getaddrinfo() returns duplicate addresses under AI_ADDRCONFIG

[Matthias Andree]

Presume I have a computer that has no global IPv6 addresses, but a few IPv4 addresses.

Applications using getaddrinfo() with hints.ai_flags = AI_ADDRCONFIG set will properly omit IPv6 addresses, HOWEVER they will return duplicate IPv4 addresses.

Demonstration code attached.

Without IPv6 addresses:

# show setup
$ grep -w localhost /etc/hosts
127.0.0.1 localhost
::1 localhost ip6-localhost ip6-loopback
$ ip a s | grep inet
    inet 127.0.0.1/8 scope host lo
    inet6 ::1/128 scope host
    inet 192.168.0.4/24 brd 192.168.0.255 scope global eth0
    inet6 fe80::226:XXXX:XXXX:XXXX/64 scope link # masked for privacy

# compile test case
$ gcc -Wall -Wextra -pedantic -O -o libcbug-demo libcbug-demo.c

# run test case
$ ./libcbug-demo localhost echo
Address 1: 127.0.0.1
Address 2: 127.0.0.1

(127.0.0.1 should be listed only once, Solaris 10 for instance achieves that)

Now configure a fake IPv6 address (don't do this on hosts with real IPv6 connectivity, it suffices to use your real IPv6 address) and re-run the application:

$ ip a a 2001::2001 dev eth0 # set up IPv6
$ ./libcbug-demo localhost echo
Address 1: ::1
Address 2: 127.0.0.1
$ ip a d 2001::2001 dev eth0 # deconfigure bogus address

(This is fine)

Practical consequence: excessive connection retries by applications that possibly take a long time.

This isn't Ubuntu specific.

ProblemType: Bug
DistroRelease: Ubuntu 10.04
Package: libc6 2.11.1-0ubuntu7
ProcVersionSignature: Ubuntu 2.6.32-22.33-generic 2.6.32.11+drm33.2
Uname: Linux 2.6.32-22-generic x86_64
NonfreeKernelModules: fglrx
Architecture: amd64
Date: Wed May 19 01:13:51 2010
ProcEnviron:
 PATH=(custom, no user)
 LANG=de_DE.UTF-8
 SHELL=/bin/bash
SourcePackage: eglibc

[Psimerda (psimerda)]

getaddrinfo() with AI_ADDRCONFIG replaces all ::1 with 127.0.0.1 on hosts without non-loopback IPv6 addresses. See the following test result:

1: lo: <LOOPBACK,UP,LOWER_UP> mtu 16436 qdisc noqueue state UNKNOWN
    link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00
    inet 127.0.0.1/8 scope host lo
    inet6 ::1/128 scope host
       valid_lft forever preferred_lft forever
2: eth0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc pfifo_fast state UP qlen 1000
    link/ether f0:de:f1:b8:6a:ee brd ff:ff:ff:ff:ff:ff
    inet 84.246.161.88/28 brd 84.246.161.95 scope global eth0

getaddrinfo host="None" hints.ai_flags=AI_ADDRCONFIG:
  ::1
  127.0.0.1
getaddrinfo host="localhost" hints.ai_flags=AI_ADDRCONFIG:
  127.0.0.1
  127.0.0.1
getaddrinfo host="127.0.0.1" hints.ai_flags=AI_ADDRCONFIG:
  127.0.0.1
getaddrinfo host="localhost4" hints.ai_flags=AI_ADDRCONFIG:
  127.0.0.1
getaddrinfo host="::1" hints.ai_flags=AI_ADDRCONFIG:
  ::1
getaddrinfo host="localhost6" hints.ai_flags=AI_ADDRCONFIG:
  127.0.0.1
getaddrinfo host="195.47.235.3" hints.ai_flags=AI_ADDRCONFIG:
  195.47.235.3
getaddrinfo host="2a02:38::1001" hints.ai_flags=AI_ADDRCONFIG:
  2a02:38::1001
getaddrinfo host="info.nix.cz" hints.ai_flags=AI_ADDRCONFIG:
  195.47.235.3
getaddrinfo host="www.google.com" hints.ai_flags=AI_ADDRCONFIG:
  173.194.39.81
  173.194.39.82
  173.194.39.80
  173.194.39.84
  173.194.39.83

Especially notable are:

getaddrinfo(localhost) -> 127.0.0.1, 127.0.0.1 (duplicate)
getaddrinfo(localhost6) -> 127.0.0.1 (wrong)

Contents of /etc/hosts:

127.0.0.1 localhost localhost4
::1 localhost localhost6

Looks like some sort of black magic someone was trying to solve something.

[Psimerda (psimerda)]

Testing script:

#!/usr/bin/python3
import sys
from socket import *
hosts = [
    None,
    "localhost",
    "127.0.0.1",
    "localhost4",
    "::1",
    "localhost6",
    "195.47.235.3",
    "2a02:38::1001",
    "info.nix.cz",
    "www.google.com",
]
for host in hosts:
    print("getaddrinfo host=\"{}\" hints.ai_flags=AI_ADDRCONFIG:".format(host))
    try:
        for item in getaddrinfo(host, "http", AF_UNSPEC, SOCK_STREAM, SOL_TCP, AI_ADDRCONFIG):
            print(" {}".format(item[4][0]))
    except gaierror as error:
     print(" !! {} !!".format(error))

[dino99 (9d9)]

This is no more a supported version now

[Matthias Andree (matthias-andree)]

That does not invalidate the bug. Either confirm the minimum required package version for the fix, or bump it to the current eglibc version.

[dino99 (9d9)]

Lucid is no more maintained, and your issue is not met then. So closing that useless report now.

[dino99 (9d9)]

need to understand why Lucid will stop getting support

[Matthias Andree (matthias-andree)]

Reopen. Bug persists in trusty.

[Matthias Andree (matthias-andree)]

Bug persists in trusty.

[dino99 (9d9)]

If that issue is not due to the coding style, then push it to bugzilla where it will get a chance to get devs attention and possibly a detailed howto.

https://sourceware.org/bugzilla/buglist.cgi?quicksearch=getaddrinfo
https://sourceware.org/bugzilla/show_bug.cgi?id=14969 (possibly yours)

[dino99 (9d9)]

Modif already requested https://sourceware.org/bugzilla/show_bug.cgi?id=14965

[dino99 (9d9)]

Some more related subjects https://sourceware.org/bugzilla/buglist.cgi?quicksearch=ai_addrconfig

[Björn Lindqvist (bjourne)]

The bug is present still in Ubuntu Trusty.

[Alex Grönholm (alex-gronholm)]

This is what the attached script gives me now (Fedora 32, Glibc 2.31):

getaddrinfo host="None" hints.ai_flags=AI_ADDRCONFIG:
  ::1
  127.0.0.1
getaddrinfo host="localhost" hints.ai_flags=AI_ADDRCONFIG:
  ::1
  127.0.0.1
getaddrinfo host="127.0.0.1" hints.ai_flags=AI_ADDRCONFIG:
  127.0.0.1
getaddrinfo host="localhost4" hints.ai_flags=AI_ADDRCONFIG:
  127.0.0.1
getaddrinfo host="::1" hints.ai_flags=AI_ADDRCONFIG:
  ::1
getaddrinfo host="localhost6" hints.ai_flags=AI_ADDRCONFIG:
  ::1
getaddrinfo host="195.47.235.3" hints.ai_flags=AI_ADDRCONFIG:
  195.47.235.3
getaddrinfo host="2a02:38::1001" hints.ai_flags=AI_ADDRCONFIG:
  2a02:38::1001
getaddrinfo host="info.nix.cz" hints.ai_flags=AI_ADDRCONFIG:
  2a02:38::1001
  195.47.235.3
getaddrinfo host="www.google.com" hints.ai_flags=AI_ADDRCONFIG:
  2a00:1450:400f:809::2004
  172.217.21.132

So the output is fine now.
HOWEVER, if I query with family=AF_INET, I still get two results:
[(<AddressFamily.AF_INET: 2>, <SocketKind.SOCK_STREAM: 1>, 6, '', ('127.0.0.1', 0)),
 (<AddressFamily.AF_INET: 2>, <SocketKind.SOCK_STREAM: 1>, 6, '', ('127.0.0.1', 0))]