On Tue, 2014-10-28 at 18:34 +0200, Sagi Grimberg wrote: > On 10/24/2014 9:02 AM, Nicholas A. Bellinger wrote: > <SNIP> > > AFAICT, it looks like the assumption in isert_disconnected_handler() to > > dereference rdma_cm_id->context as isert_conn (in all cases) is wrong, > > and the above RDMA_CM_EVENT_DEVICE_REMOVAL has iscsi_np stored in > > ->context from the original rdma_create_id() at isert_setup_np() time. > > > > So, is there a way to tell the difference how rdma_cm_id->context should > > be dereferenced when DEVICE_REMOVAL occurs..? Does DEVICE_REMOVAL occur > > on just the listener rdma_cm_id, or on all accepted children as well..? > > > > Anything else to consider wrt to other CMA events being kicked off into > > isert_disconnected_handler()..? > > > > Hey Nic, > > Terribly sorry for the late response, I'm juggling between 5 different > projects... > > This is indeed a bug, and I indeed noticed it. > > This will happen if the network portal cm id listens on a specific > address (e.g. not any - 0.0.0.0), in this case the cm id will acquire > the relevant device (see rdma_bind_addr) - hence will sense > DEVICE_REMOVAL events. And yes, all the accepted children will of course > get the event as well. > > Notice that cma_remove_one sequence (that fires DEVICE_REMOVAL event to > all the relevant cma ids) requires the cmd is owner to finish cleanup > before the end of the callback because in the end of it, it will allow > the device to remove. So I do plan to get disconnected_handler to the > callback instead of the deferred work. > > I think this should make the bug you hit go away: > iser-target: Handle DEVICE_REMOVAL event on network portal listener > correctly > > In this case the cm_id->context is the isert_np, and the cm_id->qp > is NULL, so use that to distinct the cases. > > Since we don't expect any other events on this cm_id we can > just return -1 for explicit termination of the cm_id by the > cma layer. > > Signed-off-by: Sagi Grimberg <sagig@xxxxxxxxxxxx> Just verified that this resolves the specific OOPesen. Queued up in target-pending/master, with a CC' to v3.10.y. Thanks Sagi! --nab -- To unsubscribe from this list: send the line "unsubscribe linux-rdma" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
