Bug in multipath code (Linux 2.6.x) ?

Bug in multipath code (Linux 2.6.x) ?

[Dominik Kubla]

Hi Neil,

It appears to me as if there might be a bug in the multipath code since
it is impossible to activate the multipath device properly. It doesn't
matter if one uses raidtools or mdadm. Every time i try to activate the
mp devices, the system complains about not being able to access the
drive. The result is a mp device with just one path.

Hardware configuration:

Sun Fire v65x with two LSI Logic FC-HBA's connected to two disjunct
switched fabrics.
Fabric switches are four Brocade Silkwork 3600 and storage devices are
two Adv. Unibyte JBODs.

Software configuration:

Debian GNU/Linux (unstable) with kernel 2.6.4.
Two disks are setup up as individual mp devices with a RAID1 on top of
the two mp devices.

I don't have the exact error message at the moment, but can supply it in
case it is needed.

Regards,
  Dominik Kubla

Re: Bug in multipath code (Linux 2.6.x) ?

[Neil Brown]

On Wednesday April 7, kubla@uni-mainz.de wrote:
> Hi Neil,
> 
> It appears to me as if there might be a bug in the multipath code since
> it is impossible to activate the multipath device properly. It doesn't
> matter if one uses raidtools or mdadm. Every time i try to activate the
> mp devices, the system complains about not being able to access the
> drive. The result is a mp device with just one path.

It wouldn't surprise me if there was a bug in the multipath code.
I have no hardware to test it on and little interest in it.  I'm happy
to fix obvious bugs and I try not to break it with other changes in
md, but beyond that I just leave it alone.

I can do a simple test using two loopback devices on the same file as
the two paths and that seems to work OK, so there is probably some
interesting interaction with your particular hardware and it's driver.

So, if you can get good error messages and try to figure out what is
actually happening, then I'm happy to look at fixing it, but without
lots of details, there isn't much I can do.

NeilBrown


> 
> Hardware configuration:
> 
> Sun Fire v65x with two LSI Logic FC-HBA's connected to two disjunct
> switched fabrics.
> Fabric switches are four Brocade Silkwork 3600 and storage devices are
> two Adv. Unibyte JBODs.
> 
> Software configuration:
> 
> Debian GNU/Linux (unstable) with kernel 2.6.4.
> Two disks are setup up as individual mp devices with a RAID1 on top of
> the two mp devices.
> 
> I don't have the exact error message at the moment, but can supply it in
> case it is needed.
> 
> Regards,
>   Dominik Kubla

Re: Bug in multipath code (Linux 2.6.x) ?

[Dominik Kubla]

On Thursday 08 April 2004 00:37, Neil Brown wrote:

> It wouldn't surprise me if there was a bug in the multipath code.
> I have no hardware to test it on and little interest in it.  I'm happy
> to fix obvious bugs and I try not to break it with other changes in
> md, but beyond that I just leave it alone.

Understandable. Maybe the Linux SAN-Alliance should sponsor you a couple of 
HBAs and a JBOD... ;-)

> I can do a simple test using two loopback devices on the same file as
> the two paths and that seems to work OK, so there is probably some
> interesting interaction with your particular hardware and it's driver.

Yes, that could get messy: FC-HBA <-> SCSI layer <-> MD driver

> So, if you can get good error messages and try to figure out what is
> actually happening, then I'm happy to look at fixing it, but without
> lots of details, there isn't much I can do.

I'll get you the correct error messages next week and i will try to work my 
way through the multipath code to understand what it's doing. Unfortunately 
it some years since i did serious work on the kernel internals, so i will 
most likely need to get a feel of what's changed.

Regards,
  Dominik Kubla

Re: Bug in multipath code (Linux 2.6.x) ?

[Lars Marowsky-Bree]

On 2004-04-10T10:11:58,
   Dominik Kubla <dominik.kubla@uni-mainz.de> said:

> Understandable. Maybe the Linux SAN-Alliance should sponsor you a couple of 
> HBAs and a JBOD... ;-)

A lot of work has gone into the Device-Mapper multipath target in 2.6;
maybe you should look at that one.


Sincerely,
    Lars Marowsky-Brée <lmb@suse.de>

-- 
High Availability & Clustering	      \ ever tried. ever failed. no matter.
SUSE Labs			      | try again. fail again. fail better.
Research & Development, SUSE LINUX AG \ 	-- Samuel Beckett

-
To unsubscribe from this list: send the line "unsubscribe linux-raid" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

RE: Bug in multipath code (Linux 2.6.x) ?

[Dominik Kubla]

Unfortunately very little effort seems to be spent on documenting the features of device-mapper and how to use them. I originally intended to use DM, but since i could not find any information on how to configure mirrored multipath devices i decided against it.

What is the word on MD and DM? I miss some sort of road map and a general overview (preferrably with executive summary). (I am willing to write one if i can find/get the necessary information.)

Regards,
  Dominik 

-----Original Message-----
From: Lars Marowsky-Bree [mailto:lmb@suse.de] 
Sent: Tuesday, April 13, 2004 10:04 AM
To: Dominik Kubla; Neil Brown
Cc: Dominik Kubla; linux-raid@vger.kernel.org; khoddam@netz.klinik.uni-mainz.de
Subject: Re: Bug in multipath code (Linux 2.6.x) ?

On 2004-04-10T10:11:58,
   Dominik Kubla <dominik.kubla@uni-mainz.de> said:

> Understandable. Maybe the Linux SAN-Alliance should sponsor you a 
> couple of HBAs and a JBOD... ;-)

A lot of work has gone into the Device-Mapper multipath target in 2.6; maybe you should look at that one.


Sincerely,
    Lars Marowsky-Brée <lmb@suse.de>

-- 
High Availability & Clustering	      \ ever tried. ever failed. no matter.
SUSE Labs			      | try again. fail again. fail better.
Research & Development, SUSE LINUX AG \ 	-- Samuel Beckett



-
To unsubscribe from this list: send the line "unsubscribe linux-raid" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: Bug in multipath code (Linux 2.6.x) ?

[Jure Pecar]

On Tuesday 13 of April 2004 12:46, Dominik Kubla wrote:
> Unfortunately very little effort seems to be spent on documenting the
> features of device-mapper and how to use them. I originally intended to use
> DM, but since i could not find any information on how to configure mirrored
> multipath devices i decided against it.
>
> What is the word on MD and DM? I miss some sort of road map and a general
> overview (preferrably with executive summary). (I am willing to write one
> if i can find/get the necessary information.)

Take a look at EVMS then. (http://evms.sf.net)
It has excellent documentation and a roadmap as well ;)

--

Jure Pečar
-
To unsubscribe from this list: send the line "unsubscribe linux-raid" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Bug in multipath code (Linux 2.6.x) ?

[christophe varoqui]

> Unfortunately very little effort seems to be spent on documenting the features of device-mapper and how to use them. I originally intended to use DM, but since i could not find any information on how to configure mirrored multipath devices i decided against it.
> 
> What is the word on MD and DM? I miss some sort of road map and a general overview (preferrably with executive summary). (I am willing to write one if i can find/get the necessary information.)
> 
I'm quite fond of my docs on multipath with the device-mapper subsystem.
You can find them at http://christophe.varoqui.free.fr

Hope it helps.

regards,
cvaroqui

Re: Bug in multipath code (Linux 2.6.x) ?

[Dominik Kubla]

christophe varoqui wrote:

>>Unfortunately very little effort seems to be spent on documenting the features of device-mapper and how to use them. I originally intended to use DM, but since i could not find any information on how to configure mirrored multipath devices i decided against it.
>>
>>What is the word on MD and DM? I miss some sort of road map and a general overview (preferrably with executive summary). (I am willing to write one if i can find/get the necessary information.)
>>
> 
> I'm quite fond of my docs on multipath with the device-mapper subsystem.
> You can find them at http://christophe.varoqui.free.fr
> 
> Hope it helps.
> 
> regards,
> cvaroqui
> 

Excellent! Thanks a lot!

Dominik