+
+static u32 bch2_snapshot_child(struct bch_fs *c, u32 id, unsigned child)
+{
+ return snapshot_t(c, id)->children[child];
+}
+
+static u32 bch2_snapshot_left_child(struct bch_fs *c, u32 id)
+{
+ return bch2_snapshot_child(c, id, 0);
+}
+
+static u32 bch2_snapshot_right_child(struct bch_fs *c, u32 id)
+{
+ return bch2_snapshot_child(c, id, 1);
+}
+
+static u32 bch2_snapshot_tree_next(struct bch_fs *c, u32 id)
+{
+ u32 n, parent;
+
+ n = bch2_snapshot_left_child(c, id);
+ if (n)
+ return n;
+
+ while ((parent = bch2_snapshot_parent(c, id))) {
+ n = bch2_snapshot_right_child(c, parent);
+ if (n && n != id)
+ return n;
+ id = parent;
+ }
+
+ return 0;
+}
+
+static u32 bch2_snapshot_tree_oldest_subvol(struct bch_fs *c, u32 snapshot_root)
+{
+ u32 id = snapshot_root;
+ u32 subvol = 0, s;
+
+ while (id) {
+ s = snapshot_t(c, id)->subvol;
+
+ if (s && (!subvol || s < subvol))
+ subvol = s;
+
+ id = bch2_snapshot_tree_next(c, id);
+ }
+
+ return subvol;
+}
+
+static int bch2_snapshot_tree_master_subvol(struct btree_trans *trans,
+ u32 snapshot_root, u32 *subvol_id)
+{
+ struct bch_fs *c = trans->c;
+ struct btree_iter iter;
+ struct bkey_s_c k;
+ struct bkey_s_c_subvolume s;
+ int ret;
+
+ for_each_btree_key_norestart(trans, iter, BTREE_ID_subvolumes, POS_MIN,
+ 0, k, ret) {
+ if (k.k->type != KEY_TYPE_subvolume)
+ continue;
+
+ s = bkey_s_c_to_subvolume(k);
+ if (!bch2_snapshot_is_ancestor(c, le32_to_cpu(s.v->snapshot), snapshot_root))
+ continue;
+ if (!BCH_SUBVOLUME_SNAP(s.v)) {
+ *subvol_id = s.k->p.offset;
+ goto found;
+ }
+ }
+ ret = ret ?: -ENOENT;
+found:
+ bch2_trans_iter_exit(trans, &iter);
+
+ if (ret == -ENOENT) {
+ struct bkey_i_subvolume *s;
+
+ *subvol_id = bch2_snapshot_tree_oldest_subvol(c, snapshot_root);
+
+ s = bch2_bkey_get_mut_typed(trans, &iter,
+ BTREE_ID_subvolumes, POS(0, *subvol_id),
+ 0, subvolume);
+ ret = PTR_ERR_OR_ZERO(s);
+ if (ret)
+ return ret;
+
+ SET_BCH_SUBVOLUME_SNAP(&s->v, false);
+ }
+
+ return ret;
+}
+
+static int check_snapshot_tree(struct btree_trans *trans,
+ struct btree_iter *iter,
+ struct bkey_s_c k)
+{
+ struct bch_fs *c = trans->c;
+ struct bkey_s_c_snapshot_tree st;
+ struct bch_snapshot s;
+ struct bch_subvolume subvol;
+ struct printbuf buf = PRINTBUF;
+ u32 root_id;
+ int ret;
+
+ if (k.k->type != KEY_TYPE_snapshot_tree)
+ return 0;
+
+ st = bkey_s_c_to_snapshot_tree(k);
+ root_id = le32_to_cpu(st.v->root_snapshot);
+
+ ret = snapshot_lookup(trans, root_id, &s);
+ if (ret && !bch2_err_matches(ret, ENOENT))
+ goto err;
+
+ if (fsck_err_on(ret ||
+ root_id != bch2_snapshot_root(c, root_id) ||
+ st.k->p.offset != le32_to_cpu(s.tree),
+ c,
+ "snapshot tree points to missing/incorrect snapshot:\n %s",
+ (bch2_bkey_val_to_text(&buf, c, st.s_c), buf.buf))) {
+ ret = bch2_btree_delete_at(trans, iter, 0);
+ goto err;
+ }
+
+ ret = bch2_subvolume_get(trans, le32_to_cpu(st.v->master_subvol),
+ false, 0, &subvol);
+ if (ret && !bch2_err_matches(ret, ENOENT))
+ goto err;
+
+ if (fsck_err_on(ret ||
+ !bch2_snapshot_is_ancestor(c,
+ le32_to_cpu(subvol.snapshot),
+ root_id) ||
+ !BCH_SUBVOLUME_SNAP(&subvol), c,
+ "snapshot tree points to missing/incorrect subvolume:\n %s",
+ (printbuf_reset(&buf),
+ bch2_bkey_val_to_text(&buf, c, st.s_c), buf.buf))) {
+ struct bkey_i_snapshot_tree *u;
+ u32 subvol_id;
+
+ ret = bch2_snapshot_tree_master_subvol(trans, root_id, &subvol_id);
+ if (ret)
+ goto err;
+
+ u = bch2_bkey_make_mut_typed(trans, iter, k, 0, snapshot_tree);
+ ret = PTR_ERR_OR_ZERO(u);
+ if (ret)
+ goto err;
+
+ u->v.master_subvol = cpu_to_le32(subvol_id);
+ st = snapshot_tree_i_to_s_c(u);
+ }
+err:
+fsck_err:
+ printbuf_exit(&buf);
+ return ret;
+}
+
+/*
+ * For each snapshot_tree, make sure it points to the root of a snapshot tree
+ * and that snapshot entry points back to it, or delete it.
+ *
+ * And, make sure it points to a subvolume within that snapshot tree, or correct
+ * it to point to the oldest subvolume within that snapshot tree.
+ */
+int bch2_fs_check_snapshot_trees(struct bch_fs *c)
+{
+ struct btree_iter iter;
+ struct bkey_s_c k;
+ int ret;
+
+ ret = bch2_trans_run(c,
+ for_each_btree_key_commit(&trans, iter,
+ BTREE_ID_snapshot_trees, POS_MIN,
+ BTREE_ITER_PREFETCH, k,
+ NULL, NULL, BTREE_INSERT_LAZY_RW|BTREE_INSERT_NOFAIL,
+ check_snapshot_tree(&trans, &iter, k)));
+
+ if (ret)
+ bch_err(c, "error %i checking snapshot trees", ret);
+ return ret;
+}
+
+/*
+ * Look up snapshot tree for @tree_id and find root,
+ * make sure @snap_id is a descendent:
+ */
+static int snapshot_tree_ptr_good(struct btree_trans *trans,
+ u32 snap_id, u32 tree_id)
+{
+ struct bch_snapshot_tree s_t;
+ int ret = bch2_snapshot_tree_lookup(trans, tree_id, &s_t);
+
+ if (ret)
+ return ret;
+
+ return bch2_snapshot_is_ancestor(trans->c, snap_id, le32_to_cpu(s_t.root_snapshot));
+}
+
+/*
+ * snapshot_tree pointer was incorrect: look up root snapshot node, make sure
+ * its snapshot_tree pointer is correct (allocate new one if necessary), then
+ * update this node's pointer to root node's pointer:
+ */
+static int snapshot_tree_ptr_repair(struct btree_trans *trans,
+ struct btree_iter *iter,
+ struct bkey_s_c_snapshot *s)
+{
+ struct bch_fs *c = trans->c;
+ struct btree_iter root_iter;
+ struct bch_snapshot_tree s_t;
+ struct bkey_s_c_snapshot root;
+ struct bkey_i_snapshot *u;
+ u32 root_id = bch2_snapshot_root(c, s->k->p.offset), tree_id;
+ int ret;
+
+ root = bch2_bkey_get_iter_typed(trans, &root_iter,
+ BTREE_ID_snapshots, POS(0, root_id),
+ BTREE_ITER_WITH_UPDATES, snapshot);
+ ret = bkey_err(root);
+ if (ret)
+ goto err;
+
+ tree_id = le32_to_cpu(root.v->tree);
+
+ ret = bch2_snapshot_tree_lookup(trans, tree_id, &s_t);
+ if (ret)
+ return ret;
+
+ if (le32_to_cpu(s_t.root_snapshot) != root_id) {
+ u = bch2_bkey_make_mut_typed(trans, &root_iter, root.s_c, 0, snapshot);
+ ret = PTR_ERR_OR_ZERO(u) ?:
+ snapshot_tree_create(trans, root_id,
+ bch2_snapshot_tree_oldest_subvol(c, root_id),
+ &tree_id);
+ if (ret)
+ goto err;
+
+ u->v.tree = cpu_to_le32(tree_id);
+ if (s->k->p.snapshot == root_id)
+ *s = snapshot_i_to_s_c(u);
+ }
+
+ if (s->k->p.snapshot != root_id) {
+ u = bch2_bkey_make_mut_typed(trans, iter, s->s_c, 0, snapshot);
+ ret = PTR_ERR_OR_ZERO(u);
+ if (ret)
+ goto err;
+
+ u->v.tree = cpu_to_le32(tree_id);
+ *s = snapshot_i_to_s_c(u);
+ }
+err:
+ bch2_trans_iter_exit(trans, &root_iter);
+ return ret;
+}
+