[dm-devel] rqdm: bad usage of dm_get/dm_put (Was: Re: dm mpath: fix stall when requeueing io)

[Kiyoshi Ueda]

Hi Mikulas,

On 02/25/2010 07:33 AM +0900, Mikulas Patocka wrote:
>> Indeed, we shouldn't use the current dm_put() in any interrupt-context.
>> But the "mapped_device" can disappear in request-based dm while there
>> is a request after all bios complete, so I used dm_get()/dm_put() there.
>> I'll consider another way to prevent the problem without dm_get()/dm_put().
>> E.g. wait for request completion in dm_put() instead.
> 
> How can a request-in-progress exists when all the bios complete and the 
> device is closed?

In the current request-based dm, the device opener can remove
the mapped_device while the last request is still completing,
because bios in the last request complete first and then the device
opener can remove the mapped_device before the last request completes:
 CPU0                                           CPU1
 ======================================================================
 <<INTERRUPT>>
 blk_end_request_all(clone_rq)
   blk_update_request(clone_rq)
     bio_endio(clone_bio) == end_clone_bio
       blk_update_request(orig_rq)
         bio_endio(orig_bio)
                                                <<I/O completed>>
                                                dm_blk_close()
                                                dev_remove()
                                                  dm_put(md)
                                                    <<Free md>>
   blk_finish_request(clone_rq)
     ....
     dm_end_request(clone_rq)
       free_rq_clone(clone_rq)
       blk_end_request_all(orig_rq)
       rq_completed(md)

So we need a mechanism to defer the md deletion until the last request
completes.

Thanks,
Kiyoshi Ueda