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,
414 bool snd_data_fin_enable,
416 unsigned int remaining,
417 struct mptcp_out_options *opts)
419 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
420 struct mptcp_ext *mpext;
421 unsigned int data_len;
423 /* When skb is not available, we better over-estimate the emitted
424 * options len. A full DSS option (28 bytes) is longer than
425 * TCPOLEN_MPTCP_MPC_ACK_DATA(22) or TCPOLEN_MPTCP_MPJ_ACK(24), so
426 * tell the caller to defer the estimate to
427 * mptcp_established_options_dss(), which will reserve enough space.
432 /* MPC/MPJ needed only on 3rd ack packet, DATA_FIN and TCP shutdown take precedence */
433 if (subflow->fully_established || snd_data_fin_enable ||
434 subflow->snd_isn != TCP_SKB_CB(skb)->seq ||
435 sk->sk_state != TCP_ESTABLISHED)
438 if (subflow->mp_capable) {
439 mpext = mptcp_get_ext(skb);
440 data_len = mpext ? mpext->data_len : 0;
442 /* we will check ext_copy.data_len in mptcp_write_options() to
443 * discriminate between TCPOLEN_MPTCP_MPC_ACK_DATA and
444 * TCPOLEN_MPTCP_MPC_ACK
446 opts->ext_copy.data_len = data_len;
447 opts->suboptions = OPTION_MPTCP_MPC_ACK;
448 opts->sndr_key = subflow->local_key;
449 opts->rcvr_key = subflow->remote_key;
452 * The MP_CAPABLE option is carried on the SYN, SYN/ACK, and ACK
453 * packets that start the first subflow of an MPTCP connection,
454 * as well as the first packet that carries data
457 *size = ALIGN(TCPOLEN_MPTCP_MPC_ACK_DATA, 4);
459 *size = TCPOLEN_MPTCP_MPC_ACK;
461 pr_debug("subflow=%p, local_key=%llu, remote_key=%llu map_len=%d",
462 subflow, subflow->local_key, subflow->remote_key,
466 } else if (subflow->mp_join) {
467 opts->suboptions = OPTION_MPTCP_MPJ_ACK;
468 memcpy(opts->hmac, subflow->hmac, MPTCPOPT_HMAC_LEN);
469 *size = TCPOLEN_MPTCP_MPJ_ACK;
470 pr_debug("subflow=%p", subflow);
472 schedule_3rdack_retransmission(sk);
478 static void mptcp_write_data_fin(struct mptcp_subflow_context *subflow,
479 struct sk_buff *skb, struct mptcp_ext *ext)
481 /* The write_seq value has already been incremented, so the actual
482 * sequence number for the DATA_FIN is one less.
484 u64 data_fin_tx_seq = READ_ONCE(mptcp_sk(subflow->conn)->write_seq) - 1;
486 if (!ext->use_map || !skb->len) {
487 /* RFC6824 requires a DSS mapping with specific values
488 * if DATA_FIN is set but no data payload is mapped
493 ext->data_seq = data_fin_tx_seq;
494 ext->subflow_seq = 0;
496 } else if (ext->data_seq + ext->data_len == data_fin_tx_seq) {
497 /* If there's an existing DSS mapping and it is the
498 * final mapping, DATA_FIN consumes 1 additional byte of
506 static bool mptcp_established_options_dss(struct sock *sk, struct sk_buff *skb,
507 bool snd_data_fin_enable,
509 unsigned int remaining,
510 struct mptcp_out_options *opts)
512 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
513 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
514 unsigned int dss_size = 0;
515 struct mptcp_ext *mpext;
516 unsigned int ack_size;
520 mpext = skb ? mptcp_get_ext(skb) : NULL;
522 if (!skb || (mpext && mpext->use_map) || snd_data_fin_enable) {
523 unsigned int map_size;
525 map_size = TCPOLEN_MPTCP_DSS_BASE + TCPOLEN_MPTCP_DSS_MAP64;
527 remaining -= map_size;
530 opts->ext_copy = *mpext;
532 if (skb && snd_data_fin_enable)
533 mptcp_write_data_fin(subflow, skb, &opts->ext_copy);
537 /* passive sockets msk will set the 'can_ack' after accept(), even
538 * if the first subflow may have the already the remote key handy
540 opts->ext_copy.use_ack = 0;
541 if (!READ_ONCE(msk->can_ack)) {
542 *size = ALIGN(dss_size, 4);
546 ack_seq = READ_ONCE(msk->ack_seq);
547 if (READ_ONCE(msk->use_64bit_ack)) {
548 ack_size = TCPOLEN_MPTCP_DSS_ACK64;
549 opts->ext_copy.data_ack = ack_seq;
550 opts->ext_copy.ack64 = 1;
552 ack_size = TCPOLEN_MPTCP_DSS_ACK32;
553 opts->ext_copy.data_ack32 = (uint32_t)ack_seq;
554 opts->ext_copy.ack64 = 0;
556 opts->ext_copy.use_ack = 1;
557 WRITE_ONCE(msk->old_wspace, __mptcp_space((struct sock *)msk));
559 /* Add kind/length/subtype/flag overhead if mapping is not populated */
561 ack_size += TCPOLEN_MPTCP_DSS_BASE;
563 dss_size += ack_size;
565 *size = ALIGN(dss_size, 4);
569 static u64 add_addr_generate_hmac(u64 key1, u64 key2, u8 addr_id,
570 struct in_addr *addr, u16 port)
572 u8 hmac[SHA256_DIGEST_SIZE];
576 memcpy(&msg[1], &addr->s_addr, 4);
578 msg[6] = port & 0xFF;
580 mptcp_crypto_hmac_sha(key1, key2, msg, 7, hmac);
582 return get_unaligned_be64(&hmac[SHA256_DIGEST_SIZE - sizeof(u64)]);
585 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
586 static u64 add_addr6_generate_hmac(u64 key1, u64 key2, u8 addr_id,
587 struct in6_addr *addr, u16 port)
589 u8 hmac[SHA256_DIGEST_SIZE];
593 memcpy(&msg[1], &addr->s6_addr, 16);
595 msg[18] = port & 0xFF;
597 mptcp_crypto_hmac_sha(key1, key2, msg, 19, hmac);
599 return get_unaligned_be64(&hmac[SHA256_DIGEST_SIZE - sizeof(u64)]);
603 static bool mptcp_established_options_add_addr(struct sock *sk, struct sk_buff *skb,
605 unsigned int remaining,
606 struct mptcp_out_options *opts)
608 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
609 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
610 bool drop_other_suboptions = false;
611 unsigned int opt_size = *size;
612 struct mptcp_addr_info saddr;
617 if ((mptcp_pm_should_add_signal_ipv6(msk) ||
618 mptcp_pm_should_add_signal_port(msk)) &&
619 skb && skb_is_tcp_pure_ack(skb)) {
620 pr_debug("drop other suboptions");
621 opts->suboptions = 0;
622 opts->ext_copy.use_ack = 0;
623 opts->ext_copy.use_map = 0;
624 remaining += opt_size;
625 drop_other_suboptions = true;
628 if (!mptcp_pm_should_add_signal(msk) ||
629 !(mptcp_pm_add_addr_signal(msk, remaining, &saddr, &echo, &port)))
632 len = mptcp_add_addr_len(saddr.family, echo, port);
637 if (drop_other_suboptions)
639 opts->addr_id = saddr.id;
641 opts->port = ntohs(saddr.port);
642 if (saddr.family == AF_INET) {
643 opts->suboptions |= OPTION_MPTCP_ADD_ADDR;
644 opts->addr = saddr.addr;
646 opts->ahmac = add_addr_generate_hmac(msk->local_key,
653 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
654 else if (saddr.family == AF_INET6) {
655 opts->suboptions |= OPTION_MPTCP_ADD_ADDR6;
656 opts->addr6 = saddr.addr6;
658 opts->ahmac = add_addr6_generate_hmac(msk->local_key,
666 pr_debug("addr_id=%d, ahmac=%llu, echo=%d, port=%d",
667 opts->addr_id, opts->ahmac, echo, opts->port);
672 static bool mptcp_established_options_rm_addr(struct sock *sk,
674 unsigned int remaining,
675 struct mptcp_out_options *opts)
677 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
678 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
681 if (!mptcp_pm_should_rm_signal(msk) ||
682 !(mptcp_pm_rm_addr_signal(msk, remaining, &rm_id)))
685 if (remaining < TCPOLEN_MPTCP_RM_ADDR_BASE)
688 *size = TCPOLEN_MPTCP_RM_ADDR_BASE;
689 opts->suboptions |= OPTION_MPTCP_RM_ADDR;
692 pr_debug("rm_id=%d", opts->rm_id);
697 static bool mptcp_established_options_mp_prio(struct sock *sk,
699 unsigned int remaining,
700 struct mptcp_out_options *opts)
702 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
704 if (!subflow->send_mp_prio)
707 /* account for the trailing 'nop' option */
708 if (remaining < TCPOLEN_MPTCP_PRIO_ALIGN)
711 *size = TCPOLEN_MPTCP_PRIO_ALIGN;
712 opts->suboptions |= OPTION_MPTCP_PRIO;
713 opts->backup = subflow->request_bkup;
715 pr_debug("prio=%d", opts->backup);
720 bool mptcp_established_options(struct sock *sk, struct sk_buff *skb,
721 unsigned int *size, unsigned int remaining,
722 struct mptcp_out_options *opts)
724 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
725 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
726 unsigned int opt_size = 0;
730 opts->suboptions = 0;
732 if (unlikely(__mptcp_check_fallback(msk)))
735 /* prevent adding of any MPTCP related options on reset packet
736 * until we support MP_TCPRST/MP_FASTCLOSE
738 if (unlikely(skb && TCP_SKB_CB(skb)->tcp_flags & TCPHDR_RST))
741 snd_data_fin = mptcp_data_fin_enabled(msk);
742 if (mptcp_established_options_mp(sk, skb, snd_data_fin, &opt_size, remaining, opts))
744 else if (mptcp_established_options_dss(sk, skb, snd_data_fin, &opt_size, remaining, opts))
747 /* we reserved enough space for the above options, and exceeding the
748 * TCP option space would be fatal
750 if (WARN_ON_ONCE(opt_size > remaining))
754 remaining -= opt_size;
755 if (mptcp_established_options_add_addr(sk, skb, &opt_size, remaining, opts)) {
757 remaining -= opt_size;
759 } else if (mptcp_established_options_rm_addr(sk, &opt_size, remaining, opts)) {
761 remaining -= opt_size;
765 if (mptcp_established_options_mp_prio(sk, &opt_size, remaining, opts)) {
767 remaining -= opt_size;
774 bool mptcp_synack_options(const struct request_sock *req, unsigned int *size,
775 struct mptcp_out_options *opts)
777 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
779 if (subflow_req->mp_capable) {
780 opts->suboptions = OPTION_MPTCP_MPC_SYNACK;
781 opts->sndr_key = subflow_req->local_key;
782 *size = TCPOLEN_MPTCP_MPC_SYNACK;
783 pr_debug("subflow_req=%p, local_key=%llu",
784 subflow_req, subflow_req->local_key);
786 } else if (subflow_req->mp_join) {
787 opts->suboptions = OPTION_MPTCP_MPJ_SYNACK;
788 opts->backup = subflow_req->backup;
789 opts->join_id = subflow_req->local_id;
790 opts->thmac = subflow_req->thmac;
791 opts->nonce = subflow_req->local_nonce;
792 pr_debug("req=%p, bkup=%u, id=%u, thmac=%llu, nonce=%u",
793 subflow_req, opts->backup, opts->join_id,
794 opts->thmac, opts->nonce);
795 *size = TCPOLEN_MPTCP_MPJ_SYNACK;
801 static bool check_fully_established(struct mptcp_sock *msk, struct sock *ssk,
802 struct mptcp_subflow_context *subflow,
804 struct mptcp_options_received *mp_opt)
806 /* here we can process OoO, in-window pkts, only in-sequence 4th ack
807 * will make the subflow fully established
809 if (likely(subflow->fully_established)) {
810 /* on passive sockets, check for 3rd ack retransmission
811 * note that msk is always set by subflow_syn_recv_sock()
812 * for mp_join subflows
814 if (TCP_SKB_CB(skb)->seq == subflow->ssn_offset + 1 &&
815 TCP_SKB_CB(skb)->end_seq == TCP_SKB_CB(skb)->seq &&
816 subflow->mp_join && mp_opt->mp_join &&
817 READ_ONCE(msk->pm.server_side))
819 goto fully_established;
822 /* we must process OoO packets before the first subflow is fully
823 * established. OoO packets are instead a protocol violation
824 * for MP_JOIN subflows as the peer must not send any data
825 * before receiving the forth ack - cfr. RFC 8684 section 3.2.
827 if (TCP_SKB_CB(skb)->seq != subflow->ssn_offset + 1) {
828 if (subflow->mp_join)
830 return subflow->mp_capable;
833 if (mp_opt->dss && mp_opt->use_ack) {
834 /* subflows are fully established as soon as we get any
837 subflow->fully_established = 1;
838 WRITE_ONCE(msk->fully_established, true);
839 goto fully_established;
842 if (mp_opt->add_addr) {
843 WRITE_ONCE(msk->fully_established, true);
847 /* If the first established packet does not contain MP_CAPABLE + data
848 * then fallback to TCP. Fallback scenarios requires a reset for
851 if (!mp_opt->mp_capable) {
852 if (subflow->mp_join)
854 subflow->mp_capable = 0;
856 __mptcp_do_fallback(msk);
860 if (unlikely(!READ_ONCE(msk->pm.server_side)))
861 pr_warn_once("bogus mpc option on established client sk");
862 mptcp_subflow_fully_established(subflow, mp_opt);
865 /* if the subflow is not already linked into the conn_list, we can't
866 * notify the PM: this subflow is still on the listener queue
867 * and the PM possibly acquiring the subflow lock could race with
870 if (likely(subflow->pm_notified) || list_empty(&subflow->node))
873 subflow->pm_notified = 1;
874 if (subflow->mp_join) {
875 clear_3rdack_retransmission(ssk);
876 mptcp_pm_subflow_established(msk, subflow);
878 mptcp_pm_fully_established(msk, ssk, GFP_ATOMIC);
883 mptcp_subflow_reset(ssk);
887 static u64 expand_ack(u64 old_ack, u64 cur_ack, bool use_64bit)
889 u32 old_ack32, cur_ack32;
894 old_ack32 = (u32)old_ack;
895 cur_ack32 = (u32)cur_ack;
896 cur_ack = (old_ack & GENMASK_ULL(63, 32)) + cur_ack32;
897 if (unlikely(before(cur_ack32, old_ack32)))
898 return cur_ack + (1LL << 32);
902 static void ack_update_msk(struct mptcp_sock *msk,
904 struct mptcp_options_received *mp_opt)
906 u64 new_wnd_end, new_snd_una, snd_nxt = READ_ONCE(msk->snd_nxt);
907 struct sock *sk = (struct sock *)msk;
912 /* avoid ack expansion on update conflict, to reduce the risk of
913 * wrongly expanding to a future ack sequence number, which is way
914 * more dangerous than missing an ack
916 old_snd_una = msk->snd_una;
917 new_snd_una = expand_ack(old_snd_una, mp_opt->data_ack, mp_opt->ack64);
919 /* ACK for data not even sent yet? Ignore. */
920 if (after64(new_snd_una, snd_nxt))
921 new_snd_una = old_snd_una;
923 new_wnd_end = new_snd_una + tcp_sk(ssk)->snd_wnd;
925 if (after64(new_wnd_end, msk->wnd_end))
926 msk->wnd_end = new_wnd_end;
928 /* this assumes mptcp_incoming_options() is invoked after tcp_ack() */
929 if (after64(msk->wnd_end, READ_ONCE(msk->snd_nxt)))
930 __mptcp_check_push(sk, ssk);
932 if (after64(new_snd_una, old_snd_una)) {
933 msk->snd_una = new_snd_una;
934 __mptcp_data_acked(sk);
936 mptcp_data_unlock(sk);
939 bool mptcp_update_rcv_data_fin(struct mptcp_sock *msk, u64 data_fin_seq, bool use_64bit)
941 /* Skip if DATA_FIN was already received.
942 * If updating simultaneously with the recvmsg loop, values
943 * should match. If they mismatch, the peer is misbehaving and
944 * we will prefer the most recent information.
946 if (READ_ONCE(msk->rcv_data_fin) || !READ_ONCE(msk->first))
949 WRITE_ONCE(msk->rcv_data_fin_seq,
950 expand_ack(READ_ONCE(msk->ack_seq), data_fin_seq, use_64bit));
951 WRITE_ONCE(msk->rcv_data_fin, 1);
956 static bool add_addr_hmac_valid(struct mptcp_sock *msk,
957 struct mptcp_options_received *mp_opt)
964 if (mp_opt->family == MPTCP_ADDR_IPVERSION_4)
965 hmac = add_addr_generate_hmac(msk->remote_key,
967 mp_opt->addr_id, &mp_opt->addr,
969 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
971 hmac = add_addr6_generate_hmac(msk->remote_key,
973 mp_opt->addr_id, &mp_opt->addr6,
977 pr_debug("msk=%p, ahmac=%llu, mp_opt->ahmac=%llu\n",
978 msk, (unsigned long long)hmac,
979 (unsigned long long)mp_opt->ahmac);
981 return hmac == mp_opt->ahmac;
984 void mptcp_incoming_options(struct sock *sk, struct sk_buff *skb)
986 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
987 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
988 struct mptcp_options_received mp_opt;
989 struct mptcp_ext *mpext;
991 if (__mptcp_check_fallback(msk)) {
992 /* Keep it simple and unconditionally trigger send data cleanup and
993 * pending queue spooling. We will need to acquire the data lock
994 * for more accurate checks, and once the lock is acquired, such
997 mptcp_data_lock(subflow->conn);
998 if (sk_stream_memory_free(sk))
999 __mptcp_check_push(subflow->conn, sk);
1000 __mptcp_data_acked(subflow->conn);
1001 mptcp_data_unlock(subflow->conn);
1005 mptcp_get_options(skb, &mp_opt);
1006 if (!check_fully_established(msk, sk, subflow, skb, &mp_opt))
1009 if (mp_opt.fastclose &&
1010 msk->local_key == mp_opt.rcvr_key) {
1011 WRITE_ONCE(msk->rcv_fastclose, true);
1012 mptcp_schedule_work((struct sock *)msk);
1015 if (mp_opt.add_addr && add_addr_hmac_valid(msk, &mp_opt)) {
1016 struct mptcp_addr_info addr;
1018 addr.port = htons(mp_opt.port);
1019 addr.id = mp_opt.addr_id;
1020 if (mp_opt.family == MPTCP_ADDR_IPVERSION_4) {
1021 addr.family = AF_INET;
1022 addr.addr = mp_opt.addr;
1024 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1025 else if (mp_opt.family == MPTCP_ADDR_IPVERSION_6) {
1026 addr.family = AF_INET6;
1027 addr.addr6 = mp_opt.addr6;
1031 mptcp_pm_add_addr_received(msk, &addr);
1032 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ADDADDR);
1034 mptcp_pm_del_add_timer(msk, &addr);
1035 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ECHOADD);
1039 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_PORTADD);
1041 mp_opt.add_addr = 0;
1044 if (mp_opt.rm_addr) {
1045 mptcp_pm_rm_addr_received(msk, mp_opt.rm_id);
1049 if (mp_opt.mp_prio) {
1050 mptcp_pm_mp_prio_received(sk, mp_opt.backup);
1051 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPPRIORX);
1058 /* we can't wait for recvmsg() to update the ack_seq, otherwise
1059 * monodirectional flows will stuck
1062 ack_update_msk(msk, sk, &mp_opt);
1064 /* Zero-data-length packets are dropped by the caller and not
1065 * propagated to the MPTCP layer, so the skb extension does not
1066 * need to be allocated or populated. DATA_FIN information, if
1067 * present, needs to be updated here before the skb is freed.
1069 if (TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb)->end_seq) {
1070 if (mp_opt.data_fin && mp_opt.data_len == 1 &&
1071 mptcp_update_rcv_data_fin(msk, mp_opt.data_seq, mp_opt.dsn64) &&
1072 schedule_work(&msk->work))
1073 sock_hold(subflow->conn);
1078 mpext = skb_ext_add(skb, SKB_EXT_MPTCP);
1082 memset(mpext, 0, sizeof(*mpext));
1084 if (mp_opt.use_map) {
1085 if (mp_opt.mpc_map) {
1086 /* this is an MP_CAPABLE carrying MPTCP data
1087 * we know this map the first chunk of data
1089 mptcp_crypto_key_sha(subflow->remote_key, NULL,
1092 mpext->subflow_seq = 1;
1095 mpext->data_fin = 0;
1097 mpext->data_seq = mp_opt.data_seq;
1098 mpext->subflow_seq = mp_opt.subflow_seq;
1099 mpext->dsn64 = mp_opt.dsn64;
1100 mpext->data_fin = mp_opt.data_fin;
1102 mpext->data_len = mp_opt.data_len;
1107 static void mptcp_set_rwin(const struct tcp_sock *tp)
1109 const struct sock *ssk = (const struct sock *)tp;
1110 const struct mptcp_subflow_context *subflow;
1111 struct mptcp_sock *msk;
1114 subflow = mptcp_subflow_ctx(ssk);
1115 msk = mptcp_sk(subflow->conn);
1117 ack_seq = READ_ONCE(msk->ack_seq) + tp->rcv_wnd;
1119 if (after64(ack_seq, READ_ONCE(msk->rcv_wnd_sent)))
1120 WRITE_ONCE(msk->rcv_wnd_sent, ack_seq);
1123 void mptcp_write_options(__be32 *ptr, const struct tcp_sock *tp,
1124 struct mptcp_out_options *opts)
1126 if ((OPTION_MPTCP_MPC_SYN | OPTION_MPTCP_MPC_SYNACK |
1127 OPTION_MPTCP_MPC_ACK) & opts->suboptions) {
1130 if (OPTION_MPTCP_MPC_SYN & opts->suboptions)
1131 len = TCPOLEN_MPTCP_MPC_SYN;
1132 else if (OPTION_MPTCP_MPC_SYNACK & opts->suboptions)
1133 len = TCPOLEN_MPTCP_MPC_SYNACK;
1134 else if (opts->ext_copy.data_len)
1135 len = TCPOLEN_MPTCP_MPC_ACK_DATA;
1137 len = TCPOLEN_MPTCP_MPC_ACK;
1139 *ptr++ = mptcp_option(MPTCPOPT_MP_CAPABLE, len,
1140 MPTCP_SUPPORTED_VERSION,
1141 MPTCP_CAP_HMAC_SHA256);
1143 if (!((OPTION_MPTCP_MPC_SYNACK | OPTION_MPTCP_MPC_ACK) &
1145 goto mp_capable_done;
1147 put_unaligned_be64(opts->sndr_key, ptr);
1149 if (!((OPTION_MPTCP_MPC_ACK) & opts->suboptions))
1150 goto mp_capable_done;
1152 put_unaligned_be64(opts->rcvr_key, ptr);
1154 if (!opts->ext_copy.data_len)
1155 goto mp_capable_done;
1157 put_unaligned_be32(opts->ext_copy.data_len << 16 |
1158 TCPOPT_NOP << 8 | TCPOPT_NOP, ptr);
1163 if ((OPTION_MPTCP_ADD_ADDR
1164 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1165 | OPTION_MPTCP_ADD_ADDR6
1167 ) & opts->suboptions) {
1168 u8 len = TCPOLEN_MPTCP_ADD_ADDR_BASE;
1169 u8 echo = MPTCP_ADDR_ECHO;
1171 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1172 if (OPTION_MPTCP_ADD_ADDR6 & opts->suboptions)
1173 len = TCPOLEN_MPTCP_ADD_ADDR6_BASE;
1177 len += TCPOLEN_MPTCP_PORT_LEN;
1180 len += sizeof(opts->ahmac);
1184 *ptr++ = mptcp_option(MPTCPOPT_ADD_ADDR,
1185 len, echo, opts->addr_id);
1186 if (OPTION_MPTCP_ADD_ADDR & opts->suboptions) {
1187 memcpy((u8 *)ptr, (u8 *)&opts->addr.s_addr, 4);
1190 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1191 else if (OPTION_MPTCP_ADD_ADDR6 & opts->suboptions) {
1192 memcpy((u8 *)ptr, opts->addr6.s6_addr, 16);
1199 put_unaligned_be64(opts->ahmac, ptr);
1204 u8 *bptr = (u8 *)ptr;
1206 put_unaligned_be16(opts->port, bptr);
1208 put_unaligned_be64(opts->ahmac, bptr);
1210 put_unaligned_be16(TCPOPT_NOP << 8 |
1215 put_unaligned_be32(opts->port << 16 |
1223 if (OPTION_MPTCP_RM_ADDR & opts->suboptions) {
1224 *ptr++ = mptcp_option(MPTCPOPT_RM_ADDR,
1225 TCPOLEN_MPTCP_RM_ADDR_BASE,
1229 if (OPTION_MPTCP_PRIO & opts->suboptions) {
1230 const struct sock *ssk = (const struct sock *)tp;
1231 struct mptcp_subflow_context *subflow;
1233 subflow = mptcp_subflow_ctx(ssk);
1234 subflow->send_mp_prio = 0;
1236 *ptr++ = mptcp_option(MPTCPOPT_MP_PRIO,
1238 opts->backup, TCPOPT_NOP);
1241 if (OPTION_MPTCP_MPJ_SYN & opts->suboptions) {
1242 *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1243 TCPOLEN_MPTCP_MPJ_SYN,
1244 opts->backup, opts->join_id);
1245 put_unaligned_be32(opts->token, ptr);
1247 put_unaligned_be32(opts->nonce, ptr);
1251 if (OPTION_MPTCP_MPJ_SYNACK & opts->suboptions) {
1252 *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1253 TCPOLEN_MPTCP_MPJ_SYNACK,
1254 opts->backup, opts->join_id);
1255 put_unaligned_be64(opts->thmac, ptr);
1257 put_unaligned_be32(opts->nonce, ptr);
1261 if (OPTION_MPTCP_MPJ_ACK & opts->suboptions) {
1262 *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1263 TCPOLEN_MPTCP_MPJ_ACK, 0, 0);
1264 memcpy(ptr, opts->hmac, MPTCPOPT_HMAC_LEN);
1268 if (opts->ext_copy.use_ack || opts->ext_copy.use_map) {
1269 struct mptcp_ext *mpext = &opts->ext_copy;
1270 u8 len = TCPOLEN_MPTCP_DSS_BASE;
1273 if (mpext->use_ack) {
1274 flags = MPTCP_DSS_HAS_ACK;
1276 len += TCPOLEN_MPTCP_DSS_ACK64;
1277 flags |= MPTCP_DSS_ACK64;
1279 len += TCPOLEN_MPTCP_DSS_ACK32;
1283 if (mpext->use_map) {
1284 len += TCPOLEN_MPTCP_DSS_MAP64;
1286 /* Use only 64-bit mapping flags for now, add
1287 * support for optional 32-bit mappings later.
1289 flags |= MPTCP_DSS_HAS_MAP | MPTCP_DSS_DSN64;
1290 if (mpext->data_fin)
1291 flags |= MPTCP_DSS_DATA_FIN;
1294 *ptr++ = mptcp_option(MPTCPOPT_DSS, len, 0, flags);
1296 if (mpext->use_ack) {
1298 put_unaligned_be64(mpext->data_ack, ptr);
1301 put_unaligned_be32(mpext->data_ack32, ptr);
1306 if (mpext->use_map) {
1307 put_unaligned_be64(mpext->data_seq, ptr);
1309 put_unaligned_be32(mpext->subflow_seq, ptr);
1311 put_unaligned_be32(mpext->data_len << 16 |
1312 TCPOPT_NOP << 8 | TCPOPT_NOP, ptr);