xref: /linux/net/mptcp/options.c (revision 8137a49e1567726eb10fcf55ad141ac19804ca6b)
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 		if (flags & MPTCP_CAP_DENY_JOIN_ID0)
87 			mp_opt->deny_join_id0 = 1;
88 
89 		mp_opt->mp_capable = 1;
90 		if (opsize >= TCPOLEN_MPTCP_MPC_SYNACK) {
91 			mp_opt->sndr_key = get_unaligned_be64(ptr);
92 			ptr += 8;
93 		}
94 		if (opsize >= TCPOLEN_MPTCP_MPC_ACK) {
95 			mp_opt->rcvr_key = get_unaligned_be64(ptr);
96 			ptr += 8;
97 		}
98 		if (opsize >= TCPOLEN_MPTCP_MPC_ACK_DATA) {
99 			/* Section 3.1.:
100 			 * "the data parameters in a MP_CAPABLE are semantically
101 			 * equivalent to those in a DSS option and can be used
102 			 * interchangeably."
103 			 */
104 			mp_opt->dss = 1;
105 			mp_opt->use_map = 1;
106 			mp_opt->mpc_map = 1;
107 			mp_opt->data_len = get_unaligned_be16(ptr);
108 			ptr += 2;
109 		}
110 		if (opsize == TCPOLEN_MPTCP_MPC_ACK_DATA_CSUM) {
111 			mp_opt->csum = (__force __sum16)get_unaligned_be16(ptr);
112 			mp_opt->csum_reqd = 1;
113 			ptr += 2;
114 		}
115 		pr_debug("MP_CAPABLE version=%x, flags=%x, optlen=%d sndr=%llu, rcvr=%llu len=%d csum=%u",
116 			 version, flags, opsize, mp_opt->sndr_key,
117 			 mp_opt->rcvr_key, mp_opt->data_len, mp_opt->csum);
118 		break;
119 
120 	case MPTCPOPT_MP_JOIN:
121 		mp_opt->mp_join = 1;
122 		if (opsize == TCPOLEN_MPTCP_MPJ_SYN) {
123 			mp_opt->backup = *ptr++ & MPTCPOPT_BACKUP;
124 			mp_opt->join_id = *ptr++;
125 			mp_opt->token = get_unaligned_be32(ptr);
126 			ptr += 4;
127 			mp_opt->nonce = get_unaligned_be32(ptr);
128 			ptr += 4;
129 			pr_debug("MP_JOIN bkup=%u, id=%u, token=%u, nonce=%u",
130 				 mp_opt->backup, mp_opt->join_id,
131 				 mp_opt->token, mp_opt->nonce);
132 		} else if (opsize == TCPOLEN_MPTCP_MPJ_SYNACK) {
133 			mp_opt->backup = *ptr++ & MPTCPOPT_BACKUP;
134 			mp_opt->join_id = *ptr++;
135 			mp_opt->thmac = get_unaligned_be64(ptr);
136 			ptr += 8;
137 			mp_opt->nonce = get_unaligned_be32(ptr);
138 			ptr += 4;
139 			pr_debug("MP_JOIN bkup=%u, id=%u, thmac=%llu, nonce=%u",
140 				 mp_opt->backup, mp_opt->join_id,
141 				 mp_opt->thmac, mp_opt->nonce);
142 		} else if (opsize == TCPOLEN_MPTCP_MPJ_ACK) {
143 			ptr += 2;
144 			memcpy(mp_opt->hmac, ptr, MPTCPOPT_HMAC_LEN);
145 			pr_debug("MP_JOIN hmac");
146 		} else {
147 			mp_opt->mp_join = 0;
148 		}
149 		break;
150 
151 	case MPTCPOPT_DSS:
152 		pr_debug("DSS");
153 		ptr++;
154 
155 		/* we must clear 'mpc_map' be able to detect MP_CAPABLE
156 		 * map vs DSS map in mptcp_incoming_options(), and reconstruct
157 		 * map info accordingly
158 		 */
159 		mp_opt->mpc_map = 0;
160 		flags = (*ptr++) & MPTCP_DSS_FLAG_MASK;
161 		mp_opt->data_fin = (flags & MPTCP_DSS_DATA_FIN) != 0;
162 		mp_opt->dsn64 = (flags & MPTCP_DSS_DSN64) != 0;
163 		mp_opt->use_map = (flags & MPTCP_DSS_HAS_MAP) != 0;
164 		mp_opt->ack64 = (flags & MPTCP_DSS_ACK64) != 0;
165 		mp_opt->use_ack = (flags & MPTCP_DSS_HAS_ACK);
166 
167 		pr_debug("data_fin=%d dsn64=%d use_map=%d ack64=%d use_ack=%d",
168 			 mp_opt->data_fin, mp_opt->dsn64,
169 			 mp_opt->use_map, mp_opt->ack64,
170 			 mp_opt->use_ack);
171 
172 		expected_opsize = TCPOLEN_MPTCP_DSS_BASE;
173 
174 		if (mp_opt->use_ack) {
175 			if (mp_opt->ack64)
176 				expected_opsize += TCPOLEN_MPTCP_DSS_ACK64;
177 			else
178 				expected_opsize += TCPOLEN_MPTCP_DSS_ACK32;
179 		}
180 
181 		if (mp_opt->use_map) {
182 			if (mp_opt->dsn64)
183 				expected_opsize += TCPOLEN_MPTCP_DSS_MAP64;
184 			else
185 				expected_opsize += TCPOLEN_MPTCP_DSS_MAP32;
186 		}
187 
188 		/* Always parse any csum presence combination, we will enforce
189 		 * RFC 8684 Section 3.3.0 checks later in subflow_data_ready
190 		 */
191 		if (opsize != expected_opsize &&
192 		    opsize != expected_opsize + TCPOLEN_MPTCP_DSS_CHECKSUM)
193 			break;
194 
195 		mp_opt->dss = 1;
196 
197 		if (mp_opt->use_ack) {
198 			if (mp_opt->ack64) {
199 				mp_opt->data_ack = get_unaligned_be64(ptr);
200 				ptr += 8;
201 			} else {
202 				mp_opt->data_ack = get_unaligned_be32(ptr);
203 				ptr += 4;
204 			}
205 
206 			pr_debug("data_ack=%llu", mp_opt->data_ack);
207 		}
208 
209 		if (mp_opt->use_map) {
210 			if (mp_opt->dsn64) {
211 				mp_opt->data_seq = get_unaligned_be64(ptr);
212 				ptr += 8;
213 			} else {
214 				mp_opt->data_seq = get_unaligned_be32(ptr);
215 				ptr += 4;
216 			}
217 
218 			mp_opt->subflow_seq = get_unaligned_be32(ptr);
219 			ptr += 4;
220 
221 			mp_opt->data_len = get_unaligned_be16(ptr);
222 			ptr += 2;
223 
224 			if (opsize == expected_opsize + TCPOLEN_MPTCP_DSS_CHECKSUM) {
225 				mp_opt->csum_reqd = 1;
226 				mp_opt->csum = (__force __sum16)get_unaligned_be16(ptr);
227 				ptr += 2;
228 			}
229 
230 			pr_debug("data_seq=%llu subflow_seq=%u data_len=%u csum=%d:%u",
231 				 mp_opt->data_seq, mp_opt->subflow_seq,
232 				 mp_opt->data_len, mp_opt->csum_reqd, mp_opt->csum);
233 		}
234 
235 		break;
236 
237 	case MPTCPOPT_ADD_ADDR:
238 		mp_opt->echo = (*ptr++) & MPTCP_ADDR_ECHO;
239 		if (!mp_opt->echo) {
240 			if (opsize == TCPOLEN_MPTCP_ADD_ADDR ||
241 			    opsize == TCPOLEN_MPTCP_ADD_ADDR_PORT)
242 				mp_opt->addr.family = AF_INET;
243 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
244 			else if (opsize == TCPOLEN_MPTCP_ADD_ADDR6 ||
245 				 opsize == TCPOLEN_MPTCP_ADD_ADDR6_PORT)
246 				mp_opt->addr.family = AF_INET6;
247 #endif
248 			else
249 				break;
250 		} else {
251 			if (opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE ||
252 			    opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE_PORT)
253 				mp_opt->addr.family = AF_INET;
254 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
255 			else if (opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE ||
256 				 opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE_PORT)
257 				mp_opt->addr.family = AF_INET6;
258 #endif
259 			else
260 				break;
261 		}
262 
263 		mp_opt->add_addr = 1;
264 		mp_opt->addr.id = *ptr++;
265 		if (mp_opt->addr.family == AF_INET) {
266 			memcpy((u8 *)&mp_opt->addr.addr.s_addr, (u8 *)ptr, 4);
267 			ptr += 4;
268 			if (opsize == TCPOLEN_MPTCP_ADD_ADDR_PORT ||
269 			    opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE_PORT) {
270 				mp_opt->addr.port = htons(get_unaligned_be16(ptr));
271 				ptr += 2;
272 			}
273 		}
274 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
275 		else {
276 			memcpy(mp_opt->addr.addr6.s6_addr, (u8 *)ptr, 16);
277 			ptr += 16;
278 			if (opsize == TCPOLEN_MPTCP_ADD_ADDR6_PORT ||
279 			    opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE_PORT) {
280 				mp_opt->addr.port = htons(get_unaligned_be16(ptr));
281 				ptr += 2;
282 			}
283 		}
284 #endif
285 		if (!mp_opt->echo) {
286 			mp_opt->ahmac = get_unaligned_be64(ptr);
287 			ptr += 8;
288 		}
289 		pr_debug("ADD_ADDR%s: id=%d, ahmac=%llu, echo=%d, port=%d",
290 			 (mp_opt->addr.family == AF_INET6) ? "6" : "",
291 			 mp_opt->addr.id, mp_opt->ahmac, mp_opt->echo, ntohs(mp_opt->addr.port));
292 		break;
293 
294 	case MPTCPOPT_RM_ADDR:
295 		if (opsize < TCPOLEN_MPTCP_RM_ADDR_BASE + 1 ||
296 		    opsize > TCPOLEN_MPTCP_RM_ADDR_BASE + MPTCP_RM_IDS_MAX)
297 			break;
298 
299 		ptr++;
300 
301 		mp_opt->rm_addr = 1;
302 		mp_opt->rm_list.nr = opsize - TCPOLEN_MPTCP_RM_ADDR_BASE;
303 		for (i = 0; i < mp_opt->rm_list.nr; i++)
304 			mp_opt->rm_list.ids[i] = *ptr++;
305 		pr_debug("RM_ADDR: rm_list_nr=%d", mp_opt->rm_list.nr);
306 		break;
307 
308 	case MPTCPOPT_MP_PRIO:
309 		if (opsize != TCPOLEN_MPTCP_PRIO)
310 			break;
311 
312 		mp_opt->mp_prio = 1;
313 		mp_opt->backup = *ptr++ & MPTCP_PRIO_BKUP;
314 		pr_debug("MP_PRIO: prio=%d", mp_opt->backup);
315 		break;
316 
317 	case MPTCPOPT_MP_FASTCLOSE:
318 		if (opsize != TCPOLEN_MPTCP_FASTCLOSE)
319 			break;
320 
321 		ptr += 2;
322 		mp_opt->rcvr_key = get_unaligned_be64(ptr);
323 		ptr += 8;
324 		mp_opt->fastclose = 1;
325 		break;
326 
327 	case MPTCPOPT_RST:
328 		if (opsize != TCPOLEN_MPTCP_RST)
329 			break;
330 
331 		if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_RST))
332 			break;
333 		mp_opt->reset = 1;
334 		flags = *ptr++;
335 		mp_opt->reset_transient = flags & MPTCP_RST_TRANSIENT;
336 		mp_opt->reset_reason = *ptr;
337 		break;
338 
339 	default:
340 		break;
341 	}
342 }
343 
344 void mptcp_get_options(const struct sock *sk,
345 		       const struct sk_buff *skb,
346 		       struct mptcp_options_received *mp_opt)
347 {
348 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
349 	struct mptcp_sock *msk = mptcp_sk(subflow->conn);
350 	const struct tcphdr *th = tcp_hdr(skb);
351 	const unsigned char *ptr;
352 	int length;
353 
354 	/* initialize option status */
355 	mp_opt->mp_capable = 0;
356 	mp_opt->mp_join = 0;
357 	mp_opt->add_addr = 0;
358 	mp_opt->ahmac = 0;
359 	mp_opt->fastclose = 0;
360 	mp_opt->addr.port = 0;
361 	mp_opt->rm_addr = 0;
362 	mp_opt->dss = 0;
363 	mp_opt->mp_prio = 0;
364 	mp_opt->reset = 0;
365 	mp_opt->csum_reqd = READ_ONCE(msk->csum_enabled);
366 	mp_opt->deny_join_id0 = 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 /* MP_JOIN client subflow must wait for 4th ack before sending any data:
427  * TCP can't schedule delack timer before the subflow is fully established.
428  * MPTCP uses the delack timer to do 3rd ack retransmissions
429  */
430 static void schedule_3rdack_retransmission(struct sock *sk)
431 {
432 	struct inet_connection_sock *icsk = inet_csk(sk);
433 	struct tcp_sock *tp = tcp_sk(sk);
434 	unsigned long timeout;
435 
436 	/* reschedule with a timeout above RTT, as we must look only for drop */
437 	if (tp->srtt_us)
438 		timeout = tp->srtt_us << 1;
439 	else
440 		timeout = TCP_TIMEOUT_INIT;
441 
442 	WARN_ON_ONCE(icsk->icsk_ack.pending & ICSK_ACK_TIMER);
443 	icsk->icsk_ack.pending |= ICSK_ACK_SCHED | ICSK_ACK_TIMER;
444 	icsk->icsk_ack.timeout = timeout;
445 	sk_reset_timer(sk, &icsk->icsk_delack_timer, timeout);
446 }
447 
448 static void clear_3rdack_retransmission(struct sock *sk)
449 {
450 	struct inet_connection_sock *icsk = inet_csk(sk);
451 
452 	sk_stop_timer(sk, &icsk->icsk_delack_timer);
453 	icsk->icsk_ack.timeout = 0;
454 	icsk->icsk_ack.ato = 0;
455 	icsk->icsk_ack.pending &= ~(ICSK_ACK_SCHED | ICSK_ACK_TIMER);
456 }
457 
458 static bool mptcp_established_options_mp(struct sock *sk, struct sk_buff *skb,
459 					 bool snd_data_fin_enable,
460 					 unsigned int *size,
461 					 unsigned int remaining,
462 					 struct mptcp_out_options *opts)
463 {
464 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
465 	struct mptcp_sock *msk = mptcp_sk(subflow->conn);
466 	struct mptcp_ext *mpext;
467 	unsigned int data_len;
468 	u8 len;
469 
470 	/* When skb is not available, we better over-estimate the emitted
471 	 * options len. A full DSS option (28 bytes) is longer than
472 	 * TCPOLEN_MPTCP_MPC_ACK_DATA(22) or TCPOLEN_MPTCP_MPJ_ACK(24), so
473 	 * tell the caller to defer the estimate to
474 	 * mptcp_established_options_dss(), which will reserve enough space.
475 	 */
476 	if (!skb)
477 		return false;
478 
479 	/* MPC/MPJ needed only on 3rd ack packet, DATA_FIN and TCP shutdown take precedence */
480 	if (subflow->fully_established || snd_data_fin_enable ||
481 	    subflow->snd_isn != TCP_SKB_CB(skb)->seq ||
482 	    sk->sk_state != TCP_ESTABLISHED)
483 		return false;
484 
485 	if (subflow->mp_capable) {
486 		mpext = mptcp_get_ext(skb);
487 		data_len = mpext ? mpext->data_len : 0;
488 
489 		/* we will check ext_copy.data_len in mptcp_write_options() to
490 		 * discriminate between TCPOLEN_MPTCP_MPC_ACK_DATA and
491 		 * TCPOLEN_MPTCP_MPC_ACK
492 		 */
493 		opts->ext_copy.data_len = data_len;
494 		opts->suboptions = OPTION_MPTCP_MPC_ACK;
495 		opts->sndr_key = subflow->local_key;
496 		opts->rcvr_key = subflow->remote_key;
497 		opts->csum_reqd = READ_ONCE(msk->csum_enabled);
498 		opts->allow_join_id0 = mptcp_allow_join_id0(sock_net(sk));
499 
500 		/* Section 3.1.
501 		 * The MP_CAPABLE option is carried on the SYN, SYN/ACK, and ACK
502 		 * packets that start the first subflow of an MPTCP connection,
503 		 * as well as the first packet that carries data
504 		 */
505 		if (data_len > 0) {
506 			len = TCPOLEN_MPTCP_MPC_ACK_DATA;
507 			if (opts->csum_reqd) {
508 				/* we need to propagate more info to csum the pseudo hdr */
509 				opts->ext_copy.data_seq = mpext->data_seq;
510 				opts->ext_copy.subflow_seq = mpext->subflow_seq;
511 				opts->ext_copy.csum = mpext->csum;
512 				len += TCPOLEN_MPTCP_DSS_CHECKSUM;
513 			}
514 			*size = ALIGN(len, 4);
515 		} else {
516 			*size = TCPOLEN_MPTCP_MPC_ACK;
517 		}
518 
519 		pr_debug("subflow=%p, local_key=%llu, remote_key=%llu map_len=%d",
520 			 subflow, subflow->local_key, subflow->remote_key,
521 			 data_len);
522 
523 		return true;
524 	} else if (subflow->mp_join) {
525 		opts->suboptions = OPTION_MPTCP_MPJ_ACK;
526 		memcpy(opts->hmac, subflow->hmac, MPTCPOPT_HMAC_LEN);
527 		*size = TCPOLEN_MPTCP_MPJ_ACK;
528 		pr_debug("subflow=%p", subflow);
529 
530 		schedule_3rdack_retransmission(sk);
531 		return true;
532 	}
533 	return false;
534 }
535 
536 static void mptcp_write_data_fin(struct mptcp_subflow_context *subflow,
537 				 struct sk_buff *skb, struct mptcp_ext *ext)
538 {
539 	/* The write_seq value has already been incremented, so the actual
540 	 * sequence number for the DATA_FIN is one less.
541 	 */
542 	u64 data_fin_tx_seq = READ_ONCE(mptcp_sk(subflow->conn)->write_seq) - 1;
543 
544 	if (!ext->use_map || !skb->len) {
545 		/* RFC6824 requires a DSS mapping with specific values
546 		 * if DATA_FIN is set but no data payload is mapped
547 		 */
548 		ext->data_fin = 1;
549 		ext->use_map = 1;
550 		ext->dsn64 = 1;
551 		ext->data_seq = data_fin_tx_seq;
552 		ext->subflow_seq = 0;
553 		ext->data_len = 1;
554 	} else if (ext->data_seq + ext->data_len == data_fin_tx_seq) {
555 		/* If there's an existing DSS mapping and it is the
556 		 * final mapping, DATA_FIN consumes 1 additional byte of
557 		 * mapping space.
558 		 */
559 		ext->data_fin = 1;
560 		ext->data_len++;
561 	}
562 }
563 
564 static bool mptcp_established_options_dss(struct sock *sk, struct sk_buff *skb,
565 					  bool snd_data_fin_enable,
566 					  unsigned int *size,
567 					  unsigned int remaining,
568 					  struct mptcp_out_options *opts)
569 {
570 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
571 	struct mptcp_sock *msk = mptcp_sk(subflow->conn);
572 	unsigned int dss_size = 0;
573 	struct mptcp_ext *mpext;
574 	unsigned int ack_size;
575 	bool ret = false;
576 	u64 ack_seq;
577 
578 	opts->csum_reqd = READ_ONCE(msk->csum_enabled);
579 	mpext = skb ? mptcp_get_ext(skb) : NULL;
580 
581 	if (!skb || (mpext && mpext->use_map) || snd_data_fin_enable) {
582 		unsigned int map_size = TCPOLEN_MPTCP_DSS_BASE + TCPOLEN_MPTCP_DSS_MAP64;
583 
584 		if (mpext) {
585 			if (opts->csum_reqd)
586 				map_size += TCPOLEN_MPTCP_DSS_CHECKSUM;
587 
588 			opts->ext_copy = *mpext;
589 		}
590 
591 		remaining -= map_size;
592 		dss_size = map_size;
593 		if (skb && snd_data_fin_enable)
594 			mptcp_write_data_fin(subflow, skb, &opts->ext_copy);
595 		ret = true;
596 	}
597 
598 	/* passive sockets msk will set the 'can_ack' after accept(), even
599 	 * if the first subflow may have the already the remote key handy
600 	 */
601 	opts->ext_copy.use_ack = 0;
602 	if (!READ_ONCE(msk->can_ack)) {
603 		*size = ALIGN(dss_size, 4);
604 		return ret;
605 	}
606 
607 	ack_seq = READ_ONCE(msk->ack_seq);
608 	if (READ_ONCE(msk->use_64bit_ack)) {
609 		ack_size = TCPOLEN_MPTCP_DSS_ACK64;
610 		opts->ext_copy.data_ack = ack_seq;
611 		opts->ext_copy.ack64 = 1;
612 	} else {
613 		ack_size = TCPOLEN_MPTCP_DSS_ACK32;
614 		opts->ext_copy.data_ack32 = (uint32_t)ack_seq;
615 		opts->ext_copy.ack64 = 0;
616 	}
617 	opts->ext_copy.use_ack = 1;
618 	WRITE_ONCE(msk->old_wspace, __mptcp_space((struct sock *)msk));
619 
620 	/* Add kind/length/subtype/flag overhead if mapping is not populated */
621 	if (dss_size == 0)
622 		ack_size += TCPOLEN_MPTCP_DSS_BASE;
623 
624 	dss_size += ack_size;
625 
626 	*size = ALIGN(dss_size, 4);
627 	return true;
628 }
629 
630 static u64 add_addr_generate_hmac(u64 key1, u64 key2,
631 				  struct mptcp_addr_info *addr)
632 {
633 	u16 port = ntohs(addr->port);
634 	u8 hmac[SHA256_DIGEST_SIZE];
635 	u8 msg[19];
636 	int i = 0;
637 
638 	msg[i++] = addr->id;
639 	if (addr->family == AF_INET) {
640 		memcpy(&msg[i], &addr->addr.s_addr, 4);
641 		i += 4;
642 	}
643 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
644 	else if (addr->family == AF_INET6) {
645 		memcpy(&msg[i], &addr->addr6.s6_addr, 16);
646 		i += 16;
647 	}
648 #endif
649 	msg[i++] = port >> 8;
650 	msg[i++] = port & 0xFF;
651 
652 	mptcp_crypto_hmac_sha(key1, key2, msg, i, hmac);
653 
654 	return get_unaligned_be64(&hmac[SHA256_DIGEST_SIZE - sizeof(u64)]);
655 }
656 
657 static bool mptcp_established_options_add_addr(struct sock *sk, struct sk_buff *skb,
658 					       unsigned int *size,
659 					       unsigned int remaining,
660 					       struct mptcp_out_options *opts)
661 {
662 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
663 	struct mptcp_sock *msk = mptcp_sk(subflow->conn);
664 	bool drop_other_suboptions = false;
665 	unsigned int opt_size = *size;
666 	bool echo;
667 	bool port;
668 	int len;
669 
670 	if ((mptcp_pm_should_add_signal_ipv6(msk) ||
671 	     mptcp_pm_should_add_signal_port(msk) ||
672 	     mptcp_pm_should_add_signal_echo(msk)) &&
673 	    skb && skb_is_tcp_pure_ack(skb)) {
674 		pr_debug("drop other suboptions");
675 		opts->suboptions = 0;
676 		opts->ext_copy.use_ack = 0;
677 		opts->ext_copy.use_map = 0;
678 		remaining += opt_size;
679 		drop_other_suboptions = true;
680 	}
681 
682 	if (!mptcp_pm_should_add_signal(msk) ||
683 	    !(mptcp_pm_add_addr_signal(msk, remaining, &opts->addr, &echo, &port)))
684 		return false;
685 
686 	len = mptcp_add_addr_len(opts->addr.family, echo, port);
687 	if (remaining < len)
688 		return false;
689 
690 	*size = len;
691 	if (drop_other_suboptions)
692 		*size -= opt_size;
693 	opts->suboptions |= OPTION_MPTCP_ADD_ADDR;
694 	if (!echo) {
695 		opts->ahmac = add_addr_generate_hmac(msk->local_key,
696 						     msk->remote_key,
697 						     &opts->addr);
698 	}
699 	pr_debug("addr_id=%d, ahmac=%llu, echo=%d, port=%d",
700 		 opts->addr.id, opts->ahmac, echo, ntohs(opts->addr.port));
701 
702 	return true;
703 }
704 
705 static bool mptcp_established_options_rm_addr(struct sock *sk,
706 					      unsigned int *size,
707 					      unsigned int remaining,
708 					      struct mptcp_out_options *opts)
709 {
710 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
711 	struct mptcp_sock *msk = mptcp_sk(subflow->conn);
712 	struct mptcp_rm_list rm_list;
713 	int i, len;
714 
715 	if (!mptcp_pm_should_rm_signal(msk) ||
716 	    !(mptcp_pm_rm_addr_signal(msk, remaining, &rm_list)))
717 		return false;
718 
719 	len = mptcp_rm_addr_len(&rm_list);
720 	if (len < 0)
721 		return false;
722 	if (remaining < len)
723 		return false;
724 
725 	*size = len;
726 	opts->suboptions |= OPTION_MPTCP_RM_ADDR;
727 	opts->rm_list = rm_list;
728 
729 	for (i = 0; i < opts->rm_list.nr; i++)
730 		pr_debug("rm_list_ids[%d]=%d", i, opts->rm_list.ids[i]);
731 
732 	return true;
733 }
734 
735 static bool mptcp_established_options_mp_prio(struct sock *sk,
736 					      unsigned int *size,
737 					      unsigned int remaining,
738 					      struct mptcp_out_options *opts)
739 {
740 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
741 
742 	if (!subflow->send_mp_prio)
743 		return false;
744 
745 	/* account for the trailing 'nop' option */
746 	if (remaining < TCPOLEN_MPTCP_PRIO_ALIGN)
747 		return false;
748 
749 	*size = TCPOLEN_MPTCP_PRIO_ALIGN;
750 	opts->suboptions |= OPTION_MPTCP_PRIO;
751 	opts->backup = subflow->request_bkup;
752 
753 	pr_debug("prio=%d", opts->backup);
754 
755 	return true;
756 }
757 
758 static noinline void mptcp_established_options_rst(struct sock *sk, struct sk_buff *skb,
759 						   unsigned int *size,
760 						   unsigned int remaining,
761 						   struct mptcp_out_options *opts)
762 {
763 	const struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
764 
765 	if (remaining < TCPOLEN_MPTCP_RST)
766 		return;
767 
768 	*size = TCPOLEN_MPTCP_RST;
769 	opts->suboptions |= OPTION_MPTCP_RST;
770 	opts->reset_transient = subflow->reset_transient;
771 	opts->reset_reason = subflow->reset_reason;
772 }
773 
774 bool mptcp_established_options(struct sock *sk, struct sk_buff *skb,
775 			       unsigned int *size, unsigned int remaining,
776 			       struct mptcp_out_options *opts)
777 {
778 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
779 	struct mptcp_sock *msk = mptcp_sk(subflow->conn);
780 	unsigned int opt_size = 0;
781 	bool snd_data_fin;
782 	bool ret = false;
783 
784 	opts->suboptions = 0;
785 
786 	if (unlikely(__mptcp_check_fallback(msk)))
787 		return false;
788 
789 	if (unlikely(skb && TCP_SKB_CB(skb)->tcp_flags & TCPHDR_RST)) {
790 		mptcp_established_options_rst(sk, skb, size, remaining, opts);
791 		return true;
792 	}
793 
794 	snd_data_fin = mptcp_data_fin_enabled(msk);
795 	if (mptcp_established_options_mp(sk, skb, snd_data_fin, &opt_size, remaining, opts))
796 		ret = true;
797 	else if (mptcp_established_options_dss(sk, skb, snd_data_fin, &opt_size, remaining, opts))
798 		ret = true;
799 
800 	/* we reserved enough space for the above options, and exceeding the
801 	 * TCP option space would be fatal
802 	 */
803 	if (WARN_ON_ONCE(opt_size > remaining))
804 		return false;
805 
806 	*size += opt_size;
807 	remaining -= opt_size;
808 	if (mptcp_established_options_add_addr(sk, skb, &opt_size, remaining, opts)) {
809 		*size += opt_size;
810 		remaining -= opt_size;
811 		ret = true;
812 	} else if (mptcp_established_options_rm_addr(sk, &opt_size, remaining, opts)) {
813 		*size += opt_size;
814 		remaining -= opt_size;
815 		ret = true;
816 	}
817 
818 	if (mptcp_established_options_mp_prio(sk, &opt_size, remaining, opts)) {
819 		*size += opt_size;
820 		remaining -= opt_size;
821 		ret = true;
822 	}
823 
824 	return ret;
825 }
826 
827 bool mptcp_synack_options(const struct request_sock *req, unsigned int *size,
828 			  struct mptcp_out_options *opts)
829 {
830 	struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
831 
832 	if (subflow_req->mp_capable) {
833 		opts->suboptions = OPTION_MPTCP_MPC_SYNACK;
834 		opts->sndr_key = subflow_req->local_key;
835 		opts->csum_reqd = subflow_req->csum_reqd;
836 		opts->allow_join_id0 = subflow_req->allow_join_id0;
837 		*size = TCPOLEN_MPTCP_MPC_SYNACK;
838 		pr_debug("subflow_req=%p, local_key=%llu",
839 			 subflow_req, subflow_req->local_key);
840 		return true;
841 	} else if (subflow_req->mp_join) {
842 		opts->suboptions = OPTION_MPTCP_MPJ_SYNACK;
843 		opts->backup = subflow_req->backup;
844 		opts->join_id = subflow_req->local_id;
845 		opts->thmac = subflow_req->thmac;
846 		opts->nonce = subflow_req->local_nonce;
847 		pr_debug("req=%p, bkup=%u, id=%u, thmac=%llu, nonce=%u",
848 			 subflow_req, opts->backup, opts->join_id,
849 			 opts->thmac, opts->nonce);
850 		*size = TCPOLEN_MPTCP_MPJ_SYNACK;
851 		return true;
852 	}
853 	return false;
854 }
855 
856 static bool check_fully_established(struct mptcp_sock *msk, struct sock *ssk,
857 				    struct mptcp_subflow_context *subflow,
858 				    struct sk_buff *skb,
859 				    struct mptcp_options_received *mp_opt)
860 {
861 	/* here we can process OoO, in-window pkts, only in-sequence 4th ack
862 	 * will make the subflow fully established
863 	 */
864 	if (likely(subflow->fully_established)) {
865 		/* on passive sockets, check for 3rd ack retransmission
866 		 * note that msk is always set by subflow_syn_recv_sock()
867 		 * for mp_join subflows
868 		 */
869 		if (TCP_SKB_CB(skb)->seq == subflow->ssn_offset + 1 &&
870 		    TCP_SKB_CB(skb)->end_seq == TCP_SKB_CB(skb)->seq &&
871 		    subflow->mp_join && mp_opt->mp_join &&
872 		    READ_ONCE(msk->pm.server_side))
873 			tcp_send_ack(ssk);
874 		goto fully_established;
875 	}
876 
877 	/* we must process OoO packets before the first subflow is fully
878 	 * established. OoO packets are instead a protocol violation
879 	 * for MP_JOIN subflows as the peer must not send any data
880 	 * before receiving the forth ack - cfr. RFC 8684 section 3.2.
881 	 */
882 	if (TCP_SKB_CB(skb)->seq != subflow->ssn_offset + 1) {
883 		if (subflow->mp_join)
884 			goto reset;
885 		return subflow->mp_capable;
886 	}
887 
888 	if (mp_opt->dss && mp_opt->use_ack) {
889 		/* subflows are fully established as soon as we get any
890 		 * additional ack.
891 		 */
892 		subflow->fully_established = 1;
893 		WRITE_ONCE(msk->fully_established, true);
894 		goto fully_established;
895 	}
896 
897 	if (mp_opt->add_addr) {
898 		WRITE_ONCE(msk->fully_established, true);
899 		return true;
900 	}
901 
902 	/* If the first established packet does not contain MP_CAPABLE + data
903 	 * then fallback to TCP. Fallback scenarios requires a reset for
904 	 * MP_JOIN subflows.
905 	 */
906 	if (!mp_opt->mp_capable) {
907 		if (subflow->mp_join)
908 			goto reset;
909 		subflow->mp_capable = 0;
910 		pr_fallback(msk);
911 		__mptcp_do_fallback(msk);
912 		return false;
913 	}
914 
915 	if (mp_opt->deny_join_id0)
916 		WRITE_ONCE(msk->pm.remote_deny_join_id0, true);
917 
918 	if (unlikely(!READ_ONCE(msk->pm.server_side)))
919 		pr_warn_once("bogus mpc option on established client sk");
920 	mptcp_subflow_fully_established(subflow, mp_opt);
921 
922 fully_established:
923 	/* if the subflow is not already linked into the conn_list, we can't
924 	 * notify the PM: this subflow is still on the listener queue
925 	 * and the PM possibly acquiring the subflow lock could race with
926 	 * the listener close
927 	 */
928 	if (likely(subflow->pm_notified) || list_empty(&subflow->node))
929 		return true;
930 
931 	subflow->pm_notified = 1;
932 	if (subflow->mp_join) {
933 		clear_3rdack_retransmission(ssk);
934 		mptcp_pm_subflow_established(msk);
935 	} else {
936 		mptcp_pm_fully_established(msk, ssk, GFP_ATOMIC);
937 	}
938 	return true;
939 
940 reset:
941 	mptcp_subflow_reset(ssk);
942 	return false;
943 }
944 
945 u64 __mptcp_expand_seq(u64 old_seq, u64 cur_seq)
946 {
947 	u32 old_seq32, cur_seq32;
948 
949 	old_seq32 = (u32)old_seq;
950 	cur_seq32 = (u32)cur_seq;
951 	cur_seq = (old_seq & GENMASK_ULL(63, 32)) + cur_seq32;
952 	if (unlikely(cur_seq32 < old_seq32 && before(old_seq32, cur_seq32)))
953 		return cur_seq + (1LL << 32);
954 
955 	/* reverse wrap could happen, too */
956 	if (unlikely(cur_seq32 > old_seq32 && after(old_seq32, cur_seq32)))
957 		return cur_seq - (1LL << 32);
958 	return cur_seq;
959 }
960 
961 static void ack_update_msk(struct mptcp_sock *msk,
962 			   struct sock *ssk,
963 			   struct mptcp_options_received *mp_opt)
964 {
965 	u64 new_wnd_end, new_snd_una, snd_nxt = READ_ONCE(msk->snd_nxt);
966 	struct sock *sk = (struct sock *)msk;
967 	u64 old_snd_una;
968 
969 	mptcp_data_lock(sk);
970 
971 	/* avoid ack expansion on update conflict, to reduce the risk of
972 	 * wrongly expanding to a future ack sequence number, which is way
973 	 * more dangerous than missing an ack
974 	 */
975 	old_snd_una = msk->snd_una;
976 	new_snd_una = mptcp_expand_seq(old_snd_una, mp_opt->data_ack, mp_opt->ack64);
977 
978 	/* ACK for data not even sent yet? Ignore. */
979 	if (after64(new_snd_una, snd_nxt))
980 		new_snd_una = old_snd_una;
981 
982 	new_wnd_end = new_snd_una + tcp_sk(ssk)->snd_wnd;
983 
984 	if (after64(new_wnd_end, msk->wnd_end))
985 		msk->wnd_end = new_wnd_end;
986 
987 	/* this assumes mptcp_incoming_options() is invoked after tcp_ack() */
988 	if (after64(msk->wnd_end, READ_ONCE(msk->snd_nxt)))
989 		__mptcp_check_push(sk, ssk);
990 
991 	if (after64(new_snd_una, old_snd_una)) {
992 		msk->snd_una = new_snd_una;
993 		__mptcp_data_acked(sk);
994 	}
995 	mptcp_data_unlock(sk);
996 
997 	trace_ack_update_msk(mp_opt->data_ack,
998 			     old_snd_una, new_snd_una,
999 			     new_wnd_end, msk->wnd_end);
1000 }
1001 
1002 bool mptcp_update_rcv_data_fin(struct mptcp_sock *msk, u64 data_fin_seq, bool use_64bit)
1003 {
1004 	/* Skip if DATA_FIN was already received.
1005 	 * If updating simultaneously with the recvmsg loop, values
1006 	 * should match. If they mismatch, the peer is misbehaving and
1007 	 * we will prefer the most recent information.
1008 	 */
1009 	if (READ_ONCE(msk->rcv_data_fin))
1010 		return false;
1011 
1012 	WRITE_ONCE(msk->rcv_data_fin_seq,
1013 		   mptcp_expand_seq(READ_ONCE(msk->ack_seq), data_fin_seq, use_64bit));
1014 	WRITE_ONCE(msk->rcv_data_fin, 1);
1015 
1016 	return true;
1017 }
1018 
1019 static bool add_addr_hmac_valid(struct mptcp_sock *msk,
1020 				struct mptcp_options_received *mp_opt)
1021 {
1022 	u64 hmac = 0;
1023 
1024 	if (mp_opt->echo)
1025 		return true;
1026 
1027 	hmac = add_addr_generate_hmac(msk->remote_key,
1028 				      msk->local_key,
1029 				      &mp_opt->addr);
1030 
1031 	pr_debug("msk=%p, ahmac=%llu, mp_opt->ahmac=%llu\n",
1032 		 msk, (unsigned long long)hmac,
1033 		 (unsigned long long)mp_opt->ahmac);
1034 
1035 	return hmac == mp_opt->ahmac;
1036 }
1037 
1038 void mptcp_incoming_options(struct sock *sk, struct sk_buff *skb)
1039 {
1040 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1041 	struct mptcp_sock *msk = mptcp_sk(subflow->conn);
1042 	struct mptcp_options_received mp_opt;
1043 	struct mptcp_ext *mpext;
1044 
1045 	if (__mptcp_check_fallback(msk)) {
1046 		/* Keep it simple and unconditionally trigger send data cleanup and
1047 		 * pending queue spooling. We will need to acquire the data lock
1048 		 * for more accurate checks, and once the lock is acquired, such
1049 		 * helpers are cheap.
1050 		 */
1051 		mptcp_data_lock(subflow->conn);
1052 		if (sk_stream_memory_free(sk))
1053 			__mptcp_check_push(subflow->conn, sk);
1054 		__mptcp_data_acked(subflow->conn);
1055 		mptcp_data_unlock(subflow->conn);
1056 		return;
1057 	}
1058 
1059 	mptcp_get_options(sk, skb, &mp_opt);
1060 	if (!check_fully_established(msk, sk, subflow, skb, &mp_opt))
1061 		return;
1062 
1063 	if (mp_opt.fastclose &&
1064 	    msk->local_key == mp_opt.rcvr_key) {
1065 		WRITE_ONCE(msk->rcv_fastclose, true);
1066 		mptcp_schedule_work((struct sock *)msk);
1067 	}
1068 
1069 	if (mp_opt.add_addr && add_addr_hmac_valid(msk, &mp_opt)) {
1070 		if (!mp_opt.echo) {
1071 			mptcp_pm_add_addr_received(msk, &mp_opt.addr);
1072 			MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ADDADDR);
1073 		} else {
1074 			mptcp_pm_add_addr_echoed(msk, &mp_opt.addr);
1075 			mptcp_pm_del_add_timer(msk, &mp_opt.addr, true);
1076 			MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ECHOADD);
1077 		}
1078 
1079 		if (mp_opt.addr.port)
1080 			MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_PORTADD);
1081 
1082 		mp_opt.add_addr = 0;
1083 	}
1084 
1085 	if (mp_opt.rm_addr) {
1086 		mptcp_pm_rm_addr_received(msk, &mp_opt.rm_list);
1087 		mp_opt.rm_addr = 0;
1088 	}
1089 
1090 	if (mp_opt.mp_prio) {
1091 		mptcp_pm_mp_prio_received(sk, mp_opt.backup);
1092 		MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPPRIORX);
1093 		mp_opt.mp_prio = 0;
1094 	}
1095 
1096 	if (mp_opt.reset) {
1097 		subflow->reset_seen = 1;
1098 		subflow->reset_reason = mp_opt.reset_reason;
1099 		subflow->reset_transient = mp_opt.reset_transient;
1100 	}
1101 
1102 	if (!mp_opt.dss)
1103 		return;
1104 
1105 	/* we can't wait for recvmsg() to update the ack_seq, otherwise
1106 	 * monodirectional flows will stuck
1107 	 */
1108 	if (mp_opt.use_ack)
1109 		ack_update_msk(msk, sk, &mp_opt);
1110 
1111 	/* Zero-data-length packets are dropped by the caller and not
1112 	 * propagated to the MPTCP layer, so the skb extension does not
1113 	 * need to be allocated or populated. DATA_FIN information, if
1114 	 * present, needs to be updated here before the skb is freed.
1115 	 */
1116 	if (TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb)->end_seq) {
1117 		if (mp_opt.data_fin && mp_opt.data_len == 1 &&
1118 		    mptcp_update_rcv_data_fin(msk, mp_opt.data_seq, mp_opt.dsn64) &&
1119 		    schedule_work(&msk->work))
1120 			sock_hold(subflow->conn);
1121 
1122 		return;
1123 	}
1124 
1125 	mpext = skb_ext_add(skb, SKB_EXT_MPTCP);
1126 	if (!mpext)
1127 		return;
1128 
1129 	memset(mpext, 0, sizeof(*mpext));
1130 
1131 	if (mp_opt.use_map) {
1132 		if (mp_opt.mpc_map) {
1133 			/* this is an MP_CAPABLE carrying MPTCP data
1134 			 * we know this map the first chunk of data
1135 			 */
1136 			mptcp_crypto_key_sha(subflow->remote_key, NULL,
1137 					     &mpext->data_seq);
1138 			mpext->data_seq++;
1139 			mpext->subflow_seq = 1;
1140 			mpext->dsn64 = 1;
1141 			mpext->mpc_map = 1;
1142 			mpext->data_fin = 0;
1143 		} else {
1144 			mpext->data_seq = mp_opt.data_seq;
1145 			mpext->subflow_seq = mp_opt.subflow_seq;
1146 			mpext->dsn64 = mp_opt.dsn64;
1147 			mpext->data_fin = mp_opt.data_fin;
1148 		}
1149 		mpext->data_len = mp_opt.data_len;
1150 		mpext->use_map = 1;
1151 		mpext->csum_reqd = mp_opt.csum_reqd;
1152 
1153 		if (mpext->csum_reqd)
1154 			mpext->csum = mp_opt.csum;
1155 	}
1156 }
1157 
1158 static void mptcp_set_rwin(const struct tcp_sock *tp)
1159 {
1160 	const struct sock *ssk = (const struct sock *)tp;
1161 	const struct mptcp_subflow_context *subflow;
1162 	struct mptcp_sock *msk;
1163 	u64 ack_seq;
1164 
1165 	subflow = mptcp_subflow_ctx(ssk);
1166 	msk = mptcp_sk(subflow->conn);
1167 
1168 	ack_seq = READ_ONCE(msk->ack_seq) + tp->rcv_wnd;
1169 
1170 	if (after64(ack_seq, READ_ONCE(msk->rcv_wnd_sent)))
1171 		WRITE_ONCE(msk->rcv_wnd_sent, ack_seq);
1172 }
1173 
1174 static u16 mptcp_make_csum(const struct mptcp_ext *mpext)
1175 {
1176 	struct csum_pseudo_header header;
1177 	__wsum csum;
1178 
1179 	/* cfr RFC 8684 3.3.1.:
1180 	 * the data sequence number used in the pseudo-header is
1181 	 * always the 64-bit value, irrespective of what length is used in the
1182 	 * DSS option itself.
1183 	 */
1184 	header.data_seq = cpu_to_be64(mpext->data_seq);
1185 	header.subflow_seq = htonl(mpext->subflow_seq);
1186 	header.data_len = htons(mpext->data_len);
1187 	header.csum = 0;
1188 
1189 	csum = csum_partial(&header, sizeof(header), ~csum_unfold(mpext->csum));
1190 	return (__force u16)csum_fold(csum);
1191 }
1192 
1193 void mptcp_write_options(__be32 *ptr, const struct tcp_sock *tp,
1194 			 struct mptcp_out_options *opts)
1195 {
1196 	if ((OPTION_MPTCP_MPC_SYN | OPTION_MPTCP_MPC_SYNACK |
1197 	     OPTION_MPTCP_MPC_ACK) & opts->suboptions) {
1198 		u8 len, flag = MPTCP_CAP_HMAC_SHA256;
1199 
1200 		if (OPTION_MPTCP_MPC_SYN & opts->suboptions) {
1201 			len = TCPOLEN_MPTCP_MPC_SYN;
1202 		} else if (OPTION_MPTCP_MPC_SYNACK & opts->suboptions) {
1203 			len = TCPOLEN_MPTCP_MPC_SYNACK;
1204 		} else if (opts->ext_copy.data_len) {
1205 			len = TCPOLEN_MPTCP_MPC_ACK_DATA;
1206 			if (opts->csum_reqd)
1207 				len += TCPOLEN_MPTCP_DSS_CHECKSUM;
1208 		} else {
1209 			len = TCPOLEN_MPTCP_MPC_ACK;
1210 		}
1211 
1212 		if (opts->csum_reqd)
1213 			flag |= MPTCP_CAP_CHECKSUM_REQD;
1214 
1215 		if (!opts->allow_join_id0)
1216 			flag |= MPTCP_CAP_DENY_JOIN_ID0;
1217 
1218 		*ptr++ = mptcp_option(MPTCPOPT_MP_CAPABLE, len,
1219 				      MPTCP_SUPPORTED_VERSION,
1220 				      flag);
1221 
1222 		if (!((OPTION_MPTCP_MPC_SYNACK | OPTION_MPTCP_MPC_ACK) &
1223 		    opts->suboptions))
1224 			goto mp_capable_done;
1225 
1226 		put_unaligned_be64(opts->sndr_key, ptr);
1227 		ptr += 2;
1228 		if (!((OPTION_MPTCP_MPC_ACK) & opts->suboptions))
1229 			goto mp_capable_done;
1230 
1231 		put_unaligned_be64(opts->rcvr_key, ptr);
1232 		ptr += 2;
1233 		if (!opts->ext_copy.data_len)
1234 			goto mp_capable_done;
1235 
1236 		if (opts->csum_reqd) {
1237 			put_unaligned_be32(opts->ext_copy.data_len << 16 |
1238 					   mptcp_make_csum(&opts->ext_copy), ptr);
1239 		} else {
1240 			put_unaligned_be32(opts->ext_copy.data_len << 16 |
1241 					   TCPOPT_NOP << 8 | TCPOPT_NOP, ptr);
1242 		}
1243 		ptr += 1;
1244 	}
1245 
1246 mp_capable_done:
1247 	if (OPTION_MPTCP_ADD_ADDR & opts->suboptions) {
1248 		u8 len = TCPOLEN_MPTCP_ADD_ADDR_BASE;
1249 		u8 echo = MPTCP_ADDR_ECHO;
1250 
1251 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1252 		if (opts->addr.family == AF_INET6)
1253 			len = TCPOLEN_MPTCP_ADD_ADDR6_BASE;
1254 #endif
1255 
1256 		if (opts->addr.port)
1257 			len += TCPOLEN_MPTCP_PORT_LEN;
1258 
1259 		if (opts->ahmac) {
1260 			len += sizeof(opts->ahmac);
1261 			echo = 0;
1262 		}
1263 
1264 		*ptr++ = mptcp_option(MPTCPOPT_ADD_ADDR,
1265 				      len, echo, opts->addr.id);
1266 		if (opts->addr.family == AF_INET) {
1267 			memcpy((u8 *)ptr, (u8 *)&opts->addr.addr.s_addr, 4);
1268 			ptr += 1;
1269 		}
1270 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1271 		else if (opts->addr.family == AF_INET6) {
1272 			memcpy((u8 *)ptr, opts->addr.addr6.s6_addr, 16);
1273 			ptr += 4;
1274 		}
1275 #endif
1276 
1277 		if (!opts->addr.port) {
1278 			if (opts->ahmac) {
1279 				put_unaligned_be64(opts->ahmac, ptr);
1280 				ptr += 2;
1281 			}
1282 		} else {
1283 			u16 port = ntohs(opts->addr.port);
1284 
1285 			if (opts->ahmac) {
1286 				u8 *bptr = (u8 *)ptr;
1287 
1288 				put_unaligned_be16(port, bptr);
1289 				bptr += 2;
1290 				put_unaligned_be64(opts->ahmac, bptr);
1291 				bptr += 8;
1292 				put_unaligned_be16(TCPOPT_NOP << 8 |
1293 						   TCPOPT_NOP, bptr);
1294 
1295 				ptr += 3;
1296 			} else {
1297 				put_unaligned_be32(port << 16 |
1298 						   TCPOPT_NOP << 8 |
1299 						   TCPOPT_NOP, ptr);
1300 				ptr += 1;
1301 			}
1302 		}
1303 	}
1304 
1305 	if (OPTION_MPTCP_RM_ADDR & opts->suboptions) {
1306 		u8 i = 1;
1307 
1308 		*ptr++ = mptcp_option(MPTCPOPT_RM_ADDR,
1309 				      TCPOLEN_MPTCP_RM_ADDR_BASE + opts->rm_list.nr,
1310 				      0, opts->rm_list.ids[0]);
1311 
1312 		while (i < opts->rm_list.nr) {
1313 			u8 id1, id2, id3, id4;
1314 
1315 			id1 = opts->rm_list.ids[i];
1316 			id2 = i + 1 < opts->rm_list.nr ? opts->rm_list.ids[i + 1] : TCPOPT_NOP;
1317 			id3 = i + 2 < opts->rm_list.nr ? opts->rm_list.ids[i + 2] : TCPOPT_NOP;
1318 			id4 = i + 3 < opts->rm_list.nr ? opts->rm_list.ids[i + 3] : TCPOPT_NOP;
1319 			put_unaligned_be32(id1 << 24 | id2 << 16 | id3 << 8 | id4, ptr);
1320 			ptr += 1;
1321 			i += 4;
1322 		}
1323 	}
1324 
1325 	if (OPTION_MPTCP_PRIO & opts->suboptions) {
1326 		const struct sock *ssk = (const struct sock *)tp;
1327 		struct mptcp_subflow_context *subflow;
1328 
1329 		subflow = mptcp_subflow_ctx(ssk);
1330 		subflow->send_mp_prio = 0;
1331 
1332 		*ptr++ = mptcp_option(MPTCPOPT_MP_PRIO,
1333 				      TCPOLEN_MPTCP_PRIO,
1334 				      opts->backup, TCPOPT_NOP);
1335 	}
1336 
1337 	if (OPTION_MPTCP_MPJ_SYN & opts->suboptions) {
1338 		*ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1339 				      TCPOLEN_MPTCP_MPJ_SYN,
1340 				      opts->backup, opts->join_id);
1341 		put_unaligned_be32(opts->token, ptr);
1342 		ptr += 1;
1343 		put_unaligned_be32(opts->nonce, ptr);
1344 		ptr += 1;
1345 	}
1346 
1347 	if (OPTION_MPTCP_MPJ_SYNACK & opts->suboptions) {
1348 		*ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1349 				      TCPOLEN_MPTCP_MPJ_SYNACK,
1350 				      opts->backup, opts->join_id);
1351 		put_unaligned_be64(opts->thmac, ptr);
1352 		ptr += 2;
1353 		put_unaligned_be32(opts->nonce, ptr);
1354 		ptr += 1;
1355 	}
1356 
1357 	if (OPTION_MPTCP_MPJ_ACK & opts->suboptions) {
1358 		*ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1359 				      TCPOLEN_MPTCP_MPJ_ACK, 0, 0);
1360 		memcpy(ptr, opts->hmac, MPTCPOPT_HMAC_LEN);
1361 		ptr += 5;
1362 	}
1363 
1364 	if (OPTION_MPTCP_RST & opts->suboptions)
1365 		*ptr++ = mptcp_option(MPTCPOPT_RST,
1366 				      TCPOLEN_MPTCP_RST,
1367 				      opts->reset_transient,
1368 				      opts->reset_reason);
1369 
1370 	if (opts->ext_copy.use_ack || opts->ext_copy.use_map) {
1371 		struct mptcp_ext *mpext = &opts->ext_copy;
1372 		u8 len = TCPOLEN_MPTCP_DSS_BASE;
1373 		u8 flags = 0;
1374 
1375 		if (mpext->use_ack) {
1376 			flags = MPTCP_DSS_HAS_ACK;
1377 			if (mpext->ack64) {
1378 				len += TCPOLEN_MPTCP_DSS_ACK64;
1379 				flags |= MPTCP_DSS_ACK64;
1380 			} else {
1381 				len += TCPOLEN_MPTCP_DSS_ACK32;
1382 			}
1383 		}
1384 
1385 		if (mpext->use_map) {
1386 			len += TCPOLEN_MPTCP_DSS_MAP64;
1387 
1388 			/* Use only 64-bit mapping flags for now, add
1389 			 * support for optional 32-bit mappings later.
1390 			 */
1391 			flags |= MPTCP_DSS_HAS_MAP | MPTCP_DSS_DSN64;
1392 			if (mpext->data_fin)
1393 				flags |= MPTCP_DSS_DATA_FIN;
1394 
1395 			if (opts->csum_reqd)
1396 				len += TCPOLEN_MPTCP_DSS_CHECKSUM;
1397 		}
1398 
1399 		*ptr++ = mptcp_option(MPTCPOPT_DSS, len, 0, flags);
1400 
1401 		if (mpext->use_ack) {
1402 			if (mpext->ack64) {
1403 				put_unaligned_be64(mpext->data_ack, ptr);
1404 				ptr += 2;
1405 			} else {
1406 				put_unaligned_be32(mpext->data_ack32, ptr);
1407 				ptr += 1;
1408 			}
1409 		}
1410 
1411 		if (mpext->use_map) {
1412 			put_unaligned_be64(mpext->data_seq, ptr);
1413 			ptr += 2;
1414 			put_unaligned_be32(mpext->subflow_seq, ptr);
1415 			ptr += 1;
1416 			if (opts->csum_reqd) {
1417 				put_unaligned_be32(mpext->data_len << 16 |
1418 						   mptcp_make_csum(mpext), ptr);
1419 			} else {
1420 				put_unaligned_be32(mpext->data_len << 16 |
1421 						   TCPOPT_NOP << 8 | TCPOPT_NOP, ptr);
1422 			}
1423 		}
1424 	}
1425 
1426 	if (tp)
1427 		mptcp_set_rwin(tp);
1428 }
1429 
1430 __be32 mptcp_get_reset_option(const struct sk_buff *skb)
1431 {
1432 	const struct mptcp_ext *ext = mptcp_get_ext(skb);
1433 	u8 flags, reason;
1434 
1435 	if (ext) {
1436 		flags = ext->reset_transient;
1437 		reason = ext->reset_reason;
1438 
1439 		return mptcp_option(MPTCPOPT_RST, TCPOLEN_MPTCP_RST,
1440 				    flags, reason);
1441 	}
1442 
1443 	return htonl(0u);
1444 }
1445 EXPORT_SYMBOL_GPL(mptcp_get_reset_option);
1446