Tip of the day: if Samba 3.4 fails to work…

I fought with this today… if you are running Gentoo ~arch you probably noticed that the current Samba support is “definitely suboptimal” (to use the words of the current maintainer) and indeed it failed to work on me once again (third time; the first was a broken init script; the second was missing USE deps so I was quite upset). If you find yourself unable to log-in Samba, you need to consider two possible problems.

First problem: the Samba password system seems to have either changed or moved so you have to re-apply the password to your user (and re-add the user as well!). To do so you have to use the smbpasswd command. Unfortunately this will fail to work when the system has IPv6 configured. And here comes the next problem.

Samba is likely having trouble upstream to deal with IPv6; indeed it comes down to having the smbpasswd command trying to connect to 127.0.0.1 (IPv4), but the smbd daemon is only opening :: (IPv6), so it’ll fail to connect and won’t let you set your password. To fix this, you have to change the /etc/samba/smb.conf file, and make sure that the old IPv4 addresses are listened to explicitly. If you got static IPs this is pretty simple, but if you don’t, you’ll have a little more complex situation and you’ll be forced to restart samba each time the network interface changes IP, I’m afraid (I haven’t been able to test that yet).

[global]
interfaces = 127.0.0.1 wlan0 br0
bind interfaces only = yes

As you can see we’re asking for some explicit interfaces (and the localhost address) to be used for listening; since samba uses the IPv4 localhost address for the admin utilities you explicit that to make sure it listens to that. For some reason I cannot understand, when doing this explicitly, samba knows to open different sockets for both IPv4 and IPv6, otherwise it’ll open it for IPv6 only.

I’m not even going to fight with upstream about this, I’m tired and I’m tracking down a bug in Gtk#; a nasty one that crashes the app when using custom cell renderers, and I already fixed iSCSI Target for kernel 2.6.32 (as well as version-bumped it).