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
135 /* Init mptcp request socket.
136 *
137 * Returns an error code if a JOIN has failed and a TCP reset
138 * should be sent.
139 */
subflow_check_req(struct request_sock * req,const struct sock * sk_listener,struct sk_buff * skb)140 static int subflow_check_req(struct request_sock *req,
141 const struct sock *sk_listener,
142 struct sk_buff *skb)
143 {
144 struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk_listener);
145 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
146 struct mptcp_options_received mp_opt;
147 bool opt_mp_capable, opt_mp_join;
148
149 pr_debug("subflow_req=%p, listener=%p\n", subflow_req, listener);
150
151 #ifdef CONFIG_TCP_MD5SIG
152 /* no MPTCP if MD5SIG is enabled on this socket or we may run out of
153 * TCP option space.
154 */
155 if (rcu_access_pointer(tcp_sk(sk_listener)->md5sig_info))
156 return -EINVAL;
157 #endif
158
159 mptcp_get_options(skb, &mp_opt);
160
161 opt_mp_capable = !!(mp_opt.suboptions & OPTION_MPTCP_MPC_SYN);
162 opt_mp_join = !!(mp_opt.suboptions & OPTION_MPTCP_MPJ_SYN);
163 if (opt_mp_capable) {
164 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MPCAPABLEPASSIVE);
165
166 if (opt_mp_join)
167 return 0;
168 } else if (opt_mp_join) {
169 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINSYNRX);
170
171 if (mp_opt.backup)
172 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINSYNBACKUPRX);
173 }
174
175 if (opt_mp_capable && listener->request_mptcp) {
176 int err, retries = MPTCP_TOKEN_MAX_RETRIES;
177
178 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq;
179 again:
180 do {
181 get_random_bytes(&subflow_req->local_key, sizeof(subflow_req->local_key));
182 } while (subflow_req->local_key == 0);
183
184 if (unlikely(req->syncookie)) {
185 mptcp_crypto_key_sha(subflow_req->local_key,
186 &subflow_req->token,
187 &subflow_req->idsn);
188 if (mptcp_token_exists(subflow_req->token)) {
189 if (retries-- > 0)
190 goto again;
191 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_TOKENFALLBACKINIT);
192 } else {
193 subflow_req->mp_capable = 1;
194 }
195 return 0;
196 }
197
198 err = mptcp_token_new_request(req);
199 if (err == 0)
200 subflow_req->mp_capable = 1;
201 else if (retries-- > 0)
202 goto again;
203 else
204 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_TOKENFALLBACKINIT);
205
206 } else if (opt_mp_join && listener->request_mptcp) {
207 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq;
208 subflow_req->mp_join = 1;
209 subflow_req->backup = mp_opt.backup;
210 subflow_req->remote_id = mp_opt.join_id;
211 subflow_req->token = mp_opt.token;
212 subflow_req->remote_nonce = mp_opt.nonce;
213 subflow_req->msk = subflow_token_join_request(req);
214
215 /* Can't fall back to TCP in this case. */
216 if (!subflow_req->msk) {
217 subflow_add_reset_reason(skb, MPTCP_RST_EMPTCP);
218 return -EPERM;
219 }
220
221 if (subflow_use_different_sport(subflow_req->msk, sk_listener)) {
222 pr_debug("syn inet_sport=%d %d\n",
223 ntohs(inet_sk(sk_listener)->inet_sport),
224 ntohs(inet_sk((struct sock *)subflow_req->msk)->inet_sport));
225 if (!mptcp_pm_sport_in_anno_list(subflow_req->msk, sk_listener)) {
226 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MISMATCHPORTSYNRX);
227 return -EPERM;
228 }
229 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINPORTSYNRX);
230 }
231
232 subflow_req_create_thmac(subflow_req);
233
234 if (unlikely(req->syncookie)) {
235 if (mptcp_can_accept_new_subflow(subflow_req->msk))
236 subflow_init_req_cookie_join_save(subflow_req, skb);
237 else
238 return -EPERM;
239 }
240
241 pr_debug("token=%u, remote_nonce=%u msk=%p\n", subflow_req->token,
242 subflow_req->remote_nonce, subflow_req->msk);
243 }
244
245 return 0;
246 }
247
mptcp_subflow_init_cookie_req(struct request_sock * req,const struct sock * sk_listener,struct sk_buff * skb)248 int mptcp_subflow_init_cookie_req(struct request_sock *req,
249 const struct sock *sk_listener,
250 struct sk_buff *skb)
251 {
252 struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk_listener);
253 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
254 struct mptcp_options_received mp_opt;
255 bool opt_mp_capable, opt_mp_join;
256 int err;
257
258 subflow_init_req(req, sk_listener);
259 mptcp_get_options(skb, &mp_opt);
260
261 opt_mp_capable = !!(mp_opt.suboptions & OPTION_MPTCP_MPC_ACK);
262 opt_mp_join = !!(mp_opt.suboptions & OPTION_MPTCP_MPJ_ACK);
263 if (opt_mp_capable && opt_mp_join)
264 return -EINVAL;
265
266 if (opt_mp_capable && listener->request_mptcp) {
267 if (mp_opt.sndr_key == 0)
268 return -EINVAL;
269
270 subflow_req->local_key = mp_opt.rcvr_key;
271 err = mptcp_token_new_request(req);
272 if (err)
273 return err;
274
275 subflow_req->mp_capable = 1;
276 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq - 1;
277 } else if (opt_mp_join && listener->request_mptcp) {
278 if (!mptcp_token_join_cookie_init_state(subflow_req, skb))
279 return -EINVAL;
280
281 subflow_req->mp_join = 1;
282 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq - 1;
283 }
284
285 return 0;
286 }
287 EXPORT_SYMBOL_GPL(mptcp_subflow_init_cookie_req);
288
subflow_v4_route_req(const struct sock * sk,struct sk_buff * skb,struct flowi * fl,struct request_sock * req)289 static struct dst_entry *subflow_v4_route_req(const struct sock *sk,
290 struct sk_buff *skb,
291 struct flowi *fl,
292 struct request_sock *req)
293 {
294 struct dst_entry *dst;
295 int err;
296
297 tcp_rsk(req)->is_mptcp = 1;
298 subflow_init_req(req, sk);
299
300 dst = tcp_request_sock_ipv4_ops.route_req(sk, skb, fl, req);
301 if (!dst)
302 return NULL;
303
304 err = subflow_check_req(req, sk, skb);
305 if (err == 0)
306 return dst;
307
308 dst_release(dst);
309 if (!req->syncookie)
310 tcp_request_sock_ops.send_reset(sk, skb);
311 return NULL;
312 }
313
subflow_prep_synack(const struct sock * sk,struct request_sock * req,struct tcp_fastopen_cookie * foc,enum tcp_synack_type synack_type)314 static void subflow_prep_synack(const struct sock *sk, struct request_sock *req,
315 struct tcp_fastopen_cookie *foc,
316 enum tcp_synack_type synack_type)
317 {
318 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
319 struct inet_request_sock *ireq = inet_rsk(req);
320
321 /* clear tstamp_ok, as needed depending on cookie */
322 if (foc && foc->len > -1)
323 ireq->tstamp_ok = 0;
324
325 if (synack_type == TCP_SYNACK_FASTOPEN)
326 mptcp_fastopen_subflow_synack_set_params(subflow, req);
327 }
328
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)329 static int subflow_v4_send_synack(const struct sock *sk, struct dst_entry *dst,
330 struct flowi *fl,
331 struct request_sock *req,
332 struct tcp_fastopen_cookie *foc,
333 enum tcp_synack_type synack_type,
334 struct sk_buff *syn_skb)
335 {
336 subflow_prep_synack(sk, req, foc, synack_type);
337
338 return tcp_request_sock_ipv4_ops.send_synack(sk, dst, fl, req, foc,
339 synack_type, syn_skb);
340 }
341
342 #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)343 static int subflow_v6_send_synack(const struct sock *sk, struct dst_entry *dst,
344 struct flowi *fl,
345 struct request_sock *req,
346 struct tcp_fastopen_cookie *foc,
347 enum tcp_synack_type synack_type,
348 struct sk_buff *syn_skb)
349 {
350 subflow_prep_synack(sk, req, foc, synack_type);
351
352 return tcp_request_sock_ipv6_ops.send_synack(sk, dst, fl, req, foc,
353 synack_type, syn_skb);
354 }
355
subflow_v6_route_req(const struct sock * sk,struct sk_buff * skb,struct flowi * fl,struct request_sock * req)356 static struct dst_entry *subflow_v6_route_req(const struct sock *sk,
357 struct sk_buff *skb,
358 struct flowi *fl,
359 struct request_sock *req)
360 {
361 struct dst_entry *dst;
362 int err;
363
364 tcp_rsk(req)->is_mptcp = 1;
365 subflow_init_req(req, sk);
366
367 dst = tcp_request_sock_ipv6_ops.route_req(sk, skb, fl, req);
368 if (!dst)
369 return NULL;
370
371 err = subflow_check_req(req, sk, skb);
372 if (err == 0)
373 return dst;
374
375 dst_release(dst);
376 if (!req->syncookie)
377 tcp6_request_sock_ops.send_reset(sk, skb);
378 return NULL;
379 }
380 #endif
381
382 /* validate received truncated hmac and create hmac for third ACK */
subflow_thmac_valid(struct mptcp_subflow_context * subflow)383 static bool subflow_thmac_valid(struct mptcp_subflow_context *subflow)
384 {
385 u8 hmac[SHA256_DIGEST_SIZE];
386 u64 thmac;
387
388 subflow_generate_hmac(subflow->remote_key, subflow->local_key,
389 subflow->remote_nonce, subflow->local_nonce,
390 hmac);
391
392 thmac = get_unaligned_be64(hmac);
393 pr_debug("subflow=%p, token=%u, thmac=%llu, subflow->thmac=%llu\n",
394 subflow, subflow->token, thmac, subflow->thmac);
395
396 return thmac == subflow->thmac;
397 }
398
mptcp_subflow_reset(struct sock * ssk)399 void mptcp_subflow_reset(struct sock *ssk)
400 {
401 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
402 struct sock *sk = subflow->conn;
403
404 /* mptcp_mp_fail_no_response() can reach here on an already closed
405 * socket
406 */
407 if (ssk->sk_state == TCP_CLOSE)
408 return;
409
410 /* must hold: tcp_done() could drop last reference on parent */
411 sock_hold(sk);
412
413 tcp_send_active_reset(ssk, GFP_ATOMIC);
414 tcp_done(ssk);
415 if (!test_and_set_bit(MPTCP_WORK_CLOSE_SUBFLOW, &mptcp_sk(sk)->flags))
416 mptcp_schedule_work(sk);
417
418 sock_put(sk);
419 }
420
subflow_use_different_dport(struct mptcp_sock * msk,const struct sock * sk)421 static bool subflow_use_different_dport(struct mptcp_sock *msk, const struct sock *sk)
422 {
423 return inet_sk(sk)->inet_dport != inet_sk((struct sock *)msk)->inet_dport;
424 }
425
__mptcp_sync_state(struct sock * sk,int state)426 void __mptcp_sync_state(struct sock *sk, int state)
427 {
428 struct mptcp_subflow_context *subflow;
429 struct mptcp_sock *msk = mptcp_sk(sk);
430 struct sock *ssk = msk->first;
431
432 subflow = mptcp_subflow_ctx(ssk);
433 __mptcp_propagate_sndbuf(sk, ssk);
434 if (!msk->rcvspace_init)
435 mptcp_rcv_space_init(msk, ssk);
436
437 if (sk->sk_state == TCP_SYN_SENT) {
438 /* subflow->idsn is always available is TCP_SYN_SENT state,
439 * even for the FASTOPEN scenarios
440 */
441 WRITE_ONCE(msk->write_seq, subflow->idsn + 1);
442 WRITE_ONCE(msk->snd_nxt, msk->write_seq);
443 mptcp_set_state(sk, state);
444 sk->sk_state_change(sk);
445 }
446 }
447
subflow_set_remote_key(struct mptcp_sock * msk,struct mptcp_subflow_context * subflow,const struct mptcp_options_received * mp_opt)448 static void subflow_set_remote_key(struct mptcp_sock *msk,
449 struct mptcp_subflow_context *subflow,
450 const struct mptcp_options_received *mp_opt)
451 {
452 /* active MPC subflow will reach here multiple times:
453 * at subflow_finish_connect() time and at 4th ack time
454 */
455 if (subflow->remote_key_valid)
456 return;
457
458 subflow->remote_key_valid = 1;
459 subflow->remote_key = mp_opt->sndr_key;
460 mptcp_crypto_key_sha(subflow->remote_key, NULL, &subflow->iasn);
461 subflow->iasn++;
462
463 WRITE_ONCE(msk->remote_key, subflow->remote_key);
464 WRITE_ONCE(msk->ack_seq, subflow->iasn);
465 WRITE_ONCE(msk->can_ack, true);
466 atomic64_set(&msk->rcv_wnd_sent, subflow->iasn);
467 }
468
mptcp_propagate_state(struct sock * sk,struct sock * ssk,struct mptcp_subflow_context * subflow,const struct mptcp_options_received * mp_opt)469 static void mptcp_propagate_state(struct sock *sk, struct sock *ssk,
470 struct mptcp_subflow_context *subflow,
471 const struct mptcp_options_received *mp_opt)
472 {
473 struct mptcp_sock *msk = mptcp_sk(sk);
474
475 mptcp_data_lock(sk);
476 if (mp_opt) {
477 /* Options are available only in the non fallback cases
478 * avoid updating rx path fields otherwise
479 */
480 WRITE_ONCE(msk->snd_una, subflow->idsn + 1);
481 WRITE_ONCE(msk->wnd_end, subflow->idsn + 1 + tcp_sk(ssk)->snd_wnd);
482 subflow_set_remote_key(msk, subflow, mp_opt);
483 }
484
485 if (!sock_owned_by_user(sk)) {
486 __mptcp_sync_state(sk, ssk->sk_state);
487 } else {
488 msk->pending_state = ssk->sk_state;
489 __set_bit(MPTCP_SYNC_STATE, &msk->cb_flags);
490 }
491 mptcp_data_unlock(sk);
492 }
493
subflow_finish_connect(struct sock * sk,const struct sk_buff * skb)494 static void subflow_finish_connect(struct sock *sk, const struct sk_buff *skb)
495 {
496 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
497 struct mptcp_options_received mp_opt;
498 struct sock *parent = subflow->conn;
499 struct mptcp_sock *msk;
500
501 subflow->icsk_af_ops->sk_rx_dst_set(sk, skb);
502
503 /* be sure no special action on any packet other than syn-ack */
504 if (subflow->conn_finished)
505 return;
506
507 msk = mptcp_sk(parent);
508 subflow->rel_write_seq = 1;
509 subflow->conn_finished = 1;
510 subflow->ssn_offset = TCP_SKB_CB(skb)->seq;
511 pr_debug("subflow=%p synack seq=%x\n", subflow, subflow->ssn_offset);
512
513 mptcp_get_options(skb, &mp_opt);
514 if (subflow->request_mptcp) {
515 if (!(mp_opt.suboptions & OPTION_MPTCP_MPC_SYNACK)) {
516 MPTCP_INC_STATS(sock_net(sk),
517 MPTCP_MIB_MPCAPABLEACTIVEFALLBACK);
518 mptcp_do_fallback(sk);
519 pr_fallback(msk);
520 goto fallback;
521 }
522
523 if (mp_opt.suboptions & OPTION_MPTCP_CSUMREQD)
524 WRITE_ONCE(msk->csum_enabled, true);
525 if (mp_opt.deny_join_id0)
526 WRITE_ONCE(msk->pm.remote_deny_join_id0, true);
527 subflow->mp_capable = 1;
528 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPCAPABLEACTIVEACK);
529 mptcp_finish_connect(sk);
530 mptcp_propagate_state(parent, sk, subflow, &mp_opt);
531 } else if (subflow->request_join) {
532 u8 hmac[SHA256_DIGEST_SIZE];
533
534 if (!(mp_opt.suboptions & OPTION_MPTCP_MPJ_SYNACK)) {
535 subflow->reset_reason = MPTCP_RST_EMPTCP;
536 goto do_reset;
537 }
538
539 subflow->backup = mp_opt.backup;
540 subflow->thmac = mp_opt.thmac;
541 subflow->remote_nonce = mp_opt.nonce;
542 WRITE_ONCE(subflow->remote_id, mp_opt.join_id);
543 pr_debug("subflow=%p, thmac=%llu, remote_nonce=%u backup=%d\n",
544 subflow, subflow->thmac, subflow->remote_nonce,
545 subflow->backup);
546
547 if (!subflow_thmac_valid(subflow)) {
548 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINACKMAC);
549 subflow->reset_reason = MPTCP_RST_EMPTCP;
550 goto do_reset;
551 }
552
553 if (!mptcp_finish_join(sk))
554 goto do_reset;
555
556 subflow_generate_hmac(subflow->local_key, subflow->remote_key,
557 subflow->local_nonce,
558 subflow->remote_nonce,
559 hmac);
560 memcpy(subflow->hmac, hmac, MPTCPOPT_HMAC_LEN);
561
562 subflow->mp_join = 1;
563 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINSYNACKRX);
564
565 if (subflow->backup)
566 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINSYNACKBACKUPRX);
567
568 if (subflow_use_different_dport(msk, sk)) {
569 pr_debug("synack inet_dport=%d %d\n",
570 ntohs(inet_sk(sk)->inet_dport),
571 ntohs(inet_sk(parent)->inet_dport));
572 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINPORTSYNACKRX);
573 }
574 } else if (mptcp_check_fallback(sk)) {
575 fallback:
576 mptcp_propagate_state(parent, sk, subflow, NULL);
577 }
578 return;
579
580 do_reset:
581 subflow->reset_transient = 0;
582 mptcp_subflow_reset(sk);
583 }
584
subflow_set_local_id(struct mptcp_subflow_context * subflow,int local_id)585 static void subflow_set_local_id(struct mptcp_subflow_context *subflow, int local_id)
586 {
587 WARN_ON_ONCE(local_id < 0 || local_id > 255);
588 WRITE_ONCE(subflow->local_id, local_id);
589 }
590
subflow_chk_local_id(struct sock * sk)591 static int subflow_chk_local_id(struct sock *sk)
592 {
593 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
594 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
595 int err;
596
597 if (likely(subflow->local_id >= 0))
598 return 0;
599
600 err = mptcp_pm_get_local_id(msk, (struct sock_common *)sk);
601 if (err < 0)
602 return err;
603
604 subflow_set_local_id(subflow, err);
605 subflow->request_bkup = mptcp_pm_is_backup(msk, (struct sock_common *)sk);
606
607 return 0;
608 }
609
subflow_rebuild_header(struct sock * sk)610 static int subflow_rebuild_header(struct sock *sk)
611 {
612 int err = subflow_chk_local_id(sk);
613
614 if (unlikely(err < 0))
615 return err;
616
617 return inet_sk_rebuild_header(sk);
618 }
619
620 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
subflow_v6_rebuild_header(struct sock * sk)621 static int subflow_v6_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 inet6_sk_rebuild_header(sk);
629 }
630 #endif
631
632 static struct request_sock_ops mptcp_subflow_v4_request_sock_ops __ro_after_init;
633 static struct tcp_request_sock_ops subflow_request_sock_ipv4_ops __ro_after_init;
634
subflow_v4_conn_request(struct sock * sk,struct sk_buff * skb)635 static int subflow_v4_conn_request(struct sock *sk, struct sk_buff *skb)
636 {
637 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
638
639 pr_debug("subflow=%p\n", subflow);
640
641 /* Never answer to SYNs sent to broadcast or multicast */
642 if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
643 goto drop;
644
645 return tcp_conn_request(&mptcp_subflow_v4_request_sock_ops,
646 &subflow_request_sock_ipv4_ops,
647 sk, skb);
648 drop:
649 tcp_listendrop(sk);
650 return 0;
651 }
652
subflow_v4_req_destructor(struct request_sock * req)653 static void subflow_v4_req_destructor(struct request_sock *req)
654 {
655 subflow_req_destructor(req);
656 tcp_request_sock_ops.destructor(req);
657 }
658
659 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
660 static struct request_sock_ops mptcp_subflow_v6_request_sock_ops __ro_after_init;
661 static struct tcp_request_sock_ops subflow_request_sock_ipv6_ops __ro_after_init;
662 static struct inet_connection_sock_af_ops subflow_v6_specific __ro_after_init;
663 static struct inet_connection_sock_af_ops subflow_v6m_specific __ro_after_init;
664 static struct proto tcpv6_prot_override __ro_after_init;
665
subflow_v6_conn_request(struct sock * sk,struct sk_buff * skb)666 static int subflow_v6_conn_request(struct sock *sk, struct sk_buff *skb)
667 {
668 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
669
670 pr_debug("subflow=%p\n", subflow);
671
672 if (skb->protocol == htons(ETH_P_IP))
673 return subflow_v4_conn_request(sk, skb);
674
675 if (!ipv6_unicast_destination(skb))
676 goto drop;
677
678 if (ipv6_addr_v4mapped(&ipv6_hdr(skb)->saddr)) {
679 __IP6_INC_STATS(sock_net(sk), NULL, IPSTATS_MIB_INHDRERRORS);
680 return 0;
681 }
682
683 return tcp_conn_request(&mptcp_subflow_v6_request_sock_ops,
684 &subflow_request_sock_ipv6_ops, sk, skb);
685
686 drop:
687 tcp_listendrop(sk);
688 return 0; /* don't send reset */
689 }
690
subflow_v6_req_destructor(struct request_sock * req)691 static void subflow_v6_req_destructor(struct request_sock *req)
692 {
693 subflow_req_destructor(req);
694 tcp6_request_sock_ops.destructor(req);
695 }
696 #endif
697
mptcp_subflow_reqsk_alloc(const struct request_sock_ops * ops,struct sock * sk_listener,bool attach_listener)698 struct request_sock *mptcp_subflow_reqsk_alloc(const struct request_sock_ops *ops,
699 struct sock *sk_listener,
700 bool attach_listener)
701 {
702 if (ops->family == AF_INET)
703 ops = &mptcp_subflow_v4_request_sock_ops;
704 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
705 else if (ops->family == AF_INET6)
706 ops = &mptcp_subflow_v6_request_sock_ops;
707 #endif
708
709 return inet_reqsk_alloc(ops, sk_listener, attach_listener);
710 }
711 EXPORT_SYMBOL(mptcp_subflow_reqsk_alloc);
712
713 /* validate hmac received in third ACK */
subflow_hmac_valid(const struct request_sock * req,const struct mptcp_options_received * mp_opt)714 static bool subflow_hmac_valid(const struct request_sock *req,
715 const struct mptcp_options_received *mp_opt)
716 {
717 const struct mptcp_subflow_request_sock *subflow_req;
718 u8 hmac[SHA256_DIGEST_SIZE];
719 struct mptcp_sock *msk;
720
721 subflow_req = mptcp_subflow_rsk(req);
722 msk = subflow_req->msk;
723 if (!msk)
724 return false;
725
726 subflow_generate_hmac(msk->remote_key, msk->local_key,
727 subflow_req->remote_nonce,
728 subflow_req->local_nonce, hmac);
729
730 return !crypto_memneq(hmac, mp_opt->hmac, MPTCPOPT_HMAC_LEN);
731 }
732
subflow_ulp_fallback(struct sock * sk,struct mptcp_subflow_context * old_ctx)733 static void subflow_ulp_fallback(struct sock *sk,
734 struct mptcp_subflow_context *old_ctx)
735 {
736 struct inet_connection_sock *icsk = inet_csk(sk);
737
738 mptcp_subflow_tcp_fallback(sk, old_ctx);
739 icsk->icsk_ulp_ops = NULL;
740 rcu_assign_pointer(icsk->icsk_ulp_data, NULL);
741 tcp_sk(sk)->is_mptcp = 0;
742
743 mptcp_subflow_ops_undo_override(sk);
744 }
745
mptcp_subflow_drop_ctx(struct sock * ssk)746 void mptcp_subflow_drop_ctx(struct sock *ssk)
747 {
748 struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(ssk);
749
750 if (!ctx)
751 return;
752
753 list_del(&mptcp_subflow_ctx(ssk)->node);
754 if (inet_csk(ssk)->icsk_ulp_ops) {
755 subflow_ulp_fallback(ssk, ctx);
756 if (ctx->conn)
757 sock_put(ctx->conn);
758 }
759
760 kfree_rcu(ctx, rcu);
761 }
762
__mptcp_subflow_fully_established(struct mptcp_sock * msk,struct mptcp_subflow_context * subflow,const struct mptcp_options_received * mp_opt)763 void __mptcp_subflow_fully_established(struct mptcp_sock *msk,
764 struct mptcp_subflow_context *subflow,
765 const struct mptcp_options_received *mp_opt)
766 {
767 subflow_set_remote_key(msk, subflow, mp_opt);
768 subflow->fully_established = 1;
769 WRITE_ONCE(msk->fully_established, true);
770
771 if (subflow->is_mptfo)
772 __mptcp_fastopen_gen_msk_ackseq(msk, subflow, mp_opt);
773 }
774
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)775 static struct sock *subflow_syn_recv_sock(const struct sock *sk,
776 struct sk_buff *skb,
777 struct request_sock *req,
778 struct dst_entry *dst,
779 struct request_sock *req_unhash,
780 bool *own_req)
781 {
782 struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk);
783 struct mptcp_subflow_request_sock *subflow_req;
784 struct mptcp_options_received mp_opt;
785 bool fallback, fallback_is_fatal;
786 struct mptcp_sock *owner;
787 struct sock *child;
788
789 pr_debug("listener=%p, req=%p, conn=%p\n", listener, req, listener->conn);
790
791 /* After child creation we must look for MPC even when options
792 * are not parsed
793 */
794 mp_opt.suboptions = 0;
795
796 /* hopefully temporary handling for MP_JOIN+syncookie */
797 subflow_req = mptcp_subflow_rsk(req);
798 fallback_is_fatal = tcp_rsk(req)->is_mptcp && subflow_req->mp_join;
799 fallback = !tcp_rsk(req)->is_mptcp;
800 if (fallback)
801 goto create_child;
802
803 /* if the sk is MP_CAPABLE, we try to fetch the client key */
804 if (subflow_req->mp_capable) {
805 /* we can receive and accept an in-window, out-of-order pkt,
806 * which may not carry the MP_CAPABLE opt even on mptcp enabled
807 * paths: always try to extract the peer key, and fallback
808 * for packets missing it.
809 * Even OoO DSS packets coming legitly after dropped or
810 * reordered MPC will cause fallback, but we don't have other
811 * options.
812 */
813 mptcp_get_options(skb, &mp_opt);
814 if (!(mp_opt.suboptions &
815 (OPTION_MPTCP_MPC_SYN | OPTION_MPTCP_MPC_ACK)))
816 fallback = true;
817
818 } else if (subflow_req->mp_join) {
819 mptcp_get_options(skb, &mp_opt);
820 if (!(mp_opt.suboptions & OPTION_MPTCP_MPJ_ACK) ||
821 !subflow_hmac_valid(req, &mp_opt) ||
822 !mptcp_can_accept_new_subflow(subflow_req->msk)) {
823 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKMAC);
824 fallback = true;
825 }
826 }
827
828 create_child:
829 child = listener->icsk_af_ops->syn_recv_sock(sk, skb, req, dst,
830 req_unhash, own_req);
831
832 if (child && *own_req) {
833 struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(child);
834
835 tcp_rsk(req)->drop_req = false;
836
837 /* we need to fallback on ctx allocation failure and on pre-reqs
838 * checking above. In the latter scenario we additionally need
839 * to reset the context to non MPTCP status.
840 */
841 if (!ctx || fallback) {
842 if (fallback_is_fatal) {
843 subflow_add_reset_reason(skb, MPTCP_RST_EMPTCP);
844 goto dispose_child;
845 }
846 goto fallback;
847 }
848
849 /* ssk inherits options of listener sk */
850 ctx->setsockopt_seq = listener->setsockopt_seq;
851
852 if (ctx->mp_capable) {
853 ctx->conn = mptcp_sk_clone_init(listener->conn, &mp_opt, child, req);
854 if (!ctx->conn)
855 goto fallback;
856
857 ctx->subflow_id = 1;
858 owner = mptcp_sk(ctx->conn);
859 mptcp_pm_new_connection(owner, child, 1);
860
861 /* with OoO packets we can reach here without ingress
862 * mpc option
863 */
864 if (mp_opt.suboptions & OPTION_MPTCP_MPC_ACK) {
865 mptcp_pm_fully_established(owner, child);
866 ctx->pm_notified = 1;
867 }
868 } else if (ctx->mp_join) {
869 owner = subflow_req->msk;
870 if (!owner) {
871 subflow_add_reset_reason(skb, MPTCP_RST_EPROHIBIT);
872 goto dispose_child;
873 }
874
875 /* move the msk reference ownership to the subflow */
876 subflow_req->msk = NULL;
877 ctx->conn = (struct sock *)owner;
878
879 if (subflow_use_different_sport(owner, sk)) {
880 pr_debug("ack inet_sport=%d %d\n",
881 ntohs(inet_sk(sk)->inet_sport),
882 ntohs(inet_sk((struct sock *)owner)->inet_sport));
883 if (!mptcp_pm_sport_in_anno_list(owner, sk)) {
884 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MISMATCHPORTACKRX);
885 goto dispose_child;
886 }
887 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINPORTACKRX);
888 }
889
890 if (!mptcp_finish_join(child))
891 goto dispose_child;
892
893 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKRX);
894 tcp_rsk(req)->drop_req = true;
895 }
896 }
897
898 /* check for expected invariant - should never trigger, just help
899 * catching eariler subtle bugs
900 */
901 WARN_ON_ONCE(child && *own_req && tcp_sk(child)->is_mptcp &&
902 (!mptcp_subflow_ctx(child) ||
903 !mptcp_subflow_ctx(child)->conn));
904 return child;
905
906 dispose_child:
907 mptcp_subflow_drop_ctx(child);
908 tcp_rsk(req)->drop_req = true;
909 inet_csk_prepare_for_destroy_sock(child);
910 tcp_done(child);
911 req->rsk_ops->send_reset(sk, skb);
912
913 /* The last child reference will be released by the caller */
914 return child;
915
916 fallback:
917 if (fallback)
918 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MPCAPABLEPASSIVEFALLBACK);
919 mptcp_subflow_drop_ctx(child);
920 return child;
921 }
922
923 static struct inet_connection_sock_af_ops subflow_specific __ro_after_init;
924 static struct proto tcp_prot_override __ro_after_init;
925
926 enum mapping_status {
927 MAPPING_OK,
928 MAPPING_INVALID,
929 MAPPING_EMPTY,
930 MAPPING_DATA_FIN,
931 MAPPING_DUMMY,
932 MAPPING_BAD_CSUM
933 };
934
dbg_bad_map(struct mptcp_subflow_context * subflow,u32 ssn)935 static void dbg_bad_map(struct mptcp_subflow_context *subflow, u32 ssn)
936 {
937 pr_debug("Bad mapping: ssn=%d map_seq=%d map_data_len=%d\n",
938 ssn, subflow->map_subflow_seq, subflow->map_data_len);
939 }
940
skb_is_fully_mapped(struct sock * ssk,struct sk_buff * skb)941 static bool skb_is_fully_mapped(struct sock *ssk, struct sk_buff *skb)
942 {
943 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
944 unsigned int skb_consumed;
945
946 skb_consumed = tcp_sk(ssk)->copied_seq - TCP_SKB_CB(skb)->seq;
947 if (WARN_ON_ONCE(skb_consumed >= skb->len))
948 return true;
949
950 return skb->len - skb_consumed <= subflow->map_data_len -
951 mptcp_subflow_get_map_offset(subflow);
952 }
953
validate_mapping(struct sock * ssk,struct sk_buff * skb)954 static bool validate_mapping(struct sock *ssk, struct sk_buff *skb)
955 {
956 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
957 u32 ssn = tcp_sk(ssk)->copied_seq - subflow->ssn_offset;
958
959 if (unlikely(before(ssn, subflow->map_subflow_seq))) {
960 /* Mapping covers data later in the subflow stream,
961 * currently unsupported.
962 */
963 dbg_bad_map(subflow, ssn);
964 return false;
965 }
966 if (unlikely(!before(ssn, subflow->map_subflow_seq +
967 subflow->map_data_len))) {
968 /* Mapping does covers past subflow data, invalid */
969 dbg_bad_map(subflow, ssn);
970 return false;
971 }
972 return true;
973 }
974
validate_data_csum(struct sock * ssk,struct sk_buff * skb,bool csum_reqd)975 static enum mapping_status validate_data_csum(struct sock *ssk, struct sk_buff *skb,
976 bool csum_reqd)
977 {
978 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
979 u32 offset, seq, delta;
980 __sum16 csum;
981 int len;
982
983 if (!csum_reqd)
984 return MAPPING_OK;
985
986 /* mapping already validated on previous traversal */
987 if (subflow->map_csum_len == subflow->map_data_len)
988 return MAPPING_OK;
989
990 /* traverse the receive queue, ensuring it contains a full
991 * DSS mapping and accumulating the related csum.
992 * Preserve the accoumlate csum across multiple calls, to compute
993 * the csum only once
994 */
995 delta = subflow->map_data_len - subflow->map_csum_len;
996 for (;;) {
997 seq = tcp_sk(ssk)->copied_seq + subflow->map_csum_len;
998 offset = seq - TCP_SKB_CB(skb)->seq;
999
1000 /* if the current skb has not been accounted yet, csum its contents
1001 * up to the amount covered by the current DSS
1002 */
1003 if (offset < skb->len) {
1004 __wsum csum;
1005
1006 len = min(skb->len - offset, delta);
1007 csum = skb_checksum(skb, offset, len, 0);
1008 subflow->map_data_csum = csum_block_add(subflow->map_data_csum, csum,
1009 subflow->map_csum_len);
1010
1011 delta -= len;
1012 subflow->map_csum_len += len;
1013 }
1014 if (delta == 0)
1015 break;
1016
1017 if (skb_queue_is_last(&ssk->sk_receive_queue, skb)) {
1018 /* if this subflow is closed, the partial mapping
1019 * will be never completed; flush the pending skbs, so
1020 * that subflow_sched_work_if_closed() can kick in
1021 */
1022 if (unlikely(ssk->sk_state == TCP_CLOSE))
1023 while ((skb = skb_peek(&ssk->sk_receive_queue)))
1024 sk_eat_skb(ssk, skb);
1025
1026 /* not enough data to validate the csum */
1027 return MAPPING_EMPTY;
1028 }
1029
1030 /* the DSS mapping for next skbs will be validated later,
1031 * when a get_mapping_status call will process such skb
1032 */
1033 skb = skb->next;
1034 }
1035
1036 /* note that 'map_data_len' accounts only for the carried data, does
1037 * not include the eventual seq increment due to the data fin,
1038 * while the pseudo header requires the original DSS data len,
1039 * including that
1040 */
1041 csum = __mptcp_make_csum(subflow->map_seq,
1042 subflow->map_subflow_seq,
1043 subflow->map_data_len + subflow->map_data_fin,
1044 subflow->map_data_csum);
1045 if (unlikely(csum)) {
1046 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DATACSUMERR);
1047 return MAPPING_BAD_CSUM;
1048 }
1049
1050 subflow->valid_csum_seen = 1;
1051 return MAPPING_OK;
1052 }
1053
get_mapping_status(struct sock * ssk,struct mptcp_sock * msk)1054 static enum mapping_status get_mapping_status(struct sock *ssk,
1055 struct mptcp_sock *msk)
1056 {
1057 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1058 bool csum_reqd = READ_ONCE(msk->csum_enabled);
1059 struct mptcp_ext *mpext;
1060 struct sk_buff *skb;
1061 u16 data_len;
1062 u64 map_seq;
1063
1064 skb = skb_peek(&ssk->sk_receive_queue);
1065 if (!skb)
1066 return MAPPING_EMPTY;
1067
1068 if (mptcp_check_fallback(ssk))
1069 return MAPPING_DUMMY;
1070
1071 mpext = mptcp_get_ext(skb);
1072 if (!mpext || !mpext->use_map) {
1073 if (!subflow->map_valid && !skb->len) {
1074 /* the TCP stack deliver 0 len FIN pkt to the receive
1075 * queue, that is the only 0len pkts ever expected here,
1076 * and we can admit no mapping only for 0 len pkts
1077 */
1078 if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN))
1079 WARN_ONCE(1, "0len seq %d:%d flags %x",
1080 TCP_SKB_CB(skb)->seq,
1081 TCP_SKB_CB(skb)->end_seq,
1082 TCP_SKB_CB(skb)->tcp_flags);
1083 sk_eat_skb(ssk, skb);
1084 return MAPPING_EMPTY;
1085 }
1086
1087 if (!subflow->map_valid)
1088 return MAPPING_INVALID;
1089
1090 goto validate_seq;
1091 }
1092
1093 trace_get_mapping_status(mpext);
1094
1095 data_len = mpext->data_len;
1096 if (data_len == 0) {
1097 pr_debug("infinite mapping received\n");
1098 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_INFINITEMAPRX);
1099 subflow->map_data_len = 0;
1100 return MAPPING_INVALID;
1101 }
1102
1103 if (mpext->data_fin == 1) {
1104 if (data_len == 1) {
1105 bool updated = mptcp_update_rcv_data_fin(msk, mpext->data_seq,
1106 mpext->dsn64);
1107 pr_debug("DATA_FIN with no payload seq=%llu\n", mpext->data_seq);
1108 if (subflow->map_valid) {
1109 /* A DATA_FIN might arrive in a DSS
1110 * option before the previous mapping
1111 * has been fully consumed. Continue
1112 * handling the existing mapping.
1113 */
1114 skb_ext_del(skb, SKB_EXT_MPTCP);
1115 return MAPPING_OK;
1116 } else {
1117 if (updated)
1118 mptcp_schedule_work((struct sock *)msk);
1119
1120 return MAPPING_DATA_FIN;
1121 }
1122 } else {
1123 u64 data_fin_seq = mpext->data_seq + data_len - 1;
1124
1125 /* If mpext->data_seq is a 32-bit value, data_fin_seq
1126 * must also be limited to 32 bits.
1127 */
1128 if (!mpext->dsn64)
1129 data_fin_seq &= GENMASK_ULL(31, 0);
1130
1131 mptcp_update_rcv_data_fin(msk, data_fin_seq, mpext->dsn64);
1132 pr_debug("DATA_FIN with mapping seq=%llu dsn64=%d\n",
1133 data_fin_seq, mpext->dsn64);
1134 }
1135
1136 /* Adjust for DATA_FIN using 1 byte of sequence space */
1137 data_len--;
1138 }
1139
1140 map_seq = mptcp_expand_seq(READ_ONCE(msk->ack_seq), mpext->data_seq, mpext->dsn64);
1141 WRITE_ONCE(mptcp_sk(subflow->conn)->use_64bit_ack, !!mpext->dsn64);
1142
1143 if (subflow->map_valid) {
1144 /* Allow replacing only with an identical map */
1145 if (subflow->map_seq == map_seq &&
1146 subflow->map_subflow_seq == mpext->subflow_seq &&
1147 subflow->map_data_len == data_len &&
1148 subflow->map_csum_reqd == mpext->csum_reqd) {
1149 skb_ext_del(skb, SKB_EXT_MPTCP);
1150 goto validate_csum;
1151 }
1152
1153 /* If this skb data are fully covered by the current mapping,
1154 * the new map would need caching, which is not supported
1155 */
1156 if (skb_is_fully_mapped(ssk, skb)) {
1157 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DSSNOMATCH);
1158 return MAPPING_INVALID;
1159 }
1160
1161 /* will validate the next map after consuming the current one */
1162 goto validate_csum;
1163 }
1164
1165 subflow->map_seq = map_seq;
1166 subflow->map_subflow_seq = mpext->subflow_seq;
1167 subflow->map_data_len = data_len;
1168 subflow->map_valid = 1;
1169 subflow->map_data_fin = mpext->data_fin;
1170 subflow->mpc_map = mpext->mpc_map;
1171 subflow->map_csum_reqd = mpext->csum_reqd;
1172 subflow->map_csum_len = 0;
1173 subflow->map_data_csum = csum_unfold(mpext->csum);
1174
1175 /* Cfr RFC 8684 Section 3.3.0 */
1176 if (unlikely(subflow->map_csum_reqd != csum_reqd))
1177 return MAPPING_INVALID;
1178
1179 pr_debug("new map seq=%llu subflow_seq=%u data_len=%u csum=%d:%u\n",
1180 subflow->map_seq, subflow->map_subflow_seq,
1181 subflow->map_data_len, subflow->map_csum_reqd,
1182 subflow->map_data_csum);
1183
1184 validate_seq:
1185 /* we revalidate valid mapping on new skb, because we must ensure
1186 * the current skb is completely covered by the available mapping
1187 */
1188 if (!validate_mapping(ssk, skb)) {
1189 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DSSTCPMISMATCH);
1190 return MAPPING_INVALID;
1191 }
1192
1193 skb_ext_del(skb, SKB_EXT_MPTCP);
1194
1195 validate_csum:
1196 return validate_data_csum(ssk, skb, csum_reqd);
1197 }
1198
mptcp_subflow_discard_data(struct sock * ssk,struct sk_buff * skb,u64 limit)1199 static void mptcp_subflow_discard_data(struct sock *ssk, struct sk_buff *skb,
1200 u64 limit)
1201 {
1202 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1203 bool fin = TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN;
1204 struct tcp_sock *tp = tcp_sk(ssk);
1205 u32 offset, incr, avail_len;
1206
1207 offset = tp->copied_seq - TCP_SKB_CB(skb)->seq;
1208 if (WARN_ON_ONCE(offset > skb->len))
1209 goto out;
1210
1211 avail_len = skb->len - offset;
1212 incr = limit >= avail_len ? avail_len + fin : limit;
1213
1214 pr_debug("discarding=%d len=%d offset=%d seq=%d\n", incr, skb->len,
1215 offset, subflow->map_subflow_seq);
1216 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DUPDATA);
1217 tcp_sk(ssk)->copied_seq += incr;
1218
1219 out:
1220 if (!before(tcp_sk(ssk)->copied_seq, TCP_SKB_CB(skb)->end_seq))
1221 sk_eat_skb(ssk, skb);
1222 if (mptcp_subflow_get_map_offset(subflow) >= subflow->map_data_len)
1223 subflow->map_valid = 0;
1224 }
1225
1226 /* 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)1227 static void subflow_sched_work_if_closed(struct mptcp_sock *msk, struct sock *ssk)
1228 {
1229 struct sock *sk = (struct sock *)msk;
1230
1231 if (likely(ssk->sk_state != TCP_CLOSE &&
1232 (ssk->sk_state != TCP_CLOSE_WAIT ||
1233 inet_sk_state_load(sk) != TCP_ESTABLISHED)))
1234 return;
1235
1236 if (skb_queue_empty(&ssk->sk_receive_queue) &&
1237 !test_and_set_bit(MPTCP_WORK_CLOSE_SUBFLOW, &msk->flags))
1238 mptcp_schedule_work(sk);
1239 }
1240
subflow_can_fallback(struct mptcp_subflow_context * subflow)1241 static bool subflow_can_fallback(struct mptcp_subflow_context *subflow)
1242 {
1243 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
1244
1245 if (subflow->mp_join)
1246 return false;
1247 else if (READ_ONCE(msk->csum_enabled))
1248 return !subflow->valid_csum_seen;
1249 else
1250 return !subflow->fully_established;
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 (!subflow_can_fallback(subflow) && subflow->map_data_len) {
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