summaryrefslogtreecommitdiff
path: root/net/rds
diff options
context:
space:
mode:
authorSowmini Varadhan <sowmini.varadhan@oracle.com>2015-02-10 21:33:37 +0300
committerDavid S. Miller <davem@davemloft.net>2015-02-12 01:35:44 +0300
commit80ad0d4a7a75158f2824d541e4802c88aba4f063 (patch)
tree551dd5acb605a09883085f783fb1513d1012bf74 /net/rds
parentbf250a1fa769f2eb8fc7a4e28b3b523e9cb67eef (diff)
downloadlinux-80ad0d4a7a75158f2824d541e4802c88aba4f063.tar.xz
rds: rds_cong_queue_updates needs to defer the congestion update transmission
When the RDS transport is TCP, we cannot inline the call to rds_send_xmit from rds_cong_queue_update because (a) we are already holding the sock_lock in the recv path, and will deadlock when tcp_setsockopt/tcp_sendmsg try to get the sock lock (b) cong_queue_update does an irqsave on the rds_cong_lock, and this will trigger warnings (for a good reason) from functions called out of sock_lock. This patch reverts the change introduced by 2fa57129d ("RDS: Bypass workqueue when queueing cong updates"). The patch has been verified for both RDS/TCP as well as RDS/RDMA to ensure that there are not regressions for either transport: - for verification of RDS/TCP a client-server unit-test was used, with the server blocked in gdb and thus unable to drain its rcvbuf, eventually triggering a RDS congestion update. - for RDS/RDMA, the standard IB regression tests were used Signed-off-by: Sowmini Varadhan <sowmini.varadhan@oracle.com> Signed-off-by: David S. Miller <davem@davemloft.net>
Diffstat (limited to 'net/rds')
-rw-r--r--net/rds/cong.c16
1 files changed, 15 insertions, 1 deletions
diff --git a/net/rds/cong.c b/net/rds/cong.c
index e5b65acd650b..e6144b8246fd 100644
--- a/net/rds/cong.c
+++ b/net/rds/cong.c
@@ -221,7 +221,21 @@ void rds_cong_queue_updates(struct rds_cong_map *map)
list_for_each_entry(conn, &map->m_conn_list, c_map_item) {
if (!test_and_set_bit(0, &conn->c_map_queued)) {
rds_stats_inc(s_cong_update_queued);
- rds_send_xmit(conn);
+ /* We cannot inline the call to rds_send_xmit() here
+ * for two reasons (both pertaining to a TCP transport):
+ * 1. When we get here from the receive path, we
+ * are already holding the sock_lock (held by
+ * tcp_v4_rcv()). So inlining calls to
+ * tcp_setsockopt and/or tcp_sendmsg will deadlock
+ * when it tries to get the sock_lock())
+ * 2. Interrupts are masked so that we can mark the
+ * the port congested from both send and recv paths.
+ * (See comment around declaration of rdc_cong_lock).
+ * An attempt to get the sock_lock() here will
+ * therefore trigger warnings.
+ * Defer the xmit to rds_send_worker() instead.
+ */
+ queue_delayed_work(rds_wq, &conn->c_send_w, 0);
}
}