[LU-2462] debugfs doesn't care about DIRENT_LUFID flag while inserting new entries

• Type: Bug
• Resolution: Fixed
• Priority: Critical
• Labels: e2fsprogs releases
• Severity: 3
• Assignee: Andreas Dilger
• Reporter: Mikhail Pershin

Description

Directory entries with DIRENT_LUFID flag in d_type contains packed FID right after entry name, so they have longer rec_len. Debugfs doesn't check that when inserting new entry and corrupts directory.

debugfs:  cd ROOT
debugfs:  ls
 50039  (28) .    2  (28) ..    50041  (36) .lustre    104  (32) file-0   
 105  (32) file-1    106  (32) file-2    107  (32) file-3    108  (32) file-4   
 109  (32) file-5    110  (32) file-6    111  (32) file-7    112  (32) file-8   
 113  (3716) file-9   
debugfs:  rm file-2
debugfs:  write /etc/hosts file-2
Allocated inode: 50042
debugfs:  ls
 50039  (12) .    50042  (16) file-2    2  (28) ..    50041  (36) .lustre   
 104  (32) file-0    105  (64) file-1    107  (32) file-3    108  (32) file-4   
 109  (32) file-5    110  (32) file-6    111  (32) file-7    112  (32) file-8   
 113  (3716) file-9   

Note how rec_len of '.' is changed. Now let's mount lustre back:

[root@node7 tests]# ls /mnt/lustre
ls: reading directory /mnt/lustre: Input/output error

dmesg says:

LDISKFS-fs error (device loop0): htree_dirblock_to_tree: bad entry in directory #50039: rec_len is too small for name_len - block=12610offset=0(0), inode=50039, rec_len=12, name_len=1
Aborting journal on device loop0-8.
LDISKFS-fs (loop0): Remounting filesystem read-only

Issue Links

• is related to: LU-838 "lfs path2fid /mnt/lustre" (ROOT) returns inode number