Here is a first batch of fixes for v6.5 and older.
The fixes are not linked to each others.
Patch 1 ensures subflows are unhashed before cleaning the backlog to avoid races. This fixes another recent fix from v6.4.
Patch 2 does not rely on implicit state check in mptcp_listen() to avoid races when receiving an MP_FASTCLOSE. A regression from v5.17.
The rest fixes issues in the selftests.
Patch 3 makes sure errors when setting up the environment are no longer ignored. For v5.17+.
Patch 4 uses 'iptables-legacy' if available to be able to run on older kernels. A fix for v5.13 and newer.
Patch 5 catches errors when issues are detected with packet marks. Also for v5.13+.
Patch 6 uses the correct variable instead of an undefined one. Even if there was no visible impact, it can help to find regressions later. An issue visible in v5.19+.
Patch 7 makes sure errors with some sub-tests are reported to have the selftest marked as failed as expected. Also for v5.19+.
Patch 8 adds a kernel config that is required to execute MPTCP selftests. It is valid for v5.9+.
Patch 9 fixes issues when validating the userspace path-manager with 32-bit arch, an issue affecting v5.19+.
Signed-off-by: Matthieu Baerts matthieu.baerts@tessares.net --- Matthieu Baerts (7): selftests: mptcp: connect: fail if nft supposed to work selftests: mptcp: sockopt: use 'iptables-legacy' if available selftests: mptcp: sockopt: return error if wrong mark selftests: mptcp: userspace_pm: use correct server port selftests: mptcp: userspace_pm: report errors with 'remove' tests selftests: mptcp: depend on SYN_COOKIES selftests: mptcp: pm_nl_ctl: fix 32-bit support
Paolo Abeni (2): mptcp: ensure subflow is unhashed before cleaning the backlog mptcp: do not rely on implicit state check in mptcp_listen()
net/mptcp/protocol.c | 7 +++++- tools/testing/selftests/net/mptcp/config | 1 + tools/testing/selftests/net/mptcp/mptcp_connect.sh | 3 +++ tools/testing/selftests/net/mptcp/mptcp_sockopt.sh | 29 ++++++++++++---------- tools/testing/selftests/net/mptcp/pm_nl_ctl.c | 10 ++++---- tools/testing/selftests/net/mptcp/userspace_pm.sh | 4 ++- 6 files changed, 34 insertions(+), 20 deletions(-) --- base-commit: 14bb236b29922c4f57d8c05bfdbcb82677f917c9 change-id: 20230704-upstream-net-20230704-misc-fixes-6-5-rc1-c52608649559
Best regards,
From: Paolo Abeni pabeni@redhat.com
While tacking care of the mptcp-level listener I unintentionally moved the subflow level unhash after the subflow listener backlog cleanup.
That could cause some nasty race and makes the code harder to read.
Address the issue restoring the proper order of operations.
Fixes: 57fc0f1ceaa4 ("mptcp: ensure listener is unhashed before updating the sk status") Cc: stable@vger.kernel.org Signed-off-by: Paolo Abeni pabeni@redhat.com Reviewed-by: Matthieu Baerts matthieu.baerts@tessares.net Signed-off-by: Matthieu Baerts matthieu.baerts@tessares.net --- net/mptcp/protocol.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/net/mptcp/protocol.c b/net/mptcp/protocol.c index e892673deb73..489a3defdde5 100644 --- a/net/mptcp/protocol.c +++ b/net/mptcp/protocol.c @@ -2909,10 +2909,10 @@ static void mptcp_check_listen_stop(struct sock *sk) return;
lock_sock_nested(ssk, SINGLE_DEPTH_NESTING); + tcp_set_state(ssk, TCP_CLOSE); mptcp_subflow_queue_clean(sk, ssk); inet_csk_listen_stop(ssk); mptcp_event_pm_listener(ssk, MPTCP_EVENT_LISTENER_CLOSED); - tcp_set_state(ssk, TCP_CLOSE); release_sock(ssk); }
From: Paolo Abeni pabeni@redhat.com
Since the blamed commit, closing the first subflow resets the first subflow socket state to SS_UNCONNECTED.
The current mptcp listen implementation relies only on such state to prevent touching not-fully-disconnected sockets.
Incoming mptcp fastclose (or paired endpoint removal) unconditionally closes the first subflow.
All the above allows an incoming fastclose followed by a listen() call to successfully race with a blocking recvmsg(), potentially causing the latter to hit a divide by zero bug in cleanup_rbuf/__tcp_select_window().
Address the issue explicitly checking the msk socket state in mptcp_listen(). An alternative solution would be moving the first subflow socket state update into mptcp_disconnect(), but in the long term the first subflow socket should be removed: better avoid relaying on it for internal consistency check.
Fixes: b29fcfb54cd7 ("mptcp: full disconnect implementation") Cc: stable@vger.kernel.org Reported-by: Christoph Paasch cpaasch@apple.com Closes: https://github.com/multipath-tcp/mptcp_net-next/issues/414 Signed-off-by: Paolo Abeni pabeni@redhat.com Reviewed-by: Matthieu Baerts matthieu.baerts@tessares.net Signed-off-by: Matthieu Baerts matthieu.baerts@tessares.net --- net/mptcp/protocol.c | 5 +++++ 1 file changed, 5 insertions(+)
diff --git a/net/mptcp/protocol.c b/net/mptcp/protocol.c index 489a3defdde5..3613489eb6e3 100644 --- a/net/mptcp/protocol.c +++ b/net/mptcp/protocol.c @@ -3703,6 +3703,11 @@ static int mptcp_listen(struct socket *sock, int backlog) pr_debug("msk=%p", msk);
lock_sock(sk); + + err = -EINVAL; + if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM) + goto unlock; + ssock = __mptcp_nmpc_socket(msk); if (IS_ERR(ssock)) { err = PTR_ERR(ssock);
In case of "external" errors when preparing the environment for the TProxy tests, the subtests were marked as skipped.
This is fine but it means these errors are ignored. On MPTCP Public CI, we do want to catch such issues and mark the selftest as failed if there are such issues. We can then use mptcp_lib_fail_if_expected_feature() helper that has been recently added to fail if needed.
Link: https://github.com/multipath-tcp/mptcp_net-next/issues/368 Fixes: 5fb62e9cd3ad ("selftests: mptcp: add tproxy test case") Cc: stable@vger.kernel.org Acked-by: Paolo Abeni pabeni@redhat.com Signed-off-by: Matthieu Baerts matthieu.baerts@tessares.net --- tools/testing/selftests/net/mptcp/mptcp_connect.sh | 3 +++ 1 file changed, 3 insertions(+)
diff --git a/tools/testing/selftests/net/mptcp/mptcp_connect.sh b/tools/testing/selftests/net/mptcp/mptcp_connect.sh index 13561e5bc0cd..bbae40882bfa 100755 --- a/tools/testing/selftests/net/mptcp/mptcp_connect.sh +++ b/tools/testing/selftests/net/mptcp/mptcp_connect.sh @@ -718,6 +718,7 @@ table inet mangle { EOF if [ $? -ne 0 ]; then echo "SKIP: $msg, could not load nft ruleset" + mptcp_lib_fail_if_expected_feature "nft rules" return fi
@@ -733,6 +734,7 @@ EOF if [ $? -ne 0 ]; then ip netns exec "$listener_ns" nft flush ruleset echo "SKIP: $msg, ip $r6flag rule failed" + mptcp_lib_fail_if_expected_feature "ip rule" return fi
@@ -741,6 +743,7 @@ EOF ip netns exec "$listener_ns" nft flush ruleset ip -net "$listener_ns" $r6flag rule del fwmark 1 lookup 100 echo "SKIP: $msg, ip route add local $local_addr failed" + mptcp_lib_fail_if_expected_feature "ip route" return fi
IPTables commands using 'iptables-nft' fail on old kernels, at least on v5.15 because it doesn't see the default IPTables chains:
$ iptables -L iptables/1.8.2 Failed to initialize nft: Protocol not supported
As a first step before switching to NFTables, we can use iptables-legacy if available.
Link: https://github.com/multipath-tcp/mptcp_net-next/issues/368 Fixes: dc65fe82fb07 ("selftests: mptcp: add packet mark test case") Cc: stable@vger.kernel.org Acked-by: Paolo Abeni pabeni@redhat.com Signed-off-by: Matthieu Baerts matthieu.baerts@tessares.net --- tools/testing/selftests/net/mptcp/mptcp_sockopt.sh | 20 +++++++++++--------- 1 file changed, 11 insertions(+), 9 deletions(-)
diff --git a/tools/testing/selftests/net/mptcp/mptcp_sockopt.sh b/tools/testing/selftests/net/mptcp/mptcp_sockopt.sh index f295a371ff14..c21bfd7f0c01 100755 --- a/tools/testing/selftests/net/mptcp/mptcp_sockopt.sh +++ b/tools/testing/selftests/net/mptcp/mptcp_sockopt.sh @@ -12,6 +12,8 @@ ksft_skip=4 timeout_poll=30 timeout_test=$((timeout_poll * 2 + 1)) mptcp_connect="" +iptables="iptables" +ip6tables="ip6tables"
sec=$(date +%s) rndh=$(printf %x $sec)-$(mktemp -u XXXXXX) @@ -25,7 +27,7 @@ add_mark_rules() local m=$2
local t - for t in iptables ip6tables; do + for t in ${iptables} ${ip6tables}; do # just to debug: check we have multiple subflows connection requests ip netns exec $ns $t -A OUTPUT -p tcp --syn -m mark --mark $m -j ACCEPT
@@ -95,14 +97,14 @@ if [ $? -ne 0 ];then exit $ksft_skip fi
-iptables -V > /dev/null 2>&1 -if [ $? -ne 0 ];then +# Use the legacy version if available to support old kernel versions +if iptables-legacy -V &> /dev/null; then + iptables="iptables-legacy" + ip6tables="ip6tables-legacy" +elif ! iptables -V &> /dev/null; then echo "SKIP: Could not run all tests without iptables tool" exit $ksft_skip -fi - -ip6tables -V > /dev/null 2>&1 -if [ $? -ne 0 ];then +elif ! ip6tables -V &> /dev/null; then echo "SKIP: Could not run all tests without ip6tables tool" exit $ksft_skip fi @@ -112,10 +114,10 @@ check_mark() local ns=$1 local af=$2
- local tables=iptables + local tables=${iptables}
if [ $af -eq 6 ];then - tables=ip6tables + tables=${ip6tables} fi
local counters values
When an error was detected when checking the marks, a message was correctly printed mentioning the error but followed by another one saying everything was OK and the selftest was not marked as failed as expected.
Now the 'ret' variable is directly set to 1 in order to make sure the exit is done with an error, similar to what is done in other functions. While at it, the error is correctly propagated to the caller.
Link: https://github.com/multipath-tcp/mptcp_net-next/issues/368 Fixes: dc65fe82fb07 ("selftests: mptcp: add packet mark test case") Cc: stable@vger.kernel.org Acked-by: Paolo Abeni pabeni@redhat.com Signed-off-by: Matthieu Baerts matthieu.baerts@tessares.net --- tools/testing/selftests/net/mptcp/mptcp_sockopt.sh | 9 +++++---- 1 file changed, 5 insertions(+), 4 deletions(-)
diff --git a/tools/testing/selftests/net/mptcp/mptcp_sockopt.sh b/tools/testing/selftests/net/mptcp/mptcp_sockopt.sh index c21bfd7f0c01..dc8d473fc82c 100755 --- a/tools/testing/selftests/net/mptcp/mptcp_sockopt.sh +++ b/tools/testing/selftests/net/mptcp/mptcp_sockopt.sh @@ -128,6 +128,7 @@ check_mark() for v in $values; do if [ $v -ne 0 ]; then echo "FAIL: got $tables $values in ns $ns , not 0 - not all expected packets marked" 1>&2 + ret=1 return 1 fi done @@ -227,11 +228,11 @@ do_transfer() fi
if [ $local_addr = "::" ];then - check_mark $listener_ns 6 - check_mark $connector_ns 6 + check_mark $listener_ns 6 || retc=1 + check_mark $connector_ns 6 || retc=1 else - check_mark $listener_ns 4 - check_mark $connector_ns 4 + check_mark $listener_ns 4 || retc=1 + check_mark $connector_ns 4 || retc=1 fi
check_transfer $cin $sout "file received by server"
"server4_port" variable is not set but "app4_port" is the server port in v4 and the correct variable name to use.
The port is optional so there was no visible impact.
Link: https://github.com/multipath-tcp/mptcp_net-next/issues/368 Fixes: ca188a25d43f ("selftests: mptcp: userspace PM support for MP_PRIO signals") Cc: stable@vger.kernel.org Acked-by: Paolo Abeni pabeni@redhat.com Signed-off-by: Matthieu Baerts matthieu.baerts@tessares.net --- tools/testing/selftests/net/mptcp/userspace_pm.sh | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/tools/testing/selftests/net/mptcp/userspace_pm.sh b/tools/testing/selftests/net/mptcp/userspace_pm.sh index 98d9e4d2d3fc..841a67a7d524 100755 --- a/tools/testing/selftests/net/mptcp/userspace_pm.sh +++ b/tools/testing/selftests/net/mptcp/userspace_pm.sh @@ -848,7 +848,7 @@ test_prio() local count
# Send MP_PRIO signal from client to server machine - ip netns exec "$ns2" ./pm_nl_ctl set 10.0.1.2 port "$client4_port" flags backup token "$client4_token" rip 10.0.1.1 rport "$server4_port" + ip netns exec "$ns2" ./pm_nl_ctl set 10.0.1.2 port "$client4_port" flags backup token "$client4_token" rip 10.0.1.1 rport "$app4_port" sleep 0.5
# Check TX
A message was mentioning an issue with the "remove" tests but the selftest was not marked as failed.
Directly exit with an error like it is done everywhere else in this selftest.
Link: https://github.com/multipath-tcp/mptcp_net-next/issues/368 Fixes: 259a834fadda ("selftests: mptcp: functional tests for the userspace PM type") Cc: stable@vger.kernel.org Acked-by: Paolo Abeni pabeni@redhat.com Signed-off-by: Matthieu Baerts matthieu.baerts@tessares.net --- tools/testing/selftests/net/mptcp/userspace_pm.sh | 2 ++ 1 file changed, 2 insertions(+)
diff --git a/tools/testing/selftests/net/mptcp/userspace_pm.sh b/tools/testing/selftests/net/mptcp/userspace_pm.sh index 841a67a7d524..b180133a30af 100755 --- a/tools/testing/selftests/net/mptcp/userspace_pm.sh +++ b/tools/testing/selftests/net/mptcp/userspace_pm.sh @@ -423,6 +423,7 @@ test_remove() stdbuf -o0 -e0 printf "[OK]\n" else stdbuf -o0 -e0 printf "[FAIL]\n" + exit 1 fi
# RM_ADDR using an invalid addr id should result in no action @@ -437,6 +438,7 @@ test_remove() stdbuf -o0 -e0 printf "[OK]\n" else stdbuf -o0 -e0 printf "[FAIL]\n" + exit 1 fi
# RM_ADDR from the client to server machine
MPTCP selftests are using TCP SYN Cookies for quite a while now, since v5.9.
Some CIs don't have this config option enabled and this is causing issues in the tests:
# ns1 MPTCP -> ns1 (10.0.1.1:10000 ) MPTCP (duration 167ms) sysctl: cannot stat /proc/sys/net/ipv4/tcp_syncookies: No such file or directory # [ OK ]./mptcp_connect.sh: line 554: [: -eq: unary operator expected
There is no impact in the results but the test is not doing what it is supposed to do.
Fixes: fed61c4b584c ("selftests: mptcp: make 2nd net namespace use tcp syn cookies unconditionally") Cc: stable@vger.kernel.org Signed-off-by: Matthieu Baerts matthieu.baerts@tessares.net --- tools/testing/selftests/net/mptcp/config | 1 + 1 file changed, 1 insertion(+)
diff --git a/tools/testing/selftests/net/mptcp/config b/tools/testing/selftests/net/mptcp/config index 6032f9b23c4c..e317c2e44dae 100644 --- a/tools/testing/selftests/net/mptcp/config +++ b/tools/testing/selftests/net/mptcp/config @@ -6,6 +6,7 @@ CONFIG_INET_DIAG=m CONFIG_INET_MPTCP_DIAG=m CONFIG_VETH=y CONFIG_NET_SCH_NETEM=m +CONFIG_SYN_COOKIES=y CONFIG_NETFILTER=y CONFIG_NETFILTER_ADVANCED=y CONFIG_NETFILTER_NETLINK=m
When using pm_nl_ctl to validate userspace path-manager's behaviours, it was failing on 32-bit architectures ~half of the time.
pm_nl_ctl was not reporting any error but the command was not doing what it was expected to do. As a result, the expected linked event was not triggered after and the test failed.
This is due to the fact the token given in argument to the application was parsed as an integer with atoi(): in a 32-bit arch, if the number was bigger than INT_MAX, 2147483647 was used instead.
This can simply be fixed by using strtoul() instead of atoi().
The errors have been seen "by chance" when manually looking at the results from LKFT.
Fixes: 9a0b36509df0 ("selftests: mptcp: support MPTCP_PM_CMD_ANNOUNCE") Cc: stable@vger.kernel.org Fixes: ecd2a77d672f ("selftests: mptcp: support MPTCP_PM_CMD_REMOVE") Fixes: cf8d0a6dfd64 ("selftests: mptcp: support MPTCP_PM_CMD_SUBFLOW_CREATE") Fixes: 57cc361b8d38 ("selftests: mptcp: support MPTCP_PM_CMD_SUBFLOW_DESTROY") Fixes: ca188a25d43f ("selftests: mptcp: userspace PM support for MP_PRIO signals") Signed-off-by: Matthieu Baerts matthieu.baerts@tessares.net --- tools/testing/selftests/net/mptcp/pm_nl_ctl.c | 10 +++++----- 1 file changed, 5 insertions(+), 5 deletions(-)
diff --git a/tools/testing/selftests/net/mptcp/pm_nl_ctl.c b/tools/testing/selftests/net/mptcp/pm_nl_ctl.c index abddf4c63e79..1887bd61bd9a 100644 --- a/tools/testing/selftests/net/mptcp/pm_nl_ctl.c +++ b/tools/testing/selftests/net/mptcp/pm_nl_ctl.c @@ -425,7 +425,7 @@ int dsf(int fd, int pm_family, int argc, char *argv[]) }
/* token */ - token = atoi(params[4]); + token = strtoul(params[4], NULL, 10); rta = (void *)(data + off); rta->rta_type = MPTCP_PM_ATTR_TOKEN; rta->rta_len = RTA_LENGTH(4); @@ -551,7 +551,7 @@ int csf(int fd, int pm_family, int argc, char *argv[]) }
/* token */ - token = atoi(params[4]); + token = strtoul(params[4], NULL, 10); rta = (void *)(data + off); rta->rta_type = MPTCP_PM_ATTR_TOKEN; rta->rta_len = RTA_LENGTH(4); @@ -598,7 +598,7 @@ int remove_addr(int fd, int pm_family, int argc, char *argv[]) if (++arg >= argc) error(1, 0, " missing token value");
- token = atoi(argv[arg]); + token = strtoul(argv[arg], NULL, 10); rta = (void *)(data + off); rta->rta_type = MPTCP_PM_ATTR_TOKEN; rta->rta_len = RTA_LENGTH(4); @@ -710,7 +710,7 @@ int announce_addr(int fd, int pm_family, int argc, char *argv[]) if (++arg >= argc) error(1, 0, " missing token value");
- token = atoi(argv[arg]); + token = strtoul(argv[arg], NULL, 10); } else error(1, 0, "unknown keyword %s", argv[arg]); } @@ -1347,7 +1347,7 @@ int set_flags(int fd, int pm_family, int argc, char *argv[]) error(1, 0, " missing token value");
/* token */ - token = atoi(argv[arg]); + token = strtoul(argv[arg], NULL, 10); } else if (!strcmp(argv[arg], "flags")) { char *tok, *str;
Hello:
This series was applied to netdev/net.git (main) by David S. Miller davem@davemloft.net:
On Tue, 04 Jul 2023 22:44:32 +0200 you wrote:
Here is a first batch of fixes for v6.5 and older.
The fixes are not linked to each others.
Patch 1 ensures subflows are unhashed before cleaning the backlog to avoid races. This fixes another recent fix from v6.4.
[...]
Here is the summary with links: - [net,1/9] mptcp: ensure subflow is unhashed before cleaning the backlog https://git.kernel.org/netdev/net/c/3fffa15bfef4 - [net,2/9] mptcp: do not rely on implicit state check in mptcp_listen() https://git.kernel.org/netdev/net/c/0226436acf24 - [net,3/9] selftests: mptcp: connect: fail if nft supposed to work https://git.kernel.org/netdev/net/c/221e4550454a - [net,4/9] selftests: mptcp: sockopt: use 'iptables-legacy' if available https://git.kernel.org/netdev/net/c/a5a5990c099d - [net,5/9] selftests: mptcp: sockopt: return error if wrong mark https://git.kernel.org/netdev/net/c/9ac4c28eb70c - [net,6/9] selftests: mptcp: userspace_pm: use correct server port https://git.kernel.org/netdev/net/c/d8566d0e0392 - [net,7/9] selftests: mptcp: userspace_pm: report errors with 'remove' tests https://git.kernel.org/netdev/net/c/966c6c3adfb1 - [net,8/9] selftests: mptcp: depend on SYN_COOKIES https://git.kernel.org/netdev/net/c/6c8880fcaa5c - [net,9/9] selftests: mptcp: pm_nl_ctl: fix 32-bit support https://git.kernel.org/netdev/net/c/61d965805026
You are awesome, thank you!
linux-kselftest-mirror@lists.linaro.org