Loading...
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/sha.h>
11#include <net/tcp.h>
12#include <net/mptcp.h>
13#include "protocol.h"
14
15static bool mptcp_cap_flag_sha256(u8 flags)
16{
17 return (flags & MPTCP_CAP_FLAG_MASK) == MPTCP_CAP_HMAC_SHA256;
18}
19
20static void mptcp_parse_option(const struct sk_buff *skb,
21 const unsigned char *ptr, int opsize,
22 struct mptcp_options_received *mp_opt)
23{
24 u8 subtype = *ptr >> 4;
25 int expected_opsize;
26 u8 version;
27 u8 flags;
28
29 switch (subtype) {
30 case MPTCPOPT_MP_CAPABLE:
31 /* strict size checking */
32 if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)) {
33 if (skb->len > tcp_hdr(skb)->doff << 2)
34 expected_opsize = TCPOLEN_MPTCP_MPC_ACK_DATA;
35 else
36 expected_opsize = TCPOLEN_MPTCP_MPC_ACK;
37 } else {
38 if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_ACK)
39 expected_opsize = TCPOLEN_MPTCP_MPC_SYNACK;
40 else
41 expected_opsize = TCPOLEN_MPTCP_MPC_SYN;
42 }
43 if (opsize != expected_opsize)
44 break;
45
46 /* try to be gentle vs future versions on the initial syn */
47 version = *ptr++ & MPTCP_VERSION_MASK;
48 if (opsize != TCPOLEN_MPTCP_MPC_SYN) {
49 if (version != MPTCP_SUPPORTED_VERSION)
50 break;
51 } else if (version < MPTCP_SUPPORTED_VERSION) {
52 break;
53 }
54
55 flags = *ptr++;
56 if (!mptcp_cap_flag_sha256(flags) ||
57 (flags & MPTCP_CAP_EXTENSIBILITY))
58 break;
59
60 /* RFC 6824, Section 3.1:
61 * "For the Checksum Required bit (labeled "A"), if either
62 * host requires the use of checksums, checksums MUST be used.
63 * In other words, the only way for checksums not to be used
64 * is if both hosts in their SYNs set A=0."
65 *
66 * Section 3.3.0:
67 * "If a checksum is not present when its use has been
68 * negotiated, the receiver MUST close the subflow with a RST as
69 * it is considered broken."
70 *
71 * We don't implement DSS checksum - fall back to TCP.
72 */
73 if (flags & MPTCP_CAP_CHECKSUM_REQD)
74 break;
75
76 mp_opt->mp_capable = 1;
77 if (opsize >= TCPOLEN_MPTCP_MPC_SYNACK) {
78 mp_opt->sndr_key = get_unaligned_be64(ptr);
79 ptr += 8;
80 }
81 if (opsize >= TCPOLEN_MPTCP_MPC_ACK) {
82 mp_opt->rcvr_key = get_unaligned_be64(ptr);
83 ptr += 8;
84 }
85 if (opsize == TCPOLEN_MPTCP_MPC_ACK_DATA) {
86 /* Section 3.1.:
87 * "the data parameters in a MP_CAPABLE are semantically
88 * equivalent to those in a DSS option and can be used
89 * interchangeably."
90 */
91 mp_opt->dss = 1;
92 mp_opt->use_map = 1;
93 mp_opt->mpc_map = 1;
94 mp_opt->data_len = get_unaligned_be16(ptr);
95 ptr += 2;
96 }
97 pr_debug("MP_CAPABLE version=%x, flags=%x, optlen=%d sndr=%llu, rcvr=%llu len=%d",
98 version, flags, opsize, mp_opt->sndr_key,
99 mp_opt->rcvr_key, mp_opt->data_len);
100 break;
101
102 case MPTCPOPT_MP_JOIN:
103 mp_opt->mp_join = 1;
104 if (opsize == TCPOLEN_MPTCP_MPJ_SYN) {
105 mp_opt->backup = *ptr++ & MPTCPOPT_BACKUP;
106 mp_opt->join_id = *ptr++;
107 mp_opt->token = get_unaligned_be32(ptr);
108 ptr += 4;
109 mp_opt->nonce = get_unaligned_be32(ptr);
110 ptr += 4;
111 pr_debug("MP_JOIN bkup=%u, id=%u, token=%u, nonce=%u",
112 mp_opt->backup, mp_opt->join_id,
113 mp_opt->token, mp_opt->nonce);
114 } else if (opsize == TCPOLEN_MPTCP_MPJ_SYNACK) {
115 mp_opt->backup = *ptr++ & MPTCPOPT_BACKUP;
116 mp_opt->join_id = *ptr++;
117 mp_opt->thmac = get_unaligned_be64(ptr);
118 ptr += 8;
119 mp_opt->nonce = get_unaligned_be32(ptr);
120 ptr += 4;
121 pr_debug("MP_JOIN bkup=%u, id=%u, thmac=%llu, nonce=%u",
122 mp_opt->backup, mp_opt->join_id,
123 mp_opt->thmac, mp_opt->nonce);
124 } else if (opsize == TCPOLEN_MPTCP_MPJ_ACK) {
125 ptr += 2;
126 memcpy(mp_opt->hmac, ptr, MPTCPOPT_HMAC_LEN);
127 pr_debug("MP_JOIN hmac");
128 } else {
129 pr_warn("MP_JOIN bad option size");
130 mp_opt->mp_join = 0;
131 }
132 break;
133
134 case MPTCPOPT_DSS:
135 pr_debug("DSS");
136 ptr++;
137
138 /* we must clear 'mpc_map' be able to detect MP_CAPABLE
139 * map vs DSS map in mptcp_incoming_options(), and reconstruct
140 * map info accordingly
141 */
142 mp_opt->mpc_map = 0;
143 flags = (*ptr++) & MPTCP_DSS_FLAG_MASK;
144 mp_opt->data_fin = (flags & MPTCP_DSS_DATA_FIN) != 0;
145 mp_opt->dsn64 = (flags & MPTCP_DSS_DSN64) != 0;
146 mp_opt->use_map = (flags & MPTCP_DSS_HAS_MAP) != 0;
147 mp_opt->ack64 = (flags & MPTCP_DSS_ACK64) != 0;
148 mp_opt->use_ack = (flags & MPTCP_DSS_HAS_ACK);
149
150 pr_debug("data_fin=%d dsn64=%d use_map=%d ack64=%d use_ack=%d",
151 mp_opt->data_fin, mp_opt->dsn64,
152 mp_opt->use_map, mp_opt->ack64,
153 mp_opt->use_ack);
154
155 expected_opsize = TCPOLEN_MPTCP_DSS_BASE;
156
157 if (mp_opt->use_ack) {
158 if (mp_opt->ack64)
159 expected_opsize += TCPOLEN_MPTCP_DSS_ACK64;
160 else
161 expected_opsize += TCPOLEN_MPTCP_DSS_ACK32;
162 }
163
164 if (mp_opt->use_map) {
165 if (mp_opt->dsn64)
166 expected_opsize += TCPOLEN_MPTCP_DSS_MAP64;
167 else
168 expected_opsize += TCPOLEN_MPTCP_DSS_MAP32;
169 }
170
171 /* RFC 6824, Section 3.3:
172 * If a checksum is present, but its use had
173 * not been negotiated in the MP_CAPABLE handshake,
174 * the checksum field MUST be ignored.
175 */
176 if (opsize != expected_opsize &&
177 opsize != expected_opsize + TCPOLEN_MPTCP_DSS_CHECKSUM)
178 break;
179
180 mp_opt->dss = 1;
181
182 if (mp_opt->use_ack) {
183 if (mp_opt->ack64) {
184 mp_opt->data_ack = get_unaligned_be64(ptr);
185 ptr += 8;
186 } else {
187 mp_opt->data_ack = get_unaligned_be32(ptr);
188 ptr += 4;
189 }
190
191 pr_debug("data_ack=%llu", mp_opt->data_ack);
192 }
193
194 if (mp_opt->use_map) {
195 if (mp_opt->dsn64) {
196 mp_opt->data_seq = get_unaligned_be64(ptr);
197 ptr += 8;
198 } else {
199 mp_opt->data_seq = get_unaligned_be32(ptr);
200 ptr += 4;
201 }
202
203 mp_opt->subflow_seq = get_unaligned_be32(ptr);
204 ptr += 4;
205
206 mp_opt->data_len = get_unaligned_be16(ptr);
207 ptr += 2;
208
209 pr_debug("data_seq=%llu subflow_seq=%u data_len=%u",
210 mp_opt->data_seq, mp_opt->subflow_seq,
211 mp_opt->data_len);
212 }
213
214 break;
215
216 case MPTCPOPT_ADD_ADDR:
217 mp_opt->echo = (*ptr++) & MPTCP_ADDR_ECHO;
218 if (!mp_opt->echo) {
219 if (opsize == TCPOLEN_MPTCP_ADD_ADDR ||
220 opsize == TCPOLEN_MPTCP_ADD_ADDR_PORT)
221 mp_opt->family = MPTCP_ADDR_IPVERSION_4;
222#if IS_ENABLED(CONFIG_MPTCP_IPV6)
223 else if (opsize == TCPOLEN_MPTCP_ADD_ADDR6 ||
224 opsize == TCPOLEN_MPTCP_ADD_ADDR6_PORT)
225 mp_opt->family = MPTCP_ADDR_IPVERSION_6;
226#endif
227 else
228 break;
229 } else {
230 if (opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE ||
231 opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE_PORT)
232 mp_opt->family = MPTCP_ADDR_IPVERSION_4;
233#if IS_ENABLED(CONFIG_MPTCP_IPV6)
234 else if (opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE ||
235 opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE_PORT)
236 mp_opt->family = MPTCP_ADDR_IPVERSION_6;
237#endif
238 else
239 break;
240 }
241
242 mp_opt->add_addr = 1;
243 mp_opt->port = 0;
244 mp_opt->addr_id = *ptr++;
245 pr_debug("ADD_ADDR: id=%d", mp_opt->addr_id);
246 if (mp_opt->family == MPTCP_ADDR_IPVERSION_4) {
247 memcpy((u8 *)&mp_opt->addr.s_addr, (u8 *)ptr, 4);
248 ptr += 4;
249 if (opsize == TCPOLEN_MPTCP_ADD_ADDR_PORT ||
250 opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE_PORT) {
251 mp_opt->port = get_unaligned_be16(ptr);
252 ptr += 2;
253 }
254 }
255#if IS_ENABLED(CONFIG_MPTCP_IPV6)
256 else {
257 memcpy(mp_opt->addr6.s6_addr, (u8 *)ptr, 16);
258 ptr += 16;
259 if (opsize == TCPOLEN_MPTCP_ADD_ADDR6_PORT ||
260 opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE_PORT) {
261 mp_opt->port = get_unaligned_be16(ptr);
262 ptr += 2;
263 }
264 }
265#endif
266 if (!mp_opt->echo) {
267 mp_opt->ahmac = get_unaligned_be64(ptr);
268 ptr += 8;
269 }
270 break;
271
272 case MPTCPOPT_RM_ADDR:
273 if (opsize != TCPOLEN_MPTCP_RM_ADDR_BASE)
274 break;
275
276 ptr++;
277
278 mp_opt->rm_addr = 1;
279 mp_opt->rm_id = *ptr++;
280 pr_debug("RM_ADDR: id=%d", mp_opt->rm_id);
281 break;
282
283 default:
284 break;
285 }
286}
287
288void mptcp_get_options(const struct sk_buff *skb,
289 struct mptcp_options_received *mp_opt)
290{
291 const struct tcphdr *th = tcp_hdr(skb);
292 const unsigned char *ptr;
293 int length;
294
295 /* initialize option status */
296 mp_opt->mp_capable = 0;
297 mp_opt->mp_join = 0;
298 mp_opt->add_addr = 0;
299 mp_opt->rm_addr = 0;
300 mp_opt->dss = 0;
301
302 length = (th->doff * 4) - sizeof(struct tcphdr);
303 ptr = (const unsigned char *)(th + 1);
304
305 while (length > 0) {
306 int opcode = *ptr++;
307 int opsize;
308
309 switch (opcode) {
310 case TCPOPT_EOL:
311 return;
312 case TCPOPT_NOP: /* Ref: RFC 793 section 3.1 */
313 length--;
314 continue;
315 default:
316 opsize = *ptr++;
317 if (opsize < 2) /* "silly options" */
318 return;
319 if (opsize > length)
320 return; /* don't parse partial options */
321 if (opcode == TCPOPT_MPTCP)
322 mptcp_parse_option(skb, ptr, opsize, mp_opt);
323 ptr += opsize - 2;
324 length -= opsize;
325 }
326 }
327}
328
329bool mptcp_syn_options(struct sock *sk, const struct sk_buff *skb,
330 unsigned int *size, struct mptcp_out_options *opts)
331{
332 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
333
334 /* we will use snd_isn to detect first pkt [re]transmission
335 * in mptcp_established_options_mp()
336 */
337 subflow->snd_isn = TCP_SKB_CB(skb)->end_seq;
338 if (subflow->request_mptcp) {
339 opts->suboptions = OPTION_MPTCP_MPC_SYN;
340 *size = TCPOLEN_MPTCP_MPC_SYN;
341 return true;
342 } else if (subflow->request_join) {
343 pr_debug("remote_token=%u, nonce=%u", subflow->remote_token,
344 subflow->local_nonce);
345 opts->suboptions = OPTION_MPTCP_MPJ_SYN;
346 opts->join_id = subflow->local_id;
347 opts->token = subflow->remote_token;
348 opts->nonce = subflow->local_nonce;
349 opts->backup = subflow->request_bkup;
350 *size = TCPOLEN_MPTCP_MPJ_SYN;
351 return true;
352 }
353 return false;
354}
355
356/* MP_JOIN client subflow must wait for 4th ack before sending any data:
357 * TCP can't schedule delack timer before the subflow is fully established.
358 * MPTCP uses the delack timer to do 3rd ack retransmissions
359 */
360static void schedule_3rdack_retransmission(struct sock *sk)
361{
362 struct inet_connection_sock *icsk = inet_csk(sk);
363 struct tcp_sock *tp = tcp_sk(sk);
364 unsigned long timeout;
365
366 /* reschedule with a timeout above RTT, as we must look only for drop */
367 if (tp->srtt_us)
368 timeout = tp->srtt_us << 1;
369 else
370 timeout = TCP_TIMEOUT_INIT;
371
372 WARN_ON_ONCE(icsk->icsk_ack.pending & ICSK_ACK_TIMER);
373 icsk->icsk_ack.pending |= ICSK_ACK_SCHED | ICSK_ACK_TIMER;
374 icsk->icsk_ack.timeout = timeout;
375 sk_reset_timer(sk, &icsk->icsk_delack_timer, timeout);
376}
377
378static void clear_3rdack_retransmission(struct sock *sk)
379{
380 struct inet_connection_sock *icsk = inet_csk(sk);
381
382 sk_stop_timer(sk, &icsk->icsk_delack_timer);
383 icsk->icsk_ack.timeout = 0;
384 icsk->icsk_ack.ato = 0;
385 icsk->icsk_ack.pending &= ~(ICSK_ACK_SCHED | ICSK_ACK_TIMER);
386}
387
388static bool mptcp_established_options_mp(struct sock *sk, struct sk_buff *skb,
389 unsigned int *size,
390 unsigned int remaining,
391 struct mptcp_out_options *opts)
392{
393 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
394 struct mptcp_ext *mpext;
395 unsigned int data_len;
396
397 /* When skb is not available, we better over-estimate the emitted
398 * options len. A full DSS option (28 bytes) is longer than
399 * TCPOLEN_MPTCP_MPC_ACK_DATA(22) or TCPOLEN_MPTCP_MPJ_ACK(24), so
400 * tell the caller to defer the estimate to
401 * mptcp_established_options_dss(), which will reserve enough space.
402 */
403 if (!skb)
404 return false;
405
406 /* MPC/MPJ needed only on 3rd ack packet */
407 if (subflow->fully_established ||
408 subflow->snd_isn != TCP_SKB_CB(skb)->seq)
409 return false;
410
411 if (subflow->mp_capable) {
412 mpext = mptcp_get_ext(skb);
413 data_len = mpext ? mpext->data_len : 0;
414
415 /* we will check ext_copy.data_len in mptcp_write_options() to
416 * discriminate between TCPOLEN_MPTCP_MPC_ACK_DATA and
417 * TCPOLEN_MPTCP_MPC_ACK
418 */
419 opts->ext_copy.data_len = data_len;
420 opts->suboptions = OPTION_MPTCP_MPC_ACK;
421 opts->sndr_key = subflow->local_key;
422 opts->rcvr_key = subflow->remote_key;
423
424 /* Section 3.1.
425 * The MP_CAPABLE option is carried on the SYN, SYN/ACK, and ACK
426 * packets that start the first subflow of an MPTCP connection,
427 * as well as the first packet that carries data
428 */
429 if (data_len > 0)
430 *size = ALIGN(TCPOLEN_MPTCP_MPC_ACK_DATA, 4);
431 else
432 *size = TCPOLEN_MPTCP_MPC_ACK;
433
434 pr_debug("subflow=%p, local_key=%llu, remote_key=%llu map_len=%d",
435 subflow, subflow->local_key, subflow->remote_key,
436 data_len);
437
438 return true;
439 } else if (subflow->mp_join) {
440 opts->suboptions = OPTION_MPTCP_MPJ_ACK;
441 memcpy(opts->hmac, subflow->hmac, MPTCPOPT_HMAC_LEN);
442 *size = TCPOLEN_MPTCP_MPJ_ACK;
443 pr_debug("subflow=%p", subflow);
444
445 schedule_3rdack_retransmission(sk);
446 return true;
447 }
448 return false;
449}
450
451static void mptcp_write_data_fin(struct mptcp_subflow_context *subflow,
452 struct sk_buff *skb, struct mptcp_ext *ext)
453{
454 /* The write_seq value has already been incremented, so the actual
455 * sequence number for the DATA_FIN is one less.
456 */
457 u64 data_fin_tx_seq = READ_ONCE(mptcp_sk(subflow->conn)->write_seq) - 1;
458
459 if (!ext->use_map || !skb->len) {
460 /* RFC6824 requires a DSS mapping with specific values
461 * if DATA_FIN is set but no data payload is mapped
462 */
463 ext->data_fin = 1;
464 ext->use_map = 1;
465 ext->dsn64 = 1;
466 ext->data_seq = data_fin_tx_seq;
467 ext->subflow_seq = 0;
468 ext->data_len = 1;
469 } else if (ext->data_seq + ext->data_len == data_fin_tx_seq) {
470 /* If there's an existing DSS mapping and it is the
471 * final mapping, DATA_FIN consumes 1 additional byte of
472 * mapping space.
473 */
474 ext->data_fin = 1;
475 ext->data_len++;
476 }
477}
478
479static bool mptcp_established_options_dss(struct sock *sk, struct sk_buff *skb,
480 unsigned int *size,
481 unsigned int remaining,
482 struct mptcp_out_options *opts)
483{
484 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
485 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
486 unsigned int dss_size = 0;
487 u64 snd_data_fin_enable;
488 struct mptcp_ext *mpext;
489 unsigned int ack_size;
490 bool ret = false;
491
492 mpext = skb ? mptcp_get_ext(skb) : NULL;
493 snd_data_fin_enable = READ_ONCE(msk->snd_data_fin_enable);
494
495 if (!skb || (mpext && mpext->use_map) || snd_data_fin_enable) {
496 unsigned int map_size;
497
498 map_size = TCPOLEN_MPTCP_DSS_BASE + TCPOLEN_MPTCP_DSS_MAP64;
499
500 remaining -= map_size;
501 dss_size = map_size;
502 if (mpext)
503 opts->ext_copy = *mpext;
504
505 if (skb && snd_data_fin_enable)
506 mptcp_write_data_fin(subflow, skb, &opts->ext_copy);
507 ret = true;
508 }
509
510 /* passive sockets msk will set the 'can_ack' after accept(), even
511 * if the first subflow may have the already the remote key handy
512 */
513 opts->ext_copy.use_ack = 0;
514 if (!READ_ONCE(msk->can_ack)) {
515 *size = ALIGN(dss_size, 4);
516 return ret;
517 }
518
519 if (subflow->use_64bit_ack) {
520 ack_size = TCPOLEN_MPTCP_DSS_ACK64;
521 opts->ext_copy.data_ack = READ_ONCE(msk->ack_seq);
522 opts->ext_copy.ack64 = 1;
523 } else {
524 ack_size = TCPOLEN_MPTCP_DSS_ACK32;
525 opts->ext_copy.data_ack32 = (uint32_t)READ_ONCE(msk->ack_seq);
526 opts->ext_copy.ack64 = 0;
527 }
528 opts->ext_copy.use_ack = 1;
529
530 /* Add kind/length/subtype/flag overhead if mapping is not populated */
531 if (dss_size == 0)
532 ack_size += TCPOLEN_MPTCP_DSS_BASE;
533
534 dss_size += ack_size;
535
536 *size = ALIGN(dss_size, 4);
537 return true;
538}
539
540static u64 add_addr_generate_hmac(u64 key1, u64 key2, u8 addr_id,
541 struct in_addr *addr)
542{
543 u8 hmac[SHA256_DIGEST_SIZE];
544 u8 msg[7];
545
546 msg[0] = addr_id;
547 memcpy(&msg[1], &addr->s_addr, 4);
548 msg[5] = 0;
549 msg[6] = 0;
550
551 mptcp_crypto_hmac_sha(key1, key2, msg, 7, hmac);
552
553 return get_unaligned_be64(&hmac[SHA256_DIGEST_SIZE - sizeof(u64)]);
554}
555
556#if IS_ENABLED(CONFIG_MPTCP_IPV6)
557static u64 add_addr6_generate_hmac(u64 key1, u64 key2, u8 addr_id,
558 struct in6_addr *addr)
559{
560 u8 hmac[SHA256_DIGEST_SIZE];
561 u8 msg[19];
562
563 msg[0] = addr_id;
564 memcpy(&msg[1], &addr->s6_addr, 16);
565 msg[17] = 0;
566 msg[18] = 0;
567
568 mptcp_crypto_hmac_sha(key1, key2, msg, 19, hmac);
569
570 return get_unaligned_be64(&hmac[SHA256_DIGEST_SIZE - sizeof(u64)]);
571}
572#endif
573
574static bool mptcp_established_options_addr(struct sock *sk,
575 unsigned int *size,
576 unsigned int remaining,
577 struct mptcp_out_options *opts)
578{
579 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
580 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
581 struct mptcp_addr_info saddr;
582 int len;
583
584 if (!mptcp_pm_should_signal(msk) ||
585 !(mptcp_pm_addr_signal(msk, remaining, &saddr)))
586 return false;
587
588 len = mptcp_add_addr_len(saddr.family);
589 if (remaining < len)
590 return false;
591
592 *size = len;
593 opts->addr_id = saddr.id;
594 if (saddr.family == AF_INET) {
595 opts->suboptions |= OPTION_MPTCP_ADD_ADDR;
596 opts->addr = saddr.addr;
597 opts->ahmac = add_addr_generate_hmac(msk->local_key,
598 msk->remote_key,
599 opts->addr_id,
600 &opts->addr);
601 }
602#if IS_ENABLED(CONFIG_MPTCP_IPV6)
603 else if (saddr.family == AF_INET6) {
604 opts->suboptions |= OPTION_MPTCP_ADD_ADDR6;
605 opts->addr6 = saddr.addr6;
606 opts->ahmac = add_addr6_generate_hmac(msk->local_key,
607 msk->remote_key,
608 opts->addr_id,
609 &opts->addr6);
610 }
611#endif
612 pr_debug("addr_id=%d, ahmac=%llu", opts->addr_id, opts->ahmac);
613
614 return true;
615}
616
617bool mptcp_established_options(struct sock *sk, struct sk_buff *skb,
618 unsigned int *size, unsigned int remaining,
619 struct mptcp_out_options *opts)
620{
621 unsigned int opt_size = 0;
622 bool ret = false;
623
624 opts->suboptions = 0;
625
626 if (unlikely(mptcp_check_fallback(sk)))
627 return false;
628
629 if (mptcp_established_options_mp(sk, skb, &opt_size, remaining, opts))
630 ret = true;
631 else if (mptcp_established_options_dss(sk, skb, &opt_size, remaining,
632 opts))
633 ret = true;
634
635 /* we reserved enough space for the above options, and exceeding the
636 * TCP option space would be fatal
637 */
638 if (WARN_ON_ONCE(opt_size > remaining))
639 return false;
640
641 *size += opt_size;
642 remaining -= opt_size;
643 if (mptcp_established_options_addr(sk, &opt_size, remaining, opts)) {
644 *size += opt_size;
645 remaining -= opt_size;
646 ret = true;
647 }
648
649 return ret;
650}
651
652bool mptcp_synack_options(const struct request_sock *req, unsigned int *size,
653 struct mptcp_out_options *opts)
654{
655 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
656
657 if (subflow_req->mp_capable) {
658 opts->suboptions = OPTION_MPTCP_MPC_SYNACK;
659 opts->sndr_key = subflow_req->local_key;
660 *size = TCPOLEN_MPTCP_MPC_SYNACK;
661 pr_debug("subflow_req=%p, local_key=%llu",
662 subflow_req, subflow_req->local_key);
663 return true;
664 } else if (subflow_req->mp_join) {
665 opts->suboptions = OPTION_MPTCP_MPJ_SYNACK;
666 opts->backup = subflow_req->backup;
667 opts->join_id = subflow_req->local_id;
668 opts->thmac = subflow_req->thmac;
669 opts->nonce = subflow_req->local_nonce;
670 pr_debug("req=%p, bkup=%u, id=%u, thmac=%llu, nonce=%u",
671 subflow_req, opts->backup, opts->join_id,
672 opts->thmac, opts->nonce);
673 *size = TCPOLEN_MPTCP_MPJ_SYNACK;
674 return true;
675 }
676 return false;
677}
678
679static bool check_fully_established(struct mptcp_sock *msk, struct sock *sk,
680 struct mptcp_subflow_context *subflow,
681 struct sk_buff *skb,
682 struct mptcp_options_received *mp_opt)
683{
684 /* here we can process OoO, in-window pkts, only in-sequence 4th ack
685 * will make the subflow fully established
686 */
687 if (likely(subflow->fully_established)) {
688 /* on passive sockets, check for 3rd ack retransmission
689 * note that msk is always set by subflow_syn_recv_sock()
690 * for mp_join subflows
691 */
692 if (TCP_SKB_CB(skb)->seq == subflow->ssn_offset + 1 &&
693 TCP_SKB_CB(skb)->end_seq == TCP_SKB_CB(skb)->seq &&
694 subflow->mp_join && mp_opt->mp_join &&
695 READ_ONCE(msk->pm.server_side))
696 tcp_send_ack(sk);
697 goto fully_established;
698 }
699
700 /* we should process OoO packets before the first subflow is fully
701 * established, but not expected for MP_JOIN subflows
702 */
703 if (TCP_SKB_CB(skb)->seq != subflow->ssn_offset + 1)
704 return subflow->mp_capable;
705
706 if (mp_opt->dss && mp_opt->use_ack) {
707 /* subflows are fully established as soon as we get any
708 * additional ack.
709 */
710 subflow->fully_established = 1;
711 WRITE_ONCE(msk->fully_established, true);
712 goto fully_established;
713 }
714
715 /* If the first established packet does not contain MP_CAPABLE + data
716 * then fallback to TCP
717 */
718 if (!mp_opt->mp_capable) {
719 subflow->mp_capable = 0;
720 pr_fallback(msk);
721 __mptcp_do_fallback(msk);
722 return false;
723 }
724
725 if (unlikely(!READ_ONCE(msk->pm.server_side)))
726 pr_warn_once("bogus mpc option on established client sk");
727 mptcp_subflow_fully_established(subflow, mp_opt);
728
729fully_established:
730 if (likely(subflow->pm_notified))
731 return true;
732
733 subflow->pm_notified = 1;
734 if (subflow->mp_join) {
735 clear_3rdack_retransmission(sk);
736 mptcp_pm_subflow_established(msk, subflow);
737 } else {
738 mptcp_pm_fully_established(msk);
739 }
740 return true;
741}
742
743static u64 expand_ack(u64 old_ack, u64 cur_ack, bool use_64bit)
744{
745 u32 old_ack32, cur_ack32;
746
747 if (use_64bit)
748 return cur_ack;
749
750 old_ack32 = (u32)old_ack;
751 cur_ack32 = (u32)cur_ack;
752 cur_ack = (old_ack & GENMASK_ULL(63, 32)) + cur_ack32;
753 if (unlikely(before(cur_ack32, old_ack32)))
754 return cur_ack + (1LL << 32);
755 return cur_ack;
756}
757
758static void update_una(struct mptcp_sock *msk,
759 struct mptcp_options_received *mp_opt)
760{
761 u64 new_snd_una, snd_una, old_snd_una = atomic64_read(&msk->snd_una);
762 u64 write_seq = READ_ONCE(msk->write_seq);
763
764 /* avoid ack expansion on update conflict, to reduce the risk of
765 * wrongly expanding to a future ack sequence number, which is way
766 * more dangerous than missing an ack
767 */
768 new_snd_una = expand_ack(old_snd_una, mp_opt->data_ack, mp_opt->ack64);
769
770 /* ACK for data not even sent yet? Ignore. */
771 if (after64(new_snd_una, write_seq))
772 new_snd_una = old_snd_una;
773
774 while (after64(new_snd_una, old_snd_una)) {
775 snd_una = old_snd_una;
776 old_snd_una = atomic64_cmpxchg(&msk->snd_una, snd_una,
777 new_snd_una);
778 if (old_snd_una == snd_una) {
779 mptcp_data_acked((struct sock *)msk);
780 break;
781 }
782 }
783}
784
785bool mptcp_update_rcv_data_fin(struct mptcp_sock *msk, u64 data_fin_seq, bool use_64bit)
786{
787 /* Skip if DATA_FIN was already received.
788 * If updating simultaneously with the recvmsg loop, values
789 * should match. If they mismatch, the peer is misbehaving and
790 * we will prefer the most recent information.
791 */
792 if (READ_ONCE(msk->rcv_data_fin) || !READ_ONCE(msk->first))
793 return false;
794
795 WRITE_ONCE(msk->rcv_data_fin_seq,
796 expand_ack(READ_ONCE(msk->ack_seq), data_fin_seq, use_64bit));
797 WRITE_ONCE(msk->rcv_data_fin, 1);
798
799 return true;
800}
801
802static bool add_addr_hmac_valid(struct mptcp_sock *msk,
803 struct mptcp_options_received *mp_opt)
804{
805 u64 hmac = 0;
806
807 if (mp_opt->echo)
808 return true;
809
810 if (mp_opt->family == MPTCP_ADDR_IPVERSION_4)
811 hmac = add_addr_generate_hmac(msk->remote_key,
812 msk->local_key,
813 mp_opt->addr_id, &mp_opt->addr);
814#if IS_ENABLED(CONFIG_MPTCP_IPV6)
815 else
816 hmac = add_addr6_generate_hmac(msk->remote_key,
817 msk->local_key,
818 mp_opt->addr_id, &mp_opt->addr6);
819#endif
820
821 pr_debug("msk=%p, ahmac=%llu, mp_opt->ahmac=%llu\n",
822 msk, (unsigned long long)hmac,
823 (unsigned long long)mp_opt->ahmac);
824
825 return hmac == mp_opt->ahmac;
826}
827
828void mptcp_incoming_options(struct sock *sk, struct sk_buff *skb,
829 struct tcp_options_received *opt_rx)
830{
831 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
832 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
833 struct mptcp_options_received mp_opt;
834 struct mptcp_ext *mpext;
835
836 if (__mptcp_check_fallback(msk))
837 return;
838
839 mptcp_get_options(skb, &mp_opt);
840 if (!check_fully_established(msk, sk, subflow, skb, &mp_opt))
841 return;
842
843 if (mp_opt.add_addr && add_addr_hmac_valid(msk, &mp_opt)) {
844 struct mptcp_addr_info addr;
845
846 addr.port = htons(mp_opt.port);
847 addr.id = mp_opt.addr_id;
848 if (mp_opt.family == MPTCP_ADDR_IPVERSION_4) {
849 addr.family = AF_INET;
850 addr.addr = mp_opt.addr;
851 }
852#if IS_ENABLED(CONFIG_MPTCP_IPV6)
853 else if (mp_opt.family == MPTCP_ADDR_IPVERSION_6) {
854 addr.family = AF_INET6;
855 addr.addr6 = mp_opt.addr6;
856 }
857#endif
858 if (!mp_opt.echo)
859 mptcp_pm_add_addr_received(msk, &addr);
860 mp_opt.add_addr = 0;
861 }
862
863 if (!mp_opt.dss)
864 return;
865
866 /* we can't wait for recvmsg() to update the ack_seq, otherwise
867 * monodirectional flows will stuck
868 */
869 if (mp_opt.use_ack)
870 update_una(msk, &mp_opt);
871
872 /* Zero-data-length packets are dropped by the caller and not
873 * propagated to the MPTCP layer, so the skb extension does not
874 * need to be allocated or populated. DATA_FIN information, if
875 * present, needs to be updated here before the skb is freed.
876 */
877 if (TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb)->end_seq) {
878 if (mp_opt.data_fin && mp_opt.data_len == 1 &&
879 mptcp_update_rcv_data_fin(msk, mp_opt.data_seq, mp_opt.dsn64) &&
880 schedule_work(&msk->work))
881 sock_hold(subflow->conn);
882
883 return;
884 }
885
886 mpext = skb_ext_add(skb, SKB_EXT_MPTCP);
887 if (!mpext)
888 return;
889
890 memset(mpext, 0, sizeof(*mpext));
891
892 if (mp_opt.use_map) {
893 if (mp_opt.mpc_map) {
894 /* this is an MP_CAPABLE carrying MPTCP data
895 * we know this map the first chunk of data
896 */
897 mptcp_crypto_key_sha(subflow->remote_key, NULL,
898 &mpext->data_seq);
899 mpext->data_seq++;
900 mpext->subflow_seq = 1;
901 mpext->dsn64 = 1;
902 mpext->mpc_map = 1;
903 mpext->data_fin = 0;
904 } else {
905 mpext->data_seq = mp_opt.data_seq;
906 mpext->subflow_seq = mp_opt.subflow_seq;
907 mpext->dsn64 = mp_opt.dsn64;
908 mpext->data_fin = mp_opt.data_fin;
909 }
910 mpext->data_len = mp_opt.data_len;
911 mpext->use_map = 1;
912 }
913}
914
915void mptcp_write_options(__be32 *ptr, struct mptcp_out_options *opts)
916{
917 if ((OPTION_MPTCP_MPC_SYN | OPTION_MPTCP_MPC_SYNACK |
918 OPTION_MPTCP_MPC_ACK) & opts->suboptions) {
919 u8 len;
920
921 if (OPTION_MPTCP_MPC_SYN & opts->suboptions)
922 len = TCPOLEN_MPTCP_MPC_SYN;
923 else if (OPTION_MPTCP_MPC_SYNACK & opts->suboptions)
924 len = TCPOLEN_MPTCP_MPC_SYNACK;
925 else if (opts->ext_copy.data_len)
926 len = TCPOLEN_MPTCP_MPC_ACK_DATA;
927 else
928 len = TCPOLEN_MPTCP_MPC_ACK;
929
930 *ptr++ = mptcp_option(MPTCPOPT_MP_CAPABLE, len,
931 MPTCP_SUPPORTED_VERSION,
932 MPTCP_CAP_HMAC_SHA256);
933
934 if (!((OPTION_MPTCP_MPC_SYNACK | OPTION_MPTCP_MPC_ACK) &
935 opts->suboptions))
936 goto mp_capable_done;
937
938 put_unaligned_be64(opts->sndr_key, ptr);
939 ptr += 2;
940 if (!((OPTION_MPTCP_MPC_ACK) & opts->suboptions))
941 goto mp_capable_done;
942
943 put_unaligned_be64(opts->rcvr_key, ptr);
944 ptr += 2;
945 if (!opts->ext_copy.data_len)
946 goto mp_capable_done;
947
948 put_unaligned_be32(opts->ext_copy.data_len << 16 |
949 TCPOPT_NOP << 8 | TCPOPT_NOP, ptr);
950 ptr += 1;
951 }
952
953mp_capable_done:
954 if (OPTION_MPTCP_ADD_ADDR & opts->suboptions) {
955 if (opts->ahmac)
956 *ptr++ = mptcp_option(MPTCPOPT_ADD_ADDR,
957 TCPOLEN_MPTCP_ADD_ADDR, 0,
958 opts->addr_id);
959 else
960 *ptr++ = mptcp_option(MPTCPOPT_ADD_ADDR,
961 TCPOLEN_MPTCP_ADD_ADDR_BASE,
962 MPTCP_ADDR_ECHO,
963 opts->addr_id);
964 memcpy((u8 *)ptr, (u8 *)&opts->addr.s_addr, 4);
965 ptr += 1;
966 if (opts->ahmac) {
967 put_unaligned_be64(opts->ahmac, ptr);
968 ptr += 2;
969 }
970 }
971
972#if IS_ENABLED(CONFIG_MPTCP_IPV6)
973 if (OPTION_MPTCP_ADD_ADDR6 & opts->suboptions) {
974 if (opts->ahmac)
975 *ptr++ = mptcp_option(MPTCPOPT_ADD_ADDR,
976 TCPOLEN_MPTCP_ADD_ADDR6, 0,
977 opts->addr_id);
978 else
979 *ptr++ = mptcp_option(MPTCPOPT_ADD_ADDR,
980 TCPOLEN_MPTCP_ADD_ADDR6_BASE,
981 MPTCP_ADDR_ECHO,
982 opts->addr_id);
983 memcpy((u8 *)ptr, opts->addr6.s6_addr, 16);
984 ptr += 4;
985 if (opts->ahmac) {
986 put_unaligned_be64(opts->ahmac, ptr);
987 ptr += 2;
988 }
989 }
990#endif
991
992 if (OPTION_MPTCP_RM_ADDR & opts->suboptions) {
993 *ptr++ = mptcp_option(MPTCPOPT_RM_ADDR,
994 TCPOLEN_MPTCP_RM_ADDR_BASE,
995 0, opts->rm_id);
996 }
997
998 if (OPTION_MPTCP_MPJ_SYN & opts->suboptions) {
999 *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1000 TCPOLEN_MPTCP_MPJ_SYN,
1001 opts->backup, opts->join_id);
1002 put_unaligned_be32(opts->token, ptr);
1003 ptr += 1;
1004 put_unaligned_be32(opts->nonce, ptr);
1005 ptr += 1;
1006 }
1007
1008 if (OPTION_MPTCP_MPJ_SYNACK & opts->suboptions) {
1009 *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1010 TCPOLEN_MPTCP_MPJ_SYNACK,
1011 opts->backup, opts->join_id);
1012 put_unaligned_be64(opts->thmac, ptr);
1013 ptr += 2;
1014 put_unaligned_be32(opts->nonce, ptr);
1015 ptr += 1;
1016 }
1017
1018 if (OPTION_MPTCP_MPJ_ACK & opts->suboptions) {
1019 *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1020 TCPOLEN_MPTCP_MPJ_ACK, 0, 0);
1021 memcpy(ptr, opts->hmac, MPTCPOPT_HMAC_LEN);
1022 ptr += 5;
1023 }
1024
1025 if (opts->ext_copy.use_ack || opts->ext_copy.use_map) {
1026 struct mptcp_ext *mpext = &opts->ext_copy;
1027 u8 len = TCPOLEN_MPTCP_DSS_BASE;
1028 u8 flags = 0;
1029
1030 if (mpext->use_ack) {
1031 flags = MPTCP_DSS_HAS_ACK;
1032 if (mpext->ack64) {
1033 len += TCPOLEN_MPTCP_DSS_ACK64;
1034 flags |= MPTCP_DSS_ACK64;
1035 } else {
1036 len += TCPOLEN_MPTCP_DSS_ACK32;
1037 }
1038 }
1039
1040 if (mpext->use_map) {
1041 len += TCPOLEN_MPTCP_DSS_MAP64;
1042
1043 /* Use only 64-bit mapping flags for now, add
1044 * support for optional 32-bit mappings later.
1045 */
1046 flags |= MPTCP_DSS_HAS_MAP | MPTCP_DSS_DSN64;
1047 if (mpext->data_fin)
1048 flags |= MPTCP_DSS_DATA_FIN;
1049 }
1050
1051 *ptr++ = mptcp_option(MPTCPOPT_DSS, len, 0, flags);
1052
1053 if (mpext->use_ack) {
1054 if (mpext->ack64) {
1055 put_unaligned_be64(mpext->data_ack, ptr);
1056 ptr += 2;
1057 } else {
1058 put_unaligned_be32(mpext->data_ack32, ptr);
1059 ptr += 1;
1060 }
1061 }
1062
1063 if (mpext->use_map) {
1064 put_unaligned_be64(mpext->data_seq, ptr);
1065 ptr += 2;
1066 put_unaligned_be32(mpext->subflow_seq, ptr);
1067 ptr += 1;
1068 put_unaligned_be32(mpext->data_len << 16 |
1069 TCPOPT_NOP << 8 | TCPOPT_NOP, ptr);
1070 }
1071 }
1072}
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
18static bool mptcp_cap_flag_sha256(u8 flags)
19{
20 return (flags & MPTCP_CAP_FLAG_MASK) == MPTCP_CAP_HMAC_SHA256;
21}
22
23static void mptcp_parse_option(const struct sk_buff *skb,
24 const unsigned char *ptr, int opsize,
25 struct mptcp_options_received *mp_opt)
26{
27 u8 subtype = *ptr >> 4;
28 int expected_opsize;
29 u16 subopt;
30 u8 version;
31 u8 flags;
32 u8 i;
33
34 switch (subtype) {
35 case MPTCPOPT_MP_CAPABLE:
36 /* strict size checking */
37 if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)) {
38 if (skb->len > tcp_hdr(skb)->doff << 2)
39 expected_opsize = TCPOLEN_MPTCP_MPC_ACK_DATA;
40 else
41 expected_opsize = TCPOLEN_MPTCP_MPC_ACK;
42 subopt = OPTION_MPTCP_MPC_ACK;
43 } else {
44 if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_ACK) {
45 expected_opsize = TCPOLEN_MPTCP_MPC_SYNACK;
46 subopt = OPTION_MPTCP_MPC_SYNACK;
47 } else {
48 expected_opsize = TCPOLEN_MPTCP_MPC_SYN;
49 subopt = OPTION_MPTCP_MPC_SYN;
50 }
51 }
52
53 /* Cfr RFC 8684 Section 3.3.0:
54 * If a checksum is present but its use had
55 * not been negotiated in the MP_CAPABLE handshake, the receiver MUST
56 * close the subflow with a RST, as it is not behaving as negotiated.
57 * If a checksum is not present when its use has been negotiated, the
58 * receiver MUST close the subflow with a RST, as it is considered
59 * broken
60 * We parse even option with mismatching csum presence, so that
61 * later in subflow_data_ready we can trigger the reset.
62 */
63 if (opsize != expected_opsize &&
64 (expected_opsize != TCPOLEN_MPTCP_MPC_ACK_DATA ||
65 opsize != TCPOLEN_MPTCP_MPC_ACK_DATA_CSUM))
66 break;
67
68 /* try to be gentle vs future versions on the initial syn */
69 version = *ptr++ & MPTCP_VERSION_MASK;
70 if (opsize != TCPOLEN_MPTCP_MPC_SYN) {
71 if (version != MPTCP_SUPPORTED_VERSION)
72 break;
73 } else if (version < MPTCP_SUPPORTED_VERSION) {
74 break;
75 }
76
77 flags = *ptr++;
78 if (!mptcp_cap_flag_sha256(flags) ||
79 (flags & MPTCP_CAP_EXTENSIBILITY))
80 break;
81
82 /* RFC 6824, Section 3.1:
83 * "For the Checksum Required bit (labeled "A"), if either
84 * host requires the use of checksums, checksums MUST be used.
85 * In other words, the only way for checksums not to be used
86 * is if both hosts in their SYNs set A=0."
87 */
88 if (flags & MPTCP_CAP_CHECKSUM_REQD)
89 mp_opt->suboptions |= OPTION_MPTCP_CSUMREQD;
90
91 mp_opt->deny_join_id0 = !!(flags & MPTCP_CAP_DENY_JOIN_ID0);
92
93 mp_opt->suboptions |= subopt;
94 if (opsize >= TCPOLEN_MPTCP_MPC_SYNACK) {
95 mp_opt->sndr_key = get_unaligned_be64(ptr);
96 ptr += 8;
97 }
98 if (opsize >= TCPOLEN_MPTCP_MPC_ACK) {
99 mp_opt->rcvr_key = get_unaligned_be64(ptr);
100 ptr += 8;
101 }
102 if (opsize >= TCPOLEN_MPTCP_MPC_ACK_DATA) {
103 /* Section 3.1.:
104 * "the data parameters in a MP_CAPABLE are semantically
105 * equivalent to those in a DSS option and can be used
106 * interchangeably."
107 */
108 mp_opt->suboptions |= OPTION_MPTCP_DSS;
109 mp_opt->use_map = 1;
110 mp_opt->mpc_map = 1;
111 mp_opt->use_ack = 0;
112 mp_opt->data_len = get_unaligned_be16(ptr);
113 ptr += 2;
114 }
115 if (opsize == TCPOLEN_MPTCP_MPC_ACK_DATA_CSUM) {
116 mp_opt->csum = get_unaligned((__force __sum16 *)ptr);
117 mp_opt->suboptions |= OPTION_MPTCP_CSUMREQD;
118 ptr += 2;
119 }
120 pr_debug("MP_CAPABLE version=%x, flags=%x, optlen=%d sndr=%llu, rcvr=%llu len=%d csum=%u",
121 version, flags, opsize, mp_opt->sndr_key,
122 mp_opt->rcvr_key, mp_opt->data_len, mp_opt->csum);
123 break;
124
125 case MPTCPOPT_MP_JOIN:
126 if (opsize == TCPOLEN_MPTCP_MPJ_SYN) {
127 mp_opt->suboptions |= OPTION_MPTCP_MPJ_SYN;
128 mp_opt->backup = *ptr++ & MPTCPOPT_BACKUP;
129 mp_opt->join_id = *ptr++;
130 mp_opt->token = get_unaligned_be32(ptr);
131 ptr += 4;
132 mp_opt->nonce = get_unaligned_be32(ptr);
133 ptr += 4;
134 pr_debug("MP_JOIN bkup=%u, id=%u, token=%u, nonce=%u",
135 mp_opt->backup, mp_opt->join_id,
136 mp_opt->token, mp_opt->nonce);
137 } else if (opsize == TCPOLEN_MPTCP_MPJ_SYNACK) {
138 mp_opt->suboptions |= OPTION_MPTCP_MPJ_SYNACK;
139 mp_opt->backup = *ptr++ & MPTCPOPT_BACKUP;
140 mp_opt->join_id = *ptr++;
141 mp_opt->thmac = get_unaligned_be64(ptr);
142 ptr += 8;
143 mp_opt->nonce = get_unaligned_be32(ptr);
144 ptr += 4;
145 pr_debug("MP_JOIN bkup=%u, id=%u, thmac=%llu, nonce=%u",
146 mp_opt->backup, mp_opt->join_id,
147 mp_opt->thmac, mp_opt->nonce);
148 } else if (opsize == TCPOLEN_MPTCP_MPJ_ACK) {
149 mp_opt->suboptions |= OPTION_MPTCP_MPJ_ACK;
150 ptr += 2;
151 memcpy(mp_opt->hmac, ptr, MPTCPOPT_HMAC_LEN);
152 pr_debug("MP_JOIN hmac");
153 }
154 break;
155
156 case MPTCPOPT_DSS:
157 pr_debug("DSS");
158 ptr++;
159
160 /* we must clear 'mpc_map' be able to detect MP_CAPABLE
161 * map vs DSS map in mptcp_incoming_options(), and reconstruct
162 * map info accordingly
163 */
164 mp_opt->mpc_map = 0;
165 flags = (*ptr++) & MPTCP_DSS_FLAG_MASK;
166 mp_opt->data_fin = (flags & MPTCP_DSS_DATA_FIN) != 0;
167 mp_opt->dsn64 = (flags & MPTCP_DSS_DSN64) != 0;
168 mp_opt->use_map = (flags & MPTCP_DSS_HAS_MAP) != 0;
169 mp_opt->ack64 = (flags & MPTCP_DSS_ACK64) != 0;
170 mp_opt->use_ack = (flags & MPTCP_DSS_HAS_ACK);
171
172 pr_debug("data_fin=%d dsn64=%d use_map=%d ack64=%d use_ack=%d",
173 mp_opt->data_fin, mp_opt->dsn64,
174 mp_opt->use_map, mp_opt->ack64,
175 mp_opt->use_ack);
176
177 expected_opsize = TCPOLEN_MPTCP_DSS_BASE;
178
179 if (mp_opt->use_ack) {
180 if (mp_opt->ack64)
181 expected_opsize += TCPOLEN_MPTCP_DSS_ACK64;
182 else
183 expected_opsize += TCPOLEN_MPTCP_DSS_ACK32;
184 }
185
186 if (mp_opt->use_map) {
187 if (mp_opt->dsn64)
188 expected_opsize += TCPOLEN_MPTCP_DSS_MAP64;
189 else
190 expected_opsize += TCPOLEN_MPTCP_DSS_MAP32;
191 }
192
193 /* Always parse any csum presence combination, we will enforce
194 * RFC 8684 Section 3.3.0 checks later in subflow_data_ready
195 */
196 if (opsize != expected_opsize &&
197 opsize != expected_opsize + TCPOLEN_MPTCP_DSS_CHECKSUM)
198 break;
199
200 mp_opt->suboptions |= OPTION_MPTCP_DSS;
201 if (mp_opt->use_ack) {
202 if (mp_opt->ack64) {
203 mp_opt->data_ack = get_unaligned_be64(ptr);
204 ptr += 8;
205 } else {
206 mp_opt->data_ack = get_unaligned_be32(ptr);
207 ptr += 4;
208 }
209
210 pr_debug("data_ack=%llu", mp_opt->data_ack);
211 }
212
213 if (mp_opt->use_map) {
214 if (mp_opt->dsn64) {
215 mp_opt->data_seq = get_unaligned_be64(ptr);
216 ptr += 8;
217 } else {
218 mp_opt->data_seq = get_unaligned_be32(ptr);
219 ptr += 4;
220 }
221
222 mp_opt->subflow_seq = get_unaligned_be32(ptr);
223 ptr += 4;
224
225 mp_opt->data_len = get_unaligned_be16(ptr);
226 ptr += 2;
227
228 if (opsize == expected_opsize + TCPOLEN_MPTCP_DSS_CHECKSUM) {
229 mp_opt->suboptions |= OPTION_MPTCP_CSUMREQD;
230 mp_opt->csum = get_unaligned((__force __sum16 *)ptr);
231 ptr += 2;
232 }
233
234 pr_debug("data_seq=%llu subflow_seq=%u data_len=%u csum=%d:%u",
235 mp_opt->data_seq, mp_opt->subflow_seq,
236 mp_opt->data_len, !!(mp_opt->suboptions & OPTION_MPTCP_CSUMREQD),
237 mp_opt->csum);
238 }
239
240 break;
241
242 case MPTCPOPT_ADD_ADDR:
243 mp_opt->echo = (*ptr++) & MPTCP_ADDR_ECHO;
244 if (!mp_opt->echo) {
245 if (opsize == TCPOLEN_MPTCP_ADD_ADDR ||
246 opsize == TCPOLEN_MPTCP_ADD_ADDR_PORT)
247 mp_opt->addr.family = AF_INET;
248#if IS_ENABLED(CONFIG_MPTCP_IPV6)
249 else if (opsize == TCPOLEN_MPTCP_ADD_ADDR6 ||
250 opsize == TCPOLEN_MPTCP_ADD_ADDR6_PORT)
251 mp_opt->addr.family = AF_INET6;
252#endif
253 else
254 break;
255 } else {
256 if (opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE ||
257 opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE_PORT)
258 mp_opt->addr.family = AF_INET;
259#if IS_ENABLED(CONFIG_MPTCP_IPV6)
260 else if (opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE ||
261 opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE_PORT)
262 mp_opt->addr.family = AF_INET6;
263#endif
264 else
265 break;
266 }
267
268 mp_opt->suboptions |= OPTION_MPTCP_ADD_ADDR;
269 mp_opt->addr.id = *ptr++;
270 mp_opt->addr.port = 0;
271 mp_opt->ahmac = 0;
272 if (mp_opt->addr.family == AF_INET) {
273 memcpy((u8 *)&mp_opt->addr.addr.s_addr, (u8 *)ptr, 4);
274 ptr += 4;
275 if (opsize == TCPOLEN_MPTCP_ADD_ADDR_PORT ||
276 opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE_PORT) {
277 mp_opt->addr.port = htons(get_unaligned_be16(ptr));
278 ptr += 2;
279 }
280 }
281#if IS_ENABLED(CONFIG_MPTCP_IPV6)
282 else {
283 memcpy(mp_opt->addr.addr6.s6_addr, (u8 *)ptr, 16);
284 ptr += 16;
285 if (opsize == TCPOLEN_MPTCP_ADD_ADDR6_PORT ||
286 opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE_PORT) {
287 mp_opt->addr.port = htons(get_unaligned_be16(ptr));
288 ptr += 2;
289 }
290 }
291#endif
292 if (!mp_opt->echo) {
293 mp_opt->ahmac = get_unaligned_be64(ptr);
294 ptr += 8;
295 }
296 pr_debug("ADD_ADDR%s: id=%d, ahmac=%llu, echo=%d, port=%d",
297 (mp_opt->addr.family == AF_INET6) ? "6" : "",
298 mp_opt->addr.id, mp_opt->ahmac, mp_opt->echo, ntohs(mp_opt->addr.port));
299 break;
300
301 case MPTCPOPT_RM_ADDR:
302 if (opsize < TCPOLEN_MPTCP_RM_ADDR_BASE + 1 ||
303 opsize > TCPOLEN_MPTCP_RM_ADDR_BASE + MPTCP_RM_IDS_MAX)
304 break;
305
306 ptr++;
307
308 mp_opt->suboptions |= OPTION_MPTCP_RM_ADDR;
309 mp_opt->rm_list.nr = opsize - TCPOLEN_MPTCP_RM_ADDR_BASE;
310 for (i = 0; i < mp_opt->rm_list.nr; i++)
311 mp_opt->rm_list.ids[i] = *ptr++;
312 pr_debug("RM_ADDR: rm_list_nr=%d", mp_opt->rm_list.nr);
313 break;
314
315 case MPTCPOPT_MP_PRIO:
316 if (opsize != TCPOLEN_MPTCP_PRIO)
317 break;
318
319 mp_opt->suboptions |= OPTION_MPTCP_PRIO;
320 mp_opt->backup = *ptr++ & MPTCP_PRIO_BKUP;
321 pr_debug("MP_PRIO: prio=%d", mp_opt->backup);
322 break;
323
324 case MPTCPOPT_MP_FASTCLOSE:
325 if (opsize != TCPOLEN_MPTCP_FASTCLOSE)
326 break;
327
328 ptr += 2;
329 mp_opt->rcvr_key = get_unaligned_be64(ptr);
330 ptr += 8;
331 mp_opt->suboptions |= OPTION_MPTCP_FASTCLOSE;
332 pr_debug("MP_FASTCLOSE: recv_key=%llu", mp_opt->rcvr_key);
333 break;
334
335 case MPTCPOPT_RST:
336 if (opsize != TCPOLEN_MPTCP_RST)
337 break;
338
339 if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_RST))
340 break;
341
342 mp_opt->suboptions |= OPTION_MPTCP_RST;
343 flags = *ptr++;
344 mp_opt->reset_transient = flags & MPTCP_RST_TRANSIENT;
345 mp_opt->reset_reason = *ptr;
346 pr_debug("MP_RST: transient=%u reason=%u",
347 mp_opt->reset_transient, mp_opt->reset_reason);
348 break;
349
350 case MPTCPOPT_MP_FAIL:
351 if (opsize != TCPOLEN_MPTCP_FAIL)
352 break;
353
354 ptr += 2;
355 mp_opt->suboptions |= OPTION_MPTCP_FAIL;
356 mp_opt->fail_seq = get_unaligned_be64(ptr);
357 pr_debug("MP_FAIL: data_seq=%llu", mp_opt->fail_seq);
358 break;
359
360 default:
361 break;
362 }
363}
364
365void mptcp_get_options(const struct sk_buff *skb,
366 struct mptcp_options_received *mp_opt)
367{
368 const struct tcphdr *th = tcp_hdr(skb);
369 const unsigned char *ptr;
370 int length;
371
372 /* initialize option status */
373 mp_opt->suboptions = 0;
374
375 length = (th->doff * 4) - sizeof(struct tcphdr);
376 ptr = (const unsigned char *)(th + 1);
377
378 while (length > 0) {
379 int opcode = *ptr++;
380 int opsize;
381
382 switch (opcode) {
383 case TCPOPT_EOL:
384 return;
385 case TCPOPT_NOP: /* Ref: RFC 793 section 3.1 */
386 length--;
387 continue;
388 default:
389 if (length < 2)
390 return;
391 opsize = *ptr++;
392 if (opsize < 2) /* "silly options" */
393 return;
394 if (opsize > length)
395 return; /* don't parse partial options */
396 if (opcode == TCPOPT_MPTCP)
397 mptcp_parse_option(skb, ptr, opsize, mp_opt);
398 ptr += opsize - 2;
399 length -= opsize;
400 }
401 }
402}
403
404bool mptcp_syn_options(struct sock *sk, const struct sk_buff *skb,
405 unsigned int *size, struct mptcp_out_options *opts)
406{
407 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
408
409 /* we will use snd_isn to detect first pkt [re]transmission
410 * in mptcp_established_options_mp()
411 */
412 subflow->snd_isn = TCP_SKB_CB(skb)->end_seq;
413 if (subflow->request_mptcp) {
414 opts->suboptions = OPTION_MPTCP_MPC_SYN;
415 opts->csum_reqd = mptcp_is_checksum_enabled(sock_net(sk));
416 opts->allow_join_id0 = mptcp_allow_join_id0(sock_net(sk));
417 *size = TCPOLEN_MPTCP_MPC_SYN;
418 return true;
419 } else if (subflow->request_join) {
420 pr_debug("remote_token=%u, nonce=%u", subflow->remote_token,
421 subflow->local_nonce);
422 opts->suboptions = OPTION_MPTCP_MPJ_SYN;
423 opts->join_id = subflow->local_id;
424 opts->token = subflow->remote_token;
425 opts->nonce = subflow->local_nonce;
426 opts->backup = subflow->request_bkup;
427 *size = TCPOLEN_MPTCP_MPJ_SYN;
428 return true;
429 }
430 return false;
431}
432
433static void clear_3rdack_retransmission(struct sock *sk)
434{
435 struct inet_connection_sock *icsk = inet_csk(sk);
436
437 sk_stop_timer(sk, &icsk->icsk_delack_timer);
438 icsk->icsk_ack.timeout = 0;
439 icsk->icsk_ack.ato = 0;
440 icsk->icsk_ack.pending &= ~(ICSK_ACK_SCHED | ICSK_ACK_TIMER);
441}
442
443static bool mptcp_established_options_mp(struct sock *sk, struct sk_buff *skb,
444 bool snd_data_fin_enable,
445 unsigned int *size,
446 struct mptcp_out_options *opts)
447{
448 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
449 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
450 struct mptcp_ext *mpext;
451 unsigned int data_len;
452 u8 len;
453
454 /* When skb is not available, we better over-estimate the emitted
455 * options len. A full DSS option (28 bytes) is longer than
456 * TCPOLEN_MPTCP_MPC_ACK_DATA(22) or TCPOLEN_MPTCP_MPJ_ACK(24), so
457 * tell the caller to defer the estimate to
458 * mptcp_established_options_dss(), which will reserve enough space.
459 */
460 if (!skb)
461 return false;
462
463 /* MPC/MPJ needed only on 3rd ack packet, DATA_FIN and TCP shutdown take precedence */
464 if (subflow->fully_established || snd_data_fin_enable ||
465 subflow->snd_isn != TCP_SKB_CB(skb)->seq ||
466 sk->sk_state != TCP_ESTABLISHED)
467 return false;
468
469 if (subflow->mp_capable) {
470 mpext = mptcp_get_ext(skb);
471 data_len = mpext ? mpext->data_len : 0;
472
473 /* we will check ops->data_len in mptcp_write_options() to
474 * discriminate between TCPOLEN_MPTCP_MPC_ACK_DATA and
475 * TCPOLEN_MPTCP_MPC_ACK
476 */
477 opts->data_len = data_len;
478 opts->suboptions = OPTION_MPTCP_MPC_ACK;
479 opts->sndr_key = subflow->local_key;
480 opts->rcvr_key = subflow->remote_key;
481 opts->csum_reqd = READ_ONCE(msk->csum_enabled);
482 opts->allow_join_id0 = mptcp_allow_join_id0(sock_net(sk));
483
484 /* Section 3.1.
485 * The MP_CAPABLE option is carried on the SYN, SYN/ACK, and ACK
486 * packets that start the first subflow of an MPTCP connection,
487 * as well as the first packet that carries data
488 */
489 if (data_len > 0) {
490 len = TCPOLEN_MPTCP_MPC_ACK_DATA;
491 if (opts->csum_reqd) {
492 /* we need to propagate more info to csum the pseudo hdr */
493 opts->data_seq = mpext->data_seq;
494 opts->subflow_seq = mpext->subflow_seq;
495 opts->csum = mpext->csum;
496 len += TCPOLEN_MPTCP_DSS_CHECKSUM;
497 }
498 *size = ALIGN(len, 4);
499 } else {
500 *size = TCPOLEN_MPTCP_MPC_ACK;
501 }
502
503 pr_debug("subflow=%p, local_key=%llu, remote_key=%llu map_len=%d",
504 subflow, subflow->local_key, subflow->remote_key,
505 data_len);
506
507 return true;
508 } else if (subflow->mp_join) {
509 opts->suboptions = OPTION_MPTCP_MPJ_ACK;
510 memcpy(opts->hmac, subflow->hmac, MPTCPOPT_HMAC_LEN);
511 *size = TCPOLEN_MPTCP_MPJ_ACK;
512 pr_debug("subflow=%p", subflow);
513
514 /* we can use the full delegate action helper only from BH context
515 * If we are in process context - sk is flushing the backlog at
516 * socket lock release time - just set the appropriate flag, will
517 * be handled by the release callback
518 */
519 if (sock_owned_by_user(sk))
520 set_bit(MPTCP_DELEGATE_ACK, &subflow->delegated_status);
521 else
522 mptcp_subflow_delegate(subflow, MPTCP_DELEGATE_ACK);
523 return true;
524 }
525 return false;
526}
527
528static void mptcp_write_data_fin(struct mptcp_subflow_context *subflow,
529 struct sk_buff *skb, struct mptcp_ext *ext)
530{
531 /* The write_seq value has already been incremented, so the actual
532 * sequence number for the DATA_FIN is one less.
533 */
534 u64 data_fin_tx_seq = READ_ONCE(mptcp_sk(subflow->conn)->write_seq) - 1;
535
536 if (!ext->use_map || !skb->len) {
537 /* RFC6824 requires a DSS mapping with specific values
538 * if DATA_FIN is set but no data payload is mapped
539 */
540 ext->data_fin = 1;
541 ext->use_map = 1;
542 ext->dsn64 = 1;
543 ext->data_seq = data_fin_tx_seq;
544 ext->subflow_seq = 0;
545 ext->data_len = 1;
546 } else if (ext->data_seq + ext->data_len == data_fin_tx_seq) {
547 /* If there's an existing DSS mapping and it is the
548 * final mapping, DATA_FIN consumes 1 additional byte of
549 * mapping space.
550 */
551 ext->data_fin = 1;
552 ext->data_len++;
553 }
554}
555
556static bool mptcp_established_options_dss(struct sock *sk, struct sk_buff *skb,
557 bool snd_data_fin_enable,
558 unsigned int *size,
559 struct mptcp_out_options *opts)
560{
561 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
562 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
563 unsigned int dss_size = 0;
564 struct mptcp_ext *mpext;
565 unsigned int ack_size;
566 bool ret = false;
567 u64 ack_seq;
568
569 opts->csum_reqd = READ_ONCE(msk->csum_enabled);
570 mpext = skb ? mptcp_get_ext(skb) : NULL;
571
572 if (!skb || (mpext && mpext->use_map) || snd_data_fin_enable) {
573 unsigned int map_size = TCPOLEN_MPTCP_DSS_BASE + TCPOLEN_MPTCP_DSS_MAP64;
574
575 if (mpext) {
576 if (opts->csum_reqd)
577 map_size += TCPOLEN_MPTCP_DSS_CHECKSUM;
578
579 opts->ext_copy = *mpext;
580 }
581
582 dss_size = map_size;
583 if (skb && snd_data_fin_enable)
584 mptcp_write_data_fin(subflow, skb, &opts->ext_copy);
585 opts->suboptions = OPTION_MPTCP_DSS;
586 ret = true;
587 }
588
589 /* passive sockets msk will set the 'can_ack' after accept(), even
590 * if the first subflow may have the already the remote key handy
591 */
592 opts->ext_copy.use_ack = 0;
593 if (!READ_ONCE(msk->can_ack)) {
594 *size = ALIGN(dss_size, 4);
595 return ret;
596 }
597
598 ack_seq = READ_ONCE(msk->ack_seq);
599 if (READ_ONCE(msk->use_64bit_ack)) {
600 ack_size = TCPOLEN_MPTCP_DSS_ACK64;
601 opts->ext_copy.data_ack = ack_seq;
602 opts->ext_copy.ack64 = 1;
603 } else {
604 ack_size = TCPOLEN_MPTCP_DSS_ACK32;
605 opts->ext_copy.data_ack32 = (uint32_t)ack_seq;
606 opts->ext_copy.ack64 = 0;
607 }
608 opts->ext_copy.use_ack = 1;
609 opts->suboptions = OPTION_MPTCP_DSS;
610 WRITE_ONCE(msk->old_wspace, __mptcp_space((struct sock *)msk));
611
612 /* Add kind/length/subtype/flag overhead if mapping is not populated */
613 if (dss_size == 0)
614 ack_size += TCPOLEN_MPTCP_DSS_BASE;
615
616 dss_size += ack_size;
617
618 *size = ALIGN(dss_size, 4);
619 return true;
620}
621
622static u64 add_addr_generate_hmac(u64 key1, u64 key2,
623 struct mptcp_addr_info *addr)
624{
625 u16 port = ntohs(addr->port);
626 u8 hmac[SHA256_DIGEST_SIZE];
627 u8 msg[19];
628 int i = 0;
629
630 msg[i++] = addr->id;
631 if (addr->family == AF_INET) {
632 memcpy(&msg[i], &addr->addr.s_addr, 4);
633 i += 4;
634 }
635#if IS_ENABLED(CONFIG_MPTCP_IPV6)
636 else if (addr->family == AF_INET6) {
637 memcpy(&msg[i], &addr->addr6.s6_addr, 16);
638 i += 16;
639 }
640#endif
641 msg[i++] = port >> 8;
642 msg[i++] = port & 0xFF;
643
644 mptcp_crypto_hmac_sha(key1, key2, msg, i, hmac);
645
646 return get_unaligned_be64(&hmac[SHA256_DIGEST_SIZE - sizeof(u64)]);
647}
648
649static bool mptcp_established_options_add_addr(struct sock *sk, struct sk_buff *skb,
650 unsigned int *size,
651 unsigned int remaining,
652 struct mptcp_out_options *opts)
653{
654 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
655 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
656 bool drop_other_suboptions = false;
657 unsigned int opt_size = *size;
658 bool echo;
659 int len;
660
661 /* add addr will strip the existing options, be sure to avoid breaking
662 * MPC/MPJ handshakes
663 */
664 if (!mptcp_pm_should_add_signal(msk) ||
665 (opts->suboptions & (OPTION_MPTCP_MPJ_ACK | OPTION_MPTCP_MPC_ACK)) ||
666 !mptcp_pm_add_addr_signal(msk, skb, opt_size, remaining, &opts->addr,
667 &echo, &drop_other_suboptions))
668 return false;
669
670 if (drop_other_suboptions)
671 remaining += opt_size;
672 len = mptcp_add_addr_len(opts->addr.family, echo, !!opts->addr.port);
673 if (remaining < len)
674 return false;
675
676 *size = len;
677 if (drop_other_suboptions) {
678 pr_debug("drop other suboptions");
679 opts->suboptions = 0;
680
681 /* note that e.g. DSS could have written into the memory
682 * aliased by ahmac, we must reset the field here
683 * to avoid appending the hmac even for ADD_ADDR echo
684 * options
685 */
686 opts->ahmac = 0;
687 *size -= opt_size;
688 }
689 opts->suboptions |= OPTION_MPTCP_ADD_ADDR;
690 if (!echo) {
691 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ADDADDRTX);
692 opts->ahmac = add_addr_generate_hmac(msk->local_key,
693 msk->remote_key,
694 &opts->addr);
695 } else {
696 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ECHOADDTX);
697 }
698 pr_debug("addr_id=%d, ahmac=%llu, echo=%d, port=%d",
699 opts->addr.id, opts->ahmac, echo, ntohs(opts->addr.port));
700
701 return true;
702}
703
704static bool mptcp_established_options_rm_addr(struct sock *sk,
705 unsigned int *size,
706 unsigned int remaining,
707 struct mptcp_out_options *opts)
708{
709 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
710 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
711 struct mptcp_rm_list rm_list;
712 int i, len;
713
714 if (!mptcp_pm_should_rm_signal(msk) ||
715 !(mptcp_pm_rm_addr_signal(msk, remaining, &rm_list)))
716 return false;
717
718 len = mptcp_rm_addr_len(&rm_list);
719 if (len < 0)
720 return false;
721 if (remaining < len)
722 return false;
723
724 *size = len;
725 opts->suboptions |= OPTION_MPTCP_RM_ADDR;
726 opts->rm_list = rm_list;
727
728 for (i = 0; i < opts->rm_list.nr; i++)
729 pr_debug("rm_list_ids[%d]=%d", i, opts->rm_list.ids[i]);
730 MPTCP_ADD_STATS(sock_net(sk), MPTCP_MIB_RMADDRTX, opts->rm_list.nr);
731 return true;
732}
733
734static bool mptcp_established_options_mp_prio(struct sock *sk,
735 unsigned int *size,
736 unsigned int remaining,
737 struct mptcp_out_options *opts)
738{
739 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
740
741 /* can't send MP_PRIO with MPC, as they share the same option space:
742 * 'backup'. Also it makes no sense at all
743 */
744 if (!subflow->send_mp_prio || (opts->suboptions & OPTIONS_MPTCP_MPC))
745 return false;
746
747 /* account for the trailing 'nop' option */
748 if (remaining < TCPOLEN_MPTCP_PRIO_ALIGN)
749 return false;
750
751 *size = TCPOLEN_MPTCP_PRIO_ALIGN;
752 opts->suboptions |= OPTION_MPTCP_PRIO;
753 opts->backup = subflow->request_bkup;
754
755 pr_debug("prio=%d", opts->backup);
756
757 return true;
758}
759
760static noinline bool mptcp_established_options_rst(struct sock *sk, struct sk_buff *skb,
761 unsigned int *size,
762 unsigned int remaining,
763 struct mptcp_out_options *opts)
764{
765 const struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
766
767 if (remaining < TCPOLEN_MPTCP_RST)
768 return false;
769
770 *size = TCPOLEN_MPTCP_RST;
771 opts->suboptions |= OPTION_MPTCP_RST;
772 opts->reset_transient = subflow->reset_transient;
773 opts->reset_reason = subflow->reset_reason;
774 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPRSTTX);
775
776 return true;
777}
778
779static bool mptcp_established_options_fastclose(struct sock *sk,
780 unsigned int *size,
781 unsigned int remaining,
782 struct mptcp_out_options *opts)
783{
784 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
785 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
786
787 if (likely(!subflow->send_fastclose))
788 return false;
789
790 if (remaining < TCPOLEN_MPTCP_FASTCLOSE)
791 return false;
792
793 *size = TCPOLEN_MPTCP_FASTCLOSE;
794 opts->suboptions |= OPTION_MPTCP_FASTCLOSE;
795 opts->rcvr_key = msk->remote_key;
796
797 pr_debug("FASTCLOSE key=%llu", opts->rcvr_key);
798 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPFASTCLOSETX);
799 return true;
800}
801
802static bool mptcp_established_options_mp_fail(struct sock *sk,
803 unsigned int *size,
804 unsigned int remaining,
805 struct mptcp_out_options *opts)
806{
807 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
808
809 if (likely(!subflow->send_mp_fail))
810 return false;
811
812 if (remaining < TCPOLEN_MPTCP_FAIL)
813 return false;
814
815 *size = TCPOLEN_MPTCP_FAIL;
816 opts->suboptions |= OPTION_MPTCP_FAIL;
817 opts->fail_seq = subflow->map_seq;
818
819 pr_debug("MP_FAIL fail_seq=%llu", opts->fail_seq);
820 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPFAILTX);
821
822 return true;
823}
824
825bool mptcp_established_options(struct sock *sk, struct sk_buff *skb,
826 unsigned int *size, unsigned int remaining,
827 struct mptcp_out_options *opts)
828{
829 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
830 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
831 unsigned int opt_size = 0;
832 bool snd_data_fin;
833 bool ret = false;
834
835 opts->suboptions = 0;
836
837 if (unlikely(__mptcp_check_fallback(msk) && !mptcp_check_infinite_map(skb)))
838 return false;
839
840 if (unlikely(skb && TCP_SKB_CB(skb)->tcp_flags & TCPHDR_RST)) {
841 if (mptcp_established_options_fastclose(sk, &opt_size, remaining, opts) ||
842 mptcp_established_options_mp_fail(sk, &opt_size, remaining, opts)) {
843 *size += opt_size;
844 remaining -= opt_size;
845 }
846 /* MP_RST can be used with MP_FASTCLOSE and MP_FAIL if there is room */
847 if (mptcp_established_options_rst(sk, skb, &opt_size, remaining, opts)) {
848 *size += opt_size;
849 remaining -= opt_size;
850 }
851 return true;
852 }
853
854 snd_data_fin = mptcp_data_fin_enabled(msk);
855 if (mptcp_established_options_mp(sk, skb, snd_data_fin, &opt_size, opts))
856 ret = true;
857 else if (mptcp_established_options_dss(sk, skb, snd_data_fin, &opt_size, opts)) {
858 unsigned int mp_fail_size;
859
860 ret = true;
861 if (mptcp_established_options_mp_fail(sk, &mp_fail_size,
862 remaining - opt_size, opts)) {
863 *size += opt_size + mp_fail_size;
864 remaining -= opt_size - mp_fail_size;
865 return true;
866 }
867 }
868
869 /* we reserved enough space for the above options, and exceeding the
870 * TCP option space would be fatal
871 */
872 if (WARN_ON_ONCE(opt_size > remaining))
873 return false;
874
875 *size += opt_size;
876 remaining -= opt_size;
877 if (mptcp_established_options_add_addr(sk, skb, &opt_size, remaining, opts)) {
878 *size += opt_size;
879 remaining -= opt_size;
880 ret = true;
881 } else if (mptcp_established_options_rm_addr(sk, &opt_size, remaining, opts)) {
882 *size += opt_size;
883 remaining -= opt_size;
884 ret = true;
885 }
886
887 if (mptcp_established_options_mp_prio(sk, &opt_size, remaining, opts)) {
888 *size += opt_size;
889 remaining -= opt_size;
890 ret = true;
891 }
892
893 return ret;
894}
895
896bool mptcp_synack_options(const struct request_sock *req, unsigned int *size,
897 struct mptcp_out_options *opts)
898{
899 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
900
901 if (subflow_req->mp_capable) {
902 opts->suboptions = OPTION_MPTCP_MPC_SYNACK;
903 opts->sndr_key = subflow_req->local_key;
904 opts->csum_reqd = subflow_req->csum_reqd;
905 opts->allow_join_id0 = subflow_req->allow_join_id0;
906 *size = TCPOLEN_MPTCP_MPC_SYNACK;
907 pr_debug("subflow_req=%p, local_key=%llu",
908 subflow_req, subflow_req->local_key);
909 return true;
910 } else if (subflow_req->mp_join) {
911 opts->suboptions = OPTION_MPTCP_MPJ_SYNACK;
912 opts->backup = subflow_req->backup;
913 opts->join_id = subflow_req->local_id;
914 opts->thmac = subflow_req->thmac;
915 opts->nonce = subflow_req->local_nonce;
916 pr_debug("req=%p, bkup=%u, id=%u, thmac=%llu, nonce=%u",
917 subflow_req, opts->backup, opts->join_id,
918 opts->thmac, opts->nonce);
919 *size = TCPOLEN_MPTCP_MPJ_SYNACK;
920 return true;
921 }
922 return false;
923}
924
925static bool check_fully_established(struct mptcp_sock *msk, struct sock *ssk,
926 struct mptcp_subflow_context *subflow,
927 struct sk_buff *skb,
928 struct mptcp_options_received *mp_opt)
929{
930 /* here we can process OoO, in-window pkts, only in-sequence 4th ack
931 * will make the subflow fully established
932 */
933 if (likely(subflow->fully_established)) {
934 /* on passive sockets, check for 3rd ack retransmission
935 * note that msk is always set by subflow_syn_recv_sock()
936 * for mp_join subflows
937 */
938 if (TCP_SKB_CB(skb)->seq == subflow->ssn_offset + 1 &&
939 TCP_SKB_CB(skb)->end_seq == TCP_SKB_CB(skb)->seq &&
940 subflow->mp_join && (mp_opt->suboptions & OPTIONS_MPTCP_MPJ) &&
941 !subflow->request_join)
942 tcp_send_ack(ssk);
943 goto check_notify;
944 }
945
946 /* we must process OoO packets before the first subflow is fully
947 * established. OoO packets are instead a protocol violation
948 * for MP_JOIN subflows as the peer must not send any data
949 * before receiving the forth ack - cfr. RFC 8684 section 3.2.
950 */
951 if (TCP_SKB_CB(skb)->seq != subflow->ssn_offset + 1) {
952 if (subflow->mp_join)
953 goto reset;
954 if (subflow->is_mptfo && mp_opt->suboptions & OPTION_MPTCP_MPC_ACK)
955 goto set_fully_established;
956 return subflow->mp_capable;
957 }
958
959 if (subflow->remote_key_valid &&
960 (((mp_opt->suboptions & OPTION_MPTCP_DSS) && mp_opt->use_ack) ||
961 ((mp_opt->suboptions & OPTION_MPTCP_ADD_ADDR) && !mp_opt->echo))) {
962 /* subflows are fully established as soon as we get any
963 * additional ack, including ADD_ADDR.
964 */
965 goto set_fully_established;
966 }
967
968 /* If the first established packet does not contain MP_CAPABLE + data
969 * then fallback to TCP. Fallback scenarios requires a reset for
970 * MP_JOIN subflows.
971 */
972 if (!(mp_opt->suboptions & OPTIONS_MPTCP_MPC)) {
973 if (subflow->mp_join)
974 goto reset;
975 subflow->mp_capable = 0;
976 pr_fallback(msk);
977 mptcp_do_fallback(ssk);
978 return false;
979 }
980
981 if (mp_opt->deny_join_id0)
982 WRITE_ONCE(msk->pm.remote_deny_join_id0, true);
983
984 if (unlikely(!READ_ONCE(msk->pm.server_side)))
985 pr_warn_once("bogus mpc option on established client sk");
986
987set_fully_established:
988 mptcp_data_lock((struct sock *)msk);
989 __mptcp_subflow_fully_established(msk, subflow, mp_opt);
990 mptcp_data_unlock((struct sock *)msk);
991
992check_notify:
993 /* if the subflow is not already linked into the conn_list, we can't
994 * notify the PM: this subflow is still on the listener queue
995 * and the PM possibly acquiring the subflow lock could race with
996 * the listener close
997 */
998 if (likely(subflow->pm_notified) || list_empty(&subflow->node))
999 return true;
1000
1001 subflow->pm_notified = 1;
1002 if (subflow->mp_join) {
1003 clear_3rdack_retransmission(ssk);
1004 mptcp_pm_subflow_established(msk);
1005 } else {
1006 mptcp_pm_fully_established(msk, ssk);
1007 }
1008 return true;
1009
1010reset:
1011 mptcp_subflow_reset(ssk);
1012 return false;
1013}
1014
1015u64 __mptcp_expand_seq(u64 old_seq, u64 cur_seq)
1016{
1017 u32 old_seq32, cur_seq32;
1018
1019 old_seq32 = (u32)old_seq;
1020 cur_seq32 = (u32)cur_seq;
1021 cur_seq = (old_seq & GENMASK_ULL(63, 32)) + cur_seq32;
1022 if (unlikely(cur_seq32 < old_seq32 && before(old_seq32, cur_seq32)))
1023 return cur_seq + (1LL << 32);
1024
1025 /* reverse wrap could happen, too */
1026 if (unlikely(cur_seq32 > old_seq32 && after(old_seq32, cur_seq32)))
1027 return cur_seq - (1LL << 32);
1028 return cur_seq;
1029}
1030
1031static void __mptcp_snd_una_update(struct mptcp_sock *msk, u64 new_snd_una)
1032{
1033 msk->bytes_acked += new_snd_una - msk->snd_una;
1034 msk->snd_una = new_snd_una;
1035}
1036
1037static void ack_update_msk(struct mptcp_sock *msk,
1038 struct sock *ssk,
1039 struct mptcp_options_received *mp_opt)
1040{
1041 u64 new_wnd_end, new_snd_una, snd_nxt = READ_ONCE(msk->snd_nxt);
1042 struct sock *sk = (struct sock *)msk;
1043 u64 old_snd_una;
1044
1045 mptcp_data_lock(sk);
1046
1047 /* avoid ack expansion on update conflict, to reduce the risk of
1048 * wrongly expanding to a future ack sequence number, which is way
1049 * more dangerous than missing an ack
1050 */
1051 old_snd_una = msk->snd_una;
1052 new_snd_una = mptcp_expand_seq(old_snd_una, mp_opt->data_ack, mp_opt->ack64);
1053
1054 /* ACK for data not even sent yet? Ignore.*/
1055 if (unlikely(after64(new_snd_una, snd_nxt)))
1056 new_snd_una = old_snd_una;
1057
1058 new_wnd_end = new_snd_una + tcp_sk(ssk)->snd_wnd;
1059
1060 if (after64(new_wnd_end, msk->wnd_end))
1061 msk->wnd_end = new_wnd_end;
1062
1063 /* this assumes mptcp_incoming_options() is invoked after tcp_ack() */
1064 if (after64(msk->wnd_end, READ_ONCE(msk->snd_nxt)))
1065 __mptcp_check_push(sk, ssk);
1066
1067 if (after64(new_snd_una, old_snd_una)) {
1068 __mptcp_snd_una_update(msk, new_snd_una);
1069 __mptcp_data_acked(sk);
1070 }
1071 mptcp_data_unlock(sk);
1072
1073 trace_ack_update_msk(mp_opt->data_ack,
1074 old_snd_una, new_snd_una,
1075 new_wnd_end, msk->wnd_end);
1076}
1077
1078bool mptcp_update_rcv_data_fin(struct mptcp_sock *msk, u64 data_fin_seq, bool use_64bit)
1079{
1080 /* Skip if DATA_FIN was already received.
1081 * If updating simultaneously with the recvmsg loop, values
1082 * should match. If they mismatch, the peer is misbehaving and
1083 * we will prefer the most recent information.
1084 */
1085 if (READ_ONCE(msk->rcv_data_fin))
1086 return false;
1087
1088 WRITE_ONCE(msk->rcv_data_fin_seq,
1089 mptcp_expand_seq(READ_ONCE(msk->ack_seq), data_fin_seq, use_64bit));
1090 WRITE_ONCE(msk->rcv_data_fin, 1);
1091
1092 return true;
1093}
1094
1095static bool add_addr_hmac_valid(struct mptcp_sock *msk,
1096 struct mptcp_options_received *mp_opt)
1097{
1098 u64 hmac = 0;
1099
1100 if (mp_opt->echo)
1101 return true;
1102
1103 hmac = add_addr_generate_hmac(msk->remote_key,
1104 msk->local_key,
1105 &mp_opt->addr);
1106
1107 pr_debug("msk=%p, ahmac=%llu, mp_opt->ahmac=%llu\n",
1108 msk, hmac, mp_opt->ahmac);
1109
1110 return hmac == mp_opt->ahmac;
1111}
1112
1113/* Return false if a subflow has been reset, else return true */
1114bool mptcp_incoming_options(struct sock *sk, struct sk_buff *skb)
1115{
1116 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1117 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
1118 struct mptcp_options_received mp_opt;
1119 struct mptcp_ext *mpext;
1120
1121 if (__mptcp_check_fallback(msk)) {
1122 /* Keep it simple and unconditionally trigger send data cleanup and
1123 * pending queue spooling. We will need to acquire the data lock
1124 * for more accurate checks, and once the lock is acquired, such
1125 * helpers are cheap.
1126 */
1127 mptcp_data_lock(subflow->conn);
1128 if (sk_stream_memory_free(sk))
1129 __mptcp_check_push(subflow->conn, sk);
1130
1131 /* on fallback we just need to ignore the msk-level snd_una, as
1132 * this is really plain TCP
1133 */
1134 __mptcp_snd_una_update(msk, READ_ONCE(msk->snd_nxt));
1135
1136 __mptcp_data_acked(subflow->conn);
1137 mptcp_data_unlock(subflow->conn);
1138 return true;
1139 }
1140
1141 mptcp_get_options(skb, &mp_opt);
1142
1143 /* The subflow can be in close state only if check_fully_established()
1144 * just sent a reset. If so, tell the caller to ignore the current packet.
1145 */
1146 if (!check_fully_established(msk, sk, subflow, skb, &mp_opt))
1147 return sk->sk_state != TCP_CLOSE;
1148
1149 if (unlikely(mp_opt.suboptions != OPTION_MPTCP_DSS)) {
1150 if ((mp_opt.suboptions & OPTION_MPTCP_FASTCLOSE) &&
1151 msk->local_key == mp_opt.rcvr_key) {
1152 WRITE_ONCE(msk->rcv_fastclose, true);
1153 mptcp_schedule_work((struct sock *)msk);
1154 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPFASTCLOSERX);
1155 }
1156
1157 if ((mp_opt.suboptions & OPTION_MPTCP_ADD_ADDR) &&
1158 add_addr_hmac_valid(msk, &mp_opt)) {
1159 if (!mp_opt.echo) {
1160 mptcp_pm_add_addr_received(sk, &mp_opt.addr);
1161 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ADDADDR);
1162 } else {
1163 mptcp_pm_add_addr_echoed(msk, &mp_opt.addr);
1164 mptcp_pm_del_add_timer(msk, &mp_opt.addr, true);
1165 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ECHOADD);
1166 }
1167
1168 if (mp_opt.addr.port)
1169 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_PORTADD);
1170 }
1171
1172 if (mp_opt.suboptions & OPTION_MPTCP_RM_ADDR)
1173 mptcp_pm_rm_addr_received(msk, &mp_opt.rm_list);
1174
1175 if (mp_opt.suboptions & OPTION_MPTCP_PRIO) {
1176 mptcp_pm_mp_prio_received(sk, mp_opt.backup);
1177 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPPRIORX);
1178 }
1179
1180 if (mp_opt.suboptions & OPTION_MPTCP_FAIL) {
1181 mptcp_pm_mp_fail_received(sk, mp_opt.fail_seq);
1182 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPFAILRX);
1183 }
1184
1185 if (mp_opt.suboptions & OPTION_MPTCP_RST) {
1186 subflow->reset_seen = 1;
1187 subflow->reset_reason = mp_opt.reset_reason;
1188 subflow->reset_transient = mp_opt.reset_transient;
1189 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPRSTRX);
1190 }
1191
1192 if (!(mp_opt.suboptions & OPTION_MPTCP_DSS))
1193 return true;
1194 }
1195
1196 /* we can't wait for recvmsg() to update the ack_seq, otherwise
1197 * monodirectional flows will stuck
1198 */
1199 if (mp_opt.use_ack)
1200 ack_update_msk(msk, sk, &mp_opt);
1201
1202 /* Zero-data-length packets are dropped by the caller and not
1203 * propagated to the MPTCP layer, so the skb extension does not
1204 * need to be allocated or populated. DATA_FIN information, if
1205 * present, needs to be updated here before the skb is freed.
1206 */
1207 if (TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb)->end_seq) {
1208 if (mp_opt.data_fin && mp_opt.data_len == 1 &&
1209 mptcp_update_rcv_data_fin(msk, mp_opt.data_seq, mp_opt.dsn64))
1210 mptcp_schedule_work((struct sock *)msk);
1211
1212 return true;
1213 }
1214
1215 mpext = skb_ext_add(skb, SKB_EXT_MPTCP);
1216 if (!mpext)
1217 return true;
1218
1219 memset(mpext, 0, sizeof(*mpext));
1220
1221 if (likely(mp_opt.use_map)) {
1222 if (mp_opt.mpc_map) {
1223 /* this is an MP_CAPABLE carrying MPTCP data
1224 * we know this map the first chunk of data
1225 */
1226 mptcp_crypto_key_sha(subflow->remote_key, NULL,
1227 &mpext->data_seq);
1228 mpext->data_seq++;
1229 mpext->subflow_seq = 1;
1230 mpext->dsn64 = 1;
1231 mpext->mpc_map = 1;
1232 mpext->data_fin = 0;
1233 } else {
1234 mpext->data_seq = mp_opt.data_seq;
1235 mpext->subflow_seq = mp_opt.subflow_seq;
1236 mpext->dsn64 = mp_opt.dsn64;
1237 mpext->data_fin = mp_opt.data_fin;
1238 }
1239 mpext->data_len = mp_opt.data_len;
1240 mpext->use_map = 1;
1241 mpext->csum_reqd = !!(mp_opt.suboptions & OPTION_MPTCP_CSUMREQD);
1242
1243 if (mpext->csum_reqd)
1244 mpext->csum = mp_opt.csum;
1245 }
1246
1247 return true;
1248}
1249
1250static void mptcp_set_rwin(struct tcp_sock *tp, struct tcphdr *th)
1251{
1252 const struct sock *ssk = (const struct sock *)tp;
1253 struct mptcp_subflow_context *subflow;
1254 u64 ack_seq, rcv_wnd_old, rcv_wnd_new;
1255 struct mptcp_sock *msk;
1256 u32 new_win;
1257 u64 win;
1258
1259 subflow = mptcp_subflow_ctx(ssk);
1260 msk = mptcp_sk(subflow->conn);
1261
1262 ack_seq = READ_ONCE(msk->ack_seq);
1263 rcv_wnd_new = ack_seq + tp->rcv_wnd;
1264
1265 rcv_wnd_old = atomic64_read(&msk->rcv_wnd_sent);
1266 if (after64(rcv_wnd_new, rcv_wnd_old)) {
1267 u64 rcv_wnd;
1268
1269 for (;;) {
1270 rcv_wnd = atomic64_cmpxchg(&msk->rcv_wnd_sent, rcv_wnd_old, rcv_wnd_new);
1271
1272 if (rcv_wnd == rcv_wnd_old)
1273 break;
1274
1275 rcv_wnd_old = rcv_wnd;
1276 if (before64(rcv_wnd_new, rcv_wnd_old)) {
1277 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_RCVWNDCONFLICTUPDATE);
1278 goto raise_win;
1279 }
1280 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_RCVWNDCONFLICT);
1281 }
1282 return;
1283 }
1284
1285 if (rcv_wnd_new != rcv_wnd_old) {
1286raise_win:
1287 win = rcv_wnd_old - ack_seq;
1288 tp->rcv_wnd = min_t(u64, win, U32_MAX);
1289 new_win = tp->rcv_wnd;
1290
1291 /* Make sure we do not exceed the maximum possible
1292 * scaled window.
1293 */
1294 if (unlikely(th->syn))
1295 new_win = min(new_win, 65535U) << tp->rx_opt.rcv_wscale;
1296 if (!tp->rx_opt.rcv_wscale &&
1297 READ_ONCE(sock_net(ssk)->ipv4.sysctl_tcp_workaround_signed_windows))
1298 new_win = min(new_win, MAX_TCP_WINDOW);
1299 else
1300 new_win = min(new_win, (65535U << tp->rx_opt.rcv_wscale));
1301
1302 /* RFC1323 scaling applied */
1303 new_win >>= tp->rx_opt.rcv_wscale;
1304 th->window = htons(new_win);
1305 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_RCVWNDSHARED);
1306 }
1307}
1308
1309__sum16 __mptcp_make_csum(u64 data_seq, u32 subflow_seq, u16 data_len, __wsum sum)
1310{
1311 struct csum_pseudo_header header;
1312 __wsum csum;
1313
1314 /* cfr RFC 8684 3.3.1.:
1315 * the data sequence number used in the pseudo-header is
1316 * always the 64-bit value, irrespective of what length is used in the
1317 * DSS option itself.
1318 */
1319 header.data_seq = cpu_to_be64(data_seq);
1320 header.subflow_seq = htonl(subflow_seq);
1321 header.data_len = htons(data_len);
1322 header.csum = 0;
1323
1324 csum = csum_partial(&header, sizeof(header), sum);
1325 return csum_fold(csum);
1326}
1327
1328static __sum16 mptcp_make_csum(const struct mptcp_ext *mpext)
1329{
1330 return __mptcp_make_csum(mpext->data_seq, mpext->subflow_seq, mpext->data_len,
1331 ~csum_unfold(mpext->csum));
1332}
1333
1334static void put_len_csum(u16 len, __sum16 csum, void *data)
1335{
1336 __sum16 *sumptr = data + 2;
1337 __be16 *ptr = data;
1338
1339 put_unaligned_be16(len, ptr);
1340
1341 put_unaligned(csum, sumptr);
1342}
1343
1344void mptcp_write_options(struct tcphdr *th, __be32 *ptr, struct tcp_sock *tp,
1345 struct mptcp_out_options *opts)
1346{
1347 const struct sock *ssk = (const struct sock *)tp;
1348 struct mptcp_subflow_context *subflow;
1349
1350 /* Which options can be used together?
1351 *
1352 * X: mutually exclusive
1353 * O: often used together
1354 * C: can be used together in some cases
1355 * P: could be used together but we prefer not to (optimisations)
1356 *
1357 * Opt: | MPC | MPJ | DSS | ADD | RM | PRIO | FAIL | FC |
1358 * ------|------|------|------|------|------|------|------|------|
1359 * MPC |------|------|------|------|------|------|------|------|
1360 * MPJ | X |------|------|------|------|------|------|------|
1361 * DSS | X | X |------|------|------|------|------|------|
1362 * ADD | X | X | P |------|------|------|------|------|
1363 * RM | C | C | C | P |------|------|------|------|
1364 * PRIO | X | C | C | C | C |------|------|------|
1365 * FAIL | X | X | C | X | X | X |------|------|
1366 * FC | X | X | X | X | X | X | X |------|
1367 * RST | X | X | X | X | X | X | O | O |
1368 * ------|------|------|------|------|------|------|------|------|
1369 *
1370 * The same applies in mptcp_established_options() function.
1371 */
1372 if (likely(OPTION_MPTCP_DSS & opts->suboptions)) {
1373 struct mptcp_ext *mpext = &opts->ext_copy;
1374 u8 len = TCPOLEN_MPTCP_DSS_BASE;
1375 u8 flags = 0;
1376
1377 if (mpext->use_ack) {
1378 flags = MPTCP_DSS_HAS_ACK;
1379 if (mpext->ack64) {
1380 len += TCPOLEN_MPTCP_DSS_ACK64;
1381 flags |= MPTCP_DSS_ACK64;
1382 } else {
1383 len += TCPOLEN_MPTCP_DSS_ACK32;
1384 }
1385 }
1386
1387 if (mpext->use_map) {
1388 len += TCPOLEN_MPTCP_DSS_MAP64;
1389
1390 /* Use only 64-bit mapping flags for now, add
1391 * support for optional 32-bit mappings later.
1392 */
1393 flags |= MPTCP_DSS_HAS_MAP | MPTCP_DSS_DSN64;
1394 if (mpext->data_fin)
1395 flags |= MPTCP_DSS_DATA_FIN;
1396
1397 if (opts->csum_reqd)
1398 len += TCPOLEN_MPTCP_DSS_CHECKSUM;
1399 }
1400
1401 *ptr++ = mptcp_option(MPTCPOPT_DSS, len, 0, flags);
1402
1403 if (mpext->use_ack) {
1404 if (mpext->ack64) {
1405 put_unaligned_be64(mpext->data_ack, ptr);
1406 ptr += 2;
1407 } else {
1408 put_unaligned_be32(mpext->data_ack32, ptr);
1409 ptr += 1;
1410 }
1411 }
1412
1413 if (mpext->use_map) {
1414 put_unaligned_be64(mpext->data_seq, ptr);
1415 ptr += 2;
1416 put_unaligned_be32(mpext->subflow_seq, ptr);
1417 ptr += 1;
1418 if (opts->csum_reqd) {
1419 /* data_len == 0 is reserved for the infinite mapping,
1420 * the checksum will also be set to 0.
1421 */
1422 put_len_csum(mpext->data_len,
1423 (mpext->data_len ? mptcp_make_csum(mpext) : 0),
1424 ptr);
1425 } else {
1426 put_unaligned_be32(mpext->data_len << 16 |
1427 TCPOPT_NOP << 8 | TCPOPT_NOP, ptr);
1428 }
1429 ptr += 1;
1430 }
1431
1432 /* We might need to add MP_FAIL options in rare cases */
1433 if (unlikely(OPTION_MPTCP_FAIL & opts->suboptions))
1434 goto mp_fail;
1435 } else if (OPTIONS_MPTCP_MPC & opts->suboptions) {
1436 u8 len, flag = MPTCP_CAP_HMAC_SHA256;
1437
1438 if (OPTION_MPTCP_MPC_SYN & opts->suboptions) {
1439 len = TCPOLEN_MPTCP_MPC_SYN;
1440 } else if (OPTION_MPTCP_MPC_SYNACK & opts->suboptions) {
1441 len = TCPOLEN_MPTCP_MPC_SYNACK;
1442 } else if (opts->data_len) {
1443 len = TCPOLEN_MPTCP_MPC_ACK_DATA;
1444 if (opts->csum_reqd)
1445 len += TCPOLEN_MPTCP_DSS_CHECKSUM;
1446 } else {
1447 len = TCPOLEN_MPTCP_MPC_ACK;
1448 }
1449
1450 if (opts->csum_reqd)
1451 flag |= MPTCP_CAP_CHECKSUM_REQD;
1452
1453 if (!opts->allow_join_id0)
1454 flag |= MPTCP_CAP_DENY_JOIN_ID0;
1455
1456 *ptr++ = mptcp_option(MPTCPOPT_MP_CAPABLE, len,
1457 MPTCP_SUPPORTED_VERSION,
1458 flag);
1459
1460 if (!((OPTION_MPTCP_MPC_SYNACK | OPTION_MPTCP_MPC_ACK) &
1461 opts->suboptions))
1462 goto mp_capable_done;
1463
1464 put_unaligned_be64(opts->sndr_key, ptr);
1465 ptr += 2;
1466 if (!((OPTION_MPTCP_MPC_ACK) & opts->suboptions))
1467 goto mp_capable_done;
1468
1469 put_unaligned_be64(opts->rcvr_key, ptr);
1470 ptr += 2;
1471 if (!opts->data_len)
1472 goto mp_capable_done;
1473
1474 if (opts->csum_reqd) {
1475 put_len_csum(opts->data_len,
1476 __mptcp_make_csum(opts->data_seq,
1477 opts->subflow_seq,
1478 opts->data_len,
1479 ~csum_unfold(opts->csum)),
1480 ptr);
1481 } else {
1482 put_unaligned_be32(opts->data_len << 16 |
1483 TCPOPT_NOP << 8 | TCPOPT_NOP, ptr);
1484 }
1485 ptr += 1;
1486
1487 /* MPC is additionally mutually exclusive with MP_PRIO */
1488 goto mp_capable_done;
1489 } else if (OPTIONS_MPTCP_MPJ & opts->suboptions) {
1490 if (OPTION_MPTCP_MPJ_SYN & opts->suboptions) {
1491 *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1492 TCPOLEN_MPTCP_MPJ_SYN,
1493 opts->backup, opts->join_id);
1494 put_unaligned_be32(opts->token, ptr);
1495 ptr += 1;
1496 put_unaligned_be32(opts->nonce, ptr);
1497 ptr += 1;
1498 } else if (OPTION_MPTCP_MPJ_SYNACK & opts->suboptions) {
1499 *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1500 TCPOLEN_MPTCP_MPJ_SYNACK,
1501 opts->backup, opts->join_id);
1502 put_unaligned_be64(opts->thmac, ptr);
1503 ptr += 2;
1504 put_unaligned_be32(opts->nonce, ptr);
1505 ptr += 1;
1506 } else {
1507 *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1508 TCPOLEN_MPTCP_MPJ_ACK, 0, 0);
1509 memcpy(ptr, opts->hmac, MPTCPOPT_HMAC_LEN);
1510 ptr += 5;
1511 }
1512 } else if (OPTION_MPTCP_ADD_ADDR & opts->suboptions) {
1513 u8 len = TCPOLEN_MPTCP_ADD_ADDR_BASE;
1514 u8 echo = MPTCP_ADDR_ECHO;
1515
1516#if IS_ENABLED(CONFIG_MPTCP_IPV6)
1517 if (opts->addr.family == AF_INET6)
1518 len = TCPOLEN_MPTCP_ADD_ADDR6_BASE;
1519#endif
1520
1521 if (opts->addr.port)
1522 len += TCPOLEN_MPTCP_PORT_LEN;
1523
1524 if (opts->ahmac) {
1525 len += sizeof(opts->ahmac);
1526 echo = 0;
1527 }
1528
1529 *ptr++ = mptcp_option(MPTCPOPT_ADD_ADDR,
1530 len, echo, opts->addr.id);
1531 if (opts->addr.family == AF_INET) {
1532 memcpy((u8 *)ptr, (u8 *)&opts->addr.addr.s_addr, 4);
1533 ptr += 1;
1534 }
1535#if IS_ENABLED(CONFIG_MPTCP_IPV6)
1536 else if (opts->addr.family == AF_INET6) {
1537 memcpy((u8 *)ptr, opts->addr.addr6.s6_addr, 16);
1538 ptr += 4;
1539 }
1540#endif
1541
1542 if (!opts->addr.port) {
1543 if (opts->ahmac) {
1544 put_unaligned_be64(opts->ahmac, ptr);
1545 ptr += 2;
1546 }
1547 } else {
1548 u16 port = ntohs(opts->addr.port);
1549
1550 if (opts->ahmac) {
1551 u8 *bptr = (u8 *)ptr;
1552
1553 put_unaligned_be16(port, bptr);
1554 bptr += 2;
1555 put_unaligned_be64(opts->ahmac, bptr);
1556 bptr += 8;
1557 put_unaligned_be16(TCPOPT_NOP << 8 |
1558 TCPOPT_NOP, bptr);
1559
1560 ptr += 3;
1561 } else {
1562 put_unaligned_be32(port << 16 |
1563 TCPOPT_NOP << 8 |
1564 TCPOPT_NOP, ptr);
1565 ptr += 1;
1566 }
1567 }
1568 } else if (unlikely(OPTION_MPTCP_FASTCLOSE & opts->suboptions)) {
1569 /* FASTCLOSE is mutually exclusive with others except RST */
1570 *ptr++ = mptcp_option(MPTCPOPT_MP_FASTCLOSE,
1571 TCPOLEN_MPTCP_FASTCLOSE,
1572 0, 0);
1573 put_unaligned_be64(opts->rcvr_key, ptr);
1574 ptr += 2;
1575
1576 if (OPTION_MPTCP_RST & opts->suboptions)
1577 goto mp_rst;
1578 return;
1579 } else if (unlikely(OPTION_MPTCP_FAIL & opts->suboptions)) {
1580mp_fail:
1581 /* MP_FAIL is mutually exclusive with others except RST */
1582 subflow = mptcp_subflow_ctx(ssk);
1583 subflow->send_mp_fail = 0;
1584
1585 *ptr++ = mptcp_option(MPTCPOPT_MP_FAIL,
1586 TCPOLEN_MPTCP_FAIL,
1587 0, 0);
1588 put_unaligned_be64(opts->fail_seq, ptr);
1589 ptr += 2;
1590
1591 if (OPTION_MPTCP_RST & opts->suboptions)
1592 goto mp_rst;
1593 return;
1594 } else if (unlikely(OPTION_MPTCP_RST & opts->suboptions)) {
1595mp_rst:
1596 *ptr++ = mptcp_option(MPTCPOPT_RST,
1597 TCPOLEN_MPTCP_RST,
1598 opts->reset_transient,
1599 opts->reset_reason);
1600 return;
1601 }
1602
1603 if (OPTION_MPTCP_PRIO & opts->suboptions) {
1604 subflow = mptcp_subflow_ctx(ssk);
1605 subflow->send_mp_prio = 0;
1606
1607 *ptr++ = mptcp_option(MPTCPOPT_MP_PRIO,
1608 TCPOLEN_MPTCP_PRIO,
1609 opts->backup, TCPOPT_NOP);
1610
1611 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_MPPRIOTX);
1612 }
1613
1614mp_capable_done:
1615 if (OPTION_MPTCP_RM_ADDR & opts->suboptions) {
1616 u8 i = 1;
1617
1618 *ptr++ = mptcp_option(MPTCPOPT_RM_ADDR,
1619 TCPOLEN_MPTCP_RM_ADDR_BASE + opts->rm_list.nr,
1620 0, opts->rm_list.ids[0]);
1621
1622 while (i < opts->rm_list.nr) {
1623 u8 id1, id2, id3, id4;
1624
1625 id1 = opts->rm_list.ids[i];
1626 id2 = i + 1 < opts->rm_list.nr ? opts->rm_list.ids[i + 1] : TCPOPT_NOP;
1627 id3 = i + 2 < opts->rm_list.nr ? opts->rm_list.ids[i + 2] : TCPOPT_NOP;
1628 id4 = i + 3 < opts->rm_list.nr ? opts->rm_list.ids[i + 3] : TCPOPT_NOP;
1629 put_unaligned_be32(id1 << 24 | id2 << 16 | id3 << 8 | id4, ptr);
1630 ptr += 1;
1631 i += 4;
1632 }
1633 }
1634
1635 if (tp)
1636 mptcp_set_rwin(tp, th);
1637}
1638
1639__be32 mptcp_get_reset_option(const struct sk_buff *skb)
1640{
1641 const struct mptcp_ext *ext = mptcp_get_ext(skb);
1642 u8 flags, reason;
1643
1644 if (ext) {
1645 flags = ext->reset_transient;
1646 reason = ext->reset_reason;
1647
1648 return mptcp_option(MPTCPOPT_RST, TCPOLEN_MPTCP_RST,
1649 flags, reason);
1650 }
1651
1652 return htonl(0u);
1653}
1654EXPORT_SYMBOL_GPL(mptcp_get_reset_option);