How I manage SSH connections

I have a number of machines I need to connect to via the secure shell (SSH), and typing in the IP address or host name by hand every time get tedious very quickly. I could use an application for that, but that's bloat. Let's instead see how we can leverage small universal tools to build an elegant solution of our own instead.

Since we have to use the terminal when using SSH anyway, we will create a command-line tool using FZF which lets us type the partial name of a host and make a selection. Here is what the end result will look like:

SSH >                                  ┌──────────────────────────────────────┐
  4/4                                  │ HostName     172.0.0.1               │
> Office                               │ User         johndoe                 │
  media-server                         │ Compression  yes                     │
  VM/dev-test                          │ ForwardX11   yes                     │
  VM/dev-master                        │                                      │
                                       │                                      │
                                       │                                      │
                                       │                                      │
                                       └──────────────────────────────────────┘

The goal is to create this interface and functionality using generic tools and whatever we already have installed on the system.

The big pattern

The first thing we have to do in order to tackle this large and new problem is to break it down into small and familiar problems by finding the underlying pattern. First we need a source of information and a sink into which the information eventually flows. Between the source and the sink lies a pipeline through which the data flows and gets changed along the way. In functional programming this is the filter-map-reduce pattern: we take a (potentially infinite) sequence of data, filter it down to just the information we need, transform it into another shape, and finally we collapse the sequence into one ultimate datum.

The source

For our source we ideally want something we already have and which other programs can use as well. Fortunately SSH offers us just that: a per-user configuration file which you most likely are already using. The file is located in ~/.ssh/config and the syntax is explained in ssh_config(5). Here is an example of what an entry looks like:

Host Office
    HostName 172.0.0.1
    User johndoe
    Compression yes
    ForwardX11 yes

The first like starts with Host (or Match), followed by a host pattern. The following lines until the next Host or Match are settings for the aforementioned host. We have thus the following rules:

  • Empty lines and lines starting with # are ignored

  • Each line is a record, fields of a record are separated by whitespace, leading whitespace is ignored

  • The first field is the setting (key), the second field is the value

  • A Host or Match starts the settings for a new host

For now only the value of the Host is of interest to us, but we will come back to the other rules later.

The sink

The sink is simply the ssh command. However, we cannot feed the name of the host into the standard input of ssh, we need to supply it as a command-line parameter. The xargs command allows us to splice together the command line we want:

# Same as calling `ssh Office`
echo 'Office' | xargs ssh

The pipeline

Given our source and sink we need to perform the following steps:

  • Filter out the host records from the config file

  • Map each host record to just the name of the host

  • Reduce the list of host names down to the one the user has selected

We can use grep to filter for the matching lines, cut to select the second field, and finally fzf to present the user with the menu. Go ahead and try out the commands in your shell manually to see how they work. The full command line looks like this:

grep '^[[:space:]]*Host[[:space:]]' ~/.ssh/config | cut -d ' ' -f 2 | fzf | xargs ssh

There is a small flaw though: if the user cancels the FZF selection the pipeline will keep going and ssh will be called without argument. We can fix this by storing the output of fzf in a variable and checking the exit status of the pipeline before calling ssh:

host=$(grep '^[[:space:]]*Host[[:space:]]' ~/.ssh/config | cut -d ' ' -f 2 | fzf)
[ $? -eq 0 ] && ssh "$host"

At this point we can call it a day, in two lines we have implemented a full SSH connection selector which uses the user's own SSH configuration. Let's make the presentation a bit easier on the eyes by specifying a few FZF settings. These can be passed to the fzf command, but I find it more readable to have then in an environment variable.

export FZF_DEFAULT_OPTS='
--height=20%
--reverse
--prompt="SSH > "'

host=$(grep '^[[:space:]]*Host[[:space:]]' ~/.ssh/config | cut -d ' ' -f 2 | fzf)
[ $? -eq 0 ] && ssh "$host"

You can set them to your liking of course. Here is what mine looks like:

SSH >
  4/4
> Office
  media-server
  VM/dev-test
  VM/dev-master

We can do better though: FZF will only display the host names, but if you have many connections you might want to see the details of the currently selected host in a little preview window.

FZF preview

FZF has a --preview option which allows us to specify a command line whose output will be displayed besides the selections. The command line can use the value of the current selection to construct a custom command for each selection. This means that given a host we have to pick out all its settings from the SSH config file.

A basic file processor

Recall the syntax of the config file from above. We have to loop over every line in the file, match it against the rules, and take appropriate action. We could write this file processor in any language, but in my opinion the best choice is Awk with its declarative syntax.

Awk works by reading records from a file (by default a record is a line of text), and for each line it matches it against a number of patterns and carries out the action if the pattern matches. We get the main loop and pattern matching for free, all we have to do is write the patterns actions. Here is our first attempt:

# The HOST variable is passed from the command line

/^$/ || /^#/ {
    next
}

($1 == "Host" || $1 == "Match") && did_find_host {
    exit
}

$1 == "Host" && $2 ~ HOST {
    did_find_host = 1
    next
}

did_find_host {
    print $1, $2
}

These patterns reflect the config file syntax rules from above. Let's go over them one by one.

  • An empty line (^$) or a line starting with a number sign (^#) is ignored, we move on to the next line

  • If we come across the definition of a host after we had already found our host we are done (the did_find_host variable is implicitly initialised to false in Awk)

  • If we come across the definition of our host we mark it as found and move on to the next line

  • If we come across a configuration after having found our host we print it

Now we need to add the preview command line to our FZF default options (the {} stands for the current selection).

export FZF_DEFAULT_OPTS='
--height=20%
--reverse
--prompt="SSH > "
--preview="awk -v HOST={} -f ~/.ssh/bin/host2conf.awk ~/.ssh/config"'

host=$(grep '^[[:space:]]*Host[[:space:]]' ~/.ssh/config | cut -d ' ' -f 2 | fzf)
[ $? -eq 0 ] && ssh "$host"

The result looks just like the entry from the config file and we could stop now if we wanted to.

SSH >                                  ┌──────────────────────────────────────┐
  4/4                                  │ HostName 172.0.0.1                   │
> Office                               │ User johndoe                         │
  media-server                         │ Compression yes                      │
  VM/dev-test                          │ ForwardX11 yes                       │
  VM/dev-master                        │                                      │
                                       │                                      │
                                       │                                      │
                                       │                                      │
                                       └──────────────────────────────────────┘

A better file processor

The preview looks hideous, frankly speaking. The lines are all jagged and everything is too crowded. How can we get the values to align neatly? We have to pass through the entire file first, collect the entries, find out which key is the widest, and use that to format the output in the end.

For this to work we need a new pattern END which will be run when we are done, and we will need to change our printing pattern to collect results instead.

BEGIN {  # Will match before processing the file
	n = 0  # Explicitly initialise as a number instead of empty string
}

did_find_host {
	keys[n] = $1
	values[n++] = $2
	width = max(length($1), width)  # Width of the widest key for padding
}

END {  # Will match after processing the file
	for (i = 0; i < n; ++i)
		printf "%-"width"s  %s\n", keys[i], values[i]
}

We use the variable n (set to 0 initially) to keep track of the number of lines, and we use the arrays keys and values to keep track of the individual keys and values. The reason for using two arrays is that Awk does not support nested arrays. We also keep track of the width of the widest key for alignment later.

The END pattern matches when the file has been read completely; we loop over all settings and print them out. We can specify the padding in a format string such as %-10s, where 10 is the width. Since in our case the width is dynamically computed we have to splice the format string together through string concatenation.

Putting all together we finally get our desired result:

SSH >                                  ┌──────────────────────────────────────┐
  4/4                                  │ HostName     172.0.0.1               │
> Office                               │ User         johndoe                 │
  media-server                         │ Compression  yes                     │
  VM/dev-test                          │ ForwardX11   yes                     │
  VM/dev-master                        │                                      │
                                       │                                      │
                                       │                                      │
                                       │                                      │
                                       └──────────────────────────────────────┘

Pure terminal beauty! We could now use ANSI control characters in our output to also add colour, but I am not willing to go that far down the rabbit hole.

Conclusion

Take the above shell script, put it in your $PATH, bind it to a key, write a .desktop file for it, do whatever you want. We started out with a simple shell script which did the minimum of what we need in just two lines of code, then added little bits over time to make it more pleasant to use. Aside from FZF all components come included with most Unix systems and use only what POSIX specifies.

But why not just use an integrated application off the shelf instead? Why waste time on building our own? An integrated application has one main advantage: it just works, out of the box. On the other hands, rolling our own from generic parts offers us with a number of advantages:

  • Universal programs like FZF can be used for other purposes as well, an integrated application can do only one thing

  • We can change it very easily by swapping out some of the generic components

  • The knowledge we gain while writing this one tool can be of use later (e.g. knowing how to program Awk)

  • Re-using the same application allows us to share settings (such as the SSH settings or FZF key bindings) instead of having to configure each integrated tool separately

The key to making this approach viable is that it has to be simple and cheap to put together. Unix pipes, plain text configuration and the declarative syntax of Awk allow us to focus on what is important instead of being weighted down by technical details.

But I don't want to use the terminal

If you prefer a GUI you can replace FZF with a graphical application like dmenu or rofi. This is the beauty of the pipeline: you can swap out any part of it without changing the rest of the logic. You could even write your script to use one and fall back on the other if it ist not available.

With that said, you will end up in the terminal eventually anyway, so you might as well start out in the terminal.

Whitespace in values

I made a little lie by omission: it is possible for values to contain whitespace characters if they are quoted. The above Awk script does not account for this fact. The easy solution is to simply not use whitespace, which is what I have decided to do.

Handling whitespace (or any general field separator) in Awk is not too hard, but it is not trivial either. The GNU Awk implementation supports the FPAT variable for this purpose. We could also use a more powerful programming language such as Perl or one with an SSH config parsing library.

The complete source code

The shell script ssh-select.sh:

#!/usr/bin/env sh

export FZF_DEFAULT_OPTS='
--height=20%
--reverse
--prompt="SSH > "
--preview="awk -v HOST={} -f ~/.ssh/bin/host2conf.awk ~/.ssh/config"'

host=$(grep '^[[:space:]]*Host[[:space:]]' ~/.ssh/config | cut -d ' ' -f 2 | fzf)
[ $? -eq 0 ] && ssh "$host"

The config file processor host2conf.awk

# Search the input (and SSH config file) for a given HOST and print all the
# host's settings in a tabular form to standard output. The HOST must be
# provided as a global variable to the Awk process.

BEGIN {
	n = 0  # Explicitly initialise as a number instead of empty string
}

# Skip comments
/^$/ || /^#/ {
	next
}

# A new host definition after we found our host terminates
($1 == "Host" || $1 == "Match") && did_find_host {
	exit
}

# Keep searching until we found our host
$1 == "Host" && $2 ~ HOST {
	did_find_host = 1
	next
}

# Accumulate all settings and their values for our host, ordered by their
# appearance in the input
did_find_host {
	keys[n] = $1
	values[n++] = $2
	width = max(length($1), width)  # Width of the widest key for padding
}

END {
	for (i = 0; i < n; ++i)
		printf "%-"width"s  %s\n", keys[i], values[i]
}

function max(a, b) {
	return a > b ? a : b
}