No subject
Sat Mar 5 15:32:34 UTC 2022
cloned bio and the tio structure both get freed, so there's no reason to
restore the bio after calling target->endio().
> This is an annoying overhead to have for every bio, so I
> suggest we change the semantics of target->endio and say that it may
> only trigger an io retry if there was an error. Agree ?
I do agree that my patch adds potentially worthless overhead to every cloned
bio that DM creates. The alternative is to do this only for multipath, and
have it create its own per-bio private data (which it can store in tio->info)
which it can restore itself if it needs to redrive the I/O. If you like this
idea better, I can send in a different patch later today.
But in any case, we need some fix for this, because the current code is
sending down invalid bios when a path fails. :)
--
Kevin Corry
kevcorry at us.ibm.com
http://evms.sourceforge.net/
More information about the dm-devel
mailing list