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/algapi.h>
13 #include <crypto/sha2.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>
30 static void mptcp_subflow_ops_undo_override(struct sock *ssk);
32 static void SUBFLOW_REQ_INC_STATS(struct request_sock *req,
33 enum linux_mptcp_mib_field field)
35 MPTCP_INC_STATS(sock_net(req_to_sk(req)), field);
38 static void subflow_req_destructor(struct request_sock *req)
40 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
42 pr_debug("subflow_req=%p", subflow_req);
45 sock_put((struct sock *)subflow_req->msk);
47 mptcp_token_destroy_request(req);
50 static void subflow_generate_hmac(u64 key1, u64 key2, u32 nonce1, u32 nonce2,
55 put_unaligned_be32(nonce1, &msg[0]);
56 put_unaligned_be32(nonce2, &msg[4]);
58 mptcp_crypto_hmac_sha(key1, key2, msg, 8, hmac);
61 static bool mptcp_can_accept_new_subflow(const struct mptcp_sock *msk)
63 return mptcp_is_fully_established((void *)msk) &&
64 ((mptcp_pm_is_userspace(msk) &&
65 mptcp_userspace_pm_active(msk)) ||
66 READ_ONCE(msk->pm.accept_subflow));
69 /* validate received token and create truncated hmac and nonce for SYN-ACK */
70 static void subflow_req_create_thmac(struct mptcp_subflow_request_sock *subflow_req)
72 struct mptcp_sock *msk = subflow_req->msk;
73 u8 hmac[SHA256_DIGEST_SIZE];
75 get_random_bytes(&subflow_req->local_nonce, sizeof(u32));
77 subflow_generate_hmac(msk->local_key, 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 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
310 static struct dst_entry *subflow_v6_route_req(const struct sock *sk,
313 struct request_sock *req)
315 struct dst_entry *dst;
318 tcp_rsk(req)->is_mptcp = 1;
319 subflow_init_req(req, sk);
321 dst = tcp_request_sock_ipv6_ops.route_req(sk, skb, fl, req);
325 err = subflow_check_req(req, sk, skb);
331 tcp6_request_sock_ops.send_reset(sk, skb);
336 /* validate received truncated hmac and create hmac for third ACK */
337 static bool subflow_thmac_valid(struct mptcp_subflow_context *subflow)
339 u8 hmac[SHA256_DIGEST_SIZE];
342 subflow_generate_hmac(subflow->remote_key, subflow->local_key,
343 subflow->remote_nonce, subflow->local_nonce,
346 thmac = get_unaligned_be64(hmac);
347 pr_debug("subflow=%p, token=%u, thmac=%llu, subflow->thmac=%llu\n",
348 subflow, subflow->token, thmac, subflow->thmac);
350 return thmac == subflow->thmac;
353 void mptcp_subflow_reset(struct sock *ssk)
355 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
356 struct sock *sk = subflow->conn;
358 /* mptcp_mp_fail_no_response() can reach here on an already closed
361 if (ssk->sk_state == TCP_CLOSE)
364 /* must hold: tcp_done() could drop last reference on parent */
367 tcp_send_active_reset(ssk, GFP_ATOMIC);
369 if (!test_and_set_bit(MPTCP_WORK_CLOSE_SUBFLOW, &mptcp_sk(sk)->flags))
370 mptcp_schedule_work(sk);
375 static bool subflow_use_different_dport(struct mptcp_sock *msk, const struct sock *sk)
377 return inet_sk(sk)->inet_dport != inet_sk((struct sock *)msk)->inet_dport;
380 void __mptcp_set_connected(struct sock *sk)
382 if (sk->sk_state == TCP_SYN_SENT) {
383 inet_sk_state_store(sk, TCP_ESTABLISHED);
384 sk->sk_state_change(sk);
388 static void mptcp_set_connected(struct sock *sk)
391 if (!sock_owned_by_user(sk))
392 __mptcp_set_connected(sk);
394 __set_bit(MPTCP_CONNECTED, &mptcp_sk(sk)->cb_flags);
395 mptcp_data_unlock(sk);
398 static void subflow_finish_connect(struct sock *sk, const struct sk_buff *skb)
400 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
401 struct mptcp_options_received mp_opt;
402 struct sock *parent = subflow->conn;
404 subflow->icsk_af_ops->sk_rx_dst_set(sk, skb);
406 /* be sure no special action on any packet other than syn-ack */
407 if (subflow->conn_finished)
410 mptcp_propagate_sndbuf(parent, sk);
411 subflow->rel_write_seq = 1;
412 subflow->conn_finished = 1;
413 subflow->ssn_offset = TCP_SKB_CB(skb)->seq;
414 pr_debug("subflow=%p synack seq=%x", subflow, subflow->ssn_offset);
416 mptcp_get_options(skb, &mp_opt);
417 if (subflow->request_mptcp) {
418 if (!(mp_opt.suboptions & OPTION_MPTCP_MPC_SYNACK)) {
419 MPTCP_INC_STATS(sock_net(sk),
420 MPTCP_MIB_MPCAPABLEACTIVEFALLBACK);
421 mptcp_do_fallback(sk);
422 pr_fallback(mptcp_sk(subflow->conn));
426 if (mp_opt.suboptions & OPTION_MPTCP_CSUMREQD)
427 WRITE_ONCE(mptcp_sk(parent)->csum_enabled, true);
428 if (mp_opt.deny_join_id0)
429 WRITE_ONCE(mptcp_sk(parent)->pm.remote_deny_join_id0, true);
430 subflow->mp_capable = 1;
431 subflow->can_ack = 1;
432 subflow->remote_key = mp_opt.sndr_key;
433 pr_debug("subflow=%p, remote_key=%llu", subflow,
434 subflow->remote_key);
435 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPCAPABLEACTIVEACK);
436 mptcp_finish_connect(sk);
437 mptcp_set_connected(parent);
438 } else if (subflow->request_join) {
439 u8 hmac[SHA256_DIGEST_SIZE];
441 if (!(mp_opt.suboptions & OPTION_MPTCP_MPJ_SYNACK)) {
442 subflow->reset_reason = MPTCP_RST_EMPTCP;
446 subflow->backup = mp_opt.backup;
447 subflow->thmac = mp_opt.thmac;
448 subflow->remote_nonce = mp_opt.nonce;
449 WRITE_ONCE(subflow->remote_id, mp_opt.join_id);
450 pr_debug("subflow=%p, thmac=%llu, remote_nonce=%u backup=%d",
451 subflow, subflow->thmac, subflow->remote_nonce,
454 if (!subflow_thmac_valid(subflow)) {
455 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINACKMAC);
456 subflow->reset_reason = MPTCP_RST_EMPTCP;
460 if (!mptcp_finish_join(sk))
463 subflow_generate_hmac(subflow->local_key, subflow->remote_key,
464 subflow->local_nonce,
465 subflow->remote_nonce,
467 memcpy(subflow->hmac, hmac, MPTCPOPT_HMAC_LEN);
469 subflow->mp_join = 1;
470 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINSYNACKRX);
472 if (subflow_use_different_dport(mptcp_sk(parent), sk)) {
473 pr_debug("synack inet_dport=%d %d",
474 ntohs(inet_sk(sk)->inet_dport),
475 ntohs(inet_sk(parent)->inet_dport));
476 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINPORTSYNACKRX);
478 } else if (mptcp_check_fallback(sk)) {
480 mptcp_rcv_space_init(mptcp_sk(parent), sk);
481 mptcp_set_connected(parent);
486 subflow->reset_transient = 0;
487 mptcp_subflow_reset(sk);
490 static void subflow_set_local_id(struct mptcp_subflow_context *subflow, int local_id)
492 WARN_ON_ONCE(local_id < 0 || local_id > 255);
493 WRITE_ONCE(subflow->local_id, local_id);
496 static int subflow_chk_local_id(struct sock *sk)
498 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
499 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
502 if (likely(subflow->local_id >= 0))
505 err = mptcp_pm_get_local_id(msk, (struct sock_common *)sk);
509 subflow_set_local_id(subflow, err);
513 static int subflow_rebuild_header(struct sock *sk)
515 int err = subflow_chk_local_id(sk);
517 if (unlikely(err < 0))
520 return inet_sk_rebuild_header(sk);
523 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
524 static int subflow_v6_rebuild_header(struct sock *sk)
526 int err = subflow_chk_local_id(sk);
528 if (unlikely(err < 0))
531 return inet6_sk_rebuild_header(sk);
535 static struct request_sock_ops mptcp_subflow_v4_request_sock_ops __ro_after_init;
536 static struct tcp_request_sock_ops subflow_request_sock_ipv4_ops __ro_after_init;
538 static int subflow_v4_conn_request(struct sock *sk, struct sk_buff *skb)
540 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
542 pr_debug("subflow=%p", subflow);
544 /* Never answer to SYNs sent to broadcast or multicast */
545 if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
548 return tcp_conn_request(&mptcp_subflow_v4_request_sock_ops,
549 &subflow_request_sock_ipv4_ops,
556 static void subflow_v4_req_destructor(struct request_sock *req)
558 subflow_req_destructor(req);
559 tcp_request_sock_ops.destructor(req);
562 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
563 static struct request_sock_ops mptcp_subflow_v6_request_sock_ops __ro_after_init;
564 static struct tcp_request_sock_ops subflow_request_sock_ipv6_ops __ro_after_init;
565 static struct inet_connection_sock_af_ops subflow_v6_specific __ro_after_init;
566 static struct inet_connection_sock_af_ops subflow_v6m_specific __ro_after_init;
567 static struct proto tcpv6_prot_override __ro_after_init;
569 static int subflow_v6_conn_request(struct sock *sk, struct sk_buff *skb)
571 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
573 pr_debug("subflow=%p", subflow);
575 if (skb->protocol == htons(ETH_P_IP))
576 return subflow_v4_conn_request(sk, skb);
578 if (!ipv6_unicast_destination(skb))
581 if (ipv6_addr_v4mapped(&ipv6_hdr(skb)->saddr)) {
582 __IP6_INC_STATS(sock_net(sk), NULL, IPSTATS_MIB_INHDRERRORS);
586 return tcp_conn_request(&mptcp_subflow_v6_request_sock_ops,
587 &subflow_request_sock_ipv6_ops, sk, skb);
591 return 0; /* don't send reset */
594 static void subflow_v6_req_destructor(struct request_sock *req)
596 subflow_req_destructor(req);
597 tcp6_request_sock_ops.destructor(req);
601 struct request_sock *mptcp_subflow_reqsk_alloc(const struct request_sock_ops *ops,
602 struct sock *sk_listener,
603 bool attach_listener)
605 if (ops->family == AF_INET)
606 ops = &mptcp_subflow_v4_request_sock_ops;
607 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
608 else if (ops->family == AF_INET6)
609 ops = &mptcp_subflow_v6_request_sock_ops;
612 return inet_reqsk_alloc(ops, sk_listener, attach_listener);
614 EXPORT_SYMBOL(mptcp_subflow_reqsk_alloc);
616 /* validate hmac received in third ACK */
617 static bool subflow_hmac_valid(const struct request_sock *req,
618 const struct mptcp_options_received *mp_opt)
620 const struct mptcp_subflow_request_sock *subflow_req;
621 u8 hmac[SHA256_DIGEST_SIZE];
622 struct mptcp_sock *msk;
624 subflow_req = mptcp_subflow_rsk(req);
625 msk = subflow_req->msk;
629 subflow_generate_hmac(msk->remote_key, msk->local_key,
630 subflow_req->remote_nonce,
631 subflow_req->local_nonce, hmac);
633 return !crypto_memneq(hmac, mp_opt->hmac, MPTCPOPT_HMAC_LEN);
636 static void subflow_ulp_fallback(struct sock *sk,
637 struct mptcp_subflow_context *old_ctx)
639 struct inet_connection_sock *icsk = inet_csk(sk);
641 mptcp_subflow_tcp_fallback(sk, old_ctx);
642 icsk->icsk_ulp_ops = NULL;
643 rcu_assign_pointer(icsk->icsk_ulp_data, NULL);
644 tcp_sk(sk)->is_mptcp = 0;
646 mptcp_subflow_ops_undo_override(sk);
649 void mptcp_subflow_drop_ctx(struct sock *ssk)
651 struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(ssk);
656 list_del(&mptcp_subflow_ctx(ssk)->node);
657 if (inet_csk(ssk)->icsk_ulp_ops) {
658 subflow_ulp_fallback(ssk, ctx);
666 void mptcp_subflow_fully_established(struct mptcp_subflow_context *subflow,
667 struct mptcp_options_received *mp_opt)
669 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
671 subflow->remote_key = mp_opt->sndr_key;
672 subflow->fully_established = 1;
673 subflow->can_ack = 1;
674 WRITE_ONCE(msk->fully_established, true);
677 static struct sock *subflow_syn_recv_sock(const struct sock *sk,
679 struct request_sock *req,
680 struct dst_entry *dst,
681 struct request_sock *req_unhash,
684 struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk);
685 struct mptcp_subflow_request_sock *subflow_req;
686 struct mptcp_options_received mp_opt;
687 bool fallback, fallback_is_fatal;
688 struct mptcp_sock *owner;
691 pr_debug("listener=%p, req=%p, conn=%p", listener, req, listener->conn);
693 /* After child creation we must look for MPC even when options
696 mp_opt.suboptions = 0;
698 /* hopefully temporary handling for MP_JOIN+syncookie */
699 subflow_req = mptcp_subflow_rsk(req);
700 fallback_is_fatal = tcp_rsk(req)->is_mptcp && subflow_req->mp_join;
701 fallback = !tcp_rsk(req)->is_mptcp;
705 /* if the sk is MP_CAPABLE, we try to fetch the client key */
706 if (subflow_req->mp_capable) {
707 /* we can receive and accept an in-window, out-of-order pkt,
708 * which may not carry the MP_CAPABLE opt even on mptcp enabled
709 * paths: always try to extract the peer key, and fallback
710 * for packets missing it.
711 * Even OoO DSS packets coming legitly after dropped or
712 * reordered MPC will cause fallback, but we don't have other
715 mptcp_get_options(skb, &mp_opt);
716 if (!(mp_opt.suboptions &
717 (OPTION_MPTCP_MPC_SYN | OPTION_MPTCP_MPC_ACK)))
720 } else if (subflow_req->mp_join) {
721 mptcp_get_options(skb, &mp_opt);
722 if (!(mp_opt.suboptions & OPTION_MPTCP_MPJ_ACK) ||
723 !subflow_hmac_valid(req, &mp_opt) ||
724 !mptcp_can_accept_new_subflow(subflow_req->msk)) {
725 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKMAC);
731 child = listener->icsk_af_ops->syn_recv_sock(sk, skb, req, dst,
732 req_unhash, own_req);
734 if (child && *own_req) {
735 struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(child);
737 tcp_rsk(req)->drop_req = false;
739 /* we need to fallback on ctx allocation failure and on pre-reqs
740 * checking above. In the latter scenario we additionally need
741 * to reset the context to non MPTCP status.
743 if (!ctx || fallback) {
744 if (fallback_is_fatal) {
745 subflow_add_reset_reason(skb, MPTCP_RST_EMPTCP);
751 /* ssk inherits options of listener sk */
752 ctx->setsockopt_seq = listener->setsockopt_seq;
754 if (ctx->mp_capable) {
755 ctx->conn = mptcp_sk_clone_init(listener->conn, &mp_opt, child, req);
759 owner = mptcp_sk(ctx->conn);
760 mptcp_pm_new_connection(owner, child, 1);
762 /* with OoO packets we can reach here without ingress
765 if (mp_opt.suboptions & OPTIONS_MPTCP_MPC) {
766 mptcp_subflow_fully_established(ctx, &mp_opt);
767 mptcp_pm_fully_established(owner, child, GFP_ATOMIC);
768 ctx->pm_notified = 1;
770 } else if (ctx->mp_join) {
771 owner = subflow_req->msk;
773 subflow_add_reset_reason(skb, MPTCP_RST_EPROHIBIT);
777 /* move the msk reference ownership to the subflow */
778 subflow_req->msk = NULL;
779 ctx->conn = (struct sock *)owner;
781 if (subflow_use_different_sport(owner, sk)) {
782 pr_debug("ack inet_sport=%d %d",
783 ntohs(inet_sk(sk)->inet_sport),
784 ntohs(inet_sk((struct sock *)owner)->inet_sport));
785 if (!mptcp_pm_sport_in_anno_list(owner, sk)) {
786 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MISMATCHPORTACKRX);
789 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINPORTACKRX);
792 if (!mptcp_finish_join(child))
795 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKRX);
796 tcp_rsk(req)->drop_req = true;
800 /* check for expected invariant - should never trigger, just help
801 * catching eariler subtle bugs
803 WARN_ON_ONCE(child && *own_req && tcp_sk(child)->is_mptcp &&
804 (!mptcp_subflow_ctx(child) ||
805 !mptcp_subflow_ctx(child)->conn));
809 mptcp_subflow_drop_ctx(child);
810 tcp_rsk(req)->drop_req = true;
811 inet_csk_prepare_for_destroy_sock(child);
813 req->rsk_ops->send_reset(sk, skb);
815 /* The last child reference will be released by the caller */
820 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MPCAPABLEPASSIVEFALLBACK);
821 mptcp_subflow_drop_ctx(child);
825 static struct inet_connection_sock_af_ops subflow_specific __ro_after_init;
826 static struct proto tcp_prot_override __ro_after_init;
828 enum mapping_status {
837 static void dbg_bad_map(struct mptcp_subflow_context *subflow, u32 ssn)
839 pr_debug("Bad mapping: ssn=%d map_seq=%d map_data_len=%d",
840 ssn, subflow->map_subflow_seq, subflow->map_data_len);
843 static bool skb_is_fully_mapped(struct sock *ssk, struct sk_buff *skb)
845 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
846 unsigned int skb_consumed;
848 skb_consumed = tcp_sk(ssk)->copied_seq - TCP_SKB_CB(skb)->seq;
849 if (WARN_ON_ONCE(skb_consumed >= skb->len))
852 return skb->len - skb_consumed <= subflow->map_data_len -
853 mptcp_subflow_get_map_offset(subflow);
856 static bool validate_mapping(struct sock *ssk, struct sk_buff *skb)
858 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
859 u32 ssn = tcp_sk(ssk)->copied_seq - subflow->ssn_offset;
861 if (unlikely(before(ssn, subflow->map_subflow_seq))) {
862 /* Mapping covers data later in the subflow stream,
863 * currently unsupported.
865 dbg_bad_map(subflow, ssn);
868 if (unlikely(!before(ssn, subflow->map_subflow_seq +
869 subflow->map_data_len))) {
870 /* Mapping does covers past subflow data, invalid */
871 dbg_bad_map(subflow, ssn);
877 static enum mapping_status validate_data_csum(struct sock *ssk, struct sk_buff *skb,
880 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
881 u32 offset, seq, delta;
888 /* mapping already validated on previous traversal */
889 if (subflow->map_csum_len == subflow->map_data_len)
892 /* traverse the receive queue, ensuring it contains a full
893 * DSS mapping and accumulating the related csum.
894 * Preserve the accoumlate csum across multiple calls, to compute
897 delta = subflow->map_data_len - subflow->map_csum_len;
899 seq = tcp_sk(ssk)->copied_seq + subflow->map_csum_len;
900 offset = seq - TCP_SKB_CB(skb)->seq;
902 /* if the current skb has not been accounted yet, csum its contents
903 * up to the amount covered by the current DSS
905 if (offset < skb->len) {
908 len = min(skb->len - offset, delta);
909 csum = skb_checksum(skb, offset, len, 0);
910 subflow->map_data_csum = csum_block_add(subflow->map_data_csum, csum,
911 subflow->map_csum_len);
914 subflow->map_csum_len += len;
919 if (skb_queue_is_last(&ssk->sk_receive_queue, skb)) {
920 /* if this subflow is closed, the partial mapping
921 * will be never completed; flush the pending skbs, so
922 * that subflow_sched_work_if_closed() can kick in
924 if (unlikely(ssk->sk_state == TCP_CLOSE))
925 while ((skb = skb_peek(&ssk->sk_receive_queue)))
926 sk_eat_skb(ssk, skb);
928 /* not enough data to validate the csum */
929 return MAPPING_EMPTY;
932 /* the DSS mapping for next skbs will be validated later,
933 * when a get_mapping_status call will process such skb
938 /* note that 'map_data_len' accounts only for the carried data, does
939 * not include the eventual seq increment due to the data fin,
940 * while the pseudo header requires the original DSS data len,
943 csum = __mptcp_make_csum(subflow->map_seq,
944 subflow->map_subflow_seq,
945 subflow->map_data_len + subflow->map_data_fin,
946 subflow->map_data_csum);
947 if (unlikely(csum)) {
948 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DATACSUMERR);
949 return MAPPING_BAD_CSUM;
952 subflow->valid_csum_seen = 1;
956 static enum mapping_status get_mapping_status(struct sock *ssk,
957 struct mptcp_sock *msk)
959 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
960 bool csum_reqd = READ_ONCE(msk->csum_enabled);
961 struct mptcp_ext *mpext;
966 skb = skb_peek(&ssk->sk_receive_queue);
968 return MAPPING_EMPTY;
970 if (mptcp_check_fallback(ssk))
971 return MAPPING_DUMMY;
973 mpext = mptcp_get_ext(skb);
974 if (!mpext || !mpext->use_map) {
975 if (!subflow->map_valid && !skb->len) {
976 /* the TCP stack deliver 0 len FIN pkt to the receive
977 * queue, that is the only 0len pkts ever expected here,
978 * and we can admit no mapping only for 0 len pkts
980 if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN))
981 WARN_ONCE(1, "0len seq %d:%d flags %x",
982 TCP_SKB_CB(skb)->seq,
983 TCP_SKB_CB(skb)->end_seq,
984 TCP_SKB_CB(skb)->tcp_flags);
985 sk_eat_skb(ssk, skb);
986 return MAPPING_EMPTY;
989 if (!subflow->map_valid)
990 return MAPPING_INVALID;
995 trace_get_mapping_status(mpext);
997 data_len = mpext->data_len;
999 pr_debug("infinite mapping received");
1000 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_INFINITEMAPRX);
1001 subflow->map_data_len = 0;
1002 return MAPPING_INVALID;
1005 if (mpext->data_fin == 1) {
1006 if (data_len == 1) {
1007 bool updated = mptcp_update_rcv_data_fin(msk, mpext->data_seq,
1009 pr_debug("DATA_FIN with no payload seq=%llu", mpext->data_seq);
1010 if (subflow->map_valid) {
1011 /* A DATA_FIN might arrive in a DSS
1012 * option before the previous mapping
1013 * has been fully consumed. Continue
1014 * handling the existing mapping.
1016 skb_ext_del(skb, SKB_EXT_MPTCP);
1020 mptcp_schedule_work((struct sock *)msk);
1022 return MAPPING_DATA_FIN;
1025 u64 data_fin_seq = mpext->data_seq + data_len - 1;
1027 /* If mpext->data_seq is a 32-bit value, data_fin_seq
1028 * must also be limited to 32 bits.
1031 data_fin_seq &= GENMASK_ULL(31, 0);
1033 mptcp_update_rcv_data_fin(msk, data_fin_seq, mpext->dsn64);
1034 pr_debug("DATA_FIN with mapping seq=%llu dsn64=%d",
1035 data_fin_seq, mpext->dsn64);
1038 /* Adjust for DATA_FIN using 1 byte of sequence space */
1042 map_seq = mptcp_expand_seq(READ_ONCE(msk->ack_seq), mpext->data_seq, mpext->dsn64);
1043 WRITE_ONCE(mptcp_sk(subflow->conn)->use_64bit_ack, !!mpext->dsn64);
1045 if (subflow->map_valid) {
1046 /* Allow replacing only with an identical map */
1047 if (subflow->map_seq == map_seq &&
1048 subflow->map_subflow_seq == mpext->subflow_seq &&
1049 subflow->map_data_len == data_len &&
1050 subflow->map_csum_reqd == mpext->csum_reqd) {
1051 skb_ext_del(skb, SKB_EXT_MPTCP);
1055 /* If this skb data are fully covered by the current mapping,
1056 * the new map would need caching, which is not supported
1058 if (skb_is_fully_mapped(ssk, skb)) {
1059 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DSSNOMATCH);
1060 return MAPPING_INVALID;
1063 /* will validate the next map after consuming the current one */
1067 subflow->map_seq = map_seq;
1068 subflow->map_subflow_seq = mpext->subflow_seq;
1069 subflow->map_data_len = data_len;
1070 subflow->map_valid = 1;
1071 subflow->map_data_fin = mpext->data_fin;
1072 subflow->mpc_map = mpext->mpc_map;
1073 subflow->map_csum_reqd = mpext->csum_reqd;
1074 subflow->map_csum_len = 0;
1075 subflow->map_data_csum = csum_unfold(mpext->csum);
1077 /* Cfr RFC 8684 Section 3.3.0 */
1078 if (unlikely(subflow->map_csum_reqd != csum_reqd))
1079 return MAPPING_INVALID;
1081 pr_debug("new map seq=%llu subflow_seq=%u data_len=%u csum=%d:%u",
1082 subflow->map_seq, subflow->map_subflow_seq,
1083 subflow->map_data_len, subflow->map_csum_reqd,
1084 subflow->map_data_csum);
1087 /* we revalidate valid mapping on new skb, because we must ensure
1088 * the current skb is completely covered by the available mapping
1090 if (!validate_mapping(ssk, skb)) {
1091 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DSSTCPMISMATCH);
1092 return MAPPING_INVALID;
1095 skb_ext_del(skb, SKB_EXT_MPTCP);
1098 return validate_data_csum(ssk, skb, csum_reqd);
1101 static void mptcp_subflow_discard_data(struct sock *ssk, struct sk_buff *skb,
1104 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1105 bool fin = TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN;
1108 incr = limit >= skb->len ? skb->len + fin : limit;
1110 pr_debug("discarding=%d len=%d seq=%d", incr, skb->len,
1111 subflow->map_subflow_seq);
1112 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DUPDATA);
1113 tcp_sk(ssk)->copied_seq += incr;
1114 if (!before(tcp_sk(ssk)->copied_seq, TCP_SKB_CB(skb)->end_seq))
1115 sk_eat_skb(ssk, skb);
1116 if (mptcp_subflow_get_map_offset(subflow) >= subflow->map_data_len)
1117 subflow->map_valid = 0;
1120 /* sched mptcp worker to remove the subflow if no more data is pending */
1121 static void subflow_sched_work_if_closed(struct mptcp_sock *msk, struct sock *ssk)
1123 if (likely(ssk->sk_state != TCP_CLOSE))
1126 if (skb_queue_empty(&ssk->sk_receive_queue) &&
1127 !test_and_set_bit(MPTCP_WORK_CLOSE_SUBFLOW, &msk->flags))
1128 mptcp_schedule_work((struct sock *)msk);
1131 static bool subflow_can_fallback(struct mptcp_subflow_context *subflow)
1133 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
1135 if (subflow->mp_join)
1137 else if (READ_ONCE(msk->csum_enabled))
1138 return !subflow->valid_csum_seen;
1140 return !subflow->fully_established;
1143 static void mptcp_subflow_fail(struct mptcp_sock *msk, struct sock *ssk)
1145 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1146 unsigned long fail_tout;
1148 /* greceful failure can happen only on the MPC subflow */
1149 if (WARN_ON_ONCE(ssk != READ_ONCE(msk->first)))
1152 /* since the close timeout take precedence on the fail one,
1153 * no need to start the latter when the first is already set
1155 if (sock_flag((struct sock *)msk, SOCK_DEAD))
1158 /* we don't need extreme accuracy here, use a zero fail_tout as special
1159 * value meaning no fail timeout at all;
1161 fail_tout = jiffies + TCP_RTO_MAX;
1164 WRITE_ONCE(subflow->fail_tout, fail_tout);
1167 mptcp_reset_tout_timer(msk, subflow->fail_tout);
1170 static bool subflow_check_data_avail(struct sock *ssk)
1172 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1173 enum mapping_status status;
1174 struct mptcp_sock *msk;
1175 struct sk_buff *skb;
1177 if (!skb_peek(&ssk->sk_receive_queue))
1178 WRITE_ONCE(subflow->data_avail, MPTCP_SUBFLOW_NODATA);
1179 if (subflow->data_avail)
1182 msk = mptcp_sk(subflow->conn);
1187 status = get_mapping_status(ssk, msk);
1188 trace_subflow_check_data_avail(status, skb_peek(&ssk->sk_receive_queue));
1189 if (unlikely(status == MAPPING_INVALID || status == MAPPING_DUMMY ||
1190 status == MAPPING_BAD_CSUM))
1193 if (status != MAPPING_OK)
1196 skb = skb_peek(&ssk->sk_receive_queue);
1197 if (WARN_ON_ONCE(!skb))
1200 /* if msk lacks the remote key, this subflow must provide an
1201 * MP_CAPABLE-based mapping
1203 if (unlikely(!READ_ONCE(msk->can_ack))) {
1204 if (!subflow->mpc_map)
1206 WRITE_ONCE(msk->remote_key, subflow->remote_key);
1207 WRITE_ONCE(msk->ack_seq, subflow->map_seq);
1208 WRITE_ONCE(msk->can_ack, true);
1211 old_ack = READ_ONCE(msk->ack_seq);
1212 ack_seq = mptcp_subflow_get_mapped_dsn(subflow);
1213 pr_debug("msk ack_seq=%llx subflow ack_seq=%llx", old_ack,
1215 if (unlikely(before64(ack_seq, old_ack))) {
1216 mptcp_subflow_discard_data(ssk, skb, old_ack - ack_seq);
1220 WRITE_ONCE(subflow->data_avail, MPTCP_SUBFLOW_DATA_AVAIL);
1226 subflow_sched_work_if_closed(msk, ssk);
1230 if (!__mptcp_check_fallback(msk)) {
1231 /* RFC 8684 section 3.7. */
1232 if (status == MAPPING_BAD_CSUM &&
1233 (subflow->mp_join || subflow->valid_csum_seen)) {
1234 subflow->send_mp_fail = 1;
1236 if (!READ_ONCE(msk->allow_infinite_fallback)) {
1237 subflow->reset_transient = 0;
1238 subflow->reset_reason = MPTCP_RST_EMIDDLEBOX;
1241 mptcp_subflow_fail(msk, ssk);
1242 WRITE_ONCE(subflow->data_avail, MPTCP_SUBFLOW_DATA_AVAIL);
1246 if (!subflow_can_fallback(subflow) && subflow->map_data_len) {
1247 /* fatal protocol error, close the socket.
1248 * subflow_error_report() will introduce the appropriate barriers
1250 subflow->reset_transient = 0;
1251 subflow->reset_reason = MPTCP_RST_EMPTCP;
1254 WRITE_ONCE(ssk->sk_err, EBADMSG);
1255 tcp_set_state(ssk, TCP_CLOSE);
1256 while ((skb = skb_peek(&ssk->sk_receive_queue)))
1257 sk_eat_skb(ssk, skb);
1258 tcp_send_active_reset(ssk, GFP_ATOMIC);
1259 WRITE_ONCE(subflow->data_avail, MPTCP_SUBFLOW_NODATA);
1263 mptcp_do_fallback(ssk);
1266 skb = skb_peek(&ssk->sk_receive_queue);
1267 subflow->map_valid = 1;
1268 subflow->map_seq = READ_ONCE(msk->ack_seq);
1269 subflow->map_data_len = skb->len;
1270 subflow->map_subflow_seq = tcp_sk(ssk)->copied_seq - subflow->ssn_offset;
1271 WRITE_ONCE(subflow->data_avail, MPTCP_SUBFLOW_DATA_AVAIL);
1275 bool mptcp_subflow_data_available(struct sock *sk)
1277 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1279 /* check if current mapping is still valid */
1280 if (subflow->map_valid &&
1281 mptcp_subflow_get_map_offset(subflow) >= subflow->map_data_len) {
1282 subflow->map_valid = 0;
1283 WRITE_ONCE(subflow->data_avail, MPTCP_SUBFLOW_NODATA);
1285 pr_debug("Done with mapping: seq=%u data_len=%u",
1286 subflow->map_subflow_seq,
1287 subflow->map_data_len);
1290 return subflow_check_data_avail(sk);
1293 /* If ssk has an mptcp parent socket, use the mptcp rcvbuf occupancy,
1296 * In mptcp, rwin is about the mptcp-level connection data.
1298 * Data that is still on the ssk rx queue can thus be ignored,
1299 * as far as mptcp peer is concerned that data is still inflight.
1300 * DSS ACK is updated when skb is moved to the mptcp rx queue.
1302 void mptcp_space(const struct sock *ssk, int *space, int *full_space)
1304 const struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1305 const struct sock *sk = subflow->conn;
1307 *space = __mptcp_space(sk);
1308 *full_space = tcp_full_space(sk);
1311 static void subflow_error_report(struct sock *ssk)
1313 struct sock *sk = mptcp_subflow_ctx(ssk)->conn;
1315 /* bail early if this is a no-op, so that we avoid introducing a
1316 * problematic lockdep dependency between TCP accept queue lock
1317 * and msk socket spinlock
1322 mptcp_data_lock(sk);
1323 if (!sock_owned_by_user(sk))
1324 __mptcp_error_report(sk);
1326 __set_bit(MPTCP_ERROR_REPORT, &mptcp_sk(sk)->cb_flags);
1327 mptcp_data_unlock(sk);
1330 static void subflow_data_ready(struct sock *sk)
1332 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1333 u16 state = 1 << inet_sk_state_load(sk);
1334 struct sock *parent = subflow->conn;
1335 struct mptcp_sock *msk;
1337 msk = mptcp_sk(parent);
1338 if (state & TCPF_LISTEN) {
1339 /* MPJ subflow are removed from accept queue before reaching here,
1340 * avoid stray wakeups
1342 if (reqsk_queue_empty(&inet_csk(sk)->icsk_accept_queue))
1345 parent->sk_data_ready(parent);
1349 WARN_ON_ONCE(!__mptcp_check_fallback(msk) && !subflow->mp_capable &&
1350 !subflow->mp_join && !(state & TCPF_CLOSE));
1352 if (mptcp_subflow_data_available(sk))
1353 mptcp_data_ready(parent, sk);
1354 else if (unlikely(sk->sk_err))
1355 subflow_error_report(sk);
1358 static void subflow_write_space(struct sock *ssk)
1360 struct sock *sk = mptcp_subflow_ctx(ssk)->conn;
1362 mptcp_propagate_sndbuf(sk, ssk);
1363 mptcp_write_space(sk);
1366 static const struct inet_connection_sock_af_ops *
1367 subflow_default_af_ops(struct sock *sk)
1369 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1370 if (sk->sk_family == AF_INET6)
1371 return &subflow_v6_specific;
1373 return &subflow_specific;
1376 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1377 void mptcpv6_handle_mapped(struct sock *sk, bool mapped)
1379 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1380 struct inet_connection_sock *icsk = inet_csk(sk);
1381 const struct inet_connection_sock_af_ops *target;
1383 target = mapped ? &subflow_v6m_specific : subflow_default_af_ops(sk);
1385 pr_debug("subflow=%p family=%d ops=%p target=%p mapped=%d",
1386 subflow, sk->sk_family, icsk->icsk_af_ops, target, mapped);
1388 if (likely(icsk->icsk_af_ops == target))
1391 subflow->icsk_af_ops = icsk->icsk_af_ops;
1392 icsk->icsk_af_ops = target;
1396 void mptcp_info2sockaddr(const struct mptcp_addr_info *info,
1397 struct sockaddr_storage *addr,
1398 unsigned short family)
1400 memset(addr, 0, sizeof(*addr));
1401 addr->ss_family = family;
1402 if (addr->ss_family == AF_INET) {
1403 struct sockaddr_in *in_addr = (struct sockaddr_in *)addr;
1405 if (info->family == AF_INET)
1406 in_addr->sin_addr = info->addr;
1407 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1408 else if (ipv6_addr_v4mapped(&info->addr6))
1409 in_addr->sin_addr.s_addr = info->addr6.s6_addr32[3];
1411 in_addr->sin_port = info->port;
1413 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1414 else if (addr->ss_family == AF_INET6) {
1415 struct sockaddr_in6 *in6_addr = (struct sockaddr_in6 *)addr;
1417 if (info->family == AF_INET)
1418 ipv6_addr_set_v4mapped(info->addr.s_addr,
1419 &in6_addr->sin6_addr);
1421 in6_addr->sin6_addr = info->addr6;
1422 in6_addr->sin6_port = info->port;
1427 int __mptcp_subflow_connect(struct sock *sk, const struct mptcp_addr_info *loc,
1428 const struct mptcp_addr_info *remote)
1430 struct mptcp_sock *msk = mptcp_sk(sk);
1431 struct mptcp_subflow_context *subflow;
1432 struct sockaddr_storage addr;
1433 int remote_id = remote->id;
1434 int local_id = loc->id;
1435 int err = -ENOTCONN;
1443 if (!mptcp_is_fully_established(sk))
1446 err = mptcp_subflow_create_socket(sk, loc->family, &sf);
1451 subflow = mptcp_subflow_ctx(ssk);
1453 get_random_bytes(&subflow->local_nonce, sizeof(u32));
1454 } while (!subflow->local_nonce);
1457 subflow_set_local_id(subflow, local_id);
1459 mptcp_pm_get_flags_and_ifindex_by_id(msk, local_id,
1461 subflow->remote_key = msk->remote_key;
1462 subflow->local_key = msk->local_key;
1463 subflow->token = msk->token;
1464 mptcp_info2sockaddr(loc, &addr, ssk->sk_family);
1466 addrlen = sizeof(struct sockaddr_in);
1467 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1468 if (addr.ss_family == AF_INET6)
1469 addrlen = sizeof(struct sockaddr_in6);
1471 mptcp_sockopt_sync(msk, ssk);
1473 ssk->sk_bound_dev_if = ifindex;
1474 err = kernel_bind(sf, (struct sockaddr *)&addr, addrlen);
1478 mptcp_crypto_key_sha(subflow->remote_key, &remote_token, NULL);
1479 pr_debug("msk=%p remote_token=%u local_id=%d remote_id=%d", msk,
1480 remote_token, local_id, remote_id);
1481 subflow->remote_token = remote_token;
1482 WRITE_ONCE(subflow->remote_id, remote_id);
1483 subflow->request_join = 1;
1484 subflow->request_bkup = !!(flags & MPTCP_PM_ADDR_FLAG_BACKUP);
1485 mptcp_info2sockaddr(remote, &addr, ssk->sk_family);
1488 list_add_tail(&subflow->node, &msk->conn_list);
1489 err = kernel_connect(sf, (struct sockaddr *)&addr, addrlen, O_NONBLOCK);
1490 if (err && err != -EINPROGRESS)
1493 /* discard the subflow socket */
1494 mptcp_sock_graft(ssk, sk->sk_socket);
1495 iput(SOCK_INODE(sf));
1496 WRITE_ONCE(msk->allow_infinite_fallback, false);
1497 mptcp_stop_tout_timer(sk);
1501 list_del(&subflow->node);
1502 sock_put(mptcp_subflow_tcp_sock(subflow));
1505 subflow->disposable = 1;
1509 /* we account subflows before the creation, and this failures will not
1510 * be caught by sk_state_change()
1512 mptcp_pm_close_subflow(msk);
1516 static void mptcp_attach_cgroup(struct sock *parent, struct sock *child)
1518 #ifdef CONFIG_SOCK_CGROUP_DATA
1519 struct sock_cgroup_data *parent_skcd = &parent->sk_cgrp_data,
1520 *child_skcd = &child->sk_cgrp_data;
1522 /* only the additional subflows created by kworkers have to be modified */
1523 if (cgroup_id(sock_cgroup_ptr(parent_skcd)) !=
1524 cgroup_id(sock_cgroup_ptr(child_skcd))) {
1526 struct mem_cgroup *memcg = parent->sk_memcg;
1528 mem_cgroup_sk_free(child);
1529 if (memcg && css_tryget(&memcg->css))
1530 child->sk_memcg = memcg;
1531 #endif /* CONFIG_MEMCG */
1533 cgroup_sk_free(child_skcd);
1534 *child_skcd = *parent_skcd;
1535 cgroup_sk_clone(child_skcd);
1537 #endif /* CONFIG_SOCK_CGROUP_DATA */
1540 static void mptcp_subflow_ops_override(struct sock *ssk)
1542 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1543 if (ssk->sk_prot == &tcpv6_prot)
1544 ssk->sk_prot = &tcpv6_prot_override;
1547 ssk->sk_prot = &tcp_prot_override;
1550 static void mptcp_subflow_ops_undo_override(struct sock *ssk)
1552 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1553 if (ssk->sk_prot == &tcpv6_prot_override)
1554 ssk->sk_prot = &tcpv6_prot;
1557 ssk->sk_prot = &tcp_prot;
1560 int mptcp_subflow_create_socket(struct sock *sk, unsigned short family,
1561 struct socket **new_sock)
1563 struct mptcp_subflow_context *subflow;
1564 struct net *net = sock_net(sk);
1568 /* un-accepted server sockets can reach here - on bad configuration
1569 * bail early to avoid greater trouble later
1571 if (unlikely(!sk->sk_socket))
1574 err = sock_create_kern(net, family, SOCK_STREAM, IPPROTO_TCP, &sf);
1578 lock_sock_nested(sf->sk, SINGLE_DEPTH_NESTING);
1580 /* the newly created socket has to be in the same cgroup as its parent */
1581 mptcp_attach_cgroup(sk, sf->sk);
1583 /* kernel sockets do not by default acquire net ref, but TCP timer
1586 sf->sk->sk_net_refcnt = 1;
1587 get_net_track(net, &sf->sk->ns_tracker, GFP_KERNEL);
1588 sock_inuse_add(net, 1);
1589 err = tcp_set_ulp(sf->sk, "mptcp");
1590 release_sock(sf->sk);
1597 /* the newly created socket really belongs to the owning MPTCP master
1598 * socket, even if for additional subflows the allocation is performed
1599 * by a kernel workqueue. Adjust inode references, so that the
1600 * procfs/diag interfaces really show this one belonging to the correct
1603 SOCK_INODE(sf)->i_ino = SOCK_INODE(sk->sk_socket)->i_ino;
1604 SOCK_INODE(sf)->i_uid = SOCK_INODE(sk->sk_socket)->i_uid;
1605 SOCK_INODE(sf)->i_gid = SOCK_INODE(sk->sk_socket)->i_gid;
1607 subflow = mptcp_subflow_ctx(sf->sk);
1608 pr_debug("subflow=%p", subflow);
1613 mptcp_subflow_ops_override(sf->sk);
1618 static struct mptcp_subflow_context *subflow_create_ctx(struct sock *sk,
1621 struct inet_connection_sock *icsk = inet_csk(sk);
1622 struct mptcp_subflow_context *ctx;
1624 ctx = kzalloc(sizeof(*ctx), priority);
1628 rcu_assign_pointer(icsk->icsk_ulp_data, ctx);
1629 INIT_LIST_HEAD(&ctx->node);
1630 INIT_LIST_HEAD(&ctx->delegated_node);
1632 pr_debug("subflow=%p", ctx);
1635 WRITE_ONCE(ctx->local_id, -1);
1640 static void __subflow_state_change(struct sock *sk)
1642 struct socket_wq *wq;
1645 wq = rcu_dereference(sk->sk_wq);
1646 if (skwq_has_sleeper(wq))
1647 wake_up_interruptible_all(&wq->wait);
1651 static bool subflow_is_done(const struct sock *sk)
1653 return sk->sk_shutdown & RCV_SHUTDOWN || sk->sk_state == TCP_CLOSE;
1656 static void subflow_state_change(struct sock *sk)
1658 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1659 struct sock *parent = subflow->conn;
1660 struct mptcp_sock *msk;
1662 __subflow_state_change(sk);
1664 msk = mptcp_sk(parent);
1665 if (subflow_simultaneous_connect(sk)) {
1666 mptcp_propagate_sndbuf(parent, sk);
1667 mptcp_do_fallback(sk);
1668 mptcp_rcv_space_init(msk, sk);
1670 subflow->conn_finished = 1;
1671 mptcp_set_connected(parent);
1674 /* as recvmsg() does not acquire the subflow socket for ssk selection
1675 * a fin packet carrying a DSS can be unnoticed if we don't trigger
1676 * the data available machinery here.
1678 if (mptcp_subflow_data_available(sk))
1679 mptcp_data_ready(parent, sk);
1680 else if (unlikely(sk->sk_err))
1681 subflow_error_report(sk);
1683 subflow_sched_work_if_closed(mptcp_sk(parent), sk);
1685 /* when the fallback subflow closes the rx side, trigger a 'dummy'
1686 * ingress data fin, so that the msk state will follow along
1688 if (__mptcp_check_fallback(msk) && subflow_is_done(sk) && msk->first == sk &&
1689 mptcp_update_rcv_data_fin(msk, READ_ONCE(msk->ack_seq), true))
1690 mptcp_schedule_work(parent);
1693 void mptcp_subflow_queue_clean(struct sock *listener_sk, struct sock *listener_ssk)
1695 struct request_sock_queue *queue = &inet_csk(listener_ssk)->icsk_accept_queue;
1696 struct request_sock *req, *head, *tail;
1697 struct mptcp_subflow_context *subflow;
1698 struct sock *sk, *ssk;
1700 /* Due to lock dependencies no relevant lock can be acquired under rskq_lock.
1701 * Splice the req list, so that accept() can not reach the pending ssk after
1702 * the listener socket is released below.
1704 spin_lock_bh(&queue->rskq_lock);
1705 head = queue->rskq_accept_head;
1706 tail = queue->rskq_accept_tail;
1707 queue->rskq_accept_head = NULL;
1708 queue->rskq_accept_tail = NULL;
1709 spin_unlock_bh(&queue->rskq_lock);
1714 /* can't acquire the msk socket lock under the subflow one,
1715 * or will cause ABBA deadlock
1717 release_sock(listener_ssk);
1719 for (req = head; req; req = req->dl_next) {
1721 if (!sk_is_mptcp(ssk))
1724 subflow = mptcp_subflow_ctx(ssk);
1725 if (!subflow || !subflow->conn)
1731 lock_sock_nested(sk, SINGLE_DEPTH_NESTING);
1732 __mptcp_unaccepted_force_close(sk);
1735 /* lockdep will report a false positive ABBA deadlock
1736 * between cancel_work_sync and the listener socket.
1737 * The involved locks belong to different sockets WRT
1738 * the existing AB chain.
1739 * Using a per socket key is problematic as key
1740 * deregistration requires process context and must be
1741 * performed at socket disposal time, in atomic
1743 * Just tell lockdep to consider the listener socket
1746 mutex_release(&listener_sk->sk_lock.dep_map, _RET_IP_);
1747 mptcp_cancel_work(sk);
1748 mutex_acquire(&listener_sk->sk_lock.dep_map, 0, 0, _RET_IP_);
1753 /* we are still under the listener msk socket lock */
1754 lock_sock_nested(listener_ssk, SINGLE_DEPTH_NESTING);
1756 /* restore the listener queue, to let the TCP code clean it up */
1757 spin_lock_bh(&queue->rskq_lock);
1758 WARN_ON_ONCE(queue->rskq_accept_head);
1759 queue->rskq_accept_head = head;
1760 queue->rskq_accept_tail = tail;
1761 spin_unlock_bh(&queue->rskq_lock);
1764 static int subflow_ulp_init(struct sock *sk)
1766 struct inet_connection_sock *icsk = inet_csk(sk);
1767 struct mptcp_subflow_context *ctx;
1768 struct tcp_sock *tp = tcp_sk(sk);
1771 /* disallow attaching ULP to a socket unless it has been
1772 * created with sock_create_kern()
1774 if (!sk->sk_kern_sock) {
1779 ctx = subflow_create_ctx(sk, GFP_KERNEL);
1785 pr_debug("subflow=%p, family=%d", ctx, sk->sk_family);
1788 ctx->icsk_af_ops = icsk->icsk_af_ops;
1789 icsk->icsk_af_ops = subflow_default_af_ops(sk);
1790 ctx->tcp_state_change = sk->sk_state_change;
1791 ctx->tcp_error_report = sk->sk_error_report;
1793 WARN_ON_ONCE(sk->sk_data_ready != sock_def_readable);
1794 WARN_ON_ONCE(sk->sk_write_space != sk_stream_write_space);
1796 sk->sk_data_ready = subflow_data_ready;
1797 sk->sk_write_space = subflow_write_space;
1798 sk->sk_state_change = subflow_state_change;
1799 sk->sk_error_report = subflow_error_report;
1804 static void subflow_ulp_release(struct sock *ssk)
1806 struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(ssk);
1807 bool release = true;
1815 /* if the msk has been orphaned, keep the ctx
1816 * alive, will be freed by __mptcp_close_ssk(),
1817 * when the subflow is still unaccepted
1819 release = ctx->disposable || list_empty(&ctx->node);
1821 /* inet_child_forget() does not call sk_state_change(),
1822 * explicitly trigger the socket close machinery
1824 if (!release && !test_and_set_bit(MPTCP_WORK_CLOSE_SUBFLOW,
1825 &mptcp_sk(sk)->flags))
1826 mptcp_schedule_work(sk);
1830 mptcp_subflow_ops_undo_override(ssk);
1832 kfree_rcu(ctx, rcu);
1835 static void subflow_ulp_clone(const struct request_sock *req,
1837 const gfp_t priority)
1839 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
1840 struct mptcp_subflow_context *old_ctx = mptcp_subflow_ctx(newsk);
1841 struct mptcp_subflow_context *new_ctx;
1843 if (!tcp_rsk(req)->is_mptcp ||
1844 (!subflow_req->mp_capable && !subflow_req->mp_join)) {
1845 subflow_ulp_fallback(newsk, old_ctx);
1849 new_ctx = subflow_create_ctx(newsk, priority);
1851 subflow_ulp_fallback(newsk, old_ctx);
1855 new_ctx->conn_finished = 1;
1856 new_ctx->icsk_af_ops = old_ctx->icsk_af_ops;
1857 new_ctx->tcp_state_change = old_ctx->tcp_state_change;
1858 new_ctx->tcp_error_report = old_ctx->tcp_error_report;
1859 new_ctx->rel_write_seq = 1;
1860 new_ctx->tcp_sock = newsk;
1862 if (subflow_req->mp_capable) {
1863 /* see comments in subflow_syn_recv_sock(), MPTCP connection
1864 * is fully established only after we receive the remote key
1866 new_ctx->mp_capable = 1;
1867 new_ctx->local_key = subflow_req->local_key;
1868 new_ctx->token = subflow_req->token;
1869 new_ctx->ssn_offset = subflow_req->ssn_offset;
1870 new_ctx->idsn = subflow_req->idsn;
1872 /* this is the first subflow, id is always 0 */
1873 subflow_set_local_id(new_ctx, 0);
1874 } else if (subflow_req->mp_join) {
1875 new_ctx->ssn_offset = subflow_req->ssn_offset;
1876 new_ctx->mp_join = 1;
1877 new_ctx->fully_established = 1;
1878 new_ctx->backup = subflow_req->backup;
1879 WRITE_ONCE(new_ctx->remote_id, subflow_req->remote_id);
1880 new_ctx->token = subflow_req->token;
1881 new_ctx->thmac = subflow_req->thmac;
1883 /* the subflow req id is valid, fetched via subflow_check_req()
1884 * and subflow_token_join_request()
1886 subflow_set_local_id(new_ctx, subflow_req->local_id);
1890 static void tcp_release_cb_override(struct sock *ssk)
1892 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1895 /* process and clear all the pending actions, but leave the subflow into
1896 * the napi queue. To respect locking, only the same CPU that originated
1897 * the action can touch the list. mptcp_napi_poll will take care of it.
1899 status = set_mask_bits(&subflow->delegated_status, MPTCP_DELEGATE_ACTIONS_MASK, 0);
1901 mptcp_subflow_process_delegated(ssk, status);
1903 tcp_release_cb(ssk);
1906 static int tcp_abort_override(struct sock *ssk, int err)
1908 /* closing a listener subflow requires a great deal of care.
1909 * keep it simple and just prevent such operation
1911 if (inet_sk_state_load(ssk) == TCP_LISTEN)
1914 return tcp_abort(ssk, err);
1917 static struct tcp_ulp_ops subflow_ulp_ops __read_mostly = {
1919 .owner = THIS_MODULE,
1920 .init = subflow_ulp_init,
1921 .release = subflow_ulp_release,
1922 .clone = subflow_ulp_clone,
1925 static int subflow_ops_init(struct request_sock_ops *subflow_ops)
1927 subflow_ops->obj_size = sizeof(struct mptcp_subflow_request_sock);
1929 subflow_ops->slab = kmem_cache_create(subflow_ops->slab_name,
1930 subflow_ops->obj_size, 0,
1932 SLAB_TYPESAFE_BY_RCU,
1934 if (!subflow_ops->slab)
1940 void __init mptcp_subflow_init(void)
1942 mptcp_subflow_v4_request_sock_ops = tcp_request_sock_ops;
1943 mptcp_subflow_v4_request_sock_ops.slab_name = "request_sock_subflow_v4";
1944 mptcp_subflow_v4_request_sock_ops.destructor = subflow_v4_req_destructor;
1946 if (subflow_ops_init(&mptcp_subflow_v4_request_sock_ops) != 0)
1947 panic("MPTCP: failed to init subflow v4 request sock ops\n");
1949 subflow_request_sock_ipv4_ops = tcp_request_sock_ipv4_ops;
1950 subflow_request_sock_ipv4_ops.route_req = subflow_v4_route_req;
1952 subflow_specific = ipv4_specific;
1953 subflow_specific.conn_request = subflow_v4_conn_request;
1954 subflow_specific.syn_recv_sock = subflow_syn_recv_sock;
1955 subflow_specific.sk_rx_dst_set = subflow_finish_connect;
1956 subflow_specific.rebuild_header = subflow_rebuild_header;
1958 tcp_prot_override = tcp_prot;
1959 tcp_prot_override.release_cb = tcp_release_cb_override;
1960 tcp_prot_override.diag_destroy = tcp_abort_override;
1962 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1963 /* In struct mptcp_subflow_request_sock, we assume the TCP request sock
1964 * structures for v4 and v6 have the same size. It should not changed in
1965 * the future but better to make sure to be warned if it is no longer
1968 BUILD_BUG_ON(sizeof(struct tcp_request_sock) != sizeof(struct tcp6_request_sock));
1970 mptcp_subflow_v6_request_sock_ops = tcp6_request_sock_ops;
1971 mptcp_subflow_v6_request_sock_ops.slab_name = "request_sock_subflow_v6";
1972 mptcp_subflow_v6_request_sock_ops.destructor = subflow_v6_req_destructor;
1974 if (subflow_ops_init(&mptcp_subflow_v6_request_sock_ops) != 0)
1975 panic("MPTCP: failed to init subflow v6 request sock ops\n");
1977 subflow_request_sock_ipv6_ops = tcp_request_sock_ipv6_ops;
1978 subflow_request_sock_ipv6_ops.route_req = subflow_v6_route_req;
1980 subflow_v6_specific = ipv6_specific;
1981 subflow_v6_specific.conn_request = subflow_v6_conn_request;
1982 subflow_v6_specific.syn_recv_sock = subflow_syn_recv_sock;
1983 subflow_v6_specific.sk_rx_dst_set = subflow_finish_connect;
1984 subflow_v6_specific.rebuild_header = subflow_v6_rebuild_header;
1986 subflow_v6m_specific = subflow_v6_specific;
1987 subflow_v6m_specific.queue_xmit = ipv4_specific.queue_xmit;
1988 subflow_v6m_specific.send_check = ipv4_specific.send_check;
1989 subflow_v6m_specific.net_header_len = ipv4_specific.net_header_len;
1990 subflow_v6m_specific.mtu_reduced = ipv4_specific.mtu_reduced;
1991 subflow_v6m_specific.net_frag_header_len = 0;
1992 subflow_v6m_specific.rebuild_header = subflow_rebuild_header;
1994 tcpv6_prot_override = tcpv6_prot;
1995 tcpv6_prot_override.release_cb = tcp_release_cb_override;
1996 tcpv6_prot_override.diag_destroy = tcp_abort_override;
1999 mptcp_diag_subflow_init(&subflow_ulp_ops);
2001 if (tcp_register_ulp(&subflow_ulp_ops) != 0)
2002 panic("MPTCP: failed to register subflows to ULP\n");