Re: Question on /etc/nsswitch.conf

[Zack Weinberg <zackw at panix dot com>]

On Thu, Sep 28, 2017 at 9:29 AM, Florian Weimer <fweimer@redhat.com> wrote:
> On 09/28/2017 02:28 PM, Yury Norov wrote:
>>
>> On Thu, Sep 28, 2017 at 01:44:49PM +0200, Florian Weimer wrote:
>>>
>>> On 09/28/2017 01:28 PM, Yury Norov wrote:
>>>>
>>>> Yes, it is, thanks. It means that getpwuid(), getpwnam() etc are
>>>> broken because with NUll in return value and untouched errno they
>>>> report that no user record found in passwd. Whilst the record exists.
>>>
>>> Is this because your new glibc doesn't have the compat module?
>>
>> Do you mean libnss_compat? There's no this library in Linaro toolchain.
>> AFAIR, it was deprecated and there were already issues with it.
>> https://sourceware.org/ml/libc-alpha/2017-08/msg00531.html
>
> I believe that it might still be possible that compat without NIS is useful,
> and that people use it.

I expect there are, at least, a lot of existing installations with
"passwd: compat" in their nsswitch.conf, but probably most of them
don't use NIS.

I was under the impression that the only thing compat does is look for
the +:::::: notation in /etc/passwd and forward to NIS.  If so, or
even if it's more complicated than that, maybe the Right Thing is to
build a libnss_compat.so even when !BUILD_OBSOLETE_NSL, but have it
error out on +:::::: rather than forwarding to NIS.  The big problem I
see with this idea is it might make it harder for TIRPC to drop in a
replacement.  Thoughts?

zw