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>
26 static void SUBFLOW_REQ_INC_STATS(struct request_sock *req,
27 enum linux_mptcp_mib_field field)
29 MPTCP_INC_STATS(sock_net(req_to_sk(req)), field);
32 static int subflow_rebuild_header(struct sock *sk)
34 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
35 int local_id, err = 0;
37 if (subflow->request_mptcp && !subflow->token) {
38 pr_debug("subflow=%p", sk);
39 err = mptcp_token_new_connect(sk);
40 } else if (subflow->request_join && !subflow->local_nonce) {
41 struct mptcp_sock *msk = (struct mptcp_sock *)subflow->conn;
43 pr_debug("subflow=%p", sk);
46 get_random_bytes(&subflow->local_nonce, sizeof(u32));
47 } while (!subflow->local_nonce);
49 if (subflow->local_id)
52 local_id = mptcp_pm_get_local_id(msk, (struct sock_common *)sk);
56 subflow->local_id = local_id;
63 return subflow->icsk_af_ops->rebuild_header(sk);
66 static void subflow_req_destructor(struct request_sock *req)
68 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
70 pr_debug("subflow_req=%p", subflow_req);
72 if (subflow_req->mp_capable)
73 mptcp_token_destroy_request(subflow_req->token);
74 tcp_request_sock_ops.destructor(req);
77 static void subflow_generate_hmac(u64 key1, u64 key2, u32 nonce1, u32 nonce2,
82 put_unaligned_be32(nonce1, &msg[0]);
83 put_unaligned_be32(nonce2, &msg[4]);
85 mptcp_crypto_hmac_sha(key1, key2, msg, 8, hmac);
88 /* validate received token and create truncated hmac and nonce for SYN-ACK */
89 static bool subflow_token_join_request(struct request_sock *req,
90 const struct sk_buff *skb)
92 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
93 u8 hmac[SHA256_DIGEST_SIZE];
94 struct mptcp_sock *msk;
97 msk = mptcp_token_get_sock(subflow_req->token);
99 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINNOTOKEN);
103 local_id = mptcp_pm_get_local_id(msk, (struct sock_common *)req);
105 sock_put((struct sock *)msk);
108 subflow_req->local_id = local_id;
110 get_random_bytes(&subflow_req->local_nonce, sizeof(u32));
112 subflow_generate_hmac(msk->local_key, msk->remote_key,
113 subflow_req->local_nonce,
114 subflow_req->remote_nonce, hmac);
116 subflow_req->thmac = get_unaligned_be64(hmac);
118 sock_put((struct sock *)msk);
122 static void subflow_init_req(struct request_sock *req,
123 const struct sock *sk_listener,
126 struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk_listener);
127 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
128 struct mptcp_options_received mp_opt;
130 pr_debug("subflow_req=%p, listener=%p", subflow_req, listener);
132 mptcp_get_options(skb, &mp_opt);
134 subflow_req->mp_capable = 0;
135 subflow_req->mp_join = 0;
137 #ifdef CONFIG_TCP_MD5SIG
138 /* no MPTCP if MD5SIG is enabled on this socket or we may run out of
141 if (rcu_access_pointer(tcp_sk(sk_listener)->md5sig_info))
145 if (mp_opt.mp_capable) {
146 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MPCAPABLEPASSIVE);
150 } else if (mp_opt.mp_join) {
151 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINSYNRX);
154 if (mp_opt.mp_capable && listener->request_mptcp) {
157 err = mptcp_token_new_request(req);
159 subflow_req->mp_capable = 1;
161 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq;
162 } else if (mp_opt.mp_join && listener->request_mptcp) {
163 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq;
164 subflow_req->mp_join = 1;
165 subflow_req->backup = mp_opt.backup;
166 subflow_req->remote_id = mp_opt.join_id;
167 subflow_req->token = mp_opt.token;
168 subflow_req->remote_nonce = mp_opt.nonce;
169 pr_debug("token=%u, remote_nonce=%u", subflow_req->token,
170 subflow_req->remote_nonce);
171 if (!subflow_token_join_request(req, skb)) {
172 subflow_req->mp_join = 0;
173 // @@ need to trigger RST
178 static void subflow_v4_init_req(struct request_sock *req,
179 const struct sock *sk_listener,
182 tcp_rsk(req)->is_mptcp = 1;
184 tcp_request_sock_ipv4_ops.init_req(req, sk_listener, skb);
186 subflow_init_req(req, sk_listener, skb);
189 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
190 static void subflow_v6_init_req(struct request_sock *req,
191 const struct sock *sk_listener,
194 tcp_rsk(req)->is_mptcp = 1;
196 tcp_request_sock_ipv6_ops.init_req(req, sk_listener, skb);
198 subflow_init_req(req, sk_listener, skb);
202 /* validate received truncated hmac and create hmac for third ACK */
203 static bool subflow_thmac_valid(struct mptcp_subflow_context *subflow)
205 u8 hmac[SHA256_DIGEST_SIZE];
208 subflow_generate_hmac(subflow->remote_key, subflow->local_key,
209 subflow->remote_nonce, subflow->local_nonce,
212 thmac = get_unaligned_be64(hmac);
213 pr_debug("subflow=%p, token=%u, thmac=%llu, subflow->thmac=%llu\n",
214 subflow, subflow->token,
215 (unsigned long long)thmac,
216 (unsigned long long)subflow->thmac);
218 return thmac == subflow->thmac;
221 static void subflow_finish_connect(struct sock *sk, const struct sk_buff *skb)
223 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
224 struct mptcp_options_received mp_opt;
225 struct sock *parent = subflow->conn;
226 struct tcp_sock *tp = tcp_sk(sk);
228 subflow->icsk_af_ops->sk_rx_dst_set(sk, skb);
230 if (inet_sk_state_load(parent) == TCP_SYN_SENT) {
231 inet_sk_state_store(parent, TCP_ESTABLISHED);
232 parent->sk_state_change(parent);
235 /* be sure no special action on any packet other than syn-ack */
236 if (subflow->conn_finished)
239 subflow->conn_finished = 1;
241 mptcp_get_options(skb, &mp_opt);
242 if (subflow->request_mptcp && mp_opt.mp_capable) {
243 subflow->mp_capable = 1;
244 subflow->can_ack = 1;
245 subflow->remote_key = mp_opt.sndr_key;
246 pr_debug("subflow=%p, remote_key=%llu", subflow,
247 subflow->remote_key);
248 } else if (subflow->request_join && mp_opt.mp_join) {
249 subflow->mp_join = 1;
250 subflow->thmac = mp_opt.thmac;
251 subflow->remote_nonce = mp_opt.nonce;
252 pr_debug("subflow=%p, thmac=%llu, remote_nonce=%u", subflow,
253 subflow->thmac, subflow->remote_nonce);
254 } else if (subflow->request_mptcp) {
261 if (subflow->mp_capable) {
262 pr_debug("subflow=%p, remote_key=%llu", mptcp_subflow_ctx(sk),
263 subflow->remote_key);
264 mptcp_finish_connect(sk);
267 pr_debug("synack seq=%u", TCP_SKB_CB(skb)->seq);
268 subflow->ssn_offset = TCP_SKB_CB(skb)->seq;
270 } else if (subflow->mp_join) {
271 u8 hmac[SHA256_DIGEST_SIZE];
273 pr_debug("subflow=%p, thmac=%llu, remote_nonce=%u",
274 subflow, subflow->thmac,
275 subflow->remote_nonce);
276 if (!subflow_thmac_valid(subflow)) {
277 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINACKMAC);
278 subflow->mp_join = 0;
282 subflow_generate_hmac(subflow->local_key, subflow->remote_key,
283 subflow->local_nonce,
284 subflow->remote_nonce,
287 memcpy(subflow->hmac, hmac, MPTCPOPT_HMAC_LEN);
290 subflow->ssn_offset = TCP_SKB_CB(skb)->seq;
292 if (!mptcp_finish_join(sk))
295 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINSYNACKRX);
298 tcp_send_active_reset(sk, GFP_ATOMIC);
303 static struct request_sock_ops subflow_request_sock_ops;
304 static struct tcp_request_sock_ops subflow_request_sock_ipv4_ops;
306 static int subflow_v4_conn_request(struct sock *sk, struct sk_buff *skb)
308 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
310 pr_debug("subflow=%p", subflow);
312 /* Never answer to SYNs sent to broadcast or multicast */
313 if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
316 return tcp_conn_request(&subflow_request_sock_ops,
317 &subflow_request_sock_ipv4_ops,
324 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
325 static struct tcp_request_sock_ops subflow_request_sock_ipv6_ops;
326 static struct inet_connection_sock_af_ops subflow_v6_specific;
327 static struct inet_connection_sock_af_ops subflow_v6m_specific;
329 static int subflow_v6_conn_request(struct sock *sk, struct sk_buff *skb)
331 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
333 pr_debug("subflow=%p", subflow);
335 if (skb->protocol == htons(ETH_P_IP))
336 return subflow_v4_conn_request(sk, skb);
338 if (!ipv6_unicast_destination(skb))
341 return tcp_conn_request(&subflow_request_sock_ops,
342 &subflow_request_sock_ipv6_ops, sk, skb);
346 return 0; /* don't send reset */
350 /* validate hmac received in third ACK */
351 static bool subflow_hmac_valid(const struct request_sock *req,
352 const struct mptcp_options_received *mp_opt)
354 const struct mptcp_subflow_request_sock *subflow_req;
355 u8 hmac[SHA256_DIGEST_SIZE];
356 struct mptcp_sock *msk;
359 subflow_req = mptcp_subflow_rsk(req);
360 msk = mptcp_token_get_sock(subflow_req->token);
364 subflow_generate_hmac(msk->remote_key, msk->local_key,
365 subflow_req->remote_nonce,
366 subflow_req->local_nonce, hmac);
369 if (crypto_memneq(hmac, mp_opt->hmac, MPTCPOPT_HMAC_LEN))
372 sock_put((struct sock *)msk);
376 static void mptcp_sock_destruct(struct sock *sk)
378 /* if new mptcp socket isn't accepted, it is free'd
379 * from the tcp listener sockets request queue, linked
380 * from req->sk. The tcp socket is released.
381 * This calls the ULP release function which will
382 * also remove the mptcp socket, via
383 * sock_put(ctx->conn).
385 * Problem is that the mptcp socket will not be in
386 * SYN_RECV state and doesn't have SOCK_DEAD flag.
387 * Both result in warnings from inet_sock_destruct.
390 if (sk->sk_state == TCP_SYN_RECV) {
391 sk->sk_state = TCP_CLOSE;
392 WARN_ON_ONCE(sk->sk_socket);
396 inet_sock_destruct(sk);
399 static void mptcp_force_close(struct sock *sk)
401 inet_sk_state_store(sk, TCP_CLOSE);
402 sk_common_release(sk);
405 static void subflow_ulp_fallback(struct sock *sk,
406 struct mptcp_subflow_context *old_ctx)
408 struct inet_connection_sock *icsk = inet_csk(sk);
410 mptcp_subflow_tcp_fallback(sk, old_ctx);
411 icsk->icsk_ulp_ops = NULL;
412 rcu_assign_pointer(icsk->icsk_ulp_data, NULL);
413 tcp_sk(sk)->is_mptcp = 0;
416 static void subflow_drop_ctx(struct sock *ssk)
418 struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(ssk);
423 subflow_ulp_fallback(ssk, ctx);
430 static struct sock *subflow_syn_recv_sock(const struct sock *sk,
432 struct request_sock *req,
433 struct dst_entry *dst,
434 struct request_sock *req_unhash,
437 struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk);
438 struct mptcp_subflow_request_sock *subflow_req;
439 struct mptcp_options_received mp_opt;
440 bool fallback_is_fatal = false;
441 struct sock *new_msk = NULL;
442 bool fallback = false;
445 pr_debug("listener=%p, req=%p, conn=%p", listener, req, listener->conn);
447 /* we need later a valid 'mp_capable' value even when options are not
450 mp_opt.mp_capable = 0;
451 if (tcp_rsk(req)->is_mptcp == 0)
454 /* if the sk is MP_CAPABLE, we try to fetch the client key */
455 subflow_req = mptcp_subflow_rsk(req);
456 if (subflow_req->mp_capable) {
457 if (TCP_SKB_CB(skb)->seq != subflow_req->ssn_offset + 1) {
458 /* here we can receive and accept an in-window,
459 * out-of-order pkt, which will not carry the MP_CAPABLE
460 * opt even on mptcp enabled paths
465 mptcp_get_options(skb, &mp_opt);
466 if (!mp_opt.mp_capable) {
472 new_msk = mptcp_sk_clone(listener->conn, &mp_opt, req);
475 } else if (subflow_req->mp_join) {
476 fallback_is_fatal = true;
477 mptcp_get_options(skb, &mp_opt);
478 if (!mp_opt.mp_join ||
479 !subflow_hmac_valid(req, &mp_opt)) {
480 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKMAC);
486 child = listener->icsk_af_ops->syn_recv_sock(sk, skb, req, dst,
487 req_unhash, own_req);
489 if (child && *own_req) {
490 struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(child);
492 tcp_rsk(req)->drop_req = false;
494 /* we need to fallback on ctx allocation failure and on pre-reqs
495 * checking above. In the latter scenario we additionally need
496 * to reset the context to non MPTCP status.
498 if (!ctx || fallback) {
499 if (fallback_is_fatal)
502 subflow_drop_ctx(child);
506 if (ctx->mp_capable) {
507 /* new mpc subflow takes ownership of the newly
508 * created mptcp socket
510 new_msk->sk_destruct = mptcp_sock_destruct;
511 mptcp_pm_new_connection(mptcp_sk(new_msk), 1);
515 /* with OoO packets we can reach here without ingress
518 ctx->remote_key = mp_opt.sndr_key;
519 ctx->fully_established = mp_opt.mp_capable;
520 ctx->can_ack = mp_opt.mp_capable;
521 } else if (ctx->mp_join) {
522 struct mptcp_sock *owner;
524 owner = mptcp_token_get_sock(ctx->token);
528 ctx->conn = (struct sock *)owner;
529 if (!mptcp_finish_join(child))
532 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKRX);
533 tcp_rsk(req)->drop_req = true;
538 /* dispose of the left over mptcp master, if any */
539 if (unlikely(new_msk))
540 mptcp_force_close(new_msk);
542 /* check for expected invariant - should never trigger, just help
543 * catching eariler subtle bugs
545 WARN_ON_ONCE(child && *own_req && tcp_sk(child)->is_mptcp &&
546 (!mptcp_subflow_ctx(child) ||
547 !mptcp_subflow_ctx(child)->conn));
551 subflow_drop_ctx(child);
552 tcp_rsk(req)->drop_req = true;
553 tcp_send_active_reset(child, GFP_ATOMIC);
554 inet_csk_prepare_for_destroy_sock(child);
557 /* The last child reference will be released by the caller */
561 static struct inet_connection_sock_af_ops subflow_specific;
563 enum mapping_status {
570 static u64 expand_seq(u64 old_seq, u16 old_data_len, u64 seq)
572 if ((u32)seq == (u32)old_seq)
575 /* Assume map covers data not mapped yet. */
576 return seq | ((old_seq + old_data_len + 1) & GENMASK_ULL(63, 32));
579 static void warn_bad_map(struct mptcp_subflow_context *subflow, u32 ssn)
581 WARN_ONCE(1, "Bad mapping: ssn=%d map_seq=%d map_data_len=%d",
582 ssn, subflow->map_subflow_seq, subflow->map_data_len);
585 static bool skb_is_fully_mapped(struct sock *ssk, struct sk_buff *skb)
587 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
588 unsigned int skb_consumed;
590 skb_consumed = tcp_sk(ssk)->copied_seq - TCP_SKB_CB(skb)->seq;
591 if (WARN_ON_ONCE(skb_consumed >= skb->len))
594 return skb->len - skb_consumed <= subflow->map_data_len -
595 mptcp_subflow_get_map_offset(subflow);
598 static bool validate_mapping(struct sock *ssk, struct sk_buff *skb)
600 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
601 u32 ssn = tcp_sk(ssk)->copied_seq - subflow->ssn_offset;
603 if (unlikely(before(ssn, subflow->map_subflow_seq))) {
604 /* Mapping covers data later in the subflow stream,
605 * currently unsupported.
607 warn_bad_map(subflow, ssn);
610 if (unlikely(!before(ssn, subflow->map_subflow_seq +
611 subflow->map_data_len))) {
612 /* Mapping does covers past subflow data, invalid */
613 warn_bad_map(subflow, ssn + skb->len);
619 static enum mapping_status get_mapping_status(struct sock *ssk)
621 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
622 struct mptcp_ext *mpext;
627 skb = skb_peek(&ssk->sk_receive_queue);
629 return MAPPING_EMPTY;
631 mpext = mptcp_get_ext(skb);
632 if (!mpext || !mpext->use_map) {
633 if (!subflow->map_valid && !skb->len) {
634 /* the TCP stack deliver 0 len FIN pkt to the receive
635 * queue, that is the only 0len pkts ever expected here,
636 * and we can admit no mapping only for 0 len pkts
638 if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN))
639 WARN_ONCE(1, "0len seq %d:%d flags %x",
640 TCP_SKB_CB(skb)->seq,
641 TCP_SKB_CB(skb)->end_seq,
642 TCP_SKB_CB(skb)->tcp_flags);
643 sk_eat_skb(ssk, skb);
644 return MAPPING_EMPTY;
647 if (!subflow->map_valid)
648 return MAPPING_INVALID;
653 pr_debug("seq=%llu is64=%d ssn=%u data_len=%u data_fin=%d",
654 mpext->data_seq, mpext->dsn64, mpext->subflow_seq,
655 mpext->data_len, mpext->data_fin);
657 data_len = mpext->data_len;
659 pr_err("Infinite mapping not handled");
660 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_INFINITEMAPRX);
661 return MAPPING_INVALID;
664 if (mpext->data_fin == 1) {
666 pr_debug("DATA_FIN with no payload");
667 if (subflow->map_valid) {
668 /* A DATA_FIN might arrive in a DSS
669 * option before the previous mapping
670 * has been fully consumed. Continue
671 * handling the existing mapping.
673 skb_ext_del(skb, SKB_EXT_MPTCP);
676 return MAPPING_DATA_FIN;
680 /* Adjust for DATA_FIN using 1 byte of sequence space */
685 map_seq = expand_seq(subflow->map_seq, subflow->map_data_len,
687 subflow->use_64bit_ack = 0;
688 pr_debug("expanded seq=%llu", subflow->map_seq);
690 map_seq = mpext->data_seq;
691 subflow->use_64bit_ack = 1;
694 if (subflow->map_valid) {
695 /* Allow replacing only with an identical map */
696 if (subflow->map_seq == map_seq &&
697 subflow->map_subflow_seq == mpext->subflow_seq &&
698 subflow->map_data_len == data_len) {
699 skb_ext_del(skb, SKB_EXT_MPTCP);
703 /* If this skb data are fully covered by the current mapping,
704 * the new map would need caching, which is not supported
706 if (skb_is_fully_mapped(ssk, skb)) {
707 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DSSNOMATCH);
708 return MAPPING_INVALID;
711 /* will validate the next map after consuming the current one */
715 subflow->map_seq = map_seq;
716 subflow->map_subflow_seq = mpext->subflow_seq;
717 subflow->map_data_len = data_len;
718 subflow->map_valid = 1;
719 subflow->mpc_map = mpext->mpc_map;
720 pr_debug("new map seq=%llu subflow_seq=%u data_len=%u",
721 subflow->map_seq, subflow->map_subflow_seq,
722 subflow->map_data_len);
725 /* we revalidate valid mapping on new skb, because we must ensure
726 * the current skb is completely covered by the available mapping
728 if (!validate_mapping(ssk, skb))
729 return MAPPING_INVALID;
731 skb_ext_del(skb, SKB_EXT_MPTCP);
735 static int subflow_read_actor(read_descriptor_t *desc,
737 unsigned int offset, size_t len)
739 size_t copy_len = min(desc->count, len);
741 desc->count -= copy_len;
743 pr_debug("flushed %zu bytes, %zu left", copy_len, desc->count);
747 static bool subflow_check_data_avail(struct sock *ssk)
749 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
750 enum mapping_status status;
751 struct mptcp_sock *msk;
754 pr_debug("msk=%p ssk=%p data_avail=%d skb=%p", subflow->conn, ssk,
755 subflow->data_avail, skb_peek(&ssk->sk_receive_queue));
756 if (subflow->data_avail)
759 msk = mptcp_sk(subflow->conn);
766 status = get_mapping_status(ssk);
767 pr_debug("msk=%p ssk=%p status=%d", msk, ssk, status);
768 if (status == MAPPING_INVALID) {
769 ssk->sk_err = EBADMSG;
773 if (status != MAPPING_OK)
776 skb = skb_peek(&ssk->sk_receive_queue);
777 if (WARN_ON_ONCE(!skb))
780 /* if msk lacks the remote key, this subflow must provide an
781 * MP_CAPABLE-based mapping
783 if (unlikely(!READ_ONCE(msk->can_ack))) {
784 if (!subflow->mpc_map) {
785 ssk->sk_err = EBADMSG;
788 WRITE_ONCE(msk->remote_key, subflow->remote_key);
789 WRITE_ONCE(msk->ack_seq, subflow->map_seq);
790 WRITE_ONCE(msk->can_ack, true);
793 old_ack = READ_ONCE(msk->ack_seq);
794 ack_seq = mptcp_subflow_get_mapped_dsn(subflow);
795 pr_debug("msk ack_seq=%llx subflow ack_seq=%llx", old_ack,
797 if (ack_seq == old_ack)
800 /* only accept in-sequence mapping. Old values are spurious
801 * retransmission; we can hit "future" values on active backup
802 * subflow switch, we relay on retransmissions to get
804 * Cuncurrent subflows support will require subflow data
807 map_remaining = subflow->map_data_len -
808 mptcp_subflow_get_map_offset(subflow);
809 if (before64(ack_seq, old_ack))
810 delta = min_t(size_t, old_ack - ack_seq, map_remaining);
812 delta = min_t(size_t, ack_seq - old_ack, map_remaining);
814 /* discard mapped data */
815 pr_debug("discarding %zu bytes, current map len=%d", delta,
818 read_descriptor_t desc = {
823 ret = tcp_read_sock(ssk, &desc, subflow_read_actor);
830 if (delta == map_remaining)
831 subflow->map_valid = 0;
837 /* fatal protocol error, close the socket */
838 /* This barrier is coupled with smp_rmb() in tcp_poll() */
840 ssk->sk_error_report(ssk);
841 tcp_set_state(ssk, TCP_CLOSE);
842 tcp_send_active_reset(ssk, GFP_ATOMIC);
846 bool mptcp_subflow_data_available(struct sock *sk)
848 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
851 /* check if current mapping is still valid */
852 if (subflow->map_valid &&
853 mptcp_subflow_get_map_offset(subflow) >= subflow->map_data_len) {
854 subflow->map_valid = 0;
855 subflow->data_avail = 0;
857 pr_debug("Done with mapping: seq=%u data_len=%u",
858 subflow->map_subflow_seq,
859 subflow->map_data_len);
862 if (!subflow_check_data_avail(sk)) {
863 subflow->data_avail = 0;
867 skb = skb_peek(&sk->sk_receive_queue);
868 subflow->data_avail = skb &&
869 before(tcp_sk(sk)->copied_seq, TCP_SKB_CB(skb)->end_seq);
870 return subflow->data_avail;
873 /* If ssk has an mptcp parent socket, use the mptcp rcvbuf occupancy,
876 * In mptcp, rwin is about the mptcp-level connection data.
878 * Data that is still on the ssk rx queue can thus be ignored,
879 * as far as mptcp peer is concerened that data is still inflight.
880 * DSS ACK is updated when skb is moved to the mptcp rx queue.
882 void mptcp_space(const struct sock *ssk, int *space, int *full_space)
884 const struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
885 const struct sock *sk = subflow->conn;
887 *space = tcp_space(sk);
888 *full_space = tcp_full_space(sk);
891 static void subflow_data_ready(struct sock *sk)
893 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
894 struct sock *parent = subflow->conn;
896 if (!subflow->mp_capable && !subflow->mp_join) {
897 subflow->tcp_data_ready(sk);
899 parent->sk_data_ready(parent);
903 if (mptcp_subflow_data_available(sk))
904 mptcp_data_ready(parent, sk);
907 static void subflow_write_space(struct sock *sk)
909 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
910 struct sock *parent = subflow->conn;
912 sk_stream_write_space(sk);
913 if (sk_stream_is_writeable(sk)) {
914 set_bit(MPTCP_SEND_SPACE, &mptcp_sk(parent)->flags);
915 smp_mb__after_atomic();
916 /* set SEND_SPACE before sk_stream_write_space clears NOSPACE */
917 sk_stream_write_space(parent);
921 static struct inet_connection_sock_af_ops *
922 subflow_default_af_ops(struct sock *sk)
924 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
925 if (sk->sk_family == AF_INET6)
926 return &subflow_v6_specific;
928 return &subflow_specific;
931 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
932 void mptcpv6_handle_mapped(struct sock *sk, bool mapped)
934 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
935 struct inet_connection_sock *icsk = inet_csk(sk);
936 struct inet_connection_sock_af_ops *target;
938 target = mapped ? &subflow_v6m_specific : subflow_default_af_ops(sk);
940 pr_debug("subflow=%p family=%d ops=%p target=%p mapped=%d",
941 subflow, sk->sk_family, icsk->icsk_af_ops, target, mapped);
943 if (likely(icsk->icsk_af_ops == target))
946 subflow->icsk_af_ops = icsk->icsk_af_ops;
947 icsk->icsk_af_ops = target;
951 static void mptcp_info2sockaddr(const struct mptcp_addr_info *info,
952 struct sockaddr_storage *addr)
954 memset(addr, 0, sizeof(*addr));
955 addr->ss_family = info->family;
956 if (addr->ss_family == AF_INET) {
957 struct sockaddr_in *in_addr = (struct sockaddr_in *)addr;
959 in_addr->sin_addr = info->addr;
960 in_addr->sin_port = info->port;
962 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
963 else if (addr->ss_family == AF_INET6) {
964 struct sockaddr_in6 *in6_addr = (struct sockaddr_in6 *)addr;
966 in6_addr->sin6_addr = info->addr6;
967 in6_addr->sin6_port = info->port;
972 int __mptcp_subflow_connect(struct sock *sk, int ifindex,
973 const struct mptcp_addr_info *loc,
974 const struct mptcp_addr_info *remote)
976 struct mptcp_sock *msk = mptcp_sk(sk);
977 struct mptcp_subflow_context *subflow;
978 struct sockaddr_storage addr;
984 if (sk->sk_state != TCP_ESTABLISHED)
987 err = mptcp_subflow_create_socket(sk, &sf);
991 subflow = mptcp_subflow_ctx(sf->sk);
992 subflow->remote_key = msk->remote_key;
993 subflow->local_key = msk->local_key;
994 subflow->token = msk->token;
995 mptcp_info2sockaddr(loc, &addr);
997 addrlen = sizeof(struct sockaddr_in);
998 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
999 if (loc->family == AF_INET6)
1000 addrlen = sizeof(struct sockaddr_in6);
1002 sf->sk->sk_bound_dev_if = ifindex;
1003 err = kernel_bind(sf, (struct sockaddr *)&addr, addrlen);
1007 mptcp_crypto_key_sha(subflow->remote_key, &remote_token, NULL);
1008 pr_debug("msk=%p remote_token=%u", msk, remote_token);
1009 subflow->remote_token = remote_token;
1010 subflow->local_id = loc->id;
1011 subflow->request_join = 1;
1012 subflow->request_bkup = 1;
1013 mptcp_info2sockaddr(remote, &addr);
1015 err = kernel_connect(sf, (struct sockaddr *)&addr, addrlen, O_NONBLOCK);
1016 if (err && err != -EINPROGRESS)
1019 spin_lock_bh(&msk->join_list_lock);
1020 list_add_tail(&subflow->node, &msk->join_list);
1021 spin_unlock_bh(&msk->join_list_lock);
1030 int mptcp_subflow_create_socket(struct sock *sk, struct socket **new_sock)
1032 struct mptcp_subflow_context *subflow;
1033 struct net *net = sock_net(sk);
1037 err = sock_create_kern(net, sk->sk_family, SOCK_STREAM, IPPROTO_TCP,
1044 /* kernel sockets do not by default acquire net ref, but TCP timer
1047 sf->sk->sk_net_refcnt = 1;
1049 #ifdef CONFIG_PROC_FS
1050 this_cpu_add(*net->core.sock_inuse, 1);
1052 err = tcp_set_ulp(sf->sk, "mptcp");
1053 release_sock(sf->sk);
1058 /* the newly created socket really belongs to the owning MPTCP master
1059 * socket, even if for additional subflows the allocation is performed
1060 * by a kernel workqueue. Adjust inode references, so that the
1061 * procfs/diag interaces really show this one belonging to the correct
1064 SOCK_INODE(sf)->i_ino = SOCK_INODE(sk->sk_socket)->i_ino;
1065 SOCK_INODE(sf)->i_uid = SOCK_INODE(sk->sk_socket)->i_uid;
1066 SOCK_INODE(sf)->i_gid = SOCK_INODE(sk->sk_socket)->i_gid;
1068 subflow = mptcp_subflow_ctx(sf->sk);
1069 pr_debug("subflow=%p", subflow);
1078 static struct mptcp_subflow_context *subflow_create_ctx(struct sock *sk,
1081 struct inet_connection_sock *icsk = inet_csk(sk);
1082 struct mptcp_subflow_context *ctx;
1084 ctx = kzalloc(sizeof(*ctx), priority);
1088 rcu_assign_pointer(icsk->icsk_ulp_data, ctx);
1089 INIT_LIST_HEAD(&ctx->node);
1091 pr_debug("subflow=%p", ctx);
1098 static void __subflow_state_change(struct sock *sk)
1100 struct socket_wq *wq;
1103 wq = rcu_dereference(sk->sk_wq);
1104 if (skwq_has_sleeper(wq))
1105 wake_up_interruptible_all(&wq->wait);
1109 static bool subflow_is_done(const struct sock *sk)
1111 return sk->sk_shutdown & RCV_SHUTDOWN || sk->sk_state == TCP_CLOSE;
1114 static void subflow_state_change(struct sock *sk)
1116 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1117 struct sock *parent = subflow->conn;
1119 __subflow_state_change(sk);
1121 /* as recvmsg() does not acquire the subflow socket for ssk selection
1122 * a fin packet carrying a DSS can be unnoticed if we don't trigger
1123 * the data available machinery here.
1125 if (subflow->mp_capable && mptcp_subflow_data_available(sk))
1126 mptcp_data_ready(parent, sk);
1128 if (!(parent->sk_shutdown & RCV_SHUTDOWN) &&
1129 !subflow->rx_eof && subflow_is_done(sk)) {
1130 subflow->rx_eof = 1;
1131 mptcp_subflow_eof(parent);
1135 static int subflow_ulp_init(struct sock *sk)
1137 struct inet_connection_sock *icsk = inet_csk(sk);
1138 struct mptcp_subflow_context *ctx;
1139 struct tcp_sock *tp = tcp_sk(sk);
1142 /* disallow attaching ULP to a socket unless it has been
1143 * created with sock_create_kern()
1145 if (!sk->sk_kern_sock) {
1150 ctx = subflow_create_ctx(sk, GFP_KERNEL);
1156 pr_debug("subflow=%p, family=%d", ctx, sk->sk_family);
1159 ctx->icsk_af_ops = icsk->icsk_af_ops;
1160 icsk->icsk_af_ops = subflow_default_af_ops(sk);
1161 ctx->tcp_data_ready = sk->sk_data_ready;
1162 ctx->tcp_state_change = sk->sk_state_change;
1163 ctx->tcp_write_space = sk->sk_write_space;
1164 sk->sk_data_ready = subflow_data_ready;
1165 sk->sk_write_space = subflow_write_space;
1166 sk->sk_state_change = subflow_state_change;
1171 static void subflow_ulp_release(struct sock *sk)
1173 struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(sk);
1179 sock_put(ctx->conn);
1181 kfree_rcu(ctx, rcu);
1184 static void subflow_ulp_clone(const struct request_sock *req,
1186 const gfp_t priority)
1188 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
1189 struct mptcp_subflow_context *old_ctx = mptcp_subflow_ctx(newsk);
1190 struct mptcp_subflow_context *new_ctx;
1192 if (!tcp_rsk(req)->is_mptcp ||
1193 (!subflow_req->mp_capable && !subflow_req->mp_join)) {
1194 subflow_ulp_fallback(newsk, old_ctx);
1198 new_ctx = subflow_create_ctx(newsk, priority);
1200 subflow_ulp_fallback(newsk, old_ctx);
1204 new_ctx->conn_finished = 1;
1205 new_ctx->icsk_af_ops = old_ctx->icsk_af_ops;
1206 new_ctx->tcp_data_ready = old_ctx->tcp_data_ready;
1207 new_ctx->tcp_state_change = old_ctx->tcp_state_change;
1208 new_ctx->tcp_write_space = old_ctx->tcp_write_space;
1209 new_ctx->rel_write_seq = 1;
1210 new_ctx->tcp_sock = newsk;
1212 if (subflow_req->mp_capable) {
1213 /* see comments in subflow_syn_recv_sock(), MPTCP connection
1214 * is fully established only after we receive the remote key
1216 new_ctx->mp_capable = 1;
1217 new_ctx->local_key = subflow_req->local_key;
1218 new_ctx->token = subflow_req->token;
1219 new_ctx->ssn_offset = subflow_req->ssn_offset;
1220 new_ctx->idsn = subflow_req->idsn;
1221 } else if (subflow_req->mp_join) {
1222 new_ctx->ssn_offset = subflow_req->ssn_offset;
1223 new_ctx->mp_join = 1;
1224 new_ctx->fully_established = 1;
1225 new_ctx->backup = subflow_req->backup;
1226 new_ctx->local_id = subflow_req->local_id;
1227 new_ctx->token = subflow_req->token;
1228 new_ctx->thmac = subflow_req->thmac;
1232 static struct tcp_ulp_ops subflow_ulp_ops __read_mostly = {
1234 .owner = THIS_MODULE,
1235 .init = subflow_ulp_init,
1236 .release = subflow_ulp_release,
1237 .clone = subflow_ulp_clone,
1240 static int subflow_ops_init(struct request_sock_ops *subflow_ops)
1242 subflow_ops->obj_size = sizeof(struct mptcp_subflow_request_sock);
1243 subflow_ops->slab_name = "request_sock_subflow";
1245 subflow_ops->slab = kmem_cache_create(subflow_ops->slab_name,
1246 subflow_ops->obj_size, 0,
1248 SLAB_TYPESAFE_BY_RCU,
1250 if (!subflow_ops->slab)
1253 subflow_ops->destructor = subflow_req_destructor;
1258 void mptcp_subflow_init(void)
1260 subflow_request_sock_ops = tcp_request_sock_ops;
1261 if (subflow_ops_init(&subflow_request_sock_ops) != 0)
1262 panic("MPTCP: failed to init subflow request sock ops\n");
1264 subflow_request_sock_ipv4_ops = tcp_request_sock_ipv4_ops;
1265 subflow_request_sock_ipv4_ops.init_req = subflow_v4_init_req;
1267 subflow_specific = ipv4_specific;
1268 subflow_specific.conn_request = subflow_v4_conn_request;
1269 subflow_specific.syn_recv_sock = subflow_syn_recv_sock;
1270 subflow_specific.sk_rx_dst_set = subflow_finish_connect;
1271 subflow_specific.rebuild_header = subflow_rebuild_header;
1273 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1274 subflow_request_sock_ipv6_ops = tcp_request_sock_ipv6_ops;
1275 subflow_request_sock_ipv6_ops.init_req = subflow_v6_init_req;
1277 subflow_v6_specific = ipv6_specific;
1278 subflow_v6_specific.conn_request = subflow_v6_conn_request;
1279 subflow_v6_specific.syn_recv_sock = subflow_syn_recv_sock;
1280 subflow_v6_specific.sk_rx_dst_set = subflow_finish_connect;
1281 subflow_v6_specific.rebuild_header = subflow_rebuild_header;
1283 subflow_v6m_specific = subflow_v6_specific;
1284 subflow_v6m_specific.queue_xmit = ipv4_specific.queue_xmit;
1285 subflow_v6m_specific.send_check = ipv4_specific.send_check;
1286 subflow_v6m_specific.net_header_len = ipv4_specific.net_header_len;
1287 subflow_v6m_specific.mtu_reduced = ipv4_specific.mtu_reduced;
1288 subflow_v6m_specific.net_frag_header_len = 0;
1291 mptcp_diag_subflow_init(&subflow_ulp_ops);
1293 if (tcp_register_ulp(&subflow_ulp_ops) != 0)
1294 panic("MPTCP: failed to register subflows to ULP\n");