xref: /linux/net/mptcp/options.c (revision d27656d02d85078c63f060fca9c5d99794791a75)
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 
48 		/* Cfr RFC 8684 Section 3.3.0:
49 		 * If a checksum is present but its use had
50 		 * not been negotiated in the MP_CAPABLE handshake, the receiver MUST
51 		 * close the subflow with a RST, as it is not behaving as negotiated.
52 		 * If a checksum is not present when its use has been negotiated, the
53 		 * receiver MUST close the subflow with a RST, as it is considered
54 		 * broken
55 		 * We parse even option with mismatching csum presence, so that
56 		 * later in subflow_data_ready we can trigger the reset.
57 		 */
58 		if (opsize != expected_opsize &&
59 		    (expected_opsize != TCPOLEN_MPTCP_MPC_ACK_DATA ||
60 		     opsize != TCPOLEN_MPTCP_MPC_ACK_DATA_CSUM))
61 			break;
62 
63 		/* try to be gentle vs future versions on the initial syn */
64 		version = *ptr++ & MPTCP_VERSION_MASK;
65 		if (opsize != TCPOLEN_MPTCP_MPC_SYN) {
66 			if (version != MPTCP_SUPPORTED_VERSION)
67 				break;
68 		} else if (version < MPTCP_SUPPORTED_VERSION) {
69 			break;
70 		}
71 
72 		flags = *ptr++;
73 		if (!mptcp_cap_flag_sha256(flags) ||
74 		    (flags & MPTCP_CAP_EXTENSIBILITY))
75 			break;
76 
77 		/* RFC 6824, Section 3.1:
78 		 * "For the Checksum Required bit (labeled "A"), if either
79 		 * host requires the use of checksums, checksums MUST be used.
80 		 * In other words, the only way for checksums not to be used
81 		 * is if both hosts in their SYNs set A=0."
82 		 */
83 		if (flags & MPTCP_CAP_CHECKSUM_REQD)
84 			mp_opt->suboptions |= OPTION_MPTCP_CSUMREQD;
85 
86 		mp_opt->deny_join_id0 = !!(flags & MPTCP_CAP_DENY_JOIN_ID0);
87 
88 		mp_opt->suboptions |= OPTIONS_MPTCP_MPC;
89 		if (opsize >= TCPOLEN_MPTCP_MPC_SYNACK) {
90 			mp_opt->sndr_key = get_unaligned_be64(ptr);
91 			ptr += 8;
92 		}
93 		if (opsize >= TCPOLEN_MPTCP_MPC_ACK) {
94 			mp_opt->rcvr_key = get_unaligned_be64(ptr);
95 			ptr += 8;
96 		}
97 		if (opsize >= TCPOLEN_MPTCP_MPC_ACK_DATA) {
98 			/* Section 3.1.:
99 			 * "the data parameters in a MP_CAPABLE are semantically
100 			 * equivalent to those in a DSS option and can be used
101 			 * interchangeably."
102 			 */
103 			mp_opt->suboptions |= OPTION_MPTCP_DSS;
104 			mp_opt->use_map = 1;
105 			mp_opt->mpc_map = 1;
106 			mp_opt->data_len = get_unaligned_be16(ptr);
107 			ptr += 2;
108 		}
109 		if (opsize == TCPOLEN_MPTCP_MPC_ACK_DATA_CSUM) {
110 			mp_opt->csum = (__force __sum16)get_unaligned_be16(ptr);
111 			mp_opt->suboptions |= OPTION_MPTCP_CSUMREQD;
112 			ptr += 2;
113 		}
114 		pr_debug("MP_CAPABLE version=%x, flags=%x, optlen=%d sndr=%llu, rcvr=%llu len=%d csum=%u",
115 			 version, flags, opsize, mp_opt->sndr_key,
116 			 mp_opt->rcvr_key, mp_opt->data_len, mp_opt->csum);
117 		break;
118 
119 	case MPTCPOPT_MP_JOIN:
120 		mp_opt->suboptions |= OPTIONS_MPTCP_MPJ;
121 		if (opsize == TCPOLEN_MPTCP_MPJ_SYN) {
122 			mp_opt->backup = *ptr++ & MPTCPOPT_BACKUP;
123 			mp_opt->join_id = *ptr++;
124 			mp_opt->token = get_unaligned_be32(ptr);
125 			ptr += 4;
126 			mp_opt->nonce = get_unaligned_be32(ptr);
127 			ptr += 4;
128 			pr_debug("MP_JOIN bkup=%u, id=%u, token=%u, nonce=%u",
129 				 mp_opt->backup, mp_opt->join_id,
130 				 mp_opt->token, mp_opt->nonce);
131 		} else if (opsize == TCPOLEN_MPTCP_MPJ_SYNACK) {
132 			mp_opt->backup = *ptr++ & MPTCPOPT_BACKUP;
133 			mp_opt->join_id = *ptr++;
134 			mp_opt->thmac = get_unaligned_be64(ptr);
135 			ptr += 8;
136 			mp_opt->nonce = get_unaligned_be32(ptr);
137 			ptr += 4;
138 			pr_debug("MP_JOIN bkup=%u, id=%u, thmac=%llu, nonce=%u",
139 				 mp_opt->backup, mp_opt->join_id,
140 				 mp_opt->thmac, mp_opt->nonce);
141 		} else if (opsize == TCPOLEN_MPTCP_MPJ_ACK) {
142 			ptr += 2;
143 			memcpy(mp_opt->hmac, ptr, MPTCPOPT_HMAC_LEN);
144 			pr_debug("MP_JOIN hmac");
145 		} else {
146 			mp_opt->suboptions &= ~OPTIONS_MPTCP_MPJ;
147 		}
148 		break;
149 
150 	case MPTCPOPT_DSS:
151 		pr_debug("DSS");
152 		ptr++;
153 
154 		/* we must clear 'mpc_map' be able to detect MP_CAPABLE
155 		 * map vs DSS map in mptcp_incoming_options(), and reconstruct
156 		 * map info accordingly
157 		 */
158 		mp_opt->mpc_map = 0;
159 		flags = (*ptr++) & MPTCP_DSS_FLAG_MASK;
160 		mp_opt->data_fin = (flags & MPTCP_DSS_DATA_FIN) != 0;
161 		mp_opt->dsn64 = (flags & MPTCP_DSS_DSN64) != 0;
162 		mp_opt->use_map = (flags & MPTCP_DSS_HAS_MAP) != 0;
163 		mp_opt->ack64 = (flags & MPTCP_DSS_ACK64) != 0;
164 		mp_opt->use_ack = (flags & MPTCP_DSS_HAS_ACK);
165 
166 		pr_debug("data_fin=%d dsn64=%d use_map=%d ack64=%d use_ack=%d",
167 			 mp_opt->data_fin, mp_opt->dsn64,
168 			 mp_opt->use_map, mp_opt->ack64,
169 			 mp_opt->use_ack);
170 
171 		expected_opsize = TCPOLEN_MPTCP_DSS_BASE;
172 
173 		if (mp_opt->use_ack) {
174 			if (mp_opt->ack64)
175 				expected_opsize += TCPOLEN_MPTCP_DSS_ACK64;
176 			else
177 				expected_opsize += TCPOLEN_MPTCP_DSS_ACK32;
178 		}
179 
180 		if (mp_opt->use_map) {
181 			if (mp_opt->dsn64)
182 				expected_opsize += TCPOLEN_MPTCP_DSS_MAP64;
183 			else
184 				expected_opsize += TCPOLEN_MPTCP_DSS_MAP32;
185 		}
186 
187 		/* Always parse any csum presence combination, we will enforce
188 		 * RFC 8684 Section 3.3.0 checks later in subflow_data_ready
189 		 */
190 		if (opsize != expected_opsize &&
191 		    opsize != expected_opsize + TCPOLEN_MPTCP_DSS_CHECKSUM)
192 			break;
193 
194 		mp_opt->suboptions |= OPTION_MPTCP_DSS;
195 		if (mp_opt->use_ack) {
196 			if (mp_opt->ack64) {
197 				mp_opt->data_ack = get_unaligned_be64(ptr);
198 				ptr += 8;
199 			} else {
200 				mp_opt->data_ack = get_unaligned_be32(ptr);
201 				ptr += 4;
202 			}
203 
204 			pr_debug("data_ack=%llu", mp_opt->data_ack);
205 		}
206 
207 		if (mp_opt->use_map) {
208 			if (mp_opt->dsn64) {
209 				mp_opt->data_seq = get_unaligned_be64(ptr);
210 				ptr += 8;
211 			} else {
212 				mp_opt->data_seq = get_unaligned_be32(ptr);
213 				ptr += 4;
214 			}
215 
216 			mp_opt->subflow_seq = get_unaligned_be32(ptr);
217 			ptr += 4;
218 
219 			mp_opt->data_len = get_unaligned_be16(ptr);
220 			ptr += 2;
221 
222 			if (opsize == expected_opsize + TCPOLEN_MPTCP_DSS_CHECKSUM) {
223 				mp_opt->suboptions |= OPTION_MPTCP_CSUMREQD;
224 				mp_opt->csum = (__force __sum16)get_unaligned_be16(ptr);
225 				ptr += 2;
226 			}
227 
228 			pr_debug("data_seq=%llu subflow_seq=%u data_len=%u csum=%d:%u",
229 				 mp_opt->data_seq, mp_opt->subflow_seq,
230 				 mp_opt->data_len, !!(mp_opt->suboptions & OPTION_MPTCP_CSUMREQD),
231 				 mp_opt->csum);
232 		}
233 
234 		break;
235 
236 	case MPTCPOPT_ADD_ADDR:
237 		mp_opt->echo = (*ptr++) & MPTCP_ADDR_ECHO;
238 		if (!mp_opt->echo) {
239 			if (opsize == TCPOLEN_MPTCP_ADD_ADDR ||
240 			    opsize == TCPOLEN_MPTCP_ADD_ADDR_PORT)
241 				mp_opt->addr.family = AF_INET;
242 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
243 			else if (opsize == TCPOLEN_MPTCP_ADD_ADDR6 ||
244 				 opsize == TCPOLEN_MPTCP_ADD_ADDR6_PORT)
245 				mp_opt->addr.family = AF_INET6;
246 #endif
247 			else
248 				break;
249 		} else {
250 			if (opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE ||
251 			    opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE_PORT)
252 				mp_opt->addr.family = AF_INET;
253 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
254 			else if (opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE ||
255 				 opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE_PORT)
256 				mp_opt->addr.family = AF_INET6;
257 #endif
258 			else
259 				break;
260 		}
261 
262 		mp_opt->suboptions |= OPTION_MPTCP_ADD_ADDR;
263 		mp_opt->addr.id = *ptr++;
264 		mp_opt->addr.port = 0;
265 		mp_opt->ahmac = 0;
266 		if (mp_opt->addr.family == AF_INET) {
267 			memcpy((u8 *)&mp_opt->addr.addr.s_addr, (u8 *)ptr, 4);
268 			ptr += 4;
269 			if (opsize == TCPOLEN_MPTCP_ADD_ADDR_PORT ||
270 			    opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE_PORT) {
271 				mp_opt->addr.port = htons(get_unaligned_be16(ptr));
272 				ptr += 2;
273 			}
274 		}
275 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
276 		else {
277 			memcpy(mp_opt->addr.addr6.s6_addr, (u8 *)ptr, 16);
278 			ptr += 16;
279 			if (opsize == TCPOLEN_MPTCP_ADD_ADDR6_PORT ||
280 			    opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE_PORT) {
281 				mp_opt->addr.port = htons(get_unaligned_be16(ptr));
282 				ptr += 2;
283 			}
284 		}
285 #endif
286 		if (!mp_opt->echo) {
287 			mp_opt->ahmac = get_unaligned_be64(ptr);
288 			ptr += 8;
289 		}
290 		pr_debug("ADD_ADDR%s: id=%d, ahmac=%llu, echo=%d, port=%d",
291 			 (mp_opt->addr.family == AF_INET6) ? "6" : "",
292 			 mp_opt->addr.id, mp_opt->ahmac, mp_opt->echo, ntohs(mp_opt->addr.port));
293 		break;
294 
295 	case MPTCPOPT_RM_ADDR:
296 		if (opsize < TCPOLEN_MPTCP_RM_ADDR_BASE + 1 ||
297 		    opsize > TCPOLEN_MPTCP_RM_ADDR_BASE + MPTCP_RM_IDS_MAX)
298 			break;
299 
300 		ptr++;
301 
302 		mp_opt->suboptions |= OPTION_MPTCP_RM_ADDR;
303 		mp_opt->rm_list.nr = opsize - TCPOLEN_MPTCP_RM_ADDR_BASE;
304 		for (i = 0; i < mp_opt->rm_list.nr; i++)
305 			mp_opt->rm_list.ids[i] = *ptr++;
306 		pr_debug("RM_ADDR: rm_list_nr=%d", mp_opt->rm_list.nr);
307 		break;
308 
309 	case MPTCPOPT_MP_PRIO:
310 		if (opsize != TCPOLEN_MPTCP_PRIO)
311 			break;
312 
313 		mp_opt->suboptions |= OPTION_MPTCP_PRIO;
314 		mp_opt->backup = *ptr++ & MPTCP_PRIO_BKUP;
315 		pr_debug("MP_PRIO: prio=%d", mp_opt->backup);
316 		break;
317 
318 	case MPTCPOPT_MP_FASTCLOSE:
319 		if (opsize != TCPOLEN_MPTCP_FASTCLOSE)
320 			break;
321 
322 		ptr += 2;
323 		mp_opt->rcvr_key = get_unaligned_be64(ptr);
324 		ptr += 8;
325 		mp_opt->suboptions |= OPTION_MPTCP_FASTCLOSE;
326 		break;
327 
328 	case MPTCPOPT_RST:
329 		if (opsize != TCPOLEN_MPTCP_RST)
330 			break;
331 
332 		if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_RST))
333 			break;
334 
335 		mp_opt->suboptions |= OPTION_MPTCP_RST;
336 		flags = *ptr++;
337 		mp_opt->reset_transient = flags & MPTCP_RST_TRANSIENT;
338 		mp_opt->reset_reason = *ptr;
339 		pr_debug("MP_RST: transient=%u reason=%u",
340 			 mp_opt->reset_transient, mp_opt->reset_reason);
341 		break;
342 
343 	case MPTCPOPT_MP_FAIL:
344 		if (opsize != TCPOLEN_MPTCP_FAIL)
345 			break;
346 
347 		ptr += 2;
348 		mp_opt->suboptions |= OPTION_MPTCP_FAIL;
349 		mp_opt->fail_seq = get_unaligned_be64(ptr);
350 		pr_debug("MP_FAIL: data_seq=%llu", mp_opt->fail_seq);
351 		break;
352 
353 	default:
354 		break;
355 	}
356 }
357 
358 void mptcp_get_options(const struct sk_buff *skb,
359 		       struct mptcp_options_received *mp_opt)
360 {
361 	const struct tcphdr *th = tcp_hdr(skb);
362 	const unsigned char *ptr;
363 	int length;
364 
365 	/* initialize option status */
366 	mp_opt->suboptions = 0;
367 
368 	length = (th->doff * 4) - sizeof(struct tcphdr);
369 	ptr = (const unsigned char *)(th + 1);
370 
371 	while (length > 0) {
372 		int opcode = *ptr++;
373 		int opsize;
374 
375 		switch (opcode) {
376 		case TCPOPT_EOL:
377 			return;
378 		case TCPOPT_NOP:	/* Ref: RFC 793 section 3.1 */
379 			length--;
380 			continue;
381 		default:
382 			if (length < 2)
383 				return;
384 			opsize = *ptr++;
385 			if (opsize < 2) /* "silly options" */
386 				return;
387 			if (opsize > length)
388 				return;	/* don't parse partial options */
389 			if (opcode == TCPOPT_MPTCP)
390 				mptcp_parse_option(skb, ptr, opsize, mp_opt);
391 			ptr += opsize - 2;
392 			length -= opsize;
393 		}
394 	}
395 }
396 
397 bool mptcp_syn_options(struct sock *sk, const struct sk_buff *skb,
398 		       unsigned int *size, struct mptcp_out_options *opts)
399 {
400 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
401 
402 	/* we will use snd_isn to detect first pkt [re]transmission
403 	 * in mptcp_established_options_mp()
404 	 */
405 	subflow->snd_isn = TCP_SKB_CB(skb)->end_seq;
406 	if (subflow->request_mptcp) {
407 		opts->suboptions = OPTION_MPTCP_MPC_SYN;
408 		opts->csum_reqd = mptcp_is_checksum_enabled(sock_net(sk));
409 		opts->allow_join_id0 = mptcp_allow_join_id0(sock_net(sk));
410 		*size = TCPOLEN_MPTCP_MPC_SYN;
411 		return true;
412 	} else if (subflow->request_join) {
413 		pr_debug("remote_token=%u, nonce=%u", subflow->remote_token,
414 			 subflow->local_nonce);
415 		opts->suboptions = OPTION_MPTCP_MPJ_SYN;
416 		opts->join_id = subflow->local_id;
417 		opts->token = subflow->remote_token;
418 		opts->nonce = subflow->local_nonce;
419 		opts->backup = subflow->request_bkup;
420 		*size = TCPOLEN_MPTCP_MPJ_SYN;
421 		return true;
422 	}
423 	return false;
424 }
425 
426 static void clear_3rdack_retransmission(struct sock *sk)
427 {
428 	struct inet_connection_sock *icsk = inet_csk(sk);
429 
430 	sk_stop_timer(sk, &icsk->icsk_delack_timer);
431 	icsk->icsk_ack.timeout = 0;
432 	icsk->icsk_ack.ato = 0;
433 	icsk->icsk_ack.pending &= ~(ICSK_ACK_SCHED | ICSK_ACK_TIMER);
434 }
435 
436 static bool mptcp_established_options_mp(struct sock *sk, struct sk_buff *skb,
437 					 bool snd_data_fin_enable,
438 					 unsigned int *size,
439 					 unsigned int remaining,
440 					 struct mptcp_out_options *opts)
441 {
442 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
443 	struct mptcp_sock *msk = mptcp_sk(subflow->conn);
444 	struct mptcp_ext *mpext;
445 	unsigned int data_len;
446 	u8 len;
447 
448 	/* When skb is not available, we better over-estimate the emitted
449 	 * options len. A full DSS option (28 bytes) is longer than
450 	 * TCPOLEN_MPTCP_MPC_ACK_DATA(22) or TCPOLEN_MPTCP_MPJ_ACK(24), so
451 	 * tell the caller to defer the estimate to
452 	 * mptcp_established_options_dss(), which will reserve enough space.
453 	 */
454 	if (!skb)
455 		return false;
456 
457 	/* MPC/MPJ needed only on 3rd ack packet, DATA_FIN and TCP shutdown take precedence */
458 	if (subflow->fully_established || snd_data_fin_enable ||
459 	    subflow->snd_isn != TCP_SKB_CB(skb)->seq ||
460 	    sk->sk_state != TCP_ESTABLISHED)
461 		return false;
462 
463 	if (subflow->mp_capable) {
464 		mpext = mptcp_get_ext(skb);
465 		data_len = mpext ? mpext->data_len : 0;
466 
467 		/* we will check ops->data_len in mptcp_write_options() to
468 		 * discriminate between TCPOLEN_MPTCP_MPC_ACK_DATA and
469 		 * TCPOLEN_MPTCP_MPC_ACK
470 		 */
471 		opts->data_len = data_len;
472 		opts->suboptions = OPTION_MPTCP_MPC_ACK;
473 		opts->sndr_key = subflow->local_key;
474 		opts->rcvr_key = subflow->remote_key;
475 		opts->csum_reqd = READ_ONCE(msk->csum_enabled);
476 		opts->allow_join_id0 = mptcp_allow_join_id0(sock_net(sk));
477 
478 		/* Section 3.1.
479 		 * The MP_CAPABLE option is carried on the SYN, SYN/ACK, and ACK
480 		 * packets that start the first subflow of an MPTCP connection,
481 		 * as well as the first packet that carries data
482 		 */
483 		if (data_len > 0) {
484 			len = TCPOLEN_MPTCP_MPC_ACK_DATA;
485 			if (opts->csum_reqd) {
486 				/* we need to propagate more info to csum the pseudo hdr */
487 				opts->data_seq = mpext->data_seq;
488 				opts->subflow_seq = mpext->subflow_seq;
489 				opts->csum = mpext->csum;
490 				len += TCPOLEN_MPTCP_DSS_CHECKSUM;
491 			}
492 			*size = ALIGN(len, 4);
493 		} else {
494 			*size = TCPOLEN_MPTCP_MPC_ACK;
495 		}
496 
497 		pr_debug("subflow=%p, local_key=%llu, remote_key=%llu map_len=%d",
498 			 subflow, subflow->local_key, subflow->remote_key,
499 			 data_len);
500 
501 		return true;
502 	} else if (subflow->mp_join) {
503 		opts->suboptions = OPTION_MPTCP_MPJ_ACK;
504 		memcpy(opts->hmac, subflow->hmac, MPTCPOPT_HMAC_LEN);
505 		*size = TCPOLEN_MPTCP_MPJ_ACK;
506 		pr_debug("subflow=%p", subflow);
507 
508 		/* we can use the full delegate action helper only from BH context
509 		 * If we are in process context - sk is flushing the backlog at
510 		 * socket lock release time - just set the appropriate flag, will
511 		 * be handled by the release callback
512 		 */
513 		if (sock_owned_by_user(sk))
514 			set_bit(MPTCP_DELEGATE_ACK, &subflow->delegated_status);
515 		else
516 			mptcp_subflow_delegate(subflow, MPTCP_DELEGATE_ACK);
517 		return true;
518 	}
519 	return false;
520 }
521 
522 static void mptcp_write_data_fin(struct mptcp_subflow_context *subflow,
523 				 struct sk_buff *skb, struct mptcp_ext *ext)
524 {
525 	/* The write_seq value has already been incremented, so the actual
526 	 * sequence number for the DATA_FIN is one less.
527 	 */
528 	u64 data_fin_tx_seq = READ_ONCE(mptcp_sk(subflow->conn)->write_seq) - 1;
529 
530 	if (!ext->use_map || !skb->len) {
531 		/* RFC6824 requires a DSS mapping with specific values
532 		 * if DATA_FIN is set but no data payload is mapped
533 		 */
534 		ext->data_fin = 1;
535 		ext->use_map = 1;
536 		ext->dsn64 = 1;
537 		ext->data_seq = data_fin_tx_seq;
538 		ext->subflow_seq = 0;
539 		ext->data_len = 1;
540 	} else if (ext->data_seq + ext->data_len == data_fin_tx_seq) {
541 		/* If there's an existing DSS mapping and it is the
542 		 * final mapping, DATA_FIN consumes 1 additional byte of
543 		 * mapping space.
544 		 */
545 		ext->data_fin = 1;
546 		ext->data_len++;
547 	}
548 }
549 
550 static bool mptcp_established_options_dss(struct sock *sk, struct sk_buff *skb,
551 					  bool snd_data_fin_enable,
552 					  unsigned int *size,
553 					  unsigned int remaining,
554 					  struct mptcp_out_options *opts)
555 {
556 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
557 	struct mptcp_sock *msk = mptcp_sk(subflow->conn);
558 	unsigned int dss_size = 0;
559 	struct mptcp_ext *mpext;
560 	unsigned int ack_size;
561 	bool ret = false;
562 	u64 ack_seq;
563 
564 	opts->csum_reqd = READ_ONCE(msk->csum_enabled);
565 	mpext = skb ? mptcp_get_ext(skb) : NULL;
566 
567 	if (!skb || (mpext && mpext->use_map) || snd_data_fin_enable) {
568 		unsigned int map_size = TCPOLEN_MPTCP_DSS_BASE + TCPOLEN_MPTCP_DSS_MAP64;
569 
570 		if (mpext) {
571 			if (opts->csum_reqd)
572 				map_size += TCPOLEN_MPTCP_DSS_CHECKSUM;
573 
574 			opts->ext_copy = *mpext;
575 		}
576 
577 		remaining -= map_size;
578 		dss_size = map_size;
579 		if (skb && snd_data_fin_enable)
580 			mptcp_write_data_fin(subflow, skb, &opts->ext_copy);
581 		opts->suboptions = OPTION_MPTCP_DSS;
582 		ret = true;
583 	}
584 
585 	/* passive sockets msk will set the 'can_ack' after accept(), even
586 	 * if the first subflow may have the already the remote key handy
587 	 */
588 	opts->ext_copy.use_ack = 0;
589 	if (!READ_ONCE(msk->can_ack)) {
590 		*size = ALIGN(dss_size, 4);
591 		return ret;
592 	}
593 
594 	ack_seq = READ_ONCE(msk->ack_seq);
595 	if (READ_ONCE(msk->use_64bit_ack)) {
596 		ack_size = TCPOLEN_MPTCP_DSS_ACK64;
597 		opts->ext_copy.data_ack = ack_seq;
598 		opts->ext_copy.ack64 = 1;
599 	} else {
600 		ack_size = TCPOLEN_MPTCP_DSS_ACK32;
601 		opts->ext_copy.data_ack32 = (uint32_t)ack_seq;
602 		opts->ext_copy.ack64 = 0;
603 	}
604 	opts->ext_copy.use_ack = 1;
605 	opts->suboptions = OPTION_MPTCP_DSS;
606 	WRITE_ONCE(msk->old_wspace, __mptcp_space((struct sock *)msk));
607 
608 	/* Add kind/length/subtype/flag overhead if mapping is not populated */
609 	if (dss_size == 0)
610 		ack_size += TCPOLEN_MPTCP_DSS_BASE;
611 
612 	dss_size += ack_size;
613 
614 	*size = ALIGN(dss_size, 4);
615 	return true;
616 }
617 
618 static u64 add_addr_generate_hmac(u64 key1, u64 key2,
619 				  struct mptcp_addr_info *addr)
620 {
621 	u16 port = ntohs(addr->port);
622 	u8 hmac[SHA256_DIGEST_SIZE];
623 	u8 msg[19];
624 	int i = 0;
625 
626 	msg[i++] = addr->id;
627 	if (addr->family == AF_INET) {
628 		memcpy(&msg[i], &addr->addr.s_addr, 4);
629 		i += 4;
630 	}
631 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
632 	else if (addr->family == AF_INET6) {
633 		memcpy(&msg[i], &addr->addr6.s6_addr, 16);
634 		i += 16;
635 	}
636 #endif
637 	msg[i++] = port >> 8;
638 	msg[i++] = port & 0xFF;
639 
640 	mptcp_crypto_hmac_sha(key1, key2, msg, i, hmac);
641 
642 	return get_unaligned_be64(&hmac[SHA256_DIGEST_SIZE - sizeof(u64)]);
643 }
644 
645 static bool mptcp_established_options_add_addr(struct sock *sk, struct sk_buff *skb,
646 					       unsigned int *size,
647 					       unsigned int remaining,
648 					       struct mptcp_out_options *opts)
649 {
650 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
651 	struct mptcp_sock *msk = mptcp_sk(subflow->conn);
652 	bool drop_other_suboptions = false;
653 	unsigned int opt_size = *size;
654 	bool echo;
655 	int len;
656 
657 	/* add addr will strip the existing options, be sure to avoid breaking
658 	 * MPC/MPJ handshakes
659 	 */
660 	if (!mptcp_pm_should_add_signal(msk) ||
661 	    (opts->suboptions & (OPTION_MPTCP_MPJ_ACK | OPTION_MPTCP_MPC_ACK)) ||
662 	    !mptcp_pm_add_addr_signal(msk, skb, opt_size, remaining, &opts->addr,
663 		    &echo, &drop_other_suboptions))
664 		return false;
665 
666 	if (drop_other_suboptions)
667 		remaining += opt_size;
668 	len = mptcp_add_addr_len(opts->addr.family, echo, !!opts->addr.port);
669 	if (remaining < len)
670 		return false;
671 
672 	*size = len;
673 	if (drop_other_suboptions) {
674 		pr_debug("drop other suboptions");
675 		opts->suboptions = 0;
676 
677 		/* note that e.g. DSS could have written into the memory
678 		 * aliased by ahmac, we must reset the field here
679 		 * to avoid appending the hmac even for ADD_ADDR echo
680 		 * options
681 		 */
682 		opts->ahmac = 0;
683 		*size -= opt_size;
684 	}
685 	opts->suboptions |= OPTION_MPTCP_ADD_ADDR;
686 	if (!echo) {
687 		opts->ahmac = add_addr_generate_hmac(msk->local_key,
688 						     msk->remote_key,
689 						     &opts->addr);
690 	}
691 	pr_debug("addr_id=%d, ahmac=%llu, echo=%d, port=%d",
692 		 opts->addr.id, opts->ahmac, echo, ntohs(opts->addr.port));
693 
694 	return true;
695 }
696 
697 static bool mptcp_established_options_rm_addr(struct sock *sk,
698 					      unsigned int *size,
699 					      unsigned int remaining,
700 					      struct mptcp_out_options *opts)
701 {
702 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
703 	struct mptcp_sock *msk = mptcp_sk(subflow->conn);
704 	struct mptcp_rm_list rm_list;
705 	int i, len;
706 
707 	if (!mptcp_pm_should_rm_signal(msk) ||
708 	    !(mptcp_pm_rm_addr_signal(msk, remaining, &rm_list)))
709 		return false;
710 
711 	len = mptcp_rm_addr_len(&rm_list);
712 	if (len < 0)
713 		return false;
714 	if (remaining < len)
715 		return false;
716 
717 	*size = len;
718 	opts->suboptions |= OPTION_MPTCP_RM_ADDR;
719 	opts->rm_list = rm_list;
720 
721 	for (i = 0; i < opts->rm_list.nr; i++)
722 		pr_debug("rm_list_ids[%d]=%d", i, opts->rm_list.ids[i]);
723 
724 	return true;
725 }
726 
727 static bool mptcp_established_options_mp_prio(struct sock *sk,
728 					      unsigned int *size,
729 					      unsigned int remaining,
730 					      struct mptcp_out_options *opts)
731 {
732 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
733 
734 	/* can't send MP_PRIO with MPC, as they share the same option space:
735 	 * 'backup'. Also it makes no sense at all
736 	 */
737 	if (!subflow->send_mp_prio || (opts->suboptions & OPTIONS_MPTCP_MPC))
738 		return false;
739 
740 	/* account for the trailing 'nop' option */
741 	if (remaining < TCPOLEN_MPTCP_PRIO_ALIGN)
742 		return false;
743 
744 	*size = TCPOLEN_MPTCP_PRIO_ALIGN;
745 	opts->suboptions |= OPTION_MPTCP_PRIO;
746 	opts->backup = subflow->request_bkup;
747 
748 	pr_debug("prio=%d", opts->backup);
749 
750 	return true;
751 }
752 
753 static noinline bool mptcp_established_options_rst(struct sock *sk, struct sk_buff *skb,
754 						   unsigned int *size,
755 						   unsigned int remaining,
756 						   struct mptcp_out_options *opts)
757 {
758 	const struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
759 
760 	if (remaining < TCPOLEN_MPTCP_RST)
761 		return false;
762 
763 	*size = TCPOLEN_MPTCP_RST;
764 	opts->suboptions |= OPTION_MPTCP_RST;
765 	opts->reset_transient = subflow->reset_transient;
766 	opts->reset_reason = subflow->reset_reason;
767 
768 	return true;
769 }
770 
771 static bool mptcp_established_options_fastclose(struct sock *sk,
772 						unsigned int *size,
773 						unsigned int remaining,
774 						struct mptcp_out_options *opts)
775 {
776 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
777 	struct mptcp_sock *msk = mptcp_sk(subflow->conn);
778 
779 	if (likely(!subflow->send_fastclose))
780 		return false;
781 
782 	if (remaining < TCPOLEN_MPTCP_FASTCLOSE)
783 		return false;
784 
785 	*size = TCPOLEN_MPTCP_FASTCLOSE;
786 	opts->suboptions |= OPTION_MPTCP_FASTCLOSE;
787 	opts->rcvr_key = msk->remote_key;
788 
789 	pr_debug("FASTCLOSE key=%llu", opts->rcvr_key);
790 	return true;
791 }
792 
793 static bool mptcp_established_options_mp_fail(struct sock *sk,
794 					      unsigned int *size,
795 					      unsigned int remaining,
796 					      struct mptcp_out_options *opts)
797 {
798 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
799 
800 	if (likely(!subflow->send_mp_fail))
801 		return false;
802 
803 	if (remaining < TCPOLEN_MPTCP_FAIL)
804 		return false;
805 
806 	*size = TCPOLEN_MPTCP_FAIL;
807 	opts->suboptions |= OPTION_MPTCP_FAIL;
808 	opts->fail_seq = subflow->map_seq;
809 
810 	pr_debug("MP_FAIL fail_seq=%llu", opts->fail_seq);
811 
812 	return true;
813 }
814 
815 bool mptcp_established_options(struct sock *sk, struct sk_buff *skb,
816 			       unsigned int *size, unsigned int remaining,
817 			       struct mptcp_out_options *opts)
818 {
819 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
820 	struct mptcp_sock *msk = mptcp_sk(subflow->conn);
821 	unsigned int opt_size = 0;
822 	bool snd_data_fin;
823 	bool ret = false;
824 
825 	opts->suboptions = 0;
826 
827 	if (unlikely(__mptcp_check_fallback(msk)))
828 		return false;
829 
830 	if (unlikely(skb && TCP_SKB_CB(skb)->tcp_flags & TCPHDR_RST)) {
831 		if (mptcp_established_options_fastclose(sk, &opt_size, remaining, opts) ||
832 		    mptcp_established_options_mp_fail(sk, &opt_size, remaining, opts)) {
833 			*size += opt_size;
834 			remaining -= opt_size;
835 		}
836 		/* MP_RST can be used with MP_FASTCLOSE and MP_FAIL if there is room */
837 		if (mptcp_established_options_rst(sk, skb, &opt_size, remaining, opts)) {
838 			*size += opt_size;
839 			remaining -= opt_size;
840 		}
841 		return true;
842 	}
843 
844 	snd_data_fin = mptcp_data_fin_enabled(msk);
845 	if (mptcp_established_options_mp(sk, skb, snd_data_fin, &opt_size, remaining, opts))
846 		ret = true;
847 	else if (mptcp_established_options_dss(sk, skb, snd_data_fin, &opt_size, remaining, opts)) {
848 		unsigned int mp_fail_size;
849 
850 		ret = true;
851 		if (mptcp_established_options_mp_fail(sk, &mp_fail_size,
852 						      remaining - opt_size, opts)) {
853 			*size += opt_size + mp_fail_size;
854 			remaining -= opt_size - mp_fail_size;
855 			return true;
856 		}
857 	}
858 
859 	/* we reserved enough space for the above options, and exceeding the
860 	 * TCP option space would be fatal
861 	 */
862 	if (WARN_ON_ONCE(opt_size > remaining))
863 		return false;
864 
865 	*size += opt_size;
866 	remaining -= opt_size;
867 	if (mptcp_established_options_add_addr(sk, skb, &opt_size, remaining, opts)) {
868 		*size += opt_size;
869 		remaining -= opt_size;
870 		ret = true;
871 	} else if (mptcp_established_options_rm_addr(sk, &opt_size, remaining, opts)) {
872 		*size += opt_size;
873 		remaining -= opt_size;
874 		ret = true;
875 	}
876 
877 	if (mptcp_established_options_mp_prio(sk, &opt_size, remaining, opts)) {
878 		*size += opt_size;
879 		remaining -= opt_size;
880 		ret = true;
881 	}
882 
883 	return ret;
884 }
885 
886 bool mptcp_synack_options(const struct request_sock *req, unsigned int *size,
887 			  struct mptcp_out_options *opts)
888 {
889 	struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
890 
891 	if (subflow_req->mp_capable) {
892 		opts->suboptions = OPTION_MPTCP_MPC_SYNACK;
893 		opts->sndr_key = subflow_req->local_key;
894 		opts->csum_reqd = subflow_req->csum_reqd;
895 		opts->allow_join_id0 = subflow_req->allow_join_id0;
896 		*size = TCPOLEN_MPTCP_MPC_SYNACK;
897 		pr_debug("subflow_req=%p, local_key=%llu",
898 			 subflow_req, subflow_req->local_key);
899 		return true;
900 	} else if (subflow_req->mp_join) {
901 		opts->suboptions = OPTION_MPTCP_MPJ_SYNACK;
902 		opts->backup = subflow_req->backup;
903 		opts->join_id = subflow_req->local_id;
904 		opts->thmac = subflow_req->thmac;
905 		opts->nonce = subflow_req->local_nonce;
906 		pr_debug("req=%p, bkup=%u, id=%u, thmac=%llu, nonce=%u",
907 			 subflow_req, opts->backup, opts->join_id,
908 			 opts->thmac, opts->nonce);
909 		*size = TCPOLEN_MPTCP_MPJ_SYNACK;
910 		return true;
911 	}
912 	return false;
913 }
914 
915 static bool check_fully_established(struct mptcp_sock *msk, struct sock *ssk,
916 				    struct mptcp_subflow_context *subflow,
917 				    struct sk_buff *skb,
918 				    struct mptcp_options_received *mp_opt)
919 {
920 	/* here we can process OoO, in-window pkts, only in-sequence 4th ack
921 	 * will make the subflow fully established
922 	 */
923 	if (likely(subflow->fully_established)) {
924 		/* on passive sockets, check for 3rd ack retransmission
925 		 * note that msk is always set by subflow_syn_recv_sock()
926 		 * for mp_join subflows
927 		 */
928 		if (TCP_SKB_CB(skb)->seq == subflow->ssn_offset + 1 &&
929 		    TCP_SKB_CB(skb)->end_seq == TCP_SKB_CB(skb)->seq &&
930 		    subflow->mp_join && (mp_opt->suboptions & OPTIONS_MPTCP_MPJ) &&
931 		    READ_ONCE(msk->pm.server_side))
932 			tcp_send_ack(ssk);
933 		goto fully_established;
934 	}
935 
936 	/* we must process OoO packets before the first subflow is fully
937 	 * established. OoO packets are instead a protocol violation
938 	 * for MP_JOIN subflows as the peer must not send any data
939 	 * before receiving the forth ack - cfr. RFC 8684 section 3.2.
940 	 */
941 	if (TCP_SKB_CB(skb)->seq != subflow->ssn_offset + 1) {
942 		if (subflow->mp_join)
943 			goto reset;
944 		return subflow->mp_capable;
945 	}
946 
947 	if (((mp_opt->suboptions & OPTION_MPTCP_DSS) && mp_opt->use_ack) ||
948 	    ((mp_opt->suboptions & OPTION_MPTCP_ADD_ADDR) && !mp_opt->echo)) {
949 		/* subflows are fully established as soon as we get any
950 		 * additional ack, including ADD_ADDR.
951 		 */
952 		subflow->fully_established = 1;
953 		WRITE_ONCE(msk->fully_established, true);
954 		goto fully_established;
955 	}
956 
957 	/* If the first established packet does not contain MP_CAPABLE + data
958 	 * then fallback to TCP. Fallback scenarios requires a reset for
959 	 * MP_JOIN subflows.
960 	 */
961 	if (!(mp_opt->suboptions & OPTIONS_MPTCP_MPC)) {
962 		if (subflow->mp_join)
963 			goto reset;
964 		subflow->mp_capable = 0;
965 		pr_fallback(msk);
966 		__mptcp_do_fallback(msk);
967 		return false;
968 	}
969 
970 	if (mp_opt->deny_join_id0)
971 		WRITE_ONCE(msk->pm.remote_deny_join_id0, true);
972 
973 	if (unlikely(!READ_ONCE(msk->pm.server_side)))
974 		pr_warn_once("bogus mpc option on established client sk");
975 	mptcp_subflow_fully_established(subflow, mp_opt);
976 
977 fully_established:
978 	/* if the subflow is not already linked into the conn_list, we can't
979 	 * notify the PM: this subflow is still on the listener queue
980 	 * and the PM possibly acquiring the subflow lock could race with
981 	 * the listener close
982 	 */
983 	if (likely(subflow->pm_notified) || list_empty(&subflow->node))
984 		return true;
985 
986 	subflow->pm_notified = 1;
987 	if (subflow->mp_join) {
988 		clear_3rdack_retransmission(ssk);
989 		mptcp_pm_subflow_established(msk);
990 	} else {
991 		mptcp_pm_fully_established(msk, ssk, GFP_ATOMIC);
992 	}
993 	return true;
994 
995 reset:
996 	mptcp_subflow_reset(ssk);
997 	return false;
998 }
999 
1000 u64 __mptcp_expand_seq(u64 old_seq, u64 cur_seq)
1001 {
1002 	u32 old_seq32, cur_seq32;
1003 
1004 	old_seq32 = (u32)old_seq;
1005 	cur_seq32 = (u32)cur_seq;
1006 	cur_seq = (old_seq & GENMASK_ULL(63, 32)) + cur_seq32;
1007 	if (unlikely(cur_seq32 < old_seq32 && before(old_seq32, cur_seq32)))
1008 		return cur_seq + (1LL << 32);
1009 
1010 	/* reverse wrap could happen, too */
1011 	if (unlikely(cur_seq32 > old_seq32 && after(old_seq32, cur_seq32)))
1012 		return cur_seq - (1LL << 32);
1013 	return cur_seq;
1014 }
1015 
1016 static void ack_update_msk(struct mptcp_sock *msk,
1017 			   struct sock *ssk,
1018 			   struct mptcp_options_received *mp_opt)
1019 {
1020 	u64 new_wnd_end, new_snd_una, snd_nxt = READ_ONCE(msk->snd_nxt);
1021 	struct sock *sk = (struct sock *)msk;
1022 	u64 old_snd_una;
1023 
1024 	mptcp_data_lock(sk);
1025 
1026 	/* avoid ack expansion on update conflict, to reduce the risk of
1027 	 * wrongly expanding to a future ack sequence number, which is way
1028 	 * more dangerous than missing an ack
1029 	 */
1030 	old_snd_una = msk->snd_una;
1031 	new_snd_una = mptcp_expand_seq(old_snd_una, mp_opt->data_ack, mp_opt->ack64);
1032 
1033 	/* ACK for data not even sent yet? Ignore.*/
1034 	if (unlikely(after64(new_snd_una, snd_nxt)))
1035 		new_snd_una = old_snd_una;
1036 
1037 	new_wnd_end = new_snd_una + tcp_sk(ssk)->snd_wnd;
1038 
1039 	if (after64(new_wnd_end, msk->wnd_end))
1040 		msk->wnd_end = new_wnd_end;
1041 
1042 	/* this assumes mptcp_incoming_options() is invoked after tcp_ack() */
1043 	if (after64(msk->wnd_end, READ_ONCE(msk->snd_nxt)))
1044 		__mptcp_check_push(sk, ssk);
1045 
1046 	if (after64(new_snd_una, old_snd_una)) {
1047 		msk->snd_una = new_snd_una;
1048 		__mptcp_data_acked(sk);
1049 	}
1050 	mptcp_data_unlock(sk);
1051 
1052 	trace_ack_update_msk(mp_opt->data_ack,
1053 			     old_snd_una, new_snd_una,
1054 			     new_wnd_end, msk->wnd_end);
1055 }
1056 
1057 bool mptcp_update_rcv_data_fin(struct mptcp_sock *msk, u64 data_fin_seq, bool use_64bit)
1058 {
1059 	/* Skip if DATA_FIN was already received.
1060 	 * If updating simultaneously with the recvmsg loop, values
1061 	 * should match. If they mismatch, the peer is misbehaving and
1062 	 * we will prefer the most recent information.
1063 	 */
1064 	if (READ_ONCE(msk->rcv_data_fin))
1065 		return false;
1066 
1067 	WRITE_ONCE(msk->rcv_data_fin_seq,
1068 		   mptcp_expand_seq(READ_ONCE(msk->ack_seq), data_fin_seq, use_64bit));
1069 	WRITE_ONCE(msk->rcv_data_fin, 1);
1070 
1071 	return true;
1072 }
1073 
1074 static bool add_addr_hmac_valid(struct mptcp_sock *msk,
1075 				struct mptcp_options_received *mp_opt)
1076 {
1077 	u64 hmac = 0;
1078 
1079 	if (mp_opt->echo)
1080 		return true;
1081 
1082 	hmac = add_addr_generate_hmac(msk->remote_key,
1083 				      msk->local_key,
1084 				      &mp_opt->addr);
1085 
1086 	pr_debug("msk=%p, ahmac=%llu, mp_opt->ahmac=%llu\n",
1087 		 msk, hmac, mp_opt->ahmac);
1088 
1089 	return hmac == mp_opt->ahmac;
1090 }
1091 
1092 /* Return false if a subflow has been reset, else return true */
1093 bool mptcp_incoming_options(struct sock *sk, struct sk_buff *skb)
1094 {
1095 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1096 	struct mptcp_sock *msk = mptcp_sk(subflow->conn);
1097 	struct mptcp_options_received mp_opt;
1098 	struct mptcp_ext *mpext;
1099 
1100 	if (__mptcp_check_fallback(msk)) {
1101 		/* Keep it simple and unconditionally trigger send data cleanup and
1102 		 * pending queue spooling. We will need to acquire the data lock
1103 		 * for more accurate checks, and once the lock is acquired, such
1104 		 * helpers are cheap.
1105 		 */
1106 		mptcp_data_lock(subflow->conn);
1107 		if (sk_stream_memory_free(sk))
1108 			__mptcp_check_push(subflow->conn, sk);
1109 		__mptcp_data_acked(subflow->conn);
1110 		mptcp_data_unlock(subflow->conn);
1111 		return true;
1112 	}
1113 
1114 	mptcp_get_options(skb, &mp_opt);
1115 
1116 	/* The subflow can be in close state only if check_fully_established()
1117 	 * just sent a reset. If so, tell the caller to ignore the current packet.
1118 	 */
1119 	if (!check_fully_established(msk, sk, subflow, skb, &mp_opt))
1120 		return sk->sk_state != TCP_CLOSE;
1121 
1122 	if (unlikely(mp_opt.suboptions != OPTION_MPTCP_DSS)) {
1123 		if ((mp_opt.suboptions & OPTION_MPTCP_FASTCLOSE) &&
1124 		    msk->local_key == mp_opt.rcvr_key) {
1125 			WRITE_ONCE(msk->rcv_fastclose, true);
1126 			mptcp_schedule_work((struct sock *)msk);
1127 		}
1128 
1129 		if ((mp_opt.suboptions & OPTION_MPTCP_ADD_ADDR) &&
1130 		    add_addr_hmac_valid(msk, &mp_opt)) {
1131 			if (!mp_opt.echo) {
1132 				mptcp_pm_add_addr_received(msk, &mp_opt.addr);
1133 				MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ADDADDR);
1134 			} else {
1135 				mptcp_pm_add_addr_echoed(msk, &mp_opt.addr);
1136 				mptcp_pm_del_add_timer(msk, &mp_opt.addr, true);
1137 				MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ECHOADD);
1138 			}
1139 
1140 			if (mp_opt.addr.port)
1141 				MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_PORTADD);
1142 		}
1143 
1144 		if (mp_opt.suboptions & OPTION_MPTCP_RM_ADDR)
1145 			mptcp_pm_rm_addr_received(msk, &mp_opt.rm_list);
1146 
1147 		if (mp_opt.suboptions & OPTION_MPTCP_PRIO) {
1148 			mptcp_pm_mp_prio_received(sk, mp_opt.backup);
1149 			MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPPRIORX);
1150 		}
1151 
1152 		if (mp_opt.suboptions & OPTION_MPTCP_FAIL) {
1153 			mptcp_pm_mp_fail_received(sk, mp_opt.fail_seq);
1154 			MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPFAILRX);
1155 		}
1156 
1157 		if (mp_opt.suboptions & OPTION_MPTCP_RST) {
1158 			subflow->reset_seen = 1;
1159 			subflow->reset_reason = mp_opt.reset_reason;
1160 			subflow->reset_transient = mp_opt.reset_transient;
1161 		}
1162 
1163 		if (!(mp_opt.suboptions & OPTION_MPTCP_DSS))
1164 			return true;
1165 	}
1166 
1167 	/* we can't wait for recvmsg() to update the ack_seq, otherwise
1168 	 * monodirectional flows will stuck
1169 	 */
1170 	if (mp_opt.use_ack)
1171 		ack_update_msk(msk, sk, &mp_opt);
1172 
1173 	/* Zero-data-length packets are dropped by the caller and not
1174 	 * propagated to the MPTCP layer, so the skb extension does not
1175 	 * need to be allocated or populated. DATA_FIN information, if
1176 	 * present, needs to be updated here before the skb is freed.
1177 	 */
1178 	if (TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb)->end_seq) {
1179 		if (mp_opt.data_fin && mp_opt.data_len == 1 &&
1180 		    mptcp_update_rcv_data_fin(msk, mp_opt.data_seq, mp_opt.dsn64) &&
1181 		    schedule_work(&msk->work))
1182 			sock_hold(subflow->conn);
1183 
1184 		return true;
1185 	}
1186 
1187 	mpext = skb_ext_add(skb, SKB_EXT_MPTCP);
1188 	if (!mpext)
1189 		return true;
1190 
1191 	memset(mpext, 0, sizeof(*mpext));
1192 
1193 	if (likely(mp_opt.use_map)) {
1194 		if (mp_opt.mpc_map) {
1195 			/* this is an MP_CAPABLE carrying MPTCP data
1196 			 * we know this map the first chunk of data
1197 			 */
1198 			mptcp_crypto_key_sha(subflow->remote_key, NULL,
1199 					     &mpext->data_seq);
1200 			mpext->data_seq++;
1201 			mpext->subflow_seq = 1;
1202 			mpext->dsn64 = 1;
1203 			mpext->mpc_map = 1;
1204 			mpext->data_fin = 0;
1205 		} else {
1206 			mpext->data_seq = mp_opt.data_seq;
1207 			mpext->subflow_seq = mp_opt.subflow_seq;
1208 			mpext->dsn64 = mp_opt.dsn64;
1209 			mpext->data_fin = mp_opt.data_fin;
1210 		}
1211 		mpext->data_len = mp_opt.data_len;
1212 		mpext->use_map = 1;
1213 		mpext->csum_reqd = !!(mp_opt.suboptions & OPTION_MPTCP_CSUMREQD);
1214 
1215 		if (mpext->csum_reqd)
1216 			mpext->csum = mp_opt.csum;
1217 	}
1218 
1219 	return true;
1220 }
1221 
1222 static void mptcp_set_rwin(const struct tcp_sock *tp)
1223 {
1224 	const struct sock *ssk = (const struct sock *)tp;
1225 	const struct mptcp_subflow_context *subflow;
1226 	struct mptcp_sock *msk;
1227 	u64 ack_seq;
1228 
1229 	subflow = mptcp_subflow_ctx(ssk);
1230 	msk = mptcp_sk(subflow->conn);
1231 
1232 	ack_seq = READ_ONCE(msk->ack_seq) + tp->rcv_wnd;
1233 
1234 	if (after64(ack_seq, READ_ONCE(msk->rcv_wnd_sent)))
1235 		WRITE_ONCE(msk->rcv_wnd_sent, ack_seq);
1236 }
1237 
1238 u16 __mptcp_make_csum(u64 data_seq, u32 subflow_seq, u16 data_len, __wsum sum)
1239 {
1240 	struct csum_pseudo_header header;
1241 	__wsum csum;
1242 
1243 	/* cfr RFC 8684 3.3.1.:
1244 	 * the data sequence number used in the pseudo-header is
1245 	 * always the 64-bit value, irrespective of what length is used in the
1246 	 * DSS option itself.
1247 	 */
1248 	header.data_seq = cpu_to_be64(data_seq);
1249 	header.subflow_seq = htonl(subflow_seq);
1250 	header.data_len = htons(data_len);
1251 	header.csum = 0;
1252 
1253 	csum = csum_partial(&header, sizeof(header), sum);
1254 	return (__force u16)csum_fold(csum);
1255 }
1256 
1257 static u16 mptcp_make_csum(const struct mptcp_ext *mpext)
1258 {
1259 	return __mptcp_make_csum(mpext->data_seq, mpext->subflow_seq, mpext->data_len,
1260 				 ~csum_unfold(mpext->csum));
1261 }
1262 
1263 void mptcp_write_options(__be32 *ptr, const struct tcp_sock *tp,
1264 			 struct mptcp_out_options *opts)
1265 {
1266 	const struct sock *ssk = (const struct sock *)tp;
1267 	struct mptcp_subflow_context *subflow;
1268 
1269 	/* Which options can be used together?
1270 	 *
1271 	 * X: mutually exclusive
1272 	 * O: often used together
1273 	 * C: can be used together in some cases
1274 	 * P: could be used together but we prefer not to (optimisations)
1275 	 *
1276 	 *  Opt: | MPC  | MPJ  | DSS  | ADD  |  RM  | PRIO | FAIL |  FC  |
1277 	 * ------|------|------|------|------|------|------|------|------|
1278 	 *  MPC  |------|------|------|------|------|------|------|------|
1279 	 *  MPJ  |  X   |------|------|------|------|------|------|------|
1280 	 *  DSS  |  X   |  X   |------|------|------|------|------|------|
1281 	 *  ADD  |  X   |  X   |  P   |------|------|------|------|------|
1282 	 *  RM   |  C   |  C   |  C   |  P   |------|------|------|------|
1283 	 *  PRIO |  X   |  C   |  C   |  C   |  C   |------|------|------|
1284 	 *  FAIL |  X   |  X   |  C   |  X   |  X   |  X   |------|------|
1285 	 *  FC   |  X   |  X   |  X   |  X   |  X   |  X   |  X   |------|
1286 	 *  RST  |  X   |  X   |  X   |  X   |  X   |  X   |  O   |  O   |
1287 	 * ------|------|------|------|------|------|------|------|------|
1288 	 *
1289 	 * The same applies in mptcp_established_options() function.
1290 	 */
1291 	if (likely(OPTION_MPTCP_DSS & opts->suboptions)) {
1292 		struct mptcp_ext *mpext = &opts->ext_copy;
1293 		u8 len = TCPOLEN_MPTCP_DSS_BASE;
1294 		u8 flags = 0;
1295 
1296 		if (mpext->use_ack) {
1297 			flags = MPTCP_DSS_HAS_ACK;
1298 			if (mpext->ack64) {
1299 				len += TCPOLEN_MPTCP_DSS_ACK64;
1300 				flags |= MPTCP_DSS_ACK64;
1301 			} else {
1302 				len += TCPOLEN_MPTCP_DSS_ACK32;
1303 			}
1304 		}
1305 
1306 		if (mpext->use_map) {
1307 			len += TCPOLEN_MPTCP_DSS_MAP64;
1308 
1309 			/* Use only 64-bit mapping flags for now, add
1310 			 * support for optional 32-bit mappings later.
1311 			 */
1312 			flags |= MPTCP_DSS_HAS_MAP | MPTCP_DSS_DSN64;
1313 			if (mpext->data_fin)
1314 				flags |= MPTCP_DSS_DATA_FIN;
1315 
1316 			if (opts->csum_reqd)
1317 				len += TCPOLEN_MPTCP_DSS_CHECKSUM;
1318 		}
1319 
1320 		*ptr++ = mptcp_option(MPTCPOPT_DSS, len, 0, flags);
1321 
1322 		if (mpext->use_ack) {
1323 			if (mpext->ack64) {
1324 				put_unaligned_be64(mpext->data_ack, ptr);
1325 				ptr += 2;
1326 			} else {
1327 				put_unaligned_be32(mpext->data_ack32, ptr);
1328 				ptr += 1;
1329 			}
1330 		}
1331 
1332 		if (mpext->use_map) {
1333 			put_unaligned_be64(mpext->data_seq, ptr);
1334 			ptr += 2;
1335 			put_unaligned_be32(mpext->subflow_seq, ptr);
1336 			ptr += 1;
1337 			if (opts->csum_reqd) {
1338 				put_unaligned_be32(mpext->data_len << 16 |
1339 						   mptcp_make_csum(mpext), ptr);
1340 			} else {
1341 				put_unaligned_be32(mpext->data_len << 16 |
1342 						   TCPOPT_NOP << 8 | TCPOPT_NOP, ptr);
1343 			}
1344 			ptr += 1;
1345 		}
1346 
1347 		/* We might need to add MP_FAIL options in rare cases */
1348 		if (unlikely(OPTION_MPTCP_FAIL & opts->suboptions))
1349 			goto mp_fail;
1350 	} else if (OPTIONS_MPTCP_MPC & opts->suboptions) {
1351 		u8 len, flag = MPTCP_CAP_HMAC_SHA256;
1352 
1353 		if (OPTION_MPTCP_MPC_SYN & opts->suboptions) {
1354 			len = TCPOLEN_MPTCP_MPC_SYN;
1355 		} else if (OPTION_MPTCP_MPC_SYNACK & opts->suboptions) {
1356 			len = TCPOLEN_MPTCP_MPC_SYNACK;
1357 		} else if (opts->data_len) {
1358 			len = TCPOLEN_MPTCP_MPC_ACK_DATA;
1359 			if (opts->csum_reqd)
1360 				len += TCPOLEN_MPTCP_DSS_CHECKSUM;
1361 		} else {
1362 			len = TCPOLEN_MPTCP_MPC_ACK;
1363 		}
1364 
1365 		if (opts->csum_reqd)
1366 			flag |= MPTCP_CAP_CHECKSUM_REQD;
1367 
1368 		if (!opts->allow_join_id0)
1369 			flag |= MPTCP_CAP_DENY_JOIN_ID0;
1370 
1371 		*ptr++ = mptcp_option(MPTCPOPT_MP_CAPABLE, len,
1372 				      MPTCP_SUPPORTED_VERSION,
1373 				      flag);
1374 
1375 		if (!((OPTION_MPTCP_MPC_SYNACK | OPTION_MPTCP_MPC_ACK) &
1376 		    opts->suboptions))
1377 			goto mp_capable_done;
1378 
1379 		put_unaligned_be64(opts->sndr_key, ptr);
1380 		ptr += 2;
1381 		if (!((OPTION_MPTCP_MPC_ACK) & opts->suboptions))
1382 			goto mp_capable_done;
1383 
1384 		put_unaligned_be64(opts->rcvr_key, ptr);
1385 		ptr += 2;
1386 		if (!opts->data_len)
1387 			goto mp_capable_done;
1388 
1389 		if (opts->csum_reqd) {
1390 			put_unaligned_be32(opts->data_len << 16 |
1391 					   __mptcp_make_csum(opts->data_seq,
1392 							     opts->subflow_seq,
1393 							     opts->data_len,
1394 							     ~csum_unfold(opts->csum)), ptr);
1395 		} else {
1396 			put_unaligned_be32(opts->data_len << 16 |
1397 					   TCPOPT_NOP << 8 | TCPOPT_NOP, ptr);
1398 		}
1399 		ptr += 1;
1400 
1401 		/* MPC is additionally mutually exclusive with MP_PRIO */
1402 		goto mp_capable_done;
1403 	} else if (OPTIONS_MPTCP_MPJ & opts->suboptions) {
1404 		if (OPTION_MPTCP_MPJ_SYN & opts->suboptions) {
1405 			*ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1406 					      TCPOLEN_MPTCP_MPJ_SYN,
1407 					      opts->backup, opts->join_id);
1408 			put_unaligned_be32(opts->token, ptr);
1409 			ptr += 1;
1410 			put_unaligned_be32(opts->nonce, ptr);
1411 			ptr += 1;
1412 		} else if (OPTION_MPTCP_MPJ_SYNACK & opts->suboptions) {
1413 			*ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1414 					      TCPOLEN_MPTCP_MPJ_SYNACK,
1415 					      opts->backup, opts->join_id);
1416 			put_unaligned_be64(opts->thmac, ptr);
1417 			ptr += 2;
1418 			put_unaligned_be32(opts->nonce, ptr);
1419 			ptr += 1;
1420 		} else {
1421 			*ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1422 					      TCPOLEN_MPTCP_MPJ_ACK, 0, 0);
1423 			memcpy(ptr, opts->hmac, MPTCPOPT_HMAC_LEN);
1424 			ptr += 5;
1425 		}
1426 	} else if (OPTION_MPTCP_ADD_ADDR & opts->suboptions) {
1427 		u8 len = TCPOLEN_MPTCP_ADD_ADDR_BASE;
1428 		u8 echo = MPTCP_ADDR_ECHO;
1429 
1430 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1431 		if (opts->addr.family == AF_INET6)
1432 			len = TCPOLEN_MPTCP_ADD_ADDR6_BASE;
1433 #endif
1434 
1435 		if (opts->addr.port)
1436 			len += TCPOLEN_MPTCP_PORT_LEN;
1437 
1438 		if (opts->ahmac) {
1439 			len += sizeof(opts->ahmac);
1440 			echo = 0;
1441 		}
1442 
1443 		*ptr++ = mptcp_option(MPTCPOPT_ADD_ADDR,
1444 				      len, echo, opts->addr.id);
1445 		if (opts->addr.family == AF_INET) {
1446 			memcpy((u8 *)ptr, (u8 *)&opts->addr.addr.s_addr, 4);
1447 			ptr += 1;
1448 		}
1449 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1450 		else if (opts->addr.family == AF_INET6) {
1451 			memcpy((u8 *)ptr, opts->addr.addr6.s6_addr, 16);
1452 			ptr += 4;
1453 		}
1454 #endif
1455 
1456 		if (!opts->addr.port) {
1457 			if (opts->ahmac) {
1458 				put_unaligned_be64(opts->ahmac, ptr);
1459 				ptr += 2;
1460 			}
1461 		} else {
1462 			u16 port = ntohs(opts->addr.port);
1463 
1464 			if (opts->ahmac) {
1465 				u8 *bptr = (u8 *)ptr;
1466 
1467 				put_unaligned_be16(port, bptr);
1468 				bptr += 2;
1469 				put_unaligned_be64(opts->ahmac, bptr);
1470 				bptr += 8;
1471 				put_unaligned_be16(TCPOPT_NOP << 8 |
1472 						   TCPOPT_NOP, bptr);
1473 
1474 				ptr += 3;
1475 			} else {
1476 				put_unaligned_be32(port << 16 |
1477 						   TCPOPT_NOP << 8 |
1478 						   TCPOPT_NOP, ptr);
1479 				ptr += 1;
1480 			}
1481 		}
1482 	} else if (unlikely(OPTION_MPTCP_FASTCLOSE & opts->suboptions)) {
1483 		/* FASTCLOSE is mutually exclusive with others except RST */
1484 		*ptr++ = mptcp_option(MPTCPOPT_MP_FASTCLOSE,
1485 				      TCPOLEN_MPTCP_FASTCLOSE,
1486 				      0, 0);
1487 		put_unaligned_be64(opts->rcvr_key, ptr);
1488 		ptr += 2;
1489 
1490 		if (OPTION_MPTCP_RST & opts->suboptions)
1491 			goto mp_rst;
1492 		return;
1493 	} else if (unlikely(OPTION_MPTCP_FAIL & opts->suboptions)) {
1494 mp_fail:
1495 		/* MP_FAIL is mutually exclusive with others except RST */
1496 		subflow = mptcp_subflow_ctx(ssk);
1497 		subflow->send_mp_fail = 0;
1498 
1499 		*ptr++ = mptcp_option(MPTCPOPT_MP_FAIL,
1500 				      TCPOLEN_MPTCP_FAIL,
1501 				      0, 0);
1502 		put_unaligned_be64(opts->fail_seq, ptr);
1503 		ptr += 2;
1504 
1505 		if (OPTION_MPTCP_RST & opts->suboptions)
1506 			goto mp_rst;
1507 		return;
1508 	} else if (unlikely(OPTION_MPTCP_RST & opts->suboptions)) {
1509 mp_rst:
1510 		*ptr++ = mptcp_option(MPTCPOPT_RST,
1511 				      TCPOLEN_MPTCP_RST,
1512 				      opts->reset_transient,
1513 				      opts->reset_reason);
1514 		return;
1515 	}
1516 
1517 	if (OPTION_MPTCP_PRIO & opts->suboptions) {
1518 		subflow = mptcp_subflow_ctx(ssk);
1519 		subflow->send_mp_prio = 0;
1520 
1521 		*ptr++ = mptcp_option(MPTCPOPT_MP_PRIO,
1522 				      TCPOLEN_MPTCP_PRIO,
1523 				      opts->backup, TCPOPT_NOP);
1524 	}
1525 
1526 mp_capable_done:
1527 	if (OPTION_MPTCP_RM_ADDR & opts->suboptions) {
1528 		u8 i = 1;
1529 
1530 		*ptr++ = mptcp_option(MPTCPOPT_RM_ADDR,
1531 				      TCPOLEN_MPTCP_RM_ADDR_BASE + opts->rm_list.nr,
1532 				      0, opts->rm_list.ids[0]);
1533 
1534 		while (i < opts->rm_list.nr) {
1535 			u8 id1, id2, id3, id4;
1536 
1537 			id1 = opts->rm_list.ids[i];
1538 			id2 = i + 1 < opts->rm_list.nr ? opts->rm_list.ids[i + 1] : TCPOPT_NOP;
1539 			id3 = i + 2 < opts->rm_list.nr ? opts->rm_list.ids[i + 2] : TCPOPT_NOP;
1540 			id4 = i + 3 < opts->rm_list.nr ? opts->rm_list.ids[i + 3] : TCPOPT_NOP;
1541 			put_unaligned_be32(id1 << 24 | id2 << 16 | id3 << 8 | id4, ptr);
1542 			ptr += 1;
1543 			i += 4;
1544 		}
1545 	}
1546 
1547 	if (tp)
1548 		mptcp_set_rwin(tp);
1549 }
1550 
1551 __be32 mptcp_get_reset_option(const struct sk_buff *skb)
1552 {
1553 	const struct mptcp_ext *ext = mptcp_get_ext(skb);
1554 	u8 flags, reason;
1555 
1556 	if (ext) {
1557 		flags = ext->reset_transient;
1558 		reason = ext->reset_reason;
1559 
1560 		return mptcp_option(MPTCPOPT_RST, TCPOLEN_MPTCP_RST,
1561 				    flags, reason);
1562 	}
1563 
1564 	return htonl(0u);
1565 }
1566 EXPORT_SYMBOL_GPL(mptcp_get_reset_option);
1567