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/tcp.h>
12 #include <net/inet_common.h>
13 #include <net/netns/generic.h>
14 #include <net/mptcp.h>
15 #include <net/genetlink.h>
16 #include <uapi/linux/mptcp.h>
17
18 #include "protocol.h"
19 #include "mib.h"
20
21 /* forward declaration */
22 static struct genl_family mptcp_genl_family;
23
24 static int pm_nl_pernet_id;
25
26 struct mptcp_pm_add_entry {
27 struct list_head list;
28 struct mptcp_addr_info addr;
29 u8 retrans_times;
30 struct timer_list add_timer;
31 struct mptcp_sock *sock;
32 };
33
34 struct pm_nl_pernet {
35 /* protects pernet updates */
36 spinlock_t lock;
37 struct list_head local_addr_list;
38 unsigned int addrs;
39 unsigned int stale_loss_cnt;
40 unsigned int add_addr_signal_max;
41 unsigned int add_addr_accept_max;
42 unsigned int local_addr_max;
43 unsigned int subflows_max;
44 unsigned int next_id;
45 DECLARE_BITMAP(id_bitmap, MPTCP_PM_MAX_ADDR_ID + 1);
46 };
47
48 #define MPTCP_PM_ADDR_MAX 8
49 #define ADD_ADDR_RETRANS_MAX 3
50
pm_nl_get_pernet(const struct net * net)51 static struct pm_nl_pernet *pm_nl_get_pernet(const struct net *net)
52 {
53 return net_generic(net, pm_nl_pernet_id);
54 }
55
56 static struct pm_nl_pernet *
pm_nl_get_pernet_from_msk(const struct mptcp_sock * msk)57 pm_nl_get_pernet_from_msk(const struct mptcp_sock *msk)
58 {
59 return pm_nl_get_pernet(sock_net((struct sock *)msk));
60 }
61
mptcp_addresses_equal(const struct mptcp_addr_info * a,const struct mptcp_addr_info * b,bool use_port)62 bool mptcp_addresses_equal(const struct mptcp_addr_info *a,
63 const struct mptcp_addr_info *b, bool use_port)
64 {
65 bool addr_equals = false;
66
67 if (a->family == b->family) {
68 if (a->family == AF_INET)
69 addr_equals = a->addr.s_addr == b->addr.s_addr;
70 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
71 else
72 addr_equals = !ipv6_addr_cmp(&a->addr6, &b->addr6);
73 } else if (a->family == AF_INET) {
74 if (ipv6_addr_v4mapped(&b->addr6))
75 addr_equals = a->addr.s_addr == b->addr6.s6_addr32[3];
76 } else if (b->family == AF_INET) {
77 if (ipv6_addr_v4mapped(&a->addr6))
78 addr_equals = a->addr6.s6_addr32[3] == b->addr.s_addr;
79 #endif
80 }
81
82 if (!addr_equals)
83 return false;
84 if (!use_port)
85 return true;
86
87 return a->port == b->port;
88 }
89
mptcp_local_address(const struct sock_common * skc,struct mptcp_addr_info * addr)90 void mptcp_local_address(const struct sock_common *skc, struct mptcp_addr_info *addr)
91 {
92 addr->family = skc->skc_family;
93 addr->port = htons(skc->skc_num);
94 if (addr->family == AF_INET)
95 addr->addr.s_addr = skc->skc_rcv_saddr;
96 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
97 else if (addr->family == AF_INET6)
98 addr->addr6 = skc->skc_v6_rcv_saddr;
99 #endif
100 }
101
remote_address(const struct sock_common * skc,struct mptcp_addr_info * addr)102 static void remote_address(const struct sock_common *skc,
103 struct mptcp_addr_info *addr)
104 {
105 addr->family = skc->skc_family;
106 addr->port = skc->skc_dport;
107 if (addr->family == AF_INET)
108 addr->addr.s_addr = skc->skc_daddr;
109 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
110 else if (addr->family == AF_INET6)
111 addr->addr6 = skc->skc_v6_daddr;
112 #endif
113 }
114
lookup_subflow_by_saddr(const struct list_head * list,const struct mptcp_addr_info * saddr)115 static bool lookup_subflow_by_saddr(const struct list_head *list,
116 const struct mptcp_addr_info *saddr)
117 {
118 struct mptcp_subflow_context *subflow;
119 struct mptcp_addr_info cur;
120 struct sock_common *skc;
121
122 list_for_each_entry(subflow, list, node) {
123 skc = (struct sock_common *)mptcp_subflow_tcp_sock(subflow);
124
125 mptcp_local_address(skc, &cur);
126 if (mptcp_addresses_equal(&cur, saddr, saddr->port))
127 return true;
128 }
129
130 return false;
131 }
132
lookup_subflow_by_daddr(const struct list_head * list,const struct mptcp_addr_info * daddr)133 static bool lookup_subflow_by_daddr(const struct list_head *list,
134 const struct mptcp_addr_info *daddr)
135 {
136 struct mptcp_subflow_context *subflow;
137 struct mptcp_addr_info cur;
138 struct sock_common *skc;
139
140 list_for_each_entry(subflow, list, node) {
141 skc = (struct sock_common *)mptcp_subflow_tcp_sock(subflow);
142
143 remote_address(skc, &cur);
144 if (mptcp_addresses_equal(&cur, daddr, daddr->port))
145 return true;
146 }
147
148 return false;
149 }
150
151 static struct mptcp_pm_addr_entry *
select_local_address(const struct pm_nl_pernet * pernet,const struct mptcp_sock * msk)152 select_local_address(const struct pm_nl_pernet *pernet,
153 const struct mptcp_sock *msk)
154 {
155 struct mptcp_pm_addr_entry *entry, *ret = NULL;
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 ret = entry;
168 break;
169 }
170 rcu_read_unlock();
171 return ret;
172 }
173
174 static struct mptcp_pm_addr_entry *
select_signal_address(struct pm_nl_pernet * pernet,const struct mptcp_sock * msk)175 select_signal_address(struct pm_nl_pernet *pernet, const struct mptcp_sock *msk)
176 {
177 struct mptcp_pm_addr_entry *entry, *ret = NULL;
178
179 rcu_read_lock();
180 /* do not keep any additional per socket state, just signal
181 * the address list in order.
182 * Note: removal from the local address list during the msk life-cycle
183 * can lead to additional addresses not being announced.
184 */
185 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) {
186 if (!test_bit(entry->addr.id, msk->pm.id_avail_bitmap))
187 continue;
188
189 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL))
190 continue;
191
192 ret = entry;
193 break;
194 }
195 rcu_read_unlock();
196 return ret;
197 }
198
mptcp_pm_get_add_addr_signal_max(const struct mptcp_sock * msk)199 unsigned int mptcp_pm_get_add_addr_signal_max(const struct mptcp_sock *msk)
200 {
201 const struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk);
202
203 return READ_ONCE(pernet->add_addr_signal_max);
204 }
205 EXPORT_SYMBOL_GPL(mptcp_pm_get_add_addr_signal_max);
206
mptcp_pm_get_add_addr_accept_max(const struct mptcp_sock * msk)207 unsigned int mptcp_pm_get_add_addr_accept_max(const struct mptcp_sock *msk)
208 {
209 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk);
210
211 return READ_ONCE(pernet->add_addr_accept_max);
212 }
213 EXPORT_SYMBOL_GPL(mptcp_pm_get_add_addr_accept_max);
214
mptcp_pm_get_subflows_max(const struct mptcp_sock * msk)215 unsigned int mptcp_pm_get_subflows_max(const struct mptcp_sock *msk)
216 {
217 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk);
218
219 return READ_ONCE(pernet->subflows_max);
220 }
221 EXPORT_SYMBOL_GPL(mptcp_pm_get_subflows_max);
222
mptcp_pm_get_local_addr_max(const struct mptcp_sock * msk)223 unsigned int mptcp_pm_get_local_addr_max(const struct mptcp_sock *msk)
224 {
225 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk);
226
227 return READ_ONCE(pernet->local_addr_max);
228 }
229 EXPORT_SYMBOL_GPL(mptcp_pm_get_local_addr_max);
230
mptcp_pm_nl_check_work_pending(struct mptcp_sock * msk)231 bool mptcp_pm_nl_check_work_pending(struct mptcp_sock *msk)
232 {
233 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk);
234
235 if (msk->pm.subflows == mptcp_pm_get_subflows_max(msk) ||
236 (find_next_and_bit(pernet->id_bitmap, msk->pm.id_avail_bitmap,
237 MPTCP_PM_MAX_ADDR_ID + 1, 0) == MPTCP_PM_MAX_ADDR_ID + 1)) {
238 WRITE_ONCE(msk->pm.work_pending, false);
239 return false;
240 }
241 return true;
242 }
243
244 struct mptcp_pm_add_entry *
mptcp_lookup_anno_list_by_saddr(const struct mptcp_sock * msk,const struct mptcp_addr_info * addr)245 mptcp_lookup_anno_list_by_saddr(const struct mptcp_sock *msk,
246 const struct mptcp_addr_info *addr)
247 {
248 struct mptcp_pm_add_entry *entry;
249
250 lockdep_assert_held(&msk->pm.lock);
251
252 list_for_each_entry(entry, &msk->pm.anno_list, list) {
253 if (mptcp_addresses_equal(&entry->addr, addr, true))
254 return entry;
255 }
256
257 return NULL;
258 }
259
mptcp_pm_sport_in_anno_list(struct mptcp_sock * msk,const struct sock * sk)260 bool mptcp_pm_sport_in_anno_list(struct mptcp_sock *msk, const struct sock *sk)
261 {
262 struct mptcp_pm_add_entry *entry;
263 struct mptcp_addr_info saddr;
264 bool ret = false;
265
266 mptcp_local_address((struct sock_common *)sk, &saddr);
267
268 spin_lock_bh(&msk->pm.lock);
269 list_for_each_entry(entry, &msk->pm.anno_list, list) {
270 if (mptcp_addresses_equal(&entry->addr, &saddr, true)) {
271 ret = true;
272 goto out;
273 }
274 }
275
276 out:
277 spin_unlock_bh(&msk->pm.lock);
278 return ret;
279 }
280
mptcp_pm_add_timer(struct timer_list * timer)281 static void mptcp_pm_add_timer(struct timer_list *timer)
282 {
283 struct mptcp_pm_add_entry *entry = from_timer(entry, timer, add_timer);
284 struct mptcp_sock *msk = entry->sock;
285 struct sock *sk = (struct sock *)msk;
286
287 pr_debug("msk=%p", msk);
288
289 if (!msk)
290 return;
291
292 if (inet_sk_state_load(sk) == TCP_CLOSE)
293 return;
294
295 if (!entry->addr.id)
296 return;
297
298 if (mptcp_pm_should_add_signal_addr(msk)) {
299 sk_reset_timer(sk, timer, jiffies + TCP_RTO_MAX / 8);
300 goto out;
301 }
302
303 spin_lock_bh(&msk->pm.lock);
304
305 if (!mptcp_pm_should_add_signal_addr(msk)) {
306 pr_debug("retransmit ADD_ADDR id=%d", entry->addr.id);
307 mptcp_pm_announce_addr(msk, &entry->addr, false);
308 mptcp_pm_add_addr_send_ack(msk);
309 entry->retrans_times++;
310 }
311
312 if (entry->retrans_times < ADD_ADDR_RETRANS_MAX)
313 sk_reset_timer(sk, timer,
314 jiffies + mptcp_get_add_addr_timeout(sock_net(sk)));
315
316 spin_unlock_bh(&msk->pm.lock);
317
318 if (entry->retrans_times == ADD_ADDR_RETRANS_MAX)
319 mptcp_pm_subflow_established(msk);
320
321 out:
322 __sock_put(sk);
323 }
324
325 struct mptcp_pm_add_entry *
mptcp_pm_del_add_timer(struct mptcp_sock * msk,const struct mptcp_addr_info * addr,bool check_id)326 mptcp_pm_del_add_timer(struct mptcp_sock *msk,
327 const struct mptcp_addr_info *addr, bool check_id)
328 {
329 struct mptcp_pm_add_entry *entry;
330 struct sock *sk = (struct sock *)msk;
331
332 spin_lock_bh(&msk->pm.lock);
333 entry = mptcp_lookup_anno_list_by_saddr(msk, addr);
334 if (entry && (!check_id || entry->addr.id == addr->id))
335 entry->retrans_times = ADD_ADDR_RETRANS_MAX;
336 spin_unlock_bh(&msk->pm.lock);
337
338 if (entry && (!check_id || entry->addr.id == addr->id))
339 sk_stop_timer_sync(sk, &entry->add_timer);
340
341 return entry;
342 }
343
mptcp_pm_alloc_anno_list(struct mptcp_sock * msk,const struct mptcp_addr_info * addr)344 bool mptcp_pm_alloc_anno_list(struct mptcp_sock *msk,
345 const struct mptcp_addr_info *addr)
346 {
347 struct mptcp_pm_add_entry *add_entry = NULL;
348 struct sock *sk = (struct sock *)msk;
349 struct net *net = sock_net(sk);
350
351 lockdep_assert_held(&msk->pm.lock);
352
353 add_entry = mptcp_lookup_anno_list_by_saddr(msk, addr);
354
355 if (add_entry) {
356 if (mptcp_pm_is_kernel(msk))
357 return false;
358
359 sk_reset_timer(sk, &add_entry->add_timer,
360 jiffies + mptcp_get_add_addr_timeout(net));
361 return true;
362 }
363
364 add_entry = kmalloc(sizeof(*add_entry), GFP_ATOMIC);
365 if (!add_entry)
366 return false;
367
368 list_add(&add_entry->list, &msk->pm.anno_list);
369
370 add_entry->addr = *addr;
371 add_entry->sock = msk;
372 add_entry->retrans_times = 0;
373
374 timer_setup(&add_entry->add_timer, mptcp_pm_add_timer, 0);
375 sk_reset_timer(sk, &add_entry->add_timer,
376 jiffies + mptcp_get_add_addr_timeout(net));
377
378 return true;
379 }
380
mptcp_pm_free_anno_list(struct mptcp_sock * msk)381 void mptcp_pm_free_anno_list(struct mptcp_sock *msk)
382 {
383 struct mptcp_pm_add_entry *entry, *tmp;
384 struct sock *sk = (struct sock *)msk;
385 LIST_HEAD(free_list);
386
387 pr_debug("msk=%p", msk);
388
389 spin_lock_bh(&msk->pm.lock);
390 list_splice_init(&msk->pm.anno_list, &free_list);
391 spin_unlock_bh(&msk->pm.lock);
392
393 list_for_each_entry_safe(entry, tmp, &free_list, list) {
394 sk_stop_timer_sync(sk, &entry->add_timer);
395 kfree(entry);
396 }
397 }
398
399 /* Fill all the remote addresses into the array addrs[],
400 * and return the array size.
401 */
fill_remote_addresses_vec(struct mptcp_sock * msk,struct mptcp_addr_info * local,bool fullmesh,struct mptcp_addr_info * addrs)402 static unsigned int fill_remote_addresses_vec(struct mptcp_sock *msk,
403 struct mptcp_addr_info *local,
404 bool fullmesh,
405 struct mptcp_addr_info *addrs)
406 {
407 bool deny_id0 = READ_ONCE(msk->pm.remote_deny_join_id0);
408 struct sock *sk = (struct sock *)msk, *ssk;
409 struct mptcp_subflow_context *subflow;
410 struct mptcp_addr_info remote = { 0 };
411 unsigned int subflows_max;
412 int i = 0;
413
414 subflows_max = mptcp_pm_get_subflows_max(msk);
415 remote_address((struct sock_common *)sk, &remote);
416
417 /* Non-fullmesh endpoint, fill in the single entry
418 * corresponding to the primary MPC subflow remote address
419 */
420 if (!fullmesh) {
421 if (deny_id0)
422 return 0;
423
424 if (!mptcp_pm_addr_families_match(sk, local, &remote))
425 return 0;
426
427 msk->pm.subflows++;
428 addrs[i++] = remote;
429 } else {
430 DECLARE_BITMAP(unavail_id, MPTCP_PM_MAX_ADDR_ID + 1);
431
432 /* Forbid creation of new subflows matching existing
433 * ones, possibly already created by incoming ADD_ADDR
434 */
435 bitmap_zero(unavail_id, MPTCP_PM_MAX_ADDR_ID + 1);
436 mptcp_for_each_subflow(msk, subflow)
437 if (READ_ONCE(subflow->local_id) == local->id)
438 __set_bit(subflow->remote_id, unavail_id);
439
440 mptcp_for_each_subflow(msk, subflow) {
441 ssk = mptcp_subflow_tcp_sock(subflow);
442 remote_address((struct sock_common *)ssk, &addrs[i]);
443 addrs[i].id = READ_ONCE(subflow->remote_id);
444 if (deny_id0 && !addrs[i].id)
445 continue;
446
447 if (test_bit(addrs[i].id, unavail_id))
448 continue;
449
450 if (!mptcp_pm_addr_families_match(sk, local, &addrs[i]))
451 continue;
452
453 if (msk->pm.subflows < subflows_max) {
454 /* forbid creating multiple address towards
455 * this id
456 */
457 __set_bit(addrs[i].id, unavail_id);
458 msk->pm.subflows++;
459 i++;
460 }
461 }
462 }
463
464 return i;
465 }
466
__mptcp_pm_send_ack(struct mptcp_sock * msk,struct mptcp_subflow_context * subflow,bool prio,bool backup)467 static void __mptcp_pm_send_ack(struct mptcp_sock *msk, struct mptcp_subflow_context *subflow,
468 bool prio, bool backup)
469 {
470 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
471 bool slow;
472
473 pr_debug("send ack for %s",
474 prio ? "mp_prio" : (mptcp_pm_should_add_signal(msk) ? "add_addr" : "rm_addr"));
475
476 slow = lock_sock_fast(ssk);
477 if (prio) {
478 subflow->send_mp_prio = 1;
479 subflow->backup = backup;
480 subflow->request_bkup = backup;
481 }
482
483 __mptcp_subflow_send_ack(ssk);
484 unlock_sock_fast(ssk, slow);
485 }
486
mptcp_pm_send_ack(struct mptcp_sock * msk,struct mptcp_subflow_context * subflow,bool prio,bool backup)487 static void mptcp_pm_send_ack(struct mptcp_sock *msk, struct mptcp_subflow_context *subflow,
488 bool prio, bool backup)
489 {
490 spin_unlock_bh(&msk->pm.lock);
491 __mptcp_pm_send_ack(msk, subflow, prio, backup);
492 spin_lock_bh(&msk->pm.lock);
493 }
494
495 static struct mptcp_pm_addr_entry *
__lookup_addr_by_id(struct pm_nl_pernet * pernet,unsigned int id)496 __lookup_addr_by_id(struct pm_nl_pernet *pernet, unsigned int id)
497 {
498 struct mptcp_pm_addr_entry *entry;
499
500 list_for_each_entry(entry, &pernet->local_addr_list, list) {
501 if (entry->addr.id == id)
502 return entry;
503 }
504 return NULL;
505 }
506
507 static struct mptcp_pm_addr_entry *
__lookup_addr(struct pm_nl_pernet * pernet,const struct mptcp_addr_info * info,bool lookup_by_id)508 __lookup_addr(struct pm_nl_pernet *pernet, const struct mptcp_addr_info *info,
509 bool lookup_by_id)
510 {
511 struct mptcp_pm_addr_entry *entry;
512
513 list_for_each_entry(entry, &pernet->local_addr_list, list) {
514 if ((!lookup_by_id &&
515 mptcp_addresses_equal(&entry->addr, info, entry->addr.port)) ||
516 (lookup_by_id && entry->addr.id == info->id))
517 return entry;
518 }
519 return NULL;
520 }
521
mptcp_pm_create_subflow_or_signal_addr(struct mptcp_sock * msk)522 static void mptcp_pm_create_subflow_or_signal_addr(struct mptcp_sock *msk)
523 {
524 struct sock *sk = (struct sock *)msk;
525 struct mptcp_pm_addr_entry *local;
526 unsigned int add_addr_signal_max;
527 unsigned int local_addr_max;
528 struct pm_nl_pernet *pernet;
529 unsigned int subflows_max;
530
531 pernet = pm_nl_get_pernet(sock_net(sk));
532
533 add_addr_signal_max = mptcp_pm_get_add_addr_signal_max(msk);
534 local_addr_max = mptcp_pm_get_local_addr_max(msk);
535 subflows_max = mptcp_pm_get_subflows_max(msk);
536
537 /* do lazy endpoint usage accounting for the MPC subflows */
538 if (unlikely(!(msk->pm.status & BIT(MPTCP_PM_MPC_ENDPOINT_ACCOUNTED))) && msk->first) {
539 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(msk->first);
540 struct mptcp_pm_addr_entry *entry;
541 struct mptcp_addr_info mpc_addr;
542 bool backup = false;
543
544 mptcp_local_address((struct sock_common *)msk->first, &mpc_addr);
545 rcu_read_lock();
546 entry = __lookup_addr(pernet, &mpc_addr, false);
547 if (entry) {
548 __clear_bit(entry->addr.id, msk->pm.id_avail_bitmap);
549 msk->mpc_endpoint_id = entry->addr.id;
550 backup = !!(entry->flags & MPTCP_PM_ADDR_FLAG_BACKUP);
551 }
552 rcu_read_unlock();
553
554 if (backup)
555 mptcp_pm_send_ack(msk, subflow, true, backup);
556
557 msk->pm.status |= BIT(MPTCP_PM_MPC_ENDPOINT_ACCOUNTED);
558 }
559
560 pr_debug("local %d:%d signal %d:%d subflows %d:%d\n",
561 msk->pm.local_addr_used, local_addr_max,
562 msk->pm.add_addr_signaled, add_addr_signal_max,
563 msk->pm.subflows, subflows_max);
564
565 /* check first for announce */
566 if (msk->pm.add_addr_signaled < add_addr_signal_max) {
567 local = select_signal_address(pernet, msk);
568
569 /* due to racing events on both ends we can reach here while
570 * previous add address is still running: if we invoke now
571 * mptcp_pm_announce_addr(), that will fail and the
572 * corresponding id will be marked as used.
573 * Instead let the PM machinery reschedule us when the
574 * current address announce will be completed.
575 */
576 if (msk->pm.addr_signal & BIT(MPTCP_ADD_ADDR_SIGNAL))
577 return;
578
579 if (local) {
580 if (mptcp_pm_alloc_anno_list(msk, &local->addr)) {
581 __clear_bit(local->addr.id, msk->pm.id_avail_bitmap);
582 msk->pm.add_addr_signaled++;
583 mptcp_pm_announce_addr(msk, &local->addr, false);
584 mptcp_pm_nl_addr_send_ack(msk);
585 }
586 }
587 }
588
589 /* check if should create a new subflow */
590 while (msk->pm.local_addr_used < local_addr_max &&
591 msk->pm.subflows < subflows_max) {
592 struct mptcp_addr_info addrs[MPTCP_PM_ADDR_MAX];
593 bool fullmesh;
594 int i, nr;
595
596 local = select_local_address(pernet, msk);
597 if (!local)
598 break;
599
600 fullmesh = !!(local->flags & MPTCP_PM_ADDR_FLAG_FULLMESH);
601
602 msk->pm.local_addr_used++;
603 __clear_bit(local->addr.id, msk->pm.id_avail_bitmap);
604 nr = fill_remote_addresses_vec(msk, &local->addr, fullmesh, addrs);
605 if (nr == 0)
606 continue;
607
608 spin_unlock_bh(&msk->pm.lock);
609 for (i = 0; i < nr; i++)
610 __mptcp_subflow_connect(sk, &local->addr, &addrs[i]);
611 spin_lock_bh(&msk->pm.lock);
612 }
613 mptcp_pm_nl_check_work_pending(msk);
614 }
615
mptcp_pm_nl_fully_established(struct mptcp_sock * msk)616 static void mptcp_pm_nl_fully_established(struct mptcp_sock *msk)
617 {
618 mptcp_pm_create_subflow_or_signal_addr(msk);
619 }
620
mptcp_pm_nl_subflow_established(struct mptcp_sock * msk)621 static void mptcp_pm_nl_subflow_established(struct mptcp_sock *msk)
622 {
623 mptcp_pm_create_subflow_or_signal_addr(msk);
624 }
625
626 /* Fill all the local addresses into the array addrs[],
627 * and return the array size.
628 */
fill_local_addresses_vec(struct mptcp_sock * msk,struct mptcp_addr_info * remote,struct mptcp_addr_info * addrs)629 static unsigned int fill_local_addresses_vec(struct mptcp_sock *msk,
630 struct mptcp_addr_info *remote,
631 struct mptcp_addr_info *addrs)
632 {
633 struct sock *sk = (struct sock *)msk;
634 struct mptcp_pm_addr_entry *entry;
635 struct pm_nl_pernet *pernet;
636 unsigned int subflows_max;
637 int i = 0;
638
639 pernet = pm_nl_get_pernet_from_msk(msk);
640 subflows_max = mptcp_pm_get_subflows_max(msk);
641
642 rcu_read_lock();
643 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) {
644 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_FULLMESH))
645 continue;
646
647 if (!mptcp_pm_addr_families_match(sk, &entry->addr, remote))
648 continue;
649
650 if (msk->pm.subflows < subflows_max) {
651 msk->pm.subflows++;
652 addrs[i++] = entry->addr;
653 }
654 }
655 rcu_read_unlock();
656
657 /* If the array is empty, fill in the single
658 * 'IPADDRANY' local address
659 */
660 if (!i) {
661 struct mptcp_addr_info local;
662
663 memset(&local, 0, sizeof(local));
664 local.family =
665 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
666 remote->family == AF_INET6 &&
667 ipv6_addr_v4mapped(&remote->addr6) ? AF_INET :
668 #endif
669 remote->family;
670
671 if (!mptcp_pm_addr_families_match(sk, &local, remote))
672 return 0;
673
674 msk->pm.subflows++;
675 addrs[i++] = local;
676 }
677
678 return i;
679 }
680
mptcp_pm_nl_add_addr_received(struct mptcp_sock * msk)681 static void mptcp_pm_nl_add_addr_received(struct mptcp_sock *msk)
682 {
683 struct mptcp_addr_info addrs[MPTCP_PM_ADDR_MAX];
684 struct sock *sk = (struct sock *)msk;
685 unsigned int add_addr_accept_max;
686 struct mptcp_addr_info remote;
687 unsigned int subflows_max;
688 bool sf_created = false;
689 int i, nr;
690
691 add_addr_accept_max = mptcp_pm_get_add_addr_accept_max(msk);
692 subflows_max = mptcp_pm_get_subflows_max(msk);
693
694 pr_debug("accepted %d:%d remote family %d",
695 msk->pm.add_addr_accepted, add_addr_accept_max,
696 msk->pm.remote.family);
697
698 remote = msk->pm.remote;
699 mptcp_pm_announce_addr(msk, &remote, true);
700 mptcp_pm_nl_addr_send_ack(msk);
701
702 if (lookup_subflow_by_daddr(&msk->conn_list, &remote))
703 return;
704
705 /* pick id 0 port, if none is provided the remote address */
706 if (!remote.port)
707 remote.port = sk->sk_dport;
708
709 /* connect to the specified remote address, using whatever
710 * local address the routing configuration will pick.
711 */
712 nr = fill_local_addresses_vec(msk, &remote, addrs);
713 if (nr == 0)
714 return;
715
716 spin_unlock_bh(&msk->pm.lock);
717 for (i = 0; i < nr; i++)
718 if (__mptcp_subflow_connect(sk, &addrs[i], &remote) == 0)
719 sf_created = true;
720 spin_lock_bh(&msk->pm.lock);
721
722 if (sf_created) {
723 msk->pm.add_addr_accepted++;
724 if (msk->pm.add_addr_accepted >= add_addr_accept_max ||
725 msk->pm.subflows >= subflows_max)
726 WRITE_ONCE(msk->pm.accept_addr, false);
727 }
728 }
729
mptcp_pm_nl_addr_send_ack(struct mptcp_sock * msk)730 void mptcp_pm_nl_addr_send_ack(struct mptcp_sock *msk)
731 {
732 struct mptcp_subflow_context *subflow;
733
734 msk_owned_by_me(msk);
735 lockdep_assert_held(&msk->pm.lock);
736
737 if (!mptcp_pm_should_add_signal(msk) &&
738 !mptcp_pm_should_rm_signal(msk))
739 return;
740
741 subflow = list_first_entry_or_null(&msk->conn_list, typeof(*subflow), node);
742 if (subflow)
743 mptcp_pm_send_ack(msk, subflow, false, false);
744 }
745
mptcp_pm_nl_mp_prio_send_ack(struct mptcp_sock * msk,struct mptcp_addr_info * addr,struct mptcp_addr_info * rem,u8 bkup)746 int mptcp_pm_nl_mp_prio_send_ack(struct mptcp_sock *msk,
747 struct mptcp_addr_info *addr,
748 struct mptcp_addr_info *rem,
749 u8 bkup)
750 {
751 struct mptcp_subflow_context *subflow;
752
753 pr_debug("bkup=%d", bkup);
754
755 mptcp_for_each_subflow(msk, subflow) {
756 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
757 struct mptcp_addr_info local, remote;
758
759 mptcp_local_address((struct sock_common *)ssk, &local);
760 if (!mptcp_addresses_equal(&local, addr, addr->port))
761 continue;
762
763 if (rem && rem->family != AF_UNSPEC) {
764 remote_address((struct sock_common *)ssk, &remote);
765 if (!mptcp_addresses_equal(&remote, rem, rem->port))
766 continue;
767 }
768
769 __mptcp_pm_send_ack(msk, subflow, true, bkup);
770 return 0;
771 }
772
773 return -EINVAL;
774 }
775
mptcp_local_id_match(const struct mptcp_sock * msk,u8 local_id,u8 id)776 static bool mptcp_local_id_match(const struct mptcp_sock *msk, u8 local_id, u8 id)
777 {
778 return local_id == id || (!local_id && msk->mpc_endpoint_id == id);
779 }
780
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)781 static void mptcp_pm_nl_rm_addr_or_subflow(struct mptcp_sock *msk,
782 const struct mptcp_rm_list *rm_list,
783 enum linux_mptcp_mib_field rm_type)
784 {
785 struct mptcp_subflow_context *subflow, *tmp;
786 struct sock *sk = (struct sock *)msk;
787 u8 i;
788
789 pr_debug("%s rm_list_nr %d",
790 rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow", rm_list->nr);
791
792 msk_owned_by_me(msk);
793
794 if (sk->sk_state == TCP_LISTEN)
795 return;
796
797 if (!rm_list->nr)
798 return;
799
800 if (list_empty(&msk->conn_list))
801 return;
802
803 for (i = 0; i < rm_list->nr; i++) {
804 u8 rm_id = rm_list->ids[i];
805 bool removed = false;
806
807 mptcp_for_each_subflow_safe(msk, subflow, tmp) {
808 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
809 u8 remote_id = READ_ONCE(subflow->remote_id);
810 int how = RCV_SHUTDOWN | SEND_SHUTDOWN;
811 u8 id = subflow_get_local_id(subflow);
812
813 if (rm_type == MPTCP_MIB_RMADDR && remote_id != rm_id)
814 continue;
815 if (rm_type == MPTCP_MIB_RMSUBFLOW && !mptcp_local_id_match(msk, id, rm_id))
816 continue;
817
818 pr_debug(" -> %s rm_list_ids[%d]=%u local_id=%u remote_id=%u mpc_id=%u",
819 rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow",
820 i, rm_id, id, remote_id, msk->mpc_endpoint_id);
821 spin_unlock_bh(&msk->pm.lock);
822 mptcp_subflow_shutdown(sk, ssk, how);
823
824 /* the following takes care of updating the subflows counter */
825 mptcp_close_ssk(sk, ssk, subflow);
826 spin_lock_bh(&msk->pm.lock);
827
828 removed = true;
829 if (rm_type == MPTCP_MIB_RMSUBFLOW)
830 __MPTCP_INC_STATS(sock_net(sk), rm_type);
831 }
832 if (rm_type == MPTCP_MIB_RMSUBFLOW)
833 __set_bit(rm_id ? rm_id : msk->mpc_endpoint_id, msk->pm.id_avail_bitmap);
834 else if (rm_type == MPTCP_MIB_RMADDR)
835 __MPTCP_INC_STATS(sock_net(sk), rm_type);
836 if (!removed)
837 continue;
838
839 if (!mptcp_pm_is_kernel(msk))
840 continue;
841
842 if (rm_type == MPTCP_MIB_RMADDR) {
843 msk->pm.add_addr_accepted--;
844 WRITE_ONCE(msk->pm.accept_addr, true);
845 } else if (rm_type == MPTCP_MIB_RMSUBFLOW) {
846 msk->pm.local_addr_used--;
847 }
848 }
849 }
850
mptcp_pm_nl_rm_addr_received(struct mptcp_sock * msk)851 static void mptcp_pm_nl_rm_addr_received(struct mptcp_sock *msk)
852 {
853 mptcp_pm_nl_rm_addr_or_subflow(msk, &msk->pm.rm_list_rx, MPTCP_MIB_RMADDR);
854 }
855
mptcp_pm_nl_rm_subflow_received(struct mptcp_sock * msk,const struct mptcp_rm_list * rm_list)856 void mptcp_pm_nl_rm_subflow_received(struct mptcp_sock *msk,
857 const struct mptcp_rm_list *rm_list)
858 {
859 mptcp_pm_nl_rm_addr_or_subflow(msk, rm_list, MPTCP_MIB_RMSUBFLOW);
860 }
861
mptcp_pm_nl_work(struct mptcp_sock * msk)862 void mptcp_pm_nl_work(struct mptcp_sock *msk)
863 {
864 struct mptcp_pm_data *pm = &msk->pm;
865
866 msk_owned_by_me(msk);
867
868 if (!(pm->status & MPTCP_PM_WORK_MASK))
869 return;
870
871 spin_lock_bh(&msk->pm.lock);
872
873 pr_debug("msk=%p status=%x", msk, pm->status);
874 if (pm->status & BIT(MPTCP_PM_ADD_ADDR_RECEIVED)) {
875 pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_RECEIVED);
876 mptcp_pm_nl_add_addr_received(msk);
877 }
878 if (pm->status & BIT(MPTCP_PM_ADD_ADDR_SEND_ACK)) {
879 pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_SEND_ACK);
880 mptcp_pm_nl_addr_send_ack(msk);
881 }
882 if (pm->status & BIT(MPTCP_PM_RM_ADDR_RECEIVED)) {
883 pm->status &= ~BIT(MPTCP_PM_RM_ADDR_RECEIVED);
884 mptcp_pm_nl_rm_addr_received(msk);
885 }
886 if (pm->status & BIT(MPTCP_PM_ESTABLISHED)) {
887 pm->status &= ~BIT(MPTCP_PM_ESTABLISHED);
888 mptcp_pm_nl_fully_established(msk);
889 }
890 if (pm->status & BIT(MPTCP_PM_SUBFLOW_ESTABLISHED)) {
891 pm->status &= ~BIT(MPTCP_PM_SUBFLOW_ESTABLISHED);
892 mptcp_pm_nl_subflow_established(msk);
893 }
894
895 spin_unlock_bh(&msk->pm.lock);
896 }
897
address_use_port(struct mptcp_pm_addr_entry * entry)898 static bool address_use_port(struct mptcp_pm_addr_entry *entry)
899 {
900 return (entry->flags &
901 (MPTCP_PM_ADDR_FLAG_SIGNAL | MPTCP_PM_ADDR_FLAG_SUBFLOW)) ==
902 MPTCP_PM_ADDR_FLAG_SIGNAL;
903 }
904
905 /* caller must ensure the RCU grace period is already elapsed */
__mptcp_pm_release_addr_entry(struct mptcp_pm_addr_entry * entry)906 static void __mptcp_pm_release_addr_entry(struct mptcp_pm_addr_entry *entry)
907 {
908 if (entry->lsk)
909 sock_release(entry->lsk);
910 kfree(entry);
911 }
912
mptcp_pm_nl_append_new_local_addr(struct pm_nl_pernet * pernet,struct mptcp_pm_addr_entry * entry,bool needs_id)913 static int mptcp_pm_nl_append_new_local_addr(struct pm_nl_pernet *pernet,
914 struct mptcp_pm_addr_entry *entry,
915 bool needs_id)
916 {
917 struct mptcp_pm_addr_entry *cur, *del_entry = NULL;
918 unsigned int addr_max;
919 int ret = -EINVAL;
920
921 spin_lock_bh(&pernet->lock);
922 /* to keep the code simple, don't do IDR-like allocation for address ID,
923 * just bail when we exceed limits
924 */
925 if (pernet->next_id == MPTCP_PM_MAX_ADDR_ID)
926 pernet->next_id = 1;
927 if (pernet->addrs >= MPTCP_PM_ADDR_MAX) {
928 ret = -ERANGE;
929 goto out;
930 }
931 if (test_bit(entry->addr.id, pernet->id_bitmap)) {
932 ret = -EBUSY;
933 goto out;
934 }
935
936 /* do not insert duplicate address, differentiate on port only
937 * singled addresses
938 */
939 if (!address_use_port(entry))
940 entry->addr.port = 0;
941 list_for_each_entry(cur, &pernet->local_addr_list, list) {
942 if (mptcp_addresses_equal(&cur->addr, &entry->addr,
943 cur->addr.port || entry->addr.port)) {
944 /* allow replacing the exiting endpoint only if such
945 * endpoint is an implicit one and the user-space
946 * did not provide an endpoint id
947 */
948 if (!(cur->flags & MPTCP_PM_ADDR_FLAG_IMPLICIT)) {
949 ret = -EEXIST;
950 goto out;
951 }
952 if (entry->addr.id)
953 goto out;
954
955 pernet->addrs--;
956 entry->addr.id = cur->addr.id;
957 list_del_rcu(&cur->list);
958 del_entry = cur;
959 break;
960 }
961 }
962
963 if (!entry->addr.id && needs_id) {
964 find_next:
965 entry->addr.id = find_next_zero_bit(pernet->id_bitmap,
966 MPTCP_PM_MAX_ADDR_ID + 1,
967 pernet->next_id);
968 if (!entry->addr.id && pernet->next_id != 1) {
969 pernet->next_id = 1;
970 goto find_next;
971 }
972 }
973
974 if (!entry->addr.id && needs_id)
975 goto out;
976
977 __set_bit(entry->addr.id, pernet->id_bitmap);
978 if (entry->addr.id > pernet->next_id)
979 pernet->next_id = entry->addr.id;
980
981 if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) {
982 addr_max = pernet->add_addr_signal_max;
983 WRITE_ONCE(pernet->add_addr_signal_max, addr_max + 1);
984 }
985 if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) {
986 addr_max = pernet->local_addr_max;
987 WRITE_ONCE(pernet->local_addr_max, addr_max + 1);
988 }
989
990 pernet->addrs++;
991 if (!entry->addr.port)
992 list_add_tail_rcu(&entry->list, &pernet->local_addr_list);
993 else
994 list_add_rcu(&entry->list, &pernet->local_addr_list);
995 ret = entry->addr.id;
996
997 out:
998 spin_unlock_bh(&pernet->lock);
999
1000 /* just replaced an existing entry, free it */
1001 if (del_entry) {
1002 synchronize_rcu();
1003 __mptcp_pm_release_addr_entry(del_entry);
1004 }
1005 return ret;
1006 }
1007
1008 static struct lock_class_key mptcp_slock_keys[2];
1009 static struct lock_class_key mptcp_keys[2];
1010
mptcp_pm_nl_create_listen_socket(struct sock * sk,struct mptcp_pm_addr_entry * entry)1011 static int mptcp_pm_nl_create_listen_socket(struct sock *sk,
1012 struct mptcp_pm_addr_entry *entry)
1013 {
1014 bool is_ipv6 = sk->sk_family == AF_INET6;
1015 int addrlen = sizeof(struct sockaddr_in);
1016 struct sockaddr_storage addr;
1017 struct sock *newsk, *ssk;
1018 int backlog = 1024;
1019 int err;
1020
1021 err = sock_create_kern(sock_net(sk), entry->addr.family,
1022 SOCK_STREAM, IPPROTO_MPTCP, &entry->lsk);
1023 if (err)
1024 return err;
1025
1026 newsk = entry->lsk->sk;
1027 if (!newsk)
1028 return -EINVAL;
1029
1030 /* The subflow socket lock is acquired in a nested to the msk one
1031 * in several places, even by the TCP stack, and this msk is a kernel
1032 * socket: lockdep complains. Instead of propagating the _nested
1033 * modifiers in several places, re-init the lock class for the msk
1034 * socket to an mptcp specific one.
1035 */
1036 sock_lock_init_class_and_name(newsk,
1037 is_ipv6 ? "mlock-AF_INET6" : "mlock-AF_INET",
1038 &mptcp_slock_keys[is_ipv6],
1039 is_ipv6 ? "msk_lock-AF_INET6" : "msk_lock-AF_INET",
1040 &mptcp_keys[is_ipv6]);
1041
1042 lock_sock(newsk);
1043 ssk = __mptcp_nmpc_sk(mptcp_sk(newsk));
1044 release_sock(newsk);
1045 if (IS_ERR(ssk))
1046 return PTR_ERR(ssk);
1047
1048 mptcp_info2sockaddr(&entry->addr, &addr, entry->addr.family);
1049 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1050 if (entry->addr.family == AF_INET6)
1051 addrlen = sizeof(struct sockaddr_in6);
1052 #endif
1053 if (ssk->sk_family == AF_INET)
1054 err = inet_bind_sk(ssk, (struct sockaddr *)&addr, addrlen);
1055 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1056 else if (ssk->sk_family == AF_INET6)
1057 err = inet6_bind_sk(ssk, (struct sockaddr *)&addr, addrlen);
1058 #endif
1059 if (err)
1060 return err;
1061
1062 /* We don't use mptcp_set_state() here because it needs to be called
1063 * under the msk socket lock. For the moment, that will not bring
1064 * anything more than only calling inet_sk_state_store(), because the
1065 * old status is known (TCP_CLOSE).
1066 */
1067 inet_sk_state_store(newsk, TCP_LISTEN);
1068 lock_sock(ssk);
1069 err = __inet_listen_sk(ssk, backlog);
1070 if (!err)
1071 mptcp_event_pm_listener(ssk, MPTCP_EVENT_LISTENER_CREATED);
1072 release_sock(ssk);
1073 return err;
1074 }
1075
mptcp_pm_nl_get_local_id(struct mptcp_sock * msk,struct mptcp_addr_info * skc)1076 int mptcp_pm_nl_get_local_id(struct mptcp_sock *msk, struct mptcp_addr_info *skc)
1077 {
1078 struct mptcp_pm_addr_entry *entry;
1079 struct pm_nl_pernet *pernet;
1080 int ret = -1;
1081
1082 pernet = pm_nl_get_pernet_from_msk(msk);
1083
1084 rcu_read_lock();
1085 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) {
1086 if (mptcp_addresses_equal(&entry->addr, skc, entry->addr.port)) {
1087 ret = entry->addr.id;
1088 break;
1089 }
1090 }
1091 rcu_read_unlock();
1092 if (ret >= 0)
1093 return ret;
1094
1095 /* address not found, add to local list */
1096 entry = kmalloc(sizeof(*entry), GFP_ATOMIC);
1097 if (!entry)
1098 return -ENOMEM;
1099
1100 entry->addr = *skc;
1101 entry->addr.id = 0;
1102 entry->addr.port = 0;
1103 entry->ifindex = 0;
1104 entry->flags = MPTCP_PM_ADDR_FLAG_IMPLICIT;
1105 entry->lsk = NULL;
1106 ret = mptcp_pm_nl_append_new_local_addr(pernet, entry, true);
1107 if (ret < 0)
1108 kfree(entry);
1109
1110 return ret;
1111 }
1112
1113 #define MPTCP_PM_CMD_GRP_OFFSET 0
1114 #define MPTCP_PM_EV_GRP_OFFSET 1
1115
1116 static const struct genl_multicast_group mptcp_pm_mcgrps[] = {
1117 [MPTCP_PM_CMD_GRP_OFFSET] = { .name = MPTCP_PM_CMD_GRP_NAME, },
1118 [MPTCP_PM_EV_GRP_OFFSET] = { .name = MPTCP_PM_EV_GRP_NAME,
1119 .flags = GENL_UNS_ADMIN_PERM,
1120 },
1121 };
1122
1123 static const struct nla_policy
1124 mptcp_pm_addr_policy[MPTCP_PM_ADDR_ATTR_MAX + 1] = {
1125 [MPTCP_PM_ADDR_ATTR_FAMILY] = { .type = NLA_U16, },
1126 [MPTCP_PM_ADDR_ATTR_ID] = { .type = NLA_U8, },
1127 [MPTCP_PM_ADDR_ATTR_ADDR4] = { .type = NLA_U32, },
1128 [MPTCP_PM_ADDR_ATTR_ADDR6] =
1129 NLA_POLICY_EXACT_LEN(sizeof(struct in6_addr)),
1130 [MPTCP_PM_ADDR_ATTR_PORT] = { .type = NLA_U16 },
1131 [MPTCP_PM_ADDR_ATTR_FLAGS] = { .type = NLA_U32 },
1132 [MPTCP_PM_ADDR_ATTR_IF_IDX] = { .type = NLA_S32 },
1133 };
1134
1135 static const struct nla_policy mptcp_pm_policy[MPTCP_PM_ATTR_MAX + 1] = {
1136 [MPTCP_PM_ATTR_ADDR] =
1137 NLA_POLICY_NESTED(mptcp_pm_addr_policy),
1138 [MPTCP_PM_ATTR_RCV_ADD_ADDRS] = { .type = NLA_U32, },
1139 [MPTCP_PM_ATTR_SUBFLOWS] = { .type = NLA_U32, },
1140 [MPTCP_PM_ATTR_TOKEN] = { .type = NLA_U32, },
1141 [MPTCP_PM_ATTR_LOC_ID] = { .type = NLA_U8, },
1142 [MPTCP_PM_ATTR_ADDR_REMOTE] =
1143 NLA_POLICY_NESTED(mptcp_pm_addr_policy),
1144 };
1145
mptcp_pm_nl_subflow_chk_stale(const struct mptcp_sock * msk,struct sock * ssk)1146 void mptcp_pm_nl_subflow_chk_stale(const struct mptcp_sock *msk, struct sock *ssk)
1147 {
1148 struct mptcp_subflow_context *iter, *subflow = mptcp_subflow_ctx(ssk);
1149 struct sock *sk = (struct sock *)msk;
1150 unsigned int active_max_loss_cnt;
1151 struct net *net = sock_net(sk);
1152 unsigned int stale_loss_cnt;
1153 bool slow;
1154
1155 stale_loss_cnt = mptcp_stale_loss_cnt(net);
1156 if (subflow->stale || !stale_loss_cnt || subflow->stale_count <= stale_loss_cnt)
1157 return;
1158
1159 /* look for another available subflow not in loss state */
1160 active_max_loss_cnt = max_t(int, stale_loss_cnt - 1, 1);
1161 mptcp_for_each_subflow(msk, iter) {
1162 if (iter != subflow && mptcp_subflow_active(iter) &&
1163 iter->stale_count < active_max_loss_cnt) {
1164 /* we have some alternatives, try to mark this subflow as idle ...*/
1165 slow = lock_sock_fast(ssk);
1166 if (!tcp_rtx_and_write_queues_empty(ssk)) {
1167 subflow->stale = 1;
1168 __mptcp_retransmit_pending_data(sk);
1169 MPTCP_INC_STATS(net, MPTCP_MIB_SUBFLOWSTALE);
1170 }
1171 unlock_sock_fast(ssk, slow);
1172
1173 /* always try to push the pending data regardless of re-injections:
1174 * we can possibly use backup subflows now, and subflow selection
1175 * is cheap under the msk socket lock
1176 */
1177 __mptcp_push_pending(sk, 0);
1178 return;
1179 }
1180 }
1181 }
1182
mptcp_pm_family_to_addr(int family)1183 static int mptcp_pm_family_to_addr(int family)
1184 {
1185 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1186 if (family == AF_INET6)
1187 return MPTCP_PM_ADDR_ATTR_ADDR6;
1188 #endif
1189 return MPTCP_PM_ADDR_ATTR_ADDR4;
1190 }
1191
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)1192 static int mptcp_pm_parse_pm_addr_attr(struct nlattr *tb[],
1193 const struct nlattr *attr,
1194 struct genl_info *info,
1195 struct mptcp_addr_info *addr,
1196 bool require_family)
1197 {
1198 int err, addr_addr;
1199
1200 if (!attr) {
1201 GENL_SET_ERR_MSG(info, "missing address info");
1202 return -EINVAL;
1203 }
1204
1205 /* no validation needed - was already done via nested policy */
1206 err = nla_parse_nested_deprecated(tb, MPTCP_PM_ADDR_ATTR_MAX, attr,
1207 mptcp_pm_addr_policy, info->extack);
1208 if (err)
1209 return err;
1210
1211 if (tb[MPTCP_PM_ADDR_ATTR_ID])
1212 addr->id = nla_get_u8(tb[MPTCP_PM_ADDR_ATTR_ID]);
1213
1214 if (!tb[MPTCP_PM_ADDR_ATTR_FAMILY]) {
1215 if (!require_family)
1216 return 0;
1217
1218 NL_SET_ERR_MSG_ATTR(info->extack, attr,
1219 "missing family");
1220 return -EINVAL;
1221 }
1222
1223 addr->family = nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_FAMILY]);
1224 if (addr->family != AF_INET
1225 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1226 && addr->family != AF_INET6
1227 #endif
1228 ) {
1229 NL_SET_ERR_MSG_ATTR(info->extack, attr,
1230 "unknown address family");
1231 return -EINVAL;
1232 }
1233 addr_addr = mptcp_pm_family_to_addr(addr->family);
1234 if (!tb[addr_addr]) {
1235 NL_SET_ERR_MSG_ATTR(info->extack, attr,
1236 "missing address data");
1237 return -EINVAL;
1238 }
1239
1240 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1241 if (addr->family == AF_INET6)
1242 addr->addr6 = nla_get_in6_addr(tb[addr_addr]);
1243 else
1244 #endif
1245 addr->addr.s_addr = nla_get_in_addr(tb[addr_addr]);
1246
1247 if (tb[MPTCP_PM_ADDR_ATTR_PORT])
1248 addr->port = htons(nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_PORT]));
1249
1250 return 0;
1251 }
1252
mptcp_pm_parse_addr(struct nlattr * attr,struct genl_info * info,struct mptcp_addr_info * addr)1253 int mptcp_pm_parse_addr(struct nlattr *attr, struct genl_info *info,
1254 struct mptcp_addr_info *addr)
1255 {
1256 struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1];
1257
1258 memset(addr, 0, sizeof(*addr));
1259
1260 return mptcp_pm_parse_pm_addr_attr(tb, attr, info, addr, true);
1261 }
1262
mptcp_pm_parse_entry(struct nlattr * attr,struct genl_info * info,bool require_family,struct mptcp_pm_addr_entry * entry)1263 int mptcp_pm_parse_entry(struct nlattr *attr, struct genl_info *info,
1264 bool require_family,
1265 struct mptcp_pm_addr_entry *entry)
1266 {
1267 struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1];
1268 int err;
1269
1270 memset(entry, 0, sizeof(*entry));
1271
1272 err = mptcp_pm_parse_pm_addr_attr(tb, attr, info, &entry->addr, require_family);
1273 if (err)
1274 return err;
1275
1276 if (tb[MPTCP_PM_ADDR_ATTR_IF_IDX]) {
1277 u32 val = nla_get_s32(tb[MPTCP_PM_ADDR_ATTR_IF_IDX]);
1278
1279 entry->ifindex = val;
1280 }
1281
1282 if (tb[MPTCP_PM_ADDR_ATTR_FLAGS])
1283 entry->flags = nla_get_u32(tb[MPTCP_PM_ADDR_ATTR_FLAGS]);
1284
1285 if (tb[MPTCP_PM_ADDR_ATTR_PORT])
1286 entry->addr.port = htons(nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_PORT]));
1287
1288 return 0;
1289 }
1290
genl_info_pm_nl(struct genl_info * info)1291 static struct pm_nl_pernet *genl_info_pm_nl(struct genl_info *info)
1292 {
1293 return pm_nl_get_pernet(genl_info_net(info));
1294 }
1295
mptcp_nl_add_subflow_or_signal_addr(struct net * net)1296 static int mptcp_nl_add_subflow_or_signal_addr(struct net *net)
1297 {
1298 struct mptcp_sock *msk;
1299 long s_slot = 0, s_num = 0;
1300
1301 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1302 struct sock *sk = (struct sock *)msk;
1303
1304 if (!READ_ONCE(msk->fully_established) ||
1305 mptcp_pm_is_userspace(msk))
1306 goto next;
1307
1308 lock_sock(sk);
1309 spin_lock_bh(&msk->pm.lock);
1310 mptcp_pm_create_subflow_or_signal_addr(msk);
1311 spin_unlock_bh(&msk->pm.lock);
1312 release_sock(sk);
1313
1314 next:
1315 sock_put(sk);
1316 cond_resched();
1317 }
1318
1319 return 0;
1320 }
1321
mptcp_pm_has_addr_attr_id(const struct nlattr * attr,struct genl_info * info)1322 static bool mptcp_pm_has_addr_attr_id(const struct nlattr *attr,
1323 struct genl_info *info)
1324 {
1325 struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1];
1326
1327 if (!nla_parse_nested_deprecated(tb, MPTCP_PM_ADDR_ATTR_MAX, attr,
1328 mptcp_pm_addr_policy, info->extack) &&
1329 tb[MPTCP_PM_ADDR_ATTR_ID])
1330 return true;
1331 return false;
1332 }
1333
mptcp_nl_cmd_add_addr(struct sk_buff * skb,struct genl_info * info)1334 static int mptcp_nl_cmd_add_addr(struct sk_buff *skb, struct genl_info *info)
1335 {
1336 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
1337 struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1338 struct mptcp_pm_addr_entry addr, *entry;
1339 int ret;
1340
1341 ret = mptcp_pm_parse_entry(attr, info, true, &addr);
1342 if (ret < 0)
1343 return ret;
1344
1345 if (addr.addr.port && !(addr.flags & MPTCP_PM_ADDR_FLAG_SIGNAL)) {
1346 GENL_SET_ERR_MSG(info, "flags must have signal when using port");
1347 return -EINVAL;
1348 }
1349
1350 if (addr.flags & MPTCP_PM_ADDR_FLAG_SIGNAL &&
1351 addr.flags & MPTCP_PM_ADDR_FLAG_FULLMESH) {
1352 GENL_SET_ERR_MSG(info, "flags mustn't have both signal and fullmesh");
1353 return -EINVAL;
1354 }
1355
1356 if (addr.flags & MPTCP_PM_ADDR_FLAG_IMPLICIT) {
1357 GENL_SET_ERR_MSG(info, "can't create IMPLICIT endpoint");
1358 return -EINVAL;
1359 }
1360
1361 entry = kzalloc(sizeof(*entry), GFP_KERNEL_ACCOUNT);
1362 if (!entry) {
1363 GENL_SET_ERR_MSG(info, "can't allocate addr");
1364 return -ENOMEM;
1365 }
1366
1367 *entry = addr;
1368 if (entry->addr.port) {
1369 ret = mptcp_pm_nl_create_listen_socket(skb->sk, entry);
1370 if (ret) {
1371 GENL_SET_ERR_MSG_FMT(info, "create listen socket error: %d", ret);
1372 goto out_free;
1373 }
1374 }
1375 ret = mptcp_pm_nl_append_new_local_addr(pernet, entry,
1376 !mptcp_pm_has_addr_attr_id(attr, info));
1377 if (ret < 0) {
1378 GENL_SET_ERR_MSG_FMT(info, "too many addresses or duplicate one: %d", ret);
1379 goto out_free;
1380 }
1381
1382 mptcp_nl_add_subflow_or_signal_addr(sock_net(skb->sk));
1383 return 0;
1384
1385 out_free:
1386 __mptcp_pm_release_addr_entry(entry);
1387 return ret;
1388 }
1389
mptcp_pm_nl_get_flags_and_ifindex_by_id(struct mptcp_sock * msk,unsigned int id,u8 * flags,int * ifindex)1390 int mptcp_pm_nl_get_flags_and_ifindex_by_id(struct mptcp_sock *msk, unsigned int id,
1391 u8 *flags, int *ifindex)
1392 {
1393 struct mptcp_pm_addr_entry *entry;
1394 struct sock *sk = (struct sock *)msk;
1395 struct net *net = sock_net(sk);
1396
1397 rcu_read_lock();
1398 entry = __lookup_addr_by_id(pm_nl_get_pernet(net), id);
1399 if (entry) {
1400 *flags = entry->flags;
1401 *ifindex = entry->ifindex;
1402 }
1403 rcu_read_unlock();
1404
1405 return 0;
1406 }
1407
remove_anno_list_by_saddr(struct mptcp_sock * msk,const struct mptcp_addr_info * addr)1408 static bool remove_anno_list_by_saddr(struct mptcp_sock *msk,
1409 const struct mptcp_addr_info *addr)
1410 {
1411 struct mptcp_pm_add_entry *entry;
1412
1413 entry = mptcp_pm_del_add_timer(msk, addr, false);
1414 if (entry) {
1415 list_del(&entry->list);
1416 kfree(entry);
1417 return true;
1418 }
1419
1420 return false;
1421 }
1422
mptcp_pm_remove_anno_addr(struct mptcp_sock * msk,const struct mptcp_addr_info * addr,bool force)1423 static bool mptcp_pm_remove_anno_addr(struct mptcp_sock *msk,
1424 const struct mptcp_addr_info *addr,
1425 bool force)
1426 {
1427 struct mptcp_rm_list list = { .nr = 0 };
1428 bool ret;
1429
1430 list.ids[list.nr++] = addr->id;
1431
1432 ret = remove_anno_list_by_saddr(msk, addr);
1433 if (ret || force) {
1434 spin_lock_bh(&msk->pm.lock);
1435 mptcp_pm_remove_addr(msk, &list);
1436 spin_unlock_bh(&msk->pm.lock);
1437 }
1438 return ret;
1439 }
1440
mptcp_nl_remove_subflow_and_signal_addr(struct net * net,const struct mptcp_pm_addr_entry * entry)1441 static int mptcp_nl_remove_subflow_and_signal_addr(struct net *net,
1442 const struct mptcp_pm_addr_entry *entry)
1443 {
1444 const struct mptcp_addr_info *addr = &entry->addr;
1445 struct mptcp_rm_list list = { .nr = 0 };
1446 long s_slot = 0, s_num = 0;
1447 struct mptcp_sock *msk;
1448
1449 pr_debug("remove_id=%d", addr->id);
1450
1451 list.ids[list.nr++] = addr->id;
1452
1453 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1454 struct sock *sk = (struct sock *)msk;
1455 bool remove_subflow;
1456
1457 if (mptcp_pm_is_userspace(msk))
1458 goto next;
1459
1460 if (list_empty(&msk->conn_list)) {
1461 mptcp_pm_remove_anno_addr(msk, addr, false);
1462 goto next;
1463 }
1464
1465 lock_sock(sk);
1466 remove_subflow = lookup_subflow_by_saddr(&msk->conn_list, addr);
1467 mptcp_pm_remove_anno_addr(msk, addr, remove_subflow &&
1468 !(entry->flags & MPTCP_PM_ADDR_FLAG_IMPLICIT));
1469 if (remove_subflow)
1470 mptcp_pm_remove_subflow(msk, &list);
1471 release_sock(sk);
1472
1473 next:
1474 sock_put(sk);
1475 cond_resched();
1476 }
1477
1478 return 0;
1479 }
1480
mptcp_nl_remove_id_zero_address(struct net * net,struct mptcp_addr_info * addr)1481 static int mptcp_nl_remove_id_zero_address(struct net *net,
1482 struct mptcp_addr_info *addr)
1483 {
1484 struct mptcp_rm_list list = { .nr = 0 };
1485 long s_slot = 0, s_num = 0;
1486 struct mptcp_sock *msk;
1487
1488 list.ids[list.nr++] = 0;
1489
1490 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1491 struct sock *sk = (struct sock *)msk;
1492 struct mptcp_addr_info msk_local;
1493
1494 if (list_empty(&msk->conn_list) || mptcp_pm_is_userspace(msk))
1495 goto next;
1496
1497 mptcp_local_address((struct sock_common *)msk, &msk_local);
1498 if (!mptcp_addresses_equal(&msk_local, addr, addr->port))
1499 goto next;
1500
1501 lock_sock(sk);
1502 spin_lock_bh(&msk->pm.lock);
1503 mptcp_pm_remove_addr(msk, &list);
1504 mptcp_pm_nl_rm_subflow_received(msk, &list);
1505 spin_unlock_bh(&msk->pm.lock);
1506 release_sock(sk);
1507
1508 next:
1509 sock_put(sk);
1510 cond_resched();
1511 }
1512
1513 return 0;
1514 }
1515
mptcp_nl_cmd_del_addr(struct sk_buff * skb,struct genl_info * info)1516 static int mptcp_nl_cmd_del_addr(struct sk_buff *skb, struct genl_info *info)
1517 {
1518 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
1519 struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1520 struct mptcp_pm_addr_entry addr, *entry;
1521 unsigned int addr_max;
1522 int ret;
1523
1524 ret = mptcp_pm_parse_entry(attr, info, false, &addr);
1525 if (ret < 0)
1526 return ret;
1527
1528 /* the zero id address is special: the first address used by the msk
1529 * always gets such an id, so different subflows can have different zero
1530 * id addresses. Additionally zero id is not accounted for in id_bitmap.
1531 * Let's use an 'mptcp_rm_list' instead of the common remove code.
1532 */
1533 if (addr.addr.id == 0)
1534 return mptcp_nl_remove_id_zero_address(sock_net(skb->sk), &addr.addr);
1535
1536 spin_lock_bh(&pernet->lock);
1537 entry = __lookup_addr_by_id(pernet, addr.addr.id);
1538 if (!entry) {
1539 GENL_SET_ERR_MSG(info, "address not found");
1540 spin_unlock_bh(&pernet->lock);
1541 return -EINVAL;
1542 }
1543 if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) {
1544 addr_max = pernet->add_addr_signal_max;
1545 WRITE_ONCE(pernet->add_addr_signal_max, addr_max - 1);
1546 }
1547 if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) {
1548 addr_max = pernet->local_addr_max;
1549 WRITE_ONCE(pernet->local_addr_max, addr_max - 1);
1550 }
1551
1552 pernet->addrs--;
1553 list_del_rcu(&entry->list);
1554 __clear_bit(entry->addr.id, pernet->id_bitmap);
1555 spin_unlock_bh(&pernet->lock);
1556
1557 mptcp_nl_remove_subflow_and_signal_addr(sock_net(skb->sk), entry);
1558 synchronize_rcu();
1559 __mptcp_pm_release_addr_entry(entry);
1560
1561 return ret;
1562 }
1563
mptcp_pm_remove_addrs(struct mptcp_sock * msk,struct list_head * rm_list)1564 void mptcp_pm_remove_addrs(struct mptcp_sock *msk, struct list_head *rm_list)
1565 {
1566 struct mptcp_rm_list alist = { .nr = 0 };
1567 struct mptcp_pm_addr_entry *entry;
1568
1569 list_for_each_entry(entry, rm_list, list) {
1570 if ((remove_anno_list_by_saddr(msk, &entry->addr) ||
1571 lookup_subflow_by_saddr(&msk->conn_list, &entry->addr)) &&
1572 alist.nr < MPTCP_RM_IDS_MAX)
1573 alist.ids[alist.nr++] = entry->addr.id;
1574 }
1575
1576 if (alist.nr) {
1577 spin_lock_bh(&msk->pm.lock);
1578 mptcp_pm_remove_addr(msk, &alist);
1579 spin_unlock_bh(&msk->pm.lock);
1580 }
1581 }
1582
mptcp_pm_remove_addrs_and_subflows(struct mptcp_sock * msk,struct list_head * rm_list)1583 void mptcp_pm_remove_addrs_and_subflows(struct mptcp_sock *msk,
1584 struct list_head *rm_list)
1585 {
1586 struct mptcp_rm_list alist = { .nr = 0 }, slist = { .nr = 0 };
1587 struct mptcp_pm_addr_entry *entry;
1588
1589 list_for_each_entry(entry, rm_list, list) {
1590 if (lookup_subflow_by_saddr(&msk->conn_list, &entry->addr) &&
1591 slist.nr < MPTCP_RM_IDS_MAX)
1592 slist.ids[slist.nr++] = entry->addr.id;
1593
1594 if (remove_anno_list_by_saddr(msk, &entry->addr) &&
1595 alist.nr < MPTCP_RM_IDS_MAX)
1596 alist.ids[alist.nr++] = entry->addr.id;
1597 }
1598
1599 if (alist.nr) {
1600 spin_lock_bh(&msk->pm.lock);
1601 mptcp_pm_remove_addr(msk, &alist);
1602 spin_unlock_bh(&msk->pm.lock);
1603 }
1604 if (slist.nr)
1605 mptcp_pm_remove_subflow(msk, &slist);
1606 }
1607
mptcp_nl_remove_addrs_list(struct net * net,struct list_head * rm_list)1608 static void mptcp_nl_remove_addrs_list(struct net *net,
1609 struct list_head *rm_list)
1610 {
1611 long s_slot = 0, s_num = 0;
1612 struct mptcp_sock *msk;
1613
1614 if (list_empty(rm_list))
1615 return;
1616
1617 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1618 struct sock *sk = (struct sock *)msk;
1619
1620 if (!mptcp_pm_is_userspace(msk)) {
1621 lock_sock(sk);
1622 mptcp_pm_remove_addrs_and_subflows(msk, rm_list);
1623 release_sock(sk);
1624 }
1625
1626 sock_put(sk);
1627 cond_resched();
1628 }
1629 }
1630
1631 /* caller must ensure the RCU grace period is already elapsed */
__flush_addrs(struct list_head * list)1632 static void __flush_addrs(struct list_head *list)
1633 {
1634 while (!list_empty(list)) {
1635 struct mptcp_pm_addr_entry *cur;
1636
1637 cur = list_entry(list->next,
1638 struct mptcp_pm_addr_entry, list);
1639 list_del_rcu(&cur->list);
1640 __mptcp_pm_release_addr_entry(cur);
1641 }
1642 }
1643
__reset_counters(struct pm_nl_pernet * pernet)1644 static void __reset_counters(struct pm_nl_pernet *pernet)
1645 {
1646 WRITE_ONCE(pernet->add_addr_signal_max, 0);
1647 WRITE_ONCE(pernet->add_addr_accept_max, 0);
1648 WRITE_ONCE(pernet->local_addr_max, 0);
1649 pernet->addrs = 0;
1650 }
1651
mptcp_nl_cmd_flush_addrs(struct sk_buff * skb,struct genl_info * info)1652 static int mptcp_nl_cmd_flush_addrs(struct sk_buff *skb, struct genl_info *info)
1653 {
1654 struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1655 LIST_HEAD(free_list);
1656
1657 spin_lock_bh(&pernet->lock);
1658 list_splice_init(&pernet->local_addr_list, &free_list);
1659 __reset_counters(pernet);
1660 pernet->next_id = 1;
1661 bitmap_zero(pernet->id_bitmap, MPTCP_PM_MAX_ADDR_ID + 1);
1662 spin_unlock_bh(&pernet->lock);
1663 mptcp_nl_remove_addrs_list(sock_net(skb->sk), &free_list);
1664 synchronize_rcu();
1665 __flush_addrs(&free_list);
1666 return 0;
1667 }
1668
mptcp_nl_fill_addr(struct sk_buff * skb,struct mptcp_pm_addr_entry * entry)1669 static int mptcp_nl_fill_addr(struct sk_buff *skb,
1670 struct mptcp_pm_addr_entry *entry)
1671 {
1672 struct mptcp_addr_info *addr = &entry->addr;
1673 struct nlattr *attr;
1674
1675 attr = nla_nest_start(skb, MPTCP_PM_ATTR_ADDR);
1676 if (!attr)
1677 return -EMSGSIZE;
1678
1679 if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_FAMILY, addr->family))
1680 goto nla_put_failure;
1681 if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_PORT, ntohs(addr->port)))
1682 goto nla_put_failure;
1683 if (nla_put_u8(skb, MPTCP_PM_ADDR_ATTR_ID, addr->id))
1684 goto nla_put_failure;
1685 if (nla_put_u32(skb, MPTCP_PM_ADDR_ATTR_FLAGS, entry->flags))
1686 goto nla_put_failure;
1687 if (entry->ifindex &&
1688 nla_put_s32(skb, MPTCP_PM_ADDR_ATTR_IF_IDX, entry->ifindex))
1689 goto nla_put_failure;
1690
1691 if (addr->family == AF_INET &&
1692 nla_put_in_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR4,
1693 addr->addr.s_addr))
1694 goto nla_put_failure;
1695 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1696 else if (addr->family == AF_INET6 &&
1697 nla_put_in6_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR6, &addr->addr6))
1698 goto nla_put_failure;
1699 #endif
1700 nla_nest_end(skb, attr);
1701 return 0;
1702
1703 nla_put_failure:
1704 nla_nest_cancel(skb, attr);
1705 return -EMSGSIZE;
1706 }
1707
mptcp_nl_cmd_get_addr(struct sk_buff * skb,struct genl_info * info)1708 static int mptcp_nl_cmd_get_addr(struct sk_buff *skb, struct genl_info *info)
1709 {
1710 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
1711 struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1712 struct mptcp_pm_addr_entry addr, *entry;
1713 struct sk_buff *msg;
1714 void *reply;
1715 int ret;
1716
1717 ret = mptcp_pm_parse_entry(attr, info, false, &addr);
1718 if (ret < 0)
1719 return ret;
1720
1721 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1722 if (!msg)
1723 return -ENOMEM;
1724
1725 reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0,
1726 info->genlhdr->cmd);
1727 if (!reply) {
1728 GENL_SET_ERR_MSG(info, "not enough space in Netlink message");
1729 ret = -EMSGSIZE;
1730 goto fail;
1731 }
1732
1733 spin_lock_bh(&pernet->lock);
1734 entry = __lookup_addr_by_id(pernet, addr.addr.id);
1735 if (!entry) {
1736 GENL_SET_ERR_MSG(info, "address not found");
1737 ret = -EINVAL;
1738 goto unlock_fail;
1739 }
1740
1741 ret = mptcp_nl_fill_addr(msg, entry);
1742 if (ret)
1743 goto unlock_fail;
1744
1745 genlmsg_end(msg, reply);
1746 ret = genlmsg_reply(msg, info);
1747 spin_unlock_bh(&pernet->lock);
1748 return ret;
1749
1750 unlock_fail:
1751 spin_unlock_bh(&pernet->lock);
1752
1753 fail:
1754 nlmsg_free(msg);
1755 return ret;
1756 }
1757
mptcp_nl_cmd_dump_addrs(struct sk_buff * msg,struct netlink_callback * cb)1758 static int mptcp_nl_cmd_dump_addrs(struct sk_buff *msg,
1759 struct netlink_callback *cb)
1760 {
1761 struct net *net = sock_net(msg->sk);
1762 struct mptcp_pm_addr_entry *entry;
1763 struct pm_nl_pernet *pernet;
1764 int id = cb->args[0];
1765 void *hdr;
1766 int i;
1767
1768 pernet = pm_nl_get_pernet(net);
1769
1770 spin_lock_bh(&pernet->lock);
1771 for (i = id; i < MPTCP_PM_MAX_ADDR_ID + 1; i++) {
1772 if (test_bit(i, pernet->id_bitmap)) {
1773 entry = __lookup_addr_by_id(pernet, i);
1774 if (!entry)
1775 break;
1776
1777 if (entry->addr.id <= id)
1778 continue;
1779
1780 hdr = genlmsg_put(msg, NETLINK_CB(cb->skb).portid,
1781 cb->nlh->nlmsg_seq, &mptcp_genl_family,
1782 NLM_F_MULTI, MPTCP_PM_CMD_GET_ADDR);
1783 if (!hdr)
1784 break;
1785
1786 if (mptcp_nl_fill_addr(msg, entry) < 0) {
1787 genlmsg_cancel(msg, hdr);
1788 break;
1789 }
1790
1791 id = entry->addr.id;
1792 genlmsg_end(msg, hdr);
1793 }
1794 }
1795 spin_unlock_bh(&pernet->lock);
1796
1797 cb->args[0] = id;
1798 return msg->len;
1799 }
1800
parse_limit(struct genl_info * info,int id,unsigned int * limit)1801 static int parse_limit(struct genl_info *info, int id, unsigned int *limit)
1802 {
1803 struct nlattr *attr = info->attrs[id];
1804
1805 if (!attr)
1806 return 0;
1807
1808 *limit = nla_get_u32(attr);
1809 if (*limit > MPTCP_PM_ADDR_MAX) {
1810 GENL_SET_ERR_MSG(info, "limit greater than maximum");
1811 return -EINVAL;
1812 }
1813 return 0;
1814 }
1815
1816 static int
mptcp_nl_cmd_set_limits(struct sk_buff * skb,struct genl_info * info)1817 mptcp_nl_cmd_set_limits(struct sk_buff *skb, struct genl_info *info)
1818 {
1819 struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1820 unsigned int rcv_addrs, subflows;
1821 int ret;
1822
1823 spin_lock_bh(&pernet->lock);
1824 rcv_addrs = pernet->add_addr_accept_max;
1825 ret = parse_limit(info, MPTCP_PM_ATTR_RCV_ADD_ADDRS, &rcv_addrs);
1826 if (ret)
1827 goto unlock;
1828
1829 subflows = pernet->subflows_max;
1830 ret = parse_limit(info, MPTCP_PM_ATTR_SUBFLOWS, &subflows);
1831 if (ret)
1832 goto unlock;
1833
1834 WRITE_ONCE(pernet->add_addr_accept_max, rcv_addrs);
1835 WRITE_ONCE(pernet->subflows_max, subflows);
1836
1837 unlock:
1838 spin_unlock_bh(&pernet->lock);
1839 return ret;
1840 }
1841
1842 static int
mptcp_nl_cmd_get_limits(struct sk_buff * skb,struct genl_info * info)1843 mptcp_nl_cmd_get_limits(struct sk_buff *skb, struct genl_info *info)
1844 {
1845 struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1846 struct sk_buff *msg;
1847 void *reply;
1848
1849 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1850 if (!msg)
1851 return -ENOMEM;
1852
1853 reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0,
1854 MPTCP_PM_CMD_GET_LIMITS);
1855 if (!reply)
1856 goto fail;
1857
1858 if (nla_put_u32(msg, MPTCP_PM_ATTR_RCV_ADD_ADDRS,
1859 READ_ONCE(pernet->add_addr_accept_max)))
1860 goto fail;
1861
1862 if (nla_put_u32(msg, MPTCP_PM_ATTR_SUBFLOWS,
1863 READ_ONCE(pernet->subflows_max)))
1864 goto fail;
1865
1866 genlmsg_end(msg, reply);
1867 return genlmsg_reply(msg, info);
1868
1869 fail:
1870 GENL_SET_ERR_MSG(info, "not enough space in Netlink message");
1871 nlmsg_free(msg);
1872 return -EMSGSIZE;
1873 }
1874
mptcp_pm_nl_fullmesh(struct mptcp_sock * msk,struct mptcp_addr_info * addr)1875 static void mptcp_pm_nl_fullmesh(struct mptcp_sock *msk,
1876 struct mptcp_addr_info *addr)
1877 {
1878 struct mptcp_rm_list list = { .nr = 0 };
1879
1880 list.ids[list.nr++] = addr->id;
1881
1882 spin_lock_bh(&msk->pm.lock);
1883 mptcp_pm_nl_rm_subflow_received(msk, &list);
1884 mptcp_pm_create_subflow_or_signal_addr(msk);
1885 spin_unlock_bh(&msk->pm.lock);
1886 }
1887
mptcp_nl_set_flags(struct net * net,struct mptcp_addr_info * addr,u8 bkup,u8 changed)1888 static int mptcp_nl_set_flags(struct net *net,
1889 struct mptcp_addr_info *addr,
1890 u8 bkup, u8 changed)
1891 {
1892 long s_slot = 0, s_num = 0;
1893 struct mptcp_sock *msk;
1894 int ret = -EINVAL;
1895
1896 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1897 struct sock *sk = (struct sock *)msk;
1898
1899 if (list_empty(&msk->conn_list) || mptcp_pm_is_userspace(msk))
1900 goto next;
1901
1902 lock_sock(sk);
1903 if (changed & MPTCP_PM_ADDR_FLAG_BACKUP)
1904 ret = mptcp_pm_nl_mp_prio_send_ack(msk, addr, NULL, bkup);
1905 if (changed & MPTCP_PM_ADDR_FLAG_FULLMESH)
1906 mptcp_pm_nl_fullmesh(msk, addr);
1907 release_sock(sk);
1908
1909 next:
1910 sock_put(sk);
1911 cond_resched();
1912 }
1913
1914 return ret;
1915 }
1916
mptcp_pm_nl_set_flags(struct net * net,struct mptcp_pm_addr_entry * addr,u8 bkup)1917 int mptcp_pm_nl_set_flags(struct net *net, struct mptcp_pm_addr_entry *addr, u8 bkup)
1918 {
1919 struct pm_nl_pernet *pernet = pm_nl_get_pernet(net);
1920 u8 changed, mask = MPTCP_PM_ADDR_FLAG_BACKUP |
1921 MPTCP_PM_ADDR_FLAG_FULLMESH;
1922 struct mptcp_pm_addr_entry *entry;
1923 u8 lookup_by_id = 0;
1924
1925 if (addr->addr.family == AF_UNSPEC) {
1926 lookup_by_id = 1;
1927 if (!addr->addr.id)
1928 return -EOPNOTSUPP;
1929 }
1930
1931 spin_lock_bh(&pernet->lock);
1932 entry = __lookup_addr(pernet, &addr->addr, lookup_by_id);
1933 if (!entry) {
1934 spin_unlock_bh(&pernet->lock);
1935 return -EINVAL;
1936 }
1937 if ((addr->flags & MPTCP_PM_ADDR_FLAG_FULLMESH) &&
1938 (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL)) {
1939 spin_unlock_bh(&pernet->lock);
1940 return -EINVAL;
1941 }
1942
1943 changed = (addr->flags ^ entry->flags) & mask;
1944 entry->flags = (entry->flags & ~mask) | (addr->flags & mask);
1945 *addr = *entry;
1946 spin_unlock_bh(&pernet->lock);
1947
1948 mptcp_nl_set_flags(net, &addr->addr, bkup, changed);
1949 return 0;
1950 }
1951
mptcp_nl_cmd_set_flags(struct sk_buff * skb,struct genl_info * info)1952 static int mptcp_nl_cmd_set_flags(struct sk_buff *skb, struct genl_info *info)
1953 {
1954 struct mptcp_pm_addr_entry remote = { .addr = { .family = AF_UNSPEC }, };
1955 struct mptcp_pm_addr_entry addr = { .addr = { .family = AF_UNSPEC }, };
1956 struct nlattr *attr_rem = info->attrs[MPTCP_PM_ATTR_ADDR_REMOTE];
1957 struct nlattr *token = info->attrs[MPTCP_PM_ATTR_TOKEN];
1958 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
1959 struct net *net = sock_net(skb->sk);
1960 u8 bkup = 0;
1961 int ret;
1962
1963 ret = mptcp_pm_parse_entry(attr, info, false, &addr);
1964 if (ret < 0)
1965 return ret;
1966
1967 if (attr_rem) {
1968 ret = mptcp_pm_parse_entry(attr_rem, info, false, &remote);
1969 if (ret < 0)
1970 return ret;
1971 }
1972
1973 if (addr.flags & MPTCP_PM_ADDR_FLAG_BACKUP)
1974 bkup = 1;
1975
1976 return mptcp_pm_set_flags(net, token, &addr, &remote, bkup);
1977 }
1978
mptcp_nl_mcast_send(struct net * net,struct sk_buff * nlskb,gfp_t gfp)1979 static void mptcp_nl_mcast_send(struct net *net, struct sk_buff *nlskb, gfp_t gfp)
1980 {
1981 genlmsg_multicast_netns(&mptcp_genl_family, net,
1982 nlskb, 0, MPTCP_PM_EV_GRP_OFFSET, gfp);
1983 }
1984
mptcp_userspace_pm_active(const struct mptcp_sock * msk)1985 bool mptcp_userspace_pm_active(const struct mptcp_sock *msk)
1986 {
1987 return genl_has_listeners(&mptcp_genl_family,
1988 sock_net((const struct sock *)msk),
1989 MPTCP_PM_EV_GRP_OFFSET);
1990 }
1991
mptcp_event_add_subflow(struct sk_buff * skb,const struct sock * ssk)1992 static int mptcp_event_add_subflow(struct sk_buff *skb, const struct sock *ssk)
1993 {
1994 const struct inet_sock *issk = inet_sk(ssk);
1995 const struct mptcp_subflow_context *sf;
1996
1997 if (nla_put_u16(skb, MPTCP_ATTR_FAMILY, ssk->sk_family))
1998 return -EMSGSIZE;
1999
2000 switch (ssk->sk_family) {
2001 case AF_INET:
2002 if (nla_put_in_addr(skb, MPTCP_ATTR_SADDR4, issk->inet_saddr))
2003 return -EMSGSIZE;
2004 if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, issk->inet_daddr))
2005 return -EMSGSIZE;
2006 break;
2007 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
2008 case AF_INET6: {
2009 const struct ipv6_pinfo *np = inet6_sk(ssk);
2010
2011 if (nla_put_in6_addr(skb, MPTCP_ATTR_SADDR6, &np->saddr))
2012 return -EMSGSIZE;
2013 if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &ssk->sk_v6_daddr))
2014 return -EMSGSIZE;
2015 break;
2016 }
2017 #endif
2018 default:
2019 WARN_ON_ONCE(1);
2020 return -EMSGSIZE;
2021 }
2022
2023 if (nla_put_be16(skb, MPTCP_ATTR_SPORT, issk->inet_sport))
2024 return -EMSGSIZE;
2025 if (nla_put_be16(skb, MPTCP_ATTR_DPORT, issk->inet_dport))
2026 return -EMSGSIZE;
2027
2028 sf = mptcp_subflow_ctx(ssk);
2029 if (WARN_ON_ONCE(!sf))
2030 return -EINVAL;
2031
2032 if (nla_put_u8(skb, MPTCP_ATTR_LOC_ID, subflow_get_local_id(sf)))
2033 return -EMSGSIZE;
2034
2035 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, sf->remote_id))
2036 return -EMSGSIZE;
2037
2038 return 0;
2039 }
2040
mptcp_event_put_token_and_ssk(struct sk_buff * skb,const struct mptcp_sock * msk,const struct sock * ssk)2041 static int mptcp_event_put_token_and_ssk(struct sk_buff *skb,
2042 const struct mptcp_sock *msk,
2043 const struct sock *ssk)
2044 {
2045 const struct sock *sk = (const struct sock *)msk;
2046 const struct mptcp_subflow_context *sf;
2047 u8 sk_err;
2048
2049 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token))
2050 return -EMSGSIZE;
2051
2052 if (mptcp_event_add_subflow(skb, ssk))
2053 return -EMSGSIZE;
2054
2055 sf = mptcp_subflow_ctx(ssk);
2056 if (WARN_ON_ONCE(!sf))
2057 return -EINVAL;
2058
2059 if (nla_put_u8(skb, MPTCP_ATTR_BACKUP, sf->backup))
2060 return -EMSGSIZE;
2061
2062 if (ssk->sk_bound_dev_if &&
2063 nla_put_s32(skb, MPTCP_ATTR_IF_IDX, ssk->sk_bound_dev_if))
2064 return -EMSGSIZE;
2065
2066 sk_err = READ_ONCE(ssk->sk_err);
2067 if (sk_err && sk->sk_state == TCP_ESTABLISHED &&
2068 nla_put_u8(skb, MPTCP_ATTR_ERROR, sk_err))
2069 return -EMSGSIZE;
2070
2071 return 0;
2072 }
2073
mptcp_event_sub_established(struct sk_buff * skb,const struct mptcp_sock * msk,const struct sock * ssk)2074 static int mptcp_event_sub_established(struct sk_buff *skb,
2075 const struct mptcp_sock *msk,
2076 const struct sock *ssk)
2077 {
2078 return mptcp_event_put_token_and_ssk(skb, msk, ssk);
2079 }
2080
mptcp_event_sub_closed(struct sk_buff * skb,const struct mptcp_sock * msk,const struct sock * ssk)2081 static int mptcp_event_sub_closed(struct sk_buff *skb,
2082 const struct mptcp_sock *msk,
2083 const struct sock *ssk)
2084 {
2085 const struct mptcp_subflow_context *sf;
2086
2087 if (mptcp_event_put_token_and_ssk(skb, msk, ssk))
2088 return -EMSGSIZE;
2089
2090 sf = mptcp_subflow_ctx(ssk);
2091 if (!sf->reset_seen)
2092 return 0;
2093
2094 if (nla_put_u32(skb, MPTCP_ATTR_RESET_REASON, sf->reset_reason))
2095 return -EMSGSIZE;
2096
2097 if (nla_put_u32(skb, MPTCP_ATTR_RESET_FLAGS, sf->reset_transient))
2098 return -EMSGSIZE;
2099
2100 return 0;
2101 }
2102
mptcp_event_created(struct sk_buff * skb,const struct mptcp_sock * msk,const struct sock * ssk)2103 static int mptcp_event_created(struct sk_buff *skb,
2104 const struct mptcp_sock *msk,
2105 const struct sock *ssk)
2106 {
2107 int err = nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token);
2108
2109 if (err)
2110 return err;
2111
2112 if (nla_put_u8(skb, MPTCP_ATTR_SERVER_SIDE, READ_ONCE(msk->pm.server_side)))
2113 return -EMSGSIZE;
2114
2115 return mptcp_event_add_subflow(skb, ssk);
2116 }
2117
mptcp_event_addr_removed(const struct mptcp_sock * msk,uint8_t id)2118 void mptcp_event_addr_removed(const struct mptcp_sock *msk, uint8_t id)
2119 {
2120 struct net *net = sock_net((const struct sock *)msk);
2121 struct nlmsghdr *nlh;
2122 struct sk_buff *skb;
2123
2124 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET))
2125 return;
2126
2127 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
2128 if (!skb)
2129 return;
2130
2131 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, MPTCP_EVENT_REMOVED);
2132 if (!nlh)
2133 goto nla_put_failure;
2134
2135 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token))
2136 goto nla_put_failure;
2137
2138 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, id))
2139 goto nla_put_failure;
2140
2141 genlmsg_end(skb, nlh);
2142 mptcp_nl_mcast_send(net, skb, GFP_ATOMIC);
2143 return;
2144
2145 nla_put_failure:
2146 nlmsg_free(skb);
2147 }
2148
mptcp_event_addr_announced(const struct sock * ssk,const struct mptcp_addr_info * info)2149 void mptcp_event_addr_announced(const struct sock *ssk,
2150 const struct mptcp_addr_info *info)
2151 {
2152 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
2153 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
2154 struct net *net = sock_net(ssk);
2155 struct nlmsghdr *nlh;
2156 struct sk_buff *skb;
2157
2158 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET))
2159 return;
2160
2161 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
2162 if (!skb)
2163 return;
2164
2165 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0,
2166 MPTCP_EVENT_ANNOUNCED);
2167 if (!nlh)
2168 goto nla_put_failure;
2169
2170 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token))
2171 goto nla_put_failure;
2172
2173 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, info->id))
2174 goto nla_put_failure;
2175
2176 if (nla_put_be16(skb, MPTCP_ATTR_DPORT,
2177 info->port == 0 ?
2178 inet_sk(ssk)->inet_dport :
2179 info->port))
2180 goto nla_put_failure;
2181
2182 switch (info->family) {
2183 case AF_INET:
2184 if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, info->addr.s_addr))
2185 goto nla_put_failure;
2186 break;
2187 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
2188 case AF_INET6:
2189 if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &info->addr6))
2190 goto nla_put_failure;
2191 break;
2192 #endif
2193 default:
2194 WARN_ON_ONCE(1);
2195 goto nla_put_failure;
2196 }
2197
2198 genlmsg_end(skb, nlh);
2199 mptcp_nl_mcast_send(net, skb, GFP_ATOMIC);
2200 return;
2201
2202 nla_put_failure:
2203 nlmsg_free(skb);
2204 }
2205
mptcp_event_pm_listener(const struct sock * ssk,enum mptcp_event_type event)2206 void mptcp_event_pm_listener(const struct sock *ssk,
2207 enum mptcp_event_type event)
2208 {
2209 const struct inet_sock *issk = inet_sk(ssk);
2210 struct net *net = sock_net(ssk);
2211 struct nlmsghdr *nlh;
2212 struct sk_buff *skb;
2213
2214 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET))
2215 return;
2216
2217 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
2218 if (!skb)
2219 return;
2220
2221 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, event);
2222 if (!nlh)
2223 goto nla_put_failure;
2224
2225 if (nla_put_u16(skb, MPTCP_ATTR_FAMILY, ssk->sk_family))
2226 goto nla_put_failure;
2227
2228 if (nla_put_be16(skb, MPTCP_ATTR_SPORT, issk->inet_sport))
2229 goto nla_put_failure;
2230
2231 switch (ssk->sk_family) {
2232 case AF_INET:
2233 if (nla_put_in_addr(skb, MPTCP_ATTR_SADDR4, issk->inet_saddr))
2234 goto nla_put_failure;
2235 break;
2236 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
2237 case AF_INET6: {
2238 const struct ipv6_pinfo *np = inet6_sk(ssk);
2239
2240 if (nla_put_in6_addr(skb, MPTCP_ATTR_SADDR6, &np->saddr))
2241 goto nla_put_failure;
2242 break;
2243 }
2244 #endif
2245 default:
2246 WARN_ON_ONCE(1);
2247 goto nla_put_failure;
2248 }
2249
2250 genlmsg_end(skb, nlh);
2251 mptcp_nl_mcast_send(net, skb, GFP_KERNEL);
2252 return;
2253
2254 nla_put_failure:
2255 nlmsg_free(skb);
2256 }
2257
mptcp_event(enum mptcp_event_type type,const struct mptcp_sock * msk,const struct sock * ssk,gfp_t gfp)2258 void mptcp_event(enum mptcp_event_type type, const struct mptcp_sock *msk,
2259 const struct sock *ssk, gfp_t gfp)
2260 {
2261 struct net *net = sock_net((const struct sock *)msk);
2262 struct nlmsghdr *nlh;
2263 struct sk_buff *skb;
2264
2265 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET))
2266 return;
2267
2268 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, gfp);
2269 if (!skb)
2270 return;
2271
2272 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, type);
2273 if (!nlh)
2274 goto nla_put_failure;
2275
2276 switch (type) {
2277 case MPTCP_EVENT_UNSPEC:
2278 WARN_ON_ONCE(1);
2279 break;
2280 case MPTCP_EVENT_CREATED:
2281 case MPTCP_EVENT_ESTABLISHED:
2282 if (mptcp_event_created(skb, msk, ssk) < 0)
2283 goto nla_put_failure;
2284 break;
2285 case MPTCP_EVENT_CLOSED:
2286 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token) < 0)
2287 goto nla_put_failure;
2288 break;
2289 case MPTCP_EVENT_ANNOUNCED:
2290 case MPTCP_EVENT_REMOVED:
2291 /* call mptcp_event_addr_announced()/removed instead */
2292 WARN_ON_ONCE(1);
2293 break;
2294 case MPTCP_EVENT_SUB_ESTABLISHED:
2295 case MPTCP_EVENT_SUB_PRIORITY:
2296 if (mptcp_event_sub_established(skb, msk, ssk) < 0)
2297 goto nla_put_failure;
2298 break;
2299 case MPTCP_EVENT_SUB_CLOSED:
2300 if (mptcp_event_sub_closed(skb, msk, ssk) < 0)
2301 goto nla_put_failure;
2302 break;
2303 case MPTCP_EVENT_LISTENER_CREATED:
2304 case MPTCP_EVENT_LISTENER_CLOSED:
2305 break;
2306 }
2307
2308 genlmsg_end(skb, nlh);
2309 mptcp_nl_mcast_send(net, skb, gfp);
2310 return;
2311
2312 nla_put_failure:
2313 nlmsg_free(skb);
2314 }
2315
2316 static const struct genl_small_ops mptcp_pm_ops[] = {
2317 {
2318 .cmd = MPTCP_PM_CMD_ADD_ADDR,
2319 .doit = mptcp_nl_cmd_add_addr,
2320 .flags = GENL_UNS_ADMIN_PERM,
2321 },
2322 {
2323 .cmd = MPTCP_PM_CMD_DEL_ADDR,
2324 .doit = mptcp_nl_cmd_del_addr,
2325 .flags = GENL_UNS_ADMIN_PERM,
2326 },
2327 {
2328 .cmd = MPTCP_PM_CMD_FLUSH_ADDRS,
2329 .doit = mptcp_nl_cmd_flush_addrs,
2330 .flags = GENL_UNS_ADMIN_PERM,
2331 },
2332 {
2333 .cmd = MPTCP_PM_CMD_GET_ADDR,
2334 .doit = mptcp_nl_cmd_get_addr,
2335 .dumpit = mptcp_nl_cmd_dump_addrs,
2336 },
2337 {
2338 .cmd = MPTCP_PM_CMD_SET_LIMITS,
2339 .doit = mptcp_nl_cmd_set_limits,
2340 .flags = GENL_UNS_ADMIN_PERM,
2341 },
2342 {
2343 .cmd = MPTCP_PM_CMD_GET_LIMITS,
2344 .doit = mptcp_nl_cmd_get_limits,
2345 },
2346 {
2347 .cmd = MPTCP_PM_CMD_SET_FLAGS,
2348 .doit = mptcp_nl_cmd_set_flags,
2349 .flags = GENL_UNS_ADMIN_PERM,
2350 },
2351 {
2352 .cmd = MPTCP_PM_CMD_ANNOUNCE,
2353 .doit = mptcp_nl_cmd_announce,
2354 .flags = GENL_UNS_ADMIN_PERM,
2355 },
2356 {
2357 .cmd = MPTCP_PM_CMD_REMOVE,
2358 .doit = mptcp_nl_cmd_remove,
2359 .flags = GENL_UNS_ADMIN_PERM,
2360 },
2361 {
2362 .cmd = MPTCP_PM_CMD_SUBFLOW_CREATE,
2363 .doit = mptcp_nl_cmd_sf_create,
2364 .flags = GENL_UNS_ADMIN_PERM,
2365 },
2366 {
2367 .cmd = MPTCP_PM_CMD_SUBFLOW_DESTROY,
2368 .doit = mptcp_nl_cmd_sf_destroy,
2369 .flags = GENL_UNS_ADMIN_PERM,
2370 },
2371 };
2372
2373 static struct genl_family mptcp_genl_family __ro_after_init = {
2374 .name = MPTCP_PM_NAME,
2375 .version = MPTCP_PM_VER,
2376 .maxattr = MPTCP_PM_ATTR_MAX,
2377 .policy = mptcp_pm_policy,
2378 .netnsok = true,
2379 .module = THIS_MODULE,
2380 .small_ops = mptcp_pm_ops,
2381 .n_small_ops = ARRAY_SIZE(mptcp_pm_ops),
2382 .resv_start_op = MPTCP_PM_CMD_SUBFLOW_DESTROY + 1,
2383 .mcgrps = mptcp_pm_mcgrps,
2384 .n_mcgrps = ARRAY_SIZE(mptcp_pm_mcgrps),
2385 };
2386
pm_nl_init_net(struct net * net)2387 static int __net_init pm_nl_init_net(struct net *net)
2388 {
2389 struct pm_nl_pernet *pernet = pm_nl_get_pernet(net);
2390
2391 INIT_LIST_HEAD_RCU(&pernet->local_addr_list);
2392
2393 /* Cit. 2 subflows ought to be enough for anybody. */
2394 pernet->subflows_max = 2;
2395 pernet->next_id = 1;
2396 pernet->stale_loss_cnt = 4;
2397 spin_lock_init(&pernet->lock);
2398
2399 /* No need to initialize other pernet fields, the struct is zeroed at
2400 * allocation time.
2401 */
2402
2403 return 0;
2404 }
2405
pm_nl_exit_net(struct list_head * net_list)2406 static void __net_exit pm_nl_exit_net(struct list_head *net_list)
2407 {
2408 struct net *net;
2409
2410 list_for_each_entry(net, net_list, exit_list) {
2411 struct pm_nl_pernet *pernet = pm_nl_get_pernet(net);
2412
2413 /* net is removed from namespace list, can't race with
2414 * other modifiers, also netns core already waited for a
2415 * RCU grace period.
2416 */
2417 __flush_addrs(&pernet->local_addr_list);
2418 }
2419 }
2420
2421 static struct pernet_operations mptcp_pm_pernet_ops = {
2422 .init = pm_nl_init_net,
2423 .exit_batch = pm_nl_exit_net,
2424 .id = &pm_nl_pernet_id,
2425 .size = sizeof(struct pm_nl_pernet),
2426 };
2427
mptcp_pm_nl_init(void)2428 void __init mptcp_pm_nl_init(void)
2429 {
2430 if (register_pernet_subsys(&mptcp_pm_pernet_ops) < 0)
2431 panic("Failed to register MPTCP PM pernet subsystem.\n");
2432
2433 if (genl_register_family(&mptcp_genl_family))
2434 panic("Failed to register MPTCP PM netlink family\n");
2435 }
2436