Start coding better reservation algorithm

[bacula/bacula] / bacula / kes-1.37

              Technical notes on version 1.37  
                        Kern Sibbald

General:

Changes to 1.37.36:
17Aug05
- Start coding better reservation algorithm
- Always look for slot for label command.
- Add more debug code for autochangers.
- Apply fix from Stephan Leemburg <sleemburg@jvc.nl> for
  improper scanning of schedule resource:
  Run = Level=Full Pool=Catalog daily at 1:20
- Apply patch from Chris Lee <labmonkey42@gmail.com> for 
  adding --enable-build-dird --enable-build-stored.
- Tweak datadir definition in configure.in
16Aug05
- Fix bug that missed drive=nn specification.
- Eliminate nonblocking kludge in heartbeat of FD as
  it caused high CPU usage.
- Pickup loaded slot when doing open() of tape drive.
- Make autochanger reservation code go through full list
  first pass until exact match found.
15Aug05
- Fix how FileSet is saved in job record to correct continual  
  Full save.
- Make datadir print on config.out listing.
- Move get FileSet record up in backup init to eliminate continual
  Full save seen by Peter Sjoberg.
- Add VolumeName to read-only Python variables.
- Add VolumePurged event for Python.
- Suppress /dev/ Filesystem change prohibited INFO messages.
- Do not delete FT_RAW files before restore (allows FIFOs
  to be used for restore).
13Aug05
- Add drive specification to mount, unmount, release, label,
  and relabel for Autochangers.  Note Dir<->SD protocol has
  changed.

Changes to 1.37.35:
12Aug05
- Disable parts of NLS as the configure does not work here.
- In job backup init (backup.c) define definitive Job level and
  since time, *then* apply Job Pool override selection if any.  
11Aug05
- Modified bconsole script so that it is improperly
  installed, it will refuse to execute.  This avoids
  recursive call loops.

Changes to 1.37.34:
06Aug05
- Apply David's ACL fix to src/filed/acl.c
05Aug05
- Apply patches sent by David Duchscher <kreios@gmail.com> for
  making ACLs work on MacOS X and FreeBSD.
04Aug05
- Apply patch in bug#397 that improved configure
  - find readline under $with_readline/include/readline
  - no libutil under Solaris
  - no need for -ldl under Solaris 
- Make reservation system single threaded during the
  search to avoid two threads competing for the same
  resource.
- Correct a return code in find_suitable_device_for_job()
  Possibly cause of "busy writing to another volume".
03Aug05
- Modify open() for tape so nonblocking really works.  
- Use fcntl() to reset blocking status rather than close()
  and reopen the drive.
- Make sure dev->open() is always called so that any change
  in read/write permissions will occur.
- Open drives initially in daemon in read-only mode.
- Ensure that each time the VolHdr.VolumeName is zapped
  or changed that free_volume() is called on the old name.

Changes to 1.37.33:
03Aug05
- Require 5 arguments to mtx-changer except list and slots
- Turn -EPIPE status returns from bpipe to ETIME
- Include Slot in SD status output 
- Do not term_dev() during initialization in SD if the device
  could not be opened.  In the case of a tape drive, there may
  be no tape in the drive.

Changes to 1.32.32:
02Aug05
- Correct PostgreSQL database scripts as suggested by a user.
- Add additional info to FATAL message generated when a device
  is busy writing to another volume.
- Suppress an inappropriate NULL Volume name message after a cancel.
- Correct a warning message in reserve.c
29Jul05
- Apply user's patch to make mutiple modifiers for times
  work correctly.
- Make read_dev_volume_label() handle ANSI/IBM labels
  correctly -- ie space over any label at the beginning
  of the tape.
28Jul05
- Make ANSI/IBM writing of HDR1/2 labels ignore any
  errors if at end of tape.
- Apply Martin's patch to improve Python detection in
  configure.in
- Temporarily turn off disk seeking until I find the
  cause of the problem.
27Jul05
- Add OSF1 patch supplied by user.
- Use number of files selected from write_bsr() only
  if it is not defined.
- Explicitly seek to end of file when getting size for
  restore test.
- Correctly set EndBlock position in JobMedia record
  for files.
- Remove unnecessary set StartBlock in bscan. Caused bscan
  regression error.
26Jul05
- Modify mtx-changer to wait a maximum of 300 seconds.
- Do restart of failed jobs only for Backups job types.
- A number of DVD updates from Nicolas.
24Jul05
- Turn off old service helper code in Win32.
- Correct Messages bug found by Phil in stored.
23Jul05
- Complete (almost) documentation of 1.38.
- Add error messages for error conditions with VSS.
- Fix additional problems with VSS backup that I introduced.
Changes to 1.37.31:
22Jul05
- Correct compiler complaints in wx-console and tray-monitor.
- Correct VSS problems recognizing c:
- Add VSS before job status
- Fix output of status from being one big line.
- Change cd xx; make to cd xx && make as suggested by Phil.
- Cleanup projects file
- Remove unnecessary casting of FF_PKT in filed.
- Apply Thorsten's bugfix for vss_generic.cpp
- Add check for df path for dvd_freespace
- Use df to get space used on DVD.
- Change sense of flag indicating erase DVD or not before writing.
- Fix bpipe so that it never modifies the result pointer.
- Replace more dev_name by print_name().
- Rewrite edit_device_codes_dev() so it does not overwrite the
  supplied buffer.
- Update printing of labels (for bls) so that critical information
  is printed (Job name and timestamp).
- Cleanup old spool files when starting the SD.
- Modify vss.cpp to allow C: as a path name.  Otherwise VSS
  doesn't work when only a drive name is given.
- Modify vss.c to eliminate double / in filenames.
- Update doc -- particularly the restore chapter.
18Jul05
- Make all files in working directory have .xxx at end.
- Work on DVD writing.
- Fix keepatime bug (bugs database).
- Move Python variables from Job to Bacula. They are
  DirName, Version, ConfigFile, and WorkingDir
- Fix delete of bootstrap to only occur on Bacula created
  filenames.
- Allow cancelling a Job name that is not active. It is
  sent to the daemons.
17Jul05
- Fix name space pollution by OpenSSL 0.9.8 reported by
  Matthias Kurz -- applied his patch.
- Fix bpipe.c so that it does not modify results pointer.
  ***FIXME*** calling sequence should be changed.
- Remove some remaining references to dev_name.
- Fix calls to mount_dev() and unmount_dev() to
  correspond to returned value (bool instead of int).
- Try without success to make DVD writing work.

Changes to 1.37.30 released 16 July 05:
14Jul05
- Fix "dir" command scanning field misalignment in
  wx-console.
- Switch to using the wxWidgets Unicode library.
- Include msvcr71.dll in distribution.
- Add VSS to status line in Win32 FD if enabled.
- Get VSS build scripts working with Thorsten's help.
- Unlink the bootstrap file after sending it to
  the FD.
- Remove sending include/exclude lists to the FD during
  a restore -- deprecated code.
- Cleanup the bootstrap files in the FD.

Changes to 1.37.29:
14Jul05
- Remove old commented out code from configure.in
- Add baculavssfd.mak file for building VSS version of
  Win32 FD.
- Correct date (year) in vss.cpp and vss_generic.cpp
13Jul05
- I finally found and squashed the elusive SD crash.        
  I needed to initialize the used volume list before
  firing off the device initialization thread.
Changes to 1.37.28:
11Jul05
- Make sure that bpipe results are zapped even on
  error return.
- Lots of documentation.
- Do not prune volume marked as append when needing a
  new Volume.
- Print a warning message in SD if a non-used Volume
  is specified and autolabel not turned on.
- Correct a bug in chksum.c concerning SHA1 signatures 
  (an * should have been & when checking for a bit flag).
- Print File:Block for all label records in label.c -- concerns
  primarily bls when doing Job listings (-j).
- Correct is_volume_in_use() to return false if testing
  on the same device where the Volume is already mounted.
- Define a init_done flag in the SD that is set when the
  devices are initialized and make users connecting wait.
  This prevents useless connect failure warning messages.
- Do additional device locking in ask_op_to_mount_volume()
  to prevent race conditions with a user labeling a Volume
  or autolabeling.
09Jul05
- Add a test for error return from bnet_wait... in heartbeat.c
  in FD to avoid CPU loop.
- Implement TLS in gnome console and wx-console.
08Jul05
- Correct a NULL pointer reference in the mount command.
- Correct typo in Copyright
- Add detection of EOM for IBM drives (i.e. errno == ENOSPC)
07Jul05
- Remove temp file created in mtx-changer script.
- Make fsf_dev() into a class method.
06Jul05
- Modify mtx-changer.in script to return slot:barcode for
  Volumes that are loaded in the drives.
- Correct some more places where dev->is_blocked() needs
  to be checked in dircmd.c in SD.
- Update doc.
05Jul05
- Add code to ensure that reserved but unused volumes
  are freed. 
- Correct how Volumes are mounted and handled so that the SD
  does not get stuck if multiple volumes are used (recycling,
  relabling, ...)
- Correct bug where you could relabel a volume while it
  was being acquired -- created chaos.
04Jul05
- Correct seg fault caused by open() calling sequence change.
03Jul05
- Add new rc-chio-changer script by Rudolf Cejka to 
  examples/autochangers
- Apply Rudolf's changes to bacula.in
- Expand the space from 8 to 10 characters in editing
  file sizes for restore and dir of catalog, otherwise
  GB sizes are truncated -- fixes bug report.
- Modify wx-console to know about 10 character widths.
- Allow decending into top level directory if "recurse=no"
  is set.  Fixes a bug report.
- Install pthreadVCE.dll when installing console or wx-console
  on Win32 systems. Fixes bug report.
02Jul05
- Tweak dvd-writepart script to prevent door from opening/closing
  so much.
- Remove GROUP BY in several PostgreSQL commands to prevent error.
  Resolves bug report.
- Ensure that < as first character of filename list is not treated
  as a directory for restore.
- Add debug to heartbeat in FD as it seems to go into an
  infinite loop from time to time during SD failure in DVD writing.
- Add more debug code to dvd writing.
- Attempt not to destroy existing fs on DVD. 
30Jun05
- Detect device mounted for DVD and suppress be sure to 
  mount message after label.
- Set Cleaning tape status to "Cleaning" and force no
  MediaType.
- Get DVD writing working with new standard Bacula open()
  code.
- Rename get_filename() to make more sense.
- Detect "is already mounted on" on mount command so to avoid
  error if device is already mounted.
- Eliminated guess_name() code. It may be necessary to
  add it back later.
- Eliminate seg fault from printing invalid results.
- Make dvd_write_part() bool.

29Jun05
- Attempt to fix DVD writing by eliminating a number of the 
  DVD subroutines to simplify.
- Modify DEVICE::open() to take dcr as first argument. This
  will permit providing more info to DVD opening.
- Fix scanning for time/size items which in some cases
  ate the next line.
- Eliminate read_dvd_volume_label().  New code (not yet written)
  *must* open dvd appropriately before calling 
  read_dev_volume_label.
- Modify open_first_part() open_next_part() to take DCR as 
  argument.
- Make label command from console work on DVDs.
- Make mount command from console work on DVDs.
  Unmount does not work yet.

Changes to 1.37.27:
27Jun05
- Add Database vendor to CatalogRes tuple for Python.
- Update doc
- Implement DoesVolumeExist(Vol) for Python.
- Prevent python command from seg faulting if no arg given.

Changes to 1.37.26:
26Jun05
- Add set_mode method in DEVICE.
- Correct set_mode method in DEVICE
- Add more DVD debug info 
23Jun05
- Check for incorrect duration and size modifiers in conf files.
22Jun05:
- Make Version a tuple (version, build-date)
- Add CatalogRes tuple (DBName, Address, User, Password,
    Socket, Port)
- Add Version, ConfigFile, and WorkingDir as Python attributes
  in the Director.
- Implement code (principally for Win32) that on failure to
  create a file, it will cd into the directory and attempt
  to create the file using a relative path. This avoids creating
  files with paths which fail on Win32.
- Fix parsing of times and sizes with decimal numbers.
- Make free_volume_list() in SD work if vol list is not
  initialized (./bacula-sd -t).
21Jun05:
- Add debug error printout when open() fails.
- If open() of DVD fails in mount.c, return false.
- Split open() code for DVD into separate subroutine in dev.c

Changes to 1.37.25 released on 20 Jun 05:
20Jun05:
- Remove (cd gnome-console; make clean) from src/Makefile.in
- Fix bug where Storage daemon gets confused about what
  tape is mounted. (one line of code was inadvertently 
  deleted).

Changes to 1.37.24:
18Jun05
- DVD writing/reading seems to be mostly working.
- Set execute bits on dvd-freespace and dvd-writepart
- Make dvd-freespace use existing dummy file.
- Modify dvd-freespace to pickup size from Track Size:
16Jun05
- Add Date, Job, level to updates to .bsr file in 
  dird/backup.c
- Add debug info to dvd-freespace.in
- Fix hard coded bacula.sql in make_catalog_backup reported
  by a user.
- Make sure a verify volume to catalog never reports an error
  if there are zero files to verify.
- Remove confusing debug info in filed/backup.c on network
  error.
- Make sure output from console is not sent to system log.
- Convert open_dev() into a class method.
- Change VolHdr.VolName to VolHdr.VolumeName.
- Add a flag in the device state word to indicate that
  we found Media in the drive (DVD). 
- Make mount_dev() and unmount_dev() return bool.

Changes to 1.37.22:
14Jun05
- Fix the same state variable problem in open_next_part().
- C++ify the SD code a bit more.
13Jun05
- Add more debug code and clarify debug code for DVDs.
- Do not save and restore state in open_first_part() 
  since state should be properly set after open_dev().
- Make default mandir /usr/share/man
- Install Bacula man page.
- Implement passing of FileSet Enable VSS to FD.
- Move main body of reserve drive code into subroutine
  so that it can be called multiple times.
12Jun05
- Eliminate getpass() for Win32 builds in console as the
  function does not exist.
10Jun05
- Correct some reservation problems in SD when no devices
  are available.
- Start removing #ifdef HAVE_TLS by sneaky tricks.
- Begin implementation of TLS in wx-console
- Remove ignoring SIGCHLD from console.
- Rework the dlist binary search routines for implemenation
  of the Volume reservation code -- make it more general.
- Strip double slashes // from Win32 filenames in an attempt
  to resolve restore problems on some systems.
- Fix a minor bugs in the trace code that caused the first
  line output to be lost.
- Implement a good first cut at adding Volume reservation code
  to the storage daemon (in file reserve.c).
- Remove old unused code from the tree.c routines.

Changes to 1.37.21:
06Jun05
- Fix compile problems on Win32
- Start writing Volume reservation list (already exists, but
  is not really very good).
- Implement attribute caching to put Signature into database     
  at the same time as the file attributes thus eliminating a
  number of database accesses.
- Correct a reservation problem.
- Implement full Dir Storage use.
- Reduce a bit of TLS #ifdeffing.

Changes to 1.37.20:
04Jun05
- Minor changes
01Jun05
- Add more documentation to mtx-changer.in
- Correct link to manual in authenticate.c in various
  directories.
- Create a new src/stored/reserve.c file where the 
  Use Storage command is processed and drives are
  reserved.
- Modify src/stored/autochanger.c to keep track of each
  Slot that is loaded for each device.
- Ensure that changer_command and changer_name are picked
  up from Autochanger resource if not specified, and if
  neither is specified, err.
30May05
- Fix bextract.c compile problem
- Create bacula.man 
- Make make distclean clean a bit better
29May05
- Remove old code in jcr.c
- Make testls release jcr chain when terminating.
27May05
- Implement Maximum Job Spool Size (actually DCR based)
26May05
- Use light weight non-recursive locking on jcr chain.
- Make JCR a class and implement inc_use_count() and
  dec_use_count() methods that ensure that the jcr is
  locked when inc/dec the use count.
- Remove the global jcr lock when traversing the jcr
  chain.   
- Use dlist to implement the jcr chain rather than hand
  crafted next and prev links.
- Lock the jcr chain inside each function that modifies
  the chain.

Changes to 1.37.19:
26May05
- Fix compile problem of ua_restore.c on broken compilers.
- Apply patch from bug 326 to permit bacula status by any user.
- Fix bug 325 -- conversion of 12:30pm to 24hour time.
25May05
- Put Dmsg() on inside if() to avoid calling subroutine.
- Make restore.bsr have unique name.
- Allow user to define bsr filename on restore command line
  with bootstrap=xxx.bsr
- Add limit=nnn to "list jobs" command.
- Remove old restore code that did not use .bsr file.
- unlink automatically generated bsr file.
- Cleanup heartbeat code so that duped fd is almost sure
  to be released. Previously under certain conditions, the
  memory was not released due to race conditions.
- Shorten copyright.
20May05
- Unify the reserve_device() for a single device into one subroutine.
18May05
- Modify wait during use_device to happen only after all devices
  have been examined rather than in the reserve_device code.  
- Correct updating count of number of Volumes in a pool.

Changes to 1.37.18:
16May05
- Add more debug to SD for Autochangers + status output.
- Add Scratch to PoolType in PostgreSQL make...tables and do not
  permit NULL PoolTypes. Fix for bug 319 reported by Eric.
- Update LICENSE.
- Add quotes around filename in parse_config error message. Bug
  reported by Eric.
15May05
- Change nested \include to \input so that sections are properly
  included in the pdf manual -- update the Web site.
- Set reconnect flag in MySQL packet to 1 to ensure that connection
  is re-established.  MySQL 5 changed default to 0. Fixes bug report.
- Fix Scratch pool handling as reported in a bug by Eric Bollengier
  by applying his patch.
- Remove delete job in favor of delete jobid.
- Add = NULL to configfile definitions as reported by Eric in a bug
  report.
- Update winbacula.nsi.in to reflect new manual file structure.
10May05
- Correct a minor build problem with wx-console.
- Add cancel() to Dir Python scripting.
- Re-correct bug in parse-config error handling.
- Reorganization of use_command in SD to permit
  waiting and multiple drive autochanger support.
09May05
- Correct bug in parse_config error handling.
- Where ever possible mark a volume in error or not   
  InChanger in mount.c
- Fix bug in changing tape pools after first backup. Reported
  by Peter Sjoberg.
- Enhance mtx-changer to use Working Directory as temp.
- Remove all but initial setup locking of Res in SD.
08May05
- Add Client OS type to Job report.
- Add version to manual 
- Update the Web site to have a single page for
  the documentation links.

Changes to 1.37.18 release 08May05:
08May05
- Correct attribute definition compile error in 1.37.17
- Correct inverted order of CreateTime and MD5 pointed out
  by a user in a bug report in FileSet db routine causing
  the MD5 to print in the output instead of the time/date.

Changes to 1.37.17:
07May05
- Implement cstrlen() in sql list routines.
- Implement caching of attributes to add the        
  signature so that only one DB call will be made
  per file. Not yet turned on.
- Fix Win32 build for TLS.
- Optimize File pruning to eliminate one database call. 
- Fix bug that prevented File pruning from working.
- Implement a cstrlen() which returns the character
  length of a UTF-8 string.
06May05
- Move test for MaxStartDelay as suggested by Peter.
- Implement Python methods (I had to read the Python source
  code).
- Implement run() method in Director.
- Add Priority and Scheduled time to Job report.
- Add JobInit and JobRun events.
- Add Priority as Python read/write attribute to Job.
- Correct typo in bsmtp reported by Jo.

Changes to 1.37.16 released 05May05
03May05
- Make a few tls ifdef tweaks.
- Fix create_file.c Win32 problem pointed out by
  Peter Sjoberg.
- Fix really ugly bstrncpy() but found by Thorsten.
- Move winapi.h/c from findlib to lib for inclusion in 
  multiple places.
02May05
- Thorsten Engel finished his work on Win32 Unicode. We
  now have a single executable that runs on all Win32 machines.
- Move job initialization code after job scheduling so that
  Verify jobs check for the prior JobId after they are really
  started rather than before.
- Fix lib/fnmatch.c so that it does proper testing before folding.
- More documentation -- at tls and ansi labels chapters.
- Fix fileset_convert.pl to handle empty Exclude statements.
- Turn regex back off in Win32
01May05
- Fix sign extension problem in lex.c that reads UTF-8 
  with Chinese characters incorrectly.

Changes to 1.37.14:
30Apr05
- Remove a few HAVE_TLS #ifdefs
- Implement final Python style interface. More implemention to
  be done, but the interface should change little if at all.
28Apr05
- Make default no tls support. You must add 
  --with-openssl to get tls support.
27Apr05
- Update Web header to include google search in the search
  box -- thanks to input from Michel Meyers.
- Fix md5sum so that it builds with the new openssl stuff.
- Take some enhancements to the md5sum test program proposed
  by a user.
26Apr05
- Make option 2 of restore work better (list last 20
  locations for a file).
- Make SD re-open a device with the right permissions if
  it was previously opened with something different.  This
  should allow reading read-only Volumes under all circumstances.
- Implement restore of a single directory.
- Apply Tru64 patch supplied by Pascal Pederiva <freebsd@paped.com>
- Apply Unicode fixes for Win32 from "Thorsten Engel" 
  <thorsten.engel@matrix-computer.com>
- More work on Python read feature in FD.
22Apr05
- Fix (hopefully) the ftello() overflow reported by Peter.
- Landon Fuller committed his TLS patch.
- Fixed two minor warnings in console.c with TLS turned off.
- Updated the Makefile.in for wx-windows and tray-monitor
  to handle TLS. I can only build the tray-monitor.
- Fix bscan to open tape in read-only mode (actually fix
  it so that it doesn't use the standard open routine).
- Correct what appears to be an error in setting the
  return value in dvd.c
21Apr05
- Get FD Python running -- design Python backup interface.
- Fix seg fault in SD when referencing Alert Command.
- More documentation.
- Fix one more thing in Win32 build.
20Apr05
- Doc updates
- Fix Win32 build
- Put in production here

Changes to 1.37.13:
19Apr05
- Fix SQLite and PostgreSQL table creation script syntax
  problems.
- Fix new Python code to work for Director.
- Move lib/python.c to lib/pythonlib.c so that debug output
  is easier to read (can distinguish lib from dird, ...).
- Cleanup Python build so that Python is not dragged
  into programs that don't use it.

Changes to 1.37.12:
18Apr05
- Make Bacula build without Python
- Drop Status table in drop_ scripts -- bug 283.
17Apr05
- First cut of new Python implementation.
- Doc updates
- Correct SuSE autostart installation directory name --
  supplied by a user.
- Fix director crash if Name directive not supplied
  in a Job.
- More cleanup of changing %u to %s for DB IDs.
- Replace dev_can_write(dev) with dev->can_write().
11Apr05
- Make fsr_dev() a method dev->fsr()
- Remove Created new FileSet message as it always comes out in
  the daemon messages.
- eliminate dev_is_tape() in favor of method dev->is_tape()
- Turn on disk seeking during restore for Disk Volumes. This
  required some changes to bscan -- they seem a bit suspect as
  they are not symmetrical for tape/files.
- Remove some of the recursion in stored/parse_bsr.c
- Move rescue out to a new Bacula project.
08Apr05
- Update of web site to replace old released doc with LaTeX 
  version.
- Replace logo in doc with new bat logo.
- Make a good number of updates to the manual.
- Implement Python scripting using a Python object.
  This makes interfacing with Bacula cleaner (object
  oriented).
- Add Phil's Status table to the database -- used
  for getting a text description of the status     
  codes.
- Modify FileSet so that an empty Options (no
  pattern matching) with an exclude will exclude
  all files.                       
- Modify FileSet so that no top level included File
  item can be excluded by a matching pattern.
- Suppress Created new FileSet record in daemon output.
- Implement Python in FD.
- Turn off old bfile reader code -- to be replaced
  by Python.
05Apr05
- Remove more recursion in src/dird/bsr.c as pointed out
  by Peter.
04Apr05
- Increase the index width of Filename and Path entries
  to 255 chars -- suggestion of Meno Abels.
- Change remaining VARCHARS to TINYBLOBs.
- Remove recursion from is_volume_selected() following
  bug report from Peter.
- Implement mostly correct handling of use_storage in the
  SD with full lists, ...
02Apr05
- Reset NumVols in Pool record from database on every update
  Pool.
- Modify DB to support multiple simultaneous copies and
  RAIT stiping.
- Pass copy and stripe between DIR and SD and put into
  the JobMedia DB record.
- Update and test SQLite and MySQL datebase creation and
  update scripts.
- Implement version 9 of the DB.
31Mar05
- Convert more atoi to str_to_int64() for DB.
- Implement filling in NumVols by querying DB rather
  than trying to keep track of it.
- Add storage name to string passed to in use storage=
- Fix newVolume() so that the Python script is always
  called.
- Fix handling of pool,PoolId, and storage in ua_output.
- Same fix in ua_status.c
- Remove required locking of resources
- Replace pthread_cond_signal() by pthread_cond_broadcast()
  hoping to fix the /lib/tls hang problems (lost signal).
- Move resource locking seaching from parse_conf.c to res.c
  in src/lib.
- Modify end of volume handling so that fixup_... does not
  redo what block.c has already done -- writing Vol info to
  DIR. This fixes a bug with bad numbers of files on a tape
  when it filled as reported by Peter.
- In release_device() do not update the DIR on the Volume
  info if the the information was already written at the
  end of the tape.
28Mar05
- NOTE!!!! This version has a new DIR <--> SD protocol. Both 
  must be updated at the same time.
- Begin implementation of passing all the Storage and Device
  possibilities to the SD for examination during the reserve 
  phase.
- Modify the reserve and acquire code in the SD to make a 
  job wait if the device is not available.
- Implement New Volume Each Job in DIR and pass to SD, not yet
  used.
- Remove init/update of the Device resource in DIR
- Remove passing PoolId to SD and back.
26Mar05
- Remove \a and -e from error echos in most Makefiles.
- Add more debug code when there are errors on the tape 
  to try to find Peter's tape problem.
- Add wait.c (oops forgot previously).
- Move all the reserve/acquire_device_for_read/append to have
  only a DCR as the argument.
- Rework the reserve_device_for_append() in stored to wait
  if the drive is not available.  Note! This is a short
  term solution.
25Mar05
- Comment out Multiple Connections in the document.
- Move the P() and V() to subroutines so that they can be accessed
  from class methods. The reference to strerror() caused problems.
- Implement new DEVICE class methods block() and unblock() that
  do what was previously done in 3 lines of code.
- Implement wait_for_device(), which will wait for any device
  to be released then return. This requires a new global mutex
  and condition variable, and is implemented in src/stored/wait.c
- Change the code in reserve_device_for_read(), which previously
  failed the job to use the new device wait code.
22Mar05
- Apply reschedule patch to 1.37 code.
- Add copyright to title page of manual so it is clear.
- Create patch for rescheduling problem found by Ludovic. Storage
  pointers were lost during rescheduling.
- Attempt to fix 2.6 rescue disk -- failed!
- Start working on adding a wait routine in the SD.
- Cleanup some old invalid doc in watchdog.
- Convert a number of references to dev->dev_name to dev->print_name().
- Add new wait.c file to SD.
- Add a few more methods to DEVICE in SD to cleanup code a
  bit -- implement a few of the methods.
18Mar05
- Fix more print_name()s for printing device name.
- Modify open_dev to try 10 times every 6 seconds to
  open the device if it gets an I/O error (meaning no
  volume mounted). This gives a bit of settling in time
  for an autochanger and avoids spurious messages.
- Change all yes/no to yes|no in the manual.
- Fix win32 create_file.c typo.
- Fix a typo in an error message.
17Mar05
- Detect if fseeko exists with autoconf. If so, use it 
  and ftello.
- Remove old bacula-*.conf from examples directory (out
  of date).
- Remove latex-fr index files from CVS.
- Rewrite code that stops reading the tape so that the
  tape is marked at EOT, then once the work is done,
  the EOT flag is removed.
- Flush output to file after every send in console.
- Make setting VolFiles to smaller number fatal.
- Disable Multiple Connections code.
- Add patch from user for NetBSD statvsfs() fix to
  fstype.c
- Take more care with errors in acquire.c
- Don't run through dvd code in append.c if bad status
  returned.
- Modify code so that an autochanger fault is fatal.
- Use dev->print_name() in more places.
- Implement dev->can_steal_lock() to simplify code.
- Make btape re-read first 10000 records on fill command.
- Check error return and fail job from fseeko and ftello
  in spool.c.  Don't let a -1 slip in as size.

Changes to 1.37.7:
15Mar05
- Apply NetBSD patch from kardel in bug 258.
14Mar05
- Add a second job and a second client to the default
  bacula-dir.conf file.
- Remove old style Include/Excludes.
- Fix ANSI labels to put EOF1 and EOF2 after each file mark.
- Add Python to SD and FD.
12Mar05
- Implement IBM labels
- Implement EOF and EOV labels at the end of a volume.
- Fix a rather ugly problem with the PoolId not getting
  passed correctly. Now the DIR passes the Pool name and
  Media Type to the SD, who passes them back when requesting
  the next Volume. The DIR then looks up the correct PoolId.
  This takes more time, but always works, AND allows wild
  card Media Types (i.e. the SD can decide).
- The DIR <==> SD protocol has changed.

Changes to 1.37.6:
11Mar05
- Fix scanf of PoolId in catreq to handle 64 bit Ids.
10Mar05
- Add new ua_update.c file and move update_cmd there.
- Modify "update slots" to obtain actual number of slots.
- Tweak autochanger code to handle new slots request.
- Modify autochanger code to lock/unlock around slots and
  update slots code.
09Mar05
- Patch the FD so that it does not issue an error message if
  it attempts to restore the permissions on a Win32 drive.
- Edit "Resource-name" (physical-name) for the device name
  everywhere in the SD.
- Remove .linked.tex files in preparation for cutover to
  using .tex in place of .wml.
08Mar05
- Copy latest config.sub and config.guess from autoconf.
- Try new way of identifying drives with:
   "resource-name" (physical-name)
  More work need to a complete conversion.
07Mar05
- Rework some of the autochanger data so that the DIR has   
  the number of drives.
- Modify the way the Device info is returned so that it comes
  back as a special message type and can be sent anytime the
  Device status changes.
- Copy the change name and changer command into the device
  record if none is specified.
- Require the change command and changer name to be specified in
  and AutoChanger resource.
- Force all the Media Type records of all devices in an Autochanger
  to be the same.
06Mar05
- Add new "run" command to Job resource in DIR. This permits
  cloning a job as many times as you want.
- Pass PoolId to SD on Query request. It is now used in the
  Find_media catalog request.
- Reworked the Device resource in the DIR. Eliminated num_waiting
  and use_count, but added max_writers, reserved, and PoolId.
- This DIR is nolonger compatible with previous SDs.
- Add since and cloned keywords to the Console run command
  to support cloning.
- Implemented store_alist_str() to allow multiple string items
  to be specified in a .conf file.
- Added %s (since time) to Job code editing.
- Reworked reserving drives in the SD. It now does it much simpler
  and correctly. 
05Mar05
- Integrate HP-UX patch from  Olivier Mehani <olivier.mehani@linbox.com>
- Fix FD job.c to test correctly for no level.

Changes to 1.37.4:
04Mar05
- Change Developers to Developer's Guide as requested by Michael.
- Fix developers link in manual
- Add additional dcr changes in SD to allow multiple dcrs.
02Mar05
- Fix a few problems with the MySQL table create in 1.37.
- Delete the new tables in the table delete files.
- Increase the number of items permitted in a conf table.
- Make Director loop over alternative Devices specified in the
  Storage resource until one is reserved by SD.
- Fix storing of StorageId in Media records.
- Add AutoSelect = yes|no in bacula-sd.conf
- Add Autochanger support to Label command.
- Do not autoselect devices with autoselect set false 
01Mar05
- Implement setting DIR Storage device to Autochanger
  name. 
- Select first available device in Autochanger.
- Pass back actual device name used.
- Allow Query of AutoChanger.
- Modify Query to include name of AutoChanger if 
  Device belongs to one.
- Remove old Pool code in jobq.c
- Add Autoselect flag to query and DEVICE class (still
  need Directive).
28Feb05
- Lock autochanger script when running.
- Mark Volume not InChanger if correct volume is not     
  autoloaded.
- Corrected some typos in the make_xxx_tables.in files.
- Made preliminary split of pre-run and run code for each
  job type. This will permit early opening of SD for reserving
  drives.
- Add offline and autochanger fields to Device Query record.
- Correct pthread_mutex_init() for autochanger in SD.
- Tweak Makefile for LaTeX manual, plus add nav buttons.
26Feb05
- Clean up drive reservation system. Add more sanity checks.
- Implement a few more methods for the DEVICE class in SD.
- Add latex directories to make clean
- move DEV_BSIZE to B_DEV_BSIZE to avoid conflicts with
  certain header files (FreeBSD).
24Feb05
- Fix an ASSERT that was triggering in stored/acquire.c
  attempt to fix a bug report.
23Feb05
- Corrected SunOs to SunOS in btraceback (user submitted).
- Applied patch from Roger Håkansson <hson@ludd.luth.se>
  to warn the user of defective AWKs during ./configure.
20Feb05
- Add some changes submitted by a user for HP client build.
  Not all changes accepted.
- Rework code in filed/backup.c to ease #ifdefing and make
  program flow more obvious.
- Split DVD code out of dev.c into dvd.c
- Tweak #ifdefing to add back all the performance measurement
  #defines in version.h
- Put most of MTIOCGET code in a subroutine to simplify the
  mainline code.
- Make clean remove old CVS files
- Remove unnecessary image files from Latex directory
- Implement remaining parts of Storage DB record and
  its use in the Director.
- Implement
  FullMaxWaitTime, Differential Max Wait Time, and               
  Incremental Max Wait time in Job resource.
- Start work on SD Autochanger code.
19Feb05
- Add back JobId index for MySQL as default -- speeds up 
  pruning.
- Add more database fields and fix the update scripts to
  include the new items.
- Pass actual level to FD so that ClientRun editing can reflect
  correct level -- ditto for job status. This makes the DIR
  incompatible with older clients!
- Move jobq.c acquire resources to static subroutine so that
  the code logic becomes clearer. This is in preparation for
  actually using the new Device resources.
- Fix some lower case problems in sql_cmds.c reported by
  Debian.
- Correct a seg fault in the SD reported by a user. Occurred
  only when a high debug level was set.
- Modify init_dev() in dev.c to take JCR as first arg so that
  proper error messages can be reported in next item.
- Modify the query and use device SD commands to attempt to
  open the device if it could not previously be opened.
- Correct error message for Could not reserve device.
- Correct some minor details with Autochanger resource in SD.
18Feb05
- Fix seg fault if debug level 900 set in SD.
- Truncate Win32 child return code to 8 bits.
- Remove some old lld's.

Changes to 1.37.3:
16Feb05
- Make another attempt at fixing the ClientRunXXX return code 
  bug on Win32 machines.
- Apply ua_status patch from Carsten Paeth <calle@calle.in-berlin.de>
  which enforces console ACLs in the status command for Jobs.
15Feb05
- Fix Media LabelDate and FirstWritten to be correctly set.
- Fix deadlock in multiple simultaneous jobs.
- Fix tape "truncation"/"number of files" after restore bug.
10Feb05
- Ensure that correct error messages are returned when
  reading an ANSI label.
09Feb05
- Modified ANSI label code to preserve any ANSI label 
  already found by skipping over it rather than rewriting
  it.
- Split the ANSI label code into ansi_label.c
- Do not let user relabel an ANSI labeled tape.
- Applied a patch for the console help command supplied
  in a bug report. 
- Added some new dev methods. Most notably was
  set_eof(), which handles setting all the dev variables
  when an EOF is just read. This is now used most everywhere
  in the code.
07Feb05
- Added code to detect that no files were inserted into the 
  tree for a restore. If a specific JobId was specified, the
  user has the option of restoring everything.
- More progress in implementing 64 bit DB Ids.
- Modified the daemon start messages for RH.
- Implement update scripts for all database types.
- First cut at implementing restore directory (it will not
  recurse).
04Feb05
- OK, I think ANSI labels work.
- Added Label Type = ANSI|IBM|Bacula to Device resource in SD.
  If this is set, it will force writing of the appropriate
  label type.
- Added Check Labels = yes|no to Device resource in SD. If this
  is set, Bacula will check for ANSI labels and accept them,      
  otherwise, ANSI labels will not be accepted when the tape
  is first mounted.
02Feb05
- Second cut ANSI labels.
01Feb05
- Merge Preben's patch for ACLs and for Mac OS X resource forks.
- Some doc updates.
- Display more informative message when a device was not
  found or could not be opened.
- Add the sqlite3 database scripts. 
- Add some patches for 1.36.1 (note, I have now prepared
  a 1.36.2 with all the patches and some new features --
  to be documented).
- Some minor doc updates.
- Add Arno's baculareport.pl script to the examples directory.
29Jan05 -- after vacation
- Add support for SQLite3 (it seems to run at 1/2 the speed
  of SQLite2).  Use --with-sqlite3 instead of --with-sqlite
  to get SQLite3.
- Add target for running qemu to boot Rescue CDROM
- Add code to support kernel 2.6 in Rescue CDROM -- does NOT yet
  boot correctly.
- Implement ANSI labels -- not yet tested.
  This required changes to DB format. No upgrade script yet.
  Note, more work needed to modify "update" command to handle
  changing label types, also must restrict volume name lengths
  to 6 characters.
- Add new Device, Storage, and MediaType records to DB. No
  upgrade script yet.
- Add MediaType to bsr file record types. Not yet used in SD.
- Permit multiple device specifications in Storage resource in 
  Dir conf file.
- Implement Device resources. Director requests Device resource
  info from SD on startup. 
- Note!!!! DIR->SD incompatible with previous versions.
- Remove multiple Storage definitions in Job resource.  One can
  still specify multiple Storage resources, but they all go into
  a single alist, and imply sending data to each Storage daemon 
  simultaneously.
- Implement Device query command between DIR and SD.
- Allow DIR to "reserve" a Device. It will then be acquired
  when the FD connects to the SD.
- Turn all DIR resources into classes, and implement a few class
  methods -- more to come.
- Turn DEVICE in SD into a class, and implement a number of inline
  class methods -- more to come.
- I had serious problems with ACL errors on my Laptop, and so had
  to add the following patch:
  @@ -181,7 +181,7 @@
      }
      /***** Do we really want to silently ignore errors from acl_get_file
        and acl_to_text?  *****/
  -   return -1;
  +   return 0;
   }
- Added edit_int64()
- Reworked and tested a bit the htable routines.
- Major changes to SD acquire.c -- DIR can now reserve devices. Needs
  lots of testing!!!!
- Made a special state code for DVD -- this simplifies the logic
  of the code, but I probably broke it. Testing needed!!!!
- Add AutoChanger resource to SD, but not yet used.

Changes to 1.37.2:
12Jan05
- Integrate Preben 'Peppe' Guldberg <peppe@wielders.org>'s
  acl patch. Fix case where configured but no ACL exists.
  Rework calling arguments to be shorter and positioned
  more typically in Bacula usage.
11Jan05
- Fix scripts/bacula.in to have awk on an environment variable
  and add comments for Solaris users.
- Turn off inet_aton in src/lib/address_conf.c for Win32
- Add new files to win32 build and eliminate a compiler warning.
- Add sample DVD Device resource to bacula-sd.conf
08Jan05
- Integrate Nicolas' patch for direct DVD support.        
07Jan05
- Fix fstype error returns.
- Apply Preben's cleanup.patch which puts back much of the
  cleanup code in src/filed/restore.c
06Jan05
- Apply all of Preben's patches, but revert to old backup.c
  and old restore.c in filed. Also turn off code in new
  acl.c because of errors.  The new code, when fully implemented
  moves platform specific code into acl.c.
  One of the patches also implements WildFile and WildDir -- thanks.
01Jan05
- Implement Python in the SD (no events yet though).
- Fix some typos in the previous commit.
30Dec04
- Enhance CDROM boot to include some documentation at boot time.
- NOTE!!!!! The CDROM will not boot 2.6 kernels because the
    boot sequence has changed significantly. Updates to come
    later.
- Add memtest option to CDROM boot.
- Include Nicolas' changes to fix llist JobMedia records.
- Make sure that ClientRunBefore/After messages from the program
  are terminated with a newline. Add strerror to output error
  messages.
- Return program exit status code in Win32.
29Dec04
- Add memtest86 to Bacula Rescue disk
- Enhance Rescue disk startup screen
24Dec04
- Move some variables to eliminate Solaris 2.6 compiler warnings.
- Fix the seg fault at the end of a job in the FD when using
  old style include/excludes.
22Dec04
- Apply Preben's ACL patch.
- Integrate Preben's restore patch.
- Integrate Preben's verify teaks.
- Fix doc/latex/Makefile to copy/remove .eps files when building
  html and web outputs.
21Dec04
- Fix Bacula so that it does not exit if there is a syntax error
  in its conf file during a reload command. Bug 182.
- Apply fixes suggested for old Solaris networking.
  Fixes bug 190.
- Apply Preben 'Peppe' Guldberg <peppe@wielders.org>
  three patches that clean up white space:
  ws.patch.02.strings:
    Breaks strings that span lines into concatenated strings. I am not sure
    if you like this one. Other code works with concatenated strings, though.
  ws.patch.03.trailing:
    This removes trailing whitespace. No changes resulted from this for
    my setup.
  ws.patch.04.leading:
    This replaces space runs at the start of line with tabs. No changes
    again.
- Fix overriding storage specification to be done
  through a subroutine.
- Fix autoconf so it runs with FC3. 
- Add Python4.3 to configure search paths.
- Always copy and delete storage definitions into jcr.
- Check that VolumeName supplied by Python is valid. 
  Return 0 if not.
19Dec04
- Fix undefined in non-Python build.
- Update rescue disk to include mkinitrd
- Fix umount_drives in rescue disk (only one arg to umount)
- Ensure that if SD is manually set in Console, it is used.
- Put generate_event on pointer and plug it in init. This
  permits using it in /lib
- Correct despooling size reported to be Job specific rather
  than for the whole drive.
18Dec04
- Fix bug 207. jcr use count off by one when manually
  scheduling jobs.
- Remove FNMATCH test in configure.in and always use
  the one in our library to get the FN_CASEFOLD GNU
  extensions on all platforms.
- While using the rescue CDROM after my computer would not
  boot, I realized that it would be very useful to have
  a umount_disks. So, it is not implemented, along with
  updates to the READMEs and some minor tweaks.
- Moved mounting the CDROM in the rescue boot from /cdrom
  to /mnt/cdrom (more standard location).
- Reboot in CDROM rescue should now work -- requires -d
  option (no write) to work.
- Hopefully fixed all the IPV6/4 problems and buffer
  problems with networking in lib. Bugs 190 and 204.
  Cleaned up a lot of #ifdefing problems by using routines
  in address_conf.c
17Dec04
- Apply Preben 'Peppe' Guldberg <peppe@wielders.org>
  alist fix patch.
- Remove duplicate code from chksum.h (mentioned by Preben).
13Dec04
- Integrate Tim Oberfoell <oberfoell@web.de> patch to ACLs
  to handle both the "standard" and "default" ACLs.
12Dec04
- Integrated Preben 'Peppe' Guldberg <peppe@wielders.org>
  three cleanup patches (btest, verify, find).
- Integrated Preben 'Peppe' Guldberg <peppe@wielders.org>
  three cleanup patches (backup, chksum, and verify)
09Dec04
- Integrated Preben 'Peppe' Guldberg <peppe@wielders.org>
  patch to avoid doing MTIOCGET on OSes that do not support
  it such as OpenBSD.
- Integrated Preben 'Peppe' Guldberg <peppe@wielders.org>
  patch to add filesystem type matching to FileSets in the
  Options resource.
- Integrated Preben 'Peppe' Guldberg <peppe@wielders.org>
  patch to add Mac OSX resource fork support (save/restore)
  to Bacula -- HFS Plus support.
- Add FileSet to client Job listing query.
06Dec04
- Integrated Preben 'Peppe' Guldberg <peppe@wielders.org>
  patch to backup directories skipped (due to no file system
  changes or no recursion), and to add a slash to the end
  of the directory name during the match process.
- Implement Jamie ffolliott <jamieff@inline.net>
  patch to dird_conf.c that enables Multiple Connections and
  fixes a typo in show. The rest of his patch awaits my suggested
  changes.
05Dec04 
- Implement run command in Python
04Dec04 
- Implement conversion of the manual, and some minor
  tweaks to the script tags.
- Apply a patch supplied by Preben 'Peppe' Guldberg that implements
  ignore case in wild cards and regexes.
- Fix a truncated line in the above patch due to my cut and paste.
03Dec04
- Fix it so that the InChanger flag is only changed for Volumes
  in the same Pool.
- Add PIDOF configuration path and apply to bacula.in
- Add user supplied patch to add inet_aton() of old Solaris
  systems.
- Require pools to match before allowing multiple simultaneous
  accesses to same storage resource.
- Add patch supplied by Martin to correct buffer overrun in
  bsnprintf() with no library snprintf().
02Dec04
- Apply user supplied patch that implements No Hard Links.
- Document Python interface
- Add hardlink keyword patch supplied by David R Bosso <dbosso@lsit.ucsb.edu>
01Dec04
- Fix non-python prototypes in dummy routines.
- Add python 2.3 to config search list (user submitted patch)
- Add JobStatus to Python variables.
28Nov04
- Add "python restart" command in Console.
- Make built-in variables table driven.
- First cut of Python Events for Bacula. Director only.
  StartJob, EndJob, NewVolume events.

Version 1.36.1 Released 26Nov04:
24Nov04
- Take Dan's fix to the fix_postgresql_tables (thanks Dan)
- Increase Maximum Concurrent Jobs to 20 in SD and FD!
- Fix improper handling of autochanger Volumes that are not
  marked InChanger. Bug 170.
22Nov04
- Update authentication failure message to indicate possible
  problem with Maximum Concurrent Jobs. Add to doc too.
- Commit PostgreSQL BIGINT fix -- apparently some people didn't
  get it.
- Do not use a slot unless it is InChanger in the autochanger
  code, otherwise autochanger gets upset not finding the Volume.
  Bug 170 (need fix above too).
21Nov04
- Make authentication timeout compile time configurable.
  The value is set in src/baconfig.h
- Fix removing items from watchdog queue, which apparently
  screwed up if there was more than one item.
- Rework mediaformat part of manual to separate out old
  tape format for easier reading.
- Add a lot of debug code to the authentication code.
- Add seconds to the start/end times printed in the job report.
19Nov04
- Fix mutex deadlock in dequeue of messages.
- Add debug code to FD authorization.
18Nov04
- Increase authentication timeouts to 10 minutes.  
15Nov04
- Fix cancel bug in FD on /lib/tls with zero pid in 
  pthread_kill.
- Add date/time to all messages.
- Make Qmsg use time message was queued rather than time
  printed.
- Indent job output two spaces.
13Nov04
- Fix web page links for new manual.
- Grant postgresql permission to cdimages.
- Correct crash after "list nextvol" "list media" bug 160
12Nov04
- Fix scripts/Makefile.in missing ;\  -- thanks Martin
- A bit of work on btape to keep if from going into infinite
  loops when things do not work well, and to print a bit
  better info.
11Nov04
- JobDefs Storage resource completely overrode anything
  specified in Job resource -- bug 159 -- fixed.
- Fix syntax of renaming postgresql per Dan.
- Add working_directory to be /tmp for wx_console.
10Nov04
- Allow both a JobId and a filename or list of files to be
  specified on a restore command line.
- Save old mtx-changer before installing new one: bug 156
- Fix errors in CDROM file pointed out by Scott.
09Nov04
- Fix exepath when Bacula executed without path.
- Move test for socket libraries for Solaris in configure.in
  before tcp wrapper tests so that linking works.
- Add "make copy-static-fd" to makefile in CDROM file to
  copy existing static fd to CDROM rather than building it.
08Nov04
- More doc updates
- Fix ps command for OpenBSD
- Rework the creation of indexes for SQL -- fall back
  to the old code, but document what can be added.
06Nov04
- Add new Daemon message handler in default DIR conf.
05Nov04
- Fix Seg Fault with -D100 in bpipe.c.    
- Fix Seg Fault in run specifying a JobId.
- Make mail from daemon with a Messages Resource use
  the MailCommand with editing rather than the default
  sendmail.
- Replace Jmsg in dispatch_message() with Qmsg.
- Make edit_job_codes handle NULL jcr.
04Nov04
- Add M_ALERT class and put tape alerts into it.
- Fix Verify count vs found by not double counting files
  that are split across files/Volumes.
30Oct04
- Fix count returned from write_bsr_file() to handle multiple
  volumes. This fixes most cases of the Verify VolumeToCatalog.
- Cleanup a bit the make clean for the rescue cdrom and remove
  unneeded files from the CVS.
28Oct04
- Fixed acquiring a tape so that it does not block all acquires
  when operator intervention is needed.
- Platform build script updates from Scott
- Doc updates
- Add patch to force Linux LD_ASSUME_KERNEL to avoid using the
  new /lib/tls. This is done in the startup scripts.
- Modify mtx-changer so that it checks for ONLINE while
  waiting.
- Modify make_postgresql_tables.in so that EndBlock is stored
  as a bigint. Prevents job failures when a disk volume is
  larger than 2GB.
24Oct04
- Add grep ONLINE to wait_for_drive() in mtx-changer.in
- More doc.
- Rebuild Scott's new configure