mechanism for multipath to pass information to hardware handler

mechanism for multipath to pass information to hardware handler

[Eddie Williams]

I notice in scsi_dh_emc.c that there is a comment:

TODO: need some interface so we can set trespass values

I don't see where any such interface has been developed or even any
discussion on the different mailing lists to create such an interface.
Did I miss something?

As it stands now there is no way that I can see to have the emc hardware
handler support the "honor trespass" feature short of recompiling the
module forcing the flag to be set.  This leaves a feature that worked in
earlier versions of multipath not working now.

Eddie

Re: [dm-devel] mechanism for multipath to pass information to hardware handler

[James Bottomley]

On Wed, 2009-06-24 at 13:14 -0400, Eddie Williams wrote:
> I notice in scsi_dh_emc.c that there is a comment:
> 
> TODO: need some interface so we can set trespass values
> 
> I don't see where any such interface has been developed or even any
> discussion on the different mailing lists to create such an interface.
> Did I miss something?
> 
> As it stands now there is no way that I can see to have the emc hardware
> handler support the "honor trespass" feature short of recompiling the
> module forcing the flag to be set.  This leaves a feature that worked in
> earlier versions of multipath not working now.

OK so as I understand it that means that the line

hardware handler "1 emc 0 1"

now fails to function correctly because dm-emc no longer exists.  That
makes this a functionality regression from 2.6.26 because of this
commit:


commit cb520223d7f22c5386aff27a5856a66e2c32aaac
Author: Chandra Seetharaman <sekharan@us.ibm.com>
Date:   Thu May 1 14:50:34 2008 -0700

    [SCSI] scsi_dh: Remove hardware handlers from dm
    
    This patch removes the 3 hardware handlers that currently exist
    under dm as the functionality is moved to SCSI layer in the earlier
    patches.

Because we haven't actually moved all the functionality.

James

Re: [dm-devel] mechanism for multipath to pass information to hardware handler

[Chandra Seetharaman]

Yes, Mike Christie and I were aware of this and it was one of the issue
we were trying to resolve before we pushed scsi_dh interface upstream.
(It is little complicated as we need the parameters to be set per
vendor-product tuple).

The original code I ported to scsi_dh interface was from Ed Goggin(who
was working for EMC then). IIRC, he was also aware of this issue.

When we pushed scsi_dh interface, we did get few of the EMC folks (on
Cc) to review/test the code and they did, and this issue was not seen as
a problem. 

We wanted to get back to that issue sometime later, got busy with other
things, and it disappeared from my list of things-to-do as the
regression was not seen as an issue (till now :)... I will get back to
it.

chandra

On Fri, 2009-06-26 at 14:10 -0500, James Bottomley wrote:
> On Wed, 2009-06-24 at 13:14 -0400, Eddie Williams wrote:
> > I notice in scsi_dh_emc.c that there is a comment:
> > 
> > TODO: need some interface so we can set trespass values
> > 
> > I don't see where any such interface has been developed or even any
> > discussion on the different mailing lists to create such an interface.
> > Did I miss something?
> > 
> > As it stands now there is no way that I can see to have the emc hardware
> > handler support the "honor trespass" feature short of recompiling the
> > module forcing the flag to be set.  This leaves a feature that worked in
> > earlier versions of multipath not working now.
> 
> OK so as I understand it that means that the line
> 
> hardware handler "1 emc 0 1"
> 
> now fails to function correctly because dm-emc no longer exists.  That
> makes this a functionality regression from 2.6.26 because of this
> commit:
> 
> 
> commit cb520223d7f22c5386aff27a5856a66e2c32aaac
> Author: Chandra Seetharaman <sekharan@us.ibm.com>
> Date:   Thu May 1 14:50:34 2008 -0700
> 
>     [SCSI] scsi_dh: Remove hardware handlers from dm
>     
>     This patch removes the 3 hardware handlers that currently exist
>     under dm as the functionality is moved to SCSI layer in the earlier
>     patches.
> 
> Because we haven't actually moved all the functionality.
> 
> James
> 
> 
> 

Re: [dm-devel] mechanism for multipath to pass information to hardware handler

[James Bottomley]

On Fri, 2009-06-26 at 12:45 -0700, Chandra Seetharaman wrote:
> Yes, Mike Christie and I were aware of this and it was one of the issue
> we were trying to resolve before we pushed scsi_dh interface upstream.
> (It is little complicated as we need the parameters to be set per
> vendor-product tuple).
> 
> The original code I ported to scsi_dh interface was from Ed Goggin(who
> was working for EMC then). IIRC, he was also aware of this issue.
> 
> When we pushed scsi_dh interface, we did get few of the EMC folks (on
> Cc) to review/test the code and they did, and this issue was not seen as
> a problem. 
> 
> We wanted to get back to that issue sometime later, got busy with other
> things, and it disappeared from my list of things-to-do as the
> regression was not seen as an issue (till now :)... I will get back to
> it.

One way around this might simply be to make the device_handlers create a
sysfs interface for additional parameters.  Then the multipath command
can feed them (or in a pinch, users relying on the features can pass
them in manually).  Right at the moment having no possible work around
does appear to be an issue.

James

Re: [dm-devel] mechanism for multipath to pass information to hardware handler

[Eddie Williams]

The problem is easy to duplicate using sg_persist to register/reserve
paths to a Clariion array and then fail the active path.  The "enable"
path will fail to become active when the HONOR RESERVATION flag is not
set.  Given of course that you have a Clariion array handy...

One question I have asked EMC is in what situations would you NOT want
to set this bit?  

I can test out any changes you make in my test rig and I can probably
get the folks in the EMC lab I work with to test it as well.

Eddie

On Fri, 2009-06-26 at 12:45 -0700, Chandra Seetharaman wrote:
> Yes, Mike Christie and I were aware of this and it was one of the issue
> we were trying to resolve before we pushed scsi_dh interface upstream.
> (It is little complicated as we need the parameters to be set per
> vendor-product tuple).
> 
> The original code I ported to scsi_dh interface was from Ed Goggin(who
> was working for EMC then). IIRC, he was also aware of this issue.
> 
> When we pushed scsi_dh interface, we did get few of the EMC folks (on
> Cc) to review/test the code and they did, and this issue was not seen as
> a problem. 
> 
> We wanted to get back to that issue sometime later, got busy with other
> things, and it disappeared from my list of things-to-do as the
> regression was not seen as an issue (till now :)... I will get back to
> it.
> 
> chandra
> 
> On Fri, 2009-06-26 at 14:10 -0500, James Bottomley wrote:
> > On Wed, 2009-06-24 at 13:14 -0400, Eddie Williams wrote:
> > > I notice in scsi_dh_emc.c that there is a comment:
> > > 
> > > TODO: need some interface so we can set trespass values
> > > 
> > > I don't see where any such interface has been developed or even any
> > > discussion on the different mailing lists to create such an interface.
> > > Did I miss something?
> > > 
> > > As it stands now there is no way that I can see to have the emc hardware
> > > handler support the "honor trespass" feature short of recompiling the
> > > module forcing the flag to be set.  This leaves a feature that worked in
> > > earlier versions of multipath not working now.
> > 
> > OK so as I understand it that means that the line
> > 
> > hardware handler "1 emc 0 1"
> > 
> > now fails to function correctly because dm-emc no longer exists.  That
> > makes this a functionality regression from 2.6.26 because of this
> > commit:
> > 
> > 
> > commit cb520223d7f22c5386aff27a5856a66e2c32aaac
> > Author: Chandra Seetharaman <sekharan@us.ibm.com>
> > Date:   Thu May 1 14:50:34 2008 -0700
> > 
> >     [SCSI] scsi_dh: Remove hardware handlers from dm
> >     
> >     This patch removes the 3 hardware handlers that currently exist
> >     under dm as the functionality is moved to SCSI layer in the earlier
> >     patches.
> > 
> > Because we haven't actually moved all the functionality.
> > 
> > James
> > 
> > 
> > 
> 
> --
> dm-devel mailing list
> dm-devel@redhat.com
> https://www.redhat.com/mailman/listinfo/dm-devel

Re: [dm-devel] mechanism for multipath to pass information to hardware handler

[Chandra Seetharaman]

On Fri, 2009-06-26 at 14:55 -0500, James Bottomley wrote:
> On Fri, 2009-06-26 at 12:45 -0700, Chandra Seetharaman wrote:
> > Yes, Mike Christie and I were aware of this and it was one of the issue
> > we were trying to resolve before we pushed scsi_dh interface upstream.
> > (It is little complicated as we need the parameters to be set per
> > vendor-product tuple).
> > 
> > The original code I ported to scsi_dh interface was from Ed Goggin(who
> > was working for EMC then). IIRC, he was also aware of this issue.
> > 
> > When we pushed scsi_dh interface, we did get few of the EMC folks (on
> > Cc) to review/test the code and they did, and this issue was not seen as
> > a problem. 
> > 
> > We wanted to get back to that issue sometime later, got busy with other
> > things, and it disappeared from my list of things-to-do as the
> > regression was not seen as an issue (till now :)... I will get back to
> > it.
> 
> One way around this might simply be to make the device_handlers create a
> sysfs interface for additional parameters.  Then the multipath command

Yes, having a per device interface is one of the options we considered.
Once we have that defined we can either use sysfs or a scsi_dh_8() API
to achieve the same.

But we wanted to have it per vendor-product instead of per device, I
don't recall the problems associated with it. Mike(C), do you recall ?

> can feed them (or in a pinch, users relying on the features can pass
> them in manually).  Right at the moment having no possible work around
> does appear to be an issue.

I will try to roll up a patch soon.
> 
> James
> 
> 
> --
> dm-devel mailing list
> dm-devel@redhat.com
> https://www.redhat.com/mailman/listinfo/dm-devel

Re: mechanism for multipath to pass information to hardware handler

[Mike Christie]

On 06/26/2009 07:03 PM, Chandra Seetharaman wrote:
> On Fri, 2009-06-26 at 14:55 -0500, James Bottomley wrote:
>> On Fri, 2009-06-26 at 12:45 -0700, Chandra Seetharaman wrote:
>>> Yes, Mike Christie and I were aware of this and it was one of the issue
>>> we were trying to resolve before we pushed scsi_dh interface upstream.
>>> (It is little complicated as we need the parameters to be set per
>>> vendor-product tuple).
>>>
>>> The original code I ported to scsi_dh interface was from Ed Goggin(who
>>> was working for EMC then). IIRC, he was also aware of this issue.
>>>
>>> When we pushed scsi_dh interface, we did get few of the EMC folks (on
>>> Cc) to review/test the code and they did, and this issue was not seen as
>>> a problem.
>>>
>>> We wanted to get back to that issue sometime later, got busy with other
>>> things, and it disappeared from my list of things-to-do as the
>>> regression was not seen as an issue (till now :)... I will get back to
>>> it.
>> One way around this might simply be to make the device_handlers create a
>> sysfs interface for additional parameters.  Then the multipath command
>
> Yes, having a per device interface is one of the options we considered.
> Once we have that defined we can either use sysfs or a scsi_dh_8() API
> to achieve the same.
>
> But we wanted to have it per vendor-product instead of per device, I
> don't recall the problems associated with it. Mike(C), do you recall ?
>

I do not think you want the interface to be per vendor-product in the 
kernel. You might want some devices with some feature on and some with 
it off. In userspace we can decide if it is should be per vendor-product 
or per device, and then have multipath set things up. The kernel 
interface should probably be more generic.


>> can feed them (or in a pinch, users relying on the features can pass
>> them in manually).  Right at the moment having no possible work around
>> does appear to be an issue.
>
> I will try to roll up a patch soon.
>> James
>>
>>
>> --
>> dm-devel mailing list
>> dm-devel@redhat.com
>> https://www.redhat.com/mailman/listinfo/dm-devel
>
> --
> To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html