1 // SPDX-License-Identifier: GPL-2.0
4 #include "alloc_foreground.h"
5 #include "bkey_methods.h"
6 #include "btree_cache.h"
8 #include "btree_journal_iter.h"
9 #include "btree_update.h"
10 #include "btree_update_interior.h"
12 #include "btree_iter.h"
13 #include "btree_locking.h"
19 #include "journal_reclaim.h"
25 #include <linux/random.h>
27 static int bch2_btree_insert_node(struct btree_update *, struct btree_trans *,
28 struct btree_path *, struct btree *,
29 struct keylist *, unsigned);
30 static void bch2_btree_update_add_new_node(struct btree_update *, struct btree *);
32 static struct btree_path *get_unlocked_mut_path(struct btree_trans *trans,
33 enum btree_id btree_id,
37 struct btree_path *path;
39 path = bch2_path_get(trans, btree_id, pos, level + 1, level,
40 BTREE_ITER_NOPRESERVE|
41 BTREE_ITER_INTENT, _RET_IP_);
42 path = bch2_btree_path_make_mut(trans, path, true, _RET_IP_);
43 bch2_btree_path_downgrade(trans, path);
44 __bch2_btree_path_unlock(trans, path);
51 * Verify that child nodes correctly span parent node's range:
53 static void btree_node_interior_verify(struct bch_fs *c, struct btree *b)
55 #ifdef CONFIG_BCACHEFS_DEBUG
56 struct bpos next_node = b->data->min_key;
57 struct btree_node_iter iter;
59 struct bkey_s_c_btree_ptr_v2 bp;
61 struct printbuf buf1 = PRINTBUF, buf2 = PRINTBUF;
65 if (!test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags))
68 bch2_btree_node_iter_init_from_start(&iter, b);
71 k = bch2_btree_node_iter_peek_unpack(&iter, b, &unpacked);
72 if (k.k->type != KEY_TYPE_btree_ptr_v2)
74 bp = bkey_s_c_to_btree_ptr_v2(k);
76 if (!bpos_eq(next_node, bp.v->min_key)) {
77 bch2_dump_btree_node(c, b);
78 bch2_bpos_to_text(&buf1, next_node);
79 bch2_bpos_to_text(&buf2, bp.v->min_key);
80 panic("expected next min_key %s got %s\n", buf1.buf, buf2.buf);
83 bch2_btree_node_iter_advance(&iter, b);
85 if (bch2_btree_node_iter_end(&iter)) {
86 if (!bpos_eq(k.k->p, b->key.k.p)) {
87 bch2_dump_btree_node(c, b);
88 bch2_bpos_to_text(&buf1, b->key.k.p);
89 bch2_bpos_to_text(&buf2, k.k->p);
90 panic("expected end %s got %s\n", buf1.buf, buf2.buf);
95 next_node = bpos_successor(k.k->p);
100 /* Calculate ideal packed bkey format for new btree nodes: */
102 void __bch2_btree_calc_format(struct bkey_format_state *s, struct btree *b)
104 struct bkey_packed *k;
109 bset_tree_for_each_key(b, t, k)
110 if (!bkey_deleted(k)) {
111 uk = bkey_unpack_key(b, k);
112 bch2_bkey_format_add_key(s, &uk);
116 static struct bkey_format bch2_btree_calc_format(struct btree *b)
118 struct bkey_format_state s;
120 bch2_bkey_format_init(&s);
121 bch2_bkey_format_add_pos(&s, b->data->min_key);
122 bch2_bkey_format_add_pos(&s, b->data->max_key);
123 __bch2_btree_calc_format(&s, b);
125 return bch2_bkey_format_done(&s);
128 static size_t btree_node_u64s_with_format(struct btree *b,
129 struct bkey_format *new_f)
131 struct bkey_format *old_f = &b->format;
133 /* stupid integer promotion rules */
135 (((int) new_f->key_u64s - old_f->key_u64s) *
136 (int) b->nr.packed_keys) +
137 (((int) new_f->key_u64s - BKEY_U64s) *
138 (int) b->nr.unpacked_keys);
140 BUG_ON(delta + b->nr.live_u64s < 0);
142 return b->nr.live_u64s + delta;
146 * bch2_btree_node_format_fits - check if we could rewrite node with a new format
148 * @c: filesystem handle
149 * @b: btree node to rewrite
150 * @new_f: bkey format to translate keys to
152 * Returns: true if all re-packed keys will be able to fit in a new node.
154 * Assumes all keys will successfully pack with the new format.
156 bool bch2_btree_node_format_fits(struct bch_fs *c, struct btree *b,
157 struct bkey_format *new_f)
159 size_t u64s = btree_node_u64s_with_format(b, new_f);
161 return __vstruct_bytes(struct btree_node, u64s) < btree_bytes(c);
164 /* Btree node freeing/allocation: */
166 static void __btree_node_free(struct bch_fs *c, struct btree *b)
168 trace_and_count(c, btree_node_free, c, b);
170 BUG_ON(btree_node_write_blocked(b));
171 BUG_ON(btree_node_dirty(b));
172 BUG_ON(btree_node_need_write(b));
173 BUG_ON(b == btree_node_root(c, b));
175 BUG_ON(!list_empty(&b->write_blocked));
176 BUG_ON(b->will_make_reachable);
178 clear_btree_node_noevict(b);
180 mutex_lock(&c->btree_cache.lock);
181 list_move(&b->list, &c->btree_cache.freeable);
182 mutex_unlock(&c->btree_cache.lock);
185 static void bch2_btree_node_free_inmem(struct btree_trans *trans,
186 struct btree_path *path,
189 struct bch_fs *c = trans->c;
190 unsigned level = b->c.level;
192 bch2_btree_node_lock_write_nofail(trans, path, &b->c);
193 bch2_btree_node_hash_remove(&c->btree_cache, b);
194 __btree_node_free(c, b);
195 six_unlock_write(&b->c.lock);
196 mark_btree_node_locked_noreset(path, level, BTREE_NODE_INTENT_LOCKED);
198 trans_for_each_path(trans, path)
199 if (path->l[level].b == b) {
200 btree_node_unlock(trans, path, level);
201 path->l[level].b = ERR_PTR(-BCH_ERR_no_btree_node_init);
205 static void bch2_btree_node_free_never_used(struct btree_update *as,
206 struct btree_trans *trans,
209 struct bch_fs *c = as->c;
210 struct prealloc_nodes *p = &as->prealloc_nodes[b->c.lock.readers != NULL];
211 struct btree_path *path;
212 unsigned level = b->c.level;
214 BUG_ON(!list_empty(&b->write_blocked));
215 BUG_ON(b->will_make_reachable != (1UL|(unsigned long) as));
217 b->will_make_reachable = 0;
218 closure_put(&as->cl);
220 clear_btree_node_will_make_reachable(b);
221 clear_btree_node_accessed(b);
222 clear_btree_node_dirty_acct(c, b);
223 clear_btree_node_need_write(b);
225 mutex_lock(&c->btree_cache.lock);
226 list_del_init(&b->list);
227 bch2_btree_node_hash_remove(&c->btree_cache, b);
228 mutex_unlock(&c->btree_cache.lock);
230 BUG_ON(p->nr >= ARRAY_SIZE(p->b));
233 six_unlock_intent(&b->c.lock);
235 trans_for_each_path(trans, path)
236 if (path->l[level].b == b) {
237 btree_node_unlock(trans, path, level);
238 path->l[level].b = ERR_PTR(-BCH_ERR_no_btree_node_init);
242 static struct btree *__bch2_btree_node_alloc(struct btree_trans *trans,
243 struct disk_reservation *res,
248 struct bch_fs *c = trans->c;
249 struct write_point *wp;
251 BKEY_PADDED_ONSTACK(k, BKEY_BTREE_PTR_VAL_U64s_MAX) tmp;
252 struct open_buckets obs = { .nr = 0 };
253 struct bch_devs_list devs_have = (struct bch_devs_list) { 0 };
254 enum bch_watermark watermark = flags & BCH_WATERMARK_MASK;
255 unsigned nr_reserve = watermark > BCH_WATERMARK_reclaim
260 mutex_lock(&c->btree_reserve_cache_lock);
261 if (c->btree_reserve_cache_nr > nr_reserve) {
262 struct btree_alloc *a =
263 &c->btree_reserve_cache[--c->btree_reserve_cache_nr];
266 bkey_copy(&tmp.k, &a->k);
267 mutex_unlock(&c->btree_reserve_cache_lock);
270 mutex_unlock(&c->btree_reserve_cache_lock);
273 ret = bch2_alloc_sectors_start_trans(trans,
274 c->opts.metadata_target ?:
275 c->opts.foreground_target,
277 writepoint_ptr(&c->btree_write_point),
280 c->opts.metadata_replicas_required,
281 watermark, 0, cl, &wp);
285 if (wp->sectors_free < btree_sectors(c)) {
286 struct open_bucket *ob;
289 open_bucket_for_each(c, &wp->ptrs, ob, i)
290 if (ob->sectors_free < btree_sectors(c))
291 ob->sectors_free = 0;
293 bch2_alloc_sectors_done(c, wp);
297 bkey_btree_ptr_v2_init(&tmp.k);
298 bch2_alloc_sectors_append_ptrs(c, wp, &tmp.k, btree_sectors(c), false);
300 bch2_open_bucket_get(c, wp, &obs);
301 bch2_alloc_sectors_done(c, wp);
303 b = bch2_btree_node_mem_alloc(trans, interior_node);
304 six_unlock_write(&b->c.lock);
305 six_unlock_intent(&b->c.lock);
307 /* we hold cannibalize_lock: */
311 bkey_copy(&b->key, &tmp.k);
317 static struct btree *bch2_btree_node_alloc(struct btree_update *as,
318 struct btree_trans *trans,
321 struct bch_fs *c = as->c;
323 struct prealloc_nodes *p = &as->prealloc_nodes[!!level];
326 BUG_ON(level >= BTREE_MAX_DEPTH);
331 btree_node_lock_nopath_nofail(trans, &b->c, SIX_LOCK_intent);
332 btree_node_lock_nopath_nofail(trans, &b->c, SIX_LOCK_write);
334 set_btree_node_accessed(b);
335 set_btree_node_dirty_acct(c, b);
336 set_btree_node_need_write(b);
338 bch2_bset_init_first(b, &b->data->keys);
340 b->c.btree_id = as->btree_id;
341 b->version_ondisk = c->sb.version;
343 memset(&b->nr, 0, sizeof(b->nr));
344 b->data->magic = cpu_to_le64(bset_magic(c));
345 memset(&b->data->_ptr, 0, sizeof(b->data->_ptr));
347 SET_BTREE_NODE_ID(b->data, as->btree_id);
348 SET_BTREE_NODE_LEVEL(b->data, level);
350 if (b->key.k.type == KEY_TYPE_btree_ptr_v2) {
351 struct bkey_i_btree_ptr_v2 *bp = bkey_i_to_btree_ptr_v2(&b->key);
354 bp->v.seq = b->data->keys.seq;
355 bp->v.sectors_written = 0;
358 SET_BTREE_NODE_NEW_EXTENT_OVERWRITE(b->data, true);
360 bch2_btree_build_aux_trees(b);
362 ret = bch2_btree_node_hash_insert(&c->btree_cache, b, level, as->btree_id);
365 trace_and_count(c, btree_node_alloc, c, b);
366 bch2_increment_clock(c, btree_sectors(c), WRITE);
370 static void btree_set_min(struct btree *b, struct bpos pos)
372 if (b->key.k.type == KEY_TYPE_btree_ptr_v2)
373 bkey_i_to_btree_ptr_v2(&b->key)->v.min_key = pos;
374 b->data->min_key = pos;
377 static void btree_set_max(struct btree *b, struct bpos pos)
380 b->data->max_key = pos;
383 static struct btree *bch2_btree_node_alloc_replacement(struct btree_update *as,
384 struct btree_trans *trans,
387 struct btree *n = bch2_btree_node_alloc(as, trans, b->c.level);
388 struct bkey_format format = bch2_btree_calc_format(b);
391 * The keys might expand with the new format - if they wouldn't fit in
392 * the btree node anymore, use the old format for now:
394 if (!bch2_btree_node_format_fits(as->c, b, &format))
397 SET_BTREE_NODE_SEQ(n->data, BTREE_NODE_SEQ(b->data) + 1);
399 btree_set_min(n, b->data->min_key);
400 btree_set_max(n, b->data->max_key);
402 n->data->format = format;
403 btree_node_set_format(n, format);
405 bch2_btree_sort_into(as->c, n, b);
407 btree_node_reset_sib_u64s(n);
411 static struct btree *__btree_root_alloc(struct btree_update *as,
412 struct btree_trans *trans, unsigned level)
414 struct btree *b = bch2_btree_node_alloc(as, trans, level);
416 btree_set_min(b, POS_MIN);
417 btree_set_max(b, SPOS_MAX);
418 b->data->format = bch2_btree_calc_format(b);
420 btree_node_set_format(b, b->data->format);
421 bch2_btree_build_aux_trees(b);
426 static void bch2_btree_reserve_put(struct btree_update *as, struct btree_trans *trans)
428 struct bch_fs *c = as->c;
429 struct prealloc_nodes *p;
431 for (p = as->prealloc_nodes;
432 p < as->prealloc_nodes + ARRAY_SIZE(as->prealloc_nodes);
435 struct btree *b = p->b[--p->nr];
437 mutex_lock(&c->btree_reserve_cache_lock);
439 if (c->btree_reserve_cache_nr <
440 ARRAY_SIZE(c->btree_reserve_cache)) {
441 struct btree_alloc *a =
442 &c->btree_reserve_cache[c->btree_reserve_cache_nr++];
446 bkey_copy(&a->k, &b->key);
448 bch2_open_buckets_put(c, &b->ob);
451 mutex_unlock(&c->btree_reserve_cache_lock);
453 btree_node_lock_nopath_nofail(trans, &b->c, SIX_LOCK_intent);
454 btree_node_lock_nopath_nofail(trans, &b->c, SIX_LOCK_write);
455 __btree_node_free(c, b);
456 six_unlock_write(&b->c.lock);
457 six_unlock_intent(&b->c.lock);
462 static int bch2_btree_reserve_get(struct btree_trans *trans,
463 struct btree_update *as,
464 unsigned nr_nodes[2],
468 struct bch_fs *c = as->c;
473 BUG_ON(nr_nodes[0] + nr_nodes[1] > BTREE_RESERVE_MAX);
476 * Protects reaping from the btree node cache and using the btree node
477 * open bucket reserve:
479 * BTREE_INSERT_NOWAIT only applies to btree node allocation, not
480 * blocking on this lock:
482 ret = bch2_btree_cache_cannibalize_lock(c, cl);
486 for (interior = 0; interior < 2; interior++) {
487 struct prealloc_nodes *p = as->prealloc_nodes + interior;
489 while (p->nr < nr_nodes[interior]) {
490 b = __bch2_btree_node_alloc(trans, &as->disk_res,
491 flags & BTREE_INSERT_NOWAIT ? NULL : cl,
502 bch2_btree_cache_cannibalize_unlock(c);
506 /* Asynchronous interior node update machinery */
508 static void bch2_btree_update_free(struct btree_update *as, struct btree_trans *trans)
510 struct bch_fs *c = as->c;
512 if (as->took_gc_lock)
513 up_read(&c->gc_lock);
514 as->took_gc_lock = false;
516 bch2_journal_preres_put(&c->journal, &as->journal_preres);
518 bch2_journal_pin_drop(&c->journal, &as->journal);
519 bch2_journal_pin_flush(&c->journal, &as->journal);
520 bch2_disk_reservation_put(c, &as->disk_res);
521 bch2_btree_reserve_put(as, trans);
523 bch2_time_stats_update(&c->times[BCH_TIME_btree_interior_update_total],
526 mutex_lock(&c->btree_interior_update_lock);
527 list_del(&as->unwritten_list);
530 closure_debug_destroy(&as->cl);
531 mempool_free(as, &c->btree_interior_update_pool);
534 * Have to do the wakeup with btree_interior_update_lock still held,
535 * since being on btree_interior_update_list is our ref on @c:
537 closure_wake_up(&c->btree_interior_update_wait);
539 mutex_unlock(&c->btree_interior_update_lock);
542 static void btree_update_add_key(struct btree_update *as,
543 struct keylist *keys, struct btree *b)
545 struct bkey_i *k = &b->key;
547 BUG_ON(bch2_keylist_u64s(keys) + k->k.u64s >
548 ARRAY_SIZE(as->_old_keys));
550 bkey_copy(keys->top, k);
551 bkey_i_to_btree_ptr_v2(keys->top)->v.mem_ptr = b->c.level + 1;
553 bch2_keylist_push(keys);
557 * The transactional part of an interior btree node update, where we journal the
558 * update we did to the interior node and update alloc info:
560 static int btree_update_nodes_written_trans(struct btree_trans *trans,
561 struct btree_update *as)
566 ret = darray_make_room(&trans->extra_journal_entries, as->journal_u64s);
570 memcpy(&darray_top(trans->extra_journal_entries),
572 as->journal_u64s * sizeof(u64));
573 trans->extra_journal_entries.nr += as->journal_u64s;
575 trans->journal_pin = &as->journal;
577 for_each_keylist_key(&as->old_keys, k) {
578 unsigned level = bkey_i_to_btree_ptr_v2(k)->v.mem_ptr;
580 ret = bch2_trans_mark_old(trans, as->btree_id, level, bkey_i_to_s_c(k), 0);
585 for_each_keylist_key(&as->new_keys, k) {
586 unsigned level = bkey_i_to_btree_ptr_v2(k)->v.mem_ptr;
588 ret = bch2_trans_mark_new(trans, as->btree_id, level, k, 0);
596 static void btree_update_nodes_written(struct btree_update *as)
598 struct bch_fs *c = as->c;
600 struct btree_trans *trans = bch2_trans_get(c);
606 * If we're already in an error state, it might be because a btree node
607 * was never written, and we might be trying to free that same btree
608 * node here, but it won't have been marked as allocated and we'll see
609 * spurious disk usage inconsistencies in the transactional part below
610 * if we don't skip it:
612 ret = bch2_journal_error(&c->journal);
617 * Wait for any in flight writes to finish before we free the old nodes
620 for (i = 0; i < as->nr_old_nodes; i++) {
623 b = as->old_nodes[i];
625 btree_node_lock_nopath_nofail(trans, &b->c, SIX_LOCK_read);
626 seq = b->data ? b->data->keys.seq : 0;
627 six_unlock_read(&b->c.lock);
629 if (seq == as->old_nodes_seq[i])
630 wait_on_bit_io(&b->flags, BTREE_NODE_write_in_flight_inner,
631 TASK_UNINTERRUPTIBLE);
635 * We did an update to a parent node where the pointers we added pointed
636 * to child nodes that weren't written yet: now, the child nodes have
637 * been written so we can write out the update to the interior node.
641 * We can't call into journal reclaim here: we'd block on the journal
642 * reclaim lock, but we may need to release the open buckets we have
643 * pinned in order for other btree updates to make forward progress, and
644 * journal reclaim does btree updates when flushing bkey_cached entries,
645 * which may require allocations as well.
647 ret = commit_do(trans, &as->disk_res, &journal_seq,
648 BCH_WATERMARK_reclaim|
650 BTREE_INSERT_NOCHECK_RW|
651 BTREE_INSERT_JOURNAL_RECLAIM,
652 btree_update_nodes_written_trans(trans, as));
653 bch2_trans_unlock(trans);
655 bch2_fs_fatal_err_on(ret && !bch2_journal_error(&c->journal), c,
656 "%s(): error %s", __func__, bch2_err_str(ret));
659 struct btree_path *path;
662 path = get_unlocked_mut_path(trans, as->btree_id, b->c.level, b->key.k.p);
664 * @b is the node we did the final insert into:
666 * On failure to get a journal reservation, we still have to
667 * unblock the write and allow most of the write path to happen
668 * so that shutdown works, but the i->journal_seq mechanism
669 * won't work to prevent the btree write from being visible (we
670 * didn't get a journal sequence number) - instead
671 * __bch2_btree_node_write() doesn't do the actual write if
672 * we're in journal error state:
676 * Ensure transaction is unlocked before using
677 * btree_node_lock_nopath() (the use of which is always suspect,
678 * we need to work on removing this in the future)
680 * It should be, but get_unlocked_mut_path() -> bch2_path_get()
681 * calls bch2_path_upgrade(), before we call path_make_mut(), so
682 * we may rarely end up with a locked path besides the one we
685 bch2_trans_unlock(trans);
686 btree_node_lock_nopath_nofail(trans, &b->c, SIX_LOCK_intent);
687 mark_btree_node_locked(trans, path, b->c.level, BTREE_NODE_INTENT_LOCKED);
688 path->l[b->c.level].lock_seq = six_lock_seq(&b->c.lock);
689 path->l[b->c.level].b = b;
691 bch2_btree_node_lock_write_nofail(trans, path, &b->c);
693 mutex_lock(&c->btree_interior_update_lock);
695 list_del(&as->write_blocked_list);
696 if (list_empty(&b->write_blocked))
697 clear_btree_node_write_blocked(b);
700 * Node might have been freed, recheck under
701 * btree_interior_update_lock:
705 BUG_ON(!btree_node_dirty(b));
708 struct bset *last = btree_bset_last(b);
710 last->journal_seq = cpu_to_le64(
712 le64_to_cpu(last->journal_seq)));
714 bch2_btree_add_journal_pin(c, b, journal_seq);
717 * If we didn't get a journal sequence number we
718 * can't write this btree node, because recovery
719 * won't know to ignore this write:
721 set_btree_node_never_write(b);
725 mutex_unlock(&c->btree_interior_update_lock);
727 mark_btree_node_locked_noreset(path, b->c.level, BTREE_NODE_INTENT_LOCKED);
728 six_unlock_write(&b->c.lock);
730 btree_node_write_if_need(c, b, SIX_LOCK_intent);
731 btree_node_unlock(trans, path, b->c.level);
732 bch2_path_put(trans, path, true);
735 bch2_journal_pin_drop(&c->journal, &as->journal);
737 bch2_journal_preres_put(&c->journal, &as->journal_preres);
739 mutex_lock(&c->btree_interior_update_lock);
740 for (i = 0; i < as->nr_new_nodes; i++) {
741 b = as->new_nodes[i];
743 BUG_ON(b->will_make_reachable != (unsigned long) as);
744 b->will_make_reachable = 0;
745 clear_btree_node_will_make_reachable(b);
747 mutex_unlock(&c->btree_interior_update_lock);
749 for (i = 0; i < as->nr_new_nodes; i++) {
750 b = as->new_nodes[i];
752 btree_node_lock_nopath_nofail(trans, &b->c, SIX_LOCK_read);
753 btree_node_write_if_need(c, b, SIX_LOCK_read);
754 six_unlock_read(&b->c.lock);
757 for (i = 0; i < as->nr_open_buckets; i++)
758 bch2_open_bucket_put(c, c->open_buckets + as->open_buckets[i]);
760 bch2_btree_update_free(as, trans);
761 bch2_trans_put(trans);
764 static void btree_interior_update_work(struct work_struct *work)
767 container_of(work, struct bch_fs, btree_interior_update_work);
768 struct btree_update *as;
771 mutex_lock(&c->btree_interior_update_lock);
772 as = list_first_entry_or_null(&c->btree_interior_updates_unwritten,
773 struct btree_update, unwritten_list);
774 if (as && !as->nodes_written)
776 mutex_unlock(&c->btree_interior_update_lock);
781 btree_update_nodes_written(as);
785 static void btree_update_set_nodes_written(struct closure *cl)
787 struct btree_update *as = container_of(cl, struct btree_update, cl);
788 struct bch_fs *c = as->c;
790 mutex_lock(&c->btree_interior_update_lock);
791 as->nodes_written = true;
792 mutex_unlock(&c->btree_interior_update_lock);
794 queue_work(c->btree_interior_update_worker, &c->btree_interior_update_work);
798 * We're updating @b with pointers to nodes that haven't finished writing yet:
799 * block @b from being written until @as completes
801 static void btree_update_updated_node(struct btree_update *as, struct btree *b)
803 struct bch_fs *c = as->c;
805 mutex_lock(&c->btree_interior_update_lock);
806 list_add_tail(&as->unwritten_list, &c->btree_interior_updates_unwritten);
808 BUG_ON(as->mode != BTREE_INTERIOR_NO_UPDATE);
809 BUG_ON(!btree_node_dirty(b));
812 as->mode = BTREE_INTERIOR_UPDATING_NODE;
815 set_btree_node_write_blocked(b);
816 list_add(&as->write_blocked_list, &b->write_blocked);
818 mutex_unlock(&c->btree_interior_update_lock);
821 static void btree_update_reparent(struct btree_update *as,
822 struct btree_update *child)
824 struct bch_fs *c = as->c;
826 lockdep_assert_held(&c->btree_interior_update_lock);
829 child->mode = BTREE_INTERIOR_UPDATING_AS;
831 bch2_journal_pin_copy(&c->journal, &as->journal, &child->journal, NULL);
834 static void btree_update_updated_root(struct btree_update *as, struct btree *b)
836 struct bkey_i *insert = &b->key;
837 struct bch_fs *c = as->c;
839 BUG_ON(as->mode != BTREE_INTERIOR_NO_UPDATE);
841 BUG_ON(as->journal_u64s + jset_u64s(insert->k.u64s) >
842 ARRAY_SIZE(as->journal_entries));
845 journal_entry_set((void *) &as->journal_entries[as->journal_u64s],
846 BCH_JSET_ENTRY_btree_root,
847 b->c.btree_id, b->c.level,
848 insert, insert->k.u64s);
850 mutex_lock(&c->btree_interior_update_lock);
851 list_add_tail(&as->unwritten_list, &c->btree_interior_updates_unwritten);
853 as->mode = BTREE_INTERIOR_UPDATING_ROOT;
854 mutex_unlock(&c->btree_interior_update_lock);
858 * bch2_btree_update_add_new_node:
860 * This causes @as to wait on @b to be written, before it gets to
861 * bch2_btree_update_nodes_written
863 * Additionally, it sets b->will_make_reachable to prevent any additional writes
864 * to @b from happening besides the first until @b is reachable on disk
866 * And it adds @b to the list of @as's new nodes, so that we can update sector
867 * counts in bch2_btree_update_nodes_written:
869 static void bch2_btree_update_add_new_node(struct btree_update *as, struct btree *b)
871 struct bch_fs *c = as->c;
873 closure_get(&as->cl);
875 mutex_lock(&c->btree_interior_update_lock);
876 BUG_ON(as->nr_new_nodes >= ARRAY_SIZE(as->new_nodes));
877 BUG_ON(b->will_make_reachable);
879 as->new_nodes[as->nr_new_nodes++] = b;
880 b->will_make_reachable = 1UL|(unsigned long) as;
881 set_btree_node_will_make_reachable(b);
883 mutex_unlock(&c->btree_interior_update_lock);
885 btree_update_add_key(as, &as->new_keys, b);
887 if (b->key.k.type == KEY_TYPE_btree_ptr_v2) {
888 unsigned bytes = vstruct_end(&b->data->keys) - (void *) b->data;
889 unsigned sectors = round_up(bytes, block_bytes(c)) >> 9;
891 bkey_i_to_btree_ptr_v2(&b->key)->v.sectors_written =
892 cpu_to_le16(sectors);
897 * returns true if @b was a new node
899 static void btree_update_drop_new_node(struct bch_fs *c, struct btree *b)
901 struct btree_update *as;
905 mutex_lock(&c->btree_interior_update_lock);
907 * When b->will_make_reachable != 0, it owns a ref on as->cl that's
908 * dropped when it gets written by bch2_btree_complete_write - the
909 * xchg() is for synchronization with bch2_btree_complete_write:
911 v = xchg(&b->will_make_reachable, 0);
912 clear_btree_node_will_make_reachable(b);
913 as = (struct btree_update *) (v & ~1UL);
916 mutex_unlock(&c->btree_interior_update_lock);
920 for (i = 0; i < as->nr_new_nodes; i++)
921 if (as->new_nodes[i] == b)
926 array_remove_item(as->new_nodes, as->nr_new_nodes, i);
927 mutex_unlock(&c->btree_interior_update_lock);
930 closure_put(&as->cl);
933 static void bch2_btree_update_get_open_buckets(struct btree_update *as, struct btree *b)
936 as->open_buckets[as->nr_open_buckets++] =
941 * @b is being split/rewritten: it may have pointers to not-yet-written btree
942 * nodes and thus outstanding btree_updates - redirect @b's
943 * btree_updates to point to this btree_update:
945 static void bch2_btree_interior_update_will_free_node(struct btree_update *as,
948 struct bch_fs *c = as->c;
949 struct btree_update *p, *n;
950 struct btree_write *w;
952 set_btree_node_dying(b);
954 if (btree_node_fake(b))
957 mutex_lock(&c->btree_interior_update_lock);
960 * Does this node have any btree_update operations preventing
961 * it from being written?
963 * If so, redirect them to point to this btree_update: we can
964 * write out our new nodes, but we won't make them visible until those
965 * operations complete
967 list_for_each_entry_safe(p, n, &b->write_blocked, write_blocked_list) {
968 list_del_init(&p->write_blocked_list);
969 btree_update_reparent(as, p);
972 * for flush_held_btree_writes() waiting on updates to flush or
973 * nodes to be writeable:
975 closure_wake_up(&c->btree_interior_update_wait);
978 clear_btree_node_dirty_acct(c, b);
979 clear_btree_node_need_write(b);
980 clear_btree_node_write_blocked(b);
983 * Does this node have unwritten data that has a pin on the journal?
985 * If so, transfer that pin to the btree_update operation -
986 * note that if we're freeing multiple nodes, we only need to keep the
987 * oldest pin of any of the nodes we're freeing. We'll release the pin
988 * when the new nodes are persistent and reachable on disk:
990 w = btree_current_write(b);
991 bch2_journal_pin_copy(&c->journal, &as->journal, &w->journal, NULL);
992 bch2_journal_pin_drop(&c->journal, &w->journal);
994 w = btree_prev_write(b);
995 bch2_journal_pin_copy(&c->journal, &as->journal, &w->journal, NULL);
996 bch2_journal_pin_drop(&c->journal, &w->journal);
998 mutex_unlock(&c->btree_interior_update_lock);
1001 * Is this a node that isn't reachable on disk yet?
1003 * Nodes that aren't reachable yet have writes blocked until they're
1004 * reachable - now that we've cancelled any pending writes and moved
1005 * things waiting on that write to wait on this update, we can drop this
1006 * node from the list of nodes that the other update is making
1007 * reachable, prior to freeing it:
1009 btree_update_drop_new_node(c, b);
1011 btree_update_add_key(as, &as->old_keys, b);
1013 as->old_nodes[as->nr_old_nodes] = b;
1014 as->old_nodes_seq[as->nr_old_nodes] = b->data->keys.seq;
1018 static void bch2_btree_update_done(struct btree_update *as, struct btree_trans *trans)
1020 struct bch_fs *c = as->c;
1021 u64 start_time = as->start_time;
1023 BUG_ON(as->mode == BTREE_INTERIOR_NO_UPDATE);
1025 if (as->took_gc_lock)
1026 up_read(&as->c->gc_lock);
1027 as->took_gc_lock = false;
1029 bch2_btree_reserve_put(as, trans);
1031 continue_at(&as->cl, btree_update_set_nodes_written,
1032 as->c->btree_interior_update_worker);
1034 bch2_time_stats_update(&c->times[BCH_TIME_btree_interior_update_foreground],
1038 static struct btree_update *
1039 bch2_btree_update_start(struct btree_trans *trans, struct btree_path *path,
1040 unsigned level, bool split, unsigned flags)
1042 struct bch_fs *c = trans->c;
1043 struct btree_update *as;
1044 u64 start_time = local_clock();
1045 int disk_res_flags = (flags & BTREE_INSERT_NOFAIL)
1046 ? BCH_DISK_RESERVATION_NOFAIL : 0;
1047 unsigned nr_nodes[2] = { 0, 0 };
1048 unsigned update_level = level;
1049 enum bch_watermark watermark = flags & BCH_WATERMARK_MASK;
1050 unsigned journal_flags = 0;
1052 u32 restart_count = trans->restart_count;
1054 BUG_ON(!path->should_be_locked);
1056 if (watermark == BCH_WATERMARK_copygc)
1057 watermark = BCH_WATERMARK_btree_copygc;
1058 if (watermark < BCH_WATERMARK_btree)
1059 watermark = BCH_WATERMARK_btree;
1061 flags &= ~BCH_WATERMARK_MASK;
1064 if (flags & BTREE_INSERT_JOURNAL_RECLAIM)
1065 journal_flags |= JOURNAL_RES_GET_NONBLOCK;
1066 journal_flags |= watermark;
1069 nr_nodes[!!update_level] += 1 + split;
1072 ret = bch2_btree_path_upgrade(trans, path, update_level + 1);
1074 return ERR_PTR(ret);
1076 if (!btree_path_node(path, update_level)) {
1077 /* Allocating new root? */
1078 nr_nodes[1] += split;
1079 update_level = BTREE_MAX_DEPTH;
1083 if (bch2_btree_node_insert_fits(c, path->l[update_level].b,
1084 BKEY_BTREE_PTR_U64s_MAX * (1 + split)))
1087 split = path->l[update_level].b->nr.live_u64s > BTREE_SPLIT_THRESHOLD(c);
1090 if (flags & BTREE_INSERT_GC_LOCK_HELD)
1091 lockdep_assert_held(&c->gc_lock);
1092 else if (!down_read_trylock(&c->gc_lock)) {
1093 ret = drop_locks_do(trans, (down_read(&c->gc_lock), 0));
1095 up_read(&c->gc_lock);
1096 return ERR_PTR(ret);
1100 as = mempool_alloc(&c->btree_interior_update_pool, GFP_NOFS);
1101 memset(as, 0, sizeof(*as));
1102 closure_init(&as->cl, NULL);
1104 as->start_time = start_time;
1105 as->mode = BTREE_INTERIOR_NO_UPDATE;
1106 as->took_gc_lock = !(flags & BTREE_INSERT_GC_LOCK_HELD);
1107 as->btree_id = path->btree_id;
1108 as->update_level = update_level;
1109 INIT_LIST_HEAD(&as->list);
1110 INIT_LIST_HEAD(&as->unwritten_list);
1111 INIT_LIST_HEAD(&as->write_blocked_list);
1112 bch2_keylist_init(&as->old_keys, as->_old_keys);
1113 bch2_keylist_init(&as->new_keys, as->_new_keys);
1114 bch2_keylist_init(&as->parent_keys, as->inline_keys);
1116 mutex_lock(&c->btree_interior_update_lock);
1117 list_add_tail(&as->list, &c->btree_interior_update_list);
1118 mutex_unlock(&c->btree_interior_update_lock);
1121 * We don't want to allocate if we're in an error state, that can cause
1122 * deadlock on emergency shutdown due to open buckets getting stuck in
1123 * the btree_reserve_cache after allocator shutdown has cleared it out.
1124 * This check needs to come after adding us to the btree_interior_update
1125 * list but before calling bch2_btree_reserve_get, to synchronize with
1126 * __bch2_fs_read_only().
1128 ret = bch2_journal_error(&c->journal);
1132 ret = bch2_journal_preres_get(&c->journal, &as->journal_preres,
1133 BTREE_UPDATE_JOURNAL_RES,
1134 journal_flags|JOURNAL_RES_GET_NONBLOCK);
1136 if (flags & BTREE_INSERT_JOURNAL_RECLAIM) {
1137 ret = -BCH_ERR_journal_reclaim_would_deadlock;
1141 ret = drop_locks_do(trans,
1142 bch2_journal_preres_get(&c->journal, &as->journal_preres,
1143 BTREE_UPDATE_JOURNAL_RES,
1145 if (ret == -BCH_ERR_journal_preres_get_blocked) {
1146 trace_and_count(c, trans_restart_journal_preres_get, trans, _RET_IP_, journal_flags);
1147 ret = btree_trans_restart(trans, BCH_ERR_transaction_restart_journal_preres_get);
1153 ret = bch2_disk_reservation_get(c, &as->disk_res,
1154 (nr_nodes[0] + nr_nodes[1]) * btree_sectors(c),
1155 c->opts.metadata_replicas,
1160 ret = bch2_btree_reserve_get(trans, as, nr_nodes, flags, NULL);
1161 if (bch2_err_matches(ret, ENOSPC) ||
1162 bch2_err_matches(ret, ENOMEM)) {
1166 * XXX: this should probably be a separate BTREE_INSERT_NONBLOCK
1169 if (bch2_err_matches(ret, ENOSPC) &&
1170 (flags & BTREE_INSERT_JOURNAL_RECLAIM) &&
1171 watermark != BCH_WATERMARK_reclaim) {
1172 ret = -BCH_ERR_journal_reclaim_would_deadlock;
1176 closure_init_stack(&cl);
1179 ret = bch2_btree_reserve_get(trans, as, nr_nodes, flags, &cl);
1181 bch2_trans_unlock(trans);
1183 } while (bch2_err_matches(ret, BCH_ERR_operation_blocked));
1187 trace_and_count(c, btree_reserve_get_fail, trans->fn,
1188 _RET_IP_, nr_nodes[0] + nr_nodes[1], ret);
1192 ret = bch2_trans_relock(trans);
1196 bch2_trans_verify_not_restarted(trans, restart_count);
1199 bch2_btree_update_free(as, trans);
1200 return ERR_PTR(ret);
1203 /* Btree root updates: */
1205 static void bch2_btree_set_root_inmem(struct bch_fs *c, struct btree *b)
1207 /* Root nodes cannot be reaped */
1208 mutex_lock(&c->btree_cache.lock);
1209 list_del_init(&b->list);
1210 mutex_unlock(&c->btree_cache.lock);
1212 mutex_lock(&c->btree_root_lock);
1213 BUG_ON(btree_node_root(c, b) &&
1214 (b->c.level < btree_node_root(c, b)->c.level ||
1215 !btree_node_dying(btree_node_root(c, b))));
1217 bch2_btree_id_root(c, b->c.btree_id)->b = b;
1218 mutex_unlock(&c->btree_root_lock);
1220 bch2_recalc_btree_reserve(c);
1223 static void bch2_btree_set_root(struct btree_update *as,
1224 struct btree_trans *trans,
1225 struct btree_path *path,
1228 struct bch_fs *c = as->c;
1231 trace_and_count(c, btree_node_set_root, c, b);
1233 old = btree_node_root(c, b);
1236 * Ensure no one is using the old root while we switch to the
1239 bch2_btree_node_lock_write_nofail(trans, path, &old->c);
1241 bch2_btree_set_root_inmem(c, b);
1243 btree_update_updated_root(as, b);
1246 * Unlock old root after new root is visible:
1248 * The new root isn't persistent, but that's ok: we still have
1249 * an intent lock on the new root, and any updates that would
1250 * depend on the new root would have to update the new root.
1252 bch2_btree_node_unlock_write(trans, path, old);
1255 /* Interior node updates: */
1257 static void bch2_insert_fixup_btree_ptr(struct btree_update *as,
1258 struct btree_trans *trans,
1259 struct btree_path *path,
1261 struct btree_node_iter *node_iter,
1262 struct bkey_i *insert)
1264 struct bch_fs *c = as->c;
1265 struct bkey_packed *k;
1266 struct printbuf buf = PRINTBUF;
1267 unsigned long old, new, v;
1269 BUG_ON(insert->k.type == KEY_TYPE_btree_ptr_v2 &&
1270 !btree_ptr_sectors_written(insert));
1272 if (unlikely(!test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags)))
1273 bch2_journal_key_overwritten(c, b->c.btree_id, b->c.level, insert->k.p);
1275 if (bch2_bkey_invalid(c, bkey_i_to_s_c(insert),
1276 btree_node_type(b), WRITE, &buf) ?:
1277 bch2_bkey_in_btree_node(c, b, bkey_i_to_s_c(insert), &buf)) {
1278 printbuf_reset(&buf);
1279 prt_printf(&buf, "inserting invalid bkey\n ");
1280 bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(insert));
1281 prt_printf(&buf, "\n ");
1282 bch2_bkey_invalid(c, bkey_i_to_s_c(insert),
1283 btree_node_type(b), WRITE, &buf);
1284 bch2_bkey_in_btree_node(c, b, bkey_i_to_s_c(insert), &buf);
1286 bch2_fs_inconsistent(c, "%s", buf.buf);
1290 BUG_ON(as->journal_u64s + jset_u64s(insert->k.u64s) >
1291 ARRAY_SIZE(as->journal_entries));
1294 journal_entry_set((void *) &as->journal_entries[as->journal_u64s],
1295 BCH_JSET_ENTRY_btree_keys,
1296 b->c.btree_id, b->c.level,
1297 insert, insert->k.u64s);
1299 while ((k = bch2_btree_node_iter_peek_all(node_iter, b)) &&
1300 bkey_iter_pos_cmp(b, k, &insert->k.p) < 0)
1301 bch2_btree_node_iter_advance(node_iter, b);
1303 bch2_btree_bset_insert_key(trans, path, b, node_iter, insert);
1304 set_btree_node_dirty_acct(c, b);
1306 v = READ_ONCE(b->flags);
1310 new &= ~BTREE_WRITE_TYPE_MASK;
1311 new |= BTREE_WRITE_interior;
1312 new |= 1 << BTREE_NODE_need_write;
1313 } while ((v = cmpxchg(&b->flags, old, new)) != old);
1315 printbuf_exit(&buf);
1319 __bch2_btree_insert_keys_interior(struct btree_update *as,
1320 struct btree_trans *trans,
1321 struct btree_path *path,
1323 struct btree_node_iter node_iter,
1324 struct keylist *keys)
1326 struct bkey_i *insert = bch2_keylist_front(keys);
1327 struct bkey_packed *k;
1329 BUG_ON(btree_node_type(b) != BKEY_TYPE_btree);
1331 while ((k = bch2_btree_node_iter_prev_all(&node_iter, b)) &&
1332 (bkey_cmp_left_packed(b, k, &insert->k.p) >= 0))
1335 while (!bch2_keylist_empty(keys)) {
1336 insert = bch2_keylist_front(keys);
1338 if (bpos_gt(insert->k.p, b->key.k.p))
1341 bch2_insert_fixup_btree_ptr(as, trans, path, b, &node_iter, insert);
1342 bch2_keylist_pop_front(keys);
1347 * Move keys from n1 (original replacement node, now lower node) to n2 (higher
1350 static void __btree_split_node(struct btree_update *as,
1351 struct btree_trans *trans,
1355 struct bkey_packed *k;
1356 struct bpos n1_pos = POS_MIN;
1357 struct btree_node_iter iter;
1358 struct bset *bsets[2];
1359 struct bkey_format_state format[2];
1360 struct bkey_packed *out[2];
1362 unsigned u64s, n1_u64s = (b->nr.live_u64s * 3) / 5;
1365 for (i = 0; i < 2; i++) {
1366 BUG_ON(n[i]->nsets != 1);
1368 bsets[i] = btree_bset_first(n[i]);
1369 out[i] = bsets[i]->start;
1371 SET_BTREE_NODE_SEQ(n[i]->data, BTREE_NODE_SEQ(b->data) + 1);
1372 bch2_bkey_format_init(&format[i]);
1376 for_each_btree_node_key(b, k, &iter) {
1377 if (bkey_deleted(k))
1380 i = u64s >= n1_u64s;
1382 uk = bkey_unpack_key(b, k);
1385 bch2_bkey_format_add_key(&format[i], &uk);
1388 btree_set_min(n[0], b->data->min_key);
1389 btree_set_max(n[0], n1_pos);
1390 btree_set_min(n[1], bpos_successor(n1_pos));
1391 btree_set_max(n[1], b->data->max_key);
1393 for (i = 0; i < 2; i++) {
1394 bch2_bkey_format_add_pos(&format[i], n[i]->data->min_key);
1395 bch2_bkey_format_add_pos(&format[i], n[i]->data->max_key);
1397 n[i]->data->format = bch2_bkey_format_done(&format[i]);
1398 btree_node_set_format(n[i], n[i]->data->format);
1402 for_each_btree_node_key(b, k, &iter) {
1403 if (bkey_deleted(k))
1406 i = u64s >= n1_u64s;
1409 if (bch2_bkey_transform(&n[i]->format, out[i], bkey_packed(k)
1410 ? &b->format: &bch2_bkey_format_current, k))
1411 out[i]->format = KEY_FORMAT_LOCAL_BTREE;
1413 bch2_bkey_unpack(b, (void *) out[i], k);
1415 out[i]->needs_whiteout = false;
1417 btree_keys_account_key_add(&n[i]->nr, 0, out[i]);
1418 out[i] = bkey_p_next(out[i]);
1421 for (i = 0; i < 2; i++) {
1422 bsets[i]->u64s = cpu_to_le16((u64 *) out[i] - bsets[i]->_data);
1424 BUG_ON(!bsets[i]->u64s);
1426 set_btree_bset_end(n[i], n[i]->set);
1428 btree_node_reset_sib_u64s(n[i]);
1430 bch2_verify_btree_nr_keys(n[i]);
1433 btree_node_interior_verify(as->c, n[i]);
1438 * For updates to interior nodes, we've got to do the insert before we split
1439 * because the stuff we're inserting has to be inserted atomically. Post split,
1440 * the keys might have to go in different nodes and the split would no longer be
1443 * Worse, if the insert is from btree node coalescing, if we do the insert after
1444 * we do the split (and pick the pivot) - the pivot we pick might be between
1445 * nodes that were coalesced, and thus in the middle of a child node post
1448 static void btree_split_insert_keys(struct btree_update *as,
1449 struct btree_trans *trans,
1450 struct btree_path *path,
1452 struct keylist *keys)
1454 if (!bch2_keylist_empty(keys) &&
1455 bpos_le(bch2_keylist_front(keys)->k.p, b->data->max_key)) {
1456 struct btree_node_iter node_iter;
1458 bch2_btree_node_iter_init(&node_iter, b, &bch2_keylist_front(keys)->k.p);
1460 __bch2_btree_insert_keys_interior(as, trans, path, b, node_iter, keys);
1462 btree_node_interior_verify(as->c, b);
1466 static int btree_split(struct btree_update *as, struct btree_trans *trans,
1467 struct btree_path *path, struct btree *b,
1468 struct keylist *keys, unsigned flags)
1470 struct bch_fs *c = as->c;
1471 struct btree *parent = btree_node_parent(path, b);
1472 struct btree *n1, *n2 = NULL, *n3 = NULL;
1473 struct btree_path *path1 = NULL, *path2 = NULL;
1474 u64 start_time = local_clock();
1477 BUG_ON(!parent && (b != btree_node_root(c, b)));
1478 BUG_ON(parent && !btree_node_intent_locked(path, b->c.level + 1));
1480 bch2_btree_interior_update_will_free_node(as, b);
1482 if (b->nr.live_u64s > BTREE_SPLIT_THRESHOLD(c)) {
1485 trace_and_count(c, btree_node_split, c, b);
1487 n[0] = n1 = bch2_btree_node_alloc(as, trans, b->c.level);
1488 n[1] = n2 = bch2_btree_node_alloc(as, trans, b->c.level);
1490 __btree_split_node(as, trans, b, n);
1493 btree_split_insert_keys(as, trans, path, n1, keys);
1494 btree_split_insert_keys(as, trans, path, n2, keys);
1495 BUG_ON(!bch2_keylist_empty(keys));
1498 bch2_btree_build_aux_trees(n2);
1499 bch2_btree_build_aux_trees(n1);
1501 bch2_btree_update_add_new_node(as, n1);
1502 bch2_btree_update_add_new_node(as, n2);
1503 six_unlock_write(&n2->c.lock);
1504 six_unlock_write(&n1->c.lock);
1506 path1 = get_unlocked_mut_path(trans, path->btree_id, n1->c.level, n1->key.k.p);
1507 six_lock_increment(&n1->c.lock, SIX_LOCK_intent);
1508 mark_btree_node_locked(trans, path1, n1->c.level, BTREE_NODE_INTENT_LOCKED);
1509 bch2_btree_path_level_init(trans, path1, n1);
1511 path2 = get_unlocked_mut_path(trans, path->btree_id, n2->c.level, n2->key.k.p);
1512 six_lock_increment(&n2->c.lock, SIX_LOCK_intent);
1513 mark_btree_node_locked(trans, path2, n2->c.level, BTREE_NODE_INTENT_LOCKED);
1514 bch2_btree_path_level_init(trans, path2, n2);
1517 * Note that on recursive parent_keys == keys, so we
1518 * can't start adding new keys to parent_keys before emptying it
1519 * out (which we did with btree_split_insert_keys() above)
1521 bch2_keylist_add(&as->parent_keys, &n1->key);
1522 bch2_keylist_add(&as->parent_keys, &n2->key);
1525 /* Depth increases, make a new root */
1526 n3 = __btree_root_alloc(as, trans, b->c.level + 1);
1528 bch2_btree_update_add_new_node(as, n3);
1529 six_unlock_write(&n3->c.lock);
1531 path2->locks_want++;
1532 BUG_ON(btree_node_locked(path2, n3->c.level));
1533 six_lock_increment(&n3->c.lock, SIX_LOCK_intent);
1534 mark_btree_node_locked(trans, path2, n3->c.level, BTREE_NODE_INTENT_LOCKED);
1535 bch2_btree_path_level_init(trans, path2, n3);
1537 n3->sib_u64s[0] = U16_MAX;
1538 n3->sib_u64s[1] = U16_MAX;
1540 btree_split_insert_keys(as, trans, path, n3, &as->parent_keys);
1543 trace_and_count(c, btree_node_compact, c, b);
1545 n1 = bch2_btree_node_alloc_replacement(as, trans, b);
1548 btree_split_insert_keys(as, trans, path, n1, keys);
1549 BUG_ON(!bch2_keylist_empty(keys));
1552 bch2_btree_build_aux_trees(n1);
1553 bch2_btree_update_add_new_node(as, n1);
1554 six_unlock_write(&n1->c.lock);
1556 path1 = get_unlocked_mut_path(trans, path->btree_id, n1->c.level, n1->key.k.p);
1557 six_lock_increment(&n1->c.lock, SIX_LOCK_intent);
1558 mark_btree_node_locked(trans, path1, n1->c.level, BTREE_NODE_INTENT_LOCKED);
1559 bch2_btree_path_level_init(trans, path1, n1);
1562 bch2_keylist_add(&as->parent_keys, &n1->key);
1565 /* New nodes all written, now make them visible: */
1568 /* Split a non root node */
1569 ret = bch2_btree_insert_node(as, trans, path, parent, &as->parent_keys, flags);
1573 bch2_btree_set_root(as, trans, path, n3);
1575 /* Root filled up but didn't need to be split */
1576 bch2_btree_set_root(as, trans, path, n1);
1580 bch2_btree_update_get_open_buckets(as, n3);
1581 bch2_btree_node_write(c, n3, SIX_LOCK_intent, 0);
1584 bch2_btree_update_get_open_buckets(as, n2);
1585 bch2_btree_node_write(c, n2, SIX_LOCK_intent, 0);
1587 bch2_btree_update_get_open_buckets(as, n1);
1588 bch2_btree_node_write(c, n1, SIX_LOCK_intent, 0);
1591 * The old node must be freed (in memory) _before_ unlocking the new
1592 * nodes - else another thread could re-acquire a read lock on the old
1593 * node after another thread has locked and updated the new node, thus
1594 * seeing stale data:
1596 bch2_btree_node_free_inmem(trans, path, b);
1599 bch2_trans_node_add(trans, n3);
1601 bch2_trans_node_add(trans, n2);
1602 bch2_trans_node_add(trans, n1);
1605 six_unlock_intent(&n3->c.lock);
1607 six_unlock_intent(&n2->c.lock);
1608 six_unlock_intent(&n1->c.lock);
1611 __bch2_btree_path_unlock(trans, path2);
1612 bch2_path_put(trans, path2, true);
1615 __bch2_btree_path_unlock(trans, path1);
1616 bch2_path_put(trans, path1, true);
1619 bch2_trans_verify_locks(trans);
1621 bch2_time_stats_update(&c->times[n2
1622 ? BCH_TIME_btree_node_split
1623 : BCH_TIME_btree_node_compact],
1628 bch2_btree_node_free_never_used(as, trans, n3);
1630 bch2_btree_node_free_never_used(as, trans, n2);
1631 bch2_btree_node_free_never_used(as, trans, n1);
1636 bch2_btree_insert_keys_interior(struct btree_update *as,
1637 struct btree_trans *trans,
1638 struct btree_path *path,
1640 struct keylist *keys)
1642 struct btree_path *linked;
1644 __bch2_btree_insert_keys_interior(as, trans, path, b,
1645 path->l[b->c.level].iter, keys);
1647 btree_update_updated_node(as, b);
1649 trans_for_each_path_with_node(trans, b, linked)
1650 bch2_btree_node_iter_peek(&linked->l[b->c.level].iter, b);
1652 bch2_trans_verify_paths(trans);
1656 * bch2_btree_insert_node - insert bkeys into a given btree node
1658 * @as: btree_update object
1659 * @trans: btree_trans object
1660 * @path: path that points to current node
1661 * @b: node to insert keys into
1662 * @keys: list of keys to insert
1663 * @flags: transaction commit flags
1665 * Returns: 0 on success, typically transaction restart error on failure
1667 * Inserts as many keys as it can into a given btree node, splitting it if full.
1668 * If a split occurred, this function will return early. This can only happen
1669 * for leaf nodes -- inserts into interior nodes have to be atomic.
1671 static int bch2_btree_insert_node(struct btree_update *as, struct btree_trans *trans,
1672 struct btree_path *path, struct btree *b,
1673 struct keylist *keys, unsigned flags)
1675 struct bch_fs *c = as->c;
1676 int old_u64s = le16_to_cpu(btree_bset_last(b)->u64s);
1677 int old_live_u64s = b->nr.live_u64s;
1678 int live_u64s_added, u64s_added;
1681 lockdep_assert_held(&c->gc_lock);
1682 BUG_ON(!btree_node_intent_locked(path, b->c.level));
1683 BUG_ON(!b->c.level);
1684 BUG_ON(!as || as->b);
1685 bch2_verify_keylist_sorted(keys);
1687 ret = bch2_btree_node_lock_write(trans, path, &b->c);
1691 bch2_btree_node_prep_for_write(trans, path, b);
1693 if (!bch2_btree_node_insert_fits(c, b, bch2_keylist_u64s(keys))) {
1694 bch2_btree_node_unlock_write(trans, path, b);
1698 btree_node_interior_verify(c, b);
1700 bch2_btree_insert_keys_interior(as, trans, path, b, keys);
1702 live_u64s_added = (int) b->nr.live_u64s - old_live_u64s;
1703 u64s_added = (int) le16_to_cpu(btree_bset_last(b)->u64s) - old_u64s;
1705 if (b->sib_u64s[0] != U16_MAX && live_u64s_added < 0)
1706 b->sib_u64s[0] = max(0, (int) b->sib_u64s[0] + live_u64s_added);
1707 if (b->sib_u64s[1] != U16_MAX && live_u64s_added < 0)
1708 b->sib_u64s[1] = max(0, (int) b->sib_u64s[1] + live_u64s_added);
1710 if (u64s_added > live_u64s_added &&
1711 bch2_maybe_compact_whiteouts(c, b))
1712 bch2_trans_node_reinit_iter(trans, b);
1714 bch2_btree_node_unlock_write(trans, path, b);
1716 btree_node_interior_verify(c, b);
1720 * We could attempt to avoid the transaction restart, by calling
1721 * bch2_btree_path_upgrade() and allocating more nodes:
1723 if (b->c.level >= as->update_level) {
1724 trace_and_count(c, trans_restart_split_race, trans, _THIS_IP_, b);
1725 return btree_trans_restart(trans, BCH_ERR_transaction_restart_split_race);
1728 return btree_split(as, trans, path, b, keys, flags);
1731 int bch2_btree_split_leaf(struct btree_trans *trans,
1732 struct btree_path *path,
1735 struct btree *b = path_l(path)->b;
1736 struct btree_update *as;
1740 as = bch2_btree_update_start(trans, path, path->level,
1745 ret = btree_split(as, trans, path, b, NULL, flags);
1747 bch2_btree_update_free(as, trans);
1751 bch2_btree_update_done(as, trans);
1753 for (l = path->level + 1; btree_node_intent_locked(path, l) && !ret; l++)
1754 ret = bch2_foreground_maybe_merge(trans, path, l, flags);
1759 int __bch2_foreground_maybe_merge(struct btree_trans *trans,
1760 struct btree_path *path,
1763 enum btree_node_sibling sib)
1765 struct bch_fs *c = trans->c;
1766 struct btree_path *sib_path = NULL, *new_path = NULL;
1767 struct btree_update *as;
1768 struct bkey_format_state new_s;
1769 struct bkey_format new_f;
1770 struct bkey_i delete;
1771 struct btree *b, *m, *n, *prev, *next, *parent;
1772 struct bpos sib_pos;
1774 u64 start_time = local_clock();
1777 BUG_ON(!path->should_be_locked);
1778 BUG_ON(!btree_node_locked(path, level));
1780 b = path->l[level].b;
1782 if ((sib == btree_prev_sib && bpos_eq(b->data->min_key, POS_MIN)) ||
1783 (sib == btree_next_sib && bpos_eq(b->data->max_key, SPOS_MAX))) {
1784 b->sib_u64s[sib] = U16_MAX;
1788 sib_pos = sib == btree_prev_sib
1789 ? bpos_predecessor(b->data->min_key)
1790 : bpos_successor(b->data->max_key);
1792 sib_path = bch2_path_get(trans, path->btree_id, sib_pos,
1793 U8_MAX, level, BTREE_ITER_INTENT, _THIS_IP_);
1794 ret = bch2_btree_path_traverse(trans, sib_path, false);
1798 btree_path_set_should_be_locked(sib_path);
1800 m = sib_path->l[level].b;
1802 if (btree_node_parent(path, b) !=
1803 btree_node_parent(sib_path, m)) {
1804 b->sib_u64s[sib] = U16_MAX;
1808 if (sib == btree_prev_sib) {
1816 if (!bpos_eq(bpos_successor(prev->data->max_key), next->data->min_key)) {
1817 struct printbuf buf1 = PRINTBUF, buf2 = PRINTBUF;
1819 bch2_bpos_to_text(&buf1, prev->data->max_key);
1820 bch2_bpos_to_text(&buf2, next->data->min_key);
1822 "%s(): btree topology error:\n"
1823 " prev ends at %s\n"
1824 " next starts at %s",
1825 __func__, buf1.buf, buf2.buf);
1826 printbuf_exit(&buf1);
1827 printbuf_exit(&buf2);
1828 bch2_topology_error(c);
1833 bch2_bkey_format_init(&new_s);
1834 bch2_bkey_format_add_pos(&new_s, prev->data->min_key);
1835 __bch2_btree_calc_format(&new_s, prev);
1836 __bch2_btree_calc_format(&new_s, next);
1837 bch2_bkey_format_add_pos(&new_s, next->data->max_key);
1838 new_f = bch2_bkey_format_done(&new_s);
1840 sib_u64s = btree_node_u64s_with_format(b, &new_f) +
1841 btree_node_u64s_with_format(m, &new_f);
1843 if (sib_u64s > BTREE_FOREGROUND_MERGE_HYSTERESIS(c)) {
1844 sib_u64s -= BTREE_FOREGROUND_MERGE_HYSTERESIS(c);
1846 sib_u64s += BTREE_FOREGROUND_MERGE_HYSTERESIS(c);
1849 sib_u64s = min(sib_u64s, btree_max_u64s(c));
1850 sib_u64s = min(sib_u64s, (size_t) U16_MAX - 1);
1851 b->sib_u64s[sib] = sib_u64s;
1853 if (b->sib_u64s[sib] > c->btree_foreground_merge_threshold)
1856 parent = btree_node_parent(path, b);
1857 as = bch2_btree_update_start(trans, path, level, false,
1858 BTREE_INSERT_NOFAIL|flags);
1859 ret = PTR_ERR_OR_ZERO(as);
1863 trace_and_count(c, btree_node_merge, c, b);
1865 bch2_btree_interior_update_will_free_node(as, b);
1866 bch2_btree_interior_update_will_free_node(as, m);
1868 n = bch2_btree_node_alloc(as, trans, b->c.level);
1870 SET_BTREE_NODE_SEQ(n->data,
1871 max(BTREE_NODE_SEQ(b->data),
1872 BTREE_NODE_SEQ(m->data)) + 1);
1874 btree_set_min(n, prev->data->min_key);
1875 btree_set_max(n, next->data->max_key);
1877 n->data->format = new_f;
1878 btree_node_set_format(n, new_f);
1880 bch2_btree_sort_into(c, n, prev);
1881 bch2_btree_sort_into(c, n, next);
1883 bch2_btree_build_aux_trees(n);
1884 bch2_btree_update_add_new_node(as, n);
1885 six_unlock_write(&n->c.lock);
1887 new_path = get_unlocked_mut_path(trans, path->btree_id, n->c.level, n->key.k.p);
1888 six_lock_increment(&n->c.lock, SIX_LOCK_intent);
1889 mark_btree_node_locked(trans, new_path, n->c.level, BTREE_NODE_INTENT_LOCKED);
1890 bch2_btree_path_level_init(trans, new_path, n);
1892 bkey_init(&delete.k);
1893 delete.k.p = prev->key.k.p;
1894 bch2_keylist_add(&as->parent_keys, &delete);
1895 bch2_keylist_add(&as->parent_keys, &n->key);
1897 bch2_trans_verify_paths(trans);
1899 ret = bch2_btree_insert_node(as, trans, path, parent, &as->parent_keys, flags);
1901 goto err_free_update;
1903 bch2_trans_verify_paths(trans);
1905 bch2_btree_update_get_open_buckets(as, n);
1906 bch2_btree_node_write(c, n, SIX_LOCK_intent, 0);
1908 bch2_btree_node_free_inmem(trans, path, b);
1909 bch2_btree_node_free_inmem(trans, sib_path, m);
1911 bch2_trans_node_add(trans, n);
1913 bch2_trans_verify_paths(trans);
1915 six_unlock_intent(&n->c.lock);
1917 bch2_btree_update_done(as, trans);
1919 bch2_time_stats_update(&c->times[BCH_TIME_btree_node_merge], start_time);
1923 bch2_path_put(trans, new_path, true);
1924 bch2_path_put(trans, sib_path, true);
1925 bch2_trans_verify_locks(trans);
1928 bch2_btree_node_free_never_used(as, trans, n);
1929 bch2_btree_update_free(as, trans);
1933 int bch2_btree_node_rewrite(struct btree_trans *trans,
1934 struct btree_iter *iter,
1938 struct bch_fs *c = trans->c;
1939 struct btree_path *new_path = NULL;
1940 struct btree *n, *parent;
1941 struct btree_update *as;
1944 flags |= BTREE_INSERT_NOFAIL;
1946 parent = btree_node_parent(iter->path, b);
1947 as = bch2_btree_update_start(trans, iter->path, b->c.level,
1949 ret = PTR_ERR_OR_ZERO(as);
1953 bch2_btree_interior_update_will_free_node(as, b);
1955 n = bch2_btree_node_alloc_replacement(as, trans, b);
1957 bch2_btree_build_aux_trees(n);
1958 bch2_btree_update_add_new_node(as, n);
1959 six_unlock_write(&n->c.lock);
1961 new_path = get_unlocked_mut_path(trans, iter->btree_id, n->c.level, n->key.k.p);
1962 six_lock_increment(&n->c.lock, SIX_LOCK_intent);
1963 mark_btree_node_locked(trans, new_path, n->c.level, BTREE_NODE_INTENT_LOCKED);
1964 bch2_btree_path_level_init(trans, new_path, n);
1966 trace_and_count(c, btree_node_rewrite, c, b);
1969 bch2_keylist_add(&as->parent_keys, &n->key);
1970 ret = bch2_btree_insert_node(as, trans, iter->path, parent,
1971 &as->parent_keys, flags);
1975 bch2_btree_set_root(as, trans, iter->path, n);
1978 bch2_btree_update_get_open_buckets(as, n);
1979 bch2_btree_node_write(c, n, SIX_LOCK_intent, 0);
1981 bch2_btree_node_free_inmem(trans, iter->path, b);
1983 bch2_trans_node_add(trans, n);
1984 six_unlock_intent(&n->c.lock);
1986 bch2_btree_update_done(as, trans);
1989 bch2_path_put(trans, new_path, true);
1990 bch2_trans_downgrade(trans);
1993 bch2_btree_node_free_never_used(as, trans, n);
1994 bch2_btree_update_free(as, trans);
1998 struct async_btree_rewrite {
2000 struct work_struct work;
2001 struct list_head list;
2002 enum btree_id btree_id;
2008 static int async_btree_node_rewrite_trans(struct btree_trans *trans,
2009 struct async_btree_rewrite *a)
2011 struct bch_fs *c = trans->c;
2012 struct btree_iter iter;
2016 bch2_trans_node_iter_init(trans, &iter, a->btree_id, a->pos,
2017 BTREE_MAX_DEPTH, a->level, 0);
2018 b = bch2_btree_iter_peek_node(&iter);
2019 ret = PTR_ERR_OR_ZERO(b);
2023 if (!b || b->data->keys.seq != a->seq) {
2024 struct printbuf buf = PRINTBUF;
2027 bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&b->key));
2029 prt_str(&buf, "(null");
2030 bch_info(c, "%s: node to rewrite not found:, searching for seq %llu, got\n%s",
2031 __func__, a->seq, buf.buf);
2032 printbuf_exit(&buf);
2036 ret = bch2_btree_node_rewrite(trans, &iter, b, 0);
2038 bch2_trans_iter_exit(trans, &iter);
2043 static void async_btree_node_rewrite_work(struct work_struct *work)
2045 struct async_btree_rewrite *a =
2046 container_of(work, struct async_btree_rewrite, work);
2047 struct bch_fs *c = a->c;
2050 ret = bch2_trans_do(c, NULL, NULL, 0,
2051 async_btree_node_rewrite_trans(trans, a));
2054 bch2_write_ref_put(c, BCH_WRITE_REF_node_rewrite);
2058 void bch2_btree_node_rewrite_async(struct bch_fs *c, struct btree *b)
2060 struct async_btree_rewrite *a;
2063 a = kmalloc(sizeof(*a), GFP_NOFS);
2065 bch_err(c, "%s: error allocating memory", __func__);
2070 a->btree_id = b->c.btree_id;
2071 a->level = b->c.level;
2072 a->pos = b->key.k.p;
2073 a->seq = b->data->keys.seq;
2074 INIT_WORK(&a->work, async_btree_node_rewrite_work);
2076 if (unlikely(!test_bit(BCH_FS_MAY_GO_RW, &c->flags))) {
2077 mutex_lock(&c->pending_node_rewrites_lock);
2078 list_add(&a->list, &c->pending_node_rewrites);
2079 mutex_unlock(&c->pending_node_rewrites_lock);
2083 if (!bch2_write_ref_tryget(c, BCH_WRITE_REF_node_rewrite)) {
2084 if (test_bit(BCH_FS_STARTED, &c->flags)) {
2085 bch_err(c, "%s: error getting c->writes ref", __func__);
2090 ret = bch2_fs_read_write_early(c);
2092 bch_err_msg(c, ret, "going read-write");
2097 bch2_write_ref_get(c, BCH_WRITE_REF_node_rewrite);
2100 queue_work(c->btree_interior_update_worker, &a->work);
2103 void bch2_do_pending_node_rewrites(struct bch_fs *c)
2105 struct async_btree_rewrite *a, *n;
2107 mutex_lock(&c->pending_node_rewrites_lock);
2108 list_for_each_entry_safe(a, n, &c->pending_node_rewrites, list) {
2111 bch2_write_ref_get(c, BCH_WRITE_REF_node_rewrite);
2112 queue_work(c->btree_interior_update_worker, &a->work);
2114 mutex_unlock(&c->pending_node_rewrites_lock);
2117 void bch2_free_pending_node_rewrites(struct bch_fs *c)
2119 struct async_btree_rewrite *a, *n;
2121 mutex_lock(&c->pending_node_rewrites_lock);
2122 list_for_each_entry_safe(a, n, &c->pending_node_rewrites, list) {
2127 mutex_unlock(&c->pending_node_rewrites_lock);
2130 static int __bch2_btree_node_update_key(struct btree_trans *trans,
2131 struct btree_iter *iter,
2132 struct btree *b, struct btree *new_hash,
2133 struct bkey_i *new_key,
2134 unsigned commit_flags,
2137 struct bch_fs *c = trans->c;
2138 struct btree_iter iter2 = { NULL };
2139 struct btree *parent;
2142 if (!skip_triggers) {
2143 ret = bch2_trans_mark_old(trans, b->c.btree_id, b->c.level + 1,
2144 bkey_i_to_s_c(&b->key), 0);
2148 ret = bch2_trans_mark_new(trans, b->c.btree_id, b->c.level + 1,
2155 bkey_copy(&new_hash->key, new_key);
2156 ret = bch2_btree_node_hash_insert(&c->btree_cache,
2157 new_hash, b->c.level, b->c.btree_id);
2161 parent = btree_node_parent(iter->path, b);
2163 bch2_trans_copy_iter(&iter2, iter);
2165 iter2.path = bch2_btree_path_make_mut(trans, iter2.path,
2166 iter2.flags & BTREE_ITER_INTENT,
2169 BUG_ON(iter2.path->level != b->c.level);
2170 BUG_ON(!bpos_eq(iter2.path->pos, new_key->k.p));
2172 btree_path_set_level_up(trans, iter2.path);
2174 trans->paths_sorted = false;
2176 ret = bch2_btree_iter_traverse(&iter2) ?:
2177 bch2_trans_update(trans, &iter2, new_key, BTREE_TRIGGER_NORUN);
2181 BUG_ON(btree_node_root(c, b) != b);
2183 ret = darray_make_room(&trans->extra_journal_entries,
2184 jset_u64s(new_key->k.u64s));
2188 journal_entry_set((void *) &darray_top(trans->extra_journal_entries),
2189 BCH_JSET_ENTRY_btree_root,
2190 b->c.btree_id, b->c.level,
2191 new_key, new_key->k.u64s);
2192 trans->extra_journal_entries.nr += jset_u64s(new_key->k.u64s);
2195 ret = bch2_trans_commit(trans, NULL, NULL, commit_flags);
2199 bch2_btree_node_lock_write_nofail(trans, iter->path, &b->c);
2202 mutex_lock(&c->btree_cache.lock);
2203 bch2_btree_node_hash_remove(&c->btree_cache, new_hash);
2204 bch2_btree_node_hash_remove(&c->btree_cache, b);
2206 bkey_copy(&b->key, new_key);
2207 ret = __bch2_btree_node_hash_insert(&c->btree_cache, b);
2209 mutex_unlock(&c->btree_cache.lock);
2211 bkey_copy(&b->key, new_key);
2214 bch2_btree_node_unlock_write(trans, iter->path, b);
2216 bch2_trans_iter_exit(trans, &iter2);
2220 mutex_lock(&c->btree_cache.lock);
2221 bch2_btree_node_hash_remove(&c->btree_cache, b);
2222 mutex_unlock(&c->btree_cache.lock);
2227 int bch2_btree_node_update_key(struct btree_trans *trans, struct btree_iter *iter,
2228 struct btree *b, struct bkey_i *new_key,
2229 unsigned commit_flags, bool skip_triggers)
2231 struct bch_fs *c = trans->c;
2232 struct btree *new_hash = NULL;
2233 struct btree_path *path = iter->path;
2237 ret = bch2_btree_path_upgrade(trans, path, b->c.level + 1);
2241 closure_init_stack(&cl);
2244 * check btree_ptr_hash_val() after @b is locked by
2245 * btree_iter_traverse():
2247 if (btree_ptr_hash_val(new_key) != b->hash_val) {
2248 ret = bch2_btree_cache_cannibalize_lock(c, &cl);
2250 ret = drop_locks_do(trans, (closure_sync(&cl), 0));
2255 new_hash = bch2_btree_node_mem_alloc(trans, false);
2259 ret = __bch2_btree_node_update_key(trans, iter, b, new_hash, new_key,
2260 commit_flags, skip_triggers);
2264 mutex_lock(&c->btree_cache.lock);
2265 list_move(&new_hash->list, &c->btree_cache.freeable);
2266 mutex_unlock(&c->btree_cache.lock);
2268 six_unlock_write(&new_hash->c.lock);
2269 six_unlock_intent(&new_hash->c.lock);
2272 bch2_btree_cache_cannibalize_unlock(c);
2276 int bch2_btree_node_update_key_get_iter(struct btree_trans *trans,
2277 struct btree *b, struct bkey_i *new_key,
2278 unsigned commit_flags, bool skip_triggers)
2280 struct btree_iter iter;
2283 bch2_trans_node_iter_init(trans, &iter, b->c.btree_id, b->key.k.p,
2284 BTREE_MAX_DEPTH, b->c.level,
2286 ret = bch2_btree_iter_traverse(&iter);
2290 /* has node been freed? */
2291 if (iter.path->l[b->c.level].b != b) {
2292 /* node has been freed: */
2293 BUG_ON(!btree_node_dying(b));
2297 BUG_ON(!btree_node_hashed(b));
2299 ret = bch2_btree_node_update_key(trans, &iter, b, new_key,
2300 commit_flags, skip_triggers);
2302 bch2_trans_iter_exit(trans, &iter);
2309 * Only for filesystem bringup, when first reading the btree roots or allocating
2310 * btree roots when initializing a new filesystem:
2312 void bch2_btree_set_root_for_read(struct bch_fs *c, struct btree *b)
2314 BUG_ON(btree_node_root(c, b));
2316 bch2_btree_set_root_inmem(c, b);
2319 static int __bch2_btree_root_alloc(struct btree_trans *trans, enum btree_id id)
2321 struct bch_fs *c = trans->c;
2326 closure_init_stack(&cl);
2329 ret = bch2_btree_cache_cannibalize_lock(c, &cl);
2333 b = bch2_btree_node_mem_alloc(trans, false);
2334 bch2_btree_cache_cannibalize_unlock(c);
2336 set_btree_node_fake(b);
2337 set_btree_node_need_rewrite(b);
2341 bkey_btree_ptr_init(&b->key);
2342 b->key.k.p = SPOS_MAX;
2343 *((u64 *) bkey_i_to_btree_ptr(&b->key)->v.start) = U64_MAX - id;
2345 bch2_bset_init_first(b, &b->data->keys);
2346 bch2_btree_build_aux_trees(b);
2349 btree_set_min(b, POS_MIN);
2350 btree_set_max(b, SPOS_MAX);
2351 b->data->format = bch2_btree_calc_format(b);
2352 btree_node_set_format(b, b->data->format);
2354 ret = bch2_btree_node_hash_insert(&c->btree_cache, b,
2355 b->c.level, b->c.btree_id);
2358 bch2_btree_set_root_inmem(c, b);
2360 six_unlock_write(&b->c.lock);
2361 six_unlock_intent(&b->c.lock);
2365 void bch2_btree_root_alloc(struct bch_fs *c, enum btree_id id)
2367 bch2_trans_run(c, __bch2_btree_root_alloc(trans, id));
2370 void bch2_btree_updates_to_text(struct printbuf *out, struct bch_fs *c)
2372 struct btree_update *as;
2374 mutex_lock(&c->btree_interior_update_lock);
2375 list_for_each_entry(as, &c->btree_interior_update_list, list)
2376 prt_printf(out, "%p m %u w %u r %u j %llu\n",
2380 closure_nr_remaining(&as->cl),
2382 mutex_unlock(&c->btree_interior_update_lock);
2385 static bool bch2_btree_interior_updates_pending(struct bch_fs *c)
2389 mutex_lock(&c->btree_interior_update_lock);
2390 ret = !list_empty(&c->btree_interior_update_list);
2391 mutex_unlock(&c->btree_interior_update_lock);
2396 bool bch2_btree_interior_updates_flush(struct bch_fs *c)
2398 bool ret = bch2_btree_interior_updates_pending(c);
2401 closure_wait_event(&c->btree_interior_update_wait,
2402 !bch2_btree_interior_updates_pending(c));
2406 void bch2_journal_entry_to_btree_root(struct bch_fs *c, struct jset_entry *entry)
2408 struct btree_root *r = bch2_btree_id_root(c, entry->btree_id);
2410 mutex_lock(&c->btree_root_lock);
2412 r->level = entry->level;
2414 bkey_copy(&r->key, (struct bkey_i *) entry->start);
2416 mutex_unlock(&c->btree_root_lock);
2420 bch2_btree_roots_to_journal_entries(struct bch_fs *c,
2421 struct jset_entry *start,
2422 struct jset_entry *end)
2424 struct jset_entry *entry;
2425 unsigned long have = 0;
2428 for (entry = start; entry < end; entry = vstruct_next(entry))
2429 if (entry->type == BCH_JSET_ENTRY_btree_root)
2430 __set_bit(entry->btree_id, &have);
2432 mutex_lock(&c->btree_root_lock);
2434 for (i = 0; i < btree_id_nr_alive(c); i++) {
2435 struct btree_root *r = bch2_btree_id_root(c, i);
2437 if (r->alive && !test_bit(i, &have)) {
2438 journal_entry_set(end, BCH_JSET_ENTRY_btree_root,
2439 i, r->level, &r->key, r->key.k.u64s);
2440 end = vstruct_next(end);
2444 mutex_unlock(&c->btree_root_lock);
2449 void bch2_fs_btree_interior_update_exit(struct bch_fs *c)
2451 if (c->btree_interior_update_worker)
2452 destroy_workqueue(c->btree_interior_update_worker);
2453 mempool_exit(&c->btree_interior_update_pool);
2456 void bch2_fs_btree_interior_update_init_early(struct bch_fs *c)
2458 mutex_init(&c->btree_reserve_cache_lock);
2459 INIT_LIST_HEAD(&c->btree_interior_update_list);
2460 INIT_LIST_HEAD(&c->btree_interior_updates_unwritten);
2461 mutex_init(&c->btree_interior_update_lock);
2462 INIT_WORK(&c->btree_interior_update_work, btree_interior_update_work);
2464 INIT_LIST_HEAD(&c->pending_node_rewrites);
2465 mutex_init(&c->pending_node_rewrites_lock);
2468 int bch2_fs_btree_interior_update_init(struct bch_fs *c)
2470 c->btree_interior_update_worker =
2471 alloc_workqueue("btree_update", WQ_UNBOUND|WQ_MEM_RECLAIM, 1);
2472 if (!c->btree_interior_update_worker)
2473 return -BCH_ERR_ENOMEM_btree_interior_update_worker_init;
2475 if (mempool_init_kmalloc_pool(&c->btree_interior_update_pool, 1,
2476 sizeof(struct btree_update)))
2477 return -BCH_ERR_ENOMEM_btree_interior_update_pool_init;