1 // SPDX-License-Identifier: GPL-2.0
4 #include "btree_update.h"
5 #include "btree_update_interior.h"
8 #include "btree_iter.h"
9 #include "btree_key_cache.h"
10 #include "btree_locking.h"
11 #include "btree_write_buffer.h"
16 #include "extent_update.h"
18 #include "journal_reclaim.h"
21 #include "subvolume.h"
25 #include <linux/prefetch.h>
26 #include <linux/sort.h>
29 * bch2_btree_path_peek_slot() for a cached iterator might return a key in a
32 struct bkey_s_c bch2_btree_path_peek_slot_exact(struct btree_path *path, struct bkey *u)
34 struct bkey_s_c k = bch2_btree_path_peek_slot(path, u);
36 if (k.k && bpos_eq(path->pos, k.k->p))
41 return (struct bkey_s_c) { u, NULL };
44 static void verify_update_old_key(struct btree_trans *trans, struct btree_insert_entry *i)
46 #ifdef CONFIG_BCACHEFS_DEBUG
47 struct bch_fs *c = trans->c;
49 struct bkey_s_c k = bch2_btree_path_peek_slot_exact(i->path, &u);
51 if (unlikely(trans->journal_replay_not_finished)) {
53 bch2_journal_keys_peek_slot(c, i->btree_id, i->level, i->k->k.p);
56 k = bkey_i_to_s_c(j_k);
60 u.needs_whiteout = i->old_k.needs_whiteout;
62 BUG_ON(memcmp(&i->old_k, &u, sizeof(struct bkey)));
63 BUG_ON(i->old_v != k.v);
67 static int __must_check
68 bch2_trans_update_by_path(struct btree_trans *, struct btree_path *,
69 struct bkey_i *, enum btree_update_flags);
71 static inline int btree_insert_entry_cmp(const struct btree_insert_entry *l,
72 const struct btree_insert_entry *r)
74 return cmp_int(l->btree_id, r->btree_id) ?:
75 cmp_int(l->cached, r->cached) ?:
76 -cmp_int(l->level, r->level) ?:
77 bpos_cmp(l->k->k.p, r->k->k.p);
80 static inline struct btree_path_level *insert_l(struct btree_insert_entry *i)
82 return i->path->l + i->level;
85 static inline bool same_leaf_as_prev(struct btree_trans *trans,
86 struct btree_insert_entry *i)
88 return i != trans->updates &&
89 insert_l(&i[0])->b == insert_l(&i[-1])->b;
92 static inline bool same_leaf_as_next(struct btree_trans *trans,
93 struct btree_insert_entry *i)
95 return i + 1 < trans->updates + trans->nr_updates &&
96 insert_l(&i[0])->b == insert_l(&i[1])->b;
99 inline void bch2_btree_node_prep_for_write(struct btree_trans *trans,
100 struct btree_path *path,
103 struct bch_fs *c = trans->c;
105 if (unlikely(btree_node_just_written(b)) &&
106 bch2_btree_post_write_cleanup(c, b))
107 bch2_trans_node_reinit_iter(trans, b);
110 * If the last bset has been written, or if it's gotten too big - start
111 * a new bset to insert into:
113 if (want_new_bset(c, b))
114 bch2_btree_init_next(trans, b);
117 /* Inserting into a given leaf node (last stage of insert): */
119 /* Handle overwrites and do insert, for non extents: */
120 bool bch2_btree_bset_insert_key(struct btree_trans *trans,
121 struct btree_path *path,
123 struct btree_node_iter *node_iter,
124 struct bkey_i *insert)
126 struct bkey_packed *k;
127 unsigned clobber_u64s = 0, new_u64s = 0;
129 EBUG_ON(btree_node_just_written(b));
130 EBUG_ON(bset_written(b, btree_bset_last(b)));
131 EBUG_ON(bkey_deleted(&insert->k) && bkey_val_u64s(&insert->k));
132 EBUG_ON(bpos_lt(insert->k.p, b->data->min_key));
133 EBUG_ON(bpos_gt(insert->k.p, b->data->max_key));
134 EBUG_ON(insert->k.u64s >
135 bch_btree_keys_u64s_remaining(trans->c, b));
137 k = bch2_btree_node_iter_peek_all(node_iter, b);
138 if (k && bkey_cmp_left_packed(b, k, &insert->k.p))
141 /* @k is the key being overwritten/deleted, if any: */
142 EBUG_ON(k && bkey_deleted(k));
144 /* Deleting, but not found? nothing to do: */
145 if (bkey_deleted(&insert->k) && !k)
148 if (bkey_deleted(&insert->k)) {
150 btree_account_key_drop(b, k);
151 k->type = KEY_TYPE_deleted;
153 if (k->needs_whiteout)
154 push_whiteout(trans->c, b, insert->k.p);
155 k->needs_whiteout = false;
157 if (k >= btree_bset_last(b)->start) {
158 clobber_u64s = k->u64s;
159 bch2_bset_delete(b, k, clobber_u64s);
162 bch2_btree_path_fix_key_modified(trans, b, k);
170 btree_account_key_drop(b, k);
171 k->type = KEY_TYPE_deleted;
173 insert->k.needs_whiteout = k->needs_whiteout;
174 k->needs_whiteout = false;
176 if (k >= btree_bset_last(b)->start) {
177 clobber_u64s = k->u64s;
180 bch2_btree_path_fix_key_modified(trans, b, k);
184 k = bch2_btree_node_iter_bset_pos(node_iter, b, bset_tree_last(b));
186 bch2_bset_insert(b, node_iter, k, insert, clobber_u64s);
189 if (clobber_u64s != new_u64s)
190 bch2_btree_node_iter_fix(trans, path, b, node_iter, k,
191 clobber_u64s, new_u64s);
195 static int __btree_node_flush(struct journal *j, struct journal_entry_pin *pin,
198 struct bch_fs *c = container_of(j, struct bch_fs, journal);
199 struct btree_write *w = container_of(pin, struct btree_write, journal);
200 struct btree *b = container_of(w, struct btree, writes[i]);
201 struct btree_trans trans;
202 unsigned long old, new, v;
203 unsigned idx = w - b->writes;
205 bch2_trans_init(&trans, c, 0, 0);
207 btree_node_lock_nopath_nofail(&trans, &b->c, SIX_LOCK_read);
208 v = READ_ONCE(b->flags);
213 if (!(old & (1 << BTREE_NODE_dirty)) ||
214 !!(old & (1 << BTREE_NODE_write_idx)) != idx ||
215 w->journal.seq != seq)
218 new &= ~BTREE_WRITE_TYPE_MASK;
219 new |= BTREE_WRITE_journal_reclaim;
220 new |= 1 << BTREE_NODE_need_write;
221 } while ((v = cmpxchg(&b->flags, old, new)) != old);
223 btree_node_write_if_need(c, b, SIX_LOCK_read);
224 six_unlock_read(&b->c.lock);
226 bch2_trans_exit(&trans);
230 int bch2_btree_node_flush0(struct journal *j, struct journal_entry_pin *pin, u64 seq)
232 return __btree_node_flush(j, pin, 0, seq);
235 int bch2_btree_node_flush1(struct journal *j, struct journal_entry_pin *pin, u64 seq)
237 return __btree_node_flush(j, pin, 1, seq);
240 inline void bch2_btree_add_journal_pin(struct bch_fs *c,
241 struct btree *b, u64 seq)
243 struct btree_write *w = btree_current_write(b);
245 bch2_journal_pin_add(&c->journal, seq, &w->journal,
246 btree_node_write_idx(b) == 0
247 ? bch2_btree_node_flush0
248 : bch2_btree_node_flush1);
252 * btree_insert_key - insert a key one key into a leaf node
254 inline void bch2_btree_insert_key_leaf(struct btree_trans *trans,
255 struct btree_path *path,
256 struct bkey_i *insert,
259 struct bch_fs *c = trans->c;
260 struct btree *b = path_l(path)->b;
261 struct bset_tree *t = bset_tree_last(b);
262 struct bset *i = bset(b, t);
263 int old_u64s = bset_u64s(t);
264 int old_live_u64s = b->nr.live_u64s;
265 int live_u64s_added, u64s_added;
267 if (unlikely(!bch2_btree_bset_insert_key(trans, path, b,
268 &path_l(path)->iter, insert)))
271 i->journal_seq = cpu_to_le64(max(journal_seq, le64_to_cpu(i->journal_seq)));
273 bch2_btree_add_journal_pin(c, b, journal_seq);
275 if (unlikely(!btree_node_dirty(b)))
276 set_btree_node_dirty_acct(c, b);
278 live_u64s_added = (int) b->nr.live_u64s - old_live_u64s;
279 u64s_added = (int) bset_u64s(t) - old_u64s;
281 if (b->sib_u64s[0] != U16_MAX && live_u64s_added < 0)
282 b->sib_u64s[0] = max(0, (int) b->sib_u64s[0] + live_u64s_added);
283 if (b->sib_u64s[1] != U16_MAX && live_u64s_added < 0)
284 b->sib_u64s[1] = max(0, (int) b->sib_u64s[1] + live_u64s_added);
286 if (u64s_added > live_u64s_added &&
287 bch2_maybe_compact_whiteouts(c, b))
288 bch2_trans_node_reinit_iter(trans, b);
291 static void btree_insert_key_leaf(struct btree_trans *trans,
292 struct btree_insert_entry *insert)
294 bch2_btree_insert_key_leaf(trans, insert->path, insert->k, trans->journal_res.seq);
297 /* Cached btree updates: */
299 /* Normal update interface: */
301 static inline void btree_insert_entry_checks(struct btree_trans *trans,
302 struct btree_insert_entry *i)
304 BUG_ON(!bpos_eq(i->k->k.p, i->path->pos));
305 BUG_ON(i->cached != i->path->cached);
306 BUG_ON(i->level != i->path->level);
307 BUG_ON(i->btree_id != i->path->btree_id);
309 !(i->flags & BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE) &&
310 test_bit(JOURNAL_REPLAY_DONE, &trans->c->journal.flags) &&
311 i->k->k.p.snapshot &&
312 bch2_snapshot_internal_node(trans->c, i->k->k.p.snapshot));
316 bch2_trans_journal_preres_get_cold(struct btree_trans *trans, unsigned flags,
317 unsigned long trace_ip)
319 return drop_locks_do(trans,
320 bch2_journal_preres_get(&trans->c->journal,
321 &trans->journal_preres,
322 trans->journal_preres_u64s,
323 (flags & JOURNAL_WATERMARK_MASK)));
326 static __always_inline int bch2_trans_journal_res_get(struct btree_trans *trans,
329 return bch2_journal_res_get(&trans->c->journal, &trans->journal_res,
330 trans->journal_u64s, flags);
333 #define JSET_ENTRY_LOG_U64s 4
335 static noinline void journal_transaction_name(struct btree_trans *trans)
337 struct bch_fs *c = trans->c;
338 struct journal *j = &c->journal;
339 struct jset_entry *entry =
340 bch2_journal_add_entry(j, &trans->journal_res,
341 BCH_JSET_ENTRY_log, 0, 0,
342 JSET_ENTRY_LOG_U64s);
343 struct jset_entry_log *l =
344 container_of(entry, struct jset_entry_log, entry);
346 strncpy(l->d, trans->fn, JSET_ENTRY_LOG_U64s * sizeof(u64));
349 static inline int btree_key_can_insert(struct btree_trans *trans,
350 struct btree *b, unsigned u64s)
352 struct bch_fs *c = trans->c;
354 if (!bch2_btree_node_insert_fits(c, b, u64s))
355 return -BCH_ERR_btree_insert_btree_node_full;
360 static int btree_key_can_insert_cached(struct btree_trans *trans, unsigned flags,
361 struct btree_path *path, unsigned u64s)
363 struct bch_fs *c = trans->c;
364 struct bkey_cached *ck = (void *) path->l[0].b;
365 struct btree_insert_entry *i;
367 struct bkey_i *new_k;
369 EBUG_ON(path->level);
371 if (!test_bit(BKEY_CACHED_DIRTY, &ck->flags) &&
372 bch2_btree_key_cache_must_wait(c) &&
373 !(flags & BTREE_INSERT_JOURNAL_RECLAIM))
374 return -BCH_ERR_btree_insert_need_journal_reclaim;
377 * bch2_varint_decode can read past the end of the buffer by at most 7
378 * bytes (it won't be used):
382 if (u64s <= ck->u64s)
385 new_u64s = roundup_pow_of_two(u64s);
386 new_k = krealloc(ck->k, new_u64s * sizeof(u64), GFP_NOFS);
388 bch_err(c, "error allocating memory for key cache key, btree %s u64s %u",
389 bch2_btree_ids[path->btree_id], new_u64s);
390 return -BCH_ERR_ENOMEM_btree_key_cache_insert;
393 trans_for_each_update(trans, i)
394 if (i->old_v == &ck->k->v)
395 i->old_v = &new_k->v;
404 static int run_one_mem_trigger(struct btree_trans *trans,
405 struct btree_insert_entry *i,
408 struct bkey_s_c old = { &i->old_k, i->old_v };
409 struct bkey_i *new = i->k;
412 verify_update_old_key(trans, i);
414 if (unlikely(flags & BTREE_TRIGGER_NORUN))
417 if (!btree_node_type_needs_gc(i->btree_id))
420 if (bch2_bkey_ops[old.k->type].atomic_trigger ==
421 bch2_bkey_ops[i->k->k.type].atomic_trigger &&
422 ((1U << old.k->type) & BTREE_TRIGGER_WANTS_OLD_AND_NEW)) {
423 ret = bch2_mark_key(trans, i->btree_id, i->level,
424 old, bkey_i_to_s_c(new),
425 BTREE_TRIGGER_INSERT|BTREE_TRIGGER_OVERWRITE|flags);
427 struct bkey _deleted = KEY(0, 0, 0);
428 struct bkey_s_c deleted = (struct bkey_s_c) { &_deleted, NULL };
430 _deleted.p = i->path->pos;
432 ret = bch2_mark_key(trans, i->btree_id, i->level,
433 deleted, bkey_i_to_s_c(new),
434 BTREE_TRIGGER_INSERT|flags) ?:
435 bch2_mark_key(trans, i->btree_id, i->level,
437 BTREE_TRIGGER_OVERWRITE|flags);
443 static int run_one_trans_trigger(struct btree_trans *trans, struct btree_insert_entry *i,
447 * Transactional triggers create new btree_insert_entries, so we can't
448 * pass them a pointer to a btree_insert_entry, that memory is going to
451 struct bkey old_k = i->old_k;
452 struct bkey_s_c old = { &old_k, i->old_v };
454 verify_update_old_key(trans, i);
456 if ((i->flags & BTREE_TRIGGER_NORUN) ||
457 !(BTREE_NODE_TYPE_HAS_TRANS_TRIGGERS & (1U << i->bkey_type)))
460 if (!i->insert_trigger_run &&
461 !i->overwrite_trigger_run &&
462 bch2_bkey_ops[old.k->type].trans_trigger ==
463 bch2_bkey_ops[i->k->k.type].trans_trigger &&
464 ((1U << old.k->type) & BTREE_TRIGGER_WANTS_OLD_AND_NEW)) {
465 i->overwrite_trigger_run = true;
466 i->insert_trigger_run = true;
467 return bch2_trans_mark_key(trans, i->btree_id, i->level, old, i->k,
468 BTREE_TRIGGER_INSERT|
469 BTREE_TRIGGER_OVERWRITE|
471 } else if (overwrite && !i->overwrite_trigger_run) {
472 i->overwrite_trigger_run = true;
473 return bch2_trans_mark_old(trans, i->btree_id, i->level, old, i->flags) ?: 1;
474 } else if (!overwrite && !i->insert_trigger_run) {
475 i->insert_trigger_run = true;
476 return bch2_trans_mark_new(trans, i->btree_id, i->level, i->k, i->flags) ?: 1;
482 static int run_btree_triggers(struct btree_trans *trans, enum btree_id btree_id,
483 struct btree_insert_entry *btree_id_start)
485 struct btree_insert_entry *i;
486 bool trans_trigger_run;
489 for (overwrite = 1; overwrite >= 0; --overwrite) {
492 * Running triggers will append more updates to the list of updates as
496 trans_trigger_run = false;
498 for (i = btree_id_start;
499 i < trans->updates + trans->nr_updates && i->btree_id <= btree_id;
501 if (i->btree_id != btree_id)
504 ret = run_one_trans_trigger(trans, i, overwrite);
508 trans_trigger_run = true;
510 } while (trans_trigger_run);
516 static int bch2_trans_commit_run_triggers(struct btree_trans *trans)
518 struct btree_insert_entry *i = NULL, *btree_id_start = trans->updates;
519 unsigned btree_id = 0;
524 * For a given btree, this algorithm runs insert triggers before
525 * overwrite triggers: this is so that when extents are being moved
526 * (e.g. by FALLOCATE_FL_INSERT_RANGE), we don't drop references before
529 for (btree_id = 0; btree_id < BTREE_ID_NR; btree_id++) {
530 if (btree_id == BTREE_ID_alloc)
533 while (btree_id_start < trans->updates + trans->nr_updates &&
534 btree_id_start->btree_id < btree_id)
537 ret = run_btree_triggers(trans, btree_id, btree_id_start);
542 trans_for_each_update(trans, i) {
543 if (i->btree_id > BTREE_ID_alloc)
545 if (i->btree_id == BTREE_ID_alloc) {
546 ret = run_btree_triggers(trans, BTREE_ID_alloc, i);
553 #ifdef CONFIG_BCACHEFS_DEBUG
554 trans_for_each_update(trans, i)
555 BUG_ON(!(i->flags & BTREE_TRIGGER_NORUN) &&
556 (BTREE_NODE_TYPE_HAS_TRANS_TRIGGERS & (1U << i->bkey_type)) &&
557 (!i->insert_trigger_run || !i->overwrite_trigger_run));
562 static noinline int bch2_trans_commit_run_gc_triggers(struct btree_trans *trans)
564 struct bch_fs *c = trans->c;
565 struct btree_insert_entry *i;
568 trans_for_each_update(trans, i) {
570 * XXX: synchronization of cached update triggers with gc
571 * XXX: synchronization of interior node updates with gc
573 BUG_ON(i->cached || i->level);
575 if (gc_visited(c, gc_pos_btree_node(insert_l(i)->b))) {
576 ret = run_one_mem_trigger(trans, i, i->flags|BTREE_TRIGGER_GC);
586 bch2_trans_commit_write_locked(struct btree_trans *trans, unsigned flags,
587 struct btree_insert_entry **stopped_at,
588 unsigned long trace_ip)
590 struct bch_fs *c = trans->c;
591 struct btree_insert_entry *i;
592 struct btree_write_buffered_key *wb;
593 struct btree_trans_commit_hook *h;
595 bool marking = false;
599 trace_and_count(c, trans_restart_fault_inject, trans, trace_ip);
600 return btree_trans_restart_nounlock(trans, BCH_ERR_transaction_restart_fault_inject);
604 * Check if the insert will fit in the leaf node with the write lock
605 * held, otherwise another thread could write the node changing the
606 * amount of space available:
609 prefetch(&trans->c->journal.flags);
611 trans_for_each_update(trans, i) {
612 /* Multiple inserts might go to same leaf: */
613 if (!same_leaf_as_prev(trans, i))
616 u64s += i->k->k.u64s;
618 ? btree_key_can_insert(trans, insert_l(i)->b, u64s)
619 : btree_key_can_insert_cached(trans, flags, i->path, u64s);
625 if (btree_node_type_needs_gc(i->bkey_type))
629 if (trans->nr_wb_updates &&
630 trans->nr_wb_updates + c->btree_write_buffer.state.nr > c->btree_write_buffer.size)
631 return -BCH_ERR_btree_insert_need_flush_buffer;
634 * Don't get journal reservation until after we know insert will
637 if (likely(!(flags & BTREE_INSERT_JOURNAL_REPLAY))) {
638 ret = bch2_trans_journal_res_get(trans,
639 (flags & JOURNAL_WATERMARK_MASK)|
640 JOURNAL_RES_GET_NONBLOCK);
644 if (unlikely(trans->journal_transaction_names))
645 journal_transaction_name(trans);
647 trans->journal_res.seq = c->journal.replay_journal_seq;
651 * Not allowed to fail after we've gotten our journal reservation - we
655 if (IS_ENABLED(CONFIG_BCACHEFS_DEBUG) &&
656 !(flags & BTREE_INSERT_JOURNAL_REPLAY)) {
657 if (bch2_journal_seq_verify)
658 trans_for_each_update(trans, i)
659 i->k->k.version.lo = trans->journal_res.seq;
660 else if (bch2_inject_invalid_keys)
661 trans_for_each_update(trans, i)
662 i->k->k.version = MAX_VERSION;
665 if (trans->fs_usage_deltas &&
666 bch2_trans_fs_usage_apply(trans, trans->fs_usage_deltas))
667 return -BCH_ERR_btree_insert_need_mark_replicas;
669 if (trans->nr_wb_updates) {
670 EBUG_ON(flags & BTREE_INSERT_JOURNAL_REPLAY);
672 ret = bch2_btree_insert_keys_write_buffer(trans);
674 goto revert_fs_usage;
679 ret = h->fn(trans, h);
681 goto revert_fs_usage;
685 trans_for_each_update(trans, i)
686 if (BTREE_NODE_TYPE_HAS_MEM_TRIGGERS & (1U << i->bkey_type)) {
687 ret = run_one_mem_trigger(trans, i, i->flags);
692 if (unlikely(c->gc_pos.phase)) {
693 ret = bch2_trans_commit_run_gc_triggers(trans);
698 if (unlikely(trans->extra_journal_entries.nr)) {
699 memcpy_u64s_small(journal_res_entry(&c->journal, &trans->journal_res),
700 trans->extra_journal_entries.data,
701 trans->extra_journal_entries.nr);
703 trans->journal_res.offset += trans->extra_journal_entries.nr;
704 trans->journal_res.u64s -= trans->extra_journal_entries.nr;
707 if (likely(!(flags & BTREE_INSERT_JOURNAL_REPLAY))) {
708 struct journal *j = &c->journal;
709 struct jset_entry *entry;
711 trans_for_each_update(trans, i) {
712 if (i->key_cache_already_flushed)
715 if (i->flags & BTREE_UPDATE_NOJOURNAL)
718 verify_update_old_key(trans, i);
720 if (trans->journal_transaction_names) {
721 entry = bch2_journal_add_entry(j, &trans->journal_res,
722 BCH_JSET_ENTRY_overwrite,
723 i->btree_id, i->level,
725 bkey_reassemble(&entry->start[0],
726 (struct bkey_s_c) { &i->old_k, i->old_v });
729 entry = bch2_journal_add_entry(j, &trans->journal_res,
730 BCH_JSET_ENTRY_btree_keys,
731 i->btree_id, i->level,
733 bkey_copy(&entry->start[0], i->k);
736 trans_for_each_wb_update(trans, wb) {
737 entry = bch2_journal_add_entry(j, &trans->journal_res,
738 BCH_JSET_ENTRY_btree_keys,
741 bkey_copy(&entry->start[0], &wb->k);
744 if (trans->journal_seq)
745 *trans->journal_seq = trans->journal_res.seq;
748 trans_for_each_update(trans, i) {
749 i->k->k.needs_whiteout = false;
752 btree_insert_key_leaf(trans, i);
753 else if (!i->key_cache_already_flushed)
754 bch2_btree_insert_key_cached(trans, flags, i);
756 bch2_btree_key_cache_drop(trans, i->path);
757 btree_path_set_dirty(i->path, BTREE_ITER_NEED_TRAVERSE);
765 if (trans->fs_usage_deltas)
766 bch2_trans_fs_usage_revert(trans, trans->fs_usage_deltas);
770 static noinline int trans_lock_write_fail(struct btree_trans *trans, struct btree_insert_entry *i)
772 while (--i >= trans->updates) {
773 if (same_leaf_as_prev(trans, i))
776 bch2_btree_node_unlock_write(trans, i->path, insert_l(i)->b);
779 trace_and_count(trans->c, trans_restart_would_deadlock_write, trans);
780 return btree_trans_restart(trans, BCH_ERR_transaction_restart_would_deadlock_write);
783 static inline int trans_lock_write(struct btree_trans *trans)
785 struct btree_insert_entry *i;
787 trans_for_each_update(trans, i) {
788 if (same_leaf_as_prev(trans, i))
791 if (bch2_btree_node_lock_write(trans, i->path, &insert_l(i)->b->c))
792 return trans_lock_write_fail(trans, i);
795 bch2_btree_node_prep_for_write(trans, i->path, insert_l(i)->b);
801 static noinline void bch2_drop_overwrites_from_journal(struct btree_trans *trans)
803 struct btree_insert_entry *i;
804 struct btree_write_buffered_key *wb;
806 trans_for_each_update(trans, i)
807 bch2_journal_key_overwritten(trans->c, i->btree_id, i->level, i->k->k.p);
809 trans_for_each_wb_update(trans, wb)
810 bch2_journal_key_overwritten(trans->c, wb->btree, 0, wb->k.k.p);
813 #ifdef CONFIG_BCACHEFS_DEBUG
814 static noinline int bch2_trans_commit_bkey_invalid(struct btree_trans *trans, unsigned flags,
815 struct btree_insert_entry *i,
816 struct printbuf *err)
818 struct bch_fs *c = trans->c;
819 int rw = (flags & BTREE_INSERT_JOURNAL_REPLAY) ? READ : WRITE;
822 prt_printf(err, "invalid bkey on insert from %s -> %ps",
823 trans->fn, (void *) i->ip_allocated);
825 printbuf_indent_add(err, 2);
827 bch2_bkey_val_to_text(err, c, bkey_i_to_s_c(i->k));
830 bch2_bkey_invalid(c, bkey_i_to_s_c(i->k),
831 i->bkey_type, rw, err);
832 bch2_print_string_as_lines(KERN_ERR, err->buf);
834 bch2_inconsistent_error(c);
835 bch2_dump_trans_updates(trans);
843 * Get journal reservation, take write locks, and attempt to do btree update(s):
845 static inline int do_bch2_trans_commit(struct btree_trans *trans, unsigned flags,
846 struct btree_insert_entry **stopped_at,
847 unsigned long trace_ip)
849 struct bch_fs *c = trans->c;
850 struct btree_insert_entry *i;
851 int ret, u64s_delta = 0;
853 #ifdef CONFIG_BCACHEFS_DEBUG
854 struct printbuf buf = PRINTBUF;
856 trans_for_each_update(trans, i) {
857 int rw = (flags & BTREE_INSERT_JOURNAL_REPLAY) ? READ : WRITE;
859 if (unlikely(bch2_bkey_invalid(c, bkey_i_to_s_c(i->k),
860 i->bkey_type, rw, &buf)))
861 return bch2_trans_commit_bkey_invalid(trans, flags, i, &buf);
862 btree_insert_entry_checks(trans, i);
867 trans_for_each_update(trans, i) {
871 u64s_delta += !bkey_deleted(&i->k->k) ? i->k->k.u64s : 0;
872 u64s_delta -= i->old_btree_u64s;
874 if (!same_leaf_as_next(trans, i)) {
875 if (u64s_delta <= 0) {
876 ret = bch2_foreground_maybe_merge(trans, i->path,
886 ret = bch2_journal_preres_get(&c->journal,
887 &trans->journal_preres, trans->journal_preres_u64s,
888 (flags & JOURNAL_WATERMARK_MASK)|JOURNAL_RES_GET_NONBLOCK);
889 if (unlikely(ret == -BCH_ERR_journal_preres_get_blocked))
890 ret = bch2_trans_journal_preres_get_cold(trans, flags, trace_ip);
894 ret = trans_lock_write(trans);
898 ret = bch2_trans_commit_write_locked(trans, flags, stopped_at, trace_ip);
900 if (!ret && unlikely(trans->journal_replay_not_finished))
901 bch2_drop_overwrites_from_journal(trans);
903 trans_for_each_update(trans, i)
904 if (!same_leaf_as_prev(trans, i))
905 bch2_btree_node_unlock_write_inlined(trans, i->path,
908 if (!ret && trans->journal_pin)
909 bch2_journal_pin_add(&c->journal, trans->journal_res.seq,
910 trans->journal_pin, NULL);
913 * Drop journal reservation after dropping write locks, since dropping
914 * the journal reservation may kick off a journal write:
916 bch2_journal_res_put(&c->journal, &trans->journal_res);
921 bch2_trans_downgrade(trans);
926 static int journal_reclaim_wait_done(struct bch_fs *c)
928 int ret = bch2_journal_error(&c->journal) ?:
929 !bch2_btree_key_cache_must_wait(c);
932 journal_reclaim_kick(&c->journal);
937 int bch2_trans_commit_error(struct btree_trans *trans, unsigned flags,
938 struct btree_insert_entry *i,
939 int ret, unsigned long trace_ip)
941 struct bch_fs *c = trans->c;
944 case -BCH_ERR_btree_insert_btree_node_full:
945 ret = bch2_btree_split_leaf(trans, i->path, flags);
946 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
947 trace_and_count(c, trans_restart_btree_node_split, trans, trace_ip, i->path);
949 case -BCH_ERR_btree_insert_need_mark_replicas:
950 ret = drop_locks_do(trans,
951 bch2_replicas_delta_list_mark(c, trans->fs_usage_deltas));
953 case -BCH_ERR_journal_res_get_blocked:
954 if ((flags & BTREE_INSERT_JOURNAL_RECLAIM) &&
955 !(flags & JOURNAL_WATERMARK_reserved)) {
956 ret = -BCH_ERR_journal_reclaim_would_deadlock;
960 ret = drop_locks_do(trans,
961 bch2_trans_journal_res_get(trans,
962 (flags & JOURNAL_WATERMARK_MASK)|
963 JOURNAL_RES_GET_CHECK));
965 case -BCH_ERR_btree_insert_need_journal_reclaim:
966 bch2_trans_unlock(trans);
968 trace_and_count(c, trans_blocked_journal_reclaim, trans, trace_ip);
970 wait_event_freezable(c->journal.reclaim_wait,
971 (ret = journal_reclaim_wait_done(c)));
975 ret = bch2_trans_relock(trans);
977 case -BCH_ERR_btree_insert_need_flush_buffer: {
978 struct btree_write_buffer *wb = &c->btree_write_buffer;
982 if (wb->state.nr > wb->size * 3 / 4) {
983 bch2_trans_unlock(trans);
984 mutex_lock(&wb->flush_lock);
986 if (wb->state.nr > wb->size * 3 / 4) {
987 bch2_trans_begin(trans);
988 ret = __bch2_btree_write_buffer_flush(trans,
989 flags|BTREE_INSERT_NOCHECK_RW, true);
991 trace_and_count(c, trans_restart_write_buffer_flush, trans, _THIS_IP_);
992 ret = btree_trans_restart(trans, BCH_ERR_transaction_restart_write_buffer_flush);
995 mutex_unlock(&wb->flush_lock);
996 ret = bch2_trans_relock(trans);
1006 BUG_ON(bch2_err_matches(ret, BCH_ERR_transaction_restart) != !!trans->restarted);
1008 bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOSPC) &&
1009 !(flags & BTREE_INSERT_NOWAIT) &&
1010 (flags & BTREE_INSERT_NOFAIL), c,
1011 "%s: incorrectly got %s\n", __func__, bch2_err_str(ret));
1017 bch2_trans_commit_get_rw_cold(struct btree_trans *trans, unsigned flags)
1019 struct bch_fs *c = trans->c;
1022 if (likely(!(flags & BTREE_INSERT_LAZY_RW)) ||
1023 test_bit(BCH_FS_STARTED, &c->flags))
1024 return -BCH_ERR_erofs_trans_commit;
1026 ret = drop_locks_do(trans, bch2_fs_read_write_early(c));
1030 bch2_write_ref_get(c, BCH_WRITE_REF_trans);
1035 * This is for updates done in the early part of fsck - btree_gc - before we've
1036 * gone RW. we only add the new key to the list of keys for journal replay to
1040 do_bch2_trans_commit_to_journal_replay(struct btree_trans *trans)
1042 struct bch_fs *c = trans->c;
1043 struct btree_insert_entry *i;
1046 trans_for_each_update(trans, i) {
1047 ret = bch2_journal_key_insert(c, i->btree_id, i->level, i->k);
1055 int __bch2_trans_commit(struct btree_trans *trans, unsigned flags)
1057 struct bch_fs *c = trans->c;
1058 struct btree_insert_entry *i = NULL;
1059 struct btree_write_buffered_key *wb;
1063 if (!trans->nr_updates &&
1064 !trans->nr_wb_updates &&
1065 !trans->extra_journal_entries.nr)
1068 if (flags & BTREE_INSERT_GC_LOCK_HELD)
1069 lockdep_assert_held(&c->gc_lock);
1071 ret = bch2_trans_commit_run_triggers(trans);
1075 if (unlikely(!test_bit(BCH_FS_MAY_GO_RW, &c->flags))) {
1076 ret = do_bch2_trans_commit_to_journal_replay(trans);
1080 if (!(flags & BTREE_INSERT_NOCHECK_RW) &&
1081 unlikely(!bch2_write_ref_tryget(c, BCH_WRITE_REF_trans))) {
1082 ret = bch2_trans_commit_get_rw_cold(trans, flags);
1087 if (c->btree_write_buffer.state.nr > c->btree_write_buffer.size / 2 &&
1088 mutex_trylock(&c->btree_write_buffer.flush_lock)) {
1089 bch2_trans_begin(trans);
1090 bch2_trans_unlock(trans);
1092 ret = __bch2_btree_write_buffer_flush(trans,
1093 flags|BTREE_INSERT_NOCHECK_RW, true);
1095 trace_and_count(c, trans_restart_write_buffer_flush, trans, _THIS_IP_);
1096 ret = btree_trans_restart(trans, BCH_ERR_transaction_restart_write_buffer_flush);
1101 EBUG_ON(test_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags));
1103 memset(&trans->journal_preres, 0, sizeof(trans->journal_preres));
1105 trans->journal_u64s = trans->extra_journal_entries.nr;
1106 trans->journal_preres_u64s = 0;
1108 trans->journal_transaction_names = READ_ONCE(c->opts.journal_transaction_names);
1110 if (trans->journal_transaction_names)
1111 trans->journal_u64s += jset_u64s(JSET_ENTRY_LOG_U64s);
1113 trans_for_each_update(trans, i) {
1114 EBUG_ON(!i->path->should_be_locked);
1116 ret = bch2_btree_path_upgrade(trans, i->path, i->level + 1);
1120 EBUG_ON(!btree_node_intent_locked(i->path, i->level));
1122 if (i->key_cache_already_flushed)
1125 /* we're going to journal the key being updated: */
1126 u64s = jset_u64s(i->k->k.u64s);
1128 likely(!(flags & BTREE_INSERT_JOURNAL_REPLAY)))
1129 trans->journal_preres_u64s += u64s;
1131 if (i->flags & BTREE_UPDATE_NOJOURNAL)
1134 trans->journal_u64s += u64s;
1136 /* and we're also going to log the overwrite: */
1137 if (trans->journal_transaction_names)
1138 trans->journal_u64s += jset_u64s(i->old_k.u64s);
1141 trans_for_each_wb_update(trans, wb)
1142 trans->journal_u64s += jset_u64s(wb->k.k.u64s);
1144 if (trans->extra_journal_res) {
1145 ret = bch2_disk_reservation_add(c, trans->disk_res,
1146 trans->extra_journal_res,
1147 (flags & BTREE_INSERT_NOFAIL)
1148 ? BCH_DISK_RESERVATION_NOFAIL : 0);
1153 bch2_trans_verify_not_in_restart(trans);
1154 memset(&trans->journal_res, 0, sizeof(trans->journal_res));
1156 ret = do_bch2_trans_commit(trans, flags, &i, _RET_IP_);
1158 /* make sure we didn't drop or screw up locks: */
1159 bch2_trans_verify_locks(trans);
1164 trace_and_count(c, transaction_commit, trans, _RET_IP_);
1166 bch2_journal_preres_put(&c->journal, &trans->journal_preres);
1168 if (likely(!(flags & BTREE_INSERT_NOCHECK_RW)))
1169 bch2_write_ref_put(c, BCH_WRITE_REF_trans);
1171 bch2_trans_reset_updates(trans);
1175 ret = bch2_trans_commit_error(trans, flags, i, ret, _RET_IP_);
1182 static noinline int __check_pos_snapshot_overwritten(struct btree_trans *trans,
1186 struct bch_fs *c = trans->c;
1187 struct btree_iter iter;
1191 bch2_trans_iter_init(trans, &iter, id, pos,
1192 BTREE_ITER_NOT_EXTENTS|
1193 BTREE_ITER_ALL_SNAPSHOTS);
1195 k = bch2_btree_iter_prev(&iter);
1203 if (!bkey_eq(pos, k.k->p))
1206 if (bch2_snapshot_is_ancestor(c, k.k->p.snapshot, pos.snapshot)) {
1211 bch2_trans_iter_exit(trans, &iter);
1216 static inline int check_pos_snapshot_overwritten(struct btree_trans *trans,
1220 if (!btree_type_has_snapshots(id) ||
1221 pos.snapshot == U32_MAX ||
1222 !snapshot_t(trans->c, pos.snapshot)->children[0])
1225 return __check_pos_snapshot_overwritten(trans, id, pos);
1228 static noinline int extent_front_merge(struct btree_trans *trans,
1229 struct btree_iter *iter,
1231 struct bkey_i **insert,
1232 enum btree_update_flags flags)
1234 struct bch_fs *c = trans->c;
1235 struct bkey_i *update;
1238 update = bch2_bkey_make_mut_noupdate(trans, k);
1239 ret = PTR_ERR_OR_ZERO(update);
1243 if (!bch2_bkey_merge(c, bkey_i_to_s(update), bkey_i_to_s_c(*insert)))
1246 ret = check_pos_snapshot_overwritten(trans, iter->btree_id, k.k->p) ?:
1247 check_pos_snapshot_overwritten(trans, iter->btree_id, (*insert)->k.p);
1253 ret = bch2_btree_delete_at(trans, iter, flags);
1261 static noinline int extent_back_merge(struct btree_trans *trans,
1262 struct btree_iter *iter,
1263 struct bkey_i *insert,
1266 struct bch_fs *c = trans->c;
1269 ret = check_pos_snapshot_overwritten(trans, iter->btree_id, insert->k.p) ?:
1270 check_pos_snapshot_overwritten(trans, iter->btree_id, k.k->p);
1276 bch2_bkey_merge(c, bkey_i_to_s(insert), k);
1281 * When deleting, check if we need to emit a whiteout (because we're overwriting
1282 * something in an ancestor snapshot)
1284 static int need_whiteout_for_snapshot(struct btree_trans *trans,
1285 enum btree_id btree_id, struct bpos pos)
1287 struct btree_iter iter;
1289 u32 snapshot = pos.snapshot;
1292 if (!bch2_snapshot_parent(trans->c, pos.snapshot))
1297 for_each_btree_key_norestart(trans, iter, btree_id, pos,
1298 BTREE_ITER_ALL_SNAPSHOTS|
1299 BTREE_ITER_NOPRESERVE, k, ret) {
1300 if (!bkey_eq(k.k->p, pos))
1303 if (bch2_snapshot_is_ancestor(trans->c, snapshot,
1305 ret = !bkey_whiteout(k.k);
1309 bch2_trans_iter_exit(trans, &iter);
1314 static int get_snapshot_overwrites(struct btree_trans *trans,
1315 enum btree_id btree,
1317 snapshot_id_list *overwrites)
1319 struct bch_fs *c = trans->c;
1320 struct btree_iter iter;
1322 snapshot_id_list overwrites2;
1326 darray_init(overwrites);
1327 darray_init(&overwrites2);
1329 for_each_btree_key_norestart(trans, iter, btree,
1330 SPOS(pos.inode, pos.offset, 0),
1331 BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
1332 if (bpos_ge(k.k->p, pos))
1335 if (bch2_snapshot_is_ancestor(c, k.k->p.snapshot, pos.snapshot)) {
1336 ret = snapshot_list_add(c, &overwrites2, k.k->p.snapshot);
1341 bch2_trans_iter_exit(trans, &iter);
1346 darray_for_each(overwrites2, i)
1347 if (!snapshot_list_has_ancestor(c, &overwrites2, *i)) {
1348 ret = snapshot_list_add(c, overwrites, *i);
1353 *overwrites = overwrites2;
1355 darray_exit(&overwrites2);
1358 darray_exit(overwrites);
1362 int __bch2_insert_snapshot_whiteouts(struct btree_trans *trans,
1363 enum btree_id btree,
1364 struct bpos old_pos,
1365 struct bpos new_pos)
1367 struct bch_fs *c = trans->c;
1368 snapshot_id_list old_overwrites, new_overwrites, updates;
1369 bool began_transaction = false;
1373 if (!bch2_snapshot_has_children(c, old_pos.snapshot))
1376 darray_init(&old_overwrites);
1377 darray_init(&new_overwrites);
1378 darray_init(&updates);
1380 ret = get_snapshot_overwrites(trans, btree, old_pos, &old_overwrites) ?:
1381 get_snapshot_overwrites(trans, btree, new_pos, &new_overwrites);
1385 darray_for_each(old_overwrites, i)
1386 if (!snapshot_list_has_ancestor(c, &new_overwrites, *i)) {
1387 ret = darray_push(&updates, *i);
1392 if (updates.nr > 4) {
1393 bch2_trans_begin(trans);
1394 began_transaction = true;
1397 darray_for_each(updates, i) {
1398 struct btree_iter iter;
1399 struct bkey_i *update;
1401 bch2_trans_iter_init(trans, &iter, btree,
1402 SPOS(new_pos.inode, new_pos.offset, *i),
1403 BTREE_ITER_NOT_EXTENTS|
1405 update = bch2_trans_kmalloc(trans, sizeof(struct bkey_i));
1406 ret = PTR_ERR_OR_ZERO(update);
1410 bkey_init(&update->k);
1411 update->k.p = iter.pos;
1412 update->k.type = KEY_TYPE_whiteout;
1414 ret = bch2_btree_iter_traverse(&iter) ?:
1415 bch2_trans_update(trans, &iter, update,
1416 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE) ?:
1417 (began_transaction && trans->nr_updates > 4
1418 ? bch2_trans_commit(trans, NULL, NULL, BTREE_INSERT_NOFAIL) : 0);
1420 bch2_trans_iter_exit(trans, &iter);
1426 if (began_transaction && trans->nr_updates) {
1427 ret = bch2_trans_commit(trans, NULL, NULL, BTREE_INSERT_NOFAIL);
1432 if (began_transaction)
1433 ret = -BCH_ERR_transaction_restart_nested;
1435 darray_exit(&updates);
1436 darray_exit(&new_overwrites);
1437 darray_exit(&old_overwrites);
1442 int bch2_trans_update_extent(struct btree_trans *trans,
1443 struct btree_iter *orig_iter,
1444 struct bkey_i *insert,
1445 enum btree_update_flags flags)
1447 struct btree_iter iter;
1448 struct bpos start = bkey_start_pos(&insert->k);
1449 struct bkey_i *update;
1451 enum btree_id btree_id = orig_iter->btree_id;
1452 int ret = 0, compressed_sectors;
1454 bch2_trans_iter_init(trans, &iter, btree_id, start,
1456 BTREE_ITER_WITH_UPDATES|
1457 BTREE_ITER_NOT_EXTENTS);
1458 k = bch2_btree_iter_peek_upto(&iter, POS(insert->k.p.inode, U64_MAX));
1459 if ((ret = bkey_err(k)))
1464 if (bkey_eq(k.k->p, bkey_start_pos(&insert->k))) {
1465 if (bch2_bkey_maybe_mergable(k.k, &insert->k)) {
1466 ret = extent_front_merge(trans, &iter, k, &insert, flags);
1474 while (bkey_gt(insert->k.p, bkey_start_pos(k.k))) {
1475 bool front_split = bkey_lt(bkey_start_pos(k.k), start);
1476 bool back_split = bkey_gt(k.k->p, insert->k.p);
1479 * If we're going to be splitting a compressed extent, note it
1480 * so that __bch2_trans_commit() can increase our disk
1483 if (((front_split && back_split) ||
1484 ((front_split || back_split) && k.k->p.snapshot != insert->k.p.snapshot)) &&
1485 (compressed_sectors = bch2_bkey_sectors_compressed(k)))
1486 trans->extra_journal_res += compressed_sectors;
1489 update = bch2_bkey_make_mut_noupdate(trans, k);
1490 if ((ret = PTR_ERR_OR_ZERO(update)))
1493 bch2_cut_back(start, update);
1495 ret = bch2_insert_snapshot_whiteouts(trans, btree_id,
1496 k.k->p, update->k.p) ?:
1497 bch2_btree_insert_nonextent(trans, btree_id, update,
1498 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE|flags);
1503 if (k.k->p.snapshot != insert->k.p.snapshot &&
1504 (front_split || back_split)) {
1505 update = bch2_bkey_make_mut_noupdate(trans, k);
1506 if ((ret = PTR_ERR_OR_ZERO(update)))
1509 bch2_cut_front(start, update);
1510 bch2_cut_back(insert->k.p, update);
1512 ret = bch2_insert_snapshot_whiteouts(trans, btree_id,
1513 k.k->p, update->k.p) ?:
1514 bch2_btree_insert_nonextent(trans, btree_id, update,
1515 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE|flags);
1520 if (bkey_le(k.k->p, insert->k.p)) {
1521 update = bch2_trans_kmalloc(trans, sizeof(*update));
1522 if ((ret = PTR_ERR_OR_ZERO(update)))
1525 bkey_init(&update->k);
1526 update->k.p = k.k->p;
1527 update->k.p.snapshot = insert->k.p.snapshot;
1529 if (insert->k.p.snapshot != k.k->p.snapshot) {
1530 update->k.type = KEY_TYPE_whiteout;
1531 } else if (btree_type_has_snapshots(btree_id)) {
1532 ret = need_whiteout_for_snapshot(trans, btree_id, update->k.p);
1536 update->k.type = KEY_TYPE_whiteout;
1539 ret = bch2_btree_insert_nonextent(trans, btree_id, update,
1540 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE|flags);
1546 update = bch2_bkey_make_mut_noupdate(trans, k);
1547 if ((ret = PTR_ERR_OR_ZERO(update)))
1550 bch2_cut_front(insert->k.p, update);
1552 ret = bch2_trans_update_by_path(trans, iter.path, update,
1553 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE|
1560 bch2_btree_iter_advance(&iter);
1561 k = bch2_btree_iter_peek_upto(&iter, POS(insert->k.p.inode, U64_MAX));
1562 if ((ret = bkey_err(k)))
1568 if (bch2_bkey_maybe_mergable(&insert->k, k.k)) {
1569 ret = extent_back_merge(trans, &iter, insert, k);
1574 if (!bkey_deleted(&insert->k)) {
1576 * Rewinding iterators is expensive: get a new one and the one
1577 * that points to the start of insert will be cloned from:
1579 bch2_trans_iter_exit(trans, &iter);
1580 bch2_trans_iter_init(trans, &iter, btree_id, insert->k.p,
1581 BTREE_ITER_NOT_EXTENTS|
1583 ret = bch2_btree_iter_traverse(&iter) ?:
1584 bch2_trans_update(trans, &iter, insert, flags);
1587 bch2_trans_iter_exit(trans, &iter);
1592 static int __must_check
1593 bch2_trans_update_by_path_trace(struct btree_trans *trans, struct btree_path *path,
1594 struct bkey_i *k, enum btree_update_flags flags,
1597 static noinline int flush_new_cached_update(struct btree_trans *trans,
1598 struct btree_path *path,
1599 struct btree_insert_entry *i,
1600 enum btree_update_flags flags,
1603 struct btree_path *btree_path;
1607 btree_path = bch2_path_get(trans, path->btree_id, path->pos, 1, 0,
1608 BTREE_ITER_INTENT, _THIS_IP_);
1609 ret = bch2_btree_path_traverse(trans, btree_path, 0);
1614 * The old key in the insert entry might actually refer to an existing
1615 * key in the btree that has been deleted from cache and not yet
1616 * flushed. Check for this and skip the flush so we don't run triggers
1617 * against a stale key.
1619 bch2_btree_path_peek_slot_exact(btree_path, &k);
1620 if (!bkey_deleted(&k))
1623 i->key_cache_already_flushed = true;
1624 i->flags |= BTREE_TRIGGER_NORUN;
1626 btree_path_set_should_be_locked(btree_path);
1627 ret = bch2_trans_update_by_path_trace(trans, btree_path, i->k, flags, ip);
1629 bch2_path_put(trans, btree_path, true);
1633 static int __must_check
1634 bch2_trans_update_by_path_trace(struct btree_trans *trans, struct btree_path *path,
1635 struct bkey_i *k, enum btree_update_flags flags,
1638 struct bch_fs *c = trans->c;
1639 struct btree_insert_entry *i, n;
1642 EBUG_ON(!path->should_be_locked);
1643 EBUG_ON(trans->nr_updates >= BTREE_ITER_MAX);
1644 EBUG_ON(!bpos_eq(k->k.p, path->pos));
1646 n = (struct btree_insert_entry) {
1648 .bkey_type = __btree_node_type(path->level, path->btree_id),
1649 .btree_id = path->btree_id,
1650 .level = path->level,
1651 .cached = path->cached,
1657 #ifdef CONFIG_BCACHEFS_DEBUG
1658 trans_for_each_update(trans, i)
1659 BUG_ON(i != trans->updates &&
1660 btree_insert_entry_cmp(i - 1, i) >= 0);
1664 * Pending updates are kept sorted: first, find position of new update,
1665 * then delete/trim any updates the new update overwrites:
1667 trans_for_each_update(trans, i) {
1668 cmp = btree_insert_entry_cmp(&n, i);
1673 if (!cmp && i < trans->updates + trans->nr_updates) {
1674 EBUG_ON(i->insert_trigger_run || i->overwrite_trigger_run);
1676 bch2_path_put(trans, i->path, true);
1678 i->cached = n.cached;
1681 i->ip_allocated = n.ip_allocated;
1683 array_insert_item(trans->updates, trans->nr_updates,
1684 i - trans->updates, n);
1686 i->old_v = bch2_btree_path_peek_slot_exact(path, &i->old_k).v;
1687 i->old_btree_u64s = !bkey_deleted(&i->old_k) ? i->old_k.u64s : 0;
1689 if (unlikely(trans->journal_replay_not_finished)) {
1690 struct bkey_i *j_k =
1691 bch2_journal_keys_peek_slot(c, n.btree_id, n.level, k->k.p);
1700 __btree_path_get(i->path, true);
1703 * If a key is present in the key cache, it must also exist in the
1704 * btree - this is necessary for cache coherency. When iterating over
1705 * a btree that's cached in the key cache, the btree iter code checks
1706 * the key cache - but the key has to exist in the btree for that to
1709 if (path->cached && bkey_deleted(&i->old_k))
1710 return flush_new_cached_update(trans, path, i, flags, ip);
1715 static inline int __must_check
1716 bch2_trans_update_by_path(struct btree_trans *trans, struct btree_path *path,
1717 struct bkey_i *k, enum btree_update_flags flags)
1719 return bch2_trans_update_by_path_trace(trans, path, k, flags, _RET_IP_);
1722 int __must_check bch2_trans_update(struct btree_trans *trans, struct btree_iter *iter,
1723 struct bkey_i *k, enum btree_update_flags flags)
1725 struct btree_path *path = iter->update_path ?: iter->path;
1726 struct bkey_cached *ck;
1729 if (iter->flags & BTREE_ITER_IS_EXTENTS)
1730 return bch2_trans_update_extent(trans, iter, k, flags);
1732 if (bkey_deleted(&k->k) &&
1733 !(flags & BTREE_UPDATE_KEY_CACHE_RECLAIM) &&
1734 (iter->flags & BTREE_ITER_FILTER_SNAPSHOTS)) {
1735 ret = need_whiteout_for_snapshot(trans, iter->btree_id, k->k.p);
1736 if (unlikely(ret < 0))
1740 k->k.type = KEY_TYPE_whiteout;
1744 * Ensure that updates to cached btrees go to the key cache:
1746 if (!(flags & BTREE_UPDATE_KEY_CACHE_RECLAIM) &&
1749 btree_id_cached(trans->c, path->btree_id)) {
1750 if (!iter->key_cache_path ||
1751 !iter->key_cache_path->should_be_locked ||
1752 !bpos_eq(iter->key_cache_path->pos, k->k.p)) {
1753 if (!iter->key_cache_path)
1754 iter->key_cache_path =
1755 bch2_path_get(trans, path->btree_id, path->pos, 1, 0,
1757 BTREE_ITER_CACHED, _THIS_IP_);
1759 iter->key_cache_path =
1760 bch2_btree_path_set_pos(trans, iter->key_cache_path, path->pos,
1761 iter->flags & BTREE_ITER_INTENT,
1764 ret = bch2_btree_path_traverse(trans, iter->key_cache_path,
1769 ck = (void *) iter->key_cache_path->l[0].b;
1771 if (test_bit(BKEY_CACHED_DIRTY, &ck->flags)) {
1772 trace_and_count(trans->c, trans_restart_key_cache_raced, trans, _RET_IP_);
1773 return btree_trans_restart(trans, BCH_ERR_transaction_restart_key_cache_raced);
1776 btree_path_set_should_be_locked(iter->key_cache_path);
1779 path = iter->key_cache_path;
1782 return bch2_trans_update_by_path(trans, path, k, flags);
1785 int __must_check bch2_trans_update_buffered(struct btree_trans *trans,
1786 enum btree_id btree,
1789 struct btree_write_buffered_key *i;
1792 EBUG_ON(trans->nr_wb_updates > trans->wb_updates_size);
1793 EBUG_ON(k->k.u64s > BTREE_WRITE_BUFERED_U64s_MAX);
1795 trans_for_each_wb_update(trans, i) {
1796 if (i->btree == btree && bpos_eq(i->k.k.p, k->k.p)) {
1797 bkey_copy(&i->k, k);
1802 if (!trans->wb_updates ||
1803 trans->nr_wb_updates == trans->wb_updates_size) {
1804 struct btree_write_buffered_key *u;
1806 if (trans->nr_wb_updates == trans->wb_updates_size) {
1807 struct btree_transaction_stats *s = btree_trans_stats(trans);
1809 BUG_ON(trans->wb_updates_size > U8_MAX / 2);
1810 trans->wb_updates_size = max(1, trans->wb_updates_size * 2);
1812 s->wb_updates_size = trans->wb_updates_size;
1815 u = bch2_trans_kmalloc_nomemzero(trans,
1816 trans->wb_updates_size *
1817 sizeof(struct btree_write_buffered_key));
1818 ret = PTR_ERR_OR_ZERO(u);
1822 if (trans->nr_wb_updates)
1823 memcpy(u, trans->wb_updates, trans->nr_wb_updates *
1824 sizeof(struct btree_write_buffered_key));
1825 trans->wb_updates = u;
1828 trans->wb_updates[trans->nr_wb_updates] = (struct btree_write_buffered_key) {
1832 bkey_copy(&trans->wb_updates[trans->nr_wb_updates].k, k);
1833 trans->nr_wb_updates++;
1838 int bch2_bkey_get_empty_slot(struct btree_trans *trans, struct btree_iter *iter,
1839 enum btree_id btree, struct bpos end)
1844 bch2_trans_iter_init(trans, iter, btree, POS_MAX, BTREE_ITER_INTENT);
1845 k = bch2_btree_iter_prev(iter);
1850 bch2_btree_iter_advance(iter);
1851 k = bch2_btree_iter_peek_slot(iter);
1856 BUG_ON(k.k->type != KEY_TYPE_deleted);
1858 if (bkey_gt(k.k->p, end)) {
1859 ret = -BCH_ERR_ENOSPC_btree_slot;
1865 bch2_trans_iter_exit(trans, iter);
1869 void bch2_trans_commit_hook(struct btree_trans *trans,
1870 struct btree_trans_commit_hook *h)
1872 h->next = trans->hooks;
1876 int bch2_btree_insert_nonextent(struct btree_trans *trans,
1877 enum btree_id btree, struct bkey_i *k,
1878 enum btree_update_flags flags)
1880 struct btree_iter iter;
1883 bch2_trans_iter_init(trans, &iter, btree, k->k.p,
1884 BTREE_ITER_NOT_EXTENTS|
1886 ret = bch2_btree_iter_traverse(&iter) ?:
1887 bch2_trans_update(trans, &iter, k, flags);
1888 bch2_trans_iter_exit(trans, &iter);
1892 int __bch2_btree_insert(struct btree_trans *trans, enum btree_id id,
1893 struct bkey_i *k, enum btree_update_flags flags)
1895 struct btree_iter iter;
1898 bch2_trans_iter_init(trans, &iter, id, bkey_start_pos(&k->k),
1901 ret = bch2_btree_iter_traverse(&iter) ?:
1902 bch2_trans_update(trans, &iter, k, flags);
1903 bch2_trans_iter_exit(trans, &iter);
1908 * bch2_btree_insert - insert keys into the extent btree
1909 * @c: pointer to struct bch_fs
1910 * @id: btree to insert into
1911 * @insert_keys: list of keys to insert
1912 * @hook: insert callback
1914 int bch2_btree_insert(struct bch_fs *c, enum btree_id id,
1916 struct disk_reservation *disk_res,
1917 u64 *journal_seq, int flags)
1919 return bch2_trans_do(c, disk_res, journal_seq, flags,
1920 __bch2_btree_insert(&trans, id, k, 0));
1923 int bch2_btree_delete_extent_at(struct btree_trans *trans, struct btree_iter *iter,
1924 unsigned len, unsigned update_flags)
1928 k = bch2_trans_kmalloc(trans, sizeof(*k));
1934 bch2_key_resize(&k->k, len);
1935 return bch2_trans_update(trans, iter, k, update_flags);
1938 int bch2_btree_delete_at(struct btree_trans *trans,
1939 struct btree_iter *iter, unsigned update_flags)
1941 return bch2_btree_delete_extent_at(trans, iter, 0, update_flags);
1944 int bch2_btree_delete_at_buffered(struct btree_trans *trans,
1945 enum btree_id btree, struct bpos pos)
1949 k = bch2_trans_kmalloc(trans, sizeof(*k));
1955 return bch2_trans_update_buffered(trans, btree, k);
1958 int bch2_btree_delete_range_trans(struct btree_trans *trans, enum btree_id id,
1959 struct bpos start, struct bpos end,
1960 unsigned update_flags,
1963 u32 restart_count = trans->restart_count;
1964 struct btree_iter iter;
1968 bch2_trans_iter_init(trans, &iter, id, start, BTREE_ITER_INTENT);
1969 while ((k = bch2_btree_iter_peek_upto(&iter, end)).k) {
1970 struct disk_reservation disk_res =
1971 bch2_disk_reservation_init(trans->c, 0);
1972 struct bkey_i delete;
1978 bkey_init(&delete.k);
1981 * This could probably be more efficient for extents:
1985 * For extents, iter.pos won't necessarily be the same as
1986 * bkey_start_pos(k.k) (for non extents they always will be the
1987 * same). It's important that we delete starting from iter.pos
1988 * because the range we want to delete could start in the middle
1991 * (bch2_btree_iter_peek() does guarantee that iter.pos >=
1992 * bkey_start_pos(k.k)).
1994 delete.k.p = iter.pos;
1996 if (iter.flags & BTREE_ITER_IS_EXTENTS)
1997 bch2_key_resize(&delete.k,
1998 bpos_min(end, k.k->p).offset -
2001 ret = bch2_trans_update(trans, &iter, &delete, update_flags) ?:
2002 bch2_trans_commit(trans, &disk_res, journal_seq,
2003 BTREE_INSERT_NOFAIL);
2004 bch2_disk_reservation_put(trans->c, &disk_res);
2007 * the bch2_trans_begin() call is in a weird place because we
2008 * need to call it after every transaction commit, to avoid path
2009 * overflow, but don't want to call it if the delete operation
2010 * is a no-op and we have no work to do:
2012 bch2_trans_begin(trans);
2014 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
2019 bch2_trans_iter_exit(trans, &iter);
2021 if (!ret && trans_was_restarted(trans, restart_count))
2022 ret = -BCH_ERR_transaction_restart_nested;
2027 * bch_btree_delete_range - delete everything within a given range
2029 * Range is a half open interval - [start, end)
2031 int bch2_btree_delete_range(struct bch_fs *c, enum btree_id id,
2032 struct bpos start, struct bpos end,
2033 unsigned update_flags,
2036 int ret = bch2_trans_run(c,
2037 bch2_btree_delete_range_trans(&trans, id, start, end,
2038 update_flags, journal_seq));
2039 if (ret == -BCH_ERR_transaction_restart_nested)
2044 static int __bch2_trans_log_msg(darray_u64 *entries, const char *fmt, va_list args)
2046 struct printbuf buf = PRINTBUF;
2047 struct jset_entry_log *l;
2051 prt_vprintf(&buf, fmt, args);
2052 ret = buf.allocation_failure ? -BCH_ERR_ENOMEM_trans_log_msg : 0;
2056 u64s = DIV_ROUND_UP(buf.pos, sizeof(u64));
2058 ret = darray_make_room(entries, jset_u64s(u64s));
2062 l = (void *) &darray_top(*entries);
2063 l->entry.u64s = cpu_to_le16(u64s);
2064 l->entry.btree_id = 0;
2066 l->entry.type = BCH_JSET_ENTRY_log;
2067 l->entry.pad[0] = 0;
2068 l->entry.pad[1] = 0;
2069 l->entry.pad[2] = 0;
2070 memcpy(l->d, buf.buf, buf.pos);
2072 l->d[buf.pos++] = '\0';
2074 entries->nr += jset_u64s(u64s);
2076 printbuf_exit(&buf);
2081 __bch2_fs_log_msg(struct bch_fs *c, unsigned commit_flags, const char *fmt,
2086 if (!test_bit(JOURNAL_STARTED, &c->journal.flags)) {
2087 ret = __bch2_trans_log_msg(&c->journal.early_journal_entries, fmt, args);
2089 ret = bch2_trans_do(c, NULL, NULL,
2090 BTREE_INSERT_LAZY_RW|commit_flags,
2091 __bch2_trans_log_msg(&trans.extra_journal_entries, fmt, args));
2097 int bch2_fs_log_msg(struct bch_fs *c, const char *fmt, ...)
2102 va_start(args, fmt);
2103 ret = __bch2_fs_log_msg(c, 0, fmt, args);
2109 * Use for logging messages during recovery to enable reserved space and avoid
2112 int bch2_journal_log_msg(struct bch_fs *c, const char *fmt, ...)
2117 va_start(args, fmt);
2118 ret = __bch2_fs_log_msg(c, JOURNAL_WATERMARK_reserved, fmt, args);