xref: /linux/fs/bcachefs/btree_iter.h (revision 24168c5e6dfbdd5b414f048f47f75d64533296ca)
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 void bch2_trans_verify_not_unlocked(struct btree_trans *);
220 
221 static inline int __must_check bch2_btree_path_traverse(struct btree_trans *trans,
222 					  btree_path_idx_t path, unsigned flags)
223 {
224 	bch2_trans_verify_not_unlocked(trans);
225 
226 	if (trans->paths[path].uptodate < BTREE_ITER_NEED_RELOCK)
227 		return 0;
228 
229 	return bch2_btree_path_traverse_one(trans, path, flags, _RET_IP_);
230 }
231 
232 btree_path_idx_t bch2_path_get(struct btree_trans *, enum btree_id, struct bpos,
233 				 unsigned, unsigned, unsigned, unsigned long);
234 btree_path_idx_t bch2_path_get_unlocked_mut(struct btree_trans *, enum btree_id,
235 					    unsigned, struct bpos);
236 
237 struct bkey_s_c bch2_btree_path_peek_slot(struct btree_path *, struct bkey *);
238 
239 /*
240  * bch2_btree_path_peek_slot() for a cached iterator might return a key in a
241  * different snapshot:
242  */
243 static inline struct bkey_s_c bch2_btree_path_peek_slot_exact(struct btree_path *path, struct bkey *u)
244 {
245 	struct bkey_s_c k = bch2_btree_path_peek_slot(path, u);
246 
247 	if (k.k && bpos_eq(path->pos, k.k->p))
248 		return k;
249 
250 	bkey_init(u);
251 	u->p = path->pos;
252 	return (struct bkey_s_c) { u, NULL };
253 }
254 
255 struct bkey_i *bch2_btree_journal_peek_slot(struct btree_trans *,
256 					struct btree_iter *, struct bpos);
257 
258 void bch2_btree_path_level_init(struct btree_trans *, struct btree_path *, struct btree *);
259 
260 int __bch2_trans_mutex_lock(struct btree_trans *, struct mutex *);
261 
262 static inline int bch2_trans_mutex_lock(struct btree_trans *trans, struct mutex *lock)
263 {
264 	return mutex_trylock(lock)
265 		? 0
266 		: __bch2_trans_mutex_lock(trans, lock);
267 }
268 
269 #ifdef CONFIG_BCACHEFS_DEBUG
270 void bch2_trans_verify_paths(struct btree_trans *);
271 void bch2_assert_pos_locked(struct btree_trans *, enum btree_id,
272 			    struct bpos, bool);
273 #else
274 static inline void bch2_trans_verify_paths(struct btree_trans *trans) {}
275 static inline void bch2_assert_pos_locked(struct btree_trans *trans, enum btree_id id,
276 					  struct bpos pos, bool key_cache) {}
277 #endif
278 
279 void bch2_btree_path_fix_key_modified(struct btree_trans *trans,
280 				      struct btree *, struct bkey_packed *);
281 void bch2_btree_node_iter_fix(struct btree_trans *trans, struct btree_path *,
282 			      struct btree *, struct btree_node_iter *,
283 			      struct bkey_packed *, unsigned, unsigned);
284 
285 int bch2_btree_path_relock_intent(struct btree_trans *, struct btree_path *);
286 
287 void bch2_path_put(struct btree_trans *, btree_path_idx_t, bool);
288 
289 int bch2_trans_relock(struct btree_trans *);
290 int bch2_trans_relock_notrace(struct btree_trans *);
291 void bch2_trans_unlock(struct btree_trans *);
292 void bch2_trans_unlock_long(struct btree_trans *);
293 
294 static inline int trans_was_restarted(struct btree_trans *trans, u32 restart_count)
295 {
296 	return restart_count != trans->restart_count
297 		? -BCH_ERR_transaction_restart_nested
298 		: 0;
299 }
300 
301 void __noreturn bch2_trans_restart_error(struct btree_trans *, u32);
302 
303 static inline void bch2_trans_verify_not_restarted(struct btree_trans *trans,
304 						   u32 restart_count)
305 {
306 	if (trans_was_restarted(trans, restart_count))
307 		bch2_trans_restart_error(trans, restart_count);
308 }
309 
310 void __noreturn bch2_trans_in_restart_error(struct btree_trans *);
311 
312 static inline void bch2_trans_verify_not_in_restart(struct btree_trans *trans)
313 {
314 	if (trans->restarted)
315 		bch2_trans_in_restart_error(trans);
316 }
317 
318 void __noreturn bch2_trans_unlocked_error(struct btree_trans *);
319 
320 static inline void bch2_trans_verify_not_unlocked(struct btree_trans *trans)
321 {
322 	if (!trans->locked)
323 		bch2_trans_unlocked_error(trans);
324 }
325 
326 __always_inline
327 static int btree_trans_restart_nounlock(struct btree_trans *trans, int err)
328 {
329 	BUG_ON(err <= 0);
330 	BUG_ON(!bch2_err_matches(-err, BCH_ERR_transaction_restart));
331 
332 	trans->restarted = err;
333 	trans->last_restarted_ip = _THIS_IP_;
334 	return -err;
335 }
336 
337 __always_inline
338 static int btree_trans_restart(struct btree_trans *trans, int err)
339 {
340 	btree_trans_restart_nounlock(trans, err);
341 	return -err;
342 }
343 
344 bool bch2_btree_node_upgrade(struct btree_trans *,
345 			     struct btree_path *, unsigned);
346 
347 void __bch2_btree_path_downgrade(struct btree_trans *, struct btree_path *, unsigned);
348 
349 static inline void bch2_btree_path_downgrade(struct btree_trans *trans,
350 					     struct btree_path *path)
351 {
352 	unsigned new_locks_want = path->level + !!path->intent_ref;
353 
354 	if (path->locks_want > new_locks_want)
355 		__bch2_btree_path_downgrade(trans, path, new_locks_want);
356 }
357 
358 void bch2_trans_downgrade(struct btree_trans *);
359 
360 void bch2_trans_node_add(struct btree_trans *trans, struct btree_path *, struct btree *);
361 void bch2_trans_node_reinit_iter(struct btree_trans *, struct btree *);
362 
363 int __must_check __bch2_btree_iter_traverse(struct btree_iter *iter);
364 int __must_check bch2_btree_iter_traverse(struct btree_iter *);
365 
366 struct btree *bch2_btree_iter_peek_node(struct btree_iter *);
367 struct btree *bch2_btree_iter_peek_node_and_restart(struct btree_iter *);
368 struct btree *bch2_btree_iter_next_node(struct btree_iter *);
369 
370 struct bkey_s_c bch2_btree_iter_peek_upto(struct btree_iter *, struct bpos);
371 struct bkey_s_c bch2_btree_iter_next(struct btree_iter *);
372 
373 static inline struct bkey_s_c bch2_btree_iter_peek(struct btree_iter *iter)
374 {
375 	return bch2_btree_iter_peek_upto(iter, SPOS_MAX);
376 }
377 
378 struct bkey_s_c bch2_btree_iter_peek_prev(struct btree_iter *);
379 struct bkey_s_c bch2_btree_iter_prev(struct btree_iter *);
380 
381 struct bkey_s_c bch2_btree_iter_peek_slot(struct btree_iter *);
382 struct bkey_s_c bch2_btree_iter_next_slot(struct btree_iter *);
383 struct bkey_s_c bch2_btree_iter_prev_slot(struct btree_iter *);
384 
385 bool bch2_btree_iter_advance(struct btree_iter *);
386 bool bch2_btree_iter_rewind(struct btree_iter *);
387 
388 static inline void __bch2_btree_iter_set_pos(struct btree_iter *iter, struct bpos new_pos)
389 {
390 	iter->k.type = KEY_TYPE_deleted;
391 	iter->k.p.inode		= iter->pos.inode	= new_pos.inode;
392 	iter->k.p.offset	= iter->pos.offset	= new_pos.offset;
393 	iter->k.p.snapshot	= iter->pos.snapshot	= new_pos.snapshot;
394 	iter->k.size = 0;
395 }
396 
397 static inline void bch2_btree_iter_set_pos(struct btree_iter *iter, struct bpos new_pos)
398 {
399 	struct btree_trans *trans = iter->trans;
400 
401 	if (unlikely(iter->update_path))
402 		bch2_path_put(trans, iter->update_path,
403 			      iter->flags & BTREE_ITER_intent);
404 	iter->update_path = 0;
405 
406 	if (!(iter->flags & BTREE_ITER_all_snapshots))
407 		new_pos.snapshot = iter->snapshot;
408 
409 	__bch2_btree_iter_set_pos(iter, new_pos);
410 }
411 
412 static inline void bch2_btree_iter_set_pos_to_extent_start(struct btree_iter *iter)
413 {
414 	BUG_ON(!(iter->flags & BTREE_ITER_is_extents));
415 	iter->pos = bkey_start_pos(&iter->k);
416 }
417 
418 static inline void bch2_btree_iter_set_snapshot(struct btree_iter *iter, u32 snapshot)
419 {
420 	struct bpos pos = iter->pos;
421 
422 	iter->snapshot = snapshot;
423 	pos.snapshot = snapshot;
424 	bch2_btree_iter_set_pos(iter, pos);
425 }
426 
427 void bch2_trans_iter_exit(struct btree_trans *, struct btree_iter *);
428 
429 static inline unsigned __bch2_btree_iter_flags(struct btree_trans *trans,
430 					       unsigned btree_id,
431 					       unsigned flags)
432 {
433 	if (!(flags & (BTREE_ITER_all_snapshots|BTREE_ITER_not_extents)) &&
434 	    btree_id_is_extents(btree_id))
435 		flags |= BTREE_ITER_is_extents;
436 
437 	if (!(flags & BTREE_ITER_snapshot_field) &&
438 	    !btree_type_has_snapshot_field(btree_id))
439 		flags &= ~BTREE_ITER_all_snapshots;
440 
441 	if (!(flags & BTREE_ITER_all_snapshots) &&
442 	    btree_type_has_snapshots(btree_id))
443 		flags |= BTREE_ITER_filter_snapshots;
444 
445 	if (trans->journal_replay_not_finished)
446 		flags |= BTREE_ITER_with_journal;
447 
448 	return flags;
449 }
450 
451 static inline unsigned bch2_btree_iter_flags(struct btree_trans *trans,
452 					     unsigned btree_id,
453 					     unsigned flags)
454 {
455 	if (!btree_id_cached(trans->c, btree_id)) {
456 		flags &= ~BTREE_ITER_cached;
457 		flags &= ~BTREE_ITER_with_key_cache;
458 	} else if (!(flags & BTREE_ITER_cached))
459 		flags |= BTREE_ITER_with_key_cache;
460 
461 	return __bch2_btree_iter_flags(trans, btree_id, flags);
462 }
463 
464 static inline void bch2_trans_iter_init_common(struct btree_trans *trans,
465 					  struct btree_iter *iter,
466 					  unsigned btree_id, struct bpos pos,
467 					  unsigned locks_want,
468 					  unsigned depth,
469 					  unsigned flags,
470 					  unsigned long ip)
471 {
472 	iter->trans		= trans;
473 	iter->update_path	= 0;
474 	iter->key_cache_path	= 0;
475 	iter->btree_id		= btree_id;
476 	iter->min_depth		= 0;
477 	iter->flags		= flags;
478 	iter->snapshot		= pos.snapshot;
479 	iter->pos		= pos;
480 	iter->k			= POS_KEY(pos);
481 	iter->journal_idx	= 0;
482 #ifdef CONFIG_BCACHEFS_DEBUG
483 	iter->ip_allocated = ip;
484 #endif
485 	iter->path = bch2_path_get(trans, btree_id, iter->pos,
486 				   locks_want, depth, flags, ip);
487 }
488 
489 void bch2_trans_iter_init_outlined(struct btree_trans *, struct btree_iter *,
490 			  enum btree_id, struct bpos, unsigned);
491 
492 static inline void bch2_trans_iter_init(struct btree_trans *trans,
493 			  struct btree_iter *iter,
494 			  unsigned btree_id, struct bpos pos,
495 			  unsigned flags)
496 {
497 	if (__builtin_constant_p(btree_id) &&
498 	    __builtin_constant_p(flags))
499 		bch2_trans_iter_init_common(trans, iter, btree_id, pos, 0, 0,
500 				bch2_btree_iter_flags(trans, btree_id, flags),
501 				_THIS_IP_);
502 	else
503 		bch2_trans_iter_init_outlined(trans, iter, btree_id, pos, flags);
504 }
505 
506 void bch2_trans_node_iter_init(struct btree_trans *, struct btree_iter *,
507 			       enum btree_id, struct bpos,
508 			       unsigned, unsigned, unsigned);
509 void bch2_trans_copy_iter(struct btree_iter *, struct btree_iter *);
510 
511 void bch2_set_btree_iter_dontneed(struct btree_iter *);
512 
513 void *__bch2_trans_kmalloc(struct btree_trans *, size_t);
514 
515 static inline void *bch2_trans_kmalloc(struct btree_trans *trans, size_t size)
516 {
517 	size = roundup(size, 8);
518 
519 	if (likely(trans->mem_top + size <= trans->mem_bytes)) {
520 		void *p = trans->mem + trans->mem_top;
521 
522 		trans->mem_top += size;
523 		memset(p, 0, size);
524 		return p;
525 	} else {
526 		return __bch2_trans_kmalloc(trans, size);
527 	}
528 }
529 
530 static inline void *bch2_trans_kmalloc_nomemzero(struct btree_trans *trans, size_t size)
531 {
532 	size = round_up(size, 8);
533 
534 	if (likely(trans->mem_top + size <= trans->mem_bytes)) {
535 		void *p = trans->mem + trans->mem_top;
536 
537 		trans->mem_top += size;
538 		return p;
539 	} else {
540 		return __bch2_trans_kmalloc(trans, size);
541 	}
542 }
543 
544 static inline struct bkey_s_c __bch2_bkey_get_iter(struct btree_trans *trans,
545 				struct btree_iter *iter,
546 				unsigned btree_id, struct bpos pos,
547 				unsigned flags, unsigned type)
548 {
549 	struct bkey_s_c k;
550 
551 	bch2_trans_iter_init(trans, iter, btree_id, pos, flags);
552 	k = bch2_btree_iter_peek_slot(iter);
553 
554 	if (!bkey_err(k) && type && k.k->type != type)
555 		k = bkey_s_c_err(-BCH_ERR_ENOENT_bkey_type_mismatch);
556 	if (unlikely(bkey_err(k)))
557 		bch2_trans_iter_exit(trans, iter);
558 	return k;
559 }
560 
561 static inline struct bkey_s_c bch2_bkey_get_iter(struct btree_trans *trans,
562 				struct btree_iter *iter,
563 				unsigned btree_id, struct bpos pos,
564 				unsigned flags)
565 {
566 	return __bch2_bkey_get_iter(trans, iter, btree_id, pos, flags, 0);
567 }
568 
569 #define bch2_bkey_get_iter_typed(_trans, _iter, _btree_id, _pos, _flags, _type)\
570 	bkey_s_c_to_##_type(__bch2_bkey_get_iter(_trans, _iter,			\
571 				       _btree_id, _pos, _flags, KEY_TYPE_##_type))
572 
573 static inline int __bch2_bkey_get_val_typed(struct btree_trans *trans,
574 				unsigned btree_id, struct bpos pos,
575 				unsigned flags, unsigned type,
576 				unsigned val_size, void *val)
577 {
578 	struct btree_iter iter;
579 	struct bkey_s_c k;
580 	int ret;
581 
582 	k = __bch2_bkey_get_iter(trans, &iter, btree_id, pos, flags, type);
583 	ret = bkey_err(k);
584 	if (!ret) {
585 		unsigned b = min_t(unsigned, bkey_val_bytes(k.k), val_size);
586 
587 		memcpy(val, k.v, b);
588 		if (unlikely(b < sizeof(*val)))
589 			memset((void *) val + b, 0, sizeof(*val) - b);
590 		bch2_trans_iter_exit(trans, &iter);
591 	}
592 
593 	return ret;
594 }
595 
596 #define bch2_bkey_get_val_typed(_trans, _btree_id, _pos, _flags, _type, _val)\
597 	__bch2_bkey_get_val_typed(_trans, _btree_id, _pos, _flags,	\
598 				  KEY_TYPE_##_type, sizeof(*_val), _val)
599 
600 void bch2_trans_srcu_unlock(struct btree_trans *);
601 
602 u32 bch2_trans_begin(struct btree_trans *);
603 
604 /*
605  * XXX
606  * this does not handle transaction restarts from bch2_btree_iter_next_node()
607  * correctly
608  */
609 #define __for_each_btree_node(_trans, _iter, _btree_id, _start,		\
610 			      _locks_want, _depth, _flags, _b, _ret)	\
611 	for (bch2_trans_node_iter_init((_trans), &(_iter), (_btree_id),	\
612 				_start, _locks_want, _depth, _flags);	\
613 	     (_b) = bch2_btree_iter_peek_node_and_restart(&(_iter)),	\
614 	     !((_ret) = PTR_ERR_OR_ZERO(_b)) && (_b);			\
615 	     (_b) = bch2_btree_iter_next_node(&(_iter)))
616 
617 #define for_each_btree_node(_trans, _iter, _btree_id, _start,		\
618 			    _flags, _b, _ret)				\
619 	__for_each_btree_node(_trans, _iter, _btree_id, _start,		\
620 			      0, 0, _flags, _b, _ret)
621 
622 static inline struct bkey_s_c bch2_btree_iter_peek_prev_type(struct btree_iter *iter,
623 							     unsigned flags)
624 {
625 	return  flags & BTREE_ITER_slots      ? bch2_btree_iter_peek_slot(iter) :
626 						bch2_btree_iter_peek_prev(iter);
627 }
628 
629 static inline struct bkey_s_c bch2_btree_iter_peek_type(struct btree_iter *iter,
630 							unsigned flags)
631 {
632 	return  flags & BTREE_ITER_slots      ? bch2_btree_iter_peek_slot(iter) :
633 						bch2_btree_iter_peek(iter);
634 }
635 
636 static inline struct bkey_s_c bch2_btree_iter_peek_upto_type(struct btree_iter *iter,
637 							     struct bpos end,
638 							     unsigned flags)
639 {
640 	if (!(flags & BTREE_ITER_slots))
641 		return bch2_btree_iter_peek_upto(iter, end);
642 
643 	if (bkey_gt(iter->pos, end))
644 		return bkey_s_c_null;
645 
646 	return bch2_btree_iter_peek_slot(iter);
647 }
648 
649 int __bch2_btree_trans_too_many_iters(struct btree_trans *);
650 
651 static inline int btree_trans_too_many_iters(struct btree_trans *trans)
652 {
653 	if (bitmap_weight(trans->paths_allocated, trans->nr_paths) > BTREE_ITER_NORMAL_LIMIT - 8)
654 		return __bch2_btree_trans_too_many_iters(trans);
655 
656 	return 0;
657 }
658 
659 /*
660  * goto instead of loop, so that when used inside for_each_btree_key2()
661  * break/continue work correctly
662  */
663 #define lockrestart_do(_trans, _do)					\
664 ({									\
665 	__label__ transaction_restart;					\
666 	u32 _restart_count;						\
667 	int _ret2;							\
668 transaction_restart:							\
669 	_restart_count = bch2_trans_begin(_trans);			\
670 	_ret2 = (_do);							\
671 									\
672 	if (bch2_err_matches(_ret2, BCH_ERR_transaction_restart))	\
673 		goto transaction_restart;				\
674 									\
675 	if (!_ret2)							\
676 		bch2_trans_verify_not_restarted(_trans, _restart_count);\
677 	_ret2;								\
678 })
679 
680 /*
681  * nested_lockrestart_do(), nested_commit_do():
682  *
683  * These are like lockrestart_do() and commit_do(), with two differences:
684  *
685  *  - We don't call bch2_trans_begin() unless we had a transaction restart
686  *  - We return -BCH_ERR_transaction_restart_nested if we succeeded after a
687  *  transaction restart
688  */
689 #define nested_lockrestart_do(_trans, _do)				\
690 ({									\
691 	u32 _restart_count, _orig_restart_count;			\
692 	int _ret2;							\
693 									\
694 	_restart_count = _orig_restart_count = (_trans)->restart_count;	\
695 									\
696 	while (bch2_err_matches(_ret2 = (_do), BCH_ERR_transaction_restart))\
697 		_restart_count = bch2_trans_begin(_trans);		\
698 									\
699 	if (!_ret2)							\
700 		bch2_trans_verify_not_restarted(_trans, _restart_count);\
701 									\
702 	_ret2 ?: trans_was_restarted(_trans, _restart_count);		\
703 })
704 
705 #define for_each_btree_key_upto_continue(_trans, _iter,			\
706 					 _end, _flags, _k, _do)		\
707 ({									\
708 	struct bkey_s_c _k;						\
709 	int _ret3 = 0;							\
710 									\
711 	do {								\
712 		_ret3 = lockrestart_do(_trans, ({			\
713 			(_k) = bch2_btree_iter_peek_upto_type(&(_iter),	\
714 						_end, (_flags));	\
715 			if (!(_k).k)					\
716 				break;					\
717 									\
718 			bkey_err(_k) ?: (_do);				\
719 		}));							\
720 	} while (!_ret3 && bch2_btree_iter_advance(&(_iter)));		\
721 									\
722 	bch2_trans_iter_exit((_trans), &(_iter));			\
723 	_ret3;								\
724 })
725 
726 #define for_each_btree_key_continue(_trans, _iter, _flags, _k, _do)	\
727 	for_each_btree_key_upto_continue(_trans, _iter, SPOS_MAX, _flags, _k, _do)
728 
729 #define for_each_btree_key_upto(_trans, _iter, _btree_id,		\
730 				_start, _end, _flags, _k, _do)		\
731 ({									\
732 	bch2_trans_begin(trans);					\
733 									\
734 	struct btree_iter _iter;					\
735 	bch2_trans_iter_init((_trans), &(_iter), (_btree_id),		\
736 			     (_start), (_flags));			\
737 									\
738 	for_each_btree_key_upto_continue(_trans, _iter, _end, _flags, _k, _do);\
739 })
740 
741 #define for_each_btree_key(_trans, _iter, _btree_id,			\
742 			   _start, _flags, _k, _do)			\
743 	for_each_btree_key_upto(_trans, _iter, _btree_id, _start,	\
744 				 SPOS_MAX, _flags, _k, _do)
745 
746 #define for_each_btree_key_reverse(_trans, _iter, _btree_id,		\
747 				   _start, _flags, _k, _do)		\
748 ({									\
749 	struct btree_iter _iter;					\
750 	struct bkey_s_c _k;						\
751 	int _ret3 = 0;							\
752 									\
753 	bch2_trans_iter_init((_trans), &(_iter), (_btree_id),		\
754 			     (_start), (_flags));			\
755 									\
756 	do {								\
757 		_ret3 = lockrestart_do(_trans, ({			\
758 			(_k) = bch2_btree_iter_peek_prev_type(&(_iter),	\
759 							(_flags));	\
760 			if (!(_k).k)					\
761 				break;					\
762 									\
763 			bkey_err(_k) ?: (_do);				\
764 		}));							\
765 	} while (!_ret3 && bch2_btree_iter_rewind(&(_iter)));		\
766 									\
767 	bch2_trans_iter_exit((_trans), &(_iter));			\
768 	_ret3;								\
769 })
770 
771 #define for_each_btree_key_commit(_trans, _iter, _btree_id,		\
772 				  _start, _iter_flags, _k,		\
773 				  _disk_res, _journal_seq, _commit_flags,\
774 				  _do)					\
775 	for_each_btree_key(_trans, _iter, _btree_id, _start, _iter_flags, _k,\
776 			    (_do) ?: bch2_trans_commit(_trans, (_disk_res),\
777 					(_journal_seq), (_commit_flags)))
778 
779 #define for_each_btree_key_reverse_commit(_trans, _iter, _btree_id,	\
780 				  _start, _iter_flags, _k,		\
781 				  _disk_res, _journal_seq, _commit_flags,\
782 				  _do)					\
783 	for_each_btree_key_reverse(_trans, _iter, _btree_id, _start, _iter_flags, _k,\
784 			    (_do) ?: bch2_trans_commit(_trans, (_disk_res),\
785 					(_journal_seq), (_commit_flags)))
786 
787 #define for_each_btree_key_upto_commit(_trans, _iter, _btree_id,	\
788 				  _start, _end, _iter_flags, _k,	\
789 				  _disk_res, _journal_seq, _commit_flags,\
790 				  _do)					\
791 	for_each_btree_key_upto(_trans, _iter, _btree_id, _start, _end, _iter_flags, _k,\
792 			    (_do) ?: bch2_trans_commit(_trans, (_disk_res),\
793 					(_journal_seq), (_commit_flags)))
794 
795 struct bkey_s_c bch2_btree_iter_peek_and_restart_outlined(struct btree_iter *);
796 
797 static inline struct bkey_s_c
798 __bch2_btree_iter_peek_and_restart(struct btree_trans *trans,
799 				   struct btree_iter *iter, unsigned flags)
800 {
801 	struct bkey_s_c k;
802 
803 	while (btree_trans_too_many_iters(trans) ||
804 	       (k = bch2_btree_iter_peek_type(iter, flags),
805 		bch2_err_matches(bkey_err(k), BCH_ERR_transaction_restart)))
806 		bch2_trans_begin(trans);
807 
808 	return k;
809 }
810 
811 #define for_each_btree_key_upto_norestart(_trans, _iter, _btree_id,	\
812 			   _start, _end, _flags, _k, _ret)		\
813 	for (bch2_trans_iter_init((_trans), &(_iter), (_btree_id),	\
814 				  (_start), (_flags));			\
815 	     (_k) = bch2_btree_iter_peek_upto_type(&(_iter), _end, _flags),\
816 	     !((_ret) = bkey_err(_k)) && (_k).k;			\
817 	     bch2_btree_iter_advance(&(_iter)))
818 
819 #define for_each_btree_key_upto_continue_norestart(_iter, _end, _flags, _k, _ret)\
820 	for (;									\
821 	     (_k) = bch2_btree_iter_peek_upto_type(&(_iter), _end, _flags),	\
822 	     !((_ret) = bkey_err(_k)) && (_k).k;				\
823 	     bch2_btree_iter_advance(&(_iter)))
824 
825 #define for_each_btree_key_norestart(_trans, _iter, _btree_id,		\
826 			   _start, _flags, _k, _ret)			\
827 	for_each_btree_key_upto_norestart(_trans, _iter, _btree_id, _start,\
828 					  SPOS_MAX, _flags, _k, _ret)
829 
830 #define for_each_btree_key_continue_norestart(_iter, _flags, _k, _ret)	\
831 	for_each_btree_key_upto_continue_norestart(_iter, SPOS_MAX, _flags, _k, _ret)
832 
833 /*
834  * This should not be used in a fastpath, without first trying _do in
835  * nonblocking mode - it will cause excessive transaction restarts and
836  * potentially livelocking:
837  */
838 #define drop_locks_do(_trans, _do)					\
839 ({									\
840 	bch2_trans_unlock(_trans);					\
841 	(_do) ?: bch2_trans_relock(_trans);				\
842 })
843 
844 #define allocate_dropping_locks_errcode(_trans, _do)			\
845 ({									\
846 	gfp_t _gfp = GFP_NOWAIT|__GFP_NOWARN;				\
847 	int _ret = _do;							\
848 									\
849 	if (bch2_err_matches(_ret, ENOMEM)) {				\
850 		_gfp = GFP_KERNEL;					\
851 		_ret = drop_locks_do(trans, _do);			\
852 	}								\
853 	_ret;								\
854 })
855 
856 #define allocate_dropping_locks(_trans, _ret, _do)			\
857 ({									\
858 	gfp_t _gfp = GFP_NOWAIT|__GFP_NOWARN;				\
859 	typeof(_do) _p = _do;						\
860 									\
861 	_ret = 0;							\
862 	if (unlikely(!_p)) {						\
863 		_gfp = GFP_KERNEL;					\
864 		_ret = drop_locks_do(trans, ((_p = _do), 0));		\
865 	}								\
866 	_p;								\
867 })
868 
869 void bch2_trans_updates_to_text(struct printbuf *, struct btree_trans *);
870 void bch2_btree_path_to_text(struct printbuf *, struct btree_trans *, btree_path_idx_t);
871 void bch2_trans_paths_to_text(struct printbuf *, struct btree_trans *);
872 void bch2_dump_trans_updates(struct btree_trans *);
873 void bch2_dump_trans_paths_updates(struct btree_trans *);
874 
875 struct btree_trans *__bch2_trans_get(struct bch_fs *, unsigned);
876 void bch2_trans_put(struct btree_trans *);
877 
878 extern const char *bch2_btree_transaction_fns[BCH_TRANSACTIONS_NR];
879 unsigned bch2_trans_get_fn_idx(const char *);
880 
881 #define bch2_trans_get(_c)						\
882 ({									\
883 	static unsigned trans_fn_idx;					\
884 									\
885 	if (unlikely(!trans_fn_idx))					\
886 		trans_fn_idx = bch2_trans_get_fn_idx(__func__);		\
887 	__bch2_trans_get(_c, trans_fn_idx);				\
888 })
889 
890 void bch2_btree_trans_to_text(struct printbuf *, struct btree_trans *);
891 
892 void bch2_fs_btree_iter_exit(struct bch_fs *);
893 void bch2_fs_btree_iter_init_early(struct bch_fs *);
894 int bch2_fs_btree_iter_init(struct bch_fs *);
895 
896 #endif /* _BCACHEFS_BTREE_ITER_H */
897