1 // SPDX-License-Identifier: GPL-2.0
2 /* Multipath TCP
3 *
4 * Copyright (c) 2020, Red Hat, Inc.
5 */
6
7 #define pr_fmt(fmt) "MPTCP: " fmt
8
9 #include <linux/inet.h>
10 #include <linux/kernel.h>
11 #include <net/inet_common.h>
12 #include <net/netns/generic.h>
13 #include <net/mptcp.h>
14
15 #include "protocol.h"
16 #include "mib.h"
17 #include "mptcp_pm_gen.h"
18
19 static int pm_nl_pernet_id;
20
21 struct mptcp_pm_add_entry {
22 struct list_head list;
23 struct mptcp_addr_info addr;
24 u8 retrans_times;
25 struct timer_list add_timer;
26 struct mptcp_sock *sock;
27 };
28
29 struct pm_nl_pernet {
30 /* protects pernet updates */
31 spinlock_t lock;
32 struct list_head local_addr_list;
33 unsigned int addrs;
34 unsigned int stale_loss_cnt;
35 unsigned int add_addr_signal_max;
36 unsigned int add_addr_accept_max;
37 unsigned int local_addr_max;
38 unsigned int subflows_max;
39 unsigned int next_id;
40 DECLARE_BITMAP(id_bitmap, MPTCP_PM_MAX_ADDR_ID + 1);
41 };
42
43 #define MPTCP_PM_ADDR_MAX 8
44 #define ADD_ADDR_RETRANS_MAX 3
45
pm_nl_get_pernet(const struct net * net)46 static struct pm_nl_pernet *pm_nl_get_pernet(const struct net *net)
47 {
48 return net_generic(net, pm_nl_pernet_id);
49 }
50
51 static struct pm_nl_pernet *
pm_nl_get_pernet_from_msk(const struct mptcp_sock * msk)52 pm_nl_get_pernet_from_msk(const struct mptcp_sock *msk)
53 {
54 return pm_nl_get_pernet(sock_net((struct sock *)msk));
55 }
56
mptcp_addresses_equal(const struct mptcp_addr_info * a,const struct mptcp_addr_info * b,bool use_port)57 bool mptcp_addresses_equal(const struct mptcp_addr_info *a,
58 const struct mptcp_addr_info *b, bool use_port)
59 {
60 bool addr_equals = false;
61
62 if (a->family == b->family) {
63 if (a->family == AF_INET)
64 addr_equals = a->addr.s_addr == b->addr.s_addr;
65 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
66 else
67 addr_equals = !ipv6_addr_cmp(&a->addr6, &b->addr6);
68 } else if (a->family == AF_INET) {
69 if (ipv6_addr_v4mapped(&b->addr6))
70 addr_equals = a->addr.s_addr == b->addr6.s6_addr32[3];
71 } else if (b->family == AF_INET) {
72 if (ipv6_addr_v4mapped(&a->addr6))
73 addr_equals = a->addr6.s6_addr32[3] == b->addr.s_addr;
74 #endif
75 }
76
77 if (!addr_equals)
78 return false;
79 if (!use_port)
80 return true;
81
82 return a->port == b->port;
83 }
84
mptcp_local_address(const struct sock_common * skc,struct mptcp_addr_info * addr)85 void mptcp_local_address(const struct sock_common *skc, struct mptcp_addr_info *addr)
86 {
87 addr->family = skc->skc_family;
88 addr->port = htons(skc->skc_num);
89 if (addr->family == AF_INET)
90 addr->addr.s_addr = skc->skc_rcv_saddr;
91 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
92 else if (addr->family == AF_INET6)
93 addr->addr6 = skc->skc_v6_rcv_saddr;
94 #endif
95 }
96
remote_address(const struct sock_common * skc,struct mptcp_addr_info * addr)97 static void remote_address(const struct sock_common *skc,
98 struct mptcp_addr_info *addr)
99 {
100 addr->family = skc->skc_family;
101 addr->port = skc->skc_dport;
102 if (addr->family == AF_INET)
103 addr->addr.s_addr = skc->skc_daddr;
104 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
105 else if (addr->family == AF_INET6)
106 addr->addr6 = skc->skc_v6_daddr;
107 #endif
108 }
109
lookup_subflow_by_saddr(const struct list_head * list,const struct mptcp_addr_info * saddr)110 static bool lookup_subflow_by_saddr(const struct list_head *list,
111 const struct mptcp_addr_info *saddr)
112 {
113 struct mptcp_subflow_context *subflow;
114 struct mptcp_addr_info cur;
115 struct sock_common *skc;
116
117 list_for_each_entry(subflow, list, node) {
118 skc = (struct sock_common *)mptcp_subflow_tcp_sock(subflow);
119
120 mptcp_local_address(skc, &cur);
121 if (mptcp_addresses_equal(&cur, saddr, saddr->port))
122 return true;
123 }
124
125 return false;
126 }
127
lookup_subflow_by_daddr(const struct list_head * list,const struct mptcp_addr_info * daddr)128 static bool lookup_subflow_by_daddr(const struct list_head *list,
129 const struct mptcp_addr_info *daddr)
130 {
131 struct mptcp_subflow_context *subflow;
132 struct mptcp_addr_info cur;
133
134 list_for_each_entry(subflow, list, node) {
135 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
136
137 if (!((1 << inet_sk_state_load(ssk)) &
138 (TCPF_ESTABLISHED | TCPF_SYN_SENT | TCPF_SYN_RECV)))
139 continue;
140
141 remote_address((struct sock_common *)ssk, &cur);
142 if (mptcp_addresses_equal(&cur, daddr, daddr->port))
143 return true;
144 }
145
146 return false;
147 }
148
149 static bool
select_local_address(const struct pm_nl_pernet * pernet,const struct mptcp_sock * msk,struct mptcp_pm_local * new_local)150 select_local_address(const struct pm_nl_pernet *pernet,
151 const struct mptcp_sock *msk,
152 struct mptcp_pm_local *new_local)
153 {
154 struct mptcp_pm_addr_entry *entry;
155 bool found = false;
156
157 msk_owned_by_me(msk);
158
159 rcu_read_lock();
160 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) {
161 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW))
162 continue;
163
164 if (!test_bit(entry->addr.id, msk->pm.id_avail_bitmap))
165 continue;
166
167 new_local->addr = entry->addr;
168 new_local->flags = entry->flags;
169 new_local->ifindex = entry->ifindex;
170 found = true;
171 break;
172 }
173 rcu_read_unlock();
174
175 return found;
176 }
177
178 static bool
select_signal_address(struct pm_nl_pernet * pernet,const struct mptcp_sock * msk,struct mptcp_pm_local * new_local)179 select_signal_address(struct pm_nl_pernet *pernet, const struct mptcp_sock *msk,
180 struct mptcp_pm_local *new_local)
181 {
182 struct mptcp_pm_addr_entry *entry;
183 bool found = false;
184
185 rcu_read_lock();
186 /* do not keep any additional per socket state, just signal
187 * the address list in order.
188 * Note: removal from the local address list during the msk life-cycle
189 * can lead to additional addresses not being announced.
190 */
191 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) {
192 if (!test_bit(entry->addr.id, msk->pm.id_avail_bitmap))
193 continue;
194
195 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL))
196 continue;
197
198 new_local->addr = entry->addr;
199 new_local->flags = entry->flags;
200 new_local->ifindex = entry->ifindex;
201 found = true;
202 break;
203 }
204 rcu_read_unlock();
205
206 return found;
207 }
208
mptcp_pm_get_add_addr_signal_max(const struct mptcp_sock * msk)209 unsigned int mptcp_pm_get_add_addr_signal_max(const struct mptcp_sock *msk)
210 {
211 const struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk);
212
213 return READ_ONCE(pernet->add_addr_signal_max);
214 }
215 EXPORT_SYMBOL_GPL(mptcp_pm_get_add_addr_signal_max);
216
mptcp_pm_get_add_addr_accept_max(const struct mptcp_sock * msk)217 unsigned int mptcp_pm_get_add_addr_accept_max(const struct mptcp_sock *msk)
218 {
219 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk);
220
221 return READ_ONCE(pernet->add_addr_accept_max);
222 }
223 EXPORT_SYMBOL_GPL(mptcp_pm_get_add_addr_accept_max);
224
mptcp_pm_get_subflows_max(const struct mptcp_sock * msk)225 unsigned int mptcp_pm_get_subflows_max(const struct mptcp_sock *msk)
226 {
227 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk);
228
229 return READ_ONCE(pernet->subflows_max);
230 }
231 EXPORT_SYMBOL_GPL(mptcp_pm_get_subflows_max);
232
mptcp_pm_get_local_addr_max(const struct mptcp_sock * msk)233 unsigned int mptcp_pm_get_local_addr_max(const struct mptcp_sock *msk)
234 {
235 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk);
236
237 return READ_ONCE(pernet->local_addr_max);
238 }
239 EXPORT_SYMBOL_GPL(mptcp_pm_get_local_addr_max);
240
mptcp_pm_nl_check_work_pending(struct mptcp_sock * msk)241 bool mptcp_pm_nl_check_work_pending(struct mptcp_sock *msk)
242 {
243 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk);
244
245 if (msk->pm.subflows == mptcp_pm_get_subflows_max(msk) ||
246 (find_next_and_bit(pernet->id_bitmap, msk->pm.id_avail_bitmap,
247 MPTCP_PM_MAX_ADDR_ID + 1, 0) == MPTCP_PM_MAX_ADDR_ID + 1)) {
248 WRITE_ONCE(msk->pm.work_pending, false);
249 return false;
250 }
251 return true;
252 }
253
254 struct mptcp_pm_add_entry *
mptcp_lookup_anno_list_by_saddr(const struct mptcp_sock * msk,const struct mptcp_addr_info * addr)255 mptcp_lookup_anno_list_by_saddr(const struct mptcp_sock *msk,
256 const struct mptcp_addr_info *addr)
257 {
258 struct mptcp_pm_add_entry *entry;
259
260 lockdep_assert_held(&msk->pm.lock);
261
262 list_for_each_entry(entry, &msk->pm.anno_list, list) {
263 if (mptcp_addresses_equal(&entry->addr, addr, true))
264 return entry;
265 }
266
267 return NULL;
268 }
269
mptcp_pm_sport_in_anno_list(struct mptcp_sock * msk,const struct sock * sk)270 bool mptcp_pm_sport_in_anno_list(struct mptcp_sock *msk, const struct sock *sk)
271 {
272 struct mptcp_pm_add_entry *entry;
273 struct mptcp_addr_info saddr;
274 bool ret = false;
275
276 mptcp_local_address((struct sock_common *)sk, &saddr);
277
278 spin_lock_bh(&msk->pm.lock);
279 list_for_each_entry(entry, &msk->pm.anno_list, list) {
280 if (mptcp_addresses_equal(&entry->addr, &saddr, true)) {
281 ret = true;
282 goto out;
283 }
284 }
285
286 out:
287 spin_unlock_bh(&msk->pm.lock);
288 return ret;
289 }
290
mptcp_pm_add_timer(struct timer_list * timer)291 static void mptcp_pm_add_timer(struct timer_list *timer)
292 {
293 struct mptcp_pm_add_entry *entry = from_timer(entry, timer, add_timer);
294 struct mptcp_sock *msk = entry->sock;
295 struct sock *sk = (struct sock *)msk;
296 unsigned int timeout;
297
298 pr_debug("msk=%p\n", msk);
299
300 if (!msk)
301 return;
302
303 if (inet_sk_state_load(sk) == TCP_CLOSE)
304 return;
305
306 if (!entry->addr.id)
307 return;
308
309 if (mptcp_pm_should_add_signal_addr(msk)) {
310 sk_reset_timer(sk, timer, jiffies + TCP_RTO_MAX / 8);
311 goto out;
312 }
313
314 timeout = mptcp_get_add_addr_timeout(sock_net(sk));
315 if (!timeout)
316 goto out;
317
318 spin_lock_bh(&msk->pm.lock);
319
320 if (!mptcp_pm_should_add_signal_addr(msk)) {
321 pr_debug("retransmit ADD_ADDR id=%d\n", entry->addr.id);
322 mptcp_pm_announce_addr(msk, &entry->addr, false);
323 mptcp_pm_add_addr_send_ack(msk);
324 entry->retrans_times++;
325 }
326
327 if (entry->retrans_times < ADD_ADDR_RETRANS_MAX)
328 sk_reset_timer(sk, timer,
329 jiffies + timeout);
330
331 spin_unlock_bh(&msk->pm.lock);
332
333 if (entry->retrans_times == ADD_ADDR_RETRANS_MAX)
334 mptcp_pm_subflow_established(msk);
335
336 out:
337 __sock_put(sk);
338 }
339
340 struct mptcp_pm_add_entry *
mptcp_pm_del_add_timer(struct mptcp_sock * msk,const struct mptcp_addr_info * addr,bool check_id)341 mptcp_pm_del_add_timer(struct mptcp_sock *msk,
342 const struct mptcp_addr_info *addr, bool check_id)
343 {
344 struct mptcp_pm_add_entry *entry;
345 struct sock *sk = (struct sock *)msk;
346 struct timer_list *add_timer = NULL;
347
348 spin_lock_bh(&msk->pm.lock);
349 entry = mptcp_lookup_anno_list_by_saddr(msk, addr);
350 if (entry && (!check_id || entry->addr.id == addr->id)) {
351 entry->retrans_times = ADD_ADDR_RETRANS_MAX;
352 add_timer = &entry->add_timer;
353 }
354 if (!check_id && entry)
355 list_del(&entry->list);
356 spin_unlock_bh(&msk->pm.lock);
357
358 /* no lock, because sk_stop_timer_sync() is calling del_timer_sync() */
359 if (add_timer)
360 sk_stop_timer_sync(sk, add_timer);
361
362 return entry;
363 }
364
mptcp_pm_alloc_anno_list(struct mptcp_sock * msk,const struct mptcp_addr_info * addr)365 bool mptcp_pm_alloc_anno_list(struct mptcp_sock *msk,
366 const struct mptcp_addr_info *addr)
367 {
368 struct mptcp_pm_add_entry *add_entry = NULL;
369 struct sock *sk = (struct sock *)msk;
370 struct net *net = sock_net(sk);
371 unsigned int timeout;
372
373 lockdep_assert_held(&msk->pm.lock);
374
375 add_entry = mptcp_lookup_anno_list_by_saddr(msk, addr);
376
377 if (add_entry) {
378 if (WARN_ON_ONCE(mptcp_pm_is_kernel(msk)))
379 return false;
380
381 goto reset_timer;
382 }
383
384 add_entry = kmalloc(sizeof(*add_entry), GFP_ATOMIC);
385 if (!add_entry)
386 return false;
387
388 list_add(&add_entry->list, &msk->pm.anno_list);
389
390 add_entry->addr = *addr;
391 add_entry->sock = msk;
392 add_entry->retrans_times = 0;
393
394 timer_setup(&add_entry->add_timer, mptcp_pm_add_timer, 0);
395 reset_timer:
396 timeout = mptcp_get_add_addr_timeout(net);
397 if (timeout)
398 sk_reset_timer(sk, &add_entry->add_timer, jiffies + timeout);
399
400 return true;
401 }
402
mptcp_pm_free_anno_list(struct mptcp_sock * msk)403 void mptcp_pm_free_anno_list(struct mptcp_sock *msk)
404 {
405 struct mptcp_pm_add_entry *entry, *tmp;
406 struct sock *sk = (struct sock *)msk;
407 LIST_HEAD(free_list);
408
409 pr_debug("msk=%p\n", msk);
410
411 spin_lock_bh(&msk->pm.lock);
412 list_splice_init(&msk->pm.anno_list, &free_list);
413 spin_unlock_bh(&msk->pm.lock);
414
415 list_for_each_entry_safe(entry, tmp, &free_list, list) {
416 sk_stop_timer_sync(sk, &entry->add_timer);
417 kfree(entry);
418 }
419 }
420
421 /* Fill all the remote addresses into the array addrs[],
422 * and return the array size.
423 */
fill_remote_addresses_vec(struct mptcp_sock * msk,struct mptcp_addr_info * local,bool fullmesh,struct mptcp_addr_info * addrs)424 static unsigned int fill_remote_addresses_vec(struct mptcp_sock *msk,
425 struct mptcp_addr_info *local,
426 bool fullmesh,
427 struct mptcp_addr_info *addrs)
428 {
429 bool deny_id0 = READ_ONCE(msk->pm.remote_deny_join_id0);
430 struct sock *sk = (struct sock *)msk, *ssk;
431 struct mptcp_subflow_context *subflow;
432 struct mptcp_addr_info remote = { 0 };
433 unsigned int subflows_max;
434 int i = 0;
435
436 subflows_max = mptcp_pm_get_subflows_max(msk);
437 remote_address((struct sock_common *)sk, &remote);
438
439 /* Non-fullmesh endpoint, fill in the single entry
440 * corresponding to the primary MPC subflow remote address
441 */
442 if (!fullmesh) {
443 if (deny_id0)
444 return 0;
445
446 if (!mptcp_pm_addr_families_match(sk, local, &remote))
447 return 0;
448
449 msk->pm.subflows++;
450 addrs[i++] = remote;
451 } else {
452 DECLARE_BITMAP(unavail_id, MPTCP_PM_MAX_ADDR_ID + 1);
453
454 /* Forbid creation of new subflows matching existing
455 * ones, possibly already created by incoming ADD_ADDR
456 */
457 bitmap_zero(unavail_id, MPTCP_PM_MAX_ADDR_ID + 1);
458 mptcp_for_each_subflow(msk, subflow)
459 if (READ_ONCE(subflow->local_id) == local->id)
460 __set_bit(subflow->remote_id, unavail_id);
461
462 mptcp_for_each_subflow(msk, subflow) {
463 ssk = mptcp_subflow_tcp_sock(subflow);
464 remote_address((struct sock_common *)ssk, &addrs[i]);
465 addrs[i].id = READ_ONCE(subflow->remote_id);
466 if (deny_id0 && !addrs[i].id)
467 continue;
468
469 if (test_bit(addrs[i].id, unavail_id))
470 continue;
471
472 if (!mptcp_pm_addr_families_match(sk, local, &addrs[i]))
473 continue;
474
475 if (msk->pm.subflows < subflows_max) {
476 /* forbid creating multiple address towards
477 * this id
478 */
479 __set_bit(addrs[i].id, unavail_id);
480 msk->pm.subflows++;
481 i++;
482 }
483 }
484 }
485
486 return i;
487 }
488
__mptcp_pm_send_ack(struct mptcp_sock * msk,struct mptcp_subflow_context * subflow,bool prio,bool backup)489 static void __mptcp_pm_send_ack(struct mptcp_sock *msk, struct mptcp_subflow_context *subflow,
490 bool prio, bool backup)
491 {
492 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
493 bool slow;
494
495 pr_debug("send ack for %s\n",
496 prio ? "mp_prio" : (mptcp_pm_should_add_signal(msk) ? "add_addr" : "rm_addr"));
497
498 slow = lock_sock_fast(ssk);
499 if (prio) {
500 subflow->send_mp_prio = 1;
501 subflow->request_bkup = backup;
502 }
503
504 __mptcp_subflow_send_ack(ssk);
505 unlock_sock_fast(ssk, slow);
506 }
507
mptcp_pm_send_ack(struct mptcp_sock * msk,struct mptcp_subflow_context * subflow,bool prio,bool backup)508 static void mptcp_pm_send_ack(struct mptcp_sock *msk, struct mptcp_subflow_context *subflow,
509 bool prio, bool backup)
510 {
511 spin_unlock_bh(&msk->pm.lock);
512 __mptcp_pm_send_ack(msk, subflow, prio, backup);
513 spin_lock_bh(&msk->pm.lock);
514 }
515
516 static struct mptcp_pm_addr_entry *
__lookup_addr_by_id(struct pm_nl_pernet * pernet,unsigned int id)517 __lookup_addr_by_id(struct pm_nl_pernet *pernet, unsigned int id)
518 {
519 struct mptcp_pm_addr_entry *entry;
520
521 list_for_each_entry(entry, &pernet->local_addr_list, list) {
522 if (entry->addr.id == id)
523 return entry;
524 }
525 return NULL;
526 }
527
528 static struct mptcp_pm_addr_entry *
__lookup_addr(struct pm_nl_pernet * pernet,const struct mptcp_addr_info * info)529 __lookup_addr(struct pm_nl_pernet *pernet, const struct mptcp_addr_info *info)
530 {
531 struct mptcp_pm_addr_entry *entry;
532
533 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list,
534 lockdep_is_held(&pernet->lock)) {
535 if (mptcp_addresses_equal(&entry->addr, info, entry->addr.port))
536 return entry;
537 }
538 return NULL;
539 }
540
mptcp_pm_create_subflow_or_signal_addr(struct mptcp_sock * msk)541 static void mptcp_pm_create_subflow_or_signal_addr(struct mptcp_sock *msk)
542 {
543 struct sock *sk = (struct sock *)msk;
544 unsigned int add_addr_signal_max;
545 bool signal_and_subflow = false;
546 unsigned int local_addr_max;
547 struct pm_nl_pernet *pernet;
548 struct mptcp_pm_local local;
549 unsigned int subflows_max;
550
551 pernet = pm_nl_get_pernet(sock_net(sk));
552
553 add_addr_signal_max = mptcp_pm_get_add_addr_signal_max(msk);
554 local_addr_max = mptcp_pm_get_local_addr_max(msk);
555 subflows_max = mptcp_pm_get_subflows_max(msk);
556
557 /* do lazy endpoint usage accounting for the MPC subflows */
558 if (unlikely(!(msk->pm.status & BIT(MPTCP_PM_MPC_ENDPOINT_ACCOUNTED))) && msk->first) {
559 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(msk->first);
560 struct mptcp_pm_addr_entry *entry;
561 struct mptcp_addr_info mpc_addr;
562 bool backup = false;
563
564 mptcp_local_address((struct sock_common *)msk->first, &mpc_addr);
565 rcu_read_lock();
566 entry = __lookup_addr(pernet, &mpc_addr);
567 if (entry) {
568 __clear_bit(entry->addr.id, msk->pm.id_avail_bitmap);
569 msk->mpc_endpoint_id = entry->addr.id;
570 backup = !!(entry->flags & MPTCP_PM_ADDR_FLAG_BACKUP);
571 }
572 rcu_read_unlock();
573
574 if (backup)
575 mptcp_pm_send_ack(msk, subflow, true, backup);
576
577 msk->pm.status |= BIT(MPTCP_PM_MPC_ENDPOINT_ACCOUNTED);
578 }
579
580 pr_debug("local %d:%d signal %d:%d subflows %d:%d\n",
581 msk->pm.local_addr_used, local_addr_max,
582 msk->pm.add_addr_signaled, add_addr_signal_max,
583 msk->pm.subflows, subflows_max);
584
585 /* check first for announce */
586 if (msk->pm.add_addr_signaled < add_addr_signal_max) {
587 /* due to racing events on both ends we can reach here while
588 * previous add address is still running: if we invoke now
589 * mptcp_pm_announce_addr(), that will fail and the
590 * corresponding id will be marked as used.
591 * Instead let the PM machinery reschedule us when the
592 * current address announce will be completed.
593 */
594 if (msk->pm.addr_signal & BIT(MPTCP_ADD_ADDR_SIGNAL))
595 return;
596
597 if (!select_signal_address(pernet, msk, &local))
598 goto subflow;
599
600 /* If the alloc fails, we are on memory pressure, not worth
601 * continuing, and trying to create subflows.
602 */
603 if (!mptcp_pm_alloc_anno_list(msk, &local.addr))
604 return;
605
606 __clear_bit(local.addr.id, msk->pm.id_avail_bitmap);
607 msk->pm.add_addr_signaled++;
608
609 /* Special case for ID0: set the correct ID */
610 if (local.addr.id == msk->mpc_endpoint_id)
611 local.addr.id = 0;
612
613 mptcp_pm_announce_addr(msk, &local.addr, false);
614 mptcp_pm_nl_addr_send_ack(msk);
615
616 if (local.flags & MPTCP_PM_ADDR_FLAG_SUBFLOW)
617 signal_and_subflow = true;
618 }
619
620 subflow:
621 /* check if should create a new subflow */
622 while (msk->pm.local_addr_used < local_addr_max &&
623 msk->pm.subflows < subflows_max) {
624 struct mptcp_addr_info addrs[MPTCP_PM_ADDR_MAX];
625 bool fullmesh;
626 int i, nr;
627
628 if (signal_and_subflow)
629 signal_and_subflow = false;
630 else if (!select_local_address(pernet, msk, &local))
631 break;
632
633 fullmesh = !!(local.flags & MPTCP_PM_ADDR_FLAG_FULLMESH);
634
635 __clear_bit(local.addr.id, msk->pm.id_avail_bitmap);
636
637 /* Special case for ID0: set the correct ID */
638 if (local.addr.id == msk->mpc_endpoint_id)
639 local.addr.id = 0;
640 else /* local_addr_used is not decr for ID 0 */
641 msk->pm.local_addr_used++;
642
643 nr = fill_remote_addresses_vec(msk, &local.addr, fullmesh, addrs);
644 if (nr == 0)
645 continue;
646
647 spin_unlock_bh(&msk->pm.lock);
648 for (i = 0; i < nr; i++)
649 __mptcp_subflow_connect(sk, &local, &addrs[i]);
650 spin_lock_bh(&msk->pm.lock);
651 }
652 mptcp_pm_nl_check_work_pending(msk);
653 }
654
mptcp_pm_nl_fully_established(struct mptcp_sock * msk)655 static void mptcp_pm_nl_fully_established(struct mptcp_sock *msk)
656 {
657 mptcp_pm_create_subflow_or_signal_addr(msk);
658 }
659
mptcp_pm_nl_subflow_established(struct mptcp_sock * msk)660 static void mptcp_pm_nl_subflow_established(struct mptcp_sock *msk)
661 {
662 mptcp_pm_create_subflow_or_signal_addr(msk);
663 }
664
665 /* Fill all the local addresses into the array addrs[],
666 * and return the array size.
667 */
fill_local_addresses_vec(struct mptcp_sock * msk,struct mptcp_addr_info * remote,struct mptcp_pm_local * locals)668 static unsigned int fill_local_addresses_vec(struct mptcp_sock *msk,
669 struct mptcp_addr_info *remote,
670 struct mptcp_pm_local *locals)
671 {
672 struct sock *sk = (struct sock *)msk;
673 struct mptcp_pm_addr_entry *entry;
674 struct mptcp_addr_info mpc_addr;
675 struct pm_nl_pernet *pernet;
676 unsigned int subflows_max;
677 int i = 0;
678
679 pernet = pm_nl_get_pernet_from_msk(msk);
680 subflows_max = mptcp_pm_get_subflows_max(msk);
681
682 mptcp_local_address((struct sock_common *)msk, &mpc_addr);
683
684 rcu_read_lock();
685 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) {
686 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_FULLMESH))
687 continue;
688
689 if (!mptcp_pm_addr_families_match(sk, &entry->addr, remote))
690 continue;
691
692 if (msk->pm.subflows < subflows_max) {
693 locals[i].addr = entry->addr;
694 locals[i].flags = entry->flags;
695 locals[i].ifindex = entry->ifindex;
696
697 /* Special case for ID0: set the correct ID */
698 if (mptcp_addresses_equal(&locals[i].addr, &mpc_addr, locals[i].addr.port))
699 locals[i].addr.id = 0;
700
701 msk->pm.subflows++;
702 i++;
703 }
704 }
705 rcu_read_unlock();
706
707 /* If the array is empty, fill in the single
708 * 'IPADDRANY' local address
709 */
710 if (!i) {
711 memset(&locals[i], 0, sizeof(locals[i]));
712 locals[i].addr.family =
713 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
714 remote->family == AF_INET6 &&
715 ipv6_addr_v4mapped(&remote->addr6) ? AF_INET :
716 #endif
717 remote->family;
718
719 if (!mptcp_pm_addr_families_match(sk, &locals[i].addr, remote))
720 return 0;
721
722 msk->pm.subflows++;
723 i++;
724 }
725
726 return i;
727 }
728
mptcp_pm_nl_add_addr_received(struct mptcp_sock * msk)729 static void mptcp_pm_nl_add_addr_received(struct mptcp_sock *msk)
730 {
731 struct mptcp_pm_local locals[MPTCP_PM_ADDR_MAX];
732 struct sock *sk = (struct sock *)msk;
733 unsigned int add_addr_accept_max;
734 struct mptcp_addr_info remote;
735 unsigned int subflows_max;
736 bool sf_created = false;
737 int i, nr;
738
739 add_addr_accept_max = mptcp_pm_get_add_addr_accept_max(msk);
740 subflows_max = mptcp_pm_get_subflows_max(msk);
741
742 pr_debug("accepted %d:%d remote family %d\n",
743 msk->pm.add_addr_accepted, add_addr_accept_max,
744 msk->pm.remote.family);
745
746 remote = msk->pm.remote;
747 mptcp_pm_announce_addr(msk, &remote, true);
748 mptcp_pm_nl_addr_send_ack(msk);
749
750 if (lookup_subflow_by_daddr(&msk->conn_list, &remote))
751 return;
752
753 /* pick id 0 port, if none is provided the remote address */
754 if (!remote.port)
755 remote.port = sk->sk_dport;
756
757 /* connect to the specified remote address, using whatever
758 * local address the routing configuration will pick.
759 */
760 nr = fill_local_addresses_vec(msk, &remote, locals);
761 if (nr == 0)
762 return;
763
764 spin_unlock_bh(&msk->pm.lock);
765 for (i = 0; i < nr; i++)
766 if (__mptcp_subflow_connect(sk, &locals[i], &remote) == 0)
767 sf_created = true;
768 spin_lock_bh(&msk->pm.lock);
769
770 if (sf_created) {
771 /* add_addr_accepted is not decr for ID 0 */
772 if (remote.id)
773 msk->pm.add_addr_accepted++;
774 if (msk->pm.add_addr_accepted >= add_addr_accept_max ||
775 msk->pm.subflows >= subflows_max)
776 WRITE_ONCE(msk->pm.accept_addr, false);
777 }
778 }
779
mptcp_pm_nl_is_init_remote_addr(struct mptcp_sock * msk,const struct mptcp_addr_info * remote)780 bool mptcp_pm_nl_is_init_remote_addr(struct mptcp_sock *msk,
781 const struct mptcp_addr_info *remote)
782 {
783 struct mptcp_addr_info mpc_remote;
784
785 remote_address((struct sock_common *)msk, &mpc_remote);
786 return mptcp_addresses_equal(&mpc_remote, remote, remote->port);
787 }
788
mptcp_pm_nl_addr_send_ack(struct mptcp_sock * msk)789 void mptcp_pm_nl_addr_send_ack(struct mptcp_sock *msk)
790 {
791 struct mptcp_subflow_context *subflow;
792
793 msk_owned_by_me(msk);
794 lockdep_assert_held(&msk->pm.lock);
795
796 if (!mptcp_pm_should_add_signal(msk) &&
797 !mptcp_pm_should_rm_signal(msk))
798 return;
799
800 mptcp_for_each_subflow(msk, subflow) {
801 if (__mptcp_subflow_active(subflow)) {
802 mptcp_pm_send_ack(msk, subflow, false, false);
803 break;
804 }
805 }
806 }
807
mptcp_pm_nl_mp_prio_send_ack(struct mptcp_sock * msk,struct mptcp_addr_info * addr,struct mptcp_addr_info * rem,u8 bkup)808 int mptcp_pm_nl_mp_prio_send_ack(struct mptcp_sock *msk,
809 struct mptcp_addr_info *addr,
810 struct mptcp_addr_info *rem,
811 u8 bkup)
812 {
813 struct mptcp_subflow_context *subflow;
814
815 pr_debug("bkup=%d\n", bkup);
816
817 mptcp_for_each_subflow(msk, subflow) {
818 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
819 struct mptcp_addr_info local, remote;
820
821 mptcp_local_address((struct sock_common *)ssk, &local);
822 if (!mptcp_addresses_equal(&local, addr, addr->port))
823 continue;
824
825 if (rem && rem->family != AF_UNSPEC) {
826 remote_address((struct sock_common *)ssk, &remote);
827 if (!mptcp_addresses_equal(&remote, rem, rem->port))
828 continue;
829 }
830
831 __mptcp_pm_send_ack(msk, subflow, true, bkup);
832 return 0;
833 }
834
835 return -EINVAL;
836 }
837
mptcp_pm_nl_rm_addr_or_subflow(struct mptcp_sock * msk,const struct mptcp_rm_list * rm_list,enum linux_mptcp_mib_field rm_type)838 static void mptcp_pm_nl_rm_addr_or_subflow(struct mptcp_sock *msk,
839 const struct mptcp_rm_list *rm_list,
840 enum linux_mptcp_mib_field rm_type)
841 {
842 struct mptcp_subflow_context *subflow, *tmp;
843 struct sock *sk = (struct sock *)msk;
844 u8 i;
845
846 pr_debug("%s rm_list_nr %d\n",
847 rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow", rm_list->nr);
848
849 msk_owned_by_me(msk);
850
851 if (sk->sk_state == TCP_LISTEN)
852 return;
853
854 if (!rm_list->nr)
855 return;
856
857 if (list_empty(&msk->conn_list))
858 return;
859
860 for (i = 0; i < rm_list->nr; i++) {
861 u8 rm_id = rm_list->ids[i];
862 bool removed = false;
863
864 mptcp_for_each_subflow_safe(msk, subflow, tmp) {
865 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
866 u8 remote_id = READ_ONCE(subflow->remote_id);
867 int how = RCV_SHUTDOWN | SEND_SHUTDOWN;
868 u8 id = subflow_get_local_id(subflow);
869
870 if ((1 << inet_sk_state_load(ssk)) &
871 (TCPF_FIN_WAIT1 | TCPF_FIN_WAIT2 | TCPF_CLOSING | TCPF_CLOSE))
872 continue;
873 if (rm_type == MPTCP_MIB_RMADDR && remote_id != rm_id)
874 continue;
875 if (rm_type == MPTCP_MIB_RMSUBFLOW && id != rm_id)
876 continue;
877
878 pr_debug(" -> %s rm_list_ids[%d]=%u local_id=%u remote_id=%u mpc_id=%u\n",
879 rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow",
880 i, rm_id, id, remote_id, msk->mpc_endpoint_id);
881 spin_unlock_bh(&msk->pm.lock);
882 mptcp_subflow_shutdown(sk, ssk, how);
883 removed |= subflow->request_join;
884
885 /* the following takes care of updating the subflows counter */
886 mptcp_close_ssk(sk, ssk, subflow);
887 spin_lock_bh(&msk->pm.lock);
888
889 if (rm_type == MPTCP_MIB_RMSUBFLOW)
890 __MPTCP_INC_STATS(sock_net(sk), rm_type);
891 }
892
893 if (rm_type == MPTCP_MIB_RMADDR)
894 __MPTCP_INC_STATS(sock_net(sk), rm_type);
895
896 if (!removed)
897 continue;
898
899 if (!mptcp_pm_is_kernel(msk))
900 continue;
901
902 if (rm_type == MPTCP_MIB_RMADDR && rm_id &&
903 !WARN_ON_ONCE(msk->pm.add_addr_accepted == 0)) {
904 /* Note: if the subflow has been closed before, this
905 * add_addr_accepted counter will not be decremented.
906 */
907 if (--msk->pm.add_addr_accepted < mptcp_pm_get_add_addr_accept_max(msk))
908 WRITE_ONCE(msk->pm.accept_addr, true);
909 }
910 }
911 }
912
mptcp_pm_nl_rm_addr_received(struct mptcp_sock * msk)913 static void mptcp_pm_nl_rm_addr_received(struct mptcp_sock *msk)
914 {
915 mptcp_pm_nl_rm_addr_or_subflow(msk, &msk->pm.rm_list_rx, MPTCP_MIB_RMADDR);
916 }
917
mptcp_pm_nl_rm_subflow_received(struct mptcp_sock * msk,const struct mptcp_rm_list * rm_list)918 static void mptcp_pm_nl_rm_subflow_received(struct mptcp_sock *msk,
919 const struct mptcp_rm_list *rm_list)
920 {
921 mptcp_pm_nl_rm_addr_or_subflow(msk, rm_list, MPTCP_MIB_RMSUBFLOW);
922 }
923
mptcp_pm_nl_work(struct mptcp_sock * msk)924 void mptcp_pm_nl_work(struct mptcp_sock *msk)
925 {
926 struct mptcp_pm_data *pm = &msk->pm;
927
928 msk_owned_by_me(msk);
929
930 if (!(pm->status & MPTCP_PM_WORK_MASK))
931 return;
932
933 spin_lock_bh(&msk->pm.lock);
934
935 pr_debug("msk=%p status=%x\n", msk, pm->status);
936 if (pm->status & BIT(MPTCP_PM_ADD_ADDR_RECEIVED)) {
937 pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_RECEIVED);
938 mptcp_pm_nl_add_addr_received(msk);
939 }
940 if (pm->status & BIT(MPTCP_PM_ADD_ADDR_SEND_ACK)) {
941 pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_SEND_ACK);
942 mptcp_pm_nl_addr_send_ack(msk);
943 }
944 if (pm->status & BIT(MPTCP_PM_RM_ADDR_RECEIVED)) {
945 pm->status &= ~BIT(MPTCP_PM_RM_ADDR_RECEIVED);
946 mptcp_pm_nl_rm_addr_received(msk);
947 }
948 if (pm->status & BIT(MPTCP_PM_ESTABLISHED)) {
949 pm->status &= ~BIT(MPTCP_PM_ESTABLISHED);
950 mptcp_pm_nl_fully_established(msk);
951 }
952 if (pm->status & BIT(MPTCP_PM_SUBFLOW_ESTABLISHED)) {
953 pm->status &= ~BIT(MPTCP_PM_SUBFLOW_ESTABLISHED);
954 mptcp_pm_nl_subflow_established(msk);
955 }
956
957 spin_unlock_bh(&msk->pm.lock);
958 }
959
address_use_port(struct mptcp_pm_addr_entry * entry)960 static bool address_use_port(struct mptcp_pm_addr_entry *entry)
961 {
962 return (entry->flags &
963 (MPTCP_PM_ADDR_FLAG_SIGNAL | MPTCP_PM_ADDR_FLAG_SUBFLOW)) ==
964 MPTCP_PM_ADDR_FLAG_SIGNAL;
965 }
966
967 /* caller must ensure the RCU grace period is already elapsed */
__mptcp_pm_release_addr_entry(struct mptcp_pm_addr_entry * entry)968 static void __mptcp_pm_release_addr_entry(struct mptcp_pm_addr_entry *entry)
969 {
970 if (entry->lsk)
971 sock_release(entry->lsk);
972 kfree(entry);
973 }
974
mptcp_pm_nl_append_new_local_addr(struct pm_nl_pernet * pernet,struct mptcp_pm_addr_entry * entry,bool needs_id,bool replace)975 static int mptcp_pm_nl_append_new_local_addr(struct pm_nl_pernet *pernet,
976 struct mptcp_pm_addr_entry *entry,
977 bool needs_id, bool replace)
978 {
979 struct mptcp_pm_addr_entry *cur, *del_entry = NULL;
980 unsigned int addr_max;
981 int ret = -EINVAL;
982
983 spin_lock_bh(&pernet->lock);
984 /* to keep the code simple, don't do IDR-like allocation for address ID,
985 * just bail when we exceed limits
986 */
987 if (pernet->next_id == MPTCP_PM_MAX_ADDR_ID)
988 pernet->next_id = 1;
989 if (pernet->addrs >= MPTCP_PM_ADDR_MAX) {
990 ret = -ERANGE;
991 goto out;
992 }
993 if (test_bit(entry->addr.id, pernet->id_bitmap)) {
994 ret = -EBUSY;
995 goto out;
996 }
997
998 /* do not insert duplicate address, differentiate on port only
999 * singled addresses
1000 */
1001 if (!address_use_port(entry))
1002 entry->addr.port = 0;
1003 list_for_each_entry(cur, &pernet->local_addr_list, list) {
1004 if (mptcp_addresses_equal(&cur->addr, &entry->addr,
1005 cur->addr.port || entry->addr.port)) {
1006 /* allow replacing the exiting endpoint only if such
1007 * endpoint is an implicit one and the user-space
1008 * did not provide an endpoint id
1009 */
1010 if (!(cur->flags & MPTCP_PM_ADDR_FLAG_IMPLICIT)) {
1011 ret = -EEXIST;
1012 goto out;
1013 }
1014 if (entry->addr.id)
1015 goto out;
1016
1017 /* allow callers that only need to look up the local
1018 * addr's id to skip replacement. This allows them to
1019 * avoid calling synchronize_rcu in the packet recv
1020 * path.
1021 */
1022 if (!replace) {
1023 kfree(entry);
1024 ret = cur->addr.id;
1025 goto out;
1026 }
1027
1028 pernet->addrs--;
1029 entry->addr.id = cur->addr.id;
1030 list_del_rcu(&cur->list);
1031 del_entry = cur;
1032 break;
1033 }
1034 }
1035
1036 if (!entry->addr.id && needs_id) {
1037 find_next:
1038 entry->addr.id = find_next_zero_bit(pernet->id_bitmap,
1039 MPTCP_PM_MAX_ADDR_ID + 1,
1040 pernet->next_id);
1041 if (!entry->addr.id && pernet->next_id != 1) {
1042 pernet->next_id = 1;
1043 goto find_next;
1044 }
1045 }
1046
1047 if (!entry->addr.id && needs_id)
1048 goto out;
1049
1050 __set_bit(entry->addr.id, pernet->id_bitmap);
1051 if (entry->addr.id > pernet->next_id)
1052 pernet->next_id = entry->addr.id;
1053
1054 if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) {
1055 addr_max = pernet->add_addr_signal_max;
1056 WRITE_ONCE(pernet->add_addr_signal_max, addr_max + 1);
1057 }
1058 if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) {
1059 addr_max = pernet->local_addr_max;
1060 WRITE_ONCE(pernet->local_addr_max, addr_max + 1);
1061 }
1062
1063 pernet->addrs++;
1064 if (!entry->addr.port)
1065 list_add_tail_rcu(&entry->list, &pernet->local_addr_list);
1066 else
1067 list_add_rcu(&entry->list, &pernet->local_addr_list);
1068 ret = entry->addr.id;
1069
1070 out:
1071 spin_unlock_bh(&pernet->lock);
1072
1073 /* just replaced an existing entry, free it */
1074 if (del_entry) {
1075 synchronize_rcu();
1076 __mptcp_pm_release_addr_entry(del_entry);
1077 }
1078 return ret;
1079 }
1080
1081 static struct lock_class_key mptcp_slock_keys[2];
1082 static struct lock_class_key mptcp_keys[2];
1083
mptcp_pm_nl_create_listen_socket(struct sock * sk,struct mptcp_pm_addr_entry * entry)1084 static int mptcp_pm_nl_create_listen_socket(struct sock *sk,
1085 struct mptcp_pm_addr_entry *entry)
1086 {
1087 bool is_ipv6 = sk->sk_family == AF_INET6;
1088 int addrlen = sizeof(struct sockaddr_in);
1089 struct sockaddr_storage addr;
1090 struct sock *newsk, *ssk;
1091 int backlog = 1024;
1092 int err;
1093
1094 err = sock_create_kern(sock_net(sk), entry->addr.family,
1095 SOCK_STREAM, IPPROTO_MPTCP, &entry->lsk);
1096 if (err)
1097 return err;
1098
1099 newsk = entry->lsk->sk;
1100 if (!newsk)
1101 return -EINVAL;
1102
1103 /* The subflow socket lock is acquired in a nested to the msk one
1104 * in several places, even by the TCP stack, and this msk is a kernel
1105 * socket: lockdep complains. Instead of propagating the _nested
1106 * modifiers in several places, re-init the lock class for the msk
1107 * socket to an mptcp specific one.
1108 */
1109 sock_lock_init_class_and_name(newsk,
1110 is_ipv6 ? "mlock-AF_INET6" : "mlock-AF_INET",
1111 &mptcp_slock_keys[is_ipv6],
1112 is_ipv6 ? "msk_lock-AF_INET6" : "msk_lock-AF_INET",
1113 &mptcp_keys[is_ipv6]);
1114
1115 lock_sock(newsk);
1116 ssk = __mptcp_nmpc_sk(mptcp_sk(newsk));
1117 release_sock(newsk);
1118 if (IS_ERR(ssk))
1119 return PTR_ERR(ssk);
1120
1121 mptcp_info2sockaddr(&entry->addr, &addr, entry->addr.family);
1122 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1123 if (entry->addr.family == AF_INET6)
1124 addrlen = sizeof(struct sockaddr_in6);
1125 #endif
1126 if (ssk->sk_family == AF_INET)
1127 err = inet_bind_sk(ssk, (struct sockaddr *)&addr, addrlen);
1128 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1129 else if (ssk->sk_family == AF_INET6)
1130 err = inet6_bind_sk(ssk, (struct sockaddr *)&addr, addrlen);
1131 #endif
1132 if (err)
1133 return err;
1134
1135 /* We don't use mptcp_set_state() here because it needs to be called
1136 * under the msk socket lock. For the moment, that will not bring
1137 * anything more than only calling inet_sk_state_store(), because the
1138 * old status is known (TCP_CLOSE).
1139 */
1140 inet_sk_state_store(newsk, TCP_LISTEN);
1141 lock_sock(ssk);
1142 WRITE_ONCE(mptcp_subflow_ctx(ssk)->pm_listener, true);
1143 err = __inet_listen_sk(ssk, backlog);
1144 if (!err)
1145 mptcp_event_pm_listener(ssk, MPTCP_EVENT_LISTENER_CREATED);
1146 release_sock(ssk);
1147 return err;
1148 }
1149
mptcp_pm_nl_get_local_id(struct mptcp_sock * msk,struct mptcp_addr_info * skc)1150 int mptcp_pm_nl_get_local_id(struct mptcp_sock *msk, struct mptcp_addr_info *skc)
1151 {
1152 struct mptcp_pm_addr_entry *entry;
1153 struct pm_nl_pernet *pernet;
1154 int ret = -1;
1155
1156 pernet = pm_nl_get_pernet_from_msk(msk);
1157
1158 rcu_read_lock();
1159 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) {
1160 if (mptcp_addresses_equal(&entry->addr, skc, entry->addr.port)) {
1161 ret = entry->addr.id;
1162 break;
1163 }
1164 }
1165 rcu_read_unlock();
1166 if (ret >= 0)
1167 return ret;
1168
1169 /* address not found, add to local list */
1170 entry = kmalloc(sizeof(*entry), GFP_ATOMIC);
1171 if (!entry)
1172 return -ENOMEM;
1173
1174 entry->addr = *skc;
1175 entry->addr.id = 0;
1176 entry->addr.port = 0;
1177 entry->ifindex = 0;
1178 entry->flags = MPTCP_PM_ADDR_FLAG_IMPLICIT;
1179 entry->lsk = NULL;
1180 ret = mptcp_pm_nl_append_new_local_addr(pernet, entry, true, false);
1181 if (ret < 0)
1182 kfree(entry);
1183
1184 return ret;
1185 }
1186
mptcp_pm_nl_is_backup(struct mptcp_sock * msk,struct mptcp_addr_info * skc)1187 bool mptcp_pm_nl_is_backup(struct mptcp_sock *msk, struct mptcp_addr_info *skc)
1188 {
1189 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk);
1190 struct mptcp_pm_addr_entry *entry;
1191 bool backup = false;
1192
1193 rcu_read_lock();
1194 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) {
1195 if (mptcp_addresses_equal(&entry->addr, skc, entry->addr.port)) {
1196 backup = !!(entry->flags & MPTCP_PM_ADDR_FLAG_BACKUP);
1197 break;
1198 }
1199 }
1200 rcu_read_unlock();
1201
1202 return backup;
1203 }
1204
1205 #define MPTCP_PM_CMD_GRP_OFFSET 0
1206 #define MPTCP_PM_EV_GRP_OFFSET 1
1207
1208 static const struct genl_multicast_group mptcp_pm_mcgrps[] = {
1209 [MPTCP_PM_CMD_GRP_OFFSET] = { .name = MPTCP_PM_CMD_GRP_NAME, },
1210 [MPTCP_PM_EV_GRP_OFFSET] = { .name = MPTCP_PM_EV_GRP_NAME,
1211 .flags = GENL_MCAST_CAP_NET_ADMIN,
1212 },
1213 };
1214
mptcp_pm_nl_subflow_chk_stale(const struct mptcp_sock * msk,struct sock * ssk)1215 void mptcp_pm_nl_subflow_chk_stale(const struct mptcp_sock *msk, struct sock *ssk)
1216 {
1217 struct mptcp_subflow_context *iter, *subflow = mptcp_subflow_ctx(ssk);
1218 struct sock *sk = (struct sock *)msk;
1219 unsigned int active_max_loss_cnt;
1220 struct net *net = sock_net(sk);
1221 unsigned int stale_loss_cnt;
1222 bool slow;
1223
1224 stale_loss_cnt = mptcp_stale_loss_cnt(net);
1225 if (subflow->stale || !stale_loss_cnt || subflow->stale_count <= stale_loss_cnt)
1226 return;
1227
1228 /* look for another available subflow not in loss state */
1229 active_max_loss_cnt = max_t(int, stale_loss_cnt - 1, 1);
1230 mptcp_for_each_subflow(msk, iter) {
1231 if (iter != subflow && mptcp_subflow_active(iter) &&
1232 iter->stale_count < active_max_loss_cnt) {
1233 /* we have some alternatives, try to mark this subflow as idle ...*/
1234 slow = lock_sock_fast(ssk);
1235 if (!tcp_rtx_and_write_queues_empty(ssk)) {
1236 subflow->stale = 1;
1237 __mptcp_retransmit_pending_data(sk);
1238 MPTCP_INC_STATS(net, MPTCP_MIB_SUBFLOWSTALE);
1239 }
1240 unlock_sock_fast(ssk, slow);
1241
1242 /* always try to push the pending data regardless of re-injections:
1243 * we can possibly use backup subflows now, and subflow selection
1244 * is cheap under the msk socket lock
1245 */
1246 __mptcp_push_pending(sk, 0);
1247 return;
1248 }
1249 }
1250 }
1251
mptcp_pm_family_to_addr(int family)1252 static int mptcp_pm_family_to_addr(int family)
1253 {
1254 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1255 if (family == AF_INET6)
1256 return MPTCP_PM_ADDR_ATTR_ADDR6;
1257 #endif
1258 return MPTCP_PM_ADDR_ATTR_ADDR4;
1259 }
1260
mptcp_pm_parse_pm_addr_attr(struct nlattr * tb[],const struct nlattr * attr,struct genl_info * info,struct mptcp_addr_info * addr,bool require_family)1261 static int mptcp_pm_parse_pm_addr_attr(struct nlattr *tb[],
1262 const struct nlattr *attr,
1263 struct genl_info *info,
1264 struct mptcp_addr_info *addr,
1265 bool require_family)
1266 {
1267 int err, addr_addr;
1268
1269 if (!attr) {
1270 GENL_SET_ERR_MSG(info, "missing address info");
1271 return -EINVAL;
1272 }
1273
1274 /* no validation needed - was already done via nested policy */
1275 err = nla_parse_nested_deprecated(tb, MPTCP_PM_ADDR_ATTR_MAX, attr,
1276 mptcp_pm_address_nl_policy, info->extack);
1277 if (err)
1278 return err;
1279
1280 if (tb[MPTCP_PM_ADDR_ATTR_ID])
1281 addr->id = nla_get_u8(tb[MPTCP_PM_ADDR_ATTR_ID]);
1282
1283 if (!tb[MPTCP_PM_ADDR_ATTR_FAMILY]) {
1284 if (!require_family)
1285 return 0;
1286
1287 NL_SET_ERR_MSG_ATTR(info->extack, attr,
1288 "missing family");
1289 return -EINVAL;
1290 }
1291
1292 addr->family = nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_FAMILY]);
1293 if (addr->family != AF_INET
1294 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1295 && addr->family != AF_INET6
1296 #endif
1297 ) {
1298 NL_SET_ERR_MSG_ATTR(info->extack, attr,
1299 "unknown address family");
1300 return -EINVAL;
1301 }
1302 addr_addr = mptcp_pm_family_to_addr(addr->family);
1303 if (!tb[addr_addr]) {
1304 NL_SET_ERR_MSG_ATTR(info->extack, attr,
1305 "missing address data");
1306 return -EINVAL;
1307 }
1308
1309 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1310 if (addr->family == AF_INET6)
1311 addr->addr6 = nla_get_in6_addr(tb[addr_addr]);
1312 else
1313 #endif
1314 addr->addr.s_addr = nla_get_in_addr(tb[addr_addr]);
1315
1316 if (tb[MPTCP_PM_ADDR_ATTR_PORT])
1317 addr->port = htons(nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_PORT]));
1318
1319 return 0;
1320 }
1321
mptcp_pm_parse_addr(struct nlattr * attr,struct genl_info * info,struct mptcp_addr_info * addr)1322 int mptcp_pm_parse_addr(struct nlattr *attr, struct genl_info *info,
1323 struct mptcp_addr_info *addr)
1324 {
1325 struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1];
1326
1327 memset(addr, 0, sizeof(*addr));
1328
1329 return mptcp_pm_parse_pm_addr_attr(tb, attr, info, addr, true);
1330 }
1331
mptcp_pm_parse_entry(struct nlattr * attr,struct genl_info * info,bool require_family,struct mptcp_pm_addr_entry * entry)1332 int mptcp_pm_parse_entry(struct nlattr *attr, struct genl_info *info,
1333 bool require_family,
1334 struct mptcp_pm_addr_entry *entry)
1335 {
1336 struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1];
1337 int err;
1338
1339 memset(entry, 0, sizeof(*entry));
1340
1341 err = mptcp_pm_parse_pm_addr_attr(tb, attr, info, &entry->addr, require_family);
1342 if (err)
1343 return err;
1344
1345 if (tb[MPTCP_PM_ADDR_ATTR_IF_IDX]) {
1346 u32 val = nla_get_s32(tb[MPTCP_PM_ADDR_ATTR_IF_IDX]);
1347
1348 entry->ifindex = val;
1349 }
1350
1351 if (tb[MPTCP_PM_ADDR_ATTR_FLAGS])
1352 entry->flags = nla_get_u32(tb[MPTCP_PM_ADDR_ATTR_FLAGS]);
1353
1354 if (tb[MPTCP_PM_ADDR_ATTR_PORT])
1355 entry->addr.port = htons(nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_PORT]));
1356
1357 return 0;
1358 }
1359
genl_info_pm_nl(struct genl_info * info)1360 static struct pm_nl_pernet *genl_info_pm_nl(struct genl_info *info)
1361 {
1362 return pm_nl_get_pernet(genl_info_net(info));
1363 }
1364
mptcp_nl_add_subflow_or_signal_addr(struct net * net,struct mptcp_addr_info * addr)1365 static int mptcp_nl_add_subflow_or_signal_addr(struct net *net,
1366 struct mptcp_addr_info *addr)
1367 {
1368 struct mptcp_sock *msk;
1369 long s_slot = 0, s_num = 0;
1370
1371 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1372 struct sock *sk = (struct sock *)msk;
1373 struct mptcp_addr_info mpc_addr;
1374
1375 if (!READ_ONCE(msk->fully_established) ||
1376 mptcp_pm_is_userspace(msk))
1377 goto next;
1378
1379 /* if the endp linked to the init sf is re-added with a != ID */
1380 mptcp_local_address((struct sock_common *)msk, &mpc_addr);
1381
1382 lock_sock(sk);
1383 spin_lock_bh(&msk->pm.lock);
1384 if (mptcp_addresses_equal(addr, &mpc_addr, addr->port))
1385 msk->mpc_endpoint_id = addr->id;
1386 mptcp_pm_create_subflow_or_signal_addr(msk);
1387 spin_unlock_bh(&msk->pm.lock);
1388 release_sock(sk);
1389
1390 next:
1391 sock_put(sk);
1392 cond_resched();
1393 }
1394
1395 return 0;
1396 }
1397
mptcp_pm_has_addr_attr_id(const struct nlattr * attr,struct genl_info * info)1398 static bool mptcp_pm_has_addr_attr_id(const struct nlattr *attr,
1399 struct genl_info *info)
1400 {
1401 struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1];
1402
1403 if (!nla_parse_nested_deprecated(tb, MPTCP_PM_ADDR_ATTR_MAX, attr,
1404 mptcp_pm_address_nl_policy, info->extack) &&
1405 tb[MPTCP_PM_ADDR_ATTR_ID])
1406 return true;
1407 return false;
1408 }
1409
mptcp_pm_nl_add_addr_doit(struct sk_buff * skb,struct genl_info * info)1410 int mptcp_pm_nl_add_addr_doit(struct sk_buff *skb, struct genl_info *info)
1411 {
1412 struct nlattr *attr = info->attrs[MPTCP_PM_ENDPOINT_ADDR];
1413 struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1414 struct mptcp_pm_addr_entry addr, *entry;
1415 int ret;
1416
1417 ret = mptcp_pm_parse_entry(attr, info, true, &addr);
1418 if (ret < 0)
1419 return ret;
1420
1421 if (addr.addr.port && !address_use_port(&addr)) {
1422 GENL_SET_ERR_MSG(info, "flags must have signal and not subflow when using port");
1423 return -EINVAL;
1424 }
1425
1426 if (addr.flags & MPTCP_PM_ADDR_FLAG_SIGNAL &&
1427 addr.flags & MPTCP_PM_ADDR_FLAG_FULLMESH) {
1428 GENL_SET_ERR_MSG(info, "flags mustn't have both signal and fullmesh");
1429 return -EINVAL;
1430 }
1431
1432 if (addr.flags & MPTCP_PM_ADDR_FLAG_IMPLICIT) {
1433 GENL_SET_ERR_MSG(info, "can't create IMPLICIT endpoint");
1434 return -EINVAL;
1435 }
1436
1437 entry = kzalloc(sizeof(*entry), GFP_KERNEL_ACCOUNT);
1438 if (!entry) {
1439 GENL_SET_ERR_MSG(info, "can't allocate addr");
1440 return -ENOMEM;
1441 }
1442
1443 *entry = addr;
1444 if (entry->addr.port) {
1445 ret = mptcp_pm_nl_create_listen_socket(skb->sk, entry);
1446 if (ret) {
1447 GENL_SET_ERR_MSG_FMT(info, "create listen socket error: %d", ret);
1448 goto out_free;
1449 }
1450 }
1451 ret = mptcp_pm_nl_append_new_local_addr(pernet, entry,
1452 !mptcp_pm_has_addr_attr_id(attr, info),
1453 true);
1454 if (ret < 0) {
1455 GENL_SET_ERR_MSG_FMT(info, "too many addresses or duplicate one: %d", ret);
1456 goto out_free;
1457 }
1458
1459 mptcp_nl_add_subflow_or_signal_addr(sock_net(skb->sk), &entry->addr);
1460 return 0;
1461
1462 out_free:
1463 __mptcp_pm_release_addr_entry(entry);
1464 return ret;
1465 }
1466
remove_anno_list_by_saddr(struct mptcp_sock * msk,const struct mptcp_addr_info * addr)1467 static bool remove_anno_list_by_saddr(struct mptcp_sock *msk,
1468 const struct mptcp_addr_info *addr)
1469 {
1470 struct mptcp_pm_add_entry *entry;
1471
1472 entry = mptcp_pm_del_add_timer(msk, addr, false);
1473 if (entry) {
1474 kfree(entry);
1475 return true;
1476 }
1477
1478 return false;
1479 }
1480
mptcp_endp_get_local_id(struct mptcp_sock * msk,const struct mptcp_addr_info * addr)1481 static u8 mptcp_endp_get_local_id(struct mptcp_sock *msk,
1482 const struct mptcp_addr_info *addr)
1483 {
1484 return msk->mpc_endpoint_id == addr->id ? 0 : addr->id;
1485 }
1486
mptcp_pm_remove_anno_addr(struct mptcp_sock * msk,const struct mptcp_addr_info * addr,bool force)1487 static bool mptcp_pm_remove_anno_addr(struct mptcp_sock *msk,
1488 const struct mptcp_addr_info *addr,
1489 bool force)
1490 {
1491 struct mptcp_rm_list list = { .nr = 0 };
1492 bool ret;
1493
1494 list.ids[list.nr++] = mptcp_endp_get_local_id(msk, addr);
1495
1496 ret = remove_anno_list_by_saddr(msk, addr);
1497 if (ret || force) {
1498 spin_lock_bh(&msk->pm.lock);
1499 if (ret) {
1500 __set_bit(addr->id, msk->pm.id_avail_bitmap);
1501 msk->pm.add_addr_signaled--;
1502 }
1503 mptcp_pm_remove_addr(msk, &list);
1504 spin_unlock_bh(&msk->pm.lock);
1505 }
1506 return ret;
1507 }
1508
__mark_subflow_endp_available(struct mptcp_sock * msk,u8 id)1509 static void __mark_subflow_endp_available(struct mptcp_sock *msk, u8 id)
1510 {
1511 /* If it was marked as used, and not ID 0, decrement local_addr_used */
1512 if (!__test_and_set_bit(id ? : msk->mpc_endpoint_id, msk->pm.id_avail_bitmap) &&
1513 id && !WARN_ON_ONCE(msk->pm.local_addr_used == 0))
1514 msk->pm.local_addr_used--;
1515 }
1516
mptcp_nl_remove_subflow_and_signal_addr(struct net * net,const struct mptcp_pm_addr_entry * entry)1517 static int mptcp_nl_remove_subflow_and_signal_addr(struct net *net,
1518 const struct mptcp_pm_addr_entry *entry)
1519 {
1520 const struct mptcp_addr_info *addr = &entry->addr;
1521 struct mptcp_rm_list list = { .nr = 1 };
1522 long s_slot = 0, s_num = 0;
1523 struct mptcp_sock *msk;
1524
1525 pr_debug("remove_id=%d\n", addr->id);
1526
1527 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1528 struct sock *sk = (struct sock *)msk;
1529 bool remove_subflow;
1530
1531 if (mptcp_pm_is_userspace(msk))
1532 goto next;
1533
1534 lock_sock(sk);
1535 remove_subflow = lookup_subflow_by_saddr(&msk->conn_list, addr);
1536 mptcp_pm_remove_anno_addr(msk, addr, remove_subflow &&
1537 !(entry->flags & MPTCP_PM_ADDR_FLAG_IMPLICIT));
1538
1539 list.ids[0] = mptcp_endp_get_local_id(msk, addr);
1540 if (remove_subflow) {
1541 spin_lock_bh(&msk->pm.lock);
1542 mptcp_pm_nl_rm_subflow_received(msk, &list);
1543 spin_unlock_bh(&msk->pm.lock);
1544 }
1545
1546 if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) {
1547 spin_lock_bh(&msk->pm.lock);
1548 __mark_subflow_endp_available(msk, list.ids[0]);
1549 spin_unlock_bh(&msk->pm.lock);
1550 }
1551
1552 if (msk->mpc_endpoint_id == entry->addr.id)
1553 msk->mpc_endpoint_id = 0;
1554 release_sock(sk);
1555
1556 next:
1557 sock_put(sk);
1558 cond_resched();
1559 }
1560
1561 return 0;
1562 }
1563
mptcp_nl_remove_id_zero_address(struct net * net,struct mptcp_addr_info * addr)1564 static int mptcp_nl_remove_id_zero_address(struct net *net,
1565 struct mptcp_addr_info *addr)
1566 {
1567 struct mptcp_rm_list list = { .nr = 0 };
1568 long s_slot = 0, s_num = 0;
1569 struct mptcp_sock *msk;
1570
1571 list.ids[list.nr++] = 0;
1572
1573 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1574 struct sock *sk = (struct sock *)msk;
1575 struct mptcp_addr_info msk_local;
1576
1577 if (list_empty(&msk->conn_list) || mptcp_pm_is_userspace(msk))
1578 goto next;
1579
1580 mptcp_local_address((struct sock_common *)msk, &msk_local);
1581 if (!mptcp_addresses_equal(&msk_local, addr, addr->port))
1582 goto next;
1583
1584 lock_sock(sk);
1585 spin_lock_bh(&msk->pm.lock);
1586 mptcp_pm_remove_addr(msk, &list);
1587 mptcp_pm_nl_rm_subflow_received(msk, &list);
1588 __mark_subflow_endp_available(msk, 0);
1589 spin_unlock_bh(&msk->pm.lock);
1590 release_sock(sk);
1591
1592 next:
1593 sock_put(sk);
1594 cond_resched();
1595 }
1596
1597 return 0;
1598 }
1599
mptcp_pm_nl_del_addr_doit(struct sk_buff * skb,struct genl_info * info)1600 int mptcp_pm_nl_del_addr_doit(struct sk_buff *skb, struct genl_info *info)
1601 {
1602 struct nlattr *attr = info->attrs[MPTCP_PM_ENDPOINT_ADDR];
1603 struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1604 struct mptcp_pm_addr_entry addr, *entry;
1605 unsigned int addr_max;
1606 int ret;
1607
1608 ret = mptcp_pm_parse_entry(attr, info, false, &addr);
1609 if (ret < 0)
1610 return ret;
1611
1612 /* the zero id address is special: the first address used by the msk
1613 * always gets such an id, so different subflows can have different zero
1614 * id addresses. Additionally zero id is not accounted for in id_bitmap.
1615 * Let's use an 'mptcp_rm_list' instead of the common remove code.
1616 */
1617 if (addr.addr.id == 0)
1618 return mptcp_nl_remove_id_zero_address(sock_net(skb->sk), &addr.addr);
1619
1620 spin_lock_bh(&pernet->lock);
1621 entry = __lookup_addr_by_id(pernet, addr.addr.id);
1622 if (!entry) {
1623 GENL_SET_ERR_MSG(info, "address not found");
1624 spin_unlock_bh(&pernet->lock);
1625 return -EINVAL;
1626 }
1627 if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) {
1628 addr_max = pernet->add_addr_signal_max;
1629 WRITE_ONCE(pernet->add_addr_signal_max, addr_max - 1);
1630 }
1631 if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) {
1632 addr_max = pernet->local_addr_max;
1633 WRITE_ONCE(pernet->local_addr_max, addr_max - 1);
1634 }
1635
1636 pernet->addrs--;
1637 list_del_rcu(&entry->list);
1638 __clear_bit(entry->addr.id, pernet->id_bitmap);
1639 spin_unlock_bh(&pernet->lock);
1640
1641 mptcp_nl_remove_subflow_and_signal_addr(sock_net(skb->sk), entry);
1642 synchronize_rcu();
1643 __mptcp_pm_release_addr_entry(entry);
1644
1645 return ret;
1646 }
1647
1648 /* Called from the userspace PM only */
mptcp_pm_remove_addrs(struct mptcp_sock * msk,struct list_head * rm_list)1649 void mptcp_pm_remove_addrs(struct mptcp_sock *msk, struct list_head *rm_list)
1650 {
1651 struct mptcp_rm_list alist = { .nr = 0 };
1652 struct mptcp_pm_addr_entry *entry;
1653 int anno_nr = 0;
1654
1655 list_for_each_entry(entry, rm_list, list) {
1656 if (alist.nr >= MPTCP_RM_IDS_MAX)
1657 break;
1658
1659 /* only delete if either announced or matching a subflow */
1660 if (remove_anno_list_by_saddr(msk, &entry->addr))
1661 anno_nr++;
1662 else if (!lookup_subflow_by_saddr(&msk->conn_list,
1663 &entry->addr))
1664 continue;
1665
1666 alist.ids[alist.nr++] = entry->addr.id;
1667 }
1668
1669 if (alist.nr) {
1670 spin_lock_bh(&msk->pm.lock);
1671 msk->pm.add_addr_signaled -= anno_nr;
1672 mptcp_pm_remove_addr(msk, &alist);
1673 spin_unlock_bh(&msk->pm.lock);
1674 }
1675 }
1676
1677 /* Called from the in-kernel PM only */
mptcp_pm_flush_addrs_and_subflows(struct mptcp_sock * msk,struct list_head * rm_list)1678 static void mptcp_pm_flush_addrs_and_subflows(struct mptcp_sock *msk,
1679 struct list_head *rm_list)
1680 {
1681 struct mptcp_rm_list alist = { .nr = 0 }, slist = { .nr = 0 };
1682 struct mptcp_pm_addr_entry *entry;
1683
1684 list_for_each_entry(entry, rm_list, list) {
1685 if (slist.nr < MPTCP_RM_IDS_MAX &&
1686 lookup_subflow_by_saddr(&msk->conn_list, &entry->addr))
1687 slist.ids[slist.nr++] = mptcp_endp_get_local_id(msk, &entry->addr);
1688
1689 if (alist.nr < MPTCP_RM_IDS_MAX &&
1690 remove_anno_list_by_saddr(msk, &entry->addr))
1691 alist.ids[alist.nr++] = mptcp_endp_get_local_id(msk, &entry->addr);
1692 }
1693
1694 spin_lock_bh(&msk->pm.lock);
1695 if (alist.nr) {
1696 msk->pm.add_addr_signaled -= alist.nr;
1697 mptcp_pm_remove_addr(msk, &alist);
1698 }
1699 if (slist.nr)
1700 mptcp_pm_nl_rm_subflow_received(msk, &slist);
1701 /* Reset counters: maybe some subflows have been removed before */
1702 bitmap_fill(msk->pm.id_avail_bitmap, MPTCP_PM_MAX_ADDR_ID + 1);
1703 msk->pm.local_addr_used = 0;
1704 spin_unlock_bh(&msk->pm.lock);
1705 }
1706
mptcp_nl_flush_addrs_list(struct net * net,struct list_head * rm_list)1707 static void mptcp_nl_flush_addrs_list(struct net *net,
1708 struct list_head *rm_list)
1709 {
1710 long s_slot = 0, s_num = 0;
1711 struct mptcp_sock *msk;
1712
1713 if (list_empty(rm_list))
1714 return;
1715
1716 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1717 struct sock *sk = (struct sock *)msk;
1718
1719 if (!mptcp_pm_is_userspace(msk)) {
1720 lock_sock(sk);
1721 mptcp_pm_flush_addrs_and_subflows(msk, rm_list);
1722 release_sock(sk);
1723 }
1724
1725 sock_put(sk);
1726 cond_resched();
1727 }
1728 }
1729
1730 /* caller must ensure the RCU grace period is already elapsed */
__flush_addrs(struct list_head * list)1731 static void __flush_addrs(struct list_head *list)
1732 {
1733 while (!list_empty(list)) {
1734 struct mptcp_pm_addr_entry *cur;
1735
1736 cur = list_entry(list->next,
1737 struct mptcp_pm_addr_entry, list);
1738 list_del_rcu(&cur->list);
1739 __mptcp_pm_release_addr_entry(cur);
1740 }
1741 }
1742
__reset_counters(struct pm_nl_pernet * pernet)1743 static void __reset_counters(struct pm_nl_pernet *pernet)
1744 {
1745 WRITE_ONCE(pernet->add_addr_signal_max, 0);
1746 WRITE_ONCE(pernet->local_addr_max, 0);
1747 pernet->addrs = 0;
1748 }
1749
mptcp_pm_nl_flush_addrs_doit(struct sk_buff * skb,struct genl_info * info)1750 int mptcp_pm_nl_flush_addrs_doit(struct sk_buff *skb, struct genl_info *info)
1751 {
1752 struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1753 LIST_HEAD(free_list);
1754
1755 spin_lock_bh(&pernet->lock);
1756 list_splice_init(&pernet->local_addr_list, &free_list);
1757 __reset_counters(pernet);
1758 pernet->next_id = 1;
1759 bitmap_zero(pernet->id_bitmap, MPTCP_PM_MAX_ADDR_ID + 1);
1760 spin_unlock_bh(&pernet->lock);
1761 mptcp_nl_flush_addrs_list(sock_net(skb->sk), &free_list);
1762 synchronize_rcu();
1763 __flush_addrs(&free_list);
1764 return 0;
1765 }
1766
mptcp_nl_fill_addr(struct sk_buff * skb,struct mptcp_pm_addr_entry * entry)1767 int mptcp_nl_fill_addr(struct sk_buff *skb,
1768 struct mptcp_pm_addr_entry *entry)
1769 {
1770 struct mptcp_addr_info *addr = &entry->addr;
1771 struct nlattr *attr;
1772
1773 attr = nla_nest_start(skb, MPTCP_PM_ATTR_ADDR);
1774 if (!attr)
1775 return -EMSGSIZE;
1776
1777 if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_FAMILY, addr->family))
1778 goto nla_put_failure;
1779 if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_PORT, ntohs(addr->port)))
1780 goto nla_put_failure;
1781 if (nla_put_u8(skb, MPTCP_PM_ADDR_ATTR_ID, addr->id))
1782 goto nla_put_failure;
1783 if (nla_put_u32(skb, MPTCP_PM_ADDR_ATTR_FLAGS, entry->flags))
1784 goto nla_put_failure;
1785 if (entry->ifindex &&
1786 nla_put_s32(skb, MPTCP_PM_ADDR_ATTR_IF_IDX, entry->ifindex))
1787 goto nla_put_failure;
1788
1789 if (addr->family == AF_INET &&
1790 nla_put_in_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR4,
1791 addr->addr.s_addr))
1792 goto nla_put_failure;
1793 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1794 else if (addr->family == AF_INET6 &&
1795 nla_put_in6_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR6, &addr->addr6))
1796 goto nla_put_failure;
1797 #endif
1798 nla_nest_end(skb, attr);
1799 return 0;
1800
1801 nla_put_failure:
1802 nla_nest_cancel(skb, attr);
1803 return -EMSGSIZE;
1804 }
1805
mptcp_pm_nl_get_addr(struct sk_buff * skb,struct genl_info * info)1806 int mptcp_pm_nl_get_addr(struct sk_buff *skb, struct genl_info *info)
1807 {
1808 struct nlattr *attr = info->attrs[MPTCP_PM_ENDPOINT_ADDR];
1809 struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1810 struct mptcp_pm_addr_entry addr, *entry;
1811 struct sk_buff *msg;
1812 void *reply;
1813 int ret;
1814
1815 ret = mptcp_pm_parse_entry(attr, info, false, &addr);
1816 if (ret < 0)
1817 return ret;
1818
1819 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1820 if (!msg)
1821 return -ENOMEM;
1822
1823 reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0,
1824 info->genlhdr->cmd);
1825 if (!reply) {
1826 GENL_SET_ERR_MSG(info, "not enough space in Netlink message");
1827 ret = -EMSGSIZE;
1828 goto fail;
1829 }
1830
1831 spin_lock_bh(&pernet->lock);
1832 entry = __lookup_addr_by_id(pernet, addr.addr.id);
1833 if (!entry) {
1834 GENL_SET_ERR_MSG(info, "address not found");
1835 ret = -EINVAL;
1836 goto unlock_fail;
1837 }
1838
1839 ret = mptcp_nl_fill_addr(msg, entry);
1840 if (ret)
1841 goto unlock_fail;
1842
1843 genlmsg_end(msg, reply);
1844 ret = genlmsg_reply(msg, info);
1845 spin_unlock_bh(&pernet->lock);
1846 return ret;
1847
1848 unlock_fail:
1849 spin_unlock_bh(&pernet->lock);
1850
1851 fail:
1852 nlmsg_free(msg);
1853 return ret;
1854 }
1855
mptcp_pm_nl_get_addr_doit(struct sk_buff * skb,struct genl_info * info)1856 int mptcp_pm_nl_get_addr_doit(struct sk_buff *skb, struct genl_info *info)
1857 {
1858 return mptcp_pm_get_addr(skb, info);
1859 }
1860
mptcp_pm_nl_dump_addr(struct sk_buff * msg,struct netlink_callback * cb)1861 int mptcp_pm_nl_dump_addr(struct sk_buff *msg,
1862 struct netlink_callback *cb)
1863 {
1864 struct net *net = sock_net(msg->sk);
1865 struct mptcp_pm_addr_entry *entry;
1866 struct pm_nl_pernet *pernet;
1867 int id = cb->args[0];
1868 void *hdr;
1869 int i;
1870
1871 pernet = pm_nl_get_pernet(net);
1872
1873 spin_lock_bh(&pernet->lock);
1874 for (i = id; i < MPTCP_PM_MAX_ADDR_ID + 1; i++) {
1875 if (test_bit(i, pernet->id_bitmap)) {
1876 entry = __lookup_addr_by_id(pernet, i);
1877 if (!entry)
1878 break;
1879
1880 if (entry->addr.id <= id)
1881 continue;
1882
1883 hdr = genlmsg_put(msg, NETLINK_CB(cb->skb).portid,
1884 cb->nlh->nlmsg_seq, &mptcp_genl_family,
1885 NLM_F_MULTI, MPTCP_PM_CMD_GET_ADDR);
1886 if (!hdr)
1887 break;
1888
1889 if (mptcp_nl_fill_addr(msg, entry) < 0) {
1890 genlmsg_cancel(msg, hdr);
1891 break;
1892 }
1893
1894 id = entry->addr.id;
1895 genlmsg_end(msg, hdr);
1896 }
1897 }
1898 spin_unlock_bh(&pernet->lock);
1899
1900 cb->args[0] = id;
1901 return msg->len;
1902 }
1903
mptcp_pm_nl_get_addr_dumpit(struct sk_buff * msg,struct netlink_callback * cb)1904 int mptcp_pm_nl_get_addr_dumpit(struct sk_buff *msg,
1905 struct netlink_callback *cb)
1906 {
1907 return mptcp_pm_dump_addr(msg, cb);
1908 }
1909
parse_limit(struct genl_info * info,int id,unsigned int * limit)1910 static int parse_limit(struct genl_info *info, int id, unsigned int *limit)
1911 {
1912 struct nlattr *attr = info->attrs[id];
1913
1914 if (!attr)
1915 return 0;
1916
1917 *limit = nla_get_u32(attr);
1918 if (*limit > MPTCP_PM_ADDR_MAX) {
1919 GENL_SET_ERR_MSG(info, "limit greater than maximum");
1920 return -EINVAL;
1921 }
1922 return 0;
1923 }
1924
mptcp_pm_nl_set_limits_doit(struct sk_buff * skb,struct genl_info * info)1925 int mptcp_pm_nl_set_limits_doit(struct sk_buff *skb, struct genl_info *info)
1926 {
1927 struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1928 unsigned int rcv_addrs, subflows;
1929 int ret;
1930
1931 spin_lock_bh(&pernet->lock);
1932 rcv_addrs = pernet->add_addr_accept_max;
1933 ret = parse_limit(info, MPTCP_PM_ATTR_RCV_ADD_ADDRS, &rcv_addrs);
1934 if (ret)
1935 goto unlock;
1936
1937 subflows = pernet->subflows_max;
1938 ret = parse_limit(info, MPTCP_PM_ATTR_SUBFLOWS, &subflows);
1939 if (ret)
1940 goto unlock;
1941
1942 WRITE_ONCE(pernet->add_addr_accept_max, rcv_addrs);
1943 WRITE_ONCE(pernet->subflows_max, subflows);
1944
1945 unlock:
1946 spin_unlock_bh(&pernet->lock);
1947 return ret;
1948 }
1949
mptcp_pm_nl_get_limits_doit(struct sk_buff * skb,struct genl_info * info)1950 int mptcp_pm_nl_get_limits_doit(struct sk_buff *skb, struct genl_info *info)
1951 {
1952 struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1953 struct sk_buff *msg;
1954 void *reply;
1955
1956 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1957 if (!msg)
1958 return -ENOMEM;
1959
1960 reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0,
1961 MPTCP_PM_CMD_GET_LIMITS);
1962 if (!reply)
1963 goto fail;
1964
1965 if (nla_put_u32(msg, MPTCP_PM_ATTR_RCV_ADD_ADDRS,
1966 READ_ONCE(pernet->add_addr_accept_max)))
1967 goto fail;
1968
1969 if (nla_put_u32(msg, MPTCP_PM_ATTR_SUBFLOWS,
1970 READ_ONCE(pernet->subflows_max)))
1971 goto fail;
1972
1973 genlmsg_end(msg, reply);
1974 return genlmsg_reply(msg, info);
1975
1976 fail:
1977 GENL_SET_ERR_MSG(info, "not enough space in Netlink message");
1978 nlmsg_free(msg);
1979 return -EMSGSIZE;
1980 }
1981
mptcp_pm_nl_fullmesh(struct mptcp_sock * msk,struct mptcp_addr_info * addr)1982 static void mptcp_pm_nl_fullmesh(struct mptcp_sock *msk,
1983 struct mptcp_addr_info *addr)
1984 {
1985 struct mptcp_rm_list list = { .nr = 0 };
1986
1987 list.ids[list.nr++] = mptcp_endp_get_local_id(msk, addr);
1988
1989 spin_lock_bh(&msk->pm.lock);
1990 mptcp_pm_nl_rm_subflow_received(msk, &list);
1991 __mark_subflow_endp_available(msk, list.ids[0]);
1992 mptcp_pm_create_subflow_or_signal_addr(msk);
1993 spin_unlock_bh(&msk->pm.lock);
1994 }
1995
mptcp_nl_set_flags(struct net * net,struct mptcp_addr_info * addr,u8 bkup,u8 changed)1996 static int mptcp_nl_set_flags(struct net *net,
1997 struct mptcp_addr_info *addr,
1998 u8 bkup, u8 changed)
1999 {
2000 long s_slot = 0, s_num = 0;
2001 struct mptcp_sock *msk;
2002 int ret = -EINVAL;
2003
2004 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
2005 struct sock *sk = (struct sock *)msk;
2006
2007 if (list_empty(&msk->conn_list) || mptcp_pm_is_userspace(msk))
2008 goto next;
2009
2010 lock_sock(sk);
2011 if (changed & MPTCP_PM_ADDR_FLAG_BACKUP)
2012 ret = mptcp_pm_nl_mp_prio_send_ack(msk, addr, NULL, bkup);
2013 if (changed & MPTCP_PM_ADDR_FLAG_FULLMESH)
2014 mptcp_pm_nl_fullmesh(msk, addr);
2015 release_sock(sk);
2016
2017 next:
2018 sock_put(sk);
2019 cond_resched();
2020 }
2021
2022 return ret;
2023 }
2024
mptcp_pm_nl_set_flags(struct sk_buff * skb,struct genl_info * info)2025 int mptcp_pm_nl_set_flags(struct sk_buff *skb, struct genl_info *info)
2026 {
2027 struct mptcp_pm_addr_entry addr = { .addr = { .family = AF_UNSPEC }, };
2028 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
2029 u8 changed, mask = MPTCP_PM_ADDR_FLAG_BACKUP |
2030 MPTCP_PM_ADDR_FLAG_FULLMESH;
2031 struct net *net = sock_net(skb->sk);
2032 struct mptcp_pm_addr_entry *entry;
2033 struct pm_nl_pernet *pernet;
2034 u8 lookup_by_id = 0;
2035 u8 bkup = 0;
2036 int ret;
2037
2038 pernet = pm_nl_get_pernet(net);
2039
2040 ret = mptcp_pm_parse_entry(attr, info, false, &addr);
2041 if (ret < 0)
2042 return ret;
2043
2044 if (addr.addr.family == AF_UNSPEC) {
2045 lookup_by_id = 1;
2046 if (!addr.addr.id) {
2047 GENL_SET_ERR_MSG(info, "missing required inputs");
2048 return -EOPNOTSUPP;
2049 }
2050 }
2051
2052 if (addr.flags & MPTCP_PM_ADDR_FLAG_BACKUP)
2053 bkup = 1;
2054
2055 spin_lock_bh(&pernet->lock);
2056 entry = lookup_by_id ? __lookup_addr_by_id(pernet, addr.addr.id) :
2057 __lookup_addr(pernet, &addr.addr);
2058 if (!entry) {
2059 spin_unlock_bh(&pernet->lock);
2060 GENL_SET_ERR_MSG(info, "address not found");
2061 return -EINVAL;
2062 }
2063 if ((addr.flags & MPTCP_PM_ADDR_FLAG_FULLMESH) &&
2064 (entry->flags & (MPTCP_PM_ADDR_FLAG_SIGNAL |
2065 MPTCP_PM_ADDR_FLAG_IMPLICIT))) {
2066 spin_unlock_bh(&pernet->lock);
2067 GENL_SET_ERR_MSG(info, "invalid addr flags");
2068 return -EINVAL;
2069 }
2070
2071 changed = (addr.flags ^ entry->flags) & mask;
2072 entry->flags = (entry->flags & ~mask) | (addr.flags & mask);
2073 addr = *entry;
2074 spin_unlock_bh(&pernet->lock);
2075
2076 mptcp_nl_set_flags(net, &addr.addr, bkup, changed);
2077 return 0;
2078 }
2079
mptcp_pm_nl_set_flags_doit(struct sk_buff * skb,struct genl_info * info)2080 int mptcp_pm_nl_set_flags_doit(struct sk_buff *skb, struct genl_info *info)
2081 {
2082 return mptcp_pm_set_flags(skb, info);
2083 }
2084
mptcp_nl_mcast_send(struct net * net,struct sk_buff * nlskb,gfp_t gfp)2085 static void mptcp_nl_mcast_send(struct net *net, struct sk_buff *nlskb, gfp_t gfp)
2086 {
2087 genlmsg_multicast_netns(&mptcp_genl_family, net,
2088 nlskb, 0, MPTCP_PM_EV_GRP_OFFSET, gfp);
2089 }
2090
mptcp_userspace_pm_active(const struct mptcp_sock * msk)2091 bool mptcp_userspace_pm_active(const struct mptcp_sock *msk)
2092 {
2093 return genl_has_listeners(&mptcp_genl_family,
2094 sock_net((const struct sock *)msk),
2095 MPTCP_PM_EV_GRP_OFFSET);
2096 }
2097
mptcp_event_add_subflow(struct sk_buff * skb,const struct sock * ssk)2098 static int mptcp_event_add_subflow(struct sk_buff *skb, const struct sock *ssk)
2099 {
2100 const struct inet_sock *issk = inet_sk(ssk);
2101 const struct mptcp_subflow_context *sf;
2102
2103 if (nla_put_u16(skb, MPTCP_ATTR_FAMILY, ssk->sk_family))
2104 return -EMSGSIZE;
2105
2106 switch (ssk->sk_family) {
2107 case AF_INET:
2108 if (nla_put_in_addr(skb, MPTCP_ATTR_SADDR4, issk->inet_saddr))
2109 return -EMSGSIZE;
2110 if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, issk->inet_daddr))
2111 return -EMSGSIZE;
2112 break;
2113 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
2114 case AF_INET6: {
2115 const struct ipv6_pinfo *np = inet6_sk(ssk);
2116
2117 if (nla_put_in6_addr(skb, MPTCP_ATTR_SADDR6, &np->saddr))
2118 return -EMSGSIZE;
2119 if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &ssk->sk_v6_daddr))
2120 return -EMSGSIZE;
2121 break;
2122 }
2123 #endif
2124 default:
2125 WARN_ON_ONCE(1);
2126 return -EMSGSIZE;
2127 }
2128
2129 if (nla_put_be16(skb, MPTCP_ATTR_SPORT, issk->inet_sport))
2130 return -EMSGSIZE;
2131 if (nla_put_be16(skb, MPTCP_ATTR_DPORT, issk->inet_dport))
2132 return -EMSGSIZE;
2133
2134 sf = mptcp_subflow_ctx(ssk);
2135 if (WARN_ON_ONCE(!sf))
2136 return -EINVAL;
2137
2138 if (nla_put_u8(skb, MPTCP_ATTR_LOC_ID, subflow_get_local_id(sf)))
2139 return -EMSGSIZE;
2140
2141 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, sf->remote_id))
2142 return -EMSGSIZE;
2143
2144 return 0;
2145 }
2146
mptcp_event_put_token_and_ssk(struct sk_buff * skb,const struct mptcp_sock * msk,const struct sock * ssk)2147 static int mptcp_event_put_token_and_ssk(struct sk_buff *skb,
2148 const struct mptcp_sock *msk,
2149 const struct sock *ssk)
2150 {
2151 const struct sock *sk = (const struct sock *)msk;
2152 const struct mptcp_subflow_context *sf;
2153 u8 sk_err;
2154
2155 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, READ_ONCE(msk->token)))
2156 return -EMSGSIZE;
2157
2158 if (mptcp_event_add_subflow(skb, ssk))
2159 return -EMSGSIZE;
2160
2161 sf = mptcp_subflow_ctx(ssk);
2162 if (WARN_ON_ONCE(!sf))
2163 return -EINVAL;
2164
2165 if (nla_put_u8(skb, MPTCP_ATTR_BACKUP, sf->backup))
2166 return -EMSGSIZE;
2167
2168 if (ssk->sk_bound_dev_if &&
2169 nla_put_s32(skb, MPTCP_ATTR_IF_IDX, ssk->sk_bound_dev_if))
2170 return -EMSGSIZE;
2171
2172 sk_err = READ_ONCE(ssk->sk_err);
2173 if (sk_err && sk->sk_state == TCP_ESTABLISHED &&
2174 nla_put_u8(skb, MPTCP_ATTR_ERROR, sk_err))
2175 return -EMSGSIZE;
2176
2177 return 0;
2178 }
2179
mptcp_event_sub_established(struct sk_buff * skb,const struct mptcp_sock * msk,const struct sock * ssk)2180 static int mptcp_event_sub_established(struct sk_buff *skb,
2181 const struct mptcp_sock *msk,
2182 const struct sock *ssk)
2183 {
2184 return mptcp_event_put_token_and_ssk(skb, msk, ssk);
2185 }
2186
mptcp_event_sub_closed(struct sk_buff * skb,const struct mptcp_sock * msk,const struct sock * ssk)2187 static int mptcp_event_sub_closed(struct sk_buff *skb,
2188 const struct mptcp_sock *msk,
2189 const struct sock *ssk)
2190 {
2191 const struct mptcp_subflow_context *sf;
2192
2193 if (mptcp_event_put_token_and_ssk(skb, msk, ssk))
2194 return -EMSGSIZE;
2195
2196 sf = mptcp_subflow_ctx(ssk);
2197 if (!sf->reset_seen)
2198 return 0;
2199
2200 if (nla_put_u32(skb, MPTCP_ATTR_RESET_REASON, sf->reset_reason))
2201 return -EMSGSIZE;
2202
2203 if (nla_put_u32(skb, MPTCP_ATTR_RESET_FLAGS, sf->reset_transient))
2204 return -EMSGSIZE;
2205
2206 return 0;
2207 }
2208
mptcp_event_created(struct sk_buff * skb,const struct mptcp_sock * msk,const struct sock * ssk)2209 static int mptcp_event_created(struct sk_buff *skb,
2210 const struct mptcp_sock *msk,
2211 const struct sock *ssk)
2212 {
2213 int err = nla_put_u32(skb, MPTCP_ATTR_TOKEN, READ_ONCE(msk->token));
2214 u16 flags = 0;
2215
2216 if (err)
2217 return err;
2218
2219 if (nla_put_u8(skb, MPTCP_ATTR_SERVER_SIDE, READ_ONCE(msk->pm.server_side)))
2220 return -EMSGSIZE;
2221
2222 if (READ_ONCE(msk->pm.remote_deny_join_id0))
2223 flags |= MPTCP_PM_EV_FLAG_DENY_JOIN_ID0;
2224
2225 if (flags && nla_put_u16(skb, MPTCP_ATTR_FLAGS, flags))
2226 return -EMSGSIZE;
2227
2228 return mptcp_event_add_subflow(skb, ssk);
2229 }
2230
mptcp_event_addr_removed(const struct mptcp_sock * msk,uint8_t id)2231 void mptcp_event_addr_removed(const struct mptcp_sock *msk, uint8_t id)
2232 {
2233 struct net *net = sock_net((const struct sock *)msk);
2234 struct nlmsghdr *nlh;
2235 struct sk_buff *skb;
2236
2237 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET))
2238 return;
2239
2240 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
2241 if (!skb)
2242 return;
2243
2244 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, MPTCP_EVENT_REMOVED);
2245 if (!nlh)
2246 goto nla_put_failure;
2247
2248 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, READ_ONCE(msk->token)))
2249 goto nla_put_failure;
2250
2251 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, id))
2252 goto nla_put_failure;
2253
2254 genlmsg_end(skb, nlh);
2255 mptcp_nl_mcast_send(net, skb, GFP_ATOMIC);
2256 return;
2257
2258 nla_put_failure:
2259 nlmsg_free(skb);
2260 }
2261
mptcp_event_addr_announced(const struct sock * ssk,const struct mptcp_addr_info * info)2262 void mptcp_event_addr_announced(const struct sock *ssk,
2263 const struct mptcp_addr_info *info)
2264 {
2265 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
2266 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
2267 struct net *net = sock_net(ssk);
2268 struct nlmsghdr *nlh;
2269 struct sk_buff *skb;
2270
2271 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET))
2272 return;
2273
2274 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
2275 if (!skb)
2276 return;
2277
2278 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0,
2279 MPTCP_EVENT_ANNOUNCED);
2280 if (!nlh)
2281 goto nla_put_failure;
2282
2283 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, READ_ONCE(msk->token)))
2284 goto nla_put_failure;
2285
2286 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, info->id))
2287 goto nla_put_failure;
2288
2289 if (nla_put_be16(skb, MPTCP_ATTR_DPORT,
2290 info->port == 0 ?
2291 inet_sk(ssk)->inet_dport :
2292 info->port))
2293 goto nla_put_failure;
2294
2295 switch (info->family) {
2296 case AF_INET:
2297 if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, info->addr.s_addr))
2298 goto nla_put_failure;
2299 break;
2300 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
2301 case AF_INET6:
2302 if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &info->addr6))
2303 goto nla_put_failure;
2304 break;
2305 #endif
2306 default:
2307 WARN_ON_ONCE(1);
2308 goto nla_put_failure;
2309 }
2310
2311 genlmsg_end(skb, nlh);
2312 mptcp_nl_mcast_send(net, skb, GFP_ATOMIC);
2313 return;
2314
2315 nla_put_failure:
2316 nlmsg_free(skb);
2317 }
2318
mptcp_event_pm_listener(const struct sock * ssk,enum mptcp_event_type event)2319 void mptcp_event_pm_listener(const struct sock *ssk,
2320 enum mptcp_event_type event)
2321 {
2322 const struct inet_sock *issk = inet_sk(ssk);
2323 struct net *net = sock_net(ssk);
2324 struct nlmsghdr *nlh;
2325 struct sk_buff *skb;
2326
2327 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET))
2328 return;
2329
2330 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
2331 if (!skb)
2332 return;
2333
2334 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, event);
2335 if (!nlh)
2336 goto nla_put_failure;
2337
2338 if (nla_put_u16(skb, MPTCP_ATTR_FAMILY, ssk->sk_family))
2339 goto nla_put_failure;
2340
2341 if (nla_put_be16(skb, MPTCP_ATTR_SPORT, issk->inet_sport))
2342 goto nla_put_failure;
2343
2344 switch (ssk->sk_family) {
2345 case AF_INET:
2346 if (nla_put_in_addr(skb, MPTCP_ATTR_SADDR4, issk->inet_saddr))
2347 goto nla_put_failure;
2348 break;
2349 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
2350 case AF_INET6: {
2351 const struct ipv6_pinfo *np = inet6_sk(ssk);
2352
2353 if (nla_put_in6_addr(skb, MPTCP_ATTR_SADDR6, &np->saddr))
2354 goto nla_put_failure;
2355 break;
2356 }
2357 #endif
2358 default:
2359 WARN_ON_ONCE(1);
2360 goto nla_put_failure;
2361 }
2362
2363 genlmsg_end(skb, nlh);
2364 mptcp_nl_mcast_send(net, skb, GFP_KERNEL);
2365 return;
2366
2367 nla_put_failure:
2368 nlmsg_free(skb);
2369 }
2370
mptcp_event(enum mptcp_event_type type,const struct mptcp_sock * msk,const struct sock * ssk,gfp_t gfp)2371 void mptcp_event(enum mptcp_event_type type, const struct mptcp_sock *msk,
2372 const struct sock *ssk, gfp_t gfp)
2373 {
2374 struct net *net = sock_net((const struct sock *)msk);
2375 struct nlmsghdr *nlh;
2376 struct sk_buff *skb;
2377
2378 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET))
2379 return;
2380
2381 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, gfp);
2382 if (!skb)
2383 return;
2384
2385 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, type);
2386 if (!nlh)
2387 goto nla_put_failure;
2388
2389 switch (type) {
2390 case MPTCP_EVENT_UNSPEC:
2391 WARN_ON_ONCE(1);
2392 break;
2393 case MPTCP_EVENT_CREATED:
2394 case MPTCP_EVENT_ESTABLISHED:
2395 if (mptcp_event_created(skb, msk, ssk) < 0)
2396 goto nla_put_failure;
2397 break;
2398 case MPTCP_EVENT_CLOSED:
2399 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, READ_ONCE(msk->token)) < 0)
2400 goto nla_put_failure;
2401 break;
2402 case MPTCP_EVENT_ANNOUNCED:
2403 case MPTCP_EVENT_REMOVED:
2404 /* call mptcp_event_addr_announced()/removed instead */
2405 WARN_ON_ONCE(1);
2406 break;
2407 case MPTCP_EVENT_SUB_ESTABLISHED:
2408 case MPTCP_EVENT_SUB_PRIORITY:
2409 if (mptcp_event_sub_established(skb, msk, ssk) < 0)
2410 goto nla_put_failure;
2411 break;
2412 case MPTCP_EVENT_SUB_CLOSED:
2413 if (mptcp_event_sub_closed(skb, msk, ssk) < 0)
2414 goto nla_put_failure;
2415 break;
2416 case MPTCP_EVENT_LISTENER_CREATED:
2417 case MPTCP_EVENT_LISTENER_CLOSED:
2418 break;
2419 }
2420
2421 genlmsg_end(skb, nlh);
2422 mptcp_nl_mcast_send(net, skb, gfp);
2423 return;
2424
2425 nla_put_failure:
2426 nlmsg_free(skb);
2427 }
2428
2429 struct genl_family mptcp_genl_family __ro_after_init = {
2430 .name = MPTCP_PM_NAME,
2431 .version = MPTCP_PM_VER,
2432 .netnsok = true,
2433 .module = THIS_MODULE,
2434 .ops = mptcp_pm_nl_ops,
2435 .n_ops = ARRAY_SIZE(mptcp_pm_nl_ops),
2436 .resv_start_op = MPTCP_PM_CMD_SUBFLOW_DESTROY + 1,
2437 .mcgrps = mptcp_pm_mcgrps,
2438 .n_mcgrps = ARRAY_SIZE(mptcp_pm_mcgrps),
2439 };
2440
pm_nl_init_net(struct net * net)2441 static int __net_init pm_nl_init_net(struct net *net)
2442 {
2443 struct pm_nl_pernet *pernet = pm_nl_get_pernet(net);
2444
2445 INIT_LIST_HEAD_RCU(&pernet->local_addr_list);
2446
2447 /* Cit. 2 subflows ought to be enough for anybody. */
2448 pernet->subflows_max = 2;
2449 pernet->next_id = 1;
2450 pernet->stale_loss_cnt = 4;
2451 spin_lock_init(&pernet->lock);
2452
2453 /* No need to initialize other pernet fields, the struct is zeroed at
2454 * allocation time.
2455 */
2456
2457 return 0;
2458 }
2459
pm_nl_exit_net(struct list_head * net_list)2460 static void __net_exit pm_nl_exit_net(struct list_head *net_list)
2461 {
2462 struct net *net;
2463
2464 list_for_each_entry(net, net_list, exit_list) {
2465 struct pm_nl_pernet *pernet = pm_nl_get_pernet(net);
2466
2467 /* net is removed from namespace list, can't race with
2468 * other modifiers, also netns core already waited for a
2469 * RCU grace period.
2470 */
2471 __flush_addrs(&pernet->local_addr_list);
2472 }
2473 }
2474
2475 static struct pernet_operations mptcp_pm_pernet_ops = {
2476 .init = pm_nl_init_net,
2477 .exit_batch = pm_nl_exit_net,
2478 .id = &pm_nl_pernet_id,
2479 .size = sizeof(struct pm_nl_pernet),
2480 };
2481
mptcp_pm_nl_init(void)2482 void __init mptcp_pm_nl_init(void)
2483 {
2484 if (register_pernet_subsys(&mptcp_pm_pernet_ops) < 0)
2485 panic("Failed to register MPTCP PM pernet subsystem.\n");
2486
2487 if (genl_register_family(&mptcp_genl_family))
2488 panic("Failed to register MPTCP PM netlink family\n");
2489 }
2490