mptcp: improve subflow creation on errors
Subsystems (3)
Paolo Abeni
maintainerPATCH v5 1/5 mptcp: fix per socket endpoint accounting.
In reply to Paolo AbeniSince full-mesh endpoint support, the reception of a single ADD_ADDR option can cause multiple subflows creation. When such option is accepted we increment 'add_addr_accepted' by one. When we received a paired RM_ADDR option, we deleted all the relevant subflows, decrementing 'add_addr_accepted' by one for each of them.
We have a similar issue for 'local_addr_used'
Fix them moving the pm endpoint accounting outside the subflow traversal.
--- net/mptcp/pm_netlink.c | 18 ++++++++++++------ 1 file changed, 12 insertions(+), 6 deletions(-)
net/mptcp/pm_netlink.c +12 -6
@@ -710,6 +710,8 @@ static void mptcp_pm_nl_rm_addr_or_subflow(struct mptcp_sock *msk,return ;for (i =0 ; i < rm_list->nr; i++) { +bool removed = false; +list_for_each_entry_safe (subflow, tmp, &msk->conn_list, node) {struct sock *ssk =mptcp_subflow_tcp_sock (subflow);int how =RCV_SHUTDOWN |SEND_SHUTDOWN ;
@@ -729,15 +731,19 @@ static void mptcp_pm_nl_rm_addr_or_subflow(struct mptcp_sock *msk,mptcp_close_ssk (sk, ssk, subflow);spin_lock_bh (&msk->pm.lock); -if (rm_type ==MPTCP_MIB_RMADDR ) { - msk->pm.add_addr_accepted--; -WRITE_ONCE (msk->pm.accept_addr, true); - }else if (rm_type ==MPTCP_MIB_RMSUBFLOW ) { - msk->pm.local_addr_used--; - } + removed = true; msk->pm.subflows--;__MPTCP_INC_STATS (sock_net (sk), rm_type); } +if (!removed) +continue ; + +if (rm_type ==MPTCP_MIB_RMADDR ) { + msk->pm.add_addr_accepted--; +WRITE_ONCE (msk->pm.accept_addr, true); + }else if (rm_type ==MPTCP_MIB_RMSUBFLOW ) { + msk->pm.local_addr_used--; + } } }
Signature · 1 line
2.33.1
Paolo Abeni
maintainerv5 cover mptcp: improve subflow creation on errors
This iteration adds tracking for the endpoint used by the MPC subflow, updated the self-tests accordingly (1 signal tests-case is affected) and introduces a new test case for the relevant scenario.
Currently when a subflow connection fails - either the TCP connection is reset by the peer, or the MPJ handshake never completes - the in kernel PM don't perform any further action. Notably no additional subflow creation is attempted.
This series is aimed at improving the in-kernel path manager behavior in the above scenario:
- each endpoint is used only once per connection (patch 2/4)
- on failure we try to move to the next subflow, if any (patch 3/4)
The first patch is minor cleanup, and the last patch adds specific self-tests.
This should address/close:
https://github.com/multipath-tcp/mptcp_net-next/issues/235
https://github.com/multipath-tcp/mptcp_net-next/issues/242
Paolo Abeni (5):
mptcp: fix per socket endpoint accounting.
mptcp: clean-up MPJ option writing.
mptcp: keep track of local endpoint still available for each msk
mptcp: do not block subflows creation on errors
selftests: mptcp: add tests for subflow creation failure
net/mptcp/options.c | 65 +++++--- net/mptcp/pm.c | 24 ++- net/mptcp/pm_netlink.c | 142 +++++++++++------- net/mptcp/protocol.c | 9 +- net/mptcp/protocol.h | 16 +- net/mptcp/subflow.c | 5 +- tools/testing/selftests/net/mptcp/config | 1 + .../testing/selftests/net/mptcp/mptcp_join.sh | 79 +++++++++- 8 files changed, 254 insertions(+), 87 deletions(-)
Signature · 1 line
2.33.1
Paolo Abeni
maintainerPATCH v5 2/5 mptcp: clean-up MPJ option writing.
Check for all MPJ variant at once, this reduces the number of conditionals traversed on average and will simplify the next patch.
No functional change intended.
--- net/mptcp/options.c | 44 +++++++++++++++++++++++--------------------- 1 file changed, 23 insertions(+), 21 deletions(-)
net/mptcp/options.c +23 -21
@@ -1381,27 +1381,29 @@ void mptcp_write_options(__be32 *ptr, const struct tcp_sock *tp,/* MPC is additionally mutually exclusive with MP_PRIO */ goto mp_capable_done; - }else if (OPTION_MPTCP_MPJ_SYN & opts->suboptions) { - *ptr++ =mptcp_option (MPTCPOPT_MP_JOIN , -TCPOLEN_MPTCP_MPJ_SYN , - opts->backup, opts->join_id); -put_unaligned_be32 (opts->token, ptr); - ptr +=1 ; -put_unaligned_be32 (opts->nonce, ptr); - ptr +=1 ; - }else if (OPTION_MPTCP_MPJ_SYNACK & opts->suboptions) { - *ptr++ =mptcp_option (MPTCPOPT_MP_JOIN , -TCPOLEN_MPTCP_MPJ_SYNACK , - opts->backup, opts->join_id); -put_unaligned_be64 (opts->thmac, ptr); - ptr +=2 ; -put_unaligned_be32 (opts->nonce, ptr); - ptr +=1 ; - }else if (OPTION_MPTCP_MPJ_ACK & opts->suboptions) { - *ptr++ =mptcp_option (MPTCPOPT_MP_JOIN , -TCPOLEN_MPTCP_MPJ_ACK ,0 ,0 ); -memcpy (ptr, opts->hmac,MPTCPOPT_HMAC_LEN ); - ptr +=5 ; + }else if (OPTIONS_MPTCP_MPJ & opts->suboptions) { +if (OPTION_MPTCP_MPJ_SYN & opts->suboptions) { + *ptr++ =mptcp_option (MPTCPOPT_MP_JOIN , +TCPOLEN_MPTCP_MPJ_SYN , + opts->backup, opts->join_id); +put_unaligned_be32 (opts->token, ptr); + ptr +=1 ; +put_unaligned_be32 (opts->nonce, ptr); + ptr +=1 ; + }else if (OPTION_MPTCP_MPJ_SYNACK & opts->suboptions) { + *ptr++ =mptcp_option (MPTCPOPT_MP_JOIN , +TCPOLEN_MPTCP_MPJ_SYNACK , + opts->backup, opts->join_id); +put_unaligned_be64 (opts->thmac, ptr); + ptr +=2 ; +put_unaligned_be32 (opts->nonce, ptr); + ptr +=1 ; + }else { + *ptr++ =mptcp_option (MPTCPOPT_MP_JOIN , +TCPOLEN_MPTCP_MPJ_ACK ,0 ,0 ); +memcpy (ptr, opts->hmac,MPTCPOPT_HMAC_LEN ); + ptr +=5 ; + } }else if (OPTION_MPTCP_ADD_ADDR & opts->suboptions) {u8 len =TCPOLEN_MPTCP_ADD_ADDR_BASE ;u8 echo =MPTCP_ADDR_ECHO ;
Signature · 1 line
2.33.1
Paolo Abeni
maintainerPATCH v5 4/5 mptcp: do not block subflows creation on errors
In reply to Paolo AbeniIf the MPTCP configuration allows for multiple subflows creation, and the first additional subflows never reach the fully established status - e.g. due to packets drop or reset - the in kernel path manager do not move to the next subflow.
This patch introduces a new PM helper to cope with MPJ subflow creation failure and delay and hook it where appropriate.
Such helper triggers additional subflow creation, as needed and updates the PM subflow counter, if the current one is closing.
Note that we don't need an additional timer to catch timeout and/or long delay in connection completion: we just need to measure the time elapsed since the subflow creation every time we emit an MPJ sub-option.
---
v3 -> v4:
- fix 'subflows' underflow
v2 -> v3:
- fix compile warning (CI)
v1 -> v2:
- explicitly hook on subflow close instead of error notification
- fix checkpatch issue (Mat)
- introduce and use a new pm helper to cope with subflow close
- update pm subflow counters on close
--- net/mptcp/options.c | 21 ++++++++++++++++++--- net/mptcp/pm.c | 23 +++++++++++++++++++++-- net/mptcp/pm_netlink.c | 20 ++++++++++++-------- net/mptcp/protocol.c | 6 ++++++ net/mptcp/protocol.h | 5 ++++- net/mptcp/subflow.c | 5 ++++- 6 files changed, 65 insertions(+), 15 deletions(-)
net/mptcp/options.c +18 -3
@@ -1257,10 +1257,10 @@ static u16 mptcp_make_csum(const struct mptcp_ext *mpext)void mptcp_write_options (__be32 *ptr,const struct tcp_sock *tp,struct mptcp_out_options *opts) { -if (unlikely (OPTION_MPTCP_FAIL & opts->suboptions)) { -const struct sock *ssk = (const struct sock *)tp; -struct mptcp_subflow_context *subflow; +const struct sock *ssk = (const struct sock *)tp; +struct mptcp_subflow_context *subflow; +if (unlikely (OPTION_MPTCP_FAIL & opts->suboptions)) { subflow =mptcp_subflow_ctx (ssk); subflow->send_mp_fail =0 ;
@@ -1382,6 +1382,21 @@ void mptcp_write_options(__be32 *ptr, const struct tcp_sock *tp,/* MPC is additionally mutually exclusive with MP_PRIO */ goto mp_capable_done; }else if (OPTIONS_MPTCP_MPJ & opts->suboptions) { +if (ssk) { +/* we are still in the MPJ handshake and "a lot" of time passed +* e.g. due to syn retranmissions. We can attempt next +* subflow creation +*/ + subflow =mptcp_subflow_ctx (ssk); +if (subflow->start_stamp && +unlikely (after (tcp_jiffies32, subflow->start_stamp +HZ /10 ))) { +mptcp_pm_subflow_check_next (mptcp_sk (subflow->conn), ssk, subflow); + +/* avoid triggering the PM multiple times due to timeout */ + subflow->start_stamp =0 ; + } + } +if (OPTION_MPTCP_MPJ_SYN & opts->suboptions) { *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN, TCPOLEN_MPTCP_MPJ_SYN,
net/mptcp/pm.c +21 -2
@@ -172,9 +172,28 @@ void mptcp_pm_subflow_established(struct mptcp_sock *msk)spin_unlock_bh (&pm->lock); } -void mptcp_pm_subflow_closed (struct mptcp_sock *msk,u8 id) +void mptcp_pm_subflow_check_next (struct mptcp_sock *msk,const struct sock *ssk, +const struct mptcp_subflow_context *subflow) { -pr_debug ("msk=%p" , msk); +struct mptcp_pm_data *pm = &msk->pm; +bool update_subflows; + + update_subflows = (ssk->sk_state ==TCP_CLOSE ) && + (subflow->request_join || subflow->mp_join); +if (!READ_ONCE (pm->work_pending) && !update_subflows) +return ; + +spin_lock_bh (&pm->lock); +if (update_subflows) + pm->subflows--; + +/* Even if this subflow is not really established, tell the PM to try +* to pick the next one, if possible. +*/ +if (mptcp_pm_nl_check_work_pending (msk)) +mptcp_pm_schedule_work (msk,MPTCP_PM_SUBFLOW_ESTABLISHED ); + +spin_unlock_bh (&pm->lock); } void mptcp_pm_add_addr_received(struct mptcp_sock *msk,
net/mptcp/pm_netlink.c +12 -8
@@ -251,14 +251,17 @@ unsigned int mptcp_pm_get_local_addr_max(struct mptcp_sock *msk) }EXPORT_SYMBOL_GPL (mptcp_pm_get_local_addr_max); -static void check_work_pending (struct mptcp_sock *msk) +bool mptcp_pm_nl_check_work_pending (struct mptcp_sock *msk) {struct pm_nl_pernet *pernet =net_generic (sock_net ((struct sock *)msk), pm_nl_pernet_id);if (msk->pm.subflows ==mptcp_pm_get_subflows_max (msk) || (find_next_and_bit (pernet->id_bitmap, msk->pm.id_avail_bitmap,MAX_ADDR_ID +1 ,0 ) == -MAX_ADDR_ID +1 )) +MAX_ADDR_ID +1 )) {WRITE_ONCE (msk->pm.work_pending, false); +return false; + } +return true; } struct mptcp_pm_add_entry *
@@ -503,12 +506,12 @@ static void mptcp_pm_create_subflow_or_signal_addr(struct mptcp_sock *msk)/* do lazy endpoint usage accounting for the MPC subflows */ if (unlikely (!(msk->pm.status &BIT (MPTCP_PM_MPC_ENDPOINT_ACCOUNTED ))) && msk->first) { -struct mptcp_addr_info local; +struct mptcp_addr_info mpc_addr;int mpc_id; -local_address ((struct sock_common *)msk->first, &local); - mpc_id =lookup_id_by_addr (pernet, &local); -if (mpc_id <0 ) +local_address ((struct sock_common *)msk->first, &mpc_addr); + mpc_id =lookup_id_by_addr (pernet, &mpc_addr); +if (mpc_id >=0 )__clear_bit (mpc_id, msk->pm.id_avail_bitmap); msk->pm.status |=BIT (MPTCP_PM_MPC_ENDPOINT_ACCOUNTED );
@@ -553,7 +556,7 @@ static void mptcp_pm_create_subflow_or_signal_addr(struct mptcp_sock *msk)spin_lock_bh (&msk->pm.lock); } } -check_work_pending (msk); +mptcp_pm_nl_check_work_pending (msk); }static void mptcp_pm_nl_fully_established (struct mptcp_sock *msk)
@@ -760,11 +763,12 @@ static void mptcp_pm_nl_rm_addr_or_subflow(struct mptcp_sock *msk, i, rm_list->ids[i], subflow->local_id, subflow->remote_id);spin_unlock_bh (&msk->pm.lock);mptcp_subflow_shutdown (sk, ssk, how); + +/* the following takes care of updating the subflows counter */ mptcp_close_ssk (sk, ssk, subflow);spin_lock_bh (&msk->pm.lock); removed = true; - msk->pm.subflows--;__MPTCP_INC_STATS (sock_net (sk), rm_type); }__set_bit (rm_list->ids[1 ], msk->pm.id_avail_bitmap);
net/mptcp/protocol.c +6 -0
@@ -2351,6 +2351,12 @@ void mptcp_close_ssk(struct sock *sk, struct sock *ssk, {if (sk->sk_state ==TCP_ESTABLISHED )mptcp_event (MPTCP_EVENT_SUB_CLOSED ,mptcp_sk (sk), ssk,GFP_KERNEL ); + +/* subflow aborted before reaching the fully_established status +* attempt the creation of the next subflow +*/ +mptcp_pm_subflow_check_next (mptcp_sk (sk), ssk, subflow); +__mptcp_close_ssk (sk, ssk, subflow,MPTCP_CF_PUSH ); }
net/mptcp/protocol.h +4 -1
@@ -443,6 +443,7 @@ struct mptcp_subflow_context { stale :1 ;/* unable to snd/rcv data, do not use for xmit */ enum mptcp_data_avail data_avail;u32 remote_nonce; +u32 start_stamp;u64 thmac;u32 local_nonce;u32 remote_token;
@@ -744,7 +745,9 @@ void mptcp_pm_fully_established(struct mptcp_sock *msk, const struct sock *ssk,bool mptcp_pm_allow_new_subflow (struct mptcp_sock *msk);void mptcp_pm_connection_closed (struct mptcp_sock *msk);void mptcp_pm_subflow_established (struct mptcp_sock *msk); -void mptcp_pm_subflow_closed (struct mptcp_sock *msk,u8 id); +bool mptcp_pm_nl_check_work_pending (struct mptcp_sock *msk); +void mptcp_pm_subflow_check_next (struct mptcp_sock *msk,const struct sock *ssk, +const struct mptcp_subflow_context *subflow);void mptcp_pm_add_addr_received (struct mptcp_sock *msk,const struct mptcp_addr_info *addr); void mptcp_pm_add_addr_echoed(struct mptcp_sock *msk,
net/mptcp/subflow.c +4 -1
@@ -1273,7 +1273,8 @@ void __mptcp_error_report(struct sock *sk)static void subflow_error_report (struct sock *ssk) { -struct sock *sk =mptcp_subflow_ctx (ssk)->conn; +struct mptcp_subflow_context *subflow =mptcp_subflow_ctx (ssk); +struct sock *sk = subflow->conn;mptcp_data_lock (sk);if (!sock_owned_by_user (sk))
@@ -1444,6 +1445,7 @@ int __mptcp_subflow_connect(struct sock *sk, const struct mptcp_addr_info *loc, subflow->remote_id = remote_id; subflow->request_join =1 ; subflow->request_bkup = !!(flags &MPTCP_PM_ADDR_FLAG_BACKUP ); + subflow->start_stamp = tcp_jiffies32;mptcp_info2sockaddr (remote, &addr, ssk->sk_family);mptcp_add_pending_subflow (msk, subflow);
@@ -1461,6 +1463,7 @@ int __mptcp_subflow_connect(struct sock *sk, const struct mptcp_addr_info *loc,spin_lock_bh (&msk->join_list_lock);list_del (&subflow->node);spin_unlock_bh (&msk->join_list_lock); +mptcp_pm_subflow_check_next (msk, ssk, subflow);sock_put (mptcp_subflow_tcp_sock (subflow)); failed:
Signature · 1 line
2.33.1
Paolo Abeni
maintainerPATCH v5 3/5 mptcp: keep track of local endpoint still available for each msk
In reply to Paolo AbeniInclude into the path manager status a bitmap tracking the list of local endpoints still available - not yet used - for the relavant mptcp socket.
Keep such map updated at endpoint creation/deleteion time, so that we can easily skip already used endpoint at local address selection time.
The endpoint used by the initial subflow is lazyly accounted at subflow creation time: the usage bitmap is be up2date before endpoint selection and we avoid such unneeded task in some relevant scenarions - e.g. busy servers accepting incoming subflows but not creating any additional ones nor annuncing additional addresses.
Overall this allows for fair local endpoints usage in case of subflow failure.
As a side effect, this patch also enforces that each endpoint is used at most once for each mptcp connection.
---
v4 -> v5:
- track MPC subflow, too
- update self-tests accordingly
v3 -> v4:
- track the available (not yet used) id instead of the used ones
- track both
- cleanup duplication with id_bitmap
--- net/mptcp/pm.c | 1 + net/mptcp/pm_netlink.c | 116 +++++++++++------- net/mptcp/protocol.c | 3 +- net/mptcp/protocol.h | 11 +- .../testing/selftests/net/mptcp/mptcp_join.sh | 5 +- 5 files changed, 86 insertions(+), 50 deletions(-)
net/mptcp/pm.c +1 -0
@@ -376,6 +376,7 @@ void mptcp_pm_data_reset(struct mptcp_sock *msk)WRITE_ONCE (msk->pm.accept_subflow, false);WRITE_ONCE (msk->pm.remote_deny_join_id0, false); msk->pm.status =0 ; +bitmap_fill (msk->pm.id_avail_bitmap,MAX_ADDR_ID +1 );mptcp_pm_nl_data_init (msk); }
net/mptcp/pm_netlink.c +70 -46
225 lines not on this page · raw message
net/mptcp/protocol.c +1 -2
@@ -2508,8 +2508,7 @@ static void mptcp_worker(struct work_struct *work)mptcp_check_fastclose (msk); -if (msk->pm.status) -mptcp_pm_nl_work (msk); +mptcp_pm_nl_work (msk);if (test_and_clear_bit (MPTCP_WORK_EOF , &msk->flags))mptcp_check_for_eof (msk);
net/mptcp/protocol.h +10 -1
@@ -173,16 +173,24 @@ enum mptcp_pm_status {MPTCP_PM_ADD_ADDR_SEND_ACK ,MPTCP_PM_RM_ADDR_RECEIVED ,MPTCP_PM_ESTABLISHED , -MPTCP_PM_ALREADY_ESTABLISHED ,/* persistent status, set after ESTABLISHED event */ MPTCP_PM_SUBFLOW_ESTABLISHED , +MPTCP_PM_ALREADY_ESTABLISHED ,/* persistent status, set after ESTABLISHED event */ +MPTCP_PM_MPC_ENDPOINT_ACCOUNTED /* persistent status, set after MPC local address is +* accounted int id_avail_bitmap +*/ }; +/* Status bits below MPTCP_PM_ALREADY_ESTABLISHED need pm worker actions */ +#define MPTCP_PM_WORK_MASK ((1 << MPTCP_PM_ALREADY_ESTABLISHED) - 1) +enum mptcp_addr_signal_status {MPTCP_ADD_ADDR_SIGNAL ,MPTCP_ADD_ADDR_ECHO ,MPTCP_RM_ADDR_SIGNAL , }; +#define MAX_ADDR_ID 255 +struct mptcp_pm_data {struct mptcp_addr_info local;struct mptcp_addr_info remote;
@@ -201,6 +209,7 @@ struct mptcp_pm_data {u8 local_addr_used;u8 subflows;u8 status; +DECLARE_BITMAP (id_avail_bitmap,MAX_ADDR_ID +1 );struct mptcp_rm_list rm_list_tx;struct mptcp_rm_list rm_list_rx; };
tools/testing/selftests/net/mptcp/mptcp_join.sh +4 -1
@@ -1109,7 +1109,10 @@ signal_address_tests() ip netns exec $ns2 ./pm_nl_ctl add 10.0.3.2 flags signal ip netns exec $ns2 ./pm_nl_ctl add 10.0.4.2 flags signal run_tests $ns1 $ns2 10.0.1.1 - chk_add_nr 4 4 + + # the server will not signal the address teminating + # the MPC subflow + chk_add_nr 3 3 } link_failure_tests()
Signature · 1 line
2.33.1
Paolo Abeni
maintainerPATCH v5 5/5 selftests: mptcp: add tests for subflow creation failure
In reply to Paolo AbeniVerify that, when multiple endpoints are available, subflows creation proceed even when the first additional subflow creation fails - due to packet drop on the relevant link
---
v4 -> v5:
- new test checking for wrong MPC reuse
v3 -> v4:
- use $() where applicable - Matt
- fix typo in help() - Matt
v1 -> v2:
- add missing NF kconfig
- add more test-cases (drop, later subflow creation)
--- tools/testing/selftests/net/mptcp/config | 1 + .../testing/selftests/net/mptcp/mptcp_join.sh | 74 ++++++++++++++++++- 2 files changed, 74 insertions(+), 1 deletion(-)
tools/testing/selftests/net/mptcp/config +1 -0
@@ -18,4 +18,5 @@ CONFIG_NFT_TPROXY=m CONFIG_NFT_SOCKET=m CONFIG_IP_ADVANCED_ROUTER=y CONFIG_IP_MULTIPLE_TABLES=y +CONFIG_IP_NF_TARGET_REJECT=m CONFIG_IPV6_MULTIPLE_TABLES=y
tools/testing/selftests/net/mptcp/mptcp_join.sh +73 -1
108 lines not on this page · raw message
Signature · 1 line
2.33.1
Re: [PATCH mptcp-next v5 4/5] mptcp: do not block subflows creation on errors
In reply to Paolo AbeniQuoted 67 lines · Paolo Abeni
If the MPTCP configuration allows for multiple subflows creation, and the first additional subflows never reach the fully established status - e.g. due to packets drop or reset - the in kernel path manager do not move to the next subflow.
This patch introduces a new PM helper to cope with MPJ subflow creation failure and delay and hook it where appropriate.
Such helper triggers additional subflow creation, as needed and updates the PM subflow counter, if the current one is closing.
Note that we don't need an additional timer to catch timeout and/or long delay in connection completion: we just need to measure the time elapsed since the subflow creation every time we emit an MPJ sub-option.
Signed-off-by Paolo Abeni <[email protected]>---
v3 -> v4:
- fix 'subflows' underflowv2 -> v3:
- fix compile warning (CI)v1 -> v2:
- explicitly hook on subflow close instead of error notification
- fix checkpatch issue (Mat)
- introduce and use a new pm helper to cope with subflow close
- update pm subflow counters on close
---net/mptcp/options.c | 21 ++++++++++++++++++--- net/mptcp/pm.c | 23 +++++++++++++++++++++-- net/mptcp/pm_netlink.c | 20 ++++++++++++-------- net/mptcp/protocol.c | 6 ++++++ net/mptcp/protocol.h | 5 ++++- net/mptcp/subflow.c | 5 ++++- 6 files changed, 65 insertions(+), 15 deletions(-)net/mptcp/options.c +8
-3@@ -1257,10 +1257,10 @@ static u16 mptcp_make_csum(const struct mptcp_ext *mpext)void mptcp_write_options (__be32 *ptr,const struct tcp_sock *tp,struct mptcp_out_options *opts) { -if (unlikely (OPTION_MPTCP_FAIL & opts->suboptions)) { -const struct sock *ssk = (const struct sock *)tp; -struct mptcp_subflow_context *subflow; +const struct sock *ssk = (const struct sock *)tp; +struct mptcp_subflow_context *subflow; +if (unlikely (OPTION_MPTCP_FAIL & opts->suboptions)) { subflow =mptcp_subflow_ctx (ssk); subflow->send_mp_fail =0 ;@@ -1382,6 +1382,21 @@ void mptcp_write_options(__be32 *ptr, const struct tcp_sock *tp,/* MPC is additionally mutually exclusive with MP_PRIO */ goto mp_capable_done; }else if (OPTIONS_MPTCP_MPJ & opts->suboptions) { +if (ssk) { +/* we are still in the MPJ handshake and "a lot" of time passed +* e.g. due to syn retranmissions. We can attempt next +* subflow creation +*/
net/mptcp/options.c void mptcp_write_options(__be32 *ptr, const struct tcp_sock *tp, · line 1399
Quoted 10 lines
net/mptcp/options.c +10
-0+ subflow =mptcp_subflow_ctx (ssk); +if (subflow->start_stamp && +unlikely (after (tcp_jiffies32, subflow->start_stamp +HZ /10 ))) { +mptcp_pm_subflow_check_next (mptcp_sk (subflow->conn), ssk, subflow); + +/* avoid triggering the PM multiple times due to timeout */ + subflow->start_stamp =0 ; + } + } +
tcp_options_write() and the various helpers it calls are (so far) free from side effects and locking, and I don't think we want to introduce those now (__tcp_transmit_skb() seems like a sensitive code path to tinker with, even if it is only in this narrow case for packets with MPJ suboptions).
The msk sk_timer seems well-suited for this? I think we're only using it at disconnect time so there shouldn't be conflicts. What do you think?
Mat
Quoted 171 lines
net/mptcp/options.c +0
-0if (OPTION_MPTCP_MPJ_SYN & opts->suboptions) { *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN, TCPOLEN_MPTCP_MPJ_SYN,net/mptcp/pm.c +21
-2@@ -172,9 +172,28 @@ void mptcp_pm_subflow_established(struct mptcp_sock *msk)spin_unlock_bh (&pm->lock); } -void mptcp_pm_subflow_closed (struct mptcp_sock *msk,u8 id) +void mptcp_pm_subflow_check_next (struct mptcp_sock *msk,const struct sock *ssk, +const struct mptcp_subflow_context *subflow) { -pr_debug ("msk=%p" , msk); +struct mptcp_pm_data *pm = &msk->pm; +bool update_subflows; + + update_subflows = (ssk->sk_state ==TCP_CLOSE ) && + (subflow->request_join || subflow->mp_join); +if (!READ_ONCE (pm->work_pending) && !update_subflows) +return ; + +spin_lock_bh (&pm->lock); +if (update_subflows) + pm->subflows--; + +/* Even if this subflow is not really established, tell the PM to try +* to pick the next one, if possible. +*/ +if (mptcp_pm_nl_check_work_pending (msk)) +mptcp_pm_schedule_work (msk,MPTCP_PM_SUBFLOW_ESTABLISHED ); + +spin_unlock_bh (&pm->lock); } void mptcp_pm_add_addr_received(struct mptcp_sock *msk,net/mptcp/pm_netlink.c +5
-2@@ -251,14 +251,17 @@ unsigned int mptcp_pm_get_local_addr_max(struct mptcp_sock *msk) }EXPORT_SYMBOL_GPL (mptcp_pm_get_local_addr_max); -static void check_work_pending (struct mptcp_sock *msk) +bool mptcp_pm_nl_check_work_pending (struct mptcp_sock *msk) {struct pm_nl_pernet *pernet =net_generic (sock_net ((struct sock *)msk), pm_nl_pernet_id);if (msk->pm.subflows ==mptcp_pm_get_subflows_max (msk) || (find_next_and_bit (pernet->id_bitmap, msk->pm.id_avail_bitmap,MAX_ADDR_ID +1 ,0 ) == -MAX_ADDR_ID +1 )) +MAX_ADDR_ID +1 )) {WRITE_ONCE (msk->pm.work_pending, false); +return false; + } +return true; }struct
mptcp_pm_add_entry*net/mptcp/pm_netlink.c +7
-6@@ -503,12 +506,12 @@ static void mptcp_pm_create_subflow_or_signal_addr(struct mptcp_sock *msk)/* do lazy endpoint usage accounting for the MPC subflows */ if (unlikely (!(msk->pm.status &BIT (MPTCP_PM_MPC_ENDPOINT_ACCOUNTED ))) && msk->first) { -struct mptcp_addr_info local; +struct mptcp_addr_info mpc_addr;int mpc_id; -local_address ((struct sock_common *)msk->first, &local); - mpc_id =lookup_id_by_addr (pernet, &local); -if (mpc_id <0 ) +local_address ((struct sock_common *)msk->first, &mpc_addr); + mpc_id =lookup_id_by_addr (pernet, &mpc_addr); +if (mpc_id >=0 )__clear_bit (mpc_id, msk->pm.id_avail_bitmap); msk->pm.status |=BIT (MPTCP_PM_MPC_ENDPOINT_ACCOUNTED );@@ -553,7 +556,7 @@ static void mptcp_pm_create_subflow_or_signal_addr(struct mptcp_sock *msk)spin_lock_bh (&msk->pm.lock); } } -check_work_pending (msk); +mptcp_pm_nl_check_work_pending (msk); }static void mptcp_pm_nl_fully_established (struct mptcp_sock *msk)@@ -760,11 +763,12 @@ static void mptcp_pm_nl_rm_addr_or_subflow(struct mptcp_sock *msk, i, rm_list->ids[i], subflow->local_id, subflow->remote_id);spin_unlock_bh (&msk->pm.lock);mptcp_subflow_shutdown (sk, ssk, how); + +/* the following takes care of updating the subflows counter */ mptcp_close_ssk (sk, ssk, subflow);spin_lock_bh (&msk->pm.lock); removed = true; - msk->pm.subflows--;__MPTCP_INC_STATS (sock_net (sk), rm_type); }__set_bit (rm_list->ids[1 ], msk->pm.id_avail_bitmap);net/mptcp/protocol.c +6
-0@@ -2351,6 +2351,12 @@ void mptcp_close_ssk(struct sock *sk, struct sock *ssk, {if (sk->sk_state ==TCP_ESTABLISHED )mptcp_event (MPTCP_EVENT_SUB_CLOSED ,mptcp_sk (sk), ssk,GFP_KERNEL ); + +/* subflow aborted before reaching the fully_established status +* attempt the creation of the next subflow +*/ +mptcp_pm_subflow_check_next (mptcp_sk (sk), ssk, subflow); +__mptcp_close_ssk (sk, ssk, subflow,MPTCP_CF_PUSH ); }net/mptcp/protocol.h +4
-1@@ -443,6 +443,7 @@ struct mptcp_subflow_context { stale :1 ;/* unable to snd/rcv data, do not use for xmit */ enum mptcp_data_avail data_avail;u32 remote_nonce; +u32 start_stamp;u64 thmac;u32 local_nonce;u32 remote_token;@@ -744,7 +745,9 @@ void mptcp_pm_fully_established(struct mptcp_sock *msk, const struct sock *ssk,bool mptcp_pm_allow_new_subflow (struct mptcp_sock *msk);void mptcp_pm_connection_closed (struct mptcp_sock *msk);void mptcp_pm_subflow_established (struct mptcp_sock *msk); -void mptcp_pm_subflow_closed (struct mptcp_sock *msk,u8 id); +bool mptcp_pm_nl_check_work_pending (struct mptcp_sock *msk); +void mptcp_pm_subflow_check_next (struct mptcp_sock *msk,const struct sock *ssk, +const struct mptcp_subflow_context *subflow);void mptcp_pm_add_addr_received (struct mptcp_sock *msk,const struct mptcp_addr_info *addr); void mptcp_pm_add_addr_echoed(struct mptcp_sock *msk,net/mptcp/subflow.c +4
-1@@ -1273,7 +1273,8 @@ void __mptcp_error_report(struct sock *sk)static void subflow_error_report (struct sock *ssk) { -struct sock *sk =mptcp_subflow_ctx (ssk)->conn; +struct mptcp_subflow_context *subflow =mptcp_subflow_ctx (ssk); +struct sock *sk = subflow->conn;mptcp_data_lock (sk);if (!sock_owned_by_user (sk))@@ -1444,6 +1445,7 @@ int __mptcp_subflow_connect(struct sock *sk, const struct mptcp_addr_info *loc, subflow->remote_id = remote_id; subflow->request_join =1 ; subflow->request_bkup = !!(flags &MPTCP_PM_ADDR_FLAG_BACKUP ); + subflow->start_stamp = tcp_jiffies32;mptcp_info2sockaddr (remote, &addr, ssk->sk_family);mptcp_add_pending_subflow (msk, subflow);@@ -1461,6 +1463,7 @@ int __mptcp_subflow_connect(struct sock *sk, const struct mptcp_addr_info *loc,spin_lock_bh (&msk->join_list_lock);list_del (&subflow->node);spin_unlock_bh (&msk->join_list_lock); +mptcp_pm_subflow_check_next (msk, ssk, subflow);sock_put (mptcp_subflow_tcp_sock (subflow));
Quoted 6 lines
failed:
Signature · 1 line
2.33.1
--
Mat Martineau
Intel
Re: [PATCH mptcp-next v5 5/5] selftests: mptcp: add tests for subflow creation failure
In reply to Paolo AbeniQuoted 92 lines · Paolo Abeni
Verify that, when multiple endpoints are available, subflows creation proceed even when the first additional subflow creation fails - due to packet drop on the relevant link
Signed-off-by Paolo Abeni <[email protected]>---
v4 -> v5:
- new test checking for wrong MPC reusev3 -> v4:
- use $() where applicable - Matt
- fix typo inhelp()- Mattv1 -> v2:
- add missing NF kconfig
- add more test-cases (drop, later subflow creation)
---tools/testing/selftests/net/mptcp/config| 1 +.../testing/selftests/net/mptcp/mptcp_join.sh| 74 ++++++++++++++++++-2 files changed, 74 insertions(+), 1 deletion(-)tools/testing/selftests/net/mptcp/config +0
-0@@ -18,4 +18,5 @@ CONFIG_NFT_TPROXY=m
CONFIG_NFT_SOCKET=mCONFIG_IP_ADVANCED_ROUTER=yCONFIG_IP_MULTIPLE_TABLES=y
+CONFIG_IP_NF_TARGET_REJECT=mCONFIG_IPV6_MULTIPLE_TABLES=ytools/testing/selftests/net/mptcp/mptcp_join.sh +46
-0@@ -975,6 +975,22 @@ chk_link_usage() fi } +wait_for_tw() +{ + local timeout_ms=$((timeout_poll * 1000)) + local time=0 + local ns=$1 + + while [ $time -lt $timeout_ms ]; do + local cnt=$(ip netns exec $ns ss -t state time-wait |wc -l) + + [ "$cnt" = 1 ] && return 1 + time=$((time + 100)) + sleep 0.1 + done + return 1 +} + subflows_tests() { reset@@ -1032,6 +1048,57 @@ subflows_tests() chk_join_nr "single subflow, dev" 1 1 1 } +subflows_error_tests() +{ + # If a single subflow is configured, and matches the MPC src + # address, no additional subflow should be created + reset + ip netns exec $ns1 ./pm_nl_ctl limits 0 1 + ip netns exec $ns2 ./pm_nl_ctl limits 0 1 + ip netns exec $ns2 ./pm_nl_ctl add 10.0.1.2 flags subflow + run_tests $ns1 $ns2 10.0.1.1 0 0 0 slow + chk_join_nr "no MPC address reuse with single endpoint" 0 0 0 + + # multiple subflows, with subflow creation error + reset + ip netns exec $ns1 ./pm_nl_ctl limits 0 2 + ip netns exec $ns2 ./pm_nl_ctl limits 0 2 + ip netns exec $ns2 ./pm_nl_ctl add 10.0.3.2 flags subflow + ip netns exec $ns2 ./pm_nl_ctl add 10.0.2.2 flags subflow + ip netns exec $ns1 iptables -A INPUT -s 10.0.3.2 -p tcp -j REJECT + run_tests $ns1 $ns2 10.0.1.1 0 0 0 slow + chk_join_nr "multiple subflows, with failing subflow" 1 1 1 + + # multiple subflows, with subflow timeout on MPJ + reset + ip netns exec $ns1 ./pm_nl_ctl limits 0 2 + ip netns exec $ns2 ./pm_nl_ctl limits 0 2 + ip netns exec $ns2 ./pm_nl_ctl add 10.0.3.2 flags subflow + ip netns exec $ns2 ./pm_nl_ctl add 10.0.2.2 flags subflow + ip netns exec $ns1 iptables -A INPUT -s 10.0.3.2 -p tcp -j DROP + run_tests $ns1 $ns2 10.0.1.1 0 0 0 slow + chk_join_nr "multiple subflows, with subflow timeout" 1 1 1
tools/testing/selftests/net/mptcp/mptcp_join.sh subflows_tests() · line 1090
Quoted 10 lines
tools/testing/selftests/net/mptcp/mptcp_join.sh +10
-0+ + # multiple subflows, check that the endpoint corresponding to + # closed subflow (due to reset) is not reused if additional + # subflows are added later + reset + ip netns exec $ns1 ./pm_nl_ctl limits 0 1 + ip netns exec $ns2 ./pm_nl_ctl limits 0 1 + ip netns exec $ns2 ./pm_nl_ctl add 10.0.3.2 flags subflow + ip netns exec $ns1 iptables -A INPUT -s 10.0.3.2 -p tcp -j REJECT + run_tests $ns1 $ns2 10.0.1.1 0 0 0 slow &
Still seeing the problem I reported in v4 where the test results are not correctly numbered. I think $TEST_COUNT needs to be bumped here so the next results are numbered correctly.
Mat
Quoted 43 lines
tools/testing/selftests/net/mptcp/mptcp_join.sh +13
-0+ + # mpj subflow will be in TW after the reset + wait_for_tw $ns2 + ip netns exec $ns2 ./pm_nl_ctl add 10.0.2.2 flags subflow + wait + + # additional subflow could be created only if the PM select + # the later endpoint, skipping the already used one + chk_join_nr "multiple subflows, fair usage on close" 1 1 1 +} + signal_address_tests() { # add_address, unused@@ -1843,6 +1910,7 @@ fullmesh_tests() all_tests() { subflows_tests + subflows_error_tests signal_address_tests link_failure_tests add_addr_timeout_tests@@ -1862,6 +1930,7 @@ usage() { echo "mptcp_join usage:" echo " -f subflows_tests" + echo " -e subflows_error_tests" echo " -s signal_address_tests" echo " -l link_failure_tests" echo " -t add_addr_timeout_tests"@@ -1910,11 +1979,14 @@ if [ $do_all_tests -eq 1 ]; then exit $retfi
tools/testing/selftests/net/mptcp/mptcp_join.sh +1
-1-while getopts 'fsltra64bpkdmchCS' opt; do +while getopts 'fesltra64bpkdmchCS' opt; docase $opt in f) subflows_tests ;;tools/testing/selftests/net/mptcp/mptcp_join.sh +3
-0+ e) + subflows_error_tests + ;;
Quoted 8 lines
s) signal_address_tests ;;Signature · 1 line
2.33.1
--
Mat Martineau
Intel
Paolo Abeni
maintainerRe: [PATCH mptcp-next v5 5/5] selftests: mptcp: add tests for subflow creation failure
Quoted 100 lines · Mat Martineau
Quoted 98 lines · Paolo Abeni
Verify that, when multiple endpoints are available, subflows creation proceed even when the first additional subflow creation fails - due to packet drop on the relevant link
Signed-off-by Paolo Abeni <[email protected]>---
v4 -> v5:
- new test checking for wrong MPC reusev3 -> v4:
- use $() where applicable - Matt
- fix typo inhelp()- Mattv1 -> v2:
- add missing NF kconfig
- add more test-cases (drop, later subflow creation)
---tools/testing/selftests/net/mptcp/config| 1 +.../testing/selftests/net/mptcp/mptcp_join.sh| 74 ++++++++++++++++++-2 files changed, 74 insertions(+), 1 deletion(-)tools/testing/selftests/net/mptcp/config +0
-0@@ -18,4 +18,5 @@ CONFIG_NFT_TPROXY=m
CONFIG_NFT_SOCKET=mCONFIG_IP_ADVANCED_ROUTER=yCONFIG_IP_MULTIPLE_TABLES=y
+CONFIG_IP_NF_TARGET_REJECT=mCONFIG_IPV6_MULTIPLE_TABLES=ytools/testing/selftests/net/mptcp/mptcp_join.sh +52
-0@@ -975,6 +975,22 @@ chk_link_usage() fi } +wait_for_tw() +{ + local timeout_ms=$((timeout_poll * 1000)) + local time=0 + local ns=$1 + + while [ $time -lt $timeout_ms ]; do + local cnt=$(ip netns exec $ns ss -t state time-wait |wc -l) + + [ "$cnt" = 1 ] && return 1 + time=$((time + 100)) + sleep 0.1 + done + return 1 +} + subflows_tests() { reset@@ -1032,6 +1048,57 @@ subflows_tests() chk_join_nr "single subflow, dev" 1 1 1 } +subflows_error_tests() +{ + # If a single subflow is configured, and matches the MPC src + # address, no additional subflow should be created + reset + ip netns exec $ns1 ./pm_nl_ctl limits 0 1 + ip netns exec $ns2 ./pm_nl_ctl limits 0 1 + ip netns exec $ns2 ./pm_nl_ctl add 10.0.1.2 flags subflow + run_tests $ns1 $ns2 10.0.1.1 0 0 0 slow + chk_join_nr "no MPC address reuse with single endpoint" 0 0 0 + + # multiple subflows, with subflow creation error + reset + ip netns exec $ns1 ./pm_nl_ctl limits 0 2 + ip netns exec $ns2 ./pm_nl_ctl limits 0 2 + ip netns exec $ns2 ./pm_nl_ctl add 10.0.3.2 flags subflow + ip netns exec $ns2 ./pm_nl_ctl add 10.0.2.2 flags subflow + ip netns exec $ns1 iptables -A INPUT -s 10.0.3.2 -p tcp -j REJECT + run_tests $ns1 $ns2 10.0.1.1 0 0 0 slow + chk_join_nr "multiple subflows, with failing subflow" 1 1 1 + + # multiple subflows, with subflow timeout on MPJ + reset + ip netns exec $ns1 ./pm_nl_ctl limits 0 2 + ip netns exec $ns2 ./pm_nl_ctl limits 0 2 + ip netns exec $ns2 ./pm_nl_ctl add 10.0.3.2 flags subflow + ip netns exec $ns2 ./pm_nl_ctl add 10.0.2.2 flags subflow + ip netns exec $ns1 iptables -A INPUT -s 10.0.3.2 -p tcp -j DROP + run_tests $ns1 $ns2 10.0.1.1 0 0 0 slow + chk_join_nr "multiple subflows, with subflow timeout" 1 1 1 + + # multiple subflows, check that the endpoint corresponding to + # closed subflow (due to reset) is not reused if additional + # subflows are added later + reset + ip netns exec $ns1 ./pm_nl_ctl limits 0 1
Quoted 8 lines
tools/testing/selftests/net/mptcp/mptcp_join.sh subflows_tests() · line 1090
Quoted 4 lines
tools/testing/selftests/net/mptcp/mptcp_join.sh +4
-0+ ip netns exec $ns2 ./pm_nl_ctl limits 0 1 + ip netns exec $ns2 ./pm_nl_ctl add 10.0.3.2 flags subflow + ip netns exec $ns1 iptables -A INPUT -s 10.0.3.2 -p tcp -j REJECT + run_tests $ns1 $ns2 10.0.1.1 0 0 0 slow &Still seeing the problem I reported in v4 where the test results are not correctly numbered. I think $TEST_COUNT needs to be bumped here so the next results are numbered correctly.
Oh, sorry, I missed that part. I'll cover in the next iteration.
Thanks,
Paolo
Paolo Abeni
maintainerRe: [PATCH mptcp-next v5 4/5] mptcp: do not block subflows creation on errors
In reply to Mat MartineauQuoted 86 lines · Mat Martineau
Quoted 67 lines · Paolo Abeni
If the MPTCP configuration allows for multiple subflows creation, and the first additional subflows never reach the fully established status - e.g. due to packets drop or reset - the in kernel path manager do not move to the next subflow.
This patch introduces a new PM helper to cope with MPJ subflow creation failure and delay and hook it where appropriate.
Such helper triggers additional subflow creation, as needed and updates the PM subflow counter, if the current one is closing.
Note that we don't need an additional timer to catch timeout and/or long delay in connection completion: we just need to measure the time elapsed since the subflow creation every time we emit an MPJ sub-option.
Signed-off-by Paolo Abeni <[email protected]>---
v3 -> v4:
- fix 'subflows' underflowv2 -> v3:
- fix compile warning (CI)v1 -> v2:
- explicitly hook on subflow close instead of error notification
- fix checkpatch issue (Mat)
- introduce and use a new pm helper to cope with subflow close
- update pm subflow counters on close
---net/mptcp/options.c | 21 ++++++++++++++++++--- net/mptcp/pm.c | 23 +++++++++++++++++++++-- net/mptcp/pm_netlink.c | 20 ++++++++++++-------- net/mptcp/protocol.c | 6 ++++++ net/mptcp/protocol.h | 5 ++++- net/mptcp/subflow.c | 5 ++++- 6 files changed, 65 insertions(+), 15 deletions(-)net/mptcp/options.c +8
-3@@ -1257,10 +1257,10 @@ static u16 mptcp_make_csum(const struct mptcp_ext *mpext)void mptcp_write_options (__be32 *ptr,const struct tcp_sock *tp,struct mptcp_out_options *opts) { -if (unlikely (OPTION_MPTCP_FAIL & opts->suboptions)) { -const struct sock *ssk = (const struct sock *)tp; -struct mptcp_subflow_context *subflow; +const struct sock *ssk = (const struct sock *)tp; +struct mptcp_subflow_context *subflow; +if (unlikely (OPTION_MPTCP_FAIL & opts->suboptions)) { subflow =mptcp_subflow_ctx (ssk); subflow->send_mp_fail =0 ;@@ -1382,6 +1382,21 @@ void mptcp_write_options(__be32 *ptr, const struct tcp_sock *tp,/* MPC is additionally mutually exclusive with MP_PRIO */ goto mp_capable_done; }else if (OPTIONS_MPTCP_MPJ & opts->suboptions) { +if (ssk) { +/* we are still in the MPJ handshake and "a lot" of time passed +* e.g. due to syn retranmissions. We can attempt next +* subflow creation +*/ net/mptcp/options.c void mptcp_write_options(__be32 *ptr, const struct tcp_sock *tp, · line 1399
Quoted 10 lines
net/mptcp/options.c +10
-0+ subflow =mptcp_subflow_ctx (ssk); +if (subflow->start_stamp && +unlikely (after (tcp_jiffies32, subflow->start_stamp +HZ /10 ))) { +mptcp_pm_subflow_check_next (mptcp_sk (subflow->conn), ssk, subflow); + +/* avoid triggering the PM multiple times due to timeout */ + subflow->start_stamp =0 ; + } + } +
tcp_options_write()and the various helpers it calls are (so far) free from side effects and locking, and I don't think we want to introduce those now (__tcp_transmit_skb()seems like a sensitive code path to tinker with, even if it is only in this narrow case for packets with MPJ suboptions).
Quoted 2 lines
The msk
sk_timerseems well-suited for this? I think we're only using it at disconnect time so there shouldn't be conflicts. What do you think?
I really prefer to avoid using another timer.
I'll try a different approach: creating all the subflows at once - as per Matt suggestion - so that we will not need such hook at all.
Cheers,
Paolo