xref: /openbmc/linux/net/mptcp/options.c (revision 745b7908)
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 	u16 subopt;
30 	u8 version;
31 	u8 flags;
32 	u8 i;
33 
34 	switch (subtype) {
35 	case MPTCPOPT_MP_CAPABLE:
36 		/* strict size checking */
37 		if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)) {
38 			if (skb->len > tcp_hdr(skb)->doff << 2)
39 				expected_opsize = TCPOLEN_MPTCP_MPC_ACK_DATA;
40 			else
41 				expected_opsize = TCPOLEN_MPTCP_MPC_ACK;
42 			subopt = OPTION_MPTCP_MPC_ACK;
43 		} else {
44 			if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_ACK) {
45 				expected_opsize = TCPOLEN_MPTCP_MPC_SYNACK;
46 				subopt = OPTION_MPTCP_MPC_SYNACK;
47 			} else {
48 				expected_opsize = TCPOLEN_MPTCP_MPC_SYN;
49 				subopt = OPTION_MPTCP_MPC_SYN;
50 			}
51 		}
52 
53 		/* Cfr RFC 8684 Section 3.3.0:
54 		 * If a checksum is present but its use had
55 		 * not been negotiated in the MP_CAPABLE handshake, the receiver MUST
56 		 * close the subflow with a RST, as it is not behaving as negotiated.
57 		 * If a checksum is not present when its use has been negotiated, the
58 		 * receiver MUST close the subflow with a RST, as it is considered
59 		 * broken
60 		 * We parse even option with mismatching csum presence, so that
61 		 * later in subflow_data_ready we can trigger the reset.
62 		 */
63 		if (opsize != expected_opsize &&
64 		    (expected_opsize != TCPOLEN_MPTCP_MPC_ACK_DATA ||
65 		     opsize != TCPOLEN_MPTCP_MPC_ACK_DATA_CSUM))
66 			break;
67 
68 		/* try to be gentle vs future versions on the initial syn */
69 		version = *ptr++ & MPTCP_VERSION_MASK;
70 		if (opsize != TCPOLEN_MPTCP_MPC_SYN) {
71 			if (version != MPTCP_SUPPORTED_VERSION)
72 				break;
73 		} else if (version < MPTCP_SUPPORTED_VERSION) {
74 			break;
75 		}
76 
77 		flags = *ptr++;
78 		if (!mptcp_cap_flag_sha256(flags) ||
79 		    (flags & MPTCP_CAP_EXTENSIBILITY))
80 			break;
81 
82 		/* RFC 6824, Section 3.1:
83 		 * "For the Checksum Required bit (labeled "A"), if either
84 		 * host requires the use of checksums, checksums MUST be used.
85 		 * In other words, the only way for checksums not to be used
86 		 * is if both hosts in their SYNs set A=0."
87 		 */
88 		if (flags & MPTCP_CAP_CHECKSUM_REQD)
89 			mp_opt->suboptions |= OPTION_MPTCP_CSUMREQD;
90 
91 		mp_opt->deny_join_id0 = !!(flags & MPTCP_CAP_DENY_JOIN_ID0);
92 
93 		mp_opt->suboptions |= subopt;
94 		if (opsize >= TCPOLEN_MPTCP_MPC_SYNACK) {
95 			mp_opt->sndr_key = get_unaligned_be64(ptr);
96 			ptr += 8;
97 		}
98 		if (opsize >= TCPOLEN_MPTCP_MPC_ACK) {
99 			mp_opt->rcvr_key = get_unaligned_be64(ptr);
100 			ptr += 8;
101 		}
102 		if (opsize >= TCPOLEN_MPTCP_MPC_ACK_DATA) {
103 			/* Section 3.1.:
104 			 * "the data parameters in a MP_CAPABLE are semantically
105 			 * equivalent to those in a DSS option and can be used
106 			 * interchangeably."
107 			 */
108 			mp_opt->suboptions |= OPTION_MPTCP_DSS;
109 			mp_opt->use_map = 1;
110 			mp_opt->mpc_map = 1;
111 			mp_opt->data_len = get_unaligned_be16(ptr);
112 			ptr += 2;
113 		}
114 		if (opsize == TCPOLEN_MPTCP_MPC_ACK_DATA_CSUM) {
115 			mp_opt->csum = get_unaligned((__force __sum16 *)ptr);
116 			mp_opt->suboptions |= OPTION_MPTCP_CSUMREQD;
117 			ptr += 2;
118 		}
119 		pr_debug("MP_CAPABLE version=%x, flags=%x, optlen=%d sndr=%llu, rcvr=%llu len=%d csum=%u",
120 			 version, flags, opsize, mp_opt->sndr_key,
121 			 mp_opt->rcvr_key, mp_opt->data_len, mp_opt->csum);
122 		break;
123 
124 	case MPTCPOPT_MP_JOIN:
125 		mp_opt->suboptions |= OPTIONS_MPTCP_MPJ;
126 		if (opsize == TCPOLEN_MPTCP_MPJ_SYN) {
127 			mp_opt->backup = *ptr++ & MPTCPOPT_BACKUP;
128 			mp_opt->join_id = *ptr++;
129 			mp_opt->token = get_unaligned_be32(ptr);
130 			ptr += 4;
131 			mp_opt->nonce = get_unaligned_be32(ptr);
132 			ptr += 4;
133 			pr_debug("MP_JOIN bkup=%u, id=%u, token=%u, nonce=%u",
134 				 mp_opt->backup, mp_opt->join_id,
135 				 mp_opt->token, mp_opt->nonce);
136 		} else if (opsize == TCPOLEN_MPTCP_MPJ_SYNACK) {
137 			mp_opt->backup = *ptr++ & MPTCPOPT_BACKUP;
138 			mp_opt->join_id = *ptr++;
139 			mp_opt->thmac = get_unaligned_be64(ptr);
140 			ptr += 8;
141 			mp_opt->nonce = get_unaligned_be32(ptr);
142 			ptr += 4;
143 			pr_debug("MP_JOIN bkup=%u, id=%u, thmac=%llu, nonce=%u",
144 				 mp_opt->backup, mp_opt->join_id,
145 				 mp_opt->thmac, mp_opt->nonce);
146 		} else if (opsize == TCPOLEN_MPTCP_MPJ_ACK) {
147 			ptr += 2;
148 			memcpy(mp_opt->hmac, ptr, MPTCPOPT_HMAC_LEN);
149 			pr_debug("MP_JOIN hmac");
150 		} else {
151 			mp_opt->suboptions &= ~OPTIONS_MPTCP_MPJ;
152 		}
153 		break;
154 
155 	case MPTCPOPT_DSS:
156 		pr_debug("DSS");
157 		ptr++;
158 
159 		/* we must clear 'mpc_map' be able to detect MP_CAPABLE
160 		 * map vs DSS map in mptcp_incoming_options(), and reconstruct
161 		 * map info accordingly
162 		 */
163 		mp_opt->mpc_map = 0;
164 		flags = (*ptr++) & MPTCP_DSS_FLAG_MASK;
165 		mp_opt->data_fin = (flags & MPTCP_DSS_DATA_FIN) != 0;
166 		mp_opt->dsn64 = (flags & MPTCP_DSS_DSN64) != 0;
167 		mp_opt->use_map = (flags & MPTCP_DSS_HAS_MAP) != 0;
168 		mp_opt->ack64 = (flags & MPTCP_DSS_ACK64) != 0;
169 		mp_opt->use_ack = (flags & MPTCP_DSS_HAS_ACK);
170 
171 		pr_debug("data_fin=%d dsn64=%d use_map=%d ack64=%d use_ack=%d",
172 			 mp_opt->data_fin, mp_opt->dsn64,
173 			 mp_opt->use_map, mp_opt->ack64,
174 			 mp_opt->use_ack);
175 
176 		expected_opsize = TCPOLEN_MPTCP_DSS_BASE;
177 
178 		if (mp_opt->use_ack) {
179 			if (mp_opt->ack64)
180 				expected_opsize += TCPOLEN_MPTCP_DSS_ACK64;
181 			else
182 				expected_opsize += TCPOLEN_MPTCP_DSS_ACK32;
183 		}
184 
185 		if (mp_opt->use_map) {
186 			if (mp_opt->dsn64)
187 				expected_opsize += TCPOLEN_MPTCP_DSS_MAP64;
188 			else
189 				expected_opsize += TCPOLEN_MPTCP_DSS_MAP32;
190 		}
191 
192 		/* Always parse any csum presence combination, we will enforce
193 		 * RFC 8684 Section 3.3.0 checks later in subflow_data_ready
194 		 */
195 		if (opsize != expected_opsize &&
196 		    opsize != expected_opsize + TCPOLEN_MPTCP_DSS_CHECKSUM)
197 			break;
198 
199 		mp_opt->suboptions |= OPTION_MPTCP_DSS;
200 		if (mp_opt->use_ack) {
201 			if (mp_opt->ack64) {
202 				mp_opt->data_ack = get_unaligned_be64(ptr);
203 				ptr += 8;
204 			} else {
205 				mp_opt->data_ack = get_unaligned_be32(ptr);
206 				ptr += 4;
207 			}
208 
209 			pr_debug("data_ack=%llu", mp_opt->data_ack);
210 		}
211 
212 		if (mp_opt->use_map) {
213 			if (mp_opt->dsn64) {
214 				mp_opt->data_seq = get_unaligned_be64(ptr);
215 				ptr += 8;
216 			} else {
217 				mp_opt->data_seq = get_unaligned_be32(ptr);
218 				ptr += 4;
219 			}
220 
221 			mp_opt->subflow_seq = get_unaligned_be32(ptr);
222 			ptr += 4;
223 
224 			mp_opt->data_len = get_unaligned_be16(ptr);
225 			ptr += 2;
226 
227 			if (opsize == expected_opsize + TCPOLEN_MPTCP_DSS_CHECKSUM) {
228 				mp_opt->suboptions |= OPTION_MPTCP_CSUMREQD;
229 				mp_opt->csum = get_unaligned((__force __sum16 *)ptr);
230 				ptr += 2;
231 			}
232 
233 			pr_debug("data_seq=%llu subflow_seq=%u data_len=%u csum=%d:%u",
234 				 mp_opt->data_seq, mp_opt->subflow_seq,
235 				 mp_opt->data_len, !!(mp_opt->suboptions & OPTION_MPTCP_CSUMREQD),
236 				 mp_opt->csum);
237 		}
238 
239 		break;
240 
241 	case MPTCPOPT_ADD_ADDR:
242 		mp_opt->echo = (*ptr++) & MPTCP_ADDR_ECHO;
243 		if (!mp_opt->echo) {
244 			if (opsize == TCPOLEN_MPTCP_ADD_ADDR ||
245 			    opsize == TCPOLEN_MPTCP_ADD_ADDR_PORT)
246 				mp_opt->addr.family = AF_INET;
247 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
248 			else if (opsize == TCPOLEN_MPTCP_ADD_ADDR6 ||
249 				 opsize == TCPOLEN_MPTCP_ADD_ADDR6_PORT)
250 				mp_opt->addr.family = AF_INET6;
251 #endif
252 			else
253 				break;
254 		} else {
255 			if (opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE ||
256 			    opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE_PORT)
257 				mp_opt->addr.family = AF_INET;
258 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
259 			else if (opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE ||
260 				 opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE_PORT)
261 				mp_opt->addr.family = AF_INET6;
262 #endif
263 			else
264 				break;
265 		}
266 
267 		mp_opt->suboptions |= OPTION_MPTCP_ADD_ADDR;
268 		mp_opt->addr.id = *ptr++;
269 		mp_opt->addr.port = 0;
270 		mp_opt->ahmac = 0;
271 		if (mp_opt->addr.family == AF_INET) {
272 			memcpy((u8 *)&mp_opt->addr.addr.s_addr, (u8 *)ptr, 4);
273 			ptr += 4;
274 			if (opsize == TCPOLEN_MPTCP_ADD_ADDR_PORT ||
275 			    opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE_PORT) {
276 				mp_opt->addr.port = htons(get_unaligned_be16(ptr));
277 				ptr += 2;
278 			}
279 		}
280 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
281 		else {
282 			memcpy(mp_opt->addr.addr6.s6_addr, (u8 *)ptr, 16);
283 			ptr += 16;
284 			if (opsize == TCPOLEN_MPTCP_ADD_ADDR6_PORT ||
285 			    opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE_PORT) {
286 				mp_opt->addr.port = htons(get_unaligned_be16(ptr));
287 				ptr += 2;
288 			}
289 		}
290 #endif
291 		if (!mp_opt->echo) {
292 			mp_opt->ahmac = get_unaligned_be64(ptr);
293 			ptr += 8;
294 		}
295 		pr_debug("ADD_ADDR%s: id=%d, ahmac=%llu, echo=%d, port=%d",
296 			 (mp_opt->addr.family == AF_INET6) ? "6" : "",
297 			 mp_opt->addr.id, mp_opt->ahmac, mp_opt->echo, ntohs(mp_opt->addr.port));
298 		break;
299 
300 	case MPTCPOPT_RM_ADDR:
301 		if (opsize < TCPOLEN_MPTCP_RM_ADDR_BASE + 1 ||
302 		    opsize > TCPOLEN_MPTCP_RM_ADDR_BASE + MPTCP_RM_IDS_MAX)
303 			break;
304 
305 		ptr++;
306 
307 		mp_opt->suboptions |= OPTION_MPTCP_RM_ADDR;
308 		mp_opt->rm_list.nr = opsize - TCPOLEN_MPTCP_RM_ADDR_BASE;
309 		for (i = 0; i < mp_opt->rm_list.nr; i++)
310 			mp_opt->rm_list.ids[i] = *ptr++;
311 		pr_debug("RM_ADDR: rm_list_nr=%d", mp_opt->rm_list.nr);
312 		break;
313 
314 	case MPTCPOPT_MP_PRIO:
315 		if (opsize != TCPOLEN_MPTCP_PRIO)
316 			break;
317 
318 		mp_opt->suboptions |= OPTION_MPTCP_PRIO;
319 		mp_opt->backup = *ptr++ & MPTCP_PRIO_BKUP;
320 		pr_debug("MP_PRIO: prio=%d", mp_opt->backup);
321 		break;
322 
323 	case MPTCPOPT_MP_FASTCLOSE:
324 		if (opsize != TCPOLEN_MPTCP_FASTCLOSE)
325 			break;
326 
327 		ptr += 2;
328 		mp_opt->rcvr_key = get_unaligned_be64(ptr);
329 		ptr += 8;
330 		mp_opt->suboptions |= OPTION_MPTCP_FASTCLOSE;
331 		pr_debug("MP_FASTCLOSE: recv_key=%llu", mp_opt->rcvr_key);
332 		break;
333 
334 	case MPTCPOPT_RST:
335 		if (opsize != TCPOLEN_MPTCP_RST)
336 			break;
337 
338 		if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_RST))
339 			break;
340 
341 		mp_opt->suboptions |= OPTION_MPTCP_RST;
342 		flags = *ptr++;
343 		mp_opt->reset_transient = flags & MPTCP_RST_TRANSIENT;
344 		mp_opt->reset_reason = *ptr;
345 		pr_debug("MP_RST: transient=%u reason=%u",
346 			 mp_opt->reset_transient, mp_opt->reset_reason);
347 		break;
348 
349 	case MPTCPOPT_MP_FAIL:
350 		if (opsize != TCPOLEN_MPTCP_FAIL)
351 			break;
352 
353 		ptr += 2;
354 		mp_opt->suboptions |= OPTION_MPTCP_FAIL;
355 		mp_opt->fail_seq = get_unaligned_be64(ptr);
356 		pr_debug("MP_FAIL: data_seq=%llu", mp_opt->fail_seq);
357 		break;
358 
359 	default:
360 		break;
361 	}
362 }
363 
364 void mptcp_get_options(const struct sk_buff *skb,
365 		       struct mptcp_options_received *mp_opt)
366 {
367 	const struct tcphdr *th = tcp_hdr(skb);
368 	const unsigned char *ptr;
369 	int length;
370 
371 	/* initialize option status */
372 	mp_opt->suboptions = 0;
373 
374 	length = (th->doff * 4) - sizeof(struct tcphdr);
375 	ptr = (const unsigned char *)(th + 1);
376 
377 	while (length > 0) {
378 		int opcode = *ptr++;
379 		int opsize;
380 
381 		switch (opcode) {
382 		case TCPOPT_EOL:
383 			return;
384 		case TCPOPT_NOP:	/* Ref: RFC 793 section 3.1 */
385 			length--;
386 			continue;
387 		default:
388 			if (length < 2)
389 				return;
390 			opsize = *ptr++;
391 			if (opsize < 2) /* "silly options" */
392 				return;
393 			if (opsize > length)
394 				return;	/* don't parse partial options */
395 			if (opcode == TCPOPT_MPTCP)
396 				mptcp_parse_option(skb, ptr, opsize, mp_opt);
397 			ptr += opsize - 2;
398 			length -= opsize;
399 		}
400 	}
401 }
402 
403 bool mptcp_syn_options(struct sock *sk, const struct sk_buff *skb,
404 		       unsigned int *size, struct mptcp_out_options *opts)
405 {
406 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
407 
408 	/* we will use snd_isn to detect first pkt [re]transmission
409 	 * in mptcp_established_options_mp()
410 	 */
411 	subflow->snd_isn = TCP_SKB_CB(skb)->end_seq;
412 	if (subflow->request_mptcp) {
413 		opts->suboptions = OPTION_MPTCP_MPC_SYN;
414 		opts->csum_reqd = mptcp_is_checksum_enabled(sock_net(sk));
415 		opts->allow_join_id0 = mptcp_allow_join_id0(sock_net(sk));
416 		*size = TCPOLEN_MPTCP_MPC_SYN;
417 		return true;
418 	} else if (subflow->request_join) {
419 		pr_debug("remote_token=%u, nonce=%u", subflow->remote_token,
420 			 subflow->local_nonce);
421 		opts->suboptions = OPTION_MPTCP_MPJ_SYN;
422 		opts->join_id = subflow->local_id;
423 		opts->token = subflow->remote_token;
424 		opts->nonce = subflow->local_nonce;
425 		opts->backup = subflow->request_bkup;
426 		*size = TCPOLEN_MPTCP_MPJ_SYN;
427 		return true;
428 	}
429 	return false;
430 }
431 
432 static void clear_3rdack_retransmission(struct sock *sk)
433 {
434 	struct inet_connection_sock *icsk = inet_csk(sk);
435 
436 	sk_stop_timer(sk, &icsk->icsk_delack_timer);
437 	icsk->icsk_ack.timeout = 0;
438 	icsk->icsk_ack.ato = 0;
439 	icsk->icsk_ack.pending &= ~(ICSK_ACK_SCHED | ICSK_ACK_TIMER);
440 }
441 
442 static bool mptcp_established_options_mp(struct sock *sk, struct sk_buff *skb,
443 					 bool snd_data_fin_enable,
444 					 unsigned int *size,
445 					 struct mptcp_out_options *opts)
446 {
447 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
448 	struct mptcp_sock *msk = mptcp_sk(subflow->conn);
449 	struct mptcp_ext *mpext;
450 	unsigned int data_len;
451 	u8 len;
452 
453 	/* When skb is not available, we better over-estimate the emitted
454 	 * options len. A full DSS option (28 bytes) is longer than
455 	 * TCPOLEN_MPTCP_MPC_ACK_DATA(22) or TCPOLEN_MPTCP_MPJ_ACK(24), so
456 	 * tell the caller to defer the estimate to
457 	 * mptcp_established_options_dss(), which will reserve enough space.
458 	 */
459 	if (!skb)
460 		return false;
461 
462 	/* MPC/MPJ needed only on 3rd ack packet, DATA_FIN and TCP shutdown take precedence */
463 	if (subflow->fully_established || snd_data_fin_enable ||
464 	    subflow->snd_isn != TCP_SKB_CB(skb)->seq ||
465 	    sk->sk_state != TCP_ESTABLISHED)
466 		return false;
467 
468 	if (subflow->mp_capable) {
469 		mpext = mptcp_get_ext(skb);
470 		data_len = mpext ? mpext->data_len : 0;
471 
472 		/* we will check ops->data_len in mptcp_write_options() to
473 		 * discriminate between TCPOLEN_MPTCP_MPC_ACK_DATA and
474 		 * TCPOLEN_MPTCP_MPC_ACK
475 		 */
476 		opts->data_len = data_len;
477 		opts->suboptions = OPTION_MPTCP_MPC_ACK;
478 		opts->sndr_key = subflow->local_key;
479 		opts->rcvr_key = subflow->remote_key;
480 		opts->csum_reqd = READ_ONCE(msk->csum_enabled);
481 		opts->allow_join_id0 = mptcp_allow_join_id0(sock_net(sk));
482 
483 		/* Section 3.1.
484 		 * The MP_CAPABLE option is carried on the SYN, SYN/ACK, and ACK
485 		 * packets that start the first subflow of an MPTCP connection,
486 		 * as well as the first packet that carries data
487 		 */
488 		if (data_len > 0) {
489 			len = TCPOLEN_MPTCP_MPC_ACK_DATA;
490 			if (opts->csum_reqd) {
491 				/* we need to propagate more info to csum the pseudo hdr */
492 				opts->data_seq = mpext->data_seq;
493 				opts->subflow_seq = mpext->subflow_seq;
494 				opts->csum = mpext->csum;
495 				len += TCPOLEN_MPTCP_DSS_CHECKSUM;
496 			}
497 			*size = ALIGN(len, 4);
498 		} else {
499 			*size = TCPOLEN_MPTCP_MPC_ACK;
500 		}
501 
502 		pr_debug("subflow=%p, local_key=%llu, remote_key=%llu map_len=%d",
503 			 subflow, subflow->local_key, subflow->remote_key,
504 			 data_len);
505 
506 		return true;
507 	} else if (subflow->mp_join) {
508 		opts->suboptions = OPTION_MPTCP_MPJ_ACK;
509 		memcpy(opts->hmac, subflow->hmac, MPTCPOPT_HMAC_LEN);
510 		*size = TCPOLEN_MPTCP_MPJ_ACK;
511 		pr_debug("subflow=%p", subflow);
512 
513 		/* we can use the full delegate action helper only from BH context
514 		 * If we are in process context - sk is flushing the backlog at
515 		 * socket lock release time - just set the appropriate flag, will
516 		 * be handled by the release callback
517 		 */
518 		if (sock_owned_by_user(sk))
519 			set_bit(MPTCP_DELEGATE_ACK, &subflow->delegated_status);
520 		else
521 			mptcp_subflow_delegate(subflow, MPTCP_DELEGATE_ACK);
522 		return true;
523 	}
524 	return false;
525 }
526 
527 static void mptcp_write_data_fin(struct mptcp_subflow_context *subflow,
528 				 struct sk_buff *skb, struct mptcp_ext *ext)
529 {
530 	/* The write_seq value has already been incremented, so the actual
531 	 * sequence number for the DATA_FIN is one less.
532 	 */
533 	u64 data_fin_tx_seq = READ_ONCE(mptcp_sk(subflow->conn)->write_seq) - 1;
534 
535 	if (!ext->use_map || !skb->len) {
536 		/* RFC6824 requires a DSS mapping with specific values
537 		 * if DATA_FIN is set but no data payload is mapped
538 		 */
539 		ext->data_fin = 1;
540 		ext->use_map = 1;
541 		ext->dsn64 = 1;
542 		ext->data_seq = data_fin_tx_seq;
543 		ext->subflow_seq = 0;
544 		ext->data_len = 1;
545 	} else if (ext->data_seq + ext->data_len == data_fin_tx_seq) {
546 		/* If there's an existing DSS mapping and it is the
547 		 * final mapping, DATA_FIN consumes 1 additional byte of
548 		 * mapping space.
549 		 */
550 		ext->data_fin = 1;
551 		ext->data_len++;
552 	}
553 }
554 
555 static bool mptcp_established_options_dss(struct sock *sk, struct sk_buff *skb,
556 					  bool snd_data_fin_enable,
557 					  unsigned int *size,
558 					  struct mptcp_out_options *opts)
559 {
560 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
561 	struct mptcp_sock *msk = mptcp_sk(subflow->conn);
562 	unsigned int dss_size = 0;
563 	struct mptcp_ext *mpext;
564 	unsigned int ack_size;
565 	bool ret = false;
566 	u64 ack_seq;
567 
568 	opts->csum_reqd = READ_ONCE(msk->csum_enabled);
569 	mpext = skb ? mptcp_get_ext(skb) : NULL;
570 
571 	if (!skb || (mpext && mpext->use_map) || snd_data_fin_enable) {
572 		unsigned int map_size = TCPOLEN_MPTCP_DSS_BASE + TCPOLEN_MPTCP_DSS_MAP64;
573 
574 		if (mpext) {
575 			if (opts->csum_reqd)
576 				map_size += TCPOLEN_MPTCP_DSS_CHECKSUM;
577 
578 			opts->ext_copy = *mpext;
579 		}
580 
581 		dss_size = map_size;
582 		if (skb && snd_data_fin_enable)
583 			mptcp_write_data_fin(subflow, skb, &opts->ext_copy);
584 		opts->suboptions = OPTION_MPTCP_DSS;
585 		ret = true;
586 	}
587 
588 	/* passive sockets msk will set the 'can_ack' after accept(), even
589 	 * if the first subflow may have the already the remote key handy
590 	 */
591 	opts->ext_copy.use_ack = 0;
592 	if (!READ_ONCE(msk->can_ack)) {
593 		*size = ALIGN(dss_size, 4);
594 		return ret;
595 	}
596 
597 	ack_seq = READ_ONCE(msk->ack_seq);
598 	if (READ_ONCE(msk->use_64bit_ack)) {
599 		ack_size = TCPOLEN_MPTCP_DSS_ACK64;
600 		opts->ext_copy.data_ack = ack_seq;
601 		opts->ext_copy.ack64 = 1;
602 	} else {
603 		ack_size = TCPOLEN_MPTCP_DSS_ACK32;
604 		opts->ext_copy.data_ack32 = (uint32_t)ack_seq;
605 		opts->ext_copy.ack64 = 0;
606 	}
607 	opts->ext_copy.use_ack = 1;
608 	opts->suboptions = OPTION_MPTCP_DSS;
609 	WRITE_ONCE(msk->old_wspace, __mptcp_space((struct sock *)msk));
610 
611 	/* Add kind/length/subtype/flag overhead if mapping is not populated */
612 	if (dss_size == 0)
613 		ack_size += TCPOLEN_MPTCP_DSS_BASE;
614 
615 	dss_size += ack_size;
616 
617 	*size = ALIGN(dss_size, 4);
618 	return true;
619 }
620 
621 static u64 add_addr_generate_hmac(u64 key1, u64 key2,
622 				  struct mptcp_addr_info *addr)
623 {
624 	u16 port = ntohs(addr->port);
625 	u8 hmac[SHA256_DIGEST_SIZE];
626 	u8 msg[19];
627 	int i = 0;
628 
629 	msg[i++] = addr->id;
630 	if (addr->family == AF_INET) {
631 		memcpy(&msg[i], &addr->addr.s_addr, 4);
632 		i += 4;
633 	}
634 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
635 	else if (addr->family == AF_INET6) {
636 		memcpy(&msg[i], &addr->addr6.s6_addr, 16);
637 		i += 16;
638 	}
639 #endif
640 	msg[i++] = port >> 8;
641 	msg[i++] = port & 0xFF;
642 
643 	mptcp_crypto_hmac_sha(key1, key2, msg, i, hmac);
644 
645 	return get_unaligned_be64(&hmac[SHA256_DIGEST_SIZE - sizeof(u64)]);
646 }
647 
648 static bool mptcp_established_options_add_addr(struct sock *sk, struct sk_buff *skb,
649 					       unsigned int *size,
650 					       unsigned int remaining,
651 					       struct mptcp_out_options *opts)
652 {
653 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
654 	struct mptcp_sock *msk = mptcp_sk(subflow->conn);
655 	bool drop_other_suboptions = false;
656 	unsigned int opt_size = *size;
657 	bool echo;
658 	int len;
659 
660 	/* add addr will strip the existing options, be sure to avoid breaking
661 	 * MPC/MPJ handshakes
662 	 */
663 	if (!mptcp_pm_should_add_signal(msk) ||
664 	    (opts->suboptions & (OPTION_MPTCP_MPJ_ACK | OPTION_MPTCP_MPC_ACK)) ||
665 	    !mptcp_pm_add_addr_signal(msk, skb, opt_size, remaining, &opts->addr,
666 		    &echo, &drop_other_suboptions))
667 		return false;
668 
669 	if (drop_other_suboptions)
670 		remaining += opt_size;
671 	len = mptcp_add_addr_len(opts->addr.family, echo, !!opts->addr.port);
672 	if (remaining < len)
673 		return false;
674 
675 	*size = len;
676 	if (drop_other_suboptions) {
677 		pr_debug("drop other suboptions");
678 		opts->suboptions = 0;
679 
680 		/* note that e.g. DSS could have written into the memory
681 		 * aliased by ahmac, we must reset the field here
682 		 * to avoid appending the hmac even for ADD_ADDR echo
683 		 * options
684 		 */
685 		opts->ahmac = 0;
686 		*size -= opt_size;
687 	}
688 	opts->suboptions |= OPTION_MPTCP_ADD_ADDR;
689 	if (!echo) {
690 		opts->ahmac = add_addr_generate_hmac(msk->local_key,
691 						     msk->remote_key,
692 						     &opts->addr);
693 	}
694 	pr_debug("addr_id=%d, ahmac=%llu, echo=%d, port=%d",
695 		 opts->addr.id, opts->ahmac, echo, ntohs(opts->addr.port));
696 
697 	return true;
698 }
699 
700 static bool mptcp_established_options_rm_addr(struct sock *sk,
701 					      unsigned int *size,
702 					      unsigned int remaining,
703 					      struct mptcp_out_options *opts)
704 {
705 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
706 	struct mptcp_sock *msk = mptcp_sk(subflow->conn);
707 	struct mptcp_rm_list rm_list;
708 	int i, len;
709 
710 	if (!mptcp_pm_should_rm_signal(msk) ||
711 	    !(mptcp_pm_rm_addr_signal(msk, remaining, &rm_list)))
712 		return false;
713 
714 	len = mptcp_rm_addr_len(&rm_list);
715 	if (len < 0)
716 		return false;
717 	if (remaining < len)
718 		return false;
719 
720 	*size = len;
721 	opts->suboptions |= OPTION_MPTCP_RM_ADDR;
722 	opts->rm_list = rm_list;
723 
724 	for (i = 0; i < opts->rm_list.nr; i++)
725 		pr_debug("rm_list_ids[%d]=%d", i, opts->rm_list.ids[i]);
726 
727 	return true;
728 }
729 
730 static bool mptcp_established_options_mp_prio(struct sock *sk,
731 					      unsigned int *size,
732 					      unsigned int remaining,
733 					      struct mptcp_out_options *opts)
734 {
735 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
736 
737 	/* can't send MP_PRIO with MPC, as they share the same option space:
738 	 * 'backup'. Also it makes no sense at all
739 	 */
740 	if (!subflow->send_mp_prio || (opts->suboptions & OPTIONS_MPTCP_MPC))
741 		return false;
742 
743 	/* account for the trailing 'nop' option */
744 	if (remaining < TCPOLEN_MPTCP_PRIO_ALIGN)
745 		return false;
746 
747 	*size = TCPOLEN_MPTCP_PRIO_ALIGN;
748 	opts->suboptions |= OPTION_MPTCP_PRIO;
749 	opts->backup = subflow->request_bkup;
750 
751 	pr_debug("prio=%d", opts->backup);
752 
753 	return true;
754 }
755 
756 static noinline bool mptcp_established_options_rst(struct sock *sk, struct sk_buff *skb,
757 						   unsigned int *size,
758 						   unsigned int remaining,
759 						   struct mptcp_out_options *opts)
760 {
761 	const struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
762 
763 	if (remaining < TCPOLEN_MPTCP_RST)
764 		return false;
765 
766 	*size = TCPOLEN_MPTCP_RST;
767 	opts->suboptions |= OPTION_MPTCP_RST;
768 	opts->reset_transient = subflow->reset_transient;
769 	opts->reset_reason = subflow->reset_reason;
770 	MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPRSTTX);
771 
772 	return true;
773 }
774 
775 static bool mptcp_established_options_fastclose(struct sock *sk,
776 						unsigned int *size,
777 						unsigned int remaining,
778 						struct mptcp_out_options *opts)
779 {
780 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
781 	struct mptcp_sock *msk = mptcp_sk(subflow->conn);
782 
783 	if (likely(!subflow->send_fastclose))
784 		return false;
785 
786 	if (remaining < TCPOLEN_MPTCP_FASTCLOSE)
787 		return false;
788 
789 	*size = TCPOLEN_MPTCP_FASTCLOSE;
790 	opts->suboptions |= OPTION_MPTCP_FASTCLOSE;
791 	opts->rcvr_key = msk->remote_key;
792 
793 	pr_debug("FASTCLOSE key=%llu", opts->rcvr_key);
794 	MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPFASTCLOSETX);
795 	return true;
796 }
797 
798 static bool mptcp_established_options_mp_fail(struct sock *sk,
799 					      unsigned int *size,
800 					      unsigned int remaining,
801 					      struct mptcp_out_options *opts)
802 {
803 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
804 
805 	if (likely(!subflow->send_mp_fail))
806 		return false;
807 
808 	if (remaining < TCPOLEN_MPTCP_FAIL)
809 		return false;
810 
811 	*size = TCPOLEN_MPTCP_FAIL;
812 	opts->suboptions |= OPTION_MPTCP_FAIL;
813 	opts->fail_seq = subflow->map_seq;
814 
815 	pr_debug("MP_FAIL fail_seq=%llu", opts->fail_seq);
816 	MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPFAILTX);
817 
818 	return true;
819 }
820 
821 bool mptcp_established_options(struct sock *sk, struct sk_buff *skb,
822 			       unsigned int *size, unsigned int remaining,
823 			       struct mptcp_out_options *opts)
824 {
825 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
826 	struct mptcp_sock *msk = mptcp_sk(subflow->conn);
827 	unsigned int opt_size = 0;
828 	bool snd_data_fin;
829 	bool ret = false;
830 
831 	opts->suboptions = 0;
832 
833 	if (unlikely(__mptcp_check_fallback(msk) && !mptcp_check_infinite_map(skb)))
834 		return false;
835 
836 	if (unlikely(skb && TCP_SKB_CB(skb)->tcp_flags & TCPHDR_RST)) {
837 		if (mptcp_established_options_fastclose(sk, &opt_size, remaining, opts) ||
838 		    mptcp_established_options_mp_fail(sk, &opt_size, remaining, opts)) {
839 			*size += opt_size;
840 			remaining -= opt_size;
841 		}
842 		/* MP_RST can be used with MP_FASTCLOSE and MP_FAIL if there is room */
843 		if (mptcp_established_options_rst(sk, skb, &opt_size, remaining, opts)) {
844 			*size += opt_size;
845 			remaining -= opt_size;
846 		}
847 		return true;
848 	}
849 
850 	snd_data_fin = mptcp_data_fin_enabled(msk);
851 	if (mptcp_established_options_mp(sk, skb, snd_data_fin, &opt_size, opts))
852 		ret = true;
853 	else if (mptcp_established_options_dss(sk, skb, snd_data_fin, &opt_size, opts)) {
854 		unsigned int mp_fail_size;
855 
856 		ret = true;
857 		if (mptcp_established_options_mp_fail(sk, &mp_fail_size,
858 						      remaining - opt_size, opts)) {
859 			*size += opt_size + mp_fail_size;
860 			remaining -= opt_size - mp_fail_size;
861 			return true;
862 		}
863 	}
864 
865 	/* we reserved enough space for the above options, and exceeding the
866 	 * TCP option space would be fatal
867 	 */
868 	if (WARN_ON_ONCE(opt_size > remaining))
869 		return false;
870 
871 	*size += opt_size;
872 	remaining -= opt_size;
873 	if (mptcp_established_options_add_addr(sk, skb, &opt_size, remaining, opts)) {
874 		*size += opt_size;
875 		remaining -= opt_size;
876 		ret = true;
877 	} else if (mptcp_established_options_rm_addr(sk, &opt_size, remaining, opts)) {
878 		*size += opt_size;
879 		remaining -= opt_size;
880 		ret = true;
881 	}
882 
883 	if (mptcp_established_options_mp_prio(sk, &opt_size, remaining, opts)) {
884 		*size += opt_size;
885 		remaining -= opt_size;
886 		ret = true;
887 	}
888 
889 	return ret;
890 }
891 
892 bool mptcp_synack_options(const struct request_sock *req, unsigned int *size,
893 			  struct mptcp_out_options *opts)
894 {
895 	struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
896 
897 	if (subflow_req->mp_capable) {
898 		opts->suboptions = OPTION_MPTCP_MPC_SYNACK;
899 		opts->sndr_key = subflow_req->local_key;
900 		opts->csum_reqd = subflow_req->csum_reqd;
901 		opts->allow_join_id0 = subflow_req->allow_join_id0;
902 		*size = TCPOLEN_MPTCP_MPC_SYNACK;
903 		pr_debug("subflow_req=%p, local_key=%llu",
904 			 subflow_req, subflow_req->local_key);
905 		return true;
906 	} else if (subflow_req->mp_join) {
907 		opts->suboptions = OPTION_MPTCP_MPJ_SYNACK;
908 		opts->backup = subflow_req->backup;
909 		opts->join_id = subflow_req->local_id;
910 		opts->thmac = subflow_req->thmac;
911 		opts->nonce = subflow_req->local_nonce;
912 		pr_debug("req=%p, bkup=%u, id=%u, thmac=%llu, nonce=%u",
913 			 subflow_req, opts->backup, opts->join_id,
914 			 opts->thmac, opts->nonce);
915 		*size = TCPOLEN_MPTCP_MPJ_SYNACK;
916 		return true;
917 	}
918 	return false;
919 }
920 
921 static bool check_fully_established(struct mptcp_sock *msk, struct sock *ssk,
922 				    struct mptcp_subflow_context *subflow,
923 				    struct sk_buff *skb,
924 				    struct mptcp_options_received *mp_opt)
925 {
926 	/* here we can process OoO, in-window pkts, only in-sequence 4th ack
927 	 * will make the subflow fully established
928 	 */
929 	if (likely(subflow->fully_established)) {
930 		/* on passive sockets, check for 3rd ack retransmission
931 		 * note that msk is always set by subflow_syn_recv_sock()
932 		 * for mp_join subflows
933 		 */
934 		if (TCP_SKB_CB(skb)->seq == subflow->ssn_offset + 1 &&
935 		    TCP_SKB_CB(skb)->end_seq == TCP_SKB_CB(skb)->seq &&
936 		    subflow->mp_join && (mp_opt->suboptions & OPTIONS_MPTCP_MPJ) &&
937 		    !subflow->request_join)
938 			tcp_send_ack(ssk);
939 		goto check_notify;
940 	}
941 
942 	/* we must process OoO packets before the first subflow is fully
943 	 * established. OoO packets are instead a protocol violation
944 	 * for MP_JOIN subflows as the peer must not send any data
945 	 * before receiving the forth ack - cfr. RFC 8684 section 3.2.
946 	 */
947 	if (TCP_SKB_CB(skb)->seq != subflow->ssn_offset + 1) {
948 		if (subflow->mp_join)
949 			goto reset;
950 		if (subflow->is_mptfo && mp_opt->suboptions & OPTION_MPTCP_MPC_ACK)
951 			goto set_fully_established;
952 		return subflow->mp_capable;
953 	}
954 
955 	if (subflow->remote_key_valid &&
956 	    (((mp_opt->suboptions & OPTION_MPTCP_DSS) && mp_opt->use_ack) ||
957 	     ((mp_opt->suboptions & OPTION_MPTCP_ADD_ADDR) && !mp_opt->echo))) {
958 		/* subflows are fully established as soon as we get any
959 		 * additional ack, including ADD_ADDR.
960 		 */
961 		subflow->fully_established = 1;
962 		WRITE_ONCE(msk->fully_established, true);
963 		goto check_notify;
964 	}
965 
966 	/* If the first established packet does not contain MP_CAPABLE + data
967 	 * then fallback to TCP. Fallback scenarios requires a reset for
968 	 * MP_JOIN subflows.
969 	 */
970 	if (!(mp_opt->suboptions & OPTIONS_MPTCP_MPC)) {
971 		if (subflow->mp_join)
972 			goto reset;
973 		subflow->mp_capable = 0;
974 		pr_fallback(msk);
975 		mptcp_do_fallback(ssk);
976 		return false;
977 	}
978 
979 	if (mp_opt->deny_join_id0)
980 		WRITE_ONCE(msk->pm.remote_deny_join_id0, true);
981 
982 set_fully_established:
983 	if (unlikely(!READ_ONCE(msk->pm.server_side)))
984 		pr_warn_once("bogus mpc option on established client sk");
985 	mptcp_subflow_fully_established(subflow, mp_opt);
986 
987 check_notify:
988 	/* if the subflow is not already linked into the conn_list, we can't
989 	 * notify the PM: this subflow is still on the listener queue
990 	 * and the PM possibly acquiring the subflow lock could race with
991 	 * the listener close
992 	 */
993 	if (likely(subflow->pm_notified) || list_empty(&subflow->node))
994 		return true;
995 
996 	subflow->pm_notified = 1;
997 	if (subflow->mp_join) {
998 		clear_3rdack_retransmission(ssk);
999 		mptcp_pm_subflow_established(msk);
1000 	} else {
1001 		mptcp_pm_fully_established(msk, ssk);
1002 	}
1003 	return true;
1004 
1005 reset:
1006 	mptcp_subflow_reset(ssk);
1007 	return false;
1008 }
1009 
1010 u64 __mptcp_expand_seq(u64 old_seq, u64 cur_seq)
1011 {
1012 	u32 old_seq32, cur_seq32;
1013 
1014 	old_seq32 = (u32)old_seq;
1015 	cur_seq32 = (u32)cur_seq;
1016 	cur_seq = (old_seq & GENMASK_ULL(63, 32)) + cur_seq32;
1017 	if (unlikely(cur_seq32 < old_seq32 && before(old_seq32, cur_seq32)))
1018 		return cur_seq + (1LL << 32);
1019 
1020 	/* reverse wrap could happen, too */
1021 	if (unlikely(cur_seq32 > old_seq32 && after(old_seq32, cur_seq32)))
1022 		return cur_seq - (1LL << 32);
1023 	return cur_seq;
1024 }
1025 
1026 static void ack_update_msk(struct mptcp_sock *msk,
1027 			   struct sock *ssk,
1028 			   struct mptcp_options_received *mp_opt)
1029 {
1030 	u64 new_wnd_end, new_snd_una, snd_nxt = READ_ONCE(msk->snd_nxt);
1031 	struct sock *sk = (struct sock *)msk;
1032 	u64 old_snd_una;
1033 
1034 	mptcp_data_lock(sk);
1035 
1036 	/* avoid ack expansion on update conflict, to reduce the risk of
1037 	 * wrongly expanding to a future ack sequence number, which is way
1038 	 * more dangerous than missing an ack
1039 	 */
1040 	old_snd_una = msk->snd_una;
1041 	new_snd_una = mptcp_expand_seq(old_snd_una, mp_opt->data_ack, mp_opt->ack64);
1042 
1043 	/* ACK for data not even sent yet? Ignore.*/
1044 	if (unlikely(after64(new_snd_una, snd_nxt)))
1045 		new_snd_una = old_snd_una;
1046 
1047 	new_wnd_end = new_snd_una + tcp_sk(ssk)->snd_wnd;
1048 
1049 	if (after64(new_wnd_end, msk->wnd_end))
1050 		msk->wnd_end = new_wnd_end;
1051 
1052 	/* this assumes mptcp_incoming_options() is invoked after tcp_ack() */
1053 	if (after64(msk->wnd_end, READ_ONCE(msk->snd_nxt)))
1054 		__mptcp_check_push(sk, ssk);
1055 
1056 	if (after64(new_snd_una, old_snd_una)) {
1057 		msk->snd_una = new_snd_una;
1058 		__mptcp_data_acked(sk);
1059 	}
1060 	mptcp_data_unlock(sk);
1061 
1062 	trace_ack_update_msk(mp_opt->data_ack,
1063 			     old_snd_una, new_snd_una,
1064 			     new_wnd_end, msk->wnd_end);
1065 }
1066 
1067 bool mptcp_update_rcv_data_fin(struct mptcp_sock *msk, u64 data_fin_seq, bool use_64bit)
1068 {
1069 	/* Skip if DATA_FIN was already received.
1070 	 * If updating simultaneously with the recvmsg loop, values
1071 	 * should match. If they mismatch, the peer is misbehaving and
1072 	 * we will prefer the most recent information.
1073 	 */
1074 	if (READ_ONCE(msk->rcv_data_fin))
1075 		return false;
1076 
1077 	WRITE_ONCE(msk->rcv_data_fin_seq,
1078 		   mptcp_expand_seq(READ_ONCE(msk->ack_seq), data_fin_seq, use_64bit));
1079 	WRITE_ONCE(msk->rcv_data_fin, 1);
1080 
1081 	return true;
1082 }
1083 
1084 static bool add_addr_hmac_valid(struct mptcp_sock *msk,
1085 				struct mptcp_options_received *mp_opt)
1086 {
1087 	u64 hmac = 0;
1088 
1089 	if (mp_opt->echo)
1090 		return true;
1091 
1092 	hmac = add_addr_generate_hmac(msk->remote_key,
1093 				      msk->local_key,
1094 				      &mp_opt->addr);
1095 
1096 	pr_debug("msk=%p, ahmac=%llu, mp_opt->ahmac=%llu\n",
1097 		 msk, hmac, mp_opt->ahmac);
1098 
1099 	return hmac == mp_opt->ahmac;
1100 }
1101 
1102 /* Return false if a subflow has been reset, else return true */
1103 bool mptcp_incoming_options(struct sock *sk, struct sk_buff *skb)
1104 {
1105 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1106 	struct mptcp_sock *msk = mptcp_sk(subflow->conn);
1107 	struct mptcp_options_received mp_opt;
1108 	struct mptcp_ext *mpext;
1109 
1110 	if (__mptcp_check_fallback(msk)) {
1111 		/* Keep it simple and unconditionally trigger send data cleanup and
1112 		 * pending queue spooling. We will need to acquire the data lock
1113 		 * for more accurate checks, and once the lock is acquired, such
1114 		 * helpers are cheap.
1115 		 */
1116 		mptcp_data_lock(subflow->conn);
1117 		if (sk_stream_memory_free(sk))
1118 			__mptcp_check_push(subflow->conn, sk);
1119 		__mptcp_data_acked(subflow->conn);
1120 		mptcp_data_unlock(subflow->conn);
1121 		return true;
1122 	}
1123 
1124 	mptcp_get_options(skb, &mp_opt);
1125 
1126 	/* The subflow can be in close state only if check_fully_established()
1127 	 * just sent a reset. If so, tell the caller to ignore the current packet.
1128 	 */
1129 	if (!check_fully_established(msk, sk, subflow, skb, &mp_opt))
1130 		return sk->sk_state != TCP_CLOSE;
1131 
1132 	if (unlikely(mp_opt.suboptions != OPTION_MPTCP_DSS)) {
1133 		if ((mp_opt.suboptions & OPTION_MPTCP_FASTCLOSE) &&
1134 		    msk->local_key == mp_opt.rcvr_key) {
1135 			WRITE_ONCE(msk->rcv_fastclose, true);
1136 			mptcp_schedule_work((struct sock *)msk);
1137 			MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPFASTCLOSERX);
1138 		}
1139 
1140 		if ((mp_opt.suboptions & OPTION_MPTCP_ADD_ADDR) &&
1141 		    add_addr_hmac_valid(msk, &mp_opt)) {
1142 			if (!mp_opt.echo) {
1143 				mptcp_pm_add_addr_received(sk, &mp_opt.addr);
1144 				MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ADDADDR);
1145 			} else {
1146 				mptcp_pm_add_addr_echoed(msk, &mp_opt.addr);
1147 				mptcp_pm_del_add_timer(msk, &mp_opt.addr, true);
1148 				MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ECHOADD);
1149 			}
1150 
1151 			if (mp_opt.addr.port)
1152 				MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_PORTADD);
1153 		}
1154 
1155 		if (mp_opt.suboptions & OPTION_MPTCP_RM_ADDR)
1156 			mptcp_pm_rm_addr_received(msk, &mp_opt.rm_list);
1157 
1158 		if (mp_opt.suboptions & OPTION_MPTCP_PRIO) {
1159 			mptcp_pm_mp_prio_received(sk, mp_opt.backup);
1160 			MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPPRIORX);
1161 		}
1162 
1163 		if (mp_opt.suboptions & OPTION_MPTCP_FAIL) {
1164 			mptcp_pm_mp_fail_received(sk, mp_opt.fail_seq);
1165 			MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPFAILRX);
1166 		}
1167 
1168 		if (mp_opt.suboptions & OPTION_MPTCP_RST) {
1169 			subflow->reset_seen = 1;
1170 			subflow->reset_reason = mp_opt.reset_reason;
1171 			subflow->reset_transient = mp_opt.reset_transient;
1172 			MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPRSTRX);
1173 		}
1174 
1175 		if (!(mp_opt.suboptions & OPTION_MPTCP_DSS))
1176 			return true;
1177 	}
1178 
1179 	/* we can't wait for recvmsg() to update the ack_seq, otherwise
1180 	 * monodirectional flows will stuck
1181 	 */
1182 	if (mp_opt.use_ack)
1183 		ack_update_msk(msk, sk, &mp_opt);
1184 
1185 	/* Zero-data-length packets are dropped by the caller and not
1186 	 * propagated to the MPTCP layer, so the skb extension does not
1187 	 * need to be allocated or populated. DATA_FIN information, if
1188 	 * present, needs to be updated here before the skb is freed.
1189 	 */
1190 	if (TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb)->end_seq) {
1191 		if (mp_opt.data_fin && mp_opt.data_len == 1 &&
1192 		    mptcp_update_rcv_data_fin(msk, mp_opt.data_seq, mp_opt.dsn64))
1193 			mptcp_schedule_work((struct sock *)msk);
1194 
1195 		return true;
1196 	}
1197 
1198 	mpext = skb_ext_add(skb, SKB_EXT_MPTCP);
1199 	if (!mpext)
1200 		return true;
1201 
1202 	memset(mpext, 0, sizeof(*mpext));
1203 
1204 	if (likely(mp_opt.use_map)) {
1205 		if (mp_opt.mpc_map) {
1206 			/* this is an MP_CAPABLE carrying MPTCP data
1207 			 * we know this map the first chunk of data
1208 			 */
1209 			mptcp_crypto_key_sha(subflow->remote_key, NULL,
1210 					     &mpext->data_seq);
1211 			mpext->data_seq++;
1212 			mpext->subflow_seq = 1;
1213 			mpext->dsn64 = 1;
1214 			mpext->mpc_map = 1;
1215 			mpext->data_fin = 0;
1216 		} else {
1217 			mpext->data_seq = mp_opt.data_seq;
1218 			mpext->subflow_seq = mp_opt.subflow_seq;
1219 			mpext->dsn64 = mp_opt.dsn64;
1220 			mpext->data_fin = mp_opt.data_fin;
1221 		}
1222 		mpext->data_len = mp_opt.data_len;
1223 		mpext->use_map = 1;
1224 		mpext->csum_reqd = !!(mp_opt.suboptions & OPTION_MPTCP_CSUMREQD);
1225 
1226 		if (mpext->csum_reqd)
1227 			mpext->csum = mp_opt.csum;
1228 	}
1229 
1230 	return true;
1231 }
1232 
1233 static void mptcp_set_rwin(struct tcp_sock *tp, struct tcphdr *th)
1234 {
1235 	const struct sock *ssk = (const struct sock *)tp;
1236 	struct mptcp_subflow_context *subflow;
1237 	u64 ack_seq, rcv_wnd_old, rcv_wnd_new;
1238 	struct mptcp_sock *msk;
1239 	u32 new_win;
1240 	u64 win;
1241 
1242 	subflow = mptcp_subflow_ctx(ssk);
1243 	msk = mptcp_sk(subflow->conn);
1244 
1245 	ack_seq = READ_ONCE(msk->ack_seq);
1246 	rcv_wnd_new = ack_seq + tp->rcv_wnd;
1247 
1248 	rcv_wnd_old = atomic64_read(&msk->rcv_wnd_sent);
1249 	if (after64(rcv_wnd_new, rcv_wnd_old)) {
1250 		u64 rcv_wnd;
1251 
1252 		for (;;) {
1253 			rcv_wnd = atomic64_cmpxchg(&msk->rcv_wnd_sent, rcv_wnd_old, rcv_wnd_new);
1254 
1255 			if (rcv_wnd == rcv_wnd_old)
1256 				break;
1257 			if (before64(rcv_wnd_new, rcv_wnd)) {
1258 				MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_RCVWNDCONFLICTUPDATE);
1259 				goto raise_win;
1260 			}
1261 			MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_RCVWNDCONFLICT);
1262 			rcv_wnd_old = rcv_wnd;
1263 		}
1264 		return;
1265 	}
1266 
1267 	if (rcv_wnd_new != rcv_wnd_old) {
1268 raise_win:
1269 		win = rcv_wnd_old - ack_seq;
1270 		tp->rcv_wnd = min_t(u64, win, U32_MAX);
1271 		new_win = tp->rcv_wnd;
1272 
1273 		/* Make sure we do not exceed the maximum possible
1274 		 * scaled window.
1275 		 */
1276 		if (unlikely(th->syn))
1277 			new_win = min(new_win, 65535U) << tp->rx_opt.rcv_wscale;
1278 		if (!tp->rx_opt.rcv_wscale &&
1279 		    READ_ONCE(sock_net(ssk)->ipv4.sysctl_tcp_workaround_signed_windows))
1280 			new_win = min(new_win, MAX_TCP_WINDOW);
1281 		else
1282 			new_win = min(new_win, (65535U << tp->rx_opt.rcv_wscale));
1283 
1284 		/* RFC1323 scaling applied */
1285 		new_win >>= tp->rx_opt.rcv_wscale;
1286 		th->window = htons(new_win);
1287 		MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_RCVWNDSHARED);
1288 	}
1289 }
1290 
1291 __sum16 __mptcp_make_csum(u64 data_seq, u32 subflow_seq, u16 data_len, __wsum sum)
1292 {
1293 	struct csum_pseudo_header header;
1294 	__wsum csum;
1295 
1296 	/* cfr RFC 8684 3.3.1.:
1297 	 * the data sequence number used in the pseudo-header is
1298 	 * always the 64-bit value, irrespective of what length is used in the
1299 	 * DSS option itself.
1300 	 */
1301 	header.data_seq = cpu_to_be64(data_seq);
1302 	header.subflow_seq = htonl(subflow_seq);
1303 	header.data_len = htons(data_len);
1304 	header.csum = 0;
1305 
1306 	csum = csum_partial(&header, sizeof(header), sum);
1307 	return csum_fold(csum);
1308 }
1309 
1310 static __sum16 mptcp_make_csum(const struct mptcp_ext *mpext)
1311 {
1312 	return __mptcp_make_csum(mpext->data_seq, mpext->subflow_seq, mpext->data_len,
1313 				 ~csum_unfold(mpext->csum));
1314 }
1315 
1316 static void put_len_csum(u16 len, __sum16 csum, void *data)
1317 {
1318 	__sum16 *sumptr = data + 2;
1319 	__be16 *ptr = data;
1320 
1321 	put_unaligned_be16(len, ptr);
1322 
1323 	put_unaligned(csum, sumptr);
1324 }
1325 
1326 void mptcp_write_options(struct tcphdr *th, __be32 *ptr, struct tcp_sock *tp,
1327 			 struct mptcp_out_options *opts)
1328 {
1329 	const struct sock *ssk = (const struct sock *)tp;
1330 	struct mptcp_subflow_context *subflow;
1331 
1332 	/* Which options can be used together?
1333 	 *
1334 	 * X: mutually exclusive
1335 	 * O: often used together
1336 	 * C: can be used together in some cases
1337 	 * P: could be used together but we prefer not to (optimisations)
1338 	 *
1339 	 *  Opt: | MPC  | MPJ  | DSS  | ADD  |  RM  | PRIO | FAIL |  FC  |
1340 	 * ------|------|------|------|------|------|------|------|------|
1341 	 *  MPC  |------|------|------|------|------|------|------|------|
1342 	 *  MPJ  |  X   |------|------|------|------|------|------|------|
1343 	 *  DSS  |  X   |  X   |------|------|------|------|------|------|
1344 	 *  ADD  |  X   |  X   |  P   |------|------|------|------|------|
1345 	 *  RM   |  C   |  C   |  C   |  P   |------|------|------|------|
1346 	 *  PRIO |  X   |  C   |  C   |  C   |  C   |------|------|------|
1347 	 *  FAIL |  X   |  X   |  C   |  X   |  X   |  X   |------|------|
1348 	 *  FC   |  X   |  X   |  X   |  X   |  X   |  X   |  X   |------|
1349 	 *  RST  |  X   |  X   |  X   |  X   |  X   |  X   |  O   |  O   |
1350 	 * ------|------|------|------|------|------|------|------|------|
1351 	 *
1352 	 * The same applies in mptcp_established_options() function.
1353 	 */
1354 	if (likely(OPTION_MPTCP_DSS & opts->suboptions)) {
1355 		struct mptcp_ext *mpext = &opts->ext_copy;
1356 		u8 len = TCPOLEN_MPTCP_DSS_BASE;
1357 		u8 flags = 0;
1358 
1359 		if (mpext->use_ack) {
1360 			flags = MPTCP_DSS_HAS_ACK;
1361 			if (mpext->ack64) {
1362 				len += TCPOLEN_MPTCP_DSS_ACK64;
1363 				flags |= MPTCP_DSS_ACK64;
1364 			} else {
1365 				len += TCPOLEN_MPTCP_DSS_ACK32;
1366 			}
1367 		}
1368 
1369 		if (mpext->use_map) {
1370 			len += TCPOLEN_MPTCP_DSS_MAP64;
1371 
1372 			/* Use only 64-bit mapping flags for now, add
1373 			 * support for optional 32-bit mappings later.
1374 			 */
1375 			flags |= MPTCP_DSS_HAS_MAP | MPTCP_DSS_DSN64;
1376 			if (mpext->data_fin)
1377 				flags |= MPTCP_DSS_DATA_FIN;
1378 
1379 			if (opts->csum_reqd)
1380 				len += TCPOLEN_MPTCP_DSS_CHECKSUM;
1381 		}
1382 
1383 		*ptr++ = mptcp_option(MPTCPOPT_DSS, len, 0, flags);
1384 
1385 		if (mpext->use_ack) {
1386 			if (mpext->ack64) {
1387 				put_unaligned_be64(mpext->data_ack, ptr);
1388 				ptr += 2;
1389 			} else {
1390 				put_unaligned_be32(mpext->data_ack32, ptr);
1391 				ptr += 1;
1392 			}
1393 		}
1394 
1395 		if (mpext->use_map) {
1396 			put_unaligned_be64(mpext->data_seq, ptr);
1397 			ptr += 2;
1398 			put_unaligned_be32(mpext->subflow_seq, ptr);
1399 			ptr += 1;
1400 			if (opts->csum_reqd) {
1401 				/* data_len == 0 is reserved for the infinite mapping,
1402 				 * the checksum will also be set to 0.
1403 				 */
1404 				put_len_csum(mpext->data_len,
1405 					     (mpext->data_len ? mptcp_make_csum(mpext) : 0),
1406 					     ptr);
1407 			} else {
1408 				put_unaligned_be32(mpext->data_len << 16 |
1409 						   TCPOPT_NOP << 8 | TCPOPT_NOP, ptr);
1410 			}
1411 			ptr += 1;
1412 		}
1413 
1414 		/* We might need to add MP_FAIL options in rare cases */
1415 		if (unlikely(OPTION_MPTCP_FAIL & opts->suboptions))
1416 			goto mp_fail;
1417 	} else if (OPTIONS_MPTCP_MPC & opts->suboptions) {
1418 		u8 len, flag = MPTCP_CAP_HMAC_SHA256;
1419 
1420 		if (OPTION_MPTCP_MPC_SYN & opts->suboptions) {
1421 			len = TCPOLEN_MPTCP_MPC_SYN;
1422 		} else if (OPTION_MPTCP_MPC_SYNACK & opts->suboptions) {
1423 			len = TCPOLEN_MPTCP_MPC_SYNACK;
1424 		} else if (opts->data_len) {
1425 			len = TCPOLEN_MPTCP_MPC_ACK_DATA;
1426 			if (opts->csum_reqd)
1427 				len += TCPOLEN_MPTCP_DSS_CHECKSUM;
1428 		} else {
1429 			len = TCPOLEN_MPTCP_MPC_ACK;
1430 		}
1431 
1432 		if (opts->csum_reqd)
1433 			flag |= MPTCP_CAP_CHECKSUM_REQD;
1434 
1435 		if (!opts->allow_join_id0)
1436 			flag |= MPTCP_CAP_DENY_JOIN_ID0;
1437 
1438 		*ptr++ = mptcp_option(MPTCPOPT_MP_CAPABLE, len,
1439 				      MPTCP_SUPPORTED_VERSION,
1440 				      flag);
1441 
1442 		if (!((OPTION_MPTCP_MPC_SYNACK | OPTION_MPTCP_MPC_ACK) &
1443 		    opts->suboptions))
1444 			goto mp_capable_done;
1445 
1446 		put_unaligned_be64(opts->sndr_key, ptr);
1447 		ptr += 2;
1448 		if (!((OPTION_MPTCP_MPC_ACK) & opts->suboptions))
1449 			goto mp_capable_done;
1450 
1451 		put_unaligned_be64(opts->rcvr_key, ptr);
1452 		ptr += 2;
1453 		if (!opts->data_len)
1454 			goto mp_capable_done;
1455 
1456 		if (opts->csum_reqd) {
1457 			put_len_csum(opts->data_len,
1458 				     __mptcp_make_csum(opts->data_seq,
1459 						       opts->subflow_seq,
1460 						       opts->data_len,
1461 						       ~csum_unfold(opts->csum)),
1462 				     ptr);
1463 		} else {
1464 			put_unaligned_be32(opts->data_len << 16 |
1465 					   TCPOPT_NOP << 8 | TCPOPT_NOP, ptr);
1466 		}
1467 		ptr += 1;
1468 
1469 		/* MPC is additionally mutually exclusive with MP_PRIO */
1470 		goto mp_capable_done;
1471 	} else if (OPTIONS_MPTCP_MPJ & opts->suboptions) {
1472 		if (OPTION_MPTCP_MPJ_SYN & opts->suboptions) {
1473 			*ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1474 					      TCPOLEN_MPTCP_MPJ_SYN,
1475 					      opts->backup, opts->join_id);
1476 			put_unaligned_be32(opts->token, ptr);
1477 			ptr += 1;
1478 			put_unaligned_be32(opts->nonce, ptr);
1479 			ptr += 1;
1480 		} else if (OPTION_MPTCP_MPJ_SYNACK & opts->suboptions) {
1481 			*ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1482 					      TCPOLEN_MPTCP_MPJ_SYNACK,
1483 					      opts->backup, opts->join_id);
1484 			put_unaligned_be64(opts->thmac, ptr);
1485 			ptr += 2;
1486 			put_unaligned_be32(opts->nonce, ptr);
1487 			ptr += 1;
1488 		} else {
1489 			*ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1490 					      TCPOLEN_MPTCP_MPJ_ACK, 0, 0);
1491 			memcpy(ptr, opts->hmac, MPTCPOPT_HMAC_LEN);
1492 			ptr += 5;
1493 		}
1494 	} else if (OPTION_MPTCP_ADD_ADDR & opts->suboptions) {
1495 		u8 len = TCPOLEN_MPTCP_ADD_ADDR_BASE;
1496 		u8 echo = MPTCP_ADDR_ECHO;
1497 
1498 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1499 		if (opts->addr.family == AF_INET6)
1500 			len = TCPOLEN_MPTCP_ADD_ADDR6_BASE;
1501 #endif
1502 
1503 		if (opts->addr.port)
1504 			len += TCPOLEN_MPTCP_PORT_LEN;
1505 
1506 		if (opts->ahmac) {
1507 			len += sizeof(opts->ahmac);
1508 			echo = 0;
1509 		}
1510 
1511 		*ptr++ = mptcp_option(MPTCPOPT_ADD_ADDR,
1512 				      len, echo, opts->addr.id);
1513 		if (opts->addr.family == AF_INET) {
1514 			memcpy((u8 *)ptr, (u8 *)&opts->addr.addr.s_addr, 4);
1515 			ptr += 1;
1516 		}
1517 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1518 		else if (opts->addr.family == AF_INET6) {
1519 			memcpy((u8 *)ptr, opts->addr.addr6.s6_addr, 16);
1520 			ptr += 4;
1521 		}
1522 #endif
1523 
1524 		if (!opts->addr.port) {
1525 			if (opts->ahmac) {
1526 				put_unaligned_be64(opts->ahmac, ptr);
1527 				ptr += 2;
1528 			}
1529 		} else {
1530 			u16 port = ntohs(opts->addr.port);
1531 
1532 			if (opts->ahmac) {
1533 				u8 *bptr = (u8 *)ptr;
1534 
1535 				put_unaligned_be16(port, bptr);
1536 				bptr += 2;
1537 				put_unaligned_be64(opts->ahmac, bptr);
1538 				bptr += 8;
1539 				put_unaligned_be16(TCPOPT_NOP << 8 |
1540 						   TCPOPT_NOP, bptr);
1541 
1542 				ptr += 3;
1543 			} else {
1544 				put_unaligned_be32(port << 16 |
1545 						   TCPOPT_NOP << 8 |
1546 						   TCPOPT_NOP, ptr);
1547 				ptr += 1;
1548 			}
1549 		}
1550 	} else if (unlikely(OPTION_MPTCP_FASTCLOSE & opts->suboptions)) {
1551 		/* FASTCLOSE is mutually exclusive with others except RST */
1552 		*ptr++ = mptcp_option(MPTCPOPT_MP_FASTCLOSE,
1553 				      TCPOLEN_MPTCP_FASTCLOSE,
1554 				      0, 0);
1555 		put_unaligned_be64(opts->rcvr_key, ptr);
1556 		ptr += 2;
1557 
1558 		if (OPTION_MPTCP_RST & opts->suboptions)
1559 			goto mp_rst;
1560 		return;
1561 	} else if (unlikely(OPTION_MPTCP_FAIL & opts->suboptions)) {
1562 mp_fail:
1563 		/* MP_FAIL is mutually exclusive with others except RST */
1564 		subflow = mptcp_subflow_ctx(ssk);
1565 		subflow->send_mp_fail = 0;
1566 
1567 		*ptr++ = mptcp_option(MPTCPOPT_MP_FAIL,
1568 				      TCPOLEN_MPTCP_FAIL,
1569 				      0, 0);
1570 		put_unaligned_be64(opts->fail_seq, ptr);
1571 		ptr += 2;
1572 
1573 		if (OPTION_MPTCP_RST & opts->suboptions)
1574 			goto mp_rst;
1575 		return;
1576 	} else if (unlikely(OPTION_MPTCP_RST & opts->suboptions)) {
1577 mp_rst:
1578 		*ptr++ = mptcp_option(MPTCPOPT_RST,
1579 				      TCPOLEN_MPTCP_RST,
1580 				      opts->reset_transient,
1581 				      opts->reset_reason);
1582 		return;
1583 	}
1584 
1585 	if (OPTION_MPTCP_PRIO & opts->suboptions) {
1586 		subflow = mptcp_subflow_ctx(ssk);
1587 		subflow->send_mp_prio = 0;
1588 
1589 		*ptr++ = mptcp_option(MPTCPOPT_MP_PRIO,
1590 				      TCPOLEN_MPTCP_PRIO,
1591 				      opts->backup, TCPOPT_NOP);
1592 
1593 		MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_MPPRIOTX);
1594 	}
1595 
1596 mp_capable_done:
1597 	if (OPTION_MPTCP_RM_ADDR & opts->suboptions) {
1598 		u8 i = 1;
1599 
1600 		*ptr++ = mptcp_option(MPTCPOPT_RM_ADDR,
1601 				      TCPOLEN_MPTCP_RM_ADDR_BASE + opts->rm_list.nr,
1602 				      0, opts->rm_list.ids[0]);
1603 
1604 		while (i < opts->rm_list.nr) {
1605 			u8 id1, id2, id3, id4;
1606 
1607 			id1 = opts->rm_list.ids[i];
1608 			id2 = i + 1 < opts->rm_list.nr ? opts->rm_list.ids[i + 1] : TCPOPT_NOP;
1609 			id3 = i + 2 < opts->rm_list.nr ? opts->rm_list.ids[i + 2] : TCPOPT_NOP;
1610 			id4 = i + 3 < opts->rm_list.nr ? opts->rm_list.ids[i + 3] : TCPOPT_NOP;
1611 			put_unaligned_be32(id1 << 24 | id2 << 16 | id3 << 8 | id4, ptr);
1612 			ptr += 1;
1613 			i += 4;
1614 		}
1615 	}
1616 
1617 	if (tp)
1618 		mptcp_set_rwin(tp, th);
1619 }
1620 
1621 __be32 mptcp_get_reset_option(const struct sk_buff *skb)
1622 {
1623 	const struct mptcp_ext *ext = mptcp_get_ext(skb);
1624 	u8 flags, reason;
1625 
1626 	if (ext) {
1627 		flags = ext->reset_transient;
1628 		reason = ext->reset_reason;
1629 
1630 		return mptcp_option(MPTCPOPT_RST, TCPOLEN_MPTCP_RST,
1631 				    flags, reason);
1632 	}
1633 
1634 	return htonl(0u);
1635 }
1636 EXPORT_SYMBOL_GPL(mptcp_get_reset_option);
1637