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 #include <trace/events/mptcp.h>
18 static bool mptcp_cap_flag_sha256(u8 flags)
20 return (flags & MPTCP_CAP_FLAG_MASK) == MPTCP_CAP_HMAC_SHA256;
23 static void mptcp_parse_option(const struct sk_buff *skb,
24 const unsigned char *ptr, int opsize,
25 struct mptcp_options_received *mp_opt)
27 u8 subtype = *ptr >> 4;
35 case MPTCPOPT_MP_CAPABLE:
36 /* strict size checking */
37 if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)) {
38 if (skb->len > tcp_hdr(skb)->doff << 2)
39 expected_opsize = TCPOLEN_MPTCP_MPC_ACK_DATA;
41 expected_opsize = TCPOLEN_MPTCP_MPC_ACK;
42 subopt = OPTION_MPTCP_MPC_ACK;
44 if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_ACK) {
45 expected_opsize = TCPOLEN_MPTCP_MPC_SYNACK;
46 subopt = OPTION_MPTCP_MPC_SYNACK;
48 expected_opsize = TCPOLEN_MPTCP_MPC_SYN;
49 subopt = OPTION_MPTCP_MPC_SYN;
53 /* Cfr RFC 8684 Section 3.3.0:
54 * If a checksum is present but its use had
55 * not been negotiated in the MP_CAPABLE handshake, the receiver MUST
56 * close the subflow with a RST, as it is not behaving as negotiated.
57 * If a checksum is not present when its use has been negotiated, the
58 * receiver MUST close the subflow with a RST, as it is considered
60 * We parse even option with mismatching csum presence, so that
61 * later in subflow_data_ready we can trigger the reset.
63 if (opsize != expected_opsize &&
64 (expected_opsize != TCPOLEN_MPTCP_MPC_ACK_DATA ||
65 opsize != TCPOLEN_MPTCP_MPC_ACK_DATA_CSUM))
68 /* try to be gentle vs future versions on the initial syn */
69 version = *ptr++ & MPTCP_VERSION_MASK;
70 if (opsize != TCPOLEN_MPTCP_MPC_SYN) {
71 if (version != MPTCP_SUPPORTED_VERSION)
73 } else if (version < MPTCP_SUPPORTED_VERSION) {
78 if (!mptcp_cap_flag_sha256(flags) ||
79 (flags & MPTCP_CAP_EXTENSIBILITY))
82 /* RFC 6824, Section 3.1:
83 * "For the Checksum Required bit (labeled "A"), if either
84 * host requires the use of checksums, checksums MUST be used.
85 * In other words, the only way for checksums not to be used
86 * is if both hosts in their SYNs set A=0."
88 if (flags & MPTCP_CAP_CHECKSUM_REQD)
89 mp_opt->suboptions |= OPTION_MPTCP_CSUMREQD;
91 mp_opt->deny_join_id0 = !!(flags & MPTCP_CAP_DENY_JOIN_ID0);
93 mp_opt->suboptions |= subopt;
94 if (opsize >= TCPOLEN_MPTCP_MPC_SYNACK) {
95 mp_opt->sndr_key = get_unaligned_be64(ptr);
98 if (opsize >= TCPOLEN_MPTCP_MPC_ACK) {
99 mp_opt->rcvr_key = get_unaligned_be64(ptr);
102 if (opsize >= TCPOLEN_MPTCP_MPC_ACK_DATA) {
104 * "the data parameters in a MP_CAPABLE are semantically
105 * equivalent to those in a DSS option and can be used
108 mp_opt->suboptions |= OPTION_MPTCP_DSS;
112 mp_opt->data_len = get_unaligned_be16(ptr);
115 if (opsize == TCPOLEN_MPTCP_MPC_ACK_DATA_CSUM) {
116 mp_opt->csum = get_unaligned((__force __sum16 *)ptr);
117 mp_opt->suboptions |= OPTION_MPTCP_CSUMREQD;
120 pr_debug("MP_CAPABLE version=%x, flags=%x, optlen=%d sndr=%llu, rcvr=%llu len=%d csum=%u",
121 version, flags, opsize, mp_opt->sndr_key,
122 mp_opt->rcvr_key, mp_opt->data_len, mp_opt->csum);
125 case MPTCPOPT_MP_JOIN:
126 if (opsize == TCPOLEN_MPTCP_MPJ_SYN) {
127 mp_opt->suboptions |= OPTION_MPTCP_MPJ_SYN;
128 mp_opt->backup = *ptr++ & MPTCPOPT_BACKUP;
129 mp_opt->join_id = *ptr++;
130 mp_opt->token = get_unaligned_be32(ptr);
132 mp_opt->nonce = get_unaligned_be32(ptr);
134 pr_debug("MP_JOIN bkup=%u, id=%u, token=%u, nonce=%u",
135 mp_opt->backup, mp_opt->join_id,
136 mp_opt->token, mp_opt->nonce);
137 } else if (opsize == TCPOLEN_MPTCP_MPJ_SYNACK) {
138 mp_opt->suboptions |= OPTION_MPTCP_MPJ_SYNACK;
139 mp_opt->backup = *ptr++ & MPTCPOPT_BACKUP;
140 mp_opt->join_id = *ptr++;
141 mp_opt->thmac = get_unaligned_be64(ptr);
143 mp_opt->nonce = get_unaligned_be32(ptr);
145 pr_debug("MP_JOIN bkup=%u, id=%u, thmac=%llu, nonce=%u",
146 mp_opt->backup, mp_opt->join_id,
147 mp_opt->thmac, mp_opt->nonce);
148 } else if (opsize == TCPOLEN_MPTCP_MPJ_ACK) {
149 mp_opt->suboptions |= OPTION_MPTCP_MPJ_ACK;
151 memcpy(mp_opt->hmac, ptr, MPTCPOPT_HMAC_LEN);
152 pr_debug("MP_JOIN hmac");
160 /* we must clear 'mpc_map' be able to detect MP_CAPABLE
161 * map vs DSS map in mptcp_incoming_options(), and reconstruct
162 * map info accordingly
165 flags = (*ptr++) & MPTCP_DSS_FLAG_MASK;
166 mp_opt->data_fin = (flags & MPTCP_DSS_DATA_FIN) != 0;
167 mp_opt->dsn64 = (flags & MPTCP_DSS_DSN64) != 0;
168 mp_opt->use_map = (flags & MPTCP_DSS_HAS_MAP) != 0;
169 mp_opt->ack64 = (flags & MPTCP_DSS_ACK64) != 0;
170 mp_opt->use_ack = (flags & MPTCP_DSS_HAS_ACK);
172 pr_debug("data_fin=%d dsn64=%d use_map=%d ack64=%d use_ack=%d",
173 mp_opt->data_fin, mp_opt->dsn64,
174 mp_opt->use_map, mp_opt->ack64,
177 expected_opsize = TCPOLEN_MPTCP_DSS_BASE;
179 if (mp_opt->use_ack) {
181 expected_opsize += TCPOLEN_MPTCP_DSS_ACK64;
183 expected_opsize += TCPOLEN_MPTCP_DSS_ACK32;
186 if (mp_opt->use_map) {
188 expected_opsize += TCPOLEN_MPTCP_DSS_MAP64;
190 expected_opsize += TCPOLEN_MPTCP_DSS_MAP32;
193 /* Always parse any csum presence combination, we will enforce
194 * RFC 8684 Section 3.3.0 checks later in subflow_data_ready
196 if (opsize != expected_opsize &&
197 opsize != expected_opsize + TCPOLEN_MPTCP_DSS_CHECKSUM)
200 mp_opt->suboptions |= OPTION_MPTCP_DSS;
201 if (mp_opt->use_ack) {
203 mp_opt->data_ack = get_unaligned_be64(ptr);
206 mp_opt->data_ack = get_unaligned_be32(ptr);
210 pr_debug("data_ack=%llu", mp_opt->data_ack);
213 if (mp_opt->use_map) {
215 mp_opt->data_seq = get_unaligned_be64(ptr);
218 mp_opt->data_seq = get_unaligned_be32(ptr);
222 mp_opt->subflow_seq = get_unaligned_be32(ptr);
225 mp_opt->data_len = get_unaligned_be16(ptr);
228 if (opsize == expected_opsize + TCPOLEN_MPTCP_DSS_CHECKSUM) {
229 mp_opt->suboptions |= OPTION_MPTCP_CSUMREQD;
230 mp_opt->csum = get_unaligned((__force __sum16 *)ptr);
234 pr_debug("data_seq=%llu subflow_seq=%u data_len=%u csum=%d:%u",
235 mp_opt->data_seq, mp_opt->subflow_seq,
236 mp_opt->data_len, !!(mp_opt->suboptions & OPTION_MPTCP_CSUMREQD),
242 case MPTCPOPT_ADD_ADDR:
243 mp_opt->echo = (*ptr++) & MPTCP_ADDR_ECHO;
245 if (opsize == TCPOLEN_MPTCP_ADD_ADDR ||
246 opsize == TCPOLEN_MPTCP_ADD_ADDR_PORT)
247 mp_opt->addr.family = AF_INET;
248 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
249 else if (opsize == TCPOLEN_MPTCP_ADD_ADDR6 ||
250 opsize == TCPOLEN_MPTCP_ADD_ADDR6_PORT)
251 mp_opt->addr.family = AF_INET6;
256 if (opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE ||
257 opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE_PORT)
258 mp_opt->addr.family = AF_INET;
259 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
260 else if (opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE ||
261 opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE_PORT)
262 mp_opt->addr.family = AF_INET6;
268 mp_opt->suboptions |= OPTION_MPTCP_ADD_ADDR;
269 mp_opt->addr.id = *ptr++;
270 mp_opt->addr.port = 0;
272 if (mp_opt->addr.family == AF_INET) {
273 memcpy((u8 *)&mp_opt->addr.addr.s_addr, (u8 *)ptr, 4);
275 if (opsize == TCPOLEN_MPTCP_ADD_ADDR_PORT ||
276 opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE_PORT) {
277 mp_opt->addr.port = htons(get_unaligned_be16(ptr));
281 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
283 memcpy(mp_opt->addr.addr6.s6_addr, (u8 *)ptr, 16);
285 if (opsize == TCPOLEN_MPTCP_ADD_ADDR6_PORT ||
286 opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE_PORT) {
287 mp_opt->addr.port = htons(get_unaligned_be16(ptr));
293 mp_opt->ahmac = get_unaligned_be64(ptr);
296 pr_debug("ADD_ADDR%s: id=%d, ahmac=%llu, echo=%d, port=%d",
297 (mp_opt->addr.family == AF_INET6) ? "6" : "",
298 mp_opt->addr.id, mp_opt->ahmac, mp_opt->echo, ntohs(mp_opt->addr.port));
301 case MPTCPOPT_RM_ADDR:
302 if (opsize < TCPOLEN_MPTCP_RM_ADDR_BASE + 1 ||
303 opsize > TCPOLEN_MPTCP_RM_ADDR_BASE + MPTCP_RM_IDS_MAX)
308 mp_opt->suboptions |= OPTION_MPTCP_RM_ADDR;
309 mp_opt->rm_list.nr = opsize - TCPOLEN_MPTCP_RM_ADDR_BASE;
310 for (i = 0; i < mp_opt->rm_list.nr; i++)
311 mp_opt->rm_list.ids[i] = *ptr++;
312 pr_debug("RM_ADDR: rm_list_nr=%d", mp_opt->rm_list.nr);
315 case MPTCPOPT_MP_PRIO:
316 if (opsize != TCPOLEN_MPTCP_PRIO)
319 mp_opt->suboptions |= OPTION_MPTCP_PRIO;
320 mp_opt->backup = *ptr++ & MPTCP_PRIO_BKUP;
321 pr_debug("MP_PRIO: prio=%d", mp_opt->backup);
324 case MPTCPOPT_MP_FASTCLOSE:
325 if (opsize != TCPOLEN_MPTCP_FASTCLOSE)
329 mp_opt->rcvr_key = get_unaligned_be64(ptr);
331 mp_opt->suboptions |= OPTION_MPTCP_FASTCLOSE;
332 pr_debug("MP_FASTCLOSE: recv_key=%llu", mp_opt->rcvr_key);
336 if (opsize != TCPOLEN_MPTCP_RST)
339 if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_RST))
342 mp_opt->suboptions |= OPTION_MPTCP_RST;
344 mp_opt->reset_transient = flags & MPTCP_RST_TRANSIENT;
345 mp_opt->reset_reason = *ptr;
346 pr_debug("MP_RST: transient=%u reason=%u",
347 mp_opt->reset_transient, mp_opt->reset_reason);
350 case MPTCPOPT_MP_FAIL:
351 if (opsize != TCPOLEN_MPTCP_FAIL)
355 mp_opt->suboptions |= OPTION_MPTCP_FAIL;
356 mp_opt->fail_seq = get_unaligned_be64(ptr);
357 pr_debug("MP_FAIL: data_seq=%llu", mp_opt->fail_seq);
365 void mptcp_get_options(const struct sk_buff *skb,
366 struct mptcp_options_received *mp_opt)
368 const struct tcphdr *th = tcp_hdr(skb);
369 const unsigned char *ptr;
372 /* initialize option status */
373 mp_opt->suboptions = 0;
375 length = (th->doff * 4) - sizeof(struct tcphdr);
376 ptr = (const unsigned char *)(th + 1);
385 case TCPOPT_NOP: /* Ref: RFC 793 section 3.1 */
392 if (opsize < 2) /* "silly options" */
395 return; /* don't parse partial options */
396 if (opcode == TCPOPT_MPTCP)
397 mptcp_parse_option(skb, ptr, opsize, mp_opt);
404 bool mptcp_syn_options(struct sock *sk, const struct sk_buff *skb,
405 unsigned int *size, struct mptcp_out_options *opts)
407 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
409 /* we will use snd_isn to detect first pkt [re]transmission
410 * in mptcp_established_options_mp()
412 subflow->snd_isn = TCP_SKB_CB(skb)->end_seq;
413 if (subflow->request_mptcp) {
414 opts->suboptions = OPTION_MPTCP_MPC_SYN;
415 opts->csum_reqd = mptcp_is_checksum_enabled(sock_net(sk));
416 opts->allow_join_id0 = mptcp_allow_join_id0(sock_net(sk));
417 *size = TCPOLEN_MPTCP_MPC_SYN;
419 } else if (subflow->request_join) {
420 pr_debug("remote_token=%u, nonce=%u", subflow->remote_token,
421 subflow->local_nonce);
422 opts->suboptions = OPTION_MPTCP_MPJ_SYN;
423 opts->join_id = subflow->local_id;
424 opts->token = subflow->remote_token;
425 opts->nonce = subflow->local_nonce;
426 opts->backup = subflow->request_bkup;
427 *size = TCPOLEN_MPTCP_MPJ_SYN;
433 static void clear_3rdack_retransmission(struct sock *sk)
435 struct inet_connection_sock *icsk = inet_csk(sk);
437 sk_stop_timer(sk, &icsk->icsk_delack_timer);
438 icsk->icsk_ack.timeout = 0;
439 icsk->icsk_ack.ato = 0;
440 icsk->icsk_ack.pending &= ~(ICSK_ACK_SCHED | ICSK_ACK_TIMER);
443 static bool mptcp_established_options_mp(struct sock *sk, struct sk_buff *skb,
444 bool snd_data_fin_enable,
446 struct mptcp_out_options *opts)
448 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
449 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
450 struct mptcp_ext *mpext;
451 unsigned int data_len;
454 /* When skb is not available, we better over-estimate the emitted
455 * options len. A full DSS option (28 bytes) is longer than
456 * TCPOLEN_MPTCP_MPC_ACK_DATA(22) or TCPOLEN_MPTCP_MPJ_ACK(24), so
457 * tell the caller to defer the estimate to
458 * mptcp_established_options_dss(), which will reserve enough space.
463 /* MPC/MPJ needed only on 3rd ack packet, DATA_FIN and TCP shutdown take precedence */
464 if (subflow->fully_established || snd_data_fin_enable ||
465 subflow->snd_isn != TCP_SKB_CB(skb)->seq ||
466 sk->sk_state != TCP_ESTABLISHED)
469 if (subflow->mp_capable) {
470 mpext = mptcp_get_ext(skb);
471 data_len = mpext ? mpext->data_len : 0;
473 /* we will check ops->data_len in mptcp_write_options() to
474 * discriminate between TCPOLEN_MPTCP_MPC_ACK_DATA and
475 * TCPOLEN_MPTCP_MPC_ACK
477 opts->data_len = data_len;
478 opts->suboptions = OPTION_MPTCP_MPC_ACK;
479 opts->sndr_key = subflow->local_key;
480 opts->rcvr_key = subflow->remote_key;
481 opts->csum_reqd = READ_ONCE(msk->csum_enabled);
482 opts->allow_join_id0 = mptcp_allow_join_id0(sock_net(sk));
485 * The MP_CAPABLE option is carried on the SYN, SYN/ACK, and ACK
486 * packets that start the first subflow of an MPTCP connection,
487 * as well as the first packet that carries data
490 len = TCPOLEN_MPTCP_MPC_ACK_DATA;
491 if (opts->csum_reqd) {
492 /* we need to propagate more info to csum the pseudo hdr */
493 opts->data_seq = mpext->data_seq;
494 opts->subflow_seq = mpext->subflow_seq;
495 opts->csum = mpext->csum;
496 len += TCPOLEN_MPTCP_DSS_CHECKSUM;
498 *size = ALIGN(len, 4);
500 *size = TCPOLEN_MPTCP_MPC_ACK;
503 pr_debug("subflow=%p, local_key=%llu, remote_key=%llu map_len=%d",
504 subflow, subflow->local_key, subflow->remote_key,
508 } else if (subflow->mp_join) {
509 opts->suboptions = OPTION_MPTCP_MPJ_ACK;
510 memcpy(opts->hmac, subflow->hmac, MPTCPOPT_HMAC_LEN);
511 *size = TCPOLEN_MPTCP_MPJ_ACK;
512 pr_debug("subflow=%p", subflow);
514 /* we can use the full delegate action helper only from BH context
515 * If we are in process context - sk is flushing the backlog at
516 * socket lock release time - just set the appropriate flag, will
517 * be handled by the release callback
519 if (sock_owned_by_user(sk))
520 set_bit(MPTCP_DELEGATE_ACK, &subflow->delegated_status);
522 mptcp_subflow_delegate(subflow, MPTCP_DELEGATE_ACK);
528 static void mptcp_write_data_fin(struct mptcp_subflow_context *subflow,
529 struct sk_buff *skb, struct mptcp_ext *ext)
531 /* The write_seq value has already been incremented, so the actual
532 * sequence number for the DATA_FIN is one less.
534 u64 data_fin_tx_seq = READ_ONCE(mptcp_sk(subflow->conn)->write_seq) - 1;
536 if (!ext->use_map || !skb->len) {
537 /* RFC6824 requires a DSS mapping with specific values
538 * if DATA_FIN is set but no data payload is mapped
543 ext->data_seq = data_fin_tx_seq;
544 ext->subflow_seq = 0;
546 } else if (ext->data_seq + ext->data_len == data_fin_tx_seq) {
547 /* If there's an existing DSS mapping and it is the
548 * final mapping, DATA_FIN consumes 1 additional byte of
556 static bool mptcp_established_options_dss(struct sock *sk, struct sk_buff *skb,
557 bool snd_data_fin_enable,
559 struct mptcp_out_options *opts)
561 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
562 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
563 unsigned int dss_size = 0;
564 struct mptcp_ext *mpext;
565 unsigned int ack_size;
569 opts->csum_reqd = READ_ONCE(msk->csum_enabled);
570 mpext = skb ? mptcp_get_ext(skb) : NULL;
572 if (!skb || (mpext && mpext->use_map) || snd_data_fin_enable) {
573 unsigned int map_size = TCPOLEN_MPTCP_DSS_BASE + TCPOLEN_MPTCP_DSS_MAP64;
577 map_size += TCPOLEN_MPTCP_DSS_CHECKSUM;
579 opts->ext_copy = *mpext;
583 if (skb && snd_data_fin_enable)
584 mptcp_write_data_fin(subflow, skb, &opts->ext_copy);
585 opts->suboptions = OPTION_MPTCP_DSS;
589 /* passive sockets msk will set the 'can_ack' after accept(), even
590 * if the first subflow may have the already the remote key handy
592 opts->ext_copy.use_ack = 0;
593 if (!READ_ONCE(msk->can_ack)) {
594 *size = ALIGN(dss_size, 4);
598 ack_seq = READ_ONCE(msk->ack_seq);
599 if (READ_ONCE(msk->use_64bit_ack)) {
600 ack_size = TCPOLEN_MPTCP_DSS_ACK64;
601 opts->ext_copy.data_ack = ack_seq;
602 opts->ext_copy.ack64 = 1;
604 ack_size = TCPOLEN_MPTCP_DSS_ACK32;
605 opts->ext_copy.data_ack32 = (uint32_t)ack_seq;
606 opts->ext_copy.ack64 = 0;
608 opts->ext_copy.use_ack = 1;
609 opts->suboptions = OPTION_MPTCP_DSS;
610 WRITE_ONCE(msk->old_wspace, __mptcp_space((struct sock *)msk));
612 /* Add kind/length/subtype/flag overhead if mapping is not populated */
614 ack_size += TCPOLEN_MPTCP_DSS_BASE;
616 dss_size += ack_size;
618 *size = ALIGN(dss_size, 4);
622 static u64 add_addr_generate_hmac(u64 key1, u64 key2,
623 struct mptcp_addr_info *addr)
625 u16 port = ntohs(addr->port);
626 u8 hmac[SHA256_DIGEST_SIZE];
631 if (addr->family == AF_INET) {
632 memcpy(&msg[i], &addr->addr.s_addr, 4);
635 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
636 else if (addr->family == AF_INET6) {
637 memcpy(&msg[i], &addr->addr6.s6_addr, 16);
641 msg[i++] = port >> 8;
642 msg[i++] = port & 0xFF;
644 mptcp_crypto_hmac_sha(key1, key2, msg, i, hmac);
646 return get_unaligned_be64(&hmac[SHA256_DIGEST_SIZE - sizeof(u64)]);
649 static bool mptcp_established_options_add_addr(struct sock *sk, struct sk_buff *skb,
651 unsigned int remaining,
652 struct mptcp_out_options *opts)
654 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
655 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
656 bool drop_other_suboptions = false;
657 unsigned int opt_size = *size;
661 /* add addr will strip the existing options, be sure to avoid breaking
664 if (!mptcp_pm_should_add_signal(msk) ||
665 (opts->suboptions & (OPTION_MPTCP_MPJ_ACK | OPTION_MPTCP_MPC_ACK)) ||
666 !mptcp_pm_add_addr_signal(msk, skb, opt_size, remaining, &opts->addr,
667 &echo, &drop_other_suboptions))
670 if (drop_other_suboptions)
671 remaining += opt_size;
672 len = mptcp_add_addr_len(opts->addr.family, echo, !!opts->addr.port);
677 if (drop_other_suboptions) {
678 pr_debug("drop other suboptions");
679 opts->suboptions = 0;
681 /* note that e.g. DSS could have written into the memory
682 * aliased by ahmac, we must reset the field here
683 * to avoid appending the hmac even for ADD_ADDR echo
689 opts->suboptions |= OPTION_MPTCP_ADD_ADDR;
691 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ADDADDRTX);
692 opts->ahmac = add_addr_generate_hmac(msk->local_key,
696 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ECHOADDTX);
698 pr_debug("addr_id=%d, ahmac=%llu, echo=%d, port=%d",
699 opts->addr.id, opts->ahmac, echo, ntohs(opts->addr.port));
704 static bool mptcp_established_options_rm_addr(struct sock *sk,
706 unsigned int remaining,
707 struct mptcp_out_options *opts)
709 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
710 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
711 struct mptcp_rm_list rm_list;
714 if (!mptcp_pm_should_rm_signal(msk) ||
715 !(mptcp_pm_rm_addr_signal(msk, remaining, &rm_list)))
718 len = mptcp_rm_addr_len(&rm_list);
725 opts->suboptions |= OPTION_MPTCP_RM_ADDR;
726 opts->rm_list = rm_list;
728 for (i = 0; i < opts->rm_list.nr; i++)
729 pr_debug("rm_list_ids[%d]=%d", i, opts->rm_list.ids[i]);
730 MPTCP_ADD_STATS(sock_net(sk), MPTCP_MIB_RMADDRTX, opts->rm_list.nr);
734 static bool mptcp_established_options_mp_prio(struct sock *sk,
736 unsigned int remaining,
737 struct mptcp_out_options *opts)
739 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
741 /* can't send MP_PRIO with MPC, as they share the same option space:
742 * 'backup'. Also it makes no sense at all
744 if (!subflow->send_mp_prio || (opts->suboptions & OPTIONS_MPTCP_MPC))
747 /* account for the trailing 'nop' option */
748 if (remaining < TCPOLEN_MPTCP_PRIO_ALIGN)
751 *size = TCPOLEN_MPTCP_PRIO_ALIGN;
752 opts->suboptions |= OPTION_MPTCP_PRIO;
753 opts->backup = subflow->request_bkup;
755 pr_debug("prio=%d", opts->backup);
760 static noinline bool mptcp_established_options_rst(struct sock *sk, struct sk_buff *skb,
762 unsigned int remaining,
763 struct mptcp_out_options *opts)
765 const struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
767 if (remaining < TCPOLEN_MPTCP_RST)
770 *size = TCPOLEN_MPTCP_RST;
771 opts->suboptions |= OPTION_MPTCP_RST;
772 opts->reset_transient = subflow->reset_transient;
773 opts->reset_reason = subflow->reset_reason;
774 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPRSTTX);
779 static bool mptcp_established_options_fastclose(struct sock *sk,
781 unsigned int remaining,
782 struct mptcp_out_options *opts)
784 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
785 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
787 if (likely(!subflow->send_fastclose))
790 if (remaining < TCPOLEN_MPTCP_FASTCLOSE)
793 *size = TCPOLEN_MPTCP_FASTCLOSE;
794 opts->suboptions |= OPTION_MPTCP_FASTCLOSE;
795 opts->rcvr_key = msk->remote_key;
797 pr_debug("FASTCLOSE key=%llu", opts->rcvr_key);
798 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPFASTCLOSETX);
802 static bool mptcp_established_options_mp_fail(struct sock *sk,
804 unsigned int remaining,
805 struct mptcp_out_options *opts)
807 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
809 if (likely(!subflow->send_mp_fail))
812 if (remaining < TCPOLEN_MPTCP_FAIL)
815 *size = TCPOLEN_MPTCP_FAIL;
816 opts->suboptions |= OPTION_MPTCP_FAIL;
817 opts->fail_seq = subflow->map_seq;
819 pr_debug("MP_FAIL fail_seq=%llu", opts->fail_seq);
820 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPFAILTX);
825 bool mptcp_established_options(struct sock *sk, struct sk_buff *skb,
826 unsigned int *size, unsigned int remaining,
827 struct mptcp_out_options *opts)
829 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
830 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
831 unsigned int opt_size = 0;
835 opts->suboptions = 0;
837 if (unlikely(__mptcp_check_fallback(msk) && !mptcp_check_infinite_map(skb)))
840 if (unlikely(skb && TCP_SKB_CB(skb)->tcp_flags & TCPHDR_RST)) {
841 if (mptcp_established_options_fastclose(sk, &opt_size, remaining, opts) ||
842 mptcp_established_options_mp_fail(sk, &opt_size, remaining, opts)) {
844 remaining -= opt_size;
846 /* MP_RST can be used with MP_FASTCLOSE and MP_FAIL if there is room */
847 if (mptcp_established_options_rst(sk, skb, &opt_size, remaining, opts)) {
849 remaining -= opt_size;
854 snd_data_fin = mptcp_data_fin_enabled(msk);
855 if (mptcp_established_options_mp(sk, skb, snd_data_fin, &opt_size, opts))
857 else if (mptcp_established_options_dss(sk, skb, snd_data_fin, &opt_size, opts)) {
858 unsigned int mp_fail_size;
861 if (mptcp_established_options_mp_fail(sk, &mp_fail_size,
862 remaining - opt_size, opts)) {
863 *size += opt_size + mp_fail_size;
864 remaining -= opt_size - mp_fail_size;
869 /* we reserved enough space for the above options, and exceeding the
870 * TCP option space would be fatal
872 if (WARN_ON_ONCE(opt_size > remaining))
876 remaining -= opt_size;
877 if (mptcp_established_options_add_addr(sk, skb, &opt_size, remaining, opts)) {
879 remaining -= opt_size;
881 } else if (mptcp_established_options_rm_addr(sk, &opt_size, remaining, opts)) {
883 remaining -= opt_size;
887 if (mptcp_established_options_mp_prio(sk, &opt_size, remaining, opts)) {
889 remaining -= opt_size;
896 bool mptcp_synack_options(const struct request_sock *req, unsigned int *size,
897 struct mptcp_out_options *opts)
899 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
901 if (subflow_req->mp_capable) {
902 opts->suboptions = OPTION_MPTCP_MPC_SYNACK;
903 opts->sndr_key = subflow_req->local_key;
904 opts->csum_reqd = subflow_req->csum_reqd;
905 opts->allow_join_id0 = subflow_req->allow_join_id0;
906 *size = TCPOLEN_MPTCP_MPC_SYNACK;
907 pr_debug("subflow_req=%p, local_key=%llu",
908 subflow_req, subflow_req->local_key);
910 } else if (subflow_req->mp_join) {
911 opts->suboptions = OPTION_MPTCP_MPJ_SYNACK;
912 opts->backup = subflow_req->backup;
913 opts->join_id = subflow_req->local_id;
914 opts->thmac = subflow_req->thmac;
915 opts->nonce = subflow_req->local_nonce;
916 pr_debug("req=%p, bkup=%u, id=%u, thmac=%llu, nonce=%u",
917 subflow_req, opts->backup, opts->join_id,
918 opts->thmac, opts->nonce);
919 *size = TCPOLEN_MPTCP_MPJ_SYNACK;
925 static bool check_fully_established(struct mptcp_sock *msk, struct sock *ssk,
926 struct mptcp_subflow_context *subflow,
928 struct mptcp_options_received *mp_opt)
930 /* here we can process OoO, in-window pkts, only in-sequence 4th ack
931 * will make the subflow fully established
933 if (likely(subflow->fully_established)) {
934 /* on passive sockets, check for 3rd ack retransmission
935 * note that msk is always set by subflow_syn_recv_sock()
936 * for mp_join subflows
938 if (TCP_SKB_CB(skb)->seq == subflow->ssn_offset + 1 &&
939 TCP_SKB_CB(skb)->end_seq == TCP_SKB_CB(skb)->seq &&
940 subflow->mp_join && (mp_opt->suboptions & OPTIONS_MPTCP_MPJ) &&
941 !subflow->request_join)
946 /* we must process OoO packets before the first subflow is fully
947 * established. OoO packets are instead a protocol violation
948 * for MP_JOIN subflows as the peer must not send any data
949 * before receiving the forth ack - cfr. RFC 8684 section 3.2.
951 if (TCP_SKB_CB(skb)->seq != subflow->ssn_offset + 1) {
952 if (subflow->mp_join)
954 if (subflow->is_mptfo && mp_opt->suboptions & OPTION_MPTCP_MPC_ACK)
955 goto set_fully_established;
956 return subflow->mp_capable;
959 if (subflow->remote_key_valid &&
960 (((mp_opt->suboptions & OPTION_MPTCP_DSS) && mp_opt->use_ack) ||
961 ((mp_opt->suboptions & OPTION_MPTCP_ADD_ADDR) && !mp_opt->echo))) {
962 /* subflows are fully established as soon as we get any
963 * additional ack, including ADD_ADDR.
965 goto set_fully_established;
968 /* If the first established packet does not contain MP_CAPABLE + data
969 * then fallback to TCP. Fallback scenarios requires a reset for
972 if (!(mp_opt->suboptions & OPTIONS_MPTCP_MPC)) {
973 if (subflow->mp_join)
975 subflow->mp_capable = 0;
977 mptcp_do_fallback(ssk);
981 if (mp_opt->deny_join_id0)
982 WRITE_ONCE(msk->pm.remote_deny_join_id0, true);
984 set_fully_established:
985 if (unlikely(!READ_ONCE(msk->pm.server_side)))
986 pr_warn_once("bogus mpc option on established client sk");
988 mptcp_data_lock((struct sock *)msk);
989 __mptcp_subflow_fully_established(msk, subflow, mp_opt);
990 mptcp_data_unlock((struct sock *)msk);
993 /* if the subflow is not already linked into the conn_list, we can't
994 * notify the PM: this subflow is still on the listener queue
995 * and the PM possibly acquiring the subflow lock could race with
998 if (likely(subflow->pm_notified) || list_empty(&subflow->node))
1001 subflow->pm_notified = 1;
1002 if (subflow->mp_join) {
1003 clear_3rdack_retransmission(ssk);
1004 mptcp_pm_subflow_established(msk);
1006 mptcp_pm_fully_established(msk, ssk);
1011 mptcp_subflow_reset(ssk);
1015 u64 __mptcp_expand_seq(u64 old_seq, u64 cur_seq)
1017 u32 old_seq32, cur_seq32;
1019 old_seq32 = (u32)old_seq;
1020 cur_seq32 = (u32)cur_seq;
1021 cur_seq = (old_seq & GENMASK_ULL(63, 32)) + cur_seq32;
1022 if (unlikely(cur_seq32 < old_seq32 && before(old_seq32, cur_seq32)))
1023 return cur_seq + (1LL << 32);
1025 /* reverse wrap could happen, too */
1026 if (unlikely(cur_seq32 > old_seq32 && after(old_seq32, cur_seq32)))
1027 return cur_seq - (1LL << 32);
1031 static void __mptcp_snd_una_update(struct mptcp_sock *msk, u64 new_snd_una)
1033 msk->bytes_acked += new_snd_una - msk->snd_una;
1034 msk->snd_una = new_snd_una;
1037 static void ack_update_msk(struct mptcp_sock *msk,
1039 struct mptcp_options_received *mp_opt)
1041 u64 new_wnd_end, new_snd_una, snd_nxt = READ_ONCE(msk->snd_nxt);
1042 struct sock *sk = (struct sock *)msk;
1045 mptcp_data_lock(sk);
1047 /* avoid ack expansion on update conflict, to reduce the risk of
1048 * wrongly expanding to a future ack sequence number, which is way
1049 * more dangerous than missing an ack
1051 old_snd_una = msk->snd_una;
1052 new_snd_una = mptcp_expand_seq(old_snd_una, mp_opt->data_ack, mp_opt->ack64);
1054 /* ACK for data not even sent yet? Ignore.*/
1055 if (unlikely(after64(new_snd_una, snd_nxt)))
1056 new_snd_una = old_snd_una;
1058 new_wnd_end = new_snd_una + tcp_sk(ssk)->snd_wnd;
1060 if (after64(new_wnd_end, msk->wnd_end))
1061 msk->wnd_end = new_wnd_end;
1063 /* this assumes mptcp_incoming_options() is invoked after tcp_ack() */
1064 if (after64(msk->wnd_end, READ_ONCE(msk->snd_nxt)))
1065 __mptcp_check_push(sk, ssk);
1067 if (after64(new_snd_una, old_snd_una)) {
1068 __mptcp_snd_una_update(msk, new_snd_una);
1069 __mptcp_data_acked(sk);
1071 mptcp_data_unlock(sk);
1073 trace_ack_update_msk(mp_opt->data_ack,
1074 old_snd_una, new_snd_una,
1075 new_wnd_end, msk->wnd_end);
1078 bool mptcp_update_rcv_data_fin(struct mptcp_sock *msk, u64 data_fin_seq, bool use_64bit)
1080 /* Skip if DATA_FIN was already received.
1081 * If updating simultaneously with the recvmsg loop, values
1082 * should match. If they mismatch, the peer is misbehaving and
1083 * we will prefer the most recent information.
1085 if (READ_ONCE(msk->rcv_data_fin))
1088 WRITE_ONCE(msk->rcv_data_fin_seq,
1089 mptcp_expand_seq(READ_ONCE(msk->ack_seq), data_fin_seq, use_64bit));
1090 WRITE_ONCE(msk->rcv_data_fin, 1);
1095 static bool add_addr_hmac_valid(struct mptcp_sock *msk,
1096 struct mptcp_options_received *mp_opt)
1103 hmac = add_addr_generate_hmac(msk->remote_key,
1107 pr_debug("msk=%p, ahmac=%llu, mp_opt->ahmac=%llu\n",
1108 msk, hmac, mp_opt->ahmac);
1110 return hmac == mp_opt->ahmac;
1113 /* Return false if a subflow has been reset, else return true */
1114 bool mptcp_incoming_options(struct sock *sk, struct sk_buff *skb)
1116 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1117 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
1118 struct mptcp_options_received mp_opt;
1119 struct mptcp_ext *mpext;
1121 if (__mptcp_check_fallback(msk)) {
1122 /* Keep it simple and unconditionally trigger send data cleanup and
1123 * pending queue spooling. We will need to acquire the data lock
1124 * for more accurate checks, and once the lock is acquired, such
1125 * helpers are cheap.
1127 mptcp_data_lock(subflow->conn);
1128 if (sk_stream_memory_free(sk))
1129 __mptcp_check_push(subflow->conn, sk);
1131 /* on fallback we just need to ignore the msk-level snd_una, as
1132 * this is really plain TCP
1134 __mptcp_snd_una_update(msk, READ_ONCE(msk->snd_nxt));
1136 __mptcp_data_acked(subflow->conn);
1137 mptcp_data_unlock(subflow->conn);
1141 mptcp_get_options(skb, &mp_opt);
1143 /* The subflow can be in close state only if check_fully_established()
1144 * just sent a reset. If so, tell the caller to ignore the current packet.
1146 if (!check_fully_established(msk, sk, subflow, skb, &mp_opt))
1147 return sk->sk_state != TCP_CLOSE;
1149 if (unlikely(mp_opt.suboptions != OPTION_MPTCP_DSS)) {
1150 if ((mp_opt.suboptions & OPTION_MPTCP_FASTCLOSE) &&
1151 msk->local_key == mp_opt.rcvr_key) {
1152 WRITE_ONCE(msk->rcv_fastclose, true);
1153 mptcp_schedule_work((struct sock *)msk);
1154 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPFASTCLOSERX);
1157 if ((mp_opt.suboptions & OPTION_MPTCP_ADD_ADDR) &&
1158 add_addr_hmac_valid(msk, &mp_opt)) {
1160 mptcp_pm_add_addr_received(sk, &mp_opt.addr);
1161 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ADDADDR);
1163 mptcp_pm_add_addr_echoed(msk, &mp_opt.addr);
1164 mptcp_pm_del_add_timer(msk, &mp_opt.addr, true);
1165 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ECHOADD);
1168 if (mp_opt.addr.port)
1169 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_PORTADD);
1172 if (mp_opt.suboptions & OPTION_MPTCP_RM_ADDR)
1173 mptcp_pm_rm_addr_received(msk, &mp_opt.rm_list);
1175 if (mp_opt.suboptions & OPTION_MPTCP_PRIO) {
1176 mptcp_pm_mp_prio_received(sk, mp_opt.backup);
1177 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPPRIORX);
1180 if (mp_opt.suboptions & OPTION_MPTCP_FAIL) {
1181 mptcp_pm_mp_fail_received(sk, mp_opt.fail_seq);
1182 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPFAILRX);
1185 if (mp_opt.suboptions & OPTION_MPTCP_RST) {
1186 subflow->reset_seen = 1;
1187 subflow->reset_reason = mp_opt.reset_reason;
1188 subflow->reset_transient = mp_opt.reset_transient;
1189 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPRSTRX);
1192 if (!(mp_opt.suboptions & OPTION_MPTCP_DSS))
1196 /* we can't wait for recvmsg() to update the ack_seq, otherwise
1197 * monodirectional flows will stuck
1200 ack_update_msk(msk, sk, &mp_opt);
1202 /* Zero-data-length packets are dropped by the caller and not
1203 * propagated to the MPTCP layer, so the skb extension does not
1204 * need to be allocated or populated. DATA_FIN information, if
1205 * present, needs to be updated here before the skb is freed.
1207 if (TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb)->end_seq) {
1208 if (mp_opt.data_fin && mp_opt.data_len == 1 &&
1209 mptcp_update_rcv_data_fin(msk, mp_opt.data_seq, mp_opt.dsn64))
1210 mptcp_schedule_work((struct sock *)msk);
1215 mpext = skb_ext_add(skb, SKB_EXT_MPTCP);
1219 memset(mpext, 0, sizeof(*mpext));
1221 if (likely(mp_opt.use_map)) {
1222 if (mp_opt.mpc_map) {
1223 /* this is an MP_CAPABLE carrying MPTCP data
1224 * we know this map the first chunk of data
1226 mptcp_crypto_key_sha(subflow->remote_key, NULL,
1229 mpext->subflow_seq = 1;
1232 mpext->data_fin = 0;
1234 mpext->data_seq = mp_opt.data_seq;
1235 mpext->subflow_seq = mp_opt.subflow_seq;
1236 mpext->dsn64 = mp_opt.dsn64;
1237 mpext->data_fin = mp_opt.data_fin;
1239 mpext->data_len = mp_opt.data_len;
1241 mpext->csum_reqd = !!(mp_opt.suboptions & OPTION_MPTCP_CSUMREQD);
1243 if (mpext->csum_reqd)
1244 mpext->csum = mp_opt.csum;
1250 static void mptcp_set_rwin(struct tcp_sock *tp, struct tcphdr *th)
1252 const struct sock *ssk = (const struct sock *)tp;
1253 struct mptcp_subflow_context *subflow;
1254 u64 ack_seq, rcv_wnd_old, rcv_wnd_new;
1255 struct mptcp_sock *msk;
1259 subflow = mptcp_subflow_ctx(ssk);
1260 msk = mptcp_sk(subflow->conn);
1262 ack_seq = READ_ONCE(msk->ack_seq);
1263 rcv_wnd_new = ack_seq + tp->rcv_wnd;
1265 rcv_wnd_old = atomic64_read(&msk->rcv_wnd_sent);
1266 if (after64(rcv_wnd_new, rcv_wnd_old)) {
1270 rcv_wnd = atomic64_cmpxchg(&msk->rcv_wnd_sent, rcv_wnd_old, rcv_wnd_new);
1272 if (rcv_wnd == rcv_wnd_old)
1275 rcv_wnd_old = rcv_wnd;
1276 if (before64(rcv_wnd_new, rcv_wnd_old)) {
1277 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_RCVWNDCONFLICTUPDATE);
1280 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_RCVWNDCONFLICT);
1285 if (rcv_wnd_new != rcv_wnd_old) {
1287 win = rcv_wnd_old - ack_seq;
1288 tp->rcv_wnd = min_t(u64, win, U32_MAX);
1289 new_win = tp->rcv_wnd;
1291 /* Make sure we do not exceed the maximum possible
1294 if (unlikely(th->syn))
1295 new_win = min(new_win, 65535U) << tp->rx_opt.rcv_wscale;
1296 if (!tp->rx_opt.rcv_wscale &&
1297 READ_ONCE(sock_net(ssk)->ipv4.sysctl_tcp_workaround_signed_windows))
1298 new_win = min(new_win, MAX_TCP_WINDOW);
1300 new_win = min(new_win, (65535U << tp->rx_opt.rcv_wscale));
1302 /* RFC1323 scaling applied */
1303 new_win >>= tp->rx_opt.rcv_wscale;
1304 th->window = htons(new_win);
1305 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_RCVWNDSHARED);
1309 __sum16 __mptcp_make_csum(u64 data_seq, u32 subflow_seq, u16 data_len, __wsum sum)
1311 struct csum_pseudo_header header;
1314 /* cfr RFC 8684 3.3.1.:
1315 * the data sequence number used in the pseudo-header is
1316 * always the 64-bit value, irrespective of what length is used in the
1317 * DSS option itself.
1319 header.data_seq = cpu_to_be64(data_seq);
1320 header.subflow_seq = htonl(subflow_seq);
1321 header.data_len = htons(data_len);
1324 csum = csum_partial(&header, sizeof(header), sum);
1325 return csum_fold(csum);
1328 static __sum16 mptcp_make_csum(const struct mptcp_ext *mpext)
1330 return __mptcp_make_csum(mpext->data_seq, mpext->subflow_seq, mpext->data_len,
1331 ~csum_unfold(mpext->csum));
1334 static void put_len_csum(u16 len, __sum16 csum, void *data)
1336 __sum16 *sumptr = data + 2;
1339 put_unaligned_be16(len, ptr);
1341 put_unaligned(csum, sumptr);
1344 void mptcp_write_options(struct tcphdr *th, __be32 *ptr, struct tcp_sock *tp,
1345 struct mptcp_out_options *opts)
1347 const struct sock *ssk = (const struct sock *)tp;
1348 struct mptcp_subflow_context *subflow;
1350 /* Which options can be used together?
1352 * X: mutually exclusive
1353 * O: often used together
1354 * C: can be used together in some cases
1355 * P: could be used together but we prefer not to (optimisations)
1357 * Opt: | MPC | MPJ | DSS | ADD | RM | PRIO | FAIL | FC |
1358 * ------|------|------|------|------|------|------|------|------|
1359 * MPC |------|------|------|------|------|------|------|------|
1360 * MPJ | X |------|------|------|------|------|------|------|
1361 * DSS | X | X |------|------|------|------|------|------|
1362 * ADD | X | X | P |------|------|------|------|------|
1363 * RM | C | C | C | P |------|------|------|------|
1364 * PRIO | X | C | C | C | C |------|------|------|
1365 * FAIL | X | X | C | X | X | X |------|------|
1366 * FC | X | X | X | X | X | X | X |------|
1367 * RST | X | X | X | X | X | X | O | O |
1368 * ------|------|------|------|------|------|------|------|------|
1370 * The same applies in mptcp_established_options() function.
1372 if (likely(OPTION_MPTCP_DSS & opts->suboptions)) {
1373 struct mptcp_ext *mpext = &opts->ext_copy;
1374 u8 len = TCPOLEN_MPTCP_DSS_BASE;
1377 if (mpext->use_ack) {
1378 flags = MPTCP_DSS_HAS_ACK;
1380 len += TCPOLEN_MPTCP_DSS_ACK64;
1381 flags |= MPTCP_DSS_ACK64;
1383 len += TCPOLEN_MPTCP_DSS_ACK32;
1387 if (mpext->use_map) {
1388 len += TCPOLEN_MPTCP_DSS_MAP64;
1390 /* Use only 64-bit mapping flags for now, add
1391 * support for optional 32-bit mappings later.
1393 flags |= MPTCP_DSS_HAS_MAP | MPTCP_DSS_DSN64;
1394 if (mpext->data_fin)
1395 flags |= MPTCP_DSS_DATA_FIN;
1397 if (opts->csum_reqd)
1398 len += TCPOLEN_MPTCP_DSS_CHECKSUM;
1401 *ptr++ = mptcp_option(MPTCPOPT_DSS, len, 0, flags);
1403 if (mpext->use_ack) {
1405 put_unaligned_be64(mpext->data_ack, ptr);
1408 put_unaligned_be32(mpext->data_ack32, ptr);
1413 if (mpext->use_map) {
1414 put_unaligned_be64(mpext->data_seq, ptr);
1416 put_unaligned_be32(mpext->subflow_seq, ptr);
1418 if (opts->csum_reqd) {
1419 /* data_len == 0 is reserved for the infinite mapping,
1420 * the checksum will also be set to 0.
1422 put_len_csum(mpext->data_len,
1423 (mpext->data_len ? mptcp_make_csum(mpext) : 0),
1426 put_unaligned_be32(mpext->data_len << 16 |
1427 TCPOPT_NOP << 8 | TCPOPT_NOP, ptr);
1432 /* We might need to add MP_FAIL options in rare cases */
1433 if (unlikely(OPTION_MPTCP_FAIL & opts->suboptions))
1435 } else if (OPTIONS_MPTCP_MPC & opts->suboptions) {
1436 u8 len, flag = MPTCP_CAP_HMAC_SHA256;
1438 if (OPTION_MPTCP_MPC_SYN & opts->suboptions) {
1439 len = TCPOLEN_MPTCP_MPC_SYN;
1440 } else if (OPTION_MPTCP_MPC_SYNACK & opts->suboptions) {
1441 len = TCPOLEN_MPTCP_MPC_SYNACK;
1442 } else if (opts->data_len) {
1443 len = TCPOLEN_MPTCP_MPC_ACK_DATA;
1444 if (opts->csum_reqd)
1445 len += TCPOLEN_MPTCP_DSS_CHECKSUM;
1447 len = TCPOLEN_MPTCP_MPC_ACK;
1450 if (opts->csum_reqd)
1451 flag |= MPTCP_CAP_CHECKSUM_REQD;
1453 if (!opts->allow_join_id0)
1454 flag |= MPTCP_CAP_DENY_JOIN_ID0;
1456 *ptr++ = mptcp_option(MPTCPOPT_MP_CAPABLE, len,
1457 MPTCP_SUPPORTED_VERSION,
1460 if (!((OPTION_MPTCP_MPC_SYNACK | OPTION_MPTCP_MPC_ACK) &
1462 goto mp_capable_done;
1464 put_unaligned_be64(opts->sndr_key, ptr);
1466 if (!((OPTION_MPTCP_MPC_ACK) & opts->suboptions))
1467 goto mp_capable_done;
1469 put_unaligned_be64(opts->rcvr_key, ptr);
1471 if (!opts->data_len)
1472 goto mp_capable_done;
1474 if (opts->csum_reqd) {
1475 put_len_csum(opts->data_len,
1476 __mptcp_make_csum(opts->data_seq,
1479 ~csum_unfold(opts->csum)),
1482 put_unaligned_be32(opts->data_len << 16 |
1483 TCPOPT_NOP << 8 | TCPOPT_NOP, ptr);
1487 /* MPC is additionally mutually exclusive with MP_PRIO */
1488 goto mp_capable_done;
1489 } else if (OPTIONS_MPTCP_MPJ & opts->suboptions) {
1490 if (OPTION_MPTCP_MPJ_SYN & opts->suboptions) {
1491 *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1492 TCPOLEN_MPTCP_MPJ_SYN,
1493 opts->backup, opts->join_id);
1494 put_unaligned_be32(opts->token, ptr);
1496 put_unaligned_be32(opts->nonce, ptr);
1498 } else if (OPTION_MPTCP_MPJ_SYNACK & opts->suboptions) {
1499 *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1500 TCPOLEN_MPTCP_MPJ_SYNACK,
1501 opts->backup, opts->join_id);
1502 put_unaligned_be64(opts->thmac, ptr);
1504 put_unaligned_be32(opts->nonce, ptr);
1507 *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1508 TCPOLEN_MPTCP_MPJ_ACK, 0, 0);
1509 memcpy(ptr, opts->hmac, MPTCPOPT_HMAC_LEN);
1512 } else if (OPTION_MPTCP_ADD_ADDR & opts->suboptions) {
1513 u8 len = TCPOLEN_MPTCP_ADD_ADDR_BASE;
1514 u8 echo = MPTCP_ADDR_ECHO;
1516 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1517 if (opts->addr.family == AF_INET6)
1518 len = TCPOLEN_MPTCP_ADD_ADDR6_BASE;
1521 if (opts->addr.port)
1522 len += TCPOLEN_MPTCP_PORT_LEN;
1525 len += sizeof(opts->ahmac);
1529 *ptr++ = mptcp_option(MPTCPOPT_ADD_ADDR,
1530 len, echo, opts->addr.id);
1531 if (opts->addr.family == AF_INET) {
1532 memcpy((u8 *)ptr, (u8 *)&opts->addr.addr.s_addr, 4);
1535 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1536 else if (opts->addr.family == AF_INET6) {
1537 memcpy((u8 *)ptr, opts->addr.addr6.s6_addr, 16);
1542 if (!opts->addr.port) {
1544 put_unaligned_be64(opts->ahmac, ptr);
1548 u16 port = ntohs(opts->addr.port);
1551 u8 *bptr = (u8 *)ptr;
1553 put_unaligned_be16(port, bptr);
1555 put_unaligned_be64(opts->ahmac, bptr);
1557 put_unaligned_be16(TCPOPT_NOP << 8 |
1562 put_unaligned_be32(port << 16 |
1568 } else if (unlikely(OPTION_MPTCP_FASTCLOSE & opts->suboptions)) {
1569 /* FASTCLOSE is mutually exclusive with others except RST */
1570 *ptr++ = mptcp_option(MPTCPOPT_MP_FASTCLOSE,
1571 TCPOLEN_MPTCP_FASTCLOSE,
1573 put_unaligned_be64(opts->rcvr_key, ptr);
1576 if (OPTION_MPTCP_RST & opts->suboptions)
1579 } else if (unlikely(OPTION_MPTCP_FAIL & opts->suboptions)) {
1581 /* MP_FAIL is mutually exclusive with others except RST */
1582 subflow = mptcp_subflow_ctx(ssk);
1583 subflow->send_mp_fail = 0;
1585 *ptr++ = mptcp_option(MPTCPOPT_MP_FAIL,
1588 put_unaligned_be64(opts->fail_seq, ptr);
1591 if (OPTION_MPTCP_RST & opts->suboptions)
1594 } else if (unlikely(OPTION_MPTCP_RST & opts->suboptions)) {
1596 *ptr++ = mptcp_option(MPTCPOPT_RST,
1598 opts->reset_transient,
1599 opts->reset_reason);
1603 if (OPTION_MPTCP_PRIO & opts->suboptions) {
1604 subflow = mptcp_subflow_ctx(ssk);
1605 subflow->send_mp_prio = 0;
1607 *ptr++ = mptcp_option(MPTCPOPT_MP_PRIO,
1609 opts->backup, TCPOPT_NOP);
1611 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_MPPRIOTX);
1615 if (OPTION_MPTCP_RM_ADDR & opts->suboptions) {
1618 *ptr++ = mptcp_option(MPTCPOPT_RM_ADDR,
1619 TCPOLEN_MPTCP_RM_ADDR_BASE + opts->rm_list.nr,
1620 0, opts->rm_list.ids[0]);
1622 while (i < opts->rm_list.nr) {
1623 u8 id1, id2, id3, id4;
1625 id1 = opts->rm_list.ids[i];
1626 id2 = i + 1 < opts->rm_list.nr ? opts->rm_list.ids[i + 1] : TCPOPT_NOP;
1627 id3 = i + 2 < opts->rm_list.nr ? opts->rm_list.ids[i + 2] : TCPOPT_NOP;
1628 id4 = i + 3 < opts->rm_list.nr ? opts->rm_list.ids[i + 3] : TCPOPT_NOP;
1629 put_unaligned_be32(id1 << 24 | id2 << 16 | id3 << 8 | id4, ptr);
1636 mptcp_set_rwin(tp, th);
1639 __be32 mptcp_get_reset_option(const struct sk_buff *skb)
1641 const struct mptcp_ext *ext = mptcp_get_ext(skb);
1645 flags = ext->reset_transient;
1646 reason = ext->reset_reason;
1648 return mptcp_option(MPTCPOPT_RST, TCPOLEN_MPTCP_RST,
1654 EXPORT_SYMBOL_GPL(mptcp_get_reset_option);