1 // SPDX-License-Identifier: GPL-2.0
2 /* Multipath TCP
3 *
4 * Copyright (c) 2017 - 2019, Intel Corporation.
5 */
6
7 #define pr_fmt(fmt) "MPTCP: " fmt
8
9 #include <linux/kernel.h>
10 #include <crypto/sha2.h>
11 #include <net/tcp.h>
12 #include <net/mptcp.h>
13 #include "protocol.h"
14 #include "mib.h"
15
16 #include <trace/events/mptcp.h>
17
mptcp_cap_flag_sha256(u8 flags)18 static bool mptcp_cap_flag_sha256(u8 flags)
19 {
20 return (flags & MPTCP_CAP_FLAG_MASK) == MPTCP_CAP_HMAC_SHA256;
21 }
22
mptcp_parse_option(const struct sk_buff * skb,const unsigned char * ptr,int opsize,struct mptcp_options_received * mp_opt)23 static void mptcp_parse_option(const struct sk_buff *skb,
24 const unsigned char *ptr, int opsize,
25 struct mptcp_options_received *mp_opt)
26 {
27 u8 subtype = *ptr >> 4;
28 int expected_opsize;
29 u8 version;
30 u8 flags;
31 u8 i;
32
33 switch (subtype) {
34 case MPTCPOPT_MP_CAPABLE:
35 /* strict size checking */
36 if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)) {
37 if (skb->len > tcp_hdr(skb)->doff << 2)
38 expected_opsize = TCPOLEN_MPTCP_MPC_ACK_DATA;
39 else
40 expected_opsize = TCPOLEN_MPTCP_MPC_ACK;
41 } else {
42 if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_ACK)
43 expected_opsize = TCPOLEN_MPTCP_MPC_SYNACK;
44 else
45 expected_opsize = TCPOLEN_MPTCP_MPC_SYN;
46 }
47
48 /* Cfr RFC 8684 Section 3.3.0:
49 * If a checksum is present but its use had
50 * not been negotiated in the MP_CAPABLE handshake, the receiver MUST
51 * close the subflow with a RST, as it is not behaving as negotiated.
52 * If a checksum is not present when its use has been negotiated, the
53 * receiver MUST close the subflow with a RST, as it is considered
54 * broken
55 * We parse even option with mismatching csum presence, so that
56 * later in subflow_data_ready we can trigger the reset.
57 */
58 if (opsize != expected_opsize &&
59 (expected_opsize != TCPOLEN_MPTCP_MPC_ACK_DATA ||
60 opsize != TCPOLEN_MPTCP_MPC_ACK_DATA_CSUM))
61 break;
62
63 /* try to be gentle vs future versions on the initial syn */
64 version = *ptr++ & MPTCP_VERSION_MASK;
65 if (opsize != TCPOLEN_MPTCP_MPC_SYN) {
66 if (version != MPTCP_SUPPORTED_VERSION)
67 break;
68 } else if (version < MPTCP_SUPPORTED_VERSION) {
69 break;
70 }
71
72 flags = *ptr++;
73 if (!mptcp_cap_flag_sha256(flags) ||
74 (flags & MPTCP_CAP_EXTENSIBILITY))
75 break;
76
77 /* RFC 6824, Section 3.1:
78 * "For the Checksum Required bit (labeled "A"), if either
79 * host requires the use of checksums, checksums MUST be used.
80 * In other words, the only way for checksums not to be used
81 * is if both hosts in their SYNs set A=0."
82 */
83 if (flags & MPTCP_CAP_CHECKSUM_REQD)
84 mp_opt->suboptions |= OPTION_MPTCP_CSUMREQD;
85
86 mp_opt->deny_join_id0 = !!(flags & MPTCP_CAP_DENY_JOIN_ID0);
87
88 mp_opt->suboptions |= OPTIONS_MPTCP_MPC;
89 if (opsize >= TCPOLEN_MPTCP_MPC_SYNACK) {
90 mp_opt->sndr_key = get_unaligned_be64(ptr);
91 ptr += 8;
92 }
93 if (opsize >= TCPOLEN_MPTCP_MPC_ACK) {
94 mp_opt->rcvr_key = get_unaligned_be64(ptr);
95 ptr += 8;
96 }
97 if (opsize >= TCPOLEN_MPTCP_MPC_ACK_DATA) {
98 /* Section 3.1.:
99 * "the data parameters in a MP_CAPABLE are semantically
100 * equivalent to those in a DSS option and can be used
101 * interchangeably."
102 */
103 mp_opt->suboptions |= OPTION_MPTCP_DSS;
104 mp_opt->use_map = 1;
105 mp_opt->mpc_map = 1;
106 mp_opt->data_len = get_unaligned_be16(ptr);
107 ptr += 2;
108 }
109 if (opsize == TCPOLEN_MPTCP_MPC_ACK_DATA_CSUM) {
110 mp_opt->csum = (__force __sum16)get_unaligned_be16(ptr);
111 mp_opt->suboptions |= OPTION_MPTCP_CSUMREQD;
112 ptr += 2;
113 }
114 pr_debug("MP_CAPABLE version=%x, flags=%x, optlen=%d sndr=%llu, rcvr=%llu len=%d csum=%u",
115 version, flags, opsize, mp_opt->sndr_key,
116 mp_opt->rcvr_key, mp_opt->data_len, mp_opt->csum);
117 break;
118
119 case MPTCPOPT_MP_JOIN:
120 mp_opt->suboptions |= OPTIONS_MPTCP_MPJ;
121 if (opsize == TCPOLEN_MPTCP_MPJ_SYN) {
122 mp_opt->backup = *ptr++ & MPTCPOPT_BACKUP;
123 mp_opt->join_id = *ptr++;
124 mp_opt->token = get_unaligned_be32(ptr);
125 ptr += 4;
126 mp_opt->nonce = get_unaligned_be32(ptr);
127 ptr += 4;
128 pr_debug("MP_JOIN bkup=%u, id=%u, token=%u, nonce=%u",
129 mp_opt->backup, mp_opt->join_id,
130 mp_opt->token, mp_opt->nonce);
131 } else if (opsize == TCPOLEN_MPTCP_MPJ_SYNACK) {
132 mp_opt->backup = *ptr++ & MPTCPOPT_BACKUP;
133 mp_opt->join_id = *ptr++;
134 mp_opt->thmac = get_unaligned_be64(ptr);
135 ptr += 8;
136 mp_opt->nonce = get_unaligned_be32(ptr);
137 ptr += 4;
138 pr_debug("MP_JOIN bkup=%u, id=%u, thmac=%llu, nonce=%u",
139 mp_opt->backup, mp_opt->join_id,
140 mp_opt->thmac, mp_opt->nonce);
141 } else if (opsize == TCPOLEN_MPTCP_MPJ_ACK) {
142 ptr += 2;
143 memcpy(mp_opt->hmac, ptr, MPTCPOPT_HMAC_LEN);
144 pr_debug("MP_JOIN hmac");
145 } else {
146 mp_opt->suboptions &= ~OPTIONS_MPTCP_MPJ;
147 }
148 break;
149
150 case MPTCPOPT_DSS:
151 pr_debug("DSS");
152 ptr++;
153
154 /* we must clear 'mpc_map' be able to detect MP_CAPABLE
155 * map vs DSS map in mptcp_incoming_options(), and reconstruct
156 * map info accordingly
157 */
158 mp_opt->mpc_map = 0;
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",
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", 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 = (__force __sum16)get_unaligned_be16(ptr);
225 ptr += 2;
226 }
227
228 pr_debug("data_seq=%llu subflow_seq=%u data_len=%u csum=%d:%u",
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",
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", 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", 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 break;
327
328 case MPTCPOPT_RST:
329 if (opsize != TCPOLEN_MPTCP_RST)
330 break;
331
332 if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_RST))
333 break;
334
335 mp_opt->suboptions |= OPTION_MPTCP_RST;
336 flags = *ptr++;
337 mp_opt->reset_transient = flags & MPTCP_RST_TRANSIENT;
338 mp_opt->reset_reason = *ptr;
339 break;
340
341 case MPTCPOPT_MP_FAIL:
342 if (opsize != TCPOLEN_MPTCP_FAIL)
343 break;
344
345 ptr += 2;
346 mp_opt->suboptions |= OPTION_MPTCP_FAIL;
347 mp_opt->fail_seq = get_unaligned_be64(ptr);
348 pr_debug("MP_FAIL: data_seq=%llu", mp_opt->fail_seq);
349 break;
350
351 default:
352 break;
353 }
354 }
355
mptcp_get_options(const struct sock * sk,const struct sk_buff * skb,struct mptcp_options_received * mp_opt)356 void mptcp_get_options(const struct sock *sk,
357 const struct sk_buff *skb,
358 struct mptcp_options_received *mp_opt)
359 {
360 const struct tcphdr *th = tcp_hdr(skb);
361 const unsigned char *ptr;
362 int length;
363
364 /* initialize option status */
365 mp_opt->suboptions = 0;
366
367 length = (th->doff * 4) - sizeof(struct tcphdr);
368 ptr = (const unsigned char *)(th + 1);
369
370 while (length > 0) {
371 int opcode = *ptr++;
372 int opsize;
373
374 switch (opcode) {
375 case TCPOPT_EOL:
376 return;
377 case TCPOPT_NOP: /* Ref: RFC 793 section 3.1 */
378 length--;
379 continue;
380 default:
381 if (length < 2)
382 return;
383 opsize = *ptr++;
384 if (opsize < 2) /* "silly options" */
385 return;
386 if (opsize > length)
387 return; /* don't parse partial options */
388 if (opcode == TCPOPT_MPTCP)
389 mptcp_parse_option(skb, ptr, opsize, mp_opt);
390 ptr += opsize - 2;
391 length -= opsize;
392 }
393 }
394 }
395
mptcp_syn_options(struct sock * sk,const struct sk_buff * skb,unsigned int * size,struct mptcp_out_options * opts)396 bool mptcp_syn_options(struct sock *sk, const struct sk_buff *skb,
397 unsigned int *size, struct mptcp_out_options *opts)
398 {
399 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
400
401 /* we will use snd_isn to detect first pkt [re]transmission
402 * in mptcp_established_options_mp()
403 */
404 subflow->snd_isn = TCP_SKB_CB(skb)->end_seq;
405 if (subflow->request_mptcp) {
406 opts->suboptions = OPTION_MPTCP_MPC_SYN;
407 opts->csum_reqd = mptcp_is_checksum_enabled(sock_net(sk));
408 opts->allow_join_id0 = mptcp_allow_join_id0(sock_net(sk));
409 *size = TCPOLEN_MPTCP_MPC_SYN;
410 return true;
411 } else if (subflow->request_join) {
412 pr_debug("remote_token=%u, nonce=%u", subflow->remote_token,
413 subflow->local_nonce);
414 opts->suboptions = OPTION_MPTCP_MPJ_SYN;
415 opts->join_id = subflow->local_id;
416 opts->token = subflow->remote_token;
417 opts->nonce = subflow->local_nonce;
418 opts->backup = subflow->request_bkup;
419 *size = TCPOLEN_MPTCP_MPJ_SYN;
420 return true;
421 }
422 return false;
423 }
424
clear_3rdack_retransmission(struct sock * sk)425 static void clear_3rdack_retransmission(struct sock *sk)
426 {
427 struct inet_connection_sock *icsk = inet_csk(sk);
428
429 sk_stop_timer(sk, &icsk->icsk_delack_timer);
430 icsk->icsk_ack.timeout = 0;
431 icsk->icsk_ack.ato = 0;
432 icsk->icsk_ack.pending &= ~(ICSK_ACK_SCHED | ICSK_ACK_TIMER);
433 }
434
mptcp_established_options_mp(struct sock * sk,struct sk_buff * skb,bool snd_data_fin_enable,unsigned int * size,unsigned int remaining,struct mptcp_out_options * opts)435 static bool mptcp_established_options_mp(struct sock *sk, struct sk_buff *skb,
436 bool snd_data_fin_enable,
437 unsigned int *size,
438 unsigned int remaining,
439 struct mptcp_out_options *opts)
440 {
441 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
442 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
443 struct mptcp_ext *mpext;
444 unsigned int data_len;
445 u8 len;
446
447 /* When skb is not available, we better over-estimate the emitted
448 * options len. A full DSS option (28 bytes) is longer than
449 * TCPOLEN_MPTCP_MPC_ACK_DATA(22) or TCPOLEN_MPTCP_MPJ_ACK(24), so
450 * tell the caller to defer the estimate to
451 * mptcp_established_options_dss(), which will reserve enough space.
452 */
453 if (!skb)
454 return false;
455
456 /* MPC/MPJ needed only on 3rd ack packet, DATA_FIN and TCP shutdown take precedence */
457 if (subflow->fully_established || snd_data_fin_enable ||
458 subflow->snd_isn != TCP_SKB_CB(skb)->seq ||
459 sk->sk_state != TCP_ESTABLISHED)
460 return false;
461
462 if (subflow->mp_capable) {
463 mpext = mptcp_get_ext(skb);
464 data_len = mpext ? mpext->data_len : 0;
465
466 /* we will check ops->data_len in mptcp_write_options() to
467 * discriminate between TCPOLEN_MPTCP_MPC_ACK_DATA and
468 * TCPOLEN_MPTCP_MPC_ACK
469 */
470 opts->data_len = data_len;
471 opts->suboptions = OPTION_MPTCP_MPC_ACK;
472 opts->sndr_key = subflow->local_key;
473 opts->rcvr_key = subflow->remote_key;
474 opts->csum_reqd = READ_ONCE(msk->csum_enabled);
475 opts->allow_join_id0 = mptcp_allow_join_id0(sock_net(sk));
476
477 /* Section 3.1.
478 * The MP_CAPABLE option is carried on the SYN, SYN/ACK, and ACK
479 * packets that start the first subflow of an MPTCP connection,
480 * as well as the first packet that carries data
481 */
482 if (data_len > 0) {
483 len = TCPOLEN_MPTCP_MPC_ACK_DATA;
484 if (opts->csum_reqd) {
485 /* we need to propagate more info to csum the pseudo hdr */
486 opts->data_seq = mpext->data_seq;
487 opts->subflow_seq = mpext->subflow_seq;
488 opts->csum = mpext->csum;
489 len += TCPOLEN_MPTCP_DSS_CHECKSUM;
490 }
491 *size = ALIGN(len, 4);
492 } else {
493 *size = TCPOLEN_MPTCP_MPC_ACK;
494 }
495
496 pr_debug("subflow=%p, local_key=%llu, remote_key=%llu map_len=%d",
497 subflow, subflow->local_key, subflow->remote_key,
498 data_len);
499
500 return true;
501 } else if (subflow->mp_join) {
502 opts->suboptions = OPTION_MPTCP_MPJ_ACK;
503 memcpy(opts->hmac, subflow->hmac, MPTCPOPT_HMAC_LEN);
504 *size = TCPOLEN_MPTCP_MPJ_ACK;
505 pr_debug("subflow=%p", subflow);
506
507 /* we can use the full delegate action helper only from BH context
508 * If we are in process context - sk is flushing the backlog at
509 * socket lock release time - just set the appropriate flag, will
510 * be handled by the release callback
511 */
512 if (sock_owned_by_user(sk))
513 set_bit(MPTCP_DELEGATE_ACK, &subflow->delegated_status);
514 else
515 mptcp_subflow_delegate(subflow, MPTCP_DELEGATE_ACK);
516 return true;
517 }
518 return false;
519 }
520
mptcp_write_data_fin(struct mptcp_subflow_context * subflow,struct sk_buff * skb,struct mptcp_ext * ext)521 static void mptcp_write_data_fin(struct mptcp_subflow_context *subflow,
522 struct sk_buff *skb, struct mptcp_ext *ext)
523 {
524 /* The write_seq value has already been incremented, so the actual
525 * sequence number for the DATA_FIN is one less.
526 */
527 u64 data_fin_tx_seq = READ_ONCE(mptcp_sk(subflow->conn)->write_seq) - 1;
528
529 if (!ext->use_map || !skb->len) {
530 /* RFC6824 requires a DSS mapping with specific values
531 * if DATA_FIN is set but no data payload is mapped
532 */
533 ext->data_fin = 1;
534 ext->use_map = 1;
535 ext->dsn64 = 1;
536 ext->data_seq = data_fin_tx_seq;
537 ext->subflow_seq = 0;
538 ext->data_len = 1;
539 } else if (ext->data_seq + ext->data_len == data_fin_tx_seq) {
540 /* If there's an existing DSS mapping and it is the
541 * final mapping, DATA_FIN consumes 1 additional byte of
542 * mapping space.
543 */
544 ext->data_fin = 1;
545 ext->data_len++;
546 }
547 }
548
mptcp_established_options_dss(struct sock * sk,struct sk_buff * skb,bool snd_data_fin_enable,unsigned int * size,unsigned int remaining,struct mptcp_out_options * opts)549 static bool mptcp_established_options_dss(struct sock *sk, struct sk_buff *skb,
550 bool snd_data_fin_enable,
551 unsigned int *size,
552 unsigned int remaining,
553 struct mptcp_out_options *opts)
554 {
555 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
556 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
557 unsigned int dss_size = 0;
558 struct mptcp_ext *mpext;
559 unsigned int ack_size;
560 bool ret = false;
561 u64 ack_seq;
562
563 opts->csum_reqd = READ_ONCE(msk->csum_enabled);
564 mpext = skb ? mptcp_get_ext(skb) : NULL;
565
566 if (!skb || (mpext && mpext->use_map) || snd_data_fin_enable) {
567 unsigned int map_size = TCPOLEN_MPTCP_DSS_BASE + TCPOLEN_MPTCP_DSS_MAP64;
568
569 if (mpext) {
570 if (opts->csum_reqd)
571 map_size += TCPOLEN_MPTCP_DSS_CHECKSUM;
572
573 opts->ext_copy = *mpext;
574 }
575
576 remaining -= map_size;
577 dss_size = map_size;
578 if (skb && snd_data_fin_enable)
579 mptcp_write_data_fin(subflow, skb, &opts->ext_copy);
580 opts->suboptions = OPTION_MPTCP_DSS;
581 ret = true;
582 }
583
584 /* passive sockets msk will set the 'can_ack' after accept(), even
585 * if the first subflow may have the already the remote key handy
586 */
587 opts->ext_copy.use_ack = 0;
588 if (!READ_ONCE(msk->can_ack)) {
589 *size = ALIGN(dss_size, 4);
590 return ret;
591 }
592
593 ack_seq = READ_ONCE(msk->ack_seq);
594 if (READ_ONCE(msk->use_64bit_ack)) {
595 ack_size = TCPOLEN_MPTCP_DSS_ACK64;
596 opts->ext_copy.data_ack = ack_seq;
597 opts->ext_copy.ack64 = 1;
598 } else {
599 ack_size = TCPOLEN_MPTCP_DSS_ACK32;
600 opts->ext_copy.data_ack32 = (uint32_t)ack_seq;
601 opts->ext_copy.ack64 = 0;
602 }
603 opts->ext_copy.use_ack = 1;
604 opts->suboptions = OPTION_MPTCP_DSS;
605 WRITE_ONCE(msk->old_wspace, __mptcp_space((struct sock *)msk));
606
607 /* Add kind/length/subtype/flag overhead if mapping is not populated */
608 if (dss_size == 0)
609 ack_size += TCPOLEN_MPTCP_DSS_BASE;
610
611 dss_size += ack_size;
612
613 *size = ALIGN(dss_size, 4);
614 return true;
615 }
616
add_addr_generate_hmac(u64 key1,u64 key2,struct mptcp_addr_info * addr)617 static u64 add_addr_generate_hmac(u64 key1, u64 key2,
618 struct mptcp_addr_info *addr)
619 {
620 u16 port = ntohs(addr->port);
621 u8 hmac[SHA256_DIGEST_SIZE];
622 u8 msg[19];
623 int i = 0;
624
625 msg[i++] = addr->id;
626 if (addr->family == AF_INET) {
627 memcpy(&msg[i], &addr->addr.s_addr, 4);
628 i += 4;
629 }
630 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
631 else if (addr->family == AF_INET6) {
632 memcpy(&msg[i], &addr->addr6.s6_addr, 16);
633 i += 16;
634 }
635 #endif
636 msg[i++] = port >> 8;
637 msg[i++] = port & 0xFF;
638
639 mptcp_crypto_hmac_sha(key1, key2, msg, i, hmac);
640
641 return get_unaligned_be64(&hmac[SHA256_DIGEST_SIZE - sizeof(u64)]);
642 }
643
mptcp_established_options_add_addr(struct sock * sk,struct sk_buff * skb,unsigned int * size,unsigned int remaining,struct mptcp_out_options * opts)644 static bool mptcp_established_options_add_addr(struct sock *sk, struct sk_buff *skb,
645 unsigned int *size,
646 unsigned int remaining,
647 struct mptcp_out_options *opts)
648 {
649 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
650 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
651 bool drop_other_suboptions = false;
652 unsigned int opt_size = *size;
653 bool echo;
654 bool port;
655 int len;
656
657 /* add addr will strip the existing options, be sure to avoid breaking
658 * MPC/MPJ handshakes
659 */
660 if (!mptcp_pm_should_add_signal(msk) ||
661 (opts->suboptions & (OPTION_MPTCP_MPJ_ACK | OPTION_MPTCP_MPC_ACK)) ||
662 !mptcp_pm_add_addr_signal(msk, skb, opt_size, remaining, &opts->addr,
663 &echo, &port, &drop_other_suboptions))
664 return false;
665
666 if (drop_other_suboptions)
667 remaining += opt_size;
668 len = mptcp_add_addr_len(opts->addr.family, echo, port);
669 if (remaining < len)
670 return false;
671
672 *size = len;
673 if (drop_other_suboptions) {
674 pr_debug("drop other suboptions");
675 opts->suboptions = 0;
676
677 /* note that e.g. DSS could have written into the memory
678 * aliased by ahmac, we must reset the field here
679 * to avoid appending the hmac even for ADD_ADDR echo
680 * options
681 */
682 opts->ahmac = 0;
683 *size -= opt_size;
684 }
685 opts->suboptions |= OPTION_MPTCP_ADD_ADDR;
686 if (!echo) {
687 opts->ahmac = add_addr_generate_hmac(msk->local_key,
688 msk->remote_key,
689 &opts->addr);
690 }
691 pr_debug("addr_id=%d, ahmac=%llu, echo=%d, port=%d",
692 opts->addr.id, opts->ahmac, echo, ntohs(opts->addr.port));
693
694 return true;
695 }
696
mptcp_established_options_rm_addr(struct sock * sk,unsigned int * size,unsigned int remaining,struct mptcp_out_options * opts)697 static bool mptcp_established_options_rm_addr(struct sock *sk,
698 unsigned int *size,
699 unsigned int remaining,
700 struct mptcp_out_options *opts)
701 {
702 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
703 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
704 struct mptcp_rm_list rm_list;
705 int i, len;
706
707 if (!mptcp_pm_should_rm_signal(msk) ||
708 !(mptcp_pm_rm_addr_signal(msk, remaining, &rm_list)))
709 return false;
710
711 len = mptcp_rm_addr_len(&rm_list);
712 if (len < 0)
713 return false;
714 if (remaining < len)
715 return false;
716
717 *size = len;
718 opts->suboptions |= OPTION_MPTCP_RM_ADDR;
719 opts->rm_list = rm_list;
720
721 for (i = 0; i < opts->rm_list.nr; i++)
722 pr_debug("rm_list_ids[%d]=%d", i, opts->rm_list.ids[i]);
723
724 return true;
725 }
726
mptcp_established_options_mp_prio(struct sock * sk,unsigned int * size,unsigned int remaining,struct mptcp_out_options * opts)727 static bool mptcp_established_options_mp_prio(struct sock *sk,
728 unsigned int *size,
729 unsigned int remaining,
730 struct mptcp_out_options *opts)
731 {
732 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
733
734 /* can't send MP_PRIO with MPC, as they share the same option space:
735 * 'backup'. Also it makes no sense at all
736 */
737 if (!subflow->send_mp_prio || (opts->suboptions & OPTIONS_MPTCP_MPC))
738 return false;
739
740 /* account for the trailing 'nop' option */
741 if (remaining < TCPOLEN_MPTCP_PRIO_ALIGN)
742 return false;
743
744 *size = TCPOLEN_MPTCP_PRIO_ALIGN;
745 opts->suboptions |= OPTION_MPTCP_PRIO;
746 opts->backup = subflow->request_bkup;
747
748 pr_debug("prio=%d", opts->backup);
749
750 return true;
751 }
752
mptcp_established_options_rst(struct sock * sk,struct sk_buff * skb,unsigned int * size,unsigned int remaining,struct mptcp_out_options * opts)753 static noinline bool mptcp_established_options_rst(struct sock *sk, struct sk_buff *skb,
754 unsigned int *size,
755 unsigned int remaining,
756 struct mptcp_out_options *opts)
757 {
758 const struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
759
760 if (remaining < TCPOLEN_MPTCP_RST)
761 return false;
762
763 *size = TCPOLEN_MPTCP_RST;
764 opts->suboptions |= OPTION_MPTCP_RST;
765 opts->reset_transient = subflow->reset_transient;
766 opts->reset_reason = subflow->reset_reason;
767
768 return true;
769 }
770
mptcp_established_options_mp_fail(struct sock * sk,unsigned int * size,unsigned int remaining,struct mptcp_out_options * opts)771 static bool mptcp_established_options_mp_fail(struct sock *sk,
772 unsigned int *size,
773 unsigned int remaining,
774 struct mptcp_out_options *opts)
775 {
776 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
777
778 if (likely(!subflow->send_mp_fail))
779 return false;
780
781 if (remaining < TCPOLEN_MPTCP_FAIL)
782 return false;
783
784 *size = TCPOLEN_MPTCP_FAIL;
785 opts->suboptions |= OPTION_MPTCP_FAIL;
786 opts->fail_seq = subflow->map_seq;
787
788 pr_debug("MP_FAIL fail_seq=%llu", opts->fail_seq);
789
790 return true;
791 }
792
mptcp_established_options(struct sock * sk,struct sk_buff * skb,unsigned int * size,unsigned int remaining,struct mptcp_out_options * opts)793 bool mptcp_established_options(struct sock *sk, struct sk_buff *skb,
794 unsigned int *size, unsigned int remaining,
795 struct mptcp_out_options *opts)
796 {
797 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
798 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
799 unsigned int opt_size = 0;
800 bool snd_data_fin;
801 bool ret = false;
802
803 opts->suboptions = 0;
804
805 if (unlikely(__mptcp_check_fallback(msk)))
806 return false;
807
808 if (unlikely(skb && TCP_SKB_CB(skb)->tcp_flags & TCPHDR_RST)) {
809 if (mptcp_established_options_mp_fail(sk, &opt_size, remaining, opts)) {
810 *size += opt_size;
811 remaining -= opt_size;
812 }
813 if (mptcp_established_options_rst(sk, skb, &opt_size, remaining, opts)) {
814 *size += opt_size;
815 remaining -= opt_size;
816 }
817 return true;
818 }
819
820 snd_data_fin = mptcp_data_fin_enabled(msk);
821 if (mptcp_established_options_mp(sk, skb, snd_data_fin, &opt_size, remaining, opts))
822 ret = true;
823 else if (mptcp_established_options_dss(sk, skb, snd_data_fin, &opt_size, remaining, opts)) {
824 ret = true;
825 if (mptcp_established_options_mp_fail(sk, &opt_size, remaining, opts)) {
826 *size += opt_size;
827 remaining -= opt_size;
828 return true;
829 }
830 }
831
832 /* we reserved enough space for the above options, and exceeding the
833 * TCP option space would be fatal
834 */
835 if (WARN_ON_ONCE(opt_size > remaining))
836 return false;
837
838 *size += opt_size;
839 remaining -= opt_size;
840 if (mptcp_established_options_add_addr(sk, skb, &opt_size, remaining, opts)) {
841 *size += opt_size;
842 remaining -= opt_size;
843 ret = true;
844 } else if (mptcp_established_options_rm_addr(sk, &opt_size, remaining, opts)) {
845 *size += opt_size;
846 remaining -= opt_size;
847 ret = true;
848 }
849
850 if (mptcp_established_options_mp_prio(sk, &opt_size, remaining, opts)) {
851 *size += opt_size;
852 remaining -= opt_size;
853 ret = true;
854 }
855
856 return ret;
857 }
858
mptcp_synack_options(const struct request_sock * req,unsigned int * size,struct mptcp_out_options * opts)859 bool mptcp_synack_options(const struct request_sock *req, unsigned int *size,
860 struct mptcp_out_options *opts)
861 {
862 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
863
864 if (subflow_req->mp_capable) {
865 opts->suboptions = OPTION_MPTCP_MPC_SYNACK;
866 opts->sndr_key = subflow_req->local_key;
867 opts->csum_reqd = subflow_req->csum_reqd;
868 opts->allow_join_id0 = subflow_req->allow_join_id0;
869 *size = TCPOLEN_MPTCP_MPC_SYNACK;
870 pr_debug("subflow_req=%p, local_key=%llu",
871 subflow_req, subflow_req->local_key);
872 return true;
873 } else if (subflow_req->mp_join) {
874 opts->suboptions = OPTION_MPTCP_MPJ_SYNACK;
875 opts->backup = subflow_req->backup;
876 opts->join_id = subflow_req->local_id;
877 opts->thmac = subflow_req->thmac;
878 opts->nonce = subflow_req->local_nonce;
879 pr_debug("req=%p, bkup=%u, id=%u, thmac=%llu, nonce=%u",
880 subflow_req, opts->backup, opts->join_id,
881 opts->thmac, opts->nonce);
882 *size = TCPOLEN_MPTCP_MPJ_SYNACK;
883 return true;
884 }
885 return false;
886 }
887
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)888 static bool check_fully_established(struct mptcp_sock *msk, struct sock *ssk,
889 struct mptcp_subflow_context *subflow,
890 struct sk_buff *skb,
891 struct mptcp_options_received *mp_opt)
892 {
893 /* here we can process OoO, in-window pkts, only in-sequence 4th ack
894 * will make the subflow fully established
895 */
896 if (likely(subflow->fully_established)) {
897 /* on passive sockets, check for 3rd ack retransmission
898 * note that msk is always set by subflow_syn_recv_sock()
899 * for mp_join subflows
900 */
901 if (TCP_SKB_CB(skb)->seq == subflow->ssn_offset + 1 &&
902 TCP_SKB_CB(skb)->end_seq == TCP_SKB_CB(skb)->seq &&
903 subflow->mp_join && (mp_opt->suboptions & OPTIONS_MPTCP_MPJ) &&
904 READ_ONCE(msk->pm.server_side))
905 tcp_send_ack(ssk);
906 goto fully_established;
907 }
908
909 /* we must process OoO packets before the first subflow is fully
910 * established. OoO packets are instead a protocol violation
911 * for MP_JOIN subflows as the peer must not send any data
912 * before receiving the forth ack - cfr. RFC 8684 section 3.2.
913 */
914 if (TCP_SKB_CB(skb)->seq != subflow->ssn_offset + 1) {
915 if (subflow->mp_join)
916 goto reset;
917 return subflow->mp_capable;
918 }
919
920 if (((mp_opt->suboptions & OPTION_MPTCP_DSS) && mp_opt->use_ack) ||
921 ((mp_opt->suboptions & OPTION_MPTCP_ADD_ADDR) && !mp_opt->echo)) {
922 /* subflows are fully established as soon as we get any
923 * additional ack, including ADD_ADDR.
924 */
925 subflow->fully_established = 1;
926 WRITE_ONCE(msk->fully_established, true);
927 goto fully_established;
928 }
929
930 /* If the first established packet does not contain MP_CAPABLE + data
931 * then fallback to TCP. Fallback scenarios requires a reset for
932 * MP_JOIN subflows.
933 */
934 if (!(mp_opt->suboptions & OPTIONS_MPTCP_MPC)) {
935 if (subflow->mp_join)
936 goto reset;
937 subflow->mp_capable = 0;
938 pr_fallback(msk);
939 __mptcp_do_fallback(msk);
940 return false;
941 }
942
943 if (mp_opt->deny_join_id0)
944 WRITE_ONCE(msk->pm.remote_deny_join_id0, true);
945
946 if (unlikely(!READ_ONCE(msk->pm.server_side)))
947 pr_warn_once("bogus mpc option on established client sk");
948 mptcp_subflow_fully_established(subflow, mp_opt);
949
950 fully_established:
951 /* if the subflow is not already linked into the conn_list, we can't
952 * notify the PM: this subflow is still on the listener queue
953 * and the PM possibly acquiring the subflow lock could race with
954 * the listener close
955 */
956 if (likely(subflow->pm_notified) || list_empty(&subflow->node))
957 return true;
958
959 subflow->pm_notified = 1;
960 if (subflow->mp_join) {
961 clear_3rdack_retransmission(ssk);
962 mptcp_pm_subflow_established(msk);
963 } else {
964 mptcp_pm_fully_established(msk, ssk, GFP_ATOMIC);
965 }
966 return true;
967
968 reset:
969 mptcp_subflow_reset(ssk);
970 return false;
971 }
972
__mptcp_expand_seq(u64 old_seq,u64 cur_seq)973 u64 __mptcp_expand_seq(u64 old_seq, u64 cur_seq)
974 {
975 u32 old_seq32, cur_seq32;
976
977 old_seq32 = (u32)old_seq;
978 cur_seq32 = (u32)cur_seq;
979 cur_seq = (old_seq & GENMASK_ULL(63, 32)) + cur_seq32;
980 if (unlikely(cur_seq32 < old_seq32 && before(old_seq32, cur_seq32)))
981 return cur_seq + (1LL << 32);
982
983 /* reverse wrap could happen, too */
984 if (unlikely(cur_seq32 > old_seq32 && after(old_seq32, cur_seq32)))
985 return cur_seq - (1LL << 32);
986 return cur_seq;
987 }
988
ack_update_msk(struct mptcp_sock * msk,struct sock * ssk,struct mptcp_options_received * mp_opt)989 static void ack_update_msk(struct mptcp_sock *msk,
990 struct sock *ssk,
991 struct mptcp_options_received *mp_opt)
992 {
993 u64 new_wnd_end, new_snd_una, snd_nxt = READ_ONCE(msk->snd_nxt);
994 struct sock *sk = (struct sock *)msk;
995 u64 old_snd_una;
996
997 mptcp_data_lock(sk);
998
999 /* avoid ack expansion on update conflict, to reduce the risk of
1000 * wrongly expanding to a future ack sequence number, which is way
1001 * more dangerous than missing an ack
1002 */
1003 old_snd_una = msk->snd_una;
1004 new_snd_una = mptcp_expand_seq(old_snd_una, mp_opt->data_ack, mp_opt->ack64);
1005
1006 /* ACK for data not even sent yet? Ignore.*/
1007 if (unlikely(after64(new_snd_una, snd_nxt)))
1008 new_snd_una = old_snd_una;
1009
1010 new_wnd_end = new_snd_una + tcp_sk(ssk)->snd_wnd;
1011
1012 if (after64(new_wnd_end, msk->wnd_end))
1013 msk->wnd_end = new_wnd_end;
1014
1015 /* this assumes mptcp_incoming_options() is invoked after tcp_ack() */
1016 if (after64(msk->wnd_end, READ_ONCE(msk->snd_nxt)))
1017 __mptcp_check_push(sk, ssk);
1018
1019 if (after64(new_snd_una, old_snd_una)) {
1020 msk->snd_una = new_snd_una;
1021 __mptcp_data_acked(sk);
1022 }
1023 mptcp_data_unlock(sk);
1024
1025 trace_ack_update_msk(mp_opt->data_ack,
1026 old_snd_una, new_snd_una,
1027 new_wnd_end, msk->wnd_end);
1028 }
1029
mptcp_update_rcv_data_fin(struct mptcp_sock * msk,u64 data_fin_seq,bool use_64bit)1030 bool mptcp_update_rcv_data_fin(struct mptcp_sock *msk, u64 data_fin_seq, bool use_64bit)
1031 {
1032 /* Skip if DATA_FIN was already received.
1033 * If updating simultaneously with the recvmsg loop, values
1034 * should match. If they mismatch, the peer is misbehaving and
1035 * we will prefer the most recent information.
1036 */
1037 if (READ_ONCE(msk->rcv_data_fin))
1038 return false;
1039
1040 WRITE_ONCE(msk->rcv_data_fin_seq,
1041 mptcp_expand_seq(READ_ONCE(msk->ack_seq), data_fin_seq, use_64bit));
1042 WRITE_ONCE(msk->rcv_data_fin, 1);
1043
1044 return true;
1045 }
1046
add_addr_hmac_valid(struct mptcp_sock * msk,struct mptcp_options_received * mp_opt)1047 static bool add_addr_hmac_valid(struct mptcp_sock *msk,
1048 struct mptcp_options_received *mp_opt)
1049 {
1050 u64 hmac = 0;
1051
1052 if (mp_opt->echo)
1053 return true;
1054
1055 hmac = add_addr_generate_hmac(msk->remote_key,
1056 msk->local_key,
1057 &mp_opt->addr);
1058
1059 pr_debug("msk=%p, ahmac=%llu, mp_opt->ahmac=%llu\n",
1060 msk, (unsigned long long)hmac,
1061 (unsigned long long)mp_opt->ahmac);
1062
1063 return hmac == mp_opt->ahmac;
1064 }
1065
1066 /* Return false if a subflow has been reset, else return true */
mptcp_incoming_options(struct sock * sk,struct sk_buff * skb)1067 bool mptcp_incoming_options(struct sock *sk, struct sk_buff *skb)
1068 {
1069 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1070 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
1071 struct mptcp_options_received mp_opt;
1072 struct mptcp_ext *mpext;
1073
1074 if (__mptcp_check_fallback(msk)) {
1075 /* Keep it simple and unconditionally trigger send data cleanup and
1076 * pending queue spooling. We will need to acquire the data lock
1077 * for more accurate checks, and once the lock is acquired, such
1078 * helpers are cheap.
1079 */
1080 mptcp_data_lock(subflow->conn);
1081 if (sk_stream_memory_free(sk))
1082 __mptcp_check_push(subflow->conn, sk);
1083 __mptcp_data_acked(subflow->conn);
1084 mptcp_data_unlock(subflow->conn);
1085 return true;
1086 }
1087
1088 mptcp_get_options(sk, skb, &mp_opt);
1089
1090 /* The subflow can be in close state only if check_fully_established()
1091 * just sent a reset. If so, tell the caller to ignore the current packet.
1092 */
1093 if (!check_fully_established(msk, sk, subflow, skb, &mp_opt))
1094 return sk->sk_state != TCP_CLOSE;
1095
1096 if (unlikely(mp_opt.suboptions != OPTION_MPTCP_DSS)) {
1097 if ((mp_opt.suboptions & OPTION_MPTCP_FASTCLOSE) &&
1098 msk->local_key == mp_opt.rcvr_key) {
1099 WRITE_ONCE(msk->rcv_fastclose, true);
1100 mptcp_schedule_work((struct sock *)msk);
1101 }
1102
1103 if ((mp_opt.suboptions & OPTION_MPTCP_ADD_ADDR) &&
1104 add_addr_hmac_valid(msk, &mp_opt)) {
1105 if (!mp_opt.echo) {
1106 mptcp_pm_add_addr_received(msk, &mp_opt.addr);
1107 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ADDADDR);
1108 } else {
1109 mptcp_pm_add_addr_echoed(msk, &mp_opt.addr);
1110 mptcp_pm_del_add_timer(msk, &mp_opt.addr, true);
1111 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ECHOADD);
1112 }
1113
1114 if (mp_opt.addr.port)
1115 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_PORTADD);
1116 }
1117
1118 if (mp_opt.suboptions & OPTION_MPTCP_RM_ADDR)
1119 mptcp_pm_rm_addr_received(msk, &mp_opt.rm_list);
1120
1121 if (mp_opt.suboptions & OPTION_MPTCP_PRIO) {
1122 mptcp_pm_mp_prio_received(sk, mp_opt.backup);
1123 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPPRIORX);
1124 }
1125
1126 if (mp_opt.suboptions & OPTION_MPTCP_FAIL) {
1127 mptcp_pm_mp_fail_received(sk, mp_opt.fail_seq);
1128 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPFAILRX);
1129 }
1130
1131 if (mp_opt.suboptions & OPTION_MPTCP_RST) {
1132 subflow->reset_seen = 1;
1133 subflow->reset_reason = mp_opt.reset_reason;
1134 subflow->reset_transient = mp_opt.reset_transient;
1135 }
1136
1137 if (!(mp_opt.suboptions & OPTION_MPTCP_DSS))
1138 return true;
1139 }
1140
1141 /* we can't wait for recvmsg() to update the ack_seq, otherwise
1142 * monodirectional flows will stuck
1143 */
1144 if (mp_opt.use_ack)
1145 ack_update_msk(msk, sk, &mp_opt);
1146
1147 /* Zero-data-length packets are dropped by the caller and not
1148 * propagated to the MPTCP layer, so the skb extension does not
1149 * need to be allocated or populated. DATA_FIN information, if
1150 * present, needs to be updated here before the skb is freed.
1151 */
1152 if (TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb)->end_seq) {
1153 if (mp_opt.data_fin && mp_opt.data_len == 1 &&
1154 mptcp_update_rcv_data_fin(msk, mp_opt.data_seq, mp_opt.dsn64) &&
1155 schedule_work(&msk->work))
1156 sock_hold(subflow->conn);
1157
1158 return true;
1159 }
1160
1161 mpext = skb_ext_add(skb, SKB_EXT_MPTCP);
1162 if (!mpext)
1163 return true;
1164
1165 memset(mpext, 0, sizeof(*mpext));
1166
1167 if (likely(mp_opt.use_map)) {
1168 if (mp_opt.mpc_map) {
1169 /* this is an MP_CAPABLE carrying MPTCP data
1170 * we know this map the first chunk of data
1171 */
1172 mptcp_crypto_key_sha(subflow->remote_key, NULL,
1173 &mpext->data_seq);
1174 mpext->data_seq++;
1175 mpext->subflow_seq = 1;
1176 mpext->dsn64 = 1;
1177 mpext->mpc_map = 1;
1178 mpext->data_fin = 0;
1179 } else {
1180 mpext->data_seq = mp_opt.data_seq;
1181 mpext->subflow_seq = mp_opt.subflow_seq;
1182 mpext->dsn64 = mp_opt.dsn64;
1183 mpext->data_fin = mp_opt.data_fin;
1184 }
1185 mpext->data_len = mp_opt.data_len;
1186 mpext->use_map = 1;
1187 mpext->csum_reqd = !!(mp_opt.suboptions & OPTION_MPTCP_CSUMREQD);
1188
1189 if (mpext->csum_reqd)
1190 mpext->csum = mp_opt.csum;
1191 }
1192
1193 return true;
1194 }
1195
mptcp_set_rwin(const struct tcp_sock * tp)1196 static void mptcp_set_rwin(const struct tcp_sock *tp)
1197 {
1198 const struct sock *ssk = (const struct sock *)tp;
1199 const struct mptcp_subflow_context *subflow;
1200 struct mptcp_sock *msk;
1201 u64 ack_seq;
1202
1203 subflow = mptcp_subflow_ctx(ssk);
1204 msk = mptcp_sk(subflow->conn);
1205
1206 ack_seq = READ_ONCE(msk->ack_seq) + tp->rcv_wnd;
1207
1208 if (after64(ack_seq, READ_ONCE(msk->rcv_wnd_sent)))
1209 WRITE_ONCE(msk->rcv_wnd_sent, ack_seq);
1210 }
1211
__mptcp_make_csum(u64 data_seq,u32 subflow_seq,u16 data_len,__sum16 sum)1212 static u16 __mptcp_make_csum(u64 data_seq, u32 subflow_seq, u16 data_len, __sum16 sum)
1213 {
1214 struct csum_pseudo_header header;
1215 __wsum csum;
1216
1217 /* cfr RFC 8684 3.3.1.:
1218 * the data sequence number used in the pseudo-header is
1219 * always the 64-bit value, irrespective of what length is used in the
1220 * DSS option itself.
1221 */
1222 header.data_seq = cpu_to_be64(data_seq);
1223 header.subflow_seq = htonl(subflow_seq);
1224 header.data_len = htons(data_len);
1225 header.csum = 0;
1226
1227 csum = csum_partial(&header, sizeof(header), ~csum_unfold(sum));
1228 return (__force u16)csum_fold(csum);
1229 }
1230
mptcp_make_csum(const struct mptcp_ext * mpext)1231 static u16 mptcp_make_csum(const struct mptcp_ext *mpext)
1232 {
1233 return __mptcp_make_csum(mpext->data_seq, mpext->subflow_seq, mpext->data_len,
1234 mpext->csum);
1235 }
1236
mptcp_write_options(__be32 * ptr,const struct tcp_sock * tp,struct mptcp_out_options * opts)1237 void mptcp_write_options(__be32 *ptr, const struct tcp_sock *tp,
1238 struct mptcp_out_options *opts)
1239 {
1240 if (unlikely(OPTION_MPTCP_FAIL & opts->suboptions)) {
1241 const struct sock *ssk = (const struct sock *)tp;
1242 struct mptcp_subflow_context *subflow;
1243
1244 subflow = mptcp_subflow_ctx(ssk);
1245 subflow->send_mp_fail = 0;
1246
1247 *ptr++ = mptcp_option(MPTCPOPT_MP_FAIL,
1248 TCPOLEN_MPTCP_FAIL,
1249 0, 0);
1250 put_unaligned_be64(opts->fail_seq, ptr);
1251 ptr += 2;
1252 }
1253
1254 /* RST is mutually exclusive with everything else */
1255 if (unlikely(OPTION_MPTCP_RST & opts->suboptions)) {
1256 *ptr++ = mptcp_option(MPTCPOPT_RST,
1257 TCPOLEN_MPTCP_RST,
1258 opts->reset_transient,
1259 opts->reset_reason);
1260 return;
1261 }
1262
1263 /* DSS, MPC, MPJ and ADD_ADDR are mutually exclusive, see
1264 * mptcp_established_options*()
1265 */
1266 if (likely(OPTION_MPTCP_DSS & opts->suboptions)) {
1267 struct mptcp_ext *mpext = &opts->ext_copy;
1268 u8 len = TCPOLEN_MPTCP_DSS_BASE;
1269 u8 flags = 0;
1270
1271 if (mpext->use_ack) {
1272 flags = MPTCP_DSS_HAS_ACK;
1273 if (mpext->ack64) {
1274 len += TCPOLEN_MPTCP_DSS_ACK64;
1275 flags |= MPTCP_DSS_ACK64;
1276 } else {
1277 len += TCPOLEN_MPTCP_DSS_ACK32;
1278 }
1279 }
1280
1281 if (mpext->use_map) {
1282 len += TCPOLEN_MPTCP_DSS_MAP64;
1283
1284 /* Use only 64-bit mapping flags for now, add
1285 * support for optional 32-bit mappings later.
1286 */
1287 flags |= MPTCP_DSS_HAS_MAP | MPTCP_DSS_DSN64;
1288 if (mpext->data_fin)
1289 flags |= MPTCP_DSS_DATA_FIN;
1290
1291 if (opts->csum_reqd)
1292 len += TCPOLEN_MPTCP_DSS_CHECKSUM;
1293 }
1294
1295 *ptr++ = mptcp_option(MPTCPOPT_DSS, len, 0, flags);
1296
1297 if (mpext->use_ack) {
1298 if (mpext->ack64) {
1299 put_unaligned_be64(mpext->data_ack, ptr);
1300 ptr += 2;
1301 } else {
1302 put_unaligned_be32(mpext->data_ack32, ptr);
1303 ptr += 1;
1304 }
1305 }
1306
1307 if (mpext->use_map) {
1308 put_unaligned_be64(mpext->data_seq, ptr);
1309 ptr += 2;
1310 put_unaligned_be32(mpext->subflow_seq, ptr);
1311 ptr += 1;
1312 if (opts->csum_reqd) {
1313 put_unaligned_be32(mpext->data_len << 16 |
1314 mptcp_make_csum(mpext), ptr);
1315 } else {
1316 put_unaligned_be32(mpext->data_len << 16 |
1317 TCPOPT_NOP << 8 | TCPOPT_NOP, ptr);
1318 }
1319 }
1320 } else if (OPTIONS_MPTCP_MPC & opts->suboptions) {
1321 u8 len, flag = MPTCP_CAP_HMAC_SHA256;
1322
1323 if (OPTION_MPTCP_MPC_SYN & opts->suboptions) {
1324 len = TCPOLEN_MPTCP_MPC_SYN;
1325 } else if (OPTION_MPTCP_MPC_SYNACK & opts->suboptions) {
1326 len = TCPOLEN_MPTCP_MPC_SYNACK;
1327 } else if (opts->data_len) {
1328 len = TCPOLEN_MPTCP_MPC_ACK_DATA;
1329 if (opts->csum_reqd)
1330 len += TCPOLEN_MPTCP_DSS_CHECKSUM;
1331 } else {
1332 len = TCPOLEN_MPTCP_MPC_ACK;
1333 }
1334
1335 if (opts->csum_reqd)
1336 flag |= MPTCP_CAP_CHECKSUM_REQD;
1337
1338 if (!opts->allow_join_id0)
1339 flag |= MPTCP_CAP_DENY_JOIN_ID0;
1340
1341 *ptr++ = mptcp_option(MPTCPOPT_MP_CAPABLE, len,
1342 MPTCP_SUPPORTED_VERSION,
1343 flag);
1344
1345 if (!((OPTION_MPTCP_MPC_SYNACK | OPTION_MPTCP_MPC_ACK) &
1346 opts->suboptions))
1347 goto mp_capable_done;
1348
1349 put_unaligned_be64(opts->sndr_key, ptr);
1350 ptr += 2;
1351 if (!((OPTION_MPTCP_MPC_ACK) & opts->suboptions))
1352 goto mp_capable_done;
1353
1354 put_unaligned_be64(opts->rcvr_key, ptr);
1355 ptr += 2;
1356 if (!opts->data_len)
1357 goto mp_capable_done;
1358
1359 if (opts->csum_reqd) {
1360 put_unaligned_be32(opts->data_len << 16 |
1361 __mptcp_make_csum(opts->data_seq,
1362 opts->subflow_seq,
1363 opts->data_len,
1364 opts->csum), ptr);
1365 } else {
1366 put_unaligned_be32(opts->data_len << 16 |
1367 TCPOPT_NOP << 8 | TCPOPT_NOP, ptr);
1368 }
1369 ptr += 1;
1370
1371 /* MPC is additionally mutually exclusive with MP_PRIO */
1372 goto mp_capable_done;
1373 } else if (OPTION_MPTCP_MPJ_SYN & opts->suboptions) {
1374 *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1375 TCPOLEN_MPTCP_MPJ_SYN,
1376 opts->backup, opts->join_id);
1377 put_unaligned_be32(opts->token, ptr);
1378 ptr += 1;
1379 put_unaligned_be32(opts->nonce, ptr);
1380 ptr += 1;
1381 } else if (OPTION_MPTCP_MPJ_SYNACK & opts->suboptions) {
1382 *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1383 TCPOLEN_MPTCP_MPJ_SYNACK,
1384 opts->backup, opts->join_id);
1385 put_unaligned_be64(opts->thmac, ptr);
1386 ptr += 2;
1387 put_unaligned_be32(opts->nonce, ptr);
1388 ptr += 1;
1389 } else if (OPTION_MPTCP_MPJ_ACK & opts->suboptions) {
1390 *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1391 TCPOLEN_MPTCP_MPJ_ACK, 0, 0);
1392 memcpy(ptr, opts->hmac, MPTCPOPT_HMAC_LEN);
1393 ptr += 5;
1394 } else if (OPTION_MPTCP_ADD_ADDR & opts->suboptions) {
1395 u8 len = TCPOLEN_MPTCP_ADD_ADDR_BASE;
1396 u8 echo = MPTCP_ADDR_ECHO;
1397
1398 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1399 if (opts->addr.family == AF_INET6)
1400 len = TCPOLEN_MPTCP_ADD_ADDR6_BASE;
1401 #endif
1402
1403 if (opts->addr.port)
1404 len += TCPOLEN_MPTCP_PORT_LEN;
1405
1406 if (opts->ahmac) {
1407 len += sizeof(opts->ahmac);
1408 echo = 0;
1409 }
1410
1411 *ptr++ = mptcp_option(MPTCPOPT_ADD_ADDR,
1412 len, echo, opts->addr.id);
1413 if (opts->addr.family == AF_INET) {
1414 memcpy((u8 *)ptr, (u8 *)&opts->addr.addr.s_addr, 4);
1415 ptr += 1;
1416 }
1417 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1418 else if (opts->addr.family == AF_INET6) {
1419 memcpy((u8 *)ptr, opts->addr.addr6.s6_addr, 16);
1420 ptr += 4;
1421 }
1422 #endif
1423
1424 if (!opts->addr.port) {
1425 if (opts->ahmac) {
1426 put_unaligned_be64(opts->ahmac, ptr);
1427 ptr += 2;
1428 }
1429 } else {
1430 u16 port = ntohs(opts->addr.port);
1431
1432 if (opts->ahmac) {
1433 u8 *bptr = (u8 *)ptr;
1434
1435 put_unaligned_be16(port, bptr);
1436 bptr += 2;
1437 put_unaligned_be64(opts->ahmac, bptr);
1438 bptr += 8;
1439 put_unaligned_be16(TCPOPT_NOP << 8 |
1440 TCPOPT_NOP, bptr);
1441
1442 ptr += 3;
1443 } else {
1444 put_unaligned_be32(port << 16 |
1445 TCPOPT_NOP << 8 |
1446 TCPOPT_NOP, ptr);
1447 ptr += 1;
1448 }
1449 }
1450 }
1451
1452 if (OPTION_MPTCP_PRIO & opts->suboptions) {
1453 const struct sock *ssk = (const struct sock *)tp;
1454 struct mptcp_subflow_context *subflow;
1455
1456 subflow = mptcp_subflow_ctx(ssk);
1457 subflow->send_mp_prio = 0;
1458
1459 *ptr++ = mptcp_option(MPTCPOPT_MP_PRIO,
1460 TCPOLEN_MPTCP_PRIO,
1461 opts->backup, TCPOPT_NOP);
1462 }
1463
1464 mp_capable_done:
1465 if (OPTION_MPTCP_RM_ADDR & opts->suboptions) {
1466 u8 i = 1;
1467
1468 *ptr++ = mptcp_option(MPTCPOPT_RM_ADDR,
1469 TCPOLEN_MPTCP_RM_ADDR_BASE + opts->rm_list.nr,
1470 0, opts->rm_list.ids[0]);
1471
1472 while (i < opts->rm_list.nr) {
1473 u8 id1, id2, id3, id4;
1474
1475 id1 = opts->rm_list.ids[i];
1476 id2 = i + 1 < opts->rm_list.nr ? opts->rm_list.ids[i + 1] : TCPOPT_NOP;
1477 id3 = i + 2 < opts->rm_list.nr ? opts->rm_list.ids[i + 2] : TCPOPT_NOP;
1478 id4 = i + 3 < opts->rm_list.nr ? opts->rm_list.ids[i + 3] : TCPOPT_NOP;
1479 put_unaligned_be32(id1 << 24 | id2 << 16 | id3 << 8 | id4, ptr);
1480 ptr += 1;
1481 i += 4;
1482 }
1483 }
1484
1485 if (tp)
1486 mptcp_set_rwin(tp);
1487 }
1488
mptcp_get_reset_option(const struct sk_buff * skb)1489 __be32 mptcp_get_reset_option(const struct sk_buff *skb)
1490 {
1491 const struct mptcp_ext *ext = mptcp_get_ext(skb);
1492 u8 flags, reason;
1493
1494 if (ext) {
1495 flags = ext->reset_transient;
1496 reason = ext->reset_reason;
1497
1498 return mptcp_option(MPTCPOPT_RST, TCPOLEN_MPTCP_RST,
1499 flags, reason);
1500 }
1501
1502 return htonl(0u);
1503 }
1504 EXPORT_SYMBOL_GPL(mptcp_get_reset_option);
1505