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
9static inline int __bkey_err(const struct bkey *k)
10{
11 return PTR_ERR_OR_ZERO(ptr: k);
12}
13
14#define bkey_err(_k) __bkey_err((_k).k)
15
16static inline void __btree_path_get(struct btree_path *path, bool intent)
17{
18 path->ref++;
19 path->intent_ref += intent;
20}
21
22static 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
30static 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
36static 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
42static 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(lock: &b->c.lock);
46}
47
48static inline struct btree *btree_node_parent(struct btree_path *path,
49 struct btree *b)
50{
51 return btree_path_node(path, level: b->c.level + 1);
52}
53
54/* Iterate over paths within a transaction: */
55
56void __bch2_btree_trans_sort_paths(struct btree_trans *);
57
58static 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
66static 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
71static 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
82static 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
118static 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
129static 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
144struct 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
161static 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, level: b->c.level);
166}
167
168static 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
186btree_path_idx_t __bch2_btree_path_make_mut(struct btree_trans *, btree_path_idx_t,
187 bool, unsigned long);
188
189static inline btree_path_idx_t __must_check
190bch2_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
201btree_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
205static inline btree_path_idx_t __must_check
206bch2_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(l: new_pos, r: trans->paths[path].pos)
211 ? __bch2_btree_path_set_pos(trans, path, new_pos, intent, ip)
212 : path;
213}
214
215int __must_check bch2_btree_path_traverse_one(struct btree_trans *,
216 btree_path_idx_t,
217 unsigned, unsigned long);
218
219static 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
228btree_path_idx_t bch2_path_get(struct btree_trans *, enum btree_id, struct bpos,
229 unsigned, unsigned, unsigned, unsigned long);
230struct 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 */
236static 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(l: path->pos, r: k.k->p))
241 return k;
242
243 bkey_init(k: u);
244 u->p = path->pos;
245 return (struct bkey_s_c) { u, NULL };
246}
247
248struct bkey_i *bch2_btree_journal_peek_slot(struct btree_trans *,
249 struct btree_iter *, struct bpos);
250
251void bch2_btree_path_level_init(struct btree_trans *, struct btree_path *, struct btree *);
252
253int __bch2_trans_mutex_lock(struct btree_trans *, struct mutex *);
254
255static 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
263void bch2_trans_verify_paths(struct btree_trans *);
264void bch2_assert_pos_locked(struct btree_trans *, enum btree_id,
265 struct bpos, bool);
266#else
267static inline void bch2_trans_verify_paths(struct btree_trans *trans) {}
268static inline void bch2_assert_pos_locked(struct btree_trans *trans, enum btree_id id,
269 struct bpos pos, bool key_cache) {}
270#endif
271
272void bch2_btree_path_fix_key_modified(struct btree_trans *trans,
273 struct btree *, struct bkey_packed *);
274void 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
278int bch2_btree_path_relock_intent(struct btree_trans *, struct btree_path *);
279
280void bch2_path_put(struct btree_trans *, btree_path_idx_t, bool);
281
282int bch2_trans_relock(struct btree_trans *);
283int bch2_trans_relock_notrace(struct btree_trans *);
284void bch2_trans_unlock(struct btree_trans *);
285void bch2_trans_unlock_long(struct btree_trans *);
286bool bch2_trans_locked(struct btree_trans *);
287
288static 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
295void __noreturn bch2_trans_restart_error(struct btree_trans *, u32);
296
297static 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
304void __noreturn bch2_trans_in_restart_error(struct btree_trans *);
305
306static 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
313static 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
324static int btree_trans_restart(struct btree_trans *trans, int err)
325{
326 btree_trans_restart_nounlock(trans, err);
327 return -err;
328}
329
330bool bch2_btree_node_upgrade(struct btree_trans *,
331 struct btree_path *, unsigned);
332
333void __bch2_btree_path_downgrade(struct btree_trans *, struct btree_path *, unsigned);
334
335static 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
344void bch2_trans_downgrade(struct btree_trans *);
345
346void bch2_trans_node_add(struct btree_trans *trans, struct btree_path *, struct btree *);
347void bch2_trans_node_reinit_iter(struct btree_trans *, struct btree *);
348
349int __must_check __bch2_btree_iter_traverse(struct btree_iter *iter);
350int __must_check bch2_btree_iter_traverse(struct btree_iter *);
351
352struct btree *bch2_btree_iter_peek_node(struct btree_iter *);
353struct btree *bch2_btree_iter_peek_node_and_restart(struct btree_iter *);
354struct btree *bch2_btree_iter_next_node(struct btree_iter *);
355
356struct bkey_s_c bch2_btree_iter_peek_upto(struct btree_iter *, struct bpos);
357struct bkey_s_c bch2_btree_iter_next(struct btree_iter *);
358
359static 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
364struct bkey_s_c bch2_btree_iter_peek_prev(struct btree_iter *);
365struct bkey_s_c bch2_btree_iter_prev(struct btree_iter *);
366
367struct bkey_s_c bch2_btree_iter_peek_slot(struct btree_iter *);
368struct bkey_s_c bch2_btree_iter_next_slot(struct btree_iter *);
369struct bkey_s_c bch2_btree_iter_prev_slot(struct btree_iter *);
370
371bool bch2_btree_iter_advance(struct btree_iter *);
372bool bch2_btree_iter_rewind(struct btree_iter *);
373
374static 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
383static 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
398static 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(k: &iter->k);
402}
403
404static 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, new_pos: pos);
411}
412
413void bch2_trans_iter_exit(struct btree_trans *, struct btree_iter *);
414
415static 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: btree_id))
421 flags |= BTREE_ITER_IS_EXTENTS;
422
423 if (!(flags & __BTREE_ITER_ALL_SNAPSHOTS) &&
424 !btree_type_has_snapshot_field(id: btree_id))
425 flags &= ~BTREE_ITER_ALL_SNAPSHOTS;
426
427 if (!(flags & BTREE_ITER_ALL_SNAPSHOTS) &&
428 btree_type_has_snapshots(id: 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
437static inline unsigned bch2_btree_iter_flags(struct btree_trans *trans,
438 unsigned btree_id,
439 unsigned flags)
440{
441 if (!btree_id_cached(c: trans->c, btree: 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
450static 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
475void bch2_trans_iter_init_outlined(struct btree_trans *, struct btree_iter *,
476 enum btree_id, struct bpos, unsigned);
477
478static 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, locks_want: 0, depth: 0,
486 flags: 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
492void bch2_trans_node_iter_init(struct btree_trans *, struct btree_iter *,
493 enum btree_id, struct bpos,
494 unsigned, unsigned, unsigned);
495void bch2_trans_copy_iter(struct btree_iter *, struct btree_iter *);
496
497static inline void set_btree_iter_dontneed(struct btree_iter *iter)
498{
499 struct btree_trans *trans = iter->trans;
500
501 if (!iter->path || trans->restarted)
502 return;
503
504 struct btree_path *path = btree_iter_path(trans, iter);
505 path->preserve = false;
506 if (path->ref == 1)
507 path->should_be_locked = false;
508}
509
510void *__bch2_trans_kmalloc(struct btree_trans *, size_t);
511
512static inline void *bch2_trans_kmalloc(struct btree_trans *trans, size_t size)
513{
514 size = roundup(size, 8);
515
516 if (likely(trans->mem_top + size <= trans->mem_bytes)) {
517 void *p = trans->mem + trans->mem_top;
518
519 trans->mem_top += size;
520 memset(p, 0, size);
521 return p;
522 } else {
523 return __bch2_trans_kmalloc(trans, size);
524 }
525}
526
527static inline void *bch2_trans_kmalloc_nomemzero(struct btree_trans *trans, size_t size)
528{
529 size = round_up(size, 8);
530
531 if (likely(trans->mem_top + size <= trans->mem_bytes)) {
532 void *p = trans->mem + trans->mem_top;
533
534 trans->mem_top += size;
535 return p;
536 } else {
537 return __bch2_trans_kmalloc(trans, size);
538 }
539}
540
541static inline struct bkey_s_c __bch2_bkey_get_iter(struct btree_trans *trans,
542 struct btree_iter *iter,
543 unsigned btree_id, struct bpos pos,
544 unsigned flags, unsigned type)
545{
546 struct bkey_s_c k;
547
548 bch2_trans_iter_init(trans, iter, btree_id, pos, flags);
549 k = bch2_btree_iter_peek_slot(iter);
550
551 if (!bkey_err(k) && type && k.k->type != type)
552 k = bkey_s_c_err(-BCH_ERR_ENOENT_bkey_type_mismatch);
553 if (unlikely(bkey_err(k)))
554 bch2_trans_iter_exit(trans, iter);
555 return k;
556}
557
558static inline struct bkey_s_c bch2_bkey_get_iter(struct btree_trans *trans,
559 struct btree_iter *iter,
560 unsigned btree_id, struct bpos pos,
561 unsigned flags)
562{
563 return __bch2_bkey_get_iter(trans, iter, btree_id, pos, flags, type: 0);
564}
565
566#define bch2_bkey_get_iter_typed(_trans, _iter, _btree_id, _pos, _flags, _type)\
567 bkey_s_c_to_##_type(__bch2_bkey_get_iter(_trans, _iter, \
568 _btree_id, _pos, _flags, KEY_TYPE_##_type))
569
570static inline int __bch2_bkey_get_val_typed(struct btree_trans *trans,
571 unsigned btree_id, struct bpos pos,
572 unsigned flags, unsigned type,
573 unsigned val_size, void *val)
574{
575 struct btree_iter iter;
576 struct bkey_s_c k;
577 int ret;
578
579 k = __bch2_bkey_get_iter(trans, iter: &iter, btree_id, pos, flags, type);
580 ret = bkey_err(k);
581 if (!ret) {
582 unsigned b = min_t(unsigned, bkey_val_bytes(k.k), val_size);
583
584 memcpy(val, k.v, b);
585 if (unlikely(b < sizeof(*val)))
586 memset((void *) val + b, 0, sizeof(*val) - b);
587 bch2_trans_iter_exit(trans, &iter);
588 }
589
590 return ret;
591}
592
593#define bch2_bkey_get_val_typed(_trans, _btree_id, _pos, _flags, _type, _val)\
594 __bch2_bkey_get_val_typed(_trans, _btree_id, _pos, _flags, \
595 KEY_TYPE_##_type, sizeof(*_val), _val)
596
597void bch2_trans_srcu_unlock(struct btree_trans *);
598
599u32 bch2_trans_begin(struct btree_trans *);
600
601/*
602 * XXX
603 * this does not handle transaction restarts from bch2_btree_iter_next_node()
604 * correctly
605 */
606#define __for_each_btree_node(_trans, _iter, _btree_id, _start, \
607 _locks_want, _depth, _flags, _b, _ret) \
608 for (bch2_trans_node_iter_init((_trans), &(_iter), (_btree_id), \
609 _start, _locks_want, _depth, _flags); \
610 (_b) = bch2_btree_iter_peek_node_and_restart(&(_iter)), \
611 !((_ret) = PTR_ERR_OR_ZERO(_b)) && (_b); \
612 (_b) = bch2_btree_iter_next_node(&(_iter)))
613
614#define for_each_btree_node(_trans, _iter, _btree_id, _start, \
615 _flags, _b, _ret) \
616 __for_each_btree_node(_trans, _iter, _btree_id, _start, \
617 0, 0, _flags, _b, _ret)
618
619static inline struct bkey_s_c bch2_btree_iter_peek_prev_type(struct btree_iter *iter,
620 unsigned flags)
621{
622 return flags & BTREE_ITER_SLOTS ? bch2_btree_iter_peek_slot(iter) :
623 bch2_btree_iter_peek_prev(iter);
624}
625
626static inline struct bkey_s_c bch2_btree_iter_peek_type(struct btree_iter *iter,
627 unsigned flags)
628{
629 return flags & BTREE_ITER_SLOTS ? bch2_btree_iter_peek_slot(iter) :
630 bch2_btree_iter_peek(iter);
631}
632
633static inline struct bkey_s_c bch2_btree_iter_peek_upto_type(struct btree_iter *iter,
634 struct bpos end,
635 unsigned flags)
636{
637 if (!(flags & BTREE_ITER_SLOTS))
638 return bch2_btree_iter_peek_upto(iter, end);
639
640 if (bkey_gt(l: iter->pos, r: end))
641 return bkey_s_c_null;
642
643 return bch2_btree_iter_peek_slot(iter);
644}
645
646int __bch2_btree_trans_too_many_iters(struct btree_trans *);
647
648static inline int btree_trans_too_many_iters(struct btree_trans *trans)
649{
650 if (bitmap_weight(src: trans->paths_allocated, nbits: trans->nr_paths) > BTREE_ITER_NORMAL_LIMIT - 8)
651 return __bch2_btree_trans_too_many_iters(trans);
652
653 return 0;
654}
655
656/*
657 * goto instead of loop, so that when used inside for_each_btree_key2()
658 * break/continue work correctly
659 */
660#define lockrestart_do(_trans, _do) \
661({ \
662 __label__ transaction_restart; \
663 u32 _restart_count; \
664 int _ret2; \
665transaction_restart: \
666 _restart_count = bch2_trans_begin(_trans); \
667 _ret2 = (_do); \
668 \
669 if (bch2_err_matches(_ret2, BCH_ERR_transaction_restart)) \
670 goto transaction_restart; \
671 \
672 if (!_ret2) \
673 bch2_trans_verify_not_restarted(_trans, _restart_count);\
674 _ret2; \
675})
676
677/*
678 * nested_lockrestart_do(), nested_commit_do():
679 *
680 * These are like lockrestart_do() and commit_do(), with two differences:
681 *
682 * - We don't call bch2_trans_begin() unless we had a transaction restart
683 * - We return -BCH_ERR_transaction_restart_nested if we succeeded after a
684 * transaction restart
685 */
686#define nested_lockrestart_do(_trans, _do) \
687({ \
688 u32 _restart_count, _orig_restart_count; \
689 int _ret2; \
690 \
691 _restart_count = _orig_restart_count = (_trans)->restart_count; \
692 \
693 while (bch2_err_matches(_ret2 = (_do), BCH_ERR_transaction_restart))\
694 _restart_count = bch2_trans_begin(_trans); \
695 \
696 if (!_ret2) \
697 bch2_trans_verify_not_restarted(_trans, _restart_count);\
698 \
699 _ret2 ?: trans_was_restarted(_trans, _restart_count); \
700})
701
702#define for_each_btree_key_upto(_trans, _iter, _btree_id, \
703 _start, _end, _flags, _k, _do) \
704({ \
705 struct btree_iter _iter; \
706 struct bkey_s_c _k; \
707 int _ret3 = 0; \
708 \
709 bch2_trans_iter_init((_trans), &(_iter), (_btree_id), \
710 (_start), (_flags)); \
711 \
712 do { \
713 _ret3 = lockrestart_do(_trans, ({ \
714 (_k) = bch2_btree_iter_peek_upto_type(&(_iter), \
715 _end, (_flags)); \
716 if (!(_k).k) \
717 break; \
718 \
719 bkey_err(_k) ?: (_do); \
720 })); \
721 } while (!_ret3 && bch2_btree_iter_advance(&(_iter))); \
722 \
723 bch2_trans_iter_exit((_trans), &(_iter)); \
724 _ret3; \
725})
726
727#define for_each_btree_key(_trans, _iter, _btree_id, \
728 _start, _flags, _k, _do) \
729 for_each_btree_key_upto(_trans, _iter, _btree_id, _start, \
730 SPOS_MAX, _flags, _k, _do)
731
732#define for_each_btree_key_reverse(_trans, _iter, _btree_id, \
733 _start, _flags, _k, _do) \
734({ \
735 struct btree_iter _iter; \
736 struct bkey_s_c _k; \
737 int _ret3 = 0; \
738 \
739 bch2_trans_iter_init((_trans), &(_iter), (_btree_id), \
740 (_start), (_flags)); \
741 \
742 do { \
743 _ret3 = lockrestart_do(_trans, ({ \
744 (_k) = bch2_btree_iter_peek_prev_type(&(_iter), \
745 (_flags)); \
746 if (!(_k).k) \
747 break; \
748 \
749 bkey_err(_k) ?: (_do); \
750 })); \
751 } while (!_ret3 && bch2_btree_iter_rewind(&(_iter))); \
752 \
753 bch2_trans_iter_exit((_trans), &(_iter)); \
754 _ret3; \
755})
756
757#define for_each_btree_key_commit(_trans, _iter, _btree_id, \
758 _start, _iter_flags, _k, \
759 _disk_res, _journal_seq, _commit_flags,\
760 _do) \
761 for_each_btree_key(_trans, _iter, _btree_id, _start, _iter_flags, _k,\
762 (_do) ?: bch2_trans_commit(_trans, (_disk_res),\
763 (_journal_seq), (_commit_flags)))
764
765#define for_each_btree_key_reverse_commit(_trans, _iter, _btree_id, \
766 _start, _iter_flags, _k, \
767 _disk_res, _journal_seq, _commit_flags,\
768 _do) \
769 for_each_btree_key_reverse(_trans, _iter, _btree_id, _start, _iter_flags, _k,\
770 (_do) ?: bch2_trans_commit(_trans, (_disk_res),\
771 (_journal_seq), (_commit_flags)))
772
773#define for_each_btree_key_upto_commit(_trans, _iter, _btree_id, \
774 _start, _end, _iter_flags, _k, \
775 _disk_res, _journal_seq, _commit_flags,\
776 _do) \
777 for_each_btree_key_upto(_trans, _iter, _btree_id, _start, _end, _iter_flags, _k,\
778 (_do) ?: bch2_trans_commit(_trans, (_disk_res),\
779 (_journal_seq), (_commit_flags)))
780
781struct bkey_s_c bch2_btree_iter_peek_and_restart_outlined(struct btree_iter *);
782
783static inline struct bkey_s_c
784__bch2_btree_iter_peek_and_restart(struct btree_trans *trans,
785 struct btree_iter *iter, unsigned flags)
786{
787 struct bkey_s_c k;
788
789 while (btree_trans_too_many_iters(trans) ||
790 (k = bch2_btree_iter_peek_type(iter, flags),
791 bch2_err_matches(bkey_err(k), BCH_ERR_transaction_restart)))
792 bch2_trans_begin(trans);
793
794 return k;
795}
796
797#define for_each_btree_key_old(_trans, _iter, _btree_id, \
798 _start, _flags, _k, _ret) \
799 for (bch2_trans_iter_init((_trans), &(_iter), (_btree_id), \
800 (_start), (_flags)); \
801 (_k) = __bch2_btree_iter_peek_and_restart((_trans), &(_iter), _flags),\
802 !((_ret) = bkey_err(_k)) && (_k).k; \
803 bch2_btree_iter_advance(&(_iter)))
804
805#define for_each_btree_key_upto_norestart(_trans, _iter, _btree_id, \
806 _start, _end, _flags, _k, _ret) \
807 for (bch2_trans_iter_init((_trans), &(_iter), (_btree_id), \
808 (_start), (_flags)); \
809 (_k) = bch2_btree_iter_peek_upto_type(&(_iter), _end, _flags),\
810 !((_ret) = bkey_err(_k)) && (_k).k; \
811 bch2_btree_iter_advance(&(_iter)))
812
813#define for_each_btree_key_upto_continue_norestart(_iter, _end, _flags, _k, _ret)\
814 for (; \
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_norestart(_trans, _iter, _btree_id, \
820 _start, _flags, _k, _ret) \
821 for_each_btree_key_upto_norestart(_trans, _iter, _btree_id, _start,\
822 SPOS_MAX, _flags, _k, _ret)
823
824#define for_each_btree_key_continue_norestart(_iter, _flags, _k, _ret) \
825 for_each_btree_key_upto_continue_norestart(_iter, SPOS_MAX, _flags, _k, _ret)
826
827/*
828 * This should not be used in a fastpath, without first trying _do in
829 * nonblocking mode - it will cause excessive transaction restarts and
830 * potentially livelocking:
831 */
832#define drop_locks_do(_trans, _do) \
833({ \
834 bch2_trans_unlock(_trans); \
835 _do ?: bch2_trans_relock(_trans); \
836})
837
838#define allocate_dropping_locks_errcode(_trans, _do) \
839({ \
840 gfp_t _gfp = GFP_NOWAIT|__GFP_NOWARN; \
841 int _ret = _do; \
842 \
843 if (bch2_err_matches(_ret, ENOMEM)) { \
844 _gfp = GFP_KERNEL; \
845 _ret = drop_locks_do(trans, _do); \
846 } \
847 _ret; \
848})
849
850#define allocate_dropping_locks(_trans, _ret, _do) \
851({ \
852 gfp_t _gfp = GFP_NOWAIT|__GFP_NOWARN; \
853 typeof(_do) _p = _do; \
854 \
855 _ret = 0; \
856 if (unlikely(!_p)) { \
857 _gfp = GFP_KERNEL; \
858 _ret = drop_locks_do(trans, ((_p = _do), 0)); \
859 } \
860 _p; \
861})
862
863void bch2_trans_updates_to_text(struct printbuf *, struct btree_trans *);
864void bch2_trans_paths_to_text(struct printbuf *, struct btree_trans *);
865void bch2_dump_trans_updates(struct btree_trans *);
866void bch2_dump_trans_paths_updates(struct btree_trans *);
867
868struct btree_trans *__bch2_trans_get(struct bch_fs *, unsigned);
869void bch2_trans_put(struct btree_trans *);
870
871extern const char *bch2_btree_transaction_fns[BCH_TRANSACTIONS_NR];
872unsigned bch2_trans_get_fn_idx(const char *);
873
874#define bch2_trans_get(_c) \
875({ \
876 static unsigned trans_fn_idx; \
877 \
878 if (unlikely(!trans_fn_idx)) \
879 trans_fn_idx = bch2_trans_get_fn_idx(__func__); \
880 __bch2_trans_get(_c, trans_fn_idx); \
881})
882
883void bch2_btree_trans_to_text(struct printbuf *, struct btree_trans *);
884
885void bch2_fs_btree_iter_exit(struct bch_fs *);
886void bch2_fs_btree_iter_init_early(struct bch_fs *);
887int bch2_fs_btree_iter_init(struct bch_fs *);
888
889#endif /* _BCACHEFS_BTREE_ITER_H */
890

source code of linux/fs/bcachefs/btree_iter.h