On Thu, Nov 19, 2015 at 10:18 PM, Tejun Heo <tj@xxxxxxxxxx> wrote:
> Hello, Ilya.
>
> On Thu, Nov 19, 2015 at 09:56:21PM +0100, Ilya Dryomov wrote:
>> > Yes, that's where *I* think we should be headed. Stuff in lower
>> > layers should stick around while upper layer things are around
>>
>> I think the fundamental problem is the embedding of bdi in the queue.
>> The lifetime rules (or, rather, expectations) for the two seem to be
>> completely different and, while used together, they belong to different
>> subsystems. Even if we find a way to fix this particular race, there
>> is a good chance someone will reintroduce it in the future, perhaps in
>> a more subtle way.
>
> You're right. This is nasty. Hmmm... the root problem is that beyond
> the last __blkdev_put() the bdev and disk don't really have anything
> to do with each other but the bdev is still pointing to it. We are
> already guaranteeing that the underlying disk hangs around while there
> are bdevs associated with it.
>
> We already know that the bdev is idle once bd_openers hits zero and
> the inode gets flushed, so at that point, the problem is bdev's
> inode->i_wb is still pointing to something that the bdev doesn't have
> anything to do with. So, can we do inode_detach_wb() after flushing
> the inode?
Detaching the inode earlier is what I suggested in the first email, but
I didn't know if this kind of special casing was OK. I'll try it out.
Thanks,
Ilya
--
To unsubscribe from this list: send the line "unsubscribe ceph-devel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
