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);
48 tcp_request_sock_ops.destructor(req);
51 static void subflow_generate_hmac(u64 key1, u64 key2, u32 nonce1, u32 nonce2,
56 put_unaligned_be32(nonce1, &msg[0]);
57 put_unaligned_be32(nonce2, &msg[4]);
59 mptcp_crypto_hmac_sha(key1, key2, msg, 8, hmac);
62 static bool mptcp_can_accept_new_subflow(const struct mptcp_sock *msk)
64 return mptcp_is_fully_established((void *)msk) &&
65 ((mptcp_pm_is_userspace(msk) &&
66 mptcp_userspace_pm_active(msk)) ||
67 READ_ONCE(msk->pm.accept_subflow));
70 /* validate received token and create truncated hmac and nonce for SYN-ACK */
71 static void subflow_req_create_thmac(struct mptcp_subflow_request_sock *subflow_req)
73 struct mptcp_sock *msk = subflow_req->msk;
74 u8 hmac[SHA256_DIGEST_SIZE];
76 get_random_bytes(&subflow_req->local_nonce, sizeof(u32));
78 subflow_generate_hmac(msk->local_key, msk->remote_key,
79 subflow_req->local_nonce,
80 subflow_req->remote_nonce, hmac);
82 subflow_req->thmac = get_unaligned_be64(hmac);
85 static struct mptcp_sock *subflow_token_join_request(struct request_sock *req)
87 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
88 struct mptcp_sock *msk;
91 msk = mptcp_token_get_sock(sock_net(req_to_sk(req)), subflow_req->token);
93 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINNOTOKEN);
97 local_id = mptcp_pm_get_local_id(msk, (struct sock_common *)req);
99 sock_put((struct sock *)msk);
102 subflow_req->local_id = local_id;
107 static void subflow_init_req(struct request_sock *req, const struct sock *sk_listener)
109 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
111 subflow_req->mp_capable = 0;
112 subflow_req->mp_join = 0;
113 subflow_req->csum_reqd = mptcp_is_checksum_enabled(sock_net(sk_listener));
114 subflow_req->allow_join_id0 = mptcp_allow_join_id0(sock_net(sk_listener));
115 subflow_req->msk = NULL;
116 mptcp_token_init_request(req);
119 static bool subflow_use_different_sport(struct mptcp_sock *msk, const struct sock *sk)
121 return inet_sk(sk)->inet_sport != inet_sk((struct sock *)msk)->inet_sport;
124 static void subflow_add_reset_reason(struct sk_buff *skb, u8 reason)
126 struct mptcp_ext *mpext = skb_ext_add(skb, SKB_EXT_MPTCP);
129 memset(mpext, 0, sizeof(*mpext));
130 mpext->reset_reason = reason;
134 /* Init mptcp request socket.
136 * Returns an error code if a JOIN has failed and a TCP reset
139 static int subflow_check_req(struct request_sock *req,
140 const struct sock *sk_listener,
143 struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk_listener);
144 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
145 struct mptcp_options_received mp_opt;
146 bool opt_mp_capable, opt_mp_join;
148 pr_debug("subflow_req=%p, listener=%p", subflow_req, listener);
150 #ifdef CONFIG_TCP_MD5SIG
151 /* no MPTCP if MD5SIG is enabled on this socket or we may run out of
154 if (rcu_access_pointer(tcp_sk(sk_listener)->md5sig_info))
158 mptcp_get_options(skb, &mp_opt);
160 opt_mp_capable = !!(mp_opt.suboptions & OPTIONS_MPTCP_MPC);
161 opt_mp_join = !!(mp_opt.suboptions & OPTIONS_MPTCP_MPJ);
162 if (opt_mp_capable) {
163 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MPCAPABLEPASSIVE);
167 } else if (opt_mp_join) {
168 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINSYNRX);
171 if (opt_mp_capable && listener->request_mptcp) {
172 int err, retries = MPTCP_TOKEN_MAX_RETRIES;
174 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq;
177 get_random_bytes(&subflow_req->local_key, sizeof(subflow_req->local_key));
178 } while (subflow_req->local_key == 0);
180 if (unlikely(req->syncookie)) {
181 mptcp_crypto_key_sha(subflow_req->local_key,
184 if (mptcp_token_exists(subflow_req->token)) {
187 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_TOKENFALLBACKINIT);
189 subflow_req->mp_capable = 1;
194 err = mptcp_token_new_request(req);
196 subflow_req->mp_capable = 1;
197 else if (retries-- > 0)
200 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_TOKENFALLBACKINIT);
202 } else if (opt_mp_join && listener->request_mptcp) {
203 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq;
204 subflow_req->mp_join = 1;
205 subflow_req->backup = mp_opt.backup;
206 subflow_req->remote_id = mp_opt.join_id;
207 subflow_req->token = mp_opt.token;
208 subflow_req->remote_nonce = mp_opt.nonce;
209 subflow_req->msk = subflow_token_join_request(req);
211 /* Can't fall back to TCP in this case. */
212 if (!subflow_req->msk) {
213 subflow_add_reset_reason(skb, MPTCP_RST_EMPTCP);
217 if (subflow_use_different_sport(subflow_req->msk, sk_listener)) {
218 pr_debug("syn inet_sport=%d %d",
219 ntohs(inet_sk(sk_listener)->inet_sport),
220 ntohs(inet_sk((struct sock *)subflow_req->msk)->inet_sport));
221 if (!mptcp_pm_sport_in_anno_list(subflow_req->msk, sk_listener)) {
222 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MISMATCHPORTSYNRX);
225 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINPORTSYNRX);
228 subflow_req_create_thmac(subflow_req);
230 if (unlikely(req->syncookie)) {
231 if (mptcp_can_accept_new_subflow(subflow_req->msk))
232 subflow_init_req_cookie_join_save(subflow_req, skb);
237 pr_debug("token=%u, remote_nonce=%u msk=%p", subflow_req->token,
238 subflow_req->remote_nonce, subflow_req->msk);
244 int mptcp_subflow_init_cookie_req(struct request_sock *req,
245 const struct sock *sk_listener,
248 struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk_listener);
249 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
250 struct mptcp_options_received mp_opt;
251 bool opt_mp_capable, opt_mp_join;
254 subflow_init_req(req, sk_listener);
255 mptcp_get_options(skb, &mp_opt);
257 opt_mp_capable = !!(mp_opt.suboptions & OPTIONS_MPTCP_MPC);
258 opt_mp_join = !!(mp_opt.suboptions & OPTIONS_MPTCP_MPJ);
259 if (opt_mp_capable && opt_mp_join)
262 if (opt_mp_capable && listener->request_mptcp) {
263 if (mp_opt.sndr_key == 0)
266 subflow_req->local_key = mp_opt.rcvr_key;
267 err = mptcp_token_new_request(req);
271 subflow_req->mp_capable = 1;
272 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq - 1;
273 } else if (opt_mp_join && listener->request_mptcp) {
274 if (!mptcp_token_join_cookie_init_state(subflow_req, skb))
277 subflow_req->mp_join = 1;
278 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq - 1;
283 EXPORT_SYMBOL_GPL(mptcp_subflow_init_cookie_req);
285 static struct dst_entry *subflow_v4_route_req(const struct sock *sk,
288 struct request_sock *req)
290 struct dst_entry *dst;
293 tcp_rsk(req)->is_mptcp = 1;
294 subflow_init_req(req, sk);
296 dst = tcp_request_sock_ipv4_ops.route_req(sk, skb, fl, req);
300 err = subflow_check_req(req, sk, skb);
306 tcp_request_sock_ops.send_reset(sk, skb);
310 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
311 static struct dst_entry *subflow_v6_route_req(const struct sock *sk,
314 struct request_sock *req)
316 struct dst_entry *dst;
319 tcp_rsk(req)->is_mptcp = 1;
320 subflow_init_req(req, sk);
322 dst = tcp_request_sock_ipv6_ops.route_req(sk, skb, fl, req);
326 err = subflow_check_req(req, sk, skb);
332 tcp6_request_sock_ops.send_reset(sk, skb);
337 /* validate received truncated hmac and create hmac for third ACK */
338 static bool subflow_thmac_valid(struct mptcp_subflow_context *subflow)
340 u8 hmac[SHA256_DIGEST_SIZE];
343 subflow_generate_hmac(subflow->remote_key, subflow->local_key,
344 subflow->remote_nonce, subflow->local_nonce,
347 thmac = get_unaligned_be64(hmac);
348 pr_debug("subflow=%p, token=%u, thmac=%llu, subflow->thmac=%llu\n",
349 subflow, subflow->token, thmac, subflow->thmac);
351 return thmac == subflow->thmac;
354 void mptcp_subflow_reset(struct sock *ssk)
356 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
357 struct sock *sk = subflow->conn;
359 /* must hold: tcp_done() could drop last reference on parent */
362 tcp_set_state(ssk, TCP_CLOSE);
363 tcp_send_active_reset(ssk, GFP_ATOMIC);
365 if (!test_and_set_bit(MPTCP_WORK_CLOSE_SUBFLOW, &mptcp_sk(sk)->flags) &&
366 schedule_work(&mptcp_sk(sk)->work))
367 return; /* worker will put sk for us */
372 static bool subflow_use_different_dport(struct mptcp_sock *msk, const struct sock *sk)
374 return inet_sk(sk)->inet_dport != inet_sk((struct sock *)msk)->inet_dport;
377 void __mptcp_set_connected(struct sock *sk)
379 if (sk->sk_state == TCP_SYN_SENT) {
380 inet_sk_state_store(sk, TCP_ESTABLISHED);
381 sk->sk_state_change(sk);
385 static void mptcp_set_connected(struct sock *sk)
388 if (!sock_owned_by_user(sk))
389 __mptcp_set_connected(sk);
391 __set_bit(MPTCP_CONNECTED, &mptcp_sk(sk)->cb_flags);
392 mptcp_data_unlock(sk);
395 static void subflow_finish_connect(struct sock *sk, const struct sk_buff *skb)
397 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
398 struct mptcp_options_received mp_opt;
399 struct sock *parent = subflow->conn;
401 subflow->icsk_af_ops->sk_rx_dst_set(sk, skb);
403 /* be sure no special action on any packet other than syn-ack */
404 if (subflow->conn_finished)
407 mptcp_propagate_sndbuf(parent, sk);
408 subflow->rel_write_seq = 1;
409 subflow->conn_finished = 1;
410 subflow->ssn_offset = TCP_SKB_CB(skb)->seq;
411 pr_debug("subflow=%p synack seq=%x", subflow, subflow->ssn_offset);
413 mptcp_get_options(skb, &mp_opt);
414 if (subflow->request_mptcp) {
415 if (!(mp_opt.suboptions & OPTIONS_MPTCP_MPC)) {
416 MPTCP_INC_STATS(sock_net(sk),
417 MPTCP_MIB_MPCAPABLEACTIVEFALLBACK);
418 mptcp_do_fallback(sk);
419 pr_fallback(mptcp_sk(subflow->conn));
423 if (mp_opt.suboptions & OPTION_MPTCP_CSUMREQD)
424 WRITE_ONCE(mptcp_sk(parent)->csum_enabled, true);
425 if (mp_opt.deny_join_id0)
426 WRITE_ONCE(mptcp_sk(parent)->pm.remote_deny_join_id0, true);
427 subflow->mp_capable = 1;
428 subflow->can_ack = 1;
429 subflow->remote_key = mp_opt.sndr_key;
430 pr_debug("subflow=%p, remote_key=%llu", subflow,
431 subflow->remote_key);
432 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPCAPABLEACTIVEACK);
433 mptcp_finish_connect(sk);
434 mptcp_set_connected(parent);
435 } else if (subflow->request_join) {
436 u8 hmac[SHA256_DIGEST_SIZE];
438 if (!(mp_opt.suboptions & OPTIONS_MPTCP_MPJ)) {
439 subflow->reset_reason = MPTCP_RST_EMPTCP;
443 subflow->backup = mp_opt.backup;
444 subflow->thmac = mp_opt.thmac;
445 subflow->remote_nonce = mp_opt.nonce;
446 subflow->remote_id = mp_opt.join_id;
447 pr_debug("subflow=%p, thmac=%llu, remote_nonce=%u backup=%d",
448 subflow, subflow->thmac, subflow->remote_nonce,
451 if (!subflow_thmac_valid(subflow)) {
452 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINACKMAC);
453 subflow->reset_reason = MPTCP_RST_EMPTCP;
457 if (!mptcp_finish_join(sk))
460 subflow_generate_hmac(subflow->local_key, subflow->remote_key,
461 subflow->local_nonce,
462 subflow->remote_nonce,
464 memcpy(subflow->hmac, hmac, MPTCPOPT_HMAC_LEN);
466 subflow->mp_join = 1;
467 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINSYNACKRX);
469 if (subflow_use_different_dport(mptcp_sk(parent), sk)) {
470 pr_debug("synack inet_dport=%d %d",
471 ntohs(inet_sk(sk)->inet_dport),
472 ntohs(inet_sk(parent)->inet_dport));
473 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINPORTSYNACKRX);
475 } else if (mptcp_check_fallback(sk)) {
477 mptcp_rcv_space_init(mptcp_sk(parent), sk);
478 mptcp_set_connected(parent);
483 subflow->reset_transient = 0;
484 mptcp_subflow_reset(sk);
487 static void subflow_set_local_id(struct mptcp_subflow_context *subflow, int local_id)
489 subflow->local_id = local_id;
490 subflow->local_id_valid = 1;
493 static int subflow_chk_local_id(struct sock *sk)
495 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
496 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
499 if (likely(subflow->local_id_valid))
502 err = mptcp_pm_get_local_id(msk, (struct sock_common *)sk);
506 subflow_set_local_id(subflow, err);
510 static int subflow_rebuild_header(struct sock *sk)
512 int err = subflow_chk_local_id(sk);
514 if (unlikely(err < 0))
517 return inet_sk_rebuild_header(sk);
520 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
521 static int subflow_v6_rebuild_header(struct sock *sk)
523 int err = subflow_chk_local_id(sk);
525 if (unlikely(err < 0))
528 return inet6_sk_rebuild_header(sk);
532 struct request_sock_ops mptcp_subflow_request_sock_ops;
533 static struct tcp_request_sock_ops subflow_request_sock_ipv4_ops __ro_after_init;
535 static int subflow_v4_conn_request(struct sock *sk, struct sk_buff *skb)
537 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
539 pr_debug("subflow=%p", subflow);
541 /* Never answer to SYNs sent to broadcast or multicast */
542 if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
545 return tcp_conn_request(&mptcp_subflow_request_sock_ops,
546 &subflow_request_sock_ipv4_ops,
553 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
554 static struct tcp_request_sock_ops subflow_request_sock_ipv6_ops __ro_after_init;
555 static struct inet_connection_sock_af_ops subflow_v6_specific __ro_after_init;
556 static struct inet_connection_sock_af_ops subflow_v6m_specific __ro_after_init;
557 static struct proto tcpv6_prot_override;
559 static int subflow_v6_conn_request(struct sock *sk, struct sk_buff *skb)
561 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
563 pr_debug("subflow=%p", subflow);
565 if (skb->protocol == htons(ETH_P_IP))
566 return subflow_v4_conn_request(sk, skb);
568 if (!ipv6_unicast_destination(skb))
571 if (ipv6_addr_v4mapped(&ipv6_hdr(skb)->saddr)) {
572 __IP6_INC_STATS(sock_net(sk), NULL, IPSTATS_MIB_INHDRERRORS);
576 return tcp_conn_request(&mptcp_subflow_request_sock_ops,
577 &subflow_request_sock_ipv6_ops, sk, skb);
581 return 0; /* don't send reset */
585 /* validate hmac received in third ACK */
586 static bool subflow_hmac_valid(const struct request_sock *req,
587 const struct mptcp_options_received *mp_opt)
589 const struct mptcp_subflow_request_sock *subflow_req;
590 u8 hmac[SHA256_DIGEST_SIZE];
591 struct mptcp_sock *msk;
593 subflow_req = mptcp_subflow_rsk(req);
594 msk = subflow_req->msk;
598 subflow_generate_hmac(msk->remote_key, msk->local_key,
599 subflow_req->remote_nonce,
600 subflow_req->local_nonce, hmac);
602 return !crypto_memneq(hmac, mp_opt->hmac, MPTCPOPT_HMAC_LEN);
605 static void mptcp_sock_destruct(struct sock *sk)
607 /* if new mptcp socket isn't accepted, it is free'd
608 * from the tcp listener sockets request queue, linked
609 * from req->sk. The tcp socket is released.
610 * This calls the ULP release function which will
611 * also remove the mptcp socket, via
612 * sock_put(ctx->conn).
614 * Problem is that the mptcp socket will be in
615 * ESTABLISHED state and will not have the SOCK_DEAD flag.
616 * Both result in warnings from inet_sock_destruct.
618 if ((1 << sk->sk_state) & (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) {
619 sk->sk_state = TCP_CLOSE;
620 WARN_ON_ONCE(sk->sk_socket);
624 mptcp_destroy_common(mptcp_sk(sk));
625 inet_sock_destruct(sk);
628 static void mptcp_force_close(struct sock *sk)
630 /* the msk is not yet exposed to user-space */
631 inet_sk_state_store(sk, TCP_CLOSE);
632 sk_common_release(sk);
635 static void subflow_ulp_fallback(struct sock *sk,
636 struct mptcp_subflow_context *old_ctx)
638 struct inet_connection_sock *icsk = inet_csk(sk);
640 mptcp_subflow_tcp_fallback(sk, old_ctx);
641 icsk->icsk_ulp_ops = NULL;
642 rcu_assign_pointer(icsk->icsk_ulp_data, NULL);
643 tcp_sk(sk)->is_mptcp = 0;
645 mptcp_subflow_ops_undo_override(sk);
648 static void subflow_drop_ctx(struct sock *ssk)
650 struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(ssk);
655 subflow_ulp_fallback(ssk, ctx);
662 void mptcp_subflow_fully_established(struct mptcp_subflow_context *subflow,
663 struct mptcp_options_received *mp_opt)
665 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
667 subflow->remote_key = mp_opt->sndr_key;
668 subflow->fully_established = 1;
669 subflow->can_ack = 1;
670 WRITE_ONCE(msk->fully_established, true);
673 static struct sock *subflow_syn_recv_sock(const struct sock *sk,
675 struct request_sock *req,
676 struct dst_entry *dst,
677 struct request_sock *req_unhash,
680 struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk);
681 struct mptcp_subflow_request_sock *subflow_req;
682 struct mptcp_options_received mp_opt;
683 bool fallback, fallback_is_fatal;
684 struct sock *new_msk = NULL;
687 pr_debug("listener=%p, req=%p, conn=%p", listener, req, listener->conn);
689 /* After child creation we must look for MPC even when options
692 mp_opt.suboptions = 0;
694 /* hopefully temporary handling for MP_JOIN+syncookie */
695 subflow_req = mptcp_subflow_rsk(req);
696 fallback_is_fatal = tcp_rsk(req)->is_mptcp && subflow_req->mp_join;
697 fallback = !tcp_rsk(req)->is_mptcp;
701 /* if the sk is MP_CAPABLE, we try to fetch the client key */
702 if (subflow_req->mp_capable) {
703 /* we can receive and accept an in-window, out-of-order pkt,
704 * which may not carry the MP_CAPABLE opt even on mptcp enabled
705 * paths: always try to extract the peer key, and fallback
706 * for packets missing it.
707 * Even OoO DSS packets coming legitly after dropped or
708 * reordered MPC will cause fallback, but we don't have other
711 mptcp_get_options(skb, &mp_opt);
712 if (!(mp_opt.suboptions & OPTIONS_MPTCP_MPC)) {
717 new_msk = mptcp_sk_clone(listener->conn, &mp_opt, req);
720 } else if (subflow_req->mp_join) {
721 mptcp_get_options(skb, &mp_opt);
722 if (!(mp_opt.suboptions & OPTIONS_MPTCP_MPJ) ||
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);
749 subflow_drop_ctx(child);
753 /* ssk inherits options of listener sk */
754 ctx->setsockopt_seq = listener->setsockopt_seq;
756 if (ctx->mp_capable) {
757 /* this can't race with mptcp_close(), as the msk is
758 * not yet exposted to user-space
760 inet_sk_state_store((void *)new_msk, TCP_ESTABLISHED);
762 /* record the newly created socket as the first msk
763 * subflow, but don't link it yet into conn_list
765 WRITE_ONCE(mptcp_sk(new_msk)->first, child);
767 /* new mpc subflow takes ownership of the newly
768 * created mptcp socket
770 new_msk->sk_destruct = mptcp_sock_destruct;
771 mptcp_sk(new_msk)->setsockopt_seq = ctx->setsockopt_seq;
772 mptcp_pm_new_connection(mptcp_sk(new_msk), child, 1);
773 mptcp_token_accept(subflow_req, mptcp_sk(new_msk));
777 /* with OoO packets we can reach here without ingress
780 if (mp_opt.suboptions & OPTIONS_MPTCP_MPC)
781 mptcp_subflow_fully_established(ctx, &mp_opt);
782 } else if (ctx->mp_join) {
783 struct mptcp_sock *owner;
785 owner = subflow_req->msk;
787 subflow_add_reset_reason(skb, MPTCP_RST_EPROHIBIT);
791 /* move the msk reference ownership to the subflow */
792 subflow_req->msk = NULL;
793 ctx->conn = (struct sock *)owner;
795 if (subflow_use_different_sport(owner, sk)) {
796 pr_debug("ack inet_sport=%d %d",
797 ntohs(inet_sk(sk)->inet_sport),
798 ntohs(inet_sk((struct sock *)owner)->inet_sport));
799 if (!mptcp_pm_sport_in_anno_list(owner, sk)) {
800 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MISMATCHPORTACKRX);
803 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINPORTACKRX);
806 if (!mptcp_finish_join(child))
809 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKRX);
810 tcp_rsk(req)->drop_req = true;
815 /* dispose of the left over mptcp master, if any */
816 if (unlikely(new_msk))
817 mptcp_force_close(new_msk);
819 /* check for expected invariant - should never trigger, just help
820 * catching eariler subtle bugs
822 WARN_ON_ONCE(child && *own_req && tcp_sk(child)->is_mptcp &&
823 (!mptcp_subflow_ctx(child) ||
824 !mptcp_subflow_ctx(child)->conn));
828 subflow_drop_ctx(child);
829 tcp_rsk(req)->drop_req = true;
830 inet_csk_prepare_for_destroy_sock(child);
832 req->rsk_ops->send_reset(sk, skb);
834 /* The last child reference will be released by the caller */
838 static struct inet_connection_sock_af_ops subflow_specific __ro_after_init;
839 static struct proto tcp_prot_override;
841 enum mapping_status {
849 static void dbg_bad_map(struct mptcp_subflow_context *subflow, u32 ssn)
851 pr_debug("Bad mapping: ssn=%d map_seq=%d map_data_len=%d",
852 ssn, subflow->map_subflow_seq, subflow->map_data_len);
855 static bool skb_is_fully_mapped(struct sock *ssk, struct sk_buff *skb)
857 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
858 unsigned int skb_consumed;
860 skb_consumed = tcp_sk(ssk)->copied_seq - TCP_SKB_CB(skb)->seq;
861 if (WARN_ON_ONCE(skb_consumed >= skb->len))
864 return skb->len - skb_consumed <= subflow->map_data_len -
865 mptcp_subflow_get_map_offset(subflow);
868 static bool validate_mapping(struct sock *ssk, struct sk_buff *skb)
870 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
871 u32 ssn = tcp_sk(ssk)->copied_seq - subflow->ssn_offset;
873 if (unlikely(before(ssn, subflow->map_subflow_seq))) {
874 /* Mapping covers data later in the subflow stream,
875 * currently unsupported.
877 dbg_bad_map(subflow, ssn);
880 if (unlikely(!before(ssn, subflow->map_subflow_seq +
881 subflow->map_data_len))) {
882 /* Mapping does covers past subflow data, invalid */
883 dbg_bad_map(subflow, ssn);
889 static enum mapping_status validate_data_csum(struct sock *ssk, struct sk_buff *skb,
892 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
893 u32 offset, seq, delta;
900 /* mapping already validated on previous traversal */
901 if (subflow->map_csum_len == subflow->map_data_len)
904 /* traverse the receive queue, ensuring it contains a full
905 * DSS mapping and accumulating the related csum.
906 * Preserve the accoumlate csum across multiple calls, to compute
909 delta = subflow->map_data_len - subflow->map_csum_len;
911 seq = tcp_sk(ssk)->copied_seq + subflow->map_csum_len;
912 offset = seq - TCP_SKB_CB(skb)->seq;
914 /* if the current skb has not been accounted yet, csum its contents
915 * up to the amount covered by the current DSS
917 if (offset < skb->len) {
920 len = min(skb->len - offset, delta);
921 csum = skb_checksum(skb, offset, len, 0);
922 subflow->map_data_csum = csum_block_add(subflow->map_data_csum, csum,
923 subflow->map_csum_len);
926 subflow->map_csum_len += len;
931 if (skb_queue_is_last(&ssk->sk_receive_queue, skb)) {
932 /* if this subflow is closed, the partial mapping
933 * will be never completed; flush the pending skbs, so
934 * that subflow_sched_work_if_closed() can kick in
936 if (unlikely(ssk->sk_state == TCP_CLOSE))
937 while ((skb = skb_peek(&ssk->sk_receive_queue)))
938 sk_eat_skb(ssk, skb);
940 /* not enough data to validate the csum */
941 return MAPPING_EMPTY;
944 /* the DSS mapping for next skbs will be validated later,
945 * when a get_mapping_status call will process such skb
950 /* note that 'map_data_len' accounts only for the carried data, does
951 * not include the eventual seq increment due to the data fin,
952 * while the pseudo header requires the original DSS data len,
955 csum = __mptcp_make_csum(subflow->map_seq,
956 subflow->map_subflow_seq,
957 subflow->map_data_len + subflow->map_data_fin,
958 subflow->map_data_csum);
959 if (unlikely(csum)) {
960 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DATACSUMERR);
961 if (subflow->mp_join || subflow->valid_csum_seen) {
962 subflow->send_mp_fail = 1;
963 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_MPFAILTX);
965 return subflow->mp_join ? MAPPING_INVALID : MAPPING_DUMMY;
968 subflow->valid_csum_seen = 1;
972 static enum mapping_status get_mapping_status(struct sock *ssk,
973 struct mptcp_sock *msk)
975 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
976 bool csum_reqd = READ_ONCE(msk->csum_enabled);
977 struct sock *sk = (struct sock *)msk;
978 struct mptcp_ext *mpext;
983 skb = skb_peek(&ssk->sk_receive_queue);
985 return MAPPING_EMPTY;
987 if (mptcp_check_fallback(ssk))
988 return MAPPING_DUMMY;
990 mpext = mptcp_get_ext(skb);
991 if (!mpext || !mpext->use_map) {
992 if (!subflow->map_valid && !skb->len) {
993 /* the TCP stack deliver 0 len FIN pkt to the receive
994 * queue, that is the only 0len pkts ever expected here,
995 * and we can admit no mapping only for 0 len pkts
997 if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN))
998 WARN_ONCE(1, "0len seq %d:%d flags %x",
999 TCP_SKB_CB(skb)->seq,
1000 TCP_SKB_CB(skb)->end_seq,
1001 TCP_SKB_CB(skb)->tcp_flags);
1002 sk_eat_skb(ssk, skb);
1003 return MAPPING_EMPTY;
1006 if (!subflow->map_valid)
1007 return MAPPING_INVALID;
1012 trace_get_mapping_status(mpext);
1014 data_len = mpext->data_len;
1015 if (data_len == 0) {
1016 pr_debug("infinite mapping received");
1017 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_INFINITEMAPRX);
1018 subflow->map_data_len = 0;
1019 if (sk && inet_sk_state_load(sk) != TCP_CLOSE) {
1020 mptcp_data_lock(sk);
1021 if (inet_sk_state_load(sk) != TCP_CLOSE)
1022 sk_stop_timer(sk, &sk->sk_timer);
1023 mptcp_data_unlock(sk);
1025 return MAPPING_INVALID;
1028 if (mpext->data_fin == 1) {
1029 if (data_len == 1) {
1030 bool updated = mptcp_update_rcv_data_fin(msk, mpext->data_seq,
1032 pr_debug("DATA_FIN with no payload seq=%llu", mpext->data_seq);
1033 if (subflow->map_valid) {
1034 /* A DATA_FIN might arrive in a DSS
1035 * option before the previous mapping
1036 * has been fully consumed. Continue
1037 * handling the existing mapping.
1039 skb_ext_del(skb, SKB_EXT_MPTCP);
1042 if (updated && schedule_work(&msk->work))
1043 sock_hold((struct sock *)msk);
1045 return MAPPING_DATA_FIN;
1048 u64 data_fin_seq = mpext->data_seq + data_len - 1;
1050 /* If mpext->data_seq is a 32-bit value, data_fin_seq
1051 * must also be limited to 32 bits.
1054 data_fin_seq &= GENMASK_ULL(31, 0);
1056 mptcp_update_rcv_data_fin(msk, data_fin_seq, mpext->dsn64);
1057 pr_debug("DATA_FIN with mapping seq=%llu dsn64=%d",
1058 data_fin_seq, mpext->dsn64);
1061 /* Adjust for DATA_FIN using 1 byte of sequence space */
1065 map_seq = mptcp_expand_seq(READ_ONCE(msk->ack_seq), mpext->data_seq, mpext->dsn64);
1066 WRITE_ONCE(mptcp_sk(subflow->conn)->use_64bit_ack, !!mpext->dsn64);
1068 if (subflow->map_valid) {
1069 /* Allow replacing only with an identical map */
1070 if (subflow->map_seq == map_seq &&
1071 subflow->map_subflow_seq == mpext->subflow_seq &&
1072 subflow->map_data_len == data_len &&
1073 subflow->map_csum_reqd == mpext->csum_reqd) {
1074 skb_ext_del(skb, SKB_EXT_MPTCP);
1078 /* If this skb data are fully covered by the current mapping,
1079 * the new map would need caching, which is not supported
1081 if (skb_is_fully_mapped(ssk, skb)) {
1082 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DSSNOMATCH);
1083 return MAPPING_INVALID;
1086 /* will validate the next map after consuming the current one */
1090 subflow->map_seq = map_seq;
1091 subflow->map_subflow_seq = mpext->subflow_seq;
1092 subflow->map_data_len = data_len;
1093 subflow->map_valid = 1;
1094 subflow->map_data_fin = mpext->data_fin;
1095 subflow->mpc_map = mpext->mpc_map;
1096 subflow->map_csum_reqd = mpext->csum_reqd;
1097 subflow->map_csum_len = 0;
1098 subflow->map_data_csum = csum_unfold(mpext->csum);
1100 /* Cfr RFC 8684 Section 3.3.0 */
1101 if (unlikely(subflow->map_csum_reqd != csum_reqd))
1102 return MAPPING_INVALID;
1104 pr_debug("new map seq=%llu subflow_seq=%u data_len=%u csum=%d:%u",
1105 subflow->map_seq, subflow->map_subflow_seq,
1106 subflow->map_data_len, subflow->map_csum_reqd,
1107 subflow->map_data_csum);
1110 /* we revalidate valid mapping on new skb, because we must ensure
1111 * the current skb is completely covered by the available mapping
1113 if (!validate_mapping(ssk, skb)) {
1114 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DSSTCPMISMATCH);
1115 return MAPPING_INVALID;
1118 skb_ext_del(skb, SKB_EXT_MPTCP);
1121 return validate_data_csum(ssk, skb, csum_reqd);
1124 static void mptcp_subflow_discard_data(struct sock *ssk, struct sk_buff *skb,
1127 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1128 bool fin = TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN;
1131 incr = limit >= skb->len ? skb->len + fin : limit;
1133 pr_debug("discarding=%d len=%d seq=%d", incr, skb->len,
1134 subflow->map_subflow_seq);
1135 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DUPDATA);
1136 tcp_sk(ssk)->copied_seq += incr;
1137 if (!before(tcp_sk(ssk)->copied_seq, TCP_SKB_CB(skb)->end_seq))
1138 sk_eat_skb(ssk, skb);
1139 if (mptcp_subflow_get_map_offset(subflow) >= subflow->map_data_len)
1140 subflow->map_valid = 0;
1143 /* sched mptcp worker to remove the subflow if no more data is pending */
1144 static void subflow_sched_work_if_closed(struct mptcp_sock *msk, struct sock *ssk)
1146 struct sock *sk = (struct sock *)msk;
1148 if (likely(ssk->sk_state != TCP_CLOSE))
1151 if (skb_queue_empty(&ssk->sk_receive_queue) &&
1152 !test_and_set_bit(MPTCP_WORK_CLOSE_SUBFLOW, &msk->flags)) {
1154 if (!schedule_work(&msk->work))
1159 static bool subflow_can_fallback(struct mptcp_subflow_context *subflow)
1161 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
1163 if (subflow->mp_join)
1165 else if (READ_ONCE(msk->csum_enabled))
1166 return !subflow->valid_csum_seen;
1168 return !subflow->fully_established;
1171 static bool subflow_check_data_avail(struct sock *ssk)
1173 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1174 enum mapping_status status;
1175 struct mptcp_sock *msk;
1176 struct sk_buff *skb;
1178 if (!skb_peek(&ssk->sk_receive_queue))
1179 WRITE_ONCE(subflow->data_avail, MPTCP_SUBFLOW_NODATA);
1180 if (subflow->data_avail)
1183 msk = mptcp_sk(subflow->conn);
1188 status = get_mapping_status(ssk, msk);
1189 trace_subflow_check_data_avail(status, skb_peek(&ssk->sk_receive_queue));
1190 if (unlikely(status == MAPPING_INVALID))
1193 if (unlikely(status == MAPPING_DUMMY))
1196 if (status != MAPPING_OK)
1199 skb = skb_peek(&ssk->sk_receive_queue);
1200 if (WARN_ON_ONCE(!skb))
1203 /* if msk lacks the remote key, this subflow must provide an
1204 * MP_CAPABLE-based mapping
1206 if (unlikely(!READ_ONCE(msk->can_ack))) {
1207 if (!subflow->mpc_map)
1209 WRITE_ONCE(msk->remote_key, subflow->remote_key);
1210 WRITE_ONCE(msk->ack_seq, subflow->map_seq);
1211 WRITE_ONCE(msk->can_ack, true);
1214 old_ack = READ_ONCE(msk->ack_seq);
1215 ack_seq = mptcp_subflow_get_mapped_dsn(subflow);
1216 pr_debug("msk ack_seq=%llx subflow ack_seq=%llx", old_ack,
1218 if (unlikely(before64(ack_seq, old_ack))) {
1219 mptcp_subflow_discard_data(ssk, skb, old_ack - ack_seq);
1223 WRITE_ONCE(subflow->data_avail, MPTCP_SUBFLOW_DATA_AVAIL);
1229 subflow_sched_work_if_closed(msk, ssk);
1233 if (!__mptcp_check_fallback(msk)) {
1234 /* RFC 8684 section 3.7. */
1235 if (subflow->send_mp_fail) {
1236 if (mptcp_has_another_subflow(ssk) ||
1237 !READ_ONCE(msk->allow_infinite_fallback)) {
1238 ssk->sk_err = EBADMSG;
1239 tcp_set_state(ssk, TCP_CLOSE);
1240 subflow->reset_transient = 0;
1241 subflow->reset_reason = MPTCP_RST_EMIDDLEBOX;
1242 tcp_send_active_reset(ssk, GFP_ATOMIC);
1243 while ((skb = skb_peek(&ssk->sk_receive_queue)))
1244 sk_eat_skb(ssk, skb);
1246 WRITE_ONCE(subflow->mp_fail_response_expect, true);
1247 /* The data lock is acquired in __mptcp_move_skbs() */
1248 sk_reset_timer((struct sock *)msk,
1249 &((struct sock *)msk)->sk_timer,
1250 jiffies + TCP_RTO_MAX);
1252 WRITE_ONCE(subflow->data_avail, MPTCP_SUBFLOW_NODATA);
1256 if (!subflow_can_fallback(subflow) && subflow->map_data_len) {
1257 /* fatal protocol error, close the socket.
1258 * subflow_error_report() will introduce the appropriate barriers
1260 ssk->sk_err = EBADMSG;
1261 tcp_set_state(ssk, TCP_CLOSE);
1262 subflow->reset_transient = 0;
1263 subflow->reset_reason = MPTCP_RST_EMPTCP;
1264 tcp_send_active_reset(ssk, GFP_ATOMIC);
1265 WRITE_ONCE(subflow->data_avail, MPTCP_SUBFLOW_NODATA);
1269 __mptcp_do_fallback(msk);
1272 skb = skb_peek(&ssk->sk_receive_queue);
1273 subflow->map_valid = 1;
1274 subflow->map_seq = READ_ONCE(msk->ack_seq);
1275 subflow->map_data_len = skb->len;
1276 subflow->map_subflow_seq = tcp_sk(ssk)->copied_seq - subflow->ssn_offset;
1277 WRITE_ONCE(subflow->data_avail, MPTCP_SUBFLOW_DATA_AVAIL);
1281 bool mptcp_subflow_data_available(struct sock *sk)
1283 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1285 /* check if current mapping is still valid */
1286 if (subflow->map_valid &&
1287 mptcp_subflow_get_map_offset(subflow) >= subflow->map_data_len) {
1288 subflow->map_valid = 0;
1289 WRITE_ONCE(subflow->data_avail, MPTCP_SUBFLOW_NODATA);
1291 pr_debug("Done with mapping: seq=%u data_len=%u",
1292 subflow->map_subflow_seq,
1293 subflow->map_data_len);
1296 return subflow_check_data_avail(sk);
1299 /* If ssk has an mptcp parent socket, use the mptcp rcvbuf occupancy,
1302 * In mptcp, rwin is about the mptcp-level connection data.
1304 * Data that is still on the ssk rx queue can thus be ignored,
1305 * as far as mptcp peer is concerned that data is still inflight.
1306 * DSS ACK is updated when skb is moved to the mptcp rx queue.
1308 void mptcp_space(const struct sock *ssk, int *space, int *full_space)
1310 const struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1311 const struct sock *sk = subflow->conn;
1313 *space = __mptcp_space(sk);
1314 *full_space = tcp_full_space(sk);
1317 void __mptcp_error_report(struct sock *sk)
1319 struct mptcp_subflow_context *subflow;
1320 struct mptcp_sock *msk = mptcp_sk(sk);
1322 mptcp_for_each_subflow(msk, subflow) {
1323 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
1324 int err = sock_error(ssk);
1329 /* only propagate errors on fallen-back sockets or
1332 if (sk->sk_state != TCP_SYN_SENT && !__mptcp_check_fallback(msk))
1335 inet_sk_state_store(sk, inet_sk_state_load(ssk));
1338 /* This barrier is coupled with smp_rmb() in mptcp_poll() */
1340 sk_error_report(sk);
1345 static void subflow_error_report(struct sock *ssk)
1347 struct sock *sk = mptcp_subflow_ctx(ssk)->conn;
1349 mptcp_data_lock(sk);
1350 if (!sock_owned_by_user(sk))
1351 __mptcp_error_report(sk);
1353 __set_bit(MPTCP_ERROR_REPORT, &mptcp_sk(sk)->cb_flags);
1354 mptcp_data_unlock(sk);
1357 static void subflow_data_ready(struct sock *sk)
1359 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1360 u16 state = 1 << inet_sk_state_load(sk);
1361 struct sock *parent = subflow->conn;
1362 struct mptcp_sock *msk;
1364 msk = mptcp_sk(parent);
1365 if (state & TCPF_LISTEN) {
1366 /* MPJ subflow are removed from accept queue before reaching here,
1367 * avoid stray wakeups
1369 if (reqsk_queue_empty(&inet_csk(sk)->icsk_accept_queue))
1372 parent->sk_data_ready(parent);
1376 WARN_ON_ONCE(!__mptcp_check_fallback(msk) && !subflow->mp_capable &&
1377 !subflow->mp_join && !(state & TCPF_CLOSE));
1379 if (mptcp_subflow_data_available(sk))
1380 mptcp_data_ready(parent, sk);
1381 else if (unlikely(sk->sk_err))
1382 subflow_error_report(sk);
1385 static void subflow_write_space(struct sock *ssk)
1387 struct sock *sk = mptcp_subflow_ctx(ssk)->conn;
1389 mptcp_propagate_sndbuf(sk, ssk);
1390 mptcp_write_space(sk);
1393 static const struct inet_connection_sock_af_ops *
1394 subflow_default_af_ops(struct sock *sk)
1396 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1397 if (sk->sk_family == AF_INET6)
1398 return &subflow_v6_specific;
1400 return &subflow_specific;
1403 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1404 void mptcpv6_handle_mapped(struct sock *sk, bool mapped)
1406 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1407 struct inet_connection_sock *icsk = inet_csk(sk);
1408 const struct inet_connection_sock_af_ops *target;
1410 target = mapped ? &subflow_v6m_specific : subflow_default_af_ops(sk);
1412 pr_debug("subflow=%p family=%d ops=%p target=%p mapped=%d",
1413 subflow, sk->sk_family, icsk->icsk_af_ops, target, mapped);
1415 if (likely(icsk->icsk_af_ops == target))
1418 subflow->icsk_af_ops = icsk->icsk_af_ops;
1419 icsk->icsk_af_ops = target;
1423 void mptcp_info2sockaddr(const struct mptcp_addr_info *info,
1424 struct sockaddr_storage *addr,
1425 unsigned short family)
1427 memset(addr, 0, sizeof(*addr));
1428 addr->ss_family = family;
1429 if (addr->ss_family == AF_INET) {
1430 struct sockaddr_in *in_addr = (struct sockaddr_in *)addr;
1432 if (info->family == AF_INET)
1433 in_addr->sin_addr = info->addr;
1434 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1435 else if (ipv6_addr_v4mapped(&info->addr6))
1436 in_addr->sin_addr.s_addr = info->addr6.s6_addr32[3];
1438 in_addr->sin_port = info->port;
1440 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1441 else if (addr->ss_family == AF_INET6) {
1442 struct sockaddr_in6 *in6_addr = (struct sockaddr_in6 *)addr;
1444 if (info->family == AF_INET)
1445 ipv6_addr_set_v4mapped(info->addr.s_addr,
1446 &in6_addr->sin6_addr);
1448 in6_addr->sin6_addr = info->addr6;
1449 in6_addr->sin6_port = info->port;
1454 int __mptcp_subflow_connect(struct sock *sk, const struct mptcp_addr_info *loc,
1455 const struct mptcp_addr_info *remote)
1457 struct mptcp_sock *msk = mptcp_sk(sk);
1458 struct mptcp_subflow_context *subflow;
1459 struct sockaddr_storage addr;
1460 int remote_id = remote->id;
1461 int local_id = loc->id;
1462 int err = -ENOTCONN;
1470 if (!mptcp_is_fully_established(sk))
1473 err = mptcp_subflow_create_socket(sk, &sf);
1478 subflow = mptcp_subflow_ctx(ssk);
1480 get_random_bytes(&subflow->local_nonce, sizeof(u32));
1481 } while (!subflow->local_nonce);
1484 subflow_set_local_id(subflow, local_id);
1486 mptcp_pm_get_flags_and_ifindex_by_id(msk, local_id,
1488 subflow->remote_key = msk->remote_key;
1489 subflow->local_key = msk->local_key;
1490 subflow->token = msk->token;
1491 mptcp_info2sockaddr(loc, &addr, ssk->sk_family);
1493 addrlen = sizeof(struct sockaddr_in);
1494 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1495 if (addr.ss_family == AF_INET6)
1496 addrlen = sizeof(struct sockaddr_in6);
1498 mptcp_sockopt_sync(msk, ssk);
1500 ssk->sk_bound_dev_if = ifindex;
1501 err = kernel_bind(sf, (struct sockaddr *)&addr, addrlen);
1505 mptcp_crypto_key_sha(subflow->remote_key, &remote_token, NULL);
1506 pr_debug("msk=%p remote_token=%u local_id=%d remote_id=%d", msk,
1507 remote_token, local_id, remote_id);
1508 subflow->remote_token = remote_token;
1509 subflow->remote_id = remote_id;
1510 subflow->request_join = 1;
1511 subflow->request_bkup = !!(flags & MPTCP_PM_ADDR_FLAG_BACKUP);
1512 mptcp_info2sockaddr(remote, &addr, ssk->sk_family);
1515 list_add_tail(&subflow->node, &msk->conn_list);
1516 err = kernel_connect(sf, (struct sockaddr *)&addr, addrlen, O_NONBLOCK);
1517 if (err && err != -EINPROGRESS)
1520 /* discard the subflow socket */
1521 mptcp_sock_graft(ssk, sk->sk_socket);
1522 iput(SOCK_INODE(sf));
1523 WRITE_ONCE(msk->allow_infinite_fallback, false);
1527 list_del(&subflow->node);
1528 sock_put(mptcp_subflow_tcp_sock(subflow));
1531 subflow->disposable = 1;
1535 /* we account subflows before the creation, and this failures will not
1536 * be caught by sk_state_change()
1538 mptcp_pm_close_subflow(msk);
1542 static void mptcp_attach_cgroup(struct sock *parent, struct sock *child)
1544 #ifdef CONFIG_SOCK_CGROUP_DATA
1545 struct sock_cgroup_data *parent_skcd = &parent->sk_cgrp_data,
1546 *child_skcd = &child->sk_cgrp_data;
1548 /* only the additional subflows created by kworkers have to be modified */
1549 if (cgroup_id(sock_cgroup_ptr(parent_skcd)) !=
1550 cgroup_id(sock_cgroup_ptr(child_skcd))) {
1552 struct mem_cgroup *memcg = parent->sk_memcg;
1554 mem_cgroup_sk_free(child);
1555 if (memcg && css_tryget(&memcg->css))
1556 child->sk_memcg = memcg;
1557 #endif /* CONFIG_MEMCG */
1559 cgroup_sk_free(child_skcd);
1560 *child_skcd = *parent_skcd;
1561 cgroup_sk_clone(child_skcd);
1563 #endif /* CONFIG_SOCK_CGROUP_DATA */
1566 static void mptcp_subflow_ops_override(struct sock *ssk)
1568 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1569 if (ssk->sk_prot == &tcpv6_prot)
1570 ssk->sk_prot = &tcpv6_prot_override;
1573 ssk->sk_prot = &tcp_prot_override;
1576 static void mptcp_subflow_ops_undo_override(struct sock *ssk)
1578 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1579 if (ssk->sk_prot == &tcpv6_prot_override)
1580 ssk->sk_prot = &tcpv6_prot;
1583 ssk->sk_prot = &tcp_prot;
1585 int mptcp_subflow_create_socket(struct sock *sk, struct socket **new_sock)
1587 struct mptcp_subflow_context *subflow;
1588 struct net *net = sock_net(sk);
1592 /* un-accepted server sockets can reach here - on bad configuration
1593 * bail early to avoid greater trouble later
1595 if (unlikely(!sk->sk_socket))
1598 err = sock_create_kern(net, sk->sk_family, SOCK_STREAM, IPPROTO_TCP,
1605 /* the newly created socket has to be in the same cgroup as its parent */
1606 mptcp_attach_cgroup(sk, sf->sk);
1608 /* kernel sockets do not by default acquire net ref, but TCP timer
1611 sf->sk->sk_net_refcnt = 1;
1612 get_net_track(net, &sf->sk->ns_tracker, GFP_KERNEL);
1613 sock_inuse_add(net, 1);
1614 err = tcp_set_ulp(sf->sk, "mptcp");
1615 release_sock(sf->sk);
1622 /* the newly created socket really belongs to the owning MPTCP master
1623 * socket, even if for additional subflows the allocation is performed
1624 * by a kernel workqueue. Adjust inode references, so that the
1625 * procfs/diag interaces really show this one belonging to the correct
1628 SOCK_INODE(sf)->i_ino = SOCK_INODE(sk->sk_socket)->i_ino;
1629 SOCK_INODE(sf)->i_uid = SOCK_INODE(sk->sk_socket)->i_uid;
1630 SOCK_INODE(sf)->i_gid = SOCK_INODE(sk->sk_socket)->i_gid;
1632 subflow = mptcp_subflow_ctx(sf->sk);
1633 pr_debug("subflow=%p", subflow);
1638 mptcp_subflow_ops_override(sf->sk);
1643 static struct mptcp_subflow_context *subflow_create_ctx(struct sock *sk,
1646 struct inet_connection_sock *icsk = inet_csk(sk);
1647 struct mptcp_subflow_context *ctx;
1649 ctx = kzalloc(sizeof(*ctx), priority);
1653 rcu_assign_pointer(icsk->icsk_ulp_data, ctx);
1654 INIT_LIST_HEAD(&ctx->node);
1655 INIT_LIST_HEAD(&ctx->delegated_node);
1657 pr_debug("subflow=%p", ctx);
1664 static void __subflow_state_change(struct sock *sk)
1666 struct socket_wq *wq;
1669 wq = rcu_dereference(sk->sk_wq);
1670 if (skwq_has_sleeper(wq))
1671 wake_up_interruptible_all(&wq->wait);
1675 static bool subflow_is_done(const struct sock *sk)
1677 return sk->sk_shutdown & RCV_SHUTDOWN || sk->sk_state == TCP_CLOSE;
1680 static void subflow_state_change(struct sock *sk)
1682 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1683 struct sock *parent = subflow->conn;
1685 __subflow_state_change(sk);
1687 if (subflow_simultaneous_connect(sk)) {
1688 mptcp_propagate_sndbuf(parent, sk);
1689 mptcp_do_fallback(sk);
1690 mptcp_rcv_space_init(mptcp_sk(parent), sk);
1691 pr_fallback(mptcp_sk(parent));
1692 subflow->conn_finished = 1;
1693 mptcp_set_connected(parent);
1696 /* as recvmsg() does not acquire the subflow socket for ssk selection
1697 * a fin packet carrying a DSS can be unnoticed if we don't trigger
1698 * the data available machinery here.
1700 if (mptcp_subflow_data_available(sk))
1701 mptcp_data_ready(parent, sk);
1702 else if (unlikely(sk->sk_err))
1703 subflow_error_report(sk);
1705 subflow_sched_work_if_closed(mptcp_sk(parent), sk);
1707 if (__mptcp_check_fallback(mptcp_sk(parent)) &&
1708 !subflow->rx_eof && subflow_is_done(sk)) {
1709 subflow->rx_eof = 1;
1710 mptcp_subflow_eof(parent);
1714 static int subflow_ulp_init(struct sock *sk)
1716 struct inet_connection_sock *icsk = inet_csk(sk);
1717 struct mptcp_subflow_context *ctx;
1718 struct tcp_sock *tp = tcp_sk(sk);
1721 /* disallow attaching ULP to a socket unless it has been
1722 * created with sock_create_kern()
1724 if (!sk->sk_kern_sock) {
1729 ctx = subflow_create_ctx(sk, GFP_KERNEL);
1735 pr_debug("subflow=%p, family=%d", ctx, sk->sk_family);
1738 ctx->icsk_af_ops = icsk->icsk_af_ops;
1739 icsk->icsk_af_ops = subflow_default_af_ops(sk);
1740 ctx->tcp_state_change = sk->sk_state_change;
1741 ctx->tcp_error_report = sk->sk_error_report;
1743 WARN_ON_ONCE(sk->sk_data_ready != sock_def_readable);
1744 WARN_ON_ONCE(sk->sk_write_space != sk_stream_write_space);
1746 sk->sk_data_ready = subflow_data_ready;
1747 sk->sk_write_space = subflow_write_space;
1748 sk->sk_state_change = subflow_state_change;
1749 sk->sk_error_report = subflow_error_report;
1754 static void subflow_ulp_release(struct sock *ssk)
1756 struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(ssk);
1757 bool release = true;
1765 /* if the msk has been orphaned, keep the ctx
1766 * alive, will be freed by __mptcp_close_ssk(),
1767 * when the subflow is still unaccepted
1769 release = ctx->disposable || list_empty(&ctx->node);
1773 mptcp_subflow_ops_undo_override(ssk);
1775 kfree_rcu(ctx, rcu);
1778 static void subflow_ulp_clone(const struct request_sock *req,
1780 const gfp_t priority)
1782 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
1783 struct mptcp_subflow_context *old_ctx = mptcp_subflow_ctx(newsk);
1784 struct mptcp_subflow_context *new_ctx;
1786 if (!tcp_rsk(req)->is_mptcp ||
1787 (!subflow_req->mp_capable && !subflow_req->mp_join)) {
1788 subflow_ulp_fallback(newsk, old_ctx);
1792 new_ctx = subflow_create_ctx(newsk, priority);
1794 subflow_ulp_fallback(newsk, old_ctx);
1798 new_ctx->conn_finished = 1;
1799 new_ctx->icsk_af_ops = old_ctx->icsk_af_ops;
1800 new_ctx->tcp_state_change = old_ctx->tcp_state_change;
1801 new_ctx->tcp_error_report = old_ctx->tcp_error_report;
1802 new_ctx->rel_write_seq = 1;
1803 new_ctx->tcp_sock = newsk;
1805 if (subflow_req->mp_capable) {
1806 /* see comments in subflow_syn_recv_sock(), MPTCP connection
1807 * is fully established only after we receive the remote key
1809 new_ctx->mp_capable = 1;
1810 new_ctx->local_key = subflow_req->local_key;
1811 new_ctx->token = subflow_req->token;
1812 new_ctx->ssn_offset = subflow_req->ssn_offset;
1813 new_ctx->idsn = subflow_req->idsn;
1815 /* this is the first subflow, id is always 0 */
1816 new_ctx->local_id_valid = 1;
1817 } else if (subflow_req->mp_join) {
1818 new_ctx->ssn_offset = subflow_req->ssn_offset;
1819 new_ctx->mp_join = 1;
1820 new_ctx->fully_established = 1;
1821 new_ctx->backup = subflow_req->backup;
1822 new_ctx->remote_id = subflow_req->remote_id;
1823 new_ctx->token = subflow_req->token;
1824 new_ctx->thmac = subflow_req->thmac;
1826 /* the subflow req id is valid, fetched via subflow_check_req()
1827 * and subflow_token_join_request()
1829 subflow_set_local_id(new_ctx, subflow_req->local_id);
1833 static void tcp_release_cb_override(struct sock *ssk)
1835 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1837 if (mptcp_subflow_has_delegated_action(subflow))
1838 mptcp_subflow_process_delegated(ssk);
1840 tcp_release_cb(ssk);
1843 static struct tcp_ulp_ops subflow_ulp_ops __read_mostly = {
1845 .owner = THIS_MODULE,
1846 .init = subflow_ulp_init,
1847 .release = subflow_ulp_release,
1848 .clone = subflow_ulp_clone,
1851 static int subflow_ops_init(struct request_sock_ops *subflow_ops)
1853 subflow_ops->obj_size = sizeof(struct mptcp_subflow_request_sock);
1854 subflow_ops->slab_name = "request_sock_subflow";
1856 subflow_ops->slab = kmem_cache_create(subflow_ops->slab_name,
1857 subflow_ops->obj_size, 0,
1859 SLAB_TYPESAFE_BY_RCU,
1861 if (!subflow_ops->slab)
1864 subflow_ops->destructor = subflow_req_destructor;
1869 void __init mptcp_subflow_init(void)
1871 mptcp_subflow_request_sock_ops = tcp_request_sock_ops;
1872 if (subflow_ops_init(&mptcp_subflow_request_sock_ops) != 0)
1873 panic("MPTCP: failed to init subflow request sock ops\n");
1875 subflow_request_sock_ipv4_ops = tcp_request_sock_ipv4_ops;
1876 subflow_request_sock_ipv4_ops.route_req = subflow_v4_route_req;
1878 subflow_specific = ipv4_specific;
1879 subflow_specific.conn_request = subflow_v4_conn_request;
1880 subflow_specific.syn_recv_sock = subflow_syn_recv_sock;
1881 subflow_specific.sk_rx_dst_set = subflow_finish_connect;
1882 subflow_specific.rebuild_header = subflow_rebuild_header;
1884 tcp_prot_override = tcp_prot;
1885 tcp_prot_override.release_cb = tcp_release_cb_override;
1887 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1888 subflow_request_sock_ipv6_ops = tcp_request_sock_ipv6_ops;
1889 subflow_request_sock_ipv6_ops.route_req = subflow_v6_route_req;
1891 subflow_v6_specific = ipv6_specific;
1892 subflow_v6_specific.conn_request = subflow_v6_conn_request;
1893 subflow_v6_specific.syn_recv_sock = subflow_syn_recv_sock;
1894 subflow_v6_specific.sk_rx_dst_set = subflow_finish_connect;
1895 subflow_v6_specific.rebuild_header = subflow_v6_rebuild_header;
1897 subflow_v6m_specific = subflow_v6_specific;
1898 subflow_v6m_specific.queue_xmit = ipv4_specific.queue_xmit;
1899 subflow_v6m_specific.send_check = ipv4_specific.send_check;
1900 subflow_v6m_specific.net_header_len = ipv4_specific.net_header_len;
1901 subflow_v6m_specific.mtu_reduced = ipv4_specific.mtu_reduced;
1902 subflow_v6m_specific.net_frag_header_len = 0;
1903 subflow_v6m_specific.rebuild_header = subflow_rebuild_header;
1905 tcpv6_prot_override = tcpv6_prot;
1906 tcpv6_prot_override.release_cb = tcp_release_cb_override;
1909 mptcp_diag_subflow_init(&subflow_ulp_ops);
1911 if (tcp_register_ulp(&subflow_ulp_ops) != 0)
1912 panic("MPTCP: failed to register subflows to ULP\n");