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 <crypto/sha.h>
12 #include <net/mptcp.h>
16 static bool mptcp_cap_flag_sha256(u8 flags)
18 return (flags & MPTCP_CAP_FLAG_MASK) == MPTCP_CAP_HMAC_SHA256;
21 static void mptcp_parse_option(const struct sk_buff *skb,
22 const unsigned char *ptr, int opsize,
23 struct mptcp_options_received *mp_opt)
25 u8 subtype = *ptr >> 4;
31 case MPTCPOPT_MP_CAPABLE:
32 /* strict size checking */
33 if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)) {
34 if (skb->len > tcp_hdr(skb)->doff << 2)
35 expected_opsize = TCPOLEN_MPTCP_MPC_ACK_DATA;
37 expected_opsize = TCPOLEN_MPTCP_MPC_ACK;
39 if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_ACK)
40 expected_opsize = TCPOLEN_MPTCP_MPC_SYNACK;
42 expected_opsize = TCPOLEN_MPTCP_MPC_SYN;
44 if (opsize != expected_opsize)
47 /* try to be gentle vs future versions on the initial syn */
48 version = *ptr++ & MPTCP_VERSION_MASK;
49 if (opsize != TCPOLEN_MPTCP_MPC_SYN) {
50 if (version != MPTCP_SUPPORTED_VERSION)
52 } else if (version < MPTCP_SUPPORTED_VERSION) {
57 if (!mptcp_cap_flag_sha256(flags) ||
58 (flags & MPTCP_CAP_EXTENSIBILITY))
61 /* RFC 6824, Section 3.1:
62 * "For the Checksum Required bit (labeled "A"), if either
63 * host requires the use of checksums, checksums MUST be used.
64 * In other words, the only way for checksums not to be used
65 * is if both hosts in their SYNs set A=0."
68 * "If a checksum is not present when its use has been
69 * negotiated, the receiver MUST close the subflow with a RST as
70 * it is considered broken."
72 * We don't implement DSS checksum - fall back to TCP.
74 if (flags & MPTCP_CAP_CHECKSUM_REQD)
77 mp_opt->mp_capable = 1;
78 if (opsize >= TCPOLEN_MPTCP_MPC_SYNACK) {
79 mp_opt->sndr_key = get_unaligned_be64(ptr);
82 if (opsize >= TCPOLEN_MPTCP_MPC_ACK) {
83 mp_opt->rcvr_key = get_unaligned_be64(ptr);
86 if (opsize == TCPOLEN_MPTCP_MPC_ACK_DATA) {
88 * "the data parameters in a MP_CAPABLE are semantically
89 * equivalent to those in a DSS option and can be used
95 mp_opt->data_len = get_unaligned_be16(ptr);
98 pr_debug("MP_CAPABLE version=%x, flags=%x, optlen=%d sndr=%llu, rcvr=%llu len=%d",
99 version, flags, opsize, mp_opt->sndr_key,
100 mp_opt->rcvr_key, mp_opt->data_len);
103 case MPTCPOPT_MP_JOIN:
105 if (opsize == TCPOLEN_MPTCP_MPJ_SYN) {
106 mp_opt->backup = *ptr++ & MPTCPOPT_BACKUP;
107 mp_opt->join_id = *ptr++;
108 mp_opt->token = get_unaligned_be32(ptr);
110 mp_opt->nonce = get_unaligned_be32(ptr);
112 pr_debug("MP_JOIN bkup=%u, id=%u, token=%u, nonce=%u",
113 mp_opt->backup, mp_opt->join_id,
114 mp_opt->token, mp_opt->nonce);
115 } else if (opsize == TCPOLEN_MPTCP_MPJ_SYNACK) {
116 mp_opt->backup = *ptr++ & MPTCPOPT_BACKUP;
117 mp_opt->join_id = *ptr++;
118 mp_opt->thmac = get_unaligned_be64(ptr);
120 mp_opt->nonce = get_unaligned_be32(ptr);
122 pr_debug("MP_JOIN bkup=%u, id=%u, thmac=%llu, nonce=%u",
123 mp_opt->backup, mp_opt->join_id,
124 mp_opt->thmac, mp_opt->nonce);
125 } else if (opsize == TCPOLEN_MPTCP_MPJ_ACK) {
127 memcpy(mp_opt->hmac, ptr, MPTCPOPT_HMAC_LEN);
128 pr_debug("MP_JOIN hmac");
130 pr_warn("MP_JOIN bad option size");
139 /* we must clear 'mpc_map' be able to detect MP_CAPABLE
140 * map vs DSS map in mptcp_incoming_options(), and reconstruct
141 * map info accordingly
144 flags = (*ptr++) & MPTCP_DSS_FLAG_MASK;
145 mp_opt->data_fin = (flags & MPTCP_DSS_DATA_FIN) != 0;
146 mp_opt->dsn64 = (flags & MPTCP_DSS_DSN64) != 0;
147 mp_opt->use_map = (flags & MPTCP_DSS_HAS_MAP) != 0;
148 mp_opt->ack64 = (flags & MPTCP_DSS_ACK64) != 0;
149 mp_opt->use_ack = (flags & MPTCP_DSS_HAS_ACK);
151 pr_debug("data_fin=%d dsn64=%d use_map=%d ack64=%d use_ack=%d",
152 mp_opt->data_fin, mp_opt->dsn64,
153 mp_opt->use_map, mp_opt->ack64,
156 expected_opsize = TCPOLEN_MPTCP_DSS_BASE;
158 if (mp_opt->use_ack) {
160 expected_opsize += TCPOLEN_MPTCP_DSS_ACK64;
162 expected_opsize += TCPOLEN_MPTCP_DSS_ACK32;
165 if (mp_opt->use_map) {
167 expected_opsize += TCPOLEN_MPTCP_DSS_MAP64;
169 expected_opsize += TCPOLEN_MPTCP_DSS_MAP32;
172 /* RFC 6824, Section 3.3:
173 * If a checksum is present, but its use had
174 * not been negotiated in the MP_CAPABLE handshake,
175 * the checksum field MUST be ignored.
177 if (opsize != expected_opsize &&
178 opsize != expected_opsize + TCPOLEN_MPTCP_DSS_CHECKSUM)
183 if (mp_opt->use_ack) {
185 mp_opt->data_ack = get_unaligned_be64(ptr);
188 mp_opt->data_ack = get_unaligned_be32(ptr);
192 pr_debug("data_ack=%llu", mp_opt->data_ack);
195 if (mp_opt->use_map) {
197 mp_opt->data_seq = get_unaligned_be64(ptr);
200 mp_opt->data_seq = get_unaligned_be32(ptr);
204 mp_opt->subflow_seq = get_unaligned_be32(ptr);
207 mp_opt->data_len = get_unaligned_be16(ptr);
210 pr_debug("data_seq=%llu subflow_seq=%u data_len=%u",
211 mp_opt->data_seq, mp_opt->subflow_seq,
217 case MPTCPOPT_ADD_ADDR:
218 mp_opt->echo = (*ptr++) & MPTCP_ADDR_ECHO;
220 if (opsize == TCPOLEN_MPTCP_ADD_ADDR ||
221 opsize == TCPOLEN_MPTCP_ADD_ADDR_PORT)
222 mp_opt->family = MPTCP_ADDR_IPVERSION_4;
223 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
224 else if (opsize == TCPOLEN_MPTCP_ADD_ADDR6 ||
225 opsize == TCPOLEN_MPTCP_ADD_ADDR6_PORT)
226 mp_opt->family = MPTCP_ADDR_IPVERSION_6;
231 if (opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE ||
232 opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE_PORT)
233 mp_opt->family = MPTCP_ADDR_IPVERSION_4;
234 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
235 else if (opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE ||
236 opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE_PORT)
237 mp_opt->family = MPTCP_ADDR_IPVERSION_6;
243 mp_opt->add_addr = 1;
245 mp_opt->addr_id = *ptr++;
246 pr_debug("ADD_ADDR: id=%d, echo=%d", mp_opt->addr_id, mp_opt->echo);
247 if (mp_opt->family == MPTCP_ADDR_IPVERSION_4) {
248 memcpy((u8 *)&mp_opt->addr.s_addr, (u8 *)ptr, 4);
250 if (opsize == TCPOLEN_MPTCP_ADD_ADDR_PORT ||
251 opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE_PORT) {
252 mp_opt->port = get_unaligned_be16(ptr);
256 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
258 memcpy(mp_opt->addr6.s6_addr, (u8 *)ptr, 16);
260 if (opsize == TCPOLEN_MPTCP_ADD_ADDR6_PORT ||
261 opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE_PORT) {
262 mp_opt->port = get_unaligned_be16(ptr);
268 mp_opt->ahmac = get_unaligned_be64(ptr);
273 case MPTCPOPT_RM_ADDR:
274 if (opsize != TCPOLEN_MPTCP_RM_ADDR_BASE)
280 mp_opt->rm_id = *ptr++;
281 pr_debug("RM_ADDR: id=%d", mp_opt->rm_id);
289 void mptcp_get_options(const struct sk_buff *skb,
290 struct mptcp_options_received *mp_opt)
292 const struct tcphdr *th = tcp_hdr(skb);
293 const unsigned char *ptr;
296 /* initialize option status */
297 mp_opt->mp_capable = 0;
299 mp_opt->add_addr = 0;
303 length = (th->doff * 4) - sizeof(struct tcphdr);
304 ptr = (const unsigned char *)(th + 1);
313 case TCPOPT_NOP: /* Ref: RFC 793 section 3.1 */
318 if (opsize < 2) /* "silly options" */
321 return; /* don't parse partial options */
322 if (opcode == TCPOPT_MPTCP)
323 mptcp_parse_option(skb, ptr, opsize, mp_opt);
330 bool mptcp_syn_options(struct sock *sk, const struct sk_buff *skb,
331 unsigned int *size, struct mptcp_out_options *opts)
333 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
335 /* we will use snd_isn to detect first pkt [re]transmission
336 * in mptcp_established_options_mp()
338 subflow->snd_isn = TCP_SKB_CB(skb)->end_seq;
339 if (subflow->request_mptcp) {
340 opts->suboptions = OPTION_MPTCP_MPC_SYN;
341 *size = TCPOLEN_MPTCP_MPC_SYN;
343 } else if (subflow->request_join) {
344 pr_debug("remote_token=%u, nonce=%u", subflow->remote_token,
345 subflow->local_nonce);
346 opts->suboptions = OPTION_MPTCP_MPJ_SYN;
347 opts->join_id = subflow->local_id;
348 opts->token = subflow->remote_token;
349 opts->nonce = subflow->local_nonce;
350 opts->backup = subflow->request_bkup;
351 *size = TCPOLEN_MPTCP_MPJ_SYN;
357 /* MP_JOIN client subflow must wait for 4th ack before sending any data:
358 * TCP can't schedule delack timer before the subflow is fully established.
359 * MPTCP uses the delack timer to do 3rd ack retransmissions
361 static void schedule_3rdack_retransmission(struct sock *sk)
363 struct inet_connection_sock *icsk = inet_csk(sk);
364 struct tcp_sock *tp = tcp_sk(sk);
365 unsigned long timeout;
367 /* reschedule with a timeout above RTT, as we must look only for drop */
369 timeout = tp->srtt_us << 1;
371 timeout = TCP_TIMEOUT_INIT;
373 WARN_ON_ONCE(icsk->icsk_ack.pending & ICSK_ACK_TIMER);
374 icsk->icsk_ack.pending |= ICSK_ACK_SCHED | ICSK_ACK_TIMER;
375 icsk->icsk_ack.timeout = timeout;
376 sk_reset_timer(sk, &icsk->icsk_delack_timer, timeout);
379 static void clear_3rdack_retransmission(struct sock *sk)
381 struct inet_connection_sock *icsk = inet_csk(sk);
383 sk_stop_timer(sk, &icsk->icsk_delack_timer);
384 icsk->icsk_ack.timeout = 0;
385 icsk->icsk_ack.ato = 0;
386 icsk->icsk_ack.pending &= ~(ICSK_ACK_SCHED | ICSK_ACK_TIMER);
389 static bool mptcp_established_options_mp(struct sock *sk, struct sk_buff *skb,
391 unsigned int remaining,
392 struct mptcp_out_options *opts)
394 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
395 struct mptcp_ext *mpext;
396 unsigned int data_len;
398 /* When skb is not available, we better over-estimate the emitted
399 * options len. A full DSS option (28 bytes) is longer than
400 * TCPOLEN_MPTCP_MPC_ACK_DATA(22) or TCPOLEN_MPTCP_MPJ_ACK(24), so
401 * tell the caller to defer the estimate to
402 * mptcp_established_options_dss(), which will reserve enough space.
407 /* MPC/MPJ needed only on 3rd ack packet */
408 if (subflow->fully_established ||
409 subflow->snd_isn != TCP_SKB_CB(skb)->seq)
412 if (subflow->mp_capable) {
413 mpext = mptcp_get_ext(skb);
414 data_len = mpext ? mpext->data_len : 0;
416 /* we will check ext_copy.data_len in mptcp_write_options() to
417 * discriminate between TCPOLEN_MPTCP_MPC_ACK_DATA and
418 * TCPOLEN_MPTCP_MPC_ACK
420 opts->ext_copy.data_len = data_len;
421 opts->suboptions = OPTION_MPTCP_MPC_ACK;
422 opts->sndr_key = subflow->local_key;
423 opts->rcvr_key = subflow->remote_key;
426 * The MP_CAPABLE option is carried on the SYN, SYN/ACK, and ACK
427 * packets that start the first subflow of an MPTCP connection,
428 * as well as the first packet that carries data
431 *size = ALIGN(TCPOLEN_MPTCP_MPC_ACK_DATA, 4);
433 *size = TCPOLEN_MPTCP_MPC_ACK;
435 pr_debug("subflow=%p, local_key=%llu, remote_key=%llu map_len=%d",
436 subflow, subflow->local_key, subflow->remote_key,
440 } else if (subflow->mp_join) {
441 opts->suboptions = OPTION_MPTCP_MPJ_ACK;
442 memcpy(opts->hmac, subflow->hmac, MPTCPOPT_HMAC_LEN);
443 *size = TCPOLEN_MPTCP_MPJ_ACK;
444 pr_debug("subflow=%p", subflow);
446 schedule_3rdack_retransmission(sk);
452 static void mptcp_write_data_fin(struct mptcp_subflow_context *subflow,
453 struct sk_buff *skb, struct mptcp_ext *ext)
455 /* The write_seq value has already been incremented, so the actual
456 * sequence number for the DATA_FIN is one less.
458 u64 data_fin_tx_seq = READ_ONCE(mptcp_sk(subflow->conn)->write_seq) - 1;
460 if (!ext->use_map || !skb->len) {
461 /* RFC6824 requires a DSS mapping with specific values
462 * if DATA_FIN is set but no data payload is mapped
467 ext->data_seq = data_fin_tx_seq;
468 ext->subflow_seq = 0;
470 } else if (ext->data_seq + ext->data_len == data_fin_tx_seq) {
471 /* If there's an existing DSS mapping and it is the
472 * final mapping, DATA_FIN consumes 1 additional byte of
480 static bool mptcp_established_options_dss(struct sock *sk, struct sk_buff *skb,
482 unsigned int remaining,
483 struct mptcp_out_options *opts)
485 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
486 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
487 unsigned int dss_size = 0;
488 u64 snd_data_fin_enable;
489 struct mptcp_ext *mpext;
490 unsigned int ack_size;
493 mpext = skb ? mptcp_get_ext(skb) : NULL;
494 snd_data_fin_enable = READ_ONCE(msk->snd_data_fin_enable);
496 if (!skb || (mpext && mpext->use_map) || snd_data_fin_enable) {
497 unsigned int map_size;
499 map_size = TCPOLEN_MPTCP_DSS_BASE + TCPOLEN_MPTCP_DSS_MAP64;
501 remaining -= map_size;
504 opts->ext_copy = *mpext;
506 if (skb && snd_data_fin_enable)
507 mptcp_write_data_fin(subflow, skb, &opts->ext_copy);
511 /* passive sockets msk will set the 'can_ack' after accept(), even
512 * if the first subflow may have the already the remote key handy
514 opts->ext_copy.use_ack = 0;
515 if (!READ_ONCE(msk->can_ack)) {
516 *size = ALIGN(dss_size, 4);
520 if (READ_ONCE(msk->use_64bit_ack)) {
521 ack_size = TCPOLEN_MPTCP_DSS_ACK64;
522 opts->ext_copy.data_ack = READ_ONCE(msk->ack_seq);
523 opts->ext_copy.ack64 = 1;
525 ack_size = TCPOLEN_MPTCP_DSS_ACK32;
526 opts->ext_copy.data_ack32 = (uint32_t)READ_ONCE(msk->ack_seq);
527 opts->ext_copy.ack64 = 0;
529 opts->ext_copy.use_ack = 1;
531 /* Add kind/length/subtype/flag overhead if mapping is not populated */
533 ack_size += TCPOLEN_MPTCP_DSS_BASE;
535 dss_size += ack_size;
537 *size = ALIGN(dss_size, 4);
541 static u64 add_addr_generate_hmac(u64 key1, u64 key2, u8 addr_id,
542 struct in_addr *addr)
544 u8 hmac[SHA256_DIGEST_SIZE];
548 memcpy(&msg[1], &addr->s_addr, 4);
552 mptcp_crypto_hmac_sha(key1, key2, msg, 7, hmac);
554 return get_unaligned_be64(&hmac[SHA256_DIGEST_SIZE - sizeof(u64)]);
557 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
558 static u64 add_addr6_generate_hmac(u64 key1, u64 key2, u8 addr_id,
559 struct in6_addr *addr)
561 u8 hmac[SHA256_DIGEST_SIZE];
565 memcpy(&msg[1], &addr->s6_addr, 16);
569 mptcp_crypto_hmac_sha(key1, key2, msg, 19, hmac);
571 return get_unaligned_be64(&hmac[SHA256_DIGEST_SIZE - sizeof(u64)]);
575 static bool mptcp_established_options_add_addr(struct sock *sk,
577 unsigned int remaining,
578 struct mptcp_out_options *opts)
580 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
581 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
582 struct mptcp_addr_info saddr;
586 if (!mptcp_pm_should_add_signal(msk) ||
587 !(mptcp_pm_add_addr_signal(msk, remaining, &saddr, &echo)))
590 len = mptcp_add_addr_len(saddr.family, echo);
595 opts->addr_id = saddr.id;
596 if (saddr.family == AF_INET) {
597 opts->suboptions |= OPTION_MPTCP_ADD_ADDR;
598 opts->addr = saddr.addr;
600 opts->ahmac = add_addr_generate_hmac(msk->local_key,
606 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
607 else if (saddr.family == AF_INET6) {
608 opts->suboptions |= OPTION_MPTCP_ADD_ADDR6;
609 opts->addr6 = saddr.addr6;
611 opts->ahmac = add_addr6_generate_hmac(msk->local_key,
618 pr_debug("addr_id=%d, ahmac=%llu, echo=%d", opts->addr_id, opts->ahmac, echo);
623 static bool mptcp_established_options_rm_addr(struct sock *sk,
625 unsigned int remaining,
626 struct mptcp_out_options *opts)
628 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
629 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
632 if (!mptcp_pm_should_rm_signal(msk) ||
633 !(mptcp_pm_rm_addr_signal(msk, remaining, &rm_id)))
636 if (remaining < TCPOLEN_MPTCP_RM_ADDR_BASE)
639 *size = TCPOLEN_MPTCP_RM_ADDR_BASE;
640 opts->suboptions |= OPTION_MPTCP_RM_ADDR;
643 pr_debug("rm_id=%d", opts->rm_id);
648 bool mptcp_established_options(struct sock *sk, struct sk_buff *skb,
649 unsigned int *size, unsigned int remaining,
650 struct mptcp_out_options *opts)
652 unsigned int opt_size = 0;
655 opts->suboptions = 0;
657 if (unlikely(mptcp_check_fallback(sk)))
660 /* prevent adding of any MPTCP related options on reset packet
661 * until we support MP_TCPRST/MP_FASTCLOSE
663 if (unlikely(skb && TCP_SKB_CB(skb)->tcp_flags & TCPHDR_RST))
666 if (mptcp_established_options_mp(sk, skb, &opt_size, remaining, opts))
668 else if (mptcp_established_options_dss(sk, skb, &opt_size, remaining,
672 /* we reserved enough space for the above options, and exceeding the
673 * TCP option space would be fatal
675 if (WARN_ON_ONCE(opt_size > remaining))
679 remaining -= opt_size;
680 if (mptcp_established_options_add_addr(sk, &opt_size, remaining, opts)) {
682 remaining -= opt_size;
684 } else if (mptcp_established_options_rm_addr(sk, &opt_size, remaining, opts)) {
686 remaining -= opt_size;
693 bool mptcp_synack_options(const struct request_sock *req, unsigned int *size,
694 struct mptcp_out_options *opts)
696 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
698 if (subflow_req->mp_capable) {
699 opts->suboptions = OPTION_MPTCP_MPC_SYNACK;
700 opts->sndr_key = subflow_req->local_key;
701 *size = TCPOLEN_MPTCP_MPC_SYNACK;
702 pr_debug("subflow_req=%p, local_key=%llu",
703 subflow_req, subflow_req->local_key);
705 } else if (subflow_req->mp_join) {
706 opts->suboptions = OPTION_MPTCP_MPJ_SYNACK;
707 opts->backup = subflow_req->backup;
708 opts->join_id = subflow_req->local_id;
709 opts->thmac = subflow_req->thmac;
710 opts->nonce = subflow_req->local_nonce;
711 pr_debug("req=%p, bkup=%u, id=%u, thmac=%llu, nonce=%u",
712 subflow_req, opts->backup, opts->join_id,
713 opts->thmac, opts->nonce);
714 *size = TCPOLEN_MPTCP_MPJ_SYNACK;
720 static bool check_fully_established(struct mptcp_sock *msk, struct sock *ssk,
721 struct mptcp_subflow_context *subflow,
723 struct mptcp_options_received *mp_opt)
725 /* here we can process OoO, in-window pkts, only in-sequence 4th ack
726 * will make the subflow fully established
728 if (likely(subflow->fully_established)) {
729 /* on passive sockets, check for 3rd ack retransmission
730 * note that msk is always set by subflow_syn_recv_sock()
731 * for mp_join subflows
733 if (TCP_SKB_CB(skb)->seq == subflow->ssn_offset + 1 &&
734 TCP_SKB_CB(skb)->end_seq == TCP_SKB_CB(skb)->seq &&
735 subflow->mp_join && mp_opt->mp_join &&
736 READ_ONCE(msk->pm.server_side))
738 goto fully_established;
741 /* we must process OoO packets before the first subflow is fully
742 * established. OoO packets are instead a protocol violation
743 * for MP_JOIN subflows as the peer must not send any data
744 * before receiving the forth ack - cfr. RFC 8684 section 3.2.
746 if (TCP_SKB_CB(skb)->seq != subflow->ssn_offset + 1) {
747 if (subflow->mp_join)
749 return subflow->mp_capable;
752 if (mp_opt->dss && mp_opt->use_ack) {
753 /* subflows are fully established as soon as we get any
756 subflow->fully_established = 1;
757 WRITE_ONCE(msk->fully_established, true);
758 goto fully_established;
761 /* If the first established packet does not contain MP_CAPABLE + data
762 * then fallback to TCP. Fallback scenarios requires a reset for
765 if (!mp_opt->mp_capable) {
766 if (subflow->mp_join)
768 subflow->mp_capable = 0;
770 __mptcp_do_fallback(msk);
774 if (unlikely(!READ_ONCE(msk->pm.server_side)))
775 pr_warn_once("bogus mpc option on established client sk");
776 mptcp_subflow_fully_established(subflow, mp_opt);
779 if (likely(subflow->pm_notified))
782 subflow->pm_notified = 1;
783 if (subflow->mp_join) {
784 clear_3rdack_retransmission(ssk);
785 mptcp_pm_subflow_established(msk, subflow);
787 mptcp_pm_fully_established(msk);
792 mptcp_subflow_reset(ssk);
796 static u64 expand_ack(u64 old_ack, u64 cur_ack, bool use_64bit)
798 u32 old_ack32, cur_ack32;
803 old_ack32 = (u32)old_ack;
804 cur_ack32 = (u32)cur_ack;
805 cur_ack = (old_ack & GENMASK_ULL(63, 32)) + cur_ack32;
806 if (unlikely(before(cur_ack32, old_ack32)))
807 return cur_ack + (1LL << 32);
811 static void update_una(struct mptcp_sock *msk,
812 struct mptcp_options_received *mp_opt)
814 u64 new_snd_una, snd_una, old_snd_una = atomic64_read(&msk->snd_una);
815 u64 write_seq = READ_ONCE(msk->write_seq);
817 /* avoid ack expansion on update conflict, to reduce the risk of
818 * wrongly expanding to a future ack sequence number, which is way
819 * more dangerous than missing an ack
821 new_snd_una = expand_ack(old_snd_una, mp_opt->data_ack, mp_opt->ack64);
823 /* ACK for data not even sent yet? Ignore. */
824 if (after64(new_snd_una, write_seq))
825 new_snd_una = old_snd_una;
827 while (after64(new_snd_una, old_snd_una)) {
828 snd_una = old_snd_una;
829 old_snd_una = atomic64_cmpxchg(&msk->snd_una, snd_una,
831 if (old_snd_una == snd_una) {
832 mptcp_data_acked((struct sock *)msk);
838 bool mptcp_update_rcv_data_fin(struct mptcp_sock *msk, u64 data_fin_seq, bool use_64bit)
840 /* Skip if DATA_FIN was already received.
841 * If updating simultaneously with the recvmsg loop, values
842 * should match. If they mismatch, the peer is misbehaving and
843 * we will prefer the most recent information.
845 if (READ_ONCE(msk->rcv_data_fin) || !READ_ONCE(msk->first))
848 WRITE_ONCE(msk->rcv_data_fin_seq,
849 expand_ack(READ_ONCE(msk->ack_seq), data_fin_seq, use_64bit));
850 WRITE_ONCE(msk->rcv_data_fin, 1);
855 static bool add_addr_hmac_valid(struct mptcp_sock *msk,
856 struct mptcp_options_received *mp_opt)
863 if (mp_opt->family == MPTCP_ADDR_IPVERSION_4)
864 hmac = add_addr_generate_hmac(msk->remote_key,
866 mp_opt->addr_id, &mp_opt->addr);
867 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
869 hmac = add_addr6_generate_hmac(msk->remote_key,
871 mp_opt->addr_id, &mp_opt->addr6);
874 pr_debug("msk=%p, ahmac=%llu, mp_opt->ahmac=%llu\n",
875 msk, (unsigned long long)hmac,
876 (unsigned long long)mp_opt->ahmac);
878 return hmac == mp_opt->ahmac;
881 void mptcp_incoming_options(struct sock *sk, struct sk_buff *skb)
883 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
884 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
885 struct mptcp_options_received mp_opt;
886 struct mptcp_ext *mpext;
888 if (__mptcp_check_fallback(msk))
891 mptcp_get_options(skb, &mp_opt);
892 if (!check_fully_established(msk, sk, subflow, skb, &mp_opt))
895 if (mp_opt.add_addr && add_addr_hmac_valid(msk, &mp_opt)) {
896 struct mptcp_addr_info addr;
898 addr.port = htons(mp_opt.port);
899 addr.id = mp_opt.addr_id;
900 if (mp_opt.family == MPTCP_ADDR_IPVERSION_4) {
901 addr.family = AF_INET;
902 addr.addr = mp_opt.addr;
904 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
905 else if (mp_opt.family == MPTCP_ADDR_IPVERSION_6) {
906 addr.family = AF_INET6;
907 addr.addr6 = mp_opt.addr6;
911 mptcp_pm_add_addr_received(msk, &addr);
912 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ADDADDR);
914 mptcp_pm_del_add_timer(msk, &addr);
915 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ECHOADD);
920 if (mp_opt.rm_addr) {
921 mptcp_pm_rm_addr_received(msk, mp_opt.rm_id);
928 /* we can't wait for recvmsg() to update the ack_seq, otherwise
929 * monodirectional flows will stuck
932 update_una(msk, &mp_opt);
934 /* Zero-data-length packets are dropped by the caller and not
935 * propagated to the MPTCP layer, so the skb extension does not
936 * need to be allocated or populated. DATA_FIN information, if
937 * present, needs to be updated here before the skb is freed.
939 if (TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb)->end_seq) {
940 if (mp_opt.data_fin && mp_opt.data_len == 1 &&
941 mptcp_update_rcv_data_fin(msk, mp_opt.data_seq, mp_opt.dsn64) &&
942 schedule_work(&msk->work))
943 sock_hold(subflow->conn);
948 mpext = skb_ext_add(skb, SKB_EXT_MPTCP);
952 memset(mpext, 0, sizeof(*mpext));
954 if (mp_opt.use_map) {
955 if (mp_opt.mpc_map) {
956 /* this is an MP_CAPABLE carrying MPTCP data
957 * we know this map the first chunk of data
959 mptcp_crypto_key_sha(subflow->remote_key, NULL,
962 mpext->subflow_seq = 1;
967 mpext->data_seq = mp_opt.data_seq;
968 mpext->subflow_seq = mp_opt.subflow_seq;
969 mpext->dsn64 = mp_opt.dsn64;
970 mpext->data_fin = mp_opt.data_fin;
972 mpext->data_len = mp_opt.data_len;
977 void mptcp_write_options(__be32 *ptr, struct mptcp_out_options *opts)
979 if ((OPTION_MPTCP_MPC_SYN | OPTION_MPTCP_MPC_SYNACK |
980 OPTION_MPTCP_MPC_ACK) & opts->suboptions) {
983 if (OPTION_MPTCP_MPC_SYN & opts->suboptions)
984 len = TCPOLEN_MPTCP_MPC_SYN;
985 else if (OPTION_MPTCP_MPC_SYNACK & opts->suboptions)
986 len = TCPOLEN_MPTCP_MPC_SYNACK;
987 else if (opts->ext_copy.data_len)
988 len = TCPOLEN_MPTCP_MPC_ACK_DATA;
990 len = TCPOLEN_MPTCP_MPC_ACK;
992 *ptr++ = mptcp_option(MPTCPOPT_MP_CAPABLE, len,
993 MPTCP_SUPPORTED_VERSION,
994 MPTCP_CAP_HMAC_SHA256);
996 if (!((OPTION_MPTCP_MPC_SYNACK | OPTION_MPTCP_MPC_ACK) &
998 goto mp_capable_done;
1000 put_unaligned_be64(opts->sndr_key, ptr);
1002 if (!((OPTION_MPTCP_MPC_ACK) & opts->suboptions))
1003 goto mp_capable_done;
1005 put_unaligned_be64(opts->rcvr_key, ptr);
1007 if (!opts->ext_copy.data_len)
1008 goto mp_capable_done;
1010 put_unaligned_be32(opts->ext_copy.data_len << 16 |
1011 TCPOPT_NOP << 8 | TCPOPT_NOP, ptr);
1016 if (OPTION_MPTCP_ADD_ADDR & opts->suboptions) {
1018 *ptr++ = mptcp_option(MPTCPOPT_ADD_ADDR,
1019 TCPOLEN_MPTCP_ADD_ADDR, 0,
1022 *ptr++ = mptcp_option(MPTCPOPT_ADD_ADDR,
1023 TCPOLEN_MPTCP_ADD_ADDR_BASE,
1026 memcpy((u8 *)ptr, (u8 *)&opts->addr.s_addr, 4);
1029 put_unaligned_be64(opts->ahmac, ptr);
1034 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1035 if (OPTION_MPTCP_ADD_ADDR6 & opts->suboptions) {
1037 *ptr++ = mptcp_option(MPTCPOPT_ADD_ADDR,
1038 TCPOLEN_MPTCP_ADD_ADDR6, 0,
1041 *ptr++ = mptcp_option(MPTCPOPT_ADD_ADDR,
1042 TCPOLEN_MPTCP_ADD_ADDR6_BASE,
1045 memcpy((u8 *)ptr, opts->addr6.s6_addr, 16);
1048 put_unaligned_be64(opts->ahmac, ptr);
1054 if (OPTION_MPTCP_RM_ADDR & opts->suboptions) {
1055 *ptr++ = mptcp_option(MPTCPOPT_RM_ADDR,
1056 TCPOLEN_MPTCP_RM_ADDR_BASE,
1060 if (OPTION_MPTCP_MPJ_SYN & opts->suboptions) {
1061 *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1062 TCPOLEN_MPTCP_MPJ_SYN,
1063 opts->backup, opts->join_id);
1064 put_unaligned_be32(opts->token, ptr);
1066 put_unaligned_be32(opts->nonce, ptr);
1070 if (OPTION_MPTCP_MPJ_SYNACK & opts->suboptions) {
1071 *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1072 TCPOLEN_MPTCP_MPJ_SYNACK,
1073 opts->backup, opts->join_id);
1074 put_unaligned_be64(opts->thmac, ptr);
1076 put_unaligned_be32(opts->nonce, ptr);
1080 if (OPTION_MPTCP_MPJ_ACK & opts->suboptions) {
1081 *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1082 TCPOLEN_MPTCP_MPJ_ACK, 0, 0);
1083 memcpy(ptr, opts->hmac, MPTCPOPT_HMAC_LEN);
1087 if (opts->ext_copy.use_ack || opts->ext_copy.use_map) {
1088 struct mptcp_ext *mpext = &opts->ext_copy;
1089 u8 len = TCPOLEN_MPTCP_DSS_BASE;
1092 if (mpext->use_ack) {
1093 flags = MPTCP_DSS_HAS_ACK;
1095 len += TCPOLEN_MPTCP_DSS_ACK64;
1096 flags |= MPTCP_DSS_ACK64;
1098 len += TCPOLEN_MPTCP_DSS_ACK32;
1102 if (mpext->use_map) {
1103 len += TCPOLEN_MPTCP_DSS_MAP64;
1105 /* Use only 64-bit mapping flags for now, add
1106 * support for optional 32-bit mappings later.
1108 flags |= MPTCP_DSS_HAS_MAP | MPTCP_DSS_DSN64;
1109 if (mpext->data_fin)
1110 flags |= MPTCP_DSS_DATA_FIN;
1113 *ptr++ = mptcp_option(MPTCPOPT_DSS, len, 0, flags);
1115 if (mpext->use_ack) {
1117 put_unaligned_be64(mpext->data_ack, ptr);
1120 put_unaligned_be32(mpext->data_ack32, ptr);
1125 if (mpext->use_map) {
1126 put_unaligned_be64(mpext->data_seq, ptr);
1128 put_unaligned_be32(mpext->subflow_seq, ptr);
1130 put_unaligned_be32(mpext->data_len << 16 |
1131 TCPOPT_NOP << 8 | TCPOPT_NOP, ptr);