Re: [PATCH 3/7] scsi_dh: add generic SPC-3 alua handler

[Chandra Seetharaman <sekharan@us.ibm.com>]

On Fri, 2008-05-16 at 22:18 +0200, Jens Axboe wrote:
> On Fri, May 16 2008, Chandra Seetharaman wrote:
> > It is a question to Jens :) 
> > 
> > On Thu, 2008-05-15 at 10:13 +0200, Hannes Reinecke wrote:
> > > Chandra Seetharaman wrote:
> > > > I do not know the functionality of this module. So, my comments are only
> > > > related to code.
> > > > 
> > > > blk_request_rq() frees up the request before it returns (in
> > > > blk_end_sync_rq()). So, there is no need for blk_put_request().

My bad...

After Jens's comment, looked closely at blk_execute_rq() and
blk_end_sync_rq() and realized that there is an additional reference
acquired by blk_execute_rq() which is dropped later.

> > > > 
> > > Indeed. But this opens up another question:
> > > 
> > > By the time blk_execute_rq() returns, the request is already
> > > put back onto the queue.
> > > That means that I shouldn't access rq->errors any more, as
> > > the request might have been reused already.
> > > But blk_execute_rq() returns -EIO for any error, making it
> > > impossible to signal a proper error here.
> > > So how do I get the contents of rq->errors safely?
> > > 
> > > Cheers,
> > > 
> > > Hannes
> 
> blk_get_request() and blk_put_request() when done. See how
> block/scsi_ioctl.c:sg_io() does just that.
>