xref: /linux/net/mptcp/subflow.c (revision c36461825469a9ceee2346a2e89286c522525da7)
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 	subflow_req->local_nonce = get_random_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 		SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MPJOINNOIDFOUND);
100 		sock_put((struct sock *)msk);
101 		return NULL;
102 	}
103 	subflow_req->local_id = local_id;
104 	subflow_req->request_bkup = mptcp_pm_is_backup(msk, (struct sock_common *)req);
105 
106 	return msk;
107 }
108 
109 static void subflow_init_req(struct request_sock *req, const struct sock *sk_listener)
110 {
111 	struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
112 
113 	subflow_req->mp_capable = 0;
114 	subflow_req->mp_join = 0;
115 	subflow_req->csum_reqd = mptcp_is_checksum_enabled(sock_net(sk_listener));
116 	subflow_req->allow_join_id0 = mptcp_allow_join_id0(sock_net(sk_listener));
117 	subflow_req->msk = NULL;
118 	mptcp_token_init_request(req);
119 }
120 
121 static bool subflow_use_different_sport(struct mptcp_sock *msk, const struct sock *sk)
122 {
123 	return inet_sk(sk)->inet_sport != inet_sk((struct sock *)msk)->inet_sport;
124 }
125 
126 static void subflow_add_reset_reason(struct sk_buff *skb, u8 reason)
127 {
128 	struct mptcp_ext *mpext = skb_ext_add(skb, SKB_EXT_MPTCP);
129 
130 	if (mpext) {
131 		memset(mpext, 0, sizeof(*mpext));
132 		mpext->reset_reason = reason;
133 	}
134 }
135 
136 static int subflow_reset_req_endp(struct request_sock *req, struct sk_buff *skb)
137 {
138 	SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MPCAPABLEENDPATTEMPT);
139 	subflow_add_reset_reason(skb, MPTCP_RST_EPROHIBIT);
140 	return -EPERM;
141 }
142 
143 /* Init mptcp request socket.
144  *
145  * Returns an error code if a JOIN has failed and a TCP reset
146  * should be sent.
147  */
148 static int subflow_check_req(struct request_sock *req,
149 			     const struct sock *sk_listener,
150 			     struct sk_buff *skb)
151 {
152 	struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk_listener);
153 	struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
154 	struct mptcp_options_received mp_opt;
155 	bool opt_mp_capable, opt_mp_join;
156 
157 	pr_debug("subflow_req=%p, listener=%p\n", subflow_req, listener);
158 
159 #ifdef CONFIG_TCP_MD5SIG
160 	/* no MPTCP if MD5SIG is enabled on this socket or we may run out of
161 	 * TCP option space.
162 	 */
163 	if (rcu_access_pointer(tcp_sk(sk_listener)->md5sig_info)) {
164 		MPTCP_INC_STATS(sock_net(sk_listener), MPTCP_MIB_MD5SIGRESET);
165 		subflow_add_reset_reason(skb, MPTCP_RST_EMPTCP);
166 		return -EINVAL;
167 	}
168 #endif
169 
170 	mptcp_get_options(skb, &mp_opt);
171 
172 	opt_mp_capable = !!(mp_opt.suboptions & OPTION_MPTCP_MPC_SYN);
173 	opt_mp_join = !!(mp_opt.suboptions & OPTION_MPTCP_MPJ_SYN);
174 	if (opt_mp_capable) {
175 		SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MPCAPABLEPASSIVE);
176 
177 		if (unlikely(listener->pm_listener))
178 			return subflow_reset_req_endp(req, skb);
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_announced_has_ssk(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_REQ_INC_STATS(req, MPTCP_MIB_JOINREJECTED);
251 				subflow_add_reset_reason(skb, MPTCP_RST_EPROHIBIT);
252 				return -EPERM;
253 			}
254 
255 			subflow_init_req_cookie_join_save(subflow_req, skb);
256 		}
257 
258 		pr_debug("token=%u, remote_nonce=%u msk=%p\n", subflow_req->token,
259 			 subflow_req->remote_nonce, subflow_req->msk);
260 	}
261 
262 	return 0;
263 }
264 
265 int mptcp_subflow_init_cookie_req(struct request_sock *req,
266 				  const struct sock *sk_listener,
267 				  struct sk_buff *skb)
268 {
269 	struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk_listener);
270 	struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
271 	struct mptcp_options_received mp_opt;
272 	bool opt_mp_capable, opt_mp_join;
273 	int err;
274 
275 	subflow_init_req(req, sk_listener);
276 	mptcp_get_options(skb, &mp_opt);
277 
278 	opt_mp_capable = !!(mp_opt.suboptions & OPTION_MPTCP_MPC_ACK);
279 	opt_mp_join = !!(mp_opt.suboptions & OPTION_MPTCP_MPJ_ACK);
280 	if (opt_mp_capable && listener->request_mptcp) {
281 		if (mp_opt.sndr_key == 0)
282 			return -EINVAL;
283 
284 		subflow_req->local_key = mp_opt.rcvr_key;
285 		err = mptcp_token_new_request(req);
286 		if (err)
287 			return err;
288 
289 		subflow_req->mp_capable = 1;
290 		subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq - 1;
291 	} else if (opt_mp_join && listener->request_mptcp) {
292 		if (!mptcp_token_join_cookie_init_state(subflow_req, skb))
293 			return -EINVAL;
294 
295 		subflow_req->mp_join = 1;
296 		subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq - 1;
297 	}
298 
299 	return 0;
300 }
301 EXPORT_SYMBOL_GPL(mptcp_subflow_init_cookie_req);
302 
303 static enum sk_rst_reason mptcp_get_rst_reason(const struct sk_buff *skb)
304 {
305 	const struct mptcp_ext *mpext = mptcp_get_ext(skb);
306 
307 	if (!mpext)
308 		return SK_RST_REASON_NOT_SPECIFIED;
309 
310 	return sk_rst_convert_mptcp_reason(mpext->reset_reason);
311 }
312 
313 static struct dst_entry *subflow_v4_route_req(const struct sock *sk,
314 					      struct sk_buff *skb,
315 					      struct flowi *fl,
316 					      struct request_sock *req,
317 					      u32 tw_isn)
318 {
319 	struct dst_entry *dst;
320 	int err;
321 
322 	tcp_rsk(req)->is_mptcp = 1;
323 	subflow_init_req(req, sk);
324 
325 	dst = tcp_request_sock_ipv4_ops.route_req(sk, skb, fl, req, tw_isn);
326 	if (!dst)
327 		return NULL;
328 
329 	err = subflow_check_req(req, sk, skb);
330 	if (err == 0)
331 		return dst;
332 
333 	dst_release(dst);
334 	if (!req->syncookie)
335 		tcp_request_sock_ops.send_reset(sk, skb,
336 						mptcp_get_rst_reason(skb));
337 	return NULL;
338 }
339 
340 static void subflow_prep_synack(const struct sock *sk, struct request_sock *req,
341 				struct tcp_fastopen_cookie *foc,
342 				enum tcp_synack_type synack_type)
343 {
344 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
345 	struct inet_request_sock *ireq = inet_rsk(req);
346 
347 	/* clear tstamp_ok, as needed depending on cookie */
348 	if (foc && foc->len > -1)
349 		ireq->tstamp_ok = 0;
350 
351 	if (synack_type == TCP_SYNACK_FASTOPEN)
352 		mptcp_fastopen_subflow_synack_set_params(subflow, req);
353 }
354 
355 static int subflow_v4_send_synack(const struct sock *sk, struct dst_entry *dst,
356 				  struct flowi *fl,
357 				  struct request_sock *req,
358 				  struct tcp_fastopen_cookie *foc,
359 				  enum tcp_synack_type synack_type,
360 				  struct sk_buff *syn_skb)
361 {
362 	subflow_prep_synack(sk, req, foc, synack_type);
363 
364 	return tcp_request_sock_ipv4_ops.send_synack(sk, dst, fl, req, foc,
365 						     synack_type, syn_skb);
366 }
367 
368 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
369 static int subflow_v6_send_synack(const struct sock *sk, struct dst_entry *dst,
370 				  struct flowi *fl,
371 				  struct request_sock *req,
372 				  struct tcp_fastopen_cookie *foc,
373 				  enum tcp_synack_type synack_type,
374 				  struct sk_buff *syn_skb)
375 {
376 	subflow_prep_synack(sk, req, foc, synack_type);
377 
378 	return tcp_request_sock_ipv6_ops.send_synack(sk, dst, fl, req, foc,
379 						     synack_type, syn_skb);
380 }
381 
382 static struct dst_entry *subflow_v6_route_req(const struct sock *sk,
383 					      struct sk_buff *skb,
384 					      struct flowi *fl,
385 					      struct request_sock *req,
386 					      u32 tw_isn)
387 {
388 	struct dst_entry *dst;
389 	int err;
390 
391 	tcp_rsk(req)->is_mptcp = 1;
392 	subflow_init_req(req, sk);
393 
394 	dst = tcp_request_sock_ipv6_ops.route_req(sk, skb, fl, req, tw_isn);
395 	if (!dst)
396 		return NULL;
397 
398 	err = subflow_check_req(req, sk, skb);
399 	if (err == 0)
400 		return dst;
401 
402 	dst_release(dst);
403 	if (!req->syncookie)
404 		tcp6_request_sock_ops.send_reset(sk, skb,
405 						 mptcp_get_rst_reason(skb));
406 	return NULL;
407 }
408 #endif
409 
410 /* validate received truncated hmac and create hmac for third ACK */
411 static bool subflow_thmac_valid(struct mptcp_subflow_context *subflow)
412 {
413 	u8 hmac[SHA256_DIGEST_SIZE];
414 	u64 thmac;
415 
416 	subflow_generate_hmac(subflow->remote_key, subflow->local_key,
417 			      subflow->remote_nonce, subflow->local_nonce,
418 			      hmac);
419 
420 	thmac = get_unaligned_be64(hmac);
421 	pr_debug("subflow=%p, token=%u, thmac=%llu, subflow->thmac=%llu\n",
422 		 subflow, subflow->token, thmac, subflow->thmac);
423 
424 	return thmac == subflow->thmac;
425 }
426 
427 void mptcp_subflow_reset(struct sock *ssk)
428 {
429 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
430 	struct sock *sk = subflow->conn;
431 
432 	/* mptcp_mp_fail_no_response() can reach here on an already closed
433 	 * socket
434 	 */
435 	if (ssk->sk_state == TCP_CLOSE)
436 		return;
437 
438 	/* must hold: tcp_done() could drop last reference on parent */
439 	sock_hold(sk);
440 
441 	mptcp_send_active_reset_reason(ssk);
442 	tcp_done(ssk);
443 	if (!test_and_set_bit(MPTCP_WORK_CLOSE_SUBFLOW, &mptcp_sk(sk)->flags))
444 		mptcp_schedule_work(sk);
445 
446 	sock_put(sk);
447 }
448 
449 static bool subflow_use_different_dport(struct mptcp_sock *msk, const struct sock *sk)
450 {
451 	return inet_sk(sk)->inet_dport != inet_sk((struct sock *)msk)->inet_dport;
452 }
453 
454 void __mptcp_sync_state(struct sock *sk, int state)
455 {
456 	struct mptcp_subflow_context *subflow;
457 	struct mptcp_sock *msk = mptcp_sk(sk);
458 	struct sock *ssk = msk->first;
459 
460 	subflow = mptcp_subflow_ctx(ssk);
461 	__mptcp_propagate_sndbuf(sk, ssk);
462 
463 	if (sk->sk_state == TCP_SYN_SENT) {
464 		/* subflow->idsn is always available is TCP_SYN_SENT state,
465 		 * even for the FASTOPEN scenarios
466 		 */
467 		WRITE_ONCE(msk->write_seq, subflow->idsn + 1);
468 		WRITE_ONCE(msk->snd_nxt, msk->write_seq);
469 		mptcp_set_state(sk, state);
470 		sk->sk_state_change(sk);
471 	}
472 }
473 
474 static void subflow_set_remote_key(struct mptcp_sock *msk,
475 				   struct mptcp_subflow_context *subflow,
476 				   const struct mptcp_options_received *mp_opt)
477 {
478 	/* active MPC subflow will reach here multiple times:
479 	 * at subflow_finish_connect() time and at 4th ack time
480 	 */
481 	if (subflow->remote_key_valid)
482 		return;
483 
484 	subflow->remote_key_valid = 1;
485 	subflow->remote_key = mp_opt->sndr_key;
486 	mptcp_crypto_key_sha(subflow->remote_key, NULL, &subflow->iasn);
487 	subflow->iasn++;
488 
489 	/* for fallback's sake */
490 	subflow->map_seq = subflow->iasn;
491 
492 	WRITE_ONCE(msk->remote_key, subflow->remote_key);
493 	WRITE_ONCE(msk->ack_seq, subflow->iasn);
494 	WRITE_ONCE(msk->can_ack, true);
495 	atomic64_set(&msk->rcv_wnd_sent, subflow->iasn);
496 }
497 
498 static void mptcp_propagate_state(struct sock *sk, struct sock *ssk,
499 				  struct mptcp_subflow_context *subflow,
500 				  const struct mptcp_options_received *mp_opt)
501 {
502 	struct mptcp_sock *msk = mptcp_sk(sk);
503 
504 	mptcp_data_lock(sk);
505 	if (mp_opt) {
506 		/* Options are available only in the non fallback cases
507 		 * avoid updating rx path fields otherwise
508 		 */
509 		WRITE_ONCE(msk->snd_una, subflow->idsn + 1);
510 		WRITE_ONCE(msk->wnd_end, subflow->idsn + 1 + tcp_sk(ssk)->snd_wnd);
511 		subflow_set_remote_key(msk, subflow, mp_opt);
512 	}
513 
514 	if (!sock_owned_by_user(sk)) {
515 		__mptcp_sync_state(sk, ssk->sk_state);
516 	} else {
517 		msk->pending_state = ssk->sk_state;
518 		__set_bit(MPTCP_SYNC_STATE, &msk->cb_flags);
519 	}
520 	mptcp_data_unlock(sk);
521 }
522 
523 static void subflow_finish_connect(struct sock *sk, const struct sk_buff *skb)
524 {
525 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
526 	struct mptcp_options_received mp_opt;
527 	struct sock *parent = subflow->conn;
528 	struct mptcp_sock *msk;
529 
530 	subflow->icsk_af_ops->sk_rx_dst_set(sk, skb);
531 
532 	/* be sure no special action on any packet other than syn-ack */
533 	if (subflow->conn_finished)
534 		return;
535 
536 	msk = mptcp_sk(parent);
537 	subflow->rel_write_seq = 1;
538 	subflow->conn_finished = 1;
539 	subflow->ssn_offset = TCP_SKB_CB(skb)->seq;
540 	pr_debug("subflow=%p synack seq=%x\n", subflow, subflow->ssn_offset);
541 
542 	mptcp_get_options(skb, &mp_opt);
543 	if (subflow->request_mptcp) {
544 		if (!(mp_opt.suboptions & OPTION_MPTCP_MPC_SYNACK)) {
545 			if (!mptcp_try_fallback(sk,
546 						MPTCP_MIB_MPCAPABLEACTIVEFALLBACK)) {
547 				MPTCP_INC_STATS(sock_net(sk),
548 						MPTCP_MIB_FALLBACKFAILED);
549 				goto do_reset;
550 			}
551 
552 			goto fallback;
553 		}
554 
555 		if (mp_opt.suboptions & OPTION_MPTCP_CSUMREQD)
556 			WRITE_ONCE(msk->csum_enabled, true);
557 		if (mp_opt.deny_join_id0)
558 			WRITE_ONCE(msk->pm.remote_deny_join_id0, true);
559 		subflow->mp_capable = 1;
560 		MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPCAPABLEACTIVEACK);
561 		mptcp_finish_connect(sk);
562 		mptcp_active_enable(parent);
563 		mptcp_propagate_state(parent, sk, subflow, &mp_opt);
564 	} else if (subflow->request_join) {
565 		u8 hmac[SHA256_DIGEST_SIZE];
566 
567 		if (!(mp_opt.suboptions & OPTION_MPTCP_MPJ_SYNACK)) {
568 			MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPJOINSYNACKNOMPJOIN);
569 			subflow->reset_reason = MPTCP_RST_EMPTCP;
570 			goto do_reset;
571 		}
572 
573 		subflow->backup = mp_opt.backup;
574 		subflow->thmac = mp_opt.thmac;
575 		subflow->remote_nonce = mp_opt.nonce;
576 		WRITE_ONCE(subflow->remote_id, mp_opt.join_id);
577 		pr_debug("subflow=%p, thmac=%llu, remote_nonce=%u backup=%d\n",
578 			 subflow, subflow->thmac, subflow->remote_nonce,
579 			 subflow->backup);
580 
581 		if (!subflow_thmac_valid(subflow)) {
582 			MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINSYNACKMAC);
583 			subflow->reset_reason = MPTCP_RST_EMPTCP;
584 			goto do_reset;
585 		}
586 
587 		if (!mptcp_finish_join(sk))
588 			goto do_reset;
589 
590 		subflow_generate_hmac(subflow->local_key, subflow->remote_key,
591 				      subflow->local_nonce,
592 				      subflow->remote_nonce,
593 				      hmac);
594 		memcpy(subflow->hmac, hmac, MPTCPOPT_HMAC_LEN);
595 
596 		subflow->mp_join = 1;
597 		MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINSYNACKRX);
598 
599 		if (subflow->backup)
600 			MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINSYNACKBACKUPRX);
601 
602 		if (subflow_use_different_dport(msk, sk)) {
603 			pr_debug("synack inet_dport=%d %d\n",
604 				 ntohs(inet_sk(sk)->inet_dport),
605 				 ntohs(inet_sk(parent)->inet_dport));
606 			MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINPORTSYNACKRX);
607 		}
608 	} else if (mptcp_check_fallback(sk)) {
609 		/* It looks like MPTCP is blocked, while TCP is not */
610 		if (subflow->mpc_drop)
611 			mptcp_active_disable(parent);
612 fallback:
613 		mptcp_propagate_state(parent, sk, subflow, NULL);
614 	}
615 	return;
616 
617 do_reset:
618 	subflow->reset_transient = 0;
619 	mptcp_subflow_reset(sk);
620 }
621 
622 static void subflow_set_local_id(struct mptcp_subflow_context *subflow, int local_id)
623 {
624 	WARN_ON_ONCE(local_id < 0 || local_id > 255);
625 	WRITE_ONCE(subflow->local_id, local_id);
626 }
627 
628 static int subflow_chk_local_id(struct sock *sk)
629 {
630 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
631 	struct mptcp_sock *msk = mptcp_sk(subflow->conn);
632 	int err;
633 
634 	if (likely(subflow->local_id >= 0))
635 		return 0;
636 
637 	err = mptcp_pm_get_local_id(msk, (struct sock_common *)sk);
638 	if (err < 0)
639 		return err;
640 
641 	subflow_set_local_id(subflow, err);
642 	subflow->request_bkup = mptcp_pm_is_backup(msk, (struct sock_common *)sk);
643 
644 	return 0;
645 }
646 
647 static int subflow_rebuild_header(struct sock *sk)
648 {
649 	int err = subflow_chk_local_id(sk);
650 
651 	if (unlikely(err < 0))
652 		return err;
653 
654 	return inet_sk_rebuild_header(sk);
655 }
656 
657 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
658 static int subflow_v6_rebuild_header(struct sock *sk)
659 {
660 	int err = subflow_chk_local_id(sk);
661 
662 	if (unlikely(err < 0))
663 		return err;
664 
665 	return inet6_sk_rebuild_header(sk);
666 }
667 #endif
668 
669 static struct request_sock_ops mptcp_subflow_v4_request_sock_ops __ro_after_init;
670 static struct tcp_request_sock_ops subflow_request_sock_ipv4_ops __ro_after_init;
671 
672 static int subflow_v4_conn_request(struct sock *sk, struct sk_buff *skb)
673 {
674 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
675 
676 	pr_debug("subflow=%p\n", subflow);
677 
678 	/* Never answer to SYNs sent to broadcast or multicast */
679 	if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
680 		goto drop;
681 
682 	return tcp_conn_request(&mptcp_subflow_v4_request_sock_ops,
683 				&subflow_request_sock_ipv4_ops,
684 				sk, skb);
685 drop:
686 	tcp_listendrop(sk);
687 	return 0;
688 }
689 
690 static void subflow_v4_req_destructor(struct request_sock *req)
691 {
692 	subflow_req_destructor(req);
693 	tcp_request_sock_ops.destructor(req);
694 }
695 
696 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
697 static struct request_sock_ops mptcp_subflow_v6_request_sock_ops __ro_after_init;
698 static struct tcp_request_sock_ops subflow_request_sock_ipv6_ops __ro_after_init;
699 static struct inet_connection_sock_af_ops subflow_v6_specific __ro_after_init;
700 static struct inet_connection_sock_af_ops subflow_v6m_specific __ro_after_init;
701 static struct proto tcpv6_prot_override __ro_after_init;
702 
703 static int subflow_v6_conn_request(struct sock *sk, struct sk_buff *skb)
704 {
705 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
706 
707 	pr_debug("subflow=%p\n", subflow);
708 
709 	if (skb->protocol == htons(ETH_P_IP))
710 		return subflow_v4_conn_request(sk, skb);
711 
712 	if (!ipv6_unicast_destination(skb))
713 		goto drop;
714 
715 	if (ipv6_addr_v4mapped(&ipv6_hdr(skb)->saddr)) {
716 		__IP6_INC_STATS(sock_net(sk), NULL, IPSTATS_MIB_INHDRERRORS);
717 		return 0;
718 	}
719 
720 	return tcp_conn_request(&mptcp_subflow_v6_request_sock_ops,
721 				&subflow_request_sock_ipv6_ops, sk, skb);
722 
723 drop:
724 	tcp_listendrop(sk);
725 	return 0; /* don't send reset */
726 }
727 
728 static void subflow_v6_req_destructor(struct request_sock *req)
729 {
730 	subflow_req_destructor(req);
731 	tcp6_request_sock_ops.destructor(req);
732 }
733 #endif
734 
735 struct request_sock *mptcp_subflow_reqsk_alloc(const struct request_sock_ops *ops,
736 					       struct sock *sk_listener,
737 					       bool attach_listener)
738 {
739 	if (ops->family == AF_INET)
740 		ops = &mptcp_subflow_v4_request_sock_ops;
741 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
742 	else if (ops->family == AF_INET6)
743 		ops = &mptcp_subflow_v6_request_sock_ops;
744 #endif
745 
746 	return inet_reqsk_alloc(ops, sk_listener, attach_listener);
747 }
748 EXPORT_SYMBOL(mptcp_subflow_reqsk_alloc);
749 
750 /* validate hmac received in third ACK */
751 static bool subflow_hmac_valid(const struct mptcp_subflow_request_sock *subflow_req,
752 			       const struct mptcp_options_received *mp_opt)
753 {
754 	struct mptcp_sock *msk = subflow_req->msk;
755 	u8 hmac[SHA256_DIGEST_SIZE];
756 
757 	subflow_generate_hmac(READ_ONCE(msk->remote_key),
758 			      READ_ONCE(msk->local_key),
759 			      subflow_req->remote_nonce,
760 			      subflow_req->local_nonce, hmac);
761 
762 	return !crypto_memneq(hmac, mp_opt->hmac, MPTCPOPT_HMAC_LEN);
763 }
764 
765 static void subflow_ulp_fallback(struct sock *sk,
766 				 struct mptcp_subflow_context *old_ctx)
767 {
768 	struct inet_connection_sock *icsk = inet_csk(sk);
769 
770 	mptcp_subflow_tcp_fallback(sk, old_ctx);
771 	icsk->icsk_ulp_ops = NULL;
772 	rcu_assign_pointer(icsk->icsk_ulp_data, NULL);
773 	tcp_sk(sk)->is_mptcp = 0;
774 
775 	mptcp_subflow_ops_undo_override(sk);
776 }
777 
778 void mptcp_subflow_drop_ctx(struct sock *ssk)
779 {
780 	struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(ssk);
781 
782 	if (!ctx)
783 		return;
784 
785 	list_del(&mptcp_subflow_ctx(ssk)->node);
786 	if (inet_csk(ssk)->icsk_ulp_ops) {
787 		subflow_ulp_fallback(ssk, ctx);
788 		if (ctx->conn)
789 			sock_put(ctx->conn);
790 	}
791 
792 	kfree_rcu(ctx, rcu);
793 }
794 
795 void __mptcp_subflow_fully_established(struct mptcp_sock *msk,
796 				       struct mptcp_subflow_context *subflow,
797 				       const struct mptcp_options_received *mp_opt)
798 {
799 	subflow_set_remote_key(msk, subflow, mp_opt);
800 	WRITE_ONCE(subflow->fully_established, true);
801 	WRITE_ONCE(msk->fully_established, true);
802 }
803 
804 static struct sock *subflow_syn_recv_sock(const struct sock *sk,
805 					  struct sk_buff *skb,
806 					  struct request_sock *req,
807 					  struct dst_entry *dst,
808 					  struct request_sock *req_unhash,
809 					  bool *own_req,
810 					  void (*opt_child_init)(struct sock *newsk,
811 								 const struct sock *sk))
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 			fallback = true;
854 	}
855 
856 create_child:
857 	child = listener->icsk_af_ops->syn_recv_sock(sk, skb, req, dst,
858 						     req_unhash, own_req, opt_child_init);
859 
860 	if (child && *own_req) {
861 		struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(child);
862 
863 		tcp_rsk(req)->drop_req = false;
864 
865 		/* we need to fallback on ctx allocation failure and on pre-reqs
866 		 * checking above. In the latter scenario we additionally need
867 		 * to reset the context to non MPTCP status.
868 		 */
869 		if (!ctx || fallback) {
870 			if (fallback_is_fatal) {
871 				if (!ctx)
872 					MPTCP_INC_STATS(sock_net(sk),
873 							MPTCP_MIB_MPJOINACKNOCTX);
874 				else
875 					MPTCP_INC_STATS(sock_net(sk),
876 							MPTCP_MIB_MPJOINACKNOMPJOIN);
877 				subflow_add_reset_reason(skb, MPTCP_RST_EMPTCP);
878 				goto dispose_child;
879 			}
880 			goto fallback;
881 		}
882 
883 		/* ssk inherits options of listener sk */
884 		ctx->setsockopt_seq = listener->setsockopt_seq;
885 
886 		if (ctx->mp_capable) {
887 			ctx->conn = mptcp_sk_clone_init(listener->conn, &mp_opt, child, req);
888 			if (!ctx->conn)
889 				goto fallback;
890 
891 			ctx->subflow_id = 1;
892 			owner = mptcp_sk(ctx->conn);
893 
894 			if (mp_opt.deny_join_id0)
895 				WRITE_ONCE(owner->pm.remote_deny_join_id0, true);
896 
897 			mptcp_pm_new_connection(owner, child, 1);
898 
899 			/* with OoO packets we can reach here without ingress
900 			 * mpc option
901 			 */
902 			if (mp_opt.suboptions & OPTION_MPTCP_MPC_ACK) {
903 				mptcp_pm_fully_established(owner, child);
904 				ctx->pm_notified = 1;
905 			}
906 		} else if (ctx->mp_join) {
907 			owner = subflow_req->msk;
908 			if (!owner) {
909 				subflow_add_reset_reason(skb, MPTCP_RST_EPROHIBIT);
910 				goto dispose_child;
911 			}
912 
913 			if (!subflow_hmac_valid(subflow_req, &mp_opt)) {
914 				SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKMAC);
915 				subflow_add_reset_reason(skb, MPTCP_RST_EMPTCP);
916 				goto dispose_child;
917 			}
918 
919 			if (!mptcp_can_accept_new_subflow(owner)) {
920 				SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINREJECTED);
921 				subflow_add_reset_reason(skb, MPTCP_RST_EPROHIBIT);
922 				goto dispose_child;
923 			}
924 
925 			/* move the msk reference ownership to the subflow */
926 			subflow_req->msk = NULL;
927 			ctx->conn = (struct sock *)owner;
928 
929 			if (subflow_use_different_sport(owner, sk)) {
930 				pr_debug("ack inet_sport=%d %d\n",
931 					 ntohs(inet_sk(sk)->inet_sport),
932 					 ntohs(inet_sk((struct sock *)owner)->inet_sport));
933 				if (!mptcp_pm_announced_has_ssk(owner, sk)) {
934 					SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MISMATCHPORTACKRX);
935 					subflow_add_reset_reason(skb, MPTCP_RST_EPROHIBIT);
936 					goto dispose_child;
937 				}
938 				SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINPORTACKRX);
939 			}
940 
941 			if (!mptcp_finish_join(child)) {
942 				struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(child);
943 
944 				subflow_add_reset_reason(skb, subflow->reset_reason);
945 				goto dispose_child;
946 			}
947 
948 			SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKRX);
949 			tcp_rsk(req)->drop_req = true;
950 		}
951 	}
952 
953 	/* check for expected invariant - should never trigger, just help
954 	 * catching earlier subtle bugs
955 	 */
956 	WARN_ON_ONCE(child && *own_req && tcp_sk(child)->is_mptcp &&
957 		     (!mptcp_subflow_ctx(child) ||
958 		      !mptcp_subflow_ctx(child)->conn));
959 	return child;
960 
961 dispose_child:
962 	mptcp_subflow_drop_ctx(child);
963 	tcp_rsk(req)->drop_req = true;
964 	inet_csk_prepare_for_destroy_sock(child);
965 	tcp_done(child);
966 	reason = mptcp_get_rst_reason(skb);
967 	req->rsk_ops->send_reset(sk, skb, reason);
968 
969 	/* The last child reference will be released by the caller */
970 	return child;
971 
972 fallback:
973 	if (fallback)
974 		SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MPCAPABLEPASSIVEFALLBACK);
975 	mptcp_subflow_drop_ctx(child);
976 	return child;
977 }
978 
979 static struct inet_connection_sock_af_ops subflow_specific __ro_after_init;
980 static struct proto tcp_prot_override __ro_after_init;
981 
982 enum mapping_status {
983 	MAPPING_OK,
984 	MAPPING_INVALID,
985 	MAPPING_EMPTY,
986 	MAPPING_DATA_FIN,
987 	MAPPING_DUMMY,
988 	MAPPING_BAD_CSUM,
989 	MAPPING_NODSS
990 };
991 
992 static void dbg_bad_map(struct mptcp_subflow_context *subflow, u32 ssn)
993 {
994 	pr_debug("Bad mapping: ssn=%d map_seq=%d map_data_len=%d\n",
995 		 ssn, subflow->map_subflow_seq, subflow->map_data_len);
996 }
997 
998 static bool skb_is_fully_mapped(struct sock *ssk, struct sk_buff *skb)
999 {
1000 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1001 	unsigned int skb_consumed;
1002 
1003 	skb_consumed = tcp_sk(ssk)->copied_seq - TCP_SKB_CB(skb)->seq;
1004 	if (unlikely(skb_consumed >= skb->len)) {
1005 		DEBUG_NET_WARN_ON_ONCE(1);
1006 		return true;
1007 	}
1008 
1009 	return skb->len - skb_consumed <= subflow->map_data_len -
1010 					  mptcp_subflow_get_map_offset(subflow);
1011 }
1012 
1013 static bool validate_mapping(struct sock *ssk, struct sk_buff *skb)
1014 {
1015 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1016 	u32 ssn = tcp_sk(ssk)->copied_seq - subflow->ssn_offset;
1017 
1018 	if (unlikely(before(ssn, subflow->map_subflow_seq))) {
1019 		/* Mapping covers data later in the subflow stream,
1020 		 * currently unsupported.
1021 		 */
1022 		dbg_bad_map(subflow, ssn);
1023 		return false;
1024 	}
1025 	if (unlikely(!before(ssn, subflow->map_subflow_seq +
1026 				  subflow->map_data_len))) {
1027 		/* Mapping does covers past subflow data, invalid */
1028 		dbg_bad_map(subflow, ssn);
1029 		return false;
1030 	}
1031 	return true;
1032 }
1033 
1034 static enum mapping_status validate_data_csum(struct sock *ssk, struct sk_buff *skb,
1035 					      bool csum_reqd)
1036 {
1037 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1038 	u32 offset, seq, delta;
1039 	__sum16 csum;
1040 	int len;
1041 
1042 	if (!csum_reqd)
1043 		return MAPPING_OK;
1044 
1045 	/* mapping already validated on previous traversal */
1046 	if (subflow->map_csum_len == subflow->map_data_len)
1047 		return MAPPING_OK;
1048 
1049 	/* traverse the receive queue, ensuring it contains a full
1050 	 * DSS mapping and accumulating the related csum.
1051 	 * Preserve the accoumlate csum across multiple calls, to compute
1052 	 * the csum only once
1053 	 */
1054 	delta = subflow->map_data_len - subflow->map_csum_len;
1055 	for (;;) {
1056 		seq = tcp_sk(ssk)->copied_seq + subflow->map_csum_len;
1057 		offset = seq - TCP_SKB_CB(skb)->seq;
1058 
1059 		/* if the current skb has not been accounted yet, csum its contents
1060 		 * up to the amount covered by the current DSS
1061 		 */
1062 		if (offset < skb->len) {
1063 			__wsum csum;
1064 
1065 			len = min(skb->len - offset, delta);
1066 			csum = skb_checksum(skb, offset, len, 0);
1067 			subflow->map_data_csum = csum_block_add(subflow->map_data_csum, csum,
1068 								subflow->map_csum_len);
1069 
1070 			delta -= len;
1071 			subflow->map_csum_len += len;
1072 		}
1073 		if (delta == 0)
1074 			break;
1075 
1076 		if (skb_queue_is_last(&ssk->sk_receive_queue, skb)) {
1077 			/* if this subflow is closed, the partial mapping
1078 			 * will be never completed; flush the pending skbs, so
1079 			 * that subflow_sched_work_if_closed() can kick in
1080 			 */
1081 			if (unlikely(ssk->sk_state == TCP_CLOSE))
1082 				while ((skb = skb_peek(&ssk->sk_receive_queue)))
1083 					sk_eat_skb(ssk, skb);
1084 
1085 			/* not enough data to validate the csum */
1086 			return MAPPING_EMPTY;
1087 		}
1088 
1089 		/* the DSS mapping for next skbs will be validated later,
1090 		 * when a get_mapping_status call will process such skb
1091 		 */
1092 		skb = skb->next;
1093 	}
1094 
1095 	/* note that 'map_data_len' accounts only for the carried data, does
1096 	 * not include the eventual seq increment due to the data fin,
1097 	 * while the pseudo header requires the original DSS data len,
1098 	 * including that
1099 	 */
1100 	csum = __mptcp_make_csum(subflow->map_seq,
1101 				 subflow->map_subflow_seq,
1102 				 subflow->map_data_len + subflow->map_data_fin,
1103 				 subflow->map_data_csum);
1104 	if (unlikely(csum)) {
1105 		MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DATACSUMERR);
1106 		return MAPPING_BAD_CSUM;
1107 	}
1108 
1109 	subflow->valid_csum_seen = 1;
1110 	return MAPPING_OK;
1111 }
1112 
1113 static enum mapping_status get_mapping_status(struct sock *ssk,
1114 					      struct mptcp_sock *msk)
1115 {
1116 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1117 	bool csum_reqd = READ_ONCE(msk->csum_enabled);
1118 	struct mptcp_ext *mpext;
1119 	struct sk_buff *skb;
1120 	u16 data_len;
1121 	u64 map_seq;
1122 
1123 	skb = skb_peek(&ssk->sk_receive_queue);
1124 	if (!skb)
1125 		return MAPPING_EMPTY;
1126 
1127 	if (mptcp_check_fallback(ssk))
1128 		return MAPPING_DUMMY;
1129 
1130 	mpext = mptcp_get_ext(skb);
1131 	if (!mpext || !mpext->use_map) {
1132 		if (!subflow->map_valid && !skb->len) {
1133 			/* the TCP stack deliver 0 len FIN pkt to the receive
1134 			 * queue, that is the only 0len pkts ever expected here,
1135 			 * and we can admit no mapping only for 0 len pkts
1136 			 */
1137 			if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN))
1138 				WARN_ONCE(1, "0len seq %d:%d flags %x",
1139 					  TCP_SKB_CB(skb)->seq,
1140 					  TCP_SKB_CB(skb)->end_seq,
1141 					  TCP_SKB_CB(skb)->tcp_flags);
1142 			sk_eat_skb(ssk, skb);
1143 			return MAPPING_EMPTY;
1144 		}
1145 
1146 		/* If the required DSS has likely been dropped by a middlebox */
1147 		if (!subflow->map_valid)
1148 			return MAPPING_NODSS;
1149 
1150 		goto validate_seq;
1151 	}
1152 
1153 	trace_get_mapping_status(mpext);
1154 
1155 	data_len = mpext->data_len;
1156 	if (data_len == 0) {
1157 		pr_debug("infinite mapping received\n");
1158 		MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_INFINITEMAPRX);
1159 		return MAPPING_INVALID;
1160 	}
1161 
1162 	if (mpext->data_fin == 1) {
1163 		u64 data_fin_seq;
1164 
1165 		if (data_len == 1) {
1166 			bool updated = mptcp_update_rcv_data_fin(msk, mpext->data_seq,
1167 								 mpext->dsn64);
1168 			pr_debug("DATA_FIN with no payload seq=%llu\n", mpext->data_seq);
1169 			if (subflow->map_valid) {
1170 				/* A DATA_FIN might arrive in a DSS
1171 				 * option before the previous mapping
1172 				 * has been fully consumed. Continue
1173 				 * handling the existing mapping.
1174 				 */
1175 				skb_ext_del(skb, SKB_EXT_MPTCP);
1176 				return MAPPING_OK;
1177 			}
1178 
1179 			if (updated)
1180 				mptcp_schedule_work((struct sock *)msk);
1181 
1182 			return MAPPING_DATA_FIN;
1183 		}
1184 
1185 		data_fin_seq = mpext->data_seq + data_len - 1;
1186 
1187 		/* If mpext->data_seq is a 32-bit value, data_fin_seq must also
1188 		 * be limited to 32 bits.
1189 		 */
1190 		if (!mpext->dsn64)
1191 			data_fin_seq &= GENMASK_ULL(31, 0);
1192 
1193 		mptcp_update_rcv_data_fin(msk, data_fin_seq, mpext->dsn64);
1194 		pr_debug("DATA_FIN with mapping seq=%llu dsn64=%d\n",
1195 			 data_fin_seq, mpext->dsn64);
1196 
1197 		/* Adjust for DATA_FIN using 1 byte of sequence space */
1198 		data_len--;
1199 	}
1200 
1201 	map_seq = mptcp_expand_seq(READ_ONCE(msk->ack_seq), mpext->data_seq, mpext->dsn64);
1202 	WRITE_ONCE(mptcp_sk(subflow->conn)->use_64bit_ack, !!mpext->dsn64);
1203 
1204 	if (subflow->map_valid) {
1205 		/* Allow replacing only with an identical map */
1206 		if (subflow->map_seq == map_seq &&
1207 		    subflow->map_subflow_seq == mpext->subflow_seq &&
1208 		    subflow->map_data_len == data_len &&
1209 		    subflow->map_csum_reqd == mpext->csum_reqd) {
1210 			skb_ext_del(skb, SKB_EXT_MPTCP);
1211 			goto validate_csum;
1212 		}
1213 
1214 		/* If this skb data are fully covered by the current mapping,
1215 		 * the new map would need caching, which is not supported
1216 		 */
1217 		if (skb_is_fully_mapped(ssk, skb)) {
1218 			MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DSSNOMATCH);
1219 			return MAPPING_INVALID;
1220 		}
1221 
1222 		/* will validate the next map after consuming the current one */
1223 		goto validate_csum;
1224 	}
1225 
1226 	subflow->map_seq = map_seq;
1227 	subflow->map_subflow_seq = mpext->subflow_seq;
1228 	subflow->map_data_len = data_len;
1229 	subflow->map_valid = 1;
1230 	subflow->map_data_fin = mpext->data_fin;
1231 	subflow->mpc_map = mpext->mpc_map;
1232 	subflow->map_csum_reqd = mpext->csum_reqd;
1233 	subflow->map_csum_len = 0;
1234 	subflow->map_data_csum = csum_unfold(mpext->csum);
1235 
1236 	/* Cfr RFC 8684 Section 3.3.0 */
1237 	if (unlikely(subflow->map_csum_reqd != csum_reqd))
1238 		return MAPPING_INVALID;
1239 
1240 	pr_debug("new map seq=%llu subflow_seq=%u data_len=%u csum=%d:%u\n",
1241 		 subflow->map_seq, subflow->map_subflow_seq,
1242 		 subflow->map_data_len, subflow->map_csum_reqd,
1243 		 subflow->map_data_csum);
1244 
1245 validate_seq:
1246 	/* we revalidate valid mapping on new skb, because we must ensure
1247 	 * the current skb is completely covered by the available mapping
1248 	 */
1249 	if (!validate_mapping(ssk, skb)) {
1250 		MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DSSTCPMISMATCH);
1251 		return MAPPING_INVALID;
1252 	}
1253 
1254 	skb_ext_del(skb, SKB_EXT_MPTCP);
1255 
1256 validate_csum:
1257 	return validate_data_csum(ssk, skb, csum_reqd);
1258 }
1259 
1260 static void mptcp_subflow_discard_data(struct sock *ssk, struct sk_buff *skb,
1261 				       u64 limit)
1262 {
1263 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1264 	bool fin = TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN;
1265 	struct tcp_sock *tp = tcp_sk(ssk);
1266 	u32 offset, incr, avail_len;
1267 
1268 	offset = tp->copied_seq - TCP_SKB_CB(skb)->seq;
1269 	if (WARN_ON_ONCE(offset > skb->len))
1270 		goto out;
1271 
1272 	avail_len = skb->len - offset;
1273 	incr = limit >= avail_len ? avail_len + fin : limit;
1274 
1275 	pr_debug("discarding=%d len=%d offset=%d seq=%d\n", incr, skb->len,
1276 		 offset, subflow->map_subflow_seq);
1277 	MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DUPDATA);
1278 	tcp_sk(ssk)->copied_seq += incr;
1279 
1280 out:
1281 	if (!before(tcp_sk(ssk)->copied_seq, TCP_SKB_CB(skb)->end_seq))
1282 		sk_eat_skb(ssk, skb);
1283 	if (mptcp_subflow_get_map_offset(subflow) >= subflow->map_data_len)
1284 		subflow->map_valid = 0;
1285 }
1286 
1287 static bool subflow_is_done(const struct sock *sk)
1288 {
1289 	return sk->sk_shutdown & RCV_SHUTDOWN || sk->sk_state == TCP_CLOSE;
1290 }
1291 
1292 /* sched mptcp worker for subflow cleanup if no more data is pending */
1293 static void subflow_sched_work_if_closed(struct mptcp_sock *msk, struct sock *ssk)
1294 {
1295 	const struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1296 	struct sock *sk = (struct sock *)msk;
1297 
1298 	if (likely(ssk->sk_state != TCP_CLOSE &&
1299 		   (ssk->sk_state != TCP_CLOSE_WAIT ||
1300 		    inet_sk_state_load(sk) != TCP_ESTABLISHED)))
1301 		return;
1302 
1303 	if (!skb_queue_empty(&ssk->sk_receive_queue))
1304 		return;
1305 
1306 	if (!test_and_set_bit(MPTCP_WORK_CLOSE_SUBFLOW, &msk->flags))
1307 		mptcp_schedule_work(sk);
1308 
1309 	/* when the fallback subflow closes the rx side, trigger a 'dummy'
1310 	 * ingress data fin, so that the msk state will follow along
1311 	 */
1312 	if (__mptcp_check_fallback(msk) && subflow_is_done(ssk) &&
1313 	    msk->first == ssk &&
1314 	    mptcp_update_rcv_data_fin(msk, subflow->map_seq +
1315 				      subflow->map_data_len, true))
1316 		mptcp_schedule_work(sk);
1317 }
1318 
1319 static bool mptcp_subflow_fail(struct mptcp_sock *msk, struct sock *ssk)
1320 {
1321 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1322 	unsigned long fail_tout;
1323 
1324 	/* we are really failing, prevent any later subflow join */
1325 	spin_lock_bh(&msk->fallback_lock);
1326 	if (!msk->allow_infinite_fallback) {
1327 		spin_unlock_bh(&msk->fallback_lock);
1328 		return false;
1329 	}
1330 	msk->allow_subflows = false;
1331 	spin_unlock_bh(&msk->fallback_lock);
1332 
1333 	/* graceful failure can happen only on the MPC subflow */
1334 	if (WARN_ON_ONCE(ssk != READ_ONCE(msk->first)))
1335 		return false;
1336 
1337 	/* since the close timeout take precedence on the fail one,
1338 	 * no need to start the latter when the first is already set
1339 	 */
1340 	if (sock_flag((struct sock *)msk, SOCK_DEAD))
1341 		return true;
1342 
1343 	/* we don't need extreme accuracy here, use a zero fail_tout as special
1344 	 * value meaning no fail timeout at all;
1345 	 */
1346 	fail_tout = jiffies + TCP_RTO_MAX;
1347 	if (!fail_tout)
1348 		fail_tout = 1;
1349 	WRITE_ONCE(subflow->fail_tout, fail_tout);
1350 	tcp_send_ack(ssk);
1351 
1352 	mptcp_reset_tout_timer(msk, subflow->fail_tout);
1353 	return true;
1354 }
1355 
1356 static bool subflow_check_data_avail(struct sock *ssk)
1357 {
1358 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1359 	enum mapping_status status;
1360 	struct mptcp_sock *msk;
1361 	struct sk_buff *skb;
1362 
1363 	if (!skb_peek(&ssk->sk_receive_queue))
1364 		WRITE_ONCE(subflow->data_avail, false);
1365 	if (subflow->data_avail)
1366 		return true;
1367 
1368 	msk = mptcp_sk(subflow->conn);
1369 	for (;;) {
1370 		u64 ack_seq;
1371 		u64 old_ack;
1372 
1373 		status = get_mapping_status(ssk, msk);
1374 		trace_subflow_check_data_avail(status, skb_peek(&ssk->sk_receive_queue));
1375 		if (unlikely(status == MAPPING_INVALID || status == MAPPING_DUMMY ||
1376 			     status == MAPPING_BAD_CSUM || status == MAPPING_NODSS))
1377 			goto fallback;
1378 
1379 		if (status != MAPPING_OK)
1380 			goto no_data;
1381 
1382 		skb = skb_peek(&ssk->sk_receive_queue);
1383 		if (WARN_ON_ONCE(!skb))
1384 			goto no_data;
1385 
1386 		if (unlikely(!READ_ONCE(msk->can_ack)))
1387 			goto fallback;
1388 
1389 		old_ack = READ_ONCE(msk->ack_seq);
1390 		ack_seq = mptcp_subflow_get_mapped_dsn(subflow);
1391 		pr_debug("msk ack_seq=%llx subflow ack_seq=%llx\n", old_ack,
1392 			 ack_seq);
1393 		if (unlikely(before64(ack_seq, old_ack))) {
1394 			mptcp_subflow_discard_data(ssk, skb, old_ack - ack_seq);
1395 			continue;
1396 		}
1397 
1398 		WRITE_ONCE(subflow->data_avail, true);
1399 		break;
1400 	}
1401 	return true;
1402 
1403 no_data:
1404 	subflow_sched_work_if_closed(msk, ssk);
1405 	return false;
1406 
1407 fallback:
1408 	if (!__mptcp_check_fallback(msk)) {
1409 		/* RFC 8684 section 3.7. */
1410 		if (status == MAPPING_BAD_CSUM &&
1411 		    (subflow->mp_join || subflow->valid_csum_seen)) {
1412 			subflow->send_mp_fail = 1;
1413 
1414 			if (!mptcp_subflow_fail(msk, ssk)) {
1415 				subflow->reset_transient = 0;
1416 				subflow->reset_reason = MPTCP_RST_EMIDDLEBOX;
1417 				goto reset;
1418 			}
1419 			WRITE_ONCE(subflow->data_avail, true);
1420 			return true;
1421 		}
1422 
1423 		if (!mptcp_try_fallback(ssk, MPTCP_MIB_DSSFALLBACK)) {
1424 			/* fatal protocol error, close the socket.
1425 			 * subflow_error_report() will introduce the appropriate barriers
1426 			 */
1427 			subflow->reset_transient = 0;
1428 			MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DSSRESET);
1429 			subflow->reset_reason = status == MAPPING_NODSS ?
1430 						MPTCP_RST_EMIDDLEBOX :
1431 						MPTCP_RST_EMPTCP;
1432 
1433 reset:
1434 			WRITE_ONCE(ssk->sk_err, EBADMSG);
1435 			tcp_set_state(ssk, TCP_CLOSE);
1436 			while ((skb = skb_peek(&ssk->sk_receive_queue)))
1437 				sk_eat_skb(ssk, skb);
1438 			mptcp_send_active_reset_reason(ssk);
1439 			WRITE_ONCE(subflow->data_avail, false);
1440 			return false;
1441 		}
1442 	}
1443 
1444 	skb = skb_peek(&ssk->sk_receive_queue);
1445 	subflow->map_valid = 1;
1446 	subflow->map_data_len = skb->len;
1447 	subflow->map_subflow_seq = tcp_sk(ssk)->copied_seq - subflow->ssn_offset;
1448 	subflow->map_seq = __mptcp_expand_seq(subflow->map_seq,
1449 					      subflow->iasn +
1450 					      TCP_SKB_CB(skb)->seq -
1451 					      subflow->ssn_offset - 1);
1452 	WRITE_ONCE(subflow->data_avail, true);
1453 	return true;
1454 }
1455 
1456 bool mptcp_subflow_data_available(struct sock *sk)
1457 {
1458 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1459 
1460 	/* check if current mapping is still valid */
1461 	if (subflow->map_valid &&
1462 	    mptcp_subflow_get_map_offset(subflow) >= subflow->map_data_len) {
1463 		subflow->map_valid = 0;
1464 		WRITE_ONCE(subflow->data_avail, false);
1465 
1466 		pr_debug("Done with mapping: seq=%u data_len=%u\n",
1467 			 subflow->map_subflow_seq,
1468 			 subflow->map_data_len);
1469 	}
1470 
1471 	return subflow_check_data_avail(sk);
1472 }
1473 
1474 /* If ssk has an mptcp parent socket, use the mptcp rcvbuf occupancy,
1475  * not the ssk one.
1476  *
1477  * In mptcp, rwin is about the mptcp-level connection data.
1478  *
1479  * Data that is still on the ssk rx queue can thus be ignored,
1480  * as far as mptcp peer is concerned that data is still inflight.
1481  * DSS ACK is updated when skb is moved to the mptcp rx queue.
1482  */
1483 void mptcp_space(const struct sock *ssk, int *space, int *full_space)
1484 {
1485 	const struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1486 	const struct sock *sk = subflow->conn;
1487 
1488 	*space = __mptcp_space(sk);
1489 	*full_space = mptcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf));
1490 }
1491 
1492 static void subflow_error_report(struct sock *ssk)
1493 {
1494 	struct sock *sk = mptcp_subflow_ctx(ssk)->conn;
1495 
1496 	/* bail early if this is a no-op, so that we avoid introducing a
1497 	 * problematic lockdep dependency between TCP accept queue lock
1498 	 * and msk socket spinlock
1499 	 */
1500 	if (!sk->sk_socket)
1501 		return;
1502 
1503 	mptcp_data_lock(sk);
1504 	if (!sock_owned_by_user(sk))
1505 		__mptcp_error_report(sk);
1506 	else
1507 		__set_bit(MPTCP_ERROR_REPORT,  &mptcp_sk(sk)->cb_flags);
1508 	mptcp_data_unlock(sk);
1509 }
1510 
1511 static void subflow_data_ready(struct sock *sk)
1512 {
1513 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1514 	u16 state = 1 << inet_sk_state_load(sk);
1515 	struct sock *parent = subflow->conn;
1516 	struct mptcp_sock *msk;
1517 
1518 	trace_sk_data_ready(sk);
1519 
1520 	msk = mptcp_sk(parent);
1521 	if (state & TCPF_LISTEN) {
1522 		/* MPJ subflow are removed from accept queue before reaching here,
1523 		 * avoid stray wakeups
1524 		 */
1525 		if (reqsk_queue_empty(&inet_csk(sk)->icsk_accept_queue))
1526 			return;
1527 
1528 		parent->sk_data_ready(parent);
1529 		return;
1530 	}
1531 
1532 	WARN_ON_ONCE(!__mptcp_check_fallback(msk) && !subflow->mp_capable &&
1533 		     !subflow->mp_join && !(state & TCPF_CLOSE));
1534 
1535 	if (mptcp_subflow_data_available(sk)) {
1536 		mptcp_data_ready(parent, sk);
1537 
1538 		/* subflow-level lowat test are not relevant.
1539 		 * respect the msk-level threshold eventually mandating an immediate ack
1540 		 */
1541 		if (mptcp_data_avail(msk) < parent->sk_rcvlowat &&
1542 		    (tcp_sk(sk)->rcv_nxt - tcp_sk(sk)->rcv_wup) > inet_csk(sk)->icsk_ack.rcv_mss)
1543 			inet_csk(sk)->icsk_ack.pending |= ICSK_ACK_NOW;
1544 	} else if (unlikely(sk->sk_err)) {
1545 		subflow_error_report(sk);
1546 	}
1547 }
1548 
1549 static void subflow_write_space(struct sock *ssk)
1550 {
1551 	struct sock *sk = mptcp_subflow_ctx(ssk)->conn;
1552 
1553 	mptcp_propagate_sndbuf(sk, ssk);
1554 	mptcp_write_space(sk);
1555 }
1556 
1557 static const struct inet_connection_sock_af_ops *
1558 subflow_default_af_ops(struct sock *sk)
1559 {
1560 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1561 	if (sk->sk_family == AF_INET6)
1562 		return &subflow_v6_specific;
1563 #endif
1564 	return &subflow_specific;
1565 }
1566 
1567 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1568 void mptcpv6_handle_mapped(struct sock *sk, bool mapped)
1569 {
1570 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1571 	struct inet_connection_sock *icsk = inet_csk(sk);
1572 	const struct inet_connection_sock_af_ops *target;
1573 
1574 	target = mapped ? &subflow_v6m_specific : subflow_default_af_ops(sk);
1575 
1576 	pr_debug("subflow=%p family=%d ops=%p target=%p mapped=%d\n",
1577 		 subflow, sk->sk_family, icsk->icsk_af_ops, target, mapped);
1578 
1579 	if (likely(icsk->icsk_af_ops == target))
1580 		return;
1581 
1582 	subflow->icsk_af_ops = icsk->icsk_af_ops;
1583 	icsk->icsk_af_ops = target;
1584 }
1585 #endif
1586 
1587 void mptcp_info2sockaddr(const struct mptcp_addr_info *info,
1588 			 struct sockaddr_storage *addr,
1589 			 unsigned short family)
1590 {
1591 	memset(addr, 0, sizeof(*addr));
1592 	addr->ss_family = family;
1593 	if (addr->ss_family == AF_INET) {
1594 		struct sockaddr_in *in_addr = (struct sockaddr_in *)addr;
1595 
1596 		if (info->family == AF_INET)
1597 			in_addr->sin_addr = info->addr;
1598 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1599 		else if (ipv6_addr_v4mapped(&info->addr6))
1600 			in_addr->sin_addr.s_addr = info->addr6.s6_addr32[3];
1601 #endif
1602 		in_addr->sin_port = info->port;
1603 	}
1604 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1605 	else if (addr->ss_family == AF_INET6) {
1606 		struct sockaddr_in6 *in6_addr = (struct sockaddr_in6 *)addr;
1607 
1608 		if (info->family == AF_INET)
1609 			ipv6_addr_set_v4mapped(info->addr.s_addr,
1610 					       &in6_addr->sin6_addr);
1611 		else
1612 			in6_addr->sin6_addr = info->addr6;
1613 		in6_addr->sin6_port = info->port;
1614 	}
1615 #endif
1616 }
1617 
1618 int __mptcp_subflow_connect(struct sock *sk, const struct mptcp_pm_local *local,
1619 			    const struct mptcp_addr_info *remote)
1620 {
1621 	struct mptcp_sock *msk = mptcp_sk(sk);
1622 	struct mptcp_subflow_context *subflow;
1623 	int local_id = local->addr.id;
1624 	struct sockaddr_storage addr;
1625 	int remote_id = remote->id;
1626 	int err = -ENOTCONN;
1627 	struct socket *sf;
1628 	struct sock *ssk;
1629 	u32 remote_token;
1630 	int addrlen;
1631 
1632 	/* The userspace PM sent the request too early? */
1633 	if (!mptcp_is_fully_established(sk))
1634 		goto err_out;
1635 
1636 	err = mptcp_subflow_create_socket(sk, local->addr.family, &sf);
1637 	if (err) {
1638 		MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINSYNTXCREATSKERR);
1639 		pr_debug("msk=%p local=%d remote=%d create sock error: %d\n",
1640 			 msk, local_id, remote_id, err);
1641 		goto err_out;
1642 	}
1643 
1644 	ssk = sf->sk;
1645 	subflow = mptcp_subflow_ctx(ssk);
1646 	do {
1647 		subflow->local_nonce = get_random_u32();
1648 	} while (!subflow->local_nonce);
1649 
1650 	/* if 'IPADDRANY', the ID will be set later, after the routing */
1651 	if (local->addr.family == AF_INET) {
1652 		if (!local->addr.addr.s_addr)
1653 			local_id = -1;
1654 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1655 	} else if (sk->sk_family == AF_INET6) {
1656 		if (ipv6_addr_any(&local->addr.addr6))
1657 			local_id = -1;
1658 #endif
1659 	}
1660 
1661 	if (local_id >= 0)
1662 		subflow_set_local_id(subflow, local_id);
1663 
1664 	subflow->remote_key_valid = 1;
1665 	subflow->remote_key = READ_ONCE(msk->remote_key);
1666 	subflow->local_key = READ_ONCE(msk->local_key);
1667 	subflow->token = msk->token;
1668 	mptcp_info2sockaddr(&local->addr, &addr, ssk->sk_family);
1669 
1670 	addrlen = sizeof(struct sockaddr_in);
1671 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1672 	if (addr.ss_family == AF_INET6)
1673 		addrlen = sizeof(struct sockaddr_in6);
1674 #endif
1675 	ssk->sk_bound_dev_if = local->ifindex;
1676 	err = kernel_bind(sf, (struct sockaddr_unsized *)&addr, addrlen);
1677 	if (err) {
1678 		MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINSYNTXBINDERR);
1679 		pr_debug("msk=%p local=%d remote=%d bind error: %d\n",
1680 			 msk, local_id, remote_id, err);
1681 		goto failed;
1682 	}
1683 
1684 	mptcp_crypto_key_sha(subflow->remote_key, &remote_token, NULL);
1685 	pr_debug("msk=%p remote_token=%u local_id=%d remote_id=%d\n", msk,
1686 		 remote_token, local_id, remote_id);
1687 	subflow->remote_token = remote_token;
1688 	WRITE_ONCE(subflow->remote_id, remote_id);
1689 	subflow->request_join = 1;
1690 	subflow->request_bkup = !!(local->flags & MPTCP_PM_ADDR_FLAG_BACKUP);
1691 	subflow->subflow_id = msk->subflow_id++;
1692 	mptcp_info2sockaddr(remote, &addr, ssk->sk_family);
1693 
1694 	sock_hold(ssk);
1695 	list_add_tail(&subflow->node, &msk->conn_list);
1696 	err = kernel_connect(sf, (struct sockaddr_unsized *)&addr, addrlen, O_NONBLOCK);
1697 	if (err && err != -EINPROGRESS) {
1698 		MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINSYNTXCONNECTERR);
1699 		pr_debug("msk=%p local=%d remote=%d connect error: %d\n",
1700 			 msk, local_id, remote_id, err);
1701 		goto failed_unlink;
1702 	}
1703 
1704 	MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINSYNTX);
1705 
1706 	/* discard the subflow socket */
1707 	mptcp_sock_graft(ssk, sk->sk_socket);
1708 	iput(SOCK_INODE(sf));
1709 	mptcp_stop_tout_timer(sk);
1710 	return 0;
1711 
1712 failed_unlink:
1713 	list_del(&subflow->node);
1714 	sock_put(mptcp_subflow_tcp_sock(subflow));
1715 
1716 failed:
1717 	subflow->disposable = 1;
1718 	sock_release(sf);
1719 
1720 err_out:
1721 	/* we account subflows before the creation, and this failures will not
1722 	 * be caught by sk_state_change()
1723 	 */
1724 	mptcp_pm_close_subflow(msk);
1725 	return err;
1726 }
1727 
1728 void __mptcp_inherit_memcg(struct sock *sk, struct sock *ssk, gfp_t gfp)
1729 {
1730 	/* Only if the msk has been accepted already (and not orphaned).*/
1731 	if (!mem_cgroup_sockets_enabled || !sk->sk_socket)
1732 		return;
1733 
1734 	mem_cgroup_sk_inherit(sk, ssk);
1735 	__sk_charge(ssk, gfp);
1736 }
1737 
1738 void __mptcp_inherit_cgrp_data(struct sock *sk, struct sock *ssk)
1739 {
1740 #ifdef CONFIG_SOCK_CGROUP_DATA
1741 	struct sock_cgroup_data *sk_cd = &sk->sk_cgrp_data,
1742 				*ssk_cd = &ssk->sk_cgrp_data;
1743 
1744 	/* only the additional subflows created by kworkers have to be modified */
1745 	if (cgroup_id(sock_cgroup_ptr(sk_cd)) !=
1746 	    cgroup_id(sock_cgroup_ptr(ssk_cd))) {
1747 		cgroup_sk_free(ssk_cd);
1748 		*ssk_cd = *sk_cd;
1749 		cgroup_sk_clone(sk_cd);
1750 	}
1751 #endif /* CONFIG_SOCK_CGROUP_DATA */
1752 }
1753 
1754 static void mptcp_attach_cgroup(struct sock *parent, struct sock *child)
1755 {
1756 	__mptcp_inherit_cgrp_data(parent, child);
1757 	if (mem_cgroup_sockets_enabled)
1758 		mem_cgroup_sk_inherit(parent, child);
1759 }
1760 
1761 static void mptcp_subflow_ops_override(struct sock *ssk)
1762 {
1763 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1764 	if (ssk->sk_prot == &tcpv6_prot)
1765 		ssk->sk_prot = &tcpv6_prot_override;
1766 	else
1767 #endif
1768 		ssk->sk_prot = &tcp_prot_override;
1769 }
1770 
1771 static void mptcp_subflow_ops_undo_override(struct sock *ssk)
1772 {
1773 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1774 	if (ssk->sk_prot == &tcpv6_prot_override)
1775 		ssk->sk_prot = &tcpv6_prot;
1776 	else
1777 #endif
1778 		ssk->sk_prot = &tcp_prot;
1779 }
1780 
1781 int mptcp_subflow_create_socket(struct sock *sk, unsigned short family,
1782 				struct socket **new_sock)
1783 {
1784 	struct mptcp_subflow_context *subflow;
1785 	struct net *net = sock_net(sk);
1786 	struct socket *sf;
1787 	int err;
1788 
1789 	/* un-accepted server sockets can reach here - on bad configuration
1790 	 * bail early to avoid greater trouble later
1791 	 */
1792 	if (unlikely(!sk->sk_socket))
1793 		return -EINVAL;
1794 
1795 	err = sock_create_kern(net, family, SOCK_STREAM, IPPROTO_TCP, &sf);
1796 	if (err)
1797 		return err;
1798 
1799 	lock_sock_nested(sf->sk, SINGLE_DEPTH_NESTING);
1800 
1801 	err = security_mptcp_add_subflow(sk, sf->sk);
1802 	if (err)
1803 		goto err_free;
1804 
1805 	/* the newly created socket has to be in the same cgroup as its parent */
1806 	mptcp_attach_cgroup(sk, sf->sk);
1807 
1808 	/* kernel sockets do not by default acquire net ref, but TCP timer
1809 	 * needs it.
1810 	 * Update ns_tracker to current stack trace and refcounted tracker.
1811 	 */
1812 	sk_net_refcnt_upgrade(sf->sk);
1813 	err = tcp_set_ulp(sf->sk, "mptcp");
1814 	if (err)
1815 		goto err_free;
1816 
1817 	mptcp_sockopt_sync_locked(mptcp_sk(sk), sf->sk);
1818 	release_sock(sf->sk);
1819 
1820 	/* the newly created socket really belongs to the owning MPTCP
1821 	 * socket, even if for additional subflows the allocation is performed
1822 	 * by a kernel workqueue. Adjust inode references, so that the
1823 	 * procfs/diag interfaces really show this one belonging to the correct
1824 	 * user.
1825 	 */
1826 	SOCK_INODE(sf)->i_ino = SOCK_INODE(sk->sk_socket)->i_ino;
1827 	SOCK_INODE(sf)->i_uid = SOCK_INODE(sk->sk_socket)->i_uid;
1828 	SOCK_INODE(sf)->i_gid = SOCK_INODE(sk->sk_socket)->i_gid;
1829 
1830 	subflow = mptcp_subflow_ctx(sf->sk);
1831 	pr_debug("subflow=%p\n", subflow);
1832 
1833 	*new_sock = sf;
1834 	sock_hold(sk);
1835 	subflow->conn = sk;
1836 	mptcp_subflow_ops_override(sf->sk);
1837 
1838 	return 0;
1839 
1840 err_free:
1841 	release_sock(sf->sk);
1842 	sock_release(sf);
1843 	return err;
1844 }
1845 
1846 static struct mptcp_subflow_context *subflow_create_ctx(struct sock *sk,
1847 							gfp_t priority)
1848 {
1849 	struct inet_connection_sock *icsk = inet_csk(sk);
1850 	struct mptcp_subflow_context *ctx;
1851 
1852 	ctx = kzalloc_obj(*ctx, priority);
1853 	if (!ctx)
1854 		return NULL;
1855 
1856 	rcu_assign_pointer(icsk->icsk_ulp_data, ctx);
1857 	INIT_LIST_HEAD(&ctx->node);
1858 	INIT_LIST_HEAD(&ctx->delegated_node);
1859 
1860 	pr_debug("subflow=%p\n", ctx);
1861 
1862 	ctx->tcp_sock = sk;
1863 	WRITE_ONCE(ctx->local_id, -1);
1864 
1865 	return ctx;
1866 }
1867 
1868 static void __subflow_state_change(struct sock *sk)
1869 {
1870 	struct socket_wq *wq;
1871 
1872 	rcu_read_lock();
1873 	wq = rcu_dereference(sk->sk_wq);
1874 	if (skwq_has_sleeper(wq))
1875 		wake_up_interruptible_all(&wq->wait);
1876 	rcu_read_unlock();
1877 }
1878 
1879 static void subflow_state_change(struct sock *sk)
1880 {
1881 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1882 	struct sock *parent = subflow->conn;
1883 
1884 	__subflow_state_change(sk);
1885 
1886 	/* as recvmsg() does not acquire the subflow socket for ssk selection
1887 	 * a fin packet carrying a DSS can be unnoticed if we don't trigger
1888 	 * the data available machinery here.
1889 	 */
1890 	if (mptcp_subflow_data_available(sk))
1891 		mptcp_data_ready(parent, sk);
1892 	else if (unlikely(sk->sk_err))
1893 		subflow_error_report(sk);
1894 
1895 	subflow_sched_work_if_closed(mptcp_sk(parent), sk);
1896 }
1897 
1898 void mptcp_subflow_queue_clean(struct sock *listener_sk, struct sock *listener_ssk)
1899 {
1900 	struct request_sock_queue *queue = &inet_csk(listener_ssk)->icsk_accept_queue;
1901 	struct request_sock *req, *head, *tail;
1902 	struct mptcp_subflow_context *subflow;
1903 	struct sock *sk, *ssk;
1904 
1905 	/* Due to lock dependencies no relevant lock can be acquired under rskq_lock.
1906 	 * Splice the req list, so that accept() can not reach the pending ssk after
1907 	 * the listener socket is released below.
1908 	 */
1909 	spin_lock_bh(&queue->rskq_lock);
1910 	head = queue->rskq_accept_head;
1911 	tail = queue->rskq_accept_tail;
1912 	queue->rskq_accept_head = NULL;
1913 	queue->rskq_accept_tail = NULL;
1914 	spin_unlock_bh(&queue->rskq_lock);
1915 
1916 	if (!head)
1917 		return;
1918 
1919 	/* can't acquire the msk socket lock under the subflow one,
1920 	 * or will cause ABBA deadlock
1921 	 */
1922 	release_sock(listener_ssk);
1923 
1924 	for (req = head; req; req = req->dl_next) {
1925 		ssk = req->sk;
1926 		if (!sk_is_mptcp(ssk))
1927 			continue;
1928 
1929 		subflow = mptcp_subflow_ctx(ssk);
1930 		if (!subflow || !subflow->conn)
1931 			continue;
1932 
1933 		sk = subflow->conn;
1934 		sock_hold(sk);
1935 
1936 		lock_sock_nested(sk, SINGLE_DEPTH_NESTING);
1937 		__mptcp_unaccepted_force_close(sk);
1938 		release_sock(sk);
1939 
1940 		/* lockdep will report a false positive ABBA deadlock
1941 		 * between cancel_work_sync and the listener socket.
1942 		 * The involved locks belong to different sockets WRT
1943 		 * the existing AB chain.
1944 		 * Using a per socket key is problematic as key
1945 		 * deregistration requires process context and must be
1946 		 * performed at socket disposal time, in atomic
1947 		 * context.
1948 		 * Just tell lockdep to consider the listener socket
1949 		 * released here.
1950 		 */
1951 		mutex_release(&listener_sk->sk_lock.dep_map, _RET_IP_);
1952 		mptcp_cancel_work(sk);
1953 		mutex_acquire(&listener_sk->sk_lock.dep_map, 0, 0, _RET_IP_);
1954 
1955 		sock_put(sk);
1956 	}
1957 
1958 	/* we are still under the listener msk socket lock */
1959 	lock_sock_nested(listener_ssk, SINGLE_DEPTH_NESTING);
1960 
1961 	/* restore the listener queue, to let the TCP code clean it up */
1962 	spin_lock_bh(&queue->rskq_lock);
1963 	WARN_ON_ONCE(queue->rskq_accept_head);
1964 	queue->rskq_accept_head = head;
1965 	queue->rskq_accept_tail = tail;
1966 	spin_unlock_bh(&queue->rskq_lock);
1967 }
1968 
1969 static int subflow_ulp_init(struct sock *sk)
1970 {
1971 	struct inet_connection_sock *icsk = inet_csk(sk);
1972 	struct mptcp_subflow_context *ctx;
1973 	struct tcp_sock *tp = tcp_sk(sk);
1974 	int err = 0;
1975 
1976 	/* disallow attaching ULP to a socket unless it has been
1977 	 * created with sock_create_kern()
1978 	 */
1979 	if (!sk->sk_kern_sock) {
1980 		err = -EOPNOTSUPP;
1981 		goto out;
1982 	}
1983 
1984 	ctx = subflow_create_ctx(sk, GFP_KERNEL);
1985 	if (!ctx) {
1986 		err = -ENOMEM;
1987 		goto out;
1988 	}
1989 
1990 	pr_debug("subflow=%p, family=%d\n", ctx, sk->sk_family);
1991 
1992 	tp->is_mptcp = 1;
1993 	ctx->icsk_af_ops = icsk->icsk_af_ops;
1994 	icsk->icsk_af_ops = subflow_default_af_ops(sk);
1995 	ctx->tcp_state_change = sk->sk_state_change;
1996 	ctx->tcp_error_report = sk->sk_error_report;
1997 
1998 	WARN_ON_ONCE(sk->sk_data_ready != sock_def_readable);
1999 	WARN_ON_ONCE(sk->sk_write_space != sk_stream_write_space);
2000 
2001 	sk->sk_data_ready = subflow_data_ready;
2002 	sk->sk_write_space = subflow_write_space;
2003 	sk->sk_state_change = subflow_state_change;
2004 	sk->sk_error_report = subflow_error_report;
2005 out:
2006 	return err;
2007 }
2008 
2009 static void subflow_ulp_release(struct sock *ssk)
2010 {
2011 	struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(ssk);
2012 	bool release = true;
2013 	struct sock *sk;
2014 
2015 	if (!ctx)
2016 		return;
2017 
2018 	sk = ctx->conn;
2019 	if (sk) {
2020 		/* if the msk has been orphaned, keep the ctx
2021 		 * alive, will be freed by __mptcp_close_ssk(),
2022 		 * when the subflow is still unaccepted
2023 		 */
2024 		release = ctx->disposable || list_empty(&ctx->node);
2025 
2026 		/* inet_child_forget() does not call sk_state_change(),
2027 		 * explicitly trigger the socket close machinery
2028 		 */
2029 		if (!release && !test_and_set_bit(MPTCP_WORK_CLOSE_SUBFLOW,
2030 						  &mptcp_sk(sk)->flags))
2031 			mptcp_schedule_work(sk);
2032 		sock_put(sk);
2033 	}
2034 
2035 	mptcp_subflow_ops_undo_override(ssk);
2036 	if (release)
2037 		kfree_rcu(ctx, rcu);
2038 }
2039 
2040 static void subflow_ulp_clone(const struct request_sock *req,
2041 			      struct sock *newsk,
2042 			      const gfp_t priority)
2043 {
2044 	struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
2045 	struct mptcp_subflow_context *old_ctx = mptcp_subflow_ctx(newsk);
2046 	struct mptcp_subflow_context *new_ctx;
2047 
2048 	if (!tcp_rsk(req)->is_mptcp ||
2049 	    (!subflow_req->mp_capable && !subflow_req->mp_join)) {
2050 		subflow_ulp_fallback(newsk, old_ctx);
2051 		return;
2052 	}
2053 
2054 	new_ctx = subflow_create_ctx(newsk, priority);
2055 	if (!new_ctx) {
2056 		subflow_ulp_fallback(newsk, old_ctx);
2057 		return;
2058 	}
2059 
2060 	new_ctx->conn_finished = 1;
2061 	new_ctx->icsk_af_ops = old_ctx->icsk_af_ops;
2062 	new_ctx->tcp_state_change = old_ctx->tcp_state_change;
2063 	new_ctx->tcp_error_report = old_ctx->tcp_error_report;
2064 	new_ctx->rel_write_seq = 1;
2065 
2066 	if (subflow_req->mp_capable) {
2067 		/* see comments in subflow_syn_recv_sock(), MPTCP connection
2068 		 * is fully established only after we receive the remote key
2069 		 */
2070 		new_ctx->mp_capable = 1;
2071 		new_ctx->local_key = subflow_req->local_key;
2072 		new_ctx->token = subflow_req->token;
2073 		new_ctx->ssn_offset = subflow_req->ssn_offset;
2074 		new_ctx->idsn = subflow_req->idsn;
2075 
2076 		/* this is the first subflow, id is always 0 */
2077 		subflow_set_local_id(new_ctx, 0);
2078 	} else if (subflow_req->mp_join) {
2079 		new_ctx->ssn_offset = subflow_req->ssn_offset;
2080 		new_ctx->mp_join = 1;
2081 		WRITE_ONCE(new_ctx->fully_established, true);
2082 		new_ctx->remote_key_valid = 1;
2083 		new_ctx->backup = subflow_req->backup;
2084 		new_ctx->request_bkup = subflow_req->request_bkup;
2085 		WRITE_ONCE(new_ctx->remote_id, subflow_req->remote_id);
2086 		new_ctx->token = subflow_req->token;
2087 		new_ctx->thmac = subflow_req->thmac;
2088 
2089 		/* the subflow req id is valid, fetched via subflow_check_req()
2090 		 * and subflow_token_join_request()
2091 		 */
2092 		subflow_set_local_id(new_ctx, subflow_req->local_id);
2093 	}
2094 }
2095 
2096 static void tcp_release_cb_override(struct sock *ssk)
2097 {
2098 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
2099 	long status;
2100 
2101 	/* process and clear all the pending actions, but leave the subflow into
2102 	 * the napi queue. To respect locking, only the same CPU that originated
2103 	 * the action can touch the list. mptcp_napi_poll will take care of it.
2104 	 */
2105 	status = set_mask_bits(&subflow->delegated_status, MPTCP_DELEGATE_ACTIONS_MASK, 0);
2106 	if (status)
2107 		mptcp_subflow_process_delegated(ssk, status);
2108 
2109 	tcp_release_cb(ssk);
2110 }
2111 
2112 static int tcp_abort_override(struct sock *ssk, int err)
2113 {
2114 	/* closing a listener subflow requires a great deal of care.
2115 	 * keep it simple and just prevent such operation
2116 	 */
2117 	if (inet_sk_state_load(ssk) == TCP_LISTEN)
2118 		return -EINVAL;
2119 
2120 	return tcp_abort(ssk, err);
2121 }
2122 
2123 static struct tcp_ulp_ops subflow_ulp_ops __read_mostly = {
2124 	.name		= "mptcp",
2125 	.owner		= THIS_MODULE,
2126 	.init		= subflow_ulp_init,
2127 	.release	= subflow_ulp_release,
2128 	.clone		= subflow_ulp_clone,
2129 };
2130 
2131 static int subflow_ops_init(struct request_sock_ops *subflow_ops)
2132 {
2133 	subflow_ops->obj_size = sizeof(struct mptcp_subflow_request_sock);
2134 
2135 	subflow_ops->slab = kmem_cache_create(subflow_ops->slab_name,
2136 					      subflow_ops->obj_size, 0,
2137 					      SLAB_ACCOUNT |
2138 					      SLAB_TYPESAFE_BY_RCU,
2139 					      NULL);
2140 	if (!subflow_ops->slab)
2141 		return -ENOMEM;
2142 
2143 	return 0;
2144 }
2145 
2146 void __init mptcp_subflow_init(void)
2147 {
2148 	mptcp_subflow_v4_request_sock_ops = tcp_request_sock_ops;
2149 	mptcp_subflow_v4_request_sock_ops.slab_name = "request_sock_subflow_v4";
2150 	mptcp_subflow_v4_request_sock_ops.destructor = subflow_v4_req_destructor;
2151 
2152 	if (subflow_ops_init(&mptcp_subflow_v4_request_sock_ops) != 0)
2153 		panic("MPTCP: failed to init subflow v4 request sock ops\n");
2154 
2155 	subflow_request_sock_ipv4_ops = tcp_request_sock_ipv4_ops;
2156 	subflow_request_sock_ipv4_ops.route_req = subflow_v4_route_req;
2157 	subflow_request_sock_ipv4_ops.send_synack = subflow_v4_send_synack;
2158 
2159 	subflow_specific = ipv4_specific;
2160 	subflow_specific.conn_request = subflow_v4_conn_request;
2161 	subflow_specific.syn_recv_sock = subflow_syn_recv_sock;
2162 	subflow_specific.sk_rx_dst_set = subflow_finish_connect;
2163 	subflow_specific.rebuild_header = subflow_rebuild_header;
2164 
2165 	tcp_prot_override = tcp_prot;
2166 	tcp_prot_override.release_cb = tcp_release_cb_override;
2167 	tcp_prot_override.diag_destroy = tcp_abort_override;
2168 #ifdef CONFIG_BPF_SYSCALL
2169 	/* Disable sockmap processing for subflows */
2170 	tcp_prot_override.psock_update_sk_prot = NULL;
2171 #endif
2172 
2173 	mptcp_diag_subflow_init(&subflow_ulp_ops);
2174 
2175 	if (tcp_register_ulp(&subflow_ulp_ops) != 0)
2176 		panic("MPTCP: failed to register subflows to ULP\n");
2177 }
2178 
2179 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
2180 void __init mptcp_subflow_v6_init(void)
2181 {
2182 	/* In struct mptcp_subflow_request_sock, we assume the TCP request sock
2183 	 * structures for v4 and v6 have the same size. It should not changed in
2184 	 * the future but better to make sure to be warned if it is no longer
2185 	 * the case.
2186 	 */
2187 	BUILD_BUG_ON(sizeof(struct tcp_request_sock) != sizeof(struct tcp6_request_sock));
2188 
2189 	mptcp_subflow_v6_request_sock_ops = tcp6_request_sock_ops;
2190 	mptcp_subflow_v6_request_sock_ops.slab_name = "request_sock_subflow_v6";
2191 	mptcp_subflow_v6_request_sock_ops.destructor = subflow_v6_req_destructor;
2192 
2193 	if (subflow_ops_init(&mptcp_subflow_v6_request_sock_ops) != 0)
2194 		panic("MPTCP: failed to init subflow v6 request sock ops\n");
2195 
2196 	subflow_request_sock_ipv6_ops = tcp_request_sock_ipv6_ops;
2197 	subflow_request_sock_ipv6_ops.route_req = subflow_v6_route_req;
2198 	subflow_request_sock_ipv6_ops.send_synack = subflow_v6_send_synack;
2199 
2200 	subflow_v6_specific = ipv6_specific;
2201 	subflow_v6_specific.conn_request = subflow_v6_conn_request;
2202 	subflow_v6_specific.syn_recv_sock = subflow_syn_recv_sock;
2203 	subflow_v6_specific.sk_rx_dst_set = subflow_finish_connect;
2204 	subflow_v6_specific.rebuild_header = subflow_v6_rebuild_header;
2205 
2206 	subflow_v6m_specific = subflow_v6_specific;
2207 	subflow_v6m_specific.queue_xmit = ipv4_specific.queue_xmit;
2208 	subflow_v6m_specific.net_header_len = ipv4_specific.net_header_len;
2209 	subflow_v6m_specific.mtu_reduced = ipv4_specific.mtu_reduced;
2210 	subflow_v6m_specific.rebuild_header = subflow_rebuild_header;
2211 
2212 	tcpv6_prot_override = tcpv6_prot;
2213 	tcpv6_prot_override.release_cb = tcp_release_cb_override;
2214 	tcpv6_prot_override.diag_destroy = tcp_abort_override;
2215 #ifdef CONFIG_BPF_SYSCALL
2216 	/* Disable sockmap processing for subflows */
2217 	tcpv6_prot_override.psock_update_sk_prot = NULL;
2218 #endif
2219 }
2220 #endif
2221