bcachefs: bch_subvolume::fs_path_parent

Record the filesystem path heirarchy for subvolumes in bch_subvolume

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

fs/bcachefs/bcachefs_format.h

@@ -840,7 +840,8 @@ struct bch_sb_field_downgrade {
 	x(snapshot_skiplists,		BCH_VERSION(1,  1))		\
 	x(deleted_inodes,		BCH_VERSION(1,  2))		\
 	x(rebalance_work,		BCH_VERSION(1,  3))		\
-	x(member_seq,			BCH_VERSION(1,  4))
+	x(member_seq,			BCH_VERSION(1,  4))		\
+	x(subvolume_fs_parent,		BCH_VERSION(1,  5))
 
 enum bcachefs_metadata_version {
 	bcachefs_metadata_version_min = 9,

fs/bcachefs/fs-common.c

@@ -107,6 +107,7 @@ int bch2_create_trans(struct btree_trans *trans,
 		u32 new_subvol, dir_snapshot;
 
 		ret = bch2_subvolume_create(trans, new_inode->bi_inum,
+					    dir.subvol,
 					    snapshot_src.subvol,
 					    &new_subvol, &snapshot,
 					    (flags & BCH_CREATE_SNAPSHOT_RO) != 0);
@@ -349,6 +350,22 @@ bool bch2_reinherit_attrs(struct bch_inode_unpacked *dst_u,
 	return ret;
 }
 
+static int subvol_update_parent(struct btree_trans *trans, u32 subvol, u32 new_parent)
+{
+	struct btree_iter iter;
+	struct bkey_i_subvolume *s =
+		bch2_bkey_get_mut_typed(trans, &iter,
+			BTREE_ID_subvolumes, POS(0, subvol),
+			BTREE_ITER_CACHED, subvolume);
+	int ret = PTR_ERR_OR_ZERO(s);
+	if (ret)
+		return ret;
+
+	s->v.fs_path_parent = cpu_to_le32(new_parent);
+	bch2_trans_iter_exit(trans, &iter);
+	return 0;
+}
+
 int bch2_rename_trans(struct btree_trans *trans,
 		      subvol_inum src_dir, struct bch_inode_unpacked *src_dir_u,
 		      subvol_inum dst_dir, struct bch_inode_unpacked *dst_dir_u,
@@ -410,6 +427,21 @@ int bch2_rename_trans(struct btree_trans *trans,
 			goto err;
 	}
 
+	if (src_inode_u->bi_subvol &&
+	    dst_dir.subvol != src_inode_u->bi_parent_subvol) {
+		ret = subvol_update_parent(trans, src_inode_u->bi_subvol, dst_dir.subvol);
+		if (ret)
+			goto err;
+	}
+
+	if (mode == BCH_RENAME_EXCHANGE &&
+	    dst_inode_u->bi_subvol &&
+	    src_dir.subvol != dst_inode_u->bi_parent_subvol) {
+		ret = subvol_update_parent(trans, dst_inode_u->bi_subvol, src_dir.subvol);
+		if (ret)
+			goto err;
+	}
+
 	/* Can't move across subvolumes, unless it's a subvolume root: */
 	if (src_dir.subvol != dst_dir.subvol &&
 	    (!src_inode_u->bi_subvol ||

fs/bcachefs/fsck.c

@@ -1744,11 +1744,12 @@ static int check_dirent_to_subvol(struct btree_trans *trans, struct btree_iter *
 				  struct bkey_s_c_dirent d)
 {
 	struct bch_fs *c = trans->c;
+	struct btree_iter subvol_iter = {};
 	struct bch_inode_unpacked subvol_root;
 	u32 parent_subvol = le32_to_cpu(d.v->d_parent_subvol);
 	u32 target_subvol = le32_to_cpu(d.v->d_child_subvol);
-	u32 target_snapshot, parent_snapshot;
+	u32 parent_snapshot;
-	u64 target_inum, parent_inum;
+	u64 parent_inum;
 	struct printbuf buf = PRINTBUF;
 	int ret = 0;
 
@@ -1777,8 +1778,11 @@ static int check_dirent_to_subvol(struct btree_trans *trans, struct btree_iter *
 		new_dirent->v.d_parent_subvol = cpu_to_le32(new_parent_subvol);
 	}
 
-	ret = subvol_lookup(trans, target_subvol,
+	struct bkey_s_c_subvolume s =
-			    &target_snapshot, &target_inum);
+		bch2_bkey_get_iter_typed(trans, &subvol_iter,
+					 BTREE_ID_subvolumes, POS(0, target_subvol),
+					 0, subvolume);
+	ret = bkey_err(s.s_c);
 	if (ret && !bch2_err_matches(ret, ENOENT))
 		return ret;
 
@@ -1791,8 +1795,24 @@ static int check_dirent_to_subvol(struct btree_trans *trans, struct btree_iter *
 		goto out;
 	}
 
-	ret = lookup_inode(trans, target_inum,
+	if (fsck_err_on(le32_to_cpu(s.v->fs_path_parent) != parent_subvol,
-			   &subvol_root, &target_snapshot);
+			c, subvol_fs_path_parent_wrong,
+			"subvol with wrong fs_path_parent, should be be %u\n%s",
+			parent_subvol,
+			(bch2_bkey_val_to_text(&buf, c, s.s_c), buf.buf))) {
+		struct bkey_i_subvolume *n =
+			bch2_bkey_make_mut_typed(trans, &subvol_iter, &s.s_c, 0, subvolume);
+		ret = PTR_ERR_OR_ZERO(n);
+		if (ret)
+			goto err;
+
+		n->v.fs_path_parent = cpu_to_le32(parent_subvol);
+	}
+
+	u64 target_inum = le64_to_cpu(s.v->inode);
+	u32 target_snapshot = le32_to_cpu(s.v->snapshot);
+
+	ret = lookup_inode(trans, target_inum, &subvol_root, &target_snapshot);
 	if (ret && !bch2_err_matches(ret, ENOENT))
 		return ret;
 
@@ -1814,6 +1834,7 @@ static int check_dirent_to_subvol(struct btree_trans *trans, struct btree_iter *
 out:
 err:
 fsck_err:
+	bch2_trans_iter_exit(trans, &subvol_iter);
 	printbuf_exit(&buf);
 	return ret;
 }

fs/bcachefs/sb-downgrade.c

@@ -45,7 +45,10 @@
 	  BIT_ULL(BCH_RECOVERY_PASS_check_inodes),		\
 	  BCH_FSCK_ERR_unlinked_inode_not_on_deleted_list)	\
 	x(rebalance_work,					\
-	  BIT_ULL(BCH_RECOVERY_PASS_set_fs_needs_rebalance))
+	  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)
 
 #define DOWNGRADE_TABLE()
 

fs/bcachefs/subvolume.c

@@ -20,6 +20,7 @@ static int check_subvol(struct btree_trans *trans,
 	struct bch_fs *c = trans->c;
 	struct bkey_s_c_subvolume subvol;
 	struct bch_snapshot snapshot;
+	struct printbuf buf = PRINTBUF;
 	unsigned snapid;
 	int ret = 0;
 
@@ -42,6 +43,20 @@ static int check_subvol(struct btree_trans *trans,
 		return ret ?: -BCH_ERR_transaction_restart_nested;
 	}
 
+	if (fsck_err_on(subvol.k->p.offset == BCACHEFS_ROOT_SUBVOL &&
+			subvol.v->fs_path_parent,
+			c, subvol_root_fs_path_parent_nonzero,
+			"root subvolume has nonzero fs_path_parent\n%s",
+			(bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
+		struct bkey_i_subvolume *n =
+			bch2_bkey_make_mut_typed(trans, iter, &subvol.s_c, 0, subvolume);
+		ret = PTR_ERR_OR_ZERO(n);
+		if (ret)
+			goto err;
+
+		n->v.fs_path_parent = 0;
+	}
+
 	struct bch_inode_unpacked inode;
 	struct btree_iter inode_iter = {};
 	ret = bch2_inode_peek_nowarn(trans, &inode_iter, &inode,
@@ -102,9 +117,9 @@ static int check_subvol(struct btree_trans *trans,
 			SET_BCH_SUBVOLUME_SNAP(&s->v, true);
 		}
 	}
-
 err:
 fsck_err:
+	printbuf_exit(&buf);
 	return ret;
 }
 
@@ -143,8 +158,10 @@ void bch2_subvolume_to_text(struct printbuf *out, struct bch_fs *c,
 		   le64_to_cpu(s.v->inode),
 		   le32_to_cpu(s.v->snapshot));
 
-	if (bkey_val_bytes(s.k) > offsetof(struct bch_subvolume, creation_parent))
+	if (bkey_val_bytes(s.k) > offsetof(struct bch_subvolume, creation_parent)) {
 		prt_printf(out, " creation_parent %u", le32_to_cpu(s.v->creation_parent));
+		prt_printf(out, " fs_parent %u", le32_to_cpu(s.v->fs_path_parent));
+	}
 }
 
 static __always_inline int
@@ -391,6 +408,7 @@ int bch2_subvolume_unlink(struct btree_trans *trans, u32 subvolid)
 }
 
 int bch2_subvolume_create(struct btree_trans *trans, u64 inode,
+			  u32 parent_subvolid,
 			  u32 src_subvolid,
 			  u32 *new_subvolid,
 			  u32 *new_snapshotid,
@@ -451,6 +469,7 @@ int bch2_subvolume_create(struct btree_trans *trans, u64 inode,
 	new_subvol->v.snapshot		= cpu_to_le32(new_nodes[0]);
 	new_subvol->v.inode		= cpu_to_le64(inode);
 	new_subvol->v.creation_parent	= cpu_to_le32(src_subvolid);
+	new_subvol->v.fs_path_parent	= cpu_to_le32(parent_subvolid);
 	new_subvol->v.otime.lo		= cpu_to_le64(bch2_current_time(c));
 	new_subvol->v.otime.hi		= 0;
 

fs/bcachefs/subvolume.h

@@ -30,8 +30,7 @@ int bch2_delete_dead_snapshots(struct bch_fs *);
 void bch2_delete_dead_snapshots_async(struct bch_fs *);
 
 int bch2_subvolume_unlink(struct btree_trans *, u32);
-int bch2_subvolume_create(struct btree_trans *, u64, u32,
+int bch2_subvolume_create(struct btree_trans *, u64, u32, u32, u32 *, u32 *, bool);
-			  u32 *, u32 *, bool);
 
 int bch2_fs_subvolumes_init(struct bch_fs *);
 

fs/bcachefs/subvolume_format.h

@@ -20,7 +20,7 @@ struct bch_subvolume {
 	 * this subvolume:
 	 */
 	__le32			creation_parent;
-	__le32			pad;
+	__le32			fs_path_parent;
 	bch_le128		otime;
 };