_locked &&
740 		    !btree_path_get_locks(trans, path, false, &f)) {
741 			if (trace_trans_restart_relock_enabled()) {
742 				struct printbuf buf = PRINTBUF;
743 
744 				bch2_bpos_to_text(&buf, path->pos);
745 				prt_printf(&buf, " l=%u seq=%u node seq=",
746 					   f.l, path->l[f.l].lock_seq);
747 				if (IS_ERR_OR_NULL(f.b)) {
748 					prt_str(&buf, bch2_err_str(PTR_ERR(f.b)));
749 				} else {
750 					prt_printf(&buf, "%u", f.b->c.lock.seq);
751 
752 					struct six_lock_count c =
753 						bch2_btree_node_lock_counts(trans, NULL, &f.b->c, f.l);
754 					prt_printf(&buf, " self locked %u.%u.%u", c.n[0], c.n[1], c.n[2]);
755 
756 					c = six_lock_counts(&f.b->c.lock);
757 					prt_printf(&buf, " total locked %u.%u.%u", c.n[0], c.n[1], c.n[2]);
758 				}
759 
760 				trace_trans_restart_relock(trans, _RET_IP_, buf.buf);
761 				printbuf_exit(&buf);
762 			}
763 
764 			count_event(trans->c, trans_restart_relock);
765 			return btree_trans_restart(trans, BCH_ERR_transaction_restart_relock);
766 		}
767 	}
768 
769 	return 0;
770 }
771 
772 int bch2_trans_relock_notrace(struct btree_trans *trans)
773 {
774 	struct btree_path *path;
775 	unsigned i;
776 
777 	if (unlikely(trans->restarted))
778 		return -((int) trans->restarted);
779 
780 	trans_for_each_path(trans, path, i)
781 		if (path->should_be_locked &&
782 		    !bch2_btree_path_relock_norestart(trans, path)) {
783 			return btree_trans_restart(trans, BCH_ERR_transaction_restart_relock);
784 		}
785 	return 0;
786 }
787 
788 void bch2_trans_unlock_noassert(struct btree_trans *trans)
789 {
790 	struct btree_path *path;
791 	unsigned i;
792 
793 	trans_for_each_path(trans, path, i)
794 		__bch2_btree_path_unlock(trans, path);
795 }
796 
797 void bch2_trans_unlock(struct btree_trans *trans)
798 {
799 	struct btree_path *path;
800 	unsigned i;
801 
802 	trans_for_each_path(trans, path, i)
803 		__bch2_btree_path_unlock(trans, path);
804 }
805 
806 void bch2_trans_unlock_long(struct btree_trans *trans)
807 {
808 	bch2_trans_unlock(trans);
809 	bch2_trans_srcu_unlock(trans);
810 }
811 
812 bool bch2_trans_locked(struct btree_trans *trans)
813 {
814 	struct btree_path *path;
815 	unsigned i;
816 
817 	trans_for_each_path(trans, path, i)
818 		if (path->nodes_locked)
819 			return true;
820 	return false;
821 }
822 
823 int __bch2_trans_mutex_lock(struct btree_trans *trans,
824 			    struct mutex *lock)
825 {
826 	int ret = drop_locks_do(trans, (mutex_lock(lock), 0));
827 
828 	if (ret)
829 		mutex_unlock(lock);
830 	return ret;
831 }
832 
833 /* Debug */
834 
835 #ifdef CONFIG_BCACHEFS_DEBUG
836 
837 void bch2_btree_path_verify_locks(struct btree_path *path)
838 {
839 	unsigned l;
840 
841 	if (!path->nodes_locked) {
842 		BUG_ON(path->uptodate == BTREE_ITER_UPTODATE &&
843 		       btree_path_node(path, path->level));
844 		return;
845 	}
846 
847 	for (l = 0; l < BTREE_MAX_DEPTH; l++) {
848 		int want = btree_lock_want(path, l);
849 		int have = btree_node_locked_type(path, l);
850 
851 		BUG_ON(!is_btree_node(path, l) && have != BTREE_NODE_UNLOCKED);
852 
853 		BUG_ON(is_btree_node(path, l) &&
854 		       (want == BTREE_NODE_UNLOCKED ||
855 			have != BTREE_NODE_WRITE_LOCKED) &&
856 		       want != have);
857 	}
858 }
859 
860 void bch2_trans_verify_locks(struct btree_trans *trans)
861 {
862 	struct btree_path *path;
863 	unsigned i;
864 
865 	trans_for_each_path(trans, path, i)
866 		bch2_btree_path_verify_locks(path);
867 }
868 
869 #endif
870