feat: make sshnpd execute ssh command via Process.start

[gkc]

- What I did

• feat: For perf reasons, by default we will have SSHNPDImpl directly execute the ssh command on the host rather than using SSHClient
  • writes privatekey which it received earlier from sshnp into a tmp file which is later deleted
  • added a bunch of comments explaining how the command is constructed and why
• refactor: remove _sshPublicKey instance variable from SSHNPDImpl and make it a local variable in the case 'sshpublickey' in the notification handler
• feat: Direct AtSignLogger log messages to stdErr for sshnpd, sshnp and sshrvd via AtSignLogger.defaultLoggingHandler = AtSignLogger.stdErrLoggingHandler;
• feat: Added command-line option --ssh-client which can currently have values '/usr/bin/ssh' (the default) or 'pure-dart'. If set to pure-dart then we'll do it using SSHClient as we used to.
• feat: added more command-line options to the ssh command which sshnp generates and prints to stdout
  • -o StrictHostKeyChecking=accept-new to silently accept new, but still check existing
  • -o IdentitiesOnly=yes to use only the identity in the specified identity file

[gkc]

@cconstab My local manual tests worked but the e2e tests are failing with 'Host key verification failed', because of something I'm getting wrong. Please take a look at how I'm constructing the ssh command - a fresh pair of eyes might see what the problem is straight away.

[gkc]

I believe the tests are failing because SSHClient does not do host key verification, but ssh command by default does

TerminalStudio/dartssh2#51

Fix would be to add this to the ssh command args

-o StrictHostKeyChecking=no -o UserKnownHostsFile=/dev/null

[gkc]

Applied the fix mentioned in the previous comment, and now only one of the e2e tests is failing, but it's failing in a different way.

130 sshnpd  | SHOUT|2023-07-23 18:19:51.546987| sshnpd |197de568-697a-4851-a789-9e82a722e5cb | Non-zero exit code from /usr/bin/ssh atsign@10.1.0.38 -p 40749 -t -t -i /tmp/.pem-436bdbbb-5052-4cb8-bd16-80ab0f86c376 -R 34925:localhost:22 -o StrictHostKeyChecking=no -o UserKnownHostsFile=/dev/null sleep 15 
131 sshnpd  | SHOUT|2023-07-23 18:19:51.548265| sshnpd |197de568-697a-4851-a789-9e82a722e5cb | stdout   :  
132 sshnpd  | SHOUT|2023-07-23 18:19:51.548293| sshnpd |197de568-697a-4851-a789-9e82a722e5cb | stderr   : Warning: Permanently added '[10.1.0.38]:40749' (ED25519) to the list of known hosts.
133 sshnpd  | Load key "/tmp/.pem-436bdbbb-5052-4cb8-bd16-80ab0f86c376": error in libcrypto
134 sshnpd  | atsign@10.1.0.38: Permission denied (publickey).

I think the repeat failure of the local-to-local test is an artefact of the test rig

All of the other tests which are running sshnpd from the branch (local) are passing (i.e. trunk to local, latest release to local, 3.2.0 to local are all passing)

In this run, only the first test (local to local) failed: https://github.com/atsign-foundation/sshnoports/actions/runs/5637947000 with the Permission denied (publickey) problem

But in this run, both the local to local test failed, AND the v3.2.0 to local test failed for the same (Permission denied (publickey)) reason https://github.com/atsign-foundation/sshnoports/actions/runs/5637996683/attempts/1
But then it passed in the re-run https://github.com/atsign-foundation/sshnoports/actions/runs/5637996683

The local-to-local test will not pass no matter how many times I re-run it

@JeremyTubongbanua Can you think of anything particularly special about the local-local test with regards to timings, setup, etc?

[JeremyTubongbanua]

Applied the fix mentioned in the previous comment, and now only one of the e2e tests is failing, but it's failing in a different way.

130 sshnpd  | SHOUT|2023-07-23 18:19:51.546987| sshnpd |197de568-697a-4851-a789-9e82a722e5cb | Non-zero exit code from /usr/bin/ssh atsign@10.1.0.38 -p 40749 -t -t -i /tmp/.pem-436bdbbb-5052-4cb8-bd16-80ab0f86c376 -R 34925:localhost:22 -o StrictHostKeyChecking=no -o UserKnownHostsFile=/dev/null sleep 15 
131 sshnpd  | SHOUT|2023-07-23 18:19:51.548265| sshnpd |197de568-697a-4851-a789-9e82a722e5cb | stdout   :  
132 sshnpd  | SHOUT|2023-07-23 18:19:51.548293| sshnpd |197de568-697a-4851-a789-9e82a722e5cb | stderr   : Warning: Permanently added '[10.1.0.38]:40749' (ED25519) to the list of known hosts.
133 sshnpd  | Load key "/tmp/.pem-436bdbbb-5052-4cb8-bd16-80ab0f86c376": error in libcrypto
134 sshnpd  | atsign@10.1.0.38: Permission denied (publickey).

I think the repeat failure of the local-to-local test is an artefact of the test rig

All of the other tests which are running sshnpd from the branch (local) are passing (i.e. trunk to local, latest release to local, 3.2.0 to local are all passing)

In this run, only the first test (local to local) failed: https://github.com/atsign-foundation/sshnoports/actions/runs/5637947000 with the Permission denied (publickey) problem

But in this run, both the local to local test failed, AND the v3.2.0 to local test failed for the same (Permission denied (publickey)) reason https://github.com/atsign-foundation/sshnoports/actions/runs/5637996683/attempts/1 But then it passed in the re-run https://github.com/atsign-foundation/sshnoports/actions/runs/5637996683

The local-to-local test will not pass no matter how many times I re-run it

@JeremyTubongbanua Can you think of anything particularly special about the local-local test with regards to timings, setup, etc?

Yes, @XavierChanth recently refactored the tests and I believe the bug is addressed in a comment in this ticket - #276

[JeremyTubongbanua]

Fix should be in #280

[gkc]

@JeremyTubongbanua thanks - the problem I was encountering was actually a different problem (Permission denied (publickey)) which @cconstab diagnosed here

[JeremyTubongbanua]

@JeremyTubongbanua thanks - the problem I was encountering was actually a different problem (Permission denied (publickey)) which @cconstab diagnosed here

Awesome :)

[gkc]

@cconstab Ready for final review