mptcp: mp_fail related fixes
Subsystems (2)
Paolo Abeni
maintainercover mptcp: mp_fail related fixes
While cooking a follow-up for "mptcp: invoke MP_FAIL response when needed" I stumbled upon a few other MP_FAIL related issues hit my the self-tests while mangling the above.
All the patches for for the -net tree, and should be applied in-order around to the squashed-to patch.
Note: with this series I still observe sporardic failures (once every ~200 runs) of tests 95. Given the low frequency here I did not investigate them yet...
Paolo Abeni (5):
mptcp: fix error mibs accounting
mptcp: introduce MAPPING_BAD_CSUM
Squash-to: "mptcp: invoke MP_FAIL response when needed"
mptcp: fix shutdown vs fallback race
mptcp: consistent map handling on failure
net/mptcp/options.c | 8 +++--- net/mptcp/pm.c | 5 ++-- net/mptcp/protocol.c | 56 ++++++++++++++++++++++++++++++++------- net/mptcp/protocol.h | 23 ++++++++++++---- net/mptcp/subflow.c | 63 ++++++++++++++++++++++++++++---------------- 5 files changed, 111 insertions(+), 44 deletions(-)
Signature · 1 line
2.35.3
Paolo Abeni
maintainerPATCH 1/5 mptcp: fix error mibs accounting
The current accounting for MP_FAIL and FASTCLOSE is not very
accurate: both can be increased even when the related option is not really sent. Move the accounting into the correct place.
--- net/mptcp/options.c | 6 +++--- net/mptcp/pm.c | 1 - net/mptcp/subflow.c | 4 +--- 3 files changed, 4 insertions(+), 7 deletions(-)
net/mptcp/options.c +3 -3
@@ -765,6 +765,7 @@ static noinline bool mptcp_established_options_rst(struct sock *sk, struct sk_bu opts->suboptions |=OPTION_MPTCP_RST ; opts->reset_transient = subflow->reset_transient; opts->reset_reason = subflow->reset_reason; +MPTCP_INC_STATS (sock_net (sk),MPTCP_MIB_MPRSTTX );return true; }
@@ -788,6 +789,7 @@ static bool mptcp_established_options_fastclose(struct sock *sk, opts->rcvr_key = msk->remote_key;pr_debug ("FASTCLOSE key=%llu" , opts->rcvr_key); +MPTCP_INC_STATS (sock_net (sk),MPTCP_MIB_MPFASTCLOSETX );return true; }
@@ -807,7 +809,7 @@ static bool mptcp_established_options_mp_fail(struct sock *sk, *size =TCPOLEN_MPTCP_FAIL ; opts->suboptions |=OPTION_MPTCP_FAIL ; opts->fail_seq = subflow->map_seq; - +MPTCP_INC_STATS (sock_net (sk),MPTCP_MIB_MPFAILTX );pr_debug ("MP_FAIL fail_seq=%llu" , opts->fail_seq);return true;
@@ -833,13 +835,11 @@ bool mptcp_established_options(struct sock *sk, struct sk_buff *skb, mptcp_established_options_mp_fail(sk, &opt_size, remaining, opts)) { *size += opt_size; remaining -= opt_size; -MPTCP_INC_STATS (sock_net (sk),MPTCP_MIB_MPFASTCLOSETX ); }/* MP_RST can be used with MP_FASTCLOSE and MP_FAIL if there is room */ if (mptcp_established_options_rst (sk, skb, &opt_size, remaining, opts)) { *size += opt_size; remaining -= opt_size; -MPTCP_INC_STATS (sock_net (sk),MPTCP_MIB_MPRSTTX ); }return true; }
net/mptcp/pm.c +0 -1
@@ -309,7 +309,6 @@ void mptcp_pm_mp_fail_received(struct sock *sk, u64 fail_seq)pr_debug ("send MP_FAIL response and infinite map" ); subflow->send_mp_fail =1 ; -MPTCP_INC_STATS (sock_net (sk),MPTCP_MIB_MPFAILTX ); subflow->send_infinite_map =1 ; }else {pr_debug ("MP_FAIL response received" );
net/mptcp/subflow.c +1 -3
@@ -958,10 +958,8 @@ static enum mapping_status validate_data_csum(struct sock *ssk, struct sk_buff * subflow->map_data_csum);if (unlikely (csum)) {MPTCP_INC_STATS (sock_net (ssk),MPTCP_MIB_DATACSUMERR ); -if (subflow->mp_join || subflow->valid_csum_seen) { +if (subflow->mp_join || subflow->valid_csum_seen) subflow->send_mp_fail =1 ; -MPTCP_INC_STATS (sock_net (ssk),MPTCP_MIB_MPFAILTX ); - }return subflow->mp_join ?MAPPING_INVALID :MAPPING_DUMMY ; }
Signature · 1 line
2.35.3
Paolo Abeni
maintainerPATCH 2/5 mptcp: introduce MAPPING_BAD_CSUM
In reply to Paolo AbeniThis allow moving a couple of conditional out of the fast path, making the code more easy to follow and will simplify the next patch.
--- net/mptcp/subflow.c | 18 +++++++++--------- 1 file changed, 9 insertions(+), 9 deletions(-)
net/mptcp/subflow.c +9 -9
@@ -843,7 +843,8 @@ enum mapping_status {MAPPING_INVALID ,MAPPING_EMPTY ,MAPPING_DATA_FIN , - MAPPING_DUMMY +MAPPING_DUMMY , + MAPPING_BAD_CSUM };static void dbg_bad_map (struct mptcp_subflow_context *subflow,u32 ssn)
@@ -958,9 +959,7 @@ static enum mapping_status validate_data_csum(struct sock *ssk, struct sk_buff * subflow->map_data_csum);if (unlikely (csum)) {MPTCP_INC_STATS (sock_net (ssk),MPTCP_MIB_DATACSUMERR ); -if (subflow->mp_join || subflow->valid_csum_seen) - subflow->send_mp_fail =1 ; -return subflow->mp_join ?MAPPING_INVALID :MAPPING_DUMMY ; +return MAPPING_BAD_CSUM ; } subflow->valid_csum_seen =1 ;
@@ -1178,10 +1177,8 @@ static bool subflow_check_data_avail(struct sock *ssk) status =get_mapping_status (ssk, msk);trace_subflow_check_data_avail (status,skb_peek (&ssk->sk_receive_queue)); -if (unlikely (status ==MAPPING_INVALID )) - goto fallback; - -if (unlikely (status ==MAPPING_DUMMY )) +if (unlikely (status ==MAPPING_INVALID || status ==MAPPING_DUMMY || + status ==MAPPING_BAD_CSUM )) goto fallback;if (status !=MAPPING_OK )
@@ -1223,7 +1220,10 @@ static bool subflow_check_data_avail(struct sock *ssk) fallback:if (!__mptcp_check_fallback (msk)) {/* RFC 8684 section 3.7. */ -if (subflow->send_mp_fail) { +if (status ==MAPPING_BAD_CSUM && + ((subflow->mp_join || subflow->valid_csum_seen))) { + subflow->send_mp_fail =1 ; +if (!READ_ONCE (msk->allow_infinite_fallback)) { ssk->sk_err =EBADMSG ;tcp_set_state (ssk,TCP_CLOSE );
Signature · 1 line
2.35.3
Paolo Abeni
maintainerPATCH 3/5 Squash-to: "mptcp: invoke MP_FAIL response when needed"
In reply to Paolo AbeniThis tries to address a few issues outstanding in the mentioned patch:
- we explicitly need to reset the timeout timer for mp_fail's sake
- we need to explicitly generate a tcp ack for mp_fail, otherwise there are no guarantees for suck option being sent out
- the timeout timer needs handling need some caring, as it's still shared between mp_fail and msk socket timeout.
- we can re-use msk->first for msk->fail_ssk, as only the first/mpc subflow can fail without reset. That additionally avoid the need to clear fail_ssk on the relevant subflow close.
- fail_tout would need some additional annotation. Just to be on the safe side move its manipulaiton under the ssk socket lock.
Last 2 paragraph of the squash to commit should be replaced with:
"""
It leverages the fact that only the MPC/first subflow can gracefully fail to avoid unneeded subflows traversal: the failing subflow can be only msk->first.
A new 'fail_tout' field is added to the subflow context to record the MP_FAIL response timeout and use such field to reliably share the timeout timer between the MP_FAIL event and the MPTCP socket close timeout.
Finally, a new ack is generated to send out MP_FAIL notification as soon as we hit the relevant condition, instead of waiting a possibly unbound time for the next data packet.
"""
--- net/mptcp/pm.c | 4 +++- net/mptcp/protocol.c | 54 ++++++++++++++++++++++++++++++++++++-------- net/mptcp/protocol.h | 4 ++-- net/mptcp/subflow.c | 24 ++++++++++++++++++-- 4 files changed, 72 insertions(+), 14 deletions(-)
net/mptcp/pm.c +3 -1
@@ -305,13 +305,15 @@ void mptcp_pm_mp_fail_received(struct sock *sk, u64 fail_seq)if (!READ_ONCE (msk->allow_infinite_fallback))return ; -if (!msk->fail_ssk) { +if (!subflow->fail_tout) {pr_debug ("send MP_FAIL response and infinite map" ); subflow->send_mp_fail =1 ; subflow->send_infinite_map =1 ; +tcp_send_ack (sk); }else {pr_debug ("MP_FAIL response received" ); +WRITE_ONCE (subflow->fail_tout,0 ); } }
net/mptcp/protocol.c +45 -9
121 lines not on this page · raw message
net/mptcp/protocol.h +2 -2
@@ -306,8 +306,6 @@ struct mptcp_sock {u32 setsockopt_seq;char ca_name[TCP_CA_NAME_MAX ]; -struct sock *fail_ssk; -unsigned long fail_tout; };#define mptcp_data_lock (sk) spin_lock_bh(&(sk)->sk_lock.slock)
@@ -483,6 +481,7 @@ struct mptcp_subflow_context {u8 stale_count;long delegated_status; +unsigned long fail_tout; );
@@ -663,6 +662,7 @@ void mptcp_get_options(const struct sk_buff *skb,void mptcp_finish_connect (struct sock *sk);void __mptcp_set_connected (struct sock *sk); +void mptcp_reset_timeout (struct mptcp_sock *msk,unsigned long fail_tout);static inline bool mptcp_is_fully_established (struct sock *sk) { return inet_sk_state_load(sk) == TCP_ESTABLISHED &&
net/mptcp/subflow.c +22 -2
@@ -1158,6 +1158,27 @@ static bool subflow_can_fallback(struct mptcp_subflow_context *subflow)return !subflow->fully_established; } +static void mptcp_subflow_fail (struct mptcp_sock *msk,struct sock *ssk) +{ +struct mptcp_subflow_context *subflow =mptcp_subflow_ctx (ssk); +unsigned long fail_tout; + +/* grecefull failure can happen only on the MPC subflow */ +if (WARN_ON_ONCE (ssk !=READ_ONCE (msk->first))) +return ; + +/* we don't need extreme accuracy here, use a zero fail_tout as special +* value meaning no fail timeout at all; +*/ + fail_tout = jiffies +TCP_RTO_MAX ; +if (!fail_tout) + fail_tout =1 ; +WRITE_ONCE (subflow->fail_tout, fail_tout); +tcp_send_ack (ssk); + +mptcp_reset_timeout (msk, subflow->fail_tout); +} +static bool subflow_check_data_avail (struct sock *ssk) {struct mptcp_subflow_context *subflow =mptcp_subflow_ctx (ssk);
@@ -1233,8 +1254,7 @@ static bool subflow_check_data_avail(struct sock *ssk)while ((skb =skb_peek (&ssk->sk_receive_queue)))sk_eat_skb (ssk, skb); }else { - msk->fail_ssk = ssk; - msk->fail_tout = jiffies +TCP_RTO_MAX ; +mptcp_subflow_fail (msk, ssk); }WRITE_ONCE (subflow->data_avail,MPTCP_SUBFLOW_NODATA );return true;
Signature · 1 line
2.35.3
Paolo Abeni
maintainerRe: [PATCH mptcp-net 3/5] Squash-to: "mptcp: invoke MP_FAIL response when needed"
Quoted 131 lines · Paolo Abeni
This tries to address a few issues outstanding in the mentioned patch:
- we explicitly need to reset the timeout timer for mp_fail's sake
- we need to explicitly generate a tcp ack formp_fail, otherwise there are no guarantees for suck option being sent out
- the timeout timer needs handling need some caring, as it's still shared betweenmp_failand msk socket timeout.
- we can re-usemsk->firstformsk->fail_ssk, as only the first/mpc subflow can fail without reset. That additionally avoid the need to clearfail_sskon the relevant subflow close.
-fail_toutwould need some additional annotation. Just to be on the safe side move its manipulaiton under the ssk socket lock.Last 2 paragraph of the squash to commit should be replaced with:
"""
It leverages the fact that only the MPC/first subflow can gracefully fail to avoid unneeded subflows traversal: the failing subflow can be onlymsk->first.A new '
fail_tout' field is added to the subflow context to record theMP_FAILresponse timeout and use such field to reliably share the timeout timer between theMP_FAILevent and the MPTCP socket close timeout.Finally, a new ack is generated to send out
MP_FAILnotification as soon as we hit the relevant condition, instead of waiting a possibly unbound time for the next data packet.
"""Signed-off-by Paolo Abeni <[email protected]>--- net/mptcp/pm.c | 4 +++- net/mptcp/protocol.c | 54 ++++++++++++++++++++++++++++++++++++-------- net/mptcp/protocol.h | 4 ++-- net/mptcp/subflow.c | 24 ++++++++++++++++++-- 4 files changed, 72 insertions(+), 14 deletions(-)net/mptcp/pm.c +3
-1@@ -305,13 +305,15 @@ void mptcp_pm_mp_fail_received(struct sock *sk, u64 fail_seq)if (!READ_ONCE (msk->allow_infinite_fallback))return ; -if (!msk->fail_ssk) { +if (!subflow->fail_tout) {pr_debug ("send MP_FAIL response and infinite map" ); subflow->send_mp_fail =1 ; subflow->send_infinite_map =1 ; +tcp_send_ack (sk); }else {pr_debug ("MP_FAIL response received" ); +WRITE_ONCE (subflow->fail_tout,0 ); } }net/mptcp/protocol.c +35
-3@@ -500,7 +500,7 @@ static void mptcp_set_timeout(struct sock *sk)__mptcp_set_timeout (sk, tout); } -static bool tcp_can_send_ack (const struct sock *ssk) +static inline bool tcp_can_send_ack (const struct sock *ssk) {return !((1 <<inet_sk_state_load (ssk)) & (TCPF_SYN_SENT |TCPF_SYN_RECV |TCPF_TIME_WAIT |TCPF_CLOSE |TCPF_LISTEN ));@@ -2503,24 +2503,56 @@ static void __mptcp_retrans(struct sock *sk)mptcp_reset_timer (sk); } +/* schedule the timeout timer for the nearest relevant event: either +* close timeout or mp_fail timeout. Both of them could be not +* scheduled yet +*/ +void mptcp_reset_timeout (struct mptcp_sock *msk,unsigned long fail_tout) +{ +struct sock *sk = (struct sock *)msk; +unsigned long timeout, close_timeout; + +if (!fail_tout && !sock_flag (sk,SOCK_DEAD )) +return ; + + close_timeout =inet_csk (sk)->icsk_mtup.probe_timestamp - tcp_jiffies32 + jiffies +TCP_TIMEWAIT_LEN ; + +/* the following is basically time_min(close_timeout, fail_tout) */ +if (!fail_tout) + timeout = close_timeout; +else if (!sock_flag (sk,SOCK_DEAD )) + timeout = fail_tout; +else if (time_after (close_timeout, fail_tout)) + timeout = fail_tout; +else + timeout = close_timeout; + +sk_reset_timer (sk, &sk->sk_timer, timeout); +} +static void mptcp_mp_fail_no_response (struct mptcp_sock *msk) { -struct sock *ssk = msk->fail_ssk; +struct sock *ssk = msk->first;bool slow; +if (!ssk) +return ; +pr_debug ("MP_FAIL doesn't respond, reset the subflow" ); slow =lock_sock_fast (ssk);mptcp_subflow_reset (ssk); +WRITE_ONCE (mptcp_subflow_ctx (ssk)->fail_tout,0 );unlock_sock_fast (ssk, slow); - msk->fail_ssk =NULL ; +mptcp_reset_timeout (msk,0 ); }static void mptcp_worker (struct work_struct *work) { struct mptcp_sock *msk = container_of(work, struct mptcp_sock, work);struct sock *sk = &msk->sk.icsk_inet.sk; +unsigned long fail_tout;int state;lock_sock (sk);
net/mptcp/protocol.c static void mptcp_worker(struct work_struct *work) · line 2592
Quoted 5 lines
net/mptcp/protocol.c +1
-1@@ -2557,7 +2589,8 @@ static void mptcp_worker(struct work_struct *work)if (test_and_clear_bit (MPTCP_WORK_RTX , &msk->flags))__mptcp_retrans (sk); -if (msk->fail_ssk &&time_after (jiffies, msk->fail_tout)) + fail_tout = msk->first ?READ_ONCE (mptcp_subflow_ctx (msk->first)->fail_tout) :0 ;
For the records, the above apparently trigger an UaF in the CI debug build on pktdrill tests.
AFAICS that is cause by the listener mptcp socket being destroyed while the MPC is still in accept queue: the MPC socket is freed, but msk-
Quoted 1 line
first is not updated accordingly.
Note that the above issue is not related to this patch-series, but it makes it much more easy to reproduce.
/P
Paolo Abeni
maintainerPATCH 4/5 mptcp: fix shutdown vs fallback race
In reply to Paolo AbeniIf the MPTCP socket shutdown happens before a fallback to TCP, and all the pending data have been already spooled, we never close the TCP connection.
Address the issue explicitly checking for critical condition at fallback time.
--- net/mptcp/options.c | 2 +- net/mptcp/protocol.c | 2 +- net/mptcp/protocol.h | 19 ++++++++++++++++--- net/mptcp/subflow.c | 2 +- 4 files changed, 19 insertions(+), 6 deletions(-)
net/mptcp/options.c +1 -1
@@ -966,7 +966,7 @@ static bool check_fully_established(struct mptcp_sock *msk, struct sock *ssk, goto reset; subflow->mp_capable =0 ;pr_fallback (msk); -__mptcp_do_fallback (msk); +mptcp_do_fallback (ssk);return false; }
net/mptcp/protocol.c +1 -1
@@ -1245,7 +1245,7 @@ static void mptcp_update_infinite_map(struct mptcp_sock *msk,MPTCP_INC_STATS (sock_net (ssk),MPTCP_MIB_INFINITEMAPTX );mptcp_subflow_ctx (ssk)->send_infinite_map =0 ;pr_fallback (msk); -__mptcp_do_fallback (msk); +mptcp_do_fallback (ssk); } static int mptcp_sendmsg_frag(struct sock *sk, struct sock *ssk,
net/mptcp/protocol.h +16 -3
@@ -927,12 +927,25 @@ static inline void __mptcp_do_fallback(struct mptcp_sock *msk)set_bit (MPTCP_FALLBACK_DONE , &msk->flags); } -static inline void mptcp_do_fallback (struct sock *sk) +static inline void mptcp_do_fallback (struct sock *ssk) { -struct mptcp_subflow_context *subflow =mptcp_subflow_ctx (sk); -struct mptcp_sock *msk =mptcp_sk (subflow->conn); +struct mptcp_subflow_context *subflow =mptcp_subflow_ctx (ssk); +struct sock *sk = subflow->conn; +struct mptcp_sock *msk; + msk =mptcp_sk (sk);__mptcp_do_fallback (msk); +if (READ_ONCE (msk->snd_data_fin_enable) && !(ssk->sk_shutdown &SEND_SHUTDOWN )) { +gfp_t saved_allocation = ssk->sk_allocation; + +/* we are in a atomic (BH) scope, override ssk default for data +* fin allocation +*/ + ssk->sk_allocation =GFP_ATOMIC ; + ssk->sk_shutdown |=SEND_SHUTDOWN ; +tcp_shutdown (ssk,SEND_SHUTDOWN ); + ssk->sk_allocation = saved_allocation; + } }#define pr_fallback (a) pr_debug("%s:fallback to TCP (msk=%p)", __func__, a)
net/mptcp/subflow.c +1 -1
@@ -1273,7 +1273,7 @@ static bool subflow_check_data_avail(struct sock *ssk)return false; } -__mptcp_do_fallback (msk); +mptcp_do_fallback (ssk); } skb =skb_peek (&ssk->sk_receive_queue);
Signature · 1 line
2.35.3
Paolo Abeni
maintainerPATCH 5/5 mptcp: consistent map handling on failure
In reply to Paolo AbeniWhen the MPTCP receive path reach a non fatal fall-back condition, e.g. when the MPC sockets must fall-back to TCP, the existing code is a little
self-inconsistent: it reports that new data is available - return true - but sets the MPC flag to the opposite value.
As the consequence read operations in some exceptional scenario may block unexpectedly.
Address the issue setting the correct MPC read status. Additionally avoid some code duplication in the fatal fall-back scenario.
--- net/mptcp/subflow.c | 19 +++++++++---------- 1 file changed, 9 insertions(+), 10 deletions(-)
net/mptcp/subflow.c +9 -10
@@ -1246,17 +1246,12 @@ static bool subflow_check_data_avail(struct sock *ssk) subflow->send_mp_fail =1 ;if (!READ_ONCE (msk->allow_infinite_fallback)) { - ssk->sk_err =EBADMSG ; -tcp_set_state (ssk,TCP_CLOSE ); subflow->reset_transient =0 ; subflow->reset_reason =MPTCP_RST_EMIDDLEBOX ; -tcp_send_active_reset (ssk,GFP_ATOMIC ); -while ((skb =skb_peek (&ssk->sk_receive_queue))) -sk_eat_skb (ssk, skb); - }else { -mptcp_subflow_fail (msk, ssk); + goto reset; } -WRITE_ONCE (subflow->data_avail,MPTCP_SUBFLOW_NODATA ); +mptcp_subflow_fail (msk, ssk); +WRITE_ONCE (subflow->data_avail,MPTCP_SUBFLOW_DATA_AVAIL );return true; }
@@ -1264,10 +1259,14 @@ static bool subflow_check_data_avail(struct sock *ssk)/* fatal protocol error, close the socket. * subflow_error_report() will introduce the appropriate barriers */ - ssk->sk_err =EBADMSG ; -tcp_set_state (ssk,TCP_CLOSE ); subflow->reset_transient =0 ; subflow->reset_reason =MPTCP_RST_EMPTCP ; + +reset: + ssk->sk_err =EBADMSG ; +tcp_set_state (ssk,TCP_CLOSE ); +while ((skb =skb_peek (&ssk->sk_receive_queue))) +sk_eat_skb (ssk, skb);tcp_send_active_reset (ssk,GFP_ATOMIC );WRITE_ONCE (subflow->data_avail,MPTCP_SUBFLOW_NODATA );return false;
Signature · 1 line
2.35.3
Re: mptcp: consistent map handling on failure: Tests Results
Hi Paolo,
Thank you for your modifications, that's great!
Our CI did some validations and here is its report:
- KVM Validation: normal:
- Success! ✅:
- Task: https://cirrus-ci.com/task/5751797246590976
- Summary: https://api.cirrus-ci.com/v1/artifact/task/5751797246590976/summary/summary.txt
- KVM Validation: debug:
- Critical: 1 Call Trace(s) ❌:
- Task: https://cirrus-ci.com/task/5188847293169664
- Summary: https://api.cirrus-ci.com/v1/artifact/task/5188847293169664/summary/summary.txt
Initiator: Patchew Applier
Commits: https://github.com/multipath-tcp/mptcp_net-next/commits/c95b21fa8a7b
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-debugFor 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)