xref: /openbmc/linux/net/mptcp/options.c (revision 5b448065)
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 #include <trace/events/mptcp.h>
17 
18 static bool mptcp_cap_flag_sha256(u8 flags)
19 {
20 	return (flags & MPTCP_CAP_FLAG_MASK) == MPTCP_CAP_HMAC_SHA256;
21 }
22 
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)
26 {
27 	u8 subtype = *ptr >> 4;
28 	int expected_opsize;
29 	u8 version;
30 	u8 flags;
31 	u8 i;
32 
33 	switch (subtype) {
34 	case MPTCPOPT_MP_CAPABLE:
35 		/* strict size checking */
36 		if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)) {
37 			if (skb->len > tcp_hdr(skb)->doff << 2)
38 				expected_opsize = TCPOLEN_MPTCP_MPC_ACK_DATA;
39 			else
40 				expected_opsize = TCPOLEN_MPTCP_MPC_ACK;
41 		} else {
42 			if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_ACK)
43 				expected_opsize = TCPOLEN_MPTCP_MPC_SYNACK;
44 			else
45 				expected_opsize = TCPOLEN_MPTCP_MPC_SYN;
46 		}
47 		if (opsize != expected_opsize)
48 			break;
49 
50 		/* try to be gentle vs future versions on the initial syn */
51 		version = *ptr++ & MPTCP_VERSION_MASK;
52 		if (opsize != TCPOLEN_MPTCP_MPC_SYN) {
53 			if (version != MPTCP_SUPPORTED_VERSION)
54 				break;
55 		} else if (version < MPTCP_SUPPORTED_VERSION) {
56 			break;
57 		}
58 
59 		flags = *ptr++;
60 		if (!mptcp_cap_flag_sha256(flags) ||
61 		    (flags & MPTCP_CAP_EXTENSIBILITY))
62 			break;
63 
64 		/* RFC 6824, Section 3.1:
65 		 * "For the Checksum Required bit (labeled "A"), if either
66 		 * host requires the use of checksums, checksums MUST be used.
67 		 * In other words, the only way for checksums not to be used
68 		 * is if both hosts in their SYNs set A=0."
69 		 *
70 		 * Section 3.3.0:
71 		 * "If a checksum is not present when its use has been
72 		 * negotiated, the receiver MUST close the subflow with a RST as
73 		 * it is considered broken."
74 		 *
75 		 * We don't implement DSS checksum - fall back to TCP.
76 		 */
77 		if (flags & MPTCP_CAP_CHECKSUM_REQD)
78 			break;
79 
80 		mp_opt->mp_capable = 1;
81 		if (opsize >= TCPOLEN_MPTCP_MPC_SYNACK) {
82 			mp_opt->sndr_key = get_unaligned_be64(ptr);
83 			ptr += 8;
84 		}
85 		if (opsize >= TCPOLEN_MPTCP_MPC_ACK) {
86 			mp_opt->rcvr_key = get_unaligned_be64(ptr);
87 			ptr += 8;
88 		}
89 		if (opsize == TCPOLEN_MPTCP_MPC_ACK_DATA) {
90 			/* Section 3.1.:
91 			 * "the data parameters in a MP_CAPABLE are semantically
92 			 * equivalent to those in a DSS option and can be used
93 			 * interchangeably."
94 			 */
95 			mp_opt->dss = 1;
96 			mp_opt->use_map = 1;
97 			mp_opt->mpc_map = 1;
98 			mp_opt->data_len = get_unaligned_be16(ptr);
99 			ptr += 2;
100 		}
101 		pr_debug("MP_CAPABLE version=%x, flags=%x, optlen=%d sndr=%llu, rcvr=%llu len=%d",
102 			 version, flags, opsize, mp_opt->sndr_key,
103 			 mp_opt->rcvr_key, mp_opt->data_len);
104 		break;
105 
106 	case MPTCPOPT_MP_JOIN:
107 		mp_opt->mp_join = 1;
108 		if (opsize == TCPOLEN_MPTCP_MPJ_SYN) {
109 			mp_opt->backup = *ptr++ & MPTCPOPT_BACKUP;
110 			mp_opt->join_id = *ptr++;
111 			mp_opt->token = get_unaligned_be32(ptr);
112 			ptr += 4;
113 			mp_opt->nonce = get_unaligned_be32(ptr);
114 			ptr += 4;
115 			pr_debug("MP_JOIN bkup=%u, id=%u, token=%u, nonce=%u",
116 				 mp_opt->backup, mp_opt->join_id,
117 				 mp_opt->token, mp_opt->nonce);
118 		} else if (opsize == TCPOLEN_MPTCP_MPJ_SYNACK) {
119 			mp_opt->backup = *ptr++ & MPTCPOPT_BACKUP;
120 			mp_opt->join_id = *ptr++;
121 			mp_opt->thmac = get_unaligned_be64(ptr);
122 			ptr += 8;
123 			mp_opt->nonce = get_unaligned_be32(ptr);
124 			ptr += 4;
125 			pr_debug("MP_JOIN bkup=%u, id=%u, thmac=%llu, nonce=%u",
126 				 mp_opt->backup, mp_opt->join_id,
127 				 mp_opt->thmac, mp_opt->nonce);
128 		} else if (opsize == TCPOLEN_MPTCP_MPJ_ACK) {
129 			ptr += 2;
130 			memcpy(mp_opt->hmac, ptr, MPTCPOPT_HMAC_LEN);
131 			pr_debug("MP_JOIN hmac");
132 		} else {
133 			mp_opt->mp_join = 0;
134 		}
135 		break;
136 
137 	case MPTCPOPT_DSS:
138 		pr_debug("DSS");
139 		ptr++;
140 
141 		/* we must clear 'mpc_map' be able to detect MP_CAPABLE
142 		 * map vs DSS map in mptcp_incoming_options(), and reconstruct
143 		 * map info accordingly
144 		 */
145 		mp_opt->mpc_map = 0;
146 		flags = (*ptr++) & MPTCP_DSS_FLAG_MASK;
147 		mp_opt->data_fin = (flags & MPTCP_DSS_DATA_FIN) != 0;
148 		mp_opt->dsn64 = (flags & MPTCP_DSS_DSN64) != 0;
149 		mp_opt->use_map = (flags & MPTCP_DSS_HAS_MAP) != 0;
150 		mp_opt->ack64 = (flags & MPTCP_DSS_ACK64) != 0;
151 		mp_opt->use_ack = (flags & MPTCP_DSS_HAS_ACK);
152 
153 		pr_debug("data_fin=%d dsn64=%d use_map=%d ack64=%d use_ack=%d",
154 			 mp_opt->data_fin, mp_opt->dsn64,
155 			 mp_opt->use_map, mp_opt->ack64,
156 			 mp_opt->use_ack);
157 
158 		expected_opsize = TCPOLEN_MPTCP_DSS_BASE;
159 
160 		if (mp_opt->use_ack) {
161 			if (mp_opt->ack64)
162 				expected_opsize += TCPOLEN_MPTCP_DSS_ACK64;
163 			else
164 				expected_opsize += TCPOLEN_MPTCP_DSS_ACK32;
165 		}
166 
167 		if (mp_opt->use_map) {
168 			if (mp_opt->dsn64)
169 				expected_opsize += TCPOLEN_MPTCP_DSS_MAP64;
170 			else
171 				expected_opsize += TCPOLEN_MPTCP_DSS_MAP32;
172 		}
173 
174 		/* RFC 6824, Section 3.3:
175 		 * If a checksum is present, but its use had
176 		 * not been negotiated in the MP_CAPABLE handshake,
177 		 * the checksum field MUST be ignored.
178 		 */
179 		if (opsize != expected_opsize &&
180 		    opsize != expected_opsize + TCPOLEN_MPTCP_DSS_CHECKSUM)
181 			break;
182 
183 		mp_opt->dss = 1;
184 
185 		if (mp_opt->use_ack) {
186 			if (mp_opt->ack64) {
187 				mp_opt->data_ack = get_unaligned_be64(ptr);
188 				ptr += 8;
189 			} else {
190 				mp_opt->data_ack = get_unaligned_be32(ptr);
191 				ptr += 4;
192 			}
193 
194 			pr_debug("data_ack=%llu", mp_opt->data_ack);
195 		}
196 
197 		if (mp_opt->use_map) {
198 			if (mp_opt->dsn64) {
199 				mp_opt->data_seq = get_unaligned_be64(ptr);
200 				ptr += 8;
201 			} else {
202 				mp_opt->data_seq = get_unaligned_be32(ptr);
203 				ptr += 4;
204 			}
205 
206 			mp_opt->subflow_seq = get_unaligned_be32(ptr);
207 			ptr += 4;
208 
209 			mp_opt->data_len = get_unaligned_be16(ptr);
210 			ptr += 2;
211 
212 			pr_debug("data_seq=%llu subflow_seq=%u data_len=%u",
213 				 mp_opt->data_seq, mp_opt->subflow_seq,
214 				 mp_opt->data_len);
215 		}
216 
217 		break;
218 
219 	case MPTCPOPT_ADD_ADDR:
220 		mp_opt->echo = (*ptr++) & MPTCP_ADDR_ECHO;
221 		if (!mp_opt->echo) {
222 			if (opsize == TCPOLEN_MPTCP_ADD_ADDR ||
223 			    opsize == TCPOLEN_MPTCP_ADD_ADDR_PORT)
224 				mp_opt->addr.family = AF_INET;
225 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
226 			else if (opsize == TCPOLEN_MPTCP_ADD_ADDR6 ||
227 				 opsize == TCPOLEN_MPTCP_ADD_ADDR6_PORT)
228 				mp_opt->addr.family = AF_INET6;
229 #endif
230 			else
231 				break;
232 		} else {
233 			if (opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE ||
234 			    opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE_PORT)
235 				mp_opt->addr.family = AF_INET;
236 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
237 			else if (opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE ||
238 				 opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE_PORT)
239 				mp_opt->addr.family = AF_INET6;
240 #endif
241 			else
242 				break;
243 		}
244 
245 		mp_opt->add_addr = 1;
246 		mp_opt->addr.id = *ptr++;
247 		if (mp_opt->addr.family == AF_INET) {
248 			memcpy((u8 *)&mp_opt->addr.addr.s_addr, (u8 *)ptr, 4);
249 			ptr += 4;
250 			if (opsize == TCPOLEN_MPTCP_ADD_ADDR_PORT ||
251 			    opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE_PORT) {
252 				mp_opt->addr.port = htons(get_unaligned_be16(ptr));
253 				ptr += 2;
254 			}
255 		}
256 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
257 		else {
258 			memcpy(mp_opt->addr.addr6.s6_addr, (u8 *)ptr, 16);
259 			ptr += 16;
260 			if (opsize == TCPOLEN_MPTCP_ADD_ADDR6_PORT ||
261 			    opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE_PORT) {
262 				mp_opt->addr.port = htons(get_unaligned_be16(ptr));
263 				ptr += 2;
264 			}
265 		}
266 #endif
267 		if (!mp_opt->echo) {
268 			mp_opt->ahmac = get_unaligned_be64(ptr);
269 			ptr += 8;
270 		}
271 		pr_debug("ADD_ADDR%s: id=%d, ahmac=%llu, echo=%d, port=%d",
272 			 (mp_opt->addr.family == AF_INET6) ? "6" : "",
273 			 mp_opt->addr.id, mp_opt->ahmac, mp_opt->echo, ntohs(mp_opt->addr.port));
274 		break;
275 
276 	case MPTCPOPT_RM_ADDR:
277 		if (opsize < TCPOLEN_MPTCP_RM_ADDR_BASE + 1 ||
278 		    opsize > TCPOLEN_MPTCP_RM_ADDR_BASE + MPTCP_RM_IDS_MAX)
279 			break;
280 
281 		ptr++;
282 
283 		mp_opt->rm_addr = 1;
284 		mp_opt->rm_list.nr = opsize - TCPOLEN_MPTCP_RM_ADDR_BASE;
285 		for (i = 0; i < mp_opt->rm_list.nr; i++)
286 			mp_opt->rm_list.ids[i] = *ptr++;
287 		pr_debug("RM_ADDR: rm_list_nr=%d", mp_opt->rm_list.nr);
288 		break;
289 
290 	case MPTCPOPT_MP_PRIO:
291 		if (opsize != TCPOLEN_MPTCP_PRIO)
292 			break;
293 
294 		mp_opt->mp_prio = 1;
295 		mp_opt->backup = *ptr++ & MPTCP_PRIO_BKUP;
296 		pr_debug("MP_PRIO: prio=%d", mp_opt->backup);
297 		break;
298 
299 	case MPTCPOPT_MP_FASTCLOSE:
300 		if (opsize != TCPOLEN_MPTCP_FASTCLOSE)
301 			break;
302 
303 		ptr += 2;
304 		mp_opt->rcvr_key = get_unaligned_be64(ptr);
305 		ptr += 8;
306 		mp_opt->fastclose = 1;
307 		break;
308 
309 	case MPTCPOPT_RST:
310 		if (opsize != TCPOLEN_MPTCP_RST)
311 			break;
312 
313 		if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_RST))
314 			break;
315 		mp_opt->reset = 1;
316 		flags = *ptr++;
317 		mp_opt->reset_transient = flags & MPTCP_RST_TRANSIENT;
318 		mp_opt->reset_reason = *ptr;
319 		break;
320 
321 	default:
322 		break;
323 	}
324 }
325 
326 void mptcp_get_options(const struct sk_buff *skb,
327 		       struct mptcp_options_received *mp_opt)
328 {
329 	const struct tcphdr *th = tcp_hdr(skb);
330 	const unsigned char *ptr;
331 	int length;
332 
333 	/* initialize option status */
334 	mp_opt->mp_capable = 0;
335 	mp_opt->mp_join = 0;
336 	mp_opt->add_addr = 0;
337 	mp_opt->ahmac = 0;
338 	mp_opt->fastclose = 0;
339 	mp_opt->addr.port = 0;
340 	mp_opt->rm_addr = 0;
341 	mp_opt->dss = 0;
342 	mp_opt->mp_prio = 0;
343 	mp_opt->reset = 0;
344 
345 	length = (th->doff * 4) - sizeof(struct tcphdr);
346 	ptr = (const unsigned char *)(th + 1);
347 
348 	while (length > 0) {
349 		int opcode = *ptr++;
350 		int opsize;
351 
352 		switch (opcode) {
353 		case TCPOPT_EOL:
354 			return;
355 		case TCPOPT_NOP:	/* Ref: RFC 793 section 3.1 */
356 			length--;
357 			continue;
358 		default:
359 			opsize = *ptr++;
360 			if (opsize < 2) /* "silly options" */
361 				return;
362 			if (opsize > length)
363 				return;	/* don't parse partial options */
364 			if (opcode == TCPOPT_MPTCP)
365 				mptcp_parse_option(skb, ptr, opsize, mp_opt);
366 			ptr += opsize - 2;
367 			length -= opsize;
368 		}
369 	}
370 }
371 
372 bool mptcp_syn_options(struct sock *sk, const struct sk_buff *skb,
373 		       unsigned int *size, struct mptcp_out_options *opts)
374 {
375 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
376 
377 	/* we will use snd_isn to detect first pkt [re]transmission
378 	 * in mptcp_established_options_mp()
379 	 */
380 	subflow->snd_isn = TCP_SKB_CB(skb)->end_seq;
381 	if (subflow->request_mptcp) {
382 		opts->suboptions = OPTION_MPTCP_MPC_SYN;
383 		*size = TCPOLEN_MPTCP_MPC_SYN;
384 		return true;
385 	} else if (subflow->request_join) {
386 		pr_debug("remote_token=%u, nonce=%u", subflow->remote_token,
387 			 subflow->local_nonce);
388 		opts->suboptions = OPTION_MPTCP_MPJ_SYN;
389 		opts->join_id = subflow->local_id;
390 		opts->token = subflow->remote_token;
391 		opts->nonce = subflow->local_nonce;
392 		opts->backup = subflow->request_bkup;
393 		*size = TCPOLEN_MPTCP_MPJ_SYN;
394 		return true;
395 	}
396 	return false;
397 }
398 
399 /* MP_JOIN client subflow must wait for 4th ack before sending any data:
400  * TCP can't schedule delack timer before the subflow is fully established.
401  * MPTCP uses the delack timer to do 3rd ack retransmissions
402  */
403 static void schedule_3rdack_retransmission(struct sock *sk)
404 {
405 	struct inet_connection_sock *icsk = inet_csk(sk);
406 	struct tcp_sock *tp = tcp_sk(sk);
407 	unsigned long timeout;
408 
409 	/* reschedule with a timeout above RTT, as we must look only for drop */
410 	if (tp->srtt_us)
411 		timeout = tp->srtt_us << 1;
412 	else
413 		timeout = TCP_TIMEOUT_INIT;
414 
415 	WARN_ON_ONCE(icsk->icsk_ack.pending & ICSK_ACK_TIMER);
416 	icsk->icsk_ack.pending |= ICSK_ACK_SCHED | ICSK_ACK_TIMER;
417 	icsk->icsk_ack.timeout = timeout;
418 	sk_reset_timer(sk, &icsk->icsk_delack_timer, timeout);
419 }
420 
421 static void clear_3rdack_retransmission(struct sock *sk)
422 {
423 	struct inet_connection_sock *icsk = inet_csk(sk);
424 
425 	sk_stop_timer(sk, &icsk->icsk_delack_timer);
426 	icsk->icsk_ack.timeout = 0;
427 	icsk->icsk_ack.ato = 0;
428 	icsk->icsk_ack.pending &= ~(ICSK_ACK_SCHED | ICSK_ACK_TIMER);
429 }
430 
431 static bool mptcp_established_options_mp(struct sock *sk, struct sk_buff *skb,
432 					 bool snd_data_fin_enable,
433 					 unsigned int *size,
434 					 unsigned int remaining,
435 					 struct mptcp_out_options *opts)
436 {
437 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
438 	struct mptcp_ext *mpext;
439 	unsigned int data_len;
440 
441 	/* When skb is not available, we better over-estimate the emitted
442 	 * options len. A full DSS option (28 bytes) is longer than
443 	 * TCPOLEN_MPTCP_MPC_ACK_DATA(22) or TCPOLEN_MPTCP_MPJ_ACK(24), so
444 	 * tell the caller to defer the estimate to
445 	 * mptcp_established_options_dss(), which will reserve enough space.
446 	 */
447 	if (!skb)
448 		return false;
449 
450 	/* MPC/MPJ needed only on 3rd ack packet, DATA_FIN and TCP shutdown take precedence */
451 	if (subflow->fully_established || snd_data_fin_enable ||
452 	    subflow->snd_isn != TCP_SKB_CB(skb)->seq ||
453 	    sk->sk_state != TCP_ESTABLISHED)
454 		return false;
455 
456 	if (subflow->mp_capable) {
457 		mpext = mptcp_get_ext(skb);
458 		data_len = mpext ? mpext->data_len : 0;
459 
460 		/* we will check ext_copy.data_len in mptcp_write_options() to
461 		 * discriminate between TCPOLEN_MPTCP_MPC_ACK_DATA and
462 		 * TCPOLEN_MPTCP_MPC_ACK
463 		 */
464 		opts->ext_copy.data_len = data_len;
465 		opts->suboptions = OPTION_MPTCP_MPC_ACK;
466 		opts->sndr_key = subflow->local_key;
467 		opts->rcvr_key = subflow->remote_key;
468 
469 		/* Section 3.1.
470 		 * The MP_CAPABLE option is carried on the SYN, SYN/ACK, and ACK
471 		 * packets that start the first subflow of an MPTCP connection,
472 		 * as well as the first packet that carries data
473 		 */
474 		if (data_len > 0)
475 			*size = ALIGN(TCPOLEN_MPTCP_MPC_ACK_DATA, 4);
476 		else
477 			*size = TCPOLEN_MPTCP_MPC_ACK;
478 
479 		pr_debug("subflow=%p, local_key=%llu, remote_key=%llu map_len=%d",
480 			 subflow, subflow->local_key, subflow->remote_key,
481 			 data_len);
482 
483 		return true;
484 	} else if (subflow->mp_join) {
485 		opts->suboptions = OPTION_MPTCP_MPJ_ACK;
486 		memcpy(opts->hmac, subflow->hmac, MPTCPOPT_HMAC_LEN);
487 		*size = TCPOLEN_MPTCP_MPJ_ACK;
488 		pr_debug("subflow=%p", subflow);
489 
490 		schedule_3rdack_retransmission(sk);
491 		return true;
492 	}
493 	return false;
494 }
495 
496 static void mptcp_write_data_fin(struct mptcp_subflow_context *subflow,
497 				 struct sk_buff *skb, struct mptcp_ext *ext)
498 {
499 	/* The write_seq value has already been incremented, so the actual
500 	 * sequence number for the DATA_FIN is one less.
501 	 */
502 	u64 data_fin_tx_seq = READ_ONCE(mptcp_sk(subflow->conn)->write_seq) - 1;
503 
504 	if (!ext->use_map || !skb->len) {
505 		/* RFC6824 requires a DSS mapping with specific values
506 		 * if DATA_FIN is set but no data payload is mapped
507 		 */
508 		ext->data_fin = 1;
509 		ext->use_map = 1;
510 		ext->dsn64 = 1;
511 		ext->data_seq = data_fin_tx_seq;
512 		ext->subflow_seq = 0;
513 		ext->data_len = 1;
514 	} else if (ext->data_seq + ext->data_len == data_fin_tx_seq) {
515 		/* If there's an existing DSS mapping and it is the
516 		 * final mapping, DATA_FIN consumes 1 additional byte of
517 		 * mapping space.
518 		 */
519 		ext->data_fin = 1;
520 		ext->data_len++;
521 	}
522 }
523 
524 static bool mptcp_established_options_dss(struct sock *sk, struct sk_buff *skb,
525 					  bool snd_data_fin_enable,
526 					  unsigned int *size,
527 					  unsigned int remaining,
528 					  struct mptcp_out_options *opts)
529 {
530 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
531 	struct mptcp_sock *msk = mptcp_sk(subflow->conn);
532 	unsigned int dss_size = 0;
533 	struct mptcp_ext *mpext;
534 	unsigned int ack_size;
535 	bool ret = false;
536 	u64 ack_seq;
537 
538 	mpext = skb ? mptcp_get_ext(skb) : NULL;
539 
540 	if (!skb || (mpext && mpext->use_map) || snd_data_fin_enable) {
541 		unsigned int map_size;
542 
543 		map_size = TCPOLEN_MPTCP_DSS_BASE + TCPOLEN_MPTCP_DSS_MAP64;
544 
545 		remaining -= map_size;
546 		dss_size = map_size;
547 		if (mpext)
548 			opts->ext_copy = *mpext;
549 
550 		if (skb && snd_data_fin_enable)
551 			mptcp_write_data_fin(subflow, skb, &opts->ext_copy);
552 		ret = true;
553 	}
554 
555 	/* passive sockets msk will set the 'can_ack' after accept(), even
556 	 * if the first subflow may have the already the remote key handy
557 	 */
558 	opts->ext_copy.use_ack = 0;
559 	if (!READ_ONCE(msk->can_ack)) {
560 		*size = ALIGN(dss_size, 4);
561 		return ret;
562 	}
563 
564 	ack_seq = READ_ONCE(msk->ack_seq);
565 	if (READ_ONCE(msk->use_64bit_ack)) {
566 		ack_size = TCPOLEN_MPTCP_DSS_ACK64;
567 		opts->ext_copy.data_ack = ack_seq;
568 		opts->ext_copy.ack64 = 1;
569 	} else {
570 		ack_size = TCPOLEN_MPTCP_DSS_ACK32;
571 		opts->ext_copy.data_ack32 = (uint32_t)ack_seq;
572 		opts->ext_copy.ack64 = 0;
573 	}
574 	opts->ext_copy.use_ack = 1;
575 	WRITE_ONCE(msk->old_wspace, __mptcp_space((struct sock *)msk));
576 
577 	/* Add kind/length/subtype/flag overhead if mapping is not populated */
578 	if (dss_size == 0)
579 		ack_size += TCPOLEN_MPTCP_DSS_BASE;
580 
581 	dss_size += ack_size;
582 
583 	*size = ALIGN(dss_size, 4);
584 	return true;
585 }
586 
587 static u64 add_addr_generate_hmac(u64 key1, u64 key2,
588 				  struct mptcp_addr_info *addr)
589 {
590 	u16 port = ntohs(addr->port);
591 	u8 hmac[SHA256_DIGEST_SIZE];
592 	u8 msg[19];
593 	int i = 0;
594 
595 	msg[i++] = addr->id;
596 	if (addr->family == AF_INET) {
597 		memcpy(&msg[i], &addr->addr.s_addr, 4);
598 		i += 4;
599 	}
600 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
601 	else if (addr->family == AF_INET6) {
602 		memcpy(&msg[i], &addr->addr6.s6_addr, 16);
603 		i += 16;
604 	}
605 #endif
606 	msg[i++] = port >> 8;
607 	msg[i++] = port & 0xFF;
608 
609 	mptcp_crypto_hmac_sha(key1, key2, msg, i, hmac);
610 
611 	return get_unaligned_be64(&hmac[SHA256_DIGEST_SIZE - sizeof(u64)]);
612 }
613 
614 static bool mptcp_established_options_add_addr(struct sock *sk, struct sk_buff *skb,
615 					       unsigned int *size,
616 					       unsigned int remaining,
617 					       struct mptcp_out_options *opts)
618 {
619 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
620 	struct mptcp_sock *msk = mptcp_sk(subflow->conn);
621 	bool drop_other_suboptions = false;
622 	unsigned int opt_size = *size;
623 	bool echo;
624 	bool port;
625 	int len;
626 
627 	if ((mptcp_pm_should_add_signal_ipv6(msk) ||
628 	     mptcp_pm_should_add_signal_port(msk) ||
629 	     mptcp_pm_should_add_signal_echo(msk)) &&
630 	    skb && skb_is_tcp_pure_ack(skb)) {
631 		pr_debug("drop other suboptions");
632 		opts->suboptions = 0;
633 		opts->ext_copy.use_ack = 0;
634 		opts->ext_copy.use_map = 0;
635 		remaining += opt_size;
636 		drop_other_suboptions = true;
637 	}
638 
639 	if (!mptcp_pm_should_add_signal(msk) ||
640 	    !(mptcp_pm_add_addr_signal(msk, remaining, &opts->addr, &echo, &port)))
641 		return false;
642 
643 	len = mptcp_add_addr_len(opts->addr.family, echo, port);
644 	if (remaining < len)
645 		return false;
646 
647 	*size = len;
648 	if (drop_other_suboptions)
649 		*size -= opt_size;
650 	opts->suboptions |= OPTION_MPTCP_ADD_ADDR;
651 	if (!echo) {
652 		opts->ahmac = add_addr_generate_hmac(msk->local_key,
653 						     msk->remote_key,
654 						     &opts->addr);
655 	}
656 	pr_debug("addr_id=%d, ahmac=%llu, echo=%d, port=%d",
657 		 opts->addr.id, opts->ahmac, echo, ntohs(opts->addr.port));
658 
659 	return true;
660 }
661 
662 static bool mptcp_established_options_rm_addr(struct sock *sk,
663 					      unsigned int *size,
664 					      unsigned int remaining,
665 					      struct mptcp_out_options *opts)
666 {
667 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
668 	struct mptcp_sock *msk = mptcp_sk(subflow->conn);
669 	struct mptcp_rm_list rm_list;
670 	int i, len;
671 
672 	if (!mptcp_pm_should_rm_signal(msk) ||
673 	    !(mptcp_pm_rm_addr_signal(msk, remaining, &rm_list)))
674 		return false;
675 
676 	len = mptcp_rm_addr_len(&rm_list);
677 	if (len < 0)
678 		return false;
679 	if (remaining < len)
680 		return false;
681 
682 	*size = len;
683 	opts->suboptions |= OPTION_MPTCP_RM_ADDR;
684 	opts->rm_list = rm_list;
685 
686 	for (i = 0; i < opts->rm_list.nr; i++)
687 		pr_debug("rm_list_ids[%d]=%d", i, opts->rm_list.ids[i]);
688 
689 	return true;
690 }
691 
692 static bool mptcp_established_options_mp_prio(struct sock *sk,
693 					      unsigned int *size,
694 					      unsigned int remaining,
695 					      struct mptcp_out_options *opts)
696 {
697 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
698 
699 	if (!subflow->send_mp_prio)
700 		return false;
701 
702 	/* account for the trailing 'nop' option */
703 	if (remaining < TCPOLEN_MPTCP_PRIO_ALIGN)
704 		return false;
705 
706 	*size = TCPOLEN_MPTCP_PRIO_ALIGN;
707 	opts->suboptions |= OPTION_MPTCP_PRIO;
708 	opts->backup = subflow->request_bkup;
709 
710 	pr_debug("prio=%d", opts->backup);
711 
712 	return true;
713 }
714 
715 static noinline void mptcp_established_options_rst(struct sock *sk, struct sk_buff *skb,
716 						   unsigned int *size,
717 						   unsigned int remaining,
718 						   struct mptcp_out_options *opts)
719 {
720 	const struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
721 
722 	if (remaining < TCPOLEN_MPTCP_RST)
723 		return;
724 
725 	*size = TCPOLEN_MPTCP_RST;
726 	opts->suboptions |= OPTION_MPTCP_RST;
727 	opts->reset_transient = subflow->reset_transient;
728 	opts->reset_reason = subflow->reset_reason;
729 }
730 
731 bool mptcp_established_options(struct sock *sk, struct sk_buff *skb,
732 			       unsigned int *size, unsigned int remaining,
733 			       struct mptcp_out_options *opts)
734 {
735 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
736 	struct mptcp_sock *msk = mptcp_sk(subflow->conn);
737 	unsigned int opt_size = 0;
738 	bool snd_data_fin;
739 	bool ret = false;
740 
741 	opts->suboptions = 0;
742 
743 	if (unlikely(__mptcp_check_fallback(msk)))
744 		return false;
745 
746 	if (unlikely(skb && TCP_SKB_CB(skb)->tcp_flags & TCPHDR_RST)) {
747 		mptcp_established_options_rst(sk, skb, size, remaining, opts);
748 		return true;
749 	}
750 
751 	snd_data_fin = mptcp_data_fin_enabled(msk);
752 	if (mptcp_established_options_mp(sk, skb, snd_data_fin, &opt_size, remaining, opts))
753 		ret = true;
754 	else if (mptcp_established_options_dss(sk, skb, snd_data_fin, &opt_size, remaining, opts))
755 		ret = true;
756 
757 	/* we reserved enough space for the above options, and exceeding the
758 	 * TCP option space would be fatal
759 	 */
760 	if (WARN_ON_ONCE(opt_size > remaining))
761 		return false;
762 
763 	*size += opt_size;
764 	remaining -= opt_size;
765 	if (mptcp_established_options_add_addr(sk, skb, &opt_size, remaining, opts)) {
766 		*size += opt_size;
767 		remaining -= opt_size;
768 		ret = true;
769 	} else if (mptcp_established_options_rm_addr(sk, &opt_size, remaining, opts)) {
770 		*size += opt_size;
771 		remaining -= opt_size;
772 		ret = true;
773 	}
774 
775 	if (mptcp_established_options_mp_prio(sk, &opt_size, remaining, opts)) {
776 		*size += opt_size;
777 		remaining -= opt_size;
778 		ret = true;
779 	}
780 
781 	return ret;
782 }
783 
784 bool mptcp_synack_options(const struct request_sock *req, unsigned int *size,
785 			  struct mptcp_out_options *opts)
786 {
787 	struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
788 
789 	if (subflow_req->mp_capable) {
790 		opts->suboptions = OPTION_MPTCP_MPC_SYNACK;
791 		opts->sndr_key = subflow_req->local_key;
792 		*size = TCPOLEN_MPTCP_MPC_SYNACK;
793 		pr_debug("subflow_req=%p, local_key=%llu",
794 			 subflow_req, subflow_req->local_key);
795 		return true;
796 	} else if (subflow_req->mp_join) {
797 		opts->suboptions = OPTION_MPTCP_MPJ_SYNACK;
798 		opts->backup = subflow_req->backup;
799 		opts->join_id = subflow_req->local_id;
800 		opts->thmac = subflow_req->thmac;
801 		opts->nonce = subflow_req->local_nonce;
802 		pr_debug("req=%p, bkup=%u, id=%u, thmac=%llu, nonce=%u",
803 			 subflow_req, opts->backup, opts->join_id,
804 			 opts->thmac, opts->nonce);
805 		*size = TCPOLEN_MPTCP_MPJ_SYNACK;
806 		return true;
807 	}
808 	return false;
809 }
810 
811 static bool check_fully_established(struct mptcp_sock *msk, struct sock *ssk,
812 				    struct mptcp_subflow_context *subflow,
813 				    struct sk_buff *skb,
814 				    struct mptcp_options_received *mp_opt)
815 {
816 	/* here we can process OoO, in-window pkts, only in-sequence 4th ack
817 	 * will make the subflow fully established
818 	 */
819 	if (likely(subflow->fully_established)) {
820 		/* on passive sockets, check for 3rd ack retransmission
821 		 * note that msk is always set by subflow_syn_recv_sock()
822 		 * for mp_join subflows
823 		 */
824 		if (TCP_SKB_CB(skb)->seq == subflow->ssn_offset + 1 &&
825 		    TCP_SKB_CB(skb)->end_seq == TCP_SKB_CB(skb)->seq &&
826 		    subflow->mp_join && mp_opt->mp_join &&
827 		    READ_ONCE(msk->pm.server_side))
828 			tcp_send_ack(ssk);
829 		goto fully_established;
830 	}
831 
832 	/* we must process OoO packets before the first subflow is fully
833 	 * established. OoO packets are instead a protocol violation
834 	 * for MP_JOIN subflows as the peer must not send any data
835 	 * before receiving the forth ack - cfr. RFC 8684 section 3.2.
836 	 */
837 	if (TCP_SKB_CB(skb)->seq != subflow->ssn_offset + 1) {
838 		if (subflow->mp_join)
839 			goto reset;
840 		return subflow->mp_capable;
841 	}
842 
843 	if (mp_opt->dss && mp_opt->use_ack) {
844 		/* subflows are fully established as soon as we get any
845 		 * additional ack.
846 		 */
847 		subflow->fully_established = 1;
848 		WRITE_ONCE(msk->fully_established, true);
849 		goto fully_established;
850 	}
851 
852 	if (mp_opt->add_addr) {
853 		WRITE_ONCE(msk->fully_established, true);
854 		return true;
855 	}
856 
857 	/* If the first established packet does not contain MP_CAPABLE + data
858 	 * then fallback to TCP. Fallback scenarios requires a reset for
859 	 * MP_JOIN subflows.
860 	 */
861 	if (!mp_opt->mp_capable) {
862 		if (subflow->mp_join)
863 			goto reset;
864 		subflow->mp_capable = 0;
865 		pr_fallback(msk);
866 		__mptcp_do_fallback(msk);
867 		return false;
868 	}
869 
870 	if (unlikely(!READ_ONCE(msk->pm.server_side)))
871 		pr_warn_once("bogus mpc option on established client sk");
872 	mptcp_subflow_fully_established(subflow, mp_opt);
873 
874 fully_established:
875 	/* if the subflow is not already linked into the conn_list, we can't
876 	 * notify the PM: this subflow is still on the listener queue
877 	 * and the PM possibly acquiring the subflow lock could race with
878 	 * the listener close
879 	 */
880 	if (likely(subflow->pm_notified) || list_empty(&subflow->node))
881 		return true;
882 
883 	subflow->pm_notified = 1;
884 	if (subflow->mp_join) {
885 		clear_3rdack_retransmission(ssk);
886 		mptcp_pm_subflow_established(msk);
887 	} else {
888 		mptcp_pm_fully_established(msk, ssk, GFP_ATOMIC);
889 	}
890 	return true;
891 
892 reset:
893 	mptcp_subflow_reset(ssk);
894 	return false;
895 }
896 
897 static u64 expand_ack(u64 old_ack, u64 cur_ack, bool use_64bit)
898 {
899 	u32 old_ack32, cur_ack32;
900 
901 	if (use_64bit)
902 		return cur_ack;
903 
904 	old_ack32 = (u32)old_ack;
905 	cur_ack32 = (u32)cur_ack;
906 	cur_ack = (old_ack & GENMASK_ULL(63, 32)) + cur_ack32;
907 	if (unlikely(before(cur_ack32, old_ack32)))
908 		return cur_ack + (1LL << 32);
909 	return cur_ack;
910 }
911 
912 static void ack_update_msk(struct mptcp_sock *msk,
913 			   struct sock *ssk,
914 			   struct mptcp_options_received *mp_opt)
915 {
916 	u64 new_wnd_end, new_snd_una, snd_nxt = READ_ONCE(msk->snd_nxt);
917 	struct sock *sk = (struct sock *)msk;
918 	u64 old_snd_una;
919 
920 	mptcp_data_lock(sk);
921 
922 	/* avoid ack expansion on update conflict, to reduce the risk of
923 	 * wrongly expanding to a future ack sequence number, which is way
924 	 * more dangerous than missing an ack
925 	 */
926 	old_snd_una = msk->snd_una;
927 	new_snd_una = expand_ack(old_snd_una, mp_opt->data_ack, mp_opt->ack64);
928 
929 	/* ACK for data not even sent yet? Ignore. */
930 	if (after64(new_snd_una, snd_nxt))
931 		new_snd_una = old_snd_una;
932 
933 	new_wnd_end = new_snd_una + tcp_sk(ssk)->snd_wnd;
934 
935 	if (after64(new_wnd_end, msk->wnd_end))
936 		msk->wnd_end = new_wnd_end;
937 
938 	/* this assumes mptcp_incoming_options() is invoked after tcp_ack() */
939 	if (after64(msk->wnd_end, READ_ONCE(msk->snd_nxt)))
940 		__mptcp_check_push(sk, ssk);
941 
942 	if (after64(new_snd_una, old_snd_una)) {
943 		msk->snd_una = new_snd_una;
944 		__mptcp_data_acked(sk);
945 	}
946 	mptcp_data_unlock(sk);
947 
948 	trace_ack_update_msk(mp_opt->data_ack,
949 			     old_snd_una, new_snd_una,
950 			     new_wnd_end, msk->wnd_end);
951 }
952 
953 bool mptcp_update_rcv_data_fin(struct mptcp_sock *msk, u64 data_fin_seq, bool use_64bit)
954 {
955 	/* Skip if DATA_FIN was already received.
956 	 * If updating simultaneously with the recvmsg loop, values
957 	 * should match. If they mismatch, the peer is misbehaving and
958 	 * we will prefer the most recent information.
959 	 */
960 	if (READ_ONCE(msk->rcv_data_fin))
961 		return false;
962 
963 	WRITE_ONCE(msk->rcv_data_fin_seq,
964 		   expand_ack(READ_ONCE(msk->ack_seq), data_fin_seq, use_64bit));
965 	WRITE_ONCE(msk->rcv_data_fin, 1);
966 
967 	return true;
968 }
969 
970 static bool add_addr_hmac_valid(struct mptcp_sock *msk,
971 				struct mptcp_options_received *mp_opt)
972 {
973 	u64 hmac = 0;
974 
975 	if (mp_opt->echo)
976 		return true;
977 
978 	hmac = add_addr_generate_hmac(msk->remote_key,
979 				      msk->local_key,
980 				      &mp_opt->addr);
981 
982 	pr_debug("msk=%p, ahmac=%llu, mp_opt->ahmac=%llu\n",
983 		 msk, (unsigned long long)hmac,
984 		 (unsigned long long)mp_opt->ahmac);
985 
986 	return hmac == mp_opt->ahmac;
987 }
988 
989 void mptcp_incoming_options(struct sock *sk, struct sk_buff *skb)
990 {
991 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
992 	struct mptcp_sock *msk = mptcp_sk(subflow->conn);
993 	struct mptcp_options_received mp_opt;
994 	struct mptcp_ext *mpext;
995 
996 	if (__mptcp_check_fallback(msk)) {
997 		/* Keep it simple and unconditionally trigger send data cleanup and
998 		 * pending queue spooling. We will need to acquire the data lock
999 		 * for more accurate checks, and once the lock is acquired, such
1000 		 * helpers are cheap.
1001 		 */
1002 		mptcp_data_lock(subflow->conn);
1003 		if (sk_stream_memory_free(sk))
1004 			__mptcp_check_push(subflow->conn, sk);
1005 		__mptcp_data_acked(subflow->conn);
1006 		mptcp_data_unlock(subflow->conn);
1007 		return;
1008 	}
1009 
1010 	mptcp_get_options(skb, &mp_opt);
1011 	if (!check_fully_established(msk, sk, subflow, skb, &mp_opt))
1012 		return;
1013 
1014 	if (mp_opt.fastclose &&
1015 	    msk->local_key == mp_opt.rcvr_key) {
1016 		WRITE_ONCE(msk->rcv_fastclose, true);
1017 		mptcp_schedule_work((struct sock *)msk);
1018 	}
1019 
1020 	if (mp_opt.add_addr && add_addr_hmac_valid(msk, &mp_opt)) {
1021 		if (!mp_opt.echo) {
1022 			mptcp_pm_add_addr_received(msk, &mp_opt.addr);
1023 			MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ADDADDR);
1024 		} else {
1025 			mptcp_pm_add_addr_echoed(msk, &mp_opt.addr);
1026 			mptcp_pm_del_add_timer(msk, &mp_opt.addr, true);
1027 			MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ECHOADD);
1028 		}
1029 
1030 		if (mp_opt.addr.port)
1031 			MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_PORTADD);
1032 
1033 		mp_opt.add_addr = 0;
1034 	}
1035 
1036 	if (mp_opt.rm_addr) {
1037 		mptcp_pm_rm_addr_received(msk, &mp_opt.rm_list);
1038 		mp_opt.rm_addr = 0;
1039 	}
1040 
1041 	if (mp_opt.mp_prio) {
1042 		mptcp_pm_mp_prio_received(sk, mp_opt.backup);
1043 		MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPPRIORX);
1044 		mp_opt.mp_prio = 0;
1045 	}
1046 
1047 	if (mp_opt.reset) {
1048 		subflow->reset_seen = 1;
1049 		subflow->reset_reason = mp_opt.reset_reason;
1050 		subflow->reset_transient = mp_opt.reset_transient;
1051 	}
1052 
1053 	if (!mp_opt.dss)
1054 		return;
1055 
1056 	/* we can't wait for recvmsg() to update the ack_seq, otherwise
1057 	 * monodirectional flows will stuck
1058 	 */
1059 	if (mp_opt.use_ack)
1060 		ack_update_msk(msk, sk, &mp_opt);
1061 
1062 	/* Zero-data-length packets are dropped by the caller and not
1063 	 * propagated to the MPTCP layer, so the skb extension does not
1064 	 * need to be allocated or populated. DATA_FIN information, if
1065 	 * present, needs to be updated here before the skb is freed.
1066 	 */
1067 	if (TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb)->end_seq) {
1068 		if (mp_opt.data_fin && mp_opt.data_len == 1 &&
1069 		    mptcp_update_rcv_data_fin(msk, mp_opt.data_seq, mp_opt.dsn64) &&
1070 		    schedule_work(&msk->work))
1071 			sock_hold(subflow->conn);
1072 
1073 		return;
1074 	}
1075 
1076 	mpext = skb_ext_add(skb, SKB_EXT_MPTCP);
1077 	if (!mpext)
1078 		return;
1079 
1080 	memset(mpext, 0, sizeof(*mpext));
1081 
1082 	if (mp_opt.use_map) {
1083 		if (mp_opt.mpc_map) {
1084 			/* this is an MP_CAPABLE carrying MPTCP data
1085 			 * we know this map the first chunk of data
1086 			 */
1087 			mptcp_crypto_key_sha(subflow->remote_key, NULL,
1088 					     &mpext->data_seq);
1089 			mpext->data_seq++;
1090 			mpext->subflow_seq = 1;
1091 			mpext->dsn64 = 1;
1092 			mpext->mpc_map = 1;
1093 			mpext->data_fin = 0;
1094 		} else {
1095 			mpext->data_seq = mp_opt.data_seq;
1096 			mpext->subflow_seq = mp_opt.subflow_seq;
1097 			mpext->dsn64 = mp_opt.dsn64;
1098 			mpext->data_fin = mp_opt.data_fin;
1099 		}
1100 		mpext->data_len = mp_opt.data_len;
1101 		mpext->use_map = 1;
1102 	}
1103 }
1104 
1105 static void mptcp_set_rwin(const struct tcp_sock *tp)
1106 {
1107 	const struct sock *ssk = (const struct sock *)tp;
1108 	const struct mptcp_subflow_context *subflow;
1109 	struct mptcp_sock *msk;
1110 	u64 ack_seq;
1111 
1112 	subflow = mptcp_subflow_ctx(ssk);
1113 	msk = mptcp_sk(subflow->conn);
1114 
1115 	ack_seq = READ_ONCE(msk->ack_seq) + tp->rcv_wnd;
1116 
1117 	if (after64(ack_seq, READ_ONCE(msk->rcv_wnd_sent)))
1118 		WRITE_ONCE(msk->rcv_wnd_sent, ack_seq);
1119 }
1120 
1121 void mptcp_write_options(__be32 *ptr, const struct tcp_sock *tp,
1122 			 struct mptcp_out_options *opts)
1123 {
1124 	if ((OPTION_MPTCP_MPC_SYN | OPTION_MPTCP_MPC_SYNACK |
1125 	     OPTION_MPTCP_MPC_ACK) & opts->suboptions) {
1126 		u8 len;
1127 
1128 		if (OPTION_MPTCP_MPC_SYN & opts->suboptions)
1129 			len = TCPOLEN_MPTCP_MPC_SYN;
1130 		else if (OPTION_MPTCP_MPC_SYNACK & opts->suboptions)
1131 			len = TCPOLEN_MPTCP_MPC_SYNACK;
1132 		else if (opts->ext_copy.data_len)
1133 			len = TCPOLEN_MPTCP_MPC_ACK_DATA;
1134 		else
1135 			len = TCPOLEN_MPTCP_MPC_ACK;
1136 
1137 		*ptr++ = mptcp_option(MPTCPOPT_MP_CAPABLE, len,
1138 				      MPTCP_SUPPORTED_VERSION,
1139 				      MPTCP_CAP_HMAC_SHA256);
1140 
1141 		if (!((OPTION_MPTCP_MPC_SYNACK | OPTION_MPTCP_MPC_ACK) &
1142 		    opts->suboptions))
1143 			goto mp_capable_done;
1144 
1145 		put_unaligned_be64(opts->sndr_key, ptr);
1146 		ptr += 2;
1147 		if (!((OPTION_MPTCP_MPC_ACK) & opts->suboptions))
1148 			goto mp_capable_done;
1149 
1150 		put_unaligned_be64(opts->rcvr_key, ptr);
1151 		ptr += 2;
1152 		if (!opts->ext_copy.data_len)
1153 			goto mp_capable_done;
1154 
1155 		put_unaligned_be32(opts->ext_copy.data_len << 16 |
1156 				   TCPOPT_NOP << 8 | TCPOPT_NOP, ptr);
1157 		ptr += 1;
1158 	}
1159 
1160 mp_capable_done:
1161 	if (OPTION_MPTCP_ADD_ADDR & opts->suboptions) {
1162 		u8 len = TCPOLEN_MPTCP_ADD_ADDR_BASE;
1163 		u8 echo = MPTCP_ADDR_ECHO;
1164 
1165 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1166 		if (opts->addr.family == AF_INET6)
1167 			len = TCPOLEN_MPTCP_ADD_ADDR6_BASE;
1168 #endif
1169 
1170 		if (opts->addr.port)
1171 			len += TCPOLEN_MPTCP_PORT_LEN;
1172 
1173 		if (opts->ahmac) {
1174 			len += sizeof(opts->ahmac);
1175 			echo = 0;
1176 		}
1177 
1178 		*ptr++ = mptcp_option(MPTCPOPT_ADD_ADDR,
1179 				      len, echo, opts->addr.id);
1180 		if (opts->addr.family == AF_INET) {
1181 			memcpy((u8 *)ptr, (u8 *)&opts->addr.addr.s_addr, 4);
1182 			ptr += 1;
1183 		}
1184 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1185 		else if (opts->addr.family == AF_INET6) {
1186 			memcpy((u8 *)ptr, opts->addr.addr6.s6_addr, 16);
1187 			ptr += 4;
1188 		}
1189 #endif
1190 
1191 		if (!opts->addr.port) {
1192 			if (opts->ahmac) {
1193 				put_unaligned_be64(opts->ahmac, ptr);
1194 				ptr += 2;
1195 			}
1196 		} else {
1197 			u16 port = ntohs(opts->addr.port);
1198 
1199 			if (opts->ahmac) {
1200 				u8 *bptr = (u8 *)ptr;
1201 
1202 				put_unaligned_be16(port, bptr);
1203 				bptr += 2;
1204 				put_unaligned_be64(opts->ahmac, bptr);
1205 				bptr += 8;
1206 				put_unaligned_be16(TCPOPT_NOP << 8 |
1207 						   TCPOPT_NOP, bptr);
1208 
1209 				ptr += 3;
1210 			} else {
1211 				put_unaligned_be32(port << 16 |
1212 						   TCPOPT_NOP << 8 |
1213 						   TCPOPT_NOP, ptr);
1214 				ptr += 1;
1215 			}
1216 		}
1217 	}
1218 
1219 	if (OPTION_MPTCP_RM_ADDR & opts->suboptions) {
1220 		u8 i = 1;
1221 
1222 		*ptr++ = mptcp_option(MPTCPOPT_RM_ADDR,
1223 				      TCPOLEN_MPTCP_RM_ADDR_BASE + opts->rm_list.nr,
1224 				      0, opts->rm_list.ids[0]);
1225 
1226 		while (i < opts->rm_list.nr) {
1227 			u8 id1, id2, id3, id4;
1228 
1229 			id1 = opts->rm_list.ids[i];
1230 			id2 = i + 1 < opts->rm_list.nr ? opts->rm_list.ids[i + 1] : TCPOPT_NOP;
1231 			id3 = i + 2 < opts->rm_list.nr ? opts->rm_list.ids[i + 2] : TCPOPT_NOP;
1232 			id4 = i + 3 < opts->rm_list.nr ? opts->rm_list.ids[i + 3] : TCPOPT_NOP;
1233 			put_unaligned_be32(id1 << 24 | id2 << 16 | id3 << 8 | id4, ptr);
1234 			ptr += 1;
1235 			i += 4;
1236 		}
1237 	}
1238 
1239 	if (OPTION_MPTCP_PRIO & opts->suboptions) {
1240 		const struct sock *ssk = (const struct sock *)tp;
1241 		struct mptcp_subflow_context *subflow;
1242 
1243 		subflow = mptcp_subflow_ctx(ssk);
1244 		subflow->send_mp_prio = 0;
1245 
1246 		*ptr++ = mptcp_option(MPTCPOPT_MP_PRIO,
1247 				      TCPOLEN_MPTCP_PRIO,
1248 				      opts->backup, TCPOPT_NOP);
1249 	}
1250 
1251 	if (OPTION_MPTCP_MPJ_SYN & opts->suboptions) {
1252 		*ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1253 				      TCPOLEN_MPTCP_MPJ_SYN,
1254 				      opts->backup, opts->join_id);
1255 		put_unaligned_be32(opts->token, ptr);
1256 		ptr += 1;
1257 		put_unaligned_be32(opts->nonce, ptr);
1258 		ptr += 1;
1259 	}
1260 
1261 	if (OPTION_MPTCP_MPJ_SYNACK & opts->suboptions) {
1262 		*ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1263 				      TCPOLEN_MPTCP_MPJ_SYNACK,
1264 				      opts->backup, opts->join_id);
1265 		put_unaligned_be64(opts->thmac, ptr);
1266 		ptr += 2;
1267 		put_unaligned_be32(opts->nonce, ptr);
1268 		ptr += 1;
1269 	}
1270 
1271 	if (OPTION_MPTCP_MPJ_ACK & opts->suboptions) {
1272 		*ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1273 				      TCPOLEN_MPTCP_MPJ_ACK, 0, 0);
1274 		memcpy(ptr, opts->hmac, MPTCPOPT_HMAC_LEN);
1275 		ptr += 5;
1276 	}
1277 
1278 	if (OPTION_MPTCP_RST & opts->suboptions)
1279 		*ptr++ = mptcp_option(MPTCPOPT_RST,
1280 				      TCPOLEN_MPTCP_RST,
1281 				      opts->reset_transient,
1282 				      opts->reset_reason);
1283 
1284 	if (opts->ext_copy.use_ack || opts->ext_copy.use_map) {
1285 		struct mptcp_ext *mpext = &opts->ext_copy;
1286 		u8 len = TCPOLEN_MPTCP_DSS_BASE;
1287 		u8 flags = 0;
1288 
1289 		if (mpext->use_ack) {
1290 			flags = MPTCP_DSS_HAS_ACK;
1291 			if (mpext->ack64) {
1292 				len += TCPOLEN_MPTCP_DSS_ACK64;
1293 				flags |= MPTCP_DSS_ACK64;
1294 			} else {
1295 				len += TCPOLEN_MPTCP_DSS_ACK32;
1296 			}
1297 		}
1298 
1299 		if (mpext->use_map) {
1300 			len += TCPOLEN_MPTCP_DSS_MAP64;
1301 
1302 			/* Use only 64-bit mapping flags for now, add
1303 			 * support for optional 32-bit mappings later.
1304 			 */
1305 			flags |= MPTCP_DSS_HAS_MAP | MPTCP_DSS_DSN64;
1306 			if (mpext->data_fin)
1307 				flags |= MPTCP_DSS_DATA_FIN;
1308 		}
1309 
1310 		*ptr++ = mptcp_option(MPTCPOPT_DSS, len, 0, flags);
1311 
1312 		if (mpext->use_ack) {
1313 			if (mpext->ack64) {
1314 				put_unaligned_be64(mpext->data_ack, ptr);
1315 				ptr += 2;
1316 			} else {
1317 				put_unaligned_be32(mpext->data_ack32, ptr);
1318 				ptr += 1;
1319 			}
1320 		}
1321 
1322 		if (mpext->use_map) {
1323 			put_unaligned_be64(mpext->data_seq, ptr);
1324 			ptr += 2;
1325 			put_unaligned_be32(mpext->subflow_seq, ptr);
1326 			ptr += 1;
1327 			put_unaligned_be32(mpext->data_len << 16 |
1328 					   TCPOPT_NOP << 8 | TCPOPT_NOP, ptr);
1329 		}
1330 	}
1331 
1332 	if (tp)
1333 		mptcp_set_rwin(tp);
1334 }
1335 
1336 __be32 mptcp_get_reset_option(const struct sk_buff *skb)
1337 {
1338 	const struct mptcp_ext *ext = mptcp_get_ext(skb);
1339 	u8 flags, reason;
1340 
1341 	if (ext) {
1342 		flags = ext->reset_transient;
1343 		reason = ext->reset_reason;
1344 
1345 		return mptcp_option(MPTCPOPT_RST, TCPOLEN_MPTCP_RST,
1346 				    flags, reason);
1347 	}
1348 
1349 	return htonl(0u);
1350 }
1351 EXPORT_SYMBOL_GPL(mptcp_get_reset_option);
1352