On Mon, 2023-07-17 at 09:05 +0200, Hauke Fath wrote: > On Sun, 16 Jul 2023 20:14:20 +0200, Ben Hutchings wrote: > > > > > > nfsv3 over tcp works, but is subobtimal , as described - when the router > > > goes down, the tcp mounts will hang, and the machine will have to be > > > rebooted. > > [...] > > > > Does this mean you are using the "soft" mount option? Without that, I > > would expect access to the mount to hang until the network connection > > is restored, regardless of whether the TCP or UDP transport is used. > > No, we use hard mounts. > > But the router's package filter will have lost state after a reboot, > and reject packets from tcp connections that the clients assume to > exist. This is not a problem with udp, because connection-less.
Ah, I see. You didn't mention that there was dynamic NAT involved before. If an NFS server is rebooted abruptly (so it doesn't properly close TCP connections), once it's back up it will respond to any requests from clients with a TCP RST, and they should reconnect. If a NAT router between client and server is rebooted, I think that something similar should happen, but the router would need to send the TCP RST instead. Is your router configured to send a TCP RST when receiving a packet for an unknown connection, or does it just drop those packets? (In iptables this is the difference between REJECT and DROP policies.) Ben. -- Ben Hutchings Never attribute to conspiracy what can adequately be explained by stupidity.
signature.asc
Description: This is a digitally signed message part
