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/sha2.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;
244 mp_opt->addr_id = *ptr++;
245 if (mp_opt->family == MPTCP_ADDR_IPVERSION_4) {
246 memcpy((u8 *)&mp_opt->addr.s_addr, (u8 *)ptr, 4);
248 if (opsize == TCPOLEN_MPTCP_ADD_ADDR_PORT ||
249 opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE_PORT) {
250 mp_opt->port = get_unaligned_be16(ptr);
254 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
256 memcpy(mp_opt->addr6.s6_addr, (u8 *)ptr, 16);
258 if (opsize == TCPOLEN_MPTCP_ADD_ADDR6_PORT ||
259 opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE_PORT) {
260 mp_opt->port = get_unaligned_be16(ptr);
266 mp_opt->ahmac = get_unaligned_be64(ptr);
269 pr_debug("ADD_ADDR%s: id=%d, ahmac=%llu, echo=%d, port=%d",
270 (mp_opt->family == MPTCP_ADDR_IPVERSION_6) ? "6" : "",
271 mp_opt->addr_id, mp_opt->ahmac, mp_opt->echo, mp_opt->port);
274 case MPTCPOPT_RM_ADDR:
275 if (opsize != TCPOLEN_MPTCP_RM_ADDR_BASE)
281 mp_opt->rm_id = *ptr++;
282 pr_debug("RM_ADDR: id=%d", mp_opt->rm_id);
285 case MPTCPOPT_MP_PRIO:
286 if (opsize != TCPOLEN_MPTCP_PRIO)
290 mp_opt->backup = *ptr++ & MPTCP_PRIO_BKUP;
291 pr_debug("MP_PRIO: prio=%d", mp_opt->backup);
294 case MPTCPOPT_MP_FASTCLOSE:
295 if (opsize != TCPOLEN_MPTCP_FASTCLOSE)
299 mp_opt->rcvr_key = get_unaligned_be64(ptr);
301 mp_opt->fastclose = 1;
309 void mptcp_get_options(const struct sk_buff *skb,
310 struct mptcp_options_received *mp_opt)
312 const struct tcphdr *th = tcp_hdr(skb);
313 const unsigned char *ptr;
316 /* initialize option status */
317 mp_opt->mp_capable = 0;
319 mp_opt->add_addr = 0;
321 mp_opt->fastclose = 0;
327 length = (th->doff * 4) - sizeof(struct tcphdr);
328 ptr = (const unsigned char *)(th + 1);
337 case TCPOPT_NOP: /* Ref: RFC 793 section 3.1 */
342 if (opsize < 2) /* "silly options" */
345 return; /* don't parse partial options */
346 if (opcode == TCPOPT_MPTCP)
347 mptcp_parse_option(skb, ptr, opsize, mp_opt);
354 bool mptcp_syn_options(struct sock *sk, const struct sk_buff *skb,
355 unsigned int *size, struct mptcp_out_options *opts)
357 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
359 /* we will use snd_isn to detect first pkt [re]transmission
360 * in mptcp_established_options_mp()
362 subflow->snd_isn = TCP_SKB_CB(skb)->end_seq;
363 if (subflow->request_mptcp) {
364 opts->suboptions = OPTION_MPTCP_MPC_SYN;
365 *size = TCPOLEN_MPTCP_MPC_SYN;
367 } else if (subflow->request_join) {
368 pr_debug("remote_token=%u, nonce=%u", subflow->remote_token,
369 subflow->local_nonce);
370 opts->suboptions = OPTION_MPTCP_MPJ_SYN;
371 opts->join_id = subflow->local_id;
372 opts->token = subflow->remote_token;
373 opts->nonce = subflow->local_nonce;
374 opts->backup = subflow->request_bkup;
375 *size = TCPOLEN_MPTCP_MPJ_SYN;
381 /* MP_JOIN client subflow must wait for 4th ack before sending any data:
382 * TCP can't schedule delack timer before the subflow is fully established.
383 * MPTCP uses the delack timer to do 3rd ack retransmissions
385 static void schedule_3rdack_retransmission(struct sock *sk)
387 struct inet_connection_sock *icsk = inet_csk(sk);
388 struct tcp_sock *tp = tcp_sk(sk);
389 unsigned long timeout;
391 /* reschedule with a timeout above RTT, as we must look only for drop */
393 timeout = tp->srtt_us << 1;
395 timeout = TCP_TIMEOUT_INIT;
397 WARN_ON_ONCE(icsk->icsk_ack.pending & ICSK_ACK_TIMER);
398 icsk->icsk_ack.pending |= ICSK_ACK_SCHED | ICSK_ACK_TIMER;
399 icsk->icsk_ack.timeout = timeout;
400 sk_reset_timer(sk, &icsk->icsk_delack_timer, timeout);
403 static void clear_3rdack_retransmission(struct sock *sk)
405 struct inet_connection_sock *icsk = inet_csk(sk);
407 sk_stop_timer(sk, &icsk->icsk_delack_timer);
408 icsk->icsk_ack.timeout = 0;
409 icsk->icsk_ack.ato = 0;
410 icsk->icsk_ack.pending &= ~(ICSK_ACK_SCHED | ICSK_ACK_TIMER);
413 static bool mptcp_established_options_mp(struct sock *sk, struct sk_buff *skb,
415 unsigned int remaining,
416 struct mptcp_out_options *opts)
418 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
419 struct mptcp_ext *mpext;
420 unsigned int data_len;
422 /* When skb is not available, we better over-estimate the emitted
423 * options len. A full DSS option (28 bytes) is longer than
424 * TCPOLEN_MPTCP_MPC_ACK_DATA(22) or TCPOLEN_MPTCP_MPJ_ACK(24), so
425 * tell the caller to defer the estimate to
426 * mptcp_established_options_dss(), which will reserve enough space.
431 /* MPC/MPJ needed only on 3rd ack packet */
432 if (subflow->fully_established ||
433 subflow->snd_isn != TCP_SKB_CB(skb)->seq)
436 if (subflow->mp_capable) {
437 mpext = mptcp_get_ext(skb);
438 data_len = mpext ? mpext->data_len : 0;
440 /* we will check ext_copy.data_len in mptcp_write_options() to
441 * discriminate between TCPOLEN_MPTCP_MPC_ACK_DATA and
442 * TCPOLEN_MPTCP_MPC_ACK
444 opts->ext_copy.data_len = data_len;
445 opts->suboptions = OPTION_MPTCP_MPC_ACK;
446 opts->sndr_key = subflow->local_key;
447 opts->rcvr_key = subflow->remote_key;
450 * The MP_CAPABLE option is carried on the SYN, SYN/ACK, and ACK
451 * packets that start the first subflow of an MPTCP connection,
452 * as well as the first packet that carries data
455 *size = ALIGN(TCPOLEN_MPTCP_MPC_ACK_DATA, 4);
457 *size = TCPOLEN_MPTCP_MPC_ACK;
459 pr_debug("subflow=%p, local_key=%llu, remote_key=%llu map_len=%d",
460 subflow, subflow->local_key, subflow->remote_key,
464 } else if (subflow->mp_join) {
465 opts->suboptions = OPTION_MPTCP_MPJ_ACK;
466 memcpy(opts->hmac, subflow->hmac, MPTCPOPT_HMAC_LEN);
467 *size = TCPOLEN_MPTCP_MPJ_ACK;
468 pr_debug("subflow=%p", subflow);
470 schedule_3rdack_retransmission(sk);
476 static void mptcp_write_data_fin(struct mptcp_subflow_context *subflow,
477 struct sk_buff *skb, struct mptcp_ext *ext)
479 /* The write_seq value has already been incremented, so the actual
480 * sequence number for the DATA_FIN is one less.
482 u64 data_fin_tx_seq = READ_ONCE(mptcp_sk(subflow->conn)->write_seq) - 1;
484 if (!ext->use_map || !skb->len) {
485 /* RFC6824 requires a DSS mapping with specific values
486 * if DATA_FIN is set but no data payload is mapped
491 ext->data_seq = data_fin_tx_seq;
492 ext->subflow_seq = 0;
494 } else if (ext->data_seq + ext->data_len == data_fin_tx_seq) {
495 /* If there's an existing DSS mapping and it is the
496 * final mapping, DATA_FIN consumes 1 additional byte of
504 static bool mptcp_established_options_dss(struct sock *sk, struct sk_buff *skb,
506 unsigned int remaining,
507 struct mptcp_out_options *opts)
509 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
510 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
511 u64 snd_data_fin_enable, ack_seq;
512 unsigned int dss_size = 0;
513 struct mptcp_ext *mpext;
514 unsigned int ack_size;
517 mpext = skb ? mptcp_get_ext(skb) : NULL;
518 snd_data_fin_enable = mptcp_data_fin_enabled(msk);
520 if (!skb || (mpext && mpext->use_map) || snd_data_fin_enable) {
521 unsigned int map_size;
523 map_size = TCPOLEN_MPTCP_DSS_BASE + TCPOLEN_MPTCP_DSS_MAP64;
525 remaining -= map_size;
528 opts->ext_copy = *mpext;
530 if (skb && snd_data_fin_enable)
531 mptcp_write_data_fin(subflow, skb, &opts->ext_copy);
535 /* passive sockets msk will set the 'can_ack' after accept(), even
536 * if the first subflow may have the already the remote key handy
538 opts->ext_copy.use_ack = 0;
539 if (!READ_ONCE(msk->can_ack)) {
540 *size = ALIGN(dss_size, 4);
544 ack_seq = READ_ONCE(msk->ack_seq);
545 if (READ_ONCE(msk->use_64bit_ack)) {
546 ack_size = TCPOLEN_MPTCP_DSS_ACK64;
547 opts->ext_copy.data_ack = ack_seq;
548 opts->ext_copy.ack64 = 1;
550 ack_size = TCPOLEN_MPTCP_DSS_ACK32;
551 opts->ext_copy.data_ack32 = (uint32_t)ack_seq;
552 opts->ext_copy.ack64 = 0;
554 opts->ext_copy.use_ack = 1;
555 WRITE_ONCE(msk->old_wspace, __mptcp_space((struct sock *)msk));
557 /* Add kind/length/subtype/flag overhead if mapping is not populated */
559 ack_size += TCPOLEN_MPTCP_DSS_BASE;
561 dss_size += ack_size;
563 *size = ALIGN(dss_size, 4);
567 static u64 add_addr_generate_hmac(u64 key1, u64 key2, u8 addr_id,
568 struct in_addr *addr)
570 u8 hmac[SHA256_DIGEST_SIZE];
574 memcpy(&msg[1], &addr->s_addr, 4);
578 mptcp_crypto_hmac_sha(key1, key2, msg, 7, hmac);
580 return get_unaligned_be64(&hmac[SHA256_DIGEST_SIZE - sizeof(u64)]);
583 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
584 static u64 add_addr6_generate_hmac(u64 key1, u64 key2, u8 addr_id,
585 struct in6_addr *addr)
587 u8 hmac[SHA256_DIGEST_SIZE];
591 memcpy(&msg[1], &addr->s6_addr, 16);
595 mptcp_crypto_hmac_sha(key1, key2, msg, 19, hmac);
597 return get_unaligned_be64(&hmac[SHA256_DIGEST_SIZE - sizeof(u64)]);
601 static bool mptcp_established_options_add_addr(struct sock *sk, struct sk_buff *skb,
603 unsigned int remaining,
604 struct mptcp_out_options *opts)
606 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
607 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
608 bool drop_other_suboptions = false;
609 unsigned int opt_size = *size;
610 struct mptcp_addr_info saddr;
615 if ((mptcp_pm_should_add_signal_ipv6(msk) ||
616 mptcp_pm_should_add_signal_port(msk)) &&
617 skb && skb_is_tcp_pure_ack(skb)) {
618 pr_debug("drop other suboptions");
619 opts->suboptions = 0;
620 opts->ext_copy.use_ack = 0;
621 opts->ext_copy.use_map = 0;
622 remaining += opt_size;
623 drop_other_suboptions = true;
626 if (!mptcp_pm_should_add_signal(msk) ||
627 !(mptcp_pm_add_addr_signal(msk, remaining, &saddr, &echo, &port)))
630 len = mptcp_add_addr_len(saddr.family, echo, port);
635 if (drop_other_suboptions)
637 opts->addr_id = saddr.id;
639 opts->port = ntohs(saddr.port);
640 if (saddr.family == AF_INET) {
641 opts->suboptions |= OPTION_MPTCP_ADD_ADDR;
642 opts->addr = saddr.addr;
644 opts->ahmac = add_addr_generate_hmac(msk->local_key,
650 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
651 else if (saddr.family == AF_INET6) {
652 opts->suboptions |= OPTION_MPTCP_ADD_ADDR6;
653 opts->addr6 = saddr.addr6;
655 opts->ahmac = add_addr6_generate_hmac(msk->local_key,
662 pr_debug("addr_id=%d, ahmac=%llu, echo=%d, port=%d",
663 opts->addr_id, opts->ahmac, echo, opts->port);
668 static bool mptcp_established_options_rm_addr(struct sock *sk,
670 unsigned int remaining,
671 struct mptcp_out_options *opts)
673 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
674 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
677 if (!mptcp_pm_should_rm_signal(msk) ||
678 !(mptcp_pm_rm_addr_signal(msk, remaining, &rm_id)))
681 if (remaining < TCPOLEN_MPTCP_RM_ADDR_BASE)
684 *size = TCPOLEN_MPTCP_RM_ADDR_BASE;
685 opts->suboptions |= OPTION_MPTCP_RM_ADDR;
688 pr_debug("rm_id=%d", opts->rm_id);
693 static bool mptcp_established_options_mp_prio(struct sock *sk,
695 unsigned int remaining,
696 struct mptcp_out_options *opts)
698 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
700 if (!subflow->send_mp_prio)
703 /* account for the trailing 'nop' option */
704 if (remaining < TCPOLEN_MPTCP_PRIO_ALIGN)
707 *size = TCPOLEN_MPTCP_PRIO_ALIGN;
708 opts->suboptions |= OPTION_MPTCP_PRIO;
709 opts->backup = subflow->request_bkup;
711 pr_debug("prio=%d", opts->backup);
716 bool mptcp_established_options(struct sock *sk, struct sk_buff *skb,
717 unsigned int *size, unsigned int remaining,
718 struct mptcp_out_options *opts)
720 unsigned int opt_size = 0;
723 opts->suboptions = 0;
725 if (unlikely(mptcp_check_fallback(sk)))
728 /* prevent adding of any MPTCP related options on reset packet
729 * until we support MP_TCPRST/MP_FASTCLOSE
731 if (unlikely(skb && TCP_SKB_CB(skb)->tcp_flags & TCPHDR_RST))
734 if (mptcp_established_options_mp(sk, skb, &opt_size, remaining, opts))
736 else if (mptcp_established_options_dss(sk, skb, &opt_size, remaining,
740 /* we reserved enough space for the above options, and exceeding the
741 * TCP option space would be fatal
743 if (WARN_ON_ONCE(opt_size > remaining))
747 remaining -= opt_size;
748 if (mptcp_established_options_add_addr(sk, skb, &opt_size, remaining, opts)) {
750 remaining -= opt_size;
752 } else if (mptcp_established_options_rm_addr(sk, &opt_size, remaining, opts)) {
754 remaining -= opt_size;
758 if (mptcp_established_options_mp_prio(sk, &opt_size, remaining, opts)) {
760 remaining -= opt_size;
767 bool mptcp_synack_options(const struct request_sock *req, unsigned int *size,
768 struct mptcp_out_options *opts)
770 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
772 if (subflow_req->mp_capable) {
773 opts->suboptions = OPTION_MPTCP_MPC_SYNACK;
774 opts->sndr_key = subflow_req->local_key;
775 *size = TCPOLEN_MPTCP_MPC_SYNACK;
776 pr_debug("subflow_req=%p, local_key=%llu",
777 subflow_req, subflow_req->local_key);
779 } else if (subflow_req->mp_join) {
780 opts->suboptions = OPTION_MPTCP_MPJ_SYNACK;
781 opts->backup = subflow_req->backup;
782 opts->join_id = subflow_req->local_id;
783 opts->thmac = subflow_req->thmac;
784 opts->nonce = subflow_req->local_nonce;
785 pr_debug("req=%p, bkup=%u, id=%u, thmac=%llu, nonce=%u",
786 subflow_req, opts->backup, opts->join_id,
787 opts->thmac, opts->nonce);
788 *size = TCPOLEN_MPTCP_MPJ_SYNACK;
794 static bool check_fully_established(struct mptcp_sock *msk, struct sock *ssk,
795 struct mptcp_subflow_context *subflow,
797 struct mptcp_options_received *mp_opt)
799 /* here we can process OoO, in-window pkts, only in-sequence 4th ack
800 * will make the subflow fully established
802 if (likely(subflow->fully_established)) {
803 /* on passive sockets, check for 3rd ack retransmission
804 * note that msk is always set by subflow_syn_recv_sock()
805 * for mp_join subflows
807 if (TCP_SKB_CB(skb)->seq == subflow->ssn_offset + 1 &&
808 TCP_SKB_CB(skb)->end_seq == TCP_SKB_CB(skb)->seq &&
809 subflow->mp_join && mp_opt->mp_join &&
810 READ_ONCE(msk->pm.server_side))
812 goto fully_established;
815 /* we must process OoO packets before the first subflow is fully
816 * established. OoO packets are instead a protocol violation
817 * for MP_JOIN subflows as the peer must not send any data
818 * before receiving the forth ack - cfr. RFC 8684 section 3.2.
820 if (TCP_SKB_CB(skb)->seq != subflow->ssn_offset + 1) {
821 if (subflow->mp_join)
823 return subflow->mp_capable;
826 if (mp_opt->dss && mp_opt->use_ack) {
827 /* subflows are fully established as soon as we get any
830 subflow->fully_established = 1;
831 WRITE_ONCE(msk->fully_established, true);
832 goto fully_established;
835 if (mp_opt->add_addr) {
836 WRITE_ONCE(msk->fully_established, true);
840 /* If the first established packet does not contain MP_CAPABLE + data
841 * then fallback to TCP. Fallback scenarios requires a reset for
844 if (!mp_opt->mp_capable) {
845 if (subflow->mp_join)
847 subflow->mp_capable = 0;
849 __mptcp_do_fallback(msk);
853 if (unlikely(!READ_ONCE(msk->pm.server_side)))
854 pr_warn_once("bogus mpc option on established client sk");
855 mptcp_subflow_fully_established(subflow, mp_opt);
858 /* if the subflow is not already linked into the conn_list, we can't
859 * notify the PM: this subflow is still on the listener queue
860 * and the PM possibly acquiring the subflow lock could race with
863 if (likely(subflow->pm_notified) || list_empty(&subflow->node))
866 subflow->pm_notified = 1;
867 if (subflow->mp_join) {
868 clear_3rdack_retransmission(ssk);
869 mptcp_pm_subflow_established(msk, subflow);
871 mptcp_pm_fully_established(msk, ssk, GFP_ATOMIC);
876 mptcp_subflow_reset(ssk);
880 static u64 expand_ack(u64 old_ack, u64 cur_ack, bool use_64bit)
882 u32 old_ack32, cur_ack32;
887 old_ack32 = (u32)old_ack;
888 cur_ack32 = (u32)cur_ack;
889 cur_ack = (old_ack & GENMASK_ULL(63, 32)) + cur_ack32;
890 if (unlikely(before(cur_ack32, old_ack32)))
891 return cur_ack + (1LL << 32);
895 static void ack_update_msk(struct mptcp_sock *msk,
897 struct mptcp_options_received *mp_opt)
899 u64 new_wnd_end, new_snd_una, snd_nxt = READ_ONCE(msk->snd_nxt);
900 struct sock *sk = (struct sock *)msk;
905 /* avoid ack expansion on update conflict, to reduce the risk of
906 * wrongly expanding to a future ack sequence number, which is way
907 * more dangerous than missing an ack
909 old_snd_una = msk->snd_una;
910 new_snd_una = expand_ack(old_snd_una, mp_opt->data_ack, mp_opt->ack64);
912 /* ACK for data not even sent yet? Ignore. */
913 if (after64(new_snd_una, snd_nxt))
914 new_snd_una = old_snd_una;
916 new_wnd_end = new_snd_una + tcp_sk(ssk)->snd_wnd;
918 if (after64(new_wnd_end, msk->wnd_end))
919 msk->wnd_end = new_wnd_end;
921 /* this assumes mptcp_incoming_options() is invoked after tcp_ack() */
922 if (after64(msk->wnd_end, READ_ONCE(msk->snd_nxt)))
923 __mptcp_check_push(sk, ssk);
925 if (after64(new_snd_una, old_snd_una)) {
926 msk->snd_una = new_snd_una;
927 __mptcp_data_acked(sk);
929 mptcp_data_unlock(sk);
932 bool mptcp_update_rcv_data_fin(struct mptcp_sock *msk, u64 data_fin_seq, bool use_64bit)
934 /* Skip if DATA_FIN was already received.
935 * If updating simultaneously with the recvmsg loop, values
936 * should match. If they mismatch, the peer is misbehaving and
937 * we will prefer the most recent information.
939 if (READ_ONCE(msk->rcv_data_fin) || !READ_ONCE(msk->first))
942 WRITE_ONCE(msk->rcv_data_fin_seq,
943 expand_ack(READ_ONCE(msk->ack_seq), data_fin_seq, use_64bit));
944 WRITE_ONCE(msk->rcv_data_fin, 1);
949 static bool add_addr_hmac_valid(struct mptcp_sock *msk,
950 struct mptcp_options_received *mp_opt)
957 if (mp_opt->family == MPTCP_ADDR_IPVERSION_4)
958 hmac = add_addr_generate_hmac(msk->remote_key,
960 mp_opt->addr_id, &mp_opt->addr);
961 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
963 hmac = add_addr6_generate_hmac(msk->remote_key,
965 mp_opt->addr_id, &mp_opt->addr6);
968 pr_debug("msk=%p, ahmac=%llu, mp_opt->ahmac=%llu\n",
969 msk, (unsigned long long)hmac,
970 (unsigned long long)mp_opt->ahmac);
972 return hmac == mp_opt->ahmac;
975 void mptcp_incoming_options(struct sock *sk, struct sk_buff *skb)
977 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
978 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
979 struct mptcp_options_received mp_opt;
980 struct mptcp_ext *mpext;
982 if (__mptcp_check_fallback(msk)) {
983 /* Keep it simple and unconditionally trigger send data cleanup and
984 * pending queue spooling. We will need to acquire the data lock
985 * for more accurate checks, and once the lock is acquired, such
988 mptcp_data_lock(subflow->conn);
989 if (sk_stream_memory_free(sk))
990 __mptcp_check_push(subflow->conn, sk);
991 __mptcp_data_acked(subflow->conn);
992 mptcp_data_unlock(subflow->conn);
996 mptcp_get_options(skb, &mp_opt);
997 if (!check_fully_established(msk, sk, subflow, skb, &mp_opt))
1000 if (mp_opt.fastclose &&
1001 msk->local_key == mp_opt.rcvr_key) {
1002 WRITE_ONCE(msk->rcv_fastclose, true);
1003 mptcp_schedule_work((struct sock *)msk);
1006 if (mp_opt.add_addr && add_addr_hmac_valid(msk, &mp_opt)) {
1007 struct mptcp_addr_info addr;
1009 addr.port = htons(mp_opt.port);
1010 addr.id = mp_opt.addr_id;
1011 if (mp_opt.family == MPTCP_ADDR_IPVERSION_4) {
1012 addr.family = AF_INET;
1013 addr.addr = mp_opt.addr;
1015 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1016 else if (mp_opt.family == MPTCP_ADDR_IPVERSION_6) {
1017 addr.family = AF_INET6;
1018 addr.addr6 = mp_opt.addr6;
1022 mptcp_pm_add_addr_received(msk, &addr);
1023 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ADDADDR);
1025 mptcp_pm_del_add_timer(msk, &addr);
1026 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ECHOADD);
1030 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_PORTADD);
1032 mp_opt.add_addr = 0;
1035 if (mp_opt.rm_addr) {
1036 mptcp_pm_rm_addr_received(msk, mp_opt.rm_id);
1040 if (mp_opt.mp_prio) {
1041 mptcp_pm_mp_prio_received(sk, mp_opt.backup);
1042 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPPRIORX);
1049 /* we can't wait for recvmsg() to update the ack_seq, otherwise
1050 * monodirectional flows will stuck
1053 ack_update_msk(msk, sk, &mp_opt);
1055 /* Zero-data-length packets are dropped by the caller and not
1056 * propagated to the MPTCP layer, so the skb extension does not
1057 * need to be allocated or populated. DATA_FIN information, if
1058 * present, needs to be updated here before the skb is freed.
1060 if (TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb)->end_seq) {
1061 if (mp_opt.data_fin && mp_opt.data_len == 1 &&
1062 mptcp_update_rcv_data_fin(msk, mp_opt.data_seq, mp_opt.dsn64) &&
1063 schedule_work(&msk->work))
1064 sock_hold(subflow->conn);
1069 mpext = skb_ext_add(skb, SKB_EXT_MPTCP);
1073 memset(mpext, 0, sizeof(*mpext));
1075 if (mp_opt.use_map) {
1076 if (mp_opt.mpc_map) {
1077 /* this is an MP_CAPABLE carrying MPTCP data
1078 * we know this map the first chunk of data
1080 mptcp_crypto_key_sha(subflow->remote_key, NULL,
1083 mpext->subflow_seq = 1;
1086 mpext->data_fin = 0;
1088 mpext->data_seq = mp_opt.data_seq;
1089 mpext->subflow_seq = mp_opt.subflow_seq;
1090 mpext->dsn64 = mp_opt.dsn64;
1091 mpext->data_fin = mp_opt.data_fin;
1093 mpext->data_len = mp_opt.data_len;
1098 static void mptcp_set_rwin(const struct tcp_sock *tp)
1100 const struct sock *ssk = (const struct sock *)tp;
1101 const struct mptcp_subflow_context *subflow;
1102 struct mptcp_sock *msk;
1105 subflow = mptcp_subflow_ctx(ssk);
1106 msk = mptcp_sk(subflow->conn);
1108 ack_seq = READ_ONCE(msk->ack_seq) + tp->rcv_wnd;
1110 if (after64(ack_seq, READ_ONCE(msk->rcv_wnd_sent)))
1111 WRITE_ONCE(msk->rcv_wnd_sent, ack_seq);
1114 void mptcp_write_options(__be32 *ptr, const struct tcp_sock *tp,
1115 struct mptcp_out_options *opts)
1117 if ((OPTION_MPTCP_MPC_SYN | OPTION_MPTCP_MPC_SYNACK |
1118 OPTION_MPTCP_MPC_ACK) & opts->suboptions) {
1121 if (OPTION_MPTCP_MPC_SYN & opts->suboptions)
1122 len = TCPOLEN_MPTCP_MPC_SYN;
1123 else if (OPTION_MPTCP_MPC_SYNACK & opts->suboptions)
1124 len = TCPOLEN_MPTCP_MPC_SYNACK;
1125 else if (opts->ext_copy.data_len)
1126 len = TCPOLEN_MPTCP_MPC_ACK_DATA;
1128 len = TCPOLEN_MPTCP_MPC_ACK;
1130 *ptr++ = mptcp_option(MPTCPOPT_MP_CAPABLE, len,
1131 MPTCP_SUPPORTED_VERSION,
1132 MPTCP_CAP_HMAC_SHA256);
1134 if (!((OPTION_MPTCP_MPC_SYNACK | OPTION_MPTCP_MPC_ACK) &
1136 goto mp_capable_done;
1138 put_unaligned_be64(opts->sndr_key, ptr);
1140 if (!((OPTION_MPTCP_MPC_ACK) & opts->suboptions))
1141 goto mp_capable_done;
1143 put_unaligned_be64(opts->rcvr_key, ptr);
1145 if (!opts->ext_copy.data_len)
1146 goto mp_capable_done;
1148 put_unaligned_be32(opts->ext_copy.data_len << 16 |
1149 TCPOPT_NOP << 8 | TCPOPT_NOP, ptr);
1154 if ((OPTION_MPTCP_ADD_ADDR
1155 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1156 | OPTION_MPTCP_ADD_ADDR6
1158 ) & opts->suboptions) {
1159 u8 len = TCPOLEN_MPTCP_ADD_ADDR_BASE;
1160 u8 echo = MPTCP_ADDR_ECHO;
1162 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1163 if (OPTION_MPTCP_ADD_ADDR6 & opts->suboptions)
1164 len = TCPOLEN_MPTCP_ADD_ADDR6_BASE;
1168 len += TCPOLEN_MPTCP_PORT_LEN;
1171 len += sizeof(opts->ahmac);
1175 *ptr++ = mptcp_option(MPTCPOPT_ADD_ADDR,
1176 len, echo, opts->addr_id);
1177 if (OPTION_MPTCP_ADD_ADDR & opts->suboptions) {
1178 memcpy((u8 *)ptr, (u8 *)&opts->addr.s_addr, 4);
1181 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1182 else if (OPTION_MPTCP_ADD_ADDR6 & opts->suboptions) {
1183 memcpy((u8 *)ptr, opts->addr6.s6_addr, 16);
1190 put_unaligned_be64(opts->ahmac, ptr);
1195 u8 *bptr = (u8 *)ptr;
1197 put_unaligned_be16(opts->port, bptr);
1199 put_unaligned_be64(opts->ahmac, bptr);
1201 put_unaligned_be16(TCPOPT_NOP << 8 |
1206 put_unaligned_be32(opts->port << 16 |
1214 if (OPTION_MPTCP_RM_ADDR & opts->suboptions) {
1215 *ptr++ = mptcp_option(MPTCPOPT_RM_ADDR,
1216 TCPOLEN_MPTCP_RM_ADDR_BASE,
1220 if (OPTION_MPTCP_PRIO & opts->suboptions) {
1221 const struct sock *ssk = (const struct sock *)tp;
1222 struct mptcp_subflow_context *subflow;
1224 subflow = mptcp_subflow_ctx(ssk);
1225 subflow->send_mp_prio = 0;
1227 *ptr++ = mptcp_option(MPTCPOPT_MP_PRIO,
1229 opts->backup, TCPOPT_NOP);
1232 if (OPTION_MPTCP_MPJ_SYN & opts->suboptions) {
1233 *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1234 TCPOLEN_MPTCP_MPJ_SYN,
1235 opts->backup, opts->join_id);
1236 put_unaligned_be32(opts->token, ptr);
1238 put_unaligned_be32(opts->nonce, ptr);
1242 if (OPTION_MPTCP_MPJ_SYNACK & opts->suboptions) {
1243 *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1244 TCPOLEN_MPTCP_MPJ_SYNACK,
1245 opts->backup, opts->join_id);
1246 put_unaligned_be64(opts->thmac, ptr);
1248 put_unaligned_be32(opts->nonce, ptr);
1252 if (OPTION_MPTCP_MPJ_ACK & opts->suboptions) {
1253 *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1254 TCPOLEN_MPTCP_MPJ_ACK, 0, 0);
1255 memcpy(ptr, opts->hmac, MPTCPOPT_HMAC_LEN);
1259 if (opts->ext_copy.use_ack || opts->ext_copy.use_map) {
1260 struct mptcp_ext *mpext = &opts->ext_copy;
1261 u8 len = TCPOLEN_MPTCP_DSS_BASE;
1264 if (mpext->use_ack) {
1265 flags = MPTCP_DSS_HAS_ACK;
1267 len += TCPOLEN_MPTCP_DSS_ACK64;
1268 flags |= MPTCP_DSS_ACK64;
1270 len += TCPOLEN_MPTCP_DSS_ACK32;
1274 if (mpext->use_map) {
1275 len += TCPOLEN_MPTCP_DSS_MAP64;
1277 /* Use only 64-bit mapping flags for now, add
1278 * support for optional 32-bit mappings later.
1280 flags |= MPTCP_DSS_HAS_MAP | MPTCP_DSS_DSN64;
1281 if (mpext->data_fin)
1282 flags |= MPTCP_DSS_DATA_FIN;
1285 *ptr++ = mptcp_option(MPTCPOPT_DSS, len, 0, flags);
1287 if (mpext->use_ack) {
1289 put_unaligned_be64(mpext->data_ack, ptr);
1292 put_unaligned_be32(mpext->data_ack32, ptr);
1297 if (mpext->use_map) {
1298 put_unaligned_be64(mpext->data_seq, ptr);
1300 put_unaligned_be32(mpext->subflow_seq, ptr);
1302 put_unaligned_be32(mpext->data_len << 16 |
1303 TCPOPT_NOP << 8 | TCPOPT_NOP, ptr);