* ssh-agent: lock this script with a mkdir style mutex
This script is a kind of singleton pattern and is not reentrant.
If several shells are oppened in a fast sequence, then several
independent ssh-agents would be created, which is not acceptable.
A mutex is required.
Signed-off-by: Nuno Goncalves <nunojpg@gmail.com>
* ssh-agent: only start agent if .ssh dir exists
To use the same profile system-wide, it might happen
that the .ssh directory does not exist
(typically $HOME/.ssh/). This would trigger a error.
Creating the directory would be a option, but it
usually will not make sense to do so because it means
the user doesn't have ssh keys or config.
Signed-off-by: Nuno Goncalves <nunojpg@gmail.com>
* ssh-agent: adds lazy option to disable key loading on start
Option is documented on updated README.md
Signed-off-by: Nuno Goncalves <nunojpg@gmail.com>
* ssh-agent: simplify agent-forwarding checking
Signed-off-by: Nuno Goncalves <nunojpg@gmail.com>
Co-authored-by: Robby Russell <robby@planetargon.com>
This change makes the plugin check if an identity is loaded by looking
first at the key filename reported by `ssh-add -l`. This fixes the use
case where ssh-keygen is not able to output the fingerprint of a key,
such as the one reported on #7516.
Now, for an identity to be passed onto ssh-add, it has to fail the
match for a loaded identity, both filename and signature.
With this PR the ssh-agent plugin loads all identities which are not yet
loaded in a single call to ssh-add. If a passphrase is shared between
loaded identities it only needs to be entered once.
Fixes#7506
With this PR the ssh-agent plugin checks the `ssh-add -l` output for the
identities added, and adds all those specified by the user that haven't been
added yet.
We also decouple the logic of starting ssh-agent from the logic of adding
identities, meaning that even if ssh-agent has been started by some other means
(like launchd) we can still ssh-add the user's identities.
Fixes#3019Fixes#6979
When invoking a shell as root using ```sudo -s```, the ssh-agent plugin
starts a new agent although it already exists.
The problem boils down to a check if ssh-agent is running using
```ps x```. If that is extended to ```ps ax``` for root, then the
existing ssh-agent will still work.
On systems where the shell cannot be changed because of a strict
security policy, ssh-agent will use the syntax of whatever the
default $SHELL is.
For instance, if the default shell is tcsh, ssh-agent will use the
c-shell style (setenv).
This change forces ssh-agent to use bourne-style syntax since that
has to be later interpreted by zsh. Consequently, the environment
file will contain `export' statements from now on (instead of
`setenv').
On an OS X laptop, the variable `$HOST` changes a lot depending
on what wifi network you're connected to. This causes a lot
of `~/.ssh/environment-$HOST` files to be created and
causes multiple ssh-agents to created.
Instead, use `scutil --get ComputerName` to get something
more stable.
By default, ssh-agent stores identities forever. It has an option to
set a maximum lifetime for identites (useful to expire passphrase protected
keys). Allow this option to be set using:
zstyle :omz:plugins:ssh-agent lifetime <time>
