bcachefs: Split out snapshot.c

subvolume.c has gotten a bit large, this splits out a separate file just
for managing snapshot trees - BTREE_ID_snapshots.

Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

18 files changed, 1198 insertions, 1120 deletions

diff --git a/fs/bcachefs/Makefile b/fs/bcachefs/Makefile
index 19cc4ca66356..c87be5fb7caf 100644
--- a/fs/bcachefs/Makefile
+++ b/fs/bcachefs/Makefile
@@ -68,6 +68,7 @@ bcachefs-y		:=	\
 	sb-members.o		\
 	siphash.o		\
 	six.o			\
+	snapshot.o		\
 	subvolume.o		\
 	super.o			\
 	super-io.o		\
diff --git a/fs/bcachefs/bcachefs_format.h b/fs/bcachefs/bcachefs_format.h
index 23bae622309c..20e96daf9ca1 100644
--- a/fs/bcachefs/bcachefs_format.h
+++ b/fs/bcachefs/bcachefs_format.h
@@ -1124,6 +1124,11 @@ struct bch_subvolume {
 	__le32			flags;
 	__le32			snapshot;
 	__le64			inode;
+	/*
+	 * Snapshot subvolumes form a tree, separate from the snapshot nodes
+	 * tree - if this subvolume is a snapshot, this is the ID of the
+	 * subvolume it was created from:
+	 */
 	__le32			parent;
 	__le32			pad;
 	bch_le128		otime;
diff --git a/fs/bcachefs/bkey_methods.c b/fs/bcachefs/bkey_methods.c
index 90557f4c156d..6547142db428 100644
--- a/fs/bcachefs/bkey_methods.c
+++ b/fs/bcachefs/bkey_methods.c
@@ -13,6 +13,7 @@
 #include "lru.h"
 #include "quota.h"
 #include "reflink.h"
+#include "snapshot.h"
 #include "subvolume.h"
 #include "xattr.h"
 
diff --git a/fs/bcachefs/btree_iter.c b/fs/bcachefs/btree_iter.c
index dee9d2722ace..5216d3391079 100644
--- a/fs/bcachefs/btree_iter.c
+++ b/fs/bcachefs/btree_iter.c
@@ -14,7 +14,7 @@
 #include "extents.h"
 #include "journal.h"
 #include "replicas.h"
-#include "subvolume.h"
+#include "snapshot.h"
 #include "trace.h"
 
 #include <linux/random.h>
diff --git a/fs/bcachefs/btree_trans_commit.c b/fs/bcachefs/btree_trans_commit.c
index 78a09aa050c7..83cc7f64c57c 100644
--- a/fs/bcachefs/btree_trans_commit.c
+++ b/fs/bcachefs/btree_trans_commit.c
@@ -14,7 +14,7 @@
 #include "journal.h"
 #include "journal_reclaim.h"
 #include "replicas.h"
-#include "subvolume.h"
+#include "snapshot.h"
 
 #include <linux/prefetch.h>
 
diff --git a/fs/bcachefs/btree_update.c b/fs/bcachefs/btree_update.c
index 612fba60be14..a7fa20727d4b 100644
--- a/fs/bcachefs/btree_update.c
+++ b/fs/bcachefs/btree_update.c
@@ -11,7 +11,7 @@
 #include "error.h"
 #include "extents.h"
 #include "keylist.h"
-#include "subvolume.h"
+#include "snapshot.h"
 #include "trace.h"
 
 static inline int btree_insert_entry_cmp(const struct btree_insert_entry *l,
diff --git a/fs/bcachefs/fs.c b/fs/bcachefs/fs.c
index d287db908e62..8958957b648b 100644
--- a/fs/bcachefs/fs.c
+++ b/fs/bcachefs/fs.c
@@ -23,6 +23,7 @@
 #include "journal.h"
 #include "keylist.h"
 #include "quota.h"
+#include "snapshot.h"
 #include "super.h"
 #include "xattr.h"
 
diff --git a/fs/bcachefs/fsck.c b/fs/bcachefs/fsck.c
index c3020f9e2e03..960280b706e3 100644
--- a/fs/bcachefs/fsck.c
+++ b/fs/bcachefs/fsck.c
@@ -12,7 +12,7 @@
 #include "inode.h"
 #include "keylist.h"
 #include "recovery.h"
-#include "subvolume.h"
+#include "snapshot.h"
 #include "super.h"
 #include "xattr.h"
 
diff --git a/fs/bcachefs/inode.c b/fs/bcachefs/inode.c
index 46c4012facb0..8114b6e4f202 100644
--- a/fs/bcachefs/inode.c
+++ b/fs/bcachefs/inode.c
@@ -11,6 +11,7 @@
 #include "extent_update.h"
 #include "inode.h"
 #include "str_hash.h"
+#include "snapshot.h"
 #include "subvolume.h"
 #include "varint.h"
 
diff --git a/fs/bcachefs/quota.c b/fs/bcachefs/quota.c
index 4f0654ff816f..ca99772aedc6 100644
--- a/fs/bcachefs/quota.c
+++ b/fs/bcachefs/quota.c
@@ -5,7 +5,7 @@
 #include "error.h"
 #include "inode.h"
 #include "quota.h"
-#include "subvolume.h"
+#include "snapshot.h"
 #include "super-io.h"
 
 static const char * const bch2_quota_types[] = {
diff --git a/fs/bcachefs/recovery.c b/fs/bcachefs/recovery.c
index 33a68a335be6..30efb3c90560 100644
--- a/fs/bcachefs/recovery.c
+++ b/fs/bcachefs/recovery.c
@@ -25,6 +25,7 @@
 #include "recovery.h"
 #include "replicas.h"
 #include "sb-clean.h"
+#include "snapshot.h"
 #include "subvolume.h"
 #include "super-io.h"
 
diff --git a/fs/bcachefs/snapshot.c b/fs/bcachefs/snapshot.c
new file mode 100644
index 000000000000..824259178dab
--- /dev/null
+++ b/fs/bcachefs/snapshot.c
@@ -0,0 +1,886 @@
+// SPDX-License-Identifier: GPL-2.0
+
+#include "bcachefs.h"
+#include "btree_key_cache.h"
+#include "btree_update.h"
+#include "errcode.h"
+#include "error.h"
+#include "fs.h"
+#include "snapshot.h"
+
+#include <linux/random.h>
+
+/*
+ * Snapshot trees:
+ *
+ * Keys in BTREE_ID_snapshot_trees identify a whole tree of snapshot nodes; they
+ * exist to provide a stable identifier for the whole lifetime of a snapshot
+ * tree.
+ */
+
+void bch2_snapshot_tree_to_text(struct printbuf *out, struct bch_fs *c,
+				struct bkey_s_c k)
+{
+	struct bkey_s_c_snapshot_tree t = bkey_s_c_to_snapshot_tree(k);
+
+	prt_printf(out, "subvol %u root snapshot %u",
+		   le32_to_cpu(t.v->master_subvol),
+		   le32_to_cpu(t.v->root_snapshot));
+}
+
+int bch2_snapshot_tree_invalid(const struct bch_fs *c, struct bkey_s_c k,
+			       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))) {
+		prt_printf(err, "bad pos");
+		return -BCH_ERR_invalid_bkey;
+	}
+
+	return 0;
+}
+
+int bch2_snapshot_tree_lookup(struct btree_trans *trans, u32 id,
+			      struct bch_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;
+}
+
+struct bkey_i_snapshot_tree *
+__bch2_snapshot_tree_create(struct btree_trans *trans)
+{
+	struct btree_iter iter;
+	int ret = bch2_bkey_get_empty_slot(trans, &iter,
+			BTREE_ID_snapshot_trees, POS(0, U32_MAX));
+	struct bkey_i_snapshot_tree *s_t;
+
+	if (ret == -BCH_ERR_ENOSPC_btree_slot)
+		ret = -BCH_ERR_ENOSPC_snapshot_tree;
+	if (ret)
+		return ERR_PTR(ret);
+
+	s_t = bch2_bkey_alloc(trans, &iter, 0, snapshot_tree);
+	ret = PTR_ERR_OR_ZERO(s_t);
+	bch2_trans_iter_exit(trans, &iter);
+	return ret ? ERR_PTR(ret) : s_t;
+}
+
+static int bch2_snapshot_tree_create(struct btree_trans *trans,
+				u32 root_id, u32 subvol_id, u32 *tree_id)
+{
+	struct bkey_i_snapshot_tree *n_tree =
+		__bch2_snapshot_tree_create(trans);
+
+	if (IS_ERR(n_tree))
+		return PTR_ERR(n_tree);
+
+	n_tree->v.master_subvol	= cpu_to_le32(subvol_id);
+	n_tree->v.root_snapshot	= cpu_to_le32(root_id);
+	*tree_id = n_tree->k.p.offset;
+	return 0;
+}
+
+/* Snapshot nodes: */
+
+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 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);
+}
+
+void bch2_snapshot_to_text(struct printbuf *out, struct bch_fs *c,
+			   struct bkey_s_c k)
+{
+	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 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->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,
+			  enum bkey_invalid_flags flags,
+			  struct printbuf *err)
+{
+	struct bkey_s_c_snapshot s;
+	u32 i, id;
+
+	if (bkey_gt(k.k->p, POS(0, U32_MAX)) ||
+	    bkey_lt(k.k->p, POS(0, 1))) {
+		prt_printf(err, "bad pos");
+		return -BCH_ERR_invalid_bkey;
+	}
+
+	s = bkey_s_c_to_snapshot(k);
+
+	id = le32_to_cpu(s.v->parent);
+	if (id && id <= k.k->p.offset) {
+		prt_printf(err, "bad parent node (%u <= %llu)",
+		       id, k.k->p.offset);
+		return -BCH_ERR_invalid_bkey;
+	}
+
+	if (le32_to_cpu(s.v->children[0]) < le32_to_cpu(s.v->children[1])) {
+		prt_printf(err, "children not normalized");
+		return -BCH_ERR_invalid_bkey;
+	}
+
+	if (s.v->children[0] &&
+	    s.v->children[0] == s.v->children[1]) {
+		prt_printf(err, "duplicate child nodes");
+		return -BCH_ERR_invalid_bkey;
+	}
+
+	for (i = 0; i < 2; i++) {
+		id = le32_to_cpu(s.v->children[i]);
+
+		if (id >= k.k->p.offset) {
+			prt_printf(err, "bad child node (%u >= %llu)",
+			       id, k.k->p.offset);
+			return -BCH_ERR_invalid_bkey;
+		}
+	}
+
+	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;
+}
+
+int bch2_mark_snapshot(struct btree_trans *trans,
+		       enum btree_id btree, unsigned level,
+		       struct bkey_s_c old, struct bkey_s_c new,
+		       unsigned flags)
+{
+	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 = 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 {
+		memset(t, 0, sizeof(*t));
+	}
+err:
+	mutex_unlock(&c->snapshot_table_lock);
+	return ret;
+}
+
+int bch2_snapshot_lookup(struct btree_trans *trans, u32 id,
+			 struct bch_snapshot *s)
+{
+	return bch2_bkey_get_val_typed(trans, BTREE_ID_snapshots, POS(0, id),
+				       BTREE_ITER_WITH_UPDATES, snapshot, s);
+}
+
+int bch2_snapshot_live(struct btree_trans *trans, u32 id)
+{
+	struct bch_snapshot v;
+	int ret;
+
+	if (!id)
+		return 0;
+
+	ret = bch2_snapshot_lookup(trans, id, &v);
+	if (bch2_err_matches(ret, ENOENT))
+		bch_err(trans->c, "snapshot node %u not found", id);
+	if (ret)
+		return ret;
+
+	return !BCH_SNAPSHOT_DELETED(&v);
+}
+
+/*
+ * If @k is a snapshot with just one live child, it's part of a linear chain,
+ * which we consider to be an equivalence class: and then after snapshot
+ * deletion cleanup, there should only be a single key at a given position in
+ * this equivalence class.
+ *
+ * This sets the equivalence class of @k to be the child's equivalence class, if
+ * it's part of such a linear chain: this correctly sets equivalence classes on
+ * startup if we run leaf to root (i.e. in natural key order).
+ */
+int bch2_snapshot_set_equiv(struct btree_trans *trans, struct bkey_s_c k)
+{
+	struct bch_fs *c = trans->c;
+	unsigned i, nr_live = 0, live_idx = 0;
+	struct bkey_s_c_snapshot snap;
+	u32 id = k.k->p.offset, child[2];
+
+	if (k.k->type != KEY_TYPE_snapshot)
+		return 0;
+
+	snap = bkey_s_c_to_snapshot(k);
+
+	child[0] = le32_to_cpu(snap.v->children[0]);
+	child[1] = le32_to_cpu(snap.v->children[1]);
+
+	for (i = 0; i < 2; i++) {
+		int ret = bch2_snapshot_live(trans, child[i]);
+
+		if (ret < 0)
+			return ret;
+
+		if (ret)
+			live_idx = i;
+		nr_live += ret;
+	}
+
+	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;
+}
+
+/* fsck: */
+
+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;
+	bool found = false;
+	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;
+			found = true;
+			break;
+		}
+	}
+
+	bch2_trans_iter_exit(trans, &iter);
+
+	if (!ret && !found) {
+		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 = bch2_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, c,
+			"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_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)) ||
+	    fsck_err_on(BCH_SUBVOLUME_SNAP(&subvol), c,
+			"snapshot tree points to snapshot 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_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 (bch2_err_matches(ret, ENOENT))
+		return 0;
+	if (ret)
+		return ret;
+
+	return bch2_snapshot_is_ancestor_early(trans->c, snap_id, le32_to_cpu(s_t.root_snapshot));
+}
+
+u32 bch2_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 = bch2_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;
+}
+
+/*
+ * 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 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, k.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 && !bch2_err_matches(ret, ENOENT))
+		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);
+		ret =   PTR_ERR_OR_ZERO(u) ?:
+			bch2_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 (k.k->p.offset == root_id)
+			*s = u->v;
+	}
+
+	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 = u->v;
+	}
+err:
+	bch2_trans_iter_exit(trans, &root_iter);
+	return ret;
+}
+
+static int check_snapshot(struct btree_trans *trans,
+			  struct btree_iter *iter,
+			  struct bkey_s_c k)
+{
+	struct bch_fs *c = trans->c;
+	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;
+	int ret = 0;
+
+	if (k.k->type != KEY_TYPE_snapshot)
+		return 0;
+
+	memset(&s, 0, sizeof(s));
+	memcpy(&s, k.v, bkey_val_bytes(k.k));
+
+	id = le32_to_cpu(s.parent);
+	if (id) {
+		ret = bch2_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, k), buf.buf));
+		if (ret)
+			goto err;
+
+		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, k.k->p.offset);
+			ret = -EINVAL;
+			goto err;
+		}
+	}
+
+	for (i = 0; i < 2 && s.children[i]; i++) {
+		id = le32_to_cpu(s.children[i]);
+
+		ret = bch2_snapshot_lookup(trans, id, &v);
+		if (bch2_err_matches(ret, ENOENT))
+			bch_err(c, "snapshot node %llu has nonexistent child %u",
+				k.k->p.offset, id);
+		if (ret)
+			goto err;
+
+		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), k.k->p.offset);
+			ret = -EINVAL;
+			goto err;
+		}
+	}
+
+	should_have_subvol = BCH_SNAPSHOT_SUBVOL(&s) &&
+		!BCH_SNAPSHOT_DELETED(&s);
+
+	if (should_have_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, k), buf.buf));
+		if (ret)
+			goto err;
+
+		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",
+				k.k->p.offset);
+			ret = -EINVAL;
+			goto err;
+		}
+	} else {
+		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;
+
+			u->v.subvol = 0;
+			s = u->v;
+		}
+	}
+
+	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, k), buf.buf))) {
+		ret = snapshot_tree_ptr_repair(trans, iter, k, &s);
+		if (ret)
+			goto err;
+	}
+	ret = 0;
+
+	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(bch2_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_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_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_fn(c, ret);
+	return ret;
+}
+
+int bch2_snapshots_read(struct bch_fs *c)
+{
+	struct btree_iter iter;
+	struct bkey_s_c k;
+	int ret = 0;
+
+	ret = bch2_trans_run(c,
+		for_each_btree_key2(&trans, iter, BTREE_ID_snapshots,
+			   POS_MIN, 0, k,
+			bch2_mark_snapshot(&trans, BTREE_ID_snapshots, 0, bkey_s_c_null, k, 0) ?:
+			bch2_snapshot_set_equiv(&trans, k)));
+	if (ret)
+		bch_err_fn(c, ret);
+	return ret;
+}
+
+void bch2_fs_snapshots_exit(struct bch_fs *c)
+{
+	kfree(rcu_dereference_protected(c->snapshots, true));
+}
diff --git a/fs/bcachefs/snapshot.h b/fs/bcachefs/snapshot.h
new file mode 100644
index 000000000000..c59b7e9e64fe
--- /dev/null
+++ b/fs/bcachefs/snapshot.h
@@ -0,0 +1,250 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+#ifndef _BCACHEFS_SNAPSHOT_H
+#define _BCACHEFS_SNAPSHOT_H
+
+enum bkey_invalid_flags;
+
+void bch2_snapshot_tree_to_text(struct printbuf *, struct bch_fs *, struct bkey_s_c);
+int bch2_snapshot_tree_invalid(const struct bch_fs *, struct bkey_s_c,
+			       enum bkey_invalid_flags, struct printbuf *);
+
+#define bch2_bkey_ops_snapshot_tree ((struct bkey_ops) {	\
+	.key_invalid	= bch2_snapshot_tree_invalid,		\
+	.val_to_text	= bch2_snapshot_tree_to_text,		\
+	.min_val_size	= 8,					\
+})
+
+struct bkey_i_snapshot_tree *__bch2_snapshot_tree_create(struct btree_trans *);
+
+int bch2_snapshot_tree_lookup(struct btree_trans *, u32, struct bch_snapshot_tree *);
+
+void bch2_snapshot_to_text(struct printbuf *, struct bch_fs *, struct bkey_s_c);
+int bch2_snapshot_invalid(const struct bch_fs *, struct bkey_s_c,
+			  enum bkey_invalid_flags, struct printbuf *);
+int bch2_mark_snapshot(struct btree_trans *, enum btree_id, unsigned,
+		       struct bkey_s_c, struct bkey_s_c, unsigned);
+
+#define bch2_bkey_ops_snapshot ((struct bkey_ops) {		\
+	.key_invalid	= bch2_snapshot_invalid,		\
+	.val_to_text	= bch2_snapshot_to_text,		\
+	.atomic_trigger	= bch2_mark_snapshot,			\
+	.min_val_size	= 24,					\
+})
+
+static inline struct snapshot_t *__snapshot_t(struct snapshot_table *t, u32 id)
+{
+	return &t->s[U32_MAX - id];
+}
+
+static inline const struct snapshot_t *snapshot_t(struct bch_fs *c, u32 id)
+{
+	return __snapshot_t(rcu_dereference(c->snapshots), id);
+}
+
+static inline u32 bch2_snapshot_tree(struct bch_fs *c, u32 id)
+{
+	rcu_read_lock();
+	id = snapshot_t(c, id)->tree;
+	rcu_read_unlock();
+
+	return id;
+}
+
+static inline u32 __bch2_snapshot_parent_early(struct bch_fs *c, u32 id)
+{
+	return snapshot_t(c, id)->parent;
+}
+
+static inline u32 bch2_snapshot_parent_early(struct bch_fs *c, u32 id)
+{
+	rcu_read_lock();
+	id = __bch2_snapshot_parent_early(c, id);
+	rcu_read_unlock();
+
+	return id;
+}
+
+static inline u32 __bch2_snapshot_parent(struct bch_fs *c, u32 id)
+{
+#ifdef CONFIG_BCACHEFS_DEBUG
+	u32 parent = snapshot_t(c, id)->parent;
+
+	if (parent &&
+	    snapshot_t(c, id)->depth != snapshot_t(c, parent)->depth + 1)
+		panic("id %u depth=%u parent %u depth=%u\n",
+		      id, snapshot_t(c, id)->depth,
+		      parent, snapshot_t(c, parent)->depth);
+
+	return parent;
+#else
+	return snapshot_t(c, id)->parent;
+#endif
+}
+
+static inline u32 bch2_snapshot_parent(struct bch_fs *c, u32 id)
+{
+	rcu_read_lock();
+	id = __bch2_snapshot_parent(c, id);
+	rcu_read_unlock();
+
+	return id;
+}
+
+static inline u32 bch2_snapshot_nth_parent(struct bch_fs *c, u32 id, u32 n)
+{
+	rcu_read_lock();
+	while (n--)
+		id = __bch2_snapshot_parent(c, id);
+	rcu_read_unlock();
+
+	return id;
+}
+
+u32 bch2_snapshot_skiplist_get(struct bch_fs *, u32);
+
+static inline u32 bch2_snapshot_root(struct bch_fs *c, u32 id)
+{
+	u32 parent;
+
+	rcu_read_lock();
+	while ((parent = __bch2_snapshot_parent(c, id)))
+		id = parent;
+	rcu_read_unlock();
+
+	return id;
+}
+
+static inline u32 __bch2_snapshot_equiv(struct bch_fs *c, u32 id)
+{
+	return snapshot_t(c, id)->equiv;
+}
+
+static inline u32 bch2_snapshot_equiv(struct bch_fs *c, u32 id)
+{
+	rcu_read_lock();
+	id = __bch2_snapshot_equiv(c, id);
+	rcu_read_unlock();
+
+	return id;
+}
+
+static inline bool bch2_snapshot_is_equiv(struct bch_fs *c, u32 id)
+{
+	return id == bch2_snapshot_equiv(c, id);
+}
+
+static inline bool bch2_snapshot_is_internal_node(struct bch_fs *c, u32 id)
+{
+	const struct snapshot_t *s;
+	bool ret;
+
+	rcu_read_lock();
+	s = snapshot_t(c, id);
+	ret = s->children[0];
+	rcu_read_unlock();
+
+	return ret;
+}
+
+static inline u32 bch2_snapshot_is_leaf(struct bch_fs *c, u32 id)
+{
+	return !bch2_snapshot_is_internal_node(c, id);
+}
+
+static inline u32 bch2_snapshot_sibling(struct bch_fs *c, u32 id)
+{
+	const struct snapshot_t *s;
+	u32 parent = __bch2_snapshot_parent(c, id);
+
+	if (!parent)
+		return 0;
+
+	s = snapshot_t(c, __bch2_snapshot_parent(c, id));
+	if (id == s->children[0])
+		return s->children[1];
+	if (id == s->children[1])
+		return s->children[0];
+	return 0;
+}
+
+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;
+}
+
+bool __bch2_snapshot_is_ancestor(struct bch_fs *, u32, u32);
+
+static inline bool bch2_snapshot_is_ancestor(struct bch_fs *c, u32 id, u32 ancestor)
+{
+	return id == ancestor
+		? true
+		: __bch2_snapshot_is_ancestor(c, id, ancestor);
+}
+
+static inline bool bch2_snapshot_has_children(struct bch_fs *c, u32 id)
+{
+	const struct snapshot_t *t;
+	bool ret;
+
+	rcu_read_lock();
+	t = snapshot_t(c, id);
+	ret = (t->children[0]|t->children[1]) != 0;
+	rcu_read_unlock();
+
+	return ret;
+}
+
+static inline bool snapshot_list_has_id(snapshot_id_list *s, u32 id)
+{
+	u32 *i;
+
+	darray_for_each(*s, i)
+		if (*i == id)
+			return true;
+	return false;
+}
+
+static inline bool snapshot_list_has_ancestor(struct bch_fs *c, snapshot_id_list *s, u32 id)
+{
+	u32 *i;
+
+	darray_for_each(*s, i)
+		if (bch2_snapshot_is_ancestor(c, id, *i))
+			return true;
+	return false;
+}
+
+static inline int snapshot_list_add(struct bch_fs *c, snapshot_id_list *s, u32 id)
+{
+	int ret;
+
+	BUG_ON(snapshot_list_has_id(s, id));
+	ret = darray_push(s, id);
+	if (ret)
+		bch_err(c, "error reallocating snapshot_id_list (size %zu)", s->size);
+	return ret;
+}
+
+int bch2_snapshot_lookup(struct btree_trans *trans, u32 id,
+			 struct bch_snapshot *s);
+int bch2_snapshot_get_subvol(struct btree_trans *, u32,
+			     struct bch_subvolume *);
+int bch2_snapshot_live(struct btree_trans *trans, u32 id);
+int bch2_snapshot_set_equiv(struct btree_trans *trans, struct bkey_s_c k);
+
+/* only exported for tests: */
+int bch2_snapshot_node_create(struct btree_trans *, u32,
+			      u32 *, u32 *, unsigned);
+
+int bch2_check_snapshot_trees(struct bch_fs *);
+int bch2_check_snapshots(struct bch_fs *);
+int bch2_snapshots_read(struct bch_fs *);
+void bch2_fs_snapshots_exit(struct bch_fs *);
+
+#endif /* _BCACHEFS_SNAPSHOT_H */
diff --git a/fs/bcachefs/subvolume.c b/fs/bcachefs/subvolume.c
index 736afb626a89..571cb2809c88 100644
--- a/fs/bcachefs/subvolume.c
+++ b/fs/bcachefs/subvolume.c
@@ -6,866 +6,13 @@
 #include "errcode.h"
 #include "error.h"
 #include "fs.h"
+#include "snapshot.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,
-				struct bkey_s_c k)
-{
-	struct bkey_s_c_snapshot_tree t = bkey_s_c_to_snapshot_tree(k);
-
-	prt_printf(out, "subvol %u root snapshot %u",
-		   le32_to_cpu(t.v->master_subvol),
-		   le32_to_cpu(t.v->root_snapshot));
-}
-
-int bch2_snapshot_tree_invalid(const struct bch_fs *c, struct bkey_s_c k,
-			       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))) {
-		prt_printf(err, "bad pos");
-		return -BCH_ERR_invalid_bkey;
-	}
-
-	return 0;
-}
-
-int bch2_snapshot_tree_lookup(struct btree_trans *trans, u32 id,
-			      struct bch_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 *
-__snapshot_tree_create(struct btree_trans *trans)
-{
-	struct btree_iter iter;
-	int ret = bch2_bkey_get_empty_slot(trans, &iter,
-			BTREE_ID_snapshot_trees, POS(0, U32_MAX));
-	struct bkey_i_snapshot_tree *s_t;
-
-	if (ret == -BCH_ERR_ENOSPC_btree_slot)
-		ret = -BCH_ERR_ENOSPC_snapshot_tree;
-	if (ret)
-		return ERR_PTR(ret);
-
-	s_t = bch2_bkey_alloc(trans, &iter, 0, snapshot_tree);
-	ret = PTR_ERR_OR_ZERO(s_t);
-	bch2_trans_iter_exit(trans, &iter);
-	return ret ? ERR_PTR(ret) : s_t;
-}
-
-static int snapshot_tree_create(struct btree_trans *trans,
-				u32 root_id, u32 subvol_id, u32 *tree_id)
-{
-	struct bkey_i_snapshot_tree *n_tree =
-		__snapshot_tree_create(trans);
-
-	if (IS_ERR(n_tree))
-		return PTR_ERR(n_tree);
-
-	n_tree->v.master_subvol	= cpu_to_le32(subvol_id);
-	n_tree->v.root_snapshot	= cpu_to_le32(root_id);
-	*tree_id = n_tree->k.p.offset;
-	return 0;
-}
-
-/* Snapshot nodes: */
-
-void bch2_snapshot_to_text(struct printbuf *out, struct bch_fs *c,
-			   struct bkey_s_c k)
-{
-	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 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->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,
-			  enum bkey_invalid_flags flags,
-			  struct printbuf *err)
-{
-	struct bkey_s_c_snapshot s;
-	u32 i, id;
-
-	if (bkey_gt(k.k->p, POS(0, U32_MAX)) ||
-	    bkey_lt(k.k->p, POS(0, 1))) {
-		prt_printf(err, "bad pos");
-		return -BCH_ERR_invalid_bkey;
-	}
-
-	s = bkey_s_c_to_snapshot(k);
-
-	id = le32_to_cpu(s.v->parent);
-	if (id && id <= k.k->p.offset) {
-		prt_printf(err, "bad parent node (%u <= %llu)",
-		       id, k.k->p.offset);
-		return -BCH_ERR_invalid_bkey;
-	}
-
-	if (le32_to_cpu(s.v->children[0]) < le32_to_cpu(s.v->children[1])) {
-		prt_printf(err, "children not normalized");
-		return -BCH_ERR_invalid_bkey;
-	}
-
-	if (s.v->children[0] &&
-	    s.v->children[0] == s.v->children[1]) {
-		prt_printf(err, "duplicate child nodes");
-		return -BCH_ERR_invalid_bkey;
-	}
-
-	for (i = 0; i < 2; i++) {
-		id = le32_to_cpu(s.v->children[i]);
-
-		if (id >= k.k->p.offset) {
-			prt_printf(err, "bad child node (%u >= %llu)",
-			       id, k.k->p.offset);
-			return -BCH_ERR_invalid_bkey;
-		}
-	}
-
-	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;
-}
-
-int bch2_mark_snapshot(struct btree_trans *trans,
-		       enum btree_id btree, unsigned level,
-		       struct bkey_s_c old, struct bkey_s_c new,
-		       unsigned flags)
-{
-	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 = 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 {
-		memset(t, 0, sizeof(*t));
-	}
-err:
-	mutex_unlock(&c->snapshot_table_lock);
-	return ret;
-}
-
-static int snapshot_lookup(struct btree_trans *trans, u32 id,
-			   struct bch_snapshot *s)
-{
-	return bch2_bkey_get_val_typed(trans, BTREE_ID_snapshots, POS(0, id),
-				       BTREE_ITER_WITH_UPDATES, snapshot, s);
-}
-
-static int snapshot_live(struct btree_trans *trans, u32 id)
-{
-	struct bch_snapshot v;
-	int ret;
-
-	if (!id)
-		return 0;
-
-	ret = snapshot_lookup(trans, id, &v);
-	if (bch2_err_matches(ret, ENOENT))
-		bch_err(trans->c, "snapshot node %u not found", id);
-	if (ret)
-		return ret;
-
-	return !BCH_SNAPSHOT_DELETED(&v);
-}
-
-static int bch2_snapshot_set_equiv(struct btree_trans *trans, struct bkey_s_c k)
-{
-	struct bch_fs *c = trans->c;
-	unsigned i, nr_live = 0, live_idx = 0;
-	struct bkey_s_c_snapshot snap;
-	u32 id = k.k->p.offset, child[2];
-
-	if (k.k->type != KEY_TYPE_snapshot)
-		return 0;
-
-	snap = bkey_s_c_to_snapshot(k);
-
-	child[0] = le32_to_cpu(snap.v->children[0]);
-	child[1] = le32_to_cpu(snap.v->children[1]);
-
-	for (i = 0; i < 2; i++) {
-		int ret = snapshot_live(trans, child[i]);
-
-		if (ret < 0)
-			return ret;
-
-		if (ret)
-			live_idx = i;
-		nr_live += ret;
-	}
-
-	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;
-}
-
-/* fsck: */
-
-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;
-	bool found = false;
-	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;
-			found = true;
-			break;
-		}
-	}
-
-	bch2_trans_iter_exit(trans, &iter);
-
-	if (!ret && !found) {
-		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, c,
-			"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_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)) ||
-	    fsck_err_on(BCH_SUBVOLUME_SNAP(&subvol), c,
-			"snapshot tree points to snapshot 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_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 (bch2_err_matches(ret, ENOENT))
-		return 0;
-	if (ret)
-		return ret;
-
-	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;
-}
-
-/*
- * 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 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, k.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 && !bch2_err_matches(ret, ENOENT))
-		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);
-		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 (k.k->p.offset == root_id)
-			*s = u->v;
-	}
-
-	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 = 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 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;
-	int ret = 0;
-
-	if (k.k->type != KEY_TYPE_snapshot)
-		return 0;
-
-	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, k), buf.buf));
-		if (ret)
-			goto err;
-
-		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, k.k->p.offset);
-			ret = -EINVAL;
-			goto err;
-		}
-	}
-
-	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",
-				k.k->p.offset, id);
-		if (ret)
-			goto err;
-
-		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), k.k->p.offset);
-			ret = -EINVAL;
-			goto err;
-		}
-	}
-
-	should_have_subvol = BCH_SNAPSHOT_SUBVOL(&s) &&
-		!BCH_SNAPSHOT_DELETED(&s);
-
-	if (should_have_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, k), buf.buf));
-		if (ret)
-			goto err;
-
-		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",
-				k.k->p.offset);
-			ret = -EINVAL;
-			goto err;
-		}
-	} else {
-		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;
-
-			u->v.subvol = 0;
-			s = u->v;
-		}
-	}
-
-	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, k), buf.buf))) {
-		ret = snapshot_tree_ptr_repair(trans, iter, k, &s);
-		if (ret)
-			goto err;
-	}
-	ret = 0;
-
-	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_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_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_fn(c, ret);
-	return ret;
-}
-
 static int check_subvol(struct btree_trans *trans,
 			struct btree_iter *iter,
 			struct bkey_s_c k)
@@ -881,7 +28,7 @@ static int check_subvol(struct btree_trans *trans,
 
 	subvol = bkey_s_c_to_subvolume(k);
 	snapid = le32_to_cpu(subvol.v->snapshot);
-	ret = snapshot_lookup(trans, snapid, &snapshot);
+	ret = bch2_snapshot_lookup(trans, snapid, &snapshot);
 
 	if (bch2_err_matches(ret, ENOENT))
 		bch_err(c, "subvolume %llu points to nonexistent snapshot %u",
@@ -949,27 +96,6 @@ int bch2_check_subvols(struct bch_fs *c)
 	return ret;
 }
 
-void bch2_fs_snapshots_exit(struct bch_fs *c)
-{
-	kfree(rcu_dereference_protected(c->snapshots, true));
-}
-
-int bch2_snapshots_read(struct bch_fs *c)
-{
-	struct btree_iter iter;
-	struct bkey_s_c k;
-	int ret = 0;
-
-	ret = bch2_trans_run(c,
-		for_each_btree_key2(&trans, iter, BTREE_ID_snapshots,
-			   POS_MIN, 0, k,
-			bch2_mark_snapshot(&trans, BTREE_ID_snapshots, 0, bkey_s_c_null, k, 0) ?:
-			bch2_snapshot_set_equiv(&trans, k)));
-	if (ret)
-		bch_err_fn(c, ret);
-	return ret;
-}
-
 /*
  * Mark a snapshot as deleted, for future cleanup:
  */
@@ -1126,7 +252,7 @@ static int create_snapids(struct btree_trans *trans, u32 parent, u32 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));
+			n->v.skip[j] = cpu_to_le32(bch2_snapshot_skiplist_get(c, parent));
 
 		bubble_sort(n->v.skip, ARRAY_SIZE(n->v.skip), cmp_le32);
 		SET_BCH_SNAPSHOT_SUBVOL(&n->v, true);
@@ -1196,7 +322,7 @@ static int bch2_snapshot_node_create_tree(struct btree_trans *trans,
 	struct bkey_i_snapshot_tree *n_tree;
 	int ret;
 
-	n_tree = __snapshot_tree_create(trans);
+	n_tree = __bch2_snapshot_tree_create(trans);
 	ret =   PTR_ERR_OR_ZERO(n_tree) ?:
 		create_snapids(trans, 0, n_tree->k.p.offset,
 			     new_snapids, snapshot_subvols, nr_snapids);
@@ -1224,6 +350,18 @@ int bch2_snapshot_node_create(struct btree_trans *trans, u32 parent,
 
 }
 
+/*
+ * If we have an unlinked inode in an internal snapshot node, and the inode
+ * really has been deleted in all child snapshots, how does this get cleaned up?
+ *
+ * first there is the problem of how keys that have been overwritten in all
+ * child snapshots get deleted (unimplemented?), but inodes may perhaps be
+ * special?
+ *
+ * also: unlinked inode in internal snapshot appears to not be getting deleted
+ * correctly if inode doesn't exist in leaf snapshots
+ */
+
 static int snapshot_delete_key(struct btree_trans *trans,
 			       struct btree_iter *iter,
 			       struct bkey_s_c k,
@@ -1247,6 +385,11 @@ static int snapshot_delete_key(struct btree_trans *trans,
 	}
 }
 
+/*
+ * For a given snapshot, if it doesn't have a subvolume that points to it, and
+ * it doesn't have child snapshot nodes - it's now redundant and we can mark it
+ * as deleted.
+ */
 static int bch2_delete_redundant_snapshot(struct btree_trans *trans, struct btree_iter *iter,
 					  struct bkey_s_c k)
 {
@@ -1265,8 +408,8 @@ static int bch2_delete_redundant_snapshot(struct btree_trans *trans, struct btre
 	children[0] = le32_to_cpu(snap.v->children[0]);
 	children[1] = le32_to_cpu(snap.v->children[1]);
 
-	ret   = snapshot_live(trans, children[0]) ?:
-		snapshot_live(trans, children[1]);
+	ret   = bch2_snapshot_live(trans, children[0]) ?:
+		bch2_snapshot_live(trans, children[1]);
 	if (ret < 0)
 		return ret;
 
@@ -1459,26 +602,27 @@ int bch2_snapshot_get_subvol(struct btree_trans *trans, u32 snapshot,
 {
 	struct bch_snapshot snap;
 
-	return  snapshot_lookup(trans, snapshot, &snap) ?:
+	return  bch2_snapshot_lookup(trans, snapshot, &snap) ?:
 		bch2_subvolume_get(trans, le32_to_cpu(snap.subvol), true, 0, subvol);
 }
 
-int bch2_subvolume_get_snapshot(struct btree_trans *trans, u32 subvol,
+int bch2_subvolume_get_snapshot(struct btree_trans *trans, u32 subvolid,
 				u32 *snapid)
 {
 	struct btree_iter iter;
-	struct bkey_s_c k;
+	struct bkey_s_c_subvolume subvol;
 	int ret;
 
-	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 : -BCH_ERR_ENOENT_subvolume;
+	subvol = bch2_bkey_get_iter_typed(trans, &iter,
+					  BTREE_ID_subvolumes, POS(0, subvolid),
+					  BTREE_ITER_CACHED|BTREE_ITER_WITH_UPDATES,
+					  subvolume);
+	ret = bkey_err(subvol);
+	bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOENT), trans->c,
+				"missing subvolume %u", subvolid);
 
 	if (likely(!ret))
-		*snapid = le32_to_cpu(bkey_s_c_to_subvolume(k).v->snapshot);
-	else if (bch2_err_matches(ret, ENOENT))
-		bch2_fs_inconsistent(trans->c, "missing subvolume %u", subvol);
+		*snapid = le32_to_cpu(subvol.v->snapshot);
 	bch2_trans_iter_exit(trans, &iter);
 	return ret;
 }
@@ -1508,7 +652,12 @@ static int bch2_subvolume_reparent(struct btree_trans *trans,
 }
 
 /*
- * Scan for subvolumes with parent @subvolid_to_delete, reparent:
+ * Separate from the snapshot tree in the snapshots btree, we record the tree
+ * structure of how snapshot subvolumes were created - the parent subvolume of
+ * each snapshot subvolume.
+ *
+ * When a subvolume is deleted, we scan for child subvolumes and reparant them,
+ * to avoid dangling references:
  */
 static int bch2_subvolumes_reparent(struct btree_trans *trans, u32 subvolid_to_delete)
 {
diff --git a/fs/bcachefs/subvolume.h b/fs/bcachefs/subvolume.h
index 6905e91a9470..8d4c50f4cd05 100644
--- a/fs/bcachefs/subvolume.h
+++ b/fs/bcachefs/subvolume.h
@@ -7,225 +7,8 @@
 
 enum bkey_invalid_flags;
 
-void bch2_snapshot_tree_to_text(struct printbuf *, struct bch_fs *, struct bkey_s_c);
-int bch2_snapshot_tree_invalid(const struct bch_fs *, struct bkey_s_c,
-			       enum bkey_invalid_flags, struct printbuf *);
-
-#define bch2_bkey_ops_snapshot_tree ((struct bkey_ops) {	\
-	.key_invalid	= bch2_snapshot_tree_invalid,		\
-	.val_to_text	= bch2_snapshot_tree_to_text,		\
-	.min_val_size	= 8,					\
-})
-
-int bch2_snapshot_tree_lookup(struct btree_trans *, u32, struct bch_snapshot_tree *);
-
-void bch2_snapshot_to_text(struct printbuf *, struct bch_fs *, struct bkey_s_c);
-int bch2_snapshot_invalid(const struct bch_fs *, struct bkey_s_c,
-			  enum bkey_invalid_flags, struct printbuf *);
-int bch2_mark_snapshot(struct btree_trans *, enum btree_id, unsigned,
-		       struct bkey_s_c, struct bkey_s_c, unsigned);
-
-#define bch2_bkey_ops_snapshot ((struct bkey_ops) {		\
-	.key_invalid	= bch2_snapshot_invalid,		\
-	.val_to_text	= bch2_snapshot_to_text,		\
-	.atomic_trigger	= bch2_mark_snapshot,			\
-	.min_val_size	= 24,					\
-})
-
-static inline struct snapshot_t *__snapshot_t(struct snapshot_table *t, u32 id)
-{
-	return &t->s[U32_MAX - id];
-}
-
-static inline const struct snapshot_t *snapshot_t(struct bch_fs *c, u32 id)
-{
-	return __snapshot_t(rcu_dereference(c->snapshots), id);
-}
-
-static inline u32 bch2_snapshot_tree(struct bch_fs *c, u32 id)
-{
-	rcu_read_lock();
-	id = snapshot_t(c, id)->tree;
-	rcu_read_unlock();
-
-	return id;
-}
-
-static inline u32 __bch2_snapshot_parent_early(struct bch_fs *c, u32 id)
-{
-	return snapshot_t(c, id)->parent;
-}
-
-static inline u32 bch2_snapshot_parent_early(struct bch_fs *c, u32 id)
-{
-	rcu_read_lock();
-	id = __bch2_snapshot_parent_early(c, id);
-	rcu_read_unlock();
-
-	return id;
-}
-
-static inline u32 __bch2_snapshot_parent(struct bch_fs *c, u32 id)
-{
-#ifdef CONFIG_BCACHEFS_DEBUG
-	u32 parent = snapshot_t(c, id)->parent;
-
-	if (parent &&
-	    snapshot_t(c, id)->depth != snapshot_t(c, parent)->depth + 1)
-		panic("id %u depth=%u parent %u depth=%u\n",
-		      id, snapshot_t(c, id)->depth,
-		      parent, snapshot_t(c, parent)->depth);
-
-	return parent;
-#else
-	return snapshot_t(c, id)->parent;
-#endif
-}
-
-static inline u32 bch2_snapshot_parent(struct bch_fs *c, u32 id)
-{
-	rcu_read_lock();
-	id = __bch2_snapshot_parent(c, id);
-	rcu_read_unlock();
-
-	return id;
-}
-
-static inline u32 bch2_snapshot_nth_parent(struct bch_fs *c, u32 id, u32 n)
-{
-	rcu_read_lock();
-	while (n--)
-		id = __bch2_snapshot_parent(c, id);
-	rcu_read_unlock();
-
-	return id;
-}
-
-static inline u32 bch2_snapshot_root(struct bch_fs *c, u32 id)
-{
-	u32 parent;
-
-	rcu_read_lock();
-	while ((parent = __bch2_snapshot_parent(c, id)))
-		id = parent;
-	rcu_read_unlock();
-
-	return id;
-}
-
-static inline u32 __bch2_snapshot_equiv(struct bch_fs *c, u32 id)
-{
-	return snapshot_t(c, id)->equiv;
-}
-
-static inline u32 bch2_snapshot_equiv(struct bch_fs *c, u32 id)
-{
-	rcu_read_lock();
-	id = __bch2_snapshot_equiv(c, id);
-	rcu_read_unlock();
-
-	return id;
-}
-
-static inline bool bch2_snapshot_is_equiv(struct bch_fs *c, u32 id)
-{
-	return id == bch2_snapshot_equiv(c, id);
-}
-
-static inline bool bch2_snapshot_is_internal_node(struct bch_fs *c, u32 id)
-{
-	const struct snapshot_t *s;
-	bool ret;
-
-	rcu_read_lock();
-	s = snapshot_t(c, id);
-	ret = s->children[0];
-	rcu_read_unlock();
-
-	return ret;
-}
-
-static inline u32 bch2_snapshot_is_leaf(struct bch_fs *c, u32 id)
-{
-	return !bch2_snapshot_is_internal_node(c, id);
-}
-
-static inline u32 bch2_snapshot_sibling(struct bch_fs *c, u32 id)
-{
-	const struct snapshot_t *s;
-	u32 parent = __bch2_snapshot_parent(c, id);
-
-	if (!parent)
-		return 0;
-
-	s = snapshot_t(c, __bch2_snapshot_parent(c, id));
-	if (id == s->children[0])
-		return s->children[1];
-	if (id == s->children[1])
-		return s->children[0];
-	return 0;
-}
-
-bool __bch2_snapshot_is_ancestor(struct bch_fs *, u32, u32);
-
-static inline bool bch2_snapshot_is_ancestor(struct bch_fs *c, u32 id, u32 ancestor)
-{
-	return id == ancestor
-		? true
-		: __bch2_snapshot_is_ancestor(c, id, ancestor);
-}
-
-static inline bool bch2_snapshot_has_children(struct bch_fs *c, u32 id)
-{
-	const struct snapshot_t *t;
-	bool ret;
-
-	rcu_read_lock();
-	t = snapshot_t(c, id);
-	ret = (t->children[0]|t->children[1]) != 0;
-	rcu_read_unlock();
-
-	return ret;
-}
-
-static inline bool snapshot_list_has_id(snapshot_id_list *s, u32 id)
-{
-	u32 *i;
-
-	darray_for_each(*s, i)
-		if (*i == id)
-			return true;
-	return false;
-}
-
-static inline bool snapshot_list_has_ancestor(struct bch_fs *c, snapshot_id_list *s, u32 id)
-{
-	u32 *i;
-
-	darray_for_each(*s, i)
-		if (bch2_snapshot_is_ancestor(c, id, *i))
-			return true;
-	return false;
-}
-
-static inline int snapshot_list_add(struct bch_fs *c, snapshot_id_list *s, u32 id)
-{
-	int ret;
-
-	BUG_ON(snapshot_list_has_id(s, id));
-	ret = darray_push(s, id);
-	if (ret)
-		bch_err(c, "error reallocating snapshot_id_list (size %zu)", s->size);
-	return ret;
-}
-
-int bch2_check_snapshot_trees(struct bch_fs *);
-int bch2_check_snapshots(struct bch_fs *);
 int bch2_check_subvols(struct bch_fs *);
 
-void bch2_fs_snapshots_exit(struct bch_fs *);
-int bch2_snapshots_read(struct bch_fs *);
-
 int bch2_subvolume_invalid(const struct bch_fs *, struct bkey_s_c,
 			   unsigned, struct printbuf *);
 void bch2_subvolume_to_text(struct printbuf *, struct bch_fs *, struct bkey_s_c);
@@ -238,14 +21,8 @@ void bch2_subvolume_to_text(struct printbuf *, struct bch_fs *, struct bkey_s_c)
 
 int bch2_subvolume_get(struct btree_trans *, unsigned,
 		       bool, int, struct bch_subvolume *);
-int bch2_snapshot_get_subvol(struct btree_trans *, u32,
-			     struct bch_subvolume *);
 int bch2_subvolume_get_snapshot(struct btree_trans *, u32, u32 *);
 
-/* only exported for tests: */
-int bch2_snapshot_node_create(struct btree_trans *, u32,
-			      u32 *, u32 *, unsigned);
-
 int bch2_delete_dead_snapshots(struct bch_fs *);
 void bch2_delete_dead_snapshots_async(struct bch_fs *);
 
diff --git a/fs/bcachefs/super.c b/fs/bcachefs/super.c
index 5c62fcf3afdb..48670c8b56aa 100644
--- a/fs/bcachefs/super.c
+++ b/fs/bcachefs/super.c
@@ -48,6 +48,7 @@
 #include "recovery.h"
 #include "replicas.h"
 #include "sb-clean.h"
+#include "snapshot.h"
 #include "subvolume.h"
 #include "super.h"
 #include "super-io.h"
diff --git a/fs/bcachefs/tests.c b/fs/bcachefs/tests.c
index 1d4b0a583586..72389c7376d6 100644
--- a/fs/bcachefs/tests.c
+++ b/fs/bcachefs/tests.c
@@ -4,7 +4,7 @@
 #include "bcachefs.h"
 #include "btree_update.h"
 #include "journal_reclaim.h"
-#include "subvolume.h"
+#include "snapshot.h"
 #include "tests.h"
 
 #include "linux/kthread.h"
diff --git a/fs/bcachefs/util.h b/fs/bcachefs/util.h
index 5b98ece3cf9b..d06671a09852 100644
--- a/fs/bcachefs/util.h
+++ b/fs/bcachefs/util.h
@@ -845,6 +845,11 @@ static inline int u8_cmp(u8 l, u8 r)
 	return cmp_int(l, r);
 }
 
+static inline int cmp_le32(__le32 l, __le32 r)
+{
+	return cmp_int(le32_to_cpu(l), le32_to_cpu(r));
+}
+
 #include <linux/uuid.h>
 
 #endif /* _BCACHEFS_UTIL_H */