2000-11-22 Karla Sorenson (sorenson@missioncriticallinux.com>
	* Substantial documentation additions.  SCSI, FibreChannel,
	  configs without power switches.

2000-11-13 Greg Myrdal (myrdal@missioncriticallinux.com>
	* Added numerous checks for the validity of parameters specified
	  during service configuration.
	* Svcmgr memory leak fixes.
	* Merged Eric Ayers' patch for support of log-based file systems. 
	* Do not fail stop of service if user-defined script does not exit.  
	  This allows the user to put the script on a shared file system 
	  and not have the Service Manager initialization fail.    

2000-11-13 Tim Burke (burke@missioncriticallinux.com>
	Quorumd enhancements:
	* Fixed problem preventing quorumd from exiting in the case where there 
	were no services and SM would return the SM_EXITING message to quorumd 
	quickly.  Then quorumd would get upset that SM's pid was still around.
	Now with the fix, in this case quorumd will retry looking for the 
	exit of SM. 
	* Added no power switch mode.  If the power switch type is set to NONE,
        each loop through quorumd_body, first read in the node's status block 
        to see if the other cluster member has performed a hostile takeover 
        (and marked the node state as DOWN).  In this case initiate an 
        immediate system shutdown to minimize the data corruption 
        vounerability window.   
	* Ratcheted polling rate down to a 2 second interval (formerly 3).
	* In quorumd_stop, change call sending message to powerd to terminate 
          to have a timeout; to avoid the potential to hang indefinitely. 

2000-11-12 Richard Rabbat (rabbat@MissionCriticalLinux.Com>
	* member_config enhancements: 
	- reworded some prompts
	- added new text to ask whether one wants to use cluster w/ or w/o 		  power switches. if no pswitches, then script does not ask for info 
	  about them.  if somebody's upgrading, the previous choice will show, 
	  and if somebody wants to keep the pswitches to none, then he/she 
	  will get a small warning message       

2000-10-31 Ron Lawrence (lawrence@MissionCriticalLinux.Com>
	* cluadmin enhancements: reworded prompts, fixed memory leaks,
          additional checks on service configuration parameters. Changed
	  handling of services in the error state.
	
2000-10-26  Jeff Moyer  <moyer@mclinux.com>
	* Modified Makefile.am to allow specifying the directories where
          TCL and SWIG may reside.  (--with-tcl-includes=DIR, 
	  --with-tcl-libraries=DIR)

2000-09-28 Ron Lawrence (lawrence@MissionCriticalLinux.Com>
	* Enhancements to the CFG_ library routines:
	  - more extensive error checking
	  - new memory allocation scheme, fixed parseconf allocator leakage.
	  - multi-line entry support

2000-09-22 Greg Myrdal (myrdal@MissionCriticalLinux.Com>
	* Merged in Eric Ayers' patch for support of different file system
	  types in the Service Manager and cluadmin utility.  This now
	  allows the definition of log-based file systems in a service,
	  i.e. avoid running fsck.

2000-09-19  Tim Burke  <burke@mclinux.com>
	* Anne has new kernel patches for NFS support.
	* Ron enhanced a few cluadmin help messages and minor corrections
	  to display of service state.

2000-09-15  Jeff Moyer  <moyer@mclinux.com>
	* Greg added the purgenfs command, so that we can manually fail
	  over an nfs service w/o causing ESTALE on the client side.

2000-09-07  Dave Winchell <winchell@mclinux.com>
	* Fix problems in the shadow fixup code which could cause
	  corruption or indefinite recursion. 
	  Implement a new algorithm for fixing a corrupted lock.
	  Remove the try_lock in the low level shadow read code.
	  Perform fixup at the low level only if the read is not lock data
	  and the cluster lock is held on entry.
	  Add cluster lock held assertions in the write path for all data
	  structures except lock and node status.
	  Change the disk scan code to try to take the cluster lock for
	  each data structure.
	* Improve messages when corruption occurs.
	
2000-08-23  Tim Burke  <burke@mclinux.com>
 	* Put in extra checks in quorumd cluster stop sequence to insure that
2000-08-23  Tim Burke  <burke@mclinux.com>
 	* Put in extra checks in quorumd cluster stop sequence to insure that
	  svcmgr has exited prior to initiating a quorumd stop.

2000-08-23  Greg Myrdal  <myrdal@mclinux.com>
 	* Fixed a race condition where multiple services try to configure
   	  on the same instance of an network interface.
 	* Added invalidatebuffers command in the utils directory and make use
   	  of it after unmounting filesystems to avoid data corruption.  
 	* Fixed Service Manager exit condition where it would incorrectly exit
   	  leaving a service running on that system (details too long to
   	  elaborate here).
 	* Added try attempts to the Service Manager to contact the Quorum 
	  Daemon if it is not responding.

2000-08-17  Jeff Moyer  <moyer@mclinux.com>
	* Added debian package and rpm package support.

2000-08-08  Tim Burke	  <burke@mclinux.com>
        * 2 bug fixes addressing stack overrun.  1 typically apeared as
          clustat dumping core, the other as failures of daemons to 
	  communicate via messaging.
	* Set FD_CLOEXEC bit on listen file descriptors.  Previously, 
	  quorumd would not start due to some other processes having
	  its address bound.

2000-07-30  Ron Lawrence  <lawrence@mclinux.com>

	* cluadmin is now usable for creating and modifying services.
	* Removed prior documentation and notes describing how to create
	  services in the absence of cluadmin.
 
2000-07-14  Jeff Moyer  <moyer@mclinux.com>

	* Changed method for daemon startup.  Previously, the init.d
	  script took care of starting every daemon in turn.  Now, its
	  only job is to star the quorum daemon.  quorumd then determines
	  node membership, and if the node can access the shared storage
	  partition, it will fire off the other daemons.
	* Quorumd was rewritten pretty much from scratch.  This was mainly
	  for maintainability purposes, but also has a lot to do with the
	  restructuring needed for the authentication bits.
	* Authentication bits were added at the msgsvc layer.  This
	  required a change to the msg_receive apis.  Basically, the
	  msg_receive calls just take a result parameter, auth.  This is
	  filled in with a 1 or 0 depending on whether the remote end of
	  the connection has authenticated itself.  The authentication id
	  is stored on the shared storage, and is initialized by the
	  quorum daemon, once per full cluster reboot.
	* Cluster.cfg has been renamed to cluster.conf for uniformity
	  reasons.
	* -g is now turned on by default, whereas before it was only
	  turned on if you ran ./configure --enable-debug.

2000-06-25  Jeff Moyer  <moyer@mclinux.com>

	* First open source release of kimberlite.
