Mdthbs

I have two remote nodes that I am trying to send files from one to another.

scp remote1:~/testSCP remote2@10.0.1.10:~/

I have ~/ssh/config set up in my local machine so, it is using port 2222 by default.

But the remote1's default ssh port is set to 22 in the ssh config instead of port 2222. So, to make any external connection via ssh, it uses port 22 by default.

I tried the following which did not work:

scp -P 2222 remote1:~/testSCP remote2@10.0.1.10:~/

Also tried the following, which also did not work:

scp remote1:~/testSCP -P 2222 remote2@10.0.1.10:~/

For both I got the following error:

ssh: connect to host 10.0.1.10 port 22: Connection refused

lost connection

Which is true since 10.0.1.10 is using port 2222 and not port 22.

How can I specify remote1 to use port 2222 when trying to send files to 10.0.1.10 (remote2) from remote1?

Update

After trying

 scp -3 remote1:~/testSCP remote2@10.0.1.10:~/

I get weird behaviour. Even though my password is correct, it is giving me the following output:

remote1@10.0.1.9's password: remote2@10.0.1.10's password: 

Permission denied, please try again.

remote1@10.0.1.9's password: 

Permission denied, please try again.

remote1@10.0.1.9's password:

I have not enabled key-less authentication yet.

New Update

After trying it in several ways, I was able to do it in my scripts by logging in to the remote1 via ssh from my localhost and then scp from remote1 to remote2. However, this does not answer my question. I intended upon doing it directly from my local machine then transfer files between two instances, which I think is not supported if two instances' ssh daemons are using a different port than the default for ssh connection.

You cannot do that with a simple remote-to-remote scp [1].

Instead of it, ssh to the 1st remote host and run scp with a port argument from there:

ssh -p 2222 ruser1@rhost1 scp -P 2222 /rpath/1 ruser2@rhost2:/rpath/2

If you want to do exactly what scp is doing, you can also add the -n -x -oClearAllForwardings=yes options to ssh, though that's usually not needed.

[1]: newer versions of scp support a uri spec (including a port) instead of host:path, but only when using the -3 option ("pass through the local host").

So you could probably use

scp -3 -P 2222 ruser1@rhost1:/rpath/1 scp://ruser2@rhost2:2222//rpath/2

(notice that the / after the host[:port] is not part of the path -- scp://user@host/file will refer to ./file in the user's home directory).

But copying through the localhost is both slower, and in my experience it will hide errors. For instance, this won't print any error message, despite not being able to create any /foo/bar/baz file:

scp -3 localhost:.bashrc localhost:/foo/bar/baz

I did not get to any depth into this -- just avoided it ;-)

If someone is not convinced by all this, they can look at the source code:

void

toremote(char *targ, int argc, char **argv)

{

   ...

                } else if (host) {      /* standard remote to remote */

                        if (tport != -1 && tport != SSH_DEFAULT_PORT) {

                                /* This would require the remote support URIs */

                                fatal("target port not supported with two "

                                    "remote hosts without the -3 option");

                        }

Notice that the tport variable is only set by parsing a scp:// uri, and it simply doesn't exist in versions older than 7.6p1 (Oct 2017).