OpenAFS for Debian

Introduction

  For an OpenAFS client system, install openafs-client and a kernel
  module.  See README.modules for information on how to build the kernel
  module for your system.  Then, read /etc/openafs/afs.conf to understand
  the client configuration options.  The openafs-client package will
  attempt to guess at a good cache configuration based on the size of your
  cache, but you may want to tune it further.  There are also other
  options configured in that file you may want to consider.

  FAM does not work correctly with AFS except for directories that are
  world-readable since it does not run in the user's security context and
  therefore doesn't have the user's AFS tokens.  If you are using FAM,
  you'll encounter errors from file managers such as Nautilus that use it
  if you browse restricted AFS directories.  Instead of FAM, install
  gamin, which runs in the user's security context and works correctly
  with AFS.

  For information on how to set up an OpenAFS server, read README.servers.
  You will want the openafs-fileserver package for a file server and,
  additionally, the openafs-dbserver package for a database server.

Documentation

  For the complete OpenAFS manual, install openafs-doc.  This is the same
  documentation as found at <http://www.openafs.org/>, and is
  unfortunately outdated in several respects, but it's the best that we
  have at present.

  If want to set up a new cell, read README.servers and then look at the
  example session in configuration-transcript.txt.gz in this directory.
  The procedure outlined in these two files is much simpler and more
  secure than the one in the OpenAFS documentation, but the OpenAFS
  documentation provides useful background.

Build Options

  The OpenAFS file server has been built with --enable-demand-attach-fs,
  which enables a new method of handling volume attachment.  Volumes will
  be attached only when needed, and the file server will persist state to
  disk when shutting down for faster restarts.  This file server mode is
  experimental and should not yet be used for production data.  Please
  report any bugs you encounter.

  The OpenAFS servers have been built with --enable-supergroups, which
  permits nesting of PTS groups.  Be aware that the PT database created by
  these packages is not compatible with servers not built with
  --enable-supergroups if nested PTS groups are used.  In other words, if
  you need the openafs-dbserver package to interoperate with ptservers
  that aren't built with this option, don't use this capability.

  The OpenAFS client and kernel module have been built with
  --enable-disconnected, which enables experimental support for
  disconnected operations.  This support is still experimental and known
  to have problems, but should not interfere with normal use of AFS.  If
  you want to experiment with it, see fs help discon.

Changes Relative to Stock OpenAFS

  Long-time AFS users may be confused by the directory layout.  The files
  that normally go in /usr/vice/etc go in /etc/openafs.  The cache should
  be mounted on /var/cache/openafs.  The server files have been moved
  around even more; see README.servers for the details.

  The OpenAFS kernel module is named openafs, not libafs, to better match
  normal Linux kernel module naming standards.  The Debian source package
  only builds one kernel module that matches the kernel source tree it is
  built against and does not attempt to build separate SMP and non-SMP
  modules against the same tree.  Doing so does not work on all platforms.
  To distinguish between an SMP and a non-SMP kernel module package, use
  --append_to_version; see README.modules for more information.

  The AFS up utility is installed as afs-up, since the standard name is
  rather generic.

  The libopenafs-dev package only includes static libraries and there are
  no shared library packages.  The shared libraries built by AFS are not
  compatible with Debian policy.  They do not have a stable ABI or an
  appropriate SONAME.

  kaserver is not included.  New AFS cells should use Kerberos v5 rather
  than the old K4-based kaserver KDC.

Debugging and Bug Reporting

  The current OpenAFS installation process installs fileserver and
  volserver unstripped, since backtraces and other debugging information
  for those binaries are necessary to track down file server problems.
  For the Debian packages, the fileserver and volserver binaries in the
  openafs-fileserver package are stripped, but the debugging information
  is available in the openafs-dbg package, which can be installed
  separately.  If it is installed, gdb will find that debugging
  information automatically.

  Eventually the openafs-dbg package will contain debugging information
  for all OpenAFS binaries.

  When reporting a bug in the OpenAFS client, please include your exact
  kernel version and architecture (reportbug will do this for you).  Also,
  if the client caused a kernel oops or BUG, be sure to include the
  complete kernel output, including the lines before the oops.  That's
  where the OpenAFS error message, if any, will be.

  When reporting a bug in the OpenAFS file server, please include
  backtrace information from a core dump, if any.  If the file server is
  deadlocked, you can capture a core dump using the gcore script that
  comes with the gdb package.  The file server is threaded, so use the
  command "thread apply all backtrace" in gdb to get a complete backtrace.
  It's also often useful to have the output of rxdebug <server> 7000 at
  the time of the problem and the FileLog from the file server.  You can
  increase the logging level of the file server with kill -TSTP (and reset
  it to 0 with kill -HUP).

  You can report any bug in OpenAFS against the Debian package with
  reportbug and the OpenAFS package maintainers will forward the bug
  upstream as necessary.  If you do want to report a bug directly
  upstream, see http://www.openafs.org/ for bug reporting instructions.

PAM Authentication

  Any new OpenAFS cell is strongly encouraged to use Kerberos v5 for
  authentication.  If you want PAM to automatically obtain AFS credentials
  and you are using Kerberos v5, you will want to install the libpam-krb5
  and libpam-afs-session packages and then put something like the
  following in /etc/pam.d/common-auth:

      auth    [success=ok default=1]  pam_krb5.so ignore_root
      auth    [default=done]          pam_afs_session.so
      auth    required                pam_unix.so nullok_secure try_first_pass

  and something like the following in /etc/pam.d/common-session:

      session optional        pam_krb5.so ignore_root
      session optional        pam_afs_session.so
      session required        pam_unix.so

  You'll probably also want the following in /etc/pam.d/common-account:

      account required        pam_krb5.so ignore_root
      account required        pam_unix.so

  There are, of course, many variations depending on what different
  mechanism you want to use and how you want to handle fallbacks.

  If you are still using Kerberos v4 and the OpenAFS kaserver (or a KDC
  that understands the same protocol) for authentication, you can instead
  use the libpam-openafs-kaserver package and a configuration like:

      auth    sufficient      pam_afs.so ignore_root
      auth    required        pam_unix.so nullok_secure try_first_pass

  in /etc/pam.d/common-auth and:

      session optional        pam_afs.so
      session required        pam_unix.so

  in /etc/pam.d/common-session.  Use pam_afs.krb.so instead of pam_afs.so
  if you also want the PAM module to acquire a ticket cache for you.  If
  using this configuration with sshd, you may need to disable privilege
  separation to get everything working properly.  I've had mixed results
  with that.  Obviously, converting to Kerberos v5 authentication is
  strongly preferred.

  If you are using the kaserver as your KDC, you may also want to install
  the openafs-kpasswd package to get the administrative utilities for
  managing those Kerberos accounts.

 -- Russ Allbery <rra@debian.org>, Tue,  6 Apr 2010 15:04:42 -0700