Re: [PATCH] autofs: handle old configuration format

[Ian Kent <ikent@redhat.com>]

On Thu, 2015-06-18 at 09:41 -0700, Josef Bacik wrote:
> On 06/17/2015 11:37 PM, Ian Kent wrote:
> > On Wed, 2015-06-17 at 20:52 -0700, Josef Bacik wrote:
> >> On 06/17/2015 05:39 PM, Ian Kent wrote:
> >>> Hi Josef,
> >>>
> >>> Hope you are well.
> >>>
> >>> On Wed, 2015-06-17 at 09:29 -0700, Josef Bacik wrote:
> >>>> This is a patch to fix
> >>>>
> >>>> https://urldefense.proofpoint.com/v1/url?u=http://bugzilla.centos.org/view.php?id%3D8614&k=ZVNjlDMF0FElm4dQtryO4A%3D%3D%0A&r=cKCbChRKsMpTX8ybrSkonQ%3D%3D%0A&m=2pMrg2mSVkYIRbq9PsXATsOwwHLUv7M%2FC1vSREeuAMA%3D%0A&s=bce3fa3f755b02b8849de7338fa74adb71fd4cd84a2a318820c29830ed16f699
> >>>>
> >>>> The configuration stuff was redone and it broke backwards compatiblity.
> >>>
> >>> OK, but I put quite a bit of effort in to not do that.
> >>>
> >>> I thought it more likely I'd get complaints from people not realizing an
> >>> old configuration left in place would override changes in the new
> >>> configuration.
> >>>
> >>> Can you give me an example of the problem you have seen please.
> >>>
> >>>> Unfortunately this was backported to Centos/RHEL, so configurations that were
> >>>> working in 6.4 suddenly broke in 6.6, which is not helpful.  Fix this by
> >>>> noticing if we are looking at the old configuration file, strip out the DEFAULT_
> >>>> bit of the variable if it is there and tolower the rest of the string.  This
> >>>> makes it so our old configuration works properly and now people get their home
> >>>> dirs automounted properly with either versions of autofs.  Thanks,
> >>>
> >>> The way this is supposed to work is that the new configuration file is
> >>> read and then the old one is read and anything in the old configuration
> >>> file should override anything in the new one.
> >>>
> >>> When looking up configuration values if the key isn't found and the key
> >>> starts with "DEFAULT_" the the key is looked for again without the
> >>> "DEFAULT_". Also, key comparisons are case insensitive so that shouldn't
> >>> make a difference.
> >>
> >> Oh now I realize why it's not working.  When you do defaults_get_schema
> >> you are looking up the new values, so your check in conf_lookup isn't
> >> helping anything, because we aren't looking up DEFAULT_*, we're looking
> >> up the new values.  And since we don't have those variables set in
> >> autofs.conf they don't get set until we load the old things, and so we
> >> have DEFAULT_* set in the conf but not the normal variables.  So we can
> >> do what I've done here, or change conf_lookup to add the DEFAULT_ to the
> >> front of anything we search for if we can't find it.  Thanks,
> >
> > Yes, that looks to be the case.
> >
> > I can't just strip the DEFAULT_ prefix on adding values because there
> > could be configuration key names that are meant to have the prefix.
> >
> > So I think it's best to try and handle this in the conf_lookup()
> > function.
> >
> > You may get some conflicts with this as it's against current upstream
> > but if you can resolve them, does this patch resolve the problem?
> >
> > If you have difficulty with conflicts then give me the package revision
> > and I'll back port the patch to it so you can test it.
> >
> > autofs-5.1.1 - fix config old name lookup
> >
> 
> That did it, thanks!

I'm still thinking about this so there might be more changes, not sure
though.

I'll let you know if I make more changes and will commit what I have
next time I push changes to the upstream repo (and post the patches at
the usual autofs location on kernel.org).

Ian