[Linux-cluster] cluster-1.03.00

[David Teigland]

A new source tarball from the STABLE branch has been released; it builds
and runs on 2.6.17:

  ftp://sources.redhat.com/pub/cluster/releases/cluster-1.03.00.tar.gz

Version 1.03.00 - 16 August 2006
================================
  ccsd
   * don't grab random config from network, require initial cluster.conf
   * Fix inifinite loop causing hangs in other daemons. bz#194361
  cman
   * Allow zero votes.
  cman-kernel
   * Don't try to delete AUTODELETE barriers in timer context as we can't
     get the semaphore that protects the structures. bz#177577
   * If quit_threads gets set via an incoming message, don't try to carry on.
     bz#164535
   * Don't try to print local node stuff is 'us' is NULL. bz#189605
   * Clear comms sequence number for a node when it leaves the cluster. 
     Otherwise we ignore messages when it tries to join again and causes
     cluster mayhem. bz#187777
   * If we get a Master-HELLO and we are not the master for this transition
     then kick off a new one to resolve the ambiguity. bz#194491
  dlm-kernel
   * Don't try to unlock a lock if there's no LKB. bz#188525
   * We need to allocated space for 5 ints, rather than 4 when sending a
     query reply. bz#173811
   * add printk's for the error conditions so we have some idea what
     happened before a gfs panic.
   * Kernel Oops when passing LKF_CANCEL to dlm_ls_unlock_wait.  bz#201325
  fenced
   * If there are no devices defined within a node's method, that method
     should be considered failed. bz#190661
  gfs_fsck
   * improve logging. bz#156009
   * fix to repair damaged and corrupt resource groups and resource group
     index entries that previously caused gfs_fsck to abort. bz#179069
  gfs-kernel
   * allow cman nodeid to be used with CDPN. bz#198381
   * When releasing a glock with GL_NOCACHE flag set, care was not taken to
     ensure that only one holder for the glock remained. This was corrupting
     the glock and preventing further access to the glock. FLOCKS use this
     GL_NOCACHE flag. bz#191222
   * F_GETLK was broken, always used to return zero conflicts for local
     plocks. Also bogus pid was being returned for local locks.  Added a new
     pid field to gfs posix lock to store and return actual pid.  bz#198303
  gnbd
   * Make gnbd work with device mapper multipath
   * Fix gnbd_monitor so that it will correctly restart multiple devices per
     server.
  gulm
   * Retry if initial connections attempts fail.  bz#183507
  rgmanager
   * Work around for bz#193128
   * Enable self-watchdog support (adds a second clurgmgrd process),
     bz#193247
   * Apply patch to fix bz#193128 from Navid Sheikhol-Eslami
   * Allow failover if owner dies while stopping a service.  bz#193255
   * Fix various clustat related usability & performance problems.
     bz#185952, bz#175010, bz#182454, bz#190234, bz#190408, bz#192999
   * Port clumanager's 'lock' operation to rgmanager bz#175010
   * Various internal performance improvements for large numbers of
     services.
   * Implement crude NFS lock reclaim broadcast / reclaim notifications.
   * Mark services with autostart=0 as 'disabled' instead of 'stopped'.
   * Add patch from Josef Whiter to implement no-failback option for a given
     FO domain. bz#189841