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