Existing SSH Sessions Not Reused with ControlMaster

[sghng]

Reproduction steps

Here's my sample SSH config:

Host server
	HostName example.com
	User xxx
	ControlMaster auto
        ControlPath ~/.ssh/cm-%r@%h:%p
	ForwardAgent yes
	ControlPersist yes

Launch an SSH session in terminal emulator, then connect to this remote in Zed.

I looked at the source code (ssh.rs), it seems like Zed opens a multiplexed session when connecting to remote for its terminal/SFTP file browser explicityly with -o ControlMaster=yes. It would be nice for it to detect an existing session defined in the.ssh/config to reuse.

Current vs. Expected behavior

Current behavior: Zed still prompts me to input password.

Expected behavior: Zed should reuse the existing ControlMaster SSH session.

Zed version and system specs

Zed: v0.217.2+stable.102.3d2e6ee76b4d9a932248242506611985cae6b0a8 (Zed)
OS: macOS 26.1
Memory: 36 GiB
Architecture: aarch64

Attach Zed log file

Zed.log

Relevant Zed settings

settings.json

Relevant Keymap

keymap.json

(for AI issues) Model provider details

No response

If you are using WSL on Windows, what flavor of Linux are you using?

None

[sbiquard]

Seeing the same thing with 0.219.4

[chen1plus]

Zed may be cut off if the external master connection was drop 😂
Having said that, I hope at least we can set ControlPersist

[nicosalm]

I've been looking into this and have a patch ready.

The problem is in MasterProcess::new(). Zed hardcodes ControlMaster=yes with a ControlPath in a random temp directory (/tmp/zed-ssh-sessionXXXXXX/ssh.sock), and -o flags on the command line override ~/.ssh/config. So even if you already have an authenticated ControlMaster running, Zed will never find it and prompts you again.

My patch checks for an existing master before Zed spawns its own. It runs ssh -G <destination> to get the user's resolved controlpath (tokens like %h, %p already expanded), then ssh -O check on that socket to see if it's alive. If it is, skip askpass and use the existing socket. If not, fall back to the current behavior. Both commands are local and fast (no network round trip).

I also had to change has_been_killed(), which was checking self.master_process.lock().is_none(). When reusing an external master, master_process starts as None, so the connection pool would think it's dead immediately and throw it away. I added a killed: AtomicBool to track this separately.

The change is ~220 lines in crates/remote/src/transport/ssh.rs, compiles clean, doesn't touch the Windows path. I tested it against a university SSH setup with ControlMaster auto configured. The reuse path connects without prompting when a master is already open, and the fallback still works normally when there's no existing master.

A couple things I want to check before opening a PR:

• If the external master drops mid-session, Zed's reconnect logic creates its own master on the next attempt. Seems right to me but wanted to confirm.
• kill() is a no-op when reusing (we shouldn't kill a master the user manages). It sets the killed flag for the connection pool but leaves the socket alone.
• There's some duplication between the #[cfg(not(windows))] and #[cfg(windows)] blocks now. The original code interleaved the two paths with inline #[cfg] annotations, but splitting them out was the cleanest way to add the reuse branch. Open to restructuring if you'd prefer something else.

Let me know if this approach looks right and I'll open a PR.