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
139 list_for_each_entry(subflow, list, node) {
140 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
141
142 if (!((1 << inet_sk_state_load(ssk)) &
143 (TCPF_ESTABLISHED | TCPF_SYN_SENT | TCPF_SYN_RECV)))
144 continue;
145
146 remote_address((struct sock_common *)ssk, &cur);
147 if (mptcp_addresses_equal(&cur, daddr, daddr->port))
148 return true;
149 }
150
151 return false;
152 }
153
154 static bool
select_local_address(const struct pm_nl_pernet * pernet,const struct mptcp_sock * msk,struct mptcp_pm_addr_entry * new_entry)155 select_local_address(const struct pm_nl_pernet *pernet,
156 const struct mptcp_sock *msk,
157 struct mptcp_pm_addr_entry *new_entry)
158 {
159 struct mptcp_pm_addr_entry *entry;
160 bool found = false;
161
162 msk_owned_by_me(msk);
163
164 rcu_read_lock();
165 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) {
166 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW))
167 continue;
168
169 if (!test_bit(entry->addr.id, msk->pm.id_avail_bitmap))
170 continue;
171
172 *new_entry = *entry;
173 found = true;
174 break;
175 }
176 rcu_read_unlock();
177
178 return found;
179 }
180
181 static bool
select_signal_address(struct pm_nl_pernet * pernet,const struct mptcp_sock * msk,struct mptcp_pm_addr_entry * new_entry)182 select_signal_address(struct pm_nl_pernet *pernet, const struct mptcp_sock *msk,
183 struct mptcp_pm_addr_entry *new_entry)
184 {
185 struct mptcp_pm_addr_entry *entry;
186 bool found = false;
187
188 rcu_read_lock();
189 /* do not keep any additional per socket state, just signal
190 * the address list in order.
191 * Note: removal from the local address list during the msk life-cycle
192 * can lead to additional addresses not being announced.
193 */
194 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) {
195 if (!test_bit(entry->addr.id, msk->pm.id_avail_bitmap))
196 continue;
197
198 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL))
199 continue;
200
201 *new_entry = *entry;
202 found = true;
203 break;
204 }
205 rcu_read_unlock();
206
207 return found;
208 }
209
mptcp_pm_get_add_addr_signal_max(const struct mptcp_sock * msk)210 unsigned int mptcp_pm_get_add_addr_signal_max(const struct mptcp_sock *msk)
211 {
212 const struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk);
213
214 return READ_ONCE(pernet->add_addr_signal_max);
215 }
216 EXPORT_SYMBOL_GPL(mptcp_pm_get_add_addr_signal_max);
217
mptcp_pm_get_add_addr_accept_max(const struct mptcp_sock * msk)218 unsigned int mptcp_pm_get_add_addr_accept_max(const struct mptcp_sock *msk)
219 {
220 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk);
221
222 return READ_ONCE(pernet->add_addr_accept_max);
223 }
224 EXPORT_SYMBOL_GPL(mptcp_pm_get_add_addr_accept_max);
225
mptcp_pm_get_subflows_max(const struct mptcp_sock * msk)226 unsigned int mptcp_pm_get_subflows_max(const struct mptcp_sock *msk)
227 {
228 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk);
229
230 return READ_ONCE(pernet->subflows_max);
231 }
232 EXPORT_SYMBOL_GPL(mptcp_pm_get_subflows_max);
233
mptcp_pm_get_local_addr_max(const struct mptcp_sock * msk)234 unsigned int mptcp_pm_get_local_addr_max(const struct mptcp_sock *msk)
235 {
236 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk);
237
238 return READ_ONCE(pernet->local_addr_max);
239 }
240 EXPORT_SYMBOL_GPL(mptcp_pm_get_local_addr_max);
241
mptcp_pm_nl_check_work_pending(struct mptcp_sock * msk)242 bool mptcp_pm_nl_check_work_pending(struct mptcp_sock *msk)
243 {
244 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk);
245
246 if (msk->pm.subflows == mptcp_pm_get_subflows_max(msk) ||
247 (find_next_and_bit(pernet->id_bitmap, msk->pm.id_avail_bitmap,
248 MPTCP_PM_MAX_ADDR_ID + 1, 0) == MPTCP_PM_MAX_ADDR_ID + 1)) {
249 WRITE_ONCE(msk->pm.work_pending, false);
250 return false;
251 }
252 return true;
253 }
254
255 struct mptcp_pm_add_entry *
mptcp_lookup_anno_list_by_saddr(const struct mptcp_sock * msk,const struct mptcp_addr_info * addr)256 mptcp_lookup_anno_list_by_saddr(const struct mptcp_sock *msk,
257 const struct mptcp_addr_info *addr)
258 {
259 struct mptcp_pm_add_entry *entry;
260
261 lockdep_assert_held(&msk->pm.lock);
262
263 list_for_each_entry(entry, &msk->pm.anno_list, list) {
264 if (mptcp_addresses_equal(&entry->addr, addr, true))
265 return entry;
266 }
267
268 return NULL;
269 }
270
mptcp_pm_sport_in_anno_list(struct mptcp_sock * msk,const struct sock * sk)271 bool mptcp_pm_sport_in_anno_list(struct mptcp_sock *msk, const struct sock *sk)
272 {
273 struct mptcp_pm_add_entry *entry;
274 struct mptcp_addr_info saddr;
275 bool ret = false;
276
277 mptcp_local_address((struct sock_common *)sk, &saddr);
278
279 spin_lock_bh(&msk->pm.lock);
280 list_for_each_entry(entry, &msk->pm.anno_list, list) {
281 if (mptcp_addresses_equal(&entry->addr, &saddr, true)) {
282 ret = true;
283 goto out;
284 }
285 }
286
287 out:
288 spin_unlock_bh(&msk->pm.lock);
289 return ret;
290 }
291
mptcp_pm_add_timer(struct timer_list * timer)292 static void mptcp_pm_add_timer(struct timer_list *timer)
293 {
294 struct mptcp_pm_add_entry *entry = from_timer(entry, timer, add_timer);
295 struct mptcp_sock *msk = entry->sock;
296 struct sock *sk = (struct sock *)msk;
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 spin_lock_bh(&msk->pm.lock);
315
316 if (!mptcp_pm_should_add_signal_addr(msk)) {
317 pr_debug("retransmit ADD_ADDR id=%d\n", entry->addr.id);
318 mptcp_pm_announce_addr(msk, &entry->addr, false);
319 mptcp_pm_add_addr_send_ack(msk);
320 entry->retrans_times++;
321 }
322
323 if (entry->retrans_times < ADD_ADDR_RETRANS_MAX)
324 sk_reset_timer(sk, timer,
325 jiffies + mptcp_get_add_addr_timeout(sock_net(sk)));
326
327 spin_unlock_bh(&msk->pm.lock);
328
329 if (entry->retrans_times == ADD_ADDR_RETRANS_MAX)
330 mptcp_pm_subflow_established(msk);
331
332 out:
333 __sock_put(sk);
334 }
335
336 struct mptcp_pm_add_entry *
mptcp_pm_del_add_timer(struct mptcp_sock * msk,const struct mptcp_addr_info * addr,bool check_id)337 mptcp_pm_del_add_timer(struct mptcp_sock *msk,
338 const struct mptcp_addr_info *addr, bool check_id)
339 {
340 struct mptcp_pm_add_entry *entry;
341 struct sock *sk = (struct sock *)msk;
342 struct timer_list *add_timer = NULL;
343
344 spin_lock_bh(&msk->pm.lock);
345 entry = mptcp_lookup_anno_list_by_saddr(msk, addr);
346 if (entry && (!check_id || entry->addr.id == addr->id)) {
347 entry->retrans_times = ADD_ADDR_RETRANS_MAX;
348 add_timer = &entry->add_timer;
349 }
350 if (!check_id && entry)
351 list_del(&entry->list);
352 spin_unlock_bh(&msk->pm.lock);
353
354 /* no lock, because sk_stop_timer_sync() is calling del_timer_sync() */
355 if (add_timer)
356 sk_stop_timer_sync(sk, add_timer);
357
358 return entry;
359 }
360
mptcp_pm_alloc_anno_list(struct mptcp_sock * msk,const struct mptcp_addr_info * addr)361 bool mptcp_pm_alloc_anno_list(struct mptcp_sock *msk,
362 const struct mptcp_addr_info *addr)
363 {
364 struct mptcp_pm_add_entry *add_entry = NULL;
365 struct sock *sk = (struct sock *)msk;
366 struct net *net = sock_net(sk);
367
368 lockdep_assert_held(&msk->pm.lock);
369
370 add_entry = mptcp_lookup_anno_list_by_saddr(msk, addr);
371
372 if (add_entry) {
373 if (WARN_ON_ONCE(mptcp_pm_is_kernel(msk)))
374 return false;
375
376 sk_reset_timer(sk, &add_entry->add_timer,
377 jiffies + mptcp_get_add_addr_timeout(net));
378 return true;
379 }
380
381 add_entry = kmalloc(sizeof(*add_entry), GFP_ATOMIC);
382 if (!add_entry)
383 return false;
384
385 list_add(&add_entry->list, &msk->pm.anno_list);
386
387 add_entry->addr = *addr;
388 add_entry->sock = msk;
389 add_entry->retrans_times = 0;
390
391 timer_setup(&add_entry->add_timer, mptcp_pm_add_timer, 0);
392 sk_reset_timer(sk, &add_entry->add_timer,
393 jiffies + mptcp_get_add_addr_timeout(net));
394
395 return true;
396 }
397
mptcp_pm_free_anno_list(struct mptcp_sock * msk)398 void mptcp_pm_free_anno_list(struct mptcp_sock *msk)
399 {
400 struct mptcp_pm_add_entry *entry, *tmp;
401 struct sock *sk = (struct sock *)msk;
402 LIST_HEAD(free_list);
403
404 pr_debug("msk=%p\n", msk);
405
406 spin_lock_bh(&msk->pm.lock);
407 list_splice_init(&msk->pm.anno_list, &free_list);
408 spin_unlock_bh(&msk->pm.lock);
409
410 list_for_each_entry_safe(entry, tmp, &free_list, list) {
411 sk_stop_timer_sync(sk, &entry->add_timer);
412 kfree(entry);
413 }
414 }
415
416 /* Fill all the remote addresses into the array addrs[],
417 * and return the array size.
418 */
fill_remote_addresses_vec(struct mptcp_sock * msk,struct mptcp_addr_info * local,bool fullmesh,struct mptcp_addr_info * addrs)419 static unsigned int fill_remote_addresses_vec(struct mptcp_sock *msk,
420 struct mptcp_addr_info *local,
421 bool fullmesh,
422 struct mptcp_addr_info *addrs)
423 {
424 bool deny_id0 = READ_ONCE(msk->pm.remote_deny_join_id0);
425 struct sock *sk = (struct sock *)msk, *ssk;
426 struct mptcp_subflow_context *subflow;
427 struct mptcp_addr_info remote = { 0 };
428 unsigned int subflows_max;
429 int i = 0;
430
431 subflows_max = mptcp_pm_get_subflows_max(msk);
432 remote_address((struct sock_common *)sk, &remote);
433
434 /* Non-fullmesh endpoint, fill in the single entry
435 * corresponding to the primary MPC subflow remote address
436 */
437 if (!fullmesh) {
438 if (deny_id0)
439 return 0;
440
441 if (!mptcp_pm_addr_families_match(sk, local, &remote))
442 return 0;
443
444 msk->pm.subflows++;
445 addrs[i++] = remote;
446 } else {
447 DECLARE_BITMAP(unavail_id, MPTCP_PM_MAX_ADDR_ID + 1);
448
449 /* Forbid creation of new subflows matching existing
450 * ones, possibly already created by incoming ADD_ADDR
451 */
452 bitmap_zero(unavail_id, MPTCP_PM_MAX_ADDR_ID + 1);
453 mptcp_for_each_subflow(msk, subflow)
454 if (READ_ONCE(subflow->local_id) == local->id)
455 __set_bit(subflow->remote_id, unavail_id);
456
457 mptcp_for_each_subflow(msk, subflow) {
458 ssk = mptcp_subflow_tcp_sock(subflow);
459 remote_address((struct sock_common *)ssk, &addrs[i]);
460 addrs[i].id = READ_ONCE(subflow->remote_id);
461 if (deny_id0 && !addrs[i].id)
462 continue;
463
464 if (test_bit(addrs[i].id, unavail_id))
465 continue;
466
467 if (!mptcp_pm_addr_families_match(sk, local, &addrs[i]))
468 continue;
469
470 if (msk->pm.subflows < subflows_max) {
471 /* forbid creating multiple address towards
472 * this id
473 */
474 __set_bit(addrs[i].id, unavail_id);
475 msk->pm.subflows++;
476 i++;
477 }
478 }
479 }
480
481 return i;
482 }
483
__mptcp_pm_send_ack(struct mptcp_sock * msk,struct mptcp_subflow_context * subflow,bool prio,bool backup)484 static void __mptcp_pm_send_ack(struct mptcp_sock *msk, struct mptcp_subflow_context *subflow,
485 bool prio, bool backup)
486 {
487 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
488 bool slow;
489
490 pr_debug("send ack for %s\n",
491 prio ? "mp_prio" : (mptcp_pm_should_add_signal(msk) ? "add_addr" : "rm_addr"));
492
493 slow = lock_sock_fast(ssk);
494 if (prio) {
495 subflow->send_mp_prio = 1;
496 subflow->request_bkup = backup;
497 }
498
499 __mptcp_subflow_send_ack(ssk);
500 unlock_sock_fast(ssk, slow);
501 }
502
mptcp_pm_send_ack(struct mptcp_sock * msk,struct mptcp_subflow_context * subflow,bool prio,bool backup)503 static void mptcp_pm_send_ack(struct mptcp_sock *msk, struct mptcp_subflow_context *subflow,
504 bool prio, bool backup)
505 {
506 spin_unlock_bh(&msk->pm.lock);
507 __mptcp_pm_send_ack(msk, subflow, prio, backup);
508 spin_lock_bh(&msk->pm.lock);
509 }
510
511 static struct mptcp_pm_addr_entry *
__lookup_addr_by_id(struct pm_nl_pernet * pernet,unsigned int id)512 __lookup_addr_by_id(struct pm_nl_pernet *pernet, unsigned int id)
513 {
514 struct mptcp_pm_addr_entry *entry;
515
516 list_for_each_entry(entry, &pernet->local_addr_list, list) {
517 if (entry->addr.id == id)
518 return entry;
519 }
520 return NULL;
521 }
522
523 static struct mptcp_pm_addr_entry *
__lookup_addr(struct pm_nl_pernet * pernet,const struct mptcp_addr_info * info,bool lookup_by_id)524 __lookup_addr(struct pm_nl_pernet *pernet, const struct mptcp_addr_info *info,
525 bool lookup_by_id)
526 {
527 struct mptcp_pm_addr_entry *entry;
528
529 list_for_each_entry(entry, &pernet->local_addr_list, list) {
530 if ((!lookup_by_id &&
531 mptcp_addresses_equal(&entry->addr, info, entry->addr.port)) ||
532 (lookup_by_id && entry->addr.id == info->id))
533 return entry;
534 }
535 return NULL;
536 }
537
mptcp_pm_create_subflow_or_signal_addr(struct mptcp_sock * msk)538 static void mptcp_pm_create_subflow_or_signal_addr(struct mptcp_sock *msk)
539 {
540 struct sock *sk = (struct sock *)msk;
541 struct mptcp_pm_addr_entry local;
542 unsigned int add_addr_signal_max;
543 bool signal_and_subflow = false;
544 unsigned int local_addr_max;
545 struct pm_nl_pernet *pernet;
546 unsigned int subflows_max;
547
548 pernet = pm_nl_get_pernet(sock_net(sk));
549
550 add_addr_signal_max = mptcp_pm_get_add_addr_signal_max(msk);
551 local_addr_max = mptcp_pm_get_local_addr_max(msk);
552 subflows_max = mptcp_pm_get_subflows_max(msk);
553
554 /* do lazy endpoint usage accounting for the MPC subflows */
555 if (unlikely(!(msk->pm.status & BIT(MPTCP_PM_MPC_ENDPOINT_ACCOUNTED))) && msk->first) {
556 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(msk->first);
557 struct mptcp_pm_addr_entry *entry;
558 struct mptcp_addr_info mpc_addr;
559 bool backup = false;
560
561 mptcp_local_address((struct sock_common *)msk->first, &mpc_addr);
562 rcu_read_lock();
563 entry = __lookup_addr(pernet, &mpc_addr, false);
564 if (entry) {
565 __clear_bit(entry->addr.id, msk->pm.id_avail_bitmap);
566 msk->mpc_endpoint_id = entry->addr.id;
567 backup = !!(entry->flags & MPTCP_PM_ADDR_FLAG_BACKUP);
568 }
569 rcu_read_unlock();
570
571 if (backup)
572 mptcp_pm_send_ack(msk, subflow, true, backup);
573
574 msk->pm.status |= BIT(MPTCP_PM_MPC_ENDPOINT_ACCOUNTED);
575 }
576
577 pr_debug("local %d:%d signal %d:%d subflows %d:%d\n",
578 msk->pm.local_addr_used, local_addr_max,
579 msk->pm.add_addr_signaled, add_addr_signal_max,
580 msk->pm.subflows, subflows_max);
581
582 /* check first for announce */
583 if (msk->pm.add_addr_signaled < add_addr_signal_max) {
584 /* due to racing events on both ends we can reach here while
585 * previous add address is still running: if we invoke now
586 * mptcp_pm_announce_addr(), that will fail and the
587 * corresponding id will be marked as used.
588 * Instead let the PM machinery reschedule us when the
589 * current address announce will be completed.
590 */
591 if (msk->pm.addr_signal & BIT(MPTCP_ADD_ADDR_SIGNAL))
592 return;
593
594 if (!select_signal_address(pernet, msk, &local))
595 goto subflow;
596
597 /* If the alloc fails, we are on memory pressure, not worth
598 * continuing, and trying to create subflows.
599 */
600 if (!mptcp_pm_alloc_anno_list(msk, &local.addr))
601 return;
602
603 __clear_bit(local.addr.id, msk->pm.id_avail_bitmap);
604 msk->pm.add_addr_signaled++;
605
606 /* Special case for ID0: set the correct ID */
607 if (local.addr.id == msk->mpc_endpoint_id)
608 local.addr.id = 0;
609
610 mptcp_pm_announce_addr(msk, &local.addr, false);
611 mptcp_pm_nl_addr_send_ack(msk);
612
613 if (local.flags & MPTCP_PM_ADDR_FLAG_SUBFLOW)
614 signal_and_subflow = true;
615 }
616
617 subflow:
618 /* check if should create a new subflow */
619 while (msk->pm.local_addr_used < local_addr_max &&
620 msk->pm.subflows < subflows_max) {
621 struct mptcp_addr_info addrs[MPTCP_PM_ADDR_MAX];
622 bool fullmesh;
623 int i, nr;
624
625 if (signal_and_subflow)
626 signal_and_subflow = false;
627 else if (!select_local_address(pernet, msk, &local))
628 break;
629
630 fullmesh = !!(local.flags & MPTCP_PM_ADDR_FLAG_FULLMESH);
631
632 __clear_bit(local.addr.id, msk->pm.id_avail_bitmap);
633
634 /* Special case for ID0: set the correct ID */
635 if (local.addr.id == msk->mpc_endpoint_id)
636 local.addr.id = 0;
637 else /* local_addr_used is not decr for ID 0 */
638 msk->pm.local_addr_used++;
639
640 nr = fill_remote_addresses_vec(msk, &local.addr, fullmesh, addrs);
641 if (nr == 0)
642 continue;
643
644 spin_unlock_bh(&msk->pm.lock);
645 for (i = 0; i < nr; i++)
646 __mptcp_subflow_connect(sk, &local.addr, &addrs[i]);
647 spin_lock_bh(&msk->pm.lock);
648 }
649 mptcp_pm_nl_check_work_pending(msk);
650 }
651
mptcp_pm_nl_fully_established(struct mptcp_sock * msk)652 static void mptcp_pm_nl_fully_established(struct mptcp_sock *msk)
653 {
654 mptcp_pm_create_subflow_or_signal_addr(msk);
655 }
656
mptcp_pm_nl_subflow_established(struct mptcp_sock * msk)657 static void mptcp_pm_nl_subflow_established(struct mptcp_sock *msk)
658 {
659 mptcp_pm_create_subflow_or_signal_addr(msk);
660 }
661
662 /* Fill all the local addresses into the array addrs[],
663 * and return the array size.
664 */
fill_local_addresses_vec(struct mptcp_sock * msk,struct mptcp_addr_info * remote,struct mptcp_addr_info * addrs)665 static unsigned int fill_local_addresses_vec(struct mptcp_sock *msk,
666 struct mptcp_addr_info *remote,
667 struct mptcp_addr_info *addrs)
668 {
669 struct sock *sk = (struct sock *)msk;
670 struct mptcp_pm_addr_entry *entry;
671 struct mptcp_addr_info mpc_addr;
672 struct pm_nl_pernet *pernet;
673 unsigned int subflows_max;
674 int i = 0;
675
676 pernet = pm_nl_get_pernet_from_msk(msk);
677 subflows_max = mptcp_pm_get_subflows_max(msk);
678
679 mptcp_local_address((struct sock_common *)msk, &mpc_addr);
680
681 rcu_read_lock();
682 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) {
683 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_FULLMESH))
684 continue;
685
686 if (!mptcp_pm_addr_families_match(sk, &entry->addr, remote))
687 continue;
688
689 if (msk->pm.subflows < subflows_max) {
690 msk->pm.subflows++;
691 addrs[i] = entry->addr;
692
693 /* Special case for ID0: set the correct ID */
694 if (mptcp_addresses_equal(&entry->addr, &mpc_addr, entry->addr.port))
695 addrs[i].id = 0;
696
697 i++;
698 }
699 }
700 rcu_read_unlock();
701
702 /* If the array is empty, fill in the single
703 * 'IPADDRANY' local address
704 */
705 if (!i) {
706 struct mptcp_addr_info local;
707
708 memset(&local, 0, sizeof(local));
709 local.family =
710 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
711 remote->family == AF_INET6 &&
712 ipv6_addr_v4mapped(&remote->addr6) ? AF_INET :
713 #endif
714 remote->family;
715
716 if (!mptcp_pm_addr_families_match(sk, &local, remote))
717 return 0;
718
719 msk->pm.subflows++;
720 addrs[i++] = local;
721 }
722
723 return i;
724 }
725
mptcp_pm_nl_add_addr_received(struct mptcp_sock * msk)726 static void mptcp_pm_nl_add_addr_received(struct mptcp_sock *msk)
727 {
728 struct mptcp_addr_info addrs[MPTCP_PM_ADDR_MAX];
729 struct sock *sk = (struct sock *)msk;
730 unsigned int add_addr_accept_max;
731 struct mptcp_addr_info remote;
732 unsigned int subflows_max;
733 bool sf_created = false;
734 int i, nr;
735
736 add_addr_accept_max = mptcp_pm_get_add_addr_accept_max(msk);
737 subflows_max = mptcp_pm_get_subflows_max(msk);
738
739 pr_debug("accepted %d:%d remote family %d\n",
740 msk->pm.add_addr_accepted, add_addr_accept_max,
741 msk->pm.remote.family);
742
743 remote = msk->pm.remote;
744 mptcp_pm_announce_addr(msk, &remote, true);
745 mptcp_pm_nl_addr_send_ack(msk);
746
747 if (lookup_subflow_by_daddr(&msk->conn_list, &remote))
748 return;
749
750 /* pick id 0 port, if none is provided the remote address */
751 if (!remote.port)
752 remote.port = sk->sk_dport;
753
754 /* connect to the specified remote address, using whatever
755 * local address the routing configuration will pick.
756 */
757 nr = fill_local_addresses_vec(msk, &remote, addrs);
758 if (nr == 0)
759 return;
760
761 spin_unlock_bh(&msk->pm.lock);
762 for (i = 0; i < nr; i++)
763 if (__mptcp_subflow_connect(sk, &addrs[i], &remote) == 0)
764 sf_created = true;
765 spin_lock_bh(&msk->pm.lock);
766
767 if (sf_created) {
768 /* add_addr_accepted is not decr for ID 0 */
769 if (remote.id)
770 msk->pm.add_addr_accepted++;
771 if (msk->pm.add_addr_accepted >= add_addr_accept_max ||
772 msk->pm.subflows >= subflows_max)
773 WRITE_ONCE(msk->pm.accept_addr, false);
774 }
775 }
776
mptcp_pm_nl_is_init_remote_addr(struct mptcp_sock * msk,const struct mptcp_addr_info * remote)777 bool mptcp_pm_nl_is_init_remote_addr(struct mptcp_sock *msk,
778 const struct mptcp_addr_info *remote)
779 {
780 struct mptcp_addr_info mpc_remote;
781
782 remote_address((struct sock_common *)msk, &mpc_remote);
783 return mptcp_addresses_equal(&mpc_remote, remote, remote->port);
784 }
785
mptcp_pm_nl_addr_send_ack(struct mptcp_sock * msk)786 void mptcp_pm_nl_addr_send_ack(struct mptcp_sock *msk)
787 {
788 struct mptcp_subflow_context *subflow;
789
790 msk_owned_by_me(msk);
791 lockdep_assert_held(&msk->pm.lock);
792
793 if (!mptcp_pm_should_add_signal(msk) &&
794 !mptcp_pm_should_rm_signal(msk))
795 return;
796
797 mptcp_for_each_subflow(msk, subflow) {
798 if (__mptcp_subflow_active(subflow)) {
799 mptcp_pm_send_ack(msk, subflow, false, false);
800 break;
801 }
802 }
803 }
804
mptcp_pm_nl_mp_prio_send_ack(struct mptcp_sock * msk,struct mptcp_addr_info * addr,struct mptcp_addr_info * rem,u8 bkup)805 int mptcp_pm_nl_mp_prio_send_ack(struct mptcp_sock *msk,
806 struct mptcp_addr_info *addr,
807 struct mptcp_addr_info *rem,
808 u8 bkup)
809 {
810 struct mptcp_subflow_context *subflow;
811
812 pr_debug("bkup=%d\n", bkup);
813
814 mptcp_for_each_subflow(msk, subflow) {
815 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
816 struct mptcp_addr_info local, remote;
817
818 mptcp_local_address((struct sock_common *)ssk, &local);
819 if (!mptcp_addresses_equal(&local, addr, addr->port))
820 continue;
821
822 if (rem && rem->family != AF_UNSPEC) {
823 remote_address((struct sock_common *)ssk, &remote);
824 if (!mptcp_addresses_equal(&remote, rem, rem->port))
825 continue;
826 }
827
828 __mptcp_pm_send_ack(msk, subflow, true, bkup);
829 return 0;
830 }
831
832 return -EINVAL;
833 }
834
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)835 static void mptcp_pm_nl_rm_addr_or_subflow(struct mptcp_sock *msk,
836 const struct mptcp_rm_list *rm_list,
837 enum linux_mptcp_mib_field rm_type)
838 {
839 struct mptcp_subflow_context *subflow, *tmp;
840 struct sock *sk = (struct sock *)msk;
841 u8 i;
842
843 pr_debug("%s rm_list_nr %d\n",
844 rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow", rm_list->nr);
845
846 msk_owned_by_me(msk);
847
848 if (sk->sk_state == TCP_LISTEN)
849 return;
850
851 if (!rm_list->nr)
852 return;
853
854 if (list_empty(&msk->conn_list))
855 return;
856
857 for (i = 0; i < rm_list->nr; i++) {
858 u8 rm_id = rm_list->ids[i];
859 bool removed = false;
860
861 mptcp_for_each_subflow_safe(msk, subflow, tmp) {
862 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
863 u8 remote_id = READ_ONCE(subflow->remote_id);
864 int how = RCV_SHUTDOWN | SEND_SHUTDOWN;
865 u8 id = subflow_get_local_id(subflow);
866
867 if ((1 << inet_sk_state_load(ssk)) &
868 (TCPF_FIN_WAIT1 | TCPF_FIN_WAIT2 | TCPF_CLOSING | TCPF_CLOSE))
869 continue;
870 if (rm_type == MPTCP_MIB_RMADDR && remote_id != rm_id)
871 continue;
872 if (rm_type == MPTCP_MIB_RMSUBFLOW && id != rm_id)
873 continue;
874
875 pr_debug(" -> %s rm_list_ids[%d]=%u local_id=%u remote_id=%u mpc_id=%u\n",
876 rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow",
877 i, rm_id, id, remote_id, msk->mpc_endpoint_id);
878 spin_unlock_bh(&msk->pm.lock);
879 mptcp_subflow_shutdown(sk, ssk, how);
880 removed |= subflow->request_join;
881
882 /* the following takes care of updating the subflows counter */
883 mptcp_close_ssk(sk, ssk, subflow);
884 spin_lock_bh(&msk->pm.lock);
885
886 if (rm_type == MPTCP_MIB_RMSUBFLOW)
887 __MPTCP_INC_STATS(sock_net(sk), rm_type);
888 }
889
890 if (rm_type == MPTCP_MIB_RMADDR)
891 __MPTCP_INC_STATS(sock_net(sk), rm_type);
892
893 if (!removed)
894 continue;
895
896 if (!mptcp_pm_is_kernel(msk))
897 continue;
898
899 if (rm_type == MPTCP_MIB_RMADDR && rm_id &&
900 !WARN_ON_ONCE(msk->pm.add_addr_accepted == 0)) {
901 /* Note: if the subflow has been closed before, this
902 * add_addr_accepted counter will not be decremented.
903 */
904 if (--msk->pm.add_addr_accepted < mptcp_pm_get_add_addr_accept_max(msk))
905 WRITE_ONCE(msk->pm.accept_addr, true);
906 }
907 }
908 }
909
mptcp_pm_nl_rm_addr_received(struct mptcp_sock * msk)910 static void mptcp_pm_nl_rm_addr_received(struct mptcp_sock *msk)
911 {
912 mptcp_pm_nl_rm_addr_or_subflow(msk, &msk->pm.rm_list_rx, MPTCP_MIB_RMADDR);
913 }
914
mptcp_pm_nl_rm_subflow_received(struct mptcp_sock * msk,const struct mptcp_rm_list * rm_list)915 static void mptcp_pm_nl_rm_subflow_received(struct mptcp_sock *msk,
916 const struct mptcp_rm_list *rm_list)
917 {
918 mptcp_pm_nl_rm_addr_or_subflow(msk, rm_list, MPTCP_MIB_RMSUBFLOW);
919 }
920
mptcp_pm_nl_work(struct mptcp_sock * msk)921 void mptcp_pm_nl_work(struct mptcp_sock *msk)
922 {
923 struct mptcp_pm_data *pm = &msk->pm;
924
925 msk_owned_by_me(msk);
926
927 if (!(pm->status & MPTCP_PM_WORK_MASK))
928 return;
929
930 spin_lock_bh(&msk->pm.lock);
931
932 pr_debug("msk=%p status=%x\n", msk, pm->status);
933 if (pm->status & BIT(MPTCP_PM_ADD_ADDR_RECEIVED)) {
934 pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_RECEIVED);
935 mptcp_pm_nl_add_addr_received(msk);
936 }
937 if (pm->status & BIT(MPTCP_PM_ADD_ADDR_SEND_ACK)) {
938 pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_SEND_ACK);
939 mptcp_pm_nl_addr_send_ack(msk);
940 }
941 if (pm->status & BIT(MPTCP_PM_RM_ADDR_RECEIVED)) {
942 pm->status &= ~BIT(MPTCP_PM_RM_ADDR_RECEIVED);
943 mptcp_pm_nl_rm_addr_received(msk);
944 }
945 if (pm->status & BIT(MPTCP_PM_ESTABLISHED)) {
946 pm->status &= ~BIT(MPTCP_PM_ESTABLISHED);
947 mptcp_pm_nl_fully_established(msk);
948 }
949 if (pm->status & BIT(MPTCP_PM_SUBFLOW_ESTABLISHED)) {
950 pm->status &= ~BIT(MPTCP_PM_SUBFLOW_ESTABLISHED);
951 mptcp_pm_nl_subflow_established(msk);
952 }
953
954 spin_unlock_bh(&msk->pm.lock);
955 }
956
address_use_port(struct mptcp_pm_addr_entry * entry)957 static bool address_use_port(struct mptcp_pm_addr_entry *entry)
958 {
959 return (entry->flags &
960 (MPTCP_PM_ADDR_FLAG_SIGNAL | MPTCP_PM_ADDR_FLAG_SUBFLOW)) ==
961 MPTCP_PM_ADDR_FLAG_SIGNAL;
962 }
963
964 /* caller must ensure the RCU grace period is already elapsed */
__mptcp_pm_release_addr_entry(struct mptcp_pm_addr_entry * entry)965 static void __mptcp_pm_release_addr_entry(struct mptcp_pm_addr_entry *entry)
966 {
967 if (entry->lsk)
968 sock_release(entry->lsk);
969 kfree(entry);
970 }
971
mptcp_pm_nl_append_new_local_addr(struct pm_nl_pernet * pernet,struct mptcp_pm_addr_entry * entry,bool needs_id)972 static int mptcp_pm_nl_append_new_local_addr(struct pm_nl_pernet *pernet,
973 struct mptcp_pm_addr_entry *entry,
974 bool needs_id)
975 {
976 struct mptcp_pm_addr_entry *cur, *del_entry = NULL;
977 unsigned int addr_max;
978 int ret = -EINVAL;
979
980 spin_lock_bh(&pernet->lock);
981 /* to keep the code simple, don't do IDR-like allocation for address ID,
982 * just bail when we exceed limits
983 */
984 if (pernet->next_id == MPTCP_PM_MAX_ADDR_ID)
985 pernet->next_id = 1;
986 if (pernet->addrs >= MPTCP_PM_ADDR_MAX) {
987 ret = -ERANGE;
988 goto out;
989 }
990 if (test_bit(entry->addr.id, pernet->id_bitmap)) {
991 ret = -EBUSY;
992 goto out;
993 }
994
995 /* do not insert duplicate address, differentiate on port only
996 * singled addresses
997 */
998 if (!address_use_port(entry))
999 entry->addr.port = 0;
1000 list_for_each_entry(cur, &pernet->local_addr_list, list) {
1001 if (mptcp_addresses_equal(&cur->addr, &entry->addr,
1002 cur->addr.port || entry->addr.port)) {
1003 /* allow replacing the exiting endpoint only if such
1004 * endpoint is an implicit one and the user-space
1005 * did not provide an endpoint id
1006 */
1007 if (!(cur->flags & MPTCP_PM_ADDR_FLAG_IMPLICIT)) {
1008 ret = -EEXIST;
1009 goto out;
1010 }
1011 if (entry->addr.id)
1012 goto out;
1013
1014 pernet->addrs--;
1015 entry->addr.id = cur->addr.id;
1016 list_del_rcu(&cur->list);
1017 del_entry = cur;
1018 break;
1019 }
1020 }
1021
1022 if (!entry->addr.id && needs_id) {
1023 find_next:
1024 entry->addr.id = find_next_zero_bit(pernet->id_bitmap,
1025 MPTCP_PM_MAX_ADDR_ID + 1,
1026 pernet->next_id);
1027 if (!entry->addr.id && pernet->next_id != 1) {
1028 pernet->next_id = 1;
1029 goto find_next;
1030 }
1031 }
1032
1033 if (!entry->addr.id && needs_id)
1034 goto out;
1035
1036 __set_bit(entry->addr.id, pernet->id_bitmap);
1037 if (entry->addr.id > pernet->next_id)
1038 pernet->next_id = entry->addr.id;
1039
1040 if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) {
1041 addr_max = pernet->add_addr_signal_max;
1042 WRITE_ONCE(pernet->add_addr_signal_max, addr_max + 1);
1043 }
1044 if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) {
1045 addr_max = pernet->local_addr_max;
1046 WRITE_ONCE(pernet->local_addr_max, addr_max + 1);
1047 }
1048
1049 pernet->addrs++;
1050 if (!entry->addr.port)
1051 list_add_tail_rcu(&entry->list, &pernet->local_addr_list);
1052 else
1053 list_add_rcu(&entry->list, &pernet->local_addr_list);
1054 ret = entry->addr.id;
1055
1056 out:
1057 spin_unlock_bh(&pernet->lock);
1058
1059 /* just replaced an existing entry, free it */
1060 if (del_entry) {
1061 synchronize_rcu();
1062 __mptcp_pm_release_addr_entry(del_entry);
1063 }
1064 return ret;
1065 }
1066
1067 static struct lock_class_key mptcp_slock_keys[2];
1068 static struct lock_class_key mptcp_keys[2];
1069
mptcp_pm_nl_create_listen_socket(struct sock * sk,struct mptcp_pm_addr_entry * entry)1070 static int mptcp_pm_nl_create_listen_socket(struct sock *sk,
1071 struct mptcp_pm_addr_entry *entry)
1072 {
1073 bool is_ipv6 = sk->sk_family == AF_INET6;
1074 int addrlen = sizeof(struct sockaddr_in);
1075 struct sockaddr_storage addr;
1076 struct sock *newsk, *ssk;
1077 int backlog = 1024;
1078 int err;
1079
1080 err = sock_create_kern(sock_net(sk), entry->addr.family,
1081 SOCK_STREAM, IPPROTO_MPTCP, &entry->lsk);
1082 if (err)
1083 return err;
1084
1085 newsk = entry->lsk->sk;
1086 if (!newsk)
1087 return -EINVAL;
1088
1089 /* The subflow socket lock is acquired in a nested to the msk one
1090 * in several places, even by the TCP stack, and this msk is a kernel
1091 * socket: lockdep complains. Instead of propagating the _nested
1092 * modifiers in several places, re-init the lock class for the msk
1093 * socket to an mptcp specific one.
1094 */
1095 sock_lock_init_class_and_name(newsk,
1096 is_ipv6 ? "mlock-AF_INET6" : "mlock-AF_INET",
1097 &mptcp_slock_keys[is_ipv6],
1098 is_ipv6 ? "msk_lock-AF_INET6" : "msk_lock-AF_INET",
1099 &mptcp_keys[is_ipv6]);
1100
1101 lock_sock(newsk);
1102 ssk = __mptcp_nmpc_sk(mptcp_sk(newsk));
1103 release_sock(newsk);
1104 if (IS_ERR(ssk))
1105 return PTR_ERR(ssk);
1106
1107 mptcp_info2sockaddr(&entry->addr, &addr, entry->addr.family);
1108 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1109 if (entry->addr.family == AF_INET6)
1110 addrlen = sizeof(struct sockaddr_in6);
1111 #endif
1112 if (ssk->sk_family == AF_INET)
1113 err = inet_bind_sk(ssk, (struct sockaddr *)&addr, addrlen);
1114 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1115 else if (ssk->sk_family == AF_INET6)
1116 err = inet6_bind_sk(ssk, (struct sockaddr *)&addr, addrlen);
1117 #endif
1118 if (err)
1119 return err;
1120
1121 /* We don't use mptcp_set_state() here because it needs to be called
1122 * under the msk socket lock. For the moment, that will not bring
1123 * anything more than only calling inet_sk_state_store(), because the
1124 * old status is known (TCP_CLOSE).
1125 */
1126 inet_sk_state_store(newsk, TCP_LISTEN);
1127 lock_sock(ssk);
1128 WRITE_ONCE(mptcp_subflow_ctx(ssk)->pm_listener, true);
1129 err = __inet_listen_sk(ssk, backlog);
1130 if (!err)
1131 mptcp_event_pm_listener(ssk, MPTCP_EVENT_LISTENER_CREATED);
1132 release_sock(ssk);
1133 return err;
1134 }
1135
mptcp_pm_nl_get_local_id(struct mptcp_sock * msk,struct mptcp_addr_info * skc)1136 int mptcp_pm_nl_get_local_id(struct mptcp_sock *msk, struct mptcp_addr_info *skc)
1137 {
1138 struct mptcp_pm_addr_entry *entry;
1139 struct pm_nl_pernet *pernet;
1140 int ret = -1;
1141
1142 pernet = pm_nl_get_pernet_from_msk(msk);
1143
1144 rcu_read_lock();
1145 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) {
1146 if (mptcp_addresses_equal(&entry->addr, skc, entry->addr.port)) {
1147 ret = entry->addr.id;
1148 break;
1149 }
1150 }
1151 rcu_read_unlock();
1152 if (ret >= 0)
1153 return ret;
1154
1155 /* address not found, add to local list */
1156 entry = kmalloc(sizeof(*entry), GFP_ATOMIC);
1157 if (!entry)
1158 return -ENOMEM;
1159
1160 entry->addr = *skc;
1161 entry->addr.id = 0;
1162 entry->addr.port = 0;
1163 entry->ifindex = 0;
1164 entry->flags = MPTCP_PM_ADDR_FLAG_IMPLICIT;
1165 entry->lsk = NULL;
1166 ret = mptcp_pm_nl_append_new_local_addr(pernet, entry, true);
1167 if (ret < 0)
1168 kfree(entry);
1169
1170 return ret;
1171 }
1172
mptcp_pm_nl_is_backup(struct mptcp_sock * msk,struct mptcp_addr_info * skc)1173 bool mptcp_pm_nl_is_backup(struct mptcp_sock *msk, struct mptcp_addr_info *skc)
1174 {
1175 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk);
1176 struct mptcp_pm_addr_entry *entry;
1177 bool backup = false;
1178
1179 rcu_read_lock();
1180 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) {
1181 if (mptcp_addresses_equal(&entry->addr, skc, entry->addr.port)) {
1182 backup = !!(entry->flags & MPTCP_PM_ADDR_FLAG_BACKUP);
1183 break;
1184 }
1185 }
1186 rcu_read_unlock();
1187
1188 return backup;
1189 }
1190
1191 #define MPTCP_PM_CMD_GRP_OFFSET 0
1192 #define MPTCP_PM_EV_GRP_OFFSET 1
1193
1194 static const struct genl_multicast_group mptcp_pm_mcgrps[] = {
1195 [MPTCP_PM_CMD_GRP_OFFSET] = { .name = MPTCP_PM_CMD_GRP_NAME, },
1196 [MPTCP_PM_EV_GRP_OFFSET] = { .name = MPTCP_PM_EV_GRP_NAME,
1197 .flags = GENL_UNS_ADMIN_PERM,
1198 },
1199 };
1200
1201 static const struct nla_policy
1202 mptcp_pm_addr_policy[MPTCP_PM_ADDR_ATTR_MAX + 1] = {
1203 [MPTCP_PM_ADDR_ATTR_FAMILY] = { .type = NLA_U16, },
1204 [MPTCP_PM_ADDR_ATTR_ID] = { .type = NLA_U8, },
1205 [MPTCP_PM_ADDR_ATTR_ADDR4] = { .type = NLA_U32, },
1206 [MPTCP_PM_ADDR_ATTR_ADDR6] =
1207 NLA_POLICY_EXACT_LEN(sizeof(struct in6_addr)),
1208 [MPTCP_PM_ADDR_ATTR_PORT] = { .type = NLA_U16 },
1209 [MPTCP_PM_ADDR_ATTR_FLAGS] = { .type = NLA_U32 },
1210 [MPTCP_PM_ADDR_ATTR_IF_IDX] = { .type = NLA_S32 },
1211 };
1212
1213 static const struct nla_policy mptcp_pm_policy[MPTCP_PM_ATTR_MAX + 1] = {
1214 [MPTCP_PM_ATTR_ADDR] =
1215 NLA_POLICY_NESTED(mptcp_pm_addr_policy),
1216 [MPTCP_PM_ATTR_RCV_ADD_ADDRS] = { .type = NLA_U32, },
1217 [MPTCP_PM_ATTR_SUBFLOWS] = { .type = NLA_U32, },
1218 [MPTCP_PM_ATTR_TOKEN] = { .type = NLA_U32, },
1219 [MPTCP_PM_ATTR_LOC_ID] = { .type = NLA_U8, },
1220 [MPTCP_PM_ATTR_ADDR_REMOTE] =
1221 NLA_POLICY_NESTED(mptcp_pm_addr_policy),
1222 };
1223
mptcp_pm_nl_subflow_chk_stale(const struct mptcp_sock * msk,struct sock * ssk)1224 void mptcp_pm_nl_subflow_chk_stale(const struct mptcp_sock *msk, struct sock *ssk)
1225 {
1226 struct mptcp_subflow_context *iter, *subflow = mptcp_subflow_ctx(ssk);
1227 struct sock *sk = (struct sock *)msk;
1228 unsigned int active_max_loss_cnt;
1229 struct net *net = sock_net(sk);
1230 unsigned int stale_loss_cnt;
1231 bool slow;
1232
1233 stale_loss_cnt = mptcp_stale_loss_cnt(net);
1234 if (subflow->stale || !stale_loss_cnt || subflow->stale_count <= stale_loss_cnt)
1235 return;
1236
1237 /* look for another available subflow not in loss state */
1238 active_max_loss_cnt = max_t(int, stale_loss_cnt - 1, 1);
1239 mptcp_for_each_subflow(msk, iter) {
1240 if (iter != subflow && mptcp_subflow_active(iter) &&
1241 iter->stale_count < active_max_loss_cnt) {
1242 /* we have some alternatives, try to mark this subflow as idle ...*/
1243 slow = lock_sock_fast(ssk);
1244 if (!tcp_rtx_and_write_queues_empty(ssk)) {
1245 subflow->stale = 1;
1246 __mptcp_retransmit_pending_data(sk);
1247 MPTCP_INC_STATS(net, MPTCP_MIB_SUBFLOWSTALE);
1248 }
1249 unlock_sock_fast(ssk, slow);
1250
1251 /* always try to push the pending data regardless of re-injections:
1252 * we can possibly use backup subflows now, and subflow selection
1253 * is cheap under the msk socket lock
1254 */
1255 __mptcp_push_pending(sk, 0);
1256 return;
1257 }
1258 }
1259 }
1260
mptcp_pm_family_to_addr(int family)1261 static int mptcp_pm_family_to_addr(int family)
1262 {
1263 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1264 if (family == AF_INET6)
1265 return MPTCP_PM_ADDR_ATTR_ADDR6;
1266 #endif
1267 return MPTCP_PM_ADDR_ATTR_ADDR4;
1268 }
1269
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)1270 static int mptcp_pm_parse_pm_addr_attr(struct nlattr *tb[],
1271 const struct nlattr *attr,
1272 struct genl_info *info,
1273 struct mptcp_addr_info *addr,
1274 bool require_family)
1275 {
1276 int err, addr_addr;
1277
1278 if (!attr) {
1279 GENL_SET_ERR_MSG(info, "missing address info");
1280 return -EINVAL;
1281 }
1282
1283 /* no validation needed - was already done via nested policy */
1284 err = nla_parse_nested_deprecated(tb, MPTCP_PM_ADDR_ATTR_MAX, attr,
1285 mptcp_pm_addr_policy, info->extack);
1286 if (err)
1287 return err;
1288
1289 if (tb[MPTCP_PM_ADDR_ATTR_ID])
1290 addr->id = nla_get_u8(tb[MPTCP_PM_ADDR_ATTR_ID]);
1291
1292 if (!tb[MPTCP_PM_ADDR_ATTR_FAMILY]) {
1293 if (!require_family)
1294 return 0;
1295
1296 NL_SET_ERR_MSG_ATTR(info->extack, attr,
1297 "missing family");
1298 return -EINVAL;
1299 }
1300
1301 addr->family = nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_FAMILY]);
1302 if (addr->family != AF_INET
1303 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1304 && addr->family != AF_INET6
1305 #endif
1306 ) {
1307 NL_SET_ERR_MSG_ATTR(info->extack, attr,
1308 "unknown address family");
1309 return -EINVAL;
1310 }
1311 addr_addr = mptcp_pm_family_to_addr(addr->family);
1312 if (!tb[addr_addr]) {
1313 NL_SET_ERR_MSG_ATTR(info->extack, attr,
1314 "missing address data");
1315 return -EINVAL;
1316 }
1317
1318 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1319 if (addr->family == AF_INET6)
1320 addr->addr6 = nla_get_in6_addr(tb[addr_addr]);
1321 else
1322 #endif
1323 addr->addr.s_addr = nla_get_in_addr(tb[addr_addr]);
1324
1325 if (tb[MPTCP_PM_ADDR_ATTR_PORT])
1326 addr->port = htons(nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_PORT]));
1327
1328 return 0;
1329 }
1330
mptcp_pm_parse_addr(struct nlattr * attr,struct genl_info * info,struct mptcp_addr_info * addr)1331 int mptcp_pm_parse_addr(struct nlattr *attr, struct genl_info *info,
1332 struct mptcp_addr_info *addr)
1333 {
1334 struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1];
1335
1336 memset(addr, 0, sizeof(*addr));
1337
1338 return mptcp_pm_parse_pm_addr_attr(tb, attr, info, addr, true);
1339 }
1340
mptcp_pm_parse_entry(struct nlattr * attr,struct genl_info * info,bool require_family,struct mptcp_pm_addr_entry * entry)1341 int mptcp_pm_parse_entry(struct nlattr *attr, struct genl_info *info,
1342 bool require_family,
1343 struct mptcp_pm_addr_entry *entry)
1344 {
1345 struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1];
1346 int err;
1347
1348 memset(entry, 0, sizeof(*entry));
1349
1350 err = mptcp_pm_parse_pm_addr_attr(tb, attr, info, &entry->addr, require_family);
1351 if (err)
1352 return err;
1353
1354 if (tb[MPTCP_PM_ADDR_ATTR_IF_IDX]) {
1355 u32 val = nla_get_s32(tb[MPTCP_PM_ADDR_ATTR_IF_IDX]);
1356
1357 entry->ifindex = val;
1358 }
1359
1360 if (tb[MPTCP_PM_ADDR_ATTR_FLAGS])
1361 entry->flags = nla_get_u32(tb[MPTCP_PM_ADDR_ATTR_FLAGS]);
1362
1363 if (tb[MPTCP_PM_ADDR_ATTR_PORT])
1364 entry->addr.port = htons(nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_PORT]));
1365
1366 return 0;
1367 }
1368
genl_info_pm_nl(struct genl_info * info)1369 static struct pm_nl_pernet *genl_info_pm_nl(struct genl_info *info)
1370 {
1371 return pm_nl_get_pernet(genl_info_net(info));
1372 }
1373
mptcp_nl_add_subflow_or_signal_addr(struct net * net,struct mptcp_addr_info * addr)1374 static int mptcp_nl_add_subflow_or_signal_addr(struct net *net,
1375 struct mptcp_addr_info *addr)
1376 {
1377 struct mptcp_sock *msk;
1378 long s_slot = 0, s_num = 0;
1379
1380 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1381 struct sock *sk = (struct sock *)msk;
1382 struct mptcp_addr_info mpc_addr;
1383
1384 if (!READ_ONCE(msk->fully_established) ||
1385 mptcp_pm_is_userspace(msk))
1386 goto next;
1387
1388 /* if the endp linked to the init sf is re-added with a != ID */
1389 mptcp_local_address((struct sock_common *)msk, &mpc_addr);
1390
1391 lock_sock(sk);
1392 spin_lock_bh(&msk->pm.lock);
1393 if (mptcp_addresses_equal(addr, &mpc_addr, addr->port))
1394 msk->mpc_endpoint_id = addr->id;
1395 mptcp_pm_create_subflow_or_signal_addr(msk);
1396 spin_unlock_bh(&msk->pm.lock);
1397 release_sock(sk);
1398
1399 next:
1400 sock_put(sk);
1401 cond_resched();
1402 }
1403
1404 return 0;
1405 }
1406
mptcp_pm_has_addr_attr_id(const struct nlattr * attr,struct genl_info * info)1407 static bool mptcp_pm_has_addr_attr_id(const struct nlattr *attr,
1408 struct genl_info *info)
1409 {
1410 struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1];
1411
1412 if (!nla_parse_nested_deprecated(tb, MPTCP_PM_ADDR_ATTR_MAX, attr,
1413 mptcp_pm_addr_policy, info->extack) &&
1414 tb[MPTCP_PM_ADDR_ATTR_ID])
1415 return true;
1416 return false;
1417 }
1418
mptcp_nl_cmd_add_addr(struct sk_buff * skb,struct genl_info * info)1419 static int mptcp_nl_cmd_add_addr(struct sk_buff *skb, struct genl_info *info)
1420 {
1421 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
1422 struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1423 struct mptcp_pm_addr_entry addr, *entry;
1424 int ret;
1425
1426 ret = mptcp_pm_parse_entry(attr, info, true, &addr);
1427 if (ret < 0)
1428 return ret;
1429
1430 if (addr.addr.port && !address_use_port(&addr)) {
1431 GENL_SET_ERR_MSG(info, "flags must have signal and not subflow when using port");
1432 return -EINVAL;
1433 }
1434
1435 if (addr.flags & MPTCP_PM_ADDR_FLAG_SIGNAL &&
1436 addr.flags & MPTCP_PM_ADDR_FLAG_FULLMESH) {
1437 GENL_SET_ERR_MSG(info, "flags mustn't have both signal and fullmesh");
1438 return -EINVAL;
1439 }
1440
1441 if (addr.flags & MPTCP_PM_ADDR_FLAG_IMPLICIT) {
1442 GENL_SET_ERR_MSG(info, "can't create IMPLICIT endpoint");
1443 return -EINVAL;
1444 }
1445
1446 entry = kzalloc(sizeof(*entry), GFP_KERNEL_ACCOUNT);
1447 if (!entry) {
1448 GENL_SET_ERR_MSG(info, "can't allocate addr");
1449 return -ENOMEM;
1450 }
1451
1452 *entry = addr;
1453 if (entry->addr.port) {
1454 ret = mptcp_pm_nl_create_listen_socket(skb->sk, entry);
1455 if (ret) {
1456 GENL_SET_ERR_MSG_FMT(info, "create listen socket error: %d", ret);
1457 goto out_free;
1458 }
1459 }
1460 ret = mptcp_pm_nl_append_new_local_addr(pernet, entry,
1461 !mptcp_pm_has_addr_attr_id(attr, info));
1462 if (ret < 0) {
1463 GENL_SET_ERR_MSG_FMT(info, "too many addresses or duplicate one: %d", ret);
1464 goto out_free;
1465 }
1466
1467 mptcp_nl_add_subflow_or_signal_addr(sock_net(skb->sk), &entry->addr);
1468 return 0;
1469
1470 out_free:
1471 __mptcp_pm_release_addr_entry(entry);
1472 return ret;
1473 }
1474
mptcp_pm_nl_get_flags_and_ifindex_by_id(struct mptcp_sock * msk,unsigned int id,u8 * flags,int * ifindex)1475 int mptcp_pm_nl_get_flags_and_ifindex_by_id(struct mptcp_sock *msk, unsigned int id,
1476 u8 *flags, int *ifindex)
1477 {
1478 struct mptcp_pm_addr_entry *entry;
1479 struct sock *sk = (struct sock *)msk;
1480 struct net *net = sock_net(sk);
1481
1482 /* No entries with ID 0 */
1483 if (id == 0)
1484 return 0;
1485
1486 rcu_read_lock();
1487 entry = __lookup_addr_by_id(pm_nl_get_pernet(net), id);
1488 if (entry) {
1489 *flags = entry->flags;
1490 *ifindex = entry->ifindex;
1491 }
1492 rcu_read_unlock();
1493
1494 return 0;
1495 }
1496
remove_anno_list_by_saddr(struct mptcp_sock * msk,const struct mptcp_addr_info * addr)1497 static bool remove_anno_list_by_saddr(struct mptcp_sock *msk,
1498 const struct mptcp_addr_info *addr)
1499 {
1500 struct mptcp_pm_add_entry *entry;
1501
1502 entry = mptcp_pm_del_add_timer(msk, addr, false);
1503 if (entry) {
1504 kfree(entry);
1505 return true;
1506 }
1507
1508 return false;
1509 }
1510
mptcp_endp_get_local_id(struct mptcp_sock * msk,const struct mptcp_addr_info * addr)1511 static u8 mptcp_endp_get_local_id(struct mptcp_sock *msk,
1512 const struct mptcp_addr_info *addr)
1513 {
1514 return msk->mpc_endpoint_id == addr->id ? 0 : addr->id;
1515 }
1516
mptcp_pm_remove_anno_addr(struct mptcp_sock * msk,const struct mptcp_addr_info * addr,bool force)1517 static bool mptcp_pm_remove_anno_addr(struct mptcp_sock *msk,
1518 const struct mptcp_addr_info *addr,
1519 bool force)
1520 {
1521 struct mptcp_rm_list list = { .nr = 0 };
1522 bool ret;
1523
1524 list.ids[list.nr++] = mptcp_endp_get_local_id(msk, addr);
1525
1526 ret = remove_anno_list_by_saddr(msk, addr);
1527 if (ret || force) {
1528 spin_lock_bh(&msk->pm.lock);
1529 if (ret) {
1530 __set_bit(addr->id, msk->pm.id_avail_bitmap);
1531 msk->pm.add_addr_signaled--;
1532 }
1533 mptcp_pm_remove_addr(msk, &list);
1534 spin_unlock_bh(&msk->pm.lock);
1535 }
1536 return ret;
1537 }
1538
__mark_subflow_endp_available(struct mptcp_sock * msk,u8 id)1539 static void __mark_subflow_endp_available(struct mptcp_sock *msk, u8 id)
1540 {
1541 /* If it was marked as used, and not ID 0, decrement local_addr_used */
1542 if (!__test_and_set_bit(id ? : msk->mpc_endpoint_id, msk->pm.id_avail_bitmap) &&
1543 id && !WARN_ON_ONCE(msk->pm.local_addr_used == 0))
1544 msk->pm.local_addr_used--;
1545 }
1546
mptcp_nl_remove_subflow_and_signal_addr(struct net * net,const struct mptcp_pm_addr_entry * entry)1547 static int mptcp_nl_remove_subflow_and_signal_addr(struct net *net,
1548 const struct mptcp_pm_addr_entry *entry)
1549 {
1550 const struct mptcp_addr_info *addr = &entry->addr;
1551 struct mptcp_rm_list list = { .nr = 1 };
1552 long s_slot = 0, s_num = 0;
1553 struct mptcp_sock *msk;
1554
1555 pr_debug("remove_id=%d\n", addr->id);
1556
1557 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1558 struct sock *sk = (struct sock *)msk;
1559 bool remove_subflow;
1560
1561 if (mptcp_pm_is_userspace(msk))
1562 goto next;
1563
1564 if (list_empty(&msk->conn_list)) {
1565 mptcp_pm_remove_anno_addr(msk, addr, false);
1566 goto next;
1567 }
1568
1569 lock_sock(sk);
1570 remove_subflow = lookup_subflow_by_saddr(&msk->conn_list, addr);
1571 mptcp_pm_remove_anno_addr(msk, addr, remove_subflow &&
1572 !(entry->flags & MPTCP_PM_ADDR_FLAG_IMPLICIT));
1573
1574 list.ids[0] = mptcp_endp_get_local_id(msk, addr);
1575 if (remove_subflow) {
1576 spin_lock_bh(&msk->pm.lock);
1577 mptcp_pm_nl_rm_subflow_received(msk, &list);
1578 spin_unlock_bh(&msk->pm.lock);
1579 }
1580
1581 if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) {
1582 spin_lock_bh(&msk->pm.lock);
1583 __mark_subflow_endp_available(msk, list.ids[0]);
1584 spin_unlock_bh(&msk->pm.lock);
1585 }
1586
1587 if (msk->mpc_endpoint_id == entry->addr.id)
1588 msk->mpc_endpoint_id = 0;
1589 release_sock(sk);
1590
1591 next:
1592 sock_put(sk);
1593 cond_resched();
1594 }
1595
1596 return 0;
1597 }
1598
mptcp_nl_remove_id_zero_address(struct net * net,struct mptcp_addr_info * addr)1599 static int mptcp_nl_remove_id_zero_address(struct net *net,
1600 struct mptcp_addr_info *addr)
1601 {
1602 struct mptcp_rm_list list = { .nr = 0 };
1603 long s_slot = 0, s_num = 0;
1604 struct mptcp_sock *msk;
1605
1606 list.ids[list.nr++] = 0;
1607
1608 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1609 struct sock *sk = (struct sock *)msk;
1610 struct mptcp_addr_info msk_local;
1611
1612 if (list_empty(&msk->conn_list) || mptcp_pm_is_userspace(msk))
1613 goto next;
1614
1615 mptcp_local_address((struct sock_common *)msk, &msk_local);
1616 if (!mptcp_addresses_equal(&msk_local, addr, addr->port))
1617 goto next;
1618
1619 lock_sock(sk);
1620 spin_lock_bh(&msk->pm.lock);
1621 mptcp_pm_remove_addr(msk, &list);
1622 mptcp_pm_nl_rm_subflow_received(msk, &list);
1623 __mark_subflow_endp_available(msk, 0);
1624 spin_unlock_bh(&msk->pm.lock);
1625 release_sock(sk);
1626
1627 next:
1628 sock_put(sk);
1629 cond_resched();
1630 }
1631
1632 return 0;
1633 }
1634
mptcp_nl_cmd_del_addr(struct sk_buff * skb,struct genl_info * info)1635 static int mptcp_nl_cmd_del_addr(struct sk_buff *skb, struct genl_info *info)
1636 {
1637 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
1638 struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1639 struct mptcp_pm_addr_entry addr, *entry;
1640 unsigned int addr_max;
1641 int ret;
1642
1643 ret = mptcp_pm_parse_entry(attr, info, false, &addr);
1644 if (ret < 0)
1645 return ret;
1646
1647 /* the zero id address is special: the first address used by the msk
1648 * always gets such an id, so different subflows can have different zero
1649 * id addresses. Additionally zero id is not accounted for in id_bitmap.
1650 * Let's use an 'mptcp_rm_list' instead of the common remove code.
1651 */
1652 if (addr.addr.id == 0)
1653 return mptcp_nl_remove_id_zero_address(sock_net(skb->sk), &addr.addr);
1654
1655 spin_lock_bh(&pernet->lock);
1656 entry = __lookup_addr_by_id(pernet, addr.addr.id);
1657 if (!entry) {
1658 GENL_SET_ERR_MSG(info, "address not found");
1659 spin_unlock_bh(&pernet->lock);
1660 return -EINVAL;
1661 }
1662 if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) {
1663 addr_max = pernet->add_addr_signal_max;
1664 WRITE_ONCE(pernet->add_addr_signal_max, addr_max - 1);
1665 }
1666 if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) {
1667 addr_max = pernet->local_addr_max;
1668 WRITE_ONCE(pernet->local_addr_max, addr_max - 1);
1669 }
1670
1671 pernet->addrs--;
1672 list_del_rcu(&entry->list);
1673 __clear_bit(entry->addr.id, pernet->id_bitmap);
1674 spin_unlock_bh(&pernet->lock);
1675
1676 mptcp_nl_remove_subflow_and_signal_addr(sock_net(skb->sk), entry);
1677 synchronize_rcu();
1678 __mptcp_pm_release_addr_entry(entry);
1679
1680 return ret;
1681 }
1682
1683 /* Called from the userspace PM only */
mptcp_pm_remove_addrs(struct mptcp_sock * msk,struct list_head * rm_list)1684 void mptcp_pm_remove_addrs(struct mptcp_sock *msk, struct list_head *rm_list)
1685 {
1686 struct mptcp_rm_list alist = { .nr = 0 };
1687 struct mptcp_pm_addr_entry *entry;
1688 int anno_nr = 0;
1689
1690 list_for_each_entry(entry, rm_list, list) {
1691 if (alist.nr >= MPTCP_RM_IDS_MAX)
1692 break;
1693
1694 /* only delete if either announced or matching a subflow */
1695 if (remove_anno_list_by_saddr(msk, &entry->addr))
1696 anno_nr++;
1697 else if (!lookup_subflow_by_saddr(&msk->conn_list,
1698 &entry->addr))
1699 continue;
1700
1701 alist.ids[alist.nr++] = entry->addr.id;
1702 }
1703
1704 if (alist.nr) {
1705 spin_lock_bh(&msk->pm.lock);
1706 msk->pm.add_addr_signaled -= anno_nr;
1707 mptcp_pm_remove_addr(msk, &alist);
1708 spin_unlock_bh(&msk->pm.lock);
1709 }
1710 }
1711
1712 /* Called from the in-kernel PM only */
mptcp_pm_remove_addrs_and_subflows(struct mptcp_sock * msk,struct list_head * rm_list)1713 static void mptcp_pm_remove_addrs_and_subflows(struct mptcp_sock *msk,
1714 struct list_head *rm_list)
1715 {
1716 struct mptcp_rm_list alist = { .nr = 0 }, slist = { .nr = 0 };
1717 struct mptcp_pm_addr_entry *entry;
1718
1719 list_for_each_entry(entry, rm_list, list) {
1720 if (slist.nr < MPTCP_RM_IDS_MAX &&
1721 lookup_subflow_by_saddr(&msk->conn_list, &entry->addr))
1722 slist.ids[slist.nr++] = mptcp_endp_get_local_id(msk, &entry->addr);
1723
1724 if (alist.nr < MPTCP_RM_IDS_MAX &&
1725 remove_anno_list_by_saddr(msk, &entry->addr))
1726 alist.ids[alist.nr++] = mptcp_endp_get_local_id(msk, &entry->addr);
1727 }
1728
1729 spin_lock_bh(&msk->pm.lock);
1730 if (alist.nr) {
1731 msk->pm.add_addr_signaled -= alist.nr;
1732 mptcp_pm_remove_addr(msk, &alist);
1733 }
1734 if (slist.nr)
1735 mptcp_pm_nl_rm_subflow_received(msk, &slist);
1736 /* Reset counters: maybe some subflows have been removed before */
1737 bitmap_fill(msk->pm.id_avail_bitmap, MPTCP_PM_MAX_ADDR_ID + 1);
1738 msk->pm.local_addr_used = 0;
1739 spin_unlock_bh(&msk->pm.lock);
1740 }
1741
mptcp_nl_remove_addrs_list(struct net * net,struct list_head * rm_list)1742 static void mptcp_nl_remove_addrs_list(struct net *net,
1743 struct list_head *rm_list)
1744 {
1745 long s_slot = 0, s_num = 0;
1746 struct mptcp_sock *msk;
1747
1748 if (list_empty(rm_list))
1749 return;
1750
1751 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1752 struct sock *sk = (struct sock *)msk;
1753
1754 if (!mptcp_pm_is_userspace(msk)) {
1755 lock_sock(sk);
1756 mptcp_pm_remove_addrs_and_subflows(msk, rm_list);
1757 release_sock(sk);
1758 }
1759
1760 sock_put(sk);
1761 cond_resched();
1762 }
1763 }
1764
1765 /* caller must ensure the RCU grace period is already elapsed */
__flush_addrs(struct list_head * list)1766 static void __flush_addrs(struct list_head *list)
1767 {
1768 while (!list_empty(list)) {
1769 struct mptcp_pm_addr_entry *cur;
1770
1771 cur = list_entry(list->next,
1772 struct mptcp_pm_addr_entry, list);
1773 list_del_rcu(&cur->list);
1774 __mptcp_pm_release_addr_entry(cur);
1775 }
1776 }
1777
__reset_counters(struct pm_nl_pernet * pernet)1778 static void __reset_counters(struct pm_nl_pernet *pernet)
1779 {
1780 WRITE_ONCE(pernet->add_addr_signal_max, 0);
1781 WRITE_ONCE(pernet->add_addr_accept_max, 0);
1782 WRITE_ONCE(pernet->local_addr_max, 0);
1783 pernet->addrs = 0;
1784 }
1785
mptcp_nl_cmd_flush_addrs(struct sk_buff * skb,struct genl_info * info)1786 static int mptcp_nl_cmd_flush_addrs(struct sk_buff *skb, struct genl_info *info)
1787 {
1788 struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1789 LIST_HEAD(free_list);
1790
1791 spin_lock_bh(&pernet->lock);
1792 list_splice_init(&pernet->local_addr_list, &free_list);
1793 __reset_counters(pernet);
1794 pernet->next_id = 1;
1795 bitmap_zero(pernet->id_bitmap, MPTCP_PM_MAX_ADDR_ID + 1);
1796 spin_unlock_bh(&pernet->lock);
1797 mptcp_nl_remove_addrs_list(sock_net(skb->sk), &free_list);
1798 synchronize_rcu();
1799 __flush_addrs(&free_list);
1800 return 0;
1801 }
1802
mptcp_nl_fill_addr(struct sk_buff * skb,struct mptcp_pm_addr_entry * entry)1803 static int mptcp_nl_fill_addr(struct sk_buff *skb,
1804 struct mptcp_pm_addr_entry *entry)
1805 {
1806 struct mptcp_addr_info *addr = &entry->addr;
1807 struct nlattr *attr;
1808
1809 attr = nla_nest_start(skb, MPTCP_PM_ATTR_ADDR);
1810 if (!attr)
1811 return -EMSGSIZE;
1812
1813 if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_FAMILY, addr->family))
1814 goto nla_put_failure;
1815 if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_PORT, ntohs(addr->port)))
1816 goto nla_put_failure;
1817 if (nla_put_u8(skb, MPTCP_PM_ADDR_ATTR_ID, addr->id))
1818 goto nla_put_failure;
1819 if (nla_put_u32(skb, MPTCP_PM_ADDR_ATTR_FLAGS, entry->flags))
1820 goto nla_put_failure;
1821 if (entry->ifindex &&
1822 nla_put_s32(skb, MPTCP_PM_ADDR_ATTR_IF_IDX, entry->ifindex))
1823 goto nla_put_failure;
1824
1825 if (addr->family == AF_INET &&
1826 nla_put_in_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR4,
1827 addr->addr.s_addr))
1828 goto nla_put_failure;
1829 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1830 else if (addr->family == AF_INET6 &&
1831 nla_put_in6_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR6, &addr->addr6))
1832 goto nla_put_failure;
1833 #endif
1834 nla_nest_end(skb, attr);
1835 return 0;
1836
1837 nla_put_failure:
1838 nla_nest_cancel(skb, attr);
1839 return -EMSGSIZE;
1840 }
1841
mptcp_nl_cmd_get_addr(struct sk_buff * skb,struct genl_info * info)1842 static int mptcp_nl_cmd_get_addr(struct sk_buff *skb, struct genl_info *info)
1843 {
1844 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
1845 struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1846 struct mptcp_pm_addr_entry addr, *entry;
1847 struct sk_buff *msg;
1848 void *reply;
1849 int ret;
1850
1851 ret = mptcp_pm_parse_entry(attr, info, false, &addr);
1852 if (ret < 0)
1853 return ret;
1854
1855 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1856 if (!msg)
1857 return -ENOMEM;
1858
1859 reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0,
1860 info->genlhdr->cmd);
1861 if (!reply) {
1862 GENL_SET_ERR_MSG(info, "not enough space in Netlink message");
1863 ret = -EMSGSIZE;
1864 goto fail;
1865 }
1866
1867 spin_lock_bh(&pernet->lock);
1868 entry = __lookup_addr_by_id(pernet, addr.addr.id);
1869 if (!entry) {
1870 GENL_SET_ERR_MSG(info, "address not found");
1871 ret = -EINVAL;
1872 goto unlock_fail;
1873 }
1874
1875 ret = mptcp_nl_fill_addr(msg, entry);
1876 if (ret)
1877 goto unlock_fail;
1878
1879 genlmsg_end(msg, reply);
1880 ret = genlmsg_reply(msg, info);
1881 spin_unlock_bh(&pernet->lock);
1882 return ret;
1883
1884 unlock_fail:
1885 spin_unlock_bh(&pernet->lock);
1886
1887 fail:
1888 nlmsg_free(msg);
1889 return ret;
1890 }
1891
mptcp_nl_cmd_dump_addrs(struct sk_buff * msg,struct netlink_callback * cb)1892 static int mptcp_nl_cmd_dump_addrs(struct sk_buff *msg,
1893 struct netlink_callback *cb)
1894 {
1895 struct net *net = sock_net(msg->sk);
1896 struct mptcp_pm_addr_entry *entry;
1897 struct pm_nl_pernet *pernet;
1898 int id = cb->args[0];
1899 void *hdr;
1900 int i;
1901
1902 pernet = pm_nl_get_pernet(net);
1903
1904 spin_lock_bh(&pernet->lock);
1905 for (i = id; i < MPTCP_PM_MAX_ADDR_ID + 1; i++) {
1906 if (test_bit(i, pernet->id_bitmap)) {
1907 entry = __lookup_addr_by_id(pernet, i);
1908 if (!entry)
1909 break;
1910
1911 if (entry->addr.id <= id)
1912 continue;
1913
1914 hdr = genlmsg_put(msg, NETLINK_CB(cb->skb).portid,
1915 cb->nlh->nlmsg_seq, &mptcp_genl_family,
1916 NLM_F_MULTI, MPTCP_PM_CMD_GET_ADDR);
1917 if (!hdr)
1918 break;
1919
1920 if (mptcp_nl_fill_addr(msg, entry) < 0) {
1921 genlmsg_cancel(msg, hdr);
1922 break;
1923 }
1924
1925 id = entry->addr.id;
1926 genlmsg_end(msg, hdr);
1927 }
1928 }
1929 spin_unlock_bh(&pernet->lock);
1930
1931 cb->args[0] = id;
1932 return msg->len;
1933 }
1934
parse_limit(struct genl_info * info,int id,unsigned int * limit)1935 static int parse_limit(struct genl_info *info, int id, unsigned int *limit)
1936 {
1937 struct nlattr *attr = info->attrs[id];
1938
1939 if (!attr)
1940 return 0;
1941
1942 *limit = nla_get_u32(attr);
1943 if (*limit > MPTCP_PM_ADDR_MAX) {
1944 GENL_SET_ERR_MSG(info, "limit greater than maximum");
1945 return -EINVAL;
1946 }
1947 return 0;
1948 }
1949
1950 static int
mptcp_nl_cmd_set_limits(struct sk_buff * skb,struct genl_info * info)1951 mptcp_nl_cmd_set_limits(struct sk_buff *skb, struct genl_info *info)
1952 {
1953 struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1954 unsigned int rcv_addrs, subflows;
1955 int ret;
1956
1957 spin_lock_bh(&pernet->lock);
1958 rcv_addrs = pernet->add_addr_accept_max;
1959 ret = parse_limit(info, MPTCP_PM_ATTR_RCV_ADD_ADDRS, &rcv_addrs);
1960 if (ret)
1961 goto unlock;
1962
1963 subflows = pernet->subflows_max;
1964 ret = parse_limit(info, MPTCP_PM_ATTR_SUBFLOWS, &subflows);
1965 if (ret)
1966 goto unlock;
1967
1968 WRITE_ONCE(pernet->add_addr_accept_max, rcv_addrs);
1969 WRITE_ONCE(pernet->subflows_max, subflows);
1970
1971 unlock:
1972 spin_unlock_bh(&pernet->lock);
1973 return ret;
1974 }
1975
1976 static int
mptcp_nl_cmd_get_limits(struct sk_buff * skb,struct genl_info * info)1977 mptcp_nl_cmd_get_limits(struct sk_buff *skb, struct genl_info *info)
1978 {
1979 struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1980 struct sk_buff *msg;
1981 void *reply;
1982
1983 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1984 if (!msg)
1985 return -ENOMEM;
1986
1987 reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0,
1988 MPTCP_PM_CMD_GET_LIMITS);
1989 if (!reply)
1990 goto fail;
1991
1992 if (nla_put_u32(msg, MPTCP_PM_ATTR_RCV_ADD_ADDRS,
1993 READ_ONCE(pernet->add_addr_accept_max)))
1994 goto fail;
1995
1996 if (nla_put_u32(msg, MPTCP_PM_ATTR_SUBFLOWS,
1997 READ_ONCE(pernet->subflows_max)))
1998 goto fail;
1999
2000 genlmsg_end(msg, reply);
2001 return genlmsg_reply(msg, info);
2002
2003 fail:
2004 GENL_SET_ERR_MSG(info, "not enough space in Netlink message");
2005 nlmsg_free(msg);
2006 return -EMSGSIZE;
2007 }
2008
mptcp_pm_nl_fullmesh(struct mptcp_sock * msk,struct mptcp_addr_info * addr)2009 static void mptcp_pm_nl_fullmesh(struct mptcp_sock *msk,
2010 struct mptcp_addr_info *addr)
2011 {
2012 struct mptcp_rm_list list = { .nr = 0 };
2013
2014 list.ids[list.nr++] = mptcp_endp_get_local_id(msk, addr);
2015
2016 spin_lock_bh(&msk->pm.lock);
2017 mptcp_pm_nl_rm_subflow_received(msk, &list);
2018 __mark_subflow_endp_available(msk, list.ids[0]);
2019 mptcp_pm_create_subflow_or_signal_addr(msk);
2020 spin_unlock_bh(&msk->pm.lock);
2021 }
2022
mptcp_nl_set_flags(struct net * net,struct mptcp_addr_info * addr,u8 bkup,u8 changed)2023 static int mptcp_nl_set_flags(struct net *net,
2024 struct mptcp_addr_info *addr,
2025 u8 bkup, u8 changed)
2026 {
2027 long s_slot = 0, s_num = 0;
2028 struct mptcp_sock *msk;
2029 int ret = -EINVAL;
2030
2031 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
2032 struct sock *sk = (struct sock *)msk;
2033
2034 if (list_empty(&msk->conn_list) || mptcp_pm_is_userspace(msk))
2035 goto next;
2036
2037 lock_sock(sk);
2038 if (changed & MPTCP_PM_ADDR_FLAG_BACKUP)
2039 ret = mptcp_pm_nl_mp_prio_send_ack(msk, addr, NULL, bkup);
2040 if (changed & MPTCP_PM_ADDR_FLAG_FULLMESH)
2041 mptcp_pm_nl_fullmesh(msk, addr);
2042 release_sock(sk);
2043
2044 next:
2045 sock_put(sk);
2046 cond_resched();
2047 }
2048
2049 return ret;
2050 }
2051
mptcp_pm_nl_set_flags(struct net * net,struct mptcp_pm_addr_entry * addr,u8 bkup)2052 int mptcp_pm_nl_set_flags(struct net *net, struct mptcp_pm_addr_entry *addr, u8 bkup)
2053 {
2054 struct pm_nl_pernet *pernet = pm_nl_get_pernet(net);
2055 u8 changed, mask = MPTCP_PM_ADDR_FLAG_BACKUP |
2056 MPTCP_PM_ADDR_FLAG_FULLMESH;
2057 struct mptcp_pm_addr_entry *entry;
2058 u8 lookup_by_id = 0;
2059
2060 if (addr->addr.family == AF_UNSPEC) {
2061 lookup_by_id = 1;
2062 if (!addr->addr.id)
2063 return -EOPNOTSUPP;
2064 }
2065
2066 spin_lock_bh(&pernet->lock);
2067 entry = __lookup_addr(pernet, &addr->addr, lookup_by_id);
2068 if (!entry) {
2069 spin_unlock_bh(&pernet->lock);
2070 return -EINVAL;
2071 }
2072 if ((addr->flags & MPTCP_PM_ADDR_FLAG_FULLMESH) &&
2073 (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL)) {
2074 spin_unlock_bh(&pernet->lock);
2075 return -EINVAL;
2076 }
2077
2078 changed = (addr->flags ^ entry->flags) & mask;
2079 entry->flags = (entry->flags & ~mask) | (addr->flags & mask);
2080 *addr = *entry;
2081 spin_unlock_bh(&pernet->lock);
2082
2083 mptcp_nl_set_flags(net, &addr->addr, bkup, changed);
2084 return 0;
2085 }
2086
mptcp_nl_cmd_set_flags(struct sk_buff * skb,struct genl_info * info)2087 static int mptcp_nl_cmd_set_flags(struct sk_buff *skb, struct genl_info *info)
2088 {
2089 struct mptcp_pm_addr_entry remote = { .addr = { .family = AF_UNSPEC }, };
2090 struct mptcp_pm_addr_entry addr = { .addr = { .family = AF_UNSPEC }, };
2091 struct nlattr *attr_rem = info->attrs[MPTCP_PM_ATTR_ADDR_REMOTE];
2092 struct nlattr *token = info->attrs[MPTCP_PM_ATTR_TOKEN];
2093 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
2094 struct net *net = sock_net(skb->sk);
2095 u8 bkup = 0;
2096 int ret;
2097
2098 ret = mptcp_pm_parse_entry(attr, info, false, &addr);
2099 if (ret < 0)
2100 return ret;
2101
2102 if (attr_rem) {
2103 ret = mptcp_pm_parse_entry(attr_rem, info, false, &remote);
2104 if (ret < 0)
2105 return ret;
2106 }
2107
2108 if (addr.flags & MPTCP_PM_ADDR_FLAG_BACKUP)
2109 bkup = 1;
2110
2111 return mptcp_pm_set_flags(net, token, &addr, &remote, bkup);
2112 }
2113
mptcp_nl_mcast_send(struct net * net,struct sk_buff * nlskb,gfp_t gfp)2114 static void mptcp_nl_mcast_send(struct net *net, struct sk_buff *nlskb, gfp_t gfp)
2115 {
2116 genlmsg_multicast_netns(&mptcp_genl_family, net,
2117 nlskb, 0, MPTCP_PM_EV_GRP_OFFSET, gfp);
2118 }
2119
mptcp_userspace_pm_active(const struct mptcp_sock * msk)2120 bool mptcp_userspace_pm_active(const struct mptcp_sock *msk)
2121 {
2122 return genl_has_listeners(&mptcp_genl_family,
2123 sock_net((const struct sock *)msk),
2124 MPTCP_PM_EV_GRP_OFFSET);
2125 }
2126
mptcp_event_add_subflow(struct sk_buff * skb,const struct sock * ssk)2127 static int mptcp_event_add_subflow(struct sk_buff *skb, const struct sock *ssk)
2128 {
2129 const struct inet_sock *issk = inet_sk(ssk);
2130 const struct mptcp_subflow_context *sf;
2131
2132 if (nla_put_u16(skb, MPTCP_ATTR_FAMILY, ssk->sk_family))
2133 return -EMSGSIZE;
2134
2135 switch (ssk->sk_family) {
2136 case AF_INET:
2137 if (nla_put_in_addr(skb, MPTCP_ATTR_SADDR4, issk->inet_saddr))
2138 return -EMSGSIZE;
2139 if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, issk->inet_daddr))
2140 return -EMSGSIZE;
2141 break;
2142 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
2143 case AF_INET6: {
2144 const struct ipv6_pinfo *np = inet6_sk(ssk);
2145
2146 if (nla_put_in6_addr(skb, MPTCP_ATTR_SADDR6, &np->saddr))
2147 return -EMSGSIZE;
2148 if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &ssk->sk_v6_daddr))
2149 return -EMSGSIZE;
2150 break;
2151 }
2152 #endif
2153 default:
2154 WARN_ON_ONCE(1);
2155 return -EMSGSIZE;
2156 }
2157
2158 if (nla_put_be16(skb, MPTCP_ATTR_SPORT, issk->inet_sport))
2159 return -EMSGSIZE;
2160 if (nla_put_be16(skb, MPTCP_ATTR_DPORT, issk->inet_dport))
2161 return -EMSGSIZE;
2162
2163 sf = mptcp_subflow_ctx(ssk);
2164 if (WARN_ON_ONCE(!sf))
2165 return -EINVAL;
2166
2167 if (nla_put_u8(skb, MPTCP_ATTR_LOC_ID, subflow_get_local_id(sf)))
2168 return -EMSGSIZE;
2169
2170 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, sf->remote_id))
2171 return -EMSGSIZE;
2172
2173 return 0;
2174 }
2175
mptcp_event_put_token_and_ssk(struct sk_buff * skb,const struct mptcp_sock * msk,const struct sock * ssk)2176 static int mptcp_event_put_token_and_ssk(struct sk_buff *skb,
2177 const struct mptcp_sock *msk,
2178 const struct sock *ssk)
2179 {
2180 const struct sock *sk = (const struct sock *)msk;
2181 const struct mptcp_subflow_context *sf;
2182 u8 sk_err;
2183
2184 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token))
2185 return -EMSGSIZE;
2186
2187 if (mptcp_event_add_subflow(skb, ssk))
2188 return -EMSGSIZE;
2189
2190 sf = mptcp_subflow_ctx(ssk);
2191 if (WARN_ON_ONCE(!sf))
2192 return -EINVAL;
2193
2194 if (nla_put_u8(skb, MPTCP_ATTR_BACKUP, sf->backup))
2195 return -EMSGSIZE;
2196
2197 if (ssk->sk_bound_dev_if &&
2198 nla_put_s32(skb, MPTCP_ATTR_IF_IDX, ssk->sk_bound_dev_if))
2199 return -EMSGSIZE;
2200
2201 sk_err = READ_ONCE(ssk->sk_err);
2202 if (sk_err && sk->sk_state == TCP_ESTABLISHED &&
2203 nla_put_u8(skb, MPTCP_ATTR_ERROR, sk_err))
2204 return -EMSGSIZE;
2205
2206 return 0;
2207 }
2208
mptcp_event_sub_established(struct sk_buff * skb,const struct mptcp_sock * msk,const struct sock * ssk)2209 static int mptcp_event_sub_established(struct sk_buff *skb,
2210 const struct mptcp_sock *msk,
2211 const struct sock *ssk)
2212 {
2213 return mptcp_event_put_token_and_ssk(skb, msk, ssk);
2214 }
2215
mptcp_event_sub_closed(struct sk_buff * skb,const struct mptcp_sock * msk,const struct sock * ssk)2216 static int mptcp_event_sub_closed(struct sk_buff *skb,
2217 const struct mptcp_sock *msk,
2218 const struct sock *ssk)
2219 {
2220 const struct mptcp_subflow_context *sf;
2221
2222 if (mptcp_event_put_token_and_ssk(skb, msk, ssk))
2223 return -EMSGSIZE;
2224
2225 sf = mptcp_subflow_ctx(ssk);
2226 if (!sf->reset_seen)
2227 return 0;
2228
2229 if (nla_put_u32(skb, MPTCP_ATTR_RESET_REASON, sf->reset_reason))
2230 return -EMSGSIZE;
2231
2232 if (nla_put_u32(skb, MPTCP_ATTR_RESET_FLAGS, sf->reset_transient))
2233 return -EMSGSIZE;
2234
2235 return 0;
2236 }
2237
mptcp_event_created(struct sk_buff * skb,const struct mptcp_sock * msk,const struct sock * ssk)2238 static int mptcp_event_created(struct sk_buff *skb,
2239 const struct mptcp_sock *msk,
2240 const struct sock *ssk)
2241 {
2242 int err = nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token);
2243
2244 if (err)
2245 return err;
2246
2247 if (nla_put_u8(skb, MPTCP_ATTR_SERVER_SIDE, READ_ONCE(msk->pm.server_side)))
2248 return -EMSGSIZE;
2249
2250 return mptcp_event_add_subflow(skb, ssk);
2251 }
2252
mptcp_event_addr_removed(const struct mptcp_sock * msk,uint8_t id)2253 void mptcp_event_addr_removed(const struct mptcp_sock *msk, uint8_t id)
2254 {
2255 struct net *net = sock_net((const struct sock *)msk);
2256 struct nlmsghdr *nlh;
2257 struct sk_buff *skb;
2258
2259 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET))
2260 return;
2261
2262 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
2263 if (!skb)
2264 return;
2265
2266 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, MPTCP_EVENT_REMOVED);
2267 if (!nlh)
2268 goto nla_put_failure;
2269
2270 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token))
2271 goto nla_put_failure;
2272
2273 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, id))
2274 goto nla_put_failure;
2275
2276 genlmsg_end(skb, nlh);
2277 mptcp_nl_mcast_send(net, skb, GFP_ATOMIC);
2278 return;
2279
2280 nla_put_failure:
2281 nlmsg_free(skb);
2282 }
2283
mptcp_event_addr_announced(const struct sock * ssk,const struct mptcp_addr_info * info)2284 void mptcp_event_addr_announced(const struct sock *ssk,
2285 const struct mptcp_addr_info *info)
2286 {
2287 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
2288 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
2289 struct net *net = sock_net(ssk);
2290 struct nlmsghdr *nlh;
2291 struct sk_buff *skb;
2292
2293 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET))
2294 return;
2295
2296 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
2297 if (!skb)
2298 return;
2299
2300 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0,
2301 MPTCP_EVENT_ANNOUNCED);
2302 if (!nlh)
2303 goto nla_put_failure;
2304
2305 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token))
2306 goto nla_put_failure;
2307
2308 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, info->id))
2309 goto nla_put_failure;
2310
2311 if (nla_put_be16(skb, MPTCP_ATTR_DPORT,
2312 info->port == 0 ?
2313 inet_sk(ssk)->inet_dport :
2314 info->port))
2315 goto nla_put_failure;
2316
2317 switch (info->family) {
2318 case AF_INET:
2319 if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, info->addr.s_addr))
2320 goto nla_put_failure;
2321 break;
2322 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
2323 case AF_INET6:
2324 if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &info->addr6))
2325 goto nla_put_failure;
2326 break;
2327 #endif
2328 default:
2329 WARN_ON_ONCE(1);
2330 goto nla_put_failure;
2331 }
2332
2333 genlmsg_end(skb, nlh);
2334 mptcp_nl_mcast_send(net, skb, GFP_ATOMIC);
2335 return;
2336
2337 nla_put_failure:
2338 nlmsg_free(skb);
2339 }
2340
mptcp_event_pm_listener(const struct sock * ssk,enum mptcp_event_type event)2341 void mptcp_event_pm_listener(const struct sock *ssk,
2342 enum mptcp_event_type event)
2343 {
2344 const struct inet_sock *issk = inet_sk(ssk);
2345 struct net *net = sock_net(ssk);
2346 struct nlmsghdr *nlh;
2347 struct sk_buff *skb;
2348
2349 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET))
2350 return;
2351
2352 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
2353 if (!skb)
2354 return;
2355
2356 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, event);
2357 if (!nlh)
2358 goto nla_put_failure;
2359
2360 if (nla_put_u16(skb, MPTCP_ATTR_FAMILY, ssk->sk_family))
2361 goto nla_put_failure;
2362
2363 if (nla_put_be16(skb, MPTCP_ATTR_SPORT, issk->inet_sport))
2364 goto nla_put_failure;
2365
2366 switch (ssk->sk_family) {
2367 case AF_INET:
2368 if (nla_put_in_addr(skb, MPTCP_ATTR_SADDR4, issk->inet_saddr))
2369 goto nla_put_failure;
2370 break;
2371 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
2372 case AF_INET6: {
2373 const struct ipv6_pinfo *np = inet6_sk(ssk);
2374
2375 if (nla_put_in6_addr(skb, MPTCP_ATTR_SADDR6, &np->saddr))
2376 goto nla_put_failure;
2377 break;
2378 }
2379 #endif
2380 default:
2381 WARN_ON_ONCE(1);
2382 goto nla_put_failure;
2383 }
2384
2385 genlmsg_end(skb, nlh);
2386 mptcp_nl_mcast_send(net, skb, GFP_KERNEL);
2387 return;
2388
2389 nla_put_failure:
2390 nlmsg_free(skb);
2391 }
2392
mptcp_event(enum mptcp_event_type type,const struct mptcp_sock * msk,const struct sock * ssk,gfp_t gfp)2393 void mptcp_event(enum mptcp_event_type type, const struct mptcp_sock *msk,
2394 const struct sock *ssk, gfp_t gfp)
2395 {
2396 struct net *net = sock_net((const struct sock *)msk);
2397 struct nlmsghdr *nlh;
2398 struct sk_buff *skb;
2399
2400 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET))
2401 return;
2402
2403 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, gfp);
2404 if (!skb)
2405 return;
2406
2407 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, type);
2408 if (!nlh)
2409 goto nla_put_failure;
2410
2411 switch (type) {
2412 case MPTCP_EVENT_UNSPEC:
2413 WARN_ON_ONCE(1);
2414 break;
2415 case MPTCP_EVENT_CREATED:
2416 case MPTCP_EVENT_ESTABLISHED:
2417 if (mptcp_event_created(skb, msk, ssk) < 0)
2418 goto nla_put_failure;
2419 break;
2420 case MPTCP_EVENT_CLOSED:
2421 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token) < 0)
2422 goto nla_put_failure;
2423 break;
2424 case MPTCP_EVENT_ANNOUNCED:
2425 case MPTCP_EVENT_REMOVED:
2426 /* call mptcp_event_addr_announced()/removed instead */
2427 WARN_ON_ONCE(1);
2428 break;
2429 case MPTCP_EVENT_SUB_ESTABLISHED:
2430 case MPTCP_EVENT_SUB_PRIORITY:
2431 if (mptcp_event_sub_established(skb, msk, ssk) < 0)
2432 goto nla_put_failure;
2433 break;
2434 case MPTCP_EVENT_SUB_CLOSED:
2435 if (mptcp_event_sub_closed(skb, msk, ssk) < 0)
2436 goto nla_put_failure;
2437 break;
2438 case MPTCP_EVENT_LISTENER_CREATED:
2439 case MPTCP_EVENT_LISTENER_CLOSED:
2440 break;
2441 }
2442
2443 genlmsg_end(skb, nlh);
2444 mptcp_nl_mcast_send(net, skb, gfp);
2445 return;
2446
2447 nla_put_failure:
2448 nlmsg_free(skb);
2449 }
2450
2451 static const struct genl_small_ops mptcp_pm_ops[] = {
2452 {
2453 .cmd = MPTCP_PM_CMD_ADD_ADDR,
2454 .doit = mptcp_nl_cmd_add_addr,
2455 .flags = GENL_UNS_ADMIN_PERM,
2456 },
2457 {
2458 .cmd = MPTCP_PM_CMD_DEL_ADDR,
2459 .doit = mptcp_nl_cmd_del_addr,
2460 .flags = GENL_UNS_ADMIN_PERM,
2461 },
2462 {
2463 .cmd = MPTCP_PM_CMD_FLUSH_ADDRS,
2464 .doit = mptcp_nl_cmd_flush_addrs,
2465 .flags = GENL_UNS_ADMIN_PERM,
2466 },
2467 {
2468 .cmd = MPTCP_PM_CMD_GET_ADDR,
2469 .doit = mptcp_nl_cmd_get_addr,
2470 .dumpit = mptcp_nl_cmd_dump_addrs,
2471 },
2472 {
2473 .cmd = MPTCP_PM_CMD_SET_LIMITS,
2474 .doit = mptcp_nl_cmd_set_limits,
2475 .flags = GENL_UNS_ADMIN_PERM,
2476 },
2477 {
2478 .cmd = MPTCP_PM_CMD_GET_LIMITS,
2479 .doit = mptcp_nl_cmd_get_limits,
2480 },
2481 {
2482 .cmd = MPTCP_PM_CMD_SET_FLAGS,
2483 .doit = mptcp_nl_cmd_set_flags,
2484 .flags = GENL_UNS_ADMIN_PERM,
2485 },
2486 {
2487 .cmd = MPTCP_PM_CMD_ANNOUNCE,
2488 .doit = mptcp_nl_cmd_announce,
2489 .flags = GENL_UNS_ADMIN_PERM,
2490 },
2491 {
2492 .cmd = MPTCP_PM_CMD_REMOVE,
2493 .doit = mptcp_nl_cmd_remove,
2494 .flags = GENL_UNS_ADMIN_PERM,
2495 },
2496 {
2497 .cmd = MPTCP_PM_CMD_SUBFLOW_CREATE,
2498 .doit = mptcp_nl_cmd_sf_create,
2499 .flags = GENL_UNS_ADMIN_PERM,
2500 },
2501 {
2502 .cmd = MPTCP_PM_CMD_SUBFLOW_DESTROY,
2503 .doit = mptcp_nl_cmd_sf_destroy,
2504 .flags = GENL_UNS_ADMIN_PERM,
2505 },
2506 };
2507
2508 static struct genl_family mptcp_genl_family __ro_after_init = {
2509 .name = MPTCP_PM_NAME,
2510 .version = MPTCP_PM_VER,
2511 .maxattr = MPTCP_PM_ATTR_MAX,
2512 .policy = mptcp_pm_policy,
2513 .netnsok = true,
2514 .module = THIS_MODULE,
2515 .small_ops = mptcp_pm_ops,
2516 .n_small_ops = ARRAY_SIZE(mptcp_pm_ops),
2517 .resv_start_op = MPTCP_PM_CMD_SUBFLOW_DESTROY + 1,
2518 .mcgrps = mptcp_pm_mcgrps,
2519 .n_mcgrps = ARRAY_SIZE(mptcp_pm_mcgrps),
2520 };
2521
pm_nl_init_net(struct net * net)2522 static int __net_init pm_nl_init_net(struct net *net)
2523 {
2524 struct pm_nl_pernet *pernet = pm_nl_get_pernet(net);
2525
2526 INIT_LIST_HEAD_RCU(&pernet->local_addr_list);
2527
2528 /* Cit. 2 subflows ought to be enough for anybody. */
2529 pernet->subflows_max = 2;
2530 pernet->next_id = 1;
2531 pernet->stale_loss_cnt = 4;
2532 spin_lock_init(&pernet->lock);
2533
2534 /* No need to initialize other pernet fields, the struct is zeroed at
2535 * allocation time.
2536 */
2537
2538 return 0;
2539 }
2540
pm_nl_exit_net(struct list_head * net_list)2541 static void __net_exit pm_nl_exit_net(struct list_head *net_list)
2542 {
2543 struct net *net;
2544
2545 list_for_each_entry(net, net_list, exit_list) {
2546 struct pm_nl_pernet *pernet = pm_nl_get_pernet(net);
2547
2548 /* net is removed from namespace list, can't race with
2549 * other modifiers, also netns core already waited for a
2550 * RCU grace period.
2551 */
2552 __flush_addrs(&pernet->local_addr_list);
2553 }
2554 }
2555
2556 static struct pernet_operations mptcp_pm_pernet_ops = {
2557 .init = pm_nl_init_net,
2558 .exit_batch = pm_nl_exit_net,
2559 .id = &pm_nl_pernet_id,
2560 .size = sizeof(struct pm_nl_pernet),
2561 };
2562
mptcp_pm_nl_init(void)2563 void __init mptcp_pm_nl_init(void)
2564 {
2565 if (register_pernet_subsys(&mptcp_pm_pernet_ops) < 0)
2566 panic("Failed to register MPTCP PM pernet subsystem.\n");
2567
2568 if (genl_register_family(&mptcp_genl_family))
2569 panic("Failed to register MPTCP PM netlink family\n");
2570 }
2571