Monday, March 7, 2011

Debian lenny ipv4 socket bind failing

A certain daemon that had been working elsewhere was refusing to bind to v4 on my box. Strace'ing it showed this
connect(5, {sa_family=AF_UNSPEC, sa_data="\0\0\0\0\0\0\0\0\0\0\0\0\0\0"}, 16) = 0
connect(5, {sa_family=AF_INET, sin_port=htons(10042), sin_addr=inet_addr("")}, 16) = -1 EAFNOSUPPORT (Address family not supported by protocol)
close(5)                                = 0
connect(5, {sa_family=AF_INET, sin_port=htons(10042), sin_addr=inet_addr("")}, 16) = 0
getsockname(5, {sa_family=AF_INET, sin_port=htons(60862), sin_addr=inet_addr("")}, [16]) = 0
close(5)                                = 0
setsockopt(5, SOL_SOCKET, SO_REUSEADDR, [1], 4) = 0
bind(5, {sa_family=AF_INET6, sin6_port=htons(10042), inet_pton(AF_INET6, "::", &sin6_addr), sin6_flowinfo=0, sin6_scope_id=0}, 28) = 0
listen(5, 5)                            = 0
write(2, "Config port: 10042\n", 19Config port: 10042
Notice the kernel says IPV4 binding is not supported! I am dual stacked! I had IPV4, I could ping the box and other hosts over IPV4. A quick look at my sysctl.d showed

net.ipv6.bindv6only = 1
Switching this off (0) fixed the issue. You can also do it interactively by
echo 0 > /proc/sys/net/ipv6/bindv6only 
Now everything works as expected.
This is weird. According to the kernel documentation

bindv6only - BOOLEAN
        Default value for IPV6_V6ONLY socket option,
        which restricts use of the IPv6 socket to IPv6 communication
                TRUE: disable IPv4-mapped address feature
                FALSE: enable IPv4-mapped address feature

        Default: FALSE (as specified in RFC2553bis)
Which as far as I can tell shouldn't touch the IPV4 stack. I expected to find this sysctl check under net/ipv4 but grep can only find the checks in net/ipv6. Any ideas?

No comments:

Post a Comment