[LU-16886] Reading filesystem root directory fails after setting project ID

• Status: Open
• Project: Lustre
• Affects Version/s: Lustre 2.15.2
• Type: Bug
• Priority: Critical
• Reporter: Robert Redl
• Assignee: WC Triage
• Resolution: Unresolved

Environment

Lustre: 2.15.2
ZFS: 2.1.6

• Severity: 3

Description

After accidentally setting a project ID on the root directory of a filesystem, it is not readable anymore:

lfs project -p 16 -s -r /mnt
ls /mnt
ls /mnt/
ls: reading directory '/mnt/': Input/output error
ls /mnt/
ls: reading directory '/mnt/': Invalid argument 

The lfs project command did return immediately without descending into any sub-directories. The project IDs of sub-directories are unchanged. Sub-directories also remain readable ls /mnt/subdir is not giving any errors.

 

Removing the project ID did not help:

lfs project -C /mnt
ls /mnt/
ls: reading directory '/mnt/': Input/output error 

 

LFSCK was also not able to correct the error.

Form the client log:

Jun 09 10:54:27 client kernel: LustreError: 470146:0:(statahead.c:1370:is_first_dirent()) project: reading dir [0x200000007:0x1:0x0] at 0 opendir_pid = 470146 : rc = -22
Jun 09 10:54:27 client kernel: LustreError: 470146:0:(mdc_request.c:1489:mdc_read_page()) project-MDT0000-mdc-ffff9bdde4edb800: dir page locate: [0x200000007:0x1:0x0] at 0: rc -5 

 

From the MDS log:

Jun 09 10:54:27 server kernel: LustreError: 2721866:0:(osd_index.c:217:__osd_xattr_load_by_oid()) project-MDT0000: can't get bonus, rc = -22
Jun 09 10:54:27 server kernel: LustreError: 2721866:0:(osd_index.c:217:__osd_xattr_load_by_oid()) Skipped 1 previous similar message
Jun 09 10:54:27 server kernel: Lustre: 2721866:0:(mdd_object.c:3684:mdd_dir_page_build()) build page failed: -22!
Jun 09 10:54:27 server kernel: Lustre: 2721866:0:(mdd_object.c:3684:mdd_dir_page_build()) Skipped 1 previous similar message

 

Additional info: the filesystem was created before project IDs have been available. After the upgrade setting project IDs on sub-directories failed. However, this issue was solved by the patch for LU-13189.