rport forwarding - unstable about sshj

[pottedplant]

to clarify the severity - rport forwarding is useless at the moment. at least under load (concurrent forwarded connections) the connection dies repeatably!

[pottedplant]

after some debugging i found at least one condition producing the error.

applying this (short) patch issues warnings and ignores some write requests to closed channels.
not intended for production use, just to point devs into the right direction

https://gist.github.com/1732392

for now it seems to work around the issue

[shikhar]

@pottedplant I couldn't reproduce with short-lived connections, will setup some longer ones and try

[shikhar]

thanks for the patch btw, it does point me in the right direction

[pottedplant]

to trigger the failure or warning (patched) i did setup a port forwarding (not in debug mode, debugger seems to minimize the race condition) against for example a local web server (remote server public port -> ssh -> local web server) and executed something like this:

for i in seq 100; do (curl "http://rport.server:1337/" > /dev/null &); done

this triggers the error almost always

[shikhar]

I will look into the issue this week

[sirthias]

+1 for getting remote port forwarding fixed
(Just bumped into this from scala-ssh)

[sirthias]

Actually I am currently having another issue with remote port forwarding. I can't even get the most basic setup up since the client.getRemotePortForwarder.bind call always triggers this exception:

java.lang.NullPointerException
at net.schmizz.sshj.transport.TransportImpl.write(TransportImpl.java:424)
at net.schmizz.sshj.connection.ConnectionImpl.sendGlobalRequest(ConnectionImpl.java:191)
at net.schmizz.sshj.connection.channel.forwarded.RemotePortForwarder.req(RemotePortForwarder.java:203)
at net.schmizz.sshj.connection.channel.forwarded.RemotePortForwarder.bind(RemotePortForwarder.java:170)

Is there an obvious reason as to what I might be doing wrong?

[shikhar]

@sirthias I think your issue is because you are not yet connected on the SSHClient. That said you might run into the original issue, I'll look into that one as soon as I get a chance.

[sirthias]

Ok, might be. I'll look into it.
Still, I think an NPE is always a bug.
Ideally the API shouldn't let me do things before the connection has been established.
At least it should throw a proper exception explaining the issue.
What would be the best way for me to wait for the connection to be completed before doing anything else?

[shikhar]

I agree an NPE is always a bug. I'll add exceptions for these cases where it's required that connection have been established. I am also looking at refactoring the SSHClient API.

sshClient.connect() is synchronous so as long as you are using sshClient.getRemotePortForwarder.bind after that it should be fine.

[sirthias]

Thanks, Shikhar, I was indeed calling bind before connect.
I think for the time being it'd be enough to simply render a proper error message if the user doesn't abide by the implicit API rules of calling connect before anything else.

Fixing the remote port forwarding is probably the more pressing issue.

Thanks for making and maintaining sshj!

[shikhar]

the 'Received data for nonexistent channel' bug should be fixed in 0.9.0-SNAPSHOT

Hi Shikhar. After upgrading to 0.9.0-SNAPSHOT I keep recieving: "Received data for nonexistent channel".
The problem occurs when I pass multiple connections through SSH Session. Some of them could be started "concurently" or one by one and most of them are short living. All connections are .
Server is OpenSSH-4.3

[shikhar]

Thanks for the feedback, will investigate...

Hello @shikhar is there any news regarding this issue?

[hierynomus]

Could somebody write a unit test for this so that we can solve it, it seems like a bug...