[OpenAFS-devel] 1.6 ihandle issues (was Re: Crazy DAFS problem (with log))

[Andrew Deason]

On Tue, 22 Mar 2011 17:22:26 -0400
Jeffrey Altman <jaltman@secure-endpoints.com> wrote:

> Thanks for taking the time to dive into this.  I agree that at this
> point the discussion really needs to be taken to openafs-devel in
> order to ensure that the right audience is paying attention.  This
> e-mail will hopefully take care of that.

Okay, thanks.

> On 3/22/2011 4:13 PM, Andrew Deason wrote:
>
> > A conservative fix for this I think would be to add some field to
> > FdHandle_t to indicate that a reallyclose is needed, so a subsequent
> > call to FDH_CLOSE causes an effective FDH_REALLYCLOSE. (Like how
> > ihandle_t has the IH_REALLY_CLOSED flag.) A "better" fix would be to
> > add a new fd_status type, but I'd be worried for stable 1.6.x about
> > screwing up the other fd_status checks.

This is gerrit 4274, which also happens to fix the issue reported by
Ryan. However, I'm not even really sure if this is desirable for a late
1.6.0 submission, since I don't know if it fixes any noticeable problem.
It fixes Ryan's issue, but that's only because of the ihandle leak issue
fixed in 4272.

In general, the difference between an FDH_CLOSE and FDH_REALLYCLOSE is
just performance related, due to how the fd cache is handled, isn't it?
For the cases where we actually need to close a logical "file", you need
to IH_REALLYCLOSE anyway. So, I'm not sure if FDH_REALLYCLOSE not really
closing the file is likely to be a real issue (unless something else
helps cause it, like the ihandle leak).

-- 
Andrew Deason
adeason@sinenomine.net