bcachefs: Inode backpointers

This patch adds two new inode fields, bi_dir and bi_dir_offset, that
point back to the inode's dirent.

Since we're only adding fields for a single backpointer, files that have
been hardlinked won't necessarily have valid backpointers: we also add a
new inode flag, BCH_INODE_BACKPTR_UNTRUSTED, that's set if an inode has
ever had multiple links to it. That's ok, because we only really need
this functionality for directories, which can never have multiple
hardlinks - when we add subvolumes, we'll need a way to enemurate and
print subvolumes, and this will let us reconstruct a path to a subvolume
root given a subvolume root inode.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

fs/bcachefs/bcachefs_format.h

@@ -710,7 +710,9 @@ struct bch_inode_generation {
 	x(bi_foreground_target,		16)	\
 	x(bi_background_target,		16)	\
 	x(bi_erasure_code,		16)	\
-	x(bi_fields_set,		16)
+	x(bi_fields_set,		16)	\
+	x(bi_dir,			64)	\
+	x(bi_dir_offset,		64)
 
 /* subset of BCH_INODE_FIELDS */
 #define BCH_INODE_OPTS()			\
@@ -746,6 +748,7 @@ enum {
 	__BCH_INODE_I_SIZE_DIRTY= 5,
 	__BCH_INODE_I_SECTORS_DIRTY= 6,
 	__BCH_INODE_UNLINKED	= 7,
+	__BCH_INODE_BACKPTR_UNTRUSTED = 8,
 
 	/* bits 20+ reserved for packed fields below: */
 };
@@ -758,6 +761,7 @@ enum {
 #define BCH_INODE_I_SIZE_DIRTY	(1 << __BCH_INODE_I_SIZE_DIRTY)
 #define BCH_INODE_I_SECTORS_DIRTY (1 << __BCH_INODE_I_SECTORS_DIRTY)
 #define BCH_INODE_UNLINKED	(1 << __BCH_INODE_UNLINKED)
+#define BCH_INODE_BACKPTR_UNTRUSTED (1 << __BCH_INODE_BACKPTR_UNTRUSTED)
 
 LE32_BITMASK(INODE_STR_HASH,	struct bch_inode, bi_flags, 20, 24);
 LE32_BITMASK(INODE_NR_FIELDS,	struct bch_inode, bi_flags, 24, 31);
@@ -1208,7 +1212,8 @@ enum bcachefs_metadata_version {
 	bcachefs_metadata_version_bkey_renumber		= 10,
 	bcachefs_metadata_version_inode_btree_change	= 11,
 	bcachefs_metadata_version_snapshot		= 12,
-	bcachefs_metadata_version_max			= 13,
+	bcachefs_metadata_version_inode_backpointers	= 13,
+	bcachefs_metadata_version_max			= 14,
 };
 
 #define bcachefs_metadata_version_current	(bcachefs_metadata_version_max - 1)

fs/bcachefs/dirent.c

@@ -141,7 +141,7 @@ static struct bkey_i_dirent *dirent_create_key(struct btree_trans *trans,
 int bch2_dirent_create(struct btree_trans *trans,
 		       u64 dir_inum, const struct bch_hash_info *hash_info,
 		       u8 type, const struct qstr *name, u64 dst_inum,
-		       int flags)
+		       u64 *dir_offset, int flags)
 {
 	struct bkey_i_dirent *dirent;
 	int ret;
@@ -151,8 +151,11 @@ int bch2_dirent_create(struct btree_trans *trans,
 	if (ret)
 		return ret;
 
-	return bch2_hash_set(trans, bch2_dirent_hash_desc, hash_info,
+	ret = bch2_hash_set(trans, bch2_dirent_hash_desc, hash_info,
 			    dir_inum, &dirent->k_i, flags);
+	*dir_offset = dirent->k.p.offset;
+
+	return ret;
 }
 
 static void dirent_copy_target(struct bkey_i_dirent *dst,
@@ -165,8 +168,8 @@ static void dirent_copy_target(struct bkey_i_dirent *dst,
 int bch2_dirent_rename(struct btree_trans *trans,
 		       u64 src_dir, struct bch_hash_info *src_hash,
 		       u64 dst_dir, struct bch_hash_info *dst_hash,
-		       const struct qstr *src_name, u64 *src_inum,
-		       const struct qstr *dst_name, u64 *dst_inum,
+		       const struct qstr *src_name, u64 *src_inum, u64 *src_offset,
+		       const struct qstr *dst_name, u64 *dst_inum, u64 *dst_offset,
 		       enum bch_rename_mode mode)
 {
 	struct btree_iter *src_iter = NULL, *dst_iter = NULL;
@@ -255,7 +258,7 @@ int bch2_dirent_rename(struct btree_trans *trans,
 				new_dst->k.p = src_iter->pos;
 				bch2_trans_update(trans, src_iter,
 						  &new_dst->k_i, 0);
-				goto out;
+				goto out_set_offset;
 			} else {
 				/* If we're overwriting, we can't insert new_dst
 				 * at a different slot because it has to
@@ -278,6 +281,9 @@ int bch2_dirent_rename(struct btree_trans *trans,
 
 	bch2_trans_update(trans, src_iter, &new_src->k_i, 0);
 	bch2_trans_update(trans, dst_iter, &new_dst->k_i, 0);
+out_set_offset:
+	*src_offset = new_src->k.p.offset;
+	*dst_offset = new_dst->k.p.offset;
 out:
 	bch2_trans_iter_put(trans, src_iter);
 	bch2_trans_iter_put(trans, dst_iter);

fs/bcachefs/dirent.h

@@ -31,7 +31,7 @@ static inline unsigned dirent_val_u64s(unsigned len)
 
 int bch2_dirent_create(struct btree_trans *, u64,
 		       const struct bch_hash_info *, u8,
-		       const struct qstr *, u64, int);
+		       const struct qstr *, u64, u64 *, int);
 
 int bch2_dirent_delete_at(struct btree_trans *,
 			  const struct bch_hash_info *,
@@ -46,8 +46,8 @@ enum bch_rename_mode {
 int bch2_dirent_rename(struct btree_trans *,
 		       u64, struct bch_hash_info *,
 		       u64, struct bch_hash_info *,
-		       const struct qstr *, u64 *,
-		       const struct qstr *, u64 *,
+		       const struct qstr *, u64 *, u64 *,
+		       const struct qstr *, u64 *, u64 *,
 		       enum bch_rename_mode);
 
 struct btree_iter *

fs/bcachefs/fs-common.c

@@ -20,8 +20,10 @@ int bch2_create_trans(struct btree_trans *trans, u64 dir_inum,
 {
 	struct bch_fs *c = trans->c;
 	struct btree_iter *dir_iter = NULL;
+	struct btree_iter *inode_iter = NULL;
 	struct bch_hash_info hash = bch2_hash_info_init(c, new_inode);
-	u64 now = bch2_current_time(trans->c);
+	u64 now = bch2_current_time(c);
+	u64 dir_offset = 0;
 	int ret;
 
 	dir_iter = bch2_inode_peek(trans, dir_u, dir_inum, BTREE_ITER_INTENT);
@@ -34,7 +36,8 @@ int bch2_create_trans(struct btree_trans *trans, u64 dir_inum,
 	if (!name)
 		new_inode->bi_flags |= BCH_INODE_UNLINKED;
 
-	ret = bch2_inode_create(trans, new_inode);
+	inode_iter = bch2_inode_create(trans, new_inode);
+	ret = PTR_ERR_OR_ZERO(inode_iter);
 	if (ret)
 		goto err;
 
@@ -66,11 +69,20 @@ int bch2_create_trans(struct btree_trans *trans, u64 dir_inum,
 		ret = bch2_dirent_create(trans, dir_inum, &dir_hash,
 					 mode_to_type(new_inode->bi_mode),
 					 name, new_inode->bi_inum,
+					 &dir_offset,
 					 BCH_HASH_SET_MUST_CREATE);
 		if (ret)
 			goto err;
 	}
+
+	if (c->sb.version >= bcachefs_metadata_version_inode_backpointers) {
+		new_inode->bi_dir		= dir_u->bi_inum;
+		new_inode->bi_dir_offset	= dir_offset;
+	}
+
+	ret = bch2_inode_write(trans, inode_iter, new_inode);
 err:
+	bch2_trans_iter_put(trans, inode_iter);
 	bch2_trans_iter_put(trans, dir_iter);
 	return ret;
 }
@@ -79,9 +91,11 @@ int bch2_link_trans(struct btree_trans *trans, u64 dir_inum,
 		    u64 inum, struct bch_inode_unpacked *dir_u,
 		    struct bch_inode_unpacked *inode_u, const struct qstr *name)
 {
+	struct bch_fs *c = trans->c;
 	struct btree_iter *dir_iter = NULL, *inode_iter = NULL;
 	struct bch_hash_info dir_hash;
-	u64 now = bch2_current_time(trans->c);
+	u64 now = bch2_current_time(c);
+	u64 dir_offset = 0;
 	int ret;
 
 	inode_iter = bch2_inode_peek(trans, inode_u, inum, BTREE_ITER_INTENT);
@@ -92,6 +106,8 @@ int bch2_link_trans(struct btree_trans *trans, u64 dir_inum,
 	inode_u->bi_ctime = now;
 	bch2_inode_nlink_inc(inode_u);
 
+	inode_u->bi_flags |= BCH_INODE_BACKPTR_UNTRUSTED;
+
 	dir_iter = bch2_inode_peek(trans, dir_u, dir_inum, 0);
 	ret = PTR_ERR_OR_ZERO(dir_iter);
 	if (ret)
@@ -99,12 +115,21 @@ int bch2_link_trans(struct btree_trans *trans, u64 dir_inum,
 
 	dir_u->bi_mtime = dir_u->bi_ctime = now;
 
-	dir_hash = bch2_hash_info_init(trans->c, dir_u);
+	dir_hash = bch2_hash_info_init(c, dir_u);
 
 	ret = bch2_dirent_create(trans, dir_inum, &dir_hash,
 				 mode_to_type(inode_u->bi_mode),
-				  name, inum, BCH_HASH_SET_MUST_CREATE) ?:
-		bch2_inode_write(trans, dir_iter, dir_u) ?:
+				 name, inum, &dir_offset,
+				 BCH_HASH_SET_MUST_CREATE);
+	if (ret)
+		goto err;
+
+	if (c->sb.version >= bcachefs_metadata_version_inode_backpointers) {
+		inode_u->bi_dir		= dir_inum;
+		inode_u->bi_dir_offset	= dir_offset;
+	}
+
+	ret =   bch2_inode_write(trans, dir_iter, dir_u) ?:
 		bch2_inode_write(trans, inode_iter, inode_u);
 err:
 	bch2_trans_iter_put(trans, dir_iter);
@@ -117,10 +142,11 @@ int bch2_unlink_trans(struct btree_trans *trans,
 		      struct bch_inode_unpacked *inode_u,
 		      const struct qstr *name)
 {
+	struct bch_fs *c = trans->c;
 	struct btree_iter *dir_iter = NULL, *dirent_iter = NULL,
 			  *inode_iter = NULL;
 	struct bch_hash_info dir_hash;
-	u64 inum, now = bch2_current_time(trans->c);
+	u64 inum, now = bch2_current_time(c);
 	struct bkey_s_c k;
 	int ret;
 
@@ -129,7 +155,7 @@ int bch2_unlink_trans(struct btree_trans *trans,
 	if (ret)
 		goto err;
 
-	dir_hash = bch2_hash_info_init(trans->c, dir_u);
+	dir_hash = bch2_hash_info_init(c, dir_u);
 
 	dirent_iter = __bch2_dirent_lookup_trans(trans, dir_inum, &dir_hash,
 						 name, BTREE_ITER_INTENT);
@@ -195,10 +221,12 @@ int bch2_rename_trans(struct btree_trans *trans,
 		      const struct qstr *dst_name,
 		      enum bch_rename_mode mode)
 {
+	struct bch_fs *c = trans->c;
 	struct btree_iter *src_dir_iter = NULL, *dst_dir_iter = NULL;
 	struct btree_iter *src_inode_iter = NULL, *dst_inode_iter = NULL;
 	struct bch_hash_info src_hash, dst_hash;
-	u64 src_inode, dst_inode, now = bch2_current_time(trans->c);
+	u64 src_inode, src_offset, dst_inode, dst_offset;
+	u64 now = bch2_current_time(c);
 	int ret;
 
 	src_dir_iter = bch2_inode_peek(trans, src_dir_u, src_dir,
@@ -207,7 +235,7 @@ int bch2_rename_trans(struct btree_trans *trans,
 	if (ret)
 		goto err;
 
-	src_hash = bch2_hash_info_init(trans->c, src_dir_u);
+	src_hash = bch2_hash_info_init(c, src_dir_u);
 
 	if (dst_dir != src_dir) {
 		dst_dir_iter = bch2_inode_peek(trans, dst_dir_u, dst_dir,
@@ -216,7 +244,7 @@ int bch2_rename_trans(struct btree_trans *trans,
 		if (ret)
 			goto err;
 
-		dst_hash = bch2_hash_info_init(trans->c, dst_dir_u);
+		dst_hash = bch2_hash_info_init(c, dst_dir_u);
 	} else {
 		dst_dir_u = src_dir_u;
 		dst_hash = src_hash;
@@ -225,8 +253,8 @@ int bch2_rename_trans(struct btree_trans *trans,
 	ret = bch2_dirent_rename(trans,
 				 src_dir, &src_hash,
 				 dst_dir, &dst_hash,
-				 src_name, &src_inode,
-				 dst_name, &dst_inode,
+				 src_name, &src_inode, &src_offset,
+				 dst_name, &dst_inode, &dst_offset,
 				 mode);
 	if (ret)
 		goto err;
@@ -245,6 +273,16 @@ int bch2_rename_trans(struct btree_trans *trans,
 			goto err;
 	}
 
+	if (c->sb.version >= bcachefs_metadata_version_inode_backpointers) {
+		src_inode_u->bi_dir		= dst_dir_u->bi_inum;
+		src_inode_u->bi_dir_offset	= dst_offset;
+
+		if (mode == BCH_RENAME_EXCHANGE) {
+			dst_inode_u->bi_dir		= src_dir_u->bi_inum;
+			dst_inode_u->bi_dir_offset	= src_offset;
+		}
+	}
+
 	if (mode == BCH_RENAME_OVERWRITE) {
 		if (S_ISDIR(src_inode_u->bi_mode) !=
 		    S_ISDIR(dst_inode_u->bi_mode)) {

fs/bcachefs/fsck.c

@@ -675,6 +675,39 @@ static int check_dirents(struct bch_fs *c)
 			continue;
 		}
 
+		if (!target.bi_nlink &&
+		    !(target.bi_flags & BCH_INODE_BACKPTR_UNTRUSTED) &&
+		    (target.bi_dir != k.k->p.inode ||
+		     target.bi_dir_offset != k.k->p.offset) &&
+		    (fsck_err_on(c->sb.version >= bcachefs_metadata_version_inode_backpointers, c,
+				 "inode %llu has wrong backpointer:\n"
+				 "got       %llu:%llu\n"
+				 "should be %llu:%llu",
+				 d_inum,
+				 target.bi_dir,
+				 target.bi_dir_offset,
+				 k.k->p.inode,
+				 k.k->p.offset) ||
+		     c->opts.version_upgrade)) {
+			struct bkey_inode_buf p;
+
+			target.bi_dir		= k.k->p.inode;
+			target.bi_dir_offset	= k.k->p.offset;
+			bch2_trans_unlock(&trans);
+
+			bch2_inode_pack(c, &p, &target);
+
+			ret = bch2_btree_insert(c, BTREE_ID_inodes,
+						&p.inode.k_i, NULL, NULL,
+						BTREE_INSERT_NOFAIL|
+						BTREE_INSERT_LAZY_RW);
+			if (ret) {
+				bch_err(c, "error in fsck: error %i updating inode", ret);
+				goto err;
+			}
+			continue;
+		}
+
 		if (fsck_err_on(have_target &&
 				d.v->d_type !=
 				mode_to_type(target.bi_mode), c,
@@ -1314,6 +1347,16 @@ static int check_inode(struct btree_trans *trans,
 		do_update = true;
 	}
 
+	if (!S_ISDIR(u.bi_mode) &&
+	    u.bi_nlink &&
+	    !(u.bi_flags & BCH_INODE_BACKPTR_UNTRUSTED) &&
+	    (fsck_err_on(c->sb.version >= bcachefs_metadata_version_inode_backpointers, c,
+			 "inode missing BCH_INODE_BACKPTR_UNTRUSTED flags") ||
+	     c->opts.version_upgrade)) {
+		u.bi_flags |= BCH_INODE_BACKPTR_UNTRUSTED;
+		do_update = true;
+	}
+
 	if (do_update) {
 		struct bkey_inode_buf p;
 

fs/bcachefs/inode.c

@@ -470,11 +470,10 @@ static inline u32 bkey_generation(struct bkey_s_c k)
 	}
 }
 
-int bch2_inode_create(struct btree_trans *trans,
+struct btree_iter *bch2_inode_create(struct btree_trans *trans,
 				     struct bch_inode_unpacked *inode_u)
 {
 	struct bch_fs *c = trans->c;
-	struct bkey_inode_buf *inode_p;
 	struct btree_iter *iter = NULL;
 	struct bkey_s_c k;
 	u64 min, max, start, *hint;
@@ -494,10 +493,6 @@ int bch2_inode_create(struct btree_trans *trans,
 
 	if (start >= max || start < min)
 		start = min;
-
-	inode_p = bch2_trans_kmalloc(trans, sizeof(*inode_p));
-	if (IS_ERR(inode_p))
-		return PTR_ERR(inode_p);
 again:
 	for_each_btree_key(trans, iter, BTREE_ID_inodes, POS(0, start),
 			   BTREE_ITER_SLOTS|BTREE_ITER_INTENT, k, ret) {
@@ -521,7 +516,7 @@ int bch2_inode_create(struct btree_trans *trans,
 	bch2_trans_iter_put(trans, iter);
 
 	if (ret)
-		return ret;
+		return ERR_PTR(ret);
 
 	if (start != min) {
 		/* Retry from start */
@@ -529,15 +524,12 @@ int bch2_inode_create(struct btree_trans *trans,
 		goto again;
 	}
 
-	return -ENOSPC;
+	return ERR_PTR(-ENOSPC);
 found_slot:
 	*hint			= k.k->p.offset;
 	inode_u->bi_inum	= k.k->p.offset;
 	inode_u->bi_generation	= bkey_generation(k);
-
-	ret = bch2_inode_write(trans, iter, inode_u);
-	bch2_trans_iter_put(trans, iter);
-	return ret;
+	return iter;
 }
 
 int bch2_inode_rm(struct bch_fs *c, u64 inode_nr, bool cached)

fs/bcachefs/inode.h

@@ -69,7 +69,8 @@ void bch2_inode_init(struct bch_fs *, struct bch_inode_unpacked *,
 		     uid_t, gid_t, umode_t, dev_t,
 		     struct bch_inode_unpacked *);
 
-int bch2_inode_create(struct btree_trans *, struct bch_inode_unpacked *);
+struct btree_iter *bch2_inode_create(struct btree_trans *,
+				     struct bch_inode_unpacked *);
 
 int bch2_inode_rm(struct bch_fs *, u64, bool);