10 years agoDetail: export MD_UUID from mapfile
Dan Williams [Wed, 14 Oct 2009 00:41:57 +0000 (17:41 -0700)] 
Detail: export MD_UUID from mapfile

The load_super() from an mdadm --detail call may race against an mdmon
update.  When this happens the load_super sees an inconsistent metadata
block and returns an error.  The fallback path to use the map file
contents lacks uuid reporting, so provide __fname_from_uuid for
generically printing a uuid.

Reported-by: Hans de Goede <>
Signed-off-by: Dan Williams <>
10 years agoimsm: regression test for prodigal array member scenario
Dan Williams [Wed, 14 Oct 2009 00:41:53 +0000 (17:41 -0700)] 
imsm: regression test for prodigal array member scenario

Provide a test to sanity check assembly and reassembly in the presence
of conflicting family number information.

Signed-off-by: Dan Williams <>
10 years agoimsm: add --update=uuid support
Dan Williams [Wed, 14 Oct 2009 00:41:53 +0000 (17:41 -0700)] 
imsm: add --update=uuid support

When disks have conflicting container memberships (same container ids
but incompatible member arrays) --update=uuid can be used to move
offenders to a new container id by changing 'orig_family_num'.

Note that this only supports random updates of the uuid as the actual
uuid is synthesized.  We also need to communicate the new
'orig_family_num' value to all disks involved in the update.  A new
field 'update_private' is added to struct mdinfo to allow this
information to be transmitted.

Signed-off-by: Dan Williams <>
10 years agoddf: prevent superblock being zeroed on --update
Dan Williams [Wed, 14 Oct 2009 00:41:53 +0000 (17:41 -0700)] 
ddf: prevent superblock being zeroed on --update

The full fix would be to support updating ddf metadata, but this minimal
fix just prevents the superblock from being zeroed when someone
inadvertently passes an unsupported --update option during assembly.

Reported-by: Hans de Goede <>
Signed-off-by: Dan Williams <>
10 years agoimsm: fix/support --update
Dan Williams [Wed, 14 Oct 2009 00:41:53 +0000 (17:41 -0700)] 
imsm: fix/support --update

Fix init_super_imsm() to return an empty mpb when info == NULL, and
teach store_super_imsm() to simply write out the passed in mpb.


Reported-by: Hans de Goede <>
Signed-off-by: Dan Williams <>
10 years agoimsm: fix spare record writeout race
Dan Williams [Wed, 14 Oct 2009 00:41:53 +0000 (17:41 -0700)] 
imsm: fix spare record writeout race

imsm_activate_spare() in the manager thread may race against
write_super_imsm_spares() in the monitor thread.  Give
write_super_imsm_spares() its own private mpb buffer to prevent
confusing the manager.

This change uncovered cases where spares were not being assembled due to
a failed metadata version number check.  Spares can freely associate
across metadata version number, so reduce the scope of the version check
in the spare assembly case.

Signed-off-by: Dan Williams <>
10 years agoimsm: disambiguate family_num
Dan Williams [Wed, 30 Sep 2009 18:45:41 +0000 (11:45 -0700)] 
imsm: disambiguate family_num

This is a result of trawling through the Windows implementation to learn
the mechanism of how it disambiguates family_num.  It is a continuation
of commit 148acb7b "imsm: fix family number handling" which introduced a
regression when reassembling a container with stale disks and rebuilt

When rebuilding, a new family number is assigned to protect against the
"prodigal array member" problem.  It prevents a former family member
from returning to the system and causing a rebuild to go the wrong
direction.  However, this invalidates looking at the generation number to
determine the most up-to-date disk when comparing across family numbers.
Instead the assembly logic looks for agreement between a disk's local
family membership compared against a global list of all families in the
system.  Whenever a disk's local metadata does not match a family number
on the global list that family number is marked offline.

It is possible that this logic results in multiple incompatible but
valid family numbers existing in a container.  In this case mdadm.conf
cannot be consulted because it only records the uuid which is generated
from static fields in the metadata.  The metadata lacks the data needed
to disambiguate "local" versus "foreign".  The "foreign" array in this
case requires updating to change its container-id information
(orig_family_num), and possibly the member array names.

Signed-off-by: Dan Williams <>
10 years agoimsm: kill close() of component device
Dan Williams [Wed, 30 Sep 2009 18:44:38 +0000 (11:44 -0700)] 
imsm: kill close() of component device

None of the other formats close the passed in fd at load, and this
becomes a problem when trying to support --update where we need O_EXCL
protection across the entire operation.

Signed-off-by: Dan Williams <>
10 years agoimsm: cleanup disk status tests
Dan Williams [Mon, 28 Sep 2009 21:40:59 +0000 (14:40 -0700)] 
imsm: cleanup disk status tests

Add is_failed(), is_configured(), and is_spare() helpers to clean up
disk status flag testing.

Signed-off-by: Dan Williams <>
10 years agoRelease mdadm-3.0.2 mdadm-3.0.2
NeilBrown [Fri, 25 Sep 2009 08:19:07 +0000 (18:19 +1000)] 
Release mdadm-3.0.2
Just one bugfix.

10 years agosuper0: fix crash on assemble if homehost is not set.
NeilBrown [Fri, 25 Sep 2009 07:56:22 +0000 (17:56 +1000)] 
super0: fix crash on assemble if homehost is not set.

If homehost is not set - typically during early boot,
and assemble of v0.90 metadata arrays will crash.

Reported-by: PaweĊ‚ Sikora <>
Signed-off-by: NeilBrown <>
10 years agoRelease mdadm-3.0.1 mdadm-3.0.1
NeilBrown [Fri, 25 Sep 2009 07:08:19 +0000 (17:08 +1000)] 
Release mdadm-3.0.1

Just bugfixes.

Signed-off-by: NeilBrown <>
10 years agotestreshape5 - flush devices between tests.
NeilBrown [Fri, 25 Sep 2009 06:57:01 +0000 (16:57 +1000)] 
testreshape5 - flush devices between tests.

We need to flush the block devices before reading different data.

Signed-off-by: NeilBrown <>
10 years agoMerge branch 'master' of git://
NeilBrown [Fri, 25 Sep 2009 04:11:11 +0000 (14:11 +1000)] 
Merge branch 'master' of git://

10 years agomdmon: fix freeing unallocated memory
Hans de Goede [Thu, 24 Sep 2009 13:52:06 +0000 (06:52 -0700)] 
mdmon: fix freeing unallocated memory

mdmon was creating a supertype struct with malloc, and thus not
necessarily getting zero-d memory.

This was causing it to segfault when called like this from the initrd:
/sbin/mdmon /proc/mdstat /sysroot

The problem was that  load_super_imsm would get called on the non-zero'd
super struct, whcih in turn calls free_super_imsm, which checks st->sb,
which should be zero but isn't and then starts freeing bogus memory.

Signed-off-by: Dan Williams <>
11 years agoimsm: clear CONFIGURED_DISK for failed drives
Dan Williams [Tue, 15 Sep 2009 18:35:28 +0000 (11:35 -0700)] 
imsm: clear CONFIGURED_DISK for failed drives

Synchronizing with what the Windows driver does.

Signed-off-by: Dan Williams <>
11 years agoimsm: kill USABLE_DISK flag
Dan Williams [Tue, 15 Sep 2009 18:35:28 +0000 (11:35 -0700)] 
imsm: kill USABLE_DISK flag

'USABLE_DISK' is not a 'persistent' status flag it is an internal status
flag used for the in memory representation of the disk in the Windows

Signed-off-by: Dan Williams <>
11 years agoExamine: don't count containers as spares
Dan Williams [Tue, 15 Sep 2009 18:35:28 +0000 (11:35 -0700)] 
Examine: don't count containers as spares

mdadm -Ebs will include containers in the scanned device list.
Examine() falsely thinks they are spares when MD_DISK_SYNC is not set.
This could be fixed by forcing all formats to set this flag for
container devices, but this flag is currently used by imsm to identify
free-floating spares.

Signed-off-by: Dan Williams <>
11 years agoDetail: fix for an imsm container with a spare
Dan Williams [Tue, 15 Sep 2009 18:34:20 +0000 (11:34 -0700)] 
Detail: fix for an imsm container with a spare

Spares for imsm arrays do not have any info about the container in their
metadata records.  If Detail() inadvertantly picks such a device for
->get_array_info() it will end up with less than useful info for the
container.  So, continue to read from the disks until a non-spare device
is found.

This bug was found by timeouts waiting for udev to create the
user-friendly container name.  To detect future UUID reporting problems
and a debug print to the timeout case in wait_for().

Signed-off-by: Dan Williams <>
11 years agoExamine: fixup output in the presence of containers with spares
Dan Williams [Tue, 15 Sep 2009 18:34:20 +0000 (11:34 -0700)] 
Examine: fixup output in the presence of containers with spares

If we dump any 'spare' or 'device' information for a container in the
'brief' case then we need a newline before printing member array info.

Signed-off-by: Dan Williams <>
11 years agoimsm: fix spare promotion
Dan Williams [Tue, 15 Sep 2009 18:34:20 +0000 (11:34 -0700)] 
imsm: fix spare promotion

1/ Fix an off by one error when detecting whether the device allocation
   loop succeeded or not
2/ Update ->num_raid_devs before copying to avoid a segmentation fault

Signed-off-by: Dan Williams <>
11 years agoAdd a missing 'closedir'.
NeilBrown [Fri, 11 Sep 2009 06:10:24 +0000 (16:10 +1000)] 
Add a missing 'closedir'.

Thanks to David Binderman for finding and reporting it.

Signed-off-by: NeilBrown <>
11 years agosuper1: remove fd leak when opening /dev/urandom
NeilBrown [Thu, 13 Aug 2009 05:02:39 +0000 (15:02 +1000)] 
super1: remove fd leak when opening /dev/urandom

As reported in

I forgot to close the fd after reading the random number.

Signed-off-by: NeilBrown <>
11 years agoExmaine/brief: put member arrays after container arrays.
NeilBrown [Fri, 7 Aug 2009 04:17:40 +0000 (14:17 +1000)] 
Exmaine/brief: put member arrays after container arrays.

A previous patch moved move the '--examine --brief' reporting of
member arrays to before their containers.  This breaks "mdadm -As"
assembly.  So put them back, but still fix the problem addressed by
previous patch.

Signed-off-by: NeilBrown <>
11 years agoMerge branch 'master' of git://
NeilBrown [Fri, 7 Aug 2009 03:45:38 +0000 (13:45 +1000)] 
Merge branch 'master' of git://

11 years agoAssemble: fix handling of empty container
Dan Williams [Tue, 4 Aug 2009 17:17:23 +0000 (10:17 -0700)] 
Assemble: fix handling of empty container

# mdadm --create /dev/md/ddf /dev/sd[b-e] -n 4 -e ddf
mdadm: container /dev/md/ddf prepared.
# mdadm -Ss
mdadm: stopped /dev/md126
# mdadm -As
mdadm: Container /dev/md/ddf0 has been assembled with 4 drives
Segmentation fault

Reported-by: Artur Wojcik <>
Reported-by: Jacek Danecki <>
Signed-off-by: Dan Williams <>
11 years agoimsm: fix spare-uuid assignment
Dan Williams [Sat, 1 Aug 2009 00:11:42 +0000 (17:11 -0700)] 
imsm: fix spare-uuid assignment

imsm spares do not have container membership by default so we associate
them with the first container found in the configuration file.  Some
ARRAY lines do not specify the metadata type so we cannot assume that
_cst will always be valid.

Signed-off-by: Dan Williams <>
11 years agoplatform: relax rom scanning alignment for ahci platforms
Dan Williams [Sat, 1 Aug 2009 00:11:41 +0000 (17:11 -0700)] 
platform: relax rom scanning alignment for ahci platforms

The PCI-3.0 Firmware specification allows for option-roms to have
512-byte alignment rather than 2048-byte.  As there does not appear to
be a reliable method to detect a PCI-3.0 compliant BIOS from userspace
we allow the imsm platform detection code to presume that a system
modern enough to have an Intel AHCI controller does not have
dangerous/legacy ISA regions in the option-ROM memory space.

An environment variable to disable this behaviour, IMSM_SAFE_OROM_SCAN,
is added in case this presumption is ever proven wrong.

Reported-by: Hans de Goede <>
Signed-off-by: Dan Williams <>
11 years agoimsm: fix family number handling
Dan Williams [Sat, 1 Aug 2009 00:11:41 +0000 (17:11 -0700)] 
imsm: fix family number handling

The family_number field can change.  The option-rom will change the
family number when it starts a rebuild process (flags a container for
rebuild).  This was not seen previously as mdadm would usually start the
rebuild process, preserving the family number.

This is the mechanism that helps to prevent a prodigal array member from
being returned to its original system and cause a rebuild to go in the
wrong direction.  With the change we will end up with a container that
will fail to assemble unless the device with the incompatible family
number is left out of the assembly.

So, take several actions:
1/ Convert uuid generation to use orig_family_num, being careful to
   preserve the existing uuid in the case where orig_family_num is not
   set (i.e. previous mdadm created imsm arrays)
2/ Set orig_family_num at Create.  For arrays created by mdadm prior to
   this release orig_family_num will be zero, so set it to family_num at
   the first metadata write.
3/ Add checks for orig_family_num to compare_super_imsm
4/ Update the family number when initiating rebuild
5/ The option-rom mixes some random data into the family number, add
   this functionality to the mdadm implementation.

Reported-by: Marcin Labun <>
Signed-off-by: Dan Williams <>
11 years agoimsm: fix activate_spare off-by-one
Dan Williams [Sat, 1 Aug 2009 00:11:41 +0000 (17:11 -0700)] 
imsm: fix activate_spare off-by-one

The last sector of an array is calculated by start + size - 1.

Reported-by: Rafal Marszewski <>
Reported-by: Jarema Bielanski <>
Signed-off-by: Dan Williams <>
11 years agoconditionally update uuids in the map file after Create()
Dan Williams [Sat, 1 Aug 2009 00:11:41 +0000 (17:11 -0700)] 
conditionally update uuids in the map file after Create()

The map file needs to be updated after adding the first member array to
an Intel metadata container.  The uuid for an imsm container uses the
->family_num field of the metadata.  This field is static, but is only
set after the first member array has been created.  Prior to this all
devices are free floating spares and do not have any information that
can identify specific container membership.  At Create() time we take
the uninitialized uuid from ->get_info_super() prior to updating the
metadata.  So the current result is:

# mdadm --create /dev/md/imsm /dev/sd[b-e] -n 4 -e imsm
# mdadm --create /dev/md/vol0 /dev/md/imsm -n 4 -l 0
# cat /var/run/mdadm/map
md126 /md127/0 3e03aee2:78c3c593:1e8ecaf0:eefb53ed /dev/md/vol0
md127 imsm 53d6f8b1:7a783f24:f30483c5:705c48c7 /dev/md/imsm
# mdadm -Ebs
ARRAY metadata=imsm UUID=589d2d2c:4221a54d:acb63c06:c3907f52
ARRAY /dev/md/vol0 container=589d2d2c:4221a54d:acb63c06:c3907f52
member=0 UUID=57b89b63:5cd0eae1:17dd26b3:51cc78d4

So, before we write out the new metadata check to see if the member
array uuid has changed as a result of this addition.  If it has, update
its uuid in the map file and flag its parent container for updating.  In
support of updating the container uuid the semantics of
->write_init_super are changed to clear any metadata specific member
array cursors (e.g. ddf_super.currentconf or intel_super.current_vol)
such that a subsequent call to ->getinfo_super returns container

Reported-by: Ignacy Kasperowicz <>
Signed-off-by: Dan Williams <>
11 years agoimsm: fixup examine_brief to be more descriptive in the container only case
Dan Williams [Sat, 1 Aug 2009 00:11:41 +0000 (17:11 -0700)] 
imsm: fixup examine_brief to be more descriptive in the container only case

Prior to creating any arrays in a new container the output from -Ebs for
a 4-disk imsm array returns:


We should at least display that these are imsm spares:

ARRAY metadata=imsm

Signed-off-by: Dan Williams <>
11 years agofix examine_brief segfault
Dan Williams [Sat, 1 Aug 2009 00:11:41 +0000 (17:11 -0700)] 
fix examine_brief segfault

When performing an "-Ebs -e <metadata type>" we segfault because the
superblock has been freed too early.  We also leak memory for 'ddf' and
'imsm' because, unlike super[01], we do not implicitly free when
->load_super is called on an already loaded supertype.

So, fix up imsm and ddf to match type 0 and 1 ->load_super() semantics,
and update Examine to not free the superblock until all usages have been

Signed-off-by: Dan Williams <>
11 years agofix RebuildMap() to retrieve 'subarray' info
Dan Williams [Sat, 1 Aug 2009 00:11:41 +0000 (17:11 -0700)] 
fix RebuildMap() to retrieve 'subarray' info

RebuildMap falsely returns container info for member arrays.  Retrieving
the subarray and container_dev details prior to ->load_super() changes the
result from:

md127 imsm 082c6371:74b5ce03:64972e41:6b0860d5 /dev/md/imsm
md126 imsm 082c6371:74b5ce03:64972e41:6b0860d5 /dev/md/vol0

md126 /md127/0 3e03aee2:78c3c593:1e8ecaf0:eefb53ed /dev/md/vol0
md127 imsm 082c6371:74b5ce03:64972e41:6b0860d5 /dev/md/imsm

Reported-by: Ignacy Kasperowicz <>
Signed-off-by: Dan Williams <>
11 years agoteach imsm and ddf what st->subarray means at load_super time
Dan Williams [Sat, 1 Aug 2009 00:08:22 +0000 (17:08 -0700)] 
teach imsm and ddf what st->subarray means at load_super time

RebuildMap wants to poll through mdstat and retrieve a (kernel name,
uuid, user name) tuple for each array.  Teach imsm and ddf to honor
st->sub_array at ->load_super() time to set their internal subarray
pointers to the value specified in st->subarray, or return an error if
st->subarray specifies an invalid array.

Signed-off-by: Dan Williams <>
11 years agoMonitor: use pclose rather than fclose
NeilBrown [Fri, 10 Jul 2009 04:39:20 +0000 (14:39 +1000)] 
Monitor: use pclose rather than fclose

Using pclose is probably the right thing to do seeing that we
used popen, but as there is no clear need to wait for sendmail
to finish, it isn't really important.

Signed-off-by: NeilBrown <>
11 years agoExamine: make --metadata= work with --brief
NeilBrown [Fri, 10 Jul 2009 00:44:32 +0000 (10:44 +1000)] 
Examine: make --metadata= work with --brief

They had different assumptions about the lifetime of 'st'.

Signed-off-by: NeilBrown <>
11 years agoExamine: fix --examine --brief --verbose on containers.
NeilBrown [Thu, 4 Jun 2009 02:44:32 +0000 (12:44 +1000)] 
Examine: fix --examine --brief --verbose on containers.

With --verbose, --examine --brief prints dev= information after
the personality has done its bit.
But with containers, the member array are printed in between.
So in super-ddf and super-intel, move printing of the member
arrays to before printing of the container.  This avoids

Signed-off-by: NeilBrown <>
11 years agosuper-intel: fix test on failed_disk_num.
NeilBrown [Thu, 4 Jun 2009 02:29:21 +0000 (12:29 +1000)] 
super-intel: fix test on failed_disk_num.

We sometimes set failed_disk_num to ~0.
However we cannot test for equality with that as  failed_disk_num
is 8bit and ~0 is probably 32bit with lots of 1's.
So test if ~failed_disk_num is 0 instead.

Reported-By: "Mr. James W. Laferriere" <>
Signed-off-by: NeilBrown <>
11 years agoRelease mdadm-3.0 devel-3.0 mdadm-3.0
NeilBrown [Tue, 2 Jun 2009 05:37:56 +0000 (15:37 +1000)] 
Release mdadm-3.0

 - remove lot of old 'ANNOUNCE' files
 - trim changelog
 - update version numbers

Signed-off-by: NeilBrown <>
11 years agoMerge branch 'master' of git:// into devel-3.0
NeilBrown [Tue, 2 Jun 2009 05:28:41 +0000 (15:28 +1000)] 
Merge branch 'master' of git:// into devel-3.0

11 years agoMerge branch 'master' into devel-3.0
NeilBrown [Tue, 2 Jun 2009 05:28:36 +0000 (15:28 +1000)] 
Merge branch 'master' into devel-3.0


11 years agoMove WaitClean from Monitor.c to sysfs.c
NeilBrown [Tue, 2 Jun 2009 05:27:16 +0000 (15:27 +1000)] 
Move WaitClean from Monitor.c to sysfs.c

That way mdmon doesn't need to include Monitor.o

Signed-off-by: NeilBrown <>
11 years agoUpdate copyright dates and remove references to
NeilBrown [Tue, 2 Jun 2009 04:35:45 +0000 (14:35 +1000)] 
Update copyright dates and remove references to

Also removed 'paper' addresses.

Signed-off-by: NeilBrown <>
11 years agoMonitor: reduce default poll interval if mdstat is pollable.
NeilBrown [Tue, 2 Jun 2009 04:35:44 +0000 (14:35 +1000)] 
Monitor:  reduce default poll interval if mdstat is pollable.

Since 2.6.16, mdstat responds to select/poll.
So in that case, increase the default poll interval to about 15
This ensures that the background load is insignificant.

Signed-off-by: NeilBrown <>
11 years agoMonitor: don't get confused if utime is never set.
NeilBrown [Tue, 2 Jun 2009 04:24:58 +0000 (14:24 +1000)] 
Monitor: don't get confused if utime is never set.

externally managed arrays do not (currently) cause utime in
GET_ARRAY_INFO to be updated.  So if it is zero, just assume the
current time.
This will cause GET_DISK_INFO to be called more often, but as we do
the scan only every 60 seconds normally, a few extra syscalls isn't
going to make a big difference.

Signed-off-by: NeilBrown <>
11 years agomdadm.8: Man page updates
NeilBrown [Tue, 2 Jun 2009 04:06:05 +0000 (14:06 +1000)] 
mdadm.8: Man page updates

General review and update of mdadm.8

11 years agoimsm: kill "auto=" in brief_examine_super_imsm
Dan Williams [Mon, 18 May 2009 17:02:58 +0000 (10:02 -0700)] 
imsm: kill "auto=" in brief_examine_super_imsm

The auto parameter is obsolete after kernel version 2.6.28 as all arrays
are partitionable via block device extended minor support.  Environments
that requre the mdp style of array can always edit the configuration
file to specify auto=mdp.

Signed-off-by: Dan Williams <>
11 years agoimsm: fix num_domains
Dan Williams [Mon, 18 May 2009 16:58:55 +0000 (09:58 -0700)] 
imsm: fix num_domains

The 'num_domains' field simply identifies the number of mirrors.  So it
is 2 for a 2-disk raid1 or a 4-disk raid10.  The orom does not currently
support more than 2 mirrors, but a three disk raid1 for example would
increase num_domains to 3.

Signed-off-by: Dan Williams <>
11 years agoMonitor: support spare-group manipulation for 1.x metadata.
NeilBrown [Mon, 11 May 2009 23:49:45 +0000 (09:49 +1000)] 
Monitor: support spare-group manipulation for 1.x metadata.

The code for moving spares around a spare-group currently
only works for 0.90 metadata.  Generalise it for 1.x metadata
as well.

Reported-by: "Garth Snyder" <>
Signed-off-by NeilBrown <>

11 years agoIgnore leading zeros in version number information.
NeilBrown [Mon, 11 May 2009 23:49:06 +0000 (09:49 +1000)] 
Ignore leading zeros in version number information.

--detail sometimes generates leading zero which are just noise.

11 years agoRelease mdadm-3.0-rc1 mdadm-3.0-rc1
NeilBrown [Mon, 11 May 2009 06:33:29 +0000 (16:33 +1000)] 
Release mdadm-3.0-rc1

11 years agoMerge branch 'master' into devel-3.0
NeilBrown [Mon, 11 May 2009 06:05:41 +0000 (16:05 +1000)] 
Merge branch 'master' into devel-3.0


11 years agocreate_mddev: don't replace /dev/mdX with /dev/md/X
NeilBrown [Mon, 11 May 2009 05:58:44 +0000 (15:58 +1000)] 
create_mddev: don't replace /dev/mdX with /dev/md/X

If someone creates/assemble an array called "/dev/md0", don't force
it to be "/dev/md/0".  Doing so isn't really necessary and it
likely to confuse people.

Signed-off-by: NeilBrown <>
11 years agomapfile - when rebuilding, choose an appropriate name is none is found.
NeilBrown [Mon, 11 May 2009 05:58:42 +0000 (15:58 +1000)] 
mapfile - when rebuilding, choose an appropriate name is none is found.

When rebuilding the mapfile (mdadm -Ir), if not appropriate name is
found in /dev/md/, try to find an appropriate name, either by looking
in mdadm.conf or by using the name in the metadata.

Signed-off-by: NeilBrown <>
11 years agoIncr: use devname_matches to when looking in mdadm.conf for bitmap file
NeilBrown [Mon, 11 May 2009 05:47:11 +0000 (15:47 +1000)] 
Incr: use devname_matches to when looking in mdadm.conf for bitmap file

This is more likely to always do the right thing than a strcmp.

Signed-off-by: NeilBrown <>
11 years agomapfile - Fix off-by-one error in RebuildMap
NeilBrown [Mon, 11 May 2009 05:47:11 +0000 (15:47 +1000)] 
mapfile - Fix off-by-one error in RebuildMap

"mdadm -Ir" would get the path for md0 wrong because it
went looking for mdp(-1) by mistake.

Signed-off-by NeilBrown <>

11 years agoFix printf compile warning.
NeilBrown [Mon, 11 May 2009 05:47:10 +0000 (15:47 +1000)] 
Fix printf compile warning.

It always afters to cast big things to (unsigned long long) before
printing as %llu - it seems there will always be one arch which
has something to complain about ....

Signed-off-by: NeilBrown <>
11 years agoIncremental: fix uninitialised variable.
NeilBrown [Mon, 11 May 2009 05:47:10 +0000 (15:47 +1000)] 
Incremental: fix uninitialised variable.

st2 might not be initialised at this point.  So use the
more correct 'st'.

Signed-off-by: NeilBrown <>
11 years agoIncr: cope better with possibility that mp->path might be NULL
NeilBrown [Mon, 11 May 2009 05:47:10 +0000 (15:47 +1000)] 
Incr: cope better with possibility that mp->path might be NULL

Signed-off-by: NeilBrown <>
11 years agomapfile: allow the path name to the device to be empty.
NeilBrown [Mon, 11 May 2009 05:47:10 +0000 (15:47 +1000)] 
mapfile:  allow the path name to the device to be empty.

Allowing an empty name and coping with it is less confusing
than seeing "/empty" appear.

Signed-off-by: NeilBrown <>
11 years agomap_dev: prefer names in /dev/md/
NeilBrown [Mon, 11 May 2009 05:47:10 +0000 (15:47 +1000)] 
map_dev: prefer names in /dev/md/

Rather than preferring non-standard names (of which there are
many, like /dev/block/9:1), prefer names in /dev/md/ when finding
the name of an md device.

Signed-off-by: NeilBrown <>
11 years agoudev-md-raid.rules: use string_escape
NeilBrown [Mon, 11 May 2009 05:47:10 +0000 (15:47 +1000)] 
udev-md-raid.rules: use string_escape

For safety, "MD_NAME" strings should be escaped in udev.

Signed-off-by: NeilBrown <>
11 years agosuper1 - fix brief examine.
NeilBrown [Mon, 11 May 2009 05:47:10 +0000 (15:47 +1000)] 
super1 - fix brief examine.

We manage to lose some space here...

Signed-off-by: NeilBrown <>
11 years agoman pages: note that --auto is largely deprecated and irrelevant.
NeilBrown [Mon, 11 May 2009 05:47:10 +0000 (15:47 +1000)] 
man pages: note that --auto is largely deprecated and irrelevant.

Signed-off-by: NeilBrown <>
11 years agoBe more consistent about keeping the host: prefix on array names.
NeilBrown [Mon, 11 May 2009 05:47:10 +0000 (15:47 +1000)] 
Be more consistent about keeping the host: prefix on array names.

If an array name contains a "hostname:" prefix, then
--assemble will tend to leave it there, while --incremental
will strip it off (when chosing a device name during auto-assembly).

Make this more consistent:  strip the name off if we decide that
the name will be treated as 'local'.  Leave it on if it will be
treated as 'foreign'.

Signed-off-by: NeilBrown <>
11 years agoconfig: enhance name matching in conf_get_ident.
NeilBrown [Mon, 11 May 2009 05:47:10 +0000 (15:47 +1000)] 
config: enhance name matching in conf_get_ident.

Use when searching mdadm.conf for a device, use more flexible
matching that e.g. ignores leading /dev/md/ or /dev/

As mdadm now accepts both "/dev/md/foo" and "foo" is many places as
equivalent, they should compare as the same.

Signed-off-by: NeilBrown <>
11 years agoAllow homehost to be largely ignored when assembling arrays.
NeilBrown [Mon, 11 May 2009 05:46:46 +0000 (15:46 +1000)] 
Allow homehost to be largely ignored when assembling arrays.

If mdadm.conf contains
   HOMEHOST <ignore>
or commandline contains

then the check that array metadata mentions the given homehost is
replace by a check that the name recorded in the metadata is not
already used by some other array mentioned in mdadm.conf.

This allows more arrays to use their native name rather than having
an _NN suffix added.

This should only be used during boot time if all arrays required for
normal boot are listed in mdadm.conf.

If auto-assembly is used to find all array during boot, then the
HOMEHOST feature should be used to ensure there is no room for
confusion in choosing array names, and so it should not be set
to <ignore>.

Signed-off-by: NeilBrown <>
11 years agoutil: fix test for text_version
NeilBrown [Mon, 11 May 2009 05:21:43 +0000 (15:21 +1000)] 
util: fix test for text_version

as text_version is a char array (not a pointer), testing the
address against NULL is the wrong thing to do.  Test the
content instead.

Signed-off-by: NeilBrown <>
11 years agoAssemble/Incr : minor tidy up of setting 'trustworthy'.
NeilBrown [Mon, 11 May 2009 05:19:30 +0000 (15:19 +1000)] 
Assemble/Incr : minor tidy up of setting 'trustworthy'.

Signed-off-by: NeilBrown <>
11 years agoconfig: allow Array line to contain array name without /dev/md/ prefix.
NeilBrown [Mon, 11 May 2009 05:18:35 +0000 (15:18 +1000)] 
config: allow Array line to contain array name without /dev/md/ prefix.

For consistency with --create and --assemble, allow the array name
given in mdadm.conf to exclude the "/dev/md/" prefix.  So e.g.

  ARRAY home uuid=whatever

is treated like

  ARRAY /dev/md/home uuid=whatever

Also exclude names which create_mddev will reject.

Signed-off-by: NeilBrown <>
11 years agoFix tests on ->container and ->member
NeilBrown [Mon, 11 May 2009 05:18:25 +0000 (15:18 +1000)] 
Fix tests on ->container and ->member

For container= and member= to be effective in an mdadm.conf line
they must both be present.  So when checking for their absence we
need container != NULL || member != NULL.

Signed-off-by: NeilBrown <>
11 years agoMake --brief even briefer.
NeilBrown [Mon, 11 May 2009 05:18:20 +0000 (15:18 +1000)] 
Make --brief even briefer.

Because ---examine --brief, or --detail --brief are
often used to create mdadm.conf, and because people don't want to
have to update their mdadm.conf unnecessarily, we don't want to
include information that might change.
And now that level changing is supported, that is almost everything
but UUID.

So move some more fields into the "Only print with --verbose" class.

Signed-off-by: NeilBrown <>
11 years agomdadm.8 general updates
NeilBrown [Mon, 11 May 2009 05:17:50 +0000 (15:17 +1000)] 
mdadm.8  general updates

Signed-off-by: NeilBrown <>
11 years agoconf/assemble: new config line "auto".
NeilBrown [Mon, 11 May 2009 05:17:33 +0000 (15:17 +1000)] 
conf/assemble:  new config line "auto".

The line 'auto' in mdadm.conf can be used to disable assembly
of specific metadata types, or of all arrays.

This does not affect assembly of arrays listed in mdadm.conf
or on command line.

 auto -all

will disable all auto-assembly.

 auto -ddf

will cause mdadm to ignore ddf arrays that are not explicitly
mentioned, and auto assemble anything else it finds.

Signed-off-by: NeilBrown <>
11 years agoconfig: support "ARRAY <ignore> ..." lines in mdadm.conf
NeilBrown [Mon, 11 May 2009 05:17:05 +0000 (15:17 +1000)] 
config: support "ARRAY <ignore> ..." lines in mdadm.conf

Sometimes we want to ensure particular arrays are never
assembled automatically.  This might include an array made of
devices that are shared between hosts.

To support this, allow ARRAY lines in mdadm.conf to use the word
"ignore" rather than a device name.  Arrays which match such lines
are never automatically assembled (though they can still be assembled
by explicitly giving identification information on the mdadm command

Signed-off-by: NeilBrown <>
11 years agoassemble: support arrays created with --homehost=any
NeilBrown [Mon, 11 May 2009 05:16:49 +0000 (15:16 +1000)] 
assemble:  support arrays created with --homehost=any

If an array is created with --homehost=any, then --assemble and
--incremental will treat it as being local to 'this' host, no matter
what the name of this host is.

This is useful for array that will be given unique names and be
moved between machines.

This needs to be documented.

Signed-off-by: NeilBrown <>
11 years agocreate_dev - allow array names like mdX and /dev/mdX to appear 'numeric'
NeilBrown [Mon, 11 May 2009 05:16:47 +0000 (15:16 +1000)] 
create_dev - allow array names like mdX and /dev/mdX to appear 'numeric'

When choosing the minor number to use with an array, we currently base
the number of the 'name' stored in the metadata if that name is
Extend that so that if it looks like a number md device name (/dev/md0
or just md0 or even /dev/md/0), then we use the number at the end to
suggest a minor number.

The means that if someone creates and array with "--name md0" or even
"--name /dev/md0" it will continue to do what they expect.

Signed-off-by: NeilBrown <>
11 years agore-add error check
NeilBrown [Tue, 5 May 2009 11:13:29 +0000 (21:13 +1000)] 
re-add error check

11 years agoFix gcc-4.4 compiler warning.
NeilBrown [Wed, 29 Apr 2009 01:44:02 +0000 (11:44 +1000)] 
Fix gcc-4.4 compiler warning.

Apparently the dereferencing of a type-punned pointer breaks strict
aliasing rules.   And we wouldn't want to do that.
So just make a different array of the appropriate type and use memcpy.

Resolves-Debian-bug: 505375
Signed-off-by: NeilBrown <>
11 years agoMakefile: use $(CC) more consistently.
NeilBrown [Wed, 29 Apr 2009 01:21:08 +0000 (11:21 +1000)] 
Makefile: use $(CC) more consistently.

Explicitly calling 'gcc' in some rules makes it hard to test with
other compilers.

11 years agomdadm: allow build to use --size
Paul Clements [Wed, 11 Feb 2009 18:49:26 +0000 (13:49 -0500)] 
mdadm: allow build to use --size

This patch enables the --size parameter for build operations.

Without this, if you have a raid1, for instance, where the 2 disks are
not the exact same size, and you need to build the array but one of the
disks is not available right at the moment (maybe it's USB and it's
unplugged, or maybe it's a network disk and it's unavailable), then you
have to play some weird games to get the array to size correctly (that
is, to the size of the smaller of the two components or less).

There may be other uses for this too...


Signed-off-by: NeilBrown <>
11 years agoWait for POLLPRI on /proc or /sys files.
NeilBrown [Tue, 14 Apr 2009 04:59:24 +0000 (14:59 +1000)] 
Wait for POLLPRI on /proc or /sys files.

From 2.6.30, /proc/mounts and various /sys files will
probably always returns 'readable' to select, so we will need
to wait on POLLPRI to get the 'new data is available' signal.

When using select, this corresponds to an 'exception', so
adjust calls to select accordingly.
In one case we sometimes wait on a socket and sometime on
/proc/mounts, so we need to test which.

Signed-off-by: NeilBrown <>
11 years agoMerge branch 'master' of git:// into devel-3.0
NeilBrown [Tue, 14 Apr 2009 01:11:14 +0000 (11:11 +1000)] 
Merge branch 'master' of git:// into devel-3.0

Due to independent fixes for the "mdadm hangs if reshape finishes too quickly"

11 years agoudev rules fix for partitions.
NeilBrown [Tue, 14 Apr 2009 00:58:54 +0000 (10:58 +1000)] 
udev rules fix for partitions.

sysfs directories for partitions do not have md/* files, but
should not for that reason be ignored.

Thanks to Michal Soltys for original fix.

Signed-off-by: Michal Soltys <>
Signed-off-by: NeilBrown <>
11 years agomapfile: optionally store map file in /dev
NeilBrown [Tue, 14 Apr 2009 00:49:40 +0000 (10:49 +1000)] 
mapfile:  optionally store map file in /dev

During early boot, /var/run may not exist or be writable.
If that happens, sore the mapfile (which is very important for
incremental assembly) in /dev (which should exist for udev).

Thanks to Doug Ledford <> for identify this
problem and suggesting a solution.

Signed-off-by: NeilBrown <>
11 years agotests: basic ddf tests
NeilBrown [Tue, 14 Apr 2009 00:19:06 +0000 (10:19 +1000)] 
tests: basic ddf tests

Test script to create a ddf with three different member arrays,
and assemble it in a variety of ways.

Signed-off-by: NeilBrown <>
11 years agotests: add some simple data integrity tests for raid5/raid6
NeilBrown [Tue, 14 Apr 2009 00:19:05 +0000 (10:19 +1000)] 
tests:  add some simple data integrity tests for raid5/raid6

Make sure the data is preserved even when the array is degraded.

Signed-off-by: NeilBrown <>
11 years ago.gitignore update
NeilBrown [Tue, 14 Apr 2009 00:19:04 +0000 (10:19 +1000)] 
.gitignore update

Add a few more generated files to .gitignore

Signed-off-by: NeilBrown <>
11 years agoincremental_container: preserve 'in_sync' flag when adding to existing array.
NeilBrown [Tue, 14 Apr 2009 00:19:02 +0000 (10:19 +1000)] 
incremental_container: preserve 'in_sync' flag when adding to existing array.

When building container members with -IR, we need to ensure that
devices added to an active array preserve the 'in_sync' status so they
don't needlessly get rebuilt.

So allow sysfs_add_disk to do this (only works in kernels since
2.6.30) and pass the relevant flag down.

Signed-off-by: NeilBrown <>
11 years agomdmon: fix resync completion detection
Dan Williams [Sun, 12 Apr 2009 07:58:28 +0000 (00:58 -0700)] 
mdmon: fix resync completion detection

Starting with 2.6.30 the md/resync_start attribute will no longer return
a non-sensical number when resync is complete, instead it now returns

Signed-off-by: Dan Williams <>
11 years agoGrow: fix hang when reshape completes too fast
Dan Williams [Sun, 12 Apr 2009 07:58:28 +0000 (00:58 -0700)] 
Grow: fix hang when reshape completes too fast

For short reshapes the kernel may be done before mdadm can check that
progress has passed the critical section.

Signed-off-by: Dan Williams <>
11 years agoRebuildMap: handle missing disks
Dan Williams [Sun, 12 Apr 2009 07:58:28 +0000 (00:58 -0700)] 
RebuildMap: handle missing disks

When rebuilding the map file tolerate missing/offline disks, otherwise
we will segfault on the NULL return from sysfs_read.

Reported-by: Jacek Danecki <>
Signed-off-by: Dan Williams <>
11 years agoimsm: add the ddf field
Dan Williams [Sun, 12 Apr 2009 07:58:28 +0000 (00:58 -0700)] 
imsm: add the ddf field

This field is always one in arrays created by the Windows driver / OROM,
not sure why...

Signed-off-by: Dan Williams <>
11 years agoimsm: round down array size at Create
Dan Williams [Sun, 12 Apr 2009 07:58:28 +0000 (00:58 -0700)] 
imsm: round down array size at Create

Store the 1MB rounded down size of the array at create time.

Signed-off-by: Dan Williams <>
11 years agoimsm: set array size at Create/Assemble
Dan Williams [Sun, 12 Apr 2009 07:58:28 +0000 (00:58 -0700)] 
imsm: set array size at Create/Assemble

imsm arrays round down the effective array size to the closest 1
megabyte boundary so teach get_info_super_imsm and sysfs_set_array to
set 'md/array_size' if available (and make sure ddf uses the default

Signed-off-by: Dan Williams <>
11 years agoimsm: turn off curr_migr_unit updates
Dan Williams [Sun, 12 Apr 2009 07:58:28 +0000 (00:58 -0700)] 
imsm: turn off curr_migr_unit updates

New documentation shows that this field is not equivalent to
md/resync_start.  Disable updates until full support can be developed.

Writing '0' when a migration starts/re-starts remains correct.

Signed-off-by: Dan Williams <>
11 years agoimsm: defend against unsupported migrations (temporary)
Dan Williams [Sun, 12 Apr 2009 07:58:28 +0000 (00:58 -0700)] 
imsm: defend against unsupported migrations (temporary)

Until support for higher order migrations (online capacity expansion,
raid level migration, chunk size migration...) are implemented do not
allow arrays in these states to be assembled.

Signed-off-by: Dan Williams <>
11 years agoimsm: add 'verify', 'verify with fixup', and 'general' migration types
Dan Williams [Sun, 12 Apr 2009 07:58:27 +0000 (00:58 -0700)] 
imsm: add 'verify', 'verify with fixup', and 'general' migration types

imsm distinguishes parity initialization from parity checking in the
metadata. Older option roms marked the repair operation with the
'verify' type and a 'with fixup' flag in the raid device 'status' field.

Signed-off-by: Dan Williams <>
11 years agoimsm: fix imsm_map.num_domains
Dan Williams [Sun, 12 Apr 2009 07:58:27 +0000 (00:58 -0700)] 
imsm: fix imsm_map.num_domains

'num_domains' is the number of parity domains.  I.e. 2 in the raid10
case (2-mirrors), while raid0 through raid5 have 1 parity domain (even
though raid0 does not have parity).

Signed-off-by: Dan Williams <>