Re: [PATCH 1/4] netfs: when subreq is marked for retry, do not check if it faced an error

[David Howells <dhowells@redhat.com>]

Shyam Prasad N <nspmangalore@gmail.com> wrote:

> > I think I suggested moving the check for NETFS_SREQ_NEED_RETRY up in the
> > function - above any checks of subreq->error, but after the initial stat
> > counting.
> 
> So you want to do this check regardless of whether there's an error or not?

Yes.  There's likely only to be a need to retry if there's been an error
(though there is the possibility of the filesystem returning a short read,
say, and needing a retry to complete it).  In any case, we can make the
setting of this flag up to the filesystem: does it think that a retry is
warranted?

> In that case, I think I'll still need to check if there is an error to
> set NETFS_RREQ_PAUSE only on error, right?

It needs to be set if a retry is requested as the retry thread will wait for
outstanding ops to quiesce - and if more are still being generated, that's
potentially a problem.  And for AFS, for example, the need to do a retry might
mean that we need to switch server - so we probably do want to throw a hold on
further subrequest generation until we've probed the problem.

David