ext4: don't set up encryption key during jbd2 transaction

From: Eric Biggers <ebiggers@google.com>

commit 4c0d577 upstream.

Commit a80f7fc ("ext4: fixup ext4_fc_track_* functions' signature")
extended the scope of the transaction in ext4_unlink() too far, making
it include the call to ext4_find_entry().  However, ext4_find_entry()
can deadlock when called from within a transaction because it may need
to set up the directory's encryption key.

Fix this by restoring the transaction to its original scope.

Reported-by: syzbot+1a748d0007eeac3ab079@syzkaller.appspotmail.com
Fixes: a80f7fc ("ext4: fixup ext4_fc_track_* functions' signature")
Cc: <stable@vger.kernel.org> # v5.10+
Signed-off-by: Eric Biggers <ebiggers@google.com>
Link: https://lore.kernel.org/r/20221106224841.279231-3-ebiggers@kernel.org
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

fs/ext4/ext4.h

@@ -3647,8 +3647,8 @@ extern void ext4_initialize_dirent_tail(struct buffer_head *bh,
 					unsigned int blocksize);
 extern int ext4_handle_dirty_dirblock(handle_t *handle, struct inode *inode,
 				      struct buffer_head *bh);
-extern int __ext4_unlink(handle_t *handle, struct inode *dir, const struct qstr *d_name,
-			 struct inode *inode);
+extern int __ext4_unlink(struct inode *dir, const struct qstr *d_name,
+			 struct inode *inode, struct dentry *dentry);
 extern int __ext4_link(struct inode *dir, struct inode *inode,
 		       struct dentry *dentry);
 

fs/ext4/fast_commit.c

@@ -1330,7 +1330,7 @@ static int ext4_fc_replay_unlink(struct super_block *sb, struct ext4_fc_tl *tl,
 		return 0;
 	}
 
-	ret = __ext4_unlink(NULL, old_parent, &entry, inode);
+	ret = __ext4_unlink(old_parent, &entry, inode, NULL);
 	/* -ENOENT ok coz it might not exist anymore. */
 	if (ret == -ENOENT)
 		ret = 0;

fs/ext4/namei.c

@@ -3204,14 +3204,20 @@ static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
 	return retval;
 }
 
-int __ext4_unlink(handle_t *handle, struct inode *dir, const struct qstr *d_name,
-		  struct inode *inode)
+int __ext4_unlink(struct inode *dir, const struct qstr *d_name,
+		  struct inode *inode,
+		  struct dentry *dentry /* NULL during fast_commit recovery */)
 {
 	int retval = -ENOENT;
 	struct buffer_head *bh;
 	struct ext4_dir_entry_2 *de;
+	handle_t *handle;
 	int skip_remove_dentry = 0;
 
+	/*
+	 * Keep this outside the transaction; it may have to set up the
+	 * directory's encryption key, which isn't GFP_NOFS-safe.
+	 */
 	bh = ext4_find_entry(dir, d_name, &de, NULL);
 	if (IS_ERR(bh))
 		return PTR_ERR(bh);
@@ -3228,7 +3234,14 @@ int __ext4_unlink(handle_t *handle, struct inode *dir, const struct qstr *d_name
 		if (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
 			skip_remove_dentry = 1;
 		else
-			goto out;
+			goto out_bh;
+	}
+
+	handle = ext4_journal_start(dir, EXT4_HT_DIR,
+				    EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
+	if (IS_ERR(handle)) {
+		retval = PTR_ERR(handle);
+		goto out_bh;
 	}
 
 	if (IS_DIRSYNC(dir))
@@ -3237,12 +3250,12 @@ int __ext4_unlink(handle_t *handle, struct inode *dir, const struct qstr *d_name
 	if (!skip_remove_dentry) {
 		retval = ext4_delete_entry(handle, dir, de, bh);
 		if (retval)
-			goto out;
+			goto out_handle;
 		dir->i_ctime = dir->i_mtime = current_time(dir);
 		ext4_update_dx_flag(dir);
 		retval = ext4_mark_inode_dirty(handle, dir);
 		if (retval)
-			goto out;
+			goto out_handle;
 	} else {
 		retval = 0;
 	}
@@ -3255,15 +3268,17 @@ int __ext4_unlink(handle_t *handle, struct inode *dir, const struct qstr *d_name
 		ext4_orphan_add(handle, inode);
 	inode->i_ctime = current_time(inode);
 	retval = ext4_mark_inode_dirty(handle, inode);
-
-out:
+	if (dentry && !retval)
+		ext4_fc_track_unlink(handle, dentry);
+out_handle:
+	ext4_journal_stop(handle);
+out_bh:
 	brelse(bh);
 	return retval;
 }
 
 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
 {
-	handle_t *handle;
 	int retval;
 
 	if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
@@ -3281,16 +3296,7 @@ static int ext4_unlink(struct inode *dir, struct dentry *dentry)
 	if (retval)
 		goto out_trace;
 
-	handle = ext4_journal_start(dir, EXT4_HT_DIR,
-				    EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
-	if (IS_ERR(handle)) {
-		retval = PTR_ERR(handle);
-		goto out_trace;
-	}
-
-	retval = __ext4_unlink(handle, dir, &dentry->d_name, d_inode(dentry));
-	if (!retval)
-		ext4_fc_track_unlink(handle, dentry);
+	retval = __ext4_unlink(dir, &dentry->d_name, d_inode(dentry), dentry);
 #ifdef CONFIG_UNICODE
 	/* VFS negative dentries are incompatible with Encoding and
 	 * Case-insensitiveness. Eventually we'll want avoid
@@ -3301,8 +3307,6 @@ static int ext4_unlink(struct inode *dir, struct dentry *dentry)
 	if (IS_CASEFOLDED(dir))
 		d_invalidate(dentry);
 #endif
-	if (handle)
-		ext4_journal_stop(handle);
 
 out_trace:
 	trace_ext4_unlink_exit(dentry, retval);