From: Ryusuke Konishi <konishi.ryusuke@xxxxxxxxx> commit 7373a51e7998b508af7136530f3a997b286ce81c upstream. The error handling in nilfs_empty_dir() when a directory folio/page read fails is incorrect, as in the old ext2 implementation, and if the folio/page cannot be read or nilfs_check_folio() fails, it will falsely determine the directory as empty and corrupt the file system. In addition, since nilfs_empty_dir() does not immediately return on a failed folio/page read, but continues to loop, this can cause a long loop with I/O if i_size of the directory's inode is also corrupted, causing the log writer thread to wait and hang, as reported by syzbot. Fix these issues by making nilfs_empty_dir() immediately return a false value (0) if it fails to get a directory folio/page. Link: https://lkml.kernel.org/r/20240604134255.7165-1-konishi.ryusuke@xxxxxxxxx Signed-off-by: Ryusuke Konishi <konishi.ryusuke@xxxxxxxxx> Reported-by: syzbot+c8166c541d3971bf6c87@xxxxxxxxxxxxxxxxxxxxxxxxx Closes: https://syzkaller.appspot.com/bug?extid=c8166c541d3971bf6c87 Fixes: 2ba466d74ed7 ("nilfs2: directory entry operations") Tested-by: Ryusuke Konishi <konishi.ryusuke@xxxxxxxxx> Cc: <stable@xxxxxxxxxxxxxxx> Signed-off-by: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx> Signed-off-by: Greg Kroah-Hartman <gregkh@xxxxxxxxxxxxxxxxxxx> --- fs/nilfs2/dir.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) --- a/fs/nilfs2/dir.c +++ b/fs/nilfs2/dir.c @@ -608,7 +608,7 @@ int nilfs_empty_dir(struct inode *inode) kaddr = nilfs_get_folio(inode, i, &folio); if (IS_ERR(kaddr)) - continue; + return 0; de = (struct nilfs_dir_entry *)kaddr; kaddr += nilfs_last_byte(inode, i) - NILFS_DIR_REC_LEN(1); Patches currently in stable-queue which might be from konishi.ryusuke@xxxxxxxxx are queue-6.9/nilfs2-fix-nilfs_empty_dir-misjudgment-and-long-loop-on-i-o-errors.patch queue-6.9/nilfs2-fix-potential-kernel-bug-due-to-lack-of-writeback-flag-waiting.patch
