xref: /linux/net/mptcp/options.c (revision 0c8a32eed1625a65798286fb73fea8710a908545)
1 // SPDX-License-Identifier: GPL-2.0
2 /* Multipath TCP
3  *
4  * Copyright (c) 2017 - 2019, Intel Corporation.
5  */
6 
7 #define pr_fmt(fmt) "MPTCP: " fmt
8 
9 #include <linux/kernel.h>
10 #include <crypto/sha2.h>
11 #include <net/tcp.h>
12 #include <net/mptcp.h>
13 #include "protocol.h"
14 #include "mib.h"
15 
16 static bool mptcp_cap_flag_sha256(u8 flags)
17 {
18 	return (flags & MPTCP_CAP_FLAG_MASK) == MPTCP_CAP_HMAC_SHA256;
19 }
20 
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)
24 {
25 	u8 subtype = *ptr >> 4;
26 	int expected_opsize;
27 	u8 version;
28 	u8 flags;
29 
30 	switch (subtype) {
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;
36 			else
37 				expected_opsize = TCPOLEN_MPTCP_MPC_ACK;
38 		} else {
39 			if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_ACK)
40 				expected_opsize = TCPOLEN_MPTCP_MPC_SYNACK;
41 			else
42 				expected_opsize = TCPOLEN_MPTCP_MPC_SYN;
43 		}
44 		if (opsize != expected_opsize)
45 			break;
46 
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)
51 				break;
52 		} else if (version < MPTCP_SUPPORTED_VERSION) {
53 			break;
54 		}
55 
56 		flags = *ptr++;
57 		if (!mptcp_cap_flag_sha256(flags) ||
58 		    (flags & MPTCP_CAP_EXTENSIBILITY))
59 			break;
60 
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."
66 		 *
67 		 * Section 3.3.0:
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."
71 		 *
72 		 * We don't implement DSS checksum - fall back to TCP.
73 		 */
74 		if (flags & MPTCP_CAP_CHECKSUM_REQD)
75 			break;
76 
77 		mp_opt->mp_capable = 1;
78 		if (opsize >= TCPOLEN_MPTCP_MPC_SYNACK) {
79 			mp_opt->sndr_key = get_unaligned_be64(ptr);
80 			ptr += 8;
81 		}
82 		if (opsize >= TCPOLEN_MPTCP_MPC_ACK) {
83 			mp_opt->rcvr_key = get_unaligned_be64(ptr);
84 			ptr += 8;
85 		}
86 		if (opsize == TCPOLEN_MPTCP_MPC_ACK_DATA) {
87 			/* Section 3.1.:
88 			 * "the data parameters in a MP_CAPABLE are semantically
89 			 * equivalent to those in a DSS option and can be used
90 			 * interchangeably."
91 			 */
92 			mp_opt->dss = 1;
93 			mp_opt->use_map = 1;
94 			mp_opt->mpc_map = 1;
95 			mp_opt->data_len = get_unaligned_be16(ptr);
96 			ptr += 2;
97 		}
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);
101 		break;
102 
103 	case MPTCPOPT_MP_JOIN:
104 		mp_opt->mp_join = 1;
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);
109 			ptr += 4;
110 			mp_opt->nonce = get_unaligned_be32(ptr);
111 			ptr += 4;
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);
119 			ptr += 8;
120 			mp_opt->nonce = get_unaligned_be32(ptr);
121 			ptr += 4;
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) {
126 			ptr += 2;
127 			memcpy(mp_opt->hmac, ptr, MPTCPOPT_HMAC_LEN);
128 			pr_debug("MP_JOIN hmac");
129 		} else {
130 			pr_warn("MP_JOIN bad option size");
131 			mp_opt->mp_join = 0;
132 		}
133 		break;
134 
135 	case MPTCPOPT_DSS:
136 		pr_debug("DSS");
137 		ptr++;
138 
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
142 		 */
143 		mp_opt->mpc_map = 0;
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);
150 
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,
154 			 mp_opt->use_ack);
155 
156 		expected_opsize = TCPOLEN_MPTCP_DSS_BASE;
157 
158 		if (mp_opt->use_ack) {
159 			if (mp_opt->ack64)
160 				expected_opsize += TCPOLEN_MPTCP_DSS_ACK64;
161 			else
162 				expected_opsize += TCPOLEN_MPTCP_DSS_ACK32;
163 		}
164 
165 		if (mp_opt->use_map) {
166 			if (mp_opt->dsn64)
167 				expected_opsize += TCPOLEN_MPTCP_DSS_MAP64;
168 			else
169 				expected_opsize += TCPOLEN_MPTCP_DSS_MAP32;
170 		}
171 
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.
176 		 */
177 		if (opsize != expected_opsize &&
178 		    opsize != expected_opsize + TCPOLEN_MPTCP_DSS_CHECKSUM)
179 			break;
180 
181 		mp_opt->dss = 1;
182 
183 		if (mp_opt->use_ack) {
184 			if (mp_opt->ack64) {
185 				mp_opt->data_ack = get_unaligned_be64(ptr);
186 				ptr += 8;
187 			} else {
188 				mp_opt->data_ack = get_unaligned_be32(ptr);
189 				ptr += 4;
190 			}
191 
192 			pr_debug("data_ack=%llu", mp_opt->data_ack);
193 		}
194 
195 		if (mp_opt->use_map) {
196 			if (mp_opt->dsn64) {
197 				mp_opt->data_seq = get_unaligned_be64(ptr);
198 				ptr += 8;
199 			} else {
200 				mp_opt->data_seq = get_unaligned_be32(ptr);
201 				ptr += 4;
202 			}
203 
204 			mp_opt->subflow_seq = get_unaligned_be32(ptr);
205 			ptr += 4;
206 
207 			mp_opt->data_len = get_unaligned_be16(ptr);
208 			ptr += 2;
209 
210 			pr_debug("data_seq=%llu subflow_seq=%u data_len=%u",
211 				 mp_opt->data_seq, mp_opt->subflow_seq,
212 				 mp_opt->data_len);
213 		}
214 
215 		break;
216 
217 	case MPTCPOPT_ADD_ADDR:
218 		mp_opt->echo = (*ptr++) & MPTCP_ADDR_ECHO;
219 		if (!mp_opt->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;
227 #endif
228 			else
229 				break;
230 		} else {
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;
238 #endif
239 			else
240 				break;
241 		}
242 
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);
247 			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);
251 				ptr += 2;
252 			}
253 		}
254 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
255 		else {
256 			memcpy(mp_opt->addr6.s6_addr, (u8 *)ptr, 16);
257 			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);
261 				ptr += 2;
262 			}
263 		}
264 #endif
265 		if (!mp_opt->echo) {
266 			mp_opt->ahmac = get_unaligned_be64(ptr);
267 			ptr += 8;
268 		}
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);
272 		break;
273 
274 	case MPTCPOPT_RM_ADDR:
275 		if (opsize != TCPOLEN_MPTCP_RM_ADDR_BASE)
276 			break;
277 
278 		ptr++;
279 
280 		mp_opt->rm_addr = 1;
281 		mp_opt->rm_id = *ptr++;
282 		pr_debug("RM_ADDR: id=%d", mp_opt->rm_id);
283 		break;
284 
285 	case MPTCPOPT_MP_FASTCLOSE:
286 		if (opsize != TCPOLEN_MPTCP_FASTCLOSE)
287 			break;
288 
289 		ptr += 2;
290 		mp_opt->rcvr_key = get_unaligned_be64(ptr);
291 		ptr += 8;
292 		mp_opt->fastclose = 1;
293 		break;
294 
295 	default:
296 		break;
297 	}
298 }
299 
300 void mptcp_get_options(const struct sk_buff *skb,
301 		       struct mptcp_options_received *mp_opt)
302 {
303 	const struct tcphdr *th = tcp_hdr(skb);
304 	const unsigned char *ptr;
305 	int length;
306 
307 	/* initialize option status */
308 	mp_opt->mp_capable = 0;
309 	mp_opt->mp_join = 0;
310 	mp_opt->add_addr = 0;
311 	mp_opt->ahmac = 0;
312 	mp_opt->fastclose = 0;
313 	mp_opt->port = 0;
314 	mp_opt->rm_addr = 0;
315 	mp_opt->dss = 0;
316 
317 	length = (th->doff * 4) - sizeof(struct tcphdr);
318 	ptr = (const unsigned char *)(th + 1);
319 
320 	while (length > 0) {
321 		int opcode = *ptr++;
322 		int opsize;
323 
324 		switch (opcode) {
325 		case TCPOPT_EOL:
326 			return;
327 		case TCPOPT_NOP:	/* Ref: RFC 793 section 3.1 */
328 			length--;
329 			continue;
330 		default:
331 			opsize = *ptr++;
332 			if (opsize < 2) /* "silly options" */
333 				return;
334 			if (opsize > length)
335 				return;	/* don't parse partial options */
336 			if (opcode == TCPOPT_MPTCP)
337 				mptcp_parse_option(skb, ptr, opsize, mp_opt);
338 			ptr += opsize - 2;
339 			length -= opsize;
340 		}
341 	}
342 }
343 
344 bool mptcp_syn_options(struct sock *sk, const struct sk_buff *skb,
345 		       unsigned int *size, struct mptcp_out_options *opts)
346 {
347 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
348 
349 	/* we will use snd_isn to detect first pkt [re]transmission
350 	 * in mptcp_established_options_mp()
351 	 */
352 	subflow->snd_isn = TCP_SKB_CB(skb)->end_seq;
353 	if (subflow->request_mptcp) {
354 		opts->suboptions = OPTION_MPTCP_MPC_SYN;
355 		*size = TCPOLEN_MPTCP_MPC_SYN;
356 		return true;
357 	} else if (subflow->request_join) {
358 		pr_debug("remote_token=%u, nonce=%u", subflow->remote_token,
359 			 subflow->local_nonce);
360 		opts->suboptions = OPTION_MPTCP_MPJ_SYN;
361 		opts->join_id = subflow->local_id;
362 		opts->token = subflow->remote_token;
363 		opts->nonce = subflow->local_nonce;
364 		opts->backup = subflow->request_bkup;
365 		*size = TCPOLEN_MPTCP_MPJ_SYN;
366 		return true;
367 	}
368 	return false;
369 }
370 
371 /* MP_JOIN client subflow must wait for 4th ack before sending any data:
372  * TCP can't schedule delack timer before the subflow is fully established.
373  * MPTCP uses the delack timer to do 3rd ack retransmissions
374  */
375 static void schedule_3rdack_retransmission(struct sock *sk)
376 {
377 	struct inet_connection_sock *icsk = inet_csk(sk);
378 	struct tcp_sock *tp = tcp_sk(sk);
379 	unsigned long timeout;
380 
381 	/* reschedule with a timeout above RTT, as we must look only for drop */
382 	if (tp->srtt_us)
383 		timeout = tp->srtt_us << 1;
384 	else
385 		timeout = TCP_TIMEOUT_INIT;
386 
387 	WARN_ON_ONCE(icsk->icsk_ack.pending & ICSK_ACK_TIMER);
388 	icsk->icsk_ack.pending |= ICSK_ACK_SCHED | ICSK_ACK_TIMER;
389 	icsk->icsk_ack.timeout = timeout;
390 	sk_reset_timer(sk, &icsk->icsk_delack_timer, timeout);
391 }
392 
393 static void clear_3rdack_retransmission(struct sock *sk)
394 {
395 	struct inet_connection_sock *icsk = inet_csk(sk);
396 
397 	sk_stop_timer(sk, &icsk->icsk_delack_timer);
398 	icsk->icsk_ack.timeout = 0;
399 	icsk->icsk_ack.ato = 0;
400 	icsk->icsk_ack.pending &= ~(ICSK_ACK_SCHED | ICSK_ACK_TIMER);
401 }
402 
403 static bool mptcp_established_options_mp(struct sock *sk, struct sk_buff *skb,
404 					 unsigned int *size,
405 					 unsigned int remaining,
406 					 struct mptcp_out_options *opts)
407 {
408 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
409 	struct mptcp_ext *mpext;
410 	unsigned int data_len;
411 
412 	/* When skb is not available, we better over-estimate the emitted
413 	 * options len. A full DSS option (28 bytes) is longer than
414 	 * TCPOLEN_MPTCP_MPC_ACK_DATA(22) or TCPOLEN_MPTCP_MPJ_ACK(24), so
415 	 * tell the caller to defer the estimate to
416 	 * mptcp_established_options_dss(), which will reserve enough space.
417 	 */
418 	if (!skb)
419 		return false;
420 
421 	/* MPC/MPJ needed only on 3rd ack packet */
422 	if (subflow->fully_established ||
423 	    subflow->snd_isn != TCP_SKB_CB(skb)->seq)
424 		return false;
425 
426 	if (subflow->mp_capable) {
427 		mpext = mptcp_get_ext(skb);
428 		data_len = mpext ? mpext->data_len : 0;
429 
430 		/* we will check ext_copy.data_len in mptcp_write_options() to
431 		 * discriminate between TCPOLEN_MPTCP_MPC_ACK_DATA and
432 		 * TCPOLEN_MPTCP_MPC_ACK
433 		 */
434 		opts->ext_copy.data_len = data_len;
435 		opts->suboptions = OPTION_MPTCP_MPC_ACK;
436 		opts->sndr_key = subflow->local_key;
437 		opts->rcvr_key = subflow->remote_key;
438 
439 		/* Section 3.1.
440 		 * The MP_CAPABLE option is carried on the SYN, SYN/ACK, and ACK
441 		 * packets that start the first subflow of an MPTCP connection,
442 		 * as well as the first packet that carries data
443 		 */
444 		if (data_len > 0)
445 			*size = ALIGN(TCPOLEN_MPTCP_MPC_ACK_DATA, 4);
446 		else
447 			*size = TCPOLEN_MPTCP_MPC_ACK;
448 
449 		pr_debug("subflow=%p, local_key=%llu, remote_key=%llu map_len=%d",
450 			 subflow, subflow->local_key, subflow->remote_key,
451 			 data_len);
452 
453 		return true;
454 	} else if (subflow->mp_join) {
455 		opts->suboptions = OPTION_MPTCP_MPJ_ACK;
456 		memcpy(opts->hmac, subflow->hmac, MPTCPOPT_HMAC_LEN);
457 		*size = TCPOLEN_MPTCP_MPJ_ACK;
458 		pr_debug("subflow=%p", subflow);
459 
460 		schedule_3rdack_retransmission(sk);
461 		return true;
462 	}
463 	return false;
464 }
465 
466 static void mptcp_write_data_fin(struct mptcp_subflow_context *subflow,
467 				 struct sk_buff *skb, struct mptcp_ext *ext)
468 {
469 	/* The write_seq value has already been incremented, so the actual
470 	 * sequence number for the DATA_FIN is one less.
471 	 */
472 	u64 data_fin_tx_seq = READ_ONCE(mptcp_sk(subflow->conn)->write_seq) - 1;
473 
474 	if (!ext->use_map || !skb->len) {
475 		/* RFC6824 requires a DSS mapping with specific values
476 		 * if DATA_FIN is set but no data payload is mapped
477 		 */
478 		ext->data_fin = 1;
479 		ext->use_map = 1;
480 		ext->dsn64 = 1;
481 		ext->data_seq = data_fin_tx_seq;
482 		ext->subflow_seq = 0;
483 		ext->data_len = 1;
484 	} else if (ext->data_seq + ext->data_len == data_fin_tx_seq) {
485 		/* If there's an existing DSS mapping and it is the
486 		 * final mapping, DATA_FIN consumes 1 additional byte of
487 		 * mapping space.
488 		 */
489 		ext->data_fin = 1;
490 		ext->data_len++;
491 	}
492 }
493 
494 static bool mptcp_established_options_dss(struct sock *sk, struct sk_buff *skb,
495 					  unsigned int *size,
496 					  unsigned int remaining,
497 					  struct mptcp_out_options *opts)
498 {
499 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
500 	struct mptcp_sock *msk = mptcp_sk(subflow->conn);
501 	unsigned int dss_size = 0;
502 	u64 snd_data_fin_enable;
503 	struct mptcp_ext *mpext;
504 	unsigned int ack_size;
505 	bool ret = false;
506 
507 	mpext = skb ? mptcp_get_ext(skb) : NULL;
508 	snd_data_fin_enable = mptcp_data_fin_enabled(msk);
509 
510 	if (!skb || (mpext && mpext->use_map) || snd_data_fin_enable) {
511 		unsigned int map_size;
512 
513 		map_size = TCPOLEN_MPTCP_DSS_BASE + TCPOLEN_MPTCP_DSS_MAP64;
514 
515 		remaining -= map_size;
516 		dss_size = map_size;
517 		if (mpext)
518 			opts->ext_copy = *mpext;
519 
520 		if (skb && snd_data_fin_enable)
521 			mptcp_write_data_fin(subflow, skb, &opts->ext_copy);
522 		ret = true;
523 	}
524 
525 	/* passive sockets msk will set the 'can_ack' after accept(), even
526 	 * if the first subflow may have the already the remote key handy
527 	 */
528 	opts->ext_copy.use_ack = 0;
529 	if (!READ_ONCE(msk->can_ack)) {
530 		*size = ALIGN(dss_size, 4);
531 		return ret;
532 	}
533 
534 	if (READ_ONCE(msk->use_64bit_ack)) {
535 		ack_size = TCPOLEN_MPTCP_DSS_ACK64;
536 		opts->ext_copy.data_ack = READ_ONCE(msk->ack_seq);
537 		opts->ext_copy.ack64 = 1;
538 	} else {
539 		ack_size = TCPOLEN_MPTCP_DSS_ACK32;
540 		opts->ext_copy.data_ack32 = (uint32_t)READ_ONCE(msk->ack_seq);
541 		opts->ext_copy.ack64 = 0;
542 	}
543 	opts->ext_copy.use_ack = 1;
544 	WRITE_ONCE(msk->old_wspace, __mptcp_space((struct sock *)msk));
545 
546 	/* Add kind/length/subtype/flag overhead if mapping is not populated */
547 	if (dss_size == 0)
548 		ack_size += TCPOLEN_MPTCP_DSS_BASE;
549 
550 	dss_size += ack_size;
551 
552 	*size = ALIGN(dss_size, 4);
553 	return true;
554 }
555 
556 static u64 add_addr_generate_hmac(u64 key1, u64 key2, u8 addr_id,
557 				  struct in_addr *addr)
558 {
559 	u8 hmac[SHA256_DIGEST_SIZE];
560 	u8 msg[7];
561 
562 	msg[0] = addr_id;
563 	memcpy(&msg[1], &addr->s_addr, 4);
564 	msg[5] = 0;
565 	msg[6] = 0;
566 
567 	mptcp_crypto_hmac_sha(key1, key2, msg, 7, hmac);
568 
569 	return get_unaligned_be64(&hmac[SHA256_DIGEST_SIZE - sizeof(u64)]);
570 }
571 
572 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
573 static u64 add_addr6_generate_hmac(u64 key1, u64 key2, u8 addr_id,
574 				   struct in6_addr *addr)
575 {
576 	u8 hmac[SHA256_DIGEST_SIZE];
577 	u8 msg[19];
578 
579 	msg[0] = addr_id;
580 	memcpy(&msg[1], &addr->s6_addr, 16);
581 	msg[17] = 0;
582 	msg[18] = 0;
583 
584 	mptcp_crypto_hmac_sha(key1, key2, msg, 19, hmac);
585 
586 	return get_unaligned_be64(&hmac[SHA256_DIGEST_SIZE - sizeof(u64)]);
587 }
588 #endif
589 
590 static bool mptcp_established_options_add_addr(struct sock *sk, struct sk_buff *skb,
591 					       unsigned int *size,
592 					       unsigned int remaining,
593 					       struct mptcp_out_options *opts)
594 {
595 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
596 	struct mptcp_sock *msk = mptcp_sk(subflow->conn);
597 	bool drop_other_suboptions = false;
598 	unsigned int opt_size = *size;
599 	struct mptcp_addr_info saddr;
600 	bool echo;
601 	bool port;
602 	int len;
603 
604 	if ((mptcp_pm_should_add_signal_ipv6(msk) ||
605 	     mptcp_pm_should_add_signal_port(msk)) &&
606 	    skb && skb_is_tcp_pure_ack(skb)) {
607 		pr_debug("drop other suboptions");
608 		opts->suboptions = 0;
609 		opts->ext_copy.use_ack = 0;
610 		opts->ext_copy.use_map = 0;
611 		remaining += opt_size;
612 		drop_other_suboptions = true;
613 	}
614 
615 	if (!mptcp_pm_should_add_signal(msk) ||
616 	    !(mptcp_pm_add_addr_signal(msk, remaining, &saddr, &echo, &port)))
617 		return false;
618 
619 	len = mptcp_add_addr_len(saddr.family, echo, port);
620 	if (remaining < len)
621 		return false;
622 
623 	*size = len;
624 	if (drop_other_suboptions)
625 		*size -= opt_size;
626 	opts->addr_id = saddr.id;
627 	if (port)
628 		opts->port = ntohs(saddr.port);
629 	if (saddr.family == AF_INET) {
630 		opts->suboptions |= OPTION_MPTCP_ADD_ADDR;
631 		opts->addr = saddr.addr;
632 		if (!echo) {
633 			opts->ahmac = add_addr_generate_hmac(msk->local_key,
634 							     msk->remote_key,
635 							     opts->addr_id,
636 							     &opts->addr);
637 		}
638 	}
639 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
640 	else if (saddr.family == AF_INET6) {
641 		opts->suboptions |= OPTION_MPTCP_ADD_ADDR6;
642 		opts->addr6 = saddr.addr6;
643 		if (!echo) {
644 			opts->ahmac = add_addr6_generate_hmac(msk->local_key,
645 							      msk->remote_key,
646 							      opts->addr_id,
647 							      &opts->addr6);
648 		}
649 	}
650 #endif
651 	pr_debug("addr_id=%d, ahmac=%llu, echo=%d, port=%d",
652 		 opts->addr_id, opts->ahmac, echo, opts->port);
653 
654 	return true;
655 }
656 
657 static bool mptcp_established_options_rm_addr(struct sock *sk,
658 					      unsigned int *size,
659 					      unsigned int remaining,
660 					      struct mptcp_out_options *opts)
661 {
662 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
663 	struct mptcp_sock *msk = mptcp_sk(subflow->conn);
664 	u8 rm_id;
665 
666 	if (!mptcp_pm_should_rm_signal(msk) ||
667 	    !(mptcp_pm_rm_addr_signal(msk, remaining, &rm_id)))
668 		return false;
669 
670 	if (remaining < TCPOLEN_MPTCP_RM_ADDR_BASE)
671 		return false;
672 
673 	*size = TCPOLEN_MPTCP_RM_ADDR_BASE;
674 	opts->suboptions |= OPTION_MPTCP_RM_ADDR;
675 	opts->rm_id = rm_id;
676 
677 	pr_debug("rm_id=%d", opts->rm_id);
678 
679 	return true;
680 }
681 
682 bool mptcp_established_options(struct sock *sk, struct sk_buff *skb,
683 			       unsigned int *size, unsigned int remaining,
684 			       struct mptcp_out_options *opts)
685 {
686 	unsigned int opt_size = 0;
687 	bool ret = false;
688 
689 	opts->suboptions = 0;
690 
691 	if (unlikely(mptcp_check_fallback(sk)))
692 		return false;
693 
694 	/* prevent adding of any MPTCP related options on reset packet
695 	 * until we support MP_TCPRST/MP_FASTCLOSE
696 	 */
697 	if (unlikely(skb && TCP_SKB_CB(skb)->tcp_flags & TCPHDR_RST))
698 		return false;
699 
700 	if (mptcp_established_options_mp(sk, skb, &opt_size, remaining, opts))
701 		ret = true;
702 	else if (mptcp_established_options_dss(sk, skb, &opt_size, remaining,
703 					       opts))
704 		ret = true;
705 
706 	/* we reserved enough space for the above options, and exceeding the
707 	 * TCP option space would be fatal
708 	 */
709 	if (WARN_ON_ONCE(opt_size > remaining))
710 		return false;
711 
712 	*size += opt_size;
713 	remaining -= opt_size;
714 	if (mptcp_established_options_add_addr(sk, skb, &opt_size, remaining, opts)) {
715 		*size += opt_size;
716 		remaining -= opt_size;
717 		ret = true;
718 	} else if (mptcp_established_options_rm_addr(sk, &opt_size, remaining, opts)) {
719 		*size += opt_size;
720 		remaining -= opt_size;
721 		ret = true;
722 	}
723 
724 	return ret;
725 }
726 
727 bool mptcp_synack_options(const struct request_sock *req, unsigned int *size,
728 			  struct mptcp_out_options *opts)
729 {
730 	struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
731 
732 	if (subflow_req->mp_capable) {
733 		opts->suboptions = OPTION_MPTCP_MPC_SYNACK;
734 		opts->sndr_key = subflow_req->local_key;
735 		*size = TCPOLEN_MPTCP_MPC_SYNACK;
736 		pr_debug("subflow_req=%p, local_key=%llu",
737 			 subflow_req, subflow_req->local_key);
738 		return true;
739 	} else if (subflow_req->mp_join) {
740 		opts->suboptions = OPTION_MPTCP_MPJ_SYNACK;
741 		opts->backup = subflow_req->backup;
742 		opts->join_id = subflow_req->local_id;
743 		opts->thmac = subflow_req->thmac;
744 		opts->nonce = subflow_req->local_nonce;
745 		pr_debug("req=%p, bkup=%u, id=%u, thmac=%llu, nonce=%u",
746 			 subflow_req, opts->backup, opts->join_id,
747 			 opts->thmac, opts->nonce);
748 		*size = TCPOLEN_MPTCP_MPJ_SYNACK;
749 		return true;
750 	}
751 	return false;
752 }
753 
754 static bool check_fully_established(struct mptcp_sock *msk, struct sock *ssk,
755 				    struct mptcp_subflow_context *subflow,
756 				    struct sk_buff *skb,
757 				    struct mptcp_options_received *mp_opt)
758 {
759 	/* here we can process OoO, in-window pkts, only in-sequence 4th ack
760 	 * will make the subflow fully established
761 	 */
762 	if (likely(subflow->fully_established)) {
763 		/* on passive sockets, check for 3rd ack retransmission
764 		 * note that msk is always set by subflow_syn_recv_sock()
765 		 * for mp_join subflows
766 		 */
767 		if (TCP_SKB_CB(skb)->seq == subflow->ssn_offset + 1 &&
768 		    TCP_SKB_CB(skb)->end_seq == TCP_SKB_CB(skb)->seq &&
769 		    subflow->mp_join && mp_opt->mp_join &&
770 		    READ_ONCE(msk->pm.server_side))
771 			tcp_send_ack(ssk);
772 		goto fully_established;
773 	}
774 
775 	/* we must process OoO packets before the first subflow is fully
776 	 * established. OoO packets are instead a protocol violation
777 	 * for MP_JOIN subflows as the peer must not send any data
778 	 * before receiving the forth ack - cfr. RFC 8684 section 3.2.
779 	 */
780 	if (TCP_SKB_CB(skb)->seq != subflow->ssn_offset + 1) {
781 		if (subflow->mp_join)
782 			goto reset;
783 		return subflow->mp_capable;
784 	}
785 
786 	if (mp_opt->dss && mp_opt->use_ack) {
787 		/* subflows are fully established as soon as we get any
788 		 * additional ack.
789 		 */
790 		subflow->fully_established = 1;
791 		WRITE_ONCE(msk->fully_established, true);
792 		goto fully_established;
793 	}
794 
795 	if (mp_opt->add_addr) {
796 		WRITE_ONCE(msk->fully_established, true);
797 		return true;
798 	}
799 
800 	/* If the first established packet does not contain MP_CAPABLE + data
801 	 * then fallback to TCP. Fallback scenarios requires a reset for
802 	 * MP_JOIN subflows.
803 	 */
804 	if (!mp_opt->mp_capable) {
805 		if (subflow->mp_join)
806 			goto reset;
807 		subflow->mp_capable = 0;
808 		pr_fallback(msk);
809 		__mptcp_do_fallback(msk);
810 		return false;
811 	}
812 
813 	if (unlikely(!READ_ONCE(msk->pm.server_side)))
814 		pr_warn_once("bogus mpc option on established client sk");
815 	mptcp_subflow_fully_established(subflow, mp_opt);
816 
817 fully_established:
818 	/* if the subflow is not already linked into the conn_list, we can't
819 	 * notify the PM: this subflow is still on the listener queue
820 	 * and the PM possibly acquiring the subflow lock could race with
821 	 * the listener close
822 	 */
823 	if (likely(subflow->pm_notified) || list_empty(&subflow->node))
824 		return true;
825 
826 	subflow->pm_notified = 1;
827 	if (subflow->mp_join) {
828 		clear_3rdack_retransmission(ssk);
829 		mptcp_pm_subflow_established(msk, subflow);
830 	} else {
831 		mptcp_pm_fully_established(msk);
832 	}
833 	return true;
834 
835 reset:
836 	mptcp_subflow_reset(ssk);
837 	return false;
838 }
839 
840 static u64 expand_ack(u64 old_ack, u64 cur_ack, bool use_64bit)
841 {
842 	u32 old_ack32, cur_ack32;
843 
844 	if (use_64bit)
845 		return cur_ack;
846 
847 	old_ack32 = (u32)old_ack;
848 	cur_ack32 = (u32)cur_ack;
849 	cur_ack = (old_ack & GENMASK_ULL(63, 32)) + cur_ack32;
850 	if (unlikely(before(cur_ack32, old_ack32)))
851 		return cur_ack + (1LL << 32);
852 	return cur_ack;
853 }
854 
855 static void ack_update_msk(struct mptcp_sock *msk,
856 			   struct sock *ssk,
857 			   struct mptcp_options_received *mp_opt)
858 {
859 	u64 new_wnd_end, new_snd_una, snd_nxt = READ_ONCE(msk->snd_nxt);
860 	struct sock *sk = (struct sock *)msk;
861 	u64 old_snd_una;
862 
863 	mptcp_data_lock(sk);
864 
865 	/* avoid ack expansion on update conflict, to reduce the risk of
866 	 * wrongly expanding to a future ack sequence number, which is way
867 	 * more dangerous than missing an ack
868 	 */
869 	old_snd_una = msk->snd_una;
870 	new_snd_una = expand_ack(old_snd_una, mp_opt->data_ack, mp_opt->ack64);
871 
872 	/* ACK for data not even sent yet? Ignore. */
873 	if (after64(new_snd_una, snd_nxt))
874 		new_snd_una = old_snd_una;
875 
876 	new_wnd_end = new_snd_una + tcp_sk(ssk)->snd_wnd;
877 
878 	if (after64(new_wnd_end, msk->wnd_end))
879 		msk->wnd_end = new_wnd_end;
880 
881 	/* this assumes mptcp_incoming_options() is invoked after tcp_ack() */
882 	if (after64(msk->wnd_end, READ_ONCE(msk->snd_nxt)) &&
883 	    sk_stream_memory_free(ssk))
884 		__mptcp_check_push(sk, ssk);
885 
886 	if (after64(new_snd_una, old_snd_una)) {
887 		msk->snd_una = new_snd_una;
888 		__mptcp_data_acked(sk);
889 	}
890 	mptcp_data_unlock(sk);
891 }
892 
893 bool mptcp_update_rcv_data_fin(struct mptcp_sock *msk, u64 data_fin_seq, bool use_64bit)
894 {
895 	/* Skip if DATA_FIN was already received.
896 	 * If updating simultaneously with the recvmsg loop, values
897 	 * should match. If they mismatch, the peer is misbehaving and
898 	 * we will prefer the most recent information.
899 	 */
900 	if (READ_ONCE(msk->rcv_data_fin) || !READ_ONCE(msk->first))
901 		return false;
902 
903 	WRITE_ONCE(msk->rcv_data_fin_seq,
904 		   expand_ack(READ_ONCE(msk->ack_seq), data_fin_seq, use_64bit));
905 	WRITE_ONCE(msk->rcv_data_fin, 1);
906 
907 	return true;
908 }
909 
910 static bool add_addr_hmac_valid(struct mptcp_sock *msk,
911 				struct mptcp_options_received *mp_opt)
912 {
913 	u64 hmac = 0;
914 
915 	if (mp_opt->echo)
916 		return true;
917 
918 	if (mp_opt->family == MPTCP_ADDR_IPVERSION_4)
919 		hmac = add_addr_generate_hmac(msk->remote_key,
920 					      msk->local_key,
921 					      mp_opt->addr_id, &mp_opt->addr);
922 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
923 	else
924 		hmac = add_addr6_generate_hmac(msk->remote_key,
925 					       msk->local_key,
926 					       mp_opt->addr_id, &mp_opt->addr6);
927 #endif
928 
929 	pr_debug("msk=%p, ahmac=%llu, mp_opt->ahmac=%llu\n",
930 		 msk, (unsigned long long)hmac,
931 		 (unsigned long long)mp_opt->ahmac);
932 
933 	return hmac == mp_opt->ahmac;
934 }
935 
936 void mptcp_incoming_options(struct sock *sk, struct sk_buff *skb)
937 {
938 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
939 	struct mptcp_sock *msk = mptcp_sk(subflow->conn);
940 	struct mptcp_options_received mp_opt;
941 	struct mptcp_ext *mpext;
942 
943 	if (__mptcp_check_fallback(msk)) {
944 		/* Keep it simple and unconditionally trigger send data cleanup and
945 		 * pending queue spooling. We will need to acquire the data lock
946 		 * for more accurate checks, and once the lock is acquired, such
947 		 * helpers are cheap.
948 		 */
949 		mptcp_data_lock(subflow->conn);
950 		if (sk_stream_memory_free(sk))
951 			__mptcp_check_push(subflow->conn, sk);
952 		__mptcp_data_acked(subflow->conn);
953 		mptcp_data_unlock(subflow->conn);
954 		return;
955 	}
956 
957 	mptcp_get_options(skb, &mp_opt);
958 	if (!check_fully_established(msk, sk, subflow, skb, &mp_opt))
959 		return;
960 
961 	if (mp_opt.fastclose &&
962 	    msk->local_key == mp_opt.rcvr_key) {
963 		WRITE_ONCE(msk->rcv_fastclose, true);
964 		mptcp_schedule_work((struct sock *)msk);
965 	}
966 
967 	if (mp_opt.add_addr && add_addr_hmac_valid(msk, &mp_opt)) {
968 		struct mptcp_addr_info addr;
969 
970 		addr.port = htons(mp_opt.port);
971 		addr.id = mp_opt.addr_id;
972 		if (mp_opt.family == MPTCP_ADDR_IPVERSION_4) {
973 			addr.family = AF_INET;
974 			addr.addr = mp_opt.addr;
975 		}
976 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
977 		else if (mp_opt.family == MPTCP_ADDR_IPVERSION_6) {
978 			addr.family = AF_INET6;
979 			addr.addr6 = mp_opt.addr6;
980 		}
981 #endif
982 		if (!mp_opt.echo) {
983 			mptcp_pm_add_addr_received(msk, &addr);
984 			MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ADDADDR);
985 		} else {
986 			mptcp_pm_del_add_timer(msk, &addr);
987 			MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ECHOADD);
988 		}
989 		mp_opt.add_addr = 0;
990 	}
991 
992 	if (mp_opt.rm_addr) {
993 		mptcp_pm_rm_addr_received(msk, mp_opt.rm_id);
994 		mp_opt.rm_addr = 0;
995 	}
996 
997 	if (!mp_opt.dss)
998 		return;
999 
1000 	/* we can't wait for recvmsg() to update the ack_seq, otherwise
1001 	 * monodirectional flows will stuck
1002 	 */
1003 	if (mp_opt.use_ack)
1004 		ack_update_msk(msk, sk, &mp_opt);
1005 
1006 	/* Zero-data-length packets are dropped by the caller and not
1007 	 * propagated to the MPTCP layer, so the skb extension does not
1008 	 * need to be allocated or populated. DATA_FIN information, if
1009 	 * present, needs to be updated here before the skb is freed.
1010 	 */
1011 	if (TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb)->end_seq) {
1012 		if (mp_opt.data_fin && mp_opt.data_len == 1 &&
1013 		    mptcp_update_rcv_data_fin(msk, mp_opt.data_seq, mp_opt.dsn64) &&
1014 		    schedule_work(&msk->work))
1015 			sock_hold(subflow->conn);
1016 
1017 		return;
1018 	}
1019 
1020 	mpext = skb_ext_add(skb, SKB_EXT_MPTCP);
1021 	if (!mpext)
1022 		return;
1023 
1024 	memset(mpext, 0, sizeof(*mpext));
1025 
1026 	if (mp_opt.use_map) {
1027 		if (mp_opt.mpc_map) {
1028 			/* this is an MP_CAPABLE carrying MPTCP data
1029 			 * we know this map the first chunk of data
1030 			 */
1031 			mptcp_crypto_key_sha(subflow->remote_key, NULL,
1032 					     &mpext->data_seq);
1033 			mpext->data_seq++;
1034 			mpext->subflow_seq = 1;
1035 			mpext->dsn64 = 1;
1036 			mpext->mpc_map = 1;
1037 			mpext->data_fin = 0;
1038 		} else {
1039 			mpext->data_seq = mp_opt.data_seq;
1040 			mpext->subflow_seq = mp_opt.subflow_seq;
1041 			mpext->dsn64 = mp_opt.dsn64;
1042 			mpext->data_fin = mp_opt.data_fin;
1043 		}
1044 		mpext->data_len = mp_opt.data_len;
1045 		mpext->use_map = 1;
1046 	}
1047 }
1048 
1049 static void mptcp_set_rwin(const struct tcp_sock *tp)
1050 {
1051 	const struct sock *ssk = (const struct sock *)tp;
1052 	const struct mptcp_subflow_context *subflow;
1053 	struct mptcp_sock *msk;
1054 	u64 ack_seq;
1055 
1056 	subflow = mptcp_subflow_ctx(ssk);
1057 	msk = mptcp_sk(subflow->conn);
1058 
1059 	ack_seq = READ_ONCE(msk->ack_seq) + tp->rcv_wnd;
1060 
1061 	if (after64(ack_seq, READ_ONCE(msk->rcv_wnd_sent)))
1062 		WRITE_ONCE(msk->rcv_wnd_sent, ack_seq);
1063 }
1064 
1065 void mptcp_write_options(__be32 *ptr, const struct tcp_sock *tp,
1066 			 struct mptcp_out_options *opts)
1067 {
1068 	if ((OPTION_MPTCP_MPC_SYN | OPTION_MPTCP_MPC_SYNACK |
1069 	     OPTION_MPTCP_MPC_ACK) & opts->suboptions) {
1070 		u8 len;
1071 
1072 		if (OPTION_MPTCP_MPC_SYN & opts->suboptions)
1073 			len = TCPOLEN_MPTCP_MPC_SYN;
1074 		else if (OPTION_MPTCP_MPC_SYNACK & opts->suboptions)
1075 			len = TCPOLEN_MPTCP_MPC_SYNACK;
1076 		else if (opts->ext_copy.data_len)
1077 			len = TCPOLEN_MPTCP_MPC_ACK_DATA;
1078 		else
1079 			len = TCPOLEN_MPTCP_MPC_ACK;
1080 
1081 		*ptr++ = mptcp_option(MPTCPOPT_MP_CAPABLE, len,
1082 				      MPTCP_SUPPORTED_VERSION,
1083 				      MPTCP_CAP_HMAC_SHA256);
1084 
1085 		if (!((OPTION_MPTCP_MPC_SYNACK | OPTION_MPTCP_MPC_ACK) &
1086 		    opts->suboptions))
1087 			goto mp_capable_done;
1088 
1089 		put_unaligned_be64(opts->sndr_key, ptr);
1090 		ptr += 2;
1091 		if (!((OPTION_MPTCP_MPC_ACK) & opts->suboptions))
1092 			goto mp_capable_done;
1093 
1094 		put_unaligned_be64(opts->rcvr_key, ptr);
1095 		ptr += 2;
1096 		if (!opts->ext_copy.data_len)
1097 			goto mp_capable_done;
1098 
1099 		put_unaligned_be32(opts->ext_copy.data_len << 16 |
1100 				   TCPOPT_NOP << 8 | TCPOPT_NOP, ptr);
1101 		ptr += 1;
1102 	}
1103 
1104 mp_capable_done:
1105 	if ((OPTION_MPTCP_ADD_ADDR
1106 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1107 	     | OPTION_MPTCP_ADD_ADDR6
1108 #endif
1109 	    ) & opts->suboptions) {
1110 		u8 len = TCPOLEN_MPTCP_ADD_ADDR_BASE;
1111 		u8 echo = MPTCP_ADDR_ECHO;
1112 
1113 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1114 		if (OPTION_MPTCP_ADD_ADDR6 & opts->suboptions)
1115 			len = TCPOLEN_MPTCP_ADD_ADDR6_BASE;
1116 #endif
1117 
1118 		if (opts->port)
1119 			len += TCPOLEN_MPTCP_PORT_LEN;
1120 
1121 		if (opts->ahmac) {
1122 			len += sizeof(opts->ahmac);
1123 			echo = 0;
1124 		}
1125 
1126 		*ptr++ = mptcp_option(MPTCPOPT_ADD_ADDR,
1127 				      len, echo, opts->addr_id);
1128 		if (OPTION_MPTCP_ADD_ADDR & opts->suboptions) {
1129 			memcpy((u8 *)ptr, (u8 *)&opts->addr.s_addr, 4);
1130 			ptr += 1;
1131 		}
1132 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1133 		else if (OPTION_MPTCP_ADD_ADDR6 & opts->suboptions) {
1134 			memcpy((u8 *)ptr, opts->addr6.s6_addr, 16);
1135 			ptr += 4;
1136 		}
1137 #endif
1138 
1139 		if (!opts->port) {
1140 			if (opts->ahmac) {
1141 				put_unaligned_be64(opts->ahmac, ptr);
1142 				ptr += 2;
1143 			}
1144 		} else {
1145 			if (opts->ahmac) {
1146 				u8 *bptr = (u8 *)ptr;
1147 
1148 				put_unaligned_be16(opts->port, bptr);
1149 				bptr += 2;
1150 				put_unaligned_be64(opts->ahmac, bptr);
1151 				bptr += 8;
1152 				put_unaligned_be16(TCPOPT_NOP << 8 |
1153 						   TCPOPT_NOP, bptr);
1154 
1155 				ptr += 3;
1156 			} else {
1157 				put_unaligned_be32(opts->port << 16 |
1158 						   TCPOPT_NOP << 8 |
1159 						   TCPOPT_NOP, ptr);
1160 				ptr += 1;
1161 			}
1162 		}
1163 	}
1164 
1165 	if (OPTION_MPTCP_RM_ADDR & opts->suboptions) {
1166 		*ptr++ = mptcp_option(MPTCPOPT_RM_ADDR,
1167 				      TCPOLEN_MPTCP_RM_ADDR_BASE,
1168 				      0, opts->rm_id);
1169 	}
1170 
1171 	if (OPTION_MPTCP_MPJ_SYN & opts->suboptions) {
1172 		*ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1173 				      TCPOLEN_MPTCP_MPJ_SYN,
1174 				      opts->backup, opts->join_id);
1175 		put_unaligned_be32(opts->token, ptr);
1176 		ptr += 1;
1177 		put_unaligned_be32(opts->nonce, ptr);
1178 		ptr += 1;
1179 	}
1180 
1181 	if (OPTION_MPTCP_MPJ_SYNACK & opts->suboptions) {
1182 		*ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1183 				      TCPOLEN_MPTCP_MPJ_SYNACK,
1184 				      opts->backup, opts->join_id);
1185 		put_unaligned_be64(opts->thmac, ptr);
1186 		ptr += 2;
1187 		put_unaligned_be32(opts->nonce, ptr);
1188 		ptr += 1;
1189 	}
1190 
1191 	if (OPTION_MPTCP_MPJ_ACK & opts->suboptions) {
1192 		*ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1193 				      TCPOLEN_MPTCP_MPJ_ACK, 0, 0);
1194 		memcpy(ptr, opts->hmac, MPTCPOPT_HMAC_LEN);
1195 		ptr += 5;
1196 	}
1197 
1198 	if (opts->ext_copy.use_ack || opts->ext_copy.use_map) {
1199 		struct mptcp_ext *mpext = &opts->ext_copy;
1200 		u8 len = TCPOLEN_MPTCP_DSS_BASE;
1201 		u8 flags = 0;
1202 
1203 		if (mpext->use_ack) {
1204 			flags = MPTCP_DSS_HAS_ACK;
1205 			if (mpext->ack64) {
1206 				len += TCPOLEN_MPTCP_DSS_ACK64;
1207 				flags |= MPTCP_DSS_ACK64;
1208 			} else {
1209 				len += TCPOLEN_MPTCP_DSS_ACK32;
1210 			}
1211 		}
1212 
1213 		if (mpext->use_map) {
1214 			len += TCPOLEN_MPTCP_DSS_MAP64;
1215 
1216 			/* Use only 64-bit mapping flags for now, add
1217 			 * support for optional 32-bit mappings later.
1218 			 */
1219 			flags |= MPTCP_DSS_HAS_MAP | MPTCP_DSS_DSN64;
1220 			if (mpext->data_fin)
1221 				flags |= MPTCP_DSS_DATA_FIN;
1222 		}
1223 
1224 		*ptr++ = mptcp_option(MPTCPOPT_DSS, len, 0, flags);
1225 
1226 		if (mpext->use_ack) {
1227 			if (mpext->ack64) {
1228 				put_unaligned_be64(mpext->data_ack, ptr);
1229 				ptr += 2;
1230 			} else {
1231 				put_unaligned_be32(mpext->data_ack32, ptr);
1232 				ptr += 1;
1233 			}
1234 		}
1235 
1236 		if (mpext->use_map) {
1237 			put_unaligned_be64(mpext->data_seq, ptr);
1238 			ptr += 2;
1239 			put_unaligned_be32(mpext->subflow_seq, ptr);
1240 			ptr += 1;
1241 			put_unaligned_be32(mpext->data_len << 16 |
1242 					   TCPOPT_NOP << 8 | TCPOPT_NOP, ptr);
1243 		}
1244 	}
1245 
1246 	if (tp)
1247 		mptcp_set_rwin(tp);
1248 }
1249