12 years agomore ddf stuff
Neil Brown [Thu, 15 May 2008 06:48:15 +0000 (16:48 +1000)] 
more ddf stuff
Create a BVD in a DDF

Do not actually assemble it yet...

12 years agoInitial DDF support code.
Dan Williams [Thu, 15 May 2008 06:48:14 +0000 (16:48 +1000)] 
Initial DDF support code.

Create a ddf array by naming the device /dev/ddf* or
specifying metadata 'ddf'.

If ddf is specified with no level, assume a container (indeed,
anything else would be wrong).

**Need to use text_Version to set external metadata...

More ddf support

Load a ddf container.  Now
   --examine /dev/ddf
super-ddf: fix compile warning

From: Dan Williams <>

super-ddf.c:723: format %lu expects type long unsigned int, but argument 3 has type unsigned int

Signed-off-by: Dan Williams <>
12 years agoSome support for external metadata.
Neil Brown [Thu, 15 May 2008 06:48:13 +0000 (16:48 +1000)] 
Some support for external metadata.

Allow specifying metadata type when creating arrays etc.

12 years agoChange write_init_super to be called only once.
Neil Brown [Thu, 15 May 2008 06:48:12 +0000 (16:48 +1000)] 
Change write_init_super to be called only once.

The current model for creating arrays involves writing
a superblock to each device in the array.
With containers (as with DDF), that model doesn't work.
Every device in the container may need to be updated
for an array made from just some the devices in a container.

So instead of calling write_init_super for each device,
we call it once for the array and have it iterate over
all the devices in the array.

To help with this, ->add_to_super now passes in an 'fd' and name for
the device.  These get saved for use by write_init_super.  So
add_to_super takes ownership of the fd, and write_init_super will
close it.
This information is stored in the new 'info' field of supertype.

As part of this, write_init_super now removes any old traces of raid
metadata rather than doing this in common code.

12 years agoAdd crc32 files.
Neil Brown [Thu, 15 May 2008 06:48:10 +0000 (16:48 +1000)] 
Add crc32 files.

For ddf.

12 years agoAdd cpu_to_be convertions functions.
Neil Brown [Thu, 15 May 2008 06:48:09 +0000 (16:48 +1000)] 
Add cpu_to_be convertions functions.

DDF will uses these.

12 years agoReduce openning of dev in create.
Neil Brown [Thu, 15 May 2008 06:48:08 +0000 (16:48 +1000)] 
Reduce openning of dev in create.

Now that validate_geometry opens and checks the device,
we don't need to do it as much in top level Create.
We only need it to check for old array or filesystem info.
So only open the device at that place.

12 years agoAdd 'container' level and ->validate_geometry method.
Neil Brown [Thu, 15 May 2008 06:47:41 +0000 (16:47 +1000)] 
Add 'container' level and ->validate_geometry method.

These will be used for ddf.

12 years agoRelease 2.6.5 mdadm-2.6.5
Neil Brown [Thu, 15 May 2008 05:50:59 +0000 (15:50 +1000)] 
Release 2.6.5

12 years agoGet Changelog uptodate
Neil Brown [Thu, 15 May 2008 05:50:57 +0000 (15:50 +1000)] 
Get Changelog uptodate

12 years agoCompiple fixes for mdassemble and diet-libc
Neil Brown [Thu, 15 May 2008 05:50:56 +0000 (15:50 +1000)] 
Compiple fixes for mdassemble and diet-libc

12 years agoDon't try the auto-detect test when md_mod is a module
Neil Brown [Thu, 15 May 2008 05:50:50 +0000 (15:50 +1000)] 
Don't try the auto-detect test when md_mod is a module this cannot work.

12 years agoFix support for --update=swapsuper
Neil Brown [Thu, 15 May 2008 05:50:48 +0000 (15:50 +1000)] 
Fix support for --update=swapsuper

The user of dup_super broke it.

12 years agoRecent change broken handling of metadata-less arrays.
Neil Brown [Thu, 15 May 2008 05:50:47 +0000 (15:50 +1000)] 
Recent change broken handling of metadata-less arrays.

In particular, failing a device would give a silly
error message.

12 years agoFix possible NULL dereference in super_by_fd
Neil Brown [Thu, 15 May 2008 05:50:45 +0000 (15:50 +1000)] 
Fix possible NULL dereference in super_by_fd

12 years agoAvoid possible null dereference in Assemble.
Neil Brown [Thu, 15 May 2008 05:50:42 +0000 (15:50 +1000)] 
Avoid possible null dereference in Assemble.

If the first device we look at has no superblock,
there is no 'st' to free, so don't free it.

12 years agoSet LOG_PID for syslog
Hans Lambermont [Mon, 12 May 2008 02:21:56 +0000 (12:21 +1000)] 
Set LOG_PID for syslog

From: Hans Lambermont <>

12 years agoadd --export option to --examine
Kay Sievers [Tue, 6 May 2008 00:02:38 +0000 (10:02 +1000)] 
add --export option to --examine

From: Kay Sievers <>

Cc: David Zeuthen <>
Signed-off-by: Kay Sievers <>
12 years agoMake device-special files for partitions when using --incremental
Neil Brown [Mon, 5 May 2008 11:55:39 +0000 (21:55 +1000)] 
Make device-special files for partitions when using --incremental

12 years agolet '-a' be specified for Incremental mode
Dan Williams [Mon, 5 May 2008 11:55:37 +0000 (21:55 +1000)] 
let '-a' be specified for Incremental mode

From: Dan Williams <>

Signed-off-by: Dan Williams <>
12 years agoUnify code into find_free_devnum.
Neil Brown [Mon, 5 May 2008 11:55:36 +0000 (21:55 +1000)] 
Unify code into find_free_devnum.

Two places have code to find a free md device number.  Make this
a subroutine.

12 years agofix load_super/free_super mismatch in util.c
Dan Williams [Mon, 5 May 2008 11:55:35 +0000 (21:55 +1000)] 
fix load_super/free_super mismatch in util.c

From: Dan Williams <>

Signed-off-by: Dan Williams <>
12 years agoSimplistig locking for --incremental.
Bill Nottingham [Mon, 5 May 2008 09:44:04 +0000 (19:44 +1000)] 
Simplistig locking for --incremental.

From: Bill Nottingham <>

mdadm --incremental doesn't really do any locking. If you get multiple
events in parallel for the same device (that has not yet started), they
will all go down the path to create the array. One will succeed, the
rest will have SET_ARRAY_INFO die with -EBUSY (md: array mdX already has disks!)
and will exit without adding the disk.

Original bug report is:

This is solved by adding very very rudimentary locking. Incremental() now
opens the device with O_EXCL to ensure only one invocation is frobbing the
array at once. A simple loop just tries to open 5 times a second for 5
seconds. If the array stays locked that long, you probably have bigger

12 years agoCreate /dev/md if needed and it doesn't exist.
Bill Nottingham [Mon, 5 May 2008 09:44:01 +0000 (19:44 +1000)] 
Create /dev/md if needed and it doesn't exist.

From: Bill Nottingham <>

12 years agoFix RAID acronym in man page.
Neil Brown [Mon, 5 May 2008 09:43:58 +0000 (19:43 +1000)] 
Fix RAID acronym in man page.

The 'D' in 'RAID' stands for 'DISKS' even it md supports other 'devices'.

12 years agoAdd a .gitignore file.
Neil Brown [Tue, 29 Apr 2008 07:14:01 +0000 (17:14 +1000)] 
Add a .gitignore file.

12 years agoUse .\" instead of '''
Neil Brown [Tue, 29 Apr 2008 07:13:59 +0000 (17:13 +1000)] 
Use .\" instead of '''

Debian bug 477273

12 years agodiff -ru mdadm-2.6.4-orig/Query.c mdadm-2.6.4/Query.c
Neil Brown [Tue, 29 Apr 2008 07:13:55 +0000 (17:13 +1000)] 
diff -ru mdadm-2.6.4-orig/Query.c mdadm-2.6.4/Query.c

12 years agoFix possible bug with bitmap space allocation with v1.0 metadata
Neil Brown [Tue, 29 Apr 2008 07:13:53 +0000 (17:13 +1000)] 
Fix possible bug with bitmap space allocation with v1.0 metadata

When adding a device to an array, make sure we don't reserve
so much space for the bitmap that there isn't room for the data.

12 years agoSmall improvements to --incremental for arrays that are in the middle of reshape
Neil Brown [Mon, 28 Apr 2008 06:30:31 +0000 (16:30 +1000)] 
Small improvements to --incremental for arrays that are in the middle of reshape

There is still a problem:  If array is partially assembled and started
read-only, the last device doesn't get added properly.  Probably a kernel

12 years agoAllow creation of a RAID6 with a single missing device.
Neil Brown [Mon, 28 Apr 2008 06:30:09 +0000 (16:30 +1000)] 
Allow creation of a RAID6 with a single missing device.

This did not work before as we couldn't mark it clean as there would
be some parity blocks out of sync, and raid6 will not assemble a
dirty degraded array.
So make such arrays doubly degraded (the last device becomes a spare)
and clean.

12 years agoUpdates some URLs in the man page.
Neil Brown [Mon, 28 Apr 2008 06:29:53 +0000 (16:29 +1000)] 
Updates some URLs in the man page.

Particularly, add the wiki on

12 years agoPrint 'Events' more sensibly.
Neil Brown [Mon, 28 Apr 2008 06:29:45 +0000 (16:29 +1000)] 
Print 'Events' more sensibly.

Instead of MSW.LSW, just print it as a 64bit number.

12 years agoFix problems with array.size overflowing on large arrays.
Neil Brown [Mon, 28 Apr 2008 06:29:37 +0000 (16:29 +1000)] 
Fix problems with array.size overflowing on large arrays.

array.size is 32bits and counts K.  So for arrays with
more than 4Terrabytes, it can overflow.
The correct number can be read from sysfs, but there are still
a few places that use array.size and risk truncation.  What is worse.
they compare a number of kilobytes with a number of sectors !!

So use get_component_size() to read the sysfs information, and be
more consistent about units.

12 years agoFix for segfault when reading /proc/mdstat
Neil Brown [Mon, 28 Apr 2008 06:29:12 +0000 (16:29 +1000)] 
Fix for segfault when reading /proc/mdstat

Some kernel versions don't put a space between 'active' and '(auto-read-only)'
in /proc/mdstat.  This causes a parsing problem leaving 'level' set to
NULL which causes a crash.

So synthesise a space there if it is missing, and check for 'level' to
be NULL and don't de-ref if it is.

12 years agoUse sysfs info for metadata version info in Detail and elsewhere.
Neil Brown [Fri, 14 Dec 2007 09:15:21 +0000 (20:15 +1100)] 
Use sysfs info for metadata version info in Detail and elsewhere.

12 years agoReplace sysarray with mdinfo
Neil Brown [Fri, 14 Dec 2007 09:14:59 +0000 (20:14 +1100)] 
Replace sysarray with mdinfo

Sure, mdinfo is bigger, but having a uniform structure for lots of things
will make life easier.

12 years agoUse 'mdinfo' instead of special 'sysdev' structure.
Neil Brown [Fri, 14 Dec 2007 09:14:57 +0000 (20:14 +1100)] 
Use 'mdinfo' instead of special 'sysdev' structure.

there is needless duplicatiion between mdinfo and sysdev, so discard
the latter.

12 years agoSome tidy up of 'devices' in assemble.
Neil Brown [Fri, 14 Dec 2007 09:14:53 +0000 (20:14 +1100)] 
Some tidy up of 'devices' in assemble.

use "mdinfo" in the devices list rather than just copying
ad-hoc fields in.

12 years agoDon't record oldmajor/oldminor any more during assemble.
Neil Brown [Fri, 14 Dec 2007 09:14:44 +0000 (20:14 +1100)] 
Don't record oldmajor/oldminor any more during assemble.

We don't use the information at all (any longer).

12 years agoFind super from fd on an array.
Neil Brown [Fri, 14 Dec 2007 09:14:38 +0000 (20:14 +1100)] 
Find super from fd on an array.

We used to use the major/minor numbers, but that isn't sufficient
any more, so pass the fd, and possibly check 'text' version.

12 years agoDrop the superblock arg from all metadata methods.
Neil Brown [Fri, 14 Dec 2007 09:14:33 +0000 (20:14 +1100)] 
Drop the superblock arg from all metadata methods.

It is now in the 'supertype'

12 years agoFix compare_super to take supertype instead of a superblock.
Neil Brown [Fri, 14 Dec 2007 09:14:27 +0000 (20:14 +1100)] 
Fix compare_super to take supertype instead of a superblock.

As this function takes 2 superblocks, the change is a bit more subtle,
so is done separately.

12 years agoAdd 'supertype' arg to almost all metadata methods.
Neil Brown [Fri, 14 Dec 2007 09:14:16 +0000 (20:14 +1100)] 
Add 'supertype' arg to almost all metadata methods.

The 'superblock' will be moved into this structure soon.

12 years agoAllow metadata handlers to free their own superblock.
Neil Brown [Fri, 14 Dec 2007 09:14:00 +0000 (20:14 +1100)] 
Allow metadata handlers to free their own superblock.

As the metadata handler allocates the superblock, it should free it
too.  DDF will have a more complex 'superblock' which needs more complex

12 years agoMake sure sysfs_read never succeeds for a non-MD device.
Neil Brown [Fri, 14 Dec 2007 09:13:51 +0000 (20:13 +1100)] 
Make sure sysfs_read never succeeds for a non-MD device.

... just in case.

12 years agoRemove spaces/tabs from ends of lines.
Neil Brown [Fri, 14 Dec 2007 09:13:43 +0000 (20:13 +1100)] 
Remove spaces/tabs from ends of lines.

12 years agoA recent patch broke --create --level=faulty - fix it.
Neil Brown [Thu, 29 Nov 2007 03:40:01 +0000 (14:40 +1100)] 
A recent patch broke --create --level=faulty - fix it.

Thanks to Martin Bene.

13 years agoAdd metadata version information to "--detail --brief" output.
Neil Brown [Mon, 12 Nov 2007 22:55:29 +0000 (09:55 +1100)] 
Add metadata version information to "--detail --brief" output.

13 years agoRelease 2.6.4 mdadm-2.6.4
Neil Brown [Fri, 19 Oct 2007 04:13:04 +0000 (14:13 +1000)] 
Release 2.6.4

13 years agoRemove partitions from a device before adding it to an array.
Neil Brown [Fri, 19 Oct 2007 04:03:17 +0000 (14:03 +1000)] 
Remove partitions from a device before adding it to an array.

The 'remove_partitions' call is in the wrong place and applying
to the wrong file.

13 years agoAdd some extra consistency checks based on level.
Neil Brown [Wed, 17 Oct 2007 03:08:12 +0000 (13:08 +1000)] 
Add some extra consistency checks based on level.

Also use LEVEL_LINEAR, LEVEL_MULTIPATH in place of -1 or -4 in a few places.

13 years agoClarify the avail/used devices sizes with version1 superblock.
Neil Brown [Wed, 17 Oct 2007 00:28:38 +0000 (10:28 +1000)] 
Clarify the avail/used devices sizes with version1 superblock.

Not all of the device may be available.  Of that, not all may be used
(if devices are of different sizes).

13 years agoMake sure --auto=mdp is honoured properly
Neil Brown [Wed, 17 Oct 2007 00:00:55 +0000 (10:00 +1000)] 
Make sure --auto=mdp is honoured properly

This would only create partitioned devices if the device name was a
'standard' name.  This this patch, any name will work.

13 years agoFix restarting of a reshaping array.
Neil Brown [Tue, 16 Oct 2007 03:52:35 +0000 (13:52 +1000)] 
Fix restarting of a reshaping array.

The last release broke the ability to assemble an array that
was in the middle of a reshape.
This patch adds code to test if the critical section needs
to be restored or not so that - if we have failed to restore it,
we know whether to fail or not.

13 years agoAlso use &tst instead of st later on when loading v1 superblocks
Jérémy Bobbio [Mon, 1 Oct 2007 08:17:06 +0000 (09:17 +0100)] 
Also use &tst instead of st later on when loading v1 superblocks

13 years agoFix segfault on assembly on amd64 with v1 superblocks
martin f. krafft [Sun, 30 Sep 2007 12:22:56 +0000 (13:22 +0100)] 
Fix segfault on assembly on amd64 with v1 superblocks

Commit a40b4fe introduced a temporary supertype variable tst, instead of
manipulating st directly. However, it was forgotton to pass &tst into the
recursive load_super1 call, causing an infinite recursion.

Signed-off-by: martin f. krafft <>
13 years agoMonitor.c s/MAXINT/INT_MAX/g
maximilian attems [Wed, 26 Sep 2007 16:12:46 +0000 (18:12 +0200)] 
Monitor.c s/MAXINT/INT_MAX/g

include limits.h directly instead of values.h

Signed-off-by: maximilian attems <>
Signed-off-by: Neil Brown <>
13 years agomdopen.c don't shadow major() and minor()
maximilian attems [Wed, 26 Sep 2007 16:12:47 +0000 (18:12 +0200)] 
mdopen.c don't shadow major() and minor()


Signed-off-by: maximilian attems <>
Signed-off-by: Neil Brown <>
13 years agoMonitor.c include signal.h directly
maximilian attems [Wed, 26 Sep 2007 16:12:45 +0000 (18:12 +0200)] 
Monitor.c include signal.h directly

another small step for better klibc support, glibc compile tested.

Signed-off-by: maximilian attems <>
Signed-off-by: Neil Brown <>
13 years agoAdd ANNOUNCE-2.6.1 to git
Neil Brown [Mon, 24 Sep 2007 04:40:23 +0000 (14:40 +1000)] 
Add ANNOUNCE-2.6.1 to git

Somehow this missed out of getting into .git even though it is in all
tha tar-balls.

13 years agoDon't corrupt 'supertype' when speculatively calling load_super1
Neil Brown [Mon, 24 Sep 2007 04:26:44 +0000 (14:26 +1000)] 
Don't corrupt 'supertype' when speculatively calling load_super1

When load_super1 is trying to see which sub-version of v1 superblock
is present, failure will cause it to clear st->ss, which is not good.

So use a temporary 'super_type' for the 'test if this version works'
calls, then copy that into 'st' on success.

13 years agoconfig.c include dirent.h instead sys/dir.h
maximilian attems [Fri, 21 Sep 2007 17:32:18 +0000 (19:32 +0200)] 
config.c include dirent.h instead sys/dir.h

sys/dir.h is an old BSD'ism, include dirent directly.
small step for better klibc support.

compile tested against glibc.

Signed-off-by: maximilian attems <>
Signed-off-by: Neil Brown <>
13 years agoCreate.c s/major/major_num/
maximilian attems [Fri, 21 Sep 2007 17:32:19 +0000 (19:32 +0200)] 
Create.c s/major/major_num/

don't shadow the major() function by int major.

Signed-off-by: maximilian attems <>
Signed-off-by: Neil Brown <>
13 years agoAdd klibc support to mdadm.h
maximilian attems [Fri, 21 Sep 2007 17:32:17 +0000 (19:32 +0200)] 
Add klibc support to mdadm.h

klibc still misses a lot functionality to let mdadm link against,
this small step helps to get to the real trouble.. :)

Signed-off-by: maximilian attems <>
Signed-off-by: Neil Brown <>
13 years agoMake "--write-mostly" effective when re-adding a device to an array.
Neil Brown [Mon, 24 Sep 2007 03:14:13 +0000 (13:14 +1000)] 
Make "--write-mostly" effective when re-adding a device to an array.

Fixes Debian Bug 442874

When we discover that we can 're-add' a drive, we forget to check the
write-mostly flag.
This highlights the fact that you cannot turn 'off' the write-mostly
flag at this point.  I wonder if that is a problem...

13 years agoExplain the read-balancing algorithm for RAID1 better in md.4
Iustin Pop [Tue, 11 Sep 2007 14:20:19 +0000 (16:20 +0200)] 
Explain the read-balancing algorithm for RAID1 better in md.4

From: Iustin Pop <>

There are many questions on the mailing list about the RAID1 read
performance profile. This patch adds a new paragraph to the RAID1
section in md.4 that details what kind of speed-up one should expect
from RAID1.

Signed-off-by: Iustin Pop <>
13 years agoRelease 2.6.3 mdadm-2.6.3
Neil Brown [Mon, 20 Aug 2007 04:22:15 +0000 (14:22 +1000)] 
Release 2.6.3

13 years agoUpdate changelog
Neil Brown [Mon, 20 Aug 2007 04:14:46 +0000 (14:14 +1000)] 
Update changelog

13 years agoReport error when grow cannot be restarted.
Neil Brown [Mon, 20 Aug 2007 04:14:42 +0000 (14:14 +1000)] 
Report error when grow cannot be restarted.

Make sure that if --assemble find an array in the critical region
of a reshape, and cannot find the critical data to restart the
reshape, it gives an error message.

13 years agoFix error message when adding a device that is too small.
Neil Brown [Mon, 20 Aug 2007 04:14:28 +0000 (14:14 +1000)] 
Fix error message when adding a device that is too small.

We were multiplying MD_RESERVED_SECTORS by 1024 instead of 512,
so lots of 'is it too small' checks were wrong.

13 years agoFix problem with add a device to a 1.x array created with older mdadm.
Neil Brown [Mon, 20 Aug 2007 04:14:25 +0000 (14:14 +1000)] 
Fix problem with add a device to a 1.x array created with older mdadm.

When adding new disk to an array, don't reserve so much bitmap
space that the disk cannot store the required data. (Needed when
1.x array was created with older mdadm).

13 years agoReject '--bitmap none' for build and create.
Neil Brown [Mon, 23 Jul 2007 07:28:44 +0000 (17:28 +1000)] 
Reject '--bitmap none' for build and create.

We don't want to create a file called 'none', and it isn't needed,
so just reject it.

13 years ago"--export" segfaults with non-persistent super blocks
Tim Woods [Mon, 23 Jul 2007 07:15:52 +0000 (17:15 +1000)] 
"--export" segfaults with non-persistent super blocks

From: Tim Woods <>

This patch fixes a NULL pointer dereference in Detail.c when running
'mdadm --detail --export' on a device with non-persistent super blocks.

13 years agoIn the MISC MODE section in mdadm(8), the names of the options should be
Neil Brown [Sun, 22 Jul 2007 23:32:01 +0000 (09:32 +1000)] 
In the MISC MODE section in mdadm(8), the names of the options should be
in bold.

  -- AYM 2007-01-14

diff -ur mdadm-2.6/mdadm.8 mdadm-2.6-man1/mdadm.8

13 years agoSubstantial corrections to man pages.
Scott Weikart [Fri, 13 Jul 2007 05:13:43 +0000 (15:13 +1000)] 
Substantial corrections to man pages.

From: "Scott Weikart" <>

Thanks Scott!

13 years agoTypo in man page: Multipath has an L ...
Neil Brown [Tue, 10 Jul 2007 23:08:49 +0000 (09:08 +1000)] 
Typo in man page: Multipath has an L ...

13 years agoAllow "--write-behind=" to be done in grow mode.
Ian Dall [Mon, 9 Jul 2007 01:29:04 +0000 (11:29 +1000)] 
Allow "--write-behind=" to be done in grow mode.

From: Ian Dall <>

I have a small patch to mdadm which allows the write-behind amount to be
set a array grow time (instead of currently only at grow or create
time). I have tested this fairly extensively on some arrays built out of
loop back devices, and once on a real live array.

13 years agoEnhance raid4 support: --assemble and --monitor wasn't quite happy with it.
Doug Ledford [Mon, 9 Jul 2007 00:00:02 +0000 (10:00 +1000)] 
Enhance raid4 support: --assemble and --monitor wasn't quite happy with it.

From: Doug Ledford <>

This one actually does a couple things.  Mainly related to raid4, but
kinda touches other raid levels some.

When creating a raid4 array, treat it like a raid5 array in that we
create it in degraded mode by default and add the last disk as a spare.
Besides speeding things up, this has a second effect that it makes mdadm
more consistent.  In order to create a degraded raid5 array, you need
only passing missing as one of the devices.  For a degraded raid4 array,
prior to this patch, you must pass assume-clean or else it refuses to
create the array.  Even force won't make it work without assume-clean.
With the patch, raid4 behaves identical to raid5.

Separate from that, the monitor functionality completely ignores raid4
arrays.  That seems to stem from the code that checks to see if the
array is part of a long list of types.  It seems easier to check which
array types *aren't* redundant instead of listing the ones that are
redundant and missing some of them.  This makes the monitor service
actually watch raid4 arrays.

13 years agoFix parsing of "-a" in various contexts.
Doug Ledford [Sun, 8 Jul 2007 23:59:59 +0000 (09:59 +1000)] 
Fix parsing of "-a" in various contexts.

From: Doug Ledford <>

This one fixes a bug where once manage mode is set, the -a short option
is no longer parsed correctly (true of grow mode as well).  This happens
because when you switch the short opts to the bitmap_auto version, it
specifies that the argument must follow a, yet the loop expects to get
an undecorated option and parse it as the disk dev instead of trying to
parse optarg.  So, create a new short opt array that is used for manage
and grow that doesn't list a as having an argument.

13 years agoMark some files FD_CLOEXEC to protect sendmail from them.
Doug Ledford [Sun, 8 Jul 2007 23:59:54 +0000 (09:59 +1000)] 
Mark some files FD_CLOEXEC to protect sendmail from them.

From: Doug Ledford <>

When running with SELinux enabled and using mdadm to monitor devices,
attempts to send emails to an admin will be blocked because mdadm is
holding open /proc/mdstat without setting the FD_CLOEXEC flag.  As a
result, sendmail has an open descriptor to /proc/mdstat after the
popen() call, which SELinux decides isn't really any of sendmail's
business and so sendmail gets denied.

13 years agoImprove error message when trying to create an array that already exists.
Doug Ledford [Sun, 8 Jul 2007 23:59:50 +0000 (09:59 +1000)] 
Improve error message when trying to create an array that already exists.

From: Doug Ledford <>

Simple bugfix.  If an array already exists and we are asked to create
this array, error out with an error message that makes sense to people
instead of an error that the SET_ARRAY_INFO ioctl had an invalid
argument.  Plus a typo correction.

13 years agoInterpret "--metadata=1" with --assemble to imply any version-1, not just 1.0
Doug Ledford [Sun, 8 Jul 2007 23:59:47 +0000 (09:59 +1000)] 
Interpret "--metadata=1" with --assemble to imply any version-1, not just 1.0

From: Doug Ledford <>

OK, this one fixes an issue where people were doing manual array
creation and specifying superblock types other than 1.0 (aka, 1.1, 1.2)
and then using mdadm -Ebs to populate their mdadm.conf file.  The
general problem is that if you specify a superblock type in the ARRAY
line (or on the command line), then you must specify the superblock type
*exactly*, including the minor version.  Unfortunately, mdadm -Ebs
prints out all version 1 superblocks, regardless of minor version, as
just plain old 1.  This breaks the mdadm.conf file for anything other
than plain version 1 superblock devices.

So, since I thought it was basically backwards that the mdadm -E output
was lax on specifying the location of the superblock where as the mdadm
-A input was strict, I reversed that.  With this patch, the mdadm -E
output is now exact for any given superblock.  But, in addition, the
mdadm -A input is now lax for any superblock that doesn't specifically
list the minor version, aka version 1 now means version 1, not version
0.90, but any minor version.  So does default/large.

13 years agoFix spare migration and other problems with --monitor.
Neil Brown [Sun, 8 Jul 2007 23:59:42 +0000 (09:59 +1000)] 
Fix spare migration and other problems with --monitor.

2.6 broke --monitor in various ways, including spare migration
stopped working.  This fixes it.

13 years agoCompile with various -O flags for testing.
Neil Brown [Mon, 21 May 2007 23:46:36 +0000 (09:46 +1000)] 
Compile with various -O flags for testing.

"make everything" now compiles with -O2 and -Os to increase coverage
for compiler generated warnings.

13 years agoFix compile error in Detail.c
Neil Brown [Mon, 21 May 2007 23:46:29 +0000 (09:46 +1000)] 
Fix compile error in Detail.c

'avail' is undefined if '--export', so make sure we don't try to use

13 years agoRelease 2.6.2 mdadm-2.6.2
Neil Brown [Mon, 21 May 2007 04:25:53 +0000 (14:25 +1000)] 
Release 2.6.2

13 years agoMake return code for "--detail --test" more reliable.
Neil Brown [Mon, 21 May 2007 04:25:50 +0000 (14:25 +1000)] 
Make return code for "--detail --test" more reliable.

Missing devices as well as failed devices cause an error.

13 years agoImprove error message for adding bitmap to a level that cannot support it.
Neil Brown [Mon, 21 May 2007 04:25:47 +0000 (14:25 +1000)] 
Improve error message for adding bitmap to a level that cannot support it.

Also give error on --build if no devices given.

13 years agoAdd --auto-detect for in-kernel autodetect.
Neil Brown [Mon, 21 May 2007 04:25:44 +0000 (14:25 +1000)] 
Add --auto-detect for in-kernel autodetect.

This is equivalent to raidautorun that some distros provide.

13 years agoUpdate tests and add linear-add
Neil Brown [Mon, 21 May 2007 04:25:40 +0000 (14:25 +1000)] 
Update tests and add linear-add

Update the testing scripts to allow for new space calculations
for space for bitmaps.
Add a test script for adding devices to linear arrays.

13 years agoFix --grow --add for linear arrays.
Neil Brown [Mon, 21 May 2007 04:25:37 +0000 (14:25 +1000)] 
Fix --grow --add for linear arrays.

The new superblock needs to have a new disk.number.  This is a bit of a hack...
Fix handling of negative bitmap offsets on 64bit hosts.

The bitmap offset is a signed 32bit number, so casting to (long)
isn't sufficient.  We must cast to (int32_t).
Fix various problems with --grow --add for linear.

The code to add a drive to a live linear array had never
been tested properly and so was buggy.  This tidies it up
and means that the new regression-test passes.

13 years agoFix handling of negative bitmap offsets on 64bit hosts.
Neil Brown [Mon, 21 May 2007 04:25:30 +0000 (14:25 +1000)] 
Fix handling of negative bitmap offsets on 64bit hosts.

The bitmap offset is a signed 32bit number, so casting to (long)
isn't sufficient.  We must cast to (int32_t).

13 years agoSupport failing and removed of detached and faulty devices.
Neil Brown [Fri, 11 May 2007 06:13:03 +0000 (16:13 +1000)] 
Support failing and removed of detached and faulty devices.

This if you unplug a device and udev removes the entry from /dev,
you can still remove the device.

13 years agoMinor typo in mdadm man page
David Huffman [Fri, 11 May 2007 06:12:54 +0000 (16:12 +1000)] 
Minor typo in mdadm man page

From: David Huffman <>

13 years agoAdd --export option to --detail to use key=value pairs.
Kay Sievers [Tue, 8 May 2007 07:17:33 +0000 (17:17 +1000)] 
Add --export option to --detail to use key=value pairs.

udev likes to get information about a device as key=value pairs so it
can create disk/by-id links etc.  So add --export flag which causes
the output of --detail to easily parsable.

From: Kay Sievers <>

13 years agoFix up calculation of bitmap space when creating v1 metadata.
Neil Brown [Tue, 8 May 2007 07:15:33 +0000 (17:15 +1000)] 
Fix up calculation of bitmap space when creating v1 metadata.

We have the same calculation in multiple places with subtle differences.
So unite it all.

Also fix up and endian problem in --examine.

13 years agoSend help text to stdout rather than stderr.
Neil Brown [Tue, 8 May 2007 07:14:00 +0000 (17:14 +1000)] 
Send help text to stdout rather than stderr.

This helps with piping to a pager for example.

13 years agoTypo in mdadm.conf man page
Neil Brown [Tue, 8 May 2007 07:13:51 +0000 (17:13 +1000)] 
Typo in mdadm.conf man page

13 years agoHyphens and Other Manpage Typography, 4/4: literal ASCII hyphens.
Peter Samuelson [Tue, 8 May 2007 07:13:44 +0000 (17:13 +1000)] 
Hyphens and Other Manpage Typography, 4/4: literal ASCII hyphens.

From: Peter Samuelson <>

nroff formats "-" as the very short hyphen used for hyphenated terms
and for splitting a word across two lines.  When you want a literal
ASCII "-", like for typing on a command line, you're supposed to use
"\-" instead.

Yeah, it sounds pedantic, but it actually makes a difference.  With
modern Unicode-capable terminals, "man" actually renders these with
different characters, so if you try to search for "--create" in your
favorite pager, you won't find it unless the nroff source says
"\-\-create".  This discrepancy doesn't generally show up with
non-Unicode terminals.

Signed-Off-By: Peter Samuelson <>