xref: /linux/fs/bcachefs/btree_iter.c (revision 7255fcc80d4b525cc10cfaaf7f485830d4ed2000)
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 	unsigned i;
225 
226 	EBUG_ON(path->btree_id >= BTREE_ID_NR);
227 
228 	for (i = 0; i < (!path->cached ? BTREE_MAX_DEPTH : 1); i++) {
229 		if (!path->l[i].b) {
230 			BUG_ON(!path->cached &&
231 			       bch2_btree_id_root(c, path->btree_id)->b->c.level > i);
232 			break;
233 		}
234 
235 		bch2_btree_path_verify_level(trans, path, i);
236 	}
237 
238 	bch2_btree_path_verify_locks(path);
239 }
240 
241 void bch2_trans_verify_paths(struct btree_trans *trans)
242 {
243 	struct btree_path *path;
244 	unsigned iter;
245 
246 	trans_for_each_path(trans, path, iter)
247 		bch2_btree_path_verify(trans, path);
248 }
249 
250 static void bch2_btree_iter_verify(struct btree_iter *iter)
251 {
252 	struct btree_trans *trans = iter->trans;
253 
254 	BUG_ON(iter->btree_id >= BTREE_ID_NR);
255 
256 	BUG_ON(!!(iter->flags & BTREE_ITER_CACHED) != btree_iter_path(trans, iter)->cached);
257 
258 	BUG_ON((iter->flags & BTREE_ITER_IS_EXTENTS) &&
259 	       (iter->flags & BTREE_ITER_ALL_SNAPSHOTS));
260 
261 	BUG_ON(!(iter->flags & __BTREE_ITER_ALL_SNAPSHOTS) &&
262 	       (iter->flags & BTREE_ITER_ALL_SNAPSHOTS) &&
263 	       !btree_type_has_snapshot_field(iter->btree_id));
264 
265 	if (iter->update_path)
266 		bch2_btree_path_verify(trans, &trans->paths[iter->update_path]);
267 	bch2_btree_path_verify(trans, btree_iter_path(trans, iter));
268 }
269 
270 static void bch2_btree_iter_verify_entry_exit(struct btree_iter *iter)
271 {
272 	BUG_ON((iter->flags & BTREE_ITER_FILTER_SNAPSHOTS) &&
273 	       !iter->pos.snapshot);
274 
275 	BUG_ON(!(iter->flags & BTREE_ITER_ALL_SNAPSHOTS) &&
276 	       iter->pos.snapshot != iter->snapshot);
277 
278 	BUG_ON(bkey_lt(iter->pos, bkey_start_pos(&iter->k)) ||
279 	       bkey_gt(iter->pos, iter->k.p));
280 }
281 
282 static int bch2_btree_iter_verify_ret(struct btree_iter *iter, struct bkey_s_c k)
283 {
284 	struct btree_trans *trans = iter->trans;
285 	struct btree_iter copy;
286 	struct bkey_s_c prev;
287 	int ret = 0;
288 
289 	if (!bch2_debug_check_iterators)
290 		return 0;
291 
292 	if (!(iter->flags & BTREE_ITER_FILTER_SNAPSHOTS))
293 		return 0;
294 
295 	if (bkey_err(k) || !k.k)
296 		return 0;
297 
298 	BUG_ON(!bch2_snapshot_is_ancestor(trans->c,
299 					  iter->snapshot,
300 					  k.k->p.snapshot));
301 
302 	bch2_trans_iter_init(trans, &copy, iter->btree_id, iter->pos,
303 			     BTREE_ITER_NOPRESERVE|
304 			     BTREE_ITER_ALL_SNAPSHOTS);
305 	prev = bch2_btree_iter_prev(&copy);
306 	if (!prev.k)
307 		goto out;
308 
309 	ret = bkey_err(prev);
310 	if (ret)
311 		goto out;
312 
313 	if (bkey_eq(prev.k->p, k.k->p) &&
314 	    bch2_snapshot_is_ancestor(trans->c, iter->snapshot,
315 				      prev.k->p.snapshot) > 0) {
316 		struct printbuf buf1 = PRINTBUF, buf2 = PRINTBUF;
317 
318 		bch2_bkey_to_text(&buf1, k.k);
319 		bch2_bkey_to_text(&buf2, prev.k);
320 
321 		panic("iter snap %u\n"
322 		      "k    %s\n"
323 		      "prev %s\n",
324 		      iter->snapshot,
325 		      buf1.buf, buf2.buf);
326 	}
327 out:
328 	bch2_trans_iter_exit(trans, &copy);
329 	return ret;
330 }
331 
332 void bch2_assert_pos_locked(struct btree_trans *trans, enum btree_id id,
333 			    struct bpos pos, bool key_cache)
334 {
335 	struct btree_path *path;
336 	struct trans_for_each_path_inorder_iter iter;
337 	struct printbuf buf = PRINTBUF;
338 
339 	btree_trans_sort_paths(trans);
340 
341 	trans_for_each_path_inorder(trans, path, iter) {
342 		int cmp = cmp_int(path->btree_id, id) ?:
343 			cmp_int(path->cached, key_cache);
344 
345 		if (cmp > 0)
346 			break;
347 		if (cmp < 0)
348 			continue;
349 
350 		if (!btree_node_locked(path, 0) ||
351 		    !path->should_be_locked)
352 			continue;
353 
354 		if (!key_cache) {
355 			if (bkey_ge(pos, path->l[0].b->data->min_key) &&
356 			    bkey_le(pos, path->l[0].b->key.k.p))
357 				return;
358 		} else {
359 			if (bkey_eq(pos, path->pos))
360 				return;
361 		}
362 	}
363 
364 	bch2_dump_trans_paths_updates(trans);
365 	bch2_bpos_to_text(&buf, pos);
366 
367 	panic("not locked: %s %s%s\n",
368 	      bch2_btree_id_str(id), buf.buf,
369 	      key_cache ? " cached" : "");
370 }
371 
372 #else
373 
374 static inline void bch2_btree_path_verify_level(struct btree_trans *trans,
375 						struct btree_path *path, unsigned l) {}
376 static inline void bch2_btree_path_verify(struct btree_trans *trans,
377 					  struct btree_path *path) {}
378 static inline void bch2_btree_iter_verify(struct btree_iter *iter) {}
379 static inline void bch2_btree_iter_verify_entry_exit(struct btree_iter *iter) {}
380 static inline int bch2_btree_iter_verify_ret(struct btree_iter *iter, struct bkey_s_c k) { return 0; }
381 
382 #endif
383 
384 /* Btree path: fixups after btree updates */
385 
386 static void btree_node_iter_set_set_pos(struct btree_node_iter *iter,
387 					struct btree *b,
388 					struct bset_tree *t,
389 					struct bkey_packed *k)
390 {
391 	struct btree_node_iter_set *set;
392 
393 	btree_node_iter_for_each(iter, set)
394 		if (set->end == t->end_offset) {
395 			set->k = __btree_node_key_to_offset(b, k);
396 			bch2_btree_node_iter_sort(iter, b);
397 			return;
398 		}
399 
400 	bch2_btree_node_iter_push(iter, b, k, btree_bkey_last(b, t));
401 }
402 
403 static void __bch2_btree_path_fix_key_modified(struct btree_path *path,
404 					       struct btree *b,
405 					       struct bkey_packed *where)
406 {
407 	struct btree_path_level *l = &path->l[b->c.level];
408 
409 	if (where != bch2_btree_node_iter_peek_all(&l->iter, l->b))
410 		return;
411 
412 	if (bkey_iter_pos_cmp(l->b, where, &path->pos) < 0)
413 		bch2_btree_node_iter_advance(&l->iter, l->b);
414 }
415 
416 void bch2_btree_path_fix_key_modified(struct btree_trans *trans,
417 				      struct btree *b,
418 				      struct bkey_packed *where)
419 {
420 	struct btree_path *path;
421 	unsigned i;
422 
423 	trans_for_each_path_with_node(trans, b, path, i) {
424 		__bch2_btree_path_fix_key_modified(path, b, where);
425 		bch2_btree_path_verify_level(trans, path, b->c.level);
426 	}
427 }
428 
429 static void __bch2_btree_node_iter_fix(struct btree_path *path,
430 				       struct btree *b,
431 				       struct btree_node_iter *node_iter,
432 				       struct bset_tree *t,
433 				       struct bkey_packed *where,
434 				       unsigned clobber_u64s,
435 				       unsigned new_u64s)
436 {
437 	const struct bkey_packed *end = btree_bkey_last(b, t);
438 	struct btree_node_iter_set *set;
439 	unsigned offset = __btree_node_key_to_offset(b, where);
440 	int shift = new_u64s - clobber_u64s;
441 	unsigned old_end = t->end_offset - shift;
442 	unsigned orig_iter_pos = node_iter->data[0].k;
443 	bool iter_current_key_modified =
444 		orig_iter_pos >= offset &&
445 		orig_iter_pos <= offset + clobber_u64s;
446 
447 	btree_node_iter_for_each(node_iter, set)
448 		if (set->end == old_end)
449 			goto found;
450 
451 	/* didn't find the bset in the iterator - might have to readd it: */
452 	if (new_u64s &&
453 	    bkey_iter_pos_cmp(b, where, &path->pos) >= 0) {
454 		bch2_btree_node_iter_push(node_iter, b, where, end);
455 		goto fixup_done;
456 	} else {
457 		/* Iterator is after key that changed */
458 		return;
459 	}
460 found:
461 	set->end = t->end_offset;
462 
463 	/* Iterator hasn't gotten to the key that changed yet: */
464 	if (set->k < offset)
465 		return;
466 
467 	if (new_u64s &&
468 	    bkey_iter_pos_cmp(b, where, &path->pos) >= 0) {
469 		set->k = offset;
470 	} else if (set->k < offset + clobber_u64s) {
471 		set->k = offset + new_u64s;
472 		if (set->k == set->end)
473 			bch2_btree_node_iter_set_drop(node_iter, set);
474 	} else {
475 		/* Iterator is after key that changed */
476 		set->k = (int) set->k + shift;
477 		return;
478 	}
479 
480 	bch2_btree_node_iter_sort(node_iter, b);
481 fixup_done:
482 	if (node_iter->data[0].k != orig_iter_pos)
483 		iter_current_key_modified = true;
484 
485 	/*
486 	 * When a new key is added, and the node iterator now points to that
487 	 * key, the iterator might have skipped past deleted keys that should
488 	 * come after the key the iterator now points to. We have to rewind to
489 	 * before those deleted keys - otherwise
490 	 * bch2_btree_node_iter_prev_all() breaks:
491 	 */
492 	if (!bch2_btree_node_iter_end(node_iter) &&
493 	    iter_current_key_modified &&
494 	    b->c.level) {
495 		struct bkey_packed *k, *k2, *p;
496 
497 		k = bch2_btree_node_iter_peek_all(node_iter, b);
498 
499 		for_each_bset(b, t) {
500 			bool set_pos = false;
501 
502 			if (node_iter->data[0].end == t->end_offset)
503 				continue;
504 
505 			k2 = bch2_btree_node_iter_bset_pos(node_iter, b, t);
506 
507 			while ((p = bch2_bkey_prev_all(b, t, k2)) &&
508 			       bkey_iter_cmp(b, k, p) < 0) {
509 				k2 = p;
510 				set_pos = true;
511 			}
512 
513 			if (set_pos)
514 				btree_node_iter_set_set_pos(node_iter,
515 							    b, t, k2);
516 		}
517 	}
518 }
519 
520 void bch2_btree_node_iter_fix(struct btree_trans *trans,
521 			      struct btree_path *path,
522 			      struct btree *b,
523 			      struct btree_node_iter *node_iter,
524 			      struct bkey_packed *where,
525 			      unsigned clobber_u64s,
526 			      unsigned new_u64s)
527 {
528 	struct bset_tree *t = bch2_bkey_to_bset_inlined(b, where);
529 	struct btree_path *linked;
530 	unsigned i;
531 
532 	if (node_iter != &path->l[b->c.level].iter) {
533 		__bch2_btree_node_iter_fix(path, b, node_iter, t,
534 					   where, clobber_u64s, new_u64s);
535 
536 		if (bch2_debug_check_iterators)
537 			bch2_btree_node_iter_verify(node_iter, b);
538 	}
539 
540 	trans_for_each_path_with_node(trans, b, linked, i) {
541 		__bch2_btree_node_iter_fix(linked, b,
542 					   &linked->l[b->c.level].iter, t,
543 					   where, clobber_u64s, new_u64s);
544 		bch2_btree_path_verify_level(trans, linked, b->c.level);
545 	}
546 }
547 
548 /* Btree path level: pointer to a particular btree node and node iter */
549 
550 static inline struct bkey_s_c __btree_iter_unpack(struct bch_fs *c,
551 						  struct btree_path_level *l,
552 						  struct bkey *u,
553 						  struct bkey_packed *k)
554 {
555 	if (unlikely(!k)) {
556 		/*
557 		 * signal to bch2_btree_iter_peek_slot() that we're currently at
558 		 * a hole
559 		 */
560 		u->type = KEY_TYPE_deleted;
561 		return bkey_s_c_null;
562 	}
563 
564 	return bkey_disassemble(l->b, k, u);
565 }
566 
567 static inline struct bkey_s_c btree_path_level_peek_all(struct bch_fs *c,
568 							struct btree_path_level *l,
569 							struct bkey *u)
570 {
571 	return __btree_iter_unpack(c, l, u,
572 			bch2_btree_node_iter_peek_all(&l->iter, l->b));
573 }
574 
575 static inline struct bkey_s_c btree_path_level_peek(struct btree_trans *trans,
576 						    struct btree_path *path,
577 						    struct btree_path_level *l,
578 						    struct bkey *u)
579 {
580 	struct bkey_s_c k = __btree_iter_unpack(trans->c, l, u,
581 			bch2_btree_node_iter_peek(&l->iter, l->b));
582 
583 	path->pos = k.k ? k.k->p : l->b->key.k.p;
584 	trans->paths_sorted = false;
585 	bch2_btree_path_verify_level(trans, path, l - path->l);
586 	return k;
587 }
588 
589 static inline struct bkey_s_c btree_path_level_prev(struct btree_trans *trans,
590 						    struct btree_path *path,
591 						    struct btree_path_level *l,
592 						    struct bkey *u)
593 {
594 	struct bkey_s_c k = __btree_iter_unpack(trans->c, l, u,
595 			bch2_btree_node_iter_prev(&l->iter, l->b));
596 
597 	path->pos = k.k ? k.k->p : l->b->data->min_key;
598 	trans->paths_sorted = false;
599 	bch2_btree_path_verify_level(trans, path, l - path->l);
600 	return k;
601 }
602 
603 static inline bool btree_path_advance_to_pos(struct btree_path *path,
604 					     struct btree_path_level *l,
605 					     int max_advance)
606 {
607 	struct bkey_packed *k;
608 	int nr_advanced = 0;
609 
610 	while ((k = bch2_btree_node_iter_peek_all(&l->iter, l->b)) &&
611 	       bkey_iter_pos_cmp(l->b, k, &path->pos) < 0) {
612 		if (max_advance > 0 && nr_advanced >= max_advance)
613 			return false;
614 
615 		bch2_btree_node_iter_advance(&l->iter, l->b);
616 		nr_advanced++;
617 	}
618 
619 	return true;
620 }
621 
622 static inline void __btree_path_level_init(struct btree_path *path,
623 					   unsigned level)
624 {
625 	struct btree_path_level *l = &path->l[level];
626 
627 	bch2_btree_node_iter_init(&l->iter, l->b, &path->pos);
628 
629 	/*
630 	 * Iterators to interior nodes should always be pointed at the first non
631 	 * whiteout:
632 	 */
633 	if (level)
634 		bch2_btree_node_iter_peek(&l->iter, l->b);
635 }
636 
637 void bch2_btree_path_level_init(struct btree_trans *trans,
638 				struct btree_path *path,
639 				struct btree *b)
640 {
641 	BUG_ON(path->cached);
642 
643 	EBUG_ON(!btree_path_pos_in_node(path, b));
644 
645 	path->l[b->c.level].lock_seq = six_lock_seq(&b->c.lock);
646 	path->l[b->c.level].b = b;
647 	__btree_path_level_init(path, b->c.level);
648 }
649 
650 /* Btree path: fixups after btree node updates: */
651 
652 static void bch2_trans_revalidate_updates_in_node(struct btree_trans *trans, struct btree *b)
653 {
654 	struct bch_fs *c = trans->c;
655 
656 	trans_for_each_update(trans, i)
657 		if (!i->cached &&
658 		    i->level	== b->c.level &&
659 		    i->btree_id	== b->c.btree_id &&
660 		    bpos_cmp(i->k->k.p, b->data->min_key) >= 0 &&
661 		    bpos_cmp(i->k->k.p, b->data->max_key) <= 0) {
662 			i->old_v = bch2_btree_path_peek_slot(trans->paths + i->path, &i->old_k).v;
663 
664 			if (unlikely(trans->journal_replay_not_finished)) {
665 				struct bkey_i *j_k =
666 					bch2_journal_keys_peek_slot(c, i->btree_id, i->level,
667 								    i->k->k.p);
668 
669 				if (j_k) {
670 					i->old_k = j_k->k;
671 					i->old_v = &j_k->v;
672 				}
673 			}
674 		}
675 }
676 
677 /*
678  * A btree node is being replaced - update the iterator to point to the new
679  * node:
680  */
681 void bch2_trans_node_add(struct btree_trans *trans,
682 			 struct btree_path *path,
683 			 struct btree *b)
684 {
685 	struct btree_path *prev;
686 
687 	BUG_ON(!btree_path_pos_in_node(path, b));
688 
689 	while ((prev = prev_btree_path(trans, path)) &&
690 	       btree_path_pos_in_node(prev, b))
691 		path = prev;
692 
693 	for (;
694 	     path && btree_path_pos_in_node(path, b);
695 	     path = next_btree_path(trans, path))
696 		if (path->uptodate == BTREE_ITER_UPTODATE && !path->cached) {
697 			enum btree_node_locked_type t =
698 				btree_lock_want(path, b->c.level);
699 
700 			if (t != BTREE_NODE_UNLOCKED) {
701 				btree_node_unlock(trans, path, b->c.level);
702 				six_lock_increment(&b->c.lock, (enum six_lock_type) t);
703 				mark_btree_node_locked(trans, path, b->c.level, t);
704 			}
705 
706 			bch2_btree_path_level_init(trans, path, b);
707 		}
708 
709 	bch2_trans_revalidate_updates_in_node(trans, b);
710 }
711 
712 /*
713  * A btree node has been modified in such a way as to invalidate iterators - fix
714  * them:
715  */
716 void bch2_trans_node_reinit_iter(struct btree_trans *trans, struct btree *b)
717 {
718 	struct btree_path *path;
719 	unsigned i;
720 
721 	trans_for_each_path_with_node(trans, b, path, i)
722 		__btree_path_level_init(path, b->c.level);
723 
724 	bch2_trans_revalidate_updates_in_node(trans, b);
725 }
726 
727 /* Btree path: traverse, set_pos: */
728 
729 static inline int btree_path_lock_root(struct btree_trans *trans,
730 				       struct btree_path *path,
731 				       unsigned depth_want,
732 				       unsigned long trace_ip)
733 {
734 	struct bch_fs *c = trans->c;
735 	struct btree *b, **rootp = &bch2_btree_id_root(c, path->btree_id)->b;
736 	enum six_lock_type lock_type;
737 	unsigned i;
738 	int ret;
739 
740 	EBUG_ON(path->nodes_locked);
741 
742 	while (1) {
743 		b = READ_ONCE(*rootp);
744 		path->level = READ_ONCE(b->c.level);
745 
746 		if (unlikely(path->level < depth_want)) {
747 			/*
748 			 * the root is at a lower depth than the depth we want:
749 			 * got to the end of the btree, or we're walking nodes
750 			 * greater than some depth and there are no nodes >=
751 			 * that depth
752 			 */
753 			path->level = depth_want;
754 			for (i = path->level; i < BTREE_MAX_DEPTH; i++)
755 				path->l[i].b = NULL;
756 			return 1;
757 		}
758 
759 		lock_type = __btree_lock_want(path, path->level);
760 		ret = btree_node_lock(trans, path, &b->c,
761 				      path->level, lock_type, trace_ip);
762 		if (unlikely(ret)) {
763 			if (bch2_err_matches(ret, BCH_ERR_lock_fail_root_changed))
764 				continue;
765 			if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
766 				return ret;
767 			BUG();
768 		}
769 
770 		if (likely(b == READ_ONCE(*rootp) &&
771 			   b->c.level == path->level &&
772 			   !race_fault())) {
773 			for (i = 0; i < path->level; i++)
774 				path->l[i].b = ERR_PTR(-BCH_ERR_no_btree_node_lock_root);
775 			path->l[path->level].b = b;
776 			for (i = path->level + 1; i < BTREE_MAX_DEPTH; i++)
777 				path->l[i].b = NULL;
778 
779 			mark_btree_node_locked(trans, path, path->level,
780 					       (enum btree_node_locked_type) lock_type);
781 			bch2_btree_path_level_init(trans, path, b);
782 			return 0;
783 		}
784 
785 		six_unlock_type(&b->c.lock, lock_type);
786 	}
787 }
788 
789 noinline
790 static int btree_path_prefetch(struct btree_trans *trans, struct btree_path *path)
791 {
792 	struct bch_fs *c = trans->c;
793 	struct btree_path_level *l = path_l(path);
794 	struct btree_node_iter node_iter = l->iter;
795 	struct bkey_packed *k;
796 	struct bkey_buf tmp;
797 	unsigned nr = test_bit(BCH_FS_started, &c->flags)
798 		? (path->level > 1 ? 0 :  2)
799 		: (path->level > 1 ? 1 : 16);
800 	bool was_locked = btree_node_locked(path, path->level);
801 	int ret = 0;
802 
803 	bch2_bkey_buf_init(&tmp);
804 
805 	while (nr-- && !ret) {
806 		if (!bch2_btree_node_relock(trans, path, path->level))
807 			break;
808 
809 		bch2_btree_node_iter_advance(&node_iter, l->b);
810 		k = bch2_btree_node_iter_peek(&node_iter, l->b);
811 		if (!k)
812 			break;
813 
814 		bch2_bkey_buf_unpack(&tmp, c, l->b, k);
815 		ret = bch2_btree_node_prefetch(trans, path, tmp.k, path->btree_id,
816 					       path->level - 1);
817 	}
818 
819 	if (!was_locked)
820 		btree_node_unlock(trans, path, path->level);
821 
822 	bch2_bkey_buf_exit(&tmp, c);
823 	return ret;
824 }
825 
826 static int btree_path_prefetch_j(struct btree_trans *trans, struct btree_path *path,
827 				 struct btree_and_journal_iter *jiter)
828 {
829 	struct bch_fs *c = trans->c;
830 	struct bkey_s_c k;
831 	struct bkey_buf tmp;
832 	unsigned nr = test_bit(BCH_FS_started, &c->flags)
833 		? (path->level > 1 ? 0 :  2)
834 		: (path->level > 1 ? 1 : 16);
835 	bool was_locked = btree_node_locked(path, path->level);
836 	int ret = 0;
837 
838 	bch2_bkey_buf_init(&tmp);
839 
840 	while (nr-- && !ret) {
841 		if (!bch2_btree_node_relock(trans, path, path->level))
842 			break;
843 
844 		bch2_btree_and_journal_iter_advance(jiter);
845 		k = bch2_btree_and_journal_iter_peek(jiter);
846 		if (!k.k)
847 			break;
848 
849 		bch2_bkey_buf_reassemble(&tmp, c, k);
850 		ret = bch2_btree_node_prefetch(trans, path, tmp.k, path->btree_id,
851 					       path->level - 1);
852 	}
853 
854 	if (!was_locked)
855 		btree_node_unlock(trans, path, path->level);
856 
857 	bch2_bkey_buf_exit(&tmp, c);
858 	return ret;
859 }
860 
861 static noinline void btree_node_mem_ptr_set(struct btree_trans *trans,
862 					    struct btree_path *path,
863 					    unsigned plevel, struct btree *b)
864 {
865 	struct btree_path_level *l = &path->l[plevel];
866 	bool locked = btree_node_locked(path, plevel);
867 	struct bkey_packed *k;
868 	struct bch_btree_ptr_v2 *bp;
869 
870 	if (!bch2_btree_node_relock(trans, path, plevel))
871 		return;
872 
873 	k = bch2_btree_node_iter_peek_all(&l->iter, l->b);
874 	BUG_ON(k->type != KEY_TYPE_btree_ptr_v2);
875 
876 	bp = (void *) bkeyp_val(&l->b->format, k);
877 	bp->mem_ptr = (unsigned long)b;
878 
879 	if (!locked)
880 		btree_node_unlock(trans, path, plevel);
881 }
882 
883 static noinline int btree_node_iter_and_journal_peek(struct btree_trans *trans,
884 						     struct btree_path *path,
885 						     unsigned flags,
886 						     struct bkey_buf *out)
887 {
888 	struct bch_fs *c = trans->c;
889 	struct btree_path_level *l = path_l(path);
890 	struct btree_and_journal_iter jiter;
891 	struct bkey_s_c k;
892 	int ret = 0;
893 
894 	__bch2_btree_and_journal_iter_init_node_iter(trans, &jiter, l->b, l->iter, path->pos);
895 
896 	k = bch2_btree_and_journal_iter_peek(&jiter);
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 	bch2_btree_and_journal_iter_exit(&jiter);
905 	return ret;
906 }
907 
908 static __always_inline int btree_path_down(struct btree_trans *trans,
909 					   struct btree_path *path,
910 					   unsigned flags,
911 					   unsigned long trace_ip)
912 {
913 	struct bch_fs *c = trans->c;
914 	struct btree_path_level *l = path_l(path);
915 	struct btree *b;
916 	unsigned level = path->level - 1;
917 	enum six_lock_type lock_type = __btree_lock_want(path, level);
918 	struct bkey_buf tmp;
919 	int ret;
920 
921 	EBUG_ON(!btree_node_locked(path, path->level));
922 
923 	bch2_bkey_buf_init(&tmp);
924 
925 	if (unlikely(trans->journal_replay_not_finished)) {
926 		ret = btree_node_iter_and_journal_peek(trans, path, flags, &tmp);
927 		if (ret)
928 			goto err;
929 	} else {
930 		struct bkey_packed *k = bch2_btree_node_iter_peek(&l->iter, l->b);
931 		if (!k) {
932 			struct printbuf buf = PRINTBUF;
933 
934 			prt_str(&buf, "node not found at pos ");
935 			bch2_bpos_to_text(&buf, path->pos);
936 			prt_str(&buf, " within parent node ");
937 			bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&l->b->key));
938 
939 			bch2_fs_fatal_error(c, "%s", buf.buf);
940 			printbuf_exit(&buf);
941 			ret = -BCH_ERR_btree_need_topology_repair;
942 			goto err;
943 		}
944 
945 		bch2_bkey_buf_unpack(&tmp, c, l->b, k);
946 
947 		if ((flags & BTREE_ITER_PREFETCH) &&
948 		    c->opts.btree_node_prefetch) {
949 			ret = btree_path_prefetch(trans, path);
950 			if (ret)
951 				goto err;
952 		}
953 	}
954 
955 	b = bch2_btree_node_get(trans, path, tmp.k, level, lock_type, trace_ip);
956 	ret = PTR_ERR_OR_ZERO(b);
957 	if (unlikely(ret))
958 		goto err;
959 
960 	if (likely(!trans->journal_replay_not_finished &&
961 		   tmp.k->k.type == KEY_TYPE_btree_ptr_v2) &&
962 	    unlikely(b != btree_node_mem_ptr(tmp.k)))
963 		btree_node_mem_ptr_set(trans, path, level + 1, b);
964 
965 	if (btree_node_read_locked(path, level + 1))
966 		btree_node_unlock(trans, path, level + 1);
967 
968 	mark_btree_node_locked(trans, path, level,
969 			       (enum btree_node_locked_type) lock_type);
970 	path->level = level;
971 	bch2_btree_path_level_init(trans, path, b);
972 
973 	bch2_btree_path_verify_locks(path);
974 err:
975 	bch2_bkey_buf_exit(&tmp, c);
976 	return ret;
977 }
978 
979 static int bch2_btree_path_traverse_all(struct btree_trans *trans)
980 {
981 	struct bch_fs *c = trans->c;
982 	struct btree_path *path;
983 	unsigned long trace_ip = _RET_IP_;
984 	unsigned i;
985 	int ret = 0;
986 
987 	if (trans->in_traverse_all)
988 		return -BCH_ERR_transaction_restart_in_traverse_all;
989 
990 	trans->in_traverse_all = true;
991 retry_all:
992 	trans->restarted = 0;
993 	trans->last_restarted_ip = 0;
994 
995 	trans_for_each_path(trans, path, i)
996 		path->should_be_locked = false;
997 
998 	btree_trans_sort_paths(trans);
999 
1000 	bch2_trans_unlock(trans);
1001 	cond_resched();
1002 
1003 	if (unlikely(trans->memory_allocation_failure)) {
1004 		struct closure cl;
1005 
1006 		closure_init_stack(&cl);
1007 
1008 		do {
1009 			ret = bch2_btree_cache_cannibalize_lock(trans, &cl);
1010 			closure_sync(&cl);
1011 		} while (ret);
1012 	}
1013 
1014 	/* Now, redo traversals in correct order: */
1015 	i = 0;
1016 	while (i < trans->nr_sorted) {
1017 		btree_path_idx_t idx = trans->sorted[i];
1018 
1019 		/*
1020 		 * Traversing a path can cause another path to be added at about
1021 		 * the same position:
1022 		 */
1023 		if (trans->paths[idx].uptodate) {
1024 			__btree_path_get(&trans->paths[idx], false);
1025 			ret = bch2_btree_path_traverse_one(trans, idx, 0, _THIS_IP_);
1026 			__btree_path_put(&trans->paths[idx], false);
1027 
1028 			if (bch2_err_matches(ret, BCH_ERR_transaction_restart) ||
1029 			    bch2_err_matches(ret, ENOMEM))
1030 				goto retry_all;
1031 			if (ret)
1032 				goto err;
1033 		} else {
1034 			i++;
1035 		}
1036 	}
1037 
1038 	/*
1039 	 * We used to assert that all paths had been traversed here
1040 	 * (path->uptodate < BTREE_ITER_NEED_TRAVERSE); however, since
1041 	 * path->should_be_locked is not set yet, we might have unlocked and
1042 	 * then failed to relock a path - that's fine.
1043 	 */
1044 err:
1045 	bch2_btree_cache_cannibalize_unlock(trans);
1046 
1047 	trans->in_traverse_all = false;
1048 
1049 	trace_and_count(c, trans_traverse_all, trans, trace_ip);
1050 	return ret;
1051 }
1052 
1053 static inline bool btree_path_check_pos_in_node(struct btree_path *path,
1054 						unsigned l, int check_pos)
1055 {
1056 	if (check_pos < 0 && btree_path_pos_before_node(path, path->l[l].b))
1057 		return false;
1058 	if (check_pos > 0 && btree_path_pos_after_node(path, path->l[l].b))
1059 		return false;
1060 	return true;
1061 }
1062 
1063 static inline bool btree_path_good_node(struct btree_trans *trans,
1064 					struct btree_path *path,
1065 					unsigned l, int check_pos)
1066 {
1067 	return is_btree_node(path, l) &&
1068 		bch2_btree_node_relock(trans, path, l) &&
1069 		btree_path_check_pos_in_node(path, l, check_pos);
1070 }
1071 
1072 static void btree_path_set_level_down(struct btree_trans *trans,
1073 				      struct btree_path *path,
1074 				      unsigned new_level)
1075 {
1076 	unsigned l;
1077 
1078 	path->level = new_level;
1079 
1080 	for (l = path->level + 1; l < BTREE_MAX_DEPTH; l++)
1081 		if (btree_lock_want(path, l) == BTREE_NODE_UNLOCKED)
1082 			btree_node_unlock(trans, path, l);
1083 
1084 	btree_path_set_dirty(path, BTREE_ITER_NEED_TRAVERSE);
1085 	bch2_btree_path_verify(trans, path);
1086 }
1087 
1088 static noinline unsigned __btree_path_up_until_good_node(struct btree_trans *trans,
1089 							 struct btree_path *path,
1090 							 int check_pos)
1091 {
1092 	unsigned i, l = path->level;
1093 again:
1094 	while (btree_path_node(path, l) &&
1095 	       !btree_path_good_node(trans, path, l, check_pos))
1096 		__btree_path_set_level_up(trans, path, l++);
1097 
1098 	/* If we need intent locks, take them too: */
1099 	for (i = l + 1;
1100 	     i < path->locks_want && btree_path_node(path, i);
1101 	     i++)
1102 		if (!bch2_btree_node_relock(trans, path, i)) {
1103 			while (l <= i)
1104 				__btree_path_set_level_up(trans, path, l++);
1105 			goto again;
1106 		}
1107 
1108 	return l;
1109 }
1110 
1111 static inline unsigned btree_path_up_until_good_node(struct btree_trans *trans,
1112 						     struct btree_path *path,
1113 						     int check_pos)
1114 {
1115 	return likely(btree_node_locked(path, path->level) &&
1116 		      btree_path_check_pos_in_node(path, path->level, check_pos))
1117 		? path->level
1118 		: __btree_path_up_until_good_node(trans, path, check_pos);
1119 }
1120 
1121 /*
1122  * This is the main state machine for walking down the btree - walks down to a
1123  * specified depth
1124  *
1125  * Returns 0 on success, -EIO on error (error reading in a btree node).
1126  *
1127  * On error, caller (peek_node()/peek_key()) must return NULL; the error is
1128  * stashed in the iterator and returned from bch2_trans_exit().
1129  */
1130 int bch2_btree_path_traverse_one(struct btree_trans *trans,
1131 				 btree_path_idx_t path_idx,
1132 				 unsigned flags,
1133 				 unsigned long trace_ip)
1134 {
1135 	struct btree_path *path = &trans->paths[path_idx];
1136 	unsigned depth_want = path->level;
1137 	int ret = -((int) trans->restarted);
1138 
1139 	if (unlikely(ret))
1140 		goto out;
1141 
1142 	if (unlikely(!trans->srcu_held))
1143 		bch2_trans_srcu_lock(trans);
1144 
1145 	/*
1146 	 * Ensure we obey path->should_be_locked: if it's set, we can't unlock
1147 	 * and re-traverse the path without a transaction restart:
1148 	 */
1149 	if (path->should_be_locked) {
1150 		ret = bch2_btree_path_relock(trans, path, trace_ip);
1151 		goto out;
1152 	}
1153 
1154 	if (path->cached) {
1155 		ret = bch2_btree_path_traverse_cached(trans, path, flags);
1156 		goto out;
1157 	}
1158 
1159 	path = &trans->paths[path_idx];
1160 
1161 	if (unlikely(path->level >= BTREE_MAX_DEPTH))
1162 		goto out_uptodate;
1163 
1164 	path->level = btree_path_up_until_good_node(trans, path, 0);
1165 
1166 	EBUG_ON(btree_path_node(path, path->level) &&
1167 		!btree_node_locked(path, path->level));
1168 
1169 	/*
1170 	 * Note: path->nodes[path->level] may be temporarily NULL here - that
1171 	 * would indicate to other code that we got to the end of the btree,
1172 	 * here it indicates that relocking the root failed - it's critical that
1173 	 * btree_path_lock_root() comes next and that it can't fail
1174 	 */
1175 	while (path->level > depth_want) {
1176 		ret = btree_path_node(path, path->level)
1177 			? btree_path_down(trans, path, flags, trace_ip)
1178 			: btree_path_lock_root(trans, path, depth_want, trace_ip);
1179 		if (unlikely(ret)) {
1180 			if (ret == 1) {
1181 				/*
1182 				 * No nodes at this level - got to the end of
1183 				 * the btree:
1184 				 */
1185 				ret = 0;
1186 				goto out;
1187 			}
1188 
1189 			__bch2_btree_path_unlock(trans, path);
1190 			path->level = depth_want;
1191 			path->l[path->level].b = ERR_PTR(ret);
1192 			goto out;
1193 		}
1194 	}
1195 out_uptodate:
1196 	path->uptodate = BTREE_ITER_UPTODATE;
1197 out:
1198 	if (bch2_err_matches(ret, BCH_ERR_transaction_restart) != !!trans->restarted)
1199 		panic("ret %s (%i) trans->restarted %s (%i)\n",
1200 		      bch2_err_str(ret), ret,
1201 		      bch2_err_str(trans->restarted), trans->restarted);
1202 	bch2_btree_path_verify(trans, path);
1203 	return ret;
1204 }
1205 
1206 static inline void btree_path_copy(struct btree_trans *trans, struct btree_path *dst,
1207 			    struct btree_path *src)
1208 {
1209 	unsigned i, offset = offsetof(struct btree_path, pos);
1210 
1211 	memcpy((void *) dst + offset,
1212 	       (void *) src + offset,
1213 	       sizeof(struct btree_path) - offset);
1214 
1215 	for (i = 0; i < BTREE_MAX_DEPTH; i++) {
1216 		unsigned t = btree_node_locked_type(dst, i);
1217 
1218 		if (t != BTREE_NODE_UNLOCKED)
1219 			six_lock_increment(&dst->l[i].b->c.lock, t);
1220 	}
1221 }
1222 
1223 static btree_path_idx_t btree_path_clone(struct btree_trans *trans, btree_path_idx_t src,
1224 					 bool intent)
1225 {
1226 	btree_path_idx_t new = btree_path_alloc(trans, src);
1227 	btree_path_copy(trans, trans->paths + new, trans->paths + src);
1228 	__btree_path_get(trans->paths + new, intent);
1229 	return new;
1230 }
1231 
1232 __flatten
1233 btree_path_idx_t __bch2_btree_path_make_mut(struct btree_trans *trans,
1234 			btree_path_idx_t path, bool intent, unsigned long ip)
1235 {
1236 	__btree_path_put(trans->paths + path, intent);
1237 	path = btree_path_clone(trans, path, intent);
1238 	trans->paths[path].preserve = false;
1239 	return path;
1240 }
1241 
1242 btree_path_idx_t __must_check
1243 __bch2_btree_path_set_pos(struct btree_trans *trans,
1244 			  btree_path_idx_t path_idx, struct bpos new_pos,
1245 			  bool intent, unsigned long ip)
1246 {
1247 	int cmp = bpos_cmp(new_pos, trans->paths[path_idx].pos);
1248 
1249 	bch2_trans_verify_not_in_restart(trans);
1250 	EBUG_ON(!trans->paths[path_idx].ref);
1251 
1252 	path_idx = bch2_btree_path_make_mut(trans, path_idx, intent, ip);
1253 
1254 	struct btree_path *path = trans->paths + path_idx;
1255 	path->pos		= new_pos;
1256 	trans->paths_sorted	= false;
1257 
1258 	if (unlikely(path->cached)) {
1259 		btree_node_unlock(trans, path, 0);
1260 		path->l[0].b = ERR_PTR(-BCH_ERR_no_btree_node_up);
1261 		btree_path_set_dirty(path, BTREE_ITER_NEED_TRAVERSE);
1262 		goto out;
1263 	}
1264 
1265 	unsigned level = btree_path_up_until_good_node(trans, path, cmp);
1266 
1267 	if (btree_path_node(path, level)) {
1268 		struct btree_path_level *l = &path->l[level];
1269 
1270 		BUG_ON(!btree_node_locked(path, level));
1271 		/*
1272 		 * We might have to skip over many keys, or just a few: try
1273 		 * advancing the node iterator, and if we have to skip over too
1274 		 * many keys just reinit it (or if we're rewinding, since that
1275 		 * is expensive).
1276 		 */
1277 		if (cmp < 0 ||
1278 		    !btree_path_advance_to_pos(path, l, 8))
1279 			bch2_btree_node_iter_init(&l->iter, l->b, &path->pos);
1280 
1281 		/*
1282 		 * Iterators to interior nodes should always be pointed at the first non
1283 		 * whiteout:
1284 		 */
1285 		if (unlikely(level))
1286 			bch2_btree_node_iter_peek(&l->iter, l->b);
1287 	}
1288 
1289 	if (unlikely(level != path->level)) {
1290 		btree_path_set_dirty(path, BTREE_ITER_NEED_TRAVERSE);
1291 		__bch2_btree_path_unlock(trans, path);
1292 	}
1293 out:
1294 	bch2_btree_path_verify(trans, path);
1295 	return path_idx;
1296 }
1297 
1298 /* Btree path: main interface: */
1299 
1300 static struct btree_path *have_path_at_pos(struct btree_trans *trans, struct btree_path *path)
1301 {
1302 	struct btree_path *sib;
1303 
1304 	sib = prev_btree_path(trans, path);
1305 	if (sib && !btree_path_cmp(sib, path))
1306 		return sib;
1307 
1308 	sib = next_btree_path(trans, path);
1309 	if (sib && !btree_path_cmp(sib, path))
1310 		return sib;
1311 
1312 	return NULL;
1313 }
1314 
1315 static struct btree_path *have_node_at_pos(struct btree_trans *trans, struct btree_path *path)
1316 {
1317 	struct btree_path *sib;
1318 
1319 	sib = prev_btree_path(trans, path);
1320 	if (sib && sib->level == path->level && path_l(sib)->b == path_l(path)->b)
1321 		return sib;
1322 
1323 	sib = next_btree_path(trans, path);
1324 	if (sib && sib->level == path->level && path_l(sib)->b == path_l(path)->b)
1325 		return sib;
1326 
1327 	return NULL;
1328 }
1329 
1330 static inline void __bch2_path_free(struct btree_trans *trans, btree_path_idx_t path)
1331 {
1332 	__bch2_btree_path_unlock(trans, trans->paths + path);
1333 	btree_path_list_remove(trans, trans->paths + path);
1334 	__clear_bit(path, trans->paths_allocated);
1335 }
1336 
1337 void bch2_path_put(struct btree_trans *trans, btree_path_idx_t path_idx, bool intent)
1338 {
1339 	struct btree_path *path = trans->paths + path_idx, *dup;
1340 
1341 	if (!__btree_path_put(path, intent))
1342 		return;
1343 
1344 	dup = path->preserve
1345 		? have_path_at_pos(trans, path)
1346 		: have_node_at_pos(trans, path);
1347 
1348 	if (!dup && !(!path->preserve && !is_btree_node(path, path->level)))
1349 		return;
1350 
1351 	if (path->should_be_locked &&
1352 	    !trans->restarted &&
1353 	    (!dup || !bch2_btree_path_relock_norestart(trans, dup)))
1354 		return;
1355 
1356 	if (dup) {
1357 		dup->preserve		|= path->preserve;
1358 		dup->should_be_locked	|= path->should_be_locked;
1359 	}
1360 
1361 	__bch2_path_free(trans, path_idx);
1362 }
1363 
1364 static void bch2_path_put_nokeep(struct btree_trans *trans, btree_path_idx_t path,
1365 				 bool intent)
1366 {
1367 	if (!__btree_path_put(trans->paths + path, intent))
1368 		return;
1369 
1370 	__bch2_path_free(trans, path);
1371 }
1372 
1373 void __noreturn bch2_trans_restart_error(struct btree_trans *trans, u32 restart_count)
1374 {
1375 	panic("trans->restart_count %u, should be %u, last restarted by %pS\n",
1376 	      trans->restart_count, restart_count,
1377 	      (void *) trans->last_begin_ip);
1378 }
1379 
1380 void __noreturn bch2_trans_in_restart_error(struct btree_trans *trans)
1381 {
1382 	panic("in transaction restart: %s, last restarted by %pS\n",
1383 	      bch2_err_str(trans->restarted),
1384 	      (void *) trans->last_restarted_ip);
1385 }
1386 
1387 noinline __cold
1388 void bch2_trans_updates_to_text(struct printbuf *buf, struct btree_trans *trans)
1389 {
1390 	prt_printf(buf, "transaction updates for %s journal seq %llu",
1391 	       trans->fn, trans->journal_res.seq);
1392 	prt_newline(buf);
1393 	printbuf_indent_add(buf, 2);
1394 
1395 	trans_for_each_update(trans, i) {
1396 		struct bkey_s_c old = { &i->old_k, i->old_v };
1397 
1398 		prt_printf(buf, "update: btree=%s cached=%u %pS",
1399 		       bch2_btree_id_str(i->btree_id),
1400 		       i->cached,
1401 		       (void *) i->ip_allocated);
1402 		prt_newline(buf);
1403 
1404 		prt_printf(buf, "  old ");
1405 		bch2_bkey_val_to_text(buf, trans->c, old);
1406 		prt_newline(buf);
1407 
1408 		prt_printf(buf, "  new ");
1409 		bch2_bkey_val_to_text(buf, trans->c, bkey_i_to_s_c(i->k));
1410 		prt_newline(buf);
1411 	}
1412 
1413 	for (struct jset_entry *e = trans->journal_entries;
1414 	     e != btree_trans_journal_entries_top(trans);
1415 	     e = vstruct_next(e))
1416 		bch2_journal_entry_to_text(buf, trans->c, e);
1417 
1418 	printbuf_indent_sub(buf, 2);
1419 }
1420 
1421 noinline __cold
1422 void bch2_dump_trans_updates(struct btree_trans *trans)
1423 {
1424 	struct printbuf buf = PRINTBUF;
1425 
1426 	bch2_trans_updates_to_text(&buf, trans);
1427 	bch2_print_string_as_lines(KERN_ERR, buf.buf);
1428 	printbuf_exit(&buf);
1429 }
1430 
1431 static void bch2_btree_path_to_text(struct printbuf *out, struct btree_trans *trans, btree_path_idx_t path_idx)
1432 {
1433 	struct btree_path *path = trans->paths + path_idx;
1434 
1435 	prt_printf(out, "path: idx %2u ref %u:%u %c %c btree=%s l=%u pos ",
1436 		   path_idx, path->ref, path->intent_ref,
1437 		   path->preserve ? 'P' : ' ',
1438 		   path->should_be_locked ? 'S' : ' ',
1439 		   bch2_btree_id_str(path->btree_id),
1440 		   path->level);
1441 	bch2_bpos_to_text(out, path->pos);
1442 
1443 	prt_printf(out, " locks %u", path->nodes_locked);
1444 #ifdef TRACK_PATH_ALLOCATED
1445 	prt_printf(out, " %pS", (void *) path->ip_allocated);
1446 #endif
1447 	prt_newline(out);
1448 }
1449 
1450 static noinline __cold
1451 void __bch2_trans_paths_to_text(struct printbuf *out, struct btree_trans *trans,
1452 				bool nosort)
1453 {
1454 	struct trans_for_each_path_inorder_iter iter;
1455 
1456 	if (!nosort)
1457 		btree_trans_sort_paths(trans);
1458 
1459 	trans_for_each_path_idx_inorder(trans, iter)
1460 		bch2_btree_path_to_text(out, trans, iter.path_idx);
1461 }
1462 
1463 noinline __cold
1464 void bch2_trans_paths_to_text(struct printbuf *out, struct btree_trans *trans)
1465 {
1466 	__bch2_trans_paths_to_text(out, trans, false);
1467 }
1468 
1469 static noinline __cold
1470 void __bch2_dump_trans_paths_updates(struct btree_trans *trans, bool nosort)
1471 {
1472 	struct printbuf buf = PRINTBUF;
1473 
1474 	__bch2_trans_paths_to_text(&buf, trans, nosort);
1475 	bch2_trans_updates_to_text(&buf, trans);
1476 
1477 	bch2_print_string_as_lines(KERN_ERR, buf.buf);
1478 	printbuf_exit(&buf);
1479 }
1480 
1481 noinline __cold
1482 void bch2_dump_trans_paths_updates(struct btree_trans *trans)
1483 {
1484 	__bch2_dump_trans_paths_updates(trans, false);
1485 }
1486 
1487 noinline __cold
1488 static void bch2_trans_update_max_paths(struct btree_trans *trans)
1489 {
1490 	struct btree_transaction_stats *s = btree_trans_stats(trans);
1491 	struct printbuf buf = PRINTBUF;
1492 	size_t nr = bitmap_weight(trans->paths_allocated, trans->nr_paths);
1493 
1494 	bch2_trans_paths_to_text(&buf, trans);
1495 
1496 	if (!buf.allocation_failure) {
1497 		mutex_lock(&s->lock);
1498 		if (nr > s->nr_max_paths) {
1499 			s->nr_max_paths = nr;
1500 			swap(s->max_paths_text, buf.buf);
1501 		}
1502 		mutex_unlock(&s->lock);
1503 	}
1504 
1505 	printbuf_exit(&buf);
1506 
1507 	trans->nr_paths_max = nr;
1508 }
1509 
1510 noinline __cold
1511 int __bch2_btree_trans_too_many_iters(struct btree_trans *trans)
1512 {
1513 	if (trace_trans_restart_too_many_iters_enabled()) {
1514 		struct printbuf buf = PRINTBUF;
1515 
1516 		bch2_trans_paths_to_text(&buf, trans);
1517 		trace_trans_restart_too_many_iters(trans, _THIS_IP_, buf.buf);
1518 		printbuf_exit(&buf);
1519 	}
1520 
1521 	count_event(trans->c, trans_restart_too_many_iters);
1522 
1523 	return btree_trans_restart(trans, BCH_ERR_transaction_restart_too_many_iters);
1524 }
1525 
1526 static noinline void btree_path_overflow(struct btree_trans *trans)
1527 {
1528 	bch2_dump_trans_paths_updates(trans);
1529 	bch_err(trans->c, "trans path overflow");
1530 }
1531 
1532 static noinline void btree_paths_realloc(struct btree_trans *trans)
1533 {
1534 	unsigned nr = trans->nr_paths * 2;
1535 
1536 	void *p = kvzalloc(BITS_TO_LONGS(nr) * sizeof(unsigned long) +
1537 			  sizeof(struct btree_trans_paths) +
1538 			  nr * sizeof(struct btree_path) +
1539 			  nr * sizeof(btree_path_idx_t) + 8 +
1540 			  nr * sizeof(struct btree_insert_entry), GFP_KERNEL|__GFP_NOFAIL);
1541 
1542 	unsigned long *paths_allocated = p;
1543 	memcpy(paths_allocated, trans->paths_allocated, BITS_TO_LONGS(trans->nr_paths) * sizeof(unsigned long));
1544 	p += BITS_TO_LONGS(nr) * sizeof(unsigned long);
1545 
1546 	p += sizeof(struct btree_trans_paths);
1547 	struct btree_path *paths = p;
1548 	*trans_paths_nr(paths) = nr;
1549 	memcpy(paths, trans->paths, trans->nr_paths * sizeof(struct btree_path));
1550 	p += nr * sizeof(struct btree_path);
1551 
1552 	btree_path_idx_t *sorted = p;
1553 	memcpy(sorted, trans->sorted, trans->nr_sorted * sizeof(btree_path_idx_t));
1554 	p += nr * sizeof(btree_path_idx_t) + 8;
1555 
1556 	struct btree_insert_entry *updates = p;
1557 	memcpy(updates, trans->updates, trans->nr_paths * sizeof(struct btree_insert_entry));
1558 
1559 	unsigned long *old = trans->paths_allocated;
1560 
1561 	rcu_assign_pointer(trans->paths_allocated,	paths_allocated);
1562 	rcu_assign_pointer(trans->paths,		paths);
1563 	rcu_assign_pointer(trans->sorted,		sorted);
1564 	rcu_assign_pointer(trans->updates,		updates);
1565 
1566 	trans->nr_paths		= nr;
1567 
1568 	if (old != trans->_paths_allocated)
1569 		kfree_rcu_mightsleep(old);
1570 }
1571 
1572 static inline btree_path_idx_t btree_path_alloc(struct btree_trans *trans,
1573 						btree_path_idx_t pos)
1574 {
1575 	btree_path_idx_t idx = find_first_zero_bit(trans->paths_allocated, trans->nr_paths);
1576 
1577 	if (unlikely(idx == trans->nr_paths)) {
1578 		if (trans->nr_paths == BTREE_ITER_MAX) {
1579 			btree_path_overflow(trans);
1580 			return 0;
1581 		}
1582 
1583 		btree_paths_realloc(trans);
1584 	}
1585 
1586 	/*
1587 	 * Do this before marking the new path as allocated, since it won't be
1588 	 * initialized yet:
1589 	 */
1590 	if (unlikely(idx > trans->nr_paths_max))
1591 		bch2_trans_update_max_paths(trans);
1592 
1593 	__set_bit(idx, trans->paths_allocated);
1594 
1595 	struct btree_path *path = &trans->paths[idx];
1596 	path->ref		= 0;
1597 	path->intent_ref	= 0;
1598 	path->nodes_locked	= 0;
1599 
1600 	btree_path_list_add(trans, pos, idx);
1601 	trans->paths_sorted = false;
1602 	return idx;
1603 }
1604 
1605 btree_path_idx_t bch2_path_get(struct btree_trans *trans,
1606 			     enum btree_id btree_id, struct bpos pos,
1607 			     unsigned locks_want, unsigned level,
1608 			     unsigned flags, unsigned long ip)
1609 {
1610 	struct btree_path *path;
1611 	bool cached = flags & BTREE_ITER_CACHED;
1612 	bool intent = flags & BTREE_ITER_INTENT;
1613 	struct trans_for_each_path_inorder_iter iter;
1614 	btree_path_idx_t path_pos = 0, path_idx;
1615 
1616 	bch2_trans_verify_not_in_restart(trans);
1617 	bch2_trans_verify_locks(trans);
1618 
1619 	btree_trans_sort_paths(trans);
1620 
1621 	trans_for_each_path_inorder(trans, path, iter) {
1622 		if (__btree_path_cmp(path,
1623 				     btree_id,
1624 				     cached,
1625 				     pos,
1626 				     level) > 0)
1627 			break;
1628 
1629 		path_pos = iter.path_idx;
1630 	}
1631 
1632 	if (path_pos &&
1633 	    trans->paths[path_pos].cached	== cached &&
1634 	    trans->paths[path_pos].btree_id	== btree_id &&
1635 	    trans->paths[path_pos].level	== level) {
1636 		__btree_path_get(trans->paths + path_pos, intent);
1637 		path_idx = bch2_btree_path_set_pos(trans, path_pos, pos, intent, ip);
1638 		path = trans->paths + path_idx;
1639 	} else {
1640 		path_idx = btree_path_alloc(trans, path_pos);
1641 		path = trans->paths + path_idx;
1642 
1643 		__btree_path_get(path, intent);
1644 		path->pos			= pos;
1645 		path->btree_id			= btree_id;
1646 		path->cached			= cached;
1647 		path->uptodate			= BTREE_ITER_NEED_TRAVERSE;
1648 		path->should_be_locked		= false;
1649 		path->level			= level;
1650 		path->locks_want		= locks_want;
1651 		path->nodes_locked		= 0;
1652 		for (unsigned i = 0; i < ARRAY_SIZE(path->l); i++)
1653 			path->l[i].b		= ERR_PTR(-BCH_ERR_no_btree_node_init);
1654 #ifdef TRACK_PATH_ALLOCATED
1655 		path->ip_allocated		= ip;
1656 #endif
1657 		trans->paths_sorted		= false;
1658 	}
1659 
1660 	if (!(flags & BTREE_ITER_NOPRESERVE))
1661 		path->preserve = true;
1662 
1663 	if (path->intent_ref)
1664 		locks_want = max(locks_want, level + 1);
1665 
1666 	/*
1667 	 * If the path has locks_want greater than requested, we don't downgrade
1668 	 * it here - on transaction restart because btree node split needs to
1669 	 * upgrade locks, we might be putting/getting the iterator again.
1670 	 * Downgrading iterators only happens via bch2_trans_downgrade(), after
1671 	 * a successful transaction commit.
1672 	 */
1673 
1674 	locks_want = min(locks_want, BTREE_MAX_DEPTH);
1675 	if (locks_want > path->locks_want)
1676 		bch2_btree_path_upgrade_noupgrade_sibs(trans, path, locks_want, NULL);
1677 
1678 	return path_idx;
1679 }
1680 
1681 struct bkey_s_c bch2_btree_path_peek_slot(struct btree_path *path, struct bkey *u)
1682 {
1683 
1684 	struct btree_path_level *l = path_l(path);
1685 	struct bkey_packed *_k;
1686 	struct bkey_s_c k;
1687 
1688 	if (unlikely(!l->b))
1689 		return bkey_s_c_null;
1690 
1691 	EBUG_ON(path->uptodate != BTREE_ITER_UPTODATE);
1692 	EBUG_ON(!btree_node_locked(path, path->level));
1693 
1694 	if (!path->cached) {
1695 		_k = bch2_btree_node_iter_peek_all(&l->iter, l->b);
1696 		k = _k ? bkey_disassemble(l->b, _k, u) : bkey_s_c_null;
1697 
1698 		EBUG_ON(k.k && bkey_deleted(k.k) && bpos_eq(k.k->p, path->pos));
1699 
1700 		if (!k.k || !bpos_eq(path->pos, k.k->p))
1701 			goto hole;
1702 	} else {
1703 		struct bkey_cached *ck = (void *) path->l[0].b;
1704 
1705 		EBUG_ON(ck &&
1706 			(path->btree_id != ck->key.btree_id ||
1707 			 !bkey_eq(path->pos, ck->key.pos)));
1708 		if (!ck || !ck->valid)
1709 			return bkey_s_c_null;
1710 
1711 		*u = ck->k->k;
1712 		k = bkey_i_to_s_c(ck->k);
1713 	}
1714 
1715 	return k;
1716 hole:
1717 	bkey_init(u);
1718 	u->p = path->pos;
1719 	return (struct bkey_s_c) { u, NULL };
1720 }
1721 
1722 /* Btree iterators: */
1723 
1724 int __must_check
1725 __bch2_btree_iter_traverse(struct btree_iter *iter)
1726 {
1727 	return bch2_btree_path_traverse(iter->trans, iter->path, iter->flags);
1728 }
1729 
1730 int __must_check
1731 bch2_btree_iter_traverse(struct btree_iter *iter)
1732 {
1733 	struct btree_trans *trans = iter->trans;
1734 	int ret;
1735 
1736 	iter->path = bch2_btree_path_set_pos(trans, iter->path,
1737 					btree_iter_search_key(iter),
1738 					iter->flags & BTREE_ITER_INTENT,
1739 					btree_iter_ip_allocated(iter));
1740 
1741 	ret = bch2_btree_path_traverse(iter->trans, iter->path, iter->flags);
1742 	if (ret)
1743 		return ret;
1744 
1745 	struct btree_path *path = btree_iter_path(trans, iter);
1746 	if (btree_path_node(path, path->level))
1747 		btree_path_set_should_be_locked(path);
1748 	return 0;
1749 }
1750 
1751 /* Iterate across nodes (leaf and interior nodes) */
1752 
1753 struct btree *bch2_btree_iter_peek_node(struct btree_iter *iter)
1754 {
1755 	struct btree_trans *trans = iter->trans;
1756 	struct btree *b = NULL;
1757 	int ret;
1758 
1759 	EBUG_ON(trans->paths[iter->path].cached);
1760 	bch2_btree_iter_verify(iter);
1761 
1762 	ret = bch2_btree_path_traverse(trans, iter->path, iter->flags);
1763 	if (ret)
1764 		goto err;
1765 
1766 	struct btree_path *path = btree_iter_path(trans, iter);
1767 	b = btree_path_node(path, path->level);
1768 	if (!b)
1769 		goto out;
1770 
1771 	BUG_ON(bpos_lt(b->key.k.p, iter->pos));
1772 
1773 	bkey_init(&iter->k);
1774 	iter->k.p = iter->pos = b->key.k.p;
1775 
1776 	iter->path = bch2_btree_path_set_pos(trans, iter->path, b->key.k.p,
1777 					iter->flags & BTREE_ITER_INTENT,
1778 					btree_iter_ip_allocated(iter));
1779 	btree_path_set_should_be_locked(btree_iter_path(trans, iter));
1780 out:
1781 	bch2_btree_iter_verify_entry_exit(iter);
1782 	bch2_btree_iter_verify(iter);
1783 
1784 	return b;
1785 err:
1786 	b = ERR_PTR(ret);
1787 	goto out;
1788 }
1789 
1790 struct btree *bch2_btree_iter_peek_node_and_restart(struct btree_iter *iter)
1791 {
1792 	struct btree *b;
1793 
1794 	while (b = bch2_btree_iter_peek_node(iter),
1795 	       bch2_err_matches(PTR_ERR_OR_ZERO(b), BCH_ERR_transaction_restart))
1796 		bch2_trans_begin(iter->trans);
1797 
1798 	return b;
1799 }
1800 
1801 struct btree *bch2_btree_iter_next_node(struct btree_iter *iter)
1802 {
1803 	struct btree_trans *trans = iter->trans;
1804 	struct btree *b = NULL;
1805 	int ret;
1806 
1807 	EBUG_ON(trans->paths[iter->path].cached);
1808 	bch2_trans_verify_not_in_restart(trans);
1809 	bch2_btree_iter_verify(iter);
1810 
1811 	struct btree_path *path = btree_iter_path(trans, iter);
1812 
1813 	/* already at end? */
1814 	if (!btree_path_node(path, path->level))
1815 		return NULL;
1816 
1817 	/* got to end? */
1818 	if (!btree_path_node(path, path->level + 1)) {
1819 		btree_path_set_level_up(trans, path);
1820 		return NULL;
1821 	}
1822 
1823 	if (!bch2_btree_node_relock(trans, path, path->level + 1)) {
1824 		__bch2_btree_path_unlock(trans, path);
1825 		path->l[path->level].b		= ERR_PTR(-BCH_ERR_no_btree_node_relock);
1826 		path->l[path->level + 1].b	= ERR_PTR(-BCH_ERR_no_btree_node_relock);
1827 		btree_path_set_dirty(path, BTREE_ITER_NEED_TRAVERSE);
1828 		trace_and_count(trans->c, trans_restart_relock_next_node, trans, _THIS_IP_, path);
1829 		ret = btree_trans_restart(trans, BCH_ERR_transaction_restart_relock);
1830 		goto err;
1831 	}
1832 
1833 	b = btree_path_node(path, path->level + 1);
1834 
1835 	if (bpos_eq(iter->pos, b->key.k.p)) {
1836 		__btree_path_set_level_up(trans, path, path->level++);
1837 	} else {
1838 		/*
1839 		 * Haven't gotten to the end of the parent node: go back down to
1840 		 * the next child node
1841 		 */
1842 		iter->path = bch2_btree_path_set_pos(trans, iter->path,
1843 					bpos_successor(iter->pos),
1844 					iter->flags & BTREE_ITER_INTENT,
1845 					btree_iter_ip_allocated(iter));
1846 
1847 		path = btree_iter_path(trans, iter);
1848 		btree_path_set_level_down(trans, path, iter->min_depth);
1849 
1850 		ret = bch2_btree_path_traverse(trans, iter->path, iter->flags);
1851 		if (ret)
1852 			goto err;
1853 
1854 		path = btree_iter_path(trans, iter);
1855 		b = path->l[path->level].b;
1856 	}
1857 
1858 	bkey_init(&iter->k);
1859 	iter->k.p = iter->pos = b->key.k.p;
1860 
1861 	iter->path = bch2_btree_path_set_pos(trans, iter->path, b->key.k.p,
1862 					iter->flags & BTREE_ITER_INTENT,
1863 					btree_iter_ip_allocated(iter));
1864 	btree_path_set_should_be_locked(btree_iter_path(trans, iter));
1865 	EBUG_ON(btree_iter_path(trans, iter)->uptodate);
1866 out:
1867 	bch2_btree_iter_verify_entry_exit(iter);
1868 	bch2_btree_iter_verify(iter);
1869 
1870 	return b;
1871 err:
1872 	b = ERR_PTR(ret);
1873 	goto out;
1874 }
1875 
1876 /* Iterate across keys (in leaf nodes only) */
1877 
1878 inline bool bch2_btree_iter_advance(struct btree_iter *iter)
1879 {
1880 	struct bpos pos = iter->k.p;
1881 	bool ret = !(iter->flags & BTREE_ITER_ALL_SNAPSHOTS
1882 		     ? bpos_eq(pos, SPOS_MAX)
1883 		     : bkey_eq(pos, SPOS_MAX));
1884 
1885 	if (ret && !(iter->flags & BTREE_ITER_IS_EXTENTS))
1886 		pos = bkey_successor(iter, pos);
1887 	bch2_btree_iter_set_pos(iter, pos);
1888 	return ret;
1889 }
1890 
1891 inline bool bch2_btree_iter_rewind(struct btree_iter *iter)
1892 {
1893 	struct bpos pos = bkey_start_pos(&iter->k);
1894 	bool ret = !(iter->flags & BTREE_ITER_ALL_SNAPSHOTS
1895 		     ? bpos_eq(pos, POS_MIN)
1896 		     : bkey_eq(pos, POS_MIN));
1897 
1898 	if (ret && !(iter->flags & BTREE_ITER_IS_EXTENTS))
1899 		pos = bkey_predecessor(iter, pos);
1900 	bch2_btree_iter_set_pos(iter, pos);
1901 	return ret;
1902 }
1903 
1904 static noinline
1905 void bch2_btree_trans_peek_prev_updates(struct btree_trans *trans, struct btree_iter *iter,
1906 					struct bkey_s_c *k)
1907 {
1908 	struct bpos end = path_l(btree_iter_path(trans, iter))->b->data->min_key;
1909 
1910 	trans_for_each_update(trans, i)
1911 		if (!i->key_cache_already_flushed &&
1912 		    i->btree_id == iter->btree_id &&
1913 		    bpos_le(i->k->k.p, iter->pos) &&
1914 		    bpos_ge(i->k->k.p, k->k ? k->k->p : end)) {
1915 			iter->k = i->k->k;
1916 			*k = bkey_i_to_s_c(i->k);
1917 		}
1918 }
1919 
1920 static noinline
1921 void bch2_btree_trans_peek_updates(struct btree_trans *trans, struct btree_iter *iter,
1922 				   struct bkey_s_c *k)
1923 {
1924 	struct btree_path *path = btree_iter_path(trans, iter);
1925 	struct bpos end = path_l(path)->b->key.k.p;
1926 
1927 	trans_for_each_update(trans, i)
1928 		if (!i->key_cache_already_flushed &&
1929 		    i->btree_id == iter->btree_id &&
1930 		    bpos_ge(i->k->k.p, path->pos) &&
1931 		    bpos_le(i->k->k.p, k->k ? k->k->p : end)) {
1932 			iter->k = i->k->k;
1933 			*k = bkey_i_to_s_c(i->k);
1934 		}
1935 }
1936 
1937 static noinline
1938 void bch2_btree_trans_peek_slot_updates(struct btree_trans *trans, struct btree_iter *iter,
1939 					struct bkey_s_c *k)
1940 {
1941 	trans_for_each_update(trans, i)
1942 		if (!i->key_cache_already_flushed &&
1943 		    i->btree_id == iter->btree_id &&
1944 		    bpos_eq(i->k->k.p, iter->pos)) {
1945 			iter->k = i->k->k;
1946 			*k = bkey_i_to_s_c(i->k);
1947 		}
1948 }
1949 
1950 static struct bkey_i *bch2_btree_journal_peek(struct btree_trans *trans,
1951 					      struct btree_iter *iter,
1952 					      struct bpos end_pos)
1953 {
1954 	struct btree_path *path = btree_iter_path(trans, iter);
1955 
1956 	return bch2_journal_keys_peek_upto(trans->c, iter->btree_id,
1957 					   path->level,
1958 					   path->pos,
1959 					   end_pos,
1960 					   &iter->journal_idx);
1961 }
1962 
1963 static noinline
1964 struct bkey_s_c btree_trans_peek_slot_journal(struct btree_trans *trans,
1965 					      struct btree_iter *iter)
1966 {
1967 	struct btree_path *path = btree_iter_path(trans, iter);
1968 	struct bkey_i *k = bch2_btree_journal_peek(trans, iter, path->pos);
1969 
1970 	if (k) {
1971 		iter->k = k->k;
1972 		return bkey_i_to_s_c(k);
1973 	} else {
1974 		return bkey_s_c_null;
1975 	}
1976 }
1977 
1978 static noinline
1979 struct bkey_s_c btree_trans_peek_journal(struct btree_trans *trans,
1980 					 struct btree_iter *iter,
1981 					 struct bkey_s_c k)
1982 {
1983 	struct btree_path *path = btree_iter_path(trans, iter);
1984 	struct bkey_i *next_journal =
1985 		bch2_btree_journal_peek(trans, iter,
1986 				k.k ? k.k->p : path_l(path)->b->key.k.p);
1987 
1988 	if (next_journal) {
1989 		iter->k = next_journal->k;
1990 		k = bkey_i_to_s_c(next_journal);
1991 	}
1992 
1993 	return k;
1994 }
1995 
1996 /*
1997  * Checks btree key cache for key at iter->pos and returns it if present, or
1998  * bkey_s_c_null:
1999  */
2000 static noinline
2001 struct bkey_s_c btree_trans_peek_key_cache(struct btree_iter *iter, struct bpos pos)
2002 {
2003 	struct btree_trans *trans = iter->trans;
2004 	struct bch_fs *c = trans->c;
2005 	struct bkey u;
2006 	struct bkey_s_c k;
2007 	int ret;
2008 
2009 	if ((iter->flags & BTREE_ITER_KEY_CACHE_FILL) &&
2010 	    bpos_eq(iter->pos, pos))
2011 		return bkey_s_c_null;
2012 
2013 	if (!bch2_btree_key_cache_find(c, iter->btree_id, pos))
2014 		return bkey_s_c_null;
2015 
2016 	if (!iter->key_cache_path)
2017 		iter->key_cache_path = bch2_path_get(trans, iter->btree_id, pos,
2018 						     iter->flags & BTREE_ITER_INTENT, 0,
2019 						     iter->flags|BTREE_ITER_CACHED|
2020 						     BTREE_ITER_CACHED_NOFILL,
2021 						     _THIS_IP_);
2022 
2023 	iter->key_cache_path = bch2_btree_path_set_pos(trans, iter->key_cache_path, pos,
2024 					iter->flags & BTREE_ITER_INTENT,
2025 					btree_iter_ip_allocated(iter));
2026 
2027 	ret =   bch2_btree_path_traverse(trans, iter->key_cache_path,
2028 					 iter->flags|BTREE_ITER_CACHED) ?:
2029 		bch2_btree_path_relock(trans, btree_iter_path(trans, iter), _THIS_IP_);
2030 	if (unlikely(ret))
2031 		return bkey_s_c_err(ret);
2032 
2033 	btree_path_set_should_be_locked(trans->paths + iter->key_cache_path);
2034 
2035 	k = bch2_btree_path_peek_slot(trans->paths + iter->key_cache_path, &u);
2036 	if (k.k && !bkey_err(k)) {
2037 		iter->k = u;
2038 		k.k = &iter->k;
2039 	}
2040 	return k;
2041 }
2042 
2043 static struct bkey_s_c __bch2_btree_iter_peek(struct btree_iter *iter, struct bpos search_key)
2044 {
2045 	struct btree_trans *trans = iter->trans;
2046 	struct bkey_s_c k, k2;
2047 	int ret;
2048 
2049 	EBUG_ON(btree_iter_path(trans, iter)->cached);
2050 	bch2_btree_iter_verify(iter);
2051 
2052 	while (1) {
2053 		struct btree_path_level *l;
2054 
2055 		iter->path = bch2_btree_path_set_pos(trans, iter->path, search_key,
2056 					iter->flags & BTREE_ITER_INTENT,
2057 					btree_iter_ip_allocated(iter));
2058 
2059 		ret = bch2_btree_path_traverse(trans, iter->path, iter->flags);
2060 		if (unlikely(ret)) {
2061 			/* ensure that iter->k is consistent with iter->pos: */
2062 			bch2_btree_iter_set_pos(iter, iter->pos);
2063 			k = bkey_s_c_err(ret);
2064 			goto out;
2065 		}
2066 
2067 		struct btree_path *path = btree_iter_path(trans, iter);
2068 		l = path_l(path);
2069 
2070 		if (unlikely(!l->b)) {
2071 			/* No btree nodes at requested level: */
2072 			bch2_btree_iter_set_pos(iter, SPOS_MAX);
2073 			k = bkey_s_c_null;
2074 			goto out;
2075 		}
2076 
2077 		btree_path_set_should_be_locked(path);
2078 
2079 		k = btree_path_level_peek_all(trans->c, l, &iter->k);
2080 
2081 		if (unlikely(iter->flags & BTREE_ITER_WITH_KEY_CACHE) &&
2082 		    k.k &&
2083 		    (k2 = btree_trans_peek_key_cache(iter, k.k->p)).k) {
2084 			k = k2;
2085 			ret = bkey_err(k);
2086 			if (ret) {
2087 				bch2_btree_iter_set_pos(iter, iter->pos);
2088 				goto out;
2089 			}
2090 		}
2091 
2092 		if (unlikely(iter->flags & BTREE_ITER_WITH_JOURNAL))
2093 			k = btree_trans_peek_journal(trans, iter, k);
2094 
2095 		if (unlikely((iter->flags & BTREE_ITER_WITH_UPDATES) &&
2096 			     trans->nr_updates))
2097 			bch2_btree_trans_peek_updates(trans, iter, &k);
2098 
2099 		if (k.k && bkey_deleted(k.k)) {
2100 			/*
2101 			 * If we've got a whiteout, and it's after the search
2102 			 * key, advance the search key to the whiteout instead
2103 			 * of just after the whiteout - it might be a btree
2104 			 * whiteout, with a real key at the same position, since
2105 			 * in the btree deleted keys sort before non deleted.
2106 			 */
2107 			search_key = !bpos_eq(search_key, k.k->p)
2108 				? k.k->p
2109 				: bpos_successor(k.k->p);
2110 			continue;
2111 		}
2112 
2113 		if (likely(k.k)) {
2114 			break;
2115 		} else if (likely(!bpos_eq(l->b->key.k.p, SPOS_MAX))) {
2116 			/* Advance to next leaf node: */
2117 			search_key = bpos_successor(l->b->key.k.p);
2118 		} else {
2119 			/* End of btree: */
2120 			bch2_btree_iter_set_pos(iter, SPOS_MAX);
2121 			k = bkey_s_c_null;
2122 			goto out;
2123 		}
2124 	}
2125 out:
2126 	bch2_btree_iter_verify(iter);
2127 
2128 	return k;
2129 }
2130 
2131 /**
2132  * bch2_btree_iter_peek_upto() - returns first key greater than or equal to
2133  * iterator's current position
2134  * @iter:	iterator to peek from
2135  * @end:	search limit: returns keys less than or equal to @end
2136  *
2137  * Returns:	key if found, or an error extractable with bkey_err().
2138  */
2139 struct bkey_s_c bch2_btree_iter_peek_upto(struct btree_iter *iter, struct bpos end)
2140 {
2141 	struct btree_trans *trans = iter->trans;
2142 	struct bpos search_key = btree_iter_search_key(iter);
2143 	struct bkey_s_c k;
2144 	struct bpos iter_pos;
2145 	int ret;
2146 
2147 	EBUG_ON((iter->flags & BTREE_ITER_FILTER_SNAPSHOTS) && bkey_eq(end, POS_MAX));
2148 
2149 	if (iter->update_path) {
2150 		bch2_path_put_nokeep(trans, iter->update_path,
2151 				     iter->flags & BTREE_ITER_INTENT);
2152 		iter->update_path = 0;
2153 	}
2154 
2155 	bch2_btree_iter_verify_entry_exit(iter);
2156 
2157 	while (1) {
2158 		k = __bch2_btree_iter_peek(iter, search_key);
2159 		if (unlikely(!k.k))
2160 			goto end;
2161 		if (unlikely(bkey_err(k)))
2162 			goto out_no_locked;
2163 
2164 		/*
2165 		 * We need to check against @end before FILTER_SNAPSHOTS because
2166 		 * if we get to a different inode that requested we might be
2167 		 * seeing keys for a different snapshot tree that will all be
2168 		 * filtered out.
2169 		 *
2170 		 * But we can't do the full check here, because bkey_start_pos()
2171 		 * isn't monotonically increasing before FILTER_SNAPSHOTS, and
2172 		 * that's what we check against in extents mode:
2173 		 */
2174 		if (unlikely(!(iter->flags & BTREE_ITER_IS_EXTENTS)
2175 			     ? bkey_gt(k.k->p, end)
2176 			     : k.k->p.inode > end.inode))
2177 			goto end;
2178 
2179 		if (iter->update_path &&
2180 		    !bkey_eq(trans->paths[iter->update_path].pos, k.k->p)) {
2181 			bch2_path_put_nokeep(trans, iter->update_path,
2182 					     iter->flags & BTREE_ITER_INTENT);
2183 			iter->update_path = 0;
2184 		}
2185 
2186 		if ((iter->flags & BTREE_ITER_FILTER_SNAPSHOTS) &&
2187 		    (iter->flags & BTREE_ITER_INTENT) &&
2188 		    !(iter->flags & BTREE_ITER_IS_EXTENTS) &&
2189 		    !iter->update_path) {
2190 			struct bpos pos = k.k->p;
2191 
2192 			if (pos.snapshot < iter->snapshot) {
2193 				search_key = bpos_successor(k.k->p);
2194 				continue;
2195 			}
2196 
2197 			pos.snapshot = iter->snapshot;
2198 
2199 			/*
2200 			 * advance, same as on exit for iter->path, but only up
2201 			 * to snapshot
2202 			 */
2203 			__btree_path_get(trans->paths + iter->path, iter->flags & BTREE_ITER_INTENT);
2204 			iter->update_path = iter->path;
2205 
2206 			iter->update_path = bch2_btree_path_set_pos(trans,
2207 						iter->update_path, pos,
2208 						iter->flags & BTREE_ITER_INTENT,
2209 						_THIS_IP_);
2210 			ret = bch2_btree_path_traverse(trans, iter->update_path, iter->flags);
2211 			if (unlikely(ret)) {
2212 				k = bkey_s_c_err(ret);
2213 				goto out_no_locked;
2214 			}
2215 		}
2216 
2217 		/*
2218 		 * We can never have a key in a leaf node at POS_MAX, so
2219 		 * we don't have to check these successor() calls:
2220 		 */
2221 		if ((iter->flags & BTREE_ITER_FILTER_SNAPSHOTS) &&
2222 		    !bch2_snapshot_is_ancestor(trans->c,
2223 					       iter->snapshot,
2224 					       k.k->p.snapshot)) {
2225 			search_key = bpos_successor(k.k->p);
2226 			continue;
2227 		}
2228 
2229 		if (bkey_whiteout(k.k) &&
2230 		    !(iter->flags & BTREE_ITER_ALL_SNAPSHOTS)) {
2231 			search_key = bkey_successor(iter, k.k->p);
2232 			continue;
2233 		}
2234 
2235 		/*
2236 		 * iter->pos should be mononotically increasing, and always be
2237 		 * equal to the key we just returned - except extents can
2238 		 * straddle iter->pos:
2239 		 */
2240 		if (!(iter->flags & BTREE_ITER_IS_EXTENTS))
2241 			iter_pos = k.k->p;
2242 		else
2243 			iter_pos = bkey_max(iter->pos, bkey_start_pos(k.k));
2244 
2245 		if (unlikely(!(iter->flags & BTREE_ITER_IS_EXTENTS)
2246 			     ? bkey_gt(iter_pos, end)
2247 			     : bkey_ge(iter_pos, end)))
2248 			goto end;
2249 
2250 		break;
2251 	}
2252 
2253 	iter->pos = iter_pos;
2254 
2255 	iter->path = bch2_btree_path_set_pos(trans, iter->path, k.k->p,
2256 				iter->flags & BTREE_ITER_INTENT,
2257 				btree_iter_ip_allocated(iter));
2258 
2259 	btree_path_set_should_be_locked(btree_iter_path(trans, iter));
2260 out_no_locked:
2261 	if (iter->update_path) {
2262 		ret = bch2_btree_path_relock(trans, trans->paths + iter->update_path, _THIS_IP_);
2263 		if (unlikely(ret))
2264 			k = bkey_s_c_err(ret);
2265 		else
2266 			btree_path_set_should_be_locked(trans->paths + iter->update_path);
2267 	}
2268 
2269 	if (!(iter->flags & BTREE_ITER_ALL_SNAPSHOTS))
2270 		iter->pos.snapshot = iter->snapshot;
2271 
2272 	ret = bch2_btree_iter_verify_ret(iter, k);
2273 	if (unlikely(ret)) {
2274 		bch2_btree_iter_set_pos(iter, iter->pos);
2275 		k = bkey_s_c_err(ret);
2276 	}
2277 
2278 	bch2_btree_iter_verify_entry_exit(iter);
2279 
2280 	return k;
2281 end:
2282 	bch2_btree_iter_set_pos(iter, end);
2283 	k = bkey_s_c_null;
2284 	goto out_no_locked;
2285 }
2286 
2287 /**
2288  * bch2_btree_iter_next() - returns first key greater than iterator's current
2289  * position
2290  * @iter:	iterator to peek from
2291  *
2292  * Returns:	key if found, or an error extractable with bkey_err().
2293  */
2294 struct bkey_s_c bch2_btree_iter_next(struct btree_iter *iter)
2295 {
2296 	if (!bch2_btree_iter_advance(iter))
2297 		return bkey_s_c_null;
2298 
2299 	return bch2_btree_iter_peek(iter);
2300 }
2301 
2302 /**
2303  * bch2_btree_iter_peek_prev() - returns first key less than or equal to
2304  * iterator's current position
2305  * @iter:	iterator to peek from
2306  *
2307  * Returns:	key if found, or an error extractable with bkey_err().
2308  */
2309 struct bkey_s_c bch2_btree_iter_peek_prev(struct btree_iter *iter)
2310 {
2311 	struct btree_trans *trans = iter->trans;
2312 	struct bpos search_key = iter->pos;
2313 	struct bkey_s_c k;
2314 	struct bkey saved_k;
2315 	const struct bch_val *saved_v;
2316 	btree_path_idx_t saved_path = 0;
2317 	int ret;
2318 
2319 	EBUG_ON(btree_iter_path(trans, iter)->cached ||
2320 		btree_iter_path(trans, iter)->level);
2321 
2322 	if (iter->flags & BTREE_ITER_WITH_JOURNAL)
2323 		return bkey_s_c_err(-BCH_ERR_btree_iter_with_journal_not_supported);
2324 
2325 	bch2_btree_iter_verify(iter);
2326 	bch2_btree_iter_verify_entry_exit(iter);
2327 
2328 	if (iter->flags & BTREE_ITER_FILTER_SNAPSHOTS)
2329 		search_key.snapshot = U32_MAX;
2330 
2331 	while (1) {
2332 		iter->path = bch2_btree_path_set_pos(trans, iter->path, search_key,
2333 						iter->flags & BTREE_ITER_INTENT,
2334 						btree_iter_ip_allocated(iter));
2335 
2336 		ret = bch2_btree_path_traverse(trans, iter->path, iter->flags);
2337 		if (unlikely(ret)) {
2338 			/* ensure that iter->k is consistent with iter->pos: */
2339 			bch2_btree_iter_set_pos(iter, iter->pos);
2340 			k = bkey_s_c_err(ret);
2341 			goto out_no_locked;
2342 		}
2343 
2344 		struct btree_path *path = btree_iter_path(trans, iter);
2345 
2346 		k = btree_path_level_peek(trans, path, &path->l[0], &iter->k);
2347 		if (!k.k ||
2348 		    ((iter->flags & BTREE_ITER_IS_EXTENTS)
2349 		     ? bpos_ge(bkey_start_pos(k.k), search_key)
2350 		     : bpos_gt(k.k->p, search_key)))
2351 			k = btree_path_level_prev(trans, path, &path->l[0], &iter->k);
2352 
2353 		if (unlikely((iter->flags & BTREE_ITER_WITH_UPDATES) &&
2354 			     trans->nr_updates))
2355 			bch2_btree_trans_peek_prev_updates(trans, iter, &k);
2356 
2357 		if (likely(k.k)) {
2358 			if (iter->flags & BTREE_ITER_FILTER_SNAPSHOTS) {
2359 				if (k.k->p.snapshot == iter->snapshot)
2360 					goto got_key;
2361 
2362 				/*
2363 				 * If we have a saved candidate, and we're no
2364 				 * longer at the same _key_ (not pos), return
2365 				 * that candidate
2366 				 */
2367 				if (saved_path && !bkey_eq(k.k->p, saved_k.p)) {
2368 					bch2_path_put_nokeep(trans, iter->path,
2369 						      iter->flags & BTREE_ITER_INTENT);
2370 					iter->path = saved_path;
2371 					saved_path = 0;
2372 					iter->k	= saved_k;
2373 					k.v	= saved_v;
2374 					goto got_key;
2375 				}
2376 
2377 				if (bch2_snapshot_is_ancestor(trans->c,
2378 							      iter->snapshot,
2379 							      k.k->p.snapshot)) {
2380 					if (saved_path)
2381 						bch2_path_put_nokeep(trans, saved_path,
2382 						      iter->flags & BTREE_ITER_INTENT);
2383 					saved_path = btree_path_clone(trans, iter->path,
2384 								iter->flags & BTREE_ITER_INTENT);
2385 					path = btree_iter_path(trans, iter);
2386 					saved_k = *k.k;
2387 					saved_v = k.v;
2388 				}
2389 
2390 				search_key = bpos_predecessor(k.k->p);
2391 				continue;
2392 			}
2393 got_key:
2394 			if (bkey_whiteout(k.k) &&
2395 			    !(iter->flags & BTREE_ITER_ALL_SNAPSHOTS)) {
2396 				search_key = bkey_predecessor(iter, k.k->p);
2397 				if (iter->flags & BTREE_ITER_FILTER_SNAPSHOTS)
2398 					search_key.snapshot = U32_MAX;
2399 				continue;
2400 			}
2401 
2402 			btree_path_set_should_be_locked(path);
2403 			break;
2404 		} else if (likely(!bpos_eq(path->l[0].b->data->min_key, POS_MIN))) {
2405 			/* Advance to previous leaf node: */
2406 			search_key = bpos_predecessor(path->l[0].b->data->min_key);
2407 		} else {
2408 			/* Start of btree: */
2409 			bch2_btree_iter_set_pos(iter, POS_MIN);
2410 			k = bkey_s_c_null;
2411 			goto out_no_locked;
2412 		}
2413 	}
2414 
2415 	EBUG_ON(bkey_gt(bkey_start_pos(k.k), iter->pos));
2416 
2417 	/* Extents can straddle iter->pos: */
2418 	if (bkey_lt(k.k->p, iter->pos))
2419 		iter->pos = k.k->p;
2420 
2421 	if (iter->flags & BTREE_ITER_FILTER_SNAPSHOTS)
2422 		iter->pos.snapshot = iter->snapshot;
2423 out_no_locked:
2424 	if (saved_path)
2425 		bch2_path_put_nokeep(trans, saved_path, iter->flags & BTREE_ITER_INTENT);
2426 
2427 	bch2_btree_iter_verify_entry_exit(iter);
2428 	bch2_btree_iter_verify(iter);
2429 
2430 	return k;
2431 }
2432 
2433 /**
2434  * bch2_btree_iter_prev() - returns first key less than iterator's current
2435  * position
2436  * @iter:	iterator to peek from
2437  *
2438  * Returns:	key if found, or an error extractable with bkey_err().
2439  */
2440 struct bkey_s_c bch2_btree_iter_prev(struct btree_iter *iter)
2441 {
2442 	if (!bch2_btree_iter_rewind(iter))
2443 		return bkey_s_c_null;
2444 
2445 	return bch2_btree_iter_peek_prev(iter);
2446 }
2447 
2448 struct bkey_s_c bch2_btree_iter_peek_slot(struct btree_iter *iter)
2449 {
2450 	struct btree_trans *trans = iter->trans;
2451 	struct bpos search_key;
2452 	struct bkey_s_c k;
2453 	int ret;
2454 
2455 	bch2_btree_iter_verify(iter);
2456 	bch2_btree_iter_verify_entry_exit(iter);
2457 	EBUG_ON(btree_iter_path(trans, iter)->level && (iter->flags & BTREE_ITER_WITH_KEY_CACHE));
2458 
2459 	/* extents can't span inode numbers: */
2460 	if ((iter->flags & BTREE_ITER_IS_EXTENTS) &&
2461 	    unlikely(iter->pos.offset == KEY_OFFSET_MAX)) {
2462 		if (iter->pos.inode == KEY_INODE_MAX)
2463 			return bkey_s_c_null;
2464 
2465 		bch2_btree_iter_set_pos(iter, bpos_nosnap_successor(iter->pos));
2466 	}
2467 
2468 	search_key = btree_iter_search_key(iter);
2469 	iter->path = bch2_btree_path_set_pos(trans, iter->path, search_key,
2470 					iter->flags & BTREE_ITER_INTENT,
2471 					btree_iter_ip_allocated(iter));
2472 
2473 	ret = bch2_btree_path_traverse(trans, iter->path, iter->flags);
2474 	if (unlikely(ret)) {
2475 		k = bkey_s_c_err(ret);
2476 		goto out_no_locked;
2477 	}
2478 
2479 	if ((iter->flags & BTREE_ITER_CACHED) ||
2480 	    !(iter->flags & (BTREE_ITER_IS_EXTENTS|BTREE_ITER_FILTER_SNAPSHOTS))) {
2481 		k = bkey_s_c_null;
2482 
2483 		if (unlikely((iter->flags & BTREE_ITER_WITH_UPDATES) &&
2484 			     trans->nr_updates)) {
2485 			bch2_btree_trans_peek_slot_updates(trans, iter, &k);
2486 			if (k.k)
2487 				goto out;
2488 		}
2489 
2490 		if (unlikely(iter->flags & BTREE_ITER_WITH_JOURNAL) &&
2491 		    (k = btree_trans_peek_slot_journal(trans, iter)).k)
2492 			goto out;
2493 
2494 		if (unlikely(iter->flags & BTREE_ITER_WITH_KEY_CACHE) &&
2495 		    (k = btree_trans_peek_key_cache(iter, iter->pos)).k) {
2496 			if (!bkey_err(k))
2497 				iter->k = *k.k;
2498 			/* We're not returning a key from iter->path: */
2499 			goto out_no_locked;
2500 		}
2501 
2502 		k = bch2_btree_path_peek_slot(trans->paths + iter->path, &iter->k);
2503 		if (unlikely(!k.k))
2504 			goto out_no_locked;
2505 	} else {
2506 		struct bpos next;
2507 		struct bpos end = iter->pos;
2508 
2509 		if (iter->flags & BTREE_ITER_IS_EXTENTS)
2510 			end.offset = U64_MAX;
2511 
2512 		EBUG_ON(btree_iter_path(trans, iter)->level);
2513 
2514 		if (iter->flags & BTREE_ITER_INTENT) {
2515 			struct btree_iter iter2;
2516 
2517 			bch2_trans_copy_iter(&iter2, iter);
2518 			k = bch2_btree_iter_peek_upto(&iter2, end);
2519 
2520 			if (k.k && !bkey_err(k)) {
2521 				swap(iter->key_cache_path, iter2.key_cache_path);
2522 				iter->k = iter2.k;
2523 				k.k = &iter->k;
2524 			}
2525 			bch2_trans_iter_exit(trans, &iter2);
2526 		} else {
2527 			struct bpos pos = iter->pos;
2528 
2529 			k = bch2_btree_iter_peek_upto(iter, end);
2530 			if (unlikely(bkey_err(k)))
2531 				bch2_btree_iter_set_pos(iter, pos);
2532 			else
2533 				iter->pos = pos;
2534 		}
2535 
2536 		if (unlikely(bkey_err(k)))
2537 			goto out_no_locked;
2538 
2539 		next = k.k ? bkey_start_pos(k.k) : POS_MAX;
2540 
2541 		if (bkey_lt(iter->pos, next)) {
2542 			bkey_init(&iter->k);
2543 			iter->k.p = iter->pos;
2544 
2545 			if (iter->flags & BTREE_ITER_IS_EXTENTS) {
2546 				bch2_key_resize(&iter->k,
2547 						min_t(u64, KEY_SIZE_MAX,
2548 						      (next.inode == iter->pos.inode
2549 						       ? next.offset
2550 						       : KEY_OFFSET_MAX) -
2551 						      iter->pos.offset));
2552 				EBUG_ON(!iter->k.size);
2553 			}
2554 
2555 			k = (struct bkey_s_c) { &iter->k, NULL };
2556 		}
2557 	}
2558 out:
2559 	btree_path_set_should_be_locked(btree_iter_path(trans, iter));
2560 out_no_locked:
2561 	bch2_btree_iter_verify_entry_exit(iter);
2562 	bch2_btree_iter_verify(iter);
2563 	ret = bch2_btree_iter_verify_ret(iter, k);
2564 	if (unlikely(ret))
2565 		return bkey_s_c_err(ret);
2566 
2567 	return k;
2568 }
2569 
2570 struct bkey_s_c bch2_btree_iter_next_slot(struct btree_iter *iter)
2571 {
2572 	if (!bch2_btree_iter_advance(iter))
2573 		return bkey_s_c_null;
2574 
2575 	return bch2_btree_iter_peek_slot(iter);
2576 }
2577 
2578 struct bkey_s_c bch2_btree_iter_prev_slot(struct btree_iter *iter)
2579 {
2580 	if (!bch2_btree_iter_rewind(iter))
2581 		return bkey_s_c_null;
2582 
2583 	return bch2_btree_iter_peek_slot(iter);
2584 }
2585 
2586 struct bkey_s_c bch2_btree_iter_peek_and_restart_outlined(struct btree_iter *iter)
2587 {
2588 	struct bkey_s_c k;
2589 
2590 	while (btree_trans_too_many_iters(iter->trans) ||
2591 	       (k = bch2_btree_iter_peek_type(iter, iter->flags),
2592 		bch2_err_matches(bkey_err(k), BCH_ERR_transaction_restart)))
2593 		bch2_trans_begin(iter->trans);
2594 
2595 	return k;
2596 }
2597 
2598 /* new transactional stuff: */
2599 
2600 #ifdef CONFIG_BCACHEFS_DEBUG
2601 static void btree_trans_verify_sorted_refs(struct btree_trans *trans)
2602 {
2603 	struct btree_path *path;
2604 	unsigned i;
2605 
2606 	BUG_ON(trans->nr_sorted != bitmap_weight(trans->paths_allocated, trans->nr_paths) - 1);
2607 
2608 	trans_for_each_path(trans, path, i) {
2609 		BUG_ON(path->sorted_idx >= trans->nr_sorted);
2610 		BUG_ON(trans->sorted[path->sorted_idx] != i);
2611 	}
2612 
2613 	for (i = 0; i < trans->nr_sorted; i++) {
2614 		unsigned idx = trans->sorted[i];
2615 
2616 		BUG_ON(!test_bit(idx, trans->paths_allocated));
2617 		BUG_ON(trans->paths[idx].sorted_idx != i);
2618 	}
2619 }
2620 
2621 static void btree_trans_verify_sorted(struct btree_trans *trans)
2622 {
2623 	struct btree_path *path, *prev = NULL;
2624 	struct trans_for_each_path_inorder_iter iter;
2625 
2626 	if (!bch2_debug_check_iterators)
2627 		return;
2628 
2629 	trans_for_each_path_inorder(trans, path, iter) {
2630 		if (prev && btree_path_cmp(prev, path) > 0) {
2631 			__bch2_dump_trans_paths_updates(trans, true);
2632 			panic("trans paths out of order!\n");
2633 		}
2634 		prev = path;
2635 	}
2636 }
2637 #else
2638 static inline void btree_trans_verify_sorted_refs(struct btree_trans *trans) {}
2639 static inline void btree_trans_verify_sorted(struct btree_trans *trans) {}
2640 #endif
2641 
2642 void __bch2_btree_trans_sort_paths(struct btree_trans *trans)
2643 {
2644 	int i, l = 0, r = trans->nr_sorted, inc = 1;
2645 	bool swapped;
2646 
2647 	btree_trans_verify_sorted_refs(trans);
2648 
2649 	if (trans->paths_sorted)
2650 		goto out;
2651 
2652 	/*
2653 	 * Cocktail shaker sort: this is efficient because iterators will be
2654 	 * mostly sorted.
2655 	 */
2656 	do {
2657 		swapped = false;
2658 
2659 		for (i = inc > 0 ? l : r - 2;
2660 		     i + 1 < r && i >= l;
2661 		     i += inc) {
2662 			if (btree_path_cmp(trans->paths + trans->sorted[i],
2663 					   trans->paths + trans->sorted[i + 1]) > 0) {
2664 				swap(trans->sorted[i], trans->sorted[i + 1]);
2665 				trans->paths[trans->sorted[i]].sorted_idx = i;
2666 				trans->paths[trans->sorted[i + 1]].sorted_idx = i + 1;
2667 				swapped = true;
2668 			}
2669 		}
2670 
2671 		if (inc > 0)
2672 			--r;
2673 		else
2674 			l++;
2675 		inc = -inc;
2676 	} while (swapped);
2677 
2678 	trans->paths_sorted = true;
2679 out:
2680 	btree_trans_verify_sorted(trans);
2681 }
2682 
2683 static inline void btree_path_list_remove(struct btree_trans *trans,
2684 					  struct btree_path *path)
2685 {
2686 	EBUG_ON(path->sorted_idx >= trans->nr_sorted);
2687 #ifdef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
2688 	trans->nr_sorted--;
2689 	memmove_u64s_down_small(trans->sorted + path->sorted_idx,
2690 				trans->sorted + path->sorted_idx + 1,
2691 				DIV_ROUND_UP(trans->nr_sorted - path->sorted_idx,
2692 					     sizeof(u64) / sizeof(btree_path_idx_t)));
2693 #else
2694 	array_remove_item(trans->sorted, trans->nr_sorted, path->sorted_idx);
2695 #endif
2696 	for (unsigned i = path->sorted_idx; i < trans->nr_sorted; i++)
2697 		trans->paths[trans->sorted[i]].sorted_idx = i;
2698 }
2699 
2700 static inline void btree_path_list_add(struct btree_trans *trans,
2701 				       btree_path_idx_t pos,
2702 				       btree_path_idx_t path_idx)
2703 {
2704 	struct btree_path *path = trans->paths + path_idx;
2705 
2706 	path->sorted_idx = pos ? trans->paths[pos].sorted_idx + 1 : trans->nr_sorted;
2707 
2708 #ifdef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
2709 	memmove_u64s_up_small(trans->sorted + path->sorted_idx + 1,
2710 			      trans->sorted + path->sorted_idx,
2711 			      DIV_ROUND_UP(trans->nr_sorted - path->sorted_idx,
2712 					   sizeof(u64) / sizeof(btree_path_idx_t)));
2713 	trans->nr_sorted++;
2714 	trans->sorted[path->sorted_idx] = path_idx;
2715 #else
2716 	array_insert_item(trans->sorted, trans->nr_sorted, path->sorted_idx, path_idx);
2717 #endif
2718 
2719 	for (unsigned i = path->sorted_idx; i < trans->nr_sorted; i++)
2720 		trans->paths[trans->sorted[i]].sorted_idx = i;
2721 
2722 	btree_trans_verify_sorted_refs(trans);
2723 }
2724 
2725 void bch2_trans_iter_exit(struct btree_trans *trans, struct btree_iter *iter)
2726 {
2727 	if (iter->update_path)
2728 		bch2_path_put_nokeep(trans, iter->update_path,
2729 			      iter->flags & BTREE_ITER_INTENT);
2730 	if (iter->path)
2731 		bch2_path_put(trans, iter->path,
2732 			      iter->flags & BTREE_ITER_INTENT);
2733 	if (iter->key_cache_path)
2734 		bch2_path_put(trans, iter->key_cache_path,
2735 			      iter->flags & BTREE_ITER_INTENT);
2736 	iter->path		= 0;
2737 	iter->update_path	= 0;
2738 	iter->key_cache_path	= 0;
2739 	iter->trans		= NULL;
2740 }
2741 
2742 void bch2_trans_iter_init_outlined(struct btree_trans *trans,
2743 			  struct btree_iter *iter,
2744 			  enum btree_id btree_id, struct bpos pos,
2745 			  unsigned flags)
2746 {
2747 	bch2_trans_iter_init_common(trans, iter, btree_id, pos, 0, 0,
2748 			       bch2_btree_iter_flags(trans, btree_id, flags),
2749 			       _RET_IP_);
2750 }
2751 
2752 void bch2_trans_node_iter_init(struct btree_trans *trans,
2753 			       struct btree_iter *iter,
2754 			       enum btree_id btree_id,
2755 			       struct bpos pos,
2756 			       unsigned locks_want,
2757 			       unsigned depth,
2758 			       unsigned flags)
2759 {
2760 	flags |= BTREE_ITER_NOT_EXTENTS;
2761 	flags |= __BTREE_ITER_ALL_SNAPSHOTS;
2762 	flags |= BTREE_ITER_ALL_SNAPSHOTS;
2763 
2764 	bch2_trans_iter_init_common(trans, iter, btree_id, pos, locks_want, depth,
2765 			       __bch2_btree_iter_flags(trans, btree_id, flags),
2766 			       _RET_IP_);
2767 
2768 	iter->min_depth	= depth;
2769 
2770 	struct btree_path *path = btree_iter_path(trans, iter);
2771 	BUG_ON(path->locks_want	 < min(locks_want, BTREE_MAX_DEPTH));
2772 	BUG_ON(path->level	!= depth);
2773 	BUG_ON(iter->min_depth	!= depth);
2774 }
2775 
2776 void bch2_trans_copy_iter(struct btree_iter *dst, struct btree_iter *src)
2777 {
2778 	struct btree_trans *trans = src->trans;
2779 
2780 	*dst = *src;
2781 #ifdef TRACK_PATH_ALLOCATED
2782 	dst->ip_allocated = _RET_IP_;
2783 #endif
2784 	if (src->path)
2785 		__btree_path_get(trans->paths + src->path, src->flags & BTREE_ITER_INTENT);
2786 	if (src->update_path)
2787 		__btree_path_get(trans->paths + src->update_path, src->flags & BTREE_ITER_INTENT);
2788 	dst->key_cache_path = 0;
2789 }
2790 
2791 void *__bch2_trans_kmalloc(struct btree_trans *trans, size_t size)
2792 {
2793 	struct bch_fs *c = trans->c;
2794 	unsigned new_top = trans->mem_top + size;
2795 	unsigned old_bytes = trans->mem_bytes;
2796 	unsigned new_bytes = roundup_pow_of_two(new_top);
2797 	int ret;
2798 	void *new_mem;
2799 	void *p;
2800 
2801 	WARN_ON_ONCE(new_bytes > BTREE_TRANS_MEM_MAX);
2802 
2803 	struct btree_transaction_stats *s = btree_trans_stats(trans);
2804 	s->max_mem = max(s->max_mem, new_bytes);
2805 
2806 	if (trans->used_mempool) {
2807 		if (trans->mem_bytes >= new_bytes)
2808 			goto out_change_top;
2809 
2810 		/* No more space from mempool item, need malloc new one */
2811 		new_mem = kmalloc(new_bytes, GFP_NOWAIT|__GFP_NOWARN);
2812 		if (unlikely(!new_mem)) {
2813 			bch2_trans_unlock(trans);
2814 
2815 			new_mem = kmalloc(new_bytes, GFP_KERNEL);
2816 			if (!new_mem)
2817 				return ERR_PTR(-BCH_ERR_ENOMEM_trans_kmalloc);
2818 
2819 			ret = bch2_trans_relock(trans);
2820 			if (ret) {
2821 				kfree(new_mem);
2822 				return ERR_PTR(ret);
2823 			}
2824 		}
2825 		memcpy(new_mem, trans->mem, trans->mem_top);
2826 		trans->used_mempool = false;
2827 		mempool_free(trans->mem, &c->btree_trans_mem_pool);
2828 		goto out_new_mem;
2829 	}
2830 
2831 	new_mem = krealloc(trans->mem, new_bytes, GFP_NOWAIT|__GFP_NOWARN);
2832 	if (unlikely(!new_mem)) {
2833 		bch2_trans_unlock(trans);
2834 
2835 		new_mem = krealloc(trans->mem, new_bytes, GFP_KERNEL);
2836 		if (!new_mem && new_bytes <= BTREE_TRANS_MEM_MAX) {
2837 			new_mem = mempool_alloc(&c->btree_trans_mem_pool, GFP_KERNEL);
2838 			new_bytes = BTREE_TRANS_MEM_MAX;
2839 			memcpy(new_mem, trans->mem, trans->mem_top);
2840 			trans->used_mempool = true;
2841 			kfree(trans->mem);
2842 		}
2843 
2844 		if (!new_mem)
2845 			return ERR_PTR(-BCH_ERR_ENOMEM_trans_kmalloc);
2846 
2847 		trans->mem = new_mem;
2848 		trans->mem_bytes = new_bytes;
2849 
2850 		ret = bch2_trans_relock(trans);
2851 		if (ret)
2852 			return ERR_PTR(ret);
2853 	}
2854 out_new_mem:
2855 	trans->mem = new_mem;
2856 	trans->mem_bytes = new_bytes;
2857 
2858 	if (old_bytes) {
2859 		trace_and_count(c, trans_restart_mem_realloced, trans, _RET_IP_, new_bytes);
2860 		return ERR_PTR(btree_trans_restart(trans, BCH_ERR_transaction_restart_mem_realloced));
2861 	}
2862 out_change_top:
2863 	p = trans->mem + trans->mem_top;
2864 	trans->mem_top += size;
2865 	memset(p, 0, size);
2866 	return p;
2867 }
2868 
2869 static inline void check_srcu_held_too_long(struct btree_trans *trans)
2870 {
2871 	WARN(trans->srcu_held && time_after(jiffies, trans->srcu_lock_time + HZ * 10),
2872 	     "btree trans held srcu lock (delaying memory reclaim) for %lu seconds",
2873 	     (jiffies - trans->srcu_lock_time) / HZ);
2874 }
2875 
2876 void bch2_trans_srcu_unlock(struct btree_trans *trans)
2877 {
2878 	if (trans->srcu_held) {
2879 		struct bch_fs *c = trans->c;
2880 		struct btree_path *path;
2881 		unsigned i;
2882 
2883 		trans_for_each_path(trans, path, i)
2884 			if (path->cached && !btree_node_locked(path, 0))
2885 				path->l[0].b = ERR_PTR(-BCH_ERR_no_btree_node_srcu_reset);
2886 
2887 		check_srcu_held_too_long(trans);
2888 		srcu_read_unlock(&c->btree_trans_barrier, trans->srcu_idx);
2889 		trans->srcu_held = false;
2890 	}
2891 }
2892 
2893 static void bch2_trans_srcu_lock(struct btree_trans *trans)
2894 {
2895 	if (!trans->srcu_held) {
2896 		trans->srcu_idx = srcu_read_lock(&trans->c->btree_trans_barrier);
2897 		trans->srcu_lock_time	= jiffies;
2898 		trans->srcu_held = true;
2899 	}
2900 }
2901 
2902 /**
2903  * bch2_trans_begin() - reset a transaction after a interrupted attempt
2904  * @trans: transaction to reset
2905  *
2906  * Returns:	current restart counter, to be used with trans_was_restarted()
2907  *
2908  * While iterating over nodes or updating nodes a attempt to lock a btree node
2909  * may return BCH_ERR_transaction_restart when the trylock fails. When this
2910  * occurs bch2_trans_begin() should be called and the transaction retried.
2911  */
2912 u32 bch2_trans_begin(struct btree_trans *trans)
2913 {
2914 	struct btree_path *path;
2915 	unsigned i;
2916 	u64 now;
2917 
2918 	bch2_trans_reset_updates(trans);
2919 
2920 	trans->restart_count++;
2921 	trans->mem_top			= 0;
2922 	trans->journal_entries		= NULL;
2923 
2924 	trans_for_each_path(trans, path, i) {
2925 		path->should_be_locked = false;
2926 
2927 		/*
2928 		 * If the transaction wasn't restarted, we're presuming to be
2929 		 * doing something new: dont keep iterators excpt the ones that
2930 		 * are in use - except for the subvolumes btree:
2931 		 */
2932 		if (!trans->restarted && path->btree_id != BTREE_ID_subvolumes)
2933 			path->preserve = false;
2934 
2935 		/*
2936 		 * XXX: we probably shouldn't be doing this if the transaction
2937 		 * was restarted, but currently we still overflow transaction
2938 		 * iterators if we do that
2939 		 */
2940 		if (!path->ref && !path->preserve)
2941 			__bch2_path_free(trans, i);
2942 		else
2943 			path->preserve = false;
2944 	}
2945 
2946 	now = local_clock();
2947 
2948 	if (!IS_ENABLED(CONFIG_BCACHEFS_NO_LATENCY_ACCT) &&
2949 	    time_after64(now, trans->last_begin_time + 10))
2950 		__bch2_time_stats_update(&btree_trans_stats(trans)->duration,
2951 					 trans->last_begin_time, now);
2952 
2953 	if (!trans->restarted &&
2954 	    (need_resched() ||
2955 	     time_after64(now, trans->last_begin_time + BTREE_TRANS_MAX_LOCK_HOLD_TIME_NS))) {
2956 		drop_locks_do(trans, (cond_resched(), 0));
2957 		now = local_clock();
2958 	}
2959 	trans->last_begin_time = now;
2960 
2961 	if (unlikely(trans->srcu_held &&
2962 		     time_after(jiffies, trans->srcu_lock_time + msecs_to_jiffies(10))))
2963 		bch2_trans_srcu_unlock(trans);
2964 
2965 	trans->last_begin_ip = _RET_IP_;
2966 	if (trans->restarted) {
2967 		bch2_btree_path_traverse_all(trans);
2968 		trans->notrace_relock_fail = false;
2969 	}
2970 
2971 	return trans->restart_count;
2972 }
2973 
2974 const char *bch2_btree_transaction_fns[BCH_TRANSACTIONS_NR] = { "(unknown)" };
2975 
2976 unsigned bch2_trans_get_fn_idx(const char *fn)
2977 {
2978 	for (unsigned i = 0; i < ARRAY_SIZE(bch2_btree_transaction_fns); i++)
2979 		if (!bch2_btree_transaction_fns[i] ||
2980 		    bch2_btree_transaction_fns[i] == fn) {
2981 			bch2_btree_transaction_fns[i] = fn;
2982 			return i;
2983 		}
2984 
2985 	pr_warn_once("BCH_TRANSACTIONS_NR not big enough!");
2986 	return 0;
2987 }
2988 
2989 struct btree_trans *__bch2_trans_get(struct bch_fs *c, unsigned fn_idx)
2990 	__acquires(&c->btree_trans_barrier)
2991 {
2992 	struct btree_trans *trans;
2993 
2994 	if (IS_ENABLED(__KERNEL__)) {
2995 		trans = this_cpu_xchg(c->btree_trans_bufs->trans, NULL);
2996 		if (trans) {
2997 			memset(trans, 0, offsetof(struct btree_trans, list));
2998 			goto got_trans;
2999 		}
3000 	}
3001 
3002 	trans = mempool_alloc(&c->btree_trans_pool, GFP_NOFS);
3003 	memset(trans, 0, sizeof(*trans));
3004 	closure_init_stack(&trans->ref);
3005 
3006 	seqmutex_lock(&c->btree_trans_lock);
3007 	if (IS_ENABLED(CONFIG_BCACHEFS_DEBUG)) {
3008 		struct btree_trans *pos;
3009 		pid_t pid = current->pid;
3010 
3011 		trans->locking_wait.task = current;
3012 
3013 		list_for_each_entry(pos, &c->btree_trans_list, list) {
3014 			struct task_struct *pos_task = READ_ONCE(pos->locking_wait.task);
3015 			/*
3016 			 * We'd much prefer to be stricter here and completely
3017 			 * disallow multiple btree_trans in the same thread -
3018 			 * but the data move path calls bch2_write when we
3019 			 * already have a btree_trans initialized.
3020 			 */
3021 			BUG_ON(pos_task &&
3022 			       pid == pos_task->pid &&
3023 			       bch2_trans_locked(pos));
3024 
3025 			if (pos_task && pid < pos_task->pid) {
3026 				list_add_tail(&trans->list, &pos->list);
3027 				goto list_add_done;
3028 			}
3029 		}
3030 	}
3031 	list_add_tail(&trans->list, &c->btree_trans_list);
3032 list_add_done:
3033 	seqmutex_unlock(&c->btree_trans_lock);
3034 got_trans:
3035 	trans->c		= c;
3036 	trans->last_begin_time	= local_clock();
3037 	trans->fn_idx		= fn_idx;
3038 	trans->locking_wait.task = current;
3039 	trans->journal_replay_not_finished =
3040 		unlikely(!test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags)) &&
3041 		atomic_inc_not_zero(&c->journal_keys.ref);
3042 	trans->nr_paths		= ARRAY_SIZE(trans->_paths);
3043 	trans->paths_allocated	= trans->_paths_allocated;
3044 	trans->sorted		= trans->_sorted;
3045 	trans->paths		= trans->_paths;
3046 	trans->updates		= trans->_updates;
3047 
3048 	*trans_paths_nr(trans->paths) = BTREE_ITER_INITIAL;
3049 
3050 	trans->paths_allocated[0] = 1;
3051 
3052 	if (fn_idx < BCH_TRANSACTIONS_NR) {
3053 		trans->fn = bch2_btree_transaction_fns[fn_idx];
3054 
3055 		struct btree_transaction_stats *s = &c->btree_transaction_stats[fn_idx];
3056 
3057 		if (s->max_mem) {
3058 			unsigned expected_mem_bytes = roundup_pow_of_two(s->max_mem);
3059 
3060 			trans->mem = kmalloc(expected_mem_bytes, GFP_KERNEL);
3061 			if (likely(trans->mem))
3062 				trans->mem_bytes = expected_mem_bytes;
3063 		}
3064 
3065 		trans->nr_paths_max = s->nr_max_paths;
3066 		trans->journal_entries_size = s->journal_entries_size;
3067 	}
3068 
3069 	trans->srcu_idx		= srcu_read_lock(&c->btree_trans_barrier);
3070 	trans->srcu_lock_time	= jiffies;
3071 	trans->srcu_held	= true;
3072 	return trans;
3073 }
3074 
3075 static void check_btree_paths_leaked(struct btree_trans *trans)
3076 {
3077 #ifdef CONFIG_BCACHEFS_DEBUG
3078 	struct bch_fs *c = trans->c;
3079 	struct btree_path *path;
3080 	unsigned i;
3081 
3082 	trans_for_each_path(trans, path, i)
3083 		if (path->ref)
3084 			goto leaked;
3085 	return;
3086 leaked:
3087 	bch_err(c, "btree paths leaked from %s!", trans->fn);
3088 	trans_for_each_path(trans, path, i)
3089 		if (path->ref)
3090 			printk(KERN_ERR "  btree %s %pS\n",
3091 			       bch2_btree_id_str(path->btree_id),
3092 			       (void *) path->ip_allocated);
3093 	/* Be noisy about this: */
3094 	bch2_fatal_error(c);
3095 #endif
3096 }
3097 
3098 void bch2_trans_put(struct btree_trans *trans)
3099 	__releases(&c->btree_trans_barrier)
3100 {
3101 	struct bch_fs *c = trans->c;
3102 
3103 	bch2_trans_unlock(trans);
3104 
3105 	trans_for_each_update(trans, i)
3106 		__btree_path_put(trans->paths + i->path, true);
3107 	trans->nr_updates	= 0;
3108 	trans->locking_wait.task = NULL;
3109 
3110 	check_btree_paths_leaked(trans);
3111 
3112 	if (trans->srcu_held) {
3113 		check_srcu_held_too_long(trans);
3114 		srcu_read_unlock(&c->btree_trans_barrier, trans->srcu_idx);
3115 	}
3116 
3117 	if (trans->fs_usage_deltas) {
3118 		if (trans->fs_usage_deltas->size + sizeof(trans->fs_usage_deltas) ==
3119 		    REPLICAS_DELTA_LIST_MAX)
3120 			mempool_free(trans->fs_usage_deltas,
3121 				     &c->replicas_delta_pool);
3122 		else
3123 			kfree(trans->fs_usage_deltas);
3124 	}
3125 
3126 	if (unlikely(trans->journal_replay_not_finished))
3127 		bch2_journal_keys_put(c);
3128 
3129 	unsigned long *paths_allocated = trans->paths_allocated;
3130 	trans->paths_allocated	= NULL;
3131 	trans->paths		= NULL;
3132 
3133 	if (paths_allocated != trans->_paths_allocated)
3134 		kvfree_rcu_mightsleep(paths_allocated);
3135 
3136 	if (trans->used_mempool)
3137 		mempool_free(trans->mem, &c->btree_trans_mem_pool);
3138 	else
3139 		kfree(trans->mem);
3140 
3141 	/* Userspace doesn't have a real percpu implementation: */
3142 	if (IS_ENABLED(__KERNEL__))
3143 		trans = this_cpu_xchg(c->btree_trans_bufs->trans, trans);
3144 
3145 	if (trans) {
3146 		closure_sync(&trans->ref);
3147 
3148 		seqmutex_lock(&c->btree_trans_lock);
3149 		list_del(&trans->list);
3150 		seqmutex_unlock(&c->btree_trans_lock);
3151 
3152 		mempool_free(trans, &c->btree_trans_pool);
3153 	}
3154 }
3155 
3156 static void __maybe_unused
3157 bch2_btree_bkey_cached_common_to_text(struct printbuf *out,
3158 				      struct btree_bkey_cached_common *b)
3159 {
3160 	struct six_lock_count c = six_lock_counts(&b->lock);
3161 	struct task_struct *owner;
3162 	pid_t pid;
3163 
3164 	rcu_read_lock();
3165 	owner = READ_ONCE(b->lock.owner);
3166 	pid = owner ? owner->pid : 0;
3167 	rcu_read_unlock();
3168 
3169 	prt_tab(out);
3170 	prt_printf(out, "%px %c l=%u %s:", b, b->cached ? 'c' : 'b',
3171 		   b->level, bch2_btree_id_str(b->btree_id));
3172 	bch2_bpos_to_text(out, btree_node_pos(b));
3173 
3174 	prt_tab(out);
3175 	prt_printf(out, " locks %u:%u:%u held by pid %u",
3176 		   c.n[0], c.n[1], c.n[2], pid);
3177 }
3178 
3179 void bch2_btree_trans_to_text(struct printbuf *out, struct btree_trans *trans)
3180 {
3181 	struct btree_bkey_cached_common *b;
3182 	static char lock_types[] = { 'r', 'i', 'w' };
3183 	struct task_struct *task = READ_ONCE(trans->locking_wait.task);
3184 	unsigned l, idx;
3185 
3186 	/* before rcu_read_lock(): */
3187 	bch2_printbuf_make_room(out, 4096);
3188 
3189 	if (!out->nr_tabstops) {
3190 		printbuf_tabstop_push(out, 16);
3191 		printbuf_tabstop_push(out, 32);
3192 	}
3193 
3194 	prt_printf(out, "%i %s\n", task ? task->pid : 0, trans->fn);
3195 
3196 	/* trans->paths is rcu protected vs. freeing */
3197 	rcu_read_lock();
3198 	out->atomic++;
3199 
3200 	struct btree_path *paths = rcu_dereference(trans->paths);
3201 	if (!paths)
3202 		goto out;
3203 
3204 	unsigned long *paths_allocated = trans_paths_allocated(paths);
3205 
3206 	trans_for_each_path_idx_from(paths_allocated, *trans_paths_nr(paths), idx, 1) {
3207 		struct btree_path *path = paths + idx;
3208 		if (!path->nodes_locked)
3209 			continue;
3210 
3211 		prt_printf(out, "  path %u %c l=%u %s:",
3212 		       idx,
3213 		       path->cached ? 'c' : 'b',
3214 		       path->level,
3215 		       bch2_btree_id_str(path->btree_id));
3216 		bch2_bpos_to_text(out, path->pos);
3217 		prt_newline(out);
3218 
3219 		for (l = 0; l < BTREE_MAX_DEPTH; l++) {
3220 			if (btree_node_locked(path, l) &&
3221 			    !IS_ERR_OR_NULL(b = (void *) READ_ONCE(path->l[l].b))) {
3222 				prt_printf(out, "    %c l=%u ",
3223 					   lock_types[btree_node_locked_type(path, l)], l);
3224 				bch2_btree_bkey_cached_common_to_text(out, b);
3225 				prt_newline(out);
3226 			}
3227 		}
3228 	}
3229 
3230 	b = READ_ONCE(trans->locking);
3231 	if (b) {
3232 		prt_printf(out, "  blocked for %lluus on",
3233 			   div_u64(local_clock() - trans->locking_wait.start_time,
3234 				   1000));
3235 		prt_newline(out);
3236 		prt_printf(out, "    %c", lock_types[trans->locking_wait.lock_want]);
3237 		bch2_btree_bkey_cached_common_to_text(out, b);
3238 		prt_newline(out);
3239 	}
3240 out:
3241 	--out->atomic;
3242 	rcu_read_unlock();
3243 }
3244 
3245 void bch2_fs_btree_iter_exit(struct bch_fs *c)
3246 {
3247 	struct btree_transaction_stats *s;
3248 	struct btree_trans *trans;
3249 	int cpu;
3250 
3251 	if (c->btree_trans_bufs)
3252 		for_each_possible_cpu(cpu) {
3253 			struct btree_trans *trans =
3254 				per_cpu_ptr(c->btree_trans_bufs, cpu)->trans;
3255 
3256 			if (trans) {
3257 				closure_sync(&trans->ref);
3258 
3259 				seqmutex_lock(&c->btree_trans_lock);
3260 				list_del(&trans->list);
3261 				seqmutex_unlock(&c->btree_trans_lock);
3262 			}
3263 			kfree(trans);
3264 		}
3265 	free_percpu(c->btree_trans_bufs);
3266 
3267 	trans = list_first_entry_or_null(&c->btree_trans_list, struct btree_trans, list);
3268 	if (trans)
3269 		panic("%s leaked btree_trans\n", trans->fn);
3270 
3271 	for (s = c->btree_transaction_stats;
3272 	     s < c->btree_transaction_stats + ARRAY_SIZE(c->btree_transaction_stats);
3273 	     s++) {
3274 		kfree(s->max_paths_text);
3275 		bch2_time_stats_exit(&s->lock_hold_times);
3276 	}
3277 
3278 	if (c->btree_trans_barrier_initialized)
3279 		cleanup_srcu_struct(&c->btree_trans_barrier);
3280 	mempool_exit(&c->btree_trans_mem_pool);
3281 	mempool_exit(&c->btree_trans_pool);
3282 }
3283 
3284 void bch2_fs_btree_iter_init_early(struct bch_fs *c)
3285 {
3286 	struct btree_transaction_stats *s;
3287 
3288 	for (s = c->btree_transaction_stats;
3289 	     s < c->btree_transaction_stats + ARRAY_SIZE(c->btree_transaction_stats);
3290 	     s++) {
3291 		bch2_time_stats_init(&s->duration);
3292 		bch2_time_stats_init(&s->lock_hold_times);
3293 		mutex_init(&s->lock);
3294 	}
3295 
3296 	INIT_LIST_HEAD(&c->btree_trans_list);
3297 	seqmutex_init(&c->btree_trans_lock);
3298 }
3299 
3300 int bch2_fs_btree_iter_init(struct bch_fs *c)
3301 {
3302 	int ret;
3303 
3304 	c->btree_trans_bufs = alloc_percpu(struct btree_trans_buf);
3305 	if (!c->btree_trans_bufs)
3306 		return -ENOMEM;
3307 
3308 	ret   = mempool_init_kmalloc_pool(&c->btree_trans_pool, 1,
3309 					  sizeof(struct btree_trans)) ?:
3310 		mempool_init_kmalloc_pool(&c->btree_trans_mem_pool, 1,
3311 					  BTREE_TRANS_MEM_MAX) ?:
3312 		init_srcu_struct(&c->btree_trans_barrier);
3313 	if (!ret)
3314 		c->btree_trans_barrier_initialized = true;
3315 	return ret;
3316 }
3317