On Wed, Sep 18, 2013 at 01:41:36PM -0400, Simo Sorce wrote:
On Wed, 2013-09-18 at 17:58 +0200, Jakub Hrozek wrote:
> Hi,
>
> the first patch fixes the bug that Jean-Baptiste Denis found earlier
> today. In case the lookup was not done using a FQDN, the code would skip
> setting the entries to the ncache.
>
> The second patch is an incremental improvement. I don't think we should
> abort the whole lookup if setting an entry in negcache would fail. The
> negative cache is a performance optimization after all.
It seem to me that with the first patch you are changing behavior as you
leave 'ret' unchanged to whatever error is returned instead of setting
it to ENOENT before going to 'done'.
Simo.
Ugh, that is a bug.
I was going back and forth on changing the particular return to goto
and when I made my mind, I forgot to set the errno.
Thanks Simo for catching it, I'll prepare a new version.