On Oct 18, 2016, at 7:19 PM, NeilBrown [off-list ref] wrote:

On Thu, Oct 13 2016, Song Liu wrote:

quoted

r5c_state have 4 states:
* no-cache;
* write-through (write journal only);
* write-back (w/ write cache);
* cache-broken (journal missing or Faulty)

I don't like this.

We already have a mechanism for removing a failed device from
an array: write 'remove' to the 'state' file in the relevant dev-*
subdirectory.
You can also use that file to tell if the journal has failed.

I think we would like an separate mechanism to remove "journal feature",
not "journal device".  Maybe we should put it in an different sysfs file. 

Please call the file something a little more obvious that 'r5c_state',
maybe 'journal_mode', and use it only to switch between write-through
and write-back.
If there is no journal, then either remove the file, or have writes fail
and reads return something obvious ... maybe ENODEV.

Thanks,
Song