1 // SPDX-License-Identifier: GPL-2.0
2 /* Multipath TCP
3 *
4 * Copyright (c) 2017 - 2019, Intel Corporation.
5 */
6
7 #define pr_fmt(fmt) "MPTCP: " fmt
8
9 #include <linux/kernel.h>
10 #include <crypto/sha.h>
11 #include <net/tcp.h>
12 #include <net/mptcp.h>
13 #include "protocol.h"
14 #include "mib.h"
15
mptcp_cap_flag_sha256(u8 flags)16 static bool mptcp_cap_flag_sha256(u8 flags)
17 {
18 return (flags & MPTCP_CAP_FLAG_MASK) == MPTCP_CAP_HMAC_SHA256;
19 }
20
mptcp_parse_option(const struct sk_buff * skb,const unsigned char * ptr,int opsize,struct mptcp_options_received * mp_opt)21 static void mptcp_parse_option(const struct sk_buff *skb,
22 const unsigned char *ptr, int opsize,
23 struct mptcp_options_received *mp_opt)
24 {
25 u8 subtype = *ptr >> 4;
26 int expected_opsize;
27 u8 version;
28 u8 flags;
29
30 switch (subtype) {
31 case MPTCPOPT_MP_CAPABLE:
32 /* strict size checking */
33 if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)) {
34 if (skb->len > tcp_hdr(skb)->doff << 2)
35 expected_opsize = TCPOLEN_MPTCP_MPC_ACK_DATA;
36 else
37 expected_opsize = TCPOLEN_MPTCP_MPC_ACK;
38 } else {
39 if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_ACK)
40 expected_opsize = TCPOLEN_MPTCP_MPC_SYNACK;
41 else
42 expected_opsize = TCPOLEN_MPTCP_MPC_SYN;
43 }
44 if (opsize != expected_opsize)
45 break;
46
47 /* try to be gentle vs future versions on the initial syn */
48 version = *ptr++ & MPTCP_VERSION_MASK;
49 if (opsize != TCPOLEN_MPTCP_MPC_SYN) {
50 if (version != MPTCP_SUPPORTED_VERSION)
51 break;
52 } else if (version < MPTCP_SUPPORTED_VERSION) {
53 break;
54 }
55
56 flags = *ptr++;
57 if (!mptcp_cap_flag_sha256(flags) ||
58 (flags & MPTCP_CAP_EXTENSIBILITY))
59 break;
60
61 /* RFC 6824, Section 3.1:
62 * "For the Checksum Required bit (labeled "A"), if either
63 * host requires the use of checksums, checksums MUST be used.
64 * In other words, the only way for checksums not to be used
65 * is if both hosts in their SYNs set A=0."
66 *
67 * Section 3.3.0:
68 * "If a checksum is not present when its use has been
69 * negotiated, the receiver MUST close the subflow with a RST as
70 * it is considered broken."
71 *
72 * We don't implement DSS checksum - fall back to TCP.
73 */
74 if (flags & MPTCP_CAP_CHECKSUM_REQD)
75 break;
76
77 mp_opt->mp_capable = 1;
78 if (opsize >= TCPOLEN_MPTCP_MPC_SYNACK) {
79 mp_opt->sndr_key = get_unaligned_be64(ptr);
80 ptr += 8;
81 }
82 if (opsize >= TCPOLEN_MPTCP_MPC_ACK) {
83 mp_opt->rcvr_key = get_unaligned_be64(ptr);
84 ptr += 8;
85 }
86 if (opsize == TCPOLEN_MPTCP_MPC_ACK_DATA) {
87 /* Section 3.1.:
88 * "the data parameters in a MP_CAPABLE are semantically
89 * equivalent to those in a DSS option and can be used
90 * interchangeably."
91 */
92 mp_opt->dss = 1;
93 mp_opt->use_map = 1;
94 mp_opt->mpc_map = 1;
95 mp_opt->data_len = get_unaligned_be16(ptr);
96 ptr += 2;
97 }
98 pr_debug("MP_CAPABLE version=%x, flags=%x, optlen=%d sndr=%llu, rcvr=%llu len=%d",
99 version, flags, opsize, mp_opt->sndr_key,
100 mp_opt->rcvr_key, mp_opt->data_len);
101 break;
102
103 case MPTCPOPT_MP_JOIN:
104 mp_opt->mp_join = 1;
105 if (opsize == TCPOLEN_MPTCP_MPJ_SYN) {
106 mp_opt->backup = *ptr++ & MPTCPOPT_BACKUP;
107 mp_opt->join_id = *ptr++;
108 mp_opt->token = get_unaligned_be32(ptr);
109 ptr += 4;
110 mp_opt->nonce = get_unaligned_be32(ptr);
111 ptr += 4;
112 pr_debug("MP_JOIN bkup=%u, id=%u, token=%u, nonce=%u",
113 mp_opt->backup, mp_opt->join_id,
114 mp_opt->token, mp_opt->nonce);
115 } else if (opsize == TCPOLEN_MPTCP_MPJ_SYNACK) {
116 mp_opt->backup = *ptr++ & MPTCPOPT_BACKUP;
117 mp_opt->join_id = *ptr++;
118 mp_opt->thmac = get_unaligned_be64(ptr);
119 ptr += 8;
120 mp_opt->nonce = get_unaligned_be32(ptr);
121 ptr += 4;
122 pr_debug("MP_JOIN bkup=%u, id=%u, thmac=%llu, nonce=%u",
123 mp_opt->backup, mp_opt->join_id,
124 mp_opt->thmac, mp_opt->nonce);
125 } else if (opsize == TCPOLEN_MPTCP_MPJ_ACK) {
126 ptr += 2;
127 memcpy(mp_opt->hmac, ptr, MPTCPOPT_HMAC_LEN);
128 pr_debug("MP_JOIN hmac");
129 } else {
130 pr_warn("MP_JOIN bad option size");
131 mp_opt->mp_join = 0;
132 }
133 break;
134
135 case MPTCPOPT_DSS:
136 pr_debug("DSS");
137 ptr++;
138
139 /* we must clear 'mpc_map' be able to detect MP_CAPABLE
140 * map vs DSS map in mptcp_incoming_options(), and reconstruct
141 * map info accordingly
142 */
143 mp_opt->mpc_map = 0;
144 flags = (*ptr++) & MPTCP_DSS_FLAG_MASK;
145 mp_opt->data_fin = (flags & MPTCP_DSS_DATA_FIN) != 0;
146 mp_opt->dsn64 = (flags & MPTCP_DSS_DSN64) != 0;
147 mp_opt->use_map = (flags & MPTCP_DSS_HAS_MAP) != 0;
148 mp_opt->ack64 = (flags & MPTCP_DSS_ACK64) != 0;
149 mp_opt->use_ack = (flags & MPTCP_DSS_HAS_ACK);
150
151 pr_debug("data_fin=%d dsn64=%d use_map=%d ack64=%d use_ack=%d",
152 mp_opt->data_fin, mp_opt->dsn64,
153 mp_opt->use_map, mp_opt->ack64,
154 mp_opt->use_ack);
155
156 expected_opsize = TCPOLEN_MPTCP_DSS_BASE;
157
158 if (mp_opt->use_ack) {
159 if (mp_opt->ack64)
160 expected_opsize += TCPOLEN_MPTCP_DSS_ACK64;
161 else
162 expected_opsize += TCPOLEN_MPTCP_DSS_ACK32;
163 }
164
165 if (mp_opt->use_map) {
166 if (mp_opt->dsn64)
167 expected_opsize += TCPOLEN_MPTCP_DSS_MAP64;
168 else
169 expected_opsize += TCPOLEN_MPTCP_DSS_MAP32;
170 }
171
172 /* RFC 6824, Section 3.3:
173 * If a checksum is present, but its use had
174 * not been negotiated in the MP_CAPABLE handshake,
175 * the checksum field MUST be ignored.
176 */
177 if (opsize != expected_opsize &&
178 opsize != expected_opsize + TCPOLEN_MPTCP_DSS_CHECKSUM)
179 break;
180
181 mp_opt->dss = 1;
182
183 if (mp_opt->use_ack) {
184 if (mp_opt->ack64) {
185 mp_opt->data_ack = get_unaligned_be64(ptr);
186 ptr += 8;
187 } else {
188 mp_opt->data_ack = get_unaligned_be32(ptr);
189 ptr += 4;
190 }
191
192 pr_debug("data_ack=%llu", mp_opt->data_ack);
193 }
194
195 if (mp_opt->use_map) {
196 if (mp_opt->dsn64) {
197 mp_opt->data_seq = get_unaligned_be64(ptr);
198 ptr += 8;
199 } else {
200 mp_opt->data_seq = get_unaligned_be32(ptr);
201 ptr += 4;
202 }
203
204 mp_opt->subflow_seq = get_unaligned_be32(ptr);
205 ptr += 4;
206
207 mp_opt->data_len = get_unaligned_be16(ptr);
208 ptr += 2;
209
210 pr_debug("data_seq=%llu subflow_seq=%u data_len=%u",
211 mp_opt->data_seq, mp_opt->subflow_seq,
212 mp_opt->data_len);
213 }
214
215 break;
216
217 case MPTCPOPT_ADD_ADDR:
218 mp_opt->echo = (*ptr++) & MPTCP_ADDR_ECHO;
219 if (!mp_opt->echo) {
220 if (opsize == TCPOLEN_MPTCP_ADD_ADDR ||
221 opsize == TCPOLEN_MPTCP_ADD_ADDR_PORT)
222 mp_opt->family = MPTCP_ADDR_IPVERSION_4;
223 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
224 else if (opsize == TCPOLEN_MPTCP_ADD_ADDR6 ||
225 opsize == TCPOLEN_MPTCP_ADD_ADDR6_PORT)
226 mp_opt->family = MPTCP_ADDR_IPVERSION_6;
227 #endif
228 else
229 break;
230 } else {
231 if (opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE ||
232 opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE_PORT)
233 mp_opt->family = MPTCP_ADDR_IPVERSION_4;
234 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
235 else if (opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE ||
236 opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE_PORT)
237 mp_opt->family = MPTCP_ADDR_IPVERSION_6;
238 #endif
239 else
240 break;
241 }
242
243 mp_opt->add_addr = 1;
244 mp_opt->addr_id = *ptr++;
245 pr_debug("ADD_ADDR: id=%d, echo=%d", mp_opt->addr_id, mp_opt->echo);
246 if (mp_opt->family == MPTCP_ADDR_IPVERSION_4) {
247 memcpy((u8 *)&mp_opt->addr.s_addr, (u8 *)ptr, 4);
248 ptr += 4;
249 if (opsize == TCPOLEN_MPTCP_ADD_ADDR_PORT ||
250 opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE_PORT) {
251 mp_opt->port = get_unaligned_be16(ptr);
252 ptr += 2;
253 }
254 }
255 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
256 else {
257 memcpy(mp_opt->addr6.s6_addr, (u8 *)ptr, 16);
258 ptr += 16;
259 if (opsize == TCPOLEN_MPTCP_ADD_ADDR6_PORT ||
260 opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE_PORT) {
261 mp_opt->port = get_unaligned_be16(ptr);
262 ptr += 2;
263 }
264 }
265 #endif
266 if (!mp_opt->echo) {
267 mp_opt->ahmac = get_unaligned_be64(ptr);
268 ptr += 8;
269 }
270 break;
271
272 case MPTCPOPT_RM_ADDR:
273 if (opsize != TCPOLEN_MPTCP_RM_ADDR_BASE)
274 break;
275
276 ptr++;
277
278 mp_opt->rm_addr = 1;
279 mp_opt->rm_id = *ptr++;
280 pr_debug("RM_ADDR: id=%d", mp_opt->rm_id);
281 break;
282
283 default:
284 break;
285 }
286 }
287
mptcp_get_options(const struct sk_buff * skb,struct mptcp_options_received * mp_opt)288 void mptcp_get_options(const struct sk_buff *skb,
289 struct mptcp_options_received *mp_opt)
290 {
291 const struct tcphdr *th = tcp_hdr(skb);
292 const unsigned char *ptr;
293 int length;
294
295 /* initialize option status */
296 mp_opt->mp_capable = 0;
297 mp_opt->mp_join = 0;
298 mp_opt->add_addr = 0;
299 mp_opt->ahmac = 0;
300 mp_opt->port = 0;
301 mp_opt->rm_addr = 0;
302 mp_opt->dss = 0;
303
304 length = (th->doff * 4) - sizeof(struct tcphdr);
305 ptr = (const unsigned char *)(th + 1);
306
307 while (length > 0) {
308 int opcode = *ptr++;
309 int opsize;
310
311 switch (opcode) {
312 case TCPOPT_EOL:
313 return;
314 case TCPOPT_NOP: /* Ref: RFC 793 section 3.1 */
315 length--;
316 continue;
317 default:
318 opsize = *ptr++;
319 if (opsize < 2) /* "silly options" */
320 return;
321 if (opsize > length)
322 return; /* don't parse partial options */
323 if (opcode == TCPOPT_MPTCP)
324 mptcp_parse_option(skb, ptr, opsize, mp_opt);
325 ptr += opsize - 2;
326 length -= opsize;
327 }
328 }
329 }
330
mptcp_syn_options(struct sock * sk,const struct sk_buff * skb,unsigned int * size,struct mptcp_out_options * opts)331 bool mptcp_syn_options(struct sock *sk, const struct sk_buff *skb,
332 unsigned int *size, struct mptcp_out_options *opts)
333 {
334 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
335
336 /* we will use snd_isn to detect first pkt [re]transmission
337 * in mptcp_established_options_mp()
338 */
339 subflow->snd_isn = TCP_SKB_CB(skb)->end_seq;
340 if (subflow->request_mptcp) {
341 opts->suboptions = OPTION_MPTCP_MPC_SYN;
342 *size = TCPOLEN_MPTCP_MPC_SYN;
343 return true;
344 } else if (subflow->request_join) {
345 pr_debug("remote_token=%u, nonce=%u", subflow->remote_token,
346 subflow->local_nonce);
347 opts->suboptions = OPTION_MPTCP_MPJ_SYN;
348 opts->join_id = subflow->local_id;
349 opts->token = subflow->remote_token;
350 opts->nonce = subflow->local_nonce;
351 opts->backup = subflow->request_bkup;
352 *size = TCPOLEN_MPTCP_MPJ_SYN;
353 return true;
354 }
355 return false;
356 }
357
358 /* MP_JOIN client subflow must wait for 4th ack before sending any data:
359 * TCP can't schedule delack timer before the subflow is fully established.
360 * MPTCP uses the delack timer to do 3rd ack retransmissions
361 */
schedule_3rdack_retransmission(struct sock * sk)362 static void schedule_3rdack_retransmission(struct sock *sk)
363 {
364 struct inet_connection_sock *icsk = inet_csk(sk);
365 struct tcp_sock *tp = tcp_sk(sk);
366 unsigned long timeout;
367
368 /* reschedule with a timeout above RTT, as we must look only for drop */
369 if (tp->srtt_us)
370 timeout = tp->srtt_us << 1;
371 else
372 timeout = TCP_TIMEOUT_INIT;
373
374 WARN_ON_ONCE(icsk->icsk_ack.pending & ICSK_ACK_TIMER);
375 icsk->icsk_ack.pending |= ICSK_ACK_SCHED | ICSK_ACK_TIMER;
376 icsk->icsk_ack.timeout = timeout;
377 sk_reset_timer(sk, &icsk->icsk_delack_timer, timeout);
378 }
379
clear_3rdack_retransmission(struct sock * sk)380 static void clear_3rdack_retransmission(struct sock *sk)
381 {
382 struct inet_connection_sock *icsk = inet_csk(sk);
383
384 sk_stop_timer(sk, &icsk->icsk_delack_timer);
385 icsk->icsk_ack.timeout = 0;
386 icsk->icsk_ack.ato = 0;
387 icsk->icsk_ack.pending &= ~(ICSK_ACK_SCHED | ICSK_ACK_TIMER);
388 }
389
mptcp_established_options_mp(struct sock * sk,struct sk_buff * skb,unsigned int * size,unsigned int remaining,struct mptcp_out_options * opts)390 static bool mptcp_established_options_mp(struct sock *sk, struct sk_buff *skb,
391 unsigned int *size,
392 unsigned int remaining,
393 struct mptcp_out_options *opts)
394 {
395 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
396 struct mptcp_ext *mpext;
397 unsigned int data_len;
398
399 /* When skb is not available, we better over-estimate the emitted
400 * options len. A full DSS option (28 bytes) is longer than
401 * TCPOLEN_MPTCP_MPC_ACK_DATA(22) or TCPOLEN_MPTCP_MPJ_ACK(24), so
402 * tell the caller to defer the estimate to
403 * mptcp_established_options_dss(), which will reserve enough space.
404 */
405 if (!skb)
406 return false;
407
408 /* MPC/MPJ needed only on 3rd ack packet */
409 if (subflow->fully_established ||
410 subflow->snd_isn != TCP_SKB_CB(skb)->seq)
411 return false;
412
413 if (subflow->mp_capable) {
414 mpext = mptcp_get_ext(skb);
415 data_len = mpext ? mpext->data_len : 0;
416
417 /* we will check ext_copy.data_len in mptcp_write_options() to
418 * discriminate between TCPOLEN_MPTCP_MPC_ACK_DATA and
419 * TCPOLEN_MPTCP_MPC_ACK
420 */
421 opts->ext_copy.data_len = data_len;
422 opts->suboptions = OPTION_MPTCP_MPC_ACK;
423 opts->sndr_key = subflow->local_key;
424 opts->rcvr_key = subflow->remote_key;
425
426 /* Section 3.1.
427 * The MP_CAPABLE option is carried on the SYN, SYN/ACK, and ACK
428 * packets that start the first subflow of an MPTCP connection,
429 * as well as the first packet that carries data
430 */
431 if (data_len > 0)
432 *size = ALIGN(TCPOLEN_MPTCP_MPC_ACK_DATA, 4);
433 else
434 *size = TCPOLEN_MPTCP_MPC_ACK;
435
436 pr_debug("subflow=%p, local_key=%llu, remote_key=%llu map_len=%d",
437 subflow, subflow->local_key, subflow->remote_key,
438 data_len);
439
440 return true;
441 } else if (subflow->mp_join) {
442 opts->suboptions = OPTION_MPTCP_MPJ_ACK;
443 memcpy(opts->hmac, subflow->hmac, MPTCPOPT_HMAC_LEN);
444 *size = TCPOLEN_MPTCP_MPJ_ACK;
445 pr_debug("subflow=%p", subflow);
446
447 schedule_3rdack_retransmission(sk);
448 return true;
449 }
450 return false;
451 }
452
mptcp_write_data_fin(struct mptcp_subflow_context * subflow,struct sk_buff * skb,struct mptcp_ext * ext)453 static void mptcp_write_data_fin(struct mptcp_subflow_context *subflow,
454 struct sk_buff *skb, struct mptcp_ext *ext)
455 {
456 /* The write_seq value has already been incremented, so the actual
457 * sequence number for the DATA_FIN is one less.
458 */
459 u64 data_fin_tx_seq = READ_ONCE(mptcp_sk(subflow->conn)->write_seq) - 1;
460
461 if (!ext->use_map || !skb->len) {
462 /* RFC6824 requires a DSS mapping with specific values
463 * if DATA_FIN is set but no data payload is mapped
464 */
465 ext->data_fin = 1;
466 ext->use_map = 1;
467 ext->dsn64 = 1;
468 ext->data_seq = data_fin_tx_seq;
469 ext->subflow_seq = 0;
470 ext->data_len = 1;
471 } else if (ext->data_seq + ext->data_len == data_fin_tx_seq) {
472 /* If there's an existing DSS mapping and it is the
473 * final mapping, DATA_FIN consumes 1 additional byte of
474 * mapping space.
475 */
476 ext->data_fin = 1;
477 ext->data_len++;
478 }
479 }
480
mptcp_established_options_dss(struct sock * sk,struct sk_buff * skb,unsigned int * size,unsigned int remaining,struct mptcp_out_options * opts)481 static bool mptcp_established_options_dss(struct sock *sk, struct sk_buff *skb,
482 unsigned int *size,
483 unsigned int remaining,
484 struct mptcp_out_options *opts)
485 {
486 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
487 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
488 unsigned int dss_size = 0;
489 u64 snd_data_fin_enable;
490 struct mptcp_ext *mpext;
491 unsigned int ack_size;
492 bool ret = false;
493
494 mpext = skb ? mptcp_get_ext(skb) : NULL;
495 snd_data_fin_enable = READ_ONCE(msk->snd_data_fin_enable);
496
497 if (!skb || (mpext && mpext->use_map) || snd_data_fin_enable) {
498 unsigned int map_size;
499
500 map_size = TCPOLEN_MPTCP_DSS_BASE + TCPOLEN_MPTCP_DSS_MAP64;
501
502 remaining -= map_size;
503 dss_size = map_size;
504 if (mpext)
505 opts->ext_copy = *mpext;
506
507 if (skb && snd_data_fin_enable)
508 mptcp_write_data_fin(subflow, skb, &opts->ext_copy);
509 ret = true;
510 }
511
512 /* passive sockets msk will set the 'can_ack' after accept(), even
513 * if the first subflow may have the already the remote key handy
514 */
515 opts->ext_copy.use_ack = 0;
516 if (!READ_ONCE(msk->can_ack)) {
517 *size = ALIGN(dss_size, 4);
518 return ret;
519 }
520
521 if (READ_ONCE(msk->use_64bit_ack)) {
522 ack_size = TCPOLEN_MPTCP_DSS_ACK64;
523 opts->ext_copy.data_ack = READ_ONCE(msk->ack_seq);
524 opts->ext_copy.ack64 = 1;
525 } else {
526 ack_size = TCPOLEN_MPTCP_DSS_ACK32;
527 opts->ext_copy.data_ack32 = (uint32_t)READ_ONCE(msk->ack_seq);
528 opts->ext_copy.ack64 = 0;
529 }
530 opts->ext_copy.use_ack = 1;
531
532 /* Add kind/length/subtype/flag overhead if mapping is not populated */
533 if (dss_size == 0)
534 ack_size += TCPOLEN_MPTCP_DSS_BASE;
535
536 dss_size += ack_size;
537
538 *size = ALIGN(dss_size, 4);
539 return true;
540 }
541
add_addr_generate_hmac(u64 key1,u64 key2,u8 addr_id,struct in_addr * addr)542 static u64 add_addr_generate_hmac(u64 key1, u64 key2, u8 addr_id,
543 struct in_addr *addr)
544 {
545 u8 hmac[SHA256_DIGEST_SIZE];
546 u8 msg[7];
547
548 msg[0] = addr_id;
549 memcpy(&msg[1], &addr->s_addr, 4);
550 msg[5] = 0;
551 msg[6] = 0;
552
553 mptcp_crypto_hmac_sha(key1, key2, msg, 7, hmac);
554
555 return get_unaligned_be64(&hmac[SHA256_DIGEST_SIZE - sizeof(u64)]);
556 }
557
558 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
add_addr6_generate_hmac(u64 key1,u64 key2,u8 addr_id,struct in6_addr * addr)559 static u64 add_addr6_generate_hmac(u64 key1, u64 key2, u8 addr_id,
560 struct in6_addr *addr)
561 {
562 u8 hmac[SHA256_DIGEST_SIZE];
563 u8 msg[19];
564
565 msg[0] = addr_id;
566 memcpy(&msg[1], &addr->s6_addr, 16);
567 msg[17] = 0;
568 msg[18] = 0;
569
570 mptcp_crypto_hmac_sha(key1, key2, msg, 19, hmac);
571
572 return get_unaligned_be64(&hmac[SHA256_DIGEST_SIZE - sizeof(u64)]);
573 }
574 #endif
575
mptcp_established_options_add_addr(struct sock * sk,unsigned int * size,unsigned int remaining,struct mptcp_out_options * opts)576 static bool mptcp_established_options_add_addr(struct sock *sk,
577 unsigned int *size,
578 unsigned int remaining,
579 struct mptcp_out_options *opts)
580 {
581 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
582 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
583 struct mptcp_addr_info saddr;
584 bool echo;
585 int len;
586
587 if (!mptcp_pm_should_add_signal(msk) ||
588 !(mptcp_pm_add_addr_signal(msk, remaining, &saddr, &echo)))
589 return false;
590
591 len = mptcp_add_addr_len(saddr.family, echo);
592 if (remaining < len)
593 return false;
594
595 *size = len;
596 opts->addr_id = saddr.id;
597 if (saddr.family == AF_INET) {
598 opts->suboptions |= OPTION_MPTCP_ADD_ADDR;
599 opts->addr = saddr.addr;
600 if (!echo) {
601 opts->ahmac = add_addr_generate_hmac(msk->local_key,
602 msk->remote_key,
603 opts->addr_id,
604 &opts->addr);
605 }
606 }
607 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
608 else if (saddr.family == AF_INET6) {
609 opts->suboptions |= OPTION_MPTCP_ADD_ADDR6;
610 opts->addr6 = saddr.addr6;
611 if (!echo) {
612 opts->ahmac = add_addr6_generate_hmac(msk->local_key,
613 msk->remote_key,
614 opts->addr_id,
615 &opts->addr6);
616 }
617 }
618 #endif
619 pr_debug("addr_id=%d, ahmac=%llu, echo=%d", opts->addr_id, opts->ahmac, echo);
620
621 return true;
622 }
623
mptcp_established_options_rm_addr(struct sock * sk,unsigned int * size,unsigned int remaining,struct mptcp_out_options * opts)624 static bool mptcp_established_options_rm_addr(struct sock *sk,
625 unsigned int *size,
626 unsigned int remaining,
627 struct mptcp_out_options *opts)
628 {
629 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
630 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
631 u8 rm_id;
632
633 if (!mptcp_pm_should_rm_signal(msk) ||
634 !(mptcp_pm_rm_addr_signal(msk, remaining, &rm_id)))
635 return false;
636
637 if (remaining < TCPOLEN_MPTCP_RM_ADDR_BASE)
638 return false;
639
640 *size = TCPOLEN_MPTCP_RM_ADDR_BASE;
641 opts->suboptions |= OPTION_MPTCP_RM_ADDR;
642 opts->rm_id = rm_id;
643
644 pr_debug("rm_id=%d", opts->rm_id);
645
646 return true;
647 }
648
mptcp_established_options(struct sock * sk,struct sk_buff * skb,unsigned int * size,unsigned int remaining,struct mptcp_out_options * opts)649 bool mptcp_established_options(struct sock *sk, struct sk_buff *skb,
650 unsigned int *size, unsigned int remaining,
651 struct mptcp_out_options *opts)
652 {
653 unsigned int opt_size = 0;
654 bool ret = false;
655
656 opts->suboptions = 0;
657
658 if (unlikely(mptcp_check_fallback(sk)))
659 return false;
660
661 /* prevent adding of any MPTCP related options on reset packet
662 * until we support MP_TCPRST/MP_FASTCLOSE
663 */
664 if (unlikely(skb && TCP_SKB_CB(skb)->tcp_flags & TCPHDR_RST))
665 return false;
666
667 if (mptcp_established_options_mp(sk, skb, &opt_size, remaining, opts))
668 ret = true;
669 else if (mptcp_established_options_dss(sk, skb, &opt_size, remaining,
670 opts))
671 ret = true;
672
673 /* we reserved enough space for the above options, and exceeding the
674 * TCP option space would be fatal
675 */
676 if (WARN_ON_ONCE(opt_size > remaining))
677 return false;
678
679 *size += opt_size;
680 remaining -= opt_size;
681 if (mptcp_established_options_add_addr(sk, &opt_size, remaining, opts)) {
682 *size += opt_size;
683 remaining -= opt_size;
684 ret = true;
685 } else if (mptcp_established_options_rm_addr(sk, &opt_size, remaining, opts)) {
686 *size += opt_size;
687 remaining -= opt_size;
688 ret = true;
689 }
690
691 return ret;
692 }
693
mptcp_synack_options(const struct request_sock * req,unsigned int * size,struct mptcp_out_options * opts)694 bool mptcp_synack_options(const struct request_sock *req, unsigned int *size,
695 struct mptcp_out_options *opts)
696 {
697 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
698
699 if (subflow_req->mp_capable) {
700 opts->suboptions = OPTION_MPTCP_MPC_SYNACK;
701 opts->sndr_key = subflow_req->local_key;
702 *size = TCPOLEN_MPTCP_MPC_SYNACK;
703 pr_debug("subflow_req=%p, local_key=%llu",
704 subflow_req, subflow_req->local_key);
705 return true;
706 } else if (subflow_req->mp_join) {
707 opts->suboptions = OPTION_MPTCP_MPJ_SYNACK;
708 opts->backup = subflow_req->backup;
709 opts->join_id = subflow_req->local_id;
710 opts->thmac = subflow_req->thmac;
711 opts->nonce = subflow_req->local_nonce;
712 pr_debug("req=%p, bkup=%u, id=%u, thmac=%llu, nonce=%u",
713 subflow_req, opts->backup, opts->join_id,
714 opts->thmac, opts->nonce);
715 *size = TCPOLEN_MPTCP_MPJ_SYNACK;
716 return true;
717 }
718 return false;
719 }
720
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)721 static bool check_fully_established(struct mptcp_sock *msk, struct sock *ssk,
722 struct mptcp_subflow_context *subflow,
723 struct sk_buff *skb,
724 struct mptcp_options_received *mp_opt)
725 {
726 /* here we can process OoO, in-window pkts, only in-sequence 4th ack
727 * will make the subflow fully established
728 */
729 if (likely(subflow->fully_established)) {
730 /* on passive sockets, check for 3rd ack retransmission
731 * note that msk is always set by subflow_syn_recv_sock()
732 * for mp_join subflows
733 */
734 if (TCP_SKB_CB(skb)->seq == subflow->ssn_offset + 1 &&
735 TCP_SKB_CB(skb)->end_seq == TCP_SKB_CB(skb)->seq &&
736 subflow->mp_join && mp_opt->mp_join &&
737 READ_ONCE(msk->pm.server_side))
738 tcp_send_ack(ssk);
739 goto fully_established;
740 }
741
742 /* we must process OoO packets before the first subflow is fully
743 * established. OoO packets are instead a protocol violation
744 * for MP_JOIN subflows as the peer must not send any data
745 * before receiving the forth ack - cfr. RFC 8684 section 3.2.
746 */
747 if (TCP_SKB_CB(skb)->seq != subflow->ssn_offset + 1) {
748 if (subflow->mp_join)
749 goto reset;
750 return subflow->mp_capable;
751 }
752
753 if (mp_opt->dss && mp_opt->use_ack) {
754 /* subflows are fully established as soon as we get any
755 * additional ack.
756 */
757 subflow->fully_established = 1;
758 WRITE_ONCE(msk->fully_established, true);
759 goto fully_established;
760 }
761
762 /* If the first established packet does not contain MP_CAPABLE + data
763 * then fallback to TCP. Fallback scenarios requires a reset for
764 * MP_JOIN subflows.
765 */
766 if (!mp_opt->mp_capable) {
767 if (subflow->mp_join)
768 goto reset;
769 subflow->mp_capable = 0;
770 pr_fallback(msk);
771 __mptcp_do_fallback(msk);
772 return false;
773 }
774
775 if (unlikely(!READ_ONCE(msk->pm.server_side)))
776 pr_warn_once("bogus mpc option on established client sk");
777 mptcp_subflow_fully_established(subflow, mp_opt);
778
779 fully_established:
780 if (likely(subflow->pm_notified))
781 return true;
782
783 subflow->pm_notified = 1;
784 if (subflow->mp_join) {
785 clear_3rdack_retransmission(ssk);
786 mptcp_pm_subflow_established(msk, subflow);
787 } else {
788 mptcp_pm_fully_established(msk);
789 }
790 return true;
791
792 reset:
793 mptcp_subflow_reset(ssk);
794 return false;
795 }
796
expand_ack(u64 old_ack,u64 cur_ack,bool use_64bit)797 static u64 expand_ack(u64 old_ack, u64 cur_ack, bool use_64bit)
798 {
799 u32 old_ack32, cur_ack32;
800
801 if (use_64bit)
802 return cur_ack;
803
804 old_ack32 = (u32)old_ack;
805 cur_ack32 = (u32)cur_ack;
806 cur_ack = (old_ack & GENMASK_ULL(63, 32)) + cur_ack32;
807 if (unlikely(before(cur_ack32, old_ack32)))
808 return cur_ack + (1LL << 32);
809 return cur_ack;
810 }
811
update_una(struct mptcp_sock * msk,struct mptcp_options_received * mp_opt)812 static void update_una(struct mptcp_sock *msk,
813 struct mptcp_options_received *mp_opt)
814 {
815 u64 new_snd_una, snd_una, old_snd_una = atomic64_read(&msk->snd_una);
816 u64 write_seq = READ_ONCE(msk->write_seq);
817
818 /* avoid ack expansion on update conflict, to reduce the risk of
819 * wrongly expanding to a future ack sequence number, which is way
820 * more dangerous than missing an ack
821 */
822 new_snd_una = expand_ack(old_snd_una, mp_opt->data_ack, mp_opt->ack64);
823
824 /* ACK for data not even sent yet? Ignore. */
825 if (after64(new_snd_una, write_seq))
826 new_snd_una = old_snd_una;
827
828 while (after64(new_snd_una, old_snd_una)) {
829 snd_una = old_snd_una;
830 old_snd_una = atomic64_cmpxchg(&msk->snd_una, snd_una,
831 new_snd_una);
832 if (old_snd_una == snd_una) {
833 mptcp_data_acked((struct sock *)msk);
834 break;
835 }
836 }
837 }
838
mptcp_update_rcv_data_fin(struct mptcp_sock * msk,u64 data_fin_seq,bool use_64bit)839 bool mptcp_update_rcv_data_fin(struct mptcp_sock *msk, u64 data_fin_seq, bool use_64bit)
840 {
841 /* Skip if DATA_FIN was already received.
842 * If updating simultaneously with the recvmsg loop, values
843 * should match. If they mismatch, the peer is misbehaving and
844 * we will prefer the most recent information.
845 */
846 if (READ_ONCE(msk->rcv_data_fin) || !READ_ONCE(msk->first))
847 return false;
848
849 WRITE_ONCE(msk->rcv_data_fin_seq,
850 expand_ack(READ_ONCE(msk->ack_seq), data_fin_seq, use_64bit));
851 WRITE_ONCE(msk->rcv_data_fin, 1);
852
853 return true;
854 }
855
add_addr_hmac_valid(struct mptcp_sock * msk,struct mptcp_options_received * mp_opt)856 static bool add_addr_hmac_valid(struct mptcp_sock *msk,
857 struct mptcp_options_received *mp_opt)
858 {
859 u64 hmac = 0;
860
861 if (mp_opt->echo)
862 return true;
863
864 if (mp_opt->family == MPTCP_ADDR_IPVERSION_4)
865 hmac = add_addr_generate_hmac(msk->remote_key,
866 msk->local_key,
867 mp_opt->addr_id, &mp_opt->addr);
868 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
869 else
870 hmac = add_addr6_generate_hmac(msk->remote_key,
871 msk->local_key,
872 mp_opt->addr_id, &mp_opt->addr6);
873 #endif
874
875 pr_debug("msk=%p, ahmac=%llu, mp_opt->ahmac=%llu\n",
876 msk, (unsigned long long)hmac,
877 (unsigned long long)mp_opt->ahmac);
878
879 return hmac == mp_opt->ahmac;
880 }
881
mptcp_incoming_options(struct sock * sk,struct sk_buff * skb)882 void mptcp_incoming_options(struct sock *sk, struct sk_buff *skb)
883 {
884 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
885 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
886 struct mptcp_options_received mp_opt;
887 struct mptcp_ext *mpext;
888
889 if (__mptcp_check_fallback(msk))
890 return;
891
892 mptcp_get_options(skb, &mp_opt);
893 if (!check_fully_established(msk, sk, subflow, skb, &mp_opt))
894 return;
895
896 if (mp_opt.add_addr && add_addr_hmac_valid(msk, &mp_opt)) {
897 struct mptcp_addr_info addr;
898
899 addr.port = htons(mp_opt.port);
900 addr.id = mp_opt.addr_id;
901 if (mp_opt.family == MPTCP_ADDR_IPVERSION_4) {
902 addr.family = AF_INET;
903 addr.addr = mp_opt.addr;
904 }
905 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
906 else if (mp_opt.family == MPTCP_ADDR_IPVERSION_6) {
907 addr.family = AF_INET6;
908 addr.addr6 = mp_opt.addr6;
909 }
910 #endif
911 if (!mp_opt.echo) {
912 mptcp_pm_add_addr_received(msk, &addr);
913 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ADDADDR);
914 } else {
915 mptcp_pm_del_add_timer(msk, &addr);
916 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ECHOADD);
917 }
918 mp_opt.add_addr = 0;
919 }
920
921 if (mp_opt.rm_addr) {
922 mptcp_pm_rm_addr_received(msk, mp_opt.rm_id);
923 mp_opt.rm_addr = 0;
924 }
925
926 if (!mp_opt.dss)
927 return;
928
929 /* we can't wait for recvmsg() to update the ack_seq, otherwise
930 * monodirectional flows will stuck
931 */
932 if (mp_opt.use_ack)
933 update_una(msk, &mp_opt);
934
935 /* Zero-data-length packets are dropped by the caller and not
936 * propagated to the MPTCP layer, so the skb extension does not
937 * need to be allocated or populated. DATA_FIN information, if
938 * present, needs to be updated here before the skb is freed.
939 */
940 if (TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb)->end_seq) {
941 if (mp_opt.data_fin && mp_opt.data_len == 1 &&
942 mptcp_update_rcv_data_fin(msk, mp_opt.data_seq, mp_opt.dsn64) &&
943 schedule_work(&msk->work))
944 sock_hold(subflow->conn);
945
946 return;
947 }
948
949 mpext = skb_ext_add(skb, SKB_EXT_MPTCP);
950 if (!mpext)
951 return;
952
953 memset(mpext, 0, sizeof(*mpext));
954
955 if (mp_opt.use_map) {
956 if (mp_opt.mpc_map) {
957 /* this is an MP_CAPABLE carrying MPTCP data
958 * we know this map the first chunk of data
959 */
960 mptcp_crypto_key_sha(subflow->remote_key, NULL,
961 &mpext->data_seq);
962 mpext->data_seq++;
963 mpext->subflow_seq = 1;
964 mpext->dsn64 = 1;
965 mpext->mpc_map = 1;
966 mpext->data_fin = 0;
967 } else {
968 mpext->data_seq = mp_opt.data_seq;
969 mpext->subflow_seq = mp_opt.subflow_seq;
970 mpext->dsn64 = mp_opt.dsn64;
971 mpext->data_fin = mp_opt.data_fin;
972 }
973 mpext->data_len = mp_opt.data_len;
974 mpext->use_map = 1;
975 }
976 }
977
mptcp_write_options(__be32 * ptr,struct mptcp_out_options * opts)978 void mptcp_write_options(__be32 *ptr, struct mptcp_out_options *opts)
979 {
980 if ((OPTION_MPTCP_MPC_SYN | OPTION_MPTCP_MPC_SYNACK |
981 OPTION_MPTCP_MPC_ACK) & opts->suboptions) {
982 u8 len;
983
984 if (OPTION_MPTCP_MPC_SYN & opts->suboptions)
985 len = TCPOLEN_MPTCP_MPC_SYN;
986 else if (OPTION_MPTCP_MPC_SYNACK & opts->suboptions)
987 len = TCPOLEN_MPTCP_MPC_SYNACK;
988 else if (opts->ext_copy.data_len)
989 len = TCPOLEN_MPTCP_MPC_ACK_DATA;
990 else
991 len = TCPOLEN_MPTCP_MPC_ACK;
992
993 *ptr++ = mptcp_option(MPTCPOPT_MP_CAPABLE, len,
994 MPTCP_SUPPORTED_VERSION,
995 MPTCP_CAP_HMAC_SHA256);
996
997 if (!((OPTION_MPTCP_MPC_SYNACK | OPTION_MPTCP_MPC_ACK) &
998 opts->suboptions))
999 goto mp_capable_done;
1000
1001 put_unaligned_be64(opts->sndr_key, ptr);
1002 ptr += 2;
1003 if (!((OPTION_MPTCP_MPC_ACK) & opts->suboptions))
1004 goto mp_capable_done;
1005
1006 put_unaligned_be64(opts->rcvr_key, ptr);
1007 ptr += 2;
1008 if (!opts->ext_copy.data_len)
1009 goto mp_capable_done;
1010
1011 put_unaligned_be32(opts->ext_copy.data_len << 16 |
1012 TCPOPT_NOP << 8 | TCPOPT_NOP, ptr);
1013 ptr += 1;
1014 }
1015
1016 mp_capable_done:
1017 if (OPTION_MPTCP_ADD_ADDR & opts->suboptions) {
1018 if (opts->ahmac)
1019 *ptr++ = mptcp_option(MPTCPOPT_ADD_ADDR,
1020 TCPOLEN_MPTCP_ADD_ADDR, 0,
1021 opts->addr_id);
1022 else
1023 *ptr++ = mptcp_option(MPTCPOPT_ADD_ADDR,
1024 TCPOLEN_MPTCP_ADD_ADDR_BASE,
1025 MPTCP_ADDR_ECHO,
1026 opts->addr_id);
1027 memcpy((u8 *)ptr, (u8 *)&opts->addr.s_addr, 4);
1028 ptr += 1;
1029 if (opts->ahmac) {
1030 put_unaligned_be64(opts->ahmac, ptr);
1031 ptr += 2;
1032 }
1033 }
1034
1035 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1036 if (OPTION_MPTCP_ADD_ADDR6 & opts->suboptions) {
1037 if (opts->ahmac)
1038 *ptr++ = mptcp_option(MPTCPOPT_ADD_ADDR,
1039 TCPOLEN_MPTCP_ADD_ADDR6, 0,
1040 opts->addr_id);
1041 else
1042 *ptr++ = mptcp_option(MPTCPOPT_ADD_ADDR,
1043 TCPOLEN_MPTCP_ADD_ADDR6_BASE,
1044 MPTCP_ADDR_ECHO,
1045 opts->addr_id);
1046 memcpy((u8 *)ptr, opts->addr6.s6_addr, 16);
1047 ptr += 4;
1048 if (opts->ahmac) {
1049 put_unaligned_be64(opts->ahmac, ptr);
1050 ptr += 2;
1051 }
1052 }
1053 #endif
1054
1055 if (OPTION_MPTCP_RM_ADDR & opts->suboptions) {
1056 *ptr++ = mptcp_option(MPTCPOPT_RM_ADDR,
1057 TCPOLEN_MPTCP_RM_ADDR_BASE,
1058 0, opts->rm_id);
1059 }
1060
1061 if (OPTION_MPTCP_MPJ_SYN & opts->suboptions) {
1062 *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1063 TCPOLEN_MPTCP_MPJ_SYN,
1064 opts->backup, opts->join_id);
1065 put_unaligned_be32(opts->token, ptr);
1066 ptr += 1;
1067 put_unaligned_be32(opts->nonce, ptr);
1068 ptr += 1;
1069 }
1070
1071 if (OPTION_MPTCP_MPJ_SYNACK & opts->suboptions) {
1072 *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1073 TCPOLEN_MPTCP_MPJ_SYNACK,
1074 opts->backup, opts->join_id);
1075 put_unaligned_be64(opts->thmac, ptr);
1076 ptr += 2;
1077 put_unaligned_be32(opts->nonce, ptr);
1078 ptr += 1;
1079 }
1080
1081 if (OPTION_MPTCP_MPJ_ACK & opts->suboptions) {
1082 *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1083 TCPOLEN_MPTCP_MPJ_ACK, 0, 0);
1084 memcpy(ptr, opts->hmac, MPTCPOPT_HMAC_LEN);
1085 ptr += 5;
1086 }
1087
1088 if (opts->ext_copy.use_ack || opts->ext_copy.use_map) {
1089 struct mptcp_ext *mpext = &opts->ext_copy;
1090 u8 len = TCPOLEN_MPTCP_DSS_BASE;
1091 u8 flags = 0;
1092
1093 if (mpext->use_ack) {
1094 flags = MPTCP_DSS_HAS_ACK;
1095 if (mpext->ack64) {
1096 len += TCPOLEN_MPTCP_DSS_ACK64;
1097 flags |= MPTCP_DSS_ACK64;
1098 } else {
1099 len += TCPOLEN_MPTCP_DSS_ACK32;
1100 }
1101 }
1102
1103 if (mpext->use_map) {
1104 len += TCPOLEN_MPTCP_DSS_MAP64;
1105
1106 /* Use only 64-bit mapping flags for now, add
1107 * support for optional 32-bit mappings later.
1108 */
1109 flags |= MPTCP_DSS_HAS_MAP | MPTCP_DSS_DSN64;
1110 if (mpext->data_fin)
1111 flags |= MPTCP_DSS_DATA_FIN;
1112 }
1113
1114 *ptr++ = mptcp_option(MPTCPOPT_DSS, len, 0, flags);
1115
1116 if (mpext->use_ack) {
1117 if (mpext->ack64) {
1118 put_unaligned_be64(mpext->data_ack, ptr);
1119 ptr += 2;
1120 } else {
1121 put_unaligned_be32(mpext->data_ack32, ptr);
1122 ptr += 1;
1123 }
1124 }
1125
1126 if (mpext->use_map) {
1127 put_unaligned_be64(mpext->data_seq, ptr);
1128 ptr += 2;
1129 put_unaligned_be32(mpext->subflow_seq, ptr);
1130 ptr += 1;
1131 put_unaligned_be32(mpext->data_len << 16 |
1132 TCPOPT_NOP << 8 | TCPOPT_NOP, ptr);
1133 }
1134 }
1135 }
1136
