Remote Compilation Using ssh and make

by John R. Daily

Occasionally I use my Linux machine at home to write code that I intend to compile on a remote machine.

While maintaining open FTP and TELNET connections to the remote machine to handle the transfer and compilation steps is a manageable solution, I decided to explore ssh and make to develop a more automated method.

The benefits of my solution:

  • No need to remember which files have been modified.

  • Ability to use Emacs' compilation capabilities to move to errors in the source.

  • As mentioned above, no need to use FTP or TELNET, and hence no reason to keep an open dial-up connection when not compiling.

  • Automate, automate, automate—laziness is a virtue.

Overview of Solution

My first step was to set up ssh and related tools in order to allow secure copying of files to my remote system. While setting up a .rhosts file is a (barely) acceptable solution, my IP address and name is different each time I dial in, and it would be rather awkward to change the remote system's .rhosts file each time I dialed in.

ssh gives me a much more secure form of authentication for copying files and executing remote commands.

Once I had ssh behaving properly, I used Emacs' info facility to explore implicit rules in Makefiles and wrote a simple Makefile to handle the file transfers and remote compilation.

As an example of the intended effect, assume my remote machine is called “remote” (and my local machine “local”), and I've just modified a source file called daemon.c. I would like to execute the following commands in an automated fashion (note that scp is a secure copy command packaged with ssh, and that the -C option specifies compression, useful for dialup connections):

scp -C daemon.c jdaily@remote:~/source-directory
ssh -C remote "cd source-directory && make"
Implementation

First, I needed sshd running on the remote system to handle my secure connections. Fortunately, sshd was already running on the remote system in question. According to the man pages, it can be run by any user (i.e., user does not have to be root) and is restricted to handling connections for that user (which should provide sufficient security for our needs).

Then, I needed to install the ssh tool set on my local machine. Again, ideally these would be installed in a public binary directory such as /usr/local/bin, but any user can install them in his/her home directory.

I also wanted a key which would allow me to authenticate myself between systems and eliminate the need to type my password each time I tried to run one of the ssh commands. For this, I ran ssh-keygen and made sure not to give a pass phrase, so that none would be needed when using my private key to establish the connection. Listing 1 shows an example ssh-keygen session.

Once I had a public key, I used scp to copy it to the remote machine.

$ scp -C ~/.ssh/identity.pub \jdaily@remote:~/.ssh/keyjdaily's password: <entered my remote password>

Then I logged onto the remote host and copied the key file into ~/.ssh/authorized_keys. If that file had already existed, I would have appended the key file.

Following all this, I could run ssh and scp without needing either a password or a pass phrase to connect to remote.

Now I needed a makefile to automate my system. Ideally, the files on the remote machine would be checked to see if they were older than the files on my local machine, and if so, they would be replaced. To simplify matters, I decided to keep a record of the “last transferred date” for each file by touching a corresponding file each time I copied a source file over.

As an example, when I transferred a newer copy of daemon.c over, I touched daemon.ct in the same directory. Any transfer of a .h file would be marked by the creation of a file with a .ht suffix.

After poking around the info file for make, I came up with the Makefile shown in Listing 2. This Makefile had one limitation in particular; I was unable to specify command-line arguments for make on the remote machine without writing them directly into the makefile on my local system. While this was fine for the current application, I decided to generalize it by creating a run-make shell script which would handle the remote execution of make after calling make on the local system. Listing 3 is my run-make shell script. I then removed the line from my Makefile which remotely ran make. Listing 4 is the output from a successful compilation sequence.

Resources

John Daily works for BBN (or is that GTE?) as a systems administrator/software engineer. He spends far too much time in front of computers. Whenever he can, he prefers to be outside riding his new bicycle and exploring New England. He can be reached via e-mail at jdaily@bbn.com.

Load Disqus comments