bcachefs: Snapshot depth, skiplist fields

This extents KEY_TYPE_snapshot to include some new fields:
 - depth, to indicate depth of this particular node from the root
 - skip[3], skiplist entries for quickly walking back up to the root

These are to improve bch2_snapshot_is_ancestor(), making it O(ln(n))
instead of O(n) in the snapshot tree depth.

Skiplist nodes are picked at random from the set of ancestor nodes, not
some fixed fraction.

This introduces bcachefs_metadata_version 1.1, snapshot_skiplists.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

fs/bcachefs/bcachefs_format.h

@@ -1148,6 +1148,8 @@ struct bch_snapshot {
 	__le32			children[2];
 	__le32			subvol;
 	__le32			tree;
+	__le32			depth;
+	__le32			skip[3];
 };
 
 LE32_BITMASK(BCH_SNAPSHOT_DELETED,	struct bch_snapshot, flags,  0,  1)
@@ -1625,7 +1627,9 @@ struct bch_sb_field_journal_seq_blacklist {
 	x(snapshot_trees,		BCH_VERSION(0, 29),		\
 	  RECOVERY_PASS_ALL_FSCK)					\
 	x(major_minor,			BCH_VERSION(1,  0),		\
-	  0)
+	  0)								\
+	x(snapshot_skiplists,		BCH_VERSION(1,  1),		\
+	  BIT_ULL(BCH_RECOVERY_PASS_check_snapshots))
 
 enum bcachefs_metadata_version {
 	bcachefs_metadata_version_min = 9,

fs/bcachefs/btree_iter.h

@@ -795,6 +795,14 @@ __bch2_btree_iter_peek_upto_and_restart(struct btree_trans *trans,
 			    (_do) ?: bch2_trans_commit(_trans, (_disk_res),\
 					(_journal_seq), (_commit_flags)))
 
+#define for_each_btree_key_reverse_commit(_trans, _iter, _btree_id,	\
+				  _start, _iter_flags, _k,		\
+				  _disk_res, _journal_seq, _commit_flags,\
+				  _do)					\
+	for_each_btree_key_reverse(_trans, _iter, _btree_id, _start, _iter_flags, _k,\
+			    (_do) ?: bch2_trans_commit(_trans, (_disk_res),\
+					(_journal_seq), (_commit_flags)))
+
 #define for_each_btree_key_upto_commit(_trans, _iter, _btree_id,	\
 				  _start, _end, _iter_flags, _k,	\
 				  _disk_res, _journal_seq, _commit_flags,\

fs/bcachefs/recovery.c

@@ -594,10 +594,21 @@ static int bch2_journal_replay_key(struct btree_trans *trans,
 	unsigned iter_flags =
 		BTREE_ITER_INTENT|
 		BTREE_ITER_NOT_EXTENTS;
+	unsigned update_flags = BTREE_TRIGGER_NORUN;
 	int ret;
 
+	/*
+	 * BTREE_UPDATE_KEY_CACHE_RECLAIM disables key cache lookup/update to
+	 * keep the key cache coherent with the underlying btree. Nothing
+	 * besides the allocator is doing updates yet so we don't need key cache
+	 * coherency for non-alloc btrees, and key cache fills for snapshots
+	 * btrees use BTREE_ITER_FILTER_SNAPSHOTS, which isn't available until
+	 * the snapshots recovery pass runs.
+	 */
 	if (!k->level && k->btree_id == BTREE_ID_alloc)
 		iter_flags |= BTREE_ITER_CACHED;
+	else
+		update_flags |= BTREE_UPDATE_KEY_CACHE_RECLAIM;
 
 	bch2_trans_node_iter_init(trans, &iter, k->btree_id, k->k->k.p,
 				  BTREE_MAX_DEPTH, k->level,
@@ -610,7 +621,7 @@ static int bch2_journal_replay_key(struct btree_trans *trans,
 	if (k->overwritten)
 		goto out;
 
-	ret = bch2_trans_update(trans, &iter, k->k, BTREE_TRIGGER_NORUN);
+	ret = bch2_trans_update(trans, &iter, k->k, update_flags);
 out:
 	bch2_trans_iter_exit(trans, &iter);
 	return ret;

fs/bcachefs/subvolume.h

@@ -37,9 +37,34 @@ static inline struct snapshot_t *snapshot_t(struct bch_fs *c, u32 id)
 	return genradix_ptr(&c->snapshots, U32_MAX - 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(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_nth_parent(struct bch_fs *c, u32 id, u32 n)
+{
+	while (n--)
+		id = bch2_snapshot_parent(c, id);
+
+	return id;
 }
 
 static inline u32 bch2_snapshot_root(struct bch_fs *c, u32 id)
@@ -84,13 +109,7 @@ static inline u32 bch2_snapshot_sibling(struct bch_fs *c, u32 id)
 	return 0;
 }
 
-static inline bool bch2_snapshot_is_ancestor(struct bch_fs *c, u32 id, u32 ancestor)
-{
-	while (id && id < ancestor)
-		id = bch2_snapshot_parent(c, id);
-
-	return id == ancestor;
-}
+bool bch2_snapshot_is_ancestor(struct bch_fs *, u32, u32);
 
 static inline bool bch2_snapshot_has_children(struct bch_fs *c, u32 id)
 {

fs/bcachefs/subvolume_types.h

@@ -8,6 +8,8 @@ typedef DARRAY(u32) snapshot_id_list;
 
 struct snapshot_t {
 	u32			parent;
+	u32			skip[3];
+	u32			depth;
 	u32			children[2];
 	u32			subvol; /* Nonzero only if a subvolume points to this node: */
 	u32			tree;