git-cvsserver - A CVS server emulator for Git
export CVS_SERVER="git cvsserver" cvs -d :ext:user@server/path/repo.git co <HEAD_name>
cvspserver stream tcp nowait nobody /usr/bin/git-cvsserver git-cvsserver pserver
git-cvsserver [<options>] [pserver|server] [<directory> ...]
This application is a CVS emulation layer for Git.
It is highly functional. However, not all methods are implemented, and for those methods that are implemented, not all switches are implemented.
Testing has been done using both the CLI CVS client, and the Eclipse CVS plugin. Most functionality works fine with both of these clients.
All these options obviously only make sense if enforced by the server side. They have been implemented to resemble the git-daemon(1) options as closely as possible.
-h, -H, --help
CVS clients cannot tag, branch or perform Git merges.
git-cvsserver maps Git branches to CVS modules. This is very different from what most CVS users would expect since in CVS modules usually represent one or more directories.
cvspserver stream tcp nowait nobody git-cvsserver pserver
Note: Some inetd servers let you specify the name of the executable independently of the value of argv (i.e. the name the program assumes it was executed with). In this case the correct line in /etc/inetd.conf looks like
cvspserver stream tcp nowait nobody /usr/bin/git-cvsserver git-cvsserver pserver
Only anonymous access is provided by pserver by default. To commit you will have to create pserver accounts, simply add a gitcvs.authdb setting in the config file of the repositories you want the cvsserver to allow writes to, for example:
authdb = /etc/cvsserver/passwd
The format of these files is username followed by the encrypted password, for example:
You can use the htpasswd facility that comes with Apache to make these files, but only with the -d option (or -B if your system suports it).
Preferably use the system specific utility that manages password hash creation in your platform (e.g. mkpasswd in Linux, encrypt in OpenBSD or pwhash in NetBSD) and paste it in the right location.
Then provide your password via the pserver method, for example:
cvs -d:pserver:someuser:somepassword@server:/path/repo.git co <HEAD_name>
No special setup is needed for SSH access, other than having Git tools in the PATH. If you have clients that do not accept the CVS_SERVER environment variable, you can rename git-cvsserver to cvs.
Note: Newer CVS versions (>= 1.12.11) also support specifying CVS_SERVER directly in CVSROOT like
cvs -d ":ext;CVS_SERVER=git cvsserver:user@server/path/repo.git" co <HEAD_name>
This has the advantage that it will be saved in your CVS/Root files and you don’t need to worry about always setting the correct environment variable. SSH users restricted to git-shell don’t need to override the default with CVS_SERVER (and shouldn’t) as git-shell understands cvs to mean git-cvsserver and pretends that the other end runs the real cvs better.
# optional for debugging
Note: you need to ensure each user that is going to invoke git-cvsserver has write access to the log file and to the database (see Database Backend. If you want to offer write access over SSH, the users of course also need write access to the Git repository itself.
You also need to ensure that each repository is "bare" (without a Git index file) for cvs commit to work. See gitcvs-migration(7).
All configuration variables can also be overridden for a specific method of access. Valid method names are "ext" (for SSH access) and "pserver". The following example configuration would disable pserver access while still allowing access over SSH.
export CVS_SERVER="git cvsserver"
cvs co -d project-master master
git-cvsserver uses one database per Git head (i.e. CVS module) to store information about the repository to maintain consistent CVS revision numbers. The database needs to be updated (i.e. written to) after every commit.
If the commit is done directly by using git (as opposed to using git-cvsserver) the update will need to happen on the next repository access by git-cvsserver, independent of access method and requested operation.
That means that even if you offer only read access (e.g. by using the pserver method), git-cvsserver should have write access to the database to work reliably (otherwise you need to make sure that the database is up to date any time git-cvsserver is executed).
By default it uses SQLite databases in the Git directory, named gitcvs.<module_name>.sqlite. Note that the SQLite backend creates temporary files in the same directory as the database file on write so it might not be enough to grant the users using git-cvsserver write access to the database file without granting them write access to the directory, too.
The database cannot be reliably regenerated in a consistent form after the branch it is tracking has changed. Example: For merged branches, git-cvsserver only tracks one branch of development, and after a git merge an incrementally updated database may track a different branch than a database regenerated from scratch, causing inconsistent CVS revision numbers. git-cvsserver has no way of knowing which branch it would have picked if it had been run incrementally pre-merge. So if you have to fully or partially (from old backup) regenerate the database, you should be suspicious of pre-existing CVS sandboxes.
You can configure the database backend with the following configuration variables:
git-cvsserver uses the Perl DBI module. Please also read its documentation if changing these variables, especially about DBI->connect().
All variables can also be set per access method, see above.
In dbDriver and dbUser you can use the following variables:
These variables obviate the need for command-line options in some circumstances, allowing easier restricted usage through git-shell.
GIT_CVSSERVER_BASE_PATH takes the place of the argument to --base-path.
GIT_CVSSERVER_ROOT specifies a single-directory whitelist. The repository must still be configured to allow access through git-cvsserver, as described above.
When these environment variables are set, the corresponding command-line arguments may not be used.
To get a checkout with the Eclipse CVS client:
Protocol notes: If you are using anonymous access via pserver, just select that. Those using SSH access should choose the ext protocol, and configure ext access on the Preferences→Team→CVS→ExtConnection pane. Set CVS_SERVER to "git cvsserver". Note that password support is not good when using ext, you will definitely want to have SSH keys setup.
Alternatively, you can just use the non-standard extssh protocol that Eclipse offer. In that case CVS_SERVER is ignored, and you will have to replace the cvs utility on the server with git-cvsserver or manipulate your .bashrc so that calling cvs effectively calls git-cvsserver.
All the operations required for normal use are supported, including checkout, diff, status, update, log, add, remove, commit.
Most CVS command arguments that read CVS tags or revision numbers (typically -r) work, and also support any git refspec (tag, branch, commit ID, etc). However, CVS revision numbers for non-default branches are not well emulated, and cvs log does not show tags or branches at all. (Non-main-branch CVS revision numbers superficially resemble CVS revision numbers, but they actually encode a git commit ID directly, rather than represent the number of revisions since the branch point.)
Note that there are two ways to checkout a particular branch. As described elsewhere on this page, the "module" parameter of cvs checkout is interpreted as a branch name, and it becomes the main branch. It remains the main branch for a given sandbox even if you temporarily make another branch sticky with cvs update -r. Alternatively, the -r argument can indicate some other branch to actually checkout, even though the module is still the "main" branch. Tradeoffs (as currently implemented): Each new "module" creates a new database on disk with a history for the given module, and after the database is created, operations against that main branch are fast. Or alternatively, -r doesn’t take any extra disk space, but may be significantly slower for many operations, like cvs update.
If you want to refer to a git refspec that has characters that are not allowed by CVS, you have two options. First, it may just work to supply the git refspec directly to the appropriate CVS -r argument; some CVS clients don’t seem to do much sanity checking of the argument. Second, if that fails, you can use a special character escape mechanism that only uses characters that are valid in CVS tags. A sequence of 4 or 5 characters of the form (underscore ("_"), dash ("-"), one or two characters, and dash ("-")) can encode various characters based on the one or two letters: "s" for slash ("/"), "p" for period ("."), "u" for underscore ("_"), or two hexadecimal digits for any byte value at all (typically an ASCII number, or perhaps a part of a UTF-8 encoded character).
Legacy monitoring operations are not supported (edit, watch and related). Exports and tagging (tags and branches) are not supported at this stage.
By default the server leaves the -k mode blank for all files, which causes the CVS client to treat them as a text files, subject to end-of-line conversion on some platforms.
You can make the server use the end-of-line conversion attributes to set the -k modes for files by setting the gitcvs.usecrlfattr config variable. See gitattributes(5) for more information about end-of-line conversion.
Alternatively, if gitcvs.usecrlfattr config is not enabled or the attributes do not allow automatic detection for a filename, then the server uses the gitcvs.allBinary config for the default setting. If gitcvs.allBinary is set, then file not otherwise specified will default to -kb mode. Otherwise the -k mode is left blank. But if gitcvs.allBinary is set to "guess", then the correct -k mode will be guessed based on the contents of the file.
For best consistency with cvs, it is probably best to override the defaults by setting gitcvs.usecrlfattr to true, and gitcvs.allBinary to "guess".
git-cvsserver depends on DBD::SQLite.
Part of the git(1) suite