xref: /openbmc/linux/net/mptcp/subflow.c (revision 0f9b4c3ca5fdf3e177266ef994071b1a03f07318)
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/sha2.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 #include <net/transp_v6.h>
22 #endif
23 #include <net/mptcp.h>
24 #include <uapi/linux/mptcp.h>
25 #include "protocol.h"
26 #include "mib.h"
27 
28 #include <trace/events/mptcp.h>
29 #include <trace/events/sock.h>
30 
31 static void mptcp_subflow_ops_undo_override(struct sock *ssk);
32 
SUBFLOW_REQ_INC_STATS(struct request_sock * req,enum linux_mptcp_mib_field field)33 static void SUBFLOW_REQ_INC_STATS(struct request_sock *req,
34 				  enum linux_mptcp_mib_field field)
35 {
36 	MPTCP_INC_STATS(sock_net(req_to_sk(req)), field);
37 }
38 
subflow_req_destructor(struct request_sock * req)39 static void subflow_req_destructor(struct request_sock *req)
40 {
41 	struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
42 
43 	pr_debug("subflow_req=%p\n", subflow_req);
44 
45 	if (subflow_req->msk)
46 		sock_put((struct sock *)subflow_req->msk);
47 
48 	mptcp_token_destroy_request(req);
49 }
50 
subflow_generate_hmac(u64 key1,u64 key2,u32 nonce1,u32 nonce2,void * hmac)51 static void subflow_generate_hmac(u64 key1, u64 key2, u32 nonce1, u32 nonce2,
52 				  void *hmac)
53 {
54 	u8 msg[8];
55 
56 	put_unaligned_be32(nonce1, &msg[0]);
57 	put_unaligned_be32(nonce2, &msg[4]);
58 
59 	mptcp_crypto_hmac_sha(key1, key2, msg, 8, hmac);
60 }
61 
mptcp_can_accept_new_subflow(const struct mptcp_sock * msk)62 static bool mptcp_can_accept_new_subflow(const struct mptcp_sock *msk)
63 {
64 	return mptcp_is_fully_established((void *)msk) &&
65 		((mptcp_pm_is_userspace(msk) &&
66 		  mptcp_userspace_pm_active(msk)) ||
67 		 READ_ONCE(msk->pm.accept_subflow));
68 }
69 
70 /* validate received token and create truncated hmac and nonce for SYN-ACK */
subflow_req_create_thmac(struct mptcp_subflow_request_sock * subflow_req)71 static void subflow_req_create_thmac(struct mptcp_subflow_request_sock *subflow_req)
72 {
73 	struct mptcp_sock *msk = subflow_req->msk;
74 	u8 hmac[SHA256_DIGEST_SIZE];
75 
76 	get_random_bytes(&subflow_req->local_nonce, sizeof(u32));
77 
78 	subflow_generate_hmac(msk->local_key, 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 
subflow_token_join_request(struct request_sock * req)85 static struct mptcp_sock *subflow_token_join_request(struct request_sock *req)
86 {
87 	struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
88 	struct mptcp_sock *msk;
89 	int local_id;
90 
91 	msk = mptcp_token_get_sock(sock_net(req_to_sk(req)), subflow_req->token);
92 	if (!msk) {
93 		SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINNOTOKEN);
94 		return NULL;
95 	}
96 
97 	local_id = mptcp_pm_get_local_id(msk, (struct sock_common *)req);
98 	if (local_id < 0) {
99 		sock_put((struct sock *)msk);
100 		return NULL;
101 	}
102 	subflow_req->local_id = local_id;
103 	subflow_req->request_bkup = mptcp_pm_is_backup(msk, (struct sock_common *)req);
104 
105 	return msk;
106 }
107 
subflow_init_req(struct request_sock * req,const struct sock * sk_listener)108 static void subflow_init_req(struct request_sock *req, const struct sock *sk_listener)
109 {
110 	struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
111 
112 	subflow_req->mp_capable = 0;
113 	subflow_req->mp_join = 0;
114 	subflow_req->csum_reqd = mptcp_is_checksum_enabled(sock_net(sk_listener));
115 	subflow_req->allow_join_id0 = mptcp_allow_join_id0(sock_net(sk_listener));
116 	subflow_req->msk = NULL;
117 	mptcp_token_init_request(req);
118 }
119 
subflow_use_different_sport(struct mptcp_sock * msk,const struct sock * sk)120 static bool subflow_use_different_sport(struct mptcp_sock *msk, const struct sock *sk)
121 {
122 	return inet_sk(sk)->inet_sport != inet_sk((struct sock *)msk)->inet_sport;
123 }
124 
subflow_add_reset_reason(struct sk_buff * skb,u8 reason)125 static void subflow_add_reset_reason(struct sk_buff *skb, u8 reason)
126 {
127 	struct mptcp_ext *mpext = skb_ext_add(skb, SKB_EXT_MPTCP);
128 
129 	if (mpext) {
130 		memset(mpext, 0, sizeof(*mpext));
131 		mpext->reset_reason = reason;
132 	}
133 }
134 
subflow_reset_req_endp(struct request_sock * req,struct sk_buff * skb)135 static int subflow_reset_req_endp(struct request_sock *req, struct sk_buff *skb)
136 {
137 	SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MPCAPABLEENDPATTEMPT);
138 	subflow_add_reset_reason(skb, MPTCP_RST_EPROHIBIT);
139 	return -EPERM;
140 }
141 
142 /* Init mptcp request socket.
143  *
144  * Returns an error code if a JOIN has failed and a TCP reset
145  * should be sent.
146  */
subflow_check_req(struct request_sock * req,const struct sock * sk_listener,struct sk_buff * skb)147 static int subflow_check_req(struct request_sock *req,
148 			     const struct sock *sk_listener,
149 			     struct sk_buff *skb)
150 {
151 	struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk_listener);
152 	struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
153 	struct mptcp_options_received mp_opt;
154 	bool opt_mp_capable, opt_mp_join;
155 
156 	pr_debug("subflow_req=%p, listener=%p\n", subflow_req, listener);
157 
158 #ifdef CONFIG_TCP_MD5SIG
159 	/* no MPTCP if MD5SIG is enabled on this socket or we may run out of
160 	 * TCP option space.
161 	 */
162 	if (rcu_access_pointer(tcp_sk(sk_listener)->md5sig_info))
163 		return -EINVAL;
164 #endif
165 
166 	mptcp_get_options(skb, &mp_opt);
167 
168 	opt_mp_capable = !!(mp_opt.suboptions & OPTION_MPTCP_MPC_SYN);
169 	opt_mp_join = !!(mp_opt.suboptions & OPTION_MPTCP_MPJ_SYN);
170 	if (opt_mp_capable) {
171 		SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MPCAPABLEPASSIVE);
172 
173 		if (unlikely(listener->pm_listener))
174 			return subflow_reset_req_endp(req, skb);
175 		if (opt_mp_join)
176 			return 0;
177 	} else if (opt_mp_join) {
178 		SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINSYNRX);
179 
180 		if (mp_opt.backup)
181 			SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINSYNBACKUPRX);
182 	} else if (unlikely(listener->pm_listener)) {
183 		return subflow_reset_req_endp(req, skb);
184 	}
185 
186 	if (opt_mp_capable && listener->request_mptcp) {
187 		int err, retries = MPTCP_TOKEN_MAX_RETRIES;
188 
189 		subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq;
190 again:
191 		do {
192 			get_random_bytes(&subflow_req->local_key, sizeof(subflow_req->local_key));
193 		} while (subflow_req->local_key == 0);
194 
195 		if (unlikely(req->syncookie)) {
196 			mptcp_crypto_key_sha(subflow_req->local_key,
197 					     &subflow_req->token,
198 					     &subflow_req->idsn);
199 			if (mptcp_token_exists(subflow_req->token)) {
200 				if (retries-- > 0)
201 					goto again;
202 				SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_TOKENFALLBACKINIT);
203 			} else {
204 				subflow_req->mp_capable = 1;
205 			}
206 			return 0;
207 		}
208 
209 		err = mptcp_token_new_request(req);
210 		if (err == 0)
211 			subflow_req->mp_capable = 1;
212 		else if (retries-- > 0)
213 			goto again;
214 		else
215 			SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_TOKENFALLBACKINIT);
216 
217 	} else if (opt_mp_join && listener->request_mptcp) {
218 		subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq;
219 		subflow_req->mp_join = 1;
220 		subflow_req->backup = mp_opt.backup;
221 		subflow_req->remote_id = mp_opt.join_id;
222 		subflow_req->token = mp_opt.token;
223 		subflow_req->remote_nonce = mp_opt.nonce;
224 		subflow_req->msk = subflow_token_join_request(req);
225 
226 		/* Can't fall back to TCP in this case. */
227 		if (!subflow_req->msk) {
228 			subflow_add_reset_reason(skb, MPTCP_RST_EMPTCP);
229 			return -EPERM;
230 		}
231 
232 		if (subflow_use_different_sport(subflow_req->msk, sk_listener)) {
233 			pr_debug("syn inet_sport=%d %d\n",
234 				 ntohs(inet_sk(sk_listener)->inet_sport),
235 				 ntohs(inet_sk((struct sock *)subflow_req->msk)->inet_sport));
236 			if (!mptcp_pm_sport_in_anno_list(subflow_req->msk, sk_listener)) {
237 				SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MISMATCHPORTSYNRX);
238 				return -EPERM;
239 			}
240 			SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINPORTSYNRX);
241 		}
242 
243 		subflow_req_create_thmac(subflow_req);
244 
245 		if (unlikely(req->syncookie)) {
246 			if (mptcp_can_accept_new_subflow(subflow_req->msk))
247 				subflow_init_req_cookie_join_save(subflow_req, skb);
248 			else
249 				return -EPERM;
250 		}
251 
252 		pr_debug("token=%u, remote_nonce=%u msk=%p\n", subflow_req->token,
253 			 subflow_req->remote_nonce, subflow_req->msk);
254 	}
255 
256 	return 0;
257 }
258 
mptcp_subflow_init_cookie_req(struct request_sock * req,const struct sock * sk_listener,struct sk_buff * skb)259 int mptcp_subflow_init_cookie_req(struct request_sock *req,
260 				  const struct sock *sk_listener,
261 				  struct sk_buff *skb)
262 {
263 	struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk_listener);
264 	struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
265 	struct mptcp_options_received mp_opt;
266 	bool opt_mp_capable, opt_mp_join;
267 	int err;
268 
269 	subflow_init_req(req, sk_listener);
270 	mptcp_get_options(skb, &mp_opt);
271 
272 	opt_mp_capable = !!(mp_opt.suboptions & OPTION_MPTCP_MPC_ACK);
273 	opt_mp_join = !!(mp_opt.suboptions & OPTION_MPTCP_MPJ_ACK);
274 	if (opt_mp_capable && opt_mp_join)
275 		return -EINVAL;
276 
277 	if (opt_mp_capable && listener->request_mptcp) {
278 		if (mp_opt.sndr_key == 0)
279 			return -EINVAL;
280 
281 		subflow_req->local_key = mp_opt.rcvr_key;
282 		err = mptcp_token_new_request(req);
283 		if (err)
284 			return err;
285 
286 		subflow_req->mp_capable = 1;
287 		subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq - 1;
288 	} else if (opt_mp_join && listener->request_mptcp) {
289 		if (!mptcp_token_join_cookie_init_state(subflow_req, skb))
290 			return -EINVAL;
291 
292 		subflow_req->mp_join = 1;
293 		subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq - 1;
294 	}
295 
296 	return 0;
297 }
298 EXPORT_SYMBOL_GPL(mptcp_subflow_init_cookie_req);
299 
subflow_v4_route_req(const struct sock * sk,struct sk_buff * skb,struct flowi * fl,struct request_sock * req)300 static struct dst_entry *subflow_v4_route_req(const struct sock *sk,
301 					      struct sk_buff *skb,
302 					      struct flowi *fl,
303 					      struct request_sock *req)
304 {
305 	struct dst_entry *dst;
306 	int err;
307 
308 	tcp_rsk(req)->is_mptcp = 1;
309 	subflow_init_req(req, sk);
310 
311 	dst = tcp_request_sock_ipv4_ops.route_req(sk, skb, fl, req);
312 	if (!dst)
313 		return NULL;
314 
315 	err = subflow_check_req(req, sk, skb);
316 	if (err == 0)
317 		return dst;
318 
319 	dst_release(dst);
320 	if (!req->syncookie)
321 		tcp_request_sock_ops.send_reset(sk, skb);
322 	return NULL;
323 }
324 
subflow_prep_synack(const struct sock * sk,struct request_sock * req,struct tcp_fastopen_cookie * foc,enum tcp_synack_type synack_type)325 static void subflow_prep_synack(const struct sock *sk, struct request_sock *req,
326 				struct tcp_fastopen_cookie *foc,
327 				enum tcp_synack_type synack_type)
328 {
329 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
330 	struct inet_request_sock *ireq = inet_rsk(req);
331 
332 	/* clear tstamp_ok, as needed depending on cookie */
333 	if (foc && foc->len > -1)
334 		ireq->tstamp_ok = 0;
335 
336 	if (synack_type == TCP_SYNACK_FASTOPEN)
337 		mptcp_fastopen_subflow_synack_set_params(subflow, req);
338 }
339 
subflow_v4_send_synack(const struct sock * sk,struct dst_entry * dst,struct flowi * fl,struct request_sock * req,struct tcp_fastopen_cookie * foc,enum tcp_synack_type synack_type,struct sk_buff * syn_skb)340 static int subflow_v4_send_synack(const struct sock *sk, struct dst_entry *dst,
341 				  struct flowi *fl,
342 				  struct request_sock *req,
343 				  struct tcp_fastopen_cookie *foc,
344 				  enum tcp_synack_type synack_type,
345 				  struct sk_buff *syn_skb)
346 {
347 	subflow_prep_synack(sk, req, foc, synack_type);
348 
349 	return tcp_request_sock_ipv4_ops.send_synack(sk, dst, fl, req, foc,
350 						     synack_type, syn_skb);
351 }
352 
353 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
subflow_v6_send_synack(const struct sock * sk,struct dst_entry * dst,struct flowi * fl,struct request_sock * req,struct tcp_fastopen_cookie * foc,enum tcp_synack_type synack_type,struct sk_buff * syn_skb)354 static int subflow_v6_send_synack(const struct sock *sk, struct dst_entry *dst,
355 				  struct flowi *fl,
356 				  struct request_sock *req,
357 				  struct tcp_fastopen_cookie *foc,
358 				  enum tcp_synack_type synack_type,
359 				  struct sk_buff *syn_skb)
360 {
361 	subflow_prep_synack(sk, req, foc, synack_type);
362 
363 	return tcp_request_sock_ipv6_ops.send_synack(sk, dst, fl, req, foc,
364 						     synack_type, syn_skb);
365 }
366 
subflow_v6_route_req(const struct sock * sk,struct sk_buff * skb,struct flowi * fl,struct request_sock * req)367 static struct dst_entry *subflow_v6_route_req(const struct sock *sk,
368 					      struct sk_buff *skb,
369 					      struct flowi *fl,
370 					      struct request_sock *req)
371 {
372 	struct dst_entry *dst;
373 	int err;
374 
375 	tcp_rsk(req)->is_mptcp = 1;
376 	subflow_init_req(req, sk);
377 
378 	dst = tcp_request_sock_ipv6_ops.route_req(sk, skb, fl, req);
379 	if (!dst)
380 		return NULL;
381 
382 	err = subflow_check_req(req, sk, skb);
383 	if (err == 0)
384 		return dst;
385 
386 	dst_release(dst);
387 	if (!req->syncookie)
388 		tcp6_request_sock_ops.send_reset(sk, skb);
389 	return NULL;
390 }
391 #endif
392 
393 /* validate received truncated hmac and create hmac for third ACK */
subflow_thmac_valid(struct mptcp_subflow_context * subflow)394 static bool subflow_thmac_valid(struct mptcp_subflow_context *subflow)
395 {
396 	u8 hmac[SHA256_DIGEST_SIZE];
397 	u64 thmac;
398 
399 	subflow_generate_hmac(subflow->remote_key, subflow->local_key,
400 			      subflow->remote_nonce, subflow->local_nonce,
401 			      hmac);
402 
403 	thmac = get_unaligned_be64(hmac);
404 	pr_debug("subflow=%p, token=%u, thmac=%llu, subflow->thmac=%llu\n",
405 		 subflow, subflow->token, thmac, subflow->thmac);
406 
407 	return thmac == subflow->thmac;
408 }
409 
mptcp_subflow_reset(struct sock * ssk)410 void mptcp_subflow_reset(struct sock *ssk)
411 {
412 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
413 	struct sock *sk = subflow->conn;
414 
415 	/* mptcp_mp_fail_no_response() can reach here on an already closed
416 	 * socket
417 	 */
418 	if (ssk->sk_state == TCP_CLOSE)
419 		return;
420 
421 	/* must hold: tcp_done() could drop last reference on parent */
422 	sock_hold(sk);
423 
424 	tcp_send_active_reset(ssk, GFP_ATOMIC);
425 	tcp_done(ssk);
426 	if (!test_and_set_bit(MPTCP_WORK_CLOSE_SUBFLOW, &mptcp_sk(sk)->flags))
427 		mptcp_schedule_work(sk);
428 
429 	sock_put(sk);
430 }
431 
subflow_use_different_dport(struct mptcp_sock * msk,const struct sock * sk)432 static bool subflow_use_different_dport(struct mptcp_sock *msk, const struct sock *sk)
433 {
434 	return inet_sk(sk)->inet_dport != inet_sk((struct sock *)msk)->inet_dport;
435 }
436 
__mptcp_sync_state(struct sock * sk,int state)437 void __mptcp_sync_state(struct sock *sk, int state)
438 {
439 	struct mptcp_subflow_context *subflow;
440 	struct mptcp_sock *msk = mptcp_sk(sk);
441 	struct sock *ssk = msk->first;
442 
443 	subflow = mptcp_subflow_ctx(ssk);
444 	__mptcp_propagate_sndbuf(sk, ssk);
445 	if (!msk->rcvspace_init)
446 		mptcp_rcv_space_init(msk, ssk);
447 
448 	if (sk->sk_state == TCP_SYN_SENT) {
449 		/* subflow->idsn is always available is TCP_SYN_SENT state,
450 		 * even for the FASTOPEN scenarios
451 		 */
452 		WRITE_ONCE(msk->write_seq, subflow->idsn + 1);
453 		WRITE_ONCE(msk->snd_nxt, msk->write_seq);
454 		mptcp_set_state(sk, state);
455 		sk->sk_state_change(sk);
456 	}
457 }
458 
subflow_set_remote_key(struct mptcp_sock * msk,struct mptcp_subflow_context * subflow,const struct mptcp_options_received * mp_opt)459 static void subflow_set_remote_key(struct mptcp_sock *msk,
460 				   struct mptcp_subflow_context *subflow,
461 				   const struct mptcp_options_received *mp_opt)
462 {
463 	/* active MPC subflow will reach here multiple times:
464 	 * at subflow_finish_connect() time and at 4th ack time
465 	 */
466 	if (subflow->remote_key_valid)
467 		return;
468 
469 	subflow->remote_key_valid = 1;
470 	subflow->remote_key = mp_opt->sndr_key;
471 	mptcp_crypto_key_sha(subflow->remote_key, NULL, &subflow->iasn);
472 	subflow->iasn++;
473 
474 	WRITE_ONCE(msk->remote_key, subflow->remote_key);
475 	WRITE_ONCE(msk->ack_seq, subflow->iasn);
476 	WRITE_ONCE(msk->can_ack, true);
477 	atomic64_set(&msk->rcv_wnd_sent, subflow->iasn);
478 }
479 
mptcp_propagate_state(struct sock * sk,struct sock * ssk,struct mptcp_subflow_context * subflow,const struct mptcp_options_received * mp_opt)480 static void mptcp_propagate_state(struct sock *sk, struct sock *ssk,
481 				  struct mptcp_subflow_context *subflow,
482 				  const struct mptcp_options_received *mp_opt)
483 {
484 	struct mptcp_sock *msk = mptcp_sk(sk);
485 
486 	mptcp_data_lock(sk);
487 	if (mp_opt) {
488 		/* Options are available only in the non fallback cases
489 		 * avoid updating rx path fields otherwise
490 		 */
491 		WRITE_ONCE(msk->snd_una, subflow->idsn + 1);
492 		WRITE_ONCE(msk->wnd_end, subflow->idsn + 1 + tcp_sk(ssk)->snd_wnd);
493 		subflow_set_remote_key(msk, subflow, mp_opt);
494 	}
495 
496 	if (!sock_owned_by_user(sk)) {
497 		__mptcp_sync_state(sk, ssk->sk_state);
498 	} else {
499 		msk->pending_state = ssk->sk_state;
500 		__set_bit(MPTCP_SYNC_STATE, &msk->cb_flags);
501 	}
502 	mptcp_data_unlock(sk);
503 }
504 
subflow_finish_connect(struct sock * sk,const struct sk_buff * skb)505 static void subflow_finish_connect(struct sock *sk, const struct sk_buff *skb)
506 {
507 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
508 	struct mptcp_options_received mp_opt;
509 	struct sock *parent = subflow->conn;
510 	struct mptcp_sock *msk;
511 
512 	subflow->icsk_af_ops->sk_rx_dst_set(sk, skb);
513 
514 	/* be sure no special action on any packet other than syn-ack */
515 	if (subflow->conn_finished)
516 		return;
517 
518 	msk = mptcp_sk(parent);
519 	subflow->rel_write_seq = 1;
520 	subflow->conn_finished = 1;
521 	subflow->ssn_offset = TCP_SKB_CB(skb)->seq;
522 	pr_debug("subflow=%p synack seq=%x\n", subflow, subflow->ssn_offset);
523 
524 	mptcp_get_options(skb, &mp_opt);
525 	if (subflow->request_mptcp) {
526 		if (!(mp_opt.suboptions & OPTION_MPTCP_MPC_SYNACK)) {
527 			MPTCP_INC_STATS(sock_net(sk),
528 					MPTCP_MIB_MPCAPABLEACTIVEFALLBACK);
529 			mptcp_do_fallback(sk);
530 			pr_fallback(msk);
531 			goto fallback;
532 		}
533 
534 		if (mp_opt.suboptions & OPTION_MPTCP_CSUMREQD)
535 			WRITE_ONCE(msk->csum_enabled, true);
536 		if (mp_opt.deny_join_id0)
537 			WRITE_ONCE(msk->pm.remote_deny_join_id0, true);
538 		subflow->mp_capable = 1;
539 		MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPCAPABLEACTIVEACK);
540 		mptcp_finish_connect(sk);
541 		mptcp_propagate_state(parent, sk, subflow, &mp_opt);
542 	} else if (subflow->request_join) {
543 		u8 hmac[SHA256_DIGEST_SIZE];
544 
545 		if (!(mp_opt.suboptions & OPTION_MPTCP_MPJ_SYNACK)) {
546 			subflow->reset_reason = MPTCP_RST_EMPTCP;
547 			goto do_reset;
548 		}
549 
550 		subflow->backup = mp_opt.backup;
551 		subflow->thmac = mp_opt.thmac;
552 		subflow->remote_nonce = mp_opt.nonce;
553 		WRITE_ONCE(subflow->remote_id, mp_opt.join_id);
554 		pr_debug("subflow=%p, thmac=%llu, remote_nonce=%u backup=%d\n",
555 			 subflow, subflow->thmac, subflow->remote_nonce,
556 			 subflow->backup);
557 
558 		if (!subflow_thmac_valid(subflow)) {
559 			MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINACKMAC);
560 			subflow->reset_reason = MPTCP_RST_EMPTCP;
561 			goto do_reset;
562 		}
563 
564 		if (!mptcp_finish_join(sk))
565 			goto do_reset;
566 
567 		subflow_generate_hmac(subflow->local_key, subflow->remote_key,
568 				      subflow->local_nonce,
569 				      subflow->remote_nonce,
570 				      hmac);
571 		memcpy(subflow->hmac, hmac, MPTCPOPT_HMAC_LEN);
572 
573 		subflow->mp_join = 1;
574 		MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINSYNACKRX);
575 
576 		if (subflow->backup)
577 			MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINSYNACKBACKUPRX);
578 
579 		if (subflow_use_different_dport(msk, sk)) {
580 			pr_debug("synack inet_dport=%d %d\n",
581 				 ntohs(inet_sk(sk)->inet_dport),
582 				 ntohs(inet_sk(parent)->inet_dport));
583 			MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINPORTSYNACKRX);
584 		}
585 	} else if (mptcp_check_fallback(sk)) {
586 fallback:
587 		mptcp_propagate_state(parent, sk, subflow, NULL);
588 	}
589 	return;
590 
591 do_reset:
592 	subflow->reset_transient = 0;
593 	mptcp_subflow_reset(sk);
594 }
595 
subflow_set_local_id(struct mptcp_subflow_context * subflow,int local_id)596 static void subflow_set_local_id(struct mptcp_subflow_context *subflow, int local_id)
597 {
598 	WARN_ON_ONCE(local_id < 0 || local_id > 255);
599 	WRITE_ONCE(subflow->local_id, local_id);
600 }
601 
subflow_chk_local_id(struct sock * sk)602 static int subflow_chk_local_id(struct sock *sk)
603 {
604 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
605 	struct mptcp_sock *msk = mptcp_sk(subflow->conn);
606 	int err;
607 
608 	if (likely(subflow->local_id >= 0))
609 		return 0;
610 
611 	err = mptcp_pm_get_local_id(msk, (struct sock_common *)sk);
612 	if (err < 0)
613 		return err;
614 
615 	subflow_set_local_id(subflow, err);
616 	subflow->request_bkup = mptcp_pm_is_backup(msk, (struct sock_common *)sk);
617 
618 	return 0;
619 }
620 
subflow_rebuild_header(struct sock * sk)621 static int subflow_rebuild_header(struct sock *sk)
622 {
623 	int err = subflow_chk_local_id(sk);
624 
625 	if (unlikely(err < 0))
626 		return err;
627 
628 	return inet_sk_rebuild_header(sk);
629 }
630 
631 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
subflow_v6_rebuild_header(struct sock * sk)632 static int subflow_v6_rebuild_header(struct sock *sk)
633 {
634 	int err = subflow_chk_local_id(sk);
635 
636 	if (unlikely(err < 0))
637 		return err;
638 
639 	return inet6_sk_rebuild_header(sk);
640 }
641 #endif
642 
643 static struct request_sock_ops mptcp_subflow_v4_request_sock_ops __ro_after_init;
644 static struct tcp_request_sock_ops subflow_request_sock_ipv4_ops __ro_after_init;
645 
subflow_v4_conn_request(struct sock * sk,struct sk_buff * skb)646 static int subflow_v4_conn_request(struct sock *sk, struct sk_buff *skb)
647 {
648 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
649 
650 	pr_debug("subflow=%p\n", subflow);
651 
652 	/* Never answer to SYNs sent to broadcast or multicast */
653 	if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
654 		goto drop;
655 
656 	return tcp_conn_request(&mptcp_subflow_v4_request_sock_ops,
657 				&subflow_request_sock_ipv4_ops,
658 				sk, skb);
659 drop:
660 	tcp_listendrop(sk);
661 	return 0;
662 }
663 
subflow_v4_req_destructor(struct request_sock * req)664 static void subflow_v4_req_destructor(struct request_sock *req)
665 {
666 	subflow_req_destructor(req);
667 	tcp_request_sock_ops.destructor(req);
668 }
669 
670 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
671 static struct request_sock_ops mptcp_subflow_v6_request_sock_ops __ro_after_init;
672 static struct tcp_request_sock_ops subflow_request_sock_ipv6_ops __ro_after_init;
673 static struct inet_connection_sock_af_ops subflow_v6_specific __ro_after_init;
674 static struct inet_connection_sock_af_ops subflow_v6m_specific __ro_after_init;
675 static struct proto tcpv6_prot_override __ro_after_init;
676 
subflow_v6_conn_request(struct sock * sk,struct sk_buff * skb)677 static int subflow_v6_conn_request(struct sock *sk, struct sk_buff *skb)
678 {
679 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
680 
681 	pr_debug("subflow=%p\n", subflow);
682 
683 	if (skb->protocol == htons(ETH_P_IP))
684 		return subflow_v4_conn_request(sk, skb);
685 
686 	if (!ipv6_unicast_destination(skb))
687 		goto drop;
688 
689 	if (ipv6_addr_v4mapped(&ipv6_hdr(skb)->saddr)) {
690 		__IP6_INC_STATS(sock_net(sk), NULL, IPSTATS_MIB_INHDRERRORS);
691 		return 0;
692 	}
693 
694 	return tcp_conn_request(&mptcp_subflow_v6_request_sock_ops,
695 				&subflow_request_sock_ipv6_ops, sk, skb);
696 
697 drop:
698 	tcp_listendrop(sk);
699 	return 0; /* don't send reset */
700 }
701 
subflow_v6_req_destructor(struct request_sock * req)702 static void subflow_v6_req_destructor(struct request_sock *req)
703 {
704 	subflow_req_destructor(req);
705 	tcp6_request_sock_ops.destructor(req);
706 }
707 #endif
708 
mptcp_subflow_reqsk_alloc(const struct request_sock_ops * ops,struct sock * sk_listener,bool attach_listener)709 struct request_sock *mptcp_subflow_reqsk_alloc(const struct request_sock_ops *ops,
710 					       struct sock *sk_listener,
711 					       bool attach_listener)
712 {
713 	if (ops->family == AF_INET)
714 		ops = &mptcp_subflow_v4_request_sock_ops;
715 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
716 	else if (ops->family == AF_INET6)
717 		ops = &mptcp_subflow_v6_request_sock_ops;
718 #endif
719 
720 	return inet_reqsk_alloc(ops, sk_listener, attach_listener);
721 }
722 EXPORT_SYMBOL(mptcp_subflow_reqsk_alloc);
723 
724 /* validate hmac received in third ACK */
subflow_hmac_valid(const struct request_sock * req,const struct mptcp_options_received * mp_opt)725 static bool subflow_hmac_valid(const struct request_sock *req,
726 			       const struct mptcp_options_received *mp_opt)
727 {
728 	const struct mptcp_subflow_request_sock *subflow_req;
729 	u8 hmac[SHA256_DIGEST_SIZE];
730 	struct mptcp_sock *msk;
731 
732 	subflow_req = mptcp_subflow_rsk(req);
733 	msk = subflow_req->msk;
734 	if (!msk)
735 		return false;
736 
737 	subflow_generate_hmac(msk->remote_key, msk->local_key,
738 			      subflow_req->remote_nonce,
739 			      subflow_req->local_nonce, hmac);
740 
741 	return !crypto_memneq(hmac, mp_opt->hmac, MPTCPOPT_HMAC_LEN);
742 }
743 
subflow_ulp_fallback(struct sock * sk,struct mptcp_subflow_context * old_ctx)744 static void subflow_ulp_fallback(struct sock *sk,
745 				 struct mptcp_subflow_context *old_ctx)
746 {
747 	struct inet_connection_sock *icsk = inet_csk(sk);
748 
749 	mptcp_subflow_tcp_fallback(sk, old_ctx);
750 	icsk->icsk_ulp_ops = NULL;
751 	rcu_assign_pointer(icsk->icsk_ulp_data, NULL);
752 	tcp_sk(sk)->is_mptcp = 0;
753 
754 	mptcp_subflow_ops_undo_override(sk);
755 }
756 
mptcp_subflow_drop_ctx(struct sock * ssk)757 void mptcp_subflow_drop_ctx(struct sock *ssk)
758 {
759 	struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(ssk);
760 
761 	if (!ctx)
762 		return;
763 
764 	list_del(&mptcp_subflow_ctx(ssk)->node);
765 	if (inet_csk(ssk)->icsk_ulp_ops) {
766 		subflow_ulp_fallback(ssk, ctx);
767 		if (ctx->conn)
768 			sock_put(ctx->conn);
769 	}
770 
771 	kfree_rcu(ctx, rcu);
772 }
773 
__mptcp_subflow_fully_established(struct mptcp_sock * msk,struct mptcp_subflow_context * subflow,const struct mptcp_options_received * mp_opt)774 void __mptcp_subflow_fully_established(struct mptcp_sock *msk,
775 				       struct mptcp_subflow_context *subflow,
776 				       const struct mptcp_options_received *mp_opt)
777 {
778 	subflow_set_remote_key(msk, subflow, mp_opt);
779 	subflow->fully_established = 1;
780 	WRITE_ONCE(msk->fully_established, true);
781 
782 	if (subflow->is_mptfo)
783 		__mptcp_fastopen_gen_msk_ackseq(msk, subflow, mp_opt);
784 }
785 
subflow_syn_recv_sock(const struct sock * sk,struct sk_buff * skb,struct request_sock * req,struct dst_entry * dst,struct request_sock * req_unhash,bool * own_req)786 static struct sock *subflow_syn_recv_sock(const struct sock *sk,
787 					  struct sk_buff *skb,
788 					  struct request_sock *req,
789 					  struct dst_entry *dst,
790 					  struct request_sock *req_unhash,
791 					  bool *own_req)
792 {
793 	struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk);
794 	struct mptcp_subflow_request_sock *subflow_req;
795 	struct mptcp_options_received mp_opt;
796 	bool fallback, fallback_is_fatal;
797 	struct mptcp_sock *owner;
798 	struct sock *child;
799 
800 	pr_debug("listener=%p, req=%p, conn=%p\n", listener, req, listener->conn);
801 
802 	/* After child creation we must look for MPC even when options
803 	 * are not parsed
804 	 */
805 	mp_opt.suboptions = 0;
806 
807 	/* hopefully temporary handling for MP_JOIN+syncookie */
808 	subflow_req = mptcp_subflow_rsk(req);
809 	fallback_is_fatal = tcp_rsk(req)->is_mptcp && subflow_req->mp_join;
810 	fallback = !tcp_rsk(req)->is_mptcp;
811 	if (fallback)
812 		goto create_child;
813 
814 	/* if the sk is MP_CAPABLE, we try to fetch the client key */
815 	if (subflow_req->mp_capable) {
816 		/* we can receive and accept an in-window, out-of-order pkt,
817 		 * which may not carry the MP_CAPABLE opt even on mptcp enabled
818 		 * paths: always try to extract the peer key, and fallback
819 		 * for packets missing it.
820 		 * Even OoO DSS packets coming legitly after dropped or
821 		 * reordered MPC will cause fallback, but we don't have other
822 		 * options.
823 		 */
824 		mptcp_get_options(skb, &mp_opt);
825 		if (!(mp_opt.suboptions &
826 		      (OPTION_MPTCP_MPC_SYN | OPTION_MPTCP_MPC_ACK)))
827 			fallback = true;
828 
829 	} else if (subflow_req->mp_join) {
830 		mptcp_get_options(skb, &mp_opt);
831 		if (!(mp_opt.suboptions & OPTION_MPTCP_MPJ_ACK) ||
832 		    !subflow_hmac_valid(req, &mp_opt) ||
833 		    !mptcp_can_accept_new_subflow(subflow_req->msk)) {
834 			SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKMAC);
835 			fallback = true;
836 		}
837 	}
838 
839 create_child:
840 	child = listener->icsk_af_ops->syn_recv_sock(sk, skb, req, dst,
841 						     req_unhash, own_req);
842 
843 	if (child && *own_req) {
844 		struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(child);
845 
846 		tcp_rsk(req)->drop_req = false;
847 
848 		/* we need to fallback on ctx allocation failure and on pre-reqs
849 		 * checking above. In the latter scenario we additionally need
850 		 * to reset the context to non MPTCP status.
851 		 */
852 		if (!ctx || fallback) {
853 			if (fallback_is_fatal) {
854 				subflow_add_reset_reason(skb, MPTCP_RST_EMPTCP);
855 				goto dispose_child;
856 			}
857 			goto fallback;
858 		}
859 
860 		/* ssk inherits options of listener sk */
861 		ctx->setsockopt_seq = listener->setsockopt_seq;
862 
863 		if (ctx->mp_capable) {
864 			ctx->conn = mptcp_sk_clone_init(listener->conn, &mp_opt, child, req);
865 			if (!ctx->conn)
866 				goto fallback;
867 
868 			ctx->subflow_id = 1;
869 			owner = mptcp_sk(ctx->conn);
870 			mptcp_pm_new_connection(owner, child, 1);
871 
872 			/* with OoO packets we can reach here without ingress
873 			 * mpc option
874 			 */
875 			if (mp_opt.suboptions & OPTION_MPTCP_MPC_ACK) {
876 				mptcp_pm_fully_established(owner, child);
877 				ctx->pm_notified = 1;
878 			}
879 		} else if (ctx->mp_join) {
880 			owner = subflow_req->msk;
881 			if (!owner) {
882 				subflow_add_reset_reason(skb, MPTCP_RST_EPROHIBIT);
883 				goto dispose_child;
884 			}
885 
886 			/* move the msk reference ownership to the subflow */
887 			subflow_req->msk = NULL;
888 			ctx->conn = (struct sock *)owner;
889 
890 			if (subflow_use_different_sport(owner, sk)) {
891 				pr_debug("ack inet_sport=%d %d\n",
892 					 ntohs(inet_sk(sk)->inet_sport),
893 					 ntohs(inet_sk((struct sock *)owner)->inet_sport));
894 				if (!mptcp_pm_sport_in_anno_list(owner, sk)) {
895 					SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MISMATCHPORTACKRX);
896 					goto dispose_child;
897 				}
898 				SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINPORTACKRX);
899 			}
900 
901 			if (!mptcp_finish_join(child))
902 				goto dispose_child;
903 
904 			SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKRX);
905 			tcp_rsk(req)->drop_req = true;
906 		}
907 	}
908 
909 	/* check for expected invariant - should never trigger, just help
910 	 * catching eariler subtle bugs
911 	 */
912 	WARN_ON_ONCE(child && *own_req && tcp_sk(child)->is_mptcp &&
913 		     (!mptcp_subflow_ctx(child) ||
914 		      !mptcp_subflow_ctx(child)->conn));
915 	return child;
916 
917 dispose_child:
918 	mptcp_subflow_drop_ctx(child);
919 	tcp_rsk(req)->drop_req = true;
920 	inet_csk_prepare_for_destroy_sock(child);
921 	tcp_done(child);
922 	req->rsk_ops->send_reset(sk, skb);
923 
924 	/* The last child reference will be released by the caller */
925 	return child;
926 
927 fallback:
928 	if (fallback)
929 		SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MPCAPABLEPASSIVEFALLBACK);
930 	mptcp_subflow_drop_ctx(child);
931 	return child;
932 }
933 
934 static struct inet_connection_sock_af_ops subflow_specific __ro_after_init;
935 static struct proto tcp_prot_override __ro_after_init;
936 
937 enum mapping_status {
938 	MAPPING_OK,
939 	MAPPING_INVALID,
940 	MAPPING_EMPTY,
941 	MAPPING_DATA_FIN,
942 	MAPPING_DUMMY,
943 	MAPPING_BAD_CSUM
944 };
945 
dbg_bad_map(struct mptcp_subflow_context * subflow,u32 ssn)946 static void dbg_bad_map(struct mptcp_subflow_context *subflow, u32 ssn)
947 {
948 	pr_debug("Bad mapping: ssn=%d map_seq=%d map_data_len=%d\n",
949 		 ssn, subflow->map_subflow_seq, subflow->map_data_len);
950 }
951 
skb_is_fully_mapped(struct sock * ssk,struct sk_buff * skb)952 static bool skb_is_fully_mapped(struct sock *ssk, struct sk_buff *skb)
953 {
954 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
955 	unsigned int skb_consumed;
956 
957 	skb_consumed = tcp_sk(ssk)->copied_seq - TCP_SKB_CB(skb)->seq;
958 	if (unlikely(skb_consumed >= skb->len)) {
959 		DEBUG_NET_WARN_ON_ONCE(1);
960 		return true;
961 	}
962 
963 	return skb->len - skb_consumed <= subflow->map_data_len -
964 					  mptcp_subflow_get_map_offset(subflow);
965 }
966 
validate_mapping(struct sock * ssk,struct sk_buff * skb)967 static bool validate_mapping(struct sock *ssk, struct sk_buff *skb)
968 {
969 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
970 	u32 ssn = tcp_sk(ssk)->copied_seq - subflow->ssn_offset;
971 
972 	if (unlikely(before(ssn, subflow->map_subflow_seq))) {
973 		/* Mapping covers data later in the subflow stream,
974 		 * currently unsupported.
975 		 */
976 		dbg_bad_map(subflow, ssn);
977 		return false;
978 	}
979 	if (unlikely(!before(ssn, subflow->map_subflow_seq +
980 				  subflow->map_data_len))) {
981 		/* Mapping does covers past subflow data, invalid */
982 		dbg_bad_map(subflow, ssn);
983 		return false;
984 	}
985 	return true;
986 }
987 
validate_data_csum(struct sock * ssk,struct sk_buff * skb,bool csum_reqd)988 static enum mapping_status validate_data_csum(struct sock *ssk, struct sk_buff *skb,
989 					      bool csum_reqd)
990 {
991 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
992 	u32 offset, seq, delta;
993 	__sum16 csum;
994 	int len;
995 
996 	if (!csum_reqd)
997 		return MAPPING_OK;
998 
999 	/* mapping already validated on previous traversal */
1000 	if (subflow->map_csum_len == subflow->map_data_len)
1001 		return MAPPING_OK;
1002 
1003 	/* traverse the receive queue, ensuring it contains a full
1004 	 * DSS mapping and accumulating the related csum.
1005 	 * Preserve the accoumlate csum across multiple calls, to compute
1006 	 * the csum only once
1007 	 */
1008 	delta = subflow->map_data_len - subflow->map_csum_len;
1009 	for (;;) {
1010 		seq = tcp_sk(ssk)->copied_seq + subflow->map_csum_len;
1011 		offset = seq - TCP_SKB_CB(skb)->seq;
1012 
1013 		/* if the current skb has not been accounted yet, csum its contents
1014 		 * up to the amount covered by the current DSS
1015 		 */
1016 		if (offset < skb->len) {
1017 			__wsum csum;
1018 
1019 			len = min(skb->len - offset, delta);
1020 			csum = skb_checksum(skb, offset, len, 0);
1021 			subflow->map_data_csum = csum_block_add(subflow->map_data_csum, csum,
1022 								subflow->map_csum_len);
1023 
1024 			delta -= len;
1025 			subflow->map_csum_len += len;
1026 		}
1027 		if (delta == 0)
1028 			break;
1029 
1030 		if (skb_queue_is_last(&ssk->sk_receive_queue, skb)) {
1031 			/* if this subflow is closed, the partial mapping
1032 			 * will be never completed; flush the pending skbs, so
1033 			 * that subflow_sched_work_if_closed() can kick in
1034 			 */
1035 			if (unlikely(ssk->sk_state == TCP_CLOSE))
1036 				while ((skb = skb_peek(&ssk->sk_receive_queue)))
1037 					sk_eat_skb(ssk, skb);
1038 
1039 			/* not enough data to validate the csum */
1040 			return MAPPING_EMPTY;
1041 		}
1042 
1043 		/* the DSS mapping for next skbs will be validated later,
1044 		 * when a get_mapping_status call will process such skb
1045 		 */
1046 		skb = skb->next;
1047 	}
1048 
1049 	/* note that 'map_data_len' accounts only for the carried data, does
1050 	 * not include the eventual seq increment due to the data fin,
1051 	 * while the pseudo header requires the original DSS data len,
1052 	 * including that
1053 	 */
1054 	csum = __mptcp_make_csum(subflow->map_seq,
1055 				 subflow->map_subflow_seq,
1056 				 subflow->map_data_len + subflow->map_data_fin,
1057 				 subflow->map_data_csum);
1058 	if (unlikely(csum)) {
1059 		MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DATACSUMERR);
1060 		return MAPPING_BAD_CSUM;
1061 	}
1062 
1063 	subflow->valid_csum_seen = 1;
1064 	return MAPPING_OK;
1065 }
1066 
get_mapping_status(struct sock * ssk,struct mptcp_sock * msk)1067 static enum mapping_status get_mapping_status(struct sock *ssk,
1068 					      struct mptcp_sock *msk)
1069 {
1070 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1071 	bool csum_reqd = READ_ONCE(msk->csum_enabled);
1072 	struct mptcp_ext *mpext;
1073 	struct sk_buff *skb;
1074 	u16 data_len;
1075 	u64 map_seq;
1076 
1077 	skb = skb_peek(&ssk->sk_receive_queue);
1078 	if (!skb)
1079 		return MAPPING_EMPTY;
1080 
1081 	if (mptcp_check_fallback(ssk))
1082 		return MAPPING_DUMMY;
1083 
1084 	mpext = mptcp_get_ext(skb);
1085 	if (!mpext || !mpext->use_map) {
1086 		if (!subflow->map_valid && !skb->len) {
1087 			/* the TCP stack deliver 0 len FIN pkt to the receive
1088 			 * queue, that is the only 0len pkts ever expected here,
1089 			 * and we can admit no mapping only for 0 len pkts
1090 			 */
1091 			if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN))
1092 				WARN_ONCE(1, "0len seq %d:%d flags %x",
1093 					  TCP_SKB_CB(skb)->seq,
1094 					  TCP_SKB_CB(skb)->end_seq,
1095 					  TCP_SKB_CB(skb)->tcp_flags);
1096 			sk_eat_skb(ssk, skb);
1097 			return MAPPING_EMPTY;
1098 		}
1099 
1100 		if (!subflow->map_valid)
1101 			return MAPPING_INVALID;
1102 
1103 		goto validate_seq;
1104 	}
1105 
1106 	trace_get_mapping_status(mpext);
1107 
1108 	data_len = mpext->data_len;
1109 	if (data_len == 0) {
1110 		pr_debug("infinite mapping received\n");
1111 		MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_INFINITEMAPRX);
1112 		return MAPPING_INVALID;
1113 	}
1114 
1115 	if (mpext->data_fin == 1) {
1116 		if (data_len == 1) {
1117 			bool updated = mptcp_update_rcv_data_fin(msk, mpext->data_seq,
1118 								 mpext->dsn64);
1119 			pr_debug("DATA_FIN with no payload seq=%llu\n", mpext->data_seq);
1120 			if (subflow->map_valid) {
1121 				/* A DATA_FIN might arrive in a DSS
1122 				 * option before the previous mapping
1123 				 * has been fully consumed. Continue
1124 				 * handling the existing mapping.
1125 				 */
1126 				skb_ext_del(skb, SKB_EXT_MPTCP);
1127 				return MAPPING_OK;
1128 			} else {
1129 				if (updated)
1130 					mptcp_schedule_work((struct sock *)msk);
1131 
1132 				return MAPPING_DATA_FIN;
1133 			}
1134 		} else {
1135 			u64 data_fin_seq = mpext->data_seq + data_len - 1;
1136 
1137 			/* If mpext->data_seq is a 32-bit value, data_fin_seq
1138 			 * must also be limited to 32 bits.
1139 			 */
1140 			if (!mpext->dsn64)
1141 				data_fin_seq &= GENMASK_ULL(31, 0);
1142 
1143 			mptcp_update_rcv_data_fin(msk, data_fin_seq, mpext->dsn64);
1144 			pr_debug("DATA_FIN with mapping seq=%llu dsn64=%d\n",
1145 				 data_fin_seq, mpext->dsn64);
1146 		}
1147 
1148 		/* Adjust for DATA_FIN using 1 byte of sequence space */
1149 		data_len--;
1150 	}
1151 
1152 	map_seq = mptcp_expand_seq(READ_ONCE(msk->ack_seq), mpext->data_seq, mpext->dsn64);
1153 	WRITE_ONCE(mptcp_sk(subflow->conn)->use_64bit_ack, !!mpext->dsn64);
1154 
1155 	if (subflow->map_valid) {
1156 		/* Allow replacing only with an identical map */
1157 		if (subflow->map_seq == map_seq &&
1158 		    subflow->map_subflow_seq == mpext->subflow_seq &&
1159 		    subflow->map_data_len == data_len &&
1160 		    subflow->map_csum_reqd == mpext->csum_reqd) {
1161 			skb_ext_del(skb, SKB_EXT_MPTCP);
1162 			goto validate_csum;
1163 		}
1164 
1165 		/* If this skb data are fully covered by the current mapping,
1166 		 * the new map would need caching, which is not supported
1167 		 */
1168 		if (skb_is_fully_mapped(ssk, skb)) {
1169 			MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DSSNOMATCH);
1170 			return MAPPING_INVALID;
1171 		}
1172 
1173 		/* will validate the next map after consuming the current one */
1174 		goto validate_csum;
1175 	}
1176 
1177 	subflow->map_seq = map_seq;
1178 	subflow->map_subflow_seq = mpext->subflow_seq;
1179 	subflow->map_data_len = data_len;
1180 	subflow->map_valid = 1;
1181 	subflow->map_data_fin = mpext->data_fin;
1182 	subflow->mpc_map = mpext->mpc_map;
1183 	subflow->map_csum_reqd = mpext->csum_reqd;
1184 	subflow->map_csum_len = 0;
1185 	subflow->map_data_csum = csum_unfold(mpext->csum);
1186 
1187 	/* Cfr RFC 8684 Section 3.3.0 */
1188 	if (unlikely(subflow->map_csum_reqd != csum_reqd))
1189 		return MAPPING_INVALID;
1190 
1191 	pr_debug("new map seq=%llu subflow_seq=%u data_len=%u csum=%d:%u\n",
1192 		 subflow->map_seq, subflow->map_subflow_seq,
1193 		 subflow->map_data_len, subflow->map_csum_reqd,
1194 		 subflow->map_data_csum);
1195 
1196 validate_seq:
1197 	/* we revalidate valid mapping on new skb, because we must ensure
1198 	 * the current skb is completely covered by the available mapping
1199 	 */
1200 	if (!validate_mapping(ssk, skb)) {
1201 		MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DSSTCPMISMATCH);
1202 		return MAPPING_INVALID;
1203 	}
1204 
1205 	skb_ext_del(skb, SKB_EXT_MPTCP);
1206 
1207 validate_csum:
1208 	return validate_data_csum(ssk, skb, csum_reqd);
1209 }
1210 
mptcp_subflow_discard_data(struct sock * ssk,struct sk_buff * skb,u64 limit)1211 static void mptcp_subflow_discard_data(struct sock *ssk, struct sk_buff *skb,
1212 				       u64 limit)
1213 {
1214 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1215 	bool fin = TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN;
1216 	struct tcp_sock *tp = tcp_sk(ssk);
1217 	u32 offset, incr, avail_len;
1218 
1219 	offset = tp->copied_seq - TCP_SKB_CB(skb)->seq;
1220 	if (WARN_ON_ONCE(offset > skb->len))
1221 		goto out;
1222 
1223 	avail_len = skb->len - offset;
1224 	incr = limit >= avail_len ? avail_len + fin : limit;
1225 
1226 	pr_debug("discarding=%d len=%d offset=%d seq=%d\n", incr, skb->len,
1227 		 offset, subflow->map_subflow_seq);
1228 	MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DUPDATA);
1229 	tcp_sk(ssk)->copied_seq += incr;
1230 
1231 out:
1232 	if (!before(tcp_sk(ssk)->copied_seq, TCP_SKB_CB(skb)->end_seq))
1233 		sk_eat_skb(ssk, skb);
1234 	if (mptcp_subflow_get_map_offset(subflow) >= subflow->map_data_len)
1235 		subflow->map_valid = 0;
1236 }
1237 
1238 /* sched mptcp worker to remove the subflow if no more data is pending */
subflow_sched_work_if_closed(struct mptcp_sock * msk,struct sock * ssk)1239 static void subflow_sched_work_if_closed(struct mptcp_sock *msk, struct sock *ssk)
1240 {
1241 	struct sock *sk = (struct sock *)msk;
1242 
1243 	if (likely(ssk->sk_state != TCP_CLOSE &&
1244 		   (ssk->sk_state != TCP_CLOSE_WAIT ||
1245 		    inet_sk_state_load(sk) != TCP_ESTABLISHED)))
1246 		return;
1247 
1248 	if (skb_queue_empty(&ssk->sk_receive_queue) &&
1249 	    !test_and_set_bit(MPTCP_WORK_CLOSE_SUBFLOW, &msk->flags))
1250 		mptcp_schedule_work(sk);
1251 }
1252 
mptcp_subflow_fail(struct mptcp_sock * msk,struct sock * ssk)1253 static void mptcp_subflow_fail(struct mptcp_sock *msk, struct sock *ssk)
1254 {
1255 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1256 	unsigned long fail_tout;
1257 
1258 	/* greceful failure can happen only on the MPC subflow */
1259 	if (WARN_ON_ONCE(ssk != READ_ONCE(msk->first)))
1260 		return;
1261 
1262 	/* since the close timeout take precedence on the fail one,
1263 	 * no need to start the latter when the first is already set
1264 	 */
1265 	if (sock_flag((struct sock *)msk, SOCK_DEAD))
1266 		return;
1267 
1268 	/* we don't need extreme accuracy here, use a zero fail_tout as special
1269 	 * value meaning no fail timeout at all;
1270 	 */
1271 	fail_tout = jiffies + TCP_RTO_MAX;
1272 	if (!fail_tout)
1273 		fail_tout = 1;
1274 	WRITE_ONCE(subflow->fail_tout, fail_tout);
1275 	tcp_send_ack(ssk);
1276 
1277 	mptcp_reset_tout_timer(msk, subflow->fail_tout);
1278 }
1279 
subflow_check_data_avail(struct sock * ssk)1280 static bool subflow_check_data_avail(struct sock *ssk)
1281 {
1282 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1283 	enum mapping_status status;
1284 	struct mptcp_sock *msk;
1285 	struct sk_buff *skb;
1286 
1287 	if (!skb_peek(&ssk->sk_receive_queue))
1288 		WRITE_ONCE(subflow->data_avail, MPTCP_SUBFLOW_NODATA);
1289 	if (subflow->data_avail)
1290 		return true;
1291 
1292 	msk = mptcp_sk(subflow->conn);
1293 	for (;;) {
1294 		u64 ack_seq;
1295 		u64 old_ack;
1296 
1297 		status = get_mapping_status(ssk, msk);
1298 		trace_subflow_check_data_avail(status, skb_peek(&ssk->sk_receive_queue));
1299 		if (unlikely(status == MAPPING_INVALID || status == MAPPING_DUMMY ||
1300 			     status == MAPPING_BAD_CSUM))
1301 			goto fallback;
1302 
1303 		if (status != MAPPING_OK)
1304 			goto no_data;
1305 
1306 		skb = skb_peek(&ssk->sk_receive_queue);
1307 		if (WARN_ON_ONCE(!skb))
1308 			goto no_data;
1309 
1310 		if (unlikely(!READ_ONCE(msk->can_ack)))
1311 			goto fallback;
1312 
1313 		old_ack = READ_ONCE(msk->ack_seq);
1314 		ack_seq = mptcp_subflow_get_mapped_dsn(subflow);
1315 		pr_debug("msk ack_seq=%llx subflow ack_seq=%llx\n", old_ack,
1316 			 ack_seq);
1317 		if (unlikely(before64(ack_seq, old_ack))) {
1318 			mptcp_subflow_discard_data(ssk, skb, old_ack - ack_seq);
1319 			continue;
1320 		}
1321 
1322 		WRITE_ONCE(subflow->data_avail, MPTCP_SUBFLOW_DATA_AVAIL);
1323 		break;
1324 	}
1325 	return true;
1326 
1327 no_data:
1328 	subflow_sched_work_if_closed(msk, ssk);
1329 	return false;
1330 
1331 fallback:
1332 	if (!__mptcp_check_fallback(msk)) {
1333 		/* RFC 8684 section 3.7. */
1334 		if (status == MAPPING_BAD_CSUM &&
1335 		    (subflow->mp_join || subflow->valid_csum_seen)) {
1336 			subflow->send_mp_fail = 1;
1337 
1338 			if (!READ_ONCE(msk->allow_infinite_fallback)) {
1339 				subflow->reset_transient = 0;
1340 				subflow->reset_reason = MPTCP_RST_EMIDDLEBOX;
1341 				goto reset;
1342 			}
1343 			mptcp_subflow_fail(msk, ssk);
1344 			WRITE_ONCE(subflow->data_avail, MPTCP_SUBFLOW_DATA_AVAIL);
1345 			return true;
1346 		}
1347 
1348 		if (!READ_ONCE(msk->allow_infinite_fallback)) {
1349 			/* fatal protocol error, close the socket.
1350 			 * subflow_error_report() will introduce the appropriate barriers
1351 			 */
1352 			subflow->reset_transient = 0;
1353 			subflow->reset_reason = MPTCP_RST_EMPTCP;
1354 
1355 reset:
1356 			WRITE_ONCE(ssk->sk_err, EBADMSG);
1357 			tcp_set_state(ssk, TCP_CLOSE);
1358 			while ((skb = skb_peek(&ssk->sk_receive_queue)))
1359 				sk_eat_skb(ssk, skb);
1360 			tcp_send_active_reset(ssk, GFP_ATOMIC);
1361 			WRITE_ONCE(subflow->data_avail, MPTCP_SUBFLOW_NODATA);
1362 			return false;
1363 		}
1364 
1365 		mptcp_do_fallback(ssk);
1366 	}
1367 
1368 	skb = skb_peek(&ssk->sk_receive_queue);
1369 	subflow->map_valid = 1;
1370 	subflow->map_seq = READ_ONCE(msk->ack_seq);
1371 	subflow->map_data_len = skb->len;
1372 	subflow->map_subflow_seq = tcp_sk(ssk)->copied_seq - subflow->ssn_offset;
1373 	WRITE_ONCE(subflow->data_avail, MPTCP_SUBFLOW_DATA_AVAIL);
1374 	return true;
1375 }
1376 
mptcp_subflow_data_available(struct sock * sk)1377 bool mptcp_subflow_data_available(struct sock *sk)
1378 {
1379 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1380 
1381 	/* check if current mapping is still valid */
1382 	if (subflow->map_valid &&
1383 	    mptcp_subflow_get_map_offset(subflow) >= subflow->map_data_len) {
1384 		subflow->map_valid = 0;
1385 		WRITE_ONCE(subflow->data_avail, MPTCP_SUBFLOW_NODATA);
1386 
1387 		pr_debug("Done with mapping: seq=%u data_len=%u\n",
1388 			 subflow->map_subflow_seq,
1389 			 subflow->map_data_len);
1390 	}
1391 
1392 	return subflow_check_data_avail(sk);
1393 }
1394 
1395 /* If ssk has an mptcp parent socket, use the mptcp rcvbuf occupancy,
1396  * not the ssk one.
1397  *
1398  * In mptcp, rwin is about the mptcp-level connection data.
1399  *
1400  * Data that is still on the ssk rx queue can thus be ignored,
1401  * as far as mptcp peer is concerned that data is still inflight.
1402  * DSS ACK is updated when skb is moved to the mptcp rx queue.
1403  */
mptcp_space(const struct sock * ssk,int * space,int * full_space)1404 void mptcp_space(const struct sock *ssk, int *space, int *full_space)
1405 {
1406 	const struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1407 	const struct sock *sk = subflow->conn;
1408 
1409 	*space = __mptcp_space(sk);
1410 	*full_space = mptcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf));
1411 }
1412 
subflow_error_report(struct sock * ssk)1413 static void subflow_error_report(struct sock *ssk)
1414 {
1415 	struct sock *sk = mptcp_subflow_ctx(ssk)->conn;
1416 
1417 	/* bail early if this is a no-op, so that we avoid introducing a
1418 	 * problematic lockdep dependency between TCP accept queue lock
1419 	 * and msk socket spinlock
1420 	 */
1421 	if (!sk->sk_socket)
1422 		return;
1423 
1424 	mptcp_data_lock(sk);
1425 	if (!sock_owned_by_user(sk))
1426 		__mptcp_error_report(sk);
1427 	else
1428 		__set_bit(MPTCP_ERROR_REPORT,  &mptcp_sk(sk)->cb_flags);
1429 	mptcp_data_unlock(sk);
1430 }
1431 
subflow_data_ready(struct sock * sk)1432 static void subflow_data_ready(struct sock *sk)
1433 {
1434 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1435 	u16 state = 1 << inet_sk_state_load(sk);
1436 	struct sock *parent = subflow->conn;
1437 	struct mptcp_sock *msk;
1438 
1439 	trace_sk_data_ready(sk);
1440 
1441 	msk = mptcp_sk(parent);
1442 	if (state & TCPF_LISTEN) {
1443 		/* MPJ subflow are removed from accept queue before reaching here,
1444 		 * avoid stray wakeups
1445 		 */
1446 		if (reqsk_queue_empty(&inet_csk(sk)->icsk_accept_queue))
1447 			return;
1448 
1449 		parent->sk_data_ready(parent);
1450 		return;
1451 	}
1452 
1453 	WARN_ON_ONCE(!__mptcp_check_fallback(msk) && !subflow->mp_capable &&
1454 		     !subflow->mp_join && !(state & TCPF_CLOSE));
1455 
1456 	if (mptcp_subflow_data_available(sk)) {
1457 		mptcp_data_ready(parent, sk);
1458 
1459 		/* subflow-level lowat test are not relevant.
1460 		 * respect the msk-level threshold eventually mandating an immediate ack
1461 		 */
1462 		if (mptcp_data_avail(msk) < parent->sk_rcvlowat &&
1463 		    (tcp_sk(sk)->rcv_nxt - tcp_sk(sk)->rcv_wup) > inet_csk(sk)->icsk_ack.rcv_mss)
1464 			inet_csk(sk)->icsk_ack.pending |= ICSK_ACK_NOW;
1465 	} else if (unlikely(sk->sk_err)) {
1466 		subflow_error_report(sk);
1467 	}
1468 }
1469 
subflow_write_space(struct sock * ssk)1470 static void subflow_write_space(struct sock *ssk)
1471 {
1472 	struct sock *sk = mptcp_subflow_ctx(ssk)->conn;
1473 
1474 	mptcp_propagate_sndbuf(sk, ssk);
1475 	mptcp_write_space(sk);
1476 }
1477 
1478 static const struct inet_connection_sock_af_ops *
subflow_default_af_ops(struct sock * sk)1479 subflow_default_af_ops(struct sock *sk)
1480 {
1481 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1482 	if (sk->sk_family == AF_INET6)
1483 		return &subflow_v6_specific;
1484 #endif
1485 	return &subflow_specific;
1486 }
1487 
1488 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
mptcpv6_handle_mapped(struct sock * sk,bool mapped)1489 void mptcpv6_handle_mapped(struct sock *sk, bool mapped)
1490 {
1491 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1492 	struct inet_connection_sock *icsk = inet_csk(sk);
1493 	const struct inet_connection_sock_af_ops *target;
1494 
1495 	target = mapped ? &subflow_v6m_specific : subflow_default_af_ops(sk);
1496 
1497 	pr_debug("subflow=%p family=%d ops=%p target=%p mapped=%d\n",
1498 		 subflow, sk->sk_family, icsk->icsk_af_ops, target, mapped);
1499 
1500 	if (likely(icsk->icsk_af_ops == target))
1501 		return;
1502 
1503 	subflow->icsk_af_ops = icsk->icsk_af_ops;
1504 	icsk->icsk_af_ops = target;
1505 }
1506 #endif
1507 
mptcp_info2sockaddr(const struct mptcp_addr_info * info,struct sockaddr_storage * addr,unsigned short family)1508 void mptcp_info2sockaddr(const struct mptcp_addr_info *info,
1509 			 struct sockaddr_storage *addr,
1510 			 unsigned short family)
1511 {
1512 	memset(addr, 0, sizeof(*addr));
1513 	addr->ss_family = family;
1514 	if (addr->ss_family == AF_INET) {
1515 		struct sockaddr_in *in_addr = (struct sockaddr_in *)addr;
1516 
1517 		if (info->family == AF_INET)
1518 			in_addr->sin_addr = info->addr;
1519 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1520 		else if (ipv6_addr_v4mapped(&info->addr6))
1521 			in_addr->sin_addr.s_addr = info->addr6.s6_addr32[3];
1522 #endif
1523 		in_addr->sin_port = info->port;
1524 	}
1525 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1526 	else if (addr->ss_family == AF_INET6) {
1527 		struct sockaddr_in6 *in6_addr = (struct sockaddr_in6 *)addr;
1528 
1529 		if (info->family == AF_INET)
1530 			ipv6_addr_set_v4mapped(info->addr.s_addr,
1531 					       &in6_addr->sin6_addr);
1532 		else
1533 			in6_addr->sin6_addr = info->addr6;
1534 		in6_addr->sin6_port = info->port;
1535 	}
1536 #endif
1537 }
1538 
__mptcp_subflow_connect(struct sock * sk,const struct mptcp_addr_info * loc,const struct mptcp_addr_info * remote)1539 int __mptcp_subflow_connect(struct sock *sk, const struct mptcp_addr_info *loc,
1540 			    const struct mptcp_addr_info *remote)
1541 {
1542 	struct mptcp_sock *msk = mptcp_sk(sk);
1543 	struct mptcp_subflow_context *subflow;
1544 	struct sockaddr_storage addr;
1545 	int remote_id = remote->id;
1546 	int local_id = loc->id;
1547 	int err = -ENOTCONN;
1548 	struct socket *sf;
1549 	struct sock *ssk;
1550 	u32 remote_token;
1551 	int addrlen;
1552 	int ifindex;
1553 	u8 flags;
1554 
1555 	if (!mptcp_is_fully_established(sk))
1556 		goto err_out;
1557 
1558 	err = mptcp_subflow_create_socket(sk, loc->family, &sf);
1559 	if (err)
1560 		goto err_out;
1561 
1562 	ssk = sf->sk;
1563 	subflow = mptcp_subflow_ctx(ssk);
1564 	do {
1565 		get_random_bytes(&subflow->local_nonce, sizeof(u32));
1566 	} while (!subflow->local_nonce);
1567 
1568 	if (local_id)
1569 		subflow_set_local_id(subflow, local_id);
1570 
1571 	mptcp_pm_get_flags_and_ifindex_by_id(msk, local_id,
1572 					     &flags, &ifindex);
1573 	subflow->remote_key_valid = 1;
1574 	subflow->remote_key = msk->remote_key;
1575 	subflow->local_key = msk->local_key;
1576 	subflow->token = msk->token;
1577 	mptcp_info2sockaddr(loc, &addr, ssk->sk_family);
1578 
1579 	addrlen = sizeof(struct sockaddr_in);
1580 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1581 	if (addr.ss_family == AF_INET6)
1582 		addrlen = sizeof(struct sockaddr_in6);
1583 #endif
1584 	mptcp_sockopt_sync(msk, ssk);
1585 
1586 	ssk->sk_bound_dev_if = ifindex;
1587 	err = kernel_bind(sf, (struct sockaddr *)&addr, addrlen);
1588 	if (err)
1589 		goto failed;
1590 
1591 	mptcp_crypto_key_sha(subflow->remote_key, &remote_token, NULL);
1592 	pr_debug("msk=%p remote_token=%u local_id=%d remote_id=%d\n", msk,
1593 		 remote_token, local_id, remote_id);
1594 	subflow->remote_token = remote_token;
1595 	WRITE_ONCE(subflow->remote_id, remote_id);
1596 	subflow->request_join = 1;
1597 	subflow->request_bkup = !!(flags & MPTCP_PM_ADDR_FLAG_BACKUP);
1598 	subflow->subflow_id = msk->subflow_id++;
1599 	mptcp_info2sockaddr(remote, &addr, ssk->sk_family);
1600 
1601 	sock_hold(ssk);
1602 	list_add_tail(&subflow->node, &msk->conn_list);
1603 	err = kernel_connect(sf, (struct sockaddr *)&addr, addrlen, O_NONBLOCK);
1604 	if (err && err != -EINPROGRESS)
1605 		goto failed_unlink;
1606 
1607 	/* discard the subflow socket */
1608 	mptcp_sock_graft(ssk, sk->sk_socket);
1609 	iput(SOCK_INODE(sf));
1610 	WRITE_ONCE(msk->allow_infinite_fallback, false);
1611 	mptcp_stop_tout_timer(sk);
1612 	return 0;
1613 
1614 failed_unlink:
1615 	list_del(&subflow->node);
1616 	sock_put(mptcp_subflow_tcp_sock(subflow));
1617 
1618 failed:
1619 	subflow->disposable = 1;
1620 	sock_release(sf);
1621 
1622 err_out:
1623 	/* we account subflows before the creation, and this failures will not
1624 	 * be caught by sk_state_change()
1625 	 */
1626 	mptcp_pm_close_subflow(msk);
1627 	return err;
1628 }
1629 
mptcp_attach_cgroup(struct sock * parent,struct sock * child)1630 static void mptcp_attach_cgroup(struct sock *parent, struct sock *child)
1631 {
1632 #ifdef CONFIG_SOCK_CGROUP_DATA
1633 	struct sock_cgroup_data *parent_skcd = &parent->sk_cgrp_data,
1634 				*child_skcd = &child->sk_cgrp_data;
1635 
1636 	/* only the additional subflows created by kworkers have to be modified */
1637 	if (cgroup_id(sock_cgroup_ptr(parent_skcd)) !=
1638 	    cgroup_id(sock_cgroup_ptr(child_skcd))) {
1639 #ifdef CONFIG_MEMCG
1640 		struct mem_cgroup *memcg = parent->sk_memcg;
1641 
1642 		mem_cgroup_sk_free(child);
1643 		if (memcg && css_tryget(&memcg->css))
1644 			child->sk_memcg = memcg;
1645 #endif /* CONFIG_MEMCG */
1646 
1647 		cgroup_sk_free(child_skcd);
1648 		*child_skcd = *parent_skcd;
1649 		cgroup_sk_clone(child_skcd);
1650 	}
1651 #endif /* CONFIG_SOCK_CGROUP_DATA */
1652 }
1653 
mptcp_subflow_ops_override(struct sock * ssk)1654 static void mptcp_subflow_ops_override(struct sock *ssk)
1655 {
1656 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1657 	if (ssk->sk_prot == &tcpv6_prot)
1658 		ssk->sk_prot = &tcpv6_prot_override;
1659 	else
1660 #endif
1661 		ssk->sk_prot = &tcp_prot_override;
1662 }
1663 
mptcp_subflow_ops_undo_override(struct sock * ssk)1664 static void mptcp_subflow_ops_undo_override(struct sock *ssk)
1665 {
1666 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1667 	if (ssk->sk_prot == &tcpv6_prot_override)
1668 		ssk->sk_prot = &tcpv6_prot;
1669 	else
1670 #endif
1671 		ssk->sk_prot = &tcp_prot;
1672 }
1673 
mptcp_subflow_create_socket(struct sock * sk,unsigned short family,struct socket ** new_sock)1674 int mptcp_subflow_create_socket(struct sock *sk, unsigned short family,
1675 				struct socket **new_sock)
1676 {
1677 	struct mptcp_subflow_context *subflow;
1678 	struct net *net = sock_net(sk);
1679 	struct socket *sf;
1680 	int err;
1681 
1682 	/* un-accepted server sockets can reach here - on bad configuration
1683 	 * bail early to avoid greater trouble later
1684 	 */
1685 	if (unlikely(!sk->sk_socket))
1686 		return -EINVAL;
1687 
1688 	err = sock_create_kern(net, family, SOCK_STREAM, IPPROTO_TCP, &sf);
1689 	if (err)
1690 		return err;
1691 
1692 	lock_sock_nested(sf->sk, SINGLE_DEPTH_NESTING);
1693 
1694 	err = security_mptcp_add_subflow(sk, sf->sk);
1695 	if (err)
1696 		goto release_ssk;
1697 
1698 	/* the newly created socket has to be in the same cgroup as its parent */
1699 	mptcp_attach_cgroup(sk, sf->sk);
1700 
1701 	/* kernel sockets do not by default acquire net ref, but TCP timer
1702 	 * needs it.
1703 	 * Update ns_tracker to current stack trace and refcounted tracker.
1704 	 */
1705 	__netns_tracker_free(net, &sf->sk->ns_tracker, false);
1706 	sf->sk->sk_net_refcnt = 1;
1707 	get_net_track(net, &sf->sk->ns_tracker, GFP_KERNEL);
1708 	sock_inuse_add(net, 1);
1709 	err = tcp_set_ulp(sf->sk, "mptcp");
1710 
1711 release_ssk:
1712 	release_sock(sf->sk);
1713 
1714 	if (err) {
1715 		sock_release(sf);
1716 		return err;
1717 	}
1718 
1719 	/* the newly created socket really belongs to the owning MPTCP master
1720 	 * socket, even if for additional subflows the allocation is performed
1721 	 * by a kernel workqueue. Adjust inode references, so that the
1722 	 * procfs/diag interfaces really show this one belonging to the correct
1723 	 * user.
1724 	 */
1725 	SOCK_INODE(sf)->i_ino = SOCK_INODE(sk->sk_socket)->i_ino;
1726 	SOCK_INODE(sf)->i_uid = SOCK_INODE(sk->sk_socket)->i_uid;
1727 	SOCK_INODE(sf)->i_gid = SOCK_INODE(sk->sk_socket)->i_gid;
1728 
1729 	subflow = mptcp_subflow_ctx(sf->sk);
1730 	pr_debug("subflow=%p\n", subflow);
1731 
1732 	*new_sock = sf;
1733 	sock_hold(sk);
1734 	subflow->conn = sk;
1735 	mptcp_subflow_ops_override(sf->sk);
1736 
1737 	return 0;
1738 }
1739 
subflow_create_ctx(struct sock * sk,gfp_t priority)1740 static struct mptcp_subflow_context *subflow_create_ctx(struct sock *sk,
1741 							gfp_t priority)
1742 {
1743 	struct inet_connection_sock *icsk = inet_csk(sk);
1744 	struct mptcp_subflow_context *ctx;
1745 
1746 	ctx = kzalloc(sizeof(*ctx), priority);
1747 	if (!ctx)
1748 		return NULL;
1749 
1750 	rcu_assign_pointer(icsk->icsk_ulp_data, ctx);
1751 	INIT_LIST_HEAD(&ctx->node);
1752 	INIT_LIST_HEAD(&ctx->delegated_node);
1753 
1754 	pr_debug("subflow=%p\n", ctx);
1755 
1756 	ctx->tcp_sock = sk;
1757 	WRITE_ONCE(ctx->local_id, -1);
1758 
1759 	return ctx;
1760 }
1761 
__subflow_state_change(struct sock * sk)1762 static void __subflow_state_change(struct sock *sk)
1763 {
1764 	struct socket_wq *wq;
1765 
1766 	rcu_read_lock();
1767 	wq = rcu_dereference(sk->sk_wq);
1768 	if (skwq_has_sleeper(wq))
1769 		wake_up_interruptible_all(&wq->wait);
1770 	rcu_read_unlock();
1771 }
1772 
subflow_is_done(const struct sock * sk)1773 static bool subflow_is_done(const struct sock *sk)
1774 {
1775 	return sk->sk_shutdown & RCV_SHUTDOWN || sk->sk_state == TCP_CLOSE;
1776 }
1777 
subflow_state_change(struct sock * sk)1778 static void subflow_state_change(struct sock *sk)
1779 {
1780 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1781 	struct sock *parent = subflow->conn;
1782 	struct mptcp_sock *msk;
1783 
1784 	__subflow_state_change(sk);
1785 
1786 	msk = mptcp_sk(parent);
1787 	if (subflow_simultaneous_connect(sk)) {
1788 		mptcp_do_fallback(sk);
1789 		pr_fallback(msk);
1790 		subflow->conn_finished = 1;
1791 		mptcp_propagate_state(parent, sk, subflow, NULL);
1792 	}
1793 
1794 	/* as recvmsg() does not acquire the subflow socket for ssk selection
1795 	 * a fin packet carrying a DSS can be unnoticed if we don't trigger
1796 	 * the data available machinery here.
1797 	 */
1798 	if (mptcp_subflow_data_available(sk))
1799 		mptcp_data_ready(parent, sk);
1800 	else if (unlikely(sk->sk_err))
1801 		subflow_error_report(sk);
1802 
1803 	subflow_sched_work_if_closed(mptcp_sk(parent), sk);
1804 
1805 	/* when the fallback subflow closes the rx side, trigger a 'dummy'
1806 	 * ingress data fin, so that the msk state will follow along
1807 	 */
1808 	if (__mptcp_check_fallback(msk) && subflow_is_done(sk) && msk->first == sk &&
1809 	    mptcp_update_rcv_data_fin(msk, READ_ONCE(msk->ack_seq), true))
1810 		mptcp_schedule_work(parent);
1811 }
1812 
mptcp_subflow_queue_clean(struct sock * listener_sk,struct sock * listener_ssk)1813 void mptcp_subflow_queue_clean(struct sock *listener_sk, struct sock *listener_ssk)
1814 {
1815 	struct request_sock_queue *queue = &inet_csk(listener_ssk)->icsk_accept_queue;
1816 	struct request_sock *req, *head, *tail;
1817 	struct mptcp_subflow_context *subflow;
1818 	struct sock *sk, *ssk;
1819 
1820 	/* Due to lock dependencies no relevant lock can be acquired under rskq_lock.
1821 	 * Splice the req list, so that accept() can not reach the pending ssk after
1822 	 * the listener socket is released below.
1823 	 */
1824 	spin_lock_bh(&queue->rskq_lock);
1825 	head = queue->rskq_accept_head;
1826 	tail = queue->rskq_accept_tail;
1827 	queue->rskq_accept_head = NULL;
1828 	queue->rskq_accept_tail = NULL;
1829 	spin_unlock_bh(&queue->rskq_lock);
1830 
1831 	if (!head)
1832 		return;
1833 
1834 	/* can't acquire the msk socket lock under the subflow one,
1835 	 * or will cause ABBA deadlock
1836 	 */
1837 	release_sock(listener_ssk);
1838 
1839 	for (req = head; req; req = req->dl_next) {
1840 		ssk = req->sk;
1841 		if (!sk_is_mptcp(ssk))
1842 			continue;
1843 
1844 		subflow = mptcp_subflow_ctx(ssk);
1845 		if (!subflow || !subflow->conn)
1846 			continue;
1847 
1848 		sk = subflow->conn;
1849 		sock_hold(sk);
1850 
1851 		lock_sock_nested(sk, SINGLE_DEPTH_NESTING);
1852 		__mptcp_unaccepted_force_close(sk);
1853 		release_sock(sk);
1854 
1855 		/* lockdep will report a false positive ABBA deadlock
1856 		 * between cancel_work_sync and the listener socket.
1857 		 * The involved locks belong to different sockets WRT
1858 		 * the existing AB chain.
1859 		 * Using a per socket key is problematic as key
1860 		 * deregistration requires process context and must be
1861 		 * performed at socket disposal time, in atomic
1862 		 * context.
1863 		 * Just tell lockdep to consider the listener socket
1864 		 * released here.
1865 		 */
1866 		mutex_release(&listener_sk->sk_lock.dep_map, _RET_IP_);
1867 		mptcp_cancel_work(sk);
1868 		mutex_acquire(&listener_sk->sk_lock.dep_map, 0, 0, _RET_IP_);
1869 
1870 		sock_put(sk);
1871 	}
1872 
1873 	/* we are still under the listener msk socket lock */
1874 	lock_sock_nested(listener_ssk, SINGLE_DEPTH_NESTING);
1875 
1876 	/* restore the listener queue, to let the TCP code clean it up */
1877 	spin_lock_bh(&queue->rskq_lock);
1878 	WARN_ON_ONCE(queue->rskq_accept_head);
1879 	queue->rskq_accept_head = head;
1880 	queue->rskq_accept_tail = tail;
1881 	spin_unlock_bh(&queue->rskq_lock);
1882 }
1883 
subflow_ulp_init(struct sock * sk)1884 static int subflow_ulp_init(struct sock *sk)
1885 {
1886 	struct inet_connection_sock *icsk = inet_csk(sk);
1887 	struct mptcp_subflow_context *ctx;
1888 	struct tcp_sock *tp = tcp_sk(sk);
1889 	int err = 0;
1890 
1891 	/* disallow attaching ULP to a socket unless it has been
1892 	 * created with sock_create_kern()
1893 	 */
1894 	if (!sk->sk_kern_sock) {
1895 		err = -EOPNOTSUPP;
1896 		goto out;
1897 	}
1898 
1899 	ctx = subflow_create_ctx(sk, GFP_KERNEL);
1900 	if (!ctx) {
1901 		err = -ENOMEM;
1902 		goto out;
1903 	}
1904 
1905 	pr_debug("subflow=%p, family=%d\n", ctx, sk->sk_family);
1906 
1907 	tp->is_mptcp = 1;
1908 	ctx->icsk_af_ops = icsk->icsk_af_ops;
1909 	icsk->icsk_af_ops = subflow_default_af_ops(sk);
1910 	ctx->tcp_state_change = sk->sk_state_change;
1911 	ctx->tcp_error_report = sk->sk_error_report;
1912 
1913 	WARN_ON_ONCE(sk->sk_data_ready != sock_def_readable);
1914 	WARN_ON_ONCE(sk->sk_write_space != sk_stream_write_space);
1915 
1916 	sk->sk_data_ready = subflow_data_ready;
1917 	sk->sk_write_space = subflow_write_space;
1918 	sk->sk_state_change = subflow_state_change;
1919 	sk->sk_error_report = subflow_error_report;
1920 out:
1921 	return err;
1922 }
1923 
subflow_ulp_release(struct sock * ssk)1924 static void subflow_ulp_release(struct sock *ssk)
1925 {
1926 	struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(ssk);
1927 	bool release = true;
1928 	struct sock *sk;
1929 
1930 	if (!ctx)
1931 		return;
1932 
1933 	sk = ctx->conn;
1934 	if (sk) {
1935 		/* if the msk has been orphaned, keep the ctx
1936 		 * alive, will be freed by __mptcp_close_ssk(),
1937 		 * when the subflow is still unaccepted
1938 		 */
1939 		release = ctx->disposable || list_empty(&ctx->node);
1940 
1941 		/* inet_child_forget() does not call sk_state_change(),
1942 		 * explicitly trigger the socket close machinery
1943 		 */
1944 		if (!release && !test_and_set_bit(MPTCP_WORK_CLOSE_SUBFLOW,
1945 						  &mptcp_sk(sk)->flags))
1946 			mptcp_schedule_work(sk);
1947 		sock_put(sk);
1948 	}
1949 
1950 	mptcp_subflow_ops_undo_override(ssk);
1951 	if (release)
1952 		kfree_rcu(ctx, rcu);
1953 }
1954 
subflow_ulp_clone(const struct request_sock * req,struct sock * newsk,const gfp_t priority)1955 static void subflow_ulp_clone(const struct request_sock *req,
1956 			      struct sock *newsk,
1957 			      const gfp_t priority)
1958 {
1959 	struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
1960 	struct mptcp_subflow_context *old_ctx = mptcp_subflow_ctx(newsk);
1961 	struct mptcp_subflow_context *new_ctx;
1962 
1963 	if (!tcp_rsk(req)->is_mptcp ||
1964 	    (!subflow_req->mp_capable && !subflow_req->mp_join)) {
1965 		subflow_ulp_fallback(newsk, old_ctx);
1966 		return;
1967 	}
1968 
1969 	new_ctx = subflow_create_ctx(newsk, priority);
1970 	if (!new_ctx) {
1971 		subflow_ulp_fallback(newsk, old_ctx);
1972 		return;
1973 	}
1974 
1975 	new_ctx->conn_finished = 1;
1976 	new_ctx->icsk_af_ops = old_ctx->icsk_af_ops;
1977 	new_ctx->tcp_state_change = old_ctx->tcp_state_change;
1978 	new_ctx->tcp_error_report = old_ctx->tcp_error_report;
1979 	new_ctx->rel_write_seq = 1;
1980 	new_ctx->tcp_sock = newsk;
1981 
1982 	if (subflow_req->mp_capable) {
1983 		/* see comments in subflow_syn_recv_sock(), MPTCP connection
1984 		 * is fully established only after we receive the remote key
1985 		 */
1986 		new_ctx->mp_capable = 1;
1987 		new_ctx->local_key = subflow_req->local_key;
1988 		new_ctx->token = subflow_req->token;
1989 		new_ctx->ssn_offset = subflow_req->ssn_offset;
1990 		new_ctx->idsn = subflow_req->idsn;
1991 
1992 		/* this is the first subflow, id is always 0 */
1993 		subflow_set_local_id(new_ctx, 0);
1994 	} else if (subflow_req->mp_join) {
1995 		new_ctx->ssn_offset = subflow_req->ssn_offset;
1996 		new_ctx->mp_join = 1;
1997 		new_ctx->fully_established = 1;
1998 		new_ctx->remote_key_valid = 1;
1999 		new_ctx->backup = subflow_req->backup;
2000 		new_ctx->request_bkup = subflow_req->request_bkup;
2001 		WRITE_ONCE(new_ctx->remote_id, subflow_req->remote_id);
2002 		new_ctx->token = subflow_req->token;
2003 		new_ctx->thmac = subflow_req->thmac;
2004 
2005 		/* the subflow req id is valid, fetched via subflow_check_req()
2006 		 * and subflow_token_join_request()
2007 		 */
2008 		subflow_set_local_id(new_ctx, subflow_req->local_id);
2009 	}
2010 }
2011 
tcp_release_cb_override(struct sock * ssk)2012 static void tcp_release_cb_override(struct sock *ssk)
2013 {
2014 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
2015 	long status;
2016 
2017 	/* process and clear all the pending actions, but leave the subflow into
2018 	 * the napi queue. To respect locking, only the same CPU that originated
2019 	 * the action can touch the list. mptcp_napi_poll will take care of it.
2020 	 */
2021 	status = set_mask_bits(&subflow->delegated_status, MPTCP_DELEGATE_ACTIONS_MASK, 0);
2022 	if (status)
2023 		mptcp_subflow_process_delegated(ssk, status);
2024 
2025 	tcp_release_cb(ssk);
2026 }
2027 
tcp_abort_override(struct sock * ssk,int err)2028 static int tcp_abort_override(struct sock *ssk, int err)
2029 {
2030 	/* closing a listener subflow requires a great deal of care.
2031 	 * keep it simple and just prevent such operation
2032 	 */
2033 	if (inet_sk_state_load(ssk) == TCP_LISTEN)
2034 		return -EINVAL;
2035 
2036 	return tcp_abort(ssk, err);
2037 }
2038 
2039 static struct tcp_ulp_ops subflow_ulp_ops __read_mostly = {
2040 	.name		= "mptcp",
2041 	.owner		= THIS_MODULE,
2042 	.init		= subflow_ulp_init,
2043 	.release	= subflow_ulp_release,
2044 	.clone		= subflow_ulp_clone,
2045 };
2046 
subflow_ops_init(struct request_sock_ops * subflow_ops)2047 static int subflow_ops_init(struct request_sock_ops *subflow_ops)
2048 {
2049 	subflow_ops->obj_size = sizeof(struct mptcp_subflow_request_sock);
2050 
2051 	subflow_ops->slab = kmem_cache_create(subflow_ops->slab_name,
2052 					      subflow_ops->obj_size, 0,
2053 					      SLAB_ACCOUNT |
2054 					      SLAB_TYPESAFE_BY_RCU,
2055 					      NULL);
2056 	if (!subflow_ops->slab)
2057 		return -ENOMEM;
2058 
2059 	return 0;
2060 }
2061 
mptcp_subflow_init(void)2062 void __init mptcp_subflow_init(void)
2063 {
2064 	mptcp_subflow_v4_request_sock_ops = tcp_request_sock_ops;
2065 	mptcp_subflow_v4_request_sock_ops.slab_name = "request_sock_subflow_v4";
2066 	mptcp_subflow_v4_request_sock_ops.destructor = subflow_v4_req_destructor;
2067 
2068 	if (subflow_ops_init(&mptcp_subflow_v4_request_sock_ops) != 0)
2069 		panic("MPTCP: failed to init subflow v4 request sock ops\n");
2070 
2071 	subflow_request_sock_ipv4_ops = tcp_request_sock_ipv4_ops;
2072 	subflow_request_sock_ipv4_ops.route_req = subflow_v4_route_req;
2073 	subflow_request_sock_ipv4_ops.send_synack = subflow_v4_send_synack;
2074 
2075 	subflow_specific = ipv4_specific;
2076 	subflow_specific.conn_request = subflow_v4_conn_request;
2077 	subflow_specific.syn_recv_sock = subflow_syn_recv_sock;
2078 	subflow_specific.sk_rx_dst_set = subflow_finish_connect;
2079 	subflow_specific.rebuild_header = subflow_rebuild_header;
2080 
2081 	tcp_prot_override = tcp_prot;
2082 	tcp_prot_override.release_cb = tcp_release_cb_override;
2083 	tcp_prot_override.diag_destroy = tcp_abort_override;
2084 
2085 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
2086 	/* In struct mptcp_subflow_request_sock, we assume the TCP request sock
2087 	 * structures for v4 and v6 have the same size. It should not changed in
2088 	 * the future but better to make sure to be warned if it is no longer
2089 	 * the case.
2090 	 */
2091 	BUILD_BUG_ON(sizeof(struct tcp_request_sock) != sizeof(struct tcp6_request_sock));
2092 
2093 	mptcp_subflow_v6_request_sock_ops = tcp6_request_sock_ops;
2094 	mptcp_subflow_v6_request_sock_ops.slab_name = "request_sock_subflow_v6";
2095 	mptcp_subflow_v6_request_sock_ops.destructor = subflow_v6_req_destructor;
2096 
2097 	if (subflow_ops_init(&mptcp_subflow_v6_request_sock_ops) != 0)
2098 		panic("MPTCP: failed to init subflow v6 request sock ops\n");
2099 
2100 	subflow_request_sock_ipv6_ops = tcp_request_sock_ipv6_ops;
2101 	subflow_request_sock_ipv6_ops.route_req = subflow_v6_route_req;
2102 	subflow_request_sock_ipv6_ops.send_synack = subflow_v6_send_synack;
2103 
2104 	subflow_v6_specific = ipv6_specific;
2105 	subflow_v6_specific.conn_request = subflow_v6_conn_request;
2106 	subflow_v6_specific.syn_recv_sock = subflow_syn_recv_sock;
2107 	subflow_v6_specific.sk_rx_dst_set = subflow_finish_connect;
2108 	subflow_v6_specific.rebuild_header = subflow_v6_rebuild_header;
2109 
2110 	subflow_v6m_specific = subflow_v6_specific;
2111 	subflow_v6m_specific.queue_xmit = ipv4_specific.queue_xmit;
2112 	subflow_v6m_specific.send_check = ipv4_specific.send_check;
2113 	subflow_v6m_specific.net_header_len = ipv4_specific.net_header_len;
2114 	subflow_v6m_specific.mtu_reduced = ipv4_specific.mtu_reduced;
2115 	subflow_v6m_specific.net_frag_header_len = 0;
2116 	subflow_v6m_specific.rebuild_header = subflow_rebuild_header;
2117 
2118 	tcpv6_prot_override = tcpv6_prot;
2119 	tcpv6_prot_override.release_cb = tcp_release_cb_override;
2120 	tcpv6_prot_override.diag_destroy = tcp_abort_override;
2121 #endif
2122 
2123 	mptcp_diag_subflow_init(&subflow_ulp_ops);
2124 
2125 	if (tcp_register_ulp(&subflow_ulp_ops) != 0)
2126 		panic("MPTCP: failed to register subflows to ULP\n");
2127 }
2128