[PATCH 0/21] SUSE multipath-tools resync

[PATCH 0/21] SUSE multipath-tools resync

[Hannes Reinecke]

Hi Christophe,

there have been quite some fixes accumulating here; and as I finally
found some time I thought it a nice idea to have them sent upstream.
There are fixes for various bits and pieces; mostly no big deal.
Most important of these are the following:

- Added an option '-t' to dump the internal hardware table to multipath.
  This is quite helpful when generating your own multipath-tools.conf
  and you'd like to start with some sensible defaults.
- We now got a manpage multipath.conf.5 !
- New priority checker 'hp_sw' which will allow us to group these
  machines by priority, too.

Some patches might be a bit controversial; they make sense to me, but
others might think otherwise:

- I've removed libsysfs and implemented the sysfs handling internally.
  With the current changes in sysfs the older libsysfs doesn't work
  anymore and in either case it was just a big annoyance.
- I've hacked up kpartx to handle extended partitions as well.
  Currently it's quite peculiar with kpartx; we'd be generating
  tables for all partitions _unless_ it's an extended one.
  So we won't be getting any symlinks to the extended partition
  when we're using kpartx, although the device itself is present.

Feel free to omit them if you don't agree.
And of course, feedback etc is welcome.

Christophe, please pull from

git://git.kernel.org/pub/scm/linux/kernel/hare/multipath-tools/.git
#upstream-fixes.

Cheers,

Hannes
-- 
Dr. Hannes Reinecke		      zSeries & Storage
hare@suse.de			      +49 911 74053 688
SUSE LINUX Products GmbH, Maxfeldstr. 5, 90409 Nürnberg
GF: Markus Rex, HRB 16746 (AG Nürnberg)

Re: [PATCH 0/21] SUSE multipath-tools resync

[Christophe Varoqui]

Le lundi 21 mai 2007 à 11:22 +0200, Hannes Reinecke a écrit :
> Hi Christophe,
> 
> there have been quite some fixes accumulating here; and as I finally
> found some time I thought it a nice idea to have them sent upstream.
> There are fixes for various bits and pieces; mostly no big deal.
> Most important of these are the following:
> 
Thanks, I started reading and merging this work.

> - Added an option '-t' to dump the internal hardware table to multipath.
>   This is quite helpful when generating your own multipath-tools.conf
>   and you'd like to start with some sensible defaults.

Doesn't the daemon CLI "show config" command fill this need ?
If so, I'd rather drop this patch (and the associated manpage patch).

> - We now got a manpage multipath.conf.5 !

Nice

> - New priority checker 'hp_sw' which will allow us to group these
>   machines by priority, too.
Have you pushed the kernel hardware handler upstream too ? (the
START_STOP one ?)

> 
- directio checker async mode
Interesting stuff.
1) Do we need the libaio.h copy ?
2) Have you put some thought to SG_IO async mode feasability ?
3) Shouldn't we also timeout the execv() for path prios and ids to avoid
blocking the daemon.

> Some patches might be a bit controversial; they make sense to me, but
> others might think otherwise:
> 
> - I've removed libsysfs and implemented the sysfs handling internally.
>   With the current changes in sysfs the older libsysfs doesn't work
>   anymore and in either case it was just a big annoyance.

I mostly agree, but this one will be long to review :)

You seem to have a genuine fix in there, in cli_add_map(). It might
deserve a separate patch.

> - I've hacked up kpartx to handle extended partitions as well.
>   Currently it's quite peculiar with kpartx; we'd be generating
>   tables for all partitions _unless_ it's an extended one.
>   So we won't be getting any symlinks to the extended partition
>   when we're using kpartx, although the device itself is present.
> 
I'll read those later.

> Feel free to omit them if you don't agree.
> And of course, feedback etc is welcome.
> 
I add this one, that might interest you :

commit 9ff5a680b7803d8932059b2e0d6d295c60c95370
Author: Christophe Varoqui <cvaroqui@zezette.localdomain>
Date:   Tue May 22 00:24:22 2007 +0200

    [libmultipath] new devmapper log handler needs stdarg.h
    
    As it uses the va_list type.

> Christophe, please pull from
> 
> git://git.kernel.org/pub/scm/linux/kernel/hare/multipath-tools/.git
> #upstream-fixes.
> 
> Cheers,
> 
> Hannes

Re: Re: [PATCH 0/21] SUSE multipath-tools resync

[Stefan Bader]

Hi Christophe,

> - directio checker async mode
> Interesting stuff.
> 1) Do we need the libaio.h copy ?
>
Not really. Just need some interface definitions.

> 2) Have you put some thought to SG_IO async mode feasability ?
>
That is not really an option if you have non-SCSI devices. ;-)

> 3) Shouldn't we also timeout the execv() for path prios and ids to avoid
> blocking the daemon.
>
Could be worth a thought. But maybe these are not that critical.
At least as long as you don't require these calls to re-activate
any failed paths...

Regards,
Stefan