1 // SPDX-License-Identifier: GPL-2.0
2 /* Multipath TCP
3 *
4 * Copyright (c) 2017 - 2019, Intel Corporation.
5 */
6
7 #define pr_fmt(fmt) "MPTCP: " fmt
8
9 #include <linux/kernel.h>
10 #include <linux/module.h>
11 #include <linux/netdevice.h>
12 #include <crypto/sha2.h>
13 #include <crypto/utils.h>
14 #include <net/sock.h>
15 #include <net/inet_common.h>
16 #include <net/inet_hashtables.h>
17 #include <net/protocol.h>
18 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
19 #include <net/ip6_route.h>
20 #include <net/transp_v6.h>
21 #endif
22 #include <net/mptcp.h>
23
24 #include "protocol.h"
25 #include "mib.h"
26
27 #include <trace/events/mptcp.h>
28 #include <trace/events/sock.h>
29
30 static void mptcp_subflow_ops_undo_override(struct sock *ssk);
31
SUBFLOW_REQ_INC_STATS(struct request_sock * req,enum linux_mptcp_mib_field field)32 static void SUBFLOW_REQ_INC_STATS(struct request_sock *req,
33 enum linux_mptcp_mib_field field)
34 {
35 MPTCP_INC_STATS(sock_net(req_to_sk(req)), field);
36 }
37
subflow_req_destructor(struct request_sock * req)38 static void subflow_req_destructor(struct request_sock *req)
39 {
40 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
41
42 pr_debug("subflow_req=%p\n", subflow_req);
43
44 if (subflow_req->msk)
45 sock_put((struct sock *)subflow_req->msk);
46
47 mptcp_token_destroy_request(req);
48 }
49
subflow_generate_hmac(u64 key1,u64 key2,u32 nonce1,u32 nonce2,void * hmac)50 static void subflow_generate_hmac(u64 key1, u64 key2, u32 nonce1, u32 nonce2,
51 void *hmac)
52 {
53 u8 msg[8];
54
55 put_unaligned_be32(nonce1, &msg[0]);
56 put_unaligned_be32(nonce2, &msg[4]);
57
58 mptcp_crypto_hmac_sha(key1, key2, msg, 8, hmac);
59 }
60
mptcp_can_accept_new_subflow(const struct mptcp_sock * msk)61 static bool mptcp_can_accept_new_subflow(const struct mptcp_sock *msk)
62 {
63 return mptcp_is_fully_established((void *)msk) &&
64 ((mptcp_pm_is_userspace(msk) &&
65 mptcp_userspace_pm_active(msk)) ||
66 READ_ONCE(msk->pm.accept_subflow));
67 }
68
69 /* validate received token and create truncated hmac and nonce for SYN-ACK */
subflow_req_create_thmac(struct mptcp_subflow_request_sock * subflow_req)70 static void subflow_req_create_thmac(struct mptcp_subflow_request_sock *subflow_req)
71 {
72 struct mptcp_sock *msk = subflow_req->msk;
73 u8 hmac[SHA256_DIGEST_SIZE];
74
75 get_random_bytes(&subflow_req->local_nonce, sizeof(u32));
76
77 subflow_generate_hmac(READ_ONCE(msk->local_key),
78 READ_ONCE(msk->remote_key),
79 subflow_req->local_nonce,
80 subflow_req->remote_nonce, hmac);
81
82 subflow_req->thmac = get_unaligned_be64(hmac);
83 }
84
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 subflow_add_reset_reason(skb, MPTCP_RST_EMPTCP);
164 return -EINVAL;
165 }
166 #endif
167
168 mptcp_get_options(skb, &mp_opt);
169
170 opt_mp_capable = !!(mp_opt.suboptions & OPTION_MPTCP_MPC_SYN);
171 opt_mp_join = !!(mp_opt.suboptions & OPTION_MPTCP_MPJ_SYN);
172 if (opt_mp_capable) {
173 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MPCAPABLEPASSIVE);
174
175 if (unlikely(listener->pm_listener))
176 return subflow_reset_req_endp(req, skb);
177 if (opt_mp_join)
178 return 0;
179 } else if (opt_mp_join) {
180 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINSYNRX);
181
182 if (mp_opt.backup)
183 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINSYNBACKUPRX);
184 } else if (unlikely(listener->pm_listener)) {
185 return subflow_reset_req_endp(req, skb);
186 }
187
188 if (opt_mp_capable && listener->request_mptcp) {
189 int err, retries = MPTCP_TOKEN_MAX_RETRIES;
190
191 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq;
192 again:
193 do {
194 get_random_bytes(&subflow_req->local_key, sizeof(subflow_req->local_key));
195 } while (subflow_req->local_key == 0);
196
197 if (unlikely(req->syncookie)) {
198 mptcp_crypto_key_sha(subflow_req->local_key,
199 &subflow_req->token,
200 &subflow_req->idsn);
201 if (mptcp_token_exists(subflow_req->token)) {
202 if (retries-- > 0)
203 goto again;
204 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_TOKENFALLBACKINIT);
205 } else {
206 subflow_req->mp_capable = 1;
207 }
208 return 0;
209 }
210
211 err = mptcp_token_new_request(req);
212 if (err == 0)
213 subflow_req->mp_capable = 1;
214 else if (retries-- > 0)
215 goto again;
216 else
217 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_TOKENFALLBACKINIT);
218
219 } else if (opt_mp_join && listener->request_mptcp) {
220 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq;
221 subflow_req->mp_join = 1;
222 subflow_req->backup = mp_opt.backup;
223 subflow_req->remote_id = mp_opt.join_id;
224 subflow_req->token = mp_opt.token;
225 subflow_req->remote_nonce = mp_opt.nonce;
226 subflow_req->msk = subflow_token_join_request(req);
227
228 /* Can't fall back to TCP in this case. */
229 if (!subflow_req->msk) {
230 subflow_add_reset_reason(skb, MPTCP_RST_EMPTCP);
231 return -EPERM;
232 }
233
234 if (subflow_use_different_sport(subflow_req->msk, sk_listener)) {
235 pr_debug("syn inet_sport=%d %d\n",
236 ntohs(inet_sk(sk_listener)->inet_sport),
237 ntohs(inet_sk((struct sock *)subflow_req->msk)->inet_sport));
238 if (!mptcp_pm_sport_in_anno_list(subflow_req->msk, sk_listener)) {
239 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MISMATCHPORTSYNRX);
240 subflow_add_reset_reason(skb, MPTCP_RST_EPROHIBIT);
241 return -EPERM;
242 }
243 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINPORTSYNRX);
244 }
245
246 subflow_req_create_thmac(subflow_req);
247
248 if (unlikely(req->syncookie)) {
249 if (!mptcp_can_accept_new_subflow(subflow_req->msk)) {
250 subflow_add_reset_reason(skb, MPTCP_RST_EPROHIBIT);
251 return -EPERM;
252 }
253
254 subflow_init_req_cookie_join_save(subflow_req, skb);
255 }
256
257 pr_debug("token=%u, remote_nonce=%u msk=%p\n", subflow_req->token,
258 subflow_req->remote_nonce, subflow_req->msk);
259 }
260
261 return 0;
262 }
263
mptcp_subflow_init_cookie_req(struct request_sock * req,const struct sock * sk_listener,struct sk_buff * skb)264 int mptcp_subflow_init_cookie_req(struct request_sock *req,
265 const struct sock *sk_listener,
266 struct sk_buff *skb)
267 {
268 struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk_listener);
269 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
270 struct mptcp_options_received mp_opt;
271 bool opt_mp_capable, opt_mp_join;
272 int err;
273
274 subflow_init_req(req, sk_listener);
275 mptcp_get_options(skb, &mp_opt);
276
277 opt_mp_capable = !!(mp_opt.suboptions & OPTION_MPTCP_MPC_ACK);
278 opt_mp_join = !!(mp_opt.suboptions & OPTION_MPTCP_MPJ_ACK);
279 if (opt_mp_capable && opt_mp_join)
280 return -EINVAL;
281
282 if (opt_mp_capable && listener->request_mptcp) {
283 if (mp_opt.sndr_key == 0)
284 return -EINVAL;
285
286 subflow_req->local_key = mp_opt.rcvr_key;
287 err = mptcp_token_new_request(req);
288 if (err)
289 return err;
290
291 subflow_req->mp_capable = 1;
292 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq - 1;
293 } else if (opt_mp_join && listener->request_mptcp) {
294 if (!mptcp_token_join_cookie_init_state(subflow_req, skb))
295 return -EINVAL;
296
297 subflow_req->mp_join = 1;
298 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq - 1;
299 }
300
301 return 0;
302 }
303 EXPORT_SYMBOL_GPL(mptcp_subflow_init_cookie_req);
304
mptcp_get_rst_reason(const struct sk_buff * skb)305 static enum sk_rst_reason mptcp_get_rst_reason(const struct sk_buff *skb)
306 {
307 const struct mptcp_ext *mpext = mptcp_get_ext(skb);
308
309 if (!mpext)
310 return SK_RST_REASON_NOT_SPECIFIED;
311
312 return sk_rst_convert_mptcp_reason(mpext->reset_reason);
313 }
314
subflow_v4_route_req(const struct sock * sk,struct sk_buff * skb,struct flowi * fl,struct request_sock * req,u32 tw_isn)315 static struct dst_entry *subflow_v4_route_req(const struct sock *sk,
316 struct sk_buff *skb,
317 struct flowi *fl,
318 struct request_sock *req,
319 u32 tw_isn)
320 {
321 struct dst_entry *dst;
322 int err;
323
324 tcp_rsk(req)->is_mptcp = 1;
325 subflow_init_req(req, sk);
326
327 dst = tcp_request_sock_ipv4_ops.route_req(sk, skb, fl, req, tw_isn);
328 if (!dst)
329 return NULL;
330
331 err = subflow_check_req(req, sk, skb);
332 if (err == 0)
333 return dst;
334
335 dst_release(dst);
336 if (!req->syncookie)
337 tcp_request_sock_ops.send_reset(sk, skb,
338 mptcp_get_rst_reason(skb));
339 return NULL;
340 }
341
subflow_prep_synack(const struct sock * sk,struct request_sock * req,struct tcp_fastopen_cookie * foc,enum tcp_synack_type synack_type)342 static void subflow_prep_synack(const struct sock *sk, struct request_sock *req,
343 struct tcp_fastopen_cookie *foc,
344 enum tcp_synack_type synack_type)
345 {
346 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
347 struct inet_request_sock *ireq = inet_rsk(req);
348
349 /* clear tstamp_ok, as needed depending on cookie */
350 if (foc && foc->len > -1)
351 ireq->tstamp_ok = 0;
352
353 if (synack_type == TCP_SYNACK_FASTOPEN)
354 mptcp_fastopen_subflow_synack_set_params(subflow, req);
355 }
356
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)357 static int subflow_v4_send_synack(const struct sock *sk, struct dst_entry *dst,
358 struct flowi *fl,
359 struct request_sock *req,
360 struct tcp_fastopen_cookie *foc,
361 enum tcp_synack_type synack_type,
362 struct sk_buff *syn_skb)
363 {
364 subflow_prep_synack(sk, req, foc, synack_type);
365
366 return tcp_request_sock_ipv4_ops.send_synack(sk, dst, fl, req, foc,
367 synack_type, syn_skb);
368 }
369
370 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
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)371 static int subflow_v6_send_synack(const struct sock *sk, struct dst_entry *dst,
372 struct flowi *fl,
373 struct request_sock *req,
374 struct tcp_fastopen_cookie *foc,
375 enum tcp_synack_type synack_type,
376 struct sk_buff *syn_skb)
377 {
378 subflow_prep_synack(sk, req, foc, synack_type);
379
380 return tcp_request_sock_ipv6_ops.send_synack(sk, dst, fl, req, foc,
381 synack_type, syn_skb);
382 }
383
subflow_v6_route_req(const struct sock * sk,struct sk_buff * skb,struct flowi * fl,struct request_sock * req,u32 tw_isn)384 static struct dst_entry *subflow_v6_route_req(const struct sock *sk,
385 struct sk_buff *skb,
386 struct flowi *fl,
387 struct request_sock *req,
388 u32 tw_isn)
389 {
390 struct dst_entry *dst;
391 int err;
392
393 tcp_rsk(req)->is_mptcp = 1;
394 subflow_init_req(req, sk);
395
396 dst = tcp_request_sock_ipv6_ops.route_req(sk, skb, fl, req, tw_isn);
397 if (!dst)
398 return NULL;
399
400 err = subflow_check_req(req, sk, skb);
401 if (err == 0)
402 return dst;
403
404 dst_release(dst);
405 if (!req->syncookie)
406 tcp6_request_sock_ops.send_reset(sk, skb,
407 mptcp_get_rst_reason(skb));
408 return NULL;
409 }
410 #endif
411
412 /* validate received truncated hmac and create hmac for third ACK */
subflow_thmac_valid(struct mptcp_subflow_context * subflow)413 static bool subflow_thmac_valid(struct mptcp_subflow_context *subflow)
414 {
415 u8 hmac[SHA256_DIGEST_SIZE];
416 u64 thmac;
417
418 subflow_generate_hmac(subflow->remote_key, subflow->local_key,
419 subflow->remote_nonce, subflow->local_nonce,
420 hmac);
421
422 thmac = get_unaligned_be64(hmac);
423 pr_debug("subflow=%p, token=%u, thmac=%llu, subflow->thmac=%llu\n",
424 subflow, subflow->token, thmac, subflow->thmac);
425
426 return thmac == subflow->thmac;
427 }
428
mptcp_subflow_reset(struct sock * ssk)429 void mptcp_subflow_reset(struct sock *ssk)
430 {
431 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
432 struct sock *sk = subflow->conn;
433
434 /* mptcp_mp_fail_no_response() can reach here on an already closed
435 * socket
436 */
437 if (ssk->sk_state == TCP_CLOSE)
438 return;
439
440 /* must hold: tcp_done() could drop last reference on parent */
441 sock_hold(sk);
442
443 mptcp_send_active_reset_reason(ssk);
444 tcp_done(ssk);
445 if (!test_and_set_bit(MPTCP_WORK_CLOSE_SUBFLOW, &mptcp_sk(sk)->flags))
446 mptcp_schedule_work(sk);
447
448 sock_put(sk);
449 }
450
subflow_use_different_dport(struct mptcp_sock * msk,const struct sock * sk)451 static bool subflow_use_different_dport(struct mptcp_sock *msk, const struct sock *sk)
452 {
453 return inet_sk(sk)->inet_dport != inet_sk((struct sock *)msk)->inet_dport;
454 }
455
__mptcp_sync_state(struct sock * sk,int state)456 void __mptcp_sync_state(struct sock *sk, int state)
457 {
458 struct mptcp_subflow_context *subflow;
459 struct mptcp_sock *msk = mptcp_sk(sk);
460 struct sock *ssk = msk->first;
461
462 subflow = mptcp_subflow_ctx(ssk);
463 __mptcp_propagate_sndbuf(sk, ssk);
464 if (!msk->rcvspace_init)
465 mptcp_rcv_space_init(msk, ssk);
466
467 if (sk->sk_state == TCP_SYN_SENT) {
468 /* subflow->idsn is always available is TCP_SYN_SENT state,
469 * even for the FASTOPEN scenarios
470 */
471 WRITE_ONCE(msk->write_seq, subflow->idsn + 1);
472 WRITE_ONCE(msk->snd_nxt, msk->write_seq);
473 mptcp_set_state(sk, state);
474 sk->sk_state_change(sk);
475 }
476 }
477
subflow_set_remote_key(struct mptcp_sock * msk,struct mptcp_subflow_context * subflow,const struct mptcp_options_received * mp_opt)478 static void subflow_set_remote_key(struct mptcp_sock *msk,
479 struct mptcp_subflow_context *subflow,
480 const struct mptcp_options_received *mp_opt)
481 {
482 /* active MPC subflow will reach here multiple times:
483 * at subflow_finish_connect() time and at 4th ack time
484 */
485 if (subflow->remote_key_valid)
486 return;
487
488 subflow->remote_key_valid = 1;
489 subflow->remote_key = mp_opt->sndr_key;
490 mptcp_crypto_key_sha(subflow->remote_key, NULL, &subflow->iasn);
491 subflow->iasn++;
492
493 WRITE_ONCE(msk->remote_key, subflow->remote_key);
494 WRITE_ONCE(msk->ack_seq, subflow->iasn);
495 WRITE_ONCE(msk->can_ack, true);
496 atomic64_set(&msk->rcv_wnd_sent, subflow->iasn);
497 }
498
mptcp_propagate_state(struct sock * sk,struct sock * ssk,struct mptcp_subflow_context * subflow,const struct mptcp_options_received * mp_opt)499 static void mptcp_propagate_state(struct sock *sk, struct sock *ssk,
500 struct mptcp_subflow_context *subflow,
501 const struct mptcp_options_received *mp_opt)
502 {
503 struct mptcp_sock *msk = mptcp_sk(sk);
504
505 mptcp_data_lock(sk);
506 if (mp_opt) {
507 /* Options are available only in the non fallback cases
508 * avoid updating rx path fields otherwise
509 */
510 WRITE_ONCE(msk->snd_una, subflow->idsn + 1);
511 WRITE_ONCE(msk->wnd_end, subflow->idsn + 1 + tcp_sk(ssk)->snd_wnd);
512 subflow_set_remote_key(msk, subflow, mp_opt);
513 }
514
515 if (!sock_owned_by_user(sk)) {
516 __mptcp_sync_state(sk, ssk->sk_state);
517 } else {
518 msk->pending_state = ssk->sk_state;
519 __set_bit(MPTCP_SYNC_STATE, &msk->cb_flags);
520 }
521 mptcp_data_unlock(sk);
522 }
523
subflow_finish_connect(struct sock * sk,const struct sk_buff * skb)524 static void subflow_finish_connect(struct sock *sk, const struct sk_buff *skb)
525 {
526 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
527 struct mptcp_options_received mp_opt;
528 struct sock *parent = subflow->conn;
529 struct mptcp_sock *msk;
530
531 subflow->icsk_af_ops->sk_rx_dst_set(sk, skb);
532
533 /* be sure no special action on any packet other than syn-ack */
534 if (subflow->conn_finished)
535 return;
536
537 msk = mptcp_sk(parent);
538 subflow->rel_write_seq = 1;
539 subflow->conn_finished = 1;
540 subflow->ssn_offset = TCP_SKB_CB(skb)->seq;
541 pr_debug("subflow=%p synack seq=%x\n", subflow, subflow->ssn_offset);
542
543 mptcp_get_options(skb, &mp_opt);
544 if (subflow->request_mptcp) {
545 if (!(mp_opt.suboptions & OPTION_MPTCP_MPC_SYNACK)) {
546 if (!mptcp_try_fallback(sk))
547 goto do_reset;
548
549 MPTCP_INC_STATS(sock_net(sk),
550 MPTCP_MIB_MPCAPABLEACTIVEFALLBACK);
551 pr_fallback(msk);
552 goto fallback;
553 }
554
555 if (mp_opt.suboptions & OPTION_MPTCP_CSUMREQD)
556 WRITE_ONCE(msk->csum_enabled, true);
557 if (mp_opt.deny_join_id0)
558 WRITE_ONCE(msk->pm.remote_deny_join_id0, true);
559 subflow->mp_capable = 1;
560 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPCAPABLEACTIVEACK);
561 mptcp_finish_connect(sk);
562 mptcp_active_enable(parent);
563 mptcp_propagate_state(parent, sk, subflow, &mp_opt);
564 } else if (subflow->request_join) {
565 u8 hmac[SHA256_DIGEST_SIZE];
566
567 if (!(mp_opt.suboptions & OPTION_MPTCP_MPJ_SYNACK)) {
568 subflow->reset_reason = MPTCP_RST_EMPTCP;
569 goto do_reset;
570 }
571
572 subflow->backup = mp_opt.backup;
573 subflow->thmac = mp_opt.thmac;
574 subflow->remote_nonce = mp_opt.nonce;
575 WRITE_ONCE(subflow->remote_id, mp_opt.join_id);
576 pr_debug("subflow=%p, thmac=%llu, remote_nonce=%u backup=%d\n",
577 subflow, subflow->thmac, subflow->remote_nonce,
578 subflow->backup);
579
580 if (!subflow_thmac_valid(subflow)) {
581 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINACKMAC);
582 subflow->reset_reason = MPTCP_RST_EMPTCP;
583 goto do_reset;
584 }
585
586 if (!mptcp_finish_join(sk))
587 goto do_reset;
588
589 subflow_generate_hmac(subflow->local_key, subflow->remote_key,
590 subflow->local_nonce,
591 subflow->remote_nonce,
592 hmac);
593 memcpy(subflow->hmac, hmac, MPTCPOPT_HMAC_LEN);
594
595 subflow->mp_join = 1;
596 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINSYNACKRX);
597
598 if (subflow->backup)
599 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINSYNACKBACKUPRX);
600
601 if (subflow_use_different_dport(msk, sk)) {
602 pr_debug("synack inet_dport=%d %d\n",
603 ntohs(inet_sk(sk)->inet_dport),
604 ntohs(inet_sk(parent)->inet_dport));
605 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINPORTSYNACKRX);
606 }
607 } else if (mptcp_check_fallback(sk)) {
608 /* It looks like MPTCP is blocked, while TCP is not */
609 if (subflow->mpc_drop)
610 mptcp_active_disable(parent);
611 fallback:
612 mptcp_propagate_state(parent, sk, subflow, NULL);
613 }
614 return;
615
616 do_reset:
617 subflow->reset_transient = 0;
618 mptcp_subflow_reset(sk);
619 }
620
subflow_set_local_id(struct mptcp_subflow_context * subflow,int local_id)621 static void subflow_set_local_id(struct mptcp_subflow_context *subflow, int local_id)
622 {
623 WARN_ON_ONCE(local_id < 0 || local_id > 255);
624 WRITE_ONCE(subflow->local_id, local_id);
625 }
626
subflow_chk_local_id(struct sock * sk)627 static int subflow_chk_local_id(struct sock *sk)
628 {
629 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
630 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
631 int err;
632
633 if (likely(subflow->local_id >= 0))
634 return 0;
635
636 err = mptcp_pm_get_local_id(msk, (struct sock_common *)sk);
637 if (err < 0)
638 return err;
639
640 subflow_set_local_id(subflow, err);
641 subflow->request_bkup = mptcp_pm_is_backup(msk, (struct sock_common *)sk);
642
643 return 0;
644 }
645
subflow_rebuild_header(struct sock * sk)646 static int subflow_rebuild_header(struct sock *sk)
647 {
648 int err = subflow_chk_local_id(sk);
649
650 if (unlikely(err < 0))
651 return err;
652
653 return inet_sk_rebuild_header(sk);
654 }
655
656 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
subflow_v6_rebuild_header(struct sock * sk)657 static int subflow_v6_rebuild_header(struct sock *sk)
658 {
659 int err = subflow_chk_local_id(sk);
660
661 if (unlikely(err < 0))
662 return err;
663
664 return inet6_sk_rebuild_header(sk);
665 }
666 #endif
667
668 static struct request_sock_ops mptcp_subflow_v4_request_sock_ops __ro_after_init;
669 static struct tcp_request_sock_ops subflow_request_sock_ipv4_ops __ro_after_init;
670
subflow_v4_conn_request(struct sock * sk,struct sk_buff * skb)671 static int subflow_v4_conn_request(struct sock *sk, struct sk_buff *skb)
672 {
673 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
674
675 pr_debug("subflow=%p\n", subflow);
676
677 /* Never answer to SYNs sent to broadcast or multicast */
678 if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
679 goto drop;
680
681 return tcp_conn_request(&mptcp_subflow_v4_request_sock_ops,
682 &subflow_request_sock_ipv4_ops,
683 sk, skb);
684 drop:
685 tcp_listendrop(sk);
686 return 0;
687 }
688
subflow_v4_req_destructor(struct request_sock * req)689 static void subflow_v4_req_destructor(struct request_sock *req)
690 {
691 subflow_req_destructor(req);
692 tcp_request_sock_ops.destructor(req);
693 }
694
695 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
696 static struct request_sock_ops mptcp_subflow_v6_request_sock_ops __ro_after_init;
697 static struct tcp_request_sock_ops subflow_request_sock_ipv6_ops __ro_after_init;
698 static struct inet_connection_sock_af_ops subflow_v6_specific __ro_after_init;
699 static struct inet_connection_sock_af_ops subflow_v6m_specific __ro_after_init;
700 static struct proto tcpv6_prot_override __ro_after_init;
701
subflow_v6_conn_request(struct sock * sk,struct sk_buff * skb)702 static int subflow_v6_conn_request(struct sock *sk, struct sk_buff *skb)
703 {
704 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
705
706 pr_debug("subflow=%p\n", subflow);
707
708 if (skb->protocol == htons(ETH_P_IP))
709 return subflow_v4_conn_request(sk, skb);
710
711 if (!ipv6_unicast_destination(skb))
712 goto drop;
713
714 if (ipv6_addr_v4mapped(&ipv6_hdr(skb)->saddr)) {
715 __IP6_INC_STATS(sock_net(sk), NULL, IPSTATS_MIB_INHDRERRORS);
716 return 0;
717 }
718
719 return tcp_conn_request(&mptcp_subflow_v6_request_sock_ops,
720 &subflow_request_sock_ipv6_ops, sk, skb);
721
722 drop:
723 tcp_listendrop(sk);
724 return 0; /* don't send reset */
725 }
726
subflow_v6_req_destructor(struct request_sock * req)727 static void subflow_v6_req_destructor(struct request_sock *req)
728 {
729 subflow_req_destructor(req);
730 tcp6_request_sock_ops.destructor(req);
731 }
732 #endif
733
mptcp_subflow_reqsk_alloc(const struct request_sock_ops * ops,struct sock * sk_listener,bool attach_listener)734 struct request_sock *mptcp_subflow_reqsk_alloc(const struct request_sock_ops *ops,
735 struct sock *sk_listener,
736 bool attach_listener)
737 {
738 if (ops->family == AF_INET)
739 ops = &mptcp_subflow_v4_request_sock_ops;
740 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
741 else if (ops->family == AF_INET6)
742 ops = &mptcp_subflow_v6_request_sock_ops;
743 #endif
744
745 return inet_reqsk_alloc(ops, sk_listener, attach_listener);
746 }
747 EXPORT_SYMBOL(mptcp_subflow_reqsk_alloc);
748
749 /* validate hmac received in third ACK */
subflow_hmac_valid(const struct request_sock * req,const struct mptcp_options_received * mp_opt)750 static bool subflow_hmac_valid(const struct request_sock *req,
751 const struct mptcp_options_received *mp_opt)
752 {
753 const struct mptcp_subflow_request_sock *subflow_req;
754 u8 hmac[SHA256_DIGEST_SIZE];
755 struct mptcp_sock *msk;
756
757 subflow_req = mptcp_subflow_rsk(req);
758 msk = subflow_req->msk;
759
760 subflow_generate_hmac(READ_ONCE(msk->remote_key),
761 READ_ONCE(msk->local_key),
762 subflow_req->remote_nonce,
763 subflow_req->local_nonce, hmac);
764
765 return !crypto_memneq(hmac, mp_opt->hmac, MPTCPOPT_HMAC_LEN);
766 }
767
subflow_ulp_fallback(struct sock * sk,struct mptcp_subflow_context * old_ctx)768 static void subflow_ulp_fallback(struct sock *sk,
769 struct mptcp_subflow_context *old_ctx)
770 {
771 struct inet_connection_sock *icsk = inet_csk(sk);
772
773 mptcp_subflow_tcp_fallback(sk, old_ctx);
774 icsk->icsk_ulp_ops = NULL;
775 rcu_assign_pointer(icsk->icsk_ulp_data, NULL);
776 tcp_sk(sk)->is_mptcp = 0;
777
778 mptcp_subflow_ops_undo_override(sk);
779 }
780
mptcp_subflow_drop_ctx(struct sock * ssk)781 void mptcp_subflow_drop_ctx(struct sock *ssk)
782 {
783 struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(ssk);
784
785 if (!ctx)
786 return;
787
788 list_del(&mptcp_subflow_ctx(ssk)->node);
789 if (inet_csk(ssk)->icsk_ulp_ops) {
790 subflow_ulp_fallback(ssk, ctx);
791 if (ctx->conn)
792 sock_put(ctx->conn);
793 }
794
795 kfree_rcu(ctx, rcu);
796 }
797
__mptcp_subflow_fully_established(struct mptcp_sock * msk,struct mptcp_subflow_context * subflow,const struct mptcp_options_received * mp_opt)798 void __mptcp_subflow_fully_established(struct mptcp_sock *msk,
799 struct mptcp_subflow_context *subflow,
800 const struct mptcp_options_received *mp_opt)
801 {
802 subflow_set_remote_key(msk, subflow, mp_opt);
803 WRITE_ONCE(subflow->fully_established, true);
804 WRITE_ONCE(msk->fully_established, true);
805
806 if (subflow->is_mptfo)
807 __mptcp_fastopen_gen_msk_ackseq(msk, subflow, mp_opt);
808 }
809
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)810 static struct sock *subflow_syn_recv_sock(const struct sock *sk,
811 struct sk_buff *skb,
812 struct request_sock *req,
813 struct dst_entry *dst,
814 struct request_sock *req_unhash,
815 bool *own_req)
816 {
817 struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk);
818 struct mptcp_subflow_request_sock *subflow_req;
819 struct mptcp_options_received mp_opt;
820 bool fallback, fallback_is_fatal;
821 enum sk_rst_reason reason;
822 struct mptcp_sock *owner;
823 struct sock *child;
824
825 pr_debug("listener=%p, req=%p, conn=%p\n", listener, req, listener->conn);
826
827 /* After child creation we must look for MPC even when options
828 * are not parsed
829 */
830 mp_opt.suboptions = 0;
831
832 /* hopefully temporary handling for MP_JOIN+syncookie */
833 subflow_req = mptcp_subflow_rsk(req);
834 fallback_is_fatal = tcp_rsk(req)->is_mptcp && subflow_req->mp_join;
835 fallback = !tcp_rsk(req)->is_mptcp;
836 if (fallback)
837 goto create_child;
838
839 /* if the sk is MP_CAPABLE, we try to fetch the client key */
840 if (subflow_req->mp_capable) {
841 /* we can receive and accept an in-window, out-of-order pkt,
842 * which may not carry the MP_CAPABLE opt even on mptcp enabled
843 * paths: always try to extract the peer key, and fallback
844 * for packets missing it.
845 * Even OoO DSS packets coming legitly after dropped or
846 * reordered MPC will cause fallback, but we don't have other
847 * options.
848 */
849 mptcp_get_options(skb, &mp_opt);
850 if (!(mp_opt.suboptions &
851 (OPTION_MPTCP_MPC_SYN | OPTION_MPTCP_MPC_ACK)))
852 fallback = true;
853
854 } else if (subflow_req->mp_join) {
855 mptcp_get_options(skb, &mp_opt);
856 if (!(mp_opt.suboptions & OPTION_MPTCP_MPJ_ACK))
857 fallback = true;
858 }
859
860 create_child:
861 child = listener->icsk_af_ops->syn_recv_sock(sk, skb, req, dst,
862 req_unhash, own_req);
863
864 if (child && *own_req) {
865 struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(child);
866
867 tcp_rsk(req)->drop_req = false;
868
869 /* we need to fallback on ctx allocation failure and on pre-reqs
870 * checking above. In the latter scenario we additionally need
871 * to reset the context to non MPTCP status.
872 */
873 if (!ctx || fallback) {
874 if (fallback_is_fatal) {
875 subflow_add_reset_reason(skb, MPTCP_RST_EMPTCP);
876 goto dispose_child;
877 }
878 goto fallback;
879 }
880
881 /* ssk inherits options of listener sk */
882 ctx->setsockopt_seq = listener->setsockopt_seq;
883
884 if (ctx->mp_capable) {
885 ctx->conn = mptcp_sk_clone_init(listener->conn, &mp_opt, child, req);
886 if (!ctx->conn)
887 goto fallback;
888
889 ctx->subflow_id = 1;
890 owner = mptcp_sk(ctx->conn);
891
892 if (mp_opt.deny_join_id0)
893 WRITE_ONCE(owner->pm.remote_deny_join_id0, true);
894
895 mptcp_pm_new_connection(owner, child, 1);
896
897 /* with OoO packets we can reach here without ingress
898 * mpc option
899 */
900 if (mp_opt.suboptions & OPTION_MPTCP_MPC_ACK) {
901 mptcp_pm_fully_established(owner, child);
902 ctx->pm_notified = 1;
903 }
904 } else if (ctx->mp_join) {
905 owner = subflow_req->msk;
906 if (!owner) {
907 subflow_add_reset_reason(skb, MPTCP_RST_EPROHIBIT);
908 goto dispose_child;
909 }
910
911 if (!subflow_hmac_valid(req, &mp_opt)) {
912 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKMAC);
913 subflow_add_reset_reason(skb, MPTCP_RST_EPROHIBIT);
914 goto dispose_child;
915 }
916
917 if (!mptcp_can_accept_new_subflow(owner)) {
918 subflow_add_reset_reason(skb, MPTCP_RST_EPROHIBIT);
919 goto dispose_child;
920 }
921
922 /* move the msk reference ownership to the subflow */
923 subflow_req->msk = NULL;
924 ctx->conn = (struct sock *)owner;
925
926 if (subflow_use_different_sport(owner, sk)) {
927 pr_debug("ack inet_sport=%d %d\n",
928 ntohs(inet_sk(sk)->inet_sport),
929 ntohs(inet_sk((struct sock *)owner)->inet_sport));
930 if (!mptcp_pm_sport_in_anno_list(owner, sk)) {
931 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MISMATCHPORTACKRX);
932 subflow_add_reset_reason(skb, MPTCP_RST_EPROHIBIT);
933 goto dispose_child;
934 }
935 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINPORTACKRX);
936 }
937
938 if (!mptcp_finish_join(child)) {
939 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(child);
940
941 subflow_add_reset_reason(skb, subflow->reset_reason);
942 goto dispose_child;
943 }
944
945 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKRX);
946 tcp_rsk(req)->drop_req = true;
947 }
948 }
949
950 /* check for expected invariant - should never trigger, just help
951 * catching earlier subtle bugs
952 */
953 WARN_ON_ONCE(child && *own_req && tcp_sk(child)->is_mptcp &&
954 (!mptcp_subflow_ctx(child) ||
955 !mptcp_subflow_ctx(child)->conn));
956 return child;
957
958 dispose_child:
959 mptcp_subflow_drop_ctx(child);
960 tcp_rsk(req)->drop_req = true;
961 inet_csk_prepare_for_destroy_sock(child);
962 tcp_done(child);
963 reason = mptcp_get_rst_reason(skb);
964 req->rsk_ops->send_reset(sk, skb, reason);
965
966 /* The last child reference will be released by the caller */
967 return child;
968
969 fallback:
970 if (fallback)
971 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MPCAPABLEPASSIVEFALLBACK);
972 mptcp_subflow_drop_ctx(child);
973 return child;
974 }
975
976 static struct inet_connection_sock_af_ops subflow_specific __ro_after_init;
977 static struct proto tcp_prot_override __ro_after_init;
978
979 enum mapping_status {
980 MAPPING_OK,
981 MAPPING_INVALID,
982 MAPPING_EMPTY,
983 MAPPING_DATA_FIN,
984 MAPPING_DUMMY,
985 MAPPING_BAD_CSUM
986 };
987
dbg_bad_map(struct mptcp_subflow_context * subflow,u32 ssn)988 static void dbg_bad_map(struct mptcp_subflow_context *subflow, u32 ssn)
989 {
990 pr_debug("Bad mapping: ssn=%d map_seq=%d map_data_len=%d\n",
991 ssn, subflow->map_subflow_seq, subflow->map_data_len);
992 }
993
skb_is_fully_mapped(struct sock * ssk,struct sk_buff * skb)994 static bool skb_is_fully_mapped(struct sock *ssk, struct sk_buff *skb)
995 {
996 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
997 unsigned int skb_consumed;
998
999 skb_consumed = tcp_sk(ssk)->copied_seq - TCP_SKB_CB(skb)->seq;
1000 if (unlikely(skb_consumed >= skb->len)) {
1001 DEBUG_NET_WARN_ON_ONCE(1);
1002 return true;
1003 }
1004
1005 return skb->len - skb_consumed <= subflow->map_data_len -
1006 mptcp_subflow_get_map_offset(subflow);
1007 }
1008
validate_mapping(struct sock * ssk,struct sk_buff * skb)1009 static bool validate_mapping(struct sock *ssk, struct sk_buff *skb)
1010 {
1011 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1012 u32 ssn = tcp_sk(ssk)->copied_seq - subflow->ssn_offset;
1013
1014 if (unlikely(before(ssn, subflow->map_subflow_seq))) {
1015 /* Mapping covers data later in the subflow stream,
1016 * currently unsupported.
1017 */
1018 dbg_bad_map(subflow, ssn);
1019 return false;
1020 }
1021 if (unlikely(!before(ssn, subflow->map_subflow_seq +
1022 subflow->map_data_len))) {
1023 /* Mapping does covers past subflow data, invalid */
1024 dbg_bad_map(subflow, ssn);
1025 return false;
1026 }
1027 return true;
1028 }
1029
validate_data_csum(struct sock * ssk,struct sk_buff * skb,bool csum_reqd)1030 static enum mapping_status validate_data_csum(struct sock *ssk, struct sk_buff *skb,
1031 bool csum_reqd)
1032 {
1033 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1034 u32 offset, seq, delta;
1035 __sum16 csum;
1036 int len;
1037
1038 if (!csum_reqd)
1039 return MAPPING_OK;
1040
1041 /* mapping already validated on previous traversal */
1042 if (subflow->map_csum_len == subflow->map_data_len)
1043 return MAPPING_OK;
1044
1045 /* traverse the receive queue, ensuring it contains a full
1046 * DSS mapping and accumulating the related csum.
1047 * Preserve the accoumlate csum across multiple calls, to compute
1048 * the csum only once
1049 */
1050 delta = subflow->map_data_len - subflow->map_csum_len;
1051 for (;;) {
1052 seq = tcp_sk(ssk)->copied_seq + subflow->map_csum_len;
1053 offset = seq - TCP_SKB_CB(skb)->seq;
1054
1055 /* if the current skb has not been accounted yet, csum its contents
1056 * up to the amount covered by the current DSS
1057 */
1058 if (offset < skb->len) {
1059 __wsum csum;
1060
1061 len = min(skb->len - offset, delta);
1062 csum = skb_checksum(skb, offset, len, 0);
1063 subflow->map_data_csum = csum_block_add(subflow->map_data_csum, csum,
1064 subflow->map_csum_len);
1065
1066 delta -= len;
1067 subflow->map_csum_len += len;
1068 }
1069 if (delta == 0)
1070 break;
1071
1072 if (skb_queue_is_last(&ssk->sk_receive_queue, skb)) {
1073 /* if this subflow is closed, the partial mapping
1074 * will be never completed; flush the pending skbs, so
1075 * that subflow_sched_work_if_closed() can kick in
1076 */
1077 if (unlikely(ssk->sk_state == TCP_CLOSE))
1078 while ((skb = skb_peek(&ssk->sk_receive_queue)))
1079 sk_eat_skb(ssk, skb);
1080
1081 /* not enough data to validate the csum */
1082 return MAPPING_EMPTY;
1083 }
1084
1085 /* the DSS mapping for next skbs will be validated later,
1086 * when a get_mapping_status call will process such skb
1087 */
1088 skb = skb->next;
1089 }
1090
1091 /* note that 'map_data_len' accounts only for the carried data, does
1092 * not include the eventual seq increment due to the data fin,
1093 * while the pseudo header requires the original DSS data len,
1094 * including that
1095 */
1096 csum = __mptcp_make_csum(subflow->map_seq,
1097 subflow->map_subflow_seq,
1098 subflow->map_data_len + subflow->map_data_fin,
1099 subflow->map_data_csum);
1100 if (unlikely(csum)) {
1101 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DATACSUMERR);
1102 return MAPPING_BAD_CSUM;
1103 }
1104
1105 subflow->valid_csum_seen = 1;
1106 return MAPPING_OK;
1107 }
1108
get_mapping_status(struct sock * ssk,struct mptcp_sock * msk)1109 static enum mapping_status get_mapping_status(struct sock *ssk,
1110 struct mptcp_sock *msk)
1111 {
1112 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1113 bool csum_reqd = READ_ONCE(msk->csum_enabled);
1114 struct mptcp_ext *mpext;
1115 struct sk_buff *skb;
1116 u16 data_len;
1117 u64 map_seq;
1118
1119 skb = skb_peek(&ssk->sk_receive_queue);
1120 if (!skb)
1121 return MAPPING_EMPTY;
1122
1123 if (mptcp_check_fallback(ssk))
1124 return MAPPING_DUMMY;
1125
1126 mpext = mptcp_get_ext(skb);
1127 if (!mpext || !mpext->use_map) {
1128 if (!subflow->map_valid && !skb->len) {
1129 /* the TCP stack deliver 0 len FIN pkt to the receive
1130 * queue, that is the only 0len pkts ever expected here,
1131 * and we can admit no mapping only for 0 len pkts
1132 */
1133 if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN))
1134 WARN_ONCE(1, "0len seq %d:%d flags %x",
1135 TCP_SKB_CB(skb)->seq,
1136 TCP_SKB_CB(skb)->end_seq,
1137 TCP_SKB_CB(skb)->tcp_flags);
1138 sk_eat_skb(ssk, skb);
1139 return MAPPING_EMPTY;
1140 }
1141
1142 if (!subflow->map_valid)
1143 return MAPPING_INVALID;
1144
1145 goto validate_seq;
1146 }
1147
1148 trace_get_mapping_status(mpext);
1149
1150 data_len = mpext->data_len;
1151 if (data_len == 0) {
1152 pr_debug("infinite mapping received\n");
1153 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_INFINITEMAPRX);
1154 return MAPPING_INVALID;
1155 }
1156
1157 if (mpext->data_fin == 1) {
1158 u64 data_fin_seq;
1159
1160 if (data_len == 1) {
1161 bool updated = mptcp_update_rcv_data_fin(msk, mpext->data_seq,
1162 mpext->dsn64);
1163 pr_debug("DATA_FIN with no payload seq=%llu\n", mpext->data_seq);
1164 if (subflow->map_valid) {
1165 /* A DATA_FIN might arrive in a DSS
1166 * option before the previous mapping
1167 * has been fully consumed. Continue
1168 * handling the existing mapping.
1169 */
1170 skb_ext_del(skb, SKB_EXT_MPTCP);
1171 return MAPPING_OK;
1172 }
1173
1174 if (updated)
1175 mptcp_schedule_work((struct sock *)msk);
1176
1177 return MAPPING_DATA_FIN;
1178 }
1179
1180 data_fin_seq = mpext->data_seq + data_len - 1;
1181
1182 /* If mpext->data_seq is a 32-bit value, data_fin_seq must also
1183 * be limited to 32 bits.
1184 */
1185 if (!mpext->dsn64)
1186 data_fin_seq &= GENMASK_ULL(31, 0);
1187
1188 mptcp_update_rcv_data_fin(msk, data_fin_seq, mpext->dsn64);
1189 pr_debug("DATA_FIN with mapping seq=%llu dsn64=%d\n",
1190 data_fin_seq, mpext->dsn64);
1191
1192 /* Adjust for DATA_FIN using 1 byte of sequence space */
1193 data_len--;
1194 }
1195
1196 map_seq = mptcp_expand_seq(READ_ONCE(msk->ack_seq), mpext->data_seq, mpext->dsn64);
1197 WRITE_ONCE(mptcp_sk(subflow->conn)->use_64bit_ack, !!mpext->dsn64);
1198
1199 if (subflow->map_valid) {
1200 /* Allow replacing only with an identical map */
1201 if (subflow->map_seq == map_seq &&
1202 subflow->map_subflow_seq == mpext->subflow_seq &&
1203 subflow->map_data_len == data_len &&
1204 subflow->map_csum_reqd == mpext->csum_reqd) {
1205 skb_ext_del(skb, SKB_EXT_MPTCP);
1206 goto validate_csum;
1207 }
1208
1209 /* If this skb data are fully covered by the current mapping,
1210 * the new map would need caching, which is not supported
1211 */
1212 if (skb_is_fully_mapped(ssk, skb)) {
1213 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DSSNOMATCH);
1214 return MAPPING_INVALID;
1215 }
1216
1217 /* will validate the next map after consuming the current one */
1218 goto validate_csum;
1219 }
1220
1221 subflow->map_seq = map_seq;
1222 subflow->map_subflow_seq = mpext->subflow_seq;
1223 subflow->map_data_len = data_len;
1224 subflow->map_valid = 1;
1225 subflow->map_data_fin = mpext->data_fin;
1226 subflow->mpc_map = mpext->mpc_map;
1227 subflow->map_csum_reqd = mpext->csum_reqd;
1228 subflow->map_csum_len = 0;
1229 subflow->map_data_csum = csum_unfold(mpext->csum);
1230
1231 /* Cfr RFC 8684 Section 3.3.0 */
1232 if (unlikely(subflow->map_csum_reqd != csum_reqd))
1233 return MAPPING_INVALID;
1234
1235 pr_debug("new map seq=%llu subflow_seq=%u data_len=%u csum=%d:%u\n",
1236 subflow->map_seq, subflow->map_subflow_seq,
1237 subflow->map_data_len, subflow->map_csum_reqd,
1238 subflow->map_data_csum);
1239
1240 validate_seq:
1241 /* we revalidate valid mapping on new skb, because we must ensure
1242 * the current skb is completely covered by the available mapping
1243 */
1244 if (!validate_mapping(ssk, skb)) {
1245 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DSSTCPMISMATCH);
1246 return MAPPING_INVALID;
1247 }
1248
1249 skb_ext_del(skb, SKB_EXT_MPTCP);
1250
1251 validate_csum:
1252 return validate_data_csum(ssk, skb, csum_reqd);
1253 }
1254
mptcp_subflow_discard_data(struct sock * ssk,struct sk_buff * skb,u64 limit)1255 static void mptcp_subflow_discard_data(struct sock *ssk, struct sk_buff *skb,
1256 u64 limit)
1257 {
1258 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1259 bool fin = TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN;
1260 struct tcp_sock *tp = tcp_sk(ssk);
1261 u32 offset, incr, avail_len;
1262
1263 offset = tp->copied_seq - TCP_SKB_CB(skb)->seq;
1264 if (WARN_ON_ONCE(offset > skb->len))
1265 goto out;
1266
1267 avail_len = skb->len - offset;
1268 incr = limit >= avail_len ? avail_len + fin : limit;
1269
1270 pr_debug("discarding=%d len=%d offset=%d seq=%d\n", incr, skb->len,
1271 offset, subflow->map_subflow_seq);
1272 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DUPDATA);
1273 tcp_sk(ssk)->copied_seq += incr;
1274
1275 out:
1276 if (!before(tcp_sk(ssk)->copied_seq, TCP_SKB_CB(skb)->end_seq))
1277 sk_eat_skb(ssk, skb);
1278 if (mptcp_subflow_get_map_offset(subflow) >= subflow->map_data_len)
1279 subflow->map_valid = 0;
1280 }
1281
1282 /* 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)1283 static void subflow_sched_work_if_closed(struct mptcp_sock *msk, struct sock *ssk)
1284 {
1285 struct sock *sk = (struct sock *)msk;
1286
1287 if (likely(ssk->sk_state != TCP_CLOSE &&
1288 (ssk->sk_state != TCP_CLOSE_WAIT ||
1289 inet_sk_state_load(sk) != TCP_ESTABLISHED)))
1290 return;
1291
1292 if (skb_queue_empty(&ssk->sk_receive_queue) &&
1293 !test_and_set_bit(MPTCP_WORK_CLOSE_SUBFLOW, &msk->flags))
1294 mptcp_schedule_work(sk);
1295 }
1296
mptcp_subflow_fail(struct mptcp_sock * msk,struct sock * ssk)1297 static bool mptcp_subflow_fail(struct mptcp_sock *msk, struct sock *ssk)
1298 {
1299 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1300 unsigned long fail_tout;
1301
1302 /* we are really failing, prevent any later subflow join */
1303 spin_lock_bh(&msk->fallback_lock);
1304 if (!msk->allow_infinite_fallback) {
1305 spin_unlock_bh(&msk->fallback_lock);
1306 return false;
1307 }
1308 msk->allow_subflows = false;
1309 spin_unlock_bh(&msk->fallback_lock);
1310
1311 /* graceful failure can happen only on the MPC subflow */
1312 if (WARN_ON_ONCE(ssk != READ_ONCE(msk->first)))
1313 return false;
1314
1315 /* since the close timeout take precedence on the fail one,
1316 * no need to start the latter when the first is already set
1317 */
1318 if (sock_flag((struct sock *)msk, SOCK_DEAD))
1319 return true;
1320
1321 /* we don't need extreme accuracy here, use a zero fail_tout as special
1322 * value meaning no fail timeout at all;
1323 */
1324 fail_tout = jiffies + TCP_RTO_MAX;
1325 if (!fail_tout)
1326 fail_tout = 1;
1327 WRITE_ONCE(subflow->fail_tout, fail_tout);
1328 tcp_send_ack(ssk);
1329
1330 mptcp_reset_tout_timer(msk, subflow->fail_tout);
1331 return true;
1332 }
1333
subflow_check_data_avail(struct sock * ssk)1334 static bool subflow_check_data_avail(struct sock *ssk)
1335 {
1336 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1337 enum mapping_status status;
1338 struct mptcp_sock *msk;
1339 struct sk_buff *skb;
1340
1341 if (!skb_peek(&ssk->sk_receive_queue))
1342 WRITE_ONCE(subflow->data_avail, false);
1343 if (subflow->data_avail)
1344 return true;
1345
1346 msk = mptcp_sk(subflow->conn);
1347 for (;;) {
1348 u64 ack_seq;
1349 u64 old_ack;
1350
1351 status = get_mapping_status(ssk, msk);
1352 trace_subflow_check_data_avail(status, skb_peek(&ssk->sk_receive_queue));
1353 if (unlikely(status == MAPPING_INVALID || status == MAPPING_DUMMY ||
1354 status == MAPPING_BAD_CSUM))
1355 goto fallback;
1356
1357 if (status != MAPPING_OK)
1358 goto no_data;
1359
1360 skb = skb_peek(&ssk->sk_receive_queue);
1361 if (WARN_ON_ONCE(!skb))
1362 goto no_data;
1363
1364 if (unlikely(!READ_ONCE(msk->can_ack)))
1365 goto fallback;
1366
1367 old_ack = READ_ONCE(msk->ack_seq);
1368 ack_seq = mptcp_subflow_get_mapped_dsn(subflow);
1369 pr_debug("msk ack_seq=%llx subflow ack_seq=%llx\n", old_ack,
1370 ack_seq);
1371 if (unlikely(before64(ack_seq, old_ack))) {
1372 mptcp_subflow_discard_data(ssk, skb, old_ack - ack_seq);
1373 continue;
1374 }
1375
1376 WRITE_ONCE(subflow->data_avail, true);
1377 break;
1378 }
1379 return true;
1380
1381 no_data:
1382 subflow_sched_work_if_closed(msk, ssk);
1383 return false;
1384
1385 fallback:
1386 if (!__mptcp_check_fallback(msk)) {
1387 /* RFC 8684 section 3.7. */
1388 if (status == MAPPING_BAD_CSUM &&
1389 (subflow->mp_join || subflow->valid_csum_seen)) {
1390 subflow->send_mp_fail = 1;
1391
1392 if (!mptcp_subflow_fail(msk, ssk)) {
1393 subflow->reset_transient = 0;
1394 subflow->reset_reason = MPTCP_RST_EMIDDLEBOX;
1395 goto reset;
1396 }
1397 WRITE_ONCE(subflow->data_avail, true);
1398 return true;
1399 }
1400
1401 if (!mptcp_try_fallback(ssk)) {
1402 /* fatal protocol error, close the socket.
1403 * subflow_error_report() will introduce the appropriate barriers
1404 */
1405 subflow->reset_transient = 0;
1406 subflow->reset_reason = MPTCP_RST_EMPTCP;
1407
1408 reset:
1409 WRITE_ONCE(ssk->sk_err, EBADMSG);
1410 tcp_set_state(ssk, TCP_CLOSE);
1411 while ((skb = skb_peek(&ssk->sk_receive_queue)))
1412 sk_eat_skb(ssk, skb);
1413 mptcp_send_active_reset_reason(ssk);
1414 WRITE_ONCE(subflow->data_avail, false);
1415 return false;
1416 }
1417 }
1418
1419 skb = skb_peek(&ssk->sk_receive_queue);
1420 subflow->map_valid = 1;
1421 subflow->map_seq = READ_ONCE(msk->ack_seq);
1422 subflow->map_data_len = skb->len;
1423 subflow->map_subflow_seq = tcp_sk(ssk)->copied_seq - subflow->ssn_offset;
1424 WRITE_ONCE(subflow->data_avail, true);
1425 return true;
1426 }
1427
mptcp_subflow_data_available(struct sock * sk)1428 bool mptcp_subflow_data_available(struct sock *sk)
1429 {
1430 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1431
1432 /* check if current mapping is still valid */
1433 if (subflow->map_valid &&
1434 mptcp_subflow_get_map_offset(subflow) >= subflow->map_data_len) {
1435 subflow->map_valid = 0;
1436 WRITE_ONCE(subflow->data_avail, false);
1437
1438 pr_debug("Done with mapping: seq=%u data_len=%u\n",
1439 subflow->map_subflow_seq,
1440 subflow->map_data_len);
1441 }
1442
1443 return subflow_check_data_avail(sk);
1444 }
1445
1446 /* If ssk has an mptcp parent socket, use the mptcp rcvbuf occupancy,
1447 * not the ssk one.
1448 *
1449 * In mptcp, rwin is about the mptcp-level connection data.
1450 *
1451 * Data that is still on the ssk rx queue can thus be ignored,
1452 * as far as mptcp peer is concerned that data is still inflight.
1453 * DSS ACK is updated when skb is moved to the mptcp rx queue.
1454 */
mptcp_space(const struct sock * ssk,int * space,int * full_space)1455 void mptcp_space(const struct sock *ssk, int *space, int *full_space)
1456 {
1457 const struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1458 const struct sock *sk = subflow->conn;
1459
1460 *space = __mptcp_space(sk);
1461 *full_space = mptcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf));
1462 }
1463
subflow_error_report(struct sock * ssk)1464 static void subflow_error_report(struct sock *ssk)
1465 {
1466 struct sock *sk = mptcp_subflow_ctx(ssk)->conn;
1467
1468 /* bail early if this is a no-op, so that we avoid introducing a
1469 * problematic lockdep dependency between TCP accept queue lock
1470 * and msk socket spinlock
1471 */
1472 if (!sk->sk_socket)
1473 return;
1474
1475 mptcp_data_lock(sk);
1476 if (!sock_owned_by_user(sk))
1477 __mptcp_error_report(sk);
1478 else
1479 __set_bit(MPTCP_ERROR_REPORT, &mptcp_sk(sk)->cb_flags);
1480 mptcp_data_unlock(sk);
1481 }
1482
subflow_data_ready(struct sock * sk)1483 static void subflow_data_ready(struct sock *sk)
1484 {
1485 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1486 u16 state = 1 << inet_sk_state_load(sk);
1487 struct sock *parent = subflow->conn;
1488 struct mptcp_sock *msk;
1489
1490 trace_sk_data_ready(sk);
1491
1492 msk = mptcp_sk(parent);
1493 if (state & TCPF_LISTEN) {
1494 /* MPJ subflow are removed from accept queue before reaching here,
1495 * avoid stray wakeups
1496 */
1497 if (reqsk_queue_empty(&inet_csk(sk)->icsk_accept_queue))
1498 return;
1499
1500 parent->sk_data_ready(parent);
1501 return;
1502 }
1503
1504 WARN_ON_ONCE(!__mptcp_check_fallback(msk) && !subflow->mp_capable &&
1505 !subflow->mp_join && !(state & TCPF_CLOSE));
1506
1507 if (mptcp_subflow_data_available(sk)) {
1508 mptcp_data_ready(parent, sk);
1509
1510 /* subflow-level lowat test are not relevant.
1511 * respect the msk-level threshold eventually mandating an immediate ack
1512 */
1513 if (mptcp_data_avail(msk) < parent->sk_rcvlowat &&
1514 (tcp_sk(sk)->rcv_nxt - tcp_sk(sk)->rcv_wup) > inet_csk(sk)->icsk_ack.rcv_mss)
1515 inet_csk(sk)->icsk_ack.pending |= ICSK_ACK_NOW;
1516 } else if (unlikely(sk->sk_err)) {
1517 subflow_error_report(sk);
1518 }
1519 }
1520
subflow_write_space(struct sock * ssk)1521 static void subflow_write_space(struct sock *ssk)
1522 {
1523 struct sock *sk = mptcp_subflow_ctx(ssk)->conn;
1524
1525 mptcp_propagate_sndbuf(sk, ssk);
1526 mptcp_write_space(sk);
1527 }
1528
1529 static const struct inet_connection_sock_af_ops *
subflow_default_af_ops(struct sock * sk)1530 subflow_default_af_ops(struct sock *sk)
1531 {
1532 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1533 if (sk->sk_family == AF_INET6)
1534 return &subflow_v6_specific;
1535 #endif
1536 return &subflow_specific;
1537 }
1538
1539 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
mptcpv6_handle_mapped(struct sock * sk,bool mapped)1540 void mptcpv6_handle_mapped(struct sock *sk, bool mapped)
1541 {
1542 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1543 struct inet_connection_sock *icsk = inet_csk(sk);
1544 const struct inet_connection_sock_af_ops *target;
1545
1546 target = mapped ? &subflow_v6m_specific : subflow_default_af_ops(sk);
1547
1548 pr_debug("subflow=%p family=%d ops=%p target=%p mapped=%d\n",
1549 subflow, sk->sk_family, icsk->icsk_af_ops, target, mapped);
1550
1551 if (likely(icsk->icsk_af_ops == target))
1552 return;
1553
1554 subflow->icsk_af_ops = icsk->icsk_af_ops;
1555 icsk->icsk_af_ops = target;
1556 }
1557 #endif
1558
mptcp_info2sockaddr(const struct mptcp_addr_info * info,struct sockaddr_storage * addr,unsigned short family)1559 void mptcp_info2sockaddr(const struct mptcp_addr_info *info,
1560 struct sockaddr_storage *addr,
1561 unsigned short family)
1562 {
1563 memset(addr, 0, sizeof(*addr));
1564 addr->ss_family = family;
1565 if (addr->ss_family == AF_INET) {
1566 struct sockaddr_in *in_addr = (struct sockaddr_in *)addr;
1567
1568 if (info->family == AF_INET)
1569 in_addr->sin_addr = info->addr;
1570 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1571 else if (ipv6_addr_v4mapped(&info->addr6))
1572 in_addr->sin_addr.s_addr = info->addr6.s6_addr32[3];
1573 #endif
1574 in_addr->sin_port = info->port;
1575 }
1576 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1577 else if (addr->ss_family == AF_INET6) {
1578 struct sockaddr_in6 *in6_addr = (struct sockaddr_in6 *)addr;
1579
1580 if (info->family == AF_INET)
1581 ipv6_addr_set_v4mapped(info->addr.s_addr,
1582 &in6_addr->sin6_addr);
1583 else
1584 in6_addr->sin6_addr = info->addr6;
1585 in6_addr->sin6_port = info->port;
1586 }
1587 #endif
1588 }
1589
__mptcp_subflow_connect(struct sock * sk,const struct mptcp_pm_local * local,const struct mptcp_addr_info * remote)1590 int __mptcp_subflow_connect(struct sock *sk, const struct mptcp_pm_local *local,
1591 const struct mptcp_addr_info *remote)
1592 {
1593 struct mptcp_sock *msk = mptcp_sk(sk);
1594 struct mptcp_subflow_context *subflow;
1595 int local_id = local->addr.id;
1596 struct sockaddr_storage addr;
1597 int remote_id = remote->id;
1598 int err = -ENOTCONN;
1599 struct socket *sf;
1600 struct sock *ssk;
1601 u32 remote_token;
1602 int addrlen;
1603
1604 /* The userspace PM sent the request too early? */
1605 if (!mptcp_is_fully_established(sk))
1606 goto err_out;
1607
1608 err = mptcp_subflow_create_socket(sk, local->addr.family, &sf);
1609 if (err) {
1610 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINSYNTXCREATSKERR);
1611 pr_debug("msk=%p local=%d remote=%d create sock error: %d\n",
1612 msk, local_id, remote_id, err);
1613 goto err_out;
1614 }
1615
1616 ssk = sf->sk;
1617 subflow = mptcp_subflow_ctx(ssk);
1618 do {
1619 get_random_bytes(&subflow->local_nonce, sizeof(u32));
1620 } while (!subflow->local_nonce);
1621
1622 /* if 'IPADDRANY', the ID will be set later, after the routing */
1623 if (local->addr.family == AF_INET) {
1624 if (!local->addr.addr.s_addr)
1625 local_id = -1;
1626 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1627 } else if (sk->sk_family == AF_INET6) {
1628 if (ipv6_addr_any(&local->addr.addr6))
1629 local_id = -1;
1630 #endif
1631 }
1632
1633 if (local_id >= 0)
1634 subflow_set_local_id(subflow, local_id);
1635
1636 subflow->remote_key_valid = 1;
1637 subflow->remote_key = READ_ONCE(msk->remote_key);
1638 subflow->local_key = READ_ONCE(msk->local_key);
1639 subflow->token = msk->token;
1640 mptcp_info2sockaddr(&local->addr, &addr, ssk->sk_family);
1641
1642 addrlen = sizeof(struct sockaddr_in);
1643 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1644 if (addr.ss_family == AF_INET6)
1645 addrlen = sizeof(struct sockaddr_in6);
1646 #endif
1647 ssk->sk_bound_dev_if = local->ifindex;
1648 err = kernel_bind(sf, (struct sockaddr *)&addr, addrlen);
1649 if (err) {
1650 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINSYNTXBINDERR);
1651 pr_debug("msk=%p local=%d remote=%d bind error: %d\n",
1652 msk, local_id, remote_id, err);
1653 goto failed;
1654 }
1655
1656 mptcp_crypto_key_sha(subflow->remote_key, &remote_token, NULL);
1657 pr_debug("msk=%p remote_token=%u local_id=%d remote_id=%d\n", msk,
1658 remote_token, local_id, remote_id);
1659 subflow->remote_token = remote_token;
1660 WRITE_ONCE(subflow->remote_id, remote_id);
1661 subflow->request_join = 1;
1662 subflow->request_bkup = !!(local->flags & MPTCP_PM_ADDR_FLAG_BACKUP);
1663 subflow->subflow_id = msk->subflow_id++;
1664 mptcp_info2sockaddr(remote, &addr, ssk->sk_family);
1665
1666 sock_hold(ssk);
1667 list_add_tail(&subflow->node, &msk->conn_list);
1668 err = kernel_connect(sf, (struct sockaddr *)&addr, addrlen, O_NONBLOCK);
1669 if (err && err != -EINPROGRESS) {
1670 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINSYNTXCONNECTERR);
1671 pr_debug("msk=%p local=%d remote=%d connect error: %d\n",
1672 msk, local_id, remote_id, err);
1673 goto failed_unlink;
1674 }
1675
1676 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINSYNTX);
1677
1678 /* discard the subflow socket */
1679 mptcp_sock_graft(ssk, sk->sk_socket);
1680 iput(SOCK_INODE(sf));
1681 mptcp_stop_tout_timer(sk);
1682 return 0;
1683
1684 failed_unlink:
1685 list_del(&subflow->node);
1686 sock_put(mptcp_subflow_tcp_sock(subflow));
1687
1688 failed:
1689 subflow->disposable = 1;
1690 sock_release(sf);
1691
1692 err_out:
1693 /* we account subflows before the creation, and this failures will not
1694 * be caught by sk_state_change()
1695 */
1696 mptcp_pm_close_subflow(msk);
1697 return err;
1698 }
1699
mptcp_attach_cgroup(struct sock * parent,struct sock * child)1700 static void mptcp_attach_cgroup(struct sock *parent, struct sock *child)
1701 {
1702 #ifdef CONFIG_SOCK_CGROUP_DATA
1703 struct sock_cgroup_data *parent_skcd = &parent->sk_cgrp_data,
1704 *child_skcd = &child->sk_cgrp_data;
1705
1706 /* only the additional subflows created by kworkers have to be modified */
1707 if (cgroup_id(sock_cgroup_ptr(parent_skcd)) !=
1708 cgroup_id(sock_cgroup_ptr(child_skcd))) {
1709 #ifdef CONFIG_MEMCG
1710 struct mem_cgroup *memcg = parent->sk_memcg;
1711
1712 mem_cgroup_sk_free(child);
1713 if (memcg && css_tryget(&memcg->css))
1714 child->sk_memcg = memcg;
1715 #endif /* CONFIG_MEMCG */
1716
1717 cgroup_sk_free(child_skcd);
1718 *child_skcd = *parent_skcd;
1719 cgroup_sk_clone(child_skcd);
1720 }
1721 #endif /* CONFIG_SOCK_CGROUP_DATA */
1722 }
1723
mptcp_subflow_ops_override(struct sock * ssk)1724 static void mptcp_subflow_ops_override(struct sock *ssk)
1725 {
1726 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1727 if (ssk->sk_prot == &tcpv6_prot)
1728 ssk->sk_prot = &tcpv6_prot_override;
1729 else
1730 #endif
1731 ssk->sk_prot = &tcp_prot_override;
1732 }
1733
mptcp_subflow_ops_undo_override(struct sock * ssk)1734 static void mptcp_subflow_ops_undo_override(struct sock *ssk)
1735 {
1736 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1737 if (ssk->sk_prot == &tcpv6_prot_override)
1738 ssk->sk_prot = &tcpv6_prot;
1739 else
1740 #endif
1741 ssk->sk_prot = &tcp_prot;
1742 }
1743
mptcp_subflow_create_socket(struct sock * sk,unsigned short family,struct socket ** new_sock)1744 int mptcp_subflow_create_socket(struct sock *sk, unsigned short family,
1745 struct socket **new_sock)
1746 {
1747 struct mptcp_subflow_context *subflow;
1748 struct net *net = sock_net(sk);
1749 struct socket *sf;
1750 int err;
1751
1752 /* un-accepted server sockets can reach here - on bad configuration
1753 * bail early to avoid greater trouble later
1754 */
1755 if (unlikely(!sk->sk_socket))
1756 return -EINVAL;
1757
1758 err = sock_create_kern(net, family, SOCK_STREAM, IPPROTO_TCP, &sf);
1759 if (err)
1760 return err;
1761
1762 lock_sock_nested(sf->sk, SINGLE_DEPTH_NESTING);
1763
1764 err = security_mptcp_add_subflow(sk, sf->sk);
1765 if (err)
1766 goto err_free;
1767
1768 /* the newly created socket has to be in the same cgroup as its parent */
1769 mptcp_attach_cgroup(sk, sf->sk);
1770
1771 /* kernel sockets do not by default acquire net ref, but TCP timer
1772 * needs it.
1773 * Update ns_tracker to current stack trace and refcounted tracker.
1774 */
1775 sk_net_refcnt_upgrade(sf->sk);
1776 err = tcp_set_ulp(sf->sk, "mptcp");
1777 if (err)
1778 goto err_free;
1779
1780 mptcp_sockopt_sync_locked(mptcp_sk(sk), sf->sk);
1781 release_sock(sf->sk);
1782
1783 /* the newly created socket really belongs to the owning MPTCP
1784 * socket, even if for additional subflows the allocation is performed
1785 * by a kernel workqueue. Adjust inode references, so that the
1786 * procfs/diag interfaces really show this one belonging to the correct
1787 * user.
1788 */
1789 SOCK_INODE(sf)->i_ino = SOCK_INODE(sk->sk_socket)->i_ino;
1790 SOCK_INODE(sf)->i_uid = SOCK_INODE(sk->sk_socket)->i_uid;
1791 SOCK_INODE(sf)->i_gid = SOCK_INODE(sk->sk_socket)->i_gid;
1792
1793 subflow = mptcp_subflow_ctx(sf->sk);
1794 pr_debug("subflow=%p\n", subflow);
1795
1796 *new_sock = sf;
1797 sock_hold(sk);
1798 subflow->conn = sk;
1799 mptcp_subflow_ops_override(sf->sk);
1800
1801 return 0;
1802
1803 err_free:
1804 release_sock(sf->sk);
1805 sock_release(sf);
1806 return err;
1807 }
1808
subflow_create_ctx(struct sock * sk,gfp_t priority)1809 static struct mptcp_subflow_context *subflow_create_ctx(struct sock *sk,
1810 gfp_t priority)
1811 {
1812 struct inet_connection_sock *icsk = inet_csk(sk);
1813 struct mptcp_subflow_context *ctx;
1814
1815 ctx = kzalloc(sizeof(*ctx), priority);
1816 if (!ctx)
1817 return NULL;
1818
1819 rcu_assign_pointer(icsk->icsk_ulp_data, ctx);
1820 INIT_LIST_HEAD(&ctx->node);
1821 INIT_LIST_HEAD(&ctx->delegated_node);
1822
1823 pr_debug("subflow=%p\n", ctx);
1824
1825 ctx->tcp_sock = sk;
1826 WRITE_ONCE(ctx->local_id, -1);
1827
1828 return ctx;
1829 }
1830
__subflow_state_change(struct sock * sk)1831 static void __subflow_state_change(struct sock *sk)
1832 {
1833 struct socket_wq *wq;
1834
1835 rcu_read_lock();
1836 wq = rcu_dereference(sk->sk_wq);
1837 if (skwq_has_sleeper(wq))
1838 wake_up_interruptible_all(&wq->wait);
1839 rcu_read_unlock();
1840 }
1841
subflow_is_done(const struct sock * sk)1842 static bool subflow_is_done(const struct sock *sk)
1843 {
1844 return sk->sk_shutdown & RCV_SHUTDOWN || sk->sk_state == TCP_CLOSE;
1845 }
1846
subflow_state_change(struct sock * sk)1847 static void subflow_state_change(struct sock *sk)
1848 {
1849 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1850 struct sock *parent = subflow->conn;
1851 struct mptcp_sock *msk;
1852
1853 __subflow_state_change(sk);
1854
1855 msk = mptcp_sk(parent);
1856 if (subflow_simultaneous_connect(sk)) {
1857 WARN_ON_ONCE(!mptcp_try_fallback(sk));
1858 pr_fallback(msk);
1859 subflow->conn_finished = 1;
1860 mptcp_propagate_state(parent, sk, subflow, NULL);
1861 }
1862
1863 /* as recvmsg() does not acquire the subflow socket for ssk selection
1864 * a fin packet carrying a DSS can be unnoticed if we don't trigger
1865 * the data available machinery here.
1866 */
1867 if (mptcp_subflow_data_available(sk))
1868 mptcp_data_ready(parent, sk);
1869 else if (unlikely(sk->sk_err))
1870 subflow_error_report(sk);
1871
1872 subflow_sched_work_if_closed(mptcp_sk(parent), sk);
1873
1874 /* when the fallback subflow closes the rx side, trigger a 'dummy'
1875 * ingress data fin, so that the msk state will follow along
1876 */
1877 if (__mptcp_check_fallback(msk) && subflow_is_done(sk) && msk->first == sk &&
1878 mptcp_update_rcv_data_fin(msk, READ_ONCE(msk->ack_seq), true))
1879 mptcp_schedule_work(parent);
1880 }
1881
mptcp_subflow_queue_clean(struct sock * listener_sk,struct sock * listener_ssk)1882 void mptcp_subflow_queue_clean(struct sock *listener_sk, struct sock *listener_ssk)
1883 {
1884 struct request_sock_queue *queue = &inet_csk(listener_ssk)->icsk_accept_queue;
1885 struct request_sock *req, *head, *tail;
1886 struct mptcp_subflow_context *subflow;
1887 struct sock *sk, *ssk;
1888
1889 /* Due to lock dependencies no relevant lock can be acquired under rskq_lock.
1890 * Splice the req list, so that accept() can not reach the pending ssk after
1891 * the listener socket is released below.
1892 */
1893 spin_lock_bh(&queue->rskq_lock);
1894 head = queue->rskq_accept_head;
1895 tail = queue->rskq_accept_tail;
1896 queue->rskq_accept_head = NULL;
1897 queue->rskq_accept_tail = NULL;
1898 spin_unlock_bh(&queue->rskq_lock);
1899
1900 if (!head)
1901 return;
1902
1903 /* can't acquire the msk socket lock under the subflow one,
1904 * or will cause ABBA deadlock
1905 */
1906 release_sock(listener_ssk);
1907
1908 for (req = head; req; req = req->dl_next) {
1909 ssk = req->sk;
1910 if (!sk_is_mptcp(ssk))
1911 continue;
1912
1913 subflow = mptcp_subflow_ctx(ssk);
1914 if (!subflow || !subflow->conn)
1915 continue;
1916
1917 sk = subflow->conn;
1918 sock_hold(sk);
1919
1920 lock_sock_nested(sk, SINGLE_DEPTH_NESTING);
1921 __mptcp_unaccepted_force_close(sk);
1922 release_sock(sk);
1923
1924 /* lockdep will report a false positive ABBA deadlock
1925 * between cancel_work_sync and the listener socket.
1926 * The involved locks belong to different sockets WRT
1927 * the existing AB chain.
1928 * Using a per socket key is problematic as key
1929 * deregistration requires process context and must be
1930 * performed at socket disposal time, in atomic
1931 * context.
1932 * Just tell lockdep to consider the listener socket
1933 * released here.
1934 */
1935 mutex_release(&listener_sk->sk_lock.dep_map, _RET_IP_);
1936 mptcp_cancel_work(sk);
1937 mutex_acquire(&listener_sk->sk_lock.dep_map, 0, 0, _RET_IP_);
1938
1939 sock_put(sk);
1940 }
1941
1942 /* we are still under the listener msk socket lock */
1943 lock_sock_nested(listener_ssk, SINGLE_DEPTH_NESTING);
1944
1945 /* restore the listener queue, to let the TCP code clean it up */
1946 spin_lock_bh(&queue->rskq_lock);
1947 WARN_ON_ONCE(queue->rskq_accept_head);
1948 queue->rskq_accept_head = head;
1949 queue->rskq_accept_tail = tail;
1950 spin_unlock_bh(&queue->rskq_lock);
1951 }
1952
subflow_ulp_init(struct sock * sk)1953 static int subflow_ulp_init(struct sock *sk)
1954 {
1955 struct inet_connection_sock *icsk = inet_csk(sk);
1956 struct mptcp_subflow_context *ctx;
1957 struct tcp_sock *tp = tcp_sk(sk);
1958 int err = 0;
1959
1960 /* disallow attaching ULP to a socket unless it has been
1961 * created with sock_create_kern()
1962 */
1963 if (!sk->sk_kern_sock) {
1964 err = -EOPNOTSUPP;
1965 goto out;
1966 }
1967
1968 ctx = subflow_create_ctx(sk, GFP_KERNEL);
1969 if (!ctx) {
1970 err = -ENOMEM;
1971 goto out;
1972 }
1973
1974 pr_debug("subflow=%p, family=%d\n", ctx, sk->sk_family);
1975
1976 tp->is_mptcp = 1;
1977 ctx->icsk_af_ops = icsk->icsk_af_ops;
1978 icsk->icsk_af_ops = subflow_default_af_ops(sk);
1979 ctx->tcp_state_change = sk->sk_state_change;
1980 ctx->tcp_error_report = sk->sk_error_report;
1981
1982 WARN_ON_ONCE(sk->sk_data_ready != sock_def_readable);
1983 WARN_ON_ONCE(sk->sk_write_space != sk_stream_write_space);
1984
1985 sk->sk_data_ready = subflow_data_ready;
1986 sk->sk_write_space = subflow_write_space;
1987 sk->sk_state_change = subflow_state_change;
1988 sk->sk_error_report = subflow_error_report;
1989 out:
1990 return err;
1991 }
1992
subflow_ulp_release(struct sock * ssk)1993 static void subflow_ulp_release(struct sock *ssk)
1994 {
1995 struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(ssk);
1996 bool release = true;
1997 struct sock *sk;
1998
1999 if (!ctx)
2000 return;
2001
2002 sk = ctx->conn;
2003 if (sk) {
2004 /* if the msk has been orphaned, keep the ctx
2005 * alive, will be freed by __mptcp_close_ssk(),
2006 * when the subflow is still unaccepted
2007 */
2008 release = ctx->disposable || list_empty(&ctx->node);
2009
2010 /* inet_child_forget() does not call sk_state_change(),
2011 * explicitly trigger the socket close machinery
2012 */
2013 if (!release && !test_and_set_bit(MPTCP_WORK_CLOSE_SUBFLOW,
2014 &mptcp_sk(sk)->flags))
2015 mptcp_schedule_work(sk);
2016 sock_put(sk);
2017 }
2018
2019 mptcp_subflow_ops_undo_override(ssk);
2020 if (release)
2021 kfree_rcu(ctx, rcu);
2022 }
2023
subflow_ulp_clone(const struct request_sock * req,struct sock * newsk,const gfp_t priority)2024 static void subflow_ulp_clone(const struct request_sock *req,
2025 struct sock *newsk,
2026 const gfp_t priority)
2027 {
2028 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
2029 struct mptcp_subflow_context *old_ctx = mptcp_subflow_ctx(newsk);
2030 struct mptcp_subflow_context *new_ctx;
2031
2032 if (!tcp_rsk(req)->is_mptcp ||
2033 (!subflow_req->mp_capable && !subflow_req->mp_join)) {
2034 subflow_ulp_fallback(newsk, old_ctx);
2035 return;
2036 }
2037
2038 new_ctx = subflow_create_ctx(newsk, priority);
2039 if (!new_ctx) {
2040 subflow_ulp_fallback(newsk, old_ctx);
2041 return;
2042 }
2043
2044 new_ctx->conn_finished = 1;
2045 new_ctx->icsk_af_ops = old_ctx->icsk_af_ops;
2046 new_ctx->tcp_state_change = old_ctx->tcp_state_change;
2047 new_ctx->tcp_error_report = old_ctx->tcp_error_report;
2048 new_ctx->rel_write_seq = 1;
2049 new_ctx->tcp_sock = newsk;
2050
2051 if (subflow_req->mp_capable) {
2052 /* see comments in subflow_syn_recv_sock(), MPTCP connection
2053 * is fully established only after we receive the remote key
2054 */
2055 new_ctx->mp_capable = 1;
2056 new_ctx->local_key = subflow_req->local_key;
2057 new_ctx->token = subflow_req->token;
2058 new_ctx->ssn_offset = subflow_req->ssn_offset;
2059 new_ctx->idsn = subflow_req->idsn;
2060
2061 /* this is the first subflow, id is always 0 */
2062 subflow_set_local_id(new_ctx, 0);
2063 } else if (subflow_req->mp_join) {
2064 new_ctx->ssn_offset = subflow_req->ssn_offset;
2065 new_ctx->mp_join = 1;
2066 WRITE_ONCE(new_ctx->fully_established, true);
2067 new_ctx->remote_key_valid = 1;
2068 new_ctx->backup = subflow_req->backup;
2069 new_ctx->request_bkup = subflow_req->request_bkup;
2070 WRITE_ONCE(new_ctx->remote_id, subflow_req->remote_id);
2071 new_ctx->token = subflow_req->token;
2072 new_ctx->thmac = subflow_req->thmac;
2073
2074 /* the subflow req id is valid, fetched via subflow_check_req()
2075 * and subflow_token_join_request()
2076 */
2077 subflow_set_local_id(new_ctx, subflow_req->local_id);
2078 }
2079 }
2080
tcp_release_cb_override(struct sock * ssk)2081 static void tcp_release_cb_override(struct sock *ssk)
2082 {
2083 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
2084 long status;
2085
2086 /* process and clear all the pending actions, but leave the subflow into
2087 * the napi queue. To respect locking, only the same CPU that originated
2088 * the action can touch the list. mptcp_napi_poll will take care of it.
2089 */
2090 status = set_mask_bits(&subflow->delegated_status, MPTCP_DELEGATE_ACTIONS_MASK, 0);
2091 if (status)
2092 mptcp_subflow_process_delegated(ssk, status);
2093
2094 tcp_release_cb(ssk);
2095 }
2096
tcp_abort_override(struct sock * ssk,int err)2097 static int tcp_abort_override(struct sock *ssk, int err)
2098 {
2099 /* closing a listener subflow requires a great deal of care.
2100 * keep it simple and just prevent such operation
2101 */
2102 if (inet_sk_state_load(ssk) == TCP_LISTEN)
2103 return -EINVAL;
2104
2105 return tcp_abort(ssk, err);
2106 }
2107
2108 static struct tcp_ulp_ops subflow_ulp_ops __read_mostly = {
2109 .name = "mptcp",
2110 .owner = THIS_MODULE,
2111 .init = subflow_ulp_init,
2112 .release = subflow_ulp_release,
2113 .clone = subflow_ulp_clone,
2114 };
2115
subflow_ops_init(struct request_sock_ops * subflow_ops)2116 static int subflow_ops_init(struct request_sock_ops *subflow_ops)
2117 {
2118 subflow_ops->obj_size = sizeof(struct mptcp_subflow_request_sock);
2119
2120 subflow_ops->slab = kmem_cache_create(subflow_ops->slab_name,
2121 subflow_ops->obj_size, 0,
2122 SLAB_ACCOUNT |
2123 SLAB_TYPESAFE_BY_RCU,
2124 NULL);
2125 if (!subflow_ops->slab)
2126 return -ENOMEM;
2127
2128 return 0;
2129 }
2130
mptcp_subflow_init(void)2131 void __init mptcp_subflow_init(void)
2132 {
2133 mptcp_subflow_v4_request_sock_ops = tcp_request_sock_ops;
2134 mptcp_subflow_v4_request_sock_ops.slab_name = "request_sock_subflow_v4";
2135 mptcp_subflow_v4_request_sock_ops.destructor = subflow_v4_req_destructor;
2136
2137 if (subflow_ops_init(&mptcp_subflow_v4_request_sock_ops) != 0)
2138 panic("MPTCP: failed to init subflow v4 request sock ops\n");
2139
2140 subflow_request_sock_ipv4_ops = tcp_request_sock_ipv4_ops;
2141 subflow_request_sock_ipv4_ops.route_req = subflow_v4_route_req;
2142 subflow_request_sock_ipv4_ops.send_synack = subflow_v4_send_synack;
2143
2144 subflow_specific = ipv4_specific;
2145 subflow_specific.conn_request = subflow_v4_conn_request;
2146 subflow_specific.syn_recv_sock = subflow_syn_recv_sock;
2147 subflow_specific.sk_rx_dst_set = subflow_finish_connect;
2148 subflow_specific.rebuild_header = subflow_rebuild_header;
2149
2150 tcp_prot_override = tcp_prot;
2151 tcp_prot_override.release_cb = tcp_release_cb_override;
2152 tcp_prot_override.diag_destroy = tcp_abort_override;
2153
2154 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
2155 /* In struct mptcp_subflow_request_sock, we assume the TCP request sock
2156 * structures for v4 and v6 have the same size. It should not changed in
2157 * the future but better to make sure to be warned if it is no longer
2158 * the case.
2159 */
2160 BUILD_BUG_ON(sizeof(struct tcp_request_sock) != sizeof(struct tcp6_request_sock));
2161
2162 mptcp_subflow_v6_request_sock_ops = tcp6_request_sock_ops;
2163 mptcp_subflow_v6_request_sock_ops.slab_name = "request_sock_subflow_v6";
2164 mptcp_subflow_v6_request_sock_ops.destructor = subflow_v6_req_destructor;
2165
2166 if (subflow_ops_init(&mptcp_subflow_v6_request_sock_ops) != 0)
2167 panic("MPTCP: failed to init subflow v6 request sock ops\n");
2168
2169 subflow_request_sock_ipv6_ops = tcp_request_sock_ipv6_ops;
2170 subflow_request_sock_ipv6_ops.route_req = subflow_v6_route_req;
2171 subflow_request_sock_ipv6_ops.send_synack = subflow_v6_send_synack;
2172
2173 subflow_v6_specific = ipv6_specific;
2174 subflow_v6_specific.conn_request = subflow_v6_conn_request;
2175 subflow_v6_specific.syn_recv_sock = subflow_syn_recv_sock;
2176 subflow_v6_specific.sk_rx_dst_set = subflow_finish_connect;
2177 subflow_v6_specific.rebuild_header = subflow_v6_rebuild_header;
2178
2179 subflow_v6m_specific = subflow_v6_specific;
2180 subflow_v6m_specific.queue_xmit = ipv4_specific.queue_xmit;
2181 subflow_v6m_specific.send_check = ipv4_specific.send_check;
2182 subflow_v6m_specific.net_header_len = ipv4_specific.net_header_len;
2183 subflow_v6m_specific.mtu_reduced = ipv4_specific.mtu_reduced;
2184 subflow_v6m_specific.rebuild_header = subflow_rebuild_header;
2185
2186 tcpv6_prot_override = tcpv6_prot;
2187 tcpv6_prot_override.release_cb = tcp_release_cb_override;
2188 tcpv6_prot_override.diag_destroy = tcp_abort_override;
2189 #endif
2190
2191 mptcp_diag_subflow_init(&subflow_ulp_ops);
2192
2193 if (tcp_register_ulp(&subflow_ulp_ops) != 0)
2194 panic("MPTCP: failed to register subflows to ULP\n");
2195 }
2196