1 // SPDX-License-Identifier: GPL-2.0
4 * Copyright (c) 2017 - 2019, Intel Corporation.
7 #define pr_fmt(fmt) "MPTCP: " fmt
9 #include <linux/kernel.h>
10 #include <linux/module.h>
11 #include <linux/netdevice.h>
12 #include <crypto/sha2.h>
13 #include <crypto/utils.h>
15 #include <net/inet_common.h>
16 #include <net/inet_hashtables.h>
17 #include <net/protocol.h>
19 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
20 #include <net/ip6_route.h>
21 #include <net/transp_v6.h>
23 #include <net/mptcp.h>
24 #include <uapi/linux/mptcp.h>
28 #include <trace/events/mptcp.h>
29 #include <trace/events/sock.h>
31 static void mptcp_subflow_ops_undo_override(struct sock *ssk);
33 static void SUBFLOW_REQ_INC_STATS(struct request_sock *req,
34 enum linux_mptcp_mib_field field)
36 MPTCP_INC_STATS(sock_net(req_to_sk(req)), field);
39 static void subflow_req_destructor(struct request_sock *req)
41 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
43 pr_debug("subflow_req=%p", subflow_req);
46 sock_put((struct sock *)subflow_req->msk);
48 mptcp_token_destroy_request(req);
51 static void subflow_generate_hmac(u64 key1, u64 key2, u32 nonce1, u32 nonce2,
56 put_unaligned_be32(nonce1, &msg[0]);
57 put_unaligned_be32(nonce2, &msg[4]);
59 mptcp_crypto_hmac_sha(key1, key2, msg, 8, hmac);
62 static bool mptcp_can_accept_new_subflow(const struct mptcp_sock *msk)
64 return mptcp_is_fully_established((void *)msk) &&
65 ((mptcp_pm_is_userspace(msk) &&
66 mptcp_userspace_pm_active(msk)) ||
67 READ_ONCE(msk->pm.accept_subflow));
70 /* validate received token and create truncated hmac and nonce for SYN-ACK */
71 static void subflow_req_create_thmac(struct mptcp_subflow_request_sock *subflow_req)
73 struct mptcp_sock *msk = subflow_req->msk;
74 u8 hmac[SHA256_DIGEST_SIZE];
76 get_random_bytes(&subflow_req->local_nonce, sizeof(u32));
78 subflow_generate_hmac(msk->local_key, msk->remote_key,
79 subflow_req->local_nonce,
80 subflow_req->remote_nonce, hmac);
82 subflow_req->thmac = get_unaligned_be64(hmac);
85 static struct mptcp_sock *subflow_token_join_request(struct request_sock *req)
87 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
88 struct mptcp_sock *msk;
91 msk = mptcp_token_get_sock(sock_net(req_to_sk(req)), subflow_req->token);
93 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINNOTOKEN);
97 local_id = mptcp_pm_get_local_id(msk, (struct sock_common *)req);
99 sock_put((struct sock *)msk);
102 subflow_req->local_id = local_id;
107 static void subflow_init_req(struct request_sock *req, const struct sock *sk_listener)
109 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
111 subflow_req->mp_capable = 0;
112 subflow_req->mp_join = 0;
113 subflow_req->csum_reqd = mptcp_is_checksum_enabled(sock_net(sk_listener));
114 subflow_req->allow_join_id0 = mptcp_allow_join_id0(sock_net(sk_listener));
115 subflow_req->msk = NULL;
116 mptcp_token_init_request(req);
119 static bool subflow_use_different_sport(struct mptcp_sock *msk, const struct sock *sk)
121 return inet_sk(sk)->inet_sport != inet_sk((struct sock *)msk)->inet_sport;
124 static void subflow_add_reset_reason(struct sk_buff *skb, u8 reason)
126 struct mptcp_ext *mpext = skb_ext_add(skb, SKB_EXT_MPTCP);
129 memset(mpext, 0, sizeof(*mpext));
130 mpext->reset_reason = reason;
134 /* Init mptcp request socket.
136 * Returns an error code if a JOIN has failed and a TCP reset
139 static int subflow_check_req(struct request_sock *req,
140 const struct sock *sk_listener,
143 struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk_listener);
144 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
145 struct mptcp_options_received mp_opt;
146 bool opt_mp_capable, opt_mp_join;
148 pr_debug("subflow_req=%p, listener=%p", subflow_req, listener);
150 #ifdef CONFIG_TCP_MD5SIG
151 /* no MPTCP if MD5SIG is enabled on this socket or we may run out of
154 if (rcu_access_pointer(tcp_sk(sk_listener)->md5sig_info))
158 mptcp_get_options(skb, &mp_opt);
160 opt_mp_capable = !!(mp_opt.suboptions & OPTIONS_MPTCP_MPC);
161 opt_mp_join = !!(mp_opt.suboptions & OPTIONS_MPTCP_MPJ);
162 if (opt_mp_capable) {
163 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MPCAPABLEPASSIVE);
167 } else if (opt_mp_join) {
168 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINSYNRX);
171 if (opt_mp_capable && listener->request_mptcp) {
172 int err, retries = MPTCP_TOKEN_MAX_RETRIES;
174 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq;
177 get_random_bytes(&subflow_req->local_key, sizeof(subflow_req->local_key));
178 } while (subflow_req->local_key == 0);
180 if (unlikely(req->syncookie)) {
181 mptcp_crypto_key_sha(subflow_req->local_key,
184 if (mptcp_token_exists(subflow_req->token)) {
187 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_TOKENFALLBACKINIT);
189 subflow_req->mp_capable = 1;
194 err = mptcp_token_new_request(req);
196 subflow_req->mp_capable = 1;
197 else if (retries-- > 0)
200 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_TOKENFALLBACKINIT);
202 } else if (opt_mp_join && listener->request_mptcp) {
203 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq;
204 subflow_req->mp_join = 1;
205 subflow_req->backup = mp_opt.backup;
206 subflow_req->remote_id = mp_opt.join_id;
207 subflow_req->token = mp_opt.token;
208 subflow_req->remote_nonce = mp_opt.nonce;
209 subflow_req->msk = subflow_token_join_request(req);
211 /* Can't fall back to TCP in this case. */
212 if (!subflow_req->msk) {
213 subflow_add_reset_reason(skb, MPTCP_RST_EMPTCP);
217 if (subflow_use_different_sport(subflow_req->msk, sk_listener)) {
218 pr_debug("syn inet_sport=%d %d",
219 ntohs(inet_sk(sk_listener)->inet_sport),
220 ntohs(inet_sk((struct sock *)subflow_req->msk)->inet_sport));
221 if (!mptcp_pm_sport_in_anno_list(subflow_req->msk, sk_listener)) {
222 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MISMATCHPORTSYNRX);
225 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINPORTSYNRX);
228 subflow_req_create_thmac(subflow_req);
230 if (unlikely(req->syncookie)) {
231 if (mptcp_can_accept_new_subflow(subflow_req->msk))
232 subflow_init_req_cookie_join_save(subflow_req, skb);
237 pr_debug("token=%u, remote_nonce=%u msk=%p", subflow_req->token,
238 subflow_req->remote_nonce, subflow_req->msk);
244 int mptcp_subflow_init_cookie_req(struct request_sock *req,
245 const struct sock *sk_listener,
248 struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk_listener);
249 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
250 struct mptcp_options_received mp_opt;
251 bool opt_mp_capable, opt_mp_join;
254 subflow_init_req(req, sk_listener);
255 mptcp_get_options(skb, &mp_opt);
257 opt_mp_capable = !!(mp_opt.suboptions & OPTIONS_MPTCP_MPC);
258 opt_mp_join = !!(mp_opt.suboptions & OPTIONS_MPTCP_MPJ);
259 if (opt_mp_capable && opt_mp_join)
262 if (opt_mp_capable && listener->request_mptcp) {
263 if (mp_opt.sndr_key == 0)
266 subflow_req->local_key = mp_opt.rcvr_key;
267 err = mptcp_token_new_request(req);
271 subflow_req->mp_capable = 1;
272 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq - 1;
273 } else if (opt_mp_join && listener->request_mptcp) {
274 if (!mptcp_token_join_cookie_init_state(subflow_req, skb))
277 subflow_req->mp_join = 1;
278 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq - 1;
283 EXPORT_SYMBOL_GPL(mptcp_subflow_init_cookie_req);
285 static struct dst_entry *subflow_v4_route_req(const struct sock *sk,
288 struct request_sock *req)
290 struct dst_entry *dst;
293 tcp_rsk(req)->is_mptcp = 1;
294 subflow_init_req(req, sk);
296 dst = tcp_request_sock_ipv4_ops.route_req(sk, skb, fl, req);
300 err = subflow_check_req(req, sk, skb);
306 tcp_request_sock_ops.send_reset(sk, skb);
310 static void subflow_prep_synack(const struct sock *sk, struct request_sock *req,
311 struct tcp_fastopen_cookie *foc,
312 enum tcp_synack_type synack_type)
314 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
315 struct inet_request_sock *ireq = inet_rsk(req);
317 /* clear tstamp_ok, as needed depending on cookie */
318 if (foc && foc->len > -1)
321 if (synack_type == TCP_SYNACK_FASTOPEN)
322 mptcp_fastopen_subflow_synack_set_params(subflow, req);
325 static int subflow_v4_send_synack(const struct sock *sk, struct dst_entry *dst,
327 struct request_sock *req,
328 struct tcp_fastopen_cookie *foc,
329 enum tcp_synack_type synack_type,
330 struct sk_buff *syn_skb)
332 subflow_prep_synack(sk, req, foc, synack_type);
334 return tcp_request_sock_ipv4_ops.send_synack(sk, dst, fl, req, foc,
335 synack_type, syn_skb);
338 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
339 static int subflow_v6_send_synack(const struct sock *sk, struct dst_entry *dst,
341 struct request_sock *req,
342 struct tcp_fastopen_cookie *foc,
343 enum tcp_synack_type synack_type,
344 struct sk_buff *syn_skb)
346 subflow_prep_synack(sk, req, foc, synack_type);
348 return tcp_request_sock_ipv6_ops.send_synack(sk, dst, fl, req, foc,
349 synack_type, syn_skb);
352 static struct dst_entry *subflow_v6_route_req(const struct sock *sk,
355 struct request_sock *req)
357 struct dst_entry *dst;
360 tcp_rsk(req)->is_mptcp = 1;
361 subflow_init_req(req, sk);
363 dst = tcp_request_sock_ipv6_ops.route_req(sk, skb, fl, req);
367 err = subflow_check_req(req, sk, skb);
373 tcp6_request_sock_ops.send_reset(sk, skb);
378 /* validate received truncated hmac and create hmac for third ACK */
379 static bool subflow_thmac_valid(struct mptcp_subflow_context *subflow)
381 u8 hmac[SHA256_DIGEST_SIZE];
384 subflow_generate_hmac(subflow->remote_key, subflow->local_key,
385 subflow->remote_nonce, subflow->local_nonce,
388 thmac = get_unaligned_be64(hmac);
389 pr_debug("subflow=%p, token=%u, thmac=%llu, subflow->thmac=%llu\n",
390 subflow, subflow->token, thmac, subflow->thmac);
392 return thmac == subflow->thmac;
395 void mptcp_subflow_reset(struct sock *ssk)
397 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
398 struct sock *sk = subflow->conn;
400 /* mptcp_mp_fail_no_response() can reach here on an already closed
403 if (ssk->sk_state == TCP_CLOSE)
406 /* must hold: tcp_done() could drop last reference on parent */
409 tcp_send_active_reset(ssk, GFP_ATOMIC);
411 if (!test_and_set_bit(MPTCP_WORK_CLOSE_SUBFLOW, &mptcp_sk(sk)->flags))
412 mptcp_schedule_work(sk);
417 static bool subflow_use_different_dport(struct mptcp_sock *msk, const struct sock *sk)
419 return inet_sk(sk)->inet_dport != inet_sk((struct sock *)msk)->inet_dport;
422 void __mptcp_sync_state(struct sock *sk, int state)
424 struct mptcp_sock *msk = mptcp_sk(sk);
426 __mptcp_propagate_sndbuf(sk, msk->first);
427 if (sk->sk_state == TCP_SYN_SENT) {
428 inet_sk_state_store(sk, state);
429 sk->sk_state_change(sk);
433 static void mptcp_propagate_state(struct sock *sk, struct sock *ssk)
435 struct mptcp_sock *msk = mptcp_sk(sk);
438 if (!sock_owned_by_user(sk)) {
439 __mptcp_sync_state(sk, ssk->sk_state);
441 msk->pending_state = ssk->sk_state;
442 __set_bit(MPTCP_SYNC_STATE, &msk->cb_flags);
444 mptcp_data_unlock(sk);
447 static void subflow_set_remote_key(struct mptcp_sock *msk,
448 struct mptcp_subflow_context *subflow,
449 const struct mptcp_options_received *mp_opt)
451 /* active MPC subflow will reach here multiple times:
452 * at subflow_finish_connect() time and at 4th ack time
454 if (subflow->remote_key_valid)
457 subflow->remote_key_valid = 1;
458 subflow->remote_key = mp_opt->sndr_key;
459 mptcp_crypto_key_sha(subflow->remote_key, NULL, &subflow->iasn);
462 WRITE_ONCE(msk->remote_key, subflow->remote_key);
463 WRITE_ONCE(msk->ack_seq, subflow->iasn);
464 WRITE_ONCE(msk->can_ack, true);
465 atomic64_set(&msk->rcv_wnd_sent, subflow->iasn);
468 static void subflow_finish_connect(struct sock *sk, const struct sk_buff *skb)
470 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
471 struct mptcp_options_received mp_opt;
472 struct sock *parent = subflow->conn;
473 struct mptcp_sock *msk;
475 subflow->icsk_af_ops->sk_rx_dst_set(sk, skb);
477 /* be sure no special action on any packet other than syn-ack */
478 if (subflow->conn_finished)
481 msk = mptcp_sk(parent);
482 subflow->rel_write_seq = 1;
483 subflow->conn_finished = 1;
484 subflow->ssn_offset = TCP_SKB_CB(skb)->seq;
485 pr_debug("subflow=%p synack seq=%x", subflow, subflow->ssn_offset);
487 mptcp_get_options(skb, &mp_opt);
488 if (subflow->request_mptcp) {
489 if (!(mp_opt.suboptions & OPTIONS_MPTCP_MPC)) {
490 MPTCP_INC_STATS(sock_net(sk),
491 MPTCP_MIB_MPCAPABLEACTIVEFALLBACK);
492 mptcp_do_fallback(sk);
497 if (mp_opt.suboptions & OPTION_MPTCP_CSUMREQD)
498 WRITE_ONCE(msk->csum_enabled, true);
499 if (mp_opt.deny_join_id0)
500 WRITE_ONCE(msk->pm.remote_deny_join_id0, true);
501 subflow->mp_capable = 1;
502 subflow_set_remote_key(msk, subflow, &mp_opt);
503 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPCAPABLEACTIVEACK);
504 mptcp_finish_connect(sk);
505 mptcp_propagate_state(parent, sk);
506 } else if (subflow->request_join) {
507 u8 hmac[SHA256_DIGEST_SIZE];
509 if (!(mp_opt.suboptions & OPTIONS_MPTCP_MPJ)) {
510 subflow->reset_reason = MPTCP_RST_EMPTCP;
514 subflow->backup = mp_opt.backup;
515 subflow->thmac = mp_opt.thmac;
516 subflow->remote_nonce = mp_opt.nonce;
517 subflow->remote_id = mp_opt.join_id;
518 pr_debug("subflow=%p, thmac=%llu, remote_nonce=%u backup=%d",
519 subflow, subflow->thmac, subflow->remote_nonce,
522 if (!subflow_thmac_valid(subflow)) {
523 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINACKMAC);
524 subflow->reset_reason = MPTCP_RST_EMPTCP;
528 if (!mptcp_finish_join(sk))
531 subflow_generate_hmac(subflow->local_key, subflow->remote_key,
532 subflow->local_nonce,
533 subflow->remote_nonce,
535 memcpy(subflow->hmac, hmac, MPTCPOPT_HMAC_LEN);
537 subflow->mp_join = 1;
538 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINSYNACKRX);
540 if (subflow_use_different_dport(msk, sk)) {
541 pr_debug("synack inet_dport=%d %d",
542 ntohs(inet_sk(sk)->inet_dport),
543 ntohs(inet_sk(parent)->inet_dport));
544 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINPORTSYNACKRX);
546 } else if (mptcp_check_fallback(sk)) {
548 mptcp_rcv_space_init(msk, sk);
549 mptcp_propagate_state(parent, sk);
554 subflow->reset_transient = 0;
555 mptcp_subflow_reset(sk);
558 static void subflow_set_local_id(struct mptcp_subflow_context *subflow, int local_id)
560 subflow->local_id = local_id;
561 subflow->local_id_valid = 1;
564 static int subflow_chk_local_id(struct sock *sk)
566 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
567 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
570 if (likely(subflow->local_id_valid))
573 err = mptcp_pm_get_local_id(msk, (struct sock_common *)sk);
577 subflow_set_local_id(subflow, err);
581 static int subflow_rebuild_header(struct sock *sk)
583 int err = subflow_chk_local_id(sk);
585 if (unlikely(err < 0))
588 return inet_sk_rebuild_header(sk);
591 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
592 static int subflow_v6_rebuild_header(struct sock *sk)
594 int err = subflow_chk_local_id(sk);
596 if (unlikely(err < 0))
599 return inet6_sk_rebuild_header(sk);
603 static struct request_sock_ops mptcp_subflow_v4_request_sock_ops __ro_after_init;
604 static struct tcp_request_sock_ops subflow_request_sock_ipv4_ops __ro_after_init;
606 static int subflow_v4_conn_request(struct sock *sk, struct sk_buff *skb)
608 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
610 pr_debug("subflow=%p", subflow);
612 /* Never answer to SYNs sent to broadcast or multicast */
613 if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
616 return tcp_conn_request(&mptcp_subflow_v4_request_sock_ops,
617 &subflow_request_sock_ipv4_ops,
624 static void subflow_v4_req_destructor(struct request_sock *req)
626 subflow_req_destructor(req);
627 tcp_request_sock_ops.destructor(req);
630 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
631 static struct request_sock_ops mptcp_subflow_v6_request_sock_ops __ro_after_init;
632 static struct tcp_request_sock_ops subflow_request_sock_ipv6_ops __ro_after_init;
633 static struct inet_connection_sock_af_ops subflow_v6_specific __ro_after_init;
634 static struct inet_connection_sock_af_ops subflow_v6m_specific __ro_after_init;
635 static struct proto tcpv6_prot_override __ro_after_init;
637 static int subflow_v6_conn_request(struct sock *sk, struct sk_buff *skb)
639 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
641 pr_debug("subflow=%p", subflow);
643 if (skb->protocol == htons(ETH_P_IP))
644 return subflow_v4_conn_request(sk, skb);
646 if (!ipv6_unicast_destination(skb))
649 if (ipv6_addr_v4mapped(&ipv6_hdr(skb)->saddr)) {
650 __IP6_INC_STATS(sock_net(sk), NULL, IPSTATS_MIB_INHDRERRORS);
654 return tcp_conn_request(&mptcp_subflow_v6_request_sock_ops,
655 &subflow_request_sock_ipv6_ops, sk, skb);
659 return 0; /* don't send reset */
662 static void subflow_v6_req_destructor(struct request_sock *req)
664 subflow_req_destructor(req);
665 tcp6_request_sock_ops.destructor(req);
669 struct request_sock *mptcp_subflow_reqsk_alloc(const struct request_sock_ops *ops,
670 struct sock *sk_listener,
671 bool attach_listener)
673 if (ops->family == AF_INET)
674 ops = &mptcp_subflow_v4_request_sock_ops;
675 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
676 else if (ops->family == AF_INET6)
677 ops = &mptcp_subflow_v6_request_sock_ops;
680 return inet_reqsk_alloc(ops, sk_listener, attach_listener);
682 EXPORT_SYMBOL(mptcp_subflow_reqsk_alloc);
684 /* validate hmac received in third ACK */
685 static bool subflow_hmac_valid(const struct request_sock *req,
686 const struct mptcp_options_received *mp_opt)
688 const struct mptcp_subflow_request_sock *subflow_req;
689 u8 hmac[SHA256_DIGEST_SIZE];
690 struct mptcp_sock *msk;
692 subflow_req = mptcp_subflow_rsk(req);
693 msk = subflow_req->msk;
697 subflow_generate_hmac(msk->remote_key, msk->local_key,
698 subflow_req->remote_nonce,
699 subflow_req->local_nonce, hmac);
701 return !crypto_memneq(hmac, mp_opt->hmac, MPTCPOPT_HMAC_LEN);
704 static void subflow_ulp_fallback(struct sock *sk,
705 struct mptcp_subflow_context *old_ctx)
707 struct inet_connection_sock *icsk = inet_csk(sk);
709 mptcp_subflow_tcp_fallback(sk, old_ctx);
710 icsk->icsk_ulp_ops = NULL;
711 rcu_assign_pointer(icsk->icsk_ulp_data, NULL);
712 tcp_sk(sk)->is_mptcp = 0;
714 mptcp_subflow_ops_undo_override(sk);
717 void mptcp_subflow_drop_ctx(struct sock *ssk)
719 struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(ssk);
724 list_del(&mptcp_subflow_ctx(ssk)->node);
725 if (inet_csk(ssk)->icsk_ulp_ops) {
726 subflow_ulp_fallback(ssk, ctx);
734 void mptcp_subflow_fully_established(struct mptcp_subflow_context *subflow,
735 const struct mptcp_options_received *mp_opt)
737 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
739 subflow_set_remote_key(msk, subflow, mp_opt);
740 subflow->fully_established = 1;
741 WRITE_ONCE(msk->fully_established, true);
743 if (subflow->is_mptfo)
744 mptcp_fastopen_gen_msk_ackseq(msk, subflow, mp_opt);
747 static struct sock *subflow_syn_recv_sock(const struct sock *sk,
749 struct request_sock *req,
750 struct dst_entry *dst,
751 struct request_sock *req_unhash,
754 struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk);
755 struct mptcp_subflow_request_sock *subflow_req;
756 struct mptcp_options_received mp_opt;
757 bool fallback, fallback_is_fatal;
758 struct mptcp_sock *owner;
761 pr_debug("listener=%p, req=%p, conn=%p", listener, req, listener->conn);
763 /* After child creation we must look for MPC even when options
766 mp_opt.suboptions = 0;
768 /* hopefully temporary handling for MP_JOIN+syncookie */
769 subflow_req = mptcp_subflow_rsk(req);
770 fallback_is_fatal = tcp_rsk(req)->is_mptcp && subflow_req->mp_join;
771 fallback = !tcp_rsk(req)->is_mptcp;
775 /* if the sk is MP_CAPABLE, we try to fetch the client key */
776 if (subflow_req->mp_capable) {
777 /* we can receive and accept an in-window, out-of-order pkt,
778 * which may not carry the MP_CAPABLE opt even on mptcp enabled
779 * paths: always try to extract the peer key, and fallback
780 * for packets missing it.
781 * Even OoO DSS packets coming legitly after dropped or
782 * reordered MPC will cause fallback, but we don't have other
785 mptcp_get_options(skb, &mp_opt);
786 if (!(mp_opt.suboptions & OPTIONS_MPTCP_MPC))
789 } else if (subflow_req->mp_join) {
790 mptcp_get_options(skb, &mp_opt);
791 if (!(mp_opt.suboptions & OPTIONS_MPTCP_MPJ) ||
792 !subflow_hmac_valid(req, &mp_opt) ||
793 !mptcp_can_accept_new_subflow(subflow_req->msk)) {
794 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKMAC);
800 child = listener->icsk_af_ops->syn_recv_sock(sk, skb, req, dst,
801 req_unhash, own_req);
803 if (child && *own_req) {
804 struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(child);
806 tcp_rsk(req)->drop_req = false;
808 /* we need to fallback on ctx allocation failure and on pre-reqs
809 * checking above. In the latter scenario we additionally need
810 * to reset the context to non MPTCP status.
812 if (!ctx || fallback) {
813 if (fallback_is_fatal) {
814 subflow_add_reset_reason(skb, MPTCP_RST_EMPTCP);
820 /* ssk inherits options of listener sk */
821 ctx->setsockopt_seq = listener->setsockopt_seq;
823 if (ctx->mp_capable) {
824 ctx->conn = mptcp_sk_clone_init(listener->conn, &mp_opt, child, req);
829 owner = mptcp_sk(ctx->conn);
830 mptcp_pm_new_connection(owner, child, 1);
832 /* with OoO packets we can reach here without ingress
835 if (mp_opt.suboptions & OPTION_MPTCP_MPC_ACK) {
836 mptcp_subflow_fully_established(ctx, &mp_opt);
837 mptcp_pm_fully_established(owner, child);
838 ctx->pm_notified = 1;
840 } else if (ctx->mp_join) {
841 owner = subflow_req->msk;
843 subflow_add_reset_reason(skb, MPTCP_RST_EPROHIBIT);
847 /* move the msk reference ownership to the subflow */
848 subflow_req->msk = NULL;
849 ctx->conn = (struct sock *)owner;
851 if (subflow_use_different_sport(owner, sk)) {
852 pr_debug("ack inet_sport=%d %d",
853 ntohs(inet_sk(sk)->inet_sport),
854 ntohs(inet_sk((struct sock *)owner)->inet_sport));
855 if (!mptcp_pm_sport_in_anno_list(owner, sk)) {
856 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MISMATCHPORTACKRX);
859 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINPORTACKRX);
862 if (!mptcp_finish_join(child))
865 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKRX);
866 tcp_rsk(req)->drop_req = true;
870 /* check for expected invariant - should never trigger, just help
871 * catching eariler subtle bugs
873 WARN_ON_ONCE(child && *own_req && tcp_sk(child)->is_mptcp &&
874 (!mptcp_subflow_ctx(child) ||
875 !mptcp_subflow_ctx(child)->conn));
879 mptcp_subflow_drop_ctx(child);
880 tcp_rsk(req)->drop_req = true;
881 inet_csk_prepare_for_destroy_sock(child);
883 req->rsk_ops->send_reset(sk, skb);
885 /* The last child reference will be released by the caller */
889 mptcp_subflow_drop_ctx(child);
893 static struct inet_connection_sock_af_ops subflow_specific __ro_after_init;
894 static struct proto tcp_prot_override __ro_after_init;
896 enum mapping_status {
905 static void dbg_bad_map(struct mptcp_subflow_context *subflow, u32 ssn)
907 pr_debug("Bad mapping: ssn=%d map_seq=%d map_data_len=%d",
908 ssn, subflow->map_subflow_seq, subflow->map_data_len);
911 static bool skb_is_fully_mapped(struct sock *ssk, struct sk_buff *skb)
913 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
914 unsigned int skb_consumed;
916 skb_consumed = tcp_sk(ssk)->copied_seq - TCP_SKB_CB(skb)->seq;
917 if (WARN_ON_ONCE(skb_consumed >= skb->len))
920 return skb->len - skb_consumed <= subflow->map_data_len -
921 mptcp_subflow_get_map_offset(subflow);
924 static bool validate_mapping(struct sock *ssk, struct sk_buff *skb)
926 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
927 u32 ssn = tcp_sk(ssk)->copied_seq - subflow->ssn_offset;
929 if (unlikely(before(ssn, subflow->map_subflow_seq))) {
930 /* Mapping covers data later in the subflow stream,
931 * currently unsupported.
933 dbg_bad_map(subflow, ssn);
936 if (unlikely(!before(ssn, subflow->map_subflow_seq +
937 subflow->map_data_len))) {
938 /* Mapping does covers past subflow data, invalid */
939 dbg_bad_map(subflow, ssn);
945 static enum mapping_status validate_data_csum(struct sock *ssk, struct sk_buff *skb,
948 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
949 u32 offset, seq, delta;
956 /* mapping already validated on previous traversal */
957 if (subflow->map_csum_len == subflow->map_data_len)
960 /* traverse the receive queue, ensuring it contains a full
961 * DSS mapping and accumulating the related csum.
962 * Preserve the accoumlate csum across multiple calls, to compute
965 delta = subflow->map_data_len - subflow->map_csum_len;
967 seq = tcp_sk(ssk)->copied_seq + subflow->map_csum_len;
968 offset = seq - TCP_SKB_CB(skb)->seq;
970 /* if the current skb has not been accounted yet, csum its contents
971 * up to the amount covered by the current DSS
973 if (offset < skb->len) {
976 len = min(skb->len - offset, delta);
977 csum = skb_checksum(skb, offset, len, 0);
978 subflow->map_data_csum = csum_block_add(subflow->map_data_csum, csum,
979 subflow->map_csum_len);
982 subflow->map_csum_len += len;
987 if (skb_queue_is_last(&ssk->sk_receive_queue, skb)) {
988 /* if this subflow is closed, the partial mapping
989 * will be never completed; flush the pending skbs, so
990 * that subflow_sched_work_if_closed() can kick in
992 if (unlikely(ssk->sk_state == TCP_CLOSE))
993 while ((skb = skb_peek(&ssk->sk_receive_queue)))
994 sk_eat_skb(ssk, skb);
996 /* not enough data to validate the csum */
997 return MAPPING_EMPTY;
1000 /* the DSS mapping for next skbs will be validated later,
1001 * when a get_mapping_status call will process such skb
1006 /* note that 'map_data_len' accounts only for the carried data, does
1007 * not include the eventual seq increment due to the data fin,
1008 * while the pseudo header requires the original DSS data len,
1011 csum = __mptcp_make_csum(subflow->map_seq,
1012 subflow->map_subflow_seq,
1013 subflow->map_data_len + subflow->map_data_fin,
1014 subflow->map_data_csum);
1015 if (unlikely(csum)) {
1016 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DATACSUMERR);
1017 return MAPPING_BAD_CSUM;
1020 subflow->valid_csum_seen = 1;
1024 static enum mapping_status get_mapping_status(struct sock *ssk,
1025 struct mptcp_sock *msk)
1027 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1028 bool csum_reqd = READ_ONCE(msk->csum_enabled);
1029 struct mptcp_ext *mpext;
1030 struct sk_buff *skb;
1034 skb = skb_peek(&ssk->sk_receive_queue);
1036 return MAPPING_EMPTY;
1038 if (mptcp_check_fallback(ssk))
1039 return MAPPING_DUMMY;
1041 mpext = mptcp_get_ext(skb);
1042 if (!mpext || !mpext->use_map) {
1043 if (!subflow->map_valid && !skb->len) {
1044 /* the TCP stack deliver 0 len FIN pkt to the receive
1045 * queue, that is the only 0len pkts ever expected here,
1046 * and we can admit no mapping only for 0 len pkts
1048 if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN))
1049 WARN_ONCE(1, "0len seq %d:%d flags %x",
1050 TCP_SKB_CB(skb)->seq,
1051 TCP_SKB_CB(skb)->end_seq,
1052 TCP_SKB_CB(skb)->tcp_flags);
1053 sk_eat_skb(ssk, skb);
1054 return MAPPING_EMPTY;
1057 if (!subflow->map_valid)
1058 return MAPPING_INVALID;
1063 trace_get_mapping_status(mpext);
1065 data_len = mpext->data_len;
1066 if (data_len == 0) {
1067 pr_debug("infinite mapping received");
1068 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_INFINITEMAPRX);
1069 subflow->map_data_len = 0;
1070 return MAPPING_INVALID;
1073 if (mpext->data_fin == 1) {
1074 if (data_len == 1) {
1075 bool updated = mptcp_update_rcv_data_fin(msk, mpext->data_seq,
1077 pr_debug("DATA_FIN with no payload seq=%llu", mpext->data_seq);
1078 if (subflow->map_valid) {
1079 /* A DATA_FIN might arrive in a DSS
1080 * option before the previous mapping
1081 * has been fully consumed. Continue
1082 * handling the existing mapping.
1084 skb_ext_del(skb, SKB_EXT_MPTCP);
1088 mptcp_schedule_work((struct sock *)msk);
1090 return MAPPING_DATA_FIN;
1093 u64 data_fin_seq = mpext->data_seq + data_len - 1;
1095 /* If mpext->data_seq is a 32-bit value, data_fin_seq
1096 * must also be limited to 32 bits.
1099 data_fin_seq &= GENMASK_ULL(31, 0);
1101 mptcp_update_rcv_data_fin(msk, data_fin_seq, mpext->dsn64);
1102 pr_debug("DATA_FIN with mapping seq=%llu dsn64=%d",
1103 data_fin_seq, mpext->dsn64);
1106 /* Adjust for DATA_FIN using 1 byte of sequence space */
1110 map_seq = mptcp_expand_seq(READ_ONCE(msk->ack_seq), mpext->data_seq, mpext->dsn64);
1111 WRITE_ONCE(mptcp_sk(subflow->conn)->use_64bit_ack, !!mpext->dsn64);
1113 if (subflow->map_valid) {
1114 /* Allow replacing only with an identical map */
1115 if (subflow->map_seq == map_seq &&
1116 subflow->map_subflow_seq == mpext->subflow_seq &&
1117 subflow->map_data_len == data_len &&
1118 subflow->map_csum_reqd == mpext->csum_reqd) {
1119 skb_ext_del(skb, SKB_EXT_MPTCP);
1123 /* If this skb data are fully covered by the current mapping,
1124 * the new map would need caching, which is not supported
1126 if (skb_is_fully_mapped(ssk, skb)) {
1127 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DSSNOMATCH);
1128 return MAPPING_INVALID;
1131 /* will validate the next map after consuming the current one */
1135 subflow->map_seq = map_seq;
1136 subflow->map_subflow_seq = mpext->subflow_seq;
1137 subflow->map_data_len = data_len;
1138 subflow->map_valid = 1;
1139 subflow->map_data_fin = mpext->data_fin;
1140 subflow->mpc_map = mpext->mpc_map;
1141 subflow->map_csum_reqd = mpext->csum_reqd;
1142 subflow->map_csum_len = 0;
1143 subflow->map_data_csum = csum_unfold(mpext->csum);
1145 /* Cfr RFC 8684 Section 3.3.0 */
1146 if (unlikely(subflow->map_csum_reqd != csum_reqd))
1147 return MAPPING_INVALID;
1149 pr_debug("new map seq=%llu subflow_seq=%u data_len=%u csum=%d:%u",
1150 subflow->map_seq, subflow->map_subflow_seq,
1151 subflow->map_data_len, subflow->map_csum_reqd,
1152 subflow->map_data_csum);
1155 /* we revalidate valid mapping on new skb, because we must ensure
1156 * the current skb is completely covered by the available mapping
1158 if (!validate_mapping(ssk, skb)) {
1159 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DSSTCPMISMATCH);
1160 return MAPPING_INVALID;
1163 skb_ext_del(skb, SKB_EXT_MPTCP);
1166 return validate_data_csum(ssk, skb, csum_reqd);
1169 static void mptcp_subflow_discard_data(struct sock *ssk, struct sk_buff *skb,
1172 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1173 bool fin = TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN;
1176 incr = limit >= skb->len ? skb->len + fin : limit;
1178 pr_debug("discarding=%d len=%d seq=%d", incr, skb->len,
1179 subflow->map_subflow_seq);
1180 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DUPDATA);
1181 tcp_sk(ssk)->copied_seq += incr;
1182 if (!before(tcp_sk(ssk)->copied_seq, TCP_SKB_CB(skb)->end_seq))
1183 sk_eat_skb(ssk, skb);
1184 if (mptcp_subflow_get_map_offset(subflow) >= subflow->map_data_len)
1185 subflow->map_valid = 0;
1188 /* sched mptcp worker to remove the subflow if no more data is pending */
1189 static void subflow_sched_work_if_closed(struct mptcp_sock *msk, struct sock *ssk)
1191 if (likely(ssk->sk_state != TCP_CLOSE))
1194 if (skb_queue_empty(&ssk->sk_receive_queue) &&
1195 !test_and_set_bit(MPTCP_WORK_CLOSE_SUBFLOW, &msk->flags))
1196 mptcp_schedule_work((struct sock *)msk);
1199 static bool subflow_can_fallback(struct mptcp_subflow_context *subflow)
1201 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
1203 if (subflow->mp_join)
1205 else if (READ_ONCE(msk->csum_enabled))
1206 return !subflow->valid_csum_seen;
1208 return !subflow->fully_established;
1211 static void mptcp_subflow_fail(struct mptcp_sock *msk, struct sock *ssk)
1213 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1214 unsigned long fail_tout;
1216 /* greceful failure can happen only on the MPC subflow */
1217 if (WARN_ON_ONCE(ssk != READ_ONCE(msk->first)))
1220 /* since the close timeout take precedence on the fail one,
1221 * no need to start the latter when the first is already set
1223 if (sock_flag((struct sock *)msk, SOCK_DEAD))
1226 /* we don't need extreme accuracy here, use a zero fail_tout as special
1227 * value meaning no fail timeout at all;
1229 fail_tout = jiffies + TCP_RTO_MAX;
1232 WRITE_ONCE(subflow->fail_tout, fail_tout);
1235 mptcp_reset_tout_timer(msk, subflow->fail_tout);
1238 static bool subflow_check_data_avail(struct sock *ssk)
1240 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1241 enum mapping_status status;
1242 struct mptcp_sock *msk;
1243 struct sk_buff *skb;
1245 if (!skb_peek(&ssk->sk_receive_queue))
1246 WRITE_ONCE(subflow->data_avail, false);
1247 if (subflow->data_avail)
1250 msk = mptcp_sk(subflow->conn);
1255 status = get_mapping_status(ssk, msk);
1256 trace_subflow_check_data_avail(status, skb_peek(&ssk->sk_receive_queue));
1257 if (unlikely(status == MAPPING_INVALID || status == MAPPING_DUMMY ||
1258 status == MAPPING_BAD_CSUM))
1261 if (status != MAPPING_OK)
1264 skb = skb_peek(&ssk->sk_receive_queue);
1265 if (WARN_ON_ONCE(!skb))
1268 if (unlikely(!READ_ONCE(msk->can_ack)))
1271 old_ack = READ_ONCE(msk->ack_seq);
1272 ack_seq = mptcp_subflow_get_mapped_dsn(subflow);
1273 pr_debug("msk ack_seq=%llx subflow ack_seq=%llx", old_ack,
1275 if (unlikely(before64(ack_seq, old_ack))) {
1276 mptcp_subflow_discard_data(ssk, skb, old_ack - ack_seq);
1280 WRITE_ONCE(subflow->data_avail, true);
1286 subflow_sched_work_if_closed(msk, ssk);
1290 if (!__mptcp_check_fallback(msk)) {
1291 /* RFC 8684 section 3.7. */
1292 if (status == MAPPING_BAD_CSUM &&
1293 (subflow->mp_join || subflow->valid_csum_seen)) {
1294 subflow->send_mp_fail = 1;
1296 if (!READ_ONCE(msk->allow_infinite_fallback)) {
1297 subflow->reset_transient = 0;
1298 subflow->reset_reason = MPTCP_RST_EMIDDLEBOX;
1301 mptcp_subflow_fail(msk, ssk);
1302 WRITE_ONCE(subflow->data_avail, true);
1306 if (!subflow_can_fallback(subflow) && subflow->map_data_len) {
1307 /* fatal protocol error, close the socket.
1308 * subflow_error_report() will introduce the appropriate barriers
1310 subflow->reset_transient = 0;
1311 subflow->reset_reason = MPTCP_RST_EMPTCP;
1314 WRITE_ONCE(ssk->sk_err, EBADMSG);
1315 tcp_set_state(ssk, TCP_CLOSE);
1316 while ((skb = skb_peek(&ssk->sk_receive_queue)))
1317 sk_eat_skb(ssk, skb);
1318 tcp_send_active_reset(ssk, GFP_ATOMIC);
1319 WRITE_ONCE(subflow->data_avail, false);
1323 mptcp_do_fallback(ssk);
1326 skb = skb_peek(&ssk->sk_receive_queue);
1327 subflow->map_valid = 1;
1328 subflow->map_seq = READ_ONCE(msk->ack_seq);
1329 subflow->map_data_len = skb->len;
1330 subflow->map_subflow_seq = tcp_sk(ssk)->copied_seq - subflow->ssn_offset;
1331 WRITE_ONCE(subflow->data_avail, true);
1335 bool mptcp_subflow_data_available(struct sock *sk)
1337 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1339 /* check if current mapping is still valid */
1340 if (subflow->map_valid &&
1341 mptcp_subflow_get_map_offset(subflow) >= subflow->map_data_len) {
1342 subflow->map_valid = 0;
1343 WRITE_ONCE(subflow->data_avail, false);
1345 pr_debug("Done with mapping: seq=%u data_len=%u",
1346 subflow->map_subflow_seq,
1347 subflow->map_data_len);
1350 return subflow_check_data_avail(sk);
1353 /* If ssk has an mptcp parent socket, use the mptcp rcvbuf occupancy,
1356 * In mptcp, rwin is about the mptcp-level connection data.
1358 * Data that is still on the ssk rx queue can thus be ignored,
1359 * as far as mptcp peer is concerned that data is still inflight.
1360 * DSS ACK is updated when skb is moved to the mptcp rx queue.
1362 void mptcp_space(const struct sock *ssk, int *space, int *full_space)
1364 const struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1365 const struct sock *sk = subflow->conn;
1367 *space = __mptcp_space(sk);
1368 *full_space = mptcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf));
1371 static void subflow_error_report(struct sock *ssk)
1373 struct sock *sk = mptcp_subflow_ctx(ssk)->conn;
1375 /* bail early if this is a no-op, so that we avoid introducing a
1376 * problematic lockdep dependency between TCP accept queue lock
1377 * and msk socket spinlock
1382 mptcp_data_lock(sk);
1383 if (!sock_owned_by_user(sk))
1384 __mptcp_error_report(sk);
1386 __set_bit(MPTCP_ERROR_REPORT, &mptcp_sk(sk)->cb_flags);
1387 mptcp_data_unlock(sk);
1390 static void subflow_data_ready(struct sock *sk)
1392 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1393 u16 state = 1 << inet_sk_state_load(sk);
1394 struct sock *parent = subflow->conn;
1395 struct mptcp_sock *msk;
1397 trace_sk_data_ready(sk);
1399 msk = mptcp_sk(parent);
1400 if (state & TCPF_LISTEN) {
1401 /* MPJ subflow are removed from accept queue before reaching here,
1402 * avoid stray wakeups
1404 if (reqsk_queue_empty(&inet_csk(sk)->icsk_accept_queue))
1407 parent->sk_data_ready(parent);
1411 WARN_ON_ONCE(!__mptcp_check_fallback(msk) && !subflow->mp_capable &&
1412 !subflow->mp_join && !(state & TCPF_CLOSE));
1414 if (mptcp_subflow_data_available(sk)) {
1415 mptcp_data_ready(parent, sk);
1417 /* subflow-level lowat test are not relevant.
1418 * respect the msk-level threshold eventually mandating an immediate ack
1420 if (mptcp_data_avail(msk) < parent->sk_rcvlowat &&
1421 (tcp_sk(sk)->rcv_nxt - tcp_sk(sk)->rcv_wup) > inet_csk(sk)->icsk_ack.rcv_mss)
1422 inet_csk(sk)->icsk_ack.pending |= ICSK_ACK_NOW;
1423 } else if (unlikely(sk->sk_err)) {
1424 subflow_error_report(sk);
1428 static void subflow_write_space(struct sock *ssk)
1430 struct sock *sk = mptcp_subflow_ctx(ssk)->conn;
1432 mptcp_propagate_sndbuf(sk, ssk);
1433 mptcp_write_space(sk);
1436 static const struct inet_connection_sock_af_ops *
1437 subflow_default_af_ops(struct sock *sk)
1439 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1440 if (sk->sk_family == AF_INET6)
1441 return &subflow_v6_specific;
1443 return &subflow_specific;
1446 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1447 void mptcpv6_handle_mapped(struct sock *sk, bool mapped)
1449 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1450 struct inet_connection_sock *icsk = inet_csk(sk);
1451 const struct inet_connection_sock_af_ops *target;
1453 target = mapped ? &subflow_v6m_specific : subflow_default_af_ops(sk);
1455 pr_debug("subflow=%p family=%d ops=%p target=%p mapped=%d",
1456 subflow, sk->sk_family, icsk->icsk_af_ops, target, mapped);
1458 if (likely(icsk->icsk_af_ops == target))
1461 subflow->icsk_af_ops = icsk->icsk_af_ops;
1462 icsk->icsk_af_ops = target;
1466 void mptcp_info2sockaddr(const struct mptcp_addr_info *info,
1467 struct sockaddr_storage *addr,
1468 unsigned short family)
1470 memset(addr, 0, sizeof(*addr));
1471 addr->ss_family = family;
1472 if (addr->ss_family == AF_INET) {
1473 struct sockaddr_in *in_addr = (struct sockaddr_in *)addr;
1475 if (info->family == AF_INET)
1476 in_addr->sin_addr = info->addr;
1477 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1478 else if (ipv6_addr_v4mapped(&info->addr6))
1479 in_addr->sin_addr.s_addr = info->addr6.s6_addr32[3];
1481 in_addr->sin_port = info->port;
1483 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1484 else if (addr->ss_family == AF_INET6) {
1485 struct sockaddr_in6 *in6_addr = (struct sockaddr_in6 *)addr;
1487 if (info->family == AF_INET)
1488 ipv6_addr_set_v4mapped(info->addr.s_addr,
1489 &in6_addr->sin6_addr);
1491 in6_addr->sin6_addr = info->addr6;
1492 in6_addr->sin6_port = info->port;
1497 int __mptcp_subflow_connect(struct sock *sk, const struct mptcp_addr_info *loc,
1498 const struct mptcp_addr_info *remote)
1500 struct mptcp_sock *msk = mptcp_sk(sk);
1501 struct mptcp_subflow_context *subflow;
1502 struct sockaddr_storage addr;
1503 int remote_id = remote->id;
1504 int local_id = loc->id;
1505 int err = -ENOTCONN;
1513 if (!mptcp_is_fully_established(sk))
1516 err = mptcp_subflow_create_socket(sk, loc->family, &sf);
1521 subflow = mptcp_subflow_ctx(ssk);
1523 get_random_bytes(&subflow->local_nonce, sizeof(u32));
1524 } while (!subflow->local_nonce);
1527 subflow_set_local_id(subflow, local_id);
1529 mptcp_pm_get_flags_and_ifindex_by_id(msk, local_id,
1531 subflow->remote_key_valid = 1;
1532 subflow->remote_key = msk->remote_key;
1533 subflow->local_key = msk->local_key;
1534 subflow->token = msk->token;
1535 mptcp_info2sockaddr(loc, &addr, ssk->sk_family);
1537 addrlen = sizeof(struct sockaddr_in);
1538 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1539 if (addr.ss_family == AF_INET6)
1540 addrlen = sizeof(struct sockaddr_in6);
1542 ssk->sk_bound_dev_if = ifindex;
1543 err = kernel_bind(sf, (struct sockaddr *)&addr, addrlen);
1547 mptcp_crypto_key_sha(subflow->remote_key, &remote_token, NULL);
1548 pr_debug("msk=%p remote_token=%u local_id=%d remote_id=%d", msk,
1549 remote_token, local_id, remote_id);
1550 subflow->remote_token = remote_token;
1551 subflow->remote_id = remote_id;
1552 subflow->request_join = 1;
1553 subflow->request_bkup = !!(flags & MPTCP_PM_ADDR_FLAG_BACKUP);
1554 subflow->subflow_id = msk->subflow_id++;
1555 mptcp_info2sockaddr(remote, &addr, ssk->sk_family);
1558 list_add_tail(&subflow->node, &msk->conn_list);
1559 err = kernel_connect(sf, (struct sockaddr *)&addr, addrlen, O_NONBLOCK);
1560 if (err && err != -EINPROGRESS)
1563 /* discard the subflow socket */
1564 mptcp_sock_graft(ssk, sk->sk_socket);
1565 iput(SOCK_INODE(sf));
1566 WRITE_ONCE(msk->allow_infinite_fallback, false);
1567 mptcp_stop_tout_timer(sk);
1571 list_del(&subflow->node);
1572 sock_put(mptcp_subflow_tcp_sock(subflow));
1575 subflow->disposable = 1;
1579 /* we account subflows before the creation, and this failures will not
1580 * be caught by sk_state_change()
1582 mptcp_pm_close_subflow(msk);
1586 static void mptcp_attach_cgroup(struct sock *parent, struct sock *child)
1588 #ifdef CONFIG_SOCK_CGROUP_DATA
1589 struct sock_cgroup_data *parent_skcd = &parent->sk_cgrp_data,
1590 *child_skcd = &child->sk_cgrp_data;
1592 /* only the additional subflows created by kworkers have to be modified */
1593 if (cgroup_id(sock_cgroup_ptr(parent_skcd)) !=
1594 cgroup_id(sock_cgroup_ptr(child_skcd))) {
1596 struct mem_cgroup *memcg = parent->sk_memcg;
1598 mem_cgroup_sk_free(child);
1599 if (memcg && css_tryget(&memcg->css))
1600 child->sk_memcg = memcg;
1601 #endif /* CONFIG_MEMCG */
1603 cgroup_sk_free(child_skcd);
1604 *child_skcd = *parent_skcd;
1605 cgroup_sk_clone(child_skcd);
1607 #endif /* CONFIG_SOCK_CGROUP_DATA */
1610 static void mptcp_subflow_ops_override(struct sock *ssk)
1612 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1613 if (ssk->sk_prot == &tcpv6_prot)
1614 ssk->sk_prot = &tcpv6_prot_override;
1617 ssk->sk_prot = &tcp_prot_override;
1620 static void mptcp_subflow_ops_undo_override(struct sock *ssk)
1622 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1623 if (ssk->sk_prot == &tcpv6_prot_override)
1624 ssk->sk_prot = &tcpv6_prot;
1627 ssk->sk_prot = &tcp_prot;
1630 int mptcp_subflow_create_socket(struct sock *sk, unsigned short family,
1631 struct socket **new_sock)
1633 struct mptcp_subflow_context *subflow;
1634 struct net *net = sock_net(sk);
1638 /* un-accepted server sockets can reach here - on bad configuration
1639 * bail early to avoid greater trouble later
1641 if (unlikely(!sk->sk_socket))
1644 err = sock_create_kern(net, family, SOCK_STREAM, IPPROTO_TCP, &sf);
1648 lock_sock_nested(sf->sk, SINGLE_DEPTH_NESTING);
1650 err = security_mptcp_add_subflow(sk, sf->sk);
1654 /* the newly created socket has to be in the same cgroup as its parent */
1655 mptcp_attach_cgroup(sk, sf->sk);
1657 /* kernel sockets do not by default acquire net ref, but TCP timer
1659 * Update ns_tracker to current stack trace and refcounted tracker.
1661 __netns_tracker_free(net, &sf->sk->ns_tracker, false);
1662 sf->sk->sk_net_refcnt = 1;
1663 get_net_track(net, &sf->sk->ns_tracker, GFP_KERNEL);
1664 sock_inuse_add(net, 1);
1665 err = tcp_set_ulp(sf->sk, "mptcp");
1669 mptcp_sockopt_sync_locked(mptcp_sk(sk), sf->sk);
1670 release_sock(sf->sk);
1672 /* the newly created socket really belongs to the owning MPTCP master
1673 * socket, even if for additional subflows the allocation is performed
1674 * by a kernel workqueue. Adjust inode references, so that the
1675 * procfs/diag interfaces really show this one belonging to the correct
1678 SOCK_INODE(sf)->i_ino = SOCK_INODE(sk->sk_socket)->i_ino;
1679 SOCK_INODE(sf)->i_uid = SOCK_INODE(sk->sk_socket)->i_uid;
1680 SOCK_INODE(sf)->i_gid = SOCK_INODE(sk->sk_socket)->i_gid;
1682 subflow = mptcp_subflow_ctx(sf->sk);
1683 pr_debug("subflow=%p", subflow);
1688 mptcp_subflow_ops_override(sf->sk);
1693 release_sock(sf->sk);
1698 static struct mptcp_subflow_context *subflow_create_ctx(struct sock *sk,
1701 struct inet_connection_sock *icsk = inet_csk(sk);
1702 struct mptcp_subflow_context *ctx;
1704 ctx = kzalloc(sizeof(*ctx), priority);
1708 rcu_assign_pointer(icsk->icsk_ulp_data, ctx);
1709 INIT_LIST_HEAD(&ctx->node);
1710 INIT_LIST_HEAD(&ctx->delegated_node);
1712 pr_debug("subflow=%p", ctx);
1719 static void __subflow_state_change(struct sock *sk)
1721 struct socket_wq *wq;
1724 wq = rcu_dereference(sk->sk_wq);
1725 if (skwq_has_sleeper(wq))
1726 wake_up_interruptible_all(&wq->wait);
1730 static bool subflow_is_done(const struct sock *sk)
1732 return sk->sk_shutdown & RCV_SHUTDOWN || sk->sk_state == TCP_CLOSE;
1735 static void subflow_state_change(struct sock *sk)
1737 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1738 struct sock *parent = subflow->conn;
1739 struct mptcp_sock *msk;
1741 __subflow_state_change(sk);
1743 msk = mptcp_sk(parent);
1744 if (subflow_simultaneous_connect(sk)) {
1745 mptcp_do_fallback(sk);
1746 mptcp_rcv_space_init(msk, sk);
1748 subflow->conn_finished = 1;
1749 mptcp_propagate_state(parent, sk);
1752 /* as recvmsg() does not acquire the subflow socket for ssk selection
1753 * a fin packet carrying a DSS can be unnoticed if we don't trigger
1754 * the data available machinery here.
1756 if (mptcp_subflow_data_available(sk))
1757 mptcp_data_ready(parent, sk);
1758 else if (unlikely(sk->sk_err))
1759 subflow_error_report(sk);
1761 subflow_sched_work_if_closed(mptcp_sk(parent), sk);
1763 /* when the fallback subflow closes the rx side, trigger a 'dummy'
1764 * ingress data fin, so that the msk state will follow along
1766 if (__mptcp_check_fallback(msk) && subflow_is_done(sk) && msk->first == sk &&
1767 mptcp_update_rcv_data_fin(msk, READ_ONCE(msk->ack_seq), true))
1768 mptcp_schedule_work(parent);
1771 void mptcp_subflow_queue_clean(struct sock *listener_sk, struct sock *listener_ssk)
1773 struct request_sock_queue *queue = &inet_csk(listener_ssk)->icsk_accept_queue;
1774 struct request_sock *req, *head, *tail;
1775 struct mptcp_subflow_context *subflow;
1776 struct sock *sk, *ssk;
1778 /* Due to lock dependencies no relevant lock can be acquired under rskq_lock.
1779 * Splice the req list, so that accept() can not reach the pending ssk after
1780 * the listener socket is released below.
1782 spin_lock_bh(&queue->rskq_lock);
1783 head = queue->rskq_accept_head;
1784 tail = queue->rskq_accept_tail;
1785 queue->rskq_accept_head = NULL;
1786 queue->rskq_accept_tail = NULL;
1787 spin_unlock_bh(&queue->rskq_lock);
1792 /* can't acquire the msk socket lock under the subflow one,
1793 * or will cause ABBA deadlock
1795 release_sock(listener_ssk);
1797 for (req = head; req; req = req->dl_next) {
1799 if (!sk_is_mptcp(ssk))
1802 subflow = mptcp_subflow_ctx(ssk);
1803 if (!subflow || !subflow->conn)
1809 lock_sock_nested(sk, SINGLE_DEPTH_NESTING);
1810 __mptcp_unaccepted_force_close(sk);
1813 /* lockdep will report a false positive ABBA deadlock
1814 * between cancel_work_sync and the listener socket.
1815 * The involved locks belong to different sockets WRT
1816 * the existing AB chain.
1817 * Using a per socket key is problematic as key
1818 * deregistration requires process context and must be
1819 * performed at socket disposal time, in atomic
1821 * Just tell lockdep to consider the listener socket
1824 mutex_release(&listener_sk->sk_lock.dep_map, _RET_IP_);
1825 mptcp_cancel_work(sk);
1826 mutex_acquire(&listener_sk->sk_lock.dep_map, 0, 0, _RET_IP_);
1831 /* we are still under the listener msk socket lock */
1832 lock_sock_nested(listener_ssk, SINGLE_DEPTH_NESTING);
1834 /* restore the listener queue, to let the TCP code clean it up */
1835 spin_lock_bh(&queue->rskq_lock);
1836 WARN_ON_ONCE(queue->rskq_accept_head);
1837 queue->rskq_accept_head = head;
1838 queue->rskq_accept_tail = tail;
1839 spin_unlock_bh(&queue->rskq_lock);
1842 static int subflow_ulp_init(struct sock *sk)
1844 struct inet_connection_sock *icsk = inet_csk(sk);
1845 struct mptcp_subflow_context *ctx;
1846 struct tcp_sock *tp = tcp_sk(sk);
1849 /* disallow attaching ULP to a socket unless it has been
1850 * created with sock_create_kern()
1852 if (!sk->sk_kern_sock) {
1857 ctx = subflow_create_ctx(sk, GFP_KERNEL);
1863 pr_debug("subflow=%p, family=%d", ctx, sk->sk_family);
1866 ctx->icsk_af_ops = icsk->icsk_af_ops;
1867 icsk->icsk_af_ops = subflow_default_af_ops(sk);
1868 ctx->tcp_state_change = sk->sk_state_change;
1869 ctx->tcp_error_report = sk->sk_error_report;
1871 WARN_ON_ONCE(sk->sk_data_ready != sock_def_readable);
1872 WARN_ON_ONCE(sk->sk_write_space != sk_stream_write_space);
1874 sk->sk_data_ready = subflow_data_ready;
1875 sk->sk_write_space = subflow_write_space;
1876 sk->sk_state_change = subflow_state_change;
1877 sk->sk_error_report = subflow_error_report;
1882 static void subflow_ulp_release(struct sock *ssk)
1884 struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(ssk);
1885 bool release = true;
1893 /* if the msk has been orphaned, keep the ctx
1894 * alive, will be freed by __mptcp_close_ssk(),
1895 * when the subflow is still unaccepted
1897 release = ctx->disposable || list_empty(&ctx->node);
1899 /* inet_child_forget() does not call sk_state_change(),
1900 * explicitly trigger the socket close machinery
1902 if (!release && !test_and_set_bit(MPTCP_WORK_CLOSE_SUBFLOW,
1903 &mptcp_sk(sk)->flags))
1904 mptcp_schedule_work(sk);
1908 mptcp_subflow_ops_undo_override(ssk);
1910 kfree_rcu(ctx, rcu);
1913 static void subflow_ulp_clone(const struct request_sock *req,
1915 const gfp_t priority)
1917 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
1918 struct mptcp_subflow_context *old_ctx = mptcp_subflow_ctx(newsk);
1919 struct mptcp_subflow_context *new_ctx;
1921 if (!tcp_rsk(req)->is_mptcp ||
1922 (!subflow_req->mp_capable && !subflow_req->mp_join)) {
1923 subflow_ulp_fallback(newsk, old_ctx);
1927 new_ctx = subflow_create_ctx(newsk, priority);
1929 subflow_ulp_fallback(newsk, old_ctx);
1933 new_ctx->conn_finished = 1;
1934 new_ctx->icsk_af_ops = old_ctx->icsk_af_ops;
1935 new_ctx->tcp_state_change = old_ctx->tcp_state_change;
1936 new_ctx->tcp_error_report = old_ctx->tcp_error_report;
1937 new_ctx->rel_write_seq = 1;
1938 new_ctx->tcp_sock = newsk;
1940 if (subflow_req->mp_capable) {
1941 /* see comments in subflow_syn_recv_sock(), MPTCP connection
1942 * is fully established only after we receive the remote key
1944 new_ctx->mp_capable = 1;
1945 new_ctx->local_key = subflow_req->local_key;
1946 new_ctx->token = subflow_req->token;
1947 new_ctx->ssn_offset = subflow_req->ssn_offset;
1948 new_ctx->idsn = subflow_req->idsn;
1950 /* this is the first subflow, id is always 0 */
1951 new_ctx->local_id_valid = 1;
1952 } else if (subflow_req->mp_join) {
1953 new_ctx->ssn_offset = subflow_req->ssn_offset;
1954 new_ctx->mp_join = 1;
1955 new_ctx->fully_established = 1;
1956 new_ctx->remote_key_valid = 1;
1957 new_ctx->backup = subflow_req->backup;
1958 new_ctx->remote_id = subflow_req->remote_id;
1959 new_ctx->token = subflow_req->token;
1960 new_ctx->thmac = subflow_req->thmac;
1962 /* the subflow req id is valid, fetched via subflow_check_req()
1963 * and subflow_token_join_request()
1965 subflow_set_local_id(new_ctx, subflow_req->local_id);
1969 static void tcp_release_cb_override(struct sock *ssk)
1971 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1974 /* process and clear all the pending actions, but leave the subflow into
1975 * the napi queue. To respect locking, only the same CPU that originated
1976 * the action can touch the list. mptcp_napi_poll will take care of it.
1978 status = set_mask_bits(&subflow->delegated_status, MPTCP_DELEGATE_ACTIONS_MASK, 0);
1980 mptcp_subflow_process_delegated(ssk, status);
1982 tcp_release_cb(ssk);
1985 static struct tcp_ulp_ops subflow_ulp_ops __read_mostly = {
1987 .owner = THIS_MODULE,
1988 .init = subflow_ulp_init,
1989 .release = subflow_ulp_release,
1990 .clone = subflow_ulp_clone,
1993 static int subflow_ops_init(struct request_sock_ops *subflow_ops)
1995 subflow_ops->obj_size = sizeof(struct mptcp_subflow_request_sock);
1997 subflow_ops->slab = kmem_cache_create(subflow_ops->slab_name,
1998 subflow_ops->obj_size, 0,
2000 SLAB_TYPESAFE_BY_RCU,
2002 if (!subflow_ops->slab)
2008 void __init mptcp_subflow_init(void)
2010 mptcp_subflow_v4_request_sock_ops = tcp_request_sock_ops;
2011 mptcp_subflow_v4_request_sock_ops.slab_name = "request_sock_subflow_v4";
2012 mptcp_subflow_v4_request_sock_ops.destructor = subflow_v4_req_destructor;
2014 if (subflow_ops_init(&mptcp_subflow_v4_request_sock_ops) != 0)
2015 panic("MPTCP: failed to init subflow v4 request sock ops\n");
2017 subflow_request_sock_ipv4_ops = tcp_request_sock_ipv4_ops;
2018 subflow_request_sock_ipv4_ops.route_req = subflow_v4_route_req;
2019 subflow_request_sock_ipv4_ops.send_synack = subflow_v4_send_synack;
2021 subflow_specific = ipv4_specific;
2022 subflow_specific.conn_request = subflow_v4_conn_request;
2023 subflow_specific.syn_recv_sock = subflow_syn_recv_sock;
2024 subflow_specific.sk_rx_dst_set = subflow_finish_connect;
2025 subflow_specific.rebuild_header = subflow_rebuild_header;
2027 tcp_prot_override = tcp_prot;
2028 tcp_prot_override.release_cb = tcp_release_cb_override;
2030 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
2031 /* In struct mptcp_subflow_request_sock, we assume the TCP request sock
2032 * structures for v4 and v6 have the same size. It should not changed in
2033 * the future but better to make sure to be warned if it is no longer
2036 BUILD_BUG_ON(sizeof(struct tcp_request_sock) != sizeof(struct tcp6_request_sock));
2038 mptcp_subflow_v6_request_sock_ops = tcp6_request_sock_ops;
2039 mptcp_subflow_v6_request_sock_ops.slab_name = "request_sock_subflow_v6";
2040 mptcp_subflow_v6_request_sock_ops.destructor = subflow_v6_req_destructor;
2042 if (subflow_ops_init(&mptcp_subflow_v6_request_sock_ops) != 0)
2043 panic("MPTCP: failed to init subflow v6 request sock ops\n");
2045 subflow_request_sock_ipv6_ops = tcp_request_sock_ipv6_ops;
2046 subflow_request_sock_ipv6_ops.route_req = subflow_v6_route_req;
2047 subflow_request_sock_ipv6_ops.send_synack = subflow_v6_send_synack;
2049 subflow_v6_specific = ipv6_specific;
2050 subflow_v6_specific.conn_request = subflow_v6_conn_request;
2051 subflow_v6_specific.syn_recv_sock = subflow_syn_recv_sock;
2052 subflow_v6_specific.sk_rx_dst_set = subflow_finish_connect;
2053 subflow_v6_specific.rebuild_header = subflow_v6_rebuild_header;
2055 subflow_v6m_specific = subflow_v6_specific;
2056 subflow_v6m_specific.queue_xmit = ipv4_specific.queue_xmit;
2057 subflow_v6m_specific.send_check = ipv4_specific.send_check;
2058 subflow_v6m_specific.net_header_len = ipv4_specific.net_header_len;
2059 subflow_v6m_specific.mtu_reduced = ipv4_specific.mtu_reduced;
2060 subflow_v6m_specific.rebuild_header = subflow_rebuild_header;
2062 tcpv6_prot_override = tcpv6_prot;
2063 tcpv6_prot_override.release_cb = tcp_release_cb_override;
2066 mptcp_diag_subflow_init(&subflow_ulp_ops);
2068 if (tcp_register_ulp(&subflow_ulp_ops) != 0)
2069 panic("MPTCP: failed to register subflows to ULP\n");