]> git.sesse.net Git - bcachefs-tools-debian/blobdiff - libbcachefs/subvolume.c
Update bcachefs sources to 33a60d9b05 bcachefs: Assorted fixes for clang
[bcachefs-tools-debian] / libbcachefs / subvolume.c
index 388fa12bbd8b497e9eac733461553c487547978e..736afb626a8945b75d4bc3300fb15fb60d594de8 100644 (file)
@@ -8,8 +8,107 @@
 #include "fs.h"
 #include "subvolume.h"
 
+#include <linux/random.h>
+
 static int bch2_subvolume_delete(struct btree_trans *, u32);
 
+static inline u32 get_ancestor_below(struct snapshot_table *t, u32 id, u32 ancestor)
+{
+       const struct snapshot_t *s = __snapshot_t(t, id);
+
+       if (s->skip[2] <= ancestor)
+               return s->skip[2];
+       if (s->skip[1] <= ancestor)
+               return s->skip[1];
+       if (s->skip[0] <= ancestor)
+               return s->skip[0];
+       return s->parent;
+}
+
+bool __bch2_snapshot_is_ancestor(struct bch_fs *c, u32 id, u32 ancestor)
+{
+       struct snapshot_table *t;
+       bool ret;
+
+       EBUG_ON(c->curr_recovery_pass <= BCH_RECOVERY_PASS_check_snapshots);
+
+       rcu_read_lock();
+       t = rcu_dereference(c->snapshots);
+
+       while (id && id < ancestor - IS_ANCESTOR_BITMAP)
+               id = get_ancestor_below(t, id, ancestor);
+
+       ret = id && id < ancestor
+               ? test_bit(ancestor - id - 1, __snapshot_t(t, id)->is_ancestor)
+               : id == ancestor;
+       rcu_read_unlock();
+
+       return ret;
+}
+
+static bool bch2_snapshot_is_ancestor_early(struct bch_fs *c, u32 id, u32 ancestor)
+{
+       struct snapshot_table *t;
+
+       rcu_read_lock();
+       t = rcu_dereference(c->snapshots);
+
+       while (id && id < ancestor)
+               id = __snapshot_t(t, id)->parent;
+       rcu_read_unlock();
+
+       return id == ancestor;
+}
+
+static inline u32 bch2_snapshot_depth(struct bch_fs *c, u32 parent)
+{
+       u32 depth;
+
+       rcu_read_lock();
+       depth = parent ? snapshot_t(c, parent)->depth + 1 : 0;
+       rcu_read_unlock();
+
+       return depth;
+}
+
+static noinline struct snapshot_t *__snapshot_t_mut(struct bch_fs *c, u32 id)
+{
+       size_t idx = U32_MAX - id;
+       size_t new_size;
+       struct snapshot_table *new, *old;
+
+       new_size = max(16UL, roundup_pow_of_two(idx + 1));
+
+       new = kvzalloc(struct_size(new, s, new_size), GFP_KERNEL);
+       if (!new)
+               return NULL;
+
+       old = rcu_dereference_protected(c->snapshots, true);
+       if (old)
+               memcpy(new->s,
+                      rcu_dereference_protected(c->snapshots, true)->s,
+                      sizeof(new->s[0]) * c->snapshot_table_size);
+
+       rcu_assign_pointer(c->snapshots, new);
+       c->snapshot_table_size = new_size;
+       if (old)
+               kvfree_rcu(old);
+
+       return &rcu_dereference_protected(c->snapshots, true)->s[idx];
+}
+
+static inline struct snapshot_t *snapshot_t_mut(struct bch_fs *c, u32 id)
+{
+       size_t idx = U32_MAX - id;
+
+       lockdep_assert_held(&c->snapshot_table_lock);
+
+       if (likely(idx < c->snapshot_table_size))
+               return &rcu_dereference_protected(c->snapshots, true)->s[idx];
+
+       return __snapshot_t_mut(c, id);
+}
+
 /* Snapshot tree: */
 
 void bch2_snapshot_tree_to_text(struct printbuf *out, struct bch_fs *c,
@@ -23,7 +122,8 @@ void bch2_snapshot_tree_to_text(struct printbuf *out, struct bch_fs *c,
 }
 
 int bch2_snapshot_tree_invalid(const struct bch_fs *c, struct bkey_s_c k,
-                              unsigned flags, struct printbuf *err)
+                              enum bkey_invalid_flags flags,
+                              struct printbuf *err)
 {
        if (bkey_gt(k.k->p, POS(0, U32_MAX)) ||
            bkey_lt(k.k->p, POS(0, 1))) {
@@ -37,8 +137,12 @@ int bch2_snapshot_tree_invalid(const struct bch_fs *c, struct bkey_s_c k,
 int bch2_snapshot_tree_lookup(struct btree_trans *trans, u32 id,
                              struct bch_snapshot_tree *s)
 {
-       return bch2_bkey_get_val_typed(trans, BTREE_ID_snapshot_trees, POS(0, id),
-                                      BTREE_ITER_WITH_UPDATES, snapshot_tree, s);
+       int ret = bch2_bkey_get_val_typed(trans, BTREE_ID_snapshot_trees, POS(0, id),
+                                         BTREE_ITER_WITH_UPDATES, snapshot_tree, s);
+
+       if (bch2_err_matches(ret, ENOENT))
+               ret = -BCH_ERR_ENOENT_snapshot_tree;
+       return ret;
 }
 
 static struct bkey_i_snapshot_tree *
@@ -82,17 +186,26 @@ void bch2_snapshot_to_text(struct printbuf *out, struct bch_fs *c,
 {
        struct bkey_s_c_snapshot s = bkey_s_c_to_snapshot(k);
 
-       prt_printf(out, "is_subvol %llu deleted %llu parent %10u children %10u %10u subvol %u",
+       prt_printf(out, "is_subvol %llu deleted %llu parent %10u children %10u %10u subvol %u tree %u",
               BCH_SNAPSHOT_SUBVOL(s.v),
               BCH_SNAPSHOT_DELETED(s.v),
               le32_to_cpu(s.v->parent),
               le32_to_cpu(s.v->children[0]),
               le32_to_cpu(s.v->children[1]),
-              le32_to_cpu(s.v->subvol));
+              le32_to_cpu(s.v->subvol),
+              le32_to_cpu(s.v->tree));
+
+       if (bkey_val_bytes(k.k) > offsetof(struct bch_snapshot, depth))
+               prt_printf(out, " depth %u skiplist %u %u %u",
+                          le32_to_cpu(s.v->depth),
+                          le32_to_cpu(s.v->skip[0]),
+                          le32_to_cpu(s.v->skip[1]),
+                          le32_to_cpu(s.v->skip[2]));
 }
 
 int bch2_snapshot_invalid(const struct bch_fs *c, struct bkey_s_c k,
-                         unsigned flags, struct printbuf *err)
+                         enum bkey_invalid_flags flags,
+                         struct printbuf *err)
 {
        struct bkey_s_c_snapshot s;
        u32 i, id;
@@ -133,6 +246,25 @@ int bch2_snapshot_invalid(const struct bch_fs *c, struct bkey_s_c k,
                }
        }
 
+       if (bkey_val_bytes(k.k) > offsetof(struct bch_snapshot, skip)) {
+               if (le32_to_cpu(s.v->skip[0]) > le32_to_cpu(s.v->skip[1]) ||
+                   le32_to_cpu(s.v->skip[1]) > le32_to_cpu(s.v->skip[2])) {
+                       prt_printf(err, "skiplist not normalized");
+                       return -BCH_ERR_invalid_bkey;
+               }
+
+               for (i = 0; i < ARRAY_SIZE(s.v->skip); i++) {
+                       id = le32_to_cpu(s.v->skip[i]);
+
+                       if (!id != !s.v->parent ||
+                           (s.v->parent &&
+                            id <= k.k->p.offset)) {
+                               prt_printf(err, "bad skiplist node %u)", id);
+                               return -BCH_ERR_invalid_bkey;
+                       }
+               }
+       }
+
        return 0;
 }
 
@@ -143,30 +275,53 @@ int bch2_mark_snapshot(struct btree_trans *trans,
 {
        struct bch_fs *c = trans->c;
        struct snapshot_t *t;
+       u32 id = new.k->p.offset;
+       int ret = 0;
+
+       mutex_lock(&c->snapshot_table_lock);
 
-       t = genradix_ptr_alloc(&c->snapshots,
-                              U32_MAX - new.k->p.offset,
-                              GFP_KERNEL);
-       if (!t)
-               return -BCH_ERR_ENOMEM_mark_snapshot;
+       t = snapshot_t_mut(c, id);
+       if (!t) {
+               ret = -BCH_ERR_ENOMEM_mark_snapshot;
+               goto err;
+       }
 
        if (new.k->type == KEY_TYPE_snapshot) {
                struct bkey_s_c_snapshot s = bkey_s_c_to_snapshot(new);
+               u32 parent = id;
 
                t->parent       = le32_to_cpu(s.v->parent);
                t->children[0]  = le32_to_cpu(s.v->children[0]);
                t->children[1]  = le32_to_cpu(s.v->children[1]);
                t->subvol       = BCH_SNAPSHOT_SUBVOL(s.v) ? le32_to_cpu(s.v->subvol) : 0;
                t->tree         = le32_to_cpu(s.v->tree);
+
+               if (bkey_val_bytes(s.k) > offsetof(struct bch_snapshot, depth)) {
+                       t->depth        = le32_to_cpu(s.v->depth);
+                       t->skip[0]      = le32_to_cpu(s.v->skip[0]);
+                       t->skip[1]      = le32_to_cpu(s.v->skip[1]);
+                       t->skip[2]      = le32_to_cpu(s.v->skip[2]);
+               } else {
+                       t->depth        = 0;
+                       t->skip[0]      = 0;
+                       t->skip[1]      = 0;
+                       t->skip[2]      = 0;
+               }
+
+               while ((parent = bch2_snapshot_parent_early(c, parent)) &&
+                      parent - id - 1 < IS_ANCESTOR_BITMAP)
+                       __set_bit(parent - id - 1, t->is_ancestor);
+
+               if (BCH_SNAPSHOT_DELETED(s.v)) {
+                       set_bit(BCH_FS_HAVE_DELETED_SNAPSHOTS, &c->flags);
+                       c->recovery_passes_explicit |= BIT_ULL(BCH_RECOVERY_PASS_delete_dead_snapshots);
+               }
        } else {
-               t->parent       = 0;
-               t->children[0]  = 0;
-               t->children[1]  = 0;
-               t->subvol       = 0;
-               t->tree         = 0;
+               memset(t, 0, sizeof(*t));
        }
-
-       return 0;
+err:
+       mutex_unlock(&c->snapshot_table_lock);
+       return ret;
 }
 
 static int snapshot_lookup(struct btree_trans *trans, u32 id,
@@ -219,9 +374,14 @@ static int bch2_snapshot_set_equiv(struct btree_trans *trans, struct bkey_s_c k)
                nr_live += ret;
        }
 
-       snapshot_t(c, id)->equiv = nr_live == 1
-               ? snapshot_t(c, child[live_idx])->equiv
+       mutex_lock(&c->snapshot_table_lock);
+
+       snapshot_t_mut(c, id)->equiv = nr_live == 1
+               ? snapshot_t_mut(c, child[live_idx])->equiv
                : id;
+
+       mutex_unlock(&c->snapshot_table_lock);
+
        return 0;
 }
 
@@ -284,6 +444,7 @@ static int bch2_snapshot_tree_master_subvol(struct btree_trans *trans,
        struct btree_iter iter;
        struct bkey_s_c k;
        struct bkey_s_c_subvolume s;
+       bool found = false;
        int ret;
 
        for_each_btree_key_norestart(trans, iter, BTREE_ID_subvolumes, POS_MIN,
@@ -296,14 +457,14 @@ static int bch2_snapshot_tree_master_subvol(struct btree_trans *trans,
                        continue;
                if (!BCH_SUBVOLUME_SNAP(s.v)) {
                        *subvol_id = s.k->p.offset;
-                       goto found;
+                       found = true;
+                       break;
                }
        }
-       ret = ret ?: -ENOENT;
-found:
+
        bch2_trans_iter_exit(trans, &iter);
 
-       if (bch2_err_matches(ret, ENOENT)) {
+       if (!ret && !found) {
                struct bkey_i_subvolume *s;
 
                *subvol_id = bch2_snapshot_tree_oldest_subvol(c, snapshot_root);
@@ -362,9 +523,9 @@ static int check_snapshot_tree(struct btree_trans *trans,
                        "snapshot tree points to missing subvolume:\n  %s",
                        (printbuf_reset(&buf),
                         bch2_bkey_val_to_text(&buf, c, st.s_c), buf.buf)) ||
-           fsck_err_on(!bch2_snapshot_is_ancestor(c,
-                                                  le32_to_cpu(subvol.snapshot),
-                                                  root_id), c,
+           fsck_err_on(!bch2_snapshot_is_ancestor_early(c,
+                                               le32_to_cpu(subvol.snapshot),
+                                               root_id), c,
                        "snapshot tree points to subvolume that does not point to snapshot in this tree:\n  %s",
                        (printbuf_reset(&buf),
                         bch2_bkey_val_to_text(&buf, c, st.s_c), buf.buf)) ||
@@ -379,7 +540,7 @@ static int check_snapshot_tree(struct btree_trans *trans,
                if (ret)
                        goto err;
 
-               u = bch2_bkey_make_mut_typed(trans, iter, k, 0, snapshot_tree);
+               u = bch2_bkey_make_mut_typed(trans, iter, &k, 0, snapshot_tree);
                ret = PTR_ERR_OR_ZERO(u);
                if (ret)
                        goto err;
@@ -400,7 +561,7 @@ fsck_err:
  * 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)
+int bch2_check_snapshot_trees(struct bch_fs *c)
 {
        struct btree_iter iter;
        struct bkey_s_c k;
@@ -433,7 +594,49 @@ static int snapshot_tree_ptr_good(struct btree_trans *trans,
        if (ret)
                return ret;
 
-       return bch2_snapshot_is_ancestor(trans->c, snap_id, le32_to_cpu(s_t.root_snapshot));
+       return bch2_snapshot_is_ancestor_early(trans->c, snap_id, le32_to_cpu(s_t.root_snapshot));
+}
+
+static u32 snapshot_skiplist_get(struct bch_fs *c, u32 id)
+{
+       const struct snapshot_t *s;
+
+       if (!id)
+               return 0;
+
+       rcu_read_lock();
+       s = snapshot_t(c, id);
+       if (s->parent)
+               id = bch2_snapshot_nth_parent(c, id, get_random_u32_below(s->depth));
+       rcu_read_unlock();
+
+       return id;
+}
+
+static int snapshot_skiplist_good(struct btree_trans *trans, struct bch_snapshot s)
+{
+       struct bch_snapshot a;
+       unsigned i;
+       int ret;
+
+       for (i = 0; i < 3; i++) {
+               if (!s.parent != !s.skip[i])
+                       return false;
+
+               if (!s.parent)
+                       continue;
+
+               ret = snapshot_lookup(trans, le32_to_cpu(s.skip[i]), &a);
+               if (bch2_err_matches(ret, ENOENT))
+                       return false;
+               if (ret)
+                       return ret;
+
+               if (a.tree != s.tree)
+                       return false;
+       }
+
+       return true;
 }
 
 /*
@@ -443,14 +646,15 @@ static int snapshot_tree_ptr_good(struct btree_trans *trans,
  */
 static int snapshot_tree_ptr_repair(struct btree_trans *trans,
                                    struct btree_iter *iter,
-                                   struct bkey_s_c_snapshot *s)
+                                   struct bkey_s_c k,
+                                   struct bch_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;
+       u32 root_id = bch2_snapshot_root(c, k.k->p.offset), tree_id;
        int ret;
 
        root = bch2_bkey_get_iter_typed(trans, &root_iter,
@@ -467,7 +671,7 @@ static int snapshot_tree_ptr_repair(struct btree_trans *trans,
                return ret;
 
        if (ret || le32_to_cpu(s_t.root_snapshot) != root_id) {
-               u = bch2_bkey_make_mut_typed(trans, &root_iter, root.s_c, 0, snapshot);
+               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),
@@ -476,32 +680,40 @@ static int snapshot_tree_ptr_repair(struct btree_trans *trans,
                        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 (k.k->p.offset == root_id)
+                       *s = u->v;
        }
 
-       if (s->k->p.snapshot != root_id) {
-               u = bch2_bkey_make_mut_typed(trans, iter, s->s_c, 0, snapshot);
+       if (k.k->p.offset != root_id) {
+               u = bch2_bkey_make_mut_typed(trans, iter, &k, 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);
+               *s = u->v;
        }
 err:
        bch2_trans_iter_exit(trans, &root_iter);
        return ret;
 }
 
+static int cmp_le32(__le32 l, __le32 r)
+{
+       return cmp_int(le32_to_cpu(l), le32_to_cpu(r));
+}
+
 static int check_snapshot(struct btree_trans *trans,
                          struct btree_iter *iter,
                          struct bkey_s_c k)
 {
        struct bch_fs *c = trans->c;
-       struct bkey_s_c_snapshot s;
+       struct bch_snapshot s;
        struct bch_subvolume subvol;
        struct bch_snapshot v;
+       struct bkey_i_snapshot *u;
+       u32 parent_id = bch2_snapshot_parent_early(c, k.k->p.offset);
+       u32 real_depth;
        struct printbuf buf = PRINTBUF;
        bool should_have_subvol;
        u32 i, id;
@@ -510,114 +722,147 @@ static int check_snapshot(struct btree_trans *trans,
        if (k.k->type != KEY_TYPE_snapshot)
                return 0;
 
-       s = bkey_s_c_to_snapshot(k);
-       id = le32_to_cpu(s.v->parent);
+       memset(&s, 0, sizeof(s));
+       memcpy(&s, k.v, bkey_val_bytes(k.k));
+
+       id = le32_to_cpu(s.parent);
        if (id) {
                ret = snapshot_lookup(trans, id, &v);
                if (bch2_err_matches(ret, ENOENT))
                        bch_err(c, "snapshot with nonexistent parent:\n  %s",
-                               (bch2_bkey_val_to_text(&buf, c, s.s_c), buf.buf));
+                               (bch2_bkey_val_to_text(&buf, c, k), buf.buf));
                if (ret)
                        goto err;
 
-               if (le32_to_cpu(v.children[0]) != s.k->p.offset &&
-                   le32_to_cpu(v.children[1]) != s.k->p.offset) {
+               if (le32_to_cpu(v.children[0]) != k.k->p.offset &&
+                   le32_to_cpu(v.children[1]) != k.k->p.offset) {
                        bch_err(c, "snapshot parent %u missing pointer to child %llu",
-                               id, s.k->p.offset);
+                               id, k.k->p.offset);
                        ret = -EINVAL;
                        goto err;
                }
        }
 
-       for (i = 0; i < 2 && s.v->children[i]; i++) {
-               id = le32_to_cpu(s.v->children[i]);
+       for (i = 0; i < 2 && s.children[i]; i++) {
+               id = le32_to_cpu(s.children[i]);
 
                ret = snapshot_lookup(trans, id, &v);
                if (bch2_err_matches(ret, ENOENT))
                        bch_err(c, "snapshot node %llu has nonexistent child %u",
-                               s.k->p.offset, id);
+                               k.k->p.offset, id);
                if (ret)
                        goto err;
 
-               if (le32_to_cpu(v.parent) != s.k->p.offset) {
+               if (le32_to_cpu(v.parent) != k.k->p.offset) {
                        bch_err(c, "snapshot child %u has wrong parent (got %u should be %llu)",
-                               id, le32_to_cpu(v.parent), s.k->p.offset);
+                               id, le32_to_cpu(v.parent), k.k->p.offset);
                        ret = -EINVAL;
                        goto err;
                }
        }
 
-       should_have_subvol = BCH_SNAPSHOT_SUBVOL(s.v) &&
-               !BCH_SNAPSHOT_DELETED(s.v);
+       should_have_subvol = BCH_SNAPSHOT_SUBVOL(&s) &&
+               !BCH_SNAPSHOT_DELETED(&s);
 
        if (should_have_subvol) {
-               id = le32_to_cpu(s.v->subvol);
+               id = le32_to_cpu(s.subvol);
                ret = bch2_subvolume_get(trans, id, 0, false, &subvol);
                if (bch2_err_matches(ret, ENOENT))
                        bch_err(c, "snapshot points to nonexistent subvolume:\n  %s",
-                               (bch2_bkey_val_to_text(&buf, c, s.s_c), buf.buf));
+                               (bch2_bkey_val_to_text(&buf, c, k), buf.buf));
                if (ret)
                        goto err;
 
-               if (BCH_SNAPSHOT_SUBVOL(s.v) != (le32_to_cpu(subvol.snapshot) == s.k->p.offset)) {
+               if (BCH_SNAPSHOT_SUBVOL(&s) != (le32_to_cpu(subvol.snapshot) == k.k->p.offset)) {
                        bch_err(c, "snapshot node %llu has wrong BCH_SNAPSHOT_SUBVOL",
-                               s.k->p.offset);
+                               k.k->p.offset);
                        ret = -EINVAL;
                        goto err;
                }
        } else {
-               if (fsck_err_on(s.v->subvol, c, "snapshot should not point to subvol:\n  %s",
-                               (bch2_bkey_val_to_text(&buf, c, s.s_c), buf.buf))) {
-                       struct bkey_i_snapshot *u = bch2_trans_kmalloc(trans, sizeof(*u));
-
+               if (fsck_err_on(s.subvol, c, "snapshot should not point to subvol:\n  %s",
+                               (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
+                       u = bch2_bkey_make_mut_typed(trans, iter, &k, 0, snapshot);
                        ret = PTR_ERR_OR_ZERO(u);
                        if (ret)
                                goto err;
 
-                       bkey_reassemble(&u->k_i, s.s_c);
                        u->v.subvol = 0;
-                       ret = bch2_trans_update(trans, iter, &u->k_i, 0);
-                       if (ret)
-                               goto err;
-
-                       s = snapshot_i_to_s_c(u);
+                       s = u->v;
                }
        }
 
-       ret = snapshot_tree_ptr_good(trans, s.k->p.offset, le32_to_cpu(s.v->tree));
+       ret = snapshot_tree_ptr_good(trans, k.k->p.offset, le32_to_cpu(s.tree));
        if (ret < 0)
                goto err;
 
        if (fsck_err_on(!ret, c, "snapshot points to missing/incorrect tree:\n  %s",
-                       (bch2_bkey_val_to_text(&buf, c, s.s_c), buf.buf))) {
-               ret = snapshot_tree_ptr_repair(trans, iter, &s);
+                       (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
+               ret = snapshot_tree_ptr_repair(trans, iter, k, &s);
                if (ret)
                        goto err;
        }
        ret = 0;
 
-       if (BCH_SNAPSHOT_DELETED(s.v))
-               set_bit(BCH_FS_HAVE_DELETED_SNAPSHOTS, &c->flags);
+       real_depth = bch2_snapshot_depth(c, parent_id);
+
+       if (le32_to_cpu(s.depth) != real_depth &&
+           (c->sb.version_upgrade_complete < bcachefs_metadata_version_snapshot_skiplists ||
+            fsck_err(c, "snapshot with incorrect depth field, should be %u:\n  %s",
+                     real_depth, (bch2_bkey_val_to_text(&buf, c, k), buf.buf)))) {
+               u = bch2_bkey_make_mut_typed(trans, iter, &k, 0, snapshot);
+               ret = PTR_ERR_OR_ZERO(u);
+               if (ret)
+                       goto err;
+
+               u->v.depth = cpu_to_le32(real_depth);
+               s = u->v;
+       }
+
+       ret = snapshot_skiplist_good(trans, s);
+       if (ret < 0)
+               goto err;
+
+       if (!ret &&
+           (c->sb.version_upgrade_complete < bcachefs_metadata_version_snapshot_skiplists ||
+            fsck_err(c, "snapshot with bad skiplist field:\n  %s",
+                     (bch2_bkey_val_to_text(&buf, c, k), buf.buf)))) {
+               u = bch2_bkey_make_mut_typed(trans, iter, &k, 0, snapshot);
+               ret = PTR_ERR_OR_ZERO(u);
+               if (ret)
+                       goto err;
+
+               for (i = 0; i < ARRAY_SIZE(u->v.skip); i++)
+                       u->v.skip[i] = cpu_to_le32(snapshot_skiplist_get(c, parent_id));
+
+               bubble_sort(u->v.skip, ARRAY_SIZE(u->v.skip), cmp_le32);
+               s = u->v;
+       }
+       ret = 0;
 err:
 fsck_err:
        printbuf_exit(&buf);
        return ret;
 }
 
-int bch2_fs_check_snapshots(struct bch_fs *c)
+int bch2_check_snapshots(struct bch_fs *c)
 {
        struct btree_iter iter;
        struct bkey_s_c k;
        int ret;
 
+       /*
+        * We iterate backwards as checking/fixing the depth field requires that
+        * the parent's depth already be correct:
+        */
        ret = bch2_trans_run(c,
-               for_each_btree_key_commit(&trans, iter,
-                       BTREE_ID_snapshots, POS_MIN,
+               for_each_btree_key_reverse_commit(&trans, iter,
+                       BTREE_ID_snapshots, POS_MAX,
                        BTREE_ITER_PREFETCH, k,
                        NULL, NULL, BTREE_INSERT_LAZY_RW|BTREE_INSERT_NOFAIL,
                check_snapshot(&trans, &iter, k)));
        if (ret)
-               bch_err(c, "%s: error %s", __func__, bch2_err_str(ret));
+               bch_err_fn(c, ret);
        return ret;
 }
 
@@ -656,9 +901,13 @@ static int check_subvol(struct btree_trans *trans,
 
        if (!BCH_SUBVOLUME_SNAP(subvol.v)) {
                u32 snapshot_root = bch2_snapshot_root(c, le32_to_cpu(subvol.v->snapshot));
-               u32 snapshot_tree = snapshot_t(c, snapshot_root)->tree;
+               u32 snapshot_tree;
                struct bch_snapshot_tree st;
 
+               rcu_read_lock();
+               snapshot_tree = snapshot_t(c, snapshot_root)->tree;
+               rcu_read_unlock();
+
                ret = bch2_snapshot_tree_lookup(trans, snapshot_tree, &st);
 
                bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOENT), c,
@@ -671,7 +920,7 @@ static int check_subvol(struct btree_trans *trans,
                                "subvolume %llu is not set as snapshot but is not master subvolume",
                                k.k->p.offset)) {
                        struct bkey_i_subvolume *s =
-                               bch2_bkey_make_mut_typed(trans, iter, subvol.s_c, 0, subvolume);
+                               bch2_bkey_make_mut_typed(trans, iter, &subvol.s_c, 0, subvolume);
                        ret = PTR_ERR_OR_ZERO(s);
                        if (ret)
                                return ret;
@@ -684,7 +933,7 @@ fsck_err:
        return ret;
 }
 
-int bch2_fs_check_subvols(struct bch_fs *c)
+int bch2_check_subvols(struct bch_fs *c)
 {
        struct btree_iter iter;
        struct bkey_s_c k;
@@ -696,17 +945,16 @@ int bch2_fs_check_subvols(struct bch_fs *c)
                        NULL, NULL, BTREE_INSERT_LAZY_RW|BTREE_INSERT_NOFAIL,
                check_subvol(&trans, &iter, k)));
        if (ret)
-               bch_err(c, "%s: error %s", __func__, bch2_err_str(ret));
-
+               bch_err_fn(c, ret);
        return ret;
 }
 
 void bch2_fs_snapshots_exit(struct bch_fs *c)
 {
-       genradix_free(&c->snapshots);
+       kfree(rcu_dereference_protected(c->snapshots, true));
 }
 
-int bch2_fs_snapshots_start(struct bch_fs *c)
+int bch2_snapshots_read(struct bch_fs *c)
 {
        struct btree_iter iter;
        struct bkey_s_c k;
@@ -718,7 +966,7 @@ int bch2_fs_snapshots_start(struct bch_fs *c)
                        bch2_mark_snapshot(&trans, BTREE_ID_snapshots, 0, bkey_s_c_null, k, 0) ?:
                        bch2_snapshot_set_equiv(&trans, k)));
        if (ret)
-               bch_err(c, "error starting snapshots: %s", bch2_err_str(ret));
+               bch_err_fn(c, ret);
        return ret;
 }
 
@@ -820,7 +1068,7 @@ static int bch2_snapshot_node_delete(struct btree_trans *trans, u32 id)
                        goto err;
 
                if (s.v->children[0]) {
-                       s_t->v.root_snapshot = cpu_to_le32(s.v->children[0]);
+                       s_t->v.root_snapshot = s.v->children[0];
                } else {
                        s_t->k.type = KEY_TYPE_deleted;
                        set_bkey_val_u64s(&s_t->k, 0);
@@ -840,10 +1088,12 @@ static int create_snapids(struct btree_trans *trans, u32 parent, u32 tree,
                          u32 *snapshot_subvols,
                          unsigned nr_snapids)
 {
+       struct bch_fs *c = trans->c;
        struct btree_iter iter;
        struct bkey_i_snapshot *n;
        struct bkey_s_c k;
-       unsigned i;
+       unsigned i, j;
+       u32 depth = bch2_snapshot_depth(c, parent);
        int ret;
 
        bch2_trans_iter_init(trans, &iter, BTREE_ID_snapshots,
@@ -873,6 +1123,12 @@ static int create_snapids(struct btree_trans *trans, u32 parent, u32 tree,
                n->v.parent     = cpu_to_le32(parent);
                n->v.subvol     = cpu_to_le32(snapshot_subvols[i]);
                n->v.tree       = cpu_to_le32(tree);
+               n->v.depth      = cpu_to_le32(depth);
+
+               for (j = 0; j < ARRAY_SIZE(n->v.skip); j++)
+                       n->v.skip[j] = cpu_to_le32(snapshot_skiplist_get(c, parent));
+
+               bubble_sort(n->v.skip, ARRAY_SIZE(n->v.skip), cmp_le32);
                SET_BCH_SNAPSHOT_SUBVOL(&n->v, true);
 
                ret = bch2_mark_snapshot(trans, BTREE_ID_snapshots, 0,
@@ -976,7 +1232,7 @@ static int snapshot_delete_key(struct btree_trans *trans,
                               struct bpos *last_pos)
 {
        struct bch_fs *c = trans->c;
-       u32 equiv = snapshot_t(c, k.k->p.snapshot)->equiv;
+       u32 equiv = bch2_snapshot_equiv(c, k.k->p.snapshot);
 
        if (!bkey_eq(k.k->p, *last_pos))
                equiv_seen->nr = 0;
@@ -1029,9 +1285,6 @@ int bch2_delete_dead_snapshots(struct bch_fs *c)
        u32 i, id;
        int ret = 0;
 
-       if (!test_bit(BCH_FS_HAVE_DELETED_SNAPSHOTS, &c->flags))
-               return 0;
-
        if (!test_bit(BCH_FS_STARTED, &c->flags)) {
                ret = bch2_fs_read_write_early(c);
                if (ret) {
@@ -1117,6 +1370,8 @@ int bch2_delete_dead_snapshots(struct bch_fs *c)
 err:
        darray_exit(&deleted);
        bch2_trans_exit(&trans);
+       if (ret)
+               bch_err_fn(c, ret);
        return ret;
 }
 
@@ -1124,7 +1379,8 @@ static void bch2_delete_dead_snapshots_work(struct work_struct *work)
 {
        struct bch_fs *c = container_of(work, struct bch_fs, snapshot_delete_work);
 
-       bch2_delete_dead_snapshots(c);
+       if (test_bit(BCH_FS_HAVE_DELETED_SNAPSHOTS, &c->flags))
+               bch2_delete_dead_snapshots(c);
        bch2_write_ref_put(c, BCH_WRITE_REF_delete_dead_snapshots);
 }
 
@@ -1142,7 +1398,7 @@ static int bch2_delete_dead_snapshots_hook(struct btree_trans *trans,
 
        set_bit(BCH_FS_HAVE_DELETED_SNAPSHOTS, &c->flags);
 
-       if (!test_bit(BCH_FS_FSCK_DONE, &c->flags))
+       if (c->curr_recovery_pass <= BCH_RECOVERY_PASS_delete_dead_snapshots)
                return 0;
 
        bch2_delete_dead_snapshots_async(c);
@@ -1217,7 +1473,7 @@ int bch2_subvolume_get_snapshot(struct btree_trans *trans, u32 subvol,
        k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_subvolumes, POS(0, subvol),
                               BTREE_ITER_CACHED|
                               BTREE_ITER_WITH_UPDATES);
-       ret = bkey_err(k) ?: k.k->type == KEY_TYPE_subvolume ? 0 : -ENOENT;
+       ret = bkey_err(k) ?: k.k->type == KEY_TYPE_subvolume ? 0 : -BCH_ERR_ENOENT_subvolume;
 
        if (likely(!ret))
                *snapid = le32_to_cpu(bkey_s_c_to_subvolume(k).v->snapshot);
@@ -1242,7 +1498,7 @@ static int bch2_subvolume_reparent(struct btree_trans *trans,
            le32_to_cpu(bkey_s_c_to_subvolume(k).v->parent) != old_parent)
                return 0;
 
-       s = bch2_bkey_make_mut_typed(trans, iter, k, 0, subvolume);
+       s = bch2_bkey_make_mut_typed(trans, iter, &k, 0, subvolume);
        ret = PTR_ERR_OR_ZERO(s);
        if (ret)
                return ret;
@@ -1321,7 +1577,7 @@ static int bch2_subvolume_delete(struct btree_trans *trans, u32 subvolid)
                          __bch2_subvolume_delete(trans, subvolid));
 }
 
-void bch2_subvolume_wait_for_pagecache_and_delete(struct work_struct *work)
+static void bch2_subvolume_wait_for_pagecache_and_delete(struct work_struct *work)
 {
        struct bch_fs *c = container_of(work, struct bch_fs,
                                snapshot_wait_for_pagecache_and_delete_work);
@@ -1359,7 +1615,7 @@ struct subvolume_unlink_hook {
        u32                             subvol;
 };
 
-int bch2_subvolume_wait_for_pagecache_and_delete_hook(struct btree_trans *trans,
+static int bch2_subvolume_wait_for_pagecache_and_delete_hook(struct btree_trans *trans,
                                                      struct btree_trans_commit_hook *_h)
 {
        struct subvolume_unlink_hook *h = container_of(_h, struct subvolume_unlink_hook, h);
@@ -1444,7 +1700,7 @@ int bch2_subvolume_create(struct btree_trans *trans, u64 inode,
                                BTREE_ITER_CACHED, subvolume);
                ret = PTR_ERR_OR_ZERO(src_subvol);
                if (unlikely(ret)) {
-                       bch2_fs_inconsistent_on(ret == -ENOENT, c,
+                       bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOENT), c,
                                                "subvolume %u not found", src_subvolid);
                        goto err;
                }