rxrpc: Fix error distribution
authorDavid Howells <dhowells@redhat.com>
Thu, 27 Sep 2018 14:13:09 +0000 (15:13 +0100)
committerDavid Howells <dhowells@redhat.com>
Fri, 28 Sep 2018 09:33:17 +0000 (10:33 +0100)
commitf334430316e7fd37c4821ebec627e27714bb5d76
treea8e7fb558d409f5a36ba8104975272ed91b5ab9a
parent37a675e768d7606fe8a53e0c459c9b53e121ac20
rxrpc: Fix error distribution

Fix error distribution by immediately delivering the errors to all the
affected calls rather than deferring them to a worker thread.  The problem
with the latter is that retries and things can happen in the meantime when we
want to stop that sooner.

To this end:

 (1) Stop the error distributor from removing calls from the error_targets
     list so that peer->lock isn't needed to synchronise against other adds
     and removals.

 (2) Require the peer's error_targets list to be accessed with RCU, thereby
     avoiding the need to take peer->lock over distribution.

 (3) Don't attempt to affect a call's state if it is already marked complete.

Signed-off-by: David Howells <dhowells@redhat.com>
include/trace/events/rxrpc.h
net/rxrpc/ar-internal.h
net/rxrpc/call_object.c
net/rxrpc/conn_client.c
net/rxrpc/conn_object.c
net/rxrpc/peer_event.c
net/rxrpc/peer_object.c