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_NEED_DELETE_DEAD_SNAPSHOTS, &c->flags);
329 if (c->curr_recovery_pass > BCH_RECOVERY_PASS_delete_dead_snapshots)
330 bch2_delete_dead_snapshots_async(c);
333 memset(t, 0, sizeof(*t));
336 mutex_unlock(&c->snapshot_table_lock);
340 int bch2_snapshot_lookup(struct btree_trans *trans, u32 id,
341 struct bch_snapshot *s)
343 return bch2_bkey_get_val_typed(trans, BTREE_ID_snapshots, POS(0, id),
344 BTREE_ITER_WITH_UPDATES, snapshot, s);
347 static int bch2_snapshot_live(struct btree_trans *trans, u32 id)
349 struct bch_snapshot v;
355 ret = bch2_snapshot_lookup(trans, id, &v);
356 if (bch2_err_matches(ret, ENOENT))
357 bch_err(trans->c, "snapshot node %u not found", id);
361 return !BCH_SNAPSHOT_DELETED(&v);
365 * If @k is a snapshot with just one live child, it's part of a linear chain,
366 * which we consider to be an equivalence class: and then after snapshot
367 * deletion cleanup, there should only be a single key at a given position in
368 * this equivalence class.
370 * This sets the equivalence class of @k to be the child's equivalence class, if
371 * it's part of such a linear chain: this correctly sets equivalence classes on
372 * startup if we run leaf to root (i.e. in natural key order).
374 static int bch2_snapshot_set_equiv(struct btree_trans *trans, struct bkey_s_c k)
376 struct bch_fs *c = trans->c;
377 unsigned i, nr_live = 0, live_idx = 0;
378 struct bkey_s_c_snapshot snap;
379 u32 id = k.k->p.offset, child[2];
381 if (k.k->type != KEY_TYPE_snapshot)
384 snap = bkey_s_c_to_snapshot(k);
386 child[0] = le32_to_cpu(snap.v->children[0]);
387 child[1] = le32_to_cpu(snap.v->children[1]);
389 for (i = 0; i < 2; i++) {
390 int ret = bch2_snapshot_live(trans, child[i]);
400 mutex_lock(&c->snapshot_table_lock);
402 snapshot_t_mut(c, id)->equiv = nr_live == 1
403 ? snapshot_t_mut(c, child[live_idx])->equiv
406 mutex_unlock(&c->snapshot_table_lock);
413 static u32 bch2_snapshot_child(struct bch_fs *c, u32 id, unsigned child)
415 return snapshot_t(c, id)->children[child];
418 static u32 bch2_snapshot_left_child(struct bch_fs *c, u32 id)
420 return bch2_snapshot_child(c, id, 0);
423 static u32 bch2_snapshot_right_child(struct bch_fs *c, u32 id)
425 return bch2_snapshot_child(c, id, 1);
428 static u32 bch2_snapshot_tree_next(struct bch_fs *c, u32 id)
432 n = bch2_snapshot_left_child(c, id);
436 while ((parent = bch2_snapshot_parent(c, id))) {
437 n = bch2_snapshot_right_child(c, parent);
446 static u32 bch2_snapshot_tree_oldest_subvol(struct bch_fs *c, u32 snapshot_root)
448 u32 id = snapshot_root;
452 s = snapshot_t(c, id)->subvol;
454 if (s && (!subvol || s < subvol))
457 id = bch2_snapshot_tree_next(c, id);
463 static int bch2_snapshot_tree_master_subvol(struct btree_trans *trans,
464 u32 snapshot_root, u32 *subvol_id)
466 struct bch_fs *c = trans->c;
467 struct btree_iter iter;
469 struct bkey_s_c_subvolume s;
473 for_each_btree_key_norestart(trans, iter, BTREE_ID_subvolumes, POS_MIN,
475 if (k.k->type != KEY_TYPE_subvolume)
478 s = bkey_s_c_to_subvolume(k);
479 if (!bch2_snapshot_is_ancestor(c, le32_to_cpu(s.v->snapshot), snapshot_root))
481 if (!BCH_SUBVOLUME_SNAP(s.v)) {
482 *subvol_id = s.k->p.offset;
488 bch2_trans_iter_exit(trans, &iter);
490 if (!ret && !found) {
491 struct bkey_i_subvolume *u;
493 *subvol_id = bch2_snapshot_tree_oldest_subvol(c, snapshot_root);
495 u = bch2_bkey_get_mut_typed(trans, &iter,
496 BTREE_ID_subvolumes, POS(0, *subvol_id),
498 ret = PTR_ERR_OR_ZERO(u);
502 SET_BCH_SUBVOLUME_SNAP(&u->v, false);
508 static int check_snapshot_tree(struct btree_trans *trans,
509 struct btree_iter *iter,
512 struct bch_fs *c = trans->c;
513 struct bkey_s_c_snapshot_tree st;
514 struct bch_snapshot s;
515 struct bch_subvolume subvol;
516 struct printbuf buf = PRINTBUF;
520 if (k.k->type != KEY_TYPE_snapshot_tree)
523 st = bkey_s_c_to_snapshot_tree(k);
524 root_id = le32_to_cpu(st.v->root_snapshot);
526 ret = bch2_snapshot_lookup(trans, root_id, &s);
527 if (ret && !bch2_err_matches(ret, ENOENT))
530 if (fsck_err_on(ret ||
531 root_id != bch2_snapshot_root(c, root_id) ||
532 st.k->p.offset != le32_to_cpu(s.tree),
534 "snapshot tree points to missing/incorrect snapshot:\n %s",
535 (bch2_bkey_val_to_text(&buf, c, st.s_c), buf.buf))) {
536 ret = bch2_btree_delete_at(trans, iter, 0);
540 ret = bch2_subvolume_get(trans, le32_to_cpu(st.v->master_subvol),
542 if (ret && !bch2_err_matches(ret, ENOENT))
545 if (fsck_err_on(ret, c,
546 "snapshot tree points to missing subvolume:\n %s",
547 (printbuf_reset(&buf),
548 bch2_bkey_val_to_text(&buf, c, st.s_c), buf.buf)) ||
549 fsck_err_on(!bch2_snapshot_is_ancestor_early(c,
550 le32_to_cpu(subvol.snapshot),
552 "snapshot tree points to subvolume that does not point to snapshot in this tree:\n %s",
553 (printbuf_reset(&buf),
554 bch2_bkey_val_to_text(&buf, c, st.s_c), buf.buf)) ||
555 fsck_err_on(BCH_SUBVOLUME_SNAP(&subvol), c,
556 "snapshot tree points to snapshot subvolume:\n %s",
557 (printbuf_reset(&buf),
558 bch2_bkey_val_to_text(&buf, c, st.s_c), buf.buf))) {
559 struct bkey_i_snapshot_tree *u;
562 ret = bch2_snapshot_tree_master_subvol(trans, root_id, &subvol_id);
566 u = bch2_bkey_make_mut_typed(trans, iter, &k, 0, snapshot_tree);
567 ret = PTR_ERR_OR_ZERO(u);
571 u->v.master_subvol = cpu_to_le32(subvol_id);
572 st = snapshot_tree_i_to_s_c(u);
581 * For each snapshot_tree, make sure it points to the root of a snapshot tree
582 * and that snapshot entry points back to it, or delete it.
584 * And, make sure it points to a subvolume within that snapshot tree, or correct
585 * it to point to the oldest subvolume within that snapshot tree.
587 int bch2_check_snapshot_trees(struct bch_fs *c)
589 struct btree_iter iter;
593 ret = bch2_trans_run(c,
594 for_each_btree_key_commit(trans, iter,
595 BTREE_ID_snapshot_trees, POS_MIN,
596 BTREE_ITER_PREFETCH, k,
597 NULL, NULL, BTREE_INSERT_LAZY_RW|BTREE_INSERT_NOFAIL,
598 check_snapshot_tree(trans, &iter, k)));
601 bch_err(c, "error %i checking snapshot trees", ret);
606 * Look up snapshot tree for @tree_id and find root,
607 * make sure @snap_id is a descendent:
609 static int snapshot_tree_ptr_good(struct btree_trans *trans,
610 u32 snap_id, u32 tree_id)
612 struct bch_snapshot_tree s_t;
613 int ret = bch2_snapshot_tree_lookup(trans, tree_id, &s_t);
615 if (bch2_err_matches(ret, ENOENT))
620 return bch2_snapshot_is_ancestor_early(trans->c, snap_id, le32_to_cpu(s_t.root_snapshot));
623 u32 bch2_snapshot_skiplist_get(struct bch_fs *c, u32 id)
625 const struct snapshot_t *s;
631 s = snapshot_t(c, id);
633 id = bch2_snapshot_nth_parent(c, id, get_random_u32_below(s->depth));
639 static int snapshot_skiplist_good(struct btree_trans *trans, u32 id, struct bch_snapshot s)
643 for (i = 0; i < 3; i++)
648 if (!bch2_snapshot_is_ancestor_early(trans->c, id, le32_to_cpu(s.skip[i])))
656 * snapshot_tree pointer was incorrect: look up root snapshot node, make sure
657 * its snapshot_tree pointer is correct (allocate new one if necessary), then
658 * update this node's pointer to root node's pointer:
660 static int snapshot_tree_ptr_repair(struct btree_trans *trans,
661 struct btree_iter *iter,
663 struct bch_snapshot *s)
665 struct bch_fs *c = trans->c;
666 struct btree_iter root_iter;
667 struct bch_snapshot_tree s_t;
668 struct bkey_s_c_snapshot root;
669 struct bkey_i_snapshot *u;
670 u32 root_id = bch2_snapshot_root(c, k.k->p.offset), tree_id;
673 root = bch2_bkey_get_iter_typed(trans, &root_iter,
674 BTREE_ID_snapshots, POS(0, root_id),
675 BTREE_ITER_WITH_UPDATES, snapshot);
676 ret = bkey_err(root);
680 tree_id = le32_to_cpu(root.v->tree);
682 ret = bch2_snapshot_tree_lookup(trans, tree_id, &s_t);
683 if (ret && !bch2_err_matches(ret, ENOENT))
686 if (ret || le32_to_cpu(s_t.root_snapshot) != root_id) {
687 u = bch2_bkey_make_mut_typed(trans, &root_iter, &root.s_c, 0, snapshot);
688 ret = PTR_ERR_OR_ZERO(u) ?:
689 bch2_snapshot_tree_create(trans, root_id,
690 bch2_snapshot_tree_oldest_subvol(c, root_id),
695 u->v.tree = cpu_to_le32(tree_id);
696 if (k.k->p.offset == root_id)
700 if (k.k->p.offset != root_id) {
701 u = bch2_bkey_make_mut_typed(trans, iter, &k, 0, snapshot);
702 ret = PTR_ERR_OR_ZERO(u);
706 u->v.tree = cpu_to_le32(tree_id);
710 bch2_trans_iter_exit(trans, &root_iter);
714 static int check_snapshot(struct btree_trans *trans,
715 struct btree_iter *iter,
718 struct bch_fs *c = trans->c;
719 struct bch_snapshot s;
720 struct bch_subvolume subvol;
721 struct bch_snapshot v;
722 struct bkey_i_snapshot *u;
723 u32 parent_id = bch2_snapshot_parent_early(c, k.k->p.offset);
725 struct printbuf buf = PRINTBUF;
726 bool should_have_subvol;
730 if (k.k->type != KEY_TYPE_snapshot)
733 memset(&s, 0, sizeof(s));
734 memcpy(&s, k.v, bkey_val_bytes(k.k));
736 id = le32_to_cpu(s.parent);
738 ret = bch2_snapshot_lookup(trans, id, &v);
739 if (bch2_err_matches(ret, ENOENT))
740 bch_err(c, "snapshot with nonexistent parent:\n %s",
741 (bch2_bkey_val_to_text(&buf, c, k), buf.buf));
745 if (le32_to_cpu(v.children[0]) != k.k->p.offset &&
746 le32_to_cpu(v.children[1]) != k.k->p.offset) {
747 bch_err(c, "snapshot parent %u missing pointer to child %llu",
754 for (i = 0; i < 2 && s.children[i]; i++) {
755 id = le32_to_cpu(s.children[i]);
757 ret = bch2_snapshot_lookup(trans, id, &v);
758 if (bch2_err_matches(ret, ENOENT))
759 bch_err(c, "snapshot node %llu has nonexistent child %u",
764 if (le32_to_cpu(v.parent) != k.k->p.offset) {
765 bch_err(c, "snapshot child %u has wrong parent (got %u should be %llu)",
766 id, le32_to_cpu(v.parent), k.k->p.offset);
772 should_have_subvol = BCH_SNAPSHOT_SUBVOL(&s) &&
773 !BCH_SNAPSHOT_DELETED(&s);
775 if (should_have_subvol) {
776 id = le32_to_cpu(s.subvol);
777 ret = bch2_subvolume_get(trans, id, 0, false, &subvol);
778 if (bch2_err_matches(ret, ENOENT))
779 bch_err(c, "snapshot points to nonexistent subvolume:\n %s",
780 (bch2_bkey_val_to_text(&buf, c, k), buf.buf));
784 if (BCH_SNAPSHOT_SUBVOL(&s) != (le32_to_cpu(subvol.snapshot) == k.k->p.offset)) {
785 bch_err(c, "snapshot node %llu has wrong BCH_SNAPSHOT_SUBVOL",
791 if (fsck_err_on(s.subvol, c, "snapshot should not point to subvol:\n %s",
792 (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
793 u = bch2_bkey_make_mut_typed(trans, iter, &k, 0, snapshot);
794 ret = PTR_ERR_OR_ZERO(u);
803 ret = snapshot_tree_ptr_good(trans, k.k->p.offset, le32_to_cpu(s.tree));
807 if (fsck_err_on(!ret, c, "snapshot points to missing/incorrect tree:\n %s",
808 (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
809 ret = snapshot_tree_ptr_repair(trans, iter, k, &s);
815 real_depth = bch2_snapshot_depth(c, parent_id);
817 if (le32_to_cpu(s.depth) != real_depth &&
818 (c->sb.version_upgrade_complete < bcachefs_metadata_version_snapshot_skiplists ||
819 fsck_err(c, "snapshot with incorrect depth field, should be %u:\n %s",
820 real_depth, (bch2_bkey_val_to_text(&buf, c, k), buf.buf)))) {
821 u = bch2_bkey_make_mut_typed(trans, iter, &k, 0, snapshot);
822 ret = PTR_ERR_OR_ZERO(u);
826 u->v.depth = cpu_to_le32(real_depth);
830 ret = snapshot_skiplist_good(trans, k.k->p.offset, s);
835 (c->sb.version_upgrade_complete < bcachefs_metadata_version_snapshot_skiplists ||
836 fsck_err(c, "snapshot with bad skiplist field:\n %s",
837 (bch2_bkey_val_to_text(&buf, c, k), buf.buf)))) {
838 u = bch2_bkey_make_mut_typed(trans, iter, &k, 0, snapshot);
839 ret = PTR_ERR_OR_ZERO(u);
843 for (i = 0; i < ARRAY_SIZE(u->v.skip); i++)
844 u->v.skip[i] = cpu_to_le32(bch2_snapshot_skiplist_get(c, parent_id));
846 bubble_sort(u->v.skip, ARRAY_SIZE(u->v.skip), cmp_le32);
856 int bch2_check_snapshots(struct bch_fs *c)
858 struct btree_iter iter;
863 * We iterate backwards as checking/fixing the depth field requires that
864 * the parent's depth already be correct:
866 ret = bch2_trans_run(c,
867 for_each_btree_key_reverse_commit(trans, iter,
868 BTREE_ID_snapshots, POS_MAX,
869 BTREE_ITER_PREFETCH, k,
870 NULL, NULL, BTREE_INSERT_LAZY_RW|BTREE_INSERT_NOFAIL,
871 check_snapshot(trans, &iter, k)));
878 * Mark a snapshot as deleted, for future cleanup:
880 int bch2_snapshot_node_set_deleted(struct btree_trans *trans, u32 id)
882 struct btree_iter iter;
883 struct bkey_i_snapshot *s;
886 s = bch2_bkey_get_mut_typed(trans, &iter,
887 BTREE_ID_snapshots, POS(0, id),
889 ret = PTR_ERR_OR_ZERO(s);
891 bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOENT),
892 trans->c, "missing snapshot %u", id);
896 /* already deleted? */
897 if (BCH_SNAPSHOT_DELETED(&s->v))
900 SET_BCH_SNAPSHOT_DELETED(&s->v, true);
901 SET_BCH_SNAPSHOT_SUBVOL(&s->v, false);
904 bch2_trans_iter_exit(trans, &iter);
908 static inline void normalize_snapshot_child_pointers(struct bch_snapshot *s)
910 if (le32_to_cpu(s->children[0]) < le32_to_cpu(s->children[1]))
911 swap(s->children[0], s->children[1]);
914 static int bch2_snapshot_node_delete(struct btree_trans *trans, u32 id)
916 struct bch_fs *c = trans->c;
917 struct btree_iter iter, p_iter = (struct btree_iter) { NULL };
918 struct btree_iter c_iter = (struct btree_iter) { NULL };
919 struct btree_iter tree_iter = (struct btree_iter) { NULL };
920 struct bkey_s_c_snapshot s;
921 u32 parent_id, child_id;
925 s = bch2_bkey_get_iter_typed(trans, &iter, BTREE_ID_snapshots, POS(0, id),
926 BTREE_ITER_INTENT, snapshot);
928 bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOENT), c,
929 "missing snapshot %u", id);
934 BUG_ON(s.v->children[1]);
936 parent_id = le32_to_cpu(s.v->parent);
937 child_id = le32_to_cpu(s.v->children[0]);
940 struct bkey_i_snapshot *parent;
942 parent = bch2_bkey_get_mut_typed(trans, &p_iter,
943 BTREE_ID_snapshots, POS(0, parent_id),
945 ret = PTR_ERR_OR_ZERO(parent);
946 bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOENT), c,
947 "missing snapshot %u", parent_id);
951 /* find entry in parent->children for node being deleted */
952 for (i = 0; i < 2; i++)
953 if (le32_to_cpu(parent->v.children[i]) == id)
956 if (bch2_fs_inconsistent_on(i == 2, c,
957 "snapshot %u missing child pointer to %u",
961 parent->v.children[i] = le32_to_cpu(child_id);
963 normalize_snapshot_child_pointers(&parent->v);
967 struct bkey_i_snapshot *child;
969 child = bch2_bkey_get_mut_typed(trans, &c_iter,
970 BTREE_ID_snapshots, POS(0, child_id),
972 ret = PTR_ERR_OR_ZERO(child);
973 bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOENT), c,
974 "missing snapshot %u", child_id);
978 child->v.parent = cpu_to_le32(parent_id);
980 if (!child->v.parent) {
981 child->v.skip[0] = 0;
982 child->v.skip[1] = 0;
983 child->v.skip[2] = 0;
989 * We're deleting the root of a snapshot tree: update the
990 * snapshot_tree entry to point to the new root, or delete it if
991 * this is the last snapshot ID in this tree:
993 struct bkey_i_snapshot_tree *s_t;
995 BUG_ON(s.v->children[1]);
997 s_t = bch2_bkey_get_mut_typed(trans, &tree_iter,
998 BTREE_ID_snapshot_trees, POS(0, le32_to_cpu(s.v->tree)),
1000 ret = PTR_ERR_OR_ZERO(s_t);
1004 if (s.v->children[0]) {
1005 s_t->v.root_snapshot = s.v->children[0];
1007 s_t->k.type = KEY_TYPE_deleted;
1008 set_bkey_val_u64s(&s_t->k, 0);
1012 ret = bch2_btree_delete_at(trans, &iter, 0);
1014 bch2_trans_iter_exit(trans, &tree_iter);
1015 bch2_trans_iter_exit(trans, &p_iter);
1016 bch2_trans_iter_exit(trans, &c_iter);
1017 bch2_trans_iter_exit(trans, &iter);
1021 static int create_snapids(struct btree_trans *trans, u32 parent, u32 tree,
1023 u32 *snapshot_subvols,
1024 unsigned nr_snapids)
1026 struct bch_fs *c = trans->c;
1027 struct btree_iter iter;
1028 struct bkey_i_snapshot *n;
1031 u32 depth = bch2_snapshot_depth(c, parent);
1034 bch2_trans_iter_init(trans, &iter, BTREE_ID_snapshots,
1035 POS_MIN, BTREE_ITER_INTENT);
1036 k = bch2_btree_iter_peek(&iter);
1041 for (i = 0; i < nr_snapids; i++) {
1042 k = bch2_btree_iter_prev_slot(&iter);
1047 if (!k.k || !k.k->p.offset) {
1048 ret = -BCH_ERR_ENOSPC_snapshot_create;
1052 n = bch2_bkey_alloc(trans, &iter, 0, snapshot);
1053 ret = PTR_ERR_OR_ZERO(n);
1058 n->v.parent = cpu_to_le32(parent);
1059 n->v.subvol = cpu_to_le32(snapshot_subvols[i]);
1060 n->v.tree = cpu_to_le32(tree);
1061 n->v.depth = cpu_to_le32(depth);
1063 for (j = 0; j < ARRAY_SIZE(n->v.skip); j++)
1064 n->v.skip[j] = cpu_to_le32(bch2_snapshot_skiplist_get(c, parent));
1066 bubble_sort(n->v.skip, ARRAY_SIZE(n->v.skip), cmp_le32);
1067 SET_BCH_SNAPSHOT_SUBVOL(&n->v, true);
1069 ret = bch2_mark_snapshot(trans, BTREE_ID_snapshots, 0,
1070 bkey_s_c_null, bkey_i_to_s_c(&n->k_i), 0);
1074 new_snapids[i] = iter.pos.offset;
1076 mutex_lock(&c->snapshot_table_lock);
1077 snapshot_t_mut(c, new_snapids[i])->equiv = new_snapids[i];
1078 mutex_unlock(&c->snapshot_table_lock);
1081 bch2_trans_iter_exit(trans, &iter);
1086 * Create new snapshot IDs as children of an existing snapshot ID:
1088 static int bch2_snapshot_node_create_children(struct btree_trans *trans, u32 parent,
1090 u32 *snapshot_subvols,
1091 unsigned nr_snapids)
1093 struct btree_iter iter;
1094 struct bkey_i_snapshot *n_parent;
1097 n_parent = bch2_bkey_get_mut_typed(trans, &iter,
1098 BTREE_ID_snapshots, POS(0, parent),
1100 ret = PTR_ERR_OR_ZERO(n_parent);
1101 if (unlikely(ret)) {
1102 if (bch2_err_matches(ret, ENOENT))
1103 bch_err(trans->c, "snapshot %u not found", parent);
1107 if (n_parent->v.children[0] || n_parent->v.children[1]) {
1108 bch_err(trans->c, "Trying to add child snapshot nodes to parent that already has children");
1113 ret = create_snapids(trans, parent, le32_to_cpu(n_parent->v.tree),
1114 new_snapids, snapshot_subvols, nr_snapids);
1118 n_parent->v.children[0] = cpu_to_le32(new_snapids[0]);
1119 n_parent->v.children[1] = cpu_to_le32(new_snapids[1]);
1120 n_parent->v.subvol = 0;
1121 SET_BCH_SNAPSHOT_SUBVOL(&n_parent->v, false);
1123 bch2_trans_iter_exit(trans, &iter);
1128 * Create a snapshot node that is the root of a new tree:
1130 static int bch2_snapshot_node_create_tree(struct btree_trans *trans,
1132 u32 *snapshot_subvols,
1133 unsigned nr_snapids)
1135 struct bkey_i_snapshot_tree *n_tree;
1138 n_tree = __bch2_snapshot_tree_create(trans);
1139 ret = PTR_ERR_OR_ZERO(n_tree) ?:
1140 create_snapids(trans, 0, n_tree->k.p.offset,
1141 new_snapids, snapshot_subvols, nr_snapids);
1145 n_tree->v.master_subvol = cpu_to_le32(snapshot_subvols[0]);
1146 n_tree->v.root_snapshot = cpu_to_le32(new_snapids[0]);
1150 int bch2_snapshot_node_create(struct btree_trans *trans, u32 parent,
1152 u32 *snapshot_subvols,
1153 unsigned nr_snapids)
1155 BUG_ON((parent == 0) != (nr_snapids == 1));
1156 BUG_ON((parent != 0) != (nr_snapids == 2));
1159 ? bch2_snapshot_node_create_children(trans, parent,
1160 new_snapids, snapshot_subvols, nr_snapids)
1161 : bch2_snapshot_node_create_tree(trans,
1162 new_snapids, snapshot_subvols, nr_snapids);
1167 * If we have an unlinked inode in an internal snapshot node, and the inode
1168 * really has been deleted in all child snapshots, how does this get cleaned up?
1170 * first there is the problem of how keys that have been overwritten in all
1171 * child snapshots get deleted (unimplemented?), but inodes may perhaps be
1174 * also: unlinked inode in internal snapshot appears to not be getting deleted
1175 * correctly if inode doesn't exist in leaf snapshots
1179 * for a key in an interior snapshot node that needs work to be done that
1180 * requires it to be mutated: iterate over all descendent leaf nodes and copy
1181 * that key to snapshot leaf nodes, where we can mutate it
1184 static int snapshot_delete_key(struct btree_trans *trans,
1185 struct btree_iter *iter,
1187 snapshot_id_list *deleted,
1188 snapshot_id_list *equiv_seen,
1189 struct bpos *last_pos)
1191 struct bch_fs *c = trans->c;
1192 u32 equiv = bch2_snapshot_equiv(c, k.k->p.snapshot);
1194 if (!bkey_eq(k.k->p, *last_pos))
1198 if (snapshot_list_has_id(deleted, k.k->p.snapshot) ||
1199 snapshot_list_has_id(equiv_seen, equiv)) {
1200 return bch2_btree_delete_at(trans, iter,
1201 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
1203 return snapshot_list_add(c, equiv_seen, equiv);
1207 static int move_key_to_correct_snapshot(struct btree_trans *trans,
1208 struct btree_iter *iter,
1211 struct bch_fs *c = trans->c;
1212 u32 equiv = bch2_snapshot_equiv(c, k.k->p.snapshot);
1215 * When we have a linear chain of snapshot nodes, we consider
1216 * those to form an equivalence class: we're going to collapse
1217 * them all down to a single node, and keep the leaf-most node -
1218 * which has the same id as the equivalence class id.
1220 * If there are multiple keys in different snapshots at the same
1221 * position, we're only going to keep the one in the newest
1222 * snapshot - the rest have been overwritten and are redundant,
1223 * and for the key we're going to keep we need to move it to the
1224 * equivalance class ID if it's not there already.
1226 if (equiv != k.k->p.snapshot) {
1227 struct bkey_i *new = bch2_bkey_make_mut_noupdate(trans, k);
1228 struct btree_iter new_iter;
1231 ret = PTR_ERR_OR_ZERO(new);
1235 new->k.p.snapshot = equiv;
1237 bch2_trans_iter_init(trans, &new_iter, iter->btree_id, new->k.p,
1238 BTREE_ITER_ALL_SNAPSHOTS|
1242 ret = bch2_btree_iter_traverse(&new_iter) ?:
1243 bch2_trans_update(trans, &new_iter, new,
1244 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE) ?:
1245 bch2_btree_delete_at(trans, iter,
1246 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
1247 bch2_trans_iter_exit(trans, &new_iter);
1255 static int bch2_snapshot_needs_delete(struct btree_trans *trans, struct bkey_s_c k)
1257 struct bkey_s_c_snapshot snap;
1261 if (k.k->type != KEY_TYPE_snapshot)
1264 snap = bkey_s_c_to_snapshot(k);
1265 if (BCH_SNAPSHOT_DELETED(snap.v) ||
1266 BCH_SNAPSHOT_SUBVOL(snap.v))
1269 children[0] = le32_to_cpu(snap.v->children[0]);
1270 children[1] = le32_to_cpu(snap.v->children[1]);
1272 ret = bch2_snapshot_live(trans, children[0]) ?:
1273 bch2_snapshot_live(trans, children[1]);
1280 * For a given snapshot, if it doesn't have a subvolume that points to it, and
1281 * it doesn't have child snapshot nodes - it's now redundant and we can mark it
1284 static int bch2_delete_redundant_snapshot(struct btree_trans *trans, struct bkey_s_c k)
1286 int ret = bch2_snapshot_needs_delete(trans, k);
1290 : bch2_snapshot_node_set_deleted(trans, k.k->p.offset);
1293 static inline u32 bch2_snapshot_nth_parent_skip(struct bch_fs *c, u32 id, u32 n,
1294 snapshot_id_list *skip)
1297 while (snapshot_list_has_id(skip, id))
1298 id = __bch2_snapshot_parent(c, id);
1302 id = __bch2_snapshot_parent(c, id);
1303 } while (snapshot_list_has_id(skip, id));
1310 static int bch2_fix_child_of_deleted_snapshot(struct btree_trans *trans,
1311 struct btree_iter *iter, struct bkey_s_c k,
1312 snapshot_id_list *deleted)
1314 struct bch_fs *c = trans->c;
1315 u32 nr_deleted_ancestors = 0;
1316 struct bkey_i_snapshot *s;
1320 if (k.k->type != KEY_TYPE_snapshot)
1323 if (snapshot_list_has_id(deleted, k.k->p.offset))
1326 s = bch2_bkey_make_mut_noupdate_typed(trans, k, snapshot);
1327 ret = PTR_ERR_OR_ZERO(s);
1331 darray_for_each(*deleted, i)
1332 nr_deleted_ancestors += bch2_snapshot_is_ancestor(c, s->k.p.offset, *i);
1334 if (!nr_deleted_ancestors)
1337 le32_add_cpu(&s->v.depth, -nr_deleted_ancestors);
1344 u32 depth = le32_to_cpu(s->v.depth);
1345 u32 parent = bch2_snapshot_parent(c, s->k.p.offset);
1347 for (unsigned j = 0; j < ARRAY_SIZE(s->v.skip); j++) {
1348 u32 id = le32_to_cpu(s->v.skip[j]);
1350 if (snapshot_list_has_id(deleted, id)) {
1352 ? bch2_snapshot_nth_parent_skip(c,
1354 get_random_u32_below(depth - 1),
1357 s->v.skip[j] = cpu_to_le32(id);
1361 bubble_sort(s->v.skip, ARRAY_SIZE(s->v.skip), cmp_le32);
1364 return bch2_trans_update(trans, iter, &s->k_i, 0);
1367 int bch2_delete_dead_snapshots(struct bch_fs *c)
1369 struct btree_trans *trans;
1370 struct btree_iter iter;
1372 struct bkey_s_c_snapshot snap;
1373 snapshot_id_list deleted = { 0 };
1374 snapshot_id_list deleted_interior = { 0 };
1378 if (!test_and_clear_bit(BCH_FS_NEED_DELETE_DEAD_SNAPSHOTS, &c->flags))
1381 if (!test_bit(BCH_FS_STARTED, &c->flags)) {
1382 ret = bch2_fs_read_write_early(c);
1384 bch_err_msg(c, ret, "deleting dead snapshots: error going rw");
1389 trans = bch2_trans_get(c);
1392 * For every snapshot node: If we have no live children and it's not
1393 * pointed to by a subvolume, delete it:
1395 ret = for_each_btree_key_commit(trans, iter, BTREE_ID_snapshots,
1398 bch2_delete_redundant_snapshot(trans, k));
1400 bch_err_msg(c, ret, "deleting redundant snapshots");
1404 ret = for_each_btree_key2(trans, iter, BTREE_ID_snapshots,
1406 bch2_snapshot_set_equiv(trans, k));
1408 bch_err_msg(c, ret, "in bch2_snapshots_set_equiv");
1412 for_each_btree_key(trans, iter, BTREE_ID_snapshots,
1413 POS_MIN, 0, k, ret) {
1414 if (k.k->type != KEY_TYPE_snapshot)
1417 snap = bkey_s_c_to_snapshot(k);
1418 if (BCH_SNAPSHOT_DELETED(snap.v)) {
1419 ret = snapshot_list_add(c, &deleted, k.k->p.offset);
1424 bch2_trans_iter_exit(trans, &iter);
1427 bch_err_msg(c, ret, "walking snapshots");
1431 for (id = 0; id < BTREE_ID_NR; id++) {
1432 struct bpos last_pos = POS_MIN;
1433 snapshot_id_list equiv_seen = { 0 };
1434 struct disk_reservation res = { 0 };
1436 if (!btree_type_has_snapshots(id))
1439 ret = for_each_btree_key_commit(trans, iter,
1441 BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS, k,
1442 &res, NULL, BTREE_INSERT_NOFAIL,
1443 snapshot_delete_key(trans, &iter, k, &deleted, &equiv_seen, &last_pos)) ?:
1444 for_each_btree_key_commit(trans, iter,
1446 BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS, k,
1447 &res, NULL, BTREE_INSERT_NOFAIL,
1448 move_key_to_correct_snapshot(trans, &iter, k));
1450 bch2_disk_reservation_put(c, &res);
1451 darray_exit(&equiv_seen);
1454 bch_err_msg(c, ret, "deleting keys from dying snapshots");
1459 bch2_trans_unlock(trans);
1460 down_write(&c->snapshot_create_lock);
1462 for_each_btree_key(trans, iter, BTREE_ID_snapshots,
1463 POS_MIN, 0, k, ret) {
1464 u32 snapshot = k.k->p.offset;
1465 u32 equiv = bch2_snapshot_equiv(c, snapshot);
1467 if (equiv != snapshot)
1468 snapshot_list_add(c, &deleted_interior, snapshot);
1470 bch2_trans_iter_exit(trans, &iter);
1473 goto err_create_lock;
1476 * Fixing children of deleted snapshots can't be done completely
1477 * atomically, if we crash between here and when we delete the interior
1478 * nodes some depth fields will be off:
1480 ret = for_each_btree_key_commit(trans, iter, BTREE_ID_snapshots, POS_MIN,
1481 BTREE_ITER_INTENT, k,
1482 NULL, NULL, BTREE_INSERT_NOFAIL,
1483 bch2_fix_child_of_deleted_snapshot(trans, &iter, k, &deleted_interior));
1485 goto err_create_lock;
1487 darray_for_each(deleted, i) {
1488 ret = commit_do(trans, NULL, NULL, 0,
1489 bch2_snapshot_node_delete(trans, *i));
1491 bch_err_msg(c, ret, "deleting snapshot %u", *i);
1492 goto err_create_lock;
1496 darray_for_each(deleted_interior, i) {
1497 ret = commit_do(trans, NULL, NULL, 0,
1498 bch2_snapshot_node_delete(trans, *i));
1500 bch_err_msg(c, ret, "deleting snapshot %u", *i);
1501 goto err_create_lock;
1505 up_write(&c->snapshot_create_lock);
1507 darray_exit(&deleted_interior);
1508 darray_exit(&deleted);
1509 bch2_trans_put(trans);
1515 void bch2_delete_dead_snapshots_work(struct work_struct *work)
1517 struct bch_fs *c = container_of(work, struct bch_fs, snapshot_delete_work);
1519 bch2_delete_dead_snapshots(c);
1520 bch2_write_ref_put(c, BCH_WRITE_REF_delete_dead_snapshots);
1523 void bch2_delete_dead_snapshots_async(struct bch_fs *c)
1525 if (bch2_write_ref_tryget(c, BCH_WRITE_REF_delete_dead_snapshots) &&
1526 !queue_work(c->write_ref_wq, &c->snapshot_delete_work))
1527 bch2_write_ref_put(c, BCH_WRITE_REF_delete_dead_snapshots);
1530 int __bch2_key_has_snapshot_overwrites(struct btree_trans *trans,
1534 struct bch_fs *c = trans->c;
1535 struct btree_iter iter;
1539 bch2_trans_iter_init(trans, &iter, id, pos,
1540 BTREE_ITER_NOT_EXTENTS|
1541 BTREE_ITER_ALL_SNAPSHOTS);
1543 k = bch2_btree_iter_prev(&iter);
1551 if (!bkey_eq(pos, k.k->p))
1554 if (bch2_snapshot_is_ancestor(c, k.k->p.snapshot, pos.snapshot)) {
1559 bch2_trans_iter_exit(trans, &iter);
1564 static u32 bch2_snapshot_smallest_child(struct bch_fs *c, u32 id)
1566 const struct snapshot_t *s = snapshot_t(c, id);
1568 return s->children[1] ?: s->children[0];
1571 static u32 bch2_snapshot_smallest_descendent(struct bch_fs *c, u32 id)
1575 while ((child = bch2_snapshot_smallest_child(c, id)))
1580 static int bch2_propagate_key_to_snapshot_leaf(struct btree_trans *trans,
1581 enum btree_id btree,
1582 struct bkey_s_c interior_k,
1583 u32 leaf_id, struct bpos *new_min_pos)
1585 struct btree_iter iter;
1586 struct bpos pos = interior_k.k->p;
1591 pos.snapshot = leaf_id;
1593 bch2_trans_iter_init(trans, &iter, btree, pos, BTREE_ITER_INTENT);
1594 k = bch2_btree_iter_peek_slot(&iter);
1599 /* key already overwritten in this snapshot? */
1600 if (k.k->p.snapshot != interior_k.k->p.snapshot)
1603 if (bpos_eq(*new_min_pos, POS_MIN)) {
1604 *new_min_pos = k.k->p;
1605 new_min_pos->snapshot = leaf_id;
1608 new = bch2_bkey_make_mut_noupdate(trans, interior_k);
1609 ret = PTR_ERR_OR_ZERO(new);
1613 new->k.p.snapshot = leaf_id;
1614 ret = bch2_trans_update(trans, &iter, new, 0);
1616 bch2_trans_iter_exit(trans, &iter);
1620 int bch2_propagate_key_to_snapshot_leaves(struct btree_trans *trans,
1621 enum btree_id btree,
1623 struct bpos *new_min_pos)
1625 struct bch_fs *c = trans->c;
1627 u32 restart_count = trans->restart_count;
1630 bch2_bkey_buf_init(&sk);
1631 bch2_bkey_buf_reassemble(&sk, c, k);
1632 k = bkey_i_to_s_c(sk.k);
1634 *new_min_pos = POS_MIN;
1636 for (u32 id = bch2_snapshot_smallest_descendent(c, k.k->p.snapshot);
1637 id < k.k->p.snapshot;
1639 if (!bch2_snapshot_is_ancestor(c, id, k.k->p.snapshot) ||
1640 !bch2_snapshot_is_leaf(c, id))
1643 ret = btree_trans_too_many_iters(trans) ?:
1644 bch2_propagate_key_to_snapshot_leaf(trans, btree, k, id, new_min_pos) ?:
1645 bch2_trans_commit(trans, NULL, NULL, 0);
1646 if (ret && bch2_err_matches(ret, BCH_ERR_transaction_restart)) {
1647 bch2_trans_begin(trans);
1655 bch2_bkey_buf_exit(&sk, c);
1657 return ret ?: trans_was_restarted(trans, restart_count);
1660 static int bch2_check_snapshot_needs_deletion(struct btree_trans *trans, struct bkey_s_c k)
1662 struct bch_fs *c = trans->c;
1663 struct bkey_s_c_snapshot snap;
1666 if (k.k->type != KEY_TYPE_snapshot)
1669 snap = bkey_s_c_to_snapshot(k);
1670 if (BCH_SNAPSHOT_DELETED(snap.v) ||
1671 bch2_snapshot_equiv(c, k.k->p.offset) != k.k->p.offset ||
1672 (ret = bch2_snapshot_needs_delete(trans, k)) > 0) {
1673 set_bit(BCH_FS_NEED_DELETE_DEAD_SNAPSHOTS, &c->flags);
1680 int bch2_snapshots_read(struct bch_fs *c)
1682 struct btree_iter iter;
1686 ret = bch2_trans_run(c,
1687 for_each_btree_key2(trans, iter, BTREE_ID_snapshots,
1689 bch2_mark_snapshot(trans, BTREE_ID_snapshots, 0, bkey_s_c_null, k, 0) ?:
1690 bch2_snapshot_set_equiv(trans, k) ?:
1691 bch2_check_snapshot_needs_deletion(trans, k)) ?:
1692 for_each_btree_key2(trans, iter, BTREE_ID_snapshots,
1694 (set_is_ancestor_bitmap(c, k.k->p.offset), 0)));
1700 void bch2_fs_snapshots_exit(struct bch_fs *c)
1702 kfree(rcu_dereference_protected(c->snapshots, true));