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 <linux/module.h>
11 #include <linux/netdevice.h>
12 #include <crypto/sha2.h>
13 #include <crypto/utils.h>
14 #include <net/sock.h>
15 #include <net/inet_common.h>
16 #include <net/inet_hashtables.h>
17 #include <net/protocol.h>
18 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
19 #include <net/ip6_route.h>
20 #include <net/transp_v6.h>
21 #endif
22 #include <net/mptcp.h>
23
24 #include "protocol.h"
25 #include "mib.h"
26
27 #include <trace/events/mptcp.h>
28 #include <trace/events/sock.h>
29
30 static void mptcp_subflow_ops_undo_override(struct sock *ssk);
31
32 static void SUBFLOW_REQ_INC_STATS(struct request_sock *req,
33 enum linux_mptcp_mib_field field)
34 {
35 MPTCP_INC_STATS(sock_net(req_to_sk(req)), field);
36 }
37
38 static void subflow_req_destructor(struct request_sock *req)
39 {
40 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
41
42 pr_debug("subflow_req=%p\n", subflow_req);
43
44 if (subflow_req->msk)
45 sock_put((struct sock *)subflow_req->msk);
46
47 mptcp_token_destroy_request(req);
48 }
49
50 static void subflow_generate_hmac(u64 key1, u64 key2, u32 nonce1, u32 nonce2,
51 void *hmac)
52 {
53 u8 msg[8];
54
55 put_unaligned_be32(nonce1, &msg[0]);
56 put_unaligned_be32(nonce2, &msg[4]);
57
58 mptcp_crypto_hmac_sha(key1, key2, msg, 8, hmac);
59 }
60
61 static bool mptcp_can_accept_new_subflow(const struct mptcp_sock *msk)
62 {
63 return mptcp_is_fully_established((void *)msk) &&
64 ((mptcp_pm_is_userspace(msk) &&
65 mptcp_userspace_pm_active(msk)) ||
66 READ_ONCE(msk->pm.accept_subflow));
67 }
68
69 /* validate received token and create truncated hmac and nonce for SYN-ACK */
70 static void subflow_req_create_thmac(struct mptcp_subflow_request_sock *subflow_req)
71 {
72 struct mptcp_sock *msk = subflow_req->msk;
73 u8 hmac[SHA256_DIGEST_SIZE];
74
75 get_random_bytes(&subflow_req->local_nonce, sizeof(u32));
76
77 subflow_generate_hmac(READ_ONCE(msk->local_key),
78 READ_ONCE(msk->remote_key),
79 subflow_req->local_nonce,
80 subflow_req->remote_nonce, hmac);
81
82 subflow_req->thmac = get_unaligned_be64(hmac);
83 }
84
85 static struct mptcp_sock *subflow_token_join_request(struct request_sock *req)
86 {
87 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
88 struct mptcp_sock *msk;
89 int local_id;
90
91 msk = mptcp_token_get_sock(sock_net(req_to_sk(req)), subflow_req->token);
92 if (!msk) {
93 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINNOTOKEN);
94 return NULL;
95 }
96
97 local_id = mptcp_pm_get_local_id(msk, (struct sock_common *)req);
98 if (local_id < 0) {
99 sock_put((struct sock *)msk);
100 return NULL;
101 }
102 subflow_req->local_id = local_id;
103 subflow_req->request_bkup = mptcp_pm_is_backup(msk, (struct sock_common *)req);
104
105 return msk;
106 }
107
108 static void subflow_init_req(struct request_sock *req, const struct sock *sk_listener)
109 {
110 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
111
112 subflow_req->mp_capable = 0;
113 subflow_req->mp_join = 0;
114 subflow_req->csum_reqd = mptcp_is_checksum_enabled(sock_net(sk_listener));
115 subflow_req->allow_join_id0 = mptcp_allow_join_id0(sock_net(sk_listener));
116 subflow_req->msk = NULL;
117 mptcp_token_init_request(req);
118 }
119
120 static bool subflow_use_different_sport(struct mptcp_sock *msk, const struct sock *sk)
121 {
122 return inet_sk(sk)->inet_sport != inet_sk((struct sock *)msk)->inet_sport;
123 }
124
125 static void subflow_add_reset_reason(struct sk_buff *skb, u8 reason)
126 {
127 struct mptcp_ext *mpext = skb_ext_add(skb, SKB_EXT_MPTCP);
128
129 if (mpext) {
130 memset(mpext, 0, sizeof(*mpext));
131 mpext->reset_reason = reason;
132 }
133 }
134
135 static int subflow_reset_req_endp(struct request_sock *req, struct sk_buff *skb)
136 {
137 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MPCAPABLEENDPATTEMPT);
138 subflow_add_reset_reason(skb, MPTCP_RST_EPROHIBIT);
139 return -EPERM;
140 }
141
142 /* Init mptcp request socket.
143 *
144 * Returns an error code if a JOIN has failed and a TCP reset
145 * should be sent.
146 */
147 static int subflow_check_req(struct request_sock *req,
148 const struct sock *sk_listener,
149 struct sk_buff *skb)
150 {
151 struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk_listener);
152 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
153 struct mptcp_options_received mp_opt;
154 bool opt_mp_capable, opt_mp_join;
155
156 pr_debug("subflow_req=%p, listener=%p\n", subflow_req, listener);
157
158 #ifdef CONFIG_TCP_MD5SIG
159 /* no MPTCP if MD5SIG is enabled on this socket or we may run out of
160 * TCP option space.
161 */
162 if (rcu_access_pointer(tcp_sk(sk_listener)->md5sig_info)) {
163 subflow_add_reset_reason(skb, MPTCP_RST_EMPTCP);
164 return -EINVAL;
165 }
166 #endif
167
168 mptcp_get_options(skb, &mp_opt);
169
170 opt_mp_capable = !!(mp_opt.suboptions & OPTION_MPTCP_MPC_SYN);
171 opt_mp_join = !!(mp_opt.suboptions & OPTION_MPTCP_MPJ_SYN);
172 if (opt_mp_capable) {
173 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MPCAPABLEPASSIVE);
174
175 if (unlikely(listener->pm_listener))
176 return subflow_reset_req_endp(req, skb);
177 if (opt_mp_join)
178 return 0;
179 } else if (opt_mp_join) {
180 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINSYNRX);
181
182 if (mp_opt.backup)
183 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINSYNBACKUPRX);
184 } else if (unlikely(listener->pm_listener)) {
185 return subflow_reset_req_endp(req, skb);
186 }
187
188 if (opt_mp_capable && listener->request_mptcp) {
189 int err, retries = MPTCP_TOKEN_MAX_RETRIES;
190
191 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq;
192 again:
193 do {
194 get_random_bytes(&subflow_req->local_key, sizeof(subflow_req->local_key));
195 } while (subflow_req->local_key == 0);
196
197 if (unlikely(req->syncookie)) {
198 mptcp_crypto_key_sha(subflow_req->local_key,
199 &subflow_req->token,
200 &subflow_req->idsn);
201 if (mptcp_token_exists(subflow_req->token)) {
202 if (retries-- > 0)
203 goto again;
204 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_TOKENFALLBACKINIT);
205 } else {
206 subflow_req->mp_capable = 1;
207 }
208 return 0;
209 }
210
211 err = mptcp_token_new_request(req);
212 if (err == 0)
213 subflow_req->mp_capable = 1;
214 else if (retries-- > 0)
215 goto again;
216 else
217 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_TOKENFALLBACKINIT);
218
219 } else if (opt_mp_join && listener->request_mptcp) {
220 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq;
221 subflow_req->mp_join = 1;
222 subflow_req->backup = mp_opt.backup;
223 subflow_req->remote_id = mp_opt.join_id;
224 subflow_req->token = mp_opt.token;
225 subflow_req->remote_nonce = mp_opt.nonce;
226 subflow_req->msk = subflow_token_join_request(req);
227
228 /* Can't fall back to TCP in this case. */
229 if (!subflow_req->msk) {
230 subflow_add_reset_reason(skb, MPTCP_RST_EMPTCP);
231 return -EPERM;
232 }
233
234 if (subflow_use_different_sport(subflow_req->msk, sk_listener)) {
235 pr_debug("syn inet_sport=%d %d\n",
236 ntohs(inet_sk(sk_listener)->inet_sport),
237 ntohs(inet_sk((struct sock *)subflow_req->msk)->inet_sport));
238 if (!mptcp_pm_sport_in_anno_list(subflow_req->msk, sk_listener)) {
239 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MISMATCHPORTSYNRX);
240 subflow_add_reset_reason(skb, MPTCP_RST_EPROHIBIT);
241 return -EPERM;
242 }
243 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINPORTSYNRX);
244 }
245
246 subflow_req_create_thmac(subflow_req);
247
248 if (unlikely(req->syncookie)) {
249 if (!mptcp_can_accept_new_subflow(subflow_req->msk)) {
250 subflow_add_reset_reason(skb, MPTCP_RST_EPROHIBIT);
251 return -EPERM;
252 }
253
254 subflow_init_req_cookie_join_save(subflow_req, skb);
255 }
256
257 pr_debug("token=%u, remote_nonce=%u msk=%p\n", subflow_req->token,
258 subflow_req->remote_nonce, subflow_req->msk);
259 }
260
261 return 0;
262 }
263
264 int mptcp_subflow_init_cookie_req(struct request_sock *req,
265 const struct sock *sk_listener,
266 struct sk_buff *skb)
267 {
268 struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk_listener);
269 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
270 struct mptcp_options_received mp_opt;
271 bool opt_mp_capable, opt_mp_join;
272 int err;
273
274 subflow_init_req(req, sk_listener);
275 mptcp_get_options(skb, &mp_opt);
276
277 opt_mp_capable = !!(mp_opt.suboptions & OPTION_MPTCP_MPC_ACK);
278 opt_mp_join = !!(mp_opt.suboptions & OPTION_MPTCP_MPJ_ACK);
279 if (opt_mp_capable && opt_mp_join)
280 return -EINVAL;
281
282 if (opt_mp_capable && listener->request_mptcp) {
283 if (mp_opt.sndr_key == 0)
284 return -EINVAL;
285
286 subflow_req->local_key = mp_opt.rcvr_key;
287 err = mptcp_token_new_request(req);
288 if (err)
289 return err;
290
291 subflow_req->mp_capable = 1;
292 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq - 1;
293 } else if (opt_mp_join && listener->request_mptcp) {
294 if (!mptcp_token_join_cookie_init_state(subflow_req, skb))
295 return -EINVAL;
296
297 subflow_req->mp_join = 1;
298 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq - 1;
299 }
300
301 return 0;
302 }
303 EXPORT_SYMBOL_GPL(mptcp_subflow_init_cookie_req);
304
305 static enum sk_rst_reason mptcp_get_rst_reason(const struct sk_buff *skb)
306 {
307 const struct mptcp_ext *mpext = mptcp_get_ext(skb);
308
309 if (!mpext)
310 return SK_RST_REASON_NOT_SPECIFIED;
311
312 return sk_rst_convert_mptcp_reason(mpext->reset_reason);
313 }
314
315 static struct dst_entry *subflow_v4_route_req(const struct sock *sk,
316 struct sk_buff *skb,
317 struct flowi *fl,
318 struct request_sock *req,
319 u32 tw_isn)
320 {
321 struct dst_entry *dst;
322 int err;
323
324 tcp_rsk(req)->is_mptcp = 1;
325 subflow_init_req(req, sk);
326
327 dst = tcp_request_sock_ipv4_ops.route_req(sk, skb, fl, req, tw_isn);
328 if (!dst)
329 return NULL;
330
331 err = subflow_check_req(req, sk, skb);
332 if (err == 0)
333 return dst;
334
335 dst_release(dst);
336 if (!req->syncookie)
337 tcp_request_sock_ops.send_reset(sk, skb,
338 mptcp_get_rst_reason(skb));
339 return NULL;
340 }
341
342 static void subflow_prep_synack(const struct sock *sk, struct request_sock *req,
343 struct tcp_fastopen_cookie *foc,
344 enum tcp_synack_type synack_type)
345 {
346 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
347 struct inet_request_sock *ireq = inet_rsk(req);
348
349 /* clear tstamp_ok, as needed depending on cookie */
350 if (foc && foc->len > -1)
351 ireq->tstamp_ok = 0;
352
353 if (synack_type == TCP_SYNACK_FASTOPEN)
354 mptcp_fastopen_subflow_synack_set_params(subflow, req);
355 }
356
357 static int subflow_v4_send_synack(const struct sock *sk, struct dst_entry *dst,
358 struct flowi *fl,
359 struct request_sock *req,
360 struct tcp_fastopen_cookie *foc,
361 enum tcp_synack_type synack_type,
362 struct sk_buff *syn_skb)
363 {
364 subflow_prep_synack(sk, req, foc, synack_type);
365
366 return tcp_request_sock_ipv4_ops.send_synack(sk, dst, fl, req, foc,
367 synack_type, syn_skb);
368 }
369
370 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
371 static int subflow_v6_send_synack(const struct sock *sk, struct dst_entry *dst,
372 struct flowi *fl,
373 struct request_sock *req,
374 struct tcp_fastopen_cookie *foc,
375 enum tcp_synack_type synack_type,
376 struct sk_buff *syn_skb)
377 {
378 subflow_prep_synack(sk, req, foc, synack_type);
379
380 return tcp_request_sock_ipv6_ops.send_synack(sk, dst, fl, req, foc,
381 synack_type, syn_skb);
382 }
383
384 static struct dst_entry *subflow_v6_route_req(const struct sock *sk,
385 struct sk_buff *skb,
386 struct flowi *fl,
387 struct request_sock *req,
388 u32 tw_isn)
389 {
390 struct dst_entry *dst;
391 int err;
392
393 tcp_rsk(req)->is_mptcp = 1;
394 subflow_init_req(req, sk);
395
396 dst = tcp_request_sock_ipv6_ops.route_req(sk, skb, fl, req, tw_isn);
397 if (!dst)
398 return NULL;
399
400 err = subflow_check_req(req, sk, skb);
401 if (err == 0)
402 return dst;
403
404 dst_release(dst);
405 if (!req->syncookie)
406 tcp6_request_sock_ops.send_reset(sk, skb,
407 mptcp_get_rst_reason(skb));
408 return NULL;
409 }
410 #endif
411
412 /* validate received truncated hmac and create hmac for third ACK */
413 static bool subflow_thmac_valid(struct mptcp_subflow_context *subflow)
414 {
415 u8 hmac[SHA256_DIGEST_SIZE];
416 u64 thmac;
417
418 subflow_generate_hmac(subflow->remote_key, subflow->local_key,
419 subflow->remote_nonce, subflow->local_nonce,
420 hmac);
421
422 thmac = get_unaligned_be64(hmac);
423 pr_debug("subflow=%p, token=%u, thmac=%llu, subflow->thmac=%llu\n",
424 subflow, subflow->token, thmac, subflow->thmac);
425
426 return thmac == subflow->thmac;
427 }
428
429 void mptcp_subflow_reset(struct sock *ssk)
430 {
431 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
432 struct sock *sk = subflow->conn;
433
434 /* mptcp_mp_fail_no_response() can reach here on an already closed
435 * socket
436 */
437 if (ssk->sk_state == TCP_CLOSE)
438 return;
439
440 /* must hold: tcp_done() could drop last reference on parent */
441 sock_hold(sk);
442
443 mptcp_send_active_reset_reason(ssk);
444 tcp_done(ssk);
445 if (!test_and_set_bit(MPTCP_WORK_CLOSE_SUBFLOW, &mptcp_sk(sk)->flags))
446 mptcp_schedule_work(sk);
447
448 sock_put(sk);
449 }
450
451 static bool subflow_use_different_dport(struct mptcp_sock *msk, const struct sock *sk)
452 {
453 return inet_sk(sk)->inet_dport != inet_sk((struct sock *)msk)->inet_dport;
454 }
455
456 void __mptcp_sync_state(struct sock *sk, int state)
457 {
458 struct mptcp_subflow_context *subflow;
459 struct mptcp_sock *msk = mptcp_sk(sk);
460 struct sock *ssk = msk->first;
461
462 subflow = mptcp_subflow_ctx(ssk);
463 __mptcp_propagate_sndbuf(sk, ssk);
464 if (!msk->rcvspace_init)
465 mptcp_rcv_space_init(msk, ssk);
466
467 if (sk->sk_state == TCP_SYN_SENT) {
468 /* subflow->idsn is always available is TCP_SYN_SENT state,
469 * even for the FASTOPEN scenarios
470 */
471 WRITE_ONCE(msk->write_seq, subflow->idsn + 1);
472 WRITE_ONCE(msk->snd_nxt, msk->write_seq);
473 mptcp_set_state(sk, state);
474 sk->sk_state_change(sk);
475 }
476 }
477
478 static void subflow_set_remote_key(struct mptcp_sock *msk,
479 struct mptcp_subflow_context *subflow,
480 const struct mptcp_options_received *mp_opt)
481 {
482 /* active MPC subflow will reach here multiple times:
483 * at subflow_finish_connect() time and at 4th ack time
484 */
485 if (subflow->remote_key_valid)
486 return;
487
488 subflow->remote_key_valid = 1;
489 subflow->remote_key = mp_opt->sndr_key;
490 mptcp_crypto_key_sha(subflow->remote_key, NULL, &subflow->iasn);
491 subflow->iasn++;
492
493 WRITE_ONCE(msk->remote_key, subflow->remote_key);
494 WRITE_ONCE(msk->ack_seq, subflow->iasn);
495 WRITE_ONCE(msk->can_ack, true);
496 atomic64_set(&msk->rcv_wnd_sent, subflow->iasn);
497 }
498
499 static void mptcp_propagate_state(struct sock *sk, struct sock *ssk,
500 struct mptcp_subflow_context *subflow,
501 const struct mptcp_options_received *mp_opt)
502 {
503 struct mptcp_sock *msk = mptcp_sk(sk);
504
505 mptcp_data_lock(sk);
506 if (mp_opt) {
507 /* Options are available only in the non fallback cases
508 * avoid updating rx path fields otherwise
509 */
510 WRITE_ONCE(msk->snd_una, subflow->idsn + 1);
511 WRITE_ONCE(msk->wnd_end, subflow->idsn + 1 + tcp_sk(ssk)->snd_wnd);
512 subflow_set_remote_key(msk, subflow, mp_opt);
513 }
514
515 if (!sock_owned_by_user(sk)) {
516 __mptcp_sync_state(sk, ssk->sk_state);
517 } else {
518 msk->pending_state = ssk->sk_state;
519 __set_bit(MPTCP_SYNC_STATE, &msk->cb_flags);
520 }
521 mptcp_data_unlock(sk);
522 }
523
524 static void subflow_finish_connect(struct sock *sk, const struct sk_buff *skb)
525 {
526 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
527 struct mptcp_options_received mp_opt;
528 struct sock *parent = subflow->conn;
529 struct mptcp_sock *msk;
530
531 subflow->icsk_af_ops->sk_rx_dst_set(sk, skb);
532
533 /* be sure no special action on any packet other than syn-ack */
534 if (subflow->conn_finished)
535 return;
536
537 msk = mptcp_sk(parent);
538 subflow->rel_write_seq = 1;
539 subflow->conn_finished = 1;
540 subflow->ssn_offset = TCP_SKB_CB(skb)->seq;
541 pr_debug("subflow=%p synack seq=%x\n", subflow, subflow->ssn_offset);
542
543 mptcp_get_options(skb, &mp_opt);
544 if (subflow->request_mptcp) {
545 if (!(mp_opt.suboptions & OPTION_MPTCP_MPC_SYNACK)) {
546 MPTCP_INC_STATS(sock_net(sk),
547 MPTCP_MIB_MPCAPABLEACTIVEFALLBACK);
548 mptcp_do_fallback(sk);
549 pr_fallback(msk);
550 goto fallback;
551 }
552
553 if (mp_opt.suboptions & OPTION_MPTCP_CSUMREQD)
554 WRITE_ONCE(msk->csum_enabled, true);
555 if (mp_opt.deny_join_id0)
556 WRITE_ONCE(msk->pm.remote_deny_join_id0, true);
557 subflow->mp_capable = 1;
558 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPCAPABLEACTIVEACK);
559 mptcp_finish_connect(sk);
560 mptcp_propagate_state(parent, sk, subflow, &mp_opt);
561 } else if (subflow->request_join) {
562 u8 hmac[SHA256_DIGEST_SIZE];
563
564 if (!(mp_opt.suboptions & OPTION_MPTCP_MPJ_SYNACK)) {
565 subflow->reset_reason = MPTCP_RST_EMPTCP;
566 goto do_reset;
567 }
568
569 subflow->backup = mp_opt.backup;
570 subflow->thmac = mp_opt.thmac;
571 subflow->remote_nonce = mp_opt.nonce;
572 WRITE_ONCE(subflow->remote_id, mp_opt.join_id);
573 pr_debug("subflow=%p, thmac=%llu, remote_nonce=%u backup=%d\n",
574 subflow, subflow->thmac, subflow->remote_nonce,
575 subflow->backup);
576
577 if (!subflow_thmac_valid(subflow)) {
578 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINACKMAC);
579 subflow->reset_reason = MPTCP_RST_EMPTCP;
580 goto do_reset;
581 }
582
583 if (!mptcp_finish_join(sk))
584 goto do_reset;
585
586 subflow_generate_hmac(subflow->local_key, subflow->remote_key,
587 subflow->local_nonce,
588 subflow->remote_nonce,
589 hmac);
590 memcpy(subflow->hmac, hmac, MPTCPOPT_HMAC_LEN);
591
592 subflow->mp_join = 1;
593 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINSYNACKRX);
594
595 if (subflow->backup)
596 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINSYNACKBACKUPRX);
597
598 if (subflow_use_different_dport(msk, sk)) {
599 pr_debug("synack inet_dport=%d %d\n",
600 ntohs(inet_sk(sk)->inet_dport),
601 ntohs(inet_sk(parent)->inet_dport));
602 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINPORTSYNACKRX);
603 }
604 } else if (mptcp_check_fallback(sk)) {
605 fallback:
606 mptcp_propagate_state(parent, sk, subflow, NULL);
607 }
608 return;
609
610 do_reset:
611 subflow->reset_transient = 0;
612 mptcp_subflow_reset(sk);
613 }
614
615 static void subflow_set_local_id(struct mptcp_subflow_context *subflow, int local_id)
616 {
617 WARN_ON_ONCE(local_id < 0 || local_id > 255);
618 WRITE_ONCE(subflow->local_id, local_id);
619 }
620
621 static int subflow_chk_local_id(struct sock *sk)
622 {
623 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
624 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
625 int err;
626
627 if (likely(subflow->local_id >= 0))
628 return 0;
629
630 err = mptcp_pm_get_local_id(msk, (struct sock_common *)sk);
631 if (err < 0)
632 return err;
633
634 subflow_set_local_id(subflow, err);
635 subflow->request_bkup = mptcp_pm_is_backup(msk, (struct sock_common *)sk);
636
637 return 0;
638 }
639
640 static int subflow_rebuild_header(struct sock *sk)
641 {
642 int err = subflow_chk_local_id(sk);
643
644 if (unlikely(err < 0))
645 return err;
646
647 return inet_sk_rebuild_header(sk);
648 }
649
650 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
651 static int subflow_v6_rebuild_header(struct sock *sk)
652 {
653 int err = subflow_chk_local_id(sk);
654
655 if (unlikely(err < 0))
656 return err;
657
658 return inet6_sk_rebuild_header(sk);
659 }
660 #endif
661
662 static struct request_sock_ops mptcp_subflow_v4_request_sock_ops __ro_after_init;
663 static struct tcp_request_sock_ops subflow_request_sock_ipv4_ops __ro_after_init;
664
665 static int subflow_v4_conn_request(struct sock *sk, struct sk_buff *skb)
666 {
667 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
668
669 pr_debug("subflow=%p\n", subflow);
670
671 /* Never answer to SYNs sent to broadcast or multicast */
672 if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
673 goto drop;
674
675 return tcp_conn_request(&mptcp_subflow_v4_request_sock_ops,
676 &subflow_request_sock_ipv4_ops,
677 sk, skb);
678 drop:
679 tcp_listendrop(sk);
680 return 0;
681 }
682
683 static void subflow_v4_req_destructor(struct request_sock *req)
684 {
685 subflow_req_destructor(req);
686 tcp_request_sock_ops.destructor(req);
687 }
688
689 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
690 static struct request_sock_ops mptcp_subflow_v6_request_sock_ops __ro_after_init;
691 static struct tcp_request_sock_ops subflow_request_sock_ipv6_ops __ro_after_init;
692 static struct inet_connection_sock_af_ops subflow_v6_specific __ro_after_init;
693 static struct inet_connection_sock_af_ops subflow_v6m_specific __ro_after_init;
694 static struct proto tcpv6_prot_override __ro_after_init;
695
696 static int subflow_v6_conn_request(struct sock *sk, struct sk_buff *skb)
697 {
698 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
699
700 pr_debug("subflow=%p\n", subflow);
701
702 if (skb->protocol == htons(ETH_P_IP))
703 return subflow_v4_conn_request(sk, skb);
704
705 if (!ipv6_unicast_destination(skb))
706 goto drop;
707
708 if (ipv6_addr_v4mapped(&ipv6_hdr(skb)->saddr)) {
709 __IP6_INC_STATS(sock_net(sk), NULL, IPSTATS_MIB_INHDRERRORS);
710 return 0;
711 }
712
713 return tcp_conn_request(&mptcp_subflow_v6_request_sock_ops,
714 &subflow_request_sock_ipv6_ops, sk, skb);
715
716 drop:
717 tcp_listendrop(sk);
718 return 0; /* don't send reset */
719 }
720
721 static void subflow_v6_req_destructor(struct request_sock *req)
722 {
723 subflow_req_destructor(req);
724 tcp6_request_sock_ops.destructor(req);
725 }
726 #endif
727
728 struct request_sock *mptcp_subflow_reqsk_alloc(const struct request_sock_ops *ops,
729 struct sock *sk_listener,
730 bool attach_listener)
731 {
732 if (ops->family == AF_INET)
733 ops = &mptcp_subflow_v4_request_sock_ops;
734 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
735 else if (ops->family == AF_INET6)
736 ops = &mptcp_subflow_v6_request_sock_ops;
737 #endif
738
739 return inet_reqsk_alloc(ops, sk_listener, attach_listener);
740 }
741 EXPORT_SYMBOL(mptcp_subflow_reqsk_alloc);
742
743 /* validate hmac received in third ACK */
744 static bool subflow_hmac_valid(const struct request_sock *req,
745 const struct mptcp_options_received *mp_opt)
746 {
747 const struct mptcp_subflow_request_sock *subflow_req;
748 u8 hmac[SHA256_DIGEST_SIZE];
749 struct mptcp_sock *msk;
750
751 subflow_req = mptcp_subflow_rsk(req);
752 msk = subflow_req->msk;
753 if (!msk)
754 return false;
755
756 subflow_generate_hmac(READ_ONCE(msk->remote_key),
757 READ_ONCE(msk->local_key),
758 subflow_req->remote_nonce,
759 subflow_req->local_nonce, hmac);
760
761 return !crypto_memneq(hmac, mp_opt->hmac, MPTCPOPT_HMAC_LEN);
762 }
763
764 static void subflow_ulp_fallback(struct sock *sk,
765 struct mptcp_subflow_context *old_ctx)
766 {
767 struct inet_connection_sock *icsk = inet_csk(sk);
768
769 mptcp_subflow_tcp_fallback(sk, old_ctx);
770 icsk->icsk_ulp_ops = NULL;
771 rcu_assign_pointer(icsk->icsk_ulp_data, NULL);
772 tcp_sk(sk)->is_mptcp = 0;
773
774 mptcp_subflow_ops_undo_override(sk);
775 }
776
777 void mptcp_subflow_drop_ctx(struct sock *ssk)
778 {
779 struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(ssk);
780
781 if (!ctx)
782 return;
783
784 list_del(&mptcp_subflow_ctx(ssk)->node);
785 if (inet_csk(ssk)->icsk_ulp_ops) {
786 subflow_ulp_fallback(ssk, ctx);
787 if (ctx->conn)
788 sock_put(ctx->conn);
789 }
790
791 kfree_rcu(ctx, rcu);
792 }
793
794 void __mptcp_subflow_fully_established(struct mptcp_sock *msk,
795 struct mptcp_subflow_context *subflow,
796 const struct mptcp_options_received *mp_opt)
797 {
798 subflow_set_remote_key(msk, subflow, mp_opt);
799 subflow->fully_established = 1;
800 WRITE_ONCE(msk->fully_established, true);
801
802 if (subflow->is_mptfo)
803 __mptcp_fastopen_gen_msk_ackseq(msk, subflow, mp_opt);
804 }
805
806 static struct sock *subflow_syn_recv_sock(const struct sock *sk,
807 struct sk_buff *skb,
808 struct request_sock *req,
809 struct dst_entry *dst,
810 struct request_sock *req_unhash,
811 bool *own_req)
812 {
813 struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk);
814 struct mptcp_subflow_request_sock *subflow_req;
815 struct mptcp_options_received mp_opt;
816 bool fallback, fallback_is_fatal;
817 enum sk_rst_reason reason;
818 struct mptcp_sock *owner;
819 struct sock *child;
820
821 pr_debug("listener=%p, req=%p, conn=%p\n", listener, req, listener->conn);
822
823 /* After child creation we must look for MPC even when options
824 * are not parsed
825 */
826 mp_opt.suboptions = 0;
827
828 /* hopefully temporary handling for MP_JOIN+syncookie */
829 subflow_req = mptcp_subflow_rsk(req);
830 fallback_is_fatal = tcp_rsk(req)->is_mptcp && subflow_req->mp_join;
831 fallback = !tcp_rsk(req)->is_mptcp;
832 if (fallback)
833 goto create_child;
834
835 /* if the sk is MP_CAPABLE, we try to fetch the client key */
836 if (subflow_req->mp_capable) {
837 /* we can receive and accept an in-window, out-of-order pkt,
838 * which may not carry the MP_CAPABLE opt even on mptcp enabled
839 * paths: always try to extract the peer key, and fallback
840 * for packets missing it.
841 * Even OoO DSS packets coming legitly after dropped or
842 * reordered MPC will cause fallback, but we don't have other
843 * options.
844 */
845 mptcp_get_options(skb, &mp_opt);
846 if (!(mp_opt.suboptions &
847 (OPTION_MPTCP_MPC_SYN | OPTION_MPTCP_MPC_ACK)))
848 fallback = true;
849
850 } else if (subflow_req->mp_join) {
851 mptcp_get_options(skb, &mp_opt);
852 if (!(mp_opt.suboptions & OPTION_MPTCP_MPJ_ACK) ||
853 !subflow_hmac_valid(req, &mp_opt) ||
854 !mptcp_can_accept_new_subflow(subflow_req->msk)) {
855 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKMAC);
856 fallback = true;
857 }
858 }
859
860 create_child:
861 child = listener->icsk_af_ops->syn_recv_sock(sk, skb, req, dst,
862 req_unhash, own_req);
863
864 if (child && *own_req) {
865 struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(child);
866
867 tcp_rsk(req)->drop_req = false;
868
869 /* we need to fallback on ctx allocation failure and on pre-reqs
870 * checking above. In the latter scenario we additionally need
871 * to reset the context to non MPTCP status.
872 */
873 if (!ctx || fallback) {
874 if (fallback_is_fatal) {
875 subflow_add_reset_reason(skb, MPTCP_RST_EMPTCP);
876 goto dispose_child;
877 }
878 goto fallback;
879 }
880
881 /* ssk inherits options of listener sk */
882 ctx->setsockopt_seq = listener->setsockopt_seq;
883
884 if (ctx->mp_capable) {
885 ctx->conn = mptcp_sk_clone_init(listener->conn, &mp_opt, child, req);
886 if (!ctx->conn)
887 goto fallback;
888
889 ctx->subflow_id = 1;
890 owner = mptcp_sk(ctx->conn);
891 mptcp_pm_new_connection(owner, child, 1);
892
893 /* with OoO packets we can reach here without ingress
894 * mpc option
895 */
896 if (mp_opt.suboptions & OPTION_MPTCP_MPC_ACK) {
897 mptcp_pm_fully_established(owner, child);
898 ctx->pm_notified = 1;
899 }
900 } else if (ctx->mp_join) {
901 owner = subflow_req->msk;
902 if (!owner) {
903 subflow_add_reset_reason(skb, MPTCP_RST_EPROHIBIT);
904 goto dispose_child;
905 }
906
907 /* move the msk reference ownership to the subflow */
908 subflow_req->msk = NULL;
909 ctx->conn = (struct sock *)owner;
910
911 if (subflow_use_different_sport(owner, sk)) {
912 pr_debug("ack inet_sport=%d %d\n",
913 ntohs(inet_sk(sk)->inet_sport),
914 ntohs(inet_sk((struct sock *)owner)->inet_sport));
915 if (!mptcp_pm_sport_in_anno_list(owner, sk)) {
916 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MISMATCHPORTACKRX);
917 subflow_add_reset_reason(skb, MPTCP_RST_EPROHIBIT);
918 goto dispose_child;
919 }
920 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINPORTACKRX);
921 }
922
923 if (!mptcp_finish_join(child)) {
924 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(child);
925
926 subflow_add_reset_reason(skb, subflow->reset_reason);
927 goto dispose_child;
928 }
929
930 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKRX);
931 tcp_rsk(req)->drop_req = true;
932 }
933 }
934
935 /* check for expected invariant - should never trigger, just help
936 * catching earlier subtle bugs
937 */
938 WARN_ON_ONCE(child && *own_req && tcp_sk(child)->is_mptcp &&
939 (!mptcp_subflow_ctx(child) ||
940 !mptcp_subflow_ctx(child)->conn));
941 return child;
942
943 dispose_child:
944 mptcp_subflow_drop_ctx(child);
945 tcp_rsk(req)->drop_req = true;
946 inet_csk_prepare_for_destroy_sock(child);
947 tcp_done(child);
948 reason = mptcp_get_rst_reason(skb);
949 req->rsk_ops->send_reset(sk, skb, reason);
950
951 /* The last child reference will be released by the caller */
952 return child;
953
954 fallback:
955 if (fallback)
956 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MPCAPABLEPASSIVEFALLBACK);
957 mptcp_subflow_drop_ctx(child);
958 return child;
959 }
960
961 static struct inet_connection_sock_af_ops subflow_specific __ro_after_init;
962 static struct proto tcp_prot_override __ro_after_init;
963
964 enum mapping_status {
965 MAPPING_OK,
966 MAPPING_INVALID,
967 MAPPING_EMPTY,
968 MAPPING_DATA_FIN,
969 MAPPING_DUMMY,
970 MAPPING_BAD_CSUM
971 };
972
973 static void dbg_bad_map(struct mptcp_subflow_context *subflow, u32 ssn)
974 {
975 pr_debug("Bad mapping: ssn=%d map_seq=%d map_data_len=%d\n",
976 ssn, subflow->map_subflow_seq, subflow->map_data_len);
977 }
978
979 static bool skb_is_fully_mapped(struct sock *ssk, struct sk_buff *skb)
980 {
981 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
982 unsigned int skb_consumed;
983
984 skb_consumed = tcp_sk(ssk)->copied_seq - TCP_SKB_CB(skb)->seq;
985 if (unlikely(skb_consumed >= skb->len)) {
986 DEBUG_NET_WARN_ON_ONCE(1);
987 return true;
988 }
989
990 return skb->len - skb_consumed <= subflow->map_data_len -
991 mptcp_subflow_get_map_offset(subflow);
992 }
993
994 static bool validate_mapping(struct sock *ssk, struct sk_buff *skb)
995 {
996 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
997 u32 ssn = tcp_sk(ssk)->copied_seq - subflow->ssn_offset;
998
999 if (unlikely(before(ssn, subflow->map_subflow_seq))) {
1000 /* Mapping covers data later in the subflow stream,
1001 * currently unsupported.
1002 */
1003 dbg_bad_map(subflow, ssn);
1004 return false;
1005 }
1006 if (unlikely(!before(ssn, subflow->map_subflow_seq +
1007 subflow->map_data_len))) {
1008 /* Mapping does covers past subflow data, invalid */
1009 dbg_bad_map(subflow, ssn);
1010 return false;
1011 }
1012 return true;
1013 }
1014
1015 static enum mapping_status validate_data_csum(struct sock *ssk, struct sk_buff *skb,
1016 bool csum_reqd)
1017 {
1018 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1019 u32 offset, seq, delta;
1020 __sum16 csum;
1021 int len;
1022
1023 if (!csum_reqd)
1024 return MAPPING_OK;
1025
1026 /* mapping already validated on previous traversal */
1027 if (subflow->map_csum_len == subflow->map_data_len)
1028 return MAPPING_OK;
1029
1030 /* traverse the receive queue, ensuring it contains a full
1031 * DSS mapping and accumulating the related csum.
1032 * Preserve the accoumlate csum across multiple calls, to compute
1033 * the csum only once
1034 */
1035 delta = subflow->map_data_len - subflow->map_csum_len;
1036 for (;;) {
1037 seq = tcp_sk(ssk)->copied_seq + subflow->map_csum_len;
1038 offset = seq - TCP_SKB_CB(skb)->seq;
1039
1040 /* if the current skb has not been accounted yet, csum its contents
1041 * up to the amount covered by the current DSS
1042 */
1043 if (offset < skb->len) {
1044 __wsum csum;
1045
1046 len = min(skb->len - offset, delta);
1047 csum = skb_checksum(skb, offset, len, 0);
1048 subflow->map_data_csum = csum_block_add(subflow->map_data_csum, csum,
1049 subflow->map_csum_len);
1050
1051 delta -= len;
1052 subflow->map_csum_len += len;
1053 }
1054 if (delta == 0)
1055 break;
1056
1057 if (skb_queue_is_last(&ssk->sk_receive_queue, skb)) {
1058 /* if this subflow is closed, the partial mapping
1059 * will be never completed; flush the pending skbs, so
1060 * that subflow_sched_work_if_closed() can kick in
1061 */
1062 if (unlikely(ssk->sk_state == TCP_CLOSE))
1063 while ((skb = skb_peek(&ssk->sk_receive_queue)))
1064 sk_eat_skb(ssk, skb);
1065
1066 /* not enough data to validate the csum */
1067 return MAPPING_EMPTY;
1068 }
1069
1070 /* the DSS mapping for next skbs will be validated later,
1071 * when a get_mapping_status call will process such skb
1072 */
1073 skb = skb->next;
1074 }
1075
1076 /* note that 'map_data_len' accounts only for the carried data, does
1077 * not include the eventual seq increment due to the data fin,
1078 * while the pseudo header requires the original DSS data len,
1079 * including that
1080 */
1081 csum = __mptcp_make_csum(subflow->map_seq,
1082 subflow->map_subflow_seq,
1083 subflow->map_data_len + subflow->map_data_fin,
1084 subflow->map_data_csum);
1085 if (unlikely(csum)) {
1086 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DATACSUMERR);
1087 return MAPPING_BAD_CSUM;
1088 }
1089
1090 subflow->valid_csum_seen = 1;
1091 return MAPPING_OK;
1092 }
1093
1094 static enum mapping_status get_mapping_status(struct sock *ssk,
1095 struct mptcp_sock *msk)
1096 {
1097 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1098 bool csum_reqd = READ_ONCE(msk->csum_enabled);
1099 struct mptcp_ext *mpext;
1100 struct sk_buff *skb;
1101 u16 data_len;
1102 u64 map_seq;
1103
1104 skb = skb_peek(&ssk->sk_receive_queue);
1105 if (!skb)
1106 return MAPPING_EMPTY;
1107
1108 if (mptcp_check_fallback(ssk))
1109 return MAPPING_DUMMY;
1110
1111 mpext = mptcp_get_ext(skb);
1112 if (!mpext || !mpext->use_map) {
1113 if (!subflow->map_valid && !skb->len) {
1114 /* the TCP stack deliver 0 len FIN pkt to the receive
1115 * queue, that is the only 0len pkts ever expected here,
1116 * and we can admit no mapping only for 0 len pkts
1117 */
1118 if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN))
1119 WARN_ONCE(1, "0len seq %d:%d flags %x",
1120 TCP_SKB_CB(skb)->seq,
1121 TCP_SKB_CB(skb)->end_seq,
1122 TCP_SKB_CB(skb)->tcp_flags);
1123 sk_eat_skb(ssk, skb);
1124 return MAPPING_EMPTY;
1125 }
1126
1127 if (!subflow->map_valid)
1128 return MAPPING_INVALID;
1129
1130 goto validate_seq;
1131 }
1132
1133 trace_get_mapping_status(mpext);
1134
1135 data_len = mpext->data_len;
1136 if (data_len == 0) {
1137 pr_debug("infinite mapping received\n");
1138 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_INFINITEMAPRX);
1139 subflow->map_data_len = 0;
1140 return MAPPING_INVALID;
1141 }
1142
1143 if (mpext->data_fin == 1) {
1144 u64 data_fin_seq;
1145
1146 if (data_len == 1) {
1147 bool updated = mptcp_update_rcv_data_fin(msk, mpext->data_seq,
1148 mpext->dsn64);
1149 pr_debug("DATA_FIN with no payload seq=%llu\n", mpext->data_seq);
1150 if (subflow->map_valid) {
1151 /* A DATA_FIN might arrive in a DSS
1152 * option before the previous mapping
1153 * has been fully consumed. Continue
1154 * handling the existing mapping.
1155 */
1156 skb_ext_del(skb, SKB_EXT_MPTCP);
1157 return MAPPING_OK;
1158 }
1159
1160 if (updated)
1161 mptcp_schedule_work((struct sock *)msk);
1162
1163 return MAPPING_DATA_FIN;
1164 }
1165
1166 data_fin_seq = mpext->data_seq + data_len - 1;
1167
1168 /* If mpext->data_seq is a 32-bit value, data_fin_seq must also
1169 * be limited to 32 bits.
1170 */
1171 if (!mpext->dsn64)
1172 data_fin_seq &= GENMASK_ULL(31, 0);
1173
1174 mptcp_update_rcv_data_fin(msk, data_fin_seq, mpext->dsn64);
1175 pr_debug("DATA_FIN with mapping seq=%llu dsn64=%d\n",
1176 data_fin_seq, mpext->dsn64);
1177
1178 /* Adjust for DATA_FIN using 1 byte of sequence space */
1179 data_len--;
1180 }
1181
1182 map_seq = mptcp_expand_seq(READ_ONCE(msk->ack_seq), mpext->data_seq, mpext->dsn64);
1183 WRITE_ONCE(mptcp_sk(subflow->conn)->use_64bit_ack, !!mpext->dsn64);
1184
1185 if (subflow->map_valid) {
1186 /* Allow replacing only with an identical map */
1187 if (subflow->map_seq == map_seq &&
1188 subflow->map_subflow_seq == mpext->subflow_seq &&
1189 subflow->map_data_len == data_len &&
1190 subflow->map_csum_reqd == mpext->csum_reqd) {
1191 skb_ext_del(skb, SKB_EXT_MPTCP);
1192 goto validate_csum;
1193 }
1194
1195 /* If this skb data are fully covered by the current mapping,
1196 * the new map would need caching, which is not supported
1197 */
1198 if (skb_is_fully_mapped(ssk, skb)) {
1199 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DSSNOMATCH);
1200 return MAPPING_INVALID;
1201 }
1202
1203 /* will validate the next map after consuming the current one */
1204 goto validate_csum;
1205 }
1206
1207 subflow->map_seq = map_seq;
1208 subflow->map_subflow_seq = mpext->subflow_seq;
1209 subflow->map_data_len = data_len;
1210 subflow->map_valid = 1;
1211 subflow->map_data_fin = mpext->data_fin;
1212 subflow->mpc_map = mpext->mpc_map;
1213 subflow->map_csum_reqd = mpext->csum_reqd;
1214 subflow->map_csum_len = 0;
1215 subflow->map_data_csum = csum_unfold(mpext->csum);
1216
1217 /* Cfr RFC 8684 Section 3.3.0 */
1218 if (unlikely(subflow->map_csum_reqd != csum_reqd))
1219 return MAPPING_INVALID;
1220
1221 pr_debug("new map seq=%llu subflow_seq=%u data_len=%u csum=%d:%u\n",
1222 subflow->map_seq, subflow->map_subflow_seq,
1223 subflow->map_data_len, subflow->map_csum_reqd,
1224 subflow->map_data_csum);
1225
1226 validate_seq:
1227 /* we revalidate valid mapping on new skb, because we must ensure
1228 * the current skb is completely covered by the available mapping
1229 */
1230 if (!validate_mapping(ssk, skb)) {
1231 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DSSTCPMISMATCH);
1232 return MAPPING_INVALID;
1233 }
1234
1235 skb_ext_del(skb, SKB_EXT_MPTCP);
1236
1237 validate_csum:
1238 return validate_data_csum(ssk, skb, csum_reqd);
1239 }
1240
1241 static void mptcp_subflow_discard_data(struct sock *ssk, struct sk_buff *skb,
1242 u64 limit)
1243 {
1244 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1245 bool fin = TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN;
1246 struct tcp_sock *tp = tcp_sk(ssk);
1247 u32 offset, incr, avail_len;
1248
1249 offset = tp->copied_seq - TCP_SKB_CB(skb)->seq;
1250 if (WARN_ON_ONCE(offset > skb->len))
1251 goto out;
1252
1253 avail_len = skb->len - offset;
1254 incr = limit >= avail_len ? avail_len + fin : limit;
1255
1256 pr_debug("discarding=%d len=%d offset=%d seq=%d\n", incr, skb->len,
1257 offset, subflow->map_subflow_seq);
1258 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DUPDATA);
1259 tcp_sk(ssk)->copied_seq += incr;
1260
1261 out:
1262 if (!before(tcp_sk(ssk)->copied_seq, TCP_SKB_CB(skb)->end_seq))
1263 sk_eat_skb(ssk, skb);
1264 if (mptcp_subflow_get_map_offset(subflow) >= subflow->map_data_len)
1265 subflow->map_valid = 0;
1266 }
1267
1268 /* sched mptcp worker to remove the subflow if no more data is pending */
1269 static void subflow_sched_work_if_closed(struct mptcp_sock *msk, struct sock *ssk)
1270 {
1271 struct sock *sk = (struct sock *)msk;
1272
1273 if (likely(ssk->sk_state != TCP_CLOSE &&
1274 (ssk->sk_state != TCP_CLOSE_WAIT ||
1275 inet_sk_state_load(sk) != TCP_ESTABLISHED)))
1276 return;
1277
1278 if (skb_queue_empty(&ssk->sk_receive_queue) &&
1279 !test_and_set_bit(MPTCP_WORK_CLOSE_SUBFLOW, &msk->flags))
1280 mptcp_schedule_work(sk);
1281 }
1282
1283 static bool subflow_can_fallback(struct mptcp_subflow_context *subflow)
1284 {
1285 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
1286
1287 if (subflow->mp_join)
1288 return false;
1289 else if (READ_ONCE(msk->csum_enabled))
1290 return !subflow->valid_csum_seen;
1291 else
1292 return READ_ONCE(msk->allow_infinite_fallback);
1293 }
1294
1295 static void mptcp_subflow_fail(struct mptcp_sock *msk, struct sock *ssk)
1296 {
1297 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1298 unsigned long fail_tout;
1299
1300 /* graceful failure can happen only on the MPC subflow */
1301 if (WARN_ON_ONCE(ssk != READ_ONCE(msk->first)))
1302 return;
1303
1304 /* since the close timeout take precedence on the fail one,
1305 * no need to start the latter when the first is already set
1306 */
1307 if (sock_flag((struct sock *)msk, SOCK_DEAD))
1308 return;
1309
1310 /* we don't need extreme accuracy here, use a zero fail_tout as special
1311 * value meaning no fail timeout at all;
1312 */
1313 fail_tout = jiffies + TCP_RTO_MAX;
1314 if (!fail_tout)
1315 fail_tout = 1;
1316 WRITE_ONCE(subflow->fail_tout, fail_tout);
1317 tcp_send_ack(ssk);
1318
1319 mptcp_reset_tout_timer(msk, subflow->fail_tout);
1320 }
1321
1322 static bool subflow_check_data_avail(struct sock *ssk)
1323 {
1324 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1325 enum mapping_status status;
1326 struct mptcp_sock *msk;
1327 struct sk_buff *skb;
1328
1329 if (!skb_peek(&ssk->sk_receive_queue))
1330 WRITE_ONCE(subflow->data_avail, false);
1331 if (subflow->data_avail)
1332 return true;
1333
1334 msk = mptcp_sk(subflow->conn);
1335 for (;;) {
1336 u64 ack_seq;
1337 u64 old_ack;
1338
1339 status = get_mapping_status(ssk, msk);
1340 trace_subflow_check_data_avail(status, skb_peek(&ssk->sk_receive_queue));
1341 if (unlikely(status == MAPPING_INVALID || status == MAPPING_DUMMY ||
1342 status == MAPPING_BAD_CSUM))
1343 goto fallback;
1344
1345 if (status != MAPPING_OK)
1346 goto no_data;
1347
1348 skb = skb_peek(&ssk->sk_receive_queue);
1349 if (WARN_ON_ONCE(!skb))
1350 goto no_data;
1351
1352 if (unlikely(!READ_ONCE(msk->can_ack)))
1353 goto fallback;
1354
1355 old_ack = READ_ONCE(msk->ack_seq);
1356 ack_seq = mptcp_subflow_get_mapped_dsn(subflow);
1357 pr_debug("msk ack_seq=%llx subflow ack_seq=%llx\n", old_ack,
1358 ack_seq);
1359 if (unlikely(before64(ack_seq, old_ack))) {
1360 mptcp_subflow_discard_data(ssk, skb, old_ack - ack_seq);
1361 continue;
1362 }
1363
1364 WRITE_ONCE(subflow->data_avail, true);
1365 break;
1366 }
1367 return true;
1368
1369 no_data:
1370 subflow_sched_work_if_closed(msk, ssk);
1371 return false;
1372
1373 fallback:
1374 if (!__mptcp_check_fallback(msk)) {
1375 /* RFC 8684 section 3.7. */
1376 if (status == MAPPING_BAD_CSUM &&
1377 (subflow->mp_join || subflow->valid_csum_seen)) {
1378 subflow->send_mp_fail = 1;
1379
1380 if (!READ_ONCE(msk->allow_infinite_fallback)) {
1381 subflow->reset_transient = 0;
1382 subflow->reset_reason = MPTCP_RST_EMIDDLEBOX;
1383 goto reset;
1384 }
1385 mptcp_subflow_fail(msk, ssk);
1386 WRITE_ONCE(subflow->data_avail, true);
1387 return true;
1388 }
1389
1390 if (!subflow_can_fallback(subflow) && subflow->map_data_len) {
1391 /* fatal protocol error, close the socket.
1392 * subflow_error_report() will introduce the appropriate barriers
1393 */
1394 subflow->reset_transient = 0;
1395 subflow->reset_reason = MPTCP_RST_EMPTCP;
1396
1397 reset:
1398 WRITE_ONCE(ssk->sk_err, EBADMSG);
1399 tcp_set_state(ssk, TCP_CLOSE);
1400 while ((skb = skb_peek(&ssk->sk_receive_queue)))
1401 sk_eat_skb(ssk, skb);
1402 mptcp_send_active_reset_reason(ssk);
1403 WRITE_ONCE(subflow->data_avail, false);
1404 return false;
1405 }
1406
1407 mptcp_do_fallback(ssk);
1408 }
1409
1410 skb = skb_peek(&ssk->sk_receive_queue);
1411 subflow->map_valid = 1;
1412 subflow->map_seq = READ_ONCE(msk->ack_seq);
1413 subflow->map_data_len = skb->len;
1414 subflow->map_subflow_seq = tcp_sk(ssk)->copied_seq - subflow->ssn_offset;
1415 WRITE_ONCE(subflow->data_avail, true);
1416 return true;
1417 }
1418
1419 bool mptcp_subflow_data_available(struct sock *sk)
1420 {
1421 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1422
1423 /* check if current mapping is still valid */
1424 if (subflow->map_valid &&
1425 mptcp_subflow_get_map_offset(subflow) >= subflow->map_data_len) {
1426 subflow->map_valid = 0;
1427 WRITE_ONCE(subflow->data_avail, false);
1428
1429 pr_debug("Done with mapping: seq=%u data_len=%u\n",
1430 subflow->map_subflow_seq,
1431 subflow->map_data_len);
1432 }
1433
1434 return subflow_check_data_avail(sk);
1435 }
1436
1437 /* If ssk has an mptcp parent socket, use the mptcp rcvbuf occupancy,
1438 * not the ssk one.
1439 *
1440 * In mptcp, rwin is about the mptcp-level connection data.
1441 *
1442 * Data that is still on the ssk rx queue can thus be ignored,
1443 * as far as mptcp peer is concerned that data is still inflight.
1444 * DSS ACK is updated when skb is moved to the mptcp rx queue.
1445 */
1446 void mptcp_space(const struct sock *ssk, int *space, int *full_space)
1447 {
1448 const struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1449 const struct sock *sk = subflow->conn;
1450
1451 *space = __mptcp_space(sk);
1452 *full_space = mptcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf));
1453 }
1454
1455 static void subflow_error_report(struct sock *ssk)
1456 {
1457 struct sock *sk = mptcp_subflow_ctx(ssk)->conn;
1458
1459 /* bail early if this is a no-op, so that we avoid introducing a
1460 * problematic lockdep dependency between TCP accept queue lock
1461 * and msk socket spinlock
1462 */
1463 if (!sk->sk_socket)
1464 return;
1465
1466 mptcp_data_lock(sk);
1467 if (!sock_owned_by_user(sk))
1468 __mptcp_error_report(sk);
1469 else
1470 __set_bit(MPTCP_ERROR_REPORT, &mptcp_sk(sk)->cb_flags);
1471 mptcp_data_unlock(sk);
1472 }
1473
1474 static void subflow_data_ready(struct sock *sk)
1475 {
1476 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1477 u16 state = 1 << inet_sk_state_load(sk);
1478 struct sock *parent = subflow->conn;
1479 struct mptcp_sock *msk;
1480
1481 trace_sk_data_ready(sk);
1482
1483 msk = mptcp_sk(parent);
1484 if (state & TCPF_LISTEN) {
1485 /* MPJ subflow are removed from accept queue before reaching here,
1486 * avoid stray wakeups
1487 */
1488 if (reqsk_queue_empty(&inet_csk(sk)->icsk_accept_queue))
1489 return;
1490
1491 parent->sk_data_ready(parent);
1492 return;
1493 }
1494
1495 WARN_ON_ONCE(!__mptcp_check_fallback(msk) && !subflow->mp_capable &&
1496 !subflow->mp_join && !(state & TCPF_CLOSE));
1497
1498 if (mptcp_subflow_data_available(sk)) {
1499 mptcp_data_ready(parent, sk);
1500
1501 /* subflow-level lowat test are not relevant.
1502 * respect the msk-level threshold eventually mandating an immediate ack
1503 */
1504 if (mptcp_data_avail(msk) < parent->sk_rcvlowat &&
1505 (tcp_sk(sk)->rcv_nxt - tcp_sk(sk)->rcv_wup) > inet_csk(sk)->icsk_ack.rcv_mss)
1506 inet_csk(sk)->icsk_ack.pending |= ICSK_ACK_NOW;
1507 } else if (unlikely(sk->sk_err)) {
1508 subflow_error_report(sk);
1509 }
1510 }
1511
1512 static void subflow_write_space(struct sock *ssk)
1513 {
1514 struct sock *sk = mptcp_subflow_ctx(ssk)->conn;
1515
1516 mptcp_propagate_sndbuf(sk, ssk);
1517 mptcp_write_space(sk);
1518 }
1519
1520 static const struct inet_connection_sock_af_ops *
1521 subflow_default_af_ops(struct sock *sk)
1522 {
1523 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1524 if (sk->sk_family == AF_INET6)
1525 return &subflow_v6_specific;
1526 #endif
1527 return &subflow_specific;
1528 }
1529
1530 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1531 void mptcpv6_handle_mapped(struct sock *sk, bool mapped)
1532 {
1533 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1534 struct inet_connection_sock *icsk = inet_csk(sk);
1535 const struct inet_connection_sock_af_ops *target;
1536
1537 target = mapped ? &subflow_v6m_specific : subflow_default_af_ops(sk);
1538
1539 pr_debug("subflow=%p family=%d ops=%p target=%p mapped=%d\n",
1540 subflow, sk->sk_family, icsk->icsk_af_ops, target, mapped);
1541
1542 if (likely(icsk->icsk_af_ops == target))
1543 return;
1544
1545 subflow->icsk_af_ops = icsk->icsk_af_ops;
1546 icsk->icsk_af_ops = target;
1547 }
1548 #endif
1549
1550 void mptcp_info2sockaddr(const struct mptcp_addr_info *info,
1551 struct sockaddr_storage *addr,
1552 unsigned short family)
1553 {
1554 memset(addr, 0, sizeof(*addr));
1555 addr->ss_family = family;
1556 if (addr->ss_family == AF_INET) {
1557 struct sockaddr_in *in_addr = (struct sockaddr_in *)addr;
1558
1559 if (info->family == AF_INET)
1560 in_addr->sin_addr = info->addr;
1561 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1562 else if (ipv6_addr_v4mapped(&info->addr6))
1563 in_addr->sin_addr.s_addr = info->addr6.s6_addr32[3];
1564 #endif
1565 in_addr->sin_port = info->port;
1566 }
1567 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1568 else if (addr->ss_family == AF_INET6) {
1569 struct sockaddr_in6 *in6_addr = (struct sockaddr_in6 *)addr;
1570
1571 if (info->family == AF_INET)
1572 ipv6_addr_set_v4mapped(info->addr.s_addr,
1573 &in6_addr->sin6_addr);
1574 else
1575 in6_addr->sin6_addr = info->addr6;
1576 in6_addr->sin6_port = info->port;
1577 }
1578 #endif
1579 }
1580
1581 int __mptcp_subflow_connect(struct sock *sk, const struct mptcp_addr_info *loc,
1582 const struct mptcp_addr_info *remote)
1583 {
1584 struct mptcp_sock *msk = mptcp_sk(sk);
1585 struct mptcp_subflow_context *subflow;
1586 struct sockaddr_storage addr;
1587 int remote_id = remote->id;
1588 int local_id = loc->id;
1589 int err = -ENOTCONN;
1590 struct socket *sf;
1591 struct sock *ssk;
1592 u32 remote_token;
1593 int addrlen;
1594 int ifindex;
1595 u8 flags;
1596
1597 if (!mptcp_is_fully_established(sk))
1598 goto err_out;
1599
1600 err = mptcp_subflow_create_socket(sk, loc->family, &sf);
1601 if (err)
1602 goto err_out;
1603
1604 ssk = sf->sk;
1605 subflow = mptcp_subflow_ctx(ssk);
1606 do {
1607 get_random_bytes(&subflow->local_nonce, sizeof(u32));
1608 } while (!subflow->local_nonce);
1609
1610 if (local_id)
1611 subflow_set_local_id(subflow, local_id);
1612
1613 mptcp_pm_get_flags_and_ifindex_by_id(msk, local_id,
1614 &flags, &ifindex);
1615 subflow->remote_key_valid = 1;
1616 subflow->remote_key = READ_ONCE(msk->remote_key);
1617 subflow->local_key = READ_ONCE(msk->local_key);
1618 subflow->token = msk->token;
1619 mptcp_info2sockaddr(loc, &addr, ssk->sk_family);
1620
1621 addrlen = sizeof(struct sockaddr_in);
1622 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1623 if (addr.ss_family == AF_INET6)
1624 addrlen = sizeof(struct sockaddr_in6);
1625 #endif
1626 ssk->sk_bound_dev_if = ifindex;
1627 err = kernel_bind(sf, (struct sockaddr *)&addr, addrlen);
1628 if (err)
1629 goto failed;
1630
1631 mptcp_crypto_key_sha(subflow->remote_key, &remote_token, NULL);
1632 pr_debug("msk=%p remote_token=%u local_id=%d remote_id=%d\n", msk,
1633 remote_token, local_id, remote_id);
1634 subflow->remote_token = remote_token;
1635 WRITE_ONCE(subflow->remote_id, remote_id);
1636 subflow->request_join = 1;
1637 subflow->request_bkup = !!(flags & MPTCP_PM_ADDR_FLAG_BACKUP);
1638 subflow->subflow_id = msk->subflow_id++;
1639 mptcp_info2sockaddr(remote, &addr, ssk->sk_family);
1640
1641 sock_hold(ssk);
1642 list_add_tail(&subflow->node, &msk->conn_list);
1643 err = kernel_connect(sf, (struct sockaddr *)&addr, addrlen, O_NONBLOCK);
1644 if (err && err != -EINPROGRESS)
1645 goto failed_unlink;
1646
1647 /* discard the subflow socket */
1648 mptcp_sock_graft(ssk, sk->sk_socket);
1649 iput(SOCK_INODE(sf));
1650 WRITE_ONCE(msk->allow_infinite_fallback, false);
1651 mptcp_stop_tout_timer(sk);
1652 return 0;
1653
1654 failed_unlink:
1655 list_del(&subflow->node);
1656 sock_put(mptcp_subflow_tcp_sock(subflow));
1657
1658 failed:
1659 subflow->disposable = 1;
1660 sock_release(sf);
1661
1662 err_out:
1663 /* we account subflows before the creation, and this failures will not
1664 * be caught by sk_state_change()
1665 */
1666 mptcp_pm_close_subflow(msk);
1667 return err;
1668 }
1669
1670 static void mptcp_attach_cgroup(struct sock *parent, struct sock *child)
1671 {
1672 #ifdef CONFIG_SOCK_CGROUP_DATA
1673 struct sock_cgroup_data *parent_skcd = &parent->sk_cgrp_data,
1674 *child_skcd = &child->sk_cgrp_data;
1675
1676 /* only the additional subflows created by kworkers have to be modified */
1677 if (cgroup_id(sock_cgroup_ptr(parent_skcd)) !=
1678 cgroup_id(sock_cgroup_ptr(child_skcd))) {
1679 #ifdef CONFIG_MEMCG
1680 struct mem_cgroup *memcg = parent->sk_memcg;
1681
1682 mem_cgroup_sk_free(child);
1683 if (memcg && css_tryget(&memcg->css))
1684 child->sk_memcg = memcg;
1685 #endif /* CONFIG_MEMCG */
1686
1687 cgroup_sk_free(child_skcd);
1688 *child_skcd = *parent_skcd;
1689 cgroup_sk_clone(child_skcd);
1690 }
1691 #endif /* CONFIG_SOCK_CGROUP_DATA */
1692 }
1693
1694 static void mptcp_subflow_ops_override(struct sock *ssk)
1695 {
1696 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1697 if (ssk->sk_prot == &tcpv6_prot)
1698 ssk->sk_prot = &tcpv6_prot_override;
1699 else
1700 #endif
1701 ssk->sk_prot = &tcp_prot_override;
1702 }
1703
1704 static void mptcp_subflow_ops_undo_override(struct sock *ssk)
1705 {
1706 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1707 if (ssk->sk_prot == &tcpv6_prot_override)
1708 ssk->sk_prot = &tcpv6_prot;
1709 else
1710 #endif
1711 ssk->sk_prot = &tcp_prot;
1712 }
1713
1714 int mptcp_subflow_create_socket(struct sock *sk, unsigned short family,
1715 struct socket **new_sock)
1716 {
1717 struct mptcp_subflow_context *subflow;
1718 struct net *net = sock_net(sk);
1719 struct socket *sf;
1720 int err;
1721
1722 /* un-accepted server sockets can reach here - on bad configuration
1723 * bail early to avoid greater trouble later
1724 */
1725 if (unlikely(!sk->sk_socket))
1726 return -EINVAL;
1727
1728 err = sock_create_kern(net, family, SOCK_STREAM, IPPROTO_TCP, &sf);
1729 if (err)
1730 return err;
1731
1732 lock_sock_nested(sf->sk, SINGLE_DEPTH_NESTING);
1733
1734 err = security_mptcp_add_subflow(sk, sf->sk);
1735 if (err)
1736 goto err_free;
1737
1738 /* the newly created socket has to be in the same cgroup as its parent */
1739 mptcp_attach_cgroup(sk, sf->sk);
1740
1741 /* kernel sockets do not by default acquire net ref, but TCP timer
1742 * needs it.
1743 * Update ns_tracker to current stack trace and refcounted tracker.
1744 */
1745 __netns_tracker_free(net, &sf->sk->ns_tracker, false);
1746 sf->sk->sk_net_refcnt = 1;
1747 get_net_track(net, &sf->sk->ns_tracker, GFP_KERNEL);
1748 sock_inuse_add(net, 1);
1749 err = tcp_set_ulp(sf->sk, "mptcp");
1750 if (err)
1751 goto err_free;
1752
1753 mptcp_sockopt_sync_locked(mptcp_sk(sk), sf->sk);
1754 release_sock(sf->sk);
1755
1756 /* the newly created socket really belongs to the owning MPTCP
1757 * socket, even if for additional subflows the allocation is performed
1758 * by a kernel workqueue. Adjust inode references, so that the
1759 * procfs/diag interfaces really show this one belonging to the correct
1760 * user.
1761 */
1762 SOCK_INODE(sf)->i_ino = SOCK_INODE(sk->sk_socket)->i_ino;
1763 SOCK_INODE(sf)->i_uid = SOCK_INODE(sk->sk_socket)->i_uid;
1764 SOCK_INODE(sf)->i_gid = SOCK_INODE(sk->sk_socket)->i_gid;
1765
1766 subflow = mptcp_subflow_ctx(sf->sk);
1767 pr_debug("subflow=%p\n", subflow);
1768
1769 *new_sock = sf;
1770 sock_hold(sk);
1771 subflow->conn = sk;
1772 mptcp_subflow_ops_override(sf->sk);
1773
1774 return 0;
1775
1776 err_free:
1777 release_sock(sf->sk);
1778 sock_release(sf);
1779 return err;
1780 }
1781
1782 static struct mptcp_subflow_context *subflow_create_ctx(struct sock *sk,
1783 gfp_t priority)
1784 {
1785 struct inet_connection_sock *icsk = inet_csk(sk);
1786 struct mptcp_subflow_context *ctx;
1787
1788 ctx = kzalloc(sizeof(*ctx), priority);
1789 if (!ctx)
1790 return NULL;
1791
1792 rcu_assign_pointer(icsk->icsk_ulp_data, ctx);
1793 INIT_LIST_HEAD(&ctx->node);
1794 INIT_LIST_HEAD(&ctx->delegated_node);
1795
1796 pr_debug("subflow=%p\n", ctx);
1797
1798 ctx->tcp_sock = sk;
1799 WRITE_ONCE(ctx->local_id, -1);
1800
1801 return ctx;
1802 }
1803
1804 static void __subflow_state_change(struct sock *sk)
1805 {
1806 struct socket_wq *wq;
1807
1808 rcu_read_lock();
1809 wq = rcu_dereference(sk->sk_wq);
1810 if (skwq_has_sleeper(wq))
1811 wake_up_interruptible_all(&wq->wait);
1812 rcu_read_unlock();
1813 }
1814
1815 static bool subflow_is_done(const struct sock *sk)
1816 {
1817 return sk->sk_shutdown & RCV_SHUTDOWN || sk->sk_state == TCP_CLOSE;
1818 }
1819
1820 static void subflow_state_change(struct sock *sk)
1821 {
1822 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1823 struct sock *parent = subflow->conn;
1824 struct mptcp_sock *msk;
1825
1826 __subflow_state_change(sk);
1827
1828 msk = mptcp_sk(parent);
1829 if (subflow_simultaneous_connect(sk)) {
1830 mptcp_do_fallback(sk);
1831 pr_fallback(msk);
1832 subflow->conn_finished = 1;
1833 mptcp_propagate_state(parent, sk, subflow, NULL);
1834 }
1835
1836 /* as recvmsg() does not acquire the subflow socket for ssk selection
1837 * a fin packet carrying a DSS can be unnoticed if we don't trigger
1838 * the data available machinery here.
1839 */
1840 if (mptcp_subflow_data_available(sk))
1841 mptcp_data_ready(parent, sk);
1842 else if (unlikely(sk->sk_err))
1843 subflow_error_report(sk);
1844
1845 subflow_sched_work_if_closed(mptcp_sk(parent), sk);
1846
1847 /* when the fallback subflow closes the rx side, trigger a 'dummy'
1848 * ingress data fin, so that the msk state will follow along
1849 */
1850 if (__mptcp_check_fallback(msk) && subflow_is_done(sk) && msk->first == sk &&
1851 mptcp_update_rcv_data_fin(msk, READ_ONCE(msk->ack_seq), true))
1852 mptcp_schedule_work(parent);
1853 }
1854
1855 void mptcp_subflow_queue_clean(struct sock *listener_sk, struct sock *listener_ssk)
1856 {
1857 struct request_sock_queue *queue = &inet_csk(listener_ssk)->icsk_accept_queue;
1858 struct request_sock *req, *head, *tail;
1859 struct mptcp_subflow_context *subflow;
1860 struct sock *sk, *ssk;
1861
1862 /* Due to lock dependencies no relevant lock can be acquired under rskq_lock.
1863 * Splice the req list, so that accept() can not reach the pending ssk after
1864 * the listener socket is released below.
1865 */
1866 spin_lock_bh(&queue->rskq_lock);
1867 head = queue->rskq_accept_head;
1868 tail = queue->rskq_accept_tail;
1869 queue->rskq_accept_head = NULL;
1870 queue->rskq_accept_tail = NULL;
1871 spin_unlock_bh(&queue->rskq_lock);
1872
1873 if (!head)
1874 return;
1875
1876 /* can't acquire the msk socket lock under the subflow one,
1877 * or will cause ABBA deadlock
1878 */
1879 release_sock(listener_ssk);
1880
1881 for (req = head; req; req = req->dl_next) {
1882 ssk = req->sk;
1883 if (!sk_is_mptcp(ssk))
1884 continue;
1885
1886 subflow = mptcp_subflow_ctx(ssk);
1887 if (!subflow || !subflow->conn)
1888 continue;
1889
1890 sk = subflow->conn;
1891 sock_hold(sk);
1892
1893 lock_sock_nested(sk, SINGLE_DEPTH_NESTING);
1894 __mptcp_unaccepted_force_close(sk);
1895 release_sock(sk);
1896
1897 /* lockdep will report a false positive ABBA deadlock
1898 * between cancel_work_sync and the listener socket.
1899 * The involved locks belong to different sockets WRT
1900 * the existing AB chain.
1901 * Using a per socket key is problematic as key
1902 * deregistration requires process context and must be
1903 * performed at socket disposal time, in atomic
1904 * context.
1905 * Just tell lockdep to consider the listener socket
1906 * released here.
1907 */
1908 mutex_release(&listener_sk->sk_lock.dep_map, _RET_IP_);
1909 mptcp_cancel_work(sk);
1910 mutex_acquire(&listener_sk->sk_lock.dep_map, 0, 0, _RET_IP_);
1911
1912 sock_put(sk);
1913 }
1914
1915 /* we are still under the listener msk socket lock */
1916 lock_sock_nested(listener_ssk, SINGLE_DEPTH_NESTING);
1917
1918 /* restore the listener queue, to let the TCP code clean it up */
1919 spin_lock_bh(&queue->rskq_lock);
1920 WARN_ON_ONCE(queue->rskq_accept_head);
1921 queue->rskq_accept_head = head;
1922 queue->rskq_accept_tail = tail;
1923 spin_unlock_bh(&queue->rskq_lock);
1924 }
1925
1926 static int subflow_ulp_init(struct sock *sk)
1927 {
1928 struct inet_connection_sock *icsk = inet_csk(sk);
1929 struct mptcp_subflow_context *ctx;
1930 struct tcp_sock *tp = tcp_sk(sk);
1931 int err = 0;
1932
1933 /* disallow attaching ULP to a socket unless it has been
1934 * created with sock_create_kern()
1935 */
1936 if (!sk->sk_kern_sock) {
1937 err = -EOPNOTSUPP;
1938 goto out;
1939 }
1940
1941 ctx = subflow_create_ctx(sk, GFP_KERNEL);
1942 if (!ctx) {
1943 err = -ENOMEM;
1944 goto out;
1945 }
1946
1947 pr_debug("subflow=%p, family=%d\n", ctx, sk->sk_family);
1948
1949 tp->is_mptcp = 1;
1950 ctx->icsk_af_ops = icsk->icsk_af_ops;
1951 icsk->icsk_af_ops = subflow_default_af_ops(sk);
1952 ctx->tcp_state_change = sk->sk_state_change;
1953 ctx->tcp_error_report = sk->sk_error_report;
1954
1955 WARN_ON_ONCE(sk->sk_data_ready != sock_def_readable);
1956 WARN_ON_ONCE(sk->sk_write_space != sk_stream_write_space);
1957
1958 sk->sk_data_ready = subflow_data_ready;
1959 sk->sk_write_space = subflow_write_space;
1960 sk->sk_state_change = subflow_state_change;
1961 sk->sk_error_report = subflow_error_report;
1962 out:
1963 return err;
1964 }
1965
1966 static void subflow_ulp_release(struct sock *ssk)
1967 {
1968 struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(ssk);
1969 bool release = true;
1970 struct sock *sk;
1971
1972 if (!ctx)
1973 return;
1974
1975 sk = ctx->conn;
1976 if (sk) {
1977 /* if the msk has been orphaned, keep the ctx
1978 * alive, will be freed by __mptcp_close_ssk(),
1979 * when the subflow is still unaccepted
1980 */
1981 release = ctx->disposable || list_empty(&ctx->node);
1982
1983 /* inet_child_forget() does not call sk_state_change(),
1984 * explicitly trigger the socket close machinery
1985 */
1986 if (!release && !test_and_set_bit(MPTCP_WORK_CLOSE_SUBFLOW,
1987 &mptcp_sk(sk)->flags))
1988 mptcp_schedule_work(sk);
1989 sock_put(sk);
1990 }
1991
1992 mptcp_subflow_ops_undo_override(ssk);
1993 if (release)
1994 kfree_rcu(ctx, rcu);
1995 }
1996
1997 static void subflow_ulp_clone(const struct request_sock *req,
1998 struct sock *newsk,
1999 const gfp_t priority)
2000 {
2001 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
2002 struct mptcp_subflow_context *old_ctx = mptcp_subflow_ctx(newsk);
2003 struct mptcp_subflow_context *new_ctx;
2004
2005 if (!tcp_rsk(req)->is_mptcp ||
2006 (!subflow_req->mp_capable && !subflow_req->mp_join)) {
2007 subflow_ulp_fallback(newsk, old_ctx);
2008 return;
2009 }
2010
2011 new_ctx = subflow_create_ctx(newsk, priority);
2012 if (!new_ctx) {
2013 subflow_ulp_fallback(newsk, old_ctx);
2014 return;
2015 }
2016
2017 new_ctx->conn_finished = 1;
2018 new_ctx->icsk_af_ops = old_ctx->icsk_af_ops;
2019 new_ctx->tcp_state_change = old_ctx->tcp_state_change;
2020 new_ctx->tcp_error_report = old_ctx->tcp_error_report;
2021 new_ctx->rel_write_seq = 1;
2022 new_ctx->tcp_sock = newsk;
2023
2024 if (subflow_req->mp_capable) {
2025 /* see comments in subflow_syn_recv_sock(), MPTCP connection
2026 * is fully established only after we receive the remote key
2027 */
2028 new_ctx->mp_capable = 1;
2029 new_ctx->local_key = subflow_req->local_key;
2030 new_ctx->token = subflow_req->token;
2031 new_ctx->ssn_offset = subflow_req->ssn_offset;
2032 new_ctx->idsn = subflow_req->idsn;
2033
2034 /* this is the first subflow, id is always 0 */
2035 subflow_set_local_id(new_ctx, 0);
2036 } else if (subflow_req->mp_join) {
2037 new_ctx->ssn_offset = subflow_req->ssn_offset;
2038 new_ctx->mp_join = 1;
2039 new_ctx->fully_established = 1;
2040 new_ctx->remote_key_valid = 1;
2041 new_ctx->backup = subflow_req->backup;
2042 new_ctx->request_bkup = subflow_req->request_bkup;
2043 WRITE_ONCE(new_ctx->remote_id, subflow_req->remote_id);
2044 new_ctx->token = subflow_req->token;
2045 new_ctx->thmac = subflow_req->thmac;
2046
2047 /* the subflow req id is valid, fetched via subflow_check_req()
2048 * and subflow_token_join_request()
2049 */
2050 subflow_set_local_id(new_ctx, subflow_req->local_id);
2051 }
2052 }
2053
2054 static void tcp_release_cb_override(struct sock *ssk)
2055 {
2056 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
2057 long status;
2058
2059 /* process and clear all the pending actions, but leave the subflow into
2060 * the napi queue. To respect locking, only the same CPU that originated
2061 * the action can touch the list. mptcp_napi_poll will take care of it.
2062 */
2063 status = set_mask_bits(&subflow->delegated_status, MPTCP_DELEGATE_ACTIONS_MASK, 0);
2064 if (status)
2065 mptcp_subflow_process_delegated(ssk, status);
2066
2067 tcp_release_cb(ssk);
2068 }
2069
2070 static int tcp_abort_override(struct sock *ssk, int err)
2071 {
2072 /* closing a listener subflow requires a great deal of care.
2073 * keep it simple and just prevent such operation
2074 */
2075 if (inet_sk_state_load(ssk) == TCP_LISTEN)
2076 return -EINVAL;
2077
2078 return tcp_abort(ssk, err);
2079 }
2080
2081 static struct tcp_ulp_ops subflow_ulp_ops __read_mostly = {
2082 .name = "mptcp",
2083 .owner = THIS_MODULE,
2084 .init = subflow_ulp_init,
2085 .release = subflow_ulp_release,
2086 .clone = subflow_ulp_clone,
2087 };
2088
2089 static int subflow_ops_init(struct request_sock_ops *subflow_ops)
2090 {
2091 subflow_ops->obj_size = sizeof(struct mptcp_subflow_request_sock);
2092
2093 subflow_ops->slab = kmem_cache_create(subflow_ops->slab_name,
2094 subflow_ops->obj_size, 0,
2095 SLAB_ACCOUNT |
2096 SLAB_TYPESAFE_BY_RCU,
2097 NULL);
2098 if (!subflow_ops->slab)
2099 return -ENOMEM;
2100
2101 return 0;
2102 }
2103
2104 void __init mptcp_subflow_init(void)
2105 {
2106 mptcp_subflow_v4_request_sock_ops = tcp_request_sock_ops;
2107 mptcp_subflow_v4_request_sock_ops.slab_name = "request_sock_subflow_v4";
2108 mptcp_subflow_v4_request_sock_ops.destructor = subflow_v4_req_destructor;
2109
2110 if (subflow_ops_init(&mptcp_subflow_v4_request_sock_ops) != 0)
2111 panic("MPTCP: failed to init subflow v4 request sock ops\n");
2112
2113 subflow_request_sock_ipv4_ops = tcp_request_sock_ipv4_ops;
2114 subflow_request_sock_ipv4_ops.route_req = subflow_v4_route_req;
2115 subflow_request_sock_ipv4_ops.send_synack = subflow_v4_send_synack;
2116
2117 subflow_specific = ipv4_specific;
2118 subflow_specific.conn_request = subflow_v4_conn_request;
2119 subflow_specific.syn_recv_sock = subflow_syn_recv_sock;
2120 subflow_specific.sk_rx_dst_set = subflow_finish_connect;
2121 subflow_specific.rebuild_header = subflow_rebuild_header;
2122
2123 tcp_prot_override = tcp_prot;
2124 tcp_prot_override.release_cb = tcp_release_cb_override;
2125 tcp_prot_override.diag_destroy = tcp_abort_override;
2126
2127 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
2128 /* In struct mptcp_subflow_request_sock, we assume the TCP request sock
2129 * structures for v4 and v6 have the same size. It should not changed in
2130 * the future but better to make sure to be warned if it is no longer
2131 * the case.
2132 */
2133 BUILD_BUG_ON(sizeof(struct tcp_request_sock) != sizeof(struct tcp6_request_sock));
2134
2135 mptcp_subflow_v6_request_sock_ops = tcp6_request_sock_ops;
2136 mptcp_subflow_v6_request_sock_ops.slab_name = "request_sock_subflow_v6";
2137 mptcp_subflow_v6_request_sock_ops.destructor = subflow_v6_req_destructor;
2138
2139 if (subflow_ops_init(&mptcp_subflow_v6_request_sock_ops) != 0)
2140 panic("MPTCP: failed to init subflow v6 request sock ops\n");
2141
2142 subflow_request_sock_ipv6_ops = tcp_request_sock_ipv6_ops;
2143 subflow_request_sock_ipv6_ops.route_req = subflow_v6_route_req;
2144 subflow_request_sock_ipv6_ops.send_synack = subflow_v6_send_synack;
2145
2146 subflow_v6_specific = ipv6_specific;
2147 subflow_v6_specific.conn_request = subflow_v6_conn_request;
2148 subflow_v6_specific.syn_recv_sock = subflow_syn_recv_sock;
2149 subflow_v6_specific.sk_rx_dst_set = subflow_finish_connect;
2150 subflow_v6_specific.rebuild_header = subflow_v6_rebuild_header;
2151
2152 subflow_v6m_specific = subflow_v6_specific;
2153 subflow_v6m_specific.queue_xmit = ipv4_specific.queue_xmit;
2154 subflow_v6m_specific.send_check = ipv4_specific.send_check;
2155 subflow_v6m_specific.net_header_len = ipv4_specific.net_header_len;
2156 subflow_v6m_specific.mtu_reduced = ipv4_specific.mtu_reduced;
2157 subflow_v6m_specific.rebuild_header = subflow_rebuild_header;
2158
2159 tcpv6_prot_override = tcpv6_prot;
2160 tcpv6_prot_override.release_cb = tcp_release_cb_override;
2161 tcpv6_prot_override.diag_destroy = tcp_abort_override;
2162 #endif
2163
2164 mptcp_diag_subflow_init(&subflow_ulp_ops);
2165
2166 if (tcp_register_ulp(&subflow_ulp_ops) != 0)
2167 panic("MPTCP: failed to register subflows to ULP\n");
2168 }
2169
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.