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);
73 sock_put((struct sock *)subflow_req->msk);
75 if (subflow_req->mp_capable)
76 mptcp_token_destroy_request(subflow_req->token);
77 tcp_request_sock_ops.destructor(req);
80 static void subflow_generate_hmac(u64 key1, u64 key2, u32 nonce1, u32 nonce2,
85 put_unaligned_be32(nonce1, &msg[0]);
86 put_unaligned_be32(nonce2, &msg[4]);
88 mptcp_crypto_hmac_sha(key1, key2, msg, 8, hmac);
91 /* validate received token and create truncated hmac and nonce for SYN-ACK */
92 static struct mptcp_sock *subflow_token_join_request(struct request_sock *req,
93 const struct sk_buff *skb)
95 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
96 u8 hmac[SHA256_DIGEST_SIZE];
97 struct mptcp_sock *msk;
100 msk = mptcp_token_get_sock(subflow_req->token);
102 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINNOTOKEN);
106 local_id = mptcp_pm_get_local_id(msk, (struct sock_common *)req);
108 sock_put((struct sock *)msk);
111 subflow_req->local_id = local_id;
113 get_random_bytes(&subflow_req->local_nonce, sizeof(u32));
115 subflow_generate_hmac(msk->local_key, msk->remote_key,
116 subflow_req->local_nonce,
117 subflow_req->remote_nonce, hmac);
119 subflow_req->thmac = get_unaligned_be64(hmac);
123 static void subflow_init_req(struct request_sock *req,
124 const struct sock *sk_listener,
127 struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk_listener);
128 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
129 struct mptcp_options_received mp_opt;
131 pr_debug("subflow_req=%p, listener=%p", subflow_req, listener);
133 mptcp_get_options(skb, &mp_opt);
135 subflow_req->mp_capable = 0;
136 subflow_req->mp_join = 0;
137 subflow_req->msk = NULL;
139 #ifdef CONFIG_TCP_MD5SIG
140 /* no MPTCP if MD5SIG is enabled on this socket or we may run out of
143 if (rcu_access_pointer(tcp_sk(sk_listener)->md5sig_info))
147 if (mp_opt.mp_capable) {
148 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MPCAPABLEPASSIVE);
152 } else if (mp_opt.mp_join) {
153 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINSYNRX);
156 if (mp_opt.mp_capable && listener->request_mptcp) {
159 err = mptcp_token_new_request(req);
161 subflow_req->mp_capable = 1;
163 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq;
164 } else if (mp_opt.mp_join && listener->request_mptcp) {
165 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq;
166 subflow_req->mp_join = 1;
167 subflow_req->backup = mp_opt.backup;
168 subflow_req->remote_id = mp_opt.join_id;
169 subflow_req->token = mp_opt.token;
170 subflow_req->remote_nonce = mp_opt.nonce;
171 subflow_req->msk = subflow_token_join_request(req, skb);
172 pr_debug("token=%u, remote_nonce=%u msk=%p", subflow_req->token,
173 subflow_req->remote_nonce, subflow_req->msk);
177 static void subflow_v4_init_req(struct request_sock *req,
178 const struct sock *sk_listener,
181 tcp_rsk(req)->is_mptcp = 1;
183 tcp_request_sock_ipv4_ops.init_req(req, sk_listener, skb);
185 subflow_init_req(req, sk_listener, skb);
188 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
189 static void subflow_v6_init_req(struct request_sock *req,
190 const struct sock *sk_listener,
193 tcp_rsk(req)->is_mptcp = 1;
195 tcp_request_sock_ipv6_ops.init_req(req, sk_listener, skb);
197 subflow_init_req(req, sk_listener, skb);
201 /* validate received truncated hmac and create hmac for third ACK */
202 static bool subflow_thmac_valid(struct mptcp_subflow_context *subflow)
204 u8 hmac[SHA256_DIGEST_SIZE];
207 subflow_generate_hmac(subflow->remote_key, subflow->local_key,
208 subflow->remote_nonce, subflow->local_nonce,
211 thmac = get_unaligned_be64(hmac);
212 pr_debug("subflow=%p, token=%u, thmac=%llu, subflow->thmac=%llu\n",
213 subflow, subflow->token,
214 (unsigned long long)thmac,
215 (unsigned long long)subflow->thmac);
217 return thmac == subflow->thmac;
220 static void subflow_finish_connect(struct sock *sk, const struct sk_buff *skb)
222 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
223 struct mptcp_options_received mp_opt;
224 struct sock *parent = subflow->conn;
225 struct tcp_sock *tp = tcp_sk(sk);
227 subflow->icsk_af_ops->sk_rx_dst_set(sk, skb);
229 if (inet_sk_state_load(parent) == TCP_SYN_SENT) {
230 inet_sk_state_store(parent, TCP_ESTABLISHED);
231 parent->sk_state_change(parent);
234 /* be sure no special action on any packet other than syn-ack */
235 if (subflow->conn_finished)
238 subflow->conn_finished = 1;
240 mptcp_get_options(skb, &mp_opt);
241 if (subflow->request_mptcp && mp_opt.mp_capable) {
242 subflow->mp_capable = 1;
243 subflow->can_ack = 1;
244 subflow->remote_key = mp_opt.sndr_key;
245 pr_debug("subflow=%p, remote_key=%llu", subflow,
246 subflow->remote_key);
247 } else if (subflow->request_join && mp_opt.mp_join) {
248 subflow->mp_join = 1;
249 subflow->thmac = mp_opt.thmac;
250 subflow->remote_nonce = mp_opt.nonce;
251 pr_debug("subflow=%p, thmac=%llu, remote_nonce=%u", subflow,
252 subflow->thmac, subflow->remote_nonce);
253 } else if (subflow->request_mptcp) {
260 if (subflow->mp_capable) {
261 pr_debug("subflow=%p, remote_key=%llu", mptcp_subflow_ctx(sk),
262 subflow->remote_key);
263 mptcp_finish_connect(sk);
266 pr_debug("synack seq=%u", TCP_SKB_CB(skb)->seq);
267 subflow->ssn_offset = TCP_SKB_CB(skb)->seq;
269 } else if (subflow->mp_join) {
270 u8 hmac[SHA256_DIGEST_SIZE];
272 pr_debug("subflow=%p, thmac=%llu, remote_nonce=%u",
273 subflow, subflow->thmac,
274 subflow->remote_nonce);
275 if (!subflow_thmac_valid(subflow)) {
276 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINACKMAC);
277 subflow->mp_join = 0;
281 subflow_generate_hmac(subflow->local_key, subflow->remote_key,
282 subflow->local_nonce,
283 subflow->remote_nonce,
286 memcpy(subflow->hmac, hmac, MPTCPOPT_HMAC_LEN);
289 subflow->ssn_offset = TCP_SKB_CB(skb)->seq;
291 if (!mptcp_finish_join(sk))
294 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINSYNACKRX);
297 tcp_send_active_reset(sk, GFP_ATOMIC);
302 static struct request_sock_ops subflow_request_sock_ops;
303 static struct tcp_request_sock_ops subflow_request_sock_ipv4_ops;
305 static int subflow_v4_conn_request(struct sock *sk, struct sk_buff *skb)
307 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
309 pr_debug("subflow=%p", subflow);
311 /* Never answer to SYNs sent to broadcast or multicast */
312 if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
315 return tcp_conn_request(&subflow_request_sock_ops,
316 &subflow_request_sock_ipv4_ops,
323 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
324 static struct tcp_request_sock_ops subflow_request_sock_ipv6_ops;
325 static struct inet_connection_sock_af_ops subflow_v6_specific;
326 static struct inet_connection_sock_af_ops subflow_v6m_specific;
328 static int subflow_v6_conn_request(struct sock *sk, struct sk_buff *skb)
330 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
332 pr_debug("subflow=%p", subflow);
334 if (skb->protocol == htons(ETH_P_IP))
335 return subflow_v4_conn_request(sk, skb);
337 if (!ipv6_unicast_destination(skb))
340 return tcp_conn_request(&subflow_request_sock_ops,
341 &subflow_request_sock_ipv6_ops, sk, skb);
345 return 0; /* don't send reset */
349 /* validate hmac received in third ACK */
350 static bool subflow_hmac_valid(const struct request_sock *req,
351 const struct mptcp_options_received *mp_opt)
353 const struct mptcp_subflow_request_sock *subflow_req;
354 u8 hmac[SHA256_DIGEST_SIZE];
355 struct mptcp_sock *msk;
357 subflow_req = mptcp_subflow_rsk(req);
358 msk = subflow_req->msk;
362 subflow_generate_hmac(msk->remote_key, msk->local_key,
363 subflow_req->remote_nonce,
364 subflow_req->local_nonce, hmac);
366 return !crypto_memneq(hmac, mp_opt->hmac, MPTCPOPT_HMAC_LEN);
369 static void mptcp_sock_destruct(struct sock *sk)
371 /* if new mptcp socket isn't accepted, it is free'd
372 * from the tcp listener sockets request queue, linked
373 * from req->sk. The tcp socket is released.
374 * This calls the ULP release function which will
375 * also remove the mptcp socket, via
376 * sock_put(ctx->conn).
378 * Problem is that the mptcp socket will not be in
379 * SYN_RECV state and doesn't have SOCK_DEAD flag.
380 * Both result in warnings from inet_sock_destruct.
383 if (sk->sk_state == TCP_SYN_RECV) {
384 sk->sk_state = TCP_CLOSE;
385 WARN_ON_ONCE(sk->sk_socket);
389 mptcp_token_destroy(mptcp_sk(sk)->token);
390 inet_sock_destruct(sk);
393 static void mptcp_force_close(struct sock *sk)
395 inet_sk_state_store(sk, TCP_CLOSE);
396 sk_common_release(sk);
399 static void subflow_ulp_fallback(struct sock *sk,
400 struct mptcp_subflow_context *old_ctx)
402 struct inet_connection_sock *icsk = inet_csk(sk);
404 mptcp_subflow_tcp_fallback(sk, old_ctx);
405 icsk->icsk_ulp_ops = NULL;
406 rcu_assign_pointer(icsk->icsk_ulp_data, NULL);
407 tcp_sk(sk)->is_mptcp = 0;
410 static void subflow_drop_ctx(struct sock *ssk)
412 struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(ssk);
417 subflow_ulp_fallback(ssk, ctx);
424 static struct sock *subflow_syn_recv_sock(const struct sock *sk,
426 struct request_sock *req,
427 struct dst_entry *dst,
428 struct request_sock *req_unhash,
431 struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk);
432 struct mptcp_subflow_request_sock *subflow_req;
433 struct mptcp_options_received mp_opt;
434 bool fallback, fallback_is_fatal;
435 struct sock *new_msk = NULL;
438 pr_debug("listener=%p, req=%p, conn=%p", listener, req, listener->conn);
440 /* After child creation we must look for 'mp_capable' even when options
443 mp_opt.mp_capable = 0;
445 /* hopefully temporary handling for MP_JOIN+syncookie */
446 subflow_req = mptcp_subflow_rsk(req);
447 fallback_is_fatal = subflow_req->mp_join;
448 fallback = !tcp_rsk(req)->is_mptcp;
452 /* if the sk is MP_CAPABLE, we try to fetch the client key */
453 if (subflow_req->mp_capable) {
454 if (TCP_SKB_CB(skb)->seq != subflow_req->ssn_offset + 1) {
455 /* here we can receive and accept an in-window,
456 * out-of-order pkt, which will not carry the MP_CAPABLE
457 * opt even on mptcp enabled paths
462 mptcp_get_options(skb, &mp_opt);
463 if (!mp_opt.mp_capable) {
469 new_msk = mptcp_sk_clone(listener->conn, &mp_opt, req);
472 } else if (subflow_req->mp_join) {
473 mptcp_get_options(skb, &mp_opt);
474 if (!mp_opt.mp_join ||
475 !subflow_hmac_valid(req, &mp_opt)) {
476 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKMAC);
482 child = listener->icsk_af_ops->syn_recv_sock(sk, skb, req, dst,
483 req_unhash, own_req);
485 if (child && *own_req) {
486 struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(child);
488 tcp_rsk(req)->drop_req = false;
490 /* we need to fallback on ctx allocation failure and on pre-reqs
491 * checking above. In the latter scenario we additionally need
492 * to reset the context to non MPTCP status.
494 if (!ctx || fallback) {
495 if (fallback_is_fatal)
498 subflow_drop_ctx(child);
502 if (ctx->mp_capable) {
503 /* new mpc subflow takes ownership of the newly
504 * created mptcp socket
506 new_msk->sk_destruct = mptcp_sock_destruct;
507 mptcp_pm_new_connection(mptcp_sk(new_msk), 1);
511 /* with OoO packets we can reach here without ingress
514 ctx->remote_key = mp_opt.sndr_key;
515 ctx->fully_established = mp_opt.mp_capable;
516 ctx->can_ack = mp_opt.mp_capable;
517 } else if (ctx->mp_join) {
518 struct mptcp_sock *owner;
520 owner = subflow_req->msk;
524 /* move the msk reference ownership to the subflow */
525 subflow_req->msk = NULL;
526 ctx->conn = (struct sock *)owner;
527 if (!mptcp_finish_join(child))
530 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKRX);
531 tcp_rsk(req)->drop_req = true;
536 /* dispose of the left over mptcp master, if any */
537 if (unlikely(new_msk))
538 mptcp_force_close(new_msk);
540 /* check for expected invariant - should never trigger, just help
541 * catching eariler subtle bugs
543 WARN_ON_ONCE(child && *own_req && tcp_sk(child)->is_mptcp &&
544 (!mptcp_subflow_ctx(child) ||
545 !mptcp_subflow_ctx(child)->conn));
549 subflow_drop_ctx(child);
550 tcp_rsk(req)->drop_req = true;
551 tcp_send_active_reset(child, GFP_ATOMIC);
552 inet_csk_prepare_for_destroy_sock(child);
555 /* The last child reference will be released by the caller */
559 static struct inet_connection_sock_af_ops subflow_specific;
561 enum mapping_status {
568 static u64 expand_seq(u64 old_seq, u16 old_data_len, u64 seq)
570 if ((u32)seq == (u32)old_seq)
573 /* Assume map covers data not mapped yet. */
574 return seq | ((old_seq + old_data_len + 1) & GENMASK_ULL(63, 32));
577 static void warn_bad_map(struct mptcp_subflow_context *subflow, u32 ssn)
579 WARN_ONCE(1, "Bad mapping: ssn=%d map_seq=%d map_data_len=%d",
580 ssn, subflow->map_subflow_seq, subflow->map_data_len);
583 static bool skb_is_fully_mapped(struct sock *ssk, struct sk_buff *skb)
585 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
586 unsigned int skb_consumed;
588 skb_consumed = tcp_sk(ssk)->copied_seq - TCP_SKB_CB(skb)->seq;
589 if (WARN_ON_ONCE(skb_consumed >= skb->len))
592 return skb->len - skb_consumed <= subflow->map_data_len -
593 mptcp_subflow_get_map_offset(subflow);
596 static bool validate_mapping(struct sock *ssk, struct sk_buff *skb)
598 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
599 u32 ssn = tcp_sk(ssk)->copied_seq - subflow->ssn_offset;
601 if (unlikely(before(ssn, subflow->map_subflow_seq))) {
602 /* Mapping covers data later in the subflow stream,
603 * currently unsupported.
605 warn_bad_map(subflow, ssn);
608 if (unlikely(!before(ssn, subflow->map_subflow_seq +
609 subflow->map_data_len))) {
610 /* Mapping does covers past subflow data, invalid */
611 warn_bad_map(subflow, ssn + skb->len);
617 static enum mapping_status get_mapping_status(struct sock *ssk)
619 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
620 struct mptcp_ext *mpext;
625 skb = skb_peek(&ssk->sk_receive_queue);
627 return MAPPING_EMPTY;
629 mpext = mptcp_get_ext(skb);
630 if (!mpext || !mpext->use_map) {
631 if (!subflow->map_valid && !skb->len) {
632 /* the TCP stack deliver 0 len FIN pkt to the receive
633 * queue, that is the only 0len pkts ever expected here,
634 * and we can admit no mapping only for 0 len pkts
636 if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN))
637 WARN_ONCE(1, "0len seq %d:%d flags %x",
638 TCP_SKB_CB(skb)->seq,
639 TCP_SKB_CB(skb)->end_seq,
640 TCP_SKB_CB(skb)->tcp_flags);
641 sk_eat_skb(ssk, skb);
642 return MAPPING_EMPTY;
645 if (!subflow->map_valid)
646 return MAPPING_INVALID;
651 pr_debug("seq=%llu is64=%d ssn=%u data_len=%u data_fin=%d",
652 mpext->data_seq, mpext->dsn64, mpext->subflow_seq,
653 mpext->data_len, mpext->data_fin);
655 data_len = mpext->data_len;
657 pr_err("Infinite mapping not handled");
658 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_INFINITEMAPRX);
659 return MAPPING_INVALID;
662 if (mpext->data_fin == 1) {
664 pr_debug("DATA_FIN with no payload");
665 if (subflow->map_valid) {
666 /* A DATA_FIN might arrive in a DSS
667 * option before the previous mapping
668 * has been fully consumed. Continue
669 * handling the existing mapping.
671 skb_ext_del(skb, SKB_EXT_MPTCP);
674 return MAPPING_DATA_FIN;
678 /* Adjust for DATA_FIN using 1 byte of sequence space */
683 map_seq = expand_seq(subflow->map_seq, subflow->map_data_len,
685 subflow->use_64bit_ack = 0;
686 pr_debug("expanded seq=%llu", subflow->map_seq);
688 map_seq = mpext->data_seq;
689 subflow->use_64bit_ack = 1;
692 if (subflow->map_valid) {
693 /* Allow replacing only with an identical map */
694 if (subflow->map_seq == map_seq &&
695 subflow->map_subflow_seq == mpext->subflow_seq &&
696 subflow->map_data_len == data_len) {
697 skb_ext_del(skb, SKB_EXT_MPTCP);
701 /* If this skb data are fully covered by the current mapping,
702 * the new map would need caching, which is not supported
704 if (skb_is_fully_mapped(ssk, skb)) {
705 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DSSNOMATCH);
706 return MAPPING_INVALID;
709 /* will validate the next map after consuming the current one */
713 subflow->map_seq = map_seq;
714 subflow->map_subflow_seq = mpext->subflow_seq;
715 subflow->map_data_len = data_len;
716 subflow->map_valid = 1;
717 subflow->mpc_map = mpext->mpc_map;
718 pr_debug("new map seq=%llu subflow_seq=%u data_len=%u",
719 subflow->map_seq, subflow->map_subflow_seq,
720 subflow->map_data_len);
723 /* we revalidate valid mapping on new skb, because we must ensure
724 * the current skb is completely covered by the available mapping
726 if (!validate_mapping(ssk, skb))
727 return MAPPING_INVALID;
729 skb_ext_del(skb, SKB_EXT_MPTCP);
733 static int subflow_read_actor(read_descriptor_t *desc,
735 unsigned int offset, size_t len)
737 size_t copy_len = min(desc->count, len);
739 desc->count -= copy_len;
741 pr_debug("flushed %zu bytes, %zu left", copy_len, desc->count);
745 static bool subflow_check_data_avail(struct sock *ssk)
747 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
748 enum mapping_status status;
749 struct mptcp_sock *msk;
752 pr_debug("msk=%p ssk=%p data_avail=%d skb=%p", subflow->conn, ssk,
753 subflow->data_avail, skb_peek(&ssk->sk_receive_queue));
754 if (subflow->data_avail)
757 msk = mptcp_sk(subflow->conn);
764 status = get_mapping_status(ssk);
765 pr_debug("msk=%p ssk=%p status=%d", msk, ssk, status);
766 if (status == MAPPING_INVALID) {
767 ssk->sk_err = EBADMSG;
771 if (status != MAPPING_OK)
774 skb = skb_peek(&ssk->sk_receive_queue);
775 if (WARN_ON_ONCE(!skb))
778 /* if msk lacks the remote key, this subflow must provide an
779 * MP_CAPABLE-based mapping
781 if (unlikely(!READ_ONCE(msk->can_ack))) {
782 if (!subflow->mpc_map) {
783 ssk->sk_err = EBADMSG;
786 WRITE_ONCE(msk->remote_key, subflow->remote_key);
787 WRITE_ONCE(msk->ack_seq, subflow->map_seq);
788 WRITE_ONCE(msk->can_ack, true);
791 old_ack = READ_ONCE(msk->ack_seq);
792 ack_seq = mptcp_subflow_get_mapped_dsn(subflow);
793 pr_debug("msk ack_seq=%llx subflow ack_seq=%llx", old_ack,
795 if (ack_seq == old_ack)
798 /* only accept in-sequence mapping. Old values are spurious
799 * retransmission; we can hit "future" values on active backup
800 * subflow switch, we relay on retransmissions to get
802 * Cuncurrent subflows support will require subflow data
805 map_remaining = subflow->map_data_len -
806 mptcp_subflow_get_map_offset(subflow);
807 if (before64(ack_seq, old_ack))
808 delta = min_t(size_t, old_ack - ack_seq, map_remaining);
810 delta = min_t(size_t, ack_seq - old_ack, map_remaining);
812 /* discard mapped data */
813 pr_debug("discarding %zu bytes, current map len=%d", delta,
816 read_descriptor_t desc = {
821 ret = tcp_read_sock(ssk, &desc, subflow_read_actor);
828 if (delta == map_remaining)
829 subflow->map_valid = 0;
835 /* fatal protocol error, close the socket */
836 /* This barrier is coupled with smp_rmb() in tcp_poll() */
838 ssk->sk_error_report(ssk);
839 tcp_set_state(ssk, TCP_CLOSE);
840 tcp_send_active_reset(ssk, GFP_ATOMIC);
844 bool mptcp_subflow_data_available(struct sock *sk)
846 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
849 /* check if current mapping is still valid */
850 if (subflow->map_valid &&
851 mptcp_subflow_get_map_offset(subflow) >= subflow->map_data_len) {
852 subflow->map_valid = 0;
853 subflow->data_avail = 0;
855 pr_debug("Done with mapping: seq=%u data_len=%u",
856 subflow->map_subflow_seq,
857 subflow->map_data_len);
860 if (!subflow_check_data_avail(sk)) {
861 subflow->data_avail = 0;
865 skb = skb_peek(&sk->sk_receive_queue);
866 subflow->data_avail = skb &&
867 before(tcp_sk(sk)->copied_seq, TCP_SKB_CB(skb)->end_seq);
868 return subflow->data_avail;
871 /* If ssk has an mptcp parent socket, use the mptcp rcvbuf occupancy,
874 * In mptcp, rwin is about the mptcp-level connection data.
876 * Data that is still on the ssk rx queue can thus be ignored,
877 * as far as mptcp peer is concerened that data is still inflight.
878 * DSS ACK is updated when skb is moved to the mptcp rx queue.
880 void mptcp_space(const struct sock *ssk, int *space, int *full_space)
882 const struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
883 const struct sock *sk = subflow->conn;
885 *space = tcp_space(sk);
886 *full_space = tcp_full_space(sk);
889 static void subflow_data_ready(struct sock *sk)
891 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
892 struct sock *parent = subflow->conn;
894 if (!subflow->mp_capable && !subflow->mp_join) {
895 subflow->tcp_data_ready(sk);
897 parent->sk_data_ready(parent);
901 if (mptcp_subflow_data_available(sk))
902 mptcp_data_ready(parent, sk);
905 static void subflow_write_space(struct sock *sk)
907 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
908 struct sock *parent = subflow->conn;
910 sk_stream_write_space(sk);
911 if (sk_stream_is_writeable(sk)) {
912 set_bit(MPTCP_SEND_SPACE, &mptcp_sk(parent)->flags);
913 smp_mb__after_atomic();
914 /* set SEND_SPACE before sk_stream_write_space clears NOSPACE */
915 sk_stream_write_space(parent);
919 static struct inet_connection_sock_af_ops *
920 subflow_default_af_ops(struct sock *sk)
922 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
923 if (sk->sk_family == AF_INET6)
924 return &subflow_v6_specific;
926 return &subflow_specific;
929 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
930 void mptcpv6_handle_mapped(struct sock *sk, bool mapped)
932 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
933 struct inet_connection_sock *icsk = inet_csk(sk);
934 struct inet_connection_sock_af_ops *target;
936 target = mapped ? &subflow_v6m_specific : subflow_default_af_ops(sk);
938 pr_debug("subflow=%p family=%d ops=%p target=%p mapped=%d",
939 subflow, sk->sk_family, icsk->icsk_af_ops, target, mapped);
941 if (likely(icsk->icsk_af_ops == target))
944 subflow->icsk_af_ops = icsk->icsk_af_ops;
945 icsk->icsk_af_ops = target;
949 static void mptcp_info2sockaddr(const struct mptcp_addr_info *info,
950 struct sockaddr_storage *addr)
952 memset(addr, 0, sizeof(*addr));
953 addr->ss_family = info->family;
954 if (addr->ss_family == AF_INET) {
955 struct sockaddr_in *in_addr = (struct sockaddr_in *)addr;
957 in_addr->sin_addr = info->addr;
958 in_addr->sin_port = info->port;
960 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
961 else if (addr->ss_family == AF_INET6) {
962 struct sockaddr_in6 *in6_addr = (struct sockaddr_in6 *)addr;
964 in6_addr->sin6_addr = info->addr6;
965 in6_addr->sin6_port = info->port;
970 int __mptcp_subflow_connect(struct sock *sk, int ifindex,
971 const struct mptcp_addr_info *loc,
972 const struct mptcp_addr_info *remote)
974 struct mptcp_sock *msk = mptcp_sk(sk);
975 struct mptcp_subflow_context *subflow;
976 struct sockaddr_storage addr;
982 if (sk->sk_state != TCP_ESTABLISHED)
985 err = mptcp_subflow_create_socket(sk, &sf);
989 subflow = mptcp_subflow_ctx(sf->sk);
990 subflow->remote_key = msk->remote_key;
991 subflow->local_key = msk->local_key;
992 subflow->token = msk->token;
993 mptcp_info2sockaddr(loc, &addr);
995 addrlen = sizeof(struct sockaddr_in);
996 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
997 if (loc->family == AF_INET6)
998 addrlen = sizeof(struct sockaddr_in6);
1000 sf->sk->sk_bound_dev_if = ifindex;
1001 err = kernel_bind(sf, (struct sockaddr *)&addr, addrlen);
1005 mptcp_crypto_key_sha(subflow->remote_key, &remote_token, NULL);
1006 pr_debug("msk=%p remote_token=%u", msk, remote_token);
1007 subflow->remote_token = remote_token;
1008 subflow->local_id = loc->id;
1009 subflow->request_join = 1;
1010 subflow->request_bkup = 1;
1011 mptcp_info2sockaddr(remote, &addr);
1013 err = kernel_connect(sf, (struct sockaddr *)&addr, addrlen, O_NONBLOCK);
1014 if (err && err != -EINPROGRESS)
1017 spin_lock_bh(&msk->join_list_lock);
1018 list_add_tail(&subflow->node, &msk->join_list);
1019 spin_unlock_bh(&msk->join_list_lock);
1028 int mptcp_subflow_create_socket(struct sock *sk, struct socket **new_sock)
1030 struct mptcp_subflow_context *subflow;
1031 struct net *net = sock_net(sk);
1035 err = sock_create_kern(net, sk->sk_family, SOCK_STREAM, IPPROTO_TCP,
1042 /* kernel sockets do not by default acquire net ref, but TCP timer
1045 sf->sk->sk_net_refcnt = 1;
1047 #ifdef CONFIG_PROC_FS
1048 this_cpu_add(*net->core.sock_inuse, 1);
1050 err = tcp_set_ulp(sf->sk, "mptcp");
1051 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");