Coda Distributed File System

Coda is an advanced networked filesystem. It has been developed at CMU since 1987 by the systems group of M. Satyanarayanan in the SCS department.

About this repository

This repository combines, to a best effort, the history from the official CVS repositories of Coda, as well as the supporting LWP, RPC2 and RVM libraries. The CVS repositories at this point only useful as a historical reference and further development will only be committed to the Git repository. The supporting libraries can be found in their respective directories under lib-src/.

The reasons the supporting libraries and their history have been merged with the Coda source tree are twofold. They were originally part of the Coda source tree but were separated out as over time other non-Coda users of the libraries emerged. By now Coda has outlived all other external users and merging them back to consolidate the development history seemed to make more sense.

For the past several Coda releases there have been new releases of the LWP, RPC2 and RVM libraries. Coordinating building and packaging of a single source tree is more straightforward than building four different ones, especially when there is a specific build/install ordering dependency and a new build could break a currently installed version. The current combined tree will ensure that everything builds and installs with the same library versions.

BUILDING CODA

Dependencies

Coda requires a working C/C++ development environment with gcc, gcc-c++, autotools, libtool, automake, pkg-config, flex, and bison. We require development headers for readline, ncurses5, and optionally the lua5.1 library.

On Redhat/Fedora/CentOS systems

yum install gcc gcc-c++ autoconf automake libtool pkgconfig flex bison \
    readline-devel ncurses5-devel lua-devel clang

On Debian/Ubuntu and derived systems

apt-get install build-essential automake libtool pkg-config flex bison \
    libreadline-dev libncurses5-dev liblua5.1-0-dev clang-format-6.0 valgrind \
    python3-attr python3-setuptools

Build

./bootstrap.sh
./configure --prefix=/usr --with-lua
make
sudo make install

RUNNING CODA

Configuring the Coda client

You need a configuration file in /etc/coda/venus.conf. The quickest way to get started is to run sudo venus-setup which will copy an example file with a suggested configuration, it will also create the directories that Coda uses for the local file cache and logs.

The first time the Coda client is started we need to force it to clean and initialize its 'recoverable virtual memory' (persistent state), this is done by running sudo venus -init.

Running the Coda client

The Coda client needs access to the file system services provided by the 'coda' kernel module, so to successfully start the client we have to make sure this module is loaded first. Once the kernel module is available, the client should be able to start and will automatically start running in the background as soon as it successfully mounts the file system on /coda.

sudo modprobe coda
sudo venus

Initially nothing will be visible under the /coda mountpoint, however any lookup for a Coda realm name, either through the file system, or through obtaining Coda credentials, will lead to a Coda server discovery request that, when successful, will make a permanent realm root appear.

$ ls /coda/testserver.coda.cs.cmu.edu
$ clog guest@testserver.coda.cs.cmu.edu
password: guest
$ ls /coda
testserver.coda.cs.cmu.edu

Because Coda persists its state in recoverable memory, the root will be visible even when Coda is restarted. To cleanly shut down the system, first unmount the file system and only then stop the daemon process, this way if any process has open references to files in Coda the unmount will fail and we avoid lost data.

sudo umount /coda && sudo killall venus

Configuring the Coda server

TBD

DEVELOPMENT

Pre-commit checks are run with the pre-commit framework, which is not automatically installed. You can install the pre-commit hooks with

pip3 install --user pre-commit
pre-commit install

It will automatically check and/or correct code-formatting and other issues every time a commit is made.

Although the source of LWP, RPC2, and RVM has been merged back into the main Coda repository, we are still trying to keep them mostly independent. This means that they retain their own build infrastructure and library versioning.

When updating supporting library sources make sure to properly follow the libtool library versioning guidelines. The version info is set using the CODA_LIBRARY_VERSION macro in configure.ac and consists of current:revision:age values which update according to the following basic rules.

If the library source code has changed at all since the last update, then increment revision (‘c:r:a’ becomes ‘c:r+1:a’)
If any interfaces have been added, removed, or changed since the last update, increment current, and set revision to 0.
If any interfaces have been added since the last public release, then increment age.
If any interfaces have been removed or changed since the last public release, then set age to 0.

Aside from this, we try to minimize overall version number changes so we skip updating if the corresponding version number change has already been applied since the last tagged stable release. This is different from libtool guidelines where they suggest updating only immediately before a public release, although in practice that is typically what happens.

Version updates are easily forgotten right before a release so it is better to update early, there is a helper script tools/version_check.sh that will show if there are any changes to the libraries. Developers don't have to care as much because they can run their binaries directly from the build tree in which case libtool will make sure the right library is used.

Tracing RVM transactions

Although RVM in general is (should be?) able to deal with multiple threads, the particular usage in the Coda client and server is single thread only. So there are a few strict requirements.

We must have an active RVM transaction whenever we modify data that was allocated in RVM and that we want to persist across application restarts.
We cannot begin a new transaction while there is an active one.
We should not yield our LWP thread during a transaction, because we don't know if another scheduled thread might need to start a new transaction.

Some of these constraints are now declared for function prototypes by using annotations REQUIRES_TRANSACTION, EXCLUDES_TRANSACTION, etc. These annotations can then be validated by using clang's thread safety analysis.

./configure CC=clang CXX=clang++
make

There will be compiler warnings that indicate when functions or methods are called in the wrong context, when transactions are not terminated in all possible exit branches from a function, etc.

7lima/coda