Bug#632917: libc6: getaddrinfo fails prematurely; gethostbyname does not

[Matt Mackall <mpm@selenic.com>]

Package: libc6
Version: 2.13-4
Severity: normal


If there are multiple nameservers in /etc/resolv.conf and the first of
them refuses DNS queries for some domains, getaddrinfo will fail while
gethostbyname succeeds.

This can result in VERY mysterious that-shouldn't-happen problems: 

- host works
- ping works
- browser works
- ssh and various other "properly written" apps fail

Example:

$ cat /etc/resolv.conf
# Generated by NetworkManager
nameserver 10.8.0.1    <- my VPN DNS server (any DNS server without recursion)
nameserver 192.168.43.1   <- my phone company's server

$ host bitbucket.org
bitbucket.org has address 207.223.240.182
bitbucket.org has address 207.223.240.181
bitbucket.org mail is handled by 5 alt1.aspmx.l.google.com.
bitbucket.org mail is handled by 5 alt2.aspmx.l.google.com.
bitbucket.org mail is handled by 10 aspmx2.googlemail.com.
bitbucket.org mail is handled by 10 aspmx3.googlemail.com.
bitbucket.org mail is handled by 1 aspmx.l.google.com.

$ host bitbucket.org 10.8.0.1
Using domain server:
Name: 10.8.0.1
Address: 10.8.0.1#53
Aliases: 

Host bitbucket.org not found: 5(REFUSED)

$ ping -c 1 bitbucket.org
PING bitbucket.org (207.223.240.182) 56(84) bytes of data.
64 bytes from 207.223.240.182: icmp_req=1 ttl=50 time=316 ms

--- bitbucket.org ping statistics ---
1 packets transmitted, 1 received, 0% packet loss, time 0ms
rtt min/avg/max/mdev = 316.571/316.571/316.571/0.000 ms

$ ssh bitbucket.org
ssh: Could not resolve hostname bitbucket.org: Name or service not known

$ python
>>> import socket
>>> socket.gethostbyname("bitbucket.org")
'207.223.240.181'
>>> socket.getaddrinfo("bitbucket.org", 80, 10)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
socket.gaierror: [Errno -5] No address associated with hostname

I've used ltrace to verify that Python is calling getaddrinfo directly
and that it fails. Swapping the order of DNS servers makes the above
test succeed.

Desired behavior: getaddrinfo() should have EXACTLY the same
robust DNS resolution algorithm as gethostbyname() to avoid mysterious,
hard-to-diagnose application-dependent failure modes.


-- System Information:
Debian Release: wheezy/sid
  APT prefers unstable
  APT policy: (500, 'unstable')
Architecture: amd64 (x86_64)

Kernel: Linux 2.6.39+ (SMP w/4 CPU cores)
Locale: LANG=C, LC_CTYPE=en_US.UTF-8 (charmap=UTF-8)
Shell: /bin/sh linked to /bin/bash

Versions of packages libc6 depends on:
ii  libc-bin                      2.13-4     Embedded GNU C Library: Binaries
ii  libgcc1                       1:4.6.0-10 GCC support library

libc6 recommends no packages.

Versions of packages libc6 suggests:
ii  debconf [debconf-2.0]         1.5.39     Debian configuration management sy
pn  glibc-doc                     <none>     (no description available)
ii  locales                       2.13-4     Embedded GNU C Library: National L

-- debconf information excluded

-- 
Mathematics is the supreme nostalgia of our time.