A filesystem corruption bug breaks loose

Posted Dec 10, 2018 23:20 UTC (Mon) by axboe (subscriber, #904)
In reply to: A filesystem corruption bug breaks loose by mangix
Parent article: A filesystem corruption bug breaks loose

> This seems to affect users of NVME devices, as those seem to use a scheduler of none and use blk-mq by default.

No, as the article mentions, only a subset of SCSI was affected. To hit this issue, you need:

1) Driver that can hit a BUSY condition on dispatch. SCSI fits that bill.
2) Driver that retains state over a BUSY condition. Only SCSI fits that bill.
3) Device that hits this condition outside of normal queue full. Only ATA fits that bill, with some commands not being queued.
4) Device NOT using an IO scheduler
5) Lots of bad luck

All of these conditions have to be met, or you cannot hit it.

A filesystem corruption bug breaks loose

Posted Dec 11, 2018 0:46 UTC (Tue) by mangix (guest, #126006) [Link]

Got it. Thanks for the clarification.