public inbox for stable@vger.kernel.org
 help / color / mirror / Atom feed
* [PATCH] selftests: mptcp: add test for IPv6 subflow SLAB placement
@ 2026-05-01 15:14 Vastargazing
  2026-05-01 15:45 ` Matthieu Baerts
  0 siblings, 1 reply; 5+ messages in thread
From: Vastargazing @ 2026-05-01 15:14 UTC (permalink / raw)
  To: matttbe, martineau
  Cc: mptcp, netdev, linux-kselftest, shuah, Vastargazing, stable

Add mptcp_v6_initcall.sh to verify that MPTCP IPv6 subflow child
sockets are allocated from the TCPv6 SLAB cache, not the kmalloc-4k
fallback.

tcpv6_prot_override must copy tcpv6_prot after proto_register(&tcpv6_prot)
populates tcpv6_prot.slab. If the copy runs too early, override.slab
stays NULL (frozen by __ro_after_init) and subflow children fall back
to kmalloc-4k. This lacks SLAB_TYPESAFE_BY_RCU, allowing lockless
ehash lookups in __inet_lookup_established to read freed memory.

The test exercises the IPv6 accept path via MPTCP connections between
two network namespaces, then checks that the TCPv6 slab active object
count grew. On a fixed kernel, the delta is ~2 * NR_CONNS (one subflow
per side per connection); on a broken kernel, it stays near zero because
children land in kmalloc-4k instead.

Topology: two netns connected via veth pair with /64 ULA addresses;
NR_CONNS parallel short-lived MPTCP connections are established and held
open long enough to sample /proc/slabinfo. The test skips if
CONFIG_MPTCP_IPV6 is absent (checked via kallsyms) or /proc/slabinfo is
unreadable.

Verified on Ubuntu 6.17 kernel predating the fix: TAP "not ok 1 ...
TCPv6 slab gains MPTCPv6 subflow children" with delta=0. On kernels
with the fix, delta is well above the threshold of NR_CONNS/2.

Fixes: b19bc2945b40 ("mptcp: implement delegated actions")
Cc: stable@vger.kernel.org
Signed-off-by: Vastargazing <vebohr@gmail.com>
---
 tools/testing/selftests/net/mptcp/Makefile    |   1 +
 .../selftests/net/mptcp/mptcp_v6_initcall.sh  | 140 ++++++++++++++++++
 2 files changed, 141 insertions(+)
 create mode 100644 tools/testing/selftests/net/mptcp/mptcp_v6_initcall.sh

diff --git a/tools/testing/selftests/net/mptcp/Makefile b/tools/testing/selftests/net/mptcp/Makefile
index 22ba0da2adb8..c9f329441490 100644
--- a/tools/testing/selftests/net/mptcp/Makefile
+++ b/tools/testing/selftests/net/mptcp/Makefile
@@ -14,6 +14,7 @@ TEST_PROGS := \
 	mptcp_connect_splice.sh \
 	mptcp_join.sh \
 	mptcp_sockopt.sh \
+	mptcp_v6_initcall.sh \
 	pm_netlink.sh \
 	simult_flows.sh \
 	userspace_pm.sh \
diff --git a/tools/testing/selftests/net/mptcp/mptcp_v6_initcall.sh b/tools/testing/selftests/net/mptcp/mptcp_v6_initcall.sh
new file mode 100644
index 000000000000..c55fc74b5ffb
--- /dev/null
+++ b/tools/testing/selftests/net/mptcp/mptcp_v6_initcall.sh
@@ -0,0 +1,140 @@
+#!/bin/bash
+# SPDX-License-Identifier: GPL-2.0
+#
+# Verify that MPTCP IPv6 subflow child sockets are allocated from the
+# TCPv6 slab cache.
+#
+# tcpv6_prot_override is initialised by copying tcpv6_prot, which is only
+# safe after proto_register(&tcpv6_prot) has populated tcpv6_prot.slab.
+# If the override copy runs too early during init, override.slab stays
+# NULL and __ro_after_init freezes that. Subflow child sockets then fall
+# back to kmalloc (kmalloc-4k for a TCPv6 sock), which lacks
+# SLAB_TYPESAFE_BY_RCU; lockless ehash lookups in __inet_lookup_established
+# can read freed memory.
+#
+# This test exercises the IPv6 accept path, which goes through the
+# override proto, and then asserts that the live TCPv6 slab population
+# grew. On a kernel where the override slab is NULL the delta is ~0,
+# because the children land in kmalloc-4k instead.
+
+#shellcheck disable=SC2086
+
+. "$(dirname "${0}")/mptcp_lib.sh"
+
+ns1=""
+ns2=""
+ret=0
+
+NR_CONNS=20
+TIMEOUT_POLL=30
+TIMEOUT_TEST=$((TIMEOUT_POLL * 2 + 1))
+PORT_BASE=20000
+
+# This function is used in the cleanup trap
+#shellcheck disable=SC2317,SC2329
+cleanup()
+{
+	for ns in "${ns1}" "${ns2}"; do
+		[ -n "${ns}" ] || continue
+		ip netns pids "${ns}" | xargs --no-run-if-empty kill -SIGKILL &>/dev/null
+	done
+	mptcp_lib_ns_exit "${ns1}" "${ns2}"
+}
+
+mptcp_lib_check_mptcp
+mptcp_lib_check_tools ip ss
+
+if ! mptcp_lib_kallsyms_has "tcpv6_prot_override$"; then
+	mptcp_lib_pr_skip "CONFIG_MPTCP_IPV6 not available"
+	exit ${KSFT_SKIP}
+fi
+
+if ! [ -r /proc/slabinfo ]; then
+	mptcp_lib_pr_skip "/proc/slabinfo not readable"
+	exit ${KSFT_SKIP}
+fi
+
+if ! awk '/^TCPv6 / { found = 1 } END { exit !found }' /proc/slabinfo; then
+	mptcp_lib_pr_skip "TCPv6 slab cache not present"
+	exit ${KSFT_SKIP}
+fi
+
+trap cleanup EXIT
+mptcp_lib_ns_init ns1 ns2
+
+ip -n "${ns1}" link add eth1 type veth peer name eth1 netns "${ns2}"
+ip -n "${ns1}" link set eth1 up
+ip -n "${ns1}" -6 addr add fc00::1/64 dev eth1 nodad
+ip -n "${ns2}" link set eth1 up
+ip -n "${ns2}" -6 addr add fc00::2/64 dev eth1 nodad
+
+# Wait for DAD-less addresses to settle
+ip -n "${ns1}" -6 route get fc00::2 >/dev/null 2>&1 || sleep 0.1
+
+get_tcpv6_active()
+{
+	awk '/^TCPv6 / { print $2 }' /proc/slabinfo
+}
+
+before=$(get_tcpv6_active)
+
+for i in $(seq 1 ${NR_CONNS}); do
+	echo "a" |
+		timeout ${TIMEOUT_TEST} \
+			ip netns exec "${ns2}" \
+				./mptcp_connect -6 -p $((PORT_BASE + i)) -l \
+					-t ${TIMEOUT_POLL} -w ${TIMEOUT_POLL} \
+					:: >/dev/null 2>&1 &
+done
+
+# wait_local_port_listen() only checks one port. Walk every port so we
+# do not start the connectors before all listeners are ready.
+for i in $(seq 1 ${NR_CONNS}); do
+	mptcp_lib_wait_local_port_listen "${ns2}" $((PORT_BASE + i))
+done
+
+for i in $(seq 1 ${NR_CONNS}); do
+	echo "b" |
+		timeout ${TIMEOUT_TEST} \
+			ip netns exec "${ns1}" \
+				./mptcp_connect -6 -p $((PORT_BASE + i)) \
+					-t ${TIMEOUT_POLL} -w ${TIMEOUT_POLL} \
+					fc00::2 >/dev/null 2>&1 &
+done
+
+# Wait for the accept side to materialise child sockets. ss reports the
+# number of established TCP connections in ns2 owned by mptcp_connect.
+established=0
+for _ in $(seq 20); do
+	established=$(ip netns exec "${ns2}" ss -H -t -6 state established 2>/dev/null | wc -l)
+	[ "${established}" -ge "${NR_CONNS}" ] && break
+	sleep 1
+done
+
+after=$(get_tcpv6_active)
+delta=$(( after - before ))
+
+# Conservative threshold: NR_CONNS connections (each producing at least
+# a server-side accepted child) must leave a clearly observable footprint
+# in the TCPv6 slab. On a regressed kernel, override.slab == NULL routes
+# every child into kmalloc-4k and the TCPv6 delta stays near zero.
+threshold=$(( NR_CONNS / 2 ))
+
+msg="TCPv6 slab gains MPTCPv6 subflow children"
+mptcp_lib_print_title "${msg}"
+if [ "${established}" -lt "${NR_CONNS}" ]; then
+	mptcp_lib_pr_fail "only ${established}/${NR_CONNS} connections established"
+	mptcp_lib_result_fail "${msg}"
+	ret=${KSFT_FAIL}
+elif [ "${delta}" -ge "${threshold}" ]; then
+	mptcp_lib_pr_ok "delta=${delta}"
+	mptcp_lib_result_pass "${msg}"
+else
+	mptcp_lib_pr_fail "delta=${delta} below ${threshold}:" \
+		"subflow children likely in kmalloc fallback"
+	mptcp_lib_result_fail "${msg}"
+	ret=${KSFT_FAIL}
+fi
+
+mptcp_lib_result_print_all_tap
+exit ${ret}
-- 
2.51.0


^ permalink raw reply related	[flat|nested] 5+ messages in thread
* [PATCH] selftests: mptcp: add test for IPv6 subflow SLAB placement
@ 2026-05-01 15:14 Vastargazing
  0 siblings, 0 replies; 5+ messages in thread
From: Vastargazing @ 2026-05-01 15:14 UTC (permalink / raw)
  To: vebohr; +Cc: stable

Add mptcp_v6_initcall.sh to verify that MPTCP IPv6 subflow child
sockets are allocated from the TCPv6 SLAB cache, not the kmalloc-4k
fallback.

tcpv6_prot_override must copy tcpv6_prot after proto_register(&tcpv6_prot)
populates tcpv6_prot.slab. If the copy runs too early, override.slab
stays NULL (frozen by __ro_after_init) and subflow children fall back
to kmalloc-4k. This lacks SLAB_TYPESAFE_BY_RCU, allowing lockless
ehash lookups in __inet_lookup_established to read freed memory.

The test exercises the IPv6 accept path via MPTCP connections between
two network namespaces, then checks that the TCPv6 slab active object
count grew. On a fixed kernel, the delta is ~2 * NR_CONNS (one subflow
per side per connection); on a broken kernel, it stays near zero because
children land in kmalloc-4k instead.

Topology: two netns connected via veth pair with /64 ULA addresses;
NR_CONNS parallel short-lived MPTCP connections are established and held
open long enough to sample /proc/slabinfo. The test skips if
CONFIG_MPTCP_IPV6 is absent (checked via kallsyms) or /proc/slabinfo is
unreadable.

Verified on Ubuntu 6.17 kernel predating the fix: TAP "not ok 1 ...
TCPv6 slab gains MPTCPv6 subflow children" with delta=0. On kernels
with the fix, delta is well above the threshold of NR_CONNS/2.

Fixes: b19bc2945b40 ("mptcp: implement delegated actions")
Cc: stable@vger.kernel.org
Signed-off-by: Vastargazing <vebohr@gmail.com>
---
 tools/testing/selftests/net/mptcp/Makefile    |   1 +
 .../selftests/net/mptcp/mptcp_v6_initcall.sh  | 140 ++++++++++++++++++
 2 files changed, 141 insertions(+)
 create mode 100644 tools/testing/selftests/net/mptcp/mptcp_v6_initcall.sh

diff --git a/tools/testing/selftests/net/mptcp/Makefile b/tools/testing/selftests/net/mptcp/Makefile
index 22ba0da2adb8..c9f329441490 100644
--- a/tools/testing/selftests/net/mptcp/Makefile
+++ b/tools/testing/selftests/net/mptcp/Makefile
@@ -14,6 +14,7 @@ TEST_PROGS := \
 	mptcp_connect_splice.sh \
 	mptcp_join.sh \
 	mptcp_sockopt.sh \
+	mptcp_v6_initcall.sh \
 	pm_netlink.sh \
 	simult_flows.sh \
 	userspace_pm.sh \
diff --git a/tools/testing/selftests/net/mptcp/mptcp_v6_initcall.sh b/tools/testing/selftests/net/mptcp/mptcp_v6_initcall.sh
new file mode 100644
index 000000000000..c55fc74b5ffb
--- /dev/null
+++ b/tools/testing/selftests/net/mptcp/mptcp_v6_initcall.sh
@@ -0,0 +1,140 @@
+#!/bin/bash
+# SPDX-License-Identifier: GPL-2.0
+#
+# Verify that MPTCP IPv6 subflow child sockets are allocated from the
+# TCPv6 slab cache.
+#
+# tcpv6_prot_override is initialised by copying tcpv6_prot, which is only
+# safe after proto_register(&tcpv6_prot) has populated tcpv6_prot.slab.
+# If the override copy runs too early during init, override.slab stays
+# NULL and __ro_after_init freezes that. Subflow child sockets then fall
+# back to kmalloc (kmalloc-4k for a TCPv6 sock), which lacks
+# SLAB_TYPESAFE_BY_RCU; lockless ehash lookups in __inet_lookup_established
+# can read freed memory.
+#
+# This test exercises the IPv6 accept path, which goes through the
+# override proto, and then asserts that the live TCPv6 slab population
+# grew. On a kernel where the override slab is NULL the delta is ~0,
+# because the children land in kmalloc-4k instead.
+
+#shellcheck disable=SC2086
+
+. "$(dirname "${0}")/mptcp_lib.sh"
+
+ns1=""
+ns2=""
+ret=0
+
+NR_CONNS=20
+TIMEOUT_POLL=30
+TIMEOUT_TEST=$((TIMEOUT_POLL * 2 + 1))
+PORT_BASE=20000
+
+# This function is used in the cleanup trap
+#shellcheck disable=SC2317,SC2329
+cleanup()
+{
+	for ns in "${ns1}" "${ns2}"; do
+		[ -n "${ns}" ] || continue
+		ip netns pids "${ns}" | xargs --no-run-if-empty kill -SIGKILL &>/dev/null
+	done
+	mptcp_lib_ns_exit "${ns1}" "${ns2}"
+}
+
+mptcp_lib_check_mptcp
+mptcp_lib_check_tools ip ss
+
+if ! mptcp_lib_kallsyms_has "tcpv6_prot_override$"; then
+	mptcp_lib_pr_skip "CONFIG_MPTCP_IPV6 not available"
+	exit ${KSFT_SKIP}
+fi
+
+if ! [ -r /proc/slabinfo ]; then
+	mptcp_lib_pr_skip "/proc/slabinfo not readable"
+	exit ${KSFT_SKIP}
+fi
+
+if ! awk '/^TCPv6 / { found = 1 } END { exit !found }' /proc/slabinfo; then
+	mptcp_lib_pr_skip "TCPv6 slab cache not present"
+	exit ${KSFT_SKIP}
+fi
+
+trap cleanup EXIT
+mptcp_lib_ns_init ns1 ns2
+
+ip -n "${ns1}" link add eth1 type veth peer name eth1 netns "${ns2}"
+ip -n "${ns1}" link set eth1 up
+ip -n "${ns1}" -6 addr add fc00::1/64 dev eth1 nodad
+ip -n "${ns2}" link set eth1 up
+ip -n "${ns2}" -6 addr add fc00::2/64 dev eth1 nodad
+
+# Wait for DAD-less addresses to settle
+ip -n "${ns1}" -6 route get fc00::2 >/dev/null 2>&1 || sleep 0.1
+
+get_tcpv6_active()
+{
+	awk '/^TCPv6 / { print $2 }' /proc/slabinfo
+}
+
+before=$(get_tcpv6_active)
+
+for i in $(seq 1 ${NR_CONNS}); do
+	echo "a" |
+		timeout ${TIMEOUT_TEST} \
+			ip netns exec "${ns2}" \
+				./mptcp_connect -6 -p $((PORT_BASE + i)) -l \
+					-t ${TIMEOUT_POLL} -w ${TIMEOUT_POLL} \
+					:: >/dev/null 2>&1 &
+done
+
+# wait_local_port_listen() only checks one port. Walk every port so we
+# do not start the connectors before all listeners are ready.
+for i in $(seq 1 ${NR_CONNS}); do
+	mptcp_lib_wait_local_port_listen "${ns2}" $((PORT_BASE + i))
+done
+
+for i in $(seq 1 ${NR_CONNS}); do
+	echo "b" |
+		timeout ${TIMEOUT_TEST} \
+			ip netns exec "${ns1}" \
+				./mptcp_connect -6 -p $((PORT_BASE + i)) \
+					-t ${TIMEOUT_POLL} -w ${TIMEOUT_POLL} \
+					fc00::2 >/dev/null 2>&1 &
+done
+
+# Wait for the accept side to materialise child sockets. ss reports the
+# number of established TCP connections in ns2 owned by mptcp_connect.
+established=0
+for _ in $(seq 20); do
+	established=$(ip netns exec "${ns2}" ss -H -t -6 state established 2>/dev/null | wc -l)
+	[ "${established}" -ge "${NR_CONNS}" ] && break
+	sleep 1
+done
+
+after=$(get_tcpv6_active)
+delta=$(( after - before ))
+
+# Conservative threshold: NR_CONNS connections (each producing at least
+# a server-side accepted child) must leave a clearly observable footprint
+# in the TCPv6 slab. On a regressed kernel, override.slab == NULL routes
+# every child into kmalloc-4k and the TCPv6 delta stays near zero.
+threshold=$(( NR_CONNS / 2 ))
+
+msg="TCPv6 slab gains MPTCPv6 subflow children"
+mptcp_lib_print_title "${msg}"
+if [ "${established}" -lt "${NR_CONNS}" ]; then
+	mptcp_lib_pr_fail "only ${established}/${NR_CONNS} connections established"
+	mptcp_lib_result_fail "${msg}"
+	ret=${KSFT_FAIL}
+elif [ "${delta}" -ge "${threshold}" ]; then
+	mptcp_lib_pr_ok "delta=${delta}"
+	mptcp_lib_result_pass "${msg}"
+else
+	mptcp_lib_pr_fail "delta=${delta} below ${threshold}:" \
+		"subflow children likely in kmalloc fallback"
+	mptcp_lib_result_fail "${msg}"
+	ret=${KSFT_FAIL}
+fi
+
+mptcp_lib_result_print_all_tap
+exit ${ret}
-- 
2.51.0


^ permalink raw reply related	[flat|nested] 5+ messages in thread
end of thread, other threads:[~2026-05-04 17:44 UTC | newest]

Thread overview: 5+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2026-05-01 15:14 [PATCH] selftests: mptcp: add test for IPv6 subflow SLAB placement Vastargazing
2026-05-01 15:45 ` Matthieu Baerts
2026-05-01 17:01   ` Vastargazing
2026-05-04 17:44     ` Matthieu Baerts
  -- strict thread matches above, loose matches on Subject: below --
2026-05-01 15:14 Vastargazing

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox