xref: /linux/net/mptcp/options.c (revision 1a9239bb4253f9076b5b4b2a1a4e8d7defd77a95)
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 
mptcp_cap_flag_sha256(u8 flags)18 static bool mptcp_cap_flag_sha256(u8 flags)
19 {
20 	return (flags & MPTCP_CAP_FLAG_MASK) == MPTCP_CAP_HMAC_SHA256;
21 }
22 
mptcp_parse_option(const struct sk_buff * skb,const unsigned char * ptr,int opsize,struct mptcp_options_received * mp_opt)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\n",
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 		if (opsize == TCPOLEN_MPTCP_MPJ_SYN) {
126 			mp_opt->suboptions |= OPTION_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\n",
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->suboptions |= OPTION_MPTCP_MPJ_SYNACK;
138 			mp_opt->backup = *ptr++ & MPTCPOPT_BACKUP;
139 			mp_opt->join_id = *ptr++;
140 			mp_opt->thmac = get_unaligned_be64(ptr);
141 			ptr += 8;
142 			mp_opt->nonce = get_unaligned_be32(ptr);
143 			ptr += 4;
144 			pr_debug("MP_JOIN bkup=%u, id=%u, thmac=%llu, nonce=%u\n",
145 				 mp_opt->backup, mp_opt->join_id,
146 				 mp_opt->thmac, mp_opt->nonce);
147 		} else if (opsize == TCPOLEN_MPTCP_MPJ_ACK) {
148 			mp_opt->suboptions |= OPTION_MPTCP_MPJ_ACK;
149 			ptr += 2;
150 			memcpy(mp_opt->hmac, ptr, MPTCPOPT_HMAC_LEN);
151 			pr_debug("MP_JOIN hmac\n");
152 		}
153 		break;
154 
155 	case MPTCPOPT_DSS:
156 		pr_debug("DSS\n");
157 		ptr++;
158 
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\n",
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\n", 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 = get_unaligned((__force __sum16 *)ptr);
225 				ptr += 2;
226 			}
227 
228 			pr_debug("data_seq=%llu subflow_seq=%u data_len=%u csum=%d:%u\n",
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\n",
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\n", 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\n", 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 		pr_debug("MP_FASTCLOSE: recv_key=%llu\n", mp_opt->rcvr_key);
327 		break;
328 
329 	case MPTCPOPT_RST:
330 		if (opsize != TCPOLEN_MPTCP_RST)
331 			break;
332 
333 		if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_RST))
334 			break;
335 
336 		mp_opt->suboptions |= OPTION_MPTCP_RST;
337 		flags = *ptr++;
338 		mp_opt->reset_transient = flags & MPTCP_RST_TRANSIENT;
339 		mp_opt->reset_reason = *ptr;
340 		pr_debug("MP_RST: transient=%u reason=%u\n",
341 			 mp_opt->reset_transient, mp_opt->reset_reason);
342 		break;
343 
344 	case MPTCPOPT_MP_FAIL:
345 		if (opsize != TCPOLEN_MPTCP_FAIL)
346 			break;
347 
348 		ptr += 2;
349 		mp_opt->suboptions |= OPTION_MPTCP_FAIL;
350 		mp_opt->fail_seq = get_unaligned_be64(ptr);
351 		pr_debug("MP_FAIL: data_seq=%llu\n", mp_opt->fail_seq);
352 		break;
353 
354 	default:
355 		break;
356 	}
357 }
358 
mptcp_get_options(const struct sk_buff * skb,struct mptcp_options_received * mp_opt)359 void mptcp_get_options(const struct sk_buff *skb,
360 		       struct mptcp_options_received *mp_opt)
361 {
362 	const struct tcphdr *th = tcp_hdr(skb);
363 	const unsigned char *ptr;
364 	int length;
365 
366 	/* Ensure that casting the whole status to u32 is efficient and safe */
367 	BUILD_BUG_ON(sizeof_field(struct mptcp_options_received, status) != sizeof(u32));
368 	BUILD_BUG_ON(!IS_ALIGNED(offsetof(struct mptcp_options_received, status),
369 				 sizeof(u32)));
370 	*(u32 *)&mp_opt->status = 0;
371 
372 	length = (th->doff * 4) - sizeof(struct tcphdr);
373 	ptr = (const unsigned char *)(th + 1);
374 
375 	while (length > 0) {
376 		int opcode = *ptr++;
377 		int opsize;
378 
379 		switch (opcode) {
380 		case TCPOPT_EOL:
381 			return;
382 		case TCPOPT_NOP:	/* Ref: RFC 793 section 3.1 */
383 			length--;
384 			continue;
385 		default:
386 			if (length < 2)
387 				return;
388 			opsize = *ptr++;
389 			if (opsize < 2) /* "silly options" */
390 				return;
391 			if (opsize > length)
392 				return;	/* don't parse partial options */
393 			if (opcode == TCPOPT_MPTCP)
394 				mptcp_parse_option(skb, ptr, opsize, mp_opt);
395 			ptr += opsize - 2;
396 			length -= opsize;
397 		}
398 	}
399 }
400 
mptcp_syn_options(struct sock * sk,const struct sk_buff * skb,unsigned int * size,struct mptcp_out_options * opts)401 bool mptcp_syn_options(struct sock *sk, const struct sk_buff *skb,
402 		       unsigned int *size, struct mptcp_out_options *opts)
403 {
404 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
405 
406 	/* we will use snd_isn to detect first pkt [re]transmission
407 	 * in mptcp_established_options_mp()
408 	 */
409 	subflow->snd_isn = TCP_SKB_CB(skb)->end_seq;
410 	if (subflow->request_mptcp) {
411 		opts->suboptions = OPTION_MPTCP_MPC_SYN;
412 		opts->csum_reqd = mptcp_is_checksum_enabled(sock_net(sk));
413 		opts->allow_join_id0 = mptcp_allow_join_id0(sock_net(sk));
414 		*size = TCPOLEN_MPTCP_MPC_SYN;
415 		return true;
416 	} else if (subflow->request_join) {
417 		pr_debug("remote_token=%u, nonce=%u\n", subflow->remote_token,
418 			 subflow->local_nonce);
419 		opts->suboptions = OPTION_MPTCP_MPJ_SYN;
420 		opts->join_id = subflow->local_id;
421 		opts->token = subflow->remote_token;
422 		opts->nonce = subflow->local_nonce;
423 		opts->backup = subflow->request_bkup;
424 		*size = TCPOLEN_MPTCP_MPJ_SYN;
425 		return true;
426 	}
427 	return false;
428 }
429 
clear_3rdack_retransmission(struct sock * sk)430 static void clear_3rdack_retransmission(struct sock *sk)
431 {
432 	struct inet_connection_sock *icsk = inet_csk(sk);
433 
434 	sk_stop_timer(sk, &icsk->icsk_delack_timer);
435 	icsk->icsk_ack.ato = 0;
436 	icsk->icsk_ack.pending &= ~(ICSK_ACK_SCHED | ICSK_ACK_TIMER);
437 }
438 
mptcp_established_options_mp(struct sock * sk,struct sk_buff * skb,bool snd_data_fin_enable,unsigned int * size,struct mptcp_out_options * opts)439 static bool mptcp_established_options_mp(struct sock *sk, struct sk_buff *skb,
440 					 bool snd_data_fin_enable,
441 					 unsigned int *size,
442 					 struct mptcp_out_options *opts)
443 {
444 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
445 	struct mptcp_sock *msk = mptcp_sk(subflow->conn);
446 	struct mptcp_ext *mpext;
447 	unsigned int data_len;
448 	u8 len;
449 
450 	/* When skb is not available, we better over-estimate the emitted
451 	 * options len. A full DSS option (28 bytes) is longer than
452 	 * TCPOLEN_MPTCP_MPC_ACK_DATA(22) or TCPOLEN_MPTCP_MPJ_ACK(24), so
453 	 * tell the caller to defer the estimate to
454 	 * mptcp_established_options_dss(), which will reserve enough space.
455 	 */
456 	if (!skb)
457 		return false;
458 
459 	/* MPC/MPJ needed only on 3rd ack packet, DATA_FIN and TCP shutdown take precedence */
460 	if (READ_ONCE(subflow->fully_established) || snd_data_fin_enable ||
461 	    subflow->snd_isn != TCP_SKB_CB(skb)->seq ||
462 	    sk->sk_state != TCP_ESTABLISHED)
463 		return false;
464 
465 	if (subflow->mp_capable) {
466 		mpext = mptcp_get_ext(skb);
467 		data_len = mpext ? mpext->data_len : 0;
468 
469 		/* we will check ops->data_len in mptcp_write_options() to
470 		 * discriminate between TCPOLEN_MPTCP_MPC_ACK_DATA and
471 		 * TCPOLEN_MPTCP_MPC_ACK
472 		 */
473 		opts->data_len = data_len;
474 		opts->suboptions = OPTION_MPTCP_MPC_ACK;
475 		opts->sndr_key = subflow->local_key;
476 		opts->rcvr_key = subflow->remote_key;
477 		opts->csum_reqd = READ_ONCE(msk->csum_enabled);
478 		opts->allow_join_id0 = mptcp_allow_join_id0(sock_net(sk));
479 
480 		/* Section 3.1.
481 		 * The MP_CAPABLE option is carried on the SYN, SYN/ACK, and ACK
482 		 * packets that start the first subflow of an MPTCP connection,
483 		 * as well as the first packet that carries data
484 		 */
485 		if (data_len > 0) {
486 			len = TCPOLEN_MPTCP_MPC_ACK_DATA;
487 			if (opts->csum_reqd) {
488 				/* we need to propagate more info to csum the pseudo hdr */
489 				opts->data_seq = mpext->data_seq;
490 				opts->subflow_seq = mpext->subflow_seq;
491 				opts->csum = mpext->csum;
492 				len += TCPOLEN_MPTCP_DSS_CHECKSUM;
493 			}
494 			*size = ALIGN(len, 4);
495 		} else {
496 			*size = TCPOLEN_MPTCP_MPC_ACK;
497 		}
498 
499 		pr_debug("subflow=%p, local_key=%llu, remote_key=%llu map_len=%d\n",
500 			 subflow, subflow->local_key, subflow->remote_key,
501 			 data_len);
502 
503 		return true;
504 	} else if (subflow->mp_join) {
505 		opts->suboptions = OPTION_MPTCP_MPJ_ACK;
506 		memcpy(opts->hmac, subflow->hmac, MPTCPOPT_HMAC_LEN);
507 		*size = TCPOLEN_MPTCP_MPJ_ACK;
508 		pr_debug("subflow=%p\n", subflow);
509 
510 		/* we can use the full delegate action helper only from BH context
511 		 * If we are in process context - sk is flushing the backlog at
512 		 * socket lock release time - just set the appropriate flag, will
513 		 * be handled by the release callback
514 		 */
515 		if (sock_owned_by_user(sk))
516 			set_bit(MPTCP_DELEGATE_ACK, &subflow->delegated_status);
517 		else
518 			mptcp_subflow_delegate(subflow, MPTCP_DELEGATE_ACK);
519 		return true;
520 	}
521 	return false;
522 }
523 
mptcp_write_data_fin(struct mptcp_subflow_context * subflow,struct sk_buff * skb,struct mptcp_ext * ext)524 static void mptcp_write_data_fin(struct mptcp_subflow_context *subflow,
525 				 struct sk_buff *skb, struct mptcp_ext *ext)
526 {
527 	/* The write_seq value has already been incremented, so the actual
528 	 * sequence number for the DATA_FIN is one less.
529 	 */
530 	u64 data_fin_tx_seq = READ_ONCE(mptcp_sk(subflow->conn)->write_seq) - 1;
531 
532 	if (!ext->use_map || !skb->len) {
533 		/* RFC6824 requires a DSS mapping with specific values
534 		 * if DATA_FIN is set but no data payload is mapped
535 		 */
536 		ext->data_fin = 1;
537 		ext->use_map = 1;
538 		ext->dsn64 = 1;
539 		ext->data_seq = data_fin_tx_seq;
540 		ext->subflow_seq = 0;
541 		ext->data_len = 1;
542 	} else if (ext->data_seq + ext->data_len == data_fin_tx_seq) {
543 		/* If there's an existing DSS mapping and it is the
544 		 * final mapping, DATA_FIN consumes 1 additional byte of
545 		 * mapping space.
546 		 */
547 		ext->data_fin = 1;
548 		ext->data_len++;
549 	}
550 }
551 
mptcp_established_options_dss(struct sock * sk,struct sk_buff * skb,bool snd_data_fin_enable,unsigned int * size,struct mptcp_out_options * opts)552 static bool mptcp_established_options_dss(struct sock *sk, struct sk_buff *skb,
553 					  bool snd_data_fin_enable,
554 					  unsigned int *size,
555 					  struct mptcp_out_options *opts)
556 {
557 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
558 	struct mptcp_sock *msk = mptcp_sk(subflow->conn);
559 	unsigned int dss_size = 0;
560 	struct mptcp_ext *mpext;
561 	unsigned int ack_size;
562 	bool ret = false;
563 	u64 ack_seq;
564 
565 	opts->csum_reqd = READ_ONCE(msk->csum_enabled);
566 	mpext = skb ? mptcp_get_ext(skb) : NULL;
567 
568 	if (!skb || (mpext && mpext->use_map) || snd_data_fin_enable) {
569 		unsigned int map_size = TCPOLEN_MPTCP_DSS_BASE + TCPOLEN_MPTCP_DSS_MAP64;
570 
571 		if (mpext) {
572 			if (opts->csum_reqd)
573 				map_size += TCPOLEN_MPTCP_DSS_CHECKSUM;
574 
575 			opts->ext_copy = *mpext;
576 		}
577 
578 		dss_size = map_size;
579 		if (skb && snd_data_fin_enable)
580 			mptcp_write_data_fin(subflow, skb, &opts->ext_copy);
581 		opts->suboptions = OPTION_MPTCP_DSS;
582 		ret = true;
583 	}
584 
585 	/* passive sockets msk will set the 'can_ack' after accept(), even
586 	 * if the first subflow may have the already the remote key handy
587 	 */
588 	opts->ext_copy.use_ack = 0;
589 	if (!READ_ONCE(msk->can_ack)) {
590 		*size = ALIGN(dss_size, 4);
591 		return ret;
592 	}
593 
594 	ack_seq = READ_ONCE(msk->ack_seq);
595 	if (READ_ONCE(msk->use_64bit_ack)) {
596 		ack_size = TCPOLEN_MPTCP_DSS_ACK64;
597 		opts->ext_copy.data_ack = ack_seq;
598 		opts->ext_copy.ack64 = 1;
599 	} else {
600 		ack_size = TCPOLEN_MPTCP_DSS_ACK32;
601 		opts->ext_copy.data_ack32 = (uint32_t)ack_seq;
602 		opts->ext_copy.ack64 = 0;
603 	}
604 	opts->ext_copy.use_ack = 1;
605 	opts->suboptions = OPTION_MPTCP_DSS;
606 
607 	/* Add kind/length/subtype/flag overhead if mapping is not populated */
608 	if (dss_size == 0)
609 		ack_size += TCPOLEN_MPTCP_DSS_BASE;
610 
611 	dss_size += ack_size;
612 
613 	*size = ALIGN(dss_size, 4);
614 	return true;
615 }
616 
add_addr_generate_hmac(u64 key1,u64 key2,struct mptcp_addr_info * addr)617 static u64 add_addr_generate_hmac(u64 key1, u64 key2,
618 				  struct mptcp_addr_info *addr)
619 {
620 	u16 port = ntohs(addr->port);
621 	u8 hmac[SHA256_DIGEST_SIZE];
622 	u8 msg[19];
623 	int i = 0;
624 
625 	msg[i++] = addr->id;
626 	if (addr->family == AF_INET) {
627 		memcpy(&msg[i], &addr->addr.s_addr, 4);
628 		i += 4;
629 	}
630 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
631 	else if (addr->family == AF_INET6) {
632 		memcpy(&msg[i], &addr->addr6.s6_addr, 16);
633 		i += 16;
634 	}
635 #endif
636 	msg[i++] = port >> 8;
637 	msg[i++] = port & 0xFF;
638 
639 	mptcp_crypto_hmac_sha(key1, key2, msg, i, hmac);
640 
641 	return get_unaligned_be64(&hmac[SHA256_DIGEST_SIZE - sizeof(u64)]);
642 }
643 
mptcp_established_options_add_addr(struct sock * sk,struct sk_buff * skb,unsigned int * size,unsigned int remaining,struct mptcp_out_options * opts)644 static bool mptcp_established_options_add_addr(struct sock *sk, struct sk_buff *skb,
645 					       unsigned int *size,
646 					       unsigned int remaining,
647 					       struct mptcp_out_options *opts)
648 {
649 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
650 	struct mptcp_sock *msk = mptcp_sk(subflow->conn);
651 	bool drop_other_suboptions = false;
652 	unsigned int opt_size = *size;
653 	struct mptcp_addr_info addr;
654 	bool echo;
655 	int len;
656 
657 	/* add addr will strip the existing options, be sure to avoid breaking
658 	 * MPC/MPJ handshakes
659 	 */
660 	if (!mptcp_pm_should_add_signal(msk) ||
661 	    (opts->suboptions & (OPTION_MPTCP_MPJ_ACK | OPTION_MPTCP_MPC_ACK)) ||
662 	    !mptcp_pm_add_addr_signal(msk, skb, opt_size, remaining, &addr,
663 		    &echo, &drop_other_suboptions))
664 		return false;
665 
666 	/*
667 	 * Later on, mptcp_write_options() will enforce mutually exclusion with
668 	 * DSS, bail out if such option is set and we can't drop it.
669 	 */
670 	if (drop_other_suboptions)
671 		remaining += opt_size;
672 	else if (opts->suboptions & OPTION_MPTCP_DSS)
673 		return false;
674 
675 	len = mptcp_add_addr_len(addr.family, echo, !!addr.port);
676 	if (remaining < len)
677 		return false;
678 
679 	*size = len;
680 	if (drop_other_suboptions) {
681 		pr_debug("drop other suboptions\n");
682 		opts->suboptions = 0;
683 
684 		/* note that e.g. DSS could have written into the memory
685 		 * aliased by ahmac, we must reset the field here
686 		 * to avoid appending the hmac even for ADD_ADDR echo
687 		 * options
688 		 */
689 		opts->ahmac = 0;
690 		*size -= opt_size;
691 	}
692 	opts->addr = addr;
693 	opts->suboptions |= OPTION_MPTCP_ADD_ADDR;
694 	if (!echo) {
695 		MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ADDADDRTX);
696 		opts->ahmac = add_addr_generate_hmac(READ_ONCE(msk->local_key),
697 						     READ_ONCE(msk->remote_key),
698 						     &opts->addr);
699 	} else {
700 		MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ECHOADDTX);
701 	}
702 	pr_debug("addr_id=%d, ahmac=%llu, echo=%d, port=%d\n",
703 		 opts->addr.id, opts->ahmac, echo, ntohs(opts->addr.port));
704 
705 	return true;
706 }
707 
mptcp_established_options_rm_addr(struct sock * sk,unsigned int * size,unsigned int remaining,struct mptcp_out_options * opts)708 static bool mptcp_established_options_rm_addr(struct sock *sk,
709 					      unsigned int *size,
710 					      unsigned int remaining,
711 					      struct mptcp_out_options *opts)
712 {
713 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
714 	struct mptcp_sock *msk = mptcp_sk(subflow->conn);
715 	struct mptcp_rm_list rm_list;
716 	int i, len;
717 
718 	if (!mptcp_pm_should_rm_signal(msk) ||
719 	    !(mptcp_pm_rm_addr_signal(msk, remaining, &rm_list)))
720 		return false;
721 
722 	len = mptcp_rm_addr_len(&rm_list);
723 	if (len < 0)
724 		return false;
725 	if (remaining < len)
726 		return false;
727 
728 	*size = len;
729 	opts->suboptions |= OPTION_MPTCP_RM_ADDR;
730 	opts->rm_list = rm_list;
731 
732 	for (i = 0; i < opts->rm_list.nr; i++)
733 		pr_debug("rm_list_ids[%d]=%d\n", i, opts->rm_list.ids[i]);
734 	MPTCP_ADD_STATS(sock_net(sk), MPTCP_MIB_RMADDRTX, opts->rm_list.nr);
735 	return true;
736 }
737 
mptcp_established_options_mp_prio(struct sock * sk,unsigned int * size,unsigned int remaining,struct mptcp_out_options * opts)738 static bool mptcp_established_options_mp_prio(struct sock *sk,
739 					      unsigned int *size,
740 					      unsigned int remaining,
741 					      struct mptcp_out_options *opts)
742 {
743 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
744 
745 	/* can't send MP_PRIO with MPC, as they share the same option space:
746 	 * 'backup'. Also it makes no sense at all
747 	 */
748 	if (!subflow->send_mp_prio || (opts->suboptions & OPTIONS_MPTCP_MPC))
749 		return false;
750 
751 	/* account for the trailing 'nop' option */
752 	if (remaining < TCPOLEN_MPTCP_PRIO_ALIGN)
753 		return false;
754 
755 	*size = TCPOLEN_MPTCP_PRIO_ALIGN;
756 	opts->suboptions |= OPTION_MPTCP_PRIO;
757 	opts->backup = subflow->request_bkup;
758 
759 	pr_debug("prio=%d\n", opts->backup);
760 
761 	return true;
762 }
763 
mptcp_established_options_rst(struct sock * sk,struct sk_buff * skb,unsigned int * size,unsigned int remaining,struct mptcp_out_options * opts)764 static noinline bool mptcp_established_options_rst(struct sock *sk, struct sk_buff *skb,
765 						   unsigned int *size,
766 						   unsigned int remaining,
767 						   struct mptcp_out_options *opts)
768 {
769 	const struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
770 
771 	if (remaining < TCPOLEN_MPTCP_RST)
772 		return false;
773 
774 	*size = TCPOLEN_MPTCP_RST;
775 	opts->suboptions |= OPTION_MPTCP_RST;
776 	opts->reset_transient = subflow->reset_transient;
777 	opts->reset_reason = subflow->reset_reason;
778 	MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPRSTTX);
779 
780 	return true;
781 }
782 
mptcp_established_options_fastclose(struct sock * sk,unsigned int * size,unsigned int remaining,struct mptcp_out_options * opts)783 static bool mptcp_established_options_fastclose(struct sock *sk,
784 						unsigned int *size,
785 						unsigned int remaining,
786 						struct mptcp_out_options *opts)
787 {
788 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
789 	struct mptcp_sock *msk = mptcp_sk(subflow->conn);
790 
791 	if (likely(!subflow->send_fastclose))
792 		return false;
793 
794 	if (remaining < TCPOLEN_MPTCP_FASTCLOSE)
795 		return false;
796 
797 	*size = TCPOLEN_MPTCP_FASTCLOSE;
798 	opts->suboptions |= OPTION_MPTCP_FASTCLOSE;
799 	opts->rcvr_key = READ_ONCE(msk->remote_key);
800 
801 	pr_debug("FASTCLOSE key=%llu\n", opts->rcvr_key);
802 	MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPFASTCLOSETX);
803 	return true;
804 }
805 
mptcp_established_options_mp_fail(struct sock * sk,unsigned int * size,unsigned int remaining,struct mptcp_out_options * opts)806 static bool mptcp_established_options_mp_fail(struct sock *sk,
807 					      unsigned int *size,
808 					      unsigned int remaining,
809 					      struct mptcp_out_options *opts)
810 {
811 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
812 
813 	if (likely(!subflow->send_mp_fail))
814 		return false;
815 
816 	if (remaining < TCPOLEN_MPTCP_FAIL)
817 		return false;
818 
819 	*size = TCPOLEN_MPTCP_FAIL;
820 	opts->suboptions |= OPTION_MPTCP_FAIL;
821 	opts->fail_seq = subflow->map_seq;
822 
823 	pr_debug("MP_FAIL fail_seq=%llu\n", opts->fail_seq);
824 	MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPFAILTX);
825 
826 	return true;
827 }
828 
mptcp_established_options(struct sock * sk,struct sk_buff * skb,unsigned int * size,unsigned int remaining,struct mptcp_out_options * opts)829 bool mptcp_established_options(struct sock *sk, struct sk_buff *skb,
830 			       unsigned int *size, unsigned int remaining,
831 			       struct mptcp_out_options *opts)
832 {
833 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
834 	struct mptcp_sock *msk = mptcp_sk(subflow->conn);
835 	unsigned int opt_size = 0;
836 	bool snd_data_fin;
837 	bool ret = false;
838 
839 	opts->suboptions = 0;
840 
841 	if (unlikely(__mptcp_check_fallback(msk) && !mptcp_check_infinite_map(skb)))
842 		return false;
843 
844 	if (unlikely(skb && TCP_SKB_CB(skb)->tcp_flags & TCPHDR_RST)) {
845 		if (mptcp_established_options_fastclose(sk, &opt_size, remaining, opts) ||
846 		    mptcp_established_options_mp_fail(sk, &opt_size, remaining, opts)) {
847 			*size += opt_size;
848 			remaining -= opt_size;
849 		}
850 		/* MP_RST can be used with MP_FASTCLOSE and MP_FAIL if there is room */
851 		if (mptcp_established_options_rst(sk, skb, &opt_size, remaining, opts)) {
852 			*size += opt_size;
853 			remaining -= opt_size;
854 		}
855 		return true;
856 	}
857 
858 	snd_data_fin = mptcp_data_fin_enabled(msk);
859 	if (mptcp_established_options_mp(sk, skb, snd_data_fin, &opt_size, opts))
860 		ret = true;
861 	else if (mptcp_established_options_dss(sk, skb, snd_data_fin, &opt_size, opts)) {
862 		unsigned int mp_fail_size;
863 
864 		ret = true;
865 		if (mptcp_established_options_mp_fail(sk, &mp_fail_size,
866 						      remaining - opt_size, opts)) {
867 			*size += opt_size + mp_fail_size;
868 			remaining -= opt_size - mp_fail_size;
869 			return true;
870 		}
871 	}
872 
873 	/* we reserved enough space for the above options, and exceeding the
874 	 * TCP option space would be fatal
875 	 */
876 	if (WARN_ON_ONCE(opt_size > remaining))
877 		return false;
878 
879 	*size += opt_size;
880 	remaining -= opt_size;
881 	if (mptcp_established_options_add_addr(sk, skb, &opt_size, remaining, opts)) {
882 		*size += opt_size;
883 		remaining -= opt_size;
884 		ret = true;
885 	} else if (mptcp_established_options_rm_addr(sk, &opt_size, remaining, opts)) {
886 		*size += opt_size;
887 		remaining -= opt_size;
888 		ret = true;
889 	}
890 
891 	if (mptcp_established_options_mp_prio(sk, &opt_size, remaining, opts)) {
892 		*size += opt_size;
893 		remaining -= opt_size;
894 		ret = true;
895 	}
896 
897 	return ret;
898 }
899 
mptcp_synack_options(const struct request_sock * req,unsigned int * size,struct mptcp_out_options * opts)900 bool mptcp_synack_options(const struct request_sock *req, unsigned int *size,
901 			  struct mptcp_out_options *opts)
902 {
903 	struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
904 
905 	if (subflow_req->mp_capable) {
906 		opts->suboptions = OPTION_MPTCP_MPC_SYNACK;
907 		opts->sndr_key = subflow_req->local_key;
908 		opts->csum_reqd = subflow_req->csum_reqd;
909 		opts->allow_join_id0 = subflow_req->allow_join_id0;
910 		*size = TCPOLEN_MPTCP_MPC_SYNACK;
911 		pr_debug("subflow_req=%p, local_key=%llu\n",
912 			 subflow_req, subflow_req->local_key);
913 		return true;
914 	} else if (subflow_req->mp_join) {
915 		opts->suboptions = OPTION_MPTCP_MPJ_SYNACK;
916 		opts->backup = subflow_req->request_bkup;
917 		opts->join_id = subflow_req->local_id;
918 		opts->thmac = subflow_req->thmac;
919 		opts->nonce = subflow_req->local_nonce;
920 		pr_debug("req=%p, bkup=%u, id=%u, thmac=%llu, nonce=%u\n",
921 			 subflow_req, opts->backup, opts->join_id,
922 			 opts->thmac, opts->nonce);
923 		*size = TCPOLEN_MPTCP_MPJ_SYNACK;
924 		return true;
925 	}
926 	return false;
927 }
928 
check_fully_established(struct mptcp_sock * msk,struct sock * ssk,struct mptcp_subflow_context * subflow,struct sk_buff * skb,struct mptcp_options_received * mp_opt)929 static bool check_fully_established(struct mptcp_sock *msk, struct sock *ssk,
930 				    struct mptcp_subflow_context *subflow,
931 				    struct sk_buff *skb,
932 				    struct mptcp_options_received *mp_opt)
933 {
934 	/* here we can process OoO, in-window pkts, only in-sequence 4th ack
935 	 * will make the subflow fully established
936 	 */
937 	if (likely(READ_ONCE(subflow->fully_established))) {
938 		/* on passive sockets, check for 3rd ack retransmission
939 		 * note that msk is always set by subflow_syn_recv_sock()
940 		 * for mp_join subflows
941 		 */
942 		if (TCP_SKB_CB(skb)->seq == subflow->ssn_offset + 1 &&
943 		    TCP_SKB_CB(skb)->end_seq == TCP_SKB_CB(skb)->seq &&
944 		    subflow->mp_join && (mp_opt->suboptions & OPTIONS_MPTCP_MPJ) &&
945 		    !subflow->request_join)
946 			tcp_send_ack(ssk);
947 		goto check_notify;
948 	}
949 
950 	/* we must process OoO packets before the first subflow is fully
951 	 * established. OoO packets are instead a protocol violation
952 	 * for MP_JOIN subflows as the peer must not send any data
953 	 * before receiving the forth ack - cfr. RFC 8684 section 3.2.
954 	 */
955 	if (TCP_SKB_CB(skb)->seq != subflow->ssn_offset + 1) {
956 		if (subflow->mp_join)
957 			goto reset;
958 		if (subflow->is_mptfo && mp_opt->suboptions & OPTION_MPTCP_MPC_ACK)
959 			goto set_fully_established;
960 		return subflow->mp_capable;
961 	}
962 
963 	if (subflow->remote_key_valid &&
964 	    (((mp_opt->suboptions & OPTION_MPTCP_DSS) && mp_opt->use_ack) ||
965 	     ((mp_opt->suboptions & OPTION_MPTCP_ADD_ADDR) &&
966 	      (!mp_opt->echo || subflow->mp_join)))) {
967 		/* subflows are fully established as soon as we get any
968 		 * additional ack, including ADD_ADDR.
969 		 */
970 		goto set_fully_established;
971 	}
972 
973 	/* If the first established packet does not contain MP_CAPABLE + data
974 	 * then fallback to TCP. Fallback scenarios requires a reset for
975 	 * MP_JOIN subflows.
976 	 */
977 	if (!(mp_opt->suboptions & OPTIONS_MPTCP_MPC)) {
978 		if (subflow->mp_join)
979 			goto reset;
980 		subflow->mp_capable = 0;
981 		pr_fallback(msk);
982 		mptcp_do_fallback(ssk);
983 		return false;
984 	}
985 
986 	if (mp_opt->deny_join_id0)
987 		WRITE_ONCE(msk->pm.remote_deny_join_id0, true);
988 
989 	if (unlikely(!READ_ONCE(msk->pm.server_side)))
990 		pr_warn_once("bogus mpc option on established client sk");
991 
992 set_fully_established:
993 	mptcp_data_lock((struct sock *)msk);
994 	__mptcp_subflow_fully_established(msk, subflow, mp_opt);
995 	mptcp_data_unlock((struct sock *)msk);
996 
997 check_notify:
998 	/* if the subflow is not already linked into the conn_list, we can't
999 	 * notify the PM: this subflow is still on the listener queue
1000 	 * and the PM possibly acquiring the subflow lock could race with
1001 	 * the listener close
1002 	 */
1003 	if (likely(subflow->pm_notified) || list_empty(&subflow->node))
1004 		return true;
1005 
1006 	subflow->pm_notified = 1;
1007 	if (subflow->mp_join) {
1008 		clear_3rdack_retransmission(ssk);
1009 		mptcp_pm_subflow_established(msk);
1010 	} else {
1011 		mptcp_pm_fully_established(msk, ssk);
1012 	}
1013 	return true;
1014 
1015 reset:
1016 	mptcp_subflow_reset(ssk);
1017 	return false;
1018 }
1019 
__mptcp_expand_seq(u64 old_seq,u64 cur_seq)1020 u64 __mptcp_expand_seq(u64 old_seq, u64 cur_seq)
1021 {
1022 	u32 old_seq32, cur_seq32;
1023 
1024 	old_seq32 = (u32)old_seq;
1025 	cur_seq32 = (u32)cur_seq;
1026 	cur_seq = (old_seq & GENMASK_ULL(63, 32)) + cur_seq32;
1027 	if (unlikely(cur_seq32 < old_seq32 && before(old_seq32, cur_seq32)))
1028 		return cur_seq + (1LL << 32);
1029 
1030 	/* reverse wrap could happen, too */
1031 	if (unlikely(cur_seq32 > old_seq32 && after(old_seq32, cur_seq32)))
1032 		return cur_seq - (1LL << 32);
1033 	return cur_seq;
1034 }
1035 
__mptcp_snd_una_update(struct mptcp_sock * msk,u64 new_snd_una)1036 static void __mptcp_snd_una_update(struct mptcp_sock *msk, u64 new_snd_una)
1037 {
1038 	msk->bytes_acked += new_snd_una - msk->snd_una;
1039 	WRITE_ONCE(msk->snd_una, new_snd_una);
1040 }
1041 
ack_update_msk(struct mptcp_sock * msk,struct sock * ssk,struct mptcp_options_received * mp_opt)1042 static void ack_update_msk(struct mptcp_sock *msk,
1043 			   struct sock *ssk,
1044 			   struct mptcp_options_received *mp_opt)
1045 {
1046 	u64 new_wnd_end, new_snd_una, snd_nxt = READ_ONCE(msk->snd_nxt);
1047 	struct sock *sk = (struct sock *)msk;
1048 	u64 old_snd_una;
1049 
1050 	mptcp_data_lock(sk);
1051 
1052 	/* avoid ack expansion on update conflict, to reduce the risk of
1053 	 * wrongly expanding to a future ack sequence number, which is way
1054 	 * more dangerous than missing an ack
1055 	 */
1056 	old_snd_una = msk->snd_una;
1057 	new_snd_una = mptcp_expand_seq(old_snd_una, mp_opt->data_ack, mp_opt->ack64);
1058 
1059 	/* ACK for data not even sent yet? Ignore.*/
1060 	if (unlikely(after64(new_snd_una, snd_nxt)))
1061 		new_snd_una = old_snd_una;
1062 
1063 	new_wnd_end = new_snd_una + tcp_sk(ssk)->snd_wnd;
1064 
1065 	if (after64(new_wnd_end, msk->wnd_end))
1066 		WRITE_ONCE(msk->wnd_end, new_wnd_end);
1067 
1068 	/* this assumes mptcp_incoming_options() is invoked after tcp_ack() */
1069 	if (after64(msk->wnd_end, snd_nxt))
1070 		__mptcp_check_push(sk, ssk);
1071 
1072 	if (after64(new_snd_una, old_snd_una)) {
1073 		__mptcp_snd_una_update(msk, new_snd_una);
1074 		__mptcp_data_acked(sk);
1075 	}
1076 	msk->last_ack_recv = tcp_jiffies32;
1077 	mptcp_data_unlock(sk);
1078 
1079 	trace_ack_update_msk(mp_opt->data_ack,
1080 			     old_snd_una, new_snd_una,
1081 			     new_wnd_end, READ_ONCE(msk->wnd_end));
1082 }
1083 
mptcp_update_rcv_data_fin(struct mptcp_sock * msk,u64 data_fin_seq,bool use_64bit)1084 bool mptcp_update_rcv_data_fin(struct mptcp_sock *msk, u64 data_fin_seq, bool use_64bit)
1085 {
1086 	/* Skip if DATA_FIN was already received.
1087 	 * If updating simultaneously with the recvmsg loop, values
1088 	 * should match. If they mismatch, the peer is misbehaving and
1089 	 * we will prefer the most recent information.
1090 	 */
1091 	if (READ_ONCE(msk->rcv_data_fin))
1092 		return false;
1093 
1094 	WRITE_ONCE(msk->rcv_data_fin_seq,
1095 		   mptcp_expand_seq(READ_ONCE(msk->ack_seq), data_fin_seq, use_64bit));
1096 	WRITE_ONCE(msk->rcv_data_fin, 1);
1097 
1098 	return true;
1099 }
1100 
add_addr_hmac_valid(struct mptcp_sock * msk,struct mptcp_options_received * mp_opt)1101 static bool add_addr_hmac_valid(struct mptcp_sock *msk,
1102 				struct mptcp_options_received *mp_opt)
1103 {
1104 	u64 hmac = 0;
1105 
1106 	if (mp_opt->echo)
1107 		return true;
1108 
1109 	hmac = add_addr_generate_hmac(READ_ONCE(msk->remote_key),
1110 				      READ_ONCE(msk->local_key),
1111 				      &mp_opt->addr);
1112 
1113 	pr_debug("msk=%p, ahmac=%llu, mp_opt->ahmac=%llu\n",
1114 		 msk, hmac, mp_opt->ahmac);
1115 
1116 	return hmac == mp_opt->ahmac;
1117 }
1118 
1119 /* Return false if a subflow has been reset, else return true */
mptcp_incoming_options(struct sock * sk,struct sk_buff * skb)1120 bool mptcp_incoming_options(struct sock *sk, struct sk_buff *skb)
1121 {
1122 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1123 	struct mptcp_sock *msk = mptcp_sk(subflow->conn);
1124 	struct mptcp_options_received mp_opt;
1125 	struct mptcp_ext *mpext;
1126 
1127 	if (__mptcp_check_fallback(msk)) {
1128 		/* Keep it simple and unconditionally trigger send data cleanup and
1129 		 * pending queue spooling. We will need to acquire the data lock
1130 		 * for more accurate checks, and once the lock is acquired, such
1131 		 * helpers are cheap.
1132 		 */
1133 		mptcp_data_lock(subflow->conn);
1134 		if (sk_stream_memory_free(sk))
1135 			__mptcp_check_push(subflow->conn, sk);
1136 
1137 		/* on fallback we just need to ignore the msk-level snd_una, as
1138 		 * this is really plain TCP
1139 		 */
1140 		__mptcp_snd_una_update(msk, READ_ONCE(msk->snd_nxt));
1141 
1142 		__mptcp_data_acked(subflow->conn);
1143 		mptcp_data_unlock(subflow->conn);
1144 		return true;
1145 	}
1146 
1147 	mptcp_get_options(skb, &mp_opt);
1148 
1149 	/* The subflow can be in close state only if check_fully_established()
1150 	 * just sent a reset. If so, tell the caller to ignore the current packet.
1151 	 */
1152 	if (!check_fully_established(msk, sk, subflow, skb, &mp_opt))
1153 		return sk->sk_state != TCP_CLOSE;
1154 
1155 	if (unlikely(mp_opt.suboptions != OPTION_MPTCP_DSS)) {
1156 		if ((mp_opt.suboptions & OPTION_MPTCP_FASTCLOSE) &&
1157 		    READ_ONCE(msk->local_key) == mp_opt.rcvr_key) {
1158 			WRITE_ONCE(msk->rcv_fastclose, true);
1159 			mptcp_schedule_work((struct sock *)msk);
1160 			MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPFASTCLOSERX);
1161 		}
1162 
1163 		if ((mp_opt.suboptions & OPTION_MPTCP_ADD_ADDR) &&
1164 		    add_addr_hmac_valid(msk, &mp_opt)) {
1165 			if (!mp_opt.echo) {
1166 				mptcp_pm_add_addr_received(sk, &mp_opt.addr);
1167 				MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ADDADDR);
1168 			} else {
1169 				mptcp_pm_add_addr_echoed(msk, &mp_opt.addr);
1170 				mptcp_pm_del_add_timer(msk, &mp_opt.addr, true);
1171 				MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ECHOADD);
1172 			}
1173 
1174 			if (mp_opt.addr.port)
1175 				MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_PORTADD);
1176 		}
1177 
1178 		if (mp_opt.suboptions & OPTION_MPTCP_RM_ADDR)
1179 			mptcp_pm_rm_addr_received(msk, &mp_opt.rm_list);
1180 
1181 		if (mp_opt.suboptions & OPTION_MPTCP_PRIO) {
1182 			mptcp_pm_mp_prio_received(sk, mp_opt.backup);
1183 			MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPPRIORX);
1184 		}
1185 
1186 		if (mp_opt.suboptions & OPTION_MPTCP_FAIL) {
1187 			mptcp_pm_mp_fail_received(sk, mp_opt.fail_seq);
1188 			MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPFAILRX);
1189 		}
1190 
1191 		if (mp_opt.suboptions & OPTION_MPTCP_RST) {
1192 			subflow->reset_seen = 1;
1193 			subflow->reset_reason = mp_opt.reset_reason;
1194 			subflow->reset_transient = mp_opt.reset_transient;
1195 			MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPRSTRX);
1196 		}
1197 
1198 		if (!(mp_opt.suboptions & OPTION_MPTCP_DSS))
1199 			return true;
1200 	}
1201 
1202 	/* we can't wait for recvmsg() to update the ack_seq, otherwise
1203 	 * monodirectional flows will stuck
1204 	 */
1205 	if (mp_opt.use_ack)
1206 		ack_update_msk(msk, sk, &mp_opt);
1207 
1208 	/* Zero-data-length packets are dropped by the caller and not
1209 	 * propagated to the MPTCP layer, so the skb extension does not
1210 	 * need to be allocated or populated. DATA_FIN information, if
1211 	 * present, needs to be updated here before the skb is freed.
1212 	 */
1213 	if (TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb)->end_seq) {
1214 		if (mp_opt.data_fin && mp_opt.data_len == 1 &&
1215 		    mptcp_update_rcv_data_fin(msk, mp_opt.data_seq, mp_opt.dsn64))
1216 			mptcp_schedule_work((struct sock *)msk);
1217 
1218 		return true;
1219 	}
1220 
1221 	mpext = skb_ext_add(skb, SKB_EXT_MPTCP);
1222 	if (!mpext)
1223 		return true;
1224 
1225 	memset(mpext, 0, sizeof(*mpext));
1226 
1227 	if (likely(mp_opt.use_map)) {
1228 		if (mp_opt.mpc_map) {
1229 			/* this is an MP_CAPABLE carrying MPTCP data
1230 			 * we know this map the first chunk of data
1231 			 */
1232 			mptcp_crypto_key_sha(subflow->remote_key, NULL,
1233 					     &mpext->data_seq);
1234 			mpext->data_seq++;
1235 			mpext->subflow_seq = 1;
1236 			mpext->dsn64 = 1;
1237 			mpext->mpc_map = 1;
1238 			mpext->data_fin = 0;
1239 		} else {
1240 			mpext->data_seq = mp_opt.data_seq;
1241 			mpext->subflow_seq = mp_opt.subflow_seq;
1242 			mpext->dsn64 = mp_opt.dsn64;
1243 			mpext->data_fin = mp_opt.data_fin;
1244 		}
1245 		mpext->data_len = mp_opt.data_len;
1246 		mpext->use_map = 1;
1247 		mpext->csum_reqd = !!(mp_opt.suboptions & OPTION_MPTCP_CSUMREQD);
1248 
1249 		if (mpext->csum_reqd)
1250 			mpext->csum = mp_opt.csum;
1251 	}
1252 
1253 	return true;
1254 }
1255 
mptcp_set_rwin(struct tcp_sock * tp,struct tcphdr * th)1256 static void mptcp_set_rwin(struct tcp_sock *tp, struct tcphdr *th)
1257 {
1258 	const struct sock *ssk = (const struct sock *)tp;
1259 	struct mptcp_subflow_context *subflow;
1260 	u64 ack_seq, rcv_wnd_old, rcv_wnd_new;
1261 	struct mptcp_sock *msk;
1262 	u32 new_win;
1263 	u64 win;
1264 
1265 	subflow = mptcp_subflow_ctx(ssk);
1266 	msk = mptcp_sk(subflow->conn);
1267 
1268 	ack_seq = READ_ONCE(msk->ack_seq);
1269 	rcv_wnd_new = ack_seq + tp->rcv_wnd;
1270 
1271 	rcv_wnd_old = atomic64_read(&msk->rcv_wnd_sent);
1272 	if (after64(rcv_wnd_new, rcv_wnd_old)) {
1273 		u64 rcv_wnd;
1274 
1275 		for (;;) {
1276 			rcv_wnd = atomic64_cmpxchg(&msk->rcv_wnd_sent, rcv_wnd_old, rcv_wnd_new);
1277 
1278 			if (rcv_wnd == rcv_wnd_old)
1279 				break;
1280 
1281 			rcv_wnd_old = rcv_wnd;
1282 			if (before64(rcv_wnd_new, rcv_wnd_old)) {
1283 				MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_RCVWNDCONFLICTUPDATE);
1284 				goto raise_win;
1285 			}
1286 			MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_RCVWNDCONFLICT);
1287 		}
1288 		goto update_wspace;
1289 	}
1290 
1291 	if (rcv_wnd_new != rcv_wnd_old) {
1292 raise_win:
1293 		win = rcv_wnd_old - ack_seq;
1294 		tp->rcv_wnd = min_t(u64, win, U32_MAX);
1295 		new_win = tp->rcv_wnd;
1296 
1297 		/* Make sure we do not exceed the maximum possible
1298 		 * scaled window.
1299 		 */
1300 		if (unlikely(th->syn))
1301 			new_win = min(new_win, 65535U) << tp->rx_opt.rcv_wscale;
1302 		if (!tp->rx_opt.rcv_wscale &&
1303 		    READ_ONCE(sock_net(ssk)->ipv4.sysctl_tcp_workaround_signed_windows))
1304 			new_win = min(new_win, MAX_TCP_WINDOW);
1305 		else
1306 			new_win = min(new_win, (65535U << tp->rx_opt.rcv_wscale));
1307 
1308 		/* RFC1323 scaling applied */
1309 		new_win >>= tp->rx_opt.rcv_wscale;
1310 		th->window = htons(new_win);
1311 		MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_RCVWNDSHARED);
1312 	}
1313 
1314 update_wspace:
1315 	WRITE_ONCE(msk->old_wspace, tp->rcv_wnd);
1316 }
1317 
__mptcp_make_csum(u64 data_seq,u32 subflow_seq,u16 data_len,__wsum sum)1318 __sum16 __mptcp_make_csum(u64 data_seq, u32 subflow_seq, u16 data_len, __wsum sum)
1319 {
1320 	struct csum_pseudo_header header;
1321 	__wsum csum;
1322 
1323 	/* cfr RFC 8684 3.3.1.:
1324 	 * the data sequence number used in the pseudo-header is
1325 	 * always the 64-bit value, irrespective of what length is used in the
1326 	 * DSS option itself.
1327 	 */
1328 	header.data_seq = cpu_to_be64(data_seq);
1329 	header.subflow_seq = htonl(subflow_seq);
1330 	header.data_len = htons(data_len);
1331 	header.csum = 0;
1332 
1333 	csum = csum_partial(&header, sizeof(header), sum);
1334 	return csum_fold(csum);
1335 }
1336 
mptcp_make_csum(const struct mptcp_ext * mpext)1337 static __sum16 mptcp_make_csum(const struct mptcp_ext *mpext)
1338 {
1339 	return __mptcp_make_csum(mpext->data_seq, mpext->subflow_seq, mpext->data_len,
1340 				 ~csum_unfold(mpext->csum));
1341 }
1342 
put_len_csum(u16 len,__sum16 csum,void * data)1343 static void put_len_csum(u16 len, __sum16 csum, void *data)
1344 {
1345 	__sum16 *sumptr = data + 2;
1346 	__be16 *ptr = data;
1347 
1348 	put_unaligned_be16(len, ptr);
1349 
1350 	put_unaligned(csum, sumptr);
1351 }
1352 
mptcp_write_options(struct tcphdr * th,__be32 * ptr,struct tcp_sock * tp,struct mptcp_out_options * opts)1353 void mptcp_write_options(struct tcphdr *th, __be32 *ptr, struct tcp_sock *tp,
1354 			 struct mptcp_out_options *opts)
1355 {
1356 	const struct sock *ssk = (const struct sock *)tp;
1357 	struct mptcp_subflow_context *subflow;
1358 
1359 	/* Which options can be used together?
1360 	 *
1361 	 * X: mutually exclusive
1362 	 * O: often used together
1363 	 * C: can be used together in some cases
1364 	 * P: could be used together but we prefer not to (optimisations)
1365 	 *
1366 	 *  Opt: | MPC  | MPJ  | DSS  | ADD  |  RM  | PRIO | FAIL |  FC  |
1367 	 * ------|------|------|------|------|------|------|------|------|
1368 	 *  MPC  |------|------|------|------|------|------|------|------|
1369 	 *  MPJ  |  X   |------|------|------|------|------|------|------|
1370 	 *  DSS  |  X   |  X   |------|------|------|------|------|------|
1371 	 *  ADD  |  X   |  X   |  P   |------|------|------|------|------|
1372 	 *  RM   |  C   |  C   |  C   |  P   |------|------|------|------|
1373 	 *  PRIO |  X   |  C   |  C   |  C   |  C   |------|------|------|
1374 	 *  FAIL |  X   |  X   |  C   |  X   |  X   |  X   |------|------|
1375 	 *  FC   |  X   |  X   |  X   |  X   |  X   |  X   |  X   |------|
1376 	 *  RST  |  X   |  X   |  X   |  X   |  X   |  X   |  O   |  O   |
1377 	 * ------|------|------|------|------|------|------|------|------|
1378 	 *
1379 	 * The same applies in mptcp_established_options() function.
1380 	 */
1381 	if (likely(OPTION_MPTCP_DSS & opts->suboptions)) {
1382 		struct mptcp_ext *mpext = &opts->ext_copy;
1383 		u8 len = TCPOLEN_MPTCP_DSS_BASE;
1384 		u8 flags = 0;
1385 
1386 		if (mpext->use_ack) {
1387 			flags = MPTCP_DSS_HAS_ACK;
1388 			if (mpext->ack64) {
1389 				len += TCPOLEN_MPTCP_DSS_ACK64;
1390 				flags |= MPTCP_DSS_ACK64;
1391 			} else {
1392 				len += TCPOLEN_MPTCP_DSS_ACK32;
1393 			}
1394 		}
1395 
1396 		if (mpext->use_map) {
1397 			len += TCPOLEN_MPTCP_DSS_MAP64;
1398 
1399 			/* Use only 64-bit mapping flags for now, add
1400 			 * support for optional 32-bit mappings later.
1401 			 */
1402 			flags |= MPTCP_DSS_HAS_MAP | MPTCP_DSS_DSN64;
1403 			if (mpext->data_fin)
1404 				flags |= MPTCP_DSS_DATA_FIN;
1405 
1406 			if (opts->csum_reqd)
1407 				len += TCPOLEN_MPTCP_DSS_CHECKSUM;
1408 		}
1409 
1410 		*ptr++ = mptcp_option(MPTCPOPT_DSS, len, 0, flags);
1411 
1412 		if (mpext->use_ack) {
1413 			if (mpext->ack64) {
1414 				put_unaligned_be64(mpext->data_ack, ptr);
1415 				ptr += 2;
1416 			} else {
1417 				put_unaligned_be32(mpext->data_ack32, ptr);
1418 				ptr += 1;
1419 			}
1420 		}
1421 
1422 		if (mpext->use_map) {
1423 			put_unaligned_be64(mpext->data_seq, ptr);
1424 			ptr += 2;
1425 			put_unaligned_be32(mpext->subflow_seq, ptr);
1426 			ptr += 1;
1427 			if (opts->csum_reqd) {
1428 				/* data_len == 0 is reserved for the infinite mapping,
1429 				 * the checksum will also be set to 0.
1430 				 */
1431 				put_len_csum(mpext->data_len,
1432 					     (mpext->data_len ? mptcp_make_csum(mpext) : 0),
1433 					     ptr);
1434 			} else {
1435 				put_unaligned_be32(mpext->data_len << 16 |
1436 						   TCPOPT_NOP << 8 | TCPOPT_NOP, ptr);
1437 			}
1438 			ptr += 1;
1439 		}
1440 
1441 		/* We might need to add MP_FAIL options in rare cases */
1442 		if (unlikely(OPTION_MPTCP_FAIL & opts->suboptions))
1443 			goto mp_fail;
1444 	} else if (OPTIONS_MPTCP_MPC & opts->suboptions) {
1445 		u8 len, flag = MPTCP_CAP_HMAC_SHA256;
1446 
1447 		if (OPTION_MPTCP_MPC_SYN & opts->suboptions) {
1448 			len = TCPOLEN_MPTCP_MPC_SYN;
1449 		} else if (OPTION_MPTCP_MPC_SYNACK & opts->suboptions) {
1450 			len = TCPOLEN_MPTCP_MPC_SYNACK;
1451 		} else if (opts->data_len) {
1452 			len = TCPOLEN_MPTCP_MPC_ACK_DATA;
1453 			if (opts->csum_reqd)
1454 				len += TCPOLEN_MPTCP_DSS_CHECKSUM;
1455 		} else {
1456 			len = TCPOLEN_MPTCP_MPC_ACK;
1457 		}
1458 
1459 		if (opts->csum_reqd)
1460 			flag |= MPTCP_CAP_CHECKSUM_REQD;
1461 
1462 		if (!opts->allow_join_id0)
1463 			flag |= MPTCP_CAP_DENY_JOIN_ID0;
1464 
1465 		*ptr++ = mptcp_option(MPTCPOPT_MP_CAPABLE, len,
1466 				      MPTCP_SUPPORTED_VERSION,
1467 				      flag);
1468 
1469 		if (!((OPTION_MPTCP_MPC_SYNACK | OPTION_MPTCP_MPC_ACK) &
1470 		    opts->suboptions))
1471 			goto mp_capable_done;
1472 
1473 		put_unaligned_be64(opts->sndr_key, ptr);
1474 		ptr += 2;
1475 		if (!((OPTION_MPTCP_MPC_ACK) & opts->suboptions))
1476 			goto mp_capable_done;
1477 
1478 		put_unaligned_be64(opts->rcvr_key, ptr);
1479 		ptr += 2;
1480 		if (!opts->data_len)
1481 			goto mp_capable_done;
1482 
1483 		if (opts->csum_reqd) {
1484 			put_len_csum(opts->data_len,
1485 				     __mptcp_make_csum(opts->data_seq,
1486 						       opts->subflow_seq,
1487 						       opts->data_len,
1488 						       ~csum_unfold(opts->csum)),
1489 				     ptr);
1490 		} else {
1491 			put_unaligned_be32(opts->data_len << 16 |
1492 					   TCPOPT_NOP << 8 | TCPOPT_NOP, ptr);
1493 		}
1494 		ptr += 1;
1495 
1496 		/* MPC is additionally mutually exclusive with MP_PRIO */
1497 		goto mp_capable_done;
1498 	} else if (OPTIONS_MPTCP_MPJ & opts->suboptions) {
1499 		if (OPTION_MPTCP_MPJ_SYN & opts->suboptions) {
1500 			*ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1501 					      TCPOLEN_MPTCP_MPJ_SYN,
1502 					      opts->backup, opts->join_id);
1503 			put_unaligned_be32(opts->token, ptr);
1504 			ptr += 1;
1505 			put_unaligned_be32(opts->nonce, ptr);
1506 			ptr += 1;
1507 		} else if (OPTION_MPTCP_MPJ_SYNACK & opts->suboptions) {
1508 			*ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1509 					      TCPOLEN_MPTCP_MPJ_SYNACK,
1510 					      opts->backup, opts->join_id);
1511 			put_unaligned_be64(opts->thmac, ptr);
1512 			ptr += 2;
1513 			put_unaligned_be32(opts->nonce, ptr);
1514 			ptr += 1;
1515 		} else {
1516 			*ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1517 					      TCPOLEN_MPTCP_MPJ_ACK, 0, 0);
1518 			memcpy(ptr, opts->hmac, MPTCPOPT_HMAC_LEN);
1519 			ptr += 5;
1520 		}
1521 	} else if (OPTION_MPTCP_ADD_ADDR & opts->suboptions) {
1522 		u8 len = TCPOLEN_MPTCP_ADD_ADDR_BASE;
1523 		u8 echo = MPTCP_ADDR_ECHO;
1524 
1525 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1526 		if (opts->addr.family == AF_INET6)
1527 			len = TCPOLEN_MPTCP_ADD_ADDR6_BASE;
1528 #endif
1529 
1530 		if (opts->addr.port)
1531 			len += TCPOLEN_MPTCP_PORT_LEN;
1532 
1533 		if (opts->ahmac) {
1534 			len += sizeof(opts->ahmac);
1535 			echo = 0;
1536 		}
1537 
1538 		*ptr++ = mptcp_option(MPTCPOPT_ADD_ADDR,
1539 				      len, echo, opts->addr.id);
1540 		if (opts->addr.family == AF_INET) {
1541 			memcpy((u8 *)ptr, (u8 *)&opts->addr.addr.s_addr, 4);
1542 			ptr += 1;
1543 		}
1544 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1545 		else if (opts->addr.family == AF_INET6) {
1546 			memcpy((u8 *)ptr, opts->addr.addr6.s6_addr, 16);
1547 			ptr += 4;
1548 		}
1549 #endif
1550 
1551 		if (!opts->addr.port) {
1552 			if (opts->ahmac) {
1553 				put_unaligned_be64(opts->ahmac, ptr);
1554 				ptr += 2;
1555 			}
1556 		} else {
1557 			u16 port = ntohs(opts->addr.port);
1558 
1559 			if (opts->ahmac) {
1560 				u8 *bptr = (u8 *)ptr;
1561 
1562 				put_unaligned_be16(port, bptr);
1563 				bptr += 2;
1564 				put_unaligned_be64(opts->ahmac, bptr);
1565 				bptr += 8;
1566 				put_unaligned_be16(TCPOPT_NOP << 8 |
1567 						   TCPOPT_NOP, bptr);
1568 
1569 				ptr += 3;
1570 			} else {
1571 				put_unaligned_be32(port << 16 |
1572 						   TCPOPT_NOP << 8 |
1573 						   TCPOPT_NOP, ptr);
1574 				ptr += 1;
1575 			}
1576 		}
1577 	} else if (unlikely(OPTION_MPTCP_FASTCLOSE & opts->suboptions)) {
1578 		/* FASTCLOSE is mutually exclusive with others except RST */
1579 		*ptr++ = mptcp_option(MPTCPOPT_MP_FASTCLOSE,
1580 				      TCPOLEN_MPTCP_FASTCLOSE,
1581 				      0, 0);
1582 		put_unaligned_be64(opts->rcvr_key, ptr);
1583 		ptr += 2;
1584 
1585 		if (OPTION_MPTCP_RST & opts->suboptions)
1586 			goto mp_rst;
1587 		return;
1588 	} else if (unlikely(OPTION_MPTCP_FAIL & opts->suboptions)) {
1589 mp_fail:
1590 		/* MP_FAIL is mutually exclusive with others except RST */
1591 		subflow = mptcp_subflow_ctx(ssk);
1592 		subflow->send_mp_fail = 0;
1593 
1594 		*ptr++ = mptcp_option(MPTCPOPT_MP_FAIL,
1595 				      TCPOLEN_MPTCP_FAIL,
1596 				      0, 0);
1597 		put_unaligned_be64(opts->fail_seq, ptr);
1598 		ptr += 2;
1599 
1600 		if (OPTION_MPTCP_RST & opts->suboptions)
1601 			goto mp_rst;
1602 		return;
1603 	} else if (unlikely(OPTION_MPTCP_RST & opts->suboptions)) {
1604 mp_rst:
1605 		*ptr++ = mptcp_option(MPTCPOPT_RST,
1606 				      TCPOLEN_MPTCP_RST,
1607 				      opts->reset_transient,
1608 				      opts->reset_reason);
1609 		return;
1610 	}
1611 
1612 	if (OPTION_MPTCP_PRIO & opts->suboptions) {
1613 		subflow = mptcp_subflow_ctx(ssk);
1614 		subflow->send_mp_prio = 0;
1615 
1616 		*ptr++ = mptcp_option(MPTCPOPT_MP_PRIO,
1617 				      TCPOLEN_MPTCP_PRIO,
1618 				      opts->backup, TCPOPT_NOP);
1619 
1620 		MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_MPPRIOTX);
1621 	}
1622 
1623 mp_capable_done:
1624 	if (OPTION_MPTCP_RM_ADDR & opts->suboptions) {
1625 		u8 i = 1;
1626 
1627 		*ptr++ = mptcp_option(MPTCPOPT_RM_ADDR,
1628 				      TCPOLEN_MPTCP_RM_ADDR_BASE + opts->rm_list.nr,
1629 				      0, opts->rm_list.ids[0]);
1630 
1631 		while (i < opts->rm_list.nr) {
1632 			u8 id1, id2, id3, id4;
1633 
1634 			id1 = opts->rm_list.ids[i];
1635 			id2 = i + 1 < opts->rm_list.nr ? opts->rm_list.ids[i + 1] : TCPOPT_NOP;
1636 			id3 = i + 2 < opts->rm_list.nr ? opts->rm_list.ids[i + 2] : TCPOPT_NOP;
1637 			id4 = i + 3 < opts->rm_list.nr ? opts->rm_list.ids[i + 3] : TCPOPT_NOP;
1638 			put_unaligned_be32(id1 << 24 | id2 << 16 | id3 << 8 | id4, ptr);
1639 			ptr += 1;
1640 			i += 4;
1641 		}
1642 	}
1643 
1644 	if (tp)
1645 		mptcp_set_rwin(tp, th);
1646 }
1647 
mptcp_get_reset_option(const struct sk_buff * skb)1648 __be32 mptcp_get_reset_option(const struct sk_buff *skb)
1649 {
1650 	const struct mptcp_ext *ext = mptcp_get_ext(skb);
1651 	u8 flags, reason;
1652 
1653 	if (ext) {
1654 		flags = ext->reset_transient;
1655 		reason = ext->reset_reason;
1656 
1657 		return mptcp_option(MPTCPOPT_RST, TCPOLEN_MPTCP_RST,
1658 				    flags, reason);
1659 	}
1660 
1661 	return htonl(0u);
1662 }
1663 EXPORT_SYMBOL_GPL(mptcp_get_reset_option);
1664