On Tue, Sep 7, 2010 at 21:50:14 +0200, Arthur de Jong wrote: > On Mon, 2010-09-06 at 23:37 +0200, Julien Cristau wrote: > > On Mon, Sep 6, 2010 at 18:13:59 +0200, Arthur de Jong wrote: > > > Please unblock package cvsd. The 1.0.20 contains a fix for mixing IPv4 > > > and IPv6 sockets regardless of the order of addresses returned by > > > getaddrinfo() (which changed between lenny and squeeze) and the > > > net.ipv6.bindv6only sysctl. > > > > Ignoring failure from bind(2) seems like a pretty bad idea. > > The daemon does a call to getaddrinfo() to figure out which addresses it > should listen on and tries each one. Currently at least one bind() of > the returned addresses should succeed, other failures are ignored. > I can't see any reason for bind() failure to not be fatal... > Ignoring bind() failures used to be necessary when not using > IPV6_V6ONLY. Without it both IPv6 and IPv4 addresses were returned and > the bind on the IPv4 address would always fail if the IPv6 one > succeeded. cvsd 1.0.20 only changes which failures are ignored. > They should both succeed if done in the right order, iirc (meaning the order they're returned from getaddrinfo() with AI_PASSIVE). And in any case they should both succeed with IPV6_V6ONLY set. Cheers, Julien
Attachment:
signature.asc
Description: Digital signature