[libvirt] [PATCH]: don't harcode buffer for getgrnam_r

[Guido Günther]

On Thu, Apr 16, 2009 at 11:37:24PM +0900, Ryota Ozaki wrote:
> Hi,
> 
> On Thu, Apr 16, 2009 at 10:40 PM, Daniel P. Berrange
> <berrange at redhat.com> wrote:
> > On Thu, Apr 16, 2009 at 02:14:13PM +0200, Guido G?nther wrote:
> >> On Thu, Apr 16, 2009 at 09:56:27AM +0200, Daniel Veillard wrote:
> >> > On Thu, Apr 16, 2009 at 09:19:38AM +0200, Guido Günther wrote:
> >> > > Hi,
> >> > > determines the maximum needed buffersize for getgrnam_r using sysconf
> >> > > instead of hardcoding it to 1024 and increases the buffer on ERANGE.
> >> > > The latter is needed since sysconf is allowed to return -1. Furthermore
> >> > > some glibc versions seem to return a too small buffer on amd64
> >> > > (http://bugs.debian.org/520744). O.k. to apply?
> >> >
> >> >   It looks a bit weird, using sysconf but 1/ allowing it to fail so
> >> > doing the 2/ 1024 value and loop on ERANGE , but well if I understand
> >> > correctly taht's forced by some glibc broken behaviour.
> >> Yes, sysconf is allowed to return -1 here.
> >>
> >> >   My take is that the *= 2 size loop should be bounded to avoid eating
> >> > all memory on some intermediate not size related error. Can we really
> >> glibc shouldn't return ERANGE then, but better safe than sorry. I've
> >> added that check in the attched patch.
> >
> > ACK.
> 
> I think it's more useful that such a function is implemented as a wrapper
> function like virGetUserDirectory() in util.c, because other drivers might
> use the function. Actually my patch sent just now implements
> virGetGIDByGroupname()
> in util.c ;-)
Yes, it's more useful to have this in util.c when we have more than once
caller (which we hadn't so far). I'll move remoteReadConfigFile over to
virGetGIDByGroupname once it's in.
 -- Guido