Commit d455ab64 authored Jul 06, 2010 by Zach Brown Committed by Andy Grover Sep 08, 2010

RDS/IB: always process recv completions

The recv refill path was leaking fragments because the recv event handler had
marked a ring element as free without freeing its frag. This was happening
because it wasn't processing receives when the conn wasn't marked up or
connecting, as can be the case if it races with rmmod.

Two observations support always processing receives in the callback.

First, buildup should only post receives, thus triggering recv event handler
calls, once it has built up all the state to handle them. Teardown should
destroy the CQ and drain the ring before tearing down the state needed to
process recvs. Both appear to be true today.

Second, this test was fundamentally racy. There is nothing to stop rmmod and
connection destruction from swooping in the moment after the conn state was
sampled but before real receive procesing starts.

Signed-off-by: Zach Brown <zach.brown@oracle.com>

parent 80c51be5

net/rds/ib_recv.c

+12 −8

Original line number	Diff line number	Diff line
		@@ -979,18 +979,22 @@ static inline void rds_poll_cq(struct rds_ib_connection *ic,
		* to get a recv completion _before_ the rdmacm ESTABLISHED
		* event is processed.
		*/
		if (rds_conn_up(conn) \|\| rds_conn_connecting(conn)) {
		/* We expect errors as the qp is drained during shutdown */
		if (wc.status == IB_WC_SUCCESS) {
		rds_ib_process_recv(conn, recv, wc.byte_len, state);
		} else {
		/* We expect errors as the qp is drained during shutdown */
		if (rds_conn_up(conn) \|\| rds_conn_connecting(conn))
		rds_ib_conn_error(conn, "recv completion on "
		"%pI4 had status %u, disconnecting and "
		"reconnecting\n", &conn->c_faddr,
		wc.status);
		}
		}

		/*
		* It's very important that we only free this ring entry if we've truly
		* freed the resources allocated to the entry. The refilling path can
		* leak if we don't.
		*/
		rds_ib_ring_free(&ic->i_recv_ring, 1);
		}
		}

Admin message