[dm-devel] block: be more careful about status in __bio_chain_endio

[Hannes Reinecke]

On 6/12/19 9:01 AM, Christoph Hellwig wrote:
> On Tue, Jun 11, 2019 at 10:56:42PM -0400, John Dorminy wrote:
>> I believe the second of these might, but is not guaranteed to,
>> preserve the first error observed in a child; I believe if you want to
>> definitely save the first error you need an atomic.
> 
> Is there any reason not to simply use a cmpxchg?  Yes, it is a
> relatively expensive operation, but once we are chaining bios we are out
> of the super hot path anyway.  We do something similar in xfs and iomap
> already.
> 
Agree.
Thing is, we need to check if the parent status is NULL, _and_ the
parent status might be modified asynchronously.
So even a READ_ONCE() wouldn't cut it, as it would tell us that the
parent status _was_ NULL, not that the parent status _is_ NULL by the
time we're setting it.
So cmpxchg() is it.

Cheers,

Hannes
-- 
Dr. Hannes Reinecke		   Teamlead Storage & Networking
hare at suse.de			               +49 911 74053 688
SUSE LINUX GmbH, Maxfeldstr. 5, 90409 Nürnberg
GF: Felix Imendörffer, Mary Higgins, Sri Rasiah
HRB 21284 (AG Nürnberg)