Skip to contentSkip to conversation

mptcp: rx path refactor follow-ups

PATCH In review Paolo Abeni 2 participants 5 messages Aug 24, 2022

Subsystems (2)
lore
Share & tools
lore

Paolo Abeni

author maintainer
#0 lore raw

cover mptcp: rx path refactor follow-ups

A couple of fixes trying to address the self-tests failures seen after the recent refactor.

Paolo Abeni (2):
Squash-to: "mptcp: move msk input path under full msk socket lock"
mptcp: never re-inject data on a single subflow

 net/mptcp/protocol.c | 3 ++-
 net/mptcp/sched.c    | 4 ++++
 2 files changed, 6 insertions(+), 1 deletion(-)
Signature · 1 line
2.37.1
author maintainer
#1 lore raw

PATCH 1/2 Squash-to: "mptcp: move msk input path under full msk socket lock"

whoops, I forgot to really test for pending data at release_cb time.

The above causes several recurring failures in the self-tests.

Note that this could affect badly the mptcp performance (as we now really move relevant CPU time from the subflow rx path/ksoftirqd to the user-space process), even if I haven't performed perf-tests yet on top of this change.

---
This should be placed just before: "mptcp: move RCVPRUNE event later"/ the last rx path refactor

---
 net/mptcp/protocol.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
net/mptcp/protocol.c +2 -1
index 74699bd47edf..d47c19494023 100644
@@ -3008,7 +3008,8 @@ void __mptcp_check_push(struct sock *sk, struct sock *ssk)
 
 #define MPTCP_FLAGS_PROCESS_CTX_NEED (BIT(MPTCP_PUSH_PENDING) | \
 				      BIT(MPTCP_RETRANSMIT) | \
-				      BIT(MPTCP_FLUSH_JOIN_LIST))
+				      BIT(MPTCP_FLUSH_JOIN_LIST) | \
+				      BIT(MPTCP_DEQUEUE))
 
 /* processes deferred events and flush wmem */
 static void mptcp_release_cb(struct sock *sk)
Signature · 1 line
2.37.1
author maintainer
#3 lore raw

PATCH 2/2 mptcp: never re-inject data on a single subflow

In reply to Paolo Abeni

If the MPTCP-level data ack is delayed WRT to the TCP-level ack, there is a single subflow and the user-space stops pushing data, the MPTCP-level retransmit may trigger in, fooling (disallowing) infinite mapping fallback.

All the above is triggered quite reliably by the self-tests, once we move the MPTCP receive path under the msk socket lock - delaying the MPTCP-level acks.

Plain TCP is good enough to handle retransmissions as needed.

---
Note that there are a couple of sad untold stories here. The first one is as follow: with the delayed mptcp-level ack, at MPC handshake time we have:

MPC + SYN ->

	<- MPC + SYN_ACK(1)
MPC + ACK(1) ->
	<- DSS(n bytes) + ACK(1)

DSS_ACK(1) + ACK(n+1)

and no more acks from the client side, even after reading the incoming n bytes data. So a possible alternative solution would be to tweek mptcp_cleanup_rbuf() to properly handle this scenario.

Additionally, possibly, still due to the mptcp-level delyated ack, we could need to tweek mptcp_cleanup_rbuf() more - e.g. do we need to force mptcp-level ack when there is a large enough amount of newly "ackable" data due to __mptcp_move_skbs() action? I don't know.

I guess/hope that the condition on mptcp-level window update already implemented in mptcp_cleanup_rbuf() should also cover for the above, but I'm not sure.

The other point is that I can't recall the previous discussion about avoiding re-injection with a single subflow. I now think that is bad, at least with delayed mptcp-level ack, and I don't see a need for that, but possibly/likely I'm forgetting something?!?

---
 net/mptcp/sched.c | 4 ++++
 1 file changed, 4 insertions(+)
net/mptcp/sched.c +4 -0
index 044c5ec8bbfb..409e832085c2 100644
@@ -162,6 +162,10 @@ struct sock *mptcp_sched_get_retrans(struct mptcp_sock *msk)
 	if (__mptcp_check_fallback(msk))
 		return NULL;
 
+	/* never retransmit when there is a single subflow */
+	if (list_is_singular(&msk->conn_list) && list_empty(&msk->join_list))
+		return NULL;
+
 	if (!msk->sched)
 		return mptcp_subflow_get_retrans(msk);
 
Signature · 1 line
2.37.1
author maintainer
#2 lore raw

Re: [PATCH mptcp-next 1/2] Squash-to: "mptcp: move msk input path under full msk socket lock"

In reply to Paolo Abeni
Quoted 8 lines · Paolo Abeni

whoops, I forgot to really test for pending data at release_cb time.

The above causes several recurring failures in the self-tests.

Note that this could affect badly the mptcp performance (as we now really move relevant CPU time from the subflow rx path/ksoftirqd to the user-space process), even if I haven't performed perf-tests yet on top of this change.

The first raw number are quite discouraging :/
-30% on single subflow test.

/P

#4 lore raw

Re: mptcp: never re-inject data on a single subflow: Tests Results

In reply to Paolo Abeni

Hi Paolo,

Thank you for your modifications, that's great!

Our CI did some validations and here is its report:

- KVM Validation: normal:
- Unstable: 2 failed test(s): selftest_mptcp_join selftest_simult_flows 🔴:
- Task: https://cirrus-ci.com/task/4563343720579072
- Summary: https://api.cirrus-ci.com/v1/artifact/task/4563343720579072/summary/summary.txt

- KVM Validation: debug:
- Unstable: 2 failed test(s): selftest_mptcp_connect selftest_mptcp_join 🔴:
- Task: https://cirrus-ci.com/task/5689243627421696
- Summary: https://api.cirrus-ci.com/v1/artifact/task/5689243627421696/summary/summary.txt

Initiator: Patchew Applier
Commits: https://github.com/multipath-tcp/mptcp_net-next/commits/a85c4255a43d

If there are some issues, you can reproduce them using the same environment as the one used by the CI thanks to a docker image, e.g.:

    $ cd [kernel source code]
    $ docker run -v "${PWD}:${PWD}:rw" -w "${PWD}" --privileged --rm -it \
        --pull always mptcp/mptcp-upstream-virtme-docker:latest \
        auto-debug

For more details:

https://github.com/multipath-tcp/mptcp-upstream-virtme-docker

Please note that despite all the efforts that have been already done to have a stable tests suite when executed on a public CI like here, it is possible some reported issues are not due to your modifications. Still, do not hesitate to help us improve that ;-)

Cheers,
MPTCP GH Action bot
Bot operated by Matthieu Baerts (Tessares)