changes

一套客户/服务器模式的备份系统代码,跨平台,支持linux,AIX, IRIX, FreeBSD, Digital Unix (OSF1), Solaris and HP-UX.

3.3.3pl4 - 3.3.3pl5
	- Bug fixed: The current tape blocksize must not change during online
	  server reconfiguration i.e. after a cartready command or similar
	- Bug fixed: Default lockfile was not set correctly after online
	  server reconfiguration

3.3.3 - 3.3.4
	- Option -e (eject) for cart_ctl
	- Faster cleanup on client interruption
	- Full installation support for use with xinetd
	- Option -F plus format string for afrestore to print in user-defined
	  format and possibly containing file attributes present in the index
	- Details display also available in xafrestore GUI
	- New tape position file format, needed for future handling of
	  boxes with several streamer devices
	- cartis is no longer a script, yet another link to label_tape
	- Even safer cleanup on client interruption
	- Option -l[a] for afverify to list available backups from the indexes
	- Cleverer memory allocation strategy for names of files to restore to
	  be able to hold more of them in memory
	- POSIX ACLs supported
	- OSF-1 (and derivatives) ACLs supported
	- ACLs supported on IRIX
	- When an index has moved to another client, it is no longer necessary
	  to supply the -W <clientid> option to afverify or afrestore, when
	  afrestore is used with files or filepatterns (i.e. not -aeEf)
	- Clientside process calling chain reworked -> faster startup, less
	  tape moving, easier handling of the minimum restore info, cleaner
	  implementation, cleaner error handling, but: if a backup fails, a
	  minimum restore info for this is nonetheless written to tape. But
	  the advangages outweigh this little oddity
	- print_errors removed. All errors go to the log and stderr, no longer
	  mixed into the index files
	- Emergency recovery with relocate now relocates also the var directory
	  and creates it, if not existing
	- Bug fixed: After an online server reconfiguration (e.g. following a
	  cartready command) with directory pseudotapes it might happen, that
	  the saving location of a few files was wrong in the index
	- Bug fixed: afrestore -a with -C stopped prematurely in some rare
	  cases, when there were incremantal backups with no saved files

3.3.4 - 3.3.4pl1
	- Bug fixed: Statistics output was wrong: wrong start and end tape
	  position reported. Sideeffect: Use of cartridge 0 (not valid)
	  reported and cartridge 0 protected from overwrite on the server.
	  Mainly confusing, but not a really bad bug

3.3.4pl1 - 3.3.4pl2
	- Weird problem fixed: The multi stream server failed to demultiplex
	  the data to clients when reading during interrupt check, but only
	  with many clients saved in parallel, only on linux and only when
	  compiled with -O2. Worked around it, but didn't really understand
	  the problem. Suspect a libc error in getpeername

3.3.4 - 3.3.5
	- When encountering files of type "door" (cf. Solaris), do not throw
	  warnings and do not back them up
	- Remove old index files optionally by age, not by number of existing
	- Optionally scan a maximum configured number of index files, possibly
	  by age
	- Use also builtin paths to locate configuration and other directories,
	  even with ORIG_DEFAULTS defined. Thus the full path is no longer
	  necessary in /etc/inetd.conf for argv[0] of the server entries
	- Use /dev/(u)random, if available, for random numbers
	- POSIX threads availability also recognized on FreeBSD and the like
	- Optionally check, if the user has the access permissions to restore
	  his own files, also in the case, that no relocation is done
	- cart_ctl -l filters by changer device, if option -d is given
	- Option to prefer cartridges available in a changer. Default is to
	  cycle through all tapes independently from their current location
	- Authentication key file can be configured individually for
	  clients or groups of clients e.g. by command filtering
	from a patch of Peter 'Rattacresh' Backes:
	- Added --disable-threads option to configure.in and a corresponding
	  question to Install.
	- Fixed some questionable syntax combinations in Install like backspace
	  at the end of a string literal, a double quote ended by a single one
	  and a strange method of checking for the exit code

3.3.5 - 3.3.5.1
	- Install included libintl with x_utils, if required

3.3.5 - 3.3.5.1pl1
	- Avoid bogus messages: 'Wrong tape X accepted in drive for -1' and
	  'Using cartridge N instead of the desired (N-1)'
	- Bug fixed: server might crash or overwrite tape 1 without checking
	  that the tape is free, after requesting to free tapes manually

3.3.5.1pl1 - 3.3.5.1pl2
	- Fix for 'Avoid bogus messages ...' in pl1 was bad

3.3.5.1pl2 - 3.3.5.1pl3
	- Memory allocation bug fixed in full_backup

3.3.5.1 - 3.3.6
	- Mostly RCSed
	- Modified implementation of excluding files and directories. Now
	  also toplevel subentries can be excluded writing ./whatever
	- In addition to the mtime also the ctime can be used to select files
	  for backup during incremental or level X backup. Then the atimes
	  will not be preserved during backup
	- Installation cleanup. The configuration files are now in choosable
	  configdir, what is by default .../etc. A serverside logdir and a
	  clientside confdir has been added. Upgrade will automatically move
	  the configuration files to the new place
	- Troubleshooting HowTo added
	- Fixed, that the client side might hang due to unexpected errors
	  on the server side during startup
	- Fixed: If the first thing done after a fresh installation was a
	  differential backup (?!?), the index count erroneously started with 0
	- Cosmetical bug fixed: In the afrestore output "Found N matches on X
	  servers, Y cartridges ..." X and Y were more phantasy than sth. real
	- Insufficiency fixed: When the backup device had changed, the writing
	  position was reported incorrectly at the first next client request
	- If backup is started with files and directories to backup as command
	  line arguments, don't save them in the save_entries file
	- Do not store the clientside information about crucial tapes, if
	  passing this info to the server fails (avoids confusing message)
	- When finding files with l-bit set (if supported by platform), try
	  to set read lock, then backup on success, otherwise throw warning
	- Backward compatibility with servers < version 3.2 was broken 
	- In/exclude filesystems in/from backup by type name
	- Flag -H for incr_backup to release previous backups with a lower
	  level or the previous differential backup in differential mode
	- full_backup and friends can run without a configuration file applying
	  defaults, what is not desirable in most cases except emergency i.e.
	  disaster recovery starting afrestore with option -E or -e ...
	- During disaster recovery, if an index file is directly restored from
	  backup thus colliding with the one created by afrestore, the former
	  one will be renamed and a warning issued, so a maintainer can choose,
	  which one to use further
	- Option -L for af(m)server to set a locale

3.3.6 - 3.3.6pl1:
	- Bug in the find algorithm fixed, that lead to segfault

3.3.6pl1 - 3.3.6pl2
	- Bug fixed: restore as normal user might segfault (thanks to Norbert
	  Eicker at the University of Wuppertal for analysing the problem)

3.3.6pl2 - 3.3.6pl3
	- Workaround for drand48 on SunOS-4 extended, cause it is even heavier
	  broken than assumed before
	- Bug (?) fixed: on Solaris-2 sometimes it seems impossible to find out
	  the filesystem type of a file (observed by 'customer', but not
	  reproducible on Solaris-8 systems available to AF)
	- Output of cart_ctl was wrong when requesting a manual move from a
	  location outside a changer to a drive

3.3.6pl3 - 3.3.6pl4
	- xafserverstatus did not work on new installations

3.3.6pl4 - 3.3.6pl5
	- Issue 'Fixed, that the client side might hang ...' was not really
	  fixed completely
	- Modifications for debian package incorporated

3.3.6pl5 - 3.3.6pl6
	- Skipping filesystems by type did not work in combination with other
	  restricting criteria e.g. time constraints e.g. incremental backup

3.3.6 - 3.3.7
	- Apply (configurable) watermarks to the fill rate of the tape buffer
	  during writing or reading to decide whether to read/write or pause
	- Buffer size configurable
	- Buffer can be a file (mapped into address space)
	- Tested with libcrypto from openssl instead of the libdes, hints added
	- afmserver did not set keepalive mode (client did nonetheless)
	- Extended check for erroneous set{u|g}id installation of client-side
	  programs
	- NetBSD supported
	- Workaround for strange nonstandard fcntl semantics on certain (?)
	  versions (?) of f...... HP-UX probably lacking patches (?). Don't
	  buy HP
	- Interrupt algorithm changed for {full|incr}_backup: optional fast
	  termination on SIGTERM, SIGABRT, SIGQUIT (Ctrl-\) or several Ctrl-C
	- Bug fixed: after labeling a used tape and if the current writing
	  position was on that tape, the writing position was reset incorrectly
	- Bug fixed: When removing a client or ALL it's reserved tapes from the
	  precious_tapes file on the server, the indexes were not cleaned up
	- Faster algorithm to find duplicate entries during restore, should
	  lead to notable speedup when restoring > 100000 filesystem entries
	- Multi stream server more verbose on errors, especially demultiplexing
	  format errors
	- According to the wish of a lot of people afbackup is now installing
	  without creating hard links, but symbolic links instead

3.3.7 - 3.3.7pl1
	- Bug fixed: server might crash after manual tape change

3.3.7pl1 - 3.3.7pl2
	- Bug fixed: With variable append mode turned on, when a cartridge
	  from another set was requested, but a different one inserted, the
	  client's idea of the current tape on the server was temporarily wrong
	- Bug fixed: full_backup, incr_backup, afrestore and afverify did
	  segfault before exiting with bad status, when an index file could
	  not be uncompressed

3.3.7pl2 - 3.3.7pl3
	- Throughput optimization disabled because of suspicions, that it
	  corrupts data in rare cases
	- Compiler optimization configurable interactively running Install
	- Fix for FreeBSD-4.7 because with this release the SA driver has
	  changed it's behaviour at end of tape (sigh)

3.3.7pl3 - 3.3.7pl4
	- Determine filesystems exactly regardless of duplicate device numbers
	- Always set tape writing position to file 1, when tape is reused
	- New configure flag --without-strip to disable stripping binaries
	- Directory for temporary files configurable --with-tmpdir=/what/ever
	- Fixes for UnixWare-5
	- Newer ACL implementations on Linux require -lattr during linking

3.3.7pl4 - 3.3.7pl5
	- Throughput optimizer fixed and reenabled (was disabled in pl3).
	  Numerical subtilities lead to about one corrupted block of data of
	  a size 12 - 96 bytes among 500 Meg data
	- On interrupt the server now kills subprocesses (e.g. mt commands)
	  with signal KILL
	- Bug fixed: Under certain circumstances the server had the wrong
	  idea, what tape is currently in the drive. Detected and fix suggested
	  by Vlad Solopchenko (thanks a lot)
	- Install script extended to possibly add the afbackup services entries
	  as alias names
	- postinstall/preremove/checkinstall script, pkginfo and a map file
	  for Solaris-2 packaging added to the distribution, see HOWTO Q 36
	- HOWTO, FAQ and DO-DONT are separate files now
	- files > 2 GB supported on HPUX

3.3.7pl5 - 3.3.7pl6
	- When afrestore -e has read the minimum restore information and no
	  data matching the client id has been found, no appropriate output
	  had been produced, but a senseless 'No Memory'
	- Making backup into a filesystem on changable media without a media
	  being initially available and inserted later the server did not
	  recognize it sometimes (? observed by a collegue, but didn't look
	  at the particular installation and was not able to reproduce it)
	- Configuration with backup to filesystem and no cartridge handler
	  (e.g. manual change e.g. for setting CDROM) supported
	- Slightly more detailed mail, for what a requested cartridge is needed
	- Bug fixed: if PreferCartInChanger was set, the configured cartridge
	  sets were ignored, when selecting the next cartridge when the current
	  one got full
	- Reworked HOWTO Q 7 about how to recover from disaster using
	  statically linked afbackup programs
	- Script simple_tape_barcodes.pl added from Toivo Pedaste implementing
	  a simple handling of changer device readable tape labels
	- New option -r for cart_ctl equivalent to label_tape. Now cartridges
	  are not automatically marked overwritable when labeled with option
	  -t, this has to be specified explicitely using option -r. This was
	  a feature request: it should be possible to label other tapes then
	  the productive ones with the same label numbers without freeing the
	  data on the productive tapes and without asking interactively

3.3.7 - 3.3.8
	- Modifications in the debian part contributed by the Debian afbackup
	  maintainer Torsten Werner
	- Insufficiency fixed: Restore might have segfaulted on corrupt index
	- Option -t for afrestore to only list, what tapes are required for
	  restore. This is helpful especially in the situation, that one wants
	  to know in advance, which tapes to put into a changer before actually
	  starting the restore
	- Messages from the server, that up to now were only emailed by the
	  server can now be output on the client side, if desired, configurable
	  by operation mode (backup/verify/restore/copy_tape)
	- Bug reading the minimum restore info from stdin fixed (Thanks to
	  Arthur Korn)
	- No more /tmp/afbsp_XXX files left over in /tmp after backup. Was
	  necessary before to keep verify from complaining
	- Install is now a sh script, no longer csh
	- Calling afrestore with -A and/or -B did often use too many old tapes
	  for restore unnecessarily
	- Should be IPv6 safe (can do only minimum testing unfortunately)
	- Build fixes for various operating systems
	- Should be safe to be built as 64 bit binaries, at least with gcc -m64
	- Fixes in autoconf to use gcc as preprocessor, if available
	- Several type and header issues for better portability
	- Debian upstream fix incorporated

3.3.8 - 3.3.8pl1
	- Autoconf modified to detect libacl and libattr more reliably
	- Autoconf modified to be able to build even in quite strange
	  environments by overriding system specifics (not documented)

3.3.8pl1 - 3.3.8pl2
	- Fixes in named data code, currently used only in other software, not
	  in afbackup

3.3.8pl2 - 3.3.8pl3
	- Bug in tcp multiplexer code fixed (not sure, if it's really a bug in
	  this code, could as well be considered a bug in some libc or header
	  implementations) leading to segfault of afmserver under certain
	  circumstances
	- Bug in full_backup.c putting all lines output by a subprocess into
	  one single line fixed

3.3.8pl3 - 3.3.8.1
	- Bug fixed: verify failed / hung in an endless loop, if during backup
	  with only external compression configured a file attempted to be
	  saved was not readable. With the fix still 'format error in input'
	  errors might occur during verify of backups made with the unfixed
	  software as the bug produced an invalid format to be written, what
	  can hardly be workarounded, but can be considered an uncritical, just
	  annoying warning
	- Bug fixed (why the hell did i do it this way ?): afrestore -E over
	  the multi stream port stopped after printing out the found minimum
	  restore info(s) instead of starting to restore
	- Tape device can be configured as - to prevent the server from logging
	  a bogus warning, when it is used on clients only for remote start
	- Bug fixed: Having UseCTime configured symlinks might not be written
	  to the index
	- Bug fixed: implementation error in setting the effective UIDs/GIDs,
	  caused by a man page error in an older glibc, no impact on afbackup,
	  only other software using the utilities library