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>
18 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
19 #include <net/ip6_route.h>
20 #include <net/transp_v6.h>
22 #include <net/mptcp.h>
26 #include <trace/events/mptcp.h>
27 #include <trace/events/sock.h>
29 static void mptcp_subflow_ops_undo_override(struct sock *ssk);
31 static void SUBFLOW_REQ_INC_STATS(struct request_sock *req,
32 enum linux_mptcp_mib_field field)
34 MPTCP_INC_STATS(sock_net(req_to_sk(req)), field);
37 static void subflow_req_destructor(struct request_sock *req)
39 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
41 pr_debug("subflow_req=%p", subflow_req);
44 sock_put((struct sock *)subflow_req->msk);
46 mptcp_token_destroy_request(req);
49 static void subflow_generate_hmac(u64 key1, u64 key2, u32 nonce1, u32 nonce2,
54 put_unaligned_be32(nonce1, &msg[0]);
55 put_unaligned_be32(nonce2, &msg[4]);
57 mptcp_crypto_hmac_sha(key1, key2, msg, 8, hmac);
60 static bool mptcp_can_accept_new_subflow(const struct mptcp_sock *msk)
62 return mptcp_is_fully_established((void *)msk) &&
63 ((mptcp_pm_is_userspace(msk) &&
64 mptcp_userspace_pm_active(msk)) ||
65 READ_ONCE(msk->pm.accept_subflow));
68 /* validate received token and create truncated hmac and nonce for SYN-ACK */
69 static void subflow_req_create_thmac(struct mptcp_subflow_request_sock *subflow_req)
71 struct mptcp_sock *msk = subflow_req->msk;
72 u8 hmac[SHA256_DIGEST_SIZE];
74 get_random_bytes(&subflow_req->local_nonce, sizeof(u32));
76 subflow_generate_hmac(READ_ONCE(msk->local_key),
77 READ_ONCE(msk->remote_key),
78 subflow_req->local_nonce,
79 subflow_req->remote_nonce, hmac);
81 subflow_req->thmac = get_unaligned_be64(hmac);
84 static struct mptcp_sock *subflow_token_join_request(struct request_sock *req)
86 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
87 struct mptcp_sock *msk;
90 msk = mptcp_token_get_sock(sock_net(req_to_sk(req)), subflow_req->token);
92 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINNOTOKEN);
96 local_id = mptcp_pm_get_local_id(msk, (struct sock_common *)req);
98 sock_put((struct sock *)msk);
101 subflow_req->local_id = local_id;
106 static void subflow_init_req(struct request_sock *req, const struct sock *sk_listener)
108 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
110 subflow_req->mp_capable = 0;
111 subflow_req->mp_join = 0;
112 subflow_req->csum_reqd = mptcp_is_checksum_enabled(sock_net(sk_listener));
113 subflow_req->allow_join_id0 = mptcp_allow_join_id0(sock_net(sk_listener));
114 subflow_req->msk = NULL;
115 mptcp_token_init_request(req);
118 static bool subflow_use_different_sport(struct mptcp_sock *msk, const struct sock *sk)
120 return inet_sk(sk)->inet_sport != inet_sk((struct sock *)msk)->inet_sport;
123 static void subflow_add_reset_reason(struct sk_buff *skb, u8 reason)
125 struct mptcp_ext *mpext = skb_ext_add(skb, SKB_EXT_MPTCP);
128 memset(mpext, 0, sizeof(*mpext));
129 mpext->reset_reason = reason;
133 /* Init mptcp request socket.
135 * Returns an error code if a JOIN has failed and a TCP reset
138 static int subflow_check_req(struct request_sock *req,
139 const struct sock *sk_listener,
142 struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk_listener);
143 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
144 struct mptcp_options_received mp_opt;
145 bool opt_mp_capable, opt_mp_join;
147 pr_debug("subflow_req=%p, listener=%p", subflow_req, listener);
149 #ifdef CONFIG_TCP_MD5SIG
150 /* no MPTCP if MD5SIG is enabled on this socket or we may run out of
153 if (rcu_access_pointer(tcp_sk(sk_listener)->md5sig_info))
157 mptcp_get_options(skb, &mp_opt);
159 opt_mp_capable = !!(mp_opt.suboptions & OPTION_MPTCP_MPC_SYN);
160 opt_mp_join = !!(mp_opt.suboptions & OPTION_MPTCP_MPJ_SYN);
161 if (opt_mp_capable) {
162 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MPCAPABLEPASSIVE);
166 } else if (opt_mp_join) {
167 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINSYNRX);
170 if (opt_mp_capable && listener->request_mptcp) {
171 int err, retries = MPTCP_TOKEN_MAX_RETRIES;
173 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq;
176 get_random_bytes(&subflow_req->local_key, sizeof(subflow_req->local_key));
177 } while (subflow_req->local_key == 0);
179 if (unlikely(req->syncookie)) {
180 mptcp_crypto_key_sha(subflow_req->local_key,
183 if (mptcp_token_exists(subflow_req->token)) {
186 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_TOKENFALLBACKINIT);
188 subflow_req->mp_capable = 1;
193 err = mptcp_token_new_request(req);
195 subflow_req->mp_capable = 1;
196 else if (retries-- > 0)
199 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_TOKENFALLBACKINIT);
201 } else if (opt_mp_join && listener->request_mptcp) {
202 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq;
203 subflow_req->mp_join = 1;
204 subflow_req->backup = mp_opt.backup;
205 subflow_req->remote_id = mp_opt.join_id;
206 subflow_req->token = mp_opt.token;
207 subflow_req->remote_nonce = mp_opt.nonce;
208 subflow_req->msk = subflow_token_join_request(req);
210 /* Can't fall back to TCP in this case. */
211 if (!subflow_req->msk) {
212 subflow_add_reset_reason(skb, MPTCP_RST_EMPTCP);
216 if (subflow_use_different_sport(subflow_req->msk, sk_listener)) {
217 pr_debug("syn inet_sport=%d %d",
218 ntohs(inet_sk(sk_listener)->inet_sport),
219 ntohs(inet_sk((struct sock *)subflow_req->msk)->inet_sport));
220 if (!mptcp_pm_sport_in_anno_list(subflow_req->msk, sk_listener)) {
221 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MISMATCHPORTSYNRX);
224 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINPORTSYNRX);
227 subflow_req_create_thmac(subflow_req);
229 if (unlikely(req->syncookie)) {
230 if (mptcp_can_accept_new_subflow(subflow_req->msk))
231 subflow_init_req_cookie_join_save(subflow_req, skb);
236 pr_debug("token=%u, remote_nonce=%u msk=%p", subflow_req->token,
237 subflow_req->remote_nonce, subflow_req->msk);
243 int mptcp_subflow_init_cookie_req(struct request_sock *req,
244 const struct sock *sk_listener,
247 struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk_listener);
248 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
249 struct mptcp_options_received mp_opt;
250 bool opt_mp_capable, opt_mp_join;
253 subflow_init_req(req, sk_listener);
254 mptcp_get_options(skb, &mp_opt);
256 opt_mp_capable = !!(mp_opt.suboptions & OPTION_MPTCP_MPC_ACK);
257 opt_mp_join = !!(mp_opt.suboptions & OPTION_MPTCP_MPJ_ACK);
258 if (opt_mp_capable && opt_mp_join)
261 if (opt_mp_capable && listener->request_mptcp) {
262 if (mp_opt.sndr_key == 0)
265 subflow_req->local_key = mp_opt.rcvr_key;
266 err = mptcp_token_new_request(req);
270 subflow_req->mp_capable = 1;
271 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq - 1;
272 } else if (opt_mp_join && listener->request_mptcp) {
273 if (!mptcp_token_join_cookie_init_state(subflow_req, skb))
276 subflow_req->mp_join = 1;
277 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq - 1;
282 EXPORT_SYMBOL_GPL(mptcp_subflow_init_cookie_req);
284 static struct dst_entry *subflow_v4_route_req(const struct sock *sk,
287 struct request_sock *req)
289 struct dst_entry *dst;
292 tcp_rsk(req)->is_mptcp = 1;
293 subflow_init_req(req, sk);
295 dst = tcp_request_sock_ipv4_ops.route_req(sk, skb, fl, req);
299 err = subflow_check_req(req, sk, skb);
305 tcp_request_sock_ops.send_reset(sk, skb);
309 static void subflow_prep_synack(const struct sock *sk, struct request_sock *req,
310 struct tcp_fastopen_cookie *foc,
311 enum tcp_synack_type synack_type)
313 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
314 struct inet_request_sock *ireq = inet_rsk(req);
316 /* clear tstamp_ok, as needed depending on cookie */
317 if (foc && foc->len > -1)
320 if (synack_type == TCP_SYNACK_FASTOPEN)
321 mptcp_fastopen_subflow_synack_set_params(subflow, req);
324 static int subflow_v4_send_synack(const struct sock *sk, struct dst_entry *dst,
326 struct request_sock *req,
327 struct tcp_fastopen_cookie *foc,
328 enum tcp_synack_type synack_type,
329 struct sk_buff *syn_skb)
331 subflow_prep_synack(sk, req, foc, synack_type);
333 return tcp_request_sock_ipv4_ops.send_synack(sk, dst, fl, req, foc,
334 synack_type, syn_skb);
337 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
338 static int subflow_v6_send_synack(const struct sock *sk, struct dst_entry *dst,
340 struct request_sock *req,
341 struct tcp_fastopen_cookie *foc,
342 enum tcp_synack_type synack_type,
343 struct sk_buff *syn_skb)
345 subflow_prep_synack(sk, req, foc, synack_type);
347 return tcp_request_sock_ipv6_ops.send_synack(sk, dst, fl, req, foc,
348 synack_type, syn_skb);
351 static struct dst_entry *subflow_v6_route_req(const struct sock *sk,
354 struct request_sock *req)
356 struct dst_entry *dst;
359 tcp_rsk(req)->is_mptcp = 1;
360 subflow_init_req(req, sk);
362 dst = tcp_request_sock_ipv6_ops.route_req(sk, skb, fl, req);
366 err = subflow_check_req(req, sk, skb);
372 tcp6_request_sock_ops.send_reset(sk, skb);
377 /* validate received truncated hmac and create hmac for third ACK */
378 static bool subflow_thmac_valid(struct mptcp_subflow_context *subflow)
380 u8 hmac[SHA256_DIGEST_SIZE];
383 subflow_generate_hmac(subflow->remote_key, subflow->local_key,
384 subflow->remote_nonce, subflow->local_nonce,
387 thmac = get_unaligned_be64(hmac);
388 pr_debug("subflow=%p, token=%u, thmac=%llu, subflow->thmac=%llu\n",
389 subflow, subflow->token, thmac, subflow->thmac);
391 return thmac == subflow->thmac;
394 void mptcp_subflow_reset(struct sock *ssk)
396 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
397 struct sock *sk = subflow->conn;
399 /* mptcp_mp_fail_no_response() can reach here on an already closed
402 if (ssk->sk_state == TCP_CLOSE)
405 /* must hold: tcp_done() could drop last reference on parent */
408 tcp_send_active_reset(ssk, GFP_ATOMIC);
410 if (!test_and_set_bit(MPTCP_WORK_CLOSE_SUBFLOW, &mptcp_sk(sk)->flags))
411 mptcp_schedule_work(sk);
416 static bool subflow_use_different_dport(struct mptcp_sock *msk, const struct sock *sk)
418 return inet_sk(sk)->inet_dport != inet_sk((struct sock *)msk)->inet_dport;
421 void __mptcp_sync_state(struct sock *sk, int state)
423 struct mptcp_subflow_context *subflow;
424 struct mptcp_sock *msk = mptcp_sk(sk);
425 struct sock *ssk = msk->first;
427 subflow = mptcp_subflow_ctx(ssk);
428 __mptcp_propagate_sndbuf(sk, ssk);
429 if (!msk->rcvspace_init)
430 mptcp_rcv_space_init(msk, ssk);
432 if (sk->sk_state == TCP_SYN_SENT) {
433 /* subflow->idsn is always available is TCP_SYN_SENT state,
434 * even for the FASTOPEN scenarios
436 WRITE_ONCE(msk->write_seq, subflow->idsn + 1);
437 WRITE_ONCE(msk->snd_nxt, msk->write_seq);
438 mptcp_set_state(sk, state);
439 sk->sk_state_change(sk);
443 static void subflow_set_remote_key(struct mptcp_sock *msk,
444 struct mptcp_subflow_context *subflow,
445 const struct mptcp_options_received *mp_opt)
447 /* active MPC subflow will reach here multiple times:
448 * at subflow_finish_connect() time and at 4th ack time
450 if (subflow->remote_key_valid)
453 subflow->remote_key_valid = 1;
454 subflow->remote_key = mp_opt->sndr_key;
455 mptcp_crypto_key_sha(subflow->remote_key, NULL, &subflow->iasn);
458 WRITE_ONCE(msk->remote_key, subflow->remote_key);
459 WRITE_ONCE(msk->ack_seq, subflow->iasn);
460 WRITE_ONCE(msk->can_ack, true);
461 atomic64_set(&msk->rcv_wnd_sent, subflow->iasn);
464 static void mptcp_propagate_state(struct sock *sk, struct sock *ssk,
465 struct mptcp_subflow_context *subflow,
466 const struct mptcp_options_received *mp_opt)
468 struct mptcp_sock *msk = mptcp_sk(sk);
472 /* Options are available only in the non fallback cases
473 * avoid updating rx path fields otherwise
475 WRITE_ONCE(msk->snd_una, subflow->idsn + 1);
476 WRITE_ONCE(msk->wnd_end, subflow->idsn + 1 + tcp_sk(ssk)->snd_wnd);
477 subflow_set_remote_key(msk, subflow, mp_opt);
480 if (!sock_owned_by_user(sk)) {
481 __mptcp_sync_state(sk, ssk->sk_state);
483 msk->pending_state = ssk->sk_state;
484 __set_bit(MPTCP_SYNC_STATE, &msk->cb_flags);
486 mptcp_data_unlock(sk);
489 static void subflow_finish_connect(struct sock *sk, const struct sk_buff *skb)
491 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
492 struct mptcp_options_received mp_opt;
493 struct sock *parent = subflow->conn;
494 struct mptcp_sock *msk;
496 subflow->icsk_af_ops->sk_rx_dst_set(sk, skb);
498 /* be sure no special action on any packet other than syn-ack */
499 if (subflow->conn_finished)
502 msk = mptcp_sk(parent);
503 subflow->rel_write_seq = 1;
504 subflow->conn_finished = 1;
505 subflow->ssn_offset = TCP_SKB_CB(skb)->seq;
506 pr_debug("subflow=%p synack seq=%x", subflow, subflow->ssn_offset);
508 mptcp_get_options(skb, &mp_opt);
509 if (subflow->request_mptcp) {
510 if (!(mp_opt.suboptions & OPTION_MPTCP_MPC_SYNACK)) {
511 MPTCP_INC_STATS(sock_net(sk),
512 MPTCP_MIB_MPCAPABLEACTIVEFALLBACK);
513 mptcp_do_fallback(sk);
518 if (mp_opt.suboptions & OPTION_MPTCP_CSUMREQD)
519 WRITE_ONCE(msk->csum_enabled, true);
520 if (mp_opt.deny_join_id0)
521 WRITE_ONCE(msk->pm.remote_deny_join_id0, true);
522 subflow->mp_capable = 1;
523 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPCAPABLEACTIVEACK);
524 mptcp_finish_connect(sk);
525 mptcp_propagate_state(parent, sk, subflow, &mp_opt);
526 } else if (subflow->request_join) {
527 u8 hmac[SHA256_DIGEST_SIZE];
529 if (!(mp_opt.suboptions & OPTION_MPTCP_MPJ_SYNACK)) {
530 subflow->reset_reason = MPTCP_RST_EMPTCP;
534 subflow->backup = mp_opt.backup;
535 subflow->thmac = mp_opt.thmac;
536 subflow->remote_nonce = mp_opt.nonce;
537 WRITE_ONCE(subflow->remote_id, mp_opt.join_id);
538 pr_debug("subflow=%p, thmac=%llu, remote_nonce=%u backup=%d",
539 subflow, subflow->thmac, subflow->remote_nonce,
542 if (!subflow_thmac_valid(subflow)) {
543 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINACKMAC);
544 subflow->reset_reason = MPTCP_RST_EMPTCP;
548 if (!mptcp_finish_join(sk))
551 subflow_generate_hmac(subflow->local_key, subflow->remote_key,
552 subflow->local_nonce,
553 subflow->remote_nonce,
555 memcpy(subflow->hmac, hmac, MPTCPOPT_HMAC_LEN);
557 subflow->mp_join = 1;
558 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINSYNACKRX);
560 if (subflow_use_different_dport(msk, sk)) {
561 pr_debug("synack inet_dport=%d %d",
562 ntohs(inet_sk(sk)->inet_dport),
563 ntohs(inet_sk(parent)->inet_dport));
564 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINPORTSYNACKRX);
566 } else if (mptcp_check_fallback(sk)) {
568 mptcp_propagate_state(parent, sk, subflow, NULL);
573 subflow->reset_transient = 0;
574 mptcp_subflow_reset(sk);
577 static void subflow_set_local_id(struct mptcp_subflow_context *subflow, int local_id)
579 WARN_ON_ONCE(local_id < 0 || local_id > 255);
580 WRITE_ONCE(subflow->local_id, local_id);
583 static int subflow_chk_local_id(struct sock *sk)
585 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
586 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
589 if (likely(subflow->local_id >= 0))
592 err = mptcp_pm_get_local_id(msk, (struct sock_common *)sk);
596 subflow_set_local_id(subflow, err);
600 static int subflow_rebuild_header(struct sock *sk)
602 int err = subflow_chk_local_id(sk);
604 if (unlikely(err < 0))
607 return inet_sk_rebuild_header(sk);
610 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
611 static int subflow_v6_rebuild_header(struct sock *sk)
613 int err = subflow_chk_local_id(sk);
615 if (unlikely(err < 0))
618 return inet6_sk_rebuild_header(sk);
622 static struct request_sock_ops mptcp_subflow_v4_request_sock_ops __ro_after_init;
623 static struct tcp_request_sock_ops subflow_request_sock_ipv4_ops __ro_after_init;
625 static int subflow_v4_conn_request(struct sock *sk, struct sk_buff *skb)
627 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
629 pr_debug("subflow=%p", subflow);
631 /* Never answer to SYNs sent to broadcast or multicast */
632 if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
635 return tcp_conn_request(&mptcp_subflow_v4_request_sock_ops,
636 &subflow_request_sock_ipv4_ops,
643 static void subflow_v4_req_destructor(struct request_sock *req)
645 subflow_req_destructor(req);
646 tcp_request_sock_ops.destructor(req);
649 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
650 static struct request_sock_ops mptcp_subflow_v6_request_sock_ops __ro_after_init;
651 static struct tcp_request_sock_ops subflow_request_sock_ipv6_ops __ro_after_init;
652 static struct inet_connection_sock_af_ops subflow_v6_specific __ro_after_init;
653 static struct inet_connection_sock_af_ops subflow_v6m_specific __ro_after_init;
654 static struct proto tcpv6_prot_override __ro_after_init;
656 static int subflow_v6_conn_request(struct sock *sk, struct sk_buff *skb)
658 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
660 pr_debug("subflow=%p", subflow);
662 if (skb->protocol == htons(ETH_P_IP))
663 return subflow_v4_conn_request(sk, skb);
665 if (!ipv6_unicast_destination(skb))
668 if (ipv6_addr_v4mapped(&ipv6_hdr(skb)->saddr)) {
669 __IP6_INC_STATS(sock_net(sk), NULL, IPSTATS_MIB_INHDRERRORS);
673 return tcp_conn_request(&mptcp_subflow_v6_request_sock_ops,
674 &subflow_request_sock_ipv6_ops, sk, skb);
678 return 0; /* don't send reset */
681 static void subflow_v6_req_destructor(struct request_sock *req)
683 subflow_req_destructor(req);
684 tcp6_request_sock_ops.destructor(req);
688 struct request_sock *mptcp_subflow_reqsk_alloc(const struct request_sock_ops *ops,
689 struct sock *sk_listener,
690 bool attach_listener)
692 if (ops->family == AF_INET)
693 ops = &mptcp_subflow_v4_request_sock_ops;
694 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
695 else if (ops->family == AF_INET6)
696 ops = &mptcp_subflow_v6_request_sock_ops;
699 return inet_reqsk_alloc(ops, sk_listener, attach_listener);
701 EXPORT_SYMBOL(mptcp_subflow_reqsk_alloc);
703 /* validate hmac received in third ACK */
704 static bool subflow_hmac_valid(const struct request_sock *req,
705 const struct mptcp_options_received *mp_opt)
707 const struct mptcp_subflow_request_sock *subflow_req;
708 u8 hmac[SHA256_DIGEST_SIZE];
709 struct mptcp_sock *msk;
711 subflow_req = mptcp_subflow_rsk(req);
712 msk = subflow_req->msk;
716 subflow_generate_hmac(READ_ONCE(msk->remote_key),
717 READ_ONCE(msk->local_key),
718 subflow_req->remote_nonce,
719 subflow_req->local_nonce, hmac);
721 return !crypto_memneq(hmac, mp_opt->hmac, MPTCPOPT_HMAC_LEN);
724 static void subflow_ulp_fallback(struct sock *sk,
725 struct mptcp_subflow_context *old_ctx)
727 struct inet_connection_sock *icsk = inet_csk(sk);
729 mptcp_subflow_tcp_fallback(sk, old_ctx);
730 icsk->icsk_ulp_ops = NULL;
731 rcu_assign_pointer(icsk->icsk_ulp_data, NULL);
732 tcp_sk(sk)->is_mptcp = 0;
734 mptcp_subflow_ops_undo_override(sk);
737 void mptcp_subflow_drop_ctx(struct sock *ssk)
739 struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(ssk);
744 list_del(&mptcp_subflow_ctx(ssk)->node);
745 if (inet_csk(ssk)->icsk_ulp_ops) {
746 subflow_ulp_fallback(ssk, ctx);
754 void __mptcp_subflow_fully_established(struct mptcp_sock *msk,
755 struct mptcp_subflow_context *subflow,
756 const struct mptcp_options_received *mp_opt)
758 subflow_set_remote_key(msk, subflow, mp_opt);
759 subflow->fully_established = 1;
760 WRITE_ONCE(msk->fully_established, true);
762 if (subflow->is_mptfo)
763 __mptcp_fastopen_gen_msk_ackseq(msk, subflow, mp_opt);
766 static struct sock *subflow_syn_recv_sock(const struct sock *sk,
768 struct request_sock *req,
769 struct dst_entry *dst,
770 struct request_sock *req_unhash,
773 struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk);
774 struct mptcp_subflow_request_sock *subflow_req;
775 struct mptcp_options_received mp_opt;
776 bool fallback, fallback_is_fatal;
777 struct mptcp_sock *owner;
780 pr_debug("listener=%p, req=%p, conn=%p", listener, req, listener->conn);
782 /* After child creation we must look for MPC even when options
785 mp_opt.suboptions = 0;
787 /* hopefully temporary handling for MP_JOIN+syncookie */
788 subflow_req = mptcp_subflow_rsk(req);
789 fallback_is_fatal = tcp_rsk(req)->is_mptcp && subflow_req->mp_join;
790 fallback = !tcp_rsk(req)->is_mptcp;
794 /* if the sk is MP_CAPABLE, we try to fetch the client key */
795 if (subflow_req->mp_capable) {
796 /* we can receive and accept an in-window, out-of-order pkt,
797 * which may not carry the MP_CAPABLE opt even on mptcp enabled
798 * paths: always try to extract the peer key, and fallback
799 * for packets missing it.
800 * Even OoO DSS packets coming legitly after dropped or
801 * reordered MPC will cause fallback, but we don't have other
804 mptcp_get_options(skb, &mp_opt);
805 if (!(mp_opt.suboptions &
806 (OPTION_MPTCP_MPC_SYN | OPTION_MPTCP_MPC_ACK)))
809 } else if (subflow_req->mp_join) {
810 mptcp_get_options(skb, &mp_opt);
811 if (!(mp_opt.suboptions & OPTION_MPTCP_MPJ_ACK) ||
812 !subflow_hmac_valid(req, &mp_opt) ||
813 !mptcp_can_accept_new_subflow(subflow_req->msk)) {
814 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKMAC);
820 child = listener->icsk_af_ops->syn_recv_sock(sk, skb, req, dst,
821 req_unhash, own_req);
823 if (child && *own_req) {
824 struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(child);
826 tcp_rsk(req)->drop_req = false;
828 /* we need to fallback on ctx allocation failure and on pre-reqs
829 * checking above. In the latter scenario we additionally need
830 * to reset the context to non MPTCP status.
832 if (!ctx || fallback) {
833 if (fallback_is_fatal) {
834 subflow_add_reset_reason(skb, MPTCP_RST_EMPTCP);
840 /* ssk inherits options of listener sk */
841 ctx->setsockopt_seq = listener->setsockopt_seq;
843 if (ctx->mp_capable) {
844 ctx->conn = mptcp_sk_clone_init(listener->conn, &mp_opt, child, req);
849 owner = mptcp_sk(ctx->conn);
850 mptcp_pm_new_connection(owner, child, 1);
852 /* with OoO packets we can reach here without ingress
855 if (mp_opt.suboptions & OPTION_MPTCP_MPC_ACK) {
856 mptcp_pm_fully_established(owner, child);
857 ctx->pm_notified = 1;
859 } else if (ctx->mp_join) {
860 owner = subflow_req->msk;
862 subflow_add_reset_reason(skb, MPTCP_RST_EPROHIBIT);
866 /* move the msk reference ownership to the subflow */
867 subflow_req->msk = NULL;
868 ctx->conn = (struct sock *)owner;
870 if (subflow_use_different_sport(owner, sk)) {
871 pr_debug("ack inet_sport=%d %d",
872 ntohs(inet_sk(sk)->inet_sport),
873 ntohs(inet_sk((struct sock *)owner)->inet_sport));
874 if (!mptcp_pm_sport_in_anno_list(owner, sk)) {
875 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MISMATCHPORTACKRX);
878 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINPORTACKRX);
881 if (!mptcp_finish_join(child))
884 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKRX);
885 tcp_rsk(req)->drop_req = true;
889 /* check for expected invariant - should never trigger, just help
890 * catching eariler subtle bugs
892 WARN_ON_ONCE(child && *own_req && tcp_sk(child)->is_mptcp &&
893 (!mptcp_subflow_ctx(child) ||
894 !mptcp_subflow_ctx(child)->conn));
898 mptcp_subflow_drop_ctx(child);
899 tcp_rsk(req)->drop_req = true;
900 inet_csk_prepare_for_destroy_sock(child);
902 req->rsk_ops->send_reset(sk, skb);
904 /* The last child reference will be released by the caller */
908 mptcp_subflow_drop_ctx(child);
912 static struct inet_connection_sock_af_ops subflow_specific __ro_after_init;
913 static struct proto tcp_prot_override __ro_after_init;
915 enum mapping_status {
924 static void dbg_bad_map(struct mptcp_subflow_context *subflow, u32 ssn)
926 pr_debug("Bad mapping: ssn=%d map_seq=%d map_data_len=%d",
927 ssn, subflow->map_subflow_seq, subflow->map_data_len);
930 static bool skb_is_fully_mapped(struct sock *ssk, struct sk_buff *skb)
932 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
933 unsigned int skb_consumed;
935 skb_consumed = tcp_sk(ssk)->copied_seq - TCP_SKB_CB(skb)->seq;
936 if (WARN_ON_ONCE(skb_consumed >= skb->len))
939 return skb->len - skb_consumed <= subflow->map_data_len -
940 mptcp_subflow_get_map_offset(subflow);
943 static bool validate_mapping(struct sock *ssk, struct sk_buff *skb)
945 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
946 u32 ssn = tcp_sk(ssk)->copied_seq - subflow->ssn_offset;
948 if (unlikely(before(ssn, subflow->map_subflow_seq))) {
949 /* Mapping covers data later in the subflow stream,
950 * currently unsupported.
952 dbg_bad_map(subflow, ssn);
955 if (unlikely(!before(ssn, subflow->map_subflow_seq +
956 subflow->map_data_len))) {
957 /* Mapping does covers past subflow data, invalid */
958 dbg_bad_map(subflow, ssn);
964 static enum mapping_status validate_data_csum(struct sock *ssk, struct sk_buff *skb,
967 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
968 u32 offset, seq, delta;
975 /* mapping already validated on previous traversal */
976 if (subflow->map_csum_len == subflow->map_data_len)
979 /* traverse the receive queue, ensuring it contains a full
980 * DSS mapping and accumulating the related csum.
981 * Preserve the accoumlate csum across multiple calls, to compute
984 delta = subflow->map_data_len - subflow->map_csum_len;
986 seq = tcp_sk(ssk)->copied_seq + subflow->map_csum_len;
987 offset = seq - TCP_SKB_CB(skb)->seq;
989 /* if the current skb has not been accounted yet, csum its contents
990 * up to the amount covered by the current DSS
992 if (offset < skb->len) {
995 len = min(skb->len - offset, delta);
996 csum = skb_checksum(skb, offset, len, 0);
997 subflow->map_data_csum = csum_block_add(subflow->map_data_csum, csum,
998 subflow->map_csum_len);
1001 subflow->map_csum_len += len;
1006 if (skb_queue_is_last(&ssk->sk_receive_queue, skb)) {
1007 /* if this subflow is closed, the partial mapping
1008 * will be never completed; flush the pending skbs, so
1009 * that subflow_sched_work_if_closed() can kick in
1011 if (unlikely(ssk->sk_state == TCP_CLOSE))
1012 while ((skb = skb_peek(&ssk->sk_receive_queue)))
1013 sk_eat_skb(ssk, skb);
1015 /* not enough data to validate the csum */
1016 return MAPPING_EMPTY;
1019 /* the DSS mapping for next skbs will be validated later,
1020 * when a get_mapping_status call will process such skb
1025 /* note that 'map_data_len' accounts only for the carried data, does
1026 * not include the eventual seq increment due to the data fin,
1027 * while the pseudo header requires the original DSS data len,
1030 csum = __mptcp_make_csum(subflow->map_seq,
1031 subflow->map_subflow_seq,
1032 subflow->map_data_len + subflow->map_data_fin,
1033 subflow->map_data_csum);
1034 if (unlikely(csum)) {
1035 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DATACSUMERR);
1036 return MAPPING_BAD_CSUM;
1039 subflow->valid_csum_seen = 1;
1043 static enum mapping_status get_mapping_status(struct sock *ssk,
1044 struct mptcp_sock *msk)
1046 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1047 bool csum_reqd = READ_ONCE(msk->csum_enabled);
1048 struct mptcp_ext *mpext;
1049 struct sk_buff *skb;
1053 skb = skb_peek(&ssk->sk_receive_queue);
1055 return MAPPING_EMPTY;
1057 if (mptcp_check_fallback(ssk))
1058 return MAPPING_DUMMY;
1060 mpext = mptcp_get_ext(skb);
1061 if (!mpext || !mpext->use_map) {
1062 if (!subflow->map_valid && !skb->len) {
1063 /* the TCP stack deliver 0 len FIN pkt to the receive
1064 * queue, that is the only 0len pkts ever expected here,
1065 * and we can admit no mapping only for 0 len pkts
1067 if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN))
1068 WARN_ONCE(1, "0len seq %d:%d flags %x",
1069 TCP_SKB_CB(skb)->seq,
1070 TCP_SKB_CB(skb)->end_seq,
1071 TCP_SKB_CB(skb)->tcp_flags);
1072 sk_eat_skb(ssk, skb);
1073 return MAPPING_EMPTY;
1076 if (!subflow->map_valid)
1077 return MAPPING_INVALID;
1082 trace_get_mapping_status(mpext);
1084 data_len = mpext->data_len;
1085 if (data_len == 0) {
1086 pr_debug("infinite mapping received");
1087 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_INFINITEMAPRX);
1088 subflow->map_data_len = 0;
1089 return MAPPING_INVALID;
1092 if (mpext->data_fin == 1) {
1093 if (data_len == 1) {
1094 bool updated = mptcp_update_rcv_data_fin(msk, mpext->data_seq,
1096 pr_debug("DATA_FIN with no payload seq=%llu", mpext->data_seq);
1097 if (subflow->map_valid) {
1098 /* A DATA_FIN might arrive in a DSS
1099 * option before the previous mapping
1100 * has been fully consumed. Continue
1101 * handling the existing mapping.
1103 skb_ext_del(skb, SKB_EXT_MPTCP);
1107 mptcp_schedule_work((struct sock *)msk);
1109 return MAPPING_DATA_FIN;
1112 u64 data_fin_seq = mpext->data_seq + data_len - 1;
1114 /* If mpext->data_seq is a 32-bit value, data_fin_seq
1115 * must also be limited to 32 bits.
1118 data_fin_seq &= GENMASK_ULL(31, 0);
1120 mptcp_update_rcv_data_fin(msk, data_fin_seq, mpext->dsn64);
1121 pr_debug("DATA_FIN with mapping seq=%llu dsn64=%d",
1122 data_fin_seq, mpext->dsn64);
1125 /* Adjust for DATA_FIN using 1 byte of sequence space */
1129 map_seq = mptcp_expand_seq(READ_ONCE(msk->ack_seq), mpext->data_seq, mpext->dsn64);
1130 WRITE_ONCE(mptcp_sk(subflow->conn)->use_64bit_ack, !!mpext->dsn64);
1132 if (subflow->map_valid) {
1133 /* Allow replacing only with an identical map */
1134 if (subflow->map_seq == map_seq &&
1135 subflow->map_subflow_seq == mpext->subflow_seq &&
1136 subflow->map_data_len == data_len &&
1137 subflow->map_csum_reqd == mpext->csum_reqd) {
1138 skb_ext_del(skb, SKB_EXT_MPTCP);
1142 /* If this skb data are fully covered by the current mapping,
1143 * the new map would need caching, which is not supported
1145 if (skb_is_fully_mapped(ssk, skb)) {
1146 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DSSNOMATCH);
1147 return MAPPING_INVALID;
1150 /* will validate the next map after consuming the current one */
1154 subflow->map_seq = map_seq;
1155 subflow->map_subflow_seq = mpext->subflow_seq;
1156 subflow->map_data_len = data_len;
1157 subflow->map_valid = 1;
1158 subflow->map_data_fin = mpext->data_fin;
1159 subflow->mpc_map = mpext->mpc_map;
1160 subflow->map_csum_reqd = mpext->csum_reqd;
1161 subflow->map_csum_len = 0;
1162 subflow->map_data_csum = csum_unfold(mpext->csum);
1164 /* Cfr RFC 8684 Section 3.3.0 */
1165 if (unlikely(subflow->map_csum_reqd != csum_reqd))
1166 return MAPPING_INVALID;
1168 pr_debug("new map seq=%llu subflow_seq=%u data_len=%u csum=%d:%u",
1169 subflow->map_seq, subflow->map_subflow_seq,
1170 subflow->map_data_len, subflow->map_csum_reqd,
1171 subflow->map_data_csum);
1174 /* we revalidate valid mapping on new skb, because we must ensure
1175 * the current skb is completely covered by the available mapping
1177 if (!validate_mapping(ssk, skb)) {
1178 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DSSTCPMISMATCH);
1179 return MAPPING_INVALID;
1182 skb_ext_del(skb, SKB_EXT_MPTCP);
1185 return validate_data_csum(ssk, skb, csum_reqd);
1188 static void mptcp_subflow_discard_data(struct sock *ssk, struct sk_buff *skb,
1191 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1192 bool fin = TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN;
1195 incr = limit >= skb->len ? skb->len + fin : limit;
1197 pr_debug("discarding=%d len=%d seq=%d", incr, skb->len,
1198 subflow->map_subflow_seq);
1199 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DUPDATA);
1200 tcp_sk(ssk)->copied_seq += incr;
1201 if (!before(tcp_sk(ssk)->copied_seq, TCP_SKB_CB(skb)->end_seq))
1202 sk_eat_skb(ssk, skb);
1203 if (mptcp_subflow_get_map_offset(subflow) >= subflow->map_data_len)
1204 subflow->map_valid = 0;
1207 /* sched mptcp worker to remove the subflow if no more data is pending */
1208 static void subflow_sched_work_if_closed(struct mptcp_sock *msk, struct sock *ssk)
1210 if (likely(ssk->sk_state != TCP_CLOSE))
1213 if (skb_queue_empty(&ssk->sk_receive_queue) &&
1214 !test_and_set_bit(MPTCP_WORK_CLOSE_SUBFLOW, &msk->flags))
1215 mptcp_schedule_work((struct sock *)msk);
1218 static bool subflow_can_fallback(struct mptcp_subflow_context *subflow)
1220 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
1222 if (subflow->mp_join)
1224 else if (READ_ONCE(msk->csum_enabled))
1225 return !subflow->valid_csum_seen;
1227 return !subflow->fully_established;
1230 static void mptcp_subflow_fail(struct mptcp_sock *msk, struct sock *ssk)
1232 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1233 unsigned long fail_tout;
1235 /* greceful failure can happen only on the MPC subflow */
1236 if (WARN_ON_ONCE(ssk != READ_ONCE(msk->first)))
1239 /* since the close timeout take precedence on the fail one,
1240 * no need to start the latter when the first is already set
1242 if (sock_flag((struct sock *)msk, SOCK_DEAD))
1245 /* we don't need extreme accuracy here, use a zero fail_tout as special
1246 * value meaning no fail timeout at all;
1248 fail_tout = jiffies + TCP_RTO_MAX;
1251 WRITE_ONCE(subflow->fail_tout, fail_tout);
1254 mptcp_reset_tout_timer(msk, subflow->fail_tout);
1257 static bool subflow_check_data_avail(struct sock *ssk)
1259 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1260 enum mapping_status status;
1261 struct mptcp_sock *msk;
1262 struct sk_buff *skb;
1264 if (!skb_peek(&ssk->sk_receive_queue))
1265 WRITE_ONCE(subflow->data_avail, false);
1266 if (subflow->data_avail)
1269 msk = mptcp_sk(subflow->conn);
1274 status = get_mapping_status(ssk, msk);
1275 trace_subflow_check_data_avail(status, skb_peek(&ssk->sk_receive_queue));
1276 if (unlikely(status == MAPPING_INVALID || status == MAPPING_DUMMY ||
1277 status == MAPPING_BAD_CSUM))
1280 if (status != MAPPING_OK)
1283 skb = skb_peek(&ssk->sk_receive_queue);
1284 if (WARN_ON_ONCE(!skb))
1287 if (unlikely(!READ_ONCE(msk->can_ack)))
1290 old_ack = READ_ONCE(msk->ack_seq);
1291 ack_seq = mptcp_subflow_get_mapped_dsn(subflow);
1292 pr_debug("msk ack_seq=%llx subflow ack_seq=%llx", old_ack,
1294 if (unlikely(before64(ack_seq, old_ack))) {
1295 mptcp_subflow_discard_data(ssk, skb, old_ack - ack_seq);
1299 WRITE_ONCE(subflow->data_avail, true);
1305 subflow_sched_work_if_closed(msk, ssk);
1309 if (!__mptcp_check_fallback(msk)) {
1310 /* RFC 8684 section 3.7. */
1311 if (status == MAPPING_BAD_CSUM &&
1312 (subflow->mp_join || subflow->valid_csum_seen)) {
1313 subflow->send_mp_fail = 1;
1315 if (!READ_ONCE(msk->allow_infinite_fallback)) {
1316 subflow->reset_transient = 0;
1317 subflow->reset_reason = MPTCP_RST_EMIDDLEBOX;
1320 mptcp_subflow_fail(msk, ssk);
1321 WRITE_ONCE(subflow->data_avail, true);
1325 if (!subflow_can_fallback(subflow) && subflow->map_data_len) {
1326 /* fatal protocol error, close the socket.
1327 * subflow_error_report() will introduce the appropriate barriers
1329 subflow->reset_transient = 0;
1330 subflow->reset_reason = MPTCP_RST_EMPTCP;
1333 WRITE_ONCE(ssk->sk_err, EBADMSG);
1334 tcp_set_state(ssk, TCP_CLOSE);
1335 while ((skb = skb_peek(&ssk->sk_receive_queue)))
1336 sk_eat_skb(ssk, skb);
1337 tcp_send_active_reset(ssk, GFP_ATOMIC);
1338 WRITE_ONCE(subflow->data_avail, false);
1342 mptcp_do_fallback(ssk);
1345 skb = skb_peek(&ssk->sk_receive_queue);
1346 subflow->map_valid = 1;
1347 subflow->map_seq = READ_ONCE(msk->ack_seq);
1348 subflow->map_data_len = skb->len;
1349 subflow->map_subflow_seq = tcp_sk(ssk)->copied_seq - subflow->ssn_offset;
1350 WRITE_ONCE(subflow->data_avail, true);
1354 bool mptcp_subflow_data_available(struct sock *sk)
1356 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1358 /* check if current mapping is still valid */
1359 if (subflow->map_valid &&
1360 mptcp_subflow_get_map_offset(subflow) >= subflow->map_data_len) {
1361 subflow->map_valid = 0;
1362 WRITE_ONCE(subflow->data_avail, false);
1364 pr_debug("Done with mapping: seq=%u data_len=%u",
1365 subflow->map_subflow_seq,
1366 subflow->map_data_len);
1369 return subflow_check_data_avail(sk);
1372 /* If ssk has an mptcp parent socket, use the mptcp rcvbuf occupancy,
1375 * In mptcp, rwin is about the mptcp-level connection data.
1377 * Data that is still on the ssk rx queue can thus be ignored,
1378 * as far as mptcp peer is concerned that data is still inflight.
1379 * DSS ACK is updated when skb is moved to the mptcp rx queue.
1381 void mptcp_space(const struct sock *ssk, int *space, int *full_space)
1383 const struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1384 const struct sock *sk = subflow->conn;
1386 *space = __mptcp_space(sk);
1387 *full_space = mptcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf));
1390 static void subflow_error_report(struct sock *ssk)
1392 struct sock *sk = mptcp_subflow_ctx(ssk)->conn;
1394 /* bail early if this is a no-op, so that we avoid introducing a
1395 * problematic lockdep dependency between TCP accept queue lock
1396 * and msk socket spinlock
1401 mptcp_data_lock(sk);
1402 if (!sock_owned_by_user(sk))
1403 __mptcp_error_report(sk);
1405 __set_bit(MPTCP_ERROR_REPORT, &mptcp_sk(sk)->cb_flags);
1406 mptcp_data_unlock(sk);
1409 static void subflow_data_ready(struct sock *sk)
1411 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1412 u16 state = 1 << inet_sk_state_load(sk);
1413 struct sock *parent = subflow->conn;
1414 struct mptcp_sock *msk;
1416 trace_sk_data_ready(sk);
1418 msk = mptcp_sk(parent);
1419 if (state & TCPF_LISTEN) {
1420 /* MPJ subflow are removed from accept queue before reaching here,
1421 * avoid stray wakeups
1423 if (reqsk_queue_empty(&inet_csk(sk)->icsk_accept_queue))
1426 parent->sk_data_ready(parent);
1430 WARN_ON_ONCE(!__mptcp_check_fallback(msk) && !subflow->mp_capable &&
1431 !subflow->mp_join && !(state & TCPF_CLOSE));
1433 if (mptcp_subflow_data_available(sk)) {
1434 mptcp_data_ready(parent, sk);
1436 /* subflow-level lowat test are not relevant.
1437 * respect the msk-level threshold eventually mandating an immediate ack
1439 if (mptcp_data_avail(msk) < parent->sk_rcvlowat &&
1440 (tcp_sk(sk)->rcv_nxt - tcp_sk(sk)->rcv_wup) > inet_csk(sk)->icsk_ack.rcv_mss)
1441 inet_csk(sk)->icsk_ack.pending |= ICSK_ACK_NOW;
1442 } else if (unlikely(sk->sk_err)) {
1443 subflow_error_report(sk);
1447 static void subflow_write_space(struct sock *ssk)
1449 struct sock *sk = mptcp_subflow_ctx(ssk)->conn;
1451 mptcp_propagate_sndbuf(sk, ssk);
1452 mptcp_write_space(sk);
1455 static const struct inet_connection_sock_af_ops *
1456 subflow_default_af_ops(struct sock *sk)
1458 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1459 if (sk->sk_family == AF_INET6)
1460 return &subflow_v6_specific;
1462 return &subflow_specific;
1465 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1466 void mptcpv6_handle_mapped(struct sock *sk, bool mapped)
1468 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1469 struct inet_connection_sock *icsk = inet_csk(sk);
1470 const struct inet_connection_sock_af_ops *target;
1472 target = mapped ? &subflow_v6m_specific : subflow_default_af_ops(sk);
1474 pr_debug("subflow=%p family=%d ops=%p target=%p mapped=%d",
1475 subflow, sk->sk_family, icsk->icsk_af_ops, target, mapped);
1477 if (likely(icsk->icsk_af_ops == target))
1480 subflow->icsk_af_ops = icsk->icsk_af_ops;
1481 icsk->icsk_af_ops = target;
1485 void mptcp_info2sockaddr(const struct mptcp_addr_info *info,
1486 struct sockaddr_storage *addr,
1487 unsigned short family)
1489 memset(addr, 0, sizeof(*addr));
1490 addr->ss_family = family;
1491 if (addr->ss_family == AF_INET) {
1492 struct sockaddr_in *in_addr = (struct sockaddr_in *)addr;
1494 if (info->family == AF_INET)
1495 in_addr->sin_addr = info->addr;
1496 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1497 else if (ipv6_addr_v4mapped(&info->addr6))
1498 in_addr->sin_addr.s_addr = info->addr6.s6_addr32[3];
1500 in_addr->sin_port = info->port;
1502 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1503 else if (addr->ss_family == AF_INET6) {
1504 struct sockaddr_in6 *in6_addr = (struct sockaddr_in6 *)addr;
1506 if (info->family == AF_INET)
1507 ipv6_addr_set_v4mapped(info->addr.s_addr,
1508 &in6_addr->sin6_addr);
1510 in6_addr->sin6_addr = info->addr6;
1511 in6_addr->sin6_port = info->port;
1516 int __mptcp_subflow_connect(struct sock *sk, const struct mptcp_addr_info *loc,
1517 const struct mptcp_addr_info *remote)
1519 struct mptcp_sock *msk = mptcp_sk(sk);
1520 struct mptcp_subflow_context *subflow;
1521 struct sockaddr_storage addr;
1522 int remote_id = remote->id;
1523 int local_id = loc->id;
1524 int err = -ENOTCONN;
1532 if (!mptcp_is_fully_established(sk))
1535 err = mptcp_subflow_create_socket(sk, loc->family, &sf);
1540 subflow = mptcp_subflow_ctx(ssk);
1542 get_random_bytes(&subflow->local_nonce, sizeof(u32));
1543 } while (!subflow->local_nonce);
1546 subflow_set_local_id(subflow, local_id);
1548 mptcp_pm_get_flags_and_ifindex_by_id(msk, local_id,
1550 subflow->remote_key_valid = 1;
1551 subflow->remote_key = READ_ONCE(msk->remote_key);
1552 subflow->local_key = READ_ONCE(msk->local_key);
1553 subflow->token = msk->token;
1554 mptcp_info2sockaddr(loc, &addr, ssk->sk_family);
1556 addrlen = sizeof(struct sockaddr_in);
1557 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1558 if (addr.ss_family == AF_INET6)
1559 addrlen = sizeof(struct sockaddr_in6);
1561 ssk->sk_bound_dev_if = ifindex;
1562 err = kernel_bind(sf, (struct sockaddr *)&addr, addrlen);
1566 mptcp_crypto_key_sha(subflow->remote_key, &remote_token, NULL);
1567 pr_debug("msk=%p remote_token=%u local_id=%d remote_id=%d", msk,
1568 remote_token, local_id, remote_id);
1569 subflow->remote_token = remote_token;
1570 WRITE_ONCE(subflow->remote_id, remote_id);
1571 subflow->request_join = 1;
1572 subflow->request_bkup = !!(flags & MPTCP_PM_ADDR_FLAG_BACKUP);
1573 subflow->subflow_id = msk->subflow_id++;
1574 mptcp_info2sockaddr(remote, &addr, ssk->sk_family);
1577 list_add_tail(&subflow->node, &msk->conn_list);
1578 err = kernel_connect(sf, (struct sockaddr *)&addr, addrlen, O_NONBLOCK);
1579 if (err && err != -EINPROGRESS)
1582 /* discard the subflow socket */
1583 mptcp_sock_graft(ssk, sk->sk_socket);
1584 iput(SOCK_INODE(sf));
1585 WRITE_ONCE(msk->allow_infinite_fallback, false);
1586 mptcp_stop_tout_timer(sk);
1590 list_del(&subflow->node);
1591 sock_put(mptcp_subflow_tcp_sock(subflow));
1594 subflow->disposable = 1;
1598 /* we account subflows before the creation, and this failures will not
1599 * be caught by sk_state_change()
1601 mptcp_pm_close_subflow(msk);
1605 static void mptcp_attach_cgroup(struct sock *parent, struct sock *child)
1607 #ifdef CONFIG_SOCK_CGROUP_DATA
1608 struct sock_cgroup_data *parent_skcd = &parent->sk_cgrp_data,
1609 *child_skcd = &child->sk_cgrp_data;
1611 /* only the additional subflows created by kworkers have to be modified */
1612 if (cgroup_id(sock_cgroup_ptr(parent_skcd)) !=
1613 cgroup_id(sock_cgroup_ptr(child_skcd))) {
1615 struct mem_cgroup *memcg = parent->sk_memcg;
1617 mem_cgroup_sk_free(child);
1618 if (memcg && css_tryget(&memcg->css))
1619 child->sk_memcg = memcg;
1620 #endif /* CONFIG_MEMCG */
1622 cgroup_sk_free(child_skcd);
1623 *child_skcd = *parent_skcd;
1624 cgroup_sk_clone(child_skcd);
1626 #endif /* CONFIG_SOCK_CGROUP_DATA */
1629 static void mptcp_subflow_ops_override(struct sock *ssk)
1631 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1632 if (ssk->sk_prot == &tcpv6_prot)
1633 ssk->sk_prot = &tcpv6_prot_override;
1636 ssk->sk_prot = &tcp_prot_override;
1639 static void mptcp_subflow_ops_undo_override(struct sock *ssk)
1641 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1642 if (ssk->sk_prot == &tcpv6_prot_override)
1643 ssk->sk_prot = &tcpv6_prot;
1646 ssk->sk_prot = &tcp_prot;
1649 int mptcp_subflow_create_socket(struct sock *sk, unsigned short family,
1650 struct socket **new_sock)
1652 struct mptcp_subflow_context *subflow;
1653 struct net *net = sock_net(sk);
1657 /* un-accepted server sockets can reach here - on bad configuration
1658 * bail early to avoid greater trouble later
1660 if (unlikely(!sk->sk_socket))
1663 err = sock_create_kern(net, family, SOCK_STREAM, IPPROTO_TCP, &sf);
1667 lock_sock_nested(sf->sk, SINGLE_DEPTH_NESTING);
1669 err = security_mptcp_add_subflow(sk, sf->sk);
1673 /* the newly created socket has to be in the same cgroup as its parent */
1674 mptcp_attach_cgroup(sk, sf->sk);
1676 /* kernel sockets do not by default acquire net ref, but TCP timer
1678 * Update ns_tracker to current stack trace and refcounted tracker.
1680 __netns_tracker_free(net, &sf->sk->ns_tracker, false);
1681 sf->sk->sk_net_refcnt = 1;
1682 get_net_track(net, &sf->sk->ns_tracker, GFP_KERNEL);
1683 sock_inuse_add(net, 1);
1684 err = tcp_set_ulp(sf->sk, "mptcp");
1688 mptcp_sockopt_sync_locked(mptcp_sk(sk), sf->sk);
1689 release_sock(sf->sk);
1691 /* the newly created socket really belongs to the owning MPTCP master
1692 * socket, even if for additional subflows the allocation is performed
1693 * by a kernel workqueue. Adjust inode references, so that the
1694 * procfs/diag interfaces really show this one belonging to the correct
1697 SOCK_INODE(sf)->i_ino = SOCK_INODE(sk->sk_socket)->i_ino;
1698 SOCK_INODE(sf)->i_uid = SOCK_INODE(sk->sk_socket)->i_uid;
1699 SOCK_INODE(sf)->i_gid = SOCK_INODE(sk->sk_socket)->i_gid;
1701 subflow = mptcp_subflow_ctx(sf->sk);
1702 pr_debug("subflow=%p", subflow);
1707 mptcp_subflow_ops_override(sf->sk);
1712 release_sock(sf->sk);
1717 static struct mptcp_subflow_context *subflow_create_ctx(struct sock *sk,
1720 struct inet_connection_sock *icsk = inet_csk(sk);
1721 struct mptcp_subflow_context *ctx;
1723 ctx = kzalloc(sizeof(*ctx), priority);
1727 rcu_assign_pointer(icsk->icsk_ulp_data, ctx);
1728 INIT_LIST_HEAD(&ctx->node);
1729 INIT_LIST_HEAD(&ctx->delegated_node);
1731 pr_debug("subflow=%p", ctx);
1734 WRITE_ONCE(ctx->local_id, -1);
1739 static void __subflow_state_change(struct sock *sk)
1741 struct socket_wq *wq;
1744 wq = rcu_dereference(sk->sk_wq);
1745 if (skwq_has_sleeper(wq))
1746 wake_up_interruptible_all(&wq->wait);
1750 static bool subflow_is_done(const struct sock *sk)
1752 return sk->sk_shutdown & RCV_SHUTDOWN || sk->sk_state == TCP_CLOSE;
1755 static void subflow_state_change(struct sock *sk)
1757 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1758 struct sock *parent = subflow->conn;
1759 struct mptcp_sock *msk;
1761 __subflow_state_change(sk);
1763 msk = mptcp_sk(parent);
1764 if (subflow_simultaneous_connect(sk)) {
1765 mptcp_do_fallback(sk);
1767 subflow->conn_finished = 1;
1768 mptcp_propagate_state(parent, sk, subflow, NULL);
1771 /* as recvmsg() does not acquire the subflow socket for ssk selection
1772 * a fin packet carrying a DSS can be unnoticed if we don't trigger
1773 * the data available machinery here.
1775 if (mptcp_subflow_data_available(sk))
1776 mptcp_data_ready(parent, sk);
1777 else if (unlikely(sk->sk_err))
1778 subflow_error_report(sk);
1780 subflow_sched_work_if_closed(mptcp_sk(parent), sk);
1782 /* when the fallback subflow closes the rx side, trigger a 'dummy'
1783 * ingress data fin, so that the msk state will follow along
1785 if (__mptcp_check_fallback(msk) && subflow_is_done(sk) && msk->first == sk &&
1786 mptcp_update_rcv_data_fin(msk, READ_ONCE(msk->ack_seq), true))
1787 mptcp_schedule_work(parent);
1790 void mptcp_subflow_queue_clean(struct sock *listener_sk, struct sock *listener_ssk)
1792 struct request_sock_queue *queue = &inet_csk(listener_ssk)->icsk_accept_queue;
1793 struct request_sock *req, *head, *tail;
1794 struct mptcp_subflow_context *subflow;
1795 struct sock *sk, *ssk;
1797 /* Due to lock dependencies no relevant lock can be acquired under rskq_lock.
1798 * Splice the req list, so that accept() can not reach the pending ssk after
1799 * the listener socket is released below.
1801 spin_lock_bh(&queue->rskq_lock);
1802 head = queue->rskq_accept_head;
1803 tail = queue->rskq_accept_tail;
1804 queue->rskq_accept_head = NULL;
1805 queue->rskq_accept_tail = NULL;
1806 spin_unlock_bh(&queue->rskq_lock);
1811 /* can't acquire the msk socket lock under the subflow one,
1812 * or will cause ABBA deadlock
1814 release_sock(listener_ssk);
1816 for (req = head; req; req = req->dl_next) {
1818 if (!sk_is_mptcp(ssk))
1821 subflow = mptcp_subflow_ctx(ssk);
1822 if (!subflow || !subflow->conn)
1828 lock_sock_nested(sk, SINGLE_DEPTH_NESTING);
1829 __mptcp_unaccepted_force_close(sk);
1832 /* lockdep will report a false positive ABBA deadlock
1833 * between cancel_work_sync and the listener socket.
1834 * The involved locks belong to different sockets WRT
1835 * the existing AB chain.
1836 * Using a per socket key is problematic as key
1837 * deregistration requires process context and must be
1838 * performed at socket disposal time, in atomic
1840 * Just tell lockdep to consider the listener socket
1843 mutex_release(&listener_sk->sk_lock.dep_map, _RET_IP_);
1844 mptcp_cancel_work(sk);
1845 mutex_acquire(&listener_sk->sk_lock.dep_map, 0, 0, _RET_IP_);
1850 /* we are still under the listener msk socket lock */
1851 lock_sock_nested(listener_ssk, SINGLE_DEPTH_NESTING);
1853 /* restore the listener queue, to let the TCP code clean it up */
1854 spin_lock_bh(&queue->rskq_lock);
1855 WARN_ON_ONCE(queue->rskq_accept_head);
1856 queue->rskq_accept_head = head;
1857 queue->rskq_accept_tail = tail;
1858 spin_unlock_bh(&queue->rskq_lock);
1861 static int subflow_ulp_init(struct sock *sk)
1863 struct inet_connection_sock *icsk = inet_csk(sk);
1864 struct mptcp_subflow_context *ctx;
1865 struct tcp_sock *tp = tcp_sk(sk);
1868 /* disallow attaching ULP to a socket unless it has been
1869 * created with sock_create_kern()
1871 if (!sk->sk_kern_sock) {
1876 ctx = subflow_create_ctx(sk, GFP_KERNEL);
1882 pr_debug("subflow=%p, family=%d", ctx, sk->sk_family);
1885 ctx->icsk_af_ops = icsk->icsk_af_ops;
1886 icsk->icsk_af_ops = subflow_default_af_ops(sk);
1887 ctx->tcp_state_change = sk->sk_state_change;
1888 ctx->tcp_error_report = sk->sk_error_report;
1890 WARN_ON_ONCE(sk->sk_data_ready != sock_def_readable);
1891 WARN_ON_ONCE(sk->sk_write_space != sk_stream_write_space);
1893 sk->sk_data_ready = subflow_data_ready;
1894 sk->sk_write_space = subflow_write_space;
1895 sk->sk_state_change = subflow_state_change;
1896 sk->sk_error_report = subflow_error_report;
1901 static void subflow_ulp_release(struct sock *ssk)
1903 struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(ssk);
1904 bool release = true;
1912 /* if the msk has been orphaned, keep the ctx
1913 * alive, will be freed by __mptcp_close_ssk(),
1914 * when the subflow is still unaccepted
1916 release = ctx->disposable || list_empty(&ctx->node);
1918 /* inet_child_forget() does not call sk_state_change(),
1919 * explicitly trigger the socket close machinery
1921 if (!release && !test_and_set_bit(MPTCP_WORK_CLOSE_SUBFLOW,
1922 &mptcp_sk(sk)->flags))
1923 mptcp_schedule_work(sk);
1927 mptcp_subflow_ops_undo_override(ssk);
1929 kfree_rcu(ctx, rcu);
1932 static void subflow_ulp_clone(const struct request_sock *req,
1934 const gfp_t priority)
1936 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
1937 struct mptcp_subflow_context *old_ctx = mptcp_subflow_ctx(newsk);
1938 struct mptcp_subflow_context *new_ctx;
1940 if (!tcp_rsk(req)->is_mptcp ||
1941 (!subflow_req->mp_capable && !subflow_req->mp_join)) {
1942 subflow_ulp_fallback(newsk, old_ctx);
1946 new_ctx = subflow_create_ctx(newsk, priority);
1948 subflow_ulp_fallback(newsk, old_ctx);
1952 new_ctx->conn_finished = 1;
1953 new_ctx->icsk_af_ops = old_ctx->icsk_af_ops;
1954 new_ctx->tcp_state_change = old_ctx->tcp_state_change;
1955 new_ctx->tcp_error_report = old_ctx->tcp_error_report;
1956 new_ctx->rel_write_seq = 1;
1957 new_ctx->tcp_sock = newsk;
1959 if (subflow_req->mp_capable) {
1960 /* see comments in subflow_syn_recv_sock(), MPTCP connection
1961 * is fully established only after we receive the remote key
1963 new_ctx->mp_capable = 1;
1964 new_ctx->local_key = subflow_req->local_key;
1965 new_ctx->token = subflow_req->token;
1966 new_ctx->ssn_offset = subflow_req->ssn_offset;
1967 new_ctx->idsn = subflow_req->idsn;
1969 /* this is the first subflow, id is always 0 */
1970 subflow_set_local_id(new_ctx, 0);
1971 } else if (subflow_req->mp_join) {
1972 new_ctx->ssn_offset = subflow_req->ssn_offset;
1973 new_ctx->mp_join = 1;
1974 new_ctx->fully_established = 1;
1975 new_ctx->remote_key_valid = 1;
1976 new_ctx->backup = subflow_req->backup;
1977 WRITE_ONCE(new_ctx->remote_id, subflow_req->remote_id);
1978 new_ctx->token = subflow_req->token;
1979 new_ctx->thmac = subflow_req->thmac;
1981 /* the subflow req id is valid, fetched via subflow_check_req()
1982 * and subflow_token_join_request()
1984 subflow_set_local_id(new_ctx, subflow_req->local_id);
1988 static void tcp_release_cb_override(struct sock *ssk)
1990 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1993 /* process and clear all the pending actions, but leave the subflow into
1994 * the napi queue. To respect locking, only the same CPU that originated
1995 * the action can touch the list. mptcp_napi_poll will take care of it.
1997 status = set_mask_bits(&subflow->delegated_status, MPTCP_DELEGATE_ACTIONS_MASK, 0);
1999 mptcp_subflow_process_delegated(ssk, status);
2001 tcp_release_cb(ssk);
2004 static int tcp_abort_override(struct sock *ssk, int err)
2006 /* closing a listener subflow requires a great deal of care.
2007 * keep it simple and just prevent such operation
2009 if (inet_sk_state_load(ssk) == TCP_LISTEN)
2012 return tcp_abort(ssk, err);
2015 static struct tcp_ulp_ops subflow_ulp_ops __read_mostly = {
2017 .owner = THIS_MODULE,
2018 .init = subflow_ulp_init,
2019 .release = subflow_ulp_release,
2020 .clone = subflow_ulp_clone,
2023 static int subflow_ops_init(struct request_sock_ops *subflow_ops)
2025 subflow_ops->obj_size = sizeof(struct mptcp_subflow_request_sock);
2027 subflow_ops->slab = kmem_cache_create(subflow_ops->slab_name,
2028 subflow_ops->obj_size, 0,
2030 SLAB_TYPESAFE_BY_RCU,
2032 if (!subflow_ops->slab)
2038 void __init mptcp_subflow_init(void)
2040 mptcp_subflow_v4_request_sock_ops = tcp_request_sock_ops;
2041 mptcp_subflow_v4_request_sock_ops.slab_name = "request_sock_subflow_v4";
2042 mptcp_subflow_v4_request_sock_ops.destructor = subflow_v4_req_destructor;
2044 if (subflow_ops_init(&mptcp_subflow_v4_request_sock_ops) != 0)
2045 panic("MPTCP: failed to init subflow v4 request sock ops\n");
2047 subflow_request_sock_ipv4_ops = tcp_request_sock_ipv4_ops;
2048 subflow_request_sock_ipv4_ops.route_req = subflow_v4_route_req;
2049 subflow_request_sock_ipv4_ops.send_synack = subflow_v4_send_synack;
2051 subflow_specific = ipv4_specific;
2052 subflow_specific.conn_request = subflow_v4_conn_request;
2053 subflow_specific.syn_recv_sock = subflow_syn_recv_sock;
2054 subflow_specific.sk_rx_dst_set = subflow_finish_connect;
2055 subflow_specific.rebuild_header = subflow_rebuild_header;
2057 tcp_prot_override = tcp_prot;
2058 tcp_prot_override.release_cb = tcp_release_cb_override;
2059 tcp_prot_override.diag_destroy = tcp_abort_override;
2061 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
2062 /* In struct mptcp_subflow_request_sock, we assume the TCP request sock
2063 * structures for v4 and v6 have the same size. It should not changed in
2064 * the future but better to make sure to be warned if it is no longer
2067 BUILD_BUG_ON(sizeof(struct tcp_request_sock) != sizeof(struct tcp6_request_sock));
2069 mptcp_subflow_v6_request_sock_ops = tcp6_request_sock_ops;
2070 mptcp_subflow_v6_request_sock_ops.slab_name = "request_sock_subflow_v6";
2071 mptcp_subflow_v6_request_sock_ops.destructor = subflow_v6_req_destructor;
2073 if (subflow_ops_init(&mptcp_subflow_v6_request_sock_ops) != 0)
2074 panic("MPTCP: failed to init subflow v6 request sock ops\n");
2076 subflow_request_sock_ipv6_ops = tcp_request_sock_ipv6_ops;
2077 subflow_request_sock_ipv6_ops.route_req = subflow_v6_route_req;
2078 subflow_request_sock_ipv6_ops.send_synack = subflow_v6_send_synack;
2080 subflow_v6_specific = ipv6_specific;
2081 subflow_v6_specific.conn_request = subflow_v6_conn_request;
2082 subflow_v6_specific.syn_recv_sock = subflow_syn_recv_sock;
2083 subflow_v6_specific.sk_rx_dst_set = subflow_finish_connect;
2084 subflow_v6_specific.rebuild_header = subflow_v6_rebuild_header;
2086 subflow_v6m_specific = subflow_v6_specific;
2087 subflow_v6m_specific.queue_xmit = ipv4_specific.queue_xmit;
2088 subflow_v6m_specific.send_check = ipv4_specific.send_check;
2089 subflow_v6m_specific.net_header_len = ipv4_specific.net_header_len;
2090 subflow_v6m_specific.mtu_reduced = ipv4_specific.mtu_reduced;
2091 subflow_v6m_specific.rebuild_header = subflow_rebuild_header;
2093 tcpv6_prot_override = tcpv6_prot;
2094 tcpv6_prot_override.release_cb = tcp_release_cb_override;
2095 tcpv6_prot_override.diag_destroy = tcp_abort_override;
2098 mptcp_diag_subflow_init(&subflow_ulp_ops);
2100 if (tcp_register_ulp(&subflow_ulp_ops) != 0)
2101 panic("MPTCP: failed to register subflows to ULP\n");