Subject: Re: Next try: sysklogd 1.4
From: Solar Designer (solar@false.com)
Date: Tue Sep 19 2000 - 00:36:10 CEST
> We've used this patch for about a year now in OpenLinux, and had
> no problems. I think most vendors implemented a patch like this
> when the issue was first brought up.
There're problems. Most vendors still ship compatibility libraries
(libc 5) that use stream sockets.
I think it was a bit too early to drop stream sockets support from
the official sysklogd completely. This means I will now be porting
the fixes to sysklogd-1.3 for my ~15 libc 5 systems. It would have
been nice to have a compile-time option to choose the socket type.
No, I'm not suggesting to add the support back and risk introducing
even more bugs into this overly-complicated code, but the problems
should be documented (at least mention that sysklogd 1.4+ is not to
be used on libc 5 systems).
Also, the now released(?) sysklogd-1.4 still tries to use a stream
socket in its own syslog.c (why does it need one?), which is used by
klogd. I suspect that klogd is once again broken by this change.
This is from Red Hat's sysklogd.spec:
* Mon Nov 15 1999 Bill Nottingham <notting@redhat.com>
- fix ECONNRESETs from security patch (olaf)
* Fri Sep 25 1999 Bill Nottingham <notting@redhat.com>
- eek. The security patch broke *two* things...
* Wed Sep 22 1999 Bill Nottingham <notting@redhat.com>
- make klogd actually work.
[...]
* Tue Sep 07 1999 Cristian Gafton <gafton@redhat.com>
- add patch to fix a possible DoS (thanks Olaf Kirch)
Signed,
Solar Designer
This archive was generated by hypermail 2b25 : Tue Sep 19 2000 - 00:38:40 CEST