bcachefs: BTREE_ID_subvolume_children

Add a btree to record a parent -> child subvolume relationships,
according to the filesystem heirarchy.

The subvolume_children btree is a bitset btree: if a bit is set at pos
p, that means p.offset is a child of subvolume p.inode.

This will be used for efficiently listing subvolumes, as well as
recursive deletion.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

fs/bcachefs/bcachefs.h

@@ -504,6 +504,7 @@ enum gc_phase {
 	GC_PHASE_BTREE_deleted_inodes,
 	GC_PHASE_BTREE_logged_ops,
 	GC_PHASE_BTREE_rebalance_work,
+	GC_PHASE_BTREE_subvolume_children,
 
 	GC_PHASE_PENDING_DELETE,
 };

fs/bcachefs/bcachefs_format.h

@@ -841,7 +841,8 @@ struct bch_sb_field_downgrade {
 	x(deleted_inodes,		BCH_VERSION(1,  2))		\
 	x(rebalance_work,		BCH_VERSION(1,  3))		\
 	x(member_seq,			BCH_VERSION(1,  4))		\
-	x(subvolume_fs_parent,		BCH_VERSION(1,  5))
+	x(subvolume_fs_parent,		BCH_VERSION(1,  5))		\
+	x(btree_subvolume_children,	BCH_VERSION(1,  6))
 
 enum bcachefs_metadata_version {
 	bcachefs_metadata_version_min = 9,
@@ -1489,7 +1490,9 @@ enum btree_id_flags {
 	  BIT_ULL(KEY_TYPE_logged_op_truncate)|					\
 	  BIT_ULL(KEY_TYPE_logged_op_finsert))					\
 	x(rebalance_work,	18,	BTREE_ID_SNAPSHOT_FIELD,		\
-	  BIT_ULL(KEY_TYPE_set)|BIT_ULL(KEY_TYPE_cookie))
+	  BIT_ULL(KEY_TYPE_set)|BIT_ULL(KEY_TYPE_cookie))			\
+	x(subvolume_children,	19,	0,					\
+	  BIT_ULL(KEY_TYPE_set))
 
 enum btree_id {
 #define x(name, nr, ...) BTREE_ID_##name = nr,

fs/bcachefs/btree_types.h

@@ -654,6 +654,7 @@ const char *bch2_btree_node_type_str(enum btree_node_type);
 	 BIT_ULL(BKEY_TYPE_inodes)|			\
 	 BIT_ULL(BKEY_TYPE_stripes)|			\
 	 BIT_ULL(BKEY_TYPE_reflink)|			\
+	 BIT_ULL(BKEY_TYPE_subvolumes)|			\
 	 BIT_ULL(BKEY_TYPE_btree))
 
 #define BTREE_NODE_TYPE_HAS_ATOMIC_TRIGGERS		\

fs/bcachefs/recovery_types.h

@@ -34,6 +34,7 @@
 	x(check_snapshot_trees,			18, PASS_ONLINE|PASS_FSCK)	\
 	x(check_snapshots,			19, PASS_ONLINE|PASS_FSCK)	\
 	x(check_subvols,			20, PASS_ONLINE|PASS_FSCK)	\
+	x(check_subvol_children,		35, PASS_ONLINE|PASS_FSCK)	\
 	x(delete_dead_snapshots,		21, PASS_ONLINE|PASS_FSCK)	\
 	x(fs_upgrade_for_subvolumes,		22, 0)				\
 	x(resume_logged_ops,			23, PASS_ALWAYS)		\

fs/bcachefs/sb-downgrade.c

@@ -48,7 +48,10 @@
 	  BIT_ULL(BCH_RECOVERY_PASS_set_fs_needs_rebalance))	\
 	x(subvolume_fs_parent,					\
 	  BIT_ULL(BCH_RECOVERY_PASS_check_dirents),		\
-	  BCH_FSCK_ERR_subvol_fs_path_parent_wrong)
+	  BCH_FSCK_ERR_subvol_fs_path_parent_wrong)		\
+	x(btree_subvolume_children,				\
+	  BIT_ULL(BCH_RECOVERY_PASS_check_subvols),		\
+	  BCH_FSCK_ERR_subvol_children_not_set)
 
 #define DOWNGRADE_TABLE()
 

fs/bcachefs/subvolume.c

@@ -13,12 +13,24 @@
 
 static int bch2_subvolume_delete(struct btree_trans *, u32);
 
+static struct bpos subvolume_children_pos(struct bkey_s_c k)
+{
+	if (k.k->type != KEY_TYPE_subvolume)
+		return POS_MIN;
+
+	struct bkey_s_c_subvolume s = bkey_s_c_to_subvolume(k);
+	if (!s.v->fs_path_parent)
+		return POS_MIN;
+	return POS(le32_to_cpu(s.v->fs_path_parent), s.k->p.offset);
+}
+
 static int check_subvol(struct btree_trans *trans,
 			struct btree_iter *iter,
 			struct bkey_s_c k)
 {
 	struct bch_fs *c = trans->c;
 	struct bkey_s_c_subvolume subvol;
+	struct btree_iter subvol_children_iter = {};
 	struct bch_snapshot snapshot;
 	struct printbuf buf = PRINTBUF;
 	unsigned snapid;
@@ -57,6 +69,28 @@ static int check_subvol(struct btree_trans *trans,
 		n->v.fs_path_parent = 0;
 	}
 
+	if (subvol.v->fs_path_parent) {
+		struct bpos pos = subvolume_children_pos(k);
+
+		struct bkey_s_c subvol_children_k =
+			bch2_bkey_get_iter(trans, &subvol_children_iter,
+					   BTREE_ID_subvolume_children, pos, 0);
+		ret = bkey_err(subvol_children_k);
+		if (ret)
+			goto err;
+
+		if (fsck_err_on(subvol_children_k.k->type != KEY_TYPE_set,
+				c, subvol_children_not_set,
+				"subvolume not set in subvolume_children btree at %llu:%llu\n%s",
+				pos.inode, pos.offset,
+				(printbuf_reset(&buf),
+				 bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
+			ret = bch2_btree_bit_mod(trans, BTREE_ID_subvolume_children, pos, true);
+			if (ret)
+				goto err;
+		}
+	}
+
 	struct bch_inode_unpacked inode;
 	struct btree_iter inode_iter = {};
 	ret = bch2_inode_peek_nowarn(trans, &inode_iter, &inode,
@@ -119,6 +153,7 @@ static int check_subvol(struct btree_trans *trans,
 	}
 err:
 fsck_err:
+	bch2_trans_iter_exit(trans, &subvol_children_iter);
 	printbuf_exit(&buf);
 	return ret;
 }
@@ -134,6 +169,42 @@ int bch2_check_subvols(struct bch_fs *c)
 	return ret;
 }
 
+static int check_subvol_child(struct btree_trans *trans,
+			      struct btree_iter *child_iter,
+			      struct bkey_s_c child_k)
+{
+	struct bch_fs *c = trans->c;
+	struct bch_subvolume s;
+	int ret = bch2_bkey_get_val_typed(trans, BTREE_ID_subvolumes, POS(0, child_k.k->p.offset),
+					  0, subvolume, &s);
+	if (ret && !bch2_err_matches(ret, ENOENT))
+		return ret;
+
+	if (fsck_err_on(ret ||
+			le32_to_cpu(s.fs_path_parent) != child_k.k->p.inode,
+			c, subvol_children_bad,
+			"incorrect entry in subvolume_children btree %llu:%llu",
+			child_k.k->p.inode, child_k.k->p.offset)) {
+		ret = bch2_btree_delete_at(trans, child_iter, 0);
+		if (ret)
+			goto err;
+	}
+err:
+fsck_err:
+	return ret;
+}
+
+int bch2_check_subvol_children(struct bch_fs *c)
+{
+	int ret = bch2_trans_run(c,
+		for_each_btree_key_commit(trans, iter,
+				BTREE_ID_subvolume_children, POS_MIN, BTREE_ITER_PREFETCH, k,
+				NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
+			check_subvol_child(trans, &iter, k)));
+	bch_err_fn(c, ret);
+	return 0;
+}
+
 /* Subvolumes: */
 
 int bch2_subvolume_invalid(struct bch_fs *c, struct bkey_s_c k,
@@ -164,6 +235,33 @@ void bch2_subvolume_to_text(struct printbuf *out, struct bch_fs *c,
 	}
 }
 
+static int subvolume_children_mod(struct btree_trans *trans, struct bpos pos, bool set)
+{
+	return !bpos_eq(pos, POS_MIN)
+		? bch2_btree_bit_mod(trans, BTREE_ID_subvolume_children, pos, set)
+		: 0;
+}
+
+int bch2_subvolume_trigger(struct btree_trans *trans,
+			   enum btree_id btree_id, unsigned level,
+			   struct bkey_s_c old, struct bkey_s new,
+			   unsigned flags)
+{
+	if (flags & BTREE_TRIGGER_TRANSACTIONAL) {
+		struct bpos children_pos_old = subvolume_children_pos(old);
+		struct bpos children_pos_new = subvolume_children_pos(new.s_c);
+
+		if (!bpos_eq(children_pos_old, children_pos_new)) {
+			int ret = subvolume_children_mod(trans, children_pos_old, false) ?:
+				  subvolume_children_mod(trans, children_pos_new, true);
+			if (ret)
+				return ret;
+		}
+	}
+
+	return 0;
+}
+
 static __always_inline int
 bch2_subvolume_get_inlined(struct btree_trans *trans, unsigned subvol,
 			   bool inconsistent_if_not_found,

fs/bcachefs/subvolume.h

@@ -8,14 +8,18 @@
 enum bkey_invalid_flags;
 
 int bch2_check_subvols(struct bch_fs *);
+int bch2_check_subvol_children(struct bch_fs *);
 
 int bch2_subvolume_invalid(struct bch_fs *, struct bkey_s_c,
 			   enum bkey_invalid_flags, struct printbuf *);
 void bch2_subvolume_to_text(struct printbuf *, struct bch_fs *, struct bkey_s_c);
+int bch2_subvolume_trigger(struct btree_trans *, enum btree_id, unsigned,
+			   struct bkey_s_c, struct bkey_s, unsigned);
 
 #define bch2_bkey_ops_subvolume ((struct bkey_ops) {		\
 	.key_invalid	= bch2_subvolume_invalid,		\
 	.val_to_text	= bch2_subvolume_to_text,		\
+	.trigger	= bch2_subvolume_trigger,		\
 	.min_val_size	= 16,					\
 })