> hmm, further checking 4.4.52, it was not really OK too:
> [ 1895.084325] blk_update_request: I/O error, dev sda, sector 239461608
> [ 1895.101099] sd 4:1:1:0: [sda] tag#1 FAILED Result: hostbyte=DID_OK
> driverbyte=DRIVER_SENSE
> [ 1895.101101] sd 4:1:1:0: [sda] tag#1 Sense Key : Illegal Request [current] 
> [ 1895.101103] sd 4:1:1:0: [sda] tag#1 Add. Sense: No additional sense 
> information

It's the "No additional sense information" that's the problem.

Under normal circumstances the drive would return "INVALID COMMAND
OPCODE" or "INVALID FIELD IN CDB" in the additional sense data. That
would cause us to disable WRITE SAME for the drive in question and
report to the block layer that it should zero the blocks manually.

However, because we don't get any of the expected sense data back in
your case, the failure manifests itself as a generic I/O failure to the
stack. And that's why the error is reported up to ext4 instead of the
block layer transparently doing fallback zeroing after WRITE SAME has
been disabled.

What kind of drive is attached to your controller? SATA? SAS?

Please send us the output of:

       sg_inq /dev/sda


       sg_vpd /dev/sda


Martin K. Petersen      Oracle Linux Engineering

Reply via email to