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