Re: [PATCH] nscd: Do not rebuild getaddrinfo (bug 30709)

[Siddhesh Poyarekar <siddhesh@gotplt.org>]

On 2023-08-02 10:29, Florian Weimer via Libc-alpha wrote:
> So do I.  The Git history isn't very illuminating.  The layering
> wouldn't work anyway because of the ai_flags processing.  I was
> wondering whether the __check_pf stuff was supposed to be overriding
> something, but I don't see this either (not even in the history).

Apparently there was a getaddrinfo stub in nscd/nscd.c to make sure that 
nscd never called getaddrinfo, potentially resulting in nscd talking to 
itself:

"""
/* This is an ugly hack which prevents getaddrinfo from being dragged
    into nscd.  There currently is no special getaddrinfo version for
    use in nscd.  In case it should be necessary such a version must be
    created and this dummy version should be removed.  */
extern void getaddrinfo (void) __attribute ((visibility ("hidden")));

void
getaddrinfo (void)
{
   abort ();
}
"""

which later got replaced by nscd/gai.c:

commit 3078cba2f7272dab753c45b733af2e693e86d83c
Author: Ulrich Drepper <drepper@redhat.com>
Date:   Mon Oct 4 16:35:54 2004 +0000

     Update.

             * nscd/connections.c (start_threads): Use sysconf in case
             _POSIX_CLOCK_SELECTION or _POSIX_MONOTONIC_CLOCK is not 
greater zero.

             * nscd/nscd.c: Remove getaddrinfo stub definition.

commit 91a3b2ad2e819a7fa14b8f9773856e4303a8c99a
Author: Ulrich Drepper <drepper@redhat.com>
Date:   Mon Oct 4 09:36:51 2004 +0000

     Update.

             * nscd/Makefile (nscd-modules): Add gai.
             * nscd/gai.c: New file.


but there's no indication as to why, because it doesn't look like 
getaddrinfo was actually needed at that time either.  In all I think 
this looks correct.  If we want nscd to be safe from the possibility 
that it ends up talking to itself accidentally at some point, we could 
add back the aborting stub.

Reviewed-by: Siddhesh Poyarekar <siddhesh@sourceware.org>