[LU-1367] Help finding bad LBA

• Status: Resolved
• Project: Lustre
• Type: Task
• Priority: Minor
• Reporter: Roger Spellman (Inactive)
• Assignee: Peter Jones
• Resolution: Incomplete

Description

I am on the phone with our storage vendor. We have some corruption on the disk, and the file system when readonly. We suspect that there might be a FW issue in the RAID controller.

When we run fsck, we see a couple of messages like this:

Free inodes count wrong for group #45590 (1, counted=0).

The storage vendor would like to know the LBA associated with this this data. Can you tell me how to find that?

Thanks.

Comments

[Andreas Dilger]

The "inode count wrong" is itself not associated with a particular block address. Instead, this is a side-effect of some previous error that was corrected (e.g. a corrupted or unreferenced inode being deleted), and now the summary count for the group that held that inode is incorrect.

It is possible to find the filesystem block number for a particular inode using the "debugfs -c -R 'imap <inode number>' /dev/XXX" command. Once the filesystem block number is found, this needs to be converted to a sector number (usually block_number * 8, assuming 512-byte sectors) and added to the offset of the start of the partition (0, in the recommended case of using the whole device for a Lustre target).

[Roger Spellman (Inactive)]

Andreas, Thanks for this info.
From this message, I don't see the inode number, just the group number.
If I knew the size of the group, I should would be able to calculate the LBA of that particular group, then I could use the offset of the inode count and free inodes to get their LBAs.
Do you know the size of the group?

[John Fuchs-Chesney (Inactive)]

Roger – can we mark this issue as resolved?
Thanks,
~ jfc.

[John Fuchs-Chesney (Inactive)]

Looks like we will not pursue this issue further.