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/sha.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>
22 #include <net/mptcp.h>
23 #include <uapi/linux/mptcp.h>
27 static void SUBFLOW_REQ_INC_STATS(struct request_sock *req,
28 enum linux_mptcp_mib_field field)
30 MPTCP_INC_STATS(sock_net(req_to_sk(req)), field);
33 static void subflow_req_destructor(struct request_sock *req)
35 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
37 pr_debug("subflow_req=%p", subflow_req);
40 sock_put((struct sock *)subflow_req->msk);
42 mptcp_token_destroy_request(req);
43 tcp_request_sock_ops.destructor(req);
46 static void subflow_generate_hmac(u64 key1, u64 key2, u32 nonce1, u32 nonce2,
51 put_unaligned_be32(nonce1, &msg[0]);
52 put_unaligned_be32(nonce2, &msg[4]);
54 mptcp_crypto_hmac_sha(key1, key2, msg, 8, hmac);
57 static bool mptcp_can_accept_new_subflow(const struct mptcp_sock *msk)
59 return mptcp_is_fully_established((void *)msk) &&
60 READ_ONCE(msk->pm.accept_subflow);
63 /* validate received token and create truncated hmac and nonce for SYN-ACK */
64 static struct mptcp_sock *subflow_token_join_request(struct request_sock *req,
65 const struct sk_buff *skb)
67 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
68 u8 hmac[SHA256_DIGEST_SIZE];
69 struct mptcp_sock *msk;
72 msk = mptcp_token_get_sock(subflow_req->token);
74 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINNOTOKEN);
78 local_id = mptcp_pm_get_local_id(msk, (struct sock_common *)req);
80 sock_put((struct sock *)msk);
83 subflow_req->local_id = local_id;
85 get_random_bytes(&subflow_req->local_nonce, sizeof(u32));
87 subflow_generate_hmac(msk->local_key, msk->remote_key,
88 subflow_req->local_nonce,
89 subflow_req->remote_nonce, hmac);
91 subflow_req->thmac = get_unaligned_be64(hmac);
95 static int __subflow_init_req(struct request_sock *req, const struct sock *sk_listener)
97 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
99 subflow_req->mp_capable = 0;
100 subflow_req->mp_join = 0;
101 subflow_req->msk = NULL;
102 mptcp_token_init_request(req);
104 #ifdef CONFIG_TCP_MD5SIG
105 /* no MPTCP if MD5SIG is enabled on this socket or we may run out of
108 if (rcu_access_pointer(tcp_sk(sk_listener)->md5sig_info))
115 static void subflow_init_req(struct request_sock *req,
116 const struct sock *sk_listener,
119 struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk_listener);
120 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
121 struct mptcp_options_received mp_opt;
124 pr_debug("subflow_req=%p, listener=%p", subflow_req, listener);
126 ret = __subflow_init_req(req, sk_listener);
130 mptcp_get_options(skb, &mp_opt);
132 if (mp_opt.mp_capable) {
133 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MPCAPABLEPASSIVE);
137 } else if (mp_opt.mp_join) {
138 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINSYNRX);
141 if (mp_opt.mp_capable && listener->request_mptcp) {
142 int err, retries = 4;
144 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq;
147 get_random_bytes(&subflow_req->local_key, sizeof(subflow_req->local_key));
148 } while (subflow_req->local_key == 0);
150 if (unlikely(req->syncookie)) {
151 mptcp_crypto_key_sha(subflow_req->local_key,
154 if (mptcp_token_exists(subflow_req->token)) {
158 subflow_req->mp_capable = 1;
163 err = mptcp_token_new_request(req);
165 subflow_req->mp_capable = 1;
166 else if (retries-- > 0)
169 } else if (mp_opt.mp_join && listener->request_mptcp) {
170 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq;
171 subflow_req->mp_join = 1;
172 subflow_req->backup = mp_opt.backup;
173 subflow_req->remote_id = mp_opt.join_id;
174 subflow_req->token = mp_opt.token;
175 subflow_req->remote_nonce = mp_opt.nonce;
176 subflow_req->msk = subflow_token_join_request(req, skb);
178 if (unlikely(req->syncookie) && subflow_req->msk) {
179 if (mptcp_can_accept_new_subflow(subflow_req->msk))
180 subflow_init_req_cookie_join_save(subflow_req, skb);
183 pr_debug("token=%u, remote_nonce=%u msk=%p", subflow_req->token,
184 subflow_req->remote_nonce, subflow_req->msk);
188 int mptcp_subflow_init_cookie_req(struct request_sock *req,
189 const struct sock *sk_listener,
192 struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk_listener);
193 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
194 struct mptcp_options_received mp_opt;
197 err = __subflow_init_req(req, sk_listener);
201 mptcp_get_options(skb, &mp_opt);
203 if (mp_opt.mp_capable && mp_opt.mp_join)
206 if (mp_opt.mp_capable && listener->request_mptcp) {
207 if (mp_opt.sndr_key == 0)
210 subflow_req->local_key = mp_opt.rcvr_key;
211 err = mptcp_token_new_request(req);
215 subflow_req->mp_capable = 1;
216 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq - 1;
217 } else if (mp_opt.mp_join && listener->request_mptcp) {
218 if (!mptcp_token_join_cookie_init_state(subflow_req, skb))
221 if (mptcp_can_accept_new_subflow(subflow_req->msk))
222 subflow_req->mp_join = 1;
224 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq - 1;
229 EXPORT_SYMBOL_GPL(mptcp_subflow_init_cookie_req);
231 static void subflow_v4_init_req(struct request_sock *req,
232 const struct sock *sk_listener,
235 tcp_rsk(req)->is_mptcp = 1;
237 tcp_request_sock_ipv4_ops.init_req(req, sk_listener, skb);
239 subflow_init_req(req, sk_listener, skb);
242 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
243 static void subflow_v6_init_req(struct request_sock *req,
244 const struct sock *sk_listener,
247 tcp_rsk(req)->is_mptcp = 1;
249 tcp_request_sock_ipv6_ops.init_req(req, sk_listener, skb);
251 subflow_init_req(req, sk_listener, skb);
255 /* validate received truncated hmac and create hmac for third ACK */
256 static bool subflow_thmac_valid(struct mptcp_subflow_context *subflow)
258 u8 hmac[SHA256_DIGEST_SIZE];
261 subflow_generate_hmac(subflow->remote_key, subflow->local_key,
262 subflow->remote_nonce, subflow->local_nonce,
265 thmac = get_unaligned_be64(hmac);
266 pr_debug("subflow=%p, token=%u, thmac=%llu, subflow->thmac=%llu\n",
267 subflow, subflow->token,
268 (unsigned long long)thmac,
269 (unsigned long long)subflow->thmac);
271 return thmac == subflow->thmac;
274 void mptcp_subflow_reset(struct sock *ssk)
276 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
277 struct sock *sk = subflow->conn;
279 tcp_set_state(ssk, TCP_CLOSE);
280 tcp_send_active_reset(ssk, GFP_ATOMIC);
282 if (!test_and_set_bit(MPTCP_WORK_CLOSE_SUBFLOW, &mptcp_sk(sk)->flags) &&
283 schedule_work(&mptcp_sk(sk)->work))
287 static void subflow_finish_connect(struct sock *sk, const struct sk_buff *skb)
289 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
290 struct mptcp_options_received mp_opt;
291 struct sock *parent = subflow->conn;
293 subflow->icsk_af_ops->sk_rx_dst_set(sk, skb);
295 if (inet_sk_state_load(parent) == TCP_SYN_SENT) {
296 inet_sk_state_store(parent, TCP_ESTABLISHED);
297 parent->sk_state_change(parent);
300 /* be sure no special action on any packet other than syn-ack */
301 if (subflow->conn_finished)
304 subflow->rel_write_seq = 1;
305 subflow->conn_finished = 1;
306 subflow->ssn_offset = TCP_SKB_CB(skb)->seq;
307 pr_debug("subflow=%p synack seq=%x", subflow, subflow->ssn_offset);
309 mptcp_get_options(skb, &mp_opt);
310 if (subflow->request_mptcp) {
311 if (!mp_opt.mp_capable) {
312 MPTCP_INC_STATS(sock_net(sk),
313 MPTCP_MIB_MPCAPABLEACTIVEFALLBACK);
314 mptcp_do_fallback(sk);
315 pr_fallback(mptcp_sk(subflow->conn));
319 subflow->mp_capable = 1;
320 subflow->can_ack = 1;
321 subflow->remote_key = mp_opt.sndr_key;
322 pr_debug("subflow=%p, remote_key=%llu", subflow,
323 subflow->remote_key);
324 mptcp_finish_connect(sk);
325 } else if (subflow->request_join) {
326 u8 hmac[SHA256_DIGEST_SIZE];
331 subflow->thmac = mp_opt.thmac;
332 subflow->remote_nonce = mp_opt.nonce;
333 pr_debug("subflow=%p, thmac=%llu, remote_nonce=%u", subflow,
334 subflow->thmac, subflow->remote_nonce);
336 if (!subflow_thmac_valid(subflow)) {
337 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINACKMAC);
341 subflow_generate_hmac(subflow->local_key, subflow->remote_key,
342 subflow->local_nonce,
343 subflow->remote_nonce,
345 memcpy(subflow->hmac, hmac, MPTCPOPT_HMAC_LEN);
347 if (!mptcp_finish_join(sk))
350 subflow->mp_join = 1;
351 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINSYNACKRX);
352 } else if (mptcp_check_fallback(sk)) {
354 mptcp_rcv_space_init(mptcp_sk(parent), sk);
359 mptcp_subflow_reset(sk);
362 struct request_sock_ops mptcp_subflow_request_sock_ops;
363 EXPORT_SYMBOL_GPL(mptcp_subflow_request_sock_ops);
364 static struct tcp_request_sock_ops subflow_request_sock_ipv4_ops;
366 static int subflow_v4_conn_request(struct sock *sk, struct sk_buff *skb)
368 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
370 pr_debug("subflow=%p", subflow);
372 /* Never answer to SYNs sent to broadcast or multicast */
373 if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
376 return tcp_conn_request(&mptcp_subflow_request_sock_ops,
377 &subflow_request_sock_ipv4_ops,
384 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
385 static struct tcp_request_sock_ops subflow_request_sock_ipv6_ops;
386 static struct inet_connection_sock_af_ops subflow_v6_specific;
387 static struct inet_connection_sock_af_ops subflow_v6m_specific;
389 static int subflow_v6_conn_request(struct sock *sk, struct sk_buff *skb)
391 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
393 pr_debug("subflow=%p", subflow);
395 if (skb->protocol == htons(ETH_P_IP))
396 return subflow_v4_conn_request(sk, skb);
398 if (!ipv6_unicast_destination(skb))
401 return tcp_conn_request(&mptcp_subflow_request_sock_ops,
402 &subflow_request_sock_ipv6_ops, sk, skb);
406 return 0; /* don't send reset */
410 /* validate hmac received in third ACK */
411 static bool subflow_hmac_valid(const struct request_sock *req,
412 const struct mptcp_options_received *mp_opt)
414 const struct mptcp_subflow_request_sock *subflow_req;
415 u8 hmac[SHA256_DIGEST_SIZE];
416 struct mptcp_sock *msk;
418 subflow_req = mptcp_subflow_rsk(req);
419 msk = subflow_req->msk;
423 subflow_generate_hmac(msk->remote_key, msk->local_key,
424 subflow_req->remote_nonce,
425 subflow_req->local_nonce, hmac);
427 return !crypto_memneq(hmac, mp_opt->hmac, MPTCPOPT_HMAC_LEN);
430 static void mptcp_sock_destruct(struct sock *sk)
432 /* if new mptcp socket isn't accepted, it is free'd
433 * from the tcp listener sockets request queue, linked
434 * from req->sk. The tcp socket is released.
435 * This calls the ULP release function which will
436 * also remove the mptcp socket, via
437 * sock_put(ctx->conn).
439 * Problem is that the mptcp socket will be in
440 * ESTABLISHED state and will not have the SOCK_DEAD flag.
441 * Both result in warnings from inet_sock_destruct.
444 if (sk->sk_state == TCP_ESTABLISHED) {
445 sk->sk_state = TCP_CLOSE;
446 WARN_ON_ONCE(sk->sk_socket);
450 mptcp_destroy_common(mptcp_sk(sk));
451 inet_sock_destruct(sk);
454 static void mptcp_force_close(struct sock *sk)
456 inet_sk_state_store(sk, TCP_CLOSE);
457 sk_common_release(sk);
460 static void subflow_ulp_fallback(struct sock *sk,
461 struct mptcp_subflow_context *old_ctx)
463 struct inet_connection_sock *icsk = inet_csk(sk);
465 mptcp_subflow_tcp_fallback(sk, old_ctx);
466 icsk->icsk_ulp_ops = NULL;
467 rcu_assign_pointer(icsk->icsk_ulp_data, NULL);
468 tcp_sk(sk)->is_mptcp = 0;
471 static void subflow_drop_ctx(struct sock *ssk)
473 struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(ssk);
478 subflow_ulp_fallback(ssk, ctx);
485 void mptcp_subflow_fully_established(struct mptcp_subflow_context *subflow,
486 struct mptcp_options_received *mp_opt)
488 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
490 subflow->remote_key = mp_opt->sndr_key;
491 subflow->fully_established = 1;
492 subflow->can_ack = 1;
493 WRITE_ONCE(msk->fully_established, true);
496 static struct sock *subflow_syn_recv_sock(const struct sock *sk,
498 struct request_sock *req,
499 struct dst_entry *dst,
500 struct request_sock *req_unhash,
503 struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk);
504 struct mptcp_subflow_request_sock *subflow_req;
505 struct mptcp_options_received mp_opt;
506 bool fallback, fallback_is_fatal;
507 struct sock *new_msk = NULL;
510 pr_debug("listener=%p, req=%p, conn=%p", listener, req, listener->conn);
512 /* After child creation we must look for 'mp_capable' even when options
515 mp_opt.mp_capable = 0;
517 /* hopefully temporary handling for MP_JOIN+syncookie */
518 subflow_req = mptcp_subflow_rsk(req);
519 fallback_is_fatal = tcp_rsk(req)->is_mptcp && subflow_req->mp_join;
520 fallback = !tcp_rsk(req)->is_mptcp;
524 /* if the sk is MP_CAPABLE, we try to fetch the client key */
525 if (subflow_req->mp_capable) {
526 if (TCP_SKB_CB(skb)->seq != subflow_req->ssn_offset + 1) {
527 /* here we can receive and accept an in-window,
528 * out-of-order pkt, which will not carry the MP_CAPABLE
529 * opt even on mptcp enabled paths
534 mptcp_get_options(skb, &mp_opt);
535 if (!mp_opt.mp_capable) {
541 new_msk = mptcp_sk_clone(listener->conn, &mp_opt, req);
544 } else if (subflow_req->mp_join) {
545 mptcp_get_options(skb, &mp_opt);
546 if (!mp_opt.mp_join ||
547 !mptcp_can_accept_new_subflow(subflow_req->msk) ||
548 !subflow_hmac_valid(req, &mp_opt)) {
549 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKMAC);
555 child = listener->icsk_af_ops->syn_recv_sock(sk, skb, req, dst,
556 req_unhash, own_req);
558 if (child && *own_req) {
559 struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(child);
561 tcp_rsk(req)->drop_req = false;
563 /* we need to fallback on ctx allocation failure and on pre-reqs
564 * checking above. In the latter scenario we additionally need
565 * to reset the context to non MPTCP status.
567 if (!ctx || fallback) {
568 if (fallback_is_fatal)
571 subflow_drop_ctx(child);
575 if (ctx->mp_capable) {
576 /* this can't race with mptcp_close(), as the msk is
577 * not yet exposted to user-space
579 inet_sk_state_store((void *)new_msk, TCP_ESTABLISHED);
581 /* new mpc subflow takes ownership of the newly
582 * created mptcp socket
584 new_msk->sk_destruct = mptcp_sock_destruct;
585 mptcp_pm_new_connection(mptcp_sk(new_msk), 1);
586 mptcp_token_accept(subflow_req, mptcp_sk(new_msk));
590 /* with OoO packets we can reach here without ingress
593 if (mp_opt.mp_capable)
594 mptcp_subflow_fully_established(ctx, &mp_opt);
595 } else if (ctx->mp_join) {
596 struct mptcp_sock *owner;
598 owner = subflow_req->msk;
602 /* move the msk reference ownership to the subflow */
603 subflow_req->msk = NULL;
604 ctx->conn = (struct sock *)owner;
605 if (!mptcp_finish_join(child))
608 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKRX);
609 tcp_rsk(req)->drop_req = true;
614 /* dispose of the left over mptcp master, if any */
615 if (unlikely(new_msk))
616 mptcp_force_close(new_msk);
618 /* check for expected invariant - should never trigger, just help
619 * catching eariler subtle bugs
621 WARN_ON_ONCE(child && *own_req && tcp_sk(child)->is_mptcp &&
622 (!mptcp_subflow_ctx(child) ||
623 !mptcp_subflow_ctx(child)->conn));
627 subflow_drop_ctx(child);
628 tcp_rsk(req)->drop_req = true;
629 inet_csk_prepare_for_destroy_sock(child);
631 req->rsk_ops->send_reset(sk, skb);
633 /* The last child reference will be released by the caller */
637 static struct inet_connection_sock_af_ops subflow_specific;
639 enum mapping_status {
647 static u64 expand_seq(u64 old_seq, u16 old_data_len, u64 seq)
649 if ((u32)seq == (u32)old_seq)
652 /* Assume map covers data not mapped yet. */
653 return seq | ((old_seq + old_data_len + 1) & GENMASK_ULL(63, 32));
656 static void warn_bad_map(struct mptcp_subflow_context *subflow, u32 ssn)
658 WARN_ONCE(1, "Bad mapping: ssn=%d map_seq=%d map_data_len=%d",
659 ssn, subflow->map_subflow_seq, subflow->map_data_len);
662 static bool skb_is_fully_mapped(struct sock *ssk, struct sk_buff *skb)
664 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
665 unsigned int skb_consumed;
667 skb_consumed = tcp_sk(ssk)->copied_seq - TCP_SKB_CB(skb)->seq;
668 if (WARN_ON_ONCE(skb_consumed >= skb->len))
671 return skb->len - skb_consumed <= subflow->map_data_len -
672 mptcp_subflow_get_map_offset(subflow);
675 static bool validate_mapping(struct sock *ssk, struct sk_buff *skb)
677 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
678 u32 ssn = tcp_sk(ssk)->copied_seq - subflow->ssn_offset;
680 if (unlikely(before(ssn, subflow->map_subflow_seq))) {
681 /* Mapping covers data later in the subflow stream,
682 * currently unsupported.
684 warn_bad_map(subflow, ssn);
687 if (unlikely(!before(ssn, subflow->map_subflow_seq +
688 subflow->map_data_len))) {
689 /* Mapping does covers past subflow data, invalid */
690 warn_bad_map(subflow, ssn + skb->len);
696 static enum mapping_status get_mapping_status(struct sock *ssk,
697 struct mptcp_sock *msk)
699 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
700 struct mptcp_ext *mpext;
705 skb = skb_peek(&ssk->sk_receive_queue);
707 return MAPPING_EMPTY;
709 if (mptcp_check_fallback(ssk))
710 return MAPPING_DUMMY;
712 mpext = mptcp_get_ext(skb);
713 if (!mpext || !mpext->use_map) {
714 if (!subflow->map_valid && !skb->len) {
715 /* the TCP stack deliver 0 len FIN pkt to the receive
716 * queue, that is the only 0len pkts ever expected here,
717 * and we can admit no mapping only for 0 len pkts
719 if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN))
720 WARN_ONCE(1, "0len seq %d:%d flags %x",
721 TCP_SKB_CB(skb)->seq,
722 TCP_SKB_CB(skb)->end_seq,
723 TCP_SKB_CB(skb)->tcp_flags);
724 sk_eat_skb(ssk, skb);
725 return MAPPING_EMPTY;
728 if (!subflow->map_valid)
729 return MAPPING_INVALID;
734 pr_debug("seq=%llu is64=%d ssn=%u data_len=%u data_fin=%d",
735 mpext->data_seq, mpext->dsn64, mpext->subflow_seq,
736 mpext->data_len, mpext->data_fin);
738 data_len = mpext->data_len;
740 pr_err("Infinite mapping not handled");
741 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_INFINITEMAPRX);
742 return MAPPING_INVALID;
745 if (mpext->data_fin == 1) {
747 bool updated = mptcp_update_rcv_data_fin(msk, mpext->data_seq,
749 pr_debug("DATA_FIN with no payload seq=%llu", mpext->data_seq);
750 if (subflow->map_valid) {
751 /* A DATA_FIN might arrive in a DSS
752 * option before the previous mapping
753 * has been fully consumed. Continue
754 * handling the existing mapping.
756 skb_ext_del(skb, SKB_EXT_MPTCP);
759 if (updated && schedule_work(&msk->work))
760 sock_hold((struct sock *)msk);
762 return MAPPING_DATA_FIN;
765 u64 data_fin_seq = mpext->data_seq + data_len - 1;
767 /* If mpext->data_seq is a 32-bit value, data_fin_seq
768 * must also be limited to 32 bits.
771 data_fin_seq &= GENMASK_ULL(31, 0);
773 mptcp_update_rcv_data_fin(msk, data_fin_seq, mpext->dsn64);
774 pr_debug("DATA_FIN with mapping seq=%llu dsn64=%d",
775 data_fin_seq, mpext->dsn64);
778 /* Adjust for DATA_FIN using 1 byte of sequence space */
783 map_seq = expand_seq(subflow->map_seq, subflow->map_data_len,
785 pr_debug("expanded seq=%llu", subflow->map_seq);
787 map_seq = mpext->data_seq;
789 WRITE_ONCE(mptcp_sk(subflow->conn)->use_64bit_ack, !!mpext->dsn64);
791 if (subflow->map_valid) {
792 /* Allow replacing only with an identical map */
793 if (subflow->map_seq == map_seq &&
794 subflow->map_subflow_seq == mpext->subflow_seq &&
795 subflow->map_data_len == data_len) {
796 skb_ext_del(skb, SKB_EXT_MPTCP);
800 /* If this skb data are fully covered by the current mapping,
801 * the new map would need caching, which is not supported
803 if (skb_is_fully_mapped(ssk, skb)) {
804 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DSSNOMATCH);
805 return MAPPING_INVALID;
808 /* will validate the next map after consuming the current one */
812 subflow->map_seq = map_seq;
813 subflow->map_subflow_seq = mpext->subflow_seq;
814 subflow->map_data_len = data_len;
815 subflow->map_valid = 1;
816 subflow->mpc_map = mpext->mpc_map;
817 pr_debug("new map seq=%llu subflow_seq=%u data_len=%u",
818 subflow->map_seq, subflow->map_subflow_seq,
819 subflow->map_data_len);
822 /* we revalidate valid mapping on new skb, because we must ensure
823 * the current skb is completely covered by the available mapping
825 if (!validate_mapping(ssk, skb))
826 return MAPPING_INVALID;
828 skb_ext_del(skb, SKB_EXT_MPTCP);
832 static void mptcp_subflow_discard_data(struct sock *ssk, struct sk_buff *skb,
835 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
836 bool fin = TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN;
839 incr = limit >= skb->len ? skb->len + fin : limit;
841 pr_debug("discarding=%d len=%d seq=%d", incr, skb->len,
842 subflow->map_subflow_seq);
843 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DUPDATA);
844 tcp_sk(ssk)->copied_seq += incr;
845 if (!before(tcp_sk(ssk)->copied_seq, TCP_SKB_CB(skb)->end_seq))
846 sk_eat_skb(ssk, skb);
847 if (mptcp_subflow_get_map_offset(subflow) >= subflow->map_data_len)
848 subflow->map_valid = 0;
850 tcp_cleanup_rbuf(ssk, incr);
853 static bool subflow_check_data_avail(struct sock *ssk)
855 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
856 enum mapping_status status;
857 struct mptcp_sock *msk;
860 pr_debug("msk=%p ssk=%p data_avail=%d skb=%p", subflow->conn, ssk,
861 subflow->data_avail, skb_peek(&ssk->sk_receive_queue));
862 if (!skb_peek(&ssk->sk_receive_queue))
863 subflow->data_avail = 0;
864 if (subflow->data_avail)
867 msk = mptcp_sk(subflow->conn);
872 status = get_mapping_status(ssk, msk);
873 pr_debug("msk=%p ssk=%p status=%d", msk, ssk, status);
874 if (status == MAPPING_INVALID) {
875 ssk->sk_err = EBADMSG;
878 if (status == MAPPING_DUMMY) {
879 __mptcp_do_fallback(msk);
880 skb = skb_peek(&ssk->sk_receive_queue);
881 subflow->map_valid = 1;
882 subflow->map_seq = READ_ONCE(msk->ack_seq);
883 subflow->map_data_len = skb->len;
884 subflow->map_subflow_seq = tcp_sk(ssk)->copied_seq -
886 subflow->data_avail = MPTCP_SUBFLOW_DATA_AVAIL;
890 if (status != MAPPING_OK)
893 skb = skb_peek(&ssk->sk_receive_queue);
894 if (WARN_ON_ONCE(!skb))
897 /* if msk lacks the remote key, this subflow must provide an
898 * MP_CAPABLE-based mapping
900 if (unlikely(!READ_ONCE(msk->can_ack))) {
901 if (!subflow->mpc_map) {
902 ssk->sk_err = EBADMSG;
905 WRITE_ONCE(msk->remote_key, subflow->remote_key);
906 WRITE_ONCE(msk->ack_seq, subflow->map_seq);
907 WRITE_ONCE(msk->can_ack, true);
910 old_ack = READ_ONCE(msk->ack_seq);
911 ack_seq = mptcp_subflow_get_mapped_dsn(subflow);
912 pr_debug("msk ack_seq=%llx subflow ack_seq=%llx", old_ack,
914 if (ack_seq == old_ack) {
915 subflow->data_avail = MPTCP_SUBFLOW_DATA_AVAIL;
917 } else if (after64(ack_seq, old_ack)) {
918 subflow->data_avail = MPTCP_SUBFLOW_OOO_DATA;
922 /* only accept in-sequence mapping. Old values are spurious
925 mptcp_subflow_discard_data(ssk, skb, old_ack - ack_seq);
930 /* fatal protocol error, close the socket */
931 /* This barrier is coupled with smp_rmb() in tcp_poll() */
933 ssk->sk_error_report(ssk);
934 tcp_set_state(ssk, TCP_CLOSE);
935 tcp_send_active_reset(ssk, GFP_ATOMIC);
936 subflow->data_avail = 0;
940 bool mptcp_subflow_data_available(struct sock *sk)
942 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
944 /* check if current mapping is still valid */
945 if (subflow->map_valid &&
946 mptcp_subflow_get_map_offset(subflow) >= subflow->map_data_len) {
947 subflow->map_valid = 0;
948 subflow->data_avail = 0;
950 pr_debug("Done with mapping: seq=%u data_len=%u",
951 subflow->map_subflow_seq,
952 subflow->map_data_len);
955 return subflow_check_data_avail(sk);
958 /* If ssk has an mptcp parent socket, use the mptcp rcvbuf occupancy,
961 * In mptcp, rwin is about the mptcp-level connection data.
963 * Data that is still on the ssk rx queue can thus be ignored,
964 * as far as mptcp peer is concerened that data is still inflight.
965 * DSS ACK is updated when skb is moved to the mptcp rx queue.
967 void mptcp_space(const struct sock *ssk, int *space, int *full_space)
969 const struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
970 const struct sock *sk = subflow->conn;
972 *space = tcp_space(sk);
973 *full_space = tcp_full_space(sk);
976 static void subflow_data_ready(struct sock *sk)
978 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
979 u16 state = 1 << inet_sk_state_load(sk);
980 struct sock *parent = subflow->conn;
981 struct mptcp_sock *msk;
983 msk = mptcp_sk(parent);
984 if (state & TCPF_LISTEN) {
985 set_bit(MPTCP_DATA_READY, &msk->flags);
986 parent->sk_data_ready(parent);
990 WARN_ON_ONCE(!__mptcp_check_fallback(msk) && !subflow->mp_capable &&
991 !subflow->mp_join && !(state & TCPF_CLOSE));
993 if (mptcp_subflow_data_available(sk))
994 mptcp_data_ready(parent, sk);
997 static void subflow_write_space(struct sock *sk)
999 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1000 struct sock *parent = subflow->conn;
1002 if (!sk_stream_is_writeable(sk))
1005 if (sk_stream_is_writeable(parent)) {
1006 set_bit(MPTCP_SEND_SPACE, &mptcp_sk(parent)->flags);
1007 smp_mb__after_atomic();
1008 /* set SEND_SPACE before sk_stream_write_space clears NOSPACE */
1009 sk_stream_write_space(parent);
1013 static struct inet_connection_sock_af_ops *
1014 subflow_default_af_ops(struct sock *sk)
1016 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1017 if (sk->sk_family == AF_INET6)
1018 return &subflow_v6_specific;
1020 return &subflow_specific;
1023 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1024 void mptcpv6_handle_mapped(struct sock *sk, bool mapped)
1026 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1027 struct inet_connection_sock *icsk = inet_csk(sk);
1028 struct inet_connection_sock_af_ops *target;
1030 target = mapped ? &subflow_v6m_specific : subflow_default_af_ops(sk);
1032 pr_debug("subflow=%p family=%d ops=%p target=%p mapped=%d",
1033 subflow, sk->sk_family, icsk->icsk_af_ops, target, mapped);
1035 if (likely(icsk->icsk_af_ops == target))
1038 subflow->icsk_af_ops = icsk->icsk_af_ops;
1039 icsk->icsk_af_ops = target;
1043 static void mptcp_info2sockaddr(const struct mptcp_addr_info *info,
1044 struct sockaddr_storage *addr)
1046 memset(addr, 0, sizeof(*addr));
1047 addr->ss_family = info->family;
1048 if (addr->ss_family == AF_INET) {
1049 struct sockaddr_in *in_addr = (struct sockaddr_in *)addr;
1051 in_addr->sin_addr = info->addr;
1052 in_addr->sin_port = info->port;
1054 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1055 else if (addr->ss_family == AF_INET6) {
1056 struct sockaddr_in6 *in6_addr = (struct sockaddr_in6 *)addr;
1058 in6_addr->sin6_addr = info->addr6;
1059 in6_addr->sin6_port = info->port;
1064 int __mptcp_subflow_connect(struct sock *sk, const struct mptcp_addr_info *loc,
1065 const struct mptcp_addr_info *remote)
1067 struct mptcp_sock *msk = mptcp_sk(sk);
1068 struct mptcp_subflow_context *subflow;
1069 struct sockaddr_storage addr;
1070 int remote_id = remote->id;
1071 int local_id = loc->id;
1078 if (!mptcp_is_fully_established(sk))
1081 err = mptcp_subflow_create_socket(sk, &sf);
1086 subflow = mptcp_subflow_ctx(ssk);
1088 get_random_bytes(&subflow->local_nonce, sizeof(u32));
1089 } while (!subflow->local_nonce);
1092 err = mptcp_pm_get_local_id(msk, (struct sock_common *)ssk);
1099 subflow->remote_key = msk->remote_key;
1100 subflow->local_key = msk->local_key;
1101 subflow->token = msk->token;
1102 mptcp_info2sockaddr(loc, &addr);
1104 addrlen = sizeof(struct sockaddr_in);
1105 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1106 if (loc->family == AF_INET6)
1107 addrlen = sizeof(struct sockaddr_in6);
1109 ssk->sk_bound_dev_if = loc->ifindex;
1110 err = kernel_bind(sf, (struct sockaddr *)&addr, addrlen);
1114 mptcp_crypto_key_sha(subflow->remote_key, &remote_token, NULL);
1115 pr_debug("msk=%p remote_token=%u local_id=%d remote_id=%d", msk,
1116 remote_token, local_id, remote_id);
1117 subflow->remote_token = remote_token;
1118 subflow->local_id = local_id;
1119 subflow->remote_id = remote_id;
1120 subflow->request_join = 1;
1121 subflow->request_bkup = !!(loc->flags & MPTCP_PM_ADDR_FLAG_BACKUP);
1122 mptcp_info2sockaddr(remote, &addr);
1124 err = kernel_connect(sf, (struct sockaddr *)&addr, addrlen, O_NONBLOCK);
1125 if (err && err != -EINPROGRESS)
1128 spin_lock_bh(&msk->join_list_lock);
1129 list_add_tail(&subflow->node, &msk->join_list);
1130 spin_unlock_bh(&msk->join_list_lock);
1139 int mptcp_subflow_create_socket(struct sock *sk, struct socket **new_sock)
1141 struct mptcp_subflow_context *subflow;
1142 struct net *net = sock_net(sk);
1146 /* un-accepted server sockets can reach here - on bad configuration
1147 * bail early to avoid greater trouble later
1149 if (unlikely(!sk->sk_socket))
1152 err = sock_create_kern(net, sk->sk_family, SOCK_STREAM, IPPROTO_TCP,
1159 /* kernel sockets do not by default acquire net ref, but TCP timer
1162 sf->sk->sk_net_refcnt = 1;
1164 #ifdef CONFIG_PROC_FS
1165 this_cpu_add(*net->core.sock_inuse, 1);
1167 err = tcp_set_ulp(sf->sk, "mptcp");
1168 release_sock(sf->sk);
1175 /* the newly created socket really belongs to the owning MPTCP master
1176 * socket, even if for additional subflows the allocation is performed
1177 * by a kernel workqueue. Adjust inode references, so that the
1178 * procfs/diag interaces really show this one belonging to the correct
1181 SOCK_INODE(sf)->i_ino = SOCK_INODE(sk->sk_socket)->i_ino;
1182 SOCK_INODE(sf)->i_uid = SOCK_INODE(sk->sk_socket)->i_uid;
1183 SOCK_INODE(sf)->i_gid = SOCK_INODE(sk->sk_socket)->i_gid;
1185 subflow = mptcp_subflow_ctx(sf->sk);
1186 pr_debug("subflow=%p", subflow);
1195 static struct mptcp_subflow_context *subflow_create_ctx(struct sock *sk,
1198 struct inet_connection_sock *icsk = inet_csk(sk);
1199 struct mptcp_subflow_context *ctx;
1201 ctx = kzalloc(sizeof(*ctx), priority);
1205 rcu_assign_pointer(icsk->icsk_ulp_data, ctx);
1206 INIT_LIST_HEAD(&ctx->node);
1208 pr_debug("subflow=%p", ctx);
1215 static void __subflow_state_change(struct sock *sk)
1217 struct socket_wq *wq;
1220 wq = rcu_dereference(sk->sk_wq);
1221 if (skwq_has_sleeper(wq))
1222 wake_up_interruptible_all(&wq->wait);
1226 static bool subflow_is_done(const struct sock *sk)
1228 return sk->sk_shutdown & RCV_SHUTDOWN || sk->sk_state == TCP_CLOSE;
1231 static void subflow_state_change(struct sock *sk)
1233 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1234 struct sock *parent = subflow->conn;
1236 __subflow_state_change(sk);
1238 if (subflow_simultaneous_connect(sk)) {
1239 mptcp_do_fallback(sk);
1240 mptcp_rcv_space_init(mptcp_sk(parent), sk);
1241 pr_fallback(mptcp_sk(parent));
1242 subflow->conn_finished = 1;
1243 if (inet_sk_state_load(parent) == TCP_SYN_SENT) {
1244 inet_sk_state_store(parent, TCP_ESTABLISHED);
1245 parent->sk_state_change(parent);
1249 /* as recvmsg() does not acquire the subflow socket for ssk selection
1250 * a fin packet carrying a DSS can be unnoticed if we don't trigger
1251 * the data available machinery here.
1253 if (mptcp_subflow_data_available(sk))
1254 mptcp_data_ready(parent, sk);
1256 if (__mptcp_check_fallback(mptcp_sk(parent)) &&
1257 !(parent->sk_shutdown & RCV_SHUTDOWN) &&
1258 !subflow->rx_eof && subflow_is_done(sk)) {
1259 subflow->rx_eof = 1;
1260 mptcp_subflow_eof(parent);
1264 static int subflow_ulp_init(struct sock *sk)
1266 struct inet_connection_sock *icsk = inet_csk(sk);
1267 struct mptcp_subflow_context *ctx;
1268 struct tcp_sock *tp = tcp_sk(sk);
1271 /* disallow attaching ULP to a socket unless it has been
1272 * created with sock_create_kern()
1274 if (!sk->sk_kern_sock) {
1279 ctx = subflow_create_ctx(sk, GFP_KERNEL);
1285 pr_debug("subflow=%p, family=%d", ctx, sk->sk_family);
1288 ctx->icsk_af_ops = icsk->icsk_af_ops;
1289 icsk->icsk_af_ops = subflow_default_af_ops(sk);
1290 ctx->tcp_data_ready = sk->sk_data_ready;
1291 ctx->tcp_state_change = sk->sk_state_change;
1292 ctx->tcp_write_space = sk->sk_write_space;
1293 sk->sk_data_ready = subflow_data_ready;
1294 sk->sk_write_space = subflow_write_space;
1295 sk->sk_state_change = subflow_state_change;
1300 static void subflow_ulp_release(struct sock *sk)
1302 struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(sk);
1308 sock_put(ctx->conn);
1310 kfree_rcu(ctx, rcu);
1313 static void subflow_ulp_clone(const struct request_sock *req,
1315 const gfp_t priority)
1317 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
1318 struct mptcp_subflow_context *old_ctx = mptcp_subflow_ctx(newsk);
1319 struct mptcp_subflow_context *new_ctx;
1321 if (!tcp_rsk(req)->is_mptcp ||
1322 (!subflow_req->mp_capable && !subflow_req->mp_join)) {
1323 subflow_ulp_fallback(newsk, old_ctx);
1327 new_ctx = subflow_create_ctx(newsk, priority);
1329 subflow_ulp_fallback(newsk, old_ctx);
1333 new_ctx->conn_finished = 1;
1334 new_ctx->icsk_af_ops = old_ctx->icsk_af_ops;
1335 new_ctx->tcp_data_ready = old_ctx->tcp_data_ready;
1336 new_ctx->tcp_state_change = old_ctx->tcp_state_change;
1337 new_ctx->tcp_write_space = old_ctx->tcp_write_space;
1338 new_ctx->rel_write_seq = 1;
1339 new_ctx->tcp_sock = newsk;
1341 if (subflow_req->mp_capable) {
1342 /* see comments in subflow_syn_recv_sock(), MPTCP connection
1343 * is fully established only after we receive the remote key
1345 new_ctx->mp_capable = 1;
1346 new_ctx->local_key = subflow_req->local_key;
1347 new_ctx->token = subflow_req->token;
1348 new_ctx->ssn_offset = subflow_req->ssn_offset;
1349 new_ctx->idsn = subflow_req->idsn;
1350 } else if (subflow_req->mp_join) {
1351 new_ctx->ssn_offset = subflow_req->ssn_offset;
1352 new_ctx->mp_join = 1;
1353 new_ctx->fully_established = 1;
1354 new_ctx->backup = subflow_req->backup;
1355 new_ctx->local_id = subflow_req->local_id;
1356 new_ctx->remote_id = subflow_req->remote_id;
1357 new_ctx->token = subflow_req->token;
1358 new_ctx->thmac = subflow_req->thmac;
1362 static struct tcp_ulp_ops subflow_ulp_ops __read_mostly = {
1364 .owner = THIS_MODULE,
1365 .init = subflow_ulp_init,
1366 .release = subflow_ulp_release,
1367 .clone = subflow_ulp_clone,
1370 static int subflow_ops_init(struct request_sock_ops *subflow_ops)
1372 subflow_ops->obj_size = sizeof(struct mptcp_subflow_request_sock);
1373 subflow_ops->slab_name = "request_sock_subflow";
1375 subflow_ops->slab = kmem_cache_create(subflow_ops->slab_name,
1376 subflow_ops->obj_size, 0,
1378 SLAB_TYPESAFE_BY_RCU,
1380 if (!subflow_ops->slab)
1383 subflow_ops->destructor = subflow_req_destructor;
1388 void __init mptcp_subflow_init(void)
1390 mptcp_subflow_request_sock_ops = tcp_request_sock_ops;
1391 if (subflow_ops_init(&mptcp_subflow_request_sock_ops) != 0)
1392 panic("MPTCP: failed to init subflow request sock ops\n");
1394 subflow_request_sock_ipv4_ops = tcp_request_sock_ipv4_ops;
1395 subflow_request_sock_ipv4_ops.init_req = subflow_v4_init_req;
1397 subflow_specific = ipv4_specific;
1398 subflow_specific.conn_request = subflow_v4_conn_request;
1399 subflow_specific.syn_recv_sock = subflow_syn_recv_sock;
1400 subflow_specific.sk_rx_dst_set = subflow_finish_connect;
1402 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1403 subflow_request_sock_ipv6_ops = tcp_request_sock_ipv6_ops;
1404 subflow_request_sock_ipv6_ops.init_req = subflow_v6_init_req;
1406 subflow_v6_specific = ipv6_specific;
1407 subflow_v6_specific.conn_request = subflow_v6_conn_request;
1408 subflow_v6_specific.syn_recv_sock = subflow_syn_recv_sock;
1409 subflow_v6_specific.sk_rx_dst_set = subflow_finish_connect;
1411 subflow_v6m_specific = subflow_v6_specific;
1412 subflow_v6m_specific.queue_xmit = ipv4_specific.queue_xmit;
1413 subflow_v6m_specific.send_check = ipv4_specific.send_check;
1414 subflow_v6m_specific.net_header_len = ipv4_specific.net_header_len;
1415 subflow_v6m_specific.mtu_reduced = ipv4_specific.mtu_reduced;
1416 subflow_v6m_specific.net_frag_header_len = 0;
1419 mptcp_diag_subflow_init(&subflow_ulp_ops);
1421 if (tcp_register_ulp(&subflow_ulp_ops) != 0)
1422 panic("MPTCP: failed to register subflows to ULP\n");