xref: /openbmc/linux/net/mptcp/options.c (revision f8a11425075ff11b4b5784f077cb84f3d2dfb3f0)
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->csum_reqd = 1;
85 
86 		mp_opt->mp_capable = 1;
87 		if (opsize >= TCPOLEN_MPTCP_MPC_SYNACK) {
88 			mp_opt->sndr_key = get_unaligned_be64(ptr);
89 			ptr += 8;
90 		}
91 		if (opsize >= TCPOLEN_MPTCP_MPC_ACK) {
92 			mp_opt->rcvr_key = get_unaligned_be64(ptr);
93 			ptr += 8;
94 		}
95 		if (opsize >= TCPOLEN_MPTCP_MPC_ACK_DATA) {
96 			/* Section 3.1.:
97 			 * "the data parameters in a MP_CAPABLE are semantically
98 			 * equivalent to those in a DSS option and can be used
99 			 * interchangeably."
100 			 */
101 			mp_opt->dss = 1;
102 			mp_opt->use_map = 1;
103 			mp_opt->mpc_map = 1;
104 			mp_opt->data_len = get_unaligned_be16(ptr);
105 			ptr += 2;
106 		}
107 		if (opsize == TCPOLEN_MPTCP_MPC_ACK_DATA_CSUM) {
108 			mp_opt->csum = (__force __sum16)get_unaligned_be16(ptr);
109 			mp_opt->csum_reqd = 1;
110 			ptr += 2;
111 		}
112 		pr_debug("MP_CAPABLE version=%x, flags=%x, optlen=%d sndr=%llu, rcvr=%llu len=%d csum=%u",
113 			 version, flags, opsize, mp_opt->sndr_key,
114 			 mp_opt->rcvr_key, mp_opt->data_len, mp_opt->csum);
115 		break;
116 
117 	case MPTCPOPT_MP_JOIN:
118 		mp_opt->mp_join = 1;
119 		if (opsize == TCPOLEN_MPTCP_MPJ_SYN) {
120 			mp_opt->backup = *ptr++ & MPTCPOPT_BACKUP;
121 			mp_opt->join_id = *ptr++;
122 			mp_opt->token = get_unaligned_be32(ptr);
123 			ptr += 4;
124 			mp_opt->nonce = get_unaligned_be32(ptr);
125 			ptr += 4;
126 			pr_debug("MP_JOIN bkup=%u, id=%u, token=%u, nonce=%u",
127 				 mp_opt->backup, mp_opt->join_id,
128 				 mp_opt->token, mp_opt->nonce);
129 		} else if (opsize == TCPOLEN_MPTCP_MPJ_SYNACK) {
130 			mp_opt->backup = *ptr++ & MPTCPOPT_BACKUP;
131 			mp_opt->join_id = *ptr++;
132 			mp_opt->thmac = get_unaligned_be64(ptr);
133 			ptr += 8;
134 			mp_opt->nonce = get_unaligned_be32(ptr);
135 			ptr += 4;
136 			pr_debug("MP_JOIN bkup=%u, id=%u, thmac=%llu, nonce=%u",
137 				 mp_opt->backup, mp_opt->join_id,
138 				 mp_opt->thmac, mp_opt->nonce);
139 		} else if (opsize == TCPOLEN_MPTCP_MPJ_ACK) {
140 			ptr += 2;
141 			memcpy(mp_opt->hmac, ptr, MPTCPOPT_HMAC_LEN);
142 			pr_debug("MP_JOIN hmac");
143 		} else {
144 			mp_opt->mp_join = 0;
145 		}
146 		break;
147 
148 	case MPTCPOPT_DSS:
149 		pr_debug("DSS");
150 		ptr++;
151 
152 		/* we must clear 'mpc_map' be able to detect MP_CAPABLE
153 		 * map vs DSS map in mptcp_incoming_options(), and reconstruct
154 		 * map info accordingly
155 		 */
156 		mp_opt->mpc_map = 0;
157 		flags = (*ptr++) & MPTCP_DSS_FLAG_MASK;
158 		mp_opt->data_fin = (flags & MPTCP_DSS_DATA_FIN) != 0;
159 		mp_opt->dsn64 = (flags & MPTCP_DSS_DSN64) != 0;
160 		mp_opt->use_map = (flags & MPTCP_DSS_HAS_MAP) != 0;
161 		mp_opt->ack64 = (flags & MPTCP_DSS_ACK64) != 0;
162 		mp_opt->use_ack = (flags & MPTCP_DSS_HAS_ACK);
163 
164 		pr_debug("data_fin=%d dsn64=%d use_map=%d ack64=%d use_ack=%d",
165 			 mp_opt->data_fin, mp_opt->dsn64,
166 			 mp_opt->use_map, mp_opt->ack64,
167 			 mp_opt->use_ack);
168 
169 		expected_opsize = TCPOLEN_MPTCP_DSS_BASE;
170 
171 		if (mp_opt->use_ack) {
172 			if (mp_opt->ack64)
173 				expected_opsize += TCPOLEN_MPTCP_DSS_ACK64;
174 			else
175 				expected_opsize += TCPOLEN_MPTCP_DSS_ACK32;
176 		}
177 
178 		if (mp_opt->use_map) {
179 			if (mp_opt->dsn64)
180 				expected_opsize += TCPOLEN_MPTCP_DSS_MAP64;
181 			else
182 				expected_opsize += TCPOLEN_MPTCP_DSS_MAP32;
183 		}
184 
185 		/* Always parse any csum presence combination, we will enforce
186 		 * RFC 8684 Section 3.3.0 checks later in subflow_data_ready
187 		 */
188 		if (opsize != expected_opsize &&
189 		    opsize != expected_opsize + TCPOLEN_MPTCP_DSS_CHECKSUM)
190 			break;
191 
192 		mp_opt->dss = 1;
193 
194 		if (mp_opt->use_ack) {
195 			if (mp_opt->ack64) {
196 				mp_opt->data_ack = get_unaligned_be64(ptr);
197 				ptr += 8;
198 			} else {
199 				mp_opt->data_ack = get_unaligned_be32(ptr);
200 				ptr += 4;
201 			}
202 
203 			pr_debug("data_ack=%llu", mp_opt->data_ack);
204 		}
205 
206 		if (mp_opt->use_map) {
207 			if (mp_opt->dsn64) {
208 				mp_opt->data_seq = get_unaligned_be64(ptr);
209 				ptr += 8;
210 			} else {
211 				mp_opt->data_seq = get_unaligned_be32(ptr);
212 				ptr += 4;
213 			}
214 
215 			mp_opt->subflow_seq = get_unaligned_be32(ptr);
216 			ptr += 4;
217 
218 			mp_opt->data_len = get_unaligned_be16(ptr);
219 			ptr += 2;
220 
221 			if (opsize == expected_opsize + TCPOLEN_MPTCP_DSS_CHECKSUM) {
222 				mp_opt->csum_reqd = 1;
223 				mp_opt->csum = (__force __sum16)get_unaligned_be16(ptr);
224 				ptr += 2;
225 			}
226 
227 			pr_debug("data_seq=%llu subflow_seq=%u data_len=%u csum=%d:%u",
228 				 mp_opt->data_seq, mp_opt->subflow_seq,
229 				 mp_opt->data_len, mp_opt->csum_reqd, mp_opt->csum);
230 		}
231 
232 		break;
233 
234 	case MPTCPOPT_ADD_ADDR:
235 		mp_opt->echo = (*ptr++) & MPTCP_ADDR_ECHO;
236 		if (!mp_opt->echo) {
237 			if (opsize == TCPOLEN_MPTCP_ADD_ADDR ||
238 			    opsize == TCPOLEN_MPTCP_ADD_ADDR_PORT)
239 				mp_opt->addr.family = AF_INET;
240 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
241 			else if (opsize == TCPOLEN_MPTCP_ADD_ADDR6 ||
242 				 opsize == TCPOLEN_MPTCP_ADD_ADDR6_PORT)
243 				mp_opt->addr.family = AF_INET6;
244 #endif
245 			else
246 				break;
247 		} else {
248 			if (opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE ||
249 			    opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE_PORT)
250 				mp_opt->addr.family = AF_INET;
251 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
252 			else if (opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE ||
253 				 opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE_PORT)
254 				mp_opt->addr.family = AF_INET6;
255 #endif
256 			else
257 				break;
258 		}
259 
260 		mp_opt->add_addr = 1;
261 		mp_opt->addr.id = *ptr++;
262 		if (mp_opt->addr.family == AF_INET) {
263 			memcpy((u8 *)&mp_opt->addr.addr.s_addr, (u8 *)ptr, 4);
264 			ptr += 4;
265 			if (opsize == TCPOLEN_MPTCP_ADD_ADDR_PORT ||
266 			    opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE_PORT) {
267 				mp_opt->addr.port = htons(get_unaligned_be16(ptr));
268 				ptr += 2;
269 			}
270 		}
271 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
272 		else {
273 			memcpy(mp_opt->addr.addr6.s6_addr, (u8 *)ptr, 16);
274 			ptr += 16;
275 			if (opsize == TCPOLEN_MPTCP_ADD_ADDR6_PORT ||
276 			    opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE_PORT) {
277 				mp_opt->addr.port = htons(get_unaligned_be16(ptr));
278 				ptr += 2;
279 			}
280 		}
281 #endif
282 		if (!mp_opt->echo) {
283 			mp_opt->ahmac = get_unaligned_be64(ptr);
284 			ptr += 8;
285 		}
286 		pr_debug("ADD_ADDR%s: id=%d, ahmac=%llu, echo=%d, port=%d",
287 			 (mp_opt->addr.family == AF_INET6) ? "6" : "",
288 			 mp_opt->addr.id, mp_opt->ahmac, mp_opt->echo, ntohs(mp_opt->addr.port));
289 		break;
290 
291 	case MPTCPOPT_RM_ADDR:
292 		if (opsize < TCPOLEN_MPTCP_RM_ADDR_BASE + 1 ||
293 		    opsize > TCPOLEN_MPTCP_RM_ADDR_BASE + MPTCP_RM_IDS_MAX)
294 			break;
295 
296 		ptr++;
297 
298 		mp_opt->rm_addr = 1;
299 		mp_opt->rm_list.nr = opsize - TCPOLEN_MPTCP_RM_ADDR_BASE;
300 		for (i = 0; i < mp_opt->rm_list.nr; i++)
301 			mp_opt->rm_list.ids[i] = *ptr++;
302 		pr_debug("RM_ADDR: rm_list_nr=%d", mp_opt->rm_list.nr);
303 		break;
304 
305 	case MPTCPOPT_MP_PRIO:
306 		if (opsize != TCPOLEN_MPTCP_PRIO)
307 			break;
308 
309 		mp_opt->mp_prio = 1;
310 		mp_opt->backup = *ptr++ & MPTCP_PRIO_BKUP;
311 		pr_debug("MP_PRIO: prio=%d", mp_opt->backup);
312 		break;
313 
314 	case MPTCPOPT_MP_FASTCLOSE:
315 		if (opsize != TCPOLEN_MPTCP_FASTCLOSE)
316 			break;
317 
318 		ptr += 2;
319 		mp_opt->rcvr_key = get_unaligned_be64(ptr);
320 		ptr += 8;
321 		mp_opt->fastclose = 1;
322 		break;
323 
324 	case MPTCPOPT_RST:
325 		if (opsize != TCPOLEN_MPTCP_RST)
326 			break;
327 
328 		if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_RST))
329 			break;
330 		mp_opt->reset = 1;
331 		flags = *ptr++;
332 		mp_opt->reset_transient = flags & MPTCP_RST_TRANSIENT;
333 		mp_opt->reset_reason = *ptr;
334 		break;
335 
336 	default:
337 		break;
338 	}
339 }
340 
341 void mptcp_get_options(const struct sock *sk,
342 		       const struct sk_buff *skb,
343 		       struct mptcp_options_received *mp_opt)
344 {
345 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
346 	struct mptcp_sock *msk = mptcp_sk(subflow->conn);
347 	const struct tcphdr *th = tcp_hdr(skb);
348 	const unsigned char *ptr;
349 	int length;
350 
351 	/* initialize option status */
352 	mp_opt->mp_capable = 0;
353 	mp_opt->mp_join = 0;
354 	mp_opt->add_addr = 0;
355 	mp_opt->ahmac = 0;
356 	mp_opt->fastclose = 0;
357 	mp_opt->addr.port = 0;
358 	mp_opt->rm_addr = 0;
359 	mp_opt->dss = 0;
360 	mp_opt->mp_prio = 0;
361 	mp_opt->reset = 0;
362 	mp_opt->csum_reqd = READ_ONCE(msk->csum_enabled);
363 
364 	length = (th->doff * 4) - sizeof(struct tcphdr);
365 	ptr = (const unsigned char *)(th + 1);
366 
367 	while (length > 0) {
368 		int opcode = *ptr++;
369 		int opsize;
370 
371 		switch (opcode) {
372 		case TCPOPT_EOL:
373 			return;
374 		case TCPOPT_NOP:	/* Ref: RFC 793 section 3.1 */
375 			length--;
376 			continue;
377 		default:
378 			opsize = *ptr++;
379 			if (opsize < 2) /* "silly options" */
380 				return;
381 			if (opsize > length)
382 				return;	/* don't parse partial options */
383 			if (opcode == TCPOPT_MPTCP)
384 				mptcp_parse_option(skb, ptr, opsize, mp_opt);
385 			ptr += opsize - 2;
386 			length -= opsize;
387 		}
388 	}
389 }
390 
391 bool mptcp_syn_options(struct sock *sk, const struct sk_buff *skb,
392 		       unsigned int *size, struct mptcp_out_options *opts)
393 {
394 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
395 
396 	/* we will use snd_isn to detect first pkt [re]transmission
397 	 * in mptcp_established_options_mp()
398 	 */
399 	subflow->snd_isn = TCP_SKB_CB(skb)->end_seq;
400 	if (subflow->request_mptcp) {
401 		opts->suboptions = OPTION_MPTCP_MPC_SYN;
402 		opts->csum_reqd = mptcp_is_checksum_enabled(sock_net(sk));
403 		*size = TCPOLEN_MPTCP_MPC_SYN;
404 		return true;
405 	} else if (subflow->request_join) {
406 		pr_debug("remote_token=%u, nonce=%u", subflow->remote_token,
407 			 subflow->local_nonce);
408 		opts->suboptions = OPTION_MPTCP_MPJ_SYN;
409 		opts->join_id = subflow->local_id;
410 		opts->token = subflow->remote_token;
411 		opts->nonce = subflow->local_nonce;
412 		opts->backup = subflow->request_bkup;
413 		*size = TCPOLEN_MPTCP_MPJ_SYN;
414 		return true;
415 	}
416 	return false;
417 }
418 
419 /* MP_JOIN client subflow must wait for 4th ack before sending any data:
420  * TCP can't schedule delack timer before the subflow is fully established.
421  * MPTCP uses the delack timer to do 3rd ack retransmissions
422  */
423 static void schedule_3rdack_retransmission(struct sock *sk)
424 {
425 	struct inet_connection_sock *icsk = inet_csk(sk);
426 	struct tcp_sock *tp = tcp_sk(sk);
427 	unsigned long timeout;
428 
429 	/* reschedule with a timeout above RTT, as we must look only for drop */
430 	if (tp->srtt_us)
431 		timeout = tp->srtt_us << 1;
432 	else
433 		timeout = TCP_TIMEOUT_INIT;
434 
435 	WARN_ON_ONCE(icsk->icsk_ack.pending & ICSK_ACK_TIMER);
436 	icsk->icsk_ack.pending |= ICSK_ACK_SCHED | ICSK_ACK_TIMER;
437 	icsk->icsk_ack.timeout = timeout;
438 	sk_reset_timer(sk, &icsk->icsk_delack_timer, timeout);
439 }
440 
441 static void clear_3rdack_retransmission(struct sock *sk)
442 {
443 	struct inet_connection_sock *icsk = inet_csk(sk);
444 
445 	sk_stop_timer(sk, &icsk->icsk_delack_timer);
446 	icsk->icsk_ack.timeout = 0;
447 	icsk->icsk_ack.ato = 0;
448 	icsk->icsk_ack.pending &= ~(ICSK_ACK_SCHED | ICSK_ACK_TIMER);
449 }
450 
451 static bool mptcp_established_options_mp(struct sock *sk, struct sk_buff *skb,
452 					 bool snd_data_fin_enable,
453 					 unsigned int *size,
454 					 unsigned int remaining,
455 					 struct mptcp_out_options *opts)
456 {
457 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
458 	struct mptcp_sock *msk = mptcp_sk(subflow->conn);
459 	struct mptcp_ext *mpext;
460 	unsigned int data_len;
461 	u8 len;
462 
463 	/* When skb is not available, we better over-estimate the emitted
464 	 * options len. A full DSS option (28 bytes) is longer than
465 	 * TCPOLEN_MPTCP_MPC_ACK_DATA(22) or TCPOLEN_MPTCP_MPJ_ACK(24), so
466 	 * tell the caller to defer the estimate to
467 	 * mptcp_established_options_dss(), which will reserve enough space.
468 	 */
469 	if (!skb)
470 		return false;
471 
472 	/* MPC/MPJ needed only on 3rd ack packet, DATA_FIN and TCP shutdown take precedence */
473 	if (subflow->fully_established || snd_data_fin_enable ||
474 	    subflow->snd_isn != TCP_SKB_CB(skb)->seq ||
475 	    sk->sk_state != TCP_ESTABLISHED)
476 		return false;
477 
478 	if (subflow->mp_capable) {
479 		mpext = mptcp_get_ext(skb);
480 		data_len = mpext ? mpext->data_len : 0;
481 
482 		/* we will check ext_copy.data_len in mptcp_write_options() to
483 		 * discriminate between TCPOLEN_MPTCP_MPC_ACK_DATA and
484 		 * TCPOLEN_MPTCP_MPC_ACK
485 		 */
486 		opts->ext_copy.data_len = data_len;
487 		opts->suboptions = OPTION_MPTCP_MPC_ACK;
488 		opts->sndr_key = subflow->local_key;
489 		opts->rcvr_key = subflow->remote_key;
490 		opts->csum_reqd = READ_ONCE(msk->csum_enabled);
491 
492 		/* Section 3.1.
493 		 * The MP_CAPABLE option is carried on the SYN, SYN/ACK, and ACK
494 		 * packets that start the first subflow of an MPTCP connection,
495 		 * as well as the first packet that carries data
496 		 */
497 		if (data_len > 0) {
498 			len = TCPOLEN_MPTCP_MPC_ACK_DATA;
499 			if (opts->csum_reqd) {
500 				/* we need to propagate more info to csum the pseudo hdr */
501 				opts->ext_copy.data_seq = mpext->data_seq;
502 				opts->ext_copy.subflow_seq = mpext->subflow_seq;
503 				opts->ext_copy.csum = mpext->csum;
504 				len += TCPOLEN_MPTCP_DSS_CHECKSUM;
505 			}
506 			*size = ALIGN(len, 4);
507 		} else {
508 			*size = TCPOLEN_MPTCP_MPC_ACK;
509 		}
510 
511 		pr_debug("subflow=%p, local_key=%llu, remote_key=%llu map_len=%d",
512 			 subflow, subflow->local_key, subflow->remote_key,
513 			 data_len);
514 
515 		return true;
516 	} else if (subflow->mp_join) {
517 		opts->suboptions = OPTION_MPTCP_MPJ_ACK;
518 		memcpy(opts->hmac, subflow->hmac, MPTCPOPT_HMAC_LEN);
519 		*size = TCPOLEN_MPTCP_MPJ_ACK;
520 		pr_debug("subflow=%p", subflow);
521 
522 		schedule_3rdack_retransmission(sk);
523 		return true;
524 	}
525 	return false;
526 }
527 
528 static void mptcp_write_data_fin(struct mptcp_subflow_context *subflow,
529 				 struct sk_buff *skb, struct mptcp_ext *ext)
530 {
531 	/* The write_seq value has already been incremented, so the actual
532 	 * sequence number for the DATA_FIN is one less.
533 	 */
534 	u64 data_fin_tx_seq = READ_ONCE(mptcp_sk(subflow->conn)->write_seq) - 1;
535 
536 	if (!ext->use_map || !skb->len) {
537 		/* RFC6824 requires a DSS mapping with specific values
538 		 * if DATA_FIN is set but no data payload is mapped
539 		 */
540 		ext->data_fin = 1;
541 		ext->use_map = 1;
542 		ext->dsn64 = 1;
543 		ext->data_seq = data_fin_tx_seq;
544 		ext->subflow_seq = 0;
545 		ext->data_len = 1;
546 	} else if (ext->data_seq + ext->data_len == data_fin_tx_seq) {
547 		/* If there's an existing DSS mapping and it is the
548 		 * final mapping, DATA_FIN consumes 1 additional byte of
549 		 * mapping space.
550 		 */
551 		ext->data_fin = 1;
552 		ext->data_len++;
553 	}
554 }
555 
556 static bool mptcp_established_options_dss(struct sock *sk, struct sk_buff *skb,
557 					  bool snd_data_fin_enable,
558 					  unsigned int *size,
559 					  unsigned int remaining,
560 					  struct mptcp_out_options *opts)
561 {
562 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
563 	struct mptcp_sock *msk = mptcp_sk(subflow->conn);
564 	unsigned int dss_size = 0;
565 	struct mptcp_ext *mpext;
566 	unsigned int ack_size;
567 	bool ret = false;
568 	u64 ack_seq;
569 
570 	opts->csum_reqd = READ_ONCE(msk->csum_enabled);
571 	mpext = skb ? mptcp_get_ext(skb) : NULL;
572 
573 	if (!skb || (mpext && mpext->use_map) || snd_data_fin_enable) {
574 		unsigned int map_size = TCPOLEN_MPTCP_DSS_BASE + TCPOLEN_MPTCP_DSS_MAP64;
575 
576 		if (mpext) {
577 			if (opts->csum_reqd)
578 				map_size += TCPOLEN_MPTCP_DSS_CHECKSUM;
579 
580 			opts->ext_copy = *mpext;
581 		}
582 
583 		remaining -= map_size;
584 		dss_size = map_size;
585 		if (skb && snd_data_fin_enable)
586 			mptcp_write_data_fin(subflow, skb, &opts->ext_copy);
587 		ret = true;
588 	}
589 
590 	/* passive sockets msk will set the 'can_ack' after accept(), even
591 	 * if the first subflow may have the already the remote key handy
592 	 */
593 	opts->ext_copy.use_ack = 0;
594 	if (!READ_ONCE(msk->can_ack)) {
595 		*size = ALIGN(dss_size, 4);
596 		return ret;
597 	}
598 
599 	ack_seq = READ_ONCE(msk->ack_seq);
600 	if (READ_ONCE(msk->use_64bit_ack)) {
601 		ack_size = TCPOLEN_MPTCP_DSS_ACK64;
602 		opts->ext_copy.data_ack = ack_seq;
603 		opts->ext_copy.ack64 = 1;
604 	} else {
605 		ack_size = TCPOLEN_MPTCP_DSS_ACK32;
606 		opts->ext_copy.data_ack32 = (uint32_t)ack_seq;
607 		opts->ext_copy.ack64 = 0;
608 	}
609 	opts->ext_copy.use_ack = 1;
610 	WRITE_ONCE(msk->old_wspace, __mptcp_space((struct sock *)msk));
611 
612 	/* Add kind/length/subtype/flag overhead if mapping is not populated */
613 	if (dss_size == 0)
614 		ack_size += TCPOLEN_MPTCP_DSS_BASE;
615 
616 	dss_size += ack_size;
617 
618 	*size = ALIGN(dss_size, 4);
619 	return true;
620 }
621 
622 static u64 add_addr_generate_hmac(u64 key1, u64 key2,
623 				  struct mptcp_addr_info *addr)
624 {
625 	u16 port = ntohs(addr->port);
626 	u8 hmac[SHA256_DIGEST_SIZE];
627 	u8 msg[19];
628 	int i = 0;
629 
630 	msg[i++] = addr->id;
631 	if (addr->family == AF_INET) {
632 		memcpy(&msg[i], &addr->addr.s_addr, 4);
633 		i += 4;
634 	}
635 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
636 	else if (addr->family == AF_INET6) {
637 		memcpy(&msg[i], &addr->addr6.s6_addr, 16);
638 		i += 16;
639 	}
640 #endif
641 	msg[i++] = port >> 8;
642 	msg[i++] = port & 0xFF;
643 
644 	mptcp_crypto_hmac_sha(key1, key2, msg, i, hmac);
645 
646 	return get_unaligned_be64(&hmac[SHA256_DIGEST_SIZE - sizeof(u64)]);
647 }
648 
649 static bool mptcp_established_options_add_addr(struct sock *sk, struct sk_buff *skb,
650 					       unsigned int *size,
651 					       unsigned int remaining,
652 					       struct mptcp_out_options *opts)
653 {
654 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
655 	struct mptcp_sock *msk = mptcp_sk(subflow->conn);
656 	bool drop_other_suboptions = false;
657 	unsigned int opt_size = *size;
658 	bool echo;
659 	bool port;
660 	int len;
661 
662 	if ((mptcp_pm_should_add_signal_ipv6(msk) ||
663 	     mptcp_pm_should_add_signal_port(msk) ||
664 	     mptcp_pm_should_add_signal_echo(msk)) &&
665 	    skb && skb_is_tcp_pure_ack(skb)) {
666 		pr_debug("drop other suboptions");
667 		opts->suboptions = 0;
668 		opts->ext_copy.use_ack = 0;
669 		opts->ext_copy.use_map = 0;
670 		remaining += opt_size;
671 		drop_other_suboptions = true;
672 	}
673 
674 	if (!mptcp_pm_should_add_signal(msk) ||
675 	    !(mptcp_pm_add_addr_signal(msk, remaining, &opts->addr, &echo, &port)))
676 		return false;
677 
678 	len = mptcp_add_addr_len(opts->addr.family, echo, port);
679 	if (remaining < len)
680 		return false;
681 
682 	*size = len;
683 	if (drop_other_suboptions)
684 		*size -= opt_size;
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 	if (!subflow->send_mp_prio)
735 		return false;
736 
737 	/* account for the trailing 'nop' option */
738 	if (remaining < TCPOLEN_MPTCP_PRIO_ALIGN)
739 		return false;
740 
741 	*size = TCPOLEN_MPTCP_PRIO_ALIGN;
742 	opts->suboptions |= OPTION_MPTCP_PRIO;
743 	opts->backup = subflow->request_bkup;
744 
745 	pr_debug("prio=%d", opts->backup);
746 
747 	return true;
748 }
749 
750 static noinline void mptcp_established_options_rst(struct sock *sk, struct sk_buff *skb,
751 						   unsigned int *size,
752 						   unsigned int remaining,
753 						   struct mptcp_out_options *opts)
754 {
755 	const struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
756 
757 	if (remaining < TCPOLEN_MPTCP_RST)
758 		return;
759 
760 	*size = TCPOLEN_MPTCP_RST;
761 	opts->suboptions |= OPTION_MPTCP_RST;
762 	opts->reset_transient = subflow->reset_transient;
763 	opts->reset_reason = subflow->reset_reason;
764 }
765 
766 bool mptcp_established_options(struct sock *sk, struct sk_buff *skb,
767 			       unsigned int *size, unsigned int remaining,
768 			       struct mptcp_out_options *opts)
769 {
770 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
771 	struct mptcp_sock *msk = mptcp_sk(subflow->conn);
772 	unsigned int opt_size = 0;
773 	bool snd_data_fin;
774 	bool ret = false;
775 
776 	opts->suboptions = 0;
777 
778 	if (unlikely(__mptcp_check_fallback(msk)))
779 		return false;
780 
781 	if (unlikely(skb && TCP_SKB_CB(skb)->tcp_flags & TCPHDR_RST)) {
782 		mptcp_established_options_rst(sk, skb, size, remaining, opts);
783 		return true;
784 	}
785 
786 	snd_data_fin = mptcp_data_fin_enabled(msk);
787 	if (mptcp_established_options_mp(sk, skb, snd_data_fin, &opt_size, remaining, opts))
788 		ret = true;
789 	else if (mptcp_established_options_dss(sk, skb, snd_data_fin, &opt_size, remaining, opts))
790 		ret = true;
791 
792 	/* we reserved enough space for the above options, and exceeding the
793 	 * TCP option space would be fatal
794 	 */
795 	if (WARN_ON_ONCE(opt_size > remaining))
796 		return false;
797 
798 	*size += opt_size;
799 	remaining -= opt_size;
800 	if (mptcp_established_options_add_addr(sk, skb, &opt_size, remaining, opts)) {
801 		*size += opt_size;
802 		remaining -= opt_size;
803 		ret = true;
804 	} else if (mptcp_established_options_rm_addr(sk, &opt_size, remaining, opts)) {
805 		*size += opt_size;
806 		remaining -= opt_size;
807 		ret = true;
808 	}
809 
810 	if (mptcp_established_options_mp_prio(sk, &opt_size, remaining, opts)) {
811 		*size += opt_size;
812 		remaining -= opt_size;
813 		ret = true;
814 	}
815 
816 	return ret;
817 }
818 
819 bool mptcp_synack_options(const struct request_sock *req, unsigned int *size,
820 			  struct mptcp_out_options *opts)
821 {
822 	struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
823 
824 	if (subflow_req->mp_capable) {
825 		opts->suboptions = OPTION_MPTCP_MPC_SYNACK;
826 		opts->sndr_key = subflow_req->local_key;
827 		opts->csum_reqd = subflow_req->csum_reqd;
828 		*size = TCPOLEN_MPTCP_MPC_SYNACK;
829 		pr_debug("subflow_req=%p, local_key=%llu",
830 			 subflow_req, subflow_req->local_key);
831 		return true;
832 	} else if (subflow_req->mp_join) {
833 		opts->suboptions = OPTION_MPTCP_MPJ_SYNACK;
834 		opts->backup = subflow_req->backup;
835 		opts->join_id = subflow_req->local_id;
836 		opts->thmac = subflow_req->thmac;
837 		opts->nonce = subflow_req->local_nonce;
838 		pr_debug("req=%p, bkup=%u, id=%u, thmac=%llu, nonce=%u",
839 			 subflow_req, opts->backup, opts->join_id,
840 			 opts->thmac, opts->nonce);
841 		*size = TCPOLEN_MPTCP_MPJ_SYNACK;
842 		return true;
843 	}
844 	return false;
845 }
846 
847 static bool check_fully_established(struct mptcp_sock *msk, struct sock *ssk,
848 				    struct mptcp_subflow_context *subflow,
849 				    struct sk_buff *skb,
850 				    struct mptcp_options_received *mp_opt)
851 {
852 	/* here we can process OoO, in-window pkts, only in-sequence 4th ack
853 	 * will make the subflow fully established
854 	 */
855 	if (likely(subflow->fully_established)) {
856 		/* on passive sockets, check for 3rd ack retransmission
857 		 * note that msk is always set by subflow_syn_recv_sock()
858 		 * for mp_join subflows
859 		 */
860 		if (TCP_SKB_CB(skb)->seq == subflow->ssn_offset + 1 &&
861 		    TCP_SKB_CB(skb)->end_seq == TCP_SKB_CB(skb)->seq &&
862 		    subflow->mp_join && mp_opt->mp_join &&
863 		    READ_ONCE(msk->pm.server_side))
864 			tcp_send_ack(ssk);
865 		goto fully_established;
866 	}
867 
868 	/* we must process OoO packets before the first subflow is fully
869 	 * established. OoO packets are instead a protocol violation
870 	 * for MP_JOIN subflows as the peer must not send any data
871 	 * before receiving the forth ack - cfr. RFC 8684 section 3.2.
872 	 */
873 	if (TCP_SKB_CB(skb)->seq != subflow->ssn_offset + 1) {
874 		if (subflow->mp_join)
875 			goto reset;
876 		return subflow->mp_capable;
877 	}
878 
879 	if (mp_opt->dss && mp_opt->use_ack) {
880 		/* subflows are fully established as soon as we get any
881 		 * additional ack.
882 		 */
883 		subflow->fully_established = 1;
884 		WRITE_ONCE(msk->fully_established, true);
885 		goto fully_established;
886 	}
887 
888 	if (mp_opt->add_addr) {
889 		WRITE_ONCE(msk->fully_established, true);
890 		return true;
891 	}
892 
893 	/* If the first established packet does not contain MP_CAPABLE + data
894 	 * then fallback to TCP. Fallback scenarios requires a reset for
895 	 * MP_JOIN subflows.
896 	 */
897 	if (!mp_opt->mp_capable) {
898 		if (subflow->mp_join)
899 			goto reset;
900 		subflow->mp_capable = 0;
901 		pr_fallback(msk);
902 		__mptcp_do_fallback(msk);
903 		return false;
904 	}
905 
906 	if (unlikely(!READ_ONCE(msk->pm.server_side)))
907 		pr_warn_once("bogus mpc option on established client sk");
908 	mptcp_subflow_fully_established(subflow, mp_opt);
909 
910 fully_established:
911 	/* if the subflow is not already linked into the conn_list, we can't
912 	 * notify the PM: this subflow is still on the listener queue
913 	 * and the PM possibly acquiring the subflow lock could race with
914 	 * the listener close
915 	 */
916 	if (likely(subflow->pm_notified) || list_empty(&subflow->node))
917 		return true;
918 
919 	subflow->pm_notified = 1;
920 	if (subflow->mp_join) {
921 		clear_3rdack_retransmission(ssk);
922 		mptcp_pm_subflow_established(msk);
923 	} else {
924 		mptcp_pm_fully_established(msk, ssk, GFP_ATOMIC);
925 	}
926 	return true;
927 
928 reset:
929 	mptcp_subflow_reset(ssk);
930 	return false;
931 }
932 
933 static u64 expand_ack(u64 old_ack, u64 cur_ack, bool use_64bit)
934 {
935 	u32 old_ack32, cur_ack32;
936 
937 	if (use_64bit)
938 		return cur_ack;
939 
940 	old_ack32 = (u32)old_ack;
941 	cur_ack32 = (u32)cur_ack;
942 	cur_ack = (old_ack & GENMASK_ULL(63, 32)) + cur_ack32;
943 	if (unlikely(before(cur_ack32, old_ack32)))
944 		return cur_ack + (1LL << 32);
945 	return cur_ack;
946 }
947 
948 static void ack_update_msk(struct mptcp_sock *msk,
949 			   struct sock *ssk,
950 			   struct mptcp_options_received *mp_opt)
951 {
952 	u64 new_wnd_end, new_snd_una, snd_nxt = READ_ONCE(msk->snd_nxt);
953 	struct sock *sk = (struct sock *)msk;
954 	u64 old_snd_una;
955 
956 	mptcp_data_lock(sk);
957 
958 	/* avoid ack expansion on update conflict, to reduce the risk of
959 	 * wrongly expanding to a future ack sequence number, which is way
960 	 * more dangerous than missing an ack
961 	 */
962 	old_snd_una = msk->snd_una;
963 	new_snd_una = expand_ack(old_snd_una, mp_opt->data_ack, mp_opt->ack64);
964 
965 	/* ACK for data not even sent yet? Ignore. */
966 	if (after64(new_snd_una, snd_nxt))
967 		new_snd_una = old_snd_una;
968 
969 	new_wnd_end = new_snd_una + tcp_sk(ssk)->snd_wnd;
970 
971 	if (after64(new_wnd_end, msk->wnd_end))
972 		msk->wnd_end = new_wnd_end;
973 
974 	/* this assumes mptcp_incoming_options() is invoked after tcp_ack() */
975 	if (after64(msk->wnd_end, READ_ONCE(msk->snd_nxt)))
976 		__mptcp_check_push(sk, ssk);
977 
978 	if (after64(new_snd_una, old_snd_una)) {
979 		msk->snd_una = new_snd_una;
980 		__mptcp_data_acked(sk);
981 	}
982 	mptcp_data_unlock(sk);
983 
984 	trace_ack_update_msk(mp_opt->data_ack,
985 			     old_snd_una, new_snd_una,
986 			     new_wnd_end, msk->wnd_end);
987 }
988 
989 bool mptcp_update_rcv_data_fin(struct mptcp_sock *msk, u64 data_fin_seq, bool use_64bit)
990 {
991 	/* Skip if DATA_FIN was already received.
992 	 * If updating simultaneously with the recvmsg loop, values
993 	 * should match. If they mismatch, the peer is misbehaving and
994 	 * we will prefer the most recent information.
995 	 */
996 	if (READ_ONCE(msk->rcv_data_fin))
997 		return false;
998 
999 	WRITE_ONCE(msk->rcv_data_fin_seq,
1000 		   expand_ack(READ_ONCE(msk->ack_seq), data_fin_seq, use_64bit));
1001 	WRITE_ONCE(msk->rcv_data_fin, 1);
1002 
1003 	return true;
1004 }
1005 
1006 static bool add_addr_hmac_valid(struct mptcp_sock *msk,
1007 				struct mptcp_options_received *mp_opt)
1008 {
1009 	u64 hmac = 0;
1010 
1011 	if (mp_opt->echo)
1012 		return true;
1013 
1014 	hmac = add_addr_generate_hmac(msk->remote_key,
1015 				      msk->local_key,
1016 				      &mp_opt->addr);
1017 
1018 	pr_debug("msk=%p, ahmac=%llu, mp_opt->ahmac=%llu\n",
1019 		 msk, (unsigned long long)hmac,
1020 		 (unsigned long long)mp_opt->ahmac);
1021 
1022 	return hmac == mp_opt->ahmac;
1023 }
1024 
1025 void mptcp_incoming_options(struct sock *sk, struct sk_buff *skb)
1026 {
1027 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1028 	struct mptcp_sock *msk = mptcp_sk(subflow->conn);
1029 	struct mptcp_options_received mp_opt;
1030 	struct mptcp_ext *mpext;
1031 
1032 	if (__mptcp_check_fallback(msk)) {
1033 		/* Keep it simple and unconditionally trigger send data cleanup and
1034 		 * pending queue spooling. We will need to acquire the data lock
1035 		 * for more accurate checks, and once the lock is acquired, such
1036 		 * helpers are cheap.
1037 		 */
1038 		mptcp_data_lock(subflow->conn);
1039 		if (sk_stream_memory_free(sk))
1040 			__mptcp_check_push(subflow->conn, sk);
1041 		__mptcp_data_acked(subflow->conn);
1042 		mptcp_data_unlock(subflow->conn);
1043 		return;
1044 	}
1045 
1046 	mptcp_get_options(sk, skb, &mp_opt);
1047 	if (!check_fully_established(msk, sk, subflow, skb, &mp_opt))
1048 		return;
1049 
1050 	if (mp_opt.fastclose &&
1051 	    msk->local_key == mp_opt.rcvr_key) {
1052 		WRITE_ONCE(msk->rcv_fastclose, true);
1053 		mptcp_schedule_work((struct sock *)msk);
1054 	}
1055 
1056 	if (mp_opt.add_addr && add_addr_hmac_valid(msk, &mp_opt)) {
1057 		if (!mp_opt.echo) {
1058 			mptcp_pm_add_addr_received(msk, &mp_opt.addr);
1059 			MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ADDADDR);
1060 		} else {
1061 			mptcp_pm_add_addr_echoed(msk, &mp_opt.addr);
1062 			mptcp_pm_del_add_timer(msk, &mp_opt.addr, true);
1063 			MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ECHOADD);
1064 		}
1065 
1066 		if (mp_opt.addr.port)
1067 			MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_PORTADD);
1068 
1069 		mp_opt.add_addr = 0;
1070 	}
1071 
1072 	if (mp_opt.rm_addr) {
1073 		mptcp_pm_rm_addr_received(msk, &mp_opt.rm_list);
1074 		mp_opt.rm_addr = 0;
1075 	}
1076 
1077 	if (mp_opt.mp_prio) {
1078 		mptcp_pm_mp_prio_received(sk, mp_opt.backup);
1079 		MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPPRIORX);
1080 		mp_opt.mp_prio = 0;
1081 	}
1082 
1083 	if (mp_opt.reset) {
1084 		subflow->reset_seen = 1;
1085 		subflow->reset_reason = mp_opt.reset_reason;
1086 		subflow->reset_transient = mp_opt.reset_transient;
1087 	}
1088 
1089 	if (!mp_opt.dss)
1090 		return;
1091 
1092 	/* we can't wait for recvmsg() to update the ack_seq, otherwise
1093 	 * monodirectional flows will stuck
1094 	 */
1095 	if (mp_opt.use_ack)
1096 		ack_update_msk(msk, sk, &mp_opt);
1097 
1098 	/* Zero-data-length packets are dropped by the caller and not
1099 	 * propagated to the MPTCP layer, so the skb extension does not
1100 	 * need to be allocated or populated. DATA_FIN information, if
1101 	 * present, needs to be updated here before the skb is freed.
1102 	 */
1103 	if (TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb)->end_seq) {
1104 		if (mp_opt.data_fin && mp_opt.data_len == 1 &&
1105 		    mptcp_update_rcv_data_fin(msk, mp_opt.data_seq, mp_opt.dsn64) &&
1106 		    schedule_work(&msk->work))
1107 			sock_hold(subflow->conn);
1108 
1109 		return;
1110 	}
1111 
1112 	mpext = skb_ext_add(skb, SKB_EXT_MPTCP);
1113 	if (!mpext)
1114 		return;
1115 
1116 	memset(mpext, 0, sizeof(*mpext));
1117 
1118 	if (mp_opt.use_map) {
1119 		if (mp_opt.mpc_map) {
1120 			/* this is an MP_CAPABLE carrying MPTCP data
1121 			 * we know this map the first chunk of data
1122 			 */
1123 			mptcp_crypto_key_sha(subflow->remote_key, NULL,
1124 					     &mpext->data_seq);
1125 			mpext->data_seq++;
1126 			mpext->subflow_seq = 1;
1127 			mpext->dsn64 = 1;
1128 			mpext->mpc_map = 1;
1129 			mpext->data_fin = 0;
1130 		} else {
1131 			mpext->data_seq = mp_opt.data_seq;
1132 			mpext->subflow_seq = mp_opt.subflow_seq;
1133 			mpext->dsn64 = mp_opt.dsn64;
1134 			mpext->data_fin = mp_opt.data_fin;
1135 		}
1136 		mpext->data_len = mp_opt.data_len;
1137 		mpext->use_map = 1;
1138 		mpext->csum_reqd = mp_opt.csum_reqd;
1139 
1140 		if (mpext->csum_reqd)
1141 			mpext->csum = mp_opt.csum;
1142 	}
1143 }
1144 
1145 static void mptcp_set_rwin(const struct tcp_sock *tp)
1146 {
1147 	const struct sock *ssk = (const struct sock *)tp;
1148 	const struct mptcp_subflow_context *subflow;
1149 	struct mptcp_sock *msk;
1150 	u64 ack_seq;
1151 
1152 	subflow = mptcp_subflow_ctx(ssk);
1153 	msk = mptcp_sk(subflow->conn);
1154 
1155 	ack_seq = READ_ONCE(msk->ack_seq) + tp->rcv_wnd;
1156 
1157 	if (after64(ack_seq, READ_ONCE(msk->rcv_wnd_sent)))
1158 		WRITE_ONCE(msk->rcv_wnd_sent, ack_seq);
1159 }
1160 
1161 static u16 mptcp_make_csum(const struct mptcp_ext *mpext)
1162 {
1163 	struct csum_pseudo_header header;
1164 	__wsum csum;
1165 
1166 	/* cfr RFC 8684 3.3.1.:
1167 	 * the data sequence number used in the pseudo-header is
1168 	 * always the 64-bit value, irrespective of what length is used in the
1169 	 * DSS option itself.
1170 	 */
1171 	header.data_seq = cpu_to_be64(mpext->data_seq);
1172 	header.subflow_seq = htonl(mpext->subflow_seq);
1173 	header.data_len = htons(mpext->data_len);
1174 	header.csum = 0;
1175 
1176 	csum = csum_partial(&header, sizeof(header), ~csum_unfold(mpext->csum));
1177 	return (__force u16)csum_fold(csum);
1178 }
1179 
1180 void mptcp_write_options(__be32 *ptr, const struct tcp_sock *tp,
1181 			 struct mptcp_out_options *opts)
1182 {
1183 	if ((OPTION_MPTCP_MPC_SYN | OPTION_MPTCP_MPC_SYNACK |
1184 	     OPTION_MPTCP_MPC_ACK) & opts->suboptions) {
1185 		u8 len, flag = MPTCP_CAP_HMAC_SHA256;
1186 
1187 		if (OPTION_MPTCP_MPC_SYN & opts->suboptions) {
1188 			len = TCPOLEN_MPTCP_MPC_SYN;
1189 		} else if (OPTION_MPTCP_MPC_SYNACK & opts->suboptions) {
1190 			len = TCPOLEN_MPTCP_MPC_SYNACK;
1191 		} else if (opts->ext_copy.data_len) {
1192 			len = TCPOLEN_MPTCP_MPC_ACK_DATA;
1193 			if (opts->csum_reqd)
1194 				len += TCPOLEN_MPTCP_DSS_CHECKSUM;
1195 		} else {
1196 			len = TCPOLEN_MPTCP_MPC_ACK;
1197 		}
1198 
1199 		if (opts->csum_reqd)
1200 			flag |= MPTCP_CAP_CHECKSUM_REQD;
1201 
1202 		*ptr++ = mptcp_option(MPTCPOPT_MP_CAPABLE, len,
1203 				      MPTCP_SUPPORTED_VERSION,
1204 				      flag);
1205 
1206 		if (!((OPTION_MPTCP_MPC_SYNACK | OPTION_MPTCP_MPC_ACK) &
1207 		    opts->suboptions))
1208 			goto mp_capable_done;
1209 
1210 		put_unaligned_be64(opts->sndr_key, ptr);
1211 		ptr += 2;
1212 		if (!((OPTION_MPTCP_MPC_ACK) & opts->suboptions))
1213 			goto mp_capable_done;
1214 
1215 		put_unaligned_be64(opts->rcvr_key, ptr);
1216 		ptr += 2;
1217 		if (!opts->ext_copy.data_len)
1218 			goto mp_capable_done;
1219 
1220 		if (opts->csum_reqd) {
1221 			put_unaligned_be32(opts->ext_copy.data_len << 16 |
1222 					   mptcp_make_csum(&opts->ext_copy), ptr);
1223 		} else {
1224 			put_unaligned_be32(opts->ext_copy.data_len << 16 |
1225 					   TCPOPT_NOP << 8 | TCPOPT_NOP, ptr);
1226 		}
1227 		ptr += 1;
1228 	}
1229 
1230 mp_capable_done:
1231 	if (OPTION_MPTCP_ADD_ADDR & opts->suboptions) {
1232 		u8 len = TCPOLEN_MPTCP_ADD_ADDR_BASE;
1233 		u8 echo = MPTCP_ADDR_ECHO;
1234 
1235 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1236 		if (opts->addr.family == AF_INET6)
1237 			len = TCPOLEN_MPTCP_ADD_ADDR6_BASE;
1238 #endif
1239 
1240 		if (opts->addr.port)
1241 			len += TCPOLEN_MPTCP_PORT_LEN;
1242 
1243 		if (opts->ahmac) {
1244 			len += sizeof(opts->ahmac);
1245 			echo = 0;
1246 		}
1247 
1248 		*ptr++ = mptcp_option(MPTCPOPT_ADD_ADDR,
1249 				      len, echo, opts->addr.id);
1250 		if (opts->addr.family == AF_INET) {
1251 			memcpy((u8 *)ptr, (u8 *)&opts->addr.addr.s_addr, 4);
1252 			ptr += 1;
1253 		}
1254 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1255 		else if (opts->addr.family == AF_INET6) {
1256 			memcpy((u8 *)ptr, opts->addr.addr6.s6_addr, 16);
1257 			ptr += 4;
1258 		}
1259 #endif
1260 
1261 		if (!opts->addr.port) {
1262 			if (opts->ahmac) {
1263 				put_unaligned_be64(opts->ahmac, ptr);
1264 				ptr += 2;
1265 			}
1266 		} else {
1267 			u16 port = ntohs(opts->addr.port);
1268 
1269 			if (opts->ahmac) {
1270 				u8 *bptr = (u8 *)ptr;
1271 
1272 				put_unaligned_be16(port, bptr);
1273 				bptr += 2;
1274 				put_unaligned_be64(opts->ahmac, bptr);
1275 				bptr += 8;
1276 				put_unaligned_be16(TCPOPT_NOP << 8 |
1277 						   TCPOPT_NOP, bptr);
1278 
1279 				ptr += 3;
1280 			} else {
1281 				put_unaligned_be32(port << 16 |
1282 						   TCPOPT_NOP << 8 |
1283 						   TCPOPT_NOP, ptr);
1284 				ptr += 1;
1285 			}
1286 		}
1287 	}
1288 
1289 	if (OPTION_MPTCP_RM_ADDR & opts->suboptions) {
1290 		u8 i = 1;
1291 
1292 		*ptr++ = mptcp_option(MPTCPOPT_RM_ADDR,
1293 				      TCPOLEN_MPTCP_RM_ADDR_BASE + opts->rm_list.nr,
1294 				      0, opts->rm_list.ids[0]);
1295 
1296 		while (i < opts->rm_list.nr) {
1297 			u8 id1, id2, id3, id4;
1298 
1299 			id1 = opts->rm_list.ids[i];
1300 			id2 = i + 1 < opts->rm_list.nr ? opts->rm_list.ids[i + 1] : TCPOPT_NOP;
1301 			id3 = i + 2 < opts->rm_list.nr ? opts->rm_list.ids[i + 2] : TCPOPT_NOP;
1302 			id4 = i + 3 < opts->rm_list.nr ? opts->rm_list.ids[i + 3] : TCPOPT_NOP;
1303 			put_unaligned_be32(id1 << 24 | id2 << 16 | id3 << 8 | id4, ptr);
1304 			ptr += 1;
1305 			i += 4;
1306 		}
1307 	}
1308 
1309 	if (OPTION_MPTCP_PRIO & opts->suboptions) {
1310 		const struct sock *ssk = (const struct sock *)tp;
1311 		struct mptcp_subflow_context *subflow;
1312 
1313 		subflow = mptcp_subflow_ctx(ssk);
1314 		subflow->send_mp_prio = 0;
1315 
1316 		*ptr++ = mptcp_option(MPTCPOPT_MP_PRIO,
1317 				      TCPOLEN_MPTCP_PRIO,
1318 				      opts->backup, TCPOPT_NOP);
1319 	}
1320 
1321 	if (OPTION_MPTCP_MPJ_SYN & opts->suboptions) {
1322 		*ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1323 				      TCPOLEN_MPTCP_MPJ_SYN,
1324 				      opts->backup, opts->join_id);
1325 		put_unaligned_be32(opts->token, ptr);
1326 		ptr += 1;
1327 		put_unaligned_be32(opts->nonce, ptr);
1328 		ptr += 1;
1329 	}
1330 
1331 	if (OPTION_MPTCP_MPJ_SYNACK & opts->suboptions) {
1332 		*ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1333 				      TCPOLEN_MPTCP_MPJ_SYNACK,
1334 				      opts->backup, opts->join_id);
1335 		put_unaligned_be64(opts->thmac, ptr);
1336 		ptr += 2;
1337 		put_unaligned_be32(opts->nonce, ptr);
1338 		ptr += 1;
1339 	}
1340 
1341 	if (OPTION_MPTCP_MPJ_ACK & opts->suboptions) {
1342 		*ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1343 				      TCPOLEN_MPTCP_MPJ_ACK, 0, 0);
1344 		memcpy(ptr, opts->hmac, MPTCPOPT_HMAC_LEN);
1345 		ptr += 5;
1346 	}
1347 
1348 	if (OPTION_MPTCP_RST & opts->suboptions)
1349 		*ptr++ = mptcp_option(MPTCPOPT_RST,
1350 				      TCPOLEN_MPTCP_RST,
1351 				      opts->reset_transient,
1352 				      opts->reset_reason);
1353 
1354 	if (opts->ext_copy.use_ack || opts->ext_copy.use_map) {
1355 		struct mptcp_ext *mpext = &opts->ext_copy;
1356 		u8 len = TCPOLEN_MPTCP_DSS_BASE;
1357 		u8 flags = 0;
1358 
1359 		if (mpext->use_ack) {
1360 			flags = MPTCP_DSS_HAS_ACK;
1361 			if (mpext->ack64) {
1362 				len += TCPOLEN_MPTCP_DSS_ACK64;
1363 				flags |= MPTCP_DSS_ACK64;
1364 			} else {
1365 				len += TCPOLEN_MPTCP_DSS_ACK32;
1366 			}
1367 		}
1368 
1369 		if (mpext->use_map) {
1370 			len += TCPOLEN_MPTCP_DSS_MAP64;
1371 
1372 			/* Use only 64-bit mapping flags for now, add
1373 			 * support for optional 32-bit mappings later.
1374 			 */
1375 			flags |= MPTCP_DSS_HAS_MAP | MPTCP_DSS_DSN64;
1376 			if (mpext->data_fin)
1377 				flags |= MPTCP_DSS_DATA_FIN;
1378 
1379 			if (opts->csum_reqd)
1380 				len += TCPOLEN_MPTCP_DSS_CHECKSUM;
1381 		}
1382 
1383 		*ptr++ = mptcp_option(MPTCPOPT_DSS, len, 0, flags);
1384 
1385 		if (mpext->use_ack) {
1386 			if (mpext->ack64) {
1387 				put_unaligned_be64(mpext->data_ack, ptr);
1388 				ptr += 2;
1389 			} else {
1390 				put_unaligned_be32(mpext->data_ack32, ptr);
1391 				ptr += 1;
1392 			}
1393 		}
1394 
1395 		if (mpext->use_map) {
1396 			put_unaligned_be64(mpext->data_seq, ptr);
1397 			ptr += 2;
1398 			put_unaligned_be32(mpext->subflow_seq, ptr);
1399 			ptr += 1;
1400 			if (opts->csum_reqd) {
1401 				put_unaligned_be32(mpext->data_len << 16 |
1402 						   mptcp_make_csum(mpext), ptr);
1403 			} else {
1404 				put_unaligned_be32(mpext->data_len << 16 |
1405 						   TCPOPT_NOP << 8 | TCPOPT_NOP, ptr);
1406 			}
1407 		}
1408 	}
1409 
1410 	if (tp)
1411 		mptcp_set_rwin(tp);
1412 }
1413 
1414 __be32 mptcp_get_reset_option(const struct sk_buff *skb)
1415 {
1416 	const struct mptcp_ext *ext = mptcp_get_ext(skb);
1417 	u8 flags, reason;
1418 
1419 	if (ext) {
1420 		flags = ext->reset_transient;
1421 		reason = ext->reset_reason;
1422 
1423 		return mptcp_option(MPTCPOPT_RST, TCPOLEN_MPTCP_RST,
1424 				    flags, reason);
1425 	}
1426 
1427 	return htonl(0u);
1428 }
1429 EXPORT_SYMBOL_GPL(mptcp_get_reset_option);
1430