1 // SPDX-License-Identifier: GPL-2.0
5 #include "btree_key_cache.h"
6 #include "btree_update.h"
13 #include <linux/random.h>
18 * Keys in BTREE_ID_snapshot_trees identify a whole tree of snapshot nodes; they
19 * exist to provide a stable identifier for the whole lifetime of a snapshot
23 void bch2_snapshot_tree_to_text(struct printbuf *out, struct bch_fs *c,
26 struct bkey_s_c_snapshot_tree t = bkey_s_c_to_snapshot_tree(k);
28 prt_printf(out, "subvol %u root snapshot %u",
29 le32_to_cpu(t.v->master_subvol),
30 le32_to_cpu(t.v->root_snapshot));
33 int bch2_snapshot_tree_invalid(const struct bch_fs *c, struct bkey_s_c k,
34 enum bkey_invalid_flags flags,
37 if (bkey_gt(k.k->p, POS(0, U32_MAX)) ||
38 bkey_lt(k.k->p, POS(0, 1))) {
39 prt_printf(err, "bad pos");
40 return -BCH_ERR_invalid_bkey;
46 int bch2_snapshot_tree_lookup(struct btree_trans *trans, u32 id,
47 struct bch_snapshot_tree *s)
49 int ret = bch2_bkey_get_val_typed(trans, BTREE_ID_snapshot_trees, POS(0, id),
50 BTREE_ITER_WITH_UPDATES, snapshot_tree, s);
52 if (bch2_err_matches(ret, ENOENT))
53 ret = -BCH_ERR_ENOENT_snapshot_tree;
57 struct bkey_i_snapshot_tree *
58 __bch2_snapshot_tree_create(struct btree_trans *trans)
60 struct btree_iter iter;
61 int ret = bch2_bkey_get_empty_slot(trans, &iter,
62 BTREE_ID_snapshot_trees, POS(0, U32_MAX));
63 struct bkey_i_snapshot_tree *s_t;
65 if (ret == -BCH_ERR_ENOSPC_btree_slot)
66 ret = -BCH_ERR_ENOSPC_snapshot_tree;
70 s_t = bch2_bkey_alloc(trans, &iter, 0, snapshot_tree);
71 ret = PTR_ERR_OR_ZERO(s_t);
72 bch2_trans_iter_exit(trans, &iter);
73 return ret ? ERR_PTR(ret) : s_t;
76 static int bch2_snapshot_tree_create(struct btree_trans *trans,
77 u32 root_id, u32 subvol_id, u32 *tree_id)
79 struct bkey_i_snapshot_tree *n_tree =
80 __bch2_snapshot_tree_create(trans);
83 return PTR_ERR(n_tree);
85 n_tree->v.master_subvol = cpu_to_le32(subvol_id);
86 n_tree->v.root_snapshot = cpu_to_le32(root_id);
87 *tree_id = n_tree->k.p.offset;
93 static bool bch2_snapshot_is_ancestor_early(struct bch_fs *c, u32 id, u32 ancestor)
95 struct snapshot_table *t;
98 t = rcu_dereference(c->snapshots);
100 while (id && id < ancestor)
101 id = __snapshot_t(t, id)->parent;
104 return id == ancestor;
107 static inline u32 get_ancestor_below(struct snapshot_table *t, u32 id, u32 ancestor)
109 const struct snapshot_t *s = __snapshot_t(t, id);
111 if (s->skip[2] <= ancestor)
113 if (s->skip[1] <= ancestor)
115 if (s->skip[0] <= ancestor)
120 bool __bch2_snapshot_is_ancestor(struct bch_fs *c, u32 id, u32 ancestor)
122 struct snapshot_table *t;
125 EBUG_ON(c->curr_recovery_pass <= BCH_RECOVERY_PASS_check_snapshots);
128 t = rcu_dereference(c->snapshots);
130 while (id && id < ancestor - IS_ANCESTOR_BITMAP)
131 id = get_ancestor_below(t, id, ancestor);
133 if (id && id < ancestor) {
134 ret = test_bit(ancestor - id - 1, __snapshot_t(t, id)->is_ancestor);
136 EBUG_ON(ret != bch2_snapshot_is_ancestor_early(c, id, ancestor));
138 ret = id == ancestor;
146 static noinline struct snapshot_t *__snapshot_t_mut(struct bch_fs *c, u32 id)
148 size_t idx = U32_MAX - id;
150 struct snapshot_table *new, *old;
152 new_size = max(16UL, roundup_pow_of_two(idx + 1));
154 new = kvzalloc(struct_size(new, s, new_size), GFP_KERNEL);
158 old = rcu_dereference_protected(c->snapshots, true);
161 rcu_dereference_protected(c->snapshots, true)->s,
162 sizeof(new->s[0]) * c->snapshot_table_size);
164 rcu_assign_pointer(c->snapshots, new);
165 c->snapshot_table_size = new_size;
166 kvfree_rcu_mightsleep(old);
168 return &rcu_dereference_protected(c->snapshots, true)->s[idx];
171 static inline struct snapshot_t *snapshot_t_mut(struct bch_fs *c, u32 id)
173 size_t idx = U32_MAX - id;
175 lockdep_assert_held(&c->snapshot_table_lock);
177 if (likely(idx < c->snapshot_table_size))
178 return &rcu_dereference_protected(c->snapshots, true)->s[idx];
180 return __snapshot_t_mut(c, id);
183 void bch2_snapshot_to_text(struct printbuf *out, struct bch_fs *c,
186 struct bkey_s_c_snapshot s = bkey_s_c_to_snapshot(k);
188 prt_printf(out, "is_subvol %llu deleted %llu parent %10u children %10u %10u subvol %u tree %u",
189 BCH_SNAPSHOT_SUBVOL(s.v),
190 BCH_SNAPSHOT_DELETED(s.v),
191 le32_to_cpu(s.v->parent),
192 le32_to_cpu(s.v->children[0]),
193 le32_to_cpu(s.v->children[1]),
194 le32_to_cpu(s.v->subvol),
195 le32_to_cpu(s.v->tree));
197 if (bkey_val_bytes(k.k) > offsetof(struct bch_snapshot, depth))
198 prt_printf(out, " depth %u skiplist %u %u %u",
199 le32_to_cpu(s.v->depth),
200 le32_to_cpu(s.v->skip[0]),
201 le32_to_cpu(s.v->skip[1]),
202 le32_to_cpu(s.v->skip[2]));
205 int bch2_snapshot_invalid(const struct bch_fs *c, struct bkey_s_c k,
206 enum bkey_invalid_flags flags,
207 struct printbuf *err)
209 struct bkey_s_c_snapshot s;
212 if (bkey_gt(k.k->p, POS(0, U32_MAX)) ||
213 bkey_lt(k.k->p, POS(0, 1))) {
214 prt_printf(err, "bad pos");
215 return -BCH_ERR_invalid_bkey;
218 s = bkey_s_c_to_snapshot(k);
220 id = le32_to_cpu(s.v->parent);
221 if (id && id <= k.k->p.offset) {
222 prt_printf(err, "bad parent node (%u <= %llu)",
224 return -BCH_ERR_invalid_bkey;
227 if (le32_to_cpu(s.v->children[0]) < le32_to_cpu(s.v->children[1])) {
228 prt_printf(err, "children not normalized");
229 return -BCH_ERR_invalid_bkey;
232 if (s.v->children[0] &&
233 s.v->children[0] == s.v->children[1]) {
234 prt_printf(err, "duplicate child nodes");
235 return -BCH_ERR_invalid_bkey;
238 for (i = 0; i < 2; i++) {
239 id = le32_to_cpu(s.v->children[i]);
241 if (id >= k.k->p.offset) {
242 prt_printf(err, "bad child node (%u >= %llu)",
244 return -BCH_ERR_invalid_bkey;
248 if (bkey_val_bytes(k.k) > offsetof(struct bch_snapshot, skip)) {
249 if (le32_to_cpu(s.v->skip[0]) > le32_to_cpu(s.v->skip[1]) ||
250 le32_to_cpu(s.v->skip[1]) > le32_to_cpu(s.v->skip[2])) {
251 prt_printf(err, "skiplist not normalized");
252 return -BCH_ERR_invalid_bkey;
255 for (i = 0; i < ARRAY_SIZE(s.v->skip); i++) {
256 id = le32_to_cpu(s.v->skip[i]);
258 if ((id && !s.v->parent) ||
259 (id && id <= k.k->p.offset)) {
260 prt_printf(err, "bad skiplist node %u", id);
261 return -BCH_ERR_invalid_bkey;
269 static void __set_is_ancestor_bitmap(struct bch_fs *c, u32 id)
271 struct snapshot_t *t = snapshot_t_mut(c, id);
274 while ((parent = bch2_snapshot_parent_early(c, parent)) &&
275 parent - id - 1 < IS_ANCESTOR_BITMAP)
276 __set_bit(parent - id - 1, t->is_ancestor);
279 static void set_is_ancestor_bitmap(struct bch_fs *c, u32 id)
281 mutex_lock(&c->snapshot_table_lock);
282 __set_is_ancestor_bitmap(c, id);
283 mutex_unlock(&c->snapshot_table_lock);
286 int bch2_mark_snapshot(struct btree_trans *trans,
287 enum btree_id btree, unsigned level,
288 struct bkey_s_c old, struct bkey_s_c new,
291 struct bch_fs *c = trans->c;
292 struct snapshot_t *t;
293 u32 id = new.k->p.offset;
296 mutex_lock(&c->snapshot_table_lock);
298 t = snapshot_t_mut(c, id);
300 ret = -BCH_ERR_ENOMEM_mark_snapshot;
304 if (new.k->type == KEY_TYPE_snapshot) {
305 struct bkey_s_c_snapshot s = bkey_s_c_to_snapshot(new);
307 t->parent = le32_to_cpu(s.v->parent);
308 t->children[0] = le32_to_cpu(s.v->children[0]);
309 t->children[1] = le32_to_cpu(s.v->children[1]);
310 t->subvol = BCH_SNAPSHOT_SUBVOL(s.v) ? le32_to_cpu(s.v->subvol) : 0;
311 t->tree = le32_to_cpu(s.v->tree);
313 if (bkey_val_bytes(s.k) > offsetof(struct bch_snapshot, depth)) {
314 t->depth = le32_to_cpu(s.v->depth);
315 t->skip[0] = le32_to_cpu(s.v->skip[0]);
316 t->skip[1] = le32_to_cpu(s.v->skip[1]);
317 t->skip[2] = le32_to_cpu(s.v->skip[2]);
325 __set_is_ancestor_bitmap(c, id);
327 if (BCH_SNAPSHOT_DELETED(s.v)) {
328 set_bit(BCH_FS_HAVE_DELETED_SNAPSHOTS, &c->flags);
329 c->recovery_passes_explicit |= BIT_ULL(BCH_RECOVERY_PASS_delete_dead_snapshots);
332 memset(t, 0, sizeof(*t));
335 mutex_unlock(&c->snapshot_table_lock);
339 int bch2_snapshot_lookup(struct btree_trans *trans, u32 id,
340 struct bch_snapshot *s)
342 return bch2_bkey_get_val_typed(trans, BTREE_ID_snapshots, POS(0, id),
343 BTREE_ITER_WITH_UPDATES, snapshot, s);
346 static int bch2_snapshot_live(struct btree_trans *trans, u32 id)
348 struct bch_snapshot v;
354 ret = bch2_snapshot_lookup(trans, id, &v);
355 if (bch2_err_matches(ret, ENOENT))
356 bch_err(trans->c, "snapshot node %u not found", id);
360 return !BCH_SNAPSHOT_DELETED(&v);
364 * If @k is a snapshot with just one live child, it's part of a linear chain,
365 * which we consider to be an equivalence class: and then after snapshot
366 * deletion cleanup, there should only be a single key at a given position in
367 * this equivalence class.
369 * This sets the equivalence class of @k to be the child's equivalence class, if
370 * it's part of such a linear chain: this correctly sets equivalence classes on
371 * startup if we run leaf to root (i.e. in natural key order).
373 static int bch2_snapshot_set_equiv(struct btree_trans *trans, struct bkey_s_c k)
375 struct bch_fs *c = trans->c;
376 unsigned i, nr_live = 0, live_idx = 0;
377 struct bkey_s_c_snapshot snap;
378 u32 id = k.k->p.offset, child[2];
380 if (k.k->type != KEY_TYPE_snapshot)
383 snap = bkey_s_c_to_snapshot(k);
385 child[0] = le32_to_cpu(snap.v->children[0]);
386 child[1] = le32_to_cpu(snap.v->children[1]);
388 for (i = 0; i < 2; i++) {
389 int ret = bch2_snapshot_live(trans, child[i]);
399 mutex_lock(&c->snapshot_table_lock);
401 snapshot_t_mut(c, id)->equiv = nr_live == 1
402 ? snapshot_t_mut(c, child[live_idx])->equiv
405 mutex_unlock(&c->snapshot_table_lock);
412 static u32 bch2_snapshot_child(struct bch_fs *c, u32 id, unsigned child)
414 return snapshot_t(c, id)->children[child];
417 static u32 bch2_snapshot_left_child(struct bch_fs *c, u32 id)
419 return bch2_snapshot_child(c, id, 0);
422 static u32 bch2_snapshot_right_child(struct bch_fs *c, u32 id)
424 return bch2_snapshot_child(c, id, 1);
427 static u32 bch2_snapshot_tree_next(struct bch_fs *c, u32 id)
431 n = bch2_snapshot_left_child(c, id);
435 while ((parent = bch2_snapshot_parent(c, id))) {
436 n = bch2_snapshot_right_child(c, parent);
445 static u32 bch2_snapshot_tree_oldest_subvol(struct bch_fs *c, u32 snapshot_root)
447 u32 id = snapshot_root;
451 s = snapshot_t(c, id)->subvol;
453 if (s && (!subvol || s < subvol))
456 id = bch2_snapshot_tree_next(c, id);
462 static int bch2_snapshot_tree_master_subvol(struct btree_trans *trans,
463 u32 snapshot_root, u32 *subvol_id)
465 struct bch_fs *c = trans->c;
466 struct btree_iter iter;
468 struct bkey_s_c_subvolume s;
472 for_each_btree_key_norestart(trans, iter, BTREE_ID_subvolumes, POS_MIN,
474 if (k.k->type != KEY_TYPE_subvolume)
477 s = bkey_s_c_to_subvolume(k);
478 if (!bch2_snapshot_is_ancestor(c, le32_to_cpu(s.v->snapshot), snapshot_root))
480 if (!BCH_SUBVOLUME_SNAP(s.v)) {
481 *subvol_id = s.k->p.offset;
487 bch2_trans_iter_exit(trans, &iter);
489 if (!ret && !found) {
490 struct bkey_i_subvolume *u;
492 *subvol_id = bch2_snapshot_tree_oldest_subvol(c, snapshot_root);
494 u = bch2_bkey_get_mut_typed(trans, &iter,
495 BTREE_ID_subvolumes, POS(0, *subvol_id),
497 ret = PTR_ERR_OR_ZERO(u);
501 SET_BCH_SUBVOLUME_SNAP(&u->v, false);
507 static int check_snapshot_tree(struct btree_trans *trans,
508 struct btree_iter *iter,
511 struct bch_fs *c = trans->c;
512 struct bkey_s_c_snapshot_tree st;
513 struct bch_snapshot s;
514 struct bch_subvolume subvol;
515 struct printbuf buf = PRINTBUF;
519 if (k.k->type != KEY_TYPE_snapshot_tree)
522 st = bkey_s_c_to_snapshot_tree(k);
523 root_id = le32_to_cpu(st.v->root_snapshot);
525 ret = bch2_snapshot_lookup(trans, root_id, &s);
526 if (ret && !bch2_err_matches(ret, ENOENT))
529 if (fsck_err_on(ret ||
530 root_id != bch2_snapshot_root(c, root_id) ||
531 st.k->p.offset != le32_to_cpu(s.tree),
533 "snapshot tree points to missing/incorrect snapshot:\n %s",
534 (bch2_bkey_val_to_text(&buf, c, st.s_c), buf.buf))) {
535 ret = bch2_btree_delete_at(trans, iter, 0);
539 ret = bch2_subvolume_get(trans, le32_to_cpu(st.v->master_subvol),
541 if (ret && !bch2_err_matches(ret, ENOENT))
544 if (fsck_err_on(ret, c,
545 "snapshot tree points to missing subvolume:\n %s",
546 (printbuf_reset(&buf),
547 bch2_bkey_val_to_text(&buf, c, st.s_c), buf.buf)) ||
548 fsck_err_on(!bch2_snapshot_is_ancestor_early(c,
549 le32_to_cpu(subvol.snapshot),
551 "snapshot tree points to subvolume that does not point to snapshot in this tree:\n %s",
552 (printbuf_reset(&buf),
553 bch2_bkey_val_to_text(&buf, c, st.s_c), buf.buf)) ||
554 fsck_err_on(BCH_SUBVOLUME_SNAP(&subvol), c,
555 "snapshot tree points to snapshot subvolume:\n %s",
556 (printbuf_reset(&buf),
557 bch2_bkey_val_to_text(&buf, c, st.s_c), buf.buf))) {
558 struct bkey_i_snapshot_tree *u;
561 ret = bch2_snapshot_tree_master_subvol(trans, root_id, &subvol_id);
565 u = bch2_bkey_make_mut_typed(trans, iter, &k, 0, snapshot_tree);
566 ret = PTR_ERR_OR_ZERO(u);
570 u->v.master_subvol = cpu_to_le32(subvol_id);
571 st = snapshot_tree_i_to_s_c(u);
580 * For each snapshot_tree, make sure it points to the root of a snapshot tree
581 * and that snapshot entry points back to it, or delete it.
583 * And, make sure it points to a subvolume within that snapshot tree, or correct
584 * it to point to the oldest subvolume within that snapshot tree.
586 int bch2_check_snapshot_trees(struct bch_fs *c)
588 struct btree_iter iter;
592 ret = bch2_trans_run(c,
593 for_each_btree_key_commit(trans, iter,
594 BTREE_ID_snapshot_trees, POS_MIN,
595 BTREE_ITER_PREFETCH, k,
596 NULL, NULL, BTREE_INSERT_LAZY_RW|BTREE_INSERT_NOFAIL,
597 check_snapshot_tree(trans, &iter, k)));
600 bch_err(c, "error %i checking snapshot trees", ret);
605 * Look up snapshot tree for @tree_id and find root,
606 * make sure @snap_id is a descendent:
608 static int snapshot_tree_ptr_good(struct btree_trans *trans,
609 u32 snap_id, u32 tree_id)
611 struct bch_snapshot_tree s_t;
612 int ret = bch2_snapshot_tree_lookup(trans, tree_id, &s_t);
614 if (bch2_err_matches(ret, ENOENT))
619 return bch2_snapshot_is_ancestor_early(trans->c, snap_id, le32_to_cpu(s_t.root_snapshot));
622 u32 bch2_snapshot_skiplist_get(struct bch_fs *c, u32 id)
624 const struct snapshot_t *s;
630 s = snapshot_t(c, id);
632 id = bch2_snapshot_nth_parent(c, id, get_random_u32_below(s->depth));
638 static int snapshot_skiplist_good(struct btree_trans *trans, u32 id, struct bch_snapshot s)
642 for (i = 0; i < 3; i++)
647 if (!bch2_snapshot_is_ancestor_early(trans->c, id, le32_to_cpu(s.skip[i])))
655 * snapshot_tree pointer was incorrect: look up root snapshot node, make sure
656 * its snapshot_tree pointer is correct (allocate new one if necessary), then
657 * update this node's pointer to root node's pointer:
659 static int snapshot_tree_ptr_repair(struct btree_trans *trans,
660 struct btree_iter *iter,
662 struct bch_snapshot *s)
664 struct bch_fs *c = trans->c;
665 struct btree_iter root_iter;
666 struct bch_snapshot_tree s_t;
667 struct bkey_s_c_snapshot root;
668 struct bkey_i_snapshot *u;
669 u32 root_id = bch2_snapshot_root(c, k.k->p.offset), tree_id;
672 root = bch2_bkey_get_iter_typed(trans, &root_iter,
673 BTREE_ID_snapshots, POS(0, root_id),
674 BTREE_ITER_WITH_UPDATES, snapshot);
675 ret = bkey_err(root);
679 tree_id = le32_to_cpu(root.v->tree);
681 ret = bch2_snapshot_tree_lookup(trans, tree_id, &s_t);
682 if (ret && !bch2_err_matches(ret, ENOENT))
685 if (ret || le32_to_cpu(s_t.root_snapshot) != root_id) {
686 u = bch2_bkey_make_mut_typed(trans, &root_iter, &root.s_c, 0, snapshot);
687 ret = PTR_ERR_OR_ZERO(u) ?:
688 bch2_snapshot_tree_create(trans, root_id,
689 bch2_snapshot_tree_oldest_subvol(c, root_id),
694 u->v.tree = cpu_to_le32(tree_id);
695 if (k.k->p.offset == root_id)
699 if (k.k->p.offset != root_id) {
700 u = bch2_bkey_make_mut_typed(trans, iter, &k, 0, snapshot);
701 ret = PTR_ERR_OR_ZERO(u);
705 u->v.tree = cpu_to_le32(tree_id);
709 bch2_trans_iter_exit(trans, &root_iter);
713 static int check_snapshot(struct btree_trans *trans,
714 struct btree_iter *iter,
717 struct bch_fs *c = trans->c;
718 struct bch_snapshot s;
719 struct bch_subvolume subvol;
720 struct bch_snapshot v;
721 struct bkey_i_snapshot *u;
722 u32 parent_id = bch2_snapshot_parent_early(c, k.k->p.offset);
724 struct printbuf buf = PRINTBUF;
725 bool should_have_subvol;
729 if (k.k->type != KEY_TYPE_snapshot)
732 memset(&s, 0, sizeof(s));
733 memcpy(&s, k.v, bkey_val_bytes(k.k));
735 id = le32_to_cpu(s.parent);
737 ret = bch2_snapshot_lookup(trans, id, &v);
738 if (bch2_err_matches(ret, ENOENT))
739 bch_err(c, "snapshot with nonexistent parent:\n %s",
740 (bch2_bkey_val_to_text(&buf, c, k), buf.buf));
744 if (le32_to_cpu(v.children[0]) != k.k->p.offset &&
745 le32_to_cpu(v.children[1]) != k.k->p.offset) {
746 bch_err(c, "snapshot parent %u missing pointer to child %llu",
753 for (i = 0; i < 2 && s.children[i]; i++) {
754 id = le32_to_cpu(s.children[i]);
756 ret = bch2_snapshot_lookup(trans, id, &v);
757 if (bch2_err_matches(ret, ENOENT))
758 bch_err(c, "snapshot node %llu has nonexistent child %u",
763 if (le32_to_cpu(v.parent) != k.k->p.offset) {
764 bch_err(c, "snapshot child %u has wrong parent (got %u should be %llu)",
765 id, le32_to_cpu(v.parent), k.k->p.offset);
771 should_have_subvol = BCH_SNAPSHOT_SUBVOL(&s) &&
772 !BCH_SNAPSHOT_DELETED(&s);
774 if (should_have_subvol) {
775 id = le32_to_cpu(s.subvol);
776 ret = bch2_subvolume_get(trans, id, 0, false, &subvol);
777 if (bch2_err_matches(ret, ENOENT))
778 bch_err(c, "snapshot points to nonexistent subvolume:\n %s",
779 (bch2_bkey_val_to_text(&buf, c, k), buf.buf));
783 if (BCH_SNAPSHOT_SUBVOL(&s) != (le32_to_cpu(subvol.snapshot) == k.k->p.offset)) {
784 bch_err(c, "snapshot node %llu has wrong BCH_SNAPSHOT_SUBVOL",
790 if (fsck_err_on(s.subvol, c, "snapshot should not point to subvol:\n %s",
791 (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
792 u = bch2_bkey_make_mut_typed(trans, iter, &k, 0, snapshot);
793 ret = PTR_ERR_OR_ZERO(u);
802 ret = snapshot_tree_ptr_good(trans, k.k->p.offset, le32_to_cpu(s.tree));
806 if (fsck_err_on(!ret, c, "snapshot points to missing/incorrect tree:\n %s",
807 (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
808 ret = snapshot_tree_ptr_repair(trans, iter, k, &s);
814 real_depth = bch2_snapshot_depth(c, parent_id);
816 if (le32_to_cpu(s.depth) != real_depth &&
817 (c->sb.version_upgrade_complete < bcachefs_metadata_version_snapshot_skiplists ||
818 fsck_err(c, "snapshot with incorrect depth field, should be %u:\n %s",
819 real_depth, (bch2_bkey_val_to_text(&buf, c, k), buf.buf)))) {
820 u = bch2_bkey_make_mut_typed(trans, iter, &k, 0, snapshot);
821 ret = PTR_ERR_OR_ZERO(u);
825 u->v.depth = cpu_to_le32(real_depth);
829 ret = snapshot_skiplist_good(trans, k.k->p.offset, s);
834 (c->sb.version_upgrade_complete < bcachefs_metadata_version_snapshot_skiplists ||
835 fsck_err(c, "snapshot with bad skiplist field:\n %s",
836 (bch2_bkey_val_to_text(&buf, c, k), buf.buf)))) {
837 u = bch2_bkey_make_mut_typed(trans, iter, &k, 0, snapshot);
838 ret = PTR_ERR_OR_ZERO(u);
842 for (i = 0; i < ARRAY_SIZE(u->v.skip); i++)
843 u->v.skip[i] = cpu_to_le32(bch2_snapshot_skiplist_get(c, parent_id));
845 bubble_sort(u->v.skip, ARRAY_SIZE(u->v.skip), cmp_le32);
855 int bch2_check_snapshots(struct bch_fs *c)
857 struct btree_iter iter;
862 * We iterate backwards as checking/fixing the depth field requires that
863 * the parent's depth already be correct:
865 ret = bch2_trans_run(c,
866 for_each_btree_key_reverse_commit(trans, iter,
867 BTREE_ID_snapshots, POS_MAX,
868 BTREE_ITER_PREFETCH, k,
869 NULL, NULL, BTREE_INSERT_LAZY_RW|BTREE_INSERT_NOFAIL,
870 check_snapshot(trans, &iter, k)));
877 * Mark a snapshot as deleted, for future cleanup:
879 int bch2_snapshot_node_set_deleted(struct btree_trans *trans, u32 id)
881 struct btree_iter iter;
882 struct bkey_i_snapshot *s;
885 s = bch2_bkey_get_mut_typed(trans, &iter,
886 BTREE_ID_snapshots, POS(0, id),
888 ret = PTR_ERR_OR_ZERO(s);
890 bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOENT),
891 trans->c, "missing snapshot %u", id);
895 /* already deleted? */
896 if (BCH_SNAPSHOT_DELETED(&s->v))
899 SET_BCH_SNAPSHOT_DELETED(&s->v, true);
900 SET_BCH_SNAPSHOT_SUBVOL(&s->v, false);
903 bch2_trans_iter_exit(trans, &iter);
907 static inline void normalize_snapshot_child_pointers(struct bch_snapshot *s)
909 if (le32_to_cpu(s->children[0]) < le32_to_cpu(s->children[1]))
910 swap(s->children[0], s->children[1]);
913 static int bch2_snapshot_node_delete(struct btree_trans *trans, u32 id)
915 struct bch_fs *c = trans->c;
916 struct btree_iter iter, p_iter = (struct btree_iter) { NULL };
917 struct btree_iter c_iter = (struct btree_iter) { NULL };
918 struct btree_iter tree_iter = (struct btree_iter) { NULL };
919 struct bkey_s_c_snapshot s;
920 u32 parent_id, child_id;
924 s = bch2_bkey_get_iter_typed(trans, &iter, BTREE_ID_snapshots, POS(0, id),
925 BTREE_ITER_INTENT, snapshot);
927 bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOENT), c,
928 "missing snapshot %u", id);
933 BUG_ON(s.v->children[1]);
935 parent_id = le32_to_cpu(s.v->parent);
936 child_id = le32_to_cpu(s.v->children[0]);
939 struct bkey_i_snapshot *parent;
941 parent = bch2_bkey_get_mut_typed(trans, &p_iter,
942 BTREE_ID_snapshots, POS(0, parent_id),
944 ret = PTR_ERR_OR_ZERO(parent);
945 bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOENT), c,
946 "missing snapshot %u", parent_id);
950 /* find entry in parent->children for node being deleted */
951 for (i = 0; i < 2; i++)
952 if (le32_to_cpu(parent->v.children[i]) == id)
955 if (bch2_fs_inconsistent_on(i == 2, c,
956 "snapshot %u missing child pointer to %u",
960 parent->v.children[i] = le32_to_cpu(child_id);
962 normalize_snapshot_child_pointers(&parent->v);
966 struct bkey_i_snapshot *child;
968 child = bch2_bkey_get_mut_typed(trans, &c_iter,
969 BTREE_ID_snapshots, POS(0, child_id),
971 ret = PTR_ERR_OR_ZERO(child);
972 bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOENT), c,
973 "missing snapshot %u", child_id);
977 child->v.parent = cpu_to_le32(parent_id);
979 if (!child->v.parent) {
980 child->v.skip[0] = 0;
981 child->v.skip[1] = 0;
982 child->v.skip[2] = 0;
988 * We're deleting the root of a snapshot tree: update the
989 * snapshot_tree entry to point to the new root, or delete it if
990 * this is the last snapshot ID in this tree:
992 struct bkey_i_snapshot_tree *s_t;
994 BUG_ON(s.v->children[1]);
996 s_t = bch2_bkey_get_mut_typed(trans, &tree_iter,
997 BTREE_ID_snapshot_trees, POS(0, le32_to_cpu(s.v->tree)),
999 ret = PTR_ERR_OR_ZERO(s_t);
1003 if (s.v->children[0]) {
1004 s_t->v.root_snapshot = s.v->children[0];
1006 s_t->k.type = KEY_TYPE_deleted;
1007 set_bkey_val_u64s(&s_t->k, 0);
1011 ret = bch2_btree_delete_at(trans, &iter, 0);
1013 bch2_trans_iter_exit(trans, &tree_iter);
1014 bch2_trans_iter_exit(trans, &p_iter);
1015 bch2_trans_iter_exit(trans, &c_iter);
1016 bch2_trans_iter_exit(trans, &iter);
1020 static int create_snapids(struct btree_trans *trans, u32 parent, u32 tree,
1022 u32 *snapshot_subvols,
1023 unsigned nr_snapids)
1025 struct bch_fs *c = trans->c;
1026 struct btree_iter iter;
1027 struct bkey_i_snapshot *n;
1030 u32 depth = bch2_snapshot_depth(c, parent);
1033 bch2_trans_iter_init(trans, &iter, BTREE_ID_snapshots,
1034 POS_MIN, BTREE_ITER_INTENT);
1035 k = bch2_btree_iter_peek(&iter);
1040 for (i = 0; i < nr_snapids; i++) {
1041 k = bch2_btree_iter_prev_slot(&iter);
1046 if (!k.k || !k.k->p.offset) {
1047 ret = -BCH_ERR_ENOSPC_snapshot_create;
1051 n = bch2_bkey_alloc(trans, &iter, 0, snapshot);
1052 ret = PTR_ERR_OR_ZERO(n);
1057 n->v.parent = cpu_to_le32(parent);
1058 n->v.subvol = cpu_to_le32(snapshot_subvols[i]);
1059 n->v.tree = cpu_to_le32(tree);
1060 n->v.depth = cpu_to_le32(depth);
1062 for (j = 0; j < ARRAY_SIZE(n->v.skip); j++)
1063 n->v.skip[j] = cpu_to_le32(bch2_snapshot_skiplist_get(c, parent));
1065 bubble_sort(n->v.skip, ARRAY_SIZE(n->v.skip), cmp_le32);
1066 SET_BCH_SNAPSHOT_SUBVOL(&n->v, true);
1068 ret = bch2_mark_snapshot(trans, BTREE_ID_snapshots, 0,
1069 bkey_s_c_null, bkey_i_to_s_c(&n->k_i), 0);
1073 new_snapids[i] = iter.pos.offset;
1075 mutex_lock(&c->snapshot_table_lock);
1076 snapshot_t_mut(c, new_snapids[i])->equiv = new_snapids[i];
1077 mutex_unlock(&c->snapshot_table_lock);
1080 bch2_trans_iter_exit(trans, &iter);
1085 * Create new snapshot IDs as children of an existing snapshot ID:
1087 static int bch2_snapshot_node_create_children(struct btree_trans *trans, u32 parent,
1089 u32 *snapshot_subvols,
1090 unsigned nr_snapids)
1092 struct btree_iter iter;
1093 struct bkey_i_snapshot *n_parent;
1096 n_parent = bch2_bkey_get_mut_typed(trans, &iter,
1097 BTREE_ID_snapshots, POS(0, parent),
1099 ret = PTR_ERR_OR_ZERO(n_parent);
1100 if (unlikely(ret)) {
1101 if (bch2_err_matches(ret, ENOENT))
1102 bch_err(trans->c, "snapshot %u not found", parent);
1106 if (n_parent->v.children[0] || n_parent->v.children[1]) {
1107 bch_err(trans->c, "Trying to add child snapshot nodes to parent that already has children");
1112 ret = create_snapids(trans, parent, le32_to_cpu(n_parent->v.tree),
1113 new_snapids, snapshot_subvols, nr_snapids);
1117 n_parent->v.children[0] = cpu_to_le32(new_snapids[0]);
1118 n_parent->v.children[1] = cpu_to_le32(new_snapids[1]);
1119 n_parent->v.subvol = 0;
1120 SET_BCH_SNAPSHOT_SUBVOL(&n_parent->v, false);
1122 bch2_trans_iter_exit(trans, &iter);
1127 * Create a snapshot node that is the root of a new tree:
1129 static int bch2_snapshot_node_create_tree(struct btree_trans *trans,
1131 u32 *snapshot_subvols,
1132 unsigned nr_snapids)
1134 struct bkey_i_snapshot_tree *n_tree;
1137 n_tree = __bch2_snapshot_tree_create(trans);
1138 ret = PTR_ERR_OR_ZERO(n_tree) ?:
1139 create_snapids(trans, 0, n_tree->k.p.offset,
1140 new_snapids, snapshot_subvols, nr_snapids);
1144 n_tree->v.master_subvol = cpu_to_le32(snapshot_subvols[0]);
1145 n_tree->v.root_snapshot = cpu_to_le32(new_snapids[0]);
1149 int bch2_snapshot_node_create(struct btree_trans *trans, u32 parent,
1151 u32 *snapshot_subvols,
1152 unsigned nr_snapids)
1154 BUG_ON((parent == 0) != (nr_snapids == 1));
1155 BUG_ON((parent != 0) != (nr_snapids == 2));
1158 ? bch2_snapshot_node_create_children(trans, parent,
1159 new_snapids, snapshot_subvols, nr_snapids)
1160 : bch2_snapshot_node_create_tree(trans,
1161 new_snapids, snapshot_subvols, nr_snapids);
1166 * If we have an unlinked inode in an internal snapshot node, and the inode
1167 * really has been deleted in all child snapshots, how does this get cleaned up?
1169 * first there is the problem of how keys that have been overwritten in all
1170 * child snapshots get deleted (unimplemented?), but inodes may perhaps be
1173 * also: unlinked inode in internal snapshot appears to not be getting deleted
1174 * correctly if inode doesn't exist in leaf snapshots
1178 * for a key in an interior snapshot node that needs work to be done that
1179 * requires it to be mutated: iterate over all descendent leaf nodes and copy
1180 * that key to snapshot leaf nodes, where we can mutate it
1183 static int snapshot_delete_key(struct btree_trans *trans,
1184 struct btree_iter *iter,
1186 snapshot_id_list *deleted,
1187 snapshot_id_list *equiv_seen,
1188 struct bpos *last_pos)
1190 struct bch_fs *c = trans->c;
1191 u32 equiv = bch2_snapshot_equiv(c, k.k->p.snapshot);
1193 if (!bkey_eq(k.k->p, *last_pos))
1197 if (snapshot_list_has_id(deleted, k.k->p.snapshot) ||
1198 snapshot_list_has_id(equiv_seen, equiv)) {
1199 return bch2_btree_delete_at(trans, iter,
1200 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
1202 return snapshot_list_add(c, equiv_seen, equiv);
1206 static int move_key_to_correct_snapshot(struct btree_trans *trans,
1207 struct btree_iter *iter,
1210 struct bch_fs *c = trans->c;
1211 u32 equiv = bch2_snapshot_equiv(c, k.k->p.snapshot);
1214 * When we have a linear chain of snapshot nodes, we consider
1215 * those to form an equivalence class: we're going to collapse
1216 * them all down to a single node, and keep the leaf-most node -
1217 * which has the same id as the equivalence class id.
1219 * If there are multiple keys in different snapshots at the same
1220 * position, we're only going to keep the one in the newest
1221 * snapshot - the rest have been overwritten and are redundant,
1222 * and for the key we're going to keep we need to move it to the
1223 * equivalance class ID if it's not there already.
1225 if (equiv != k.k->p.snapshot) {
1226 struct bkey_i *new = bch2_bkey_make_mut_noupdate(trans, k);
1227 struct btree_iter new_iter;
1230 ret = PTR_ERR_OR_ZERO(new);
1234 new->k.p.snapshot = equiv;
1236 bch2_trans_iter_init(trans, &new_iter, iter->btree_id, new->k.p,
1237 BTREE_ITER_ALL_SNAPSHOTS|
1241 ret = bch2_btree_iter_traverse(&new_iter) ?:
1242 bch2_trans_update(trans, &new_iter, new,
1243 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE) ?:
1244 bch2_btree_delete_at(trans, iter,
1245 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
1246 bch2_trans_iter_exit(trans, &new_iter);
1255 * For a given snapshot, if it doesn't have a subvolume that points to it, and
1256 * it doesn't have child snapshot nodes - it's now redundant and we can mark it
1259 static int bch2_delete_redundant_snapshot(struct btree_trans *trans, struct btree_iter *iter,
1262 struct bkey_s_c_snapshot snap;
1266 if (k.k->type != KEY_TYPE_snapshot)
1269 snap = bkey_s_c_to_snapshot(k);
1270 if (BCH_SNAPSHOT_DELETED(snap.v) ||
1271 BCH_SNAPSHOT_SUBVOL(snap.v))
1274 children[0] = le32_to_cpu(snap.v->children[0]);
1275 children[1] = le32_to_cpu(snap.v->children[1]);
1277 ret = bch2_snapshot_live(trans, children[0]) ?:
1278 bch2_snapshot_live(trans, children[1]);
1283 return bch2_snapshot_node_set_deleted(trans, k.k->p.offset);
1287 static inline u32 bch2_snapshot_nth_parent_skip(struct bch_fs *c, u32 id, u32 n,
1288 snapshot_id_list *skip)
1291 while (snapshot_list_has_id(skip, id))
1292 id = __bch2_snapshot_parent(c, id);
1296 id = __bch2_snapshot_parent(c, id);
1297 } while (snapshot_list_has_id(skip, id));
1304 static int bch2_fix_child_of_deleted_snapshot(struct btree_trans *trans,
1305 struct btree_iter *iter, struct bkey_s_c k,
1306 snapshot_id_list *deleted)
1308 struct bch_fs *c = trans->c;
1309 u32 nr_deleted_ancestors = 0;
1310 struct bkey_i_snapshot *s;
1314 if (k.k->type != KEY_TYPE_snapshot)
1317 if (snapshot_list_has_id(deleted, k.k->p.offset))
1320 s = bch2_bkey_make_mut_noupdate_typed(trans, k, snapshot);
1321 ret = PTR_ERR_OR_ZERO(s);
1325 darray_for_each(*deleted, i)
1326 nr_deleted_ancestors += bch2_snapshot_is_ancestor(c, s->k.p.offset, *i);
1328 if (!nr_deleted_ancestors)
1331 le32_add_cpu(&s->v.depth, -nr_deleted_ancestors);
1338 u32 depth = le32_to_cpu(s->v.depth);
1339 u32 parent = bch2_snapshot_parent(c, s->k.p.offset);
1341 for (unsigned j = 0; j < ARRAY_SIZE(s->v.skip); j++) {
1342 u32 id = le32_to_cpu(s->v.skip[j]);
1344 if (snapshot_list_has_id(deleted, id)) {
1346 ? bch2_snapshot_nth_parent_skip(c,
1348 get_random_u32_below(depth - 1),
1351 s->v.skip[j] = cpu_to_le32(id);
1355 bubble_sort(s->v.skip, ARRAY_SIZE(s->v.skip), cmp_le32);
1358 return bch2_trans_update(trans, iter, &s->k_i, 0);
1361 int bch2_delete_dead_snapshots(struct bch_fs *c)
1363 struct btree_trans *trans;
1364 struct btree_iter iter;
1366 struct bkey_s_c_snapshot snap;
1367 snapshot_id_list deleted = { 0 };
1368 snapshot_id_list deleted_interior = { 0 };
1372 if (!test_bit(BCH_FS_STARTED, &c->flags)) {
1373 ret = bch2_fs_read_write_early(c);
1375 bch_err_msg(c, ret, "deleting dead snapshots: error going rw");
1380 trans = bch2_trans_get(c);
1383 * For every snapshot node: If we have no live children and it's not
1384 * pointed to by a subvolume, delete it:
1386 ret = for_each_btree_key_commit(trans, iter, BTREE_ID_snapshots,
1389 bch2_delete_redundant_snapshot(trans, &iter, k));
1391 bch_err_msg(c, ret, "deleting redundant snapshots");
1395 ret = for_each_btree_key2(trans, iter, BTREE_ID_snapshots,
1397 bch2_snapshot_set_equiv(trans, k));
1399 bch_err_msg(c, ret, "in bch2_snapshots_set_equiv");
1403 for_each_btree_key(trans, iter, BTREE_ID_snapshots,
1404 POS_MIN, 0, k, ret) {
1405 if (k.k->type != KEY_TYPE_snapshot)
1408 snap = bkey_s_c_to_snapshot(k);
1409 if (BCH_SNAPSHOT_DELETED(snap.v)) {
1410 ret = snapshot_list_add(c, &deleted, k.k->p.offset);
1415 bch2_trans_iter_exit(trans, &iter);
1418 bch_err_msg(c, ret, "walking snapshots");
1422 for (id = 0; id < BTREE_ID_NR; id++) {
1423 struct bpos last_pos = POS_MIN;
1424 snapshot_id_list equiv_seen = { 0 };
1425 struct disk_reservation res = { 0 };
1427 if (!btree_type_has_snapshots(id))
1430 ret = for_each_btree_key_commit(trans, iter,
1432 BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS, k,
1433 &res, NULL, BTREE_INSERT_NOFAIL,
1434 snapshot_delete_key(trans, &iter, k, &deleted, &equiv_seen, &last_pos)) ?:
1435 for_each_btree_key_commit(trans, iter,
1437 BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS, k,
1438 &res, NULL, BTREE_INSERT_NOFAIL,
1439 move_key_to_correct_snapshot(trans, &iter, k));
1441 bch2_disk_reservation_put(c, &res);
1442 darray_exit(&equiv_seen);
1445 bch_err_msg(c, ret, "deleting keys from dying snapshots");
1450 down_write(&c->snapshot_create_lock);
1452 for_each_btree_key(trans, iter, BTREE_ID_snapshots,
1453 POS_MIN, 0, k, ret) {
1454 u32 snapshot = k.k->p.offset;
1455 u32 equiv = bch2_snapshot_equiv(c, snapshot);
1457 if (equiv != snapshot)
1458 snapshot_list_add(c, &deleted_interior, snapshot);
1460 bch2_trans_iter_exit(trans, &iter);
1463 goto err_create_lock;
1466 * Fixing children of deleted snapshots can't be done completely
1467 * atomically, if we crash between here and when we delete the interior
1468 * nodes some depth fields will be off:
1470 ret = for_each_btree_key_commit(trans, iter, BTREE_ID_snapshots, POS_MIN,
1471 BTREE_ITER_INTENT, k,
1472 NULL, NULL, BTREE_INSERT_NOFAIL,
1473 bch2_fix_child_of_deleted_snapshot(trans, &iter, k, &deleted_interior));
1475 goto err_create_lock;
1477 darray_for_each(deleted, i) {
1478 ret = commit_do(trans, NULL, NULL, 0,
1479 bch2_snapshot_node_delete(trans, *i));
1481 bch_err_msg(c, ret, "deleting snapshot %u", *i);
1482 goto err_create_lock;
1486 darray_for_each(deleted_interior, i) {
1487 ret = commit_do(trans, NULL, NULL, 0,
1488 bch2_snapshot_node_delete(trans, *i));
1490 bch_err_msg(c, ret, "deleting snapshot %u", *i);
1491 goto err_create_lock;
1495 clear_bit(BCH_FS_HAVE_DELETED_SNAPSHOTS, &c->flags);
1497 up_write(&c->snapshot_create_lock);
1499 darray_exit(&deleted_interior);
1500 darray_exit(&deleted);
1501 bch2_trans_put(trans);
1507 void bch2_delete_dead_snapshots_work(struct work_struct *work)
1509 struct bch_fs *c = container_of(work, struct bch_fs, snapshot_delete_work);
1511 if (test_bit(BCH_FS_HAVE_DELETED_SNAPSHOTS, &c->flags))
1512 bch2_delete_dead_snapshots(c);
1513 bch2_write_ref_put(c, BCH_WRITE_REF_delete_dead_snapshots);
1516 void bch2_delete_dead_snapshots_async(struct bch_fs *c)
1518 if (bch2_write_ref_tryget(c, BCH_WRITE_REF_delete_dead_snapshots) &&
1519 !queue_work(c->write_ref_wq, &c->snapshot_delete_work))
1520 bch2_write_ref_put(c, BCH_WRITE_REF_delete_dead_snapshots);
1523 int bch2_delete_dead_snapshots_hook(struct btree_trans *trans,
1524 struct btree_trans_commit_hook *h)
1526 struct bch_fs *c = trans->c;
1528 set_bit(BCH_FS_HAVE_DELETED_SNAPSHOTS, &c->flags);
1530 if (c->curr_recovery_pass <= BCH_RECOVERY_PASS_delete_dead_snapshots)
1533 bch2_delete_dead_snapshots_async(c);
1537 int __bch2_key_has_snapshot_overwrites(struct btree_trans *trans,
1541 struct bch_fs *c = trans->c;
1542 struct btree_iter iter;
1546 bch2_trans_iter_init(trans, &iter, id, pos,
1547 BTREE_ITER_NOT_EXTENTS|
1548 BTREE_ITER_ALL_SNAPSHOTS);
1550 k = bch2_btree_iter_prev(&iter);
1558 if (!bkey_eq(pos, k.k->p))
1561 if (bch2_snapshot_is_ancestor(c, k.k->p.snapshot, pos.snapshot)) {
1566 bch2_trans_iter_exit(trans, &iter);
1571 static u32 bch2_snapshot_smallest_child(struct bch_fs *c, u32 id)
1573 const struct snapshot_t *s = snapshot_t(c, id);
1575 return s->children[1] ?: s->children[0];
1578 static u32 bch2_snapshot_smallest_descendent(struct bch_fs *c, u32 id)
1582 while ((child = bch2_snapshot_smallest_child(c, id)))
1587 static int bch2_propagate_key_to_snapshot_leaf(struct btree_trans *trans,
1588 enum btree_id btree,
1589 struct bkey_s_c interior_k,
1590 u32 leaf_id, struct bpos *new_min_pos)
1592 struct btree_iter iter;
1593 struct bpos pos = interior_k.k->p;
1598 pos.snapshot = leaf_id;
1600 bch2_trans_iter_init(trans, &iter, btree, pos, BTREE_ITER_INTENT);
1601 k = bch2_btree_iter_peek_slot(&iter);
1606 /* key already overwritten in this snapshot? */
1607 if (k.k->p.snapshot != interior_k.k->p.snapshot)
1610 if (bpos_eq(*new_min_pos, POS_MIN)) {
1611 *new_min_pos = k.k->p;
1612 new_min_pos->snapshot = leaf_id;
1615 new = bch2_bkey_make_mut_noupdate(trans, interior_k);
1616 ret = PTR_ERR_OR_ZERO(new);
1620 new->k.p.snapshot = leaf_id;
1621 ret = bch2_trans_update(trans, &iter, new, 0);
1623 bch2_trans_iter_exit(trans, &iter);
1627 int bch2_propagate_key_to_snapshot_leaves(struct btree_trans *trans,
1628 enum btree_id btree,
1630 struct bpos *new_min_pos)
1632 struct bch_fs *c = trans->c;
1634 u32 restart_count = trans->restart_count;
1637 bch2_bkey_buf_init(&sk);
1638 bch2_bkey_buf_reassemble(&sk, c, k);
1639 k = bkey_i_to_s_c(sk.k);
1641 *new_min_pos = POS_MIN;
1643 for (u32 id = bch2_snapshot_smallest_descendent(c, k.k->p.snapshot);
1644 id < k.k->p.snapshot;
1646 if (!bch2_snapshot_is_ancestor(c, id, k.k->p.snapshot) ||
1647 !bch2_snapshot_is_leaf(c, id))
1650 ret = btree_trans_too_many_iters(trans) ?:
1651 bch2_propagate_key_to_snapshot_leaf(trans, btree, k, id, new_min_pos) ?:
1652 bch2_trans_commit(trans, NULL, NULL, 0);
1653 if (ret && bch2_err_matches(ret, BCH_ERR_transaction_restart)) {
1654 bch2_trans_begin(trans);
1662 bch2_bkey_buf_exit(&sk, c);
1664 return ret ?: trans_was_restarted(trans, restart_count);
1667 int bch2_snapshots_read(struct bch_fs *c)
1669 struct btree_iter iter;
1673 ret = bch2_trans_run(c,
1674 for_each_btree_key2(trans, iter, BTREE_ID_snapshots,
1676 bch2_mark_snapshot(trans, BTREE_ID_snapshots, 0, bkey_s_c_null, k, 0) ?:
1677 bch2_snapshot_set_equiv(trans, k)) ?:
1678 for_each_btree_key2(trans, iter, BTREE_ID_snapshots,
1680 (set_is_ancestor_bitmap(c, k.k->p.offset), 0)));
1686 void bch2_fs_snapshots_exit(struct bch_fs *c)
1688 kfree(rcu_dereference_protected(c->snapshots, true));