Skip to contentSkip to conversation

mptcp: some self-tests fixup

PATCH In review Paolo Abeni 3 participants 6 messages Aug 11, 2021 – Aug 12, 2021

Subsystems (3)
lore
Share & tools
lore

Paolo Abeni

author maintainer
#0 lore raw

cover mptcp: some self-tests fixup

The 2 squash-to patches try to address issues/221, at least for the root cause reported in the github tracker.

Note, I can still reproduce some self-test failures, on other link_failure test case, but with lower rate and different cause - to be investigated.

Paolo Abeni (2):
Squash-to: "selftests: mptcp: add testcase for active-back"
Squash-to: "mptcp: move drop_other_suboptions check under pm lock"

 net/mptcp/options.c                             | 4 ++++
 tools/testing/selftests/net/mptcp/mptcp_join.sh | 8 ++++++--
 2 files changed, 10 insertions(+), 2 deletions(-)
Signature · 1 line
2.26.3
author maintainer
#1 lore raw

PATCH 1/2 Squash-to: "selftests: mptcp: add testcase for active-back"

we need to use big-enough client file on link failure test.

We compute the expected link usage assuming half of the client data will be sent after the link failure. The link failure happens after sending once the client file, and than the client file is sent again. After the link failure, a whole cwin amount of data can be re-inected: to be able to predict correctly the link utilization the client file size must be much greater then max cwin.

---
 tools/testing/selftests/net/mptcp/mptcp_join.sh | 8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)
tools/testing/selftests/net/mptcp/mptcp_join.sh +6 -2
index ae406b56c684..255793c5ac4f 100755
@@ -513,9 +513,13 @@ run_tests()
 	# create the input file for the failure test when
 	# the first failure test run
 	if [ "$test_linkfail" -ne 0 -a -z "$cinfail" ]; then
-		size=$((RANDOM%8))
+		# the client file must be considerably larger
+		# of the maximum expected cwin value, or the
+		# link utilization will be not predicable
+		size=$((RANDOM%2))
 		size=$((size+1))
-		size=$((size*2048))
+		size=$((size*8192))
+		size=$((size + ( $RANDOM % 8192) ))
 
 		cinfail=$(mktemp)
 		make_file "$cinfail" "client" $size
Signature · 1 line
2.26.3
author maintainer
#2 lore raw

PATCH 2/2 Squash-to: "mptcp: move drop_other_suboptions check under pm lock"

In reply to Paolo Abeni

MPJ ACK is a pure TCP ack on an established TCP connection, so is eligble to carry MPTCP ADD_ADDR option. If that happens, the MPJ subopt will be stripped and the MPJ handshake will be broken.

Avoid adding ADD_ADDR subopt if the current packet is MPJ ACK (or MPC ACK, just to be safe: the added check it's cheap).

This fixes issues/221

---
 net/mptcp/options.c | 4 ++++
 1 file changed, 4 insertions(+)
net/mptcp/options.c +4 -0
index b2af81515f20..b3d5547fdb61 100644
@@ -678,7 +678,11 @@ static bool mptcp_established_options_add_addr(struct sock *sk, struct sk_buff *
 	bool port;
 	int len;
 
+	/* add addr will strip the existing options, be sure to avoid breaking
+	 * MPC/MPJ handshakes
+	 */
 	if (!mptcp_pm_should_add_signal(msk) ||
+	    (opts->suboptions & (OPTION_MPTCP_MPJ_ACK | OPTION_MPTCP_MPC_ACK)) ||
 	    !mptcp_pm_add_addr_signal(msk, skb, opt_size, remaining, &opts->addr,
 		    &echo, &port, &drop_other_suboptions))
 		return false;
Signature · 1 line
2.26.3
Tested-by on the v1 series
#3 lore raw
In reply to Paolo Abeni

Hi Paolo,

Quoted 2 lines · Paolo Abeni

The 2 squash-to patches try to address issues/221, at least for the root cause reported in the github tracker.

I confirm that it is much better, thank you!

Before the patches, it took me 2 attempts to have "mptcp_join.sh -l" failing with a debug kernel.

With the patches, I have 30 successful runs so far!

Cheers,
Matt

Signature · 2 lines
Tessares | Belgium | Hybrid Access Solutions
www.tessares.net
#4 lore raw
Quoted 6 lines · Matthieu Baerts

Hi Paolo,

Quoted 2 lines · Paolo Abeni

The 2 squash-to patches try to address issues/221, at least for the root cause reported in the github tracker.

Quoted 8 lines

I confirm that it is much better, thank you!

Before the patches, it took me 2 attempts to have "mptcp_join.sh -l" failing with a debug kernel.

With the patches, I have 30 successful runs so far!

Thanks for the patches Paolo, and thanks for testing Matthieu. Looks good to squash.

--
Mat Martineau
Intel

#5 lore raw
In reply to Paolo Abeni

Hi Paolo, Mat,

Quoted 2 lines · Paolo Abeni

The 2 squash-to patches try to address issues/221, at least for the root cause reported in the github tracker.

Thank you for the patches and the review!

Now in our tree:

- 3b6b8c96b67c: "squashed" patch 1/2 in "selftests: mptcp: add testcase for active-back"
- Results: e41bc27ecfa9..c636d849945e

- 32c3e4d50806: "squashed" patch 2/2 in "mptcp: move drop_other_suboptions check under pm lock"
- Results: c636d849945e..3e7443d82746

Builds and tests are now in progress:

https://cirrus-ci.com/github/multipath-tcp/mptcp_net-next/export/20210812T065059
https://github.com/multipath-tcp/mptcp_net-next/actions/workflows/build-validation.yml?query=branch:export/20210812T065059

Quoted 3 lines

Note, I can still reproduce some self-test failures, on other link_failure test case, but with lower rate and different cause - to be investigated.

Do you prefer to keep the issue #221 open or should we open a new one?

Cheers,
Matt

Signature · 2 lines
Tessares | Belgium | Hybrid Access Solutions
www.tessares.net