Use O_DIRECT to read bitmap files.

[thirdparty/mdadm.git] / ChangeLog

Changes Prior to 2.4.1 release
    -   Honour --write-mostly when adding to an array without persistent
        superblocks.
    -   Fix alignment problem in version-1 superblocks.
	NOTE: This is an incompatable change affecting raid5 reshape.
	If you want to reshape a raid5 using version-1 superblocks,
	use 2.6.17-rc2 or later, and mdadm-2.4.1 or later.

Changes Prior to 2.4 release
    -   Rewrite 'reshape' support including performing a backup
	of the critical region for a raid5 growth, and restoring that
	backup after a crash.
    -   Put a 'canary' at each end of the backup so a corruption
	can be more easily detected.
    -   Remove useless 'ident' arguement from ->getinfo_super method.
    -   Support --backup-file for backing-up critical section during
        growth.
    -   Erase old superblocks (of different versions) when creating new
        array.
    -   Allow --monitor to work with arrays with >28 devices
    -   Report reshape information in --detail
    -   Handle symlinks in /dev better
    -   Fix mess in --detail output which a device is missing.
    -   Manpage tidyup
    -   Support 'bitmap=' in mdadm.conf for auto-assembling arrays with
	write-intent bitmaps in separate files.
    -   Updates to md.4 man page including section on RESTRIPING and SYSFS

Changes Prior to 2.3.1 release
    -   Fixed -O2 compile so I could make and RPM.
    -   Type cast number to be printed %llu so it compiles on 64bit
        machines. (Thanks Luca).
    -   Stop using asm/byteorder.h - to make Redhat happy :-(
    -   Require bitmap files to have a '/' in their name.
    -   Error-check a few syscalls - code from SuSE package.

Changes Prior to 2.3 release
    -   Try /etc/mdadm/mdadm.conf if /etc/mdadm.conf doesn't exist.
	This provided compatability for Debian.
    -   Fixed for version-1 superblock:
	 report chunksize for raid6 and raid10
	 make sure device size used is a multiple of chunksize
    -   Fix "--assemble --scan" crash.
    -   Fix completely failure to create array on ppc64
    -   Fix memcmp in place of memcpy
    -   A few minor improvements to online help
    -   Clean up usage of 'long long' for used-size of devices, so
	that it is possible to create a raid1 of 7TB devices!
    -   Make internal bitmaps work on 7TB raid1 arrays.
    -   Provide error message if --examine doesn't find any superblock.
    -   Report 'reshape' status in --examine - this depends on kernel
	patches that are not yet finalised.
    -   Report bitmap status in --detail and --examine
    -   Default to v1 superblocks instead of v0.90 if the array
	is too big for 0.90 to handle.
    -   Sort the output of "mdadm --detail --scan" so that it is
	in a suitable order for assembling arrays. i.e. components come
	before an array that they are part of.
    -   Print size of large reiserfs array properly went warning of
        possible confilcts.

Changes Prior to 2.2 release
    -   Assorted bug fixes
    -   Support write-intent-bitmaps on raid10
    -   Support little-endian (Rather than hostendian) bitmaps.
    -   Return correct error code from 'mdadm -S'
    -   Remove extra blank line from 'mdadm -Eb' output.
    -   Improve option parsing so that -a and -b do not have
	optional arguements: the arg is either required or not
	depending on context.
    -   Allow scanning of devices listed in /proc/partitions even
	if they don't appear in /dev.
    -   Support --assume-clean in --create mode as well as --build
    -   Add support for --monitor to report to syslog: -y or --syslog.
	Thanks to Ross Vandegrift
    -   --monitor now reports which device failed in a 'Fail' message
	   This broke with 2.6
    -   Improve chance of array starting properly after a crash.
	mdadm was insisting the event numbers were identical, but this
	isn't needed, and is a problem if the crash was while the metadata
	was being updated.
    -   Support --update==uuid
    -   Added README.initramfs and mkinitramfs to help people use an
	initram for starting md arrays at boot.

Changes Prior to 2.1 release
    -   Fix assembling of raid10 array when devices are missing.
	  mdadm now correctly detects if a array is workable or not
	  depending on which devices are present, and so will correctly
	  handle "--assemble --force"  if multiple devices have failed.
    -   Report raid10 layout in --examine output.
    -   Fix assembling of arrays that use the version-1 superblock and
	  have spares.  Previously the spares would be ignored.
    -   Fix bug so that multiple drives can be re-added at once.
    -   Fix problem with hot-adding a bitmap to version-1-superblock
	  arrays.

Changes Prior to 2.0
    -   Support assembling from byte-swapped superblocks
	metadata type "0.swap" and --update=byteorder
    -   write-mostly and write-behind support for raid1.
    -   Support --name= and 'name=' config entry for identifying
	arrays be name.
    -   RAID10 in man pages.
    -   Lot of minor manpage updates

Changes Prior to 2.0-devel-3 release
    -   Assorted fixes for multiple bugs...
    -   Add test suite
	
Changes Prior to 1.12.0 release
  Several of these are backported from the Debian package
    -   Don't use 'lstat' to check for blockdevices, use stat.
    -   Document --size=max option for --grow
    -   Document SparesMissing event and DeviceDisappeared/WrongLevel
    -   --stop --scan repeatly cycles until no more progress can be made
	so that stack devices are stopped properly
    -   Minor type rationalisation for ident->uuid - now always 'int[]'
    -   Fix type in online help for --grow (was -F, now -G)
    -   Allow --auto command line option to set default auto=
	value when running "--assemble --scan".  Previously
	--auto was ignored if --scan was given
    -   Fix a few type casts
    -   Fix parsing of /dev/md/N in is_standard
    -   Fix rounding errors in human_size()
    -   Fix silly example in mdadm.conf-examples
    -   When finding a /dev name for a device, prefer shorter names
    -   Suppress listing of devices= in --brief output of -D or -E,
	unless -v is given (-vv gives the old behaviour with -Dsv).
	This is because the device list can change and so is not a
	stable aspect of the array
    -   Allow --force with --grow so '-Gfn1' works (on raid1)
    -   Replace sprintf calls with snprintf (to quiet diet-libc)
    -   Add man page for mdassemble
    -   Fix compilation with tinyc
	
Changes Prior to 1.11.0 release
    -   Fix embarassing bug which causes --add to always fail.

Changes Prior to 1.10.0 release
    -   Fix bug with --config=partitions
    -   Open sub-devices with O_EXCL to detect if already in use
    -   Make sure superblock updates are flushed directly to disk.
	
Changes Prior to 2.0-devel-1 release
    -   Support for version-1 superblock.  See --metadata option.
    -   Support for bitmap based intent logging.
    -   Minor fixes.
	
Changes Prior to 1.9.0 release
    -   Fix rpm build problem (stray %)
    -   Minor manpage updates
    -   Change "dirty" status to "active" as it was confusing people.
    -   --assemble --auto recognises 'standard' name and insists on using
        the appropriate major/minor number for them.
    -   Remove underscore from partition names, so partitions of 
	"foo" are "foo1", "foo2" etc (unchanged) and partitions of
	"f00" are "f00p1", "f00p2" etc rather than "f00_p1"...
    -   Use "major", "minor", "makedev" macros instead of 
	"MAJOR", "MINOR", "MKDEV" so that large device numbers work
	on 2.6 (providing you have glibc 2.3.3 or later).
    -   Add some missing closes of open file descriptors.
    -   Reread /proc/partition for every array assembled when using
	it to find devices, rather than only once.
    -   Make "mdadm -Ss" stop stacked devices properly, by reversing the
	order in which arrays are stopped.
    -   Improve some error messages.
    -   Allow device name to appear before first option, so e.g.
	mdadm /dev/md0 -A /dev/sd[ab]
	works.
    -   Assume '-Q' if just a device is given, rather than being silent.
    -   Change "dirty" status to "active" as it was confusing people.
	
Changes Prior to 1.8.0 release
    -   Makefile cleanup from  Luca Berra <bluca@comedia.it>
    -   --pid-file (-i) to set a pid file to use with --monitor --daemonise
    -   Fix typo in mdadm man page
    -   Fix coredump when "-s" used with no config file present.
    -   Support new "faulty" personality which can inject synthetic
        faults. (Not in kernel.org yet at 1Nov2004)
    -   Support raid0/linear on devices > 2 Terabytes
    -   Make sure raid6 resyncs when created with one missing device

Changes Prior to 1.7.0 release
    -   Support "--grow --add" to add a device to a linear array, if the
        kernel supports it.  Not documented yet.
    -   Restore support for uclibc which was broken recently.
    -   Several improvements to the output of --detail, including
        reporting "resyncing" or "recovering" in the state.
    -   Close filedescriptor at end of --detail (exit would have closed it 
	anyway, so this isn't abig deal).
    -   Report "Sync checkpoint" in --examine output if appropriate.
    -   Add --update=resync for --assemble mode to for a resync when the
        array is assembled.	 
    -   Add support for "raid10", which is under development in 2.6.
	Not documented yet.
    -   --monitor now reads spare-group and spares info from config file
	even when names of arrays to scan are given on the command line
	
Changes Prior to 1.6.0 release
    -	Device name given in -Eb is determined by examining /dev rather
	than assuming /dev/md%d
    -   Fix bug in --monitor where an array could be held open an so
	could not be stopped without killing mdadm.
    -   Add --grow mode.  Currently only --size and --raid-disks can be
	changed. Both require kernel support which, at the time of
	writing, is not in a release kernel yet.
    -   Don't print out "errors" or "no-errors" in -D and -E, as the bit
	is never set or used.
    -   Use md event notification in 2.6.??? to make --monitor mode
        respond instantly to events.
    -   Add --auto= option and auto= configfile entry to tell mdadm to
	create device files as needed.  This is particularly useful
	with partitioned arrays where the major device number can change.
    -   When generating --brief listing, if the standard name doesn't
        exist, search /dev for one rather than using a temp name.
    -   Allow --build to build raid1 and multipath arrays.
    -   Add "--assume-clean" for Create and Build, particularly for raid1
	Note: this is dangerous. Only use it if you are certain.
    -   Fix bug so that Rebuild status monitoring works again.
    -   Add "degraded" and "recovering" options to the "Status:"
	entry for --detail

Changes Prior to 1.5.0 release
    -   new commands "mdassemble" which is a stripped-down equivalent of
	"mdadm -As", that can be compiled with dietlibc.
	Thanks to Luca Berra <bluca@comedia.it>.
	It can be using in an initramfs or initrd.
    -   Fix compiling error with BLKGETSIZE64 and some signed/unsigned
	comparison warnings.
    -   Add Rebuild Status (% complete) to --detail output.
    -   Support "--monitor --test" which will generate a test alert
	for each array once, to test notification paths.
    -   Generate RebuildFinished event when rebuild finishes.
    -   Support for raid6 as found in 2.6.2 - thanks to  
	H. Peter Anvin  <hpa@zytor.com>
    -   Support partitioned md arrays with a different major number and
	naming scheme (md_dX in /proc/mdstat, /dev/md/dXpY in /dev).
	
Changes Prior to 1.4.0 release
    -   Document fact that creating a raid5 array really creates a
        degraded array with a spare.	
    -   Add "spares=" tag to config file and generate it wit --detail and
        --examine
    -   Add "SparesMissing" event when --monitor first sees an array and
	it doesn't have the enough spare devices.
    -   Add --update=summaries for --assemble to update summary
        information in superblock, and correct other inconsistancies in
	the superblock.
    -   Add --test option to --detail to set a meaningful exit status.
	
Changes Prior to 1.3.0 release
    -   Make 'size' and unsigned long in Create to allow creation of
        larger arrays.
    -   Explicitly flag spare devices as 'spare' in --detail and --examine
        output.  Previously they simply had no flags lists.
    -   Make MailCmd (for monitor) configurable in Makefile, and default
        to "/usr/sbin/sendmail -t". Also split out the warning related
	flags into CWFLAGS for easier build configurability.
    -   Minor bugfix in Manage code.
    -   --monitor now notices and reports degraded arrays at startup using
        "DegradedArray" event, and also has a --oneshot option to only
	report DegradedArrays, and then exit.
    -   Small man-page clarification w.r.t. raid levels and raid4 in
        particular.
    -   Disallow creation of arrays with only one device as this is
        probably a mistake.  --force will override this check.
    -   Correct some misleading documentation in the "mdadm --create --help"
	message.
    -   Ignore chunksize if raid1 or multipath.
    -   Explicit statement in man page that raid-disks cannot be changed
        after array is created.
    -   Improve message when attempting to start an array with
        insufficient devices.  Instead of required the array to be full,
	we only require it has as many active devices as last time.
	
Changes Prior to 1.2.0 release
    -   Fix bug where --daemonise required an argument.
    -   In --assemble --verbose, print appropriate message if device is
        not in devices= list
    -   Updated mdadm.conf.5 to reflect fact that device= takes wildcards
    -   Typos:  componenet -> component
    -   Reduce size of "--help" message put excess into "--help-options"
    -   Fix bug introduced when MD_SB_DISKS dependancy removed, and which
        caused spares not be assembled properly.
    -   Print appropriate message if --monitor --scan decides not to
	monitor anything.  
Changes Prior to 1.1.0 release
    -   add --deamonise flag for --monitor - forks and prints pid to stdout
    -   Fix bug so we REALLY clear dirty flag with -Af
    -   -Db now prints a 'devices=' word for each array.
    -   "mdadm -A /dev/md0" will get info from configfile, even without scan
    -   When assembling multipath arrays, ignore devices which are flagged
        as having errors.
    -   take --super-minor=dev  to mean "use the minor number of the mddev 
	being assembled.
    -   take --config=none to mean "completely ignore config file"
    -   Make --monitor require --scan or a device list.
Changes Prior to 1.0.9 release
    -   Documentation updates including kernel parameters documented
	in md.4
    -   --assemble --force for raid4/5 will mark clean, needed for 2.5
    -   --detail prints out the events counter as well
    -   flush device before reading superblock to be sure to get
	current data
    -   added mdadm.static target to makefile for static linking
    -	--monitor was ignoring  /dev/md0 due to off-by-one error
    -	Fix assorted typos
    -   Fix printing of Gibibytes - calc was wrong.
    -   Fix printing of Array Size in --detail when very big.
    -   --monitor no longer tries to work for raid0 or linear as these
	have nothing to be monitored.
    -   The word 'partitions' on a DEVICE line will cause all partitions
	listed in /proc/partitions to be considered
    -   If the config file is called 'partitions' then it will be treated
        as though it contained exactly 'device partitions' so e.g.
	  mdadm -Ebsc partitions
	will find all raid partitions easily.	
    -	successfully assemble multipath devices by ignoring raid_disk
	value from superblock (it is always the same).
    -   --assemble not tied to MD_SB_DISKS limit quite so much
    -   Support compiling with tcc
    -   Support compiling with uclibc - just skip scan of /dev
    -   Add --update= option for Assemble mode.  Either sparc2.2
	or super-minor updates are possible.  See mdadm.8
	
Changes Prior to 1.0.1 release
    -   Round off MB/GiB etc values instead of round down.
    -   Add --sparc2.2 option to examine to shift superblock around
	and --sparc2.2update to rewrite the superblock
    -   Fix assorted typos in online help
	
Changes Prior to 1.0.0 release
    -   Allow --config with Misc mode (for --examine --scan)
    -   Add $(CXFLAGS) to end of CFLAGS in makefile
    -   When making an N disk raid5 array, the Nth drive
	is moved to the end of the array as a spare rather than
	being shifted up one place.  This means that when the
	kernel builds onto the last spare and inserts it,
	the devices will be in the expected order.
    -   Man page improvements
Changes Prior to 0.8.2 release
    -   Correct spelling of persist[ae]nce/persist[ae]nt.
    -   Change "disk" to "device" in options and config file
    -   convert array size to "long long" *before* shift-left in -D and -Q
	
Changes Prior to 0.8.1 release
    -   Add "INSTALL" file.
    -   Fix some "i" variables that were not being set properly
    -   Initialise minsize and maxsize so that compilers don't complain.
    -   Tidy up Makefile and mdadm.spec installations
    -   Add "multipath" to documentation of valid levels
	
Changes Prior to 0.8 release
    -   Fix another bug in Assemble.c due to confusing 'i' with 'j'
    -   Minimal, untested, support for multipath
    -   re-write of argument parsing to have more coherent modes,
    -   add --query,-Q option
    -   Update mdadm.8 to reflect arg processing change and --query
    -   Change "long" to "unsigned long" for device sizes
    -   Handle "mailaddr" and "program" lines in config file for follow/scan mode.
    -   --follow --scan will exit if no program or mail found
    -   Add MAILADDR and PROGRAM to mdadm.conf-example
    -   Spell check man pages
    -   consistently use "component devices" instead of "subdevices"
    -   Make -Wall -Werror really work and fix lots of errors.
    -   --detail and --stop can have --scan which chooses devices from /proc/mdstat
    -   --monitor detects 20% changes in resync, failed spares,
        disappearing arrays,
    -   --monitor --scan will automatically add any devices found in /proc/mdstat
    -   --monitor will move spares between arrays with same spare-group if necessary
    -   Documentation for Monitor Mode
    -   --query notes if the array containing the given device is active or not
    -   Finished md.4 man page.
	
Changes Prior to 0.7.2 release
    -   mdadm.spec updates and ifdef BLKGETSIZE64 from  Luca Berra -- bluca@comedia.it
    -   more mdadm.spec updates from  Gregory Leblanc <gleblanc@linuxweasel.com>
    -   make directory for mdadm.conf configurable in Makefile
    -   Finished mdadm.conf.5.  Removed details of conf file from
	mdadm.8 leaving a reference to mdadm.conf.5.
    -   Fix bug in Assemble.c, thanks to Junaid Rizvi <domdev@sat.net.pk>
    -   Get --assemble --force to make sure old major/minor numbers are
        consistant, as md.c worries about this :-(
	
	
Changes Prior to 0.7.1 release
    -   update mdadm.spec
    -   use BLKGETSIZE64 if available for array size
    -   give human readable as GiB/MiB and GB and MB, with 2 decimal point precision
    -   Only warn about size variation for raid1/4/5.
    -   Started md.4 man page
    -   Started mdadm.conf.5 man page
	
Changes Prior to 0.7 release	

    -   Fix makefile to install binary at /sbin and not /sbin/sbin
	Also install man page.
    -   Add --zero-superblock based on --destroywithextremeprejudice
	from Dale Stephenson <steph@snapserver.com>
    -   change name to mdadm.  It is palandromic, and much nicer to pronouce.
	
Changes Prior to 0.6 release

    -   Remove the limit on the number of device names that can be
	given on the command line.
    -   Fix bug in --assemble --force where it would only update a 
	single superblock.
    -   Fix bogus printing of big numbers not being block devices
	when given names of devices that don't exist.
    -   When --assemble --force, consider superblocks with an event
	count that is 1 behind as out-of-date.  Normally they are
	considered up-to-date (as the kernel assumes this too).
    -   When marking drives as not-failed in the superblock, 
	we also mark them as ACTIVE and SYNC.
    -   Don't start arrays for which not all drives are available unless:
	 --scan   which implies that all drives were found automatically
	 --run    which means the user knows what they want
	 --force  which means that we are fixing something broken
    -   Make sure all device numbers passed as 3rd arg of ioctl
	are passed as unsigned lock, so that it works on SPARC
    -   If HOT_ADD_DISK failes for -a, then only try ADD_NEW_DISK
	if we cannot read from the array, i.e. if the array is
	not started yet.
    -   man page update
    -   Taught Examine to handle --scan. It examines all devices listed
	on DEVICE lines in the config file.  
    -   Added --brief (-b) flag for Examine and Detail to print out
	and mdctl.conf compatible description with uuid=, level=,
	disks= and  - for Examine - devices=
	--examine --brief collects all devices the make the one array and 
	list them as one entry.
    -   Added level= and disks= options to ARRAY lines in config files
	so --brief output could be used as-is.
    -   Make parity style ({left,right}-{,a}symmetric) consistantly use -,
	never _.
    -   Add "Array Size" to --detail output
    -   Change "Size" to "Device Size" and exclude from Detail of arrays
	that do not have a consistent device size.
    -   Add Human readable MiB or GiB value on size lines of Detail and Examine
    -   --assemble --scan  doesn't complain about active drives
    -   require number of spares given in -x to be listed.
    -   Made --build actually work.
Changes Prior to 0.5 release

  --assemble:
     spare drives are handled properly.

     --force can be used to recover from 2-drive failures on RAID5
     If you belive that /dev/hda1 /dev/hdb1 /dev/hdc1 /dev/hdd1 should
     make a raid5 array, but it has experienced multiple failures and
     wont start, then

       mdctl --assemble --force /dev/md0 /dev/hd[abcd]1

     Should update the superblock on the newest failed drive and
     restart the array in degraded mode.  You should then remove the
     remaining failed drive and re-add it (if you are happy that it
     might work).

     Ofcourse whenever you have a 2-drive failure, you have a risk
     of corruption in data that hasn't be changed for a long time.  So
     this doesn't give you your array back all nice and happy, but it
     does allow you to recover data that might not be corrupt.

     More flexibility in identifying a RAID array in the mdctl.conf
     e.g.
         array /dev/md4  super-minor=4

      assembles /dev/md4 from all devices found that have a raid
      superblock that says the minor number of the array is 4.
      If the blocks with the right minor number do not all have the
      same UUID, an error is flags and no assembly happens.
 
        array /dev/md3  devices=/dev/hd[abc]2

      Assembles /dev/md3 drom /dev/hda2 /dev/hdb2 and/dev/hdc2.  All
      devices must exist and have raid superblock with the same uuid.

      If two identity specifiers are used, only devices that match all
      of them are considered, so

        array /dev/md2 devices=/dev/hd?2 super-minor=2

      will assemble /dev/md2 using all /dev/hd?2 devices which have a 
      raid superblock with minor number 2.
 
  --create:
      When listing devices for --create, the word "missing" can be
      used to indicate that the respective slot does not have a
      working drive currently.  This is similar to the "failed-disk"
      directive in mkraid/raidtab.
      e.g.
         mdctl --create --level=5 -raid-disks=4 --spare-disks=2
                 /dev/md0 /dev/sda /dev/sdb missing /dev/sdc /dev/sdd  /dev/sde

      will create a raid5 array with the third slot empty, and two
      spares.

      By default, raid5 arrays are created with the last slot empty
      and drive listed for the last slot added as a spare.  If a
      "missing" slot is given, or if --force is given, then --create
      does exactly what you ask and doesn't try to be clever.
   

   --follow / --monitor:

      This is a new mode.  I couldn't stop my self from picking a name
      starting with F (as current modes start A,B,C,D,E) but I
      relented and provided an alternate name that is somewhat more
      meaningful. 
 
      In this mode, mdctl does not exit, but runs continuously and
      periodically polls all the md devices to see if they have had
      any interested state change.
      The changes that it currently notices are:
           Fail      -  an active disc fails
           FailSpare -  a spare, that was presumably being build, fails
           ActiveSpare - a spare becomes active, presumably after a rebuild.

      Options:
         --mail mailaddress  - send Email on any Fail* event
         --program program   - run the program on any event.  
		   Args are: eventname mddevice subdevice(if-known)
         --delay  seconds    - change from the default 60second pause
			       between polls.

      I plan to add functionality to this mode to allow sharing of
      spare drives. If an array is marks "spare-group=fred", and it
      has a failed drive and no spares, and if some other array is
      also "spare-group=fred" and it has no failed drives, but does
      have a spare drive that is big enough, the spare will be moved
      to the first array.

  I also have the idea of adding a --grow mode which will re-organise
  the data on an N disk raid0/4/5 array to be on an N+M disk array.
  I have no concrete plans for this though.

  I got rid of the "v" in the archive file name, and include the
  version number in the directory created by the archive.

  There is now a man page and mdctl.spec (for rpm) thanks to
  Danilo Godec <danci@agenda.si>.
  
  Ofcourse, the man page is now out of date and despite being based on
  the --help output, is not wholy correct.  After I get --follow
  working properly, I plan to revise the various documentation and/or
  the code to make sure the two match.