Use sys_clone under Linux to share I/O contexts between dump processes.

[dump.git] / CHANGES

$Id: CHANGES,v 1.300 2009/06/18 09:50:54 stelian Exp $

Changes between versions 0.4b41 and 0.4b42 (released xxxxxxxxxxxxxxx)
=====================================================================

1.	Fix printout of the 'Connection to' message in dump/restore.
	Thanks to Dale Wiles <dwiles@users.sourceforge.net> for the
	bug report and its fix.

2.	Fix dump -w/-W output for filesystems having the last backup
	level equal to 0. Thanks to Pascal Bouchareine
	<kalou@users.sourceforge.net> for reporting the bug.

3.	Enable restore to handle restoring onto a different SELinux
	policy from the dump. Thanks to Tony Nelson 
	<tony_nelson@users.sourceforge.net> for the patch.
	(see http://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=189845
	for details)

4.	Fix EA set failures when restoring immutable files. Thanks
	to Andrew Kroeger for the patch.

5.	Fix "mode file too large" errors when restoring huge backups.
	Thanks to Kenneth Porter <shiva@sewingwitch.com> for the bug
	report.

6.	Add the ability to link against libtinfo library if present,
	instead of libtermcap or libncurses libraries. Thanks to
	Adam Tkac <atkac@redhat.com> for the patch.

7.	When comparing a backup, do not consider to be an error the fact
	that we encounter files created while dump was in progress.
	Thanks to Kenneth Porter <shiva@sewingwitch.com> for the bug
	report.

8.	Force '-a' to be the default value when running 'restore -C'.
	Thanks to Kenneth Porter <shiva@sewingwitch.com> for the bug
	report.

9.	Many improvements to the 'cron_dump_to_disk' example.
	Thanks to Aaron S. Hawley <ashawley@users.sourceforge.net> for
	the patch.

10.	Fix the QFA generation when extended attributes are backuped
	along with the inodes. Prior to this fix, the entries for
	some inodes may get corrupted (duplicated entries or incorrect
	ones), making the QFA file unusable for repositionning in
	restore. Many thanks to Kenneth Porter <shiva@sewingwitch.com>
	for the bug report and the testing of the fix.

11.	Fix EA restoration in debug mode (Sourceforge bug #1986216).

12.	Stefan Auracher <star7@onlinehome.de> noticed that the Tower
	of Hanoi backup strategy documented in the man page wasn't
	actually based on the Tower of Hanoi algorithm. The man page
	was updated accordingly.

13.	Fix missing level information from dump output when using
	the default level (RedHat bug #493635). Thanks to Adam Tkac
	<atkac@redhat.com> for the patch.

14.	Fix the dump man page to reflect the fact that the default
	dump level is 0 not 9 (RedHat bug #356121). Thanks to Adam Tkac
	<atkac@redhat.com> for the patch.

15.	Fixed a few spelling errors in the man pages (RedHat bug #489853).
	Thanks to Adam Tkac <atkac@redhat.com> for the patch.

16.	Fixed restore man page and program usage which incorrectly stated
	that -P and -A were both allowed in the same invocation (RedHat
	bug #490627). Thanks to Adam Tkac <atkac@redhat.com> for the patch.

17.	Use sys_clone under Linux to share I/O contexts between dump
	processes, thus drastically increasing the performance of dump
	under CFQ (which is the default I/O scheduler used in a number of
	distributions). Thanks to Jeff Moyer <jmoyer@redhat.com> for the
	patch.

Changes between versions 0.4b40 and 0.4b41 (released January 2, 2006)
=====================================================================

1.	Fix restore of dumped Access Control Lists. The previous
	code for EA works fine, but ACLs needed conversion from the
	ext2/3 disk format to posix_acl format before restoring.

2.	Fix some issues with restoration of EA on big endian
	platforms.

3.	Fix restore when the symtab is over 2GB in size.

4.	Made the directory hash indexing an optional feature,
	accessible by the '-H' option of restore, and disable it
	by default.

5.	Fixed dump to not include extended attributes information
	in the toc (archive) file which confused restore -t.

Changes between versions 0.4b39 and 0.4b40 (released May 2, 2005)
=================================================================

1.	Changed restore to emit warnings (instead of emitting a fatal
	error) if a file (or a directory) is unavailable for a 
	comparision (if the user doesn't have the necessary permissions
	to access it for example). Thanks to Kenneth Porter
	<shiva@sewingwitch.com> for the bug report.

2.	Re-done the 'do not save directory entries to non-dumped inodes
	(excluded from dump)' feature. The previous implementation
	worked well for excluded directories but not for regular files.
	Thanks to Kenneth Porter <shiva@sewingwitch.com> for the bug
	report.

3.	Fixed a bug in dump where the tape size was miscalculated when
	the user used -d/-s to specify the tape characteristics. Thanks
	to Philip Goisman <goisman@physics.arizona.edu> for reporting
	the bug.

4.	Fixed another bug introduced in restore with the hashtree
	implementation. This one caused restore to stop saying
	"removenode: non-empty directory" in some cases.

5.	Added support for dumping and restoring ext2/3 extended
	attributes (EA), like the access control lists (ACL) or
	the security labels used by SELinux.

Changes between versions 0.4b38 and 0.4b39 (released January 21, 2005)
======================================================================

1.	The newly added dump_on_cd_3 example was buggy, replace it
	with an updated version from Andrew Basterfield
	<bob@cemetery.homeunix.org>.

2.	Made restore to chdir() back into the initial directory when
	dumping core while aborting a comparision operation. The
	previous behaviour was to write the corefile at the root of
	the directory being compared, which could very well be 
	read only and preventing the corefile generation. Thanks
	to Kenneth Porter <shiva@sewingwitch.com> for the bug report.

3.	Silenced the failure to call fgetflags() when comparing an
	entry which has no ext2 attributes (as in lsattr()).

4.	Fix a brown paper bug in restore -C which broke restore and
	caused modifications on the filesystem being compared
	(directories containing a file with the same name as the
	directory get renamed to RSTTMP...). Thanks to Kenneth Porter
	<shiva@sewingwitch.com> for finding the bug and helping me
	reproduce it.

5.	Made restore -C force the -N flag (no writing allowed on
	the disk) in order to prevent more bugs like the above one.

Changes between versions 0.4b37 and 0.4b38 (released January 7, 2005)
=====================================================================

1.	Fix a couple of troff syntax bugs in the man pages.
	Thanks to Eric Raymond <esr@thyrsus.com> for the patch.

2.	Made restore use either libncurses or libtermcap, depending
	on which one is available at configure time.

3.	Fixed restore negative size display bug when comparing a
	dump containing files over 2GB. Thanks to Steve Bonds
	<sbonds@users.sourceforge.net> for the bug report.

4.	Do not save directory entries to non-dumped inodes 
	(excluded from dump). This will eliminate the 'missing
	file' warnings when doing 'restore -C'.

5.	Fix dump crash when backuping a huge (2TB) filesystem,
	due to a bogus calculation on the inode map size.
	Thanks to Kevin B. Haines <K.B.Haines@rl.ac.uk> for 
	submitting the bug and testing the fix.

6.	Fix a problem in restore where the final \0 in the symbolic
	link names could have been lost, generating corrupt filenames.
	Thanks to Kyle Wilson <kyle.wilson@amd.com> for reporting the
	bug.

7.	Implemented a hash list for the directory names in restore.
	The linear list used before caused problems in interactive
	restores when dealing with directories having thousands of
	entries. Thanks to Brian Ristuccia <bristuccia@starentnetworks.com>
	for reporting the bug.

8.	Improved restore -C, this time including the directory
	attributes into the comparision.

9.	Made restore understand tapes containing EA/ACLs (which will
	be dumped by the next version of dump). In this version 
	extended attributes on the tape are ignored, for full EA/ACL
	support wait for the next version or try the experimental EA
	patch.

Changes between versions 0.4b36 and 0.4b37 (released July 7, 2004)
==================================================================

1.	Added the --enable-staticz configure option which enables
	dump and restore to be linked with static versions of
	libz and libbz2 (and dynamic versions of all other libraries).
	This will make Debian users happy, because libz and libbz2
	were the only needed libraries living in /usr, all the
	others live in /lib. In case of system emergency, it is 
	better not to have to rely on an extra filesystem.
	Thanks to Bdale Garbee <bdale@gag.com> for the suggestion.

2.	Fix compilation on (at least the Linux Debian port to) AMD64.
	(<ext2fs/ext2_types.h> defines some types (__s64 and __u64)
	that are also defined by <linux/types.h> (<asm/types.h>) and 
	they conflict).

3.	Make dump's reading of the dumpdates file a bit more robust,
	preventing dump from crashing when the dumpdates file has
	been modified by hand.

4.	Fixed some offset calculations in dump code which could 
	lead to "bread lseek errors" on large filesystems. Thanks
	to Bruce Lowekamp <lowekamp@users.sourceforge.net> for
	reporting this bug and debugging the issue.

5.	Made dump use the blkid library when searching for devices
	by label or uuid instead of dump's own routines.

6.	Corrected a bug in dump where a wrong LABEL=... line in
	/etc/fstab could prevent dump from dumping unrelated 
	filesystems. Thanks to Bruce Lowekamp
	<lowekamp@users.sourceforge.net> for reporting the bug.

Changes between versions 0.4b35 and 0.4b36 (released April 21, 2004)
====================================================================

1.	Fixed dump compilation with old gcc versions. Thanks to
	Mike Castle <dalgoda@users.sourceforge.net> for the patch.

2.	Fixed some warnings (howmany, roundup, powerof2 redefined)
	when compiling against a recent glibc version.

3.	Fixed a bug in restore preventing the read of a dump tape
	written with Solaris 7 ufsdump. Thanks to Patrick Higgins
	<phiggins@transzap.com> for reporting the bug and providing 
	the test case.

4.	Changed dump to enable the creation of volumes of different
	sizes in a single run (make -B accept a list of values).
	Patch contributed by Florian Zumbiehl <florz@gmx.de>.

5.	Use the glibc provided minor() and major() macros instead
	of our own bitmask implementation. This should be safe for
	when the major/minor namespace will migrate to 32 bits.
	Thanks to Zhang Jun <zhangjun@nanjing-fnst.com> for reporting
	the bug.

6.	Made explicit in the dump man page that dump will not create
	a remote file, it will only write to an already existing one.

7.	Another try at making size estimates better again.

8.	Put back the inconditional running of the end-of-tape
	script which was changed in 0.434 to be run only when -M
	or multiple -f were NOT specified. Some users rely on this
	feature even when it is combined with -M/-f.

9.	Fixed restore when restoring huge backups (where rstdir...
	temporary files are over 2GB). Thanks to 
	Raphael Jaffey <rjaffey@artic.edu> for reporting this, 
	debugging the issue and making the patch.

10.	Made restore understand FreeBSD UFS2 tapes. Thanks to
	David <vrtin@users.sourceforge.net> for submitting the bug
	and providing a test case.

11.	Made dump work with an arbitrary integer level (no more 10
	levels only limitation). Thanks to Matthew
	<msvincen@midway.uchicago.edu> for the patch.
	
Changes between versions 0.4b34 and 0.4b35 (released December 21, 2003)
=======================================================================

1.	Added a note in the dump man page saying that the default
	blocksize can be 32 if -d is used with a high density argument.
	Thanks to Antonios Christofides <A.Christofides@itia.ntua.gr>
	for the patch.

2.	Fixed configure to correctly understand CPPFLAGS, CFLAGS,
	LDFLAGS environment variables. Thanks to Arcady Genkin
	<antipode@users.sourceforge.net> for reporting the bug.

3.	Made -e/-E options of dump accept an unlimited number of inodes
	to be excluded, rather than a hardcoded maximum. Thanks to 
	Dietrich Rothe <d-rothe@users.sourceforge.net> for the patch.

4.	Updated the autoconf system to 2.50. Forced the -D_BSD_SOURCE
	and -D_USE_BSD_SIGNAL defines in configure in order to solve
	64bit build problems because quad_t is redifined with a 
	different signature. Thanks to Mike Harris <mharris@redhat.com>
	for reporting this bug.

5.	Made restore build on Solaris, making possible to
	restore Linux's "enhanced" tapes. Thanks to Uwe Gohlke
	<uwe@ugsoft.de> for the patch.

6.	Made an extension in the dump tape format capable of saving
	MacOSX specific inode extensions. Uwe Gohlke <uwe@ugsoft.de>
	wrote the extension and contributed the restore code back
	into this codebase. The same extension mechanism will be
	used in the future to save ACLs...

7.	Made rmt work correctly with regard to QFA and local/remote
	files and tapes. The remote access will however work only 
	when the dump provided rmt version is used. If you want to
	use another rmt server, please do not use the QFA feature.
	Thanks to Uwe Gohlke <uwe@ugsoft.de> for the patch.

Changes between versions 0.4b33 and 0.4b34 (released April 18, 2003)
====================================================================

1.	Fixed the INSTALL file to reflect the actual install paths.
	Thanks to David Ronis <ronis@ronispc.chem.mcgill.ca> for
	reporting the bug.

2.	Fixed the configure script to only check for headers presence
	instead of trying to compile them. This should fix issues
	with old build environments. Thanks to Kari Lempiainen 
	<kari@funky.pp.fi> for reporting the bug.

3.	Fixed restore to correctly ignore sockets when comparing
	a dump (as socket cannot be properly restored anyway). Thanks
	to Gunther Reiszig <gunther@mit.edu> for reporting the bug.

4.	Fixed restore to correctly access the archive file (-A argument)
	even when using a remote tape. Thanks to Fabrice Bellet 
	<fabrice@bellet.info> for reporting the bug.

5.	Fixed (again) handling of long (largefile) seeks in rmt.
	Thanks to Fabrice Bellet <fabrice@bellet.info> for reporting
	the bug.

6.	Fixed restore corner case when dealing with large block sizes
	dump is able to write now (-b 1024). Thanks to Fabrice Bellet
	<fabrice@bellet.info> for reporting the bug.

7.	Fixed a bug preventing dump to access a filesystem having 
	a label exactly 16 bytes in length. Thanks to <minduni@ti-edu.ch>
	for reporting the bug.

8.	Made dump store dump dates using explicit timezones, fixing a
	problem with incremental dumps if the timezone is modified 
	between the dumps. Thanks to Philippe Troin <phil@fifi.org> for
	the bug report and the patch.

9.	Fixed a bug encountered when dumping individual files (not full
	filesystems or directories) and dangling symbolic links happen
	to be in the list of files. For as far as dump is concerned,
	dangling symbolic links are allowed, and are dumped as is.
	Thanks to Jin-su Ahn <jsahn@ee.snu.ac.kr> for reporting the
	bug and providing the fix.

10.	Fixed open and creation modes and permissions for QFA and
	table-of-contents files in dump and restore. Thanks to 
	Philippe Troin <phil@fifi.org> for the patch.

11.	Fixed the archive file descriptor handling enabling it to be 0.
	This can happen in some cases when shell redirections are used.
	Thanks to Philippe Troin <phil@fifi.org> for the patch.

12.	Delayed the opening of archive file until after suid had been
	dropped (fixing a possible security issue if dump is suid). 
	Thanks to Philippe Troin <phil@fifi.org> for the patch.

13.	Fixed the 'S' command handling in the rmt client part.
	Thanks to Philippe Troin <phil@fifi.org> for the patch.
	
14.	Modified the end-of-tape script handling to print out statistics
	(and stop the timer) before launching the eot script. Also, the eot
	script does not get run anymore when using -M (which makes sense) or
	when multiple tapes are listed on the command line 
	(-f tape0,tape1,tapen) (which also makes sense).
	Thanks to Philippe Troin <phil@fifi.org> for the patch.

15.	Relicensed dump/restore under the 'revised' BSD license, as per
	ftp://ftp.cs.berkeley.edu/ucb/4bsd/README.Impt.License.Change.

16.	Added LZO compression to dump. This new compression method has
	the advantage of being super fast, thus not killing tape streaming
	on slow machines. Thanks to Helmut Jarausch 
	<jarausch@igpm.rwth-aachen.de> for the patch and to 
	Markus Oberhumer <markus@oberhumer.com> for giving special permission
	to include his miniLZO project (GPL licensed) in dump/restore.

17.	Some small buffer overruns fixes in rmt. Thanks to Antonomasia
	<ant@notatla.demon.co.uk> for reporting the bugs.

18.	Added a special rmt version which can do encryption when writing
	to tape. Read examples/encrypted_rmt/README for details on
	how to enable and configure it. Thanks to Ken Lalonde
	<ken@globalremit.com> for the patch.

19.	Made dump work with 2.5 kernel end of tape early warning semantics.
	Thanks to Kai Makisara <Kai.Makisara@kolumbus.fi> for the patch.

20.	Fixed a bug which caused dump -w|-W not to work anymore, because
	the fs_freq and fs_passno fields in /etc/mtab are always set
	to 0 0. Thanks to Trent Piepho <xyzzy@speakeasy.org> for 
	reporting the bug.

Changes between versions 0.4b32 and 0.4b33 (released February 10, 2003)
=======================================================================

1.	Added a note in the restore man page clarifying the question
	'set the permissions on the current directory ?' asked by
	restore at the end of treatment in -i and -x modes.

2.	Fixed the endianess issues when reading compressed tapes.
	Thanks to Dark Force <daq4th@users.sourceforge.net> for
	reporting this bug and providing test cases.

3.	Fixed the "ACL won't be dumped" warning message (which showed
	an extra, unrelated error message). Thanks to Dragan Krnic
	<dkrnic@lycos.com> for reporting this bug.

4.	Made dump look first into /etc/mtab, then into /etc/fstab 
	when searching for filesystem paths. Also fixed some problems
	caused by binding mounts or multiple block device mounts.
	Thanks to Matus Uhlar <uhlar@fantomas.sk>, Elliott Mitchell
	<ehem@m5p.com>, Greg Edwards <gedwards@users.sourceforge.net>,
	Brian Hoy <brian.hoy@opus.co.nz>. (fixes Debian bugs #147086
	and #160305, Sourceforge bugs #618699 and #679832).

5.	Made dump's -I option accept the value '0' meaning all the
	read errors will be ignored. This can be useful when running
	dump from unattended sessions (like cron jobs). Thanks to
	John I Wang <jiwang@users.sourceforge.net> for the suggestion.

6.	Fixed the output of dump to indicate 'blocks' instead of
	'tape blocks' in the various messages (blocks are always
	1 Kilobyte, tape blocks are 1 BK * '-b' argument), and 
	made it clearly print the current blocksize at the start of
	a dump. Thanks to Michal Szymanski <msz@astrouw.edu.pl> for
	the suggestions.

7.	Made rmt understand long (largefiles) seeks.

8.	Fixed build with very old versions of libext2fs, where
	EXT2_FT_* constants were undefined.

9.	Made dump accept the dumpdates path on the command line
	(-D file option) instead of using only the hardcoded one.
	Thanks to Piete Brooks <pb22@users.sourceforge.net> for the
	suggestion.

10.	Enabled rmt, LFS, readline, QFA options by default in
	./configure. Updated the configure process (new versions
	of config.guess, config.sub etc).

Changes between versions 0.4b31 and 0.4b32 (released November 15, 2002)
=======================================================================

1.	Changed dump to use fcntl(F_SETLK) style locking instead
	of flock() when locking the dumpdates file. With the old 
	locking scheme, a local user having read rights on the 
	dumpdates file could be able to do a Denial of Service attack
	on dump. In order to lock the dumpdates file with the new
	scheme, the user would need to have write access on the file.
	Thanks to Richard Johnson <Richard.Johnson3@ey.com> for 
	reporting the bug (originally a bugtraq post).

2.	Fixed interactive 'ls' which caused spurious errors warnings
	about 'undefined filetypes' detected. Thanks to Jorgen Ostling 
	<jorgen_ostling@users.sourceforge.net> for reporting this 
	bug.

3.	Fixed dump's estimate when dealing with sparse inodes.

4.	Modified dump to allow setting a blocksize bigger than 32kB
	(raised the limit to 1024kB), because newer hardware needs
	this for proper operation. Thanks to Dirk Traenapp
	<dtraenapp@users.sourceforge.net> for reporting this.

5.	Fixed a bug causing Dump to stop and report an error if an
	inode number in the exclude file was followed by some amount
	of whitespace. Thanks to Jeffrey Sofferin 
	<sofferin@users.sourceforge.net> for reporting this bug.

6.	Fixed a bug which caused restore, in some particular cases,
	to ask some 'scary' questions and leave a bunch of RSTTMP
	directories behind when restoring incremental tapes. Thanks
	to Philippe Troin <phil@fifi.org> for reporting this bug and
	providing the test cases.

7.	Changed the wording when inodes are excluded from dump:
	replaced 'Added inode 7 to exclude list' with
	'Excluding inode 7 (resize inode) from dump', as suggested
	by Elliott Mitchell <ehem@m5p.com> in a Debian bug report.

Changes between versions 0.4b30 and 0.4b31 (released July 30, 2002)
===================================================================

1.	Fixed rmt open flags transmission (GNU's symbolic syntax over
	rmt) which I broke in 0.4b29. Thanks to Eros Albertazzi
	<eros@lamel.bo.cnr.it> for reporting the bug.

Changes between versions 0.4b29 and 0.4b30 (released July 25, 2002)
===================================================================

1.	Made dump print out the ext2 logical block number in case of a read
	error, which can be given as an argument to debugfs' ncheck command,
	in order to find out the affected inode. Added note about this 
	usage in dump's man page.

2.	Fixed a problem in restore when reading tapes written on big
	endian machines with very old versions of dump. The patch was
	contributed by George Helffrich <george@geology.bristol.ac.uk>.

3.	Fixed the tape length calculation when using large tapes
	and compression. Thanks to Georg Lippold 
	<g_lippold@sourceforge.net> for reporting the bug.

4.	Added a new set of examples in dump_on_cd_2 directory, based
	on dump_on_cd examples but somewhat enhanced, supporting DVD
	media, and localized in english and german. Thanks to
	Georg Lippold <g_lippold@sourceforge.net> for the new scripts.

5.	Made dump save 32 bit UID/GID. Internally, this was achieved
	by switching from the old BSD inode format to the new BSD 
	inode format, which means that the tape format was changed.
	However, since all restore versions out there should 
	transparently support both inode formats, the change should
	have no side effects. Thanks to John Yu <jky@cs.bu.edu> for
	reporting the bug.

6.	Fixed a lot of warnings in the code shown when compiling
	with 'gcc -W'. Thanks to Matthias Andree 
	<matthias.andree@stud.uni-dortmund.de> for reporting this.

7.	Fixed a small markup bug in the dump man page. Thanks to
	Eric S. Raymond <esr@minx.thyrsus.com> for submitting the
	patch.

8.	Rewrote entirely the man pages using the tmac.an macro
	package (Linux man page format) instead of the original BSD
	format. They should be now cleaner and easier to modify.

Changes between versions 0.4b28 and 0.4b29 (released June 8, 2002)
==================================================================

1.	Fixed a problem in the rmt ioctl command, where ioctl's issued from
	non Linux clients were misinterpreted. The description of the problem
	(incompatible numbering in Linux mtio opcodes) is documented at
	ftp://ftp.fokus.gmd.de/pub/unix/star/README.mtio . Thanks to