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(sock_net(req_to_sk(req)), 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 if (!mptcp_finish_join(sk))
344 subflow_generate_hmac(subflow->local_key, subflow->remote_key,
345 subflow->local_nonce,
346 subflow->remote_nonce,
348 memcpy(subflow->hmac, hmac, MPTCPOPT_HMAC_LEN);
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 if (ipv6_addr_v4mapped(&ipv6_hdr(skb)->saddr)) {
402 __IP6_INC_STATS(sock_net(sk), NULL, IPSTATS_MIB_INHDRERRORS);
406 return tcp_conn_request(&mptcp_subflow_request_sock_ops,
407 &subflow_request_sock_ipv6_ops, sk, skb);
411 return 0; /* don't send reset */
415 /* validate hmac received in third ACK */
416 static bool subflow_hmac_valid(const struct request_sock *req,
417 const struct mptcp_options_received *mp_opt)
419 const struct mptcp_subflow_request_sock *subflow_req;
420 u8 hmac[SHA256_DIGEST_SIZE];
421 struct mptcp_sock *msk;
423 subflow_req = mptcp_subflow_rsk(req);
424 msk = subflow_req->msk;
428 subflow_generate_hmac(msk->remote_key, msk->local_key,
429 subflow_req->remote_nonce,
430 subflow_req->local_nonce, hmac);
432 return !crypto_memneq(hmac, mp_opt->hmac, MPTCPOPT_HMAC_LEN);
435 static void mptcp_sock_destruct(struct sock *sk)
437 /* if new mptcp socket isn't accepted, it is free'd
438 * from the tcp listener sockets request queue, linked
439 * from req->sk. The tcp socket is released.
440 * This calls the ULP release function which will
441 * also remove the mptcp socket, via
442 * sock_put(ctx->conn).
444 * Problem is that the mptcp socket will be in
445 * ESTABLISHED state and will not have the SOCK_DEAD flag.
446 * Both result in warnings from inet_sock_destruct.
448 if ((1 << sk->sk_state) & (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) {
449 sk->sk_state = TCP_CLOSE;
450 WARN_ON_ONCE(sk->sk_socket);
454 mptcp_destroy_common(mptcp_sk(sk));
455 inet_sock_destruct(sk);
458 static void mptcp_force_close(struct sock *sk)
460 inet_sk_state_store(sk, TCP_CLOSE);
461 sk_common_release(sk);
464 static void subflow_ulp_fallback(struct sock *sk,
465 struct mptcp_subflow_context *old_ctx)
467 struct inet_connection_sock *icsk = inet_csk(sk);
469 mptcp_subflow_tcp_fallback(sk, old_ctx);
470 icsk->icsk_ulp_ops = NULL;
471 rcu_assign_pointer(icsk->icsk_ulp_data, NULL);
472 tcp_sk(sk)->is_mptcp = 0;
475 static void subflow_drop_ctx(struct sock *ssk)
477 struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(ssk);
482 subflow_ulp_fallback(ssk, ctx);
489 void mptcp_subflow_fully_established(struct mptcp_subflow_context *subflow,
490 struct mptcp_options_received *mp_opt)
492 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
494 subflow->remote_key = mp_opt->sndr_key;
495 subflow->fully_established = 1;
496 subflow->can_ack = 1;
497 WRITE_ONCE(msk->fully_established, true);
500 static struct sock *subflow_syn_recv_sock(const struct sock *sk,
502 struct request_sock *req,
503 struct dst_entry *dst,
504 struct request_sock *req_unhash,
507 struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk);
508 struct mptcp_subflow_request_sock *subflow_req;
509 struct mptcp_options_received mp_opt;
510 bool fallback, fallback_is_fatal;
511 struct sock *new_msk = NULL;
514 pr_debug("listener=%p, req=%p, conn=%p", listener, req, listener->conn);
516 /* After child creation we must look for 'mp_capable' even when options
519 mp_opt.mp_capable = 0;
521 /* hopefully temporary handling for MP_JOIN+syncookie */
522 subflow_req = mptcp_subflow_rsk(req);
523 fallback_is_fatal = tcp_rsk(req)->is_mptcp && subflow_req->mp_join;
524 fallback = !tcp_rsk(req)->is_mptcp;
528 /* if the sk is MP_CAPABLE, we try to fetch the client key */
529 if (subflow_req->mp_capable) {
530 /* we can receive and accept an in-window, out-of-order pkt,
531 * which may not carry the MP_CAPABLE opt even on mptcp enabled
532 * paths: always try to extract the peer key, and fallback
533 * for packets missing it.
534 * Even OoO DSS packets coming legitly after dropped or
535 * reordered MPC will cause fallback, but we don't have other
538 mptcp_get_options(skb, &mp_opt);
539 if (!mp_opt.mp_capable) {
544 new_msk = mptcp_sk_clone(listener->conn, &mp_opt, req);
547 } else if (subflow_req->mp_join) {
548 mptcp_get_options(skb, &mp_opt);
549 if (!mp_opt.mp_join || !subflow_hmac_valid(req, &mp_opt) ||
550 !mptcp_can_accept_new_subflow(subflow_req->msk)) {
551 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKMAC);
557 child = listener->icsk_af_ops->syn_recv_sock(sk, skb, req, dst,
558 req_unhash, own_req);
560 if (child && *own_req) {
561 struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(child);
563 tcp_rsk(req)->drop_req = false;
565 /* we need to fallback on ctx allocation failure and on pre-reqs
566 * checking above. In the latter scenario we additionally need
567 * to reset the context to non MPTCP status.
569 if (!ctx || fallback) {
570 if (fallback_is_fatal)
573 subflow_drop_ctx(child);
577 if (ctx->mp_capable) {
578 /* this can't race with mptcp_close(), as the msk is
579 * not yet exposted to user-space
581 inet_sk_state_store((void *)new_msk, TCP_ESTABLISHED);
583 /* new mpc subflow takes ownership of the newly
584 * created mptcp socket
586 new_msk->sk_destruct = mptcp_sock_destruct;
587 mptcp_pm_new_connection(mptcp_sk(new_msk), 1);
588 mptcp_token_accept(subflow_req, mptcp_sk(new_msk));
592 /* with OoO packets we can reach here without ingress
595 if (mp_opt.mp_capable)
596 mptcp_subflow_fully_established(ctx, &mp_opt);
597 } else if (ctx->mp_join) {
598 struct mptcp_sock *owner;
600 owner = subflow_req->msk;
604 /* move the msk reference ownership to the subflow */
605 subflow_req->msk = NULL;
606 ctx->conn = (struct sock *)owner;
607 if (!mptcp_finish_join(child))
610 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKRX);
611 tcp_rsk(req)->drop_req = true;
616 /* dispose of the left over mptcp master, if any */
617 if (unlikely(new_msk))
618 mptcp_force_close(new_msk);
620 /* check for expected invariant - should never trigger, just help
621 * catching eariler subtle bugs
623 WARN_ON_ONCE(child && *own_req && tcp_sk(child)->is_mptcp &&
624 (!mptcp_subflow_ctx(child) ||
625 !mptcp_subflow_ctx(child)->conn));
629 subflow_drop_ctx(child);
630 tcp_rsk(req)->drop_req = true;
631 inet_csk_prepare_for_destroy_sock(child);
633 req->rsk_ops->send_reset(sk, skb);
635 /* The last child reference will be released by the caller */
639 static struct inet_connection_sock_af_ops subflow_specific;
641 enum mapping_status {
649 static u64 expand_seq(u64 old_seq, u16 old_data_len, u64 seq)
651 if ((u32)seq == (u32)old_seq)
654 /* Assume map covers data not mapped yet. */
655 return seq | ((old_seq + old_data_len + 1) & GENMASK_ULL(63, 32));
658 static void dbg_bad_map(struct mptcp_subflow_context *subflow, u32 ssn)
660 pr_debug("Bad mapping: ssn=%d map_seq=%d map_data_len=%d",
661 ssn, subflow->map_subflow_seq, subflow->map_data_len);
664 static bool skb_is_fully_mapped(struct sock *ssk, struct sk_buff *skb)
666 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
667 unsigned int skb_consumed;
669 skb_consumed = tcp_sk(ssk)->copied_seq - TCP_SKB_CB(skb)->seq;
670 if (WARN_ON_ONCE(skb_consumed >= skb->len))
673 return skb->len - skb_consumed <= subflow->map_data_len -
674 mptcp_subflow_get_map_offset(subflow);
677 static bool validate_mapping(struct sock *ssk, struct sk_buff *skb)
679 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
680 u32 ssn = tcp_sk(ssk)->copied_seq - subflow->ssn_offset;
682 if (unlikely(before(ssn, subflow->map_subflow_seq))) {
683 /* Mapping covers data later in the subflow stream,
684 * currently unsupported.
686 dbg_bad_map(subflow, ssn);
689 if (unlikely(!before(ssn, subflow->map_subflow_seq +
690 subflow->map_data_len))) {
691 /* Mapping does covers past subflow data, invalid */
692 dbg_bad_map(subflow, ssn);
698 static enum mapping_status get_mapping_status(struct sock *ssk,
699 struct mptcp_sock *msk)
701 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
702 struct mptcp_ext *mpext;
707 skb = skb_peek(&ssk->sk_receive_queue);
709 return MAPPING_EMPTY;
711 if (mptcp_check_fallback(ssk))
712 return MAPPING_DUMMY;
714 mpext = mptcp_get_ext(skb);
715 if (!mpext || !mpext->use_map) {
716 if (!subflow->map_valid && !skb->len) {
717 /* the TCP stack deliver 0 len FIN pkt to the receive
718 * queue, that is the only 0len pkts ever expected here,
719 * and we can admit no mapping only for 0 len pkts
721 if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN))
722 WARN_ONCE(1, "0len seq %d:%d flags %x",
723 TCP_SKB_CB(skb)->seq,
724 TCP_SKB_CB(skb)->end_seq,
725 TCP_SKB_CB(skb)->tcp_flags);
726 sk_eat_skb(ssk, skb);
727 return MAPPING_EMPTY;
730 if (!subflow->map_valid)
731 return MAPPING_INVALID;
736 pr_debug("seq=%llu is64=%d ssn=%u data_len=%u data_fin=%d",
737 mpext->data_seq, mpext->dsn64, mpext->subflow_seq,
738 mpext->data_len, mpext->data_fin);
740 data_len = mpext->data_len;
742 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_INFINITEMAPRX);
743 return MAPPING_INVALID;
746 if (mpext->data_fin == 1) {
748 bool updated = mptcp_update_rcv_data_fin(msk, mpext->data_seq,
750 pr_debug("DATA_FIN with no payload seq=%llu", mpext->data_seq);
751 if (subflow->map_valid) {
752 /* A DATA_FIN might arrive in a DSS
753 * option before the previous mapping
754 * has been fully consumed. Continue
755 * handling the existing mapping.
757 skb_ext_del(skb, SKB_EXT_MPTCP);
760 if (updated && schedule_work(&msk->work))
761 sock_hold((struct sock *)msk);
763 return MAPPING_DATA_FIN;
766 u64 data_fin_seq = mpext->data_seq + data_len - 1;
768 /* If mpext->data_seq is a 32-bit value, data_fin_seq
769 * must also be limited to 32 bits.
772 data_fin_seq &= GENMASK_ULL(31, 0);
774 mptcp_update_rcv_data_fin(msk, data_fin_seq, mpext->dsn64);
775 pr_debug("DATA_FIN with mapping seq=%llu dsn64=%d",
776 data_fin_seq, mpext->dsn64);
779 /* Adjust for DATA_FIN using 1 byte of sequence space */
784 map_seq = expand_seq(subflow->map_seq, subflow->map_data_len,
786 pr_debug("expanded seq=%llu", subflow->map_seq);
788 map_seq = mpext->data_seq;
790 WRITE_ONCE(mptcp_sk(subflow->conn)->use_64bit_ack, !!mpext->dsn64);
792 if (subflow->map_valid) {
793 /* Allow replacing only with an identical map */
794 if (subflow->map_seq == map_seq &&
795 subflow->map_subflow_seq == mpext->subflow_seq &&
796 subflow->map_data_len == data_len) {
797 skb_ext_del(skb, SKB_EXT_MPTCP);
801 /* If this skb data are fully covered by the current mapping,
802 * the new map would need caching, which is not supported
804 if (skb_is_fully_mapped(ssk, skb)) {
805 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DSSNOMATCH);
806 return MAPPING_INVALID;
809 /* will validate the next map after consuming the current one */
813 subflow->map_seq = map_seq;
814 subflow->map_subflow_seq = mpext->subflow_seq;
815 subflow->map_data_len = data_len;
816 subflow->map_valid = 1;
817 subflow->mpc_map = mpext->mpc_map;
818 pr_debug("new map seq=%llu subflow_seq=%u data_len=%u",
819 subflow->map_seq, subflow->map_subflow_seq,
820 subflow->map_data_len);
823 /* we revalidate valid mapping on new skb, because we must ensure
824 * the current skb is completely covered by the available mapping
826 if (!validate_mapping(ssk, skb))
827 return MAPPING_INVALID;
829 skb_ext_del(skb, SKB_EXT_MPTCP);
833 static void mptcp_subflow_discard_data(struct sock *ssk, struct sk_buff *skb,
836 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
837 bool fin = TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN;
840 incr = limit >= skb->len ? skb->len + fin : limit;
842 pr_debug("discarding=%d len=%d seq=%d", incr, skb->len,
843 subflow->map_subflow_seq);
844 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DUPDATA);
845 tcp_sk(ssk)->copied_seq += incr;
846 if (!before(tcp_sk(ssk)->copied_seq, TCP_SKB_CB(skb)->end_seq))
847 sk_eat_skb(ssk, skb);
848 if (mptcp_subflow_get_map_offset(subflow) >= subflow->map_data_len)
849 subflow->map_valid = 0;
851 tcp_cleanup_rbuf(ssk, incr);
854 static bool subflow_check_data_avail(struct sock *ssk)
856 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
857 enum mapping_status status;
858 struct mptcp_sock *msk;
861 pr_debug("msk=%p ssk=%p data_avail=%d skb=%p", subflow->conn, ssk,
862 subflow->data_avail, skb_peek(&ssk->sk_receive_queue));
863 if (!skb_peek(&ssk->sk_receive_queue))
864 subflow->data_avail = 0;
865 if (subflow->data_avail)
868 msk = mptcp_sk(subflow->conn);
873 status = get_mapping_status(ssk, msk);
874 pr_debug("msk=%p ssk=%p status=%d", msk, ssk, status);
875 if (status == MAPPING_INVALID) {
876 ssk->sk_err = EBADMSG;
879 if (status == MAPPING_DUMMY) {
880 __mptcp_do_fallback(msk);
881 skb = skb_peek(&ssk->sk_receive_queue);
882 subflow->map_valid = 1;
883 subflow->map_seq = READ_ONCE(msk->ack_seq);
884 subflow->map_data_len = skb->len;
885 subflow->map_subflow_seq = tcp_sk(ssk)->copied_seq -
887 subflow->data_avail = MPTCP_SUBFLOW_DATA_AVAIL;
891 if (status != MAPPING_OK)
894 skb = skb_peek(&ssk->sk_receive_queue);
895 if (WARN_ON_ONCE(!skb))
898 /* if msk lacks the remote key, this subflow must provide an
899 * MP_CAPABLE-based mapping
901 if (unlikely(!READ_ONCE(msk->can_ack))) {
902 if (!subflow->mpc_map) {
903 ssk->sk_err = EBADMSG;
906 WRITE_ONCE(msk->remote_key, subflow->remote_key);
907 WRITE_ONCE(msk->ack_seq, subflow->map_seq);
908 WRITE_ONCE(msk->can_ack, true);
911 old_ack = READ_ONCE(msk->ack_seq);
912 ack_seq = mptcp_subflow_get_mapped_dsn(subflow);
913 pr_debug("msk ack_seq=%llx subflow ack_seq=%llx", old_ack,
915 if (ack_seq == old_ack) {
916 subflow->data_avail = MPTCP_SUBFLOW_DATA_AVAIL;
918 } else if (after64(ack_seq, old_ack)) {
919 subflow->data_avail = MPTCP_SUBFLOW_OOO_DATA;
923 /* only accept in-sequence mapping. Old values are spurious
926 mptcp_subflow_discard_data(ssk, skb, old_ack - ack_seq);
931 /* fatal protocol error, close the socket */
932 /* This barrier is coupled with smp_rmb() in tcp_poll() */
934 ssk->sk_error_report(ssk);
935 tcp_set_state(ssk, TCP_CLOSE);
936 tcp_send_active_reset(ssk, GFP_ATOMIC);
937 subflow->data_avail = 0;
941 bool mptcp_subflow_data_available(struct sock *sk)
943 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
945 /* check if current mapping is still valid */
946 if (subflow->map_valid &&
947 mptcp_subflow_get_map_offset(subflow) >= subflow->map_data_len) {
948 subflow->map_valid = 0;
949 subflow->data_avail = 0;
951 pr_debug("Done with mapping: seq=%u data_len=%u",
952 subflow->map_subflow_seq,
953 subflow->map_data_len);
956 return subflow_check_data_avail(sk);
959 /* If ssk has an mptcp parent socket, use the mptcp rcvbuf occupancy,
962 * In mptcp, rwin is about the mptcp-level connection data.
964 * Data that is still on the ssk rx queue can thus be ignored,
965 * as far as mptcp peer is concerened that data is still inflight.
966 * DSS ACK is updated when skb is moved to the mptcp rx queue.
968 void mptcp_space(const struct sock *ssk, int *space, int *full_space)
970 const struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
971 const struct sock *sk = subflow->conn;
973 *space = tcp_space(sk);
974 *full_space = tcp_full_space(sk);
977 static void subflow_data_ready(struct sock *sk)
979 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
980 u16 state = 1 << inet_sk_state_load(sk);
981 struct sock *parent = subflow->conn;
982 struct mptcp_sock *msk;
984 msk = mptcp_sk(parent);
985 if (state & TCPF_LISTEN) {
986 /* MPJ subflow are removed from accept queue before reaching here,
987 * avoid stray wakeups
989 if (reqsk_queue_empty(&inet_csk(sk)->icsk_accept_queue))
992 set_bit(MPTCP_DATA_READY, &msk->flags);
993 parent->sk_data_ready(parent);
997 WARN_ON_ONCE(!__mptcp_check_fallback(msk) && !subflow->mp_capable &&
998 !subflow->mp_join && !(state & TCPF_CLOSE));
1000 if (mptcp_subflow_data_available(sk))
1001 mptcp_data_ready(parent, sk);
1004 static void subflow_write_space(struct sock *sk)
1006 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1007 struct sock *parent = subflow->conn;
1009 if (!sk_stream_is_writeable(sk))
1012 if (sk_stream_is_writeable(parent)) {
1013 set_bit(MPTCP_SEND_SPACE, &mptcp_sk(parent)->flags);
1014 smp_mb__after_atomic();
1015 /* set SEND_SPACE before sk_stream_write_space clears NOSPACE */
1016 sk_stream_write_space(parent);
1020 static struct inet_connection_sock_af_ops *
1021 subflow_default_af_ops(struct sock *sk)
1023 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1024 if (sk->sk_family == AF_INET6)
1025 return &subflow_v6_specific;
1027 return &subflow_specific;
1030 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1031 void mptcpv6_handle_mapped(struct sock *sk, bool mapped)
1033 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1034 struct inet_connection_sock *icsk = inet_csk(sk);
1035 struct inet_connection_sock_af_ops *target;
1037 target = mapped ? &subflow_v6m_specific : subflow_default_af_ops(sk);
1039 pr_debug("subflow=%p family=%d ops=%p target=%p mapped=%d",
1040 subflow, sk->sk_family, icsk->icsk_af_ops, target, mapped);
1042 if (likely(icsk->icsk_af_ops == target))
1045 subflow->icsk_af_ops = icsk->icsk_af_ops;
1046 icsk->icsk_af_ops = target;
1050 static void mptcp_info2sockaddr(const struct mptcp_addr_info *info,
1051 struct sockaddr_storage *addr)
1053 memset(addr, 0, sizeof(*addr));
1054 addr->ss_family = info->family;
1055 if (addr->ss_family == AF_INET) {
1056 struct sockaddr_in *in_addr = (struct sockaddr_in *)addr;
1058 in_addr->sin_addr = info->addr;
1059 in_addr->sin_port = info->port;
1061 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1062 else if (addr->ss_family == AF_INET6) {
1063 struct sockaddr_in6 *in6_addr = (struct sockaddr_in6 *)addr;
1065 in6_addr->sin6_addr = info->addr6;
1066 in6_addr->sin6_port = info->port;
1071 int __mptcp_subflow_connect(struct sock *sk, const struct mptcp_addr_info *loc,
1072 const struct mptcp_addr_info *remote)
1074 struct mptcp_sock *msk = mptcp_sk(sk);
1075 struct mptcp_subflow_context *subflow;
1076 struct sockaddr_storage addr;
1077 int remote_id = remote->id;
1078 int local_id = loc->id;
1085 if (!mptcp_is_fully_established(sk))
1088 err = mptcp_subflow_create_socket(sk, &sf);
1093 subflow = mptcp_subflow_ctx(ssk);
1095 get_random_bytes(&subflow->local_nonce, sizeof(u32));
1096 } while (!subflow->local_nonce);
1099 err = mptcp_pm_get_local_id(msk, (struct sock_common *)ssk);
1106 subflow->remote_key = msk->remote_key;
1107 subflow->local_key = msk->local_key;
1108 subflow->token = msk->token;
1109 mptcp_info2sockaddr(loc, &addr);
1111 addrlen = sizeof(struct sockaddr_in);
1112 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1113 if (loc->family == AF_INET6)
1114 addrlen = sizeof(struct sockaddr_in6);
1116 ssk->sk_bound_dev_if = loc->ifindex;
1117 err = kernel_bind(sf, (struct sockaddr *)&addr, addrlen);
1121 mptcp_crypto_key_sha(subflow->remote_key, &remote_token, NULL);
1122 pr_debug("msk=%p remote_token=%u local_id=%d remote_id=%d", msk,
1123 remote_token, local_id, remote_id);
1124 subflow->remote_token = remote_token;
1125 subflow->local_id = local_id;
1126 subflow->remote_id = remote_id;
1127 subflow->request_join = 1;
1128 subflow->request_bkup = !!(loc->flags & MPTCP_PM_ADDR_FLAG_BACKUP);
1129 mptcp_info2sockaddr(remote, &addr);
1131 err = kernel_connect(sf, (struct sockaddr *)&addr, addrlen, O_NONBLOCK);
1132 if (err && err != -EINPROGRESS)
1135 spin_lock_bh(&msk->join_list_lock);
1136 list_add_tail(&subflow->node, &msk->join_list);
1137 spin_unlock_bh(&msk->join_list_lock);
1146 int mptcp_subflow_create_socket(struct sock *sk, struct socket **new_sock)
1148 struct mptcp_subflow_context *subflow;
1149 struct net *net = sock_net(sk);
1153 /* un-accepted server sockets can reach here - on bad configuration
1154 * bail early to avoid greater trouble later
1156 if (unlikely(!sk->sk_socket))
1159 err = sock_create_kern(net, sk->sk_family, SOCK_STREAM, IPPROTO_TCP,
1166 /* kernel sockets do not by default acquire net ref, but TCP timer
1169 sf->sk->sk_net_refcnt = 1;
1171 #ifdef CONFIG_PROC_FS
1172 this_cpu_add(*net->core.sock_inuse, 1);
1174 err = tcp_set_ulp(sf->sk, "mptcp");
1175 release_sock(sf->sk);
1182 /* the newly created socket really belongs to the owning MPTCP master
1183 * socket, even if for additional subflows the allocation is performed
1184 * by a kernel workqueue. Adjust inode references, so that the
1185 * procfs/diag interaces really show this one belonging to the correct
1188 SOCK_INODE(sf)->i_ino = SOCK_INODE(sk->sk_socket)->i_ino;
1189 SOCK_INODE(sf)->i_uid = SOCK_INODE(sk->sk_socket)->i_uid;
1190 SOCK_INODE(sf)->i_gid = SOCK_INODE(sk->sk_socket)->i_gid;
1192 subflow = mptcp_subflow_ctx(sf->sk);
1193 pr_debug("subflow=%p", subflow);
1202 static struct mptcp_subflow_context *subflow_create_ctx(struct sock *sk,
1205 struct inet_connection_sock *icsk = inet_csk(sk);
1206 struct mptcp_subflow_context *ctx;
1208 ctx = kzalloc(sizeof(*ctx), priority);
1212 rcu_assign_pointer(icsk->icsk_ulp_data, ctx);
1213 INIT_LIST_HEAD(&ctx->node);
1215 pr_debug("subflow=%p", ctx);
1222 static void __subflow_state_change(struct sock *sk)
1224 struct socket_wq *wq;
1227 wq = rcu_dereference(sk->sk_wq);
1228 if (skwq_has_sleeper(wq))
1229 wake_up_interruptible_all(&wq->wait);
1233 static bool subflow_is_done(const struct sock *sk)
1235 return sk->sk_shutdown & RCV_SHUTDOWN || sk->sk_state == TCP_CLOSE;
1238 static void subflow_state_change(struct sock *sk)
1240 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1241 struct sock *parent = subflow->conn;
1243 __subflow_state_change(sk);
1245 if (subflow_simultaneous_connect(sk)) {
1246 mptcp_do_fallback(sk);
1247 mptcp_rcv_space_init(mptcp_sk(parent), sk);
1248 pr_fallback(mptcp_sk(parent));
1249 subflow->conn_finished = 1;
1250 if (inet_sk_state_load(parent) == TCP_SYN_SENT) {
1251 inet_sk_state_store(parent, TCP_ESTABLISHED);
1252 parent->sk_state_change(parent);
1256 /* as recvmsg() does not acquire the subflow socket for ssk selection
1257 * a fin packet carrying a DSS can be unnoticed if we don't trigger
1258 * the data available machinery here.
1260 if (mptcp_subflow_data_available(sk))
1261 mptcp_data_ready(parent, sk);
1263 if (__mptcp_check_fallback(mptcp_sk(parent)) &&
1264 !(parent->sk_shutdown & RCV_SHUTDOWN) &&
1265 !subflow->rx_eof && subflow_is_done(sk)) {
1266 subflow->rx_eof = 1;
1267 mptcp_subflow_eof(parent);
1271 static int subflow_ulp_init(struct sock *sk)
1273 struct inet_connection_sock *icsk = inet_csk(sk);
1274 struct mptcp_subflow_context *ctx;
1275 struct tcp_sock *tp = tcp_sk(sk);
1278 /* disallow attaching ULP to a socket unless it has been
1279 * created with sock_create_kern()
1281 if (!sk->sk_kern_sock) {
1286 ctx = subflow_create_ctx(sk, GFP_KERNEL);
1292 pr_debug("subflow=%p, family=%d", ctx, sk->sk_family);
1295 ctx->icsk_af_ops = icsk->icsk_af_ops;
1296 icsk->icsk_af_ops = subflow_default_af_ops(sk);
1297 ctx->tcp_data_ready = sk->sk_data_ready;
1298 ctx->tcp_state_change = sk->sk_state_change;
1299 ctx->tcp_write_space = sk->sk_write_space;
1300 sk->sk_data_ready = subflow_data_ready;
1301 sk->sk_write_space = subflow_write_space;
1302 sk->sk_state_change = subflow_state_change;
1307 static void subflow_ulp_release(struct sock *sk)
1309 struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(sk);
1315 sock_put(ctx->conn);
1317 kfree_rcu(ctx, rcu);
1320 static void subflow_ulp_clone(const struct request_sock *req,
1322 const gfp_t priority)
1324 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
1325 struct mptcp_subflow_context *old_ctx = mptcp_subflow_ctx(newsk);
1326 struct mptcp_subflow_context *new_ctx;
1328 if (!tcp_rsk(req)->is_mptcp ||
1329 (!subflow_req->mp_capable && !subflow_req->mp_join)) {
1330 subflow_ulp_fallback(newsk, old_ctx);
1334 new_ctx = subflow_create_ctx(newsk, priority);
1336 subflow_ulp_fallback(newsk, old_ctx);
1340 new_ctx->conn_finished = 1;
1341 new_ctx->icsk_af_ops = old_ctx->icsk_af_ops;
1342 new_ctx->tcp_data_ready = old_ctx->tcp_data_ready;
1343 new_ctx->tcp_state_change = old_ctx->tcp_state_change;
1344 new_ctx->tcp_write_space = old_ctx->tcp_write_space;
1345 new_ctx->rel_write_seq = 1;
1346 new_ctx->tcp_sock = newsk;
1348 if (subflow_req->mp_capable) {
1349 /* see comments in subflow_syn_recv_sock(), MPTCP connection
1350 * is fully established only after we receive the remote key
1352 new_ctx->mp_capable = 1;
1353 new_ctx->local_key = subflow_req->local_key;
1354 new_ctx->token = subflow_req->token;
1355 new_ctx->ssn_offset = subflow_req->ssn_offset;
1356 new_ctx->idsn = subflow_req->idsn;
1357 } else if (subflow_req->mp_join) {
1358 new_ctx->ssn_offset = subflow_req->ssn_offset;
1359 new_ctx->mp_join = 1;
1360 new_ctx->fully_established = 1;
1361 new_ctx->backup = subflow_req->backup;
1362 new_ctx->local_id = subflow_req->local_id;
1363 new_ctx->remote_id = subflow_req->remote_id;
1364 new_ctx->token = subflow_req->token;
1365 new_ctx->thmac = subflow_req->thmac;
1369 static struct tcp_ulp_ops subflow_ulp_ops __read_mostly = {
1371 .owner = THIS_MODULE,
1372 .init = subflow_ulp_init,
1373 .release = subflow_ulp_release,
1374 .clone = subflow_ulp_clone,
1377 static int subflow_ops_init(struct request_sock_ops *subflow_ops)
1379 subflow_ops->obj_size = sizeof(struct mptcp_subflow_request_sock);
1380 subflow_ops->slab_name = "request_sock_subflow";
1382 subflow_ops->slab = kmem_cache_create(subflow_ops->slab_name,
1383 subflow_ops->obj_size, 0,
1385 SLAB_TYPESAFE_BY_RCU,
1387 if (!subflow_ops->slab)
1390 subflow_ops->destructor = subflow_req_destructor;
1395 void __init mptcp_subflow_init(void)
1397 mptcp_subflow_request_sock_ops = tcp_request_sock_ops;
1398 if (subflow_ops_init(&mptcp_subflow_request_sock_ops) != 0)
1399 panic("MPTCP: failed to init subflow request sock ops\n");
1401 subflow_request_sock_ipv4_ops = tcp_request_sock_ipv4_ops;
1402 subflow_request_sock_ipv4_ops.init_req = subflow_v4_init_req;
1404 subflow_specific = ipv4_specific;
1405 subflow_specific.conn_request = subflow_v4_conn_request;
1406 subflow_specific.syn_recv_sock = subflow_syn_recv_sock;
1407 subflow_specific.sk_rx_dst_set = subflow_finish_connect;
1409 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1410 subflow_request_sock_ipv6_ops = tcp_request_sock_ipv6_ops;
1411 subflow_request_sock_ipv6_ops.init_req = subflow_v6_init_req;
1413 subflow_v6_specific = ipv6_specific;
1414 subflow_v6_specific.conn_request = subflow_v6_conn_request;
1415 subflow_v6_specific.syn_recv_sock = subflow_syn_recv_sock;
1416 subflow_v6_specific.sk_rx_dst_set = subflow_finish_connect;
1418 subflow_v6m_specific = subflow_v6_specific;
1419 subflow_v6m_specific.queue_xmit = ipv4_specific.queue_xmit;
1420 subflow_v6m_specific.send_check = ipv4_specific.send_check;
1421 subflow_v6m_specific.net_header_len = ipv4_specific.net_header_len;
1422 subflow_v6m_specific.mtu_reduced = ipv4_specific.mtu_reduced;
1423 subflow_v6m_specific.net_frag_header_len = 0;
1426 mptcp_diag_subflow_init(&subflow_ulp_ops);
1428 if (tcp_register_ulp(&subflow_ulp_ops) != 0)
1429 panic("MPTCP: failed to register subflows to ULP\n");