Heinz Mauelshagen | 1 Sep 11:45 2008
Picon

Re: Dmraid development status - 1.0.0rc15?

Am Sonntag, den 31.08.2008, 20:46 +0300 schrieb Antti Mäkelä:
> Hi,

Hello.

ASAP ;-)
We're settling the Intel contributions to allow for set create, remove,
rebuild and device failure event handling aiming fro F10 and RHEL5.3.

Heinz

> 
>    Anyone know when is dmraid 1.0.0rc15 going to be released? I'm 
> especially interested about the Intel RAID5 rebuild support and also the 
> (apparent) RAID10 support - at least it's hinted to in the sources. (I'm 
> running a Intel Matrix setup where I have most of my system - both Linux 
> and Windows - in a RAID5 and swap/temp on RAID0.)
> 
>    http://sourceware.org/cgi-bin/cvsweb.cgi/dmraid/?cvsroot=dm shows that
> last changes were 2 months ago...when will it be released?
> 
>    Also, kernel patches for dmraid45 module (at 
> http://people.redhat.com/~heinzm/sw/dm/dm-raid45/) stop at 2.6.25rc2 - at 
> the moment I'm running kernel 2.6.24 and it works, but if I ever need to 
> upgrade kernel - what am I supposed to do? Can dmraid utilize the kernel's 
> own raid456 module and doesn't need patches anymore? Or what is the 
> situation?
> 
>    Thanks.
> 
(Continue reading)

Antti Mäkelä | 1 Sep 20:35 2008
Picon
Picon

Re: Dmraid development status - 1.0.0rc15?

On Mon, 1 Sep 2008, Heinz Mauelshagen wrote:

> Am Sonntag, den 31.08.2008, 20:46 +0300 schrieb Antti Mäkelä:
>> Hi,
>
> Hello.
>
> ASAP ;-)
> We're settling the Intel contributions to allow for set create, remove,
> rebuild and device failure event handling aiming fro F10 and RHEL5.3.

   What about the kernel-specific stuff - does dmraid still require it's 
own module or can it use the raid456 from main tree? (I'm running with 
Gentoo's kernel with your patch added in).

   My understanding has so far been than all that dmraid really does is 
some device-mapper magic that the kernel can use it's normal software raid 
functionalities on fakeraid arrays - eg. it just bootstraps the standard 
kernel drivers - raid45 has been a special case since even Linux's own 
software raid added support only recently. Or what's the deal?

   (At least the fact that I've run into some patches for dmraid that make 
it use raid456 instead suggest this...)

   Anyway, thanks for answers.

--

-- 
- Antti
Heinz Mauelshagen | 2 Sep 21:15 2008
Picon

Re: Dmraid development status - 1.0.0rc15?

Am Montag, den 01.09.2008, 21:35 +0300 schrieb Antti Mäkelä:
> On Mon, 1 Sep 2008, Heinz Mauelshagen wrote:
> 
> > Am Sonntag, den 31.08.2008, 20:46 +0300 schrieb Antti Mäkelä:
> >> Hi,
> >
> > Hello.
> >
> > ASAP ;-)
> > We're settling the Intel contributions to allow for set create, remove,
> > rebuild and device failure event handling aiming fro F10 and RHEL5.3.
> 
>    What about the kernel-specific stuff - does dmraid still require it's 
> own module or can it use the raid456 from main tree? (I'm running with 
> Gentoo's kernel with your patch added in).

It needs the dm-raid45 target, which we're aiming to upstream shortly.

> 
>    My understanding has so far been than all that dmraid really does is 
> some device-mapper magic that the kernel can use it's normal software raid 
> functionalities on fakeraid arrays - eg. it just bootstraps the standard 
> kernel drivers - raid45 has been a special case since even Linux's own 
> software raid added support only recently. Or what's the deal?

It uses device-mapper to activate the respective mappings in the kernel
for discovered ATARAID+DDF1 mappings.

> 
>    (At least the fact that I've run into some patches for dmraid that make 
(Continue reading)

Antti Mäkelä | 2 Sep 21:55 2008
Picon
Picon

Re: Dmraid development status - 1.0.0rc15?

On Tue, 2 Sep 2008, Heinz Mauelshagen wrote:
>>    What about the kernel-specific stuff - does dmraid still require it's
>> own module or can it use the raid456 from main tree? (I'm running with
>> Gentoo's kernel with your patch added in).
>
> It needs the dm-raid45 target, which we're aiming to upstream shortly.

   Ok, thanks for all the answers!

--

-- 
- Antti
Antti Mäkelä | 5 Sep 17:45 2008
Picon
Picon

Re: Dmraid development status - 1.0.0rc15?

On Fri, 5 Sep 2008, Thomas Backlund wrote:
> # dm-raid45-2.6.25-rc2_20080221.patch.bz2 rediffed for 2.6.26:
> http://tmb.kkc.fi/Kernels/Cooker/patches/dm-raid45-2.6.26-20080801.patch
>
> # make the above patch build under 2.6.26:
> http://tmb.kkc.fi/Kernels/Cooker/patches/dm-raid45-2.6.26-20080801-buildfixes.patch
>
> # make it build with 2.6.27-rc5 (after 2.6.26 buildfix is applied):
> http://tmb.kkc.fi/Kernels/Cooker/patches/dm-raid45-2.6.27-20080829-buildfix.patch

   Thanks for these!

--

-- 
- Antti
Phillip Susi | 5 Sep 21:18 2008
Picon

Re: Why are spaces removed from the disk serial number?

It has been two months, Heinz, have you caught up from your vacation 
yet?  Is there a reason for the whitespace to be stripped from the disk 
serial number, or can we remove that?

Cetnerowski, Adam wrote:
> Cetnerowski, Adam wrote:
>> The serial number for sdb does look fishy and it could be interacting
>> strangely with the library isspace() function used to strip out the
>> whitespace (if some of the strange characters are control characters).
>> This may be the reason why the raid set is recognized under Windows.
> 
> The question is, should the whitespace even be stripped?  If the disk 
> has a space in its serial number, then shouldn't it be recorded in the 
> metadata with the space?  If the space is supposed to be stripped then I
> 
> can see the isspace() behaving incorrectly on control characters, but it
> 
> seems to me that if the disk reports a space, the metadata should have a
> 
> space in it as well.
> 
> We will need a reply from Heinz on this.
Phillip Susi | 5 Sep 21:36 2008
Picon

Re: isw brokenness

I thought that I would post an update to this thread as I now have a 
better understanding of what is going on.  It seems that the old Linux 
IDE driver was broken and failed to respect the HPA.  This was fixed in 
the new libata driver, but because a system that had been installed 
under an old kernel would no longer work properly with the new driver 
since it could not access the entire disk, Ubuntu decided to disable the 
HPA by default.  Since the bios and windows respect the HPA, and Ubuntu 
Linux does not, that lead to the dmraid problem.  The workaround is to 
set the libata parameter directing the kernel to respect the HPA.  The 
vanilla kernel respects it by default, so this should not be an issue 
for non Ubuntu users.

Phillip Susi wrote:
> It appears to be related to the use of a Host Protected Area ( HPA ). 
> Both disks are the same size, but one has some space reserved as an HPA. 
>  The metadata in that disk reports its size as reduced by the size of 
> the HPA.  The metadata in the second disk appears to report the size of 
> the first disk in its entirety, not respecting the HPA.  Because the 
> metadata on each disk disagrees, their checksums were different when 
> created which lead to them showing different values for family_num, 
> which causes dmraid to identify each disk as belonging to different 
> unrelated raid sets.
> 
> So it appears that the bios has a bug that caused it to generate 
> incorrect metadata when the raid set was created, yet the windows driver 
> appears to ignore this inconsistency and operate anyhow.
Antti Mäkelä | 5 Sep 23:07 2008
Picon
Picon

Re: Dmraid development status - 1.0.0rc15?

On Fri, 5 Sep 2008, Thomas Backlund wrote:

> # dm-raid45-2.6.25-rc2_20080221.patch.bz2 rediffed for 2.6.26:
> http://tmb.kkc.fi/Kernels/Cooker/patches/dm-raid45-2.6.26-20080801.patch
>
> # make the above patch build under 2.6.26:
> http://tmb.kkc.fi/Kernels/Cooker/patches/dm-raid45-2.6.26-20080801-buildfixes.patch
>
> # make it build with 2.6.27-rc5 (after 2.6.26 buildfix is applied):
> http://tmb.kkc.fi/Kernels/Cooker/patches/dm-raid45-2.6.27-20080829-buildfix.patch

   As with the 2.6.24 patch, I had to do a find/replace to replace all 
instances of "dmraid4-5" to "dmraid45" since otherwise it complains

# dmraid -ay
ERROR: device-mapper target type "raid45" not in kernel

   Otherwise, seems to work.

--

-- 
- Antti
Antti Mäkelä | 5 Sep 23:52 2008
Picon
Picon

Re: Dmraid development status - 1.0.0rc15?

On Sat, 6 Sep 2008, Thomas Backlund wrote:
> Antti Mäkelä skrev:
>>    As with the 2.6.24 patch, I had to do a find/replace to replace all
>> instances of "dmraid4-5" to "dmraid45" since otherwise it complains
>> # dmraid -ay
>> ERROR: device-mapper target type "raid45" not in kernel
>>
>>    Otherwise, seems to work.
>> 
>
> That's weird...
> I have never had to do that...
>
> Are you using it built-in or as module ?

   For my self-compiled kernel, built-in. This actually seemed *very* 
common problem, see e.g. 
https://bugs.launchpad.net/ubuntu/+source/linux/+bug/220493/
   (the comments actually helped me a lot when I tried getting Gentoo 
installed and Intel RAID5 going).

   Comment 5

https://bugs.launchpad.net/ubuntu/+source/linux/+bug/220493/comments/8

happens to mention that there's been a rename of the module - 
device-mapper/dmraid/something is expecting dm-raid45 but module is 
dm-raid4-5.

   Anyway, in the case of Ubuntu liveCD, I used the instructions in comment 
(Continue reading)

Heinz Mauelshagen | 10 Sep 14:36 2008
Picon

Re: Why are spaces removed from the disk serial number?

Am Freitag, den 05.09.2008, 15:18 -0400 schrieb Phillip Susi:
> It has been two months, Heinz, have you caught up from your vacation 
> yet?  Is there a reason for the whitespace to be stripped from the disk 
> serial number, or can we remove that?

I haven't been able to look to run regressions with all metadata samples
yet because of other duties. If Intel doesn't see an issue with keeping
the white space, fine with me.

Heinz

> 
> Cetnerowski, Adam wrote:
> > Cetnerowski, Adam wrote:
> >> The serial number for sdb does look fishy and it could be interacting
> >> strangely with the library isspace() function used to strip out the
> >> whitespace (if some of the strange characters are control characters).
> >> This may be the reason why the raid set is recognized under Windows.
> > 
> > The question is, should the whitespace even be stripped?  If the disk 
> > has a space in its serial number, then shouldn't it be recorded in the 
> > metadata with the space?  If the space is supposed to be stripped then I
> > 
> > can see the isspace() behaving incorrectly on control characters, but it
> > 
> > seems to me that if the disk reports a space, the metadata should have a
> > 
> > space in it as well.
> > 
> > We will need a reply from Heinz on this.
(Continue reading)


Gmane