1 // SPDX-License-Identifier: GPL-2.0
2 /* Multipath TCP
3 *
4 * Copyright (c) 2017 - 2019, Intel Corporation.
5 */
6
7 #define pr_fmt(fmt) "MPTCP: " fmt
8
9 #include <linux/kernel.h>
10 #include <crypto/sha2.h>
11 #include <net/tcp.h>
12 #include <net/mptcp.h>
13 #include "protocol.h"
14 #include "mib.h"
15
16 #include <trace/events/mptcp.h>
17
mptcp_cap_flag_sha256(u8 flags)18 static bool mptcp_cap_flag_sha256(u8 flags)
19 {
20 return (flags & MPTCP_CAP_FLAG_MASK) == MPTCP_CAP_HMAC_SHA256;
21 }
22
mptcp_parse_option(const struct sk_buff * skb,const unsigned char * ptr,int opsize,struct mptcp_options_received * mp_opt)23 static void mptcp_parse_option(const struct sk_buff *skb,
24 const unsigned char *ptr, int opsize,
25 struct mptcp_options_received *mp_opt)
26 {
27 u8 subtype = *ptr >> 4;
28 int expected_opsize;
29 u16 subopt;
30 u8 version;
31 u8 flags;
32 u8 i;
33
34 switch (subtype) {
35 case MPTCPOPT_MP_CAPABLE:
36 /* strict size checking */
37 if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)) {
38 if (skb->len > tcp_hdr(skb)->doff << 2)
39 expected_opsize = TCPOLEN_MPTCP_MPC_ACK_DATA;
40 else
41 expected_opsize = TCPOLEN_MPTCP_MPC_ACK;
42 subopt = OPTION_MPTCP_MPC_ACK;
43 } else {
44 if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_ACK) {
45 expected_opsize = TCPOLEN_MPTCP_MPC_SYNACK;
46 subopt = OPTION_MPTCP_MPC_SYNACK;
47 } else {
48 expected_opsize = TCPOLEN_MPTCP_MPC_SYN;
49 subopt = OPTION_MPTCP_MPC_SYN;
50 }
51 }
52
53 /* Cfr RFC 8684 Section 3.3.0:
54 * If a checksum is present but its use had
55 * not been negotiated in the MP_CAPABLE handshake, the receiver MUST
56 * close the subflow with a RST, as it is not behaving as negotiated.
57 * If a checksum is not present when its use has been negotiated, the
58 * receiver MUST close the subflow with a RST, as it is considered
59 * broken
60 * We parse even option with mismatching csum presence, so that
61 * later in subflow_data_ready we can trigger the reset.
62 */
63 if (opsize != expected_opsize &&
64 (expected_opsize != TCPOLEN_MPTCP_MPC_ACK_DATA ||
65 opsize != TCPOLEN_MPTCP_MPC_ACK_DATA_CSUM))
66 break;
67
68 /* try to be gentle vs future versions on the initial syn */
69 version = *ptr++ & MPTCP_VERSION_MASK;
70 if (opsize != TCPOLEN_MPTCP_MPC_SYN) {
71 if (version != MPTCP_SUPPORTED_VERSION)
72 break;
73 } else if (version < MPTCP_SUPPORTED_VERSION) {
74 break;
75 }
76
77 flags = *ptr++;
78 if (!mptcp_cap_flag_sha256(flags) ||
79 (flags & MPTCP_CAP_EXTENSIBILITY))
80 break;
81
82 /* RFC 6824, Section 3.1:
83 * "For the Checksum Required bit (labeled "A"), if either
84 * host requires the use of checksums, checksums MUST be used.
85 * In other words, the only way for checksums not to be used
86 * is if both hosts in their SYNs set A=0."
87 */
88 if (flags & MPTCP_CAP_CHECKSUM_REQD)
89 mp_opt->suboptions |= OPTION_MPTCP_CSUMREQD;
90
91 mp_opt->deny_join_id0 = !!(flags & MPTCP_CAP_DENY_JOIN_ID0);
92
93 mp_opt->suboptions |= subopt;
94 if (opsize >= TCPOLEN_MPTCP_MPC_SYNACK) {
95 mp_opt->sndr_key = get_unaligned_be64(ptr);
96 ptr += 8;
97 }
98 if (opsize >= TCPOLEN_MPTCP_MPC_ACK) {
99 mp_opt->rcvr_key = get_unaligned_be64(ptr);
100 ptr += 8;
101 }
102 if (opsize >= TCPOLEN_MPTCP_MPC_ACK_DATA) {
103 /* Section 3.1.:
104 * "the data parameters in a MP_CAPABLE are semantically
105 * equivalent to those in a DSS option and can be used
106 * interchangeably."
107 */
108 mp_opt->suboptions |= OPTION_MPTCP_DSS;
109 mp_opt->use_map = 1;
110 mp_opt->mpc_map = 1;
111 mp_opt->data_len = get_unaligned_be16(ptr);
112 ptr += 2;
113 }
114 if (opsize == TCPOLEN_MPTCP_MPC_ACK_DATA_CSUM) {
115 mp_opt->csum = get_unaligned((__force __sum16 *)ptr);
116 mp_opt->suboptions |= OPTION_MPTCP_CSUMREQD;
117 ptr += 2;
118 }
119 pr_debug("MP_CAPABLE version=%x, flags=%x, optlen=%d sndr=%llu, rcvr=%llu len=%d csum=%u\n",
120 version, flags, opsize, mp_opt->sndr_key,
121 mp_opt->rcvr_key, mp_opt->data_len, mp_opt->csum);
122 break;
123
124 case MPTCPOPT_MP_JOIN:
125 if (opsize == TCPOLEN_MPTCP_MPJ_SYN) {
126 mp_opt->suboptions |= OPTION_MPTCP_MPJ_SYN;
127 mp_opt->backup = *ptr++ & MPTCPOPT_BACKUP;
128 mp_opt->join_id = *ptr++;
129 mp_opt->token = get_unaligned_be32(ptr);
130 ptr += 4;
131 mp_opt->nonce = get_unaligned_be32(ptr);
132 ptr += 4;
133 pr_debug("MP_JOIN bkup=%u, id=%u, token=%u, nonce=%u\n",
134 mp_opt->backup, mp_opt->join_id,
135 mp_opt->token, mp_opt->nonce);
136 } else if (opsize == TCPOLEN_MPTCP_MPJ_SYNACK) {
137 mp_opt->suboptions |= OPTION_MPTCP_MPJ_SYNACK;
138 mp_opt->backup = *ptr++ & MPTCPOPT_BACKUP;
139 mp_opt->join_id = *ptr++;
140 mp_opt->thmac = get_unaligned_be64(ptr);
141 ptr += 8;
142 mp_opt->nonce = get_unaligned_be32(ptr);
143 ptr += 4;
144 pr_debug("MP_JOIN bkup=%u, id=%u, thmac=%llu, nonce=%u\n",
145 mp_opt->backup, mp_opt->join_id,
146 mp_opt->thmac, mp_opt->nonce);
147 } else if (opsize == TCPOLEN_MPTCP_MPJ_ACK) {
148 mp_opt->suboptions |= OPTION_MPTCP_MPJ_ACK;
149 ptr += 2;
150 memcpy(mp_opt->hmac, ptr, MPTCPOPT_HMAC_LEN);
151 pr_debug("MP_JOIN hmac\n");
152 }
153 break;
154
155 case MPTCPOPT_DSS:
156 pr_debug("DSS\n");
157 ptr++;
158
159 flags = (*ptr++) & MPTCP_DSS_FLAG_MASK;
160 mp_opt->data_fin = (flags & MPTCP_DSS_DATA_FIN) != 0;
161 mp_opt->dsn64 = (flags & MPTCP_DSS_DSN64) != 0;
162 mp_opt->use_map = (flags & MPTCP_DSS_HAS_MAP) != 0;
163 mp_opt->ack64 = (flags & MPTCP_DSS_ACK64) != 0;
164 mp_opt->use_ack = (flags & MPTCP_DSS_HAS_ACK);
165
166 pr_debug("data_fin=%d dsn64=%d use_map=%d ack64=%d use_ack=%d\n",
167 mp_opt->data_fin, mp_opt->dsn64,
168 mp_opt->use_map, mp_opt->ack64,
169 mp_opt->use_ack);
170
171 expected_opsize = TCPOLEN_MPTCP_DSS_BASE;
172
173 if (mp_opt->use_ack) {
174 if (mp_opt->ack64)
175 expected_opsize += TCPOLEN_MPTCP_DSS_ACK64;
176 else
177 expected_opsize += TCPOLEN_MPTCP_DSS_ACK32;
178 }
179
180 if (mp_opt->use_map) {
181 if (mp_opt->dsn64)
182 expected_opsize += TCPOLEN_MPTCP_DSS_MAP64;
183 else
184 expected_opsize += TCPOLEN_MPTCP_DSS_MAP32;
185 }
186
187 /* Always parse any csum presence combination, we will enforce
188 * RFC 8684 Section 3.3.0 checks later in subflow_data_ready
189 */
190 if (opsize != expected_opsize &&
191 opsize != expected_opsize + TCPOLEN_MPTCP_DSS_CHECKSUM)
192 break;
193
194 mp_opt->suboptions |= OPTION_MPTCP_DSS;
195 if (mp_opt->use_ack) {
196 if (mp_opt->ack64) {
197 mp_opt->data_ack = get_unaligned_be64(ptr);
198 ptr += 8;
199 } else {
200 mp_opt->data_ack = get_unaligned_be32(ptr);
201 ptr += 4;
202 }
203
204 pr_debug("data_ack=%llu\n", mp_opt->data_ack);
205 }
206
207 if (mp_opt->use_map) {
208 if (mp_opt->dsn64) {
209 mp_opt->data_seq = get_unaligned_be64(ptr);
210 ptr += 8;
211 } else {
212 mp_opt->data_seq = get_unaligned_be32(ptr);
213 ptr += 4;
214 }
215
216 mp_opt->subflow_seq = get_unaligned_be32(ptr);
217 ptr += 4;
218
219 mp_opt->data_len = get_unaligned_be16(ptr);
220 ptr += 2;
221
222 if (opsize == expected_opsize + TCPOLEN_MPTCP_DSS_CHECKSUM) {
223 mp_opt->suboptions |= OPTION_MPTCP_CSUMREQD;
224 mp_opt->csum = get_unaligned((__force __sum16 *)ptr);
225 ptr += 2;
226 }
227
228 pr_debug("data_seq=%llu subflow_seq=%u data_len=%u csum=%d:%u\n",
229 mp_opt->data_seq, mp_opt->subflow_seq,
230 mp_opt->data_len, !!(mp_opt->suboptions & OPTION_MPTCP_CSUMREQD),
231 mp_opt->csum);
232 }
233
234 break;
235
236 case MPTCPOPT_ADD_ADDR:
237 mp_opt->echo = (*ptr++) & MPTCP_ADDR_ECHO;
238 if (!mp_opt->echo) {
239 if (opsize == TCPOLEN_MPTCP_ADD_ADDR ||
240 opsize == TCPOLEN_MPTCP_ADD_ADDR_PORT)
241 mp_opt->addr.family = AF_INET;
242 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
243 else if (opsize == TCPOLEN_MPTCP_ADD_ADDR6 ||
244 opsize == TCPOLEN_MPTCP_ADD_ADDR6_PORT)
245 mp_opt->addr.family = AF_INET6;
246 #endif
247 else
248 break;
249 } else {
250 if (opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE ||
251 opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE_PORT)
252 mp_opt->addr.family = AF_INET;
253 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
254 else if (opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE ||
255 opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE_PORT)
256 mp_opt->addr.family = AF_INET6;
257 #endif
258 else
259 break;
260 }
261
262 mp_opt->suboptions |= OPTION_MPTCP_ADD_ADDR;
263 mp_opt->addr.id = *ptr++;
264 mp_opt->addr.port = 0;
265 mp_opt->ahmac = 0;
266 if (mp_opt->addr.family == AF_INET) {
267 memcpy((u8 *)&mp_opt->addr.addr.s_addr, (u8 *)ptr, 4);
268 ptr += 4;
269 if (opsize == TCPOLEN_MPTCP_ADD_ADDR_PORT ||
270 opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE_PORT) {
271 mp_opt->addr.port = htons(get_unaligned_be16(ptr));
272 ptr += 2;
273 }
274 }
275 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
276 else {
277 memcpy(mp_opt->addr.addr6.s6_addr, (u8 *)ptr, 16);
278 ptr += 16;
279 if (opsize == TCPOLEN_MPTCP_ADD_ADDR6_PORT ||
280 opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE_PORT) {
281 mp_opt->addr.port = htons(get_unaligned_be16(ptr));
282 ptr += 2;
283 }
284 }
285 #endif
286 if (!mp_opt->echo) {
287 mp_opt->ahmac = get_unaligned_be64(ptr);
288 ptr += 8;
289 }
290 pr_debug("ADD_ADDR%s: id=%d, ahmac=%llu, echo=%d, port=%d\n",
291 (mp_opt->addr.family == AF_INET6) ? "6" : "",
292 mp_opt->addr.id, mp_opt->ahmac, mp_opt->echo, ntohs(mp_opt->addr.port));
293 break;
294
295 case MPTCPOPT_RM_ADDR:
296 if (opsize < TCPOLEN_MPTCP_RM_ADDR_BASE + 1 ||
297 opsize > TCPOLEN_MPTCP_RM_ADDR_BASE + MPTCP_RM_IDS_MAX)
298 break;
299
300 ptr++;
301
302 mp_opt->suboptions |= OPTION_MPTCP_RM_ADDR;
303 mp_opt->rm_list.nr = opsize - TCPOLEN_MPTCP_RM_ADDR_BASE;
304 for (i = 0; i < mp_opt->rm_list.nr; i++)
305 mp_opt->rm_list.ids[i] = *ptr++;
306 pr_debug("RM_ADDR: rm_list_nr=%d\n", mp_opt->rm_list.nr);
307 break;
308
309 case MPTCPOPT_MP_PRIO:
310 if (opsize != TCPOLEN_MPTCP_PRIO)
311 break;
312
313 mp_opt->suboptions |= OPTION_MPTCP_PRIO;
314 mp_opt->backup = *ptr++ & MPTCP_PRIO_BKUP;
315 pr_debug("MP_PRIO: prio=%d\n", mp_opt->backup);
316 break;
317
318 case MPTCPOPT_MP_FASTCLOSE:
319 if (opsize != TCPOLEN_MPTCP_FASTCLOSE)
320 break;
321
322 ptr += 2;
323 mp_opt->rcvr_key = get_unaligned_be64(ptr);
324 ptr += 8;
325 mp_opt->suboptions |= OPTION_MPTCP_FASTCLOSE;
326 pr_debug("MP_FASTCLOSE: recv_key=%llu\n", mp_opt->rcvr_key);
327 break;
328
329 case MPTCPOPT_RST:
330 if (opsize != TCPOLEN_MPTCP_RST)
331 break;
332
333 if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_RST))
334 break;
335
336 mp_opt->suboptions |= OPTION_MPTCP_RST;
337 flags = *ptr++;
338 mp_opt->reset_transient = flags & MPTCP_RST_TRANSIENT;
339 mp_opt->reset_reason = *ptr;
340 pr_debug("MP_RST: transient=%u reason=%u\n",
341 mp_opt->reset_transient, mp_opt->reset_reason);
342 break;
343
344 case MPTCPOPT_MP_FAIL:
345 if (opsize != TCPOLEN_MPTCP_FAIL)
346 break;
347
348 ptr += 2;
349 mp_opt->suboptions |= OPTION_MPTCP_FAIL;
350 mp_opt->fail_seq = get_unaligned_be64(ptr);
351 pr_debug("MP_FAIL: data_seq=%llu\n", mp_opt->fail_seq);
352 break;
353
354 default:
355 break;
356 }
357 }
358
mptcp_get_options(const struct sk_buff * skb,struct mptcp_options_received * mp_opt)359 void mptcp_get_options(const struct sk_buff *skb,
360 struct mptcp_options_received *mp_opt)
361 {
362 const struct tcphdr *th = tcp_hdr(skb);
363 const unsigned char *ptr;
364 int length;
365
366 /* Ensure that casting the whole status to u32 is efficient and safe */
367 BUILD_BUG_ON(sizeof_field(struct mptcp_options_received, status) != sizeof(u32));
368 BUILD_BUG_ON(!IS_ALIGNED(offsetof(struct mptcp_options_received, status),
369 sizeof(u32)));
370 *(u32 *)&mp_opt->status = 0;
371
372 length = (th->doff * 4) - sizeof(struct tcphdr);
373 ptr = (const unsigned char *)(th + 1);
374
375 while (length > 0) {
376 int opcode = *ptr++;
377 int opsize;
378
379 switch (opcode) {
380 case TCPOPT_EOL:
381 return;
382 case TCPOPT_NOP: /* Ref: RFC 793 section 3.1 */
383 length--;
384 continue;
385 default:
386 if (length < 2)
387 return;
388 opsize = *ptr++;
389 if (opsize < 2) /* "silly options" */
390 return;
391 if (opsize > length)
392 return; /* don't parse partial options */
393 if (opcode == TCPOPT_MPTCP)
394 mptcp_parse_option(skb, ptr, opsize, mp_opt);
395 ptr += opsize - 2;
396 length -= opsize;
397 }
398 }
399 }
400
mptcp_syn_options(struct sock * sk,const struct sk_buff * skb,unsigned int * size,struct mptcp_out_options * opts)401 bool mptcp_syn_options(struct sock *sk, const struct sk_buff *skb,
402 unsigned int *size, struct mptcp_out_options *opts)
403 {
404 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
405
406 /* we will use snd_isn to detect first pkt [re]transmission
407 * in mptcp_established_options_mp()
408 */
409 subflow->snd_isn = TCP_SKB_CB(skb)->end_seq;
410 if (subflow->request_mptcp) {
411 if (unlikely(subflow_simultaneous_connect(sk))) {
412 WARN_ON_ONCE(!mptcp_try_fallback(sk, MPTCP_MIB_SIMULTCONNFALLBACK));
413
414 /* Ensure mptcp_finish_connect() will not process the
415 * MPC handshake.
416 */
417 subflow->request_mptcp = 0;
418 return false;
419 }
420
421 opts->suboptions = OPTION_MPTCP_MPC_SYN;
422 opts->csum_reqd = mptcp_is_checksum_enabled(sock_net(sk));
423 opts->allow_join_id0 = mptcp_allow_join_id0(sock_net(sk));
424 *size = TCPOLEN_MPTCP_MPC_SYN;
425 return true;
426 } else if (subflow->request_join) {
427 pr_debug("remote_token=%u, nonce=%u\n", subflow->remote_token,
428 subflow->local_nonce);
429 opts->suboptions = OPTION_MPTCP_MPJ_SYN;
430 opts->join_id = subflow->local_id;
431 opts->token = subflow->remote_token;
432 opts->nonce = subflow->local_nonce;
433 opts->backup = subflow->request_bkup;
434 *size = TCPOLEN_MPTCP_MPJ_SYN;
435 return true;
436 }
437 return false;
438 }
439
clear_3rdack_retransmission(struct sock * sk)440 static void clear_3rdack_retransmission(struct sock *sk)
441 {
442 struct inet_connection_sock *icsk = inet_csk(sk);
443
444 sk_stop_timer(sk, &icsk->icsk_delack_timer);
445 icsk->icsk_ack.ato = 0;
446 icsk->icsk_ack.pending &= ~(ICSK_ACK_SCHED | ICSK_ACK_TIMER);
447 }
448
mptcp_established_options_mp(struct sock * sk,struct sk_buff * skb,bool snd_data_fin_enable,unsigned int * size,struct mptcp_out_options * opts)449 static bool mptcp_established_options_mp(struct sock *sk, struct sk_buff *skb,
450 bool snd_data_fin_enable,
451 unsigned int *size,
452 struct mptcp_out_options *opts)
453 {
454 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
455 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
456 struct mptcp_ext *mpext;
457 unsigned int data_len;
458 u8 len;
459
460 /* When skb is not available, we better over-estimate the emitted
461 * options len. A full DSS option (28 bytes) is longer than
462 * TCPOLEN_MPTCP_MPC_ACK_DATA(22) or TCPOLEN_MPTCP_MPJ_ACK(24), so
463 * tell the caller to defer the estimate to
464 * mptcp_established_options_dss(), which will reserve enough space.
465 */
466 if (!skb)
467 return false;
468
469 /* MPC/MPJ needed only on 3rd ack packet, DATA_FIN and TCP shutdown take precedence */
470 if (READ_ONCE(subflow->fully_established) || snd_data_fin_enable ||
471 subflow->snd_isn != TCP_SKB_CB(skb)->seq ||
472 sk->sk_state != TCP_ESTABLISHED)
473 return false;
474
475 if (subflow->mp_capable) {
476 mpext = mptcp_get_ext(skb);
477 data_len = mpext ? mpext->data_len : 0;
478
479 /* we will check ops->data_len in mptcp_write_options() to
480 * discriminate between TCPOLEN_MPTCP_MPC_ACK_DATA and
481 * TCPOLEN_MPTCP_MPC_ACK
482 */
483 opts->data_len = data_len;
484 opts->suboptions = OPTION_MPTCP_MPC_ACK;
485 opts->sndr_key = subflow->local_key;
486 opts->rcvr_key = subflow->remote_key;
487 opts->csum_reqd = READ_ONCE(msk->csum_enabled);
488 opts->allow_join_id0 = mptcp_allow_join_id0(sock_net(sk));
489
490 /* Section 3.1.
491 * The MP_CAPABLE option is carried on the SYN, SYN/ACK, and ACK
492 * packets that start the first subflow of an MPTCP connection,
493 * as well as the first packet that carries data
494 */
495 if (data_len > 0) {
496 len = TCPOLEN_MPTCP_MPC_ACK_DATA;
497 if (opts->csum_reqd) {
498 /* we need to propagate more info to csum the pseudo hdr */
499 opts->data_seq = mpext->data_seq;
500 opts->subflow_seq = mpext->subflow_seq;
501 opts->csum = mpext->csum;
502 len += TCPOLEN_MPTCP_DSS_CHECKSUM;
503 }
504 *size = ALIGN(len, 4);
505 } else {
506 *size = TCPOLEN_MPTCP_MPC_ACK;
507 }
508
509 pr_debug("subflow=%p, local_key=%llu, remote_key=%llu map_len=%d\n",
510 subflow, subflow->local_key, subflow->remote_key,
511 data_len);
512
513 return true;
514 } else if (subflow->mp_join) {
515 opts->suboptions = OPTION_MPTCP_MPJ_ACK;
516 memcpy(opts->hmac, subflow->hmac, MPTCPOPT_HMAC_LEN);
517 *size = TCPOLEN_MPTCP_MPJ_ACK;
518 pr_debug("subflow=%p\n", subflow);
519
520 /* we can use the full delegate action helper only from BH context
521 * If we are in process context - sk is flushing the backlog at
522 * socket lock release time - just set the appropriate flag, will
523 * be handled by the release callback
524 */
525 if (sock_owned_by_user(sk))
526 set_bit(MPTCP_DELEGATE_ACK, &subflow->delegated_status);
527 else
528 mptcp_subflow_delegate(subflow, MPTCP_DELEGATE_ACK);
529 return true;
530 }
531 return false;
532 }
533
mptcp_write_data_fin(struct mptcp_subflow_context * subflow,struct sk_buff * skb,struct mptcp_ext * ext)534 static void mptcp_write_data_fin(struct mptcp_subflow_context *subflow,
535 struct sk_buff *skb, struct mptcp_ext *ext)
536 {
537 /* The write_seq value has already been incremented, so the actual
538 * sequence number for the DATA_FIN is one less.
539 */
540 u64 data_fin_tx_seq = READ_ONCE(mptcp_sk(subflow->conn)->write_seq) - 1;
541
542 if (!ext->use_map || !skb->len) {
543 /* RFC6824 requires a DSS mapping with specific values
544 * if DATA_FIN is set but no data payload is mapped
545 */
546 ext->data_fin = 1;
547 ext->use_map = 1;
548 ext->dsn64 = 1;
549 ext->data_seq = data_fin_tx_seq;
550 ext->subflow_seq = 0;
551 ext->data_len = 1;
552 } else if (ext->data_seq + ext->data_len == data_fin_tx_seq) {
553 /* If there's an existing DSS mapping and it is the
554 * final mapping, DATA_FIN consumes 1 additional byte of
555 * mapping space.
556 */
557 ext->data_fin = 1;
558 ext->data_len++;
559 }
560 }
561
mptcp_established_options_dss(struct sock * sk,struct sk_buff * skb,bool snd_data_fin_enable,unsigned int * size,struct mptcp_out_options * opts)562 static bool mptcp_established_options_dss(struct sock *sk, struct sk_buff *skb,
563 bool snd_data_fin_enable,
564 unsigned int *size,
565 struct mptcp_out_options *opts)
566 {
567 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
568 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
569 unsigned int dss_size = 0;
570 struct mptcp_ext *mpext;
571 unsigned int ack_size;
572 bool ret = false;
573 u64 ack_seq;
574
575 opts->csum_reqd = READ_ONCE(msk->csum_enabled);
576 mpext = skb ? mptcp_get_ext(skb) : NULL;
577
578 if (!skb || (mpext && mpext->use_map) || snd_data_fin_enable) {
579 unsigned int map_size = TCPOLEN_MPTCP_DSS_BASE + TCPOLEN_MPTCP_DSS_MAP64;
580
581 if (mpext) {
582 if (opts->csum_reqd)
583 map_size += TCPOLEN_MPTCP_DSS_CHECKSUM;
584
585 opts->ext_copy = *mpext;
586 }
587
588 dss_size = map_size;
589 if (skb && snd_data_fin_enable)
590 mptcp_write_data_fin(subflow, skb, &opts->ext_copy);
591 opts->suboptions = OPTION_MPTCP_DSS;
592 ret = true;
593 }
594
595 /* passive sockets msk will set the 'can_ack' after accept(), even
596 * if the first subflow may have the already the remote key handy
597 */
598 opts->ext_copy.use_ack = 0;
599 if (!READ_ONCE(msk->can_ack)) {
600 *size = ALIGN(dss_size, 4);
601 return ret;
602 }
603
604 ack_seq = READ_ONCE(msk->ack_seq);
605 if (READ_ONCE(msk->use_64bit_ack)) {
606 ack_size = TCPOLEN_MPTCP_DSS_ACK64;
607 opts->ext_copy.data_ack = ack_seq;
608 opts->ext_copy.ack64 = 1;
609 } else {
610 ack_size = TCPOLEN_MPTCP_DSS_ACK32;
611 opts->ext_copy.data_ack32 = (uint32_t)ack_seq;
612 opts->ext_copy.ack64 = 0;
613 }
614 opts->ext_copy.use_ack = 1;
615 opts->suboptions = OPTION_MPTCP_DSS;
616
617 /* Add kind/length/subtype/flag overhead if mapping is not populated */
618 if (dss_size == 0)
619 ack_size += TCPOLEN_MPTCP_DSS_BASE;
620
621 dss_size += ack_size;
622
623 *size = ALIGN(dss_size, 4);
624 return true;
625 }
626
add_addr_generate_hmac(u64 key1,u64 key2,struct mptcp_addr_info * addr)627 static u64 add_addr_generate_hmac(u64 key1, u64 key2,
628 struct mptcp_addr_info *addr)
629 {
630 u16 port = ntohs(addr->port);
631 u8 hmac[SHA256_DIGEST_SIZE];
632 u8 msg[19];
633 int i = 0;
634
635 msg[i++] = addr->id;
636 if (addr->family == AF_INET) {
637 memcpy(&msg[i], &addr->addr.s_addr, 4);
638 i += 4;
639 }
640 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
641 else if (addr->family == AF_INET6) {
642 memcpy(&msg[i], &addr->addr6.s6_addr, 16);
643 i += 16;
644 }
645 #endif
646 msg[i++] = port >> 8;
647 msg[i++] = port & 0xFF;
648
649 mptcp_crypto_hmac_sha(key1, key2, msg, i, hmac);
650
651 return get_unaligned_be64(&hmac[SHA256_DIGEST_SIZE - sizeof(u64)]);
652 }
653
mptcp_established_options_add_addr(struct sock * sk,struct sk_buff * skb,unsigned int * size,unsigned int remaining,struct mptcp_out_options * opts)654 static bool mptcp_established_options_add_addr(struct sock *sk, struct sk_buff *skb,
655 unsigned int *size,
656 unsigned int remaining,
657 struct mptcp_out_options *opts)
658 {
659 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
660 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
661 bool drop_other_suboptions = false;
662 unsigned int opt_size = *size;
663 struct mptcp_addr_info addr;
664 bool echo;
665 int len;
666
667 /* add addr will strip the existing options, be sure to avoid breaking
668 * MPC/MPJ handshakes
669 */
670 if (!mptcp_pm_should_add_signal(msk) ||
671 (opts->suboptions & (OPTION_MPTCP_MPJ_ACK | OPTION_MPTCP_MPC_ACK)) ||
672 !mptcp_pm_add_addr_signal(msk, skb, opt_size, remaining, &addr,
673 &echo, &drop_other_suboptions))
674 return false;
675
676 /*
677 * Later on, mptcp_write_options() will enforce mutually exclusion with
678 * DSS, bail out if such option is set and we can't drop it.
679 */
680 if (drop_other_suboptions)
681 remaining += opt_size;
682 else if (opts->suboptions & OPTION_MPTCP_DSS)
683 return false;
684
685 len = mptcp_add_addr_len(addr.family, echo, !!addr.port);
686 if (remaining < len)
687 return false;
688
689 *size = len;
690 if (drop_other_suboptions) {
691 pr_debug("drop other suboptions\n");
692 opts->suboptions = 0;
693
694 /* note that e.g. DSS could have written into the memory
695 * aliased by ahmac, we must reset the field here
696 * to avoid appending the hmac even for ADD_ADDR echo
697 * options
698 */
699 opts->ahmac = 0;
700 *size -= opt_size;
701 }
702 opts->addr = addr;
703 opts->suboptions |= OPTION_MPTCP_ADD_ADDR;
704 if (!echo) {
705 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ADDADDRTX);
706 opts->ahmac = add_addr_generate_hmac(READ_ONCE(msk->local_key),
707 READ_ONCE(msk->remote_key),
708 &opts->addr);
709 } else {
710 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ECHOADDTX);
711 }
712 pr_debug("addr_id=%d, ahmac=%llu, echo=%d, port=%d\n",
713 opts->addr.id, opts->ahmac, echo, ntohs(opts->addr.port));
714
715 return true;
716 }
717
mptcp_established_options_rm_addr(struct sock * sk,unsigned int * size,unsigned int remaining,struct mptcp_out_options * opts)718 static bool mptcp_established_options_rm_addr(struct sock *sk,
719 unsigned int *size,
720 unsigned int remaining,
721 struct mptcp_out_options *opts)
722 {
723 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
724 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
725 struct mptcp_rm_list rm_list;
726 int i, len;
727
728 if (!mptcp_pm_should_rm_signal(msk) ||
729 !(mptcp_pm_rm_addr_signal(msk, remaining, &rm_list)))
730 return false;
731
732 len = mptcp_rm_addr_len(&rm_list);
733 if (len < 0)
734 return false;
735 if (remaining < len)
736 return false;
737
738 *size = len;
739 opts->suboptions |= OPTION_MPTCP_RM_ADDR;
740 opts->rm_list = rm_list;
741
742 for (i = 0; i < opts->rm_list.nr; i++)
743 pr_debug("rm_list_ids[%d]=%d\n", i, opts->rm_list.ids[i]);
744 MPTCP_ADD_STATS(sock_net(sk), MPTCP_MIB_RMADDRTX, opts->rm_list.nr);
745 return true;
746 }
747
mptcp_established_options_mp_prio(struct sock * sk,unsigned int * size,unsigned int remaining,struct mptcp_out_options * opts)748 static bool mptcp_established_options_mp_prio(struct sock *sk,
749 unsigned int *size,
750 unsigned int remaining,
751 struct mptcp_out_options *opts)
752 {
753 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
754
755 /* can't send MP_PRIO with MPC, as they share the same option space:
756 * 'backup'. Also it makes no sense at all
757 */
758 if (!subflow->send_mp_prio || (opts->suboptions & OPTIONS_MPTCP_MPC))
759 return false;
760
761 /* account for the trailing 'nop' option */
762 if (remaining < TCPOLEN_MPTCP_PRIO_ALIGN)
763 return false;
764
765 *size = TCPOLEN_MPTCP_PRIO_ALIGN;
766 opts->suboptions |= OPTION_MPTCP_PRIO;
767 opts->backup = subflow->request_bkup;
768
769 pr_debug("prio=%d\n", opts->backup);
770
771 return true;
772 }
773
mptcp_established_options_rst(struct sock * sk,struct sk_buff * skb,unsigned int * size,unsigned int remaining,struct mptcp_out_options * opts)774 static noinline bool mptcp_established_options_rst(struct sock *sk, struct sk_buff *skb,
775 unsigned int *size,
776 unsigned int remaining,
777 struct mptcp_out_options *opts)
778 {
779 const struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
780
781 if (remaining < TCPOLEN_MPTCP_RST)
782 return false;
783
784 *size = TCPOLEN_MPTCP_RST;
785 opts->suboptions |= OPTION_MPTCP_RST;
786 opts->reset_transient = subflow->reset_transient;
787 opts->reset_reason = subflow->reset_reason;
788 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPRSTTX);
789
790 return true;
791 }
792
mptcp_established_options_fastclose(struct sock * sk,unsigned int * size,unsigned int remaining,struct mptcp_out_options * opts)793 static bool mptcp_established_options_fastclose(struct sock *sk,
794 unsigned int *size,
795 unsigned int remaining,
796 struct mptcp_out_options *opts)
797 {
798 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
799 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
800
801 if (likely(!subflow->send_fastclose))
802 return false;
803
804 if (remaining < TCPOLEN_MPTCP_FASTCLOSE)
805 return false;
806
807 *size = TCPOLEN_MPTCP_FASTCLOSE;
808 opts->suboptions |= OPTION_MPTCP_FASTCLOSE;
809 opts->rcvr_key = READ_ONCE(msk->remote_key);
810
811 pr_debug("FASTCLOSE key=%llu\n", opts->rcvr_key);
812 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPFASTCLOSETX);
813 return true;
814 }
815
mptcp_established_options_mp_fail(struct sock * sk,unsigned int * size,unsigned int remaining,struct mptcp_out_options * opts)816 static bool mptcp_established_options_mp_fail(struct sock *sk,
817 unsigned int *size,
818 unsigned int remaining,
819 struct mptcp_out_options *opts)
820 {
821 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
822
823 if (likely(!subflow->send_mp_fail))
824 return false;
825
826 if (remaining < TCPOLEN_MPTCP_FAIL)
827 return false;
828
829 *size = TCPOLEN_MPTCP_FAIL;
830 opts->suboptions |= OPTION_MPTCP_FAIL;
831 opts->fail_seq = subflow->map_seq;
832
833 pr_debug("MP_FAIL fail_seq=%llu\n", opts->fail_seq);
834 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPFAILTX);
835
836 return true;
837 }
838
mptcp_established_options(struct sock * sk,struct sk_buff * skb,unsigned int * size,unsigned int remaining,struct mptcp_out_options * opts)839 bool mptcp_established_options(struct sock *sk, struct sk_buff *skb,
840 unsigned int *size, unsigned int remaining,
841 struct mptcp_out_options *opts)
842 {
843 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
844 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
845 unsigned int opt_size = 0;
846 bool snd_data_fin;
847 bool ret = false;
848
849 opts->suboptions = 0;
850
851 /* Force later mptcp_write_options(), but do not use any actual
852 * option space.
853 */
854 if (unlikely(__mptcp_check_fallback(msk) && !mptcp_check_infinite_map(skb)))
855 return true;
856
857 if (unlikely(skb && TCP_SKB_CB(skb)->tcp_flags & TCPHDR_RST)) {
858 if (mptcp_established_options_fastclose(sk, &opt_size, remaining, opts) ||
859 mptcp_established_options_mp_fail(sk, &opt_size, remaining, opts)) {
860 *size += opt_size;
861 remaining -= opt_size;
862 }
863 /* MP_RST can be used with MP_FASTCLOSE and MP_FAIL if there is room */
864 if (mptcp_established_options_rst(sk, skb, &opt_size, remaining, opts)) {
865 *size += opt_size;
866 remaining -= opt_size;
867 }
868 return true;
869 }
870
871 snd_data_fin = mptcp_data_fin_enabled(msk);
872 if (mptcp_established_options_mp(sk, skb, snd_data_fin, &opt_size, opts))
873 ret = true;
874 else if (mptcp_established_options_dss(sk, skb, snd_data_fin, &opt_size, opts)) {
875 unsigned int mp_fail_size;
876
877 ret = true;
878 if (mptcp_established_options_mp_fail(sk, &mp_fail_size,
879 remaining - opt_size, opts)) {
880 *size += opt_size + mp_fail_size;
881 remaining -= opt_size - mp_fail_size;
882 return true;
883 }
884 }
885
886 /* we reserved enough space for the above options, and exceeding the
887 * TCP option space would be fatal
888 */
889 if (WARN_ON_ONCE(opt_size > remaining))
890 return false;
891
892 *size += opt_size;
893 remaining -= opt_size;
894 if (mptcp_established_options_add_addr(sk, skb, &opt_size, remaining, opts)) {
895 *size += opt_size;
896 remaining -= opt_size;
897 ret = true;
898 } else if (mptcp_established_options_rm_addr(sk, &opt_size, remaining, opts)) {
899 *size += opt_size;
900 remaining -= opt_size;
901 ret = true;
902 }
903
904 if (mptcp_established_options_mp_prio(sk, &opt_size, remaining, opts)) {
905 *size += opt_size;
906 remaining -= opt_size;
907 ret = true;
908 }
909
910 return ret;
911 }
912
mptcp_synack_options(const struct request_sock * req,unsigned int * size,struct mptcp_out_options * opts)913 bool mptcp_synack_options(const struct request_sock *req, unsigned int *size,
914 struct mptcp_out_options *opts)
915 {
916 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
917
918 if (subflow_req->mp_capable) {
919 opts->suboptions = OPTION_MPTCP_MPC_SYNACK;
920 opts->sndr_key = subflow_req->local_key;
921 opts->csum_reqd = subflow_req->csum_reqd;
922 opts->allow_join_id0 = subflow_req->allow_join_id0;
923 *size = TCPOLEN_MPTCP_MPC_SYNACK;
924 pr_debug("subflow_req=%p, local_key=%llu\n",
925 subflow_req, subflow_req->local_key);
926 return true;
927 } else if (subflow_req->mp_join) {
928 opts->suboptions = OPTION_MPTCP_MPJ_SYNACK;
929 opts->backup = subflow_req->request_bkup;
930 opts->join_id = subflow_req->local_id;
931 opts->thmac = subflow_req->thmac;
932 opts->nonce = subflow_req->local_nonce;
933 pr_debug("req=%p, bkup=%u, id=%u, thmac=%llu, nonce=%u\n",
934 subflow_req, opts->backup, opts->join_id,
935 opts->thmac, opts->nonce);
936 *size = TCPOLEN_MPTCP_MPJ_SYNACK;
937 return true;
938 }
939 return false;
940 }
941
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)942 static bool check_fully_established(struct mptcp_sock *msk, struct sock *ssk,
943 struct mptcp_subflow_context *subflow,
944 struct sk_buff *skb,
945 struct mptcp_options_received *mp_opt)
946 {
947 /* here we can process OoO, in-window pkts, only in-sequence 4th ack
948 * will make the subflow fully established
949 */
950 if (likely(READ_ONCE(subflow->fully_established))) {
951 /* on passive sockets, check for 3rd ack retransmission
952 * note that msk is always set by subflow_syn_recv_sock()
953 * for mp_join subflows
954 */
955 if (TCP_SKB_CB(skb)->seq == subflow->ssn_offset + 1 &&
956 TCP_SKB_CB(skb)->end_seq == TCP_SKB_CB(skb)->seq &&
957 subflow->mp_join && (mp_opt->suboptions & OPTIONS_MPTCP_MPJ) &&
958 !subflow->request_join)
959 tcp_send_ack(ssk);
960 goto check_notify;
961 }
962
963 /* we must process OoO packets before the first subflow is fully
964 * established. OoO packets are instead a protocol violation
965 * for MP_JOIN subflows as the peer must not send any data
966 * before receiving the forth ack - cfr. RFC 8684 section 3.2.
967 */
968 if (TCP_SKB_CB(skb)->seq != subflow->ssn_offset + 1) {
969 if (subflow->mp_join)
970 goto reset;
971 if (subflow->is_mptfo && mp_opt->suboptions & OPTION_MPTCP_MPC_ACK)
972 goto set_fully_established;
973 return subflow->mp_capable;
974 }
975
976 if (subflow->remote_key_valid &&
977 (((mp_opt->suboptions & OPTION_MPTCP_DSS) && mp_opt->use_ack) ||
978 ((mp_opt->suboptions & OPTION_MPTCP_ADD_ADDR) &&
979 (!mp_opt->echo || subflow->mp_join)))) {
980 /* subflows are fully established as soon as we get any
981 * additional ack, including ADD_ADDR.
982 */
983 goto set_fully_established;
984 }
985
986 /* If the first established packet does not contain MP_CAPABLE + data
987 * then fallback to TCP. Fallback scenarios requires a reset for
988 * MP_JOIN subflows.
989 */
990 if (!(mp_opt->suboptions & OPTIONS_MPTCP_MPC)) {
991 if (subflow->mp_join)
992 goto reset;
993 subflow->mp_capable = 0;
994 if (!mptcp_try_fallback(ssk, MPTCP_MIB_MPCAPABLEDATAFALLBACK)) {
995 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_FALLBACKFAILED);
996 goto reset;
997 }
998 return false;
999 }
1000
1001 if (unlikely(!READ_ONCE(msk->pm.server_side)))
1002 pr_warn_once("bogus mpc option on established client sk");
1003
1004 set_fully_established:
1005 if (mp_opt->deny_join_id0)
1006 WRITE_ONCE(msk->pm.remote_deny_join_id0, true);
1007
1008 mptcp_data_lock((struct sock *)msk);
1009 __mptcp_subflow_fully_established(msk, subflow, mp_opt);
1010 mptcp_data_unlock((struct sock *)msk);
1011
1012 check_notify:
1013 /* if the subflow is not already linked into the conn_list, we can't
1014 * notify the PM: this subflow is still on the listener queue
1015 * and the PM possibly acquiring the subflow lock could race with
1016 * the listener close
1017 */
1018 if (likely(subflow->pm_notified) || list_empty(&subflow->node))
1019 return true;
1020
1021 subflow->pm_notified = 1;
1022 if (subflow->mp_join) {
1023 clear_3rdack_retransmission(ssk);
1024 mptcp_pm_subflow_established(msk);
1025 } else {
1026 mptcp_pm_fully_established(msk, ssk);
1027 }
1028 return true;
1029
1030 reset:
1031 mptcp_subflow_reset(ssk);
1032 return false;
1033 }
1034
__mptcp_expand_seq(u64 old_seq,u64 cur_seq)1035 u64 __mptcp_expand_seq(u64 old_seq, u64 cur_seq)
1036 {
1037 u32 old_seq32, cur_seq32;
1038
1039 old_seq32 = (u32)old_seq;
1040 cur_seq32 = (u32)cur_seq;
1041 cur_seq = (old_seq & GENMASK_ULL(63, 32)) + cur_seq32;
1042 if (unlikely(cur_seq32 < old_seq32 && before(old_seq32, cur_seq32)))
1043 return cur_seq + (1LL << 32);
1044
1045 /* reverse wrap could happen, too */
1046 if (unlikely(cur_seq32 > old_seq32 && after(old_seq32, cur_seq32)))
1047 return cur_seq - (1LL << 32);
1048 return cur_seq;
1049 }
1050
__mptcp_snd_una_update(struct mptcp_sock * msk,u64 new_snd_una)1051 static void __mptcp_snd_una_update(struct mptcp_sock *msk, u64 new_snd_una)
1052 {
1053 msk->bytes_acked += new_snd_una - msk->snd_una;
1054 WRITE_ONCE(msk->snd_una, new_snd_una);
1055 }
1056
rwin_update(struct mptcp_sock * msk,struct sock * ssk,struct sk_buff * skb)1057 static void rwin_update(struct mptcp_sock *msk, struct sock *ssk,
1058 struct sk_buff *skb)
1059 {
1060 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1061 struct tcp_sock *tp = tcp_sk(ssk);
1062 u64 mptcp_rcv_wnd;
1063
1064 /* Avoid touching extra cachelines if TCP is going to accept this
1065 * skb without filling the TCP-level window even with a possibly
1066 * outdated mptcp-level rwin.
1067 */
1068 if (!skb->len || skb->len < tcp_receive_window(tp))
1069 return;
1070
1071 mptcp_rcv_wnd = atomic64_read(&msk->rcv_wnd_sent);
1072 if (!after64(mptcp_rcv_wnd, subflow->rcv_wnd_sent))
1073 return;
1074
1075 /* Some other subflow grew the mptcp-level rwin since rcv_wup,
1076 * resync.
1077 */
1078 tp->rcv_wnd += mptcp_rcv_wnd - subflow->rcv_wnd_sent;
1079 subflow->rcv_wnd_sent = mptcp_rcv_wnd;
1080 }
1081
ack_update_msk(struct mptcp_sock * msk,struct sock * ssk,struct mptcp_options_received * mp_opt)1082 static void ack_update_msk(struct mptcp_sock *msk,
1083 struct sock *ssk,
1084 struct mptcp_options_received *mp_opt)
1085 {
1086 u64 new_wnd_end, new_snd_una, snd_nxt = READ_ONCE(msk->snd_nxt);
1087 struct sock *sk = (struct sock *)msk;
1088 u64 old_snd_una;
1089
1090 mptcp_data_lock(sk);
1091
1092 /* avoid ack expansion on update conflict, to reduce the risk of
1093 * wrongly expanding to a future ack sequence number, which is way
1094 * more dangerous than missing an ack
1095 */
1096 old_snd_una = msk->snd_una;
1097 new_snd_una = mptcp_expand_seq(old_snd_una, mp_opt->data_ack, mp_opt->ack64);
1098
1099 /* ACK for data not even sent yet? Ignore.*/
1100 if (unlikely(after64(new_snd_una, snd_nxt)))
1101 new_snd_una = old_snd_una;
1102
1103 new_wnd_end = new_snd_una + tcp_sk(ssk)->snd_wnd;
1104
1105 if (after64(new_wnd_end, msk->wnd_end))
1106 WRITE_ONCE(msk->wnd_end, new_wnd_end);
1107
1108 /* this assumes mptcp_incoming_options() is invoked after tcp_ack() */
1109 if (after64(msk->wnd_end, snd_nxt))
1110 __mptcp_check_push(sk, ssk);
1111
1112 if (after64(new_snd_una, old_snd_una)) {
1113 __mptcp_snd_una_update(msk, new_snd_una);
1114 __mptcp_data_acked(sk);
1115 }
1116 msk->last_ack_recv = tcp_jiffies32;
1117 mptcp_data_unlock(sk);
1118
1119 trace_ack_update_msk(mp_opt->data_ack,
1120 old_snd_una, new_snd_una,
1121 new_wnd_end, READ_ONCE(msk->wnd_end));
1122 }
1123
mptcp_update_rcv_data_fin(struct mptcp_sock * msk,u64 data_fin_seq,bool use_64bit)1124 bool mptcp_update_rcv_data_fin(struct mptcp_sock *msk, u64 data_fin_seq, bool use_64bit)
1125 {
1126 /* Skip if DATA_FIN was already received.
1127 * If updating simultaneously with the recvmsg loop, values
1128 * should match. If they mismatch, the peer is misbehaving and
1129 * we will prefer the most recent information.
1130 */
1131 if (READ_ONCE(msk->rcv_data_fin))
1132 return false;
1133
1134 WRITE_ONCE(msk->rcv_data_fin_seq,
1135 mptcp_expand_seq(READ_ONCE(msk->ack_seq), data_fin_seq, use_64bit));
1136 WRITE_ONCE(msk->rcv_data_fin, 1);
1137
1138 return true;
1139 }
1140
add_addr_hmac_valid(struct mptcp_sock * msk,struct mptcp_options_received * mp_opt)1141 static bool add_addr_hmac_valid(struct mptcp_sock *msk,
1142 struct mptcp_options_received *mp_opt)
1143 {
1144 u64 hmac = 0;
1145
1146 if (mp_opt->echo)
1147 return true;
1148
1149 hmac = add_addr_generate_hmac(READ_ONCE(msk->remote_key),
1150 READ_ONCE(msk->local_key),
1151 &mp_opt->addr);
1152
1153 pr_debug("msk=%p, ahmac=%llu, mp_opt->ahmac=%llu\n",
1154 msk, hmac, mp_opt->ahmac);
1155
1156 return hmac == mp_opt->ahmac;
1157 }
1158
1159 /* Return false in case of error (or subflow has been reset),
1160 * else return true.
1161 */
mptcp_incoming_options(struct sock * sk,struct sk_buff * skb)1162 bool mptcp_incoming_options(struct sock *sk, struct sk_buff *skb)
1163 {
1164 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1165 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
1166 struct mptcp_options_received mp_opt;
1167 struct mptcp_ext *mpext;
1168
1169 if (__mptcp_check_fallback(msk)) {
1170 /* Keep it simple and unconditionally trigger send data cleanup and
1171 * pending queue spooling. We will need to acquire the data lock
1172 * for more accurate checks, and once the lock is acquired, such
1173 * helpers are cheap.
1174 */
1175 mptcp_data_lock(subflow->conn);
1176 if (sk_stream_memory_free(sk))
1177 __mptcp_check_push(subflow->conn, sk);
1178
1179 /* on fallback we just need to ignore the msk-level snd_una, as
1180 * this is really plain TCP
1181 */
1182 __mptcp_snd_una_update(msk, READ_ONCE(msk->snd_nxt));
1183
1184 __mptcp_data_acked(subflow->conn);
1185 mptcp_data_unlock(subflow->conn);
1186 return true;
1187 }
1188
1189 mptcp_get_options(skb, &mp_opt);
1190
1191 /* The subflow can be in close state only if check_fully_established()
1192 * just sent a reset. If so, tell the caller to ignore the current packet.
1193 */
1194 if (!check_fully_established(msk, sk, subflow, skb, &mp_opt))
1195 return sk->sk_state != TCP_CLOSE;
1196
1197 if (unlikely(mp_opt.suboptions != OPTION_MPTCP_DSS)) {
1198 if ((mp_opt.suboptions & OPTION_MPTCP_FASTCLOSE) &&
1199 READ_ONCE(msk->local_key) == mp_opt.rcvr_key) {
1200 WRITE_ONCE(msk->rcv_fastclose, true);
1201 mptcp_schedule_work((struct sock *)msk);
1202 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPFASTCLOSERX);
1203 }
1204
1205 if ((mp_opt.suboptions & OPTION_MPTCP_ADD_ADDR) &&
1206 add_addr_hmac_valid(msk, &mp_opt)) {
1207 if (!mp_opt.echo) {
1208 mptcp_pm_add_addr_received(sk, &mp_opt.addr);
1209 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ADDADDR);
1210 } else {
1211 mptcp_pm_add_addr_echoed(msk, &mp_opt.addr);
1212 mptcp_pm_del_add_timer(msk, &mp_opt.addr, true);
1213 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ECHOADD);
1214 }
1215
1216 if (mp_opt.addr.port)
1217 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_PORTADD);
1218 }
1219
1220 if (mp_opt.suboptions & OPTION_MPTCP_RM_ADDR)
1221 mptcp_pm_rm_addr_received(msk, &mp_opt.rm_list);
1222
1223 if (mp_opt.suboptions & OPTION_MPTCP_PRIO) {
1224 mptcp_pm_mp_prio_received(sk, mp_opt.backup);
1225 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPPRIORX);
1226 }
1227
1228 if (mp_opt.suboptions & OPTION_MPTCP_FAIL) {
1229 mptcp_pm_mp_fail_received(sk, mp_opt.fail_seq);
1230 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPFAILRX);
1231 }
1232
1233 if (mp_opt.suboptions & OPTION_MPTCP_RST) {
1234 subflow->reset_seen = 1;
1235 subflow->reset_reason = mp_opt.reset_reason;
1236 subflow->reset_transient = mp_opt.reset_transient;
1237 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPRSTRX);
1238 }
1239
1240 if (!(mp_opt.suboptions & OPTION_MPTCP_DSS))
1241 return true;
1242 }
1243
1244 /* we can't wait for recvmsg() to update the ack_seq, otherwise
1245 * monodirectional flows will stuck
1246 */
1247 if (mp_opt.use_ack)
1248 ack_update_msk(msk, sk, &mp_opt);
1249 rwin_update(msk, sk, skb);
1250
1251 /* Zero-data-length packets are dropped by the caller and not
1252 * propagated to the MPTCP layer, so the skb extension does not
1253 * need to be allocated or populated. DATA_FIN information, if
1254 * present, needs to be updated here before the skb is freed.
1255 */
1256 if (TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb)->end_seq) {
1257 if (mp_opt.data_fin && mp_opt.data_len == 1 &&
1258 mptcp_update_rcv_data_fin(msk, mp_opt.data_seq, mp_opt.dsn64))
1259 mptcp_schedule_work((struct sock *)msk);
1260
1261 return true;
1262 }
1263
1264 mpext = skb_ext_add(skb, SKB_EXT_MPTCP);
1265 if (!mpext)
1266 return false;
1267
1268 memset(mpext, 0, sizeof(*mpext));
1269
1270 if (likely(mp_opt.use_map)) {
1271 if (mp_opt.mpc_map) {
1272 /* this is an MP_CAPABLE carrying MPTCP data
1273 * we know this map the first chunk of data
1274 */
1275 mptcp_crypto_key_sha(subflow->remote_key, NULL,
1276 &mpext->data_seq);
1277 mpext->data_seq++;
1278 mpext->subflow_seq = 1;
1279 mpext->dsn64 = 1;
1280 mpext->mpc_map = 1;
1281 mpext->data_fin = 0;
1282 } else {
1283 mpext->data_seq = mp_opt.data_seq;
1284 mpext->subflow_seq = mp_opt.subflow_seq;
1285 mpext->dsn64 = mp_opt.dsn64;
1286 mpext->data_fin = mp_opt.data_fin;
1287 }
1288 mpext->data_len = mp_opt.data_len;
1289 mpext->use_map = 1;
1290 mpext->csum_reqd = !!(mp_opt.suboptions & OPTION_MPTCP_CSUMREQD);
1291
1292 if (mpext->csum_reqd)
1293 mpext->csum = mp_opt.csum;
1294 }
1295
1296 return true;
1297 }
1298
mptcp_set_rwin(struct tcp_sock * tp,struct tcphdr * th)1299 static void mptcp_set_rwin(struct tcp_sock *tp, struct tcphdr *th)
1300 {
1301 const struct sock *ssk = (const struct sock *)tp;
1302 struct mptcp_subflow_context *subflow;
1303 u64 ack_seq, rcv_wnd_old, rcv_wnd_new;
1304 struct mptcp_sock *msk;
1305 u32 new_win;
1306 u64 win;
1307
1308 subflow = mptcp_subflow_ctx(ssk);
1309 msk = mptcp_sk(subflow->conn);
1310
1311 ack_seq = READ_ONCE(msk->ack_seq);
1312 rcv_wnd_new = ack_seq + tp->rcv_wnd;
1313
1314 rcv_wnd_old = atomic64_read(&msk->rcv_wnd_sent);
1315 if (after64(rcv_wnd_new, rcv_wnd_old)) {
1316 u64 rcv_wnd;
1317
1318 for (;;) {
1319 rcv_wnd = atomic64_cmpxchg(&msk->rcv_wnd_sent, rcv_wnd_old, rcv_wnd_new);
1320
1321 if (rcv_wnd == rcv_wnd_old)
1322 break;
1323
1324 rcv_wnd_old = rcv_wnd;
1325 if (before64(rcv_wnd_new, rcv_wnd_old)) {
1326 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_RCVWNDCONFLICTUPDATE);
1327 goto raise_win;
1328 }
1329 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_RCVWNDCONFLICT);
1330 }
1331 goto update_wspace;
1332 }
1333
1334 if (rcv_wnd_new != rcv_wnd_old) {
1335 raise_win:
1336 /* The msk-level rcv wnd is after the tcp level one,
1337 * sync the latter.
1338 */
1339 rcv_wnd_new = rcv_wnd_old;
1340 win = rcv_wnd_old - ack_seq;
1341 tp->rcv_wnd = min_t(u64, win, U32_MAX);
1342 new_win = tp->rcv_wnd;
1343
1344 /* Make sure we do not exceed the maximum possible
1345 * scaled window.
1346 */
1347 if (unlikely(th->syn))
1348 new_win = min(new_win, 65535U) << tp->rx_opt.rcv_wscale;
1349 if (!tp->rx_opt.rcv_wscale &&
1350 READ_ONCE(sock_net(ssk)->ipv4.sysctl_tcp_workaround_signed_windows))
1351 new_win = min(new_win, MAX_TCP_WINDOW);
1352 else
1353 new_win = min(new_win, (65535U << tp->rx_opt.rcv_wscale));
1354
1355 /* RFC1323 scaling applied */
1356 new_win >>= tp->rx_opt.rcv_wscale;
1357 th->window = htons(new_win);
1358 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_RCVWNDSHARED);
1359 }
1360
1361 update_wspace:
1362 WRITE_ONCE(msk->old_wspace, tp->rcv_wnd);
1363 subflow->rcv_wnd_sent = rcv_wnd_new;
1364 }
1365
mptcp_track_rwin(struct tcp_sock * tp)1366 static void mptcp_track_rwin(struct tcp_sock *tp)
1367 {
1368 const struct sock *ssk = (const struct sock *)tp;
1369 struct mptcp_subflow_context *subflow;
1370 struct mptcp_sock *msk;
1371
1372 if (!ssk)
1373 return;
1374
1375 subflow = mptcp_subflow_ctx(ssk);
1376 msk = mptcp_sk(subflow->conn);
1377 WRITE_ONCE(msk->old_wspace, tp->rcv_wnd);
1378 }
1379
__mptcp_make_csum(u64 data_seq,u32 subflow_seq,u16 data_len,__wsum sum)1380 __sum16 __mptcp_make_csum(u64 data_seq, u32 subflow_seq, u16 data_len, __wsum sum)
1381 {
1382 struct csum_pseudo_header header;
1383 __wsum csum;
1384
1385 /* cfr RFC 8684 3.3.1.:
1386 * the data sequence number used in the pseudo-header is
1387 * always the 64-bit value, irrespective of what length is used in the
1388 * DSS option itself.
1389 */
1390 header.data_seq = cpu_to_be64(data_seq);
1391 header.subflow_seq = htonl(subflow_seq);
1392 header.data_len = htons(data_len);
1393 header.csum = 0;
1394
1395 csum = csum_partial(&header, sizeof(header), sum);
1396 return csum_fold(csum);
1397 }
1398
mptcp_make_csum(const struct mptcp_ext * mpext)1399 static __sum16 mptcp_make_csum(const struct mptcp_ext *mpext)
1400 {
1401 return __mptcp_make_csum(mpext->data_seq, mpext->subflow_seq, mpext->data_len,
1402 ~csum_unfold(mpext->csum));
1403 }
1404
put_len_csum(u16 len,__sum16 csum,void * data)1405 static void put_len_csum(u16 len, __sum16 csum, void *data)
1406 {
1407 __sum16 *sumptr = data + 2;
1408 __be16 *ptr = data;
1409
1410 put_unaligned_be16(len, ptr);
1411
1412 put_unaligned(csum, sumptr);
1413 }
1414
mptcp_write_options(struct tcphdr * th,__be32 * ptr,struct tcp_sock * tp,struct mptcp_out_options * opts)1415 void mptcp_write_options(struct tcphdr *th, __be32 *ptr, struct tcp_sock *tp,
1416 struct mptcp_out_options *opts)
1417 {
1418 const struct sock *ssk = (const struct sock *)tp;
1419 struct mptcp_subflow_context *subflow;
1420
1421 /* Which options can be used together?
1422 *
1423 * X: mutually exclusive
1424 * O: often used together
1425 * C: can be used together in some cases
1426 * P: could be used together but we prefer not to (optimisations)
1427 *
1428 * Opt: | MPC | MPJ | DSS | ADD | RM | PRIO | FAIL | FC |
1429 * ------|------|------|------|------|------|------|------|------|
1430 * MPC |------|------|------|------|------|------|------|------|
1431 * MPJ | X |------|------|------|------|------|------|------|
1432 * DSS | X | X |------|------|------|------|------|------|
1433 * ADD | X | X | P |------|------|------|------|------|
1434 * RM | C | C | C | P |------|------|------|------|
1435 * PRIO | X | C | C | C | C |------|------|------|
1436 * FAIL | X | X | C | X | X | X |------|------|
1437 * FC | X | X | X | X | X | X | X |------|
1438 * RST | X | X | X | X | X | X | O | O |
1439 * ------|------|------|------|------|------|------|------|------|
1440 *
1441 * The same applies in mptcp_established_options() function.
1442 */
1443 if (likely(OPTION_MPTCP_DSS & opts->suboptions)) {
1444 struct mptcp_ext *mpext = &opts->ext_copy;
1445 u8 len = TCPOLEN_MPTCP_DSS_BASE;
1446 u8 flags = 0;
1447
1448 if (mpext->use_ack) {
1449 flags = MPTCP_DSS_HAS_ACK;
1450 if (mpext->ack64) {
1451 len += TCPOLEN_MPTCP_DSS_ACK64;
1452 flags |= MPTCP_DSS_ACK64;
1453 } else {
1454 len += TCPOLEN_MPTCP_DSS_ACK32;
1455 }
1456 }
1457
1458 if (mpext->use_map) {
1459 len += TCPOLEN_MPTCP_DSS_MAP64;
1460
1461 /* Use only 64-bit mapping flags for now, add
1462 * support for optional 32-bit mappings later.
1463 */
1464 flags |= MPTCP_DSS_HAS_MAP | MPTCP_DSS_DSN64;
1465 if (mpext->data_fin)
1466 flags |= MPTCP_DSS_DATA_FIN;
1467
1468 if (opts->csum_reqd)
1469 len += TCPOLEN_MPTCP_DSS_CHECKSUM;
1470 }
1471
1472 *ptr++ = mptcp_option(MPTCPOPT_DSS, len, 0, flags);
1473
1474 if (mpext->use_ack) {
1475 if (mpext->ack64) {
1476 put_unaligned_be64(mpext->data_ack, ptr);
1477 ptr += 2;
1478 } else {
1479 put_unaligned_be32(mpext->data_ack32, ptr);
1480 ptr += 1;
1481 }
1482 }
1483
1484 if (mpext->use_map) {
1485 put_unaligned_be64(mpext->data_seq, ptr);
1486 ptr += 2;
1487 put_unaligned_be32(mpext->subflow_seq, ptr);
1488 ptr += 1;
1489 if (opts->csum_reqd) {
1490 /* data_len == 0 is reserved for the infinite mapping,
1491 * the checksum will also be set to 0.
1492 */
1493 put_len_csum(mpext->data_len,
1494 (mpext->data_len ? mptcp_make_csum(mpext) : 0),
1495 ptr);
1496 } else {
1497 put_unaligned_be32(mpext->data_len << 16 |
1498 TCPOPT_NOP << 8 | TCPOPT_NOP, ptr);
1499 }
1500 ptr += 1;
1501 }
1502
1503 /* We might need to add MP_FAIL options in rare cases */
1504 if (unlikely(OPTION_MPTCP_FAIL & opts->suboptions))
1505 goto mp_fail;
1506 } else if (OPTIONS_MPTCP_MPC & opts->suboptions) {
1507 u8 len, flag = MPTCP_CAP_HMAC_SHA256;
1508
1509 if (OPTION_MPTCP_MPC_SYN & opts->suboptions) {
1510 len = TCPOLEN_MPTCP_MPC_SYN;
1511 } else if (OPTION_MPTCP_MPC_SYNACK & opts->suboptions) {
1512 len = TCPOLEN_MPTCP_MPC_SYNACK;
1513 } else if (opts->data_len) {
1514 len = TCPOLEN_MPTCP_MPC_ACK_DATA;
1515 if (opts->csum_reqd)
1516 len += TCPOLEN_MPTCP_DSS_CHECKSUM;
1517 } else {
1518 len = TCPOLEN_MPTCP_MPC_ACK;
1519 }
1520
1521 if (opts->csum_reqd)
1522 flag |= MPTCP_CAP_CHECKSUM_REQD;
1523
1524 if (!opts->allow_join_id0)
1525 flag |= MPTCP_CAP_DENY_JOIN_ID0;
1526
1527 *ptr++ = mptcp_option(MPTCPOPT_MP_CAPABLE, len,
1528 MPTCP_SUPPORTED_VERSION,
1529 flag);
1530
1531 if (!((OPTION_MPTCP_MPC_SYNACK | OPTION_MPTCP_MPC_ACK) &
1532 opts->suboptions))
1533 goto mp_capable_done;
1534
1535 put_unaligned_be64(opts->sndr_key, ptr);
1536 ptr += 2;
1537 if (!((OPTION_MPTCP_MPC_ACK) & opts->suboptions))
1538 goto mp_capable_done;
1539
1540 put_unaligned_be64(opts->rcvr_key, ptr);
1541 ptr += 2;
1542 if (!opts->data_len)
1543 goto mp_capable_done;
1544
1545 if (opts->csum_reqd) {
1546 put_len_csum(opts->data_len,
1547 __mptcp_make_csum(opts->data_seq,
1548 opts->subflow_seq,
1549 opts->data_len,
1550 ~csum_unfold(opts->csum)),
1551 ptr);
1552 } else {
1553 put_unaligned_be32(opts->data_len << 16 |
1554 TCPOPT_NOP << 8 | TCPOPT_NOP, ptr);
1555 }
1556 ptr += 1;
1557
1558 /* MPC is additionally mutually exclusive with MP_PRIO */
1559 goto mp_capable_done;
1560 } else if (OPTIONS_MPTCP_MPJ & opts->suboptions) {
1561 if (OPTION_MPTCP_MPJ_SYN & opts->suboptions) {
1562 *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1563 TCPOLEN_MPTCP_MPJ_SYN,
1564 opts->backup, opts->join_id);
1565 put_unaligned_be32(opts->token, ptr);
1566 ptr += 1;
1567 put_unaligned_be32(opts->nonce, ptr);
1568 ptr += 1;
1569 } else if (OPTION_MPTCP_MPJ_SYNACK & opts->suboptions) {
1570 *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1571 TCPOLEN_MPTCP_MPJ_SYNACK,
1572 opts->backup, opts->join_id);
1573 put_unaligned_be64(opts->thmac, ptr);
1574 ptr += 2;
1575 put_unaligned_be32(opts->nonce, ptr);
1576 ptr += 1;
1577 } else {
1578 *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1579 TCPOLEN_MPTCP_MPJ_ACK, 0, 0);
1580 memcpy(ptr, opts->hmac, MPTCPOPT_HMAC_LEN);
1581 ptr += 5;
1582 }
1583 } else if (OPTION_MPTCP_ADD_ADDR & opts->suboptions) {
1584 u8 len = TCPOLEN_MPTCP_ADD_ADDR_BASE;
1585 u8 echo = MPTCP_ADDR_ECHO;
1586
1587 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1588 if (opts->addr.family == AF_INET6)
1589 len = TCPOLEN_MPTCP_ADD_ADDR6_BASE;
1590 #endif
1591
1592 if (opts->addr.port)
1593 len += TCPOLEN_MPTCP_PORT_LEN;
1594
1595 if (opts->ahmac) {
1596 len += sizeof(opts->ahmac);
1597 echo = 0;
1598 }
1599
1600 *ptr++ = mptcp_option(MPTCPOPT_ADD_ADDR,
1601 len, echo, opts->addr.id);
1602 if (opts->addr.family == AF_INET) {
1603 memcpy((u8 *)ptr, (u8 *)&opts->addr.addr.s_addr, 4);
1604 ptr += 1;
1605 }
1606 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1607 else if (opts->addr.family == AF_INET6) {
1608 memcpy((u8 *)ptr, opts->addr.addr6.s6_addr, 16);
1609 ptr += 4;
1610 }
1611 #endif
1612
1613 if (!opts->addr.port) {
1614 if (opts->ahmac) {
1615 put_unaligned_be64(opts->ahmac, ptr);
1616 ptr += 2;
1617 }
1618 } else {
1619 u16 port = ntohs(opts->addr.port);
1620
1621 if (opts->ahmac) {
1622 u8 *bptr = (u8 *)ptr;
1623
1624 put_unaligned_be16(port, bptr);
1625 bptr += 2;
1626 put_unaligned_be64(opts->ahmac, bptr);
1627 bptr += 8;
1628 put_unaligned_be16(TCPOPT_NOP << 8 |
1629 TCPOPT_NOP, bptr);
1630
1631 ptr += 3;
1632 } else {
1633 put_unaligned_be32(port << 16 |
1634 TCPOPT_NOP << 8 |
1635 TCPOPT_NOP, ptr);
1636 ptr += 1;
1637 }
1638 }
1639 } else if (unlikely(OPTION_MPTCP_FASTCLOSE & opts->suboptions)) {
1640 /* FASTCLOSE is mutually exclusive with others except RST */
1641 *ptr++ = mptcp_option(MPTCPOPT_MP_FASTCLOSE,
1642 TCPOLEN_MPTCP_FASTCLOSE,
1643 0, 0);
1644 put_unaligned_be64(opts->rcvr_key, ptr);
1645 ptr += 2;
1646
1647 if (OPTION_MPTCP_RST & opts->suboptions)
1648 goto mp_rst;
1649 return;
1650 } else if (unlikely(OPTION_MPTCP_FAIL & opts->suboptions)) {
1651 mp_fail:
1652 /* MP_FAIL is mutually exclusive with others except RST */
1653 subflow = mptcp_subflow_ctx(ssk);
1654 subflow->send_mp_fail = 0;
1655
1656 *ptr++ = mptcp_option(MPTCPOPT_MP_FAIL,
1657 TCPOLEN_MPTCP_FAIL,
1658 0, 0);
1659 put_unaligned_be64(opts->fail_seq, ptr);
1660 ptr += 2;
1661
1662 if (OPTION_MPTCP_RST & opts->suboptions)
1663 goto mp_rst;
1664 return;
1665 } else if (unlikely(OPTION_MPTCP_RST & opts->suboptions)) {
1666 mp_rst:
1667 *ptr++ = mptcp_option(MPTCPOPT_RST,
1668 TCPOLEN_MPTCP_RST,
1669 opts->reset_transient,
1670 opts->reset_reason);
1671 return;
1672 } else if (unlikely(!opts->suboptions)) {
1673 /* Fallback to TCP */
1674 mptcp_track_rwin(tp);
1675 return;
1676 }
1677
1678 if (OPTION_MPTCP_PRIO & opts->suboptions) {
1679 subflow = mptcp_subflow_ctx(ssk);
1680 subflow->send_mp_prio = 0;
1681
1682 *ptr++ = mptcp_option(MPTCPOPT_MP_PRIO,
1683 TCPOLEN_MPTCP_PRIO,
1684 opts->backup, TCPOPT_NOP);
1685
1686 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_MPPRIOTX);
1687 }
1688
1689 mp_capable_done:
1690 if (OPTION_MPTCP_RM_ADDR & opts->suboptions) {
1691 u8 i = 1;
1692
1693 *ptr++ = mptcp_option(MPTCPOPT_RM_ADDR,
1694 TCPOLEN_MPTCP_RM_ADDR_BASE + opts->rm_list.nr,
1695 0, opts->rm_list.ids[0]);
1696
1697 while (i < opts->rm_list.nr) {
1698 u8 id1, id2, id3, id4;
1699
1700 id1 = opts->rm_list.ids[i];
1701 id2 = i + 1 < opts->rm_list.nr ? opts->rm_list.ids[i + 1] : TCPOPT_NOP;
1702 id3 = i + 2 < opts->rm_list.nr ? opts->rm_list.ids[i + 2] : TCPOPT_NOP;
1703 id4 = i + 3 < opts->rm_list.nr ? opts->rm_list.ids[i + 3] : TCPOPT_NOP;
1704 put_unaligned_be32(id1 << 24 | id2 << 16 | id3 << 8 | id4, ptr);
1705 ptr += 1;
1706 i += 4;
1707 }
1708 }
1709
1710 if (tp)
1711 mptcp_set_rwin(tp, th);
1712 }
1713
mptcp_get_reset_option(const struct sk_buff * skb)1714 __be32 mptcp_get_reset_option(const struct sk_buff *skb)
1715 {
1716 const struct mptcp_ext *ext = mptcp_get_ext(skb);
1717 u8 flags, reason;
1718
1719 if (ext) {
1720 flags = ext->reset_transient;
1721 reason = ext->reset_reason;
1722
1723 return mptcp_option(MPTCPOPT_RST, TCPOLEN_MPTCP_RST,
1724 flags, reason);
1725 }
1726
1727 return htonl(0u);
1728 }
1729 EXPORT_SYMBOL_GPL(mptcp_get_reset_option);
1730