dm-multipath MP_RETRY_PG_INIT patch

dm-multipath MP_RETRY_PG_INIT patch

[Jitendra Bhivare]

[-- Attachment #1.1: Type: text/plain, Size: 638 bytes --]

Hi,

In dm-multipath framework, using the hw-handler's error-function entry
point, nothing can be known about the reason for IO failure. From this
function MP_FAIL_PATH/MP_BYPASS_PG/MP_ERROR_IO can returned.
If the same path needs to be tried (i.e. in pg_init) to figure out the
reason for failure there is no way to tell that to dm-multipath.

One way to do so, is to return MP_RETRY_PG_INIT from hw_handler's
error function. This would set current_pgpath and current_pg to NULL
which would trigger pg_init. The path won't fail until pg_init says so.

Should this option MP_RETRY_PG_INIT be added to dm-multipath?

Thanks & Regards,

JB

[-- Attachment #1.2: Type: text/html, Size: 816 bytes --]

[-- Attachment #2: Type: text/plain, Size: 0 bytes --]

Re: dm-multipath MP_RETRY_PG_INIT patch

[Mike Christie]

Jitendra Bhivare wrote:
> Hi,
> 
> In dm-multipath framework, using the hw-handler's error-function entry
> point, nothing can be known about the reason for IO failure. From this
> function MP_FAIL_PATH/MP_BYPASS_PG/MP_ERROR_IO can returned.
> If the same path needs to be tried (i.e. in pg_init) to figure out the
> reason for failure there is no way to tell that to dm-multipath.
> 
> One way to do so, is to return MP_RETRY_PG_INIT from hw_handler's
> error function. This would set current_pgpath and current_pg to NULL
> which would trigger pg_init. The path won't fail until pg_init says so.
> 
> Should this option MP_RETRY_PG_INIT be added to dm-multipath?
> 

I think we need something like it. What hw handler is this for? Dave W, 
sent a hw handler for old boxes that need START_STOP and he sent a patch 
which worked like what you describe? Have you seen his patch to this list?