Feb 162001

Accessing a cvs repository

This article shows how to use a cvs repository to check out a local copy, modify the contents, then check the working copy back into the main repository.  It also shows how to control the access to the main repository and how to use ssh for the connection.

I will assume you have already read cvs – create your own repository and know how to create a repository.  In this article I will use the term working to refer to a repository which exists on your local computer.  The main repository will exist somewhere else.  The working copy is s q what you check out, modify, then commit back to the main repository.  This strategy is common practice and is The Way To Go(tm).

When I started writing this article, I was trying to get access to a cvs repository from another box.  I promptly became very confused between cvs and cvsup.   cvs is an Concurrent Versions System.  Basically, it’s a repository.  cvsup is a network distribution package for CVS repositories.  One is the actual repository.  The other deals with distributing that repository.

You use cvsup to update your ports or get your source code.

The main resource for this article is http://cvsbook.red-bean.com/cvsbook.html.

Setting up the cvs server

Most of this section was taken from The Password-Authenticating Server and describes how to configure a cvs server.  I will assume the repository is at cvs.example.org.  Here’s a command I could use for the login to that server:

$ cvs -d :pserver:dan@cvs.example.org:/usr/local/cvs login

I’ll come back to the details of this command in a later section.  For now, I will deal with the cvs server setup.  The reason for showing this command is that it is directing cvs to use the pserver method for logging in.  This is by far the most common access method.  I’ll now describe how to do enable pserver on cvs.example.org.  This is accomplished with the following steps:

  1. Ensure the service is in the service name database.  This entry should already be in /etc/services, but you should verify that it is:

    cvspserver 2401/tcp
  2. Enable the service via inetd by adding the following line (which will already be present but will require modification) to /etc/inetd.conf:

    cvspserver stream tcp nowait root /usr/bin/cvs cvs --allow-root=/usr/local/cvs pserver

    Note that the example above differs from that found in the URL above because cvs is located in a different location under FreeBSD.  Use the above.  Not that found in the URL.

    In this example, the repository resides at /usr/local/cvs.
  3. HUP inetd so that it takes notice of the above changes:

    # killall -hup inetd

With the above changes, you should be able to connect to the cvs server.  Provided any firewall rules allow it.  Here’s a simple test.

$ telnet cvsup.example.org 2401
Connected to cvsup.example.org.
Escape character is '^]'.

If that’s what you see, then you’ve succeeded in setting up the cvs server.   Congratulations.  Now let’s add a bit of security.

Adding passwords to the server

When adding passwords to the server, make sure you don’t use the same passwords people use to login into the system.  Keep them different.  If the cvs passwords are ever compromised, they can’t be used to get into the rest of the system.   I put my cvs repository files under the control of a dedicated user.  I created the user especially for this task.  Just use adduser.  Read The Password-Authenticating Server for more detail here.

Here’s what I added to /usr/local/cvs/CVSROOT/passwd:


Entries in this file are of the format:


There’s a script at the above URL for generating such passwords.  I suggest you use it.  Note that the output from the password script is time sensitive.   That’s OK.  It’s supposed to be that way.  It’s not a problem.  It’s also not a bug.  That’s by design.  And it does work.

Accessing the server using the password

Let’s go back to where I started.  This was the command I used to access the cvs server:

$ cvs -d :pserver:dan@cvs.example.org:/usr/local/cvs login
(Logging in to dan@cvs.example.org)
CVS password:

And that’s it.  You’re logged into the cvs server.  You can issue all the normal commands.  Such as a checkout:

$ cvs co scripts
cvs server: Updating scripts
U scripts/.procmailrc
U scripts/archive-logs.sh
U scripts/crontab
U scripts/fp-updates.dtd
U scripts/load_xml_into_db.pl
U scripts/log-catcher.awk
U scripts/log-munger.awk
U scripts/openbsd-cvs-parse.pl
U scripts/process_cvs_mail.pl

On the cvs server, the repository resides at /usr/local/cvs.   We are obtaining the contents of /usr/local/cvs/scripts.  The above checkout obtains a working copy of that repository.  You can treat those files exactly as you would any other files.  See cvs – create your own repository for command examples.

And when you’re finished, don’t forget to log out.

$ cvs logout
(Logging out of dan@cvs.example.org)

using ssh for added secrecy

Since writing this article, I started using stunnel for my cvs connections.

So far I have been using the pserver method for connecting to the cvs server.  This method works fine and is acceptable over trusted networks (e.g. over your own LAN).  But if the Internet or other shared network lies between you and the cvs server, you might want to consider using a more secure method.  Luckily, cvs is designed to use other external programs for client-server communication.

In this example, I’m going to use ssh.  In my previous examples, cvs has been told that the repository exists on another box (e.g. pserver:dan@cvs.example.org).   That works well.  But for this example, I’m going to make cvs think the repository is local, but it will actually reside remotely.  I will use an SSH tunnel for this.  This method requires that you have a shell account on the cvs server and that you an access that account from your workstation.  The main reason I’m using a tunnel is because my cvs server sits behind my gateway and is NAT’d.   If your cvs server is not NAT’d you can use just plain straight ssh.

If you use this method, you’ll also need to use two terminal session on the workstation.  One will be used for your cvs operations.  The second will be used for an ssh tunnel.

My reference for this configuration is http://jakarta.apache.org/site/cvsonunix.html.

  1. Create the ssh tunnel to the cvs server:

    ssh -L 2401:localhost:2401 cvsup.example.org
    dan@cvsup.example.org's password:
    Last login: Fri Feb 16 09:05:30 2001 from box.example
    Copyright (c) 1980, 1983, 1986, 1988, 1990, 1991, 1993, 1994
    The Regents of the University of California. All rights reserved.
    FreeBSD 4.2-STABLE (XEON) #0: Sun Feb 11 14:45:37 NZDT 2001

    Welcome to the dual XEON.

    You will remember port 2401 from above.  That’s the port on which cvs connects to the cvs server.  The above command redirects local port 2401 (-L 2401) to port 2401 on cvsup.example.org.
  2. login into the cvs server using pserver:

    $ cvs -d :pserver:dan@localhost:/usr/local/cvs login
    (Logging in to dan@localhost)
    CVS password:

    Note that I am using localhost here.  Not cvsup.example.org as I have in previous examples.  I can use localhost because of the ssh tunnel set up in step 1.  This tunnel redirects port 2401 on localhost (which the login will use) to port 2401 on the cvs.example.org.

  3. use cvs as normal:

    $ cvs checkout www
    cvs server: Updating www
    cvs server: Updating www/images

    That’s it.

I use screen to keep that second connection in the background.  Note that this command connects to port 2022 because that port is redirected to the cvs server which is behind a NAT’d gateway.

screen -S tunnel ssh -p 2022 -L 2401:localhost:2401 cvsup.example.org

Then I detach from the screen session, leaving the ssh tunnel in the background.

Even better than screen, is this approach:

$ ssh -L 2401:localhost:2401 -N -f -g cvsup.example.org
The -N means you are not supplying a command (e.g. ls, sleep). The -f makes it go into the background. And -g is to allow hosts other than localhost to connect (e.g. other hosts on your network). Ensure only your network can connect. You don’t want others getting in there….

plain old ssh

As mentioned before, I used the tunnel because of NAT.  But you can do this:

export CVS_RSH=ssh
export CVSROOT=:ext:dan@cvs.example.org:/usr/local/cvs

I have tried:

CVS_RSH="ssh -p 2022"

because of my NAT’d gateway, but I had no success:

$ cvs checkout www
cvs [checkout aborted]: cannot exec ssh -p 2022: No such file or directory
cvs [checkout aborted]: end of file from server (consult above messages if any)

That’s why I used the ssh tunnel instead.

But since I wrote the above, David MacKenzie wrote in with this solution:

Create a shell script, ssh2022 which contains this:

exec ssh -p 2022 "$@"

Then chmod that file:

chmod 755 ssh2022

Then your export statement is:

export CVS_RSH=/usr/local/bin/ssh2022

David says this technique also works with rdist6 -P and rsync -e.

Keep it short and sweet

You don’t have to specify the -d option with the cvs command.  You can set an environment variable:

export CVSROOT=:pserver:dan@cvs.example.org:/usr/local/cvs

Then you can just issue this command:

$ cvs login
(Logging in to dan@cvs.example.org)
CVS password:

But if you are using an ssh tunnel, you should specify local host.  If you don’t, cvs will ignore the ssh tunnel and connect directly to the remote host and transmissions will be in clear text.

export CVSROOT=:pserver:dan@localhost:/usr/local/cvs

Errors I have known (added on 19 June 2002)

I encountered this error when setting up my FreshSource repository:

$ cvs co configuration
cvs checkout: authorization failed: server xeon rejected access to /usr/repo/freshsource for user mike
cvs checkout: used empty password; try "cvs login" with a real password
To fix this problem I had to do two things:
  1. Remember to add an entry to /etc/inetd.conf for the new repository (/usr/repo/freshsource).
  2. Add an entry to ~/.cvspass which related to /usr/repo/freshsource.
I could have just done a cvs login instead of step two above. But the problem was I could not remember my cvs password… So I cheated. Read more about this in The Red Bean Book section on login.

Well, I just noticed that I never mentioned ~/.cvspass. I suggest reading The Red Bean Book and look for Run Control Files.

Other errors I have known (added on 8 January 2003)

I was setting up a new repo today and I really messed something up. Here was the error I was getting on the server:
Jan 8 13:13:33 xeon cvs: login refused for /home/repo/mynewrepo
I was playing with ~/.cvspass and /home/repo/mynewrepo to ensure the passwords were correct. Nothing worked. Finally, I went to /etc/inetd.conf and found the problem. The line for cvspserver contained this entry:
I had missed a /. I fixed that. Then things worked. Here is another situation I recently encountered:
$ cvs co www
cvs [checkout aborted]: unrecognized auth response from xeon: cvs [pserver aborted]: descramble: unknown scrambling method
This was a case of the password in ~/.cvspass on the client not matching the password in /home/repo/mynewrepo on the server.

  One Response to “Accessing a cvs repository”

  1. This and the other article on CVS tought me pretty much everything I needed to get going with CVS. The one thing that I did change was to use public keys for establishing the ssh session from my windows machines with a combination of putty, pagent and TurtleCVS.

    Works very nicely!