xref: /linux/net/mptcp/options.c (revision 9052e9c95d908d6c3d7570aadc8898e1d871c8bb)
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 		break;
340 
341 	case MPTCPOPT_MP_FAIL:
342 		if (opsize != TCPOLEN_MPTCP_FAIL)
343 			break;
344 
345 		ptr += 2;
346 		mp_opt->suboptions |= OPTION_MPTCP_FAIL;
347 		mp_opt->fail_seq = get_unaligned_be64(ptr);
348 		pr_debug("MP_FAIL: data_seq=%llu", mp_opt->fail_seq);
349 		break;
350 
351 	default:
352 		break;
353 	}
354 }
355 
356 void mptcp_get_options(const struct sock *sk,
357 		       const struct sk_buff *skb,
358 		       struct mptcp_options_received *mp_opt)
359 {
360 	const struct tcphdr *th = tcp_hdr(skb);
361 	const unsigned char *ptr;
362 	int length;
363 
364 	/* initialize option status */
365 	mp_opt->suboptions = 0;
366 
367 	length = (th->doff * 4) - sizeof(struct tcphdr);
368 	ptr = (const unsigned char *)(th + 1);
369 
370 	while (length > 0) {
371 		int opcode = *ptr++;
372 		int opsize;
373 
374 		switch (opcode) {
375 		case TCPOPT_EOL:
376 			return;
377 		case TCPOPT_NOP:	/* Ref: RFC 793 section 3.1 */
378 			length--;
379 			continue;
380 		default:
381 			if (length < 2)
382 				return;
383 			opsize = *ptr++;
384 			if (opsize < 2) /* "silly options" */
385 				return;
386 			if (opsize > length)
387 				return;	/* don't parse partial options */
388 			if (opcode == TCPOPT_MPTCP)
389 				mptcp_parse_option(skb, ptr, opsize, mp_opt);
390 			ptr += opsize - 2;
391 			length -= opsize;
392 		}
393 	}
394 }
395 
396 bool mptcp_syn_options(struct sock *sk, const struct sk_buff *skb,
397 		       unsigned int *size, struct mptcp_out_options *opts)
398 {
399 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
400 
401 	/* we will use snd_isn to detect first pkt [re]transmission
402 	 * in mptcp_established_options_mp()
403 	 */
404 	subflow->snd_isn = TCP_SKB_CB(skb)->end_seq;
405 	if (subflow->request_mptcp) {
406 		opts->suboptions = OPTION_MPTCP_MPC_SYN;
407 		opts->csum_reqd = mptcp_is_checksum_enabled(sock_net(sk));
408 		opts->allow_join_id0 = mptcp_allow_join_id0(sock_net(sk));
409 		*size = TCPOLEN_MPTCP_MPC_SYN;
410 		return true;
411 	} else if (subflow->request_join) {
412 		pr_debug("remote_token=%u, nonce=%u", subflow->remote_token,
413 			 subflow->local_nonce);
414 		opts->suboptions = OPTION_MPTCP_MPJ_SYN;
415 		opts->join_id = subflow->local_id;
416 		opts->token = subflow->remote_token;
417 		opts->nonce = subflow->local_nonce;
418 		opts->backup = subflow->request_bkup;
419 		*size = TCPOLEN_MPTCP_MPJ_SYN;
420 		return true;
421 	}
422 	return false;
423 }
424 
425 /* MP_JOIN client subflow must wait for 4th ack before sending any data:
426  * TCP can't schedule delack timer before the subflow is fully established.
427  * MPTCP uses the delack timer to do 3rd ack retransmissions
428  */
429 static void schedule_3rdack_retransmission(struct sock *sk)
430 {
431 	struct inet_connection_sock *icsk = inet_csk(sk);
432 	struct tcp_sock *tp = tcp_sk(sk);
433 	unsigned long timeout;
434 
435 	/* reschedule with a timeout above RTT, as we must look only for drop */
436 	if (tp->srtt_us)
437 		timeout = tp->srtt_us << 1;
438 	else
439 		timeout = TCP_TIMEOUT_INIT;
440 
441 	WARN_ON_ONCE(icsk->icsk_ack.pending & ICSK_ACK_TIMER);
442 	icsk->icsk_ack.pending |= ICSK_ACK_SCHED | ICSK_ACK_TIMER;
443 	icsk->icsk_ack.timeout = timeout;
444 	sk_reset_timer(sk, &icsk->icsk_delack_timer, timeout);
445 }
446 
447 static void clear_3rdack_retransmission(struct sock *sk)
448 {
449 	struct inet_connection_sock *icsk = inet_csk(sk);
450 
451 	sk_stop_timer(sk, &icsk->icsk_delack_timer);
452 	icsk->icsk_ack.timeout = 0;
453 	icsk->icsk_ack.ato = 0;
454 	icsk->icsk_ack.pending &= ~(ICSK_ACK_SCHED | ICSK_ACK_TIMER);
455 }
456 
457 static bool mptcp_established_options_mp(struct sock *sk, struct sk_buff *skb,
458 					 bool snd_data_fin_enable,
459 					 unsigned int *size,
460 					 unsigned int remaining,
461 					 struct mptcp_out_options *opts)
462 {
463 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
464 	struct mptcp_sock *msk = mptcp_sk(subflow->conn);
465 	struct mptcp_ext *mpext;
466 	unsigned int data_len;
467 	u8 len;
468 
469 	/* When skb is not available, we better over-estimate the emitted
470 	 * options len. A full DSS option (28 bytes) is longer than
471 	 * TCPOLEN_MPTCP_MPC_ACK_DATA(22) or TCPOLEN_MPTCP_MPJ_ACK(24), so
472 	 * tell the caller to defer the estimate to
473 	 * mptcp_established_options_dss(), which will reserve enough space.
474 	 */
475 	if (!skb)
476 		return false;
477 
478 	/* MPC/MPJ needed only on 3rd ack packet, DATA_FIN and TCP shutdown take precedence */
479 	if (subflow->fully_established || snd_data_fin_enable ||
480 	    subflow->snd_isn != TCP_SKB_CB(skb)->seq ||
481 	    sk->sk_state != TCP_ESTABLISHED)
482 		return false;
483 
484 	if (subflow->mp_capable) {
485 		mpext = mptcp_get_ext(skb);
486 		data_len = mpext ? mpext->data_len : 0;
487 
488 		/* we will check ext_copy.data_len in mptcp_write_options() to
489 		 * discriminate between TCPOLEN_MPTCP_MPC_ACK_DATA and
490 		 * TCPOLEN_MPTCP_MPC_ACK
491 		 */
492 		opts->ext_copy.data_len = data_len;
493 		opts->suboptions = OPTION_MPTCP_MPC_ACK;
494 		opts->sndr_key = subflow->local_key;
495 		opts->rcvr_key = subflow->remote_key;
496 		opts->csum_reqd = READ_ONCE(msk->csum_enabled);
497 		opts->allow_join_id0 = mptcp_allow_join_id0(sock_net(sk));
498 
499 		/* Section 3.1.
500 		 * The MP_CAPABLE option is carried on the SYN, SYN/ACK, and ACK
501 		 * packets that start the first subflow of an MPTCP connection,
502 		 * as well as the first packet that carries data
503 		 */
504 		if (data_len > 0) {
505 			len = TCPOLEN_MPTCP_MPC_ACK_DATA;
506 			if (opts->csum_reqd) {
507 				/* we need to propagate more info to csum the pseudo hdr */
508 				opts->ext_copy.data_seq = mpext->data_seq;
509 				opts->ext_copy.subflow_seq = mpext->subflow_seq;
510 				opts->ext_copy.csum = mpext->csum;
511 				len += TCPOLEN_MPTCP_DSS_CHECKSUM;
512 			}
513 			*size = ALIGN(len, 4);
514 		} else {
515 			*size = TCPOLEN_MPTCP_MPC_ACK;
516 		}
517 
518 		pr_debug("subflow=%p, local_key=%llu, remote_key=%llu map_len=%d",
519 			 subflow, subflow->local_key, subflow->remote_key,
520 			 data_len);
521 
522 		return true;
523 	} else if (subflow->mp_join) {
524 		opts->suboptions = OPTION_MPTCP_MPJ_ACK;
525 		memcpy(opts->hmac, subflow->hmac, MPTCPOPT_HMAC_LEN);
526 		*size = TCPOLEN_MPTCP_MPJ_ACK;
527 		pr_debug("subflow=%p", subflow);
528 
529 		schedule_3rdack_retransmission(sk);
530 		return true;
531 	}
532 	return false;
533 }
534 
535 static void mptcp_write_data_fin(struct mptcp_subflow_context *subflow,
536 				 struct sk_buff *skb, struct mptcp_ext *ext)
537 {
538 	/* The write_seq value has already been incremented, so the actual
539 	 * sequence number for the DATA_FIN is one less.
540 	 */
541 	u64 data_fin_tx_seq = READ_ONCE(mptcp_sk(subflow->conn)->write_seq) - 1;
542 
543 	if (!ext->use_map || !skb->len) {
544 		/* RFC6824 requires a DSS mapping with specific values
545 		 * if DATA_FIN is set but no data payload is mapped
546 		 */
547 		ext->data_fin = 1;
548 		ext->use_map = 1;
549 		ext->dsn64 = 1;
550 		ext->data_seq = data_fin_tx_seq;
551 		ext->subflow_seq = 0;
552 		ext->data_len = 1;
553 	} else if (ext->data_seq + ext->data_len == data_fin_tx_seq) {
554 		/* If there's an existing DSS mapping and it is the
555 		 * final mapping, DATA_FIN consumes 1 additional byte of
556 		 * mapping space.
557 		 */
558 		ext->data_fin = 1;
559 		ext->data_len++;
560 	}
561 }
562 
563 static bool mptcp_established_options_dss(struct sock *sk, struct sk_buff *skb,
564 					  bool snd_data_fin_enable,
565 					  unsigned int *size,
566 					  unsigned int remaining,
567 					  struct mptcp_out_options *opts)
568 {
569 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
570 	struct mptcp_sock *msk = mptcp_sk(subflow->conn);
571 	unsigned int dss_size = 0;
572 	struct mptcp_ext *mpext;
573 	unsigned int ack_size;
574 	bool ret = false;
575 	u64 ack_seq;
576 
577 	opts->csum_reqd = READ_ONCE(msk->csum_enabled);
578 	mpext = skb ? mptcp_get_ext(skb) : NULL;
579 
580 	if (!skb || (mpext && mpext->use_map) || snd_data_fin_enable) {
581 		unsigned int map_size = TCPOLEN_MPTCP_DSS_BASE + TCPOLEN_MPTCP_DSS_MAP64;
582 
583 		if (mpext) {
584 			if (opts->csum_reqd)
585 				map_size += TCPOLEN_MPTCP_DSS_CHECKSUM;
586 
587 			opts->ext_copy = *mpext;
588 		}
589 
590 		remaining -= map_size;
591 		dss_size = map_size;
592 		if (skb && snd_data_fin_enable)
593 			mptcp_write_data_fin(subflow, skb, &opts->ext_copy);
594 		opts->suboptions = OPTION_MPTCP_DSS;
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 	opts->suboptions = OPTION_MPTCP_DSS;
619 	WRITE_ONCE(msk->old_wspace, __mptcp_space((struct sock *)msk));
620 
621 	/* Add kind/length/subtype/flag overhead if mapping is not populated */
622 	if (dss_size == 0)
623 		ack_size += TCPOLEN_MPTCP_DSS_BASE;
624 
625 	dss_size += ack_size;
626 
627 	*size = ALIGN(dss_size, 4);
628 	return true;
629 }
630 
631 static u64 add_addr_generate_hmac(u64 key1, u64 key2,
632 				  struct mptcp_addr_info *addr)
633 {
634 	u16 port = ntohs(addr->port);
635 	u8 hmac[SHA256_DIGEST_SIZE];
636 	u8 msg[19];
637 	int i = 0;
638 
639 	msg[i++] = addr->id;
640 	if (addr->family == AF_INET) {
641 		memcpy(&msg[i], &addr->addr.s_addr, 4);
642 		i += 4;
643 	}
644 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
645 	else if (addr->family == AF_INET6) {
646 		memcpy(&msg[i], &addr->addr6.s6_addr, 16);
647 		i += 16;
648 	}
649 #endif
650 	msg[i++] = port >> 8;
651 	msg[i++] = port & 0xFF;
652 
653 	mptcp_crypto_hmac_sha(key1, key2, msg, i, hmac);
654 
655 	return get_unaligned_be64(&hmac[SHA256_DIGEST_SIZE - sizeof(u64)]);
656 }
657 
658 static bool mptcp_established_options_add_addr(struct sock *sk, struct sk_buff *skb,
659 					       unsigned int *size,
660 					       unsigned int remaining,
661 					       struct mptcp_out_options *opts)
662 {
663 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
664 	struct mptcp_sock *msk = mptcp_sk(subflow->conn);
665 	bool drop_other_suboptions = false;
666 	unsigned int opt_size = *size;
667 	bool echo;
668 	bool port;
669 	int len;
670 
671 	/* add addr will strip the existing options, be sure to avoid breaking
672 	 * MPC/MPJ handshakes
673 	 */
674 	if (!mptcp_pm_should_add_signal(msk) ||
675 	    (opts->suboptions & (OPTION_MPTCP_MPJ_ACK | OPTION_MPTCP_MPC_ACK)) ||
676 	    !mptcp_pm_add_addr_signal(msk, skb, opt_size, remaining, &opts->addr,
677 		    &echo, &port, &drop_other_suboptions))
678 		return false;
679 
680 	if (drop_other_suboptions)
681 		remaining += opt_size;
682 	len = mptcp_add_addr_len(opts->addr.family, echo, port);
683 	if (remaining < len)
684 		return false;
685 
686 	*size = len;
687 	if (drop_other_suboptions) {
688 		pr_debug("drop other suboptions");
689 		opts->suboptions = 0;
690 
691 		/* note that e.g. DSS could have written into the memory
692 		 * aliased by ahmac, we must reset the field here
693 		 * to avoid appending the hmac even for ADD_ADDR echo
694 		 * options
695 		 */
696 		opts->ahmac = 0;
697 		*size -= opt_size;
698 	}
699 	opts->suboptions |= OPTION_MPTCP_ADD_ADDR;
700 	if (!echo) {
701 		opts->ahmac = add_addr_generate_hmac(msk->local_key,
702 						     msk->remote_key,
703 						     &opts->addr);
704 	}
705 	pr_debug("addr_id=%d, ahmac=%llu, echo=%d, port=%d",
706 		 opts->addr.id, opts->ahmac, echo, ntohs(opts->addr.port));
707 
708 	return true;
709 }
710 
711 static bool mptcp_established_options_rm_addr(struct sock *sk,
712 					      unsigned int *size,
713 					      unsigned int remaining,
714 					      struct mptcp_out_options *opts)
715 {
716 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
717 	struct mptcp_sock *msk = mptcp_sk(subflow->conn);
718 	struct mptcp_rm_list rm_list;
719 	int i, len;
720 
721 	if (!mptcp_pm_should_rm_signal(msk) ||
722 	    !(mptcp_pm_rm_addr_signal(msk, remaining, &rm_list)))
723 		return false;
724 
725 	len = mptcp_rm_addr_len(&rm_list);
726 	if (len < 0)
727 		return false;
728 	if (remaining < len)
729 		return false;
730 
731 	*size = len;
732 	opts->suboptions |= OPTION_MPTCP_RM_ADDR;
733 	opts->rm_list = rm_list;
734 
735 	for (i = 0; i < opts->rm_list.nr; i++)
736 		pr_debug("rm_list_ids[%d]=%d", i, opts->rm_list.ids[i]);
737 
738 	return true;
739 }
740 
741 static bool mptcp_established_options_mp_prio(struct sock *sk,
742 					      unsigned int *size,
743 					      unsigned int remaining,
744 					      struct mptcp_out_options *opts)
745 {
746 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
747 
748 	/* can't send MP_PRIO with MPC, as they share the same option space:
749 	 * 'backup'. Also it makes no sense at all
750 	 */
751 	if (!subflow->send_mp_prio ||
752 	    ((OPTION_MPTCP_MPC_SYN | OPTION_MPTCP_MPC_SYNACK |
753 	      OPTION_MPTCP_MPC_ACK) & opts->suboptions))
754 		return false;
755 
756 	/* account for the trailing 'nop' option */
757 	if (remaining < TCPOLEN_MPTCP_PRIO_ALIGN)
758 		return false;
759 
760 	*size = TCPOLEN_MPTCP_PRIO_ALIGN;
761 	opts->suboptions |= OPTION_MPTCP_PRIO;
762 	opts->backup = subflow->request_bkup;
763 
764 	pr_debug("prio=%d", opts->backup);
765 
766 	return true;
767 }
768 
769 static noinline bool mptcp_established_options_rst(struct sock *sk, struct sk_buff *skb,
770 						   unsigned int *size,
771 						   unsigned int remaining,
772 						   struct mptcp_out_options *opts)
773 {
774 	const struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
775 
776 	if (remaining < TCPOLEN_MPTCP_RST)
777 		return false;
778 
779 	*size = TCPOLEN_MPTCP_RST;
780 	opts->suboptions |= OPTION_MPTCP_RST;
781 	opts->reset_transient = subflow->reset_transient;
782 	opts->reset_reason = subflow->reset_reason;
783 
784 	return true;
785 }
786 
787 static bool mptcp_established_options_mp_fail(struct sock *sk,
788 					      unsigned int *size,
789 					      unsigned int remaining,
790 					      struct mptcp_out_options *opts)
791 {
792 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
793 
794 	if (likely(!subflow->send_mp_fail))
795 		return false;
796 
797 	if (remaining < TCPOLEN_MPTCP_FAIL)
798 		return false;
799 
800 	*size = TCPOLEN_MPTCP_FAIL;
801 	opts->suboptions |= OPTION_MPTCP_FAIL;
802 	opts->fail_seq = subflow->map_seq;
803 
804 	pr_debug("MP_FAIL fail_seq=%llu", opts->fail_seq);
805 
806 	return true;
807 }
808 
809 bool mptcp_established_options(struct sock *sk, struct sk_buff *skb,
810 			       unsigned int *size, unsigned int remaining,
811 			       struct mptcp_out_options *opts)
812 {
813 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
814 	struct mptcp_sock *msk = mptcp_sk(subflow->conn);
815 	unsigned int opt_size = 0;
816 	bool snd_data_fin;
817 	bool ret = false;
818 
819 	opts->suboptions = 0;
820 
821 	if (unlikely(__mptcp_check_fallback(msk)))
822 		return false;
823 
824 	if (unlikely(skb && TCP_SKB_CB(skb)->tcp_flags & TCPHDR_RST)) {
825 		if (mptcp_established_options_mp_fail(sk, &opt_size, remaining, opts)) {
826 			*size += opt_size;
827 			remaining -= opt_size;
828 		}
829 		if (mptcp_established_options_rst(sk, skb, &opt_size, remaining, opts)) {
830 			*size += opt_size;
831 			remaining -= opt_size;
832 		}
833 		return true;
834 	}
835 
836 	snd_data_fin = mptcp_data_fin_enabled(msk);
837 	if (mptcp_established_options_mp(sk, skb, snd_data_fin, &opt_size, remaining, opts))
838 		ret = true;
839 	else if (mptcp_established_options_dss(sk, skb, snd_data_fin, &opt_size, remaining, opts)) {
840 		ret = true;
841 		if (mptcp_established_options_mp_fail(sk, &opt_size, remaining, opts)) {
842 			*size += opt_size;
843 			remaining -= opt_size;
844 			return true;
845 		}
846 	}
847 
848 	/* we reserved enough space for the above options, and exceeding the
849 	 * TCP option space would be fatal
850 	 */
851 	if (WARN_ON_ONCE(opt_size > remaining))
852 		return false;
853 
854 	*size += opt_size;
855 	remaining -= opt_size;
856 	if (mptcp_established_options_add_addr(sk, skb, &opt_size, remaining, opts)) {
857 		*size += opt_size;
858 		remaining -= opt_size;
859 		ret = true;
860 	} else if (mptcp_established_options_rm_addr(sk, &opt_size, remaining, opts)) {
861 		*size += opt_size;
862 		remaining -= opt_size;
863 		ret = true;
864 	}
865 
866 	if (mptcp_established_options_mp_prio(sk, &opt_size, remaining, opts)) {
867 		*size += opt_size;
868 		remaining -= opt_size;
869 		ret = true;
870 	}
871 
872 	return ret;
873 }
874 
875 bool mptcp_synack_options(const struct request_sock *req, unsigned int *size,
876 			  struct mptcp_out_options *opts)
877 {
878 	struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
879 
880 	if (subflow_req->mp_capable) {
881 		opts->suboptions = OPTION_MPTCP_MPC_SYNACK;
882 		opts->sndr_key = subflow_req->local_key;
883 		opts->csum_reqd = subflow_req->csum_reqd;
884 		opts->allow_join_id0 = subflow_req->allow_join_id0;
885 		*size = TCPOLEN_MPTCP_MPC_SYNACK;
886 		pr_debug("subflow_req=%p, local_key=%llu",
887 			 subflow_req, subflow_req->local_key);
888 		return true;
889 	} else if (subflow_req->mp_join) {
890 		opts->suboptions = OPTION_MPTCP_MPJ_SYNACK;
891 		opts->backup = subflow_req->backup;
892 		opts->join_id = subflow_req->local_id;
893 		opts->thmac = subflow_req->thmac;
894 		opts->nonce = subflow_req->local_nonce;
895 		pr_debug("req=%p, bkup=%u, id=%u, thmac=%llu, nonce=%u",
896 			 subflow_req, opts->backup, opts->join_id,
897 			 opts->thmac, opts->nonce);
898 		*size = TCPOLEN_MPTCP_MPJ_SYNACK;
899 		return true;
900 	}
901 	return false;
902 }
903 
904 static bool check_fully_established(struct mptcp_sock *msk, struct sock *ssk,
905 				    struct mptcp_subflow_context *subflow,
906 				    struct sk_buff *skb,
907 				    struct mptcp_options_received *mp_opt)
908 {
909 	/* here we can process OoO, in-window pkts, only in-sequence 4th ack
910 	 * will make the subflow fully established
911 	 */
912 	if (likely(subflow->fully_established)) {
913 		/* on passive sockets, check for 3rd ack retransmission
914 		 * note that msk is always set by subflow_syn_recv_sock()
915 		 * for mp_join subflows
916 		 */
917 		if (TCP_SKB_CB(skb)->seq == subflow->ssn_offset + 1 &&
918 		    TCP_SKB_CB(skb)->end_seq == TCP_SKB_CB(skb)->seq &&
919 		    subflow->mp_join && (mp_opt->suboptions & OPTIONS_MPTCP_MPJ) &&
920 		    READ_ONCE(msk->pm.server_side))
921 			tcp_send_ack(ssk);
922 		goto fully_established;
923 	}
924 
925 	/* we must process OoO packets before the first subflow is fully
926 	 * established. OoO packets are instead a protocol violation
927 	 * for MP_JOIN subflows as the peer must not send any data
928 	 * before receiving the forth ack - cfr. RFC 8684 section 3.2.
929 	 */
930 	if (TCP_SKB_CB(skb)->seq != subflow->ssn_offset + 1) {
931 		if (subflow->mp_join)
932 			goto reset;
933 		return subflow->mp_capable;
934 	}
935 
936 	if (((mp_opt->suboptions & OPTION_MPTCP_DSS) && mp_opt->use_ack) ||
937 	    ((mp_opt->suboptions & OPTION_MPTCP_ADD_ADDR) && !mp_opt->echo)) {
938 		/* subflows are fully established as soon as we get any
939 		 * additional ack, including ADD_ADDR.
940 		 */
941 		subflow->fully_established = 1;
942 		WRITE_ONCE(msk->fully_established, true);
943 		goto fully_established;
944 	}
945 
946 	/* If the first established packet does not contain MP_CAPABLE + data
947 	 * then fallback to TCP. Fallback scenarios requires a reset for
948 	 * MP_JOIN subflows.
949 	 */
950 	if (!(mp_opt->suboptions & OPTIONS_MPTCP_MPC)) {
951 		if (subflow->mp_join)
952 			goto reset;
953 		subflow->mp_capable = 0;
954 		pr_fallback(msk);
955 		__mptcp_do_fallback(msk);
956 		return false;
957 	}
958 
959 	if (mp_opt->deny_join_id0)
960 		WRITE_ONCE(msk->pm.remote_deny_join_id0, true);
961 
962 	if (unlikely(!READ_ONCE(msk->pm.server_side)))
963 		pr_warn_once("bogus mpc option on established client sk");
964 	mptcp_subflow_fully_established(subflow, mp_opt);
965 
966 fully_established:
967 	/* if the subflow is not already linked into the conn_list, we can't
968 	 * notify the PM: this subflow is still on the listener queue
969 	 * and the PM possibly acquiring the subflow lock could race with
970 	 * the listener close
971 	 */
972 	if (likely(subflow->pm_notified) || list_empty(&subflow->node))
973 		return true;
974 
975 	subflow->pm_notified = 1;
976 	if (subflow->mp_join) {
977 		clear_3rdack_retransmission(ssk);
978 		mptcp_pm_subflow_established(msk);
979 	} else {
980 		mptcp_pm_fully_established(msk, ssk, GFP_ATOMIC);
981 	}
982 	return true;
983 
984 reset:
985 	mptcp_subflow_reset(ssk);
986 	return false;
987 }
988 
989 u64 __mptcp_expand_seq(u64 old_seq, u64 cur_seq)
990 {
991 	u32 old_seq32, cur_seq32;
992 
993 	old_seq32 = (u32)old_seq;
994 	cur_seq32 = (u32)cur_seq;
995 	cur_seq = (old_seq & GENMASK_ULL(63, 32)) + cur_seq32;
996 	if (unlikely(cur_seq32 < old_seq32 && before(old_seq32, cur_seq32)))
997 		return cur_seq + (1LL << 32);
998 
999 	/* reverse wrap could happen, too */
1000 	if (unlikely(cur_seq32 > old_seq32 && after(old_seq32, cur_seq32)))
1001 		return cur_seq - (1LL << 32);
1002 	return cur_seq;
1003 }
1004 
1005 static void ack_update_msk(struct mptcp_sock *msk,
1006 			   struct sock *ssk,
1007 			   struct mptcp_options_received *mp_opt)
1008 {
1009 	u64 new_wnd_end, new_snd_una, snd_nxt = READ_ONCE(msk->snd_nxt);
1010 	struct sock *sk = (struct sock *)msk;
1011 	u64 old_snd_una;
1012 
1013 	mptcp_data_lock(sk);
1014 
1015 	/* avoid ack expansion on update conflict, to reduce the risk of
1016 	 * wrongly expanding to a future ack sequence number, which is way
1017 	 * more dangerous than missing an ack
1018 	 */
1019 	old_snd_una = msk->snd_una;
1020 	new_snd_una = mptcp_expand_seq(old_snd_una, mp_opt->data_ack, mp_opt->ack64);
1021 
1022 	/* ACK for data not even sent yet and even above recovery bound? Ignore.*/
1023 	if (unlikely(after64(new_snd_una, snd_nxt))) {
1024 		if (!msk->recovery || after64(new_snd_una, msk->recovery_snd_nxt))
1025 			new_snd_una = old_snd_una;
1026 	}
1027 
1028 	new_wnd_end = new_snd_una + tcp_sk(ssk)->snd_wnd;
1029 
1030 	if (after64(new_wnd_end, msk->wnd_end))
1031 		msk->wnd_end = new_wnd_end;
1032 
1033 	/* this assumes mptcp_incoming_options() is invoked after tcp_ack() */
1034 	if (after64(msk->wnd_end, READ_ONCE(msk->snd_nxt)))
1035 		__mptcp_check_push(sk, ssk);
1036 
1037 	if (after64(new_snd_una, old_snd_una)) {
1038 		msk->snd_una = new_snd_una;
1039 		__mptcp_data_acked(sk);
1040 	}
1041 	mptcp_data_unlock(sk);
1042 
1043 	trace_ack_update_msk(mp_opt->data_ack,
1044 			     old_snd_una, new_snd_una,
1045 			     new_wnd_end, msk->wnd_end);
1046 }
1047 
1048 bool mptcp_update_rcv_data_fin(struct mptcp_sock *msk, u64 data_fin_seq, bool use_64bit)
1049 {
1050 	/* Skip if DATA_FIN was already received.
1051 	 * If updating simultaneously with the recvmsg loop, values
1052 	 * should match. If they mismatch, the peer is misbehaving and
1053 	 * we will prefer the most recent information.
1054 	 */
1055 	if (READ_ONCE(msk->rcv_data_fin))
1056 		return false;
1057 
1058 	WRITE_ONCE(msk->rcv_data_fin_seq,
1059 		   mptcp_expand_seq(READ_ONCE(msk->ack_seq), data_fin_seq, use_64bit));
1060 	WRITE_ONCE(msk->rcv_data_fin, 1);
1061 
1062 	return true;
1063 }
1064 
1065 static bool add_addr_hmac_valid(struct mptcp_sock *msk,
1066 				struct mptcp_options_received *mp_opt)
1067 {
1068 	u64 hmac = 0;
1069 
1070 	if (mp_opt->echo)
1071 		return true;
1072 
1073 	hmac = add_addr_generate_hmac(msk->remote_key,
1074 				      msk->local_key,
1075 				      &mp_opt->addr);
1076 
1077 	pr_debug("msk=%p, ahmac=%llu, mp_opt->ahmac=%llu\n",
1078 		 msk, (unsigned long long)hmac,
1079 		 (unsigned long long)mp_opt->ahmac);
1080 
1081 	return hmac == mp_opt->ahmac;
1082 }
1083 
1084 /* Return false if a subflow has been reset, else return true */
1085 bool mptcp_incoming_options(struct sock *sk, struct sk_buff *skb)
1086 {
1087 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1088 	struct mptcp_sock *msk = mptcp_sk(subflow->conn);
1089 	struct mptcp_options_received mp_opt;
1090 	struct mptcp_ext *mpext;
1091 
1092 	if (__mptcp_check_fallback(msk)) {
1093 		/* Keep it simple and unconditionally trigger send data cleanup and
1094 		 * pending queue spooling. We will need to acquire the data lock
1095 		 * for more accurate checks, and once the lock is acquired, such
1096 		 * helpers are cheap.
1097 		 */
1098 		mptcp_data_lock(subflow->conn);
1099 		if (sk_stream_memory_free(sk))
1100 			__mptcp_check_push(subflow->conn, sk);
1101 		__mptcp_data_acked(subflow->conn);
1102 		mptcp_data_unlock(subflow->conn);
1103 		return true;
1104 	}
1105 
1106 	mptcp_get_options(sk, skb, &mp_opt);
1107 
1108 	/* The subflow can be in close state only if check_fully_established()
1109 	 * just sent a reset. If so, tell the caller to ignore the current packet.
1110 	 */
1111 	if (!check_fully_established(msk, sk, subflow, skb, &mp_opt))
1112 		return sk->sk_state != TCP_CLOSE;
1113 
1114 	if (unlikely(mp_opt.suboptions != OPTION_MPTCP_DSS)) {
1115 		if ((mp_opt.suboptions & OPTION_MPTCP_FASTCLOSE) &&
1116 		    msk->local_key == mp_opt.rcvr_key) {
1117 			WRITE_ONCE(msk->rcv_fastclose, true);
1118 			mptcp_schedule_work((struct sock *)msk);
1119 		}
1120 
1121 		if ((mp_opt.suboptions & OPTION_MPTCP_ADD_ADDR) &&
1122 		    add_addr_hmac_valid(msk, &mp_opt)) {
1123 			if (!mp_opt.echo) {
1124 				mptcp_pm_add_addr_received(msk, &mp_opt.addr);
1125 				MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ADDADDR);
1126 			} else {
1127 				mptcp_pm_add_addr_echoed(msk, &mp_opt.addr);
1128 				mptcp_pm_del_add_timer(msk, &mp_opt.addr, true);
1129 				MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ECHOADD);
1130 			}
1131 
1132 			if (mp_opt.addr.port)
1133 				MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_PORTADD);
1134 		}
1135 
1136 		if (mp_opt.suboptions & OPTION_MPTCP_RM_ADDR)
1137 			mptcp_pm_rm_addr_received(msk, &mp_opt.rm_list);
1138 
1139 		if (mp_opt.suboptions & OPTION_MPTCP_PRIO) {
1140 			mptcp_pm_mp_prio_received(sk, mp_opt.backup);
1141 			MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPPRIORX);
1142 		}
1143 
1144 		if (mp_opt.suboptions & OPTION_MPTCP_FAIL) {
1145 			mptcp_pm_mp_fail_received(sk, mp_opt.fail_seq);
1146 			MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPFAILRX);
1147 		}
1148 
1149 		if (mp_opt.suboptions & OPTION_MPTCP_RST) {
1150 			subflow->reset_seen = 1;
1151 			subflow->reset_reason = mp_opt.reset_reason;
1152 			subflow->reset_transient = mp_opt.reset_transient;
1153 		}
1154 
1155 		if (!(mp_opt.suboptions & OPTION_MPTCP_DSS))
1156 			return true;
1157 	}
1158 
1159 	/* we can't wait for recvmsg() to update the ack_seq, otherwise
1160 	 * monodirectional flows will stuck
1161 	 */
1162 	if (mp_opt.use_ack)
1163 		ack_update_msk(msk, sk, &mp_opt);
1164 
1165 	/* Zero-data-length packets are dropped by the caller and not
1166 	 * propagated to the MPTCP layer, so the skb extension does not
1167 	 * need to be allocated or populated. DATA_FIN information, if
1168 	 * present, needs to be updated here before the skb is freed.
1169 	 */
1170 	if (TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb)->end_seq) {
1171 		if (mp_opt.data_fin && mp_opt.data_len == 1 &&
1172 		    mptcp_update_rcv_data_fin(msk, mp_opt.data_seq, mp_opt.dsn64) &&
1173 		    schedule_work(&msk->work))
1174 			sock_hold(subflow->conn);
1175 
1176 		return true;
1177 	}
1178 
1179 	mpext = skb_ext_add(skb, SKB_EXT_MPTCP);
1180 	if (!mpext)
1181 		return true;
1182 
1183 	memset(mpext, 0, sizeof(*mpext));
1184 
1185 	if (likely(mp_opt.use_map)) {
1186 		if (mp_opt.mpc_map) {
1187 			/* this is an MP_CAPABLE carrying MPTCP data
1188 			 * we know this map the first chunk of data
1189 			 */
1190 			mptcp_crypto_key_sha(subflow->remote_key, NULL,
1191 					     &mpext->data_seq);
1192 			mpext->data_seq++;
1193 			mpext->subflow_seq = 1;
1194 			mpext->dsn64 = 1;
1195 			mpext->mpc_map = 1;
1196 			mpext->data_fin = 0;
1197 		} else {
1198 			mpext->data_seq = mp_opt.data_seq;
1199 			mpext->subflow_seq = mp_opt.subflow_seq;
1200 			mpext->dsn64 = mp_opt.dsn64;
1201 			mpext->data_fin = mp_opt.data_fin;
1202 		}
1203 		mpext->data_len = mp_opt.data_len;
1204 		mpext->use_map = 1;
1205 		mpext->csum_reqd = !!(mp_opt.suboptions & OPTION_MPTCP_CSUMREQD);
1206 
1207 		if (mpext->csum_reqd)
1208 			mpext->csum = mp_opt.csum;
1209 	}
1210 
1211 	return true;
1212 }
1213 
1214 static void mptcp_set_rwin(const struct tcp_sock *tp)
1215 {
1216 	const struct sock *ssk = (const struct sock *)tp;
1217 	const struct mptcp_subflow_context *subflow;
1218 	struct mptcp_sock *msk;
1219 	u64 ack_seq;
1220 
1221 	subflow = mptcp_subflow_ctx(ssk);
1222 	msk = mptcp_sk(subflow->conn);
1223 
1224 	ack_seq = READ_ONCE(msk->ack_seq) + tp->rcv_wnd;
1225 
1226 	if (after64(ack_seq, READ_ONCE(msk->rcv_wnd_sent)))
1227 		WRITE_ONCE(msk->rcv_wnd_sent, ack_seq);
1228 }
1229 
1230 static u16 mptcp_make_csum(const struct mptcp_ext *mpext)
1231 {
1232 	struct csum_pseudo_header header;
1233 	__wsum csum;
1234 
1235 	/* cfr RFC 8684 3.3.1.:
1236 	 * the data sequence number used in the pseudo-header is
1237 	 * always the 64-bit value, irrespective of what length is used in the
1238 	 * DSS option itself.
1239 	 */
1240 	header.data_seq = cpu_to_be64(mpext->data_seq);
1241 	header.subflow_seq = htonl(mpext->subflow_seq);
1242 	header.data_len = htons(mpext->data_len);
1243 	header.csum = 0;
1244 
1245 	csum = csum_partial(&header, sizeof(header), ~csum_unfold(mpext->csum));
1246 	return (__force u16)csum_fold(csum);
1247 }
1248 
1249 void mptcp_write_options(__be32 *ptr, const struct tcp_sock *tp,
1250 			 struct mptcp_out_options *opts)
1251 {
1252 	if (unlikely(OPTION_MPTCP_FAIL & opts->suboptions)) {
1253 		const struct sock *ssk = (const struct sock *)tp;
1254 		struct mptcp_subflow_context *subflow;
1255 
1256 		subflow = mptcp_subflow_ctx(ssk);
1257 		subflow->send_mp_fail = 0;
1258 
1259 		*ptr++ = mptcp_option(MPTCPOPT_MP_FAIL,
1260 				      TCPOLEN_MPTCP_FAIL,
1261 				      0, 0);
1262 		put_unaligned_be64(opts->fail_seq, ptr);
1263 		ptr += 2;
1264 	}
1265 
1266 	/* RST is mutually exclusive with everything else */
1267 	if (unlikely(OPTION_MPTCP_RST & opts->suboptions)) {
1268 		*ptr++ = mptcp_option(MPTCPOPT_RST,
1269 				      TCPOLEN_MPTCP_RST,
1270 				      opts->reset_transient,
1271 				      opts->reset_reason);
1272 		return;
1273 	}
1274 
1275 	/* DSS, MPC, MPJ and ADD_ADDR are mutually exclusive, see
1276 	 * mptcp_established_options*()
1277 	 */
1278 	if (likely(OPTION_MPTCP_DSS & opts->suboptions)) {
1279 		struct mptcp_ext *mpext = &opts->ext_copy;
1280 		u8 len = TCPOLEN_MPTCP_DSS_BASE;
1281 		u8 flags = 0;
1282 
1283 		if (mpext->use_ack) {
1284 			flags = MPTCP_DSS_HAS_ACK;
1285 			if (mpext->ack64) {
1286 				len += TCPOLEN_MPTCP_DSS_ACK64;
1287 				flags |= MPTCP_DSS_ACK64;
1288 			} else {
1289 				len += TCPOLEN_MPTCP_DSS_ACK32;
1290 			}
1291 		}
1292 
1293 		if (mpext->use_map) {
1294 			len += TCPOLEN_MPTCP_DSS_MAP64;
1295 
1296 			/* Use only 64-bit mapping flags for now, add
1297 			 * support for optional 32-bit mappings later.
1298 			 */
1299 			flags |= MPTCP_DSS_HAS_MAP | MPTCP_DSS_DSN64;
1300 			if (mpext->data_fin)
1301 				flags |= MPTCP_DSS_DATA_FIN;
1302 
1303 			if (opts->csum_reqd)
1304 				len += TCPOLEN_MPTCP_DSS_CHECKSUM;
1305 		}
1306 
1307 		*ptr++ = mptcp_option(MPTCPOPT_DSS, len, 0, flags);
1308 
1309 		if (mpext->use_ack) {
1310 			if (mpext->ack64) {
1311 				put_unaligned_be64(mpext->data_ack, ptr);
1312 				ptr += 2;
1313 			} else {
1314 				put_unaligned_be32(mpext->data_ack32, ptr);
1315 				ptr += 1;
1316 			}
1317 		}
1318 
1319 		if (mpext->use_map) {
1320 			put_unaligned_be64(mpext->data_seq, ptr);
1321 			ptr += 2;
1322 			put_unaligned_be32(mpext->subflow_seq, ptr);
1323 			ptr += 1;
1324 			if (opts->csum_reqd) {
1325 				put_unaligned_be32(mpext->data_len << 16 |
1326 						   mptcp_make_csum(mpext), ptr);
1327 			} else {
1328 				put_unaligned_be32(mpext->data_len << 16 |
1329 						   TCPOPT_NOP << 8 | TCPOPT_NOP, ptr);
1330 			}
1331 		}
1332 	} else if ((OPTION_MPTCP_MPC_SYN | OPTION_MPTCP_MPC_SYNACK |
1333 		    OPTION_MPTCP_MPC_ACK) & opts->suboptions) {
1334 		u8 len, flag = MPTCP_CAP_HMAC_SHA256;
1335 
1336 		if (OPTION_MPTCP_MPC_SYN & opts->suboptions) {
1337 			len = TCPOLEN_MPTCP_MPC_SYN;
1338 		} else if (OPTION_MPTCP_MPC_SYNACK & opts->suboptions) {
1339 			len = TCPOLEN_MPTCP_MPC_SYNACK;
1340 		} else if (opts->ext_copy.data_len) {
1341 			len = TCPOLEN_MPTCP_MPC_ACK_DATA;
1342 			if (opts->csum_reqd)
1343 				len += TCPOLEN_MPTCP_DSS_CHECKSUM;
1344 		} else {
1345 			len = TCPOLEN_MPTCP_MPC_ACK;
1346 		}
1347 
1348 		if (opts->csum_reqd)
1349 			flag |= MPTCP_CAP_CHECKSUM_REQD;
1350 
1351 		if (!opts->allow_join_id0)
1352 			flag |= MPTCP_CAP_DENY_JOIN_ID0;
1353 
1354 		*ptr++ = mptcp_option(MPTCPOPT_MP_CAPABLE, len,
1355 				      MPTCP_SUPPORTED_VERSION,
1356 				      flag);
1357 
1358 		if (!((OPTION_MPTCP_MPC_SYNACK | OPTION_MPTCP_MPC_ACK) &
1359 		    opts->suboptions))
1360 			goto mp_capable_done;
1361 
1362 		put_unaligned_be64(opts->sndr_key, ptr);
1363 		ptr += 2;
1364 		if (!((OPTION_MPTCP_MPC_ACK) & opts->suboptions))
1365 			goto mp_capable_done;
1366 
1367 		put_unaligned_be64(opts->rcvr_key, ptr);
1368 		ptr += 2;
1369 		if (!opts->ext_copy.data_len)
1370 			goto mp_capable_done;
1371 
1372 		if (opts->csum_reqd) {
1373 			put_unaligned_be32(opts->ext_copy.data_len << 16 |
1374 					   mptcp_make_csum(&opts->ext_copy), ptr);
1375 		} else {
1376 			put_unaligned_be32(opts->ext_copy.data_len << 16 |
1377 					   TCPOPT_NOP << 8 | TCPOPT_NOP, ptr);
1378 		}
1379 		ptr += 1;
1380 
1381 		/* MPC is additionally mutually exclusive with MP_PRIO */
1382 		goto mp_capable_done;
1383 	} else if (OPTION_MPTCP_MPJ_SYN & opts->suboptions) {
1384 		*ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1385 				      TCPOLEN_MPTCP_MPJ_SYN,
1386 				      opts->backup, opts->join_id);
1387 		put_unaligned_be32(opts->token, ptr);
1388 		ptr += 1;
1389 		put_unaligned_be32(opts->nonce, ptr);
1390 		ptr += 1;
1391 	} else if (OPTION_MPTCP_MPJ_SYNACK & opts->suboptions) {
1392 		*ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1393 				      TCPOLEN_MPTCP_MPJ_SYNACK,
1394 				      opts->backup, opts->join_id);
1395 		put_unaligned_be64(opts->thmac, ptr);
1396 		ptr += 2;
1397 		put_unaligned_be32(opts->nonce, ptr);
1398 		ptr += 1;
1399 	} else if (OPTION_MPTCP_MPJ_ACK & opts->suboptions) {
1400 		*ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1401 				      TCPOLEN_MPTCP_MPJ_ACK, 0, 0);
1402 		memcpy(ptr, opts->hmac, MPTCPOPT_HMAC_LEN);
1403 		ptr += 5;
1404 	} else if (OPTION_MPTCP_ADD_ADDR & opts->suboptions) {
1405 		u8 len = TCPOLEN_MPTCP_ADD_ADDR_BASE;
1406 		u8 echo = MPTCP_ADDR_ECHO;
1407 
1408 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1409 		if (opts->addr.family == AF_INET6)
1410 			len = TCPOLEN_MPTCP_ADD_ADDR6_BASE;
1411 #endif
1412 
1413 		if (opts->addr.port)
1414 			len += TCPOLEN_MPTCP_PORT_LEN;
1415 
1416 		if (opts->ahmac) {
1417 			len += sizeof(opts->ahmac);
1418 			echo = 0;
1419 		}
1420 
1421 		*ptr++ = mptcp_option(MPTCPOPT_ADD_ADDR,
1422 				      len, echo, opts->addr.id);
1423 		if (opts->addr.family == AF_INET) {
1424 			memcpy((u8 *)ptr, (u8 *)&opts->addr.addr.s_addr, 4);
1425 			ptr += 1;
1426 		}
1427 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1428 		else if (opts->addr.family == AF_INET6) {
1429 			memcpy((u8 *)ptr, opts->addr.addr6.s6_addr, 16);
1430 			ptr += 4;
1431 		}
1432 #endif
1433 
1434 		if (!opts->addr.port) {
1435 			if (opts->ahmac) {
1436 				put_unaligned_be64(opts->ahmac, ptr);
1437 				ptr += 2;
1438 			}
1439 		} else {
1440 			u16 port = ntohs(opts->addr.port);
1441 
1442 			if (opts->ahmac) {
1443 				u8 *bptr = (u8 *)ptr;
1444 
1445 				put_unaligned_be16(port, bptr);
1446 				bptr += 2;
1447 				put_unaligned_be64(opts->ahmac, bptr);
1448 				bptr += 8;
1449 				put_unaligned_be16(TCPOPT_NOP << 8 |
1450 						   TCPOPT_NOP, bptr);
1451 
1452 				ptr += 3;
1453 			} else {
1454 				put_unaligned_be32(port << 16 |
1455 						   TCPOPT_NOP << 8 |
1456 						   TCPOPT_NOP, ptr);
1457 				ptr += 1;
1458 			}
1459 		}
1460 	}
1461 
1462 	if (OPTION_MPTCP_PRIO & opts->suboptions) {
1463 		const struct sock *ssk = (const struct sock *)tp;
1464 		struct mptcp_subflow_context *subflow;
1465 
1466 		subflow = mptcp_subflow_ctx(ssk);
1467 		subflow->send_mp_prio = 0;
1468 
1469 		*ptr++ = mptcp_option(MPTCPOPT_MP_PRIO,
1470 				      TCPOLEN_MPTCP_PRIO,
1471 				      opts->backup, TCPOPT_NOP);
1472 	}
1473 
1474 mp_capable_done:
1475 	if (OPTION_MPTCP_RM_ADDR & opts->suboptions) {
1476 		u8 i = 1;
1477 
1478 		*ptr++ = mptcp_option(MPTCPOPT_RM_ADDR,
1479 				      TCPOLEN_MPTCP_RM_ADDR_BASE + opts->rm_list.nr,
1480 				      0, opts->rm_list.ids[0]);
1481 
1482 		while (i < opts->rm_list.nr) {
1483 			u8 id1, id2, id3, id4;
1484 
1485 			id1 = opts->rm_list.ids[i];
1486 			id2 = i + 1 < opts->rm_list.nr ? opts->rm_list.ids[i + 1] : TCPOPT_NOP;
1487 			id3 = i + 2 < opts->rm_list.nr ? opts->rm_list.ids[i + 2] : TCPOPT_NOP;
1488 			id4 = i + 3 < opts->rm_list.nr ? opts->rm_list.ids[i + 3] : TCPOPT_NOP;
1489 			put_unaligned_be32(id1 << 24 | id2 << 16 | id3 << 8 | id4, ptr);
1490 			ptr += 1;
1491 			i += 4;
1492 		}
1493 	}
1494 
1495 	if (tp)
1496 		mptcp_set_rwin(tp);
1497 }
1498 
1499 __be32 mptcp_get_reset_option(const struct sk_buff *skb)
1500 {
1501 	const struct mptcp_ext *ext = mptcp_get_ext(skb);
1502 	u8 flags, reason;
1503 
1504 	if (ext) {
1505 		flags = ext->reset_transient;
1506 		reason = ext->reset_reason;
1507 
1508 		return mptcp_option(MPTCPOPT_RST, TCPOLEN_MPTCP_RST,
1509 				    flags, reason);
1510 	}
1511 
1512 	return htonl(0u);
1513 }
1514 EXPORT_SYMBOL_GPL(mptcp_get_reset_option);
1515