Re: 2.6.20-rc6: known unfixed regressions (v2) (part 2)

[Mike Christie <michaelc@cs.wisc.edu>]

Mike Christie wrote:
> Linus Torvalds wrote:
>> On Mon, 29 Jan 2007, Mike Christie wrote:
>>> rq->bio is NULL here, so no data is coped back to userspace and it seems
>>> nero just stops trying to talk to the drive after this.
>> Well, except that's what we used to do in 2.6.19 too. So what changed?
> 

Actually, I do not think we did this in 2.6.19. Tomo added a bug when he
ported a patch and mixed up some things so we did something weird for
2.6.20-rc1.

> Oops, you are right. I thought you reverted the place where rq->bio was
> getting set to bio. Ignore my comment.
> 

I think I am right now :)

In 2.6.19, we did:

bio = rq->bio;
blk_execute_rq() <- the execution sets rq->bio to null so that is why we
save a bio pointer.
blk_rq_unmap_user(bio);


For a while in 2.6.20-rc1, we basically did

blk_execute_rq()
blk_rq_unmap_user(rq->bio) <- this was a bug that caused the mem leak
and caused data to not be copied because rq->bio was null.

Tomo and Jens then fixes this in 2.6.20-rc2 or rc3 to what we have in rc6:

bio = rq->bio;
blk_execute_rq() <- the execution sets rq->bio to null so that is why we
save a bio pointer.
blk_rq_unmap_user(bio);