[dm-devel] [PATCH/rfc] dm: revise 'rescue' strategy for md->bs allocations

[NeilBrown]

On Thu, Aug 31 2017, Mikulas Patocka wrote:

>> 
>> Note that only current->bio_list[0] is offloaded.  current->bio_list[1]
>> contains bios that were scheduled *before* the current one started, so
>
> These bios need to be offloaded too, otherwise you re-introduce this 
> deadlock: https://www.redhat.com/archives/dm-devel/2014-May/msg00089.html

Thanks for the link.
In the example the bio that is stuck was created in step 4.  The call
to generic_make_request() will have placed it on current->bio_list[0].
The top-level generic_make_request call by Process A is still running,
so nothing will have moved the bio to ->bio_list[1].  That only happens
after the ->make_request_fn completes, which must be after step 7.
So the problem bio is on ->bio_list[0] and the code in my patch will
pass it to a workqueue for handling.

So I don't think the deadlock would be reintroduced.  Can you see
something that I am missing?

Thanks,
NeilBrown
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 832 bytes
Desc: not available
URL: <http://listman.redhat.com/archives/dm-devel/attachments/20170901/8b2da3f6/attachment.sig>