Bug 395

Summary:	resolver updates causes resolver to stop resolving some hosts
Product:	glibc	Reporter:	Mike Frysinger <vapier>
Component:	libc	Assignee:	GOTO Masanori <gotom>
Status:	RESOLVED WORKSFORME
Severity:	normal	CC:	glibc-bugs, lv
Priority:	P2
Version:	unspecified
Target Milestone:	---
Host:	i686-pc-linux-gnu	Target:
Build:		Last reconfirmed:

Description Mike Frysinger 2004-09-21 04:38:09 UTC

original report here:
http://sources.redhat.com/ml/libc-alpha/2004-09/msg00146.html

basically somewhere between Aug 08 2004 and Sep 18 2004, the resolver library's
behavior broke when dealing with 'domain' entries in resolv.conf

when looking up a hostname, the old behavior was to try to search for the
hostname in each of the domains found in resolv.conf ... if those lookups
failed, the hostname itself would be tried
the new behavior is to only search in the domains found in resolv.conf

for example, i have the host 'aurora' setup to resolve to '192.168.0.7' on my
dns server ... on a client machine, i would look up 'aurora' via gethostbyname()
... in the resolv.conf on the client machine, i have one domain entry, 'domain
wh0rd.org' ...

the old glibc would query 'aurora.wh0rd.org' and fail, and then fall back and
try to look up 'aurora' which would work
the new glibc only tries 'aurora.wh0rd.org' and then fails

sniffing dns traffic via ethereal backed up this analysis

Comment 1 Mike Frysinger 2004-09-21 04:40:23 UTC

i neglected to mention that if i remove the 'domain wh0rd.org' from the
resolv.conf on my client, the new glibc will attempt to lookup just 'aurora' and
thus work 'properly'

Comment 2 Ulrich Drepper 2004-09-27 04:22:44 UTC

I know which change this was, revision 1.20 of resolv/res_query.c (minus the
accidentally left in debug stuff).  But I maintain that this is the correct
change and that it worked for you by accident.

Try adding

  options ndots:0

to your resolv.conf.

Comment 3 Mike Frysinger 2004-09-28 03:56:19 UTC

'options ndots:0' restored the previous behavior like i described and like you said

i'm not sure if there are any specs which cover the 'proper' behavior with
respect to resolv.conf/domain/etc... (SuSv3 and such) but like you say, when
using the 'domain' option in resolv.conf, this seems like 'proper' behavior,
even it does take 'more' effort to get back to the older/simpler behavior

perhaps we can get the man page to make a note of the 'expected' behavior under
the ndots section so that people (such as myself) wont file anymore 'invalid'
bugs unknowingly ?

Comment 4 Ulrich Drepper 2005-09-23 16:36:25 UTC

Man pages are not part of glibc.  Bring this up with the appropriate person.