On Mon, 2016-06-27 at 07:35 +0300, Andrei Borzenkov wrote: > The problem is that current implementation of RAID56 puts exactly CoW > data at risk. I.e. writing new (copy of) data may suddenly make old > (copy of) data inaccessible, even though it had been safely committed > to > disk and is now in read-only snapshot. Sure,... mine was just a general thing to be added. No checksums => no way to tell which block is valid in case of silent block errors => no way to recover unless by chance => should be included as a warning, especially as userland software starts to automatically set nodatacow (IIRC systemd does so), thereby silently breaking functionality (integrity+recoverability) assumed by the user. Cheers, Chris-
Attachment:
smime.p7s
Description: S/MIME cryptographic signature
