On Wed, Sep 18, 2013 at 10:40:13PM +0200, Jakub Hrozek wrote:
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.
I think we were both confused by the poor label naming. The label actually,
despite being named "done", made the function return ENOENT so the code
was correct. But even I was confused couple of hours after sending the
patch so the code had to be improved :)
So in the attached patch I renamed the label to notfound. I hope that's
OK if we don't use the commonly used "done" in this case. I think the most
important thing is that there is only a single label we jump to.
Alternatively, if you prefer strictly one exit point, I could convert
the function to only use goto done and set negcache based on errno value
(if ENOENT->ncache) but I didn't see that as necessary.