▲ 381 points
back
77 comments
My favorite use of this is peer-to-peer transfer of Docker images. The Docker CLI only allows you to use registries authenticated with HTTPS but there's an exception where it allows HTTP transfers over localhost.
So, if you use SSH tunneling to forward a port from localhost to a remote, then Docker unwittingly pushes to a remote. This is super useful "off the grid" with robotics/embedded applications where you don't want to bother with a registry and a good Internet connection.
Example, docker pussh: https://github.com/psviderski/unregistry
This is what kamal does when you use localhost as registry [0] [1].
Pretty cool thing, I ship plenty of services to my tiny $5/mo vps with it without having to pay for a docker registry.
[0] https://kamal-deploy.org/docs/configuration/docker-registry/...
[1] https://github.com/basecamp/kamal/blob/eee0083b38661c3707c6b...
That's not quite true, you just need to add the `insecure-registries`[1] option with a list of either IP (or ip ranges) or hostnames that you want to allow without TLS.
```/etc/docker/daemon.json
{
"insecure-registries": ["10.100.0.0/24", "registry.yourmom.example.com:5000"]
}
[1] https://docs.docker.com/reference/cli/dockerd/#insecure-regi...
This is really useful as you don't have to add an entry under insecure-registries for local registries that don't have valid certificates.
iirc there's a setting to allow docker to trust and use http registries
i set it up a few years ago for my homelab
I'll mention it here, because I learned about it here.
"~C" will drop you into the SSH command line, allowing you to, among other things, effect port forwarding
-L8080:localhost:443
But for some reason this has been disabled by default in more recent ssh configurations... to ensure its available
-o EnableEscapeCommandline=yes
EnableEscapeCommandline yes
Important to note that `~` SSH commands work only right after you press Enter - it doesn’t trigger everywhere you press `~`.
Also EnableEscapeCommandline fortunately only affects `~C` - the all-important `~.` to kill a hung SSH session still works with it disabled.
Eh, once I started using master sockets, I never went back. Problem with ~C is it's hard to keep track of what you have open.
I never pass up an opportunity to recommend the Cyber Plumber's Handbook: https://github.com/opsdisk/the_cyber_plumbers_handbook
Goes over similar content as TFA, in perhaps a little more depth. Indispensable sysadmin knowledge.
Not mentioning the ssh -w option in that book should be a crime.
Appreciate the mention wbadart!
so good, I learn new things.
The article mentions bastions, but no jumphosting?
ssh -J user1@bastion1,user2@bastion2 targetuser@targethost
It is surprising how many times I see this content (this version might be marked “Published: Jun 19, 2026” but I've definitely seen those exact diagrams before, starting at least a few years ago, and the same content around them in many tutorials before that) without it being updated to mention jump-hosts.
Support was added to OpenSSH about a decade ago? Even on a low moving Linux distro like Debian/LTS everyone should have support by now.
What I've found beautiful about -J is the host you jump through requires no privileges on the final host. Only my laptop has the SSH key to access my home server, not my cheap VPS.
And this allows me to have zero open ports on my home internet. I do a reverse tunnel to my VPS from my home server (in a FreeBSD jail), and that port is what my laptop client jumps through.
For me, this is always used via ProxyJump rules in my ~/.ssh/config
It is also nice that it works recursively, so I can logically structure my rules so that the one for my regular targets say to use bastion1, then the rule for bastion1 says to go via bastion 2, etc.
I find this easier to reason about and maintain rather than juggling a bunch of these multi-step rules.
And with match/exec rules you can always connect to MyHost and make it conditional whether to use a jumphost or not, so it's like an on demand vpn.. only with ssh.
Match host="MyHost" exec "! grep Home ~/.wifi-loc-control/.current"
ProxyJump home-jumphost.mydomain.tld>ssh -J user1@bastion1,user2@bastion2 targetuser@targethost
Are you using SSH key auth or password authenticating three times when you do this?
Need to mention sshuttle [0] here, as it magically solves a bunch of these problems without constant reconfiguration
Yup, use it regularly in order to jump through networks where a VPN/Wireguard setup is not possible. It can also forward DNS requests and handle local NS operations reasonably well, such that it can be used as a low-key split-VPN client (i.e. only forwards traffic for a specific domain or IP range without redirecting any other traffic). Note that for integration with `systemd-resolved`, one needs to jump through a few hoops, but I feel its works very nicely: https://github.com/sshuttle/sshuttle/issues/688#issuecomment....
sshuttle is amazing. I've used it extensively on stupidly configured networks, super useful tool.
Learning how SSH port forwarding is great as a pseudo-vpn for everything from GUI-client database access to (in physical infra) access to web-admin tools for appliances.
The socks proxy support can also deal with bad web filtering and privacy issues on public wifi networks (though nowadays if you're ssh'ing to a cloud IP, you'll get lots of "bot" restrictions).
Yeah, I get use out of the SOCKS proxy mode in combination with a "split VPN" at work.
I need VPN to get into some internal resources via SSH, but there are lots of external/public/AWS resources I also need to access, and the full VPN adds too much overhead and fragility for those.
Using the available split VPN, I can point a browser instance at a localhost SOCKS proxy port to relay over SSH + VPN for other web resources I need to access internally.
Unfortunately, Firefox proxy config rules are sort of backwards for my needs. I want to say "only use proxy for these 3 domains" whereas it wants to use the proxy by default and only allow me to bypass specific domains.
> The socks proxy support ...
I just love SOCKS proxy in SSH tunnels: at some point I had a dedicated server (on a fixed IP) with countless machines (usually headless Pis dropped at a family member's place and/or SME office) automatically setting up, 24/7, reverse tunnels to that dedicated server.
Then I could, from anywhere, both access their LANs (to fix stuff) and have a browser, running locally, pretending to be in this or that country.
Basically because I had all those reverse tunnels always there, I could always decide how to use them (just SSH in or SOCKS in etc.).
Hak5 video that helped me understand many years ago about the practical usage of SSH forwarding
Hak5 - SSH Forwarding: Local vs Remote with examples, Hak5 1113 part1 https://www.youtube.com/watch?v=g_Row8zEJZc
BTW I use this and a systemd unit file and SSH tunnel my Jellyfin to a public VPS to its local host.
I then use nginx to proxy it.
Because its a unit file, sshd reconnects if my ISP's IP changes. Does so within 30s. Also hides my ISP IP in case I have to turn it off.
And no data is effectively on the VPS. Its just a mostly empty machine.
If you have many different remote devices behind NATs or firewalls, a cool trick to access them all via EC2 server (or such) is to setup Remote Forwarding via UNIX socket on the server side, to devices' port 22. Preferably, UNIX socket filenames should start with a common prefix, so an SSH config can be written that will use ssh+socat in a ProxyCommand to establish the connection.
It's amazing how lightweight this method actually is. I have managed to connect hundreds of devices using a single EC2 nano instance.
I think the more modern ProxyJump rule is superior for this. Just let it manage the actual TCP forwarding for you automatically. It's just the normal "bastion host" concept.
Particularly, you can use name patterns to apply the same rule broadly, assuming you have some systematic naming scheme for your eventual target devices.
Do you have more info on this method? How is the remote forwarding actually done?
It’s amazing what you can learn by reading the manual.
It is, because manuals are often not the best way to learn things. Most software manuals are reference manuals. SSH man page isn't too bad. I learned most of my SSH knowledge from it, but I'm not sure it's the best way to do it.
It's amazing how badly written most manuals are.
Reminiscent of Steve Friedl's excellent work: http://www.unixwiz.net/techtips/ssh-agent-forwarding.html
When I see one of these with obvious AI tells at the top (sentences lacking a subject or verb), I ask myself:
Can’t I just open up a harness and prompt “Teach me how to do X?”
I do this all the time, I have a skill/gem with instructions on how I want to receive info, how to format and so on. Really helps to go fast to get the point.
I personally do this, ask claude code to teach me about concepts I don't know about when it codes something, and only then I accept what it suggests to me
The best part of this is you can daisy chain tunnels and have a network connection between any machines you can ssh into.
If you have a private server with a public IP, you don't need localtunnel etc.. you can just use ssh tunnels to expose your home network services over public IP.
Or you could just install something like Tailscale and never have to think about it again.
or tell an llm to do it for you
There's a asymmetry here that "-R" works both for reverse static and dynamic (using SOCKS protocol) forwarding, but "-D" is required for dynamic forwarding which "-L" cannot do.
Why is that?
It's historical. Some older flags could be easily extended for dynamic port support and others could not.
As a sysadmin, one of your biggest ROI is learning the ins and outs of SSH.
Thanks! I will keep this for reference. I use ssh alot but thie reminds me that I can learn new ways of tunneling :-)
I think that the start of the article is at least a bit exaggerated:
SSH is yet another example of an ancient technology that is still in wide use today.
Weird.. I have almost this exact same "cheat sheet" printed out and stuck up on my whiteboard.. except it's slightly different. Only 4 panels instead of 6 and the panels don't have titles.
I know this is a solid "cool story bro" moment, but whatever hah
Another option that I never see mentioned anywhere is -w which allows you to create either layer2 or layer3 tunnels via a tun(4) interface.
Always annoys me -p(port) for ssh clashes with -p(reserve permissions) for sftp and scp. Unavoidable given history.
Also annoys me we "invented" ssh:// url format after the tools were baked so it's a somewhat odd bonding into the model.
Very refreshing to see a utilitarian series such as this. Disappointing that the latest ai drama gets 20x more discussion and visibility on this site
Should add how to bypass MFA using phishing and SSH Multiplexing to the article.