xref: /linux/fs/bcachefs/btree_iter.h (revision ae22a94997b8a03dcb3c922857c203246711f9d4)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _BCACHEFS_BTREE_ITER_H
3 #define _BCACHEFS_BTREE_ITER_H
4 
5 #include "bset.h"
6 #include "btree_types.h"
7 #include "trace.h"
8 
9 static inline int __bkey_err(const struct bkey *k)
10 {
11 	return PTR_ERR_OR_ZERO(k);
12 }
13 
14 #define bkey_err(_k)	__bkey_err((_k).k)
15 
16 static inline void __btree_path_get(struct btree_path *path, bool intent)
17 {
18 	path->ref++;
19 	path->intent_ref += intent;
20 }
21 
22 static inline bool __btree_path_put(struct btree_path *path, bool intent)
23 {
24 	EBUG_ON(!path->ref);
25 	EBUG_ON(!path->intent_ref && intent);
26 	path->intent_ref -= intent;
27 	return --path->ref == 0;
28 }
29 
30 static inline void btree_path_set_dirty(struct btree_path *path,
31 					enum btree_path_uptodate u)
32 {
33 	path->uptodate = max_t(unsigned, path->uptodate, u);
34 }
35 
36 static inline struct btree *btree_path_node(struct btree_path *path,
37 					    unsigned level)
38 {
39 	return level < BTREE_MAX_DEPTH ? path->l[level].b : NULL;
40 }
41 
42 static inline bool btree_node_lock_seq_matches(const struct btree_path *path,
43 					const struct btree *b, unsigned level)
44 {
45 	return path->l[level].lock_seq == six_lock_seq(&b->c.lock);
46 }
47 
48 static inline struct btree *btree_node_parent(struct btree_path *path,
49 					      struct btree *b)
50 {
51 	return btree_path_node(path, b->c.level + 1);
52 }
53 
54 /* Iterate over paths within a transaction: */
55 
56 void __bch2_btree_trans_sort_paths(struct btree_trans *);
57 
58 static inline void btree_trans_sort_paths(struct btree_trans *trans)
59 {
60 	if (!IS_ENABLED(CONFIG_BCACHEFS_DEBUG) &&
61 	    trans->paths_sorted)
62 		return;
63 	__bch2_btree_trans_sort_paths(trans);
64 }
65 
66 static inline unsigned long *trans_paths_nr(struct btree_path *paths)
67 {
68 	return &container_of(paths, struct btree_trans_paths, paths[0])->nr_paths;
69 }
70 
71 static inline unsigned long *trans_paths_allocated(struct btree_path *paths)
72 {
73 	unsigned long *v = trans_paths_nr(paths);
74 	return v - BITS_TO_LONGS(*v);
75 }
76 
77 #define trans_for_each_path_idx_from(_paths_allocated, _nr, _idx, _start)\
78 	for (_idx = _start;						\
79 	     (_idx = find_next_bit(_paths_allocated, _nr, _idx)) < _nr;	\
80 	     _idx++)
81 
82 static inline struct btree_path *
83 __trans_next_path(struct btree_trans *trans, unsigned *idx)
84 {
85 	unsigned long *w = trans->paths_allocated + *idx / BITS_PER_LONG;
86 	/*
87 	 * Open coded find_next_bit(), because
88 	 *  - this is fast path, we can't afford the function call
89 	 *  - and we know that nr_paths is a multiple of BITS_PER_LONG,
90 	 */
91 	while (*idx < trans->nr_paths) {
92 		unsigned long v = *w >> (*idx & (BITS_PER_LONG - 1));
93 		if (v) {
94 			*idx += __ffs(v);
95 			return trans->paths + *idx;
96 		}
97 
98 		*idx += BITS_PER_LONG;
99 		*idx &= ~(BITS_PER_LONG - 1);
100 		w++;
101 	}
102 
103 	return NULL;
104 }
105 
106 /*
107  * This version is intended to be safe for use on a btree_trans that is owned by
108  * another thread, for bch2_btree_trans_to_text();
109  */
110 #define trans_for_each_path_from(_trans, _path, _idx, _start)		\
111 	for (_idx = _start;						\
112 	     (_path = __trans_next_path((_trans), &_idx));		\
113 	     _idx++)
114 
115 #define trans_for_each_path(_trans, _path, _idx)			\
116 	trans_for_each_path_from(_trans, _path, _idx, 1)
117 
118 static inline struct btree_path *next_btree_path(struct btree_trans *trans, struct btree_path *path)
119 {
120 	unsigned idx = path ? path->sorted_idx + 1 : 0;
121 
122 	EBUG_ON(idx > trans->nr_sorted);
123 
124 	return idx < trans->nr_sorted
125 		? trans->paths + trans->sorted[idx]
126 		: NULL;
127 }
128 
129 static inline struct btree_path *prev_btree_path(struct btree_trans *trans, struct btree_path *path)
130 {
131 	unsigned idx = path ? path->sorted_idx : trans->nr_sorted;
132 
133 	return idx
134 		? trans->paths + trans->sorted[idx - 1]
135 		: NULL;
136 }
137 
138 #define trans_for_each_path_idx_inorder(_trans, _iter)			\
139 	for (_iter = (struct trans_for_each_path_inorder_iter) { 0 };	\
140 	     (_iter.path_idx = trans->sorted[_iter.sorted_idx],		\
141 	      _iter.sorted_idx < (_trans)->nr_sorted);			\
142 	     _iter.sorted_idx++)
143 
144 struct trans_for_each_path_inorder_iter {
145 	btree_path_idx_t	sorted_idx;
146 	btree_path_idx_t	path_idx;
147 };
148 
149 #define trans_for_each_path_inorder(_trans, _path, _iter)		\
150 	for (_iter = (struct trans_for_each_path_inorder_iter) { 0 };	\
151 	     (_iter.path_idx = trans->sorted[_iter.sorted_idx],		\
152 	      _path = (_trans)->paths + _iter.path_idx,			\
153 	      _iter.sorted_idx < (_trans)->nr_sorted);			\
154 	     _iter.sorted_idx++)
155 
156 #define trans_for_each_path_inorder_reverse(_trans, _path, _i)		\
157 	for (_i = trans->nr_sorted - 1;					\
158 	     ((_path) = (_trans)->paths + trans->sorted[_i]), (_i) >= 0;\
159 	     --_i)
160 
161 static inline bool __path_has_node(const struct btree_path *path,
162 				   const struct btree *b)
163 {
164 	return path->l[b->c.level].b == b &&
165 		btree_node_lock_seq_matches(path, b, b->c.level);
166 }
167 
168 static inline struct btree_path *
169 __trans_next_path_with_node(struct btree_trans *trans, struct btree *b,
170 			    unsigned *idx)
171 {
172 	struct btree_path *path;
173 
174 	while ((path = __trans_next_path(trans, idx)) &&
175 		!__path_has_node(path, b))
176 	       (*idx)++;
177 
178 	return path;
179 }
180 
181 #define trans_for_each_path_with_node(_trans, _b, _path, _iter)		\
182 	for (_iter = 1;							\
183 	     (_path = __trans_next_path_with_node((_trans), (_b), &_iter));\
184 	     _iter++)
185 
186 btree_path_idx_t __bch2_btree_path_make_mut(struct btree_trans *, btree_path_idx_t,
187 					    bool, unsigned long);
188 
189 static inline btree_path_idx_t __must_check
190 bch2_btree_path_make_mut(struct btree_trans *trans,
191 			 btree_path_idx_t path, bool intent,
192 			 unsigned long ip)
193 {
194 	if (trans->paths[path].ref > 1 ||
195 	    trans->paths[path].preserve)
196 		path = __bch2_btree_path_make_mut(trans, path, intent, ip);
197 	trans->paths[path].should_be_locked = false;
198 	return path;
199 }
200 
201 btree_path_idx_t __must_check
202 __bch2_btree_path_set_pos(struct btree_trans *, btree_path_idx_t,
203 			  struct bpos, bool, unsigned long);
204 
205 static inline btree_path_idx_t __must_check
206 bch2_btree_path_set_pos(struct btree_trans *trans,
207 			btree_path_idx_t path, struct bpos new_pos,
208 			bool intent, unsigned long ip)
209 {
210 	return !bpos_eq(new_pos, trans->paths[path].pos)
211 		? __bch2_btree_path_set_pos(trans, path, new_pos, intent, ip)
212 		: path;
213 }
214 
215 int __must_check bch2_btree_path_traverse_one(struct btree_trans *,
216 					      btree_path_idx_t,
217 					      unsigned, unsigned long);
218 
219 static inline int __must_check bch2_btree_path_traverse(struct btree_trans *trans,
220 					  btree_path_idx_t path, unsigned flags)
221 {
222 	if (trans->paths[path].uptodate < BTREE_ITER_NEED_RELOCK)
223 		return 0;
224 
225 	return bch2_btree_path_traverse_one(trans, path, flags, _RET_IP_);
226 }
227 
228 btree_path_idx_t bch2_path_get(struct btree_trans *, enum btree_id, struct bpos,
229 				 unsigned, unsigned, unsigned, unsigned long);
230 struct bkey_s_c bch2_btree_path_peek_slot(struct btree_path *, struct bkey *);
231 
232 /*
233  * bch2_btree_path_peek_slot() for a cached iterator might return a key in a
234  * different snapshot:
235  */
236 static inline struct bkey_s_c bch2_btree_path_peek_slot_exact(struct btree_path *path, struct bkey *u)
237 {
238 	struct bkey_s_c k = bch2_btree_path_peek_slot(path, u);
239 
240 	if (k.k && bpos_eq(path->pos, k.k->p))
241 		return k;
242 
243 	bkey_init(u);
244 	u->p = path->pos;
245 	return (struct bkey_s_c) { u, NULL };
246 }
247 
248 struct bkey_i *bch2_btree_journal_peek_slot(struct btree_trans *,
249 					struct btree_iter *, struct bpos);
250 
251 void bch2_btree_path_level_init(struct btree_trans *, struct btree_path *, struct btree *);
252 
253 int __bch2_trans_mutex_lock(struct btree_trans *, struct mutex *);
254 
255 static inline int bch2_trans_mutex_lock(struct btree_trans *trans, struct mutex *lock)
256 {
257 	return mutex_trylock(lock)
258 		? 0
259 		: __bch2_trans_mutex_lock(trans, lock);
260 }
261 
262 #ifdef CONFIG_BCACHEFS_DEBUG
263 void bch2_trans_verify_paths(struct btree_trans *);
264 void bch2_assert_pos_locked(struct btree_trans *, enum btree_id,
265 			    struct bpos, bool);
266 #else
267 static inline void bch2_trans_verify_paths(struct btree_trans *trans) {}
268 static inline void bch2_assert_pos_locked(struct btree_trans *trans, enum btree_id id,
269 					  struct bpos pos, bool key_cache) {}
270 #endif
271 
272 void bch2_btree_path_fix_key_modified(struct btree_trans *trans,
273 				      struct btree *, struct bkey_packed *);
274 void bch2_btree_node_iter_fix(struct btree_trans *trans, struct btree_path *,
275 			      struct btree *, struct btree_node_iter *,
276 			      struct bkey_packed *, unsigned, unsigned);
277 
278 int bch2_btree_path_relock_intent(struct btree_trans *, struct btree_path *);
279 
280 void bch2_path_put(struct btree_trans *, btree_path_idx_t, bool);
281 
282 int bch2_trans_relock(struct btree_trans *);
283 int bch2_trans_relock_notrace(struct btree_trans *);
284 void bch2_trans_unlock(struct btree_trans *);
285 void bch2_trans_unlock_long(struct btree_trans *);
286 bool bch2_trans_locked(struct btree_trans *);
287 
288 static inline int trans_was_restarted(struct btree_trans *trans, u32 restart_count)
289 {
290 	return restart_count != trans->restart_count
291 		? -BCH_ERR_transaction_restart_nested
292 		: 0;
293 }
294 
295 void __noreturn bch2_trans_restart_error(struct btree_trans *, u32);
296 
297 static inline void bch2_trans_verify_not_restarted(struct btree_trans *trans,
298 						   u32 restart_count)
299 {
300 	if (trans_was_restarted(trans, restart_count))
301 		bch2_trans_restart_error(trans, restart_count);
302 }
303 
304 void __noreturn bch2_trans_in_restart_error(struct btree_trans *);
305 
306 static inline void bch2_trans_verify_not_in_restart(struct btree_trans *trans)
307 {
308 	if (trans->restarted)
309 		bch2_trans_in_restart_error(trans);
310 }
311 
312 __always_inline
313 static int btree_trans_restart_nounlock(struct btree_trans *trans, int err)
314 {
315 	BUG_ON(err <= 0);
316 	BUG_ON(!bch2_err_matches(-err, BCH_ERR_transaction_restart));
317 
318 	trans->restarted = err;
319 	trans->last_restarted_ip = _THIS_IP_;
320 	return -err;
321 }
322 
323 __always_inline
324 static int btree_trans_restart(struct btree_trans *trans, int err)
325 {
326 	btree_trans_restart_nounlock(trans, err);
327 	return -err;
328 }
329 
330 bool bch2_btree_node_upgrade(struct btree_trans *,
331 			     struct btree_path *, unsigned);
332 
333 void __bch2_btree_path_downgrade(struct btree_trans *, struct btree_path *, unsigned);
334 
335 static inline void bch2_btree_path_downgrade(struct btree_trans *trans,
336 					     struct btree_path *path)
337 {
338 	unsigned new_locks_want = path->level + !!path->intent_ref;
339 
340 	if (path->locks_want > new_locks_want)
341 		__bch2_btree_path_downgrade(trans, path, new_locks_want);
342 }
343 
344 void bch2_trans_downgrade(struct btree_trans *);
345 
346 void bch2_trans_node_add(struct btree_trans *trans, struct btree_path *, struct btree *);
347 void bch2_trans_node_reinit_iter(struct btree_trans *, struct btree *);
348 
349 int __must_check __bch2_btree_iter_traverse(struct btree_iter *iter);
350 int __must_check bch2_btree_iter_traverse(struct btree_iter *);
351 
352 struct btree *bch2_btree_iter_peek_node(struct btree_iter *);
353 struct btree *bch2_btree_iter_peek_node_and_restart(struct btree_iter *);
354 struct btree *bch2_btree_iter_next_node(struct btree_iter *);
355 
356 struct bkey_s_c bch2_btree_iter_peek_upto(struct btree_iter *, struct bpos);
357 struct bkey_s_c bch2_btree_iter_next(struct btree_iter *);
358 
359 static inline struct bkey_s_c bch2_btree_iter_peek(struct btree_iter *iter)
360 {
361 	return bch2_btree_iter_peek_upto(iter, SPOS_MAX);
362 }
363 
364 struct bkey_s_c bch2_btree_iter_peek_prev(struct btree_iter *);
365 struct bkey_s_c bch2_btree_iter_prev(struct btree_iter *);
366 
367 struct bkey_s_c bch2_btree_iter_peek_slot(struct btree_iter *);
368 struct bkey_s_c bch2_btree_iter_next_slot(struct btree_iter *);
369 struct bkey_s_c bch2_btree_iter_prev_slot(struct btree_iter *);
370 
371 bool bch2_btree_iter_advance(struct btree_iter *);
372 bool bch2_btree_iter_rewind(struct btree_iter *);
373 
374 static inline void __bch2_btree_iter_set_pos(struct btree_iter *iter, struct bpos new_pos)
375 {
376 	iter->k.type = KEY_TYPE_deleted;
377 	iter->k.p.inode		= iter->pos.inode	= new_pos.inode;
378 	iter->k.p.offset	= iter->pos.offset	= new_pos.offset;
379 	iter->k.p.snapshot	= iter->pos.snapshot	= new_pos.snapshot;
380 	iter->k.size = 0;
381 }
382 
383 static inline void bch2_btree_iter_set_pos(struct btree_iter *iter, struct bpos new_pos)
384 {
385 	struct btree_trans *trans = iter->trans;
386 
387 	if (unlikely(iter->update_path))
388 		bch2_path_put(trans, iter->update_path,
389 			      iter->flags & BTREE_ITER_INTENT);
390 	iter->update_path = 0;
391 
392 	if (!(iter->flags & BTREE_ITER_ALL_SNAPSHOTS))
393 		new_pos.snapshot = iter->snapshot;
394 
395 	__bch2_btree_iter_set_pos(iter, new_pos);
396 }
397 
398 static inline void bch2_btree_iter_set_pos_to_extent_start(struct btree_iter *iter)
399 {
400 	BUG_ON(!(iter->flags & BTREE_ITER_IS_EXTENTS));
401 	iter->pos = bkey_start_pos(&iter->k);
402 }
403 
404 static inline void bch2_btree_iter_set_snapshot(struct btree_iter *iter, u32 snapshot)
405 {
406 	struct bpos pos = iter->pos;
407 
408 	iter->snapshot = snapshot;
409 	pos.snapshot = snapshot;
410 	bch2_btree_iter_set_pos(iter, pos);
411 }
412 
413 void bch2_trans_iter_exit(struct btree_trans *, struct btree_iter *);
414 
415 static inline unsigned __bch2_btree_iter_flags(struct btree_trans *trans,
416 					       unsigned btree_id,
417 					       unsigned flags)
418 {
419 	if (!(flags & (BTREE_ITER_ALL_SNAPSHOTS|BTREE_ITER_NOT_EXTENTS)) &&
420 	    btree_id_is_extents(btree_id))
421 		flags |= BTREE_ITER_IS_EXTENTS;
422 
423 	if (!(flags & __BTREE_ITER_ALL_SNAPSHOTS) &&
424 	    !btree_type_has_snapshot_field(btree_id))
425 		flags &= ~BTREE_ITER_ALL_SNAPSHOTS;
426 
427 	if (!(flags & BTREE_ITER_ALL_SNAPSHOTS) &&
428 	    btree_type_has_snapshots(btree_id))
429 		flags |= BTREE_ITER_FILTER_SNAPSHOTS;
430 
431 	if (trans->journal_replay_not_finished)
432 		flags |= BTREE_ITER_WITH_JOURNAL;
433 
434 	return flags;
435 }
436 
437 static inline unsigned bch2_btree_iter_flags(struct btree_trans *trans,
438 					     unsigned btree_id,
439 					     unsigned flags)
440 {
441 	if (!btree_id_cached(trans->c, btree_id)) {
442 		flags &= ~BTREE_ITER_CACHED;
443 		flags &= ~BTREE_ITER_WITH_KEY_CACHE;
444 	} else if (!(flags & BTREE_ITER_CACHED))
445 		flags |= BTREE_ITER_WITH_KEY_CACHE;
446 
447 	return __bch2_btree_iter_flags(trans, btree_id, flags);
448 }
449 
450 static inline void bch2_trans_iter_init_common(struct btree_trans *trans,
451 					  struct btree_iter *iter,
452 					  unsigned btree_id, struct bpos pos,
453 					  unsigned locks_want,
454 					  unsigned depth,
455 					  unsigned flags,
456 					  unsigned long ip)
457 {
458 	iter->trans		= trans;
459 	iter->update_path	= 0;
460 	iter->key_cache_path	= 0;
461 	iter->btree_id		= btree_id;
462 	iter->min_depth		= 0;
463 	iter->flags		= flags;
464 	iter->snapshot		= pos.snapshot;
465 	iter->pos		= pos;
466 	iter->k			= POS_KEY(pos);
467 	iter->journal_idx	= 0;
468 #ifdef CONFIG_BCACHEFS_DEBUG
469 	iter->ip_allocated = ip;
470 #endif
471 	iter->path = bch2_path_get(trans, btree_id, iter->pos,
472 				   locks_want, depth, flags, ip);
473 }
474 
475 void bch2_trans_iter_init_outlined(struct btree_trans *, struct btree_iter *,
476 			  enum btree_id, struct bpos, unsigned);
477 
478 static inline void bch2_trans_iter_init(struct btree_trans *trans,
479 			  struct btree_iter *iter,
480 			  unsigned btree_id, struct bpos pos,
481 			  unsigned flags)
482 {
483 	if (__builtin_constant_p(btree_id) &&
484 	    __builtin_constant_p(flags))
485 		bch2_trans_iter_init_common(trans, iter, btree_id, pos, 0, 0,
486 				bch2_btree_iter_flags(trans, btree_id, flags),
487 				_THIS_IP_);
488 	else
489 		bch2_trans_iter_init_outlined(trans, iter, btree_id, pos, flags);
490 }
491 
492 void bch2_trans_node_iter_init(struct btree_trans *, struct btree_iter *,
493 			       enum btree_id, struct bpos,
494 			       unsigned, unsigned, unsigned);
495 void bch2_trans_copy_iter(struct btree_iter *, struct btree_iter *);
496 
497 static inline void set_btree_iter_dontneed(struct btree_iter *iter)
498 {
499 	struct btree_trans *trans = iter->trans;
500 
501 	if (!trans->restarted)
502 		btree_iter_path(trans, iter)->preserve = false;
503 }
504 
505 void *__bch2_trans_kmalloc(struct btree_trans *, size_t);
506 
507 static inline void *bch2_trans_kmalloc(struct btree_trans *trans, size_t size)
508 {
509 	size = roundup(size, 8);
510 
511 	if (likely(trans->mem_top + size <= trans->mem_bytes)) {
512 		void *p = trans->mem + trans->mem_top;
513 
514 		trans->mem_top += size;
515 		memset(p, 0, size);
516 		return p;
517 	} else {
518 		return __bch2_trans_kmalloc(trans, size);
519 	}
520 }
521 
522 static inline void *bch2_trans_kmalloc_nomemzero(struct btree_trans *trans, size_t size)
523 {
524 	size = round_up(size, 8);
525 
526 	if (likely(trans->mem_top + size <= trans->mem_bytes)) {
527 		void *p = trans->mem + trans->mem_top;
528 
529 		trans->mem_top += size;
530 		return p;
531 	} else {
532 		return __bch2_trans_kmalloc(trans, size);
533 	}
534 }
535 
536 static inline struct bkey_s_c __bch2_bkey_get_iter(struct btree_trans *trans,
537 				struct btree_iter *iter,
538 				unsigned btree_id, struct bpos pos,
539 				unsigned flags, unsigned type)
540 {
541 	struct bkey_s_c k;
542 
543 	bch2_trans_iter_init(trans, iter, btree_id, pos, flags);
544 	k = bch2_btree_iter_peek_slot(iter);
545 
546 	if (!bkey_err(k) && type && k.k->type != type)
547 		k = bkey_s_c_err(-BCH_ERR_ENOENT_bkey_type_mismatch);
548 	if (unlikely(bkey_err(k)))
549 		bch2_trans_iter_exit(trans, iter);
550 	return k;
551 }
552 
553 static inline struct bkey_s_c bch2_bkey_get_iter(struct btree_trans *trans,
554 				struct btree_iter *iter,
555 				unsigned btree_id, struct bpos pos,
556 				unsigned flags)
557 {
558 	return __bch2_bkey_get_iter(trans, iter, btree_id, pos, flags, 0);
559 }
560 
561 #define bch2_bkey_get_iter_typed(_trans, _iter, _btree_id, _pos, _flags, _type)\
562 	bkey_s_c_to_##_type(__bch2_bkey_get_iter(_trans, _iter,			\
563 				       _btree_id, _pos, _flags, KEY_TYPE_##_type))
564 
565 static inline int __bch2_bkey_get_val_typed(struct btree_trans *trans,
566 				unsigned btree_id, struct bpos pos,
567 				unsigned flags, unsigned type,
568 				unsigned val_size, void *val)
569 {
570 	struct btree_iter iter;
571 	struct bkey_s_c k;
572 	int ret;
573 
574 	k = __bch2_bkey_get_iter(trans, &iter, btree_id, pos, flags, type);
575 	ret = bkey_err(k);
576 	if (!ret) {
577 		unsigned b = min_t(unsigned, bkey_val_bytes(k.k), val_size);
578 
579 		memcpy(val, k.v, b);
580 		if (unlikely(b < sizeof(*val)))
581 			memset((void *) val + b, 0, sizeof(*val) - b);
582 		bch2_trans_iter_exit(trans, &iter);
583 	}
584 
585 	return ret;
586 }
587 
588 #define bch2_bkey_get_val_typed(_trans, _btree_id, _pos, _flags, _type, _val)\
589 	__bch2_bkey_get_val_typed(_trans, _btree_id, _pos, _flags,	\
590 				  KEY_TYPE_##_type, sizeof(*_val), _val)
591 
592 void bch2_trans_srcu_unlock(struct btree_trans *);
593 
594 u32 bch2_trans_begin(struct btree_trans *);
595 
596 /*
597  * XXX
598  * this does not handle transaction restarts from bch2_btree_iter_next_node()
599  * correctly
600  */
601 #define __for_each_btree_node(_trans, _iter, _btree_id, _start,		\
602 			      _locks_want, _depth, _flags, _b, _ret)	\
603 	for (bch2_trans_node_iter_init((_trans), &(_iter), (_btree_id),	\
604 				_start, _locks_want, _depth, _flags);	\
605 	     (_b) = bch2_btree_iter_peek_node_and_restart(&(_iter)),	\
606 	     !((_ret) = PTR_ERR_OR_ZERO(_b)) && (_b);			\
607 	     (_b) = bch2_btree_iter_next_node(&(_iter)))
608 
609 #define for_each_btree_node(_trans, _iter, _btree_id, _start,		\
610 			    _flags, _b, _ret)				\
611 	__for_each_btree_node(_trans, _iter, _btree_id, _start,		\
612 			      0, 0, _flags, _b, _ret)
613 
614 static inline struct bkey_s_c bch2_btree_iter_peek_prev_type(struct btree_iter *iter,
615 							     unsigned flags)
616 {
617 	return  flags & BTREE_ITER_SLOTS      ? bch2_btree_iter_peek_slot(iter) :
618 						bch2_btree_iter_peek_prev(iter);
619 }
620 
621 static inline struct bkey_s_c bch2_btree_iter_peek_type(struct btree_iter *iter,
622 							unsigned flags)
623 {
624 	return  flags & BTREE_ITER_SLOTS      ? bch2_btree_iter_peek_slot(iter) :
625 						bch2_btree_iter_peek(iter);
626 }
627 
628 static inline struct bkey_s_c bch2_btree_iter_peek_upto_type(struct btree_iter *iter,
629 							     struct bpos end,
630 							     unsigned flags)
631 {
632 	if (!(flags & BTREE_ITER_SLOTS))
633 		return bch2_btree_iter_peek_upto(iter, end);
634 
635 	if (bkey_gt(iter->pos, end))
636 		return bkey_s_c_null;
637 
638 	return bch2_btree_iter_peek_slot(iter);
639 }
640 
641 int __bch2_btree_trans_too_many_iters(struct btree_trans *);
642 
643 static inline int btree_trans_too_many_iters(struct btree_trans *trans)
644 {
645 	if (bitmap_weight(trans->paths_allocated, trans->nr_paths) > BTREE_ITER_INITIAL - 8)
646 		return __bch2_btree_trans_too_many_iters(trans);
647 
648 	return 0;
649 }
650 
651 /*
652  * goto instead of loop, so that when used inside for_each_btree_key2()
653  * break/continue work correctly
654  */
655 #define lockrestart_do(_trans, _do)					\
656 ({									\
657 	__label__ transaction_restart;					\
658 	u32 _restart_count;						\
659 	int _ret2;							\
660 transaction_restart:							\
661 	_restart_count = bch2_trans_begin(_trans);			\
662 	_ret2 = (_do);							\
663 									\
664 	if (bch2_err_matches(_ret2, BCH_ERR_transaction_restart))	\
665 		goto transaction_restart;				\
666 									\
667 	if (!_ret2)							\
668 		bch2_trans_verify_not_restarted(_trans, _restart_count);\
669 	_ret2;								\
670 })
671 
672 /*
673  * nested_lockrestart_do(), nested_commit_do():
674  *
675  * These are like lockrestart_do() and commit_do(), with two differences:
676  *
677  *  - We don't call bch2_trans_begin() unless we had a transaction restart
678  *  - We return -BCH_ERR_transaction_restart_nested if we succeeded after a
679  *  transaction restart
680  */
681 #define nested_lockrestart_do(_trans, _do)				\
682 ({									\
683 	u32 _restart_count, _orig_restart_count;			\
684 	int _ret2;							\
685 									\
686 	_restart_count = _orig_restart_count = (_trans)->restart_count;	\
687 									\
688 	while (bch2_err_matches(_ret2 = (_do), BCH_ERR_transaction_restart))\
689 		_restart_count = bch2_trans_begin(_trans);		\
690 									\
691 	if (!_ret2)							\
692 		bch2_trans_verify_not_restarted(_trans, _restart_count);\
693 									\
694 	_ret2 ?: trans_was_restarted(_trans, _restart_count);		\
695 })
696 
697 #define for_each_btree_key_upto(_trans, _iter, _btree_id,		\
698 				_start, _end, _flags, _k, _do)		\
699 ({									\
700 	struct btree_iter _iter;					\
701 	struct bkey_s_c _k;						\
702 	int _ret3 = 0;							\
703 									\
704 	bch2_trans_iter_init((_trans), &(_iter), (_btree_id),		\
705 			     (_start), (_flags));			\
706 									\
707 	do {								\
708 		_ret3 = lockrestart_do(_trans, ({			\
709 			(_k) = bch2_btree_iter_peek_upto_type(&(_iter),	\
710 						_end, (_flags));	\
711 			if (!(_k).k)					\
712 				break;					\
713 									\
714 			bkey_err(_k) ?: (_do);				\
715 		}));							\
716 	} while (!_ret3 && bch2_btree_iter_advance(&(_iter)));		\
717 									\
718 	bch2_trans_iter_exit((_trans), &(_iter));			\
719 	_ret3;								\
720 })
721 
722 #define for_each_btree_key(_trans, _iter, _btree_id,			\
723 			   _start, _flags, _k, _do)			\
724 	for_each_btree_key_upto(_trans, _iter, _btree_id, _start,	\
725 				 SPOS_MAX, _flags, _k, _do)
726 
727 #define for_each_btree_key_reverse(_trans, _iter, _btree_id,		\
728 				   _start, _flags, _k, _do)		\
729 ({									\
730 	struct btree_iter _iter;					\
731 	struct bkey_s_c _k;						\
732 	int _ret3 = 0;							\
733 									\
734 	bch2_trans_iter_init((_trans), &(_iter), (_btree_id),		\
735 			     (_start), (_flags));			\
736 									\
737 	do {								\
738 		_ret3 = lockrestart_do(_trans, ({			\
739 			(_k) = bch2_btree_iter_peek_prev_type(&(_iter),	\
740 							(_flags));	\
741 			if (!(_k).k)					\
742 				break;					\
743 									\
744 			bkey_err(_k) ?: (_do);				\
745 		}));							\
746 	} while (!_ret3 && bch2_btree_iter_rewind(&(_iter)));		\
747 									\
748 	bch2_trans_iter_exit((_trans), &(_iter));			\
749 	_ret3;								\
750 })
751 
752 #define for_each_btree_key_commit(_trans, _iter, _btree_id,		\
753 				  _start, _iter_flags, _k,		\
754 				  _disk_res, _journal_seq, _commit_flags,\
755 				  _do)					\
756 	for_each_btree_key(_trans, _iter, _btree_id, _start, _iter_flags, _k,\
757 			    (_do) ?: bch2_trans_commit(_trans, (_disk_res),\
758 					(_journal_seq), (_commit_flags)))
759 
760 #define for_each_btree_key_reverse_commit(_trans, _iter, _btree_id,	\
761 				  _start, _iter_flags, _k,		\
762 				  _disk_res, _journal_seq, _commit_flags,\
763 				  _do)					\
764 	for_each_btree_key_reverse(_trans, _iter, _btree_id, _start, _iter_flags, _k,\
765 			    (_do) ?: bch2_trans_commit(_trans, (_disk_res),\
766 					(_journal_seq), (_commit_flags)))
767 
768 #define for_each_btree_key_upto_commit(_trans, _iter, _btree_id,	\
769 				  _start, _end, _iter_flags, _k,	\
770 				  _disk_res, _journal_seq, _commit_flags,\
771 				  _do)					\
772 	for_each_btree_key_upto(_trans, _iter, _btree_id, _start, _end, _iter_flags, _k,\
773 			    (_do) ?: bch2_trans_commit(_trans, (_disk_res),\
774 					(_journal_seq), (_commit_flags)))
775 
776 struct bkey_s_c bch2_btree_iter_peek_and_restart_outlined(struct btree_iter *);
777 
778 static inline struct bkey_s_c
779 __bch2_btree_iter_peek_and_restart(struct btree_trans *trans,
780 				   struct btree_iter *iter, unsigned flags)
781 {
782 	struct bkey_s_c k;
783 
784 	while (btree_trans_too_many_iters(trans) ||
785 	       (k = bch2_btree_iter_peek_type(iter, flags),
786 		bch2_err_matches(bkey_err(k), BCH_ERR_transaction_restart)))
787 		bch2_trans_begin(trans);
788 
789 	return k;
790 }
791 
792 #define for_each_btree_key_old(_trans, _iter, _btree_id,		\
793 			   _start, _flags, _k, _ret)			\
794 	for (bch2_trans_iter_init((_trans), &(_iter), (_btree_id),	\
795 				  (_start), (_flags));			\
796 	     (_k) = __bch2_btree_iter_peek_and_restart((_trans), &(_iter), _flags),\
797 	     !((_ret) = bkey_err(_k)) && (_k).k;			\
798 	     bch2_btree_iter_advance(&(_iter)))
799 
800 #define for_each_btree_key_upto_norestart(_trans, _iter, _btree_id,	\
801 			   _start, _end, _flags, _k, _ret)		\
802 	for (bch2_trans_iter_init((_trans), &(_iter), (_btree_id),	\
803 				  (_start), (_flags));			\
804 	     (_k) = bch2_btree_iter_peek_upto_type(&(_iter), _end, _flags),\
805 	     !((_ret) = bkey_err(_k)) && (_k).k;			\
806 	     bch2_btree_iter_advance(&(_iter)))
807 
808 #define for_each_btree_key_upto_continue_norestart(_iter, _end, _flags, _k, _ret)\
809 	for (;									\
810 	     (_k) = bch2_btree_iter_peek_upto_type(&(_iter), _end, _flags),	\
811 	     !((_ret) = bkey_err(_k)) && (_k).k;				\
812 	     bch2_btree_iter_advance(&(_iter)))
813 
814 #define for_each_btree_key_norestart(_trans, _iter, _btree_id,		\
815 			   _start, _flags, _k, _ret)			\
816 	for_each_btree_key_upto_norestart(_trans, _iter, _btree_id, _start,\
817 					  SPOS_MAX, _flags, _k, _ret)
818 
819 #define for_each_btree_key_continue_norestart(_iter, _flags, _k, _ret)	\
820 	for_each_btree_key_upto_continue_norestart(_iter, SPOS_MAX, _flags, _k, _ret)
821 
822 /*
823  * This should not be used in a fastpath, without first trying _do in
824  * nonblocking mode - it will cause excessive transaction restarts and
825  * potentially livelocking:
826  */
827 #define drop_locks_do(_trans, _do)					\
828 ({									\
829 	bch2_trans_unlock(_trans);					\
830 	_do ?: bch2_trans_relock(_trans);				\
831 })
832 
833 #define allocate_dropping_locks_errcode(_trans, _do)			\
834 ({									\
835 	gfp_t _gfp = GFP_NOWAIT|__GFP_NOWARN;				\
836 	int _ret = _do;							\
837 									\
838 	if (bch2_err_matches(_ret, ENOMEM)) {				\
839 		_gfp = GFP_KERNEL;					\
840 		_ret = drop_locks_do(trans, _do);			\
841 	}								\
842 	_ret;								\
843 })
844 
845 #define allocate_dropping_locks(_trans, _ret, _do)			\
846 ({									\
847 	gfp_t _gfp = GFP_NOWAIT|__GFP_NOWARN;				\
848 	typeof(_do) _p = _do;						\
849 									\
850 	_ret = 0;							\
851 	if (unlikely(!_p)) {						\
852 		_gfp = GFP_KERNEL;					\
853 		_ret = drop_locks_do(trans, ((_p = _do), 0));		\
854 	}								\
855 	_p;								\
856 })
857 
858 void bch2_trans_updates_to_text(struct printbuf *, struct btree_trans *);
859 void bch2_trans_paths_to_text(struct printbuf *, struct btree_trans *);
860 void bch2_dump_trans_updates(struct btree_trans *);
861 void bch2_dump_trans_paths_updates(struct btree_trans *);
862 
863 struct btree_trans *__bch2_trans_get(struct bch_fs *, unsigned);
864 void bch2_trans_put(struct btree_trans *);
865 
866 extern const char *bch2_btree_transaction_fns[BCH_TRANSACTIONS_NR];
867 unsigned bch2_trans_get_fn_idx(const char *);
868 
869 #define bch2_trans_get(_c)						\
870 ({									\
871 	static unsigned trans_fn_idx;					\
872 									\
873 	if (unlikely(!trans_fn_idx))					\
874 		trans_fn_idx = bch2_trans_get_fn_idx(__func__);		\
875 	__bch2_trans_get(_c, trans_fn_idx);				\
876 })
877 
878 void bch2_btree_trans_to_text(struct printbuf *, struct btree_trans *);
879 
880 void bch2_fs_btree_iter_exit(struct bch_fs *);
881 void bch2_fs_btree_iter_init_early(struct bch_fs *);
882 int bch2_fs_btree_iter_init(struct bch_fs *);
883 
884 #endif /* _BCACHEFS_BTREE_ITER_H */
885