Cannot prevent rpcbind from listening on all interfaces for TCP connections

Status changed to 'Confirmed' because the bug affects multiple users.

There is no way to prevent rpcbind from listening on all interfaces for TCP connections. The setting "-h" only applies to UDP connections.

Without the attached patch, starting rpcbind with -h makes no effect to listen on a specific address:

$ rpcbind -h 192.168.100.2

tcp 0 0 0.0.0.0:111 0.0.0.0:* LISTEN 22011/rpcbind
tcp6 0 0 :::111 :::* LISTEN 22011/rpcbind
udp 0 0 127.0.0.1:111 0.0.0.0:* 22011/rpcbind
udp 0 0 0.0.0.0:986 0.0.0.0:* 22011/rpcbind
udp6 0 0 ::1:111 :::* 22011/rpcbind
udp6 0 0 :::986 :::* 22011/rpcbind

With this patch, starting rpcbind with -h makes effect to listen to a specific address

$ rpcbind -h 192.168.100.2

tcp 0 0 192.168.100.2:111 0.0.0.0:* LISTEN 24826/rpcbind
tcp 0 0 127.0.0.1:111 0.0.0.0:* LISTEN 24826/rpcbind
tcp6 0 0 ::1:111 :::* LISTEN 24826/rpcbind

Thanks for the patch! Just two questions:

1. why does rpcbind not listen to UDP any more?
2. have you noticed that a similar problem exists in rpc.statd?

Hello Kurt,

1) I am not seeing the reported issue, it still works on UDP/TCP properly.
2) rpc.statd is part of another package nfs-common, I can review that specifically.

16:36 <rbasak> niedbalski: I'm not sure that bug 1430181 is appropriate to fix in an SRU or during feature freeze in Vivid (without an exception).

16:36 <rbasak> niedbalski: seems to me that TCP binding is a new feature because the switch is documented to support UDP only.

16:37 <rbasak> niedbalski: the patch looks pretty extensive too.

So I think this needs to wait for Vivid+1? We should have the time, in that case, to get this patch upstreamed, instead of needing a delta in Ubuntu specifically. Or if this isn't possible, can the reason be explained in this bug, please?

unsuscribing sponsors, it seems it's not to upload for vivid and already being handled (it has a merge request as well)

I looked at this as well, it looks like it probably would benefit from preparing the 0.2.3 release instead of patching things. I haven't looked very closely, but it seems as though the patches may already be included in uptream git (and thus in 0.2.3, which seems to be the latest revision on git://linux-nfs.org/~steved/rpcbind, which is the correct upstream branch if I'm to believe the sourceforge link set as homepage in the package). If they aren't, then they probably wouldn't apply cleanly anyway.

I would suggest discussing the update with Debian; from there we could merge the package to the latest version.

Marking Won't Fix for the SRU tasks as the current position is that we have no intention of SRUing this.

Laurent,

Thank you for adding the Debian bug watch. I think the Debian bug refers to UDP only though, and the complaint in this bug is for TCP support, so I don't think the two bugs track the same issue. Consequently I'm deleting the bug watch. If this is wrong, please comment.