Skip to contentSkip to conversation

mptcp: fix another deadlock issue

PATCH Superseded Paolo Abeni 4 participants 8 messages Feb 21, 2024 – Feb 26, 2024

Subsystems (4)

v1 of 3 Superseded Latest is v3 →

lore
Share & tools
lore

Paolo Abeni

author maintainer
#0 lore raw

cover mptcp: fix another deadlock issue

Eric reported and diagnosed another possible deadlock problem with the MPTCP diag code. The first patch address the issue, and the second adds self-test coverage for the relevant codepath, to get lockdep help to catch future similar issues.

Paolo Abeni (2):
mptcp: fix possible deadlock in subflow diag
selftests: mptcp: explicitly trigger the listener diag code-path

 net/mptcp/diag.c                          |  3 +++
 tools/testing/selftests/net/mptcp/diag.sh | 30 +++++++++++++++++++++++
 2 files changed, 33 insertions(+)
Signature · 1 line
2.43.0
author maintainer
#1 lore raw

PATCH 1/2 mptcp: fix possible deadlock in subflow diag

Syzbot and Eric reported a lockdep splat in the subflow diag:

WARNING: possible circular locking dependency detected 6.8.0-rc4-syzkaller-00212-g40b9385dd8e6 #0 Not tainted

syz-executor.2/24141 is trying to acquire lock:
ffff888045870130 (k-sk_lock-AF_INET6){+.+.}-{0:0}, at:

   tcp_diag_put_ulp net/ipv4/tcp_diag.c:100 [inline]
   ffff888045870130 (k-sk_lock-AF_INET6){+.+.}-{0:0}, at:
   tcp_diag_get_aux+0x738/0x830 net/ipv4/tcp_diag.c:137

but task is already holding lock:
ffffc9000135e488 (&h->lhash2[i].lock){+.+.}-{2:2}, at: spin_lock include/linux/spinlock.h:351 [inline] ffffc9000135e488 (&h->lhash2[i].lock){+.+.}-{2:2}, at:
inet_diag_dump_icsk+0x39f/0x1f80 net/ipv4/inet_diag.c:1038

which lock already depends on the new lock.

the existing dependency chain (in reverse order) is:

-> #1 (&h->lhash2[i].lock){+.+.}-{2:2}:

   lock_acquire+0x1e3/0x530 kernel/locking/lockdep.c:5754
   __raw_spin_lock include/linux/spinlock_api_smp.h:133 [inline]
   _raw_spin_lock+0x2e/0x40 kernel/locking/spinlock.c:154
   spin_lock include/linux/spinlock.h:351 [inline]
   __inet_hash+0x335/0xbe0 net/ipv4/inet_hashtables.c:743
   inet_csk_listen_start+0x23a/0x320 net/ipv4/inet_connection_sock.c:1261
   __inet_listen_sk+0x2a2/0x770 net/ipv4/af_inet.c:217
   inet_listen+0xa3/0x110 net/ipv4/af_inet.c:239
   rds_tcp_listen_init+0x3fd/0x5a0 net/rds/tcp_listen.c:316
   rds_tcp_init_net+0x141/0x320 net/rds/tcp.c:577
   ops_init+0x352/0x610 net/core/net_namespace.c:136
   __register_pernet_operations net/core/net_namespace.c:1214 [inline]
   register_pernet_operations+0x2cb/0x660 net/core/net_namespace.c:1283
   register_pernet_device+0x33/0x80 net/core/net_namespace.c:1370
   rds_tcp_init+0x62/0xd0 net/rds/tcp.c:735
   do_one_initcall+0x238/0x830 init/main.c:1236
   do_initcall_level+0x157/0x210 init/main.c:1298
   do_initcalls+0x3f/0x80 init/main.c:1314
   kernel_init_freeable+0x42f/0x5d0 init/main.c:1551
   kernel_init+0x1d/0x2a0 init/main.c:1441
   ret_from_fork+0x4b/0x80 arch/x86/kernel/process.c:147
   ret_from_fork_asm+0x1b/0x30 arch/x86/entry/entry_64.S:242

-> #0 (k-sk_lock-AF_INET6){+.+.}-{0:0}:

   check_prev_add kernel/locking/lockdep.c:3134 [inline]
   check_prevs_add kernel/locking/lockdep.c:3253 [inline]
   validate_chain+0x18ca/0x58e0 kernel/locking/lockdep.c:3869
   __lock_acquire+0x1345/0x1fd0 kernel/locking/lockdep.c:5137
   lock_acquire+0x1e3/0x530 kernel/locking/lockdep.c:5754
   lock_sock_fast include/net/sock.h:1723 [inline]
   subflow_get_info+0x166/0xd20 net/mptcp/diag.c:28
   tcp_diag_put_ulp net/ipv4/tcp_diag.c:100 [inline]
   tcp_diag_get_aux+0x738/0x830 net/ipv4/tcp_diag.c:137
   inet_sk_diag_fill+0x10ed/0x1e00 net/ipv4/inet_diag.c:345
   inet_diag_dump_icsk+0x55b/0x1f80 net/ipv4/inet_diag.c:1061
   __inet_diag_dump+0x211/0x3a0 net/ipv4/inet_diag.c:1263
   inet_diag_dump_compat+0x1c1/0x2d0 net/ipv4/inet_diag.c:1371
   netlink_dump+0x59b/0xc80 net/netlink/af_netlink.c:2264
   __netlink_dump_start+0x5df/0x790 net/netlink/af_netlink.c:2370
   netlink_dump_start include/linux/netlink.h:338 [inline]
   inet_diag_rcv_msg_compat+0x209/0x4c0 net/ipv4/inet_diag.c:1405
   sock_diag_rcv_msg+0xe7/0x410
   netlink_rcv_skb+0x1e3/0x430 net/netlink/af_netlink.c:2543
   sock_diag_rcv+0x2a/0x40 net/core/sock_diag.c:280
   netlink_unicast_kernel net/netlink/af_netlink.c:1341 [inline]
   netlink_unicast+0x7ea/0x980 net/netlink/af_netlink.c:1367
   netlink_sendmsg+0xa3b/0xd70 net/netlink/af_netlink.c:1908
   sock_sendmsg_nosec net/socket.c:730 [inline]
   __sock_sendmsg+0x221/0x270 net/socket.c:745
   ____sys_sendmsg+0x525/0x7d0 net/socket.c:2584
   ___sys_sendmsg net/socket.c:2638 [inline]
   __sys_sendmsg+0x2b0/0x3a0 net/socket.c:2667
   do_syscall_64+0xf9/0x240
   entry_SYSCALL_64_after_hwframe+0x6f/0x77

As noted by Eric we can break the lock dependency chain avoid dumping any extended info for the mptcp subflow listener:
nothing actually useful is presented there.

Fixes b8adb69a7d29 ("mptcp: fix lockless access in subflow ULP diag")
---
 net/mptcp/diag.c | 3 +++
 1 file changed, 3 insertions(+)
net/mptcp/diag.c +3 -0
index 6ff6f14674aa..7017dd60659d 100644
@@ -21,6 +21,9 @@ static int subflow_get_info(struct sock *sk, struct sk_buff *skb)
 	bool slow;
 	int err;
 
+	if (inet_sk_state_load(sk) == TCP_LISTEN)
+		return 0;
+
 	start = nla_nest_start_noflag(skb, INET_ULP_INFO_MPTCP);
 	if (!start)
 		return -EMSGSIZE;
Signature · 1 line
2.43.0
author maintainer
#2 lore raw

PATCH 2/2 selftests: mptcp: explicitly trigger the listener diag code-path

In reply to Paolo Abeni

The mptcp diag interface already experienced a few locking bugs that lockdep and appropriate coverage have detected in advance.

Let's add a test-case triggering the relevant code path, to prevent similar issues in the future.

---
 tools/testing/selftests/net/mptcp/diag.sh | 30 +++++++++++++++++++++++
 1 file changed, 30 insertions(+)
tools/testing/selftests/net/mptcp/diag.sh +30 -0
index 60a7009ce1b5..3ab584b38566 100755
@@ -81,6 +81,21 @@ chk_msk_nr()
 	__chk_msk_nr "grep -c token:" "$@"
 }
 
+chk_listener_nr()
+{
+	local expected=$1
+	local msg="$2"
+
+	if [ $expected -gt 0 ] && \
+	   ! mptcp_lib_kallsyms_has "mptcp_diag_dump_listeners"; then
+		printf "%-50s%s\n" "$msg - mptcp" "[ skip ]"
+		mptcp_lib_result_skip "many listener sockets"
+	else
+		__chk_nr "ss -inmlHMON $ns | wc -l" "$expected" "$msg - mptcp"
+	fi
+	__chk_nr "ss -inmlHtON $ns | wc -l" "$expected" "$msg - subflows"
+}
+
 wait_msk_nr()
 {
 	local condition="grep -c token:"
@@ -279,5 +294,20 @@ flush_pids
 chk_msk_inuse 0 "many->0"
 chk_msk_cestab 0 "many->0"
 
+chk_listener_nr 0 "no listener sockets"
+NR_SERVERS=100
+for I in $(seq 1 $NR_SERVERS); do
+	ip netns exec $ns ./mptcp_connect -p $((I + 20001)) -l 0.0.0.0 2>&1 >/dev/null &
+	mptcp_lib_wait_local_port_listen $ns $((I + 20001))
+done
+
+chk_listener_nr $NR_SERVERS "many listener sockets"
+
+# gracefull termination
+for I in $(seq 1 $NR_SERVERS); do
+	echo a | ip netns exec $ns ./mptcp_connect -p $((I + 20001)) 127.0.0.1 2>&1 >/dev/null
+done
+flush_pids
+
 mptcp_lib_result_print_all_tap
 exit $ret
Signature · 1 line
2.43.0
#3 lore raw

Re: selftests: mptcp: explicitly trigger the listener diag code-path: Tests Results

Hi Paolo,

Thank you for your modifications, that's great!

Our CI (GitHub Action) did some validations and here is its report:

- KVM Validation: normal:
- Unstable: 2 failed test(s): selftest_diag selftest_mptcp_join 🔴:
- Task: https://github.com/multipath-tcp/mptcp_net-next/actions/runs/7988706773

Initiator: Patchew Applier
Commits: https://github.com/multipath-tcp/mptcp_net-next/commits/dac9a523a0e5

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-normal

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 (NGI0 Core)

author maintainer
#4 lore raw

Re: [PATCH net 2/2] selftests: mptcp: explicitly trigger the listener diag code-path

In reply to Paolo Abeni
Quoted 18 lines · Paolo Abeni

The mptcp diag interface already experienced a few locking bugs that lockdep and appropriate coverage have detected in advance.

Let's add a test-case triggering the relevant code path, to prevent similar issues in the future.

---
 tools/testing/selftests/net/mptcp/diag.sh | 30 +++++++++++++++++++++++
 1 file changed, 30 insertions(+)
tools/testing/selftests/net/mptcp/diag.sh +0 -0
index 60a7009ce1b5..3ab584b38566 100755
@@ -81,6 +81,21 @@ chk_msk_nr()
 	__chk_msk_nr "grep -c token:" "$@"
 }

tools/testing/selftests/net/mptcp/diag.sh chk_msk_nr() · line 92

Quoted 10 lines
tools/testing/selftests/net/mptcp/diag.sh +9 -0
 
+chk_listener_nr()
+{
+	local expected=$1
+	local msg="$2"
+
+	if [ $expected -gt 0 ] && \
+	   ! mptcp_lib_kallsyms_has "mptcp_diag_dump_listeners"; then
+		printf "%-50s%s\n" "$msg - mptcp" "[ skip ]"
+		mptcp_lib_result_skip "many listener sockets"

I was too hasty and this is too fragile/requires debug. A v2 with something more reliable is needed. Given this is not going to land into this week PR, we will probably send it with a burst with other fixes.

Thanks,

Paolo

reviewer
#5 lore raw

Re: [PATCH net 2/2] selftests: mptcp: explicitly trigger the listener diag code-path

In reply to Paolo Abeni
Quoted 41 lines · Paolo Abeni

l The mptcp diag interface already experienced a few locking bugs that lockdep and appropriate coverage have detected in advance.

Let's add a test-case triggering the relevant code path, to prevent similar issues in the future.

---
 tools/testing/selftests/net/mptcp/diag.sh | 30 +++++++++++++++++++++++
 1 file changed, 30 insertions(+)
tools/testing/selftests/net/mptcp/diag.sh +15 -0
index 60a7009ce1b5..3ab584b38566 100755
@@ -81,6 +81,21 @@ chk_msk_nr()
 	__chk_msk_nr "grep -c token:" "$@"
 }
 
+chk_listener_nr()
+{
+	local expected=$1
+	local msg="$2"
+
+	if [ $expected -gt 0 ] && \
+	   ! mptcp_lib_kallsyms_has "mptcp_diag_dump_listeners"; then
+		printf "%-50s%s\n" "$msg - mptcp" "[ skip ]"
+		mptcp_lib_result_skip "many listener sockets"
+	else
+		__chk_nr "ss -inmlHMON $ns | wc -l" "$expected" "$msg - mptcp"
+	fi
+	__chk_nr "ss -inmlHtON $ns | wc -l" "$expected" "$msg - subflows"
+}
+
 wait_msk_nr()
 {
 	local condition="grep -c token:"
@@ -279,5 +294,20 @@ flush_pids
 chk_msk_inuse 0 "many->0"
 chk_msk_cestab 0 "many->0"
 

tools/testing/selftests/net/mptcp/diag.sh flush_pids · line 306

Quoted 10 lines
tools/testing/selftests/net/mptcp/diag.sh +10 -0
+chk_listener_nr 0 "no listener sockets"
+NR_SERVERS=100
+for I in $(seq 1 $NR_SERVERS); do
+	ip netns exec $ns ./mptcp_connect -p $((I + 20001)) -l 0.0.0.0 2>&1 >/dev/null &
+	mptcp_lib_wait_local_port_listen $ns $((I + 20001))
+done
+
+chk_listener_nr $NR_SERVERS "many listener sockets"
+
+# gracefull termination

nit, as you plan to re-spin anyway: graceful

tools/testing/selftests/net/mptcp/diag.sh flush_pids · line 308

Quoted 2 lines
tools/testing/selftests/net/mptcp/diag.sh +2 -0
+for I in $(seq 1 $NR_SERVERS); do
+	echo a | ip netns exec $ns ./mptcp_connect -p $((I + 20001)) 127.0.0.1 2>&1 >/dev/null

I'm not sure of the validity of this but shellcheck complains that:

SC2069 (warning): To redirect stdout+stderr, 2>&1 must be last (or use '{ cmd > file; } 2>&1' to clarify).

Also for the same constriction slightly earlier in this patch.

Quoted 9 lines
tools/testing/selftests/net/mptcp/diag.sh +3 -0
+done
+flush_pids
+
 mptcp_lib_result_print_all_tap
 exit $ret
Signature · 1 line
2.43.0
maintainer
#6 lore raw

Re: [PATCH net 2/2] selftests: mptcp: explicitly trigger the listener diag code-path

Hi Simon,

Thank you for the review!

Quoted 42 lines · Simon Horman
Quoted 41 lines · Paolo Abeni

l The mptcp diag interface already experienced a few locking bugs that lockdep and appropriate coverage have detected in advance.

Let's add a test-case triggering the relevant code path, to prevent similar issues in the future.

---
 tools/testing/selftests/net/mptcp/diag.sh | 30 +++++++++++++++++++++++
 1 file changed, 30 insertions(+)
tools/testing/selftests/net/mptcp/diag.sh +15 -0
index 60a7009ce1b5..3ab584b38566 100755
@@ -81,6 +81,21 @@ chk_msk_nr()
 	__chk_msk_nr "grep -c token:" "$@"
 }
 
+chk_listener_nr()
+{
+	local expected=$1
+	local msg="$2"
+
+	if [ $expected -gt 0 ] && \
+	   ! mptcp_lib_kallsyms_has "mptcp_diag_dump_listeners"; then
+		printf "%-50s%s\n" "$msg - mptcp" "[ skip ]"
+		mptcp_lib_result_skip "many listener sockets"
+	else
+		__chk_nr "ss -inmlHMON $ns | wc -l" "$expected" "$msg - mptcp"
+	fi
+	__chk_nr "ss -inmlHtON $ns | wc -l" "$expected" "$msg - subflows"
+}
+
 wait_msk_nr()
 {
 	local condition="grep -c token:"
@@ -279,5 +294,20 @@ flush_pids
 chk_msk_inuse 0 "many->0"
 chk_msk_cestab 0 "many->0"
 
Quoted 12 lines

tools/testing/selftests/net/mptcp/diag.sh flush_pids · line 306

Quoted 10 lines
tools/testing/selftests/net/mptcp/diag.sh +10 -0
+chk_listener_nr 0 "no listener sockets"
+NR_SERVERS=100
+for I in $(seq 1 $NR_SERVERS); do
+	ip netns exec $ns ./mptcp_connect -p $((I + 20001)) -l 0.0.0.0 2>&1 >/dev/null &
+	mptcp_lib_wait_local_port_listen $ns $((I + 20001))
+done
+
+chk_listener_nr $NR_SERVERS "many listener sockets"
+
+# gracefull termination

nit, as you plan to re-spin anyway: graceful

Paolo sent a new version on MPTCP ML only. When applying the new version, I fixed this:

https://lore.kernel.org/mptcp/[email protected]/

Quoted 8 lines

tools/testing/selftests/net/mptcp/diag.sh flush_pids · line 308

Quoted 2 lines
tools/testing/selftests/net/mptcp/diag.sh +2 -0
+for I in $(seq 1 $NR_SERVERS); do
+	echo a | ip netns exec $ns ./mptcp_connect -p $((I + 20001)) 127.0.0.1 2>&1 >/dev/null

I'm not sure of the validity of this but shellcheck complains that:

SC2069 (warning): To redirect stdout+stderr, 2>&1 must be last (or use '{ cmd > file; } 2>&1' to clarify).

Also for the same constriction slightly earlier in this patch.

Good catch!

I should add shellcheck support in our CI to catch that :)

I just applied the suggested modification in our tree [1]. The new version that will be sent to netdev will no longer have this issue.

https://github.com/multipath-tcp/mptcp_net-next/commit/31f1460d552c

Cheers,
Matt

Signature · 1 line
Sponsored by the NGI0 Core fund.
reviewer
#7 lore raw

Re: [PATCH net 2/2] selftests: mptcp: explicitly trigger the listener diag code-path

Quoted 74 lines · Matthieu Baerts

Hi Simon,

Thank you for the review!

Quoted 42 lines · Simon Horman
Quoted 41 lines · Paolo Abeni

l The mptcp diag interface already experienced a few locking bugs that lockdep and appropriate coverage have detected in advance.

Let's add a test-case triggering the relevant code path, to prevent similar issues in the future.

---
 tools/testing/selftests/net/mptcp/diag.sh | 30 +++++++++++++++++++++++
 1 file changed, 30 insertions(+)
tools/testing/selftests/net/mptcp/diag.sh +15 -0
index 60a7009ce1b5..3ab584b38566 100755
@@ -81,6 +81,21 @@ chk_msk_nr()
 	__chk_msk_nr "grep -c token:" "$@"
 }
 
+chk_listener_nr()
+{
+	local expected=$1
+	local msg="$2"
+
+	if [ $expected -gt 0 ] && \
+	   ! mptcp_lib_kallsyms_has "mptcp_diag_dump_listeners"; then
+		printf "%-50s%s\n" "$msg - mptcp" "[ skip ]"
+		mptcp_lib_result_skip "many listener sockets"
+	else
+		__chk_nr "ss -inmlHMON $ns | wc -l" "$expected" "$msg - mptcp"
+	fi
+	__chk_nr "ss -inmlHtON $ns | wc -l" "$expected" "$msg - subflows"
+}
+
 wait_msk_nr()
 {
 	local condition="grep -c token:"
@@ -279,5 +294,20 @@ flush_pids
 chk_msk_inuse 0 "many->0"
 chk_msk_cestab 0 "many->0"
 
Quoted 12 lines

tools/testing/selftests/net/mptcp/diag.sh flush_pids · line 306

Quoted 10 lines
tools/testing/selftests/net/mptcp/diag.sh +10 -0
+chk_listener_nr 0 "no listener sockets"
+NR_SERVERS=100
+for I in $(seq 1 $NR_SERVERS); do
+	ip netns exec $ns ./mptcp_connect -p $((I + 20001)) -l 0.0.0.0 2>&1 >/dev/null &
+	mptcp_lib_wait_local_port_listen $ns $((I + 20001))
+done
+
+chk_listener_nr $NR_SERVERS "many listener sockets"
+
+# gracefull termination

nit, as you plan to re-spin anyway: graceful

Paolo sent a new version on MPTCP ML only. When applying the new version, I fixed this:

https://lore.kernel.org/mptcp/[email protected]/

Quoted 8 lines

tools/testing/selftests/net/mptcp/diag.sh flush_pids · line 308

Quoted 2 lines
tools/testing/selftests/net/mptcp/diag.sh +2 -0
+for I in $(seq 1 $NR_SERVERS); do
+	echo a | ip netns exec $ns ./mptcp_connect -p $((I + 20001)) 127.0.0.1 2>&1 >/dev/null

I'm not sure of the validity of this but shellcheck complains that:

SC2069 (warning): To redirect stdout+stderr, 2>&1 must be last (or use '{ cmd > file; } 2>&1' to clarify).

Also for the same constriction slightly earlier in this patch.

Quoted 8 lines

Good catch!

I should add shellcheck support in our CI to catch that :)

I just applied the suggested modification in our tree [1]. The new version that will be sent to netdev will no longer have this issue.

https://github.com/multipath-tcp/mptcp_net-next/commit/31f1460d552c

Thanks Matt,

much appreciated.