针对IPv4和IPv6地址的connect（）行为

Question

I create a socket with following params:

hints.ai_family = AF_UNSPEC;
hints.ai_socktype = SOCK_STREAM;
getaddrinfo(serverName, port, &hints, &res);
sFd = socket (res_node->ai_family, SOCK_STREAM, 0);

I then make connect calls for IPv4 and Ipv6 server addresses which may or may not be present. When the server is not reachable, it has different behavior in case of IPv4 and IPv6 addresses.

In case of v4 it gets stuck for a very long time (some internal default timeout) before returning error. However, in case of IPv6 the behavior is changed. For some addresses the call returns error immediately eg address like 1111::22, whereas for some it takes longer eg fec0:60:69bc:94:211:25ff:fec4:6 but still lot lesser than IPv4 timeout.

Can anyone explain the difference in behavior?

Answer 1

When the route lookup to the particular target host (either protocol) yields unreachable/prohibited/reject/etc., connect() can immediately fail. Otherwise it has to send probes (ARP/NDISC, and potentially L4 packets like TCP SYN/SCTP INIT).

Route lookup example of a system without a route to global IPv6 unicast:

$ ip r g 2a01::1
unreachable 2a01::1 from :: dev lo  table unspec  proto kernel  src fe80::224:beff:fec2:7f16  metric -1  error -101 hoplimit 255

Note the difference to an IPv6-connected host:

$ ip r g 2a01::1
2a01::1 from :: via fe80::2d0:3ff:fe93:d123 dev eth0  src 2001:527:500:770f:19e:84ff:fe9e:878  metric 0