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 	u8 version;
30 	u8 flags;
31 	u8 i;
32 
33 	switch (subtype) {
34 	case MPTCPOPT_MP_CAPABLE:
35 		/* strict size checking */
36 		if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)) {
37 			if (skb->len > tcp_hdr(skb)->doff << 2)
38 				expected_opsize = TCPOLEN_MPTCP_MPC_ACK_DATA;
39 			else
40 				expected_opsize = TCPOLEN_MPTCP_MPC_ACK;
41 		} else {
42 			if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_ACK)
43 				expected_opsize = TCPOLEN_MPTCP_MPC_SYNACK;
44 			else
45 				expected_opsize = TCPOLEN_MPTCP_MPC_SYN;
46 		}
47 		if (opsize != expected_opsize)
48 			break;
49 
50 		/* try to be gentle vs future versions on the initial syn */
51 		version = *ptr++ & MPTCP_VERSION_MASK;
52 		if (opsize != TCPOLEN_MPTCP_MPC_SYN) {
53 			if (version != MPTCP_SUPPORTED_VERSION)
54 				break;
55 		} else if (version < MPTCP_SUPPORTED_VERSION) {
56 			break;
57 		}
58 
59 		flags = *ptr++;
60 		if (!mptcp_cap_flag_sha256(flags) ||
61 		    (flags & MPTCP_CAP_EXTENSIBILITY))
62 			break;
63 
64 		/* RFC 6824, Section 3.1:
65 		 * "For the Checksum Required bit (labeled "A"), if either
66 		 * host requires the use of checksums, checksums MUST be used.
67 		 * In other words, the only way for checksums not to be used
68 		 * is if both hosts in their SYNs set A=0."
69 		 *
70 		 * Section 3.3.0:
71 		 * "If a checksum is not present when its use has been
72 		 * negotiated, the receiver MUST close the subflow with a RST as
73 		 * it is considered broken."
74 		 *
75 		 * We don't implement DSS checksum - fall back to TCP.
76 		 */
77 		if (flags & MPTCP_CAP_CHECKSUM_REQD)
78 			break;
79 
80 		mp_opt->mp_capable = 1;
81 		if (opsize >= TCPOLEN_MPTCP_MPC_SYNACK) {
82 			mp_opt->sndr_key = get_unaligned_be64(ptr);
83 			ptr += 8;
84 		}
85 		if (opsize >= TCPOLEN_MPTCP_MPC_ACK) {
86 			mp_opt->rcvr_key = get_unaligned_be64(ptr);
87 			ptr += 8;
88 		}
89 		if (opsize == TCPOLEN_MPTCP_MPC_ACK_DATA) {
90 			/* Section 3.1.:
91 			 * "the data parameters in a MP_CAPABLE are semantically
92 			 * equivalent to those in a DSS option and can be used
93 			 * interchangeably."
94 			 */
95 			mp_opt->dss = 1;
96 			mp_opt->use_map = 1;
97 			mp_opt->mpc_map = 1;
98 			mp_opt->data_len = get_unaligned_be16(ptr);
99 			ptr += 2;
100 		}
101 		pr_debug("MP_CAPABLE version=%x, flags=%x, optlen=%d sndr=%llu, rcvr=%llu len=%d",
102 			 version, flags, opsize, mp_opt->sndr_key,
103 			 mp_opt->rcvr_key, mp_opt->data_len);
104 		break;
105 
106 	case MPTCPOPT_MP_JOIN:
107 		mp_opt->mp_join = 1;
108 		if (opsize == TCPOLEN_MPTCP_MPJ_SYN) {
109 			mp_opt->backup = *ptr++ & MPTCPOPT_BACKUP;
110 			mp_opt->join_id = *ptr++;
111 			mp_opt->token = get_unaligned_be32(ptr);
112 			ptr += 4;
113 			mp_opt->nonce = get_unaligned_be32(ptr);
114 			ptr += 4;
115 			pr_debug("MP_JOIN bkup=%u, id=%u, token=%u, nonce=%u",
116 				 mp_opt->backup, mp_opt->join_id,
117 				 mp_opt->token, mp_opt->nonce);
118 		} else if (opsize == TCPOLEN_MPTCP_MPJ_SYNACK) {
119 			mp_opt->backup = *ptr++ & MPTCPOPT_BACKUP;
120 			mp_opt->join_id = *ptr++;
121 			mp_opt->thmac = get_unaligned_be64(ptr);
122 			ptr += 8;
123 			mp_opt->nonce = get_unaligned_be32(ptr);
124 			ptr += 4;
125 			pr_debug("MP_JOIN bkup=%u, id=%u, thmac=%llu, nonce=%u",
126 				 mp_opt->backup, mp_opt->join_id,
127 				 mp_opt->thmac, mp_opt->nonce);
128 		} else if (opsize == TCPOLEN_MPTCP_MPJ_ACK) {
129 			ptr += 2;
130 			memcpy(mp_opt->hmac, ptr, MPTCPOPT_HMAC_LEN);
131 			pr_debug("MP_JOIN hmac");
132 		} else {
133 			pr_warn("MP_JOIN bad option size");
134 			mp_opt->mp_join = 0;
135 		}
136 		break;
137 
138 	case MPTCPOPT_DSS:
139 		pr_debug("DSS");
140 		ptr++;
141 
142 		/* we must clear 'mpc_map' be able to detect MP_CAPABLE
143 		 * map vs DSS map in mptcp_incoming_options(), and reconstruct
144 		 * map info accordingly
145 		 */
146 		mp_opt->mpc_map = 0;
147 		flags = (*ptr++) & MPTCP_DSS_FLAG_MASK;
148 		mp_opt->data_fin = (flags & MPTCP_DSS_DATA_FIN) != 0;
149 		mp_opt->dsn64 = (flags & MPTCP_DSS_DSN64) != 0;
150 		mp_opt->use_map = (flags & MPTCP_DSS_HAS_MAP) != 0;
151 		mp_opt->ack64 = (flags & MPTCP_DSS_ACK64) != 0;
152 		mp_opt->use_ack = (flags & MPTCP_DSS_HAS_ACK);
153 
154 		pr_debug("data_fin=%d dsn64=%d use_map=%d ack64=%d use_ack=%d",
155 			 mp_opt->data_fin, mp_opt->dsn64,
156 			 mp_opt->use_map, mp_opt->ack64,
157 			 mp_opt->use_ack);
158 
159 		expected_opsize = TCPOLEN_MPTCP_DSS_BASE;
160 
161 		if (mp_opt->use_ack) {
162 			if (mp_opt->ack64)
163 				expected_opsize += TCPOLEN_MPTCP_DSS_ACK64;
164 			else
165 				expected_opsize += TCPOLEN_MPTCP_DSS_ACK32;
166 		}
167 
168 		if (mp_opt->use_map) {
169 			if (mp_opt->dsn64)
170 				expected_opsize += TCPOLEN_MPTCP_DSS_MAP64;
171 			else
172 				expected_opsize += TCPOLEN_MPTCP_DSS_MAP32;
173 		}
174 
175 		/* RFC 6824, Section 3.3:
176 		 * If a checksum is present, but its use had
177 		 * not been negotiated in the MP_CAPABLE handshake,
178 		 * the checksum field MUST be ignored.
179 		 */
180 		if (opsize != expected_opsize &&
181 		    opsize != expected_opsize + TCPOLEN_MPTCP_DSS_CHECKSUM)
182 			break;
183 
184 		mp_opt->dss = 1;
185 
186 		if (mp_opt->use_ack) {
187 			if (mp_opt->ack64) {
188 				mp_opt->data_ack = get_unaligned_be64(ptr);
189 				ptr += 8;
190 			} else {
191 				mp_opt->data_ack = get_unaligned_be32(ptr);
192 				ptr += 4;
193 			}
194 
195 			pr_debug("data_ack=%llu", mp_opt->data_ack);
196 		}
197 
198 		if (mp_opt->use_map) {
199 			if (mp_opt->dsn64) {
200 				mp_opt->data_seq = get_unaligned_be64(ptr);
201 				ptr += 8;
202 			} else {
203 				mp_opt->data_seq = get_unaligned_be32(ptr);
204 				ptr += 4;
205 			}
206 
207 			mp_opt->subflow_seq = get_unaligned_be32(ptr);
208 			ptr += 4;
209 
210 			mp_opt->data_len = get_unaligned_be16(ptr);
211 			ptr += 2;
212 
213 			pr_debug("data_seq=%llu subflow_seq=%u data_len=%u",
214 				 mp_opt->data_seq, mp_opt->subflow_seq,
215 				 mp_opt->data_len);
216 		}
217 
218 		break;
219 
220 	case MPTCPOPT_ADD_ADDR:
221 		mp_opt->echo = (*ptr++) & MPTCP_ADDR_ECHO;
222 		if (!mp_opt->echo) {
223 			if (opsize == TCPOLEN_MPTCP_ADD_ADDR ||
224 			    opsize == TCPOLEN_MPTCP_ADD_ADDR_PORT)
225 				mp_opt->addr.family = AF_INET;
226 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
227 			else if (opsize == TCPOLEN_MPTCP_ADD_ADDR6 ||
228 				 opsize == TCPOLEN_MPTCP_ADD_ADDR6_PORT)
229 				mp_opt->addr.family = AF_INET6;
230 #endif
231 			else
232 				break;
233 		} else {
234 			if (opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE ||
235 			    opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE_PORT)
236 				mp_opt->addr.family = AF_INET;
237 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
238 			else if (opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE ||
239 				 opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE_PORT)
240 				mp_opt->addr.family = AF_INET6;
241 #endif
242 			else
243 				break;
244 		}
245 
246 		mp_opt->add_addr = 1;
247 		mp_opt->addr.id = *ptr++;
248 		if (mp_opt->addr.family == AF_INET) {
249 			memcpy((u8 *)&mp_opt->addr.addr.s_addr, (u8 *)ptr, 4);
250 			ptr += 4;
251 			if (opsize == TCPOLEN_MPTCP_ADD_ADDR_PORT ||
252 			    opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE_PORT) {
253 				mp_opt->addr.port = htons(get_unaligned_be16(ptr));
254 				ptr += 2;
255 			}
256 		}
257 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
258 		else {
259 			memcpy(mp_opt->addr.addr6.s6_addr, (u8 *)ptr, 16);
260 			ptr += 16;
261 			if (opsize == TCPOLEN_MPTCP_ADD_ADDR6_PORT ||
262 			    opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE_PORT) {
263 				mp_opt->addr.port = htons(get_unaligned_be16(ptr));
264 				ptr += 2;
265 			}
266 		}
267 #endif
268 		if (!mp_opt->echo) {
269 			mp_opt->ahmac = get_unaligned_be64(ptr);
270 			ptr += 8;
271 		}
272 		pr_debug("ADD_ADDR%s: id=%d, ahmac=%llu, echo=%d, port=%d",
273 			 (mp_opt->addr.family == AF_INET6) ? "6" : "",
274 			 mp_opt->addr.id, mp_opt->ahmac, mp_opt->echo, ntohs(mp_opt->addr.port));
275 		break;
276 
277 	case MPTCPOPT_RM_ADDR:
278 		if (opsize < TCPOLEN_MPTCP_RM_ADDR_BASE + 1 ||
279 		    opsize > TCPOLEN_MPTCP_RM_ADDR_BASE + MPTCP_RM_IDS_MAX)
280 			break;
281 
282 		ptr++;
283 
284 		mp_opt->rm_addr = 1;
285 		mp_opt->rm_list.nr = opsize - TCPOLEN_MPTCP_RM_ADDR_BASE;
286 		for (i = 0; i < mp_opt->rm_list.nr; i++)
287 			mp_opt->rm_list.ids[i] = *ptr++;
288 		pr_debug("RM_ADDR: rm_list_nr=%d", mp_opt->rm_list.nr);
289 		break;
290 
291 	case MPTCPOPT_MP_PRIO:
292 		if (opsize != TCPOLEN_MPTCP_PRIO)
293 			break;
294 
295 		mp_opt->mp_prio = 1;
296 		mp_opt->backup = *ptr++ & MPTCP_PRIO_BKUP;
297 		pr_debug("MP_PRIO: prio=%d", mp_opt->backup);
298 		break;
299 
300 	case MPTCPOPT_MP_FASTCLOSE:
301 		if (opsize != TCPOLEN_MPTCP_FASTCLOSE)
302 			break;
303 
304 		ptr += 2;
305 		mp_opt->rcvr_key = get_unaligned_be64(ptr);
306 		ptr += 8;
307 		mp_opt->fastclose = 1;
308 		break;
309 
310 	case MPTCPOPT_RST:
311 		if (opsize != TCPOLEN_MPTCP_RST)
312 			break;
313 
314 		if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_RST))
315 			break;
316 		mp_opt->reset = 1;
317 		flags = *ptr++;
318 		mp_opt->reset_transient = flags & MPTCP_RST_TRANSIENT;
319 		mp_opt->reset_reason = *ptr;
320 		break;
321 
322 	default:
323 		break;
324 	}
325 }
326 
mptcp_get_options(const struct sk_buff * skb,struct mptcp_options_received * mp_opt)327 void mptcp_get_options(const struct sk_buff *skb,
328 		       struct mptcp_options_received *mp_opt)
329 {
330 	const struct tcphdr *th = tcp_hdr(skb);
331 	const unsigned char *ptr;
332 	int length;
333 
334 	/* initialize option status */
335 	mp_opt->mp_capable = 0;
336 	mp_opt->mp_join = 0;
337 	mp_opt->add_addr = 0;
338 	mp_opt->ahmac = 0;
339 	mp_opt->fastclose = 0;
340 	mp_opt->addr.port = 0;
341 	mp_opt->rm_addr = 0;
342 	mp_opt->dss = 0;
343 	mp_opt->mp_prio = 0;
344 	mp_opt->reset = 0;
345 
346 	length = (th->doff * 4) - sizeof(struct tcphdr);
347 	ptr = (const unsigned char *)(th + 1);
348 
349 	while (length > 0) {
350 		int opcode = *ptr++;
351 		int opsize;
352 
353 		switch (opcode) {
354 		case TCPOPT_EOL:
355 			return;
356 		case TCPOPT_NOP:	/* Ref: RFC 793 section 3.1 */
357 			length--;
358 			continue;
359 		default:
360 			opsize = *ptr++;
361 			if (opsize < 2) /* "silly options" */
362 				return;
363 			if (opsize > length)
364 				return;	/* don't parse partial options */
365 			if (opcode == TCPOPT_MPTCP)
366 				mptcp_parse_option(skb, ptr, opsize, mp_opt);
367 			ptr += opsize - 2;
368 			length -= opsize;
369 		}
370 	}
371 }
372 
mptcp_syn_options(struct sock * sk,const struct sk_buff * skb,unsigned int * size,struct mptcp_out_options * opts)373 bool mptcp_syn_options(struct sock *sk, const struct sk_buff *skb,
374 		       unsigned int *size, struct mptcp_out_options *opts)
375 {
376 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
377 
378 	/* we will use snd_isn to detect first pkt [re]transmission
379 	 * in mptcp_established_options_mp()
380 	 */
381 	subflow->snd_isn = TCP_SKB_CB(skb)->end_seq;
382 	if (subflow->request_mptcp) {
383 		opts->suboptions = OPTION_MPTCP_MPC_SYN;
384 		*size = TCPOLEN_MPTCP_MPC_SYN;
385 		return true;
386 	} else if (subflow->request_join) {
387 		pr_debug("remote_token=%u, nonce=%u", subflow->remote_token,
388 			 subflow->local_nonce);
389 		opts->suboptions = OPTION_MPTCP_MPJ_SYN;
390 		opts->join_id = subflow->local_id;
391 		opts->token = subflow->remote_token;
392 		opts->nonce = subflow->local_nonce;
393 		opts->backup = subflow->request_bkup;
394 		*size = TCPOLEN_MPTCP_MPJ_SYN;
395 		return true;
396 	}
397 	return false;
398 }
399 
400 /* MP_JOIN client subflow must wait for 4th ack before sending any data:
401  * TCP can't schedule delack timer before the subflow is fully established.
402  * MPTCP uses the delack timer to do 3rd ack retransmissions
403  */
schedule_3rdack_retransmission(struct sock * sk)404 static void schedule_3rdack_retransmission(struct sock *sk)
405 {
406 	struct inet_connection_sock *icsk = inet_csk(sk);
407 	struct tcp_sock *tp = tcp_sk(sk);
408 	unsigned long timeout;
409 
410 	/* reschedule with a timeout above RTT, as we must look only for drop */
411 	if (tp->srtt_us)
412 		timeout = tp->srtt_us << 1;
413 	else
414 		timeout = TCP_TIMEOUT_INIT;
415 
416 	WARN_ON_ONCE(icsk->icsk_ack.pending & ICSK_ACK_TIMER);
417 	icsk->icsk_ack.pending |= ICSK_ACK_SCHED | ICSK_ACK_TIMER;
418 	icsk->icsk_ack.timeout = timeout;
419 	sk_reset_timer(sk, &icsk->icsk_delack_timer, timeout);
420 }
421 
clear_3rdack_retransmission(struct sock * sk)422 static void clear_3rdack_retransmission(struct sock *sk)
423 {
424 	struct inet_connection_sock *icsk = inet_csk(sk);
425 
426 	sk_stop_timer(sk, &icsk->icsk_delack_timer);
427 	icsk->icsk_ack.timeout = 0;
428 	icsk->icsk_ack.ato = 0;
429 	icsk->icsk_ack.pending &= ~(ICSK_ACK_SCHED | ICSK_ACK_TIMER);
430 }
431 
mptcp_established_options_mp(struct sock * sk,struct sk_buff * skb,bool snd_data_fin_enable,unsigned int * size,unsigned int remaining,struct mptcp_out_options * opts)432 static bool mptcp_established_options_mp(struct sock *sk, struct sk_buff *skb,
433 					 bool snd_data_fin_enable,
434 					 unsigned int *size,
435 					 unsigned int remaining,
436 					 struct mptcp_out_options *opts)
437 {
438 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
439 	struct mptcp_ext *mpext;
440 	unsigned int data_len;
441 
442 	/* When skb is not available, we better over-estimate the emitted
443 	 * options len. A full DSS option (28 bytes) is longer than
444 	 * TCPOLEN_MPTCP_MPC_ACK_DATA(22) or TCPOLEN_MPTCP_MPJ_ACK(24), so
445 	 * tell the caller to defer the estimate to
446 	 * mptcp_established_options_dss(), which will reserve enough space.
447 	 */
448 	if (!skb)
449 		return false;
450 
451 	/* MPC/MPJ needed only on 3rd ack packet, DATA_FIN and TCP shutdown take precedence */
452 	if (subflow->fully_established || snd_data_fin_enable ||
453 	    subflow->snd_isn != TCP_SKB_CB(skb)->seq ||
454 	    sk->sk_state != TCP_ESTABLISHED)
455 		return false;
456 
457 	if (subflow->mp_capable) {
458 		mpext = mptcp_get_ext(skb);
459 		data_len = mpext ? mpext->data_len : 0;
460 
461 		/* we will check ext_copy.data_len in mptcp_write_options() to
462 		 * discriminate between TCPOLEN_MPTCP_MPC_ACK_DATA and
463 		 * TCPOLEN_MPTCP_MPC_ACK
464 		 */
465 		opts->ext_copy.data_len = data_len;
466 		opts->suboptions = OPTION_MPTCP_MPC_ACK;
467 		opts->sndr_key = subflow->local_key;
468 		opts->rcvr_key = subflow->remote_key;
469 
470 		/* Section 3.1.
471 		 * The MP_CAPABLE option is carried on the SYN, SYN/ACK, and ACK
472 		 * packets that start the first subflow of an MPTCP connection,
473 		 * as well as the first packet that carries data
474 		 */
475 		if (data_len > 0)
476 			*size = ALIGN(TCPOLEN_MPTCP_MPC_ACK_DATA, 4);
477 		else
478 			*size = TCPOLEN_MPTCP_MPC_ACK;
479 
480 		pr_debug("subflow=%p, local_key=%llu, remote_key=%llu map_len=%d",
481 			 subflow, subflow->local_key, subflow->remote_key,
482 			 data_len);
483 
484 		return true;
485 	} else if (subflow->mp_join) {
486 		opts->suboptions = OPTION_MPTCP_MPJ_ACK;
487 		memcpy(opts->hmac, subflow->hmac, MPTCPOPT_HMAC_LEN);
488 		*size = TCPOLEN_MPTCP_MPJ_ACK;
489 		pr_debug("subflow=%p", subflow);
490 
491 		schedule_3rdack_retransmission(sk);
492 		return true;
493 	}
494 	return false;
495 }
496 
mptcp_write_data_fin(struct mptcp_subflow_context * subflow,struct sk_buff * skb,struct mptcp_ext * ext)497 static void mptcp_write_data_fin(struct mptcp_subflow_context *subflow,
498 				 struct sk_buff *skb, struct mptcp_ext *ext)
499 {
500 	/* The write_seq value has already been incremented, so the actual
501 	 * sequence number for the DATA_FIN is one less.
502 	 */
503 	u64 data_fin_tx_seq = READ_ONCE(mptcp_sk(subflow->conn)->write_seq) - 1;
504 
505 	if (!ext->use_map || !skb->len) {
506 		/* RFC6824 requires a DSS mapping with specific values
507 		 * if DATA_FIN is set but no data payload is mapped
508 		 */
509 		ext->data_fin = 1;
510 		ext->use_map = 1;
511 		ext->dsn64 = 1;
512 		ext->data_seq = data_fin_tx_seq;
513 		ext->subflow_seq = 0;
514 		ext->data_len = 1;
515 	} else if (ext->data_seq + ext->data_len == data_fin_tx_seq) {
516 		/* If there's an existing DSS mapping and it is the
517 		 * final mapping, DATA_FIN consumes 1 additional byte of
518 		 * mapping space.
519 		 */
520 		ext->data_fin = 1;
521 		ext->data_len++;
522 	}
523 }
524 
mptcp_established_options_dss(struct sock * sk,struct sk_buff * skb,bool snd_data_fin_enable,unsigned int * size,unsigned int remaining,struct mptcp_out_options * opts)525 static bool mptcp_established_options_dss(struct sock *sk, struct sk_buff *skb,
526 					  bool snd_data_fin_enable,
527 					  unsigned int *size,
528 					  unsigned int remaining,
529 					  struct mptcp_out_options *opts)
530 {
531 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
532 	struct mptcp_sock *msk = mptcp_sk(subflow->conn);
533 	unsigned int dss_size = 0;
534 	struct mptcp_ext *mpext;
535 	unsigned int ack_size;
536 	bool ret = false;
537 	u64 ack_seq;
538 
539 	mpext = skb ? mptcp_get_ext(skb) : NULL;
540 
541 	if (!skb || (mpext && mpext->use_map) || snd_data_fin_enable) {
542 		unsigned int map_size;
543 
544 		map_size = TCPOLEN_MPTCP_DSS_BASE + TCPOLEN_MPTCP_DSS_MAP64;
545 
546 		remaining -= map_size;
547 		dss_size = map_size;
548 		if (mpext)
549 			opts->ext_copy = *mpext;
550 
551 		if (skb && snd_data_fin_enable)
552 			mptcp_write_data_fin(subflow, skb, &opts->ext_copy);
553 		ret = true;
554 	}
555 
556 	/* passive sockets msk will set the 'can_ack' after accept(), even
557 	 * if the first subflow may have the already the remote key handy
558 	 */
559 	opts->ext_copy.use_ack = 0;
560 	if (!READ_ONCE(msk->can_ack)) {
561 		*size = ALIGN(dss_size, 4);
562 		return ret;
563 	}
564 
565 	ack_seq = READ_ONCE(msk->ack_seq);
566 	if (READ_ONCE(msk->use_64bit_ack)) {
567 		ack_size = TCPOLEN_MPTCP_DSS_ACK64;
568 		opts->ext_copy.data_ack = ack_seq;
569 		opts->ext_copy.ack64 = 1;
570 	} else {
571 		ack_size = TCPOLEN_MPTCP_DSS_ACK32;
572 		opts->ext_copy.data_ack32 = (uint32_t)ack_seq;
573 		opts->ext_copy.ack64 = 0;
574 	}
575 	opts->ext_copy.use_ack = 1;
576 	WRITE_ONCE(msk->old_wspace, __mptcp_space((struct sock *)msk));
577 
578 	/* Add kind/length/subtype/flag overhead if mapping is not populated */
579 	if (dss_size == 0)
580 		ack_size += TCPOLEN_MPTCP_DSS_BASE;
581 
582 	dss_size += ack_size;
583 
584 	*size = ALIGN(dss_size, 4);
585 	return true;
586 }
587 
add_addr_generate_hmac(u64 key1,u64 key2,struct mptcp_addr_info * addr)588 static u64 add_addr_generate_hmac(u64 key1, u64 key2,
589 				  struct mptcp_addr_info *addr)
590 {
591 	u16 port = ntohs(addr->port);
592 	u8 hmac[SHA256_DIGEST_SIZE];
593 	u8 msg[19];
594 	int i = 0;
595 
596 	msg[i++] = addr->id;
597 	if (addr->family == AF_INET) {
598 		memcpy(&msg[i], &addr->addr.s_addr, 4);
599 		i += 4;
600 	}
601 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
602 	else if (addr->family == AF_INET6) {
603 		memcpy(&msg[i], &addr->addr6.s6_addr, 16);
604 		i += 16;
605 	}
606 #endif
607 	msg[i++] = port >> 8;
608 	msg[i++] = port & 0xFF;
609 
610 	mptcp_crypto_hmac_sha(key1, key2, msg, i, hmac);
611 
612 	return get_unaligned_be64(&hmac[SHA256_DIGEST_SIZE - sizeof(u64)]);
613 }
614 
mptcp_established_options_add_addr(struct sock * sk,struct sk_buff * skb,unsigned int * size,unsigned int remaining,struct mptcp_out_options * opts)615 static bool mptcp_established_options_add_addr(struct sock *sk, struct sk_buff *skb,
616 					       unsigned int *size,
617 					       unsigned int remaining,
618 					       struct mptcp_out_options *opts)
619 {
620 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
621 	struct mptcp_sock *msk = mptcp_sk(subflow->conn);
622 	bool drop_other_suboptions = false;
623 	unsigned int opt_size = *size;
624 	bool echo;
625 	bool port;
626 	int len;
627 
628 	if ((mptcp_pm_should_add_signal_ipv6(msk) ||
629 	     mptcp_pm_should_add_signal_port(msk) ||
630 	     mptcp_pm_should_add_signal_echo(msk)) &&
631 	    skb && skb_is_tcp_pure_ack(skb)) {
632 		pr_debug("drop other suboptions");
633 		opts->suboptions = 0;
634 		opts->ext_copy.use_ack = 0;
635 		opts->ext_copy.use_map = 0;
636 		remaining += opt_size;
637 		drop_other_suboptions = true;
638 	}
639 
640 	if (!mptcp_pm_should_add_signal(msk) ||
641 	    !(mptcp_pm_add_addr_signal(msk, remaining, &opts->addr, &echo, &port)))
642 		return false;
643 
644 	len = mptcp_add_addr_len(opts->addr.family, echo, port);
645 	if (remaining < len)
646 		return false;
647 
648 	*size = len;
649 	if (drop_other_suboptions)
650 		*size -= opt_size;
651 	opts->suboptions |= OPTION_MPTCP_ADD_ADDR;
652 	if (!echo) {
653 		opts->ahmac = add_addr_generate_hmac(msk->local_key,
654 						     msk->remote_key,
655 						     &opts->addr);
656 	}
657 	pr_debug("addr_id=%d, ahmac=%llu, echo=%d, port=%d",
658 		 opts->addr.id, opts->ahmac, echo, ntohs(opts->addr.port));
659 
660 	return true;
661 }
662 
mptcp_established_options_rm_addr(struct sock * sk,unsigned int * size,unsigned int remaining,struct mptcp_out_options * opts)663 static bool mptcp_established_options_rm_addr(struct sock *sk,
664 					      unsigned int *size,
665 					      unsigned int remaining,
666 					      struct mptcp_out_options *opts)
667 {
668 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
669 	struct mptcp_sock *msk = mptcp_sk(subflow->conn);
670 	struct mptcp_rm_list rm_list;
671 	int i, len;
672 
673 	if (!mptcp_pm_should_rm_signal(msk) ||
674 	    !(mptcp_pm_rm_addr_signal(msk, remaining, &rm_list)))
675 		return false;
676 
677 	len = mptcp_rm_addr_len(&rm_list);
678 	if (len < 0)
679 		return false;
680 	if (remaining < len)
681 		return false;
682 
683 	*size = len;
684 	opts->suboptions |= OPTION_MPTCP_RM_ADDR;
685 	opts->rm_list = rm_list;
686 
687 	for (i = 0; i < opts->rm_list.nr; i++)
688 		pr_debug("rm_list_ids[%d]=%d", i, opts->rm_list.ids[i]);
689 
690 	return true;
691 }
692 
mptcp_established_options_mp_prio(struct sock * sk,unsigned int * size,unsigned int remaining,struct mptcp_out_options * opts)693 static bool mptcp_established_options_mp_prio(struct sock *sk,
694 					      unsigned int *size,
695 					      unsigned int remaining,
696 					      struct mptcp_out_options *opts)
697 {
698 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
699 
700 	if (!subflow->send_mp_prio)
701 		return false;
702 
703 	/* account for the trailing 'nop' option */
704 	if (remaining < TCPOLEN_MPTCP_PRIO_ALIGN)
705 		return false;
706 
707 	*size = TCPOLEN_MPTCP_PRIO_ALIGN;
708 	opts->suboptions |= OPTION_MPTCP_PRIO;
709 	opts->backup = subflow->request_bkup;
710 
711 	pr_debug("prio=%d", opts->backup);
712 
713 	return true;
714 }
715 
mptcp_established_options_rst(struct sock * sk,struct sk_buff * skb,unsigned int * size,unsigned int remaining,struct mptcp_out_options * opts)716 static noinline void mptcp_established_options_rst(struct sock *sk, struct sk_buff *skb,
717 						   unsigned int *size,
718 						   unsigned int remaining,
719 						   struct mptcp_out_options *opts)
720 {
721 	const struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
722 
723 	if (remaining < TCPOLEN_MPTCP_RST)
724 		return;
725 
726 	*size = TCPOLEN_MPTCP_RST;
727 	opts->suboptions |= OPTION_MPTCP_RST;
728 	opts->reset_transient = subflow->reset_transient;
729 	opts->reset_reason = subflow->reset_reason;
730 }
731 
mptcp_established_options(struct sock * sk,struct sk_buff * skb,unsigned int * size,unsigned int remaining,struct mptcp_out_options * opts)732 bool mptcp_established_options(struct sock *sk, struct sk_buff *skb,
733 			       unsigned int *size, unsigned int remaining,
734 			       struct mptcp_out_options *opts)
735 {
736 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
737 	struct mptcp_sock *msk = mptcp_sk(subflow->conn);
738 	unsigned int opt_size = 0;
739 	bool snd_data_fin;
740 	bool ret = false;
741 
742 	opts->suboptions = 0;
743 
744 	if (unlikely(__mptcp_check_fallback(msk)))
745 		return false;
746 
747 	if (unlikely(skb && TCP_SKB_CB(skb)->tcp_flags & TCPHDR_RST)) {
748 		mptcp_established_options_rst(sk, skb, size, remaining, opts);
749 		return true;
750 	}
751 
752 	snd_data_fin = mptcp_data_fin_enabled(msk);
753 	if (mptcp_established_options_mp(sk, skb, snd_data_fin, &opt_size, remaining, opts))
754 		ret = true;
755 	else if (mptcp_established_options_dss(sk, skb, snd_data_fin, &opt_size, remaining, opts))
756 		ret = true;
757 
758 	/* we reserved enough space for the above options, and exceeding the
759 	 * TCP option space would be fatal
760 	 */
761 	if (WARN_ON_ONCE(opt_size > remaining))
762 		return false;
763 
764 	*size += opt_size;
765 	remaining -= opt_size;
766 	if (mptcp_established_options_add_addr(sk, skb, &opt_size, remaining, opts)) {
767 		*size += opt_size;
768 		remaining -= opt_size;
769 		ret = true;
770 	} else if (mptcp_established_options_rm_addr(sk, &opt_size, remaining, opts)) {
771 		*size += opt_size;
772 		remaining -= opt_size;
773 		ret = true;
774 	}
775 
776 	if (mptcp_established_options_mp_prio(sk, &opt_size, remaining, opts)) {
777 		*size += opt_size;
778 		remaining -= opt_size;
779 		ret = true;
780 	}
781 
782 	return ret;
783 }
784 
mptcp_synack_options(const struct request_sock * req,unsigned int * size,struct mptcp_out_options * opts)785 bool mptcp_synack_options(const struct request_sock *req, unsigned int *size,
786 			  struct mptcp_out_options *opts)
787 {
788 	struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
789 
790 	if (subflow_req->mp_capable) {
791 		opts->suboptions = OPTION_MPTCP_MPC_SYNACK;
792 		opts->sndr_key = subflow_req->local_key;
793 		*size = TCPOLEN_MPTCP_MPC_SYNACK;
794 		pr_debug("subflow_req=%p, local_key=%llu",
795 			 subflow_req, subflow_req->local_key);
796 		return true;
797 	} else if (subflow_req->mp_join) {
798 		opts->suboptions = OPTION_MPTCP_MPJ_SYNACK;
799 		opts->backup = subflow_req->backup;
800 		opts->join_id = subflow_req->local_id;
801 		opts->thmac = subflow_req->thmac;
802 		opts->nonce = subflow_req->local_nonce;
803 		pr_debug("req=%p, bkup=%u, id=%u, thmac=%llu, nonce=%u",
804 			 subflow_req, opts->backup, opts->join_id,
805 			 opts->thmac, opts->nonce);
806 		*size = TCPOLEN_MPTCP_MPJ_SYNACK;
807 		return true;
808 	}
809 	return false;
810 }
811 
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)812 static bool check_fully_established(struct mptcp_sock *msk, struct sock *ssk,
813 				    struct mptcp_subflow_context *subflow,
814 				    struct sk_buff *skb,
815 				    struct mptcp_options_received *mp_opt)
816 {
817 	/* here we can process OoO, in-window pkts, only in-sequence 4th ack
818 	 * will make the subflow fully established
819 	 */
820 	if (likely(subflow->fully_established)) {
821 		/* on passive sockets, check for 3rd ack retransmission
822 		 * note that msk is always set by subflow_syn_recv_sock()
823 		 * for mp_join subflows
824 		 */
825 		if (TCP_SKB_CB(skb)->seq == subflow->ssn_offset + 1 &&
826 		    TCP_SKB_CB(skb)->end_seq == TCP_SKB_CB(skb)->seq &&
827 		    subflow->mp_join && mp_opt->mp_join &&
828 		    READ_ONCE(msk->pm.server_side))
829 			tcp_send_ack(ssk);
830 		goto fully_established;
831 	}
832 
833 	/* we must process OoO packets before the first subflow is fully
834 	 * established. OoO packets are instead a protocol violation
835 	 * for MP_JOIN subflows as the peer must not send any data
836 	 * before receiving the forth ack - cfr. RFC 8684 section 3.2.
837 	 */
838 	if (TCP_SKB_CB(skb)->seq != subflow->ssn_offset + 1) {
839 		if (subflow->mp_join)
840 			goto reset;
841 		return subflow->mp_capable;
842 	}
843 
844 	if (mp_opt->dss && mp_opt->use_ack) {
845 		/* subflows are fully established as soon as we get any
846 		 * additional ack.
847 		 */
848 		subflow->fully_established = 1;
849 		WRITE_ONCE(msk->fully_established, true);
850 		goto fully_established;
851 	}
852 
853 	if (mp_opt->add_addr) {
854 		WRITE_ONCE(msk->fully_established, true);
855 		return true;
856 	}
857 
858 	/* If the first established packet does not contain MP_CAPABLE + data
859 	 * then fallback to TCP. Fallback scenarios requires a reset for
860 	 * MP_JOIN subflows.
861 	 */
862 	if (!mp_opt->mp_capable) {
863 		if (subflow->mp_join)
864 			goto reset;
865 		subflow->mp_capable = 0;
866 		pr_fallback(msk);
867 		__mptcp_do_fallback(msk);
868 		return false;
869 	}
870 
871 	if (unlikely(!READ_ONCE(msk->pm.server_side)))
872 		pr_warn_once("bogus mpc option on established client sk");
873 	mptcp_subflow_fully_established(subflow, mp_opt);
874 
875 fully_established:
876 	/* if the subflow is not already linked into the conn_list, we can't
877 	 * notify the PM: this subflow is still on the listener queue
878 	 * and the PM possibly acquiring the subflow lock could race with
879 	 * the listener close
880 	 */
881 	if (likely(subflow->pm_notified) || list_empty(&subflow->node))
882 		return true;
883 
884 	subflow->pm_notified = 1;
885 	if (subflow->mp_join) {
886 		clear_3rdack_retransmission(ssk);
887 		mptcp_pm_subflow_established(msk);
888 	} else {
889 		mptcp_pm_fully_established(msk, ssk, GFP_ATOMIC);
890 	}
891 	return true;
892 
893 reset:
894 	mptcp_subflow_reset(ssk);
895 	return false;
896 }
897 
expand_ack(u64 old_ack,u64 cur_ack,bool use_64bit)898 static u64 expand_ack(u64 old_ack, u64 cur_ack, bool use_64bit)
899 {
900 	u32 old_ack32, cur_ack32;
901 
902 	if (use_64bit)
903 		return cur_ack;
904 
905 	old_ack32 = (u32)old_ack;
906 	cur_ack32 = (u32)cur_ack;
907 	cur_ack = (old_ack & GENMASK_ULL(63, 32)) + cur_ack32;
908 	if (unlikely(before(cur_ack32, old_ack32)))
909 		return cur_ack + (1LL << 32);
910 	return cur_ack;
911 }
912 
ack_update_msk(struct mptcp_sock * msk,struct sock * ssk,struct mptcp_options_received * mp_opt)913 static void ack_update_msk(struct mptcp_sock *msk,
914 			   struct sock *ssk,
915 			   struct mptcp_options_received *mp_opt)
916 {
917 	u64 new_wnd_end, new_snd_una, snd_nxt = READ_ONCE(msk->snd_nxt);
918 	struct sock *sk = (struct sock *)msk;
919 	u64 old_snd_una;
920 
921 	mptcp_data_lock(sk);
922 
923 	/* avoid ack expansion on update conflict, to reduce the risk of
924 	 * wrongly expanding to a future ack sequence number, which is way
925 	 * more dangerous than missing an ack
926 	 */
927 	old_snd_una = msk->snd_una;
928 	new_snd_una = expand_ack(old_snd_una, mp_opt->data_ack, mp_opt->ack64);
929 
930 	/* ACK for data not even sent yet? Ignore. */
931 	if (after64(new_snd_una, snd_nxt))
932 		new_snd_una = old_snd_una;
933 
934 	new_wnd_end = new_snd_una + tcp_sk(ssk)->snd_wnd;
935 
936 	if (after64(new_wnd_end, msk->wnd_end))
937 		msk->wnd_end = new_wnd_end;
938 
939 	/* this assumes mptcp_incoming_options() is invoked after tcp_ack() */
940 	if (after64(msk->wnd_end, READ_ONCE(msk->snd_nxt)))
941 		__mptcp_check_push(sk, ssk);
942 
943 	if (after64(new_snd_una, old_snd_una)) {
944 		msk->snd_una = new_snd_una;
945 		__mptcp_data_acked(sk);
946 	}
947 	mptcp_data_unlock(sk);
948 
949 	trace_ack_update_msk(mp_opt->data_ack,
950 			     old_snd_una, new_snd_una,
951 			     new_wnd_end, msk->wnd_end);
952 }
953 
mptcp_update_rcv_data_fin(struct mptcp_sock * msk,u64 data_fin_seq,bool use_64bit)954 bool mptcp_update_rcv_data_fin(struct mptcp_sock *msk, u64 data_fin_seq, bool use_64bit)
955 {
956 	/* Skip if DATA_FIN was already received.
957 	 * If updating simultaneously with the recvmsg loop, values
958 	 * should match. If they mismatch, the peer is misbehaving and
959 	 * we will prefer the most recent information.
960 	 */
961 	if (READ_ONCE(msk->rcv_data_fin))
962 		return false;
963 
964 	WRITE_ONCE(msk->rcv_data_fin_seq,
965 		   expand_ack(READ_ONCE(msk->ack_seq), data_fin_seq, use_64bit));
966 	WRITE_ONCE(msk->rcv_data_fin, 1);
967 
968 	return true;
969 }
970 
add_addr_hmac_valid(struct mptcp_sock * msk,struct mptcp_options_received * mp_opt)971 static bool add_addr_hmac_valid(struct mptcp_sock *msk,
972 				struct mptcp_options_received *mp_opt)
973 {
974 	u64 hmac = 0;
975 
976 	if (mp_opt->echo)
977 		return true;
978 
979 	hmac = add_addr_generate_hmac(msk->remote_key,
980 				      msk->local_key,
981 				      &mp_opt->addr);
982 
983 	pr_debug("msk=%p, ahmac=%llu, mp_opt->ahmac=%llu\n",
984 		 msk, (unsigned long long)hmac,
985 		 (unsigned long long)mp_opt->ahmac);
986 
987 	return hmac == mp_opt->ahmac;
988 }
989 
mptcp_incoming_options(struct sock * sk,struct sk_buff * skb)990 void mptcp_incoming_options(struct sock *sk, struct sk_buff *skb)
991 {
992 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
993 	struct mptcp_sock *msk = mptcp_sk(subflow->conn);
994 	struct mptcp_options_received mp_opt;
995 	struct mptcp_ext *mpext;
996 
997 	if (__mptcp_check_fallback(msk)) {
998 		/* Keep it simple and unconditionally trigger send data cleanup and
999 		 * pending queue spooling. We will need to acquire the data lock
1000 		 * for more accurate checks, and once the lock is acquired, such
1001 		 * helpers are cheap.
1002 		 */
1003 		mptcp_data_lock(subflow->conn);
1004 		if (sk_stream_memory_free(sk))
1005 			__mptcp_check_push(subflow->conn, sk);
1006 		__mptcp_data_acked(subflow->conn);
1007 		mptcp_data_unlock(subflow->conn);
1008 		return;
1009 	}
1010 
1011 	mptcp_get_options(skb, &mp_opt);
1012 	if (!check_fully_established(msk, sk, subflow, skb, &mp_opt))
1013 		return;
1014 
1015 	if (mp_opt.fastclose &&
1016 	    msk->local_key == mp_opt.rcvr_key) {
1017 		WRITE_ONCE(msk->rcv_fastclose, true);
1018 		mptcp_schedule_work((struct sock *)msk);
1019 	}
1020 
1021 	if (mp_opt.add_addr && add_addr_hmac_valid(msk, &mp_opt)) {
1022 		if (!mp_opt.echo) {
1023 			mptcp_pm_add_addr_received(msk, &mp_opt.addr);
1024 			MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ADDADDR);
1025 		} else {
1026 			mptcp_pm_add_addr_echoed(msk, &mp_opt.addr);
1027 			mptcp_pm_del_add_timer(msk, &mp_opt.addr);
1028 			MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ECHOADD);
1029 		}
1030 
1031 		if (mp_opt.addr.port)
1032 			MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_PORTADD);
1033 
1034 		mp_opt.add_addr = 0;
1035 	}
1036 
1037 	if (mp_opt.rm_addr) {
1038 		mptcp_pm_rm_addr_received(msk, &mp_opt.rm_list);
1039 		mp_opt.rm_addr = 0;
1040 	}
1041 
1042 	if (mp_opt.mp_prio) {
1043 		mptcp_pm_mp_prio_received(sk, mp_opt.backup);
1044 		MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPPRIORX);
1045 		mp_opt.mp_prio = 0;
1046 	}
1047 
1048 	if (mp_opt.reset) {
1049 		subflow->reset_seen = 1;
1050 		subflow->reset_reason = mp_opt.reset_reason;
1051 		subflow->reset_transient = mp_opt.reset_transient;
1052 	}
1053 
1054 	if (!mp_opt.dss)
1055 		return;
1056 
1057 	/* we can't wait for recvmsg() to update the ack_seq, otherwise
1058 	 * monodirectional flows will stuck
1059 	 */
1060 	if (mp_opt.use_ack)
1061 		ack_update_msk(msk, sk, &mp_opt);
1062 
1063 	/* Zero-data-length packets are dropped by the caller and not
1064 	 * propagated to the MPTCP layer, so the skb extension does not
1065 	 * need to be allocated or populated. DATA_FIN information, if
1066 	 * present, needs to be updated here before the skb is freed.
1067 	 */
1068 	if (TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb)->end_seq) {
1069 		if (mp_opt.data_fin && mp_opt.data_len == 1 &&
1070 		    mptcp_update_rcv_data_fin(msk, mp_opt.data_seq, mp_opt.dsn64) &&
1071 		    schedule_work(&msk->work))
1072 			sock_hold(subflow->conn);
1073 
1074 		return;
1075 	}
1076 
1077 	mpext = skb_ext_add(skb, SKB_EXT_MPTCP);
1078 	if (!mpext)
1079 		return;
1080 
1081 	memset(mpext, 0, sizeof(*mpext));
1082 
1083 	if (mp_opt.use_map) {
1084 		if (mp_opt.mpc_map) {
1085 			/* this is an MP_CAPABLE carrying MPTCP data
1086 			 * we know this map the first chunk of data
1087 			 */
1088 			mptcp_crypto_key_sha(subflow->remote_key, NULL,
1089 					     &mpext->data_seq);
1090 			mpext->data_seq++;
1091 			mpext->subflow_seq = 1;
1092 			mpext->dsn64 = 1;
1093 			mpext->mpc_map = 1;
1094 			mpext->data_fin = 0;
1095 		} else {
1096 			mpext->data_seq = mp_opt.data_seq;
1097 			mpext->subflow_seq = mp_opt.subflow_seq;
1098 			mpext->dsn64 = mp_opt.dsn64;
1099 			mpext->data_fin = mp_opt.data_fin;
1100 		}
1101 		mpext->data_len = mp_opt.data_len;
1102 		mpext->use_map = 1;
1103 	}
1104 }
1105 
mptcp_set_rwin(const struct tcp_sock * tp)1106 static void mptcp_set_rwin(const struct tcp_sock *tp)
1107 {
1108 	const struct sock *ssk = (const struct sock *)tp;
1109 	const struct mptcp_subflow_context *subflow;
1110 	struct mptcp_sock *msk;
1111 	u64 ack_seq;
1112 
1113 	subflow = mptcp_subflow_ctx(ssk);
1114 	msk = mptcp_sk(subflow->conn);
1115 
1116 	ack_seq = READ_ONCE(msk->ack_seq) + tp->rcv_wnd;
1117 
1118 	if (after64(ack_seq, READ_ONCE(msk->rcv_wnd_sent)))
1119 		WRITE_ONCE(msk->rcv_wnd_sent, ack_seq);
1120 }
1121 
mptcp_write_options(__be32 * ptr,const struct tcp_sock * tp,struct mptcp_out_options * opts)1122 void mptcp_write_options(__be32 *ptr, const struct tcp_sock *tp,
1123 			 struct mptcp_out_options *opts)
1124 {
1125 	if ((OPTION_MPTCP_MPC_SYN | OPTION_MPTCP_MPC_SYNACK |
1126 	     OPTION_MPTCP_MPC_ACK) & opts->suboptions) {
1127 		u8 len;
1128 
1129 		if (OPTION_MPTCP_MPC_SYN & opts->suboptions)
1130 			len = TCPOLEN_MPTCP_MPC_SYN;
1131 		else if (OPTION_MPTCP_MPC_SYNACK & opts->suboptions)
1132 			len = TCPOLEN_MPTCP_MPC_SYNACK;
1133 		else if (opts->ext_copy.data_len)
1134 			len = TCPOLEN_MPTCP_MPC_ACK_DATA;
1135 		else
1136 			len = TCPOLEN_MPTCP_MPC_ACK;
1137 
1138 		*ptr++ = mptcp_option(MPTCPOPT_MP_CAPABLE, len,
1139 				      MPTCP_SUPPORTED_VERSION,
1140 				      MPTCP_CAP_HMAC_SHA256);
1141 
1142 		if (!((OPTION_MPTCP_MPC_SYNACK | OPTION_MPTCP_MPC_ACK) &
1143 		    opts->suboptions))
1144 			goto mp_capable_done;
1145 
1146 		put_unaligned_be64(opts->sndr_key, ptr);
1147 		ptr += 2;
1148 		if (!((OPTION_MPTCP_MPC_ACK) & opts->suboptions))
1149 			goto mp_capable_done;
1150 
1151 		put_unaligned_be64(opts->rcvr_key, ptr);
1152 		ptr += 2;
1153 		if (!opts->ext_copy.data_len)
1154 			goto mp_capable_done;
1155 
1156 		put_unaligned_be32(opts->ext_copy.data_len << 16 |
1157 				   TCPOPT_NOP << 8 | TCPOPT_NOP, ptr);
1158 		ptr += 1;
1159 	}
1160 
1161 mp_capable_done:
1162 	if (OPTION_MPTCP_ADD_ADDR & opts->suboptions) {
1163 		u8 len = TCPOLEN_MPTCP_ADD_ADDR_BASE;
1164 		u8 echo = MPTCP_ADDR_ECHO;
1165 
1166 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1167 		if (opts->addr.family == AF_INET6)
1168 			len = TCPOLEN_MPTCP_ADD_ADDR6_BASE;
1169 #endif
1170 
1171 		if (opts->addr.port)
1172 			len += TCPOLEN_MPTCP_PORT_LEN;
1173 
1174 		if (opts->ahmac) {
1175 			len += sizeof(opts->ahmac);
1176 			echo = 0;
1177 		}
1178 
1179 		*ptr++ = mptcp_option(MPTCPOPT_ADD_ADDR,
1180 				      len, echo, opts->addr.id);
1181 		if (opts->addr.family == AF_INET) {
1182 			memcpy((u8 *)ptr, (u8 *)&opts->addr.addr.s_addr, 4);
1183 			ptr += 1;
1184 		}
1185 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1186 		else if (opts->addr.family == AF_INET6) {
1187 			memcpy((u8 *)ptr, opts->addr.addr6.s6_addr, 16);
1188 			ptr += 4;
1189 		}
1190 #endif
1191 
1192 		if (!opts->addr.port) {
1193 			if (opts->ahmac) {
1194 				put_unaligned_be64(opts->ahmac, ptr);
1195 				ptr += 2;
1196 			}
1197 		} else {
1198 			u16 port = ntohs(opts->addr.port);
1199 
1200 			if (opts->ahmac) {
1201 				u8 *bptr = (u8 *)ptr;
1202 
1203 				put_unaligned_be16(port, bptr);
1204 				bptr += 2;
1205 				put_unaligned_be64(opts->ahmac, bptr);
1206 				bptr += 8;
1207 				put_unaligned_be16(TCPOPT_NOP << 8 |
1208 						   TCPOPT_NOP, bptr);
1209 
1210 				ptr += 3;
1211 			} else {
1212 				put_unaligned_be32(port << 16 |
1213 						   TCPOPT_NOP << 8 |
1214 						   TCPOPT_NOP, ptr);
1215 				ptr += 1;
1216 			}
1217 		}
1218 	}
1219 
1220 	if (OPTION_MPTCP_RM_ADDR & opts->suboptions) {
1221 		u8 i = 1;
1222 
1223 		*ptr++ = mptcp_option(MPTCPOPT_RM_ADDR,
1224 				      TCPOLEN_MPTCP_RM_ADDR_BASE + opts->rm_list.nr,
1225 				      0, opts->rm_list.ids[0]);
1226 
1227 		while (i < opts->rm_list.nr) {
1228 			u8 id1, id2, id3, id4;
1229 
1230 			id1 = opts->rm_list.ids[i];
1231 			id2 = i + 1 < opts->rm_list.nr ? opts->rm_list.ids[i + 1] : TCPOPT_NOP;
1232 			id3 = i + 2 < opts->rm_list.nr ? opts->rm_list.ids[i + 2] : TCPOPT_NOP;
1233 			id4 = i + 3 < opts->rm_list.nr ? opts->rm_list.ids[i + 3] : TCPOPT_NOP;
1234 			put_unaligned_be32(id1 << 24 | id2 << 16 | id3 << 8 | id4, ptr);
1235 			ptr += 1;
1236 			i += 4;
1237 		}
1238 	}
1239 
1240 	if (OPTION_MPTCP_PRIO & opts->suboptions) {
1241 		const struct sock *ssk = (const struct sock *)tp;
1242 		struct mptcp_subflow_context *subflow;
1243 
1244 		subflow = mptcp_subflow_ctx(ssk);
1245 		subflow->send_mp_prio = 0;
1246 
1247 		*ptr++ = mptcp_option(MPTCPOPT_MP_PRIO,
1248 				      TCPOLEN_MPTCP_PRIO,
1249 				      opts->backup, TCPOPT_NOP);
1250 	}
1251 
1252 	if (OPTION_MPTCP_MPJ_SYN & opts->suboptions) {
1253 		*ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1254 				      TCPOLEN_MPTCP_MPJ_SYN,
1255 				      opts->backup, opts->join_id);
1256 		put_unaligned_be32(opts->token, ptr);
1257 		ptr += 1;
1258 		put_unaligned_be32(opts->nonce, ptr);
1259 		ptr += 1;
1260 	}
1261 
1262 	if (OPTION_MPTCP_MPJ_SYNACK & opts->suboptions) {
1263 		*ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1264 				      TCPOLEN_MPTCP_MPJ_SYNACK,
1265 				      opts->backup, opts->join_id);
1266 		put_unaligned_be64(opts->thmac, ptr);
1267 		ptr += 2;
1268 		put_unaligned_be32(opts->nonce, ptr);
1269 		ptr += 1;
1270 	}
1271 
1272 	if (OPTION_MPTCP_MPJ_ACK & opts->suboptions) {
1273 		*ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1274 				      TCPOLEN_MPTCP_MPJ_ACK, 0, 0);
1275 		memcpy(ptr, opts->hmac, MPTCPOPT_HMAC_LEN);
1276 		ptr += 5;
1277 	}
1278 
1279 	if (OPTION_MPTCP_RST & opts->suboptions)
1280 		*ptr++ = mptcp_option(MPTCPOPT_RST,
1281 				      TCPOLEN_MPTCP_RST,
1282 				      opts->reset_transient,
1283 				      opts->reset_reason);
1284 
1285 	if (opts->ext_copy.use_ack || opts->ext_copy.use_map) {
1286 		struct mptcp_ext *mpext = &opts->ext_copy;
1287 		u8 len = TCPOLEN_MPTCP_DSS_BASE;
1288 		u8 flags = 0;
1289 
1290 		if (mpext->use_ack) {
1291 			flags = MPTCP_DSS_HAS_ACK;
1292 			if (mpext->ack64) {
1293 				len += TCPOLEN_MPTCP_DSS_ACK64;
1294 				flags |= MPTCP_DSS_ACK64;
1295 			} else {
1296 				len += TCPOLEN_MPTCP_DSS_ACK32;
1297 			}
1298 		}
1299 
1300 		if (mpext->use_map) {
1301 			len += TCPOLEN_MPTCP_DSS_MAP64;
1302 
1303 			/* Use only 64-bit mapping flags for now, add
1304 			 * support for optional 32-bit mappings later.
1305 			 */
1306 			flags |= MPTCP_DSS_HAS_MAP | MPTCP_DSS_DSN64;
1307 			if (mpext->data_fin)
1308 				flags |= MPTCP_DSS_DATA_FIN;
1309 		}
1310 
1311 		*ptr++ = mptcp_option(MPTCPOPT_DSS, len, 0, flags);
1312 
1313 		if (mpext->use_ack) {
1314 			if (mpext->ack64) {
1315 				put_unaligned_be64(mpext->data_ack, ptr);
1316 				ptr += 2;
1317 			} else {
1318 				put_unaligned_be32(mpext->data_ack32, ptr);
1319 				ptr += 1;
1320 			}
1321 		}
1322 
1323 		if (mpext->use_map) {
1324 			put_unaligned_be64(mpext->data_seq, ptr);
1325 			ptr += 2;
1326 			put_unaligned_be32(mpext->subflow_seq, ptr);
1327 			ptr += 1;
1328 			put_unaligned_be32(mpext->data_len << 16 |
1329 					   TCPOPT_NOP << 8 | TCPOPT_NOP, ptr);
1330 		}
1331 	}
1332 
1333 	if (tp)
1334 		mptcp_set_rwin(tp);
1335 }
1336 
mptcp_get_reset_option(const struct sk_buff * skb)1337 __be32 mptcp_get_reset_option(const struct sk_buff *skb)
1338 {
1339 	const struct mptcp_ext *ext = mptcp_get_ext(skb);
1340 	u8 flags, reason;
1341 
1342 	if (ext) {
1343 		flags = ext->reset_transient;
1344 		reason = ext->reset_reason;
1345 
1346 		return mptcp_option(MPTCPOPT_RST, TCPOLEN_MPTCP_RST,
1347 				    flags, reason);
1348 	}
1349 
1350 	return htonl(0u);
1351 }
1352 EXPORT_SYMBOL_GPL(mptcp_get_reset_option);
1353