xref: /linux/net/mptcp/subflow.c (revision 93a3545d812ae7cfe4426374e00a7d8f64ac02e0)
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/algapi.h>
13 #include <crypto/sha.h>
14 #include <net/sock.h>
15 #include <net/inet_common.h>
16 #include <net/inet_hashtables.h>
17 #include <net/protocol.h>
18 #include <net/tcp.h>
19 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
20 #include <net/ip6_route.h>
21 #endif
22 #include <net/mptcp.h>
23 #include "protocol.h"
24 #include "mib.h"
25 
26 static void SUBFLOW_REQ_INC_STATS(struct request_sock *req,
27 				  enum linux_mptcp_mib_field field)
28 {
29 	MPTCP_INC_STATS(sock_net(req_to_sk(req)), field);
30 }
31 
32 static void subflow_req_destructor(struct request_sock *req)
33 {
34 	struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
35 
36 	pr_debug("subflow_req=%p", subflow_req);
37 
38 	if (subflow_req->msk)
39 		sock_put((struct sock *)subflow_req->msk);
40 
41 	mptcp_token_destroy_request(req);
42 	tcp_request_sock_ops.destructor(req);
43 }
44 
45 static void subflow_generate_hmac(u64 key1, u64 key2, u32 nonce1, u32 nonce2,
46 				  void *hmac)
47 {
48 	u8 msg[8];
49 
50 	put_unaligned_be32(nonce1, &msg[0]);
51 	put_unaligned_be32(nonce2, &msg[4]);
52 
53 	mptcp_crypto_hmac_sha(key1, key2, msg, 8, hmac);
54 }
55 
56 /* validate received token and create truncated hmac and nonce for SYN-ACK */
57 static struct mptcp_sock *subflow_token_join_request(struct request_sock *req,
58 						     const struct sk_buff *skb)
59 {
60 	struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
61 	u8 hmac[SHA256_DIGEST_SIZE];
62 	struct mptcp_sock *msk;
63 	int local_id;
64 
65 	msk = mptcp_token_get_sock(subflow_req->token);
66 	if (!msk) {
67 		SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINNOTOKEN);
68 		return NULL;
69 	}
70 
71 	local_id = mptcp_pm_get_local_id(msk, (struct sock_common *)req);
72 	if (local_id < 0) {
73 		sock_put((struct sock *)msk);
74 		return NULL;
75 	}
76 	subflow_req->local_id = local_id;
77 
78 	get_random_bytes(&subflow_req->local_nonce, sizeof(u32));
79 
80 	subflow_generate_hmac(msk->local_key, msk->remote_key,
81 			      subflow_req->local_nonce,
82 			      subflow_req->remote_nonce, hmac);
83 
84 	subflow_req->thmac = get_unaligned_be64(hmac);
85 	return msk;
86 }
87 
88 static void subflow_init_req(struct request_sock *req,
89 			     const struct sock *sk_listener,
90 			     struct sk_buff *skb)
91 {
92 	struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk_listener);
93 	struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
94 	struct mptcp_options_received mp_opt;
95 
96 	pr_debug("subflow_req=%p, listener=%p", subflow_req, listener);
97 
98 	mptcp_get_options(skb, &mp_opt);
99 
100 	subflow_req->mp_capable = 0;
101 	subflow_req->mp_join = 0;
102 	subflow_req->msk = NULL;
103 	mptcp_token_init_request(req);
104 
105 #ifdef CONFIG_TCP_MD5SIG
106 	/* no MPTCP if MD5SIG is enabled on this socket or we may run out of
107 	 * TCP option space.
108 	 */
109 	if (rcu_access_pointer(tcp_sk(sk_listener)->md5sig_info))
110 		return;
111 #endif
112 
113 	if (mp_opt.mp_capable) {
114 		SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MPCAPABLEPASSIVE);
115 
116 		if (mp_opt.mp_join)
117 			return;
118 	} else if (mp_opt.mp_join) {
119 		SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINSYNRX);
120 	}
121 
122 	if (mp_opt.mp_capable && listener->request_mptcp) {
123 		int err;
124 
125 		err = mptcp_token_new_request(req);
126 		if (err == 0)
127 			subflow_req->mp_capable = 1;
128 
129 		subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq;
130 	} else if (mp_opt.mp_join && listener->request_mptcp) {
131 		subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq;
132 		subflow_req->mp_join = 1;
133 		subflow_req->backup = mp_opt.backup;
134 		subflow_req->remote_id = mp_opt.join_id;
135 		subflow_req->token = mp_opt.token;
136 		subflow_req->remote_nonce = mp_opt.nonce;
137 		subflow_req->msk = subflow_token_join_request(req, skb);
138 		pr_debug("token=%u, remote_nonce=%u msk=%p", subflow_req->token,
139 			 subflow_req->remote_nonce, subflow_req->msk);
140 	}
141 }
142 
143 static void subflow_v4_init_req(struct request_sock *req,
144 				const struct sock *sk_listener,
145 				struct sk_buff *skb)
146 {
147 	tcp_rsk(req)->is_mptcp = 1;
148 
149 	tcp_request_sock_ipv4_ops.init_req(req, sk_listener, skb);
150 
151 	subflow_init_req(req, sk_listener, skb);
152 }
153 
154 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
155 static void subflow_v6_init_req(struct request_sock *req,
156 				const struct sock *sk_listener,
157 				struct sk_buff *skb)
158 {
159 	tcp_rsk(req)->is_mptcp = 1;
160 
161 	tcp_request_sock_ipv6_ops.init_req(req, sk_listener, skb);
162 
163 	subflow_init_req(req, sk_listener, skb);
164 }
165 #endif
166 
167 /* validate received truncated hmac and create hmac for third ACK */
168 static bool subflow_thmac_valid(struct mptcp_subflow_context *subflow)
169 {
170 	u8 hmac[SHA256_DIGEST_SIZE];
171 	u64 thmac;
172 
173 	subflow_generate_hmac(subflow->remote_key, subflow->local_key,
174 			      subflow->remote_nonce, subflow->local_nonce,
175 			      hmac);
176 
177 	thmac = get_unaligned_be64(hmac);
178 	pr_debug("subflow=%p, token=%u, thmac=%llu, subflow->thmac=%llu\n",
179 		 subflow, subflow->token,
180 		 (unsigned long long)thmac,
181 		 (unsigned long long)subflow->thmac);
182 
183 	return thmac == subflow->thmac;
184 }
185 
186 static void subflow_finish_connect(struct sock *sk, const struct sk_buff *skb)
187 {
188 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
189 	struct mptcp_options_received mp_opt;
190 	struct sock *parent = subflow->conn;
191 
192 	subflow->icsk_af_ops->sk_rx_dst_set(sk, skb);
193 
194 	if (inet_sk_state_load(parent) == TCP_SYN_SENT) {
195 		inet_sk_state_store(parent, TCP_ESTABLISHED);
196 		parent->sk_state_change(parent);
197 	}
198 
199 	/* be sure no special action on any packet other than syn-ack */
200 	if (subflow->conn_finished)
201 		return;
202 
203 	subflow->conn_finished = 1;
204 	subflow->ssn_offset = TCP_SKB_CB(skb)->seq;
205 	pr_debug("subflow=%p synack seq=%x", subflow, subflow->ssn_offset);
206 
207 	mptcp_get_options(skb, &mp_opt);
208 	if (subflow->request_mptcp && mp_opt.mp_capable) {
209 		subflow->mp_capable = 1;
210 		subflow->can_ack = 1;
211 		subflow->remote_key = mp_opt.sndr_key;
212 		pr_debug("subflow=%p, remote_key=%llu", subflow,
213 			 subflow->remote_key);
214 	} else if (subflow->request_join && mp_opt.mp_join) {
215 		subflow->mp_join = 1;
216 		subflow->thmac = mp_opt.thmac;
217 		subflow->remote_nonce = mp_opt.nonce;
218 		pr_debug("subflow=%p, thmac=%llu, remote_nonce=%u", subflow,
219 			 subflow->thmac, subflow->remote_nonce);
220 	} else {
221 		if (subflow->request_mptcp)
222 			MPTCP_INC_STATS(sock_net(sk),
223 					MPTCP_MIB_MPCAPABLEACTIVEFALLBACK);
224 		mptcp_do_fallback(sk);
225 		pr_fallback(mptcp_sk(subflow->conn));
226 	}
227 
228 	if (mptcp_check_fallback(sk)) {
229 		mptcp_rcv_space_init(mptcp_sk(parent), sk);
230 		return;
231 	}
232 
233 	if (subflow->mp_capable) {
234 		pr_debug("subflow=%p, remote_key=%llu", mptcp_subflow_ctx(sk),
235 			 subflow->remote_key);
236 		mptcp_finish_connect(sk);
237 	} else if (subflow->mp_join) {
238 		u8 hmac[SHA256_DIGEST_SIZE];
239 
240 		pr_debug("subflow=%p, thmac=%llu, remote_nonce=%u",
241 			 subflow, subflow->thmac,
242 			 subflow->remote_nonce);
243 		if (!subflow_thmac_valid(subflow)) {
244 			MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINACKMAC);
245 			subflow->mp_join = 0;
246 			goto do_reset;
247 		}
248 
249 		subflow_generate_hmac(subflow->local_key, subflow->remote_key,
250 				      subflow->local_nonce,
251 				      subflow->remote_nonce,
252 				      hmac);
253 
254 		memcpy(subflow->hmac, hmac, MPTCPOPT_HMAC_LEN);
255 
256 		if (!mptcp_finish_join(sk))
257 			goto do_reset;
258 
259 		MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINSYNACKRX);
260 	} else {
261 do_reset:
262 		tcp_send_active_reset(sk, GFP_ATOMIC);
263 		tcp_done(sk);
264 	}
265 }
266 
267 static struct request_sock_ops subflow_request_sock_ops;
268 static struct tcp_request_sock_ops subflow_request_sock_ipv4_ops;
269 
270 static int subflow_v4_conn_request(struct sock *sk, struct sk_buff *skb)
271 {
272 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
273 
274 	pr_debug("subflow=%p", subflow);
275 
276 	/* Never answer to SYNs sent to broadcast or multicast */
277 	if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
278 		goto drop;
279 
280 	return tcp_conn_request(&subflow_request_sock_ops,
281 				&subflow_request_sock_ipv4_ops,
282 				sk, skb);
283 drop:
284 	tcp_listendrop(sk);
285 	return 0;
286 }
287 
288 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
289 static struct tcp_request_sock_ops subflow_request_sock_ipv6_ops;
290 static struct inet_connection_sock_af_ops subflow_v6_specific;
291 static struct inet_connection_sock_af_ops subflow_v6m_specific;
292 
293 static int subflow_v6_conn_request(struct sock *sk, struct sk_buff *skb)
294 {
295 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
296 
297 	pr_debug("subflow=%p", subflow);
298 
299 	if (skb->protocol == htons(ETH_P_IP))
300 		return subflow_v4_conn_request(sk, skb);
301 
302 	if (!ipv6_unicast_destination(skb))
303 		goto drop;
304 
305 	return tcp_conn_request(&subflow_request_sock_ops,
306 				&subflow_request_sock_ipv6_ops, sk, skb);
307 
308 drop:
309 	tcp_listendrop(sk);
310 	return 0; /* don't send reset */
311 }
312 #endif
313 
314 /* validate hmac received in third ACK */
315 static bool subflow_hmac_valid(const struct request_sock *req,
316 			       const struct mptcp_options_received *mp_opt)
317 {
318 	const struct mptcp_subflow_request_sock *subflow_req;
319 	u8 hmac[SHA256_DIGEST_SIZE];
320 	struct mptcp_sock *msk;
321 
322 	subflow_req = mptcp_subflow_rsk(req);
323 	msk = subflow_req->msk;
324 	if (!msk)
325 		return false;
326 
327 	subflow_generate_hmac(msk->remote_key, msk->local_key,
328 			      subflow_req->remote_nonce,
329 			      subflow_req->local_nonce, hmac);
330 
331 	return !crypto_memneq(hmac, mp_opt->hmac, MPTCPOPT_HMAC_LEN);
332 }
333 
334 static void mptcp_sock_destruct(struct sock *sk)
335 {
336 	/* if new mptcp socket isn't accepted, it is free'd
337 	 * from the tcp listener sockets request queue, linked
338 	 * from req->sk.  The tcp socket is released.
339 	 * This calls the ULP release function which will
340 	 * also remove the mptcp socket, via
341 	 * sock_put(ctx->conn).
342 	 *
343 	 * Problem is that the mptcp socket will not be in
344 	 * SYN_RECV state and doesn't have SOCK_DEAD flag.
345 	 * Both result in warnings from inet_sock_destruct.
346 	 */
347 
348 	if (sk->sk_state == TCP_SYN_RECV) {
349 		sk->sk_state = TCP_CLOSE;
350 		WARN_ON_ONCE(sk->sk_socket);
351 		sock_orphan(sk);
352 	}
353 
354 	mptcp_token_destroy(mptcp_sk(sk));
355 	inet_sock_destruct(sk);
356 }
357 
358 static void mptcp_force_close(struct sock *sk)
359 {
360 	inet_sk_state_store(sk, TCP_CLOSE);
361 	sk_common_release(sk);
362 }
363 
364 static void subflow_ulp_fallback(struct sock *sk,
365 				 struct mptcp_subflow_context *old_ctx)
366 {
367 	struct inet_connection_sock *icsk = inet_csk(sk);
368 
369 	mptcp_subflow_tcp_fallback(sk, old_ctx);
370 	icsk->icsk_ulp_ops = NULL;
371 	rcu_assign_pointer(icsk->icsk_ulp_data, NULL);
372 	tcp_sk(sk)->is_mptcp = 0;
373 }
374 
375 static void subflow_drop_ctx(struct sock *ssk)
376 {
377 	struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(ssk);
378 
379 	if (!ctx)
380 		return;
381 
382 	subflow_ulp_fallback(ssk, ctx);
383 	if (ctx->conn)
384 		sock_put(ctx->conn);
385 
386 	kfree_rcu(ctx, rcu);
387 }
388 
389 static struct sock *subflow_syn_recv_sock(const struct sock *sk,
390 					  struct sk_buff *skb,
391 					  struct request_sock *req,
392 					  struct dst_entry *dst,
393 					  struct request_sock *req_unhash,
394 					  bool *own_req)
395 {
396 	struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk);
397 	struct mptcp_subflow_request_sock *subflow_req;
398 	struct mptcp_options_received mp_opt;
399 	bool fallback, fallback_is_fatal;
400 	struct sock *new_msk = NULL;
401 	struct sock *child;
402 
403 	pr_debug("listener=%p, req=%p, conn=%p", listener, req, listener->conn);
404 
405 	/* After child creation we must look for 'mp_capable' even when options
406 	 * are not parsed
407 	 */
408 	mp_opt.mp_capable = 0;
409 
410 	/* hopefully temporary handling for MP_JOIN+syncookie */
411 	subflow_req = mptcp_subflow_rsk(req);
412 	fallback_is_fatal = subflow_req->mp_join;
413 	fallback = !tcp_rsk(req)->is_mptcp;
414 	if (fallback)
415 		goto create_child;
416 
417 	/* if the sk is MP_CAPABLE, we try to fetch the client key */
418 	if (subflow_req->mp_capable) {
419 		if (TCP_SKB_CB(skb)->seq != subflow_req->ssn_offset + 1) {
420 			/* here we can receive and accept an in-window,
421 			 * out-of-order pkt, which will not carry the MP_CAPABLE
422 			 * opt even on mptcp enabled paths
423 			 */
424 			goto create_msk;
425 		}
426 
427 		mptcp_get_options(skb, &mp_opt);
428 		if (!mp_opt.mp_capable) {
429 			fallback = true;
430 			goto create_child;
431 		}
432 
433 create_msk:
434 		new_msk = mptcp_sk_clone(listener->conn, &mp_opt, req);
435 		if (!new_msk)
436 			fallback = true;
437 	} else if (subflow_req->mp_join) {
438 		mptcp_get_options(skb, &mp_opt);
439 		if (!mp_opt.mp_join ||
440 		    !subflow_hmac_valid(req, &mp_opt)) {
441 			SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKMAC);
442 			fallback = true;
443 		}
444 	}
445 
446 create_child:
447 	child = listener->icsk_af_ops->syn_recv_sock(sk, skb, req, dst,
448 						     req_unhash, own_req);
449 
450 	if (child && *own_req) {
451 		struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(child);
452 
453 		tcp_rsk(req)->drop_req = false;
454 
455 		/* we need to fallback on ctx allocation failure and on pre-reqs
456 		 * checking above. In the latter scenario we additionally need
457 		 * to reset the context to non MPTCP status.
458 		 */
459 		if (!ctx || fallback) {
460 			if (fallback_is_fatal)
461 				goto dispose_child;
462 
463 			subflow_drop_ctx(child);
464 			goto out;
465 		}
466 
467 		if (ctx->mp_capable) {
468 			/* new mpc subflow takes ownership of the newly
469 			 * created mptcp socket
470 			 */
471 			new_msk->sk_destruct = mptcp_sock_destruct;
472 			mptcp_pm_new_connection(mptcp_sk(new_msk), 1);
473 			mptcp_token_accept(subflow_req, mptcp_sk(new_msk));
474 			ctx->conn = new_msk;
475 			new_msk = NULL;
476 
477 			/* with OoO packets we can reach here without ingress
478 			 * mpc option
479 			 */
480 			ctx->remote_key = mp_opt.sndr_key;
481 			ctx->fully_established = mp_opt.mp_capable;
482 			ctx->can_ack = mp_opt.mp_capable;
483 		} else if (ctx->mp_join) {
484 			struct mptcp_sock *owner;
485 
486 			owner = subflow_req->msk;
487 			if (!owner)
488 				goto dispose_child;
489 
490 			/* move the msk reference ownership to the subflow */
491 			subflow_req->msk = NULL;
492 			ctx->conn = (struct sock *)owner;
493 			if (!mptcp_finish_join(child))
494 				goto dispose_child;
495 
496 			SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKRX);
497 			tcp_rsk(req)->drop_req = true;
498 		}
499 	}
500 
501 out:
502 	/* dispose of the left over mptcp master, if any */
503 	if (unlikely(new_msk))
504 		mptcp_force_close(new_msk);
505 
506 	/* check for expected invariant - should never trigger, just help
507 	 * catching eariler subtle bugs
508 	 */
509 	WARN_ON_ONCE(child && *own_req && tcp_sk(child)->is_mptcp &&
510 		     (!mptcp_subflow_ctx(child) ||
511 		      !mptcp_subflow_ctx(child)->conn));
512 	return child;
513 
514 dispose_child:
515 	subflow_drop_ctx(child);
516 	tcp_rsk(req)->drop_req = true;
517 	tcp_send_active_reset(child, GFP_ATOMIC);
518 	inet_csk_prepare_for_destroy_sock(child);
519 	tcp_done(child);
520 
521 	/* The last child reference will be released by the caller */
522 	return child;
523 }
524 
525 static struct inet_connection_sock_af_ops subflow_specific;
526 
527 enum mapping_status {
528 	MAPPING_OK,
529 	MAPPING_INVALID,
530 	MAPPING_EMPTY,
531 	MAPPING_DATA_FIN,
532 	MAPPING_DUMMY
533 };
534 
535 static u64 expand_seq(u64 old_seq, u16 old_data_len, u64 seq)
536 {
537 	if ((u32)seq == (u32)old_seq)
538 		return old_seq;
539 
540 	/* Assume map covers data not mapped yet. */
541 	return seq | ((old_seq + old_data_len + 1) & GENMASK_ULL(63, 32));
542 }
543 
544 static void warn_bad_map(struct mptcp_subflow_context *subflow, u32 ssn)
545 {
546 	WARN_ONCE(1, "Bad mapping: ssn=%d map_seq=%d map_data_len=%d",
547 		  ssn, subflow->map_subflow_seq, subflow->map_data_len);
548 }
549 
550 static bool skb_is_fully_mapped(struct sock *ssk, struct sk_buff *skb)
551 {
552 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
553 	unsigned int skb_consumed;
554 
555 	skb_consumed = tcp_sk(ssk)->copied_seq - TCP_SKB_CB(skb)->seq;
556 	if (WARN_ON_ONCE(skb_consumed >= skb->len))
557 		return true;
558 
559 	return skb->len - skb_consumed <= subflow->map_data_len -
560 					  mptcp_subflow_get_map_offset(subflow);
561 }
562 
563 static bool validate_mapping(struct sock *ssk, struct sk_buff *skb)
564 {
565 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
566 	u32 ssn = tcp_sk(ssk)->copied_seq - subflow->ssn_offset;
567 
568 	if (unlikely(before(ssn, subflow->map_subflow_seq))) {
569 		/* Mapping covers data later in the subflow stream,
570 		 * currently unsupported.
571 		 */
572 		warn_bad_map(subflow, ssn);
573 		return false;
574 	}
575 	if (unlikely(!before(ssn, subflow->map_subflow_seq +
576 				  subflow->map_data_len))) {
577 		/* Mapping does covers past subflow data, invalid */
578 		warn_bad_map(subflow, ssn + skb->len);
579 		return false;
580 	}
581 	return true;
582 }
583 
584 static enum mapping_status get_mapping_status(struct sock *ssk)
585 {
586 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
587 	struct mptcp_ext *mpext;
588 	struct sk_buff *skb;
589 	u16 data_len;
590 	u64 map_seq;
591 
592 	skb = skb_peek(&ssk->sk_receive_queue);
593 	if (!skb)
594 		return MAPPING_EMPTY;
595 
596 	if (mptcp_check_fallback(ssk))
597 		return MAPPING_DUMMY;
598 
599 	mpext = mptcp_get_ext(skb);
600 	if (!mpext || !mpext->use_map) {
601 		if (!subflow->map_valid && !skb->len) {
602 			/* the TCP stack deliver 0 len FIN pkt to the receive
603 			 * queue, that is the only 0len pkts ever expected here,
604 			 * and we can admit no mapping only for 0 len pkts
605 			 */
606 			if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN))
607 				WARN_ONCE(1, "0len seq %d:%d flags %x",
608 					  TCP_SKB_CB(skb)->seq,
609 					  TCP_SKB_CB(skb)->end_seq,
610 					  TCP_SKB_CB(skb)->tcp_flags);
611 			sk_eat_skb(ssk, skb);
612 			return MAPPING_EMPTY;
613 		}
614 
615 		if (!subflow->map_valid)
616 			return MAPPING_INVALID;
617 
618 		goto validate_seq;
619 	}
620 
621 	pr_debug("seq=%llu is64=%d ssn=%u data_len=%u data_fin=%d",
622 		 mpext->data_seq, mpext->dsn64, mpext->subflow_seq,
623 		 mpext->data_len, mpext->data_fin);
624 
625 	data_len = mpext->data_len;
626 	if (data_len == 0) {
627 		pr_err("Infinite mapping not handled");
628 		MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_INFINITEMAPRX);
629 		return MAPPING_INVALID;
630 	}
631 
632 	if (mpext->data_fin == 1) {
633 		if (data_len == 1) {
634 			pr_debug("DATA_FIN with no payload");
635 			if (subflow->map_valid) {
636 				/* A DATA_FIN might arrive in a DSS
637 				 * option before the previous mapping
638 				 * has been fully consumed. Continue
639 				 * handling the existing mapping.
640 				 */
641 				skb_ext_del(skb, SKB_EXT_MPTCP);
642 				return MAPPING_OK;
643 			} else {
644 				return MAPPING_DATA_FIN;
645 			}
646 		}
647 
648 		/* Adjust for DATA_FIN using 1 byte of sequence space */
649 		data_len--;
650 	}
651 
652 	if (!mpext->dsn64) {
653 		map_seq = expand_seq(subflow->map_seq, subflow->map_data_len,
654 				     mpext->data_seq);
655 		subflow->use_64bit_ack = 0;
656 		pr_debug("expanded seq=%llu", subflow->map_seq);
657 	} else {
658 		map_seq = mpext->data_seq;
659 		subflow->use_64bit_ack = 1;
660 	}
661 
662 	if (subflow->map_valid) {
663 		/* Allow replacing only with an identical map */
664 		if (subflow->map_seq == map_seq &&
665 		    subflow->map_subflow_seq == mpext->subflow_seq &&
666 		    subflow->map_data_len == data_len) {
667 			skb_ext_del(skb, SKB_EXT_MPTCP);
668 			return MAPPING_OK;
669 		}
670 
671 		/* If this skb data are fully covered by the current mapping,
672 		 * the new map would need caching, which is not supported
673 		 */
674 		if (skb_is_fully_mapped(ssk, skb)) {
675 			MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DSSNOMATCH);
676 			return MAPPING_INVALID;
677 		}
678 
679 		/* will validate the next map after consuming the current one */
680 		return MAPPING_OK;
681 	}
682 
683 	subflow->map_seq = map_seq;
684 	subflow->map_subflow_seq = mpext->subflow_seq;
685 	subflow->map_data_len = data_len;
686 	subflow->map_valid = 1;
687 	subflow->mpc_map = mpext->mpc_map;
688 	pr_debug("new map seq=%llu subflow_seq=%u data_len=%u",
689 		 subflow->map_seq, subflow->map_subflow_seq,
690 		 subflow->map_data_len);
691 
692 validate_seq:
693 	/* we revalidate valid mapping on new skb, because we must ensure
694 	 * the current skb is completely covered by the available mapping
695 	 */
696 	if (!validate_mapping(ssk, skb))
697 		return MAPPING_INVALID;
698 
699 	skb_ext_del(skb, SKB_EXT_MPTCP);
700 	return MAPPING_OK;
701 }
702 
703 static int subflow_read_actor(read_descriptor_t *desc,
704 			      struct sk_buff *skb,
705 			      unsigned int offset, size_t len)
706 {
707 	size_t copy_len = min(desc->count, len);
708 
709 	desc->count -= copy_len;
710 
711 	pr_debug("flushed %zu bytes, %zu left", copy_len, desc->count);
712 	return copy_len;
713 }
714 
715 static bool subflow_check_data_avail(struct sock *ssk)
716 {
717 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
718 	enum mapping_status status;
719 	struct mptcp_sock *msk;
720 	struct sk_buff *skb;
721 
722 	pr_debug("msk=%p ssk=%p data_avail=%d skb=%p", subflow->conn, ssk,
723 		 subflow->data_avail, skb_peek(&ssk->sk_receive_queue));
724 	if (subflow->data_avail)
725 		return true;
726 
727 	msk = mptcp_sk(subflow->conn);
728 	for (;;) {
729 		u32 map_remaining;
730 		size_t delta;
731 		u64 ack_seq;
732 		u64 old_ack;
733 
734 		status = get_mapping_status(ssk);
735 		pr_debug("msk=%p ssk=%p status=%d", msk, ssk, status);
736 		if (status == MAPPING_INVALID) {
737 			ssk->sk_err = EBADMSG;
738 			goto fatal;
739 		}
740 		if (status == MAPPING_DUMMY) {
741 			__mptcp_do_fallback(msk);
742 			skb = skb_peek(&ssk->sk_receive_queue);
743 			subflow->map_valid = 1;
744 			subflow->map_seq = READ_ONCE(msk->ack_seq);
745 			subflow->map_data_len = skb->len;
746 			subflow->map_subflow_seq = tcp_sk(ssk)->copied_seq -
747 						   subflow->ssn_offset;
748 			return true;
749 		}
750 
751 		if (status != MAPPING_OK)
752 			return false;
753 
754 		skb = skb_peek(&ssk->sk_receive_queue);
755 		if (WARN_ON_ONCE(!skb))
756 			return false;
757 
758 		/* if msk lacks the remote key, this subflow must provide an
759 		 * MP_CAPABLE-based mapping
760 		 */
761 		if (unlikely(!READ_ONCE(msk->can_ack))) {
762 			if (!subflow->mpc_map) {
763 				ssk->sk_err = EBADMSG;
764 				goto fatal;
765 			}
766 			WRITE_ONCE(msk->remote_key, subflow->remote_key);
767 			WRITE_ONCE(msk->ack_seq, subflow->map_seq);
768 			WRITE_ONCE(msk->can_ack, true);
769 		}
770 
771 		old_ack = READ_ONCE(msk->ack_seq);
772 		ack_seq = mptcp_subflow_get_mapped_dsn(subflow);
773 		pr_debug("msk ack_seq=%llx subflow ack_seq=%llx", old_ack,
774 			 ack_seq);
775 		if (ack_seq == old_ack)
776 			break;
777 
778 		/* only accept in-sequence mapping. Old values are spurious
779 		 * retransmission; we can hit "future" values on active backup
780 		 * subflow switch, we relay on retransmissions to get
781 		 * in-sequence data.
782 		 * Cuncurrent subflows support will require subflow data
783 		 * reordering
784 		 */
785 		map_remaining = subflow->map_data_len -
786 				mptcp_subflow_get_map_offset(subflow);
787 		if (before64(ack_seq, old_ack))
788 			delta = min_t(size_t, old_ack - ack_seq, map_remaining);
789 		else
790 			delta = min_t(size_t, ack_seq - old_ack, map_remaining);
791 
792 		/* discard mapped data */
793 		pr_debug("discarding %zu bytes, current map len=%d", delta,
794 			 map_remaining);
795 		if (delta) {
796 			read_descriptor_t desc = {
797 				.count = delta,
798 			};
799 			int ret;
800 
801 			ret = tcp_read_sock(ssk, &desc, subflow_read_actor);
802 			if (ret < 0) {
803 				ssk->sk_err = -ret;
804 				goto fatal;
805 			}
806 			if (ret < delta)
807 				return false;
808 			if (delta == map_remaining)
809 				subflow->map_valid = 0;
810 		}
811 	}
812 	return true;
813 
814 fatal:
815 	/* fatal protocol error, close the socket */
816 	/* This barrier is coupled with smp_rmb() in tcp_poll() */
817 	smp_wmb();
818 	ssk->sk_error_report(ssk);
819 	tcp_set_state(ssk, TCP_CLOSE);
820 	tcp_send_active_reset(ssk, GFP_ATOMIC);
821 	return false;
822 }
823 
824 bool mptcp_subflow_data_available(struct sock *sk)
825 {
826 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
827 	struct sk_buff *skb;
828 
829 	/* check if current mapping is still valid */
830 	if (subflow->map_valid &&
831 	    mptcp_subflow_get_map_offset(subflow) >= subflow->map_data_len) {
832 		subflow->map_valid = 0;
833 		subflow->data_avail = 0;
834 
835 		pr_debug("Done with mapping: seq=%u data_len=%u",
836 			 subflow->map_subflow_seq,
837 			 subflow->map_data_len);
838 	}
839 
840 	if (!subflow_check_data_avail(sk)) {
841 		subflow->data_avail = 0;
842 		return false;
843 	}
844 
845 	skb = skb_peek(&sk->sk_receive_queue);
846 	subflow->data_avail = skb &&
847 		       before(tcp_sk(sk)->copied_seq, TCP_SKB_CB(skb)->end_seq);
848 	return subflow->data_avail;
849 }
850 
851 /* If ssk has an mptcp parent socket, use the mptcp rcvbuf occupancy,
852  * not the ssk one.
853  *
854  * In mptcp, rwin is about the mptcp-level connection data.
855  *
856  * Data that is still on the ssk rx queue can thus be ignored,
857  * as far as mptcp peer is concerened that data is still inflight.
858  * DSS ACK is updated when skb is moved to the mptcp rx queue.
859  */
860 void mptcp_space(const struct sock *ssk, int *space, int *full_space)
861 {
862 	const struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
863 	const struct sock *sk = subflow->conn;
864 
865 	*space = tcp_space(sk);
866 	*full_space = tcp_full_space(sk);
867 }
868 
869 static void subflow_data_ready(struct sock *sk)
870 {
871 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
872 	struct sock *parent = subflow->conn;
873 	struct mptcp_sock *msk;
874 
875 	msk = mptcp_sk(parent);
876 	if ((1 << inet_sk_state_load(sk)) & (TCPF_LISTEN | TCPF_CLOSE)) {
877 		set_bit(MPTCP_DATA_READY, &msk->flags);
878 		parent->sk_data_ready(parent);
879 		return;
880 	}
881 
882 	WARN_ON_ONCE(!__mptcp_check_fallback(msk) && !subflow->mp_capable &&
883 		     !subflow->mp_join);
884 
885 	if (mptcp_subflow_data_available(sk))
886 		mptcp_data_ready(parent, sk);
887 }
888 
889 static void subflow_write_space(struct sock *sk)
890 {
891 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
892 	struct sock *parent = subflow->conn;
893 
894 	sk_stream_write_space(sk);
895 	if (sk_stream_is_writeable(sk)) {
896 		set_bit(MPTCP_SEND_SPACE, &mptcp_sk(parent)->flags);
897 		smp_mb__after_atomic();
898 		/* set SEND_SPACE before sk_stream_write_space clears NOSPACE */
899 		sk_stream_write_space(parent);
900 	}
901 }
902 
903 static struct inet_connection_sock_af_ops *
904 subflow_default_af_ops(struct sock *sk)
905 {
906 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
907 	if (sk->sk_family == AF_INET6)
908 		return &subflow_v6_specific;
909 #endif
910 	return &subflow_specific;
911 }
912 
913 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
914 void mptcpv6_handle_mapped(struct sock *sk, bool mapped)
915 {
916 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
917 	struct inet_connection_sock *icsk = inet_csk(sk);
918 	struct inet_connection_sock_af_ops *target;
919 
920 	target = mapped ? &subflow_v6m_specific : subflow_default_af_ops(sk);
921 
922 	pr_debug("subflow=%p family=%d ops=%p target=%p mapped=%d",
923 		 subflow, sk->sk_family, icsk->icsk_af_ops, target, mapped);
924 
925 	if (likely(icsk->icsk_af_ops == target))
926 		return;
927 
928 	subflow->icsk_af_ops = icsk->icsk_af_ops;
929 	icsk->icsk_af_ops = target;
930 }
931 #endif
932 
933 static void mptcp_info2sockaddr(const struct mptcp_addr_info *info,
934 				struct sockaddr_storage *addr)
935 {
936 	memset(addr, 0, sizeof(*addr));
937 	addr->ss_family = info->family;
938 	if (addr->ss_family == AF_INET) {
939 		struct sockaddr_in *in_addr = (struct sockaddr_in *)addr;
940 
941 		in_addr->sin_addr = info->addr;
942 		in_addr->sin_port = info->port;
943 	}
944 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
945 	else if (addr->ss_family == AF_INET6) {
946 		struct sockaddr_in6 *in6_addr = (struct sockaddr_in6 *)addr;
947 
948 		in6_addr->sin6_addr = info->addr6;
949 		in6_addr->sin6_port = info->port;
950 	}
951 #endif
952 }
953 
954 int __mptcp_subflow_connect(struct sock *sk, int ifindex,
955 			    const struct mptcp_addr_info *loc,
956 			    const struct mptcp_addr_info *remote)
957 {
958 	struct mptcp_sock *msk = mptcp_sk(sk);
959 	struct mptcp_subflow_context *subflow;
960 	struct sockaddr_storage addr;
961 	int local_id = loc->id;
962 	struct socket *sf;
963 	struct sock *ssk;
964 	u32 remote_token;
965 	int addrlen;
966 	int err;
967 
968 	if (sk->sk_state != TCP_ESTABLISHED)
969 		return -ENOTCONN;
970 
971 	err = mptcp_subflow_create_socket(sk, &sf);
972 	if (err)
973 		return err;
974 
975 	ssk = sf->sk;
976 	subflow = mptcp_subflow_ctx(ssk);
977 	do {
978 		get_random_bytes(&subflow->local_nonce, sizeof(u32));
979 	} while (!subflow->local_nonce);
980 
981 	if (!local_id) {
982 		err = mptcp_pm_get_local_id(msk, (struct sock_common *)ssk);
983 		if (err < 0)
984 			goto failed;
985 
986 		local_id = err;
987 	}
988 
989 	subflow->remote_key = msk->remote_key;
990 	subflow->local_key = msk->local_key;
991 	subflow->token = msk->token;
992 	mptcp_info2sockaddr(loc, &addr);
993 
994 	addrlen = sizeof(struct sockaddr_in);
995 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
996 	if (loc->family == AF_INET6)
997 		addrlen = sizeof(struct sockaddr_in6);
998 #endif
999 	ssk->sk_bound_dev_if = ifindex;
1000 	err = kernel_bind(sf, (struct sockaddr *)&addr, addrlen);
1001 	if (err)
1002 		goto failed;
1003 
1004 	mptcp_crypto_key_sha(subflow->remote_key, &remote_token, NULL);
1005 	pr_debug("msk=%p remote_token=%u local_id=%d", msk, remote_token,
1006 		 local_id);
1007 	subflow->remote_token = remote_token;
1008 	subflow->local_id = local_id;
1009 	subflow->request_join = 1;
1010 	subflow->request_bkup = 1;
1011 	mptcp_info2sockaddr(remote, &addr);
1012 
1013 	err = kernel_connect(sf, (struct sockaddr *)&addr, addrlen, O_NONBLOCK);
1014 	if (err && err != -EINPROGRESS)
1015 		goto failed;
1016 
1017 	spin_lock_bh(&msk->join_list_lock);
1018 	list_add_tail(&subflow->node, &msk->join_list);
1019 	spin_unlock_bh(&msk->join_list_lock);
1020 
1021 	return err;
1022 
1023 failed:
1024 	sock_release(sf);
1025 	return err;
1026 }
1027 
1028 int mptcp_subflow_create_socket(struct sock *sk, struct socket **new_sock)
1029 {
1030 	struct mptcp_subflow_context *subflow;
1031 	struct net *net = sock_net(sk);
1032 	struct socket *sf;
1033 	int err;
1034 
1035 	err = sock_create_kern(net, sk->sk_family, SOCK_STREAM, IPPROTO_TCP,
1036 			       &sf);
1037 	if (err)
1038 		return err;
1039 
1040 	lock_sock(sf->sk);
1041 
1042 	/* kernel sockets do not by default acquire net ref, but TCP timer
1043 	 * needs it.
1044 	 */
1045 	sf->sk->sk_net_refcnt = 1;
1046 	get_net(net);
1047 #ifdef CONFIG_PROC_FS
1048 	this_cpu_add(*net->core.sock_inuse, 1);
1049 #endif
1050 	err = tcp_set_ulp(sf->sk, "mptcp");
1051 	release_sock(sf->sk);
1052 
1053 	if (err) {
1054 		sock_release(sf);
1055 		return err;
1056 	}
1057 
1058 	/* the newly created socket really belongs to the owning MPTCP master
1059 	 * socket, even if for additional subflows the allocation is performed
1060 	 * by a kernel workqueue. Adjust inode references, so that the
1061 	 * procfs/diag interaces really show this one belonging to the correct
1062 	 * user.
1063 	 */
1064 	SOCK_INODE(sf)->i_ino = SOCK_INODE(sk->sk_socket)->i_ino;
1065 	SOCK_INODE(sf)->i_uid = SOCK_INODE(sk->sk_socket)->i_uid;
1066 	SOCK_INODE(sf)->i_gid = SOCK_INODE(sk->sk_socket)->i_gid;
1067 
1068 	subflow = mptcp_subflow_ctx(sf->sk);
1069 	pr_debug("subflow=%p", subflow);
1070 
1071 	*new_sock = sf;
1072 	sock_hold(sk);
1073 	subflow->conn = sk;
1074 
1075 	return 0;
1076 }
1077 
1078 static struct mptcp_subflow_context *subflow_create_ctx(struct sock *sk,
1079 							gfp_t priority)
1080 {
1081 	struct inet_connection_sock *icsk = inet_csk(sk);
1082 	struct mptcp_subflow_context *ctx;
1083 
1084 	ctx = kzalloc(sizeof(*ctx), priority);
1085 	if (!ctx)
1086 		return NULL;
1087 
1088 	rcu_assign_pointer(icsk->icsk_ulp_data, ctx);
1089 	INIT_LIST_HEAD(&ctx->node);
1090 
1091 	pr_debug("subflow=%p", ctx);
1092 
1093 	ctx->tcp_sock = sk;
1094 
1095 	return ctx;
1096 }
1097 
1098 static void __subflow_state_change(struct sock *sk)
1099 {
1100 	struct socket_wq *wq;
1101 
1102 	rcu_read_lock();
1103 	wq = rcu_dereference(sk->sk_wq);
1104 	if (skwq_has_sleeper(wq))
1105 		wake_up_interruptible_all(&wq->wait);
1106 	rcu_read_unlock();
1107 }
1108 
1109 static bool subflow_is_done(const struct sock *sk)
1110 {
1111 	return sk->sk_shutdown & RCV_SHUTDOWN || sk->sk_state == TCP_CLOSE;
1112 }
1113 
1114 static void subflow_state_change(struct sock *sk)
1115 {
1116 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1117 	struct sock *parent = subflow->conn;
1118 
1119 	__subflow_state_change(sk);
1120 
1121 	if (subflow_simultaneous_connect(sk)) {
1122 		mptcp_do_fallback(sk);
1123 		mptcp_rcv_space_init(mptcp_sk(parent), sk);
1124 		pr_fallback(mptcp_sk(parent));
1125 		subflow->conn_finished = 1;
1126 		if (inet_sk_state_load(parent) == TCP_SYN_SENT) {
1127 			inet_sk_state_store(parent, TCP_ESTABLISHED);
1128 			parent->sk_state_change(parent);
1129 		}
1130 	}
1131 
1132 	/* as recvmsg() does not acquire the subflow socket for ssk selection
1133 	 * a fin packet carrying a DSS can be unnoticed if we don't trigger
1134 	 * the data available machinery here.
1135 	 */
1136 	if (mptcp_subflow_data_available(sk))
1137 		mptcp_data_ready(parent, sk);
1138 
1139 	if (!(parent->sk_shutdown & RCV_SHUTDOWN) &&
1140 	    !subflow->rx_eof && subflow_is_done(sk)) {
1141 		subflow->rx_eof = 1;
1142 		mptcp_subflow_eof(parent);
1143 	}
1144 }
1145 
1146 static int subflow_ulp_init(struct sock *sk)
1147 {
1148 	struct inet_connection_sock *icsk = inet_csk(sk);
1149 	struct mptcp_subflow_context *ctx;
1150 	struct tcp_sock *tp = tcp_sk(sk);
1151 	int err = 0;
1152 
1153 	/* disallow attaching ULP to a socket unless it has been
1154 	 * created with sock_create_kern()
1155 	 */
1156 	if (!sk->sk_kern_sock) {
1157 		err = -EOPNOTSUPP;
1158 		goto out;
1159 	}
1160 
1161 	ctx = subflow_create_ctx(sk, GFP_KERNEL);
1162 	if (!ctx) {
1163 		err = -ENOMEM;
1164 		goto out;
1165 	}
1166 
1167 	pr_debug("subflow=%p, family=%d", ctx, sk->sk_family);
1168 
1169 	tp->is_mptcp = 1;
1170 	ctx->icsk_af_ops = icsk->icsk_af_ops;
1171 	icsk->icsk_af_ops = subflow_default_af_ops(sk);
1172 	ctx->tcp_data_ready = sk->sk_data_ready;
1173 	ctx->tcp_state_change = sk->sk_state_change;
1174 	ctx->tcp_write_space = sk->sk_write_space;
1175 	sk->sk_data_ready = subflow_data_ready;
1176 	sk->sk_write_space = subflow_write_space;
1177 	sk->sk_state_change = subflow_state_change;
1178 out:
1179 	return err;
1180 }
1181 
1182 static void subflow_ulp_release(struct sock *sk)
1183 {
1184 	struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(sk);
1185 
1186 	if (!ctx)
1187 		return;
1188 
1189 	if (ctx->conn)
1190 		sock_put(ctx->conn);
1191 
1192 	kfree_rcu(ctx, rcu);
1193 }
1194 
1195 static void subflow_ulp_clone(const struct request_sock *req,
1196 			      struct sock *newsk,
1197 			      const gfp_t priority)
1198 {
1199 	struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
1200 	struct mptcp_subflow_context *old_ctx = mptcp_subflow_ctx(newsk);
1201 	struct mptcp_subflow_context *new_ctx;
1202 
1203 	if (!tcp_rsk(req)->is_mptcp ||
1204 	    (!subflow_req->mp_capable && !subflow_req->mp_join)) {
1205 		subflow_ulp_fallback(newsk, old_ctx);
1206 		return;
1207 	}
1208 
1209 	new_ctx = subflow_create_ctx(newsk, priority);
1210 	if (!new_ctx) {
1211 		subflow_ulp_fallback(newsk, old_ctx);
1212 		return;
1213 	}
1214 
1215 	new_ctx->conn_finished = 1;
1216 	new_ctx->icsk_af_ops = old_ctx->icsk_af_ops;
1217 	new_ctx->tcp_data_ready = old_ctx->tcp_data_ready;
1218 	new_ctx->tcp_state_change = old_ctx->tcp_state_change;
1219 	new_ctx->tcp_write_space = old_ctx->tcp_write_space;
1220 	new_ctx->rel_write_seq = 1;
1221 	new_ctx->tcp_sock = newsk;
1222 
1223 	if (subflow_req->mp_capable) {
1224 		/* see comments in subflow_syn_recv_sock(), MPTCP connection
1225 		 * is fully established only after we receive the remote key
1226 		 */
1227 		new_ctx->mp_capable = 1;
1228 		new_ctx->local_key = subflow_req->local_key;
1229 		new_ctx->token = subflow_req->token;
1230 		new_ctx->ssn_offset = subflow_req->ssn_offset;
1231 		new_ctx->idsn = subflow_req->idsn;
1232 	} else if (subflow_req->mp_join) {
1233 		new_ctx->ssn_offset = subflow_req->ssn_offset;
1234 		new_ctx->mp_join = 1;
1235 		new_ctx->fully_established = 1;
1236 		new_ctx->backup = subflow_req->backup;
1237 		new_ctx->local_id = subflow_req->local_id;
1238 		new_ctx->token = subflow_req->token;
1239 		new_ctx->thmac = subflow_req->thmac;
1240 	}
1241 }
1242 
1243 static struct tcp_ulp_ops subflow_ulp_ops __read_mostly = {
1244 	.name		= "mptcp",
1245 	.owner		= THIS_MODULE,
1246 	.init		= subflow_ulp_init,
1247 	.release	= subflow_ulp_release,
1248 	.clone		= subflow_ulp_clone,
1249 };
1250 
1251 static int subflow_ops_init(struct request_sock_ops *subflow_ops)
1252 {
1253 	subflow_ops->obj_size = sizeof(struct mptcp_subflow_request_sock);
1254 	subflow_ops->slab_name = "request_sock_subflow";
1255 
1256 	subflow_ops->slab = kmem_cache_create(subflow_ops->slab_name,
1257 					      subflow_ops->obj_size, 0,
1258 					      SLAB_ACCOUNT |
1259 					      SLAB_TYPESAFE_BY_RCU,
1260 					      NULL);
1261 	if (!subflow_ops->slab)
1262 		return -ENOMEM;
1263 
1264 	subflow_ops->destructor = subflow_req_destructor;
1265 
1266 	return 0;
1267 }
1268 
1269 void __init mptcp_subflow_init(void)
1270 {
1271 	subflow_request_sock_ops = tcp_request_sock_ops;
1272 	if (subflow_ops_init(&subflow_request_sock_ops) != 0)
1273 		panic("MPTCP: failed to init subflow request sock ops\n");
1274 
1275 	subflow_request_sock_ipv4_ops = tcp_request_sock_ipv4_ops;
1276 	subflow_request_sock_ipv4_ops.init_req = subflow_v4_init_req;
1277 
1278 	subflow_specific = ipv4_specific;
1279 	subflow_specific.conn_request = subflow_v4_conn_request;
1280 	subflow_specific.syn_recv_sock = subflow_syn_recv_sock;
1281 	subflow_specific.sk_rx_dst_set = subflow_finish_connect;
1282 
1283 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1284 	subflow_request_sock_ipv6_ops = tcp_request_sock_ipv6_ops;
1285 	subflow_request_sock_ipv6_ops.init_req = subflow_v6_init_req;
1286 
1287 	subflow_v6_specific = ipv6_specific;
1288 	subflow_v6_specific.conn_request = subflow_v6_conn_request;
1289 	subflow_v6_specific.syn_recv_sock = subflow_syn_recv_sock;
1290 	subflow_v6_specific.sk_rx_dst_set = subflow_finish_connect;
1291 
1292 	subflow_v6m_specific = subflow_v6_specific;
1293 	subflow_v6m_specific.queue_xmit = ipv4_specific.queue_xmit;
1294 	subflow_v6m_specific.send_check = ipv4_specific.send_check;
1295 	subflow_v6m_specific.net_header_len = ipv4_specific.net_header_len;
1296 	subflow_v6m_specific.mtu_reduced = ipv4_specific.mtu_reduced;
1297 	subflow_v6m_specific.net_frag_header_len = 0;
1298 #endif
1299 
1300 	mptcp_diag_subflow_init(&subflow_ulp_ops);
1301 
1302 	if (tcp_register_ulp(&subflow_ulp_ops) != 0)
1303 		panic("MPTCP: failed to register subflows to ULP\n");
1304 }
1305