1 // SPDX-License-Identifier: GPL-2.0
2 /* Multipath TCP
3 *
4 * Copyright (c) 2019, Intel Corporation.
5 */
6 #define pr_fmt(fmt) "MPTCP: " fmt
7
8 #include <linux/kernel.h>
9 #include <net/mptcp.h>
10 #include "protocol.h"
11
12 #include "mib.h"
13
14 /* path manager command handlers */
15
16 int mptcp_pm_announce_addr(struct mptcp_sock *msk,
17 const struct mptcp_addr_info *addr,
18 bool echo)
19 {
20 u8 add_addr = READ_ONCE(msk->pm.addr_signal);
21
22 pr_debug("msk=%p, local_id=%d, echo=%d\n", msk, addr->id, echo);
23
24 lockdep_assert_held(&msk->pm.lock);
25
26 if (add_addr &
27 (echo ? BIT(MPTCP_ADD_ADDR_ECHO) : BIT(MPTCP_ADD_ADDR_SIGNAL))) {
28 MPTCP_INC_STATS(sock_net((struct sock *)msk),
29 echo ? MPTCP_MIB_ECHOADDTXDROP : MPTCP_MIB_ADDADDRTXDROP);
30 return -EINVAL;
31 }
32
33 if (echo) {
34 msk->pm.remote = *addr;
35 add_addr |= BIT(MPTCP_ADD_ADDR_ECHO);
36 } else {
37 msk->pm.local = *addr;
38 add_addr |= BIT(MPTCP_ADD_ADDR_SIGNAL);
39 }
40 WRITE_ONCE(msk->pm.addr_signal, add_addr);
41 return 0;
42 }
43
44 int mptcp_pm_remove_addr(struct mptcp_sock *msk, const struct mptcp_rm_list *rm_list)
45 {
46 u8 rm_addr = READ_ONCE(msk->pm.addr_signal);
47
48 pr_debug("msk=%p, rm_list_nr=%d\n", msk, rm_list->nr);
49
50 if (rm_addr) {
51 MPTCP_ADD_STATS(sock_net((struct sock *)msk),
52 MPTCP_MIB_RMADDRTXDROP, rm_list->nr);
53 return -EINVAL;
54 }
55
56 msk->pm.rm_list_tx = *rm_list;
57 rm_addr |= BIT(MPTCP_RM_ADDR_SIGNAL);
58 WRITE_ONCE(msk->pm.addr_signal, rm_addr);
59 mptcp_pm_nl_addr_send_ack(msk);
60 return 0;
61 }
62
63 /* path manager event handlers */
64
65 void mptcp_pm_new_connection(struct mptcp_sock *msk, const struct sock *ssk, int server_side)
66 {
67 struct mptcp_pm_data *pm = &msk->pm;
68
69 pr_debug("msk=%p, token=%u side=%d\n", msk, READ_ONCE(msk->token), server_side);
70
71 WRITE_ONCE(pm->server_side, server_side);
72 mptcp_event(MPTCP_EVENT_CREATED, msk, ssk, GFP_ATOMIC);
73 }
74
75 bool mptcp_pm_allow_new_subflow(struct mptcp_sock *msk)
76 {
77 struct mptcp_pm_data *pm = &msk->pm;
78 unsigned int subflows_max;
79 int ret = 0;
80
81 if (mptcp_pm_is_userspace(msk)) {
82 if (mptcp_userspace_pm_active(msk)) {
83 spin_lock_bh(&pm->lock);
84 pm->subflows++;
85 spin_unlock_bh(&pm->lock);
86 return true;
87 }
88 return false;
89 }
90
91 subflows_max = mptcp_pm_get_subflows_max(msk);
92
93 pr_debug("msk=%p subflows=%d max=%d allow=%d\n", msk, pm->subflows,
94 subflows_max, READ_ONCE(pm->accept_subflow));
95
96 /* try to avoid acquiring the lock below */
97 if (!READ_ONCE(pm->accept_subflow))
98 return false;
99
100 spin_lock_bh(&pm->lock);
101 if (READ_ONCE(pm->accept_subflow)) {
102 ret = pm->subflows < subflows_max;
103 if (ret && ++pm->subflows == subflows_max)
104 WRITE_ONCE(pm->accept_subflow, false);
105 }
106 spin_unlock_bh(&pm->lock);
107
108 return ret;
109 }
110
111 /* return true if the new status bit is currently cleared, that is, this event
112 * can be server, eventually by an already scheduled work
113 */
114 static bool mptcp_pm_schedule_work(struct mptcp_sock *msk,
115 enum mptcp_pm_status new_status)
116 {
117 pr_debug("msk=%p status=%x new=%lx\n", msk, msk->pm.status,
118 BIT(new_status));
119 if (msk->pm.status & BIT(new_status))
120 return false;
121
122 msk->pm.status |= BIT(new_status);
123 mptcp_schedule_work((struct sock *)msk);
124 return true;
125 }
126
127 void mptcp_pm_fully_established(struct mptcp_sock *msk, const struct sock *ssk)
128 {
129 struct mptcp_pm_data *pm = &msk->pm;
130 bool announce = false;
131
132 pr_debug("msk=%p\n", msk);
133
134 spin_lock_bh(&pm->lock);
135
136 /* mptcp_pm_fully_established() can be invoked by multiple
137 * racing paths - accept() and check_fully_established()
138 * be sure to serve this event only once.
139 */
140 if (READ_ONCE(pm->work_pending) &&
141 !(msk->pm.status & BIT(MPTCP_PM_ALREADY_ESTABLISHED)))
142 mptcp_pm_schedule_work(msk, MPTCP_PM_ESTABLISHED);
143
144 if ((msk->pm.status & BIT(MPTCP_PM_ALREADY_ESTABLISHED)) == 0)
145 announce = true;
146
147 msk->pm.status |= BIT(MPTCP_PM_ALREADY_ESTABLISHED);
148 spin_unlock_bh(&pm->lock);
149
150 if (announce)
151 mptcp_event(MPTCP_EVENT_ESTABLISHED, msk, ssk, GFP_ATOMIC);
152 }
153
154 void mptcp_pm_connection_closed(struct mptcp_sock *msk)
155 {
156 pr_debug("msk=%p\n", msk);
157 }
158
159 void mptcp_pm_subflow_established(struct mptcp_sock *msk)
160 {
161 struct mptcp_pm_data *pm = &msk->pm;
162
163 pr_debug("msk=%p\n", msk);
164
165 if (!READ_ONCE(pm->work_pending))
166 return;
167
168 spin_lock_bh(&pm->lock);
169
170 if (READ_ONCE(pm->work_pending))
171 mptcp_pm_schedule_work(msk, MPTCP_PM_SUBFLOW_ESTABLISHED);
172
173 spin_unlock_bh(&pm->lock);
174 }
175
176 void mptcp_pm_subflow_check_next(struct mptcp_sock *msk,
177 const struct mptcp_subflow_context *subflow)
178 {
179 struct mptcp_pm_data *pm = &msk->pm;
180 bool update_subflows;
181
182 update_subflows = subflow->request_join || subflow->mp_join;
183 if (mptcp_pm_is_userspace(msk)) {
184 if (update_subflows) {
185 spin_lock_bh(&pm->lock);
186 pm->subflows--;
187 spin_unlock_bh(&pm->lock);
188 }
189 return;
190 }
191
192 if (!READ_ONCE(pm->work_pending) && !update_subflows)
193 return;
194
195 spin_lock_bh(&pm->lock);
196 if (update_subflows)
197 __mptcp_pm_close_subflow(msk);
198
199 /* Even if this subflow is not really established, tell the PM to try
200 * to pick the next ones, if possible.
201 */
202 if (mptcp_pm_nl_check_work_pending(msk))
203 mptcp_pm_schedule_work(msk, MPTCP_PM_SUBFLOW_ESTABLISHED);
204
205 spin_unlock_bh(&pm->lock);
206 }
207
208 void mptcp_pm_add_addr_received(const struct sock *ssk,
209 const struct mptcp_addr_info *addr)
210 {
211 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
212 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
213 struct mptcp_pm_data *pm = &msk->pm;
214
215 pr_debug("msk=%p remote_id=%d accept=%d\n", msk, addr->id,
216 READ_ONCE(pm->accept_addr));
217
218 mptcp_event_addr_announced(ssk, addr);
219
220 spin_lock_bh(&pm->lock);
221
222 if (mptcp_pm_is_userspace(msk)) {
223 if (mptcp_userspace_pm_active(msk)) {
224 mptcp_pm_announce_addr(msk, addr, true);
225 mptcp_pm_add_addr_send_ack(msk);
226 } else {
227 __MPTCP_INC_STATS(sock_net((struct sock *)msk), MPTCP_MIB_ADDADDRDROP);
228 }
229 /* id0 should not have a different address */
230 } else if ((addr->id == 0 && !mptcp_pm_nl_is_init_remote_addr(msk, addr)) ||
231 (addr->id > 0 && !READ_ONCE(pm->accept_addr))) {
232 mptcp_pm_announce_addr(msk, addr, true);
233 mptcp_pm_add_addr_send_ack(msk);
234 } else if (mptcp_pm_schedule_work(msk, MPTCP_PM_ADD_ADDR_RECEIVED)) {
235 pm->remote = *addr;
236 } else {
237 __MPTCP_INC_STATS(sock_net((struct sock *)msk), MPTCP_MIB_ADDADDRDROP);
238 }
239
240 spin_unlock_bh(&pm->lock);
241 }
242
243 void mptcp_pm_add_addr_echoed(struct mptcp_sock *msk,
244 const struct mptcp_addr_info *addr)
245 {
246 struct mptcp_pm_data *pm = &msk->pm;
247
248 pr_debug("msk=%p\n", msk);
249
250 spin_lock_bh(&pm->lock);
251
252 if (mptcp_lookup_anno_list_by_saddr(msk, addr) && READ_ONCE(pm->work_pending))
253 mptcp_pm_schedule_work(msk, MPTCP_PM_SUBFLOW_ESTABLISHED);
254
255 spin_unlock_bh(&pm->lock);
256 }
257
258 void mptcp_pm_add_addr_send_ack(struct mptcp_sock *msk)
259 {
260 if (!mptcp_pm_should_add_signal(msk))
261 return;
262
263 mptcp_pm_schedule_work(msk, MPTCP_PM_ADD_ADDR_SEND_ACK);
264 }
265
266 void mptcp_pm_rm_addr_received(struct mptcp_sock *msk,
267 const struct mptcp_rm_list *rm_list)
268 {
269 struct mptcp_pm_data *pm = &msk->pm;
270 u8 i;
271
272 pr_debug("msk=%p remote_ids_nr=%d\n", msk, rm_list->nr);
273
274 for (i = 0; i < rm_list->nr; i++)
275 mptcp_event_addr_removed(msk, rm_list->ids[i]);
276
277 spin_lock_bh(&pm->lock);
278 if (mptcp_pm_schedule_work(msk, MPTCP_PM_RM_ADDR_RECEIVED))
279 pm->rm_list_rx = *rm_list;
280 else
281 __MPTCP_INC_STATS(sock_net((struct sock *)msk), MPTCP_MIB_RMADDRDROP);
282 spin_unlock_bh(&pm->lock);
283 }
284
285 void mptcp_pm_mp_prio_received(struct sock *ssk, u8 bkup)
286 {
287 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
288 struct sock *sk = subflow->conn;
289 struct mptcp_sock *msk;
290
291 pr_debug("subflow->backup=%d, bkup=%d\n", subflow->backup, bkup);
292 msk = mptcp_sk(sk);
293 if (subflow->backup != bkup)
294 subflow->backup = bkup;
295
296 mptcp_event(MPTCP_EVENT_SUB_PRIORITY, msk, ssk, GFP_ATOMIC);
297 }
298
299 void mptcp_pm_mp_fail_received(struct sock *sk, u64 fail_seq)
300 {
301 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
302 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
303
304 pr_debug("fail_seq=%llu\n", fail_seq);
305
306 if (!READ_ONCE(msk->allow_infinite_fallback))
307 return;
308
309 if (!subflow->fail_tout) {
310 pr_debug("send MP_FAIL response and infinite map\n");
311
312 subflow->send_mp_fail = 1;
313 subflow->send_infinite_map = 1;
314 tcp_send_ack(sk);
315 } else {
316 pr_debug("MP_FAIL response received\n");
317 WRITE_ONCE(subflow->fail_tout, 0);
318 }
319 }
320
321 /* path manager helpers */
322
323 bool mptcp_pm_add_addr_signal(struct mptcp_sock *msk, const struct sk_buff *skb,
324 unsigned int opt_size, unsigned int remaining,
325 struct mptcp_addr_info *addr, bool *echo,
326 bool *drop_other_suboptions)
327 {
328 int ret = false;
329 u8 add_addr;
330 u8 family;
331 bool port;
332
333 spin_lock_bh(&msk->pm.lock);
334
335 /* double check after the lock is acquired */
336 if (!mptcp_pm_should_add_signal(msk))
337 goto out_unlock;
338
339 /* always drop every other options for pure ack ADD_ADDR; this is a
340 * plain dup-ack from TCP perspective. The other MPTCP-relevant info,
341 * if any, will be carried by the 'original' TCP ack
342 */
343 if (skb && skb_is_tcp_pure_ack(skb)) {
344 remaining += opt_size;
345 *drop_other_suboptions = true;
346 }
347
348 *echo = mptcp_pm_should_add_signal_echo(msk);
349 port = !!(*echo ? msk->pm.remote.port : msk->pm.local.port);
350
351 family = *echo ? msk->pm.remote.family : msk->pm.local.family;
352 if (remaining < mptcp_add_addr_len(family, *echo, port))
353 goto out_unlock;
354
355 if (*echo) {
356 *addr = msk->pm.remote;
357 add_addr = msk->pm.addr_signal & ~BIT(MPTCP_ADD_ADDR_ECHO);
358 } else {
359 *addr = msk->pm.local;
360 add_addr = msk->pm.addr_signal & ~BIT(MPTCP_ADD_ADDR_SIGNAL);
361 }
362 WRITE_ONCE(msk->pm.addr_signal, add_addr);
363 ret = true;
364
365 out_unlock:
366 spin_unlock_bh(&msk->pm.lock);
367 return ret;
368 }
369
370 bool mptcp_pm_rm_addr_signal(struct mptcp_sock *msk, unsigned int remaining,
371 struct mptcp_rm_list *rm_list)
372 {
373 int ret = false, len;
374 u8 rm_addr;
375
376 spin_lock_bh(&msk->pm.lock);
377
378 /* double check after the lock is acquired */
379 if (!mptcp_pm_should_rm_signal(msk))
380 goto out_unlock;
381
382 rm_addr = msk->pm.addr_signal & ~BIT(MPTCP_RM_ADDR_SIGNAL);
383 len = mptcp_rm_addr_len(&msk->pm.rm_list_tx);
384 if (len < 0) {
385 WRITE_ONCE(msk->pm.addr_signal, rm_addr);
386 goto out_unlock;
387 }
388 if (remaining < len)
389 goto out_unlock;
390
391 *rm_list = msk->pm.rm_list_tx;
392 WRITE_ONCE(msk->pm.addr_signal, rm_addr);
393 ret = true;
394
395 out_unlock:
396 spin_unlock_bh(&msk->pm.lock);
397 return ret;
398 }
399
400 int mptcp_pm_get_local_id(struct mptcp_sock *msk, struct sock_common *skc)
401 {
402 struct mptcp_addr_info skc_local;
403 struct mptcp_addr_info msk_local;
404
405 if (WARN_ON_ONCE(!msk))
406 return -1;
407
408 /* The 0 ID mapping is defined by the first subflow, copied into the msk
409 * addr
410 */
411 mptcp_local_address((struct sock_common *)msk, &msk_local);
412 mptcp_local_address((struct sock_common *)skc, &skc_local);
413 if (mptcp_addresses_equal(&msk_local, &skc_local, false))
414 return 0;
415
416 if (mptcp_pm_is_userspace(msk))
417 return mptcp_userspace_pm_get_local_id(msk, &skc_local);
418 return mptcp_pm_nl_get_local_id(msk, &skc_local);
419 }
420
421 bool mptcp_pm_is_backup(struct mptcp_sock *msk, struct sock_common *skc)
422 {
423 struct mptcp_addr_info skc_local;
424
425 mptcp_local_address((struct sock_common *)skc, &skc_local);
426
427 if (mptcp_pm_is_userspace(msk))
428 return mptcp_userspace_pm_is_backup(msk, &skc_local);
429
430 return mptcp_pm_nl_is_backup(msk, &skc_local);
431 }
432
433 int mptcp_pm_get_flags_and_ifindex_by_id(struct mptcp_sock *msk, unsigned int id,
434 u8 *flags, int *ifindex)
435 {
436 *flags = 0;
437 *ifindex = 0;
438
439 if (mptcp_pm_is_userspace(msk))
440 return mptcp_userspace_pm_get_flags_and_ifindex_by_id(msk, id, flags, ifindex);
441 return mptcp_pm_nl_get_flags_and_ifindex_by_id(msk, id, flags, ifindex);
442 }
443
444 int mptcp_pm_get_addr(struct sk_buff *skb, struct genl_info *info)
445 {
446 if (info->attrs[MPTCP_PM_ATTR_TOKEN])
447 return mptcp_userspace_pm_get_addr(skb, info);
448 return mptcp_pm_nl_get_addr(skb, info);
449 }
450
451 int mptcp_pm_dump_addr(struct sk_buff *msg, struct netlink_callback *cb)
452 {
453 const struct genl_info *info = genl_info_dump(cb);
454
455 if (info->attrs[MPTCP_PM_ATTR_TOKEN])
456 return mptcp_userspace_pm_dump_addr(msg, cb);
457 return mptcp_pm_nl_dump_addr(msg, cb);
458 }
459
460 int mptcp_pm_set_flags(struct sk_buff *skb, struct genl_info *info)
461 {
462 if (info->attrs[MPTCP_PM_ATTR_TOKEN])
463 return mptcp_userspace_pm_set_flags(skb, info);
464 return mptcp_pm_nl_set_flags(skb, info);
465 }
466
467 void mptcp_pm_subflow_chk_stale(const struct mptcp_sock *msk, struct sock *ssk)
468 {
469 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
470 u32 rcv_tstamp = READ_ONCE(tcp_sk(ssk)->rcv_tstamp);
471
472 /* keep track of rtx periods with no progress */
473 if (!subflow->stale_count) {
474 subflow->stale_rcv_tstamp = rcv_tstamp;
475 subflow->stale_count++;
476 } else if (subflow->stale_rcv_tstamp == rcv_tstamp) {
477 if (subflow->stale_count < U8_MAX)
478 subflow->stale_count++;
479 mptcp_pm_nl_subflow_chk_stale(msk, ssk);
480 } else {
481 subflow->stale_count = 0;
482 mptcp_subflow_set_active(subflow);
483 }
484 }
485
486 /* if sk is ipv4 or ipv6_only allows only same-family local and remote addresses,
487 * otherwise allow any matching local/remote pair
488 */
489 bool mptcp_pm_addr_families_match(const struct sock *sk,
490 const struct mptcp_addr_info *loc,
491 const struct mptcp_addr_info *rem)
492 {
493 bool mptcp_is_v4 = sk->sk_family == AF_INET;
494
495 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
496 bool loc_is_v4 = loc->family == AF_INET || ipv6_addr_v4mapped(&loc->addr6);
497 bool rem_is_v4 = rem->family == AF_INET || ipv6_addr_v4mapped(&rem->addr6);
498
499 if (mptcp_is_v4)
500 return loc_is_v4 && rem_is_v4;
501
502 if (ipv6_only_sock(sk))
503 return !loc_is_v4 && !rem_is_v4;
504
505 return loc_is_v4 == rem_is_v4;
506 #else
507 return mptcp_is_v4 && loc->family == AF_INET && rem->family == AF_INET;
508 #endif
509 }
510
511 void mptcp_pm_data_reset(struct mptcp_sock *msk)
512 {
513 u8 pm_type = mptcp_get_pm_type(sock_net((struct sock *)msk));
514 struct mptcp_pm_data *pm = &msk->pm;
515
516 pm->add_addr_signaled = 0;
517 pm->add_addr_accepted = 0;
518 pm->local_addr_used = 0;
519 pm->subflows = 0;
520 pm->rm_list_tx.nr = 0;
521 pm->rm_list_rx.nr = 0;
522 WRITE_ONCE(pm->pm_type, pm_type);
523
524 if (pm_type == MPTCP_PM_TYPE_KERNEL) {
525 bool subflows_allowed = !!mptcp_pm_get_subflows_max(msk);
526
527 /* pm->work_pending must be only be set to 'true' when
528 * pm->pm_type is set to MPTCP_PM_TYPE_KERNEL
529 */
530 WRITE_ONCE(pm->work_pending,
531 (!!mptcp_pm_get_local_addr_max(msk) &&
532 subflows_allowed) ||
533 !!mptcp_pm_get_add_addr_signal_max(msk));
534 WRITE_ONCE(pm->accept_addr,
535 !!mptcp_pm_get_add_addr_accept_max(msk) &&
536 subflows_allowed);
537 WRITE_ONCE(pm->accept_subflow, subflows_allowed);
538 } else {
539 WRITE_ONCE(pm->work_pending, 0);
540 WRITE_ONCE(pm->accept_addr, 0);
541 WRITE_ONCE(pm->accept_subflow, 0);
542 }
543
544 WRITE_ONCE(pm->addr_signal, 0);
545 WRITE_ONCE(pm->remote_deny_join_id0, false);
546 pm->status = 0;
547 bitmap_fill(msk->pm.id_avail_bitmap, MPTCP_PM_MAX_ADDR_ID + 1);
548 }
549
550 void mptcp_pm_data_init(struct mptcp_sock *msk)
551 {
552 spin_lock_init(&msk->pm.lock);
553 INIT_LIST_HEAD(&msk->pm.anno_list);
554 INIT_LIST_HEAD(&msk->pm.userspace_pm_local_addr_list);
555 mptcp_pm_data_reset(msk);
556 }
557
558 void __init mptcp_pm_init(void)
559 {
560 mptcp_pm_nl_init();
561 }
562
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.