Re: [PATCH v2 net-next] RDS: fix rds-ping deadlock over TCP transport

[David Miller]

From: Santosh Shilimkar 
Date: Fri, 16 Oct 2015 22:13:21 -0400

> Sowmini found hang with rds-ping while testing RDS over TCP. Its
> a corner case and doesn't happen always. The issue is not reproducible
> with IB transport. Its clear from below dump why we see it with RDS TCP.
...
> This happens because rds_send_xmit() chain wants to take
> sock_lock which is already taken by tcp_v4_rcv() on its
> way to rds_tcp_data_ready(). Commit db6526dcb51b ("RDS: use
> rds_send_xmit() state instead of RDS_LL_SEND_FULL") which
> was trying to opportunistically finish the send request
> in same thread context.
> 
> But because of above recursive lock hang with RDS TCP,
> the send work from rds_send_pong() needs to deferred to
> worker to avoid lock up. Given RDS ping is more of connectivity
> test than performance critical path, its should be ok even
> for transport like IB.
> 
> Reported-by: Sowmini Varadhan 
> Acked-by:  Sowmini Varadhan 
> Signed-off-by: Santosh Shilimkar 
> Signed-off-by: Santosh Shilimkar 
> ---
> v2: Dropped the confusing SEND_LL_FULL check from v1

Applied, thank you.
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majord...@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH v2 net-next] RDS: fix rds-ping deadlock over TCP transport

[Sowmini Varadhan]

On 10/16/2015 10:13 PM, Santosh Shilimkar wrote:



But because of above recursive lock hang with RDS TCP,
the send work from rds_send_pong() needs to deferred to
worker to avoid lock up. Given RDS ping is more of connectivity
test than performance critical path, its should be ok even
for transport like IB.



Acked-by: Sowmini Varadhan 
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majord...@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH v2 net-next] RDS: fix rds-ping deadlock over TCP transport

[Santosh Shilimkar]

Sowmini found hang with rds-ping while testing RDS over TCP. Its
a corner case and doesn't happen always. The issue is not reproducible
with IB transport. Its clear from below dump why we see it with RDS TCP.

 [] do_tcp_setsockopt+0xb5/0x740
 [] tcp_setsockopt+0x24/0x30
 [] sock_common_setsockopt+0x14/0x20
 [] rds_tcp_xmit_prepare+0x5d/0x70 [rds_tcp]
 [] rds_send_xmit+0xd7/0x740 [rds]
 [] rds_send_pong+0x142/0x180 [rds]
 [] rds_recv_incoming+0x274/0x330 [rds]
 [] ? ttwu_queue+0x11e/0x130
 [] ? skb_copy_bits+0x6d/0x2c0
 [] rds_tcp_data_recv+0x2f0/0x3d0 [rds_tcp]
 [] tcp_read_sock+0x96/0x1c0
 [] ? rds_tcp_recv_init+0x40/0x40 [rds_tcp]
 [] ? sock_def_write_space+0xa0/0xa0
 [] rds_tcp_data_ready+0xa1/0xf0 [rds_tcp]
 [] tcp_data_queue+0x379/0x5b0
 [] ? rds_tcp_write_space+0xbb/0x110 [rds_tcp]
 [] tcp_rcv_established+0x2e2/0x6e0
 [] tcp_v4_do_rcv+0x122/0x220
 [] tcp_v4_rcv+0x867/0x880
 [] ip_local_deliver_finish+0xa3/0x220

This happens because rds_send_xmit() chain wants to take
sock_lock which is already taken by tcp_v4_rcv() on its
way to rds_tcp_data_ready(). Commit db6526dcb51b ("RDS: use
rds_send_xmit() state instead of RDS_LL_SEND_FULL") which
was trying to opportunistically finish the send request
in same thread context.

But because of above recursive lock hang with RDS TCP,
the send work from rds_send_pong() needs to deferred to
worker to avoid lock up. Given RDS ping is more of connectivity
test than performance critical path, its should be ok even
for transport like IB.

Reported-by: Sowmini Varadhan 
Acked-by:  Sowmini Varadhan 
Signed-off-by: Santosh Shilimkar 
Signed-off-by: Santosh Shilimkar 
---
v2: Dropped the confusing SEND_LL_FULL check from v1

 net/rds/send.c | 5 ++---
 1 file changed, 2 insertions(+), 3 deletions(-)

diff --git a/net/rds/send.c b/net/rds/send.c
index ee49c25..827155c 100644
--- a/net/rds/send.c
+++ b/net/rds/send.c
@@ -1182,9 +1182,8 @@ rds_send_pong(struct rds_connection *conn, __be16 dport)
rds_stats_inc(s_send_queued);
rds_stats_inc(s_send_pong);
 
-   ret = rds_send_xmit(conn);
-   if (ret == -ENOMEM || ret == -EAGAIN)
-   queue_delayed_work(rds_wq, >c_send_w, 1);
+   /* schedule the send work on rds_wq */
+   queue_delayed_work(rds_wq, >c_send_w, 1);
 
rds_message_put(rm);
return 0;
-- 
1.9.1

--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majord...@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html