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 READ_ONCE(msk->pm.accept_subflow);
68 /* validate received token and create truncated hmac and nonce for SYN-ACK */
69 static void subflow_req_create_thmac(struct mptcp_subflow_request_sock *subflow_req)
71 struct mptcp_sock *msk = subflow_req->msk;
72 u8 hmac[SHA256_DIGEST_SIZE];
74 get_random_bytes(&subflow_req->local_nonce, sizeof(u32));
76 subflow_generate_hmac(msk->local_key, msk->remote_key,
77 subflow_req->local_nonce,
78 subflow_req->remote_nonce, hmac);
80 subflow_req->thmac = get_unaligned_be64(hmac);
83 static struct mptcp_sock *subflow_token_join_request(struct request_sock *req)
85 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
86 struct mptcp_sock *msk;
89 msk = mptcp_token_get_sock(subflow_req->token);
91 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINNOTOKEN);
95 local_id = mptcp_pm_get_local_id(msk, (struct sock_common *)req);
97 sock_put((struct sock *)msk);
100 subflow_req->local_id = local_id;
105 static void subflow_init_req(struct request_sock *req, const struct sock *sk_listener)
107 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
109 subflow_req->mp_capable = 0;
110 subflow_req->mp_join = 0;
111 subflow_req->csum_reqd = mptcp_is_checksum_enabled(sock_net(sk_listener));
112 subflow_req->allow_join_id0 = mptcp_allow_join_id0(sock_net(sk_listener));
113 subflow_req->msk = NULL;
114 mptcp_token_init_request(req);
117 static bool subflow_use_different_sport(struct mptcp_sock *msk, const struct sock *sk)
119 return inet_sk(sk)->inet_sport != inet_sk((struct sock *)msk)->inet_sport;
122 static void subflow_add_reset_reason(struct sk_buff *skb, u8 reason)
124 struct mptcp_ext *mpext = skb_ext_add(skb, SKB_EXT_MPTCP);
127 memset(mpext, 0, sizeof(*mpext));
128 mpext->reset_reason = reason;
132 /* Init mptcp request socket.
134 * Returns an error code if a JOIN has failed and a TCP reset
137 static int subflow_check_req(struct request_sock *req,
138 const struct sock *sk_listener,
141 struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk_listener);
142 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
143 struct mptcp_options_received mp_opt;
145 pr_debug("subflow_req=%p, listener=%p", subflow_req, listener);
147 #ifdef CONFIG_TCP_MD5SIG
148 /* no MPTCP if MD5SIG is enabled on this socket or we may run out of
151 if (rcu_access_pointer(tcp_sk(sk_listener)->md5sig_info))
155 mptcp_get_options(sk_listener, skb, &mp_opt);
157 if (mp_opt.mp_capable) {
158 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MPCAPABLEPASSIVE);
162 } else if (mp_opt.mp_join) {
163 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINSYNRX);
166 if (mp_opt.mp_capable && listener->request_mptcp) {
167 int err, retries = MPTCP_TOKEN_MAX_RETRIES;
169 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq;
172 get_random_bytes(&subflow_req->local_key, sizeof(subflow_req->local_key));
173 } while (subflow_req->local_key == 0);
175 if (unlikely(req->syncookie)) {
176 mptcp_crypto_key_sha(subflow_req->local_key,
179 if (mptcp_token_exists(subflow_req->token)) {
182 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_TOKENFALLBACKINIT);
184 subflow_req->mp_capable = 1;
189 err = mptcp_token_new_request(req);
191 subflow_req->mp_capable = 1;
192 else if (retries-- > 0)
195 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_TOKENFALLBACKINIT);
197 } else if (mp_opt.mp_join && listener->request_mptcp) {
198 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq;
199 subflow_req->mp_join = 1;
200 subflow_req->backup = mp_opt.backup;
201 subflow_req->remote_id = mp_opt.join_id;
202 subflow_req->token = mp_opt.token;
203 subflow_req->remote_nonce = mp_opt.nonce;
204 subflow_req->msk = subflow_token_join_request(req);
206 /* Can't fall back to TCP in this case. */
207 if (!subflow_req->msk) {
208 subflow_add_reset_reason(skb, MPTCP_RST_EMPTCP);
212 if (subflow_use_different_sport(subflow_req->msk, sk_listener)) {
213 pr_debug("syn inet_sport=%d %d",
214 ntohs(inet_sk(sk_listener)->inet_sport),
215 ntohs(inet_sk((struct sock *)subflow_req->msk)->inet_sport));
216 if (!mptcp_pm_sport_in_anno_list(subflow_req->msk, sk_listener)) {
217 sock_put((struct sock *)subflow_req->msk);
218 mptcp_token_destroy_request(req);
219 tcp_request_sock_ops.destructor(req);
220 subflow_req->msk = NULL;
221 subflow_req->mp_join = 0;
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);
235 pr_debug("token=%u, remote_nonce=%u msk=%p", subflow_req->token,
236 subflow_req->remote_nonce, subflow_req->msk);
242 int mptcp_subflow_init_cookie_req(struct request_sock *req,
243 const struct sock *sk_listener,
246 struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk_listener);
247 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
248 struct mptcp_options_received mp_opt;
251 subflow_init_req(req, sk_listener);
252 mptcp_get_options(sk_listener, skb, &mp_opt);
254 if (mp_opt.mp_capable && mp_opt.mp_join)
257 if (mp_opt.mp_capable && listener->request_mptcp) {
258 if (mp_opt.sndr_key == 0)
261 subflow_req->local_key = mp_opt.rcvr_key;
262 err = mptcp_token_new_request(req);
266 subflow_req->mp_capable = 1;
267 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq - 1;
268 } else if (mp_opt.mp_join && listener->request_mptcp) {
269 if (!mptcp_token_join_cookie_init_state(subflow_req, skb))
272 if (mptcp_can_accept_new_subflow(subflow_req->msk))
273 subflow_req->mp_join = 1;
275 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq - 1;
280 EXPORT_SYMBOL_GPL(mptcp_subflow_init_cookie_req);
282 static struct dst_entry *subflow_v4_route_req(const struct sock *sk,
285 struct request_sock *req)
287 struct dst_entry *dst;
290 tcp_rsk(req)->is_mptcp = 1;
291 subflow_init_req(req, sk);
293 dst = tcp_request_sock_ipv4_ops.route_req(sk, skb, fl, req);
297 err = subflow_check_req(req, sk, skb);
303 tcp_request_sock_ops.send_reset(sk, skb);
307 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
308 static struct dst_entry *subflow_v6_route_req(const struct sock *sk,
311 struct request_sock *req)
313 struct dst_entry *dst;
316 tcp_rsk(req)->is_mptcp = 1;
317 subflow_init_req(req, sk);
319 dst = tcp_request_sock_ipv6_ops.route_req(sk, skb, fl, req);
323 err = subflow_check_req(req, sk, skb);
329 tcp6_request_sock_ops.send_reset(sk, skb);
334 /* validate received truncated hmac and create hmac for third ACK */
335 static bool subflow_thmac_valid(struct mptcp_subflow_context *subflow)
337 u8 hmac[SHA256_DIGEST_SIZE];
340 subflow_generate_hmac(subflow->remote_key, subflow->local_key,
341 subflow->remote_nonce, subflow->local_nonce,
344 thmac = get_unaligned_be64(hmac);
345 pr_debug("subflow=%p, token=%u, thmac=%llu, subflow->thmac=%llu\n",
346 subflow, subflow->token,
347 (unsigned long long)thmac,
348 (unsigned long long)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 /* must hold: tcp_done() could drop last reference on parent */
361 tcp_set_state(ssk, TCP_CLOSE);
362 tcp_send_active_reset(ssk, GFP_ATOMIC);
364 if (!test_and_set_bit(MPTCP_WORK_CLOSE_SUBFLOW, &mptcp_sk(sk)->flags) &&
365 schedule_work(&mptcp_sk(sk)->work))
366 return; /* worker will put sk for us */
371 static bool subflow_use_different_dport(struct mptcp_sock *msk, const struct sock *sk)
373 return inet_sk(sk)->inet_dport != inet_sk((struct sock *)msk)->inet_dport;
376 void __mptcp_set_connected(struct sock *sk)
378 if (sk->sk_state == TCP_SYN_SENT) {
379 inet_sk_state_store(sk, TCP_ESTABLISHED);
380 sk->sk_state_change(sk);
384 static void mptcp_set_connected(struct sock *sk)
387 if (!sock_owned_by_user(sk))
388 __mptcp_set_connected(sk);
390 set_bit(MPTCP_CONNECTED, &mptcp_sk(sk)->flags);
391 mptcp_data_unlock(sk);
394 static void subflow_finish_connect(struct sock *sk, const struct sk_buff *skb)
396 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
397 struct mptcp_options_received mp_opt;
398 struct sock *parent = subflow->conn;
400 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(sk, skb, &mp_opt);
414 if (subflow->request_mptcp) {
415 if (!mp_opt.mp_capable) {
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.csum_reqd)
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.mp_join) {
439 subflow->reset_reason = MPTCP_RST_EMPTCP;
443 subflow->thmac = mp_opt.thmac;
444 subflow->remote_nonce = mp_opt.nonce;
445 pr_debug("subflow=%p, thmac=%llu, remote_nonce=%u", subflow,
446 subflow->thmac, subflow->remote_nonce);
448 if (!subflow_thmac_valid(subflow)) {
449 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINACKMAC);
450 subflow->reset_reason = MPTCP_RST_EMPTCP;
454 if (!mptcp_finish_join(sk))
457 subflow_generate_hmac(subflow->local_key, subflow->remote_key,
458 subflow->local_nonce,
459 subflow->remote_nonce,
461 memcpy(subflow->hmac, hmac, MPTCPOPT_HMAC_LEN);
463 subflow->mp_join = 1;
464 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINSYNACKRX);
466 if (subflow_use_different_dport(mptcp_sk(parent), sk)) {
467 pr_debug("synack inet_dport=%d %d",
468 ntohs(inet_sk(sk)->inet_dport),
469 ntohs(inet_sk(parent)->inet_dport));
470 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINPORTSYNACKRX);
472 } else if (mptcp_check_fallback(sk)) {
474 mptcp_rcv_space_init(mptcp_sk(parent), sk);
475 mptcp_set_connected(parent);
480 subflow->reset_transient = 0;
481 mptcp_subflow_reset(sk);
484 struct request_sock_ops mptcp_subflow_request_sock_ops;
485 EXPORT_SYMBOL_GPL(mptcp_subflow_request_sock_ops);
486 static struct tcp_request_sock_ops subflow_request_sock_ipv4_ops;
488 static int subflow_v4_conn_request(struct sock *sk, struct sk_buff *skb)
490 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
492 pr_debug("subflow=%p", subflow);
494 /* Never answer to SYNs sent to broadcast or multicast */
495 if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
498 return tcp_conn_request(&mptcp_subflow_request_sock_ops,
499 &subflow_request_sock_ipv4_ops,
506 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
507 static struct tcp_request_sock_ops subflow_request_sock_ipv6_ops;
508 static struct inet_connection_sock_af_ops subflow_v6_specific;
509 static struct inet_connection_sock_af_ops subflow_v6m_specific;
510 static struct proto tcpv6_prot_override;
512 static int subflow_v6_conn_request(struct sock *sk, struct sk_buff *skb)
514 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
516 pr_debug("subflow=%p", subflow);
518 if (skb->protocol == htons(ETH_P_IP))
519 return subflow_v4_conn_request(sk, skb);
521 if (!ipv6_unicast_destination(skb))
524 if (ipv6_addr_v4mapped(&ipv6_hdr(skb)->saddr)) {
525 __IP6_INC_STATS(sock_net(sk), NULL, IPSTATS_MIB_INHDRERRORS);
529 return tcp_conn_request(&mptcp_subflow_request_sock_ops,
530 &subflow_request_sock_ipv6_ops, sk, skb);
534 return 0; /* don't send reset */
538 /* validate hmac received in third ACK */
539 static bool subflow_hmac_valid(const struct request_sock *req,
540 const struct mptcp_options_received *mp_opt)
542 const struct mptcp_subflow_request_sock *subflow_req;
543 u8 hmac[SHA256_DIGEST_SIZE];
544 struct mptcp_sock *msk;
546 subflow_req = mptcp_subflow_rsk(req);
547 msk = subflow_req->msk;
551 subflow_generate_hmac(msk->remote_key, msk->local_key,
552 subflow_req->remote_nonce,
553 subflow_req->local_nonce, hmac);
555 return !crypto_memneq(hmac, mp_opt->hmac, MPTCPOPT_HMAC_LEN);
558 static void mptcp_sock_destruct(struct sock *sk)
560 /* if new mptcp socket isn't accepted, it is free'd
561 * from the tcp listener sockets request queue, linked
562 * from req->sk. The tcp socket is released.
563 * This calls the ULP release function which will
564 * also remove the mptcp socket, via
565 * sock_put(ctx->conn).
567 * Problem is that the mptcp socket will be in
568 * ESTABLISHED state and will not have the SOCK_DEAD flag.
569 * Both result in warnings from inet_sock_destruct.
571 if ((1 << sk->sk_state) & (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) {
572 sk->sk_state = TCP_CLOSE;
573 WARN_ON_ONCE(sk->sk_socket);
577 mptcp_destroy_common(mptcp_sk(sk));
578 inet_sock_destruct(sk);
581 static void mptcp_force_close(struct sock *sk)
583 /* the msk is not yet exposed to user-space */
584 inet_sk_state_store(sk, TCP_CLOSE);
585 sk_common_release(sk);
588 static void subflow_ulp_fallback(struct sock *sk,
589 struct mptcp_subflow_context *old_ctx)
591 struct inet_connection_sock *icsk = inet_csk(sk);
593 mptcp_subflow_tcp_fallback(sk, old_ctx);
594 icsk->icsk_ulp_ops = NULL;
595 rcu_assign_pointer(icsk->icsk_ulp_data, NULL);
596 tcp_sk(sk)->is_mptcp = 0;
598 mptcp_subflow_ops_undo_override(sk);
601 static void subflow_drop_ctx(struct sock *ssk)
603 struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(ssk);
608 subflow_ulp_fallback(ssk, ctx);
615 void mptcp_subflow_fully_established(struct mptcp_subflow_context *subflow,
616 struct mptcp_options_received *mp_opt)
618 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
620 subflow->remote_key = mp_opt->sndr_key;
621 subflow->fully_established = 1;
622 subflow->can_ack = 1;
623 WRITE_ONCE(msk->fully_established, true);
626 static struct sock *subflow_syn_recv_sock(const struct sock *sk,
628 struct request_sock *req,
629 struct dst_entry *dst,
630 struct request_sock *req_unhash,
633 struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk);
634 struct mptcp_subflow_request_sock *subflow_req;
635 struct mptcp_options_received mp_opt;
636 bool fallback, fallback_is_fatal;
637 struct sock *new_msk = NULL;
640 pr_debug("listener=%p, req=%p, conn=%p", listener, req, listener->conn);
642 /* After child creation we must look for 'mp_capable' even when options
645 mp_opt.mp_capable = 0;
647 /* hopefully temporary handling for MP_JOIN+syncookie */
648 subflow_req = mptcp_subflow_rsk(req);
649 fallback_is_fatal = tcp_rsk(req)->is_mptcp && subflow_req->mp_join;
650 fallback = !tcp_rsk(req)->is_mptcp;
654 /* if the sk is MP_CAPABLE, we try to fetch the client key */
655 if (subflow_req->mp_capable) {
656 /* we can receive and accept an in-window, out-of-order pkt,
657 * which may not carry the MP_CAPABLE opt even on mptcp enabled
658 * paths: always try to extract the peer key, and fallback
659 * for packets missing it.
660 * Even OoO DSS packets coming legitly after dropped or
661 * reordered MPC will cause fallback, but we don't have other
664 mptcp_get_options(sk, skb, &mp_opt);
665 if (!mp_opt.mp_capable) {
670 new_msk = mptcp_sk_clone(listener->conn, &mp_opt, req);
673 } else if (subflow_req->mp_join) {
674 mptcp_get_options(sk, skb, &mp_opt);
675 if (!mp_opt.mp_join || !subflow_hmac_valid(req, &mp_opt) ||
676 !mptcp_can_accept_new_subflow(subflow_req->msk)) {
677 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKMAC);
683 child = listener->icsk_af_ops->syn_recv_sock(sk, skb, req, dst,
684 req_unhash, own_req);
686 if (child && *own_req) {
687 struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(child);
689 tcp_rsk(req)->drop_req = false;
691 /* we need to fallback on ctx allocation failure and on pre-reqs
692 * checking above. In the latter scenario we additionally need
693 * to reset the context to non MPTCP status.
695 if (!ctx || fallback) {
696 if (fallback_is_fatal) {
697 subflow_add_reset_reason(skb, MPTCP_RST_EMPTCP);
701 subflow_drop_ctx(child);
705 /* ssk inherits options of listener sk */
706 ctx->setsockopt_seq = listener->setsockopt_seq;
708 if (ctx->mp_capable) {
709 /* this can't race with mptcp_close(), as the msk is
710 * not yet exposted to user-space
712 inet_sk_state_store((void *)new_msk, TCP_ESTABLISHED);
714 /* record the newly created socket as the first msk
715 * subflow, but don't link it yet into conn_list
717 WRITE_ONCE(mptcp_sk(new_msk)->first, child);
719 /* new mpc subflow takes ownership of the newly
720 * created mptcp socket
722 new_msk->sk_destruct = mptcp_sock_destruct;
723 mptcp_sk(new_msk)->setsockopt_seq = ctx->setsockopt_seq;
724 mptcp_pm_new_connection(mptcp_sk(new_msk), child, 1);
725 mptcp_token_accept(subflow_req, mptcp_sk(new_msk));
729 /* with OoO packets we can reach here without ingress
732 if (mp_opt.mp_capable)
733 mptcp_subflow_fully_established(ctx, &mp_opt);
734 } else if (ctx->mp_join) {
735 struct mptcp_sock *owner;
737 owner = subflow_req->msk;
739 subflow_add_reset_reason(skb, MPTCP_RST_EPROHIBIT);
743 /* move the msk reference ownership to the subflow */
744 subflow_req->msk = NULL;
745 ctx->conn = (struct sock *)owner;
747 if (subflow_use_different_sport(owner, sk)) {
748 pr_debug("ack inet_sport=%d %d",
749 ntohs(inet_sk(sk)->inet_sport),
750 ntohs(inet_sk((struct sock *)owner)->inet_sport));
751 if (!mptcp_pm_sport_in_anno_list(owner, sk)) {
752 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MISMATCHPORTACKRX);
755 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINPORTACKRX);
758 if (!mptcp_finish_join(child))
761 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKRX);
762 tcp_rsk(req)->drop_req = true;
767 /* dispose of the left over mptcp master, if any */
768 if (unlikely(new_msk))
769 mptcp_force_close(new_msk);
771 /* check for expected invariant - should never trigger, just help
772 * catching eariler subtle bugs
774 WARN_ON_ONCE(child && *own_req && tcp_sk(child)->is_mptcp &&
775 (!mptcp_subflow_ctx(child) ||
776 !mptcp_subflow_ctx(child)->conn));
780 subflow_drop_ctx(child);
781 tcp_rsk(req)->drop_req = true;
782 inet_csk_prepare_for_destroy_sock(child);
784 req->rsk_ops->send_reset(sk, skb);
786 /* The last child reference will be released by the caller */
790 static struct inet_connection_sock_af_ops subflow_specific;
791 static struct proto tcp_prot_override;
793 enum mapping_status {
801 static void dbg_bad_map(struct mptcp_subflow_context *subflow, u32 ssn)
803 pr_debug("Bad mapping: ssn=%d map_seq=%d map_data_len=%d",
804 ssn, subflow->map_subflow_seq, subflow->map_data_len);
807 static bool skb_is_fully_mapped(struct sock *ssk, struct sk_buff *skb)
809 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
810 unsigned int skb_consumed;
812 skb_consumed = tcp_sk(ssk)->copied_seq - TCP_SKB_CB(skb)->seq;
813 if (WARN_ON_ONCE(skb_consumed >= skb->len))
816 return skb->len - skb_consumed <= subflow->map_data_len -
817 mptcp_subflow_get_map_offset(subflow);
820 static bool validate_mapping(struct sock *ssk, struct sk_buff *skb)
822 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
823 u32 ssn = tcp_sk(ssk)->copied_seq - subflow->ssn_offset;
825 if (unlikely(before(ssn, subflow->map_subflow_seq))) {
826 /* Mapping covers data later in the subflow stream,
827 * currently unsupported.
829 dbg_bad_map(subflow, ssn);
832 if (unlikely(!before(ssn, subflow->map_subflow_seq +
833 subflow->map_data_len))) {
834 /* Mapping does covers past subflow data, invalid */
835 dbg_bad_map(subflow, ssn);
841 static enum mapping_status validate_data_csum(struct sock *ssk, struct sk_buff *skb,
844 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
845 struct csum_pseudo_header header;
846 u32 offset, seq, delta;
853 /* mapping already validated on previous traversal */
854 if (subflow->map_csum_len == subflow->map_data_len)
857 /* traverse the receive queue, ensuring it contains a full
858 * DSS mapping and accumulating the related csum.
859 * Preserve the accoumlate csum across multiple calls, to compute
862 delta = subflow->map_data_len - subflow->map_csum_len;
864 seq = tcp_sk(ssk)->copied_seq + subflow->map_csum_len;
865 offset = seq - TCP_SKB_CB(skb)->seq;
867 /* if the current skb has not been accounted yet, csum its contents
868 * up to the amount covered by the current DSS
870 if (offset < skb->len) {
873 len = min(skb->len - offset, delta);
874 csum = skb_checksum(skb, offset, len, 0);
875 subflow->map_data_csum = csum_block_add(subflow->map_data_csum, csum,
876 subflow->map_csum_len);
879 subflow->map_csum_len += len;
884 if (skb_queue_is_last(&ssk->sk_receive_queue, skb)) {
885 /* if this subflow is closed, the partial mapping
886 * will be never completed; flush the pending skbs, so
887 * that subflow_sched_work_if_closed() can kick in
889 if (unlikely(ssk->sk_state == TCP_CLOSE))
890 while ((skb = skb_peek(&ssk->sk_receive_queue)))
891 sk_eat_skb(ssk, skb);
893 /* not enough data to validate the csum */
894 return MAPPING_EMPTY;
897 /* the DSS mapping for next skbs will be validated later,
898 * when a get_mapping_status call will process such skb
903 /* note that 'map_data_len' accounts only for the carried data, does
904 * not include the eventual seq increment due to the data fin,
905 * while the pseudo header requires the original DSS data len,
908 header.data_seq = cpu_to_be64(subflow->map_seq);
909 header.subflow_seq = htonl(subflow->map_subflow_seq);
910 header.data_len = htons(subflow->map_data_len + subflow->map_data_fin);
913 csum = csum_partial(&header, sizeof(header), subflow->map_data_csum);
914 if (unlikely(csum_fold(csum))) {
915 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DATACSUMERR);
916 return subflow->mp_join ? MAPPING_INVALID : MAPPING_DUMMY;
922 static enum mapping_status get_mapping_status(struct sock *ssk,
923 struct mptcp_sock *msk)
925 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
926 bool csum_reqd = READ_ONCE(msk->csum_enabled);
927 struct mptcp_ext *mpext;
932 skb = skb_peek(&ssk->sk_receive_queue);
934 return MAPPING_EMPTY;
936 if (mptcp_check_fallback(ssk))
937 return MAPPING_DUMMY;
939 mpext = mptcp_get_ext(skb);
940 if (!mpext || !mpext->use_map) {
941 if (!subflow->map_valid && !skb->len) {
942 /* the TCP stack deliver 0 len FIN pkt to the receive
943 * queue, that is the only 0len pkts ever expected here,
944 * and we can admit no mapping only for 0 len pkts
946 if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN))
947 WARN_ONCE(1, "0len seq %d:%d flags %x",
948 TCP_SKB_CB(skb)->seq,
949 TCP_SKB_CB(skb)->end_seq,
950 TCP_SKB_CB(skb)->tcp_flags);
951 sk_eat_skb(ssk, skb);
952 return MAPPING_EMPTY;
955 if (!subflow->map_valid)
956 return MAPPING_INVALID;
961 trace_get_mapping_status(mpext);
963 data_len = mpext->data_len;
965 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_INFINITEMAPRX);
966 return MAPPING_INVALID;
969 if (mpext->data_fin == 1) {
971 bool updated = mptcp_update_rcv_data_fin(msk, mpext->data_seq,
973 pr_debug("DATA_FIN with no payload seq=%llu", mpext->data_seq);
974 if (subflow->map_valid) {
975 /* A DATA_FIN might arrive in a DSS
976 * option before the previous mapping
977 * has been fully consumed. Continue
978 * handling the existing mapping.
980 skb_ext_del(skb, SKB_EXT_MPTCP);
983 if (updated && schedule_work(&msk->work))
984 sock_hold((struct sock *)msk);
986 return MAPPING_DATA_FIN;
989 u64 data_fin_seq = mpext->data_seq + data_len - 1;
991 /* If mpext->data_seq is a 32-bit value, data_fin_seq
992 * must also be limited to 32 bits.
995 data_fin_seq &= GENMASK_ULL(31, 0);
997 mptcp_update_rcv_data_fin(msk, data_fin_seq, mpext->dsn64);
998 pr_debug("DATA_FIN with mapping seq=%llu dsn64=%d",
999 data_fin_seq, mpext->dsn64);
1002 /* Adjust for DATA_FIN using 1 byte of sequence space */
1006 map_seq = mptcp_expand_seq(READ_ONCE(msk->ack_seq), mpext->data_seq, mpext->dsn64);
1007 WRITE_ONCE(mptcp_sk(subflow->conn)->use_64bit_ack, !!mpext->dsn64);
1009 if (subflow->map_valid) {
1010 /* Allow replacing only with an identical map */
1011 if (subflow->map_seq == map_seq &&
1012 subflow->map_subflow_seq == mpext->subflow_seq &&
1013 subflow->map_data_len == data_len &&
1014 subflow->map_csum_reqd == mpext->csum_reqd) {
1015 skb_ext_del(skb, SKB_EXT_MPTCP);
1019 /* If this skb data are fully covered by the current mapping,
1020 * the new map would need caching, which is not supported
1022 if (skb_is_fully_mapped(ssk, skb)) {
1023 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DSSNOMATCH);
1024 return MAPPING_INVALID;
1027 /* will validate the next map after consuming the current one */
1031 subflow->map_seq = map_seq;
1032 subflow->map_subflow_seq = mpext->subflow_seq;
1033 subflow->map_data_len = data_len;
1034 subflow->map_valid = 1;
1035 subflow->map_data_fin = mpext->data_fin;
1036 subflow->mpc_map = mpext->mpc_map;
1037 subflow->map_csum_reqd = mpext->csum_reqd;
1038 subflow->map_csum_len = 0;
1039 subflow->map_data_csum = csum_unfold(mpext->csum);
1041 /* Cfr RFC 8684 Section 3.3.0 */
1042 if (unlikely(subflow->map_csum_reqd != csum_reqd))
1043 return MAPPING_INVALID;
1045 pr_debug("new map seq=%llu subflow_seq=%u data_len=%u csum=%d:%u",
1046 subflow->map_seq, subflow->map_subflow_seq,
1047 subflow->map_data_len, subflow->map_csum_reqd,
1048 subflow->map_data_csum);
1051 /* we revalidate valid mapping on new skb, because we must ensure
1052 * the current skb is completely covered by the available mapping
1054 if (!validate_mapping(ssk, skb)) {
1055 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DSSTCPMISMATCH);
1056 return MAPPING_INVALID;
1059 skb_ext_del(skb, SKB_EXT_MPTCP);
1062 return validate_data_csum(ssk, skb, csum_reqd);
1065 static void mptcp_subflow_discard_data(struct sock *ssk, struct sk_buff *skb,
1068 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1069 bool fin = TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN;
1072 incr = limit >= skb->len ? skb->len + fin : limit;
1074 pr_debug("discarding=%d len=%d seq=%d", incr, skb->len,
1075 subflow->map_subflow_seq);
1076 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DUPDATA);
1077 tcp_sk(ssk)->copied_seq += incr;
1078 if (!before(tcp_sk(ssk)->copied_seq, TCP_SKB_CB(skb)->end_seq))
1079 sk_eat_skb(ssk, skb);
1080 if (mptcp_subflow_get_map_offset(subflow) >= subflow->map_data_len)
1081 subflow->map_valid = 0;
1084 /* sched mptcp worker to remove the subflow if no more data is pending */
1085 static void subflow_sched_work_if_closed(struct mptcp_sock *msk, struct sock *ssk)
1087 struct sock *sk = (struct sock *)msk;
1089 if (likely(ssk->sk_state != TCP_CLOSE))
1092 if (skb_queue_empty(&ssk->sk_receive_queue) &&
1093 !test_and_set_bit(MPTCP_WORK_CLOSE_SUBFLOW, &msk->flags)) {
1095 if (!schedule_work(&msk->work))
1100 static bool subflow_check_data_avail(struct sock *ssk)
1102 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1103 enum mapping_status status;
1104 struct mptcp_sock *msk;
1105 struct sk_buff *skb;
1107 if (!skb_peek(&ssk->sk_receive_queue))
1108 WRITE_ONCE(subflow->data_avail, 0);
1109 if (subflow->data_avail)
1112 msk = mptcp_sk(subflow->conn);
1117 status = get_mapping_status(ssk, msk);
1118 trace_subflow_check_data_avail(status, skb_peek(&ssk->sk_receive_queue));
1119 if (unlikely(status == MAPPING_INVALID))
1122 if (unlikely(status == MAPPING_DUMMY))
1125 if (status != MAPPING_OK)
1128 skb = skb_peek(&ssk->sk_receive_queue);
1129 if (WARN_ON_ONCE(!skb))
1132 /* if msk lacks the remote key, this subflow must provide an
1133 * MP_CAPABLE-based mapping
1135 if (unlikely(!READ_ONCE(msk->can_ack))) {
1136 if (!subflow->mpc_map)
1138 WRITE_ONCE(msk->remote_key, subflow->remote_key);
1139 WRITE_ONCE(msk->ack_seq, subflow->map_seq);
1140 WRITE_ONCE(msk->can_ack, true);
1143 old_ack = READ_ONCE(msk->ack_seq);
1144 ack_seq = mptcp_subflow_get_mapped_dsn(subflow);
1145 pr_debug("msk ack_seq=%llx subflow ack_seq=%llx", old_ack,
1147 if (unlikely(before64(ack_seq, old_ack))) {
1148 mptcp_subflow_discard_data(ssk, skb, old_ack - ack_seq);
1152 WRITE_ONCE(subflow->data_avail, MPTCP_SUBFLOW_DATA_AVAIL);
1158 subflow_sched_work_if_closed(msk, ssk);
1162 /* RFC 8684 section 3.7. */
1163 if (subflow->mp_join || subflow->fully_established) {
1164 /* fatal protocol error, close the socket.
1165 * subflow_error_report() will introduce the appropriate barriers
1167 ssk->sk_err = EBADMSG;
1168 tcp_set_state(ssk, TCP_CLOSE);
1169 subflow->reset_transient = 0;
1170 subflow->reset_reason = MPTCP_RST_EMPTCP;
1171 tcp_send_active_reset(ssk, GFP_ATOMIC);
1172 WRITE_ONCE(subflow->data_avail, 0);
1176 __mptcp_do_fallback(msk);
1177 skb = skb_peek(&ssk->sk_receive_queue);
1178 subflow->map_valid = 1;
1179 subflow->map_seq = READ_ONCE(msk->ack_seq);
1180 subflow->map_data_len = skb->len;
1181 subflow->map_subflow_seq = tcp_sk(ssk)->copied_seq - subflow->ssn_offset;
1182 WRITE_ONCE(subflow->data_avail, MPTCP_SUBFLOW_DATA_AVAIL);
1186 bool mptcp_subflow_data_available(struct sock *sk)
1188 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1190 /* check if current mapping is still valid */
1191 if (subflow->map_valid &&
1192 mptcp_subflow_get_map_offset(subflow) >= subflow->map_data_len) {
1193 subflow->map_valid = 0;
1194 WRITE_ONCE(subflow->data_avail, 0);
1196 pr_debug("Done with mapping: seq=%u data_len=%u",
1197 subflow->map_subflow_seq,
1198 subflow->map_data_len);
1201 return subflow_check_data_avail(sk);
1204 /* If ssk has an mptcp parent socket, use the mptcp rcvbuf occupancy,
1207 * In mptcp, rwin is about the mptcp-level connection data.
1209 * Data that is still on the ssk rx queue can thus be ignored,
1210 * as far as mptcp peer is concerned that data is still inflight.
1211 * DSS ACK is updated when skb is moved to the mptcp rx queue.
1213 void mptcp_space(const struct sock *ssk, int *space, int *full_space)
1215 const struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1216 const struct sock *sk = subflow->conn;
1218 *space = __mptcp_space(sk);
1219 *full_space = tcp_full_space(sk);
1222 void __mptcp_error_report(struct sock *sk)
1224 struct mptcp_subflow_context *subflow;
1225 struct mptcp_sock *msk = mptcp_sk(sk);
1227 mptcp_for_each_subflow(msk, subflow) {
1228 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
1229 int err = sock_error(ssk);
1234 /* only propagate errors on fallen-back sockets or
1237 if (sk->sk_state != TCP_SYN_SENT && !__mptcp_check_fallback(msk))
1240 inet_sk_state_store(sk, inet_sk_state_load(ssk));
1243 /* This barrier is coupled with smp_rmb() in mptcp_poll() */
1245 sk_error_report(sk);
1250 static void subflow_error_report(struct sock *ssk)
1252 struct sock *sk = mptcp_subflow_ctx(ssk)->conn;
1254 mptcp_data_lock(sk);
1255 if (!sock_owned_by_user(sk))
1256 __mptcp_error_report(sk);
1258 set_bit(MPTCP_ERROR_REPORT, &mptcp_sk(sk)->flags);
1259 mptcp_data_unlock(sk);
1262 static void subflow_data_ready(struct sock *sk)
1264 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1265 u16 state = 1 << inet_sk_state_load(sk);
1266 struct sock *parent = subflow->conn;
1267 struct mptcp_sock *msk;
1269 msk = mptcp_sk(parent);
1270 if (state & TCPF_LISTEN) {
1271 /* MPJ subflow are removed from accept queue before reaching here,
1272 * avoid stray wakeups
1274 if (reqsk_queue_empty(&inet_csk(sk)->icsk_accept_queue))
1277 set_bit(MPTCP_DATA_READY, &msk->flags);
1278 parent->sk_data_ready(parent);
1282 WARN_ON_ONCE(!__mptcp_check_fallback(msk) && !subflow->mp_capable &&
1283 !subflow->mp_join && !(state & TCPF_CLOSE));
1285 if (mptcp_subflow_data_available(sk))
1286 mptcp_data_ready(parent, sk);
1287 else if (unlikely(sk->sk_err))
1288 subflow_error_report(sk);
1291 static void subflow_write_space(struct sock *ssk)
1293 struct sock *sk = mptcp_subflow_ctx(ssk)->conn;
1295 mptcp_propagate_sndbuf(sk, ssk);
1296 mptcp_write_space(sk);
1299 static struct inet_connection_sock_af_ops *
1300 subflow_default_af_ops(struct sock *sk)
1302 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1303 if (sk->sk_family == AF_INET6)
1304 return &subflow_v6_specific;
1306 return &subflow_specific;
1309 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1310 void mptcpv6_handle_mapped(struct sock *sk, bool mapped)
1312 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1313 struct inet_connection_sock *icsk = inet_csk(sk);
1314 struct inet_connection_sock_af_ops *target;
1316 target = mapped ? &subflow_v6m_specific : subflow_default_af_ops(sk);
1318 pr_debug("subflow=%p family=%d ops=%p target=%p mapped=%d",
1319 subflow, sk->sk_family, icsk->icsk_af_ops, target, mapped);
1321 if (likely(icsk->icsk_af_ops == target))
1324 subflow->icsk_af_ops = icsk->icsk_af_ops;
1325 icsk->icsk_af_ops = target;
1329 void mptcp_info2sockaddr(const struct mptcp_addr_info *info,
1330 struct sockaddr_storage *addr,
1331 unsigned short family)
1333 memset(addr, 0, sizeof(*addr));
1334 addr->ss_family = family;
1335 if (addr->ss_family == AF_INET) {
1336 struct sockaddr_in *in_addr = (struct sockaddr_in *)addr;
1338 if (info->family == AF_INET)
1339 in_addr->sin_addr = info->addr;
1340 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1341 else if (ipv6_addr_v4mapped(&info->addr6))
1342 in_addr->sin_addr.s_addr = info->addr6.s6_addr32[3];
1344 in_addr->sin_port = info->port;
1346 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1347 else if (addr->ss_family == AF_INET6) {
1348 struct sockaddr_in6 *in6_addr = (struct sockaddr_in6 *)addr;
1350 if (info->family == AF_INET)
1351 ipv6_addr_set_v4mapped(info->addr.s_addr,
1352 &in6_addr->sin6_addr);
1354 in6_addr->sin6_addr = info->addr6;
1355 in6_addr->sin6_port = info->port;
1360 int __mptcp_subflow_connect(struct sock *sk, const struct mptcp_addr_info *loc,
1361 const struct mptcp_addr_info *remote,
1362 u8 flags, int ifindex)
1364 struct mptcp_sock *msk = mptcp_sk(sk);
1365 struct mptcp_subflow_context *subflow;
1366 struct sockaddr_storage addr;
1367 int remote_id = remote->id;
1368 int local_id = loc->id;
1375 if (!mptcp_is_fully_established(sk))
1378 err = mptcp_subflow_create_socket(sk, &sf);
1383 subflow = mptcp_subflow_ctx(ssk);
1385 get_random_bytes(&subflow->local_nonce, sizeof(u32));
1386 } while (!subflow->local_nonce);
1389 err = mptcp_pm_get_local_id(msk, (struct sock_common *)ssk);
1396 subflow->remote_key = msk->remote_key;
1397 subflow->local_key = msk->local_key;
1398 subflow->token = msk->token;
1399 mptcp_info2sockaddr(loc, &addr, ssk->sk_family);
1401 addrlen = sizeof(struct sockaddr_in);
1402 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1403 if (addr.ss_family == AF_INET6)
1404 addrlen = sizeof(struct sockaddr_in6);
1406 ssk->sk_bound_dev_if = ifindex;
1407 err = kernel_bind(sf, (struct sockaddr *)&addr, addrlen);
1411 mptcp_crypto_key_sha(subflow->remote_key, &remote_token, NULL);
1412 pr_debug("msk=%p remote_token=%u local_id=%d remote_id=%d", msk,
1413 remote_token, local_id, remote_id);
1414 subflow->remote_token = remote_token;
1415 subflow->local_id = local_id;
1416 subflow->remote_id = remote_id;
1417 subflow->request_join = 1;
1418 subflow->request_bkup = !!(flags & MPTCP_PM_ADDR_FLAG_BACKUP);
1419 mptcp_info2sockaddr(remote, &addr, ssk->sk_family);
1421 mptcp_add_pending_subflow(msk, subflow);
1422 mptcp_sockopt_sync(msk, ssk);
1423 err = kernel_connect(sf, (struct sockaddr *)&addr, addrlen, O_NONBLOCK);
1424 if (err && err != -EINPROGRESS)
1427 /* discard the subflow socket */
1428 mptcp_sock_graft(ssk, sk->sk_socket);
1429 iput(SOCK_INODE(sf));
1433 spin_lock_bh(&msk->join_list_lock);
1434 list_del(&subflow->node);
1435 spin_unlock_bh(&msk->join_list_lock);
1436 sock_put(mptcp_subflow_tcp_sock(subflow));
1439 subflow->disposable = 1;
1444 static void mptcp_attach_cgroup(struct sock *parent, struct sock *child)
1446 #ifdef CONFIG_SOCK_CGROUP_DATA
1447 struct sock_cgroup_data *parent_skcd = &parent->sk_cgrp_data,
1448 *child_skcd = &child->sk_cgrp_data;
1450 /* only the additional subflows created by kworkers have to be modified */
1451 if (cgroup_id(sock_cgroup_ptr(parent_skcd)) !=
1452 cgroup_id(sock_cgroup_ptr(child_skcd))) {
1454 struct mem_cgroup *memcg = parent->sk_memcg;
1456 mem_cgroup_sk_free(child);
1457 if (memcg && css_tryget(&memcg->css))
1458 child->sk_memcg = memcg;
1459 #endif /* CONFIG_MEMCG */
1461 cgroup_sk_free(child_skcd);
1462 *child_skcd = *parent_skcd;
1463 cgroup_sk_clone(child_skcd);
1465 #endif /* CONFIG_SOCK_CGROUP_DATA */
1468 static void mptcp_subflow_ops_override(struct sock *ssk)
1470 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1471 if (ssk->sk_prot == &tcpv6_prot)
1472 ssk->sk_prot = &tcpv6_prot_override;
1475 ssk->sk_prot = &tcp_prot_override;
1478 static void mptcp_subflow_ops_undo_override(struct sock *ssk)
1480 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1481 if (ssk->sk_prot == &tcpv6_prot_override)
1482 ssk->sk_prot = &tcpv6_prot;
1485 ssk->sk_prot = &tcp_prot;
1487 int mptcp_subflow_create_socket(struct sock *sk, struct socket **new_sock)
1489 struct mptcp_subflow_context *subflow;
1490 struct net *net = sock_net(sk);
1494 /* un-accepted server sockets can reach here - on bad configuration
1495 * bail early to avoid greater trouble later
1497 if (unlikely(!sk->sk_socket))
1500 err = sock_create_kern(net, sk->sk_family, SOCK_STREAM, IPPROTO_TCP,
1507 /* the newly created socket has to be in the same cgroup as its parent */
1508 mptcp_attach_cgroup(sk, sf->sk);
1510 /* kernel sockets do not by default acquire net ref, but TCP timer
1513 sf->sk->sk_net_refcnt = 1;
1515 #ifdef CONFIG_PROC_FS
1516 this_cpu_add(*net->core.sock_inuse, 1);
1518 err = tcp_set_ulp(sf->sk, "mptcp");
1519 release_sock(sf->sk);
1526 /* the newly created socket really belongs to the owning MPTCP master
1527 * socket, even if for additional subflows the allocation is performed
1528 * by a kernel workqueue. Adjust inode references, so that the
1529 * procfs/diag interaces really show this one belonging to the correct
1532 SOCK_INODE(sf)->i_ino = SOCK_INODE(sk->sk_socket)->i_ino;
1533 SOCK_INODE(sf)->i_uid = SOCK_INODE(sk->sk_socket)->i_uid;
1534 SOCK_INODE(sf)->i_gid = SOCK_INODE(sk->sk_socket)->i_gid;
1536 subflow = mptcp_subflow_ctx(sf->sk);
1537 pr_debug("subflow=%p", subflow);
1542 mptcp_subflow_ops_override(sf->sk);
1547 static struct mptcp_subflow_context *subflow_create_ctx(struct sock *sk,
1550 struct inet_connection_sock *icsk = inet_csk(sk);
1551 struct mptcp_subflow_context *ctx;
1553 ctx = kzalloc(sizeof(*ctx), priority);
1557 rcu_assign_pointer(icsk->icsk_ulp_data, ctx);
1558 INIT_LIST_HEAD(&ctx->node);
1559 INIT_LIST_HEAD(&ctx->delegated_node);
1561 pr_debug("subflow=%p", ctx);
1568 static void __subflow_state_change(struct sock *sk)
1570 struct socket_wq *wq;
1573 wq = rcu_dereference(sk->sk_wq);
1574 if (skwq_has_sleeper(wq))
1575 wake_up_interruptible_all(&wq->wait);
1579 static bool subflow_is_done(const struct sock *sk)
1581 return sk->sk_shutdown & RCV_SHUTDOWN || sk->sk_state == TCP_CLOSE;
1584 static void subflow_state_change(struct sock *sk)
1586 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1587 struct sock *parent = subflow->conn;
1589 __subflow_state_change(sk);
1591 if (subflow_simultaneous_connect(sk)) {
1592 mptcp_propagate_sndbuf(parent, sk);
1593 mptcp_do_fallback(sk);
1594 mptcp_rcv_space_init(mptcp_sk(parent), sk);
1595 pr_fallback(mptcp_sk(parent));
1596 subflow->conn_finished = 1;
1597 mptcp_set_connected(parent);
1600 /* as recvmsg() does not acquire the subflow socket for ssk selection
1601 * a fin packet carrying a DSS can be unnoticed if we don't trigger
1602 * the data available machinery here.
1604 if (mptcp_subflow_data_available(sk))
1605 mptcp_data_ready(parent, sk);
1606 else if (unlikely(sk->sk_err))
1607 subflow_error_report(sk);
1609 subflow_sched_work_if_closed(mptcp_sk(parent), sk);
1611 if (__mptcp_check_fallback(mptcp_sk(parent)) &&
1612 !subflow->rx_eof && subflow_is_done(sk)) {
1613 subflow->rx_eof = 1;
1614 mptcp_subflow_eof(parent);
1618 static int subflow_ulp_init(struct sock *sk)
1620 struct inet_connection_sock *icsk = inet_csk(sk);
1621 struct mptcp_subflow_context *ctx;
1622 struct tcp_sock *tp = tcp_sk(sk);
1625 /* disallow attaching ULP to a socket unless it has been
1626 * created with sock_create_kern()
1628 if (!sk->sk_kern_sock) {
1633 ctx = subflow_create_ctx(sk, GFP_KERNEL);
1639 pr_debug("subflow=%p, family=%d", ctx, sk->sk_family);
1642 ctx->icsk_af_ops = icsk->icsk_af_ops;
1643 icsk->icsk_af_ops = subflow_default_af_ops(sk);
1644 ctx->tcp_data_ready = sk->sk_data_ready;
1645 ctx->tcp_state_change = sk->sk_state_change;
1646 ctx->tcp_write_space = sk->sk_write_space;
1647 ctx->tcp_error_report = sk->sk_error_report;
1648 sk->sk_data_ready = subflow_data_ready;
1649 sk->sk_write_space = subflow_write_space;
1650 sk->sk_state_change = subflow_state_change;
1651 sk->sk_error_report = subflow_error_report;
1656 static void subflow_ulp_release(struct sock *ssk)
1658 struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(ssk);
1659 bool release = true;
1667 /* if the msk has been orphaned, keep the ctx
1668 * alive, will be freed by __mptcp_close_ssk(),
1669 * when the subflow is still unaccepted
1671 release = ctx->disposable || list_empty(&ctx->node);
1675 mptcp_subflow_ops_undo_override(ssk);
1677 kfree_rcu(ctx, rcu);
1680 static void subflow_ulp_clone(const struct request_sock *req,
1682 const gfp_t priority)
1684 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
1685 struct mptcp_subflow_context *old_ctx = mptcp_subflow_ctx(newsk);
1686 struct mptcp_subflow_context *new_ctx;
1688 if (!tcp_rsk(req)->is_mptcp ||
1689 (!subflow_req->mp_capable && !subflow_req->mp_join)) {
1690 subflow_ulp_fallback(newsk, old_ctx);
1694 new_ctx = subflow_create_ctx(newsk, priority);
1696 subflow_ulp_fallback(newsk, old_ctx);
1700 new_ctx->conn_finished = 1;
1701 new_ctx->icsk_af_ops = old_ctx->icsk_af_ops;
1702 new_ctx->tcp_data_ready = old_ctx->tcp_data_ready;
1703 new_ctx->tcp_state_change = old_ctx->tcp_state_change;
1704 new_ctx->tcp_write_space = old_ctx->tcp_write_space;
1705 new_ctx->tcp_error_report = old_ctx->tcp_error_report;
1706 new_ctx->rel_write_seq = 1;
1707 new_ctx->tcp_sock = newsk;
1709 if (subflow_req->mp_capable) {
1710 /* see comments in subflow_syn_recv_sock(), MPTCP connection
1711 * is fully established only after we receive the remote key
1713 new_ctx->mp_capable = 1;
1714 new_ctx->local_key = subflow_req->local_key;
1715 new_ctx->token = subflow_req->token;
1716 new_ctx->ssn_offset = subflow_req->ssn_offset;
1717 new_ctx->idsn = subflow_req->idsn;
1718 } else if (subflow_req->mp_join) {
1719 new_ctx->ssn_offset = subflow_req->ssn_offset;
1720 new_ctx->mp_join = 1;
1721 new_ctx->fully_established = 1;
1722 new_ctx->backup = subflow_req->backup;
1723 new_ctx->local_id = subflow_req->local_id;
1724 new_ctx->remote_id = subflow_req->remote_id;
1725 new_ctx->token = subflow_req->token;
1726 new_ctx->thmac = subflow_req->thmac;
1730 static void tcp_release_cb_override(struct sock *ssk)
1732 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1734 if (mptcp_subflow_has_delegated_action(subflow))
1735 mptcp_subflow_process_delegated(ssk);
1737 tcp_release_cb(ssk);
1740 static struct tcp_ulp_ops subflow_ulp_ops __read_mostly = {
1742 .owner = THIS_MODULE,
1743 .init = subflow_ulp_init,
1744 .release = subflow_ulp_release,
1745 .clone = subflow_ulp_clone,
1748 static int subflow_ops_init(struct request_sock_ops *subflow_ops)
1750 subflow_ops->obj_size = sizeof(struct mptcp_subflow_request_sock);
1751 subflow_ops->slab_name = "request_sock_subflow";
1753 subflow_ops->slab = kmem_cache_create(subflow_ops->slab_name,
1754 subflow_ops->obj_size, 0,
1756 SLAB_TYPESAFE_BY_RCU,
1758 if (!subflow_ops->slab)
1761 subflow_ops->destructor = subflow_req_destructor;
1766 void __init mptcp_subflow_init(void)
1768 mptcp_subflow_request_sock_ops = tcp_request_sock_ops;
1769 if (subflow_ops_init(&mptcp_subflow_request_sock_ops) != 0)
1770 panic("MPTCP: failed to init subflow request sock ops\n");
1772 subflow_request_sock_ipv4_ops = tcp_request_sock_ipv4_ops;
1773 subflow_request_sock_ipv4_ops.route_req = subflow_v4_route_req;
1775 subflow_specific = ipv4_specific;
1776 subflow_specific.conn_request = subflow_v4_conn_request;
1777 subflow_specific.syn_recv_sock = subflow_syn_recv_sock;
1778 subflow_specific.sk_rx_dst_set = subflow_finish_connect;
1780 tcp_prot_override = tcp_prot;
1781 tcp_prot_override.release_cb = tcp_release_cb_override;
1783 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1784 subflow_request_sock_ipv6_ops = tcp_request_sock_ipv6_ops;
1785 subflow_request_sock_ipv6_ops.route_req = subflow_v6_route_req;
1787 subflow_v6_specific = ipv6_specific;
1788 subflow_v6_specific.conn_request = subflow_v6_conn_request;
1789 subflow_v6_specific.syn_recv_sock = subflow_syn_recv_sock;
1790 subflow_v6_specific.sk_rx_dst_set = subflow_finish_connect;
1792 subflow_v6m_specific = subflow_v6_specific;
1793 subflow_v6m_specific.queue_xmit = ipv4_specific.queue_xmit;
1794 subflow_v6m_specific.send_check = ipv4_specific.send_check;
1795 subflow_v6m_specific.net_header_len = ipv4_specific.net_header_len;
1796 subflow_v6m_specific.mtu_reduced = ipv4_specific.mtu_reduced;
1797 subflow_v6m_specific.net_frag_header_len = 0;
1799 tcpv6_prot_override = tcpv6_prot;
1800 tcpv6_prot_override.release_cb = tcp_release_cb_override;
1803 mptcp_diag_subflow_init(&subflow_ulp_ops);
1805 if (tcp_register_ulp(&subflow_ulp_ops) != 0)
1806 panic("MPTCP: failed to register subflows to ULP\n");